diff --git a/include/sys/vdev_impl.h b/include/sys/vdev_impl.h index ad9dc3aefd8e..3f2312c23438 100644 --- a/include/sys/vdev_impl.h +++ b/include/sys/vdev_impl.h @@ -1,649 +1,652 @@ /* * 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 https://opensource.org/licenses/CDDL-1.0. * 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; struct abd; extern uint_t zfs_vdev_queue_depth_pct; extern uint_t zfs_vdev_def_queue_depth; extern uint_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_kobj_post_evt_func_t(vdev_t *vd); 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; vdev_kobj_post_evt_func_t *vdev_op_kobj_evt_post; char vdev_op_type[16]; boolean_t vdev_op_leaf; } vdev_ops_t; /* * Virtual device properties */ typedef union vdev_queue_class { - list_t vqc_list; + struct { + ulong_t vqc_list_numnodes; + list_t vqc_list; + }; avl_tree_t vqc_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_read_offset_tree; avl_tree_t vq_write_offset_tree; uint64_t vq_last_offset; zio_priority_t vq_last_prio; /* Last sent I/O priority. */ uint32_t vq_cqueued; /* Classes with queued I/Os. */ uint32_t vq_cactive[ZIO_PRIORITY_NUM_QUEUEABLE]; uint32_t vq_active; /* Number of active I/Os. */ uint32_t vq_ia_active; /* Active interactive I/Os. */ uint32_t vq_nia_credit; /* Non-interactive I/Os credit. */ list_t vq_active_list; /* List of active I/Os. */ 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 */ int vdev_validate_error; /* error on last validate */ kthread_t *vdev_open_thread; /* thread opening children */ kthread_t *vdev_validate_thread; /* thread validating children */ uint64_t vdev_crtxg; /* txg when top-level was added */ uint64_t vdev_root_zap; /* * 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 */ 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_noalloc; /* device is passivated? */ uint64_t vdev_removing; /* device is being removed? */ uint64_t vdev_failfast; /* device failfast setting */ 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; kcondvar_t vdev_autotrim_kick_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 */ boolean_t vdev_kobj_flag; /* kobj event record */ boolean_t vdev_attaching; /* vdev attach ashift handling */ vdev_queue_t vdev_queue; /* I/O deadline schedule queue */ 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, deadman, and 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_deadman_rl; zfs_ratelimit_t vdev_checksum_rl; /* * Checksum and IO thresholds for tuning ZED */ uint64_t vdev_checksum_n; uint64_t vdev_checksum_t; uint64_t vdev_io_n; uint64_t vdev_io_t; }; #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 overridden 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; _Static_assert(sizeof (vdev_boot_envblock_t) == VDEV_PAD_SIZE, "vdev_boot_envblock_t wrong 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; /* * 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); uint64_t vdev_best_ashift(uint64_t logical, uint64_t a, uint64_t b); /* * Vdev ashift optimization tunables */ extern uint_t zfs_vdev_min_auto_ashift; extern uint_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/vdev_queue.c b/module/zfs/vdev_queue.c index 08d918467d03..092b3f375be0 100644 --- a/module/zfs/vdev_queue.c +++ b/module/zfs/vdev_queue.c @@ -1,1162 +1,1165 @@ /* * 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 https://opensource.org/licenses/CDDL-1.0. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* * Copyright (c) 2012, 2018 by Delphix. All rights reserved. */ #include #include #include #include #include #include #include #include #include /* * ZFS I/O Scheduler * --------------- * * ZFS issues I/O operations to leaf vdevs to satisfy and complete zios. The * I/O scheduler determines when and in what order those operations are * issued. The I/O scheduler divides operations into five I/O classes * prioritized in the following order: sync read, sync write, async read, * async write, and scrub/resilver. Each queue defines the minimum and * maximum number of concurrent operations that may be issued to the device. * In addition, the device has an aggregate maximum. Note that the sum of the * per-queue minimums must not exceed the aggregate maximum. If the * sum of the per-queue maximums exceeds the aggregate maximum, then the * number of active i/os may reach zfs_vdev_max_active, in which case no * further i/os will be issued regardless of whether all per-queue * minimums have been met. * * For many physical devices, throughput increases with the number of * concurrent operations, but latency typically suffers. Further, physical * devices typically have a limit at which more concurrent operations have no * effect on throughput or can actually cause it to decrease. * * The scheduler selects the next operation to issue by first looking for an * I/O class whose minimum has not been satisfied. Once all are satisfied and * the aggregate