diff --git a/sys/contrib/openzfs/etc/systemd/system/zfs-import-cache.service.in b/sys/contrib/openzfs/etc/systemd/system/zfs-import-cache.service.in index b09961a8bb84..5e5c6281c9d7 100644 --- a/sys/contrib/openzfs/etc/systemd/system/zfs-import-cache.service.in +++ b/sys/contrib/openzfs/etc/systemd/system/zfs-import-cache.service.in @@ -1,20 +1,20 @@ [Unit] Description=Import ZFS pools by cache file Documentation=man:zpool(8) DefaultDependencies=no Requires=systemd-udev-settle.service After=systemd-udev-settle.service After=cryptsetup.target After=multipathd.target After=systemd-remount-fs.service Before=zfs-import.target ConditionFileNotEmpty=@sysconfdir@/zfs/zpool.cache ConditionPathIsDirectory=/sys/module/zfs [Service] Type=oneshot RemainAfterExit=yes -ExecStart=@sbindir@/zpool import -c @sysconfdir@/zfs/zpool.cache -aN ${ZPOOL_IMPORT_OPTS} +ExecStart=@sbindir@/zpool import -c @sysconfdir@/zfs/zpool.cache -aN $ZPOOL_IMPORT_OPTS [Install] WantedBy=zfs-import.target diff --git a/sys/contrib/openzfs/etc/systemd/system/zfs-import-scan.service.in b/sys/contrib/openzfs/etc/systemd/system/zfs-import-scan.service.in index c1111c73a607..d3c083f7e9cf 100644 --- a/sys/contrib/openzfs/etc/systemd/system/zfs-import-scan.service.in +++ b/sys/contrib/openzfs/etc/systemd/system/zfs-import-scan.service.in @@ -1,19 +1,19 @@ [Unit] Description=Import ZFS pools by device scanning Documentation=man:zpool(8) DefaultDependencies=no Requires=systemd-udev-settle.service After=systemd-udev-settle.service After=cryptsetup.target After=multipathd.target Before=zfs-import.target ConditionFileNotEmpty=!@sysconfdir@/zfs/zpool.cache ConditionPathIsDirectory=/sys/module/zfs [Service] Type=oneshot RemainAfterExit=yes -ExecStart=@sbindir@/zpool import -aN -o cachefile=none ${ZPOOL_IMPORT_OPTS} +ExecStart=@sbindir@/zpool import -aN -o cachefile=none $ZPOOL_IMPORT_OPTS [Install] WantedBy=zfs-import.target diff --git a/sys/contrib/openzfs/include/sys/abd_impl.h b/sys/contrib/openzfs/include/sys/abd_impl.h index 435a8dc6d9ce..6bce08cfa343 100644 --- a/sys/contrib/openzfs/include/sys/abd_impl.h +++ b/sys/contrib/openzfs/include/sys/abd_impl.h @@ -1,112 +1,112 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2014 by Chunwei Chen. All rights reserved. * Copyright (c) 2016, 2019 by Delphix. All rights reserved. */ #ifndef _ABD_IMPL_H #define _ABD_IMPL_H #include +#include #ifdef __cplusplus extern "C" { #endif typedef enum abd_stats_op { ABDSTAT_INCR, /* Increase abdstat values */ ABDSTAT_DECR /* Decrease abdstat values */ } abd_stats_op_t; struct scatterlist; /* forward declaration */ struct abd_iter { /* public interface */ void *iter_mapaddr; /* addr corresponding to iter_pos */ size_t iter_mapsize; /* length of data valid at mapaddr */ /* private */ abd_t *iter_abd; /* ABD being iterated through */ size_t iter_pos; size_t iter_offset; /* offset in current sg/abd_buf, */ /* abd_offset included */ struct scatterlist *iter_sg; /* current sg */ }; extern abd_t *abd_zero_scatter; abd_t *abd_gang_get_offset(abd_t *, size_t *); abd_t *abd_alloc_struct(size_t); void abd_free_struct(abd_t *); /* * OS specific functions */ abd_t *abd_alloc_struct_impl(size_t); abd_t *abd_get_offset_scatter(abd_t *, abd_t *, size_t); void abd_free_struct_impl(abd_t *); void abd_alloc_chunks(abd_t *, size_t); void abd_free_chunks(abd_t *); boolean_t abd_size_alloc_linear(size_t); void abd_update_scatter_stats(abd_t *, abd_stats_op_t); void abd_update_linear_stats(abd_t *, abd_stats_op_t); void abd_verify_scatter(abd_t *); void abd_free_linear_page(abd_t *); /* OS specific abd_iter functions */ void abd_iter_init(struct abd_iter *, abd_t *); boolean_t abd_iter_at_end(struct abd_iter *); void abd_iter_advance(struct abd_iter *, size_t); void abd_iter_map(struct abd_iter *); void abd_iter_unmap(struct abd_iter *); /* * Helper macros */ -#define ABDSTAT(stat) (abd_stats.stat.value.ui64) #define ABDSTAT_INCR(stat, val) \ - atomic_add_64(&abd_stats.stat.value.ui64, (val)) + wmsum_add(&abd_sums.stat, (val)) #define ABDSTAT_BUMP(stat) ABDSTAT_INCR(stat, 1) #define ABDSTAT_BUMPDOWN(stat) ABDSTAT_INCR(stat, -1) #define ABD_SCATTER(abd) (abd->abd_u.abd_scatter) #define ABD_LINEAR_BUF(abd) (abd->abd_u.abd_linear.abd_buf) #define ABD_GANG(abd) (abd->abd_u.abd_gang) #if defined(_KERNEL) #if defined(__FreeBSD__) #define abd_enter_critical(flags) critical_enter() #define abd_exit_critical(flags) critical_exit() #else #define abd_enter_critical(flags) local_irq_save(flags) #define abd_exit_critical(flags) local_irq_restore(flags) #endif #else /* !_KERNEL */ #define abd_enter_critical(flags) ((void)0) #define abd_exit_critical(flags) ((void)0) #endif #ifdef __cplusplus } #endif #endif /* _ABD_IMPL_H */ diff --git a/sys/contrib/openzfs/include/sys/arc_impl.h b/sys/contrib/openzfs/include/sys/arc_impl.h index c01da46e01e3..1f341ec94faf 100644 --- a/sys/contrib/openzfs/include/sys/arc_impl.h +++ b/sys/contrib/openzfs/include/sys/arc_impl.h @@ -1,948 +1,1015 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2013, Delphix. All rights reserved. * Copyright (c) 2013, Saso Kiselkov. All rights reserved. * Copyright (c) 2013, Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2020, George Amanakis. All rights reserved. */ #ifndef _SYS_ARC_IMPL_H #define _SYS_ARC_IMPL_H #include #include #include #include +#include #ifdef __cplusplus extern "C" { #endif /* * Note that buffers can be in one of 6 states: * ARC_anon - anonymous (discussed below) * ARC_mru - recently used, currently cached * ARC_mru_ghost - recently used, no longer in cache * ARC_mfu - frequently used, currently cached * ARC_mfu_ghost - frequently used, no longer in cache * ARC_l2c_only - exists in L2ARC but not other states * When there are no active references to the buffer, they are * are linked onto a list in one of these arc states. These are * the only buffers that can be evicted or deleted. Within each * state there are multiple lists, one for meta-data and one for * non-meta-data. Meta-data (indirect blocks, blocks of dnodes, * etc.) is tracked separately so that it can be managed more * explicitly: favored over data, limited explicitly. * * Anonymous buffers are buffers that are not associated with * a DVA. These are buffers that hold dirty block copies * before they are written to stable storage. By definition, * they are "ref'd" and are considered part of arc_mru * that cannot be freed. Generally, they will acquire a DVA * as they are written and migrate onto the arc_mru list. * * The ARC_l2c_only state is for buffers that are in the second * level ARC but no longer in any of the ARC_m* lists. The second * level ARC itself may also contain buffers that are in any of * the ARC_m* states - meaning that a buffer can exist in two * places. The reason for the ARC_l2c_only state is to keep the * buffer header in the hash table, so that reads that hit the * second level ARC benefit from these fast lookups. */ typedef struct arc_state { /* * list of evictable buffers */ multilist_t arcs_list[ARC_BUFC_NUMTYPES]; /* * supports the "dbufs" kstat */ arc_state_type_t arcs_state; /* * total amount of evictable data in this state */ zfs_refcount_t arcs_esize[ARC_BUFC_NUMTYPES] ____cacheline_aligned; /* * total amount of data in this state; this includes: evictable, * non-evictable, ARC_BUFC_DATA, and ARC_BUFC_METADATA. */ zfs_refcount_t arcs_size; } arc_state_t; typedef struct arc_callback arc_callback_t; struct arc_callback { void *acb_private; arc_read_done_func_t *acb_done; arc_buf_t *acb_buf; boolean_t acb_encrypted; boolean_t acb_compressed; boolean_t acb_noauth; boolean_t acb_nobuf; zbookmark_phys_t acb_zb; zio_t *acb_zio_dummy; zio_t *acb_zio_head; arc_callback_t *acb_next; }; typedef struct arc_write_callback arc_write_callback_t; struct arc_write_callback { void *awcb_private; arc_write_done_func_t *awcb_ready; arc_write_done_func_t *awcb_children_ready; arc_write_done_func_t *awcb_physdone; arc_write_done_func_t *awcb_done; arc_buf_t *awcb_buf; }; /* * ARC buffers are separated into multiple structs as a memory saving measure: * - Common fields struct, always defined, and embedded within it: * - L2-only fields, always allocated but undefined when not in L2ARC * - L1-only fields, only allocated when in L1ARC * * Buffer in L1 Buffer only in L2 * +------------------------+ +------------------------+ * | arc_buf_hdr_t | | arc_buf_hdr_t | * | | | | * | | | | * | | | | * +------------------------+ +------------------------+ * | l2arc_buf_hdr_t | | l2arc_buf_hdr_t | * | (undefined if L1-only) | | | * +------------------------+ +------------------------+ * | l1arc_buf_hdr_t | * | | * | | * | | * | | * +------------------------+ * * Because it's possible for the L2ARC to become extremely large, we can wind * up eating a lot of memory in L2ARC buffer headers, so the size of a header * is minimized by only allocating the fields necessary for an L1-cached buffer * when a header is actually in the L1 cache. The sub-headers (l1arc_buf_hdr and * l2arc_buf_hdr) are embedded rather than allocated separately to save a couple * words in pointers. arc_hdr_realloc() is used to switch a header between * these two allocation states. */ typedef struct l1arc_buf_hdr { kmutex_t b_freeze_lock; zio_cksum_t *b_freeze_cksum; arc_buf_t *b_buf; uint32_t b_bufcnt; /* for waiting on writes to complete */ kcondvar_t b_cv; uint8_t b_byteswap; /* protected by arc state mutex */ arc_state_t *b_state; multilist_node_t b_arc_node; /* updated atomically */ clock_t b_arc_access; uint32_t b_mru_hits; uint32_t b_mru_ghost_hits; uint32_t b_mfu_hits; uint32_t b_mfu_ghost_hits; uint32_t b_l2_hits; /* self protecting */ zfs_refcount_t b_refcnt; arc_callback_t *b_acb; abd_t *b_pabd; } l1arc_buf_hdr_t; typedef enum l2arc_dev_hdr_flags_t { L2ARC_DEV_HDR_EVICT_FIRST = (1 << 0) /* mirror of l2ad_first */ } l2arc_dev_hdr_flags_t; /* * Pointer used in persistent L2ARC (for pointing to log blocks). */ typedef struct l2arc_log_blkptr { /* * Offset of log block within the device, in bytes */ uint64_t lbp_daddr; /* * Aligned payload size (in bytes) of the log block */ uint64_t lbp_payload_asize; /* * Offset in bytes of the first buffer in the payload */ uint64_t lbp_payload_start; /* * lbp_prop has the following format: * * logical size (in bytes) * * aligned (after compression) size (in bytes) * * compression algorithm (we always LZ4-compress l2arc logs) * * checksum algorithm (used for lbp_cksum) */ uint64_t lbp_prop; zio_cksum_t lbp_cksum; /* checksum of log */ } l2arc_log_blkptr_t; /* * The persistent L2ARC device header. * Byte order of magic determines whether 64-bit bswap of fields is necessary. */ typedef struct l2arc_dev_hdr_phys { uint64_t dh_magic; /* L2ARC_DEV_HDR_MAGIC */ uint64_t dh_version; /* Persistent L2ARC version */ /* * Global L2ARC device state and metadata. */ uint64_t dh_spa_guid; uint64_t dh_vdev_guid; uint64_t dh_log_entries; /* mirror of l2ad_log_entries */ uint64_t dh_evict; /* evicted offset in bytes */ uint64_t dh_flags; /* l2arc_dev_hdr_flags_t */ /* * Used in zdb.c for determining if a log block is valid, in the same * way that l2arc_rebuild() does. */ uint64_t dh_start; /* mirror of l2ad_start */ uint64_t dh_end; /* mirror of l2ad_end */ /* * Start of log block chain. [0] -> newest log, [1] -> one older (used * for initiating prefetch). */ l2arc_log_blkptr_t dh_start_lbps[2]; /* * Aligned size of all log blocks as accounted by vdev_space_update(). */ uint64_t dh_lb_asize; /* mirror of l2ad_lb_asize */ uint64_t dh_lb_count; /* mirror of l2ad_lb_count */ /* * Mirrors of vdev_trim_action_time and vdev_trim_state, used to * display when the cache device was fully trimmed for the last * time. */ uint64_t dh_trim_action_time; uint64_t dh_trim_state; const uint64_t dh_pad[30]; /* pad to 512 bytes */ zio_eck_t dh_tail; } l2arc_dev_hdr_phys_t; CTASSERT_GLOBAL(sizeof (l2arc_dev_hdr_phys_t) == SPA_MINBLOCKSIZE); /* * A single ARC buffer header entry in a l2arc_log_blk_phys_t. */ typedef struct l2arc_log_ent_phys { dva_t le_dva; /* dva of buffer */ uint64_t le_birth; /* birth txg of buffer */ /* * le_prop has the following format: * * logical size (in bytes) * * physical (compressed) size (in bytes) * * compression algorithm * * object type (used to restore arc_buf_contents_t) * * protected status (used for encryption) * * prefetch status (used in l2arc_read_done()) */ uint64_t le_prop; uint64_t le_daddr; /* buf location on l2dev */ uint64_t le_complevel; /* * We pad the size of each entry to a power of 2 so that the size of * l2arc_log_blk_phys_t is power-of-2 aligned with SPA_MINBLOCKSHIFT, * because of the L2ARC_SET_*SIZE macros. */ const uint64_t le_pad[2]; /* pad to 64 bytes */ } l2arc_log_ent_phys_t; #define L2ARC_LOG_BLK_MAX_ENTRIES (1022) /* * A log block of up to 1022 ARC buffer log entries, chained into the * persistent L2ARC metadata linked list. Byte order of magic determines * whether 64-bit bswap of fields is necessary. */ typedef struct l2arc_log_blk_phys { uint64_t lb_magic; /* L2ARC_LOG_BLK_MAGIC */ /* * There are 2 chains (headed by dh_start_lbps[2]), and this field * points back to the previous block in this chain. We alternate * which chain we append to, so they are time-wise and offset-wise * interleaved, but that is an optimization rather than for * correctness. */ l2arc_log_blkptr_t lb_prev_lbp; /* pointer to prev log block */ /* * Pad header section to 128 bytes */ uint64_t lb_pad[7]; /* Payload */ l2arc_log_ent_phys_t lb_entries[L2ARC_LOG_BLK_MAX_ENTRIES]; } l2arc_log_blk_phys_t; /* 64K total */ /* * The size of l2arc_log_blk_phys_t has to be power-of-2 aligned with * SPA_MINBLOCKSHIFT because of L2BLK_SET_*SIZE macros. */ CTASSERT_GLOBAL(IS_P2ALIGNED(sizeof (l2arc_log_blk_phys_t), 1ULL << SPA_MINBLOCKSHIFT)); CTASSERT_GLOBAL(sizeof (l2arc_log_blk_phys_t) >= SPA_MINBLOCKSIZE); CTASSERT_GLOBAL(sizeof (l2arc_log_blk_phys_t) <= SPA_MAXBLOCKSIZE); /* * These structures hold in-flight abd buffers for log blocks as they're being * written to the L2ARC device. */ typedef struct l2arc_lb_abd_buf { abd_t *abd; list_node_t node; } l2arc_lb_abd_buf_t; /* * These structures hold pointers to log blocks present on the L2ARC device. */ typedef struct l2arc_lb_ptr_buf { l2arc_log_blkptr_t *lb_ptr; list_node_t node; } l2arc_lb_ptr_buf_t; /* Macros for setting fields in le_prop and lbp_prop */ #define L2BLK_GET_LSIZE(field) \ BF64_GET_SB((field), 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1) #define L2BLK_SET_LSIZE(field, x) \ BF64_SET_SB((field), 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x) #define L2BLK_GET_PSIZE(field) \ BF64_GET_SB((field), 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1) #define L2BLK_SET_PSIZE(field, x) \ BF64_SET_SB((field), 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x) #define L2BLK_GET_COMPRESS(field) \ BF64_GET((field), 32, SPA_COMPRESSBITS) #define L2BLK_SET_COMPRESS(field, x) \ BF64_SET((field), 32, SPA_COMPRESSBITS, x) #define L2BLK_GET_PREFETCH(field) BF64_GET((field), 39, 1) #define L2BLK_SET_PREFETCH(field, x) BF64_SET((field), 39, 1, x) #define L2BLK_GET_CHECKSUM(field) BF64_GET((field), 40, 8) #define L2BLK_SET_CHECKSUM(field, x) BF64_SET((field), 40, 8, x) #define L2BLK_GET_TYPE(field) BF64_GET((field), 48, 8) #define L2BLK_SET_TYPE(field, x) BF64_SET((field), 48, 8, x) #define L2BLK_GET_PROTECTED(field) BF64_GET((field), 56, 1) #define L2BLK_SET_PROTECTED(field, x) BF64_SET((field), 56, 1, x) #define L2BLK_GET_STATE(field) BF64_GET((field), 57, 4) #define L2BLK_SET_STATE(field, x) BF64_SET((field), 57, 4, x) #define PTR_SWAP(x, y) \ do { \ void *tmp = (x);\ x = y; \ y = tmp; \ _NOTE(CONSTCOND)\ } while (0) #define L2ARC_DEV_HDR_MAGIC 0x5a46534341434845LLU /* ASCII: "ZFSCACHE" */ #define L2ARC_LOG_BLK_MAGIC 0x4c4f47424c4b4844LLU /* ASCII: "LOGBLKHD" */ /* * L2ARC Internals */ typedef struct l2arc_dev { vdev_t *l2ad_vdev; /* vdev */ spa_t *l2ad_spa; /* spa */ uint64_t l2ad_hand; /* next write location */ uint64_t l2ad_start; /* first addr on device */ uint64_t l2ad_end; /* last addr on device */ boolean_t l2ad_first; /* first sweep through */ boolean_t l2ad_writing; /* currently writing */ kmutex_t l2ad_mtx; /* lock for buffer list */ list_t l2ad_buflist; /* buffer list */ list_node_t l2ad_node; /* device list node */ zfs_refcount_t l2ad_alloc; /* allocated bytes */ /* * Persistence-related stuff */ l2arc_dev_hdr_phys_t *l2ad_dev_hdr; /* persistent device header */ uint64_t l2ad_dev_hdr_asize; /* aligned hdr size */ l2arc_log_blk_phys_t l2ad_log_blk; /* currently open log block */ int l2ad_log_ent_idx; /* index into cur log blk */ /* Number of bytes in current log block's payload */ uint64_t l2ad_log_blk_payload_asize; /* * Offset (in bytes) of the first buffer in current log block's * payload. */ uint64_t l2ad_log_blk_payload_start; /* Flag indicating whether a rebuild is scheduled or is going on */ boolean_t l2ad_rebuild; boolean_t l2ad_rebuild_cancel; boolean_t l2ad_rebuild_began; uint64_t l2ad_log_entries; /* entries per log blk */ uint64_t l2ad_evict; /* evicted offset in bytes */ /* List of pointers to log blocks present in the L2ARC device */ list_t l2ad_lbptr_list; /* * Aligned size of all log blocks as accounted by vdev_space_update(). */ zfs_refcount_t l2ad_lb_asize; /* * Number of log blocks present on the device. */ zfs_refcount_t l2ad_lb_count; boolean_t l2ad_trim_all; /* TRIM whole device */ } l2arc_dev_t; /* * Encrypted blocks will need to be stored encrypted on the L2ARC * disk as they appear in the main pool. In order for this to work we * need to pass around the encryption parameters so they can be used * to write data to the L2ARC. This struct is only defined in the * arc_buf_hdr_t if the L1 header is defined and has the ARC_FLAG_ENCRYPTED * flag set. */ typedef struct arc_buf_hdr_crypt { abd_t *b_rabd; /* raw encrypted data */ dmu_object_type_t b_ot; /* object type */ uint32_t b_ebufcnt; /* count of encrypted buffers */ /* dsobj for looking up encryption key for l2arc encryption */ uint64_t b_dsobj; /* encryption parameters */ uint8_t b_salt[ZIO_DATA_SALT_LEN]; uint8_t b_iv[ZIO_DATA_IV_LEN]; /* * Technically this could be removed since we will always be able to * get the mac from the bp when we need it. However, it is inconvenient * for callers of arc code to have to pass a bp in all the time. This * also allows us to assert that L2ARC data is properly encrypted to * match the data in the main storage pool. */ uint8_t b_mac[ZIO_DATA_MAC_LEN]; } arc_buf_hdr_crypt_t; typedef struct l2arc_buf_hdr { /* protected by arc_buf_hdr mutex */ l2arc_dev_t *b_dev; /* L2ARC device */ uint64_t b_daddr; /* disk address, offset byte */ uint32_t b_hits; arc_state_type_t b_arcs_state; list_node_t b_l2node; } l2arc_buf_hdr_t; typedef struct l2arc_write_callback { l2arc_dev_t *l2wcb_dev; /* device info */ arc_buf_hdr_t *l2wcb_head; /* head of write buflist */ /* in-flight list of log blocks */ list_t l2wcb_abd_list; } l2arc_write_callback_t; struct arc_buf_hdr { /* protected by hash lock */ dva_t b_dva; uint64_t b_birth; arc_buf_contents_t b_type; uint8_t b_complevel; uint8_t b_reserved1; /* used for 4 byte alignment */ uint16_t b_reserved2; /* used for 4 byte alignment */ arc_buf_hdr_t *b_hash_next; arc_flags_t b_flags; /* * This field stores the size of the data buffer after * compression, and is set in the arc's zio completion handlers. * It is in units of SPA_MINBLOCKSIZE (e.g. 1 == 512 bytes). * * While the block pointers can store up to 32MB in their psize * field, we can only store up to 32MB minus 512B. This is due * to the bp using a bias of 1, whereas we use a bias of 0 (i.e. * a field of zeros represents 512B in the bp). We can't use a * bias of 1 since we need to reserve a psize of zero, here, to * represent holes and embedded blocks. * * This isn't a problem in practice, since the maximum size of a * buffer is limited to 16MB, so we never need to store 32MB in * this field. Even in the upstream illumos code base, the * maximum size of a buffer is limited to 16MB. */ uint16_t b_psize; /* * This field stores the size of the data buffer before * compression, and cannot change once set. It is in units * of SPA_MINBLOCKSIZE (e.g. 2 == 1024 bytes) */ uint16_t b_lsize; /* immutable */ uint64_t b_spa; /* immutable */ /* L2ARC fields. Undefined when not in L2ARC. */ l2arc_buf_hdr_t b_l2hdr; /* L1ARC fields. Undefined when in l2arc_only state */ l1arc_buf_hdr_t b_l1hdr; /* * Encryption parameters. Defined only when ARC_FLAG_ENCRYPTED * is set and the L1 header exists. */ arc_buf_hdr_crypt_t b_crypt_hdr; }; typedef struct arc_stats { kstat_named_t arcstat_hits; kstat_named_t arcstat_misses; kstat_named_t arcstat_demand_data_hits; kstat_named_t arcstat_demand_data_misses; kstat_named_t arcstat_demand_metadata_hits; kstat_named_t arcstat_demand_metadata_misses; kstat_named_t arcstat_prefetch_data_hits; kstat_named_t arcstat_prefetch_data_misses; kstat_named_t arcstat_prefetch_metadata_hits; kstat_named_t arcstat_prefetch_metadata_misses; kstat_named_t arcstat_mru_hits; kstat_named_t arcstat_mru_ghost_hits; kstat_named_t arcstat_mfu_hits; kstat_named_t arcstat_mfu_ghost_hits; kstat_named_t arcstat_deleted; /* * Number of buffers that could not be evicted because the hash lock * was held by another thread. The lock may not necessarily be held * by something using the same buffer, since hash locks are shared * by multiple buffers. */ kstat_named_t arcstat_mutex_miss; /* * Number of buffers skipped when updating the access state due to the * header having already been released after acquiring the hash lock. */ kstat_named_t arcstat_access_skip; /* * Number of buffers skipped because they have I/O in progress, are * indirect prefetch buffers that have not lived long enough, or are * not from the spa we're trying to evict from. */ kstat_named_t arcstat_evict_skip; /* * Number of times arc_evict_state() was unable to evict enough * buffers to reach its target amount. */ kstat_named_t arcstat_evict_not_enough; kstat_named_t arcstat_evict_l2_cached; kstat_named_t arcstat_evict_l2_eligible; kstat_named_t arcstat_evict_l2_eligible_mfu; kstat_named_t arcstat_evict_l2_eligible_mru; kstat_named_t arcstat_evict_l2_ineligible; kstat_named_t arcstat_evict_l2_skip; kstat_named_t arcstat_hash_elements; kstat_named_t arcstat_hash_elements_max; kstat_named_t arcstat_hash_collisions; kstat_named_t arcstat_hash_chains; kstat_named_t arcstat_hash_chain_max; kstat_named_t arcstat_p; kstat_named_t arcstat_c; kstat_named_t arcstat_c_min; kstat_named_t arcstat_c_max; - /* Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_size; /* * Number of compressed bytes stored in the arc_buf_hdr_t's b_pabd. * Note that the compressed bytes may match the uncompressed bytes * if the block is either not compressed or compressed arc is disabled. */ kstat_named_t arcstat_compressed_size; /* * Uncompressed size of the data stored in b_pabd. If compressed * arc is disabled then this value will be identical to the stat * above. */ kstat_named_t arcstat_uncompressed_size; /* * Number of bytes stored in all the arc_buf_t's. This is classified * as "overhead" since this data is typically short-lived and will * be evicted from the arc when it becomes unreferenced unless the * zfs_keep_uncompressed_metadata or zfs_keep_uncompressed_level * values have been set (see comment in dbuf.c for more information). */ kstat_named_t arcstat_overhead_size; /* * Number of bytes consumed by internal ARC structures necessary * for tracking purposes; these structures are not actually * backed by ARC buffers. This includes arc_buf_hdr_t structures * (allocated via arc_buf_hdr_t_full and arc_buf_hdr_t_l2only * caches), and arc_buf_t structures (allocated via arc_buf_t * cache). - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_hdr_size; /* * Number of bytes consumed by ARC buffers of type equal to * ARC_BUFC_DATA. This is generally consumed by buffers backing * on disk user data (e.g. plain file contents). - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_data_size; /* * Number of bytes consumed by ARC buffers of type equal to * ARC_BUFC_METADATA. This is generally consumed by buffers * backing on disk data that is used for internal ZFS * structures (e.g. ZAP, dnode, indirect blocks, etc). - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_metadata_size; /* * Number of bytes consumed by dmu_buf_impl_t objects. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_dbuf_size; /* * Number of bytes consumed by dnode_t objects. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_dnode_size; /* * Number of bytes consumed by bonus buffers. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_bonus_size; #if defined(COMPAT_FREEBSD11) /* * Sum of the previous three counters, provided for compatibility. */ kstat_named_t arcstat_other_size; #endif /* * Total number of bytes consumed by ARC buffers residing in the * arc_anon state. This includes *all* buffers in the arc_anon * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_anon_size; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_DATA, * residing in the arc_anon state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_anon_evictable_data; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_METADATA, * residing in the arc_anon state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_anon_evictable_metadata; /* * Total number of bytes consumed by ARC buffers residing in the * arc_mru state. This includes *all* buffers in the arc_mru * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mru_size; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_DATA, * residing in the arc_mru state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mru_evictable_data; /* * Number of bytes consumed by ARC buffers that meet the * following criteria: backing buffers of type ARC_BUFC_METADATA, * residing in the arc_mru state, and are eligible for eviction * (e.g. have no outstanding holds on the buffer). - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mru_evictable_metadata; /* * Total number of bytes that *would have been* consumed by ARC * buffers in the arc_mru_ghost state. The key thing to note * here, is the fact that this size doesn't actually indicate * RAM consumption. The ghost lists only consist of headers and * don't actually have ARC buffers linked off of these headers. * Thus, *if* the headers had associated ARC buffers, these * buffers *would have* consumed this number of bytes. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mru_ghost_size; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_DATA, and linked off the arc_mru_ghost state. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mru_ghost_evictable_data; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mru_ghost_evictable_metadata; /* * Total number of bytes consumed by ARC buffers residing in the * arc_mfu state. This includes *all* buffers in the arc_mfu * state; e.g. data, metadata, evictable, and unevictable buffers * are all included in this value. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mfu_size; /* * Number of bytes consumed by ARC buffers that are eligible for * eviction, of type ARC_BUFC_DATA, and reside in the arc_mfu * state. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mfu_evictable_data; /* * Number of bytes consumed by ARC buffers that are eligible for * eviction, of type ARC_BUFC_METADATA, and reside in the * arc_mfu state. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mfu_evictable_metadata; /* * Total number of bytes that *would have been* consumed by ARC * buffers in the arc_mfu_ghost state. See the comment above * arcstat_mru_ghost_size for more details. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mfu_ghost_size; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_DATA, and linked off the arc_mfu_ghost state. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mfu_ghost_evictable_data; /* * Number of bytes that *would have been* consumed by ARC * buffers that are eligible for eviction, of type * ARC_BUFC_METADATA, and linked off the arc_mru_ghost state. - * Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_mfu_ghost_evictable_metadata; kstat_named_t arcstat_l2_hits; kstat_named_t arcstat_l2_misses; /* * Allocated size (in bytes) of L2ARC cached buffers by ARC state. */ kstat_named_t arcstat_l2_prefetch_asize; kstat_named_t arcstat_l2_mru_asize; kstat_named_t arcstat_l2_mfu_asize; /* * Allocated size (in bytes) of L2ARC cached buffers by buffer content * type. */ kstat_named_t arcstat_l2_bufc_data_asize; kstat_named_t arcstat_l2_bufc_metadata_asize; kstat_named_t arcstat_l2_feeds; kstat_named_t arcstat_l2_rw_clash; kstat_named_t arcstat_l2_read_bytes; kstat_named_t arcstat_l2_write_bytes; kstat_named_t arcstat_l2_writes_sent; kstat_named_t arcstat_l2_writes_done; kstat_named_t arcstat_l2_writes_error; kstat_named_t arcstat_l2_writes_lock_retry; kstat_named_t arcstat_l2_evict_lock_retry; kstat_named_t arcstat_l2_evict_reading; kstat_named_t arcstat_l2_evict_l1cached; kstat_named_t arcstat_l2_free_on_write; kstat_named_t arcstat_l2_abort_lowmem; kstat_named_t arcstat_l2_cksum_bad; kstat_named_t arcstat_l2_io_error; kstat_named_t arcstat_l2_lsize; kstat_named_t arcstat_l2_psize; - /* Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_l2_hdr_size; /* * Number of L2ARC log blocks written. These are used for restoring the * L2ARC. Updated during writing of L2ARC log blocks. */ kstat_named_t arcstat_l2_log_blk_writes; /* * Moving average of the aligned size of the L2ARC log blocks, in * bytes. Updated during L2ARC rebuild and during writing of L2ARC * log blocks. */ kstat_named_t arcstat_l2_log_blk_avg_asize; /* Aligned size of L2ARC log blocks on L2ARC devices. */ kstat_named_t arcstat_l2_log_blk_asize; /* Number of L2ARC log blocks present on L2ARC devices. */ kstat_named_t arcstat_l2_log_blk_count; /* * Moving average of the aligned size of L2ARC restored data, in bytes, * to the aligned size of their metadata in L2ARC, in bytes. * Updated during L2ARC rebuild and during writing of L2ARC log blocks. */ kstat_named_t arcstat_l2_data_to_meta_ratio; /* * Number of times the L2ARC rebuild was successful for an L2ARC device. */ kstat_named_t arcstat_l2_rebuild_success; /* * Number of times the L2ARC rebuild failed because the device header * was in an unsupported format or corrupted. */ kstat_named_t arcstat_l2_rebuild_abort_unsupported; /* * Number of times the L2ARC rebuild failed because of IO errors * while reading a log block. */ kstat_named_t arcstat_l2_rebuild_abort_io_errors; /* * Number of times the L2ARC rebuild failed because of IO errors when * reading the device header. */ kstat_named_t arcstat_l2_rebuild_abort_dh_errors; /* * Number of L2ARC log blocks which failed to be restored due to * checksum errors. */ kstat_named_t arcstat_l2_rebuild_abort_cksum_lb_errors; /* * Number of times the L2ARC rebuild was aborted due to low system * memory. */ kstat_named_t arcstat_l2_rebuild_abort_lowmem; /* Logical size of L2ARC restored data, in bytes. */ kstat_named_t arcstat_l2_rebuild_size; /* Aligned size of L2ARC restored data, in bytes. */ kstat_named_t arcstat_l2_rebuild_asize; /* * Number of L2ARC log entries (buffers) that were successfully * restored in ARC. */ kstat_named_t arcstat_l2_rebuild_bufs; /* * Number of L2ARC log entries (buffers) already cached in ARC. These * were not restored again. */ kstat_named_t arcstat_l2_rebuild_bufs_precached; /* * Number of L2ARC log blocks that were restored successfully. Each * log block may hold up to L2ARC_LOG_BLK_MAX_ENTRIES buffers. */ kstat_named_t arcstat_l2_rebuild_log_blks; kstat_named_t arcstat_memory_throttle_count; kstat_named_t arcstat_memory_direct_count; kstat_named_t arcstat_memory_indirect_count; kstat_named_t arcstat_memory_all_bytes; kstat_named_t arcstat_memory_free_bytes; kstat_named_t arcstat_memory_available_bytes; kstat_named_t arcstat_no_grow; kstat_named_t arcstat_tempreserve; kstat_named_t arcstat_loaned_bytes; kstat_named_t arcstat_prune; - /* Not updated directly; only synced in arc_kstat_update. */ kstat_named_t arcstat_meta_used; kstat_named_t arcstat_meta_limit; kstat_named_t arcstat_dnode_limit; kstat_named_t arcstat_meta_max; kstat_named_t arcstat_meta_min; kstat_named_t arcstat_async_upgrade_sync; kstat_named_t arcstat_demand_hit_predictive_prefetch; kstat_named_t arcstat_demand_hit_prescient_prefetch; kstat_named_t arcstat_need_free; kstat_named_t arcstat_sys_free; kstat_named_t arcstat_raw_size; kstat_named_t arcstat_cached_only_in_progress; kstat_named_t arcstat_abd_chunk_waste_size; } arc_stats_t; +typedef struct arc_sums { + wmsum_t arcstat_hits; + wmsum_t arcstat_misses; + wmsum_t arcstat_demand_data_hits; + wmsum_t arcstat_demand_data_misses; + wmsum_t arcstat_demand_metadata_hits; + wmsum_t arcstat_demand_metadata_misses; + wmsum_t arcstat_prefetch_data_hits; + wmsum_t arcstat_prefetch_data_misses; + wmsum_t arcstat_prefetch_metadata_hits; + wmsum_t arcstat_prefetch_metadata_misses; + wmsum_t arcstat_mru_hits; + wmsum_t arcstat_mru_ghost_hits; + wmsum_t arcstat_mfu_hits; + wmsum_t arcstat_mfu_ghost_hits; + wmsum_t arcstat_deleted; + wmsum_t arcstat_mutex_miss; + wmsum_t arcstat_access_skip; + wmsum_t arcstat_evict_skip; + wmsum_t arcstat_evict_not_enough; + wmsum_t arcstat_evict_l2_cached; + wmsum_t arcstat_evict_l2_eligible; + wmsum_t arcstat_evict_l2_eligible_mfu; + wmsum_t arcstat_evict_l2_eligible_mru; + wmsum_t arcstat_evict_l2_ineligible; + wmsum_t arcstat_evict_l2_skip; + wmsum_t arcstat_hash_collisions; + wmsum_t arcstat_hash_chains; + aggsum_t arcstat_size; + wmsum_t arcstat_compressed_size; + wmsum_t arcstat_uncompressed_size; + wmsum_t arcstat_overhead_size; + wmsum_t arcstat_hdr_size; + wmsum_t arcstat_data_size; + wmsum_t arcstat_metadata_size; + wmsum_t arcstat_dbuf_size; + aggsum_t arcstat_dnode_size; + wmsum_t arcstat_bonus_size; + wmsum_t arcstat_l2_hits; + wmsum_t arcstat_l2_misses; + wmsum_t arcstat_l2_prefetch_asize; + wmsum_t arcstat_l2_mru_asize; + wmsum_t arcstat_l2_mfu_asize; + wmsum_t arcstat_l2_bufc_data_asize; + wmsum_t arcstat_l2_bufc_metadata_asize; + wmsum_t arcstat_l2_feeds; + wmsum_t arcstat_l2_rw_clash; + wmsum_t arcstat_l2_read_bytes; + wmsum_t arcstat_l2_write_bytes; + wmsum_t arcstat_l2_writes_sent; + wmsum_t arcstat_l2_writes_done; + wmsum_t arcstat_l2_writes_error; + wmsum_t arcstat_l2_writes_lock_retry; + wmsum_t arcstat_l2_evict_lock_retry; + wmsum_t arcstat_l2_evict_reading; + wmsum_t arcstat_l2_evict_l1cached; + wmsum_t arcstat_l2_free_on_write; + wmsum_t arcstat_l2_abort_lowmem; + wmsum_t arcstat_l2_cksum_bad; + wmsum_t arcstat_l2_io_error; + wmsum_t arcstat_l2_lsize; + wmsum_t arcstat_l2_psize; + aggsum_t arcstat_l2_hdr_size; + wmsum_t arcstat_l2_log_blk_writes; + wmsum_t arcstat_l2_log_blk_asize; + wmsum_t arcstat_l2_log_blk_count; + wmsum_t arcstat_l2_rebuild_success; + wmsum_t arcstat_l2_rebuild_abort_unsupported; + wmsum_t arcstat_l2_rebuild_abort_io_errors; + wmsum_t arcstat_l2_rebuild_abort_dh_errors; + wmsum_t arcstat_l2_rebuild_abort_cksum_lb_errors; + wmsum_t arcstat_l2_rebuild_abort_lowmem; + wmsum_t arcstat_l2_rebuild_size; + wmsum_t arcstat_l2_rebuild_asize; + wmsum_t arcstat_l2_rebuild_bufs; + wmsum_t arcstat_l2_rebuild_bufs_precached; + wmsum_t arcstat_l2_rebuild_log_blks; + wmsum_t arcstat_memory_throttle_count; + wmsum_t arcstat_memory_direct_count; + wmsum_t arcstat_memory_indirect_count; + wmsum_t arcstat_prune; + aggsum_t arcstat_meta_used; + wmsum_t arcstat_async_upgrade_sync; + wmsum_t arcstat_demand_hit_predictive_prefetch; + wmsum_t arcstat_demand_hit_prescient_prefetch; + wmsum_t arcstat_raw_size; + wmsum_t arcstat_cached_only_in_progress; + wmsum_t arcstat_abd_chunk_waste_size; +} arc_sums_t; + typedef struct arc_evict_waiter { list_node_t aew_node; kcondvar_t aew_cv; uint64_t aew_count; } arc_evict_waiter_t; #define ARCSTAT(stat) (arc_stats.stat.value.ui64) #define ARCSTAT_INCR(stat, val) \ - atomic_add_64(&arc_stats.stat.value.ui64, (val)) + wmsum_add(&arc_sums.stat, (val)) #define ARCSTAT_BUMP(stat) ARCSTAT_INCR(stat, 1) #define ARCSTAT_BUMPDOWN(stat) ARCSTAT_INCR(stat, -1) #define arc_no_grow ARCSTAT(arcstat_no_grow) /* do not grow cache size */ #define arc_p ARCSTAT(arcstat_p) /* target size of MRU */ #define arc_c ARCSTAT(arcstat_c) /* target size of cache */ #define arc_c_min ARCSTAT(arcstat_c_min) /* min target cache size */ #define arc_c_max ARCSTAT(arcstat_c_max) /* max target cache size */ #define arc_sys_free ARCSTAT(arcstat_sys_free) /* target system free bytes */ extern taskq_t *arc_prune_taskq; extern arc_stats_t arc_stats; +extern arc_sums_t arc_sums; extern hrtime_t arc_growtime; extern boolean_t arc_warm; extern int arc_grow_retry; extern int arc_no_grow_shift; extern int arc_shrink_shift; extern kmutex_t arc_prune_mtx; extern list_t arc_prune_list; -extern aggsum_t arc_size; extern arc_state_t *arc_mfu; extern arc_state_t *arc_mru; extern uint_t zfs_arc_pc_percent; extern int arc_lotsfree_percent; extern unsigned long zfs_arc_min; extern unsigned long zfs_arc_max; extern void arc_reduce_target_size(int64_t to_free); extern boolean_t arc_reclaim_needed(void); extern void arc_kmem_reap_soon(void); extern boolean_t arc_is_overflowing(void); extern void arc_wait_for_eviction(uint64_t); extern void arc_lowmem_init(void); extern void arc_lowmem_fini(void); extern void arc_prune_async(int64_t); extern int arc_memory_throttle(spa_t *spa, uint64_t reserve, uint64_t txg); extern uint64_t arc_free_memory(void); extern int64_t arc_available_memory(void); extern void arc_tuning_update(boolean_t); extern void arc_register_hotplug(void); extern void arc_unregister_hotplug(void); extern int param_set_arc_long(ZFS_MODULE_PARAM_ARGS); extern int param_set_arc_int(ZFS_MODULE_PARAM_ARGS); /* used in zdb.c */ boolean_t l2arc_log_blkptr_valid(l2arc_dev_t *dev, const l2arc_log_blkptr_t *lbp); /* used in vdev_trim.c */ void l2arc_dev_hdr_update(l2arc_dev_t *dev); l2arc_dev_t *l2arc_vdev_get(vdev_t *vd); #ifdef __cplusplus } #endif #endif /* _SYS_ARC_IMPL_H */ diff --git a/sys/contrib/openzfs/lib/libzutil/zutil_import.c b/sys/contrib/openzfs/lib/libzutil/zutil_import.c index 9bd12973fc51..b5b2d7dbed91 100644 --- a/sys/contrib/openzfs/lib/libzutil/zutil_import.c +++ b/sys/contrib/openzfs/lib/libzutil/zutil_import.c @@ -1,1860 +1,1861 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2015 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2018 by Delphix. All rights reserved. * Copyright 2015 RackTop Systems. * Copyright (c) 2016, Intel Corporation. * Copyright (c) 2021, Colm Buckley */ /* * Pool import support functions. * * Used by zpool, ztest, zdb, and zhack to locate importable configs. Since * these commands are expected to run in the global zone, we can assume * that the devices are all readable when called. * * To import a pool, we rely on reading the configuration information from the * ZFS label of each device. If we successfully read the label, then we * organize the configuration information in the following hierarchy: * * pool guid -> toplevel vdev guid -> label txg * * Duplicate entries matching this same tuple will be discarded. Once we have * examined every device, we pick the best label txg config for each toplevel * vdev. We then arrange these toplevel vdevs into a complete pool config, and * update any paths that have changed. Finally, we attempt to import the pool * using our derived config, and record the results. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "zutil_import.h" /*PRINTFLIKE2*/ static void zutil_error_aux(libpc_handle_t *hdl, const char *fmt, ...) { va_list ap; va_start(ap, fmt); (void) vsnprintf(hdl->lpc_desc, sizeof (hdl->lpc_desc), fmt, ap); hdl->lpc_desc_active = B_TRUE; va_end(ap); } static void zutil_verror(libpc_handle_t *hdl, const char *error, const char *fmt, va_list ap) { char action[1024]; (void) vsnprintf(action, sizeof (action), fmt, ap); if (hdl->lpc_desc_active) hdl->lpc_desc_active = B_FALSE; else hdl->lpc_desc[0] = '\0'; if (hdl->lpc_printerr) { if (hdl->lpc_desc[0] != '\0') error = hdl->lpc_desc; (void) fprintf(stderr, "%s: %s\n", action, error); } } /*PRINTFLIKE3*/ static int zutil_error_fmt(libpc_handle_t *hdl, const char *error, const char *fmt, ...) { va_list ap; va_start(ap, fmt); zutil_verror(hdl, error, fmt, ap); va_end(ap); return (-1); } static int zutil_error(libpc_handle_t *hdl, const char *error, const char *msg) { return (zutil_error_fmt(hdl, error, "%s", msg)); } static int zutil_no_memory(libpc_handle_t *hdl) { zutil_error(hdl, EZFS_NOMEM, "internal error"); exit(1); } void * zutil_alloc(libpc_handle_t *hdl, size_t size) { void *data; if ((data = calloc(1, size)) == NULL) (void) zutil_no_memory(hdl); return (data); } char * zutil_strdup(libpc_handle_t *hdl, const char *str) { char *ret; if ((ret = strdup(str)) == NULL) (void) zutil_no_memory(hdl); return (ret); } static char * zutil_strndup(libpc_handle_t *hdl, const char *str, size_t n) { char *ret; if ((ret = strndup(str, n)) == NULL) (void) zutil_no_memory(hdl); return (ret); } /* * Intermediate structures used to gather configuration information. */ typedef struct config_entry { uint64_t ce_txg; nvlist_t *ce_config; struct config_entry *ce_next; } config_entry_t; typedef struct vdev_entry { uint64_t ve_guid; config_entry_t *ve_configs; struct vdev_entry *ve_next; } vdev_entry_t; typedef struct pool_entry { uint64_t pe_guid; vdev_entry_t *pe_vdevs; struct pool_entry *pe_next; } pool_entry_t; typedef struct name_entry { char *ne_name; uint64_t ne_guid; uint64_t ne_order; uint64_t ne_num_labels; struct name_entry *ne_next; } name_entry_t; typedef struct pool_list { pool_entry_t *pools; name_entry_t *names; } pool_list_t; /* * Go through and fix up any path and/or devid information for the given vdev * configuration. */ static int fix_paths(libpc_handle_t *hdl, nvlist_t *nv, name_entry_t *names) { nvlist_t **child; uint_t c, children; uint64_t guid; name_entry_t *ne, *best; char *path; if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, &child, &children) == 0) { for (c = 0; c < children; c++) if (fix_paths(hdl, child[c], names) != 0) return (-1); return (0); } /* * This is a leaf (file or disk) vdev. In either case, go through * the name list and see if we find a matching guid. If so, replace * the path and see if we can calculate a new devid. * * There may be multiple names associated with a particular guid, in * which case we have overlapping partitions or multiple paths to the * same disk. In this case we prefer to use the path name which * matches the ZPOOL_CONFIG_PATH. If no matching entry is found we * use the lowest order device which corresponds to the first match * while traversing the ZPOOL_IMPORT_PATH search path. */ verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0) path = NULL; best = NULL; for (ne = names; ne != NULL; ne = ne->ne_next) { if (ne->ne_guid == guid) { if (path == NULL) { best = ne; break; } if ((strlen(path) == strlen(ne->ne_name)) && strncmp(path, ne->ne_name, strlen(path)) == 0) { best = ne; break; } if (best == NULL) { best = ne; continue; } /* Prefer paths with move vdev labels. */ if (ne->ne_num_labels > best->ne_num_labels) { best = ne; continue; } /* Prefer paths earlier in the search order. */ if (ne->ne_num_labels == best->ne_num_labels && ne->ne_order < best->ne_order) { best = ne; continue; } } } if (best == NULL) return (0); if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0) return (-1); update_vdev_config_dev_strs(nv); return (0); } /* * Add the given configuration to the list of known devices. */ static int add_config(libpc_handle_t *hdl, pool_list_t *pl, const char *path, int order, int num_labels, nvlist_t *config) { uint64_t pool_guid, vdev_guid, top_guid, txg, state; pool_entry_t *pe; vdev_entry_t *ve; config_entry_t *ce; name_entry_t *ne; /* * If this is a hot spare not currently in use or level 2 cache * device, add it to the list of names to translate, but don't do * anything else. */ if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state) == 0 && (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) && nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) { if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL) return (-1); if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) { free(ne); return (-1); } ne->ne_guid = vdev_guid; ne->ne_order = order; ne->ne_num_labels = num_labels; ne->ne_next = pl->names; pl->names = ne; return (0); } /* * If we have a valid config but cannot read any of these fields, then * it means we have a half-initialized label. In vdev_label_init() * we write a label with txg == 0 so that we can identify the device * in case the user refers to the same disk later on. If we fail to * create the pool, we'll be left with a label in this state * which should not be considered part of a valid pool. */ if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid) != 0 || nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) != 0 || nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID, &top_guid) != 0 || nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) != 0 || txg == 0) { return (0); } /* * First, see if we know about this pool. If not, then add it to the * list of known pools. */ for (pe = pl->pools; pe != NULL; pe = pe->pe_next) { if (pe->pe_guid == pool_guid) break; } if (pe == NULL) { if ((pe = zutil_alloc(hdl, sizeof (pool_entry_t))) == NULL) { return (-1); } pe->pe_guid = pool_guid; pe->pe_next = pl->pools; pl->pools = pe; } /* * Second, see if we know about this toplevel vdev. Add it if its * missing. */ for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) { if (ve->ve_guid == top_guid) break; } if (ve == NULL) { if ((ve = zutil_alloc(hdl, sizeof (vdev_entry_t))) == NULL) { return (-1); } ve->ve_guid = top_guid; ve->ve_next = pe->pe_vdevs; pe->pe_vdevs = ve; } /* * Third, see if we have a config with a matching transaction group. If * so, then we do nothing. Otherwise, add it to the list of known * configs. */ for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) { if (ce->ce_txg == txg) break; } if (ce == NULL) { if ((ce = zutil_alloc(hdl, sizeof (config_entry_t))) == NULL) { return (-1); } ce->ce_txg = txg; ce->ce_config = fnvlist_dup(config); ce->ce_next = ve->ve_configs; ve->ve_configs = ce; } /* * At this point we've successfully added our config to the list of * known configs. The last thing to do is add the vdev guid -> path * mappings so that we can fix up the configuration as necessary before * doing the import. */ if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL) return (-1); if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) { free(ne); return (-1); } ne->ne_guid = vdev_guid; ne->ne_order = order; ne->ne_num_labels = num_labels; ne->ne_next = pl->names; pl->names = ne; return (0); } static int zutil_pool_active(libpc_handle_t *hdl, const char *name, uint64_t guid, boolean_t *isactive) { ASSERT(hdl->lpc_ops->pco_pool_active != NULL); int error = hdl->lpc_ops->pco_pool_active(hdl->lpc_lib_handle, name, guid, isactive); return (error); } static nvlist_t * zutil_refresh_config(libpc_handle_t *hdl, nvlist_t *tryconfig) { ASSERT(hdl->lpc_ops->pco_refresh_config != NULL); return (hdl->lpc_ops->pco_refresh_config(hdl->lpc_lib_handle, tryconfig)); } /* * Determine if the vdev id is a hole in the namespace. */ static boolean_t vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id) { int c; for (c = 0; c < holes; c++) { /* Top-level is a hole */ if (hole_array[c] == id) return (B_TRUE); } return (B_FALSE); } /* * Convert our list of pools into the definitive set of configurations. We * start by picking the best config for each toplevel vdev. Once that's done, * we assemble the toplevel vdevs into a full config for the pool. We make a * pass to fix up any incorrect paths, and then add it to the main list to * return to the user. */ static nvlist_t * get_configs(libpc_handle_t *hdl, pool_list_t *pl, boolean_t active_ok, nvlist_t *policy) { pool_entry_t *pe; vdev_entry_t *ve; config_entry_t *ce; nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot; nvlist_t **spares, **l2cache; uint_t i, nspares, nl2cache; boolean_t config_seen; uint64_t best_txg; char *name, *hostname = NULL; uint64_t guid; uint_t children = 0; nvlist_t **child = NULL; uint_t holes; uint64_t *hole_array, max_id; uint_t c; boolean_t isactive; uint64_t hostid; nvlist_t *nvl; boolean_t valid_top_config = B_FALSE; if (nvlist_alloc(&ret, 0, 0) != 0) goto nomem; for (pe = pl->pools; pe != NULL; pe = pe->pe_next) { uint64_t id, max_txg = 0; if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0) goto nomem; config_seen = B_FALSE; /* * Iterate over all toplevel vdevs. Grab the pool configuration * from the first one we find, and then go through the rest and * add them as necessary to the 'vdevs' member of the config. */ for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) { /* * Determine the best configuration for this vdev by * selecting the config with the latest transaction * group. */ best_txg = 0; for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) { if (ce->ce_txg > best_txg) { tmp = ce->ce_config; best_txg = ce->ce_txg; } } /* * We rely on the fact that the max txg for the * pool will contain the most up-to-date information * about the valid top-levels in the vdev namespace. */ if (best_txg > max_txg) { (void) nvlist_remove(config, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64); (void) nvlist_remove(config, ZPOOL_CONFIG_HOLE_ARRAY, DATA_TYPE_UINT64_ARRAY); max_txg = best_txg; hole_array = NULL; holes = 0; max_id = 0; valid_top_config = B_FALSE; if (nvlist_lookup_uint64(tmp, ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) { verify(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, max_id) == 0); valid_top_config = B_TRUE; } if (nvlist_lookup_uint64_array(tmp, ZPOOL_CONFIG_HOLE_ARRAY, &hole_array, &holes) == 0) { verify(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, hole_array, holes) == 0); } } if (!config_seen) { /* * Copy the relevant pieces of data to the pool * configuration: * * version * pool guid * name * comment (if available) * compatibility features (if available) * pool state * hostid (if available) * hostname (if available) */ uint64_t state, version; char *comment = NULL; char *compatibility = NULL; version = fnvlist_lookup_uint64(tmp, ZPOOL_CONFIG_VERSION); fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, version); guid = fnvlist_lookup_uint64(tmp, ZPOOL_CONFIG_POOL_GUID); fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, guid); name = fnvlist_lookup_string(tmp, ZPOOL_CONFIG_POOL_NAME); fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, name); if (nvlist_lookup_string(tmp, ZPOOL_CONFIG_COMMENT, &comment) == 0) fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT, comment); if (nvlist_lookup_string(tmp, ZPOOL_CONFIG_COMPATIBILITY, &compatibility) == 0) fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY, compatibility); state = fnvlist_lookup_uint64(tmp, ZPOOL_CONFIG_POOL_STATE); fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state); hostid = 0; if (nvlist_lookup_uint64(tmp, ZPOOL_CONFIG_HOSTID, &hostid) == 0) { fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid); hostname = fnvlist_lookup_string(tmp, ZPOOL_CONFIG_HOSTNAME); fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, hostname); } config_seen = B_TRUE; } /* * Add this top-level vdev to the child array. */ verify(nvlist_lookup_nvlist(tmp, ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID, &id) == 0); if (id >= children) { nvlist_t **newchild; newchild = zutil_alloc(hdl, (id + 1) * sizeof (nvlist_t *)); if (newchild == NULL) goto nomem; for (c = 0; c < children; c++) newchild[c] = child[c]; free(child); child = newchild; children = id + 1; } if (nvlist_dup(nvtop, &child[id], 0) != 0) goto nomem; } /* * If we have information about all the top-levels then * clean up the nvlist which we've constructed. This * means removing any extraneous devices that are * beyond the valid range or adding devices to the end * of our array which appear to be missing. */ if (valid_top_config) { if (max_id < children) { for (c = max_id; c < children; c++) nvlist_free(child[c]); children = max_id; } else if (max_id > children) { nvlist_t **newchild; newchild = zutil_alloc(hdl, (max_id) * sizeof (nvlist_t *)); if (newchild == NULL) goto nomem; for (c = 0; c < children; c++) newchild[c] = child[c]; free(child); child = newchild; children = max_id; } } verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &guid) == 0); /* * The vdev namespace may contain holes as a result of * device removal. We must add them back into the vdev * tree before we process any missing devices. */ if (holes > 0) { ASSERT(valid_top_config); for (c = 0; c < children; c++) { nvlist_t *holey; if (child[c] != NULL || !vdev_is_hole(hole_array, holes, c)) continue; if (nvlist_alloc(&holey, NV_UNIQUE_NAME, 0) != 0) goto nomem; /* * Holes in the namespace are treated as * "hole" top-level vdevs and have a * special flag set on them. */ if (nvlist_add_string(holey, ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) != 0 || nvlist_add_uint64(holey, ZPOOL_CONFIG_ID, c) != 0 || nvlist_add_uint64(holey, ZPOOL_CONFIG_GUID, 0ULL) != 0) { nvlist_free(holey); goto nomem; } child[c] = holey; } } /* * Look for any missing top-level vdevs. If this is the case, * create a faked up 'missing' vdev as a placeholder. We cannot * simply compress the child array, because the kernel performs * certain checks to make sure the vdev IDs match their location * in the configuration. */ for (c = 0; c < children; c++) { if (child[c] == NULL) { nvlist_t *missing; if (nvlist_alloc(&missing, NV_UNIQUE_NAME, 0) != 0) goto nomem; if (nvlist_add_string(missing, ZPOOL_CONFIG_TYPE, VDEV_TYPE_MISSING) != 0 || nvlist_add_uint64(missing, ZPOOL_CONFIG_ID, c) != 0 || nvlist_add_uint64(missing, ZPOOL_CONFIG_GUID, 0ULL) != 0) { nvlist_free(missing); goto nomem; } child[c] = missing; } } /* * Put all of this pool's top-level vdevs into a root vdev. */ if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0) goto nomem; if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 || nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 || nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 || nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, child, children) != 0) { nvlist_free(nvroot); goto nomem; } for (c = 0; c < children; c++) nvlist_free(child[c]); free(child); children = 0; child = NULL; /* * Go through and fix up any paths and/or devids based on our * known list of vdev GUID -> path mappings. */ if (fix_paths(hdl, nvroot, pl->names) != 0) { nvlist_free(nvroot); goto nomem; } /* * Add the root vdev to this pool's configuration. */ if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) != 0) { nvlist_free(nvroot); goto nomem; } nvlist_free(nvroot); /* * zdb uses this path to report on active pools that were * imported or created using -R. */ if (active_ok) goto add_pool; /* * Determine if this pool is currently active, in which case we * can't actually import it. */ verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, &name) == 0); verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &guid) == 0); if (zutil_pool_active(hdl, name, guid, &isactive) != 0) goto error; if (isactive) { nvlist_free(config); config = NULL; continue; } if (policy != NULL) { if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY, policy) != 0) goto nomem; } if ((nvl = zutil_refresh_config(hdl, config)) == NULL) { nvlist_free(config); config = NULL; continue; } nvlist_free(config); config = nvl; /* * Go through and update the paths for spares, now that we have * them. */ verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) { for (i = 0; i < nspares; i++) { if (fix_paths(hdl, spares[i], pl->names) != 0) goto nomem; } } /* * Update the paths for l2cache devices. */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0) { for (i = 0; i < nl2cache; i++) { if (fix_paths(hdl, l2cache[i], pl->names) != 0) goto nomem; } } /* * Restore the original information read from the actual label. */ (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID, DATA_TYPE_UINT64); (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME, DATA_TYPE_STRING); if (hostid != 0) { verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid) == 0); verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, hostname) == 0); } add_pool: /* * Add this pool to the list of configs. */ verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, &name) == 0); if (nvlist_add_nvlist(ret, name, config) != 0) goto nomem; nvlist_free(config); config = NULL; } return (ret); nomem: (void) zutil_no_memory(hdl); error: nvlist_free(config); nvlist_free(ret); for (c = 0; c < children; c++) nvlist_free(child[c]); free(child); return (NULL); } /* * Return the offset of the given label. */ static uint64_t label_offset(uint64_t size, int l) { ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0); return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 0 : size - VDEV_LABELS * sizeof (vdev_label_t))); } /* * The same description applies as to zpool_read_label below, * except here we do it without aio, presumably because an aio call * errored out in a way we think not using it could circumvent. */ static int zpool_read_label_slow(int fd, nvlist_t **config, int *num_labels) { struct stat64 statbuf; int l, count = 0; vdev_phys_t *label; nvlist_t *expected_config = NULL; uint64_t expected_guid = 0, size; int error; *config = NULL; if (fstat64_blk(fd, &statbuf) == -1) return (0); size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t); error = posix_memalign((void **)&label, PAGESIZE, sizeof (*label)); if (error) return (-1); for (l = 0; l < VDEV_LABELS; l++) { uint64_t state, guid, txg; off_t offset = label_offset(size, l) + VDEV_SKIP_SIZE; if (pread64(fd, label, sizeof (vdev_phys_t), offset) != sizeof (vdev_phys_t)) continue; if (nvlist_unpack(label->vp_nvlist, sizeof (label->vp_nvlist), config, 0) != 0) continue; if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID, &guid) != 0 || guid == 0) { nvlist_free(*config); continue; } if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE, &state) != 0 || state > POOL_STATE_L2CACHE) { nvlist_free(*config); continue; } if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG, &txg) != 0 || txg == 0)) { nvlist_free(*config); continue; } if (expected_guid) { if (expected_guid == guid) count++; nvlist_free(*config); } else { expected_config = *config; expected_guid = guid; count++; } } if (num_labels != NULL) *num_labels = count; free(label); *config = expected_config; return (0); } /* * Given a file descriptor, read the label information and return an nvlist * describing the configuration, if there is one. The number of valid * labels found will be returned in num_labels when non-NULL. */ int zpool_read_label(int fd, nvlist_t **config, int *num_labels) { struct stat64 statbuf; struct aiocb aiocbs[VDEV_LABELS]; struct aiocb *aiocbps[VDEV_LABELS]; vdev_phys_t *labels; nvlist_t *expected_config = NULL; uint64_t expected_guid = 0, size; int error, l, count = 0; *config = NULL; if (fstat64_blk(fd, &statbuf) == -1) return (0); size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t); error = posix_memalign((void **)&labels, PAGESIZE, VDEV_LABELS * sizeof (*labels)); if (error) return (-1); memset(aiocbs, 0, sizeof (aiocbs)); for (l = 0; l < VDEV_LABELS; l++) { off_t offset = label_offset(size, l) + VDEV_SKIP_SIZE; aiocbs[l].aio_fildes = fd; aiocbs[l].aio_offset = offset; aiocbs[l].aio_buf = &labels[l]; aiocbs[l].aio_nbytes = sizeof (vdev_phys_t); aiocbs[l].aio_lio_opcode = LIO_READ; aiocbps[l] = &aiocbs[l]; } if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) { int saved_errno = errno; boolean_t do_slow = B_FALSE; error = -1; if (errno == EAGAIN || errno == EINTR || errno == EIO) { /* * A portion of the requests may have been submitted. * Clean them up. */ for (l = 0; l < VDEV_LABELS; l++) { errno = 0; switch (aio_error(&aiocbs[l])) { case EINVAL: break; case EINPROGRESS: // This shouldn't be possible to // encounter, die if we do. ASSERT(B_FALSE); case EOPNOTSUPP: case ENOSYS: do_slow = B_TRUE; case 0: default: (void) aio_return(&aiocbs[l]); } } } if (do_slow) { /* * At least some IO involved access unsafe-for-AIO * files. Let's try again, without AIO this time. */ error = zpool_read_label_slow(fd, config, num_labels); saved_errno = errno; } free(labels); errno = saved_errno; return (error); } for (l = 0; l < VDEV_LABELS; l++) { uint64_t state, guid, txg; if (aio_return(&aiocbs[l]) != sizeof (vdev_phys_t)) continue; if (nvlist_unpack(labels[l].vp_nvlist, sizeof (labels[l].vp_nvlist), config, 0) != 0) continue; if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID, &guid) != 0 || guid == 0) { nvlist_free(*config); continue; } if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE, &state) != 0 || state > POOL_STATE_L2CACHE) { nvlist_free(*config); continue; } if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG, &txg) != 0 || txg == 0)) { nvlist_free(*config); continue; } if (expected_guid) { if (expected_guid == guid) count++; nvlist_free(*config); } else { expected_config = *config; expected_guid = guid; count++; } } if (num_labels != NULL) *num_labels = count; free(labels); *config = expected_config; return (0); } /* * Sorted by full path and then vdev guid to allow for multiple entries with * the same full path name. This is required because it's possible to * have multiple block devices with labels that refer to the same * ZPOOL_CONFIG_PATH yet have different vdev guids. In this case both * entries need to be added to the cache. Scenarios where this can occur * include overwritten pool labels, devices which are visible from multiple * hosts and multipath devices. */ int slice_cache_compare(const void *arg1, const void *arg2) { const char *nm1 = ((rdsk_node_t *)arg1)->rn_name; const char *nm2 = ((rdsk_node_t *)arg2)->rn_name; uint64_t guid1 = ((rdsk_node_t *)arg1)->rn_vdev_guid; uint64_t guid2 = ((rdsk_node_t *)arg2)->rn_vdev_guid; int rv; rv = TREE_ISIGN(strcmp(nm1, nm2)); if (rv) return (rv); return (TREE_CMP(guid1, guid2)); } static int label_paths_impl(libpc_handle_t *hdl, nvlist_t *nvroot, uint64_t pool_guid, uint64_t vdev_guid, char **path, char **devid) { nvlist_t **child; uint_t c, children; uint64_t guid; char *val; int error; if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &child, &children) == 0) { for (c = 0; c < children; c++) { error = label_paths_impl(hdl, child[c], pool_guid, vdev_guid, path, devid); if (error) return (error); } return (0); } if (nvroot == NULL) return (0); error = nvlist_lookup_uint64(nvroot, ZPOOL_CONFIG_GUID, &guid); if ((error != 0) || (guid != vdev_guid)) return (0); error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_PATH, &val); if (error == 0) *path = val; error = nvlist_lookup_string(nvroot, ZPOOL_CONFIG_DEVID, &val); if (error == 0) *devid = val; return (0); } /* * Given a disk label fetch the ZPOOL_CONFIG_PATH and ZPOOL_CONFIG_DEVID * and store these strings as config_path and devid_path respectively. * The returned pointers are only valid as long as label remains valid. */ int label_paths(libpc_handle_t *hdl, nvlist_t *label, char **path, char **devid) { nvlist_t *nvroot; uint64_t pool_guid; uint64_t vdev_guid; *path = NULL; *devid = NULL; if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvroot) || nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &pool_guid) || nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &vdev_guid)) return (ENOENT); return (label_paths_impl(hdl, nvroot, pool_guid, vdev_guid, path, devid)); } static void zpool_find_import_scan_add_slice(libpc_handle_t *hdl, pthread_mutex_t *lock, avl_tree_t *cache, const char *path, const char *name, int order) { avl_index_t where; rdsk_node_t *slice; slice = zutil_alloc(hdl, sizeof (rdsk_node_t)); if (asprintf(&slice->rn_name, "%s/%s", path, name) == -1) { free(slice); return; } slice->rn_vdev_guid = 0; slice->rn_lock = lock; slice->rn_avl = cache; slice->rn_hdl = hdl; slice->rn_order = order + IMPORT_ORDER_SCAN_OFFSET; slice->rn_labelpaths = B_FALSE; pthread_mutex_lock(lock); if (avl_find(cache, slice, &where)) { free(slice->rn_name); free(slice); } else { avl_insert(cache, slice, where); } pthread_mutex_unlock(lock); } static int zpool_find_import_scan_dir(libpc_handle_t *hdl, pthread_mutex_t *lock, avl_tree_t *cache, const char *dir, int order) { int error; char path[MAXPATHLEN]; struct dirent64 *dp; DIR *dirp; if (realpath(dir, path) == NULL) { error = errno; if (error == ENOENT) return (0); zutil_error_aux(hdl, strerror(error)); (void) zutil_error_fmt(hdl, EZFS_BADPATH, dgettext( TEXT_DOMAIN, "cannot resolve path '%s'"), dir); return (error); } dirp = opendir(path); if (dirp == NULL) { error = errno; zutil_error_aux(hdl, strerror(error)); (void) zutil_error_fmt(hdl, EZFS_BADPATH, dgettext(TEXT_DOMAIN, "cannot open '%s'"), path); return (error); } while ((dp = readdir64(dirp)) != NULL) { const char *name = dp->d_name; if (strcmp(name, ".") == 0 || strcmp(name, "..") == 0) continue; switch (dp->d_type) { case DT_UNKNOWN: case DT_BLK: + case DT_LNK: #ifdef __FreeBSD__ case DT_CHR: #endif case DT_REG: break; default: continue; } zpool_find_import_scan_add_slice(hdl, lock, cache, path, name, order); } (void) closedir(dirp); return (0); } static int zpool_find_import_scan_path(libpc_handle_t *hdl, pthread_mutex_t *lock, avl_tree_t *cache, const char *dir, int order) { int error = 0; char path[MAXPATHLEN]; char *d = NULL; ssize_t dl; const char *dpath, *name; /* * Separate the directory and the basename. * We do this so that we can get the realpath of * the directory. We don't get the realpath on the * whole path because if it's a symlink, we want the * path of the symlink not where it points to. */ name = zfs_basename(dir); if ((dl = zfs_dirnamelen(dir)) == -1) dpath = "."; else dpath = d = zutil_strndup(hdl, dir, dl); if (realpath(dpath, path) == NULL) { error = errno; if (error == ENOENT) { error = 0; goto out; } zutil_error_aux(hdl, strerror(error)); (void) zutil_error_fmt(hdl, EZFS_BADPATH, dgettext( TEXT_DOMAIN, "cannot resolve path '%s'"), dir); goto out; } zpool_find_import_scan_add_slice(hdl, lock, cache, path, name, order); out: free(d); return (error); } /* * Scan a list of directories for zfs devices. */ static int zpool_find_import_scan(libpc_handle_t *hdl, pthread_mutex_t *lock, avl_tree_t **slice_cache, const char * const *dir, size_t dirs) { avl_tree_t *cache; rdsk_node_t *slice; void *cookie; int i, error; *slice_cache = NULL; cache = zutil_alloc(hdl, sizeof (avl_tree_t)); avl_create(cache, slice_cache_compare, sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node)); for (i = 0; i < dirs; i++) { struct stat sbuf; if (stat(dir[i], &sbuf) != 0) { error = errno; if (error == ENOENT) continue; zutil_error_aux(hdl, strerror(error)); (void) zutil_error_fmt(hdl, EZFS_BADPATH, dgettext( TEXT_DOMAIN, "cannot resolve path '%s'"), dir[i]); goto error; } /* * If dir[i] is a directory, we walk through it and add all * the entries to the cache. If it's not a directory, we just * add it to the cache. */ if (S_ISDIR(sbuf.st_mode)) { if ((error = zpool_find_import_scan_dir(hdl, lock, cache, dir[i], i)) != 0) goto error; } else { if ((error = zpool_find_import_scan_path(hdl, lock, cache, dir[i], i)) != 0) goto error; } } *slice_cache = cache; return (0); error: cookie = NULL; while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) { free(slice->rn_name); free(slice); } free(cache); return (error); } /* * Given a list of directories to search, find all pools stored on disk. This * includes partial pools which are not available to import. If no args are * given (argc is 0), then the default directory (/dev/dsk) is searched. * poolname or guid (but not both) are provided by the caller when trying * to import a specific pool. */ static nvlist_t * zpool_find_import_impl(libpc_handle_t *hdl, importargs_t *iarg, pthread_mutex_t *lock, avl_tree_t *cache) { nvlist_t *ret = NULL; pool_list_t pools = { 0 }; pool_entry_t *pe, *penext; vdev_entry_t *ve, *venext; config_entry_t *ce, *cenext; name_entry_t *ne, *nenext; rdsk_node_t *slice; void *cookie; tpool_t *t; verify(iarg->poolname == NULL || iarg->guid == 0); /* * Create a thread pool to parallelize the process of reading and * validating labels, a large number of threads can be used due to * minimal contention. */ t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN), 0, NULL); for (slice = avl_first(cache); slice; (slice = avl_walk(cache, slice, AVL_AFTER))) (void) tpool_dispatch(t, zpool_open_func, slice); tpool_wait(t); tpool_destroy(t); /* * Process the cache, filtering out any entries which are not * for the specified pool then adding matching label configs. */ cookie = NULL; while ((slice = avl_destroy_nodes(cache, &cookie)) != NULL) { if (slice->rn_config != NULL) { nvlist_t *config = slice->rn_config; boolean_t matched = B_TRUE; boolean_t aux = B_FALSE; int fd; /* * Check if it's a spare or l2cache device. If it is, * we need to skip the name and guid check since they * don't exist on aux device label. */ if (iarg->poolname != NULL || iarg->guid != 0) { uint64_t state; aux = nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state) == 0 && (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE); } if (iarg->poolname != NULL && !aux) { char *pname; matched = nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, &pname) == 0 && strcmp(iarg->poolname, pname) == 0; } else if (iarg->guid != 0 && !aux) { uint64_t this_guid; matched = nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &this_guid) == 0 && iarg->guid == this_guid; } if (matched) { /* * Verify all remaining entries can be opened * exclusively. This will prune all underlying * multipath devices which otherwise could * result in the vdev appearing as UNAVAIL. * * Under zdb, this step isn't required and * would prevent a zdb -e of active pools with * no cachefile. */ fd = open(slice->rn_name, O_RDONLY | O_EXCL | O_CLOEXEC); if (fd >= 0 || iarg->can_be_active) { if (fd >= 0) close(fd); add_config(hdl, &pools, slice->rn_name, slice->rn_order, slice->rn_num_labels, config); } } nvlist_free(config); } free(slice->rn_name); free(slice); } avl_destroy(cache); free(cache); ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy); for (pe = pools.pools; pe != NULL; pe = penext) { penext = pe->pe_next; for (ve = pe->pe_vdevs; ve != NULL; ve = venext) { venext = ve->ve_next; for (ce = ve->ve_configs; ce != NULL; ce = cenext) { cenext = ce->ce_next; nvlist_free(ce->ce_config); free(ce); } free(ve); } free(pe); } for (ne = pools.names; ne != NULL; ne = nenext) { nenext = ne->ne_next; free(ne->ne_name); free(ne); } return (ret); } /* * Given a config, discover the paths for the devices which * exist in the config. */ static int discover_cached_paths(libpc_handle_t *hdl, nvlist_t *nv, avl_tree_t *cache, pthread_mutex_t *lock) { char *path = NULL; ssize_t dl; uint_t children; nvlist_t **child; if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, &child, &children) == 0) { for (int c = 0; c < children; c++) { discover_cached_paths(hdl, child[c], cache, lock); } } /* * Once we have the path, we need to add the directory to * our directory cache. */ if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) { if ((dl = zfs_dirnamelen(path)) == -1) path = "."; else path[dl] = '\0'; return (zpool_find_import_scan_dir(hdl, lock, cache, path, 0)); } return (0); } /* * Given a cache file, return the contents as a list of importable pools. * poolname or guid (but not both) are provided by the caller when trying * to import a specific pool. */ static nvlist_t * zpool_find_import_cached(libpc_handle_t *hdl, importargs_t *iarg) { char *buf; int fd; struct stat64 statbuf; nvlist_t *raw, *src, *dst; nvlist_t *pools; nvpair_t *elem; char *name; uint64_t this_guid; boolean_t active; verify(iarg->poolname == NULL || iarg->guid == 0); if ((fd = open(iarg->cachefile, O_RDONLY | O_CLOEXEC)) < 0) { zutil_error_aux(hdl, "%s", strerror(errno)); (void) zutil_error(hdl, EZFS_BADCACHE, dgettext(TEXT_DOMAIN, "failed to open cache file")); return (NULL); } if (fstat64(fd, &statbuf) != 0) { zutil_error_aux(hdl, "%s", strerror(errno)); (void) close(fd); (void) zutil_error(hdl, EZFS_BADCACHE, dgettext(TEXT_DOMAIN, "failed to get size of cache file")); return (NULL); } if ((buf = zutil_alloc(hdl, statbuf.st_size)) == NULL) { (void) close(fd); return (NULL); } if (read(fd, buf, statbuf.st_size) != statbuf.st_size) { (void) close(fd); free(buf); (void) zutil_error(hdl, EZFS_BADCACHE, dgettext(TEXT_DOMAIN, "failed to read cache file contents")); return (NULL); } (void) close(fd); if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) { free(buf); (void) zutil_error(hdl, EZFS_BADCACHE, dgettext(TEXT_DOMAIN, "invalid or corrupt cache file contents")); return (NULL); } free(buf); /* * Go through and get the current state of the pools and refresh their * state. */ if (nvlist_alloc(&pools, 0, 0) != 0) { (void) zutil_no_memory(hdl); nvlist_free(raw); return (NULL); } elem = NULL; while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) { src = fnvpair_value_nvlist(elem); name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME); if (iarg->poolname != NULL && strcmp(iarg->poolname, name) != 0) continue; this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID); if (iarg->guid != 0 && iarg->guid != this_guid) continue; if (zutil_pool_active(hdl, name, this_guid, &active) != 0) { nvlist_free(raw); nvlist_free(pools); return (NULL); } if (active) continue; if (iarg->scan) { uint64_t saved_guid = iarg->guid; const char *saved_poolname = iarg->poolname; pthread_mutex_t lock; /* * Create the device cache that will hold the * devices we will scan based on the cachefile. * This will get destroyed and freed by * zpool_find_import_impl. */ avl_tree_t *cache = zutil_alloc(hdl, sizeof (avl_tree_t)); avl_create(cache, slice_cache_compare, sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node)); nvlist_t *nvroot = fnvlist_lookup_nvlist(src, ZPOOL_CONFIG_VDEV_TREE); /* * We only want to find the pool with this_guid. * We will reset these values back later. */ iarg->guid = this_guid; iarg->poolname = NULL; /* * We need to build up a cache of devices that exists * in the paths pointed to by the cachefile. This allows * us to preserve the device namespace that was * originally specified by the user but also lets us * scan devices in those directories in case they had * been renamed. */ pthread_mutex_init(&lock, NULL); discover_cached_paths(hdl, nvroot, cache, &lock); nvlist_t *nv = zpool_find_import_impl(hdl, iarg, &lock, cache); pthread_mutex_destroy(&lock); /* * zpool_find_import_impl will return back * a list of pools that it found based on the * device cache. There should only be one pool * since we're looking for a specific guid. * We will use that pool to build up the final * pool nvlist which is returned back to the * caller. */ nvpair_t *pair = nvlist_next_nvpair(nv, NULL); fnvlist_add_nvlist(pools, nvpair_name(pair), fnvpair_value_nvlist(pair)); VERIFY3P(nvlist_next_nvpair(nv, pair), ==, NULL); iarg->guid = saved_guid; iarg->poolname = saved_poolname; continue; } if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE, iarg->cachefile) != 0) { (void) zutil_no_memory(hdl); nvlist_free(raw); nvlist_free(pools); return (NULL); } if ((dst = zutil_refresh_config(hdl, src)) == NULL) { nvlist_free(raw); nvlist_free(pools); return (NULL); } if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) { (void) zutil_no_memory(hdl); nvlist_free(dst); nvlist_free(raw); nvlist_free(pools); return (NULL); } nvlist_free(dst); } nvlist_free(raw); return (pools); } static nvlist_t * zpool_find_import(libpc_handle_t *hdl, importargs_t *iarg) { pthread_mutex_t lock; avl_tree_t *cache; nvlist_t *pools = NULL; verify(iarg->poolname == NULL || iarg->guid == 0); pthread_mutex_init(&lock, NULL); /* * Locate pool member vdevs by blkid or by directory scanning. * On success a newly allocated AVL tree which is populated with an * entry for each discovered vdev will be returned in the cache. * It's the caller's responsibility to consume and destroy this tree. */ if (iarg->scan || iarg->paths != 0) { size_t dirs = iarg->paths; const char * const *dir = (const char * const *)iarg->path; if (dirs == 0) dir = zpool_default_search_paths(&dirs); if (zpool_find_import_scan(hdl, &lock, &cache, dir, dirs) != 0) { pthread_mutex_destroy(&lock); return (NULL); } } else { if (zpool_find_import_blkid(hdl, &lock, &cache) != 0) { pthread_mutex_destroy(&lock); return (NULL); } } pools = zpool_find_import_impl(hdl, iarg, &lock, cache); pthread_mutex_destroy(&lock); return (pools); } nvlist_t * zpool_search_import(void *hdl, importargs_t *import, const pool_config_ops_t *pco) { libpc_handle_t handle = { 0 }; nvlist_t *pools = NULL; handle.lpc_lib_handle = hdl; handle.lpc_ops = pco; handle.lpc_printerr = B_TRUE; verify(import->poolname == NULL || import->guid == 0); if (import->cachefile != NULL) pools = zpool_find_import_cached(&handle, import); else pools = zpool_find_import(&handle, import); if ((pools == NULL || nvlist_empty(pools)) && handle.lpc_open_access_error && geteuid() != 0) { (void) zutil_error(&handle, EZFS_EACESS, dgettext(TEXT_DOMAIN, "no pools found")); } return (pools); } static boolean_t pool_match(nvlist_t *cfg, char *tgt) { uint64_t v, guid = strtoull(tgt, NULL, 0); char *s; if (guid != 0) { if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0) return (v == guid); } else { if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0) return (strcmp(s, tgt) == 0); } return (B_FALSE); } int zpool_find_config(void *hdl, const char *target, nvlist_t **configp, importargs_t *args, const pool_config_ops_t *pco) { nvlist_t *pools; nvlist_t *match = NULL; nvlist_t *config = NULL; char *sepp = NULL; int count = 0; char *targetdup = strdup(target); *configp = NULL; if ((sepp = strpbrk(targetdup, "/@")) != NULL) *sepp = '\0'; pools = zpool_search_import(hdl, args, pco); if (pools != NULL) { nvpair_t *elem = NULL; while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) { VERIFY0(nvpair_value_nvlist(elem, &config)); if (pool_match(config, targetdup)) { count++; if (match != NULL) { /* multiple matches found */ continue; } else { match = fnvlist_dup(config); } } } fnvlist_free(pools); } if (count == 0) { free(targetdup); return (ENOENT); } if (count > 1) { free(targetdup); fnvlist_free(match); return (EINVAL); } *configp = match; free(targetdup); return (0); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/abd_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/abd_os.c index 00aa21c6c9ec..47adc2278df2 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/abd_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/abd_os.c @@ -1,486 +1,532 @@ /* * This file and its contents are supplied under the terms of the * Common Development and Distribution License ("CDDL"), version 1.0. * You may only use this file in accordance with the terms of version * 1.0 of the CDDL. * * A full copy of the text of the CDDL should have accompanied this * source. A copy of the CDDL is also available via the Internet at * http://www.illumos.org/license/CDDL. */ /* * Copyright (c) 2014 by Chunwei Chen. All rights reserved. * Copyright (c) 2016 by Delphix. All rights reserved. */ /* * See abd.c for a general overview of the arc buffered data (ABD). * * Using a large proportion of scattered ABDs decreases ARC fragmentation since * when we are at the limit of allocatable space, using equal-size chunks will * allow us to quickly reclaim enough space for a new large allocation (assuming * it is also scattered). * * ABDs are allocated scattered by default unless the caller uses * abd_alloc_linear() or zfs_abd_scatter_enabled is disabled. */ #include #include #include #include #include #include typedef struct abd_stats { kstat_named_t abdstat_struct_size; kstat_named_t abdstat_scatter_cnt; kstat_named_t abdstat_scatter_data_size; kstat_named_t abdstat_scatter_chunk_waste; kstat_named_t abdstat_linear_cnt; kstat_named_t abdstat_linear_data_size; } abd_stats_t; static abd_stats_t abd_stats = { /* Amount of memory occupied by all of the abd_t struct allocations */ { "struct_size", KSTAT_DATA_UINT64 }, /* * The number of scatter ABDs which are currently allocated, excluding * ABDs which don't own their data (for instance the ones which were * allocated through abd_get_offset()). */ { "scatter_cnt", KSTAT_DATA_UINT64 }, /* Amount of data stored in all scatter ABDs tracked by scatter_cnt */ { "scatter_data_size", KSTAT_DATA_UINT64 }, /* * The amount of space wasted at the end of the last chunk across all * scatter ABDs tracked by scatter_cnt. */ { "scatter_chunk_waste", KSTAT_DATA_UINT64 }, /* * The number of linear ABDs which are currently allocated, excluding * ABDs which don't own their data (for instance the ones which were * allocated through abd_get_offset() and abd_get_from_buf()). If an * ABD takes ownership of its buf then it will become tracked. */ { "linear_cnt", KSTAT_DATA_UINT64 }, /* Amount of data stored in all linear ABDs tracked by linear_cnt */ { "linear_data_size", KSTAT_DATA_UINT64 }, }; +struct { + wmsum_t abdstat_struct_size; + wmsum_t abdstat_scatter_cnt; + wmsum_t abdstat_scatter_data_size; + wmsum_t abdstat_scatter_chunk_waste; + wmsum_t abdstat_linear_cnt; + wmsum_t abdstat_linear_data_size; +} abd_sums; + /* * The size of the chunks ABD allocates. Because the sizes allocated from the * kmem_cache can't change, this tunable can only be modified at boot. Changing * it at runtime would cause ABD iteration to work incorrectly for ABDs which * were allocated with the old size, so a safeguard has been put in place which * will cause the machine to panic if you change it and try to access the data * within a scattered ABD. */ size_t zfs_abd_chunk_size = 4096; #if defined(_KERNEL) SYSCTL_DECL(_vfs_zfs); SYSCTL_INT(_vfs_zfs, OID_AUTO, abd_scatter_enabled, CTLFLAG_RWTUN, &zfs_abd_scatter_enabled, 0, "Enable scattered ARC data buffers"); SYSCTL_ULONG(_vfs_zfs, OID_AUTO, abd_chunk_size, CTLFLAG_RDTUN, &zfs_abd_chunk_size, 0, "The size of the chunks ABD allocates"); #endif kmem_cache_t *abd_chunk_cache; static kstat_t *abd_ksp; /* * We use a scattered SPA_MAXBLOCKSIZE sized ABD whose chunks are * just a single zero'd sized zfs_abd_chunk_size buffer. This * allows us to conserve memory by only using a single zero buffer * for the scatter chunks. */ abd_t *abd_zero_scatter = NULL; static char *abd_zero_buf = NULL; static void abd_free_chunk(void *c) { kmem_cache_free(abd_chunk_cache, c); } static uint_t abd_chunkcnt_for_bytes(size_t size) { return (P2ROUNDUP(size, zfs_abd_chunk_size) / zfs_abd_chunk_size); } static inline uint_t abd_scatter_chunkcnt(abd_t *abd) { ASSERT(!abd_is_linear(abd)); return (abd_chunkcnt_for_bytes( ABD_SCATTER(abd).abd_offset + abd->abd_size)); } boolean_t abd_size_alloc_linear(size_t size) { return (size <= zfs_abd_chunk_size ? B_TRUE : B_FALSE); } void abd_update_scatter_stats(abd_t *abd, abd_stats_op_t op) { uint_t n = abd_scatter_chunkcnt(abd); ASSERT(op == ABDSTAT_INCR || op == ABDSTAT_DECR); int waste = n * zfs_abd_chunk_size - abd->abd_size; if (op == ABDSTAT_INCR) { ABDSTAT_BUMP(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, abd->abd_size); ABDSTAT_INCR(abdstat_scatter_chunk_waste, waste); arc_space_consume(waste, ARC_SPACE_ABD_CHUNK_WASTE); } else { ABDSTAT_BUMPDOWN(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, -(int)abd->abd_size); ABDSTAT_INCR(abdstat_scatter_chunk_waste, -waste); arc_space_return(waste, ARC_SPACE_ABD_CHUNK_WASTE); } } void abd_update_linear_stats(abd_t *abd, abd_stats_op_t op) { ASSERT(op == ABDSTAT_INCR || op == ABDSTAT_DECR); if (op == ABDSTAT_INCR) { ABDSTAT_BUMP(abdstat_linear_cnt); ABDSTAT_INCR(abdstat_linear_data_size, abd->abd_size); } else { ABDSTAT_BUMPDOWN(abdstat_linear_cnt); ABDSTAT_INCR(abdstat_linear_data_size, -(int)abd->abd_size); } } void abd_verify_scatter(abd_t *abd) { uint_t i, n; /* * There is no scatter linear pages in FreeBSD so there is an * if an error if the ABD has been marked as a linear page. */ ASSERT(!abd_is_linear_page(abd)); ASSERT3U(ABD_SCATTER(abd).abd_offset, <, zfs_abd_chunk_size); n = abd_scatter_chunkcnt(abd); for (i = 0; i < n; i++) { ASSERT3P(ABD_SCATTER(abd).abd_chunks[i], !=, NULL); } } void abd_alloc_chunks(abd_t *abd, size_t size) { uint_t i, n; n = abd_chunkcnt_for_bytes(size); for (i = 0; i < n; i++) { void *c = kmem_cache_alloc(abd_chunk_cache, KM_PUSHPAGE); ASSERT3P(c, !=, NULL); ABD_SCATTER(abd).abd_chunks[i] = c; } ABD_SCATTER(abd).abd_chunk_size = zfs_abd_chunk_size; } void abd_free_chunks(abd_t *abd) { uint_t i, n; n = abd_scatter_chunkcnt(abd); for (i = 0; i < n; i++) { abd_free_chunk(ABD_SCATTER(abd).abd_chunks[i]); } } abd_t * abd_alloc_struct_impl(size_t size) { uint_t chunkcnt = abd_chunkcnt_for_bytes(size); /* * In the event we are allocating a gang ABD, the size passed in * will be 0. We must make sure to set abd_size to the size of an * ABD struct as opposed to an ABD scatter with 0 chunks. The gang * ABD struct allocation accounts for an additional 24 bytes over * a scatter ABD with 0 chunks. */ size_t abd_size = MAX(sizeof (abd_t), offsetof(abd_t, abd_u.abd_scatter.abd_chunks[chunkcnt])); abd_t *abd = kmem_alloc(abd_size, KM_PUSHPAGE); ASSERT3P(abd, !=, NULL); ABDSTAT_INCR(abdstat_struct_size, abd_size); return (abd); } void abd_free_struct_impl(abd_t *abd) { uint_t chunkcnt = abd_is_linear(abd) || abd_is_gang(abd) ? 0 : abd_scatter_chunkcnt(abd); ssize_t size = MAX(sizeof (abd_t), offsetof(abd_t, abd_u.abd_scatter.abd_chunks[chunkcnt])); kmem_free(abd, size); ABDSTAT_INCR(abdstat_struct_size, -size); } /* * Allocate scatter ABD of size SPA_MAXBLOCKSIZE, where * each chunk in the scatterlist will be set to abd_zero_buf. */ static void abd_alloc_zero_scatter(void) { uint_t i, n; n = abd_chunkcnt_for_bytes(SPA_MAXBLOCKSIZE); abd_zero_buf = kmem_zalloc(zfs_abd_chunk_size, KM_SLEEP); abd_zero_scatter = abd_alloc_struct(SPA_MAXBLOCKSIZE); abd_zero_scatter->abd_flags |= ABD_FLAG_OWNER | ABD_FLAG_ZEROS; abd_zero_scatter->abd_size = SPA_MAXBLOCKSIZE; ABD_SCATTER(abd_zero_scatter).abd_offset = 0; ABD_SCATTER(abd_zero_scatter).abd_chunk_size = zfs_abd_chunk_size; for (i = 0; i < n; i++) { ABD_SCATTER(abd_zero_scatter).abd_chunks[i] = abd_zero_buf; } ABDSTAT_BUMP(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, zfs_abd_chunk_size); } static void abd_free_zero_scatter(void) { ABDSTAT_BUMPDOWN(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, -(int)zfs_abd_chunk_size); abd_free_struct(abd_zero_scatter); abd_zero_scatter = NULL; kmem_free(abd_zero_buf, zfs_abd_chunk_size); } +static int +abd_kstats_update(kstat_t *ksp, int rw) +{ + abd_stats_t *as = ksp->ks_data; + + if (rw == KSTAT_WRITE) + return (EACCES); + as->abdstat_struct_size.value.ui64 = + wmsum_value(&abd_sums.abdstat_struct_size); + as->abdstat_scatter_cnt.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_cnt); + as->abdstat_scatter_data_size.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_data_size); + as->abdstat_scatter_chunk_waste.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_chunk_waste); + as->abdstat_linear_cnt.value.ui64 = + wmsum_value(&abd_sums.abdstat_linear_cnt); + as->abdstat_linear_data_size.value.ui64 = + wmsum_value(&abd_sums.abdstat_linear_data_size); + return (0); +} + void abd_init(void) { abd_chunk_cache = kmem_cache_create("abd_chunk", zfs_abd_chunk_size, 0, NULL, NULL, NULL, NULL, 0, KMC_NODEBUG); + wmsum_init(&abd_sums.abdstat_struct_size, 0); + wmsum_init(&abd_sums.abdstat_scatter_cnt, 0); + wmsum_init(&abd_sums.abdstat_scatter_data_size, 0); + wmsum_init(&abd_sums.abdstat_scatter_chunk_waste, 0); + wmsum_init(&abd_sums.abdstat_linear_cnt, 0); + wmsum_init(&abd_sums.abdstat_linear_data_size, 0); + abd_ksp = kstat_create("zfs", 0, "abdstats", "misc", KSTAT_TYPE_NAMED, sizeof (abd_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (abd_ksp != NULL) { abd_ksp->ks_data = &abd_stats; + abd_ksp->ks_update = abd_kstats_update; kstat_install(abd_ksp); } abd_alloc_zero_scatter(); } void abd_fini(void) { abd_free_zero_scatter(); if (abd_ksp != NULL) { kstat_delete(abd_ksp); abd_ksp = NULL; } + wmsum_fini(&abd_sums.abdstat_struct_size); + wmsum_fini(&abd_sums.abdstat_scatter_cnt); + wmsum_fini(&abd_sums.abdstat_scatter_data_size); + wmsum_fini(&abd_sums.abdstat_scatter_chunk_waste); + wmsum_fini(&abd_sums.abdstat_linear_cnt); + wmsum_fini(&abd_sums.abdstat_linear_data_size); + kmem_cache_destroy(abd_chunk_cache); abd_chunk_cache = NULL; } void abd_free_linear_page(abd_t *abd) { /* * FreeBSD does not have scatter linear pages * so there is an error. */ VERIFY(0); } /* * If we're going to use this ABD for doing I/O using the block layer, the * consumer of the ABD data doesn't care if it's scattered or not, and we don't * plan to store this ABD in memory for a long period of time, we should * allocate the ABD type that requires the least data copying to do the I/O. * * Currently this is linear ABDs, however if ldi_strategy() can ever issue I/Os * using a scatter/gather list we should switch to that and replace this call * with vanilla abd_alloc(). */ abd_t * abd_alloc_for_io(size_t size, boolean_t is_metadata) { return (abd_alloc_linear(size, is_metadata)); } abd_t * abd_get_offset_scatter(abd_t *abd, abd_t *sabd, size_t off) { abd_verify(sabd); ASSERT3U(off, <=, sabd->abd_size); size_t new_offset = ABD_SCATTER(sabd).abd_offset + off; uint_t chunkcnt = abd_scatter_chunkcnt(sabd) - (new_offset / zfs_abd_chunk_size); /* * If an abd struct is provided, it is only the minimum size. If we * need additional chunks, we need to allocate a new struct. */ if (abd != NULL && offsetof(abd_t, abd_u.abd_scatter.abd_chunks[chunkcnt]) > sizeof (abd_t)) { abd = NULL; } if (abd == NULL) abd = abd_alloc_struct(chunkcnt * zfs_abd_chunk_size); /* * Even if this buf is filesystem metadata, we only track that * if we own the underlying data buffer, which is not true in * this case. Therefore, we don't ever use ABD_FLAG_META here. */ ABD_SCATTER(abd).abd_offset = new_offset % zfs_abd_chunk_size; ABD_SCATTER(abd).abd_chunk_size = zfs_abd_chunk_size; /* Copy the scatterlist starting at the correct offset */ (void) memcpy(&ABD_SCATTER(abd).abd_chunks, &ABD_SCATTER(sabd).abd_chunks[new_offset / zfs_abd_chunk_size], chunkcnt * sizeof (void *)); return (abd); } static inline size_t abd_iter_scatter_chunk_offset(struct abd_iter *aiter) { ASSERT(!abd_is_linear(aiter->iter_abd)); return ((ABD_SCATTER(aiter->iter_abd).abd_offset + aiter->iter_pos) % zfs_abd_chunk_size); } static inline size_t abd_iter_scatter_chunk_index(struct abd_iter *aiter) { ASSERT(!abd_is_linear(aiter->iter_abd)); return ((ABD_SCATTER(aiter->iter_abd).abd_offset + aiter->iter_pos) / zfs_abd_chunk_size); } /* * Initialize the abd_iter. */ void abd_iter_init(struct abd_iter *aiter, abd_t *abd) { ASSERT(!abd_is_gang(abd)); abd_verify(abd); aiter->iter_abd = abd; aiter->iter_pos = 0; aiter->iter_mapaddr = NULL; aiter->iter_mapsize = 0; } /* * This is just a helper function to see if we have exhausted the * abd_iter and reached the end. */ boolean_t abd_iter_at_end(struct abd_iter *aiter) { return (aiter->iter_pos == aiter->iter_abd->abd_size); } /* * Advance the iterator by a certain amount. Cannot be called when a chunk is * in use. This can be safely called when the aiter has already exhausted, in * which case this does nothing. */ void abd_iter_advance(struct abd_iter *aiter, size_t amount) { ASSERT3P(aiter->iter_mapaddr, ==, NULL); ASSERT0(aiter->iter_mapsize); /* There's nothing left to advance to, so do nothing */ if (abd_iter_at_end(aiter)) return; aiter->iter_pos += amount; } /* * Map the current chunk into aiter. This can be safely called when the aiter * has already exhausted, in which case this does nothing. */ void abd_iter_map(struct abd_iter *aiter) { void *paddr; size_t offset = 0; ASSERT3P(aiter->iter_mapaddr, ==, NULL); ASSERT0(aiter->iter_mapsize); /* Panic if someone has changed zfs_abd_chunk_size */ IMPLY(!abd_is_linear(aiter->iter_abd), zfs_abd_chunk_size == ABD_SCATTER(aiter->iter_abd).abd_chunk_size); /* There's nothing left to iterate over, so do nothing */ if (abd_iter_at_end(aiter)) return; if (abd_is_linear(aiter->iter_abd)) { offset = aiter->iter_pos; aiter->iter_mapsize = aiter->iter_abd->abd_size - offset; paddr = ABD_LINEAR_BUF(aiter->iter_abd); } else { size_t index = abd_iter_scatter_chunk_index(aiter); offset = abd_iter_scatter_chunk_offset(aiter); aiter->iter_mapsize = MIN(zfs_abd_chunk_size - offset, aiter->iter_abd->abd_size - aiter->iter_pos); paddr = ABD_SCATTER(aiter->iter_abd).abd_chunks[index]; } aiter->iter_mapaddr = (char *)paddr + offset; } /* * Unmap the current chunk from aiter. This can be safely called when the aiter * has already exhausted, in which case this does nothing. */ void abd_iter_unmap(struct abd_iter *aiter) { /* There's nothing left to unmap, so do nothing */ if (abd_iter_at_end(aiter)) return; ASSERT3P(aiter->iter_mapaddr, !=, NULL); ASSERT3U(aiter->iter_mapsize, >, 0); aiter->iter_mapaddr = NULL; aiter->iter_mapsize = 0; } void abd_cache_reap_now(void) { kmem_cache_reap_soon(abd_chunk_cache); } diff --git a/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c b/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c index 551a3cc8d1db..af543d6e3f7e 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c @@ -1,1071 +1,1146 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2014 by Chunwei Chen. All rights reserved. * Copyright (c) 2019 by Delphix. All rights reserved. */ /* * See abd.c for a general overview of the arc buffered data (ABD). * * Linear buffers act exactly like normal buffers and are always mapped into the * kernel's virtual memory space, while scattered ABD data chunks are allocated * as physical pages and then mapped in only while they are actually being * accessed through one of the abd_* library functions. Using scattered ABDs * provides several benefits: * * (1) They avoid use of kmem_*, preventing performance problems where running * kmem_reap on very large memory systems never finishes and causes * constant TLB shootdowns. * * (2) Fragmentation is less of an issue since when we are at the limit of * allocatable space, we won't have to search around for a long free * hole in the VA space for large ARC allocations. Each chunk is mapped in * individually, so even if we are using HIGHMEM (see next point) we * wouldn't need to worry about finding a contiguous address range. * * (3) If we are not using HIGHMEM, then all physical memory is always * mapped into the kernel's address space, so we also avoid the map / * unmap costs on each ABD access. * * If we are not using HIGHMEM, scattered buffers which have only one chunk * can be treated as linear buffers, because they are contiguous in the * kernel's virtual address space. See abd_alloc_chunks() for details. */ #include #include #include #include #include #include #ifdef _KERNEL #include #include #else #define MAX_ORDER 1 #endif typedef struct abd_stats { kstat_named_t abdstat_struct_size; kstat_named_t abdstat_linear_cnt; kstat_named_t abdstat_linear_data_size; kstat_named_t abdstat_scatter_cnt; kstat_named_t abdstat_scatter_data_size; kstat_named_t abdstat_scatter_chunk_waste; kstat_named_t abdstat_scatter_orders[MAX_ORDER]; kstat_named_t abdstat_scatter_page_multi_chunk; kstat_named_t abdstat_scatter_page_multi_zone; kstat_named_t abdstat_scatter_page_alloc_retry; kstat_named_t abdstat_scatter_sg_table_retry; } abd_stats_t; static abd_stats_t abd_stats = { /* Amount of memory occupied by all of the abd_t struct allocations */ { "struct_size", KSTAT_DATA_UINT64 }, /* * The number of linear ABDs which are currently allocated, excluding * ABDs which don't own their data (for instance the ones which were * allocated through abd_get_offset() and abd_get_from_buf()). If an * ABD takes ownership of its buf then it will become tracked. */ { "linear_cnt", KSTAT_DATA_UINT64 }, /* Amount of data stored in all linear ABDs tracked by linear_cnt */ { "linear_data_size", KSTAT_DATA_UINT64 }, /* * The number of scatter ABDs which are currently allocated, excluding * ABDs which don't own their data (for instance the ones which were * allocated through abd_get_offset()). */ { "scatter_cnt", KSTAT_DATA_UINT64 }, /* Amount of data stored in all scatter ABDs tracked by scatter_cnt */ { "scatter_data_size", KSTAT_DATA_UINT64 }, /* * The amount of space wasted at the end of the last chunk across all * scatter ABDs tracked by scatter_cnt. */ { "scatter_chunk_waste", KSTAT_DATA_UINT64 }, /* * The number of compound allocations of a given order. These * allocations are spread over all currently allocated ABDs, and * act as a measure of memory fragmentation. */ { { "scatter_order_N", KSTAT_DATA_UINT64 } }, /* * The number of scatter ABDs which contain multiple chunks. * ABDs are preferentially allocated from the minimum number of * contiguous multi-page chunks, a single chunk is optimal. */ { "scatter_page_multi_chunk", KSTAT_DATA_UINT64 }, /* * The number of scatter ABDs which are split across memory zones. * ABDs are preferentially allocated using pages from a single zone. */ { "scatter_page_multi_zone", KSTAT_DATA_UINT64 }, /* * The total number of retries encountered when attempting to * allocate the pages to populate the scatter ABD. */ { "scatter_page_alloc_retry", KSTAT_DATA_UINT64 }, /* * The total number of retries encountered when attempting to * allocate the sg table for an ABD. */ { "scatter_sg_table_retry", KSTAT_DATA_UINT64 }, }; +struct { + wmsum_t abdstat_struct_size; + wmsum_t abdstat_linear_cnt; + wmsum_t abdstat_linear_data_size; + wmsum_t abdstat_scatter_cnt; + wmsum_t abdstat_scatter_data_size; + wmsum_t abdstat_scatter_chunk_waste; + wmsum_t abdstat_scatter_orders[MAX_ORDER]; + wmsum_t abdstat_scatter_page_multi_chunk; + wmsum_t abdstat_scatter_page_multi_zone; + wmsum_t abdstat_scatter_page_alloc_retry; + wmsum_t abdstat_scatter_sg_table_retry; +} abd_sums; + #define abd_for_each_sg(abd, sg, n, i) \ for_each_sg(ABD_SCATTER(abd).abd_sgl, sg, n, i) unsigned zfs_abd_scatter_max_order = MAX_ORDER - 1; /* * zfs_abd_scatter_min_size is the minimum allocation size to use scatter * ABD's. Smaller allocations will use linear ABD's which uses * zio_[data_]buf_alloc(). * * Scatter ABD's use at least one page each, so sub-page allocations waste * some space when allocated as scatter (e.g. 2KB scatter allocation wastes * half of each page). Using linear ABD's for small allocations means that * they will be put on slabs which contain many allocations. This can * improve memory efficiency, but it also makes it much harder for ARC * evictions to actually free pages, because all the buffers on one slab need * to be freed in order for the slab (and underlying pages) to be freed. * Typically, 512B and 1KB kmem caches have 16 buffers per slab, so it's * possible for them to actually waste more memory than scatter (one page per * buf = wasting 3/4 or 7/8th; one buf per slab = wasting 15/16th). * * Spill blocks are typically 512B and are heavily used on systems running * selinux with the default dnode size and the `xattr=sa` property set. * * By default we use linear allocations for 512B and 1KB, and scatter * allocations for larger (1.5KB and up). */ int zfs_abd_scatter_min_size = 512 * 3; /* * We use a scattered SPA_MAXBLOCKSIZE sized ABD whose pages are * just a single zero'd page. This allows us to conserve memory by * only using a single zero page for the scatterlist. */ abd_t *abd_zero_scatter = NULL; struct page; /* * abd_zero_page we will be an allocated zero'd PAGESIZE buffer, which is * assigned to set each of the pages of abd_zero_scatter. */ static struct page *abd_zero_page = NULL; static kmem_cache_t *abd_cache = NULL; static kstat_t *abd_ksp; static uint_t abd_chunkcnt_for_bytes(size_t size) { return (P2ROUNDUP(size, PAGESIZE) / PAGESIZE); } abd_t * abd_alloc_struct_impl(size_t size) { /* * In Linux we do not use the size passed in during ABD * allocation, so we just ignore it. */ abd_t *abd = kmem_cache_alloc(abd_cache, KM_PUSHPAGE); ASSERT3P(abd, !