diff --git a/include/sys/vdev_impl.h b/include/sys/vdev_impl.h index 1239451bf428..c509d390fc76 100644 --- a/include/sys/vdev_impl.h +++ b/include/sys/vdev_impl.h @@ -1,654 +1,655 @@ /* * 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) 2017, Intel Corporation. */ #ifndef _SYS_VDEV_IMPL_H #define _SYS_VDEV_IMPL_H #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* * Virtual device descriptors. * * All storage pool operations go through the virtual device framework, * which provides data replication and I/O scheduling. */ /* * Forward declarations that lots of things need. */ typedef struct vdev_queue vdev_queue_t; typedef struct vdev_cache vdev_cache_t; typedef struct vdev_cache_entry vdev_cache_entry_t; struct abd; extern int zfs_vdev_queue_depth_pct; extern int zfs_vdev_def_queue_depth; extern uint32_t zfs_vdev_async_write_max_active; /* * Virtual device operations */ typedef int vdev_init_func_t(spa_t *spa, nvlist_t *nv, void **tsd); typedef void vdev_fini_func_t(vdev_t *vd); typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size, uint64_t *ashift, uint64_t *pshift); typedef void vdev_close_func_t(vdev_t *vd); typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize); typedef uint64_t vdev_min_asize_func_t(vdev_t *vd); typedef uint64_t vdev_min_alloc_func_t(vdev_t *vd); typedef void vdev_io_start_func_t(zio_t *zio); typedef void vdev_io_done_func_t(zio_t *zio); typedef void vdev_state_change_func_t(vdev_t *vd, int, int); typedef boolean_t vdev_need_resilver_func_t(vdev_t *vd, const dva_t *dva, size_t psize, uint64_t phys_birth); typedef void vdev_hold_func_t(vdev_t *vd); typedef void vdev_rele_func_t(vdev_t *vd); typedef void vdev_remap_cb_t(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg); typedef void vdev_remap_func_t(vdev_t *vd, uint64_t offset, uint64_t size, vdev_remap_cb_t callback, void *arg); /* * Given a target vdev, translates the logical range "in" to the physical * range "res" */ typedef void vdev_xlation_func_t(vdev_t *cvd, const range_seg64_t *logical, range_seg64_t *physical, range_seg64_t *remain); typedef uint64_t vdev_rebuild_asize_func_t(vdev_t *vd, uint64_t start, uint64_t size, uint64_t max_segment); typedef void vdev_metaslab_init_func_t(vdev_t *vd, uint64_t *startp, uint64_t *sizep); typedef void vdev_config_generate_func_t(vdev_t *vd, nvlist_t *nv); typedef uint64_t vdev_nparity_func_t(vdev_t *vd); typedef uint64_t vdev_ndisks_func_t(vdev_t *vd); typedef const struct vdev_ops { vdev_init_func_t *vdev_op_init; vdev_fini_func_t *vdev_op_fini; vdev_open_func_t *vdev_op_open; vdev_close_func_t *vdev_op_close; vdev_asize_func_t *vdev_op_asize; vdev_min_asize_func_t *vdev_op_min_asize; vdev_min_alloc_func_t *vdev_op_min_alloc; vdev_io_start_func_t *vdev_op_io_start; vdev_io_done_func_t *vdev_op_io_done; vdev_state_change_func_t *vdev_op_state_change; vdev_need_resilver_func_t *vdev_op_need_resilver; vdev_hold_func_t *vdev_op_hold; vdev_rele_func_t *vdev_op_rele; vdev_remap_func_t *vdev_op_remap; vdev_xlation_func_t *vdev_op_xlate; vdev_rebuild_asize_func_t *vdev_op_rebuild_asize; vdev_metaslab_init_func_t *vdev_op_metaslab_init; vdev_config_generate_func_t *vdev_op_config_generate; vdev_nparity_func_t *vdev_op_nparity; vdev_ndisks_func_t *vdev_op_ndisks; char vdev_op_type[16]; boolean_t vdev_op_leaf; } vdev_ops_t; /* * Virtual device properties */ struct vdev_cache_entry { struct abd *ve_abd; uint64_t ve_offset; clock_t ve_lastused; avl_node_t ve_offset_node; avl_node_t ve_lastused_node; uint32_t ve_hits; uint16_t ve_missed_update; zio_t *ve_fill_io; }; struct vdev_cache { avl_tree_t vc_offset_tree; avl_tree_t vc_lastused_tree; kmutex_t vc_lock; }; typedef struct vdev_queue_class { uint32_t vqc_active; /* * Sorted by offset or timestamp, depending on if the queue is * LBA-ordered vs FIFO. */ avl_tree_t vqc_queued_tree; } vdev_queue_class_t; struct vdev_queue { vdev_t *vq_vdev; vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE]; avl_tree_t vq_active_tree; avl_tree_t vq_read_offset_tree; avl_tree_t vq_write_offset_tree; avl_tree_t vq_trim_offset_tree; uint64_t vq_last_offset; zio_priority_t vq_last_prio; /* Last sent I/O priority. */ uint32_t vq_ia_active; /* Active interactive I/Os. */ uint32_t vq_nia_credit; /* Non-interactive I/Os credit. */ hrtime_t vq_io_complete_ts; /* time last i/o completed */ hrtime_t vq_io_delta_ts; zio_t vq_io_search; /* used as local for stack reduction */ kmutex_t vq_lock; }; typedef enum vdev_alloc_bias { VDEV_BIAS_NONE, VDEV_BIAS_LOG, /* dedicated to ZIL data (SLOG) */ VDEV_BIAS_SPECIAL, /* dedicated to ddt, metadata, and small blks */ VDEV_BIAS_DEDUP /* dedicated to dedup metadata */ } vdev_alloc_bias_t; /* * On-disk indirect vdev state. * * An indirect vdev is described exclusively in the MOS config of a pool. * The config for an indirect vdev includes several fields, which are * accessed in memory by a vdev_indirect_config_t. */ typedef struct vdev_indirect_config { /* * Object (in MOS) which contains the indirect mapping. This object * contains an array of vdev_indirect_mapping_entry_phys_t ordered by * vimep_src. The bonus buffer for this object is a * vdev_indirect_mapping_phys_t. This object is allocated when a vdev * removal is initiated. * * Note that this object can be empty if none of the data on the vdev * has been copied yet. */ uint64_t vic_mapping_object; /* * Object (in MOS) which contains the birth times for the mapping * entries. This object contains an array of * vdev_indirect_birth_entry_phys_t sorted by vibe_offset. The bonus * buffer for this object is a vdev_indirect_birth_phys_t. This object * is allocated when a vdev removal is initiated. * * Note that this object can be empty if none of the vdev has yet been * copied. */ uint64_t vic_births_object; /* * This is the vdev ID which was removed previous to this vdev, or * UINT64_MAX if there are no previously removed vdevs. */ uint64_t vic_prev_indirect_vdev; } vdev_indirect_config_t; /* * Virtual device descriptor */ struct vdev { /* * Common to all vdev types. */ uint64_t vdev_id; /* child number in vdev parent */ uint64_t vdev_guid; /* unique ID for this vdev */ uint64_t vdev_guid_sum; /* self guid + all child guids */ uint64_t vdev_orig_guid; /* orig. guid prior to remove */ uint64_t vdev_asize; /* allocatable device capacity */ uint64_t vdev_min_asize; /* min acceptable asize */ uint64_t vdev_max_asize; /* max acceptable asize */ uint64_t vdev_ashift; /* block alignment shift */ /* * Logical block alignment shift * * The smallest sized/aligned I/O supported by the device. */ uint64_t vdev_logical_ashift; /* * Physical block alignment shift * * The device supports logical I/Os with vdev_logical_ashift * size/alignment, but optimum performance will be achieved by * aligning/sizing requests to vdev_physical_ashift. Smaller * requests may be inflated or incur device level read-modify-write * operations. * * May be 0 to indicate no preference (i.e. use vdev_logical_ashift). */ uint64_t vdev_physical_ashift; uint64_t vdev_state; /* see VDEV_STATE_* #defines */ uint64_t vdev_prevstate; /* used when reopening a vdev */ vdev_ops_t *vdev_ops; /* vdev operations */ spa_t *vdev_spa; /* spa for this vdev */ void *vdev_tsd; /* type-specific data */ vdev_t *vdev_top; /* top-level vdev */ vdev_t *vdev_parent; /* parent vdev */ vdev_t **vdev_child; /* array of children */ uint64_t vdev_children; /* number of children */ vdev_stat_t vdev_stat; /* virtual device statistics */ vdev_stat_ex_t vdev_stat_ex; /* extended statistics */ boolean_t vdev_expanding; /* expand the vdev? */ boolean_t vdev_reopening; /* reopen in progress? */ boolean_t vdev_nonrot; /* true if solid state */ + int vdev_load_error; /* error on last load */ int vdev_open_error; /* error on last open */ kthread_t *vdev_open_thread; /* thread opening children */ uint64_t vdev_crtxg; /* txg when top-level was added */ /* * Top-level vdev state. */ uint64_t vdev_ms_array; /* metaslab array object */ uint64_t vdev_ms_shift; /* metaslab size shift */ uint64_t vdev_ms_count; /* number of metaslabs */ metaslab_group_t *vdev_mg; /* metaslab group */ metaslab_group_t *vdev_log_mg; /* embedded slog metaslab group */ metaslab_t **vdev_ms; /* metaslab array */ uint64_t vdev_pending_fastwrite; /* allocated fastwrites */ txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */ txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */ txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */ boolean_t vdev_remove_wanted; /* async remove wanted? */ boolean_t vdev_probe_wanted; /* async probe wanted? */ list_node_t vdev_config_dirty_node; /* config dirty list */ list_node_t vdev_state_dirty_node; /* state dirty list */ uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */ uint64_t vdev_islog; /* is an intent log device */ uint64_t vdev_removing; /* device is being removed? */ boolean_t vdev_ishole; /* is a hole in the namespace */ uint64_t vdev_top_zap; vdev_alloc_bias_t vdev_alloc_bias; /* metaslab allocation bias */ /* pool checkpoint related */ space_map_t *vdev_checkpoint_sm; /* contains reserved blocks */ /* Initialize related */ boolean_t vdev_initialize_exit_wanted; vdev_initializing_state_t vdev_initialize_state; list_node_t vdev_initialize_node; kthread_t *vdev_initialize_thread; /* Protects vdev_initialize_thread and vdev_initialize_state. */ kmutex_t vdev_initialize_lock; kcondvar_t vdev_initialize_cv; uint64_t vdev_initialize_offset[TXG_SIZE]; uint64_t vdev_initialize_last_offset; range_tree_t *vdev_initialize_tree; /* valid while initializing */ uint64_t vdev_initialize_bytes_est; uint64_t vdev_initialize_bytes_done; uint64_t vdev_initialize_action_time; /* start and end time */ /* TRIM related */ boolean_t vdev_trim_exit_wanted; boolean_t vdev_autotrim_exit_wanted; vdev_trim_state_t vdev_trim_state; list_node_t vdev_trim_node; kmutex_t vdev_autotrim_lock; kcondvar_t vdev_autotrim_cv; kthread_t *vdev_autotrim_thread; /* Protects vdev_trim_thread and vdev_trim_state. */ kmutex_t vdev_trim_lock; kcondvar_t vdev_trim_cv; kthread_t *vdev_trim_thread; uint64_t vdev_trim_offset[TXG_SIZE]; uint64_t vdev_trim_last_offset; uint64_t vdev_trim_bytes_est; uint64_t vdev_trim_bytes_done; uint64_t vdev_trim_rate; /* requested rate (bytes/sec) */ uint64_t vdev_trim_partial; /* requested partial TRIM */ uint64_t vdev_trim_secure; /* requested secure TRIM */ uint64_t vdev_trim_action_time; /* start and end time */ /* Rebuild related */ boolean_t vdev_rebuilding; boolean_t vdev_rebuild_exit_wanted; boolean_t vdev_rebuild_cancel_wanted; boolean_t vdev_rebuild_reset_wanted; kmutex_t vdev_rebuild_lock; kcondvar_t vdev_rebuild_cv; kthread_t *vdev_rebuild_thread; vdev_rebuild_t vdev_rebuild_config; /* For limiting outstanding I/Os (initialize, TRIM) */ kmutex_t vdev_initialize_io_lock; kcondvar_t vdev_initialize_io_cv; uint64_t vdev_initialize_inflight; kmutex_t vdev_trim_io_lock; kcondvar_t vdev_trim_io_cv; uint64_t vdev_trim_inflight[3]; /* * Values stored in the config for an indirect or removing vdev. */ vdev_indirect_config_t vdev_indirect_config; /* * The vdev_indirect_rwlock protects the vdev_indirect_mapping * pointer from changing on indirect vdevs (when it is condensed). * Note that removing (not yet indirect) vdevs have different * access patterns (the mapping is not accessed from open context, * e.g. from zio_read) and locking strategy (e.g. svr_lock). */ krwlock_t vdev_indirect_rwlock; vdev_indirect_mapping_t *vdev_indirect_mapping; vdev_indirect_births_t *vdev_indirect_births; /* * In memory data structures used to manage the obsolete sm, for * indirect or removing vdevs. * * The vdev_obsolete_segments is the in-core record of the segments * that are no longer referenced anywhere in the pool (due to * being freed or remapped and not referenced by any snapshots). * During a sync, segments are added to vdev_obsolete_segments * via vdev_indirect_mark_obsolete(); at the end of each sync * pass, this is appended to vdev_obsolete_sm via * vdev_indirect_sync_obsolete(). The vdev_obsolete_lock * protects against concurrent modifications of vdev_obsolete_segments * from multiple zio threads. */ kmutex_t vdev_obsolete_lock; range_tree_t *vdev_obsolete_segments; space_map_t *vdev_obsolete_sm; /* * Protects the vdev_scan_io_queue field itself as well as the * structure's contents (when present). */ kmutex_t vdev_scan_io_queue_lock; struct dsl_scan_io_queue *vdev_scan_io_queue; /* * Leaf vdev state. */ range_tree_t *vdev_dtl[DTL_TYPES]; /* dirty time logs */ space_map_t *vdev_dtl_sm; /* dirty time log space map */ txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */ uint64_t vdev_dtl_object; /* DTL object */ uint64_t vdev_psize; /* physical device capacity */ uint64_t vdev_wholedisk; /* true if this is a whole disk */ uint64_t vdev_offline; /* persistent offline state */ uint64_t vdev_faulted; /* persistent faulted state */ uint64_t vdev_degraded; /* persistent degraded state */ uint64_t vdev_removed; /* persistent removed state */ uint64_t vdev_resilver_txg; /* persistent resilvering state */ uint64_t vdev_rebuild_txg; /* persistent rebuilding state */ char *vdev_path; /* vdev path (if any) */ char *vdev_devid; /* vdev devid (if any) */ char *vdev_physpath; /* vdev device path (if any) */ char *vdev_enc_sysfs_path; /* enclosure sysfs path */ char *vdev_fru; /* physical FRU location */ uint64_t vdev_not_present; /* not present during import */ uint64_t vdev_unspare; /* unspare when resilvering done */ boolean_t vdev_nowritecache; /* true if flushwritecache failed */ boolean_t vdev_has_trim; /* TRIM is supported */ boolean_t vdev_has_securetrim; /* secure TRIM is supported */ boolean_t vdev_checkremove; /* temporary online test */ boolean_t vdev_forcefault; /* force online fault */ boolean_t vdev_splitting; /* split or repair in progress */ boolean_t vdev_delayed_close; /* delayed device close? */ boolean_t vdev_tmpoffline; /* device taken offline temporarily? */ boolean_t vdev_detached; /* device detached? */ boolean_t vdev_cant_read; /* vdev is failing all reads */ boolean_t vdev_cant_write; /* vdev is failing all writes */ boolean_t vdev_isspare; /* was a hot spare */ boolean_t vdev_isl2cache; /* was a l2cache device */ boolean_t vdev_copy_uberblocks; /* post expand copy uberblocks */ boolean_t vdev_resilver_deferred; /* resilver deferred */ vdev_queue_t vdev_queue; /* I/O deadline schedule queue */ vdev_cache_t vdev_cache; /* physical block cache */ spa_aux_vdev_t *vdev_aux; /* for l2cache and spares vdevs */ zio_t *vdev_probe_zio; /* root of current probe */ vdev_aux_t vdev_label_aux; /* on-disk aux state */ uint64_t vdev_leaf_zap; hrtime_t vdev_mmp_pending; /* 0 if write finished */ uint64_t vdev_mmp_kstat_id; /* to find kstat entry */ uint64_t vdev_expansion_time; /* vdev's last expansion time */ list_node_t vdev_leaf_node; /* leaf vdev list */ /* * For DTrace to work in userland (libzpool) context, these fields must * remain at the end of the structure. DTrace will use the kernel's * CTF definition for 'struct vdev', and since the size of a kmutex_t is * larger in userland, the offsets for the rest of the fields would be * incorrect. */ kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */ kmutex_t vdev_stat_lock; /* vdev_stat */ kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */ /* * We rate limit ZIO delay and ZIO checksum events, since they * can flood ZED with tons of events when a drive is acting up. */ zfs_ratelimit_t vdev_delay_rl; zfs_ratelimit_t vdev_checksum_rl; }; #define VDEV_PAD_SIZE (8 << 10) /* 2 padding areas (vl_pad1 and vl_be) to skip */ #define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2 #define VDEV_PHYS_SIZE (112 << 10) #define VDEV_UBERBLOCK_RING (128 << 10) /* * MMP blocks occupy the last MMP_BLOCKS_PER_LABEL slots in the uberblock * ring when MMP is enabled. */ #define MMP_BLOCKS_PER_LABEL 1 /* The largest uberblock we support is 8k. */ #define MAX_UBERBLOCK_SHIFT (13) #define VDEV_UBERBLOCK_SHIFT(vd) \ MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \ MAX_UBERBLOCK_SHIFT) #define VDEV_UBERBLOCK_COUNT(vd) \ (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd)) #define VDEV_UBERBLOCK_OFFSET(vd, n) \ offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)]) #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd)) typedef struct vdev_phys { char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)]; zio_eck_t vp_zbt; } vdev_phys_t; typedef enum vbe_vers { /* * The bootenv file is stored as ascii text in the envblock. * It is used by the GRUB bootloader used on Linux to store the * contents of the grubenv file. The file is stored as raw ASCII, * and is protected by an embedded checksum. By default, GRUB will * check if the boot filesystem supports storing the environment data * in a special location, and if so, will invoke filesystem specific * logic to retrieve it. This can be overriden by a variable, should * the user so desire. */ VB_RAW = 0, /* * The bootenv file is converted to an nvlist and then packed into the * envblock. */ VB_NVLIST = 1 } vbe_vers_t; typedef struct vdev_boot_envblock { uint64_t vbe_version; char vbe_bootenv[VDEV_PAD_SIZE - sizeof (uint64_t) - sizeof (zio_eck_t)]; zio_eck_t vbe_zbt; } vdev_boot_envblock_t; CTASSERT_GLOBAL(sizeof (vdev_boot_envblock_t) == VDEV_PAD_SIZE); typedef struct vdev_label { char vl_pad1[VDEV_PAD_SIZE]; /* 8K */ vdev_boot_envblock_t vl_be; /* 8K */ vdev_phys_t vl_vdev_phys; /* 112K */ char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */ } vdev_label_t; /* 256K total */ /* * vdev_dirty() flags */ #define VDD_METASLAB 0x01 #define VDD_DTL 0x02 /* Offset of embedded boot loader region on each label */ #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t)) /* * Size of embedded boot loader region on each label. * The total size of the first two labels plus the boot area is 4MB. */ #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */ /* * Size of label regions at the start and end of each leaf device. */ #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE) #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t)) #define VDEV_LABELS 4 #define VDEV_BEST_LABEL VDEV_LABELS #define VDEV_OFFSET_IS_LABEL(vd, off) \ (((off) < VDEV_LABEL_START_SIZE) || \ ((off) >= ((vd)->vdev_psize - VDEV_LABEL_END_SIZE))) #define VDEV_ALLOC_LOAD 0 #define VDEV_ALLOC_ADD 1 #define VDEV_ALLOC_SPARE 2 #define VDEV_ALLOC_L2CACHE 3 #define VDEV_ALLOC_ROOTPOOL 4 #define VDEV_ALLOC_SPLIT 5 #define VDEV_ALLOC_ATTACH 6 /* * Allocate or free a vdev */ extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops); extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config, vdev_t *parent, uint_t id, int alloctype); extern void vdev_free(vdev_t *vd); /* * Add or remove children and parents */ extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd); extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd); extern void vdev_compact_children(vdev_t *pvd); extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops); extern void vdev_remove_parent(vdev_t *cvd); /* * vdev sync load and sync */ extern boolean_t vdev_log_state_valid(vdev_t *vd); extern int vdev_load(vdev_t *vd); extern int vdev_dtl_load(vdev_t *vd); extern void vdev_sync(vdev_t *vd, uint64_t txg); extern void vdev_sync_done(vdev_t *vd, uint64_t txg); extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg); extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg); /* * Available vdev types. */ extern vdev_ops_t vdev_root_ops; extern vdev_ops_t vdev_mirror_ops; extern vdev_ops_t vdev_replacing_ops; extern vdev_ops_t vdev_raidz_ops; extern vdev_ops_t vdev_draid_ops; extern vdev_ops_t vdev_draid_spare_ops; extern vdev_ops_t vdev_disk_ops; extern vdev_ops_t vdev_file_ops; extern vdev_ops_t vdev_missing_ops; extern vdev_ops_t vdev_hole_ops; extern vdev_ops_t vdev_spare_ops; extern vdev_ops_t vdev_indirect_ops; /* * Common size functions */ extern void vdev_default_xlate(vdev_t *vd, const range_seg64_t *logical_rs, range_seg64_t *physical_rs, range_seg64_t *remain_rs); extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize); extern uint64_t vdev_default_min_asize(vdev_t *vd); extern uint64_t vdev_get_min_asize(vdev_t *vd); extern void vdev_set_min_asize(vdev_t *vd); extern uint64_t vdev_get_min_alloc(vdev_t *vd); extern uint64_t vdev_get_nparity(vdev_t *vd); extern uint64_t vdev_get_ndisks(vdev_t *vd); /* * Global variables */ extern int zfs_vdev_standard_sm_blksz; /* zdb uses this tunable, so it must be declared here to make lint happy. */ extern int zfs_vdev_cache_size; /* * Functions from vdev_indirect.c */ extern void vdev_indirect_sync_obsolete(vdev_t *vd, dmu_tx_t *tx); extern boolean_t vdev_indirect_should_condense(vdev_t *vd); extern void spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx); extern int vdev_obsolete_sm_object(vdev_t *vd, uint64_t *sm_obj); extern int vdev_obsolete_counts_are_precise(vdev_t *vd, boolean_t *are_precise); /* * Other miscellaneous functions */ int vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj); void vdev_metaslab_group_create(vdev_t *vd); /* * Vdev ashift optimization tunables */ extern uint64_t zfs_vdev_min_auto_ashift; extern uint64_t zfs_vdev_max_auto_ashift; int param_set_min_auto_ashift(ZFS_MODULE_PARAM_ARGS); int param_set_max_auto_ashift(ZFS_MODULE_PARAM_ARGS); #ifdef __cplusplus } #endif #endif /* _SYS_VDEV_IMPL_H */ diff --git a/module/zfs/metaslab.c b/module/zfs/metaslab.c index fdc6922b0011..bc4f007b61a1 100644 --- a/module/zfs/metaslab.c +++ b/module/zfs/metaslab.c @@ -1,6281 +1,6287 @@ /* * 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, 2019 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2017, Intel Corporation. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #define WITH_DF_BLOCK_ALLOCATOR #define GANG_ALLOCATION(flags) \ ((flags) & (METASLAB_GANG_CHILD | METASLAB_GANG_HEADER)) /* * Metaslab granularity, in bytes. This is roughly similar to what would be * referred to as the "stripe size" in traditional RAID arrays. In normal * operation, we will try to write this amount of data to a top-level vdev * before moving on to the next one. */ unsigned long metaslab_aliquot = 512 << 10; /* * For testing, make some blocks above a certain size be gang blocks. */ unsigned long metaslab_force_ganging = SPA_MAXBLOCKSIZE + 1; /* * In pools where the log space map feature is not enabled we touch * multiple metaslabs (and their respective space maps) with each * transaction group. Thus, we benefit from having a small space map * block size since it allows us to issue more I/O operations scattered * around the disk. So a sane default for the space map block size * is 8~16K. */ int zfs_metaslab_sm_blksz_no_log = (1 << 14); /* * When the log space map feature is enabled, we accumulate a lot of * changes per metaslab that are flushed once in a while so we benefit * from a bigger block size like 128K for the metaslab space maps. */ int zfs_metaslab_sm_blksz_with_log = (1 << 17); /* * The in-core space map representation is more compact than its on-disk form. * The zfs_condense_pct determines how much more compact the in-core * space map representation must be before we compact it on-disk. * Values should be greater than or equal to 100. */ int zfs_condense_pct = 200; /* * Condensing a metaslab is not guaranteed to actually reduce the amount of * space used on disk. In particular, a space map uses data in increments of * MAX(1 << ashift, space_map_blksz), so a metaslab might use the * same number of blocks after condensing. Since the goal of condensing is to * reduce the number of IOPs required to read the space map, we only want to * condense when we can be sure we will reduce the number of blocks used by the * space map. Unfortunately, we cannot precisely compute whether or not this is * the case in metaslab_should_condense since we are holding ms_lock. Instead, * we apply the following heuristic: do not condense a spacemap unless the * uncondensed size consumes greater than zfs_metaslab_condense_block_threshold * blocks. */ int zfs_metaslab_condense_block_threshold = 4; /* * The zfs_mg_noalloc_threshold defines which metaslab groups should * be eligible for allocation. The value is defined as a percentage of * free space. Metaslab groups that have more free space than * zfs_mg_noalloc_threshold are always eligible for allocations. Once * a metaslab group's free space is less than or equal to the * zfs_mg_noalloc_threshold the allocator will avoid allocating to that * group unless all groups in the pool have reached zfs_mg_noalloc_threshold. * Once all groups in the pool reach zfs_mg_noalloc_threshold then all * groups are allowed to accept allocations. Gang blocks are always * eligible to allocate on any metaslab group. The default value of 0 means * no metaslab group will be excluded based on this criterion. */ int zfs_mg_noalloc_threshold = 0; /* * Metaslab groups are considered eligible for allocations if their * fragmentation metric (measured as a percentage) is less than or * equal to zfs_mg_fragmentation_threshold. If a metaslab group * exceeds this threshold then it will be skipped unless all metaslab * groups within the metaslab class have also crossed this threshold. * * This tunable was introduced to avoid edge cases where we continue * allocating from very fragmented disks in our pool while other, less * fragmented disks, exists. On the other hand, if all disks in the * pool are uniformly approaching the threshold, the threshold can * be a speed bump in performance, where we keep switching the disks * that we allocate from (e.g. we allocate some segments from disk A * making it bypassing the threshold while freeing segments from disk * B getting its fragmentation below the threshold). * * Empirically, we've seen that our vdev selection for allocations is * good enough that fragmentation increases uniformly across all vdevs * the majority of the time. Thus we set the threshold percentage high * enough to avoid hitting the speed bump on pools that are being pushed * to the edge. */ int zfs_mg_fragmentation_threshold = 95; /* * Allow metaslabs to keep their active state as long as their fragmentation * percentage is less than or equal to zfs_metaslab_fragmentation_threshold. An * active metaslab that exceeds this threshold will no longer keep its active * status allowing better metaslabs to be selected. */ int zfs_metaslab_fragmentation_threshold = 70; /* * When set will load all metaslabs when pool is first opened. */ int metaslab_debug_load = 0; /* * When set will prevent metaslabs from being unloaded. */ int metaslab_debug_unload = 0; /* * Minimum size which forces the dynamic allocator to change * it's allocation strategy. Once the space map cannot satisfy * an allocation of this size then it switches to using more * aggressive strategy (i.e search by size rather than offset). */ uint64_t metaslab_df_alloc_threshold = SPA_OLD_MAXBLOCKSIZE; /* * The minimum free space, in percent, which must be available * in a space map to continue allocations in a first-fit fashion. * Once the space map's free space drops below this level we dynamically * switch to using best-fit allocations. */ int metaslab_df_free_pct = 4; /* * Maximum distance to search forward from the last offset. Without this * limit, fragmented pools can see >100,000 iterations and * metaslab_block_picker() becomes the performance limiting factor on * high-performance storage. * * With the default setting of 16MB, we typically see less than 500 * iterations, even with very fragmented, ashift=9 pools. The maximum number * of iterations possible is: * metaslab_df_max_search / (2 * (1<60KB (but fewer segments in this * bucket, and therefore a lower weight). */ int zfs_metaslab_find_max_tries = 100; static uint64_t metaslab_weight(metaslab_t *, boolean_t); static void metaslab_set_fragmentation(metaslab_t *, boolean_t); static void metaslab_free_impl(vdev_t *, uint64_t, uint64_t, boolean_t); static void metaslab_check_free_impl(vdev_t *, uint64_t, uint64_t); static void metaslab_passivate(metaslab_t *msp, uint64_t weight); static uint64_t metaslab_weight_from_range_tree(metaslab_t *msp); static void metaslab_flush_update(metaslab_t *, dmu_tx_t *); static unsigned int metaslab_idx_func(multilist_t *, void *); static void metaslab_evict(metaslab_t *, uint64_t); static void metaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg); kmem_cache_t *metaslab_alloc_trace_cache; typedef struct metaslab_stats { kstat_named_t metaslabstat_trace_over_limit; kstat_named_t metaslabstat_reload_tree; kstat_named_t metaslabstat_too_many_tries; kstat_named_t metaslabstat_try_hard; } metaslab_stats_t; static metaslab_stats_t metaslab_stats = { { "trace_over_limit", KSTAT_DATA_UINT64 }, { "reload_tree", KSTAT_DATA_UINT64 }, { "too_many_tries", KSTAT_DATA_UINT64 }, { "try_hard", KSTAT_DATA_UINT64 }, }; #define METASLABSTAT_BUMP(stat) \ atomic_inc_64(&metaslab_stats.stat.value.ui64); kstat_t *metaslab_ksp; void metaslab_stat_init(void) { ASSERT(metaslab_alloc_trace_cache == NULL); metaslab_alloc_trace_cache = kmem_cache_create( "metaslab_alloc_trace_cache", sizeof (metaslab_alloc_trace_t), 0, NULL, NULL, NULL, NULL, NULL, 0); metaslab_ksp = kstat_create("zfs", 0, "metaslab_stats", "misc", KSTAT_TYPE_NAMED, sizeof (metaslab_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (metaslab_ksp != NULL) { metaslab_ksp->ks_data = &metaslab_stats; kstat_install(metaslab_ksp); } } void metaslab_stat_fini(void) { if (metaslab_ksp != NULL) { kstat_delete(metaslab_ksp); metaslab_ksp = NULL; } kmem_cache_destroy(metaslab_alloc_trace_cache); metaslab_alloc_trace_cache = NULL; } /* * ========================================================================== * Metaslab classes * ========================================================================== */ metaslab_class_t * metaslab_class_create(spa_t *spa, metaslab_ops_t *ops) { metaslab_class_t *mc; mc = kmem_zalloc(offsetof(metaslab_class_t, mc_allocator[spa->spa_alloc_count]), KM_SLEEP); mc->mc_spa = spa; mc->mc_ops = ops; mutex_init(&mc->mc_lock, NULL, MUTEX_DEFAULT, NULL); mc->mc_metaslab_txg_list = multilist_create(sizeof (metaslab_t), offsetof(metaslab_t, ms_class_txg_node), metaslab_idx_func); for (int i = 0; i < spa->spa_alloc_count; i++) { metaslab_class_allocator_t *mca = &mc->mc_allocator[i]; mca->mca_rotor = NULL; zfs_refcount_create_tracked(&mca->mca_alloc_slots); } return (mc); } void metaslab_class_destroy(metaslab_class_t *mc) { spa_t *spa = mc->mc_spa; ASSERT(mc->mc_alloc == 0); ASSERT(mc->mc_deferred == 0); ASSERT(mc->mc_space == 0); ASSERT(mc->mc_dspace == 0); for (int i = 0; i < spa->spa_alloc_count; i++) { metaslab_class_allocator_t *mca = &mc->mc_allocator[i]; ASSERT(mca->mca_rotor == NULL); zfs_refcount_destroy(&mca->mca_alloc_slots); } mutex_destroy(&mc->mc_lock); multilist_destroy(mc->mc_metaslab_txg_list); kmem_free(mc, offsetof(metaslab_class_t, mc_allocator[spa->spa_alloc_count])); } int metaslab_class_validate(metaslab_class_t *mc) { metaslab_group_t *mg; vdev_t *vd; /* * Must hold one of the spa_config locks. */ ASSERT(spa_config_held(mc->mc_spa, SCL_ALL, RW_READER) || spa_config_held(mc->mc_spa, SCL_ALL, RW_WRITER)); if ((mg = mc->mc_allocator[0].mca_rotor) == NULL) return (0); do { vd = mg->mg_vd; ASSERT(vd->vdev_mg != NULL); ASSERT3P(vd->vdev_top, ==, vd); ASSERT3P(mg->mg_class, ==, mc); ASSERT3P(vd->vdev_ops, !=, &vdev_hole_ops); } while ((mg = mg->mg_next) != mc->mc_allocator[0].mca_rotor); return (0); } static void metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta, int64_t defer_delta, int64_t space_delta, int64_t dspace_delta) { atomic_add_64(&mc->mc_alloc, alloc_delta); atomic_add_64(&mc->mc_deferred, defer_delta); atomic_add_64(&mc->mc_space, space_delta); atomic_add_64(&mc->mc_dspace, dspace_delta); } uint64_t metaslab_class_get_alloc(metaslab_class_t *mc) { return (mc->mc_alloc); } uint64_t metaslab_class_get_deferred(metaslab_class_t *mc) { return (mc->mc_deferred); } uint64_t metaslab_class_get_space(metaslab_class_t *mc) { return (mc->mc_space); } uint64_t metaslab_class_get_dspace(metaslab_class_t *mc) { return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space); } void metaslab_class_histogram_verify(metaslab_class_t *mc) { spa_t *spa = mc->mc_spa; vdev_t *rvd = spa->spa_root_vdev; uint64_t *mc_hist; int i; if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0) return; mc_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE, KM_SLEEP); + mutex_enter(&mc->mc_lock); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = vdev_get_mg(tvd, mc); /* * Skip any holes, uninitialized top-levels, or * vdevs that are not in this metalab class. */ if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } IMPLY(mg == mg->mg_vd->vdev_log_mg, mc == spa_embedded_log_class(mg->mg_vd->vdev_spa)); for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) mc_hist[i] += mg->mg_histogram[i]; } for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { VERIFY3U(mc_hist[i], ==, mc->mc_histogram[i]); } + mutex_exit(&mc->mc_lock); kmem_free(mc_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE); } /* * Calculate the metaslab class's fragmentation metric. The metric * is weighted based on the space contribution of each metaslab group. * The return value will be a number between 0 and 100 (inclusive), or * ZFS_FRAG_INVALID if the metric has not been set. See comment above the * zfs_frag_table for more information about the metric. */ uint64_t metaslab_class_fragmentation(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t fragmentation = 0; spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; /* * Skip any holes, uninitialized top-levels, * or vdevs that are not in this metalab class. */ if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } /* * If a metaslab group does not contain a fragmentation * metric then just bail out. */ if (mg->mg_fragmentation == ZFS_FRAG_INVALID) { spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (ZFS_FRAG_INVALID); } /* * Determine how much this metaslab_group is contributing * to the overall pool fragmentation metric. */ fragmentation += mg->mg_fragmentation * metaslab_group_get_space(mg); } fragmentation /= metaslab_class_get_space(mc); ASSERT3U(fragmentation, <=, 100); spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (fragmentation); } /* * Calculate the amount of expandable space that is available in * this metaslab class. If a device is expanded then its expandable * space will be the amount of allocatable space that is currently not * part of this metaslab class. */ uint64_t metaslab_class_expandable_space(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t space = 0; spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } /* * Calculate if we have enough space to add additional * metaslabs. We report the expandable space in terms * of the metaslab size since that's the unit of expansion. */ space += P2ALIGN(tvd->vdev_max_asize - tvd->vdev_asize, 1ULL << tvd->vdev_ms_shift); } spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (space); } void metaslab_class_evict_old(metaslab_class_t *mc, uint64_t txg) { multilist_t *ml = mc->mc_metaslab_txg_list; for (int i = 0; i < multilist_get_num_sublists(ml); i++) { multilist_sublist_t *mls = multilist_sublist_lock(ml, i); metaslab_t *msp = multilist_sublist_head(mls); multilist_sublist_unlock(mls); while (msp != NULL) { mutex_enter(&msp->ms_lock); /* * If the metaslab has been removed from the list * (which could happen if we were at the memory limit * and it was evicted during this loop), then we can't * proceed and we should restart the sublist. */ if (!multilist_link_active(&msp->ms_class_txg_node)) { mutex_exit(&msp->ms_lock); i--; break; } mls = multilist_sublist_lock(ml, i); metaslab_t *next_msp = multilist_sublist_next(mls, msp); multilist_sublist_unlock(mls); if (txg > msp->ms_selected_txg + metaslab_unload_delay && gethrtime() > msp->ms_selected_time + (uint64_t)MSEC2NSEC(metaslab_unload_delay_ms)) { metaslab_evict(msp, txg); } else { /* * Once we've hit a metaslab selected too * recently to evict, we're done evicting for * now. */ mutex_exit(&msp->ms_lock); break; } mutex_exit(&msp->ms_lock); msp = next_msp; } } } static int metaslab_compare(const void *x1, const void *x2) { const metaslab_t *m1 = (const metaslab_t *)x1; const metaslab_t *m2 = (const metaslab_t *)x2; int sort1 = 0; int sort2 = 0; if (m1->ms_allocator != -1 && m1->ms_primary) sort1 = 1; else if (m1->ms_allocator != -1 && !m1->ms_primary) sort1 = 2; if (m2->ms_allocator != -1 && m2->ms_primary) sort2 = 1; else if (m2->ms_allocator != -1 && !m2->ms_primary) sort2 = 2; /* * Sort inactive metaslabs first, then primaries, then secondaries. When * selecting a metaslab to allocate from, an allocator first tries its * primary, then secondary active metaslab. If it doesn't have active * metaslabs, or can't allocate from them, it searches for an inactive * metaslab to activate. If it can't find a suitable one, it will steal * a primary or secondary metaslab from another allocator. */ if (sort1 < sort2) return (-1); if (sort1 > sort2) return (1); int cmp = TREE_CMP(m2->ms_weight, m1->ms_weight); if (likely(cmp)) return (cmp); IMPLY(TREE_CMP(m1->ms_start, m2->ms_start) == 0, m1 == m2); return (TREE_CMP(m1->ms_start, m2->ms_start)); } /* * ========================================================================== * Metaslab groups * ========================================================================== */ /* * Update the allocatable flag and the metaslab group's capacity. * The allocatable flag is set to true if the capacity is below * the zfs_mg_noalloc_threshold or has a fragmentation value that is * greater than zfs_mg_fragmentation_threshold. If a metaslab group * transitions from allocatable to non-allocatable or vice versa then the * metaslab group's class is updated to reflect the transition. */ static void metaslab_group_alloc_update(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; metaslab_class_t *mc = mg->mg_class; vdev_stat_t *vs = &vd->vdev_stat; boolean_t was_allocatable; boolean_t was_initialized; ASSERT(vd == vd->vdev_top); ASSERT3U(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_READER), ==, SCL_ALLOC); mutex_enter(&mg->mg_lock); was_allocatable = mg->mg_allocatable; was_initialized = mg->mg_initialized; mg->mg_free_capacity = ((vs->vs_space - vs->vs_alloc) * 100) / (vs->vs_space + 1); mutex_enter(&mc->mc_lock); /* * If the metaslab group was just added then it won't * have any space until we finish syncing out this txg. * At that point we will consider it initialized and available * for allocations. We also don't consider non-activated * metaslab groups (e.g. vdevs that are in the middle of being removed) * to be initialized, because they can't be used for allocation. */ mg->mg_initialized = metaslab_group_initialized(mg); if (!was_initialized && mg->mg_initialized) { mc->mc_groups++; } else if (was_initialized && !mg->mg_initialized) { ASSERT3U(mc->mc_groups, >, 0); mc->mc_groups--; } if (mg->mg_initialized) mg->mg_no_free_space = B_FALSE; /* * A metaslab group is considered allocatable if it has plenty * of free space or is not heavily fragmented. We only take * fragmentation into account if the metaslab group has a valid * fragmentation metric (i.e. a value between 0 and 100). */ mg->mg_allocatable = (mg->mg_activation_count > 0 && mg->mg_free_capacity > zfs_mg_noalloc_threshold && (mg->mg_fragmentation == ZFS_FRAG_INVALID || mg->mg_fragmentation <= zfs_mg_fragmentation_threshold)); /* * The mc_alloc_groups maintains a count of the number of * groups in this metaslab class that are still above the * zfs_mg_noalloc_threshold. This is used by the allocating * threads to determine if they should avoid allocations to * a given group. The allocator will avoid allocations to a group * if that group has reached or is below the zfs_mg_noalloc_threshold * and there are still other groups that are above the threshold. * When a group transitions from allocatable to non-allocatable or * vice versa we update the metaslab class to reflect that change. * When the mc_alloc_groups value drops to 0 that means that all * groups have reached the zfs_mg_noalloc_threshold making all groups * eligible for allocations. This effectively means that all devices * are balanced again. */ if (was_allocatable && !mg->mg_allocatable) mc->mc_alloc_groups--; else if (!was_allocatable && mg->mg_allocatable) mc->mc_alloc_groups++; mutex_exit(&mc->mc_lock); mutex_exit(&mg->mg_lock); } int metaslab_sort_by_flushed(const void *va, const void *vb) { const metaslab_t *a = va; const metaslab_t *b = vb; int cmp = TREE_CMP(a->ms_unflushed_txg, b->ms_unflushed_txg); if (likely(cmp)) return (cmp); uint64_t a_vdev_id = a->ms_group->mg_vd->vdev_id; uint64_t b_vdev_id = b->ms_group->mg_vd->vdev_id; cmp = TREE_CMP(a_vdev_id, b_vdev_id); if (cmp) return (cmp); return (TREE_CMP(a->ms_id, b->ms_id)); } metaslab_group_t * metaslab_group_create(metaslab_class_t *mc, vdev_t *vd, int allocators) { metaslab_group_t *mg; mg = kmem_zalloc(offsetof(metaslab_group_t, mg_allocator[allocators]), KM_SLEEP); mutex_init(&mg->mg_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&mg->mg_ms_disabled_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&mg->mg_ms_disabled_cv, NULL, CV_DEFAULT, NULL); avl_create(&mg->mg_metaslab_tree, metaslab_compare, sizeof (metaslab_t), offsetof(metaslab_t, ms_group_node)); mg->mg_vd = vd; mg->mg_class = mc; mg->mg_activation_count = 0; mg->mg_initialized = B_FALSE; mg->mg_no_free_space = B_TRUE; mg->mg_allocators = allocators; for (int i = 0; i < allocators; i++) { metaslab_group_allocator_t *mga = &mg->mg_allocator[i]; zfs_refcount_create_tracked(&mga->mga_alloc_queue_depth); } mg->mg_taskq = taskq_create("metaslab_group_taskq", metaslab_load_pct, maxclsyspri, 10, INT_MAX, TASKQ_THREADS_CPU_PCT | TASKQ_DYNAMIC); return (mg); } void metaslab_group_destroy(metaslab_group_t *mg) { ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); /* * We may have gone below zero with the activation count * either because we never activated in the first place or * because we're done, and possibly removing the vdev. */ ASSERT(mg->mg_activation_count <= 0); taskq_destroy(mg->mg_taskq); avl_destroy(&mg->mg_metaslab_tree); mutex_destroy(&mg->mg_lock); mutex_destroy(&mg->mg_ms_disabled_lock); cv_destroy(&mg->mg_ms_disabled_cv); for (int i = 0; i < mg->mg_allocators; i++) { metaslab_group_allocator_t *mga = &mg->mg_allocator[i]; zfs_refcount_destroy(&mga->mga_alloc_queue_depth); } kmem_free(mg, offsetof(metaslab_group_t, mg_allocator[mg->mg_allocators])); } void metaslab_group_activate(metaslab_group_t *mg) { metaslab_class_t *mc = mg->mg_class; spa_t *spa = mc->mc_spa; metaslab_group_t *mgprev, *mgnext; ASSERT3U(spa_config_held(spa, SCL_ALLOC, RW_WRITER), !