maximum has not been hit, the scheduler looks for classes * whose maximum has not been satisfied. Iteration through the I/O classes is * done in the order specified above. No further operations are issued if the * aggregate maximum number of concurrent operations has been hit or if there * are no operations queued for an I/O class that has not hit its maximum. * Every time an i/o is queued or an operation completes, the I/O scheduler * looks for new operations to issue. * * All I/O classes have a fixed maximum number of outstanding operations * except for the async write class. Asynchronous writes represent the data * that is committed to stable storage during the syncing stage for * transaction groups (see txg.c). Transaction groups enter the syncing state * periodically so the number of queued async writes will quickly burst up and * then bleed down to zero. Rather than servicing them as quickly as possible, * the I/O scheduler changes the maximum number of active async write i/os * according to the amount of dirty data in the pool (see dsl_pool.c). Since * both throughput and latency typically increase with the number of * concurrent operations issued to physical devices, reducing the burstiness * in the number of concurrent operations also stabilizes the response time of * operations from other -- and in particular synchronous -- queues. In broad * strokes, the I/O scheduler will issue more concurrent operations from the * async write queue as there's more dirty data in the pool. * * Async Writes * * The number of concurrent operations issued for the async write I/O class * follows a piece-wise linear function defined by a few adjustable points. * * | o---------| <-- zfs_vdev_async_write_max_active * ^ | /^ | * | | / | | * active | / | | * I/O | / | | * count | / | | * | / | | * |------------o | | <-- zfs_vdev_async_write_min_active * 0|____________^______|_________| * 0% | | 100% of zfs_dirty_data_max * | | * | `-- zfs_vdev_async_write_active_max_dirty_percent * `--------- zfs_vdev_async_write_active_min_dirty_percent * * Until the amount of dirty data exceeds a minimum percentage of the dirty * data allowed in the pool, the I/O scheduler will limit the number of * concurrent operations to the minimum. As that threshold is crossed, the * number of concurrent operations issued increases linearly to the maximum at * the specified maximum percentage of the dirty data allowed in the pool. * * Ideally, the amount of dirty data on a busy pool will stay in the sloped * part of the function between zfs_vdev_async_write_active_min_dirty_percent * and zfs_vdev_async_write_active_max_dirty_percent. If it exceeds the * maximum percentage, this indicates that the rate of incoming data is * greater than the rate that the backend storage can handle. In this case, we * must further throttle incoming writes (see dmu_tx_delay() for details). */ /* * The maximum number of i/os active to each device. Ideally, this will be >= * the sum of each queue's max_active. */ uint_t zfs_vdev_max_active = 1000; /* * Per-queue limits on the number of i/os active to each device. If the * number of active i/os is < zfs_vdev_max_active, then the min_active comes * into play. We will send min_active from each queue round-robin, and then * send from queues in the order defined by zio_priority_t up to max_active. * Some queues have additional mechanisms to limit number of active I/Os in * addition to min_active and max_active, see below. * * In general, smaller max_active's will lead to lower latency of synchronous * operations. Larger max_active's may lead to higher overall throughput, * depending on underlying storage. * * The ratio of the queues' max_actives determines the balance of performance * between reads, writes, and scrubs. E.g., increasing * zfs_vdev_scrub_max_active will cause the scrub or resilver to complete * more quickly, but reads and writes to have higher latency and lower * throughput. */ static uint_t zfs_vdev_sync_read_min_active = 10; static uint_t zfs_vdev_sync_read_max_active = 10; static uint_t zfs_vdev_sync_write_min_active = 10; static uint_t zfs_vdev_sync_write_max_active = 10; static uint_t zfs_vdev_async_read_min_active = 1; /* */ uint_t zfs_vdev_async_read_max_active = 3; static uint_t zfs_vdev_async_write_min_active = 2; /* */ uint_t zfs_vdev_async_write_max_active = 10; static uint_t zfs_vdev_scrub_min_active = 1; static uint_t zfs_vdev_scrub_max_active = 3; static uint_t zfs_vdev_removal_min_active = 1; static uint_t zfs_vdev_removal_max_active = 2; static uint_t zfs_vdev_initializing_min_active = 1; static uint_t zfs_vdev_initializing_max_active = 1; static uint_t zfs_vdev_trim_min_active = 1; static uint_t zfs_vdev_trim_max_active = 2; static uint_t zfs_vdev_rebuild_min_active = 1; static uint_t zfs_vdev_rebuild_max_active = 3; /* * When the pool has less than zfs_vdev_async_write_active_min_dirty_percent * dirty data, use zfs_vdev_async_write_min_active. When it has more than * zfs_vdev_async_write_active_max_dirty_percent, use * zfs_vdev_async_write_max_active. The value is linearly interpolated * between min and max. */ uint_t zfs_vdev_async_write_active_min_dirty_percent = 30; uint_t zfs_vdev_async_write_active_max_dirty_percent = 60; /* * For non-interactive I/O (scrub, resilver, removal, initialize and rebuild), * the number of concurrently-active I/O's is limited to *_min_active, unless * the vdev is "idle". When there are no interactive I/Os active (sync or * async), and zfs_vdev_nia_delay I/Os have completed since the last * interactive I/O, then the vdev is considered to be "idle", and the number * of concurrently-active non-interactive I/O's is increased to *_max_active. */ static uint_t zfs_vdev_nia_delay = 5; /* * Some HDDs tend to prioritize sequential I/O so high that concurrent * random I/O latency reaches several seconds. On some HDDs it happens * even if sequential I/Os are submitted one at a time, and so setting * *_max_active to 1 does not help. To prevent non-interactive I/Os, like * scrub, from monopolizing the device no more than zfs_vdev_nia_credit * I/Os can be sent while there are outstanding incomplete interactive * I/Os. This enforced wait ensures the HDD services the interactive I/O * within a reasonable amount of time. */ static uint_t zfs_vdev_nia_credit = 5; /* * To reduce IOPs, we aggregate small adjacent I/Os into one large I/O. * For read I/Os, we also aggregate across small adjacency gaps; for writes * we include spans of optional I/Os to aid aggregation at the disk even when * they aren't able to help us aggregate at this level. */ static uint_t zfs_vdev_aggregation_limit = 1 << 20; static uint_t zfs_vdev_aggregation_limit_non_rotating = SPA_OLD_MAXBLOCKSIZE; static uint_t zfs_vdev_read_gap_limit = 32 << 10; static uint_t zfs_vdev_write_gap_limit = 4 << 10; /* * Define the queue depth percentage for each top-level. This percentage is * used in conjunction with zfs_vdev_async_max_active to determine how many * allocations a specific top-level vdev should handle. Once the queue depth * reaches zfs_vdev_queue_depth_pct * zfs_vdev_async_write_max_active / 100 * then allocator will stop allocating blocks on that top-level device. * The default kernel setting is 1000% which will yield 100 allocations per * device. For userland testing, the default setting is 300% which equates * to 30 allocations per device. */ #ifdef _KERNEL uint_t zfs_vdev_queue_depth_pct = 1000; #else uint_t zfs_vdev_queue_depth_pct = 300; #endif /* * When performing allocations for a given metaslab, we want to make sure that * there are enough IOs to aggregate together to improve throughput. We want to * ensure that there are at least 128k worth of IOs that can be aggregated, and * we assume that the average allocation size is 4k, so we need the queue depth * to be 32 per allocator to get good aggregation of sequential writes. */ uint_t zfs_vdev_def_queue_depth = 32; static int vdev_queue_offset_compare(const void *x1, const void *x2) { const zio_t *z1 = (const zio_t *)x1; const zio_t *z2 = (const zio_t *)x2; int cmp = TREE_CMP(z1->io_offset, z2->io_offset); if (likely(cmp)) return (cmp); return (TREE_PCMP(z1, z2)); } #define VDQ_T_SHIFT 29 static int vdev_queue_to_compare(const void *x1, const void *x2) { const zio_t *z1 = (const zio_t *)x1; const zio_t *z2 = (const zio_t *)x2; int tcmp = TREE_CMP(z1->io_timestamp >> VDQ_T_SHIFT, z2->io_timestamp >> VDQ_T_SHIFT); int ocmp = TREE_CMP(z1->io_offset, z2->io_offset); int cmp = tcmp ? tcmp : ocmp; if (likely(cmp | (z1->io_queue_state == ZIO_QS_NONE))) return (cmp); return (TREE_PCMP(z1, z2)); } static inline boolean_t vdev_queue_class_fifo(zio_priority_t p) { return (p == ZIO_PRIORITY_SYNC_READ || p == ZIO_PRIORITY_SYNC_WRITE || p == ZIO_PRIORITY_TRIM); } static void vdev_queue_class_add(vdev_queue_t *vq, zio_t *zio) { zio_priority_t p = zio->io_priority; vq->vq_cqueued |= 1U << p; - if (vdev_queue_class_fifo(p)) + if (vdev_queue_class_fifo(p)) { list_insert_tail(&vq->vq_class[p].vqc_list, zio); + vq->vq_class[p].vqc_list_numnodes++; + } else avl_add(&vq->vq_class[p].vqc_tree, zio); } static void vdev_queue_class_remove(vdev_queue_t *vq, zio_t *zio) { zio_priority_t p = zio->io_priority; uint32_t empty; if (vdev_queue_class_fifo(p)) { list_t *list = &vq->vq_class[p].vqc_list; list_remove(list, zio); empty = list_is_empty(list); + vq->vq_class[p].vqc_list_numnodes--; } else { avl_tree_t *tree = &vq->vq_class[p].vqc_tree; avl_remove(tree, zio); empty = avl_is_empty(tree); } vq->vq_cqueued &= ~(empty << p); } static uint_t vdev_queue_class_min_active(vdev_queue_t *vq, zio_priority_t p) { switch (p) { case ZIO_PRIORITY_SYNC_READ: return (zfs_vdev_sync_read_min_active); case ZIO_PRIORITY_SYNC_WRITE: return (zfs_vdev_sync_write_min_active); case