=, NULL); ABDSTAT_INCR(abdstat_struct_size, sizeof (abd_t)); return (abd); } void abd_free_struct_impl(abd_t *abd) { kmem_cache_free(abd_cache, abd); ABDSTAT_INCR(abdstat_struct_size, -(int)sizeof (abd_t)); } #ifdef _KERNEL /* * Mark zfs data pages so they can be excluded from kernel crash dumps */ #ifdef _LP64 #define ABD_FILE_CACHE_PAGE 0x2F5ABDF11ECAC4E static inline void abd_mark_zfs_page(struct page *page) { get_page(page); SetPagePrivate(page); set_page_private(page, ABD_FILE_CACHE_PAGE); } static inline void abd_unmark_zfs_page(struct page *page) { set_page_private(page, 0UL); ClearPagePrivate(page); put_page(page); } #else #define abd_mark_zfs_page(page) #define abd_unmark_zfs_page(page) #endif /* _LP64 */ #ifndef CONFIG_HIGHMEM #ifndef __GFP_RECLAIM #define __GFP_RECLAIM __GFP_WAIT #endif /* * The goal is to minimize fragmentation by preferentially populating ABDs * with higher order compound pages from a single zone. Allocation size is * progressively decreased until it can be satisfied without performing * reclaim or compaction. When necessary this function will degenerate to * allocating individual pages and allowing reclaim to satisfy allocations. */ void abd_alloc_chunks(abd_t *abd, size_t size) { struct list_head pages; struct sg_table table; struct scatterlist *sg; struct page *page, *tmp_page = NULL; gfp_t gfp = __GFP_NOWARN | GFP_NOIO; gfp_t gfp_comp = (gfp | __GFP_NORETRY | __GFP_COMP) & ~__GFP_RECLAIM; int max_order = MIN(zfs_abd_scatter_max_order, MAX_ORDER - 1); int nr_pages = abd_chunkcnt_for_bytes(size); int chunks = 0, zones = 0; size_t remaining_size; int nid = NUMA_NO_NODE; int alloc_pages = 0; INIT_LIST_HEAD(&pages); while (alloc_pages < nr_pages) { unsigned chunk_pages; int order; order = MIN(highbit64(nr_pages - alloc_pages) - 1, max_order); chunk_pages = (1U << order); page = alloc_pages_node(nid, order ? gfp_comp : gfp, order); if (page == NULL) { if (order == 0) { ABDSTAT_BUMP(abdstat_scatter_page_alloc_retry); schedule_timeout_interruptible(1); } else { max_order = MAX(0, order - 1); } continue; } list_add_tail(&page->lru, &pages); if ((nid != NUMA_NO_NODE) && (page_to_nid(page) != nid)) zones++; nid = page_to_nid(page); ABDSTAT_BUMP(abdstat_scatter_orders[order]); chunks++; alloc_pages += chunk_pages; } ASSERT3S(alloc_pages, ==, nr_pages); while (sg_alloc_table(&table, chunks, gfp)) { ABDSTAT_BUMP(abdstat_scatter_sg_table_retry); schedule_timeout_interruptible(1); } sg = table.sgl; remaining_size = size; list_for_each_entry_safe(page, tmp_page, &pages, lru) { size_t sg_size = MIN(PAGESIZE << compound_order(page), remaining_size); sg_set_page(sg, page, sg_size, 0); abd_mark_zfs_page(page); remaining_size -= sg_size; sg = sg_next(sg); list_del(&page->lru); } /* * These conditions ensure that a possible transformation to a linear * ABD would be valid. */ ASSERT(!PageHighMem(sg_page(table.sgl))); ASSERT0(ABD_SCATTER(abd).abd_offset); if (table.nents == 1) { /* * Since there is only one entry, this ABD can be represented * as a linear buffer. All single-page (4K) ABD's can be * represented this way. Some multi-page ABD's can also be * represented this way, if we were able to allocate a single * "chunk" (higher-order "page" which represents a power-of-2 * series of physically-contiguous pages). This is often the * case for 2-page (8K) ABD's. * * Representing a single-entry scatter ABD as a linear ABD * has the performance advantage of avoiding the copy (and * allocation) in abd_borrow_buf_copy / abd_return_buf_copy. * A performance increase of around 5% has been observed for * ARC-cached reads (of small blocks which can take advantage * of this). * * Note that this optimization is only possible because the * pages are always mapped into the kernel's address space. * This is not the case for highmem pages, so the * optimization can not be made there. */ abd->abd_flags |= ABD_FLAG_LINEAR; abd->abd_flags |= ABD_FLAG_LINEAR_PAGE; abd->abd_u.abd_linear.abd_sgl = table.sgl; ABD_LINEAR_BUF(abd) = page_address(sg_page(table.sgl)); } else if (table.nents > 1) { ABDSTAT_BUMP(abdstat_scatter_page_multi_chunk); abd->abd_flags |= ABD_FLAG_MULTI_CHUNK; if (zones) { ABDSTAT_BUMP(abdstat_scatter_page_multi_zone); abd->abd_flags |= ABD_FLAG_MULTI_ZONE; } ABD_SCATTER(abd).abd_sgl = table.sgl; ABD_SCATTER(abd).abd_nents = table.nents; } } #else /* * Allocate N individual pages to construct a scatter ABD. This function * makes no attempt to request contiguous pages and requires the minimal * number of kernel interfaces. It's designed for maximum compatibility. */ void abd_alloc_chunks(abd_t *abd, size_t size) { struct scatterlist *sg = NULL; struct sg_table table; struct page *page; gfp_t gfp = __GFP_NOWARN | GFP_NOIO; int nr_pages = abd_chunkcnt_for_bytes(size); int i = 0; while (sg_alloc_table(&table, nr_pages, gfp)) { ABDSTAT_BUMP(abdstat_scatter_sg_table_retry); schedule_timeout_interruptible(1); } ASSERT3U(table.nents, ==, nr_pages); ABD_SCATTER(abd).abd_sgl = table.sgl; ABD_SCATTER(abd).abd_nents = nr_pages; abd_for_each_sg(abd, sg, nr_pages, i) { while ((page = __page_cache_alloc(gfp)) == NULL) { ABDSTAT_BUMP(abdstat_scatter_page_alloc_retry); schedule_timeout_interruptible(1); } ABDSTAT_BUMP(abdstat_scatter_orders[0]); sg_set_page(sg, page, PAGESIZE, 0); abd_mark_zfs_page(page); } if (nr_pages > 1) { ABDSTAT_BUMP(abdstat_scatter_page_multi_chunk); abd->abd_flags |= ABD_FLAG_MULTI_CHUNK; } } #endif /* !CONFIG_HIGHMEM */ /* * This must be called if any of the sg_table allocation functions * are called. */ static void abd_free_sg_table(abd_t *abd) { struct sg_table table; table.sgl = ABD_SCATTER(abd).abd_sgl; table.nents = table.orig_nents = ABD_SCATTER(abd).abd_nents; sg_free_table(&table); } void abd_free_chunks(abd_t *abd) { struct scatterlist *sg = NULL; struct page *page; int nr_pages = ABD_SCATTER(abd).abd_nents; int order, i = 0; if (abd->abd_flags & ABD_FLAG_MULTI_ZONE) ABDSTAT_BUMPDOWN(abdstat_scatter_page_multi_zone); if (abd->abd_flags & ABD_FLAG_MULTI_CHUNK) ABDSTAT_BUMPDOWN(abdstat_scatter_page_multi_chunk); abd_for_each_sg(abd, sg, nr_pages, i) { page = sg_page(sg); abd_unmark_zfs_page(page); order = compound_order(page); __free_pages(page, order); ASSERT3U(sg->length, <=, PAGE_SIZE << order); ABDSTAT_BUMPDOWN(abdstat_scatter_orders[order]); } abd_free_sg_table(abd); } /* * Allocate scatter ABD of size SPA_MAXBLOCKSIZE, where each page in * the scatterlist will be set to the zero'd out buffer abd_zero_page. */ static void abd_alloc_zero_scatter(void) { struct scatterlist *sg = NULL; struct sg_table table; gfp_t gfp = __GFP_NOWARN | GFP_NOIO; gfp_t gfp_zero_page = gfp | __GFP_ZERO; int nr_pages = abd_chunkcnt_for_bytes(SPA_MAXBLOCKSIZE); int i = 0; while ((abd_zero_page = __page_cache_alloc(gfp_zero_page)) == NULL) { ABDSTAT_BUMP(abdstat_scatter_page_alloc_retry); schedule_timeout_interruptible(1); } abd_mark_zfs_page(abd_zero_page); while (sg_alloc_table(&table, nr_pages, gfp)) { ABDSTAT_BUMP(abdstat_scatter_sg_table_retry); schedule_timeout_interruptible(1); } ASSERT3U(table.nents, ==, nr_pages); abd_zero_scatter = abd_alloc_struct(SPA_MAXBLOCKSIZE); abd_zero_scatter->abd_flags |= ABD_FLAG_OWNER; ABD_SCATTER(abd_zero_scatter).abd_offset = 0; ABD_SCATTER(abd_zero_scatter).abd_sgl = table.sgl; ABD_SCATTER(abd_zero_scatter).abd_nents = nr_pages; abd_zero_scatter->abd_size = SPA_MAXBLOCKSIZE; abd_zero_scatter->abd_flags |= ABD_FLAG_MULTI_CHUNK | ABD_FLAG_ZEROS; abd_for_each_sg(abd_zero_scatter, sg, nr_pages, i) { sg_set_page(sg, abd_zero_page, PAGESIZE, 0); } ABDSTAT_BUMP(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, PAGESIZE); ABDSTAT_BUMP(abdstat_scatter_page_multi_chunk); } #else /* _KERNEL */ #ifndef PAGE_SHIFT #define PAGE_SHIFT (highbit64(PAGESIZE)-1) #endif #define zfs_kmap_atomic(chunk) ((void *)chunk) #define zfs_kunmap_atomic(addr) do { (void)(addr); } while (0) #define local_irq_save(flags) do { (void)(flags); } while (0) #define local_irq_restore(flags) do { (void)(flags); } while (0) #define nth_page(pg, i) \ ((struct page *)((void *)(pg) + (i) * PAGESIZE)) struct scatterlist { struct page *page; int length; int end; }; static void sg_init_table(struct scatterlist *sg, int nr) { memset(sg, 0, nr * sizeof (struct scatterlist)); sg[nr - 1].end = 1; } /* * This must be called if any of the sg_table allocation functions * are called. */ static void abd_free_sg_table(abd_t *abd) { int nents = ABD_SCATTER(abd).abd_nents; vmem_free(ABD_SCATTER(abd).abd_sgl, nents * sizeof (struct scatterlist)); } #define for_each_sg(sgl, sg, nr, i) \ for ((i) = 0, (sg) = (sgl); (i) < (nr); (i)++, (sg) = sg_next(sg)) static inline void sg_set_page(struct scatterlist *sg, struct page *page, unsigned int len, unsigned int offset) { /* currently we don't use offset */ ASSERT(offset == 0); sg->page = page; sg->length = len; } static inline struct page * sg_page(struct scatterlist *sg) { return (sg->page); } static inline struct scatterlist * sg_next(struct scatterlist *sg) { if (sg->end) return (NULL); return (sg + 1); } void abd_alloc_chunks(abd_t *abd, size_t size) { unsigned nr_pages = abd_chunkcnt_for_bytes(size); struct scatterlist *sg; int i; ABD_SCATTER(abd).abd_sgl = vmem_alloc(nr_pages * sizeof (struct scatterlist), KM_SLEEP); sg_init_table(ABD_SCATTER(abd).abd_sgl, nr_pages); abd_for_each_sg(abd, sg, nr_pages, i) { struct page *p = umem_alloc_aligned(PAGESIZE, 64, KM_SLEEP); sg_set_page(sg, p, PAGESIZE, 0); } ABD_SCATTER(abd).abd_nents = nr_pages; } void abd_free_chunks(abd_t *abd) { int i, n = ABD_SCATTER(abd).abd_nents; struct scatterlist *sg; abd_for_each_sg(abd, sg, n, i) { for (int j = 0; j < sg->length; j += PAGESIZE) { struct page *p = nth_page(sg_page(sg), j >> PAGE_SHIFT); umem_free(p, PAGESIZE); } } abd_free_sg_table(abd); } static void abd_alloc_zero_scatter(void) { unsigned nr_pages = abd_chunkcnt_for_bytes(SPA_MAXBLOCKSIZE); struct scatterlist *sg; int i; abd_zero_page = umem_alloc_aligned(PAGESIZE, 64, KM_SLEEP); memset(abd_zero_page, 0, PAGESIZE); abd_zero_scatter = abd_alloc_struct(SPA_MAXBLOCKSIZE); abd_zero_scatter->abd_flags |= ABD_FLAG_OWNER; abd_zero_scatter->abd_flags |= ABD_FLAG_MULTI_CHUNK | ABD_FLAG_ZEROS; ABD_SCATTER(abd_zero_scatter).abd_offset = 0; ABD_SCATTER(abd_zero_scatter).abd_nents = nr_pages; abd_zero_scatter->abd_size = SPA_MAXBLOCKSIZE; zfs_refcount_create(&abd_zero_scatter->abd_children); ABD_SCATTER(abd_zero_scatter).abd_sgl = vmem_alloc(nr_pages * sizeof (struct scatterlist), KM_SLEEP); sg_init_table(ABD_SCATTER(abd_zero_scatter).abd_sgl, nr_pages); abd_for_each_sg(abd_zero_scatter, sg, nr_pages, i) { sg_set_page(sg, abd_zero_page, PAGESIZE, 0); } ABDSTAT_BUMP(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, PAGESIZE); ABDSTAT_BUMP(abdstat_scatter_page_multi_chunk); } #endif /* _KERNEL */ boolean_t abd_size_alloc_linear(size_t size) { return (size < zfs_abd_scatter_min_size ? B_TRUE : B_FALSE); } void abd_update_scatter_stats(abd_t *abd, abd_stats_op_t op) { ASSERT(op == ABDSTAT_INCR || op == ABDSTAT_DECR); int waste = P2ROUNDUP(abd->abd_size, PAGESIZE) - abd->abd_size; if (op == ABDSTAT_INCR) { ABDSTAT_BUMP(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, abd->abd_size); ABDSTAT_INCR(abdstat_scatter_chunk_waste, waste); arc_space_consume(waste, ARC_SPACE_ABD_CHUNK_WASTE); } else { ABDSTAT_BUMPDOWN(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, -(int)abd->abd_size); ABDSTAT_INCR(abdstat_scatter_chunk_waste, -waste); arc_space_return(waste, ARC_SPACE_ABD_CHUNK_WASTE); } } void abd_update_linear_stats(abd_t *abd, abd_stats_op_t op) { ASSERT(op == ABDSTAT_INCR || op == ABDSTAT_DECR); if (op == ABDSTAT_INCR) { ABDSTAT_BUMP(abdstat_linear_cnt); ABDSTAT_INCR(abdstat_linear_data_size, abd->abd_size); } else { ABDSTAT_BUMPDOWN(abdstat_linear_cnt); ABDSTAT_INCR(abdstat_linear_data_size, -(int)abd->abd_size); } } void abd_verify_scatter(abd_t *abd) { size_t n; int i = 0; struct scatterlist *sg = NULL; ASSERT3U(ABD_SCATTER(abd).abd_nents, >, 0); ASSERT3U(ABD_SCATTER(abd).abd_offset, <, ABD_SCATTER(abd).abd_sgl->length); n = ABD_SCATTER(abd).abd_nents; abd_for_each_sg(abd, sg, n, i) { ASSERT3P(sg_page(sg), !=, NULL); } } static void abd_free_zero_scatter(void) { ABDSTAT_BUMPDOWN(abdstat_scatter_cnt); ABDSTAT_INCR(abdstat_scatter_data_size, -(int)PAGESIZE); ABDSTAT_BUMPDOWN(abdstat_scatter_page_multi_chunk); abd_free_sg_table(abd_zero_scatter); abd_free_struct(abd_zero_scatter); abd_zero_scatter = NULL; ASSERT3P(abd_zero_page, !=, NULL); #if defined(_KERNEL) abd_unmark_zfs_page(abd_zero_page); __free_page(abd_zero_page); #else umem_free(abd_zero_page, PAGESIZE); #endif /* _KERNEL */ } +static int +abd_kstats_update(kstat_t *ksp, int rw) +{ + abd_stats_t *as = ksp->ks_data; + + if (rw == KSTAT_WRITE) + return (EACCES); + as->abdstat_struct_size.value.ui64 = + wmsum_value(&abd_sums.abdstat_struct_size); + as->abdstat_linear_cnt.value.ui64 = + wmsum_value(&abd_sums.abdstat_linear_cnt); + as->abdstat_linear_data_size.value.ui64 = + wmsum_value(&abd_sums.abdstat_linear_data_size); + as->abdstat_scatter_cnt.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_cnt); + as->abdstat_scatter_data_size.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_data_size); + as->abdstat_scatter_chunk_waste.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_chunk_waste); + for (int i = 0; i < MAX_ORDER; i++) { + as->abdstat_scatter_orders[i].value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_orders[i]); + } + as->abdstat_scatter_page_multi_chunk.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_page_multi_chunk); + as->abdstat_scatter_page_multi_zone.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_page_multi_zone); + as->abdstat_scatter_page_alloc_retry.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_page_alloc_retry); + as->abdstat_scatter_sg_table_retry.value.ui64 = + wmsum_value(&abd_sums.abdstat_scatter_sg_table_retry); + return (0); +} + void abd_init(void) { int i; abd_cache = kmem_cache_create("abd_t", sizeof (abd_t), 0, NULL, NULL, NULL, NULL, NULL, 0); + wmsum_init(&abd_sums.abdstat_struct_size, 0); + wmsum_init(&abd_sums.abdstat_linear_cnt, 0); + wmsum_init(&abd_sums.abdstat_linear_data_size, 0); + wmsum_init(&abd_sums.abdstat_scatter_cnt, 0); + wmsum_init(&abd_sums.abdstat_scatter_data_size, 0); + wmsum_init(&abd_sums.abdstat_scatter_chunk_waste, 0); + for (i = 0; i < MAX_ORDER; i++) + wmsum_init(&abd_sums.abdstat_scatter_orders[i], 0); + wmsum_init(&abd_sums.abdstat_scatter_page_multi_chunk, 0); + wmsum_init(&abd_sums.abdstat_scatter_page_multi_zone, 0); + wmsum_init(&abd_sums.abdstat_scatter_page_alloc_retry, 0); + wmsum_init(&abd_sums.abdstat_scatter_sg_table_retry, 0); + abd_ksp = kstat_create("zfs", 0, "abdstats", "misc", KSTAT_TYPE_NAMED, sizeof (abd_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (abd_ksp != NULL) { for (i = 0; i < MAX_ORDER; i++) { snprintf(abd_stats.abdstat_scatter_orders[i].name, KSTAT_STRLEN, "scatter_order_%d", i); abd_stats.abdstat_scatter_orders[i].data_type = KSTAT_DATA_UINT64; } abd_ksp->ks_data = &abd_stats; + abd_ksp->ks_update = abd_kstats_update; kstat_install(abd_ksp); } abd_alloc_zero_scatter(); } void abd_fini(void) { abd_free_zero_scatter(); if (abd_ksp != NULL) { kstat_delete(abd_ksp); abd_ksp = NULL; } + wmsum_fini(&abd_sums.abdstat_struct_size); + wmsum_fini(&abd_sums.abdstat_linear_cnt); + wmsum_fini(&abd_sums.abdstat_linear_data_size); + wmsum_fini(&abd_sums.abdstat_scatter_cnt); + wmsum_fini(&abd_sums.abdstat_scatter_data_size); + wmsum_fini(&abd_sums.abdstat_scatter_chunk_waste); + for (int i = 0; i < MAX_ORDER; i++) + wmsum_fini(&abd_sums.abdstat_scatter_orders[i]); + wmsum_fini(&abd_sums.abdstat_scatter_page_multi_chunk); + wmsum_fini(&abd_sums.abdstat_scatter_page_multi_zone); + wmsum_fini(&abd_sums.abdstat_scatter_page_alloc_retry); + wmsum_fini(&abd_sums.abdstat_scatter_sg_table_retry); + if (abd_cache) { kmem_cache_destroy(abd_cache); abd_cache = NULL; } } void abd_free_linear_page(abd_t *abd) { /* Transform it back into a scatter ABD for freeing */ struct scatterlist *sg = abd->abd_u.abd_linear.abd_sgl; abd->abd_flags &= ~ABD_FLAG_LINEAR; abd->abd_flags &= ~ABD_FLAG_LINEAR_PAGE; ABD_SCATTER(abd).abd_nents = 1; ABD_SCATTER(abd).abd_offset = 0; ABD_SCATTER(abd).abd_sgl = sg; abd_free_chunks(abd); abd_update_scatter_stats(abd, ABDSTAT_DECR); } /* * If we're going to use this ABD for doing I/O using the block layer, the * consumer of the ABD data doesn't care if it's scattered or not, and we don't * plan to store this ABD in memory for a long period of time, we should * allocate the ABD type that requires the least data copying to do the I/O. * * On Linux the optimal thing to do would be to use abd_get_offset() and * construct a new ABD which shares the original pages thereby eliminating * the copy. But for the moment a new linear ABD is allocated until this * performance optimization can be implemented. */ abd_t * abd_alloc_for_io(size_t size, boolean_t is_metadata) { return (abd_alloc(size, is_metadata)); } abd_t * abd_get_offset_scatter(abd_t *abd, abd_t *sabd, size_t off) { int i = 0; struct scatterlist *sg = NULL; abd_verify(sabd); ASSERT3U(off, <=, sabd->abd_size); size_t new_offset = ABD_SCATTER(sabd).abd_offset + off; if (abd == NULL) abd = abd_alloc_struct(0); /* * Even if this buf is filesystem metadata, we only track that * if we own the underlying data buffer, which is not true in * this case. Therefore, we don't ever use ABD_FLAG_META here. */ abd_for_each_sg(sabd, sg, ABD_SCATTER(sabd).abd_nents, i) { if (new_offset < sg->length) break; new_offset -= sg->length; } ABD_SCATTER(abd).abd_sgl = sg; ABD_SCATTER(abd).abd_offset = new_offset; ABD_SCATTER(abd).abd_nents = ABD_SCATTER(sabd).abd_nents - i; return (abd); } /* * Initialize the abd_iter. */ void abd_iter_init(struct abd_iter *aiter, abd_t *abd) { ASSERT(!abd_is_gang(abd)); abd_verify(abd); aiter->iter_abd = abd; aiter->iter_mapaddr = NULL; aiter->iter_mapsize = 0; aiter->iter_pos = 0; if (abd_is_linear(abd)) { aiter->iter_offset = 0; aiter->iter_sg = NULL; } else { aiter->iter_offset = ABD_SCATTER(abd).abd_offset; aiter->iter_sg = ABD_SCATTER(abd).abd_sgl; } } /* * This is just a helper function to see if we have exhausted the * abd_iter and reached the end. */ boolean_t abd_iter_at_end(struct abd_iter *aiter) { return (aiter->iter_pos == aiter->iter_abd->abd_size); } /* * Advance the iterator by a certain amount. Cannot be called when a chunk is * in use. This can be safely called when the aiter has already exhausted, in * which case this does nothing. */ void abd_iter_advance(struct abd_iter *aiter, size_t amount) { ASSERT3P(aiter->iter_mapaddr, ==, NULL); ASSERT0(aiter->iter_mapsize); /* There's nothing left to advance to, so do nothing */ if (abd_iter_at_end(aiter)) return; aiter->iter_pos += amount; aiter->iter_offset += amount; if (!abd_is_linear(aiter->iter_abd)) { while (aiter->iter_offset >= aiter->iter_sg->length) { aiter->iter_offset -= aiter->iter_sg->length; aiter->iter_sg = sg_next(aiter->iter_sg); if (aiter->iter_sg == NULL) { ASSERT0(aiter->iter_offset); break; } } } } /* * Map the current chunk into aiter. This can be safely called when the aiter * has already exhausted, in which case this does nothing. */ void abd_iter_map(struct abd_iter *aiter) { void *paddr; size_t offset = 0; ASSERT3P(aiter->iter_mapaddr, ==, NULL); ASSERT0(aiter->iter_mapsize); /* There's nothing left to iterate over, so do nothing */ if (abd_iter_at_end(aiter)) return; if (abd_is_linear(aiter->iter_abd)) { ASSERT3U(aiter->iter_pos, ==, aiter->iter_offset); offset = aiter->iter_offset; aiter->iter_mapsize = aiter->iter_abd->abd_size - offset; paddr = ABD_LINEAR_BUF(aiter->iter_abd); } else { offset = aiter->iter_offset; aiter->iter_mapsize = MIN(aiter->iter_sg->length - offset, aiter->iter_abd->abd_size - aiter->iter_pos); paddr = zfs_kmap_atomic(sg_page(aiter->iter_sg)); } aiter->iter_mapaddr = (char *)paddr + offset; } /* * Unmap the current chunk from aiter. This can be safely called when the aiter * has already exhausted, in which case this does nothing. */ void abd_iter_unmap(struct abd_iter *aiter) { /* There's nothing left to unmap, so do nothing */ if (abd_iter_at_end(aiter)) return; if (!abd_is_linear(aiter->iter_abd)) { /* LINTED E_FUNC_SET_NOT_USED */ zfs_kunmap_atomic(aiter->iter_mapaddr - aiter->iter_offset); } ASSERT3P(aiter->iter_mapaddr, !=, NULL); ASSERT3U(aiter->iter_mapsize, >, 0); aiter->iter_mapaddr = NULL; aiter->iter_mapsize = 0; } void abd_cache_reap_now(void) { } #if defined(_KERNEL) /* * bio_nr_pages for ABD. * @off is the offset in @abd */ unsigned long abd_nr_pages_off(abd_t *abd, unsigned int size, size_t off) { unsigned long pos; if (abd_is_gang(abd)) { unsigned long count = 0; for (abd_t *cabd = abd_gang_get_offset(abd, &off); cabd != NULL && size != 0; cabd = list_next(&ABD_GANG(abd).abd_gang_chain, cabd)) { ASSERT3U(off, <, cabd->abd_size); int mysize = MIN(size, cabd->abd_size - off); count += abd_nr_pages_off(cabd, mysize, off); size -= mysize; off = 0; } return (count); } if (abd_is_linear(abd)) pos = (unsigned long)abd_to_buf(abd) + off; else pos = ABD_SCATTER(abd).abd_offset + off; return (((pos + size + PAGESIZE - 1) >> PAGE_SHIFT) - (pos >> PAGE_SHIFT)); } static unsigned int bio_map(struct bio *bio, void *buf_ptr, unsigned int bio_size) { unsigned int offset, size, i; struct page *page; offset = offset_in_page(buf_ptr); for (i = 0; i < bio->bi_max_vecs; i++) { size = PAGE_SIZE - offset; if (bio_size <= 0) break; if (size > bio_size) size = bio_size; if (is_vmalloc_addr(buf_ptr)) page = vmalloc_to_page(buf_ptr); else page = virt_to_page(buf_ptr); /* * Some network related block device uses tcp_sendpage, which * doesn't behave well when using 0-count page, this is a * safety net to catch them. */ ASSERT3S(page_count(page), >, 0); if (bio_add_page(bio, page, size, offset) != size) break; buf_ptr += size; bio_size -= size; offset = 0; } return (bio_size); } /* * bio_map for gang ABD. */ static unsigned int abd_gang_bio_map_off(struct bio *bio, abd_t *abd, unsigned int io_size, size_t off) { ASSERT(abd_is_gang(abd)); for (abd_t *cabd = abd_gang_get_offset(abd, &off); cabd != NULL; cabd = list_next(&ABD_GANG(abd).abd_gang_chain, cabd)) { ASSERT3U(off, <, cabd->abd_size); int size = MIN(io_size, cabd->abd_size - off); int remainder = abd_bio_map_off(bio, cabd, size, off); io_size -= (size - remainder); if (io_size == 0 || remainder > 0) return (io_size); off = 0; } ASSERT0(io_size); return (io_size); } /* * bio_map for ABD. * @off is the offset in @abd * Remaining IO size is returned */ unsigned int abd_bio_map_off(struct bio *bio, abd_t *abd, unsigned int io_size, size_t off) { struct abd_iter aiter; ASSERT3U(io_size, <=, abd->abd_size - off); if (abd_is_linear(abd)) return (bio_map(bio, ((char *)abd_to_buf(abd)) + off, io_size)); ASSERT(!abd_is_linear(abd)); if (abd_is_gang(abd)) return (abd_gang_bio_map_off(bio, abd, io_size, off)); abd_iter_init(&aiter, abd); abd_iter_advance(&aiter, off); for (int i = 0; i < bio->bi_max_vecs; i++) { struct page *pg; size_t len, sgoff, pgoff; struct scatterlist *sg; if (io_size <= 0) break; sg = aiter.iter_sg; sgoff = aiter.iter_offset; pgoff = sgoff & (PAGESIZE - 1); len = MIN(io_size, PAGESIZE - pgoff); ASSERT(len > 0); pg = nth_page(sg_page(sg), sgoff >> PAGE_SHIFT); if (bio_add_page(bio, pg, len, pgoff) != len) break; io_size -= len; abd_iter_advance(&aiter, len); } return (io_size); } /* Tunable Parameters */ module_param(zfs_abd_scatter_enabled, int, 0644); MODULE_PARM_DESC(zfs_abd_scatter_enabled, "Toggle whether ABD allocations must be linear."); module_param(zfs_abd_scatter_min_size, int, 0644); MODULE_PARM_DESC(zfs_abd_scatter_min_size, "Minimum size of scatter allocations."); /* CSTYLED */ module_param(zfs_abd_scatter_max_order, uint, 0644); MODULE_PARM_DESC(zfs_abd_scatter_max_order, "Maximum order allocation used for a scatter ABD."); #endif diff --git a/sys/contrib/openzfs/module/os/linux/zfs/arc_os.c b/sys/contrib/openzfs/module/os/linux/zfs/arc_os.c index 465775a6748e..b03ad8318d1d 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/arc_os.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/arc_os.c @@ -1,529 +1,529 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2018, Joyent, Inc. * Copyright (c) 2011, 2019 by Delphix. All rights reserved. * Copyright (c) 2014 by Saso Kiselkov. All rights reserved. * Copyright 2017 Nexenta Systems, Inc. All rights reserved. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #include #include #include #include #include #endif #include #include #include #include #include #include #include /* * This is a limit on how many pages the ARC shrinker makes available for * eviction in response to one page allocation attempt. Note that in * practice, the kernel's shrinker can ask us to evict up to about 4x this * for one allocation attempt. * * The default limit of 10,000 (in practice, 160MB per allocation attempt * with 4K pages) limits the amount of time spent attempting to reclaim ARC * memory to less than 100ms per allocation attempt, even with a small * average compressed block size of ~8KB. * * See also the comment in arc_shrinker_count(). * Set to 0 to disable limit. */ int zfs_arc_shrinker_limit = 10000; #ifdef CONFIG_MEMORY_HOTPLUG static struct notifier_block arc_hotplug_callback_mem_nb; #endif /* * Return a default max arc size based on the amount of physical memory. */ uint64_t arc_default_max(uint64_t min, uint64_t allmem) { /* Default to 1/2 of all memory. */ return (MAX(allmem / 2, min)); } #ifdef _KERNEL /* * Return maximum amount of memory that we could possibly use. Reduced * to half of all memory in user space which is primarily used for testing. */ uint64_t arc_all_memory(void) { #ifdef CONFIG_HIGHMEM return (ptob(zfs_totalram_pages - zfs_totalhigh_pages)); #else return (ptob(zfs_totalram_pages)); #endif /* CONFIG_HIGHMEM */ } /* * Return the amount of memory that is considered free. In user space * which is primarily used for testing we pretend that free memory ranges * from 0-20% of all memory. */ uint64_t arc_free_memory(void) { #ifdef CONFIG_HIGHMEM struct sysinfo si; si_meminfo(&si); return (ptob(si.freeram - si.freehigh)); #else return (ptob(nr_free_pages() + nr_inactive_file_pages())); #endif /* CONFIG_HIGHMEM */ } /* * Return the amount of memory that can be consumed before reclaim will be * needed. Positive if there is sufficient free memory, negative indicates * the amount of memory that needs to be freed up. */ int64_t arc_available_memory(void) { return (arc_free_memory() - arc_sys_free); } static uint64_t arc_evictable_memory(void) { - int64_t asize = aggsum_value(&arc_size); + int64_t asize = aggsum_value(&arc_sums.arcstat_size); uint64_t arc_clean = zfs_refcount_count(&arc_mru->arcs_esize[ARC_BUFC_DATA]) + zfs_refcount_count(&arc_mru->arcs_esize[ARC_BUFC_METADATA]) + zfs_refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_DATA]) + zfs_refcount_count(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); uint64_t arc_dirty = MAX((int64_t)asize - (int64_t)arc_clean, 0); /* * Scale reported evictable memory in proportion to page cache, cap * at specified min/max. */ uint64_t min = (ptob(nr_file_pages()) / 100) * zfs_arc_pc_percent; min = MAX(arc_c_min, MIN(arc_c_max, min)); if (arc_dirty >= min) return (arc_clean); return (MAX((int64_t)asize - (int64_t)min, 0)); } /* * The _count() function returns the number of free-able objects. * The _scan() function returns the number of objects that were freed. */ static unsigned long arc_shrinker_count(struct shrinker *shrink, struct shrink_control *sc) { /* * __GFP_FS won't be set if we are called from ZFS code (see * kmem_flags_convert(), which removes it). To avoid a deadlock, we * don't allow evicting in this case. We return 0 rather than * SHRINK_STOP so that the shrinker logic doesn't accumulate a * deficit against us. */ if (!(sc->gfp_mask & __GFP_FS)) { return (0); } /* * This code is reached in the "direct reclaim" case, where the * kernel (outside ZFS) is trying to allocate a page, and the system * is low on memory. * * The kernel's shrinker code doesn't understand how many pages the * ARC's callback actually frees, so it may ask the ARC to shrink a * lot for one page allocation. This is problematic because it may * take a long time, thus delaying the page allocation, and because * it may force the ARC to unnecessarily shrink very small. * * Therefore, we limit the amount of data that we say is evictable, * which limits the amount that the shrinker will ask us to evict for * one page allocation attempt. * * In practice, we may be asked to shrink 4x the limit to satisfy one * page allocation, before the kernel's shrinker code gives up on us. * When that happens, we rely on the kernel code to find the pages * that we freed before invoking the OOM killer. This happens in * __alloc_pages_slowpath(), which retries and finds the pages we * freed when it calls get_page_from_freelist(). * * See also the comment above zfs_arc_shrinker_limit. */ int64_t limit = zfs_arc_shrinker_limit != 0 ? zfs_arc_shrinker_limit : INT64_MAX; return (MIN(limit, btop((int64_t)arc_evictable_memory()))); } static unsigned long arc_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc) { ASSERT((sc->gfp_mask & __GFP_FS) != 0); /* The arc is considered warm once reclaim has occurred */ if (unlikely(arc_warm == B_FALSE)) arc_warm = B_TRUE; /* * Evict the requested number of pages by reducing arc_c and waiting * for the requested amount of data to be evicted. */ arc_reduce_target_size(ptob(sc->nr_to_scan)); arc_wait_for_eviction(ptob(sc->nr_to_scan)); if (current->reclaim_state != NULL) current->reclaim_state->reclaimed_slab += sc->nr_to_scan; /* * We are experiencing memory pressure which the arc_evict_zthr was * unable to keep up with. Set arc_no_grow to briefly pause arc * growth to avoid compounding the memory pressure. */ arc_no_grow = B_TRUE; /* * When direct reclaim is observed it usually indicates a rapid * increase in memory pressure. This occurs because the kswapd * threads were unable to asynchronously keep enough free memory * available. */ if (current_is_kswapd()) { ARCSTAT_BUMP(arcstat_memory_indirect_count); } else { ARCSTAT_BUMP(arcstat_memory_direct_count); } return (sc->nr_to_scan); } SPL_SHRINKER_DECLARE(arc_shrinker, arc_shrinker_count, arc_shrinker_scan, DEFAULT_SEEKS); int arc_memory_throttle(spa_t *spa, uint64_t reserve, uint64_t txg) { uint64_t free_memory = arc_free_memory(); if (free_memory > arc_all_memory() * arc_lotsfree_percent / 100) return (0); if (txg > spa->spa_lowmem_last_txg) { spa->spa_lowmem_last_txg = txg; spa->spa_lowmem_page_load = 0; } /* * If we are in pageout, we know that memory is already tight, * the arc is already going to be evicting, so we just want to * continue to let page writes occur as quickly as possible. */ if (current_is_kswapd()) { if (spa->spa_lowmem_page_load > MAX(arc_sys_free / 4, free_memory) / 4) { DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); return (SET_ERROR(ERESTART)); } /* Note: reserve is inflated, so we deflate */ atomic_add_64(&spa->spa_lowmem_page_load, reserve / 8); return (0); } else if (spa->spa_lowmem_page_load > 0 && arc_reclaim_needed()) { /* memory is low, delay before restarting */ ARCSTAT_INCR(arcstat_memory_throttle_count, 1); DMU_TX_STAT_BUMP(dmu_tx_memory_reclaim); return (SET_ERROR(EAGAIN)); } spa->spa_lowmem_page_load = 0; return (0); } static void arc_set_sys_free(uint64_t allmem) { /* * The ARC tries to keep at least this much memory available for the * system. This gives the ARC time to shrink in response to memory * pressure, before running completely out of memory and invoking the * direct-reclaim ARC shrinker. * * This should be more than twice high_wmark_pages(), so that * arc_wait_for_eviction() will wait until at least the * high_wmark_pages() are free (see arc_evict_state_impl()). * * Note: Even when the system is very low on memory, the kernel's * shrinker code may only ask for one "batch" of pages (512KB) to be * evicted. If concurrent allocations consume these pages, there may * still be insufficient free pages, and the OOM killer takes action. * * By setting arc_sys_free large enough, and having * arc_wait_for_eviction() wait until there is at least arc_sys_free/2 * free memory, it is much less likely that concurrent allocations can * consume all the memory that was evicted before checking for * OOM. * * It's hard to iterate the zones from a linux kernel module, which * makes it difficult to determine the watermark dynamically. Instead * we compute the maximum high watermark for this system, based * on the amount of memory, assuming default parameters on Linux kernel * 5.3. */ /* * Base wmark_low is 4 * the square root of Kbytes of RAM. */ long wmark = 4 * int_sqrt(allmem/1024) * 1024; /* * Clamp to between 128K and 64MB. */ wmark = MAX(wmark, 128 * 1024); wmark = MIN(wmark, 64 * 1024 * 1024); /* * watermark_boost can increase the wmark by up to 150%. */ wmark += wmark * 150 / 100; /* * arc_sys_free needs to be more than 2x the watermark, because * arc_wait_for_eviction() waits for half of arc_sys_free. Bump this up * to 3x to ensure we're above it. */ arc_sys_free = wmark * 3 + allmem / 32; } void arc_lowmem_init(void) { uint64_t allmem = arc_all_memory(); /* * Register a shrinker to support synchronous (direct) memory * reclaim from the arc. This is done to prevent kswapd from * swapping out pages when it is preferable to shrink the arc. */ spl_register_shrinker(&arc_shrinker); arc_set_sys_free(allmem); } void arc_lowmem_fini(void) { spl_unregister_shrinker(&arc_shrinker); } int param_set_arc_long(const char *buf, zfs_kernel_param_t *kp) { int error; error = param_set_long(buf, kp); if (error < 0) return (SET_ERROR(error)); arc_tuning_update(B_TRUE); return (0); } int param_set_arc_int(const char *buf, zfs_kernel_param_t *kp) { int error; error = param_set_int(buf, kp); if (error < 0) return (SET_ERROR(error)); arc_tuning_update(B_TRUE); return (0); } #ifdef CONFIG_MEMORY_HOTPLUG /* ARGSUSED */ static int arc_hotplug_callback(struct notifier_block *self, unsigned long action, void *arg) { uint64_t allmem = arc_all_memory(); if (action != MEM_ONLINE) return (NOTIFY_OK); arc_set_limits(allmem); #ifdef __LP64__ if (zfs_dirty_data_max_max == 0) zfs_dirty_data_max_max = MIN(4ULL * 1024 * 1024 * 1024, allmem * zfs_dirty_data_max_max_percent / 100); #else if (zfs_dirty_data_max_max == 0) zfs_dirty_data_max_max = MIN(1ULL * 1024 * 1024 * 1024, allmem * zfs_dirty_data_max_max_percent / 100); #endif arc_set_sys_free(allmem); return (NOTIFY_OK); } #endif void arc_register_hotplug(void) { #ifdef CONFIG_MEMORY_HOTPLUG arc_hotplug_callback_mem_nb.notifier_call = arc_hotplug_callback; /* There is no significance to the value 100 */ arc_hotplug_callback_mem_nb.priority = 100; register_memory_notifier(&arc_hotplug_callback_mem_nb); #endif } void arc_unregister_hotplug(void) { #ifdef CONFIG_MEMORY_HOTPLUG unregister_memory_notifier(&arc_hotplug_callback_mem_nb); #endif } #else /* _KERNEL */ int64_t arc_available_memory(void) { int64_t lowest = INT64_MAX; /* Every 100 calls, free a small amount */ if (spa_get_random(100) == 0) lowest = -1024; return (lowest); } int arc_memory_throttle(spa_t *spa, uint64_t reserve, uint64_t txg) { return (0); } uint64_t arc_all_memory(void) { return (ptob(physmem) / 2); } uint64_t arc_free_memory(void) { return (spa_get_random(arc_all_memory() * 20 / 100)); } void arc_register_hotplug(void) { } void arc_unregister_hotplug(void) { } #endif /* _KERNEL */ /* * Helper function for arc_prune_async() it is responsible for safely * handling the execution of a registered arc_prune_func_t. */ static void arc_prune_task(void *ptr) { arc_prune_t *ap = (arc_prune_t *)ptr; arc_prune_func_t *func = ap->p_pfunc; if (func != NULL) func(ap->p_adjust, ap->p_private); zfs_refcount_remove(&ap->p_refcnt, func); } /* * Notify registered consumers they must drop holds on a portion of the ARC * buffered they reference. This provides a mechanism to ensure the ARC can * honor the arc_meta_limit and reclaim otherwise pinned ARC buffers. This * is analogous to dnlc_reduce_cache() but more generic. * * This operation is performed asynchronously so it may be safely called * in the context of the arc_reclaim_thread(). A reference is taken here * for each registered arc_prune_t and the arc_prune_task() is responsible * for releasing it once the registered arc_prune_func_t has completed. */ void arc_prune_async(int64_t adjust) { arc_prune_t *ap; mutex_enter(&arc_prune_mtx); for (ap = list_head(&arc_prune_list); ap != NULL; ap = list_next(&arc_prune_list, ap)) { if (zfs_refcount_count(&ap->p_refcnt) >= 2) continue; zfs_refcount_add(&ap->p_refcnt, ap->p_pfunc); ap->p_adjust = adjust; if (taskq_dispatch(arc_prune_taskq, arc_prune_task, ap, TQ_SLEEP) == TASKQID_INVALID) { zfs_refcount_remove(&ap->p_refcnt, ap->p_pfunc); continue; } ARCSTAT_BUMP(arcstat_prune); } mutex_exit(&arc_prune_mtx); } /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, shrinker_limit, INT, ZMOD_RW, "Limit on number of pages that ARC shrinker can reclaim at once"); /* END CSTYLED */ diff --git a/sys/contrib/openzfs/module/zfs/arc.c b/sys/contrib/openzfs/module/zfs/arc.c index 737904f34855..b0468159d2e6 100644 --- a/sys/contrib/openzfs/module/zfs/arc.c +++ b/sys/contrib/openzfs/module/zfs/arc.c @@ -1,10765 +1,11039 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2018, Joyent, Inc. * Copyright (c) 2011, 2020, Delphix. All rights reserved. * Copyright (c) 2014, Saso Kiselkov. All rights reserved. * Copyright (c) 2017, Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2019, loli10K . All rights reserved. * Copyright (c) 2020, George Amanakis. All rights reserved. * Copyright (c) 2019, Klara Inc. * Copyright (c) 2019, Allan Jude * Copyright (c) 2020, The FreeBSD Foundation [1] * * [1] Portions of this software were developed by Allan Jude * under sponsorship from the FreeBSD Foundation. */ /* * DVA-based Adjustable Replacement Cache * * While much of the theory of operation used here is * based on the self-tuning, low overhead replacement cache * presented by Megiddo and Modha at FAST 2003, there are some * significant differences: * * 1. The Megiddo and Modha model assumes any page is evictable. * Pages in its cache cannot be "locked" into memory. This makes * the eviction algorithm simple: evict the last page in the list. * This also make the performance characteristics easy to reason * about. Our cache is not so simple. At any given moment, some * subset of the blocks in the cache are un-evictable because we * have handed out a reference to them. Blocks are only evictable * when there are no external references active. This makes * eviction far more problematic: we choose to evict the evictable * blocks that are the "lowest" in the list. * * There are times when it is not possible to evict the requested * space. In these circumstances we are unable to adjust the cache * size. To prevent the cache growing unbounded at these times we * implement a "cache throttle" that slows the flow of new data * into the cache until we can make space available. * * 2. The Megiddo and Modha model assumes a fixed cache size. * Pages are evicted when the cache is full and there is a cache * miss. Our model has a variable sized cache. It grows with * high use, but also tries to react to memory pressure from the * operating system: decreasing its size when system memory is * tight. * * 3. The Megiddo and Modha model assumes a fixed page size. All * elements of the cache are therefore exactly the same size. So * when adjusting the cache size following a cache miss, its simply * a matter of choosing a single page to evict. In our model, we * have variable sized cache blocks (ranging from 512 bytes to * 128K bytes). We therefore choose a set of blocks to evict to make * space for a cache miss that approximates as closely as possible * the space used by the new block. * * See also: "ARC: A Self-Tuning, Low Overhead Replacement Cache" * by N. Megiddo & D. Modha, FAST 2003 */ /* * The locking model: * * A new reference to a cache buffer can be obtained in two * ways: 1) via a hash table lookup using the DVA as a key, * or 2) via one of the ARC lists. The arc_read() interface * uses method 1, while the internal ARC algorithms for * adjusting the cache use method 2. We therefore provide two * types of locks: 1) the hash table lock array, and 2) the * ARC list locks. * * Buffers do not have their own mutexes, rather they rely on the * hash table mutexes for the bulk of their protection (i.e. most * fields in the arc_buf_hdr_t are protected by these mutexes). * * buf_hash_find() returns the appropriate mutex (held) when it * locates the requested buffer in the hash table. It returns * NULL for the mutex if the buffer was not in the table. * * buf_hash_remove() expects the appropriate hash mutex to be * already held before it is invoked. * * Each ARC state also has a mutex which is used to protect the * buffer list associated with the state. When attempting to * obtain a hash table lock while holding an ARC list lock you * must use: mutex_tryenter() to avoid deadlock. Also note that * the active state mutex must be held before the ghost state mutex. * * It as also possible to register a callback which is run when the * arc_meta_limit is reached and no buffers can be safely evicted. In * this case the arc user should drop a reference on some arc buffers so * they can be reclaimed and the arc_meta_limit honored. For example, * when using the ZPL each dentry holds a references on a znode. These * dentries must be pruned before the arc buffer holding the znode can * be safely evicted. * * Note that the majority of the performance stats are manipulated * with atomic operations. * * The L2ARC uses the l2ad_mtx on each vdev for the following: * * - L2ARC buflist creation * - L2ARC buflist eviction * - L2ARC write completion, which walks L2ARC buflists * - ARC header destruction, as it removes from L2ARC buflists * - ARC header release, as it removes from L2ARC buflists */ /* * ARC operation: * * Every block that is in the ARC is tracked by an arc_buf_hdr_t structure. * This structure can point either to a block that is still in the cache or to * one that is only accessible in an L2 ARC device, or it can provide * information about a block that was recently evicted. If a block is * only accessible in the L2ARC, then the arc_buf_hdr_t only has enough * information to retrieve it from the L2ARC device. This information is * stored in the l2arc_buf_hdr_t sub-structure of the arc_buf_hdr_t. A block * that is in this state cannot access the data directly. * * Blocks that are actively being referenced or have not been evicted * are cached in the L1ARC. The L1ARC (l1arc_buf_hdr_t) is a structure within * the arc_buf_hdr_t that will point to the data block in memory. A block can * only be read by a consumer if it has an l1arc_buf_hdr_t. The L1ARC * caches data in two ways -- in a list of ARC buffers (arc_buf_t) and * also in the arc_buf_hdr_t's private physical data block pointer (b_pabd). * * The L1ARC's data pointer may or may not be uncompressed. The ARC has the * ability to store the physical data (b_pabd) associated with the DVA of the * arc_buf_hdr_t. Since the b_pabd is a copy of the on-disk physical block, * it will match its on-disk compression characteristics. This behavior can be * disabled by setting 'zfs_compressed_arc_enabled' to B_FALSE. When the * compressed ARC functionality is disabled, the b_pabd will point to an * uncompressed version of the on-disk data. * * Data in the L1ARC is not accessed by consumers of the ARC directly. Each * arc_buf_hdr_t can have multiple ARC buffers (arc_buf_t) which reference it. * Each ARC buffer (arc_buf_t) is being actively accessed by a specific ARC * consumer. The ARC will provide references to this data and will keep it * cached until it is no longer in use. The ARC caches only the L1ARC's physical * data block and will evict any arc_buf_t that is no longer referenced. The * amount of memory consumed by the arc_buf_ts' data buffers can be seen via the * "overhead_size" kstat. * * Depending on the consumer, an arc_buf_t can be requested in uncompressed or * compressed form. The typical case is that consumers will want uncompressed * data, and when that happens a new data buffer is allocated where the data is * decompressed for them to use. Currently the only consumer who wants * compressed arc_buf_t's is "zfs send", when it streams data exactly as it * exists on disk. When this happens, the arc_buf_t's data buffer is shared * with the arc_buf_hdr_t. * * Here is a diagram showing an arc_buf_hdr_t referenced by two arc_buf_t's. The * first one is owned by a compressed send consumer (and therefore references * the same compressed data buffer as the arc_buf_hdr_t) and the second could be * used by any other consumer (and has its own uncompressed copy of the data * buffer). * * arc_buf_hdr_t * +-----------+ * | fields | * | common to | * | L1- and | * | L2ARC | * +-----------+ * | l2arc_buf_hdr_t * | | * +-----------+ * | l1arc_buf_hdr_t * | | arc_buf_t * | b_buf +------------>+-----------+ arc_buf_t * | b_pabd +-+ |b_next +---->+-----------+ * +-----------+ | |-----------| |b_next +-->NULL * | |b_comp = T | +-----------+ * | |b_data +-+ |b_comp = F | * | +-----------+ | |b_data +-+ * +->+------+ | +-----------+ | * compressed | | | | * data | |<--------------+ | uncompressed * +------+ compressed, | data * shared +-->+------+ * data | | * | | * +------+ * * When a consumer reads a block, the ARC must first look to see if the * arc_buf_hdr_t is cached. If the hdr is cached then the ARC allocates a new * arc_buf_t and either copies uncompressed data into a new data buffer from an * existing uncompressed arc_buf_t, decompresses the hdr's b_pabd buffer into a * new data buffer, or shares the hdr's b_pabd buffer, depending on whether the * hdr is compressed and the desired compression characteristics of the * arc_buf_t consumer. If the arc_buf_t ends up sharing data with the * arc_buf_hdr_t and both of them are uncompressed then the arc_buf_t must be * the last buffer in the hdr's b_buf list, however a shared compressed buf can * be anywhere in the hdr's list. * * The diagram below shows an example of an uncompressed ARC hdr that is * sharing its data with an arc_buf_t (note that the shared uncompressed buf is * the last element in the buf list): * * arc_buf_hdr_t * +-----------+ * | | * | | * | | * +-----------+ * l2arc_buf_hdr_t| | * | | * +-----------+ * l1arc_buf_hdr_t| | * | | arc_buf_t (shared) * | b_buf +------------>+---------+ arc_buf_t * | | |b_next +---->+---------+ * | b_pabd +-+ |---------| |b_next +-->NULL * +-----------+ | | | +---------+ * | |b_data +-+ | | * | +---------+ | |b_data +-+ * +->+------+ | +---------+ | * | | | | * uncompressed | | | | * data +------+ | | * ^ +->+------+ | * | uncompressed | | | * | data | | | * | +------+ | * +---------------------------------+ * * Writing to the ARC requires that the ARC first discard the hdr's b_pabd * since the physical block is about to be rewritten. The new data contents * will be contained in the arc_buf_t. As the I/O pipeline performs the write, * it may compress the data before writing it to disk. The ARC will be called * with the transformed data and will bcopy the transformed on-disk block into * a newly allocated b_pabd. Writes are always done into buffers which have * either been loaned (and hence are new and don't have other readers) or * buffers which have been released (and hence have their own hdr, if there * were originally other readers of the buf's original hdr). This ensures that * the ARC only needs to update a single buf and its hdr after a write occurs. * * When the L2ARC is in use, it will also take advantage of the b_pabd. The * L2ARC will always write the contents of b_pabd to the L2ARC. This means * that when compressed ARC is enabled that the L2ARC blocks are identical * to the on-disk block in the main data pool. This provides a significant * advantage since the ARC can leverage the bp's checksum when reading from the * L2ARC to determine if the contents are valid. However, if the compressed * ARC is disabled, then the L2ARC's block must be transformed to look * like the physical block in the main data pool before comparing the * checksum and determining its validity. * * The L1ARC has a slightly different system for storing encrypted data. * Raw (encrypted + possibly compressed) data has a few subtle differences from * data that is just compressed. The biggest difference is that it is not * possible to decrypt encrypted data (or vice-versa) if the keys aren't loaded. * The other difference is that encryption cannot be treated as a suggestion. * If a caller would prefer compressed data, but they actually wind up with * uncompressed data the worst thing that could happen is there might be a * performance hit. If the caller requests encrypted data, however, we must be * sure they actually get it or else secret information could be leaked. Raw * data is stored in hdr->b_crypt_hdr.b_rabd. An encrypted header, therefore, * may have both an encrypted version and a decrypted version of its data at * once. When a caller needs a raw arc_buf_t, it is allocated and the data is * copied out of this header. To avoid complications with b_pabd, raw buffers * cannot be shared. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef _KERNEL /* set with ZFS_DEBUG=watch, to enable watchpoints on frozen buffers */ boolean_t arc_watch = B_FALSE; #endif /* * This thread's job is to keep enough free memory in the system, by * calling arc_kmem_reap_soon() plus arc_reduce_target_size(), which improves * arc_available_memory(). */ static zthr_t *arc_reap_zthr; /* * This thread's job is to keep arc_size under arc_c, by calling * arc_evict(), which improves arc_is_overflowing(). */ static zthr_t *arc_evict_zthr; static kmutex_t arc_evict_lock; static boolean_t arc_evict_needed = B_FALSE; /* * Count of bytes evicted since boot. */ static uint64_t arc_evict_count; /* * List of arc_evict_waiter_t's, representing threads waiting for the * arc_evict_count to reach specific values. */ static list_t arc_evict_waiters; /* * When arc_is_overflowing(), arc_get_data_impl() waits for this percent of * the requested amount of data to be evicted. For example, by default for * every 2KB that's evicted, 1KB of it may be "reused" by a new allocation. * Since this is above 100%, it ensures that progress is made towards getting * arc_size under arc_c. Since this is finite, it ensures that allocations * can still happen, even during the potentially long time that arc_size is * more than arc_c. */ int zfs_arc_eviction_pct = 200; /* * The number of headers to evict in arc_evict_state_impl() before * dropping the sublist lock and evicting from another sublist. A lower * value means we're more likely to evict the "correct" header (i.e. the * oldest header in the arc state), but comes with higher overhead * (i.e. more invocations of arc_evict_state_impl()). */ int zfs_arc_evict_batch_limit = 10; /* number of seconds before growing cache again */ int arc_grow_retry = 5; /* * Minimum time between calls to arc_kmem_reap_soon(). */ int arc_kmem_cache_reap_retry_ms = 1000; /* shift of arc_c for calculating overflow limit in arc_get_data_impl */ int zfs_arc_overflow_shift = 8; /* shift of arc_c for calculating both min and max arc_p */ int arc_p_min_shift = 4; /* log2(fraction of arc to reclaim) */ int arc_shrink_shift = 7; /* percent of pagecache to reclaim arc to */ #ifdef _KERNEL uint_t zfs_arc_pc_percent = 0; #endif /* * log2(fraction of ARC which must be free to allow growing). * I.e. If there is less than arc_c >> arc_no_grow_shift free memory, * when reading a new block into the ARC, we will evict an equal-sized block * from the ARC. * * This must be less than arc_shrink_shift, so that when we shrink the ARC, * we will still not allow it to grow. */ int arc_no_grow_shift = 5; /* * minimum lifespan of a prefetch block in clock ticks * (initialized in arc_init()) */ static int arc_min_prefetch_ms; static int arc_min_prescient_prefetch_ms; /* * If this percent of memory is free, don't throttle. */ int arc_lotsfree_percent = 10; /* * The arc has filled available memory and has now warmed up. */ boolean_t arc_warm; /* * These tunables are for performance analysis. */ unsigned long zfs_arc_max = 0; unsigned long zfs_arc_min = 0; unsigned long zfs_arc_meta_limit = 0; unsigned long zfs_arc_meta_min = 0; unsigned long zfs_arc_dnode_limit = 0; unsigned long zfs_arc_dnode_reduce_percent = 10; int zfs_arc_grow_retry = 0; int zfs_arc_shrink_shift = 0; int zfs_arc_p_min_shift = 0; int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */ /* * ARC dirty data constraints for arc_tempreserve_space() throttle. */ unsigned long zfs_arc_dirty_limit_percent = 50; /* total dirty data limit */ unsigned long zfs_arc_anon_limit_percent = 25; /* anon block dirty limit */ unsigned long zfs_arc_pool_dirty_percent = 20; /* each pool's anon allowance */ /* * Enable or disable compressed arc buffers. */ int zfs_compressed_arc_enabled = B_TRUE; /* * ARC will evict meta buffers that exceed arc_meta_limit. This * tunable make arc_meta_limit adjustable for different workloads. */ unsigned long zfs_arc_meta_limit_percent = 75; /* * Percentage that can be consumed by dnodes of ARC meta buffers. */ unsigned long zfs_arc_dnode_limit_percent = 10; /* * These tunables are Linux specific */ unsigned long zfs_arc_sys_free = 0; int zfs_arc_min_prefetch_ms = 0; int zfs_arc_min_prescient_prefetch_ms = 0; int zfs_arc_p_dampener_disable = 1; int zfs_arc_meta_prune = 10000; int zfs_arc_meta_strategy = ARC_STRATEGY_META_BALANCED; int zfs_arc_meta_adjust_restarts = 4096; int zfs_arc_lotsfree_percent = 10; /* The 6 states: */ arc_state_t ARC_anon; arc_state_t ARC_mru; arc_state_t ARC_mru_ghost; arc_state_t ARC_mfu; arc_state_t ARC_mfu_ghost; arc_state_t ARC_l2c_only; arc_stats_t arc_stats = { { "hits", KSTAT_DATA_UINT64 }, { "misses", KSTAT_DATA_UINT64 }, { "demand_data_hits", KSTAT_DATA_UINT64 }, { "demand_data_misses", KSTAT_DATA_UINT64 }, { "demand_metadata_hits", KSTAT_DATA_UINT64 }, { "demand_metadata_misses", KSTAT_DATA_UINT64 }, { "prefetch_data_hits", KSTAT_DATA_UINT64 }, { "prefetch_data_misses", KSTAT_DATA_UINT64 }, { "prefetch_metadata_hits", KSTAT_DATA_UINT64 }, { "prefetch_metadata_misses", KSTAT_DATA_UINT64 }, { "mru_hits", KSTAT_DATA_UINT64 }, { "mru_ghost_hits", KSTAT_DATA_UINT64 }, { "mfu_hits", KSTAT_DATA_UINT64 }, { "mfu_ghost_hits", KSTAT_DATA_UINT64 }, { "deleted", KSTAT_DATA_UINT64 }, { "mutex_miss", KSTAT_DATA_UINT64 }, { "access_skip", KSTAT_DATA_UINT64 }, { "evict_skip", KSTAT_DATA_UINT64 }, { "evict_not_enough", KSTAT_DATA_UINT64 }, { "evict_l2_cached", KSTAT_DATA_UINT64 }, { "evict_l2_eligible", KSTAT_DATA_UINT64 }, { "evict_l2_eligible_mfu", KSTAT_DATA_UINT64 }, { "evict_l2_eligible_mru", KSTAT_DATA_UINT64 }, { "evict_l2_ineligible", KSTAT_DATA_UINT64 }, { "evict_l2_skip", KSTAT_DATA_UINT64 }, { "hash_elements", KSTAT_DATA_UINT64 }, { "hash_elements_max", KSTAT_DATA_UINT64 }, { "hash_collisions", KSTAT_DATA_UINT64 }, { "hash_chains", KSTAT_DATA_UINT64 }, { "hash_chain_max", KSTAT_DATA_UINT64 }, { "p", KSTAT_DATA_UINT64 }, { "c", KSTAT_DATA_UINT64 }, { "c_min", KSTAT_DATA_UINT64 }, { "c_max", KSTAT_DATA_UINT64 }, { "size", KSTAT_DATA_UINT64 }, { "compressed_size", KSTAT_DATA_UINT64 }, { "uncompressed_size", KSTAT_DATA_UINT64 }, { "overhead_size", KSTAT_DATA_UINT64 }, { "hdr_size", KSTAT_DATA_UINT64 }, { "data_size", KSTAT_DATA_UINT64 }, { "metadata_size", KSTAT_DATA_UINT64 }, { "dbuf_size", KSTAT_DATA_UINT64 }, { "dnode_size", KSTAT_DATA_UINT64 }, { "bonus_size", KSTAT_DATA_UINT64 }, #if defined(COMPAT_FREEBSD11) { "other_size", KSTAT_DATA_UINT64 }, #endif { "anon_size", KSTAT_DATA_UINT64 }, { "anon_evictable_data", KSTAT_DATA_UINT64 }, { "anon_evictable_metadata", KSTAT_DATA_UINT64 }, { "mru_size", KSTAT_DATA_UINT64 }, { "mru_evictable_data", KSTAT_DATA_UINT64 }, { "mru_evictable_metadata", KSTAT_DATA_UINT64 }, { "mru_ghost_size", KSTAT_DATA_UINT64 }, { "mru_ghost_evictable_data", KSTAT_DATA_UINT64 }, { "mru_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, { "mfu_size", KSTAT_DATA_UINT64 }, { "mfu_evictable_data", KSTAT_DATA_UINT64 }, { "mfu_evictable_metadata", KSTAT_DATA_UINT64 }, { "mfu_ghost_size", KSTAT_DATA_UINT64 }, { "mfu_ghost_evictable_data", KSTAT_DATA_UINT64 }, { "mfu_ghost_evictable_metadata", KSTAT_DATA_UINT64 }, { "l2_hits", KSTAT_DATA_UINT64 }, { "l2_misses", KSTAT_DATA_UINT64 }, { "l2_prefetch_asize", KSTAT_DATA_UINT64 }, { "l2_mru_asize", KSTAT_DATA_UINT64 }, { "l2_mfu_asize", KSTAT_DATA_UINT64 }, { "l2_bufc_data_asize", KSTAT_DATA_UINT64 }, { "l2_bufc_metadata_asize", KSTAT_DATA_UINT64 }, { "l2_feeds", KSTAT_DATA_UINT64 }, { "l2_rw_clash", KSTAT_DATA_UINT64 }, { "l2_read_bytes", KSTAT_DATA_UINT64 }, { "l2_write_bytes", KSTAT_DATA_UINT64 }, { "l2_writes_sent", KSTAT_DATA_UINT64 }, { "l2_writes_done", KSTAT_DATA_UINT64 }, { "l2_writes_error", KSTAT_DATA_UINT64 }, { "l2_writes_lock_retry", KSTAT_DATA_UINT64 }, { "l2_evict_lock_retry", KSTAT_DATA_UINT64 }, { "l2_evict_reading", KSTAT_DATA_UINT64 }, { "l2_evict_l1cached", KSTAT_DATA_UINT64 }, { "l2_free_on_write", KSTAT_DATA_UINT64 }, { "l2_abort_lowmem", KSTAT_DATA_UINT64 }, { "l2_cksum_bad", KSTAT_DATA_UINT64 }, { "l2_io_error", KSTAT_DATA_UINT64 }, { "l2_size", KSTAT_DATA_UINT64 }, { "l2_asize", KSTAT_DATA_UINT64 }, { "l2_hdr_size", KSTAT_DATA_UINT64 }, { "l2_log_blk_writes", KSTAT_DATA_UINT64 }, { "l2_log_blk_avg_asize", KSTAT_DATA_UINT64 }, { "l2_log_blk_asize", KSTAT_DATA_UINT64 }, { "l2_log_blk_count", KSTAT_DATA_UINT64 }, { "l2_data_to_meta_ratio", KSTAT_DATA_UINT64 }, { "l2_rebuild_success", KSTAT_DATA_UINT64 }, { "l2_rebuild_unsupported", KSTAT_DATA_UINT64 }, { "l2_rebuild_io_errors", KSTAT_DATA_UINT64 }, { "l2_rebuild_dh_errors", KSTAT_DATA_UINT64 }, { "l2_rebuild_cksum_lb_errors", KSTAT_DATA_UINT64 }, { "l2_rebuild_lowmem", KSTAT_DATA_UINT64 }, { "l2_rebuild_size", KSTAT_DATA_UINT64 }, { "l2_rebuild_asize", KSTAT_DATA_UINT64 }, { "l2_rebuild_bufs", KSTAT_DATA_UINT64 }, { "l2_rebuild_bufs_precached", KSTAT_DATA_UINT64 }, { "l2_rebuild_log_blks", KSTAT_DATA_UINT64 }, { "memory_throttle_count", KSTAT_DATA_UINT64 }, { "memory_direct_count", KSTAT_DATA_UINT64 }, { "memory_indirect_count", KSTAT_DATA_UINT64 }, { "memory_all_bytes", KSTAT_DATA_UINT64 }, { "memory_free_bytes", KSTAT_DATA_UINT64 }, { "memory_available_bytes", KSTAT_DATA_INT64 }, { "arc_no_grow", KSTAT_DATA_UINT64 }, { "arc_tempreserve", KSTAT_DATA_UINT64 }, { "arc_loaned_bytes", KSTAT_DATA_UINT64 }, { "arc_prune", KSTAT_DATA_UINT64 }, { "arc_meta_used", KSTAT_DATA_UINT64 }, { "arc_meta_limit", KSTAT_DATA_UINT64 }, { "arc_dnode_limit", KSTAT_DATA_UINT64 }, { "arc_meta_max", KSTAT_DATA_UINT64 }, { "arc_meta_min", KSTAT_DATA_UINT64 }, { "async_upgrade_sync", KSTAT_DATA_UINT64 }, { "demand_hit_predictive_prefetch", KSTAT_DATA_UINT64 }, { "demand_hit_prescient_prefetch", KSTAT_DATA_UINT64 }, { "arc_need_free", KSTAT_DATA_UINT64 }, { "arc_sys_free", KSTAT_DATA_UINT64 }, { "arc_raw_size", KSTAT_DATA_UINT64 }, { "cached_only_in_progress", KSTAT_DATA_UINT64 }, { "abd_chunk_waste_size", KSTAT_DATA_UINT64 }, }; +arc_sums_t arc_sums; + #define ARCSTAT_MAX(stat, val) { \ uint64_t m; \ while ((val) > (m = arc_stats.stat.value.ui64) && \ (m != atomic_cas_64(&arc_stats.stat.value.ui64, m, (val)))) \ continue; \ } -#define ARCSTAT_MAXSTAT(stat) \ - ARCSTAT_MAX(stat##_max, arc_stats.stat.value.ui64) - /* * We define a macro to allow ARC hits/misses to be easily broken down by * two separate conditions, giving a total of four different subtypes for * each of hits and misses (so eight statistics total). */ #define ARCSTAT_CONDSTAT(cond1, stat1, notstat1, cond2, stat2, notstat2, stat) \ if (cond1) { \ if (cond2) { \ ARCSTAT_BUMP(arcstat_##stat1##_##stat2##_##stat); \ } else { \ ARCSTAT_BUMP(arcstat_##stat1##_##notstat2##_##stat); \ } \ } else { \ if (cond2) { \ ARCSTAT_BUMP(arcstat_##notstat1##_##stat2##_##stat); \ } else { \ ARCSTAT_BUMP(arcstat_##notstat1##_##notstat2##_##stat);\ } \ } /* * This macro allows us to use kstats as floating averages. Each time we * update this kstat, we first factor it and the update value by * ARCSTAT_AVG_FACTOR to shrink the new value's contribution to the overall * average. This macro assumes that integer loads and stores are atomic, but * is not safe for multiple writers updating the kstat in parallel (only the * last writer's update will remain). */ #define ARCSTAT_F_AVG_FACTOR 3 #define ARCSTAT_F_AVG(stat, value) \ do { \ uint64_t x = ARCSTAT(stat); \ x = x - x / ARCSTAT_F_AVG_FACTOR + \ (value) / ARCSTAT_F_AVG_FACTOR; \ ARCSTAT(stat) = x; \ _NOTE(CONSTCOND) \ } while (0) kstat_t *arc_ksp; static arc_state_t *arc_anon; static arc_state_t *arc_mru_ghost; static arc_state_t *arc_mfu_ghost; static arc_state_t *arc_l2c_only; arc_state_t *arc_mru; arc_state_t *arc_mfu; /* * There are several ARC variables that are critical to export as kstats -- * but we don't want to have to grovel around in the kstat whenever we wish to * manipulate them. For these variables, we therefore define them to be in * terms of the statistic variable. This assures that we are not introducing * the possibility of inconsistency by having shadow copies of the variables, * while still allowing the code to be readable. */ #define arc_tempreserve ARCSTAT(arcstat_tempreserve) #define arc_loaned_bytes ARCSTAT(arcstat_loaned_bytes) #define arc_meta_limit ARCSTAT(arcstat_meta_limit) /* max size for metadata */ /* max size for dnodes */ #define arc_dnode_size_limit ARCSTAT(arcstat_dnode_limit) #define arc_meta_min ARCSTAT(arcstat_meta_min) /* min size for metadata */ -#define arc_meta_max ARCSTAT(arcstat_meta_max) /* max size of metadata */ #define arc_need_free ARCSTAT(arcstat_need_free) /* waiting to be evicted */ -/* size of all b_rabd's in entire arc */ -#define arc_raw_size ARCSTAT(arcstat_raw_size) -/* compressed size of entire arc */ -#define arc_compressed_size ARCSTAT(arcstat_compressed_size) -/* uncompressed size of entire arc */ -#define arc_uncompressed_size ARCSTAT(arcstat_uncompressed_size) -/* number of bytes in the arc from arc_buf_t's */ -#define arc_overhead_size ARCSTAT(arcstat_overhead_size) - -/* - * There are also some ARC variables that we want to export, but that are - * updated so often that having the canonical representation be the statistic - * variable causes a performance bottleneck. We want to use aggsum_t's for these - * instead, but still be able to export the kstat in the same way as before. - * The solution is to always use the aggsum version, except in the kstat update - * callback. - */ -aggsum_t arc_size; -aggsum_t arc_meta_used; -wmsum_t astat_data_size; -wmsum_t astat_metadata_size; -wmsum_t astat_dbuf_size; -aggsum_t astat_dnode_size; -wmsum_t astat_bonus_size; -wmsum_t astat_hdr_size; -aggsum_t astat_l2_hdr_size; -wmsum_t astat_abd_chunk_waste_size; - hrtime_t arc_growtime; list_t arc_prune_list; kmutex_t arc_prune_mtx; taskq_t *arc_prune_taskq; #define GHOST_STATE(state) \ ((state) == arc_mru_ghost || (state) == arc_mfu_ghost || \ (state) == arc_l2c_only) #define HDR_IN_HASH_TABLE(hdr) ((hdr)->b_flags & ARC_FLAG_IN_HASH_TABLE) #define HDR_IO_IN_PROGRESS(hdr) ((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) #define HDR_IO_ERROR(hdr) ((hdr)->b_flags & ARC_FLAG_IO_ERROR) #define HDR_PREFETCH(hdr) ((hdr)->b_flags & ARC_FLAG_PREFETCH) #define HDR_PRESCIENT_PREFETCH(hdr) \ ((hdr)->b_flags & ARC_FLAG_PRESCIENT_PREFETCH) #define HDR_COMPRESSION_ENABLED(hdr) \ ((hdr)->b_flags & ARC_FLAG_COMPRESSED_ARC) #define HDR_L2CACHE(hdr) ((hdr)->b_flags & ARC_FLAG_L2CACHE) #define HDR_L2_READING(hdr) \ (((hdr)->b_flags & ARC_FLAG_IO_IN_PROGRESS) && \ ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR)) #define HDR_L2_WRITING(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITING) #define HDR_L2_EVICTED(hdr) ((hdr)->b_flags & ARC_FLAG_L2_EVICTED) #define HDR_L2_WRITE_HEAD(hdr) ((hdr)->b_flags & ARC_FLAG_L2_WRITE_HEAD) #define HDR_PROTECTED(hdr) ((hdr)->b_flags & ARC_FLAG_PROTECTED) #define HDR_NOAUTH(hdr) ((hdr)->b_flags & ARC_FLAG_NOAUTH) #define HDR_SHARED_DATA(hdr) ((hdr)->b_flags & ARC_FLAG_SHARED_DATA) #define HDR_ISTYPE_METADATA(hdr) \ ((hdr)->b_flags & ARC_FLAG_BUFC_METADATA) #define HDR_ISTYPE_DATA(hdr) (!HDR_ISTYPE_METADATA(hdr)) #define HDR_HAS_L1HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L1HDR) #define HDR_HAS_L2HDR(hdr) ((hdr)->b_flags & ARC_FLAG_HAS_L2HDR) #define HDR_HAS_RABD(hdr) \ (HDR_HAS_L1HDR(hdr) && HDR_PROTECTED(hdr) && \ (hdr)->b_crypt_hdr.b_rabd != NULL) #define HDR_ENCRYPTED(hdr) \ (HDR_PROTECTED(hdr) && DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot)) #define HDR_AUTHENTICATED(hdr) \ (HDR_PROTECTED(hdr) && !DMU_OT_IS_ENCRYPTED((hdr)->b_crypt_hdr.b_ot)) /* For storing compression mode in b_flags */ #define HDR_COMPRESS_OFFSET (highbit64(ARC_FLAG_COMPRESS_0) - 1) #define HDR_GET_COMPRESS(hdr) ((enum zio_compress)BF32_GET((hdr)->b_flags, \ HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS)) #define HDR_SET_COMPRESS(hdr, cmp) BF32_SET((hdr)->b_flags, \ HDR_COMPRESS_OFFSET, SPA_COMPRESSBITS, (cmp)); #define ARC_BUF_LAST(buf) ((buf)->b_next == NULL) #define ARC_BUF_SHARED(buf) ((buf)->b_flags & ARC_BUF_FLAG_SHARED) #define ARC_BUF_COMPRESSED(buf) ((buf)->b_flags & ARC_BUF_FLAG_COMPRESSED) #define ARC_BUF_ENCRYPTED(buf) ((buf)->b_flags & ARC_BUF_FLAG_ENCRYPTED) /* * Other sizes */ #define HDR_FULL_CRYPT_SIZE ((int64_t)sizeof (arc_buf_hdr_t)) #define HDR_FULL_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_crypt_hdr)) #define HDR_L2ONLY_SIZE ((int64_t)offsetof(arc_buf_hdr_t, b_l1hdr)) /* * Hash table routines */ #define HT_LOCK_ALIGN 64 #define HT_LOCK_PAD (P2NPHASE(sizeof (kmutex_t), (HT_LOCK_ALIGN))) struct ht_lock { kmutex_t ht_lock; #ifdef _KERNEL unsigned char pad[HT_LOCK_PAD]; #endif }; #define BUF_LOCKS 8192 typedef struct buf_hash_table { uint64_t ht_mask; arc_buf_hdr_t **ht_table; struct ht_lock ht_locks[BUF_LOCKS]; } buf_hash_table_t; static buf_hash_table_t buf_hash_table; #define BUF_HASH_INDEX(spa, dva, birth) \ (buf_hash(spa, dva, birth) & buf_hash_table.ht_mask) #define BUF_HASH_LOCK_NTRY(idx) (buf_hash_table.ht_locks[idx & (BUF_LOCKS-1)]) #define BUF_HASH_LOCK(idx) (&(BUF_HASH_LOCK_NTRY(idx).ht_lock)) #define HDR_LOCK(hdr) \ (BUF_HASH_LOCK(BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth))) uint64_t zfs_crc64_table[256]; /* * Level 2 ARC */ #define L2ARC_WRITE_SIZE (8 * 1024 * 1024) /* initial write max */ #define L2ARC_HEADROOM 2 /* num of writes */ /* * If we discover during ARC scan any buffers to be compressed, we boost * our headroom for the next scanning cycle by this percentage multiple. */ #define L2ARC_HEADROOM_BOOST 200 #define L2ARC_FEED_SECS 1 /* caching interval secs */ #define L2ARC_FEED_MIN_MS 200 /* min caching interval ms */ /* * We can feed L2ARC from two states of ARC buffers, mru and mfu, * and each of the state has two types: data and metadata. */ #define L2ARC_FEED_TYPES 4 -#define l2arc_writes_sent ARCSTAT(arcstat_l2_writes_sent) -#define l2arc_writes_done ARCSTAT(arcstat_l2_writes_done) - /* L2ARC Performance Tunables */ unsigned long l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */ unsigned long l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */ unsigned long l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */ unsigned long l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; unsigned long l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ unsigned long l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */ int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ int l2arc_feed_again = B_TRUE; /* turbo warmup */ int l2arc_norw = B_FALSE; /* no reads during writes */ int l2arc_meta_percent = 33; /* limit on headers size */ /* * L2ARC Internals */ static list_t L2ARC_dev_list; /* device list */ static list_t *l2arc_dev_list; /* device list pointer */ static kmutex_t l2arc_dev_mtx; /* device list mutex */ static l2arc_dev_t *l2arc_dev_last; /* last device used */ static list_t L2ARC_free_on_write; /* free after write buf list */ static list_t *l2arc_free_on_write; /* free after write list ptr */ static kmutex_t l2arc_free_on_write_mtx; /* mutex for list */ static uint64_t l2arc_ndev; /* number of devices */ typedef struct l2arc_read_callback { arc_buf_hdr_t *l2rcb_hdr; /* read header */ blkptr_t l2rcb_bp; /* original blkptr */ zbookmark_phys_t l2rcb_zb; /* original bookmark */ int l2rcb_flags; /* original flags */ abd_t *l2rcb_abd; /* temporary buffer */ } l2arc_read_callback_t; typedef struct l2arc_data_free { /* protected by l2arc_free_on_write_mtx */ abd_t *l2df_abd; size_t l2df_size; arc_buf_contents_t l2df_type; list_node_t l2df_list_node; } l2arc_data_free_t; typedef enum arc_fill_flags { ARC_FILL_LOCKED = 1 << 0, /* hdr lock is held */ ARC_FILL_COMPRESSED = 1 << 1, /* fill with compressed data */ ARC_FILL_ENCRYPTED = 1 << 2, /* fill with encrypted data */ ARC_FILL_NOAUTH = 1 << 3, /* don't attempt to authenticate */ ARC_FILL_IN_PLACE = 1 << 4 /* fill in place (special case) */ } arc_fill_flags_t; static kmutex_t l2arc_feed_thr_lock; static kcondvar_t l2arc_feed_thr_cv; static uint8_t l2arc_thread_exit; static kmutex_t l2arc_rebuild_thr_lock; static kcondvar_t l2arc_rebuild_thr_cv; enum arc_hdr_alloc_flags { ARC_HDR_ALLOC_RDATA = 0x1, ARC_HDR_DO_ADAPT = 0x2, }; static abd_t *arc_get_data_abd(arc_buf_hdr_t *, uint64_t, void *, boolean_t); static void *arc_get_data_buf(arc_buf_hdr_t *, uint64_t, void *); static void arc_get_data_impl(arc_buf_hdr_t *, uint64_t, void *, boolean_t); static void arc_free_data_abd(arc_buf_hdr_t *, abd_t *, uint64_t, void *); static void arc_free_data_buf(arc_buf_hdr_t *, void *, uint64_t, void *); static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag); static void arc_hdr_free_abd(arc_buf_hdr_t *, boolean_t); static void arc_hdr_alloc_abd(arc_buf_hdr_t *, int); static void arc_access(arc_buf_hdr_t *, kmutex_t *); static void arc_buf_watch(arc_buf_t *); static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *); static uint32_t arc_bufc_to_flags(arc_buf_contents_t); static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags); static boolean_t l2arc_write_eligible(uint64_t, arc_buf_hdr_t *); static void l2arc_read_done(zio_t *); static void l2arc_do_free_on_write(void); static void l2arc_hdr_arcstats_update(arc_buf_hdr_t *hdr, boolean_t incr, boolean_t state_only); #define l2arc_hdr_arcstats_increment(hdr) \ l2arc_hdr_arcstats_update((hdr), B_TRUE, B_FALSE) #define l2arc_hdr_arcstats_decrement(hdr) \ l2arc_hdr_arcstats_update((hdr), B_FALSE, B_FALSE) #define l2arc_hdr_arcstats_increment_state(hdr) \ l2arc_hdr_arcstats_update((hdr), B_TRUE, B_TRUE) #define l2arc_hdr_arcstats_decrement_state(hdr) \ l2arc_hdr_arcstats_update((hdr), B_FALSE, B_TRUE) /* * l2arc_mfuonly : A ZFS module parameter that controls whether only MFU * metadata and data are cached from ARC into L2ARC. */ int l2arc_mfuonly = 0; /* * L2ARC TRIM * l2arc_trim_ahead : A ZFS module parameter that controls how much ahead of * the current write size (l2arc_write_max) we should TRIM if we * have filled the device. It is defined as a percentage of the * write size. If set to 100 we trim twice the space required to * accommodate upcoming writes. A minimum of 64MB will be trimmed. * It also enables TRIM of the whole L2ARC device upon creation or * addition to an existing pool or if the header of the device is * invalid upon importing a pool or onlining a cache device. The * default is 0, which disables TRIM on L2ARC altogether as it can * put significant stress on the underlying storage devices. This * will vary depending of how well the specific device handles * these commands. */ unsigned long l2arc_trim_ahead = 0; /* * Performance tuning of L2ARC persistence: * * l2arc_rebuild_enabled : A ZFS module parameter that controls whether adding * an L2ARC device (either at pool import or later) will attempt * to rebuild L2ARC buffer contents. * l2arc_rebuild_blocks_min_l2size : A ZFS module parameter that controls * whether log blocks are written to the L2ARC device. If the L2ARC * device is less than 1GB, the amount of data l2arc_evict() * evicts is significant compared to the amount of restored L2ARC * data. In this case do not write log blocks in L2ARC in order * not to waste space. */ int l2arc_rebuild_enabled = B_TRUE; unsigned long l2arc_rebuild_blocks_min_l2size = 1024 * 1024 * 1024; /* L2ARC persistence rebuild control routines. */ void l2arc_rebuild_vdev(vdev_t *vd, boolean_t reopen); static void l2arc_dev_rebuild_thread(void *arg); static int l2arc_rebuild(l2arc_dev_t *dev); /* L2ARC persistence read I/O routines. */ static int l2arc_dev_hdr_read(l2arc_dev_t *dev); static int l2arc_log_blk_read(l2arc_dev_t *dev, const l2arc_log_blkptr_t *this_lp, const l2arc_log_blkptr_t *next_lp, l2arc_log_blk_phys_t *this_lb, l2arc_log_blk_phys_t *next_lb, zio_t *this_io, zio_t **next_io); static zio_t *l2arc_log_blk_fetch(vdev_t *vd, const l2arc_log_blkptr_t *lp, l2arc_log_blk_phys_t *lb); static void l2arc_log_blk_fetch_abort(zio_t *zio); /* L2ARC persistence block restoration routines. */ static void l2arc_log_blk_restore(l2arc_dev_t *dev, const l2arc_log_blk_phys_t *lb, uint64_t lb_asize); static void l2arc_hdr_restore(const l2arc_log_ent_phys_t *le, l2arc_dev_t *dev); /* L2ARC persistence write I/O routines. */ static void l2arc_log_blk_commit(l2arc_dev_t *dev, zio_t *pio, l2arc_write_callback_t *cb); /* L2ARC persistence auxiliary routines. */ boolean_t l2arc_log_blkptr_valid(l2arc_dev_t *dev, const l2arc_log_blkptr_t *lbp); static boolean_t l2arc_log_blk_insert(l2arc_dev_t *dev, const arc_buf_hdr_t *ab); boolean_t l2arc_range_check_overlap(uint64_t bottom, uint64_t top, uint64_t check); static void l2arc_blk_fetch_done(zio_t *zio); static inline uint64_t l2arc_log_blk_overhead(uint64_t write_sz, l2arc_dev_t *dev); /* * We use Cityhash for this. It's fast, and has good hash properties without * requiring any large static buffers. */ static uint64_t buf_hash(uint64_t spa, const dva_t *dva, uint64_t birth) { return (cityhash4(spa, dva->dva_word[0], dva->dva_word[1], birth)); } #define HDR_EMPTY(hdr) \ ((hdr)->b_dva.dva_word[0] == 0 && \ (hdr)->b_dva.dva_word[1] == 0) #define HDR_EMPTY_OR_LOCKED(hdr) \ (HDR_EMPTY(hdr) || MUTEX_HELD(HDR_LOCK(hdr))) #define HDR_EQUAL(spa, dva, birth, hdr) \ ((hdr)->b_dva.dva_word[0] == (dva)->dva_word[0]) && \ ((hdr)->b_dva.dva_word[1] == (dva)->dva_word[1]) && \ ((hdr)->b_birth == birth) && ((hdr)->b_spa == spa) static void buf_discard_identity(arc_buf_hdr_t *hdr) { hdr->b_dva.dva_word[0] = 0; hdr->b_dva.dva_word[1] = 0; hdr->b_birth = 0; } static arc_buf_hdr_t * buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp) { const dva_t *dva = BP_IDENTITY(bp); uint64_t birth = BP_PHYSICAL_BIRTH(bp); uint64_t idx = BUF_HASH_INDEX(spa, dva, birth); kmutex_t *hash_lock = BUF_HASH_LOCK(idx); arc_buf_hdr_t *hdr; mutex_enter(hash_lock); for (hdr = buf_hash_table.ht_table[idx]; hdr != NULL; hdr = hdr->b_hash_next) { if (HDR_EQUAL(spa, dva, birth, hdr)) { *lockp = hash_lock; return (hdr); } } mutex_exit(hash_lock); *lockp = NULL; return (NULL); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. * If lockp == NULL, the caller is assumed to already hold the hash lock. */ static arc_buf_hdr_t * buf_hash_insert(arc_buf_hdr_t *hdr, kmutex_t **lockp) { uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); kmutex_t *hash_lock = BUF_HASH_LOCK(idx); arc_buf_hdr_t *fhdr; uint32_t i; ASSERT(!DVA_IS_EMPTY(&hdr->b_dva)); ASSERT(hdr->b_birth != 0); ASSERT(!HDR_IN_HASH_TABLE(hdr)); if (lockp != NULL) { *lockp = hash_lock; mutex_enter(hash_lock); } else { ASSERT(MUTEX_HELD(hash_lock)); } for (fhdr = buf_hash_table.ht_table[idx], i = 0; fhdr != NULL; fhdr = fhdr->b_hash_next, i++) { if (HDR_EQUAL(hdr->b_spa, &hdr->b_dva, hdr->b_birth, fhdr)) return (fhdr); } hdr->b_hash_next = buf_hash_table.ht_table[idx]; buf_hash_table.ht_table[idx] = hdr; arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); /* collect some hash table performance data */ if (i > 0) { ARCSTAT_BUMP(arcstat_hash_collisions); if (i == 1) ARCSTAT_BUMP(arcstat_hash_chains); ARCSTAT_MAX(arcstat_hash_chain_max, i); } - - ARCSTAT_BUMP(arcstat_hash_elements); - ARCSTAT_MAXSTAT(arcstat_hash_elements); + uint64_t he = atomic_inc_64_nv( + &arc_stats.arcstat_hash_elements.value.ui64); + ARCSTAT_MAX(arcstat_hash_elements_max, he); return (NULL); } static void buf_hash_remove(arc_buf_hdr_t *hdr) { arc_buf_hdr_t *fhdr, **hdrp; uint64_t idx = BUF_HASH_INDEX(hdr->b_spa, &hdr->b_dva, hdr->b_birth); ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx))); ASSERT(HDR_IN_HASH_TABLE(hdr)); hdrp = &buf_hash_table.ht_table[idx]; while ((fhdr = *hdrp) != hdr) { ASSERT3P(fhdr, !=, NULL); hdrp = &fhdr->b_hash_next; } *hdrp = hdr->b_hash_next; hdr->b_hash_next = NULL; arc_hdr_clear_flags(hdr, ARC_FLAG_IN_HASH_TABLE); /* collect some hash table performance data */ - ARCSTAT_BUMPDOWN(arcstat_hash_elements); + atomic_dec_64(&arc_stats.arcstat_hash_elements.value.ui64); if (buf_hash_table.ht_table[idx] && buf_hash_table.ht_table[idx]->b_hash_next == NULL) ARCSTAT_BUMPDOWN(arcstat_hash_chains); } /* * Global data structures and functions for the buf kmem cache. */ static kmem_cache_t *hdr_full_cache; static kmem_cache_t *hdr_full_crypt_cache; static kmem_cache_t *hdr_l2only_cache; static kmem_cache_t *buf_cache; static void buf_fini(void) { int i; #if defined(_KERNEL) /* * Large allocations which do not require contiguous pages * should be using vmem_free() in the linux kernel\ */ vmem_free(buf_hash_table.ht_table, (buf_hash_table.ht_mask + 1) * sizeof (void *)); #else kmem_free(buf_hash_table.ht_table, (buf_hash_table.ht_mask + 1) * sizeof (void *)); #endif for (i = 0; i < BUF_LOCKS; i++) mutex_destroy(&buf_hash_table.ht_locks[i].ht_lock); kmem_cache_destroy(hdr_full_cache); kmem_cache_destroy(hdr_full_crypt_cache); kmem_cache_destroy(hdr_l2only_cache); kmem_cache_destroy(buf_cache); } /* * Constructor callback - called when the cache is empty * and a new buf is requested. */ /* ARGSUSED */ static int hdr_full_cons(void *vbuf, void *unused, int kmflag) { arc_buf_hdr_t *hdr = vbuf; bzero(hdr, HDR_FULL_SIZE); hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; cv_init(&hdr->b_l1hdr.b_cv, NULL, CV_DEFAULT, NULL); zfs_refcount_create(&hdr->b_l1hdr.b_refcnt); mutex_init(&hdr->b_l1hdr.b_freeze_lock, NULL, MUTEX_DEFAULT, NULL); list_link_init(&hdr->b_l1hdr.b_arc_node); list_link_init(&hdr->b_l2hdr.b_l2node); multilist_link_init(&hdr->b_l1hdr.b_arc_node); arc_space_consume(HDR_FULL_SIZE, ARC_SPACE_HDRS); return (0); } /* ARGSUSED */ static int hdr_full_crypt_cons(void *vbuf, void *unused, int kmflag) { arc_buf_hdr_t *hdr = vbuf; hdr_full_cons(vbuf, unused, kmflag); bzero(&hdr->b_crypt_hdr, sizeof (hdr->b_crypt_hdr)); arc_space_consume(sizeof (hdr->b_crypt_hdr), ARC_SPACE_HDRS); return (0); } /* ARGSUSED */ static int hdr_l2only_cons(void *vbuf, void *unused, int kmflag) { arc_buf_hdr_t *hdr = vbuf; bzero(hdr, HDR_L2ONLY_SIZE); arc_space_consume(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); return (0); } /* ARGSUSED */ static int buf_cons(void *vbuf, void *unused, int kmflag) { arc_buf_t *buf = vbuf; bzero(buf, sizeof (arc_buf_t)); mutex_init(&buf->b_evict_lock, NULL, MUTEX_DEFAULT, NULL); arc_space_consume(sizeof (arc_buf_t), ARC_SPACE_HDRS); return (0); } /* * Destructor callback - called when a cached buf is * no longer required. */ /* ARGSUSED */ static void hdr_full_dest(void *vbuf, void *unused) { arc_buf_hdr_t *hdr = vbuf; ASSERT(HDR_EMPTY(hdr)); cv_destroy(&hdr->b_l1hdr.b_cv); zfs_refcount_destroy(&hdr->b_l1hdr.b_refcnt); mutex_destroy(&hdr->b_l1hdr.b_freeze_lock); ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); arc_space_return(HDR_FULL_SIZE, ARC_SPACE_HDRS); } /* ARGSUSED */ static void hdr_full_crypt_dest(void *vbuf, void *unused) { arc_buf_hdr_t *hdr = vbuf; hdr_full_dest(vbuf, unused); arc_space_return(sizeof (hdr->b_crypt_hdr), ARC_SPACE_HDRS); } /* ARGSUSED */ static void hdr_l2only_dest(void *vbuf, void *unused) { arc_buf_hdr_t *hdr __maybe_unused = vbuf; ASSERT(HDR_EMPTY(hdr)); arc_space_return(HDR_L2ONLY_SIZE, ARC_SPACE_L2HDRS); } /* ARGSUSED */ static void buf_dest(void *vbuf, void *unused) { arc_buf_t *buf = vbuf; mutex_destroy(&buf->b_evict_lock); arc_space_return(sizeof (arc_buf_t), ARC_SPACE_HDRS); } static void buf_init(void) { uint64_t *ct = NULL; uint64_t hsize = 1ULL << 12; int i, j; /* * The hash table is big enough to fill all of physical memory * with an average block size of zfs_arc_average_blocksize (default 8K). * By default, the table will take up * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). */ while (hsize * zfs_arc_average_blocksize < arc_all_memory()) hsize <<= 1; retry: buf_hash_table.ht_mask = hsize - 1; #if defined(_KERNEL) /* * Large allocations which do not require contiguous pages * should be using vmem_alloc() in the linux kernel */ buf_hash_table.ht_table = vmem_zalloc(hsize * sizeof (void*), KM_SLEEP); #else buf_hash_table.ht_table = kmem_zalloc(hsize * sizeof (void*), KM_NOSLEEP); #endif if (buf_hash_table.ht_table == NULL) { ASSERT(hsize > (1ULL << 8)); hsize >>= 1; goto retry; } hdr_full_cache = kmem_cache_create("arc_buf_hdr_t_full", HDR_FULL_SIZE, 0, hdr_full_cons, hdr_full_dest, NULL, NULL, NULL, 0); hdr_full_crypt_cache = kmem_cache_create("arc_buf_hdr_t_full_crypt", HDR_FULL_CRYPT_SIZE, 0, hdr_full_crypt_cons, hdr_full_crypt_dest, NULL, NULL, NULL, 0); hdr_l2only_cache = kmem_cache_create("arc_buf_hdr_t_l2only", HDR_L2ONLY_SIZE, 0, hdr_l2only_cons, hdr_l2only_dest, NULL, NULL, NULL, 0); buf_cache = kmem_cache_create("arc_buf_t", sizeof (arc_buf_t), 0, buf_cons, buf_dest, NULL, NULL, NULL, 0); for (i = 0; i < 256; i++) for (ct = zfs_crc64_table + i, *ct = i, j = 8; j > 0; j--) *ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY); for (i = 0; i < BUF_LOCKS; i++) { mutex_init(&buf_hash_table.ht_locks[i].ht_lock, NULL, MUTEX_DEFAULT, NULL); } } #define ARC_MINTIME (hz>>4) /* 62 ms */ /* * This is the size that the buf occupies in memory. If the buf is compressed, * it will correspond to the compressed size. You should use this method of * getting the buf size unless you explicitly need the logical size. */ uint64_t arc_buf_size(arc_buf_t *buf) { return (ARC_BUF_COMPRESSED(buf) ? HDR_GET_PSIZE(buf->b_hdr) : HDR_GET_LSIZE(buf->b_hdr)); } uint64_t arc_buf_lsize(arc_buf_t *buf) { return (HDR_GET_LSIZE(buf->b_hdr)); } /* * This function will return B_TRUE if the buffer is encrypted in memory. * This buffer can be decrypted by calling arc_untransform(). */ boolean_t arc_is_encrypted(arc_buf_t *buf) { return (ARC_BUF_ENCRYPTED(buf) != 0); } /* * Returns B_TRUE if the buffer represents data that has not had its MAC * verified yet. */ boolean_t arc_is_unauthenticated(arc_buf_t *buf) { return (HDR_NOAUTH(buf->b_hdr) != 0); } void arc_get_raw_params(arc_buf_t *buf, boolean_t *byteorder, uint8_t *salt, uint8_t *iv, uint8_t *mac) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT(HDR_PROTECTED(hdr)); bcopy(hdr->b_crypt_hdr.b_salt, salt, ZIO_DATA_SALT_LEN); bcopy(hdr->b_crypt_hdr.b_iv, iv, ZIO_DATA_IV_LEN); bcopy(hdr->b_crypt_hdr.b_mac, mac, ZIO_DATA_MAC_LEN); *byteorder = (hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS) ? ZFS_HOST_BYTEORDER : !ZFS_HOST_BYTEORDER; } /* * Indicates how this buffer is compressed in memory. If it is not compressed * the value will be ZIO_COMPRESS_OFF. It can be made normally readable with * arc_untransform() as long as it is also unencrypted. */ enum zio_compress arc_get_compression(arc_buf_t *buf) { return (ARC_BUF_COMPRESSED(buf) ? HDR_GET_COMPRESS(buf->b_hdr) : ZIO_COMPRESS_OFF); } /* * Return the compression algorithm used to store this data in the ARC. If ARC * compression is enabled or this is an encrypted block, this will be the same * as what's used to store it on-disk. Otherwise, this will be ZIO_COMPRESS_OFF. */ static inline enum zio_compress arc_hdr_get_compress(arc_buf_hdr_t *hdr) { return (HDR_COMPRESSION_ENABLED(hdr) ? HDR_GET_COMPRESS(hdr) : ZIO_COMPRESS_OFF); } uint8_t arc_get_complevel(arc_buf_t *buf) { return (buf->b_hdr->b_complevel); } static inline boolean_t arc_buf_is_shared(arc_buf_t *buf) { boolean_t shared = (buf->b_data != NULL && buf->b_hdr->b_l1hdr.b_pabd != NULL && abd_is_linear(buf->b_hdr->b_l1hdr.b_pabd) && buf->b_data == abd_to_buf(buf->b_hdr->b_l1hdr.b_pabd)); IMPLY(shared, HDR_SHARED_DATA(buf->b_hdr)); IMPLY(shared, ARC_BUF_SHARED(buf)); IMPLY(shared, ARC_BUF_COMPRESSED(buf) || ARC_BUF_LAST(buf)); /* * It would be nice to assert arc_can_share() too, but the "hdr isn't * already being shared" requirement prevents us from doing that. */ return (shared); } /* * Free the checksum associated with this header. If there is no checksum, this * is a no-op. */ static inline void arc_cksum_free(arc_buf_hdr_t *hdr) { ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum != NULL) { kmem_free(hdr->b_l1hdr.b_freeze_cksum, sizeof (zio_cksum_t)); hdr->b_l1hdr.b_freeze_cksum = NULL; } mutex_exit(&hdr->b_l1hdr.b_freeze_lock); } /* * Return true iff at least one of the bufs on hdr is not compressed. * Encrypted buffers count as compressed. */ static boolean_t arc_hdr_has_uncompressed_buf(arc_buf_hdr_t *hdr) { ASSERT(hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY_OR_LOCKED(hdr)); for (arc_buf_t *b = hdr->b_l1hdr.b_buf; b != NULL; b = b->b_next) { if (!ARC_BUF_COMPRESSED(b)) { return (B_TRUE); } } return (B_FALSE); } /* * If we've turned on the ZFS_DEBUG_MODIFY flag, verify that the buf's data * matches the checksum that is stored in the hdr. If there is no checksum, * or if the buf is compressed, this is a no-op. */ static void arc_cksum_verify(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; zio_cksum_t zc; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; if (ARC_BUF_COMPRESSED(buf)) return; ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum == NULL || HDR_IO_ERROR(hdr)) { mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, &zc); if (!ZIO_CHECKSUM_EQUAL(*hdr->b_l1hdr.b_freeze_cksum, zc)) panic("buffer modified while frozen!"); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); } /* * This function makes the assumption that data stored in the L2ARC * will be transformed exactly as it is in the main pool. Because of * this we can verify the checksum against the reading process's bp. */ static boolean_t arc_cksum_is_equal(arc_buf_hdr_t *hdr, zio_t *zio) { ASSERT(!BP_IS_EMBEDDED(zio->io_bp)); VERIFY3U(BP_GET_PSIZE(zio->io_bp), ==, HDR_GET_PSIZE(hdr)); /* * Block pointers always store the checksum for the logical data. * If the block pointer has the gang bit set, then the checksum * it represents is for the reconstituted data and not for an * individual gang member. The zio pipeline, however, must be able to * determine the checksum of each of the gang constituents so it * treats the checksum comparison differently than what we need * for l2arc blocks. This prevents us from using the * zio_checksum_error() interface directly. Instead we must call the * zio_checksum_error_impl() so that we can ensure the checksum is * generated using the correct checksum algorithm and accounts for the * logical I/O size and not just a gang fragment. */ return (zio_checksum_error_impl(zio->io_spa, zio->io_bp, BP_GET_CHECKSUM(zio->io_bp), zio->io_abd, zio->io_size, zio->io_offset, NULL) == 0); } /* * Given a buf full of data, if ZFS_DEBUG_MODIFY is enabled this computes a * checksum and attaches it to the buf's hdr so that we can ensure that the buf * isn't modified later on. If buf is compressed or there is already a checksum * on the hdr, this is a no-op (we only checksum uncompressed bufs). */ static void arc_cksum_compute(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; ASSERT(HDR_HAS_L1HDR(hdr)); mutex_enter(&buf->b_hdr->b_l1hdr.b_freeze_lock); if (hdr->b_l1hdr.b_freeze_cksum != NULL || ARC_BUF_COMPRESSED(buf)) { mutex_exit(&hdr->b_l1hdr.b_freeze_lock); return; } ASSERT(!ARC_BUF_ENCRYPTED(buf)); ASSERT(!ARC_BUF_COMPRESSED(buf)); hdr->b_l1hdr.b_freeze_cksum = kmem_alloc(sizeof (zio_cksum_t), KM_SLEEP); fletcher_2_native(buf->b_data, arc_buf_size(buf), NULL, hdr->b_l1hdr.b_freeze_cksum); mutex_exit(&hdr->b_l1hdr.b_freeze_lock); arc_buf_watch(buf); } #ifndef _KERNEL void arc_buf_sigsegv(int sig, siginfo_t *si, void *unused) { panic("Got SIGSEGV at address: 0x%lx\n", (long)si->si_addr); } #endif /* ARGSUSED */ static void arc_buf_unwatch(arc_buf_t *buf) { #ifndef _KERNEL if (arc_watch) { ASSERT0(mprotect(buf->b_data, arc_buf_size(buf), PROT_READ | PROT_WRITE)); } #endif } /* ARGSUSED */ static void arc_buf_watch(arc_buf_t *buf) { #ifndef _KERNEL if (arc_watch) ASSERT0(mprotect(buf->b_data, arc_buf_size(buf), PROT_READ)); #endif } static arc_buf_contents_t arc_buf_type(arc_buf_hdr_t *hdr) { arc_buf_contents_t type; if (HDR_ISTYPE_METADATA(hdr)) { type = ARC_BUFC_METADATA; } else { type = ARC_BUFC_DATA; } VERIFY3U(hdr->b_type, ==, type); return (type); } boolean_t arc_is_metadata(arc_buf_t *buf) { return (HDR_ISTYPE_METADATA(buf->b_hdr) != 0); } static uint32_t arc_bufc_to_flags(arc_buf_contents_t type) { switch (type) { case ARC_BUFC_DATA: /* metadata field is 0 if buffer contains normal data */ return (0); case ARC_BUFC_METADATA: return (ARC_FLAG_BUFC_METADATA); default: break; } panic("undefined ARC buffer type!"); return ((uint32_t)-1); } void arc_buf_thaw(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); arc_cksum_verify(buf); /* * Compressed buffers do not manipulate the b_freeze_cksum. */ if (ARC_BUF_COMPRESSED(buf)) return; ASSERT(HDR_HAS_L1HDR(hdr)); arc_cksum_free(hdr); arc_buf_unwatch(buf); } void arc_buf_freeze(arc_buf_t *buf) { if (!(zfs_flags & ZFS_DEBUG_MODIFY)) return; if (ARC_BUF_COMPRESSED(buf)) return; ASSERT(HDR_HAS_L1HDR(buf->b_hdr)); arc_cksum_compute(buf); } /* * The arc_buf_hdr_t's b_flags should never be modified directly. Instead, * the following functions should be used to ensure that the flags are * updated in a thread-safe way. When manipulating the flags either * the hash_lock must be held or the hdr must be undiscoverable. This * ensures that we're not racing with any other threads when updating * the flags. */ static inline void arc_hdr_set_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) { ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); hdr->b_flags |= flags; } static inline void arc_hdr_clear_flags(arc_buf_hdr_t *hdr, arc_flags_t flags) { ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); hdr->b_flags &= ~flags; } /* * Setting the compression bits in the arc_buf_hdr_t's b_flags is * done in a special way since we have to clear and set bits * at the same time. Consumers that wish to set the compression bits * must use this function to ensure that the flags are updated in * thread-safe manner. */ static void arc_hdr_set_compress(arc_buf_hdr_t *hdr, enum zio_compress cmp) { ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); /* * Holes and embedded blocks will always have a psize = 0 so * we ignore the compression of the blkptr and set the * want to uncompress them. Mark them as uncompressed. */ if (!zfs_compressed_arc_enabled || HDR_GET_PSIZE(hdr) == 0) { arc_hdr_clear_flags(hdr, ARC_FLAG_COMPRESSED_ARC); ASSERT(!HDR_COMPRESSION_ENABLED(hdr)); } else { arc_hdr_set_flags(hdr, ARC_FLAG_COMPRESSED_ARC); ASSERT(HDR_COMPRESSION_ENABLED(hdr)); } HDR_SET_COMPRESS(hdr, cmp); ASSERT3U(HDR_GET_COMPRESS(hdr), ==, cmp); } /* * Looks for another buf on the same hdr which has the data decompressed, copies * from it, and returns true. If no such buf exists, returns false. */ static boolean_t arc_buf_try_copy_decompressed_data(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; boolean_t copied = B_FALSE; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3P(buf->b_data, !=, NULL); ASSERT(!ARC_BUF_COMPRESSED(buf)); for (arc_buf_t *from = hdr->b_l1hdr.b_buf; from != NULL; from = from->b_next) { /* can't use our own data buffer */ if (from == buf) { continue; } if (!ARC_BUF_COMPRESSED(from)) { bcopy(from->b_data, buf->b_data, arc_buf_size(buf)); copied = B_TRUE; break; } } /* * There were no decompressed bufs, so there should not be a * checksum on the hdr either. */ if (zfs_flags & ZFS_DEBUG_MODIFY) EQUIV(!copied, hdr->b_l1hdr.b_freeze_cksum == NULL); return (copied); } /* * Allocates an ARC buf header that's in an evicted & L2-cached state. * This is used during l2arc reconstruction to make empty ARC buffers * which circumvent the regular disk->arc->l2arc path and instead come * into being in the reverse order, i.e. l2arc->arc. */ static arc_buf_hdr_t * arc_buf_alloc_l2only(size_t size, arc_buf_contents_t type, l2arc_dev_t *dev, dva_t dva, uint64_t daddr, int32_t psize, uint64_t birth, enum zio_compress compress, uint8_t complevel, boolean_t protected, boolean_t prefetch, arc_state_type_t arcs_state) { arc_buf_hdr_t *hdr; ASSERT(size != 0); hdr = kmem_cache_alloc(hdr_l2only_cache, KM_SLEEP); hdr->b_birth = birth; hdr->b_type = type; hdr->b_flags = 0; arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L2HDR); HDR_SET_LSIZE(hdr, size); HDR_SET_PSIZE(hdr, psize); arc_hdr_set_compress(hdr, compress); hdr->b_complevel = complevel; if (protected) arc_hdr_set_flags(hdr, ARC_FLAG_PROTECTED); if (prefetch) arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); hdr->b_spa = spa_load_guid(dev->l2ad_vdev->vdev_spa); hdr->b_dva = dva; hdr->b_l2hdr.b_dev = dev; hdr->b_l2hdr.b_daddr = daddr; hdr->b_l2hdr.b_arcs_state = arcs_state; return (hdr); } /* * Return the size of the block, b_pabd, that is stored in the arc_buf_hdr_t. */ static uint64_t arc_hdr_size(arc_buf_hdr_t *hdr) { uint64_t size; if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF && HDR_GET_PSIZE(hdr) > 0) { size = HDR_GET_PSIZE(hdr); } else { ASSERT3U(HDR_GET_LSIZE(hdr), !=, 0); size = HDR_GET_LSIZE(hdr); } return (size); } static int arc_hdr_authenticate(arc_buf_hdr_t *hdr, spa_t *spa, uint64_t dsobj) { int ret; uint64_t csize; uint64_t lsize = HDR_GET_LSIZE(hdr); uint64_t psize = HDR_GET_PSIZE(hdr); void *tmpbuf = NULL; abd_t *abd = hdr->b_l1hdr.b_pabd; ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); ASSERT(HDR_AUTHENTICATED(hdr)); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); /* * The MAC is calculated on the compressed data that is stored on disk. * However, if compressed arc is disabled we will only have the * decompressed data available to us now. Compress it into a temporary * abd so we can verify the MAC. The performance overhead of this will * be relatively low, since most objects in an encrypted objset will * be encrypted (instead of authenticated) anyway. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) { tmpbuf = zio_buf_alloc(lsize); abd = abd_get_from_buf(tmpbuf, lsize); abd_take_ownership_of_buf(abd, B_TRUE); csize = zio_compress_data(HDR_GET_COMPRESS(hdr), hdr->b_l1hdr.b_pabd, tmpbuf, lsize, hdr->b_complevel); ASSERT3U(csize, <=, psize); abd_zero_off(abd, csize, psize - csize); } /* * Authentication is best effort. We authenticate whenever the key is * available. If we succeed we clear ARC_FLAG_NOAUTH. */ if (hdr->b_crypt_hdr.b_ot == DMU_OT_OBJSET) { ASSERT3U(HDR_GET_COMPRESS(hdr), ==, ZIO_COMPRESS_OFF); ASSERT3U(lsize, ==, psize); ret = spa_do_crypt_objset_mac_abd(B_FALSE, spa, dsobj, abd, psize, hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS); } else { ret = spa_do_crypt_mac_abd(B_FALSE, spa, dsobj, abd, psize, hdr->b_crypt_hdr.b_mac); } if (ret == 0) arc_hdr_clear_flags(hdr, ARC_FLAG_NOAUTH); else if (ret != ENOENT) goto error; if (tmpbuf != NULL) abd_free(abd); return (0); error: if (tmpbuf != NULL) abd_free(abd); return (ret); } /* * This function will take a header that only has raw encrypted data in * b_crypt_hdr.b_rabd and decrypt it into a new buffer which is stored in * b_l1hdr.b_pabd. If designated in the header flags, this function will * also decompress the data. */ static int arc_hdr_decrypt(arc_buf_hdr_t *hdr, spa_t *spa, const zbookmark_phys_t *zb) { int ret; abd_t *cabd = NULL; void *tmp = NULL; boolean_t no_crypt = B_FALSE; boolean_t bswap = (hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS); ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); ASSERT(HDR_ENCRYPTED(hdr)); arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT); ret = spa_do_crypt_abd(B_FALSE, spa, zb, hdr->b_crypt_hdr.b_ot, B_FALSE, bswap, hdr->b_crypt_hdr.b_salt, hdr->b_crypt_hdr.b_iv, hdr->b_crypt_hdr.b_mac, HDR_GET_PSIZE(hdr), hdr->b_l1hdr.b_pabd, hdr->b_crypt_hdr.b_rabd, &no_crypt); if (ret != 0) goto error; if (no_crypt) { abd_copy(hdr->b_l1hdr.b_pabd, hdr->b_crypt_hdr.b_rabd, HDR_GET_PSIZE(hdr)); } /* * If this header has disabled arc compression but the b_pabd is * compressed after decrypting it, we need to decompress the newly * decrypted data. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) { /* * We want to make sure that we are correctly honoring the * zfs_abd_scatter_enabled setting, so we allocate an abd here * and then loan a buffer from it, rather than allocating a * linear buffer and wrapping it in an abd later. */ cabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr, B_TRUE); tmp = abd_borrow_buf(cabd, arc_hdr_size(hdr)); ret = zio_decompress_data(HDR_GET_COMPRESS(hdr), hdr->b_l1hdr.b_pabd, tmp, HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr), &hdr->b_complevel); if (ret != 0) { abd_return_buf(cabd, tmp, arc_hdr_size(hdr)); goto error; } abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr)); arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, arc_hdr_size(hdr), hdr); hdr->b_l1hdr.b_pabd = cabd; } return (0); error: arc_hdr_free_abd(hdr, B_FALSE); if (cabd != NULL) arc_free_data_buf(hdr, cabd, arc_hdr_size(hdr), hdr); return (ret); } /* * This function is called during arc_buf_fill() to prepare the header's * abd plaintext pointer for use. This involves authenticated protected * data and decrypting encrypted data into the plaintext abd. */ static int arc_fill_hdr_crypt(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, spa_t *spa, const zbookmark_phys_t *zb, boolean_t noauth) { int ret; ASSERT(HDR_PROTECTED(hdr)); if (hash_lock != NULL) mutex_enter(hash_lock); if (HDR_NOAUTH(hdr) && !noauth) { /* * The caller requested authenticated data but our data has * not been authenticated yet. Verify the MAC now if we can. */ ret = arc_hdr_authenticate(hdr, spa, zb->zb_objset); if (ret != 0) goto error; } else if (HDR_HAS_RABD(hdr) && hdr->b_l1hdr.b_pabd == NULL) { /* * If we only have the encrypted version of the data, but the * unencrypted version was requested we take this opportunity * to store the decrypted version in the header for future use. */ ret = arc_hdr_decrypt(hdr, spa, zb); if (ret != 0) goto error; } ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); if (hash_lock != NULL) mutex_exit(hash_lock); return (0); error: if (hash_lock != NULL) mutex_exit(hash_lock); return (ret); } /* * This function is used by the dbuf code to decrypt bonus buffers in place. * The dbuf code itself doesn't have any locking for decrypting a shared dnode * block, so we use the hash lock here to protect against concurrent calls to * arc_buf_fill(). */ static void arc_buf_untransform_in_place(arc_buf_t *buf, kmutex_t *hash_lock) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT(HDR_ENCRYPTED(hdr)); ASSERT3U(hdr->b_crypt_hdr.b_ot, ==, DMU_OT_DNODE); ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); zio_crypt_copy_dnode_bonus(hdr->b_l1hdr.b_pabd, buf->b_data, arc_buf_size(buf)); buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED; buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; hdr->b_crypt_hdr.b_ebufcnt -= 1; } /* * Given a buf that has a data buffer attached to it, this function will * efficiently fill the buf with data of the specified compression setting from * the hdr and update the hdr's b_freeze_cksum if necessary. If the buf and hdr * are already sharing a data buf, no copy is performed. * * If the buf is marked as compressed but uncompressed data was requested, this * will allocate a new data buffer for the buf, remove that flag, and fill the * buf with uncompressed data. You can't request a compressed buf on a hdr with * uncompressed data, and (since we haven't added support for it yet) if you * want compressed data your buf must already be marked as compressed and have * the correct-sized data buffer. */ static int arc_buf_fill(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb, arc_fill_flags_t flags) { int error = 0; arc_buf_hdr_t *hdr = buf->b_hdr; boolean_t hdr_compressed = (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF); boolean_t compressed = (flags & ARC_FILL_COMPRESSED) != 0; boolean_t encrypted = (flags & ARC_FILL_ENCRYPTED) != 0; dmu_object_byteswap_t bswap = hdr->b_l1hdr.b_byteswap; kmutex_t *hash_lock = (flags & ARC_FILL_LOCKED) ? NULL : HDR_LOCK(hdr); ASSERT3P(buf->b_data, !=, NULL); IMPLY(compressed, hdr_compressed || ARC_BUF_ENCRYPTED(buf)); IMPLY(compressed, ARC_BUF_COMPRESSED(buf)); IMPLY(encrypted, HDR_ENCRYPTED(hdr)); IMPLY(encrypted, ARC_BUF_ENCRYPTED(buf)); IMPLY(encrypted, ARC_BUF_COMPRESSED(buf)); IMPLY(encrypted, !ARC_BUF_SHARED(buf)); /* * If the caller wanted encrypted data we just need to copy it from * b_rabd and potentially byteswap it. We won't be able to do any * further transforms on it. */ if (encrypted) { ASSERT(HDR_HAS_RABD(hdr)); abd_copy_to_buf(buf->b_data, hdr->b_crypt_hdr.b_rabd, HDR_GET_PSIZE(hdr)); goto byteswap; } /* * Adjust encrypted and authenticated headers to accommodate * the request if needed. Dnode blocks (ARC_FILL_IN_PLACE) are * allowed to fail decryption due to keys not being loaded * without being marked as an IO error. */ if (HDR_PROTECTED(hdr)) { error = arc_fill_hdr_crypt(hdr, hash_lock, spa, zb, !!(flags & ARC_FILL_NOAUTH)); if (error == EACCES && (flags & ARC_FILL_IN_PLACE) != 0) { return (error); } else if (error != 0) { if (hash_lock != NULL) mutex_enter(hash_lock); arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR); if (hash_lock != NULL) mutex_exit(hash_lock); return (error); } } /* * There is a special case here for dnode blocks which are * decrypting their bonus buffers. These blocks may request to * be decrypted in-place. This is necessary because there may * be many dnodes pointing into this buffer and there is * currently no method to synchronize replacing the backing * b_data buffer and updating all of the pointers. Here we use * the hash lock to ensure there are no races. If the need * arises for other types to be decrypted in-place, they must * add handling here as well. */ if ((flags & ARC_FILL_IN_PLACE) != 0) { ASSERT(!hdr_compressed); ASSERT(!compressed); ASSERT(!encrypted); if (HDR_ENCRYPTED(hdr) && ARC_BUF_ENCRYPTED(buf)) { ASSERT3U(hdr->b_crypt_hdr.b_ot, ==, DMU_OT_DNODE); if (hash_lock != NULL) mutex_enter(hash_lock); arc_buf_untransform_in_place(buf, hash_lock); if (hash_lock != NULL) mutex_exit(hash_lock); /* Compute the hdr's checksum if necessary */ arc_cksum_compute(buf); } return (0); } if (hdr_compressed == compressed) { if (!arc_buf_is_shared(buf)) { abd_copy_to_buf(buf->b_data, hdr->b_l1hdr.b_pabd, arc_buf_size(buf)); } } else { ASSERT(hdr_compressed); ASSERT(!compressed); ASSERT3U(HDR_GET_LSIZE(hdr), !=, HDR_GET_PSIZE(hdr)); /* * If the buf is sharing its data with the hdr, unlink it and * allocate a new data buffer for the buf. */ if (arc_buf_is_shared(buf)) { ASSERT(ARC_BUF_COMPRESSED(buf)); /* We need to give the buf its own b_data */ buf->b_flags &= ~ARC_BUF_FLAG_SHARED; buf->b_data = arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); /* Previously overhead was 0; just add new overhead */ ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr)); } else if (ARC_BUF_COMPRESSED(buf)) { /* We need to reallocate the buf's b_data */ arc_free_data_buf(hdr, buf->b_data, HDR_GET_PSIZE(hdr), buf); buf->b_data = arc_get_data_buf(hdr, HDR_GET_LSIZE(hdr), buf); /* We increased the size of b_data; update overhead */ ARCSTAT_INCR(arcstat_overhead_size, HDR_GET_LSIZE(hdr) - HDR_GET_PSIZE(hdr)); } /* * Regardless of the buf's previous compression settings, it * should not be compressed at the end of this function. */ buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; /* * Try copying the data from another buf which already has a * decompressed version. If that's not possible, it's time to * bite the bullet and decompress the data from the hdr. */ if (arc_buf_try_copy_decompressed_data(buf)) { /* Skip byteswapping and checksumming (already done) */ return (0); } else { error = zio_decompress_data(HDR_GET_COMPRESS(hdr), hdr->b_l1hdr.b_pabd, buf->b_data, HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr), &hdr->b_complevel); /* * Absent hardware errors or software bugs, this should * be impossible, but log it anyway so we can debug it. */ if (error != 0) { zfs_dbgmsg( "hdr %px, compress %d, psize %d, lsize %d", hdr, arc_hdr_get_compress(hdr), HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr)); if (hash_lock != NULL) mutex_enter(hash_lock); arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR); if (hash_lock != NULL) mutex_exit(hash_lock); return (SET_ERROR(EIO)); } } } byteswap: /* Byteswap the buf's data if necessary */ if (bswap != DMU_BSWAP_NUMFUNCS) { ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT3U(bswap, <, DMU_BSWAP_NUMFUNCS); dmu_ot_byteswap[bswap].ob_func(buf->b_data, HDR_GET_LSIZE(hdr)); } /* Compute the hdr's checksum if necessary */ arc_cksum_compute(buf); return (0); } /* * If this function is being called to decrypt an encrypted buffer or verify an * authenticated one, the key must be loaded and a mapping must be made * available in the keystore via spa_keystore_create_mapping() or one of its * callers. */ int arc_untransform(arc_buf_t *buf, spa_t *spa, const zbookmark_phys_t *zb, boolean_t in_place) { int ret; arc_fill_flags_t flags = 0; if (in_place) flags |= ARC_FILL_IN_PLACE; ret = arc_buf_fill(buf, spa, zb, flags); if (ret == ECKSUM) { /* * Convert authentication and decryption errors to EIO * (and generate an ereport) before leaving the ARC. */ ret = SET_ERROR(EIO); spa_log_error(spa, zb); (void) zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION, spa, NULL, zb, NULL, 0); } return (ret); } /* * Increment the amount of evictable space in the arc_state_t's refcount. * We account for the space used by the hdr and the arc buf individually * so that we can add and remove them from the refcount individually. */ static void arc_evictable_space_increment(arc_buf_hdr_t *hdr, arc_state_t *state) { arc_buf_contents_t type = arc_buf_type(hdr); ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(state)) { ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); (void) zfs_refcount_add_many(&state->arcs_esize[type], HDR_GET_LSIZE(hdr), hdr); return; } ASSERT(!GHOST_STATE(state)); if (hdr->b_l1hdr.b_pabd != NULL) { (void) zfs_refcount_add_many(&state->arcs_esize[type], arc_hdr_size(hdr), hdr); } if (HDR_HAS_RABD(hdr)) { (void) zfs_refcount_add_many(&state->arcs_esize[type], HDR_GET_PSIZE(hdr), hdr); } for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { if (arc_buf_is_shared(buf)) continue; (void) zfs_refcount_add_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } } /* * Decrement the amount of evictable space in the arc_state_t's refcount. * We account for the space used by the hdr and the arc buf individually * so that we can add and remove them from the refcount individually. */ static void arc_evictable_space_decrement(arc_buf_hdr_t *hdr, arc_state_t *state) { arc_buf_contents_t type = arc_buf_type(hdr); ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(state)) { ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); (void) zfs_refcount_remove_many(&state->arcs_esize[type], HDR_GET_LSIZE(hdr), hdr); return; } ASSERT(!GHOST_STATE(state)); if (hdr->b_l1hdr.b_pabd != NULL) { (void) zfs_refcount_remove_many(&state->arcs_esize[type], arc_hdr_size(hdr), hdr); } if (HDR_HAS_RABD(hdr)) { (void) zfs_refcount_remove_many(&state->arcs_esize[type], HDR_GET_PSIZE(hdr), hdr); } for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { if (arc_buf_is_shared(buf)) continue; (void) zfs_refcount_remove_many(&state->arcs_esize[type], arc_buf_size(buf), buf); } } /* * Add a reference to this hdr indicating that someone is actively * referencing that memory. When the refcount transitions from 0 to 1, * we remove it from the respective arc_state_t list to indicate that * it is not evictable. */ static void add_reference(arc_buf_hdr_t *hdr, void *tag) { arc_state_t *state; ASSERT(HDR_HAS_L1HDR(hdr)); if (!HDR_EMPTY(hdr) && !MUTEX_HELD(HDR_LOCK(hdr))) { ASSERT(hdr->b_l1hdr.b_state == arc_anon); ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); } state = hdr->b_l1hdr.b_state; if ((zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, tag) == 1) && (state != arc_anon)) { /* We don't use the L2-only state list. */ if (state != arc_l2c_only) { multilist_remove(&state->arcs_list[arc_buf_type(hdr)], hdr); arc_evictable_space_decrement(hdr, state); } /* remove the prefetch flag if we get a reference */ if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_decrement_state(hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH); if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_increment_state(hdr); } } /* * Remove a reference from this hdr. When the reference transitions from * 1 to 0 and we're not anonymous, then we add this hdr to the arc_state_t's * list making it eligible for eviction. */ static int remove_reference(arc_buf_hdr_t *hdr, kmutex_t *hash_lock, void *tag) { int cnt; arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(state == arc_anon || MUTEX_HELD(hash_lock)); ASSERT(!GHOST_STATE(state)); /* * arc_l2c_only counts as a ghost state so we don't need to explicitly * check to prevent usage of the arc_l2c_only list. */ if (((cnt = zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, tag)) == 0) && (state != arc_anon)) { multilist_insert(&state->arcs_list[arc_buf_type(hdr)], hdr); ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); arc_evictable_space_increment(hdr, state); } return (cnt); } /* * Returns detailed information about a specific arc buffer. When the * state_index argument is set the function will calculate the arc header * list position for its arc state. Since this requires a linear traversal * callers are strongly encourage not to do this. However, it can be helpful * for targeted analysis so the functionality is provided. */ void arc_buf_info(arc_buf_t *ab, arc_buf_info_t *abi, int state_index) { arc_buf_hdr_t *hdr = ab->b_hdr; l1arc_buf_hdr_t *l1hdr = NULL; l2arc_buf_hdr_t *l2hdr = NULL; arc_state_t *state = NULL; memset(abi, 0, sizeof (arc_buf_info_t)); if (hdr == NULL) return; abi->abi_flags = hdr->b_flags; if (HDR_HAS_L1HDR(hdr)) { l1hdr = &hdr->b_l1hdr; state = l1hdr->b_state; } if (HDR_HAS_L2HDR(hdr)) l2hdr = &hdr->b_l2hdr; if (l1hdr) { abi->abi_bufcnt = l1hdr->b_bufcnt; abi->abi_access = l1hdr->b_arc_access; abi->abi_mru_hits = l1hdr->b_mru_hits; abi->abi_mru_ghost_hits = l1hdr->b_mru_ghost_hits; abi->abi_mfu_hits = l1hdr->b_mfu_hits; abi->abi_mfu_ghost_hits = l1hdr->b_mfu_ghost_hits; abi->abi_holds = zfs_refcount_count(&l1hdr->b_refcnt); } if (l2hdr) { abi->abi_l2arc_dattr = l2hdr->b_daddr; abi->abi_l2arc_hits = l2hdr->b_hits; } abi->abi_state_type = state ? state->arcs_state : ARC_STATE_ANON; abi->abi_state_contents = arc_buf_type(hdr); abi->abi_size = arc_hdr_size(hdr); } /* * Move the supplied buffer to the indicated state. The hash lock * for the buffer must be held by the caller. */ static void arc_change_state(arc_state_t *new_state, arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { arc_state_t *old_state; int64_t refcnt; uint32_t bufcnt; boolean_t update_old, update_new; arc_buf_contents_t buftype = arc_buf_type(hdr); /* * We almost always have an L1 hdr here, since we call arc_hdr_realloc() * in arc_read() when bringing a buffer out of the L2ARC. However, the * L1 hdr doesn't always exist when we change state to arc_anon before * destroying a header, in which case reallocating to add the L1 hdr is * pointless. */ if (HDR_HAS_L1HDR(hdr)) { old_state = hdr->b_l1hdr.b_state; refcnt = zfs_refcount_count(&hdr->b_l1hdr.b_refcnt); bufcnt = hdr->b_l1hdr.b_bufcnt; update_old = (bufcnt > 0 || hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr)); } else { old_state = arc_l2c_only; refcnt = 0; bufcnt = 0; update_old = B_FALSE; } update_new = update_old; ASSERT(MUTEX_HELD(hash_lock)); ASSERT3P(new_state, !