=, 0); ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); ASSERT(mg->mg_activation_count <= 0); if (++mg->mg_activation_count <= 0) return; mg->mg_aliquot = metaslab_aliquot * MAX(1, mg->mg_vd->vdev_children); metaslab_group_alloc_update(mg); if ((mgprev = mc->mc_allocator[0].mca_rotor) == NULL) { mg->mg_prev = mg; mg->mg_next = mg; } else { mgnext = mgprev->mg_next; mg->mg_prev = mgprev; mg->mg_next = mgnext; mgprev->mg_next = mg; mgnext->mg_prev = mg; } for (int i = 0; i < spa->spa_alloc_count; i++) { mc->mc_allocator[i].mca_rotor = mg; mg = mg->mg_next; } } /* * Passivate a metaslab group and remove it from the allocation rotor. * Callers must hold both the SCL_ALLOC and SCL_ZIO lock prior to passivating * a metaslab group. This function will momentarily drop spa_config_locks * that are lower than the SCL_ALLOC lock (see comment below). */ void metaslab_group_passivate(metaslab_group_t *mg) { metaslab_class_t *mc = mg->mg_class; spa_t *spa = mc->mc_spa; metaslab_group_t *mgprev, *mgnext; int locks = spa_config_held(spa, SCL_ALL, RW_WRITER); ASSERT3U(spa_config_held(spa, SCL_ALLOC | SCL_ZIO, RW_WRITER), ==, (SCL_ALLOC | SCL_ZIO)); if (--mg->mg_activation_count != 0) { for (int i = 0; i < spa->spa_alloc_count; i++) ASSERT(mc->mc_allocator[i].mca_rotor != mg); ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); ASSERT(mg->mg_activation_count < 0); return; } /* * The spa_config_lock is an array of rwlocks, ordered as * follows (from highest to lowest): * SCL_CONFIG > SCL_STATE > SCL_L2ARC > SCL_ALLOC > * SCL_ZIO > SCL_FREE > SCL_VDEV * (For more information about the spa_config_lock see spa_misc.c) * The higher the lock, the broader its coverage. When we passivate * a metaslab group, we must hold both the SCL_ALLOC and the SCL_ZIO * config locks. However, the metaslab group's taskq might be trying * to preload metaslabs so we must drop the SCL_ZIO lock and any * lower locks to allow the I/O to complete. At a minimum, * we continue to hold the SCL_ALLOC lock, which prevents any future * allocations from taking place and any changes to the vdev tree. */ spa_config_exit(spa, locks & ~(SCL_ZIO - 1), spa); taskq_wait_outstanding(mg->mg_taskq, 0); spa_config_enter(spa, locks & ~(SCL_ZIO - 1), spa, RW_WRITER); metaslab_group_alloc_update(mg); for (int i = 0; i < mg->mg_allocators; i++) { metaslab_group_allocator_t *mga = &mg->mg_allocator[i]; metaslab_t *msp = mga->mga_primary; if (msp != NULL) { mutex_enter(&msp->ms_lock); metaslab_passivate(msp, metaslab_weight_from_range_tree(msp)); mutex_exit(&msp->ms_lock); } msp = mga->mga_secondary; if (msp != NULL) { mutex_enter(&msp->ms_lock); metaslab_passivate(msp, metaslab_weight_from_range_tree(msp)); mutex_exit(&msp->ms_lock); } } mgprev = mg->mg_prev; mgnext = mg->mg_next; if (mg == mgnext) { mgnext = NULL; } else { mgprev->mg_next = mgnext; mgnext->mg_prev = mgprev; } for (int i = 0; i < spa->spa_alloc_count; i++) { if (mc->mc_allocator[i].mca_rotor == mg) mc->mc_allocator[i].mca_rotor = mgnext; } mg->mg_prev = NULL; mg->mg_next = NULL; } boolean_t metaslab_group_initialized(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; vdev_stat_t *vs = &vd->vdev_stat; return (vs->vs_space != 0 && mg->mg_activation_count > 0); } uint64_t metaslab_group_get_space(metaslab_group_t *mg) { /* * Note that the number of nodes in mg_metaslab_tree may be one less * than vdev_ms_count, due to the embedded log metaslab. */ mutex_enter(&mg->mg_lock); uint64_t ms_count = avl_numnodes(&mg->mg_metaslab_tree); mutex_exit(&mg->mg_lock); return ((1ULL << mg->mg_vd->vdev_ms_shift) * ms_count); } void metaslab_group_histogram_verify(metaslab_group_t *mg) { uint64_t *mg_hist; avl_tree_t *t = &mg->mg_metaslab_tree; uint64_t ashift = mg->mg_vd->vdev_ashift; if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0) return; mg_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE, KM_SLEEP); ASSERT3U(RANGE_TREE_HISTOGRAM_SIZE, >=, SPACE_MAP_HISTOGRAM_SIZE + ashift); mutex_enter(&mg->mg_lock); for (metaslab_t *msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) { VERIFY3P(msp->ms_group, ==, mg); /* skip if not active */ if (msp->ms_sm == NULL) continue; for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { mg_hist[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; } } for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i ++) VERIFY3U(mg_hist[i], ==, mg->mg_histogram[i]); mutex_exit(&mg->mg_lock); kmem_free(mg_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE); } static void metaslab_group_histogram_add(metaslab_group_t *mg, metaslab_t *msp) { metaslab_class_t *mc = mg->mg_class; uint64_t ashift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); if (msp->ms_sm == NULL) return; mutex_enter(&mg->mg_lock); + mutex_enter(&mc->mc_lock); for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { IMPLY(mg == mg->mg_vd->vdev_log_mg, mc == spa_embedded_log_class(mg->mg_vd->vdev_spa)); mg->mg_histogram[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; mc->mc_histogram[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; } + mutex_exit(&mc->mc_lock); mutex_exit(&mg->mg_lock); } void metaslab_group_histogram_remove(metaslab_group_t *mg, metaslab_t *msp) { metaslab_class_t *mc = mg->mg_class; uint64_t ashift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); if (msp->ms_sm == NULL) return; mutex_enter(&mg->mg_lock); + mutex_enter(&mc->mc_lock); for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { ASSERT3U(mg->mg_histogram[i + ashift], >=, msp->ms_sm->sm_phys->smp_histogram[i]); ASSERT3U(mc->mc_histogram[i + ashift], >=, msp->ms_sm->sm_phys->smp_histogram[i]); IMPLY(mg == mg->mg_vd->vdev_log_mg, mc == spa_embedded_log_class(mg->mg_vd->vdev_spa)); mg->mg_histogram[i + ashift] -= msp->ms_sm->sm_phys->smp_histogram[i]; mc->mc_histogram[i + ashift] -= msp->ms_sm->sm_phys->smp_histogram[i]; } + mutex_exit(&mc->mc_lock); mutex_exit(&mg->mg_lock); } static void metaslab_group_add(metaslab_group_t *mg, metaslab_t *msp) { ASSERT(msp->ms_group == NULL); mutex_enter(&mg->mg_lock); msp->ms_group = mg; msp->ms_weight = 0; avl_add(&mg->mg_metaslab_tree, msp); mutex_exit(&mg->mg_lock); mutex_enter(&msp->ms_lock); metaslab_group_histogram_add(mg, msp); mutex_exit(&msp->ms_lock); } static void metaslab_group_remove(metaslab_group_t *mg, metaslab_t *msp) { mutex_enter(&msp->ms_lock); metaslab_group_histogram_remove(mg, msp); mutex_exit(&msp->ms_lock); mutex_enter(&mg->mg_lock); ASSERT(msp->ms_group == mg); avl_remove(&mg->mg_metaslab_tree, msp); metaslab_class_t *mc = msp->ms_group->mg_class; multilist_sublist_t *mls = multilist_sublist_lock_obj(mc->mc_metaslab_txg_list, msp); if (multilist_link_active(&msp->ms_class_txg_node)) multilist_sublist_remove(mls, msp); multilist_sublist_unlock(mls); msp->ms_group = NULL; mutex_exit(&mg->mg_lock); } static void metaslab_group_sort_impl(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight) { ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(MUTEX_HELD(&mg->mg_lock)); ASSERT(msp->ms_group == mg); avl_remove(&mg->mg_metaslab_tree, msp); msp->ms_weight = weight; avl_add(&mg->mg_metaslab_tree, msp); } static void metaslab_group_sort(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight) { /* * Although in principle the weight can be any value, in * practice we do not use values in the range [1, 511]. */ ASSERT(weight >= SPA_MINBLOCKSIZE || weight == 0); ASSERT(MUTEX_HELD(&msp->ms_lock)); mutex_enter(&mg->mg_lock); metaslab_group_sort_impl(mg, msp, weight); mutex_exit(&mg->mg_lock); } /* * Calculate the fragmentation for a given metaslab group. We can use * a simple average here since all metaslabs within the group must have * the same size. The return value will be a value between 0 and 100 * (inclusive), or ZFS_FRAG_INVALID if less than half of the metaslab in this * group have a fragmentation metric. */ uint64_t metaslab_group_fragmentation(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; uint64_t fragmentation = 0; uint64_t valid_ms = 0; for (int m = 0; m < vd->vdev_ms_count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp->ms_fragmentation == ZFS_FRAG_INVALID) continue; if (msp->ms_group != mg) continue; valid_ms++; fragmentation += msp->ms_fragmentation; } if (valid_ms <= mg->mg_vd->vdev_ms_count / 2) return (ZFS_FRAG_INVALID); fragmentation /= valid_ms; ASSERT3U(fragmentation, <=, 100); return (fragmentation); } /* * Determine if a given metaslab group should skip allocations. A metaslab * group should avoid allocations if its free capacity is less than the * zfs_mg_noalloc_threshold or its fragmentation metric is greater than * zfs_mg_fragmentation_threshold and there is at least one metaslab group * that can still handle allocations. If the allocation throttle is enabled * then we skip allocations to devices that have reached their maximum * allocation queue depth unless the selected metaslab group is the only * eligible group remaining. */ static boolean_t metaslab_group_allocatable(metaslab_group_t *mg, metaslab_group_t *rotor, uint64_t psize, int allocator, int d) { spa_t *spa = mg->mg_vd->vdev_spa; metaslab_class_t *mc = mg->mg_class; /* * We can only consider skipping this metaslab group if it's * in the normal metaslab class and there are other metaslab * groups to select from. Otherwise, we always consider it eligible * for allocations. */ if ((mc != spa_normal_class(spa) && mc != spa_special_class(spa) && mc != spa_dedup_class(spa)) || mc->mc_groups <= 1) return (B_TRUE); /* * If the metaslab group's mg_allocatable flag is set (see comments * in metaslab_group_alloc_update() for more information) and * the allocation throttle is disabled then allow allocations to this * device. However, if the allocation throttle is enabled then * check if we have reached our allocation limit (mga_alloc_queue_depth) * to determine if we should allow allocations to this metaslab group. * If all metaslab groups are no longer considered allocatable * (mc_alloc_groups == 0) or we're trying to allocate the smallest * gang block size then we allow allocations on this metaslab group * regardless of the mg_allocatable or throttle settings. */ if (mg->mg_allocatable) { metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator]; int64_t qdepth; uint64_t qmax = mga->mga_cur_max_alloc_queue_depth; if (!mc->mc_alloc_throttle_enabled) return (B_TRUE); /* * If this metaslab group does not have any free space, then * there is no point in looking further. */ if (mg->mg_no_free_space) return (B_FALSE); /* * Relax allocation throttling for ditto blocks. Due to * random imbalances in allocation it tends to push copies * to one vdev, that looks a bit better at the moment. */ qmax = qmax * (4 + d) / 4; qdepth = zfs_refcount_count(&mga->mga_alloc_queue_depth); /* * If this metaslab group is below its qmax or it's * the only allocatable metasable group, then attempt * to allocate from it. */ if (qdepth < qmax || mc->mc_alloc_groups == 1) return (B_TRUE); ASSERT3U(mc->mc_alloc_groups, >, 1); /* * Since this metaslab group is at or over its qmax, we * need to determine if there are metaslab groups after this * one that might be able to handle this allocation. This is * racy since we can't hold the locks for all metaslab * groups at the same time when we make this check. */ for (metaslab_group_t *mgp = mg->mg_next; mgp != rotor; mgp = mgp->mg_next) { metaslab_group_allocator_t *mgap = &mgp->mg_allocator[allocator]; qmax = mgap->mga_cur_max_alloc_queue_depth; qmax = qmax * (4 + d) / 4; qdepth = zfs_refcount_count(&mgap->mga_alloc_queue_depth); /* * If there is another metaslab group that * might be able to handle the allocation, then * we return false so that we skip this group. */ if (qdepth < qmax && !mgp->mg_no_free_space) return (B_FALSE); } /* * We didn't find another group to handle the allocation * so we can't skip this metaslab group even though * we are at or over our qmax. */ return (B_TRUE); } else if (mc->mc_alloc_groups == 0 || psize == SPA_MINBLOCKSIZE) { return (B_TRUE); } return (B_FALSE); } /* * ========================================================================== * Range tree callbacks * ========================================================================== */ /* * Comparison function for the private size-ordered tree using 32-bit * ranges. Tree is sorted by size, larger sizes at the end of the tree. */ static int metaslab_rangesize32_compare(const void *x1, const void *x2) { const range_seg32_t *r1 = x1; const range_seg32_t *r2 = x2; uint64_t rs_size1 = r1->rs_end - r1->rs_start; uint64_t rs_size2 = r2->rs_end - r2->rs_start; int cmp = TREE_CMP(rs_size1, rs_size2); if (likely(cmp)) return (cmp); return (TREE_CMP(r1->rs_start, r2->rs_start)); } /* * Comparison function for the private size-ordered tree using 64-bit * ranges. Tree is sorted by size, larger sizes at the end of the tree. */ static int metaslab_rangesize64_compare(const void *x1, const void *x2) { const range_seg64_t *r1 = x1; const range_seg64_t *r2 = x2; uint64_t rs_size1 = r1->rs_end - r1->rs_start; uint64_t rs_size2 = r2->rs_end - r2->rs_start; int cmp = TREE_CMP(rs_size1, rs_size2); if (likely(cmp)) return (cmp); return (TREE_CMP(r1->rs_start, r2->rs_start)); } typedef struct metaslab_rt_arg { zfs_btree_t *mra_bt; uint32_t mra_floor_shift; } metaslab_rt_arg_t; struct mssa_arg { range_tree_t *rt; metaslab_rt_arg_t *mra; }; static void metaslab_size_sorted_add(void *arg, uint64_t start, uint64_t size) { struct mssa_arg *mssap = arg; range_tree_t *rt = mssap->rt; metaslab_rt_arg_t *mrap = mssap->mra; range_seg_max_t seg = {0}; rs_set_start(&seg, rt, start); rs_set_end(&seg, rt, start + size); metaslab_rt_add(rt, &seg, mrap); } static void metaslab_size_tree_full_load(range_tree_t *rt) { metaslab_rt_arg_t *mrap = rt->rt_arg; METASLABSTAT_BUMP(metaslabstat_reload_tree); ASSERT0(zfs_btree_numnodes(mrap->mra_bt)); mrap->mra_floor_shift = 0; struct mssa_arg arg = {0}; arg.rt = rt; arg.mra = mrap; range_tree_walk(rt, metaslab_size_sorted_add, &arg); } /* * Create any block allocator specific components. The current allocators * rely on using both a size-ordered range_tree_t and an array of uint64_t's. */ /* ARGSUSED */ static void metaslab_rt_create(range_tree_t *rt, void *arg) { metaslab_rt_arg_t *mrap = arg; zfs_btree_t *size_tree = mrap->mra_bt; size_t size; int (*compare) (const void *, const void *); switch (rt->rt_type) { case RANGE_SEG32: size = sizeof (range_seg32_t); compare = metaslab_rangesize32_compare; break; case RANGE_SEG64: size = sizeof (range_seg64_t); compare = metaslab_rangesize64_compare; break; default: panic("Invalid range seg type %d", rt->rt_type); } zfs_btree_create(size_tree, compare, size); mrap->mra_floor_shift = metaslab_by_size_min_shift; } /* ARGSUSED */ static void metaslab_rt_destroy(range_tree_t *rt, void *arg) { metaslab_rt_arg_t *mrap = arg; zfs_btree_t *size_tree = mrap->mra_bt; zfs_btree_destroy(size_tree); kmem_free(mrap, sizeof (*mrap)); } /* ARGSUSED */ static void metaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg) { metaslab_rt_arg_t *mrap = arg; zfs_btree_t *size_tree = mrap->mra_bt; if (rs_get_end(rs, rt) - rs_get_start(rs, rt) < (1 << mrap->mra_floor_shift)) return; zfs_btree_add(size_tree, rs); } /* ARGSUSED */ static void metaslab_rt_remove(range_tree_t *rt, range_seg_t *rs, void *arg) { metaslab_rt_arg_t *mrap = arg; zfs_btree_t *size_tree = mrap->mra_bt; if (rs_get_end(rs, rt) - rs_get_start(rs, rt) < (1 << mrap->mra_floor_shift)) return; zfs_btree_remove(size_tree, rs); } /* ARGSUSED */ static void metaslab_rt_vacate(range_tree_t *rt, void *arg) { metaslab_rt_arg_t *mrap = arg; zfs_btree_t *size_tree = mrap->mra_bt; zfs_btree_clear(size_tree); zfs_btree_destroy(size_tree); metaslab_rt_create(rt, arg); } static range_tree_ops_t metaslab_rt_ops = { .rtop_create = metaslab_rt_create, .rtop_destroy = metaslab_rt_destroy, .rtop_add = metaslab_rt_add, .rtop_remove = metaslab_rt_remove, .rtop_vacate = metaslab_rt_vacate }; /* * ========================================================================== * Common allocator routines * ========================================================================== */ /* * Return the maximum contiguous segment within the metaslab. */ uint64_t metaslab_largest_allocatable(metaslab_t *msp) { zfs_btree_t *t = &msp->ms_allocatable_by_size; range_seg_t *rs; if (t == NULL) return (0); if (zfs_btree_numnodes(t) == 0) metaslab_size_tree_full_load(msp->ms_allocatable); rs = zfs_btree_last(t, NULL); if (rs == NULL) return (0); return (rs_get_end(rs, msp->ms_allocatable) - rs_get_start(rs, msp->ms_allocatable)); } /* * Return the maximum contiguous segment within the unflushed frees of this * metaslab. */ static uint64_t metaslab_largest_unflushed_free(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); if (msp->ms_unflushed_frees == NULL) return (0); if (zfs_btree_numnodes(&msp->ms_unflushed_frees_by_size) == 0) metaslab_size_tree_full_load(msp->ms_unflushed_frees); range_seg_t *rs = zfs_btree_last(&msp->ms_unflushed_frees_by_size, NULL); if (rs == NULL) return (0); /* * When a range is freed from the metaslab, that range is added to * both the unflushed frees and the deferred frees. While the block * will eventually be usable, if the metaslab were loaded the range * would not be added to the ms_allocatable tree until TXG_DEFER_SIZE * txgs had passed. As a result, when attempting to estimate an upper * bound for the largest currently-usable free segment in the * metaslab, we need to not consider any ranges currently in the defer * trees. This algorithm approximates the largest available chunk in * the largest range in the unflushed_frees tree by taking the first * chunk. While this may be a poor estimate, it should only remain so * briefly and should eventually self-correct as frees are no longer * deferred. Similar logic applies to the ms_freed tree. See * metaslab_load() for more details. * * There are two primary sources of inaccuracy in this estimate. Both * are tolerated for performance reasons. The first source is that we * only check the largest segment for overlaps. Smaller segments may * have more favorable overlaps with the other trees, resulting in * larger usable chunks. Second, we only look at the first chunk in * the largest segment; there may be other usable chunks in the * largest segment, but we ignore them. */ uint64_t rstart = rs_get_start(rs, msp->ms_unflushed_frees); uint64_t rsize = rs_get_end(rs, msp->ms_unflushed_frees) - rstart; for (int t = 0; t < TXG_DEFER_SIZE; t++) { uint64_t start = 0; uint64_t size = 0; boolean_t found = range_tree_find_in(msp->ms_defer[t], rstart, rsize, &start, &size); if (found) { if (rstart == start) return (0); rsize = start - rstart; } } uint64_t start = 0; uint64_t size = 0; boolean_t found = range_tree_find_in(msp->ms_freed, rstart, rsize, &start, &size); if (found) rsize = start - rstart; return (rsize); } static range_seg_t * metaslab_block_find(zfs_btree_t *t, range_tree_t *rt, uint64_t start, uint64_t size, zfs_btree_index_t *where) { range_seg_t *rs; range_seg_max_t rsearch; rs_set_start(&rsearch, rt, start); rs_set_end(&rsearch, rt, start + size); rs = zfs_btree_find(t, &rsearch, where); if (rs == NULL) { rs = zfs_btree_next(t, where, where); } return (rs); } #if defined(WITH_DF_BLOCK_ALLOCATOR) || \ defined(WITH_CF_BLOCK_ALLOCATOR) /* * This is a helper function that can be used by the allocator to find a * suitable block to allocate. This will search the specified B-tree looking * for a block that matches the specified criteria. */ static uint64_t metaslab_block_picker(range_tree_t *rt, uint64_t *cursor, uint64_t size, uint64_t max_search) { if (*cursor == 0) *cursor = rt->rt_start; zfs_btree_t *bt = &rt->rt_root; zfs_btree_index_t where; range_seg_t *rs = metaslab_block_find(bt, rt, *cursor, size, &where); uint64_t first_found; int count_searched = 0; if (rs != NULL) first_found = rs_get_start(rs, rt); while (rs != NULL && (rs_get_start(rs, rt) - first_found <= max_search || count_searched < metaslab_min_search_count)) { uint64_t offset = rs_get_start(rs, rt); if (offset + size <= rs_get_end(rs, rt)) { *cursor = offset + size; return (offset); } rs = zfs_btree_next(bt, &where, &where); count_searched++; } *cursor = 0; return (-1ULL); } #endif /* WITH_DF/CF_BLOCK_ALLOCATOR */ #if defined(WITH_DF_BLOCK_ALLOCATOR) /* * ========================================================================== * Dynamic Fit (df) block allocator * * Search for a free chunk of at least this size, starting from the last * offset (for this alignment of block) looking for up to * metaslab_df_max_search bytes (16MB). If a large enough free chunk is not * found within 16MB, then return a free chunk of exactly the requested size (or * larger). * * If it seems like searching from the last offset will be unproductive, skip * that and just return a free chunk of exactly the requested size (or larger). * This is based on metaslab_df_alloc_threshold and metaslab_df_free_pct. This * mechanism is probably not very useful and may be removed in the future. * * The behavior when not searching can be changed to return the largest free * chunk, instead of a free chunk of exactly the requested size, by setting * metaslab_df_use_largest_segment. * ========================================================================== */ static uint64_t metaslab_df_alloc(metaslab_t *msp, uint64_t size) { /* * Find the largest power of 2 block size that evenly divides the * requested size. This is used to try to allocate blocks with similar * alignment from the same area of the metaslab (i.e. same cursor * bucket) but it does not guarantee that other allocations sizes * may exist in the same region. */ uint64_t align = size & -size; uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1]; range_tree_t *rt = msp->ms_allocatable; int free_pct = range_tree_space(rt) * 100 / msp->ms_size; uint64_t offset; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * If we're running low on space, find a segment based on size, * rather than iterating based on offset. */ if (metaslab_largest_allocatable(msp) < metaslab_df_alloc_threshold || free_pct < metaslab_df_free_pct) { offset = -1; } else { offset = metaslab_block_picker(rt, cursor, size, metaslab_df_max_search); } if (offset == -1) { range_seg_t *rs; if (zfs_btree_numnodes(&msp->ms_allocatable_by_size) == 0) metaslab_size_tree_full_load(msp->ms_allocatable); if (metaslab_df_use_largest_segment) { /* use largest free segment */ rs = zfs_btree_last(&msp->ms_allocatable_by_size, NULL); } else { zfs_btree_index_t where; /* use segment of this size, or next largest */ rs = metaslab_block_find(&msp->ms_allocatable_by_size, rt, msp->ms_start, size, &where); } if (rs != NULL && rs_get_start(rs, rt) + size <= rs_get_end(rs, rt)) { offset = rs_get_start(rs, rt); *cursor = offset + size; } } return (offset); } static metaslab_ops_t metaslab_df_ops = { metaslab_df_alloc }; metaslab_ops_t *zfs_metaslab_ops = &metaslab_df_ops; #endif /* WITH_DF_BLOCK_ALLOCATOR */ #if defined(WITH_CF_BLOCK_ALLOCATOR) /* * ========================================================================== * Cursor fit block allocator - * Select the largest region in the metaslab, set the cursor to the beginning * of the range and the cursor_end to the end of the range. As allocations * are made advance the cursor. Continue allocating from the cursor until * the range is exhausted and then find a new range. * ========================================================================== */ static uint64_t metaslab_cf_alloc(metaslab_t *msp, uint64_t size) { range_tree_t *rt = msp->ms_allocatable; zfs_btree_t *t = &msp->ms_allocatable_by_size; uint64_t *cursor = &msp->ms_lbas[0]; uint64_t *cursor_end = &msp->ms_lbas[1]; uint64_t offset = 0; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(*cursor_end, >=, *cursor); if ((*cursor + size) > *cursor_end) { range_seg_t *rs; if (zfs_btree_numnodes(t) == 0) metaslab_size_tree_full_load(msp->ms_allocatable); rs = zfs_btree_last(t, NULL); if (rs == NULL || (rs_get_end(rs, rt) - rs_get_start(rs, rt)) < size) return (-1ULL); *cursor = rs_get_start(rs, rt); *cursor_end = rs_get_end(rs, rt); } offset = *cursor; *cursor += size; return (offset); } static metaslab_ops_t metaslab_cf_ops = { metaslab_cf_alloc }; metaslab_ops_t *zfs_metaslab_ops = &metaslab_cf_ops; #endif /* WITH_CF_BLOCK_ALLOCATOR */ #if defined(WITH_NDF_BLOCK_ALLOCATOR) /* * ========================================================================== * New dynamic fit allocator - * Select a region that is large enough to allocate 2^metaslab_ndf_clump_shift * contiguous blocks. If no region is found then just use the largest segment * that remains. * ========================================================================== */ /* * Determines desired number of contiguous blocks (2^metaslab_ndf_clump_shift) * to request from the allocator. */ uint64_t metaslab_ndf_clump_shift = 4; static uint64_t metaslab_ndf_alloc(metaslab_t *msp, uint64_t size) { zfs_btree_t *t = &msp->ms_allocatable->rt_root; range_tree_t *rt = msp->ms_allocatable; zfs_btree_index_t where; range_seg_t *rs; range_seg_max_t rsearch; uint64_t hbit = highbit64(size); uint64_t *cursor = &msp->ms_lbas[hbit - 1]; uint64_t max_size = metaslab_largest_allocatable(msp); ASSERT(MUTEX_HELD(&msp->ms_lock)); if (max_size < size) return (-1ULL); rs_set_start(&rsearch, rt, *cursor); rs_set_end(&rsearch, rt, *cursor + size); rs = zfs_btree_find(t, &rsearch, &where); if (rs == NULL || (rs_get_end(rs, rt) - rs_get_start(rs, rt)) < size) { t = &msp->ms_allocatable_by_size; rs_set_start(&rsearch, rt, 0); rs_set_end(&rsearch, rt, MIN(max_size, 1ULL << (hbit + metaslab_ndf_clump_shift))); rs = zfs_btree_find(t, &rsearch, &where); if (rs == NULL) rs = zfs_btree_next(t, &where, &where); ASSERT(rs != NULL); } if ((rs_get_end(rs, rt) - rs_get_start(rs, rt)) >= size) { *cursor = rs_get_start(rs, rt) + size; return (rs_get_start(rs, rt)); } return (-1ULL); } static metaslab_ops_t metaslab_ndf_ops = { metaslab_ndf_alloc }; metaslab_ops_t *zfs_metaslab_ops = &metaslab_ndf_ops; #endif /* WITH_NDF_BLOCK_ALLOCATOR */ /* * ========================================================================== * Metaslabs * ========================================================================== */ /* * Wait for any in-progress metaslab loads to complete. */ static void metaslab_load_wait(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); while (msp->ms_loading) { ASSERT(!msp->ms_loaded); cv_wait(&msp->ms_load_cv, &msp->ms_lock); } } /* * Wait for any in-progress flushing to complete. */ static void metaslab_flush_wait(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); while (msp->ms_flushing) cv_wait(&msp->ms_flush_cv, &msp->ms_lock); } static unsigned int metaslab_idx_func(multilist_t *ml, void *arg) { metaslab_t *msp = arg; return (msp->ms_id % multilist_get_num_sublists(ml)); } uint64_t metaslab_allocated_space(metaslab_t *msp) { return (msp->ms_allocated_space); } /* * Verify that the space accounting on disk matches the in-core range_trees. */ static void metaslab_verify_space(metaslab_t *msp, uint64_t txg) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; uint64_t allocating = 0; uint64_t sm_free_space, msp_free_space; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(!msp->ms_condensing); if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0) return; /* * We can only verify the metaslab space when we're called * from syncing context with a loaded metaslab that has an * allocated space map. Calling this in non-syncing context * does not provide a consistent view of the metaslab since * we're performing allocations in the future. */ if (txg != spa_syncing_txg(spa) || msp->ms_sm == NULL || !msp->ms_loaded) return; /* * Even though the smp_alloc field can get negative, * when it comes to a metaslab's space map, that should * never be the case. */ ASSERT3S(space_map_allocated(msp->ms_sm), >=, 0); ASSERT3U(space_map_allocated(msp->ms_sm), >=, range_tree_space(msp->ms_unflushed_frees)); ASSERT3U(metaslab_allocated_space(msp), ==, space_map_allocated(msp->ms_sm) + range_tree_space(msp->ms_unflushed_allocs) - range_tree_space(msp->ms_unflushed_frees)); sm_free_space = msp->ms_size - metaslab_allocated_space(msp); /* * Account for future allocations since we would have * already deducted that space from the ms_allocatable. */ for (int t = 0; t < TXG_CONCURRENT_STATES; t++) { allocating += range_tree_space(msp->ms_allocating[(txg + t) & TXG_MASK]); } ASSERT3U(allocating + msp->ms_allocated_this_txg, ==, msp->ms_allocating_total); ASSERT3U(msp->ms_deferspace, ==, range_tree_space(msp->ms_defer[0]) + range_tree_space(msp->ms_defer[1])); msp_free_space = range_tree_space(msp->ms_allocatable) + allocating + msp->ms_deferspace + range_tree_space(msp->ms_freed); VERIFY3U(sm_free_space, ==, msp_free_space); } static void metaslab_aux_histograms_clear(metaslab_t *msp) { /* * Auxiliary histograms are only cleared when resetting them, * which can only happen while the metaslab is loaded. */ ASSERT(msp->ms_loaded); bzero(msp->ms_synchist, sizeof (msp->ms_synchist)); for (int t = 0; t < TXG_DEFER_SIZE; t++) bzero(msp->ms_deferhist[t], sizeof (msp->ms_deferhist[t])); } static void metaslab_aux_histogram_add(uint64_t *histogram, uint64_t shift, range_tree_t *rt) { /* * This is modeled after space_map_histogram_add(), so refer to that * function for implementation details. We want this to work like * the space map histogram, and not the range tree histogram, as we * are essentially constructing a delta that will be later subtracted * from the space map histogram. */ int idx = 0; for (int i = shift; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { ASSERT3U(i, >=, idx + shift); histogram[idx] += rt->rt_histogram[i] << (i - idx - shift); if (idx < SPACE_MAP_HISTOGRAM_SIZE - 1) { ASSERT3U(idx + shift, ==, i); idx++; ASSERT3U(idx, <, SPACE_MAP_HISTOGRAM_SIZE); } } } /* * Called at every sync pass that the metaslab gets synced. * * The reason is that we want our auxiliary histograms to be updated * wherever the metaslab's space map histogram is updated. This way * we stay consistent on which parts of the metaslab space map's * histogram are currently not available for allocations (e.g because * they are in the defer, freed, and freeing trees). */ static void metaslab_aux_histograms_update(metaslab_t *msp) { space_map_t *sm = msp->ms_sm; ASSERT(sm != NULL); /* * This is similar to the metaslab's space map histogram updates * that take place in metaslab_sync(). The only difference is that * we only care about segments that haven't made it into the * ms_allocatable tree yet. */ if (msp->ms_loaded) { metaslab_aux_histograms_clear(msp); metaslab_aux_histogram_add(msp->ms_synchist, sm->sm_shift, msp->ms_freed); for (int t = 0; t < TXG_DEFER_SIZE; t++) { metaslab_aux_histogram_add(msp->ms_deferhist[t], sm->sm_shift, msp->ms_defer[t]); } } metaslab_aux_histogram_add(msp->ms_synchist, sm->sm_shift, msp->ms_freeing); } /* * Called every time we are done syncing (writing to) the metaslab, * i.e. at the end of each sync pass. * [see the comment in metaslab_impl.h for ms_synchist, ms_deferhist] */ static void metaslab_aux_histograms_update_done(metaslab_t *msp, boolean_t defer_allowed) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; space_map_t *sm = msp->ms_sm; if (sm == NULL) { /* * We came here from metaslab_init() when creating/opening a * pool, looking at a metaslab that hasn't had any allocations * yet. */ return; } /* * This is similar to the actions that we take for the ms_freed * and ms_defer trees in metaslab_sync_done(). */ uint64_t hist_index = spa_syncing_txg(spa) % TXG_DEFER_SIZE; if (defer_allowed) { bcopy(msp->ms_synchist, msp->ms_deferhist[hist_index], sizeof (msp->ms_synchist)); } else { bzero(msp->ms_deferhist[hist_index], sizeof (msp->ms_deferhist[hist_index])); } bzero(msp->ms_synchist, sizeof (msp->ms_synchist)); } /* * Ensure that the metaslab's weight and fragmentation are consistent * with the contents of the histogram (either the range tree's histogram * or the space map's depending whether the metaslab is loaded). */ static void metaslab_verify_weight_and_frag(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0) return; /* * We can end up here from vdev_remove_complete(), in which case we * cannot do these assertions because we hold spa config locks and * thus we are not allowed to read from the DMU. * * We check if the metaslab group has been removed and if that's * the case we return immediately as that would mean that we are * here from the aforementioned code path. */ if (msp->ms_group == NULL) return; /* * Devices being removed always return a weight of 0 and leave * fragmentation and ms_max_size as is - there is nothing for * us to verify here. */ vdev_t *vd = msp->ms_group->mg_vd; if (vd->vdev_removing) return; /* * If the metaslab is dirty it probably means that we've done * some allocations or frees that have changed our histograms * and thus the weight. */ for (int t = 0; t < TXG_SIZE; t++) { if (txg_list_member(&vd->vdev_ms_list, msp, t)) return; } /* * This verification checks that our in-memory state is consistent * with what's on disk. If the pool is read-only then there aren't * any changes and we just have the initially-loaded state. */ if (!spa_writeable(msp->ms_group->mg_vd->vdev_spa)) return; /* some extra verification for in-core tree if you can */ if (msp->ms_loaded) { range_tree_stat_verify(msp->ms_allocatable); VERIFY(space_map_histogram_verify(msp->ms_sm, msp->ms_allocatable)); } uint64_t weight = msp->ms_weight; uint64_t was_active = msp->ms_weight & METASLAB_ACTIVE_MASK; boolean_t space_based = WEIGHT_IS_SPACEBASED(msp->ms_weight); uint64_t frag = msp->ms_fragmentation; uint64_t max_segsize = msp->ms_max_size; msp->ms_weight = 0; msp->ms_fragmentation = 0; /* * This function is used for verification purposes and thus should * not introduce any side-effects/mutations on the system's state. * * Regardless of whether metaslab_weight() thinks this metaslab * should be active or not, we want to ensure that the actual weight * (and therefore the value of ms_weight) would be the same if it * was to be recalculated at this point. * * In addition we set the nodirty flag so metaslab_weight() does * not dirty the metaslab for future TXGs (e.g. when trying to * force condensing to upgrade the metaslab spacemaps). */ msp->ms_weight = metaslab_weight(msp, B_TRUE) | was_active; VERIFY3U(max_segsize, ==, msp->ms_max_size); /* * If the weight type changed then there is no point in doing * verification. Revert fields to their original values. */ if ((space_based && !WEIGHT_IS_SPACEBASED(msp->ms_weight)) || (!space_based && WEIGHT_IS_SPACEBASED(msp->ms_weight))) { msp->ms_fragmentation = frag; msp->ms_weight = weight; return; } VERIFY3U(msp->ms_fragmentation, ==, frag); VERIFY3U(msp->ms_weight, ==, weight); } /* * If we're over the zfs_metaslab_mem_limit, select the loaded metaslab from * this class that was used longest ago, and attempt to unload it. We don't * want to spend too much time in this loop to prevent performance * degradation, and we expect that most of the time this operation will * succeed. Between that and the normal unloading processing during txg sync, * we expect this to keep the metaslab memory usage under control. */ static void metaslab_potentially_evict(metaslab_class_t *mc) { #ifdef _KERNEL uint64_t allmem = arc_all_memory(); uint64_t inuse = spl_kmem_cache_inuse(zfs_btree_leaf_cache); uint64_t size = spl_kmem_cache_entry_size(zfs_btree_leaf_cache); int tries = 0; for (; allmem * zfs_metaslab_mem_limit / 100 < inuse * size && tries < multilist_get_num_sublists(mc->mc_metaslab_txg_list) * 2; tries++) { unsigned int idx = multilist_get_random_index( mc->mc_metaslab_txg_list); multilist_sublist_t *mls = multilist_sublist_lock(mc->mc_metaslab_txg_list, idx); metaslab_t *msp = multilist_sublist_head(mls); multilist_sublist_unlock(mls); while (msp != NULL && allmem * zfs_metaslab_mem_limit / 100 < inuse * size) { VERIFY3P(mls, ==, multilist_sublist_lock( mc->mc_metaslab_txg_list, idx)); ASSERT3U(idx, ==, metaslab_idx_func(mc->mc_metaslab_txg_list, msp)); if (!multilist_link_active(&msp->ms_class_txg_node)) { multilist_sublist_unlock(mls); break; } metaslab_t *next_msp = multilist_sublist_next(mls, msp); multilist_sublist_unlock(mls); /* * If the metaslab is currently loading there are two * cases. If it's the metaslab we're evicting, we * can't continue on or we'll panic when we attempt to * recursively lock the mutex. If it's another * metaslab that's loading, it can be safely skipped, * since we know it's very new and therefore not a * good eviction candidate. We check later once the * lock is held that the metaslab is fully loaded * before actually unloading it. */ if (msp->ms_loading) { msp = next_msp; inuse = spl_kmem_cache_inuse(zfs_btree_leaf_cache); continue; } /* * We can't unload metaslabs with no spacemap because * they're not ready to be unloaded yet. We can't * unload metaslabs with outstanding allocations * because doing so could cause the metaslab's weight * to decrease while it's unloaded, which violates an * invariant that we use to prevent unnecessary * loading. We also don't unload metaslabs that are * currently active because they are high-weight * metaslabs that are likely to be used in the near * future. */ mutex_enter(&msp->ms_lock); if (msp->ms_allocator == -1 && msp->ms_sm != NULL && msp->ms_allocating_total == 0) { metaslab_unload(msp); } mutex_exit(&msp->ms_lock); msp = next_msp; inuse = spl_kmem_cache_inuse(zfs_btree_leaf_cache); } } #endif } static int metaslab_load_impl(metaslab_t *msp) { int error = 0; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(msp->ms_loading); ASSERT(!msp->ms_condensing); /* * We temporarily drop the lock to unblock other operations while we * are reading the space map. Therefore, metaslab_sync() and * metaslab_sync_done() can run at the same time as we do. * * If we are using the log space maps, metaslab_sync() can't write to * the metaslab's space map while we are loading as we only write to * it when we are flushing the metaslab, and that can't happen while * we are loading it. * * If we are not using log space maps though, metaslab_sync() can * append to the space map while we are loading. Therefore we load * only entries that existed when we started the load. Additionally, * metaslab_sync_done() has to wait for the load to complete because * there are potential races like metaslab_load() loading parts of the * space map that are currently being appended by metaslab_sync(). If * we didn't, the ms_allocatable would have entries that * metaslab_sync_done() would try to re-add later. * * That's why before dropping the lock we remember the synced length * of the metaslab and read up to that point of the space map, * ignoring entries appended by metaslab_sync() that happen after we * drop the lock. */ uint64_t length = msp->ms_synced_length; mutex_exit(&msp->ms_lock); hrtime_t load_start = gethrtime(); metaslab_rt_arg_t *mrap; if (msp->ms_allocatable->rt_arg == NULL) { mrap = kmem_zalloc(sizeof (*mrap), KM_SLEEP); } else { mrap = msp->ms_allocatable->rt_arg; msp->ms_allocatable->rt_ops = NULL; msp->ms_allocatable->rt_arg = NULL; } mrap->mra_bt = &msp->ms_allocatable_by_size; mrap->mra_floor_shift = metaslab_by_size_min_shift; if (msp->ms_sm != NULL) { error = space_map_load_length(msp->ms_sm, msp->ms_allocatable, SM_FREE, length); /* Now, populate the size-sorted tree. */ metaslab_rt_create(msp->ms_allocatable, mrap); msp->ms_allocatable->rt_ops = &metaslab_rt_ops; msp->ms_allocatable->rt_arg = mrap; struct mssa_arg arg = {0}; arg.rt = msp->ms_allocatable; arg.mra = mrap; range_tree_walk(msp->ms_allocatable, metaslab_size_sorted_add, &arg); } else { /* * Add the size-sorted tree first, since we don't need to load * the metaslab from the spacemap. */ metaslab_rt_create(msp->ms_allocatable, mrap); msp->ms_allocatable->rt_ops = &metaslab_rt_ops; msp->ms_allocatable->rt_arg = mrap; /* * The space map has not been allocated yet, so treat * all the space in the metaslab as free and add it to the * ms_allocatable tree. */ range_tree_add(msp->ms_allocatable, msp->ms_start, msp->ms_size); if (msp->ms_freed != NULL) { /* * If the ms_sm doesn't exist, this means that this * metaslab hasn't gone through metaslab_sync() and * thus has never been dirtied. So we shouldn't * expect any unflushed allocs or frees from previous * TXGs. * * Note: ms_freed and all the other trees except for * the ms_allocatable, can be NULL at this point only * if this is a new metaslab of a vdev that just got * expanded. */ ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs)); ASSERT(range_tree_is_empty(msp->ms_unflushed_frees)); } } /* * We need to grab the ms_sync_lock to prevent metaslab_sync() from * changing the ms_sm (or log_sm) and the metaslab's range trees * while we are about to use them and populate the ms_allocatable. * The ms_lock is insufficient for this because metaslab_sync() doesn't * hold the ms_lock while writing the ms_checkpointing tree to disk. */ mutex_enter(&msp->ms_sync_lock); mutex_enter(&msp->ms_lock); ASSERT(!msp->ms_condensing); ASSERT(!msp->ms_flushing); if (error != 0) { mutex_exit(&msp->ms_sync_lock); return (error); } ASSERT3P(msp->ms_group, !