diff options
author | Brian Behlendorf <[email protected]> | 2011-12-22 12:20:43 -0800 |
---|---|---|
committer | Brian Behlendorf <[email protected]> | 2012-01-11 11:46:02 -0800 |
commit | ab26409db753bb087842ab6f1af943f3386c764f (patch) | |
tree | 7baa51b121ef9a866371bd750b9949688f0c29e8 /module | |
parent | 8eaa020b46e80b09cc5f924b90809e219ce08e75 (diff) |
Linux 3.1 compat, super_block->s_shrink
The Linux 3.1 kernel has introduced the concept of per-filesystem
shrinkers which are directly assoicated with a super block. Prior
to this change there was one shared global shrinker.
The zfs code relied on being able to call the global shrinker when
the arc_meta_limit was exceeded. This would cause the VFS to drop
references on a fraction of the dentries in the dcache. The ARC
could then safely reclaim the memory used by these entries and
honor the arc_meta_limit. Unfortunately, when per-filesystem
shrinkers were added the old interfaces were made unavailable.
This change adds support to use the new per-filesystem shrinker
interface so we can continue to honor the arc_meta_limit. The
major benefit of the new interface is that we can now target
only the zfs filesystem for dentry and inode pruning. Thus we
can minimize any impact on the caching of other filesystems.
In the context of making this change several other important
issues related to managing the ARC were addressed, they include:
* The dnlc_reduce_cache() function which was called by the ARC
to drop dentries for the Posix layer was replaced with a generic
zfs_prune_t callback. The ZPL layer now registers a callback to
drop these dentries removing a layering violation which dates
back to the Solaris code. This callback can also be used by
other ARC consumers such as Lustre.
arc_add_prune_callback()
arc_remove_prune_callback()
* The arc_reduce_dnlc_percent module option has been changed to
arc_meta_prune for clarity. The dnlc functions are specific to
Solaris's VFS and have already been largely eliminated already.
The replacement tunable now represents the number of bytes the
prune callback will request when invoked.
* Less aggressively invoke the prune callback. We used to call
this whenever we exceeded the arc_meta_limit however that's not
strictly correct since it results in over zeleous reclaim of
dentries and inodes. It is now only called once the arc_meta_limit
is exceeded and every effort has been made to evict other data from
the ARC cache.
* More promptly manage exceeding the arc_meta_limit. When reading
meta data in to the cache if a buffer was unable to be recycled
notify the arc_reclaim thread to invoke the required prune.
* Added arcstat_prune kstat which is incremented when the ARC
is forced to request that a consumer prune its cache. Remember
this will only occur when the ARC has no other choice. If it
can evict buffers safely without invoking the prune callback
it will.
* This change is also expected to resolve the unexpect collapses
of the ARC cache. This would occur because when exceeded just the
arc_meta_limit reclaim presure would be excerted on the arc_c
value via arc_shrink(). This effectively shrunk the entire cache
when really we just needed to reclaim meta data.
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #466
Closes #292
Diffstat (limited to 'module')
-rw-r--r-- | module/zfs/arc.c | 188 | ||||
-rw-r--r-- | module/zfs/zfs_vfsops.c | 21 | ||||
-rw-r--r-- | module/zfs/zfs_znode.c | 2 | ||||
-rw-r--r-- | module/zfs/zpl_super.c | 126 |
4 files changed, 287 insertions, 50 deletions
diff --git a/module/zfs/arc.c b/module/zfs/arc.c index f71c24844..95d14a9e7 100644 --- a/module/zfs/arc.c +++ b/module/zfs/arc.c @@ -104,6 +104,14 @@ * protected from simultaneous callbacks from arc_buf_evict() * and arc_do_user_evicts(). * + * It as also possible to register a callback which is run when the + * arc_meta_limit is reached and no buffers can be safely evicted. In + * this case the arc user should drop a reference on some arc buffers so + * they can be reclaimed and the arc_meta_limit honored. For example, + * when using the ZPL each dentry holds a references on a znode. These + * dentries must be pruned before the arc buffer holding the znode can + * be safely evicted. + * * Note that the majority of the performance stats are manipulated * with atomic operations. * @@ -120,14 +128,13 @@ #include <sys/zio.h> #include <sys/zfs_context.h> #include <sys/arc.h> -#include <sys/refcount.h> #include <sys/vdev.h> #include <sys/vdev_impl.h> #ifdef _KERNEL #include <sys/vmsystm.h> #include <vm/anon.h> #include <sys/fs/swapnode.h> -#include <sys/dnlc.h> +#include <sys/zpl.h> #endif #include <sys/callb.h> #include <sys/kstat.h> @@ -141,8 +148,8 @@ extern int zfs_write_limit_shift; extern uint64_t zfs_write_limit_max; extern kmutex_t zfs_write_limit_lock; -#define ARC_REDUCE_DNLC_PERCENT 