=, old_state); ASSERT(!GHOST_STATE(new_state) || bufcnt == 0); ASSERT(old_state != arc_anon || bufcnt <= 1); /* * If this buffer is evictable, transfer it from the * old state list to the new state list. */ if (refcnt == 0) { if (old_state != arc_anon && old_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); multilist_remove(&old_state->arcs_list[buftype], hdr); if (GHOST_STATE(old_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); update_old = B_TRUE; } arc_evictable_space_decrement(hdr, old_state); } if (new_state != arc_anon && new_state != arc_l2c_only) { /* * An L1 header always exists here, since if we're * moving to some L1-cached state (i.e. not l2c_only or * anonymous), we realloc the header to add an L1hdr * beforehand. */ ASSERT(HDR_HAS_L1HDR(hdr)); multilist_insert(&new_state->arcs_list[buftype], hdr); if (GHOST_STATE(new_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); update_new = B_TRUE; } arc_evictable_space_increment(hdr, new_state); } } ASSERT(!HDR_EMPTY(hdr)); if (new_state == arc_anon && HDR_IN_HASH_TABLE(hdr)) buf_hash_remove(hdr); /* adjust state sizes (ignore arc_l2c_only) */ if (update_new && new_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(new_state)) { ASSERT0(bufcnt); /* * When moving a header to a ghost state, we first * remove all arc buffers. Thus, we'll have a * bufcnt of zero, and no arc buffer to use for * the reference. As a result, we use the arc * header pointer for the reference. */ (void) zfs_refcount_add_many(&new_state->arcs_size, HDR_GET_LSIZE(hdr), hdr); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); } else { uint32_t buffers = 0; /* * Each individual buffer holds a unique reference, * thus we must remove each of these references one * at a time. */ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { ASSERT3U(bufcnt, !=, 0); buffers++; /* * When the arc_buf_t is sharing the data * block with the hdr, the owner of the * reference belongs to the hdr. Only * add to the refcount if the arc_buf_t is * not shared. */ if (arc_buf_is_shared(buf)) continue; (void) zfs_refcount_add_many( &new_state->arcs_size, arc_buf_size(buf), buf); } ASSERT3U(bufcnt, ==, buffers); if (hdr->b_l1hdr.b_pabd != NULL) { (void) zfs_refcount_add_many( &new_state->arcs_size, arc_hdr_size(hdr), hdr); } if (HDR_HAS_RABD(hdr)) { (void) zfs_refcount_add_many( &new_state->arcs_size, HDR_GET_PSIZE(hdr), hdr); } } } if (update_old && old_state != arc_l2c_only) { ASSERT(HDR_HAS_L1HDR(hdr)); if (GHOST_STATE(old_state)) { ASSERT0(bufcnt); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); /* * When moving a header off of a ghost state, * the header will not contain any arc buffers. * We use the arc header pointer for the reference * which is exactly what we did when we put the * header on the ghost state. */ (void) zfs_refcount_remove_many(&old_state->arcs_size, HDR_GET_LSIZE(hdr), hdr); } else { uint32_t buffers = 0; /* * Each individual buffer holds a unique reference, * thus we must remove each of these references one * at a time. */ for (arc_buf_t *buf = hdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { ASSERT3U(bufcnt, !=, 0); buffers++; /* * When the arc_buf_t is sharing the data * block with the hdr, the owner of the * reference belongs to the hdr. Only * add to the refcount if the arc_buf_t is * not shared. */ if (arc_buf_is_shared(buf)) continue; (void) zfs_refcount_remove_many( &old_state->arcs_size, arc_buf_size(buf), buf); } ASSERT3U(bufcnt, ==, buffers); ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr)); if (hdr->b_l1hdr.b_pabd != NULL) { (void) zfs_refcount_remove_many( &old_state->arcs_size, arc_hdr_size(hdr), hdr); } if (HDR_HAS_RABD(hdr)) { (void) zfs_refcount_remove_many( &old_state->arcs_size, HDR_GET_PSIZE(hdr), hdr); } } } if (HDR_HAS_L1HDR(hdr)) { hdr->b_l1hdr.b_state = new_state; if (HDR_HAS_L2HDR(hdr) && new_state != arc_l2c_only) { l2arc_hdr_arcstats_decrement_state(hdr); hdr->b_l2hdr.b_arcs_state = new_state->arcs_state; l2arc_hdr_arcstats_increment_state(hdr); } } /* * L2 headers should never be on the L2 state list since they don't * have L1 headers allocated. */ ASSERT(multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]) && multilist_is_empty(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA])); } void arc_space_consume(uint64_t space, arc_space_type_t type) { ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); switch (type) { default: break; case ARC_SPACE_DATA: - wmsum_add(&astat_data_size, space); + ARCSTAT_INCR(arcstat_data_size, space); break; case ARC_SPACE_META: - wmsum_add(&astat_metadata_size, space); + ARCSTAT_INCR(arcstat_metadata_size, space); break; case ARC_SPACE_BONUS: - wmsum_add(&astat_bonus_size, space); + ARCSTAT_INCR(arcstat_bonus_size, space); break; case ARC_SPACE_DNODE: - aggsum_add(&astat_dnode_size, space); + aggsum_add(&arc_sums.arcstat_dnode_size, space); break; case ARC_SPACE_DBUF: - wmsum_add(&astat_dbuf_size, space); + ARCSTAT_INCR(arcstat_dbuf_size, space); break; case ARC_SPACE_HDRS: - wmsum_add(&astat_hdr_size, space); + ARCSTAT_INCR(arcstat_hdr_size, space); break; case ARC_SPACE_L2HDRS: - aggsum_add(&astat_l2_hdr_size, space); + aggsum_add(&arc_sums.arcstat_l2_hdr_size, space); break; case ARC_SPACE_ABD_CHUNK_WASTE: /* * Note: this includes space wasted by all scatter ABD's, not * just those allocated by the ARC. But the vast majority of * scatter ABD's come from the ARC, because other users are * very short-lived. */ - wmsum_add(&astat_abd_chunk_waste_size, space); + ARCSTAT_INCR(arcstat_abd_chunk_waste_size, space); break; } if (type != ARC_SPACE_DATA && type != ARC_SPACE_ABD_CHUNK_WASTE) - aggsum_add(&arc_meta_used, space); + aggsum_add(&arc_sums.arcstat_meta_used, space); - aggsum_add(&arc_size, space); + aggsum_add(&arc_sums.arcstat_size, space); } void arc_space_return(uint64_t space, arc_space_type_t type) { ASSERT(type >= 0 && type < ARC_SPACE_NUMTYPES); switch (type) { default: break; case ARC_SPACE_DATA: - wmsum_add(&astat_data_size, -space); + ARCSTAT_INCR(arcstat_data_size, -space); break; case ARC_SPACE_META: - wmsum_add(&astat_metadata_size, -space); + ARCSTAT_INCR(arcstat_metadata_size, -space); break; case ARC_SPACE_BONUS: - wmsum_add(&astat_bonus_size, -space); + ARCSTAT_INCR(arcstat_bonus_size, -space); break; case ARC_SPACE_DNODE: - aggsum_add(&astat_dnode_size, -space); + aggsum_add(&arc_sums.arcstat_dnode_size, -space); break; case ARC_SPACE_DBUF: - wmsum_add(&astat_dbuf_size, -space); + ARCSTAT_INCR(arcstat_dbuf_size, -space); break; case ARC_SPACE_HDRS: - wmsum_add(&astat_hdr_size, -space); + ARCSTAT_INCR(arcstat_hdr_size, -space); break; case ARC_SPACE_L2HDRS: - aggsum_add(&astat_l2_hdr_size, -space); + aggsum_add(&arc_sums.arcstat_l2_hdr_size, -space); break; case ARC_SPACE_ABD_CHUNK_WASTE: - wmsum_add(&astat_abd_chunk_waste_size, -space); + ARCSTAT_INCR(arcstat_abd_chunk_waste_size, -space); break; } if (type != ARC_SPACE_DATA && type != ARC_SPACE_ABD_CHUNK_WASTE) { - ASSERT(aggsum_compare(&arc_meta_used, space) >= 0); - /* - * We use the upper bound here rather than the precise value - * because the arc_meta_max value doesn't need to be - * precise. It's only consumed by humans via arcstats. - */ - if (arc_meta_max < aggsum_upper_bound(&arc_meta_used)) - arc_meta_max = aggsum_upper_bound(&arc_meta_used); - aggsum_add(&arc_meta_used, -space); + ASSERT(aggsum_compare(&arc_sums.arcstat_meta_used, + space) >= 0); + ARCSTAT_MAX(arcstat_meta_max, + aggsum_upper_bound(&arc_sums.arcstat_meta_used)); + aggsum_add(&arc_sums.arcstat_meta_used, -space); } - ASSERT(aggsum_compare(&arc_size, space) >= 0); - aggsum_add(&arc_size, -space); + ASSERT(aggsum_compare(&arc_sums.arcstat_size, space) >= 0); + aggsum_add(&arc_sums.arcstat_size, -space); } /* * Given a hdr and a buf, returns whether that buf can share its b_data buffer * with the hdr's b_pabd. */ static boolean_t arc_can_share(arc_buf_hdr_t *hdr, arc_buf_t *buf) { /* * The criteria for sharing a hdr's data are: * 1. the buffer is not encrypted * 2. the hdr's compression matches the buf's compression * 3. the hdr doesn't need to be byteswapped * 4. the hdr isn't already being shared * 5. the buf is either compressed or it is the last buf in the hdr list * * Criterion #5 maintains the invariant that shared uncompressed * bufs must be the final buf in the hdr's b_buf list. Reading this, you * might ask, "if a compressed buf is allocated first, won't that be the * last thing in the list?", but in that case it's impossible to create * a shared uncompressed buf anyway (because the hdr must be compressed * to have the compressed buf). You might also think that #3 is * sufficient to make this guarantee, however it's possible * (specifically in the rare L2ARC write race mentioned in * arc_buf_alloc_impl()) there will be an existing uncompressed buf that * is shareable, but wasn't at the time of its allocation. Rather than * allow a new shared uncompressed buf to be created and then shuffle * the list around to make it the last element, this simply disallows * sharing if the new buf isn't the first to be added. */ ASSERT3P(buf->b_hdr, ==, hdr); boolean_t hdr_compressed = arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF; boolean_t buf_compressed = ARC_BUF_COMPRESSED(buf) != 0; return (!ARC_BUF_ENCRYPTED(buf) && buf_compressed == hdr_compressed && hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS && !HDR_SHARED_DATA(hdr) && (ARC_BUF_LAST(buf) || ARC_BUF_COMPRESSED(buf))); } /* * Allocate a buf for this hdr. If you care about the data that's in the hdr, * or if you want a compressed buffer, pass those flags in. Returns 0 if the * copy was made successfully, or an error code otherwise. */ static int arc_buf_alloc_impl(arc_buf_hdr_t *hdr, spa_t *spa, const zbookmark_phys_t *zb, void *tag, boolean_t encrypted, boolean_t compressed, boolean_t noauth, boolean_t fill, arc_buf_t **ret) { arc_buf_t *buf; arc_fill_flags_t flags = ARC_FILL_LOCKED; ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); VERIFY(hdr->b_type == ARC_BUFC_DATA || hdr->b_type == ARC_BUFC_METADATA); ASSERT3P(ret, !=, NULL); ASSERT3P(*ret, ==, NULL); IMPLY(encrypted, compressed); hdr->b_l1hdr.b_mru_hits = 0; hdr->b_l1hdr.b_mru_ghost_hits = 0; hdr->b_l1hdr.b_mfu_hits = 0; hdr->b_l1hdr.b_mfu_ghost_hits = 0; hdr->b_l1hdr.b_l2_hits = 0; buf = *ret = kmem_cache_alloc(buf_cache, KM_PUSHPAGE); buf->b_hdr = hdr; buf->b_data = NULL; buf->b_next = hdr->b_l1hdr.b_buf; buf->b_flags = 0; add_reference(hdr, tag); /* * We're about to change the hdr's b_flags. We must either * hold the hash_lock or be undiscoverable. */ ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); /* * Only honor requests for compressed bufs if the hdr is actually * compressed. This must be overridden if the buffer is encrypted since * encrypted buffers cannot be decompressed. */ if (encrypted) { buf->b_flags |= ARC_BUF_FLAG_COMPRESSED; buf->b_flags |= ARC_BUF_FLAG_ENCRYPTED; flags |= ARC_FILL_COMPRESSED | ARC_FILL_ENCRYPTED; } else if (compressed && arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF) { buf->b_flags |= ARC_BUF_FLAG_COMPRESSED; flags |= ARC_FILL_COMPRESSED; } if (noauth) { ASSERT0(encrypted); flags |= ARC_FILL_NOAUTH; } /* * If the hdr's data can be shared then we share the data buffer and * set the appropriate bit in the hdr's b_flags to indicate the hdr is * sharing it's b_pabd with the arc_buf_t. Otherwise, we allocate a new * buffer to store the buf's data. * * There are two additional restrictions here because we're sharing * hdr -> buf instead of the usual buf -> hdr. First, the hdr can't be * actively involved in an L2ARC write, because if this buf is used by * an arc_write() then the hdr's data buffer will be released when the * write completes, even though the L2ARC write might still be using it. * Second, the hdr's ABD must be linear so that the buf's user doesn't * need to be ABD-aware. It must be allocated via * zio_[data_]buf_alloc(), not as a page, because we need to be able * to abd_release_ownership_of_buf(), which isn't allowed on "linear * page" buffers because the ABD code needs to handle freeing them * specially. */ boolean_t can_share = arc_can_share(hdr, buf) && !HDR_L2_WRITING(hdr) && hdr->b_l1hdr.b_pabd != NULL && abd_is_linear(hdr->b_l1hdr.b_pabd) && !abd_is_linear_page(hdr->b_l1hdr.b_pabd); /* Set up b_data and sharing */ if (can_share) { buf->b_data = abd_to_buf(hdr->b_l1hdr.b_pabd); buf->b_flags |= ARC_BUF_FLAG_SHARED; arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); } else { buf->b_data = arc_get_data_buf(hdr, arc_buf_size(buf), buf); ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); } VERIFY3P(buf->b_data, !=, NULL); hdr->b_l1hdr.b_buf = buf; hdr->b_l1hdr.b_bufcnt += 1; if (encrypted) hdr->b_crypt_hdr.b_ebufcnt += 1; /* * If the user wants the data from the hdr, we need to either copy or * decompress the data. */ if (fill) { ASSERT3P(zb, !=, NULL); return (arc_buf_fill(buf, spa, zb, flags)); } return (0); } static char *arc_onloan_tag = "onloan"; static inline void arc_loaned_bytes_update(int64_t delta) { atomic_add_64(&arc_loaned_bytes, delta); /* assert that it did not wrap around */ ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); } /* * Loan out an anonymous arc buffer. Loaned buffers are not counted as in * flight data by arc_tempreserve_space() until they are "returned". Loaned * buffers must be returned to the arc before they can be used by the DMU or * freed. */ arc_buf_t * arc_loan_buf(spa_t *spa, boolean_t is_metadata, int size) { arc_buf_t *buf = arc_alloc_buf(spa, arc_onloan_tag, is_metadata ? ARC_BUFC_METADATA : ARC_BUFC_DATA, size); arc_loaned_bytes_update(arc_buf_size(buf)); return (buf); } arc_buf_t * arc_loan_compressed_buf(spa_t *spa, uint64_t psize, uint64_t lsize, enum zio_compress compression_type, uint8_t complevel) { arc_buf_t *buf = arc_alloc_compressed_buf(spa, arc_onloan_tag, psize, lsize, compression_type, complevel); arc_loaned_bytes_update(arc_buf_size(buf)); return (buf); } arc_buf_t * arc_loan_raw_buf(spa_t *spa, uint64_t dsobj, boolean_t byteorder, const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_object_type_t ot, uint64_t psize, uint64_t lsize, enum zio_compress compression_type, uint8_t complevel) { arc_buf_t *buf = arc_alloc_raw_buf(spa, arc_onloan_tag, dsobj, byteorder, salt, iv, mac, ot, psize, lsize, compression_type, complevel); atomic_add_64(&arc_loaned_bytes, psize); return (buf); } /* * Return a loaned arc buffer to the arc. */ void arc_return_buf(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(buf->b_data, !=, NULL); ASSERT(HDR_HAS_L1HDR(hdr)); (void) zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, tag); (void) zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); arc_loaned_bytes_update(-arc_buf_size(buf)); } /* Detach an arc_buf from a dbuf (tag) */ void arc_loan_inuse_buf(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(buf->b_data, !=, NULL); ASSERT(HDR_HAS_L1HDR(hdr)); (void) zfs_refcount_add(&hdr->b_l1hdr.b_refcnt, arc_onloan_tag); (void) zfs_refcount_remove(&hdr->b_l1hdr.b_refcnt, tag); arc_loaned_bytes_update(arc_buf_size(buf)); } static void l2arc_free_abd_on_write(abd_t *abd, size_t size, arc_buf_contents_t type) { l2arc_data_free_t *df = kmem_alloc(sizeof (*df), KM_SLEEP); df->l2df_abd = abd; df->l2df_size = size; df->l2df_type = type; mutex_enter(&l2arc_free_on_write_mtx); list_insert_head(l2arc_free_on_write, df); mutex_exit(&l2arc_free_on_write_mtx); } static void arc_hdr_free_on_write(arc_buf_hdr_t *hdr, boolean_t free_rdata) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); uint64_t size = (free_rdata) ? HDR_GET_PSIZE(hdr) : arc_hdr_size(hdr); /* protected by hash lock, if in the hash table */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT(state != arc_anon && state != arc_l2c_only); (void) zfs_refcount_remove_many(&state->arcs_esize[type], size, hdr); } (void) zfs_refcount_remove_many(&state->arcs_size, size, hdr); if (type == ARC_BUFC_METADATA) { arc_space_return(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); arc_space_return(size, ARC_SPACE_DATA); } if (free_rdata) { l2arc_free_abd_on_write(hdr->b_crypt_hdr.b_rabd, size, type); } else { l2arc_free_abd_on_write(hdr->b_l1hdr.b_pabd, size, type); } } /* * Share the arc_buf_t's data with the hdr. Whenever we are sharing the * data buffer, we transfer the refcount ownership to the hdr and update * the appropriate kstats. */ static void arc_share_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) { ASSERT(arc_can_share(hdr, buf)); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!ARC_BUF_ENCRYPTED(buf)); ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); /* * Start sharing the data buffer. We transfer the * refcount ownership to the hdr since it always owns * the refcount whenever an arc_buf_t is shared. */ zfs_refcount_transfer_ownership_many(&hdr->b_l1hdr.b_state->arcs_size, arc_hdr_size(hdr), buf, hdr); hdr->b_l1hdr.b_pabd = abd_get_from_buf(buf->b_data, arc_buf_size(buf)); abd_take_ownership_of_buf(hdr->b_l1hdr.b_pabd, HDR_ISTYPE_METADATA(hdr)); arc_hdr_set_flags(hdr, ARC_FLAG_SHARED_DATA); buf->b_flags |= ARC_BUF_FLAG_SHARED; /* * Since we've transferred ownership to the hdr we need * to increment its compressed and uncompressed kstats and * decrement the overhead size. */ ARCSTAT_INCR(arcstat_compressed_size, arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); ARCSTAT_INCR(arcstat_overhead_size, -arc_buf_size(buf)); } static void arc_unshare_buf(arc_buf_hdr_t *hdr, arc_buf_t *buf) { ASSERT(arc_buf_is_shared(buf)); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); /* * We are no longer sharing this buffer so we need * to transfer its ownership to the rightful owner. */ zfs_refcount_transfer_ownership_many(&hdr->b_l1hdr.b_state->arcs_size, arc_hdr_size(hdr), hdr, buf); arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); abd_release_ownership_of_buf(hdr->b_l1hdr.b_pabd); abd_free(hdr->b_l1hdr.b_pabd); hdr->b_l1hdr.b_pabd = NULL; buf->b_flags &= ~ARC_BUF_FLAG_SHARED; /* * Since the buffer is no longer shared between * the arc buf and the hdr, count it as overhead. */ ARCSTAT_INCR(arcstat_compressed_size, -arc_hdr_size(hdr)); ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); ARCSTAT_INCR(arcstat_overhead_size, arc_buf_size(buf)); } /* * Remove an arc_buf_t from the hdr's buf list and return the last * arc_buf_t on the list. If no buffers remain on the list then return * NULL. */ static arc_buf_t * arc_buf_remove(arc_buf_hdr_t *hdr, arc_buf_t *buf) { ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); arc_buf_t **bufp = &hdr->b_l1hdr.b_buf; arc_buf_t *lastbuf = NULL; /* * Remove the buf from the hdr list and locate the last * remaining buffer on the list. */ while (*bufp != NULL) { if (*bufp == buf) *bufp = buf->b_next; /* * If we've removed a buffer in the middle of * the list then update the lastbuf and update * bufp. */ if (*bufp != NULL) { lastbuf = *bufp; bufp = &(*bufp)->b_next; } } buf->b_next = NULL; ASSERT3P(lastbuf, !=, buf); IMPLY(hdr->b_l1hdr.b_bufcnt > 0, lastbuf != NULL); IMPLY(hdr->b_l1hdr.b_bufcnt > 0, hdr->b_l1hdr.b_buf != NULL); IMPLY(lastbuf != NULL, ARC_BUF_LAST(lastbuf)); return (lastbuf); } /* * Free up buf->b_data and pull the arc_buf_t off of the arc_buf_hdr_t's * list and free it. */ static void arc_buf_destroy_impl(arc_buf_t *buf) { arc_buf_hdr_t *hdr = buf->b_hdr; /* * Free up the data associated with the buf but only if we're not * sharing this with the hdr. If we are sharing it with the hdr, the * hdr is responsible for doing the free. */ if (buf->b_data != NULL) { /* * We're about to change the hdr's b_flags. We must either * hold the hash_lock or be undiscoverable. */ ASSERT(HDR_EMPTY_OR_LOCKED(hdr)); arc_cksum_verify(buf); arc_buf_unwatch(buf); if (arc_buf_is_shared(buf)) { arc_hdr_clear_flags(hdr, ARC_FLAG_SHARED_DATA); } else { uint64_t size = arc_buf_size(buf); arc_free_data_buf(hdr, buf->b_data, size, buf); ARCSTAT_INCR(arcstat_overhead_size, -size); } buf->b_data = NULL; ASSERT(hdr->b_l1hdr.b_bufcnt > 0); hdr->b_l1hdr.b_bufcnt -= 1; if (ARC_BUF_ENCRYPTED(buf)) { hdr->b_crypt_hdr.b_ebufcnt -= 1; /* * If we have no more encrypted buffers and we've * already gotten a copy of the decrypted data we can * free b_rabd to save some space. */ if (hdr->b_crypt_hdr.b_ebufcnt == 0 && HDR_HAS_RABD(hdr) && hdr->b_l1hdr.b_pabd != NULL && !HDR_IO_IN_PROGRESS(hdr)) { arc_hdr_free_abd(hdr, B_TRUE); } } } arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); if (ARC_BUF_SHARED(buf) && !ARC_BUF_COMPRESSED(buf)) { /* * If the current arc_buf_t is sharing its data buffer with the * hdr, then reassign the hdr's b_pabd to share it with the new * buffer at the end of the list. The shared buffer is always * the last one on the hdr's buffer list. * * There is an equivalent case for compressed bufs, but since * they aren't guaranteed to be the last buf in the list and * that is an exceedingly rare case, we just allow that space be * wasted temporarily. We must also be careful not to share * encrypted buffers, since they cannot be shared. */ if (lastbuf != NULL && !ARC_BUF_ENCRYPTED(lastbuf)) { /* Only one buf can be shared at once */ VERIFY(!arc_buf_is_shared(lastbuf)); /* hdr is uncompressed so can't have compressed buf */ VERIFY(!ARC_BUF_COMPRESSED(lastbuf)); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); arc_hdr_free_abd(hdr, B_FALSE); /* * We must setup a new shared block between the * last buffer and the hdr. The data would have * been allocated by the arc buf so we need to transfer * ownership to the hdr since it's now being shared. */ arc_share_buf(hdr, lastbuf); } } else if (HDR_SHARED_DATA(hdr)) { /* * Uncompressed shared buffers are always at the end * of the list. Compressed buffers don't have the * same requirements. This makes it hard to * simply assert that the lastbuf is shared so * we rely on the hdr's compression flags to determine * if we have a compressed, shared buffer. */ ASSERT3P(lastbuf, !=, NULL); ASSERT(arc_buf_is_shared(lastbuf) || arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF); } /* * Free the checksum if we're removing the last uncompressed buf from * this hdr. */ if (!arc_hdr_has_uncompressed_buf(hdr)) { arc_cksum_free(hdr); } /* clean up the buf */ buf->b_hdr = NULL; kmem_cache_free(buf_cache, buf); } static void arc_hdr_alloc_abd(arc_buf_hdr_t *hdr, int alloc_flags) { uint64_t size; boolean_t alloc_rdata = ((alloc_flags & ARC_HDR_ALLOC_RDATA) != 0); boolean_t do_adapt = ((alloc_flags & ARC_HDR_DO_ADAPT) != 0); ASSERT3U(HDR_GET_LSIZE(hdr), >, 0); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!HDR_SHARED_DATA(hdr) || alloc_rdata); IMPLY(alloc_rdata, HDR_PROTECTED(hdr)); if (alloc_rdata) { size = HDR_GET_PSIZE(hdr); ASSERT3P(hdr->b_crypt_hdr.b_rabd, ==, NULL); hdr->b_crypt_hdr.b_rabd = arc_get_data_abd(hdr, size, hdr, do_adapt); ASSERT3P(hdr->b_crypt_hdr.b_rabd, !=, NULL); ARCSTAT_INCR(arcstat_raw_size, size); } else { size = arc_hdr_size(hdr); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); hdr->b_l1hdr.b_pabd = arc_get_data_abd(hdr, size, hdr, do_adapt); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); } ARCSTAT_INCR(arcstat_compressed_size, size); ARCSTAT_INCR(arcstat_uncompressed_size, HDR_GET_LSIZE(hdr)); } static void arc_hdr_free_abd(arc_buf_hdr_t *hdr, boolean_t free_rdata) { uint64_t size = (free_rdata) ? HDR_GET_PSIZE(hdr) : arc_hdr_size(hdr); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr)); IMPLY(free_rdata, HDR_HAS_RABD(hdr)); /* * If the hdr is currently being written to the l2arc then * we defer freeing the data by adding it to the l2arc_free_on_write * list. The l2arc will free the data once it's finished * writing it to the l2arc device. */ if (HDR_L2_WRITING(hdr)) { arc_hdr_free_on_write(hdr, free_rdata); ARCSTAT_BUMP(arcstat_l2_free_on_write); } else if (free_rdata) { arc_free_data_abd(hdr, hdr->b_crypt_hdr.b_rabd, size, hdr); } else { arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, size, hdr); } if (free_rdata) { hdr->b_crypt_hdr.b_rabd = NULL; ARCSTAT_INCR(arcstat_raw_size, -size); } else { hdr->b_l1hdr.b_pabd = NULL; } if (hdr->b_l1hdr.b_pabd == NULL && !HDR_HAS_RABD(hdr)) hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; ARCSTAT_INCR(arcstat_compressed_size, -size); ARCSTAT_INCR(arcstat_uncompressed_size, -HDR_GET_LSIZE(hdr)); } static arc_buf_hdr_t * arc_hdr_alloc(uint64_t spa, int32_t psize, int32_t lsize, boolean_t protected, enum zio_compress compression_type, uint8_t complevel, arc_buf_contents_t type, boolean_t alloc_rdata) { arc_buf_hdr_t *hdr; int flags = ARC_HDR_DO_ADAPT; VERIFY(type == ARC_BUFC_DATA || type == ARC_BUFC_METADATA); if (protected) { hdr = kmem_cache_alloc(hdr_full_crypt_cache, KM_PUSHPAGE); } else { hdr = kmem_cache_alloc(hdr_full_cache, KM_PUSHPAGE); } flags |= alloc_rdata ? ARC_HDR_ALLOC_RDATA : 0; ASSERT(HDR_EMPTY(hdr)); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); HDR_SET_PSIZE(hdr, psize); HDR_SET_LSIZE(hdr, lsize); hdr->b_spa = spa; hdr->b_type = type; hdr->b_flags = 0; arc_hdr_set_flags(hdr, arc_bufc_to_flags(type) | ARC_FLAG_HAS_L1HDR); arc_hdr_set_compress(hdr, compression_type); hdr->b_complevel = complevel; if (protected) arc_hdr_set_flags(hdr, ARC_FLAG_PROTECTED); hdr->b_l1hdr.b_state = arc_anon; hdr->b_l1hdr.b_arc_access = 0; hdr->b_l1hdr.b_bufcnt = 0; hdr->b_l1hdr.b_buf = NULL; /* * Allocate the hdr's buffer. This will contain either * the compressed or uncompressed data depending on the block * it references and compressed arc enablement. */ arc_hdr_alloc_abd(hdr, flags); ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); return (hdr); } /* * Transition between the two allocation states for the arc_buf_hdr struct. * The arc_buf_hdr struct can be allocated with (hdr_full_cache) or without * (hdr_l2only_cache) the fields necessary for the L1 cache - the smaller * version is used when a cache buffer is only in the L2ARC in order to reduce * memory usage. */ static arc_buf_hdr_t * arc_hdr_realloc(arc_buf_hdr_t *hdr, kmem_cache_t *old, kmem_cache_t *new) { ASSERT(HDR_HAS_L2HDR(hdr)); arc_buf_hdr_t *nhdr; l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; ASSERT((old == hdr_full_cache && new == hdr_l2only_cache) || (old == hdr_l2only_cache && new == hdr_full_cache)); /* * if the caller wanted a new full header and the header is to be * encrypted we will actually allocate the header from the full crypt * cache instead. The same applies to freeing from the old cache. */ if (HDR_PROTECTED(hdr) && new == hdr_full_cache) new = hdr_full_crypt_cache; if (HDR_PROTECTED(hdr) && old == hdr_full_cache) old = hdr_full_crypt_cache; nhdr = kmem_cache_alloc(new, KM_PUSHPAGE); ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); buf_hash_remove(hdr); bcopy(hdr, nhdr, HDR_L2ONLY_SIZE); if (new == hdr_full_cache || new == hdr_full_crypt_cache) { arc_hdr_set_flags(nhdr, ARC_FLAG_HAS_L1HDR); /* * arc_access and arc_change_state need to be aware that a * header has just come out of L2ARC, so we set its state to * l2c_only even though it's about to change. */ nhdr->b_l1hdr.b_state = arc_l2c_only; /* Verify previous threads set to NULL before freeing */ ASSERT3P(nhdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); } else { ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT0(hdr->b_l1hdr.b_bufcnt); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); /* * If we've reached here, We must have been called from * arc_evict_hdr(), as such we should have already been * removed from any ghost list we were previously on * (which protects us from racing with arc_evict_state), * thus no locking is needed during this check. */ ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); /* * A buffer must not be moved into the arc_l2c_only * state if it's not finished being written out to the * l2arc device. Otherwise, the b_l1hdr.b_pabd field * might try to be accessed, even though it was removed. */ VERIFY(!HDR_L2_WRITING(hdr)); VERIFY3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); arc_hdr_clear_flags(nhdr, ARC_FLAG_HAS_L1HDR); } /* * The header has been reallocated so we need to re-insert it into any * lists it was on. */ (void) buf_hash_insert(nhdr, NULL); ASSERT(list_link_active(&hdr->b_l2hdr.b_l2node)); mutex_enter(&dev->l2ad_mtx); /* * We must place the realloc'ed header back into the list at * the same spot. Otherwise, if it's placed earlier in the list, * l2arc_write_buffers() could find it during the function's * write phase, and try to write it out to the l2arc. */ list_insert_after(&dev->l2ad_buflist, hdr, nhdr); list_remove(&dev->l2ad_buflist, hdr); mutex_exit(&dev->l2ad_mtx); /* * Since we're using the pointer address as the tag when * incrementing and decrementing the l2ad_alloc refcount, we * must remove the old pointer (that we're about to destroy) and * add the new pointer to the refcount. Otherwise we'd remove * the wrong pointer address when calling arc_hdr_destroy() later. */ (void) zfs_refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); (void) zfs_refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(nhdr), nhdr); buf_discard_identity(hdr); kmem_cache_free(old, hdr); return (nhdr); } /* * This function allows an L1 header to be reallocated as a crypt * header and vice versa. If we are going to a crypt header, the * new fields will be zeroed out. */ static arc_buf_hdr_t * arc_hdr_realloc_crypt(arc_buf_hdr_t *hdr, boolean_t need_crypt) { arc_buf_hdr_t *nhdr; arc_buf_t *buf; kmem_cache_t *ncache, *ocache; /* * This function requires that hdr is in the arc_anon state. * Therefore it won't have any L2ARC data for us to worry * about copying. */ ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!HDR_HAS_L2HDR(hdr)); ASSERT3U(!!HDR_PROTECTED(hdr), !=, need_crypt); ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT(!list_link_active(&hdr->b_l2hdr.b_l2node)); ASSERT3P(hdr->b_hash_next, ==, NULL); if (need_crypt) { ncache = hdr_full_crypt_cache; ocache = hdr_full_cache; } else { ncache = hdr_full_cache; ocache = hdr_full_crypt_cache; } nhdr = kmem_cache_alloc(ncache, KM_PUSHPAGE); /* * Copy all members that aren't locks or condvars to the new header. * No lists are pointing to us (as we asserted above), so we don't * need to worry about the list nodes. */ nhdr->b_dva = hdr->b_dva; nhdr->b_birth = hdr->b_birth; nhdr->b_type = hdr->b_type; nhdr->b_flags = hdr->b_flags; nhdr->b_psize = hdr->b_psize; nhdr->b_lsize = hdr->b_lsize; nhdr->b_spa = hdr->b_spa; nhdr->b_l1hdr.b_freeze_cksum = hdr->b_l1hdr.b_freeze_cksum; nhdr->b_l1hdr.b_bufcnt = hdr->b_l1hdr.b_bufcnt; nhdr->b_l1hdr.b_byteswap = hdr->b_l1hdr.b_byteswap; nhdr->b_l1hdr.b_state = hdr->b_l1hdr.b_state; nhdr->b_l1hdr.b_arc_access = hdr->b_l1hdr.b_arc_access; nhdr->b_l1hdr.b_mru_hits = hdr->b_l1hdr.b_mru_hits; nhdr->b_l1hdr.b_mru_ghost_hits = hdr->b_l1hdr.b_mru_ghost_hits; nhdr->b_l1hdr.b_mfu_hits = hdr->b_l1hdr.b_mfu_hits; nhdr->b_l1hdr.b_mfu_ghost_hits = hdr->b_l1hdr.b_mfu_ghost_hits; nhdr->b_l1hdr.b_l2_hits = hdr->b_l1hdr.b_l2_hits; nhdr->b_l1hdr.b_acb = hdr->b_l1hdr.b_acb; nhdr->b_l1hdr.b_pabd = hdr->b_l1hdr.b_pabd; /* * This zfs_refcount_add() exists only to ensure that the individual * arc buffers always point to a header that is referenced, avoiding * a small race condition that could trigger ASSERTs. */ (void) zfs_refcount_add(&nhdr->b_l1hdr.b_refcnt, FTAG); nhdr->b_l1hdr.b_buf = hdr->b_l1hdr.b_buf; for (buf = nhdr->b_l1hdr.b_buf; buf != NULL; buf = buf->b_next) { mutex_enter(&buf->b_evict_lock); buf->b_hdr = nhdr; mutex_exit(&buf->b_evict_lock); } zfs_refcount_transfer(&nhdr->b_l1hdr.b_refcnt, &hdr->b_l1hdr.b_refcnt); (void) zfs_refcount_remove(&nhdr->b_l1hdr.b_refcnt, FTAG); ASSERT0(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt)); if (need_crypt) { arc_hdr_set_flags(nhdr, ARC_FLAG_PROTECTED); } else { arc_hdr_clear_flags(nhdr, ARC_FLAG_PROTECTED); } /* unset all members of the original hdr */ bzero(&hdr->b_dva, sizeof (dva_t)); hdr->b_birth = 0; hdr->b_type = ARC_BUFC_INVALID; hdr->b_flags = 0; hdr->b_psize = 0; hdr->b_lsize = 0; hdr->b_spa = 0; hdr->b_l1hdr.b_freeze_cksum = NULL; hdr->b_l1hdr.b_buf = NULL; hdr->b_l1hdr.b_bufcnt = 0; hdr->b_l1hdr.b_byteswap = 0; hdr->b_l1hdr.b_state = NULL; hdr->b_l1hdr.b_arc_access = 0; hdr->b_l1hdr.b_mru_hits = 0; hdr->b_l1hdr.b_mru_ghost_hits = 0; hdr->b_l1hdr.b_mfu_hits = 0; hdr->b_l1hdr.b_mfu_ghost_hits = 0; hdr->b_l1hdr.b_l2_hits = 0; hdr->b_l1hdr.b_acb = NULL; hdr->b_l1hdr.b_pabd = NULL; if (ocache == hdr_full_crypt_cache) { ASSERT(!HDR_HAS_RABD(hdr)); hdr->b_crypt_hdr.b_ot = DMU_OT_NONE; hdr->b_crypt_hdr.b_ebufcnt = 0; hdr->b_crypt_hdr.b_dsobj = 0; bzero(hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN); bzero(hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN); bzero(hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN); } buf_discard_identity(hdr); kmem_cache_free(ocache, hdr); return (nhdr); } /* * This function is used by the send / receive code to convert a newly * allocated arc_buf_t to one that is suitable for a raw encrypted write. It * is also used to allow the root objset block to be updated without altering * its embedded MACs. Both block types will always be uncompressed so we do not * have to worry about compression type or psize. */ void arc_convert_to_raw(arc_buf_t *buf, uint64_t dsobj, boolean_t byteorder, dmu_object_type_t ot, const uint8_t *salt, const uint8_t *iv, const uint8_t *mac) { arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT(ot == DMU_OT_DNODE || ot == DMU_OT_OBJSET); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); buf->b_flags |= (ARC_BUF_FLAG_COMPRESSED | ARC_BUF_FLAG_ENCRYPTED); if (!HDR_PROTECTED(hdr)) hdr = arc_hdr_realloc_crypt(hdr, B_TRUE); hdr->b_crypt_hdr.b_dsobj = dsobj; hdr->b_crypt_hdr.b_ot = ot; hdr->b_l1hdr.b_byteswap = (byteorder == ZFS_HOST_BYTEORDER) ? DMU_BSWAP_NUMFUNCS : DMU_OT_BYTESWAP(ot); if (!arc_hdr_has_uncompressed_buf(hdr)) arc_cksum_free(hdr); if (salt != NULL) bcopy(salt, hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN); if (iv != NULL) bcopy(iv, hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN); if (mac != NULL) bcopy(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN); } /* * Allocate a new arc_buf_hdr_t and arc_buf_t and return the buf to the caller. * The buf is returned thawed since we expect the consumer to modify it. */ arc_buf_t * arc_alloc_buf(spa_t *spa, void *tag, arc_buf_contents_t type, int32_t size) { arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), size, size, B_FALSE, ZIO_COMPRESS_OFF, 0, type, B_FALSE); arc_buf_t *buf = NULL; VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_FALSE, B_FALSE, B_FALSE, B_FALSE, &buf)); arc_buf_thaw(buf); return (buf); } /* * Allocate a compressed buf in the same manner as arc_alloc_buf. Don't use this * for bufs containing metadata. */ arc_buf_t * arc_alloc_compressed_buf(spa_t *spa, void *tag, uint64_t psize, uint64_t lsize, enum zio_compress compression_type, uint8_t complevel) { ASSERT3U(lsize, >, 0); ASSERT3U(lsize, >=, psize); ASSERT3U(compression_type, >, ZIO_COMPRESS_OFF); ASSERT3U(compression_type, <, ZIO_COMPRESS_FUNCTIONS); arc_buf_hdr_t *hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, B_FALSE, compression_type, complevel, ARC_BUFC_DATA, B_FALSE); arc_buf_t *buf = NULL; VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_FALSE, B_TRUE, B_FALSE, B_FALSE, &buf)); arc_buf_thaw(buf); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); if (!arc_buf_is_shared(buf)) { /* * To ensure that the hdr has the correct data in it if we call * arc_untransform() on this buf before it's been written to * disk, it's easiest if we just set up sharing between the * buf and the hdr. */ arc_hdr_free_abd(hdr, B_FALSE); arc_share_buf(hdr, buf); } return (buf); } arc_buf_t * arc_alloc_raw_buf(spa_t *spa, void *tag, uint64_t dsobj, boolean_t byteorder, const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_object_type_t ot, uint64_t psize, uint64_t lsize, enum zio_compress compression_type, uint8_t complevel) { arc_buf_hdr_t *hdr; arc_buf_t *buf; arc_buf_contents_t type = DMU_OT_IS_METADATA(ot) ? ARC_BUFC_METADATA : ARC_BUFC_DATA; ASSERT3U(lsize, >, 0); ASSERT3U(lsize, >=, psize); ASSERT3U(compression_type, >=, ZIO_COMPRESS_OFF); ASSERT3U(compression_type, <, ZIO_COMPRESS_FUNCTIONS); hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, B_TRUE, compression_type, complevel, type, B_TRUE); hdr->b_crypt_hdr.b_dsobj = dsobj; hdr->b_crypt_hdr.b_ot = ot; hdr->b_l1hdr.b_byteswap = (byteorder == ZFS_HOST_BYTEORDER) ? DMU_BSWAP_NUMFUNCS : DMU_OT_BYTESWAP(ot); bcopy(salt, hdr->b_crypt_hdr.b_salt, ZIO_DATA_SALT_LEN); bcopy(iv, hdr->b_crypt_hdr.b_iv, ZIO_DATA_IV_LEN); bcopy(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN); /* * This buffer will be considered encrypted even if the ot is not an * encrypted type. It will become authenticated instead in * arc_write_ready(). */ buf = NULL; VERIFY0(arc_buf_alloc_impl(hdr, spa, NULL, tag, B_TRUE, B_TRUE, B_FALSE, B_FALSE, &buf)); arc_buf_thaw(buf); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); return (buf); } static void l2arc_hdr_arcstats_update(arc_buf_hdr_t *hdr, boolean_t incr, boolean_t state_only) { l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; l2arc_dev_t *dev = l2hdr->b_dev; uint64_t lsize = HDR_GET_LSIZE(hdr); uint64_t psize = HDR_GET_PSIZE(hdr); uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, psize); arc_buf_contents_t type = hdr->b_type; int64_t lsize_s; int64_t psize_s; int64_t asize_s; if (incr) { lsize_s = lsize; psize_s = psize; asize_s = asize; } else { lsize_s = -lsize; psize_s = -psize; asize_s = -asize; } /* If the buffer is a prefetch, count it as such. */ if (HDR_PREFETCH(hdr)) { ARCSTAT_INCR(arcstat_l2_prefetch_asize, asize_s); } else { /* * We use the value stored in the L2 header upon initial * caching in L2ARC. This value will be updated in case * an MRU/MRU_ghost buffer transitions to MFU but the L2ARC * metadata (log entry) cannot currently be updated. Having * the ARC state in the L2 header solves the problem of a * possibly absent L1 header (apparent in buffers restored * from persistent L2ARC). */ switch (hdr->b_l2hdr.b_arcs_state) { case ARC_STATE_MRU_GHOST: case ARC_STATE_MRU: ARCSTAT_INCR(arcstat_l2_mru_asize, asize_s); break; case ARC_STATE_MFU_GHOST: case ARC_STATE_MFU: ARCSTAT_INCR(arcstat_l2_mfu_asize, asize_s); break; default: break; } } if (state_only) return; ARCSTAT_INCR(arcstat_l2_psize, psize_s); ARCSTAT_INCR(arcstat_l2_lsize, lsize_s); switch (type) { case ARC_BUFC_DATA: ARCSTAT_INCR(arcstat_l2_bufc_data_asize, asize_s); break; case ARC_BUFC_METADATA: ARCSTAT_INCR(arcstat_l2_bufc_metadata_asize, asize_s); break; default: break; } } static void arc_hdr_l2hdr_destroy(arc_buf_hdr_t *hdr) { l2arc_buf_hdr_t *l2hdr = &hdr->b_l2hdr; l2arc_dev_t *dev = l2hdr->b_dev; uint64_t psize = HDR_GET_PSIZE(hdr); uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, psize); ASSERT(MUTEX_HELD(&dev->l2ad_mtx)); ASSERT(HDR_HAS_L2HDR(hdr)); list_remove(&dev->l2ad_buflist, hdr); l2arc_hdr_arcstats_decrement(hdr); vdev_space_update(dev->l2ad_vdev, -asize, 0, 0); (void) zfs_refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); } static void arc_hdr_destroy(arc_buf_hdr_t *hdr) { if (HDR_HAS_L1HDR(hdr)) { ASSERT(hdr->b_l1hdr.b_buf == NULL || hdr->b_l1hdr.b_bufcnt > 0); ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); } ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(!HDR_IN_HASH_TABLE(hdr)); if (HDR_HAS_L2HDR(hdr)) { l2arc_dev_t *dev = hdr->b_l2hdr.b_dev; boolean_t buflist_held = MUTEX_HELD(&dev->l2ad_mtx); if (!buflist_held) mutex_enter(&dev->l2ad_mtx); /* * Even though we checked this conditional above, we * need to check this again now that we have the * l2ad_mtx. This is because we could be racing with * another thread calling l2arc_evict() which might have * destroyed this header's L2 portion as we were waiting * to acquire the l2ad_mtx. If that happens, we don't * want to re-destroy the header's L2 portion. */ if (HDR_HAS_L2HDR(hdr)) arc_hdr_l2hdr_destroy(hdr); if (!buflist_held) mutex_exit(&dev->l2ad_mtx); } /* * The header's identify can only be safely discarded once it is no * longer discoverable. This requires removing it from the hash table * and the l2arc header list. After this point the hash lock can not * be used to protect the header. */ if (!HDR_EMPTY(hdr)) buf_discard_identity(hdr); if (HDR_HAS_L1HDR(hdr)) { arc_cksum_free(hdr); while (hdr->b_l1hdr.b_buf != NULL) arc_buf_destroy_impl(hdr->b_l1hdr.b_buf); if (hdr->b_l1hdr.b_pabd != NULL) arc_hdr_free_abd(hdr, B_FALSE); if (HDR_HAS_RABD(hdr)) arc_hdr_free_abd(hdr, B_TRUE); } ASSERT3P(hdr->b_hash_next, ==, NULL); if (HDR_HAS_L1HDR(hdr)) { ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); if (!HDR_PROTECTED(hdr)) { kmem_cache_free(hdr_full_cache, hdr); } else { kmem_cache_free(hdr_full_crypt_cache, hdr); } } else { kmem_cache_free(hdr_l2only_cache, hdr); } } void arc_buf_destroy(arc_buf_t *buf, void* tag) { arc_buf_hdr_t *hdr = buf->b_hdr; if (hdr->b_l1hdr.b_state == arc_anon) { ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); VERIFY0(remove_reference(hdr, NULL, tag)); arc_hdr_destroy(hdr); return; } kmutex_t *hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); ASSERT3P(hdr, ==, buf->b_hdr); ASSERT(hdr->b_l1hdr.b_bufcnt > 0); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3P(hdr->b_l1hdr.b_state, !=, arc_anon); ASSERT3P(buf->b_data, !=, NULL); (void) remove_reference(hdr, hash_lock, tag); arc_buf_destroy_impl(buf); mutex_exit(hash_lock); } /* * Evict the arc_buf_hdr that is provided as a parameter. The resultant * state of the header is dependent on its state prior to entering this * function. The following transitions are possible: * * - arc_mru -> arc_mru_ghost * - arc_mfu -> arc_mfu_ghost * - arc_mru_ghost -> arc_l2c_only * - arc_mru_ghost -> deleted * - arc_mfu_ghost -> arc_l2c_only * - arc_mfu_ghost -> deleted */ static int64_t arc_evict_hdr(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { arc_state_t *evicted_state, *state; int64_t bytes_evicted = 0; int min_lifetime = HDR_PRESCIENT_PREFETCH(hdr) ? arc_min_prescient_prefetch_ms : arc_min_prefetch_ms; ASSERT(MUTEX_HELD(hash_lock)); ASSERT(HDR_HAS_L1HDR(hdr)); state = hdr->b_l1hdr.b_state; if (GHOST_STATE(state)) { ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); /* * l2arc_write_buffers() relies on a header's L1 portion * (i.e. its b_pabd field) during it's write phase. * Thus, we cannot push a header onto the arc_l2c_only * state (removing its L1 piece) until the header is * done being written to the l2arc. */ if (HDR_HAS_L2HDR(hdr) && HDR_L2_WRITING(hdr)) { ARCSTAT_BUMP(arcstat_evict_l2_skip); return (bytes_evicted); } ARCSTAT_BUMP(arcstat_deleted); bytes_evicted += HDR_GET_LSIZE(hdr); DTRACE_PROBE1(arc__delete, arc_buf_hdr_t *, hdr); if (HDR_HAS_L2HDR(hdr)) { ASSERT(hdr->b_l1hdr.b_pabd == NULL); ASSERT(!HDR_HAS_RABD(hdr)); /* * This buffer is cached on the 2nd Level ARC; * don't destroy the header. */ arc_change_state(arc_l2c_only, hdr, hash_lock); /* * dropping from L1+L2 cached to L2-only, * realloc to remove the L1 header. */ hdr = arc_hdr_realloc(hdr, hdr_full_cache, hdr_l2only_cache); } else { arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); } return (bytes_evicted); } ASSERT(state == arc_mru || state == arc_mfu); evicted_state = (state == arc_mru) ? arc_mru_ghost : arc_mfu_ghost; /* prefetch buffers have a minimum lifespan */ if (HDR_IO_IN_PROGRESS(hdr) || ((hdr->b_flags & (ARC_FLAG_PREFETCH | ARC_FLAG_INDIRECT)) && ddi_get_lbolt() - hdr->b_l1hdr.b_arc_access < MSEC_TO_TICK(min_lifetime))) { ARCSTAT_BUMP(arcstat_evict_skip); return (bytes_evicted); } ASSERT0(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt)); while (hdr->b_l1hdr.b_buf) { arc_buf_t *buf = hdr->b_l1hdr.b_buf; if (!mutex_tryenter(&buf->b_evict_lock)) { ARCSTAT_BUMP(arcstat_mutex_miss); break; } if (buf->b_data != NULL) bytes_evicted += HDR_GET_LSIZE(hdr); mutex_exit(&buf->b_evict_lock); arc_buf_destroy_impl(buf); } if (HDR_HAS_L2HDR(hdr)) { ARCSTAT_INCR(arcstat_evict_l2_cached, HDR_GET_LSIZE(hdr)); } else { if (l2arc_write_eligible(hdr->b_spa, hdr)) { ARCSTAT_INCR(arcstat_evict_l2_eligible, HDR_GET_LSIZE(hdr)); switch (state->arcs_state) { case ARC_STATE_MRU: ARCSTAT_INCR( arcstat_evict_l2_eligible_mru, HDR_GET_LSIZE(hdr)); break; case ARC_STATE_MFU: ARCSTAT_INCR( arcstat_evict_l2_eligible_mfu, HDR_GET_LSIZE(hdr)); break; default: break; } } else { ARCSTAT_INCR(arcstat_evict_l2_ineligible, HDR_GET_LSIZE(hdr)); } } if (hdr->b_l1hdr.b_bufcnt == 0) { arc_cksum_free(hdr); bytes_evicted += arc_hdr_size(hdr); /* * If this hdr is being evicted and has a compressed * buffer then we discard it here before we change states. * This ensures that the accounting is updated correctly * in arc_free_data_impl(). */ if (hdr->b_l1hdr.b_pabd != NULL) arc_hdr_free_abd(hdr, B_FALSE); if (HDR_HAS_RABD(hdr)) arc_hdr_free_abd(hdr, B_TRUE); arc_change_state(evicted_state, hdr, hash_lock); ASSERT(HDR_IN_HASH_TABLE(hdr)); arc_hdr_set_flags(hdr, ARC_FLAG_IN_HASH_TABLE); DTRACE_PROBE1(arc__evict, arc_buf_hdr_t *, hdr); } return (bytes_evicted); } static void arc_set_need_free(void) { ASSERT(MUTEX_HELD(&arc_evict_lock)); int64_t remaining = arc_free_memory() - arc_sys_free / 2; arc_evict_waiter_t *aw = list_tail(&arc_evict_waiters); if (aw == NULL) { arc_need_free = MAX(-remaining, 0); } else { arc_need_free = MAX(-remaining, (int64_t)(aw->aew_count - arc_evict_count)); } } static uint64_t arc_evict_state_impl(multilist_t *ml, int idx, arc_buf_hdr_t *marker, uint64_t spa, int64_t bytes) { multilist_sublist_t *mls; uint64_t bytes_evicted = 0; arc_buf_hdr_t *hdr; kmutex_t *hash_lock; int evict_count = 0; ASSERT3P(marker, !=, NULL); IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); mls = multilist_sublist_lock(ml, idx); for (hdr = multilist_sublist_prev(mls, marker); hdr != NULL; hdr = multilist_sublist_prev(mls, marker)) { if ((bytes != ARC_EVICT_ALL && bytes_evicted >= bytes) || (evict_count >= zfs_arc_evict_batch_limit)) break; /* * To keep our iteration location, move the marker * forward. Since we're not holding hdr's hash lock, we * must be very careful and not remove 'hdr' from the * sublist. Otherwise, other consumers might mistake the * 'hdr' as not being on a sublist when they call the * multilist_link_active() function (they all rely on * the hash lock protecting concurrent insertions and * removals). multilist_sublist_move_forward() was * specifically implemented to ensure this is the case * (only 'marker' will be removed and re-inserted). */ multilist_sublist_move_forward(mls, marker); /* * The only case where the b_spa field should ever be * zero, is the marker headers inserted by * arc_evict_state(). It's possible for multiple threads * to be calling arc_evict_state() concurrently (e.g. * dsl_pool_close() and zio_inject_fault()), so we must * skip any markers we see from these other threads. */ if (hdr->b_spa == 0) continue; /* we're only interested in evicting buffers of a certain spa */ if (spa != 0 && hdr->b_spa != spa) { ARCSTAT_BUMP(arcstat_evict_skip); continue; } hash_lock = HDR_LOCK(hdr); /* * We aren't calling this function from any code path * that would already be holding a hash lock, so we're * asserting on this assumption to be defensive in case * this ever changes. Without this check, it would be * possible to incorrectly increment arcstat_mutex_miss * below (e.g. if the code changed such that we called * this function with a hash lock held). */ ASSERT(!MUTEX_HELD(hash_lock)); if (mutex_tryenter(hash_lock)) { uint64_t evicted = arc_evict_hdr(hdr, hash_lock); mutex_exit(hash_lock); bytes_evicted += evicted; /* * If evicted is zero, arc_evict_hdr() must have * decided to skip this header, don't increment * evict_count in this case. */ if (evicted != 0) evict_count++; } else { ARCSTAT_BUMP(arcstat_mutex_miss); } } multilist_sublist_unlock(mls); /* * Increment the count of evicted bytes, and wake up any threads that * are waiting for the count to reach this value. Since the list is * ordered by ascending aew_count, we pop off the beginning of the * list until we reach the end, or a waiter that's past the current * "count". Doing this outside the loop reduces the number of times * we need to acquire the global arc_evict_lock. * * Only wake when there's sufficient free memory in the system * (specifically, arc_sys_free/2, which by default is a bit more than * 1/64th of RAM). See the comments in arc_wait_for_eviction(). */ mutex_enter(&arc_evict_lock); arc_evict_count += bytes_evicted; if (arc_free_memory() > arc_sys_free / 2) { arc_evict_waiter_t *aw; while ((aw = list_head(&arc_evict_waiters)) != NULL && aw->aew_count <= arc_evict_count) { list_remove(&arc_evict_waiters, aw); cv_broadcast(&aw->aew_cv); } } arc_set_need_free(); mutex_exit(&arc_evict_lock); /* * If the ARC size is reduced from arc_c_max to arc_c_min (especially * if the average cached block is small), eviction can be on-CPU for * many seconds. To ensure that other threads that may be bound to * this CPU are able to make progress, make a voluntary preemption * call here. */ cond_resched(); return (bytes_evicted); } /* * Evict buffers from the given arc state, until we've removed the * specified number of bytes. Move the removed buffers to the * appropriate evict state. * * This function makes a "best effort". It skips over any buffers * it can't get a hash_lock on, and so, may not catch all candidates. * It may also return without evicting as much space as requested. * * If bytes is specified using the special value ARC_EVICT_ALL, this * will evict all available (i.e. unlocked and evictable) buffers from * the given arc state; which is used by arc_flush(). */ static uint64_t arc_evict_state(arc_state_t *state, uint64_t spa, int64_t bytes, arc_buf_contents_t type) { uint64_t total_evicted = 0; multilist_t *ml = &state->arcs_list[type]; int num_sublists; arc_buf_hdr_t **markers; IMPLY(bytes < 0, bytes == ARC_EVICT_ALL); num_sublists = multilist_get_num_sublists(ml); /* * If we've tried to evict from each sublist, made some * progress, but still have not hit the target number of bytes * to evict, we want to keep trying. The markers allow us to * pick up where we left off for each individual sublist, rather * than starting from the tail each time. */ markers = kmem_zalloc(sizeof (*markers) * num_sublists, KM_SLEEP); for (int i = 0; i < num_sublists; i++) { multilist_sublist_t *mls; markers[i] = kmem_cache_alloc(hdr_full_cache, KM_SLEEP); /* * A b_spa of 0 is used to indicate that this header is * a marker. This fact is used in arc_evict_type() and * arc_evict_state_impl(). */ markers[i]->b_spa = 0; mls = multilist_sublist_lock(ml, i); multilist_sublist_insert_tail(mls, markers[i]); multilist_sublist_unlock(mls); } /* * While we haven't hit our target number of bytes to evict, or * we're evicting all available buffers. */ while (total_evicted < bytes || bytes == ARC_EVICT_ALL) { int sublist_idx = multilist_get_random_index(ml); uint64_t scan_evicted = 0; /* * Try to reduce pinned dnodes with a floor of arc_dnode_limit. * Request that 10% of the LRUs be scanned by the superblock * shrinker. */ - if (type == ARC_BUFC_DATA && aggsum_compare(&astat_dnode_size, - arc_dnode_size_limit) > 0) { - arc_prune_async((aggsum_upper_bound(&astat_dnode_size) - + if (type == ARC_BUFC_DATA && aggsum_compare( + &arc_sums.arcstat_dnode_size, arc_dnode_size_limit) > 0) { + arc_prune_async((aggsum_upper_bound( + &arc_sums.arcstat_dnode_size) - arc_dnode_size_limit) / sizeof (dnode_t) / zfs_arc_dnode_reduce_percent); } /* * Start eviction using a randomly selected sublist, * this is to try and evenly balance eviction across all * sublists. Always starting at the same sublist * (e.g. index 0) would cause evictions to favor certain * sublists over others. */ for (int i = 0; i < num_sublists; i++) { uint64_t bytes_remaining; uint64_t bytes_evicted; if (bytes == ARC_EVICT_ALL) bytes_remaining = ARC_EVICT_ALL; else if (total_evicted < bytes) bytes_remaining = bytes - total_evicted; else break; bytes_evicted = arc_evict_state_impl(ml, sublist_idx, markers[sublist_idx], spa, bytes_remaining); scan_evicted += bytes_evicted; total_evicted += bytes_evicted; /* we've reached the end, wrap to the beginning */ if (++sublist_idx >= num_sublists) sublist_idx = 0; } /* * If we didn't evict anything during this scan, we have * no reason to believe we'll evict more during another * scan, so break the loop. */ if (scan_evicted == 0) { /* This isn't possible, let's make that obvious */ ASSERT3S(bytes, !=, 0); /* * When bytes is ARC_EVICT_ALL, the only way to * break the loop is when scan_evicted is zero. * In that case, we actually have evicted enough, * so we don't want to increment the kstat. */ if (bytes != ARC_EVICT_ALL) { ASSERT3S(total_evicted, <, bytes); ARCSTAT_BUMP(arcstat_evict_not_enough); } break; } } for (int i = 0; i < num_sublists; i++) { multilist_sublist_t *mls = multilist_sublist_lock(ml, i); multilist_sublist_remove(mls, markers[i]); multilist_sublist_unlock(mls); kmem_cache_free(hdr_full_cache, markers[i]); } kmem_free(markers, sizeof (*markers) * num_sublists); return (total_evicted); } /* * Flush all "evictable" data of the given type from the arc state * specified. This will not evict any "active" buffers (i.e. referenced). * * When 'retry' is set to B_FALSE, the function will make a single pass * over the state and evict any buffers that it can. Since it doesn't * continually retry the eviction, it might end up leaving some buffers * in the ARC due to lock misses. * * When 'retry' is set to B_TRUE, the function will continually retry the * eviction until *all* evictable buffers have been removed from the * state. As a result, if concurrent insertions into the state are * allowed (e.g. if the ARC isn't shutting down), this function might * wind up in an infinite loop, continually trying to evict buffers. */ static uint64_t arc_flush_state(arc_state_t *state, uint64_t spa, arc_buf_contents_t type, boolean_t retry) { uint64_t evicted = 0; while (zfs_refcount_count(&state->arcs_esize[type]) != 0) { evicted += arc_evict_state(state, spa, ARC_EVICT_ALL, type); if (!retry) break; } return (evicted); } /* * Evict the specified number of bytes from the state specified, * restricting eviction to the spa and type given. This function * prevents us from trying to evict more from a state's list than * is "evictable", and to skip evicting altogether when passed a * negative value for "bytes". In contrast, arc_evict_state() will * evict everything it can, when passed a negative value for "bytes". */ static uint64_t arc_evict_impl(arc_state_t *state, uint64_t spa, int64_t bytes, arc_buf_contents_t type) { int64_t delta; if (bytes > 0 && zfs_refcount_count(&state->arcs_esize[type]) > 0) { delta = MIN(zfs_refcount_count(&state->arcs_esize[type]), bytes); return (arc_evict_state(state, spa, delta, type)); } return (0); } /* * The goal of this function is to evict enough meta data buffers from the * ARC in order to enforce the arc_meta_limit. Achieving this is slightly * more complicated than it appears because it is common for data buffers * to have holds on meta data buffers. In addition, dnode meta data buffers * will be held by the dnodes in the block preventing them from being freed. * This means we can't simply traverse the ARC and expect to always find * enough unheld meta data buffer to release. * * Therefore, this function has been updated to make alternating passes * over the ARC releasing data buffers and then newly unheld meta data * buffers. This ensures forward progress is maintained and meta_used * will decrease. Normally this is sufficient, but if required the ARC * will call the registered prune callbacks causing dentry and inodes to * be dropped from the VFS cache. This will make dnode meta data buffers * available for reclaim. */ static uint64_t arc_evict_meta_balanced(uint64_t meta_used) { int64_t delta, prune = 0, adjustmnt; uint64_t total_evicted = 0; arc_buf_contents_t type = ARC_BUFC_DATA; int restarts = MAX(zfs_arc_meta_adjust_restarts, 0); restart: /* * This slightly differs than the way we evict from the mru in * arc_evict because we don't have a "target" value (i.e. no * "meta" arc_p). As a result, I think we can completely * cannibalize the metadata in the MRU before we evict the * metadata from the MFU. I think we probably need to implement a * "metadata arc_p" value to do this properly. */ adjustmnt = meta_used - arc_meta_limit; if (adjustmnt > 0 && zfs_refcount_count(&arc_mru->arcs_esize[type]) > 0) { delta = MIN(zfs_refcount_count(&arc_mru->arcs_esize[type]), adjustmnt); total_evicted += arc_evict_impl(arc_mru, 0, delta, type); adjustmnt -= delta; } /* * We can't afford to recalculate adjustmnt here. If we do, * new metadata buffers can sneak into the MRU or ANON lists, * thus penalize the MFU metadata. Although the fudge factor is * small, it has been empirically shown to be significant for * certain workloads (e.g. creating many empty directories). As * such, we use the original calculation for adjustmnt, and * simply decrement the amount of data evicted from the MRU. */ if (adjustmnt > 0 && zfs_refcount_count(&arc_mfu->arcs_esize[type]) > 0) { delta = MIN(zfs_refcount_count(&arc_mfu->arcs_esize[type]), adjustmnt); total_evicted += arc_evict_impl(arc_mfu, 0, delta, type); } adjustmnt = meta_used - arc_meta_limit; if (adjustmnt > 0 && zfs_refcount_count(&arc_mru_ghost->arcs_esize[type]) > 0) { delta = MIN(adjustmnt, zfs_refcount_count(&arc_mru_ghost->arcs_esize[type])); total_evicted += arc_evict_impl(arc_mru_ghost, 0, delta, type); adjustmnt -= delta; } if (adjustmnt > 0 && zfs_refcount_count(&arc_mfu_ghost->arcs_esize[type]) > 0) { delta = MIN(adjustmnt, zfs_refcount_count(&arc_mfu_ghost->arcs_esize[type])); total_evicted += arc_evict_impl(arc_mfu_ghost, 0, delta, type); } /* * If after attempting to make the requested adjustment to the ARC * the meta limit is still being exceeded then request that the * higher layers drop some cached objects which have holds on ARC * meta buffers. Requests to the upper layers will be made with * increasingly large scan sizes until the ARC is below the limit. */ if (meta_used > arc_meta_limit) { if (type == ARC_BUFC_DATA) { type = ARC_BUFC_METADATA; } else { type = ARC_BUFC_DATA; if (zfs_arc_meta_prune) { prune += zfs_arc_meta_prune; arc_prune_async(prune); } } if (restarts > 0) { restarts--; goto restart; } } return (total_evicted); } /* - * Evict metadata buffers from the cache, such that arc_meta_used is + * Evict metadata buffers from the cache, such that arcstat_meta_used is * capped by the arc_meta_limit tunable. */ static uint64_t arc_evict_meta_only(uint64_t meta_used) { uint64_t total_evicted = 0; int64_t target; /* * If we're over the meta limit, we want to evict enough * metadata to get back under the meta limit. We don't want to * evict so much that we drop the MRU below arc_p, though. If * we're over the meta limit more than we're over arc_p, we * evict some from the MRU here, and some from the MFU below. */ target = MIN((int64_t)(meta_used - arc_meta_limit), (int64_t)(zfs_refcount_count(&arc_anon->arcs_size) + zfs_refcount_count(&arc_mru->arcs_size) - arc_p)); total_evicted += arc_evict_impl(arc_mru, 0, target, ARC_BUFC_METADATA); /* * Similar to the above, we want to evict enough bytes to get us * below the meta limit, but not so much as to drop us below the * space allotted to the MFU (which is defined as arc_c - arc_p). */ target = MIN((int64_t)(meta_used - arc_meta_limit), (int64_t)(zfs_refcount_count(&arc_mfu->arcs_size) - (arc_c - arc_p))); total_evicted += arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); return (total_evicted); } static uint64_t arc_evict_meta(uint64_t meta_used) { if (zfs_arc_meta_strategy == ARC_STRATEGY_META_ONLY) return (arc_evict_meta_only(meta_used)); else return (arc_evict_meta_balanced(meta_used)); } /* * Return the type of the oldest buffer in the given arc state * * This function will select a random sublist of type ARC_BUFC_DATA and * a random sublist of type ARC_BUFC_METADATA. The tail of each sublist * is compared, and the type which contains the "older" buffer will be * returned. */ static arc_buf_contents_t arc_evict_type(arc_state_t *state) { multilist_t *data_ml = &state->arcs_list[ARC_BUFC_DATA]; multilist_t *meta_ml = &state->arcs_list[ARC_BUFC_METADATA]; int data_idx = multilist_get_random_index(data_ml); int meta_idx = multilist_get_random_index(meta_ml); multilist_sublist_t *data_mls; multilist_sublist_t *meta_mls; arc_buf_contents_t type; arc_buf_hdr_t *data_hdr; arc_buf_hdr_t *meta_hdr; /* * We keep the sublist lock until we're finished, to prevent * the headers from being destroyed via arc_evict_state(). */ data_mls = multilist_sublist_lock(data_ml, data_idx); meta_mls = multilist_sublist_lock(meta_ml, meta_idx); /* * These two loops are to ensure we skip any markers that * might be at the tail of the lists due to arc_evict_state(). */ for (data_hdr = multilist_sublist_tail(data_mls); data_hdr != NULL; data_hdr = multilist_sublist_prev(data_mls, data_hdr)) { if (data_hdr->b_spa != 0) break; } for (meta_hdr = multilist_sublist_tail(meta_mls); meta_hdr != NULL; meta_hdr = multilist_sublist_prev(meta_mls, meta_hdr)) { if (meta_hdr->b_spa != 0) break; } if (data_hdr == NULL && meta_hdr == NULL) { type = ARC_BUFC_DATA; } else if (data_hdr == NULL) { ASSERT3P(meta_hdr, !=, NULL); type = ARC_BUFC_METADATA; } else if (meta_hdr == NULL) { ASSERT3P(data_hdr, !=, NULL); type = ARC_BUFC_DATA; } else { ASSERT3P(data_hdr, !=, NULL); ASSERT3P(meta_hdr, !=, NULL); /* The headers can't be on the sublist without an L1 header */ ASSERT(HDR_HAS_L1HDR(data_hdr)); ASSERT(HDR_HAS_L1HDR(meta_hdr)); if (data_hdr->b_l1hdr.b_arc_access < meta_hdr->b_l1hdr.b_arc_access) { type = ARC_BUFC_DATA; } else { type = ARC_BUFC_METADATA; } } multilist_sublist_unlock(meta_mls); multilist_sublist_unlock(data_mls); return (type); } /* - * Evict buffers from the cache, such that arc_size is capped by arc_c. + * Evict buffers from the cache, such that arcstat_size is capped by arc_c. */ static uint64_t arc_evict(void) { uint64_t total_evicted = 0; uint64_t bytes; int64_t target; - uint64_t asize = aggsum_value(&arc_size); - uint64_t ameta = aggsum_value(&arc_meta_used); + uint64_t asize = aggsum_value(&arc_sums.arcstat_size); + uint64_t ameta = aggsum_value(&arc_sums.arcstat_meta_used); /* * If we're over arc_meta_limit, we want to correct that before * potentially evicting data buffers below. */ total_evicted += arc_evict_meta(ameta); /* * Adjust MRU size * * If we're over the target cache size, we want to evict enough * from the list to get back to our target size. We don't want * to evict too much from the MRU, such that it drops below * arc_p. So, if we're over our target cache size more than * the MRU is over arc_p, we'll evict enough to get back to * arc_p here, and then evict more from the MFU below. */ target = MIN((int64_t)(asize - arc_c), (int64_t)(zfs_refcount_count(&arc_anon->arcs_size) + zfs_refcount_count(&arc_mru->arcs_size) + ameta - arc_p)); /* * If we're below arc_meta_min, always prefer to evict data. * Otherwise, try to satisfy the requested number of bytes to * evict from the type which contains older buffers; in an * effort to keep newer buffers in the cache regardless of their * type. If we cannot satisfy the number of bytes from this * type, spill over into the next type. */ if (arc_evict_type(arc_mru) == ARC_BUFC_METADATA && ameta > arc_meta_min) { bytes = arc_evict_impl(arc_mru, 0, target, ARC_BUFC_METADATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * metadata, we try to get the rest from data. */ target -= bytes; total_evicted += arc_evict_impl(arc_mru, 0, target, ARC_BUFC_DATA); } else { bytes = arc_evict_impl(arc_mru, 0, target, ARC_BUFC_DATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * data, we try to get the rest from metadata. */ target -= bytes; total_evicted += arc_evict_impl(arc_mru, 0, target, ARC_BUFC_METADATA); } /* * Re-sum ARC stats after the first round of evictions. */ - asize = aggsum_value(&arc_size); - ameta = aggsum_value(&arc_meta_used); + asize = aggsum_value(&arc_sums.arcstat_size); + ameta = aggsum_value(&arc_sums.arcstat_meta_used); /* * Adjust MFU size * * Now that we've tried to evict enough from the MRU to get its * size back to arc_p, if we're still above the target cache * size, we evict the rest from the MFU. */ target = asize - arc_c; if (arc_evict_type(arc_mfu) == ARC_BUFC_METADATA && ameta > arc_meta_min) { bytes = arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * metadata, we try to get the rest from data. */ target -= bytes; total_evicted += arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_DATA); } else { bytes = arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_DATA); total_evicted += bytes; /* * If we couldn't evict our target number of bytes from * data, we try to get the rest from data. */ target -= bytes; total_evicted += arc_evict_impl(arc_mfu, 0, target, ARC_BUFC_METADATA); } /* * Adjust ghost lists * * In addition to the above, the ARC also defines target values * for the ghost lists. The sum of the mru list and mru ghost * list should never exceed the target size of the cache, and * the sum of the mru list, mfu list, mru ghost list, and mfu * ghost list should never exceed twice the target size of the * cache. The following logic enforces these limits on the ghost * caches, and evicts from them as needed. */ target = zfs_refcount_count(&arc_mru->arcs_size) + zfs_refcount_count(&arc_mru_ghost->arcs_size) - arc_c; bytes = arc_evict_impl(arc_mru_ghost, 0, target, ARC_BUFC_DATA); total_evicted += bytes; target -= bytes; total_evicted += arc_evict_impl(arc_mru_ghost, 0, target, ARC_BUFC_METADATA); /* * We assume the sum of the mru list and mfu list is less than * or equal to arc_c (we enforced this above), which means we * can use the simpler of the two equations below: * * mru + mfu + mru ghost + mfu ghost <= 2 * arc_c * mru ghost + mfu ghost <= arc_c */ target = zfs_refcount_count(&arc_mru_ghost->arcs_size) + zfs_refcount_count(&arc_mfu_ghost->arcs_size) - arc_c; bytes = arc_evict_impl(arc_mfu_ghost, 0, target, ARC_BUFC_DATA); total_evicted += bytes; target -= bytes; total_evicted += arc_evict_impl(arc_mfu_ghost, 0, target, ARC_BUFC_METADATA); return (total_evicted); } void arc_flush(spa_t *spa, boolean_t retry) { uint64_t guid = 0; /* * If retry is B_TRUE, a spa must not be specified since we have * no good way to determine if all of a spa's buffers have been * evicted from an arc state. */ ASSERT(!retry || spa == 0); if (spa != NULL) guid = spa_load_guid(spa); (void) arc_flush_state(arc_mru, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mru, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mfu, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mru_ghost, guid, ARC_BUFC_METADATA, retry); (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_DATA, retry); (void) arc_flush_state(arc_mfu_ghost, guid, ARC_BUFC_METADATA, retry); } void arc_reduce_target_size(int64_t to_free) { - uint64_t asize = aggsum_value(&arc_size); + uint64_t asize = aggsum_value(&arc_sums.arcstat_size); /* * All callers want the ARC to actually evict (at least) this much * memory. Therefore we reduce from the lower of the current size and * the target size. This way, even if arc_c is much higher than * arc_size (as can be the case after many calls to arc_freed(), we will * immediately have arc_c < arc_size and therefore the arc_evict_zthr * will evict. */ uint64_t c = MIN(arc_c, asize); if (c > to_free && c - to_free > arc_c_min) { arc_c = c - to_free; atomic_add_64(&arc_p, -(arc_p >> arc_shrink_shift)); if (arc_p > arc_c) arc_p = (arc_c >> 1); ASSERT(arc_c >= arc_c_min); ASSERT((int64_t)arc_p >= 0); } else { arc_c = arc_c_min; } if (asize > arc_c) { /* See comment in arc_evict_cb_check() on why lock+flag */ mutex_enter(&arc_evict_lock); arc_evict_needed = B_TRUE; mutex_exit(&arc_evict_lock); zthr_wakeup(arc_evict_zthr); } } /* * Determine if the system is under memory pressure and is asking * to reclaim memory. A return value of B_TRUE indicates that the system * is under memory pressure and that the arc should adjust accordingly. */ boolean_t arc_reclaim_needed(void) { return (arc_available_memory() < 0); } void arc_kmem_reap_soon(void) { size_t i; kmem_cache_t *prev_cache = NULL; kmem_cache_t *prev_data_cache = NULL; extern kmem_cache_t *zio_buf_cache[]; extern kmem_cache_t *zio_data_buf_cache[]; #ifdef _KERNEL - if ((aggsum_compare(&arc_meta_used, arc_meta_limit) >= 0) && - zfs_arc_meta_prune) { + if ((aggsum_compare(&arc_sums.arcstat_meta_used, + arc_meta_limit) >= 0) && zfs_arc_meta_prune) { /* * We are exceeding our meta-data cache limit. * Prune some entries to release holds on meta-data. */ arc_prune_async(zfs_arc_meta_prune); } #if defined(_ILP32) /* * Reclaim unused memory from all kmem caches. */ kmem_reap(); #endif #endif for (i = 0; i < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; i++) { #if defined(_ILP32) /* reach upper limit of cache size on 32-bit */ if (zio_buf_cache[i] == NULL) break; #endif if (zio_buf_cache[i] != prev_cache) { prev_cache = zio_buf_cache[i]; kmem_cache_reap_now(zio_buf_cache[i]); } if (zio_data_buf_cache[i] != prev_data_cache) { prev_data_cache = zio_data_buf_cache[i]; kmem_cache_reap_now(zio_data_buf_cache[i]); } } kmem_cache_reap_now(buf_cache); kmem_cache_reap_now(hdr_full_cache); kmem_cache_reap_now(hdr_l2only_cache); kmem_cache_reap_now(zfs_btree_leaf_cache); abd_cache_reap_now(); } /* ARGSUSED */ static boolean_t arc_evict_cb_check(void *arg, zthr_t *zthr) { #ifdef ZFS_DEBUG /* * This is necessary in order to keep the kstat information * up to date for tools that display kstat data such as the * mdb ::arc dcmd and the Linux crash utility. These tools * typically do not call kstat's update function, but simply * dump out stats from the most recent update. Without * this call, these commands may show stale stats for the * anon, mru, mru_ghost, mfu, and mfu_ghost lists. Even * with this call, the data might be out of date if the * evict thread hasn't been woken recently; but that should * suffice. The arc_state_t structures can be queried * directly if more accurate information is needed. */ if (arc_ksp != NULL) arc_ksp->ks_update(arc_ksp, KSTAT_READ); #endif /* * We have to rely on arc_wait_for_eviction() to tell us when to * evict, rather than checking if we are overflowing here, so that we * are sure to not leave arc_wait_for_eviction() waiting on aew_cv. * If we have become "not overflowing" since arc_wait_for_eviction() * checked, we need to wake it up. We could broadcast the CV here, * but arc_wait_for_eviction() may have not yet gone to sleep. We * would need to use a mutex to ensure that this function doesn't * broadcast until arc_wait_for_eviction() has gone to sleep (e.g. * the arc_evict_lock). However, the lock ordering of such a lock * would necessarily be incorrect with respect to the zthr_lock, * which is held before this function is called, and is held by * arc_wait_for_eviction() when it calls zthr_wakeup(). */ return (arc_evict_needed); } /* * Keep arc_size under arc_c by running arc_evict which evicts data * from the ARC. */ /* ARGSUSED */ static void arc_evict_cb(void *arg, zthr_t *zthr) { uint64_t evicted = 0; fstrans_cookie_t cookie = spl_fstrans_mark(); /* Evict from cache */ evicted = arc_evict(); /* * If evicted is zero, we couldn't evict anything * via arc_evict(). This could be due to hash lock * collisions, but more likely due to the majority of * arc buffers being unevictable. Therefore, even if * arc_size is above arc_c, another pass is unlikely to * be helpful and could potentially cause us to enter an * infinite loop. Additionally, zthr_iscancelled() is * checked here so that if the arc is shutting down, the * broadcast will wake any remaining arc evict waiters. */ mutex_enter(&arc_evict_lock); arc_evict_needed = !zthr_iscancelled(arc_evict_zthr) && - evicted > 0 && aggsum_compare(&arc_size, arc_c) > 0; + evicted > 0 && aggsum_compare(&arc_sums.arcstat_size, arc_c) > 0; if (!arc_evict_needed) { /* * We're either no longer overflowing, or we * can't evict anything more, so we should wake * arc_get_data_impl() sooner. */ arc_evict_waiter_t *aw; while ((aw = list_remove_head(&arc_evict_waiters)) != NULL) { cv_broadcast(&aw->aew_cv); } arc_set_need_free(); } mutex_exit(&arc_evict_lock); spl_fstrans_unmark(cookie); } /* ARGSUSED */ static boolean_t arc_reap_cb_check(void *arg, zthr_t *zthr) { int64_t free_memory = arc_available_memory(); static int reap_cb_check_counter = 0; /* * If a kmem reap is already active, don't schedule more. We must * check for this because kmem_cache_reap_soon() won't actually * block on the cache being reaped (this is to prevent callers from * becoming implicitly blocked by a system-wide kmem reap -- which, * on a system with many, many full magazines, can take minutes). */ if (!kmem_cache_reap_active() && free_memory < 0) { arc_no_grow = B_TRUE; arc_warm = B_TRUE; /* * Wait at least zfs_grow_retry (default 5) seconds * before considering growing. */ arc_growtime = gethrtime() + SEC2NSEC(arc_grow_retry); return (B_TRUE); } else if (free_memory < arc_c >> arc_no_grow_shift) { arc_no_grow = B_TRUE; } else if (gethrtime() >= arc_growtime) { arc_no_grow = B_FALSE; } /* * Called unconditionally every 60 seconds to reclaim unused * zstd compression and decompression context. This is done * here to avoid the need for an independent thread. */ if (!((reap_cb_check_counter++) % 60)) zfs_zstd_cache_reap_now(); return (B_FALSE); } /* * Keep enough free memory in the system by reaping the ARC's kmem * caches. To cause more slabs to be reapable, we may reduce the * target size of the cache (arc_c), causing the arc_evict_cb() * to free more buffers. */ /* ARGSUSED */ static void arc_reap_cb(void *arg, zthr_t *zthr) { int64_t free_memory; fstrans_cookie_t cookie = spl_fstrans_mark(); /* * Kick off asynchronous kmem_reap()'s of all our caches. */ arc_kmem_reap_soon(); /* * Wait at least arc_kmem_cache_reap_retry_ms between * arc_kmem_reap_soon() calls. Without this check it is possible to * end up in a situation where we spend lots of time reaping * caches, while we're near arc_c_min. Waiting here also gives the * subsequent free memory check a chance of finding that the * asynchronous reap has already freed enough memory, and we don't * need to call arc_reduce_target_size(). */ delay((hz * arc_kmem_cache_reap_retry_ms + 999) / 1000); /* * Reduce the target size as needed to maintain the amount of free * memory in the system at a fraction of the arc_size (1/128th by * default). If oversubscribed (free_memory < 0) then reduce the * target arc_size by the deficit amount plus the fractional * amount. If free memory is positive but less than the fractional * amount, reduce by what is needed to hit the fractional amount. */ free_memory = arc_available_memory(); int64_t to_free = (arc_c >> arc_shrink_shift) - free_memory; if (to_free > 0) { arc_reduce_target_size(to_free); } spl_fstrans_unmark(cookie); } #ifdef _KERNEL /* * Determine the amount of memory eligible for eviction contained in the * ARC. All clean data reported by the ghost lists can always be safely * evicted. Due to arc_c_min, the same does not hold for all clean data * contained by the regular mru and mfu lists. * * In the case of the regular mru and mfu lists, we need to report as * much clean data as possible, such that evicting that same reported * data will not bring arc_size below arc_c_min. Thus, in certain * circumstances, the total amount of clean data in the mru and mfu * lists might not actually be evictable. * * The following two distinct cases are accounted for: * * 1. The sum of the amount of dirty data contained by both the mru and * mfu lists, plus the ARC's other accounting (e.g. the anon list), * is greater than or equal to arc_c_min. * (i.e. amount of dirty data >= arc_c_min) * * This is the easy case; all clean data contained by the mru and mfu * lists is evictable. Evicting all clean data can only drop arc_size * to the amount of dirty data, which is greater than arc_c_min. * * 2. The sum of the amount of dirty data contained by both the mru and * mfu lists, plus the ARC's other accounting (e.g. the anon list), * is less than arc_c_min. * (i.e. arc_c_min > amount of dirty data) * * 2.1. arc_size is greater than or equal arc_c_min. * (i.e. arc_size >= arc_c_min > amount of dirty data) * * In this case, not all clean data from the regular mru and mfu * lists is actually evictable; we must leave enough clean data * to keep arc_size above arc_c_min. Thus, the maximum amount of * evictable data from the two lists combined, is exactly the * difference between arc_size and arc_c_min. * * 2.2. arc_size is less than arc_c_min * (i.e. arc_c_min > arc_size > amount of dirty data) * * In this case, none of the data contained in the mru and mfu * lists is evictable, even if it's clean. Since arc_size is * already below arc_c_min, evicting any more would only * increase this negative difference. */ #endif /* _KERNEL */ /* * Adapt arc info given the number of bytes we are trying to add and * the state that we are coming from. This function is only called * when we are adding new content to the cache. */ static void arc_adapt(int bytes, arc_state_t *state) { int mult; uint64_t arc_p_min = (arc_c >> arc_p_min_shift); int64_t mrug_size = zfs_refcount_count(&arc_mru_ghost->arcs_size); int64_t mfug_size = zfs_refcount_count(&arc_mfu_ghost->arcs_size); ASSERT(bytes > 0); /* * Adapt the target size of the MRU list: * - if we just hit in the MRU ghost list, then increase * the target size of the MRU list. * - if we just hit in the MFU ghost list, then increase * the target size of the MFU list by decreasing the * target size of the MRU list. */ if (state == arc_mru_ghost) { mult = (mrug_size >= mfug_size) ? 1 : (mfug_size / mrug_size); if (!zfs_arc_p_dampener_disable) mult = MIN(mult, 10); /* avoid wild arc_p adjustment */ arc_p = MIN(arc_c - arc_p_min, arc_p + bytes * mult); } else if (state == arc_mfu_ghost) { uint64_t delta; mult = (mfug_size >= mrug_size) ? 1 : (mrug_size / mfug_size); if (!zfs_arc_p_dampener_disable) mult = MIN(mult, 10); delta = MIN(bytes * mult, arc_p); arc_p = MAX(arc_p_min, arc_p - delta); } ASSERT((int64_t)arc_p >= 0); /* * Wake reap thread if we do not have any available memory */ if (arc_reclaim_needed()) { zthr_wakeup(arc_reap_zthr); return; } if (arc_no_grow) return; if (arc_c >= arc_c_max) return; /* * If we're within (2 * maxblocksize) bytes of the target * cache size, increment the target cache size */ ASSERT3U(arc_c, >=, 2ULL << SPA_MAXBLOCKSHIFT); - if (aggsum_upper_bound(&arc_size) >= + if (aggsum_upper_bound(&arc_sums.arcstat_size) >= arc_c - (2ULL << SPA_MAXBLOCKSHIFT)) { atomic_add_64(&arc_c, (int64_t)bytes); if (arc_c > arc_c_max) arc_c = arc_c_max; else if (state == arc_anon) atomic_add_64(&arc_p, (int64_t)bytes); if (arc_p > arc_c) arc_p = arc_c; } ASSERT((int64_t)arc_p >= 0); } /* * Check if arc_size has grown past our upper threshold, determined by * zfs_arc_overflow_shift. */ boolean_t arc_is_overflowing(void) { /* Always allow at least one block of overflow */ int64_t overflow = MAX(SPA_MAXBLOCKSIZE, arc_c >> zfs_arc_overflow_shift); /* * We just compare the lower bound here for performance reasons. Our * primary goals are to make sure that the arc never grows without * bound, and that it can reach its maximum size. This check * accomplishes both goals. The maximum amount we could run over by is * 2 * aggsum_borrow_multiplier * NUM_CPUS * the average size of a block * in the ARC. In practice, that's in the tens of MB, which is low * enough to be safe. */ - return (aggsum_lower_bound(&arc_size) >= (int64_t)arc_c + overflow); + return (aggsum_lower_bound(&arc_sums.arcstat_size) >= + (int64_t)arc_c + overflow); } static abd_t * arc_get_data_abd(arc_buf_hdr_t *hdr, uint64_t size, void *tag, boolean_t do_adapt) { arc_buf_contents_t type = arc_buf_type(hdr); arc_get_data_impl(hdr, size, tag, do_adapt); if (type == ARC_BUFC_METADATA) { return (abd_alloc(size, B_TRUE)); } else { ASSERT(type == ARC_BUFC_DATA); return (abd_alloc(size, B_FALSE)); } } static void * arc_get_data_buf(arc_buf_hdr_t *hdr, uint64_t size, void *tag) { arc_buf_contents_t type = arc_buf_type(hdr); arc_get_data_impl(hdr, size, tag, B_TRUE); if (type == ARC_BUFC_METADATA) { return (zio_buf_alloc(size)); } else { ASSERT(type == ARC_BUFC_DATA); return (zio_data_buf_alloc(size)); } } /* * Wait for the specified amount of data (in bytes) to be evicted from the * ARC, and for there to be sufficient free memory in the system. Waiting for * eviction ensures that the memory used by the ARC decreases. Waiting for * free memory ensures that the system won't run out of free pages, regardless * of ARC behavior and settings. See arc_lowmem_init(). */ void arc_wait_for_eviction(uint64_t amount) { mutex_enter(&arc_evict_lock); if (arc_is_overflowing()) { arc_evict_needed = B_TRUE; zthr_wakeup(arc_evict_zthr); if (amount != 0) { arc_evict_waiter_t aw; list_link_init(&aw.aew_node); cv_init(&aw.aew_cv, NULL, CV_DEFAULT, NULL); uint64_t last_count = 0; if (!list_is_empty(&arc_evict_waiters)) { arc_evict_waiter_t *last = list_tail(&arc_evict_waiters); last_count = last->aew_count; } /* * Note, the last waiter's count may be less than * arc_evict_count if we are low on memory in which * case arc_evict_state_impl() may have deferred * wakeups (but still incremented arc_evict_count). */ aw.aew_count = MAX(last_count, arc_evict_count) + amount; list_insert_tail(&arc_evict_waiters, &aw); arc_set_need_free(); DTRACE_PROBE3(arc__wait__for__eviction, uint64_t, amount, uint64_t, arc_evict_count, uint64_t, aw.aew_count); /* * We will be woken up either when arc_evict_count * reaches aew_count, or when the ARC is no longer * overflowing and eviction completes. */ cv_wait(&aw.aew_cv, &arc_evict_lock); /* * In case of "false" wakeup, we will still be on the * list. */ if (list_link_active(&aw.aew_node)) list_remove(&arc_evict_waiters, &aw); cv_destroy(&aw.aew_cv); } } mutex_exit(&arc_evict_lock); } /* * Allocate a block and return it to the caller. If we are hitting the * hard limit for the cache size, we must sleep, waiting for the eviction * thread to catch up. If we're past the target size but below the hard * limit, we'll only signal the reclaim thread and continue on. */ static void arc_get_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag, boolean_t do_adapt) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); if (do_adapt) arc_adapt(size, state); /* * If arc_size is currently overflowing, we must be adding data * faster than we are evicting. To ensure we don't compound the * problem by adding more data and forcing arc_size to grow even * further past it's target size, we wait for the eviction thread to * make some progress. We also wait for there to be sufficient free * memory in the system, as measured by arc_free_memory(). * * Specifically, we wait for zfs_arc_eviction_pct percent of the * requested size to be evicted. This should be more than 100%, to * ensure that that progress is also made towards getting arc_size * under arc_c. See the comment above zfs_arc_eviction_pct. * * We do the overflowing check without holding the arc_evict_lock to * reduce lock contention in this hot path. Note that * arc_wait_for_eviction() will acquire the lock and check again to * ensure we are truly overflowing before blocking. */ if (arc_is_overflowing()) { arc_wait_for_eviction(size * zfs_arc_eviction_pct / 100); } VERIFY3U(hdr->b_type, ==, type); if (type == ARC_BUFC_METADATA) { arc_space_consume(size, ARC_SPACE_META); } else { arc_space_consume(size, ARC_SPACE_DATA); } /* * Update the state size. Note that ghost states have a * "ghost size" and so don't need to be updated. */ if (!GHOST_STATE(state)) { (void) zfs_refcount_add_many(&state->arcs_size, size, tag); /* * If this is reached via arc_read, the link is * protected by the hash lock. If reached via * arc_buf_alloc, the header should not be accessed by * any other thread. And, if reached via arc_read_done, * the hash lock will protect it if it's found in the * hash table; otherwise no other thread should be * trying to [add|remove]_reference it. */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); (void) zfs_refcount_add_many(&state->arcs_esize[type], size, tag); } /* * If we are growing the cache, and we are adding anonymous * data, and we have outgrown arc_p, update arc_p */ - if (aggsum_upper_bound(&arc_size) < arc_c && + if (aggsum_upper_bound(&arc_sums.arcstat_size) < arc_c && hdr->b_l1hdr.b_state == arc_anon && (zfs_refcount_count(&arc_anon->arcs_size) + zfs_refcount_count(&arc_mru->arcs_size) > arc_p)) arc_p = MIN(arc_c, arc_p + size); } } static void arc_free_data_abd(arc_buf_hdr_t *hdr, abd_t *abd, uint64_t size, void *tag) { arc_free_data_impl(hdr, size, tag); abd_free(abd); } static void arc_free_data_buf(arc_buf_hdr_t *hdr, void *buf, uint64_t size, void *tag) { arc_buf_contents_t type = arc_buf_type(hdr); arc_free_data_impl(hdr, size, tag); if (type == ARC_BUFC_METADATA) { zio_buf_free(buf, size); } else { ASSERT(type == ARC_BUFC_DATA); zio_data_buf_free(buf, size); } } /* * Free the arc data buffer. */ static void arc_free_data_impl(arc_buf_hdr_t *hdr, uint64_t size, void *tag) { arc_state_t *state = hdr->b_l1hdr.b_state; arc_buf_contents_t type = arc_buf_type(hdr); /* protected by hash lock, if in the hash table */ if (multilist_link_active(&hdr->b_l1hdr.b_arc_node)) { ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); ASSERT(state != arc_anon && state != arc_l2c_only); (void) zfs_refcount_remove_many(&state->arcs_esize[type], size, tag); } (void) zfs_refcount_remove_many(&state->arcs_size, size, tag); VERIFY3U(hdr->b_type, ==, type); if (type == ARC_BUFC_METADATA) { arc_space_return(size, ARC_SPACE_META); } else { ASSERT(type == ARC_BUFC_DATA); arc_space_return(size, ARC_SPACE_DATA); } } /* * This routine is called whenever a buffer is accessed. * NOTE: the hash lock is dropped in this function. */ static void arc_access(arc_buf_hdr_t *hdr, kmutex_t *hash_lock) { clock_t now; ASSERT(MUTEX_HELD(hash_lock)); ASSERT(HDR_HAS_L1HDR(hdr)); if (hdr->b_l1hdr.b_state == arc_anon) { /* * This buffer is not in the cache, and does not * appear in our "ghost" list. Add the new buffer * to the MRU state. */ ASSERT0(hdr->b_l1hdr.b_arc_access); hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); arc_change_state(arc_mru, hdr, hash_lock); } else if (hdr->b_l1hdr.b_state == arc_mru) { now = ddi_get_lbolt(); /* * If this buffer is here because of a prefetch, then either: * - clear the flag if this is a "referencing" read * (any subsequent access will bump this into the MFU state). * or * - move the buffer to the head of the list if this is * another prefetch (to make it less likely to be evicted). */ if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) { if (zfs_refcount_count(&hdr->b_l1hdr.b_refcnt) == 0) { /* link protected by hash lock */ ASSERT(multilist_link_active( &hdr->b_l1hdr.b_arc_node)); } else { if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_decrement_state(hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH | ARC_FLAG_PRESCIENT_PREFETCH); atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); ARCSTAT_BUMP(arcstat_mru_hits); if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_increment_state(hdr); } hdr->b_l1hdr.b_arc_access = now; return; } /* * This buffer has been "accessed" only once so far, * but it is still in the cache. Move it to the MFU * state. */ if (ddi_time_after(now, hdr->b_l1hdr.b_arc_access + ARC_MINTIME)) { /* * More than 125ms have passed since we * instantiated this buffer. Move it to the * most frequently used state. */ hdr->b_l1hdr.b_arc_access = now; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(arc_mfu, hdr, hash_lock); } atomic_inc_32(&hdr->b_l1hdr.b_mru_hits); ARCSTAT_BUMP(arcstat_mru_hits); } else if (hdr->b_l1hdr.b_state == arc_mru_ghost) { arc_state_t *new_state; /* * This buffer has been "accessed" recently, but * was evicted from the cache. Move it to the * MFU state. */ if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) { new_state = arc_mru; if (zfs_refcount_count(&hdr->b_l1hdr.b_refcnt) > 0) { if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_decrement_state(hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_PREFETCH | ARC_FLAG_PRESCIENT_PREFETCH); if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_increment_state(hdr); } DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, hdr); } else { new_state = arc_mfu; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); } hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); arc_change_state(new_state, hdr, hash_lock); atomic_inc_32(&hdr->b_l1hdr.b_mru_ghost_hits); ARCSTAT_BUMP(arcstat_mru_ghost_hits); } else if (hdr->b_l1hdr.b_state == arc_mfu) { /* * This buffer has been accessed more than once and is * still in the cache. Keep it in the MFU state. * * NOTE: an add_reference() that occurred when we did * the arc_read() will have kicked this off the list. * If it was a prefetch, we will explicitly move it to * the head of the list now. */ atomic_inc_32(&hdr->b_l1hdr.b_mfu_hits); ARCSTAT_BUMP(arcstat_mfu_hits); hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); } else if (hdr->b_l1hdr.b_state == arc_mfu_ghost) { arc_state_t *new_state = arc_mfu; /* * This buffer has been accessed more than once but has * been evicted from the cache. Move it back to the * MFU state. */ if (HDR_PREFETCH(hdr) || HDR_PRESCIENT_PREFETCH(hdr)) { /* * This is a prefetch access... * move this block back to the MRU state. */ new_state = arc_mru; } hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(new_state, hdr, hash_lock); atomic_inc_32(&hdr->b_l1hdr.b_mfu_ghost_hits); ARCSTAT_BUMP(arcstat_mfu_ghost_hits); } else if (hdr->b_l1hdr.b_state == arc_l2c_only) { /* * This buffer is on the 2nd Level ARC. */ hdr->b_l1hdr.b_arc_access = ddi_get_lbolt(); DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, hdr); arc_change_state(arc_mfu, hdr, hash_lock); } else { cmn_err(CE_PANIC, "invalid arc state 0x%p", hdr->b_l1hdr.b_state); } } /* * This routine is called by dbuf_hold() to update the arc_access() state * which otherwise would be skipped for entries in the dbuf cache. */ void arc_buf_access(arc_buf_t *buf) { mutex_enter(&buf->b_evict_lock); arc_buf_hdr_t *hdr = buf->b_hdr; /* * Avoid taking the hash_lock when possible as an optimization. * The header must be checked again under the hash_lock in order * to handle the case where it is concurrently being released. */ if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) { mutex_exit(&buf->b_evict_lock); return; } kmutex_t *hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); if (hdr->b_l1hdr.b_state == arc_anon || HDR_EMPTY(hdr)) { mutex_exit(hash_lock); mutex_exit(&buf->b_evict_lock); ARCSTAT_BUMP(arcstat_access_skip); return; } mutex_exit(&buf->b_evict_lock); ASSERT(hdr->b_l1hdr.b_state == arc_mru || hdr->b_l1hdr.b_state == arc_mfu); DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); arc_access(hdr, hash_lock); mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_hits); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr) && !HDR_PRESCIENT_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, hits); } /* a generic arc_read_done_func_t which you can use */ /* ARGSUSED */ void arc_bcopy_func(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp, arc_buf_t *buf, void *arg) { if (buf == NULL) return; bcopy(buf->b_data, arg, arc_buf_size(buf)); arc_buf_destroy(buf, arg); } /* a generic arc_read_done_func_t */ /* ARGSUSED */ void arc_getbuf_func(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp, arc_buf_t *buf, void *arg) { arc_buf_t **bufp = arg; if (buf == NULL) { ASSERT(zio == NULL || zio->io_error != 0); *bufp = NULL; } else { ASSERT(zio == NULL || zio->io_error == 0); *bufp = buf; ASSERT(buf->b_data != NULL); } } static void arc_hdr_verify(arc_buf_hdr_t *hdr, blkptr_t *bp) { if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) { ASSERT3U(HDR_GET_PSIZE(hdr), ==, 0); ASSERT3U(arc_hdr_get_compress(hdr), ==, ZIO_COMPRESS_OFF); } else { if (HDR_COMPRESSION_ENABLED(hdr)) { ASSERT3U(arc_hdr_get_compress(hdr), ==, BP_GET_COMPRESS(bp)); } ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp)); ASSERT3U(HDR_GET_PSIZE(hdr), ==, BP_GET_PSIZE(bp)); ASSERT3U(!!HDR_PROTECTED(hdr), ==, BP_IS_PROTECTED(bp)); } } static void arc_read_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; arc_buf_hdr_t *hdr = zio->io_private; kmutex_t *hash_lock = NULL; arc_callback_t *callback_list; arc_callback_t *acb; boolean_t freeable = B_FALSE; /* * The hdr was inserted into hash-table and removed from lists * prior to starting I/O. We should find this header, since * it's in the hash table, and it should be legit since it's * not possible to evict it during the I/O. The only possible * reason for it not to be found is if we were freed during the * read. */ if (HDR_IN_HASH_TABLE(hdr)) { arc_buf_hdr_t *found; ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp)); ASSERT3U(hdr->b_dva.dva_word[0], ==, BP_IDENTITY(zio->io_bp)->dva_word[0]); ASSERT3U(hdr->b_dva.dva_word[1], ==, BP_IDENTITY(zio->io_bp)->dva_word[1]); found = buf_hash_find(hdr->b_spa, zio->io_bp, &hash_lock); ASSERT((found == hdr && DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) || (found == hdr && HDR_L2_READING(hdr))); ASSERT3P(hash_lock, !=, NULL); } if (BP_IS_PROTECTED(bp)) { hdr->b_crypt_hdr.b_ot = BP_GET_TYPE(bp); hdr->b_crypt_hdr.b_dsobj = zio->io_bookmark.zb_objset; zio_crypt_decode_params_bp(bp, hdr->b_crypt_hdr.b_salt, hdr->b_crypt_hdr.b_iv); if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) { void *tmpbuf; tmpbuf = abd_borrow_buf_copy(zio->io_abd, sizeof (zil_chain_t)); zio_crypt_decode_mac_zil(tmpbuf, hdr->b_crypt_hdr.b_mac); abd_return_buf(zio->io_abd, tmpbuf, sizeof (zil_chain_t)); } else { zio_crypt_decode_mac_bp(bp, hdr->b_crypt_hdr.b_mac); } } if (zio->io_error == 0) { /* byteswap if necessary */ if (BP_SHOULD_BYTESWAP(zio->io_bp)) { if (BP_GET_LEVEL(zio->io_bp) > 0) { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64; } else { hdr->b_l1hdr.b_byteswap = DMU_OT_BYTESWAP(BP_GET_TYPE(zio->io_bp)); } } else { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; } if (!HDR_L2_READING(hdr)) { hdr->b_complevel = zio->io_prop.zp_complevel; } } arc_hdr_clear_flags(hdr, ARC_FLAG_L2_EVICTED); if (l2arc_noprefetch && HDR_PREFETCH(hdr)) arc_hdr_clear_flags(hdr, ARC_FLAG_L2CACHE); callback_list = hdr->b_l1hdr.b_acb; ASSERT3P(callback_list, !=, NULL); if (hash_lock && zio->io_error == 0 && hdr->b_l1hdr.b_state == arc_anon) { /* * Only call arc_access on anonymous buffers. This is because * if we've issued an I/O for an evicted buffer, we've already * called arc_access (to prevent any simultaneous readers from * getting confused). */ arc_access(hdr, hash_lock); } /* * If a read request has a callback (i.e. acb_done is not NULL), then we * make a buf containing the data according to the parameters which were * passed in. The implementation of arc_buf_alloc_impl() ensures that we * aren't needlessly decompressing the data multiple times. */ int callback_cnt = 0; for (acb = callback_list; acb != NULL; acb = acb->acb_next) { if (!acb->acb_done || acb->acb_nobuf) continue; callback_cnt++; if (zio->io_error != 0) continue; int error = arc_buf_alloc_impl(hdr, zio->io_spa, &acb->acb_zb, acb->acb_private, acb->acb_encrypted, acb->acb_compressed, acb->acb_noauth, B_TRUE, &acb->acb_buf); /* * Assert non-speculative zios didn't fail because an * encryption key wasn't loaded */ ASSERT((zio->io_flags & ZIO_FLAG_SPECULATIVE) || error != EACCES); /* * If we failed to decrypt, report an error now (as the zio * layer would have done if it had done the transforms). */ if (error == ECKSUM) { ASSERT(BP_IS_PROTECTED(bp)); error = SET_ERROR(EIO); if ((zio->io_flags & ZIO_FLAG_SPECULATIVE) == 0) { spa_log_error(zio->io_spa, &acb->acb_zb); (void) zfs_ereport_post( FM_EREPORT_ZFS_AUTHENTICATION, zio->io_spa, NULL, &acb->acb_zb, zio, 0); } } if (error != 0) { /* * Decompression or decryption failed. Set * io_error so that when we call acb_done * (below), we will indicate that the read * failed. Note that in the unusual case * where one callback is compressed and another * uncompressed, we will mark all of them * as failed, even though the uncompressed * one can't actually fail. In this case, * the hdr will not be anonymous, because * if there are multiple callbacks, it's * because multiple threads found the same * arc buf in the hash table. */ zio->io_error = error; } } /* * If there are multiple callbacks, we must have the hash lock, * because the only way for multiple threads to find this hdr is * in the hash table. This ensures that if there are multiple * callbacks, the hdr is not anonymous. If it were anonymous, * we couldn't use arc_buf_destroy() in the error case below. */ ASSERT(callback_cnt < 2 || hash_lock != NULL); hdr->b_l1hdr.b_acb = NULL; arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); if (callback_cnt == 0) ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr)); ASSERT(zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt) || callback_list != NULL); if (zio->io_error == 0) { arc_hdr_verify(hdr, zio->io_bp); } else { arc_hdr_set_flags(hdr, ARC_FLAG_IO_ERROR); if (hdr->b_l1hdr.b_state != arc_anon) arc_change_state(arc_anon, hdr, hash_lock); if (HDR_IN_HASH_TABLE(hdr)) buf_hash_remove(hdr); freeable = zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt); } /* * Broadcast before we drop the hash_lock to avoid the possibility * that the hdr (and hence the cv) might be freed before we get to * the cv_broadcast(). */ cv_broadcast(&hdr->b_l1hdr.b_cv); if (hash_lock != NULL) { mutex_exit(hash_lock); } else { /* * This block was freed while we waited for the read to * complete. It has been removed from the hash table and * moved to the anonymous state (so that it won't show up * in the cache). */ ASSERT3P(hdr->b_l1hdr.b_state, ==, arc_anon); freeable = zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt); } /* execute each callback and free its structure */ while ((acb = callback_list) != NULL) { if (acb->acb_done != NULL) { if (zio->io_error != 0 && acb->acb_buf != NULL) { /* * If arc_buf_alloc_impl() fails during * decompression, the buf will still be * allocated, and needs to be freed here. */ arc_buf_destroy(acb->acb_buf, acb->acb_private); acb->acb_buf = NULL; } acb->acb_done(zio, &zio->io_bookmark, zio->io_bp, acb->acb_buf, acb->acb_private); } if (acb->acb_zio_dummy != NULL) { acb->acb_zio_dummy->io_error = zio->io_error; zio_nowait(acb->acb_zio_dummy); } callback_list = acb->acb_next; kmem_free(acb, sizeof (arc_callback_t)); } if (freeable) arc_hdr_destroy(hdr); } /* * "Read" the block at the specified DVA (in bp) via the * cache. If the block is found in the cache, invoke the provided * callback immediately and return. Note that the `zio' parameter * in the callback will be NULL in this case, since no IO was * required. If the block is not in the cache pass the read request * on to the spa with a substitute callback function, so that the * requested block will be added to the cache. * * If a read request arrives for a block that has a read in-progress, * either wait for the in-progress read to complete (and return the * results); or, if this is a read with a "done" func, add a record * to the read to invoke the "done" func when the read completes, * and return; or just return. * * arc_read_done() will invoke all the requested "done" functions * for readers of this block. */ int arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_read_done_func_t *done, void *private, zio_priority_t priority, int zio_flags, arc_flags_t *arc_flags, const zbookmark_phys_t *zb) { arc_buf_hdr_t *hdr = NULL; kmutex_t *hash_lock = NULL; zio_t *rzio; uint64_t guid = spa_load_guid(spa); boolean_t compressed_read = (zio_flags & ZIO_FLAG_RAW_COMPRESS) != 0; boolean_t encrypted_read = BP_IS_ENCRYPTED(bp) && (zio_flags & ZIO_FLAG_RAW_ENCRYPT) != 0; boolean_t noauth_read = BP_IS_AUTHENTICATED(bp) && (zio_flags & ZIO_FLAG_RAW_ENCRYPT) != 0; boolean_t embedded_bp = !!BP_IS_EMBEDDED(bp); boolean_t no_buf = *arc_flags & ARC_FLAG_NO_BUF; int rc = 0; ASSERT(!embedded_bp || BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA); ASSERT(!BP_IS_HOLE(bp)); ASSERT(!BP_IS_REDACTED(bp)); /* * Normally SPL_FSTRANS will already be set since kernel threads which * expect to call the DMU interfaces will set it when created. System * calls are similarly handled by setting/cleaning the bit in the * registered callback (module/os/.../zfs/zpl_*). * * External consumers such as Lustre which call the exported DMU * interfaces may not have set SPL_FSTRANS. To avoid a deadlock * on the hash_lock always set and clear the bit. */ fstrans_cookie_t cookie = spl_fstrans_mark(); top: if (!embedded_bp) { /* * Embedded BP's have no DVA and require no I/O to "read". * Create an anonymous arc buf to back it. */ + if (!zfs_blkptr_verify(spa, bp, zio_flags & + ZIO_FLAG_CONFIG_WRITER, BLK_VERIFY_LOG)) { + rc = SET_ERROR(ECKSUM); + goto out; + } + hdr = buf_hash_find(guid, bp, &hash_lock); } /* * Determine if we have an L1 cache hit or a cache miss. For simplicity * we maintain encrypted data separately from compressed / uncompressed * data. If the user is requesting raw encrypted data and we don't have * that in the header we will read from disk to guarantee that we can * get it even if the encryption keys aren't loaded. */ if (hdr != NULL && HDR_HAS_L1HDR(hdr) && (HDR_HAS_RABD(hdr) || (hdr->b_l1hdr.b_pabd != NULL && !encrypted_read))) { arc_buf_t *buf = NULL; *arc_flags |= ARC_FLAG_CACHED; if (HDR_IO_IN_PROGRESS(hdr)) { zio_t *head_zio = hdr->b_l1hdr.b_acb->acb_zio_head; if (*arc_flags & ARC_FLAG_CACHED_ONLY) { mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_cached_only_in_progress); rc = SET_ERROR(ENOENT); goto out; } ASSERT3P(head_zio, !=, NULL); if ((hdr->b_flags & ARC_FLAG_PRIO_ASYNC_READ) && priority == ZIO_PRIORITY_SYNC_READ) { /* * This is a sync read that needs to wait for * an in-flight async read. Request that the * zio have its priority upgraded. */ zio_change_priority(head_zio, priority); DTRACE_PROBE1(arc__async__upgrade__sync, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_async_upgrade_sync); } if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { arc_hdr_clear_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); } if (*arc_flags & ARC_FLAG_WAIT) { cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); mutex_exit(hash_lock); goto top; } ASSERT(*arc_flags & ARC_FLAG_NOWAIT); if (done) { arc_callback_t *acb = NULL; acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); acb->acb_done = done; acb->acb_private = private; acb->acb_compressed = compressed_read; acb->acb_encrypted = encrypted_read; acb->acb_noauth = noauth_read; acb->acb_nobuf = no_buf; acb->acb_zb = *zb; if (pio != NULL) acb->acb_zio_dummy = zio_null(pio, spa, NULL, NULL, NULL, zio_flags); ASSERT3P(acb->acb_done, !