=, NULL); msp->ms_loaded = B_TRUE; /* * Apply all the unflushed changes to ms_allocatable right * away so any manipulations we do below have a clear view * of what is allocated and what is free. */ range_tree_walk(msp->ms_unflushed_allocs, range_tree_remove, msp->ms_allocatable); range_tree_walk(msp->ms_unflushed_frees, range_tree_add, msp->ms_allocatable); msp->ms_loaded = B_TRUE; ASSERT3P(msp->ms_group, !=, NULL); spa_t *spa = msp->ms_group->mg_vd->vdev_spa; if (spa_syncing_log_sm(spa) != NULL) { ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP)); /* * If we use a log space map we add all the segments * that are in ms_unflushed_frees so they are available * for allocation. * * ms_allocatable needs to contain all free segments * that are ready for allocations (thus not segments * from ms_freeing, ms_freed, and the ms_defer trees). * But if we grab the lock in this code path at a sync * pass later that 1, then it also contains the * segments of ms_freed (they were added to it earlier * in this path through ms_unflushed_frees). So we * need to remove all the segments that exist in * ms_freed from ms_allocatable as they will be added * later in metaslab_sync_done(). * * When there's no log space map, the ms_allocatable * correctly doesn't contain any segments that exist * in ms_freed [see ms_synced_length]. */ range_tree_walk(msp->ms_freed, range_tree_remove, msp->ms_allocatable); } /* * If we are not using the log space map, ms_allocatable * contains the segments that exist in the ms_defer trees * [see ms_synced_length]. Thus we need to remove them * from ms_allocatable as they will be added again in * metaslab_sync_done(). * * If we are using the log space map, ms_allocatable still * contains the segments that exist in the ms_defer trees. * Not because it read them through the ms_sm though. But * because these segments are part of ms_unflushed_frees * whose segments we add to ms_allocatable earlier in this * code path. */ for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_walk(msp->ms_defer[t], range_tree_remove, msp->ms_allocatable); } /* * Call metaslab_recalculate_weight_and_sort() now that the * metaslab is loaded so we get the metaslab's real weight. * * Unless this metaslab was created with older software and * has not yet been converted to use segment-based weight, we * expect the new weight to be better or equal to the weight * that the metaslab had while it was not loaded. This is * because the old weight does not take into account the * consolidation of adjacent segments between TXGs. [see * comment for ms_synchist and ms_deferhist[] for more info] */ uint64_t weight = msp->ms_weight; uint64_t max_size = msp->ms_max_size; metaslab_recalculate_weight_and_sort(msp); if (!WEIGHT_IS_SPACEBASED(weight)) ASSERT3U(weight, <=, msp->ms_weight); msp->ms_max_size = metaslab_largest_allocatable(msp); ASSERT3U(max_size, <=, msp->ms_max_size); hrtime_t load_end = gethrtime(); msp->ms_load_time = load_end; zfs_dbgmsg("metaslab_load: txg %llu, spa %s, vdev_id %llu, " "ms_id %llu, smp_length %llu, " "unflushed_allocs %llu, unflushed_frees %llu, " "freed %llu, defer %llu + %llu, unloaded time %llu ms, " "loading_time %lld ms, ms_max_size %llu, " "max size error %lld, " "old_weight %llx, new_weight %llx", spa_syncing_txg(spa), spa_name(spa), msp->ms_group->mg_vd->vdev_id, msp->ms_id, space_map_length(msp->ms_sm), range_tree_space(msp->ms_unflushed_allocs), range_tree_space(msp->ms_unflushed_frees), range_tree_space(msp->ms_freed), range_tree_space(msp->ms_defer[0]), range_tree_space(msp->ms_defer[1]), (longlong_t)((load_start - msp->ms_unload_time) / 1000000), (longlong_t)((load_end - load_start) / 1000000), msp->ms_max_size, msp->ms_max_size - max_size, weight, msp->ms_weight); metaslab_verify_space(msp, spa_syncing_txg(spa)); mutex_exit(&msp->ms_sync_lock); return (0); } int metaslab_load(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * There may be another thread loading the same metaslab, if that's * the case just wait until the other thread is done and return. */ metaslab_load_wait(msp); if (msp->ms_loaded) return (0); VERIFY(!msp->ms_loading); ASSERT(!msp->ms_condensing); /* * We set the loading flag BEFORE potentially dropping the lock to * wait for an ongoing flush (see ms_flushing below). This way other * threads know that there is already a thread that is loading this * metaslab. */ msp->ms_loading = B_TRUE; /* * Wait for any in-progress flushing to finish as we drop the ms_lock * both here (during space_map_load()) and in metaslab_flush() (when * we flush our changes to the ms_sm). */ if (msp->ms_flushing) metaslab_flush_wait(msp); /* * In the possibility that we were waiting for the metaslab to be * flushed (where we temporarily dropped the ms_lock), ensure that * no one else loaded the metaslab somehow. */ ASSERT(!msp->ms_loaded); /* * If we're loading a metaslab in the normal class, consider evicting * another one to keep our memory usage under the limit defined by the * zfs_metaslab_mem_limit tunable. */ if (spa_normal_class(msp->ms_group->mg_class->mc_spa) == msp->ms_group->mg_class) { metaslab_potentially_evict(msp->ms_group->mg_class); } int error = metaslab_load_impl(msp); ASSERT(MUTEX_HELD(&msp->ms_lock)); msp->ms_loading = B_FALSE; cv_broadcast(&msp->ms_load_cv); return (error); } void metaslab_unload(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * This can happen if a metaslab is selected for eviction (in * metaslab_potentially_evict) and then unloaded during spa_sync (via * metaslab_class_evict_old). */ if (!msp->ms_loaded) return; range_tree_vacate(msp->ms_allocatable, NULL, NULL); msp->ms_loaded = B_FALSE; msp->ms_unload_time = gethrtime(); msp->ms_activation_weight = 0; msp->ms_weight &= ~METASLAB_ACTIVE_MASK; if (msp->ms_group != NULL) { metaslab_class_t *mc = msp->ms_group->mg_class; multilist_sublist_t *mls = multilist_sublist_lock_obj(mc->mc_metaslab_txg_list, msp); if (multilist_link_active(&msp->ms_class_txg_node)) multilist_sublist_remove(mls, msp); multilist_sublist_unlock(mls); spa_t *spa = msp->ms_group->mg_vd->vdev_spa; zfs_dbgmsg("metaslab_unload: txg %llu, spa %s, vdev_id %llu, " "ms_id %llu, weight %llx, " "selected txg %llu (%llu ms ago), alloc_txg %llu, " "loaded %llu ms ago, max_size %llu", spa_syncing_txg(spa), spa_name(spa), msp->ms_group->mg_vd->vdev_id, msp->ms_id, msp->ms_weight, msp->ms_selected_txg, (msp->ms_unload_time - msp->ms_selected_time) / 1000 / 1000, msp->ms_alloc_txg, (msp->ms_unload_time - msp->ms_load_time) / 1000 / 1000, msp->ms_max_size); } /* * We explicitly recalculate the metaslab's weight based on its space * map (as it is now not loaded). We want unload metaslabs to always * have their weights calculated from the space map histograms, while * loaded ones have it calculated from their in-core range tree * [see metaslab_load()]. This way, the weight reflects the information * available in-core, whether it is loaded or not. * * If ms_group == NULL means that we came here from metaslab_fini(), * at which point it doesn't make sense for us to do the recalculation * and the sorting. */ if (msp->ms_group != NULL) metaslab_recalculate_weight_and_sort(msp); } /* * We want to optimize the memory use of the per-metaslab range * trees. To do this, we store the segments in the range trees in * units of sectors, zero-indexing from the start of the metaslab. If * the vdev_ms_shift - the vdev_ashift is less than 32, we can store * the ranges using two uint32_ts, rather than two uint64_ts. */ range_seg_type_t metaslab_calculate_range_tree_type(vdev_t *vdev, metaslab_t *msp, uint64_t *start, uint64_t *shift) { if (vdev->vdev_ms_shift - vdev->vdev_ashift < 32 && !zfs_metaslab_force_large_segs) { *shift = vdev->vdev_ashift; *start = msp->ms_start; return (RANGE_SEG32); } else { *shift = 0; *start = 0; return (RANGE_SEG64); } } void metaslab_set_selected_txg(metaslab_t *msp, uint64_t txg) { ASSERT(MUTEX_HELD(&msp->ms_lock)); metaslab_class_t *mc = msp->ms_group->mg_class; multilist_sublist_t *mls = multilist_sublist_lock_obj(mc->mc_metaslab_txg_list, msp); if (multilist_link_active(&msp->ms_class_txg_node)) multilist_sublist_remove(mls, msp); msp->ms_selected_txg = txg; msp->ms_selected_time = gethrtime(); multilist_sublist_insert_tail(mls, msp); multilist_sublist_unlock(mls); } void metaslab_space_update(vdev_t *vd, metaslab_class_t *mc, int64_t alloc_delta, int64_t defer_delta, int64_t space_delta) { vdev_space_update(vd, alloc_delta, defer_delta, space_delta); ASSERT3P(vd->vdev_spa->spa_root_vdev, ==, vd->vdev_parent); ASSERT(vd->vdev_ms_count != 0); metaslab_class_space_update(mc, alloc_delta, defer_delta, space_delta, vdev_deflated_space(vd, space_delta)); } int metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg, metaslab_t **msp) { vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; objset_t *mos = spa->spa_meta_objset; metaslab_t *ms; int error; ms = kmem_zalloc(sizeof (metaslab_t), KM_SLEEP); mutex_init(&ms->ms_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ms->ms_sync_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&ms->ms_load_cv, NULL, CV_DEFAULT, NULL); cv_init(&ms->ms_flush_cv, NULL, CV_DEFAULT, NULL); multilist_link_init(&ms->ms_class_txg_node); ms->ms_id = id; ms->ms_start = id << vd->vdev_ms_shift; ms->ms_size = 1ULL << vd->vdev_ms_shift; ms->ms_allocator = -1; ms->ms_new = B_TRUE; vdev_ops_t *ops = vd->vdev_ops; if (ops->vdev_op_metaslab_init != NULL) ops->vdev_op_metaslab_init(vd, &ms->ms_start, &ms->ms_size); /* * We only open space map objects that already exist. All others * will be opened when we finally allocate an object for it. * * Note: * When called from vdev_expand(), we can't call into the DMU as * we are holding the spa_config_lock as a writer and we would * deadlock [see relevant comment in vdev_metaslab_init()]. in * that case, the object parameter is zero though, so we won't * call into the DMU. */ if (object != 0) { error = space_map_open(&ms->ms_sm, mos, object, ms->ms_start, ms->ms_size, vd->vdev_ashift); if (error != 0) { kmem_free(ms, sizeof (metaslab_t)); return (error); } ASSERT(ms->ms_sm != NULL); ms->ms_allocated_space = space_map_allocated(ms->ms_sm); } range_seg_type_t type; uint64_t shift, start; type = metaslab_calculate_range_tree_type(vd, ms, &start, &shift); /* * We create the ms_allocatable here, but we don't create the * other range trees until metaslab_sync_done(). This serves * two purposes: it allows metaslab_sync_done() to detect the * addition of new space; and for debugging, it ensures that * we'd data fault on any attempt to use this metaslab before * it's ready. */ ms->ms_allocatable = range_tree_create(NULL, type, NULL, start, shift); ms->ms_trim = range_tree_create(NULL, type, NULL, start, shift); metaslab_group_add(mg, ms); metaslab_set_fragmentation(ms, B_FALSE); /* * If we're opening an existing pool (txg == 0) or creating * a new one (txg == TXG_INITIAL), all space is available now. * If we're adding space to an existing pool, the new space * does not become available until after this txg has synced. * The metaslab's weight will also be initialized when we sync * out this txg. This ensures that we don't attempt to allocate * from it before we have initialized it completely. */ if (txg <= TXG_INITIAL) { metaslab_sync_done(ms, 0); metaslab_space_update(vd, mg->mg_class, metaslab_allocated_space(ms), 0, 0); } if (txg != 0) { vdev_dirty(vd, 0, NULL, txg); vdev_dirty(vd, VDD_METASLAB, ms, txg); } *msp = ms; return (0); } static void metaslab_fini_flush_data(metaslab_t *msp) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; if (metaslab_unflushed_txg(msp) == 0) { ASSERT3P(avl_find(&spa->spa_metaslabs_by_flushed, msp, NULL), ==, NULL); return; } ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)); mutex_enter(&spa->spa_flushed_ms_lock); avl_remove(&spa->spa_metaslabs_by_flushed, msp); mutex_exit(&spa->spa_flushed_ms_lock); spa_log_sm_decrement_mscount(spa, metaslab_unflushed_txg(msp)); spa_log_summary_decrement_mscount(spa, metaslab_unflushed_txg(msp)); } uint64_t metaslab_unflushed_changes_memused(metaslab_t *ms) { return ((range_tree_numsegs(ms->ms_unflushed_allocs) + range_tree_numsegs(ms->ms_unflushed_frees)) * ms->ms_unflushed_allocs->rt_root.bt_elem_size); } void metaslab_fini(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; metaslab_fini_flush_data(msp); metaslab_group_remove(mg, msp); mutex_enter(&msp->ms_lock); VERIFY(msp->ms_group == NULL); /* * If the range trees haven't been allocated, this metaslab hasn't * been through metaslab_sync_done() for the first time yet, so its * space hasn't been accounted for in its vdev and doesn't need to be * subtracted. */ if (msp->ms_freed != NULL) { metaslab_space_update(vd, mg->mg_class, -metaslab_allocated_space(msp), 0, -msp->ms_size); } space_map_close(msp->ms_sm); msp->ms_sm = NULL; metaslab_unload(msp); range_tree_destroy(msp->ms_allocatable); if (msp->ms_freed != NULL) { range_tree_destroy(msp->ms_freeing); range_tree_destroy(msp->ms_freed); ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=, metaslab_unflushed_changes_memused(msp)); spa->spa_unflushed_stats.sus_memused -= metaslab_unflushed_changes_memused(msp); range_tree_vacate(msp->ms_unflushed_allocs, NULL, NULL); range_tree_destroy(msp->ms_unflushed_allocs); range_tree_destroy(msp->ms_checkpointing); range_tree_vacate(msp->ms_unflushed_frees, NULL, NULL); range_tree_destroy(msp->ms_unflushed_frees); for (int t = 0; t < TXG_SIZE; t++) { range_tree_destroy(msp->ms_allocating[t]); } for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_destroy(msp->ms_defer[t]); } } ASSERT0(msp->ms_deferspace); for (int t = 0; t < TXG_SIZE; t++) ASSERT(!txg_list_member(&vd->vdev_ms_list, msp, t)); range_tree_vacate(msp->ms_trim, NULL, NULL); range_tree_destroy(msp->ms_trim); mutex_exit(&msp->ms_lock); cv_destroy(&msp->ms_load_cv); cv_destroy(&msp->ms_flush_cv); mutex_destroy(&msp->ms_lock); mutex_destroy(&msp->ms_sync_lock); ASSERT3U(msp->ms_allocator, ==, -1); kmem_free(msp, sizeof (metaslab_t)); } #define FRAGMENTATION_TABLE_SIZE 17 /* * This table defines a segment size based fragmentation metric that will * allow each metaslab to derive its own fragmentation value. This is done * by calculating the space in each bucket of the spacemap histogram and * multiplying that by the fragmentation metric in this table. Doing * this for all buckets and dividing it by the total amount of free * space in this metaslab (i.e. the total free space in all buckets) gives * us the fragmentation metric. This means that a high fragmentation metric * equates to most of the free space being comprised of small segments. * Conversely, if the metric is low, then most of the free space is in * large segments. A 10% change in fragmentation equates to approximately * double the number of segments. * * This table defines 0% fragmented space using 16MB segments. Testing has * shown that segments that are greater than or equal to 16MB do not suffer * from drastic performance problems. Using this value, we derive the rest * of the table. Since the fragmentation value is never stored on disk, it * is possible to change these calculations in the future. */ int zfs_frag_table[FRAGMENTATION_TABLE_SIZE] = { 100, /* 512B */ 100, /* 1K */ 98, /* 2K */ 95, /* 4K */ 90, /* 8K */ 80, /* 16K */ 70, /* 32K */ 60, /* 64K */ 50, /* 128K */ 40, /* 256K */ 30, /* 512K */ 20, /* 1M */ 15, /* 2M */ 10, /* 4M */ 5, /* 8M */ 0 /* 16M */ }; /* * Calculate the metaslab's fragmentation metric and set ms_fragmentation. * Setting this value to ZFS_FRAG_INVALID means that the metaslab has not * been upgraded and does not support this metric. Otherwise, the return * value should be in the range [0, 100]. */ static void metaslab_set_fragmentation(metaslab_t *msp, boolean_t nodirty) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; uint64_t fragmentation = 0; uint64_t total = 0; boolean_t feature_enabled = spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM); if (!feature_enabled) { msp->ms_fragmentation = ZFS_FRAG_INVALID; return; } /* * A null space map means that the entire metaslab is free * and thus is not fragmented. */ if (msp->ms_sm == NULL) { msp->ms_fragmentation = 0; return; } /* * If this metaslab's space map has not been upgraded, flag it * so that we upgrade next time we encounter it. */ if (msp->ms_sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) { uint64_t txg = spa_syncing_txg(spa); vdev_t *vd = msp->ms_group->mg_vd; /* * If we've reached the final dirty txg, then we must * be shutting down the pool. We don't want to dirty * any data past this point so skip setting the condense * flag. We can retry this action the next time the pool * is imported. We also skip marking this metaslab for * condensing if the caller has explicitly set nodirty. */ if (!nodirty && spa_writeable(spa) && txg < spa_final_dirty_txg(spa)) { msp->ms_condense_wanted = B_TRUE; vdev_dirty(vd, VDD_METASLAB, msp, txg + 1); zfs_dbgmsg("txg %llu, requesting force condense: " "ms_id %llu, vdev_id %llu", txg, msp->ms_id, vd->vdev_id); } msp->ms_fragmentation = ZFS_FRAG_INVALID; return; } for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { uint64_t space = 0; uint8_t shift = msp->ms_sm->sm_shift; int idx = MIN(shift - SPA_MINBLOCKSHIFT + i, FRAGMENTATION_TABLE_SIZE - 1); if (msp->ms_sm->sm_phys->smp_histogram[i] == 0) continue; space = msp->ms_sm->sm_phys->smp_histogram[i] << (i + shift); total += space; ASSERT3U(idx, <, FRAGMENTATION_TABLE_SIZE); fragmentation += space * zfs_frag_table[idx]; } if (total > 0) fragmentation /= total; ASSERT3U(fragmentation, <=, 100); msp->ms_fragmentation = fragmentation; } /* * Compute a weight -- a selection preference value -- for the given metaslab. * This is based on the amount of free space, the level of fragmentation, * the LBA range, and whether the metaslab is loaded. */ static uint64_t metaslab_space_weight(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; uint64_t weight, space; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * The baseline weight is the metaslab's free space. */ space = msp->ms_size - metaslab_allocated_space(msp); if (metaslab_fragmentation_factor_enabled && msp->ms_fragmentation != ZFS_FRAG_INVALID) { /* * Use the fragmentation information to inversely scale * down the baseline weight. We need to ensure that we * don't exclude this metaslab completely when it's 100% * fragmented. To avoid this we reduce the fragmented value * by 1. */ space = (space * (100 - (msp->ms_fragmentation - 1))) / 100; /* * If space < SPA_MINBLOCKSIZE, then we will not allocate from * this metaslab again. The fragmentation metric may have * decreased the space to something smaller than * SPA_MINBLOCKSIZE, so reset the space to SPA_MINBLOCKSIZE * so that we can consume any remaining space. */ if (space > 0 && space < SPA_MINBLOCKSIZE) space = SPA_MINBLOCKSIZE; } weight = space; /* * Modern disks have uniform bit density and constant angular velocity. * Therefore, the outer recording zones are faster (higher bandwidth) * than the inner zones by the ratio of outer to inner track diameter, * which is typically around 2:1. We account for this by assigning * higher weight to lower metaslabs (multiplier ranging from 2x to 1x). * In effect, this means that we'll select the metaslab with the most * free bandwidth rather than simply the one with the most free space. */ if (!vd->vdev_nonrot && metaslab_lba_weighting_enabled) { weight = 2 * weight - (msp->ms_id * weight) / vd->vdev_ms_count; ASSERT(weight >= space && weight <= 2 * space); } /* * If this metaslab is one we're actively using, adjust its * weight to make it preferable to any inactive metaslab so * we'll polish it off. If the fragmentation on this metaslab * has exceed our threshold, then don't mark it active. */ if (msp->ms_loaded && msp->ms_fragmentation != ZFS_FRAG_INVALID && msp->ms_fragmentation <= zfs_metaslab_fragmentation_threshold) { weight |= (msp->ms_weight & METASLAB_ACTIVE_MASK); } WEIGHT_SET_SPACEBASED(weight); return (weight); } /* * Return the weight of the specified metaslab, according to the segment-based * weighting algorithm. The metaslab must be loaded. This function can * be called within a sync pass since it relies only on the metaslab's * range tree which is always accurate when the metaslab is loaded. */ static uint64_t metaslab_weight_from_range_tree(metaslab_t *msp) { uint64_t weight = 0; uint32_t segments = 0; ASSERT(msp->ms_loaded); for (int i = RANGE_TREE_HISTOGRAM_SIZE - 1; i >= SPA_MINBLOCKSHIFT; i--) { uint8_t shift = msp->ms_group->mg_vd->vdev_ashift; int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1; segments <<= 1; segments += msp->ms_allocatable->rt_histogram[i]; /* * The range tree provides more precision than the space map * and must be downgraded so that all values fit within the * space map's histogram. This allows us to compare loaded * vs. unloaded metaslabs to determine which metaslab is * considered "best". */ if (i > max_idx) continue; if (segments != 0) { WEIGHT_SET_COUNT(weight, segments); WEIGHT_SET_INDEX(weight, i); WEIGHT_SET_ACTIVE(weight, 0); break; } } return (weight); } /* * Calculate the weight based on the on-disk histogram. Should be applied * only to unloaded metaslabs (i.e no incoming allocations) in-order to * give results consistent with the on-disk state */ static uint64_t metaslab_weight_from_spacemap(metaslab_t *msp) { space_map_t *sm = msp->ms_sm; ASSERT(!msp->ms_loaded); ASSERT(sm != NULL); ASSERT3U(space_map_object(sm), !=, 0); ASSERT3U(sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t)); /* * Create a joint histogram from all the segments that have made * it to the metaslab's space map histogram, that are not yet * available for allocation because they are still in the freeing * pipeline (e.g. freeing, freed, and defer trees). Then subtract * these segments from the space map's histogram to get a more * accurate weight. */ uint64_t deferspace_histogram[SPACE_MAP_HISTOGRAM_SIZE] = {0}; for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) deferspace_histogram[i] += msp->ms_synchist[i]; for (int t = 0; t < TXG_DEFER_SIZE; t++) { for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { deferspace_histogram[i] += msp->ms_deferhist[t][i]; } } uint64_t weight = 0; for (int i = SPACE_MAP_HISTOGRAM_SIZE - 1; i >= 0; i--) { ASSERT3U(sm->sm_phys->smp_histogram[i], >=, deferspace_histogram[i]); uint64_t count = sm->sm_phys->smp_histogram[i] - deferspace_histogram[i]; if (count != 0) { WEIGHT_SET_COUNT(weight, count); WEIGHT_SET_INDEX(weight, i + sm->sm_shift); WEIGHT_SET_ACTIVE(weight, 0); break; } } return (weight); } /* * Compute a segment-based weight for the specified metaslab. The weight * is determined by highest bucket in the histogram. The information * for the highest bucket is encoded into the weight value. */ static uint64_t metaslab_segment_weight(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; uint64_t weight = 0; uint8_t shift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * The metaslab is completely free. */ if (metaslab_allocated_space(msp) == 0) { int idx = highbit64(msp->ms_size) - 1; int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1; if (idx < max_idx) { WEIGHT_SET_COUNT(weight, 1ULL); WEIGHT_SET_INDEX(weight, idx); } else { WEIGHT_SET_COUNT(weight, 1ULL << (idx - max_idx)); WEIGHT_SET_INDEX(weight, max_idx); } WEIGHT_SET_ACTIVE(weight, 0); ASSERT(!WEIGHT_IS_SPACEBASED(weight)); return (weight); } ASSERT3U(msp->ms_sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t)); /* * If the metaslab is fully allocated then just make the weight 0. */ if (metaslab_allocated_space(msp) == msp->ms_size) return (0); /* * If the metaslab is already loaded, then use the range tree to * determine the weight. Otherwise, we rely on the space map information * to generate the weight. */ if (msp->ms_loaded) { weight = metaslab_weight_from_range_tree(msp); } else { weight = metaslab_weight_from_spacemap(msp); } /* * If the metaslab was active the last time we calculated its weight * then keep it active. We want to consume the entire region that * is associated with this weight. */ if (msp->ms_activation_weight != 0 && weight != 0) WEIGHT_SET_ACTIVE(weight, WEIGHT_GET_ACTIVE(msp->ms_weight)); return (weight); } /* * Determine if we should attempt to allocate from this metaslab. If the * metaslab is loaded, then we can determine if the desired allocation * can be satisfied by looking at the size of the maximum free segment * on that metaslab. Otherwise, we make our decision based on the metaslab's * weight. For segment-based weighting we can determine the maximum * allocation based on the index encoded in its value. For space-based * weights we rely on the entire weight (excluding the weight-type bit). */ static boolean_t metaslab_should_allocate(metaslab_t *msp, uint64_t asize, boolean_t try_hard) { /* * If the metaslab is loaded, ms_max_size is definitive and we can use * the fast check. If it's not, the ms_max_size is a lower bound (once * set), and we should use the fast check as long as we're not in * try_hard and it's been less than zfs_metaslab_max_size_cache_sec * seconds since the metaslab was unloaded. */ if (msp->ms_loaded || (msp->ms_max_size != 0 && !try_hard && gethrtime() < msp->ms_unload_time + SEC2NSEC(zfs_metaslab_max_size_cache_sec))) return (msp->ms_max_size >= asize); boolean_t should_allocate; if (!WEIGHT_IS_SPACEBASED(msp->ms_weight)) { /* * The metaslab segment weight indicates segments in the * range [2^i, 2^(i+1)), where i is the index in the weight. * Since the asize might be in the middle of the range, we * should attempt the allocation if asize < 2^(i+1). */ should_allocate = (asize < 1ULL << (WEIGHT_GET_INDEX(msp->ms_weight) + 1)); } else { should_allocate = (asize <= (msp->ms_weight & ~METASLAB_WEIGHT_TYPE)); } return (should_allocate); } static uint64_t metaslab_weight(metaslab_t *msp, boolean_t nodirty) { vdev_t *vd = msp->ms_group->mg_vd; spa_t *spa = vd->vdev_spa; uint64_t weight; ASSERT(MUTEX_HELD(&msp->ms_lock)); metaslab_set_fragmentation(msp, nodirty); /* * Update the maximum size. If the metaslab is loaded, this will * ensure that we get an accurate maximum size if newly freed space * has been added back into the free tree. If the metaslab is * unloaded, we check if there's a larger free segment in the * unflushed frees. This is a lower bound on the largest allocatable * segment size. Coalescing of adjacent entries may reveal larger * allocatable segments, but we aren't aware of those until loading * the space map into a range tree. */ if (msp->ms_loaded) { msp->ms_max_size = metaslab_largest_allocatable(msp); } else { msp->ms_max_size = MAX(msp->ms_max_size, metaslab_largest_unflushed_free(msp)); } /* * Segment-based weighting requires space map histogram support. */ if (zfs_metaslab_segment_weight_enabled && spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM) && (msp->ms_sm == NULL || msp->ms_sm->sm_dbuf->db_size == sizeof (space_map_phys_t))) { weight = metaslab_segment_weight(msp); } else { weight = metaslab_space_weight(msp); } return (weight); } void metaslab_recalculate_weight_and_sort(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); /* note: we preserve the mask (e.g. indication of primary, etc..) */ uint64_t was_active = msp->ms_weight & METASLAB_ACTIVE_MASK; metaslab_group_sort(msp->ms_group, msp, metaslab_weight(msp, B_FALSE) | was_active); } static int metaslab_activate_allocator(metaslab_group_t *mg, metaslab_t *msp, int allocator, uint64_t activation_weight) { metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator]; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * If we're activating for the claim code, we don't want to actually * set the metaslab up for a specific allocator. */ if (activation_weight == METASLAB_WEIGHT_CLAIM) { ASSERT0(msp->ms_activation_weight); msp->ms_activation_weight = msp->ms_weight; metaslab_group_sort(mg, msp, msp->ms_weight | activation_weight); return (0); } metaslab_t **mspp = (activation_weight == METASLAB_WEIGHT_PRIMARY ? &mga->mga_primary : &mga->mga_secondary); mutex_enter(&mg->mg_lock); if (*mspp != NULL) { mutex_exit(&mg->mg_lock); return (EEXIST); } *mspp = msp; ASSERT3S(msp->ms_allocator, ==, -1); msp->ms_allocator = allocator; msp->ms_primary = (activation_weight == METASLAB_WEIGHT_PRIMARY); ASSERT0(msp->ms_activation_weight); msp->ms_activation_weight = msp->ms_weight; metaslab_group_sort_impl(mg, msp, msp->ms_weight | activation_weight); mutex_exit(&mg->mg_lock); return (0); } static int metaslab_activate(metaslab_t *msp, int allocator, uint64_t activation_weight) { ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * The current metaslab is already activated for us so there * is nothing to do. Already activated though, doesn't mean * that this metaslab is activated for our allocator nor our * requested activation weight. The metaslab could have started * as an active one for our allocator but changed allocators * while we were waiting to grab its ms_lock or we stole it * [see find_valid_metaslab()]. This means that there is a * possibility of passivating a metaslab of another allocator * or from a different activation mask, from this thread. */ if ((msp->ms_weight & METASLAB_ACTIVE_MASK) != 0) { ASSERT(msp->ms_loaded); return (0); } int error = metaslab_load(msp); if (error != 0) { metaslab_group_sort(msp->ms_group, msp, 0); return (error); } /* * When entering metaslab_load() we may have dropped the * ms_lock because we were loading this metaslab, or we * were waiting for another thread to load it for us. In * that scenario, we recheck the weight of the metaslab * to see if it was activated by another thread. * * If the metaslab was activated for another allocator or * it was activated with a different activation weight (e.g. * we wanted to make it a primary but it was activated as * secondary) we return error (EBUSY). * * If the metaslab was activated for the same allocator * and requested activation mask, skip activating it. */ if ((msp->ms_weight & METASLAB_ACTIVE_MASK) != 0) { if (msp->ms_allocator != allocator) return (EBUSY); if ((msp->ms_weight & activation_weight) == 0) return (SET_ERROR(EBUSY)); EQUIV((activation_weight == METASLAB_WEIGHT_PRIMARY), msp->ms_primary); return (0); } /* * If the metaslab has literally 0 space, it will have weight 0. In * that case, don't bother activating it. This can happen if the * metaslab had space during find_valid_metaslab, but another thread * loaded it and used all that space while we were waiting to grab the * lock. */ if (msp->ms_weight == 0) { ASSERT0(range_tree_space(msp->ms_allocatable)); return (SET_ERROR(ENOSPC)); } if ((error = metaslab_activate_allocator(msp->ms_group, msp, allocator, activation_weight)) != 0) { return (error); } ASSERT(msp->ms_loaded); ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK); return (0); } static void metaslab_passivate_allocator(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight) { ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(msp->ms_loaded); if (msp->ms_weight & METASLAB_WEIGHT_CLAIM) { metaslab_group_sort(mg, msp, weight); return; } mutex_enter(&mg->mg_lock); ASSERT3P(msp->ms_group, ==, mg); ASSERT3S(0, <=, msp->ms_allocator); ASSERT3U(msp->ms_allocator, <, mg->mg_allocators); metaslab_group_allocator_t *mga = &mg->mg_allocator[msp->ms_allocator]; if (msp->ms_primary) { ASSERT3P(mga->mga_primary, ==, msp); ASSERT(msp->ms_weight & METASLAB_WEIGHT_PRIMARY); mga->mga_primary = NULL; } else { ASSERT3P(mga->mga_secondary, ==, msp); ASSERT(msp->ms_weight & METASLAB_WEIGHT_SECONDARY); mga->mga_secondary = NULL; } msp->ms_allocator = -1; metaslab_group_sort_impl(mg, msp, weight); mutex_exit(&mg->mg_lock); } static void metaslab_passivate(metaslab_t *msp, uint64_t weight) { uint64_t size __maybe_unused = weight & ~METASLAB_WEIGHT_TYPE; /* * If size < SPA_MINBLOCKSIZE, then we will not allocate from * this metaslab again. In that case, it had better be empty, * or we would be leaving space on the table. */ ASSERT(!WEIGHT_IS_SPACEBASED(msp->ms_weight) || size >= SPA_MINBLOCKSIZE || range_tree_space(msp->ms_allocatable) == 0); ASSERT0(weight & METASLAB_ACTIVE_MASK); ASSERT(msp->ms_activation_weight != 0); msp->ms_activation_weight = 0; metaslab_passivate_allocator(msp->ms_group, msp, weight); ASSERT0(msp->ms_weight & METASLAB_ACTIVE_MASK); } /* * Segment-based metaslabs are activated once and remain active until * we either fail an allocation attempt (similar to space-based metaslabs) * or have exhausted the free space in zfs_metaslab_switch_threshold * buckets since the metaslab was activated. This function checks to see * if we've exhausted the zfs_metaslab_switch_threshold buckets in the * metaslab and passivates it proactively. This will allow us to select a * metaslab with a larger contiguous region, if any, remaining within this * metaslab group. If we're in sync pass > 1, then we continue using this * metaslab so that we don't dirty more block and cause more sync passes. */ static void metaslab_segment_may_passivate(metaslab_t *msp) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; if (WEIGHT_IS_SPACEBASED(msp->ms_weight) || spa_sync_pass(spa) > 1) return; /* * Since we are in the middle of a sync pass, the most accurate * information that is accessible to us is the in-core range tree * histogram; calculate the new weight based on that information. */ uint64_t weight = metaslab_weight_from_range_tree(msp); int activation_idx = WEIGHT_GET_INDEX(msp->ms_activation_weight); int current_idx = WEIGHT_GET_INDEX(weight); if (current_idx <= activation_idx - zfs_metaslab_switch_threshold) metaslab_passivate(msp, weight); } static void metaslab_preload(void *arg) { metaslab_t *msp = arg; metaslab_class_t *mc = msp->ms_group->mg_class; spa_t *spa = mc->mc_spa; fstrans_cookie_t cookie = spl_fstrans_mark(); ASSERT(!MUTEX_HELD(&msp->ms_group->mg_lock)); mutex_enter(&msp->ms_lock); (void) metaslab_load(msp); metaslab_set_selected_txg(msp, spa_syncing_txg(spa)); mutex_exit(&msp->ms_lock); spl_fstrans_unmark(cookie); } static void metaslab_group_preload(metaslab_group_t *mg) { spa_t *spa = mg->mg_vd->vdev_spa; metaslab_t *msp; avl_tree_t *t = &mg->mg_metaslab_tree; int m = 0; if (spa_shutting_down(spa) || !metaslab_preload_enabled) { taskq_wait_outstanding(mg->mg_taskq, 0); return; } mutex_enter(&mg->mg_lock); /* * Load the next potential metaslabs */ for (msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) { ASSERT3P(msp->ms_group, ==, mg); /* * We preload only the maximum number of metaslabs specified * by metaslab_preload_limit. If a metaslab is being forced * to condense then we preload it too. This will ensure * that force condensing happens in the next txg. */ if (++m > metaslab_preload_limit && !msp->ms_condense_wanted) { continue; } VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload, msp, TQ_SLEEP) != TASKQID_INVALID); } mutex_exit(&mg->mg_lock); } /* * Determine if the space map's on-disk footprint is past our tolerance for * inefficiency. We would like to use the following criteria to make our * decision: * * 1. Do not condense if the size of the space map object would dramatically * increase as a result of writing out the free space range tree. * * 2. Condense if the on on-disk space map representation is at least * zfs_condense_pct/100 times the size of the optimal representation * (i.e. zfs_condense_pct = 110 and in-core = 1MB, optimal = 1.1MB). * * 3. Do not condense if the on-disk size of the space map does not actually * decrease. * * Unfortunately, we cannot compute the on-disk size of the space map in this * context because we cannot accurately compute the effects of compression, etc. * Instead, we apply the heuristic described in the block comment for * zfs_metaslab_condense_block_threshold - we only condense if the space used * is greater than a threshold number of blocks. */ static boolean_t metaslab_should_condense(metaslab_t *msp) { space_map_t *sm = msp->ms_sm; vdev_t *vd = msp->ms_group->mg_vd; uint64_t vdev_blocksize = 1 << vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(msp->ms_loaded); ASSERT(sm != NULL); ASSERT3U(spa_sync_pass(vd->vdev_spa), ==, 1); /* * We always condense metaslabs that are empty and metaslabs for * which a condense request has been made. */ if (range_tree_numsegs(msp->ms_allocatable) == 0 || msp->ms_condense_wanted) return (B_TRUE); uint64_t record_size = MAX(sm->sm_blksz, vdev_blocksize); uint64_t object_size = space_map_length(sm); uint64_t optimal_size = space_map_estimate_optimal_size(sm, msp->ms_allocatable, SM_NO_VDEVID); return (object_size >= (optimal_size * zfs_condense_pct / 100) && object_size > zfs_metaslab_condense_block_threshold * record_size); } /* * Condense the on-disk space map representation to its minimized form. * The minimized form consists of a small number of allocations followed * by the entries of the free range tree (ms_allocatable). The condensed * spacemap contains all the entries of previous TXGs (including those in * the pool-wide log spacemaps; thus this is effectively a superset of * metaslab_flush()), but this TXG's entries still need to be written. */ static void metaslab_condense(metaslab_t *msp, dmu_tx_t *tx) { range_tree_t *condense_tree; space_map_t *sm = msp->ms_sm; uint64_t txg = dmu_tx_get_txg(tx); spa_t *spa = msp->ms_group->mg_vd->vdev_spa; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(msp->ms_loaded); ASSERT(msp->ms_sm != NULL); /* * In order to condense the space map, we need to change it so it * only describes which segments are currently allocated and free. * * All the current free space resides in the ms_allocatable, all * the ms_defer trees, and all the ms_allocating trees. We ignore * ms_freed because it is empty because we're in sync pass 1. We * ignore ms_freeing because these changes are not yet reflected * in the spacemap (they will be written later this txg). * * So to truncate the space map to represent all the entries of * previous TXGs we do the following: * * 1] We create a range tree (condense tree) that is 100% empty. * 2] We add to it all segments found in the ms_defer trees * as those segments are marked as free in the original space * map. We do the same with the ms_allocating trees for the same * reason. Adding these segments should be a relatively * inexpensive operation since we expect these trees to have a * small number of nodes. * 3] We vacate any unflushed allocs, since they are not frees we * need to add to the condense tree. Then we vacate any * unflushed frees as they should already be part of ms_allocatable. * 4] At this point, we would ideally like to add all segments * in the ms_allocatable tree from the condense tree. This way * we would write all the entries of the condense tree as the * condensed space map, which would only contain freed * segments with everything else assumed to be allocated. * * Doing so can be prohibitively expensive as ms_allocatable can * be large, and therefore computationally expensive to add to * the condense_tree. Instead we first sync out an entry marking * everything as allocated, then the condense_tree and then the * ms_allocatable, in the condensed space map. While this is not * optimal, it is typically close to optimal and more importantly * much cheaper to compute. * * 5] Finally, as both of the unflushed trees were written to our * new and condensed metaslab space map, we basically flushed * all the unflushed changes to disk, thus we call * metaslab_flush_update(). */ ASSERT3U(spa_sync_pass(spa), ==, 1); ASSERT(range_tree_is_empty(msp->ms_freed)); /* since it is pass 1 */ zfs_dbgmsg("condensing: txg %llu, msp[%llu] %px, vdev id %llu, " "spa %s, smp size %llu, segments %lu, forcing condense=%s", txg, msp->ms_id, msp, msp->ms_group->mg_vd->vdev_id, spa->spa_name, space_map_length(msp->ms_sm), range_tree_numsegs(msp->ms_allocatable), msp->ms_condense_wanted ? "TRUE" : "FALSE"); msp->ms_condense_wanted = B_FALSE; range_seg_type_t type; uint64_t shift, start; type = metaslab_calculate_range_tree_type(msp->ms_group->mg_vd, msp, &start, &shift); condense_tree = range_tree_create(NULL, type, NULL, start, shift); for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_walk(msp->ms_defer[t], range_tree_add, condense_tree); } for (int t = 0; t < TXG_CONCURRENT_STATES; t++) { range_tree_walk(msp->ms_allocating[(txg + t) & TXG_MASK], range_tree_add, condense_tree); } ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=, metaslab_unflushed_changes_memused(msp)); spa->spa_unflushed_stats.sus_memused -= metaslab_unflushed_changes_memused(msp); range_tree_vacate(msp->ms_unflushed_allocs, NULL, NULL); range_tree_vacate(msp->ms_unflushed_frees, NULL, NULL); /* * We're