ZIO_PRIORITY_ASYNC_READ: return (zfs_vdev_async_read_min_active); case ZIO_PRIORITY_ASYNC_WRITE: return (zfs_vdev_async_write_min_active); case ZIO_PRIORITY_SCRUB: return (vq->vq_ia_active == 0 ? zfs_vdev_scrub_min_active : MIN(vq->vq_nia_credit, zfs_vdev_scrub_min_active)); case ZIO_PRIORITY_REMOVAL: return (vq->vq_ia_active == 0 ? zfs_vdev_removal_min_active : MIN(vq->vq_nia_credit, zfs_vdev_removal_min_active)); case ZIO_PRIORITY_INITIALIZING: return (vq->vq_ia_active == 0 ?zfs_vdev_initializing_min_active: MIN(vq->vq_nia_credit, zfs_vdev_initializing_min_active)); case ZIO_PRIORITY_TRIM: return (zfs_vdev_trim_min_active); case ZIO_PRIORITY_REBUILD: return (vq->vq_ia_active == 0 ? zfs_vdev_rebuild_min_active : MIN(vq->vq_nia_credit, zfs_vdev_rebuild_min_active)); default: panic("invalid priority %u", p); return (0); } } static uint_t vdev_queue_max_async_writes(spa_t *spa) { uint_t writes; uint64_t dirty = 0; dsl_pool_t *dp = spa_get_dsl(spa); uint64_t min_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_min_dirty_percent / 100; uint64_t max_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_max_dirty_percent / 100; /* * Async writes may occur before the assignment of the spa's * dsl_pool_t if a self-healing zio is issued prior to the * completion of dmu_objset_open_impl(). */ if (dp == NULL) return (zfs_vdev_async_write_max_active); /* * Sync tasks correspond to interactive user actions. To reduce the * execution time of those actions we push data out as fast as possible. */ dirty = dp->dp_dirty_total; if (dirty > max_bytes || spa_has_pending_synctask(spa)) return (zfs_vdev_async_write_max_active); if (dirty < min_bytes) return (zfs_vdev_async_write_min_active); /* * linear interpolation: * slope = (max_writes - min_writes) / (max_bytes - min_bytes) * move right by min_bytes * move up by min_writes */ writes = (dirty - min_bytes) * (zfs_vdev_async_write_max_active - zfs_vdev_async_write_min_active) / (max_bytes - min_bytes) + zfs_vdev_async_write_min_active; ASSERT3U(writes, >=, zfs_vdev_async_write_min_active); ASSERT3U(writes, <=, zfs_vdev_async_write_max_active); return (writes); } static uint_t vdev_queue_class_max_active(vdev_queue_t *vq, zio_priority_t p) { switch (p) { case ZIO_PRIORITY_SYNC_READ: return (zfs_vdev_sync_read_max_active); case ZIO_PRIORITY_SYNC_WRITE: return (zfs_vdev_sync_write_max_active); case ZIO_PRIORITY_ASYNC_READ: return (zfs_vdev_async_read_max_active); case ZIO_PRIORITY_ASYNC_WRITE: return (vdev_queue_max_async_writes(vq->vq_vdev->vdev_spa)); case ZIO_PRIORITY_SCRUB: if (vq->vq_ia_active > 0) { return (MIN(vq->vq_nia_credit, zfs_vdev_scrub_min_active)); } else if (vq->vq_nia_credit < zfs_vdev_nia_delay) return (MAX(1, zfs_vdev_scrub_min_active)); return (zfs_vdev_scrub_max_active); case ZIO_PRIORITY_REMOVAL: if (vq->vq_ia_active > 0) { return (MIN(vq->vq_nia_credit, zfs_vdev_removal_min_active)); } else if (vq->vq_nia_credit < zfs_vdev_nia_delay) return (MAX(1, zfs_vdev_removal_min_active)); return (zfs_vdev_removal_max_active); case ZIO_PRIORITY_INITIALIZING: if (vq->vq_ia_active > 0) { return (MIN(vq->vq_nia_credit, zfs_vdev_initializing_min_active)); } else if (vq->vq_nia_credit < zfs_vdev_nia_delay) return (MAX(1, zfs_vdev_initializing_min_active)); return (zfs_vdev_initializing_max_active); case ZIO_PRIORITY_TRIM: return (zfs_vdev_trim_max_active); case ZIO_PRIORITY_REBUILD: if (vq->vq_ia_active > 0) { return (MIN(vq->vq_nia_credit, zfs_vdev_rebuild_min_active)); } else if (vq->vq_nia_credit < zfs_vdev_nia_delay) return (MAX(1, zfs_vdev_rebuild_min_active)); return (zfs_vdev_rebuild_max_active); default: panic("invalid priority %u", p); return (0); } } /* * Return the i/o class to issue from, or ZIO_PRIORITY_NUM_QUEUEABLE if * there is no eligible class. */ static zio_priority_t vdev_queue_class_to_issue(vdev_queue_t *vq) { uint32_t cq = vq->vq_cqueued; zio_priority_t p, p1; if (cq == 0 || vq->vq_active >= zfs_vdev_max_active) return (ZIO_PRIORITY_NUM_QUEUEABLE); /* * Find a queue that has not reached its minimum # outstanding i/os. * Do round-robin to reduce starvation due to zfs_vdev_max_active * and vq_nia_credit limits. */ p1 = vq->vq_last_prio + 1; if (p1 >= ZIO_PRIORITY_NUM_QUEUEABLE) p1 = 0; for (p = p1; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) { if ((cq & (1U << p)) != 0 && vq->vq_cactive[p] < vdev_queue_class_min_active(vq, p)) goto found; } for (p = 0; p < p1; p++) { if ((cq & (1U << p)) != 0 && vq->vq_cactive[p] < vdev_queue_class_min_active(vq, p)) goto found; } /* * If we haven't found a queue, look for one that hasn't reached its * maximum # outstanding i/os. */ for (p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) { if ((cq & (1U << p)) != 0 && vq->vq_cactive[p] < vdev_queue_class_max_active(vq, p)) break; } found: vq->vq_last_prio = p; return (p); } void vdev_queue_init(vdev_t *vd) { vdev_queue_t *vq = &vd->vdev_queue; zio_priority_t p; vq->vq_vdev = vd; for (p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) { if (vdev_queue_class_fifo(p)) { list_create(&vq->vq_class[p].vqc_list, sizeof (zio_t), offsetof(struct zio, io_queue_node.l)); } else { avl_create(&vq->vq_class[p].vqc_tree, vdev_queue_to_compare, sizeof (zio_t), offsetof(struct zio, io_queue_node.a)); } } avl_create(&vq->vq_read_offset_tree, vdev_queue_offset_compare, sizeof (zio_t), offsetof(struct zio, io_offset_node)); avl_create(&vq->vq_write_offset_tree, vdev_queue_offset_compare, sizeof (zio_t), offsetof(struct zio, io_offset_node)); vq->vq_last_offset = 0; list_create(&vq->vq_active_list, sizeof (struct zio), offsetof(struct zio, io_queue_node.l)); mutex_init(&vq->vq_lock, NULL, MUTEX_DEFAULT, NULL); } void vdev_queue_fini(vdev_t *vd) { vdev_queue_t *vq = &vd->vdev_queue; for (zio_priority_t p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) { if (vdev_queue_class_fifo(p)) list_destroy(&vq->vq_class[p].vqc_list); else avl_destroy(&vq->vq_class[p].vqc_tree); } avl_destroy(&vq->vq_read_offset_tree); avl_destroy(&vq->vq_write_offset_tree); list_destroy(&vq->vq_active_list); mutex_destroy(&vq->vq_lock); } static void vdev_queue_io_add(vdev_queue_t *vq, zio_t *zio) { zio->io_queue_state = ZIO_QS_QUEUED; vdev_queue_class_add(vq, zio); if (zio->io_type == ZIO_TYPE_READ) avl_add(&vq->vq_read_offset_tree, zio); else if (zio->io_type == ZIO_TYPE_WRITE) avl_add(&vq->vq_write_offset_tree, zio); } static void vdev_queue_io_remove(vdev_queue_t *vq, zio_t *zio) { vdev_queue_class_remove(vq, zio); if (zio->io_type == ZIO_TYPE_READ) avl_remove(&vq->vq_read_offset_tree, zio); else if (zio->io_type == ZIO_TYPE_WRITE) avl_remove(&vq->vq_write_offset_tree, zio); zio->io_queue_state = ZIO_QS_NONE; } static boolean_t vdev_queue_is_interactive(zio_priority_t p) { switch (p) { case ZIO_PRIORITY_SCRUB: case ZIO_PRIORITY_REMOVAL: case ZIO_PRIORITY_INITIALIZING: case ZIO_PRIORITY_REBUILD: return (B_FALSE); default: return (B_TRUE); } } static void vdev_queue_pending_add(vdev_queue_t *vq, zio_t *zio) { ASSERT(MUTEX_HELD(&vq->vq_lock)); ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); vq->vq_cactive[zio->io_priority]++; vq->vq_active++; if (vdev_queue_is_interactive(zio->io_priority)) { if (++vq->vq_ia_active == 1) vq->vq_nia_credit = 1; } else if (vq->vq_ia_active > 0) { vq->vq_nia_credit--; } zio->io_queue_state = ZIO_QS_ACTIVE; list_insert_tail(&vq->vq_active_list, zio); } static void vdev_queue_pending_remove(vdev_queue_t *vq, zio_t *zio) { ASSERT(MUTEX_HELD(&vq->vq_lock)); ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); vq->vq_cactive[zio->io_priority]--; vq->vq_active--; if (vdev_queue_is_interactive(zio->io_priority)) { if (--vq->vq_ia_active == 0) vq->vq_nia_credit = 0; else vq->vq_nia_credit = zfs_vdev_nia_credit; } else if (vq->vq_ia_active == 0) vq->vq_nia_credit++; list_remove(&vq->vq_active_list, zio); zio->io_queue_state = ZIO_QS_NONE; } static void vdev_queue_agg_io_done(zio_t *aio) { abd_free(aio->io_abd); } /* * Compute the range spanned by two i/os, which is the endpoint of the last * (lio->io_offset + lio->io_size) minus start of the first (fio->io_offset). * Conveniently, the gap between fio and lio is given by -IO_SPAN(lio, fio); * thus fio and lio are adjacent if and only if IO_SPAN(lio, fio) == 0. */ #define IO_SPAN(fio, lio) ((lio)->io_offset + (lio)->io_size - (fio)->io_offset) #define IO_GAP(fio, lio) (-IO_SPAN(lio, fio)) /* * Sufficiently adjacent io_offset's in ZIOs will be aggregated. We do this * by creating a gang ABD from the adjacent ZIOs io_abd's. By using * a gang ABD we avoid doing memory copies to and from the parent, * child ZIOs. The gang ABD also accounts for gaps between adjacent * io_offsets by simply getting the zero ABD for writes or allocating * a new ABD for reads and placing them in the gang ABD as well. */ static zio_t * vdev_queue_aggregate(vdev_queue_t *vq, zio_t *zio) { zio_t *first, *last, *aio, *dio, *mandatory, *nio; uint64_t maxgap = 0; uint64_t size; uint64_t limit; boolean_t stretch = B_FALSE; uint64_t next_offset; abd_t *abd; avl_tree_t *t; /* * TRIM aggregation should not be needed since code in zfs_trim.c can * submit TRIM I/O for extents up to zfs_trim_extent_bytes_max (128M). */ if (zio->io_type == ZIO_TYPE_TRIM) return (NULL); if (zio->io_flags & ZIO_FLAG_DONT_AGGREGATE) return (NULL); if (vq->vq_vdev->vdev_nonrot) limit = zfs_vdev_aggregation_limit_non_rotating; else limit = zfs_vdev_aggregation_limit; if (limit == 0) return (NULL); limit = MIN(limit, SPA_MAXBLOCKSIZE); /* * I/Os to distributed spares are directly dispatched to the dRAID * leaf vdevs for aggregation. See the comment at the end of the * zio_vdev_io_start() function. */ ASSERT(vq->vq_vdev->vdev_ops != &vdev_draid_spare_ops); first = last = zio; if (zio->io_type == ZIO_TYPE_READ) { maxgap = zfs_vdev_read_gap_limit; t = &vq->vq_read_offset_tree; } else { ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE); t = &vq->vq_write_offset_tree; } /* * We can aggregate I/Os that are sufficiently adjacent and of * the same flavor, as expressed by the AGG_INHERIT flags. * The latter requirement is necessary so that certain * attributes of the I/O, such as whether it's a normal I/O * or a scrub/resilver, can be preserved in the aggregate. * We can include optional I/Os, but don't allow them * to begin a range as they add no benefit in that situation. */ /* * We keep track of the last non-optional I/O. */ mandatory = (first->io_flags & ZIO_FLAG_OPTIONAL) ? NULL : first; /* * Walk backwards through sufficiently contiguous I/Os * recording the last non-optional I/O. */ zio_flag_t flags = zio->io_flags & ZIO_FLAG_AGG_INHERIT; while ((dio = AVL_PREV(t, first)) != NULL && (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags && IO_SPAN(dio, last) <= limit && IO_GAP(dio, first) <= maxgap && dio->io_type == zio->io_type) { first = dio; if (mandatory == NULL && !(first->io_flags & ZIO_FLAG_OPTIONAL)) mandatory = first; } /* * Skip any initial optional I/Os. */ while ((first->io_flags & ZIO_FLAG_OPTIONAL) && first != last) { first = AVL_NEXT(t, first); ASSERT(first != NULL); } /* * Walk forward through sufficiently contiguous I/Os. * The aggregation limit does not apply to optional i/os, so that * we can issue contiguous writes even if they are larger than the * aggregation limit. */ while ((dio = AVL_NEXT(t, last)) != NULL && (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags && (IO_SPAN(first, dio) <= limit || (dio->io_flags & ZIO_FLAG_OPTIONAL)) && IO_SPAN(first, dio) <= SPA_MAXBLOCKSIZE && IO_GAP(last, dio) <= maxgap && dio->io_type == zio->io_type) { last = dio; if (!(last->io_flags & ZIO_FLAG_OPTIONAL)) mandatory = last; } /* * Now that we've established the range of the I/O aggregation * we must decide what to do with trailing optional I/Os. * For reads, there's nothing to do. While we are unable to * aggregate further, it's possible that a trailing optional * I/O would allow the underlying device to aggregate with * subsequent I/Os. We must therefore determine if the next * non-optional I/O is close enough to make aggregation * worthwhile. */ if (zio->io_type == ZIO_TYPE_WRITE && mandatory != NULL) { zio_t *nio = last; while ((dio = AVL_NEXT(t, nio)) != NULL && IO_GAP(nio, dio) == 0 && IO_GAP(mandatory, dio) <= zfs_vdev_write_gap_limit) { nio = dio; if (!(nio->io_flags & ZIO_FLAG_OPTIONAL)) { stretch = B_TRUE; break; } } } if (stretch) { /* * We are going to include an optional io in our aggregated * span, thus closing the write gap. Only mandatory i/os can * start aggregated spans, so make sure that the next i/o * after our span is mandatory. */ dio = AVL_NEXT(t, last); ASSERT3P(dio, !=, NULL); dio->io_flags &= ~ZIO_FLAG_OPTIONAL; } else { /* do not include the optional i/o */ while (last != mandatory && last != first) { ASSERT(last->io_flags & ZIO_FLAG_OPTIONAL); last = AVL_PREV(t, last); ASSERT(last != NULL); } } if (first == last) return (NULL); size = IO_SPAN(first, last); ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); abd = abd_alloc_gang(); if (abd == NULL) return (NULL); aio = zio_vdev_delegated_io(first->io_vd, first->io_offset, abd, size, first->io_type, zio->io_priority, flags | ZIO_FLAG_DONT_QUEUE, vdev_queue_agg_io_done, NULL); aio->io_timestamp = first->io_timestamp; nio = first; next_offset = first->io_offset; do { dio = nio; nio = AVL_NEXT(t, dio); ASSERT3P(dio, !=, NULL); zio_add_child(dio, aio); vdev_queue_io_remove(vq, dio); if (dio->io_offset != next_offset) { /* allocate a buffer for a read gap */ ASSERT3U(dio->io_type, ==, ZIO_TYPE_READ); ASSERT3U(dio->io_offset, >, next_offset); abd = abd_alloc_for_io( dio->io_offset - next_offset, B_TRUE); abd_gang_add(aio->io_abd, abd, B_TRUE); } if (dio->io_abd && (dio->io_size != abd_get_size(dio->io_abd))) { /* abd size not the same as IO size */ ASSERT3U(abd_get_size(dio->io_abd), >, dio->io_size); abd = abd_get_offset_size(dio->io_abd, 0, dio->io_size); abd_gang_add(aio->io_abd, abd, B_TRUE); } else { if (dio->io_flags & ZIO_FLAG_NODATA) { /* allocate a buffer for a write gap */ ASSERT3U(dio->io_type, ==, ZIO_TYPE_WRITE); ASSERT3P(dio->io_abd, ==, NULL); abd_gang_add(aio->io_abd, abd_get_zeros(dio->io_size), B_TRUE); } else { /* * We pass B_FALSE to abd_gang_add() * because we did not allocate a new * ABD, so it is assumed the caller * will free this ABD. */ abd_gang_add(aio->io_abd, dio->io_abd, B_FALSE); } } next_offset = dio->io_offset + dio->io_size; } while (dio != last); ASSERT3U(abd_get_size(aio->io_abd), ==, aio->io_size); /* * Callers must call zio_vdev_io_bypass() and zio_execute() for * aggregated (parent) I/Os so that we could avoid dropping the * queue's lock here to avoid a deadlock that we