=, NULL); acb->acb_zio_head = head_zio; acb->acb_next = hdr->b_l1hdr.b_acb; hdr->b_l1hdr.b_acb = acb; } mutex_exit(hash_lock); goto out; } ASSERT(hdr->b_l1hdr.b_state == arc_mru || hdr->b_l1hdr.b_state == arc_mfu); if (done && !no_buf) { if (hdr->b_flags & ARC_FLAG_PREDICTIVE_PREFETCH) { /* * This is a demand read which does not have to * wait for i/o because we did a predictive * prefetch i/o for it, which has completed. */ DTRACE_PROBE1( arc__demand__hit__predictive__prefetch, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP( arcstat_demand_hit_predictive_prefetch); arc_hdr_clear_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); } if (hdr->b_flags & ARC_FLAG_PRESCIENT_PREFETCH) { ARCSTAT_BUMP( arcstat_demand_hit_prescient_prefetch); arc_hdr_clear_flags(hdr, ARC_FLAG_PRESCIENT_PREFETCH); } ASSERT(!embedded_bp || !BP_IS_HOLE(bp)); /* Get a buf with the desired data in it. */ rc = arc_buf_alloc_impl(hdr, spa, zb, private, encrypted_read, compressed_read, noauth_read, B_TRUE, &buf); if (rc == ECKSUM) { /* * Convert authentication and decryption errors * to EIO (and generate an ereport if needed) * before leaving the ARC. */ rc = SET_ERROR(EIO); if ((zio_flags & ZIO_FLAG_SPECULATIVE) == 0) { spa_log_error(spa, zb); (void) zfs_ereport_post( FM_EREPORT_ZFS_AUTHENTICATION, spa, NULL, zb, NULL, 0); } } if (rc != 0) { (void) remove_reference(hdr, hash_lock, private); arc_buf_destroy_impl(buf); buf = NULL; } /* assert any errors weren't due to unloaded keys */ ASSERT((zio_flags & ZIO_FLAG_SPECULATIVE) || rc != EACCES); } else if (*arc_flags & ARC_FLAG_PREFETCH && zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_decrement_state(hdr); arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_increment_state(hdr); } DTRACE_PROBE1(arc__hit, arc_buf_hdr_t *, hdr); arc_access(hdr, hash_lock); if (*arc_flags & ARC_FLAG_PRESCIENT_PREFETCH) arc_hdr_set_flags(hdr, ARC_FLAG_PRESCIENT_PREFETCH); if (*arc_flags & ARC_FLAG_L2CACHE) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); mutex_exit(hash_lock); ARCSTAT_BUMP(arcstat_hits); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, hits); if (done) done(NULL, zb, bp, buf, private); } else { uint64_t lsize = BP_GET_LSIZE(bp); uint64_t psize = BP_GET_PSIZE(bp); arc_callback_t *acb; vdev_t *vd = NULL; uint64_t addr = 0; boolean_t devw = B_FALSE; uint64_t size; abd_t *hdr_abd; int alloc_flags = encrypted_read ? ARC_HDR_ALLOC_RDATA : 0; if (*arc_flags & ARC_FLAG_CACHED_ONLY) { rc = SET_ERROR(ENOENT); if (hash_lock != NULL) mutex_exit(hash_lock); goto out; } - /* - * Gracefully handle a damaged logical block size as a - * checksum error. - */ - if (lsize > spa_maxblocksize(spa)) { - rc = SET_ERROR(ECKSUM); - if (hash_lock != NULL) - mutex_exit(hash_lock); - goto out; - } - if (hdr == NULL) { /* * This block is not in the cache or it has * embedded data. */ arc_buf_hdr_t *exists = NULL; arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp); hdr = arc_hdr_alloc(spa_load_guid(spa), psize, lsize, BP_IS_PROTECTED(bp), BP_GET_COMPRESS(bp), 0, type, encrypted_read); if (!embedded_bp) { hdr->b_dva = *BP_IDENTITY(bp); hdr->b_birth = BP_PHYSICAL_BIRTH(bp); exists = buf_hash_insert(hdr, &hash_lock); } if (exists != NULL) { /* somebody beat us to the hash insert */ mutex_exit(hash_lock); buf_discard_identity(hdr); arc_hdr_destroy(hdr); goto top; /* restart the IO request */ } } else { /* * This block is in the ghost cache or encrypted data * was requested and we didn't have it. If it was * L2-only (and thus didn't have an L1 hdr), * we realloc the header to add an L1 hdr. */ if (!HDR_HAS_L1HDR(hdr)) { hdr = arc_hdr_realloc(hdr, hdr_l2only_cache, hdr_full_cache); } if (GHOST_STATE(hdr->b_l1hdr.b_state)) { ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT0(zfs_refcount_count( &hdr->b_l1hdr.b_refcnt)); ASSERT3P(hdr->b_l1hdr.b_buf, ==, NULL); ASSERT3P(hdr->b_l1hdr.b_freeze_cksum, ==, NULL); } else if (HDR_IO_IN_PROGRESS(hdr)) { /* * If this header already had an IO in progress * and we are performing another IO to fetch * encrypted data we must wait until the first * IO completes so as not to confuse * arc_read_done(). This should be very rare * and so the performance impact shouldn't * matter. */ cv_wait(&hdr->b_l1hdr.b_cv, hash_lock); mutex_exit(hash_lock); goto top; } /* * This is a delicate dance that we play here. * This hdr might be in the ghost list so we access * it to move it out of the ghost list before we * initiate the read. If it's a prefetch then * it won't have a callback so we'll remove the * reference that arc_buf_alloc_impl() created. We * do this after we've called arc_access() to * avoid hitting an assert in remove_reference(). */ arc_adapt(arc_hdr_size(hdr), hdr->b_l1hdr.b_state); arc_access(hdr, hash_lock); arc_hdr_alloc_abd(hdr, alloc_flags); } if (encrypted_read) { ASSERT(HDR_HAS_RABD(hdr)); size = HDR_GET_PSIZE(hdr); hdr_abd = hdr->b_crypt_hdr.b_rabd; zio_flags |= ZIO_FLAG_RAW; } else { ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); size = arc_hdr_size(hdr); hdr_abd = hdr->b_l1hdr.b_pabd; if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF) { zio_flags |= ZIO_FLAG_RAW_COMPRESS; } /* * For authenticated bp's, we do not ask the ZIO layer * to authenticate them since this will cause the entire * IO to fail if the key isn't loaded. Instead, we * defer authentication until arc_buf_fill(), which will * verify the data when the key is available. */ if (BP_IS_AUTHENTICATED(bp)) zio_flags |= ZIO_FLAG_RAW_ENCRYPT; } if (*arc_flags & ARC_FLAG_PREFETCH && zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_decrement_state(hdr); arc_hdr_set_flags(hdr, ARC_FLAG_PREFETCH); if (HDR_HAS_L2HDR(hdr)) l2arc_hdr_arcstats_increment_state(hdr); } if (*arc_flags & ARC_FLAG_PRESCIENT_PREFETCH) arc_hdr_set_flags(hdr, ARC_FLAG_PRESCIENT_PREFETCH); if (*arc_flags & ARC_FLAG_L2CACHE) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); if (BP_IS_AUTHENTICATED(bp)) arc_hdr_set_flags(hdr, ARC_FLAG_NOAUTH); if (BP_GET_LEVEL(bp) > 0) arc_hdr_set_flags(hdr, ARC_FLAG_INDIRECT); if (*arc_flags & ARC_FLAG_PREDICTIVE_PREFETCH) arc_hdr_set_flags(hdr, ARC_FLAG_PREDICTIVE_PREFETCH); ASSERT(!GHOST_STATE(hdr->b_l1hdr.b_state)); acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP); acb->acb_done = done; acb->acb_private = private; acb->acb_compressed = compressed_read; acb->acb_encrypted = encrypted_read; acb->acb_noauth = noauth_read; acb->acb_zb = *zb; ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); hdr->b_l1hdr.b_acb = acb; arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); if (HDR_HAS_L2HDR(hdr) && (vd = hdr->b_l2hdr.b_dev->l2ad_vdev) != NULL) { devw = hdr->b_l2hdr.b_dev->l2ad_writing; addr = hdr->b_l2hdr.b_daddr; /* * Lock out L2ARC device removal. */ if (vdev_is_dead(vd) || !spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) vd = NULL; } /* * We count both async reads and scrub IOs as asynchronous so * that both can be upgraded in the event of a cache hit while * the read IO is still in-flight. */ if (priority == ZIO_PRIORITY_ASYNC_READ || priority == ZIO_PRIORITY_SCRUB) arc_hdr_set_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); else arc_hdr_clear_flags(hdr, ARC_FLAG_PRIO_ASYNC_READ); /* * At this point, we have a level 1 cache miss or a blkptr * with embedded data. Try again in L2ARC if possible. */ ASSERT3U(HDR_GET_LSIZE(hdr), ==, lsize); /* * Skip ARC stat bump for block pointers with embedded * data. The data are read from the blkptr itself via * decode_embedded_bp_compressed(). */ if (!embedded_bp) { DTRACE_PROBE4(arc__miss, arc_buf_hdr_t *, hdr, blkptr_t *, bp, uint64_t, lsize, zbookmark_phys_t *, zb); ARCSTAT_BUMP(arcstat_misses); ARCSTAT_CONDSTAT(!HDR_PREFETCH(hdr), demand, prefetch, !HDR_ISTYPE_METADATA(hdr), data, metadata, misses); zfs_racct_read(size, 1); } /* Check if the spa even has l2 configured */ const boolean_t spa_has_l2 = l2arc_ndev != 0 && spa->spa_l2cache.sav_count > 0; if (vd != NULL && spa_has_l2 && !(l2arc_norw && devw)) { /* * Read from the L2ARC if the following are true: * 1. The L2ARC vdev was previously cached. * 2. This buffer still has L2ARC metadata. * 3. This buffer isn't currently writing to the L2ARC. * 4. The L2ARC entry wasn't evicted, which may * also have invalidated the vdev. * 5. This isn't prefetch or l2arc_noprefetch is 0. */ if (HDR_HAS_L2HDR(hdr) && !HDR_L2_WRITING(hdr) && !HDR_L2_EVICTED(hdr) && !(l2arc_noprefetch && HDR_PREFETCH(hdr))) { l2arc_read_callback_t *cb; abd_t *abd; uint64_t asize; DTRACE_PROBE1(l2arc__hit, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_hits); atomic_inc_32(&hdr->b_l2hdr.b_hits); cb = kmem_zalloc(sizeof (l2arc_read_callback_t), KM_SLEEP); cb->l2rcb_hdr = hdr; cb->l2rcb_bp = *bp; cb->l2rcb_zb = *zb; cb->l2rcb_flags = zio_flags; /* * When Compressed ARC is disabled, but the * L2ARC block is compressed, arc_hdr_size() * will have returned LSIZE rather than PSIZE. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr) && HDR_GET_PSIZE(hdr) != 0) { size = HDR_GET_PSIZE(hdr); } asize = vdev_psize_to_asize(vd, size); if (asize != size) { abd = abd_alloc_for_io(asize, HDR_ISTYPE_METADATA(hdr)); cb->l2rcb_abd = abd; } else { abd = hdr_abd; } ASSERT(addr >= VDEV_LABEL_START_SIZE && addr + asize <= vd->vdev_psize - VDEV_LABEL_END_SIZE); /* * l2arc read. The SCL_L2ARC lock will be * released by l2arc_read_done(). * Issue a null zio if the underlying buffer * was squashed to zero size by compression. */ ASSERT3U(arc_hdr_get_compress(hdr), !=, ZIO_COMPRESS_EMPTY); rzio = zio_read_phys(pio, vd, addr, asize, abd, ZIO_CHECKSUM_OFF, l2arc_read_done, cb, priority, zio_flags | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY, B_FALSE); acb->acb_zio_head = rzio; if (hash_lock != NULL) mutex_exit(hash_lock); DTRACE_PROBE2(l2arc__read, vdev_t *, vd, zio_t *, rzio); ARCSTAT_INCR(arcstat_l2_read_bytes, HDR_GET_PSIZE(hdr)); if (*arc_flags & ARC_FLAG_NOWAIT) { zio_nowait(rzio); goto out; } ASSERT(*arc_flags & ARC_FLAG_WAIT); if (zio_wait(rzio) == 0) goto out; /* l2arc read error; goto zio_read() */ if (hash_lock != NULL) mutex_enter(hash_lock); } else { DTRACE_PROBE1(l2arc__miss, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_misses); if (HDR_L2_WRITING(hdr)) ARCSTAT_BUMP(arcstat_l2_rw_clash); spa_config_exit(spa, SCL_L2ARC, vd); } } else { if (vd != NULL) spa_config_exit(spa, SCL_L2ARC, vd); /* * Only a spa with l2 should contribute to l2 * miss stats. (Including the case of having a * faulted cache device - that's also a miss.) */ if (spa_has_l2) { /* * Skip ARC stat bump for block pointers with * embedded data. The data are read from the * blkptr itself via * decode_embedded_bp_compressed(). */ if (!embedded_bp) { DTRACE_PROBE1(l2arc__miss, arc_buf_hdr_t *, hdr); ARCSTAT_BUMP(arcstat_l2_misses); } } } rzio = zio_read(pio, spa, bp, hdr_abd, size, arc_read_done, hdr, priority, zio_flags, zb); acb->acb_zio_head = rzio; if (hash_lock != NULL) mutex_exit(hash_lock); if (*arc_flags & ARC_FLAG_WAIT) { rc = zio_wait(rzio); goto out; } ASSERT(*arc_flags & ARC_FLAG_NOWAIT); zio_nowait(rzio); } out: /* embedded bps don't actually go to disk */ if (!embedded_bp) spa_read_history_add(spa, zb, *arc_flags); spl_fstrans_unmark(cookie); return (rc); } arc_prune_t * arc_add_prune_callback(arc_prune_func_t *func, void *private) { arc_prune_t *p; p = kmem_alloc(sizeof (*p), KM_SLEEP); p->p_pfunc = func; p->p_private = private; list_link_init(&p->p_node); zfs_refcount_create(&p->p_refcnt); mutex_enter(&arc_prune_mtx); zfs_refcount_add(&p->p_refcnt, &arc_prune_list); list_insert_head(&arc_prune_list, p); mutex_exit(&arc_prune_mtx); return (p); } void arc_remove_prune_callback(arc_prune_t *p) { boolean_t wait = B_FALSE; mutex_enter(&arc_prune_mtx); list_remove(&arc_prune_list, p); if (zfs_refcount_remove(&p->p_refcnt, &arc_prune_list) > 0) wait = B_TRUE; mutex_exit(&arc_prune_mtx); /* wait for arc_prune_task to finish */ if (wait) taskq_wait_outstanding(arc_prune_taskq, 0); ASSERT0(zfs_refcount_count(&p->p_refcnt)); zfs_refcount_destroy(&p->p_refcnt); kmem_free(p, sizeof (*p)); } /* * Notify the arc that a block was freed, and thus will never be used again. */ void arc_freed(spa_t *spa, const blkptr_t *bp) { arc_buf_hdr_t *hdr; kmutex_t *hash_lock; uint64_t guid = spa_load_guid(spa); ASSERT(!BP_IS_EMBEDDED(bp)); hdr = buf_hash_find(guid, bp, &hash_lock); if (hdr == NULL) return; /* * We might be trying to free a block that is still doing I/O * (i.e. prefetch) or has a reference (i.e. a dedup-ed, * dmu_sync-ed block). If this block is being prefetched, then it * would still have the ARC_FLAG_IO_IN_PROGRESS flag set on the hdr * until the I/O completes. A block may also have a reference if it is * part of a dedup-ed, dmu_synced write. The dmu_sync() function would * have written the new block to its final resting place on disk but * without the dedup flag set. This would have left the hdr in the MRU * state and discoverable. When the txg finally syncs it detects that * the block was overridden in open context and issues an override I/O. * Since this is a dedup block, the override I/O will determine if the * block is already in the DDT. If so, then it will replace the io_bp * with the bp from the DDT and allow the I/O to finish. When the I/O * reaches the done callback, dbuf_write_override_done, it will * check to see if the io_bp and io_bp_override are identical. * If they are not, then it indicates that the bp was replaced with * the bp in the DDT and the override bp is freed. This allows * us to arrive here with a reference on a block that is being * freed. So if we have an I/O in progress, or a reference to * this hdr, then we don't destroy the hdr. */ if (!HDR_HAS_L1HDR(hdr) || (!HDR_IO_IN_PROGRESS(hdr) && zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt))) { arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); mutex_exit(hash_lock); } else { mutex_exit(hash_lock); } } /* * Release this buffer from the cache, making it an anonymous buffer. This * must be done after a read and prior to modifying the buffer contents. * If the buffer has more than one reference, we must make * a new hdr for the buffer. */ void arc_release(arc_buf_t *buf, void *tag) { arc_buf_hdr_t *hdr = buf->b_hdr; /* * It would be nice to assert that if its DMU metadata (level > * 0 || it's the dnode file), then it must be syncing context. * But we don't know that information at this level. */ mutex_enter(&buf->b_evict_lock); ASSERT(HDR_HAS_L1HDR(hdr)); /* * We don't grab the hash lock prior to this check, because if * the buffer's header is in the arc_anon state, it won't be * linked into the hash table. */ if (hdr->b_l1hdr.b_state == arc_anon) { mutex_exit(&buf->b_evict_lock); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT(!HDR_IN_HASH_TABLE(hdr)); ASSERT(!HDR_HAS_L2HDR(hdr)); ASSERT(HDR_EMPTY(hdr)); ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); ASSERT3S(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt), ==, 1); ASSERT(!list_link_active(&hdr->b_l1hdr.b_arc_node)); hdr->b_l1hdr.b_arc_access = 0; /* * If the buf is being overridden then it may already * have a hdr that is not empty. */ buf_discard_identity(hdr); arc_buf_thaw(buf); return; } kmutex_t *hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); /* * This assignment is only valid as long as the hash_lock is * held, we must be careful not to reference state or the * b_state field after dropping the lock. */ arc_state_t *state = hdr->b_l1hdr.b_state; ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); ASSERT3P(state, !=, arc_anon); /* this buffer is not on any list */ ASSERT3S(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt), >, 0); if (HDR_HAS_L2HDR(hdr)) { mutex_enter(&hdr->b_l2hdr.b_dev->l2ad_mtx); /* * We have to recheck this conditional again now that * we're holding the l2ad_mtx to prevent a race with * another thread which might be concurrently calling * l2arc_evict(). In that case, l2arc_evict() might have * destroyed the header's L2 portion as we were waiting * to acquire the l2ad_mtx. */ if (HDR_HAS_L2HDR(hdr)) arc_hdr_l2hdr_destroy(hdr); mutex_exit(&hdr->b_l2hdr.b_dev->l2ad_mtx); } /* * Do we have more than one buf? */ if (hdr->b_l1hdr.b_bufcnt > 1) { arc_buf_hdr_t *nhdr; uint64_t spa = hdr->b_spa; uint64_t psize = HDR_GET_PSIZE(hdr); uint64_t lsize = HDR_GET_LSIZE(hdr); boolean_t protected = HDR_PROTECTED(hdr); enum zio_compress compress = arc_hdr_get_compress(hdr); arc_buf_contents_t type = arc_buf_type(hdr); VERIFY3U(hdr->b_type, ==, type); ASSERT(hdr->b_l1hdr.b_buf != buf || buf->b_next != NULL); (void) remove_reference(hdr, hash_lock, tag); if (arc_buf_is_shared(buf) && !ARC_BUF_COMPRESSED(buf)) { ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); ASSERT(ARC_BUF_LAST(buf)); } /* * Pull the data off of this hdr and attach it to * a new anonymous hdr. Also find the last buffer * in the hdr's buffer list. */ arc_buf_t *lastbuf = arc_buf_remove(hdr, buf); ASSERT3P(lastbuf, !=, NULL); /* * If the current arc_buf_t and the hdr are sharing their data * buffer, then we must stop sharing that block. */ if (arc_buf_is_shared(buf)) { ASSERT3P(hdr->b_l1hdr.b_buf, !=, buf); VERIFY(!arc_buf_is_shared(lastbuf)); /* * First, sever the block sharing relationship between * buf and the arc_buf_hdr_t. */ arc_unshare_buf(hdr, buf); /* * Now we need to recreate the hdr's b_pabd. Since we * have lastbuf handy, we try to share with it, but if * we can't then we allocate a new b_pabd and copy the * data from buf into it. */ if (arc_can_share(hdr, lastbuf)) { arc_share_buf(hdr, lastbuf); } else { arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT); abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data, psize); } VERIFY3P(lastbuf->b_data, !=, NULL); } else if (HDR_SHARED_DATA(hdr)) { /* * Uncompressed shared buffers are always at the end * of the list. Compressed buffers don't have the * same requirements. This makes it hard to * simply assert that the lastbuf is shared so * we rely on the hdr's compression flags to determine * if we have a compressed, shared buffer. */ ASSERT(arc_buf_is_shared(lastbuf) || arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF); ASSERT(!ARC_BUF_SHARED(buf)); } ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr)); ASSERT3P(state, !=, arc_l2c_only); (void) zfs_refcount_remove_many(&state->arcs_size, arc_buf_size(buf), buf); if (zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)) { ASSERT3P(state, !=, arc_l2c_only); (void) zfs_refcount_remove_many( &state->arcs_esize[type], arc_buf_size(buf), buf); } hdr->b_l1hdr.b_bufcnt -= 1; if (ARC_BUF_ENCRYPTED(buf)) hdr->b_crypt_hdr.b_ebufcnt -= 1; arc_cksum_verify(buf); arc_buf_unwatch(buf); /* if this is the last uncompressed buf free the checksum */ if (!arc_hdr_has_uncompressed_buf(hdr)) arc_cksum_free(hdr); mutex_exit(hash_lock); /* * Allocate a new hdr. The new hdr will contain a b_pabd * buffer which will be freed in arc_write(). */ nhdr = arc_hdr_alloc(spa, psize, lsize, protected, compress, hdr->b_complevel, type, HDR_HAS_RABD(hdr)); ASSERT3P(nhdr->b_l1hdr.b_buf, ==, NULL); ASSERT0(nhdr->b_l1hdr.b_bufcnt); ASSERT0(zfs_refcount_count(&nhdr->b_l1hdr.b_refcnt)); VERIFY3U(nhdr->b_type, ==, type); ASSERT(!HDR_SHARED_DATA(nhdr)); nhdr->b_l1hdr.b_buf = buf; nhdr->b_l1hdr.b_bufcnt = 1; if (ARC_BUF_ENCRYPTED(buf)) nhdr->b_crypt_hdr.b_ebufcnt = 1; nhdr->b_l1hdr.b_mru_hits = 0; nhdr->b_l1hdr.b_mru_ghost_hits = 0; nhdr->b_l1hdr.b_mfu_hits = 0; nhdr->b_l1hdr.b_mfu_ghost_hits = 0; nhdr->b_l1hdr.b_l2_hits = 0; (void) zfs_refcount_add(&nhdr->b_l1hdr.b_refcnt, tag); buf->b_hdr = nhdr; mutex_exit(&buf->b_evict_lock); (void) zfs_refcount_add_many(&arc_anon->arcs_size, arc_buf_size(buf), buf); } else { mutex_exit(&buf->b_evict_lock); ASSERT(zfs_refcount_count(&hdr->b_l1hdr.b_refcnt) == 1); /* protected by hash lock, or hdr is on arc_anon */ ASSERT(!multilist_link_active(&hdr->b_l1hdr.b_arc_node)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); hdr->b_l1hdr.b_mru_hits = 0; hdr->b_l1hdr.b_mru_ghost_hits = 0; hdr->b_l1hdr.b_mfu_hits = 0; hdr->b_l1hdr.b_mfu_ghost_hits = 0; hdr->b_l1hdr.b_l2_hits = 0; arc_change_state(arc_anon, hdr, hash_lock); hdr->b_l1hdr.b_arc_access = 0; mutex_exit(hash_lock); buf_discard_identity(hdr); arc_buf_thaw(buf); } } int arc_released(arc_buf_t *buf) { int released; mutex_enter(&buf->b_evict_lock); released = (buf->b_data != NULL && buf->b_hdr->b_l1hdr.b_state == arc_anon); mutex_exit(&buf->b_evict_lock); return (released); } #ifdef ZFS_DEBUG int arc_referenced(arc_buf_t *buf) { int referenced; mutex_enter(&buf->b_evict_lock); referenced = (zfs_refcount_count(&buf->b_hdr->b_l1hdr.b_refcnt)); mutex_exit(&buf->b_evict_lock); return (referenced); } #endif static void arc_write_ready(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; arc_buf_hdr_t *hdr = buf->b_hdr; blkptr_t *bp = zio->io_bp; uint64_t psize = BP_IS_HOLE(bp) ? 0 : BP_GET_PSIZE(bp); fstrans_cookie_t cookie = spl_fstrans_mark(); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT(!zfs_refcount_is_zero(&buf->b_hdr->b_l1hdr.b_refcnt)); ASSERT(hdr->b_l1hdr.b_bufcnt > 0); /* * If we're reexecuting this zio because the pool suspended, then * cleanup any state that was previously set the first time the * callback was invoked. */ if (zio->io_flags & ZIO_FLAG_REEXECUTED) { arc_cksum_free(hdr); arc_buf_unwatch(buf); if (hdr->b_l1hdr.b_pabd != NULL) { if (arc_buf_is_shared(buf)) { arc_unshare_buf(hdr, buf); } else { arc_hdr_free_abd(hdr, B_FALSE); } } if (HDR_HAS_RABD(hdr)) arc_hdr_free_abd(hdr, B_TRUE); } ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); ASSERT(!HDR_HAS_RABD(hdr)); ASSERT(!HDR_SHARED_DATA(hdr)); ASSERT(!arc_buf_is_shared(buf)); callback->awcb_ready(zio, buf, callback->awcb_private); if (HDR_IO_IN_PROGRESS(hdr)) ASSERT(zio->io_flags & ZIO_FLAG_REEXECUTED); arc_hdr_set_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); if (BP_IS_PROTECTED(bp) != !!HDR_PROTECTED(hdr)) hdr = arc_hdr_realloc_crypt(hdr, BP_IS_PROTECTED(bp)); if (BP_IS_PROTECTED(bp)) { /* ZIL blocks are written through zio_rewrite */ ASSERT3U(BP_GET_TYPE(bp), !=, DMU_OT_INTENT_LOG); ASSERT(HDR_PROTECTED(hdr)); if (BP_SHOULD_BYTESWAP(bp)) { if (BP_GET_LEVEL(bp) > 0) { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_UINT64; } else { hdr->b_l1hdr.b_byteswap = DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); } } else { hdr->b_l1hdr.b_byteswap = DMU_BSWAP_NUMFUNCS; } hdr->b_crypt_hdr.b_ot = BP_GET_TYPE(bp); hdr->b_crypt_hdr.b_dsobj = zio->io_bookmark.zb_objset; zio_crypt_decode_params_bp(bp, hdr->b_crypt_hdr.b_salt, hdr->b_crypt_hdr.b_iv); zio_crypt_decode_mac_bp(bp, hdr->b_crypt_hdr.b_mac); } /* * If this block was written for raw encryption but the zio layer * ended up only authenticating it, adjust the buffer flags now. */ if (BP_IS_AUTHENTICATED(bp) && ARC_BUF_ENCRYPTED(buf)) { arc_hdr_set_flags(hdr, ARC_FLAG_NOAUTH); buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED; if (BP_GET_COMPRESS(bp) == ZIO_COMPRESS_OFF) buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; } else if (BP_IS_HOLE(bp) && ARC_BUF_ENCRYPTED(buf)) { buf->b_flags &= ~ARC_BUF_FLAG_ENCRYPTED; buf->b_flags &= ~ARC_BUF_FLAG_COMPRESSED; } /* this must be done after the buffer flags are adjusted */ arc_cksum_compute(buf); enum zio_compress compress; if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) { compress = ZIO_COMPRESS_OFF; } else { ASSERT3U(HDR_GET_LSIZE(hdr), ==, BP_GET_LSIZE(bp)); compress = BP_GET_COMPRESS(bp); } HDR_SET_PSIZE(hdr, psize); arc_hdr_set_compress(hdr, compress); hdr->b_complevel = zio->io_prop.zp_complevel; if (zio->io_error != 0 || psize == 0) goto out; /* * Fill the hdr with data. If the buffer is encrypted we have no choice * but to copy the data into b_radb. If the hdr is compressed, the data * we want is available from the zio, otherwise we can take it from * the buf. * * We might be able to share the buf's data with the hdr here. However, * doing so would cause the ARC to be full of linear ABDs if we write a * lot of shareable data. As a compromise, we check whether scattered * ABDs are allowed, and assume that if they are then the user wants * the ARC to be primarily filled with them regardless of the data being * written. Therefore, if they're allowed then we allocate one and copy * the data into it; otherwise, we share the data directly if we can. */ if (ARC_BUF_ENCRYPTED(buf)) { ASSERT3U(psize, >, 0); ASSERT(ARC_BUF_COMPRESSED(buf)); arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT|ARC_HDR_ALLOC_RDATA); abd_copy(hdr->b_crypt_hdr.b_rabd, zio->io_abd, psize); } else if (zfs_abd_scatter_enabled || !arc_can_share(hdr, buf)) { /* * Ideally, we would always copy the io_abd into b_pabd, but the * user may have disabled compressed ARC, thus we must check the * hdr's compression setting rather than the io_bp's. */ if (BP_IS_ENCRYPTED(bp)) { ASSERT3U(psize, >, 0); arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT|ARC_HDR_ALLOC_RDATA); abd_copy(hdr->b_crypt_hdr.b_rabd, zio->io_abd, psize); } else if (arc_hdr_get_compress(hdr) != ZIO_COMPRESS_OFF && !ARC_BUF_COMPRESSED(buf)) { ASSERT3U(psize, >, 0); arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT); abd_copy(hdr->b_l1hdr.b_pabd, zio->io_abd, psize); } else { ASSERT3U(zio->io_orig_size, ==, arc_hdr_size(hdr)); arc_hdr_alloc_abd(hdr, ARC_HDR_DO_ADAPT); abd_copy_from_buf(hdr->b_l1hdr.b_pabd, buf->b_data, arc_buf_size(buf)); } } else { ASSERT3P(buf->b_data, ==, abd_to_buf(zio->io_orig_abd)); ASSERT3U(zio->io_orig_size, ==, arc_buf_size(buf)); ASSERT3U(hdr->b_l1hdr.b_bufcnt, ==, 1); arc_share_buf(hdr, buf); } out: arc_hdr_verify(hdr, bp); spl_fstrans_unmark(cookie); } static void arc_write_children_ready(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; callback->awcb_children_ready(zio, buf, callback->awcb_private); } /* * The SPA calls this callback for each physical write that happens on behalf * of a logical write. See the comment in dbuf_write_physdone() for details. */ static void arc_write_physdone(zio_t *zio) { arc_write_callback_t *cb = zio->io_private; if (cb->awcb_physdone != NULL) cb->awcb_physdone(zio, cb->awcb_buf, cb->awcb_private); } static void arc_write_done(zio_t *zio) { arc_write_callback_t *callback = zio->io_private; arc_buf_t *buf = callback->awcb_buf; arc_buf_hdr_t *hdr = buf->b_hdr; ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); if (zio->io_error == 0) { arc_hdr_verify(hdr, zio->io_bp); if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) { buf_discard_identity(hdr); } else { hdr->b_dva = *BP_IDENTITY(zio->io_bp); hdr->b_birth = BP_PHYSICAL_BIRTH(zio->io_bp); } } else { ASSERT(HDR_EMPTY(hdr)); } /* * If the block to be written was all-zero or compressed enough to be * embedded in the BP, no write was performed so there will be no * dva/birth/checksum. The buffer must therefore remain anonymous * (and uncached). */ if (!HDR_EMPTY(hdr)) { arc_buf_hdr_t *exists; kmutex_t *hash_lock; ASSERT3U(zio->io_error, ==, 0); arc_cksum_verify(buf); exists = buf_hash_insert(hdr, &hash_lock); if (exists != NULL) { /* * This can only happen if we overwrite for * sync-to-convergence, because we remove * buffers from the hash table when we arc_free(). */ if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) panic("bad overwrite, hdr=%p exists=%p", (void *)hdr, (void *)exists); ASSERT(zfs_refcount_is_zero( &exists->b_l1hdr.b_refcnt)); arc_change_state(arc_anon, exists, hash_lock); arc_hdr_destroy(exists); mutex_exit(hash_lock); exists = buf_hash_insert(hdr, &hash_lock); ASSERT3P(exists, ==, NULL); } else if (zio->io_flags & ZIO_FLAG_NOPWRITE) { /* nopwrite */ ASSERT(zio->io_prop.zp_nopwrite); if (!BP_EQUAL(&zio->io_bp_orig, zio->io_bp)) panic("bad nopwrite, hdr=%p exists=%p", (void *)hdr, (void *)exists); } else { /* Dedup */ ASSERT(hdr->b_l1hdr.b_bufcnt == 1); ASSERT(hdr->b_l1hdr.b_state == arc_anon); ASSERT(BP_GET_DEDUP(zio->io_bp)); ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); } } arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); /* if it's not anon, we are doing a scrub */ if (exists == NULL && hdr->b_l1hdr.b_state == arc_anon) arc_access(hdr, hash_lock); mutex_exit(hash_lock); } else { arc_hdr_clear_flags(hdr, ARC_FLAG_IO_IN_PROGRESS); } ASSERT(!zfs_refcount_is_zero(&hdr->b_l1hdr.b_refcnt)); callback->awcb_done(zio, buf, callback->awcb_private); abd_free(zio->io_abd); kmem_free(callback, sizeof (arc_write_callback_t)); } zio_t * arc_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, arc_buf_t *buf, boolean_t l2arc, const zio_prop_t *zp, arc_write_done_func_t *ready, arc_write_done_func_t *children_ready, arc_write_done_func_t *physdone, arc_write_done_func_t *done, void *private, zio_priority_t priority, int zio_flags, const zbookmark_phys_t *zb) { arc_buf_hdr_t *hdr = buf->b_hdr; arc_write_callback_t *callback; zio_t *zio; zio_prop_t localprop = *zp; ASSERT3P(ready, !=, NULL); ASSERT3P(done, !=, NULL); ASSERT(!HDR_IO_ERROR(hdr)); ASSERT(!HDR_IO_IN_PROGRESS(hdr)); ASSERT3P(hdr->b_l1hdr.b_acb, ==, NULL); ASSERT3U(hdr->b_l1hdr.b_bufcnt, >, 0); if (l2arc) arc_hdr_set_flags(hdr, ARC_FLAG_L2CACHE); if (ARC_BUF_ENCRYPTED(buf)) { ASSERT(ARC_BUF_COMPRESSED(buf)); localprop.zp_encrypt = B_TRUE; localprop.zp_compress = HDR_GET_COMPRESS(hdr); localprop.zp_complevel = hdr->b_complevel; localprop.zp_byteorder = (hdr->b_l1hdr.b_byteswap == DMU_BSWAP_NUMFUNCS) ? ZFS_HOST_BYTEORDER : !ZFS_HOST_BYTEORDER; bcopy(hdr->b_crypt_hdr.b_salt, localprop.zp_salt, ZIO_DATA_SALT_LEN); bcopy(hdr->b_crypt_hdr.b_iv, localprop.zp_iv, ZIO_DATA_IV_LEN); bcopy(hdr->b_crypt_hdr.b_mac, localprop.zp_mac, ZIO_DATA_MAC_LEN); if (DMU_OT_IS_ENCRYPTED(localprop.zp_type)) { localprop.zp_nopwrite = B_FALSE; localprop.zp_copies = MIN(localprop.zp_copies, SPA_DVAS_PER_BP - 1); } zio_flags |= ZIO_FLAG_RAW; } else if (ARC_BUF_COMPRESSED(buf)) { ASSERT3U(HDR_GET_LSIZE(hdr), !=, arc_buf_size(buf)); localprop.zp_compress = HDR_GET_COMPRESS(hdr); localprop.zp_complevel = hdr->b_complevel; zio_flags |= ZIO_FLAG_RAW_COMPRESS; } callback = kmem_zalloc(sizeof (arc_write_callback_t), KM_SLEEP); callback->awcb_ready = ready; callback->awcb_children_ready = children_ready; callback->awcb_physdone = physdone; callback->awcb_done = done; callback->awcb_private = private; callback->awcb_buf = buf; /* * The hdr's b_pabd is now stale, free it now. A new data block * will be allocated when the zio pipeline calls arc_write_ready(). */ if (hdr->b_l1hdr.b_pabd != NULL) { /* * If the buf is currently sharing the data block with * the hdr then we need to break that relationship here. * The hdr will remain with a NULL data pointer and the * buf will take sole ownership of the block. */ if (arc_buf_is_shared(buf)) { arc_unshare_buf(hdr, buf); } else { arc_hdr_free_abd(hdr, B_FALSE); } VERIFY3P(buf->b_data, !=, NULL); } if (HDR_HAS_RABD(hdr)) arc_hdr_free_abd(hdr, B_TRUE); if (!(zio_flags & ZIO_FLAG_RAW)) arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF); ASSERT(!arc_buf_is_shared(buf)); ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL); zio = zio_write(pio, spa, txg, bp, abd_get_from_buf(buf->b_data, HDR_GET_LSIZE(hdr)), HDR_GET_LSIZE(hdr), arc_buf_size(buf), &localprop, arc_write_ready, (children_ready != NULL) ? arc_write_children_ready : NULL, arc_write_physdone, arc_write_done, callback, priority, zio_flags, zb); return (zio); } void arc_tempreserve_clear(uint64_t reserve) { atomic_add_64(&arc_tempreserve, -reserve); ASSERT((int64_t)arc_tempreserve >= 0); } int arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg) { int error; uint64_t anon_size; if (!arc_no_grow && reserve > arc_c/4 && reserve * 4 > (2ULL << SPA_MAXBLOCKSHIFT)) arc_c = MIN(arc_c_max, reserve * 4); /* * Throttle when the calculated memory footprint for the TXG * exceeds the target ARC size. */ if (reserve > arc_c) { DMU_TX_STAT_BUMP(dmu_tx_memory_reserve); return (SET_ERROR(ERESTART)); } /* * Don't count loaned bufs as in flight dirty data to prevent long * network delays from blocking transactions that are ready to be * assigned to a txg. */ /* assert that it has not wrapped around */ ASSERT3S(atomic_add_64_nv(&arc_loaned_bytes, 0), >=, 0); anon_size = MAX((int64_t)(zfs_refcount_count(&arc_anon->arcs_size) - arc_loaned_bytes), 0); /* * Writes will, almost always, require additional memory allocations * in order to compress/encrypt/etc the data. We therefore need to * make sure that there is sufficient available memory for this. */ error = arc_memory_throttle(spa, reserve, txg); if (error != 0) return (error); /* * Throttle writes when the amount of dirty data in the cache * gets too large. We try to keep the cache less than half full * of dirty blocks so that our sync times don't grow too large. * * In the case of one pool being built on another pool, we want * to make sure we don't end up throttling the lower (backing) * pool when the upper pool is the majority contributor to dirty * data. To insure we make forward progress during throttling, we * also check the current pool's net dirty data and only throttle * if it exceeds zfs_arc_pool_dirty_percent of the anonymous dirty * data in the cache. * * Note: if two requests come in concurrently, we might let them * both succeed, when one of them should fail. Not a huge deal. */ uint64_t total_dirty = reserve + arc_tempreserve + anon_size; uint64_t spa_dirty_anon = spa_dirty_data(spa); uint64_t rarc_c = arc_warm ? arc_c : arc_c_max; if (total_dirty > rarc_c * zfs_arc_dirty_limit_percent / 100 && anon_size > rarc_c * zfs_arc_anon_limit_percent / 100 && spa_dirty_anon > anon_size * zfs_arc_pool_dirty_percent / 100) { #ifdef ZFS_DEBUG uint64_t meta_esize = zfs_refcount_count( &arc_anon->arcs_esize[ARC_BUFC_METADATA]); uint64_t data_esize = zfs_refcount_count(&arc_anon->arcs_esize[ARC_BUFC_DATA]); dprintf("failing, arc_tempreserve=%lluK anon_meta=%lluK " "anon_data=%lluK tempreserve=%lluK rarc_c=%lluK\n", arc_tempreserve >> 10, meta_esize >> 10, data_esize >> 10, reserve >> 10, rarc_c >> 10); #endif DMU_TX_STAT_BUMP(dmu_tx_dirty_throttle); return (SET_ERROR(ERESTART)); } atomic_add_64(&arc_tempreserve, reserve); return (0); } static void arc_kstat_update_state(arc_state_t *state, kstat_named_t *size, kstat_named_t *evict_data, kstat_named_t *evict_metadata) { size->value.ui64 = zfs_refcount_count(&state->arcs_size); evict_data->value.ui64 = zfs_refcount_count(&state->arcs_esize[ARC_BUFC_DATA]); evict_metadata->value.ui64 = zfs_refcount_count(&state->arcs_esize[ARC_BUFC_METADATA]); } static int arc_kstat_update(kstat_t *ksp, int rw) { arc_stats_t *as = ksp->ks_data; - if (rw == KSTAT_WRITE) { + if (rw == KSTAT_WRITE) return (SET_ERROR(EACCES)); - } else { - arc_kstat_update_state(arc_anon, - &as->arcstat_anon_size, - &as->arcstat_anon_evictable_data, - &as->arcstat_anon_evictable_metadata); - arc_kstat_update_state(arc_mru, - &as->arcstat_mru_size, - &as->arcstat_mru_evictable_data, - &as->arcstat_mru_evictable_metadata); - arc_kstat_update_state(arc_mru_ghost, - &as->arcstat_mru_ghost_size, - &as->arcstat_mru_ghost_evictable_data, - &as->arcstat_mru_ghost_evictable_metadata); - arc_kstat_update_state(arc_mfu, - &as->arcstat_mfu_size, - &as->arcstat_mfu_evictable_data, - &as->arcstat_mfu_evictable_metadata); - arc_kstat_update_state(arc_mfu_ghost, - &as->arcstat_mfu_ghost_size, - &as->arcstat_mfu_ghost_evictable_data, - &as->arcstat_mfu_ghost_evictable_metadata); - - ARCSTAT(arcstat_size) = aggsum_value(&arc_size); - ARCSTAT(arcstat_meta_used) = aggsum_value(&arc_meta_used); - ARCSTAT(arcstat_data_size) = wmsum_value(&astat_data_size); - ARCSTAT(arcstat_metadata_size) = - wmsum_value(&astat_metadata_size); - ARCSTAT(arcstat_hdr_size) = wmsum_value(&astat_hdr_size); - ARCSTAT(arcstat_l2_hdr_size) = aggsum_value(&astat_l2_hdr_size); - ARCSTAT(arcstat_dbuf_size) = wmsum_value(&astat_dbuf_size); + + as->arcstat_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_hits); + as->arcstat_misses.value.ui64 = + wmsum_value(&arc_sums.arcstat_misses); + as->arcstat_demand_data_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_demand_data_hits); + as->arcstat_demand_data_misses.value.ui64 = + wmsum_value(&arc_sums.arcstat_demand_data_misses); + as->arcstat_demand_metadata_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_demand_metadata_hits); + as->arcstat_demand_metadata_misses.value.ui64 = + wmsum_value(&arc_sums.arcstat_demand_metadata_misses); + as->arcstat_prefetch_data_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_prefetch_data_hits); + as->arcstat_prefetch_data_misses.value.ui64 = + wmsum_value(&arc_sums.arcstat_prefetch_data_misses); + as->arcstat_prefetch_metadata_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_prefetch_metadata_hits); + as->arcstat_prefetch_metadata_misses.value.ui64 = + wmsum_value(&arc_sums.arcstat_prefetch_metadata_misses); + as->arcstat_mru_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_mru_hits); + as->arcstat_mru_ghost_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_mru_ghost_hits); + as->arcstat_mfu_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_mfu_hits); + as->arcstat_mfu_ghost_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_mfu_ghost_hits); + as->arcstat_deleted.value.ui64 = + wmsum_value(&arc_sums.arcstat_deleted); + as->arcstat_mutex_miss.value.ui64 = + wmsum_value(&arc_sums.arcstat_mutex_miss); + as->arcstat_access_skip.value.ui64 = + wmsum_value(&arc_sums.arcstat_access_skip); + as->arcstat_evict_skip.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_skip); + as->arcstat_evict_not_enough.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_not_enough); + as->arcstat_evict_l2_cached.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_l2_cached); + as->arcstat_evict_l2_eligible.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_l2_eligible); + as->arcstat_evict_l2_eligible_mfu.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_l2_eligible_mfu); + as->arcstat_evict_l2_eligible_mru.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_l2_eligible_mru); + as->arcstat_evict_l2_ineligible.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_l2_ineligible); + as->arcstat_evict_l2_skip.value.ui64 = + wmsum_value(&arc_sums.arcstat_evict_l2_skip); + as->arcstat_hash_collisions.value.ui64 = + wmsum_value(&arc_sums.arcstat_hash_collisions); + as->arcstat_hash_chains.value.ui64 = + wmsum_value(&arc_sums.arcstat_hash_chains); + as->arcstat_size.value.ui64 = + aggsum_value(&arc_sums.arcstat_size); + as->arcstat_compressed_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_compressed_size); + as->arcstat_uncompressed_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_uncompressed_size); + as->arcstat_overhead_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_overhead_size); + as->arcstat_hdr_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_hdr_size); + as->arcstat_data_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_data_size); + as->arcstat_metadata_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_metadata_size); + as->arcstat_dbuf_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_dbuf_size); #if defined(COMPAT_FREEBSD11) - ARCSTAT(arcstat_other_size) = wmsum_value(&astat_bonus_size) + - aggsum_value(&astat_dnode_size) + - wmsum_value(&astat_dbuf_size); + as->arcstat_other_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_bonus_size) + + aggsum_value(&arc_sums.arcstat_dnode_size) + + wmsum_value(&arc_sums.arcstat_dbuf_size); #endif - ARCSTAT(arcstat_dnode_size) = aggsum_value(&astat_dnode_size); - ARCSTAT(arcstat_bonus_size) = wmsum_value(&astat_bonus_size); - ARCSTAT(arcstat_abd_chunk_waste_size) = - wmsum_value(&astat_abd_chunk_waste_size); - as->arcstat_memory_all_bytes.value.ui64 = - arc_all_memory(); - as->arcstat_memory_free_bytes.value.ui64 = - arc_free_memory(); - as->arcstat_memory_available_bytes.value.i64 = - arc_available_memory(); - } + arc_kstat_update_state(arc_anon, + &as->arcstat_anon_size, + &as->arcstat_anon_evictable_data, + &as->arcstat_anon_evictable_metadata); + arc_kstat_update_state(arc_mru, + &as->arcstat_mru_size, + &as->arcstat_mru_evictable_data, + &as->arcstat_mru_evictable_metadata); + arc_kstat_update_state(arc_mru_ghost, + &as->arcstat_mru_ghost_size, + &as->arcstat_mru_ghost_evictable_data, + &as->arcstat_mru_ghost_evictable_metadata); + arc_kstat_update_state(arc_mfu, + &as->arcstat_mfu_size, + &as->arcstat_mfu_evictable_data, + &as->arcstat_mfu_evictable_metadata); + arc_kstat_update_state(arc_mfu_ghost, + &as->arcstat_mfu_ghost_size, + &as->arcstat_mfu_ghost_evictable_data, + &as->arcstat_mfu_ghost_evictable_metadata); + + as->arcstat_dnode_size.value.ui64 = + aggsum_value(&arc_sums.arcstat_dnode_size); + as->arcstat_bonus_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_bonus_size); + as->arcstat_l2_hits.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_hits); + as->arcstat_l2_misses.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_misses); + as->arcstat_l2_prefetch_asize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_prefetch_asize); + as->arcstat_l2_mru_asize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_mru_asize); + as->arcstat_l2_mfu_asize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_mfu_asize); + as->arcstat_l2_bufc_data_asize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_bufc_data_asize); + as->arcstat_l2_bufc_metadata_asize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_bufc_metadata_asize); + as->arcstat_l2_feeds.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_feeds); + as->arcstat_l2_rw_clash.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rw_clash); + as->arcstat_l2_read_bytes.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_read_bytes); + as->arcstat_l2_write_bytes.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_write_bytes); + as->arcstat_l2_writes_sent.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_writes_sent); + as->arcstat_l2_writes_done.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_writes_done); + as->arcstat_l2_writes_error.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_writes_error); + as->arcstat_l2_writes_lock_retry.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_writes_lock_retry); + as->arcstat_l2_evict_lock_retry.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_evict_lock_retry); + as->arcstat_l2_evict_reading.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_evict_reading); + as->arcstat_l2_evict_l1cached.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_evict_l1cached); + as->arcstat_l2_free_on_write.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_free_on_write); + as->arcstat_l2_abort_lowmem.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_abort_lowmem); + as->arcstat_l2_cksum_bad.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_cksum_bad); + as->arcstat_l2_io_error.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_io_error); + as->arcstat_l2_lsize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_lsize); + as->arcstat_l2_psize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_psize); + as->arcstat_l2_hdr_size.value.ui64 = + aggsum_value(&arc_sums.arcstat_l2_hdr_size); + as->arcstat_l2_log_blk_writes.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_log_blk_writes); + as->arcstat_l2_log_blk_asize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_log_blk_asize); + as->arcstat_l2_log_blk_count.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_log_blk_count); + as->arcstat_l2_rebuild_success.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_success); + as->arcstat_l2_rebuild_abort_unsupported.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_unsupported); + as->arcstat_l2_rebuild_abort_io_errors.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_io_errors); + as->arcstat_l2_rebuild_abort_dh_errors.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_dh_errors); + as->arcstat_l2_rebuild_abort_cksum_lb_errors.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_cksum_lb_errors); + as->arcstat_l2_rebuild_abort_lowmem.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_abort_lowmem); + as->arcstat_l2_rebuild_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_size); + as->arcstat_l2_rebuild_asize.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_asize); + as->arcstat_l2_rebuild_bufs.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_bufs); + as->arcstat_l2_rebuild_bufs_precached.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_bufs_precached); + as->arcstat_l2_rebuild_log_blks.value.ui64 = + wmsum_value(&arc_sums.arcstat_l2_rebuild_log_blks); + as->arcstat_memory_throttle_count.value.ui64 = + wmsum_value(&arc_sums.arcstat_memory_throttle_count); + as->arcstat_memory_direct_count.value.ui64 = + wmsum_value(&arc_sums.arcstat_memory_direct_count); + as->arcstat_memory_indirect_count.value.ui64 = + wmsum_value(&arc_sums.arcstat_memory_indirect_count); + + as->arcstat_memory_all_bytes.value.ui64 = + arc_all_memory(); + as->arcstat_memory_free_bytes.value.ui64 = + arc_free_memory(); + as->arcstat_memory_available_bytes.value.i64 = + arc_available_memory(); + + as->arcstat_prune.value.ui64 = + wmsum_value(&arc_sums.arcstat_prune); + as->arcstat_meta_used.value.ui64 = + aggsum_value(&arc_sums.arcstat_meta_used); + as->arcstat_async_upgrade_sync.value.ui64 = + wmsum_value(&arc_sums.arcstat_async_upgrade_sync); + as->arcstat_demand_hit_predictive_prefetch.value.ui64 = + wmsum_value(&arc_sums.arcstat_demand_hit_predictive_prefetch); + as->arcstat_demand_hit_prescient_prefetch.value.ui64 = + wmsum_value(&arc_sums.arcstat_demand_hit_prescient_prefetch); + as->arcstat_raw_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_raw_size); + as->arcstat_cached_only_in_progress.value.ui64 = + wmsum_value(&arc_sums.arcstat_cached_only_in_progress); + as->arcstat_abd_chunk_waste_size.value.ui64 = + wmsum_value(&arc_sums.arcstat_abd_chunk_waste_size); return (0); } /* * This function *must* return indices evenly distributed between all * sublists of the multilist. This is needed due to how the ARC eviction * code is laid out; arc_evict_state() assumes ARC buffers are evenly * distributed between all sublists and uses this assumption when * deciding which sublist to evict from and how much to evict from it. */ static unsigned int arc_state_multilist_index_func(multilist_t *ml, void *obj) { arc_buf_hdr_t *hdr = obj; /* * We rely on b_dva to generate evenly distributed index * numbers using buf_hash below. So, as an added precaution, * let's make sure we never add empty buffers to the arc lists. */ ASSERT(!HDR_EMPTY(hdr)); /* * The assumption here, is the hash value for a given * arc_buf_hdr_t will remain constant throughout its lifetime * (i.e. its b_spa, b_dva, and b_birth fields don't change). * Thus, we don't need to store the header's sublist index * on insertion, as this index can be recalculated on removal. * * Also, the low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. */ return (buf_hash(hdr->b_spa, &hdr->b_dva, hdr->b_birth) % multilist_get_num_sublists(ml)); } #define WARN_IF_TUNING_IGNORED(tuning, value, do_warn) do { \ if ((do_warn) && (tuning) && ((tuning) != (value))) { \ cmn_err(CE_WARN, \ "ignoring tunable %s (using %llu instead)", \ (#tuning), (value)); \ } \ } while (0) /* * Called during module initialization and periodically thereafter to * apply reasonable changes to the exposed performance tunings. Can also be * called explicitly by param_set_arc_*() functions when ARC tunables are * updated manually. Non-zero zfs_* values which differ from the currently set * values will be applied. */ void arc_tuning_update(boolean_t verbose) { uint64_t allmem = arc_all_memory(); unsigned long limit; /* Valid range: 32M - */ if ((zfs_arc_min) && (zfs_arc_min != arc_c_min) && (zfs_arc_min >= 2ULL << SPA_MAXBLOCKSHIFT) && (zfs_arc_min <= arc_c_max)) { arc_c_min = zfs_arc_min; arc_c = MAX(arc_c, arc_c_min); } WARN_IF_TUNING_IGNORED(zfs_arc_min, arc_c_min, verbose); /* Valid range: 64M - */ if ((zfs_arc_max) && (zfs_arc_max != arc_c_max) && (zfs_arc_max >= 64 << 20) && (zfs_arc_max < allmem) && (zfs_arc_max > arc_c_min)) { arc_c_max = zfs_arc_max; arc_c = MIN(arc_c, arc_c_max); arc_p = (arc_c >> 1); if (arc_meta_limit > arc_c_max) arc_meta_limit = arc_c_max; if (arc_dnode_size_limit > arc_meta_limit) arc_dnode_size_limit = arc_meta_limit; } WARN_IF_TUNING_IGNORED(zfs_arc_max, arc_c_max, verbose); /* Valid range: 16M - */ if ((zfs_arc_meta_min) && (zfs_arc_meta_min != arc_meta_min) && (zfs_arc_meta_min >= 1ULL << SPA_MAXBLOCKSHIFT) && (zfs_arc_meta_min <= arc_c_max)) { arc_meta_min = zfs_arc_meta_min; if (arc_meta_limit < arc_meta_min) arc_meta_limit = arc_meta_min; if (arc_dnode_size_limit < arc_meta_min) arc_dnode_size_limit = arc_meta_min; } WARN_IF_TUNING_IGNORED(zfs_arc_meta_min, arc_meta_min, verbose); /* Valid range: - */ limit = zfs_arc_meta_limit ? zfs_arc_meta_limit : MIN(zfs_arc_meta_limit_percent, 100) * arc_c_max / 100; if ((limit != arc_meta_limit) && (limit >= arc_meta_min) && (limit <= arc_c_max)) arc_meta_limit = limit; WARN_IF_TUNING_IGNORED(zfs_arc_meta_limit, arc_meta_limit, verbose); /* Valid range: - */ limit = zfs_arc_dnode_limit ? zfs_arc_dnode_limit : MIN(zfs_arc_dnode_limit_percent, 100) * arc_meta_limit / 100; if ((limit != arc_dnode_size_limit) && (limit >= arc_meta_min) && (limit <= arc_meta_limit)) arc_dnode_size_limit = limit; WARN_IF_TUNING_IGNORED(zfs_arc_dnode_limit, arc_dnode_size_limit, verbose); /* Valid range: 1 - N */ if (zfs_arc_grow_retry) arc_grow_retry = zfs_arc_grow_retry; /* Valid range: 1 - N */ if (zfs_arc_shrink_shift) { arc_shrink_shift = zfs_arc_shrink_shift; arc_no_grow_shift = MIN(arc_no_grow_shift, arc_shrink_shift -1); } /* Valid range: 1 - N */ if (zfs_arc_p_min_shift) arc_p_min_shift = zfs_arc_p_min_shift; /* Valid range: 1 - N ms */ if (zfs_arc_min_prefetch_ms) arc_min_prefetch_ms = zfs_arc_min_prefetch_ms; /* Valid range: 1 - N ms */ if (zfs_arc_min_prescient_prefetch_ms) { arc_min_prescient_prefetch_ms = zfs_arc_min_prescient_prefetch_ms; } /* Valid range: 0 - 100 */ if ((zfs_arc_lotsfree_percent >= 0) && (zfs_arc_lotsfree_percent <= 100)) arc_lotsfree_percent = zfs_arc_lotsfree_percent; WARN_IF_TUNING_IGNORED(zfs_arc_lotsfree_percent, arc_lotsfree_percent, verbose); /* Valid range: 0 - */ if ((zfs_arc_sys_free) && (zfs_arc_sys_free != arc_sys_free)) arc_sys_free = MIN(MAX(zfs_arc_sys_free, 0), allmem); WARN_IF_TUNING_IGNORED(zfs_arc_sys_free, arc_sys_free, verbose); } static void arc_state_init(void) { arc_anon = &ARC_anon; arc_mru = &ARC_mru; arc_mru_ghost = &ARC_mru_ghost; arc_mfu = &ARC_mfu; arc_mfu_ghost = &ARC_mfu_ghost; arc_l2c_only = &ARC_l2c_only; multilist_create(&arc_mru->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_mru->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_mfu->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_mfu->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); multilist_create(&arc_l2c_only->arcs_list[ARC_BUFC_DATA], sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l1hdr.b_arc_node), arc_state_multilist_index_func); zfs_refcount_create(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_create(&arc_anon->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_create(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_create(&arc_mru->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_create(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_create(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_create(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_create(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_create(&arc_anon->arcs_size); zfs_refcount_create(&arc_mru->arcs_size); zfs_refcount_create(&arc_mru_ghost->arcs_size); zfs_refcount_create(&arc_mfu->arcs_size); zfs_refcount_create(&arc_mfu_ghost->arcs_size); zfs_refcount_create(&arc_l2c_only->arcs_size); - aggsum_init(&arc_meta_used, 0); - aggsum_init(&arc_size, 0); - wmsum_init(&astat_data_size, 0); - wmsum_init(&astat_metadata_size, 0); - wmsum_init(&astat_hdr_size, 0); - aggsum_init(&astat_l2_hdr_size, 0); - wmsum_init(&astat_bonus_size, 0); - aggsum_init(&astat_dnode_size, 0); - wmsum_init(&astat_dbuf_size, 0); - wmsum_init(&astat_abd_chunk_waste_size, 0); + wmsum_init(&arc_sums.arcstat_hits, 0); + wmsum_init(&arc_sums.arcstat_misses, 0); + wmsum_init(&arc_sums.arcstat_demand_data_hits, 0); + wmsum_init(&arc_sums.arcstat_demand_data_misses, 0); + wmsum_init(&arc_sums.arcstat_demand_metadata_hits, 0); + wmsum_init(&arc_sums.arcstat_demand_metadata_misses, 0); + wmsum_init(&arc_sums.arcstat_prefetch_data_hits, 0); + wmsum_init(&arc_sums.arcstat_prefetch_data_misses, 0); + wmsum_init(&arc_sums.arcstat_prefetch_metadata_hits, 0); + wmsum_init(&arc_sums.arcstat_prefetch_metadata_misses, 0); + wmsum_init(&arc_sums.arcstat_mru_hits, 0); + wmsum_init(&arc_sums.arcstat_mru_ghost_hits, 0); + wmsum_init(&arc_sums.arcstat_mfu_hits, 0); + wmsum_init(&arc_sums.arcstat_mfu_ghost_hits, 0); + wmsum_init(&arc_sums.arcstat_deleted, 0); + wmsum_init(&arc_sums.arcstat_mutex_miss, 0); + wmsum_init(&arc_sums.arcstat_access_skip, 0); + wmsum_init(&arc_sums.arcstat_evict_skip, 0); + wmsum_init(&arc_sums.arcstat_evict_not_enough, 0); + wmsum_init(&arc_sums.arcstat_evict_l2_cached, 0); + wmsum_init(&arc_sums.arcstat_evict_l2_eligible, 0); + wmsum_init(&arc_sums.arcstat_evict_l2_eligible_mfu, 0); + wmsum_init(&arc_sums.arcstat_evict_l2_eligible_mru, 0); + wmsum_init(&arc_sums.arcstat_evict_l2_ineligible, 0); + wmsum_init(&arc_sums.arcstat_evict_l2_skip, 0); + wmsum_init(&arc_sums.arcstat_hash_collisions, 0); + wmsum_init(&arc_sums.arcstat_hash_chains, 0); + aggsum_init(&arc_sums.arcstat_size, 0); + wmsum_init(&arc_sums.arcstat_compressed_size, 0); + wmsum_init(&arc_sums.arcstat_uncompressed_size, 0); + wmsum_init(&arc_sums.arcstat_overhead_size, 0); + wmsum_init(&arc_sums.arcstat_hdr_size, 0); + wmsum_init(&arc_sums.arcstat_data_size, 0); + wmsum_init(&arc_sums.arcstat_metadata_size, 0); + wmsum_init(&arc_sums.arcstat_dbuf_size, 0); + aggsum_init(&arc_sums.arcstat_dnode_size, 0); + wmsum_init(&arc_sums.arcstat_bonus_size, 0); + wmsum_init(&arc_sums.arcstat_l2_hits, 0); + wmsum_init(&arc_sums.arcstat_l2_misses, 0); + wmsum_init(&arc_sums.arcstat_l2_prefetch_asize, 0); + wmsum_init(&arc_sums.arcstat_l2_mru_asize, 0); + wmsum_init(&arc_sums.arcstat_l2_mfu_asize, 0); + wmsum_init(&arc_sums.arcstat_l2_bufc_data_asize, 0); + wmsum_init(&arc_sums.arcstat_l2_bufc_metadata_asize, 0); + wmsum_init(&arc_sums.arcstat_l2_feeds, 0); + wmsum_init(&arc_sums.arcstat_l2_rw_clash, 0); + wmsum_init(&arc_sums.arcstat_l2_read_bytes, 0); + wmsum_init(&arc_sums.arcstat_l2_write_bytes, 0); + wmsum_init(&arc_sums.arcstat_l2_writes_sent, 0); + wmsum_init(&arc_sums.arcstat_l2_writes_done, 0); + wmsum_init(&arc_sums.arcstat_l2_writes_error, 0); + wmsum_init(&arc_sums.arcstat_l2_writes_lock_retry, 0); + wmsum_init(&arc_sums.arcstat_l2_evict_lock_retry, 0); + wmsum_init(&arc_sums.arcstat_l2_evict_reading, 0); + wmsum_init(&arc_sums.arcstat_l2_evict_l1cached, 0); + wmsum_init(&arc_sums.arcstat_l2_free_on_write, 0); + wmsum_init(&arc_sums.arcstat_l2_abort_lowmem, 0); + wmsum_init(&arc_sums.arcstat_l2_cksum_bad, 0); + wmsum_init(&arc_sums.arcstat_l2_io_error, 0); + wmsum_init(&arc_sums.arcstat_l2_lsize, 0); + wmsum_init(&arc_sums.arcstat_l2_psize, 0); + aggsum_init(&arc_sums.arcstat_l2_hdr_size, 0); + wmsum_init(&arc_sums.arcstat_l2_log_blk_writes, 0); + wmsum_init(&arc_sums.arcstat_l2_log_blk_asize, 0); + wmsum_init(&arc_sums.arcstat_l2_log_blk_count, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_success, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_unsupported, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_io_errors, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_dh_errors, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_cksum_lb_errors, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_abort_lowmem, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_size, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_asize, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_bufs, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_bufs_precached, 0); + wmsum_init(&arc_sums.arcstat_l2_rebuild_log_blks, 0); + wmsum_init(&arc_sums.arcstat_memory_throttle_count, 0); + wmsum_init(&arc_sums.arcstat_memory_direct_count, 0); + wmsum_init(&arc_sums.arcstat_memory_indirect_count, 0); + wmsum_init(&arc_sums.arcstat_prune, 0); + aggsum_init(&arc_sums.arcstat_meta_used, 0); + wmsum_init(&arc_sums.arcstat_async_upgrade_sync, 0); + wmsum_init(&arc_sums.arcstat_demand_hit_predictive_prefetch, 0); + wmsum_init(&arc_sums.arcstat_demand_hit_prescient_prefetch, 0); + wmsum_init(&arc_sums.arcstat_raw_size, 0); + wmsum_init(&arc_sums.arcstat_cached_only_in_progress, 0); + wmsum_init(&arc_sums.arcstat_abd_chunk_waste_size, 0); arc_anon->arcs_state = ARC_STATE_ANON; arc_mru->arcs_state = ARC_STATE_MRU; arc_mru_ghost->arcs_state = ARC_STATE_MRU_GHOST; arc_mfu->arcs_state = ARC_STATE_MFU; arc_mfu_ghost->arcs_state = ARC_STATE_MFU_GHOST; arc_l2c_only->arcs_state = ARC_STATE_L2C_ONLY; } static void arc_state_fini(void) { zfs_refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_destroy(&arc_anon->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_destroy(&arc_mru->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_destroy(&arc_mru_ghost->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_destroy(&arc_mfu->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_destroy(&arc_mfu_ghost->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_METADATA]); zfs_refcount_destroy(&arc_l2c_only->arcs_esize[ARC_BUFC_DATA]); zfs_refcount_destroy(&arc_anon->arcs_size); zfs_refcount_destroy(&arc_mru->arcs_size); zfs_refcount_destroy(&arc_mru_ghost->arcs_size); zfs_refcount_destroy(&arc_mfu->arcs_size); zfs_refcount_destroy(&arc_mfu_ghost->arcs_size); zfs_refcount_destroy(&arc_l2c_only->arcs_size); multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(&arc_mru->arcs_list[ARC_BUFC_DATA]); multilist_destroy(&arc_mru_ghost->arcs_list[ARC_BUFC_DATA]); multilist_destroy(&arc_mfu->arcs_list[ARC_BUFC_DATA]); multilist_destroy(&arc_mfu_ghost->arcs_list[ARC_BUFC_DATA]); multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_METADATA]); multilist_destroy(&arc_l2c_only->arcs_list[ARC_BUFC_DATA]); - aggsum_fini(&arc_meta_used); - aggsum_fini(&arc_size); - wmsum_fini(&astat_data_size); - wmsum_fini(&astat_metadata_size); - wmsum_fini(&astat_hdr_size); - aggsum_fini(&astat_l2_hdr_size); - wmsum_fini(&astat_bonus_size); - aggsum_fini(&astat_dnode_size); - wmsum_fini(&astat_dbuf_size); - wmsum_fini(&astat_abd_chunk_waste_size); + wmsum_fini(&arc_sums.arcstat_hits); + wmsum_fini(&arc_sums.arcstat_misses); + wmsum_fini(&arc_sums.arcstat_demand_data_hits); + wmsum_fini(&arc_sums.arcstat_demand_data_misses); + wmsum_fini(&arc_sums.arcstat_demand_metadata_hits); + wmsum_fini(&arc_sums.arcstat_demand_metadata_misses); + wmsum_fini(&arc_sums.arcstat_prefetch_data_hits); + wmsum_fini(&arc_sums.arcstat_prefetch_data_misses); + wmsum_fini(&arc_sums.arcstat_prefetch_metadata_hits); + wmsum_fini(&arc_sums.arcstat_prefetch_metadata_misses); + wmsum_fini(&arc_sums.arcstat_mru_hits); + wmsum_fini(&arc_sums.arcstat_mru_ghost_hits); + wmsum_fini(&arc_sums.arcstat_mfu_hits); + wmsum_fini(&arc_sums.arcstat_mfu_ghost_hits); + wmsum_fini(&arc_sums.arcstat_deleted); + wmsum_fini(&arc_sums.arcstat_mutex_miss); + wmsum_fini(&arc_sums.arcstat_access_skip); + wmsum_fini(&arc_sums.arcstat_evict_skip); + wmsum_fini(&arc_sums.arcstat_evict_not_enough); + wmsum_fini(&arc_sums.arcstat_evict_l2_cached); + wmsum_fini(&arc_sums.arcstat_evict_l2_eligible); + wmsum_fini(&arc_sums.arcstat_evict_l2_eligible_mfu); + wmsum_fini(&arc_sums.arcstat_evict_l2_eligible_mru); + wmsum_fini(&arc_sums.arcstat_evict_l2_ineligible); + wmsum_fini(&arc_sums.arcstat_evict_l2_skip); + wmsum_fini(&arc_sums.arcstat_hash_collisions); + wmsum_fini(&arc_sums.arcstat_hash_chains); + aggsum_fini(&arc_sums.arcstat_size); + wmsum_fini(&arc_sums.arcstat_compressed_size); + wmsum_fini(&arc_sums.arcstat_uncompressed_size); + wmsum_fini(&arc_sums.arcstat_overhead_size); + wmsum_fini(&arc_sums.arcstat_hdr_size); + wmsum_fini(&arc_sums.arcstat_data_size); + wmsum_fini(&arc_sums.arcstat_metadata_size); + wmsum_fini(&arc_sums.arcstat_dbuf_size); + aggsum_fini(&arc_sums.arcstat_dnode_size); + wmsum_fini(&arc_sums.arcstat_bonus_size); + wmsum_fini(&arc_sums.arcstat_l2_hits); + wmsum_fini(&arc_sums.arcstat_l2_misses); + wmsum_fini(&arc_sums.arcstat_l2_prefetch_asize); + wmsum_fini(&arc_sums.arcstat_l2_mru_asize); + wmsum_fini(&arc_sums.arcstat_l2_mfu_asize); + wmsum_fini(&arc_sums.arcstat_l2_bufc_data_asize); + wmsum_fini(&arc_sums.arcstat_l2_bufc_metadata_asize); + wmsum_fini(&arc_sums.arcstat_l2_feeds); + wmsum_fini(&arc_sums.arcstat_l2_rw_clash); + wmsum_fini(&arc_sums.arcstat_l2_read_bytes); + wmsum_fini(&arc_sums.arcstat_l2_write_bytes); + wmsum_fini(&arc_sums.arcstat_l2_writes_sent); + wmsum_fini(&arc_sums.arcstat_l2_writes_done); + wmsum_fini(&arc_sums.arcstat_l2_writes_error); + wmsum_fini(&arc_sums.arcstat_l2_writes_lock_retry); + wmsum_fini(&arc_sums.arcstat_l2_evict_lock_retry); + wmsum_fini(&arc_sums.arcstat_l2_evict_reading); + wmsum_fini(&arc_sums.arcstat_l2_evict_l1cached); + wmsum_fini(&arc_sums.arcstat_l2_free_on_write); + wmsum_fini(&arc_sums.arcstat_l2_abort_lowmem); + wmsum_fini(&arc_sums.arcstat_l2_cksum_bad); + wmsum_fini(&arc_sums.arcstat_l2_io_error); + wmsum_fini(&arc_sums.arcstat_l2_lsize); + wmsum_fini(&arc_sums.arcstat_l2_psize); + aggsum_fini(&arc_sums.arcstat_l2_hdr_size); + wmsum_fini(&arc_sums.arcstat_l2_log_blk_writes); + wmsum_fini(&arc_sums.arcstat_l2_log_blk_asize); + wmsum_fini(&arc_sums.arcstat_l2_log_blk_count); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_success); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_unsupported); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_io_errors); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_dh_errors); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_cksum_lb_errors); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_abort_lowmem); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_size); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_asize); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_bufs); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_bufs_precached); + wmsum_fini(&arc_sums.arcstat_l2_rebuild_log_blks); + wmsum_fini(&arc_sums.arcstat_memory_throttle_count); + wmsum_fini(&arc_sums.arcstat_memory_direct_count); + wmsum_fini(&arc_sums.arcstat_memory_indirect_count); + wmsum_fini(&arc_sums.arcstat_prune); + aggsum_fini(&arc_sums.arcstat_meta_used); + wmsum_fini(&arc_sums.arcstat_async_upgrade_sync); + wmsum_fini(&arc_sums.arcstat_demand_hit_predictive_prefetch); + wmsum_fini(&arc_sums.arcstat_demand_hit_prescient_prefetch); + wmsum_fini(&arc_sums.arcstat_raw_size); + wmsum_fini(&arc_sums.arcstat_cached_only_in_progress); + wmsum_fini(&arc_sums.arcstat_abd_chunk_waste_size); } uint64_t arc_target_bytes(void) { return (arc_c); } void arc_set_limits(uint64_t allmem) { /* Set min cache to 1/32 of all memory, or 32MB, whichever is more. */ arc_c_min = MAX(allmem / 32, 2ULL << SPA_MAXBLOCKSHIFT); /* How to set default max varies by platform. */ arc_c_max = arc_default_max(arc_c_min, allmem); } void arc_init(void) { uint64_t percent, allmem = arc_all_memory(); mutex_init(&arc_evict_lock, NULL, MUTEX_DEFAULT, NULL); list_create(&arc_evict_waiters, sizeof (arc_evict_waiter_t), offsetof(arc_evict_waiter_t, aew_node)); arc_min_prefetch_ms = 1000; arc_min_prescient_prefetch_ms = 6000; #if defined(_KERNEL) arc_lowmem_init(); #endif arc_set_limits(allmem); #ifndef _KERNEL /* * In userland, there's only the memory pressure that we artificially * create (see arc_available_memory()). Don't let arc_c get too * small, because it can cause transactions to be larger than * arc_c, causing arc_tempreserve_space() to fail. */ arc_c_min = MAX(arc_c_max / 2, 2ULL << SPA_MAXBLOCKSHIFT); #endif arc_c = arc_c_min; arc_p = (arc_c >> 1); /* Set min to 1/2 of arc_c_min */ arc_meta_min = 1ULL << SPA_MAXBLOCKSHIFT; - /* Initialize maximum observed usage to zero */ - arc_meta_max = 0; /* * Set arc_meta_limit to a percent of arc_c_max with a floor of * arc_meta_min, and a ceiling of arc_c_max. */ percent = MIN(zfs_arc_meta_limit_percent, 100); arc_meta_limit = MAX(arc_meta_min, (percent * arc_c_max) / 100); percent = MIN(zfs_arc_dnode_limit_percent, 100); arc_dnode_size_limit = (percent * arc_meta_limit) / 100; /* Apply user specified tunings */ arc_tuning_update(B_TRUE); /* if kmem_flags are set, lets try to use less memory */ if (kmem_debugging()) arc_c = arc_c / 2; if (arc_c < arc_c_min) arc_c = arc_c_min; arc_register_hotplug(); arc_state_init(); buf_init(); list_create(&arc_prune_list, sizeof (arc_prune_t), offsetof(arc_prune_t, p_node)); mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL); arc_prune_taskq = taskq_create("arc_prune", 100, defclsyspri, boot_ncpus, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT); arc_ksp = kstat_create("zfs", 0, "arcstats", "misc", KSTAT_TYPE_NAMED, sizeof (arc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (arc_ksp != NULL) { arc_ksp->ks_data = &arc_stats; arc_ksp->ks_update = arc_kstat_update; kstat_install(arc_ksp); } arc_evict_zthr = zthr_create("arc_evict", arc_evict_cb_check, arc_evict_cb, NULL); arc_reap_zthr = zthr_create_timer("arc_reap", arc_reap_cb_check, arc_reap_cb, NULL, SEC2NSEC(1)); arc_warm = B_FALSE; /* * Calculate maximum amount of dirty data per pool. * * If it has been set by a module parameter, take that. * Otherwise, use a percentage of physical memory defined by * zfs_dirty_data_max_percent (default 10%) with a cap at * zfs_dirty_data_max_max (default 4G or 25% of physical memory). */ #ifdef __LP64__ if (zfs_dirty_data_max_max == 0) zfs_dirty_data_max_max = MIN(4ULL * 1024 * 1024 * 1024, allmem * zfs_dirty_data_max_max_percent / 100); #else if (zfs_dirty_data_max_max == 0) zfs_dirty_data_max_max = MIN(1ULL * 1024 * 1024 * 1024, allmem * zfs_dirty_data_max_max_percent / 100); #endif if (zfs_dirty_data_max == 0) { zfs_dirty_data_max = allmem * zfs_dirty_data_max_percent / 100; zfs_dirty_data_max = MIN(zfs_dirty_data_max, zfs_dirty_data_max_max); } } void arc_fini(void) { arc_prune_t *p; #ifdef _KERNEL arc_lowmem_fini(); #endif /* _KERNEL */ /* Use B_TRUE to ensure *all* buffers are evicted */ arc_flush(NULL, B_TRUE); if (arc_ksp != NULL) { kstat_delete(arc_ksp); arc_ksp = NULL; } taskq_wait(arc_prune_taskq); taskq_destroy(arc_prune_taskq); mutex_enter(&arc_prune_mtx); while ((p = list_head(&arc_prune_list)) != NULL) { list_remove(&arc_prune_list, p); zfs_refcount_remove(&p->p_refcnt, &arc_prune_list); zfs_refcount_destroy(&p->p_refcnt); kmem_free(p, sizeof (*p)); } mutex_exit(&arc_prune_mtx); list_destroy(&arc_prune_list); mutex_destroy(&arc_prune_mtx); (void) zthr_cancel(arc_evict_zthr); (void) zthr_cancel(arc_reap_zthr); mutex_destroy(&arc_evict_lock); list_destroy(&arc_evict_waiters); /* * Free any buffers that were tagged for destruction. This needs * to occur before arc_state_fini() runs and destroys the aggsum * values which are updated when freeing scatter ABDs. */ l2arc_do_free_on_write(); /* * buf_fini() must proceed arc_state_fini() because buf_fin() may * trigger the release of kmem magazines, which can callback to * arc_space_return() which accesses aggsums freed in act_state_fini(). */ buf_fini(); arc_state_fini(); arc_unregister_hotplug(); /* * We destroy the zthrs after all the ARC state has been * torn down to avoid the case of them receiving any * wakeup() signals after they are destroyed. */ zthr_destroy(arc_evict_zthr); zthr_destroy(arc_reap_zthr); ASSERT0(arc_loaned_bytes); } /* * Level 2 ARC * * The level 2 ARC (L2ARC) is a cache layer in-between main memory and disk. * It uses dedicated storage devices to hold cached data, which are populated * using large infrequent writes. The main role of this cache is to boost * the performance of random read workloads. The intended L2ARC devices * include short-stroked disks, solid state disks, and other media with * substantially faster read latency than disk. * * +-----------------------+ * | ARC | * +-----------------------+ * | ^ ^ * | | | * l2arc_feed_thread() arc_read() * | | | * | l2arc read | * V | | * +---------------+ | * | L2ARC | | * +---------------+ | * | ^ | * l2arc_write() | | * | | | * V | | * +-------+ +-------+ * | vdev | | vdev | * | cache | | cache | * +-------+ +-------+ * +=========+ .-----. * : L2ARC : |-_____-| * : devices : | Disks | * +=========+ `-_____-' * * Read requests are satisfied from the following sources, in order: * * 1) ARC * 2) vdev cache of L2ARC devices * 3) L2ARC devices * 4) vdev cache of disks * 5) disks * * Some L2ARC device types exhibit extremely slow write performance. * To accommodate for this there are some significant differences between * the L2ARC and traditional cache design: * * 1. There is no eviction path from the ARC to the L2ARC. Evictions from * the ARC behave as usual, freeing buffers and placing headers on ghost * lists. The ARC does not send buffers to the L2ARC during eviction as * this would add inflated write latencies for all ARC memory pressure. * * 2. The L2ARC attempts to cache data from the ARC before it is evicted. * It does this by periodically scanning buffers from the eviction-end of * the MFU and MRU ARC lists, copying them to the L2ARC devices if they are * not already there. It scans until a headroom of buffers is satisfied, * which itself is a buffer for ARC eviction. If a compressible buffer is * found during scanning and selected for writing to an L2ARC device, we * temporarily boost scanning headroom during the next scan cycle to make * sure we adapt to compression effects (which might significantly reduce * the data volume we write to L2ARC). The thread that does this is * l2arc_feed_thread(), illustrated below; example sizes are included to * provide a better sense of ratio than this diagram: * * head --> tail * +---------------------+----------+ * ARC_mfu |:::::#:::::::::::::::|o#o###o###|-->. # already on L2ARC * +---------------------+----------+ | o L2ARC eligible * ARC_mru |:#:::::::::::::::::::|#o#ooo####|-->| : ARC buffer * +---------------------+----------+ | * 15.9 Gbytes ^ 32 Mbytes | * headroom | * l2arc_feed_thread() * | * l2arc write hand <--[oooo]--' * | 8 Mbyte * | write max * V * +==============================+ * L2ARC dev |####|#|###|###| |####| ... | * +==============================+ * 32 Gbytes * * 3. If an ARC buffer is copied to the L2ARC but then hit instead of * evicted, then the L2ARC has cached a buffer much sooner than it probably * needed to, potentially wasting L2ARC device bandwidth and storage. It is * safe to say that this is an uncommon case, since buffers at the end of * the ARC lists have moved there due to inactivity. * * 4. If the ARC evicts faster than the L2ARC can maintain a headroom, * then the L2ARC simply misses copying some buffers. This serves as a * pressure valve to prevent heavy read workloads from both stalling the ARC * with waits and clogging the L2ARC with writes. This also helps prevent * the potential for the L2ARC to churn if it attempts to cache content too * quickly, such as during backups of the entire pool. * * 5. After system boot and before the ARC has filled main memory, there are * no evictions from the ARC and so the tails of the ARC_mfu and ARC_mru * lists can remain mostly static. Instead of searching from tail of these * lists as pictured, the l2arc_feed_thread() will search from the list heads * for eligible buffers, greatly increasing its chance of finding them. * * The L2ARC device write speed is also boosted during this time so that * the L2ARC warms up faster. Since there have been no ARC evictions yet, * there are no L2ARC reads, and no fear of degrading read performance * through increased writes. * * 6. Writes to the L2ARC devices are grouped and sent in-sequence, so that * the vdev queue can aggregate them into larger and fewer writes. Each * device is written to in a rotor fashion, sweeping writes through * available space then repeating. * * 7. The L2ARC does not store dirty content. It never needs to flush * write buffers back to disk based storage. * * 8. If an ARC buffer is written (and dirtied) which also exists in the * L2ARC, the now stale L2ARC buffer is immediately dropped. * * The performance of the L2ARC can be tweaked by a number of tunables, which * may be necessary for different workloads: * * l2arc_write_max max write bytes per interval * l2arc_write_boost extra write bytes during device warmup * l2arc_noprefetch skip caching prefetched buffers * l2arc_headroom number of max device writes to precache * l2arc_headroom_boost when we find compressed buffers during ARC * scanning, we multiply headroom by this * percentage factor for the next scan cycle, * since more compressed buffers are likely to * be present * l2arc_feed_secs seconds between L2ARC writing * * Tunables may be removed or added as future performance improvements are * integrated, and also may become zpool properties. * * There are three key functions that control how the L2ARC warms up: * * l2arc_write_eligible() check if a buffer is eligible to cache * l2arc_write_size() calculate how much to write * l2arc_write_interval() calculate sleep delay between writes * * These three functions determine what to write, how much, and how quickly * to send writes. * * L2ARC persistence: * * When writing buffers to L2ARC, we periodically add some metadata to * make sure we can pick them up after reboot, thus dramatically reducing * the impact that any downtime has on the performance of storage systems * with large caches. * * The implementation works fairly simply by integrating the following two * modifications: * * *) When writing to the L2ARC, we occasionally write a "l2arc log block", * which is an additional piece of metadata which describes what's been * written. This allows us to rebuild the arc_buf_hdr_t structures of the * main ARC buffers. There are 2 linked-lists of log blocks headed by * dh_start_lbps[2]. We alternate which chain we append to, so they are * time-wise and offset-wise interleaved, but that is an optimization rather * than for correctness. The log block also includes a pointer to the * previous block in its chain. * * *) We reserve SPA_MINBLOCKSIZE of space at the start of each L2ARC device * for our header bookkeeping purposes. This contains a device header, * which contains our top-level reference structures. We update it each * time we write a new log block, so that we're able to locate it in the * L2ARC device. If this write results in an inconsistent device header * (e.g. due to power failure), we detect this by verifying the header's * checksum and simply fail to reconstruct the L2ARC after reboot. * * Implementation diagram: * * +=== L2ARC device (not to scale) ======================================+ * | ___two newest log block pointers__.__________ | * | / \dh_start_lbps[1] | * | / \ \dh_start_lbps[0]| * |.___/__. V V | * ||L2 dev|....|lb |bufs |lb |bufs |lb |bufs |lb |bufs |lb |---(empty)---| * || hdr| ^ /^ /^ / / | * |+------+ ...--\-------/ \-----/--\------/ / | * | \--------------/ \--------------/ | * +======================================================================+ * * As can be seen on the diagram, rather than using a simple linked list, * we use a pair of linked lists with alternating elements. This is a * performance enhancement due to the fact that we only find out the * address of the next log block access once the current block has been * completely read in. Obviously, this hurts performance, because we'd be * keeping the device's I/O queue at only a 1 operation deep, thus * incurring a large amount of I/O round-trip latency. Having two lists * allows us to fetch two log blocks ahead of where we are currently * rebuilding L2ARC buffers. * * On-device data structures: * * L2ARC device header: l2arc_dev_hdr_phys_t * L2ARC log block: l2arc_log_blk_phys_t * * L2ARC reconstruction: * * When writing data, we simply write in the standard rotary fashion, * evicting buffers as we go and simply writing new data over them (writing * a new log block every now and then). This obviously means that once we * loop around the end of the device, we will start cutting into an already * committed log block (and its referenced data buffers), like so: * * current write head__ __old tail * \ / * V V * <--|bufs |lb |bufs |lb | |bufs |lb |bufs |lb |--> * ^ ^^^^^^^^^___________________________________ * | \ * <> may overwrite this blk and/or its bufs --' * * When importing the pool, we detect this situation and use it to stop * our scanning process (see l2arc_rebuild). * * There is one significant caveat to consider when rebuilding ARC contents * from an L2ARC device: what about invalidated buffers? Given the above * construction, we cannot update blocks which we've already written to amend * them to remove buffers which were invalidated. Thus, during reconstruction, * we might be populating the cache with buffers for data that's not on the * main pool anymore, or may have been overwritten! * * As it turns out, this isn't a problem. Every arc_read request includes * both the DVA and, crucially, the birth TXG of the BP the caller is * looking for. So even if the cache were populated by completely rotten * blocks for data that had been long deleted and/or overwritten, we'll * never actually return bad data from the cache, since the DVA with the * birth TXG uniquely identify a block in space and time - once created, * a block is immutable on disk. The worst thing we have done is wasted * some time and memory at l2arc rebuild to reconstruct outdated ARC * entries that will get dropped from the l2arc as it is being updated * with new blocks. * * L2ARC buffers that have been evicted by l2arc_evict() ahead of the write * hand are not restored. This is done by saving the offset (in bytes) * l2arc_evict() has evicted to in the L2ARC device header and taking it * into account when restoring buffers. */ static boolean_t l2arc_write_eligible(uint64_t spa_guid, arc_buf_hdr_t *hdr) { /* * A buffer is *not* eligible for the L2ARC if it: * 1. belongs to a different spa. * 2. is already cached on the L2ARC. * 3. has an I/O in progress (it may be an incomplete read). * 4. is flagged not eligible (zfs property). */ if (hdr->b_spa != spa_guid || HDR_HAS_L2HDR(hdr) || HDR_IO_IN_PROGRESS(hdr) || !HDR_L2CACHE(hdr)) return (B_FALSE); return (B_TRUE); } static uint64_t l2arc_write_size(l2arc_dev_t *dev) { uint64_t size, dev_size, tsize; /* * Make sure our globals have meaningful values in case the user * altered them. */ size = l2arc_write_max; if (size == 0) { cmn_err(CE_NOTE, "Bad value for l2arc_write_max, value must " "be greater than zero, resetting it to the default (%d)", L2ARC_WRITE_SIZE); size = l2arc_write_max = L2ARC_WRITE_SIZE; } if (arc_warm == B_FALSE) size += l2arc_write_boost; /* * Make sure the write size does not exceed the size of the cache * device. This is important in l2arc_evict(), otherwise infinite * iteration can occur. */ dev_size = dev->l2ad_end - dev->l2ad_start; tsize = size + l2arc_log_blk_overhead(size, dev); if (dev->l2ad_vdev->vdev_has_trim && l2arc_trim_ahead > 0) tsize += MAX(64 * 1024 * 1024, (tsize * l2arc_trim_ahead) / 100); if (tsize >= dev_size) { cmn_err(CE_NOTE, "l2arc_write_max or l2arc_write_boost " "plus the overhead of log blocks (persistent L2ARC, " "%llu bytes) exceeds the size of the cache device " "(guid %llu), resetting them to the default (%d)", l2arc_log_blk_overhead(size, dev), dev->l2ad_vdev->vdev_guid, L2ARC_WRITE_SIZE); size = l2arc_write_max = l2arc_write_boost = L2ARC_WRITE_SIZE; if (arc_warm == B_FALSE) size += l2arc_write_boost; } return (size); } static clock_t l2arc_write_interval(clock_t began, uint64_t wanted, uint64_t wrote) { clock_t interval, next, now; /* * If the ARC lists are busy, increase our write rate; if the * lists are stale, idle back. This is achieved by checking * how much we previously wrote - if it was more than half of * what we wanted, schedule the next write much sooner. */ if (l2arc_feed_again && wrote > (wanted / 2)) interval = (hz * l2arc_feed_min_ms) / 1000; else interval = hz * l2arc_feed_secs; now = ddi_get_lbolt(); next = MAX(now, MIN(now + interval, began + interval)); return (next); } /* * Cycle through L2ARC devices. This is how L2ARC load balances. * If a device is returned, this also returns holding the spa config lock. */ static l2arc_dev_t * l2arc_dev_get_next(void) { l2arc_dev_t *first, *next = NULL; /* * Lock out the removal of spas (spa_namespace_lock), then removal * of cache devices (l2arc_dev_mtx). Once a device has been selected, * both locks will be dropped and a spa config lock held instead. */ mutex_enter(&spa_namespace_lock); mutex_enter(&l2arc_dev_mtx); /* if there are no vdevs, there is nothing to do */ if (l2arc_ndev == 0) goto out; first = NULL; next = l2arc_dev_last; do { /* loop around the list looking for a non-faulted vdev */ if (next == NULL) { next = list_head(l2arc_dev_list); } else { next = list_next(l2arc_dev_list, next); if (next == NULL) next = list_head(l2arc_dev_list); } /* if we have come back to the start, bail out */ if (first == NULL) first = next; else if (next == first) break; } while (vdev_is_dead(next->l2ad_vdev) || next->l2ad_rebuild || next->l2ad_trim_all); /* if we were unable to find any usable vdevs, return NULL */ if (vdev_is_dead(next->l2ad_vdev) || next->l2ad_rebuild || next->l2ad_trim_all) next = NULL; l2arc_dev_last = next; out: mutex_exit(&l2arc_dev_mtx); /* * Grab the config lock to prevent the 'next' device from being * removed while we are writing to it. */ if (next != NULL) spa_config_enter(next->l2ad_spa, SCL_L2ARC, next, RW_READER); mutex_exit(&spa_namespace_lock); return (next); } /* * Free buffers that were tagged for destruction. */ static void l2arc_do_free_on_write(void) { list_t *buflist; l2arc_data_free_t *df, *df_prev; mutex_enter(&l2arc_free_on_write_mtx); buflist = l2arc_free_on_write; for (df = list_tail(buflist); df; df = df_prev) { df_prev = list_prev(buflist, df); ASSERT3P(df->l2df_abd, !=, NULL); abd_free(df->l2df_abd); list_remove(buflist, df); kmem_free(df, sizeof (l2arc_data_free_t)); } mutex_exit(&l2arc_free_on_write_mtx); } /* * A write to a cache device has completed. Update all headers to allow * reads from these buffers to begin. */ static void l2arc_write_done(zio_t *zio) { l2arc_write_callback_t *cb; l2arc_lb_abd_buf_t *abd_buf; l2arc_lb_ptr_buf_t *lb_ptr_buf; l2arc_dev_t *dev; l2arc_dev_hdr_phys_t *l2dhdr; list_t *buflist; arc_buf_hdr_t *head, *hdr, *hdr_prev; kmutex_t *hash_lock; int64_t bytes_dropped = 0; cb = zio->io_private; ASSERT3P(cb, !=, NULL); dev = cb->l2wcb_dev; l2dhdr = dev->l2ad_dev_hdr; ASSERT3P(dev, !=, NULL); head = cb->l2wcb_head; ASSERT3P(head, !=, NULL); buflist = &dev->l2ad_buflist; ASSERT3P(buflist, !=, NULL); DTRACE_PROBE2(l2arc__iodone, zio_t *, zio, l2arc_write_callback_t *, cb); /* * All writes completed, or an error was hit. */ top: mutex_enter(&dev->l2ad_mtx); for (hdr = list_prev(buflist, head); hdr; hdr = hdr_prev) { hdr_prev = list_prev(buflist, hdr); hash_lock = HDR_LOCK(hdr); /* * We cannot use mutex_enter or else we can deadlock * with l2arc_write_buffers (due to swapping the order * the hash lock and l2ad_mtx are taken). */ if (!mutex_tryenter(hash_lock)) { /* * Missed the hash lock. We must retry so we * don't leave the ARC_FLAG_L2_WRITING bit set. */ ARCSTAT_BUMP(arcstat_l2_writes_lock_retry); /* * We don't want to rescan the headers we've * already marked as having been written out, so * we reinsert the head node so we can pick up * where we left off. */ list_remove(buflist, head); list_insert_after(buflist, hdr, head); mutex_exit(&dev->l2ad_mtx); /* * We wait for the hash lock to become available * to try and prevent busy waiting, and increase * the chance we'll be able to acquire the lock * the next time around. */ mutex_enter(hash_lock); mutex_exit(hash_lock); goto top; } /* * We could not have been moved into the arc_l2c_only * state while in-flight due to our ARC_FLAG_L2_WRITING * bit being set. Let's just ensure that's being enforced. */ ASSERT(HDR_HAS_L1HDR(hdr)); /* * Skipped - drop L2ARC entry and mark the header as no * longer L2 eligibile. */ if (zio->io_error != 0) { /* * Error - drop L2ARC entry. */ list_remove(buflist, hdr); arc_hdr_clear_flags(hdr, ARC_FLAG_HAS_L2HDR); uint64_t psize = HDR_GET_PSIZE(hdr); l2arc_hdr_arcstats_decrement(hdr); bytes_dropped += vdev_psize_to_asize(dev->l2ad_vdev, psize); (void) zfs_refcount_remove_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); } /* * Allow ARC to begin reads and ghost list evictions to * this L2ARC entry. */ arc_hdr_clear_flags(hdr, ARC_FLAG_L2_WRITING); mutex_exit(hash_lock); } /* * Free the allocated abd buffers for writing the log blocks. * If the zio failed reclaim the allocated space and remove the * pointers to these log blocks from the log block pointer list * of the L2ARC device. */ while ((abd_buf = list_remove_tail(&cb->l2wcb_abd_list)) != NULL) { abd_free(abd_buf->abd); zio_buf_free(abd_buf, sizeof (*abd_buf)); if (zio->io_error != 0) { lb_ptr_buf = list_remove_head(&dev->l2ad_lbptr_list); /* * L2BLK_GET_PSIZE returns aligned size for log * blocks. */ uint64_t asize = L2BLK_GET_PSIZE((lb_ptr_buf->lb_ptr)->lbp_prop); bytes_dropped += asize; ARCSTAT_INCR(arcstat_l2_log_blk_asize, -asize); ARCSTAT_BUMPDOWN(arcstat_l2_log_blk_count); zfs_refcount_remove_many(&dev->l2ad_lb_asize, asize, lb_ptr_buf); zfs_refcount_remove(&dev->l2ad_lb_count, lb_ptr_buf); kmem_free(lb_ptr_buf->lb_ptr, sizeof (l2arc_log_blkptr_t)); kmem_free(lb_ptr_buf, sizeof (l2arc_lb_ptr_buf_t)); } } list_destroy(&cb->l2wcb_abd_list); if (zio->io_error != 0) { ARCSTAT_BUMP(arcstat_l2_writes_error); /* * Restore the lbps array in the header to its previous state. * If the list of log block pointers is empty, zero out the * log block pointers in the device header. */ lb_ptr_buf = list_head(&dev->l2ad_lbptr_list); for (int i = 0; i < 2; i++) { if (lb_ptr_buf == NULL) { /* * If the list is empty zero out the device * header. Otherwise zero out the second log * block pointer in the header. */ if (i == 0) { bzero(l2dhdr, dev->l2ad_dev_hdr_asize); } else { bzero(&l2dhdr->dh_start_lbps[i], sizeof (l2arc_log_blkptr_t)); } break; } bcopy(lb_ptr_buf->lb_ptr, &l2dhdr->dh_start_lbps[i], sizeof (l2arc_log_blkptr_t)); lb_ptr_buf = list_next(&dev->l2ad_lbptr_list, lb_ptr_buf); } } - atomic_inc_64(&l2arc_writes_done); + ARCSTAT_BUMP(arcstat_l2_writes_done); list_remove(buflist, head); ASSERT(!HDR_HAS_L1HDR(head)); kmem_cache_free(hdr_l2only_cache, head); mutex_exit(&dev->l2ad_mtx); ASSERT(dev->l2ad_vdev != NULL); vdev_space_update(dev->l2ad_vdev, -bytes_dropped, 0, 0); l2arc_do_free_on_write(); kmem_free(cb, sizeof (l2arc_write_callback_t)); } static int l2arc_untransform(zio_t *zio, l2arc_read_callback_t *cb) { int ret; spa_t *spa = zio->io_spa; arc_buf_hdr_t *hdr = cb->l2rcb_hdr; blkptr_t *bp = zio->io_bp; uint8_t salt[ZIO_DATA_SALT_LEN]; uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t mac[ZIO_DATA_MAC_LEN]; boolean_t no_crypt = B_FALSE; /* * ZIL data is never be written to the L2ARC, so we don't need * special handling for its unique MAC storage. */ ASSERT3U(BP_GET_TYPE(bp), !=, DMU_OT_INTENT_LOG); ASSERT(MUTEX_HELD(HDR_LOCK(hdr))); ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); /* * If the data was encrypted, decrypt it now. Note that * we must check the bp here and not the hdr, since the * hdr does not have its encryption parameters updated * until arc_read_done(). */ if (BP_IS_ENCRYPTED(bp)) { abd_t *eabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr, B_TRUE); zio_crypt_decode_params_bp(bp, salt, iv); zio_crypt_decode_mac_bp(bp, mac); ret = spa_do_crypt_abd(B_FALSE, spa, &cb->l2rcb_zb, BP_GET_TYPE(bp), BP_GET_DEDUP(bp), BP_SHOULD_BYTESWAP(bp), salt, iv, mac, HDR_GET_PSIZE(hdr), eabd, hdr->b_l1hdr.b_pabd, &no_crypt); if (ret != 0) { arc_free_data_abd(hdr, eabd, arc_hdr_size(hdr), hdr); goto error; } /* * If we actually performed decryption, replace b_pabd * with the decrypted data. Otherwise we can just throw * our decryption buffer away. */ if (!no_crypt) { arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, arc_hdr_size(hdr), hdr); hdr->b_l1hdr.b_pabd = eabd; zio->io_abd = eabd; } else { arc_free_data_abd(hdr, eabd, arc_hdr_size(hdr), hdr); } } /* * If the L2ARC block was compressed, but ARC compression * is disabled we decompress the data into a new buffer and * replace the existing data. */ if (HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) { abd_t *cabd = arc_get_data_abd(hdr, arc_hdr_size(hdr), hdr, B_TRUE); void *tmp = abd_borrow_buf(cabd, arc_hdr_size(hdr)); ret = zio_decompress_data(HDR_GET_COMPRESS(hdr), hdr->b_l1hdr.b_pabd, tmp, HDR_GET_PSIZE(hdr), HDR_GET_LSIZE(hdr), &hdr->b_complevel); if (ret != 0) { abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr)); arc_free_data_abd(hdr, cabd, arc_hdr_size(hdr), hdr); goto error; } abd_return_buf_copy(cabd, tmp, arc_hdr_size(hdr)); arc_free_data_abd(hdr, hdr->b_l1hdr.b_pabd, arc_hdr_size(hdr), hdr); hdr->b_l1hdr.b_pabd = cabd; zio->io_abd = cabd; zio->io_size = HDR_GET_LSIZE(hdr); } return (0); error: return (ret); } /* * A read to a cache device completed. Validate buffer contents before * handing over to the regular ARC routines. */ static void l2arc_read_done(zio_t *zio) { int tfm_error = 0; l2arc_read_callback_t *cb = zio->io_private; arc_buf_hdr_t *hdr; kmutex_t *hash_lock; boolean_t valid_cksum; boolean_t using_rdata = (BP_IS_ENCRYPTED(&cb->l2rcb_bp) && (cb->l2rcb_flags & ZIO_FLAG_RAW_ENCRYPT)); ASSERT3P(zio->io_vd, !=, NULL); ASSERT(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE); spa_config_exit(zio->io_spa, SCL_L2ARC, zio->io_vd); ASSERT3P(cb, !=, NULL); hdr = cb->l2rcb_hdr; ASSERT3P(hdr, !=, NULL); hash_lock = HDR_LOCK(hdr); mutex_enter(hash_lock); ASSERT3P(hash_lock, ==, HDR_LOCK(hdr)); /* * If the data was read into a temporary buffer, * move it and free the buffer. */ if (cb->l2rcb_abd != NULL) { ASSERT3U(arc_hdr_size(hdr), <, zio->io_size); if (zio->io_error == 0) { if (using_rdata) { abd_copy(hdr->b_crypt_hdr.b_rabd, cb->l2rcb_abd, arc_hdr_size(hdr)); } else { abd_copy(hdr->b_l1hdr.b_pabd, cb->l2rcb_abd, arc_hdr_size(hdr)); } } /* * The following must be done regardless of whether * there was an error: * - free the temporary buffer * - point zio to the real ARC buffer * - set zio size accordingly * These are required because zio is either re-used for * an I/O of the block in the case of the error * or the zio is passed to arc_read_done() and it * needs real data. */ abd_free(cb->l2rcb_abd); zio->io_size = zio->io_orig_size = arc_hdr_size(hdr); if (using_rdata) { ASSERT(HDR_HAS_RABD(hdr)); zio->io_abd = zio->io_orig_abd = hdr->b_crypt_hdr.b_rabd; } else { ASSERT3P(hdr->b_l1hdr.b_pabd, !=, NULL); zio->io_abd = zio->io_orig_abd = hdr->b_l1hdr.b_pabd; } } ASSERT3P(zio->io_abd, !=, NULL); /* * Check this survived the L2ARC journey. */ ASSERT(zio->io_abd == hdr->b_l1hdr.b_pabd || (HDR_HAS_RABD(hdr) && zio->io_abd == hdr->b_crypt_hdr.b_rabd)); zio->io_bp_copy = cb->l2rcb_bp; /* XXX fix in L2ARC 2.0 */ zio->io_bp = &zio->io_bp_copy; /* XXX fix in L2ARC 2.0 */ zio->io_prop.zp_complevel = hdr->b_complevel; valid_cksum = arc_cksum_is_equal(hdr, zio); /* * b_rabd will always match the data as it exists on disk if it is * being used. Therefore if we are reading into b_rabd we do not * attempt to untransform the data. */ if (valid_cksum && !using_rdata) tfm_error = l2arc_untransform(zio, cb); if (valid_cksum && tfm_error == 0 && zio->io_error == 0 && !HDR_L2_EVICTED(hdr)) { mutex_exit(hash_lock); zio->io_private = hdr; arc_read_done(zio); } else { /* * Buffer didn't survive caching. Increment stats and * reissue to the original storage device. */ if (zio->io_error != 0) { ARCSTAT_BUMP(arcstat_l2_io_error); } else { zio->io_error = SET_ERROR(EIO); } if (!valid_cksum || tfm_error != 0) ARCSTAT_BUMP(arcstat_l2_cksum_bad); /* * If there's no waiter, issue an async i/o to the primary * storage now. If there *is* a waiter, the caller must * issue the i/o in a context where it's OK to block. */ if (zio->io_waiter == NULL) { zio_t *pio = zio_unique_parent(zio); void *abd = (using_rdata) ? hdr->b_crypt_hdr.b_rabd : hdr->b_l1hdr.b_pabd; ASSERT(!pio || pio->io_child_type == ZIO_CHILD_LOGICAL); zio = zio_read(pio, zio->io_spa, zio->io_bp, abd, zio->io_size, arc_read_done, hdr, zio->io_priority, cb->l2rcb_flags, &cb->l2rcb_zb); /* * Original ZIO will be freed, so we need to update * ARC header with the new ZIO pointer to be used * by zio_change_priority() in arc_read(). */ for (struct arc_callback *acb = hdr->b_l1hdr.b_acb; acb != NULL; acb = acb->acb_next) acb->acb_zio_head = zio; mutex_exit(hash_lock); zio_nowait(zio); } else { mutex_exit(hash_lock); } } kmem_free(cb, sizeof (l2arc_read_callback_t)); } /* * This is the list priority from which the L2ARC will search for pages to * cache. This is used within loops (0..3) to cycle through lists in the * desired order. This order can have a significant effect on cache * performance. * * Currently the metadata lists are hit first, MFU then MRU, followed by * the data lists. This function returns a locked list, and also returns * the lock pointer. */ static multilist_sublist_t * l2arc_sublist_lock(int list_num) { multilist_t *ml = NULL; unsigned int idx; ASSERT(list_num >= 0 && list_num < L2ARC_FEED_TYPES); switch (list_num) { case 0: ml = &arc_mfu->arcs_list[ARC_BUFC_METADATA]; break; case 1: ml = &arc_mru->arcs_list[ARC_BUFC_METADATA]; break; case 2: ml = &arc_mfu->arcs_list[ARC_BUFC_DATA]; break; case 3: ml = &arc_mru->arcs_list[ARC_BUFC_DATA]; break; default: return (NULL); } /* * Return a randomly-selected sublist. This is acceptable * because the caller feeds only a little bit of data for each * call (8MB). Subsequent calls will result in different * sublists being selected. */ idx = multilist_get_random_index(ml); return (multilist_sublist_lock(ml, idx)); } /* * Calculates the maximum overhead of L2ARC metadata log blocks for a given * L2ARC write size. l2arc_evict and l2arc_write_size need to include this * overhead in processing to make sure there is enough headroom available * when writing buffers. */ static inline uint64_t l2arc_log_blk_overhead(uint64_t write_sz, l2arc_dev_t *dev) { if (dev->l2ad_log_entries == 0) { return (0); } else { uint64_t log_entries = write_sz >> SPA_MINBLOCKSHIFT; uint64_t log_blocks = (log_entries + dev->l2ad_log_entries - 1) / dev->l2ad_log_entries; return (vdev_psize_to_asize(dev->l2ad_vdev, sizeof (l2arc_log_blk_phys_t)) * log_blocks); } } /* * Evict buffers from the device write hand to the distance specified in * bytes. This distance may span populated buffers, it may span nothing. * This is clearing a region on the L2ARC device ready for writing. * If the 'all' boolean is set, every buffer is evicted. */ static void l2arc_evict(l2arc_dev_t *dev, uint64_t distance, boolean_t all) { list_t *buflist; arc_buf_hdr_t *hdr, *hdr_prev; kmutex_t *hash_lock; uint64_t taddr; l2arc_lb_ptr_buf_t *lb_ptr_buf, *lb_ptr_buf_prev; vdev_t *vd = dev->l2ad_vdev; boolean_t rerun; buflist = &dev->l2ad_buflist; /* * We need to add in the worst case scenario of log block overhead. */ distance += l2arc_log_blk_overhead(distance, dev); if (vd->vdev_has_trim && l2arc_trim_ahead > 0) { /* * Trim ahead of the write size 64MB or (l2arc_trim_ahead/100) * times the write size, whichever is greater. */ distance += MAX(64 * 1024 * 1024, (distance * l2arc_trim_ahead) / 100); } top: rerun = B_FALSE; if (dev->l2ad_hand >= (dev->l2ad_end - distance)) { /* * When there is no space to accommodate upcoming writes, * evict to the end. Then bump the write and evict hands * to the start and iterate. This iteration does not * happen indefinitely as we make sure in * l2arc_write_size() that when the write hand is reset, * the write size does not exceed the end of the device. */ rerun = B_TRUE; taddr = dev->l2ad_end; } else { taddr = dev->l2ad_hand + distance; } DTRACE_PROBE4(l2arc__evict, l2arc_dev_t *, dev, list_t *, buflist, uint64_t, taddr, boolean_t, all); if (!all) { /* * This check has to be placed after deciding whether to * iterate (rerun). */ if (dev->l2ad_first) { /* * This is the first sweep through the device. There is * nothing to evict. We have already trimmmed the * whole device. */ goto out; } else { /* * Trim the space to be evicted. */ if (vd->vdev_has_trim && dev->l2ad_evict < taddr && l2arc_trim_ahead > 0) { /* * We have to drop the spa_config lock because * vdev_trim_range() will acquire it. * l2ad_evict already accounts for the label * size. To prevent vdev_trim_ranges() from * adding it again, we subtract it from * l2ad_evict. */ spa_config_exit(dev->l2ad_spa, SCL_L2ARC, dev); vdev_trim_simple(vd, dev->l2ad_evict - VDEV_LABEL_START_SIZE, taddr - dev->l2ad_evict); spa_config_enter(dev->l2ad_spa, SCL_L2ARC, dev, RW_READER); } /* * When rebuilding L2ARC we retrieve the evict hand * from the header of the device. Of note, l2arc_evict() * does not actually delete buffers from the cache * device, but trimming may do so depending on the * hardware implementation. Thus keeping track of the * evict hand is useful. */ dev->l2ad_evict = MAX(dev->l2ad_evict, taddr); } } retry: mutex_enter(&dev->l2ad_mtx); /* * We have to account for evicted log blocks. Run vdev_space_update() * on log blocks whose offset (in bytes) is before the evicted offset * (in bytes) by searching in the list of pointers to log blocks * present in the L2ARC device. */ for (lb_ptr_buf = list_tail(&dev->l2ad_lbptr_list); lb_ptr_buf; lb_ptr_buf = lb_ptr_buf_prev) { lb_ptr_buf_prev = list_prev(&dev->l2ad_lbptr_list, lb_ptr_buf); /* L2BLK_GET_PSIZE returns aligned size for log blocks */ uint64_t asize = L2BLK_GET_PSIZE( (lb_ptr_buf->lb_ptr)->lbp_prop); /* * We don't worry about log blocks left behind (ie * lbp_payload_start < l2ad_hand) because l2arc_write_buffers() * will never write more than l2arc_evict() evicts. */ if (!all && l2arc_log_blkptr_valid(dev, lb_ptr_buf->lb_ptr)) { break; } else { vdev_space_update(vd, -asize, 0, 0); ARCSTAT_INCR(arcstat_l2_log_blk_asize, -asize); ARCSTAT_BUMPDOWN(arcstat_l2_log_blk_count); zfs_refcount_remove_many(&dev->l2ad_lb_asize, asize, lb_ptr_buf); zfs_refcount_remove(&dev->l2ad_lb_count, lb_ptr_buf); list_remove(&dev->l2ad_lbptr_list, lb_ptr_buf); kmem_free(lb_ptr_buf->lb_ptr, sizeof (l2arc_log_blkptr_t)); kmem_free(lb_ptr_buf, sizeof (l2arc_lb_ptr_buf_t)); } } for (hdr = list_tail(buflist); hdr; hdr = hdr_prev) { hdr_prev = list_prev(buflist, hdr); ASSERT(!HDR_EMPTY(hdr)); hash_lock = HDR_LOCK(hdr); /* * We cannot use mutex_enter or else we can deadlock * with l2arc_write_buffers (due to swapping the order * the hash lock and l2ad_mtx are taken). */ if (!mutex_tryenter(hash_lock)) { /* * Missed the hash lock. Retry. */ ARCSTAT_BUMP(arcstat_l2_evict_lock_retry); mutex_exit(&dev->l2ad_mtx); mutex_enter(hash_lock); mutex_exit(hash_lock); goto retry; } /* * A header can't be on this list if it doesn't have L2 header. */ ASSERT(HDR_HAS_L2HDR(hdr)); /* Ensure this header has finished being written. */ ASSERT(!HDR_L2_WRITING(hdr)); ASSERT(!HDR_L2_WRITE_HEAD(hdr)); if (!all && (hdr->b_l2hdr.b_daddr >= dev->l2ad_evict || hdr->b_l2hdr.b_daddr < dev->l2ad_hand)) { /* * We've evicted to the target address, * or the end of the device. */ mutex_exit(hash_lock); break; } if (!HDR_HAS_L1HDR(hdr)) { ASSERT(!HDR_L2_READING(hdr)); /* * This doesn't exist in the ARC. Destroy. * arc_hdr_destroy() will call list_remove() * and decrement arcstat_l2_lsize. */ arc_change_state(arc_anon, hdr, hash_lock); arc_hdr_destroy(hdr); } else { ASSERT(hdr->b_l1hdr.b_state != arc_l2c_only); ARCSTAT_BUMP(arcstat_l2_evict_l1cached); /* * Invalidate issued or about to be issued * reads, since we may be about to write * over this location. */ if (HDR_L2_READING(hdr)) { ARCSTAT_BUMP(arcstat_l2_evict_reading); arc_hdr_set_flags(hdr, ARC_FLAG_L2_EVICTED); } arc_hdr_l2hdr_destroy(hdr); } mutex_exit(hash_lock); } mutex_exit(&dev->l2ad_mtx); out: /* * We need to check if we evict all buffers, otherwise we may iterate * unnecessarily. */ if (!all && rerun) { /* * Bump device hand to the device start if it is approaching the * end. l2arc_evict() has already evicted ahead for this case. */ dev->l2ad_hand = dev->l2ad_start; dev->l2ad_evict = dev->l2ad_start; dev->l2ad_first = B_FALSE; goto top; } if (!all) { /* * In case of cache device removal (all) the following * assertions may be violated without functional consequences * as the device is about to be removed. */ ASSERT3U(dev->l2ad_hand + distance, <, dev->l2ad_end); if (!dev->l2ad_first) ASSERT3U(dev->l2ad_hand, <, dev->l2ad_evict); } } /* * Handle any abd transforms that might be required for writing to the L2ARC. * If successful, this function will always return an abd with the data * transformed as it is on disk in a new abd of asize bytes. */ static int l2arc_apply_transforms(spa_t *spa, arc_buf_hdr_t *hdr, uint64_t asize, abd_t **abd_out) { int ret; void *tmp = NULL; abd_t *cabd = NULL, *eabd = NULL, *to_write = hdr->b_l1hdr.b_pabd; enum zio_compress compress = HDR_GET_COMPRESS(hdr); uint64_t psize = HDR_GET_PSIZE(hdr); uint64_t size = arc_hdr_size(hdr); boolean_t ismd = HDR_ISTYPE_METADATA(hdr); boolean_t bswap = (hdr->b_l1hdr.b_byteswap != DMU_BSWAP_NUMFUNCS); dsl_crypto_key_t *dck = NULL; uint8_t mac[ZIO_DATA_MAC_LEN] = { 0 }; boolean_t no_crypt = B_FALSE; ASSERT((HDR_GET_COMPRESS(hdr) != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) || HDR_ENCRYPTED(hdr) || HDR_SHARED_DATA(hdr) || psize != asize); ASSERT3U(psize, <=, asize); /* * If this data simply needs its own buffer, we simply allocate it * and copy the data. This may be done to eliminate a dependency on a * shared buffer or to reallocate the buffer to match asize. */ if (HDR_HAS_RABD(hdr) && asize != psize) { ASSERT3U(asize, >=, psize); to_write = abd_alloc_for_io(asize, ismd); abd_copy(to_write, hdr->b_crypt_hdr.b_rabd, psize); if (psize != asize) abd_zero_off(to_write, psize, asize - psize); goto out; } if ((compress == ZIO_COMPRESS_OFF || HDR_COMPRESSION_ENABLED(hdr)) && !HDR_ENCRYPTED(hdr)) { ASSERT3U(size, ==, psize); to_write = abd_alloc_for_io(asize, ismd); abd_copy(to_write, hdr->b_l1hdr.b_pabd, size); if (size != asize) abd_zero_off(to_write, size, asize - size); goto out; } if (compress != ZIO_COMPRESS_OFF && !HDR_COMPRESSION_ENABLED(hdr)) { cabd = abd_alloc_for_io(asize, ismd); tmp = abd_borrow_buf(cabd, asize); psize = zio_compress_data(compress, to_write, tmp, size, hdr->b_complevel); if (psize >= size) { abd_return_buf(cabd, tmp, asize); HDR_SET_COMPRESS(hdr, ZIO_COMPRESS_OFF); to_write = cabd; abd_copy(to_write, hdr->b_l1hdr.b_pabd, size); if (size != asize) abd_zero_off(to_write, size, asize - size); goto encrypt; } ASSERT3U(psize, <=, HDR_GET_PSIZE(hdr)); if (psize < asize) bzero((char *)tmp + psize, asize - psize); psize = HDR_GET_PSIZE(hdr); abd_return_buf_copy(cabd, tmp, asize); to_write = cabd; } encrypt: if (HDR_ENCRYPTED(hdr)) { eabd = abd_alloc_for_io(asize, ismd); /* * If the dataset was disowned before the buffer * made it to this point, the key to re-encrypt * it won't be available. In this case we simply * won't write the buffer to the L2ARC. */ ret = spa_keystore_lookup_key(spa, hdr->b_crypt_hdr.b_dsobj, FTAG, &dck); if (ret != 0) goto error; ret = zio_do_crypt_abd(B_TRUE, &dck->dck_key, hdr->b_crypt_hdr.b_ot, bswap, hdr->b_crypt_hdr.b_salt, hdr->b_crypt_hdr.b_iv, mac, psize, to_write, eabd, &no_crypt); if (ret != 0) goto error; if (no_crypt) abd_copy(eabd, to_write, psize); if (psize != asize) abd_zero_off(eabd, psize, asize - psize); /* assert that the MAC we got here matches the one we saved */ ASSERT0(bcmp(mac, hdr->b_crypt_hdr.b_mac, ZIO_DATA_MAC_LEN)); spa_keystore_dsl_key_rele(spa, dck, FTAG); if (to_write == cabd) abd_free(cabd); to_write = eabd; } out: ASSERT3P(to_write, !=, hdr->b_l1hdr.b_pabd); *abd_out = to_write; return (0); error: if (dck != NULL) spa_keystore_dsl_key_rele(spa, dck, FTAG); if (cabd != NULL) abd_free(cabd); if (eabd != NULL) abd_free(eabd); *abd_out = NULL; return (ret); } static void l2arc_blk_fetch_done(zio_t *zio) { l2arc_read_callback_t *cb; cb = zio->io_private; if (cb->l2rcb_abd != NULL) abd_free(cb->l2rcb_abd); kmem_free(cb, sizeof (l2arc_read_callback_t)); } /* * Find and write ARC buffers to the L2ARC device. * * An ARC_FLAG_L2_WRITING flag is set so that the L2ARC buffers are not valid * for reading until they have completed writing. * The headroom_boost is an in-out parameter used to maintain headroom boost * state between calls to this function. * * Returns the number of bytes actually written (which may be smaller than * the delta by which the device hand has changed due to alignment and the * writing of log blocks). */ static uint64_t l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz) { arc_buf_hdr_t *hdr, *hdr_prev, *head; uint64_t write_asize, write_psize, write_lsize, headroom; boolean_t full; l2arc_write_callback_t *cb = NULL; zio_t *pio, *wzio; uint64_t guid = spa_load_guid(spa); l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr; ASSERT3P(dev->l2ad_vdev, !=, NULL); pio = NULL; write_lsize = write_asize = write_psize = 0; full = B_FALSE; head = kmem_cache_alloc(hdr_l2only_cache, KM_PUSHPAGE); arc_hdr_set_flags(head, ARC_FLAG_L2_WRITE_HEAD | ARC_FLAG_HAS_L2HDR); /* * Copy buffers for L2ARC writing. */ for (int pass = 0; pass < L2ARC_FEED_TYPES; pass++) { /* * If pass == 1 or 3, we cache MRU metadata and data * respectively. */ if (l2arc_mfuonly) { if (pass == 1 || pass == 3) continue; } multilist_sublist_t *mls = l2arc_sublist_lock(pass); uint64_t passed_sz = 0; VERIFY3P(mls, !