about to drop the metaslab's lock thus allowing other * consumers to change it's content. Set the metaslab's ms_condensing * flag to ensure that allocations on this metaslab do not occur * while we're in the middle of committing it to disk. This is only * critical for ms_allocatable as all other range trees use per TXG * views of their content. */ msp->ms_condensing = B_TRUE; mutex_exit(&msp->ms_lock); uint64_t object = space_map_object(msp->ms_sm); space_map_truncate(sm, spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP) ? zfs_metaslab_sm_blksz_with_log : zfs_metaslab_sm_blksz_no_log, tx); /* * space_map_truncate() may have reallocated the spacemap object. * If so, update the vdev_ms_array. */ if (space_map_object(msp->ms_sm) != object) { object = space_map_object(msp->ms_sm); dmu_write(spa->spa_meta_objset, msp->ms_group->mg_vd->vdev_ms_array, sizeof (uint64_t) * msp->ms_id, sizeof (uint64_t), &object, tx); } /* * Note: * When the log space map feature is enabled, each space map will * always have ALLOCS followed by FREES for each sync pass. This is * typically true even when the log space map feature is disabled, * except from the case where a metaslab goes through metaslab_sync() * and gets condensed. In that case the metaslab's space map will have * ALLOCS followed by FREES (due to condensing) followed by ALLOCS * followed by FREES (due to space_map_write() in metaslab_sync()) for * sync pass 1. */ range_tree_t *tmp_tree = range_tree_create(NULL, type, NULL, start, shift); range_tree_add(tmp_tree, msp->ms_start, msp->ms_size); space_map_write(sm, tmp_tree, SM_ALLOC, SM_NO_VDEVID, tx); space_map_write(sm, msp->ms_allocatable, SM_FREE, SM_NO_VDEVID, tx); space_map_write(sm, condense_tree, SM_FREE, SM_NO_VDEVID, tx); range_tree_vacate(condense_tree, NULL, NULL); range_tree_destroy(condense_tree); range_tree_vacate(tmp_tree, NULL, NULL); range_tree_destroy(tmp_tree); mutex_enter(&msp->ms_lock); msp->ms_condensing = B_FALSE; metaslab_flush_update(msp, tx); } /* * Called when the metaslab has been flushed (its own spacemap now reflects * all the contents of the pool-wide spacemap log). Updates the metaslab's * metadata and any pool-wide related log space map data (e.g. summary, * obsolete logs, etc..) to reflect that. */ static void metaslab_flush_update(metaslab_t *msp, dmu_tx_t *tx) { metaslab_group_t *mg = msp->ms_group; spa_t *spa = mg->mg_vd->vdev_spa; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(spa_sync_pass(spa), ==, 1); ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs)); ASSERT(range_tree_is_empty(msp->ms_unflushed_frees)); /* * Just because a metaslab got flushed, that doesn't mean that * it will pass through metaslab_sync_done(). Thus, make sure to * update ms_synced_length here in case it doesn't. */ msp->ms_synced_length = space_map_length(msp->ms_sm); /* * We may end up here from metaslab_condense() without the * feature being active. In that case this is a no-op. */ if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) return; ASSERT(spa_syncing_log_sm(spa) != NULL); ASSERT(msp->ms_sm != NULL); ASSERT(metaslab_unflushed_txg(msp) != 0); ASSERT3P(avl_find(&spa->spa_metaslabs_by_flushed, msp, NULL), ==, msp); VERIFY3U(tx->tx_txg, <=, spa_final_dirty_txg(spa)); /* update metaslab's position in our flushing tree */ uint64_t ms_prev_flushed_txg = metaslab_unflushed_txg(msp); mutex_enter(&spa->spa_flushed_ms_lock); avl_remove(&spa->spa_metaslabs_by_flushed, msp); metaslab_set_unflushed_txg(msp, spa_syncing_txg(spa), tx); avl_add(&spa->spa_metaslabs_by_flushed, msp); mutex_exit(&spa->spa_flushed_ms_lock); /* update metaslab counts of spa_log_sm_t nodes */ spa_log_sm_decrement_mscount(spa, ms_prev_flushed_txg); spa_log_sm_increment_current_mscount(spa); /* cleanup obsolete logs if any */ uint64_t log_blocks_before = spa_log_sm_nblocks(spa); spa_cleanup_old_sm_logs(spa, tx); uint64_t log_blocks_after = spa_log_sm_nblocks(spa); VERIFY3U(log_blocks_after, <=, log_blocks_before); /* update log space map summary */ uint64_t blocks_gone = log_blocks_before - log_blocks_after; spa_log_summary_add_flushed_metaslab(spa); spa_log_summary_decrement_mscount(spa, ms_prev_flushed_txg); spa_log_summary_decrement_blkcount(spa, blocks_gone); } boolean_t metaslab_flush(metaslab_t *msp, dmu_tx_t *tx) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(spa_sync_pass(spa), ==, 1); ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)); ASSERT(msp->ms_sm != NULL); ASSERT(metaslab_unflushed_txg(msp) != 0); ASSERT(avl_find(&spa->spa_metaslabs_by_flushed, msp, NULL) != NULL); /* * There is nothing wrong with flushing the same metaslab twice, as * this codepath should work on that case. However, the current * flushing scheme makes sure to avoid this situation as we would be * making all these calls without having anything meaningful to write * to disk. We assert this behavior here. */ ASSERT3U(metaslab_unflushed_txg(msp), <, dmu_tx_get_txg(tx)); /* * We can not flush while loading, because then we would * not load the ms_unflushed_{allocs,frees}. */ if (msp->ms_loading) return (B_FALSE); metaslab_verify_space(msp, dmu_tx_get_txg(tx)); metaslab_verify_weight_and_frag(msp); /* * Metaslab condensing is effectively flushing. Therefore if the * metaslab can be condensed we can just condense it instead of * flushing it. * * Note that metaslab_condense() does call metaslab_flush_update() * so we can just return immediately after condensing. We also * don't need to care about setting ms_flushing or broadcasting * ms_flush_cv, even if we temporarily drop the ms_lock in * metaslab_condense(), as the metaslab is already loaded. */ if (msp->ms_loaded && metaslab_should_condense(msp)) { metaslab_group_t *mg = msp->ms_group; /* * For all histogram operations below refer to the * comments of metaslab_sync() where we follow a * similar procedure. */ metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); metaslab_group_histogram_remove(mg, msp); metaslab_condense(msp, tx); space_map_histogram_clear(msp->ms_sm); space_map_histogram_add(msp->ms_sm, msp->ms_allocatable, tx); ASSERT(range_tree_is_empty(msp->ms_freed)); for (int t = 0; t < TXG_DEFER_SIZE; t++) { space_map_histogram_add(msp->ms_sm, msp->ms_defer[t], tx); } metaslab_aux_histograms_update(msp); metaslab_group_histogram_add(mg, msp); metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); metaslab_verify_space(msp, dmu_tx_get_txg(tx)); /* * Since we recreated the histogram (and potentially * the ms_sm too while condensing) ensure that the * weight is updated too because we are not guaranteed * that this metaslab is dirty and will go through * metaslab_sync_done(). */ metaslab_recalculate_weight_and_sort(msp); return (B_TRUE); } msp->ms_flushing = B_TRUE; uint64_t sm_len_before = space_map_length(msp->ms_sm); mutex_exit(&msp->ms_lock); space_map_write(msp->ms_sm, msp->ms_unflushed_allocs, SM_ALLOC, SM_NO_VDEVID, tx); space_map_write(msp->ms_sm, msp->ms_unflushed_frees, SM_FREE, SM_NO_VDEVID, tx); mutex_enter(&msp->ms_lock); uint64_t sm_len_after = space_map_length(msp->ms_sm); if (zfs_flags & ZFS_DEBUG_LOG_SPACEMAP) { zfs_dbgmsg("flushing: txg %llu, spa %s, vdev_id %llu, " "ms_id %llu, unflushed_allocs %llu, unflushed_frees %llu, " "appended %llu bytes", dmu_tx_get_txg(tx), spa_name(spa), msp->ms_group->mg_vd->vdev_id, msp->ms_id, range_tree_space(msp->ms_unflushed_allocs), range_tree_space(msp->ms_unflushed_frees), (sm_len_after - sm_len_before)); } ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=, metaslab_unflushed_changes_memused(msp)); spa->spa_unflushed_stats.sus_memused -= metaslab_unflushed_changes_memused(msp); range_tree_vacate(msp->ms_unflushed_allocs, NULL, NULL); range_tree_vacate(msp->ms_unflushed_frees, NULL, NULL); metaslab_verify_space(msp, dmu_tx_get_txg(tx)); metaslab_verify_weight_and_frag(msp); metaslab_flush_update(msp, tx); metaslab_verify_space(msp, dmu_tx_get_txg(tx)); metaslab_verify_weight_and_frag(msp); msp->ms_flushing = B_FALSE; cv_broadcast(&msp->ms_flush_cv); return (B_TRUE); } /* * Write a metaslab to disk in the context of the specified transaction group. */ void metaslab_sync(metaslab_t *msp, uint64_t txg) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; objset_t *mos = spa_meta_objset(spa); range_tree_t *alloctree = msp->ms_allocating[txg & TXG_MASK]; dmu_tx_t *tx; ASSERT(!vd->vdev_ishole); /* * This metaslab has just been added so there's no work to do now. */ if (msp->ms_freeing == NULL) { ASSERT3P(alloctree, ==, NULL); return; } ASSERT3P(alloctree, !=, NULL); ASSERT3P(msp->ms_freeing, !=, NULL); ASSERT3P(msp->ms_freed, !=, NULL); ASSERT3P(msp->ms_checkpointing, !=, NULL); ASSERT3P(msp->ms_trim, !=, NULL); /* * Normally, we don't want to process a metaslab if there are no * allocations or frees to perform. However, if the metaslab is being * forced to condense, it's loaded and we're not beyond the final * dirty txg, we need to let it through. Not condensing beyond the * final dirty txg prevents an issue where metaslabs that need to be * condensed but were loaded for other reasons could cause a panic * here. By only checking the txg in that branch of the conditional, * we preserve the utility of the VERIFY statements in all other * cases. */ if (range_tree_is_empty(alloctree) && range_tree_is_empty(msp->ms_freeing) && range_tree_is_empty(msp->ms_checkpointing) && !(msp->ms_loaded && msp->ms_condense_wanted && txg <= spa_final_dirty_txg(spa))) return; VERIFY3U(txg, <=, spa_final_dirty_txg(spa)); /* * The only state that can actually be changing concurrently * with metaslab_sync() is the metaslab's ms_allocatable. No * other thread can be modifying this txg's alloc, freeing, * freed, or space_map_phys_t. We drop ms_lock whenever we * could call into the DMU, because the DMU can call down to * us (e.g. via zio_free()) at any time. * * The spa_vdev_remove_thread() can be reading metaslab state * concurrently, and it is locked out by the ms_sync_lock. * Note that the ms_lock is insufficient for this, because it * is dropped by space_map_write(). */ tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); /* * Generate a log space map if one doesn't exist already. */ spa_generate_syncing_log_sm(spa, tx); if (msp->ms_sm == NULL) { uint64_t new_object = space_map_alloc(mos, spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP) ? zfs_metaslab_sm_blksz_with_log : zfs_metaslab_sm_blksz_no_log, tx); VERIFY3U(new_object, !=, 0); dmu_write(mos, vd->vdev_ms_array, sizeof (uint64_t) * msp->ms_id, sizeof (uint64_t), &new_object, tx); VERIFY0(space_map_open(&msp->ms_sm, mos, new_object, msp->ms_start, msp->ms_size, vd->vdev_ashift)); ASSERT(msp->ms_sm != NULL); ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs)); ASSERT(range_tree_is_empty(msp->ms_unflushed_frees)); ASSERT0(metaslab_allocated_space(msp)); } if (metaslab_unflushed_txg(msp) == 0 && spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) { ASSERT(spa_syncing_log_sm(spa) != NULL); metaslab_set_unflushed_txg(msp, spa_syncing_txg(spa), tx); spa_log_sm_increment_current_mscount(spa); spa_log_summary_add_flushed_metaslab(spa); ASSERT(msp->ms_sm != NULL); mutex_enter(&spa->spa_flushed_ms_lock); avl_add(&spa->spa_metaslabs_by_flushed, msp); mutex_exit(&spa->spa_flushed_ms_lock); ASSERT(range_tree_is_empty(msp->ms_unflushed_allocs)); ASSERT(range_tree_is_empty(msp->ms_unflushed_frees)); } if (!range_tree_is_empty(msp->ms_checkpointing) && vd->vdev_checkpoint_sm == NULL) { ASSERT(spa_has_checkpoint(spa)); uint64_t new_object = space_map_alloc(mos, zfs_vdev_standard_sm_blksz, tx); VERIFY3U(new_object, !=, 0); VERIFY0(space_map_open(&vd->vdev_checkpoint_sm, mos, new_object, 0, vd->vdev_asize, vd->vdev_ashift)); ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); /* * We save the space map object as an entry in vdev_top_zap * so it can be retrieved when the pool is reopened after an * export or through zdb. */ VERIFY0(zap_add(vd->vdev_spa->spa_meta_objset, vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (new_object), 1, &new_object, tx)); } mutex_enter(&msp->ms_sync_lock); mutex_enter(&msp->ms_lock); /* * Note: metaslab_condense() clears the space map's histogram. * Therefore we must verify and remove this histogram before * condensing. */ metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); metaslab_group_histogram_remove(mg, msp); if (spa->spa_sync_pass == 1 && msp->ms_loaded && metaslab_should_condense(msp)) metaslab_condense(msp, tx); /* * We'll be going to disk to sync our space accounting, thus we * drop the ms_lock during that time so allocations coming from * open-context (ZIL) for future TXGs do not block. */ mutex_exit(&msp->ms_lock); space_map_t *log_sm = spa_syncing_log_sm(spa); if (log_sm != NULL) { ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP)); space_map_write(log_sm, alloctree, SM_ALLOC, vd->vdev_id, tx); space_map_write(log_sm, msp->ms_freeing, SM_FREE, vd->vdev_id, tx); mutex_enter(&msp->ms_lock); ASSERT3U(spa->spa_unflushed_stats.sus_memused, >=, metaslab_unflushed_changes_memused(msp)); spa->spa_unflushed_stats.sus_memused -= metaslab_unflushed_changes_memused(msp); range_tree_remove_xor_add(alloctree, msp->ms_unflushed_frees, msp->ms_unflushed_allocs); range_tree_remove_xor_add(msp->ms_freeing, msp->ms_unflushed_allocs, msp->ms_unflushed_frees); spa->spa_unflushed_stats.sus_memused += metaslab_unflushed_changes_memused(msp); } else { ASSERT(!spa_feature_is_enabled(spa, SPA_FEATURE_LOG_SPACEMAP)); space_map_write(msp->ms_sm, alloctree, SM_ALLOC, SM_NO_VDEVID, tx); space_map_write(msp->ms_sm, msp->ms_freeing, SM_FREE, SM_NO_VDEVID, tx); mutex_enter(&msp->ms_lock); } msp->ms_allocated_space += range_tree_space(alloctree); ASSERT3U(msp->ms_allocated_space, >=, range_tree_space(msp->ms_freeing)); msp->ms_allocated_space -= range_tree_space(msp->ms_freeing); if (!range_tree_is_empty(msp->ms_checkpointing)) { ASSERT(spa_has_checkpoint(spa)); ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); /* * Since we are doing writes to disk and the ms_checkpointing * tree won't be changing during that time, we drop the * ms_lock while writing to the checkpoint space map, for the * same reason mentioned above. */ mutex_exit(&msp->ms_lock); space_map_write(vd->vdev_checkpoint_sm, msp->ms_checkpointing, SM_FREE, SM_NO_VDEVID, tx); mutex_enter(&msp->ms_lock); spa->spa_checkpoint_info.sci_dspace += range_tree_space(msp->ms_checkpointing); vd->vdev_stat.vs_checkpoint_space += range_tree_space(msp->ms_checkpointing); ASSERT3U(vd->vdev_stat.vs_checkpoint_space, ==, -space_map_allocated(vd->vdev_checkpoint_sm)); range_tree_vacate(msp->ms_checkpointing, NULL, NULL); } if (msp->ms_loaded) { /* * When the space map is loaded, we have an accurate * histogram in the range tree. This gives us an opportunity * to bring the space map's histogram up-to-date so we clear * it first before updating it. */ space_map_histogram_clear(msp->ms_sm); space_map_histogram_add(msp->ms_sm, msp->ms_allocatable, tx); /* * Since we've cleared the histogram we need to add back * any free space that has already been processed, plus * any deferred space. This allows the on-disk histogram * to accurately reflect all free space even if some space * is not yet available for allocation (i.e. deferred). */ space_map_histogram_add(msp->ms_sm, msp->ms_freed, tx); /* * Add back any deferred free space that has not been * added back into the in-core free tree yet. This will * ensure that we don't end up with a space map histogram * that is completely empty unless the metaslab is fully * allocated. */ for (int t = 0; t < TXG_DEFER_SIZE; t++) { space_map_histogram_add(msp->ms_sm, msp->ms_defer[t], tx); } } /* * Always add the free space from this sync pass to the space * map histogram. We want to make sure that the on-disk histogram * accounts for all free space. If the space map is not loaded, * then we will lose some accuracy but will correct it the next * time we load the space map. */ space_map_histogram_add(msp->ms_sm, msp->ms_freeing, tx); metaslab_aux_histograms_update(msp); metaslab_group_histogram_add(mg, msp); metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); /* * For sync pass 1, we avoid traversing this txg's free range tree * and instead will just swap the pointers for freeing and freed. * We can safely do this since the freed_tree is guaranteed to be * empty on the initial pass. * * Keep in mind that even if we are currently using a log spacemap * we want current frees to end up in the ms_allocatable (but not * get appended to the ms_sm) so their ranges can be reused as usual. */ if (spa_sync_pass(spa) == 1) { range_tree_swap(&msp->ms_freeing, &msp->ms_freed); ASSERT0(msp->ms_allocated_this_txg); } else { range_tree_vacate(msp->ms_freeing, range_tree_add, msp->ms_freed); } msp->ms_allocated_this_txg += range_tree_space(alloctree); range_tree_vacate(alloctree, NULL, NULL); ASSERT0(range_tree_space(msp->ms_allocating[txg & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_allocating[TXG_CLEAN(txg) & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_freeing)); ASSERT0(range_tree_space(msp->ms_checkpointing)); mutex_exit(&msp->ms_lock); /* * Verify that the space map object ID has been recorded in the * vdev_ms_array. */ uint64_t object; VERIFY0(dmu_read(mos, vd->vdev_ms_array, msp->ms_id * sizeof (uint64_t), sizeof (uint64_t), &object, 0)); VERIFY3U(object, ==, space_map_object(msp->ms_sm)); mutex_exit(&msp->ms_sync_lock); dmu_tx_commit(tx); } static void metaslab_evict(metaslab_t *msp, uint64_t txg) { if (!msp->ms_loaded || msp->ms_disabled != 0) return; for (int t = 1; t < TXG_CONCURRENT_STATES; t++) { VERIFY0(range_tree_space( msp->ms_allocating[(txg + t) & TXG_MASK])); } if (msp->ms_allocator != -1) metaslab_passivate(msp, msp->ms_weight & ~METASLAB_ACTIVE_MASK); if (!metaslab_debug_unload) metaslab_unload(msp); } /* * Called after a transaction group has completely synced to mark * all of the metaslab's free space as usable. */ void metaslab_sync_done(metaslab_t *msp, uint64_t txg) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; range_tree_t **defer_tree; int64_t alloc_delta, defer_delta; boolean_t defer_allowed = B_TRUE; ASSERT(!vd->vdev_ishole); mutex_enter(&msp->ms_lock); /* * If this metaslab is just becoming available, initialize its * range trees and add its capacity to the vdev. */ if (msp->ms_freed == NULL) { range_seg_type_t type; uint64_t shift, start; type = metaslab_calculate_range_tree_type(vd, msp, &start, &shift); for (int t = 0; t < TXG_SIZE; t++) { ASSERT(msp->ms_allocating[t] == NULL); msp->ms_allocating[t] = range_tree_create(NULL, type, NULL, start, shift); } ASSERT3P(msp->ms_freeing, ==, NULL); msp->ms_freeing = range_tree_create(NULL, type, NULL, start, shift); ASSERT3P(msp->ms_freed, ==, NULL); msp->ms_freed = range_tree_create(NULL, type, NULL, start, shift); for (int t = 0; t < TXG_DEFER_SIZE; t++) { ASSERT3P(msp->ms_defer[t], ==, NULL); msp->ms_defer[t] = range_tree_create(NULL, type, NULL, start, shift); } ASSERT3P(msp->ms_checkpointing, ==, NULL); msp->ms_checkpointing = range_tree_create(NULL, type, NULL, start, shift); ASSERT3P(msp->ms_unflushed_allocs, ==, NULL); msp->ms_unflushed_allocs = range_tree_create(NULL, type, NULL, start, shift); metaslab_rt_arg_t *mrap = kmem_zalloc(sizeof (*mrap), KM_SLEEP); mrap->mra_bt = &msp->ms_unflushed_frees_by_size; mrap->mra_floor_shift = metaslab_by_size_min_shift; ASSERT3P(msp->ms_unflushed_frees, ==, NULL); msp->ms_unflushed_frees = range_tree_create(&metaslab_rt_ops, type, mrap, start, shift); metaslab_space_update(vd, mg->mg_class, 0, 0, msp->ms_size); } ASSERT0(range_tree_space(msp->ms_freeing)); ASSERT0(range_tree_space(msp->ms_checkpointing)); defer_tree = &msp->ms_defer[txg % TXG_DEFER_SIZE]; uint64_t free_space = metaslab_class_get_space(spa_normal_class(spa)) - metaslab_class_get_alloc(spa_normal_class(spa)); if (free_space <= spa_get_slop_space(spa) || vd->vdev_removing) { defer_allowed = B_FALSE; } defer_delta = 0; alloc_delta = msp->ms_allocated_this_txg - range_tree_space(msp->ms_freed); if (defer_allowed) { defer_delta = range_tree_space(msp->ms_freed) - range_tree_space(*defer_tree); } else { defer_delta -= range_tree_space(*defer_tree); } metaslab_space_update(vd, mg->mg_class, alloc_delta + defer_delta, defer_delta, 0); if (spa_syncing_log_sm(spa) == NULL) { /* * If there's a metaslab_load() in progress and we don't have * a log space map, it means that we probably wrote to the * metaslab's space map. If this is the case, we need to * make sure that we wait for the load to complete so that we * have a consistent view at the in-core side of the metaslab. */ metaslab_load_wait(msp); } else { ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)); } /* * When auto-trimming is enabled, free ranges which are added to * ms_allocatable are also be added to ms_trim. The ms_trim tree is * periodically consumed by the vdev_autotrim_thread() which issues * trims for all ranges and then vacates the tree. The ms_trim tree * can be discarded at any time with the sole consequence of recent * frees not being trimmed. */ if (spa_get_autotrim(spa) == SPA_AUTOTRIM_ON) { range_tree_walk(*defer_tree, range_tree_add, msp->ms_trim); if (!defer_allowed) { range_tree_walk(msp->ms_freed, range_tree_add, msp->ms_trim); } } else { range_tree_vacate(msp->ms_trim, NULL, NULL); } /* * Move the frees from the defer_tree back to the free * range tree (if it's loaded). Swap the freed_tree and * the defer_tree -- this is safe to do because we've * just emptied out the defer_tree. */ range_tree_vacate(*defer_tree, msp->ms_loaded ? range_tree_add : NULL, msp->ms_allocatable); if (defer_allowed) { range_tree_swap(&msp->ms_freed, defer_tree); } else { range_tree_vacate(msp->ms_freed, msp->ms_loaded ? range_tree_add : NULL, msp->ms_allocatable); } msp->ms_synced_length = space_map_length(msp->ms_sm); msp->ms_deferspace += defer_delta; ASSERT3S(msp->ms_deferspace, >=, 0); ASSERT3S(msp->ms_deferspace, <=, msp->ms_size); if (msp->ms_deferspace != 0) { /* * Keep syncing this metaslab until all deferred frees * are back in circulation. */ vdev_dirty(vd, VDD_METASLAB, msp, txg + 1); } metaslab_aux_histograms_update_done(msp, defer_allowed); if (msp->ms_new) { msp->ms_new = B_FALSE; mutex_enter(&mg->mg_lock); mg->mg_ms_ready++; mutex_exit(&mg->mg_lock); } /* * Re-sort metaslab within its group now that we've adjusted * its allocatable space. */ metaslab_recalculate_weight_and_sort(msp); ASSERT0(range_tree_space(msp->ms_allocating[txg & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_freeing)); ASSERT0(range_tree_space(msp->ms_freed)); ASSERT0(range_tree_space(msp->ms_checkpointing)); msp->ms_allocating_total -= msp->ms_allocated_this_txg; msp->ms_allocated_this_txg = 0; mutex_exit(&msp->ms_lock); } void metaslab_sync_reassess(metaslab_group_t *mg) { spa_t *spa = mg->mg_class->mc_spa; spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); metaslab_group_alloc_update(mg); mg->mg_fragmentation = metaslab_group_fragmentation(mg); /* * Preload the next potential metaslabs but only on active * metaslab groups. We can get into a state where the metaslab * is no longer active since we dirty metaslabs as we remove a * a device, thus potentially making the metaslab group eligible * for preloading. */ if (mg->mg_activation_count > 0) { metaslab_group_preload(mg); } spa_config_exit(spa, SCL_ALLOC, FTAG); } /* * When writing a ditto block (i.e. more than one DVA for a given BP) on * the same vdev as an existing DVA of this BP, then try to allocate it * on a different metaslab than existing DVAs (i.e. a unique metaslab). */ static boolean_t metaslab_is_unique(metaslab_t *msp, dva_t *dva) { uint64_t dva_ms_id; if (DVA_GET_ASIZE(dva) == 0) return (B_TRUE); if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva)) return (B_TRUE); dva_ms_id = DVA_GET_OFFSET(dva) >> msp->ms_group->mg_vd->vdev_ms_shift; return (msp->ms_id != dva_ms_id); } /* * ========================================================================== * Metaslab allocation tracing facility * ========================================================================== */ /* * Add an allocation trace element to the allocation tracing list. */ static void metaslab_trace_add(zio_alloc_list_t *zal, metaslab_group_t *mg, metaslab_t *msp, uint64_t psize, uint32_t dva_id, uint64_t offset, int allocator) { metaslab_alloc_trace_t *mat; if (!metaslab_trace_enabled) return; /* * When the tracing list reaches its maximum we remove * the second element in the list before adding a new one. * By removing the second element we preserve the original * entry as a clue to what allocations steps have already been * performed. */ if (zal->zal_size == metaslab_trace_max_entries) { metaslab_alloc_trace_t *mat_next; #ifdef ZFS_DEBUG panic("too many entries in allocation list"); #endif METASLABSTAT_BUMP(metaslabstat_trace_over_limit); zal->zal_size--; mat_next = list_next(&zal->zal_list, list_head(&zal->zal_list)); list_remove(&zal->zal_list, mat_next); kmem_cache_free(metaslab_alloc_trace_cache, mat_next); } mat = kmem_cache_alloc(metaslab_alloc_trace_cache, KM_SLEEP); list_link_init(&mat->mat_list_node); mat->mat_mg = mg; mat->mat_msp = msp; mat->mat_size = psize; mat->mat_dva_id = dva_id; mat->mat_offset = offset; mat->mat_weight = 0; mat->mat_allocator = allocator; if (msp != NULL) mat->mat_weight = msp->ms_weight; /* * The list is part of the zio so locking is not required. Only * a single thread will perform allocations for a given zio. */ list_insert_tail(&zal->zal_list, mat); zal->zal_size++; ASSERT3U(zal->zal_size, <=, metaslab_trace_max_entries); } void metaslab_trace_init(zio_alloc_list_t *zal) { list_create(&zal->zal_list, sizeof (metaslab_alloc_trace_t), offsetof(metaslab_alloc_trace_t, mat_list_node)); zal->zal_size = 0; } void metaslab_trace_fini(zio_alloc_list_t *zal) { metaslab_alloc_trace_t *mat; while ((mat = list_remove_head(&zal->zal_list)) != NULL) kmem_cache_free(metaslab_alloc_trace_cache, mat); list_destroy(&zal->zal_list); zal->zal_size = 0; } /* * ========================================================================== * Metaslab block operations * ========================================================================== */ static void metaslab_group_alloc_increment(spa_t *spa, uint64_t vdev, void *tag, int flags, int allocator) { if (!(flags & METASLAB_ASYNC_ALLOC) || (flags & METASLAB_DONT_THROTTLE)) return; metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; if (!mg->mg_class->mc_alloc_throttle_enabled) return; metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator]; (void) zfs_refcount_add(&mga->mga_alloc_queue_depth, tag); } static void metaslab_group_increment_qdepth(metaslab_group_t *mg, int allocator) { metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator]; metaslab_class_allocator_t *mca = &mg->mg_class->mc_allocator[allocator]; uint64_t max = mg->mg_max_alloc_queue_depth; uint64_t cur = mga->mga_cur_max_alloc_queue_depth; while (cur < max) { if (atomic_cas_64(&mga->mga_cur_max_alloc_queue_depth, cur, cur + 1) == cur) { atomic_inc_64(&mca->mca_alloc_max_slots); return; } cur = mga->mga_cur_max_alloc_queue_depth; } } void metaslab_group_alloc_decrement(spa_t *spa, uint64_t vdev, void *tag, int flags, int allocator, boolean_t io_complete) { if (!(flags & METASLAB_ASYNC_ALLOC) || (flags & METASLAB_DONT_THROTTLE)) return; metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; if (!mg->mg_class->mc_alloc_throttle_enabled) return; metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator]; (void) zfs_refcount_remove(&mga->mga_alloc_queue_depth, tag); if (io_complete) metaslab_group_increment_qdepth(mg, allocator); } void metaslab_group_alloc_verify(spa_t *spa, const blkptr_t *bp, void *tag, int allocator) { #ifdef ZFS_DEBUG const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); for (int d = 0; d < ndvas; d++) { uint64_t vdev = DVA_GET_VDEV(&dva[d]); metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator]; VERIFY(zfs_refcount_not_held(&mga->mga_alloc_queue_depth, tag)); } #endif } static uint64_t metaslab_block_alloc(metaslab_t *msp, uint64_t size, uint64_t txg) { uint64_t start; range_tree_t *rt = msp->ms_allocatable; metaslab_class_t *mc = msp->ms_group->mg_class; ASSERT(MUTEX_HELD(&msp->ms_lock)); VERIFY(!msp->ms_condensing); VERIFY0(msp->ms_disabled); start = mc->mc_ops->msop_alloc(msp, size); if (start != -1ULL) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; VERIFY0(P2PHASE(start, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); VERIFY3U(range_tree_space(rt) - size, <=, msp->ms_size); range_tree_remove(rt, start, size); range_tree_clear(msp->ms_trim, start, size); if (range_tree_is_empty(msp->ms_allocating[txg & TXG_MASK])) vdev_dirty(mg->mg_vd, VDD_METASLAB, msp, txg); range_tree_add(msp->ms_allocating[txg & TXG_MASK], start, size); msp->ms_allocating_total += size; /* Track the last successful allocation */ msp->ms_alloc_txg = txg; metaslab_verify_space(msp, txg); } /* * Now that we've attempted the allocation we need to update the * metaslab's maximum block size since it may have changed. */ msp->ms_max_size = metaslab_largest_allocatable(msp); return (start); } /* * Find the metaslab with the highest weight that is less than what we've * already tried. In the common case, this means that we will examine each * metaslab at most once. Note that concurrent callers could reorder metaslabs * by activation/passivation once we have dropped the mg_lock. If a metaslab is * activated by another thread, and we fail to allocate from the metaslab we * have selected, we may not try the newly-activated metaslab, and instead * activate another metaslab. This is not optimal, but generally does not cause * any problems (a possible exception being if every metaslab is completely full * except for the newly-activated metaslab which we fail to examine). */ static metaslab_t * find_valid_metaslab(metaslab_group_t *mg, uint64_t activation_weight, dva_t *dva, int d, boolean_t want_unique, uint64_t asize, int allocator, boolean_t try_hard, zio_alloc_list_t *zal, metaslab_t *search, boolean_t *was_active) { avl_index_t idx; avl_tree_t *t = &mg->mg_metaslab_tree; metaslab_t *msp = avl_find(t, search, &idx); if (msp == NULL) msp = avl_nearest(t, idx, AVL_AFTER); int tries = 0; for (; msp != NULL; msp = AVL_NEXT(t, msp)) { int i; if (!try_hard && tries > zfs_metaslab_find_max_tries) { METASLABSTAT_BUMP(metaslabstat_too_many_tries); return (NULL); } tries++; if (!metaslab_should_allocate(msp, asize, try_hard)) { metaslab_trace_add(zal, mg, msp, asize, d, TRACE_TOO_SMALL, allocator); continue; } /* * If the selected metaslab is condensing or disabled, * skip it. */ if (msp->ms_condensing || msp->ms_disabled > 0) continue; *was_active = msp->ms_allocator != -1; /* * If we're activating as primary, this is our first allocation * from this disk, so we don't need to check how close we are. * If the metaslab under consideration was already active, * we're getting desperate enough to steal another allocator's * metaslab, so we still don't care about distances. */ if (activation_weight == METASLAB_WEIGHT_PRIMARY || *was_active) break; for (i = 0; i < d; i++) { if (want_unique && !metaslab_is_unique(msp, &dva[i])) break; /* try another metaslab */ } if (i == d) break; } if (msp != NULL) { search->ms_weight = msp->ms_weight; search->ms_start = msp->ms_start + 1; search->ms_allocator = msp->ms_allocator; search->ms_primary = msp->ms_primary; } return (msp); } static void metaslab_active_mask_verify(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0) return; if ((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0) return; if (msp->ms_weight & METASLAB_WEIGHT_PRIMARY) { VERIFY0(msp->ms_weight & METASLAB_WEIGHT_SECONDARY); VERIFY0(msp->ms_weight & METASLAB_WEIGHT_CLAIM); VERIFY3S(msp->ms_allocator, !=, -1); VERIFY(msp->ms_primary); return; } if (msp->ms_weight & METASLAB_WEIGHT_SECONDARY) { VERIFY0(msp->ms_weight & METASLAB_WEIGHT_PRIMARY); VERIFY0(msp->ms_weight & METASLAB_WEIGHT_CLAIM); VERIFY3S(msp->ms_allocator, !=, -1); VERIFY(!msp->ms_primary); return; } if (msp->ms_weight & METASLAB_WEIGHT_CLAIM) { VERIFY0(msp->ms_weight & METASLAB_WEIGHT_PRIMARY); VERIFY0(msp->ms_weight & METASLAB_WEIGHT_SECONDARY); VERIFY3S(msp->ms_allocator, ==, -1); return; } } /* ARGSUSED */ static uint64_t metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal, uint64_t asize, uint64_t txg, boolean_t want_unique, dva_t *dva, int d, int allocator, boolean_t try_hard) { metaslab_t *msp = NULL; uint64_t offset = -1ULL; uint64_t activation_weight = METASLAB_WEIGHT_PRIMARY; for (int i = 0; i < d; i++) { if (activation_weight == METASLAB_WEIGHT_PRIMARY && DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) { activation_weight = METASLAB_WEIGHT_SECONDARY; } else if (activation_weight == METASLAB_WEIGHT_SECONDARY && DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) { activation_weight = METASLAB_WEIGHT_CLAIM; break; } } /* * If we don't have enough metaslabs active to fill the entire array, we * just use the 0th slot. */ if (mg->mg_ms_ready < mg->mg_allocators * 3) allocator = 0; metaslab_group_allocator_t *mga = &mg->mg_allocator[allocator]; ASSERT3U(mg->mg_vd->vdev_ms_count, >=, 2); metaslab_t *search = kmem_alloc(sizeof (*search), KM_SLEEP); search->ms_weight = UINT64_MAX; search->ms_start = 0; /* * At the end of the metaslab tree are the already-active metaslabs, * first the primaries, then the secondaries. When we resume searching * through the tree, we need to consider ms_allocator and ms_primary so * we start in the location right after where we left off, and don't * accidentally loop forever considering the same metaslabs. */ search->ms_allocator = -1; search->ms_primary = B_TRUE; for (;;) { boolean_t was_active = B_FALSE; mutex_enter(&mg->mg_lock); if (activation_weight == METASLAB_WEIGHT_PRIMARY && mga->mga_primary != NULL) { msp = mga->mga_primary; /* * Even though we don't hold the ms_lock for the * primary metaslab, those fields should not * change while we hold the mg_lock. Thus it is * safe to make assertions on them. */ ASSERT(msp->ms_primary); ASSERT3S(msp->ms_allocator, ==, allocator); ASSERT(msp->ms_loaded); was_active = B_TRUE; ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK); } else if (activation_weight == METASLAB_WEIGHT_SECONDARY && mga->mga_secondary != NULL) { msp = mga->mga_secondary; /* * See comment above about the similar assertions * for the primary metaslab. */ ASSERT(!msp->ms_primary); ASSERT3S(msp->ms_allocator, ==, allocator); ASSERT(msp->ms_loaded); was_active = B_TRUE; ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK); } else { msp = find_valid_metaslab(mg, activation_weight, dva, d, want_unique, asize, allocator, try_hard, zal, search, &was_active); } mutex_exit(&mg->mg_lock); if (msp == NULL) { kmem_free(search, sizeof (*search)); return (-1ULL); } mutex_enter(&msp->ms_lock); metaslab_active_mask_verify(msp); /* * This code is disabled out because of issues with * tracepoints in non-gpl kernel modules. */ #if 0 DTRACE_PROBE3(ms__activation__attempt, metaslab_t *, msp, uint64_t, activation_weight, boolean_t, was_active); #endif /* * Ensure that the metaslab we have selected is still * capable of handling our request. It's possible that * another thread may have changed the weight while we * were blocked on the metaslab lock. We check the * active status first to see if we need to set_selected_txg * a new metaslab. */ if (was_active && !