could encounter * due to lock order reversal between vq_lock and io_lock in * zio_change_priority(). */ return (aio); } static zio_t * vdev_queue_io_to_issue(vdev_queue_t *vq) { zio_t *zio, *aio; zio_priority_t p; avl_index_t idx; avl_tree_t *tree; again: ASSERT(MUTEX_HELD(&vq->vq_lock)); p = vdev_queue_class_to_issue(vq); if (p == ZIO_PRIORITY_NUM_QUEUEABLE) { /* No eligible queued i/os */ return (NULL); } if (vdev_queue_class_fifo(p)) { zio = list_head(&vq->vq_class[p].vqc_list); } else { /* * For LBA-ordered queues (async / scrub / initializing), * issue the I/O which follows the most recently issued I/O * in LBA (offset) order, but to avoid starvation only within * the same 0.5 second interval as the first I/O. */ tree = &vq->vq_class[p].vqc_tree; zio = aio = avl_first(tree); if (zio->io_offset < vq->vq_last_offset) { vq->vq_io_search.io_timestamp = zio->io_timestamp; vq->vq_io_search.io_offset = vq->vq_last_offset; zio = avl_find(tree, &vq->vq_io_search, &idx); if (zio == NULL) { zio = avl_nearest(tree, idx, AVL_AFTER); if (zio == NULL || (zio->io_timestamp >> VDQ_T_SHIFT) != (aio->io_timestamp >> VDQ_T_SHIFT)) zio = aio; } } } ASSERT3U(zio->io_priority, ==, p); aio = vdev_queue_aggregate(vq, zio); if (aio != NULL) { zio = aio; } else { vdev_queue_io_remove(vq, zio); /* * If the I/O is or was optional and therefore has no data, we * need to simply discard it. We need to drop the vdev queue's * lock to avoid a deadlock that we could encounter since this * I/O will complete immediately. */ if (zio->io_flags & ZIO_FLAG_NODATA) { mutex_exit(&vq->vq_lock); zio_vdev_io_bypass(zio); zio_execute(zio); mutex_enter(&vq->vq_lock); goto again; } } vdev_queue_pending_add(vq, zio); vq->vq_last_offset = zio->io_offset + zio->io_size; return (zio); } zio_t * vdev_queue_io(zio_t *zio) { vdev_queue_t *vq = &zio->io_vd->vdev_queue; zio_t *dio, *nio; zio_link_t *zl = NULL; if (zio->io_flags & ZIO_FLAG_DONT_QUEUE) return (zio); /* * Children i/os inherent their parent's priority, which might * not match the child's i/o type. Fix it up here. */ if (zio->io_type == ZIO_TYPE_READ) { ASSERT(zio->io_priority != ZIO_PRIORITY_TRIM); if (zio->io_priority != ZIO_PRIORITY_SYNC_READ && zio->io_priority != ZIO_PRIORITY_ASYNC_READ && zio->io_priority != ZIO_PRIORITY_SCRUB && zio->io_priority != ZIO_PRIORITY_REMOVAL && zio->io_priority != ZIO_PRIORITY_INITIALIZING && zio->io_priority != ZIO_PRIORITY_REBUILD) { zio->io_priority = ZIO_PRIORITY_ASYNC_READ; } } else if (zio->io_type == ZIO_TYPE_WRITE) { ASSERT(zio->io_priority != ZIO_PRIORITY_TRIM); if (zio->io_priority != ZIO_PRIORITY_SYNC_WRITE && zio->io_priority != ZIO_PRIORITY_ASYNC_WRITE && zio->io_priority != ZIO_PRIORITY_REMOVAL && zio->io_priority != ZIO_PRIORITY_INITIALIZING && zio->io_priority != ZIO_PRIORITY_REBUILD) { zio->io_priority = ZIO_PRIORITY_ASYNC_WRITE; } } else { ASSERT(zio->io_type == ZIO_TYPE_TRIM); ASSERT(zio->io_priority == ZIO_PRIORITY_TRIM); } zio->io_flags |= ZIO_FLAG_DONT_QUEUE; zio->io_timestamp = gethrtime(); mutex_enter(&vq->vq_lock); vdev_queue_io_add(vq, zio); nio = vdev_queue_io_to_issue(vq); mutex_exit(&vq->vq_lock); if (nio == NULL) return (NULL); if (nio->io_done == vdev_queue_agg_io_done) { while ((dio = zio_walk_parents(nio, &zl)) != NULL) { ASSERT3U(dio->io_type, ==, nio->io_type); zio_vdev_io_bypass(dio); zio_execute(dio); } zio_nowait(nio); return (NULL); } return (nio); } void vdev_queue_io_done(zio_t *zio) { vdev_queue_t *vq = &zio->io_vd->vdev_queue; zio_t *dio, *nio; zio_link_t *zl = NULL; hrtime_t now = gethrtime(); vq->vq_io_complete_ts = now; vq->vq_io_delta_ts = zio->io_delta = now - zio->io_timestamp; mutex_enter(&vq->vq_lock); vdev_queue_pending_remove(vq, zio); while ((nio = vdev_queue_io_to_issue(vq)) != NULL) { mutex_exit(&vq->vq_lock); if (nio->io_done == vdev_queue_agg_io_done) { while ((dio = zio_walk_parents(nio, &zl)) != NULL) { ASSERT3U(dio->io_type, ==, nio->io_type); zio_vdev_io_bypass(dio); zio_execute(dio); } zio_nowait(nio); } else { zio_vdev_io_reissue(nio); zio_execute(nio); } mutex_enter(&vq->vq_lock); } mutex_exit(&vq->vq_lock); } void vdev_queue_change_io_priority(zio_t *zio, zio_priority_t priority) { vdev_queue_t *vq = &zio->io_vd->vdev_queue; /* * ZIO_PRIORITY_NOW is used by the vdev cache code and the aggregate zio * code to issue IOs without adding them to the vdev queue. In this * case, the zio is already going to be issued as quickly as possible * and so it doesn't need any reprioritization to help. */ if (zio->io_priority == ZIO_PRIORITY_NOW) return; ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); ASSERT3U(priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); if (zio->io_type == ZIO_TYPE_READ) { if (priority != ZIO_PRIORITY_SYNC_READ && priority != ZIO_PRIORITY_ASYNC_READ && priority != ZIO_PRIORITY_SCRUB) priority = ZIO_PRIORITY_ASYNC_READ; } else { ASSERT(zio->io_type == ZIO_TYPE_WRITE); if (priority != ZIO_PRIORITY_SYNC_WRITE && priority != ZIO_PRIORITY_ASYNC_WRITE) priority = ZIO_PRIORITY_ASYNC_WRITE; } mutex_enter(&vq->vq_lock); /* * If the zio is in none of the queues we can simply change * the priority. If the zio is waiting to be submitted we must * remove it from the queue and re-insert it with the new priority. * Otherwise, the zio is currently active and we cannot change its * priority. */ if (zio->io_queue_state == ZIO_QS_QUEUED) { vdev_queue_class_remove(vq, zio); zio->io_priority = priority; vdev_queue_class_add(vq, zio); } else if (zio->io_queue_state == ZIO_QS_NONE) { zio->io_priority = priority; } mutex_exit(&vq->vq_lock); } /* * As these two methods are only used for load calculations we're not * concerned if we get an incorrect value on 32bit platforms due to lack of * vq_lock mutex use here, instead we prefer to keep it lock free for * performance. */ uint32_t vdev_queue_length(vdev_t *vd) { return (vd->vdev_queue.vq_active); } uint64_t vdev_queue_last_offset(vdev_t *vd) { return (vd->vdev_queue.vq_last_offset); } uint64_t vdev_queue_class_length(vdev_t *vd, zio_priority_t p) { vdev_queue_t *vq = &vd->vdev_queue; if (vdev_queue_class_fifo(p)) - return (list_is_empty(&vq->vq_class[p].vqc_list) == 0); + return (vq->vq_class[p].vqc_list_numnodes); else return (avl_numnodes(&vq->vq_class[p].vqc_tree)); } ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, aggregation_limit, UINT, ZMOD_RW, "Max vdev I/O aggregation size"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, aggregation_limit_non_rotating, UINT, ZMOD_RW, "Max vdev I/O aggregation size for non-rotating media"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, read_gap_limit, UINT, ZMOD_RW, "Aggregate read I/O over gap"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, write_gap_limit, UINT, ZMOD_RW, "Aggregate write I/O over gap"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, max_active, UINT, ZMOD_RW, "Maximum number of active I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, async_write_active_max_dirty_percent, UINT, ZMOD_RW, "Async write concurrency max threshold"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, async_write_active_min_dirty_percent, UINT, ZMOD_RW, "Async write concurrency min threshold"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, async_read_max_active, UINT, ZMOD_RW, "Max active async read I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, async_read_min_active, UINT, ZMOD_RW, "Min active async read I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, async_write_max_active, UINT, ZMOD_RW, "Max active async write I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, async_write_min_active, UINT, ZMOD_RW, "Min active async write I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, initializing_max_active, UINT, ZMOD_RW, "Max active initializing I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, initializing_min_active, UINT, ZMOD_RW, "Min active initializing I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, removal_max_active, UINT, ZMOD_RW, "Max active removal I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, removal_min_active, UINT, ZMOD_RW, "Min active removal I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, scrub_max_active, UINT, ZMOD_RW, "Max active scrub I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, scrub_min_active, UINT, ZMOD_RW, "Min active scrub I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, sync_read_max_active, UINT, ZMOD_RW, "Max active sync read I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, sync_read_min_active, UINT, ZMOD_RW, "Min active sync read I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, sync_write_max_active, UINT, ZMOD_RW, "Max active sync write I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, sync_write_min_active, UINT, ZMOD_RW, "Min active sync write I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, trim_max_active, UINT, ZMOD_RW, "Max active trim/discard I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, trim_min_active, UINT, ZMOD_RW, "Min active trim/discard I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, rebuild_max_active, UINT, ZMOD_RW, "Max active rebuild I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, rebuild_min_active, UINT, ZMOD_RW, "Min active rebuild I/Os per vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, nia_credit, UINT, ZMOD_RW, "Number of non-interactive I/Os to allow in sequence"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, nia_delay, UINT, ZMOD_RW, "Number of non-interactive I/Os before _max_active"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, queue_depth_pct, UINT, ZMOD_RW, "Queue depth percentage for each top-level vdev"); ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, def_queue_depth, UINT, ZMOD_RW, "Default queue depth for each allocator");