=, NULL); /* * L2ARC fast warmup. * * Until the ARC is warm and starts to evict, read from the * head of the ARC lists rather than the tail. */ if (arc_warm == B_FALSE) hdr = multilist_sublist_head(mls); else hdr = multilist_sublist_tail(mls); headroom = target_sz * l2arc_headroom; if (zfs_compressed_arc_enabled) headroom = (headroom * l2arc_headroom_boost) / 100; for (; hdr; hdr = hdr_prev) { kmutex_t *hash_lock; abd_t *to_write = NULL; if (arc_warm == B_FALSE) hdr_prev = multilist_sublist_next(mls, hdr); else hdr_prev = multilist_sublist_prev(mls, hdr); hash_lock = HDR_LOCK(hdr); if (!mutex_tryenter(hash_lock)) { /* * Skip this buffer rather than waiting. */ continue; } passed_sz += HDR_GET_LSIZE(hdr); if (l2arc_headroom != 0 && passed_sz > headroom) { /* * Searched too far. */ mutex_exit(hash_lock); break; } if (!l2arc_write_eligible(guid, hdr)) { mutex_exit(hash_lock); continue; } /* * We rely on the L1 portion of the header below, so * it's invalid for this header to have been evicted out * of the ghost cache, prior to being written out. The * ARC_FLAG_L2_WRITING bit ensures this won't happen. */ ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3U(HDR_GET_PSIZE(hdr), >, 0); ASSERT3U(arc_hdr_size(hdr), >, 0); ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr)); uint64_t psize = HDR_GET_PSIZE(hdr); uint64_t asize = vdev_psize_to_asize(dev->l2ad_vdev, psize); if ((write_asize + asize) > target_sz) { full = B_TRUE; mutex_exit(hash_lock); break; } /* * We rely on the L1 portion of the header below, so * it's invalid for this header to have been evicted out * of the ghost cache, prior to being written out. The * ARC_FLAG_L2_WRITING bit ensures this won't happen. */ arc_hdr_set_flags(hdr, ARC_FLAG_L2_WRITING); ASSERT(HDR_HAS_L1HDR(hdr)); ASSERT3U(HDR_GET_PSIZE(hdr), >, 0); ASSERT(hdr->b_l1hdr.b_pabd != NULL || HDR_HAS_RABD(hdr)); ASSERT3U(arc_hdr_size(hdr), >, 0); /* * If this header has b_rabd, we can use this since it * must always match the data exactly as it exists on * disk. Otherwise, the L2ARC can normally use the * hdr's data, but if we're sharing data between the * hdr and one of its bufs, L2ARC needs its own copy of * the data so that the ZIO below can't race with the * buf consumer. To ensure that this copy will be * available for the lifetime of the ZIO and be cleaned * up afterwards, we add it to the l2arc_free_on_write * queue. If we need to apply any transforms to the * data (compression, encryption) we will also need the * extra buffer. */ if (HDR_HAS_RABD(hdr) && psize == asize) { to_write = hdr->b_crypt_hdr.b_rabd; } else if ((HDR_COMPRESSION_ENABLED(hdr) || HDR_GET_COMPRESS(hdr) == ZIO_COMPRESS_OFF) && !HDR_ENCRYPTED(hdr) && !HDR_SHARED_DATA(hdr) && psize == asize) { to_write = hdr->b_l1hdr.b_pabd; } else { int ret; arc_buf_contents_t type = arc_buf_type(hdr); ret = l2arc_apply_transforms(spa, hdr, asize, &to_write); if (ret != 0) { arc_hdr_clear_flags(hdr, ARC_FLAG_L2_WRITING); mutex_exit(hash_lock); continue; } l2arc_free_abd_on_write(to_write, asize, type); } if (pio == NULL) { /* * Insert a dummy header on the buflist so * l2arc_write_done() can find where the * write buffers begin without searching. */ mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_buflist, head); mutex_exit(&dev->l2ad_mtx); cb = kmem_alloc( sizeof (l2arc_write_callback_t), KM_SLEEP); cb->l2wcb_dev = dev; cb->l2wcb_head = head; /* * Create a list to save allocated abd buffers * for l2arc_log_blk_commit(). */ list_create(&cb->l2wcb_abd_list, sizeof (l2arc_lb_abd_buf_t), offsetof(l2arc_lb_abd_buf_t, node)); pio = zio_root(spa, l2arc_write_done, cb, ZIO_FLAG_CANFAIL); } hdr->b_l2hdr.b_dev = dev; hdr->b_l2hdr.b_hits = 0; hdr->b_l2hdr.b_daddr = dev->l2ad_hand; hdr->b_l2hdr.b_arcs_state = hdr->b_l1hdr.b_state->arcs_state; arc_hdr_set_flags(hdr, ARC_FLAG_HAS_L2HDR); mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_buflist, hdr); mutex_exit(&dev->l2ad_mtx); (void) zfs_refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); wzio = zio_write_phys(pio, dev->l2ad_vdev, hdr->b_l2hdr.b_daddr, asize, to_write, ZIO_CHECKSUM_OFF, NULL, hdr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE); write_lsize += HDR_GET_LSIZE(hdr); DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, zio_t *, wzio); write_psize += psize; write_asize += asize; dev->l2ad_hand += asize; l2arc_hdr_arcstats_increment(hdr); vdev_space_update(dev->l2ad_vdev, asize, 0, 0); mutex_exit(hash_lock); /* * Append buf info to current log and commit if full. * arcstat_l2_{size,asize} kstats are updated * internally. */ if (l2arc_log_blk_insert(dev, hdr)) l2arc_log_blk_commit(dev, pio, cb); zio_nowait(wzio); } multilist_sublist_unlock(mls); if (full == B_TRUE) break; } /* No buffers selected for writing? */ if (pio == NULL) { ASSERT0(write_lsize); ASSERT(!HDR_HAS_L1HDR(head)); kmem_cache_free(hdr_l2only_cache, head); /* * Although we did not write any buffers l2ad_evict may * have advanced. */ if (dev->l2ad_evict != l2dhdr->dh_evict) l2arc_dev_hdr_update(dev); return (0); } if (!dev->l2ad_first) ASSERT3U(dev->l2ad_hand, <=, dev->l2ad_evict); ASSERT3U(write_asize, <=, target_sz); ARCSTAT_BUMP(arcstat_l2_writes_sent); ARCSTAT_INCR(arcstat_l2_write_bytes, write_psize); dev->l2ad_writing = B_TRUE; (void) zio_wait(pio); dev->l2ad_writing = B_FALSE; /* * Update the device header after the zio completes as * l2arc_write_done() may have updated the memory holding the log block * pointers in the device header. */ l2arc_dev_hdr_update(dev); return (write_asize); } static boolean_t l2arc_hdr_limit_reached(void) { - int64_t s = aggsum_upper_bound(&astat_l2_hdr_size); + int64_t s = aggsum_upper_bound(&arc_sums.arcstat_l2_hdr_size); return (arc_reclaim_needed() || (s > arc_meta_limit * 3 / 4) || (s > (arc_warm ? arc_c : arc_c_max) * l2arc_meta_percent / 100)); } /* * This thread feeds the L2ARC at regular intervals. This is the beating * heart of the L2ARC. */ /* ARGSUSED */ static void l2arc_feed_thread(void *unused) { callb_cpr_t cpr; l2arc_dev_t *dev; spa_t *spa; uint64_t size, wrote; clock_t begin, next = ddi_get_lbolt(); fstrans_cookie_t cookie; CALLB_CPR_INIT(&cpr, &l2arc_feed_thr_lock, callb_generic_cpr, FTAG); mutex_enter(&l2arc_feed_thr_lock); cookie = spl_fstrans_mark(); while (l2arc_thread_exit == 0) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait_idle(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock, next); CALLB_CPR_SAFE_END(&cpr, &l2arc_feed_thr_lock); next = ddi_get_lbolt() + hz; /* * Quick check for L2ARC devices. */ mutex_enter(&l2arc_dev_mtx); if (l2arc_ndev == 0) { mutex_exit(&l2arc_dev_mtx); continue; } mutex_exit(&l2arc_dev_mtx); begin = ddi_get_lbolt(); /* * This selects the next l2arc device to write to, and in * doing so the next spa to feed from: dev->l2ad_spa. This * will return NULL if there are now no l2arc devices or if * they are all faulted. * * If a device is returned, its spa's config lock is also * held to prevent device removal. l2arc_dev_get_next() * will grab and release l2arc_dev_mtx. */ if ((dev = l2arc_dev_get_next()) == NULL) continue; spa = dev->l2ad_spa; ASSERT3P(spa, !=, NULL); /* * If the pool is read-only then force the feed thread to * sleep a little longer. */ if (!spa_writeable(spa)) { next = ddi_get_lbolt() + 5 * l2arc_feed_secs * hz; spa_config_exit(spa, SCL_L2ARC, dev); continue; } /* * Avoid contributing to memory pressure. */ if (l2arc_hdr_limit_reached()) { ARCSTAT_BUMP(arcstat_l2_abort_lowmem); spa_config_exit(spa, SCL_L2ARC, dev); continue; } ARCSTAT_BUMP(arcstat_l2_feeds); size = l2arc_write_size(dev); /* * Evict L2ARC buffers that will be overwritten. */ l2arc_evict(dev, size, B_FALSE); /* * Write ARC buffers. */ wrote = l2arc_write_buffers(spa, dev, size); /* * Calculate interval between writes. */ next = l2arc_write_interval(begin, size, wrote); spa_config_exit(spa, SCL_L2ARC, dev); } spl_fstrans_unmark(cookie); l2arc_thread_exit = 0; cv_broadcast(&l2arc_feed_thr_cv); CALLB_CPR_EXIT(&cpr); /* drops l2arc_feed_thr_lock */ thread_exit(); } boolean_t l2arc_vdev_present(vdev_t *vd) { return (l2arc_vdev_get(vd) != NULL); } /* * Returns the l2arc_dev_t associated with a particular vdev_t or NULL if * the vdev_t isn't an L2ARC device. */ l2arc_dev_t * l2arc_vdev_get(vdev_t *vd) { l2arc_dev_t *dev; mutex_enter(&l2arc_dev_mtx); for (dev = list_head(l2arc_dev_list); dev != NULL; dev = list_next(l2arc_dev_list, dev)) { if (dev->l2ad_vdev == vd) break; } mutex_exit(&l2arc_dev_mtx); return (dev); } /* * Add a vdev for use by the L2ARC. By this point the spa has already * validated the vdev and opened it. */ void l2arc_add_vdev(spa_t *spa, vdev_t *vd) { l2arc_dev_t *adddev; uint64_t l2dhdr_asize; ASSERT(!l2arc_vdev_present(vd)); /* * Create a new l2arc device entry. */ adddev = vmem_zalloc(sizeof (l2arc_dev_t), KM_SLEEP); adddev->l2ad_spa = spa; adddev->l2ad_vdev = vd; /* leave extra size for an l2arc device header */ l2dhdr_asize = adddev->l2ad_dev_hdr_asize = MAX(sizeof (*adddev->l2ad_dev_hdr), 1 << vd->vdev_ashift); adddev->l2ad_start = VDEV_LABEL_START_SIZE + l2dhdr_asize; adddev->l2ad_end = VDEV_LABEL_START_SIZE + vdev_get_min_asize(vd); ASSERT3U(adddev->l2ad_start, <, adddev->l2ad_end); adddev->l2ad_hand = adddev->l2ad_start; adddev->l2ad_evict = adddev->l2ad_start; adddev->l2ad_first = B_TRUE; adddev->l2ad_writing = B_FALSE; adddev->l2ad_trim_all = B_FALSE; list_link_init(&adddev->l2ad_node); adddev->l2ad_dev_hdr = kmem_zalloc(l2dhdr_asize, KM_SLEEP); mutex_init(&adddev->l2ad_mtx, NULL, MUTEX_DEFAULT, NULL); /* * This is a list of all ARC buffers that are still valid on the * device. */ list_create(&adddev->l2ad_buflist, sizeof (arc_buf_hdr_t), offsetof(arc_buf_hdr_t, b_l2hdr.b_l2node)); /* * This is a list of pointers to log blocks that are still present * on the device. */ list_create(&adddev->l2ad_lbptr_list, sizeof (l2arc_lb_ptr_buf_t), offsetof(l2arc_lb_ptr_buf_t, node)); vdev_space_update(vd, 0, 0, adddev->l2ad_end - adddev->l2ad_hand); zfs_refcount_create(&adddev->l2ad_alloc); zfs_refcount_create(&adddev->l2ad_lb_asize); zfs_refcount_create(&adddev->l2ad_lb_count); /* * Add device to global list */ mutex_enter(&l2arc_dev_mtx); list_insert_head(l2arc_dev_list, adddev); atomic_inc_64(&l2arc_ndev); mutex_exit(&l2arc_dev_mtx); /* * Decide if vdev is eligible for L2ARC rebuild */ l2arc_rebuild_vdev(adddev->l2ad_vdev, B_FALSE); } void l2arc_rebuild_vdev(vdev_t *vd, boolean_t reopen) { l2arc_dev_t *dev = NULL; l2arc_dev_hdr_phys_t *l2dhdr; uint64_t l2dhdr_asize; spa_t *spa; dev = l2arc_vdev_get(vd); ASSERT3P(dev, !=, NULL); spa = dev->l2ad_spa; l2dhdr = dev->l2ad_dev_hdr; l2dhdr_asize = dev->l2ad_dev_hdr_asize; /* * The L2ARC has to hold at least the payload of one log block for * them to be restored (persistent L2ARC). The payload of a log block * depends on the amount of its log entries. We always write log blocks * with 1022 entries. How many of them are committed or restored depends * on the size of the L2ARC device. Thus the maximum payload of * one log block is 1022 * SPA_MAXBLOCKSIZE = 16GB. If the L2ARC device * is less than that, we reduce the amount of committed and restored * log entries per block so as to enable persistence. */ if (dev->l2ad_end < l2arc_rebuild_blocks_min_l2size) { dev->l2ad_log_entries = 0; } else { dev->l2ad_log_entries = MIN((dev->l2ad_end - dev->l2ad_start) >> SPA_MAXBLOCKSHIFT, L2ARC_LOG_BLK_MAX_ENTRIES); } /* * Read the device header, if an error is returned do not rebuild L2ARC. */ if (l2arc_dev_hdr_read(dev) == 0 && dev->l2ad_log_entries > 0) { /* * If we are onlining a cache device (vdev_reopen) that was * still present (l2arc_vdev_present()) and rebuild is enabled, * we should evict all ARC buffers and pointers to log blocks * and reclaim their space before restoring its contents to * L2ARC. */ if (reopen) { if (!l2arc_rebuild_enabled) { return; } else { l2arc_evict(dev, 0, B_TRUE); /* start a new log block */ dev->l2ad_log_ent_idx = 0; dev->l2ad_log_blk_payload_asize = 0; dev->l2ad_log_blk_payload_start = 0; } } /* * Just mark the device as pending for a rebuild. We won't * be starting a rebuild in line here as it would block pool * import. Instead spa_load_impl will hand that off to an * async task which will call l2arc_spa_rebuild_start. */ dev->l2ad_rebuild = B_TRUE; } else if (spa_writeable(spa)) { /* * In this case TRIM the whole device if l2arc_trim_ahead > 0, * otherwise create a new header. We zero out the memory holding * the header to reset dh_start_lbps. If we TRIM the whole * device the new header will be written by * vdev_trim_l2arc_thread() at the end of the TRIM to update the * trim_state in the header too. When reading the header, if * trim_state is not VDEV_TRIM_COMPLETE and l2arc_trim_ahead > 0 * we opt to TRIM the whole device again. */ if (l2arc_trim_ahead > 0) { dev->l2ad_trim_all = B_TRUE; } else { bzero(l2dhdr, l2dhdr_asize); l2arc_dev_hdr_update(dev); } } } /* * Remove a vdev from the L2ARC. */ void l2arc_remove_vdev(vdev_t *vd) { l2arc_dev_t *remdev = NULL; /* * Find the device by vdev */ remdev = l2arc_vdev_get(vd); ASSERT3P(remdev, !=, NULL); /* * Cancel any ongoing or scheduled rebuild. */ mutex_enter(&l2arc_rebuild_thr_lock); if (remdev->l2ad_rebuild_began == B_TRUE) { remdev->l2ad_rebuild_cancel = B_TRUE; while (remdev->l2ad_rebuild == B_TRUE) cv_wait(&l2arc_rebuild_thr_cv, &l2arc_rebuild_thr_lock); } mutex_exit(&l2arc_rebuild_thr_lock); /* * Remove device from global list */ mutex_enter(&l2arc_dev_mtx); list_remove(l2arc_dev_list, remdev); l2arc_dev_last = NULL; /* may have been invalidated */ atomic_dec_64(&l2arc_ndev); mutex_exit(&l2arc_dev_mtx); /* * Clear all buflists and ARC references. L2ARC device flush. */ l2arc_evict(remdev, 0, B_TRUE); list_destroy(&remdev->l2ad_buflist); ASSERT(list_is_empty(&remdev->l2ad_lbptr_list)); list_destroy(&remdev->l2ad_lbptr_list); mutex_destroy(&remdev->l2ad_mtx); zfs_refcount_destroy(&remdev->l2ad_alloc); zfs_refcount_destroy(&remdev->l2ad_lb_asize); zfs_refcount_destroy(&remdev->l2ad_lb_count); kmem_free(remdev->l2ad_dev_hdr, remdev->l2ad_dev_hdr_asize); vmem_free(remdev, sizeof (l2arc_dev_t)); } void l2arc_init(void) { l2arc_thread_exit = 0; l2arc_ndev = 0; - l2arc_writes_sent = 0; - l2arc_writes_done = 0; mutex_init(&l2arc_feed_thr_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&l2arc_feed_thr_cv, NULL, CV_DEFAULT, NULL); mutex_init(&l2arc_rebuild_thr_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&l2arc_rebuild_thr_cv, NULL, CV_DEFAULT, NULL); mutex_init(&l2arc_dev_mtx, NULL, MUTEX_DEFAULT, NULL); mutex_init(&l2arc_free_on_write_mtx, NULL, MUTEX_DEFAULT, NULL); l2arc_dev_list = &L2ARC_dev_list; l2arc_free_on_write = &L2ARC_free_on_write; list_create(l2arc_dev_list, sizeof (l2arc_dev_t), offsetof(l2arc_dev_t, l2ad_node)); list_create(l2arc_free_on_write, sizeof (l2arc_data_free_t), offsetof(l2arc_data_free_t, l2df_list_node)); } void l2arc_fini(void) { mutex_destroy(&l2arc_feed_thr_lock); cv_destroy(&l2arc_feed_thr_cv); mutex_destroy(&l2arc_rebuild_thr_lock); cv_destroy(&l2arc_rebuild_thr_cv); mutex_destroy(&l2arc_dev_mtx); mutex_destroy(&l2arc_free_on_write_mtx); list_destroy(l2arc_dev_list); list_destroy(l2arc_free_on_write); } void l2arc_start(void) { if (!(spa_mode_global & SPA_MODE_WRITE)) return; (void) thread_create(NULL, 0, l2arc_feed_thread, NULL, 0, &p0, TS_RUN, defclsyspri); } void l2arc_stop(void) { if (!(spa_mode_global & SPA_MODE_WRITE)) return; mutex_enter(&l2arc_feed_thr_lock); cv_signal(&l2arc_feed_thr_cv); /* kick thread out of startup */ l2arc_thread_exit = 1; while (l2arc_thread_exit != 0) cv_wait(&l2arc_feed_thr_cv, &l2arc_feed_thr_lock); mutex_exit(&l2arc_feed_thr_lock); } /* * Punches out rebuild threads for the L2ARC devices in a spa. This should * be called after pool import from the spa async thread, since starting * these threads directly from spa_import() will make them part of the * "zpool import" context and delay process exit (and thus pool import). */ void l2arc_spa_rebuild_start(spa_t *spa) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); /* * Locate the spa's l2arc devices and kick off rebuild threads. */ for (int i = 0; i < spa->spa_l2cache.sav_count; i++) { l2arc_dev_t *dev = l2arc_vdev_get(spa->spa_l2cache.sav_vdevs[i]); if (dev == NULL) { /* Don't attempt a rebuild if the vdev is UNAVAIL */ continue; } mutex_enter(&l2arc_rebuild_thr_lock); if (dev->l2ad_rebuild && !dev->l2ad_rebuild_cancel) { dev->l2ad_rebuild_began = B_TRUE; (void) thread_create(NULL, 0, l2arc_dev_rebuild_thread, dev, 0, &p0, TS_RUN, minclsyspri); } mutex_exit(&l2arc_rebuild_thr_lock); } } /* * Main entry point for L2ARC rebuilding. */ static void l2arc_dev_rebuild_thread(void *arg) { l2arc_dev_t *dev = arg; VERIFY(!dev->l2ad_rebuild_cancel); VERIFY(dev->l2ad_rebuild); (void) l2arc_rebuild(dev); mutex_enter(&l2arc_rebuild_thr_lock); dev->l2ad_rebuild_began = B_FALSE; dev->l2ad_rebuild = B_FALSE; mutex_exit(&l2arc_rebuild_thr_lock); thread_exit(); } /* * This function implements the actual L2ARC metadata rebuild. It: * starts reading the log block chain and restores each block's contents * to memory (reconstructing arc_buf_hdr_t's). * * Operation stops under any of the following conditions: * * 1) We reach the end of the log block chain. * 2) We encounter *any* error condition (cksum errors, io errors) */ static int l2arc_rebuild(l2arc_dev_t *dev) { vdev_t *vd = dev->l2ad_vdev; spa_t *spa = vd->vdev_spa; int err = 0; l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr; l2arc_log_blk_phys_t *this_lb, *next_lb; zio_t *this_io = NULL, *next_io = NULL; l2arc_log_blkptr_t lbps[2]; l2arc_lb_ptr_buf_t *lb_ptr_buf; boolean_t lock_held; this_lb = vmem_zalloc(sizeof (*this_lb), KM_SLEEP); next_lb = vmem_zalloc(sizeof (*next_lb), KM_SLEEP); /* * We prevent device removal while issuing reads to the device, * then during the rebuilding phases we drop this lock again so * that a spa_unload or device remove can be initiated - this is * safe, because the spa will signal us to stop before removing * our device and wait for us to stop. */ spa_config_enter(spa, SCL_L2ARC, vd, RW_READER); lock_held = B_TRUE; /* * Retrieve the persistent L2ARC device state. * L2BLK_GET_PSIZE returns aligned size for log blocks. */ dev->l2ad_evict = MAX(l2dhdr->dh_evict, dev->l2ad_start); dev->l2ad_hand = MAX(l2dhdr->dh_start_lbps[0].lbp_daddr + L2BLK_GET_PSIZE((&l2dhdr->dh_start_lbps[0])->lbp_prop), dev->l2ad_start); dev->l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST); vd->vdev_trim_action_time = l2dhdr->dh_trim_action_time; vd->vdev_trim_state = l2dhdr->dh_trim_state; /* * In case the zfs module parameter l2arc_rebuild_enabled is false * we do not start the rebuild process. */ if (!l2arc_rebuild_enabled) goto out; /* Prepare the rebuild process */ bcopy(l2dhdr->dh_start_lbps, lbps, sizeof (lbps)); /* Start the rebuild process */ for (;;) { if (!l2arc_log_blkptr_valid(dev, &lbps[0])) break; if ((err = l2arc_log_blk_read(dev, &lbps[0], &lbps[1], this_lb, next_lb, this_io, &next_io)) != 0) goto out; /* * Our memory pressure valve. If the system is running low * on memory, rather than swamping memory with new ARC buf * hdrs, we opt not to rebuild the L2ARC. At this point, * however, we have already set up our L2ARC dev to chain in * new metadata log blocks, so the user may choose to offline/ * online the L2ARC dev at a later time (or re-import the pool) * to reconstruct it (when there's less memory pressure). */ if (l2arc_hdr_limit_reached()) { ARCSTAT_BUMP(arcstat_l2_rebuild_abort_lowmem); cmn_err(CE_NOTE, "System running low on memory, " "aborting L2ARC rebuild."); err = SET_ERROR(ENOMEM); goto out; } spa_config_exit(spa, SCL_L2ARC, vd); lock_held = B_FALSE; /* * Now that we know that the next_lb checks out alright, we * can start reconstruction from this log block. * L2BLK_GET_PSIZE returns aligned size for log blocks. */ uint64_t asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop); l2arc_log_blk_restore(dev, this_lb, asize); /* * log block restored, include its pointer in the list of * pointers to log blocks present in the L2ARC device. */ lb_ptr_buf = kmem_zalloc(sizeof (l2arc_lb_ptr_buf_t), KM_SLEEP); lb_ptr_buf->lb_ptr = kmem_zalloc(sizeof (l2arc_log_blkptr_t), KM_SLEEP); bcopy(&lbps[0], lb_ptr_buf->lb_ptr, sizeof (l2arc_log_blkptr_t)); mutex_enter(&dev->l2ad_mtx); list_insert_tail(&dev->l2ad_lbptr_list, lb_ptr_buf); ARCSTAT_INCR(arcstat_l2_log_blk_asize, asize); ARCSTAT_BUMP(arcstat_l2_log_blk_count); zfs_refcount_add_many(&dev->l2ad_lb_asize, asize, lb_ptr_buf); zfs_refcount_add(&dev->l2ad_lb_count, lb_ptr_buf); mutex_exit(&dev->l2ad_mtx); vdev_space_update(vd, asize, 0, 0); /* * Protection against loops of log blocks: * * l2ad_hand l2ad_evict * V V * l2ad_start |=======================================| l2ad_end * -----|||----|||---|||----||| * (3) (2) (1) (0) * ---|||---|||----|||---||| * (7) (6) (5) (4) * * In this situation the pointer of log block (4) passes * l2arc_log_blkptr_valid() but the log block should not be * restored as it is overwritten by the payload of log block * (0). Only log blocks (0)-(3) should be restored. We check * whether l2ad_evict lies in between the payload starting * offset of the next log block (lbps[1].lbp_payload_start) * and the payload starting offset of the present log block * (lbps[0].lbp_payload_start). If true and this isn't the * first pass, we are looping from the beginning and we should * stop. */ if (l2arc_range_check_overlap(lbps[1].lbp_payload_start, lbps[0].lbp_payload_start, dev->l2ad_evict) && !dev->l2ad_first) goto out; cond_resched(); for (;;) { mutex_enter(&l2arc_rebuild_thr_lock); if (dev->l2ad_rebuild_cancel) { dev->l2ad_rebuild = B_FALSE; cv_signal(&l2arc_rebuild_thr_cv); mutex_exit(&l2arc_rebuild_thr_lock); err = SET_ERROR(ECANCELED); goto out; } mutex_exit(&l2arc_rebuild_thr_lock); if (spa_config_tryenter(spa, SCL_L2ARC, vd, RW_READER)) { lock_held = B_TRUE; break; } /* * L2ARC config lock held by somebody in writer, * possibly due to them trying to remove us. They'll * likely to want us to shut down, so after a little * delay, we check l2ad_rebuild_cancel and retry * the lock again. */ delay(1); } /* * Continue with the next log block. */ lbps[0] = lbps[1]; lbps[1] = this_lb->lb_prev_lbp; PTR_SWAP(this_lb, next_lb); this_io = next_io; next_io = NULL; } if (this_io != NULL) l2arc_log_blk_fetch_abort(this_io); out: if (next_io != NULL) l2arc_log_blk_fetch_abort(next_io); vmem_free(this_lb, sizeof (*this_lb)); vmem_free(next_lb, sizeof (*next_lb)); if (!l2arc_rebuild_enabled) { spa_history_log_internal(spa, "L2ARC rebuild", NULL, "disabled"); } else if (err == 0 && zfs_refcount_count(&dev->l2ad_lb_count) > 0) { ARCSTAT_BUMP(arcstat_l2_rebuild_success); spa_history_log_internal(spa, "L2ARC rebuild", NULL, "successful, restored %llu blocks", (u_longlong_t)zfs_refcount_count(&dev->l2ad_lb_count)); } else if (err == 0 && zfs_refcount_count(&dev->l2ad_lb_count) == 0) { /* * No error but also nothing restored, meaning the lbps array * in the device header points to invalid/non-present log * blocks. Reset the header. */ spa_history_log_internal(spa, "L2ARC rebuild", NULL, "no valid log blocks"); bzero(l2dhdr, dev->l2ad_dev_hdr_asize); l2arc_dev_hdr_update(dev); } else if (err == ECANCELED) { /* * In case the rebuild was canceled do not log to spa history * log as the pool may be in the process of being removed. */ zfs_dbgmsg("L2ARC rebuild aborted, restored %llu blocks", zfs_refcount_count(&dev->l2ad_lb_count)); } else if (err != 0) { spa_history_log_internal(spa, "L2ARC rebuild", NULL, "aborted, restored %llu blocks", (u_longlong_t)zfs_refcount_count(&dev->l2ad_lb_count)); } if (lock_held) spa_config_exit(spa, SCL_L2ARC, vd); return (err); } /* * Attempts to read the device header on the provided L2ARC device and writes * it to `hdr'. On success, this function returns 0, otherwise the appropriate * error code is returned. */ static int l2arc_dev_hdr_read(l2arc_dev_t *dev) { int err; uint64_t guid; l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr; const uint64_t l2dhdr_asize = dev->l2ad_dev_hdr_asize; abd_t *abd; guid = spa_guid(dev->l2ad_vdev->vdev_spa); abd = abd_get_from_buf(l2dhdr, l2dhdr_asize); err = zio_wait(zio_read_phys(NULL, dev->l2ad_vdev, VDEV_LABEL_START_SIZE, l2dhdr_asize, abd, ZIO_CHECKSUM_LABEL, NULL, NULL, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_SPECULATIVE, B_FALSE)); abd_free(abd); if (err != 0) { ARCSTAT_BUMP(arcstat_l2_rebuild_abort_dh_errors); zfs_dbgmsg("L2ARC IO error (%d) while reading device header, " "vdev guid: %llu", err, dev->l2ad_vdev->vdev_guid); return (err); } if (l2dhdr->dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC)) byteswap_uint64_array(l2dhdr, sizeof (*l2dhdr)); if (l2dhdr->dh_magic != L2ARC_DEV_HDR_MAGIC || l2dhdr->dh_spa_guid != guid || l2dhdr->dh_vdev_guid != dev->l2ad_vdev->vdev_guid || l2dhdr->dh_version != L2ARC_PERSISTENT_VERSION || l2dhdr->dh_log_entries != dev->l2ad_log_entries || l2dhdr->dh_end != dev->l2ad_end || !l2arc_range_check_overlap(dev->l2ad_start, dev->l2ad_end, l2dhdr->dh_evict) || (l2dhdr->dh_trim_state != VDEV_TRIM_COMPLETE && l2arc_trim_ahead > 0)) { /* * Attempt to rebuild a device containing no actual dev hdr * or containing a header from some other pool or from another * version of persistent L2ARC. */ ARCSTAT_BUMP(arcstat_l2_rebuild_abort_unsupported); return (SET_ERROR(ENOTSUP)); } return (0); } /* * Reads L2ARC log blocks from storage and validates their contents. * * This function implements a simple fetcher to make sure that while * we're processing one buffer the L2ARC is already fetching the next * one in the chain. * * The arguments this_lp and next_lp point to the current and next log block * address in the block chain. Similarly, this_lb and next_lb hold the * l2arc_log_blk_phys_t's of the current and next L2ARC blk. * * The `this_io' and `next_io' arguments are used for block fetching. * When issuing the first blk IO during rebuild, you should pass NULL for * `this_io'. This function will then issue a sync IO to read the block and * also issue an async IO to fetch the next block in the block chain. The * fetched IO is returned in `next_io'. On subsequent calls to this * function, pass the value returned in `next_io' from the previous call * as `this_io' and a fresh `next_io' pointer to hold the next fetch IO. * Prior to the call, you should initialize your `next_io' pointer to be * NULL. If no fetch IO was issued, the pointer is left set at NULL. * * On success, this function returns 0, otherwise it returns an appropriate * error code. On error the fetching IO is aborted and cleared before * returning from this function. Therefore, if we return `success', the * caller can assume that we have taken care of cleanup of fetch IOs. */ static int l2arc_log_blk_read(l2arc_dev_t *dev, const l2arc_log_blkptr_t *this_lbp, const l2arc_log_blkptr_t *next_lbp, l2arc_log_blk_phys_t *this_lb, l2arc_log_blk_phys_t *next_lb, zio_t *this_io, zio_t **next_io) { int err = 0; zio_cksum_t cksum; abd_t *abd = NULL; uint64_t asize; ASSERT(this_lbp != NULL && next_lbp != NULL); ASSERT(this_lb != NULL && next_lb != NULL); ASSERT(next_io != NULL && *next_io == NULL); ASSERT(l2arc_log_blkptr_valid(dev, this_lbp)); /* * Check to see if we have issued the IO for this log block in a * previous run. If not, this is the first call, so issue it now. */ if (this_io == NULL) { this_io = l2arc_log_blk_fetch(dev->l2ad_vdev, this_lbp, this_lb); } /* * Peek to see if we can start issuing the next IO immediately. */ if (l2arc_log_blkptr_valid(dev, next_lbp)) { /* * Start issuing IO for the next log block early - this * should help keep the L2ARC device busy while we * decompress and restore this log block. */ *next_io = l2arc_log_blk_fetch(dev->l2ad_vdev, next_lbp, next_lb); } /* Wait for the IO to read this log block to complete */ if ((err = zio_wait(this_io)) != 0) { ARCSTAT_BUMP(arcstat_l2_rebuild_abort_io_errors); zfs_dbgmsg("L2ARC IO error (%d) while reading log block, " "offset: %llu, vdev guid: %llu", err, this_lbp->lbp_daddr, dev->l2ad_vdev->vdev_guid); goto cleanup; } /* * Make sure the buffer checks out. * L2BLK_GET_PSIZE returns aligned size for log blocks. */ asize = L2BLK_GET_PSIZE((this_lbp)->lbp_prop); fletcher_4_native(this_lb, asize, NULL, &cksum); if (!ZIO_CHECKSUM_EQUAL(cksum, this_lbp->lbp_cksum)) { ARCSTAT_BUMP(arcstat_l2_rebuild_abort_cksum_lb_errors); zfs_dbgmsg("L2ARC log block cksum failed, offset: %llu, " "vdev guid: %llu, l2ad_hand: %llu, l2ad_evict: %llu", this_lbp->lbp_daddr, dev->l2ad_vdev->vdev_guid, dev->l2ad_hand, dev->l2ad_evict); err = SET_ERROR(ECKSUM); goto cleanup; } /* Now we can take our time decoding this buffer */ switch (L2BLK_GET_COMPRESS((this_lbp)->lbp_prop)) { case ZIO_COMPRESS_OFF: break; case ZIO_COMPRESS_LZ4: abd = abd_alloc_for_io(asize, B_TRUE); abd_copy_from_buf_off(abd, this_lb, 0, asize); if ((err = zio_decompress_data( L2BLK_GET_COMPRESS((this_lbp)->lbp_prop), abd, this_lb, asize, sizeof (*this_lb), NULL)) != 0) { err = SET_ERROR(EINVAL); goto cleanup; } break; default: err = SET_ERROR(EINVAL); goto cleanup; } if (this_lb->lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC)) byteswap_uint64_array(this_lb, sizeof (*this_lb)); if (this_lb->lb_magic != L2ARC_LOG_BLK_MAGIC) { err = SET_ERROR(EINVAL); goto cleanup; } cleanup: /* Abort an in-flight fetch I/O in case of error */ if (err != 0 && *next_io != NULL) { l2arc_log_blk_fetch_abort(*next_io); *next_io = NULL; } if (abd != NULL) abd_free(abd); return (err); } /* * Restores the payload of a log block to ARC. This creates empty ARC hdr * entries which only contain an l2arc hdr, essentially restoring the * buffers to their L2ARC evicted state. This function also updates space * usage on the L2ARC vdev to make sure it tracks restored buffers. */ static void l2arc_log_blk_restore(l2arc_dev_t *dev, const l2arc_log_blk_phys_t *lb, uint64_t lb_asize) { uint64_t size = 0, asize = 0; uint64_t log_entries = dev->l2ad_log_entries; /* * Usually arc_adapt() is called only for data, not headers, but * since we may allocate significant amount of memory here, let ARC * grow its arc_c. */ arc_adapt(log_entries * HDR_L2ONLY_SIZE, arc_l2c_only); for (int i = log_entries - 1; i >= 0; i--) { /* * Restore goes in the reverse temporal direction to preserve * correct temporal ordering of buffers in the l2ad_buflist. * l2arc_hdr_restore also does a list_insert_tail instead of * list_insert_head on the l2ad_buflist: * * LIST l2ad_buflist LIST * HEAD <------ (time) ------ TAIL * direction +-----+-----+-----+-----+-----+ direction * of l2arc <== | buf | buf | buf | buf | buf | ===> of rebuild * fill +-----+-----+-----+-----+-----+ * ^ ^ * | | * | | * l2arc_feed_thread l2arc_rebuild * will place new bufs here restores bufs here * * During l2arc_rebuild() the device is not used by * l2arc_feed_thread() as dev->l2ad_rebuild is set to true. */ size += L2BLK_GET_LSIZE((&lb->lb_entries[i])->le_prop); asize += vdev_psize_to_asize(dev->l2ad_vdev, L2BLK_GET_PSIZE((&lb->lb_entries[i])->le_prop)); l2arc_hdr_restore(&lb->lb_entries[i], dev); } /* * Record rebuild stats: * size Logical size of restored buffers in the L2ARC * asize Aligned size of restored buffers in the L2ARC */ ARCSTAT_INCR(arcstat_l2_rebuild_size, size); ARCSTAT_INCR(arcstat_l2_rebuild_asize, asize); ARCSTAT_INCR(arcstat_l2_rebuild_bufs, log_entries); ARCSTAT_F_AVG(arcstat_l2_log_blk_avg_asize, lb_asize); ARCSTAT_F_AVG(arcstat_l2_data_to_meta_ratio, asize / lb_asize); ARCSTAT_BUMP(arcstat_l2_rebuild_log_blks); } /* * Restores a single ARC buf hdr from a log entry. The ARC buffer is put * into a state indicating that it has been evicted to L2ARC. */ static void l2arc_hdr_restore(const l2arc_log_ent_phys_t *le, l2arc_dev_t *dev) { arc_buf_hdr_t *hdr, *exists; kmutex_t *hash_lock; arc_buf_contents_t type = L2BLK_GET_TYPE((le)->le_prop); uint64_t asize; /* * Do all the allocation before grabbing any locks, this lets us * sleep if memory is full and we don't have to deal with failed * allocations. */ hdr = arc_buf_alloc_l2only(L2BLK_GET_LSIZE((le)->le_prop), type, dev, le->le_dva, le->le_daddr, L2BLK_GET_PSIZE((le)->le_prop), le->le_birth, L2BLK_GET_COMPRESS((le)->le_prop), le->le_complevel, L2BLK_GET_PROTECTED((le)->le_prop), L2BLK_GET_PREFETCH((le)->le_prop), L2BLK_GET_STATE((le)->le_prop)); asize = vdev_psize_to_asize(dev->l2ad_vdev, L2BLK_GET_PSIZE((le)->le_prop)); /* * vdev_space_update() has to be called before arc_hdr_destroy() to * avoid underflow since the latter also calls vdev_space_update(). */ l2arc_hdr_arcstats_increment(hdr); vdev_space_update(dev->l2ad_vdev, asize, 0, 0); mutex_enter(&dev->l2ad_mtx); list_insert_tail(&dev->l2ad_buflist, hdr); (void) zfs_refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(hdr), hdr); mutex_exit(&dev->l2ad_mtx); exists = buf_hash_insert(hdr, &hash_lock); if (exists) { /* Buffer was already cached, no need to restore it. */ arc_hdr_destroy(hdr); /* * If the buffer is already cached, check whether it has * L2ARC metadata. If not, enter them and update the flag. * This is important is case of onlining a cache device, since * we previously evicted all L2ARC metadata from ARC. */ if (!HDR_HAS_L2HDR(exists)) { arc_hdr_set_flags(exists, ARC_FLAG_HAS_L2HDR); exists->b_l2hdr.b_dev = dev; exists->b_l2hdr.b_daddr = le->le_daddr; exists->b_l2hdr.b_arcs_state = L2BLK_GET_STATE((le)->le_prop); mutex_enter(&dev->l2ad_mtx); list_insert_tail(&dev->l2ad_buflist, exists); (void) zfs_refcount_add_many(&dev->l2ad_alloc, arc_hdr_size(exists), exists); mutex_exit(&dev->l2ad_mtx); l2arc_hdr_arcstats_increment(exists); vdev_space_update(dev->l2ad_vdev, asize, 0, 0); } ARCSTAT_BUMP(arcstat_l2_rebuild_bufs_precached); } mutex_exit(hash_lock); } /* * Starts an asynchronous read IO to read a log block. This is used in log * block reconstruction to start reading the next block before we are done * decoding and reconstructing the current block, to keep the l2arc device * nice and hot with read IO to process. * The returned zio will contain a newly allocated memory buffers for the IO * data which should then be freed by the caller once the zio is no longer * needed (i.e. due to it having completed). If you wish to abort this * zio, you should do so using l2arc_log_blk_fetch_abort, which takes * care of disposing of the allocated buffers correctly. */ static zio_t * l2arc_log_blk_fetch(vdev_t *vd, const l2arc_log_blkptr_t *lbp, l2arc_log_blk_phys_t *lb) { uint32_t asize; zio_t *pio; l2arc_read_callback_t *cb; /* L2BLK_GET_PSIZE returns aligned size for log blocks */ asize = L2BLK_GET_PSIZE((lbp)->lbp_prop); ASSERT(asize <= sizeof (l2arc_log_blk_phys_t)); cb = kmem_zalloc(sizeof (l2arc_read_callback_t), KM_SLEEP); cb->l2rcb_abd = abd_get_from_buf(lb, asize); pio = zio_root(vd->vdev_spa, l2arc_blk_fetch_done, cb, ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY); (void) zio_nowait(zio_read_phys(pio, vd, lbp->lbp_daddr, asize, cb->l2rcb_abd, ZIO_CHECKSUM_OFF, NULL, NULL, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_DONT_CACHE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY, B_FALSE)); return (pio); } /* * Aborts a zio returned from l2arc_log_blk_fetch and frees the data * buffers allocated for it. */ static void l2arc_log_blk_fetch_abort(zio_t *zio) { (void) zio_wait(zio); } /* * Creates a zio to update the device header on an l2arc device. */ void l2arc_dev_hdr_update(l2arc_dev_t *dev) { l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr; const uint64_t l2dhdr_asize = dev->l2ad_dev_hdr_asize; abd_t *abd; int err; VERIFY(spa_config_held(dev->l2ad_spa, SCL_STATE_ALL, RW_READER)); l2dhdr->dh_magic = L2ARC_DEV_HDR_MAGIC; l2dhdr->dh_version = L2ARC_PERSISTENT_VERSION; l2dhdr->dh_spa_guid = spa_guid(dev->l2ad_vdev->vdev_spa); l2dhdr->dh_vdev_guid = dev->l2ad_vdev->vdev_guid; l2dhdr->dh_log_entries = dev->l2ad_log_entries; l2dhdr->dh_evict = dev->l2ad_evict; l2dhdr->dh_start = dev->l2ad_start; l2dhdr->dh_end = dev->l2ad_end; l2dhdr->dh_lb_asize = zfs_refcount_count(&dev->l2ad_lb_asize); l2dhdr->dh_lb_count = zfs_refcount_count(&dev->l2ad_lb_count); l2dhdr->dh_flags = 0; l2dhdr->dh_trim_action_time = dev->l2ad_vdev->vdev_trim_action_time; l2dhdr->dh_trim_state = dev->l2ad_vdev->vdev_trim_state; if (dev->l2ad_first) l2dhdr->dh_flags |= L2ARC_DEV_HDR_EVICT_FIRST; abd = abd_get_from_buf(l2dhdr, l2dhdr_asize); err = zio_wait(zio_write_phys(NULL, dev->l2ad_vdev, VDEV_LABEL_START_SIZE, l2dhdr_asize, abd, ZIO_CHECKSUM_LABEL, NULL, NULL, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE)); abd_free(abd); if (err != 0) { zfs_dbgmsg("L2ARC IO error (%d) while writing device header, " "vdev guid: %llu", err, dev->l2ad_vdev->vdev_guid); } } /* * Commits a log block to the L2ARC device. This routine is invoked from * l2arc_write_buffers when the log block fills up. * This function allocates some memory to temporarily hold the serialized * buffer to be written. This is then released in l2arc_write_done. */ static void l2arc_log_blk_commit(l2arc_dev_t *dev, zio_t *pio, l2arc_write_callback_t *cb) { l2arc_log_blk_phys_t *lb = &dev->l2ad_log_blk; l2arc_dev_hdr_phys_t *l2dhdr = dev->l2ad_dev_hdr; uint64_t psize, asize; zio_t *wzio; l2arc_lb_abd_buf_t *abd_buf; uint8_t *tmpbuf; l2arc_lb_ptr_buf_t *lb_ptr_buf; VERIFY3S(dev->l2ad_log_ent_idx, ==, dev->l2ad_log_entries); tmpbuf = zio_buf_alloc(sizeof (*lb)); abd_buf = zio_buf_alloc(sizeof (*abd_buf)); abd_buf->abd = abd_get_from_buf(lb, sizeof (*lb)); lb_ptr_buf = kmem_zalloc(sizeof (l2arc_lb_ptr_buf_t), KM_SLEEP); lb_ptr_buf->lb_ptr = kmem_zalloc(sizeof (l2arc_log_blkptr_t), KM_SLEEP); /* link the buffer into the block chain */ lb->lb_prev_lbp = l2dhdr->dh_start_lbps[1]; lb->lb_magic = L2ARC_LOG_BLK_MAGIC; /* * l2arc_log_blk_commit() may be called multiple times during a single * l2arc_write_buffers() call. Save the allocated abd buffers in a list * so we can free them in l2arc_write_done() later on. */ list_insert_tail(&cb->l2wcb_abd_list, abd_buf); /* try to compress the buffer */ psize = zio_compress_data(ZIO_COMPRESS_LZ4, abd_buf->abd, tmpbuf, sizeof (*lb), 0); /* a log block is never entirely zero */ ASSERT(psize != 0); asize = vdev_psize_to_asize(dev->l2ad_vdev, psize); ASSERT(asize <= sizeof (*lb)); /* * Update the start log block pointer in the device header to point * to the log block we're about to write. */ l2dhdr->dh_start_lbps[1] = l2dhdr->dh_start_lbps[0]; l2dhdr->dh_start_lbps[0].lbp_daddr = dev->l2ad_hand; l2dhdr->dh_start_lbps[0].lbp_payload_asize = dev->l2ad_log_blk_payload_asize; l2dhdr->dh_start_lbps[0].lbp_payload_start = dev->l2ad_log_blk_payload_start; _NOTE(CONSTCOND) L2BLK_SET_LSIZE( (&l2dhdr->dh_start_lbps[0])->lbp_prop, sizeof (*lb)); L2BLK_SET_PSIZE( (&l2dhdr->dh_start_lbps[0])->lbp_prop, asize); L2BLK_SET_CHECKSUM( (&l2dhdr->dh_start_lbps[0])->lbp_prop, ZIO_CHECKSUM_FLETCHER_4); if (asize < sizeof (*lb)) { /* compression succeeded */ bzero(tmpbuf + psize, asize - psize); L2BLK_SET_COMPRESS( (&l2dhdr->dh_start_lbps[0])->lbp_prop, ZIO_COMPRESS_LZ4); } else { /* compression failed */ bcopy(lb, tmpbuf, sizeof (*lb)); L2BLK_SET_COMPRESS( (&l2dhdr->dh_start_lbps[0])->lbp_prop, ZIO_COMPRESS_OFF); } /* checksum what we're about to write */ fletcher_4_native(tmpbuf, asize, NULL, &l2dhdr->dh_start_lbps[0].lbp_cksum); abd_free(abd_buf->abd); /* perform the write itself */ abd_buf->abd = abd_get_from_buf(tmpbuf, sizeof (*lb)); abd_take_ownership_of_buf(abd_buf->abd, B_TRUE); wzio = zio_write_phys(pio, dev->l2ad_vdev, dev->l2ad_hand, asize, abd_buf->abd, ZIO_CHECKSUM_OFF, NULL, NULL, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_CANFAIL, B_FALSE); DTRACE_PROBE2(l2arc__write, vdev_t *, dev->l2ad_vdev, zio_t *, wzio); (void) zio_nowait(wzio); dev->l2ad_hand += asize; /* * Include the committed log block's pointer in the list of pointers * to log blocks present in the L2ARC device. */ bcopy(&l2dhdr->dh_start_lbps[0], lb_ptr_buf->lb_ptr, sizeof (l2arc_log_blkptr_t)); mutex_enter(&dev->l2ad_mtx); list_insert_head(&dev->l2ad_lbptr_list, lb_ptr_buf); ARCSTAT_INCR(arcstat_l2_log_blk_asize, asize); ARCSTAT_BUMP(arcstat_l2_log_blk_count); zfs_refcount_add_many(&dev->l2ad_lb_asize, asize, lb_ptr_buf); zfs_refcount_add(&dev->l2ad_lb_count, lb_ptr_buf); mutex_exit(&dev->l2ad_mtx); vdev_space_update(dev->l2ad_vdev, asize, 0, 0); /* bump the kstats */ ARCSTAT_INCR(arcstat_l2_write_bytes, asize); ARCSTAT_BUMP(arcstat_l2_log_blk_writes); ARCSTAT_F_AVG(arcstat_l2_log_blk_avg_asize, asize); ARCSTAT_F_AVG(arcstat_l2_data_to_meta_ratio, dev->l2ad_log_blk_payload_asize / asize); /* start a new log block */ dev->l2ad_log_ent_idx = 0; dev->l2ad_log_blk_payload_asize = 0; dev->l2ad_log_blk_payload_start = 0; } /* * Validates an L2ARC log block address to make sure that it can be read * from the provided L2ARC device. */ boolean_t l2arc_log_blkptr_valid(l2arc_dev_t *dev, const l2arc_log_blkptr_t *lbp) { /* L2BLK_GET_PSIZE returns aligned size for log blocks */ uint64_t asize = L2BLK_GET_PSIZE((lbp)->lbp_prop); uint64_t end = lbp->lbp_daddr + asize - 1; uint64_t start = lbp->lbp_payload_start; boolean_t evicted = B_FALSE; /* * A log block is valid if all of the following conditions are true: * - it fits entirely (including its payload) between l2ad_start and * l2ad_end * - it has a valid size * - neither the log block itself nor part of its payload was evicted * by l2arc_evict(): * * l2ad_hand l2ad_evict * | | lbp_daddr * | start | | end * | | | | | * V V V V V * l2ad_start ============================================ l2ad_end * --------------------------|||| * ^ ^ * | log block * payload */ evicted = l2arc_range_check_overlap(start, end, dev->l2ad_hand) || l2arc_range_check_overlap(start, end, dev->l2ad_evict) || l2arc_range_check_overlap(dev->l2ad_hand, dev->l2ad_evict, start) || l2arc_range_check_overlap(dev->l2ad_hand, dev->l2ad_evict, end); return (start >= dev->l2ad_start && end <= dev->l2ad_end && asize > 0 && asize <= sizeof (l2arc_log_blk_phys_t) && (!evicted || dev->l2ad_first)); } /* * Inserts ARC buffer header `hdr' into the current L2ARC log block on * the device. The buffer being inserted must be present in L2ARC. * Returns B_TRUE if the L2ARC log block is full and needs to be committed * to L2ARC, or B_FALSE if it still has room for more ARC buffers. */ static boolean_t l2arc_log_blk_insert(l2arc_dev_t *dev, const arc_buf_hdr_t *hdr) { l2arc_log_blk_phys_t *lb = &dev->l2ad_log_blk; l2arc_log_ent_phys_t *le; if (dev->l2ad_log_entries == 0) return (B_FALSE); int index = dev->l2ad_log_ent_idx++; ASSERT3S(index, <, dev->l2ad_log_entries); ASSERT(HDR_HAS_L2HDR(hdr)); le = &lb->lb_entries[index]; bzero(le, sizeof (*le)); le->le_dva = hdr->b_dva; le->le_birth = hdr->b_birth; le->le_daddr = hdr->b_l2hdr.b_daddr; if (index == 0) dev->l2ad_log_blk_payload_start = le->le_daddr; L2BLK_SET_LSIZE((le)->le_prop, HDR_GET_LSIZE(hdr)); L2BLK_SET_PSIZE((le)->le_prop, HDR_GET_PSIZE(hdr)); L2BLK_SET_COMPRESS((le)->le_prop, HDR_GET_COMPRESS(hdr)); le->le_complevel = hdr->b_complevel; L2BLK_SET_TYPE((le)->le_prop, hdr->b_type); L2BLK_SET_PROTECTED((le)->le_prop, !!(HDR_PROTECTED(hdr))); L2BLK_SET_PREFETCH((le)->le_prop, !!(HDR_PREFETCH(hdr))); L2BLK_SET_STATE((le)->le_prop, hdr->b_l1hdr.b_state->arcs_state); dev->l2ad_log_blk_payload_asize += vdev_psize_to_asize(dev->l2ad_vdev, HDR_GET_PSIZE(hdr)); return (dev->l2ad_log_ent_idx == dev->l2ad_log_entries); } /* * Checks whether a given L2ARC device address sits in a time-sequential * range. The trick here is that the L2ARC is a rotary buffer, so we can't * just do a range comparison, we need to handle the situation in which the * range wraps around the end of the L2ARC device. Arguments: * bottom -- Lower end of the range to check (written to earlier). * top -- Upper end of the range to check (written to later). * check -- The address for which we want to determine if it sits in * between the top and bottom. * * The 3-way conditional below represents the following cases: * * bottom < top : Sequentially ordered case: * --------+-------------------+ * | (overlap here?) | * L2ARC dev V V * |---------------============--------------| * * bottom > top: Looped-around case: * --------+------------------+ * | (overlap here?) | * L2ARC dev V V * |===============---------------===========| * ^ ^ * | (or here?) | * +---------------+--------- * * top == bottom : Just a single address comparison. */ boolean_t l2arc_range_check_overlap(uint64_t bottom, uint64_t top, uint64_t check) { if (bottom < top) return (bottom <= check && check <= top); else if (bottom > top) return (check <= top || bottom <= check); else return (check == top); } EXPORT_SYMBOL(arc_buf_size); EXPORT_SYMBOL(arc_write); EXPORT_SYMBOL(arc_read); EXPORT_SYMBOL(arc_buf_info); EXPORT_SYMBOL(arc_getbuf_func); EXPORT_SYMBOL(arc_add_prune_callback); EXPORT_SYMBOL(arc_remove_prune_callback); /* BEGIN CSTYLED */ ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min, param_set_arc_long, param_get_long, ZMOD_RW, "Min arc size"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, max, param_set_arc_long, param_get_long, ZMOD_RW, "Max arc size"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_limit, param_set_arc_long, param_get_long, ZMOD_RW, "Metadata limit for arc size"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_limit_percent, param_set_arc_long, param_get_long, ZMOD_RW, "Percent of arc size for arc meta limit"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, meta_min, param_set_arc_long, param_get_long, ZMOD_RW, "Min arc metadata"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_prune, INT, ZMOD_RW, "Meta objects to scan for prune"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_adjust_restarts, INT, ZMOD_RW, "Limit number of restarts in arc_evict_meta"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, meta_strategy, INT, ZMOD_RW, "Meta reclaim strategy"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, grow_retry, param_set_arc_int, param_get_int, ZMOD_RW, "Seconds before growing arc size"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, p_dampener_disable, INT, ZMOD_RW, "Disable arc_p adapt dampener"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, shrink_shift, param_set_arc_int, param_get_int, ZMOD_RW, "log2(fraction of arc to reclaim)"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, pc_percent, UINT, ZMOD_RW, "Percent of pagecache to reclaim arc to"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, p_min_shift, param_set_arc_int, param_get_int, ZMOD_RW, "arc_c shift to calc min/max arc_p"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, average_blocksize, INT, ZMOD_RD, "Target average block size"); ZFS_MODULE_PARAM(zfs, zfs_, compressed_arc_enabled, INT, ZMOD_RW, "Disable compressed arc buffers"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min_prefetch_ms, param_set_arc_int, param_get_int, ZMOD_RW, "Min life of prefetch block in ms"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, min_prescient_prefetch_ms, param_set_arc_int, param_get_int, ZMOD_RW, "Min life of prescient prefetched block in ms"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, write_max, ULONG, ZMOD_RW, "Max write bytes per interval"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, write_boost, ULONG, ZMOD_RW, "Extra write bytes during device warmup"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, headroom, ULONG, ZMOD_RW, "Number of max device writes to precache"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, headroom_boost, ULONG, ZMOD_RW, "Compressed l2arc_headroom multiplier"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, trim_ahead, ULONG, ZMOD_RW, "TRIM ahead L2ARC write size multiplier"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, feed_secs, ULONG, ZMOD_RW, "Seconds between L2ARC writing"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, feed_min_ms, ULONG, ZMOD_RW, "Min feed interval in milliseconds"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, noprefetch, INT, ZMOD_RW, "Skip caching prefetched buffers"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, feed_again, INT, ZMOD_RW, "Turbo L2ARC warmup"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, norw, INT, ZMOD_RW, "No reads during writes"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, meta_percent, INT, ZMOD_RW, "Percent of ARC size allowed for L2ARC-only headers"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, rebuild_enabled, INT, ZMOD_RW, "Rebuild the L2ARC when importing a pool"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, rebuild_blocks_min_l2size, ULONG, ZMOD_RW, "Min size in bytes to write rebuild log blocks in L2ARC"); ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, mfuonly, INT, ZMOD_RW, "Cache only MFU data from ARC into L2ARC"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, lotsfree_percent, param_set_arc_int, param_get_int, ZMOD_RW, "System free memory I/O throttle in bytes"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, sys_free, param_set_arc_long, param_get_long, ZMOD_RW, "System free memory target size in bytes"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, dnode_limit, param_set_arc_long, param_get_long, ZMOD_RW, "Minimum bytes of dnodes in arc"); ZFS_MODULE_PARAM_CALL(zfs_arc, zfs_arc_, dnode_limit_percent, param_set_arc_long, param_get_long, ZMOD_RW, "Percent of ARC meta buffers for dnodes"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, dnode_reduce_percent, ULONG, ZMOD_RW, "Percentage of excess dnodes to try to unpin"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, eviction_pct, INT, ZMOD_RW, "When full, ARC allocation waits for eviction of this % of alloc size"); ZFS_MODULE_PARAM(zfs_arc, zfs_arc_, evict_batch_limit, INT, ZMOD_RW, "The number of headers to evict per sublist before moving to the next"); /* END CSTYLED */ diff --git a/sys/contrib/openzfs/module/zfs/dbuf.c b/sys/contrib/openzfs/module/zfs/dbuf.c index 9d741545fae3..5b072f02613b 100644 --- a/sys/contrib/openzfs/module/zfs/dbuf.c +++ b/sys/contrib/openzfs/module/zfs/dbuf.c @@ -1,4957 +1,5019 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2012, 2020 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2019, Klara Inc. * Copyright (c) 2019, Allan Jude */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include kstat_t *dbuf_ksp; typedef struct dbuf_stats { /* * Various statistics about the size of the dbuf cache. */ kstat_named_t cache_count; kstat_named_t cache_size_bytes; kstat_named_t cache_size_bytes_max; /* * Statistics regarding the bounds on the dbuf cache size. */ kstat_named_t cache_target_bytes; kstat_named_t cache_lowater_bytes; kstat_named_t cache_hiwater_bytes; /* * Total number of dbuf cache evictions that have occurred. */ kstat_named_t cache_total_evicts; /* * The distribution of dbuf levels in the dbuf cache and * the total size of all dbufs at each level. */ kstat_named_t cache_levels[DN_MAX_LEVELS]; kstat_named_t cache_levels_bytes[DN_MAX_LEVELS]; /* * Statistics about the dbuf hash table. */ kstat_named_t hash_hits; kstat_named_t hash_misses; kstat_named_t hash_collisions; kstat_named_t hash_elements; kstat_named_t hash_elements_max; /* * Number of sublists containing more than one dbuf in the dbuf * hash table. Keep track of the longest hash chain. */ kstat_named_t hash_chains; kstat_named_t hash_chain_max; /* * Number of times a dbuf_create() discovers that a dbuf was * already created and in the dbuf hash table. */ kstat_named_t hash_insert_race; /* * Statistics about the size of the metadata dbuf cache. */ kstat_named_t metadata_cache_count; kstat_named_t metadata_cache_size_bytes; kstat_named_t metadata_cache_size_bytes_max; /* * For diagnostic purposes, this is incremented whenever we can't add * something to the metadata cache because it's full, and instead put * the data in the regular dbuf cache. */ kstat_named_t metadata_cache_overflow; } dbuf_stats_t; dbuf_stats_t dbuf_stats = { { "cache_count", KSTAT_DATA_UINT64 }, { "cache_size_bytes", KSTAT_DATA_UINT64 }, { "cache_size_bytes_max", KSTAT_DATA_UINT64 }, { "cache_target_bytes", KSTAT_DATA_UINT64 }, { "cache_lowater_bytes", KSTAT_DATA_UINT64 }, { "cache_hiwater_bytes", KSTAT_DATA_UINT64 }, { "cache_total_evicts", KSTAT_DATA_UINT64 }, { { "cache_levels_N", KSTAT_DATA_UINT64 } }, { { "cache_levels_bytes_N", KSTAT_DATA_UINT64 } }, { "hash_hits", KSTAT_DATA_UINT64 }, { "hash_misses", KSTAT_DATA_UINT64 }, { "hash_collisions", KSTAT_DATA_UINT64 }, { "hash_elements", KSTAT_DATA_UINT64 }, { "hash_elements_max", KSTAT_DATA_UINT64 }, { "hash_chains", KSTAT_DATA_UINT64 }, { "hash_chain_max", KSTAT_DATA_UINT64 }, { "hash_insert_race", KSTAT_DATA_UINT64 }, { "metadata_cache_count", KSTAT_DATA_UINT64 }, { "metadata_cache_size_bytes", KSTAT_DATA_UINT64 }, { "metadata_cache_size_bytes_max", KSTAT_DATA_UINT64 }, { "metadata_cache_overflow", KSTAT_DATA_UINT64 } }; +struct { + wmsum_t cache_count; + wmsum_t cache_total_evicts; + wmsum_t cache_levels[DN_MAX_LEVELS]; + wmsum_t cache_levels_bytes[DN_MAX_LEVELS]; + wmsum_t hash_hits; + wmsum_t hash_misses; + wmsum_t hash_collisions; + wmsum_t hash_chains; + wmsum_t hash_insert_race; + wmsum_t metadata_cache_count; + wmsum_t metadata_cache_overflow; +} dbuf_sums; + #define DBUF_STAT_INCR(stat, val) \ - atomic_add_64(&dbuf_stats.stat.value.ui64, (val)); + wmsum_add(&dbuf_sums.stat, val); #define DBUF_STAT_DECR(stat, val) \ DBUF_STAT_INCR(stat, -(val)); #define DBUF_STAT_BUMP(stat) \ DBUF_STAT_INCR(stat, 1); #define DBUF_STAT_BUMPDOWN(stat) \ DBUF_STAT_INCR(stat, -1); #define DBUF_STAT_MAX(stat, v) { \ uint64_t _m; \ while ((v) > (_m = dbuf_stats.stat.value.ui64) && \ (_m != atomic_cas_64(&dbuf_stats.stat.value.ui64, _m, (v))))\ continue; \ } static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx); static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx); static void dbuf_sync_leaf_verify_bonus_dnode(dbuf_dirty_record_t *dr); static int dbuf_read_verify_dnode_crypt(dmu_buf_impl_t *db, uint32_t flags); extern inline void dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync, dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp); /* * Global data structures and functions for the dbuf cache. */ static kmem_cache_t *dbuf_kmem_cache; static taskq_t *dbu_evict_taskq; static kthread_t *dbuf_cache_evict_thread; static kmutex_t dbuf_evict_lock; static kcondvar_t dbuf_evict_cv; static boolean_t dbuf_evict_thread_exit; /* * There are two dbuf caches; each dbuf can only be in one of them at a time. * * 1. Cache of metadata dbufs, to help make read-heavy administrative commands * from /sbin/zfs run faster. The "metadata cache" specifically stores dbufs * that represent the metadata that describes filesystems/snapshots/ * bookmarks/properties/etc. We only evict from this cache when we export a * pool, to short-circuit as much I/O as possible for all administrative * commands that need the metadata. There is no eviction policy for this * cache, because we try to only include types in it which would occupy a * very small amount of space per object but create a large impact on the * performance of these commands. Instead, after it reaches a maximum size * (which should only happen on very small memory systems with a very large * number of filesystem objects), we stop taking new dbufs into the * metadata cache, instead putting them in the normal dbuf cache. * * 2. LRU cache of dbufs. The dbuf cache maintains a list of dbufs that * are not currently held but have been recently released. These dbufs * are not eligible for arc eviction until they are aged out of the cache. * Dbufs that are aged out of the cache will be immediately destroyed and * become eligible for arc eviction. * * Dbufs are added to these caches once the last hold is released. If a dbuf is * later accessed and still exists in the dbuf cache, then it will be removed * from the cache and later re-added to the head of the cache. * * If a given dbuf meets the requirements for the metadata cache, it will go * there, otherwise it will be considered for the generic LRU dbuf cache. The * caches and the refcounts tracking their sizes are stored in an array indexed * by those caches' matching enum values (from dbuf_cached_state_t). */ typedef struct dbuf_cache { multilist_t cache; zfs_refcount_t size ____cacheline_aligned; } dbuf_cache_t; dbuf_cache_t dbuf_caches[DB_CACHE_MAX]; /* Size limits for the caches */ unsigned long dbuf_cache_max_bytes = ULONG_MAX; unsigned long dbuf_metadata_cache_max_bytes = ULONG_MAX; /* Set the default sizes of the caches to log2 fraction of arc size */ int dbuf_cache_shift = 5; int dbuf_metadata_cache_shift = 6; static unsigned long dbuf_cache_target_bytes(void); static unsigned long dbuf_metadata_cache_target_bytes(void); /* * The LRU dbuf cache uses a three-stage eviction policy: * - A low water marker designates when the dbuf eviction thread * should stop evicting from the dbuf cache. * - When we reach the maximum size (aka mid water mark), we * signal the eviction thread to run. * - The high water mark indicates when the eviction thread * is unable to keep up with the incoming load and eviction must * happen in the context of the calling thread. * * The dbuf cache: * (max size) * low water mid water hi water * +----------------------------------------+----------+----------+ * | | | | * | | | | * | | | | * | | | | * +----------------------------------------+----------+----------+ * stop signal evict * evicting eviction directly * thread * * The high and low water marks indicate the operating range for the eviction * thread. The low water mark is, by default, 90% of the total size of the * cache and the high water mark is at 110% (both of these percentages can be * changed by setting dbuf_cache_lowater_pct and dbuf_cache_hiwater_pct, * respectively). The eviction thread will try to ensure that the cache remains * within this range by waking up every second and checking if the cache is * above the low water mark. The thread can also be woken up by callers adding * elements into the cache if the cache is larger than the mid water (i.e max * cache size). Once the eviction thread is woken up and eviction is required, * it will continue evicting buffers until it's able to reduce the cache size * to the low water mark. If the cache size continues to grow and hits the high * water mark, then callers adding elements to the cache will begin to evict * directly from the cache until the cache is no longer above the high water * mark. */ /* * The percentage above and below the maximum cache size. */ uint_t dbuf_cache_hiwater_pct = 10; uint_t dbuf_cache_lowater_pct = 10; /* ARGSUSED */ static int dbuf_cons(void *vdb, void *unused, int kmflag) { dmu_buf_impl_t *db = vdb; bzero(db, sizeof (dmu_buf_impl_t)); mutex_init(&db->db_mtx, NULL, MUTEX_DEFAULT, NULL); rw_init(&db->db_rwlock, NULL, RW_DEFAULT, NULL); cv_init(&db->db_changed, NULL, CV_DEFAULT, NULL); multilist_link_init(&db->db_cache_link); zfs_refcount_create(&db->db_holds); return (0); } /* ARGSUSED */ static void dbuf_dest(void *vdb, void *unused) { dmu_buf_impl_t *db = vdb; mutex_destroy(&db->db_mtx); rw_destroy(&db->db_rwlock); cv_destroy(&db->db_changed); ASSERT(!multilist_link_active(&db->db_cache_link)); zfs_refcount_destroy(&db->db_holds); } /* * dbuf hash table routines */ static dbuf_hash_table_t dbuf_hash_table; -static uint64_t dbuf_hash_count; - /* * We use Cityhash for this. It's fast, and has good hash properties without * requiring any large static buffers. */ static uint64_t dbuf_hash(void *os, uint64_t obj, uint8_t lvl, uint64_t blkid) { return (cityhash4((uintptr_t)os, obj, (uint64_t)lvl, blkid)); } #define DTRACE_SET_STATE(db, why) \ DTRACE_PROBE2(dbuf__state_change, dmu_buf_impl_t *, db, \ const char *, why) #define DBUF_EQUAL(dbuf, os, obj, level, blkid) \ ((dbuf)->db.db_object == (obj) && \ (dbuf)->db_objset == (os) && \ (dbuf)->db_level == (level) && \ (dbuf)->db_blkid == (blkid)) dmu_buf_impl_t * dbuf_find(objset_t *os, uint64_t obj, uint8_t level, uint64_t blkid) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv; uint64_t idx; dmu_buf_impl_t *db; hv = dbuf_hash(os, obj, level, blkid); idx = hv & h->hash_table_mask; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (db = h->hash_table[idx]; db != NULL; db = db->db_hash_next) { if (DBUF_EQUAL(db, os, obj, level, blkid)) { mutex_enter(&db->db_mtx); if (db->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (db); } mutex_exit(&db->db_mtx); } } mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (NULL); } static dmu_buf_impl_t * dbuf_find_bonus(objset_t *os, uint64_t object) { dnode_t *dn; dmu_buf_impl_t *db = NULL; if (dnode_hold(os, object, FTAG, &dn) == 0) { rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_bonus != NULL) { db = dn->dn_bonus; mutex_enter(&db->db_mtx); } rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); } return (db); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. */ static dmu_buf_impl_t * dbuf_hash_insert(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; objset_t *os = db->db_objset; uint64_t obj = db->db.db_object; int level = db->db_level; uint64_t blkid, hv, idx; dmu_buf_impl_t *dbf; uint32_t i; blkid = db->db_blkid; hv = dbuf_hash(os, obj, level, blkid); idx = hv & h->hash_table_mask; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (dbf = h->hash_table[idx], i = 0; dbf != NULL; dbf = dbf->db_hash_next, i++) { if (DBUF_EQUAL(dbf, os, obj, level, blkid)) { mutex_enter(&dbf->db_mtx); if (dbf->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (dbf); } mutex_exit(&dbf->db_mtx); } } if (i > 0) { DBUF_STAT_BUMP(hash_collisions); if (i == 1) DBUF_STAT_BUMP(hash_chains); DBUF_STAT_MAX(hash_chain_max, i); } mutex_enter(&db->db_mtx); db->db_hash_next = h->hash_table[idx]; h->hash_table[idx] = db; mutex_exit(DBUF_HASH_MUTEX(h, idx)); - atomic_inc_64(&dbuf_hash_count); - DBUF_STAT_MAX(hash_elements_max, dbuf_hash_count); + uint64_t he = atomic_inc_64_nv(&dbuf_stats.hash_elements.value.ui64); + DBUF_STAT_MAX(hash_elements_max, he); return (NULL); } /* * This returns whether this dbuf should be stored in the metadata cache, which * is based on whether it's from one of the dnode types that store data related * to traversing dataset hierarchies. */ static boolean_t dbuf_include_in_metadata_cache(dmu_buf_impl_t *db) { DB_DNODE_ENTER(db); dmu_object_type_t type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); /* Check if this dbuf is one of the types we care about */ if (DMU_OT_IS_METADATA_CACHED(type)) { /* If we hit this, then we set something up wrong in dmu_ot */ ASSERT(DMU_OT_IS_METADATA(type)); /* * Sanity check for small-memory systems: don't allocate too * much memory for this purpose. */ if (zfs_refcount_count( &dbuf_caches[DB_DBUF_METADATA_CACHE].size) > dbuf_metadata_cache_target_bytes()) { DBUF_STAT_BUMP(metadata_cache_overflow); return (B_FALSE); } return (B_TRUE); } return (B_FALSE); } /* * Remove an entry from the hash table. It must be in the EVICTING state. */ static void dbuf_hash_remove(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv, idx; dmu_buf_impl_t *dbf, **dbp; hv = dbuf_hash(db->db_objset, db->db.db_object, db->db_level, db->db_blkid); idx = hv & h->hash_table_mask; /* * We mustn't hold db_mtx to maintain lock ordering: * DBUF_HASH_MUTEX > db_mtx. */ ASSERT(zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_state == DB_EVICTING); ASSERT(!MUTEX_HELD(&db->db_mtx)); mutex_enter(DBUF_HASH_MUTEX(h, idx)); dbp = &h->hash_table[idx]; while ((dbf = *dbp) != db) { dbp = &dbf->db_hash_next; ASSERT(dbf != NULL); } *dbp = db->db_hash_next; db->db_hash_next = NULL; if (h->hash_table[idx] && h->hash_table[idx]->db_hash_next == NULL) DBUF_STAT_BUMPDOWN(hash_chains); mutex_exit(DBUF_HASH_MUTEX(h, idx)); - atomic_dec_64(&dbuf_hash_count); + atomic_dec_64(&dbuf_stats.hash_elements.value.ui64); } typedef enum { DBVU_EVICTING, DBVU_NOT_EVICTING } dbvu_verify_type_t; static void dbuf_verify_user(dmu_buf_impl_t *db, dbvu_verify_type_t verify_type) { #ifdef ZFS_DEBUG int64_t holds; if (db->db_user == NULL) return; /* Only data blocks support the attachment of user data. */ ASSERT(db->db_level == 0); /* Clients must resolve a dbuf before attaching user data. */ ASSERT(db->db.db_data != NULL); ASSERT3U(db->db_state, ==, DB_CACHED); holds = zfs_refcount_count(&db->db_holds); if (verify_type == DBVU_EVICTING) { /* * Immediate eviction occurs when holds == dirtycnt. * For normal eviction buffers, holds is zero on * eviction, except when dbuf_fix_old_data() calls * dbuf_clear_data(). However, the hold count can grow * during eviction even though db_mtx is held (see * dmu_bonus_hold() for an example), so we can only * test the generic invariant that holds >= dirtycnt. */ ASSERT3U(holds, >=, db->db_dirtycnt); } else { if (db->db_user_immediate_evict == TRUE) ASSERT3U(holds, >=, db->db_dirtycnt); else ASSERT3U(holds, >, 0); } #endif } static void dbuf_evict_user(dmu_buf_impl_t *db) { dmu_buf_user_t *dbu = db->db_user; ASSERT(MUTEX_HELD(&db->db_mtx)); if (dbu == NULL) return; dbuf_verify_user(db, DBVU_EVICTING); db->db_user = NULL; #ifdef ZFS_DEBUG if (dbu->dbu_clear_on_evict_dbufp != NULL) *dbu->dbu_clear_on_evict_dbufp = NULL; #endif /* * There are two eviction callbacks - one that we call synchronously * and one that we invoke via a taskq. The async one is useful for * avoiding lock order reversals and limiting stack depth. * * Note that if we have a sync callback but no async callback, * it's likely that the sync callback will free the structure * containing the dbu. In that case we need to take care to not * dereference dbu after calling the sync evict func. */ boolean_t has_async = (dbu->dbu_evict_func_async != NULL); if (dbu->dbu_evict_func_sync != NULL) dbu->dbu_evict_func_sync(dbu); if (has_async) { taskq_dispatch_ent(dbu_evict_taskq, dbu->dbu_evict_func_async, dbu, 0, &dbu->dbu_tqent); } } boolean_t dbuf_is_metadata(dmu_buf_impl_t *db) { /* * Consider indirect blocks and spill blocks to be meta data. */ if (db->db_level > 0 || db->db_blkid == DMU_SPILL_BLKID) { return (B_TRUE); } else { boolean_t is_metadata; DB_DNODE_ENTER(db); is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type); DB_DNODE_EXIT(db); return (is_metadata); } } /* * This function *must* return indices evenly distributed between all * sublists of the multilist. This is needed due to how the dbuf eviction * code is laid out; dbuf_evict_thread() assumes dbufs are evenly * distributed between all sublists and uses this assumption when * deciding which sublist to evict from and how much to evict from it. */ static unsigned int dbuf_cache_multilist_index_func(multilist_t *ml, void *obj) { dmu_buf_impl_t *db = obj; /* * The assumption here, is the hash value for a given * dmu_buf_impl_t will remain constant throughout it's lifetime * (i.e. it's objset, object, level and blkid fields don't change). * Thus, we don't need to store the dbuf's sublist index * on insertion, as this index can be recalculated on removal. * * Also, the low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. */ return (dbuf_hash(db->db_objset, db->db.db_object, db->db_level, db->db_blkid) % multilist_get_num_sublists(ml)); } /* * The target size of the dbuf cache can grow with the ARC target, * unless limited by the tunable dbuf_cache_max_bytes. */ static inline unsigned long dbuf_cache_target_bytes(void) { return (MIN(dbuf_cache_max_bytes, arc_target_bytes() >> dbuf_cache_shift)); } /* * The target size of the dbuf metadata cache can grow with the ARC target, * unless limited by the tunable dbuf_metadata_cache_max_bytes. */ static inline unsigned long dbuf_metadata_cache_target_bytes(void) { return (MIN(dbuf_metadata_cache_max_bytes, arc_target_bytes() >> dbuf_metadata_cache_shift)); } static inline uint64_t dbuf_cache_hiwater_bytes(void) { uint64_t dbuf_cache_target = dbuf_cache_target_bytes(); return (dbuf_cache_target + (dbuf_cache_target * dbuf_cache_hiwater_pct) / 100); } static inline uint64_t dbuf_cache_lowater_bytes(void) { uint64_t dbuf_cache_target = dbuf_cache_target_bytes(); return (dbuf_cache_target - (dbuf_cache_target * dbuf_cache_lowater_pct) / 100); } static inline boolean_t dbuf_cache_above_lowater(void) { return (zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size) > dbuf_cache_lowater_bytes()); } /* * Evict the oldest eligible dbuf from the dbuf cache. */ static void dbuf_evict_one(void) { int idx = multilist_get_random_index(&dbuf_caches[DB_DBUF_CACHE].cache); multilist_sublist_t *mls = multilist_sublist_lock( &dbuf_caches[DB_DBUF_CACHE].cache, idx); ASSERT(!MUTEX_HELD(&dbuf_evict_lock)); dmu_buf_impl_t *db = multilist_sublist_tail(mls); while (db != NULL && mutex_tryenter(&db->db_mtx) == 0) { db = multilist_sublist_prev(mls, db); } DTRACE_PROBE2(dbuf__evict__one, dmu_buf_impl_t *, db, multilist_sublist_t *, mls); if (db != NULL) { multilist_sublist_remove(mls, db); multilist_sublist_unlock(mls); (void) zfs_refcount_remove_many( &dbuf_caches[DB_DBUF_CACHE].size, db->db.db_size, db); DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]); DBUF_STAT_BUMPDOWN(cache_count); DBUF_STAT_DECR(cache_levels_bytes[db->db_level], db->db.db_size); ASSERT3U(db->db_caching_status, ==, DB_DBUF_CACHE); db->db_caching_status = DB_NO_CACHE; dbuf_destroy(db); DBUF_STAT_BUMP(cache_total_evicts); } else { multilist_sublist_unlock(mls); } } /* * The dbuf evict thread is responsible for aging out dbufs from the * cache. Once the cache has reached it's maximum size, dbufs are removed * and destroyed. The eviction thread will continue running until the size * of the dbuf cache is at or below the maximum size. Once the dbuf is aged * out of the cache it is destroyed and becomes eligible for arc eviction. */ /* ARGSUSED */ static void dbuf_evict_thread(void *unused) { callb_cpr_t cpr; CALLB_CPR_INIT(&cpr, &dbuf_evict_lock, callb_generic_cpr, FTAG); mutex_enter(&dbuf_evict_lock); while (!dbuf_evict_thread_exit) { while (!dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait_idle_hires(&dbuf_evict_cv, &dbuf_evict_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); CALLB_CPR_SAFE_END(&cpr, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); /* * Keep evicting as long as we're above the low water mark * for the cache. We do this without holding the locks to * minimize lock contention. */ while (dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { dbuf_evict_one(); } mutex_enter(&dbuf_evict_lock); } dbuf_evict_thread_exit = B_FALSE; cv_broadcast(&dbuf_evict_cv); CALLB_CPR_EXIT(&cpr); /* drops dbuf_evict_lock */ thread_exit(); } /* * Wake up the dbuf eviction thread if the dbuf cache is at its max size. * If the dbuf cache is at its high water mark, then evict a dbuf from the * dbuf cache using the callers context. */ static void dbuf_evict_notify(uint64_t size) { /* * We check if we should evict without holding the dbuf_evict_lock, * because it's OK to occasionally make the wrong decision here, * and grabbing the lock results in massive lock contention. */ if (size > dbuf_cache_target_bytes()) { if (size > dbuf_cache_hiwater_bytes()) dbuf_evict_one(); cv_signal(&dbuf_evict_cv); } } static int dbuf_kstat_update(kstat_t *ksp, int rw) { dbuf_stats_t *ds = ksp->ks_data; - if (rw == KSTAT_WRITE) { + if (rw == KSTAT_WRITE) return (SET_ERROR(EACCES)); - } else { - ds->metadata_cache_size_bytes.value.ui64 = zfs_refcount_count( - &dbuf_caches[DB_DBUF_METADATA_CACHE].size); - ds->cache_size_bytes.value.ui64 = - zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size); - ds->cache_target_bytes.value.ui64 = dbuf_cache_target_bytes(); - ds->cache_hiwater_bytes.value.ui64 = dbuf_cache_hiwater_bytes(); - ds->cache_lowater_bytes.value.ui64 = dbuf_cache_lowater_bytes(); - ds->hash_elements.value.ui64 = dbuf_hash_count; - } + ds->cache_count.value.ui64 = + wmsum_value(&dbuf_sums.cache_count); + ds->cache_size_bytes.value.ui64 = + zfs_refcount_count(&dbuf_caches[DB_DBUF_CACHE].size); + ds->cache_target_bytes.value.ui64 = dbuf_cache_target_bytes(); + ds->cache_hiwater_bytes.value.ui64 = dbuf_cache_hiwater_bytes(); + ds->cache_lowater_bytes.value.ui64 = dbuf_cache_lowater_bytes(); + ds->cache_total_evicts.value.ui64 = + wmsum_value(&dbuf_sums.cache_total_evicts); + for (int i = 0; i < DN_MAX_LEVELS; i++) { + ds->cache_levels[i].value.ui64 = + wmsum_value(&dbuf_sums.cache_levels[i]); + ds->cache_levels_bytes[i].value.ui64 = + wmsum_value(&dbuf_sums.cache_levels_bytes[i]); + } + ds->hash_hits.value.ui64 = + wmsum_value(&dbuf_sums.hash_hits); + ds->hash_misses.value.ui64 = + wmsum_value(&dbuf_sums.hash_misses); + ds->hash_collisions.value.ui64 = + wmsum_value(&dbuf_sums.hash_collisions); + ds->hash_chains.value.ui64 = + wmsum_value(&dbuf_sums.hash_chains); + ds->hash_insert_race.value.ui64 = + wmsum_value(&dbuf_sums.hash_insert_race); + ds->metadata_cache_count.value.ui64 = + wmsum_value(&dbuf_sums.metadata_cache_count); + ds->metadata_cache_size_bytes.value.ui64 = zfs_refcount_count( + &dbuf_caches[DB_DBUF_METADATA_CACHE].size); + ds->metadata_cache_overflow.value.ui64 = + wmsum_value(&dbuf_sums.metadata_cache_overflow); return (0); } void dbuf_init(void) { uint64_t hsize = 1ULL << 16; dbuf_hash_table_t *h = &dbuf_hash_table; int i; /* * The hash table is big enough to fill all of physical memory * with an average block size of zfs_arc_average_blocksize (default 8K). * By default, the table will take up * totalmem * sizeof(void*) / 8K (1MB per GB with 8-byte pointers). */ while (hsize * zfs_arc_average_blocksize < physmem * PAGESIZE) hsize <<= 1; retry: h->hash_table_mask = hsize - 1; #if defined(_KERNEL) /* * Large allocations which do not require contiguous pages * should be using vmem_alloc() in the linux kernel */ h->hash_table = vmem_zalloc(hsize * sizeof (void *), KM_SLEEP); #else h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP); #endif if (h->hash_table == NULL) { /* XXX - we should really return an error instead of assert */ ASSERT(hsize > (1ULL << 10)); hsize >>= 1; goto retry; } dbuf_kmem_cache = kmem_cache_create("dmu_buf_impl_t", sizeof (dmu_buf_impl_t), 0, dbuf_cons, dbuf_dest, NULL, NULL, NULL, 0); for (i = 0; i < DBUF_MUTEXES; i++) mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL); dbuf_stats_init(h); /* * All entries are queued via taskq_dispatch_ent(), so min/maxalloc * configuration is not required. */ dbu_evict_taskq = taskq_create("dbu_evict", 1, defclsyspri, 0, 0, 0); for (dbuf_cached_state_t dcs = 0; dcs < DB_CACHE_MAX; dcs++) { multilist_create(&dbuf_caches[dcs].cache, sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_cache_link), dbuf_cache_multilist_index_func); zfs_refcount_create(&dbuf_caches[dcs].size); } dbuf_evict_thread_exit = B_FALSE; mutex_init(&dbuf_evict_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&dbuf_evict_cv, NULL, CV_DEFAULT, NULL); dbuf_cache_evict_thread = thread_create(NULL, 0, dbuf_evict_thread, NULL, 0, &p0, TS_RUN, minclsyspri); + wmsum_init(&dbuf_sums.cache_count, 0); + wmsum_init(&dbuf_sums.cache_total_evicts, 0); + for (i = 0; i < DN_MAX_LEVELS; i++) { + wmsum_init(&dbuf_sums.cache_levels[i], 0); + wmsum_init(&dbuf_sums.cache_levels_bytes[i], 0); + } + wmsum_init(&dbuf_sums.hash_hits, 0); + wmsum_init(&dbuf_sums.hash_misses, 0); + wmsum_init(&dbuf_sums.hash_collisions, 0); + wmsum_init(&dbuf_sums.hash_chains, 0); + wmsum_init(&dbuf_sums.hash_insert_race, 0); + wmsum_init(&dbuf_sums.metadata_cache_count, 0); + wmsum_init(&dbuf_sums.metadata_cache_overflow, 0); + dbuf_ksp = kstat_create("zfs", 0, "dbufstats", "misc", KSTAT_TYPE_NAMED, sizeof (dbuf_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (dbuf_ksp != NULL) { for (i = 0; i < DN_MAX_LEVELS; i++) { snprintf(dbuf_stats.cache_levels[i].name, KSTAT_STRLEN, "cache_level_%d", i); dbuf_stats.cache_levels[i].data_type = KSTAT_DATA_UINT64; snprintf(dbuf_stats.cache_levels_bytes[i].name, KSTAT_STRLEN, "cache_level_%d_bytes", i); dbuf_stats.cache_levels_bytes[i].data_type = KSTAT_DATA_UINT64; } dbuf_ksp->ks_data = &dbuf_stats; dbuf_ksp->ks_update = dbuf_kstat_update; kstat_install(dbuf_ksp); } } void dbuf_fini(void) { dbuf_hash_table_t *h = &dbuf_hash_table; int i; dbuf_stats_destroy(); for (i = 0; i < DBUF_MUTEXES; i++) mutex_destroy(&h->hash_mutexes[i]); #if defined(_KERNEL) /* * Large allocations which do not require contiguous pages * should be using vmem_free() in the linux kernel */ vmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *)); #else kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *)); #endif kmem_cache_destroy(dbuf_kmem_cache); taskq_destroy(dbu_evict_taskq); mutex_enter(&dbuf_evict_lock); dbuf_evict_thread_exit = B_TRUE; while (dbuf_evict_thread_exit) { cv_signal(&dbuf_evict_cv); cv_wait(&dbuf_evict_cv, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); mutex_destroy(&dbuf_evict_lock); cv_destroy(&dbuf_evict_cv); for (dbuf_cached_state_t dcs = 0; dcs < DB_CACHE_MAX; dcs++) { zfs_refcount_destroy(&dbuf_caches[dcs].size); multilist_destroy(&dbuf_caches[dcs].cache); } if (dbuf_ksp != NULL) { kstat_delete(dbuf_ksp); dbuf_ksp = NULL; } + + wmsum_fini(&dbuf_sums.cache_count); + wmsum_fini(&dbuf_sums.cache_total_evicts); + for (i = 0; i < DN_MAX_LEVELS; i++) { + wmsum_fini(&dbuf_sums.cache_levels[i]); + wmsum_fini(&dbuf_sums.cache_levels_bytes[i]); + } + wmsum_fini(&dbuf_sums.hash_hits); + wmsum_fini(&dbuf_sums.hash_misses); + wmsum_fini(&dbuf_sums.hash_collisions); + wmsum_fini(&dbuf_sums.hash_chains); + wmsum_fini(&dbuf_sums.hash_insert_race); + wmsum_fini(&dbuf_sums.metadata_cache_count); + wmsum_fini(&dbuf_sums.metadata_cache_overflow); } /* * Other stuff. */ #ifdef ZFS_DEBUG static void dbuf_verify(dmu_buf_impl_t *db) { dnode_t *dn; dbuf_dirty_record_t *dr; uint32_t txg_prev; ASSERT(MUTEX_HELD(&db->db_mtx)); if (!(zfs_flags & ZFS_DEBUG_DBUF_VERIFY)) return; ASSERT(db->db_objset != NULL); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dn == NULL) { ASSERT(db->db_parent == NULL); ASSERT(db->db_blkptr == NULL); } else { ASSERT3U(db->db.db_object, ==, dn->dn_object); ASSERT3P(db->db_objset, ==, dn->dn_objset); ASSERT3U(db->db_level, <, dn->dn_nlevels); ASSERT(db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID || !avl_is_empty(&dn->dn_dbufs)); } if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT3U(db->db.db_offset, ==, DMU_BONUS_BLKID); } else if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn != NULL); ASSERT0(db->db.db_offset); } else { ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size); } if ((dr = list_head(&db->db_dirty_records)) != NULL) { ASSERT(dr->dr_dbuf == db); txg_prev = dr->dr_txg; for (dr = list_next(&db->db_dirty_records, dr); dr != NULL; dr = list_next(&db->db_dirty_records, dr)) { ASSERT(dr->dr_dbuf == db); ASSERT(txg_prev > dr->dr_txg); txg_prev = dr->dr_txg; } } /* * We can't assert that db_size matches dn_datablksz because it * can be momentarily different when another thread is doing * dnode_set_blksz(). */ if (db->db_level == 0 && db->db.db_object == DMU_META_DNODE_OBJECT) { dr = db->db_data_pending; /* * It should only be modified in syncing context, so * make sure we only have one copy of the data. */ ASSERT(dr == NULL || dr->dt.dl.dr_data == db->db_buf); } /* verify db->db_blkptr */ if (db->db_blkptr) { if (db->db_parent == dn->dn_dbuf) { /* db is pointed to by the dnode */ /* ASSERT3U(db->db_blkid, <, dn->dn_nblkptr); */ if (DMU_OBJECT_IS_SPECIAL(db->db.db_object)) ASSERT(db->db_parent == NULL); else ASSERT(db->db_parent != NULL); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); } else { /* db is pointed to by an indirect block */ int epb __maybe_unused = db->db_parent->db.db_size >> SPA_BLKPTRSHIFT; ASSERT3U(db->db_parent->db_level, ==, db->db_level+1); ASSERT3U(db->db_parent->db.db_object, ==, db->db.db_object); /* * dnode_grow_indblksz() can make this fail if we don't * have the parent's rwlock. XXX indblksz no longer * grows. safe to do this now? */ if (RW_LOCK_HELD(&db->db_parent->db_rwlock)) { ASSERT3P(db->db_blkptr, ==, ((blkptr_t *)db->db_parent->db.db_data + db->db_blkid % epb)); } } } if ((db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr)) && (db->db_buf == NULL || db->db_buf->b_data) && db->db.db_data && db->db_blkid != DMU_BONUS_BLKID && db->db_state != DB_FILL && !dn->dn_free_txg) { /* * If the blkptr isn't set but they have nonzero data, * it had better be dirty, otherwise we'll lose that * data when we evict this buffer. * * There is an exception to this rule for indirect blocks; in * this case, if the indirect block is a hole, we fill in a few * fields on each of the child blocks (importantly, birth time) * to prevent hole birth times from being lost when you * partially fill in a hole. */ if (db->db_dirtycnt == 0) { if (db->db_level == 0) { uint64_t *buf = db->db.db_data; int i; for (i = 0; i < db->db.db_size >> 3; i++) { ASSERT(buf[i] == 0); } } else { blkptr_t *bps = db->db.db_data; ASSERT3U(1 << DB_DNODE(db)->dn_indblkshift, ==, db->db.db_size); /* * We want to verify that all the blkptrs in the * indirect block are holes, but we may have * automatically set up a few fields for them. * We iterate through each blkptr and verify * they only have those fields set. */ for (int i = 0; i < db->db.db_size / sizeof (blkptr_t); i++) { blkptr_t *bp = &bps[i]; ASSERT(ZIO_CHECKSUM_IS_ZERO( &bp->blk_cksum)); ASSERT( DVA_IS_EMPTY(&bp->blk_dva[0]) && DVA_IS_EMPTY(&bp->blk_dva[1]) && DVA_IS_EMPTY(&bp->blk_dva[2])); ASSERT0(bp->blk_fill); ASSERT0(bp->blk_pad[0]); ASSERT0(bp->blk_pad[1]); ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(BP_IS_HOLE(bp)); ASSERT0(bp->blk_phys_birth); } } } } DB_DNODE_EXIT(db); } #endif static void dbuf_clear_data(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); dbuf_evict_user(db); ASSERT3P(db->db_buf, ==, NULL); db->db.db_data = NULL; if (db->db_state != DB_NOFILL) { db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "clear data"); } } static void dbuf_set_data(dmu_buf_impl_t *db, arc_buf_t *buf) { ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(buf != NULL); db->db_buf = buf; ASSERT(buf->b_data != NULL); db->db.db_data = buf->b_data; } static arc_buf_t * dbuf_alloc_arcbuf_from_arcbuf(dmu_buf_impl_t *db, arc_buf_t *data) { objset_t *os = db->db_objset; spa_t *spa = os->os_spa; arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); enum zio_compress compress_type; uint8_t complevel; int psize, lsize; psize = arc_buf_size(data); lsize = arc_buf_lsize(data); compress_type = arc_get_compression(data); complevel = arc_get_complevel(data); if (arc_is_encrypted(data)) { boolean_t byteorder; uint8_t salt[ZIO_DATA_SALT_LEN]; uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t mac[ZIO_DATA_MAC_LEN]; dnode_t *dn = DB_DNODE(db); arc_get_raw_params(data, &byteorder, salt, iv, mac); data = arc_alloc_raw_buf(spa, db, dmu_objset_id(os), byteorder, salt, iv, mac, dn->dn_type, psize, lsize, compress_type, complevel); } else if (compress_type != ZIO_COMPRESS_OFF) { ASSERT3U(type, ==, ARC_BUFC_DATA); data = arc_alloc_compressed_buf(spa, db, psize, lsize, compress_type, complevel); } else { data = arc_alloc_buf(spa, db, type, psize); } return (data); } static arc_buf_t * dbuf_alloc_arcbuf(dmu_buf_impl_t *db) { spa_t *spa = db->db_objset->os_spa; return (arc_alloc_buf(spa, db, DBUF_GET_BUFC_TYPE(db), db->db.db_size)); } /* * Loan out an arc_buf for read. Return the loaned arc_buf. */ arc_buf_t * dbuf_loan_arcbuf(dmu_buf_impl_t *db) { arc_buf_t *abuf; ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); if (arc_released(db->db_buf) || zfs_refcount_count(&db->db_holds) > 1) { int blksz = db->db.db_size; spa_t *spa = db->db_objset->os_spa; mutex_exit(&db->db_mtx); abuf = arc_loan_buf(spa, B_FALSE, blksz); bcopy(db->db.db_data, abuf->b_data, blksz); } else { abuf = db->db_buf; arc_loan_inuse_buf(abuf, db); db->db_buf = NULL; dbuf_clear_data(db); mutex_exit(&db->db_mtx); } return (abuf); } /* * Calculate which level n block references the data at the level 0 offset * provided. */ uint64_t dbuf_whichblock(const dnode_t *dn, const int64_t level, const uint64_t offset) { if (dn->dn_datablkshift != 0 && dn->dn_indblkshift != 0) { /* * The level n blkid is equal to the level 0 blkid divided by * the number of level 0s in a level n block. * * The level 0 blkid is offset >> datablkshift = * offset / 2^datablkshift. * * The number of level 0s in a level n is the number of block * pointers in an indirect block, raised to the power of level. * This is 2^(indblkshift - SPA_BLKPTRSHIFT)^level = * 2^(level*(indblkshift - SPA_BLKPTRSHIFT)). * * Thus, the level n blkid is: offset / * ((2^datablkshift)*(2^(level*(indblkshift-SPA_BLKPTRSHIFT)))) * = offset / 2^(datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) * = offset >> (datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) */ const unsigned exp = dn->dn_datablkshift + level * (dn->dn_indblkshift - SPA_BLKPTRSHIFT); if (exp >= 8 * sizeof (offset)) { /* This only happens on the highest indirection level */ ASSERT3U(level, ==, dn->dn_nlevels - 1); return (0); } ASSERT3U(exp, <, 8 * sizeof (offset)); return (offset >> exp); } else { ASSERT3U(offset, <, dn->dn_datablksz); return (0); } } /* * This function is used to lock the parent of the provided dbuf. This should be * used when modifying or reading db_blkptr. */ db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, void *tag) { enum db_lock_type ret = DLT_NONE; if (db->db_parent != NULL) { rw_enter(&db->db_parent->db_rwlock, rw); ret = DLT_PARENT; } else if (dmu_objset_ds(db->db_objset) != NULL) { rrw_enter(&dmu_objset_ds(db->db_objset)->ds_bp_rwlock, rw, tag); ret = DLT_OBJSET; } /* * We only return a DLT_NONE lock when it's the top-most indirect block * of the meta-dnode of the MOS. */ return (ret); } /* * We need to pass the lock type in because it's possible that the block will * move from being the topmost indirect block in a dnode (and thus, have no * parent) to not the top-most via an indirection increase. This would cause a * panic if we didn't pass the lock type in. */ void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, void *tag) { if (type == DLT_PARENT) rw_exit(&db->db_parent->db_rwlock); else if (type == DLT_OBJSET) rrw_exit(&dmu_objset_ds(db->db_objset)->ds_bp_rwlock, tag); } static void dbuf_read_done(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; mutex_enter(&db->db_mtx); ASSERT3U(db->db_state, ==, DB_READ); /* * All reads are synchronous, so we must have a hold on the dbuf */ ASSERT(zfs_refcount_count(&db->db_holds) > 0); ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); if (buf == NULL) { /* i/o error */ ASSERT(zio == NULL || zio->io_error != 0); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT3P(db->db_buf, ==, NULL); db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "i/o error"); } else if (db->db_level == 0 && db->db_freed_in_flight) { /* freed in flight */ ASSERT(zio == NULL || zio->io_error == 0); arc_release(buf, db); bzero(buf->b_data, db->db.db_size); arc_buf_freeze(buf); db->db_freed_in_flight = FALSE; dbuf_set_data(db, buf); db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "freed in flight"); } else { /* success */ ASSERT(zio == NULL || zio->io_error == 0); dbuf_set_data(db, buf); db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "successful read"); } cv_broadcast(&db->db_changed); dbuf_rele_and_unlock(db, NULL, B_FALSE); } /* * Shortcut for performing reads on bonus dbufs. Returns * an error if we fail to verify the dnode associated with * a decrypted block. Otherwise success. */ static int dbuf_read_bonus(dmu_buf_impl_t *db, dnode_t *dn, uint32_t flags) { int bonuslen, max_bonuslen, err; err = dbuf_read_verify_dnode_crypt(db, flags); if (err) return (err); bonuslen = MIN(dn->dn_bonuslen, dn->dn_phys->dn_bonuslen); max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(DB_DNODE_HELD(db)); ASSERT3U(bonuslen, <=, db->db.db_size); db->db.db_data = kmem_alloc(max_bonuslen, KM_SLEEP); arc_space_consume(max_bonuslen, ARC_SPACE_BONUS); if (bonuslen < max_bonuslen) bzero(db->db.db_data, max_bonuslen); if (bonuslen) bcopy(DN_BONUS(dn->dn_phys), db->db.db_data, bonuslen); db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "bonus buffer filled"); return (0); } static void dbuf_handle_indirect_hole(dmu_buf_impl_t *db, dnode_t *dn) { blkptr_t *bps = db->db.db_data; uint32_t indbs = 1ULL << dn->dn_indblkshift; int n_bps = indbs >> SPA_BLKPTRSHIFT; for (int i = 0; i < n_bps; i++) { blkptr_t *bp = &bps[i]; ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, indbs); BP_SET_LSIZE(bp, BP_GET_LEVEL(db->db_blkptr) == 1 ? dn->dn_datablksz : BP_GET_LSIZE(db->db_blkptr)); BP_SET_TYPE(bp, BP_GET_TYPE(db->db_blkptr)); BP_SET_LEVEL(bp, BP_GET_LEVEL(db->db_blkptr) - 1); BP_SET_BIRTH(bp, db->db_blkptr->blk_birth, 0); } } /* * Handle reads on dbufs that are holes, if necessary. This function * requires that the dbuf's mutex is held. Returns success (0) if action * was taken, ENOENT if no action was taken. */ static int dbuf_read_hole(dmu_buf_impl_t *db, dnode_t *dn, uint32_t flags) { ASSERT(MUTEX_HELD(&db->db_mtx)); int is_hole = db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr); /* * For level 0 blocks only, if the above check fails: * Recheck BP_IS_HOLE() after dnode_block_freed() in case dnode_sync() * processes the delete record and clears the bp while we are waiting * for the dn_mtx (resulting in a "no" from block_freed). */ if (!is_hole && db->db_level == 0) { is_hole = dnode_block_freed(dn, db->db_blkid) || BP_IS_HOLE(db->db_blkptr); } if (is_hole) { dbuf_set_data(db, dbuf_alloc_arcbuf(db)); bzero(db->db.db_data, db->db.db_size); if (db->db_blkptr != NULL && db->db_level > 0 && BP_IS_HOLE(db->db_blkptr) && db->db_blkptr->blk_birth != 0) { dbuf_handle_indirect_hole(db, dn); } db->db_state = DB_CACHED; DTRACE_SET_STATE(db, "hole read satisfied"); return (0); } return (ENOENT); } /* * This function ensures that, when doing a decrypting read of a block, * we make sure we have decrypted the dnode associated with it. We must do * this so that we ensure we are fully authenticating the checksum-of-MACs * tree from the root of the objset down to this block. Indirect blocks are * always verified against their secure checksum-of-MACs assuming that the * dnode containing them is correct. Now that we are doing a decrypting read, * we can be sure that the key is loaded and verify that assumption. This is * especially important considering that we always read encrypted dnode * blocks as raw data (without verifying their MACs) to start, and * decrypt / authenticate them when we need to read an encrypted bonus buffer. */ static int dbuf_read_verify_dnode_crypt(dmu_buf_impl_t *db, uint32_t flags) { int err = 0; objset_t *os = db->db_objset; arc_buf_t *dnode_abuf; dnode_t *dn; zbookmark_phys_t zb; ASSERT(MUTEX_HELD(&db->db_mtx)); if (!os->os_encrypted || os->os_raw_receive || (flags & DB_RF_NO_DECRYPT) != 0) return (0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dnode_abuf = (dn->dn_dbuf != NULL) ? dn->dn_dbuf->db_buf : NULL; if (dnode_abuf == NULL || !arc_is_encrypted(dnode_abuf)) { DB_DNODE_EXIT(db); return (0); } SET_BOOKMARK(&zb, dmu_objset_id(os), DMU_META_DNODE_OBJECT, 0, dn->dn_dbuf->db_blkid); err = arc_untransform(dnode_abuf, os->os_spa, &zb, B_TRUE); /* * An error code of EACCES tells us that the key is still not * available. This is ok if we are only reading authenticated * (and therefore non-encrypted) blocks. */ if (err == EACCES && ((db->db_blkid != DMU_BONUS_BLKID && !DMU_OT_IS_ENCRYPTED(dn->dn_type)) || (db->db_blkid == DMU_BONUS_BLKID && !DMU_OT_IS_ENCRYPTED(dn->dn_bonustype)))) err = 0; DB_DNODE_EXIT(db); return (err); } /* * Drops db_mtx and the parent lock specified by dblt and tag before * returning. */ static int dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags, db_lock_type_t dblt, void *tag) { dnode_t *dn; zbookmark_phys_t zb; uint32_t aflags = ARC_FLAG_NOWAIT; int err, zio_flags; err = zio_flags = 0; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_state == DB_UNCACHED); ASSERT(db->db_buf == NULL); ASSERT(db->db_parent == NULL || RW_LOCK_HELD(&db->db_parent->db_rwlock)); if (db->db_blkid == DMU_BONUS_BLKID) { err = dbuf_read_bonus(db, dn, flags); goto early_unlock; } err = dbuf_read_hole(db, dn, flags); if (err == 0) goto early_unlock; /* * Any attempt to read a redacted block should result in an error. This * will never happen under normal conditions, but can be useful for * debugging purposes. */ if (BP_IS_REDACTED(db->db_blkptr)) { ASSERT(dsl_dataset_feature_is_active( db->db_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); err = SET_ERROR(EIO); goto early_unlock; } SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset), db->db.db_object, db->db_level, db->db_blkid); /* * All bps of an encrypted os should have the encryption bit set. * If this is not true it indicates tampering and we report an error. */ if (db->db_objset->os_encrypted && !BP_USES_CRYPT(db->db_blkptr)) { spa_log_error(db->db_objset->os_spa, &zb); zfs_panic_recover("unencrypted block in encrypted " "object set %llu", dmu_objset_id(db->db_objset)); err = SET_ERROR(EIO); goto early_unlock; } err = dbuf_read_verify_dnode_crypt(db, flags); if (err != 0) goto early_unlock; DB_DNODE_EXIT(db); db->db_state = DB_READ; DTRACE_SET_STATE(db, "read issued"); mutex_exit(&db->db_mtx); if (DBUF_IS_L2CACHEABLE(db)) aflags |= ARC_FLAG_L2CACHE; dbuf_add_ref(db, NULL); zio_flags = (flags & DB_RF_CANFAIL) ? ZIO_FLAG_CANFAIL : ZIO_FLAG_MUSTSUCCEED; if ((flags & DB_RF_NO_DECRYPT) && BP_IS_PROTECTED(db->db_blkptr)) zio_flags |= ZIO_FLAG_RAW; /* * The zio layer will copy the provided blkptr later, but we need to * do this now so that we can release the parent's rwlock. We have to * do that now so that if dbuf_read_done is called synchronously (on * an l1 cache hit) we don't acquire the db_mtx while holding the * parent's rwlock, which would be a lock ordering violation. */ blkptr_t bp = *db->db_blkptr; dmu_buf_unlock_parent(db, dblt, tag); (void) arc_read(zio, db->db_objset->os_spa, &bp, dbuf_read_done, db, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); return (err); early_unlock: DB_DNODE_EXIT(db); mutex_exit(&db->db_mtx); dmu_buf_unlock_parent(db, dblt, tag); return (err); } /* * This is our just-in-time copy function. It makes a copy of buffers that * have been modified in a previous transaction group before we access them in * the current active group. * * This function is used in three places: when we are dirtying a buffer for the * first time in a txg, when we are freeing a range in a dnode that includes * this buffer, and when we are accessing a buffer which was received compressed * and later referenced in a WRITE_BYREF record. * * Note that when we are called from dbuf_free_range() we do not put a hold on * the buffer, we just traverse the active dbuf list for the dnode. */ static void dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg) { dbuf_dirty_record_t *dr = list_head(&db->db_dirty_records); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db.db_data != NULL); ASSERT(db->db_level == 0); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT); if (dr == NULL || (dr->dt.dl.dr_data != ((db->db_blkid == DMU_BONUS_BLKID) ? db->db.db_data : db->db_buf))) return; /* * If the last dirty record for this dbuf has not yet synced * and its referencing the dbuf data, either: * reset the reference to point to a new copy, * or (if there a no active holders) * just null out the current db_data pointer. */ ASSERT3U(dr->dr_txg, >=, txg - 2); if (db->db_blkid == DMU_BONUS_BLKID) { dnode_t *dn = DB_DNODE(db); int bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); dr->dt.dl.dr_data = kmem_alloc(bonuslen, KM_SLEEP); arc_space_consume(bonuslen, ARC_SPACE_BONUS); bcopy(db->db.db_data, dr->dt.dl.dr_data, bonuslen); } else if (zfs_refcount_count(&db->db_holds) > db->db_dirtycnt) { arc_buf_t *buf = dbuf_alloc_arcbuf_from_arcbuf(db, db->db_buf); dr->dt.dl.dr_data = buf; bcopy(db->db.db_data, buf->b_data, arc_buf_size(buf)); } else { db->db_buf = NULL; dbuf_clear_data(db); } } int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags) { int err = 0; boolean_t prefetch; dnode_t *dn; /* * We don't have to hold the mutex to check db_state because it * can't be freed while we have a hold on the buffer. */ ASSERT(!zfs_refcount_is_zero(&db->db_holds)); if (db->db_state == DB_NOFILL) return (SET_ERROR(EIO)); DB_DNODE_ENTER(db); dn = DB_DNODE(db); prefetch = db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && (flags & DB_RF_NOPREFETCH) == 0 && dn != NULL && DBUF_IS_CACHEABLE(db); mutex_enter(&db->db_mtx); if (db->db_state == DB_CACHED) { spa_t *spa = dn->dn_objset->os_spa; /* * Ensure that this block's dnode has been decrypted if * the caller has requested decrypted data. */ err = dbuf_read_verify_dnode_crypt(db, flags); /* * If the arc buf is compressed or encrypted and the caller * requested uncompressed data, we need to untransform it * before returning. We also call arc_untransform() on any * unauthenticated blocks, which will verify their MAC if * the key is now available. */ if (err == 0 && db->db_buf != NULL && (flags & DB_RF_NO_DECRYPT) == 0 && (arc_is_encrypted(db->db_buf) || arc_is_unauthenticated(db->db_buf) || arc_get_compression(db->db_buf) != ZIO_COMPRESS_OFF)) { zbookmark_phys_t zb; SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset), db->db.db_object, db->db_level, db->db_blkid); dbuf_fix_old_data(db, spa_syncing_txg(spa)); err = arc_untransform(db->db_buf, spa, &zb, B_FALSE); dbuf_set_data(db, db->db_buf); } mutex_exit(&db->db_mtx); if (err == 0 && prefetch) { dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE, B_FALSE, flags & DB_RF_HAVESTRUCT); } DB_DNODE_EXIT(db); DBUF_STAT_BUMP(hash_hits); } else if (db->db_state == DB_UNCACHED) { spa_t *spa = dn->dn_objset->os_spa; boolean_t need_wait = B_FALSE; db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_READER, FTAG); if (zio == NULL && db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr)) { zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); need_wait = B_TRUE; } err = dbuf_read_impl(db, zio, flags, dblt, FTAG); /* * dbuf_read_impl has dropped db_mtx and our parent's rwlock * for us */ if (!err && prefetch) { dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE, db->db_state != DB_CACHED, flags & DB_RF_HAVESTRUCT); } DB_DNODE_EXIT(db); DBUF_STAT_BUMP(hash_misses); /* * If we created a zio_root we must execute it to avoid * leaking it, even if it isn't attached to any work due * to an error in dbuf_read_impl(). */ if (need_wait) { if (err == 0) err = zio_wait(zio); else VERIFY0(zio_wait(zio)); } } else { /* * Another reader came in while the dbuf was in flight * between UNCACHED and CACHED. Either a writer will finish * writing the buffer (sending the dbuf to CACHED) or the * first reader's request will reach the read_done callback * and send the dbuf to CACHED. Otherwise, a failure * occurred and the dbuf went to UNCACHED. */ mutex_exit(&db->db_mtx); if (prefetch) { dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE, B_TRUE, flags & DB_RF_HAVESTRUCT); } DB_DNODE_EXIT(db); DBUF_STAT_BUMP(hash_misses); /* Skip the wait per the caller's request. */ if ((flags & DB_RF_NEVERWAIT) == 0) { mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) { ASSERT(db->db_state == DB_READ || (flags & DB_RF_HAVESTRUCT) == 0); DTRACE_PROBE2(blocked__read, dmu_buf_impl_t *, db, zio_t *, zio); cv_wait(&db->db_changed, &db->db_mtx); } if (db->db_state == DB_UNCACHED) err = SET_ERROR(EIO); mutex_exit(&db->db_mtx); } } return (err); } static void dbuf_noread(dmu_buf_impl_t *db) { ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); if (db->db_state == DB_UNCACHED) { ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); dbuf_set_data(db, dbuf_alloc_arcbuf(db)); db->db_state = DB_FILL; DTRACE_SET_STATE(db, "assigning filled buffer"); } else if (db->db_state == DB_NOFILL) { dbuf_clear_data(db); } else { ASSERT3U(db->db_state, ==, DB_CACHED); } mutex_exit(&db->db_mtx); } void dbuf_unoverride(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *bp = &dr->dt.dl.dr_overridden_by; uint64_t txg = dr->dr_txg; ASSERT(MUTEX_HELD(&db->db_mtx)); /* * This assert is valid because dmu_sync() expects to be called by * a zilog's get_data while holding a range lock. This call only * comes from dbuf_dirty() callers who must also hold a range lock. */ ASSERT(dr->dt.dl.dr_override_state != DR_IN_DMU_SYNC); ASSERT(db->db_level == 0); if (db->db_blkid == DMU_BONUS_BLKID || dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN) return; ASSERT(db->db_data_pending != dr); /* free this block */ if (!BP_IS_HOLE(bp) && !dr->dt.dl.dr_nopwrite) zio_free(db->db_objset->os_spa, txg, bp); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; dr->dt.dl.dr_nopwrite = B_FALSE; dr->dt.dl.dr_has_raw_params = B_FALSE; /* * Release the already-written buffer, so we leave it in * a consistent dirty state. Note that all callers are * modifying the buffer, so they will immediately do * another (redundant) arc_release(). Therefore, leave * the buf thawed to save the effort of freezing & * immediately re-thawing it. */ arc_release(dr->dt.dl.dr_data, db); } /* * Evict (if its unreferenced) or clear (if its referenced) any level-0 * data blocks in the free range, so that any future readers will find * empty blocks. */ void dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid, dmu_tx_t *tx) { dmu_buf_impl_t *db_search; dmu_buf_impl_t *db, *db_next; uint64_t txg = tx->tx_txg; avl_index_t where; dbuf_dirty_record_t *dr; if (end_blkid > dn->dn_maxblkid && !