(msp->ms_weight & METASLAB_ACTIVE_MASK)) { ASSERT3S(msp->ms_allocator, ==, -1); mutex_exit(&msp->ms_lock); continue; } /* * If the metaslab was activated for another allocator * while we were waiting in the ms_lock above, or it's * a primary and we're seeking a secondary (or vice versa), * we go back and select a new metaslab. */ if (!was_active && (msp->ms_weight & METASLAB_ACTIVE_MASK) && (msp->ms_allocator != -1) && (msp->ms_allocator != allocator || ((activation_weight == METASLAB_WEIGHT_PRIMARY) != msp->ms_primary))) { ASSERT(msp->ms_loaded); ASSERT((msp->ms_weight & METASLAB_WEIGHT_CLAIM) || msp->ms_allocator != -1); mutex_exit(&msp->ms_lock); continue; } /* * This metaslab was used for claiming regions allocated * by the ZIL during pool import. Once these regions are * claimed we don't need to keep the CLAIM bit set * anymore. Passivate this metaslab to zero its activation * mask. */ if (msp->ms_weight & METASLAB_WEIGHT_CLAIM && activation_weight != METASLAB_WEIGHT_CLAIM) { ASSERT(msp->ms_loaded); ASSERT3S(msp->ms_allocator, ==, -1); metaslab_passivate(msp, msp->ms_weight & ~METASLAB_WEIGHT_CLAIM); mutex_exit(&msp->ms_lock); continue; } metaslab_set_selected_txg(msp, txg); int activation_error = metaslab_activate(msp, allocator, activation_weight); metaslab_active_mask_verify(msp); /* * If the metaslab was activated by another thread for * another allocator or activation_weight (EBUSY), or it * failed because another metaslab was assigned as primary * for this allocator (EEXIST) we continue using this * metaslab for our allocation, rather than going on to a * worse metaslab (we waited for that metaslab to be loaded * after all). * * If the activation failed due to an I/O error or ENOSPC we * skip to the next metaslab. */ boolean_t activated; if (activation_error == 0) { activated = B_TRUE; } else if (activation_error == EBUSY || activation_error == EEXIST) { activated = B_FALSE; } else { mutex_exit(&msp->ms_lock); continue; } ASSERT(msp->ms_loaded); /* * Now that we have the lock, recheck to see if we should * continue to use this metaslab for this allocation. The * the metaslab is now loaded so metaslab_should_allocate() * can accurately determine if the allocation attempt should * proceed. */ if (!metaslab_should_allocate(msp, asize, try_hard)) { /* Passivate this metaslab and select a new one. */ metaslab_trace_add(zal, mg, msp, asize, d, TRACE_TOO_SMALL, allocator); goto next; } /* * If this metaslab is currently condensing then pick again * as we can't manipulate this metaslab until it's committed * to disk. If this metaslab is being initialized, we shouldn't * allocate from it since the allocated region might be * overwritten after allocation. */ if (msp->ms_condensing) { metaslab_trace_add(zal, mg, msp, asize, d, TRACE_CONDENSING, allocator); if (activated) { metaslab_passivate(msp, msp->ms_weight & ~METASLAB_ACTIVE_MASK); } mutex_exit(&msp->ms_lock); continue; } else if (msp->ms_disabled > 0) { metaslab_trace_add(zal, mg, msp, asize, d, TRACE_DISABLED, allocator); if (activated) { metaslab_passivate(msp, msp->ms_weight & ~METASLAB_ACTIVE_MASK); } mutex_exit(&msp->ms_lock); continue; } offset = metaslab_block_alloc(msp, asize, txg); metaslab_trace_add(zal, mg, msp, asize, d, offset, allocator); if (offset != -1ULL) { /* Proactively passivate the metaslab, if needed */ if (activated) metaslab_segment_may_passivate(msp); break; } next: ASSERT(msp->ms_loaded); /* * This code is disabled out because of issues with * tracepoints in non-gpl kernel modules. */ #if 0 DTRACE_PROBE2(ms__alloc__failure, metaslab_t *, msp, uint64_t, asize); #endif /* * We were unable to allocate from this metaslab so determine * a new weight for this metaslab. Now that we have loaded * the metaslab we can provide a better hint to the metaslab * selector. * * For space-based metaslabs, we use the maximum block size. * This information is only available when the metaslab * is loaded and is more accurate than the generic free * space weight that was calculated by metaslab_weight(). * This information allows us to quickly compare the maximum * available allocation in the metaslab to the allocation * size being requested. * * For segment-based metaslabs, determine the new weight * based on the highest bucket in the range tree. We * explicitly use the loaded segment weight (i.e. the range * tree histogram) since it contains the space that is * currently available for allocation and is accurate * even within a sync pass. */ uint64_t weight; if (WEIGHT_IS_SPACEBASED(msp->ms_weight)) { weight = metaslab_largest_allocatable(msp); WEIGHT_SET_SPACEBASED(weight); } else { weight = metaslab_weight_from_range_tree(msp); } if (activated) { metaslab_passivate(msp, weight); } else { /* * For the case where we use the metaslab that is * active for another allocator we want to make * sure that we retain the activation mask. * * Note that we could attempt to use something like * metaslab_recalculate_weight_and_sort() that * retains the activation mask here. That function * uses metaslab_weight() to set the weight though * which is not as accurate as the calculations * above. */ weight |= msp->ms_weight & METASLAB_ACTIVE_MASK; metaslab_group_sort(mg, msp, weight); } metaslab_active_mask_verify(msp); /* * We have just failed an allocation attempt, check * that metaslab_should_allocate() agrees. Otherwise, * we may end up in an infinite loop retrying the same * metaslab. */ ASSERT(!metaslab_should_allocate(msp, asize, try_hard)); mutex_exit(&msp->ms_lock); } mutex_exit(&msp->ms_lock); kmem_free(search, sizeof (*search)); return (offset); } static uint64_t metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal, uint64_t asize, uint64_t txg, boolean_t want_unique, dva_t *dva, int d, int allocator, boolean_t try_hard) { uint64_t offset; ASSERT(mg->mg_initialized); offset = metaslab_group_alloc_normal(mg, zal, asize, txg, want_unique, dva, d, allocator, try_hard); mutex_enter(&mg->mg_lock); if (offset == -1ULL) { mg->mg_failed_allocations++; metaslab_trace_add(zal, mg, NULL, asize, d, TRACE_GROUP_FAILURE, allocator); if (asize == SPA_GANGBLOCKSIZE) { /* * This metaslab group was unable to allocate * the minimum gang block size so it must be out of * space. We must notify the allocation throttle * to start skipping allocation attempts to this * metaslab group until more space becomes available. * Note: this failure cannot be caused by the * allocation throttle since the allocation throttle * is only responsible for skipping devices and * not failing block allocations. */ mg->mg_no_free_space = B_TRUE; } } mg->mg_allocations++; mutex_exit(&mg->mg_lock); return (offset); } /* * Allocate a block for the specified i/o. */ int metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize, dva_t *dva, int d, dva_t *hintdva, uint64_t txg, int flags, zio_alloc_list_t *zal, int allocator) { metaslab_class_allocator_t *mca = &mc->mc_allocator[allocator]; metaslab_group_t *mg, *fast_mg, *rotor; vdev_t *vd; boolean_t try_hard = B_FALSE; ASSERT(!DVA_IS_VALID(&dva[d])); /* * For testing, make some blocks above a certain size be gang blocks. * This will result in more split blocks when using device removal, * and a large number of split blocks coupled with ztest-induced * damage can result in extremely long reconstruction times. This * will also test spilling from special to normal. */ if (psize >= metaslab_force_ganging && (spa_get_random(100) < 3)) { metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG, allocator); return (SET_ERROR(ENOSPC)); } /* * Start at the rotor and loop through all mgs until we find something. * Note that there's no locking on mca_rotor or mca_aliquot because * nothing actually breaks if we miss a few updates -- we just won't * allocate quite as evenly. It all balances out over time. * * If we are doing ditto or log blocks, try to spread them across * consecutive vdevs. If we're forced to reuse a vdev before we've * allocated all of our ditto blocks, then try and spread them out on * that vdev as much as possible. If it turns out to not be possible, * gradually lower our standards until anything becomes acceptable. * Also, allocating on consecutive vdevs (as opposed to random vdevs) * gives us hope of containing our fault domains to something we're * able to reason about. Otherwise, any two top-level vdev failures * will guarantee the loss of data. With consecutive allocation, * only two adjacent top-level vdev failures will result in data loss. * * If we are doing gang blocks (hintdva is non-NULL), try to keep * ourselves on the same vdev as our gang block header. That * way, we can hope for locality in vdev_cache, plus it makes our * fault domains something tractable. */ if (hintdva) { vd = vdev_lookup_top(spa, DVA_GET_VDEV(&hintdva[d])); /* * It's possible the vdev we're using as the hint no * longer exists or its mg has been closed (e.g. by * device removal). Consult the rotor when * all else fails. */ if (vd != NULL && vd->vdev_mg != NULL) { mg = vdev_get_mg(vd, mc); if (flags & METASLAB_HINTBP_AVOID && mg->mg_next != NULL) mg = mg->mg_next; } else { mg = mca->mca_rotor; } } else if (d != 0) { vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d - 1])); mg = vd->vdev_mg->mg_next; } else if (flags & METASLAB_FASTWRITE) { mg = fast_mg = mca->mca_rotor; do { if (fast_mg->mg_vd->vdev_pending_fastwrite < mg->mg_vd->vdev_pending_fastwrite) mg = fast_mg; } while ((fast_mg = fast_mg->mg_next) != mca->mca_rotor); } else { ASSERT(mca->mca_rotor != NULL); mg = mca->mca_rotor; } /* * If the hint put us into the wrong metaslab class, or into a * metaslab group that has been passivated, just follow the rotor. */ if (mg->mg_class != mc || mg->mg_activation_count <= 0) mg = mca->mca_rotor; rotor = mg; top: do { boolean_t allocatable; ASSERT(mg->mg_activation_count == 1); vd = mg->mg_vd; /* * Don't allocate from faulted devices. */ if (try_hard) { spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER); allocatable = vdev_allocatable(vd); spa_config_exit(spa, SCL_ZIO, FTAG); } else { allocatable = vdev_allocatable(vd); } /* * Determine if the selected metaslab group is eligible * for allocations. If we're ganging then don't allow * this metaslab group to skip allocations since that would * inadvertently return ENOSPC and suspend the pool * even though space is still available. */ if (allocatable && !GANG_ALLOCATION(flags) && !try_hard) { allocatable = metaslab_group_allocatable(mg, rotor, psize, allocator, d); } if (!allocatable) { metaslab_trace_add(zal, mg, NULL, psize, d, TRACE_NOT_ALLOCATABLE, allocator); goto next; } ASSERT(mg->mg_initialized); /* * Avoid writing single-copy data to a failing, * non-redundant vdev, unless we've already tried all * other vdevs. */ if ((vd->vdev_stat.vs_write_errors > 0 || vd->vdev_state < VDEV_STATE_HEALTHY) && d == 0 && !try_hard && vd->vdev_children == 0) { metaslab_trace_add(zal, mg, NULL, psize, d, TRACE_VDEV_ERROR, allocator); goto next; } ASSERT(mg->mg_class == mc); uint64_t asize = vdev_psize_to_asize(vd, psize); ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0); /* * If we don't need to try hard, then require that the * block be on a different metaslab from any other DVAs * in this BP (unique=true). If we are trying hard, then * allow any metaslab to be used (unique=false). */ uint64_t offset = metaslab_group_alloc(mg, zal, asize, txg, !try_hard, dva, d, allocator, try_hard); if (offset != -1ULL) { /* * If we've just selected this metaslab group, * figure out whether the corresponding vdev is * over- or under-used relative to the pool, * and set an allocation bias to even it out. * * Bias is also used to compensate for unequally * sized vdevs so that space is allocated fairly. */ if (mca->mca_aliquot == 0 && metaslab_bias_enabled) { vdev_stat_t *vs = &vd->vdev_stat; int64_t vs_free = vs->vs_space - vs->vs_alloc; int64_t mc_free = mc->mc_space - mc->mc_alloc; int64_t ratio; /* * Calculate how much more or less we should * try to allocate from this device during * this iteration around the rotor. * * This basically introduces a zero-centered * bias towards the devices with the most * free space, while compensating for vdev * size differences. * * Examples: * vdev V1 = 16M/128M * vdev V2 = 16M/128M * ratio(V1) = 100% ratio(V2) = 100% * * vdev V1 = 16M/128M * vdev V2 = 64M/128M * ratio(V1) = 127% ratio(V2) = 72% * * vdev V1 = 16M/128M * vdev V2 = 64M/512M * ratio(V1) = 40% ratio(V2) = 160% */ ratio = (vs_free * mc->mc_alloc_groups * 100) / (mc_free + 1); mg->mg_bias = ((ratio - 100) * (int64_t)mg->mg_aliquot) / 100; } else if (!metaslab_bias_enabled) { mg->mg_bias = 0; } if ((flags & METASLAB_FASTWRITE) || atomic_add_64_nv(&mca->mca_aliquot, asize) >= mg->mg_aliquot + mg->mg_bias) { mca->mca_rotor = mg->mg_next; mca->mca_aliquot = 0; } DVA_SET_VDEV(&dva[d], vd->vdev_id); DVA_SET_OFFSET(&dva[d], offset); DVA_SET_GANG(&dva[d], ((flags & METASLAB_GANG_HEADER) ? 1 : 0)); DVA_SET_ASIZE(&dva[d], asize); if (flags & METASLAB_FASTWRITE) { atomic_add_64(&vd->vdev_pending_fastwrite, psize); } return (0); } next: mca->mca_rotor = mg->mg_next; mca->mca_aliquot = 0; } while ((mg = mg->mg_next) != rotor); /* * If we haven't tried hard, perhaps do so now. */ if (!try_hard && (zfs_metaslab_try_hard_before_gang || GANG_ALLOCATION(flags) || (flags & METASLAB_ZIL) != 0 || psize <= 1 << spa->spa_min_ashift)) { METASLABSTAT_BUMP(metaslabstat_try_hard); try_hard = B_TRUE; goto top; } bzero(&dva[d], sizeof (dva_t)); metaslab_trace_add(zal, rotor, NULL, psize, d, TRACE_ENOSPC, allocator); return (SET_ERROR(ENOSPC)); } void metaslab_free_concrete(vdev_t *vd, uint64_t offset, uint64_t asize, boolean_t checkpoint) { metaslab_t *msp; spa_t *spa = vd->vdev_spa; ASSERT(vdev_is_concrete(vd)); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); ASSERT3U(offset >> vd->vdev_ms_shift, <, vd->vdev_ms_count); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; VERIFY(!msp->ms_condensing); VERIFY3U(offset, >=, msp->ms_start); VERIFY3U(offset + asize, <=, msp->ms_start + msp->ms_size); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(asize, 1ULL << vd->vdev_ashift)); metaslab_check_free_impl(vd, offset, asize); mutex_enter(&msp->ms_lock); if (range_tree_is_empty(msp->ms_freeing) && range_tree_is_empty(msp->ms_checkpointing)) { vdev_dirty(vd, VDD_METASLAB, msp, spa_syncing_txg(spa)); } if (checkpoint) { ASSERT(spa_has_checkpoint(spa)); range_tree_add(msp->ms_checkpointing, offset, asize); } else { range_tree_add(msp->ms_freeing, offset, asize); } mutex_exit(&msp->ms_lock); } /* ARGSUSED */ void metaslab_free_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { boolean_t *checkpoint = arg; ASSERT3P(checkpoint, !=, NULL); if (vd->vdev_ops->vdev_op_remap != NULL) vdev_indirect_mark_obsolete(vd, offset, size); else metaslab_free_impl(vd, offset, size, *checkpoint); } static void metaslab_free_impl(vdev_t *vd, uint64_t offset, uint64_t size, boolean_t checkpoint) { spa_t *spa = vd->vdev_spa; ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); if (spa_syncing_txg(spa) > spa_freeze_txg(spa)) return; if (spa->spa_vdev_removal != NULL && spa->spa_vdev_removal->svr_vdev_id == vd->vdev_id && vdev_is_concrete(vd)) { /* * Note: we check if the vdev is concrete because when * we complete the removal, we first change the vdev to be * an indirect vdev (in open context), and then (in syncing * context) clear spa_vdev_removal. */ free_from_removing_vdev(vd, offset, size); } else if (vd->vdev_ops->vdev_op_remap != NULL) { vdev_indirect_mark_obsolete(vd, offset, size); vd->vdev_ops->vdev_op_remap(vd, offset, size, metaslab_free_impl_cb, &checkpoint); } else { metaslab_free_concrete(vd, offset, size, checkpoint); } } typedef struct remap_blkptr_cb_arg { blkptr_t *rbca_bp; spa_remap_cb_t rbca_cb; vdev_t *rbca_remap_vd; uint64_t rbca_remap_offset; void *rbca_cb_arg; } remap_blkptr_cb_arg_t; static void remap_blkptr_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { remap_blkptr_cb_arg_t *rbca = arg; blkptr_t *bp = rbca->rbca_bp; /* We can not remap split blocks. */ if (size != DVA_GET_ASIZE(&bp->blk_dva[0])) return; ASSERT0(inner_offset); if (rbca->rbca_cb != NULL) { /* * At this point we know that we are not handling split * blocks and we invoke the callback on the previous * vdev which must be indirect. */ ASSERT3P(rbca->rbca_remap_vd->vdev_ops, ==, &vdev_indirect_ops); rbca->rbca_cb(rbca->rbca_remap_vd->vdev_id, rbca->rbca_remap_offset, size, rbca->rbca_cb_arg); /* set up remap_blkptr_cb_arg for the next call */ rbca->rbca_remap_vd = vd; rbca->rbca_remap_offset = offset; } /* * The phys birth time is that of dva[0]. This ensures that we know * when each dva was written, so that resilver can determine which * blocks need to be scrubbed (i.e. those written during the time * the vdev was offline). It also ensures that the key used in * the ARC hash table is unique (i.e. dva[0] + phys_birth). If * we didn't change the phys_birth, a lookup in the ARC for a * remapped BP could find the data that was previously stored at * this vdev + offset. */ vdev_t *oldvd = vdev_lookup_top(vd->vdev_spa, DVA_GET_VDEV(&bp->blk_dva[0])); vdev_indirect_births_t *vib = oldvd->vdev_indirect_births; bp->blk_phys_birth = vdev_indirect_births_physbirth(vib, DVA_GET_OFFSET(&bp->blk_dva[0]), DVA_GET_ASIZE(&bp->blk_dva[0])); DVA_SET_VDEV(&bp->blk_dva[0], vd->vdev_id); DVA_SET_OFFSET(&bp->blk_dva[0], offset); } /* * If the block pointer contains any indirect DVAs, modify them to refer to * concrete DVAs. Note that this will sometimes not be possible, leaving * the indirect DVA in place. This happens if the indirect DVA spans multiple * segments in the mapping (i.e. it is a "split block"). * * If the BP was remapped, calls the callback on the original dva (note the * callback can be called multiple times if the original indirect DVA refers * to another indirect DVA, etc). * * Returns TRUE if the BP was remapped. */ boolean_t spa_remap_blkptr(spa_t *spa, blkptr_t *bp, spa_remap_cb_t callback, void *arg) { remap_blkptr_cb_arg_t rbca; if (!zfs_remap_blkptr_enable) return (B_FALSE); if (!spa_feature_is_enabled(spa, SPA_FEATURE_OBSOLETE_COUNTS)) return (B_FALSE); /* * Dedup BP's can not be remapped, because ddt_phys_select() depends * on DVA[0] being the same in the BP as in the DDT (dedup table). */ if (BP_GET_DEDUP(bp)) return (B_FALSE); /* * Gang blocks can not be remapped, because * zio_checksum_gang_verifier() depends on the DVA[0] that's in * the BP used to read the gang block header (GBH) being the same * as the DVA[0] that we allocated for the GBH. */ if (BP_IS_GANG(bp)) return (B_FALSE); /* * Embedded BP's have no DVA to remap. */ if (BP_GET_NDVAS(bp) < 1) return (B_FALSE); /* * Note: we only remap dva[0]. If we remapped other dvas, we * would no longer know what their phys birth txg is. */ dva_t *dva = &bp->blk_dva[0]; uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); if (vd->vdev_ops->vdev_op_remap == NULL) return (B_FALSE); rbca.rbca_bp = bp; rbca.rbca_cb = callback; rbca.rbca_remap_vd = vd; rbca.rbca_remap_offset = offset; rbca.rbca_cb_arg = arg; /* * remap_blkptr_cb() will be called in order for each level of * indirection, until a concrete vdev is reached or a split block is * encountered. old_vd and old_offset are updated within the callback * as we go from the one indirect vdev to the next one (either concrete * or indirect again) in that order. */ vd->vdev_ops->vdev_op_remap(vd, offset, size, remap_blkptr_cb, &rbca); /* Check if the DVA wasn't remapped because it is a split block */ if (DVA_GET_VDEV(&rbca.rbca_bp->blk_dva[0]) == vd->vdev_id) return (B_FALSE); return (B_TRUE); } /* * Undo the allocation of a DVA which happened in the given transaction group. */ void metaslab_unalloc_dva(spa_t *spa, const dva_t *dva, uint64_t txg) { metaslab_t *msp; vdev_t *vd; uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); ASSERT(DVA_IS_VALID(dva)); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); if (txg > spa_freeze_txg(spa)) return; if ((vd = vdev_lookup_top(spa, vdev)) == NULL || !DVA_IS_VALID(dva) || (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) { zfs_panic_recover("metaslab_free_dva(): bad DVA %llu:%llu:%llu", (u_longlong_t)vdev, (u_longlong_t)offset, (u_longlong_t)size); return; } ASSERT(!vd->vdev_removing); ASSERT(vdev_is_concrete(vd)); ASSERT0(vd->vdev_indirect_config.vic_mapping_object); ASSERT3P(vd->vdev_indirect_mapping, ==, NULL); if (DVA_GET_GANG(dva)) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; mutex_enter(&msp->ms_lock); range_tree_remove(msp->ms_allocating[txg & TXG_MASK], offset, size); msp->ms_allocating_total -= size; VERIFY(!msp->ms_condensing); VERIFY3U(offset, >=, msp->ms_start); VERIFY3U(offset + size, <=, msp->ms_start + msp->ms_size); VERIFY3U(range_tree_space(msp->ms_allocatable) + size, <=, msp->ms_size); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); range_tree_add(msp->ms_allocatable, offset, size); mutex_exit(&msp->ms_lock); } /* * Free the block represented by the given DVA. */ void metaslab_free_dva(spa_t *spa, const dva_t *dva, boolean_t checkpoint) { uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd = vdev_lookup_top(spa, vdev); ASSERT(DVA_IS_VALID(dva)); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); if (DVA_GET_GANG(dva)) { size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); } metaslab_free_impl(vd, offset, size, checkpoint); } /* * Reserve some allocation slots. The reservation system must be called * before we call into the allocator. If there aren't any available slots * then the I/O will be throttled until an I/O completes and its slots are * freed up. The function returns true if it was successful in placing * the reservation. */ boolean_t metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, int allocator, zio_t *zio, int flags) { metaslab_class_allocator_t *mca = &mc->mc_allocator[allocator]; uint64_t available_slots = 0; boolean_t slot_reserved = B_FALSE; uint64_t max = mca->mca_alloc_max_slots; ASSERT(mc->mc_alloc_throttle_enabled); mutex_enter(&mc->mc_lock); uint64_t reserved_slots = zfs_refcount_count(&mca->mca_alloc_slots); if (reserved_slots < max) available_slots = max - reserved_slots; if (slots <= available_slots || GANG_ALLOCATION(flags) || flags & METASLAB_MUST_RESERVE) { /* * We reserve the slots individually so that we can unreserve * them individually when an I/O completes. */ for (int d = 0; d < slots; d++) zfs_refcount_add(&mca->mca_alloc_slots, zio); zio->io_flags |= ZIO_FLAG_IO_ALLOCATING; slot_reserved = B_TRUE; } mutex_exit(&mc->mc_lock); return (slot_reserved); } void metaslab_class_throttle_unreserve(metaslab_class_t *mc, int slots, int allocator, zio_t *zio) { metaslab_class_allocator_t *mca = &mc->mc_allocator[allocator]; ASSERT(mc->mc_alloc_throttle_enabled); mutex_enter(&mc->mc_lock); for (int d = 0; d < slots; d++) zfs_refcount_remove(&mca->mca_alloc_slots, zio); mutex_exit(&mc->mc_lock); } static int metaslab_claim_concrete(vdev_t *vd, uint64_t offset, uint64_t size, uint64_t txg) { metaslab_t *msp; spa_t *spa = vd->vdev_spa; int error = 0; if (offset >> vd->vdev_ms_shift >= vd->vdev_ms_count) return (SET_ERROR(ENXIO)); ASSERT3P(vd->vdev_ms, !=, NULL); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; mutex_enter(&msp->ms_lock); if ((txg != 0 && spa_writeable(spa)) || !msp->ms_loaded) { error = metaslab_activate(msp, 0, METASLAB_WEIGHT_CLAIM); if (error == EBUSY) { ASSERT(msp->ms_loaded); ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK); error = 0; } } if (error == 0 && !range_tree_contains(msp->ms_allocatable, offset, size)) error = SET_ERROR(ENOENT); if (error || txg == 0) { /* txg == 0 indicates dry run */ mutex_exit(&msp->ms_lock); return (error); } VERIFY(!msp->ms_condensing); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); VERIFY3U(range_tree_space(msp->ms_allocatable) - size, <=, msp->ms_size); range_tree_remove(msp->ms_allocatable, offset, size); range_tree_clear(msp->ms_trim, offset, size); if (spa_writeable(spa)) { /* don't dirty if we're zdb(8) */ metaslab_class_t *mc = msp->ms_group->mg_class; multilist_sublist_t *mls = multilist_sublist_lock_obj(mc->mc_metaslab_txg_list, msp); if (!multilist_link_active(&msp->ms_class_txg_node)) { msp->ms_selected_txg = txg; multilist_sublist_insert_head(mls, msp); } multilist_sublist_unlock(mls); if (range_tree_is_empty(msp->ms_allocating[txg & TXG_MASK])) vdev_dirty(vd, VDD_METASLAB, msp, txg); range_tree_add(msp->ms_allocating[txg & TXG_MASK], offset, size); msp->ms_allocating_total += size; } mutex_exit(&msp->ms_lock); return (0); } typedef struct metaslab_claim_cb_arg_t { uint64_t mcca_txg; int mcca_error; } metaslab_claim_cb_arg_t; /* ARGSUSED */ static void metaslab_claim_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { metaslab_claim_cb_arg_t *mcca_arg = arg; if (mcca_arg->mcca_error == 0) { mcca_arg->mcca_error = metaslab_claim_concrete(vd, offset, size, mcca_arg->mcca_txg); } } int metaslab_claim_impl(vdev_t *vd, uint64_t offset, uint64_t size, uint64_t txg) { if (vd->vdev_ops->vdev_op_remap != NULL) { metaslab_claim_cb_arg_t arg; /* * Only zdb(8) can claim on indirect vdevs. This is used * to detect leaks of mapped space (that are not accounted * for in the obsolete counts, spacemap, or bpobj). */ ASSERT(!spa_writeable(vd->vdev_spa)); arg.mcca_error = 0; arg.mcca_txg = txg; vd->vdev_ops->vdev_op_remap(vd, offset, size, metaslab_claim_impl_cb, &arg); if (arg.mcca_error == 0) { arg.mcca_error = metaslab_claim_concrete(vd, offset, size, txg); } return (arg.mcca_error); } else { return (metaslab_claim_concrete(vd, offset, size, txg)); } } /* * Intent log support: upon opening the pool after a crash, notify the SPA * of blocks that the intent log has allocated for immediate write, but * which are still considered free by the SPA because the last transaction * group didn't commit yet. */ static int metaslab_claim_dva(spa_t *spa, const dva_t *dva, uint64_t txg) { uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd; if ((vd = vdev_lookup_top(spa, vdev)) == NULL) { return (SET_ERROR(ENXIO)); } ASSERT(DVA_IS_VALID(dva)); if (DVA_GET_GANG(dva)) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); return (metaslab_claim_impl(vd, offset, size, txg)); } int metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, blkptr_t *bp, int ndvas, uint64_t txg, blkptr_t *hintbp, int flags, zio_alloc_list_t *zal, zio_t *zio, int allocator) { dva_t *dva = bp->blk_dva; dva_t *hintdva = (hintbp != NULL) ? hintbp->blk_dva : NULL; int error = 0; ASSERT(bp->blk_birth == 0); ASSERT(BP_PHYSICAL_BIRTH(bp) == 0); spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); if (mc->mc_allocator[allocator].mca_rotor == NULL) { /* no vdevs in this class */ spa_config_exit(spa, SCL_ALLOC, FTAG); return (SET_ERROR(ENOSPC)); } ASSERT(ndvas > 0 && ndvas <= spa_max_replication(spa)); ASSERT(BP_GET_NDVAS(bp) == 0); ASSERT(hintbp == NULL || ndvas <= BP_GET_NDVAS(hintbp)); ASSERT3P(zal, !=, NULL); for (int d = 0; d < ndvas; d++) { error = metaslab_alloc_dva(spa, mc, psize, dva, d, hintdva, txg, flags, zal, allocator); if (error != 0) { for (d--; d >= 0; d--) { metaslab_unalloc_dva(spa, &dva[d], txg); metaslab_group_alloc_decrement(spa, DVA_GET_VDEV(&dva[d]), zio, flags, allocator, B_FALSE); bzero(&dva[d], sizeof (dva_t)); } spa_config_exit(spa, SCL_ALLOC, FTAG); return (error); } else { /* * Update the metaslab group's queue depth * based on the newly allocated dva. */ metaslab_group_alloc_increment(spa, DVA_GET_VDEV(&dva[d]), zio, flags, allocator); } } ASSERT(error == 0); ASSERT(BP_GET_NDVAS(bp) == ndvas); spa_config_exit(spa, SCL_ALLOC, FTAG); BP_SET_BIRTH(bp, txg, 0); return (0); } void metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); ASSERT(!BP_IS_HOLE(bp)); ASSERT(!now || bp->blk_birth >= spa_syncing_txg(spa)); /* * If we have a checkpoint for the pool we need to make sure that * the blocks that we free that are part of the checkpoint won't be * reused until the checkpoint is discarded or we revert to it. * * The checkpoint flag is passed down the metaslab_free code path * and is set whenever we want to add a block to the checkpoint's * accounting. That is, we "checkpoint" blocks that existed at the * time the checkpoint was created and are therefore referenced by * the checkpointed uberblock. * * Note that, we don't checkpoint any blocks if the current * syncing txg <= spa_checkpoint_txg. We want these frees to sync * normally as they will be referenced by the checkpointed uberblock. */ boolean_t checkpoint = B_FALSE; if (bp->blk_birth <= spa->spa_checkpoint_txg && spa_syncing_txg(spa) > spa->spa_checkpoint_txg) { /* * At this point, if the block is part of the checkpoint * there is no way it was created in the current txg. */ ASSERT(!now); ASSERT3U(spa_syncing_txg(spa), ==, txg); checkpoint = B_TRUE; } spa_config_enter(spa, SCL_FREE, FTAG, RW_READER); for (int d = 0; d < ndvas; d++) { if (now) { metaslab_unalloc_dva(spa, &dva[d], txg); } else { ASSERT3U(txg, ==, spa_syncing_txg(spa)); metaslab_free_dva(spa, &dva[d], checkpoint); } } spa_config_exit(spa, SCL_FREE, FTAG); } int metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); int error = 0; ASSERT(!BP_IS_HOLE(bp)); if (txg != 0) { /* * First do a dry run to make sure all DVAs are claimable, * so we don't have to unwind from partial failures below. */ if ((error = metaslab_claim(spa, bp, 0)) != 0) return (error); } spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); for (int d = 0; d < ndvas; d++) { error = metaslab_claim_dva(spa, &dva[d], txg); if (error != 0) break; } spa_config_exit(spa, SCL_ALLOC, FTAG); ASSERT(error == 0 || txg == 0); return (error); } void metaslab_fastwrite_mark(spa_t *spa, const blkptr_t *bp) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); uint64_t psize = BP_GET_PSIZE(bp); int d; vdev_t *vd; ASSERT(!BP_IS_HOLE(bp)); ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(psize > 0); spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (d = 0; d < ndvas; d++) { if ((vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]))) == NULL) continue; atomic_add_64(&vd->vdev_pending_fastwrite, psize); } spa_config_exit(spa, SCL_VDEV, FTAG); } void metaslab_fastwrite_unmark(spa_t *spa, const blkptr_t *bp) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); uint64_t psize = BP_GET_PSIZE(bp); int d; vdev_t *vd; ASSERT(!BP_IS_HOLE(bp)); ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(psize > 0); spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (d = 0; d < ndvas; d++) { if ((vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]))) == NULL) continue; ASSERT3U(vd->vdev_pending_fastwrite, >=, psize); atomic_sub_64(&vd->vdev_pending_fastwrite, psize); } spa_config_exit(spa, SCL_VDEV, FTAG); } /* ARGSUSED */ static void metaslab_check_free_impl_cb(uint64_t inner, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { if (vd->vdev_ops == &vdev_indirect_ops) return; metaslab_check_free_impl(vd, offset, size); } static void metaslab_check_free_impl(vdev_t *vd, uint64_t offset, uint64_t size) { metaslab_t *msp; spa_t *spa __maybe_unused = vd->vdev_spa; if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0) return; if (vd->vdev_ops->vdev_op_remap != NULL) { vd->vdev_ops->vdev_op_remap(vd, offset, size, metaslab_check_free_impl_cb, NULL); return; } ASSERT(vdev_is_concrete(vd)); ASSERT3U(offset >> vd->vdev_ms_shift, <, vd->vdev_ms_count); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; mutex_enter(&msp->ms_lock); if (msp->ms_loaded) { range_tree_verify_not_present(msp->ms_allocatable, offset, size); } /* * Check all segments that currently exist in the freeing pipeline. * * It would intuitively make sense to also check the current allocating * tree since metaslab_unalloc_dva() exists for extents that are * allocated and freed in the same sync pass within the same txg. * Unfortunately there are places (e.g. the ZIL) where we allocate a * segment but then we free part of it within the same txg * [see zil_sync()]. Thus, we don't call range_tree_verify() in the * current allocating tree. */ range_tree_verify_not_present(msp->ms_freeing, offset, size); range_tree_verify_not_present(msp->ms_checkpointing, offset, size); range_tree_verify_not_present(msp->ms_freed, offset, size); for (int j = 0; j < TXG_DEFER_SIZE; j++) range_tree_verify_not_present(msp->ms_defer[j], offset, size); range_tree_verify_not_present(msp->ms_trim, offset, size); mutex_exit(&msp->ms_lock); } void metaslab_check_free(spa_t *spa, const blkptr_t *bp) { if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0) return; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (int i = 0; i < BP_GET_NDVAS(bp); i++) { uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[i]); vdev_t *vd = vdev_lookup_top(spa, vdev); uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]); uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]); if (DVA_GET_GANG(&bp->blk_dva[i])) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); ASSERT3P(vd, !=, NULL); metaslab_check_free_impl(vd, offset, size); } spa_config_exit(spa, SCL_VDEV, FTAG); } static void metaslab_group_disable_wait(metaslab_group_t *mg) { ASSERT(MUTEX_HELD(&mg->mg_ms_disabled_lock)); while (mg->mg_disabled_updating) { cv_wait(&mg->mg_ms_disabled_cv, &mg->mg_ms_disabled_lock); } } static void metaslab_group_disabled_increment(metaslab_group_t *mg) { ASSERT(MUTEX_HELD(&mg->mg_ms_disabled_lock)); ASSERT(mg->mg_disabled_updating); while (mg->mg_ms_disabled >= max_disabled_ms) { cv_wait(&mg->mg_ms_disabled_cv, &mg->mg_ms_disabled_lock); } mg->mg_ms_disabled++; ASSERT3U(mg->mg_ms_disabled, <=, max_disabled_ms); } /* * Mark the metaslab as disabled to prevent any allocations on this metaslab. * We must also track how many metaslabs are currently disabled within a * metaslab group and limit them to prevent allocation failures from * occurring because all metaslabs are disabled. */ void metaslab_disable(metaslab_t *msp) { ASSERT(!MUTEX_HELD(&msp->ms_lock)); metaslab_group_t *mg = msp->ms_group; mutex_enter(&mg->mg_ms_disabled_lock); /* * To keep an accurate count of how many threads have disabled * a specific metaslab group, we only allow one thread to mark * the metaslab group at a time. This ensures that the value of * ms_disabled will be accurate when we decide to mark a metaslab * group as disabled. To do this we force all other threads * to wait till the metaslab's mg_disabled_updating flag is no * longer set. */ metaslab_group_disable_wait(mg); mg->mg_disabled_updating = B_TRUE; if (msp->ms_disabled == 0) { metaslab_group_disabled_increment(mg); } mutex_enter(&msp->ms_lock); msp->ms_disabled++; mutex_exit(&msp->ms_lock); mg->mg_disabled_updating = B_FALSE; cv_broadcast(&mg->mg_ms_disabled_cv); mutex_exit(&mg->mg_ms_disabled_lock); } void metaslab_enable(metaslab_t *msp, boolean_t sync, boolean_t unload) { metaslab_group_t *mg = msp->ms_group; spa_t *spa = mg->mg_vd->vdev_spa; /* * Wait for the outstanding IO to be synced to prevent newly * allocated blocks from being overwritten. This used by * initialize and TRIM which are modifying unallocated space. */ if (sync) txg_wait_synced(spa_get_dsl(spa), 0); mutex_enter(&mg->mg_ms_disabled_lock); mutex_enter(&msp->ms_lock); if (--msp->ms_disabled == 0) { mg->mg_ms_disabled--; cv_broadcast(&mg->mg_ms_disabled_cv); if (unload) metaslab_unload(msp); } mutex_exit(&msp->ms_lock); mutex_exit(&mg->mg_ms_disabled_lock); } static void metaslab_update_ondisk_flush_data(metaslab_t *ms, dmu_tx_t *tx) { vdev_t *vd = ms->ms_group->mg_vd; spa_t *spa = vd->vdev_spa; objset_t *mos = spa_meta_objset(spa); ASSERT(spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)); metaslab_unflushed_phys_t entry = { .msp_unflushed_txg = metaslab_unflushed_txg(ms), }; uint64_t entry_size = sizeof (entry); uint64_t entry_offset = ms->ms_id * entry_size; uint64_t object = 0; int err = zap_lookup(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object); if (err == ENOENT) { object = dmu_object_alloc(mos, DMU_OTN_UINT64_METADATA, SPA_OLD_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx); VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object, tx)); } else { VERIFY0(err); } dmu_write(spa_meta_objset(spa), object, entry_offset, entry_size, &entry, tx); } void metaslab_set_unflushed_txg(metaslab_t *ms, uint64_t txg, dmu_tx_t *tx) { spa_t *spa = ms->ms_group->mg_vd->vdev_spa; if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) return; ms->ms_unflushed_txg = txg; metaslab_update_ondisk_flush_data(ms, tx); } uint64_t metaslab_unflushed_txg(metaslab_t *ms) { return (ms->ms_unflushed_txg); } ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, aliquot, ULONG, ZMOD_RW, "Allocation granularity (a.k.a. stripe size)"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, debug_load, INT, ZMOD_RW, "Load all metaslabs when pool is first opened"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, debug_unload, INT, ZMOD_RW, "Prevent metaslabs from being unloaded"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, preload_enabled, INT, ZMOD_RW, "Preload potential metaslabs during reassessment"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, unload_delay, INT, ZMOD_RW, "Delay in txgs after metaslab was last used before unloading"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, unload_delay_ms, INT, ZMOD_RW, "Delay in milliseconds after metaslab was last used before unloading"); /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs_mg, zfs_mg_, noalloc_threshold, INT, ZMOD_RW, "Percentage of metaslab group size that should be free to make it " "eligible for allocation"); ZFS_MODULE_PARAM(zfs_mg, zfs_mg_, fragmentation_threshold, INT, ZMOD_RW, "Percentage of metaslab group size that should be considered eligible " "for allocations unless all metaslab groups within the metaslab class " "have also crossed this threshold"); ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, fragmentation_threshold, INT, ZMOD_RW, "Fragmentation for metaslab to allow allocation"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, fragmentation_factor_enabled, INT, ZMOD_RW, "Use the fragmentation metric to prefer less fragmented metaslabs"); /* END CSTYLED */ ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, lba_weighting_enabled, INT, ZMOD_RW, "Prefer metaslabs with lower LBAs"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, bias_enabled, INT, ZMOD_RW, "Enable metaslab group biasing"); ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, segment_weight_enabled, INT, ZMOD_RW, "Enable segment-based metaslab selection"); ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, switch_threshold, INT, ZMOD_RW, "Segment-based metaslab selection maximum buckets before switching"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, force_ganging, ULONG, ZMOD_RW, "Blocks larger than this size are forced to be gang blocks"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, df_max_search, INT, ZMOD_RW, "Max distance (bytes) to search forward before using size tree"); ZFS_MODULE_PARAM(zfs_metaslab, metaslab_, df_use_largest_segment, INT, ZMOD_RW, "When looking in size tree, use largest segment instead of exact fit"); ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, max_size_cache_sec, ULONG, ZMOD_RW, "How long to trust the cached max chunk size of a metaslab"); ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, mem_limit, INT, ZMOD_RW, "Percentage of memory that can be used to store metaslab range trees"); ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, try_hard_before_gang, INT, ZMOD_RW, "Try hard to allocate before ganging"); ZFS_MODULE_PARAM(zfs_metaslab, zfs_metaslab_, find_max_tries, INT, ZMOD_RW, "Normally only consider this many of the best metaslabs in each vdev"); diff --git a/module/zfs/vdev.c b/module/zfs/vdev.c index f305da6f56e7..018e48c3867f 100644 --- a/module/zfs/vdev.c +++ b/module/zfs/vdev.c @@ -1,5349 +1,5385 @@ /* * 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 2017 Nexenta Systems, Inc. * Copyright (c) 2014 Integros [integros.com] * Copyright 2016 Toomas Soome * Copyright 2017 Joyent, Inc. * Copyright (c) 2017, Intel Corporation. * Copyright (c) 2019, Datto Inc. All rights reserved. */ #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 #include /* * One metaslab from each (normal-class) vdev is used by the ZIL. These are * called "embedded slog metaslabs", are referenced by vdev_log_mg, and are * part of the spa_embedded_log_class. The metaslab with the most free space * in each vdev is selected for this purpose when the pool is opened (or a * vdev is added). See vdev_metaslab_init(). * * Log blocks can be allocated from the following locations. Each one is tried * in order until the allocation succeeds: * 1. dedicated log vdevs, aka "slog" (spa_log_class) * 2. embedded slog metaslabs (spa_embedded_log_class) * 3. other metaslabs in normal vdevs (spa_normal_class) * * zfs_embedded_slog_min_ms disables the embedded slog if there are fewer * than this number of metaslabs in the vdev. This ensures that we don't set * aside an unreasonable amount of space for the ZIL. If set to less than * 1 << (spa_slop_shift + 1), on small pools the usable space may be reduced * (by more than 1<vdev_path != NULL) { zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, vd->vdev_path, buf); } else { zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", vd->vdev_ops->vdev_op_type, (u_longlong_t)vd->vdev_id, (u_longlong_t)vd->vdev_guid, buf); } } void vdev_dbgmsg_print_tree(vdev_t *vd, int indent) { char state[20]; if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, vd->vdev_ops->vdev_op_type); return; } switch (vd->vdev_state) { case VDEV_STATE_UNKNOWN: (void) snprintf(state, sizeof (state), "unknown"); break; case VDEV_STATE_CLOSED: (void) snprintf(state, sizeof (state), "closed"); break; case VDEV_STATE_OFFLINE: (void) snprintf(state, sizeof (state), "offline"); break; case VDEV_STATE_REMOVED: (void) snprintf(state, sizeof (state), "removed"); break; case VDEV_STATE_CANT_OPEN: (void) snprintf(state, sizeof (state), "can't open"); break; case VDEV_STATE_FAULTED: (void) snprintf(state, sizeof (state), "faulted"); break; case VDEV_STATE_DEGRADED: (void) snprintf(state, sizeof (state), "degraded"); break; case VDEV_STATE_HEALTHY: (void) snprintf(state, sizeof (state), "healthy"); break; default: (void) snprintf(state, sizeof (state), "", (uint_t)vd->vdev_state); } zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, vd->vdev_islog ? " (log)" : "", (u_longlong_t)vd->vdev_guid, vd->vdev_path ? vd->vdev_path : "N/A", state); for (uint64_t i = 0; i < vd->vdev_children; i++) vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); } /* * Virtual device management. */ static vdev_ops_t *vdev_ops_table[] = { &vdev_root_ops, &vdev_raidz_ops, &vdev_draid_ops, &vdev_draid_spare_ops, &vdev_mirror_ops, &vdev_replacing_ops, &vdev_spare_ops, &vdev_disk_ops, &vdev_file_ops, &vdev_missing_ops, &vdev_hole_ops, &vdev_indirect_ops, NULL }; /* * Given a vdev type, return the appropriate ops vector. */ static vdev_ops_t * vdev_getops(const char *type) { vdev_ops_t *ops, **opspp; for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) if (strcmp(ops->vdev_op_type, type) == 0) break; return (ops); } /* * Given a vdev and a metaslab class, find which metaslab group we're * interested in. All vdevs may belong to two different metaslab classes. * Dedicated slog devices use only the primary metaslab group, rather than a * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL. */ metaslab_group_t * vdev_get_mg(vdev_t *vd, metaslab_class_t *mc) { if (mc == spa_embedded_log_class(vd->vdev_spa) && vd->vdev_log_mg != NULL) return (vd->vdev_log_mg); else return (vd->vdev_mg); } /* ARGSUSED */ void vdev_default_xlate(vdev_t *vd, const range_seg64_t *logical_rs, range_seg64_t *physical_rs, range_seg64_t *remain_rs) { physical_rs->rs_start = logical_rs->rs_start; physical_rs->rs_end = logical_rs->rs_end; } /* * Derive the enumerated allocation bias from string input. * String origin is either the per-vdev zap or zpool(8). */ static vdev_alloc_bias_t vdev_derive_alloc_bias(const char *bias) { vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0) alloc_bias = VDEV_BIAS_LOG; else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) alloc_bias = VDEV_BIAS_SPECIAL; else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0) alloc_bias = VDEV_BIAS_DEDUP; return (alloc_bias); } /* * Default asize function: return the MAX of psize with the asize of * all children. This is what's used by anything other than RAID-Z. */ uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize) { uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); uint64_t csize; for (int c = 0; c < vd->vdev_children; c++) { csize = vdev_psize_to_asize(vd->vdev_child[c], psize); asize = MAX(asize, csize); } return (asize); } uint64_t vdev_default_min_asize(vdev_t *vd) { return (vd->vdev_min_asize); } /* * Get the minimum allocatable size. We define the allocatable size as * the vdev's asize rounded to the nearest metaslab. This allows us to * replace or attach devices which don't have the same physical size but * can still satisfy the same number of allocations. */ uint64_t vdev_get_min_asize(vdev_t *vd) { vdev_t *pvd = vd->vdev_parent; /* * If our parent is NULL (inactive spare or cache) or is the root, * just return our own asize. */ if (pvd == NULL) return (vd->vdev_asize); /* * The top-level vdev just returns the allocatable size rounded * to the nearest metaslab. */ if (vd == vd->vdev_top) return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); return (pvd->vdev_ops->vdev_op_min_asize(pvd)); } void vdev_set_min_asize(vdev_t *vd) { vd->vdev_min_asize = vdev_get_min_asize(vd); for (int c = 0; c < vd->vdev_children; c++) vdev_set_min_asize(vd->vdev_child[c]); } /* * Get the minimal allocation size for the top-level vdev. */ uint64_t vdev_get_min_alloc(vdev_t *vd) { uint64_t min_alloc = 1ULL << vd->vdev_ashift; if (vd->vdev_ops->vdev_op_min_alloc != NULL) min_alloc = vd->vdev_ops->vdev_op_min_alloc(vd); return (min_alloc); } /* * Get the parity level for a top-level vdev. */ uint64_t vdev_get_nparity(vdev_t *vd) { uint64_t nparity = 0; if (vd->vdev_ops->vdev_op_nparity != NULL) nparity = vd->vdev_ops->vdev_op_nparity(vd); return (nparity); } /* * Get the number of data disks for a top-level vdev. */ uint64_t vdev_get_ndisks(vdev_t *vd) { uint64_t ndisks = 1; if (vd->vdev_ops->vdev_op_ndisks != NULL) ndisks = vd->vdev_ops->vdev_op_ndisks(vd); return (ndisks); } vdev_t * vdev_lookup_top(spa_t *spa, uint64_t vdev) { vdev_t *rvd = spa->spa_root_vdev; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); if (vdev < rvd->vdev_children) { ASSERT(rvd->vdev_child[vdev] != NULL); return (rvd->vdev_child[vdev]); } return (NULL); } vdev_t * vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) { vdev_t *mvd; if (vd->vdev_guid == guid) return (vd); for (int c = 0; c < vd->vdev_children; c++) if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != NULL) return (mvd); return (NULL); } static int vdev_count_leaves_impl(vdev_t *vd) { int n = 0; if (vd->vdev_ops->vdev_op_leaf) return (1); for (int c = 0; c < vd->vdev_children; c++) n += vdev_count_leaves_impl(vd->vdev_child[c]); return (n); } int vdev_count_leaves(spa_t *spa) { int rc; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); rc = vdev_count_leaves_impl(spa->spa_root_vdev); spa_config_exit(spa, SCL_VDEV, FTAG); return (rc); } void vdev_add_child(vdev_t *pvd, vdev_t *cvd) { size_t oldsize, newsize; uint64_t id = cvd->vdev_id; vdev_t **newchild; ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); ASSERT(cvd->vdev_parent == NULL); cvd->vdev_parent = pvd; if (pvd == NULL) return; ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); oldsize = pvd->vdev_children * sizeof (vdev_t *); pvd->vdev_children = MAX(pvd->vdev_children, id + 1); newsize = pvd->vdev_children * sizeof (vdev_t *); newchild = kmem_alloc(newsize, KM_SLEEP); if (pvd->vdev_child != NULL) { bcopy(pvd->vdev_child, newchild, oldsize); kmem_free(pvd->vdev_child, oldsize); } pvd->vdev_child = newchild; pvd->vdev_child[id] = cvd; cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); /* * Walk up all ancestors to update guid sum. */ for (; pvd != NULL; pvd = pvd->vdev_parent) pvd->vdev_guid_sum += cvd->vdev_guid_sum; if (cvd->vdev_ops->vdev_op_leaf) { list_insert_head(&cvd->vdev_spa->spa_leaf_list, cvd); cvd->vdev_spa->spa_leaf_list_gen++; } } void vdev_remove_child(vdev_t *pvd, vdev_t *cvd) { int c; uint_t id = cvd->vdev_id; ASSERT(cvd->vdev_parent == pvd); if (pvd == NULL) return; ASSERT(id < pvd->vdev_children); ASSERT(pvd->vdev_child[id] == cvd); pvd->vdev_child[id] = NULL; cvd->vdev_parent = NULL; for (c = 0; c < pvd->vdev_children; c++) if (pvd->vdev_child[c]) break; if (c == pvd->vdev_children) { kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); pvd->vdev_child = NULL; pvd->vdev_children = 0; } if (cvd->vdev_ops->vdev_op_leaf) { spa_t *spa = cvd->vdev_spa; list_remove(&spa->spa_leaf_list, cvd); spa->spa_leaf_list_gen++; } /* * Walk up all ancestors to update guid sum. */ for (; pvd != NULL; pvd = pvd->vdev_parent) pvd->vdev_guid_sum -= cvd->vdev_guid_sum; } /* * Remove any holes in the child array. */ void vdev_compact_children(vdev_t *pvd) { vdev_t **newchild, *cvd; int oldc = pvd->vdev_children; int newc; ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); if (oldc == 0) return; for (int c = newc = 0; c < oldc; c++) if (pvd->vdev_child[c]) newc++; if (newc > 0) { newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_SLEEP); for (int c = newc = 0; c < oldc; c++) { if ((cvd = pvd->vdev_child[c]) != NULL) { newchild[newc] = cvd; cvd->vdev_id = newc++; } } } else { newchild = NULL; } kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); pvd->vdev_child = newchild; pvd->vdev_children = newc; } /* * Allocate and minimally initialize a vdev_t. */ vdev_t * vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) { vdev_t *vd; vdev_indirect_config_t *vic; vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); vic = &vd->vdev_indirect_config; if (spa->spa_root_vdev == NULL) { ASSERT(ops == &vdev_root_ops); spa->spa_root_vdev = vd; spa->spa_load_guid = spa_generate_guid(NULL); } if (guid == 0 && ops != &vdev_hole_ops) { if (spa->spa_root_vdev == vd) { /* * The root vdev's guid will also be the pool guid, * which must be unique among all pools. */ guid = spa_generate_guid(NULL); } else { /* * Any other vdev's guid must be unique within the pool. */ guid = spa_generate_guid(spa); } ASSERT(!spa_guid_exists(spa_guid(spa), guid)); } vd->vdev_spa = spa; vd->vdev_id = id; vd->vdev_guid = guid; vd->vdev_guid_sum = guid; vd->vdev_ops = ops; vd->vdev_state = VDEV_STATE_CLOSED; vd->vdev_ishole = (ops == &vdev_hole_ops); vic->vic_prev_indirect_vdev = UINT64_MAX; rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); vd->vdev_obsolete_segments = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); /* * Initialize rate limit structs for events. We rate limit ZIO delay * and checksum events so that we don't overwhelm ZED with thousands * of events when a disk is acting up. */ zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_slow_io_events_per_second, 1); zfs_ratelimit_init(&vd->vdev_checksum_rl, &zfs_checksum_events_per_second, 1); list_link_init(&vd->vdev_config_dirty_node); list_link_init(&vd->vdev_state_dirty_node); list_link_init(&vd->vdev_initialize_node); list_link_init(&vd->vdev_leaf_node); list_link_init(&vd->vdev_trim_node); mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_NOLOCKDEP, NULL); mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_initialize_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_initialize_io_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&vd->vdev_initialize_cv, NULL, CV_DEFAULT, NULL); cv_init(&vd->vdev_initialize_io_cv, NULL, CV_DEFAULT, NULL); mutex_init(&vd->vdev_trim_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_autotrim_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_trim_io_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&vd->vdev_trim_cv, NULL, CV_DEFAULT, NULL); cv_init(&vd->vdev_autotrim_cv, NULL, CV_DEFAULT, NULL); cv_init(&vd->vdev_trim_io_cv, NULL, CV_DEFAULT, NULL); mutex_init(&vd->vdev_rebuild_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&vd->vdev_rebuild_cv, NULL, CV_DEFAULT, NULL); for (int t = 0; t < DTL_TYPES; t++) { vd->vdev_dtl[t] = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); } txg_list_create(&vd->vdev_ms_list, spa, offsetof(struct metaslab, ms_txg_node)); txg_list_create(&vd->vdev_dtl_list, spa, offsetof(struct vdev, vdev_dtl_node)); vd->vdev_stat.vs_timestamp = gethrtime(); vdev_queue_init(vd); vdev_cache_init(vd); return (vd); } /* * Allocate a new vdev. The 'alloctype' is used to control whether we are * creating a new vdev or loading an existing one - the behavior is slightly * different for each case. */ int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, int alloctype) { vdev_ops_t *ops; char *type; uint64_t guid = 0, islog; vdev_t *vd; vdev_indirect_config_t *vic; char *tmp = NULL; int rc; vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE; boolean_t top_level = (parent && !parent->vdev_parent); ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) return (SET_ERROR(EINVAL)); if ((ops = vdev_getops(type)) == NULL) return (SET_ERROR(EINVAL)); /* * If this is a load, get the vdev guid from the nvlist. * Otherwise, vdev_alloc_common() will generate one for us. */ if (alloctype == VDEV_ALLOC_LOAD) { uint64_t label_id; if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || label_id != id) return (SET_ERROR(EINVAL)); if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } else if (alloctype == VDEV_ALLOC_SPARE) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } else if (alloctype == VDEV_ALLOC_L2CACHE) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } /* * The first allocated vdev must be of type 'root'. */ if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) return (SET_ERROR(EINVAL)); /* * Determine whether we're a log vdev. */ islog = 0; (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); if (islog && spa_version(spa) < SPA_VERSION_SLOGS) return (SET_ERROR(ENOTSUP)); if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) return (SET_ERROR(ENOTSUP)); if (top_level && alloctype == VDEV_ALLOC_ADD) { char *bias; /* * If creating a top-level vdev, check for allocation * classes input. */ if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS, &bias) == 0) { alloc_bias = vdev_derive_alloc_bias(bias); /* spa_vdev_add() expects feature to be enabled */ if (spa->spa_load_state != SPA_LOAD_CREATE && !spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) { return (SET_ERROR(ENOTSUP)); } } /* spa_vdev_add() expects feature to be enabled */ if (ops == &vdev_draid_ops && spa->spa_load_state != SPA_LOAD_CREATE && !spa_feature_is_enabled(spa, SPA_FEATURE_DRAID)) { return (SET_ERROR(ENOTSUP)); } } /* * Initialize the vdev specific data. This is done before calling * vdev_alloc_common() since it may fail and this simplifies the * error reporting and cleanup code paths. */ void *tsd = NULL; if (ops->vdev_op_init != NULL) { rc = ops->vdev_op_init(spa, nv, &tsd); if (rc != 0) { return (rc); } } vd = vdev_alloc_common(spa, id, guid, ops); vd->vdev_tsd = tsd; vd->vdev_islog = islog; if (top_level && alloc_bias != VDEV_BIAS_NONE) vd->vdev_alloc_bias = alloc_bias; if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) vd->vdev_path = spa_strdup(vd->vdev_path); /* * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a * fault on a vdev and want it to persist across imports (like with * zpool offline -f). */ rc = nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &tmp); if (rc == 0 && tmp != NULL && strcmp(tmp, "external") == 0) { vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; vd->vdev_faulted = 1; vd->vdev_label_aux = VDEV_AUX_EXTERNAL; } if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) vd->vdev_devid = spa_strdup(vd->vdev_devid); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, &vd->vdev_physpath) == 0) vd->vdev_physpath = spa_strdup(vd->vdev_physpath); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, &vd->vdev_enc_sysfs_path) == 0) vd->vdev_enc_sysfs_path = spa_strdup(vd->vdev_enc_sysfs_path); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) vd->vdev_fru = spa_strdup(vd->vdev_fru); /* * Set the whole_disk property. If it's not specified, leave the value * as -1. */ if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &vd->vdev_wholedisk) != 0) vd->vdev_wholedisk = -1ULL; vic = &vd->vdev_indirect_config; ASSERT0(vic->vic_mapping_object); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, &vic->vic_mapping_object); ASSERT0(vic->vic_births_object); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, &vic->vic_births_object); ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, &vic->vic_prev_indirect_vdev); /* * Look for the 'not present' flag. This will only be set if the device * was not present at the time of import. */ (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &vd->vdev_not_present); /* * Get the alignment requirement. */ (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); /* * Retrieve the vdev creation time. */ (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, &vd->vdev_crtxg); /* * If we're a top-level vdev, try to load the allocation parameters. */ if (top_level && (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, &vd->vdev_ms_array); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, &vd->vdev_ms_shift); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, &vd->vdev_asize); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, &vd->vdev_removing); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, &vd->vdev_top_zap); } else { ASSERT0(vd->vdev_top_zap); } if (top_level && alloctype != VDEV_ALLOC_ATTACH) { ASSERT(alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_ADD || alloctype == VDEV_ALLOC_SPLIT || alloctype == VDEV_ALLOC_ROOTPOOL); /* Note: metaslab_group_create() is now deferred */ } if (vd->vdev_ops->vdev_op_leaf && (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); } else { ASSERT0(vd->vdev_leaf_zap); } /* * If we're a leaf vdev, try to load the DTL object and other state. */ if (vd->vdev_ops->vdev_op_leaf && (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || alloctype == VDEV_ALLOC_ROOTPOOL)) { if (alloctype == VDEV_ALLOC_LOAD) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, &vd->vdev_dtl_object); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, &vd->vdev_unspare); } if (alloctype == VDEV_ALLOC_ROOTPOOL) { uint64_t spare = 0; if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, &spare) == 0 && spare) spa_spare_add(vd); } (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &vd->vdev_offline); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, &vd->vdev_resilver_txg); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG, &vd->vdev_rebuild_txg); if (nvlist_exists(nv, ZPOOL_CONFIG_RESILVER_DEFER)) vdev_defer_resilver(vd); /* * In general, when importing a pool we want to ignore the * persistent fault state, as the diagnosis made on another * system may not be valid in the current context. The only * exception is if we forced a vdev to a persistently faulted * state with 'zpool offline -f'. The persistent fault will * remain across imports until cleared. * * Local vdevs will remain in the faulted state. */ if (spa_load_state(spa) == SPA_LOAD_OPEN || spa_load_state(spa) == SPA_LOAD_IMPORT) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &vd->vdev_faulted); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, &vd->vdev_degraded); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &vd->vdev_removed); if (vd->vdev_faulted || vd->vdev_degraded) { char *aux; vd->vdev_label_aux = VDEV_AUX_ERR_EXCEEDED; if (nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && strcmp(aux, "external") == 0) vd->vdev_label_aux = VDEV_AUX_EXTERNAL; else vd->vdev_faulted = 0ULL; } } } /* * Add ourselves to the parent's list of children. */ vdev_add_child(parent, vd); *vdp = vd; return (0); } void vdev_free(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT3P(vd->vdev_initialize_thread, ==, NULL); ASSERT3P(vd->vdev_trim_thread, ==, NULL); ASSERT3P(vd->vdev_autotrim_thread, ==, NULL); ASSERT3P(vd->vdev_rebuild_thread, ==, NULL); /* * Scan queues are normally destroyed at the end of a scan. If the * queue exists here, that implies the vdev is being removed while * the scan is still running. */ if (vd->vdev_scan_io_queue != NULL) { mutex_enter(&vd->vdev_scan_io_queue_lock); dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); vd->vdev_scan_io_queue = NULL; mutex_exit(&vd->vdev_scan_io_queue_lock); } /* * vdev_free() implies closing the vdev first. This is simpler than * trying to ensure complicated semantics for all callers. */ vdev_close(vd); ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); /* * Free all children. */ for (int c = 0; c < vd->vdev_children; c++) vdev_free(vd->vdev_child[c]); ASSERT(vd->vdev_child == NULL); ASSERT(vd->vdev_guid_sum == vd->vdev_guid); if (vd->vdev_ops->vdev_op_fini != NULL) vd->vdev_ops->vdev_op_fini(vd); /* * Discard allocation state. */ if (vd->vdev_mg != NULL) { vdev_metaslab_fini(vd); metaslab_group_destroy(vd->vdev_mg); vd->vdev_mg = NULL; } if (vd->vdev_log_mg != NULL) { ASSERT0(vd->vdev_ms_count); metaslab_group_destroy(vd->vdev_log_mg); vd->vdev_log_mg = NULL; } ASSERT0(vd->vdev_stat.vs_space); ASSERT0(vd->vdev_stat.vs_dspace); ASSERT0(vd->vdev_stat.vs_alloc); /* * Remove this vdev from its parent's child list. */ vdev_remove_child(vd->vdev_parent, vd); ASSERT(vd->vdev_parent == NULL); ASSERT(!list_link_active(&vd->vdev_leaf_node)); /* * Clean up vdev structure. */ vdev_queue_fini(vd); vdev_cache_fini(vd); if (vd->vdev_path) spa_strfree(vd->vdev_path); if (vd->vdev_devid) spa_strfree(vd->vdev_devid); if (vd->vdev_physpath) spa_strfree(vd->vdev_physpath); if (vd->vdev_enc_sysfs_path) spa_strfree(vd->vdev_enc_sysfs_path); if (vd->vdev_fru) spa_strfree(vd->vdev_fru); if (vd->vdev_isspare) spa_spare_remove(vd); if (vd->vdev_isl2cache) spa_l2cache_remove(vd); txg_list_destroy(&vd->vdev_ms_list); txg_list_destroy(&vd->vdev_dtl_list); mutex_enter(&vd->vdev_dtl_lock); space_map_close(vd->vdev_dtl_sm); for (int t = 0; t < DTL_TYPES; t++) { range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); range_tree_destroy(vd->vdev_dtl[t]); } mutex_exit(&vd->vdev_dtl_lock); EQUIV(vd->vdev_indirect_births != NULL, vd->vdev_indirect_mapping != NULL); if (vd->vdev_indirect_births != NULL) { vdev_indirect_mapping_close(vd->vdev_indirect_mapping); vdev_indirect_births_close(vd->vdev_indirect_births); } if (vd->vdev_obsolete_sm != NULL) { ASSERT(vd->vdev_removing || vd->vdev_ops == &vdev_indirect_ops); space_map_close(vd->vdev_obsolete_sm); vd->vdev_obsolete_sm = NULL; } range_tree_destroy(vd->vdev_obsolete_segments); rw_destroy(&vd->vdev_indirect_rwlock); mutex_destroy(&vd->vdev_obsolete_lock); mutex_destroy(&vd->vdev_dtl_lock); mutex_destroy(&vd->vdev_stat_lock); mutex_destroy(&vd->vdev_probe_lock); mutex_destroy(&vd->vdev_scan_io_queue_lock); mutex_destroy(&vd->vdev_initialize_lock); mutex_destroy(&vd->vdev_initialize_io_lock); cv_destroy(&vd->vdev_initialize_io_cv); cv_destroy(&vd->vdev_initialize_cv); mutex_destroy(&vd->vdev_trim_lock); mutex_destroy(&vd->vdev_autotrim_lock); mutex_destroy(&vd->vdev_trim_io_lock); cv_destroy(&vd->vdev_trim_cv); cv_destroy(&vd->vdev_autotrim_cv); cv_destroy(&vd->vdev_trim_io_cv); mutex_destroy(&vd->vdev_rebuild_lock); cv_destroy(&vd->vdev_rebuild_cv); zfs_ratelimit_fini(&vd->vdev_delay_rl); zfs_ratelimit_fini(&vd->vdev_checksum_rl); if (vd == spa->spa_root_vdev) spa->spa_root_vdev = NULL; kmem_free(vd, sizeof (vdev_t)); } /* * Transfer top-level vdev state from svd to tvd. */ static void vdev_top_transfer(vdev_t *svd, vdev_t *tvd) { spa_t *spa = svd->vdev_spa; metaslab_t *msp; vdev_t *vd; int t; ASSERT(tvd == tvd->vdev_top); tvd->vdev_pending_fastwrite = svd->vdev_pending_fastwrite; tvd->vdev_ms_array = svd->vdev_ms_array; tvd->vdev_ms_shift = svd->vdev_ms_shift; tvd->vdev_ms_count = svd->vdev_ms_count; tvd->vdev_top_zap = svd->vdev_top_zap; svd->vdev_ms_array = 0; svd->vdev_ms_shift = 0; svd->vdev_ms_count = 0; svd->vdev_top_zap = 0; if (tvd->vdev_mg) ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); if (tvd->vdev_log_mg) ASSERT3P(tvd->vdev_log_mg, ==, svd->vdev_log_mg); tvd->vdev_mg = svd->vdev_mg; tvd->vdev_log_mg = svd->vdev_log_mg; tvd->vdev_ms = svd->vdev_ms; svd->vdev_mg = NULL; svd->vdev_log_mg = NULL; svd->vdev_ms = NULL; if (tvd->vdev_mg != NULL) tvd->vdev_mg->mg_vd = tvd; if (tvd->vdev_log_mg != NULL) tvd->vdev_log_mg->mg_vd = tvd; tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; svd->vdev_checkpoint_sm = NULL; tvd->vdev_alloc_bias = svd->vdev_alloc_bias; svd->vdev_alloc_bias = VDEV_BIAS_NONE; tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; svd->vdev_stat.vs_alloc = 0; svd->vdev_stat.vs_space = 0; svd->vdev_stat.vs_dspace = 0; /* * State which may be set on a top-level vdev that's in the * process of being removed. */ ASSERT0(tvd->vdev_indirect_config.vic_births_object); ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); ASSERT3P(tvd->vdev_indirect_births, ==, NULL); ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); ASSERT0(tvd->vdev_removing); ASSERT0(tvd->vdev_rebuilding); tvd->vdev_removing = svd->vdev_removing; tvd->vdev_rebuilding = svd->vdev_rebuilding; tvd->vdev_rebuild_config = svd->vdev_rebuild_config; tvd->vdev_indirect_config = svd->vdev_indirect_config; tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; tvd->vdev_indirect_births = svd->vdev_indirect_births; range_tree_swap(&svd->vdev_obsolete_segments, &tvd->vdev_obsolete_segments); tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; svd->vdev_indirect_config.vic_mapping_object = 0; svd->vdev_indirect_config.vic_births_object = 0; svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; svd->vdev_indirect_mapping = NULL; svd->vdev_indirect_births = NULL; svd->vdev_obsolete_sm = NULL; svd->vdev_removing = 0; svd->vdev_rebuilding = 0; for (t = 0; t < TXG_SIZE; t++) { while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) (void) txg_list_add(&tvd->vdev_ms_list, msp, t); while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); } if (list_link_active(&svd->vdev_config_dirty_node)) { vdev_config_clean(svd); vdev_config_dirty(tvd); } if (list_link_active(&svd->vdev_state_dirty_node)) { vdev_state_clean(svd); vdev_state_dirty(tvd); } tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; svd->vdev_deflate_ratio = 0; tvd->vdev_islog = svd->vdev_islog; svd->vdev_islog = 0; dsl_scan_io_queue_vdev_xfer(svd, tvd); } static void vdev_top_update(vdev_t *tvd, vdev_t *vd) { if (vd == NULL) return; vd->vdev_top = tvd; for (int c = 0; c < vd->vdev_children; c++) vdev_top_update(tvd, vd->vdev_child[c]); } /* * Add a mirror/replacing vdev above an existing vdev. There is no need to * call .vdev_op_init() since mirror/replacing vdevs do not have private state. */ vdev_t * vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) { spa_t *spa = cvd->vdev_spa; vdev_t *pvd = cvd->vdev_parent; vdev_t *mvd; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); mvd->vdev_asize = cvd->vdev_asize; mvd->vdev_min_asize = cvd->vdev_min_asize; mvd->vdev_max_asize = cvd->vdev_max_asize; mvd->vdev_psize = cvd->vdev_psize; mvd->vdev_ashift = cvd->vdev_ashift; mvd->vdev_logical_ashift = cvd->vdev_logical_ashift; mvd->vdev_physical_ashift = cvd->vdev_physical_ashift; mvd->vdev_state = cvd->vdev_state; mvd->vdev_crtxg = cvd->vdev_crtxg; vdev_remove_child(pvd, cvd); vdev_add_child(pvd, mvd); cvd->vdev_id = mvd->vdev_children; vdev_add_child(mvd, cvd); vdev_top_update(cvd->vdev_top, cvd->vdev_top); if (mvd == mvd->vdev_top) vdev_top_transfer(cvd, mvd); return (mvd); } /* * Remove a 1-way mirror/replacing vdev from the tree. */ void vdev_remove_parent(vdev_t *cvd) { vdev_t *mvd = cvd->vdev_parent; vdev_t *pvd = mvd->vdev_parent; ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); ASSERT(mvd->vdev_children == 1); ASSERT(mvd->vdev_ops == &vdev_mirror_ops || mvd->vdev_ops == &vdev_replacing_ops || mvd->vdev_ops == &vdev_spare_ops); cvd->vdev_ashift = mvd->vdev_ashift; cvd->vdev_logical_ashift = mvd->vdev_logical_ashift; cvd->vdev_physical_ashift = mvd->vdev_physical_ashift; vdev_remove_child(mvd, cvd); vdev_remove_child(pvd, mvd); /* * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. * Otherwise, we could have detached an offline device, and when we * go to import the pool we'll think we have two top-level vdevs, * instead of a different version of the same top-level vdev. */ if (mvd->vdev_top == mvd) { uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; cvd->vdev_orig_guid = cvd->vdev_guid; cvd->vdev_guid += guid_delta; cvd->vdev_guid_sum += guid_delta; /* * If pool not set for autoexpand, we need to also preserve * mvd's asize to prevent automatic expansion of cvd. * Otherwise if we are adjusting the mirror by attaching and * detaching children of non-uniform sizes, the mirror could * autoexpand, unexpectedly requiring larger devices to * re-establish the mirror. */ if (!cvd->vdev_spa->spa_autoexpand) cvd->vdev_asize = mvd->vdev_asize; } cvd->vdev_id = mvd->vdev_id; vdev_add_child(pvd, cvd); vdev_top_update(cvd->vdev_top, cvd->vdev_top); if (cvd == cvd->vdev_top) vdev_top_transfer(mvd, cvd); ASSERT(mvd->vdev_children == 0); vdev_free(mvd); } void vdev_metaslab_group_create(vdev_t *vd) { spa_t *spa = vd->vdev_spa; /* * metaslab_group_create was delayed until allocation bias was available */ if (vd->vdev_mg == NULL) { metaslab_class_t *mc; if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE) vd->vdev_alloc_bias = VDEV_BIAS_LOG; ASSERT3U(vd->vdev_islog, ==, (vd->vdev_alloc_bias == VDEV_BIAS_LOG)); switch (vd->vdev_alloc_bias) { case VDEV_BIAS_LOG: mc = spa_log_class(spa); break; case VDEV_BIAS_SPECIAL: mc = spa_special_class(spa); break; case VDEV_BIAS_DEDUP: mc = spa_dedup_class(spa); break; default: mc = spa_normal_class(spa); } vd->vdev_mg = metaslab_group_create(mc, vd, spa->spa_alloc_count); if (!vd->vdev_islog) { vd->vdev_log_mg = metaslab_group_create( spa_embedded_log_class(spa), vd, 1); } /* * The spa ashift min/max only apply for the normal metaslab * class. Class destination is late binding so ashift boundry * setting had to wait until now. */ if (vd->vdev_top == vd && vd->vdev_ashift != 0 && mc == spa_normal_class(spa) && vd->vdev_aux == NULL) { if (vd->vdev_ashift > spa->spa_max_ashift) spa->spa_max_ashift = vd->vdev_ashift; if (vd->vdev_ashift < spa->spa_min_ashift) spa->spa_min_ashift = vd->vdev_ashift; uint64_t min_alloc = vdev_get_min_alloc(vd); if (min_alloc < spa->spa_min_alloc) spa->spa_min_alloc = min_alloc; } } } int vdev_metaslab_init(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; uint64_t oldc = vd->vdev_ms_count; uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; metaslab_t **mspp; int error; boolean_t expanding = (oldc != 0); ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); /* * This vdev is not being allocated from yet or is a hole. */ if (vd->vdev_ms_shift == 0) return (0); ASSERT(!vd->vdev_ishole); ASSERT(oldc <= newc); mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); if (expanding) { bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); vmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); } vd->vdev_ms = mspp; vd->vdev_ms_count = newc; for (uint64_t m = oldc; m < newc; m++) { uint64_t object = 0; /* * vdev_ms_array may be 0 if we are creating the "fake" * metaslabs for an indirect vdev for zdb's leak detection. * See zdb_leak_init(). */ if (txg == 0 && vd->vdev_ms_array != 0) { error = dmu_read(spa->spa_meta_objset, vd->vdev_ms_array, m * sizeof (uint64_t), sizeof (uint64_t), &object, DMU_READ_PREFETCH); if (error != 0) { vdev_dbgmsg(vd, "unable to read the metaslab " "array [error=%d]", error); return (error); } } error = metaslab_init(vd->vdev_mg, m, object, txg, &(vd->vdev_ms[m])); if (error != 0) { vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", error); return (error); } } /* * Find the emptiest metaslab on the vdev and mark it for use for * embedded slog by moving it from the regular to the log metaslab * group. */ if (vd->vdev_mg->mg_class == spa_normal_class(spa) && vd->vdev_ms_count > zfs_embedded_slog_min_ms && avl_is_empty(&vd->vdev_log_mg->mg_metaslab_tree)) { uint64_t slog_msid = 0; uint64_t smallest = UINT64_MAX; /* * Note, we only search the new metaslabs, because the old * (pre-existing) ones may be active (e.g. have non-empty * range_tree's), and we don't move them to the new * metaslab_t. */ for (uint64_t m = oldc; m < newc; m++) { uint64_t alloc = space_map_allocated(vd->vdev_ms[m]->ms_sm); if (alloc < smallest) { slog_msid = m; smallest = alloc; } } metaslab_t *slog_ms = vd->vdev_ms[slog_msid]; /* * The metaslab was marked as dirty at the end of * metaslab_init(). Remove it from the dirty list so that we * can uninitialize and reinitialize it to the new class. */ if (txg != 0) { (void) txg_list_remove_this(&vd->vdev_ms_list, slog_ms, txg); } uint64_t sm_obj = space_map_object(slog_ms->ms_sm); metaslab_fini(slog_ms); VERIFY0(metaslab_init(vd->vdev_log_mg, slog_msid, sm_obj, txg, &vd->vdev_ms[slog_msid])); } if (txg == 0) spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); /* * If the vdev is being removed we don't activate * the metaslabs since we want to ensure that no new * allocations are performed on this device. */ if (!expanding && !vd->vdev_removing) { metaslab_group_activate(vd->vdev_mg); if (vd->vdev_log_mg != NULL) metaslab_group_activate(vd->vdev_log_mg); } if (txg == 0) spa_config_exit(spa, SCL_ALLOC, FTAG); /* * Regardless whether this vdev was just added or it is being * expanded, the metaslab count has changed. Recalculate the * block limit. */ spa_log_sm_set_blocklimit(spa); return (0); } void vdev_metaslab_fini(vdev_t *vd) { if (vd->vdev_checkpoint_sm != NULL) { ASSERT(spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_POOL_CHECKPOINT)); space_map_close(vd->vdev_checkpoint_sm); /* * Even though we close the space map, we need to set its * pointer to NULL. The reason is that vdev_metaslab_fini() * may be called multiple times for certain operations * (i.e. when destroying a pool) so we need to ensure that * this clause never executes twice. This logic is similar * to the one used for the vdev_ms clause below. */ vd->vdev_checkpoint_sm = NULL; } if (vd->vdev_ms != NULL) { metaslab_group_t *mg = vd->vdev_mg; metaslab_group_passivate(mg); if (vd->vdev_log_mg != NULL) { ASSERT(!vd->vdev_islog); metaslab_group_passivate(vd->vdev_log_mg); } uint64_t count = vd->vdev_ms_count; for (uint64_t m = 0; m < count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp != NULL) metaslab_fini(msp); } vmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); vd->vdev_ms = NULL; vd->vdev_ms_count = 0; for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { ASSERT0(mg->mg_histogram[i]); if (vd->vdev_log_mg != NULL) ASSERT0(vd->vdev_log_mg->mg_histogram[i]); } } ASSERT0(vd->vdev_ms_count); ASSERT3U(vd->vdev_pending_fastwrite, ==, 0); } typedef struct vdev_probe_stats { boolean_t vps_readable; boolean_t vps_writeable; int vps_flags; } vdev_probe_stats_t; static void vdev_probe_done(zio_t *zio) { spa_t *spa = zio->io_spa; vdev_t *vd = zio->io_vd; vdev_probe_stats_t *vps = zio->io_private; ASSERT(vd->vdev_probe_zio != NULL); if (zio->io_type == ZIO_TYPE_READ) { if (zio->io_error == 0) vps->vps_readable = 1; if (zio->io_error == 0 && spa_writeable(spa)) { zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, zio->io_offset, zio->io_size, zio->io_abd, ZIO_CHECKSUM_OFF, vdev_probe_done, vps, ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); } else { abd_free(zio->io_abd); } } else if (zio->io_type == ZIO_TYPE_WRITE) { if (zio->io_error == 0) vps->vps_writeable = 1; abd_free(zio->io_abd); } else if (zio->io_type == ZIO_TYPE_NULL) { zio_t *pio; zio_link_t *zl; vd->vdev_cant_read |= !vps->vps_readable; vd->vdev_cant_write |= !vps->vps_writeable; if (vdev_readable(vd) && (vdev_writeable(vd) || !spa_writeable(spa))) { zio->io_error = 0; } else { ASSERT(zio->io_error != 0); vdev_dbgmsg(vd, "failed probe"); (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, spa, vd, NULL, NULL, 0); zio->io_error = SET_ERROR(ENXIO); } mutex_enter(&vd->vdev_probe_lock); ASSERT(vd->vdev_probe_zio == zio); vd->vdev_probe_zio = NULL; mutex_exit(&vd->vdev_probe_lock); zl = NULL; while ((pio = zio_walk_parents(zio, &zl)) != NULL) if (!vdev_accessible(vd, pio)) pio->io_error = SET_ERROR(ENXIO); kmem_free(vps, sizeof (*vps)); } } /* * Determine whether this device is accessible. * * Read and write to several known locations: the pad regions of each * vdev label but the first, which we leave alone in case it contains * a VTOC. */ zio_t * vdev_probe(vdev_t *vd, zio_t *zio) { spa_t *spa = vd->vdev_spa; vdev_probe_stats_t *vps = NULL; zio_t *pio; ASSERT(vd->vdev_ops->vdev_op_leaf); /* * Don't probe the probe. */ if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) return (NULL); /* * To prevent 'probe storms' when a device fails, we create * just one probe i/o at a time. All zios that want to probe * this vdev will become parents of the probe io. */ mutex_enter(&vd->vdev_probe_lock); if ((pio = vd->vdev_probe_zio) == NULL) { vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | ZIO_FLAG_TRYHARD; if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { /* * vdev_cant_read and vdev_cant_write can only * transition from TRUE to FALSE when we have the * SCL_ZIO lock as writer; otherwise they can only * transition from FALSE to TRUE. This ensures that * any zio looking at these values can assume that * failures persist for the life of the I/O. That's * important because when a device has intermittent * connectivity problems, we want to ensure that * they're ascribed to the device (ENXIO) and not * the zio (EIO). * * Since we hold SCL_ZIO as writer here, clear both * values so the probe can reevaluate from first * principles. */ vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; vd->vdev_cant_read = B_FALSE; vd->vdev_cant_write = B_FALSE; } vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, vdev_probe_done, vps, vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); /* * We can't change the vdev state in this context, so we * kick off an async task to do it on our behalf. */ if (zio != NULL) { vd->vdev_probe_wanted = B_TRUE; spa_async_request(spa, SPA_ASYNC_PROBE); } } if (zio != NULL) zio_add_child(zio, pio); mutex_exit(&vd->vdev_probe_lock); if (vps == NULL) { ASSERT(zio != NULL); return (NULL); } for (int l = 1; l < VDEV_LABELS; l++) { zio_nowait(zio_read_phys(pio, vd, vdev_label_offset(vd->vdev_psize, l, offsetof(vdev_label_t, vl_be)), VDEV_PAD_SIZE, abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), ZIO_CHECKSUM_OFF, vdev_probe_done, vps, ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); } if (zio == NULL) return (pio); zio_nowait(pio); return (NULL); } +static void +vdev_load_child(void *arg) +{ + vdev_t *vd = arg; + + vd->vdev_load_error = vdev_load(vd); +} + static void vdev_open_child(void *arg) { vdev_t *vd = arg; vd->vdev_open_thread = curthread; vd->vdev_open_error = vdev_open(vd); vd->vdev_open_thread = NULL; } static boolean_t vdev_uses_zvols(vdev_t *vd) { #ifdef _KERNEL if (zvol_is_zvol(vd->vdev_path)) return (B_TRUE); #endif for (int c = 0; c < vd->vdev_children; c++) if (vdev_uses_zvols(vd->vdev_child[c])) return (B_TRUE); return (B_FALSE); } /* * Returns B_TRUE if the passed child should be opened. */ static boolean_t vdev_default_open_children_func(vdev_t *vd) { return (B_TRUE); } /* * Open the requested child vdevs. If any of the leaf vdevs are using * a ZFS volume then do the opens in a single thread. This avoids a * deadlock when the current thread is holding the spa_namespace_lock. */ static void vdev_open_children_impl(vdev_t *vd, vdev_open_children_func_t *open_func) { int children = vd->vdev_children; taskq_t *tq = taskq_create("vdev_open", children, minclsyspri, children, children, TASKQ_PREPOPULATE); vd->vdev_nonrot = B_TRUE; for (int c = 0; c < children; c++) { vdev_t *cvd = vd->vdev_child[c]; if (open_func(cvd) == B_FALSE) continue; if (tq == NULL || vdev_uses_zvols(vd)) { cvd->vdev_open_error = vdev_open(cvd); } else { VERIFY(taskq_dispatch(tq, vdev_open_child, cvd, TQ_SLEEP) != TASKQID_INVALID); } vd->vdev_nonrot &= cvd->vdev_nonrot; } if (tq != NULL) { taskq_wait(tq); taskq_destroy(tq); } } /* * Open all child vdevs. */ void vdev_open_children(vdev_t *vd) { vdev_open_children_impl(vd, vdev_default_open_children_func); } /* * Conditionally open a subset of child vdevs. */ void vdev_open_children_subset(vdev_t *vd, vdev_open_children_func_t *open_func) { vdev_open_children_impl(vd, open_func); } /* * Compute the raidz-deflation ratio. Note, we hard-code * in 128k (1 << 17) because it is the "typical" blocksize. * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, * otherwise it would inconsistently account for existing bp's. */ static void vdev_set_deflate_ratio(vdev_t *vd) { if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { vd->vdev_deflate_ratio = (1 << 17) / (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); } } /* * Maximize performance by inflating the configured ashift for top level * vdevs to be as close to the physical ashift as possible while maintaining * administrator defined limits and ensuring it doesn't go below the * logical ashift. */ static void vdev_ashift_optimize(vdev_t *vd) { ASSERT(vd == vd->vdev_top); if (vd->vdev_ashift < vd->vdev_physical_ashift) { vd->vdev_ashift = MIN( MAX(zfs_vdev_max_auto_ashift, vd->vdev_ashift), MAX(zfs_vdev_min_auto_ashift, vd->vdev_physical_ashift)); } else { /* * If the logical and physical ashifts are the same, then * we ensure that the top-level vdev's ashift is not smaller * than our minimum ashift value. For the unusual case * where logical ashift > physical ashift, we can't cap * the calculated ashift based on max ashift as that * would cause failures. * We still check if we need to increase it to match * the min ashift. */ vd->vdev_ashift = MAX(zfs_vdev_min_auto_ashift, vd->vdev_ashift); } } /* * Prepare a virtual device for access. */ int vdev_open(vdev_t *vd) { spa_t *spa = vd->vdev_spa; int error; uint64_t osize = 0; uint64_t max_osize = 0; uint64_t asize, max_asize, psize; uint64_t logical_ashift = 0; uint64_t physical_ashift = 0; ASSERT(vd->vdev_open_thread == curthread || spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || vd->vdev_state == VDEV_STATE_CANT_OPEN || vd->vdev_state == VDEV_STATE_OFFLINE); vd->vdev_stat.vs_aux = VDEV_AUX_NONE; vd->vdev_cant_read = B_FALSE; vd->vdev_cant_write = B_FALSE; vd->vdev_min_asize = vdev_get_min_asize(vd); /* * If this vdev is not removed, check its fault status. If it's * faulted, bail out of the open. */ if (!vd->vdev_removed && vd->vdev_faulted) { ASSERT(vd->vdev_children == 0); ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || vd->vdev_label_aux == VDEV_AUX_EXTERNAL); vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, vd->vdev_label_aux); return (SET_ERROR(ENXIO)); } else if (vd->vdev_offline) { ASSERT(vd->vdev_children == 0); vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); return (SET_ERROR(ENXIO)); } error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &logical_ashift, &physical_ashift); /* * Physical volume size should never be larger than its max size, unless * the disk has shrunk while we were reading it or the device is buggy * or damaged: either way it's not safe for use, bail out of the open. */ if (osize > max_osize) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_OPEN_FAILED); return (SET_ERROR(ENXIO)); } /* * Reset the vdev_reopening flag so that we actually close * the vdev on error. */ vd->vdev_reopening = B_FALSE; if (zio_injection_enabled && error == 0) error = zio_handle_device_injection(vd, NULL, SET_ERROR(ENXIO)); if (error) { if (vd->vdev_removed && vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) vd->vdev_removed = B_FALSE; if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, vd->vdev_stat.vs_aux); } else { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, vd->vdev_stat.vs_aux); } return (error); } vd->vdev_removed = B_FALSE; /* * Recheck the faulted flag now that we have confirmed that * the vdev is accessible. If we're faulted, bail. */ if (vd->vdev_faulted) { ASSERT(vd->vdev_children == 0); ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || vd->vdev_label_aux == VDEV_AUX_EXTERNAL); vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, vd->vdev_label_aux); return (SET_ERROR(ENXIO)); } if (vd->vdev_degraded) { ASSERT(vd->vdev_children == 0); vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, VDEV_AUX_ERR_EXCEEDED); } else { vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); } /* * For hole or missing vdevs we just return success. */ if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) return (0); for (int c = 0; c < vd->vdev_children; c++) { if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, VDEV_AUX_NONE); break; } } osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); if (vd->vdev_children == 0) { if (osize < SPA_MINDEVSIZE) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_TOO_SMALL); return (SET_ERROR(EOVERFLOW)); } psize = osize; asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); max_asize = max_osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); } else { if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_TOO_SMALL); return (SET_ERROR(EOVERFLOW)); } psize = 0; asize = osize; max_asize = max_osize; } /* * If the vdev was expanded, record this so that we can re-create the * uberblock rings in labels {2,3}, during the next sync. */ if ((psize > vd->vdev_psize) && (vd->vdev_psize != 0)) vd->vdev_copy_uberblocks = B_TRUE; vd->vdev_psize = psize; /* * Make sure the allocatable size hasn't shrunk too much. */ if (asize < vd->vdev_min_asize) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_LABEL); return (SET_ERROR(EINVAL)); } /* * We can always set the logical/physical ashift members since * their values are only used to calculate the vdev_ashift when * the device is first added to the config. These values should * not be used for anything else since they may change whenever * the device is reopened and we don't store them in the label. */ vd->vdev_physical_ashift = MAX(physical_ashift, vd->vdev_physical_ashift); vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift); if (vd->vdev_asize == 0) { /* * This is the first-ever open, so use the computed values. * For compatibility, a different ashift can be requested. */ vd->vdev_asize = asize; vd->vdev_max_asize = max_asize; /* * If the vdev_ashift was not overriden at creation time, * then set it the logical ashift and optimize the ashift. */ if (vd->vdev_ashift == 0) { vd->vdev_ashift = vd->vdev_logical_ashift; if (vd->vdev_logical_ashift > ASHIFT_MAX) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_ASHIFT_TOO_BIG); return (SET_ERROR(EDOM)); } if (vd->vdev_top == vd) { vdev_ashift_optimize(vd); } } if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN || vd->vdev_ashift > ASHIFT_MAX)) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_ASHIFT); return (SET_ERROR(EDOM)); } } else { /* * Make sure the alignment required hasn't increased. */ if (vd->vdev_ashift > vd->vdev_top->vdev_ashift && vd->vdev_ops->vdev_op_leaf) { (void) zfs_ereport_post( FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT, spa, vd, NULL, NULL, 0); vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_LABEL); return (SET_ERROR(EDOM)); } vd->vdev_max_asize = max_asize; } /* * If all children are healthy we update asize if either: * The asize has increased, due to a device expansion caused by dynamic * LUN growth or vdev replacement, and automatic expansion is enabled; * making the additional space available. * * The asize has decreased, due to a device shrink usually caused by a * vdev replace with a smaller device. This ensures that calculations * based of max_asize and asize e.g. esize are always valid. It's safe * to do this as we've already validated that asize is greater than * vdev_min_asize. */ if (vd->vdev_state == VDEV_STATE_HEALTHY && ((asize > vd->vdev_asize && (vd->vdev_expanding || spa->spa_autoexpand)) || (asize < vd->vdev_asize))) vd->vdev_asize = asize; vdev_set_min_asize(vd); /* * Ensure we can issue some IO before declaring the * vdev open for business. */ if (vd->vdev_ops->vdev_op_leaf && (error = zio_wait(vdev_probe(vd, NULL))) != 0) { vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED); return (error); } /* * Track the the minimum allocation size. */ if (vd->vdev_top == vd && vd->vdev_ashift != 0 && vd->vdev_islog == 0 && vd->vdev_aux == NULL) { uint64_t min_alloc = vdev_get_min_alloc(vd); if (min_alloc < spa->spa_min_alloc) spa->spa_min_alloc = min_alloc; } /* * If this is a leaf vdev, assess whether a resilver is needed. * But don't do this if we are doing a reopen for a scrub, since * this would just restart the scrub we are already doing. */ if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen) dsl_scan_assess_vdev(spa->spa_dsl_pool, vd); return (0); } /* * Called once the vdevs are all opened, this routine validates the label * contents. This needs to be done before vdev_load() so that we don't * inadvertently do repair I/Os to the wrong device. * * This function will only return failure if one of the vdevs indicates that it * has since been destroyed or exported. This is only possible if * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state * will be updated but the function will return 0. */ int vdev_validate(vdev_t *vd) { spa_t *spa = vd->vdev_spa; nvlist_t *label; uint64_t guid = 0, aux_guid = 0, top_guid; uint64_t state; nvlist_t *nvl; uint64_t txg; if (vdev_validate_skip) return (0); for (uint64_t c = 0; c < vd->vdev_children; c++) if (vdev_validate(vd->vdev_child[c]) != 0) return (SET_ERROR(EBADF)); /* * If the device has already failed, or was marked offline, don't do * any further validation. Otherwise, label I/O will fail and we will * overwrite the previous state. */ if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) return (0); /* * If we are performing an extreme rewind, we allow for a label that * was modified at a point after the current txg. * If config lock is not held do not check for the txg. spa_sync could * be updating the vdev's label before updating spa_last_synced_txg. */ if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) txg = UINT64_MAX; else txg = spa_last_synced_txg(spa); if ((label = vdev_label_read_config(vd, txg)) == NULL) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_LABEL); vdev_dbgmsg(vd, "vdev_validate: failed reading config for " "txg %llu", (u_longlong_t)txg); return (0); } /* * Determine if this vdev has been split off into another * pool. If so, then refuse to open it. */ if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, &aux_guid) == 0 && aux_guid == spa_guid(spa)) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_SPLIT_POOL); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); return (0); } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_POOL_GUID); return (0); } /* * If config is not trusted then ignore the spa guid check. This is * necessary because if the machine crashed during a re-guid the new * guid might have been written to all of the vdev labels, but not the * cached config. The check will be performed again once we have the * trusted config from the MOS. */ if (spa->spa_trust_config && guid != spa_guid(spa)) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " "match config (%llu != %llu)", (u_longlong_t)guid, (u_longlong_t)spa_guid(spa)); return (0); } if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, &aux_guid) != 0) aux_guid = 0; if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_GUID); return (0); } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_TOP_GUID); return (0); } /* * If this vdev just became a top-level vdev because its sibling was * detached, it will have adopted the parent's vdev guid -- but the * label may or may not be on disk yet. Fortunately, either version * of the label will have the same top guid, so if we're a top-level * vdev, we can safely compare to that instead. * However, if the config comes from a cachefile that failed to update * after the detach, a top-level vdev will appear as a non top-level * vdev in the config. Also relax the constraints if we perform an * extreme rewind. * * If we split this vdev off instead, then we also check the * original pool's guid. We don't want to consider the vdev * corrupt if it is partway through a split operation. */ if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { boolean_t mismatch = B_FALSE; if (spa->spa_trust_config && !spa->spa_extreme_rewind) { if (vd != vd->vdev_top || vd->vdev_guid != top_guid) mismatch = B_TRUE; } else { if (vd->vdev_guid != top_guid && vd->vdev_top->vdev_guid != guid) mismatch = B_TRUE; } if (mismatch) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: config guid " "doesn't match label guid"); vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", (u_longlong_t)vd->vdev_guid, (u_longlong_t)vd->vdev_top->vdev_guid); vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " "aux_guid %llu", (u_longlong_t)guid, (u_longlong_t)top_guid, (u_longlong_t)aux_guid); return (0); } } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_POOL_STATE); return (0); } nvlist_free(label); /* * If this is a verbatim import, no need to check the * state of the pool. */ if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && spa_load_state(spa) == SPA_LOAD_OPEN && state != POOL_STATE_ACTIVE) { vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " "for spa %s", (u_longlong_t)state, spa->spa_name); return (SET_ERROR(EBADF)); } /* * If we were able to open and validate a vdev that was * previously marked permanently unavailable, clear that state * now. */ if (vd->vdev_not_present) vd->vdev_not_present = 0; return (0); } static void vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) { if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, dvd->vdev_path, svd->vdev_path); spa_strfree(dvd->vdev_path); dvd->vdev_path = spa_strdup(svd->vdev_path); } } else if (svd->vdev_path != NULL) { dvd->vdev_path = spa_strdup(svd->vdev_path); zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); } } /* * Recursively copy vdev paths from one vdev to another. Source and destination * vdev trees must have same geometry otherwise return error. Intended to copy * paths from userland config into MOS config. */ int vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) { if ((svd->vdev_ops == &vdev_missing_ops) || (svd->vdev_ishole && dvd->vdev_ishole) || (dvd->vdev_ops == &vdev_indirect_ops)) return (0); if (svd->vdev_ops != dvd->vdev_ops) { vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); return (SET_ERROR(EINVAL)); } if (svd->vdev_guid != dvd->vdev_guid) { vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " "%llu)", (u_longlong_t)svd->vdev_guid, (u_longlong_t)dvd->vdev_guid); return (SET_ERROR(EINVAL)); } if (svd->vdev_children != dvd->vdev_children) { vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " "%llu != %llu", (u_longlong_t)svd->vdev_children, (u_longlong_t)dvd->vdev_children); return (SET_ERROR(EINVAL)); } for (uint64_t i = 0; i < svd->vdev_children; i++) { int error = vdev_copy_path_strict(svd->vdev_child[i], dvd->vdev_child[i]); if (error != 0) return (error); } if (svd->vdev_ops->vdev_op_leaf) vdev_copy_path_impl(svd, dvd); return (0); } static void vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) { ASSERT(stvd->vdev_top == stvd); ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); for (uint64_t i = 0; i < dvd->vdev_children; i++) { vdev_copy_path_search(stvd, dvd->vdev_child[i]); } if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) return; /* * The idea here is that while a vdev can shift positions within * a top vdev (when replacing, attaching mirror, etc.) it cannot * step outside of it. */ vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) return; ASSERT(vd->vdev_ops->vdev_op_leaf); vdev_copy_path_impl(vd, dvd); } /* * Recursively copy vdev paths from one root vdev to another. Source and * destination vdev trees may differ in geometry. For each destination leaf * vdev, search a vdev with the same guid and top vdev id in the source. * Intended to copy paths from userland config into MOS config. */ void vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) { uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); ASSERT(srvd->vdev_ops == &vdev_root_ops); ASSERT(drvd->vdev_ops == &vdev_root_ops); for (uint64_t i = 0; i < children; i++) { vdev_copy_path_search(srvd->vdev_child[i], drvd->vdev_child[i]); } } /* * Close a virtual device. */ void vdev_close(vdev_t *vd) { vdev_t *pvd = vd->vdev_parent; spa_t *spa __maybe_unused = vd->vdev_spa; ASSERT(vd != NULL); ASSERT(vd->vdev_open_thread == curthread || spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); /* * If our parent is reopening, then we are as well, unless we are * going offline. */ if (pvd != NULL && pvd->vdev_reopening) vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); vd->vdev_ops->vdev_op_close(vd); vdev_cache_purge(vd); /* * We record the previous state before we close it, so that if we are * doing a reopen(), we don't generate FMA ereports if we notice that * it's still faulted. */ vd->vdev_prevstate = vd->vdev_state; if (vd->vdev_offline) vd->vdev_state = VDEV_STATE_OFFLINE; else vd->vdev_state = VDEV_STATE_CLOSED; vd->vdev_stat.vs_aux = VDEV_AUX_NONE; } void vdev_hold(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_is_root(spa)); if (spa->spa_state == POOL_STATE_UNINITIALIZED) return; for (int c = 0; c < vd->vdev_children; c++) vdev_hold(vd->vdev_child[c]); if (vd->vdev_ops->vdev_op_leaf) vd->vdev_ops->vdev_op_hold(vd); } void vdev_rele(vdev_t *vd) { ASSERT(spa_is_root(vd->vdev_spa)); for (int c = 0; c < vd->vdev_children; c++) vdev_rele(vd->vdev_child[c]); if (vd->vdev_ops->vdev_op_leaf) vd->vdev_ops->vdev_op_rele(vd); } /* * Reopen all interior vdevs and any unopened leaves. We don't actually * reopen leaf vdevs which had previously been opened as they might deadlock * on the spa_config_lock. Instead we only obtain the leaf's physical size. * If the leaf has never been opened then open it, as usual. */ void vdev_reopen(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); /* set the reopening flag unless we're taking the vdev offline */ vd->vdev_reopening = !vd->vdev_offline; vdev_close(vd); (void) vdev_open(vd); /* * Call vdev_validate() here to make sure we have the same device. * Otherwise, a device with an invalid label could be successfully * opened in response to vdev_reopen(). */ if (vd->vdev_aux) { (void) vdev_validate_aux(vd); if (vdev_readable(vd) && vdev_writeable(vd) && vd->vdev_aux == &spa->spa_l2cache) { /* * In case the vdev is present we should evict all ARC * buffers and pointers to log blocks and reclaim their * space before restoring its contents to L2ARC. */ if (l2arc_vdev_present(vd)) { l2arc_rebuild_vdev(vd, B_TRUE); } else { l2arc_add_vdev(spa, vd); } spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); spa_async_request(spa, SPA_ASYNC_L2CACHE_TRIM); } } else { (void) vdev_validate(vd); } /* * Reassess parent vdev's health. */ vdev_propagate_state(vd); } int vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) { int error; /* * Normally, partial opens (e.g. of a mirror) are allowed. * For a create, however, we want to fail the request if * there are any components we can't open. */ error = vdev_open(vd); if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { vdev_close(vd); return (error ? error : SET_ERROR(ENXIO)); } /* * Recursively load DTLs and initialize all labels. */ if ((error = vdev_dtl_load(vd)) != 0 || (error = vdev_label_init(vd, txg, isreplacing ? VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { vdev_close(vd); return (error); } return (0); } void vdev_metaslab_set_size(vdev_t *vd) { uint64_t asize = vd->vdev_asize; uint64_t ms_count = asize >> zfs_vdev_default_ms_shift; uint64_t ms_shift; /* * There are two dimensions to the metaslab sizing calculation: * the size of the metaslab and the count of metaslabs per vdev. * * The default values used below are a good balance between memory * usage (larger metaslab size means more memory needed for loaded * metaslabs; more metaslabs means more memory needed for the * metaslab_t structs), metaslab load time (larger metaslabs take * longer to load), and metaslab sync time (more metaslabs means * more time spent syncing all of them). * * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs. * The range of the dimensions are as follows: * * 2^29 <= ms_size <= 2^34 * 16 <= ms_count <= 131,072 * * On the lower end of vdev sizes, we aim for metaslabs sizes of * at least 512MB (2^29) to minimize fragmentation effects when * testing with smaller devices. However, the count constraint * of at least 16 metaslabs will override this minimum size goal. * * On the upper end of vdev sizes, we aim for a maximum metaslab * size of 16GB. However, we will cap the total count to 2^17 * metaslabs to keep our memory footprint in check and let the * metaslab size grow from there if that limit is hit. * * The net effect of applying above constrains is summarized below. * * vdev size metaslab count * --------------|----------------- * < 8GB ~16 * 8GB - 100GB one per 512MB * 100GB - 3TB ~200 * 3TB - 2PB one per 16GB * > 2PB ~131,072 * -------------------------------- * * Finally, note that all of the above calculate the initial * number of metaslabs. Expanding a top-level vdev will result * in additional metaslabs being allocated making it possible * to exceed the zfs_vdev_ms_count_limit. */ if (ms_count < zfs_vdev_min_ms_count) ms_shift = highbit64(asize / zfs_vdev_min_ms_count); else if (ms_count > zfs_vdev_default_ms_count) ms_shift = highbit64(asize / zfs_vdev_default_ms_count); else ms_shift = zfs_vdev_default_ms_shift; if (ms_shift < SPA_MAXBLOCKSHIFT) { ms_shift = SPA_MAXBLOCKSHIFT; } else if (ms_shift > zfs_vdev_max_ms_shift) { ms_shift = zfs_vdev_max_ms_shift; /* cap the total count to constrain memory footprint */ if ((asize >> ms_shift) > zfs_vdev_ms_count_limit) ms_shift = highbit64(asize / zfs_vdev_ms_count_limit); } vd->vdev_ms_shift = ms_shift; ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); } void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) { ASSERT(vd == vd->vdev_top); /* indirect vdevs don't have metaslabs or dtls */ ASSERT(vdev_is_concrete(vd) || flags == 0); ASSERT(ISP2(flags)); ASSERT(spa_writeable(vd->vdev_spa)); if (flags & VDD_METASLAB) (void) txg_list_add(&vd->vdev_ms_list, arg, txg); if (flags & VDD_DTL) (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); } void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) { for (int c = 0; c < vd->vdev_children; c++) vdev_dirty_leaves(vd->vdev_child[c], flags, txg); if (vd->vdev_ops->vdev_op_leaf) vdev_dirty(vd->vdev_top, flags, vd, txg); } /* * DTLs. * * A vdev's DTL (dirty time log) is the set of transaction groups for which * the vdev has less than perfect replication. There are four kinds of DTL: * * DTL_MISSING: txgs for which the vdev has no valid copies of the data * * DTL_PARTIAL: txgs for which data is available, but not fully replicated * * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of * txgs that was scrubbed. * * DTL_OUTAGE: txgs which cannot currently be read, whether due to * persistent errors or just some device being offline. * Unlike the other three, the DTL_OUTAGE map is not generally * maintained; it's only computed when needed, typically to * determine whether a device can be detached. * * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device * either has the data or it doesn't. * * For interior vdevs such as mirror and RAID-Z the picture is more complex. * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because * if any child is less than fully replicated, then so is its parent. * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, * comprising only those txgs which appear in 'maxfaults' or more children; * those are the txgs we don't have enough replication to read. For example, * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); * thus, its DTL_MISSING consists of the set of txgs that appear in more than * two child DTL_MISSING maps. * * It should be clear from the above that to compute the DTLs and outage maps * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. * Therefore, that is all we keep on disk. When loading the pool, or after * a configuration change, we generate all other DTLs from first principles. */ void vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) { range_tree_t *rt = vd->vdev_dtl[t]; ASSERT(t < DTL_TYPES); ASSERT(vd != vd->vdev_spa->spa_root_vdev); ASSERT(spa_writeable(vd->vdev_spa)); mutex_enter(&vd->vdev_dtl_lock); if (!range_tree_contains(rt, txg, size)) range_tree_add(rt, txg, size); mutex_exit(&vd->vdev_dtl_lock); } boolean_t vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) { range_tree_t *rt = vd->vdev_dtl[t]; boolean_t dirty = B_FALSE; ASSERT(t < DTL_TYPES); ASSERT(vd != vd->vdev_spa->spa_root_vdev); /* * While we are loading the pool, the DTLs have not been loaded yet. * This isn't a problem but it can result in devices being tried * which are known to not have the data. In which case, the import * is relying on the checksum to ensure that we get the right data. * Note that while importing we are only reading the MOS, which is * always checksummed. */ mutex_enter(&vd->vdev_dtl_lock); if (!range_tree_is_empty(rt)) dirty = range_tree_contains(rt, txg, size); mutex_exit(&vd->vdev_dtl_lock); return (dirty); } boolean_t vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) { range_tree_t *rt = vd->vdev_dtl[t]; boolean_t empty; mutex_enter(&vd->vdev_dtl_lock); empty = range_tree_is_empty(rt); mutex_exit(&vd->vdev_dtl_lock); return (empty); } /* * Check if the txg falls within the range which must be * resilvered. DVAs outside this range can always be skipped. */ boolean_t vdev_default_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize, uint64_t phys_birth) { /* Set by sequential resilver. */ if (phys_birth == TXG_UNKNOWN) return (B_TRUE); return (vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1)); } /* * Returns B_TRUE if the vdev determines the DVA needs to be resilvered. */ boolean_t vdev_dtl_need_resilver(vdev_t *vd, const dva_t *dva, size_t psize, uint64_t phys_birth) { ASSERT(vd != vd->vdev_spa->spa_root_vdev); if (vd->vdev_ops->vdev_op_need_resilver == NULL || vd->vdev_ops->vdev_op_leaf) return (B_TRUE); return (vd->vdev_ops->vdev_op_need_resilver(vd, dva, psize, phys_birth)); } /* * Returns the lowest txg in the DTL range. */ static uint64_t vdev_dtl_min(vdev_t *vd) { ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); ASSERT0(vd->vdev_children); return (range_tree_min(vd->vdev_dtl[DTL_MISSING]) - 1); } /* * Returns the highest txg in the DTL. */ static uint64_t vdev_dtl_max(vdev_t *vd) { ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); ASSERT0(vd->vdev_children); return (range_tree_max(vd->vdev_dtl[DTL_MISSING])); } /* * Determine if a resilvering vdev should remove any DTL entries from * its range. If the vdev was resilvering for the entire duration of the * scan then it should excise that range from its DTLs. Otherwise, this * vdev is considered partially resilvered and should leave its DTL * entries intact. The comment in vdev_dtl_reassess() describes how we * excise the DTLs. */ static boolean_t vdev_dtl_should_excise(vdev_t *vd, boolean_t rebuild_done) { ASSERT0(vd->vdev_children); if (vd->vdev_state < VDEV_STATE_DEGRADED) return (B_FALSE); if (vd->vdev_resilver_deferred) return (B_FALSE); if (range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) return (B_TRUE); if (rebuild_done) { vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config; vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys; /* Rebuild not initiated by attach */ if (vd->vdev_rebuild_txg == 0) return (B_TRUE); /* * When a rebuild completes without error then all missing data * up to the rebuild max txg has been reconstructed and the DTL * is eligible for excision. */ if (vrp->vrp_rebuild_state == VDEV_REBUILD_COMPLETE && vdev_dtl_max(vd) <= vrp->vrp_max_txg) { ASSERT3U(vrp->vrp_min_txg, <=, vdev_dtl_min(vd)); ASSERT3U(vrp->vrp_min_txg, <, vd->vdev_rebuild_txg); ASSERT3U(vd->vdev_rebuild_txg, <=, vrp->vrp_max_txg); return (B_TRUE); } } else { dsl_scan_t *scn = vd->vdev_spa->spa_dsl_pool->dp_scan; dsl_scan_phys_t *scnp __maybe_unused = &scn->scn_phys; /* Resilver not initiated by attach */ if (vd->vdev_resilver_txg == 0) return (B_TRUE); /* * When a resilver is initiated the scan will assign the * scn_max_txg value to the highest txg value that exists * in all DTLs. If this device's max DTL is not part of this * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg] * then it is not eligible for excision. */ if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { ASSERT3U(scnp->scn_min_txg, <=, vdev_dtl_min(vd)); ASSERT3U(scnp->scn_min_txg, <, vd->vdev_resilver_txg); ASSERT3U(vd->vdev_resilver_txg, <=, scnp->scn_max_txg); return (B_TRUE); } } return (B_FALSE); } /* * Reassess DTLs after a config change or scrub completion. If txg == 0 no * write operations will be issued to the pool. */ void vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, boolean_t scrub_done, boolean_t rebuild_done) { spa_t *spa = vd->vdev_spa; avl_tree_t reftree; int minref; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); for (int c = 0; c < vd->vdev_children; c++) vdev_dtl_reassess(vd->vdev_child[c], txg, scrub_txg, scrub_done, rebuild_done); if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) return; if (vd->vdev_ops->vdev_op_leaf) { dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; vdev_rebuild_t *vr = &vd->vdev_top->vdev_rebuild_config; boolean_t check_excise = B_FALSE; boolean_t wasempty = B_TRUE; mutex_enter(&vd->vdev_dtl_lock); /* * If requested, pretend the scan or rebuild completed cleanly. */ if (zfs_scan_ignore_errors) { if (scn != NULL) scn->scn_phys.scn_errors = 0; if (vr != NULL) vr->vr_rebuild_phys.vrp_errors = 0; } if (scrub_txg != 0 && !range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { wasempty = B_FALSE; zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d " "dtl:%llu/%llu errors:%llu", (u_longlong_t)vd->vdev_guid, (u_longlong_t)txg, (u_longlong_t)scrub_txg, spa->spa_scrub_started, (u_longlong_t)vdev_dtl_min(vd), (u_longlong_t)vdev_dtl_max(vd), (u_longlong_t)(scn ? scn->scn_phys.scn_errors : 0)); } /* * If we've completed a scrub/resilver or a rebuild cleanly * then determine if this vdev should remove any DTLs. We * only want to excise regions on vdevs that were available * during the entire duration of this scan. */ if (rebuild_done && vr != NULL && vr->vr_rebuild_phys.vrp_errors == 0) { check_excise = B_TRUE; } else { if (spa->spa_scrub_started || (scn != NULL && scn->scn_phys.scn_errors == 0)) { check_excise = B_TRUE; } } if (scrub_txg && check_excise && vdev_dtl_should_excise(vd, rebuild_done)) { /* * We completed a scrub, resilver or rebuild up to * scrub_txg. If we did it without rebooting, then * the scrub dtl will be valid, so excise the old * region and fold in the scrub dtl. Otherwise, * leave the dtl as-is if there was an error. * * There's little trick here: to excise the beginning * of the DTL_MISSING map, we put it into a reference * tree and then add a segment with refcnt -1 that * covers the range [0, scrub_txg). This means * that each txg in that range has refcnt -1 or 0. * We then add DTL_SCRUB with a refcnt of 2, so that * entries in the range [0, scrub_txg) will have a * positive refcnt -- either 1 or 2. We then convert * the reference tree into the new DTL_MISSING map. */ space_reftree_create(&reftree); space_reftree_add_map(&reftree, vd->vdev_dtl[DTL_MISSING], 1); space_reftree_add_seg(&reftree, 0, scrub_txg, -1); space_reftree_add_map(&reftree, vd->vdev_dtl[DTL_SCRUB], 2); space_reftree_generate_map(&reftree, vd->vdev_dtl[DTL_MISSING], 1); space_reftree_destroy(&reftree); if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) { zfs_dbgmsg("update DTL_MISSING:%llu/%llu", (u_longlong_t)vdev_dtl_min(vd), (u_longlong_t)vdev_dtl_max(vd)); } else if (!wasempty) { zfs_dbgmsg("DTL_MISSING is now empty"); } } range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); range_tree_walk(vd->vdev_dtl[DTL_MISSING], range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); if (scrub_done) range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); if (!vdev_readable(vd)) range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); else range_tree_walk(vd->vdev_dtl[DTL_MISSING], range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); /* * If the vdev was resilvering or rebuilding and no longer * has any DTLs then reset the appropriate flag and dirty * the top level so that we persist the change. */ if (txg != 0 && range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) { if (vd->vdev_rebuild_txg != 0) { vd->vdev_rebuild_txg = 0; vdev_config_dirty(vd->vdev_top); } else if (vd->vdev_resilver_txg != 0) { vd->vdev_resilver_txg = 0; vdev_config_dirty(vd->vdev_top); } } mutex_exit(&vd->vdev_dtl_lock); if (txg != 0) vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); return; } mutex_enter(&vd->vdev_dtl_lock); for (int t = 0; t < DTL_TYPES; t++) { /* account for child's outage in parent's missing map */ int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; if (t == DTL_SCRUB) continue; /* leaf vdevs only */ if (t == DTL_PARTIAL) minref = 1; /* i.e. non-zero */ else if (vdev_get_nparity(vd) != 0) minref = vdev_get_nparity(vd) + 1; /* RAID-Z, dRAID */ else minref = vd->vdev_children; /* any kind of mirror */ space_reftree_create(&reftree); for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; mutex_enter(&cvd->vdev_dtl_lock); space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); mutex_exit(&cvd->vdev_dtl_lock); } space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); space_reftree_destroy(&reftree); } mutex_exit(&vd->vdev_dtl_lock); } int vdev_dtl_load(vdev_t *vd) { spa_t *spa = vd->vdev_spa; objset_t *mos = spa->spa_meta_objset; range_tree_t *rt; int error = 0; if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { ASSERT(vdev_is_concrete(vd)); error = space_map_open(&vd->vdev_dtl_sm, mos, vd->vdev_dtl_object, 0, -1ULL, 0); if (error) return (error); ASSERT(vd->vdev_dtl_sm != NULL); rt = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); error = space_map_load(vd->vdev_dtl_sm, rt, SM_ALLOC); if (error == 0) { mutex_enter(&vd->vdev_dtl_lock); range_tree_walk(rt, range_tree_add, vd->vdev_dtl[DTL_MISSING]); mutex_exit(&vd->vdev_dtl_lock); } range_tree_vacate(rt, NULL, NULL); range_tree_destroy(rt); return (error); } for (int c = 0; c < vd->vdev_children; c++) { error = vdev_dtl_load(vd->vdev_child[c]); if (error != 0) break; } return (error); } static void vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx) { spa_t *spa = vd->vdev_spa; objset_t *mos = spa->spa_meta_objset; vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; const char *string; ASSERT(alloc_bias != VDEV_BIAS_NONE); string = (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG : (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL : (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL; ASSERT(string != NULL); VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, strlen(string) + 1, string, tx)); if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) { spa_activate_allocation_classes(spa, tx); } } void vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) { spa_t *spa = vd->vdev_spa; VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, zapobj, tx)); } uint64_t vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) { spa_t *spa = vd->vdev_spa; uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); ASSERT(zap != 0); VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, zap, tx)); return (zap); } void vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) { if (vd->vdev_ops != &vdev_hole_ops && vd->vdev_ops != &vdev_missing_ops && vd->vdev_ops != &vdev_root_ops && !vd->vdev_top->vdev_removing) { if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); } if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { vd->vdev_top_zap = vdev_create_link_zap(vd, tx); if (vd->vdev_alloc_bias != VDEV_BIAS_NONE) vdev_zap_allocation_data(vd, tx); } } for (uint64_t i = 0; i < vd->vdev_children; i++) { vdev_construct_zaps(vd->vdev_child[i], tx); } } static void vdev_dtl_sync(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; objset_t *mos = spa->spa_meta_objset; range_tree_t *rtsync; dmu_tx_t *tx; uint64_t object = space_map_object(vd->vdev_dtl_sm); ASSERT(vdev_is_concrete(vd)); ASSERT(vd->vdev_ops->vdev_op_leaf); tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); if (vd->vdev_detached || vd->vdev_top->vdev_removing) { mutex_enter(&vd->vdev_dtl_lock); space_map_free(vd->vdev_dtl_sm, tx); space_map_close(vd->vdev_dtl_sm); vd->vdev_dtl_sm = NULL; mutex_exit(&vd->vdev_dtl_lock); /* * We only destroy the leaf ZAP for detached leaves or for * removed log devices. Removed data devices handle leaf ZAP * cleanup later, once cancellation is no longer possible. */ if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || vd->vdev_top->vdev_islog)) { vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); vd->vdev_leaf_zap = 0; } dmu_tx_commit(tx); return; } if (vd->vdev_dtl_sm == NULL) { uint64_t new_object; new_object = space_map_alloc(mos, zfs_vdev_dtl_sm_blksz, tx); VERIFY3U(new_object, !=, 0); VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 0, -1ULL, 0)); ASSERT(vd->vdev_dtl_sm != NULL); } rtsync = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0); mutex_enter(&vd->vdev_dtl_lock); range_tree_walk(rt, range_tree_add, rtsync); mutex_exit(&vd->vdev_dtl_lock); space_map_truncate(vd->vdev_dtl_sm, zfs_vdev_dtl_sm_blksz, tx); space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); range_tree_vacate(rtsync, NULL, NULL); range_tree_destroy(rtsync); /* * If the object for the space map has changed then dirty * the top level so that we update the config. */ if (object != space_map_object(vd->vdev_dtl_sm)) { vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " "new object %llu", (u_longlong_t)txg, spa_name(spa), (u_longlong_t)object, (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); vdev_config_dirty(vd->vdev_top); } dmu_tx_commit(tx); } /* * Determine whether the specified vdev can be offlined/detached/removed * without losing data. */ boolean_t vdev_dtl_required(vdev_t *vd) { spa_t *spa = vd->vdev_spa; vdev_t *tvd = vd->vdev_top; uint8_t cant_read = vd->vdev_cant_read; boolean_t required; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); if (vd == spa->spa_root_vdev || vd == tvd) return (B_TRUE); /* * Temporarily mark the device as unreadable, and then determine * whether this results in any DTL outages in the top-level vdev. * If not, we can safely offline/detach/remove the device. */ vd->vdev_cant_read = B_TRUE; vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE); required = !vdev_dtl_empty(tvd, DTL_OUTAGE); vd->vdev_cant_read = cant_read; vdev_dtl_reassess(tvd, 0, 0, B_FALSE, B_FALSE); if (!required && zio_injection_enabled) { required = !!zio_handle_device_injection(vd, NULL, SET_ERROR(ECHILD)); } return (required); } /* * Determine if resilver is needed, and if so the txg range. */ boolean_t vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) { boolean_t needed = B_FALSE; uint64_t thismin = UINT64_MAX; uint64_t thismax = 0; if (vd->vdev_children == 0) { mutex_enter(&vd->vdev_dtl_lock); if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && vdev_writeable(vd)) { thismin = vdev_dtl_min(vd); thismax = vdev_dtl_max(vd); needed = B_TRUE; } mutex_exit(&vd->vdev_dtl_lock); } else { for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; uint64_t cmin, cmax; if (vdev_resilver_needed(cvd, &cmin, &cmax)) { thismin = MIN(thismin, cmin); thismax = MAX(thismax, cmax); needed = B_TRUE; } } } if (needed && minp) { *minp = thismin; *maxp = thismax; } return (needed); } /* * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj * will contain either the checkpoint spacemap object or zero if none exists. * All other errors are returned to the caller. */ int vdev_checkpoint_sm_object(vdev_t *vd, uint64_t *sm_obj) { ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); if (vd->vdev_top_zap == 0) { *sm_obj = 0; return (0); } int error = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, sm_obj); if (error == ENOENT) { *sm_obj = 0; error = 0; } return (error); } int vdev_load(vdev_t *vd) { + int children = vd->vdev_children; int error = 0; + taskq_t *tq = NULL; + + /* + * It's only worthwhile to use the taskq for the root vdev, because the + * slow part is metaslab_init, and that only happens for top-level + * vdevs. + */ + if (vd->vdev_ops == &vdev_root_ops && vd->vdev_children > 0) { + tq = taskq_create("vdev_load", children, minclsyspri, + children, children, TASKQ_PREPOPULATE); + } /* * Recursively load all children. */ for (int c = 0; c < vd->vdev_children; c++) { - error = vdev_load(vd->vdev_child[c]); - if (error != 0) { - return (error); + vdev_t *cvd = vd->vdev_child[c]; + + if (tq == NULL || vdev_uses_zvols(cvd)) { + cvd->vdev_load_error = vdev_load(cvd); + } else { + VERIFY(taskq_dispatch(tq, vdev_load_child, + cvd, TQ_SLEEP) != TASKQID_INVALID); } } + if (tq != NULL) { + taskq_wait(tq); + taskq_destroy(tq); + } + + for (int c = 0; c < vd->vdev_children; c++) { + int error = vd->vdev_child[c]->vdev_load_error; + + if (error != 0) + return (error); + } + vdev_set_deflate_ratio(vd); /* * On spa_load path, grab the allocation bias from our zap */ if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { spa_t *spa = vd->vdev_spa; char bias_str[64]; error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str), bias_str); if (error == 0) { ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE); vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str); } else if (error != ENOENT) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: zap_lookup(top_zap=%llu) " "failed [error=%d]", vd->vdev_top_zap, error); return (error); } } /* * Load any rebuild state from the top-level vdev zap. */ if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { error = vdev_rebuild_load(vd); if (error && error != ENOTSUP) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: vdev_rebuild_load " "failed [error=%d]", error); return (error); } } /* * If this is a top-level vdev, initialize its metaslabs. */ if (vd == vd->vdev_top && vdev_is_concrete(vd)) { vdev_metaslab_group_create(vd); if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " "asize=%llu", (u_longlong_t)vd->vdev_ashift, (u_longlong_t)vd->vdev_asize); return (SET_ERROR(ENXIO)); } error = vdev_metaslab_init(vd, 0); if (error != 0) { vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " "[error=%d]", error); vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); return (error); } uint64_t checkpoint_sm_obj; error = vdev_checkpoint_sm_object(vd, &checkpoint_sm_obj); if (error == 0 && checkpoint_sm_obj != 0) { objset_t *mos = spa_meta_objset(vd->vdev_spa); ASSERT(vd->vdev_asize != 0); ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); error = space_map_open(&vd->vdev_checkpoint_sm, mos, checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift); if (error != 0) { vdev_dbgmsg(vd, "vdev_load: space_map_open " "failed for checkpoint spacemap (obj %llu) " "[error=%d]", (u_longlong_t)checkpoint_sm_obj, error); return (error); } ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); /* * Since the checkpoint_sm contains free entries * exclusively we can use space_map_allocated() to * indicate the cumulative checkpointed space that * has been freed. */ vd->vdev_stat.vs_checkpoint_space = -space_map_allocated(vd->vdev_checkpoint_sm); vd->vdev_spa->spa_checkpoint_info.sci_dspace += vd->vdev_stat.vs_checkpoint_space; } else if (error != 0) { vdev_dbgmsg(vd, "vdev_load: failed to retrieve " "checkpoint space map object from vdev ZAP " "[error=%d]", error); return (error); } } /* * If this is a leaf vdev, load its DTL. */ if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " "[error=%d]", error); return (error); } uint64_t obsolete_sm_object; error = vdev_obsolete_sm_object(vd, &obsolete_sm_object); if (error == 0 && obsolete_sm_object != 0) { objset_t *mos = vd->vdev_spa->spa_meta_objset; ASSERT(vd->vdev_asize != 0); ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, obsolete_sm_object, 0, vd->vdev_asize, 0))) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " "obsolete spacemap (obj %llu) [error=%d]", (u_longlong_t)obsolete_sm_object, error); return (error); } } else if (error != 0) { vdev_dbgmsg(vd, "vdev_load: failed to retrieve obsolete " "space map object from vdev ZAP [error=%d]", error); return (error); } return (0); } /* * The special vdev case is used for hot spares and l2cache devices. Its * sole purpose it to set the vdev state for the associated vdev. To do this, * we make sure that we can open the underlying device, then try to read the * label, and make sure that the label is sane and that it hasn't been * repurposed to another pool. */ int vdev_validate_aux(vdev_t *vd) { nvlist_t *label; uint64_t guid, version; uint64_t state; if (!vdev_readable(vd)) return (0); if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); return (-1); } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || !SPA_VERSION_IS_SUPPORTED(version) || nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || guid != vd->vdev_guid || nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); return (-1); } /* * We don't actually check the pool state here. If it's in fact in * use by another pool, we update this fact on the fly when requested. */ nvlist_free(label); return (0); } static void vdev_destroy_ms_flush_data(vdev_t *vd, dmu_tx_t *tx) { objset_t *mos = spa_meta_objset(vd->vdev_spa); if (vd->vdev_top_zap == 0) return; uint64_t object = 0; int err = zap_lookup(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, sizeof (uint64_t), 1, &object); if (err == ENOENT) return; VERIFY0(err); VERIFY0(dmu_object_free(mos, object, tx)); VERIFY0(zap_remove(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, tx)); } /* * Free the objects used to store this vdev's spacemaps, and the array * that points to them. */ void vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) { if (vd->vdev_ms_array == 0) return; objset_t *mos = vd->vdev_spa->spa_meta_objset; uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; size_t array_bytes = array_count * sizeof (uint64_t); uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, array_bytes, smobj_array, 0)); for (uint64_t i = 0; i < array_count; i++) { uint64_t smobj = smobj_array[i]; if (smobj == 0) continue; space_map_free_obj(mos, smobj, tx); } kmem_free(smobj_array, array_bytes); VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); vdev_destroy_ms_flush_data(vd, tx); vd->vdev_ms_array = 0; } static void vdev_remove_empty_log(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; ASSERT(vd->vdev_islog); ASSERT(vd == vd->vdev_top); ASSERT3U(txg, ==, spa_syncing_txg(spa)); dmu_tx_t *tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); vdev_destroy_spacemaps(vd, tx); if (vd->vdev_top_zap != 0) { vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); vd->vdev_top_zap = 0; } dmu_tx_commit(tx); } void vdev_sync_done(vdev_t *vd, uint64_t txg) { metaslab_t *msp; boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); ASSERT(vdev_is_concrete(vd)); while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) != NULL) metaslab_sync_done(msp, txg); if (reassess) { metaslab_sync_reassess(vd->vdev_mg); if (vd->vdev_log_mg != NULL) metaslab_sync_reassess(vd->vdev_log_mg); } } void vdev_sync(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; vdev_t *lvd; metaslab_t *msp; ASSERT3U(txg, ==, spa->spa_syncing_txg); dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); if (range_tree_space(vd->vdev_obsolete_segments) > 0) { ASSERT(vd->vdev_removing || vd->vdev_ops == &vdev_indirect_ops); vdev_indirect_sync_obsolete(vd, tx); /* * If the vdev is indirect, it can't have dirty * metaslabs or DTLs. */ if (vd->vdev_ops == &vdev_indirect_ops) { ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); dmu_tx_commit(tx); return; } } ASSERT(vdev_is_concrete(vd)); if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && !vd->vdev_removing) { ASSERT(vd == vd->vdev_top); ASSERT0(vd->vdev_indirect_config.vic_mapping_object); vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); ASSERT(vd->vdev_ms_array != 0); vdev_config_dirty(vd); } while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { metaslab_sync(msp, txg); (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); } while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) vdev_dtl_sync(lvd, txg); /* * If this is an empty log device being removed, destroy the * metadata associated with it. */ if (vd->vdev_islog && vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) vdev_remove_empty_log(vd, txg); (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); dmu_tx_commit(tx); } uint64_t vdev_psize_to_asize(vdev_t *vd, uint64_t psize) { return (vd->vdev_ops->vdev_op_asize(vd, psize)); } /* * Mark the given vdev faulted. A faulted vdev behaves as if the device could * not be opened, and no I/O is attempted. */ int vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) { vdev_t *vd, *tvd; spa_vdev_state_enter(spa, SCL_NONE); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); tvd = vd->vdev_top; /* * If user did a 'zpool offline -f' then make the fault persist across * reboots. */ if (aux == VDEV_AUX_EXTERNAL_PERSIST) { /* * There are two kinds of forced faults: temporary and * persistent. Temporary faults go away at pool import, while * persistent faults stay set. Both types of faults can be * cleared with a zpool clear. * * We tell if a vdev is persistently faulted by looking at the * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at * import then it's a persistent fault. Otherwise, it's * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external" * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This * tells vdev_config_generate() (which gets run later) to set * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist. */ vd->vdev_stat.vs_aux = VDEV_AUX_EXTERNAL; vd->vdev_tmpoffline = B_FALSE; aux = VDEV_AUX_EXTERNAL; } else { vd->vdev_tmpoffline = B_TRUE; } /* * We don't directly use the aux state here, but if we do a * vdev_reopen(), we need this value to be present to remember why we * were faulted. */ vd->vdev_label_aux = aux; /* * Faulted state takes precedence over degraded. */ vd->vdev_delayed_close = B_FALSE; vd->vdev_faulted = 1ULL; vd->vdev_degraded = 0ULL; vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); /* * If this device has the only valid copy of the data, then * back off and simply mark the vdev as degraded instead. */ if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { vd->vdev_degraded = 1ULL; vd->vdev_faulted = 0ULL; /* * If we reopen the device and it's not dead, only then do we * mark it degraded. */ vdev_reopen(tvd); if (vdev_readable(vd)) vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); } return (spa_vdev_state_exit(spa, vd, 0)); } /* * Mark the given vdev degraded. A degraded vdev is purely an indication to the * user that something is wrong. The vdev continues to operate as normal as far * as I/O is concerned. */ int vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) { vdev_t *vd; spa_vdev_state_enter(spa, SCL_NONE); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); /* * If the vdev is already faulted, then don't do anything. */ if (vd->vdev_faulted || vd->vdev_degraded) return (spa_vdev_state_exit(spa, NULL, 0)); vd->vdev_degraded = 1ULL; if (!vdev_is_dead(vd)) vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); return (spa_vdev_state_exit(spa, vd, 0)); } /* * Online the given vdev. * * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached * spare device should be detached when the device finishes resilvering. * Second, the online should be treated like a 'test' online case, so no FMA * events are generated if the device fails to open. */ int vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) { vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; boolean_t wasoffline; vdev_state_t oldstate; spa_vdev_state_enter(spa, SCL_NONE); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); oldstate = vd->vdev_state; tvd = vd->vdev_top; vd->vdev_offline = B_FALSE; vd->vdev_tmpoffline = B_FALSE; vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); /* XXX - L2ARC 1.0 does not support expansion */ if (!vd->vdev_aux) { for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand); vd->vdev_expansion_time = gethrestime_sec(); } vdev_reopen(tvd); vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; if (!vd->vdev_aux) { for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) pvd->vdev_expanding = B_FALSE; } if (newstate) *newstate = vd->vdev_state; if ((flags & ZFS_ONLINE_UNSPARE) && !vdev_is_dead(vd) && vd->vdev_parent && vd->vdev_parent->vdev_ops == &vdev_spare_ops && vd->vdev_parent->vdev_child[0] == vd) vd->vdev_unspare = B_TRUE; if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { /* XXX - L2ARC 1.0 does not support expansion */ if (vd->vdev_aux) return (spa_vdev_state_exit(spa, vd, ENOTSUP)); spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); } /* Restart initializing if necessary */ mutex_enter(&vd->vdev_initialize_lock); if (vdev_writeable(vd) && vd->vdev_initialize_thread == NULL && vd->vdev_initialize_state == VDEV_INITIALIZE_ACTIVE) { (void) vdev_initialize(vd); } mutex_exit(&vd->vdev_initialize_lock); /* * Restart trimming if necessary. We do not restart trimming for cache * devices here. This is triggered by l2arc_rebuild_vdev() * asynchronously for the whole device or in l2arc_evict() as it evicts * space for upcoming writes. */ mutex_enter(&vd->vdev_trim_lock); if (vdev_writeable(vd) && !vd->vdev_isl2cache && vd->vdev_trim_thread == NULL && vd->vdev_trim_state == VDEV_TRIM_ACTIVE) { (void) vdev_trim(vd, vd->vdev_trim_rate, vd->vdev_trim_partial, vd->vdev_trim_secure); } mutex_exit(&vd->vdev_trim_lock); if (wasoffline || (oldstate < VDEV_STATE_DEGRADED && vd->vdev_state >= VDEV_STATE_DEGRADED)) spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); return (spa_vdev_state_exit(spa, vd, 0)); } static int vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) { vdev_t *vd, *tvd; int error = 0; uint64_t generation; metaslab_group_t *mg; top: spa_vdev_state_enter(spa, SCL_ALLOC); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENODEV))); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(ENOTSUP))); if (vd->vdev_ops == &vdev_draid_spare_ops) return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); tvd = vd->vdev_top; mg = tvd->vdev_mg; generation = spa->spa_config_generation + 1; /* * If the device isn't already offline, try to offline it. */ if (!vd->vdev_offline) { /* * If this device has the only valid copy of some data, * don't allow it to be offlined. Log devices are always * expendable. */ if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) return (spa_vdev_state_exit(spa, NULL, SET_ERROR(EBUSY))); /* * If the top-level is a slog and it has had allocations * then proceed. We check that the vdev's metaslab group * is not NULL since it's possible that we may have just * added this vdev but not yet initialized its metaslabs. */ if (tvd->vdev_islog && mg != NULL) { /* * Prevent any future allocations. */ ASSERT3P(tvd->vdev_log_mg, ==, NULL); metaslab_group_passivate(mg); (void) spa_vdev_state_exit(spa, vd, 0); error = spa_reset_logs(spa); /* * If the log device was successfully reset but has * checkpointed data, do not offline it. */ if (error == 0 && tvd->vdev_checkpoint_sm != NULL) { ASSERT3U(space_map_allocated( tvd->vdev_checkpoint_sm), !