(start_blkid == DMU_SPILL_BLKID || end_blkid == DMU_SPILL_BLKID)) end_blkid = dn->dn_maxblkid; dprintf_dnode(dn, "start=%llu end=%llu\n", start_blkid, end_blkid); db_search = kmem_alloc(sizeof (dmu_buf_impl_t), KM_SLEEP); db_search->db_level = 0; db_search->db_blkid = start_blkid; db_search->db_state = DB_SEARCH; mutex_enter(&dn->dn_dbufs_mtx); db = avl_find(&dn->dn_dbufs, db_search, &where); ASSERT3P(db, ==, NULL); db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); for (; db != NULL; db = db_next) { db_next = AVL_NEXT(&dn->dn_dbufs, db); ASSERT(db->db_blkid != DMU_BONUS_BLKID); if (db->db_level != 0 || db->db_blkid > end_blkid) { break; } ASSERT3U(db->db_blkid, >=, start_blkid); /* found a level 0 buffer in the range */ mutex_enter(&db->db_mtx); if (dbuf_undirty(db, tx)) { /* mutex has been dropped and dbuf destroyed */ continue; } if (db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL || db->db_state == DB_EVICTING) { ASSERT(db->db.db_data == NULL); mutex_exit(&db->db_mtx); continue; } if (db->db_state == DB_READ || db->db_state == DB_FILL) { /* will be handled in dbuf_read_done or dbuf_rele */ db->db_freed_in_flight = TRUE; mutex_exit(&db->db_mtx); continue; } if (zfs_refcount_count(&db->db_holds) == 0) { ASSERT(db->db_buf); dbuf_destroy(db); continue; } /* The dbuf is referenced */ dr = list_head(&db->db_dirty_records); if (dr != NULL) { if (dr->dr_txg == txg) { /* * This buffer is "in-use", re-adjust the file * size to reflect that this buffer may * contain new data when we sync. */ if (db->db_blkid != DMU_SPILL_BLKID && db->db_blkid > dn->dn_maxblkid) dn->dn_maxblkid = db->db_blkid; dbuf_unoverride(dr); } else { /* * This dbuf is not dirty in the open context. * Either uncache it (if its not referenced in * the open context) or reset its contents to * empty. */ dbuf_fix_old_data(db, txg); } } /* clear the contents if its cached */ if (db->db_state == DB_CACHED) { ASSERT(db->db.db_data != NULL); arc_release(db->db_buf, db); rw_enter(&db->db_rwlock, RW_WRITER); bzero(db->db.db_data, db->db.db_size); rw_exit(&db->db_rwlock); arc_buf_freeze(db->db_buf); } mutex_exit(&db->db_mtx); } kmem_free(db_search, sizeof (dmu_buf_impl_t)); mutex_exit(&dn->dn_dbufs_mtx); } void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx) { arc_buf_t *buf, *old_buf; dbuf_dirty_record_t *dr; int osize = db->db.db_size; arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dnode_t *dn; ASSERT(db->db_blkid != DMU_BONUS_BLKID); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* * XXX we should be doing a dbuf_read, checking the return * value and returning that up to our callers */ dmu_buf_will_dirty(&db->db, tx); /* create the data buffer for the new block */ buf = arc_alloc_buf(dn->dn_objset->os_spa, db, type, size); /* copy old block data to the new block */ old_buf = db->db_buf; bcopy(old_buf->b_data, buf->b_data, MIN(osize, size)); /* zero the remainder */ if (size > osize) bzero((uint8_t *)buf->b_data + osize, size - osize); mutex_enter(&db->db_mtx); dbuf_set_data(db, buf); arc_buf_destroy(old_buf, db); db->db.db_size = size; dr = list_head(&db->db_dirty_records); /* dirty record added by dmu_buf_will_dirty() */ VERIFY(dr != NULL); if (db->db_level == 0) dr->dt.dl.dr_data = buf; ASSERT3U(dr->dr_txg, ==, tx->tx_txg); ASSERT3U(dr->dr_accounted, ==, osize); dr->dr_accounted = size; mutex_exit(&db->db_mtx); dmu_objset_willuse_space(dn->dn_objset, size - osize, tx); DB_DNODE_EXIT(db); } void dbuf_release_bp(dmu_buf_impl_t *db) { objset_t *os __maybe_unused = db->db_objset; ASSERT(dsl_pool_sync_context(dmu_objset_pool(os))); ASSERT(arc_released(os->os_phys_buf) || list_link_active(&os->os_dsl_dataset->ds_synced_link)); ASSERT(db->db_parent == NULL || arc_released(db->db_parent->db_buf)); (void) arc_release(db->db_buf, db); } /* * We already have a dirty record for this TXG, and we are being * dirtied again. */ static void dbuf_redirty(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID) { /* * If this buffer has already been written out, * we now need to reset its state. */ dbuf_unoverride(dr); if (db->db.db_object != DMU_META_DNODE_OBJECT && db->db_state != DB_NOFILL) { /* Already released on initial dirty, so just thaw. */ ASSERT(arc_released(db->db_buf)); arc_buf_thaw(db->db_buf); } } } dbuf_dirty_record_t * dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx) { rw_enter(&dn->dn_struct_rwlock, RW_READER); IMPLY(dn->dn_objset->os_raw_receive, dn->dn_maxblkid >= blkid); dnode_new_blkid(dn, blkid, tx, B_TRUE, B_FALSE); ASSERT(dn->dn_maxblkid >= blkid); dbuf_dirty_record_t *dr = kmem_zalloc(sizeof (*dr), KM_SLEEP); list_link_init(&dr->dr_dirty_node); list_link_init(&dr->dr_dbuf_node); dr->dr_dnode = dn; dr->dr_txg = tx->tx_txg; dr->dt.dll.dr_blkid = blkid; dr->dr_accounted = dn->dn_datablksz; /* * There should not be any dbuf for the block that we're dirtying. * Otherwise the buffer contents could be inconsistent between the * dbuf and the lightweight dirty record. */ ASSERT3P(NULL, ==, dbuf_find(dn->dn_objset, dn->dn_object, 0, blkid)); mutex_enter(&dn->dn_mtx); int txgoff = tx->tx_txg & TXG_MASK; if (dn->dn_free_ranges[txgoff] != NULL) { range_tree_clear(dn->dn_free_ranges[txgoff], blkid, 1); } if (dn->dn_nlevels == 1) { ASSERT3U(blkid, <, dn->dn_nblkptr); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); rw_exit(&dn->dn_struct_rwlock); dnode_setdirty(dn, tx); } else { mutex_exit(&dn->dn_mtx); int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; dmu_buf_impl_t *parent_db = dbuf_hold_level(dn, 1, blkid >> epbs, FTAG); rw_exit(&dn->dn_struct_rwlock); if (parent_db == NULL) { kmem_free(dr, sizeof (*dr)); return (NULL); } int err = dbuf_read(parent_db, NULL, (DB_RF_NOPREFETCH | DB_RF_CANFAIL)); if (err != 0) { dbuf_rele(parent_db, FTAG); kmem_free(dr, sizeof (*dr)); return (NULL); } dbuf_dirty_record_t *parent_dr = dbuf_dirty(parent_db, tx); dbuf_rele(parent_db, FTAG); mutex_enter(&parent_dr->dt.di.dr_mtx); ASSERT3U(parent_dr->dr_txg, ==, tx->tx_txg); list_insert_tail(&parent_dr->dt.di.dr_children, dr); mutex_exit(&parent_dr->dt.di.dr_mtx); dr->dr_parent = parent_dr; } dmu_objset_willuse_space(dn->dn_objset, dr->dr_accounted, tx); return (dr); } dbuf_dirty_record_t * dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; objset_t *os; dbuf_dirty_record_t *dr, *dr_next, *dr_head; int txgoff = tx->tx_txg & TXG_MASK; boolean_t drop_struct_rwlock = B_FALSE; ASSERT(tx->tx_txg != 0); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); DMU_TX_DIRTY_BUF(tx, db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* * Shouldn't dirty a regular buffer in syncing context. Private * objects may be dirtied in syncing context, but only if they * were already pre-dirtied in open context. */ #ifdef ZFS_DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_enter(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); } ASSERT(!dmu_tx_is_syncing(tx) || BP_IS_HOLE(dn->dn_objset->os_rootbp) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_objset->os_dsl_dataset == NULL); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif /* * We make this assert for private objects as well, but after we * check if we're already dirty. They are allowed to re-dirty * in syncing context. */ ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); mutex_enter(&db->db_mtx); /* * XXX make this true for indirects too? The problem is that * transactions created with dmu_tx_create_assigned() from * syncing context don't bother holding ahead. */ ASSERT(db->db_level != 0 || db->db_state == DB_CACHED || db->db_state == DB_FILL || db->db_state == DB_NOFILL); mutex_enter(&dn->dn_mtx); dnode_set_dirtyctx(dn, tx, db); if (tx->tx_txg > dn->dn_dirty_txg) dn->dn_dirty_txg = tx->tx_txg; mutex_exit(&dn->dn_mtx); if (db->db_blkid == DMU_SPILL_BLKID) dn->dn_have_spill = B_TRUE; /* * If this buffer is already dirty, we're done. */ dr_head = list_head(&db->db_dirty_records); ASSERT(dr_head == NULL || dr_head->dr_txg <= tx->tx_txg || db->db.db_object == DMU_META_DNODE_OBJECT); dr_next = dbuf_find_dirty_lte(db, tx->tx_txg); if (dr_next && dr_next->dr_txg == tx->tx_txg) { DB_DNODE_EXIT(db); dbuf_redirty(dr_next); mutex_exit(&db->db_mtx); return (dr_next); } /* * Only valid if not already dirty. */ ASSERT(dn->dn_object == 0 || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); ASSERT3U(dn->dn_nlevels, >, db->db_level); /* * We should only be dirtying in syncing context if it's the * mos or we're initializing the os or it's a special object. * However, we are allowed to dirty in syncing context provided * we already dirtied it in open context. Hence we must make * this assertion only if we're not already dirty. */ os = dn->dn_objset; VERIFY3U(tx->tx_txg, <=, spa_final_dirty_txg(os->os_spa)); #ifdef ZFS_DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp)); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif ASSERT(db->db.db_size != 0); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); if (db->db_blkid != DMU_BONUS_BLKID) { dmu_objset_willuse_space(os, db->db.db_size, tx); } /* * If this buffer is dirty in an old transaction group we need * to make a copy of it so that the changes we make in this * transaction group won't leak out when we sync the older txg. */ dr = kmem_zalloc(sizeof (dbuf_dirty_record_t), KM_SLEEP); list_link_init(&dr->dr_dirty_node); list_link_init(&dr->dr_dbuf_node); dr->dr_dnode = dn; if (db->db_level == 0) { void *data_old = db->db_buf; if (db->db_state != DB_NOFILL) { if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db.db_data; } else if (db->db.db_object != DMU_META_DNODE_OBJECT) { /* * Release the data buffer from the cache so * that we can modify it without impacting * possible other users of this cached data * block. Note that indirect blocks and * private objects are not released until the * syncing state (since they are only modified * then). */ arc_release(db->db_buf, db); dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db_buf; } ASSERT(data_old != NULL); } dr->dt.dl.dr_data = data_old; } else { mutex_init(&dr->dt.di.dr_mtx, NULL, MUTEX_NOLOCKDEP, NULL); list_create(&dr->dt.di.dr_children, sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dirty_node)); } if (db->db_blkid != DMU_BONUS_BLKID) dr->dr_accounted = db->db.db_size; dr->dr_dbuf = db; dr->dr_txg = tx->tx_txg; list_insert_before(&db->db_dirty_records, dr_next, dr); /* * We could have been freed_in_flight between the dbuf_noread * and dbuf_dirty. We win, as though the dbuf_noread() had * happened after the free. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && db->db_blkid != DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_free_ranges[txgoff] != NULL) { range_tree_clear(dn->dn_free_ranges[txgoff], db->db_blkid, 1); } mutex_exit(&dn->dn_mtx); db->db_freed_in_flight = FALSE; } /* * This buffer is now part of this txg */ dbuf_add_ref(db, (void *)(uintptr_t)tx->tx_txg); db->db_dirtycnt += 1; ASSERT3U(db->db_dirtycnt, <=, 3); mutex_exit(&db->db_mtx); if (db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { rw_enter(&dn->dn_struct_rwlock, RW_READER); drop_struct_rwlock = B_TRUE; } /* * If we are overwriting a dedup BP, then unless it is snapshotted, * when we get to syncing context we will need to decrement its * refcount in the DDT. Prefetch the relevant DDT block so that * syncing context won't have to wait for the i/o. */ if (db->db_blkptr != NULL) { db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_READER, FTAG); ddt_prefetch(os->os_spa, db->db_blkptr); dmu_buf_unlock_parent(db, dblt, FTAG); } /* * We need to hold the dn_struct_rwlock to make this assertion, * because it protects dn_phys / dn_next_nlevels from changing. */ ASSERT((dn->dn_phys->dn_nlevels == 0 && db->db_level == 0) || dn->dn_phys->dn_nlevels > db->db_level || dn->dn_next_nlevels[txgoff] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-1) & TXG_MASK] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-2) & TXG_MASK] > db->db_level); if (db->db_level == 0) { ASSERT(!db->db_objset->os_raw_receive || dn->dn_maxblkid >= db->db_blkid); dnode_new_blkid(dn, db->db_blkid, tx, drop_struct_rwlock, B_FALSE); ASSERT(dn->dn_maxblkid >= db->db_blkid); } if (db->db_level+1 < dn->dn_nlevels) { dmu_buf_impl_t *parent = db->db_parent; dbuf_dirty_record_t *di; int parent_held = FALSE; if (db->db_parent == NULL || db->db_parent == dn->dn_dbuf) { int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; parent = dbuf_hold_level(dn, db->db_level + 1, db->db_blkid >> epbs, FTAG); ASSERT(parent != NULL); parent_held = TRUE; } if (drop_struct_rwlock) rw_exit(&dn->dn_struct_rwlock); ASSERT3U(db->db_level + 1, ==, parent->db_level); di = dbuf_dirty(parent, tx); if (parent_held) dbuf_rele(parent, FTAG); mutex_enter(&db->db_mtx); /* * Since we've dropped the mutex, it's possible that * dbuf_undirty() might have changed this out from under us. */ if (list_head(&db->db_dirty_records) == dr || dn->dn_object == DMU_META_DNODE_OBJECT) { mutex_enter(&di->dt.di.dr_mtx); ASSERT3U(di->dr_txg, ==, tx->tx_txg); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&di->dt.di.dr_children, dr); mutex_exit(&di->dt.di.dr_mtx); dr->dr_parent = di; } mutex_exit(&db->db_mtx); } else { ASSERT(db->db_level + 1 == dn->dn_nlevels); ASSERT(db->db_blkid < dn->dn_nblkptr); ASSERT(db->db_parent == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); if (drop_struct_rwlock) rw_exit(&dn->dn_struct_rwlock); } dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } static void dbuf_undirty_bonus(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; if (dr->dt.dl.dr_data != db->db.db_data) { struct dnode *dn = dr->dr_dnode; int max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); kmem_free(dr->dt.dl.dr_data, max_bonuslen); arc_space_return(max_bonuslen, ARC_SPACE_BONUS); } db->db_data_pending = NULL; ASSERT(list_next(&db->db_dirty_records, dr) == NULL); list_remove(&db->db_dirty_records, dr); if (dr->dr_dbuf->db_level != 0) { mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT3U(db->db_dirtycnt, >, 0); db->db_dirtycnt -= 1; } /* * Undirty a buffer in the transaction group referenced by the given * transaction. Return whether this evicted the dbuf. */ static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { uint64_t txg = tx->tx_txg; ASSERT(txg != 0); /* * Due to our use of dn_nlevels below, this can only be called * in open context, unless we are operating on the MOS. * From syncing context, dn_nlevels may be different from the * dn_nlevels used when dbuf was dirtied. */ ASSERT(db->db_objset == dmu_objset_pool(db->db_objset)->dp_meta_objset || txg != spa_syncing_txg(dmu_objset_spa(db->db_objset))); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT0(db->db_level); ASSERT(MUTEX_HELD(&db->db_mtx)); /* * If this buffer is not dirty, we're done. */ dbuf_dirty_record_t *dr = dbuf_find_dirty_eq(db, txg); if (dr == NULL) return (B_FALSE); ASSERT(dr->dr_dbuf == db); dnode_t *dn = dr->dr_dnode; dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); ASSERT(db->db.db_size != 0); dsl_pool_undirty_space(dmu_objset_pool(dn->dn_objset), dr->dr_accounted, txg); list_remove(&db->db_dirty_records, dr); /* * Note that there are three places in dbuf_dirty() * where this dirty record may be put on a list. * Make sure to do a list_remove corresponding to * every one of those list_insert calls. */ if (dr->dr_parent) { mutex_enter(&dr->dr_parent->dt.di.dr_mtx); list_remove(&dr->dr_parent->dt.di.dr_children, dr); mutex_exit(&dr->dr_parent->dt.di.dr_mtx); } else if (db->db_blkid == DMU_SPILL_BLKID || db->db_level + 1 == dn->dn_nlevels) { ASSERT(db->db_blkptr == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); list_remove(&dn->dn_dirty_records[txg & TXG_MASK], dr); mutex_exit(&dn->dn_mtx); } if (db->db_state != DB_NOFILL) { dbuf_unoverride(dr); ASSERT(db->db_buf != NULL); ASSERT(dr->dt.dl.dr_data != NULL); if (dr->dt.dl.dr_data != db->db_buf) arc_buf_destroy(dr->dt.dl.dr_data, db); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; if (zfs_refcount_remove(&db->db_holds, (void *)(uintptr_t)txg) == 0) { ASSERT(db->db_state == DB_NOFILL || arc_released(db->db_buf)); dbuf_destroy(db); return (B_TRUE); } return (B_FALSE); } static void dmu_buf_will_dirty_impl(dmu_buf_t *db_fake, int flags, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(tx->tx_txg != 0); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); /* * Quick check for dirtiness. For already dirty blocks, this * reduces runtime of this function by >90%, and overall performance * by 50% for some workloads (e.g. file deletion with indirect blocks * cached). */ mutex_enter(&db->db_mtx); if (db->db_state == DB_CACHED) { dbuf_dirty_record_t *dr = dbuf_find_dirty_eq(db, tx->tx_txg); /* * It's possible that it is already dirty but not cached, * because there are some calls to dbuf_dirty() that don't * go through dmu_buf_will_dirty(). */ if (dr != NULL) { /* This dbuf is already dirty and cached. */ dbuf_redirty(dr); mutex_exit(&db->db_mtx); return; } } mutex_exit(&db->db_mtx); DB_DNODE_ENTER(db); if (RW_WRITE_HELD(&DB_DNODE(db)->dn_struct_rwlock)) flags |= DB_RF_HAVESTRUCT; DB_DNODE_EXIT(db); (void) dbuf_read(db, NULL, flags); (void) dbuf_dirty(db, tx); } void dmu_buf_will_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_will_dirty_impl(db_fake, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH, tx); } boolean_t dmu_buf_is_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_dirty_record_t *dr; mutex_enter(&db->db_mtx); dr = dbuf_find_dirty_eq(db, tx->tx_txg); mutex_exit(&db->db_mtx); return (dr != NULL); } void dmu_buf_will_not_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_state = DB_NOFILL; DTRACE_SET_STATE(db, "allocating NOFILL buffer"); dmu_buf_will_fill(db_fake, tx); } void dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(tx->tx_txg != 0); ASSERT(db->db_level == 0); ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); dbuf_noread(db); (void) dbuf_dirty(db, tx); } /* * This function is effectively the same as dmu_buf_will_dirty(), but * indicates the caller expects raw encrypted data in the db, and provides * the crypt params (byteorder, salt, iv, mac) which should be stored in the * blkptr_t when this dbuf is written. This is only used for blocks of * dnodes, during raw receive. */ void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder, const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_dirty_record_t *dr; /* * dr_has_raw_params is only processed for blocks of dnodes * (see dbuf_sync_dnode_leaf_crypt()). */ ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT); ASSERT3U(db->db_level, ==, 0); ASSERT(db->db_objset->os_raw_receive); dmu_buf_will_dirty_impl(db_fake, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_NO_DECRYPT, tx); dr = dbuf_find_dirty_eq(db, tx->tx_txg); ASSERT3P(dr, !=, NULL); dr->dt.dl.dr_has_raw_params = B_TRUE; dr->dt.dl.dr_byteorder = byteorder; bcopy(salt, dr->dt.dl.dr_salt, ZIO_DATA_SALT_LEN); bcopy(iv, dr->dt.dl.dr_iv, ZIO_DATA_IV_LEN); bcopy(mac, dr->dt.dl.dr_mac, ZIO_DATA_MAC_LEN); } static void dbuf_override_impl(dmu_buf_impl_t *db, const blkptr_t *bp, dmu_tx_t *tx) { struct dirty_leaf *dl; dbuf_dirty_record_t *dr; dr = list_head(&db->db_dirty_records); ASSERT3U(dr->dr_txg, ==, tx->tx_txg); dl = &dr->dt.dl; dl->dr_overridden_by = *bp; dl->dr_override_state = DR_OVERRIDDEN; dl->dr_overridden_by.blk_birth = dr->dr_txg; } /* ARGSUSED */ void dmu_buf_fill_done(dmu_buf_t *dbuf, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; dbuf_states_t old_state; mutex_enter(&db->db_mtx); DBUF_VERIFY(db); old_state = db->db_state; db->db_state = DB_CACHED; if (old_state == DB_FILL) { if (db->db_level == 0 && db->db_freed_in_flight) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); /* we were freed while filling */ /* XXX dbuf_undirty? */ bzero(db->db.db_data, db->db.db_size); db->db_freed_in_flight = FALSE; DTRACE_SET_STATE(db, "fill done handling freed in flight"); } else { DTRACE_SET_STATE(db, "fill done"); } cv_broadcast(&db->db_changed); } mutex_exit(&db->db_mtx); } void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, bp_embedded_type_t etype, enum zio_compress comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; struct dirty_leaf *dl; dmu_object_type_t type; dbuf_dirty_record_t *dr; if (etype == BP_EMBEDDED_TYPE_DATA) { ASSERT(spa_feature_is_active(dmu_objset_spa(db->db_objset), SPA_FEATURE_EMBEDDED_DATA)); } DB_DNODE_ENTER(db); type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); ASSERT0(db->db_level); ASSERT(db->db_blkid != DMU_BONUS_BLKID); dmu_buf_will_not_fill(dbuf, tx); dr = list_head(&db->db_dirty_records); ASSERT3U(dr->dr_txg, ==, tx->tx_txg); dl = &dr->dt.dl; encode_embedded_bp_compressed(&dl->dr_overridden_by, data, comp, uncompressed_size, compressed_size); BPE_SET_ETYPE(&dl->dr_overridden_by, etype); BP_SET_TYPE(&dl->dr_overridden_by, type); BP_SET_LEVEL(&dl->dr_overridden_by, 0); BP_SET_BYTEORDER(&dl->dr_overridden_by, byteorder); dl->dr_override_state = DR_OVERRIDDEN; dl->dr_overridden_by.blk_birth = dr->dr_txg; } void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; dmu_object_type_t type; ASSERT(dsl_dataset_feature_is_active(db->db_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); DB_DNODE_ENTER(db); type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); ASSERT0(db->db_level); dmu_buf_will_not_fill(dbuf, tx); blkptr_t bp = { { { {0} } } }; BP_SET_TYPE(&bp, type); BP_SET_LEVEL(&bp, 0); BP_SET_BIRTH(&bp, tx->tx_txg, 0); BP_SET_REDACTED(&bp); BPE_SET_LSIZE(&bp, dbuf->db_size); dbuf_override_impl(db, &bp, tx); } /* * Directly assign a provided arc buf to a given dbuf if it's not referenced * by anybody except our caller. Otherwise copy arcbuf's contents to dbuf. */ void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx) { ASSERT(!zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(db->db_level == 0); ASSERT3U(dbuf_is_metadata(db), ==, arc_is_metadata(buf)); ASSERT(buf != NULL); ASSERT3U(arc_buf_lsize(buf), ==, db->db.db_size); ASSERT(tx->tx_txg != 0); arc_return_buf(buf, db); ASSERT(arc_released(buf)); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); ASSERT(db->db_state == DB_CACHED || db->db_state == DB_UNCACHED); if (db->db_state == DB_CACHED && zfs_refcount_count(&db->db_holds) - 1 > db->db_dirtycnt) { /* * In practice, we will never have a case where we have an * encrypted arc buffer while additional holds exist on the * dbuf. We don't handle this here so we simply assert that * fact instead. */ ASSERT(!arc_is_encrypted(buf)); mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); bcopy(buf->b_data, db->db.db_data, db->db.db_size); arc_buf_destroy(buf, db); return; } if (db->db_state == DB_CACHED) { dbuf_dirty_record_t *dr = list_head(&db->db_dirty_records); ASSERT(db->db_buf != NULL); if (dr != NULL && dr->dr_txg == tx->tx_txg) { ASSERT(dr->dt.dl.dr_data == db->db_buf); if (!arc_released(db->db_buf)) { ASSERT(dr->dt.dl.dr_override_state == DR_OVERRIDDEN); arc_release(db->db_buf, db); } dr->dt.dl.dr_data = buf; arc_buf_destroy(db->db_buf, db); } else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) { arc_release(db->db_buf, db); arc_buf_destroy(db->db_buf, db); } db->db_buf = NULL; } ASSERT(db->db_buf == NULL); dbuf_set_data(db, buf); db->db_state = DB_FILL; DTRACE_SET_STATE(db, "filling assigned arcbuf"); mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); dmu_buf_fill_done(&db->db, tx); } void dbuf_destroy(dmu_buf_impl_t *db) { dnode_t *dn; dmu_buf_impl_t *parent = db->db_parent; dmu_buf_impl_t *dndb; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(zfs_refcount_is_zero(&db->db_holds)); if (db->db_buf != NULL) { arc_buf_destroy(db->db_buf, db); db->db_buf = NULL; } if (db->db_blkid == DMU_BONUS_BLKID) { int slots = DB_DNODE(db)->dn_num_slots; int bonuslen = DN_SLOTS_TO_BONUSLEN(slots); if (db->db.db_data != NULL) { kmem_free(db->db.db_data, bonuslen); arc_space_return(bonuslen, ARC_SPACE_BONUS); db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "buffer cleared"); } } dbuf_clear_data(db); if (multilist_link_active(&db->db_cache_link)) { ASSERT(db->db_caching_status == DB_DBUF_CACHE || db->db_caching_status == DB_DBUF_METADATA_CACHE); multilist_remove(&dbuf_caches[db->db_caching_status].cache, db); (void) zfs_refcount_remove_many( &dbuf_caches[db->db_caching_status].size, db->db.db_size, db); if (db->db_caching_status == DB_DBUF_METADATA_CACHE) { DBUF_STAT_BUMPDOWN(metadata_cache_count); } else { DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]); DBUF_STAT_BUMPDOWN(cache_count); DBUF_STAT_DECR(cache_levels_bytes[db->db_level], db->db.db_size); } db->db_caching_status = DB_NO_CACHE; } ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); ASSERT(db->db_data_pending == NULL); ASSERT(list_is_empty(&db->db_dirty_records)); db->db_state = DB_EVICTING; DTRACE_SET_STATE(db, "buffer eviction started"); db->db_blkptr = NULL; /* * Now that db_state is DB_EVICTING, nobody else can find this via * the hash table. We can now drop db_mtx, which allows us to * acquire the dn_dbufs_mtx. */ mutex_exit(&db->db_mtx); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dndb = dn->dn_dbuf; if (db->db_blkid != DMU_BONUS_BLKID) { boolean_t needlock = !MUTEX_HELD(&dn->dn_dbufs_mtx); if (needlock) mutex_enter_nested(&dn->dn_dbufs_mtx, NESTED_SINGLE); avl_remove(&dn->dn_dbufs, db); membar_producer(); DB_DNODE_EXIT(db); if (needlock) mutex_exit(&dn->dn_dbufs_mtx); /* * Decrementing the dbuf count means that the hold corresponding * to the removed dbuf is no longer discounted in dnode_move(), * so the dnode cannot be moved until after we release the hold. * The membar_producer() ensures visibility of the decremented * value in dnode_move(), since DB_DNODE_EXIT doesn't actually * release any lock. */ mutex_enter(&dn->dn_mtx); dnode_rele_and_unlock(dn, db, B_TRUE); db->db_dnode_handle = NULL; dbuf_hash_remove(db); } else { DB_DNODE_EXIT(db); } ASSERT(zfs_refcount_is_zero(&db->db_holds)); db->db_parent = NULL; ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); ASSERT(db->db_hash_next == NULL); ASSERT(db->db_blkptr == NULL); ASSERT(db->db_data_pending == NULL); ASSERT3U(db->db_caching_status, ==, DB_NO_CACHE); ASSERT(!multilist_link_active(&db->db_cache_link)); kmem_cache_free(dbuf_kmem_cache, db); arc_space_return(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF); /* * If this dbuf is referenced from an indirect dbuf, * decrement the ref count on the indirect dbuf. */ if (parent && parent != dndb) { mutex_enter(&parent->db_mtx); dbuf_rele_and_unlock(parent, db, B_TRUE); } } /* * Note: While bpp will always be updated if the function returns success, * parentp will not be updated if the dnode does not have dn_dbuf filled in; * this happens when the dnode is the meta-dnode, or {user|group|project}used * object. */ __attribute__((always_inline)) static inline int dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse, dmu_buf_impl_t **parentp, blkptr_t **bpp) { *parentp = NULL; *bpp = NULL; ASSERT(blkid != DMU_BONUS_BLKID); if (blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_have_spill && (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) *bpp = DN_SPILL_BLKPTR(dn->dn_phys); else *bpp = NULL; dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; mutex_exit(&dn->dn_mtx); return (0); } int nlevels = (dn->dn_phys->dn_nlevels == 0) ? 1 : dn->dn_phys->dn_nlevels; int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(level * epbs, <, 64); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); /* * This assertion shouldn't trip as long as the max indirect block size * is less than 1M. The reason for this is that up to that point, * the number of levels required to address an entire object with blocks * of size SPA_MINBLOCKSIZE satisfies nlevels * epbs + 1 <= 64. In * other words, if N * epbs + 1 > 64, then if (N-1) * epbs + 1 > 55 * (i.e. we can address the entire object), objects will all use at most * N-1 levels and the assertion won't overflow. However, once epbs is * 13, 4 * 13 + 1 = 53, but 5 * 13 + 1 = 66. Then, 4 levels will not be * enough to address an entire object, so objects will have 5 levels, * but then this assertion will overflow. * * All this is to say that if we ever increase DN_MAX_INDBLKSHIFT, we * need to redo this logic to handle overflows. */ ASSERT(level >= nlevels || ((nlevels - level - 1) * epbs) + highbit64(dn->dn_phys->dn_nblkptr) <= 64); if (level >= nlevels || blkid >= ((uint64_t)dn->dn_phys->dn_nblkptr << ((nlevels - level - 1) * epbs)) || (fail_sparse && blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) { /* the buffer has no parent yet */ return (SET_ERROR(ENOENT)); } else if (level < nlevels-1) { /* this block is referenced from an indirect block */ int err; err = dbuf_hold_impl(dn, level + 1, blkid >> epbs, fail_sparse, FALSE, NULL, parentp); if (err) return (err); err = dbuf_read(*parentp, NULL, (DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH | DB_RF_CANFAIL)); if (err) { dbuf_rele(*parentp, NULL); *parentp = NULL; return (err); } rw_enter(&(*parentp)->db_rwlock, RW_READER); *bpp = ((blkptr_t *)(*parentp)->db.db_data) + (blkid & ((1ULL << epbs) - 1)); if (blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs))) ASSERT(BP_IS_HOLE(*bpp)); rw_exit(&(*parentp)->db_rwlock); return (0); } else { /* the block is referenced from the dnode */ ASSERT3U(level, ==, nlevels-1); ASSERT(dn->dn_phys->dn_nblkptr == 0 || blkid < dn->dn_phys->dn_nblkptr); if (dn->dn_dbuf) { dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; } *bpp = &dn->dn_phys->dn_blkptr[blkid]; return (0); } } static dmu_buf_impl_t * dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid, dmu_buf_impl_t *parent, blkptr_t *blkptr) { objset_t *os = dn->dn_objset; dmu_buf_impl_t *db, *odb; ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_type != DMU_OT_NONE); db = kmem_cache_alloc(dbuf_kmem_cache, KM_SLEEP); list_create(&db->db_dirty_records, sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dbuf_node)); db->db_objset = os; db->db.db_object = dn->dn_object; db->db_level = level; db->db_blkid = blkid; db->db_dirtycnt = 0; db->db_dnode_handle = dn->dn_handle; db->db_parent = parent; db->db_blkptr = blkptr; db->db_user = NULL; db->db_user_immediate_evict = FALSE; db->db_freed_in_flight = FALSE; db->db_pending_evict = FALSE; if (blkid == DMU_BONUS_BLKID) { ASSERT3P(parent, ==, dn->dn_dbuf); db->db.db_size = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) - (dn->dn_nblkptr-1) * sizeof (blkptr_t); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); db->db.db_offset = DMU_BONUS_BLKID; db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "bonus buffer created"); db->db_caching_status = DB_NO_CACHE; /* the bonus dbuf is not placed in the hash table */ arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF); return (db); } else if (blkid == DMU_SPILL_BLKID) { db->db.db_size = (blkptr != NULL) ? BP_GET_LSIZE(blkptr) : SPA_MINBLOCKSIZE; db->db.db_offset = 0; } else { int blocksize = db->db_level ? 1 << dn->dn_indblkshift : dn->dn_datablksz; db->db.db_size = blocksize; db->db.db_offset = db->db_blkid * blocksize; } /* * Hold the dn_dbufs_mtx while we get the new dbuf * in the hash table *and* added to the dbufs list. * This prevents a possible deadlock with someone * trying to look up this dbuf before it's added to the * dn_dbufs list. */ mutex_enter(&dn->dn_dbufs_mtx); db->db_state = DB_EVICTING; /* not worth logging this state change */ if ((odb = dbuf_hash_insert(db)) != NULL) { /* someone else inserted it first */ kmem_cache_free(dbuf_kmem_cache, db); mutex_exit(&dn->dn_dbufs_mtx); DBUF_STAT_BUMP(hash_insert_race); return (odb); } avl_add(&dn->dn_dbufs, db); db->db_state = DB_UNCACHED; DTRACE_SET_STATE(db, "regular buffer created"); db->db_caching_status = DB_NO_CACHE; mutex_exit(&dn->dn_dbufs_mtx); arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_DBUF); if (parent && parent != dn->dn_dbuf) dbuf_add_ref(parent, db); ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || zfs_refcount_count(&dn->dn_holds) > 0); (void) zfs_refcount_add(&dn->dn_holds, db); dprintf_dbuf(db, "db=%p\n", db); return (db); } /* * This function returns a block pointer and information about the object, * given a dnode and a block. This is a publicly accessible version of * dbuf_findbp that only returns some information, rather than the * dbuf. Note that the dnode passed in must be held, and the dn_struct_rwlock * should be locked as (at least) a reader. */ int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid, blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift) { dmu_buf_impl_t *dbp = NULL; blkptr_t *bp2; int err = 0; ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); err = dbuf_findbp(dn, level, blkid, B_FALSE, &dbp, &bp2); if (err == 0) { *bp = *bp2; if (dbp != NULL) dbuf_rele(dbp, NULL); if (datablkszsec != NULL) *datablkszsec = dn->dn_phys->dn_datablkszsec; if (indblkshift != NULL) *indblkshift = dn->dn_phys->dn_indblkshift; } return (err); } typedef struct dbuf_prefetch_arg { spa_t *dpa_spa; /* The spa to issue the prefetch in. */ zbookmark_phys_t dpa_zb; /* The target block to prefetch. */ int dpa_epbs; /* Entries (blkptr_t's) Per Block Shift. */ int dpa_curlevel; /* The current level that we're reading */ dnode_t *dpa_dnode; /* The dnode associated with the prefetch */ zio_priority_t dpa_prio; /* The priority I/Os should be issued at. */ zio_t *dpa_zio; /* The parent zio_t for all prefetches. */ arc_flags_t dpa_aflags; /* Flags to pass to the final prefetch. */ dbuf_prefetch_fn dpa_cb; /* prefetch completion callback */ void *dpa_arg; /* prefetch completion arg */ } dbuf_prefetch_arg_t; static void dbuf_prefetch_fini(dbuf_prefetch_arg_t *dpa, boolean_t io_done) { if (dpa->dpa_cb != NULL) dpa->dpa_cb(dpa->dpa_arg, io_done); kmem_free(dpa, sizeof (*dpa)); } static void dbuf_issue_final_prefetch_done(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *iobp, arc_buf_t *abuf, void *private) { dbuf_prefetch_arg_t *dpa = private; dbuf_prefetch_fini(dpa, B_TRUE); if (abuf != NULL) arc_buf_destroy(abuf, private); } /* * Actually issue the prefetch read for the block given. */ static void dbuf_issue_final_prefetch(dbuf_prefetch_arg_t *dpa, blkptr_t *bp) { ASSERT(!BP_IS_REDACTED(bp) || dsl_dataset_feature_is_active( dpa->dpa_dnode->dn_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) || BP_IS_REDACTED(bp)) return (dbuf_prefetch_fini(dpa, B_FALSE)); int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE; arc_flags_t aflags = dpa->dpa_aflags | ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH | ARC_FLAG_NO_BUF; /* dnodes are always read as raw and then converted later */ if (BP_GET_TYPE(bp) == DMU_OT_DNODE && BP_IS_PROTECTED(bp) && dpa->dpa_curlevel == 0) zio_flags |= ZIO_FLAG_RAW; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); ASSERT3U(dpa->dpa_curlevel, ==, dpa->dpa_zb.zb_level); ASSERT(dpa->dpa_zio != NULL); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, dbuf_issue_final_prefetch_done, dpa, dpa->dpa_prio, zio_flags, &aflags, &dpa->dpa_zb); } /* * Called when an indirect block above our prefetch target is read in. This * will either read in the next indirect block down the tree or issue the actual * prefetch if the next block down is our target. */ static void dbuf_prefetch_indirect_done(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *iobp, arc_buf_t *abuf, void *private) { dbuf_prefetch_arg_t *dpa = private; ASSERT3S(dpa->dpa_zb.zb_level, <, dpa->dpa_curlevel); ASSERT3S(dpa->dpa_curlevel, >, 0); if (abuf == NULL) { ASSERT(zio == NULL || zio->io_error != 0); return (dbuf_prefetch_fini(dpa, B_TRUE)); } ASSERT(zio == NULL || zio->io_error == 0); /* * The dpa_dnode is only valid if we are called with a NULL * zio. This indicates that the arc_read() returned without * first calling zio_read() to issue a physical read. Once * a physical read is made the dpa_dnode must be invalidated * as the locks guarding it may have been dropped. If the * dpa_dnode is still valid, then we want to add it to the dbuf * cache. To do so, we must hold the dbuf associated with the block * we just prefetched, read its contents so that we associate it * with an arc_buf_t, and then release it. */ if (zio != NULL) { ASSERT3S(BP_GET_LEVEL(zio->io_bp), ==, dpa->dpa_curlevel); if (zio->io_flags & ZIO_FLAG_RAW_COMPRESS) { ASSERT3U(BP_GET_PSIZE(zio->io_bp), ==, zio->io_size); } else { ASSERT3U(BP_GET_LSIZE(zio->io_bp), ==, zio->io_size); } ASSERT3P(zio->io_spa, ==, dpa->dpa_spa); dpa->dpa_dnode = NULL; } else if (dpa->dpa_dnode != NULL) { uint64_t curblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); dmu_buf_impl_t *db = dbuf_hold_level(dpa->dpa_dnode, dpa->dpa_curlevel, curblkid, FTAG); if (db == NULL) { arc_buf_destroy(abuf, private); return (dbuf_prefetch_fini(dpa, B_TRUE)); } (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_HAVESTRUCT); dbuf_rele(db, FTAG); } dpa->dpa_curlevel--; uint64_t nextblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); blkptr_t *bp = ((blkptr_t *)abuf->b_data) + P2PHASE(nextblkid, 1ULL << dpa->dpa_epbs); ASSERT(!BP_IS_REDACTED(bp) || dsl_dataset_feature_is_active( dpa->dpa_dnode->dn_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) { dbuf_prefetch_fini(dpa, B_TRUE); } else if (dpa->dpa_curlevel == dpa->dpa_zb.zb_level) { ASSERT3U(nextblkid, ==, dpa->dpa_zb.zb_blkid); dbuf_issue_final_prefetch(dpa, bp); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; /* flag if L2ARC eligible, l2arc_noprefetch then decides */ if (dpa->dpa_aflags & ARC_FLAG_L2CACHE) iter_aflags |= ARC_FLAG_L2CACHE; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); SET_BOOKMARK(&zb, dpa->dpa_zb.zb_objset, dpa->dpa_zb.zb_object, dpa->dpa_curlevel, nextblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, dbuf_prefetch_indirect_done, dpa, dpa->dpa_prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } arc_buf_destroy(abuf, private); } /* * Issue prefetch reads for the given block on the given level. If the indirect * blocks above that block are not in memory, we will read them in * asynchronously. As a result, this call never blocks waiting for a read to * complete. Note that the prefetch might fail if the dataset is encrypted and * the encryption key is unmapped before the IO completes. */ int dbuf_prefetch_impl(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb, void *arg) { blkptr_t bp; int epbs, nlevels, curlevel; uint64_t curblkid; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); if (blkid > dn->dn_maxblkid) goto no_issue; if (level == 0 && dnode_block_freed(dn, blkid)) goto no_issue; /* * This dnode hasn't been written to disk yet, so there's nothing to * prefetch. */ nlevels = dn->dn_phys->dn_nlevels; if (level >= nlevels || dn->dn_phys->dn_nblkptr == 0) goto no_issue; epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; if (dn->dn_phys->dn_maxblkid < blkid << (epbs * level)) goto no_issue; dmu_buf_impl_t *db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db != NULL) { mutex_exit(&db->db_mtx); /* * This dbuf already exists. It is either CACHED, or * (we assume) about to be read or filled. */ goto no_issue; } /* * Find the closest ancestor (indirect block) of the target block * that is present in the cache. In this indirect block, we will * find the bp that is at curlevel, curblkid. */ curlevel = level; curblkid = blkid; while (curlevel < nlevels - 1) { int parent_level = curlevel + 1; uint64_t parent_blkid = curblkid >> epbs; dmu_buf_impl_t *db; if (dbuf_hold_impl(dn, parent_level, parent_blkid, FALSE, TRUE, FTAG, &db) == 0) { blkptr_t *bpp = db->db_buf->b_data; bp = bpp[P2PHASE(curblkid, 1 << epbs)]; dbuf_rele(db, FTAG); break; } curlevel = parent_level; curblkid = parent_blkid; } if (curlevel == nlevels - 1) { /* No cached indirect blocks found. */ ASSERT3U(curblkid, <, dn->dn_phys->dn_nblkptr); bp = dn->dn_phys->dn_blkptr[curblkid]; } ASSERT(!BP_IS_REDACTED(&bp) || dsl_dataset_feature_is_active(dn->dn_objset->os_dsl_dataset, SPA_FEATURE_REDACTED_DATASETS)); if (BP_IS_HOLE(&bp) || BP_IS_REDACTED(&bp)) goto no_issue; ASSERT3U(curlevel, ==, BP_GET_LEVEL(&bp)); zio_t *pio = zio_root(dmu_objset_spa(dn->dn_objset), NULL, NULL, ZIO_FLAG_CANFAIL); dbuf_prefetch_arg_t *dpa = kmem_zalloc(sizeof (*dpa), KM_SLEEP); dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; SET_BOOKMARK(&dpa->dpa_zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, level, blkid); dpa->dpa_curlevel = curlevel; dpa->dpa_prio = prio; dpa->dpa_aflags = aflags; dpa->dpa_spa = dn->dn_objset->os_spa; dpa->dpa_dnode = dn; dpa->dpa_epbs = epbs; dpa->dpa_zio = pio; dpa->dpa_cb = cb; dpa->dpa_arg = arg; /* flag if L2ARC eligible, l2arc_noprefetch then decides */ if (DNODE_LEVEL_IS_L2CACHEABLE(dn, level)) dpa->dpa_aflags |= ARC_FLAG_L2CACHE; /* * If we have the indirect just above us, no need to do the asynchronous * prefetch chain; we'll just run the last step ourselves. If we're at * a higher level, though, we want to issue the prefetches for all the * indirect blocks asynchronously, so we can go on with whatever we were * doing. */ if (curlevel == level) { ASSERT3U(curblkid, ==, blkid); dbuf_issue_final_prefetch(dpa, &bp); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; /* flag if L2ARC eligible, l2arc_noprefetch then decides */ if (DNODE_LEVEL_IS_L2CACHEABLE(dn, level)) iter_aflags |= ARC_FLAG_L2CACHE; SET_BOOKMARK(&zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, curlevel, curblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, &bp, dbuf_prefetch_indirect_done, dpa, prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } /* * We use pio here instead of dpa_zio since it's possible that * dpa may have already been freed. */ zio_nowait(pio); return (1); no_issue: if (cb != NULL) cb(arg, B_FALSE); return (0); } int dbuf_prefetch(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags) { return (dbuf_prefetch_impl(dn, level, blkid, prio, aflags, NULL, NULL)); } /* * Helper function for dbuf_hold_impl() to copy a buffer. Handles * the case of encrypted, compressed and uncompressed buffers by * allocating the new buffer, respectively, with arc_alloc_raw_buf(), * arc_alloc_compressed_buf() or arc_alloc_buf().* * * NOTE: Declared noinline to avoid stack bloat in dbuf_hold_impl(). */ noinline static void dbuf_hold_copy(dnode_t *dn, dmu_buf_impl_t *db) { dbuf_dirty_record_t *dr = db->db_data_pending; arc_buf_t *newdata, *data = dr->dt.dl.dr_data; newdata = dbuf_alloc_arcbuf_from_arcbuf(db, data); dbuf_set_data(db, newdata); rw_enter(&db->db_rwlock, RW_WRITER); bcopy(data->b_data, db->db.db_data, arc_buf_size(data)); rw_exit(&db->db_rwlock); } /* * Returns with db_holds incremented, and db_mtx not held. * Note: dn_struct_rwlock must be held. */ int dbuf_hold_impl(dnode_t *dn, uint8_t level, uint64_t blkid, boolean_t fail_sparse, boolean_t fail_uncached, void *tag, dmu_buf_impl_t **dbp) { dmu_buf_impl_t *db, *parent = NULL; /* If the pool has been created, verify the tx_sync_lock is not held */ spa_t *spa = dn->dn_objset->os_spa; dsl_pool_t *dp = spa->spa_dsl_pool; if (dp != NULL) { ASSERT(!MUTEX_HELD(&dp->dp_tx.tx_sync_lock)); } ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT3U(dn->dn_nlevels, >, level); *dbp = NULL; /* dbuf_find() returns with db_mtx held */ db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db == NULL) { blkptr_t *bp = NULL; int err; if (fail_uncached) return (SET_ERROR(ENOENT)); ASSERT3P(parent, ==, NULL); err = dbuf_findbp(dn, level, blkid, fail_sparse, &parent, &bp); if (fail_sparse) { if (err == 0 && bp && BP_IS_HOLE(bp)) err = SET_ERROR(ENOENT); if (err) { if (parent) dbuf_rele(parent, NULL); return (err); } } if (err && err != ENOENT) return (err); db = dbuf_create(dn, level, blkid, parent, bp); } if (fail_uncached && db->db_state != DB_CACHED) { mutex_exit(&db->db_mtx); return (SET_ERROR(ENOENT)); } if (db->db_buf != NULL) { arc_buf_access(db->db_buf); ASSERT3P(db->db.db_data, ==, db->db_buf->b_data); } ASSERT(db->db_buf == NULL || arc_referenced(db->db_buf)); /* * If this buffer is currently syncing out, and we are * still referencing it from db_data, we need to make a copy * of it in case we decide we want to dirty it again in this txg. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && dn->dn_object != DMU_META_DNODE_OBJECT && db->db_state == DB_CACHED && db->db_data_pending) { dbuf_dirty_record_t *dr = db->db_data_pending; if (dr->dt.dl.dr_data == db->db_buf) dbuf_hold_copy(dn, db); } if (multilist_link_active(&db->db_cache_link)) { ASSERT(zfs_refcount_is_zero(&db->db_holds)); ASSERT(db->db_caching_status == DB_DBUF_CACHE || db->db_caching_status == DB_DBUF_METADATA_CACHE); multilist_remove(&dbuf_caches[db->db_caching_status].cache, db); (void) zfs_refcount_remove_many( &dbuf_caches[db->db_caching_status].size, db->db.db_size, db); if (db->db_caching_status == DB_DBUF_METADATA_CACHE) { DBUF_STAT_BUMPDOWN(metadata_cache_count); } else { DBUF_STAT_BUMPDOWN(cache_levels[db->db_level]); DBUF_STAT_BUMPDOWN(cache_count); DBUF_STAT_DECR(cache_levels_bytes[db->db_level], db->db.db_size); } db->db_caching_status = DB_NO_CACHE; } (void) zfs_refcount_add(&db->db_holds, tag); DBUF_VERIFY(db); mutex_exit(&db->db_mtx); /* NOTE: we can't rele the parent until after we drop the db_mtx */ if (parent) dbuf_rele(parent, NULL); ASSERT3P(DB_DNODE(db), ==, dn); ASSERT3U(db->db_blkid, ==, blkid); ASSERT3U(db->db_level, ==, level); *dbp = db; return (0); } dmu_buf_impl_t * dbuf_hold(dnode_t *dn, uint64_t blkid, void *tag) { return (dbuf_hold_level(dn, 0, blkid, tag)); } dmu_buf_impl_t * dbuf_hold_level(dnode_t *dn, int level, uint64_t blkid, void *tag) { dmu_buf_impl_t *db; int err = dbuf_hold_impl(dn, level, blkid, FALSE, FALSE, tag, &db); return (err ? NULL : db); } void dbuf_create_bonus(dnode_t *dn) { ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_bonus == NULL); dn->dn_bonus = dbuf_create(dn, 0, DMU_BONUS_BLKID, dn->dn_dbuf, NULL); } int dbuf_spill_set_blksz(dmu_buf_t *db_fake, uint64_t blksz, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; if (db->db_blkid != DMU_SPILL_BLKID) return (SET_ERROR(ENOTSUP)); if (blksz == 0) blksz = SPA_MINBLOCKSIZE; ASSERT3U(blksz, <=, spa_maxblocksize(dmu_objset_spa(db->db_objset))); blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE); dbuf_new_size(db, blksz, tx); return (0); } void dbuf_rm_spill(dnode_t *dn, dmu_tx_t *tx) { dbuf_free_range(dn, DMU_SPILL_BLKID, DMU_SPILL_BLKID, tx); } #pragma weak dmu_buf_add_ref = dbuf_add_ref void dbuf_add_ref(dmu_buf_impl_t *db, void *tag) { int64_t holds = zfs_refcount_add(&db->db_holds, tag); VERIFY3S(holds, >, 1); } #pragma weak dmu_buf_try_add_ref = dbuf_try_add_ref boolean_t dbuf_try_add_ref(dmu_buf_t *db_fake, objset_t *os, uint64_t obj, uint64_t blkid, void *tag) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dmu_buf_impl_t *found_db; boolean_t result = B_FALSE; if (blkid == DMU_BONUS_BLKID) found_db = dbuf_find_bonus(os, obj); else found_db = dbuf_find(os, obj, 0, blkid); if (found_db != NULL) { if (db == found_db && dbuf_refcount(db) > db->db_dirtycnt) { (void) zfs_refcount_add(&db->db_holds, tag); result = B_TRUE; } mutex_exit(&found_db->db_mtx); } return (result); } /* * If you call dbuf_rele() you had better not be referencing the dnode handle * unless you have some other direct or indirect hold on the dnode. (An indirect * hold is a hold on one of the dnode's dbufs, including the bonus buffer.) * Without that, the dbuf_rele() could lead to a dnode_rele() followed by the * dnode's parent dbuf evicting its dnode handles. */ void dbuf_rele(dmu_buf_impl_t *db, void *tag) { mutex_enter(&db->db_mtx); dbuf_rele_and_unlock(db, tag, B_FALSE); } void dmu_buf_rele(dmu_buf_t *db, void *tag) { dbuf_rele((dmu_buf_impl_t *)db, tag); } /* * dbuf_rele() for an already-locked dbuf. This is necessary to allow * db_dirtycnt and db_holds to be updated atomically. The 'evicting' * argument should be set if we are already in the dbuf-evicting code * path, in which case we don't want to recursively evict. This allows us to * avoid deeply nested stacks that would have a call flow similar to this: * * dbuf_rele()-->dbuf_rele_and_unlock()-->dbuf_evict_notify() * ^ | * | | * +-----dbuf_destroy()<--dbuf_evict_one()<--------+ * */ void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag, boolean_t evicting) { int64_t holds; uint64_t size; ASSERT(MUTEX_HELD(&db->db_mtx)); DBUF_VERIFY(db); /* * Remove the reference to the dbuf before removing its hold on the * dnode so we can guarantee in dnode_move() that a referenced bonus * buffer has a corresponding dnode hold. */ holds = zfs_refcount_remove(&db->db_holds, tag); ASSERT(holds >= 0); /* * We can't freeze indirects if there is a possibility that they * may be modified in the current syncing context. */ if (db->db_buf != NULL && holds == (db->db_level == 0 ? db->db_dirtycnt : 0)) { arc_buf_freeze(db->db_buf); } if (holds == db->db_dirtycnt && db->db_level == 0 && db->db_user_immediate_evict) dbuf_evict_user(db); if (holds == 0) { if (db->db_blkid == DMU_BONUS_BLKID) { dnode_t *dn; boolean_t evict_dbuf = db->db_pending_evict; /* * If the dnode moves here, we cannot cross this * barrier until the move completes. */ DB_DNODE_ENTER(db); dn = DB_DNODE(db); atomic_dec_32(&dn->dn_dbufs_count); /* * Decrementing the dbuf count means that the bonus * buffer's dnode hold is no longer discounted in * dnode_move(). The dnode cannot move until after * the dnode_rele() below. */ DB_DNODE_EXIT(db); /* * Do not reference db after its lock is dropped. * Another thread may evict it. */ mutex_exit(&db->db_mtx); if (evict_dbuf) dnode_evict_bonus(dn); dnode_rele(dn, db); } else if (db->db_buf == NULL) { /* * This is a special case: we never associated this * dbuf with any data allocated from the ARC. */ ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); dbuf_destroy(db); } else if (arc_released(db->db_buf)) { /* * This dbuf has anonymous data associated with it. */ dbuf_destroy(db); } else { boolean_t do_arc_evict = B_FALSE; blkptr_t bp; spa_t *spa = dmu_objset_spa(db->db_objset); if (!DBUF_IS_CACHEABLE(db) && db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr) && !BP_IS_EMBEDDED(db->db_blkptr)) { do_arc_evict = B_TRUE; bp = *db->db_blkptr; } if (!DBUF_IS_CACHEABLE(db) || db->db_pending_evict) { dbuf_destroy(db); } else if (!multilist_link_active(&db->db_cache_link)) { ASSERT3U(db->db_caching_status, ==, DB_NO_CACHE); dbuf_cached_state_t dcs = dbuf_include_in_metadata_cache(db) ? DB_DBUF_METADATA_CACHE : DB_DBUF_CACHE; db->db_caching_status = dcs; multilist_insert(&dbuf_caches[dcs].cache, db); + uint64_t db_size = db->db.db_size; size = zfs_refcount_add_many( - &dbuf_caches[dcs].size, - db->db.db_size, db); + &dbuf_caches[dcs].size, db_size, db); + uint8_t db_level = db->db_level; + mutex_exit(&db->db_mtx); if (dcs == DB_DBUF_METADATA_CACHE) { DBUF_STAT_BUMP(metadata_cache_count); DBUF_STAT_MAX( metadata_cache_size_bytes_max, size); } else { - DBUF_STAT_BUMP( - cache_levels[db->db_level]); DBUF_STAT_BUMP(cache_count); - DBUF_STAT_INCR( - cache_levels_bytes[db->db_level], - db->db.db_size); DBUF_STAT_MAX(cache_size_bytes_max, size); + DBUF_STAT_BUMP(cache_levels[db_level]); + DBUF_STAT_INCR( + cache_levels_bytes[db_level], + db_size); } - mutex_exit(&db->db_mtx); if (dcs == DB_DBUF_CACHE && !evicting) dbuf_evict_notify(size); } if (do_arc_evict) arc_freed(spa, &bp); } } else { mutex_exit(&db->db_mtx); } } #pragma weak dmu_buf_refcount = dbuf_refcount uint64_t dbuf_refcount(dmu_buf_impl_t *db) { return (zfs_refcount_count(&db->db_holds)); } uint64_t dmu_buf_user_refcount(dmu_buf_t *db_fake) { uint64_t holds; dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; mutex_enter(&db->db_mtx); ASSERT3U(zfs_refcount_count(&db->db_holds), >=, db->db_dirtycnt); holds = zfs_refcount_count(&db->db_holds) - db->db_dirtycnt; mutex_exit(&db->db_mtx); return (holds); } void * dmu_buf_replace_user(dmu_buf_t *db_fake, dmu_buf_user_t *old_user, dmu_buf_user_t *new_user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; mutex_enter(&db->db_mtx); dbuf_verify_user(db, DBVU_NOT_EVICTING); if (db->db_user == old_user) db->db_user = new_user; else old_user = db->db_user; dbuf_verify_user(db, DBVU_NOT_EVICTING); mutex_exit(&db->db_mtx); return (old_user); } void * dmu_buf_set_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, NULL, user)); } void * dmu_buf_set_user_ie(dmu_buf_t *db_fake, dmu_buf_user_t *user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_user_immediate_evict = TRUE; return (dmu_buf_set_user(db_fake, user)); } void * dmu_buf_remove_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, user, NULL)); } void * dmu_buf_get_user(dmu_buf_t *db_fake) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_verify_user(db, DBVU_NOT_EVICTING); return (db->db_user); } void dmu_buf_user_evict_wait() { taskq_wait(dbu_evict_taskq); } blkptr_t * dmu_buf_get_blkptr(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_blkptr); } objset_t * dmu_buf_get_objset(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_objset); } dnode_t * dmu_buf_dnode_enter(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_ENTER(dbi); return (DB_DNODE(dbi)); } void dmu_buf_dnode_exit(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_EXIT(dbi); } static void dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db) { /* ASSERT(dmu_tx_is_syncing(tx) */ ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_blkptr != NULL) return; if (db->db_blkid == DMU_SPILL_BLKID) { db->db_blkptr = DN_SPILL_BLKPTR(dn->dn_phys); BP_ZERO(db->db_blkptr); return; } if (db->db_level == dn->dn_phys->dn_nlevels-1) { /* * This buffer was allocated at a time when there was * no available blkptrs from the dnode, or it was * inappropriate to hook it in (i.e., nlevels mismatch). */ ASSERT(db->db_blkid < dn->dn_phys->dn_nblkptr); ASSERT(db->db_parent == NULL); db->db_parent = dn->dn_dbuf; db->db_blkptr = &dn->dn_phys->dn_blkptr[db->db_blkid]; DBUF_VERIFY(db); } else { dmu_buf_impl_t *parent = db->db_parent; int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT(dn->dn_phys->dn_nlevels > 1); if (parent == NULL) { mutex_exit(&db->db_mtx); rw_enter(&dn->dn_struct_rwlock, RW_READER); parent = dbuf_hold_level(dn, db->db_level + 1, db->db_blkid >> epbs, db); rw_exit(&dn->dn_struct_rwlock); mutex_enter(&db->db_mtx); db->db_parent = parent; } db->db_blkptr = (blkptr_t *)parent->db.db_data + (db->db_blkid & ((1ULL << epbs) - 1)); DBUF_VERIFY(db); } } static void dbuf_sync_bonus(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; void *data = dr->dt.dl.dr_data; ASSERT0(db->db_level); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_blkid == DMU_BONUS_BLKID); ASSERT(data != NULL); dnode_t *dn = dr->dr_dnode; ASSERT3U(DN_MAX_BONUS_LEN(dn->dn_phys), <=, DN_SLOTS_TO_BONUSLEN(dn->dn_phys->dn_extra_slots + 1)); bcopy(data, DN_BONUS(dn->dn_phys), DN_MAX_BONUS_LEN(dn->dn_phys)); dbuf_sync_leaf_verify_bonus_dnode(dr); dbuf_undirty_bonus(dr); dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg, B_FALSE); } /* * When syncing out a blocks of dnodes, adjust the block to deal with * encryption. Normally, we make sure the block is decrypted before writing * it. If we have crypt params, then we are writing a raw (encrypted) block, * from a raw receive. In this case, set the ARC buf's crypt params so * that the BP will be filled with the correct byteorder, salt, iv, and mac. */ static void dbuf_prepare_encrypted_dnode_leaf(dbuf_dirty_record_t *dr) { int err; dmu_buf_impl_t *db = dr->dr_dbuf; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT); ASSERT3U(db->db_level, ==, 0); if (!db->db_objset->os_raw_receive && arc_is_encrypted(db->db_buf)) { zbookmark_phys_t zb; /* * Unfortunately, there is currently no mechanism for * syncing context to handle decryption errors. An error * here is only possible if an attacker maliciously * changed a dnode block and updated the associated * checksums going up the block tree. */ SET_BOOKMARK(&zb, dmu_objset_id(db->db_objset), db->db.db_object, db->db_level, db->db_blkid); err = arc_untransform(db->db_buf, db->db_objset->os_spa, &zb, B_TRUE); if (err) panic("Invalid dnode block MAC"); } else if (dr->dt.dl.dr_has_raw_params) { (void) arc_release(dr->dt.dl.dr_data, db); arc_convert_to_raw(dr->dt.dl.dr_data, dmu_objset_id(db->db_objset), dr->dt.dl.dr_byteorder, DMU_OT_DNODE, dr->dt.dl.dr_salt, dr->dt.dl.dr_iv, dr->dt.dl.dr_mac); } } /* * dbuf_sync_indirect() is called recursively from dbuf_sync_list() so it * is critical the we not allow the compiler to inline this function in to * dbuf_sync_list() thereby drastically bloating the stack usage. */ noinline static void dbuf_sync_indirect(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn = dr->dr_dnode; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); ASSERT(db->db_level > 0); DBUF_VERIFY(db); /* Read the block if it hasn't been read yet. */ if (db->db_buf == NULL) { mutex_exit(&db->db_mtx); (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); mutex_enter(&db->db_mtx); } ASSERT3U(db->db_state, ==, DB_CACHED); ASSERT(db->db_buf != NULL); /* Indirect block size must match what the dnode thinks it is. */ ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); dbuf_check_blkptr(dn, db); /* Provide the pending dirty record to child dbufs */ db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, db->db_buf, tx); zio_t *zio = dr->dr_zio; mutex_enter(&dr->dt.di.dr_mtx); dbuf_sync_list(&dr->dt.di.dr_children, db->db_level - 1, tx); ASSERT(list_head(&dr->dt.di.dr_children) == NULL); mutex_exit(&dr->dt.di.dr_mtx); zio_nowait(zio); } /* * Verify that the size of the data in our bonus buffer does not exceed * its recorded size. * * The purpose of this verification is to catch any cases in development * where the size of a phys structure (i.e space_map_phys_t) grows and, * due to incorrect feature management, older pools expect to read more * data even though they didn't actually write it to begin with. * * For a example, this would catch an error in the feature logic where we * open an older pool and we expect to write the space map histogram of * a space map with size SPACE_MAP_SIZE_V0. */ static void dbuf_sync_leaf_verify_bonus_dnode(dbuf_dirty_record_t *dr) { #ifdef ZFS_DEBUG dnode_t *dn = dr->dr_dnode; /* * Encrypted bonus buffers can have data past their bonuslen. * Skip the verification of these blocks. */ if (DMU_OT_IS_ENCRYPTED(dn->dn_bonustype)) return; uint16_t bonuslen = dn->dn_phys->dn_bonuslen; uint16_t maxbonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots); ASSERT3U(bonuslen, <=, maxbonuslen); arc_buf_t *datap = dr->dt.dl.dr_data; char *datap_end = ((char *)datap) + bonuslen; char *datap_max = ((char *)datap) + maxbonuslen; /* ensure that everything is zero after our data */ for (; datap_end < datap_max; datap_end++) ASSERT(*datap_end == 0); #endif } static blkptr_t * dbuf_lightweight_bp(dbuf_dirty_record_t *dr) { /* This must be a lightweight dirty record. */ ASSERT3P(dr->dr_dbuf, ==, NULL); dnode_t *dn = dr->dr_dnode; if (dn->dn_phys->dn_nlevels == 1) { VERIFY3U(dr->dt.dll.dr_blkid, <, dn->dn_phys->dn_nblkptr); return (&dn->dn_phys->dn_blkptr[dr->dt.dll.dr_blkid]); } else { dmu_buf_impl_t *parent_db = dr->dr_parent->dr_dbuf; int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; VERIFY3U(parent_db->db_level, ==, 1); VERIFY3P(parent_db->db_dnode_handle->dnh_dnode, ==, dn); VERIFY3U(dr->dt.dll.dr_blkid >> epbs, ==, parent_db->db_blkid); blkptr_t *bp = parent_db->db.db_data; return (&bp[dr->dt.dll.dr_blkid & ((1 << epbs) - 1)]); } } static void dbuf_lightweight_ready(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; blkptr_t *bp = zio->io_bp; if (zio->io_error != 0) return; dnode_t *dn = dr->dr_dnode; blkptr_t *bp_orig = dbuf_lightweight_bp(dr); spa_t *spa = dmu_objset_spa(dn->dn_objset); int64_t delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig); dnode_diduse_space(dn, delta); uint64_t blkid = dr->dt.dll.dr_blkid; mutex_enter(&dn->dn_mtx); if (blkid > dn->dn_phys->dn_maxblkid) { ASSERT0(dn->dn_objset->os_raw_receive); dn->dn_phys->dn_maxblkid = blkid; } mutex_exit(&dn->dn_mtx); if (!BP_IS_EMBEDDED(bp)) { uint64_t fill = BP_IS_HOLE(bp) ? 0 : 1; BP_SET_FILL(bp, fill); } dmu_buf_impl_t *parent_db; EQUIV(dr->dr_parent == NULL, dn->dn_phys->dn_nlevels == 1); if (dr->dr_parent == NULL) { parent_db = dn->dn_dbuf; } else { parent_db = dr->dr_parent->dr_dbuf; } rw_enter(&parent_db->db_rwlock, RW_WRITER); *bp_orig = *bp; rw_exit(&parent_db->db_rwlock); } static void dbuf_lightweight_physdone(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dsl_pool_t *dp = spa_get_dsl(zio->io_spa); ASSERT3U(dr->dr_txg, ==, zio->io_txg); /* * The callback will be called io_phys_children times. Retire one * portion of our dirty space each time we are called. Any rounding * error will be cleaned up by dbuf_lightweight_done(). */ int delta = dr->dr_accounted / zio->io_phys_children; dsl_pool_undirty_space(dp, delta, zio->io_txg); } static void dbuf_lightweight_done(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; VERIFY0(zio->io_error); objset_t *os = dr->dr_dnode->dn_objset; dmu_tx_t *tx = os->os_synctx; if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) { ASSERT(BP_EQUAL(zio->io_bp, &zio->io_bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; (void) dsl_dataset_block_kill(ds, &zio->io_bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, zio->io_bp, tx); } /* * See comment in dbuf_write_done(). */ if (zio->io_phys_children == 0) { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted, zio->io_txg); } else { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted % zio->io_phys_children, zio->io_txg); } abd_free(dr->dt.dll.dr_abd); kmem_free(dr, sizeof (*dr)); } noinline static void dbuf_sync_lightweight(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dnode_t *dn = dr->dr_dnode; zio_t *pio; if (dn->dn_phys->dn_nlevels == 1) { pio = dn->dn_zio; } else { pio = dr->dr_parent->dr_zio; } zbookmark_phys_t zb = { .zb_objset = dmu_objset_id(dn->dn_objset), .zb_object = dn->dn_object, .zb_level = 0, .zb_blkid = dr->dt.dll.dr_blkid, }; /* * See comment in dbuf_write(). This is so that zio->io_bp_orig * will have the old BP in dbuf_lightweight_done(). */ dr->dr_bp_copy = *dbuf_lightweight_bp(dr); dr->dr_zio = zio_write(pio, dmu_objset_spa(dn->dn_objset), dmu_tx_get_txg(tx), &dr->dr_bp_copy, dr->dt.dll.dr_abd, dn->dn_datablksz, abd_get_size(dr->dt.dll.dr_abd), &dr->dt.dll.dr_props, dbuf_lightweight_ready, NULL, dbuf_lightweight_physdone, dbuf_lightweight_done, dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | dr->dt.dll.dr_flags, &zb); zio_nowait(dr->dr_zio); } /* * dbuf_sync_leaf() is called recursively from dbuf_sync_list() so it is * critical the we not allow the compiler to inline this function in to * dbuf_sync_list() thereby drastically bloating the stack usage. */ noinline static void dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { arc_buf_t **datap = &dr->dt.dl.dr_data; dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn = dr->dr_dnode; objset_t *os; uint64_t txg = tx->tx_txg; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); /* * To be synced, we must be dirtied. But we * might have been freed after the dirty. */ if (db->db_state == DB_UNCACHED) { /* This buffer has been freed since it was dirtied */ ASSERT(db->db.db_data == NULL); } else if (db->db_state == DB_FILL) { /* This buffer was freed and is now being re-filled */ ASSERT(db->db.db_data != dr->dt.dl.dr_data); } else { ASSERT(db->db_state == DB_CACHED || db->db_state == DB_NOFILL); } DBUF_VERIFY(db); if (db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (!(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) { /* * In the previous transaction group, the bonus buffer * was entirely used to store the attributes for the * dnode which overrode the dn_spill field. However, * when adding more attributes to the file a spill * block was required to hold the extra attributes. * * Make sure to clear the garbage left in the dn_spill * field from the previous attributes in the bonus * buffer. Otherwise, after writing out the spill * block to the new allocated dva, it will free * the old block pointed to by the invalid dn_spill. */ db->db_blkptr = NULL; } dn->dn_phys->dn_flags |= DNODE_FLAG_SPILL_BLKPTR; mutex_exit(&dn->dn_mtx); } /* * If this is a bonus buffer, simply copy the bonus data into the * dnode. It will be written out when the dnode is synced (and it * will be synced, since it must have been dirty for dbuf_sync to * be called). */ if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(dr->dr_dbuf == db); dbuf_sync_bonus(dr, tx); return; } os = dn->dn_objset; /* * This function may have dropped the db_mtx lock allowing a dmu_sync * operation to sneak in. As a result, we need to ensure that we * don't check the dr_override_state until we have returned from * dbuf_check_blkptr. */ dbuf_check_blkptr(dn, db); /* * If this buffer is in the middle of an immediate write, * wait for the synchronous IO to complete. */ while (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) { ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); cv_wait(&db->db_changed, &db->db_mtx); ASSERT(dr->dt.dl.dr_override_state != DR_NOT_OVERRIDDEN); } /* * If this is a dnode block, ensure it is appropriately encrypted * or decrypted, depending on what we are writing to it this txg. */ if (os->os_encrypted && dn->dn_object == DMU_META_DNODE_OBJECT) dbuf_prepare_encrypted_dnode_leaf(dr); if (db->db_state != DB_NOFILL && dn->dn_object != DMU_META_DNODE_OBJECT && zfs_refcount_count(&db->db_holds) > 1 && dr->dt.dl.dr_override_state != DR_OVERRIDDEN && *datap == db->db_buf) { /* * If this buffer is currently "in use" (i.e., there * are active holds and db_data still references it), * then make a copy before we start the write so that * any modifications from the open txg will not leak * into this write. * * NOTE: this copy does not need to be made for * objects only modified in the syncing context (e.g. * DNONE_DNODE blocks). */ *datap = dbuf_alloc_arcbuf_from_arcbuf(db, db->db_buf); bcopy(db->db.db_data, (*datap)->b_data, arc_buf_size(*datap)); } db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, *datap, tx); ASSERT(!list_link_active(&dr->dr_dirty_node)); if (dn->dn_object == DMU_META_DNODE_OBJECT) { list_insert_tail(&dn->dn_dirty_records[txg & TXG_MASK], dr); } else { zio_nowait(dr->dr_zio); } } void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx) { dbuf_dirty_record_t *dr; while ((dr = list_head(list))) { if (dr->dr_zio != NULL) { /* * If we find an already initialized zio then we * are processing the meta-dnode, and we have finished. * The dbufs for all dnodes are put back on the list * during processing, so that we can zio_wait() * these IOs after initiating all child IOs. */ ASSERT3U(dr->dr_dbuf->db.db_object, ==, DMU_META_DNODE_OBJECT); break; } list_remove(list, dr); if (dr->dr_dbuf == NULL) { dbuf_sync_lightweight(dr, tx); } else { if (dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { VERIFY3U(dr->dr_dbuf->db_level, ==, level); } if (dr->dr_dbuf->db_level > 0) dbuf_sync_indirect(dr, tx); else dbuf_sync_leaf(dr, tx); } } } /* ARGSUSED */ static void dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; spa_t *spa = zio->io_spa; int64_t delta; uint64_t fill = 0; int i; ASSERT3P(db->db_blkptr, !=, NULL); ASSERT3P(&db->db_data_pending->dr_bp_copy, ==, bp); DB_DNODE_ENTER(db); dn = DB_DNODE(db); delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig); dnode_diduse_space(dn, delta - zio->io_prev_space_delta); zio->io_prev_space_delta = delta; if (bp->blk_birth != 0) { ASSERT((db->db_blkid != DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_type) || (db->db_blkid == DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_bonustype) || BP_IS_EMBEDDED(bp)); ASSERT(BP_GET_LEVEL(bp) == db->db_level); } mutex_enter(&db->db_mtx); #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(bp)) && db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys)); } #endif if (db->db_level == 0) { mutex_enter(&dn->dn_mtx); if (db->db_blkid > dn->dn_phys->dn_maxblkid && db->db_blkid != DMU_SPILL_BLKID) { ASSERT0(db->db_objset->os_raw_receive); dn->dn_phys->dn_maxblkid = db->db_blkid; } mutex_exit(&dn->dn_mtx); if (dn->dn_type == DMU_OT_DNODE) { i = 0; while (i < db->db.db_size) { dnode_phys_t *dnp = (void *)(((char *)db->db.db_data) + i); i += DNODE_MIN_SIZE; if (dnp->dn_type != DMU_OT_NONE) { fill++; i += dnp->dn_extra_slots * DNODE_MIN_SIZE; } } } else { if (BP_IS_HOLE(bp)) { fill = 0; } else { fill = 1; } } } else { blkptr_t *ibp = db->db.db_data; ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); for (i = db->db.db_size >> SPA_BLKPTRSHIFT; i > 0; i--, ibp++) { if (BP_IS_HOLE(ibp)) continue; fill += BP_GET_FILL(ibp); } } DB_DNODE_EXIT(db); if (!BP_IS_EMBEDDED(bp)) BP_SET_FILL(bp, fill); mutex_exit(&db->db_mtx); db_lock_type_t dblt = dmu_buf_lock_parent(db, RW_WRITER, FTAG); *db->db_blkptr = *bp; dmu_buf_unlock_parent(db, dblt, FTAG); } /* ARGSUSED */ /* * This function gets called just prior to running through the compression * stage of the zio pipeline. If we're an indirect block comprised of only * holes, then we want this indirect to be compressed away to a hole. In * order to do that we must zero out any information about the holes that * this indirect points to prior to before we try to compress it. */ static void dbuf_write_children_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp; unsigned int epbs, i; ASSERT3U(db->db_level, >, 0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(epbs, <, 31); /* Determine if all our children are holes */ for (i = 0, bp = db->db.db_data; i < 1ULL << epbs; i++, bp++) { if (!BP_IS_HOLE(bp)) break; } /* * If all the children are holes, then zero them all out so that * we may get compressed away. */ if (i == 1ULL << epbs) { /* * We only found holes. Grab the rwlock to prevent * anybody from reading the blocks we're about to * zero out. */ rw_enter(&db->db_rwlock, RW_WRITER); bzero(db->db.db_data, db->db.db_size); rw_exit(&db->db_rwlock); } DB_DNODE_EXIT(db); } /* * The SPA will call this callback several times for each zio - once * for every physical child i/o (zio->io_phys_children times). This * allows the DMU to monitor the progress of each logical i/o. For example, * there may be 2 copies of an indirect block, or many fragments of a RAID-Z * block. There may be a long delay before all copies/fragments are completed, * so this callback allows us to retire dirty space gradually, as the physical * i/os complete. */ /* ARGSUSED */ static void dbuf_write_physdone(zio_t *zio, arc_buf_t *buf, void *arg) { dmu_buf_impl_t *db = arg; objset_t *os = db->db_objset; dsl_pool_t *dp = dmu_objset_pool(os); dbuf_dirty_record_t *dr; int delta = 0; dr = db->db_data_pending; ASSERT3U(dr->dr_txg, ==, zio->io_txg); /* * The callback will be called io_phys_children times. Retire one * portion of our dirty space each time we are called. Any rounding * error will be cleaned up by dbuf_write_done(). */ delta = dr->dr_accounted / zio->io_phys_children; dsl_pool_undirty_space(dp, delta, zio->io_txg); } /* ARGSUSED */ static void dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; blkptr_t *bp_orig = &zio->io_bp_orig; blkptr_t *bp = db->db_blkptr; objset_t *os = db->db_objset; dmu_tx_t *tx = os->os_synctx; ASSERT0(zio->io_error); ASSERT(db->db_blkptr == bp); /* * For nopwrites and rewrites we ensure that the bp matches our * original and bypass all the accounting. */ if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) { ASSERT(BP_EQUAL(bp, bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, bp, tx); } mutex_enter(&db->db_mtx); DBUF_VERIFY(db); dbuf_dirty_record_t *dr = db->db_data_pending; dnode_t *dn = dr->dr_dnode; ASSERT(!list_link_active(&dr->dr_dirty_node)); ASSERT(dr->dr_dbuf == db); ASSERT(list_next(&db->db_dirty_records, dr) == NULL); list_remove(&db->db_dirty_records, dr); #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(db->db_blkptr)) && db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys)); } #endif if (db->db_level == 0) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); if (db->db_state != DB_NOFILL) { if (dr->dt.dl.dr_data != db->db_buf) arc_buf_destroy(dr->dt.dl.dr_data, db); } } else { ASSERT(list_head(&dr->dt.di.dr_children) == NULL); ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); if (!BP_IS_HOLE(db->db_blkptr)) { int epbs __maybe_unused = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(db->db_blkid, <=, dn->dn_phys->dn_maxblkid >> (db->db_level * epbs)); ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, db->db.db_size); } mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } cv_broadcast(&db->db_changed); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; db->db_data_pending = NULL; dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg, B_FALSE); /* * If we didn't do a physical write in this ZIO and we * still ended up here, it means that the space of the * dbuf that we just released (and undirtied) above hasn't * been marked as undirtied in the pool's accounting. * * Thus, we undirty that space in the pool's view of the * world here. For physical writes this type of update * happens in dbuf_write_physdone(). * * If we did a physical write, cleanup any rounding errors * that came up due to writing multiple copies of a block * on disk [see dbuf_write_physdone()]. */ if (zio->io_phys_children == 0) { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted, zio->io_txg); } else { dsl_pool_undirty_space(dmu_objset_pool(os), dr->dr_accounted % zio->io_phys_children, zio->io_txg); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); } static void dbuf_write_nofill_ready(zio_t *zio) { dbuf_write_ready(zio, NULL, zio->io_private); } static void dbuf_write_nofill_done(zio_t *zio) { dbuf_write_done(zio, NULL, zio->io_private); } static void dbuf_write_override_ready(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; dbuf_write_ready(zio, NULL, db); } static void dbuf_write_override_done(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *obp = &dr->dt.dl.dr_overridden_by; mutex_enter(&db->db_mtx); if (!BP_EQUAL(zio->io_bp, obp)) { if (!BP_IS_HOLE(obp)) dsl_free(spa_get_dsl(zio->io_spa), zio->io_txg, obp); arc_release(dr->dt.dl.dr_data, db); } mutex_exit(&db->db_mtx); dbuf_write_done(zio, NULL, db); if (zio->io_abd != NULL) abd_free(zio->io_abd); } typedef struct dbuf_remap_impl_callback_arg { objset_t *drica_os; uint64_t drica_blk_birth; dmu_tx_t *drica_tx; } dbuf_remap_impl_callback_arg_t; static void dbuf_remap_impl_callback(uint64_t vdev, uint64_t offset, uint64_t size, void *arg) { dbuf_remap_impl_callback_arg_t *drica = arg; objset_t *os = drica->drica_os; spa_t *spa = dmu_objset_spa(os); dmu_tx_t *tx = drica->drica_tx; ASSERT(dsl_pool_sync_context(spa_get_dsl(spa))); if (os == spa_meta_objset(spa)) { spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx); } else { dsl_dataset_block_remapped(dmu_objset_ds(os), vdev, offset, size, drica->drica_blk_birth, tx); } } static void dbuf_remap_impl(dnode_t *dn, blkptr_t *bp, krwlock_t *rw, dmu_tx_t *tx) { blkptr_t bp_copy = *bp; spa_t *spa = dmu_objset_spa(dn->dn_objset); dbuf_remap_impl_callback_arg_t drica; ASSERT(dsl_pool_sync_context(spa_get_dsl(spa))); drica.drica_os = dn->dn_objset; drica.drica_blk_birth = bp->blk_birth; drica.drica_tx = tx; if (spa_remap_blkptr(spa, &bp_copy, dbuf_remap_impl_callback, &drica)) { /* * If the blkptr being remapped is tracked by a livelist, * then we need to make sure the livelist reflects the update. * First, cancel out the old blkptr by appending a 'FREE' * entry. Next, add an 'ALLOC' to track the new version. This * way we avoid trying to free an inaccurate blkptr at delete. * Note that embedded blkptrs are not tracked in livelists. */ if (dn->dn_objset != spa_meta_objset(spa)) { dsl_dataset_t *ds = dmu_objset_ds(dn->dn_objset); if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && bp->blk_birth > ds->ds_dir->dd_origin_txg) { ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(dsl_dir_is_clone(ds->ds_dir)); ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_LIVELIST)); bplist_append(&ds->ds_dir->dd_pending_frees, bp); bplist_append(&ds->ds_dir->dd_pending_allocs, &bp_copy); } } /* * The db_rwlock prevents dbuf_read_impl() from * dereferencing the BP while we are changing it. To * avoid lock contention, only grab it when we are actually * changing the BP. */ if (rw != NULL) rw_enter(rw, RW_WRITER); *bp = bp_copy; if (rw != NULL) rw_exit(rw); } } /* * Remap any existing BP's to concrete vdevs, if possible. */ static void dbuf_remap(dnode_t *dn, dmu_buf_impl_t *db, dmu_tx_t *tx) { spa_t *spa = dmu_objset_spa(db->db_objset); ASSERT(dsl_pool_sync_context(spa_get_dsl(spa))); if (!spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)) return; if (db->db_level > 0) { blkptr_t *bp = db->db.db_data; for (int i = 0; i < db->db.db_size >> SPA_BLKPTRSHIFT; i++) { dbuf_remap_impl(dn, &bp[i], &db->db_rwlock, tx); } } else if (db->db.db_object == DMU_META_DNODE_OBJECT) { dnode_phys_t *dnp = db->db.db_data; ASSERT3U(db->db_dnode_handle->dnh_dnode->dn_type, ==, DMU_OT_DNODE); for (int i = 0; i < db->db.db_size >> DNODE_SHIFT; i += dnp[i].dn_extra_slots + 1) { for (int j = 0; j < dnp[i].dn_nblkptr; j++) { krwlock_t *lock = (dn->dn_dbuf == NULL ? NULL : &dn->dn_dbuf->db_rwlock); dbuf_remap_impl(dn, &dnp[i].dn_blkptr[j], lock, tx); } } } } /* Issue I/O to commit a dirty buffer to disk. */ static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn = dr->dr_dnode; objset_t *os; dmu_buf_impl_t *parent = db->db_parent; uint64_t txg = tx->tx_txg; zbookmark_phys_t zb; zio_prop_t zp; zio_t *pio; /* parent I/O */ int wp_flag = 0; ASSERT(dmu_tx_is_syncing(tx)); os = dn->dn_objset; if (db->db_state != DB_NOFILL) { if (db->db_level > 0 || dn->dn_type == DMU_OT_DNODE) { /* * Private object buffers are released here rather * than in dbuf_dirty() since they are only modified * in the syncing context and we don't want the * overhead of making multiple copies of the data. */ if (BP_IS_HOLE(db->db_blkptr)) { arc_buf_thaw(data); } else { dbuf_release_bp(db); } dbuf_remap(dn, db, tx); } } if (parent != dn->dn_dbuf) { /* Our parent is an indirect block. */ /* We have a dirty parent that has been scheduled for write. */ ASSERT(parent && parent->db_data_pending); /* Our parent's buffer is one level closer to the dnode. */ ASSERT(db->db_level == parent->db_level-1); /* * We're about to modify our parent's db_data by modifying * our block pointer, so the parent must be released. */ ASSERT(arc_released(parent->db_buf)); pio = parent->db_data_pending->dr_zio; } else { /* Our parent is the dnode itself. */ ASSERT((db->db_level == dn->dn_phys->dn_nlevels-1 && db->db_blkid != DMU_SPILL_BLKID) || (db->db_blkid == DMU_SPILL_BLKID && db->db_level == 0)); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); pio = dn->dn_zio; } ASSERT(db->db_level == 0 || data == db->db_buf); ASSERT3U(db->db_blkptr->blk_birth, <=, txg); ASSERT(pio); SET_BOOKMARK(&zb, os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : DMU_META_OBJSET, db->db.db_object, db->db_level, db->db_blkid); if (db->db_blkid == DMU_SPILL_BLKID) wp_flag = WP_SPILL; wp_flag |= (db->db_state == DB_NOFILL) ? WP_NOFILL : 0; dmu_write_policy(os, dn, db->db_level, wp_flag, &zp); /* * We copy the blkptr now (rather than when we instantiate the dirty * record), because its value can change between open context and * syncing context. We do not need to hold dn_struct_rwlock to read * db_blkptr because we are in syncing context. */ dr->dr_bp_copy = *db->db_blkptr; if (db->db_level == 0 && dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { /* * The BP for this block has been provided by open context * (by dmu_sync() or dmu_buf_write_embedded()). */ abd_t *contents = (data != NULL) ? abd_get_from_buf(data->b_data, arc_buf_size(data)) : NULL; dr->dr_zio = zio_write(pio, os->os_spa, txg, &dr->dr_bp_copy, contents, db->db.db_size, db->db.db_size, &zp, dbuf_write_override_ready, NULL, NULL, dbuf_write_override_done, dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); mutex_enter(&db->db_mtx); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; zio_write_override(dr->dr_zio, &dr->dt.dl.dr_overridden_by, dr->dt.dl.dr_copies, dr->dt.dl.dr_nopwrite); mutex_exit(&db->db_mtx); } else if (db->db_state == DB_NOFILL) { ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF || zp.zp_checksum == ZIO_CHECKSUM_NOPARITY); dr->dr_zio = zio_write(pio, os->os_spa, txg, &dr->dr_bp_copy, NULL, db->db.db_size, db->db.db_size, &zp, dbuf_write_nofill_ready, NULL, NULL, dbuf_write_nofill_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb); } else { ASSERT(arc_released(data)); /* * For indirect blocks, we want to setup the children * ready callback so that we can properly handle an indirect * block that only contains holes. */ arc_write_done_func_t *children_ready_cb = NULL; if (db->db_level != 0) children_ready_cb = dbuf_write_children_ready; dr->dr_zio = arc_write(pio, os->os_spa, txg, &dr->dr_bp_copy, data, DBUF_IS_L2CACHEABLE(db), &zp, dbuf_write_ready, children_ready_cb, dbuf_write_physdone, dbuf_write_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); } } EXPORT_SYMBOL(dbuf_find); EXPORT_SYMBOL(dbuf_is_metadata); EXPORT_SYMBOL(dbuf_destroy); EXPORT_SYMBOL(dbuf_loan_arcbuf); EXPORT_SYMBOL(dbuf_whichblock); EXPORT_SYMBOL(dbuf_read); EXPORT_SYMBOL(dbuf_unoverride); EXPORT_SYMBOL(dbuf_free_range); EXPORT_SYMBOL(dbuf_new_size); EXPORT_SYMBOL(dbuf_release_bp); EXPORT_SYMBOL(dbuf_dirty); EXPORT_SYMBOL(dmu_buf_set_crypt_params); EXPORT_SYMBOL(dmu_buf_will_dirty); EXPORT_SYMBOL(dmu_buf_is_dirty); EXPORT_SYMBOL(dmu_buf_will_not_fill); EXPORT_SYMBOL(dmu_buf_will_fill); EXPORT_SYMBOL(dmu_buf_fill_done); EXPORT_SYMBOL(dmu_buf_rele); EXPORT_SYMBOL(dbuf_assign_arcbuf); EXPORT_SYMBOL(dbuf_prefetch); EXPORT_SYMBOL(dbuf_hold_impl); EXPORT_SYMBOL(dbuf_hold); EXPORT_SYMBOL(dbuf_hold_level); EXPORT_SYMBOL(dbuf_create_bonus); EXPORT_SYMBOL(dbuf_spill_set_blksz); EXPORT_SYMBOL(dbuf_rm_spill); EXPORT_SYMBOL(dbuf_add_ref); EXPORT_SYMBOL(dbuf_rele); EXPORT_SYMBOL(dbuf_rele_and_unlock); EXPORT_SYMBOL(dbuf_refcount); EXPORT_SYMBOL(dbuf_sync_list); EXPORT_SYMBOL(dmu_buf_set_user); EXPORT_SYMBOL(dmu_buf_set_user_ie); EXPORT_SYMBOL(dmu_buf_get_user); EXPORT_SYMBOL(dmu_buf_get_blkptr); /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, max_bytes, ULONG, ZMOD_RW, "Maximum size in bytes of the dbuf cache."); ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, hiwater_pct, UINT, ZMOD_RW, "Percentage over dbuf_cache_max_bytes when dbufs must be evicted " "directly."); ZFS_MODULE_PARAM(zfs_dbuf_cache, dbuf_cache_, lowater_pct, UINT, ZMOD_RW, "Percentage below dbuf_cache_max_bytes when the evict thread stops " "evicting dbufs."); ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, metadata_cache_max_bytes, ULONG, ZMOD_RW, "Maximum size in bytes of the dbuf metadata cache."); ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, cache_shift, INT, ZMOD_RW, "Set the size of the dbuf cache to a log2 fraction of arc size."); ZFS_MODULE_PARAM(zfs_dbuf, dbuf_, metadata_cache_shift, INT, ZMOD_RW, "Set the size of the dbuf metadata cache to a log2 fraction of arc " "size."); /* END CSTYLED */ diff --git a/sys/contrib/openzfs/module/zfs/dmu_zfetch.c b/sys/contrib/openzfs/module/zfs/dmu_zfetch.c index 3d7407016d2c..4a323fa990fe 100644 --- a/sys/contrib/openzfs/module/zfs/dmu_zfetch.c +++ b/sys/contrib/openzfs/module/zfs/dmu_zfetch.c @@ -1,515 +1,548 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* * Copyright (c) 2013, 2017 by Delphix. All rights reserved. */ #include #include #include #include #include #include #include +#include /* * This tunable disables predictive prefetch. Note that it leaves "prescient" * prefetch (e.g. prefetch for zfs send) intact. Unlike predictive prefetch, * prescient prefetch never issues i/os that end up not being needed, * so it can't hurt performance. */ int zfs_prefetch_disable = B_FALSE; /* max # of streams per zfetch */ unsigned int zfetch_max_streams = 8; /* min time before stream reclaim */ unsigned int zfetch_min_sec_reap = 2; /* max bytes to prefetch per stream (default 8MB) */ unsigned int zfetch_max_distance = 8 * 1024 * 1024; /* max bytes to prefetch indirects for per stream (default 64MB) */ unsigned int zfetch_max_idistance = 64 * 1024 * 1024; /* max number of bytes in an array_read in which we allow prefetching (1MB) */ unsigned long zfetch_array_rd_sz = 1024 * 1024; typedef struct zfetch_stats { kstat_named_t zfetchstat_hits; kstat_named_t zfetchstat_misses; kstat_named_t zfetchstat_max_streams; kstat_named_t zfetchstat_io_issued; } zfetch_stats_t; static zfetch_stats_t zfetch_stats = { { "hits", KSTAT_DATA_UINT64 }, { "misses", KSTAT_DATA_UINT64 }, { "max_streams", KSTAT_DATA_UINT64 }, { "io_issued", KSTAT_DATA_UINT64 }, }; -#define ZFETCHSTAT_BUMP(stat) \ - atomic_inc_64(&zfetch_stats.stat.value.ui64) +struct { + wmsum_t zfetchstat_hits; + wmsum_t zfetchstat_misses; + wmsum_t zfetchstat_max_streams; + wmsum_t zfetchstat_io_issued; +} zfetch_sums; + +#define ZFETCHSTAT_BUMP(stat) \ + wmsum_add(&zfetch_sums.stat, 1) #define ZFETCHSTAT_ADD(stat, val) \ - atomic_add_64(&zfetch_stats.stat.value.ui64, val) -#define ZFETCHSTAT_SET(stat, val) \ - zfetch_stats.stat.value.ui64 = val -#define ZFETCHSTAT_GET(stat) \ - zfetch_stats.stat.value.ui64 + wmsum_add(&zfetch_sums.stat, val) kstat_t *zfetch_ksp; +static int +zfetch_kstats_update(kstat_t *ksp, int rw) +{ + zfetch_stats_t *zs = ksp->ks_data; + + if (rw == KSTAT_WRITE) + return (EACCES); + zs->zfetchstat_hits.value.ui64 = + wmsum_value(&zfetch_sums.zfetchstat_hits); + zs->zfetchstat_misses.value.ui64 = + wmsum_value(&zfetch_sums.zfetchstat_misses); + zs->zfetchstat_max_streams.value.ui64 = + wmsum_value(&zfetch_sums.zfetchstat_max_streams); + zs->zfetchstat_io_issued.value.ui64 = + wmsum_value(&zfetch_sums.zfetchstat_io_issued); + return (0); +} + void zfetch_init(void) { + wmsum_init(&zfetch_sums.zfetchstat_hits, 0); + wmsum_init(&zfetch_sums.zfetchstat_misses, 0); + wmsum_init(&zfetch_sums.zfetchstat_max_streams, 0); + wmsum_init(&zfetch_sums.zfetchstat_io_issued, 0); + zfetch_ksp = kstat_create("zfs", 0, "zfetchstats", "misc", KSTAT_TYPE_NAMED, sizeof (zfetch_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (zfetch_ksp != NULL) { zfetch_ksp->ks_data = &zfetch_stats; + zfetch_ksp->ks_update = zfetch_kstats_update; kstat_install(zfetch_ksp); } } void zfetch_fini(void) { if (zfetch_ksp != NULL) { kstat_delete(zfetch_ksp); zfetch_ksp = NULL; } + + wmsum_fini(&zfetch_sums.zfetchstat_hits); + wmsum_fini(&zfetch_sums.zfetchstat_misses); + wmsum_fini(&zfetch_sums.zfetchstat_max_streams); + wmsum_fini(&zfetch_sums.zfetchstat_io_issued); } /* * This takes a pointer to a zfetch structure and a dnode. It performs the * necessary setup for the zfetch structure, grokking data from the * associated dnode. */ void dmu_zfetch_init(zfetch_t *zf, dnode_t *dno) { if (zf == NULL) return; zf->zf_dnode = dno; zf->zf_numstreams = 0; list_create(&zf->zf_stream, sizeof (zstream_t), offsetof(zstream_t, zs_node)); mutex_init(&zf->zf_lock, NULL, MUTEX_DEFAULT, NULL); } static void dmu_zfetch_stream_fini(zstream_t *zs) { ASSERT(!list_link_active(&zs->zs_node)); kmem_free(zs, sizeof (*zs)); } static void dmu_zfetch_stream_remove(zfetch_t *zf, zstream_t *zs) { ASSERT(MUTEX_HELD(&zf->zf_lock)); list_remove(&zf->zf_stream, zs); zf->zf_numstreams--; membar_producer(); if (zfs_refcount_remove(&zs->zs_refs, NULL) == 0) dmu_zfetch_stream_fini(zs); } /* * Clean-up state associated with a zfetch structure (e.g. destroy the * streams). This doesn't free the zfetch_t itself, that's left to the caller. */ void dmu_zfetch_fini(zfetch_t *zf) { zstream_t *zs; mutex_enter(&zf->zf_lock); while ((zs = list_head(&zf->zf_stream)) != NULL) dmu_zfetch_stream_remove(zf, zs); mutex_exit(&zf->zf_lock); list_destroy(&zf->zf_stream); mutex_destroy(&zf->zf_lock); zf->zf_dnode = NULL; } /* * If there aren't too many streams already, create a new stream. * The "blkid" argument is the next block that we expect this stream to access. * While we're here, clean up old streams (which haven't been * accessed for at least zfetch_min_sec_reap seconds). */ static void dmu_zfetch_stream_create(zfetch_t *zf, uint64_t blkid) { zstream_t *zs_next; hrtime_t now = gethrtime(); ASSERT(MUTEX_HELD(&zf->zf_lock)); /* * Clean up old streams. */ for (zstream_t *zs = list_head(&zf->zf_stream); zs != NULL; zs = zs_next) { zs_next = list_next(&zf->zf_stream, zs); /* * Skip if still active. 1 -- zf_stream reference. */ if (zfs_refcount_count(&zs->zs_refs) != 1) continue; if (((now - zs->zs_atime) / NANOSEC) > zfetch_min_sec_reap) dmu_zfetch_stream_remove(zf, zs); } /* * The maximum number of streams is normally zfetch_max_streams, * but for small files we lower it such that it's at least possible * for all the streams to be non-overlapping. * * If we are already at the maximum number of streams for this file, * even after removing old streams, then don't create this stream. */ uint32_t max_streams = MAX(1, MIN(zfetch_max_streams, zf->zf_dnode->dn_maxblkid * zf->zf_dnode->dn_datablksz / zfetch_max_distance)); if (zf->zf_numstreams >= max_streams) { ZFETCHSTAT_BUMP(zfetchstat_max_streams); return; } zstream_t *zs = kmem_zalloc(sizeof (*zs), KM_SLEEP); zs->zs_blkid = blkid; zs->zs_pf_blkid1 = blkid; zs->zs_pf_blkid = blkid; zs->zs_ipf_blkid1 = blkid; zs->zs_ipf_blkid = blkid; zs->zs_atime = now; zs->zs_fetch = zf; zs->zs_missed = B_FALSE; zfs_refcount_create(&zs->zs_callers); zfs_refcount_create(&zs->zs_refs); /* One reference for zf_stream. */ zfs_refcount_add(&zs->zs_refs, NULL); zf->zf_numstreams++; list_insert_head(&zf->zf_stream, zs); } static void dmu_zfetch_stream_done(void *arg, boolean_t io_issued) { zstream_t *zs = arg; if (zfs_refcount_remove(&zs->zs_refs, NULL) == 0) dmu_zfetch_stream_fini(zs); } /* * This is the predictive prefetch entry point. dmu_zfetch_prepare() * associates dnode access specified with blkid and nblks arguments with * prefetch stream, predicts further accesses based on that stats and returns * the stream pointer on success. That pointer must later be passed to * dmu_zfetch_run() to initiate the speculative prefetch for the stream and * release it. dmu_zfetch() is a wrapper for simple cases when window between * prediction and prefetch initiation is not needed. * fetch_data argument specifies whether actual data blocks should be fetched: * FALSE -- prefetch only indirect blocks for predicted data blocks; * TRUE -- prefetch predicted data blocks plus following indirect blocks. */ zstream_t * dmu_zfetch_prepare(zfetch_t *zf, uint64_t blkid, uint64_t nblks, boolean_t fetch_data, boolean_t have_lock) { zstream_t *zs; int64_t pf_start, ipf_start; int64_t pf_ahead_blks, max_blks; int max_dist_blks, pf_nblks, ipf_nblks; uint64_t end_of_access_blkid, maxblkid; end_of_access_blkid = blkid + nblks; spa_t *spa = zf->zf_dnode->dn_objset->os_spa; if (zfs_prefetch_disable) return (NULL); /* * If we haven't yet loaded the indirect vdevs' mappings, we * can only read from blocks that we carefully ensure are on * concrete vdevs (or previously-loaded indirect vdevs). So we * can't allow the predictive prefetcher to attempt reads of other * blocks (e.g. of the MOS's dnode object). */ if (!spa_indirect_vdevs_loaded(spa)) return (NULL); /* * As a fast path for small (single-block) files, ignore access * to the first block. */ if (!have_lock && blkid == 0) return (NULL); if (!have_lock) rw_enter(&zf->zf_dnode->dn_struct_rwlock, RW_READER); /* * A fast path for small files for which no prefetch will * happen. */ maxblkid = zf->zf_dnode->dn_maxblkid; if (maxblkid < 2) { if (!have_lock) rw_exit(&zf->zf_dnode->dn_struct_rwlock); return (NULL); } mutex_enter(&zf->zf_lock); /* * Find matching prefetch stream. Depending on whether the accesses * are block-aligned, first block of the new access may either follow * the last block of the previous access, or be equal to it. */ for (zs = list_head(&zf->zf_stream); zs != NULL; zs = list_next(&zf->zf_stream, zs)) { if (blkid == zs->zs_blkid) { break; } else if (blkid + 1 == zs->zs_blkid) { blkid++; nblks--; break; } } /* * If the file is ending, remove the matching stream if found. * If not found then it is too late to create a new one now. */ if (end_of_access_blkid >= maxblkid) { if (zs != NULL) dmu_zfetch_stream_remove(zf, zs); mutex_exit(&zf->zf_lock); if (!have_lock) rw_exit(&zf->zf_dnode->dn_struct_rwlock); return (NULL); } /* Exit if we already prefetched this block before. */ if (nblks == 0) { mutex_exit(&zf->zf_lock); if (!have_lock) rw_exit(&zf->zf_dnode->dn_struct_rwlock); return (NULL); } if (zs == NULL) { /* * This access is not part of any existing stream. Create * a new stream for it. */ dmu_zfetch_stream_create(zf, end_of_access_blkid); mutex_exit(&zf->zf_lock); if (!have_lock) rw_exit(&zf->zf_dnode->dn_struct_rwlock); ZFETCHSTAT_BUMP(zfetchstat_misses); return (NULL); } /* * This access was to a block that we issued a prefetch for on * behalf of this stream. Issue further prefetches for this stream. * * Normally, we start prefetching where we stopped * prefetching last (zs_pf_blkid). But when we get our first * hit on this stream, zs_pf_blkid == zs_blkid, we don't * want to prefetch the block we just accessed. In this case, * start just after the block we just accessed. */ pf_start = MAX(zs->zs_pf_blkid, end_of_access_blkid); if (zs->zs_pf_blkid1 < end_of_access_blkid) zs->zs_pf_blkid1 = end_of_access_blkid; if (zs->zs_ipf_blkid1 < end_of_access_blkid) zs->zs_ipf_blkid1 = end_of_access_blkid; /* * Double our amount of prefetched data, but don't let the * prefetch get further ahead than zfetch_max_distance. */ if (fetch_data) { max_dist_blks = zfetch_max_distance >> zf->zf_dnode->dn_datablkshift; /* * Previously, we were (zs_pf_blkid - blkid) ahead. We * want to now be double that, so read that amount again, * plus the amount we are catching up by (i.e. the amount * read just now). */ pf_ahead_blks = zs->zs_pf_blkid - blkid + nblks; max_blks = max_dist_blks - (pf_start - end_of_access_blkid); pf_nblks = MIN(pf_ahead_blks, max_blks); } else { pf_nblks = 0; } zs->zs_pf_blkid = pf_start + pf_nblks; /* * Do the same for indirects, starting from where we stopped last, * or where we will stop reading data blocks (and the indirects * that point to them). */ ipf_start = MAX(zs->zs_ipf_blkid, zs->zs_pf_blkid); max_dist_blks = zfetch_max_idistance >> zf->zf_dnode->dn_datablkshift; /* * We want to double our distance ahead of the data prefetch * (or reader, if we are not prefetching data). Previously, we * were (zs_ipf_blkid - blkid) ahead. To double that, we read * that amount again, plus the amount we are catching up by * (i.e. the amount read now + the amount of data prefetched now). */ pf_ahead_blks = zs->zs_ipf_blkid - blkid + nblks + pf_nblks; max_blks = max_dist_blks - (ipf_start - zs->zs_pf_blkid); ipf_nblks = MIN(pf_ahead_blks, max_blks); zs->zs_ipf_blkid = ipf_start + ipf_nblks; zs->zs_blkid = end_of_access_blkid; /* Protect the stream from reclamation. */ zs->zs_atime = gethrtime(); zfs_refcount_add(&zs->zs_refs, NULL); /* Count concurrent callers. */ zfs_refcount_add(&zs->zs_callers, NULL); mutex_exit(&zf->zf_lock); if (!have_lock) rw_exit(&zf->zf_dnode->dn_struct_rwlock); ZFETCHSTAT_BUMP(zfetchstat_hits); return (zs); } void dmu_zfetch_run(zstream_t *zs, boolean_t missed, boolean_t have_lock) { zfetch_t *zf = zs->zs_fetch; int64_t pf_start, pf_end, ipf_start, ipf_end; int epbs, issued; if (missed) zs->zs_missed = missed; /* * Postpone the prefetch if there are more concurrent callers. * It happens when multiple requests are waiting for the same * indirect block. The last one will run the prefetch for all. */ if (zfs_refcount_remove(&zs->zs_callers, NULL) != 0) { /* Drop reference taken in dmu_zfetch_prepare(). */ if (zfs_refcount_remove(&zs->zs_refs, NULL) == 0) dmu_zfetch_stream_fini(zs); return; } mutex_enter(&zf->zf_lock); if (zs->zs_missed) { pf_start = zs->zs_pf_blkid1; pf_end = zs->zs_pf_blkid1 = zs->zs_pf_blkid; } else { pf_start = pf_end = 0; } ipf_start = MAX(zs->zs_pf_blkid1, zs->zs_ipf_blkid1); ipf_end = zs->zs_ipf_blkid1 = zs->zs_ipf_blkid; mutex_exit(&zf->zf_lock); ASSERT3S(pf_start, <=, pf_end); ASSERT3S(ipf_start, <=, ipf_end); epbs = zf->zf_dnode->dn_indblkshift - SPA_BLKPTRSHIFT; ipf_start = P2ROUNDUP(ipf_start, 1 << epbs) >> epbs; ipf_end = P2ROUNDUP(ipf_end, 1 << epbs) >> epbs; ASSERT3S(ipf_start, <=, ipf_end); issued = pf_end - pf_start + ipf_end - ipf_start; if (issued > 1) { /* More references on top of taken in dmu_zfetch_prepare(). */ zfs_refcount_add_many(&zs->zs_refs, issued - 1, NULL); } else if (issued == 0) { /* Some other thread has done our work, so drop the ref. */ if (zfs_refcount_remove(&zs->zs_refs, NULL) == 0) dmu_zfetch_stream_fini(zs); return; } if (!have_lock) rw_enter(&zf->zf_dnode->dn_struct_rwlock, RW_READER); issued = 0; for (int64_t blk = pf_start; blk < pf_end; blk++) { issued += dbuf_prefetch_impl(zf->zf_dnode, 0, blk, ZIO_PRIORITY_ASYNC_READ, ARC_FLAG_PREDICTIVE_PREFETCH, dmu_zfetch_stream_done, zs); } for (int64_t iblk = ipf_start; iblk < ipf_end; iblk++) { issued += dbuf_prefetch_impl(zf->zf_dnode, 1, iblk, ZIO_PRIORITY_ASYNC_READ, ARC_FLAG_PREDICTIVE_PREFETCH, dmu_zfetch_stream_done, zs); } if (!have_lock) rw_exit(&zf->zf_dnode->dn_struct_rwlock); if (issued) ZFETCHSTAT_ADD(zfetchstat_io_issued, issued); } void dmu_zfetch(zfetch_t *zf, uint64_t blkid, uint64_t nblks, boolean_t fetch_data, boolean_t missed, boolean_t have_lock) { zstream_t *zs; zs = dmu_zfetch_prepare(zf, blkid, nblks, fetch_data, have_lock); if (zs) dmu_zfetch_run(zs, missed, have_lock); } /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs_prefetch, zfs_prefetch_, disable, INT, ZMOD_RW, "Disable all ZFS prefetching"); ZFS_MODULE_PARAM(zfs_prefetch, zfetch_, max_streams, UINT, ZMOD_RW, "Max number of streams per zfetch"); ZFS_MODULE_PARAM(zfs_prefetch, zfetch_, min_sec_reap, UINT, ZMOD_RW, "Min time before stream reclaim"); ZFS_MODULE_PARAM(zfs_prefetch, zfetch_, max_distance, UINT, ZMOD_RW, "Max bytes to prefetch per stream"); ZFS_MODULE_PARAM(zfs_prefetch, zfetch_, max_idistance, UINT, ZMOD_RW, "Max bytes to prefetch indirects for per stream"); ZFS_MODULE_PARAM(zfs_prefetch, zfetch_, array_rd_sz, ULONG, ZMOD_RW, "Number of bytes in a array_read"); /* END CSTYLED */ diff --git a/sys/contrib/openzfs/module/zfs/vdev_draid.c b/sys/contrib/openzfs/module/zfs/vdev_draid.c index 20b1457f0ce8..b8f82d52e8f0 100644 --- a/sys/contrib/openzfs/module/zfs/vdev_draid.c +++ b/sys/contrib/openzfs/module/zfs/vdev_draid.c @@ -1,2783 +1,2784 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2018 Intel Corporation. * Copyright (c) 2020 by Lawrence Livermore National Security, LLC. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef ZFS_DEBUG #include /* For vdev_xlate() in vdev_draid_io_verify() */ #endif /* * dRAID is a distributed spare implementation for ZFS. A dRAID vdev is * comprised of multiple raidz redundancy groups which are spread over the * dRAID children. To ensure an even distribution, and avoid hot spots, a * permutation mapping is applied to the order of the dRAID children. * This mixing effectively distributes the parity columns evenly over all * of the disks in the dRAID. * * This is beneficial because it means when resilvering all of the disks * can participate thereby increasing the available IOPs and bandwidth. * Furthermore, by reserving a small fraction of each child's total capacity * virtual distributed spare disks can be created. These spares similarly * benefit from the performance gains of spanning all of the children. The * consequence of which is that resilvering to a distributed spare can * substantially reduce the time required to restore full parity to pool * with a failed disks. * * === dRAID group layout === * * First, let's define a "row" in the configuration to be a 16M chunk from * each physical drive at the same offset. This is the minimum allowable * size since it must be possible to store a full 16M block when there is * only a single data column. Next, we define a "group" to be a set of * sequential disks containing both the parity and data columns. We allow * groups to span multiple rows in order to align any group size to any * number of physical drives. Finally, a "slice" is comprised of the rows * which contain the target number of groups. The permutation mappings * are applied in a round robin fashion to each slice. * * Given D+P drives in a group (including parity drives) and C-S physical * drives (not including the spare drives), we can distribute the groups * across R rows without remainder by selecting the least common multiple * of D+P and C-S as the number of groups; i.e. ngroups = LCM(D+P, C-S). * * In the example below, there are C=14 physical drives in the configuration * with S=2 drives worth of spare capacity. Each group has a width of 9 * which includes D=8 data and P=1 parity drive. There are 4 groups and * 3 rows per slice. Each group has a size of 144M (16M * 9) and a slice * size is 576M (144M * 4). When allocating from a dRAID each group is * filled before moving on to the next as show in slice0 below. * * data disks (8 data + 1 parity) spares (2) * +===+===+===+===+===+===+===+===+===+===+===+===+===+===+ * ^ | 2 | 6 | 1 | 11| 4 | 0 | 7 | 10| 8 | 9 | 13| 5 | 12| 3 | device map 0 * | +===+===+===+===+===+===+===+===+===+===+===+===+===+===+ * | | group 0 | group 1..| | * | +-----------------------------------+-----------+-------| * | | 0 1 2 3 4 5 6 7 8 | 36 37 38| | r * | | 9 10 11 12 13 14 15 16 17| 45 46 47| | o * | | 18 19 20 21 22 23 24 25 26| 54 55 56| | w * | 27 28 29 30 31 32 33 34 35| 63 64 65| | 0 * s +-----------------------+-----------------------+-------+ * l | ..group 1 | group 2.. | | * i +-----------------------+-----------------------+-------+ * c | 39 40 41 42 43 44| 72 73 74 75 76 77| | r * e | 48 49 50 51 52 53| 81 82 83 84 85 86| | o * 0 | 57 58 59 60 61 62| 90 91 92 93 94 95| | w * | 66 67 68 69 70 71| 99 100 101 102 103 104| | 1 * | +-----------+-----------+-----------------------+-------+ * | |..group 2 | group 3 | | * | +-----------+-----------+-----------------------+-------+ * | | 78 79 80|108 109 110 111 112 113 114 115 116| | r * | | 87 88 89|117 118 119 120 121 122 123 124 125| | o * | | 96 97 98|126 127 128 129 130 131 132 133 134| | w * v |105 106 107|135 136 137 138 139 140 141 142 143| | 2 * +===+===+===+===+===+===+===+===+===+===+===+===+===+===+ * | 9 | 11| 12| 2 | 4 | 1 | 3 | 0 | 10| 13| 8 | 5 | 6 | 7 | device map 1 * s +===+===+===+===+===+===+===+===+===+===+===+===+===+===+ * l | group 4 | group 5..| | row 3 * i +-----------------------+-----------+-----------+-------| * c | ..group 5 | group 6.. | | row 4 * e +-----------+-----------+-----------------------+-------+ * 1 |..group 6 | group 7 | | row 5 * +===+===+===+===+===+===+===+===+===+===+===+===+===+===+ * | 3 | 5 | 10| 8 | 6 | 11| 12| 0 | 2 | 4 | 7 | 1 | 9 | 13| device map 2 * s +===+===+===+===+===+===+===+===+===+===+===+===+===+===+ * l | group 8 | group 9..| | row 6 * i +-----------------------------------------------+-------| * c | ..group 9 | group 10.. | | row 7 * e +-----------------------+-----------------------+-------+ * 2 |..group 10 | group 11 | | row 8 * +-----------+-----------------------------------+-------+ * * This layout has several advantages over requiring that each row contain * a whole number of groups. * * 1. The group count is not a relevant parameter when defining a dRAID * layout. Only the group width is needed, and *all* groups will have * the desired size. * * 2. All possible group widths (<= physical disk count) can be supported. * * 3. The logic within vdev_draid.c is simplified when the group width is * the same for all groups (although some of the logic around computing * permutation numbers and drive offsets is more complicated). * * N.B. The following array describes all valid dRAID permutation maps. * Each row is used to generate a permutation map for a different number * of children from a unique seed. The seeds were generated and carefully * evaluated by the 'draid' utility in order to provide balanced mappings. * In addition to the seed a checksum of the in-memory mapping is stored * for verification. * * The imbalance ratio of a given failure (e.g. 5 disks wide, child 3 failed, * with a given permutation map) is the ratio of the amounts of I/O that will * be sent to the least and most busy disks when resilvering. The average * imbalance ratio (of a given number of disks and permutation map) is the * average of the ratios of all possible single and double disk failures. * * In order to achieve a low imbalance ratio the number of permutations in * the mapping must be significantly larger than the number of children. * For dRAID the number of permutations has been limited to 512 to minimize * the map size. This does result in a gradually increasing imbalance ratio * as seen in the table below. Increasing the number of permutations for * larger child counts would reduce the imbalance ratio. However, in practice * when there are a large number of children each child is responsible for * fewer total IOs so it's less of a concern. * * Note these values are hard coded and must never be changed. Existing * pools depend on the same mapping always being generated in order to * read and write from the correct locations. Any change would make * existing pools completely inaccessible. */ static const draid_map_t draid_maps[VDEV_DRAID_MAX_MAPS] = { { 2, 256, 0x89ef3dabbcc7de37, 0x00000000433d433d }, /* 1.000 */ { 3, 256, 0x89a57f3de98121b4, 0x00000000bcd8b7b5 }, /* 1.000 */ { 4, 256, 0xc9ea9ec82340c885, 0x00000001819d7c69 }, /* 1.000 */ { 5, 256, 0xf46733b7f4d47dfd, 0x00000002a1648d74 }, /* 1.010 */ { 6, 256, 0x88c3c62d8585b362, 0x00000003d3b0c2c4 }, /* 1.031 */ { 7, 256, 0x3a65d809b4d1b9d5, 0x000000055c4183ee }, /* 1.043 */ { 8, 256, 0xe98930e3c5d2e90a, 0x00000006edfb0329 }, /* 1.059 */ { 9, 256, 0x5a5430036b982ccb, 0x00000008ceaf6934 }, /* 1.056 */ { 10, 256, 0x92bf389e9eadac74, 0x0000000b26668c09 }, /* 1.072 */ { 11, 256, 0x74ccebf1dcf3ae80, 0x0000000dd691358c }, /* 1.083 */ { 12, 256, 0x8847e41a1a9f5671, 0x00000010a0c63c8e }, /* 1.097 */ { 13, 256, 0x7481b56debf0e637, 0x0000001424121fe4 }, /* 1.100 */ { 14, 256, 0x559b8c44065f8967, 0x00000016ab2ff079 }, /* 1.121 */ { 15, 256, 0x34c49545a2ee7f01, 0x0000001a6028efd6 }, /* 1.103 */ { 16, 256, 0xb85f4fa81a7698f7, 0x0000001e95ff5e66 }, /* 1.111 */ { 17, 256, 0x6353e47b7e47aba0, 0x00000021a81fa0fe }, /* 1.133 */ { 18, 256, 0xaa549746b1cbb81c, 0x00000026f02494c9 }, /* 1.131 */ { 19, 256, 0x892e343f2f31d690, 0x00000029eb392835 }, /* 1.130 */ { 20, 256, 0x76914824db98cc3f, 0x0000003004f31a7c }, /* 1.141 */ { 21, 256, 0x4b3cbabf9cfb1d0f, 0x00000036363a2408 }, /* 1.139 */ { 22, 256, 0xf45c77abb4f035d4, 0x00000038dd0f3e84 }, /* 1.150 */ { 23, 256, 0x5e18bd7f3fd4baf4, 0x0000003f0660391f }, /* 1.174 */ { 24, 256, 0xa7b3a4d285d6503b, 0x000000443dfc9ff6 }, /* 1.168 */ { 25, 256, 0x56ac7dd967521f5a, 0x0000004b03a87eb7 }, /* 1.180 */ { 26, 256, 0x3a42dfda4eb880f7, 0x000000522c719bba }, /* 1.226 */ { 27, 256, 0xd200d2fc6b54bf60, 0x0000005760b4fdf5 }, /* 1.228 */ { 28, 256, 0xc52605bbd486c546, 0x0000005e00d8f74c }, /* 1.217 */ { 29, 256, 0xc761779e63cd762f, 0x00000067be3cd85c }, /* 1.239 */ { 30, 256, 0xca577b1e07f85ca5, 0x0000006f5517f3e4 }, /* 1.238 */ { 31, 256, 0xfd50a593c518b3d4, 0x0000007370e7778f }, /* 1.273 */ { 32, 512, 0xc6c87ba5b042650b, 0x000000f7eb08a156 }, /* 1.191 */ { 33, 512, 0xc3880d0c9d458304, 0x0000010734b5d160 }, /* 1.199 */ { 34, 512, 0xe920927e4d8b2c97, 0x00000118c1edbce0 }, /* 1.195 */ { 35, 512, 0x8da7fcda87bde316, 0x0000012a3e9f9110 }, /* 1.201 */ { 36, 512, 0xcf09937491514a29, 0x0000013bd6a24bef }, /* 1.194 */ { 37, 512, 0x9b5abbf345cbd7cc, 0x0000014b9d90fac3 }, /* 1.237 */ { 38, 512, 0x506312a44668d6a9, 0x0000015e1b5f6148 }, /* 1.242 */ { 39, 512, 0x71659ede62b4755f, 0x00000173ef029bcd }, /* 1.231 */ { 40, 512, 0xa7fde73fb74cf2d7, 0x000001866fb72748 }, /* 1.233 */ { 41, 512, 0x19e8b461a1dea1d3, 0x000001a046f76b23 }, /* 1.271 */ { 42, 512, 0x031c9b868cc3e976, 0x000001afa64c49d3 }, /* 1.263 */ { 43, 512, 0xbaa5125faa781854, 0x000001c76789e278 }, /* 1.270 */ { 44, 512, 0x4ed55052550d721b, 0x000001d800ccd8eb }, /* 1.281 */ { 45, 512, 0x0fd63ddbdff90677, 0x000001f08ad59ed2 }, /* 1.282 */ { 46, 512, 0x36d66546de7fdd6f, 0x000002016f09574b }, /* 1.286 */ { 47, 512, 0x99f997e7eafb69d7, 0x0000021e42e47cb6 }, /* 1.329 */ { 48, 512, 0xbecd9c2571312c5d, 0x000002320fe2872b }, /* 1.286 */ { 49, 512, 0xd97371329e488a32, 0x0000024cd73f2ca7 }, /* 1.322 */ { 50, 512, 0x30e9b136670749ee, 0x000002681c83b0e0 }, /* 1.335 */ { 51, 512, 0x11ad6bc8f47aaeb4, 0x0000027e9261b5d5 }, /* 1.305 */ { 52, 512, 0x68e445300af432c1, 0x0000029aa0eb7dbf }, /* 1.330 */ { 53, 512, 0x910fb561657ea98c, 0x000002b3dca04853 }, /* 1.365 */ { 54, 512, 0xd619693d8ce5e7a5, 0x000002cc280e9c97 }, /* 1.334 */ { 55, 512, 0x24e281f564dbb60a, 0x000002e9fa842713 }, /* 1.364 */ { 56, 512, 0x947a7d3bdaab44c5, 0x000003046680f72e }, /* 1.374 */ { 57, 512, 0x2d44fec9c093e0de, 0x00000324198ba810 }, /* 1.363 */ { 58, 512, 0x87743c272d29bb4c, 0x0000033ec48c9ac9 }, /* 1.401 */ { 59, 512, 0x96aa3b6f67f5d923, 0x0000034faead902c }, /* 1.392 */ { 60, 512, 0x94a4f1faf520b0d3, 0x0000037d713ab005 }, /* 1.360 */ { 61, 512, 0xb13ed3a272f711a2, 0x00000397368f3cbd }, /* 1.396 */ { 62, 512, 0x3b1b11805fa4a64a, 0x000003b8a5e2840c }, /* 1.453 */ { 63, 512, 0x4c74caad9172ba71, 0x000003d4be280290 }, /* 1.437 */ { 64, 512, 0x035ff643923dd29e, 0x000003fad6c355e1 }, /* 1.402 */ { 65, 512, 0x768e9171b11abd3c, 0x0000040eb07fed20 }, /* 1.459 */ { 66, 512, 0x75880e6f78a13ddd, 0x000004433d6acf14 }, /* 1.423 */ { 67, 512, 0x910b9714f698a877, 0x00000451ea65d5db }, /* 1.447 */ { 68, 512, 0x87f5db6f9fdcf5c7, 0x000004732169e3f7 }, /* 1.450 */ { 69, 512, 0x836d4968fbaa3706, 0x000004954068a380 }, /* 1.455 */ { 70, 512, 0xc567d73a036421ab, 0x000004bd7cb7bd3d }, /* 1.463 */ { 71, 512, 0x619df40f240b8fed, 0x000004e376c2e972 }, /* 1.463 */ { 72, 512, 0x42763a680d5bed8e, 0x000005084275c680 }, /* 1.452 */ { 73, 512, 0x5866f064b3230431, 0x0000052906f2c9ab }, /* 1.498 */ { 74, 512, 0x9fa08548b1621a44, 0x0000054708019247 }, /* 1.526 */ { 75, 512, 0xb6053078ce0fc303, 0x00000572cc5c72b0 }, /* 1.491 */ { 76, 512, 0x4a7aad7bf3890923, 0x0000058e987bc8e9 }, /* 1.470 */ { 77, 512, 0xe165613fd75b5a53, 0x000005c20473a211 }, /* 1.527 */ { 78, 512, 0x3ff154ac878163a6, 0x000005d659194bf3 }, /* 1.509 */ { 79, 512, 0x24b93ade0aa8a532, 0x0000060a201c4f8e }, /* 1.569 */ { 80, 512, 0xc18e2d14cd9bb554, 0x0000062c55cfe48c }, /* 1.555 */ { 81, 512, 0x98cc78302feb58b6, 0x0000066656a07194 }, /* 1.509 */ { 82, 512, 0xc6c5fd5a2abc0543, 0x0000067cff94fbf8 }, /* 1.596 */ { 83, 512, 0xa7962f514acbba21, 0x000006ab7b5afa2e }, /* 1.568 */ { 84, 512, 0xba02545069ddc6dc, 0x000006d19861364f }, /* 1.541 */ { 85, 512, 0x447c73192c35073e, 0x000006fce315ce35 }, /* 1.623 */ { 86, 512, 0x48beef9e2d42b0c2, 0x00000720a8e38b6b }, /* 1.620 */ { 87, 512, 0x4874cf98541a35e0, 0x00000758382a2273 }, /* 1.597 */ { 88, 512, 0xad4cf8333a31127a, 0x00000781e1651b1b }, /* 1.575 */ { 89, 512, 0x47ae4859d57888c1, 0x000007b27edbe5bc }, /* 1.627 */ { 90, 512, 0x06f7723cfe5d1891, 0x000007dc2a96d8eb }, /* 1.596 */ { 91, 512, 0xd4e44218d660576d, 0x0000080ac46f02d5 }, /* 1.622 */ { 92, 512, 0x7066702b0d5be1f2, 0x00000832c96d154e }, /* 1.695 */ { 93, 512, 0x011209b4f9e11fb9, 0x0000085eefda104c }, /* 1.605 */ { 94, 512, 0x47ffba30a0b35708, 0x00000899badc32dc }, /* 1.625 */ { 95, 512, 0x1a95a6ac4538aaa8, 0x000008b6b69a42b2 }, /* 1.687 */ { 96, 512, 0xbda2b239bb2008eb, 0x000008f22d2de38a }, /* 1.621 */ { 97, 512, 0x7ffa0bea90355c6c, 0x0000092e5b23b816 }, /* 1.699 */ { 98, 512, 0x1d56ba34be426795, 0x0000094f482e5d1b }, /* 1.688 */ { 99, 512, 0x0aa89d45c502e93d, 0x00000977d94a98ce }, /* 1.642 */ { 100, 512, 0x54369449f6857774, 0x000009c06c9b34cc }, /* 1.683 */ { 101, 512, 0xf7d4dd8445b46765, 0x000009e5dc542259 }, /* 1.755 */ { 102, 512, 0xfa8866312f169469, 0x00000a16b54eae93 }, /* 1.692 */ { 103, 512, 0xd8a5aea08aef3ff9, 0x00000a381d2cbfe7 }, /* 1.747 */ { 104, 512, 0x66bcd2c3d5f9ef0e, 0x00000a8191817be7 }, /* 1.751 */ { 105, 512, 0x3fb13a47a012ec81, 0x00000ab562b9a254 }, /* 1.751 */ { 106, 512, 0x43100f01c9e5e3ca, 0x00000aeee84c185f }, /* 1.726 */ { 107, 512, 0xca09c50ccee2d054, 0x00000b1c359c047d }, /* 1.788 */ { 108, 512, 0xd7176732ac503f9b, 0x00000b578bc52a73 }, /* 1.740 */ { 109, 512, 0xed206e51f8d9422d, 0x00000b8083e0d960 }, /* 1.780 */ { 110, 512, 0x17ead5dc6ba0dcd6, 0x00000bcfb1a32ca8 }, /* 1.836 */ { 111, 512, 0x5f1dc21e38a969eb, 0x00000c0171becdd6 }, /* 1.778 */ { 112, 512, 0xddaa973de33ec528, 0x00000c3edaba4b95 }, /* 1.831 */ { 113, 512, 0x2a5eccd7735a3630, 0x00000c630664e7df }, /* 1.825 */ { 114, 512, 0xafcccee5c0b71446, 0x00000cb65392f6e4 }, /* 1.826 */ { 115, 512, 0x8fa30c5e7b147e27, 0x00000cd4db391e55 }, /* 1.843 */ { 116, 512, 0x5afe0711fdfafd82, 0x00000d08cb4ec35d }, /* 1.826 */ { 117, 512, 0x533a6090238afd4c, 0x00000d336f115d1b }, /* 1.803 */ { 118, 512, 0x90cf11b595e39a84, 0x00000d8e041c2048 }, /* 1.857 */ { 119, 512, 0x0d61a3b809444009, 0x00000dcb798afe35 }, /* 1.877 */ { 120, 512, 0x7f34da0f54b0d114, 0x00000df3922664e1 }, /* 1.849 */ { 121, 512, 0xa52258d5b72f6551, 0x00000e4d37a9872d }, /* 1.867 */ { 122, 512, 0xc1de54d7672878db, 0x00000e6583a94cf6 }, /* 1.978 */ { 123, 512, 0x1d03354316a414ab, 0x00000ebffc50308d }, /* 1.947 */ { 124, 512, 0xcebdcc377665412c, 0x00000edee1997cea }, /* 1.865 */ { 125, 512, 0x4ddd4c04b1a12344, 0x00000f21d64b373f }, /* 1.881 */ { 126, 512, 0x64fc8f94e3973658, 0x00000f8f87a8896b }, /* 1.882 */ { 127, 512, 0x68765f78034a334e, 0x00000fb8fe62197e }, /* 1.867 */ { 128, 512, 0xaf36b871a303e816, 0x00000fec6f3afb1e }, /* 1.972 */ { 129, 512, 0x2a4cbf73866c3a28, 0x00001027febfe4e5 }, /* 1.896 */ { 130, 512, 0x9cb128aacdcd3b2f, 0x0000106aa8ac569d }, /* 1.965 */ { 131, 512, 0x5511d41c55869124, 0x000010bbd755ddf1 }, /* 1.963 */ { 132, 512, 0x42f92461937f284a, 0x000010fb8bceb3b5 }, /* 1.925 */ { 133, 512, 0xe2d89a1cf6f1f287, 0x0000114cf5331e34 }, /* 1.862 */ { 134, 512, 0xdc631a038956200e, 0x0000116428d2adc5 }, /* 2.042 */ { 135, 512, 0xb2e5ac222cd236be, 0x000011ca88e4d4d2 }, /* 1.935 */ { 136, 512, 0xbc7d8236655d88e7, 0x000011e39cb94e66 }, /* 2.005 */ { 137, 512, 0x073e02d88d2d8e75, 0x0000123136c7933c }, /* 2.041 */ { 138, 512, 0x3ddb9c3873166be0, 0x00001280e4ec6d52 }, /* 1.997 */ { 139, 512, 0x7d3b1a845420e1b5, 0x000012c2e7cd6a44 }, /* 1.996 */ { 140, 512, 0x60102308aa7b2a6c, 0x000012fc490e6c7d }, /* 2.053 */ { 141, 512, 0xdb22bb2f9eb894aa, 0x00001343f5a85a1a }, /* 1.971 */ { 142, 512, 0xd853f879a13b1606, 0x000013bb7d5f9048 }, /* 2.018 */ { 143, 512, 0x001620a03f804b1d, 0x000013e74cc794fd }, /* 1.961 */ { 144, 512, 0xfdb52dda76fbf667, 0x00001442d2f22480 }, /* 2.046 */ { 145, 512, 0xa9160110f66e24ff, 0x0000144b899f9dbb }, /* 1.968 */ { 146, 512, 0x77306a30379ae03b, 0x000014cb98eb1f81 }, /* 2.143 */ { 147, 512, 0x14f5985d2752319d, 0x000014feab821fc9 }, /* 2.064 */ { 148, 512, 0xa4b8ff11de7863f8, 0x0000154a0e60b9c9 }, /* 2.023 */ { 149, 512, 0x44b345426455c1b3, 0x000015999c3c569c }, /* 2.136 */ { 150, 512, 0x272677826049b46c, 0x000015c9697f4b92 }, /* 2.063 */ { 151, 512, 0x2f9216e2cd74fe40, 0x0000162b1f7bbd39 }, /* 1.974 */ { 152, 512, 0x706ae3e763ad8771, 0x00001661371c55e1 }, /* 2.210 */ { 153, 512, 0xf7fd345307c2480e, 0x000016e251f28b6a }, /* 2.006 */ { 154, 512, 0x6e94e3d26b3139eb, 0x000016f2429bb8c6 }, /* 2.193 */ { 155, 512, 0x5458bbfbb781fcba, 0x0000173efdeca1b9 }, /* 2.163 */ { 156, 512, 0xa80e2afeccd93b33, 0x000017bfdcb78adc }, /* 2.046 */ { 157, 512, 0x1e4ccbb22796cf9d, 0x00001826fdcc39c9 }, /* 2.084 */ { 158, 512, 0x8fba4b676aaa3663, 0x00001841a1379480 }, /* 2.264 */ { 159, 512, 0xf82b843814b315fa, 0x000018886e19b8a3 }, /* 2.074 */ { 160, 512, 0x7f21e920ecf753a3, 0x0000191812ca0ea7 }, /* 2.282 */ { 161, 512, 0x48bb8ea2c4caa620, 0x0000192f310faccf }, /* 2.148 */ { 162, 512, 0x5cdb652b4952c91b, 0x0000199e1d7437c7 }, /* 2.355 */ { 163, 512, 0x6ac1ba6f78c06cd4, 0x000019cd11f82c70 }, /* 2.164 */ { 164, 512, 0x9faf5f9ca2669a56, 0x00001a18d5431f6a }, /* 2.393 */ { 165, 512, 0xaa57e9383eb01194, 0x00001a9e7d253d85 }, /* 2.178 */ { 166, 512, 0x896967bf495c34d2, 0x00001afb8319b9fc }, /* 2.334 */ { 167, 512, 0xdfad5f05de225f1b, 0x00001b3a59c3093b }, /* 2.266 */ { 168, 512, 0xfd299a99f9f2abdd, 0x00001bb6f1a10799 }, /* 2.304 */ { 169, 512, 0xdda239e798fe9fd4, 0x00001bfae0c9692d }, /* 2.218 */ { 170, 512, 0x5fca670414a32c3e, 0x00001c22129dbcff }, /* 2.377 */ { 171, 512, 0x1bb8934314b087de, 0x00001c955db36cd0 }, /* 2.155 */ { 172, 512, 0xd96394b4b082200d, 0x00001cfc8619b7e6 }, /* 2.404 */ { 173, 512, 0xb612a7735b1c8cbc, 0x00001d303acdd585 }, /* 2.205 */ { 174, 512, 0x28e7430fe5875fe1, 0x00001d7ed5b3697d }, /* 2.359 */ { 175, 512, 0x5038e89efdd981b9, 0x00001dc40ec35c59 }, /* 2.158 */ { 176, 512, 0x075fd78f1d14db7c, 0x00001e31c83b4a2b }, /* 2.614 */ { 177, 512, 0xc50fafdb5021be15, 0x00001e7cdac82fbc }, /* 2.239 */ { 178, 512, 0xe6dc7572ce7b91c7, 0x00001edd8bb454fc }, /* 2.493 */ { 179, 512, 0x21f7843e7beda537, 0x00001f3a8e019d6c }, /* 2.327 */ { 180, 512, 0xc83385e20b43ec82, 0x00001f70735ec137 }, /* 2.231 */ { 181, 512, 0xca818217dddb21fd, 0x0000201ca44c5a3c }, /* 2.237 */ { 182, 512, 0xe6035defea48f933, 0x00002038e3346658 }, /* 2.691 */ { 183, 512, 0x47262a4f953dac5a, 0x000020c2e554314e }, /* 2.170 */ { 184, 512, 0xe24c7246260873ea, 0x000021197e618d64 }, /* 2.600 */ { 185, 512, 0xeef6b57c9b58e9e1, 0x0000217ea48ecddc }, /* 2.391 */ { 186, 512, 0x2becd3346e386142, 0x000021c496d4a5f9 }, /* 2.677 */ { 187, 512, 0x63c6207bdf3b40a3, 0x0000220e0f2eec0c }, /* 2.410 */ { 188, 512, 0x3056ce8989767d4b, 0x0000228eb76cd137 }, /* 2.776 */ { 189, 512, 0x91af61c307cee780, 0x000022e17e2ea501 }, /* 2.266 */ { 190, 512, 0xda359da225f6d54f, 0x00002358a2debc19 }, /* 2.717 */ { 191, 512, 0x0a5f7a2a55607ba0, 0x0000238a79dac18c }, /* 2.474 */ { 192, 512, 0x27bb75bf5224638a, 0x00002403a58e2351 }, /* 2.673 */ { 193, 512, 0x1ebfdb94630f5d0f, 0x00002492a10cb339 }, /* 2.420 */ { 194, 512, 0x6eae5e51d9c5f6fb, 0x000024ce4bf98715 }, /* 2.898 */ { 195, 512, 0x08d903b4daedc2e0, 0x0000250d1e15886c }, /* 2.363 */ { 196, 512, 0xc722a2f7fa7cd686, 0x0000258a99ed0c9e }, /* 2.747 */ { 197, 512, 0x8f71faf0e54e361d, 0x000025dee11976f5 }, /* 2.531 */ { 198, 512, 0x87f64695c91a54e7, 0x0000264e00a43da0 }, /* 2.707 */ { 199, 512, 0xc719cbac2c336b92, 0x000026d327277ac1 }, /* 2.315 */ { 200, 512, 0xe7e647afaf771ade, 0x000027523a5c44bf }, /* 3.012 */ { 201, 512, 0x12d4b5c38ce8c946, 0x0000273898432545 }, /* 2.378 */ { 202, 512, 0xf2e0cd4067bdc94a, 0x000027e47bb2c935 }, /* 2.969 */ { 203, 512, 0x21b79f14d6d947d3, 0x0000281e64977f0d }, /* 2.594 */ { 204, 512, 0x515093f952f18cd6, 0x0000289691a473fd }, /* 2.763 */ { 205, 512, 0xd47b160a1b1022c8, 0x00002903e8b52411 }, /* 2.457 */ { 206, 512, 0xc02fc96684715a16, 0x0000297515608601 }, /* 3.057 */ { 207, 512, 0xef51e68efba72ed0, 0x000029ef73604804 }, /* 2.590 */ { 208, 512, 0x9e3be6e5448b4f33, 0x00002a2846ed074b }, /* 3.047 */ { 209, 512, 0x81d446c6d5fec063, 0x00002a92ca693455 }, /* 2.676 */ { 210, 512, 0xff215de8224e57d5, 0x00002b2271fe3729 }, /* 2.993 */ { 211, 512, 0xe2524d9ba8f69796, 0x00002b64b99c3ba2 }, /* 2.457 */ { 212, 512, 0xf6b28e26097b7e4b, 0x00002bd768b6e068 }, /* 3.182 */ { 213, 512, 0x893a487f30ce1644, 0x00002c67f722b4b2 }, /* 2.563 */ { 214, 512, 0x386566c3fc9871df, 0x00002cc1cf8b4037 }, /* 3.025 */ { 215, 512, 0x1e0ed78edf1f558a, 0x00002d3948d36c7f }, /* 2.730 */ { 216, 512, 0xe3bc20c31e61f113, 0x00002d6d6b12e025 }, /* 3.036 */ { 217, 512, 0xd6c3ad2e23021882, 0x00002deff7572241 }, /* 2.722 */ { 218, 512, 0xb4a9f95cf0f69c5a, 0x00002e67d537aa36 }, /* 3.356 */ { 219, 512, 0x6e98ed6f6c38e82f, 0x00002e9720626789 }, /* 2.697 */ { 220, 512, 0x2e01edba33fddac7, 0x00002f407c6b0198 }, /* 2.979 */ { 221, 512, 0x559d02e1f5f57ccc, 0x00002fb6a5ab4f24 }, /* 2.858 */ { 222, 512, 0xac18f5a916adcd8e, 0x0000304ae1c5c57e }, /* 3.258 */ { 223, 512, 0x15789fbaddb86f4b, 0x0000306f6e019c78 }, /* 2.693 */ { 224, 512, 0xf4a9c36d5bc4c408, 0x000030da40434213 }, /* 3.259 */ { 225, 512, 0xf640f90fd2727f44, 0x00003189ed37b90c }, /* 2.733 */ { 226, 512, 0xb5313d390d61884a, 0x000031e152616b37 }, /* 3.235 */ { 227, 512, 0x4bae6b3ce9160939, 0x0000321f40aeac42 }, /* 2.983 */ { 228, 512, 0x838c34480f1a66a1, 0x000032f389c0f78e }, /* 3.308 */ { 229, 512, 0xb1c4a52c8e3d6060, 0x0000330062a40284 }, /* 2.715 */ { 230, 512, 0xe0f1110c6d0ed822, 0x0000338be435644f }, /* 3.540 */ { 231, 512, 0x9f1a8ccdcea68d4b, 0x000034045a4e97e1 }, /* 2.779 */ { 232, 512, 0x3261ed62223f3099, 0x000034702cfc401c }, /* 3.084 */ { 233, 512, 0xf2191e2311022d65, 0x00003509dd19c9fc }, /* 2.987 */ { 234, 512, 0xf102a395c2033abc, 0x000035654dc96fae }, /* 3.341 */ { 235, 512, 0x11fe378f027906b6, 0x000035b5193b0264 }, /* 2.793 */ { 236, 512, 0xf777f2c026b337aa, 0x000036704f5d9297 }, /* 3.518 */ { 237, 512, 0x1b04e9c2ee143f32, 0x000036dfbb7af218 }, /* 2.962 */ { 238, 512, 0x2fcec95266f9352c, 0x00003785c8df24a9 }, /* 3.196 */ { 239, 512, 0xfe2b0e47e427dd85, 0x000037cbdf5da729 }, /* 2.914 */ { 240, 512, 0x72b49bf2225f6c6d, 0x0000382227c15855 }, /* 3.408 */ { 241, 512, 0x50486b43df7df9c7, 0x0000389b88be6453 }, /* 2.903 */ { 242, 512, 0x5192a3e53181c8ab, 0x000038ddf3d67263 }, /* 3.778 */ { 243, 512, 0xe9f5d8365296fd5e, 0x0000399f1c6c9e9c }, /* 3.026 */ { 244, 512, 0xc740263f0301efa8, 0x00003a147146512d }, /* 3.347 */ { 245, 512, 0x23cd0f2b5671e67d, 0x00003ab10bcc0d9d }, /* 3.212 */ { 246, 512, 0x002ccc7e5cd41390, 0x00003ad6cd14a6c0 }, /* 3.482 */ { 247, 512, 0x9aafb3c02544b31b, 0x00003b8cb8779fb0 }, /* 3.146 */ { 248, 512, 0x72ba07a78b121999, 0x00003c24142a5a3f }, /* 3.626 */ { 249, 512, 0x3d784aa58edfc7b4, 0x00003cd084817d99 }, /* 2.952 */ { 250, 512, 0xaab750424d8004af, 0x00003d506a8e098e }, /* 3.463 */ { 251, 512, 0x84403fcf8e6b5ca2, 0x00003d4c54c2aec4 }, /* 3.131 */ { 252, 512, 0x71eb7455ec98e207, 0x00003e655715cf2c }, /* 3.538 */ { 253, 512, 0xd752b4f19301595b, 0x00003ecd7b2ca5ac }, /* 2.974 */ { 254, 512, 0xc4674129750499de, 0x00003e99e86d3e95 }, /* 3.843 */ { 255, 512, 0x9772baff5cd12ef5, 0x00003f895c019841 }, /* 3.088 */ }; /* * Verify the map is valid. Each device index must appear exactly * once in every row, and the permutation array checksum must match. */ static int verify_perms(uint8_t *perms, uint64_t children, uint64_t nperms, uint64_t checksum) { int countssz = sizeof (uint16_t) * children; uint16_t *counts = kmem_zalloc(countssz, KM_SLEEP); for (int i = 0; i < nperms; i++) { for (int j = 0; j < children; j++) { uint8_t val = perms[(i * children) + j]; if (val >= children || counts[val] != i) { kmem_free(counts, countssz); return (EINVAL); } counts[val]++; } } if (checksum != 0) { int permssz = sizeof (uint8_t) * children * nperms; zio_cksum_t cksum; fletcher_4_native_varsize(perms, permssz, &cksum); if (checksum != cksum.zc_word[0]) { kmem_free(counts, countssz); return (ECKSUM); } } kmem_free(counts, countssz); return (0); } /* * Generate the permutation array for the draid_map_t. These maps control * the placement of all data in a dRAID. Therefore it's critical that the * seed always generates the same mapping. We provide our own pseudo-random * number generator for this purpose. */ int vdev_draid_generate_perms(const draid_map_t *map, uint8_t **permsp) { VERIFY3U(map->dm_children, >=, VDEV_DRAID_MIN_CHILDREN); VERIFY3U(map->dm_children, <=, VDEV_DRAID_MAX_CHILDREN); VERIFY3U(map->dm_seed, !=, 0); VERIFY3U(map->dm_nperms, !=, 0); VERIFY3P(map->dm_perms, ==, NULL); #ifdef _KERNEL /* * The kernel code always provides both a map_seed and checksum. * Only the tests/zfs-tests/cmd/draid/draid.c utility will provide * a zero checksum when generating new candidate maps. */ VERIFY3U(map->dm_checksum, !=, 0); #endif uint64_t children = map->dm_children; uint64_t nperms = map->dm_nperms; int rowsz = sizeof (uint8_t) * children; int permssz = rowsz * nperms; uint8_t *perms; /* Allocate the permutation array */ perms = vmem_alloc(permssz, KM_SLEEP); /* Setup an initial row with a known pattern */ uint8_t *initial_row = kmem_alloc(rowsz, KM_SLEEP); for (int i = 0; i < children; i++) initial_row[i] = i; uint64_t draid_seed[2] = { VDEV_DRAID_SEED, map->dm_seed }; uint8_t *current_row, *previous_row = initial_row; /* * Perform a Fisher-Yates shuffle of each row using the previous * row as the starting point. An initial_row with known pattern * is used as the input for the first row. */ for (int i = 0; i < nperms; i++) { current_row = &perms[i * children]; memcpy(current_row, previous_row, rowsz); for (int j = children - 1; j > 0; j--) { uint64_t k = vdev_draid_rand(draid_seed) % (j + 1); uint8_t val = current_row[j]; current_row[j] = current_row[k]; current_row[k] = val; } previous_row = current_row; } kmem_free(initial_row, rowsz); int error = verify_perms(perms, children, nperms, map->dm_checksum); if (error) { vmem_free(perms, permssz); return (error); } *permsp = perms; return (0); } /* * Lookup the fixed draid_map_t for the requested number of children. */ int vdev_draid_lookup_map(uint64_t children, const draid_map_t **mapp) { for (int i = 0; i <= VDEV_DRAID_MAX_MAPS; i++) { if (draid_maps[i].dm_children == children) { *mapp = &draid_maps[i]; return (0); } } return (ENOENT); } /* * Lookup the permutation array and iteration id for the provided offset. */ static void vdev_draid_get_perm(vdev_draid_config_t *vdc, uint64_t pindex, uint8_t **base, uint64_t *iter) { uint64_t ncols = vdc->vdc_children; uint64_t poff = pindex % (vdc->vdc_nperms * ncols); *base = vdc->vdc_perms + (poff / ncols) * ncols; *iter = poff % ncols; } static inline uint64_t vdev_draid_permute_id(vdev_draid_config_t *vdc, uint8_t *base, uint64_t iter, uint64_t index) { return ((base[index] + iter) % vdc->vdc_children); } /* * Return the asize which is the psize rounded up to a full group width. * i.e. vdev_draid_psize_to_asize(). */ static uint64_t vdev_draid_asize(vdev_t *vd, uint64_t psize) { vdev_draid_config_t *vdc = vd->vdev_tsd; uint64_t ashift = vd->vdev_ashift; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); uint64_t rows = ((psize - 1) / (vdc->vdc_ndata << ashift)) + 1; uint64_t asize = (rows * vdc->vdc_groupwidth) << ashift; ASSERT3U(asize, !=, 0); ASSERT3U(asize % (vdc->vdc_groupwidth), ==, 0); return (asize); } /* * Deflate the asize to the psize, this includes stripping parity. */ uint64_t vdev_draid_asize_to_psize(vdev_t *vd, uint64_t asize) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT0(asize % vdc->vdc_groupwidth); return ((asize / vdc->vdc_groupwidth) * vdc->vdc_ndata); } /* * Convert a logical offset to the corresponding group number. */ static uint64_t vdev_draid_offset_to_group(vdev_t *vd, uint64_t offset) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); return (offset / vdc->vdc_groupsz); } /* * Convert a group number to the logical starting offset for that group. */ static uint64_t vdev_draid_group_to_offset(vdev_t *vd, uint64_t group) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); return (group * vdc->vdc_groupsz); } /* * Full stripe writes. When writing, all columns (D+P) are required. Parity * is calculated over all the columns, including empty zero filled sectors, * and each is written to disk. While only the data columns are needed for * a normal read, all of the columns are required for reconstruction when * performing a sequential resilver. * * For "big columns" it's sufficient to map the correct range of the zio ABD. * Partial columns require allocating a gang ABD in order to zero fill the * empty sectors. When the column is empty a zero filled sector must be * mapped. In all cases the data ABDs must be the same size as the parity * ABDs (e.g. rc->rc_size == parity_size). */ static void vdev_draid_map_alloc_write(zio_t *zio, uint64_t abd_offset, raidz_row_t *rr) { uint64_t skip_size = 1ULL << zio->io_vd->vdev_top->vdev_ashift; uint64_t parity_size = rr->rr_col[0].rc_size; uint64_t abd_off = abd_offset; ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE); ASSERT3U(parity_size, ==, abd_get_size(rr->rr_col[0].rc_abd)); for (uint64_t c = rr->rr_firstdatacol; c < rr->rr_cols; c++) { raidz_col_t *rc = &rr->rr_col[c]; if (rc->rc_size == 0) { /* empty data column (small write), add a skip sector */ ASSERT3U(skip_size, ==, parity_size); rc->rc_abd = abd_get_zeros(skip_size); } else if (rc->rc_size == parity_size) { /* this is a "big column" */ rc->rc_abd = abd_get_offset_struct(&rc->rc_abdstruct, zio->io_abd, abd_off, rc->rc_size); } else { /* short data column, add a skip sector */ ASSERT3U(rc->rc_size + skip_size, ==, parity_size); rc->rc_abd = abd_alloc_gang(); abd_gang_add(rc->rc_abd, abd_get_offset_size( zio->io_abd, abd_off, rc->rc_size), B_TRUE); abd_gang_add(rc->rc_abd, abd_get_zeros(skip_size), B_TRUE); } ASSERT3U(abd_get_size(rc->rc_abd), ==, parity_size); abd_off += rc->rc_size; rc->rc_size = parity_size; } IMPLY(abd_offset != 0, abd_off == zio->io_size); } /* * Scrub/resilver reads. In order to store the contents of the skip sectors * an additional ABD is allocated. The columns are handled in the same way * as a full stripe write except instead of using the zero ABD the newly * allocated skip ABD is used to back the skip sectors. In all cases the * data ABD must be the same size as the parity ABDs. */ static void vdev_draid_map_alloc_scrub(zio_t *zio, uint64_t abd_offset, raidz_row_t *rr) { uint64_t skip_size = 1ULL << zio->io_vd->vdev_top->vdev_ashift; uint64_t parity_size = rr->rr_col[0].rc_size; uint64_t abd_off = abd_offset; uint64_t skip_off = 0; ASSERT3U(zio->io_type, ==, ZIO_TYPE_READ); ASSERT3P(rr->rr_abd_empty, ==, NULL); if (rr->rr_nempty > 0) { rr->rr_abd_empty = abd_alloc_linear(rr->rr_nempty * skip_size, B_FALSE); } for (uint64_t c = rr->rr_firstdatacol; c < rr->rr_cols; c++) { raidz_col_t *rc = &rr->rr_col[c]; if (rc->rc_size == 0) { /* empty data column (small read), add a skip sector */ ASSERT3U(skip_size, ==, parity_size); ASSERT3U(rr->rr_nempty, !=, 0); rc->rc_abd = abd_get_offset_size(rr->rr_abd_empty, skip_off, skip_size); skip_off += skip_size; } else if (rc->rc_size == parity_size) { /* this is a "big column" */ rc->rc_abd = abd_get_offset_struct(&rc->rc_abdstruct, zio->io_abd, abd_off, rc->rc_size); } else { /* short data column, add a skip sector */ ASSERT3U(rc->rc_size + skip_size, ==, parity_size); ASSERT3U(rr->rr_nempty, !=, 0); rc->rc_abd = abd_alloc_gang(); abd_gang_add(rc->rc_abd, abd_get_offset_size( zio->io_abd, abd_off, rc->rc_size), B_TRUE); abd_gang_add(rc->rc_abd, abd_get_offset_size( rr->rr_abd_empty, skip_off, skip_size), B_TRUE); skip_off += skip_size; } uint64_t abd_size = abd_get_size(rc->rc_abd); ASSERT3U(abd_size, ==, abd_get_size(rr->rr_col[0].rc_abd)); /* * Increase rc_size so the skip ABD is included in subsequent * parity calculations. */ abd_off += rc->rc_size; rc->rc_size = abd_size; } IMPLY(abd_offset != 0, abd_off == zio->io_size); ASSERT3U(skip_off, ==, rr->rr_nempty * skip_size); } /* * Normal reads. In this common case only the columns containing data * are read in to the zio ABDs. Neither the parity columns or empty skip * sectors are read unless the checksum fails verification. In which case * vdev_raidz_read_all() will call vdev_draid_map_alloc_empty() to expand * the raid map in order to allow reconstruction using the parity data and * skip sectors. */ static void vdev_draid_map_alloc_read(zio_t *zio, uint64_t abd_offset, raidz_row_t *rr) { uint64_t abd_off = abd_offset; ASSERT3U(zio->io_type, ==, ZIO_TYPE_READ); for (uint64_t c = rr->rr_firstdatacol; c < rr->rr_cols; c++) { raidz_col_t *rc = &rr->rr_col[c]; if (rc->rc_size > 0) { rc->rc_abd = abd_get_offset_struct(&rc->rc_abdstruct, zio->io_abd, abd_off, rc->rc_size); abd_off += rc->rc_size; } } IMPLY(abd_offset != 0, abd_off == zio->io_size); } /* * Converts a normal "read" raidz_row_t to a "scrub" raidz_row_t. The key * difference is that an ABD is allocated to back skip sectors so they may * be read in to memory, verified, and repaired if needed. */ void vdev_draid_map_alloc_empty(zio_t *zio, raidz_row_t *rr) { uint64_t skip_size = 1ULL << zio->io_vd->vdev_top->vdev_ashift; uint64_t parity_size = rr->rr_col[0].rc_size; uint64_t skip_off = 0; ASSERT3U(zio->io_type, ==, ZIO_TYPE_READ); ASSERT3P(rr->rr_abd_empty, ==, NULL); if (rr->rr_nempty > 0) { rr->rr_abd_empty = abd_alloc_linear(rr->rr_nempty * skip_size, B_FALSE); } for (uint64_t c = rr->rr_firstdatacol; c < rr->rr_cols; c++) { raidz_col_t *rc = &rr->rr_col[c]; if (rc->rc_size == 0) { /* empty data column (small read), add a skip sector */ ASSERT3U(skip_size, ==, parity_size); ASSERT3U(rr->rr_nempty, !=, 0); ASSERT3P(rc->rc_abd, ==, NULL); rc->rc_abd = abd_get_offset_size(rr->rr_abd_empty, skip_off, skip_size); skip_off += skip_size; } else if (rc->rc_size == parity_size) { /* this is a "big column", nothing to add */ ASSERT3P(rc->rc_abd, !=, NULL); } else { /* * short data column, add a skip sector and clear * rc_tried to force the entire column to be re-read * thereby including the missing skip sector data * which is needed for reconstruction. */ ASSERT3U(rc->rc_size + skip_size, ==, parity_size); ASSERT3U(rr->rr_nempty, !=, 0); ASSERT3P(rc->rc_abd, !=, NULL); ASSERT(!abd_is_gang(rc->rc_abd)); abd_t *read_abd = rc->rc_abd; rc->rc_abd = abd_alloc_gang(); abd_gang_add(rc->rc_abd, read_abd, B_TRUE); abd_gang_add(rc->rc_abd, abd_get_offset_size( rr->rr_abd_empty, skip_off, skip_size), B_TRUE); skip_off += skip_size; rc->rc_tried = 0; } /* * Increase rc_size so the empty ABD is included in subsequent * parity calculations. */ rc->rc_size = parity_size; } ASSERT3U(skip_off, ==, rr->rr_nempty * skip_size); } /* * Given a logical address within a dRAID configuration, return the physical * address on the first drive in the group that this address maps to * (at position 'start' in permutation number 'perm'). */ static uint64_t vdev_draid_logical_to_physical(vdev_t *vd, uint64_t logical_offset, uint64_t *perm, uint64_t *start) { vdev_draid_config_t *vdc = vd->vdev_tsd; /* b is the dRAID (parent) sector offset. */ uint64_t ashift = vd->vdev_top->vdev_ashift; uint64_t b_offset = logical_offset >> ashift; /* * The height of a row in units of the vdev's minimum sector size. * This is the amount of data written to each disk of each group * in a given permutation. */ uint64_t rowheight_sectors = VDEV_DRAID_ROWHEIGHT >> ashift; /* * We cycle through a disk permutation every groupsz * ngroups chunk * of address space. Note that ngroups * groupsz must be a multiple * of the number of data drives (ndisks) in order to guarantee * alignment. So, for example, if our row height is 16MB, our group * size is 10, and there are 13 data drives in the draid, then ngroups * will be 13, we will change permutation every 2.08GB and each * disk will have 160MB of data per chunk. */ uint64_t groupwidth = vdc->vdc_groupwidth; uint64_t ngroups = vdc->vdc_ngroups; uint64_t ndisks = vdc->vdc_ndisks; /* * groupstart is where the group this IO will land in "starts" in * the permutation array. */ uint64_t group = logical_offset / vdc->vdc_groupsz; uint64_t groupstart = (group * groupwidth) % ndisks; ASSERT3U(groupstart + groupwidth, <=, ndisks + groupstart); *start = groupstart; /* b_offset is the sector offset within a group chunk */ b_offset = b_offset % (rowheight_sectors * groupwidth); ASSERT0(b_offset % groupwidth); /* * Find the starting byte offset on each child vdev: * - within a permutation there are ngroups groups spread over the * rows, where each row covers a slice portion of the disk * - each permutation has (groupwidth * ngroups) / ndisks rows * - so each permutation covers rows * slice portion of the disk * - so we need to find the row where this IO group target begins */ *perm = group / ngroups; uint64_t row = (*perm * ((groupwidth * ngroups) / ndisks)) + (((group % ngroups) * groupwidth) / ndisks); return (((rowheight_sectors * row) + (b_offset / groupwidth)) << ashift); } static uint64_t vdev_draid_map_alloc_row(zio_t *zio, raidz_row_t **rrp, uint64_t io_offset, uint64_t abd_offset, uint64_t abd_size) { vdev_t *vd = zio->io_vd; vdev_draid_config_t *vdc = vd->vdev_tsd; uint64_t ashift = vd->vdev_top->vdev_ashift; uint64_t io_size = abd_size; uint64_t io_asize = vdev_draid_asize(vd, io_size); uint64_t group = vdev_draid_offset_to_group(vd, io_offset); uint64_t start_offset = vdev_draid_group_to_offset(vd, group + 1); /* * Limit the io_size to the space remaining in the group. A second * row in the raidz_map_t is created for the remainder. */ if (io_offset + io_asize > start_offset) { io_size = vdev_draid_asize_to_psize(vd, start_offset - io_offset); } /* * At most a block may span the logical end of one group and the start * of the next group. Therefore, at the end of a group the io_size must * span the group width evenly and the remainder must be aligned to the * start of the next group. */ IMPLY(abd_offset == 0 && io_size < zio->io_size, (io_asize >> ashift) % vdc->vdc_groupwidth == 0); IMPLY(abd_offset != 0, vdev_draid_group_to_offset(vd, group) == io_offset); /* Lookup starting byte offset on each child vdev */ uint64_t groupstart, perm; uint64_t physical_offset = vdev_draid_logical_to_physical(vd, io_offset, &perm, &groupstart); /* * If there is less than groupwidth drives available after the group * start, the group is going to wrap onto the next row. 'wrap' is the * group disk number that starts on the next row. */ uint64_t ndisks = vdc->vdc_ndisks; uint64_t groupwidth = vdc->vdc_groupwidth; uint64_t wrap = groupwidth; if (groupstart + groupwidth > ndisks) wrap = ndisks - groupstart; /* The io size in units of the vdev's minimum sector size. */ const uint64_t psize = io_size >> ashift; /* * "Quotient": The number of data sectors for this stripe on all but * the "big column" child vdevs that also contain "remainder" data. */ uint64_t q = psize / vdc->vdc_ndata; /* * "Remainder": The number of partial stripe data sectors in this I/O. * This will add a sector to some, but not all, child vdevs. */ uint64_t r = psize - q * vdc->vdc_ndata; /* The number of "big columns" - those which contain remainder data. */ uint64_t bc = (r == 0 ? 0 : r + vdc->vdc_nparity); ASSERT3U(bc, <, groupwidth); /* The total number of data and parity sectors for this I/O. */ uint64_t tot = psize + (vdc->vdc_nparity * (q + (r == 0 ? 0 : 1))); raidz_row_t *rr; rr = kmem_alloc(offsetof(raidz_row_t, rr_col[groupwidth]), KM_SLEEP); rr->rr_cols = groupwidth; rr->rr_scols = groupwidth; rr->rr_bigcols = bc; rr->rr_missingdata = 0; rr->rr_missingparity = 0; rr->rr_firstdatacol = vdc->vdc_nparity; rr->rr_abd_empty = NULL; #ifdef ZFS_DEBUG rr->rr_offset = io_offset; rr->rr_size = io_size; #endif *rrp = rr; uint8_t *base; uint64_t iter, asize = 0; vdev_draid_get_perm(vdc, perm, &base, &iter); for (uint64_t i = 0; i < groupwidth; i++) { raidz_col_t *rc = &rr->rr_col[i]; uint64_t c = (groupstart + i) % ndisks; /* increment the offset if we wrap to the next row */ if (i == wrap) physical_offset += VDEV_DRAID_ROWHEIGHT; rc->rc_devidx = vdev_draid_permute_id(vdc, base, iter, c); rc->rc_offset = physical_offset; rc->rc_abd = NULL; rc->rc_orig_data = NULL; rc->rc_error = 0; rc->rc_tried = 0; rc->rc_skipped = 0; rc->rc_force_repair = 0; rc->rc_allow_repair = 1; rc->rc_need_orig_restore = B_FALSE; if (q == 0 && i >= bc) rc->rc_size = 0; else if (i < bc) rc->rc_size = (q + 1) << ashift; else rc->rc_size = q << ashift; asize += rc->rc_size; } ASSERT3U(asize, ==, tot << ashift); rr->rr_nempty = roundup(tot, groupwidth) - tot; IMPLY(bc > 0, rr->rr_nempty == groupwidth - bc); /* Allocate buffers for the parity columns */ for (uint64_t c = 0; c < rr->rr_firstdatacol; c++) { raidz_col_t *rc = &rr->rr_col[c]; rc->rc_abd = abd_alloc_linear(rc->rc_size, B_FALSE); } /* * Map buffers for data columns and allocate/map buffers for skip * sectors. There are three distinct cases for dRAID which are * required to support sequential rebuild. */ if (zio->io_type == ZIO_TYPE_WRITE) { vdev_draid_map_alloc_write(zio, abd_offset, rr); } else if ((rr->rr_nempty > 0) && (zio->io_flags & (ZIO_FLAG_SCRUB | ZIO_FLAG_RESILVER))) { vdev_draid_map_alloc_scrub(zio, abd_offset, rr); } else { ASSERT3U(zio->io_type, ==, ZIO_TYPE_READ); vdev_draid_map_alloc_read(zio, abd_offset, rr); } return (io_size); } /* * Allocate the raidz mapping to be applied to the dRAID I/O. The parity * calculations for dRAID are identical to raidz however there are a few * differences in the layout. * * - dRAID always allocates a full stripe width. Any extra sectors due * this padding are zero filled and written to disk. They will be read * back during a scrub or repair operation since they are included in * the parity calculation. This property enables sequential resilvering. * * - When the block at the logical offset spans redundancy groups then two * rows are allocated in the raidz_map_t. One row resides at the end of * the first group and the other at the start of the following group. */ static raidz_map_t * vdev_draid_map_alloc(zio_t *zio) { raidz_row_t *rr[2]; uint64_t abd_offset = 0; uint64_t abd_size = zio->io_size; uint64_t io_offset = zio->io_offset; uint64_t size; int nrows = 1; size = vdev_draid_map_alloc_row(zio, &rr[0], io_offset, abd_offset, abd_size); if (size < abd_size) { vdev_t *vd = zio->io_vd; io_offset += vdev_draid_asize(vd, size); abd_offset += size; abd_size -= size; nrows++; ASSERT3U(io_offset, ==, vdev_draid_group_to_offset( vd, vdev_draid_offset_to_group(vd, io_offset))); ASSERT3U(abd_offset, <, zio->io_size); ASSERT3U(abd_size, !=, 0); size = vdev_draid_map_alloc_row(zio, &rr[1], io_offset, abd_offset, abd_size); VERIFY3U(size, ==, abd_size); } raidz_map_t *rm; rm = kmem_zalloc(offsetof(raidz_map_t, rm_row[nrows]), KM_SLEEP); rm->rm_ops = vdev_raidz_math_get_ops(); rm->rm_nrows = nrows; rm->rm_row[0] = rr[0]; if (nrows == 2) rm->rm_row[1] = rr[1]; return (rm); } /* * Given an offset into a dRAID return the next group width aligned offset * which can be used to start an allocation. */ static uint64_t vdev_draid_get_astart(vdev_t *vd, const uint64_t start) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); return (roundup(start, vdc->vdc_groupwidth << vd->vdev_ashift)); } /* * Allocatable space for dRAID is (children - nspares) * sizeof(smallest child) * rounded down to the last full slice. So each child must provide at least * 1 / (children - nspares) of its asize. */ static uint64_t vdev_draid_min_asize(vdev_t *vd) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); - return ((vd->vdev_min_asize + vdc->vdc_ndisks - 1) / (vdc->vdc_ndisks)); + return (VDEV_DRAID_REFLOW_RESERVE + + (vd->vdev_min_asize + vdc->vdc_ndisks - 1) / (vdc->vdc_ndisks)); } /* * When using dRAID the minimum allocation size is determined by the number * of data disks in the redundancy group. Full stripes are always used. */ static uint64_t vdev_draid_min_alloc(vdev_t *vd) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); return (vdc->vdc_ndata << vd->vdev_ashift); } /* * Returns true if the txg range does not exist on any leaf vdev. * * A dRAID spare does not fit into the DTL model. While it has child vdevs * there is no redundancy among them, and the effective child vdev is * determined by offset. Essentially we do a vdev_dtl_reassess() on the * fly by replacing a dRAID spare with the child vdev under the offset. * Note that it is a recursive process because the child vdev can be * another dRAID spare and so on. */ boolean_t vdev_draid_missing(vdev_t *vd, uint64_t physical_offset, uint64_t txg, uint64_t size) { if (vd->vdev_ops == &vdev_spare_ops || vd->vdev_ops == &vdev_replacing_ops) { /* * Check all of the readable children, if any child * contains the txg range the data it is not missing. */ for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; if (!vdev_readable(cvd)) continue; if (!vdev_draid_missing(cvd, physical_offset, txg, size)) return (B_FALSE); } return (B_TRUE); } if (vd->vdev_ops == &vdev_draid_spare_ops) { /* * When sequentially resilvering we don't have a proper * txg range so instead we must presume all txgs are * missing on this vdev until the resilver completes. */ if (vd->vdev_rebuild_txg != 0) return (B_TRUE); /* * DTL_MISSING is set for all prior txgs when a resilver * is started in spa_vdev_attach(). */ if (vdev_dtl_contains(vd, DTL_MISSING, txg, size)) return (B_TRUE); /* * Consult the DTL on the relevant vdev. Either a vdev * leaf or spare/replace mirror child may be returned so * we must recursively call vdev_draid_missing_impl(). */ vd = vdev_draid_spare_get_child(vd, physical_offset); if (vd == NULL) return (B_TRUE); return (vdev_draid_missing(vd, physical_offset, txg, size)); } return (vdev_dtl_contains(vd, DTL_MISSING, txg, size)); } /* * Returns true if the txg is only partially replicated on the leaf vdevs. */ static boolean_t vdev_draid_partial(vdev_t *vd, uint64_t physical_offset, uint64_t txg, uint64_t size) { if (vd->vdev_ops == &vdev_spare_ops || vd->vdev_ops == &vdev_replacing_ops) { /* * Check all of the readable children, if any child is * missing the txg range then it is partially replicated. */ for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; if (!vdev_readable(cvd)) continue; if (vdev_draid_partial(cvd, physical_offset, txg, size)) return (B_TRUE); } return (B_FALSE); } if (vd->vdev_ops == &vdev_draid_spare_ops) { /* * When sequentially resilvering we don't have a proper * txg range so instead we must presume all txgs are * missing on this vdev until the resilver completes. */ if (vd->vdev_rebuild_txg != 0) return (B_TRUE); /* * DTL_MISSING is set for all prior txgs when a resilver * is started in spa_vdev_attach(). */ if (vdev_dtl_contains(vd, DTL_MISSING, txg, size)) return (B_TRUE); /* * Consult the DTL on the relevant vdev. Either a vdev * leaf or spare/replace mirror child may be returned so * we must recursively call vdev_draid_missing_impl(). */ vd = vdev_draid_spare_get_child(vd, physical_offset); if (vd == NULL) return (B_TRUE); return (vdev_draid_partial(vd, physical_offset, txg, size)); } return (vdev_dtl_contains(vd, DTL_MISSING, txg, size)); } /* * Determine if the vdev is readable at the given offset. */ boolean_t vdev_draid_readable(vdev_t *vd, uint64_t physical_offset) { if (vd->vdev_ops == &vdev_draid_spare_ops) { vd = vdev_draid_spare_get_child(vd, physical_offset); if (vd == NULL) return (B_FALSE); } if (vd->vdev_ops == &vdev_spare_ops || vd->vdev_ops == &vdev_replacing_ops) { for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; if (!vdev_readable(cvd)) continue; if (vdev_draid_readable(cvd, physical_offset)) return (B_TRUE); } return (B_FALSE); } return (vdev_readable(vd)); } /* * Returns the first distributed spare found under the provided vdev tree. */ static vdev_t * vdev_draid_find_spare(vdev_t *vd) { if (vd->vdev_ops == &vdev_draid_spare_ops) return (vd); for (int c = 0; c < vd->vdev_children; c++) { vdev_t *svd = vdev_draid_find_spare(vd->vdev_child[c]); if (svd != NULL) return (svd); } return (NULL); } /* * Returns B_TRUE if the passed in vdev is currently "faulted". * Faulted, in this context, means that the vdev represents a * replacing or sparing vdev tree. */ static boolean_t vdev_draid_faulted(vdev_t *vd, uint64_t physical_offset) { if (vd->vdev_ops == &vdev_draid_spare_ops) { vd = vdev_draid_spare_get_child(vd, physical_offset); if (vd == NULL) return (B_FALSE); /* * After resolving the distributed spare to a leaf vdev * check the parent to determine if it's "faulted". */ vd = vd->vdev_parent; } return (vd->vdev_ops == &vdev_replacing_ops || vd->vdev_ops == &vdev_spare_ops); } /* * Determine if the dRAID block at the logical offset is degraded. * Used by sequential resilver. */ static boolean_t vdev_draid_group_degraded(vdev_t *vd, uint64_t offset) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); ASSERT3U(vdev_draid_get_astart(vd, offset), ==, offset); uint64_t groupstart, perm; uint64_t physical_offset = vdev_draid_logical_to_physical(vd, offset, &perm, &groupstart); uint8_t *base; uint64_t iter; vdev_draid_get_perm(vdc, perm, &base, &iter); for (uint64_t i = 0; i < vdc->vdc_groupwidth; i++) { uint64_t c = (groupstart + i) % vdc->vdc_ndisks; uint64_t cid = vdev_draid_permute_id(vdc, base, iter, c); vdev_t *cvd = vd->vdev_child[cid]; /* Group contains a faulted vdev. */ if (vdev_draid_faulted(cvd, physical_offset)) return (B_TRUE); /* * Always check groups with active distributed spares * because any vdev failure in the pool will affect them. */ if (vdev_draid_find_spare(cvd) != NULL) return (B_TRUE); } return (B_FALSE); } /* * Determine if the txg is missing. Used by healing resilver. */ static boolean_t vdev_draid_group_missing(vdev_t *vd, uint64_t offset, uint64_t txg, uint64_t size) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); ASSERT3U(vdev_draid_get_astart(vd, offset), ==, offset); uint64_t groupstart, perm; uint64_t physical_offset = vdev_draid_logical_to_physical(vd, offset, &perm, &groupstart); uint8_t *base; uint64_t iter; vdev_draid_get_perm(vdc, perm, &base, &iter); for (uint64_t i = 0; i < vdc->vdc_groupwidth; i++) { uint64_t c = (groupstart + i) % vdc->vdc_ndisks; uint64_t cid = vdev_draid_permute_id(vdc, base, iter, c); vdev_t *cvd = vd->vdev_child[cid]; /* Transaction group is known to be partially replicated. */ if (vdev_draid_partial(cvd, physical_offset, txg, size)) return (B_TRUE); /* * Always check groups with active distributed spares * because any vdev failure in the pool will affect them. */ if (vdev_draid_find_spare(cvd) != NULL) return (B_TRUE); } return (B_FALSE); } /* * Find the smallest child asize and largest sector size to calculate the * available capacity. Distributed spares are ignored since their capacity * is also based of the minimum child size in the top-level dRAID. */ static void vdev_draid_calculate_asize(vdev_t *vd, uint64_t *asizep, uint64_t *max_asizep, uint64_t *logical_ashiftp, uint64_t *physical_ashiftp) { uint64_t logical_ashift = 0, physical_ashift = 0; uint64_t asize = 0, max_asize = 0; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; if (cvd->vdev_ops == &vdev_draid_spare_ops) continue; asize = MIN(asize - 1, cvd->vdev_asize - 1) + 1; max_asize = MIN(max_asize - 1, cvd->vdev_max_asize - 1) + 1; logical_ashift = MAX(logical_ashift, cvd->vdev_ashift); physical_ashift = MAX(physical_ashift, cvd->vdev_physical_ashift); } *asizep = asize; *max_asizep = max_asize; *logical_ashiftp = logical_ashift; *physical_ashiftp = physical_ashift; } /* * Open spare vdevs. */ static boolean_t vdev_draid_open_spares(vdev_t *vd) { return (vd->vdev_ops == &vdev_draid_spare_ops || vd->vdev_ops == &vdev_replacing_ops || vd->vdev_ops == &vdev_spare_ops); } /* * Open all children, excluding spares. */ static boolean_t vdev_draid_open_children(vdev_t *vd) { return (!vdev_draid_open_spares(vd)); } /* * Open a top-level dRAID vdev. */ static int vdev_draid_open(vdev_t *vd, uint64_t *asize, uint64_t *max_asize, uint64_t *logical_ashift, uint64_t *physical_ashift) { vdev_draid_config_t *vdc = vd->vdev_tsd; uint64_t nparity = vdc->vdc_nparity; int open_errors = 0; if (nparity > VDEV_DRAID_MAXPARITY || vd->vdev_children < nparity + 1) { vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; return (SET_ERROR(EINVAL)); } /* * First open the normal children then the distributed spares. This * ordering is important to ensure the distributed spares calculate * the correct psize in the event that the dRAID vdevs were expanded. */ vdev_open_children_subset(vd, vdev_draid_open_children); vdev_open_children_subset(vd, vdev_draid_open_spares); /* Verify enough of the children are available to continue. */ for (int c = 0; c < vd->vdev_children; c++) { if (vd->vdev_child[c]->vdev_open_error != 0) { if ((++open_errors) > nparity) { vd->vdev_stat.vs_aux = VDEV_AUX_NO_REPLICAS; return (SET_ERROR(ENXIO)); } } } /* * Allocatable capacity is the sum of the space on all children less * the number of distributed spares rounded down to last full row * and then to the last full group. An additional 32MB of scratch * space is reserved at the end of each child for use by the dRAID * expansion feature. */ uint64_t child_asize, child_max_asize; vdev_draid_calculate_asize(vd, &child_asize, &child_max_asize, logical_ashift, physical_ashift); /* * Should be unreachable since the minimum child size is 64MB, but * we want to make sure an underflow absolutely cannot occur here. */ if (child_asize < VDEV_DRAID_REFLOW_RESERVE || child_max_asize < VDEV_DRAID_REFLOW_RESERVE) { return (SET_ERROR(ENXIO)); } child_asize = ((child_asize - VDEV_DRAID_REFLOW_RESERVE) / VDEV_DRAID_ROWHEIGHT) * VDEV_DRAID_ROWHEIGHT; child_max_asize = ((child_max_asize - VDEV_DRAID_REFLOW_RESERVE) / VDEV_DRAID_ROWHEIGHT) * VDEV_DRAID_ROWHEIGHT; *asize = (((child_asize * vdc->vdc_ndisks) / vdc->vdc_groupsz) * vdc->vdc_groupsz); *max_asize = (((child_max_asize * vdc->vdc_ndisks) / vdc->vdc_groupsz) * vdc->vdc_groupsz); return (0); } /* * Close a top-level dRAID vdev. */ static void vdev_draid_close(vdev_t *vd) { for (int c = 0; c < vd->vdev_children; c++) { if (vd->vdev_child[c] != NULL) vdev_close(vd->vdev_child[c]); } } /* * Return the maximum asize for a rebuild zio in the provided range * given the following constraints. A dRAID chunks may not: * * - Exceed the maximum allowed block size (SPA_MAXBLOCKSIZE), or * - Span dRAID redundancy groups. */ static uint64_t vdev_draid_rebuild_asize(vdev_t *vd, uint64_t start, uint64_t asize, uint64_t max_segment) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); uint64_t ashift = vd->vdev_ashift; uint64_t ndata = vdc->vdc_ndata; uint64_t psize = MIN(P2ROUNDUP(max_segment * ndata, 1 << ashift), SPA_MAXBLOCKSIZE); ASSERT3U(vdev_draid_get_astart(vd, start), ==, start); ASSERT3U(asize % (vdc->vdc_groupwidth << ashift), ==, 0); /* Chunks must evenly span all data columns in the group. */ psize = (((psize >> ashift) / ndata) * ndata) << ashift; uint64_t chunk_size = MIN(asize, vdev_psize_to_asize(vd, psize)); /* Reduce the chunk size to the group space remaining. */ uint64_t group = vdev_draid_offset_to_group(vd, start); uint64_t left = vdev_draid_group_to_offset(vd, group + 1) - start; chunk_size = MIN(chunk_size, left); ASSERT3U(chunk_size % (vdc->vdc_groupwidth << ashift), ==, 0); ASSERT3U(vdev_draid_offset_to_group(vd, start), ==, vdev_draid_offset_to_group(vd, start + chunk_size - 1)); return (chunk_size); } /* * Align the start of the metaslab to the group width and slightly reduce * its size to a multiple of the group width. Since full stripe writes are * required by dRAID this space is unallocable. Furthermore, aligning the * metaslab start is important for vdev initialize and TRIM which both operate * on metaslab boundaries which vdev_xlate() expects to be aligned. */ static void vdev_draid_metaslab_init(vdev_t *vd, uint64_t *ms_start, uint64_t *ms_size) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); uint64_t sz = vdc->vdc_groupwidth << vd->vdev_ashift; uint64_t astart = vdev_draid_get_astart(vd, *ms_start); uint64_t asize = ((*ms_size - (astart - *ms_start)) / sz) * sz; *ms_start = astart; *ms_size = asize; ASSERT0(*ms_start % sz); ASSERT0(*ms_size % sz); } /* * Add virtual dRAID spares to the list of valid spares. In order to accomplish * this the existing array must be freed and reallocated with the additional * entries. */ int vdev_draid_spare_create(nvlist_t *nvroot, vdev_t *vd, uint64_t *ndraidp, uint64_t next_vdev_id) { uint64_t draid_nspares = 0; uint64_t ndraid = 0; int error; for (uint64_t i = 0; i < vd->vdev_children; i++) { vdev_t *cvd = vd->vdev_child[i]; if (cvd->vdev_ops == &vdev_draid_ops) { vdev_draid_config_t *vdc = cvd->vdev_tsd; draid_nspares += vdc->vdc_nspares; ndraid++; } } if (draid_nspares == 0) { *ndraidp = ndraid; return (0); } nvlist_t **old_spares, **new_spares; uint_t old_nspares; error = nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &old_spares, &old_nspares); if (error) old_nspares = 0; /* Allocate memory and copy of the existing spares. */ new_spares = kmem_alloc(sizeof (nvlist_t *) * (draid_nspares + old_nspares), KM_SLEEP); for (uint_t i = 0; i < old_nspares; i++) new_spares[i] = fnvlist_dup(old_spares[i]); /* Add new distributed spares to ZPOOL_CONFIG_SPARES. */ uint64_t n = old_nspares; for (uint64_t vdev_id = 0; vdev_id < vd->vdev_children; vdev_id++) { vdev_t *cvd = vd->vdev_child[vdev_id]; char path[64]; if (cvd->vdev_ops != &vdev_draid_ops) continue; vdev_draid_config_t *vdc = cvd->vdev_tsd; uint64_t nspares = vdc->vdc_nspares; uint64_t nparity = vdc->vdc_nparity; for (uint64_t spare_id = 0; spare_id < nspares; spare_id++) { bzero(path, sizeof (path)); (void) snprintf(path, sizeof (path) - 1, "%s%llu-%llu-%llu", VDEV_TYPE_DRAID, (u_longlong_t)nparity, (u_longlong_t)next_vdev_id + vdev_id, (u_longlong_t)spare_id); nvlist_t *spare = fnvlist_alloc(); fnvlist_add_string(spare, ZPOOL_CONFIG_PATH, path); fnvlist_add_string(spare, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DRAID_SPARE); fnvlist_add_uint64(spare, ZPOOL_CONFIG_TOP_GUID, cvd->vdev_guid); fnvlist_add_uint64(spare, ZPOOL_CONFIG_SPARE_ID, spare_id); fnvlist_add_uint64(spare, ZPOOL_CONFIG_IS_LOG, 0); fnvlist_add_uint64(spare, ZPOOL_CONFIG_IS_SPARE, 1); fnvlist_add_uint64(spare, ZPOOL_CONFIG_WHOLE_DISK, 1); fnvlist_add_uint64(spare, ZPOOL_CONFIG_ASHIFT, cvd->vdev_ashift); new_spares[n] = spare; n++; } } if (n > 0) { (void) nvlist_remove_all(nvroot, ZPOOL_CONFIG_SPARES); fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, new_spares, n); } for (int i = 0; i < n; i++) nvlist_free(new_spares[i]); kmem_free(new_spares, sizeof (*new_spares) * n); *ndraidp = ndraid; return (0); } /* * Determine if any portion of the provided block resides on a child vdev * with a dirty DTL and therefore needs to be resilvered. */ static boolean_t vdev_draid_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize, uint64_t phys_birth) { uint64_t offset = DVA_GET_OFFSET(dva); uint64_t asize = vdev_draid_asize(vd, psize); if (phys_birth == TXG_UNKNOWN) { /* * Sequential resilver. There is no meaningful phys_birth * for this block, we can only determine if block resides * in a degraded group in which case it must be resilvered. */ ASSERT3U(vdev_draid_offset_to_group(vd, offset), ==, vdev_draid_offset_to_group(vd, offset + asize - 1)); return (vdev_draid_group_degraded(vd, offset)); } else { /* * Healing resilver. TXGs not in DTL_PARTIAL are intact, * as are blocks in non-degraded groups. */ if (!vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1)) return (B_FALSE); if (vdev_draid_group_missing(vd, offset, phys_birth, 1)) return (B_TRUE); /* The block may span groups in which case check both. */ if (vdev_draid_offset_to_group(vd, offset) != vdev_draid_offset_to_group(vd, offset + asize - 1)) { if (vdev_draid_group_missing(vd, offset + asize, phys_birth, 1)) return (B_TRUE); } return (B_FALSE); } } static boolean_t vdev_draid_rebuilding(vdev_t *vd) { if (vd->vdev_ops->vdev_op_leaf && vd->vdev_rebuild_txg) return (B_TRUE); for (int i = 0; i < vd->vdev_children; i++) { if (vdev_draid_rebuilding(vd->vdev_child[i])) { return (B_TRUE); } } return (B_FALSE); } static void vdev_draid_io_verify(vdev_t *vd, raidz_row_t *rr, int col) { #ifdef ZFS_DEBUG range_seg64_t logical_rs, physical_rs, remain_rs; logical_rs.rs_start = rr->rr_offset; logical_rs.rs_end = logical_rs.rs_start + vdev_draid_asize(vd, rr->rr_size); raidz_col_t *rc = &rr->rr_col[col]; vdev_t *cvd = vd->vdev_child[rc->rc_devidx]; vdev_xlate(cvd, &logical_rs, &physical_rs, &remain_rs); ASSERT(vdev_xlate_is_empty(&remain_rs)); ASSERT3U(rc->rc_offset, ==, physical_rs.rs_start); ASSERT3U(rc->rc_offset, <, physical_rs.rs_end); ASSERT3U(rc->rc_offset + rc->rc_size, ==, physical_rs.rs_end); #endif } /* * For write operations: * 1. Generate the parity data * 2. Create child zio write operations to each column's vdev, for both * data and parity. A gang ABD is allocated by vdev_draid_map_alloc() * if a skip sector needs to be added to a column. */ static void vdev_draid_io_start_write(zio_t *zio, raidz_row_t *rr) { vdev_t *vd = zio->io_vd; raidz_map_t *rm = zio->io_vsd; vdev_raidz_generate_parity_row(rm, rr); for (int c = 0; c < rr->rr_cols; c++) { raidz_col_t *rc = &rr->rr_col[c]; /* * Empty columns are zero filled and included in the parity * calculation and therefore must be written. */ ASSERT3U(rc->rc_size, !=, 0); /* Verify physical to logical translation */ vdev_draid_io_verify(vd, rr, c); zio_nowait(zio_vdev_child_io(zio, NULL, vd->vdev_child[rc->rc_devidx], rc->rc_offset, rc->rc_abd, rc->rc_size, zio->io_type, zio->io_priority, 0, vdev_raidz_child_done, rc)); } } /* * For read operations: * 1. The vdev_draid_map_alloc() function will create a minimal raidz * mapping for the read based on the zio->io_flags. There are two * possible mappings either 1) a normal read, or 2) a scrub/resilver. * 2. Create the zio read operations. This will include all parity * columns and skip sectors for a scrub/resilver. */ static void vdev_draid_io_start_read(zio_t *zio, raidz_row_t *rr) { vdev_t *vd = zio->io_vd; /* Sequential rebuild must do IO at redundancy group boundary. */ IMPLY(zio->io_priority == ZIO_PRIORITY_REBUILD, rr->rr_nempty == 0); /* * Iterate over the columns in reverse order so that we hit the parity * last. Any errors along the way will force us to read the parity. * For scrub/resilver IOs which verify skip sectors, a gang ABD will * have been allocated to store them and rc->rc_size is increased. */ for (int c = rr->rr_cols - 1; c >= 0; c--) { raidz_col_t *rc = &rr->rr_col[c]; vdev_t *cvd = vd->vdev_child[rc->rc_devidx]; if (!vdev_draid_readable(cvd, rc->rc_offset)) { if (c >= rr->rr_firstdatacol) rr->rr_missingdata++; else rr->rr_missingparity++; rc->rc_error = SET_ERROR(ENXIO); rc->rc_tried = 1; rc->rc_skipped = 1; continue; } if (vdev_draid_missing(cvd, rc->rc_offset, zio->io_txg, 1)) { if (c >= rr->rr_firstdatacol) rr->rr_missingdata++; else rr->rr_missingparity++; rc->rc_error = SET_ERROR(ESTALE); rc->rc_skipped = 1; continue; } /* * Empty columns may be read during vdev_draid_io_done(). * Only skip them after the readable and missing checks * verify they are available. */ if (rc->rc_size == 0) { rc->rc_skipped = 1; continue; } if (zio->io_flags & ZIO_FLAG_RESILVER) { vdev_t *svd; /* * Sequential rebuilds need to always consider the data * on the child being rebuilt to be stale. This is * important when all columns are available to aid * known reconstruction in identifing which columns * contain incorrect data. * * Furthermore, all repairs need to be constrained to * the devices being rebuilt because without a checksum * we cannot verify the data is actually correct and * performing an incorrect repair could result in * locking in damage and making the data unrecoverable. */ if (zio->io_priority == ZIO_PRIORITY_REBUILD) { if (vdev_draid_rebuilding(cvd)) { if (c >= rr->rr_firstdatacol) rr->rr_missingdata++; else rr->rr_missingparity++; rc->rc_error = SET_ERROR(ESTALE); rc->rc_skipped = 1; rc->rc_allow_repair = 1; continue; } else { rc->rc_allow_repair = 0; } } else { rc->rc_allow_repair = 1; } /* * If this child is a distributed spare then the * offset might reside on the vdev being replaced. * In which case this data must be written to the * new device. Failure to do so would result in * checksum errors when the old device is detached * and the pool is scrubbed. */ if ((svd = vdev_draid_find_spare(cvd)) != NULL) { svd = vdev_draid_spare_get_child(svd, rc->rc_offset); if (svd && (svd->vdev_ops == &vdev_spare_ops || svd->vdev_ops == &vdev_replacing_ops)) { rc->rc_force_repair = 1; if (vdev_draid_rebuilding(svd)) rc->rc_allow_repair = 1; } } /* * Always issue a repair IO to this child when its * a spare or replacing vdev with an active rebuild. */ if ((cvd->vdev_ops == &vdev_spare_ops || cvd->vdev_ops == &vdev_replacing_ops) && vdev_draid_rebuilding(cvd)) { rc->rc_force_repair = 1; rc->rc_allow_repair = 1; } } } /* * Either a parity or data column is missing this means a repair * may be attempted by vdev_draid_io_done(). Expand the raid map * to read in empty columns which are needed along with the parity * during reconstruction. */ if ((rr->rr_missingdata > 0 || rr->rr_missingparity > 0) && rr->rr_nempty > 0 && rr->rr_abd_empty == NULL) { vdev_draid_map_alloc_empty(zio, rr); } for (int c = rr->rr_cols - 1; c >= 0; c--) { raidz_col_t *rc = &rr->rr_col[c]; vdev_t *cvd = vd->vdev_child[rc->rc_devidx]; if (rc->rc_error || rc->rc_size == 0) continue; if (c >= rr->rr_firstdatacol || rr->rr_missingdata > 0 || (zio->io_flags & (ZIO_FLAG_SCRUB | ZIO_FLAG_RESILVER))) { zio_nowait(zio_vdev_child_io(zio, NULL, cvd, rc->rc_offset, rc->rc_abd, rc->rc_size, zio->io_type, zio->io_priority, 0, vdev_raidz_child_done, rc)); } } } /* * Start an IO operation to a dRAID vdev. */ static void vdev_draid_io_start(zio_t *zio) { vdev_t *vd __maybe_unused = zio->io_vd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); ASSERT3U(zio->io_offset, ==, vdev_draid_get_astart(vd, zio->io_offset)); raidz_map_t *rm = vdev_draid_map_alloc(zio); zio->io_vsd = rm; zio->io_vsd_ops = &vdev_raidz_vsd_ops; if (zio->io_type == ZIO_TYPE_WRITE) { for (int i = 0; i < rm->rm_nrows; i++) { vdev_draid_io_start_write(zio, rm->rm_row[i]); } } else { ASSERT(zio->io_type == ZIO_TYPE_READ); for (int i = 0; i < rm->rm_nrows; i++) { vdev_draid_io_start_read(zio, rm->rm_row[i]); } } zio_execute(zio); } /* * Complete an IO operation on a dRAID vdev. The raidz logic can be applied * to dRAID since the layout is fully described by the raidz_map_t. */ static void vdev_draid_io_done(zio_t *zio) { vdev_raidz_io_done(zio); } static void vdev_draid_state_change(vdev_t *vd, int faulted, int degraded) { vdev_draid_config_t *vdc = vd->vdev_tsd; ASSERT(vd->vdev_ops == &vdev_draid_ops); if (faulted > vdc->vdc_nparity) vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_NO_REPLICAS); else if (degraded + faulted != 0) vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, VDEV_AUX_NONE); else vdev_set_state(vd, B_FALSE, VDEV_STATE_HEALTHY, VDEV_AUX_NONE); } static void vdev_draid_xlate(vdev_t *cvd, const range_seg64_t *logical_rs, range_seg64_t *physical_rs, range_seg64_t *remain_rs) { vdev_t *raidvd = cvd->vdev_parent; ASSERT(raidvd->vdev_ops == &vdev_draid_ops); vdev_draid_config_t *vdc = raidvd->vdev_tsd; uint64_t ashift = raidvd->vdev_top->vdev_ashift; /* Make sure the offsets are block-aligned */ ASSERT0(logical_rs->rs_start % (1 << ashift)); ASSERT0(logical_rs->rs_end % (1 << ashift)); uint64_t logical_start = logical_rs->rs_start; uint64_t logical_end = logical_rs->rs_end; /* * Unaligned ranges must be skipped. All metaslabs are correctly * aligned so this should not happen, but this case is handled in * case it's needed by future callers. */ uint64_t astart = vdev_draid_get_astart(raidvd, logical_start); if (astart != logical_start) { physical_rs->rs_start = logical_start; physical_rs->rs_end = logical_start; remain_rs->rs_start = MIN(astart, logical_end); remain_rs->rs_end = logical_end; return; } /* * Unlike with mirrors and raidz a dRAID logical range can map * to multiple non-contiguous physical ranges. This is handled by * limiting the size of the logical range to a single group and * setting the remain argument such that it describes the remaining * unmapped logical range. This is stricter than absolutely * necessary but helps simplify the logic below. */ uint64_t group = vdev_draid_offset_to_group(raidvd, logical_start); uint64_t nextstart = vdev_draid_group_to_offset(raidvd, group + 1); if (logical_end > nextstart) logical_end = nextstart; /* Find the starting offset for each vdev in the group */ uint64_t perm, groupstart; uint64_t start = vdev_draid_logical_to_physical(raidvd, logical_start, &perm, &groupstart); uint64_t end = start; uint8_t *base; uint64_t iter, id; vdev_draid_get_perm(vdc, perm, &base, &iter); /* * Check if the passed child falls within the group. If it does * update the start and end to reflect the physical range. * Otherwise, leave them unmodified which will result in an empty * (zero-length) physical range being returned. */ for (uint64_t i = 0; i < vdc->vdc_groupwidth; i++) { uint64_t c = (groupstart + i) % vdc->vdc_ndisks; if (c == 0 && i != 0) { /* the group wrapped, increment the start */ start += VDEV_DRAID_ROWHEIGHT; end = start; } id = vdev_draid_permute_id(vdc, base, iter, c); if (id == cvd->vdev_id) { uint64_t b_size = (logical_end >> ashift) - (logical_start >> ashift); ASSERT3U(b_size, >, 0); end = start + ((((b_size - 1) / vdc->vdc_groupwidth) + 1) << ashift); break; } } physical_rs->rs_start = start; physical_rs->rs_end = end; /* * Only top-level vdevs are allowed to set remain_rs because * when .vdev_op_xlate() is called for their children the full * logical range is not provided by vdev_xlate(). */ remain_rs->rs_start = logical_end; remain_rs->rs_end = logical_rs->rs_end; ASSERT3U(physical_rs->rs_start, <=, logical_start); ASSERT3U(physical_rs->rs_end - physical_rs->rs_start, <=, logical_end - logical_start); } /* * Add dRAID specific fields to the config nvlist. */ static void vdev_draid_config_generate(vdev_t *vd, nvlist_t *nv) { ASSERT3P(vd->vdev_ops, ==, &vdev_draid_ops); vdev_draid_config_t *vdc = vd->vdev_tsd; fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vdc->vdc_nparity); fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NDATA, vdc->vdc_ndata); fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NSPARES, vdc->vdc_nspares); fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NGROUPS, vdc->vdc_ngroups); } /* * Initialize private dRAID specific fields from the nvlist. */ static int vdev_draid_init(spa_t *spa, nvlist_t *nv, void **tsd) { uint64_t ndata, nparity, nspares, ngroups; int error; if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DRAID_NDATA, &ndata)) return (SET_ERROR(EINVAL)); if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, &nparity) || nparity == 0 || nparity > VDEV_DRAID_MAXPARITY) { return (SET_ERROR(EINVAL)); } uint_t children; nvlist_t **child; if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, &child, &children) != 0 || children == 0 || children > VDEV_DRAID_MAX_CHILDREN) { return (SET_ERROR(EINVAL)); } if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DRAID_NSPARES, &nspares) || nspares > 100 || nspares > (children - (ndata + nparity))) { return (SET_ERROR(EINVAL)); } if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DRAID_NGROUPS, &ngroups) || ngroups == 0 || ngroups > VDEV_DRAID_MAX_CHILDREN) { return (SET_ERROR(EINVAL)); } /* * Validate the minimum number of children exist per group for the * specified parity level (draid1 >= 2, draid2 >= 3, draid3 >= 4). */ if (children < (ndata + nparity + nspares)) return (SET_ERROR(EINVAL)); /* * Create the dRAID configuration using the pool nvlist configuration * and the fixed mapping for the correct number of children. */ vdev_draid_config_t *vdc; const draid_map_t *map; error = vdev_draid_lookup_map(children, &map); if (error) return (SET_ERROR(EINVAL)); vdc = kmem_zalloc(sizeof (*vdc), KM_SLEEP); vdc->vdc_ndata = ndata; vdc->vdc_nparity = nparity; vdc->vdc_nspares = nspares; vdc->vdc_children = children; vdc->vdc_ngroups = ngroups; vdc->vdc_nperms = map->dm_nperms; error = vdev_draid_generate_perms(map, &vdc->vdc_perms); if (error) { kmem_free(vdc, sizeof (*vdc)); return (SET_ERROR(EINVAL)); } /* * Derived constants. */ vdc->vdc_groupwidth = vdc->vdc_ndata + vdc->vdc_nparity; vdc->vdc_ndisks = vdc->vdc_children - vdc->vdc_nspares; vdc->vdc_groupsz = vdc->vdc_groupwidth * VDEV_DRAID_ROWHEIGHT; vdc->vdc_devslicesz = (vdc->vdc_groupsz * vdc->vdc_ngroups) / vdc->vdc_ndisks; ASSERT3U(vdc->vdc_groupwidth, >=, 2); ASSERT3U(vdc->vdc_groupwidth, <=, vdc->vdc_ndisks); ASSERT3U(vdc->vdc_groupsz, >=, 2 * VDEV_DRAID_ROWHEIGHT); ASSERT3U(vdc->vdc_devslicesz, >=, VDEV_DRAID_ROWHEIGHT); ASSERT3U(vdc->vdc_devslicesz % VDEV_DRAID_ROWHEIGHT, ==, 0); ASSERT3U((vdc->vdc_groupwidth * vdc->vdc_ngroups) % vdc->vdc_ndisks, ==, 0); *tsd = vdc; return (0); } static void vdev_draid_fini(vdev_t *vd) { vdev_draid_config_t *vdc = vd->vdev_tsd; vmem_free(vdc->vdc_perms, sizeof (uint8_t) * vdc->vdc_children * vdc->vdc_nperms); kmem_free(vdc, sizeof (*vdc)); } static uint64_t vdev_draid_nparity(vdev_t *vd) { vdev_draid_config_t *vdc = vd->vdev_tsd; return (vdc->vdc_nparity); } static uint64_t vdev_draid_ndisks(vdev_t *vd) { vdev_draid_config_t *vdc = vd->vdev_tsd; return (vdc->vdc_ndisks); } vdev_ops_t vdev_draid_ops = { .vdev_op_init = vdev_draid_init, .vdev_op_fini = vdev_draid_fini, .vdev_op_open = vdev_draid_open, .vdev_op_close = vdev_draid_close, .vdev_op_asize = vdev_draid_asize, .vdev_op_min_asize = vdev_draid_min_asize, .vdev_op_min_alloc = vdev_draid_min_alloc, .vdev_op_io_start = vdev_draid_io_start, .vdev_op_io_done = vdev_draid_io_done, .vdev_op_state_change = vdev_draid_state_change, .vdev_op_need_resilver = vdev_draid_need_resilver, .vdev_op_hold = NULL, .vdev_op_rele = NULL, .vdev_op_remap = NULL, .vdev_op_xlate = vdev_draid_xlate, .vdev_op_rebuild_asize = vdev_draid_rebuild_asize, .vdev_op_metaslab_init = vdev_draid_metaslab_init, .vdev_op_config_generate = vdev_draid_config_generate, .vdev_op_nparity = vdev_draid_nparity, .vdev_op_ndisks = vdev_draid_ndisks, .vdev_op_type = VDEV_TYPE_DRAID, .vdev_op_leaf = B_FALSE, }; /* * A dRAID distributed spare is a virtual leaf vdev which is included in the * parent dRAID configuration. The last N columns of the dRAID permutation * table are used to determine on which dRAID children a specific offset * should be written. These spare leaf vdevs can only be used to replace * faulted children in the same dRAID configuration. */ /* * Distributed spare state. All fields are set when the distributed spare is * first opened and are immutable. */ typedef struct { vdev_t *vds_draid_vdev; /* top-level parent dRAID vdev */ uint64_t vds_top_guid; /* top-level parent dRAID guid */ uint64_t vds_spare_id; /* spare id (0 - vdc->vdc_nspares-1) */ } vdev_draid_spare_t; /* * Returns the parent dRAID vdev to which the distributed spare belongs. * This may be safely called even when the vdev is not open. */ vdev_t * vdev_draid_spare_get_parent(vdev_t *vd) { vdev_draid_spare_t *vds = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_spare_ops); if (vds->vds_draid_vdev != NULL) return (vds->vds_draid_vdev); return (vdev_lookup_by_guid(vd->vdev_spa->spa_root_vdev, vds->vds_top_guid)); } /* * A dRAID space is active when it's the child of a vdev using the * vdev_spare_ops, vdev_replacing_ops or vdev_draid_ops. */ static boolean_t vdev_draid_spare_is_active(vdev_t *vd) { vdev_t *pvd = vd->vdev_parent; if (pvd != NULL && (pvd->vdev_ops == &vdev_spare_ops || pvd->vdev_ops == &vdev_replacing_ops || pvd->vdev_ops == &vdev_draid_ops)) { return (B_TRUE); } else { return (B_FALSE); } } /* * Given a dRAID distribute spare vdev, returns the physical child vdev * on which the provided offset resides. This may involve recursing through * multiple layers of distributed spares. Note that offset is relative to * this vdev. */ vdev_t * vdev_draid_spare_get_child(vdev_t *vd, uint64_t physical_offset) { vdev_draid_spare_t *vds = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_spare_ops); /* The vdev is closed */ if (vds->vds_draid_vdev == NULL) return (NULL); vdev_t *tvd = vds->vds_draid_vdev; vdev_draid_config_t *vdc = tvd->vdev_tsd; ASSERT3P(tvd->vdev_ops, ==, &vdev_draid_ops); ASSERT3U(vds->vds_spare_id, <, vdc->vdc_nspares); uint8_t *base; uint64_t iter; uint64_t perm = physical_offset / vdc->vdc_devslicesz; vdev_draid_get_perm(vdc, perm, &base, &iter); uint64_t cid = vdev_draid_permute_id(vdc, base, iter, (tvd->vdev_children - 1) - vds->vds_spare_id); vdev_t *cvd = tvd->vdev_child[cid]; if (cvd->vdev_ops == &vdev_draid_spare_ops) return (vdev_draid_spare_get_child(cvd, physical_offset)); return (cvd); } /* ARGSUSED */ static void vdev_draid_spare_close(vdev_t *vd) { vdev_draid_spare_t *vds = vd->vdev_tsd; vds->vds_draid_vdev = NULL; } /* * Opening a dRAID spare device is done by looking up the associated dRAID * top-level vdev guid from the spare configuration. */ static int vdev_draid_spare_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize, uint64_t *logical_ashift, uint64_t *physical_ashift) { vdev_draid_spare_t *vds = vd->vdev_tsd; vdev_t *rvd = vd->vdev_spa->spa_root_vdev; uint64_t asize, max_asize; vdev_t *tvd = vdev_lookup_by_guid(rvd, vds->vds_top_guid); if (tvd == NULL) { /* * When spa_vdev_add() is labeling new spares the * associated dRAID is not attached to the root vdev * nor does this spare have a parent. Simulate a valid * device in order to allow the label to be initialized * and the distributed spare added to the configuration. */ if (vd->vdev_parent == NULL) { *psize = *max_psize = SPA_MINDEVSIZE; *logical_ashift = *physical_ashift = ASHIFT_MIN; return (0); } return (SET_ERROR(EINVAL)); } vdev_draid_config_t *vdc = tvd->vdev_tsd; if (tvd->vdev_ops != &vdev_draid_ops || vdc == NULL) return (SET_ERROR(EINVAL)); if (vds->vds_spare_id >= vdc->vdc_nspares) return (SET_ERROR(EINVAL)); /* * Neither tvd->vdev_asize or tvd->vdev_max_asize can be used here * because the caller may be vdev_draid_open() in which case the * values are stale as they haven't yet been updated by vdev_open(). * To avoid this always recalculate the dRAID asize and max_asize. */ vdev_draid_calculate_asize(tvd, &asize, &max_asize, logical_ashift, physical_ashift); *psize = asize + VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; *max_psize = max_asize + VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; vds->vds_draid_vdev = tvd; return (0); } /* * Completed distributed spare IO. Store the result in the parent zio * as if it had performed the operation itself. Only the first error is * preserved if there are multiple errors. */ static void vdev_draid_spare_child_done(zio_t *zio) { zio_t *pio = zio->io_private; /* * IOs are issued to non-writable vdevs in order to keep their * DTLs accurate. However, we don't want to propagate the * error in to the distributed spare's DTL. When resilvering * vdev_draid_need_resilver() will consult the relevant DTL * to determine if the data is missing and must be repaired. */ if (!vdev_writeable(zio->io_vd)) return; if (pio->io_error == 0) pio->io_error = zio->io_error; } /* * Returns a valid label nvlist for the distributed spare vdev. This is * used to bypass the IO pipeline to avoid the complexity of constructing * a complete label with valid checksum to return when read. */ nvlist_t * vdev_draid_read_config_spare(vdev_t *vd) { spa_t *spa = vd->vdev_spa; spa_aux_vdev_t *sav = &spa->spa_spares; uint64_t guid = vd->vdev_guid; nvlist_t *nv = fnvlist_alloc(); fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); fnvlist_add_uint64(nv, ZPOOL_CONFIG_VERSION, spa_version(spa)); fnvlist_add_string(nv, ZPOOL_CONFIG_POOL_NAME, spa_name(spa)); fnvlist_add_uint64(nv, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa)); fnvlist_add_uint64(nv, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg); fnvlist_add_uint64(nv, ZPOOL_CONFIG_TOP_GUID, vd->vdev_top->vdev_guid); fnvlist_add_uint64(nv, ZPOOL_CONFIG_POOL_STATE, vdev_draid_spare_is_active(vd) ? POOL_STATE_ACTIVE : POOL_STATE_SPARE); /* Set the vdev guid based on the vdev list in sav_count. */ for (int i = 0; i < sav->sav_count; i++) { if (sav->sav_vdevs[i]->vdev_ops == &vdev_draid_spare_ops && strcmp(sav->sav_vdevs[i]->vdev_path, vd->vdev_path) == 0) { guid = sav->sav_vdevs[i]->vdev_guid; break; } } fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, guid); return (nv); } /* * Handle any ioctl requested of the distributed spare. Only flushes * are supported in which case all children must be flushed. */ static int vdev_draid_spare_ioctl(zio_t *zio) { vdev_t *vd = zio->io_vd; int error = 0; if (zio->io_cmd == DKIOCFLUSHWRITECACHE) { for (int c = 0; c < vd->vdev_children; c++) { zio_nowait(zio_vdev_child_io(zio, NULL, vd->vdev_child[c], zio->io_offset, zio->io_abd, zio->io_size, zio->io_type, zio->io_priority, 0, vdev_draid_spare_child_done, zio)); } } else { error = SET_ERROR(ENOTSUP); } return (error); } /* * Initiate an IO to the distributed spare. For normal IOs this entails using * the zio->io_offset and permutation table to calculate which child dRAID vdev * is responsible for the data. Then passing along the zio to that child to * perform the actual IO. The label ranges are not stored on disk and require * some special handling which is described below. */ static void vdev_draid_spare_io_start(zio_t *zio) { vdev_t *cvd = NULL, *vd = zio->io_vd; vdev_draid_spare_t *vds = vd->vdev_tsd; uint64_t offset = zio->io_offset - VDEV_LABEL_START_SIZE; /* * If the vdev is closed, it's likely in the REMOVED or FAULTED state. * Nothing to be done here but return failure. */ if (vds == NULL) { zio->io_error = ENXIO; zio_interrupt(zio); return; } switch (zio->io_type) { case ZIO_TYPE_IOCTL: zio->io_error = vdev_draid_spare_ioctl(zio); break; case ZIO_TYPE_WRITE: if (VDEV_OFFSET_IS_LABEL(vd, zio->io_offset)) { /* * Accept probe IOs and config writers to simulate the * existence of an on disk label. vdev_label_sync(), * vdev_uberblock_sync() and vdev_copy_uberblocks() * skip the distributed spares. This only leaves * vdev_label_init() which is allowed to succeed to * avoid adding special cases the function. */ if (zio->io_flags & ZIO_FLAG_PROBE || zio->io_flags & ZIO_FLAG_CONFIG_WRITER) { zio->io_error = 0; } else { zio->io_error = SET_ERROR(EIO); } } else { cvd = vdev_draid_spare_get_child(vd, offset); if (cvd == NULL) { zio->io_error = SET_ERROR(ENXIO); } else { zio_nowait(zio_vdev_child_io(zio, NULL, cvd, offset, zio->io_abd, zio->io_size, zio->io_type, zio->io_priority, 0, vdev_draid_spare_child_done, zio)); } } break; case ZIO_TYPE_READ: if (VDEV_OFFSET_IS_LABEL(vd, zio->io_offset)) { /* * Accept probe IOs to simulate the existence of a * label. vdev_label_read_config() bypasses the * pipeline to read the label configuration and * vdev_uberblock_load() skips distributed spares * when attempting to locate the best uberblock. */ if (zio->io_flags & ZIO_FLAG_PROBE) { zio->io_error = 0; } else { zio->io_error = SET_ERROR(EIO); } } else { cvd = vdev_draid_spare_get_child(vd, offset); if (cvd == NULL || !vdev_readable(cvd)) { zio->io_error = SET_ERROR(ENXIO); } else { zio_nowait(zio_vdev_child_io(zio, NULL, cvd, offset, zio->io_abd, zio->io_size, zio->io_type, zio->io_priority, 0, vdev_draid_spare_child_done, zio)); } } break; case ZIO_TYPE_TRIM: /* The vdev label ranges are never trimmed */ ASSERT0(VDEV_OFFSET_IS_LABEL(vd, zio->io_offset)); cvd = vdev_draid_spare_get_child(vd, offset); if (cvd == NULL || !cvd->vdev_has_trim) { zio->io_error = SET_ERROR(ENXIO); } else { zio_nowait(zio_vdev_child_io(zio, NULL, cvd, offset, zio->io_abd, zio->io_size, zio->io_type, zio->io_priority, 0, vdev_draid_spare_child_done, zio)); } break; default: zio->io_error = SET_ERROR(ENOTSUP); break; } zio_execute(zio); } /* ARGSUSED */ static void vdev_draid_spare_io_done(zio_t *zio) { } /* * Lookup the full spare config in spa->spa_spares.sav_config and * return the top_guid and spare_id for the named spare. */ static int vdev_draid_spare_lookup(spa_t *spa, nvlist_t *nv, uint64_t *top_guidp, uint64_t *spare_idp) { nvlist_t **spares; uint_t nspares; int error; if ((spa->spa_spares.sav_config == NULL) || (nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0)) { return (SET_ERROR(ENOENT)); } char *spare_name; error = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &spare_name); if (error != 0) return (SET_ERROR(EINVAL)); for (int i = 0; i < nspares; i++) { nvlist_t *spare = spares[i]; uint64_t top_guid, spare_id; char *type, *path; /* Skip non-distributed spares */ error = nvlist_lookup_string(spare, ZPOOL_CONFIG_TYPE, &type); if (error != 0 || strcmp(type, VDEV_TYPE_DRAID_SPARE) != 0) continue; /* Skip spares with the wrong name */ error = nvlist_lookup_string(spare, ZPOOL_CONFIG_PATH, &path); if (error != 0 || strcmp(path, spare_name) != 0) continue; /* Found the matching spare */ error = nvlist_lookup_uint64(spare, ZPOOL_CONFIG_TOP_GUID, &top_guid); if (error == 0) { error = nvlist_lookup_uint64(spare, ZPOOL_CONFIG_SPARE_ID, &spare_id); } if (error != 0) { return (SET_ERROR(EINVAL)); } else { *top_guidp = top_guid; *spare_idp = spare_id; return (0); } } return (SET_ERROR(ENOENT)); } /* * Initialize private dRAID spare specific fields from the nvlist. */ static int vdev_draid_spare_init(spa_t *spa, nvlist_t *nv, void **tsd) { vdev_draid_spare_t *vds; uint64_t top_guid = 0; uint64_t spare_id; /* * In the normal case check the list of spares stored in the spa * to lookup the top_guid and spare_id for provided spare config. * When creating a new pool or adding vdevs the spare list is not * yet populated and the values are provided in the passed config. */ if (vdev_draid_spare_lookup(spa, nv, &top_guid, &spare_id) != 0) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_TOP_GUID, &top_guid) != 0) return (SET_ERROR(EINVAL)); if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_SPARE_ID, &spare_id) != 0) return (SET_ERROR(EINVAL)); } vds = kmem_alloc(sizeof (vdev_draid_spare_t), KM_SLEEP); vds->vds_draid_vdev = NULL; vds->vds_top_guid = top_guid; vds->vds_spare_id = spare_id; *tsd = vds; return (0); } static void vdev_draid_spare_fini(vdev_t *vd) { kmem_free(vd->vdev_tsd, sizeof (vdev_draid_spare_t)); } static void vdev_draid_spare_config_generate(vdev_t *vd, nvlist_t *nv) { vdev_draid_spare_t *vds = vd->vdev_tsd; ASSERT3P(vd->vdev_ops, ==, &vdev_draid_spare_ops); fnvlist_add_uint64(nv, ZPOOL_CONFIG_TOP_GUID, vds->vds_top_guid); fnvlist_add_uint64(nv, ZPOOL_CONFIG_SPARE_ID, vds->vds_spare_id); } vdev_ops_t vdev_draid_spare_ops = { .vdev_op_init = vdev_draid_spare_init, .vdev_op_fini = vdev_draid_spare_fini, .vdev_op_open = vdev_draid_spare_open, .vdev_op_close = vdev_draid_spare_close, .vdev_op_asize = vdev_default_asize, .vdev_op_min_asize = vdev_default_min_asize, .vdev_op_min_alloc = NULL, .vdev_op_io_start = vdev_draid_spare_io_start, .vdev_op_io_done = vdev_draid_spare_io_done, .vdev_op_state_change = NULL, .vdev_op_need_resilver = NULL, .vdev_op_hold = NULL, .vdev_op_rele = NULL, .vdev_op_remap = NULL, .vdev_op_xlate = vdev_default_xlate, .vdev_op_rebuild_asize = NULL, .vdev_op_metaslab_init = NULL, .vdev_op_config_generate = vdev_draid_spare_config_generate, .vdev_op_nparity = NULL, .vdev_op_ndisks = NULL, .vdev_op_type = VDEV_TYPE_DRAID_SPARE, .vdev_op_leaf = B_TRUE, }; diff --git a/sys/contrib/openzfs/module/zfs/zio.c b/sys/contrib/openzfs/module/zfs/zio.c index 87ccb6861850..66ac545c7981 100644 --- a/sys/contrib/openzfs/module/zfs/zio.c +++ b/sys/contrib/openzfs/module/zfs/zio.c @@ -1,5036 +1,5033 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2020 by Delphix. All rights reserved. * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2017, Intel Corporation. * Copyright (c) 2019, Klara Inc. * Copyright (c) 2019, Allan Jude * Copyright (c) 2021, Datto, Inc. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * ========================================================================== * I/O type descriptions * ========================================================================== */ const char *zio_type_name[ZIO_TYPES] = { /* * Note: Linux kernel thread name length is limited * so these names will differ from upstream open zfs. */ "z_null", "z_rd", "z_wr", "z_fr", "z_cl", "z_ioctl", "z_trim" }; int zio_dva_throttle_enabled = B_TRUE; int zio_deadman_log_all = B_FALSE; /* * ========================================================================== * I/O kmem caches * ========================================================================== */ kmem_cache_t *zio_cache; kmem_cache_t *zio_link_cache; kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; #if defined(ZFS_DEBUG) && !defined(_KERNEL) uint64_t zio_buf_cache_allocs[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; uint64_t zio_buf_cache_frees[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; #endif /* Mark IOs as "slow" if they take longer than 30 seconds */ int zio_slow_io_ms = (30 * MILLISEC); #define BP_SPANB(indblkshift, level) \ (((uint64_t)1) << ((level) * ((indblkshift) - SPA_BLKPTRSHIFT))) #define COMPARE_META_LEVEL 0x80000000ul /* * The following actions directly effect the spa's sync-to-convergence logic. * The values below define the sync pass when we start performing the action. * Care should be taken when changing these values as they directly impact * spa_sync() performance. Tuning these values may introduce subtle performance * pathologies and should only be done in the context of performance analysis. * These tunables will eventually be removed and replaced with #defines once * enough analysis has been done to determine optimal values. * * The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that * regular blocks are not deferred. * * Starting in sync pass 8 (zfs_sync_pass_dont_compress), we disable * compression (including of metadata). In practice, we don't have this * many sync passes, so this has no effect. * * The original intent was that disabling compression would help the sync * passes to converge. However, in practice disabling compression increases * the average number of sync passes, because when we turn compression off, a * lot of block's size will change and thus we have to re-allocate (not * overwrite) them. It also increases the number of 128KB allocations (e.g. * for indirect blocks and spacemaps) because these will not be compressed. * The 128K allocations are especially detrimental to performance on highly * fragmented systems, which may have very few free segments of this size, * and may need to load new metaslabs to satisfy 128K allocations. */ int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */ int zfs_sync_pass_dont_compress = 8; /* don't compress starting in this pass */ int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */ /* * An allocating zio is one that either currently has the DVA allocate * stage set or will have it later in its lifetime. */ #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) /* * Enable smaller cores by excluding metadata * allocations as well. */ int zio_exclude_metadata = 0; int zio_requeue_io_start_cut_in_line = 1; #ifdef ZFS_DEBUG int zio_buf_debug_limit = 16384; #else int zio_buf_debug_limit = 0; #endif static inline void __zio_execute(zio_t *zio); static void zio_taskq_dispatch(zio_t *, zio_taskq_type_t, boolean_t); void zio_init(void) { size_t c; zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, 0); zio_link_cache = kmem_cache_create("zio_link_cache", sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0); /* * For small buffers, we want a cache for each multiple of * SPA_MINBLOCKSIZE. For larger buffers, we want a cache * for each quarter-power of 2. */ for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { size_t size = (c + 1) << SPA_MINBLOCKSHIFT; size_t p2 = size; size_t align = 0; size_t data_cflags, cflags; data_cflags = KMC_NODEBUG; cflags = (zio_exclude_metadata || size > zio_buf_debug_limit) ? KMC_NODEBUG : 0; #if defined(_ILP32) && defined(_KERNEL) /* * Cache size limited to 1M on 32-bit platforms until ARC * buffers no longer require virtual address space. */ if (size > zfs_max_recordsize) break; #endif while (!ISP2(p2)) p2 &= p2 - 1; #ifndef _KERNEL /* * If we are using watchpoints, put each buffer on its own page, * to eliminate the performance overhead of trapping to the * kernel when modifying a non-watched buffer that shares the * page with a watched buffer. */ if (arc_watch && !IS_P2ALIGNED(size, PAGESIZE)) continue; /* * Here's the problem - on 4K native devices in userland on * Linux using O_DIRECT, buffers must be 4K aligned or I/O * will fail with EINVAL, causing zdb (and others) to coredump. * Since userland probably doesn't need optimized buffer caches, * we just force 4K alignment on everything. */ align = 8 * SPA_MINBLOCKSIZE; #else if (size < PAGESIZE) { align = SPA_MINBLOCKSIZE; } else if (IS_P2ALIGNED(size, p2 >> 2)) { align = PAGESIZE; } #endif if (align != 0) { char name[36]; if (cflags == data_cflags) { /* * Resulting kmem caches would be identical. * Save memory by creating only one. */ (void) snprintf(name, sizeof (name), "zio_buf_comb_%lu", (ulong_t)size); zio_buf_cache[c] = kmem_cache_create(name, size, align, NULL, NULL, NULL, NULL, NULL, cflags); zio_data_buf_cache[c] = zio_buf_cache[c]; continue; } (void) snprintf(name, sizeof (name), "zio_buf_%lu", (ulong_t)size); zio_buf_cache[c] = kmem_cache_create(name, size, align, NULL, NULL, NULL, NULL, NULL, cflags); (void) snprintf(name, sizeof (name), "zio_data_buf_%lu", (ulong_t)size); zio_data_buf_cache[c] = kmem_cache_create(name, size, align, NULL, NULL, NULL, NULL, NULL, data_cflags); } } while (--c != 0) { ASSERT(zio_buf_cache[c] != NULL); if (zio_buf_cache[c - 1] == NULL) zio_buf_cache[c - 1] = zio_buf_cache[c]; ASSERT(zio_data_buf_cache[c] != NULL); if (zio_data_buf_cache[c - 1] == NULL) zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; } zio_inject_init(); lz4_init(); } void zio_fini(void) { size_t n = SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; #if defined(ZFS_DEBUG) && !defined(_KERNEL) for (size_t i = 0; i < n; i++) { if (zio_buf_cache_allocs[i] != zio_buf_cache_frees[i]) (void) printf("zio_fini: [%d] %llu != %llu\n", (int)((i + 1) << SPA_MINBLOCKSHIFT), (long long unsigned)zio_buf_cache_allocs[i], (long long unsigned)zio_buf_cache_frees[i]); } #endif /* * The same kmem cache can show up multiple times in both zio_buf_cache * and zio_data_buf_cache. Do a wasteful but trivially correct scan to * sort it out. */ for (size_t i = 0; i < n; i++) { kmem_cache_t *cache = zio_buf_cache[i]; if (cache == NULL) continue; for (size_t j = i; j < n; j++) { if (cache == zio_buf_cache[j]) zio_buf_cache[j] = NULL; if (cache == zio_data_buf_cache[j]) zio_data_buf_cache[j] = NULL; } kmem_cache_destroy(cache); } for (size_t i = 0; i < n; i++) { kmem_cache_t *cache = zio_data_buf_cache[i]; if (cache == NULL) continue; for (size_t j = i; j < n; j++) { if (cache == zio_data_buf_cache[j]) zio_data_buf_cache[j] = NULL; } kmem_cache_destroy(cache); } for (size_t i = 0; i < n; i++) { VERIFY3P(zio_buf_cache[i], ==, NULL); VERIFY3P(zio_data_buf_cache[i], ==, NULL); } kmem_cache_destroy(zio_link_cache); kmem_cache_destroy(zio_cache); zio_inject_fini(); lz4_fini(); } /* * ========================================================================== * Allocate and free I/O buffers * ========================================================================== */ /* * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a * crashdump if the kernel panics, so use it judiciously. Obviously, it's * useful to inspect ZFS metadata, but if possible, we should avoid keeping * excess / transient data in-core during a crashdump. */ void * zio_buf_alloc(size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); #if defined(ZFS_DEBUG) && !defined(_KERNEL) atomic_add_64(&zio_buf_cache_allocs[c], 1); #endif return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); } /* * Use zio_data_buf_alloc to allocate data. The data will not appear in a * crashdump if the kernel panics. This exists so that we will limit the amount * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount * of kernel heap dumped to disk when the kernel panics) */ void * zio_data_buf_alloc(size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); } void zio_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); #if defined(ZFS_DEBUG) && !defined(_KERNEL) atomic_add_64(&zio_buf_cache_frees[c], 1); #endif kmem_cache_free(zio_buf_cache[c], buf); } void zio_data_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); kmem_cache_free(zio_data_buf_cache[c], buf); } static void zio_abd_free(void *abd, size_t size) { abd_free((abd_t *)abd); } /* * ========================================================================== * Push and pop I/O transform buffers * ========================================================================== */ void zio_push_transform(zio_t *zio, abd_t *data, uint64_t size, uint64_t bufsize, zio_transform_func_t *transform) { zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP); zt->zt_orig_abd = zio->io_abd; zt->zt_orig_size = zio->io_size; zt->zt_bufsize = bufsize; zt->zt_transform = transform; zt->zt_next = zio->io_transform_stack; zio->io_transform_stack = zt; zio->io_abd = data; zio->io_size = size; } void zio_pop_transforms(zio_t *zio) { zio_transform_t *zt; while ((zt = zio->io_transform_stack) != NULL) { if (zt->zt_transform != NULL) zt->zt_transform(zio, zt->zt_orig_abd, zt->zt_orig_size); if (zt->zt_bufsize != 0) abd_free(zio->io_abd); zio->io_abd = zt->zt_orig_abd; zio->io_size = zt->zt_orig_size; zio->io_transform_stack = zt->zt_next; kmem_free(zt, sizeof (zio_transform_t)); } } /* * ========================================================================== * I/O transform callbacks for subblocks, decompression, and decryption * ========================================================================== */ static void zio_subblock(zio_t *zio, abd_t *data, uint64_t size) { ASSERT(zio->io_size > size); if (zio->io_type == ZIO_TYPE_READ) abd_copy(data, zio->io_abd, size); } static void zio_decompress(zio_t *zio, abd_t *data, uint64_t size) { if (zio->io_error == 0) { void *tmp = abd_borrow_buf(data, size); int ret = zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), zio->io_abd, tmp, zio->io_size, size, &zio->io_prop.zp_complevel); abd_return_buf_copy(data, tmp, size); if (zio_injection_enabled && ret == 0) ret = zio_handle_fault_injection(zio, EINVAL); if (ret != 0) zio->io_error = SET_ERROR(EIO); } } static void zio_decrypt(zio_t *zio, abd_t *data, uint64_t size) { int ret; void *tmp; blkptr_t *bp = zio->io_bp; spa_t *spa = zio->io_spa; uint64_t dsobj = zio->io_bookmark.zb_objset; uint64_t lsize = BP_GET_LSIZE(bp); dmu_object_type_t ot = BP_GET_TYPE(bp); uint8_t salt[ZIO_DATA_SALT_LEN]; uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t mac[ZIO_DATA_MAC_LEN]; boolean_t no_crypt = B_FALSE; ASSERT(BP_USES_CRYPT(bp)); ASSERT3U(size, !=, 0); if (zio->io_error != 0) return; /* * Verify the cksum of MACs stored in an indirect bp. It will always * be possible to verify this since it does not require an encryption * key. */ if (BP_HAS_INDIRECT_MAC_CKSUM(bp)) { zio_crypt_decode_mac_bp(bp, mac); if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF) { /* * We haven't decompressed the data yet, but * zio_crypt_do_indirect_mac_checksum() requires * decompressed data to be able to parse out the MACs * from the indirect block. We decompress it now and * throw away the result after we are finished. */ tmp = zio_buf_alloc(lsize); ret = zio_decompress_data(BP_GET_COMPRESS(bp), zio->io_abd, tmp, zio->io_size, lsize, &zio->io_prop.zp_complevel); if (ret != 0) { ret = SET_ERROR(EIO); goto error; } ret = zio_crypt_do_indirect_mac_checksum(B_FALSE, tmp, lsize, BP_SHOULD_BYTESWAP(bp), mac); zio_buf_free(tmp, lsize); } else { ret = zio_crypt_do_indirect_mac_checksum_abd(B_FALSE, zio->io_abd, size, BP_SHOULD_BYTESWAP(bp), mac); } abd_copy(data, zio->io_abd, size); if (zio_injection_enabled && ot != DMU_OT_DNODE && ret == 0) { ret = zio_handle_decrypt_injection(spa, &zio->io_bookmark, ot, ECKSUM); } if (ret != 0) goto error; return; } /* * If this is an authenticated block, just check the MAC. It would be * nice to separate this out into its own flag, but for the moment * enum zio_flag is out of bits. */ if (BP_IS_AUTHENTICATED(bp)) { if (ot == DMU_OT_OBJSET) { ret = spa_do_crypt_objset_mac_abd(B_FALSE, spa, dsobj, zio->io_abd, size, BP_SHOULD_BYTESWAP(bp)); } else { zio_crypt_decode_mac_bp(bp, mac); ret = spa_do_crypt_mac_abd(B_FALSE, spa, dsobj, zio->io_abd, size, mac); if (zio_injection_enabled && ret == 0) { ret = zio_handle_decrypt_injection(spa, &zio->io_bookmark, ot, ECKSUM); } } abd_copy(data, zio->io_abd, size); if (ret != 0) goto error; return; } zio_crypt_decode_params_bp(bp, salt, iv); if (ot == DMU_OT_INTENT_LOG) { tmp = abd_borrow_buf_copy(zio->io_abd, sizeof (zil_chain_t)); zio_crypt_decode_mac_zil(tmp, mac); abd_return_buf(zio->io_abd, tmp, sizeof (zil_chain_t)); } else { zio_crypt_decode_mac_bp(bp, mac); } ret = spa_do_crypt_abd(B_FALSE, spa, &zio->io_bookmark, BP_GET_TYPE(bp), BP_GET_DEDUP(bp), BP_SHOULD_BYTESWAP(bp), salt, iv, mac, size, data, zio->io_abd, &no_crypt); if (no_crypt) abd_copy(data, zio->io_abd, size); if (ret != 0) goto error; return; error: /* assert that the key was found unless this was speculative */ ASSERT(ret != EACCES || (zio->io_flags & ZIO_FLAG_SPECULATIVE)); /* * If there was a decryption / authentication error return EIO as * the io_error. If this was not a speculative zio, create an ereport. */ if (ret == ECKSUM) { zio->io_error = SET_ERROR(EIO); if ((zio->io_flags & ZIO_FLAG_SPECULATIVE) == 0) { spa_log_error(spa, &zio->io_bookmark); (void) zfs_ereport_post(FM_EREPORT_ZFS_AUTHENTICATION, spa, NULL, &zio->io_bookmark, zio, 0); } } else { zio->io_error = ret; } } /* * ========================================================================== * I/O parent/child relationships and pipeline interlocks * ========================================================================== */ zio_t * zio_walk_parents(zio_t *cio, zio_link_t **zl) { list_t *pl = &cio->io_parent_list; *zl = (*zl == NULL) ? list_head(pl) : list_next(pl, *zl); if (*zl == NULL) return (NULL); ASSERT((*zl)->zl_child == cio); return ((*zl)->zl_parent); } zio_t * zio_walk_children(zio_t *pio, zio_link_t **zl) { list_t *cl = &pio->io_child_list; ASSERT(MUTEX_HELD(&pio->io_lock)); *zl = (*zl == NULL) ? list_head(cl) : list_next(cl, *zl); if (*zl == NULL) return (NULL); ASSERT((*zl)->zl_parent == pio); return ((*zl)->zl_child); } zio_t * zio_unique_parent(zio_t *cio) { zio_link_t *zl = NULL; zio_t *pio = zio_walk_parents(cio, &zl); VERIFY3P(zio_walk_parents(cio, &zl), ==, NULL); return (pio); } void zio_add_child(zio_t *pio, zio_t *cio) { zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP); /* * Logical I/Os can have logical, gang, or vdev children. * Gang I/Os can have gang or vdev children. * Vdev I/Os can only have vdev children. * The following ASSERT captures all of these constraints. */ ASSERT3S(cio->io_child_type, <=, pio->io_child_type); zl->zl_parent = pio; zl->zl_child = cio; mutex_enter(&pio->io_lock); mutex_enter(&cio->io_lock); ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; list_insert_head(&pio->io_child_list, zl); list_insert_head(&cio->io_parent_list, zl); pio->io_child_count++; cio->io_parent_count++; mutex_exit(&cio->io_lock); mutex_exit(&pio->io_lock); } static void zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) { ASSERT(zl->zl_parent == pio); ASSERT(zl->zl_child == cio); mutex_enter(&pio->io_lock); mutex_enter(&cio->io_lock); list_remove(&pio->io_child_list, zl); list_remove(&cio->io_parent_list, zl); pio->io_child_count--; cio->io_parent_count--; mutex_exit(&cio->io_lock); mutex_exit(&pio->io_lock); kmem_cache_free(zio_link_cache, zl); } static boolean_t zio_wait_for_children(zio_t *zio, uint8_t childbits, enum zio_wait_type wait) { boolean_t waiting = B_FALSE; mutex_enter(&zio->io_lock); ASSERT(zio->io_stall == NULL); for (int c = 0; c < ZIO_CHILD_TYPES; c++) { if (!(ZIO_CHILD_BIT_IS_SET(childbits, c))) continue; uint64_t *countp = &zio->io_children[c][wait]; if (*countp != 0) { zio->io_stage >>= 1; ASSERT3U(zio->io_stage, !=, ZIO_STAGE_OPEN); zio->io_stall = countp; waiting = B_TRUE; break; } } mutex_exit(&zio->io_lock); return (waiting); } __attribute__((always_inline)) static inline void zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait, zio_t **next_to_executep) { uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; int *errorp = &pio->io_child_error[zio->io_child_type]; mutex_enter(&pio->io_lock); if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) *errorp = zio_worst_error(*errorp, zio->io_error); pio->io_reexecute |= zio->io_reexecute; ASSERT3U(*countp, >, 0); (*countp)--; if (*countp == 0 && pio->io_stall == countp) { zio_taskq_type_t type = pio->io_stage < ZIO_STAGE_VDEV_IO_START ? ZIO_TASKQ_ISSUE : ZIO_TASKQ_INTERRUPT; pio->io_stall = NULL; mutex_exit(&pio->io_lock); /* * If we can tell the caller to execute this parent next, do * so. Otherwise dispatch the parent zio as its own task. * * Having the caller execute the parent when possible reduces * locking on the zio taskq's, reduces context switch * overhead, and has no recursion penalty. Note that one * read from disk typically causes at least 3 zio's: a * zio_null(), the logical zio_read(), and then a physical * zio. When the physical ZIO completes, we are able to call * zio_done() on all 3 of these zio's from one invocation of * zio_execute() by returning the parent back to * zio_execute(). Since the parent isn't executed until this * thread returns back to zio_execute(), the caller should do * so promptly. * * In other cases, dispatching the parent prevents * overflowing the stack when we have deeply nested * parent-child relationships, as we do with the "mega zio" * of writes for spa_sync(), and the chain of ZIL blocks. */ if (next_to_executep != NULL && *next_to_executep == NULL) { *next_to_executep = pio; } else { zio_taskq_dispatch(pio, type, B_FALSE); } } else { mutex_exit(&pio->io_lock); } } static void zio_inherit_child_errors(zio_t *zio, enum zio_child c) { if (zio->io_child_error[c] != 0 && zio->io_error == 0) zio->io_error = zio->io_child_error[c]; } int zio_bookmark_compare(const void *x1, const void *x2) { const zio_t *z1 = x1; const zio_t *z2 = x2; if (z1->io_bookmark.zb_objset < z2->io_bookmark.zb_objset) return (-1); if (z1->io_bookmark.zb_objset > z2->io_bookmark.zb_objset) return (1); if (z1->io_bookmark.zb_object < z2->io_bookmark.zb_object) return (-1); if (z1->io_bookmark.zb_object > z2->io_bookmark.zb_object) return (1); if (z1->io_bookmark.zb_level < z2->io_bookmark.zb_level) return (-1); if (z1->io_bookmark.zb_level > z2->io_bookmark.zb_level) return (1); if (z1->io_bookmark.zb_blkid < z2->io_bookmark.zb_blkid) return (-1); if (z1->io_bookmark.zb_blkid > z2->io_bookmark.zb_blkid) return (1); if (z1 < z2) return (-1); if (z1 > z2) return (1); return (0); } /* * ========================================================================== * Create the various types of I/O (read, write, free, etc) * ========================================================================== */ static zio_t * zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, abd_t *data, uint64_t lsize, uint64_t psize, zio_done_func_t *done, void *private, zio_type_t type, zio_priority_t priority, enum zio_flag flags, vdev_t *vd, uint64_t offset, const zbookmark_phys_t *zb, enum zio_stage stage, enum zio_stage pipeline) { zio_t *zio; IMPLY(type != ZIO_TYPE_TRIM, psize <= SPA_MAXBLOCKSIZE); ASSERT(P2PHASE(psize, SPA_MINBLOCKSIZE) == 0); ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); ASSERT(vd || stage == ZIO_STAGE_OPEN); IMPLY(lsize != psize, (flags & ZIO_FLAG_RAW_COMPRESS) != 0); zio = kmem_cache_alloc(zio_cache, KM_SLEEP); bzero(zio, sizeof (zio_t)); mutex_init(&zio->io_lock, NULL, MUTEX_NOLOCKDEP, NULL); cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); list_create(&zio->io_parent_list, sizeof (zio_link_t), offsetof(zio_link_t, zl_parent_node)); list_create(&zio->io_child_list, sizeof (zio_link_t), offsetof(zio_link_t, zl_child_node)); metaslab_trace_init(&zio->io_alloc_list); if (vd != NULL) zio->io_child_type = ZIO_CHILD_VDEV; else if (flags & ZIO_FLAG_GANG_CHILD) zio->io_child_type = ZIO_CHILD_GANG; else if (flags & ZIO_FLAG_DDT_CHILD) zio->io_child_type = ZIO_CHILD_DDT; else zio->io_child_type = ZIO_CHILD_LOGICAL; if (bp != NULL) { zio->io_bp = (blkptr_t *)bp; zio->io_bp_copy = *bp; zio->io_bp_orig = *bp; if (type != ZIO_TYPE_WRITE || zio->io_child_type == ZIO_CHILD_DDT) zio->io_bp = &zio->io_bp_copy; /* so caller can free */ if (zio->io_child_type == ZIO_CHILD_LOGICAL) zio->io_logical = zio; if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) pipeline |= ZIO_GANG_STAGES; } zio->io_spa = spa; zio->io_txg = txg; zio->io_done = done; zio->io_private = private; zio->io_type = type; zio->io_priority = priority; zio->io_vd = vd; zio->io_offset = offset; zio->io_orig_abd = zio->io_abd = data; zio->io_orig_size = zio->io_size = psize; zio->io_lsize = lsize; zio->io_orig_flags = zio->io_flags = flags; zio->io_orig_stage = zio->io_stage = stage; zio->io_orig_pipeline = zio->io_pipeline = pipeline; zio->io_pipeline_trace = ZIO_STAGE_OPEN; zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); if (zb != NULL) zio->io_bookmark = *zb; if (pio != NULL) { if (zio->io_metaslab_class == NULL) zio->io_metaslab_class = pio->io_metaslab_class; if (zio->io_logical == NULL) zio->io_logical = pio->io_logical; if (zio->io_child_type == ZIO_CHILD_GANG) zio->io_gang_leader = pio->io_gang_leader; zio_add_child(pio, zio); } taskq_init_ent(&zio->io_tqent); return (zio); } static void zio_destroy(zio_t *zio) { metaslab_trace_fini(&zio->io_alloc_list); list_destroy(&zio->io_parent_list); list_destroy(&zio->io_child_list); mutex_destroy(&zio->io_lock); cv_destroy(&zio->io_cv); kmem_cache_free(zio_cache, zio); } zio_t * zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; zio = zio_create(pio, spa, 0, NULL, NULL, 0, 0, done, private, ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); return (zio); } zio_t * zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) { return (zio_null(NULL, spa, NULL, done, private, flags)); } static int zfs_blkptr_verify_log(spa_t *spa, const blkptr_t *bp, enum blk_verify_flag blk_verify, const char *fmt, ...) { va_list adx; char buf[256]; va_start(adx, fmt); (void) vsnprintf(buf, sizeof (buf), fmt, adx); va_end(adx); switch (blk_verify) { case BLK_VERIFY_HALT: dprintf_bp(bp, "blkptr at %p dprintf_bp():", bp); zfs_panic_recover("%s: %s", spa_name(spa), buf); break; case BLK_VERIFY_LOG: zfs_dbgmsg("%s: %s", spa_name(spa), buf); break; case BLK_VERIFY_ONLY: break; } return (1); } /* * Verify the block pointer fields contain reasonable values. This means * it only contains known object types, checksum/compression identifiers, * block sizes within the maximum allowed limits, valid DVAs, etc. * * If everything checks out B_TRUE is returned. The zfs_blkptr_verify * argument controls the behavior when an invalid field is detected. * * Modes for zfs_blkptr_verify: * 1) BLK_VERIFY_ONLY (evaluate the block) * 2) BLK_VERIFY_LOG (evaluate the block and log problems) * 3) BLK_VERIFY_HALT (call zfs_panic_recover on error) */ boolean_t zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp, boolean_t config_held, enum blk_verify_flag blk_verify) { int errors = 0; if (!DMU_OT_IS_VALID(BP_GET_TYPE(bp))) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p has invalid TYPE %llu", bp, (longlong_t)BP_GET_TYPE(bp)); } if (BP_GET_CHECKSUM(bp) >= ZIO_CHECKSUM_FUNCTIONS || BP_GET_CHECKSUM(bp) <= ZIO_CHECKSUM_ON) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p has invalid CHECKSUM %llu", bp, (longlong_t)BP_GET_CHECKSUM(bp)); } if (BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_FUNCTIONS || BP_GET_COMPRESS(bp) <= ZIO_COMPRESS_ON) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p has invalid COMPRESS %llu", bp, (longlong_t)BP_GET_COMPRESS(bp)); } if (BP_GET_LSIZE(bp) > SPA_MAXBLOCKSIZE) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p has invalid LSIZE %llu", bp, (longlong_t)BP_GET_LSIZE(bp)); } if (BP_GET_PSIZE(bp) > SPA_MAXBLOCKSIZE) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p has invalid PSIZE %llu", bp, (longlong_t)BP_GET_PSIZE(bp)); } if (BP_IS_EMBEDDED(bp)) { if (BPE_GET_ETYPE(bp) >= NUM_BP_EMBEDDED_TYPES) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p has invalid ETYPE %llu", bp, (longlong_t)BPE_GET_ETYPE(bp)); } } /* * Do not verify individual DVAs if the config is not trusted. This * will be done once the zio is executed in vdev_mirror_map_alloc. */ if (!spa->spa_trust_config) return (B_TRUE); if (!config_held) spa_config_enter(spa, SCL_VDEV, bp, RW_READER); else ASSERT(spa_config_held(spa, SCL_VDEV, RW_WRITER)); /* * Pool-specific checks. * * Note: it would be nice to verify that the blk_birth and * BP_PHYSICAL_BIRTH() are not too large. However, spa_freeze() * allows the birth time of log blocks (and dmu_sync()-ed blocks * that are in the log) to be arbitrarily large. */ for (int i = 0; i < BP_GET_NDVAS(bp); i++) { const dva_t *dva = &bp->blk_dva[i]; uint64_t vdevid = DVA_GET_VDEV(dva); if (vdevid >= spa->spa_root_vdev->vdev_children) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p DVA %u has invalid VDEV %llu", bp, i, (longlong_t)vdevid); continue; } vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid]; if (vd == NULL) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p DVA %u has invalid VDEV %llu", bp, i, (longlong_t)vdevid); continue; } if (vd->vdev_ops == &vdev_hole_ops) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p DVA %u has hole VDEV %llu", bp, i, (longlong_t)vdevid); continue; } if (vd->vdev_ops == &vdev_missing_ops) { /* * "missing" vdevs are valid during import, but we * don't have their detailed info (e.g. asize), so * we can't perform any more checks on them. */ continue; } uint64_t offset = DVA_GET_OFFSET(dva); uint64_t asize = DVA_GET_ASIZE(dva); if (DVA_GET_GANG(dva)) asize = vdev_gang_header_asize(vd); if (offset + asize > vd->vdev_asize) { errors += zfs_blkptr_verify_log(spa, bp, blk_verify, "blkptr at %p DVA %u has invalid OFFSET %llu", bp, i, (longlong_t)offset); } } if (errors > 0) dprintf_bp(bp, "blkptr at %p dprintf_bp():", bp); if (!config_held) spa_config_exit(spa, SCL_VDEV, bp); return (errors == 0); } boolean_t zfs_dva_valid(spa_t *spa, const dva_t *dva, const blkptr_t *bp) { uint64_t vdevid = DVA_GET_VDEV(dva); if (vdevid >= spa->spa_root_vdev->vdev_children) return (B_FALSE); vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid]; if (vd == NULL) return (B_FALSE); if (vd->vdev_ops == &vdev_hole_ops) return (B_FALSE); if (vd->vdev_ops == &vdev_missing_ops) { return (B_FALSE); } uint64_t offset = DVA_GET_OFFSET(dva); uint64_t asize = DVA_GET_ASIZE(dva); if (DVA_GET_GANG(dva)) asize = vdev_gang_header_asize(vd); if (offset + asize > vd->vdev_asize) return (B_FALSE); return (B_TRUE); } zio_t * zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, abd_t *data, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) { zio_t *zio; - (void) zfs_blkptr_verify(spa, bp, flags & ZIO_FLAG_CONFIG_WRITER, - BLK_VERIFY_HALT); - zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, data, size, size, done, private, ZIO_TYPE_READ, priority, flags, NULL, 0, zb, ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); return (zio); } zio_t * zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, abd_t *data, uint64_t lsize, uint64_t psize, const zio_prop_t *zp, zio_done_func_t *ready, zio_done_func_t *children_ready, zio_done_func_t *physdone, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) { zio_t *zio; ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && zp->zp_compress >= ZIO_COMPRESS_OFF && zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && DMU_OT_IS_VALID(zp->zp_type) && zp->zp_level < 32 && zp->zp_copies > 0 && zp->zp_copies <= spa_max_replication(spa)); zio = zio_create(pio, spa, txg, bp, data, lsize, psize, done, private, ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); zio->io_ready = ready; zio->io_children_ready = children_ready; zio->io_physdone = physdone; zio->io_prop = *zp; /* * Data can be NULL if we are going to call zio_write_override() to * provide the already-allocated BP. But we may need the data to * verify a dedup hit (if requested). In this case, don't try to * dedup (just take the already-allocated BP verbatim). Encrypted * dedup blocks need data as well so we also disable dedup in this * case. */ if (data == NULL && (zio->io_prop.zp_dedup_verify || zio->io_prop.zp_encrypt)) { zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE; } return (zio); } zio_t * zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, abd_t *data, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb) { zio_t *zio; zio = zio_create(pio, spa, txg, bp, data, size, size, done, private, ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_IO_REWRITE, NULL, 0, zb, ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); return (zio); } void zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(zio->io_stage == ZIO_STAGE_OPEN); ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); /* * We must reset the io_prop to match the values that existed * when the bp was first written by dmu_sync() keeping in mind * that nopwrite and dedup are mutually exclusive. */ zio->io_prop.zp_dedup = nopwrite ? B_FALSE : zio->io_prop.zp_dedup; zio->io_prop.zp_nopwrite = nopwrite; zio->io_prop.zp_copies = copies; zio->io_bp_override = bp; } void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) { (void) zfs_blkptr_verify(spa, bp, B_FALSE, BLK_VERIFY_HALT); /* * The check for EMBEDDED is a performance optimization. We * process the free here (by ignoring it) rather than * putting it on the list and then processing it in zio_free_sync(). */ if (BP_IS_EMBEDDED(bp)) return; metaslab_check_free(spa, bp); /* * Frees that are for the currently-syncing txg, are not going to be * deferred, and which will not need to do a read (i.e. not GANG or * DEDUP), can be processed immediately. Otherwise, put them on the * in-memory list for later processing. * * Note that we only defer frees after zfs_sync_pass_deferred_free * when the log space map feature is disabled. [see relevant comment * in spa_sync_iterate_to_convergence()] */ if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp) || txg != spa->spa_syncing_txg || (spa_sync_pass(spa) >= zfs_sync_pass_deferred_free && !spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))) { bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); } else { VERIFY3P(zio_free_sync(NULL, spa, txg, bp, 0), ==, NULL); } } /* * To improve performance, this function may return NULL if we were able * to do the free immediately. This avoids the cost of creating a zio * (and linking it to the parent, etc). */ zio_t * zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, enum zio_flag flags) { ASSERT(!BP_IS_HOLE(bp)); ASSERT(spa_syncing_txg(spa) == txg); if (BP_IS_EMBEDDED(bp)) return (NULL); metaslab_check_free(spa, bp); arc_freed(spa, bp); dsl_scan_freed(spa, bp); if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp)) { /* * GANG and DEDUP blocks can induce a read (for the gang block * header, or the DDT), so issue them asynchronously so that * this thread is not tied up. */ enum zio_stage stage = ZIO_FREE_PIPELINE | ZIO_STAGE_ISSUE_ASYNC; return (zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), BP_GET_PSIZE(bp), NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, stage)); } else { metaslab_free(spa, bp, txg, B_FALSE); return (NULL); } } zio_t * zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; (void) zfs_blkptr_verify(spa, bp, flags & ZIO_FLAG_CONFIG_WRITER, BLK_VERIFY_HALT); if (BP_IS_EMBEDDED(bp)) return (zio_null(pio, spa, NULL, NULL, NULL, 0)); /* * A claim is an allocation of a specific block. Claims are needed * to support immediate writes in the intent log. The issue is that * immediate writes contain committed data, but in a txg that was * *not* committed. Upon opening the pool after an unclean shutdown, * the intent log claims all blocks that contain immediate write data * so that the SPA knows they're in use. * * All claims *must* be resolved in the first txg -- before the SPA * starts allocating blocks -- so that nothing is allocated twice. * If txg == 0 we just verify that the block is claimable. */ ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_min_claim_txg(spa)); ASSERT(txg == spa_min_claim_txg(spa) || txg == 0); ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(8) */ zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), BP_GET_PSIZE(bp), done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); ASSERT0(zio->io_queued_timestamp); return (zio); } zio_t * zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; int c; if (vd->vdev_children == 0) { zio = zio_create(pio, spa, 0, NULL, NULL, 0, 0, done, private, ZIO_TYPE_IOCTL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); zio->io_cmd = cmd; } else { zio = zio_null(pio, spa, NULL, NULL, NULL, flags); for (c = 0; c < vd->vdev_children; c++) zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, done, private, flags)); } return (zio); } zio_t * zio_trim(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, enum trim_flag trim_flags) { zio_t *zio; ASSERT0(vd->vdev_children); ASSERT0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); ASSERT0(P2PHASE(size, 1ULL << vd->vdev_ashift)); ASSERT3U(size, !