=, 0); error = ZFS_ERR_CHECKPOINT_EXISTS; } spa_vdev_state_enter(spa, SCL_ALLOC); /* * Check to see if the config has changed. */ if (error || generation != spa->spa_config_generation) { metaslab_group_activate(mg); if (error) return (spa_vdev_state_exit(spa, vd, error)); (void) spa_vdev_state_exit(spa, vd, 0); goto top; } ASSERT0(tvd->vdev_stat.vs_alloc); } /* * Offline this device and reopen its top-level vdev. * If the top-level vdev is a log device then just offline * it. Otherwise, if this action results in the top-level * vdev becoming unusable, undo it and fail the request. */ vd->vdev_offline = B_TRUE; vdev_reopen(tvd); if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_is_dead(tvd)) { vd->vdev_offline = B_FALSE; vdev_reopen(tvd); return (spa_vdev_state_exit(spa, NULL, SET_ERROR(EBUSY))); } /* * Add the device back into the metaslab rotor so that * once we online the device it's open for business. */ if (tvd->vdev_islog && mg != NULL) metaslab_group_activate(mg); } vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); return (spa_vdev_state_exit(spa, vd, 0)); } int vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) { int error; mutex_enter(&spa->spa_vdev_top_lock); error = vdev_offline_locked(spa, guid, flags); mutex_exit(&spa->spa_vdev_top_lock); return (error); } /* * Clear the error counts associated with this vdev. Unlike vdev_online() and * vdev_offline(), we assume the spa config is locked. We also clear all * children. If 'vd' is NULL, then the user wants to clear all vdevs. */ void vdev_clear(spa_t *spa, vdev_t *vd) { vdev_t *rvd = spa->spa_root_vdev; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); if (vd == NULL) vd = rvd; vd->vdev_stat.vs_read_errors = 0; vd->vdev_stat.vs_write_errors = 0; vd->vdev_stat.vs_checksum_errors = 0; vd->vdev_stat.vs_slow_ios = 0; for (int c = 0; c < vd->vdev_children; c++) vdev_clear(spa, vd->vdev_child[c]); /* * It makes no sense to "clear" an indirect vdev. */ if (!vdev_is_concrete(vd)) return; /* * If we're in the FAULTED state or have experienced failed I/O, then * clear the persistent state and attempt to reopen the device. We * also mark the vdev config dirty, so that the new faulted state is * written out to disk. */ if (vd->vdev_faulted || vd->vdev_degraded || !vdev_readable(vd) || !vdev_writeable(vd)) { /* * When reopening in response to a clear event, it may be due to * a fmadm repair request. In this case, if the device is * still broken, we want to still post the ereport again. */ vd->vdev_forcefault = B_TRUE; vd->vdev_faulted = vd->vdev_degraded = 0ULL; vd->vdev_cant_read = B_FALSE; vd->vdev_cant_write = B_FALSE; vd->vdev_stat.vs_aux = 0; vdev_reopen(vd == rvd ? rvd : vd->vdev_top); vd->vdev_forcefault = B_FALSE; if (vd != rvd && vdev_writeable(vd->vdev_top)) vdev_state_dirty(vd->vdev_top); /* If a resilver isn't required, check if vdevs can be culled */ if (vd->vdev_aux == NULL && !vdev_is_dead(vd) && !dsl_scan_resilvering(spa->spa_dsl_pool) && !dsl_scan_resilver_scheduled(spa->spa_dsl_pool)) spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); } /* * When clearing a FMA-diagnosed fault, we always want to * unspare the device, as we assume that the original spare was * done in response to the FMA fault. */ if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && vd->vdev_parent->vdev_ops == &vdev_spare_ops && vd->vdev_parent->vdev_child[0] == vd) vd->vdev_unspare = B_TRUE; } boolean_t vdev_is_dead(vdev_t *vd) { /* * Holes and missing devices are always considered "dead". * This simplifies the code since we don't have to check for * these types of devices in the various code paths. * Instead we rely on the fact that we skip over dead devices * before issuing I/O to them. */ return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ops == &vdev_hole_ops || vd->vdev_ops == &vdev_missing_ops); } boolean_t vdev_readable(vdev_t *vd) { return (!vdev_is_dead(vd) && !vd->vdev_cant_read); } boolean_t vdev_writeable(vdev_t *vd) { return (!vdev_is_dead(vd) && !vd->vdev_cant_write && vdev_is_concrete(vd)); } boolean_t vdev_allocatable(vdev_t *vd) { uint64_t state = vd->vdev_state; /* * We currently allow allocations from vdevs which may be in the * process of reopening (i.e. VDEV_STATE_CLOSED). If the device * fails to reopen then we'll catch it later when we're holding * the proper locks. Note that we have to get the vdev state * in a local variable because although it changes atomically, * we're asking two separate questions about it. */ return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && !vd->vdev_cant_write && vdev_is_concrete(vd) && vd->vdev_mg->mg_initialized); } boolean_t vdev_accessible(vdev_t *vd, zio_t *zio) { ASSERT(zio->io_vd == vd); if (vdev_is_dead(vd) || vd->vdev_remove_wanted) return (B_FALSE); if (zio->io_type == ZIO_TYPE_READ) return (!vd->vdev_cant_read); if (zio->io_type == ZIO_TYPE_WRITE) return (!vd->vdev_cant_write); return (B_TRUE); } static void vdev_get_child_stat(vdev_t *cvd, vdev_stat_t *vs, vdev_stat_t *cvs) { /* * Exclude the dRAID spare when aggregating to avoid double counting * the ops and bytes. These IOs are counted by the physical leaves. */ if (cvd->vdev_ops == &vdev_draid_spare_ops) return; for (int t = 0; t < VS_ZIO_TYPES; t++) { vs->vs_ops[t] += cvs->vs_ops[t]; vs->vs_bytes[t] += cvs->vs_bytes[t]; } cvs->vs_scan_removing = cvd->vdev_removing; } /* * Get extended stats */ static void vdev_get_child_stat_ex(vdev_t *cvd, vdev_stat_ex_t *vsx, vdev_stat_ex_t *cvsx) { int t, b; for (t = 0; t < ZIO_TYPES; t++) { for (b = 0; b < ARRAY_SIZE(vsx->vsx_disk_histo[0]); b++) vsx->vsx_disk_histo[t][b] += cvsx->vsx_disk_histo[t][b]; for (b = 0; b < ARRAY_SIZE(vsx->vsx_total_histo[0]); b++) { vsx->vsx_total_histo[t][b] += cvsx->vsx_total_histo[t][b]; } } for (t = 0; t < ZIO_PRIORITY_NUM_QUEUEABLE; t++) { for (b = 0; b < ARRAY_SIZE(vsx->vsx_queue_histo[0]); b++) { vsx->vsx_queue_histo[t][b] += cvsx->vsx_queue_histo[t][b]; } vsx->vsx_active_queue[t] += cvsx->vsx_active_queue[t]; vsx->vsx_pend_queue[t] += cvsx->vsx_pend_queue[t]; for (b = 0; b < ARRAY_SIZE(vsx->vsx_ind_histo[0]); b++) vsx->vsx_ind_histo[t][b] += cvsx->vsx_ind_histo[t][b]; for (b = 0; b < ARRAY_SIZE(vsx->vsx_agg_histo[0]); b++) vsx->vsx_agg_histo[t][b] += cvsx->vsx_agg_histo[t][b]; } } boolean_t vdev_is_spacemap_addressable(vdev_t *vd) { if (spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_SPACEMAP_V2)) return (B_TRUE); /* * If double-word space map entries are not enabled we assume * 47 bits of the space map entry are dedicated to the entry's * offset (see SM_OFFSET_BITS in space_map.h). We then use that * to calculate the maximum address that can be described by a * space map entry for the given device. */ uint64_t shift = vd->vdev_ashift + SM_OFFSET_BITS; if (shift >= 63) /* detect potential overflow */ return (B_TRUE); return (vd->vdev_asize < (1ULL << shift)); } /* * Get statistics for the given vdev. */ static void vdev_get_stats_ex_impl(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) { int t; /* * If we're getting stats on the root vdev, aggregate the I/O counts * over all top-level vdevs (i.e. the direct children of the root). */ if (!vd->vdev_ops->vdev_op_leaf) { if (vs) { memset(vs->vs_ops, 0, sizeof (vs->vs_ops)); memset(vs->vs_bytes, 0, sizeof (vs->vs_bytes)); } if (vsx) memset(vsx, 0, sizeof (*vsx)); for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; vdev_stat_t *cvs = &cvd->vdev_stat; vdev_stat_ex_t *cvsx = &cvd->vdev_stat_ex; vdev_get_stats_ex_impl(cvd, cvs, cvsx); if (vs) vdev_get_child_stat(cvd, vs, cvs); if (vsx) vdev_get_child_stat_ex(cvd, vsx, cvsx); } } else { /* * We're a leaf. Just copy our ZIO active queue stats in. The * other leaf stats are updated in vdev_stat_update(). */ if (!vsx) return; memcpy(vsx, &vd->vdev_stat_ex, sizeof (vd->vdev_stat_ex)); for (t = 0; t < ARRAY_SIZE(vd->vdev_queue.vq_class); t++) { vsx->vsx_active_queue[t] = vd->vdev_queue.vq_class[t].vqc_active; vsx->vsx_pend_queue[t] = avl_numnodes( &vd->vdev_queue.vq_class[t].vqc_queued_tree); } } } void vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx) { vdev_t *tvd = vd->vdev_top; mutex_enter(&vd->vdev_stat_lock); if (vs) { bcopy(&vd->vdev_stat, vs, sizeof (*vs)); vs->vs_timestamp = gethrtime() - vs->vs_timestamp; vs->vs_state = vd->vdev_state; vs->vs_rsize = vdev_get_min_asize(vd); if (vd->vdev_ops->vdev_op_leaf) { vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; /* * Report initializing progress. Since we don't * have the initializing locks held, this is only * an estimate (although a fairly accurate one). */ vs->vs_initialize_bytes_done = vd->vdev_initialize_bytes_done; vs->vs_initialize_bytes_est = vd->vdev_initialize_bytes_est; vs->vs_initialize_state = vd->vdev_initialize_state; vs->vs_initialize_action_time = vd->vdev_initialize_action_time; /* * Report manual TRIM progress. Since we don't have * the manual TRIM locks held, this is only an * estimate (although fairly accurate one). */ vs->vs_trim_notsup = !vd->vdev_has_trim; vs->vs_trim_bytes_done = vd->vdev_trim_bytes_done; vs->vs_trim_bytes_est = vd->vdev_trim_bytes_est; vs->vs_trim_state = vd->vdev_trim_state; vs->vs_trim_action_time = vd->vdev_trim_action_time; /* Set when there is a deferred resilver. */ vs->vs_resilver_deferred = vd->vdev_resilver_deferred; } /* * Report expandable space on top-level, non-auxiliary devices * only. The expandable space is reported in terms of metaslab * sized units since that determines how much space the pool * can expand. */ if (vd->vdev_aux == NULL && tvd != NULL) { vs->vs_esize = P2ALIGN( vd->vdev_max_asize - vd->vdev_asize, 1ULL << tvd->vdev_ms_shift); } vs->vs_configured_ashift = vd->vdev_top != NULL ? vd->vdev_top->vdev_ashift : vd->vdev_ashift; vs->vs_logical_ashift = vd->vdev_logical_ashift; vs->vs_physical_ashift = vd->vdev_physical_ashift; /* * Report fragmentation and rebuild progress for top-level, * non-auxiliary, concrete devices. */ if (vd->vdev_aux == NULL && vd == vd->vdev_top && vdev_is_concrete(vd)) { /* * The vdev fragmentation rating doesn't take into * account the embedded slog metaslab (vdev_log_mg). * Since it's only one metaslab, it would have a tiny * impact on the overall fragmentation. */ vs->vs_fragmentation = (vd->vdev_mg != NULL) ? vd->vdev_mg->mg_fragmentation : 0; } } vdev_get_stats_ex_impl(vd, vs, vsx); mutex_exit(&vd->vdev_stat_lock); } void vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) { return (vdev_get_stats_ex(vd, vs, NULL)); } void vdev_clear_stats(vdev_t *vd) { mutex_enter(&vd->vdev_stat_lock); vd->vdev_stat.vs_space = 0; vd->vdev_stat.vs_dspace = 0; vd->vdev_stat.vs_alloc = 0; mutex_exit(&vd->vdev_stat_lock); } void vdev_scan_stat_init(vdev_t *vd) { vdev_stat_t *vs = &vd->vdev_stat; for (int c = 0; c < vd->vdev_children; c++) vdev_scan_stat_init(vd->vdev_child[c]); mutex_enter(&vd->vdev_stat_lock); vs->vs_scan_processed = 0; mutex_exit(&vd->vdev_stat_lock); } void vdev_stat_update(zio_t *zio, uint64_t psize) { spa_t *spa = zio->io_spa; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; vdev_t *pvd; uint64_t txg = zio->io_txg; vdev_stat_t *vs = &vd->vdev_stat; vdev_stat_ex_t *vsx = &vd->vdev_stat_ex; zio_type_t type = zio->io_type; int flags = zio->io_flags; /* * If this i/o is a gang leader, it didn't do any actual work. */ if (zio->io_gang_tree) return; if (zio->io_error == 0) { /* * If this is a root i/o, don't count it -- we've already * counted the top-level vdevs, and vdev_get_stats() will * aggregate them when asked. This reduces contention on * the root vdev_stat_lock and implicitly handles blocks * that compress away to holes, for which there is no i/o. * (Holes never create vdev children, so all the counters * remain zero, which is what we want.) * * Note: this only applies to successful i/o (io_error == 0) * because unlike i/o counts, errors are not additive. * When reading a ditto block, for example, failure of * one top-level vdev does not imply a root-level error. */ if (vd == rvd) return; ASSERT(vd == zio->io_vd); if (flags & ZIO_FLAG_IO_BYPASS) return; mutex_enter(&vd->vdev_stat_lock); if (flags & ZIO_FLAG_IO_REPAIR) { /* * Repair is the result of a resilver issued by the * scan thread (spa_sync). */ if (flags & ZIO_FLAG_SCAN_THREAD) { dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; dsl_scan_phys_t *scn_phys = &scn->scn_phys; uint64_t *processed = &scn_phys->scn_processed; if (vd->vdev_ops->vdev_op_leaf) atomic_add_64(processed, psize); vs->vs_scan_processed += psize; } /* * Repair is the result of a rebuild issued by the * rebuild thread (vdev_rebuild_thread). To avoid * double counting repaired bytes the virtual dRAID * spare vdev is excluded from the processed bytes. */ if (zio->io_priority == ZIO_PRIORITY_REBUILD) { vdev_t *tvd = vd->vdev_top; vdev_rebuild_t *vr = &tvd->vdev_rebuild_config; vdev_rebuild_phys_t *vrp = &vr->vr_rebuild_phys; uint64_t *rebuilt = &vrp->vrp_bytes_rebuilt; if (vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_draid_spare_ops) { atomic_add_64(rebuilt, psize); } vs->vs_rebuild_processed += psize; } if (flags & ZIO_FLAG_SELF_HEAL) vs->vs_self_healed += psize; } /* * The bytes/ops/histograms are recorded at the leaf level and * aggregated into the higher level vdevs in vdev_get_stats(). */ if (vd->vdev_ops->vdev_op_leaf && (zio->io_priority < ZIO_PRIORITY_NUM_QUEUEABLE)) { zio_type_t vs_type = type; zio_priority_t priority = zio->io_priority; /* * TRIM ops and bytes are reported to user space as * ZIO_TYPE_IOCTL. This is done to preserve the * vdev_stat_t structure layout for user space. */ if (type == ZIO_TYPE_TRIM) vs_type = ZIO_TYPE_IOCTL; /* * Solely for the purposes of 'zpool iostat -lqrw' * reporting use the priority to catagorize the IO. * Only the following are reported to user space: * * ZIO_PRIORITY_SYNC_READ, * ZIO_PRIORITY_SYNC_WRITE, * ZIO_PRIORITY_ASYNC_READ, * ZIO_PRIORITY_ASYNC_WRITE, * ZIO_PRIORITY_SCRUB, * ZIO_PRIORITY_TRIM. */ if (priority == ZIO_PRIORITY_REBUILD) { priority = ((type == ZIO_TYPE_WRITE) ? ZIO_PRIORITY_ASYNC_WRITE : ZIO_PRIORITY_SCRUB); } else if (priority == ZIO_PRIORITY_INITIALIZING) { ASSERT3U(type, ==, ZIO_TYPE_WRITE); priority = ZIO_PRIORITY_ASYNC_WRITE; } else if (priority == ZIO_PRIORITY_REMOVAL) { priority = ((type == ZIO_TYPE_WRITE) ? ZIO_PRIORITY_ASYNC_WRITE : ZIO_PRIORITY_ASYNC_READ); } vs->vs_ops[vs_type]++; vs->vs_bytes[vs_type] += psize; if (flags & ZIO_FLAG_DELEGATED) { vsx->vsx_agg_histo[priority] [RQ_HISTO(zio->io_size)]++; } else { vsx->vsx_ind_histo[priority] [RQ_HISTO(zio->io_size)]++; } if (zio->io_delta && zio->io_delay) { vsx->vsx_queue_histo[priority] [L_HISTO(zio->io_delta - zio->io_delay)]++; vsx->vsx_disk_histo[type] [L_HISTO(zio->io_delay)]++; vsx->vsx_total_histo[type] [L_HISTO(zio->io_delta)]++; } } mutex_exit(&vd->vdev_stat_lock); return; } if (flags & ZIO_FLAG_SPECULATIVE) return; /* * If this is an I/O error that is going to be retried, then ignore the * error. Otherwise, the user may interpret B_FAILFAST I/O errors as * hard errors, when in reality they can happen for any number of * innocuous reasons (bus resets, MPxIO link failure, etc). */ if (zio->io_error == EIO && !(zio->io_flags & ZIO_FLAG_IO_RETRY)) return; /* * Intent logs writes won't propagate their error to the root * I/O so don't mark these types of failures as pool-level * errors. */ if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) return; if (type == ZIO_TYPE_WRITE && txg != 0 && (!(flags & ZIO_FLAG_IO_REPAIR) || (flags & ZIO_FLAG_SCAN_THREAD) || spa->spa_claiming)) { /* * This is either a normal write (not a repair), or it's * a repair induced by the scrub thread, or it's a repair * made by zil_claim() during spa_load() in the first txg. * In the normal case, we commit the DTL change in the same * txg as the block was born. In the scrub-induced repair * case, we know that scrubs run in first-pass syncing context, * so we commit the DTL change in spa_syncing_txg(spa). * In the zil_claim() case, we commit in spa_first_txg(spa). * * We currently do not make DTL entries for failed spontaneous * self-healing writes triggered by normal (non-scrubbing) * reads, because we have no transactional context in which to * do so -- and it's not clear that it'd be desirable anyway. */ if (vd->vdev_ops->vdev_op_leaf) { uint64_t commit_txg = txg; if (flags & ZIO_FLAG_SCAN_THREAD) { ASSERT(flags & ZIO_FLAG_IO_REPAIR); ASSERT(spa_sync_pass(spa) == 1); vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); commit_txg = spa_syncing_txg(spa); } else if (spa->spa_claiming) { ASSERT(flags & ZIO_FLAG_IO_REPAIR); commit_txg = spa_first_txg(spa); } ASSERT(commit_txg >= spa_syncing_txg(spa)); if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) return; for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); } if (vd != rvd) vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); } } int64_t vdev_deflated_space(vdev_t *vd, int64_t space) { ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0); ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio); } /* * Update the in-core space usage stats for this vdev, its metaslab class, * and the root vdev. */ void vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, int64_t space_delta) { int64_t dspace_delta; spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; ASSERT(vd == vd->vdev_top); /* * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion * factor. We must calculate this here and not at the root vdev * because the root vdev's psize-to-asize is simply the max of its * children's, thus not accurate enough for us. */ dspace_delta = vdev_deflated_space(vd, space_delta); mutex_enter(&vd->vdev_stat_lock); /* ensure we won't underflow */ if (alloc_delta < 0) { ASSERT3U(vd->vdev_stat.vs_alloc, >=, -alloc_delta); } vd->vdev_stat.vs_alloc += alloc_delta; vd->vdev_stat.vs_space += space_delta; vd->vdev_stat.vs_dspace += dspace_delta; mutex_exit(&vd->vdev_stat_lock); /* every class but log contributes to root space stats */ if (vd->vdev_mg != NULL && !vd->vdev_islog) { ASSERT(!vd->vdev_isl2cache); mutex_enter(&rvd->vdev_stat_lock); rvd->vdev_stat.vs_alloc += alloc_delta; rvd->vdev_stat.vs_space += space_delta; rvd->vdev_stat.vs_dspace += dspace_delta; mutex_exit(&rvd->vdev_stat_lock); } /* Note: metaslab_class_space_update moved to metaslab_space_update */ } /* * Mark a top-level vdev's config as dirty, placing it on the dirty list * so that it will be written out next time the vdev configuration is synced. * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. */ void vdev_config_dirty(vdev_t *vd) { spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; int c; ASSERT(spa_writeable(spa)); /* * If this is an aux vdev (as with l2cache and spare devices), then we * update the vdev config manually and set the sync flag. */ if (vd->vdev_aux != NULL) { spa_aux_vdev_t *sav = vd->vdev_aux; nvlist_t **aux; uint_t naux; for (c = 0; c < sav->sav_count; c++) { if (sav->sav_vdevs[c] == vd) break; } if (c == sav->sav_count) { /* * We're being removed. There's nothing more to do. */ ASSERT(sav->sav_sync == B_TRUE); return; } sav->sav_sync = B_TRUE; if (nvlist_lookup_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); } ASSERT(c < naux); /* * Setting the nvlist in the middle if the array is a little * sketchy, but it will work. */ nvlist_free(aux[c]); aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); return; } /* * The dirty list is protected by the SCL_CONFIG lock. The caller * must either hold SCL_CONFIG as writer, or must be the sync thread * (which holds SCL_CONFIG as reader). There's only one sync thread, * so this is sufficient to ensure mutual exclusion. */ ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_CONFIG, RW_READER))); if (vd == rvd) { for (c = 0; c < rvd->vdev_children; c++) vdev_config_dirty(rvd->vdev_child[c]); } else { ASSERT(vd == vd->vdev_top); if (!list_link_active(&vd->vdev_config_dirty_node) && vdev_is_concrete(vd)) { list_insert_head(&spa->spa_config_dirty_list, vd); } } } void vdev_config_clean(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_CONFIG, RW_READER))); ASSERT(list_link_active(&vd->vdev_config_dirty_node)); list_remove(&spa->spa_config_dirty_list, vd); } /* * Mark a top-level vdev's state as dirty, so that the next pass of * spa_sync() can convert this into vdev_config_dirty(). We distinguish * the state changes from larger config changes because they require * much less locking, and are often needed for administrative actions. */ void vdev_state_dirty(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_writeable(spa)); ASSERT(vd == vd->vdev_top); /* * The state list is protected by the SCL_STATE lock. The caller * must either hold SCL_STATE as writer, or must be the sync thread * (which holds SCL_STATE as reader). There's only one sync thread, * so this is sufficient to ensure mutual exclusion. */ ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_STATE, RW_READER))); if (!list_link_active(&vd->vdev_state_dirty_node) && vdev_is_concrete(vd)) list_insert_head(&spa->spa_state_dirty_list, vd); } void vdev_state_clean(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_STATE, RW_READER))); ASSERT(list_link_active(&vd->vdev_state_dirty_node)); list_remove(&spa->spa_state_dirty_list, vd); } /* * Propagate vdev state up from children to parent. */ void vdev_propagate_state(vdev_t *vd) { spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; int degraded = 0, faulted = 0; int corrupted = 0; vdev_t *child; if (vd->vdev_children > 0) { for (int c = 0; c < vd->vdev_children; c++) { child = vd->vdev_child[c]; /* * Don't factor holes or indirect vdevs into the * decision. */ if (!vdev_is_concrete(child)) continue; if (!vdev_readable(child) || (!vdev_writeable(child) && spa_writeable(spa))) { /* * Root special: if there is a top-level log * device, treat the root vdev as if it were * degraded. */ if (child->vdev_islog && vd == rvd) degraded++; else faulted++; } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { degraded++; } if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) corrupted++; } vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); /* * Root special: if there is a top-level vdev that cannot be * opened due to corrupted metadata, then propagate the root * vdev's aux state as 'corrupt' rather than 'insufficient * replicas'. */ if (corrupted && vd == rvd && rvd->vdev_state == VDEV_STATE_CANT_OPEN) vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); } if (vd->vdev_parent) vdev_propagate_state(vd->vdev_parent); } /* * Set a vdev's state. If this is during an open, we don't update the parent * state, because we're in the process of opening children depth-first. * Otherwise, we propagate the change to the parent. * * If this routine places a device in a faulted state, an appropriate ereport is * generated. */ void vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) { uint64_t save_state; spa_t *spa = vd->vdev_spa; if (state == vd->vdev_state) { /* * Since vdev_offline() code path is already in an offline * state we can miss a statechange event to OFFLINE. Check * the previous state to catch this condition. */ if (vd->vdev_ops->vdev_op_leaf && (state == VDEV_STATE_OFFLINE) && (vd->vdev_prevstate >= VDEV_STATE_FAULTED)) { /* post an offline state change */ zfs_post_state_change(spa, vd, vd->vdev_prevstate); } vd->vdev_stat.vs_aux = aux; return; } save_state = vd->vdev_state; vd->vdev_state = state; vd->vdev_stat.vs_aux = aux; /* * If we are setting the vdev state to anything but an open state, then * always close the underlying device unless the device has requested * a delayed close (i.e. we're about to remove or fault the device). * Otherwise, we keep accessible but invalid devices open forever. * We don't call vdev_close() itself, because that implies some extra * checks (offline, etc) that we don't want here. This is limited to * leaf devices, because otherwise closing the device will affect other * children. */ if (!vd->vdev_delayed_close && vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf) vd->vdev_ops->vdev_op_close(vd); if (vd->vdev_removed && state == VDEV_STATE_CANT_OPEN && (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { /* * If the previous state is set to VDEV_STATE_REMOVED, then this * device was previously marked removed and someone attempted to * reopen it. If this failed due to a nonexistent device, then * keep the device in the REMOVED state. We also let this be if * it is one of our special test online cases, which is only * attempting to online the device and shouldn't generate an FMA * fault. */ vd->vdev_state = VDEV_STATE_REMOVED; vd->vdev_stat.vs_aux = VDEV_AUX_NONE; } else if (state == VDEV_STATE_REMOVED) { vd->vdev_removed = B_TRUE; } else if (state == VDEV_STATE_CANT_OPEN) { /* * If we fail to open a vdev during an import or recovery, we * mark it as "not available", which signifies that it was * never there to begin with. Failure to open such a device * is not considered an error. */ if ((spa_load_state(spa) == SPA_LOAD_IMPORT || spa_load_state(spa) == SPA_LOAD_RECOVER) && vd->vdev_ops->vdev_op_leaf) vd->vdev_not_present = 1; /* * Post the appropriate ereport. If the 'prevstate' field is * set to something other than VDEV_STATE_UNKNOWN, it indicates * that this is part of a vdev_reopen(). In this case, we don't * want to post the ereport if the device was already in the * CANT_OPEN state beforehand. * * If the 'checkremove' flag is set, then this is an attempt to * online the device in response to an insertion event. If we * hit this case, then we have detected an insertion event for a * faulted or offline device that wasn't in the removed state. * In this scenario, we don't post an ereport because we are * about to replace the device, or attempt an online with * vdev_forcefault, which will generate the fault for us. */ if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && !vd->vdev_not_present && !vd->vdev_checkremove && vd != spa->spa_root_vdev) { const char *class; switch (aux) { case VDEV_AUX_OPEN_FAILED: class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; break; case VDEV_AUX_CORRUPT_DATA: class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; break; case VDEV_AUX_NO_REPLICAS: class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; break; case VDEV_AUX_BAD_GUID_SUM: class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; break; case VDEV_AUX_TOO_SMALL: class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; break; case VDEV_AUX_BAD_LABEL: class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; break; case VDEV_AUX_BAD_ASHIFT: class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT; break; default: class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; } (void) zfs_ereport_post(class, spa, vd, NULL, NULL, save_state); } /* Erase any notion of persistent removed state */ vd->vdev_removed = B_FALSE; } else { vd->vdev_removed = B_FALSE; } /* * Notify ZED of any significant state-change on a leaf vdev. * */ if (vd->vdev_ops->vdev_op_leaf) { /* preserve original state from a vdev_reopen() */ if ((vd->vdev_prevstate != VDEV_STATE_UNKNOWN) && (vd->vdev_prevstate != vd->vdev_state) && (save_state <= VDEV_STATE_CLOSED)) save_state = vd->vdev_prevstate; /* filter out state change due to initial vdev_open */ if (save_state > VDEV_STATE_CLOSED) zfs_post_state_change(spa, vd, save_state); } if (!isopen && vd->vdev_parent) vdev_propagate_state(vd->vdev_parent); } boolean_t vdev_children_are_offline(vdev_t *vd) { ASSERT(!vd->vdev_ops->vdev_op_leaf); for (uint64_t i = 0; i < vd->vdev_children; i++) { if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) return (B_FALSE); } return (B_TRUE); } /* * Check the vdev configuration to ensure that it's capable of supporting * a root pool. We do not support partial configuration. */ boolean_t vdev_is_bootable(vdev_t *vd) { if (!vd->vdev_ops->vdev_op_leaf) { const char *vdev_type = vd->vdev_ops->vdev_op_type; if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { return (B_FALSE); } } for (int c = 0; c < vd->vdev_children; c++) { if (!vdev_is_bootable(vd->vdev_child[c])) return (B_FALSE); } return (B_TRUE); } boolean_t vdev_is_concrete(vdev_t *vd) { vdev_ops_t *ops = vd->vdev_ops; if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || ops == &vdev_missing_ops || ops == &vdev_root_ops) { return (B_FALSE); } else { return (B_TRUE); } } /* * Determine if a log device has valid content. If the vdev was * removed or faulted in the MOS config then we know that * the content on the log device has already been written to the pool. */ boolean_t vdev_log_state_valid(vdev_t *vd) { if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && !vd->vdev_removed) return (B_TRUE); for (int c = 0; c < vd->vdev_children; c++) if (vdev_log_state_valid(vd->vdev_child[c])) return (B_TRUE); return (B_FALSE); } /* * Expand a vdev if possible. */ void vdev_expand(vdev_t *vd, uint64_t txg) { ASSERT(vd->vdev_top == vd); ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); ASSERT(vdev_is_concrete(vd)); vdev_set_deflate_ratio(vd); if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && vdev_is_concrete(vd)) { vdev_metaslab_group_create(vd); VERIFY(vdev_metaslab_init(vd, txg) == 0); vdev_config_dirty(vd); } } /* * Split a vdev. */ void vdev_split(vdev_t *vd) { vdev_t *cvd, *pvd = vd->vdev_parent; vdev_remove_child(pvd, vd); vdev_compact_children(pvd); cvd = pvd->vdev_child[0]; if (pvd->vdev_children == 1) { vdev_remove_parent(cvd); cvd->vdev_splitting = B_TRUE; } vdev_propagate_state(cvd); } void vdev_deadman(vdev_t *vd, char *tag) { for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; vdev_deadman(cvd, tag); } if (vd->vdev_ops->vdev_op_leaf) { vdev_queue_t *vq = &vd->vdev_queue; mutex_enter(&vq->vq_lock); if (avl_numnodes(&vq->vq_active_tree) > 0) { spa_t *spa = vd->vdev_spa; zio_t *fio; uint64_t delta; zfs_dbgmsg("slow vdev: %s has %d active IOs", vd->vdev_path, avl_numnodes(&vq->vq_active_tree)); /* * Look at the head of all the pending queues, * if any I/O has been outstanding for longer than * the spa_deadman_synctime invoke the deadman logic. */ fio = avl_first(&vq->vq_active_tree); delta = gethrtime() - fio->io_timestamp; if (delta > spa_deadman_synctime(spa)) zio_deadman(fio, tag); } mutex_exit(&vq->vq_lock); } } void vdev_defer_resilver(vdev_t *vd) { ASSERT(vd->vdev_ops->vdev_op_leaf); vd->vdev_resilver_deferred = B_TRUE; vd->vdev_spa->spa_resilver_deferred = B_TRUE; } /* * Clears the resilver deferred flag on all leaf devs under vd. Returns * B_TRUE if we have devices that need to be resilvered and are available to * accept resilver I/Os. */ boolean_t vdev_clear_resilver_deferred(vdev_t *vd, dmu_tx_t *tx) { boolean_t resilver_needed = B_FALSE; spa_t *spa = vd->vdev_spa; for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; resilver_needed |= vdev_clear_resilver_deferred(cvd, tx); } if (vd == spa->spa_root_vdev && spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) { spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx); vdev_config_dirty(vd); spa->spa_resilver_deferred = B_FALSE; return (resilver_needed); } if (!vdev_is_concrete(vd) || vd->vdev_aux || !vd->vdev_ops->vdev_op_leaf) return (resilver_needed); vd->vdev_resilver_deferred = B_FALSE; return (!vdev_is_dead(vd) && !vd->vdev_offline && vdev_resilver_needed(vd, NULL, NULL)); } boolean_t vdev_xlate_is_empty(range_seg64_t *rs) { return (rs->rs_start == rs->rs_end); } /* * Translate a logical range to the first contiguous physical range for the * specified vdev_t. This function is initially called with a leaf vdev and * will walk each parent vdev until it reaches a top-level vdev. Once the * top-level is reached the physical range is initialized and the recursive * function begins to unwind. As it unwinds it calls the parent's vdev * specific translation function to do the real conversion. */ void vdev_xlate(vdev_t *vd, const range_seg64_t *logical_rs, range_seg64_t *physical_rs, range_seg64_t *remain_rs) { /* * Walk up the vdev tree */ if (vd != vd->vdev_top) { vdev_xlate(vd->vdev_parent, logical_rs, physical_rs, remain_rs); } else { /* * We've reached the top-level vdev, initialize the physical * range to the logical range and set an empty remaining * range then start to unwind. */ physical_rs->rs_start = logical_rs->rs_start; physical_rs->rs_end = logical_rs->rs_end; remain_rs->rs_start = logical_rs->rs_start; remain_rs->rs_end = logical_rs->rs_start; return; } vdev_t *pvd = vd->vdev_parent; ASSERT3P(pvd, !=, NULL); ASSERT3P(pvd->vdev_ops->vdev_op_xlate, !=, NULL); /* * As this recursive function unwinds, translate the logical * range into its physical and any remaining components by calling * the vdev specific translate function. */ range_seg64_t intermediate = { 0 }; pvd->vdev_ops->vdev_op_xlate(vd, physical_rs, &intermediate, remain_rs); physical_rs->rs_start = intermediate.rs_start; physical_rs->rs_end = intermediate.rs_end; } void vdev_xlate_walk(vdev_t *vd, const range_seg64_t *logical_rs, vdev_xlate_func_t *func, void *arg) { range_seg64_t iter_rs = *logical_rs; range_seg64_t physical_rs; range_seg64_t remain_rs; while (!vdev_xlate_is_empty(&iter_rs)) { vdev_xlate(vd, &iter_rs, &physical_rs, &remain_rs); /* * With raidz and dRAID, it's possible that the logical range * does not live on this leaf vdev. Only when there is a non- * zero physical size call the provided function. */ if (!vdev_xlate_is_empty(&physical_rs)) func(arg, &physical_rs); iter_rs = remain_rs; } } /* * Look at the vdev tree and determine whether any devices are currently being * replaced. */ boolean_t vdev_replace_in_progress(vdev_t *vdev) { ASSERT(spa_config_held(vdev->vdev_spa, SCL_ALL, RW_READER) != 0); if (vdev->vdev_ops == &vdev_replacing_ops) return (B_TRUE); /* * A 'spare' vdev indicates that we have a replace in progress, unless * it has exactly two children, and the second, the hot spare, has * finished being resilvered. */ if (vdev->vdev_ops == &vdev_spare_ops && (vdev->vdev_children > 2 || !vdev_dtl_empty(vdev->vdev_child[1], DTL_MISSING))) return (B_TRUE); for (int i = 0; i < vdev->vdev_children; i++) { if (vdev_replace_in_progress(vdev->vdev_child[i])) return (B_TRUE); } return (B_FALSE); } EXPORT_SYMBOL(vdev_fault); EXPORT_SYMBOL(vdev_degrade); EXPORT_SYMBOL(vdev_online); EXPORT_SYMBOL(vdev_offline); EXPORT_SYMBOL(vdev_clear); /* BEGIN CSTYLED */ ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_count, INT, ZMOD_RW, "Target number of metaslabs per top-level vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, default_ms_shift, INT, ZMOD_RW, "Default limit for metaslab size"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, min_ms_count, INT, ZMOD_RW, "Minimum number of metaslabs per top-level vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, ms_count_limit, INT, ZMOD_RW, "Practical upper limit of total metaslabs per top-level vdev"); ZFS_MODULE_PARAM(zfs, zfs_, slow_io_events_per_second, UINT, ZMOD_RW, "Rate limit slow IO (delay) events to this many per second"); ZFS_MODULE_PARAM(zfs, zfs_, checksum_events_per_second, UINT, ZMOD_RW, "Rate limit checksum events to this many checksum errors per second " "(do not set below zed threshold)."); ZFS_MODULE_PARAM(zfs, zfs_, scan_ignore_errors, INT, ZMOD_RW, "Ignore errors during resilver/scrub"); ZFS_MODULE_PARAM(zfs_vdev, vdev_, validate_skip, INT, ZMOD_RW, "Bypass vdev_validate()"); ZFS_MODULE_PARAM(zfs, zfs_, nocacheflush, INT, ZMOD_RW, "Disable cache flushes"); ZFS_MODULE_PARAM(zfs, zfs_, embedded_slog_min_ms, INT, ZMOD_RW, "Minimum number of metaslabs required to dedicate one for log blocks"); ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, min_auto_ashift, param_set_min_auto_ashift, param_get_ulong, ZMOD_RW, "Minimum ashift used when creating new top-level vdevs"); ZFS_MODULE_PARAM_CALL(zfs_vdev, zfs_vdev_, max_auto_ashift, param_set_max_auto_ashift, param_get_ulong, ZMOD_RW, "Maximum ashift used when optimizing for logical -> physical sector " "size on new top-level vdevs"); /* END CSTYLED */