=, 0); zio = zio_create(pio, vd->vdev_spa, 0, NULL, NULL, size, size, done, private, ZIO_TYPE_TRIM, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_TRIM_PIPELINE); zio->io_trim_flags = trim_flags; return (zio); } zio_t * zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, abd_t *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels) { zio_t *zio; ASSERT(vd->vdev_children == 0); ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); ASSERT3U(offset + size, <=, vd->vdev_psize); zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, size, done, private, ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); zio->io_prop.zp_checksum = checksum; return (zio); } zio_t * zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, abd_t *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels) { zio_t *zio; ASSERT(vd->vdev_children == 0); ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); ASSERT3U(offset + size, <=, vd->vdev_psize); zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, size, done, private, ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); zio->io_prop.zp_checksum = checksum; if (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_EMBEDDED) { /* * zec checksums are necessarily destructive -- they modify * the end of the write buffer to hold the verifier/checksum. * Therefore, we must make a local copy in case the data is * being written to multiple places in parallel. */ abd_t *wbuf = abd_alloc_sametype(data, size); abd_copy(wbuf, data, size); zio_push_transform(zio, wbuf, size, size, NULL); } return (zio); } /* * Create a child I/O to do some work for us. */ zio_t * zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, abd_t *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private) { enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; zio_t *zio; /* * vdev child I/Os do not propagate their error to the parent. * Therefore, for correct operation the caller *must* check for * and handle the error in the child i/o's done callback. * The only exceptions are i/os that we don't care about * (OPTIONAL or REPAIR). */ ASSERT((flags & ZIO_FLAG_OPTIONAL) || (flags & ZIO_FLAG_IO_REPAIR) || done != NULL); if (type == ZIO_TYPE_READ && bp != NULL) { /* * If we have the bp, then the child should perform the * checksum and the parent need not. This pushes error * detection as close to the leaves as possible and * eliminates redundant checksums in the interior nodes. */ pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; } if (vd->vdev_ops->vdev_op_leaf) { ASSERT0(vd->vdev_children); offset += VDEV_LABEL_START_SIZE; } flags |= ZIO_VDEV_CHILD_FLAGS(pio); /* * If we've decided to do a repair, the write is not speculative -- * even if the original read was. */ if (flags & ZIO_FLAG_IO_REPAIR) flags &= ~ZIO_FLAG_SPECULATIVE; /* * If we're creating a child I/O that is not associated with a * top-level vdev, then the child zio is not an allocating I/O. * If this is a retried I/O then we ignore it since we will * have already processed the original allocating I/O. */ if (flags & ZIO_FLAG_IO_ALLOCATING && (vd != vd->vdev_top || (flags & ZIO_FLAG_IO_RETRY))) { ASSERT(pio->io_metaslab_class != NULL); ASSERT(pio->io_metaslab_class->mc_alloc_throttle_enabled); ASSERT(type == ZIO_TYPE_WRITE); ASSERT(priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(flags & ZIO_FLAG_IO_REPAIR)); ASSERT(!(pio->io_flags & ZIO_FLAG_IO_REWRITE) || pio->io_child_type == ZIO_CHILD_GANG); flags &= ~ZIO_FLAG_IO_ALLOCATING; } zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, size, done, private, type, priority, flags, vd, offset, &pio->io_bookmark, ZIO_STAGE_VDEV_IO_START >> 1, pipeline); ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV); zio->io_physdone = pio->io_physdone; if (vd->vdev_ops->vdev_op_leaf && zio->io_logical != NULL) zio->io_logical->io_phys_children++; return (zio); } zio_t * zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, abd_t *data, uint64_t size, zio_type_t type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private) { zio_t *zio; ASSERT(vd->vdev_ops->vdev_op_leaf); zio = zio_create(NULL, vd->vdev_spa, 0, NULL, data, size, size, done, private, type, priority, flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_DELEGATED, vd, offset, NULL, ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); return (zio); } void zio_flush(zio_t *zio, vdev_t *vd) { zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); } void zio_shrink(zio_t *zio, uint64_t size) { ASSERT3P(zio->io_executor, ==, NULL); ASSERT3U(zio->io_orig_size, ==, zio->io_size); ASSERT3U(size, <=, zio->io_size); /* * We don't shrink for raidz because of problems with the * reconstruction when reading back less than the block size. * Note, BP_IS_RAIDZ() assumes no compression. */ ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); if (!BP_IS_RAIDZ(zio->io_bp)) { /* we are not doing a raw write */ ASSERT3U(zio->io_size, ==, zio->io_lsize); zio->io_orig_size = zio->io_size = zio->io_lsize = size; } } /* * ========================================================================== * Prepare to read and write logical blocks * ========================================================================== */ static zio_t * zio_read_bp_init(zio_t *zio) { blkptr_t *bp = zio->io_bp; uint64_t psize = BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp); ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy); if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && zio->io_child_type == ZIO_CHILD_LOGICAL && !(zio->io_flags & ZIO_FLAG_RAW_COMPRESS)) { zio_push_transform(zio, abd_alloc_sametype(zio->io_abd, psize), psize, psize, zio_decompress); } if (((BP_IS_PROTECTED(bp) && !(zio->io_flags & ZIO_FLAG_RAW_ENCRYPT)) || BP_HAS_INDIRECT_MAC_CKSUM(bp)) && zio->io_child_type == ZIO_CHILD_LOGICAL) { zio_push_transform(zio, abd_alloc_sametype(zio->io_abd, psize), psize, psize, zio_decrypt); } if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) { int psize = BPE_GET_PSIZE(bp); void *data = abd_borrow_buf(zio->io_abd, psize); zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; decode_embedded_bp_compressed(bp, data); abd_return_buf_copy(zio->io_abd, data, psize); } else { ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy); } if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) zio->io_flags |= ZIO_FLAG_DONT_CACHE; if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) zio->io_flags |= ZIO_FLAG_DONT_CACHE; if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) zio->io_pipeline = ZIO_DDT_READ_PIPELINE; return (zio); } static zio_t * zio_write_bp_init(zio_t *zio) { if (!IO_IS_ALLOCATING(zio)) return (zio); ASSERT(zio->io_child_type != ZIO_CHILD_DDT); if (zio->io_bp_override) { blkptr_t *bp = zio->io_bp; zio_prop_t *zp = &zio->io_prop; ASSERT(bp->blk_birth != zio->io_txg); ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); *bp = *zio->io_bp_override; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (BP_IS_EMBEDDED(bp)) return (zio); /* * If we've been overridden and nopwrite is set then * set the flag accordingly to indicate that a nopwrite * has already occurred. */ if (!BP_IS_HOLE(bp) && zp->zp_nopwrite) { ASSERT(!zp->zp_dedup); ASSERT3U(BP_GET_CHECKSUM(bp), ==, zp->zp_checksum); zio->io_flags |= ZIO_FLAG_NOPWRITE; return (zio); } ASSERT(!zp->zp_nopwrite); if (BP_IS_HOLE(bp) || !zp->zp_dedup) return (zio); ASSERT((zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP) || zp->zp_dedup_verify); if (BP_GET_CHECKSUM(bp) == zp->zp_checksum && !zp->zp_encrypt) { BP_SET_DEDUP(bp, 1); zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; return (zio); } /* * We were unable to handle this as an override bp, treat * it as a regular write I/O. */ zio->io_bp_override = NULL; *bp = zio->io_bp_orig; zio->io_pipeline = zio->io_orig_pipeline; } return (zio); } static zio_t * zio_write_compress(zio_t *zio) { spa_t *spa = zio->io_spa; zio_prop_t *zp = &zio->io_prop; enum zio_compress compress = zp->zp_compress; blkptr_t *bp = zio->io_bp; uint64_t lsize = zio->io_lsize; uint64_t psize = zio->io_size; int pass = 1; /* * If our children haven't all reached the ready stage, * wait for them and then repeat this pipeline stage. */ if (zio_wait_for_children(zio, ZIO_CHILD_LOGICAL_BIT | ZIO_CHILD_GANG_BIT, ZIO_WAIT_READY)) { return (NULL); } if (!IO_IS_ALLOCATING(zio)) return (zio); if (zio->io_children_ready != NULL) { /* * Now that all our children are ready, run the callback * associated with this zio in case it wants to modify the * data to be written. */ ASSERT3U(zp->zp_level, >, 0); zio->io_children_ready(zio); } ASSERT(zio->io_child_type != ZIO_CHILD_DDT); ASSERT(zio->io_bp_override == NULL); if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg) { /* * We're rewriting an existing block, which means we're * working on behalf of spa_sync(). For spa_sync() to * converge, it must eventually be the case that we don't * have to allocate new blocks. But compression changes * the blocksize, which forces a reallocate, and makes * convergence take longer. Therefore, after the first * few passes, stop compressing to ensure convergence. */ pass = spa_sync_pass(spa); ASSERT(zio->io_txg == spa_syncing_txg(spa)); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!BP_GET_DEDUP(bp)); if (pass >= zfs_sync_pass_dont_compress) compress = ZIO_COMPRESS_OFF; /* Make sure someone doesn't change their mind on overwrites */ ASSERT(BP_IS_EMBEDDED(bp) || MIN(zp->zp_copies + BP_IS_GANG(bp), spa_max_replication(spa)) == BP_GET_NDVAS(bp)); } /* If it's a compressed write that is not raw, compress the buffer. */ if (compress != ZIO_COMPRESS_OFF && !(zio->io_flags & ZIO_FLAG_RAW_COMPRESS)) { void *cbuf = zio_buf_alloc(lsize); psize = zio_compress_data(compress, zio->io_abd, cbuf, lsize, zp->zp_complevel); if (psize == 0 || psize >= lsize) { compress = ZIO_COMPRESS_OFF; zio_buf_free(cbuf, lsize); } else if (!zp->zp_dedup && !zp->zp_encrypt && psize <= BPE_PAYLOAD_SIZE && zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) && spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) { encode_embedded_bp_compressed(bp, cbuf, compress, lsize, psize); BPE_SET_ETYPE(bp, BP_EMBEDDED_TYPE_DATA); BP_SET_TYPE(bp, zio->io_prop.zp_type); BP_SET_LEVEL(bp, zio->io_prop.zp_level); zio_buf_free(cbuf, lsize); bp->blk_birth = zio->io_txg; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; ASSERT(spa_feature_is_active(spa, SPA_FEATURE_EMBEDDED_DATA)); return (zio); } else { /* * Round compressed size up to the minimum allocation * size of the smallest-ashift device, and zero the * tail. This ensures that the compressed size of the * BP (and thus compressratio property) are correct, * in that we charge for the padding used to fill out * the last sector. */ ASSERT3U(spa->spa_min_alloc, >=, SPA_MINBLOCKSHIFT); size_t rounded = (size_t)roundup(psize, spa->spa_min_alloc); if (rounded >= lsize) { compress = ZIO_COMPRESS_OFF; zio_buf_free(cbuf, lsize); psize = lsize; } else { abd_t *cdata = abd_get_from_buf(cbuf, lsize); abd_take_ownership_of_buf(cdata, B_TRUE); abd_zero_off(cdata, psize, rounded - psize); psize = rounded; zio_push_transform(zio, cdata, psize, lsize, NULL); } } /* * We were unable to handle this as an override bp, treat * it as a regular write I/O. */ zio->io_bp_override = NULL; *bp = zio->io_bp_orig; zio->io_pipeline = zio->io_orig_pipeline; } else if ((zio->io_flags & ZIO_FLAG_RAW_ENCRYPT) != 0 && zp->zp_type == DMU_OT_DNODE) { /* * The DMU actually relies on the zio layer's compression * to free metadnode blocks that have had all contained * dnodes freed. As a result, even when doing a raw * receive, we must check whether the block can be compressed * to a hole. */ psize = zio_compress_data(ZIO_COMPRESS_EMPTY, zio->io_abd, NULL, lsize, zp->zp_complevel); if (psize == 0 || psize >= lsize) compress = ZIO_COMPRESS_OFF; } else { ASSERT3U(psize, !=, 0); } /* * The final pass of spa_sync() must be all rewrites, but the first * few passes offer a trade-off: allocating blocks defers convergence, * but newly allocated blocks are sequential, so they can be written * to disk faster. Therefore, we allow the first few passes of * spa_sync() to allocate new blocks, but force rewrites after that. * There should only be a handful of blocks after pass 1 in any case. */ if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && pass >= zfs_sync_pass_rewrite) { VERIFY3U(psize, !=, 0); enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; zio->io_flags |= ZIO_FLAG_IO_REWRITE; } else { BP_ZERO(bp); zio->io_pipeline = ZIO_WRITE_PIPELINE; } if (psize == 0) { if (zio->io_bp_orig.blk_birth != 0 && spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { BP_SET_LSIZE(bp, lsize); BP_SET_TYPE(bp, zp->zp_type); BP_SET_LEVEL(bp, zp->zp_level); BP_SET_BIRTH(bp, zio->io_txg, 0); } zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; } else { ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); BP_SET_LSIZE(bp, lsize); BP_SET_TYPE(bp, zp->zp_type); BP_SET_LEVEL(bp, zp->zp_level); BP_SET_PSIZE(bp, psize); BP_SET_COMPRESS(bp, compress); BP_SET_CHECKSUM(bp, zp->zp_checksum); BP_SET_DEDUP(bp, zp->zp_dedup); BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); if (zp->zp_dedup) { ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); ASSERT(!zp->zp_encrypt || DMU_OT_IS_ENCRYPTED(zp->zp_type)); zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; } if (zp->zp_nopwrite) { ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); zio->io_pipeline |= ZIO_STAGE_NOP_WRITE; } } return (zio); } static zio_t * zio_free_bp_init(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio->io_child_type == ZIO_CHILD_LOGICAL) { if (BP_GET_DEDUP(bp)) zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; } ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy); return (zio); } /* * ========================================================================== * Execute the I/O pipeline * ========================================================================== */ static void zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) { spa_t *spa = zio->io_spa; zio_type_t t = zio->io_type; int flags = (cutinline ? TQ_FRONT : 0); /* * If we're a config writer or a probe, the normal issue and * interrupt threads may all be blocked waiting for the config lock. * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. */ if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) t = ZIO_TYPE_NULL; /* * A similar issue exists for the L2ARC write thread until L2ARC 2.0. */ if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) t = ZIO_TYPE_NULL; /* * If this is a high priority I/O, then use the high priority taskq if * available. */ if ((zio->io_priority == ZIO_PRIORITY_NOW || zio->io_priority == ZIO_PRIORITY_SYNC_WRITE) && spa->spa_zio_taskq[t][q + 1].stqs_count != 0) q++; ASSERT3U(q, <, ZIO_TASKQ_TYPES); /* * NB: We are assuming that the zio can only be dispatched * to a single taskq at a time. It would be a grievous error * to dispatch the zio to another taskq at the same time. */ ASSERT(taskq_empty_ent(&zio->io_tqent)); spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, flags, &zio->io_tqent); } static boolean_t zio_taskq_member(zio_t *zio, zio_taskq_type_t q) { spa_t *spa = zio->io_spa; taskq_t *tq = taskq_of_curthread(); for (zio_type_t t = 0; t < ZIO_TYPES; t++) { spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; uint_t i; for (i = 0; i < tqs->stqs_count; i++) { if (tqs->stqs_taskq[i] == tq) return (B_TRUE); } } return (B_FALSE); } static zio_t * zio_issue_async(zio_t *zio) { zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); return (NULL); } void zio_interrupt(zio_t *zio) { zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); } void zio_delay_interrupt(zio_t *zio) { /* * The timeout_generic() function isn't defined in userspace, so * rather than trying to implement the function, the zio delay * functionality has been disabled for userspace builds. */ #ifdef _KERNEL /* * If io_target_timestamp is zero, then no delay has been registered * for this IO, thus jump to the end of this function and "skip" the * delay; issuing it directly to the zio layer. */ if (zio->io_target_timestamp != 0) { hrtime_t now = gethrtime(); if (now >= zio->io_target_timestamp) { /* * This IO has already taken longer than the target * delay to complete, so we don't want to delay it * any longer; we "miss" the delay and issue it * directly to the zio layer. This is likely due to * the target latency being set to a value less than * the underlying hardware can satisfy (e.g. delay * set to 1ms, but the disks take 10ms to complete an * IO request). */ DTRACE_PROBE2(zio__delay__miss, zio_t *, zio, hrtime_t, now); zio_interrupt(zio); } else { taskqid_t tid; hrtime_t diff = zio->io_target_timestamp - now; clock_t expire_at_tick = ddi_get_lbolt() + NSEC_TO_TICK(diff); DTRACE_PROBE3(zio__delay__hit, zio_t *, zio, hrtime_t, now, hrtime_t, diff); if (NSEC_TO_TICK(diff) == 0) { /* Our delay is less than a jiffy - just spin */ zfs_sleep_until(zio->io_target_timestamp); zio_interrupt(zio); } else { /* * Use taskq_dispatch_delay() in the place of * OpenZFS's timeout_generic(). */ tid = taskq_dispatch_delay(system_taskq, (task_func_t *)zio_interrupt, zio, TQ_NOSLEEP, expire_at_tick); if (tid == TASKQID_INVALID) { /* * Couldn't allocate a task. Just * finish the zio without a delay. */ zio_interrupt(zio); } } } return; } #endif DTRACE_PROBE1(zio__delay__skip, zio_t *, zio); zio_interrupt(zio); } static void zio_deadman_impl(zio_t *pio, int ziodepth) { zio_t *cio, *cio_next; zio_link_t *zl = NULL; vdev_t *vd = pio->io_vd; if (zio_deadman_log_all || (vd != NULL && vd->vdev_ops->vdev_op_leaf)) { vdev_queue_t *vq = vd ? &vd->vdev_queue : NULL; zbookmark_phys_t *zb = &pio->io_bookmark; uint64_t delta = gethrtime() - pio->io_timestamp; uint64_t failmode = spa_get_deadman_failmode(pio->io_spa); zfs_dbgmsg("slow zio[%d]: zio=%px timestamp=%llu " "delta=%llu queued=%llu io=%llu " "path=%s last=%llu " "type=%d priority=%d flags=0x%x " "stage=0x%x pipeline=0x%x pipeline-trace=0x%x " "objset=%llu object=%llu level=%llu blkid=%llu " "offset=%llu size=%llu error=%d", ziodepth, pio, pio->io_timestamp, delta, pio->io_delta, pio->io_delay, vd ? vd->vdev_path : "NULL", vq ? vq->vq_io_complete_ts : 0, pio->io_type, pio->io_priority, pio->io_flags, pio->io_stage, pio->io_pipeline, pio->io_pipeline_trace, zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid, pio->io_offset, pio->io_size, pio->io_error); (void) zfs_ereport_post(FM_EREPORT_ZFS_DEADMAN, pio->io_spa, vd, zb, pio, 0); if (failmode == ZIO_FAILURE_MODE_CONTINUE && taskq_empty_ent(&pio->io_tqent)) { zio_interrupt(pio); } } mutex_enter(&pio->io_lock); for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) { cio_next = zio_walk_children(pio, &zl); zio_deadman_impl(cio, ziodepth + 1); } mutex_exit(&pio->io_lock); } /* * Log the critical information describing this zio and all of its children * using the zfs_dbgmsg() interface then post deadman event for the ZED. */ void zio_deadman(zio_t *pio, char *tag) { spa_t *spa = pio->io_spa; char *name = spa_name(spa); if (!zfs_deadman_enabled || spa_suspended(spa)) return; zio_deadman_impl(pio, 0); switch (spa_get_deadman_failmode(spa)) { case ZIO_FAILURE_MODE_WAIT: zfs_dbgmsg("%s waiting for hung I/O to pool '%s'", tag, name); break; case ZIO_FAILURE_MODE_CONTINUE: zfs_dbgmsg("%s restarting hung I/O for pool '%s'", tag, name); break; case ZIO_FAILURE_MODE_PANIC: fm_panic("%s determined I/O to pool '%s' is hung.", tag, name); break; } } /* * Execute the I/O pipeline until one of the following occurs: * (1) the I/O completes; (2) the pipeline stalls waiting for * dependent child I/Os; (3) the I/O issues, so we're waiting * for an I/O completion interrupt; (4) the I/O is delegated by * vdev-level caching or aggregation; (5) the I/O is deferred * due to vdev-level queueing; (6) the I/O is handed off to * another thread. In all cases, the pipeline stops whenever * there's no CPU work; it never burns a thread in cv_wait_io(). * * There's no locking on io_stage because there's no legitimate way * for multiple threads to be attempting to process the same I/O. */ static zio_pipe_stage_t *zio_pipeline[]; /* * zio_execute() is a wrapper around the static function * __zio_execute() so that we can force __zio_execute() to be * inlined. This reduces stack overhead which is important * because __zio_execute() is called recursively in several zio * code paths. zio_execute() itself cannot be inlined because * it is externally visible. */ void zio_execute(zio_t *zio) { fstrans_cookie_t cookie; cookie = spl_fstrans_mark(); __zio_execute(zio); spl_fstrans_unmark(cookie); } /* * Used to determine if in the current context the stack is sized large * enough to allow zio_execute() to be called recursively. A minimum * stack size of 16K is required to avoid needing to re-dispatch the zio. */ static boolean_t zio_execute_stack_check(zio_t *zio) { #if !defined(HAVE_LARGE_STACKS) dsl_pool_t *dp = spa_get_dsl(zio->io_spa); /* Executing in txg_sync_thread() context. */ if (dp && curthread == dp->dp_tx.tx_sync_thread) return (B_TRUE); /* Pool initialization outside of zio_taskq context. */ if (dp && spa_is_initializing(dp->dp_spa) && !zio_taskq_member(zio, ZIO_TASKQ_ISSUE) && !zio_taskq_member(zio, ZIO_TASKQ_ISSUE_HIGH)) return (B_TRUE); #endif /* HAVE_LARGE_STACKS */ return (B_FALSE); } __attribute__((always_inline)) static inline void __zio_execute(zio_t *zio) { ASSERT3U(zio->io_queued_timestamp, >, 0); while (zio->io_stage < ZIO_STAGE_DONE) { enum zio_stage pipeline = zio->io_pipeline; enum zio_stage stage = zio->io_stage; zio->io_executor = curthread; ASSERT(!MUTEX_HELD(&zio->io_lock)); ASSERT(ISP2(stage)); ASSERT(zio->io_stall == NULL); do { stage <<= 1; } while ((stage & pipeline) == 0); ASSERT(stage <= ZIO_STAGE_DONE); /* * If we are in interrupt context and this pipeline stage * will grab a config lock that is held across I/O, * or may wait for an I/O that needs an interrupt thread * to complete, issue async to avoid deadlock. * * For VDEV_IO_START, we cut in line so that the io will * be sent to disk promptly. */ if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ? zio_requeue_io_start_cut_in_line : B_FALSE; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); return; } /* * If the current context doesn't have large enough stacks * the zio must be issued asynchronously to prevent overflow. */ if (zio_execute_stack_check(zio)) { boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ? zio_requeue_io_start_cut_in_line : B_FALSE; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); return; } zio->io_stage = stage; zio->io_pipeline_trace |= zio->io_stage; /* * The zio pipeline stage returns the next zio to execute * (typically the same as this one), or NULL if we should * stop. */ zio = zio_pipeline[highbit64(stage) - 1](zio); if (zio == NULL) return; } } /* * ========================================================================== * Initiate I/O, either sync or async * ========================================================================== */ int zio_wait(zio_t *zio) { /* * Some routines, like zio_free_sync(), may return a NULL zio * to avoid the performance overhead of creating and then destroying * an unneeded zio. For the callers' simplicity, we accept a NULL * zio and ignore it. */ if (zio == NULL) return (0); long timeout = MSEC_TO_TICK(zfs_deadman_ziotime_ms); int error; ASSERT3S(zio->io_stage, ==, ZIO_STAGE_OPEN); ASSERT3P(zio->io_executor, ==, NULL); zio->io_waiter = curthread; ASSERT0(zio->io_queued_timestamp); zio->io_queued_timestamp = gethrtime(); __zio_execute(zio); mutex_enter(&zio->io_lock); while (zio->io_executor != NULL) { error = cv_timedwait_io(&zio->io_cv, &zio->io_lock, ddi_get_lbolt() + timeout); if (zfs_deadman_enabled && error == -1 && gethrtime() - zio->io_queued_timestamp > spa_deadman_ziotime(zio->io_spa)) { mutex_exit(&zio->io_lock); timeout = MSEC_TO_TICK(zfs_deadman_checktime_ms); zio_deadman(zio, FTAG); mutex_enter(&zio->io_lock); } } mutex_exit(&zio->io_lock); error = zio->io_error; zio_destroy(zio); return (error); } void zio_nowait(zio_t *zio) { /* * See comment in zio_wait(). */ if (zio == NULL) return; ASSERT3P(zio->io_executor, ==, NULL); if (zio->io_child_type == ZIO_CHILD_LOGICAL && zio_unique_parent(zio) == NULL) { zio_t *pio; /* * This is a logical async I/O with no parent to wait for it. * We add it to the spa_async_root_zio "Godfather" I/O which * will ensure they complete prior to unloading the pool. */ spa_t *spa = zio->io_spa; pio = spa->spa_async_zio_root[CPU_SEQID_UNSTABLE]; zio_add_child(pio, zio); } ASSERT0(zio->io_queued_timestamp); zio->io_queued_timestamp = gethrtime(); __zio_execute(zio); } /* * ========================================================================== * Reexecute, cancel, or suspend/resume failed I/O * ========================================================================== */ static void zio_reexecute(zio_t *pio) { zio_t *cio, *cio_next; ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); ASSERT(pio->io_gang_leader == NULL); ASSERT(pio->io_gang_tree == NULL); pio->io_flags = pio->io_orig_flags; pio->io_stage = pio->io_orig_stage; pio->io_pipeline = pio->io_orig_pipeline; pio->io_reexecute = 0; pio->io_flags |= ZIO_FLAG_REEXECUTED; pio->io_pipeline_trace = 0; pio->io_error = 0; for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_state[w] = 0; for (int c = 0; c < ZIO_CHILD_TYPES; c++) pio->io_child_error[c] = 0; if (IO_IS_ALLOCATING(pio)) BP_ZERO(pio->io_bp); /* * As we reexecute pio's children, new children could be created. * New children go to the head of pio's io_child_list, however, * so we will (correctly) not reexecute them. The key is that * the remainder of pio's io_child_list, from 'cio_next' onward, * cannot be affected by any side effects of reexecuting 'cio'. */ zio_link_t *zl = NULL; mutex_enter(&pio->io_lock); for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) { cio_next = zio_walk_children(pio, &zl); for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_children[cio->io_child_type][w]++; mutex_exit(&pio->io_lock); zio_reexecute(cio); mutex_enter(&pio->io_lock); } mutex_exit(&pio->io_lock); /* * Now that all children have been reexecuted, execute the parent. * We don't reexecute "The Godfather" I/O here as it's the * responsibility of the caller to wait on it. */ if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) { pio->io_queued_timestamp = gethrtime(); __zio_execute(pio); } } void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t reason) { if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) fm_panic("Pool '%s' has encountered an uncorrectable I/O " "failure and the failure mode property for this pool " "is set to panic.", spa_name(spa)); cmn_err(CE_WARN, "Pool '%s' has encountered an uncorrectable I/O " "failure and has been suspended.\n", spa_name(spa)); (void) zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, NULL, 0); mutex_enter(&spa->spa_suspend_lock); if (spa->spa_suspend_zio_root == NULL) spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); spa->spa_suspended = reason; if (zio != NULL) { ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); ASSERT(zio != spa->spa_suspend_zio_root); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(zio_unique_parent(zio) == NULL); ASSERT(zio->io_stage == ZIO_STAGE_DONE); zio_add_child(spa->spa_suspend_zio_root, zio); } mutex_exit(&spa->spa_suspend_lock); } int zio_resume(spa_t *spa) { zio_t *pio; /* * Reexecute all previously suspended i/o. */ mutex_enter(&spa->spa_suspend_lock); spa->spa_suspended = ZIO_SUSPEND_NONE; cv_broadcast(&spa->spa_suspend_cv); pio = spa->spa_suspend_zio_root; spa->spa_suspend_zio_root = NULL; mutex_exit(&spa->spa_suspend_lock); if (pio == NULL) return (0); zio_reexecute(pio); return (zio_wait(pio)); } void zio_resume_wait(spa_t *spa) { mutex_enter(&spa->spa_suspend_lock); while (spa_suspended(spa)) cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); mutex_exit(&spa->spa_suspend_lock); } /* * ========================================================================== * Gang blocks. * * A gang block is a collection of small blocks that looks to the DMU * like one large block. When zio_dva_allocate() cannot find a block * of the requested size, due to either severe fragmentation or the pool * being nearly full, it calls zio_write_gang_block() to construct the * block from smaller fragments. * * A gang block consists of a gang header (zio_gbh_phys_t) and up to * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like * an indirect block: it's an array of block pointers. It consumes * only one sector and hence is allocatable regardless of fragmentation. * The gang header's bps point to its gang members, which hold the data. * * Gang blocks are self-checksumming, using the bp's * as the verifier to ensure uniqueness of the SHA256 checksum. * Critically, the gang block bp's blk_cksum is the checksum of the data, * not the gang header. This ensures that data block signatures (needed for * deduplication) are independent of how the block is physically stored. * * Gang blocks can be nested: a gang member may itself be a gang block. * Thus every gang block is a tree in which root and all interior nodes are * gang headers, and the leaves are normal blocks that contain user data. * The root of the gang tree is called the gang leader. * * To perform any operation (read, rewrite, free, claim) on a gang block, * zio_gang_assemble() first assembles the gang tree (minus data leaves) * in the io_gang_tree field of the original logical i/o by recursively * reading the gang leader and all gang headers below it. This yields * an in-core tree containing the contents of every gang header and the * bps for every constituent of the gang block. * * With the gang tree now assembled, zio_gang_issue() just walks the gang tree * and invokes a callback on each bp. To free a gang block, zio_gang_issue() * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). * zio_read_gang() is a wrapper around zio_read() that omits reading gang * headers, since we already have those in io_gang_tree. zio_rewrite_gang() * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() * of the gang header plus zio_checksum_compute() of the data to update the * gang header's blk_cksum as described above. * * The two-phase assemble/issue model solves the problem of partial failure -- * what if you'd freed part of a gang block but then couldn't read the * gang header for another part? Assembling the entire gang tree first * ensures that all the necessary gang header I/O has succeeded before * starting the actual work of free, claim, or write. Once the gang tree * is assembled, free and claim are in-memory operations that cannot fail. * * In the event that a gang write fails, zio_dva_unallocate() walks the * gang tree to immediately free (i.e. insert back into the space map) * everything we've allocated. This ensures that we don't get ENOSPC * errors during repeated suspend/resume cycles due to a flaky device. * * Gang rewrites only happen during sync-to-convergence. If we can't assemble * the gang tree, we won't modify the block, so we can safely defer the free * (knowing that the block is still intact). If we *can* assemble the gang * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free * each constituent bp and we can allocate a new block on the next sync pass. * * In all cases, the gang tree allows complete recovery from partial failure. * ========================================================================== */ static void zio_gang_issue_func_done(zio_t *zio) { abd_free(zio->io_abd); } static zio_t * zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { if (gn != NULL) return (pio); return (zio_read(pio, pio->io_spa, bp, abd_get_offset(data, offset), BP_GET_PSIZE(bp), zio_gang_issue_func_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark)); } static zio_t * zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { zio_t *zio; if (gn != NULL) { abd_t *gbh_abd = abd_get_from_buf(gn->gn_gbh, SPA_GANGBLOCKSIZE); zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, gbh_abd, SPA_GANGBLOCKSIZE, zio_gang_issue_func_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); /* * As we rewrite each gang header, the pipeline will compute * a new gang block header checksum for it; but no one will * compute a new data checksum, so we do that here. The one * exception is the gang leader: the pipeline already computed * its data checksum because that stage precedes gang assembly. * (Presently, nothing actually uses interior data checksums; * this is just good hygiene.) */ if (gn != pio->io_gang_leader->io_gang_tree) { abd_t *buf = abd_get_offset(data, offset); zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), buf, BP_GET_PSIZE(bp)); abd_free(buf); } /* * If we are here to damage data for testing purposes, * leave the GBH alone so that we can detect the damage. */ if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; } else { zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, abd_get_offset(data, offset), BP_GET_PSIZE(bp), zio_gang_issue_func_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); } return (zio); } /* ARGSUSED */ static zio_t * zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { zio_t *zio = zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, ZIO_GANG_CHILD_FLAGS(pio)); if (zio == NULL) { zio = zio_null(pio, pio->io_spa, NULL, NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio)); } return (zio); } /* ARGSUSED */ static zio_t * zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); } static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { NULL, zio_read_gang, zio_rewrite_gang, zio_free_gang, zio_claim_gang, NULL }; static void zio_gang_tree_assemble_done(zio_t *zio); static zio_gang_node_t * zio_gang_node_alloc(zio_gang_node_t **gnpp) { zio_gang_node_t *gn; ASSERT(*gnpp == NULL); gn = kmem_zalloc(sizeof (*gn), KM_SLEEP); gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); *gnpp = gn; return (gn); } static void zio_gang_node_free(zio_gang_node_t **gnpp) { zio_gang_node_t *gn = *gnpp; for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) ASSERT(gn->gn_child[g] == NULL); zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); kmem_free(gn, sizeof (*gn)); *gnpp = NULL; } static void zio_gang_tree_free(zio_gang_node_t **gnpp) { zio_gang_node_t *gn = *gnpp; if (gn == NULL) return; for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) zio_gang_tree_free(&gn->gn_child[g]); zio_gang_node_free(gnpp); } static void zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) { zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); abd_t *gbh_abd = abd_get_from_buf(gn->gn_gbh, SPA_GANGBLOCKSIZE); ASSERT(gio->io_gang_leader == gio); ASSERT(BP_IS_GANG(bp)); zio_nowait(zio_read(gio, gio->io_spa, bp, gbh_abd, SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); } static void zio_gang_tree_assemble_done(zio_t *zio) { zio_t *gio = zio->io_gang_leader; zio_gang_node_t *gn = zio->io_private; blkptr_t *bp = zio->io_bp; ASSERT(gio == zio_unique_parent(zio)); ASSERT(zio->io_child_count == 0); if (zio->io_error) return; /* this ABD was created from a linear buf in zio_gang_tree_assemble */ if (BP_SHOULD_BYTESWAP(bp)) byteswap_uint64_array(abd_to_buf(zio->io_abd), zio->io_size); ASSERT3P(abd_to_buf(zio->io_abd), ==, gn->gn_gbh); ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); abd_free(zio->io_abd); for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; if (!BP_IS_GANG(gbp)) continue; zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); } } static void zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, abd_t *data, uint64_t offset) { zio_t *gio = pio->io_gang_leader; zio_t *zio; ASSERT(BP_IS_GANG(bp) == !!gn); ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); /* * If you're a gang header, your data is in gn->gn_gbh. * If you're a gang member, your data is in 'data' and gn == NULL. */ zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data, offset); if (gn != NULL) { ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; if (BP_IS_HOLE(gbp)) continue; zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data, offset); offset += BP_GET_PSIZE(gbp); } } if (gn == gio->io_gang_tree) ASSERT3U(gio->io_size, ==, offset); if (zio != pio) zio_nowait(zio); } static zio_t * zio_gang_assemble(zio_t *zio) { blkptr_t *bp = zio->io_bp; ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); ASSERT(zio->io_child_type > ZIO_CHILD_GANG); zio->io_gang_leader = zio; zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); return (zio); } static zio_t * zio_gang_issue(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio_wait_for_children(zio, ZIO_CHILD_GANG_BIT, ZIO_WAIT_DONE)) { return (NULL); } ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); ASSERT(zio->io_child_type > ZIO_CHILD_GANG); if (zio->io_child_error[ZIO_CHILD_GANG] == 0) zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_abd, 0); else zio_gang_tree_free(&zio->io_gang_tree); zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; return (zio); } static void zio_write_gang_member_ready(zio_t *zio) { zio_t *pio = zio_unique_parent(zio); dva_t *cdva = zio->io_bp->blk_dva; dva_t *pdva = pio->io_bp->blk_dva; uint64_t asize; zio_t *gio __maybe_unused = zio->io_gang_leader; if (BP_IS_HOLE(zio->io_bp)) return; ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); ASSERT(zio->io_child_type == ZIO_CHILD_GANG); ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); mutex_enter(&pio->io_lock); for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { ASSERT(DVA_GET_GANG(&pdva[d])); asize = DVA_GET_ASIZE(&pdva[d]); asize += DVA_GET_ASIZE(&cdva[d]); DVA_SET_ASIZE(&pdva[d], asize); } mutex_exit(&pio->io_lock); } static void zio_write_gang_done(zio_t *zio) { /* * The io_abd field will be NULL for a zio with no data. The io_flags * will initially have the ZIO_FLAG_NODATA bit flag set, but we can't * check for it here as it is cleared in zio_ready. */ if (zio->io_abd != NULL) abd_free(zio->io_abd); } static zio_t * zio_write_gang_block(zio_t *pio, metaslab_class_t *mc) { spa_t *spa = pio->io_spa; blkptr_t *bp = pio->io_bp; zio_t *gio = pio->io_gang_leader; zio_t *zio; zio_gang_node_t *gn, **gnpp; zio_gbh_phys_t *gbh; abd_t *gbh_abd; uint64_t txg = pio->io_txg; uint64_t resid = pio->io_size; uint64_t lsize; int copies = gio->io_prop.zp_copies; int gbh_copies; zio_prop_t zp; int error; boolean_t has_data = !(pio->io_flags & ZIO_FLAG_NODATA); /* * encrypted blocks need DVA[2] free so encrypted gang headers can't * have a third copy. */ gbh_copies = MIN(copies + 1, spa_max_replication(spa)); if (gio->io_prop.zp_encrypt && gbh_copies >= SPA_DVAS_PER_BP) gbh_copies = SPA_DVAS_PER_BP - 1; int flags = METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER; if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(has_data); flags |= METASLAB_ASYNC_ALLOC; VERIFY(zfs_refcount_held(&mc->mc_allocator[pio->io_allocator]. mca_alloc_slots, pio)); /* * The logical zio has already placed a reservation for * 'copies' allocation slots but gang blocks may require * additional copies. These additional copies * (i.e. gbh_copies - copies) are guaranteed to succeed * since metaslab_class_throttle_reserve() always allows * additional reservations for gang blocks. */ VERIFY(metaslab_class_throttle_reserve(mc, gbh_copies - copies, pio->io_allocator, pio, flags)); } error = metaslab_alloc(spa, mc, SPA_GANGBLOCKSIZE, bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, flags, &pio->io_alloc_list, pio, pio->io_allocator); if (error) { if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(has_data); /* * If we failed to allocate the gang block header then * we remove any additional allocation reservations that * we placed here. The original reservation will * be removed when the logical I/O goes to the ready * stage. */ metaslab_class_throttle_unreserve(mc, gbh_copies - copies, pio->io_allocator, pio); } pio->io_error = error; return (pio); } if (pio == gio) { gnpp = &gio->io_gang_tree; } else { gnpp = pio->io_private; ASSERT(pio->io_ready == zio_write_gang_member_ready); } gn = zio_gang_node_alloc(gnpp); gbh = gn->gn_gbh; bzero(gbh, SPA_GANGBLOCKSIZE); gbh_abd = abd_get_from_buf(gbh, SPA_GANGBLOCKSIZE); /* * Create the gang header. */ zio = zio_rewrite(pio, spa, txg, bp, gbh_abd, SPA_GANGBLOCKSIZE, zio_write_gang_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); /* * Create and nowait the gang children. */ for (int g = 0; resid != 0; resid -= lsize, g++) { lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), SPA_MINBLOCKSIZE); ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); zp.zp_checksum = gio->io_prop.zp_checksum; zp.zp_compress = ZIO_COMPRESS_OFF; zp.zp_complevel = gio->io_prop.zp_complevel; zp.zp_type = DMU_OT_NONE; zp.zp_level = 0; zp.zp_copies = gio->io_prop.zp_copies; zp.zp_dedup = B_FALSE; zp.zp_dedup_verify = B_FALSE; zp.zp_nopwrite = B_FALSE; zp.zp_encrypt = gio->io_prop.zp_encrypt; zp.zp_byteorder = gio->io_prop.zp_byteorder; bzero(zp.zp_salt, ZIO_DATA_SALT_LEN); bzero(zp.zp_iv, ZIO_DATA_IV_LEN); bzero(zp.zp_mac, ZIO_DATA_MAC_LEN); zio_t *cio = zio_write(zio, spa, txg, &gbh->zg_blkptr[g], has_data ? abd_get_offset(pio->io_abd, pio->io_size - resid) : NULL, lsize, lsize, &zp, zio_write_gang_member_ready, NULL, NULL, zio_write_gang_done, &gn->gn_child[g], pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(has_data); /* * Gang children won't throttle but we should * account for their work, so reserve an allocation * slot for them here. */ VERIFY(metaslab_class_throttle_reserve(mc, zp.zp_copies, cio->io_allocator, cio, flags)); } zio_nowait(cio); } /* * Set pio's pipeline to just wait for zio to finish. */ pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; /* * We didn't allocate this bp, so make sure it doesn't get unmarked. */ pio->io_flags &= ~ZIO_FLAG_FASTWRITE; zio_nowait(zio); return (pio); } /* * The zio_nop_write stage in the pipeline determines if allocating a * new bp is necessary. The nopwrite feature can handle writes in * either syncing or open context (i.e. zil writes) and as a result is * mutually exclusive with dedup. * * By leveraging a cryptographically secure checksum, such as SHA256, we * can compare the checksums of the new data and the old to determine if * allocating a new block is required. Note that our requirements for * cryptographic strength are fairly weak: there can't be any accidental * hash collisions, but we don't need to be secure against intentional * (malicious) collisions. To trigger a nopwrite, you have to be able * to write the file to begin with, and triggering an incorrect (hash * collision) nopwrite is no worse than simply writing to the file. * That said, there are no known attacks against the checksum algorithms * used for nopwrite, assuming that the salt and the checksums * themselves remain secret. */ static zio_t * zio_nop_write(zio_t *zio) { blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; zio_prop_t *zp = &zio->io_prop; ASSERT(BP_GET_LEVEL(bp) == 0); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); ASSERT(zp->zp_nopwrite); ASSERT(!zp->zp_dedup); ASSERT(zio->io_bp_override == NULL); ASSERT(IO_IS_ALLOCATING(zio)); /* * Check to see if the original bp and the new bp have matching * characteristics (i.e. same checksum, compression algorithms, etc). * If they don't then just continue with the pipeline which will * allocate a new bp. */ if (BP_IS_HOLE(bp_orig) || !(zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_flags & ZCHECKSUM_FLAG_NOPWRITE) || BP_IS_ENCRYPTED(bp) || BP_IS_ENCRYPTED(bp_orig) || BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) || BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) || BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) || zp->zp_copies != BP_GET_NDVAS(bp_orig)) return (zio); /* * If the checksums match then reset the pipeline so that we * avoid allocating a new bp and issuing any I/O. */ if (ZIO_CHECKSUM_EQUAL(bp->blk_cksum, bp_orig->blk_cksum)) { ASSERT(zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_NOPWRITE); ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(bp_orig)); ASSERT3U(BP_GET_LSIZE(bp), ==, BP_GET_LSIZE(bp_orig)); ASSERT(zp->zp_compress != ZIO_COMPRESS_OFF); ASSERT(bcmp(&bp->blk_prop, &bp_orig->blk_prop, sizeof (uint64_t)) == 0); /* * If we're overwriting a block that is currently on an * indirect vdev, then ignore the nopwrite request and * allow a new block to be allocated on a concrete vdev. */ spa_config_enter(zio->io_spa, SCL_VDEV, FTAG, RW_READER); vdev_t *tvd = vdev_lookup_top(zio->io_spa, DVA_GET_VDEV(&bp->blk_dva[0])); if (tvd->vdev_ops == &vdev_indirect_ops) { spa_config_exit(zio->io_spa, SCL_VDEV, FTAG); return (zio); } spa_config_exit(zio->io_spa, SCL_VDEV, FTAG); *bp = *bp_orig; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; zio->io_flags |= ZIO_FLAG_NOPWRITE; } return (zio); } /* * ========================================================================== * Dedup * ========================================================================== */ static void zio_ddt_child_read_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp; zio_t *pio = zio_unique_parent(zio); mutex_enter(&pio->io_lock); ddp = ddt_phys_select(dde, bp); if (zio->io_error == 0) ddt_phys_clear(ddp); /* this ddp doesn't need repair */ if (zio->io_error == 0 && dde->dde_repair_abd == NULL) dde->dde_repair_abd = zio->io_abd; else abd_free(zio->io_abd); mutex_exit(&pio->io_lock); } static zio_t * zio_ddt_read_start(zio_t *zio) { blkptr_t *bp = zio->io_bp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_PSIZE(bp) == zio->io_size); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (zio->io_child_error[ZIO_CHILD_DDT]) { ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = ddt_repair_start(ddt, bp); ddt_phys_t *ddp = dde->dde_phys; ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); blkptr_t blk; ASSERT(zio->io_vsd == NULL); zio->io_vsd = dde; if (ddp_self == NULL) return (zio); for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) continue; ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, &blk); zio_nowait(zio_read(zio, zio->io_spa, &blk, abd_alloc_for_io(zio->io_size, B_TRUE), zio->io_size, zio_ddt_child_read_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, &zio->io_bookmark)); } return (zio); } zio_nowait(zio_read(zio, zio->io_spa, bp, zio->io_abd, zio->io_size, NULL, NULL, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); return (zio); } static zio_t * zio_ddt_read_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio_wait_for_children(zio, ZIO_CHILD_DDT_BIT, ZIO_WAIT_DONE)) { return (NULL); } ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_PSIZE(bp) == zio->io_size); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (zio->io_child_error[ZIO_CHILD_DDT]) { ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = zio->io_vsd; if (ddt == NULL) { ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); return (zio); } if (dde == NULL) { zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); return (NULL); } if (dde->dde_repair_abd != NULL) { abd_copy(zio->io_abd, dde->dde_repair_abd, zio->io_size); zio->io_child_error[ZIO_CHILD_DDT] = 0; } ddt_repair_done(ddt, dde); zio->io_vsd = NULL; } ASSERT(zio->io_vsd == NULL); return (zio); } static boolean_t zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) { spa_t *spa = zio->io_spa; boolean_t do_raw = !!(zio->io_flags & ZIO_FLAG_RAW); ASSERT(!(zio->io_bp_override && do_raw)); /* * Note: we compare the original data, not the transformed data, * because when zio->io_bp is an override bp, we will not have * pushed the I/O transforms. That's an important optimization * because otherwise we'd compress/encrypt all dmu_sync() data twice. * However, we should never get a raw, override zio so in these * cases we can compare the io_abd directly. This is useful because * it allows us to do dedup verification even if we don't have access * to the original data (for instance, if the encryption keys aren't * loaded). */ for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { zio_t *lio = dde->dde_lead_zio[p]; if (lio != NULL && do_raw) { return (lio->io_size != zio->io_size || abd_cmp(zio->io_abd, lio->io_abd) != 0); } else if (lio != NULL) { return (lio->io_orig_size != zio->io_orig_size || abd_cmp(zio->io_orig_abd, lio->io_orig_abd) != 0); } } for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { ddt_phys_t *ddp = &dde->dde_phys[p]; if (ddp->ddp_phys_birth != 0 && do_raw) { blkptr_t blk = *zio->io_bp; uint64_t psize; abd_t *tmpabd; int error; ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); psize = BP_GET_PSIZE(&blk); if (psize != zio->io_size) return (B_TRUE); ddt_exit(ddt); tmpabd = abd_alloc_for_io(psize, B_TRUE); error = zio_wait(zio_read(NULL, spa, &blk, tmpabd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_RAW, &zio->io_bookmark)); if (error == 0) { if (abd_cmp(tmpabd, zio->io_abd) != 0) error = SET_ERROR(ENOENT); } abd_free(tmpabd); ddt_enter(ddt); return (error != 0); } else if (ddp->ddp_phys_birth != 0) { arc_buf_t *abuf = NULL; arc_flags_t aflags = ARC_FLAG_WAIT; blkptr_t blk = *zio->io_bp; int error; ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); if (BP_GET_LSIZE(&blk) != zio->io_orig_size) return (B_TRUE); ddt_exit(ddt); error = arc_read(NULL, spa, &blk, arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, &zio->io_bookmark); if (error == 0) { if (abd_cmp_buf(zio->io_orig_abd, abuf->b_data, zio->io_orig_size) != 0) error = SET_ERROR(ENOENT); arc_buf_destroy(abuf, &abuf); } ddt_enter(ddt); return (error != 0); } } return (B_FALSE); } static void zio_ddt_child_write_ready(zio_t *zio) { int p = zio->io_prop.zp_copies; ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; zio_t *pio; if (zio->io_error) return; ddt_enter(ddt); ASSERT(dde->dde_lead_zio[p] == zio); ddt_phys_fill(ddp, zio->io_bp); zio_link_t *zl = NULL; while ((pio = zio_walk_parents(zio, &zl)) != NULL) ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); ddt_exit(ddt); } static void zio_ddt_child_write_done(zio_t *zio) { int p = zio->io_prop.zp_copies; ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; ddt_enter(ddt); ASSERT(ddp->ddp_refcnt == 0); ASSERT(dde->dde_lead_zio[p] == zio); dde->dde_lead_zio[p] = NULL; if (zio->io_error == 0) { zio_link_t *zl = NULL; while (zio_walk_parents(zio, &zl) != NULL) ddt_phys_addref(ddp); } else { ddt_phys_clear(ddp); } ddt_exit(ddt); } static zio_t * zio_ddt_write(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; uint64_t txg = zio->io_txg; zio_prop_t *zp = &zio->io_prop; int p = zp->zp_copies; zio_t *cio = NULL; ddt_t *ddt = ddt_select(spa, bp); ddt_entry_t *dde; ddt_phys_t *ddp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); ASSERT(!(zio->io_bp_override && (zio->io_flags & ZIO_FLAG_RAW))); ddt_enter(ddt); dde = ddt_lookup(ddt, bp, B_TRUE); ddp = &dde->dde_phys[p]; if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { /* * If we're using a weak checksum, upgrade to a strong checksum * and try again. If we're already using a strong checksum, * we can't resolve it, so just convert to an ordinary write. * (And automatically e-mail a paper to Nature?) */ if (!(zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP)) { zp->zp_checksum = spa_dedup_checksum(spa); zio_pop_transforms(zio); zio->io_stage = ZIO_STAGE_OPEN; BP_ZERO(bp); } else { zp->zp_dedup = B_FALSE; BP_SET_DEDUP(bp, B_FALSE); } ASSERT(!BP_GET_DEDUP(bp)); zio->io_pipeline = ZIO_WRITE_PIPELINE; ddt_exit(ddt); return (zio); } if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { if (ddp->ddp_phys_birth != 0) ddt_bp_fill(ddp, bp, txg); if (dde->dde_lead_zio[p] != NULL) zio_add_child(zio, dde->dde_lead_zio[p]); else ddt_phys_addref(ddp); } else if (zio->io_bp_override) { ASSERT(bp->blk_birth == txg); ASSERT(BP_EQUAL(bp, zio->io_bp_override)); ddt_phys_fill(ddp, bp); ddt_phys_addref(ddp); } else { cio = zio_write(zio, spa, txg, bp, zio->io_orig_abd, zio->io_orig_size, zio->io_orig_size, zp, zio_ddt_child_write_ready, NULL, NULL, zio_ddt_child_write_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); zio_push_transform(cio, zio->io_abd, zio->io_size, 0, NULL); dde->dde_lead_zio[p] = cio; } ddt_exit(ddt); zio_nowait(cio); return (zio); } ddt_entry_t *freedde; /* for debugging */ static zio_t * zio_ddt_free(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; ddt_t *ddt = ddt_select(spa, bp); ddt_entry_t *dde; ddt_phys_t *ddp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ddt_enter(ddt); freedde = dde = ddt_lookup(ddt, bp, B_TRUE); if (dde) { ddp = ddt_phys_select(dde, bp); if (ddp) ddt_phys_decref(ddp); } ddt_exit(ddt); return (zio); } /* * ========================================================================== * Allocate and free blocks * ========================================================================== */ static zio_t * zio_io_to_allocate(spa_t *spa, int allocator) { zio_t *zio; ASSERT(MUTEX_HELD(&spa->spa_alloc_locks[allocator])); zio = avl_first(&spa->spa_alloc_trees[allocator]); if (zio == NULL) return (NULL); ASSERT(IO_IS_ALLOCATING(zio)); /* * Try to place a reservation for this zio. If we're unable to * reserve then we throttle. */ ASSERT3U(zio->io_allocator, ==, allocator); if (!metaslab_class_throttle_reserve(zio->io_metaslab_class, zio->io_prop.zp_copies, zio->io_allocator, zio, 0)) { return (NULL); } avl_remove(&spa->spa_alloc_trees[allocator], zio); ASSERT3U(zio->io_stage, <, ZIO_STAGE_DVA_ALLOCATE); return (zio); } static zio_t * zio_dva_throttle(zio_t *zio) { spa_t *spa = zio->io_spa; zio_t *nio; metaslab_class_t *mc; /* locate an appropriate allocation class */ mc = spa_preferred_class(spa, zio->io_size, zio->io_prop.zp_type, zio->io_prop.zp_level, zio->io_prop.zp_zpl_smallblk); if (zio->io_priority == ZIO_PRIORITY_SYNC_WRITE || !mc->mc_alloc_throttle_enabled || zio->io_child_type == ZIO_CHILD_GANG || zio->io_flags & ZIO_FLAG_NODATA) { return (zio); } ASSERT(zio->io_child_type > ZIO_CHILD_GANG); ASSERT3U(zio->io_queued_timestamp, >, 0); ASSERT(zio->io_stage == ZIO_STAGE_DVA_THROTTLE); zbookmark_phys_t *bm = &zio->io_bookmark; /* * We want to try to use as many allocators as possible to help improve * performance, but we also want logically adjacent IOs to be physically * adjacent to improve sequential read performance. We chunk each object * into 2^20 block regions, and then hash based on the objset, object, * level, and region to accomplish both of these goals. */ zio->io_allocator = cityhash4(bm->zb_objset, bm->zb_object, bm->zb_level, bm->zb_blkid >> 20) % spa->spa_alloc_count; mutex_enter(&spa->spa_alloc_locks[zio->io_allocator]); ASSERT(zio->io_type == ZIO_TYPE_WRITE); zio->io_metaslab_class = mc; avl_add(&spa->spa_alloc_trees[zio->io_allocator], zio); nio = zio_io_to_allocate(spa, zio->io_allocator); mutex_exit(&spa->spa_alloc_locks[zio->io_allocator]); return (nio); } static void zio_allocate_dispatch(spa_t *spa, int allocator) { zio_t *zio; mutex_enter(&spa->spa_alloc_locks[allocator]); zio = zio_io_to_allocate(spa, allocator); mutex_exit(&spa->spa_alloc_locks[allocator]); if (zio == NULL) return; ASSERT3U(zio->io_stage, ==, ZIO_STAGE_DVA_THROTTLE); ASSERT0(zio->io_error); zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_TRUE); } static zio_t * zio_dva_allocate(zio_t *zio) { spa_t *spa = zio->io_spa; metaslab_class_t *mc; blkptr_t *bp = zio->io_bp; int error; int flags = 0; if (zio->io_gang_leader == NULL) { ASSERT(zio->io_child_type > ZIO_CHILD_GANG); zio->io_gang_leader = zio; } ASSERT(BP_IS_HOLE(bp)); ASSERT0(BP_GET_NDVAS(bp)); ASSERT3U(zio->io_prop.zp_copies, >, 0); ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); flags |= (zio->io_flags & ZIO_FLAG_FASTWRITE) ? METASLAB_FASTWRITE : 0; if (zio->io_flags & ZIO_FLAG_NODATA) flags |= METASLAB_DONT_THROTTLE; if (zio->io_flags & ZIO_FLAG_GANG_CHILD) flags |= METASLAB_GANG_CHILD; if (zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE) flags |= METASLAB_ASYNC_ALLOC; /* * if not already chosen, locate an appropriate allocation class */ mc = zio->io_metaslab_class; if (mc == NULL) { mc = spa_preferred_class(spa, zio->io_size, zio->io_prop.zp_type, zio->io_prop.zp_level, zio->io_prop.zp_zpl_smallblk); zio->io_metaslab_class = mc; } /* * Try allocating the block in the usual metaslab class. * If that's full, allocate it in the normal class. * If that's full, allocate as a gang block, * and if all are full, the allocation fails (which shouldn't happen). * * Note that we do not fall back on embedded slog (ZIL) space, to * preserve unfragmented slog space, which is critical for decent * sync write performance. If a log allocation fails, we will fall * back to spa_sync() which is abysmal for performance. */ error = metaslab_alloc(spa, mc, zio->io_size, bp, zio->io_prop.zp_copies, zio->io_txg, NULL, flags, &zio->io_alloc_list, zio, zio->io_allocator); /* * Fallback to normal class when an alloc class is full */ if (error == ENOSPC && mc != spa_normal_class(spa)) { /* * If throttling, transfer reservation over to normal class. * The io_allocator slot can remain the same even though we * are switching classes. */ if (mc->mc_alloc_throttle_enabled && (zio->io_flags & ZIO_FLAG_IO_ALLOCATING)) { metaslab_class_throttle_unreserve(mc, zio->io_prop.zp_copies, zio->io_allocator, zio); zio->io_flags &= ~ZIO_FLAG_IO_ALLOCATING; VERIFY(metaslab_class_throttle_reserve( spa_normal_class(spa), zio->io_prop.zp_copies, zio->io_allocator, zio, flags | METASLAB_MUST_RESERVE)); } zio->io_metaslab_class = mc = spa_normal_class(spa); if (zfs_flags & ZFS_DEBUG_METASLAB_ALLOC) { zfs_dbgmsg("%s: metaslab allocation failure, " "trying normal class: zio %px, size %llu, error %d", spa_name(spa), zio, zio->io_size, error); } error = metaslab_alloc(spa, mc, zio->io_size, bp, zio->io_prop.zp_copies, zio->io_txg, NULL, flags, &zio->io_alloc_list, zio, zio->io_allocator); } if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) { if (zfs_flags & ZFS_DEBUG_METASLAB_ALLOC) { zfs_dbgmsg("%s: metaslab allocation failure, " "trying ganging: zio %px, size %llu, error %d", spa_name(spa), zio, zio->io_size, error); } return (zio_write_gang_block(zio, mc)); } if (error != 0) { if (error != ENOSPC || (zfs_flags & ZFS_DEBUG_METASLAB_ALLOC)) { zfs_dbgmsg("%s: metaslab allocation failure: zio %px, " "size %llu, error %d", spa_name(spa), zio, zio->io_size, error); } zio->io_error = error; } return (zio); } static zio_t * zio_dva_free(zio_t *zio) { metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); return (zio); } static zio_t * zio_dva_claim(zio_t *zio) { int error; error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); if (error) zio->io_error = error; return (zio); } /* * Undo an allocation. This is used by zio_done() when an I/O fails * and we want to give back the block we just allocated. * This handles both normal blocks and gang blocks. */ static void zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) { ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); ASSERT(zio->io_bp_override == NULL); if (!BP_IS_HOLE(bp)) metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); if (gn != NULL) { for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { zio_dva_unallocate(zio, gn->gn_child[g], &gn->gn_gbh->zg_blkptr[g]); } } } /* * Try to allocate an intent log block. Return 0 on success, errno on failure. */ int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp, uint64_t size, boolean_t *slog) { int error = 1; zio_alloc_list_t io_alloc_list; ASSERT(txg > spa_syncing_txg(spa)); metaslab_trace_init(&io_alloc_list); /* * Block pointer fields are useful to metaslabs for stats and debugging. * Fill in the obvious ones before calling into metaslab_alloc(). */ BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); BP_SET_PSIZE(new_bp, size); BP_SET_LEVEL(new_bp, 0); /* * When allocating a zil block, we don't have information about * the final destination of the block except the objset it's part * of, so we just hash the objset ID to pick the allocator to get * some parallelism. */ int flags = METASLAB_FASTWRITE | METASLAB_ZIL; int allocator = cityhash4(0, 0, 0, os->os_dsl_dataset->ds_object) % spa->spa_alloc_count; error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1, txg, NULL, flags, &io_alloc_list, NULL, allocator); *slog = (error == 0); if (error != 0) { error = metaslab_alloc(spa, spa_embedded_log_class(spa), size, new_bp, 1, txg, NULL, flags, &io_alloc_list, NULL, allocator); } if (error != 0) { error = metaslab_alloc(spa, spa_normal_class(spa), size, new_bp, 1, txg, NULL, flags, &io_alloc_list, NULL, allocator); } metaslab_trace_fini(&io_alloc_list); if (error == 0) { BP_SET_LSIZE(new_bp, size); BP_SET_PSIZE(new_bp, size); BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); BP_SET_CHECKSUM(new_bp, spa_version(spa) >= SPA_VERSION_SLIM_ZIL ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); BP_SET_LEVEL(new_bp, 0); BP_SET_DEDUP(new_bp, 0); BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); /* * encrypted blocks will require an IV and salt. We generate * these now since we will not be rewriting the bp at * rewrite time. */ if (os->os_encrypted) { uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t salt[ZIO_DATA_SALT_LEN]; BP_SET_CRYPT(new_bp, B_TRUE); VERIFY0(spa_crypt_get_salt(spa, dmu_objset_id(os), salt)); VERIFY0(zio_crypt_generate_iv(iv)); zio_crypt_encode_params_bp(new_bp, salt, iv); } } else { zfs_dbgmsg("%s: zil block allocation failure: " "size %llu, error %d", spa_name(spa), size, error); } return (error); } /* * ========================================================================== * Read and write to physical devices * ========================================================================== */ /* * Issue an I/O to the underlying vdev. Typically the issue pipeline * stops after this stage and will resume upon I/O completion. * However, there are instances where the vdev layer may need to * continue the pipeline when an I/O was not issued. Since the I/O * that was sent to the vdev layer might be different than the one * currently active in the pipeline (see vdev_queue_io()), we explicitly * force the underlying vdev layers to call either zio_execute() or * zio_interrupt() to ensure that the pipeline continues with the correct I/O. */ static zio_t * zio_vdev_io_start(zio_t *zio) { vdev_t *vd = zio->io_vd; uint64_t align; spa_t *spa = zio->io_spa; zio->io_delay = 0; ASSERT(zio->io_error == 0); ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); if (vd == NULL) { if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) spa_config_enter(spa, SCL_ZIO, zio, RW_READER); /* * The mirror_ops handle multiple DVAs in a single BP. */ vdev_mirror_ops.vdev_op_io_start(zio); return (NULL); } ASSERT3P(zio->io_logical, !=, zio); if (zio->io_type == ZIO_TYPE_WRITE) { ASSERT(spa->spa_trust_config); /* * Note: the code can handle other kinds of writes, * but we don't expect them. */ if (zio->io_vd->vdev_removing) { ASSERT(zio->io_flags & (ZIO_FLAG_PHYSICAL | ZIO_FLAG_SELF_HEAL | ZIO_FLAG_RESILVER | ZIO_FLAG_INDUCE_DAMAGE)); } } align = 1ULL << vd->vdev_top->vdev_ashift; if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) && P2PHASE(zio->io_size, align) != 0) { /* Transform logical writes to be a full physical block size. */ uint64_t asize = P2ROUNDUP(zio->io_size, align); abd_t *abuf = abd_alloc_sametype(zio->io_abd, asize); ASSERT(vd == vd->vdev_top); if (zio->io_type == ZIO_TYPE_WRITE) { abd_copy(abuf, zio->io_abd, zio->io_size); abd_zero_off(abuf, zio->io_size, asize - zio->io_size); } zio_push_transform(zio, abuf, asize, asize, zio_subblock); } /* * If this is not a physical io, make sure that it is properly aligned * before proceeding. */ if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) { ASSERT0(P2PHASE(zio->io_offset, align)); ASSERT0(P2PHASE(zio->io_size, align)); } else { /* * For physical writes, we allow 512b aligned writes and assume * the device will perform a read-modify-write as necessary. */ ASSERT0(P2PHASE(zio->io_offset, SPA_MINBLOCKSIZE)); ASSERT0(P2PHASE(zio->io_size, SPA_MINBLOCKSIZE)); } VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); /* * If this is a repair I/O, and there's no self-healing involved -- * that is, we're just resilvering what we expect to resilver -- * then don't do the I/O unless zio's txg is actually in vd's DTL. * This prevents spurious resilvering. * * There are a few ways that we can end up creating these spurious * resilver i/os: * * 1. A resilver i/o will be issued if any DVA in the BP has a * dirty DTL. The mirror code will issue resilver writes to * each DVA, including the one(s) that are not on vdevs with dirty * DTLs. * * 2. With nested replication, which happens when we have a * "replacing" or "spare" vdev that's a child of a mirror or raidz. * For example, given mirror(replacing(A+B), C), it's likely that * only A is out of date (it's the new device). In this case, we'll * read from C, then use the data to resilver A+B -- but we don't * actually want to resilver B, just A. The top-level mirror has no * way to know this, so instead we just discard unnecessary repairs * as we work our way down the vdev tree. * * 3. ZTEST also creates mirrors of mirrors, mirrors of raidz, etc. * The same logic applies to any form of nested replication: ditto * + mirror, RAID-Z + replacing, etc. * * However, indirect vdevs point off to other vdevs which may have * DTL's, so we never bypass them. The child i/os on concrete vdevs * will be properly bypassed instead. * * Leaf DTL_PARTIAL can be empty when a legitimate write comes from * a dRAID spare vdev. For example, when a dRAID spare is first * used, its spare blocks need to be written to but the leaf vdev's * of such blocks can have empty DTL_PARTIAL. * * There seemed no clean way to allow such writes while bypassing * spurious ones. At this point, just avoid all bypassing for dRAID * for correctness. */ if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && zio->io_txg != 0 && /* not a delegated i/o */ vd->vdev_ops != &vdev_indirect_ops && vd->vdev_top->vdev_ops != &vdev_draid_ops && !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); zio_vdev_io_bypass(zio); return (zio); } /* * Select the next best leaf I/O to process. Distributed spares are * excluded since they dispatch the I/O directly to a leaf vdev after * applying the dRAID mapping. */ if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_draid_spare_ops && (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE || zio->io_type == ZIO_TYPE_TRIM)) { if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio)) return (zio); if ((zio = vdev_queue_io(zio)) == NULL) return (NULL); if (!vdev_accessible(vd, zio)) { zio->io_error = SET_ERROR(ENXIO); zio_interrupt(zio); return (NULL); } zio->io_delay = gethrtime(); } vd->vdev_ops->vdev_op_io_start(zio); return (NULL); } static zio_t * zio_vdev_io_done(zio_t *zio) { vdev_t *vd = zio->io_vd; vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; boolean_t unexpected_error = B_FALSE; if (zio_wait_for_children(zio, ZIO_CHILD_VDEV_BIT, ZIO_WAIT_DONE)) { return (NULL); } ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE || zio->io_type == ZIO_TYPE_TRIM); if (zio->io_delay) zio->io_delay = gethrtime() - zio->io_delay; if (vd != NULL && vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_draid_spare_ops) { vdev_queue_io_done(zio); if (zio->io_type == ZIO_TYPE_WRITE) vdev_cache_write(zio); if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_device_injections(vd, zio, EIO, EILSEQ); if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_label_injection(zio, EIO); if (zio->io_error && zio->io_type != ZIO_TYPE_TRIM) { if (!vdev_accessible(vd, zio)) { zio->io_error = SET_ERROR(ENXIO); } else { unexpected_error = B_TRUE; } } } ops->vdev_op_io_done(zio); if (unexpected_error) VERIFY(vdev_probe(vd, zio) == NULL); return (zio); } /* * This function is used to change the priority of an existing zio that is * currently in-flight. This is used by the arc to upgrade priority in the * event that a demand read is made for a block that is currently queued * as a scrub or async read IO. Otherwise, the high priority read request * would end up having to wait for the lower priority IO. */ void zio_change_priority(zio_t *pio, zio_priority_t priority) { zio_t *cio, *cio_next; zio_link_t *zl = NULL; ASSERT3U(priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); if (pio->io_vd != NULL && pio->io_vd->vdev_ops->vdev_op_leaf) { vdev_queue_change_io_priority(pio, priority); } else { pio->io_priority = priority; } mutex_enter(&pio->io_lock); for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) { cio_next = zio_walk_children(pio, &zl); zio_change_priority(cio, priority); } mutex_exit(&pio->io_lock); } /* * For non-raidz ZIOs, we can just copy aside the bad data read from the * disk, and use that to finish the checksum ereport later. */ static void zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, const abd_t *good_buf) { /* no processing needed */ zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); } /*ARGSUSED*/ void zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr) { void *abd = abd_alloc_sametype(zio->io_abd, zio->io_size); abd_copy(abd, zio->io_abd, zio->io_size); zcr->zcr_cbinfo = zio->io_size; zcr->zcr_cbdata = abd; zcr->zcr_finish = zio_vsd_default_cksum_finish; zcr->zcr_free = zio_abd_free; } static zio_t * zio_vdev_io_assess(zio_t *zio) { vdev_t *vd = zio->io_vd; if (zio_wait_for_children(zio, ZIO_CHILD_VDEV_BIT, ZIO_WAIT_DONE)) { return (NULL); } if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) spa_config_exit(zio->io_spa, SCL_ZIO, zio); if (zio->io_vsd != NULL) { zio->io_vsd_ops->vsd_free(zio); zio->io_vsd = NULL; } if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_fault_injection(zio, EIO); /* * If the I/O failed, determine whether we should attempt to retry it. * * On retry, we cut in line in the issue queue, since we don't want * compression/checksumming/etc. work to prevent our (cheap) IO reissue. */ if (zio->io_error && vd == NULL && !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ zio->io_error = 0; zio->io_flags |= ZIO_FLAG_IO_RETRY | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, zio_requeue_io_start_cut_in_line); return (NULL); } /* * If we got an error on a leaf device, convert it to ENXIO * if the device is not accessible at all. */ if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && !vdev_accessible(vd, zio)) zio->io_error = SET_ERROR(ENXIO); /* * If we can't write to an interior vdev (mirror or RAID-Z), * set vdev_cant_write so that we stop trying to allocate from it. */ if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && vd != NULL && !vd->vdev_ops->vdev_op_leaf) { vdev_dbgmsg(vd, "zio_vdev_io_assess(zio=%px) setting " "cant_write=TRUE due to write failure with ENXIO", zio); vd->vdev_cant_write = B_TRUE; } /* * If a cache flush returns ENOTSUP or ENOTTY, we know that no future * attempts will ever succeed. In this case we set a persistent * boolean flag so that we don't bother with it in the future. */ if ((zio->io_error == ENOTSUP || zio->io_error == ENOTTY) && zio->io_type == ZIO_TYPE_IOCTL && zio->io_cmd == DKIOCFLUSHWRITECACHE && vd != NULL) vd->vdev_nowritecache = B_TRUE; if (zio->io_error) zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (vd != NULL && vd->vdev_ops->vdev_op_leaf && zio->io_physdone != NULL) { ASSERT(!(zio->io_flags & ZIO_FLAG_DELEGATED)); ASSERT(zio->io_child_type == ZIO_CHILD_VDEV); zio->io_physdone(zio->io_logical); } return (zio); } void zio_vdev_io_reissue(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); ASSERT(zio->io_error == 0); zio->io_stage >>= 1; } void zio_vdev_io_redone(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); zio->io_stage >>= 1; } void zio_vdev_io_bypass(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); ASSERT(zio->io_error == 0); zio->io_flags |= ZIO_FLAG_IO_BYPASS; zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; } /* * ========================================================================== * Encrypt and store encryption parameters * ========================================================================== */ /* * This function is used for ZIO_STAGE_ENCRYPT. It is responsible for * managing the storage of encryption parameters and passing them to the * lower-level encryption functions. */ static zio_t * zio_encrypt(zio_t *zio) { zio_prop_t *zp = &zio->io_prop; spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; uint64_t psize = BP_GET_PSIZE(bp); uint64_t dsobj = zio->io_bookmark.zb_objset; dmu_object_type_t ot = BP_GET_TYPE(bp); void *enc_buf = NULL; abd_t *eabd = NULL; uint8_t salt[ZIO_DATA_SALT_LEN]; uint8_t iv[ZIO_DATA_IV_LEN]; uint8_t mac[ZIO_DATA_MAC_LEN]; boolean_t no_crypt = B_FALSE; /* the root zio already encrypted the data */ if (zio->io_child_type == ZIO_CHILD_GANG) return (zio); /* only ZIL blocks are re-encrypted on rewrite */ if (!IO_IS_ALLOCATING(zio) && ot != DMU_OT_INTENT_LOG) return (zio); if (!(zp->zp_encrypt || BP_IS_ENCRYPTED(bp))) { BP_SET_CRYPT(bp, B_FALSE); return (zio); } /* if we are doing raw encryption set the provided encryption params */ if (zio->io_flags & ZIO_FLAG_RAW_ENCRYPT) { ASSERT0(BP_GET_LEVEL(bp)); BP_SET_CRYPT(bp, B_TRUE); BP_SET_BYTEORDER(bp, zp->zp_byteorder); if (ot != DMU_OT_OBJSET) zio_crypt_encode_mac_bp(bp, zp->zp_mac); /* dnode blocks must be written out in the provided byteorder */ if (zp->zp_byteorder != ZFS_HOST_BYTEORDER && ot == DMU_OT_DNODE) { void *bswap_buf = zio_buf_alloc(psize); abd_t *babd = abd_get_from_buf(bswap_buf, psize); ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); abd_copy_to_buf(bswap_buf, zio->io_abd, psize); dmu_ot_byteswap[DMU_OT_BYTESWAP(ot)].ob_func(bswap_buf, psize); abd_take_ownership_of_buf(babd, B_TRUE); zio_push_transform(zio, babd, psize, psize, NULL); } if (DMU_OT_IS_ENCRYPTED(ot)) zio_crypt_encode_params_bp(bp, zp->zp_salt, zp->zp_iv); return (zio); } /* indirect blocks only maintain a cksum of the lower level MACs */ if (BP_GET_LEVEL(bp) > 0) { BP_SET_CRYPT(bp, B_TRUE); VERIFY0(zio_crypt_do_indirect_mac_checksum_abd(B_TRUE, zio->io_orig_abd, BP_GET_LSIZE(bp), BP_SHOULD_BYTESWAP(bp), mac)); zio_crypt_encode_mac_bp(bp, mac); return (zio); } /* * Objset blocks are a special case since they have 2 256-bit MACs * embedded within them. */ if (ot == DMU_OT_OBJSET) { ASSERT0(DMU_OT_IS_ENCRYPTED(ot)); ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); BP_SET_CRYPT(bp, B_TRUE); VERIFY0(spa_do_crypt_objset_mac_abd(B_TRUE, spa, dsobj, zio->io_abd, psize, BP_SHOULD_BYTESWAP(bp))); return (zio); } /* unencrypted object types are only authenticated with a MAC */ if (!DMU_OT_IS_ENCRYPTED(ot)) { BP_SET_CRYPT(bp, B_TRUE); VERIFY0(spa_do_crypt_mac_abd(B_TRUE, spa, dsobj, zio->io_abd, psize, mac)); zio_crypt_encode_mac_bp(bp, mac); return (zio); } /* * Later passes of sync-to-convergence may decide to rewrite data * in place to avoid more disk reallocations. This presents a problem * for encryption because this constitutes rewriting the new data with * the same encryption key and IV. However, this only applies to blocks * in the MOS (particularly the spacemaps) and we do not encrypt the * MOS. We assert that the zio is allocating or an intent log write * to enforce this. */ ASSERT(IO_IS_ALLOCATING(zio) || ot == DMU_OT_INTENT_LOG); ASSERT(BP_GET_LEVEL(bp) == 0 || ot == DMU_OT_INTENT_LOG); ASSERT(spa_feature_is_active(spa, SPA_FEATURE_ENCRYPTION)); ASSERT3U(psize, !=, 0); enc_buf = zio_buf_alloc(psize); eabd = abd_get_from_buf(enc_buf, psize); abd_take_ownership_of_buf(eabd, B_TRUE); /* * For an explanation of what encryption parameters are stored * where, see the block comment in zio_crypt.c. */ if (ot == DMU_OT_INTENT_LOG) { zio_crypt_decode_params_bp(bp, salt, iv); } else { BP_SET_CRYPT(bp, B_TRUE); } /* Perform the encryption. This should not fail */ VERIFY0(spa_do_crypt_abd(B_TRUE, spa, &zio->io_bookmark, BP_GET_TYPE(bp), BP_GET_DEDUP(bp), BP_SHOULD_BYTESWAP(bp), salt, iv, mac, psize, zio->io_abd, eabd, &no_crypt)); /* encode encryption metadata into the bp */ if (ot == DMU_OT_INTENT_LOG) { /* * ZIL blocks store the MAC in the embedded checksum, so the * transform must always be applied. */ zio_crypt_encode_mac_zil(enc_buf, mac); zio_push_transform(zio, eabd, psize, psize, NULL); } else { BP_SET_CRYPT(bp, B_TRUE); zio_crypt_encode_params_bp(bp, salt, iv); zio_crypt_encode_mac_bp(bp, mac); if (no_crypt) { ASSERT3U(ot, ==, DMU_OT_DNODE); abd_free(eabd); } else { zio_push_transform(zio, eabd, psize, psize, NULL); } } return (zio); } /* * ========================================================================== * Generate and verify checksums * ========================================================================== */ static zio_t * zio_checksum_generate(zio_t *zio) { blkptr_t *bp = zio->io_bp; enum zio_checksum checksum; if (bp == NULL) { /* * This is zio_write_phys(). * We're either generating a label checksum, or none at all. */ checksum = zio->io_prop.zp_checksum; if (checksum == ZIO_CHECKSUM_OFF) return (zio); ASSERT(checksum == ZIO_CHECKSUM_LABEL); } else { if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { ASSERT(!IO_IS_ALLOCATING(zio)); checksum = ZIO_CHECKSUM_GANG_HEADER; } else { checksum = BP_GET_CHECKSUM(bp); } } zio_checksum_compute(zio, checksum, zio->io_abd, zio->io_size); return (zio); } static zio_t * zio_checksum_verify(zio_t *zio) { zio_bad_cksum_t info; blkptr_t *bp = zio->io_bp; int error; ASSERT(zio->io_vd != NULL); if (bp == NULL) { /* * This is zio_read_phys(). * We're either verifying a label checksum, or nothing at all. */ if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) return (zio); ASSERT3U(zio->io_prop.zp_checksum, ==, ZIO_CHECKSUM_LABEL); } if ((error = zio_checksum_error(zio, &info)) != 0) { zio->io_error = error; if (error == ECKSUM && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { (void) zfs_ereport_start_checksum(zio->io_spa, zio->io_vd, &zio->io_bookmark, zio, zio->io_offset, zio->io_size, &info); mutex_enter(&zio->io_vd->vdev_stat_lock); zio->io_vd->vdev_stat.vs_checksum_errors++; mutex_exit(&zio->io_vd->vdev_stat_lock); } } return (zio); } /* * Called by RAID-Z to ensure we don't compute the checksum twice. */ void zio_checksum_verified(zio_t *zio) { zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; } /* * ========================================================================== * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. * An error of 0 indicates success. ENXIO indicates whole-device failure, * which may be transient (e.g. unplugged) or permanent. ECKSUM and EIO * indicate errors that are specific to one I/O, and most likely permanent. * Any other error is presumed to be worse because we weren't expecting it. * ========================================================================== */ int zio_worst_error(int e1, int e2) { static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; int r1, r2; for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) if (e1 == zio_error_rank[r1]) break; for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) if (e2 == zio_error_rank[r2]) break; return (r1 > r2 ? e1 : e2); } /* * ========================================================================== * I/O completion * ========================================================================== */ static zio_t * zio_ready(zio_t *zio) { blkptr_t *bp = zio->io_bp; zio_t *pio, *pio_next; zio_link_t *zl = NULL; if (zio_wait_for_children(zio, ZIO_CHILD_GANG_BIT | ZIO_CHILD_DDT_BIT, ZIO_WAIT_READY)) { return (NULL); } if (zio->io_ready) { ASSERT(IO_IS_ALLOCATING(zio)); ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp) || (zio->io_flags & ZIO_FLAG_NOPWRITE)); ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); zio->io_ready(zio); } if (bp != NULL && bp != &zio->io_bp_copy) zio->io_bp_copy = *bp; if (zio->io_error != 0) { zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(IO_IS_ALLOCATING(zio)); ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(zio->io_metaslab_class != NULL); /* * We were unable to allocate anything, unreserve and * issue the next I/O to allocate. */ metaslab_class_throttle_unreserve( zio->io_metaslab_class, zio->io_prop.zp_copies, zio->io_allocator, zio); zio_allocate_dispatch(zio->io_spa, zio->io_allocator); } } mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_READY] = 1; pio = zio_walk_parents(zio, &zl); mutex_exit(&zio->io_lock); /* * As we notify zio's parents, new parents could be added. * New parents go to the head of zio's io_parent_list, however, * so we will (correctly) not notify them. The remainder of zio's * io_parent_list, from 'pio_next' onward, cannot change because * all parents must wait for us to be done before they can be done. */ for (; pio != NULL; pio = pio_next) { pio_next = zio_walk_parents(zio, &zl); zio_notify_parent(pio, zio, ZIO_WAIT_READY, NULL); } if (zio->io_flags & ZIO_FLAG_NODATA) { if (BP_IS_GANG(bp)) { zio->io_flags &= ~ZIO_FLAG_NODATA; } else { ASSERT((uintptr_t)zio->io_abd < SPA_MAXBLOCKSIZE); zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; } } if (zio_injection_enabled && zio->io_spa->spa_syncing_txg == zio->io_txg) zio_handle_ignored_writes(zio); return (zio); } /* * Update the allocation throttle accounting. */ static void zio_dva_throttle_done(zio_t *zio) { zio_t *lio __maybe_unused = zio->io_logical; zio_t *pio = zio_unique_parent(zio); vdev_t *vd = zio->io_vd; int flags = METASLAB_ASYNC_ALLOC; ASSERT3P(zio->io_bp, !=, NULL); ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE); ASSERT3U(zio->io_priority, ==, ZIO_PRIORITY_ASYNC_WRITE); ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV); ASSERT(vd != NULL); ASSERT3P(vd, ==, vd->vdev_top); ASSERT(zio_injection_enabled || !(zio->io_flags & ZIO_FLAG_IO_RETRY)); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REPAIR)); ASSERT(zio->io_flags & ZIO_FLAG_IO_ALLOCATING); ASSERT(!(lio->io_flags & ZIO_FLAG_IO_REWRITE)); ASSERT(!(lio->io_orig_flags & ZIO_FLAG_NODATA)); /* * Parents of gang children can have two flavors -- ones that * allocated the gang header (will have ZIO_FLAG_IO_REWRITE set) * and ones that allocated the constituent blocks. The allocation * throttle needs to know the allocating parent zio so we must find * it here. */ if (pio->io_child_type == ZIO_CHILD_GANG) { /* * If our parent is a rewrite gang child then our grandparent * would have been the one that performed the allocation. */ if (pio->io_flags & ZIO_FLAG_IO_REWRITE) pio = zio_unique_parent(pio); flags |= METASLAB_GANG_CHILD; } ASSERT(IO_IS_ALLOCATING(pio)); ASSERT3P(zio, !=, zio->io_logical); ASSERT(zio->io_logical != NULL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REPAIR)); ASSERT0(zio->io_flags & ZIO_FLAG_NOPWRITE); ASSERT(zio->io_metaslab_class != NULL); mutex_enter(&pio->io_lock); metaslab_group_alloc_decrement(zio->io_spa, vd->vdev_id, pio, flags, pio->io_allocator, B_TRUE); mutex_exit(&pio->io_lock); metaslab_class_throttle_unreserve(zio->io_metaslab_class, 1, pio->io_allocator, pio); /* * Call into the pipeline to see if there is more work that * needs to be done. If there is work to be done it will be * dispatched to another taskq thread. */ zio_allocate_dispatch(zio->io_spa, pio->io_allocator); } static zio_t * zio_done(zio_t *zio) { /* * Always attempt to keep stack usage minimal here since * we can be called recursively up to 19 levels deep. */ const uint64_t psize = zio->io_size; zio_t *pio, *pio_next; zio_link_t *zl = NULL; /* * If our children haven't all completed, * wait for them and then repeat this pipeline stage. */ if (zio_wait_for_children(zio, ZIO_CHILD_ALL_BITS, ZIO_WAIT_DONE)) { return (NULL); } /* * If the allocation throttle is enabled, then update the accounting. * We only track child I/Os that are part of an allocating async * write. We must do this since the allocation is performed * by the logical I/O but the actual write is done by child I/Os. */ if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING && zio->io_child_type == ZIO_CHILD_VDEV) { ASSERT(zio->io_metaslab_class != NULL); ASSERT(zio->io_metaslab_class->mc_alloc_throttle_enabled); zio_dva_throttle_done(zio); } /* * If the allocation throttle is enabled, verify that * we have decremented the refcounts for every I/O that was throttled. */ if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(zio->io_bp != NULL); metaslab_group_alloc_verify(zio->io_spa, zio->io_bp, zio, zio->io_allocator); VERIFY(zfs_refcount_not_held(&zio->io_metaslab_class-> mc_allocator[zio->io_allocator].mca_alloc_slots, zio)); } for (int c = 0; c < ZIO_CHILD_TYPES; c++) for (int w = 0; w < ZIO_WAIT_TYPES; w++) ASSERT(zio->io_children[c][w] == 0); if (zio->io_bp != NULL && !BP_IS_EMBEDDED(zio->io_bp)) { ASSERT(zio->io_bp->blk_pad[0] == 0); ASSERT(zio->io_bp->blk_pad[1] == 0); ASSERT(bcmp(zio->io_bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || (zio->io_bp == zio_unique_parent(zio)->io_bp)); if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) && zio->io_bp_override == NULL && !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 || (BP_COUNT_GANG(zio->io_bp) == BP_GET_NDVAS(zio->io_bp))); } if (zio->io_flags & ZIO_FLAG_NOPWRITE) VERIFY(BP_EQUAL(zio->io_bp, &zio->io_bp_orig)); } /* * If there were child vdev/gang/ddt errors, they apply to us now. */ zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); zio_inherit_child_errors(zio, ZIO_CHILD_GANG); zio_inherit_child_errors(zio, ZIO_CHILD_DDT); /* * If the I/O on the transformed data was successful, generate any * checksum reports now while we still have the transformed data. */ if (zio->io_error == 0) { while (zio->io_cksum_report != NULL) { zio_cksum_report_t *zcr = zio->io_cksum_report; uint64_t align = zcr->zcr_align; uint64_t asize = P2ROUNDUP(psize, align); abd_t *adata = zio->io_abd; if (adata != NULL && asize != psize) { adata = abd_alloc(asize, B_TRUE); abd_copy(adata, zio->io_abd, psize); abd_zero_off(adata, psize, asize - psize); } zio->io_cksum_report = zcr->zcr_next; zcr->zcr_next = NULL; zcr->zcr_finish(zcr, adata); zfs_ereport_free_checksum(zcr); if (adata != NULL && asize != psize) abd_free(adata); } } zio_pop_transforms(zio); /* note: may set zio->io_error */ vdev_stat_update(zio, psize); /* * If this I/O is attached to a particular vdev is slow, exceeding * 30 seconds to complete, post an error described the I/O delay. * We ignore these errors if the device is currently unavailable. */ if (zio->io_delay >= MSEC2NSEC(zio_slow_io_ms)) { if (zio->io_vd != NULL && !vdev_is_dead(zio->io_vd)) { /* * We want to only increment our slow IO counters if * the IO is valid (i.e. not if the drive is removed). * * zfs_ereport_post() will also do these checks, but * it can also ratelimit and have other failures, so we * need to increment the slow_io counters independent * of it. */ if (zfs_ereport_is_valid(FM_EREPORT_ZFS_DELAY, zio->io_spa, zio->io_vd, zio)) { mutex_enter(&zio->io_vd->vdev_stat_lock); zio->io_vd->vdev_stat.vs_slow_ios++; mutex_exit(&zio->io_vd->vdev_stat_lock); (void) zfs_ereport_post(FM_EREPORT_ZFS_DELAY, zio->io_spa, zio->io_vd, &zio->io_bookmark, zio, 0); } } } if (zio->io_error) { /* * If this I/O is attached to a particular vdev, * generate an error message describing the I/O failure * at the block level. We ignore these errors if the * device is currently unavailable. */ if (zio->io_error != ECKSUM && zio->io_vd != NULL && !vdev_is_dead(zio->io_vd)) { int ret = zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa, zio->io_vd, &zio->io_bookmark, zio, 0); if (ret != EALREADY) { mutex_enter(&zio->io_vd->vdev_stat_lock); if (zio->io_type == ZIO_TYPE_READ) zio->io_vd->vdev_stat.vs_read_errors++; else if (zio->io_type == ZIO_TYPE_WRITE) zio->io_vd->vdev_stat.vs_write_errors++; mutex_exit(&zio->io_vd->vdev_stat_lock); } } if ((zio->io_error == EIO || !(zio->io_flags & (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && zio == zio->io_logical) { /* * For logical I/O requests, tell the SPA to log the * error and generate a logical data ereport. */ spa_log_error(zio->io_spa, &zio->io_bookmark); (void) zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa, NULL, &zio->io_bookmark, zio, 0); } } if (zio->io_error && zio == zio->io_logical) { /* * Determine whether zio should be reexecuted. This will * propagate all the way to the root via zio_notify_parent(). */ ASSERT(zio->io_vd == NULL && zio->io_bp != NULL); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (IO_IS_ALLOCATING(zio) && !(zio->io_flags & ZIO_FLAG_CANFAIL)) { if (zio->io_error != ENOSPC) zio->io_reexecute |= ZIO_REEXECUTE_NOW; else zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; } if ((zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_FREE) && !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_error == ENXIO && spa_load_state(zio->io_spa) == SPA_LOAD_NONE && spa_get_failmode(zio->io_spa) != ZIO_FAILURE_MODE_CONTINUE) zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; /* * Here is a possibly good place to attempt to do * either combinatorial reconstruction or error correction * based on checksums. It also might be a good place * to send out preliminary ereports before we suspend * processing. */ } /* * If there were logical child errors, they apply to us now. * We defer this until now to avoid conflating logical child * errors with errors that happened to the zio itself when * updating vdev stats and reporting FMA events above. */ zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); if ((zio->io_error || zio->io_reexecute) && IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && !(zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE))) zio_dva_unallocate(zio, zio->io_gang_tree, zio->io_bp); zio_gang_tree_free(&zio->io_gang_tree); /* * Godfather I/Os should never suspend. */ if ((zio->io_flags & ZIO_FLAG_GODFATHER) && (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) zio->io_reexecute &= ~ZIO_REEXECUTE_SUSPEND; if (zio->io_reexecute) { /* * This is a logical I/O that wants to reexecute. * * Reexecute is top-down. When an i/o fails, if it's not * the root, it simply notifies its parent and sticks around. * The parent, seeing that it still has children in zio_done(), * does the same. This percolates all the way up to the root. * The root i/o will reexecute or suspend the entire tree. * * This approach ensures that zio_reexecute() honors * all the original i/o dependency relationships, e.g. * parents not executing until children are ready. */ ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); zio->io_gang_leader = NULL; mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_DONE] = 1; mutex_exit(&zio->io_lock); /* * "The Godfather" I/O monitors its children but is * not a true parent to them. It will track them through * the pipeline but severs its ties whenever they get into * trouble (e.g. suspended). This allows "The Godfather" * I/O to return status without blocking. */ zl = NULL; for (pio = zio_walk_parents(zio, &zl); pio != NULL; pio = pio_next) { zio_link_t *remove_zl = zl; pio_next = zio_walk_parents(zio, &zl); if ((pio->io_flags & ZIO_FLAG_GODFATHER) && (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { zio_remove_child(pio, zio, remove_zl); /* * This is a rare code path, so we don't * bother with "next_to_execute". */ zio_notify_parent(pio, zio, ZIO_WAIT_DONE, NULL); } } if ((pio = zio_unique_parent(zio)) != NULL) { /* * We're not a root i/o, so there's nothing to do * but notify our parent. Don't propagate errors * upward since we haven't permanently failed yet. */ ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; /* * This is a rare code path, so we don't bother with * "next_to_execute". */ zio_notify_parent(pio, zio, ZIO_WAIT_DONE, NULL); } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { /* * We'd fail again if we reexecuted now, so suspend * until conditions improve (e.g. device comes online). */ zio_suspend(zio->io_spa, zio, ZIO_SUSPEND_IOERR); } else { /* * Reexecution is potentially a huge amount of work. * Hand it off to the otherwise-unused claim taskq. */ ASSERT(taskq_empty_ent(&zio->io_tqent)); spa_taskq_dispatch_ent(zio->io_spa, ZIO_TYPE_CLAIM, ZIO_TASKQ_ISSUE, (task_func_t *)zio_reexecute, zio, 0, &zio->io_tqent); } return (NULL); } ASSERT(zio->io_child_count == 0); ASSERT(zio->io_reexecute == 0); ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); /* * Report any checksum errors, since the I/O is complete. */ while (zio->io_cksum_report != NULL) { zio_cksum_report_t *zcr = zio->io_cksum_report; zio->io_cksum_report = zcr->zcr_next; zcr->zcr_next = NULL; zcr->zcr_finish(zcr, NULL); zfs_ereport_free_checksum(zcr); } if (zio->io_flags & ZIO_FLAG_FASTWRITE && zio->io_bp && !BP_IS_HOLE(zio->io_bp) && !BP_IS_EMBEDDED(zio->io_bp) && !(zio->io_flags & ZIO_FLAG_NOPWRITE)) { metaslab_fastwrite_unmark(zio->io_spa, zio->io_bp); } /* * It is the responsibility of the done callback to ensure that this * particular zio is no longer discoverable for adoption, and as * such, cannot acquire any new parents. */ if (zio->io_done) zio->io_done(zio); mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_DONE] = 1; mutex_exit(&zio->io_lock); /* * We are done executing this zio. We may want to execute a parent * next. See the comment in zio_notify_parent(). */ zio_t *next_to_execute = NULL; zl = NULL; for (pio = zio_walk_parents(zio, &zl); pio != NULL; pio = pio_next) { zio_link_t *remove_zl = zl; pio_next = zio_walk_parents(zio, &zl); zio_remove_child(pio, zio, remove_zl); zio_notify_parent(pio, zio, ZIO_WAIT_DONE, &next_to_execute); } if (zio->io_waiter != NULL) { mutex_enter(&zio->io_lock); zio->io_executor = NULL; cv_broadcast(&zio->io_cv); mutex_exit(&zio->io_lock); } else { zio_destroy(zio); } return (next_to_execute); } /* * ========================================================================== * I/O pipeline definition * ========================================================================== */ static zio_pipe_stage_t *zio_pipeline[] = { NULL, zio_read_bp_init, zio_write_bp_init, zio_free_bp_init, zio_issue_async, zio_write_compress, zio_encrypt, zio_checksum_generate, zio_nop_write, zio_ddt_read_start, zio_ddt_read_done, zio_ddt_write, zio_ddt_free, zio_gang_assemble, zio_gang_issue, zio_dva_throttle, zio_dva_allocate, zio_dva_free, zio_dva_claim, zio_ready, zio_vdev_io_start, zio_vdev_io_done, zio_vdev_io_assess, zio_checksum_verify, zio_done }; /* * Compare two zbookmark_phys_t's to see which we would reach first in a * pre-order traversal of the object tree. * * This is simple in every case aside from the meta-dnode object. For all other * objects, we traverse them in order (object 1 before object 2, and so on). * However, all of these objects are traversed while traversing object 0, since * the data it points to is the list of objects. Thus, we need to convert to a * canonical representation so we can compare meta-dnode bookmarks to * non-meta-dnode bookmarks. * * We do this by calculating "equivalents" for each field of the zbookmark. * zbookmarks outside of the meta-dnode use their own object and level, and * calculate the level 0 equivalent (the first L0 blkid that is contained in the * blocks this bookmark refers to) by multiplying their blkid by their span * (the number of L0 blocks contained within one block at their level). * zbookmarks inside the meta-dnode calculate their object equivalent * (which is L0equiv * dnodes per data block), use 0 for their L0equiv, and use * level + 1<<31 (any value larger than a level could ever be) for their level. * This causes them to always compare before a bookmark in their object * equivalent, compare appropriately to bookmarks in other objects, and to * compare appropriately to other bookmarks in the meta-dnode. */ int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2) { /* * These variables represent the "equivalent" values for the zbookmark, * after converting zbookmarks inside the meta dnode to their * normal-object equivalents. */ uint64_t zb1obj, zb2obj; uint64_t zb1L0, zb2L0; uint64_t zb1level, zb2level; if (zb1->zb_object == zb2->zb_object && zb1->zb_level == zb2->zb_level && zb1->zb_blkid == zb2->zb_blkid) return (0); IMPLY(zb1->zb_level > 0, ibs1 >= SPA_MINBLOCKSHIFT); IMPLY(zb2->zb_level > 0, ibs2 >= SPA_MINBLOCKSHIFT); /* * BP_SPANB calculates the span in blocks. */ zb1L0 = (zb1->zb_blkid) * BP_SPANB(ibs1, zb1->zb_level); zb2L0 = (zb2->zb_blkid) * BP_SPANB(ibs2, zb2->zb_level); if (zb1->zb_object == DMU_META_DNODE_OBJECT) { zb1obj = zb1L0 * (dbss1 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); zb1L0 = 0; zb1level = zb1->zb_level + COMPARE_META_LEVEL; } else { zb1obj = zb1->zb_object; zb1level = zb1->zb_level; } if (zb2->zb_object == DMU_META_DNODE_OBJECT) { zb2obj = zb2L0 * (dbss2 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); zb2L0 = 0; zb2level = zb2->zb_level + COMPARE_META_LEVEL; } else { zb2obj = zb2->zb_object; zb2level = zb2->zb_level; } /* Now that we have a canonical representation, do the comparison. */ if (zb1obj != zb2obj) return (zb1obj < zb2obj ? -1 : 1); else if (zb1L0 != zb2L0) return (zb1L0 < zb2L0 ? -1 : 1); else if (zb1level != zb2level) return (zb1level > zb2level ? -1 : 1); /* * This can (theoretically) happen if the bookmarks have the same object * and level, but different blkids, if the block sizes are not the same. * There is presently no way to change the indirect block sizes */ return (0); } /* * This function checks the following: given that last_block is the place that * our traversal stopped last time, does that guarantee that we've visited * every node under subtree_root? Therefore, we can't just use the raw output * of zbookmark_compare. We have to pass in a modified version of * subtree_root; by incrementing the block id, and then checking whether * last_block is before or equal to that, we can tell whether or not having * visited last_block implies that all of subtree_root's children have been * visited. */ boolean_t zbookmark_subtree_completed(const dnode_phys_t *dnp, const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block) { zbookmark_phys_t mod_zb = *subtree_root; mod_zb.zb_blkid++; ASSERT(last_block->zb_level == 0); /* The objset_phys_t isn't before anything. */ if (dnp == NULL) return (B_FALSE); /* * We pass in 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT) for the * data block size in sectors, because that variable is only used if * the bookmark refers to a block in the meta-dnode. Since we don't * know without examining it what object it refers to, and there's no * harm in passing in this value in other cases, we always pass it in. * * We pass in 0 for the indirect block size shift because zb2 must be * level 0. The indirect block size is only used to calculate the span * of the bookmark, but since the bookmark must be level 0, the span is * always 1, so the math works out. * * If you make changes to how the zbookmark_compare code works, be sure * to make sure that this code still works afterwards. */ return (zbookmark_compare(dnp->dn_datablkszsec, dnp->dn_indblkshift, 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT), 0, &mod_zb, last_block) <= 0); } EXPORT_SYMBOL(zio_type_name); EXPORT_SYMBOL(zio_buf_alloc); EXPORT_SYMBOL(zio_data_buf_alloc); EXPORT_SYMBOL(zio_buf_free); EXPORT_SYMBOL(zio_data_buf_free); /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs_zio, zio_, slow_io_ms, INT, ZMOD_RW, "Max I/O completion time (milliseconds) before marking it as slow"); ZFS_MODULE_PARAM(zfs_zio, zio_, requeue_io_start_cut_in_line, INT, ZMOD_RW, "Prioritize requeued I/O"); ZFS_MODULE_PARAM(zfs, zfs_, sync_pass_deferred_free, INT, ZMOD_RW, "Defer frees starting in this pass"); ZFS_MODULE_PARAM(zfs, zfs_, sync_pass_dont_compress, INT, ZMOD_RW, "Don't compress starting in this pass"); ZFS_MODULE_PARAM(zfs, zfs_, sync_pass_rewrite, INT, ZMOD_RW, "Rewrite new bps starting in this pass"); ZFS_MODULE_PARAM(zfs_zio, zio_, dva_throttle_enabled, INT, ZMOD_RW, "Throttle block allocations in the ZIO pipeline"); ZFS_MODULE_PARAM(zfs_zio, zio_, deadman_log_all, INT, ZMOD_RW, "Log all slow ZIOs, not just those with vdevs"); /* END CSTYLED */