3 -uint_t arc_reduce_dnlc_percent = ARC_REDUCE_DNLC_PERCENT; +/* number of bytes to prune from caches when at arc_meta_limit is reached */ +uint_t arc_meta_prune = 1048576; typedef enum arc_reclaim_strategy { ARC_RECLAIM_AGGR, /* Aggressive reclaim strategy */ @@ -180,7 +187,7 @@ unsigned long zfs_arc_meta_limit = 0; int zfs_arc_grow_retry = 0; int zfs_arc_shrink_shift = 0; int zfs_arc_p_min_shift = 0; -int zfs_arc_reduce_dnlc_percent = 0; +int zfs_arc_meta_prune = 0; /* * Note that buffers can be in one of 6 states: @@ -288,6 +295,7 @@ typedef struct arc_stats { kstat_named_t arcstat_no_grow; kstat_named_t arcstat_tempreserve; kstat_named_t arcstat_loaned_bytes; + kstat_named_t arcstat_prune; kstat_named_t arcstat_meta_used; kstat_named_t arcstat_meta_limit; kstat_named_t arcstat_meta_max; @@ -352,6 +360,7 @@ static arc_stats_t arc_stats = { { "arc_no_grow", KSTAT_DATA_UINT64 }, { "arc_tempreserve", KSTAT_DATA_UINT64 }, { "arc_loaned_bytes", KSTAT_DATA_UINT64 }, + { "arc_prune", KSTAT_DATA_UINT64 }, { "arc_meta_used", KSTAT_DATA_UINT64 }, { "arc_meta_limit", KSTAT_DATA_UINT64 }, { "arc_meta_max", KSTAT_DATA_UINT64 }, @@ -481,6 +490,8 @@ struct arc_buf_hdr { list_node_t b_l2node; }; +static list_t arc_prune_list; +static kmutex_t arc_prune_mtx; static arc_buf_t *arc_eviction_list; static kmutex_t arc_eviction_mtx; static arc_buf_hdr_t arc_eviction_hdr; @@ -1925,6 +1936,48 @@ arc_adjust(void) } } +/* + * Request that arc user drop references so that N bytes can be released + * from the cache. This provides a mechanism to ensure the arc can honor + * the arc_meta_limit and reclaim buffers which are pinned in the cache + * by higher layers. (i.e. the zpl) + */ +static void +arc_do_user_prune(int64_t adjustment) +{ + arc_prune_func_t *func; + void *private; + arc_prune_t *cp, *np; + + mutex_enter(&arc_prune_mtx); + + cp = list_head(&arc_prune_list); + while (cp != NULL) { + func = cp->p_pfunc; + private = cp->p_private; + np = list_next(&arc_prune_list, cp); + refcount_add(&cp->p_refcnt, func); + mutex_exit(&arc_prune_mtx); + + if (func != NULL) + func(adjustment, private); + + mutex_enter(&arc_prune_mtx); + + /* User removed prune callback concurrently with execution */ + if (refcount_remove(&cp->p_refcnt, func) == 0) { + ASSERT(!list_link_active(&cp->p_node)); + refcount_destroy(&cp->p_refcnt); + kmem_free(cp, sizeof (*cp)); + } + + cp = np; + } + + ARCSTAT_BUMP(arcstat_prune); + mutex_exit(&arc_prune_mtx); +} + static void arc_do_user_evicts(void) { @@ -1949,6 +2002,32 @@ arc_do_user_evicts(void) } /* + * Evict only meta data objects from the cache leaving the data objects. + * This is only used to enforce the tunable arc_meta_limit, if we are + * unable to evict enough buffers notify the user via the prune callback. + */ +void +arc_adjust_meta(int64_t adjustment, boolean_t may_prune) +{ + int64_t delta; + + if (adjustment > 0 && arc_mru->arcs_lsize[ARC_BUFC_METADATA] > 0) { + delta = MIN(arc_mru->arcs_lsize[ARC_BUFC_METADATA], adjustment); + arc_evict(arc_mru, 0, delta, FALSE, ARC_BUFC_METADATA); + adjustment -= delta; + } + + if (adjustment > 0 && arc_mfu->arcs_lsize[ARC_BUFC_METADATA] > 0) { + delta = MIN(arc_mfu->arcs_lsize[ARC_BUFC_METADATA], adjustment); + arc_evict(arc_mfu, 0, delta, FALSE, ARC_BUFC_METADATA); + adjustment -= delta; + } + + if (may_prune && (adjustment > 0) && (arc_meta_used > arc_meta_limit)) + arc_do_user_prune(arc_meta_prune); +} + +/* * Flush all *evictable* data from the cache for the given spa. * NOTE: this will not touch "active" (i.e. referenced) data. */ @@ -2085,24 +2164,6 @@ arc_kmem_reap_now(arc_reclaim_strategy_t strat) kmem_cache_t *prev_data_cache = NULL; extern kmem_cache_t *zio_buf_cache[]; extern kmem_cache_t *zio_data_buf_cache[]; -#ifdef _KERNEL - int retry = 0; - - while ((arc_meta_used >= arc_meta_limit) && (retry < 10)) { - /* - * We are exceeding our meta-data cache limit. - * Purge some DNLC entries to release holds on meta-data. - */ - dnlc_reduce_cache((void *)(uintptr_t)arc_reduce_dnlc_percent); - retry++; - } -#if defined(__i386) - /* - * Reclaim unused memory from all kmem caches. - */ - kmem_reap(); -#endif -#endif /* * An aggressive reclamation will shrink the cache size as well as @@ -2121,6 +2182,7 @@ arc_kmem_reap_now(arc_reclaim_strategy_t strat) kmem_cache_reap_now(zio_data_buf_cache[i]); } } + kmem_cache_reap_now(buf_cache); kmem_cache_reap_now(hdr_cache); } @@ -2131,6 +2193,7 @@ arc_reclaim_thread(void) clock_t growtime = 0; arc_reclaim_strategy_t last_reclaim = ARC_RECLAIM_CONS; callb_cpr_t cpr; + int64_t prune; CALLB_CPR_INIT(&cpr, &arc_reclaim_thr_lock, callb_generic_cpr, FTAG); @@ -2160,9 +2223,14 @@ arc_reclaim_thread(void) arc_no_grow = FALSE; } - /* Keep meta data usage within limits */ - if (arc_meta_used >= arc_meta_limit) - arc_kmem_reap_now(ARC_RECLAIM_CONS); + /* + * Keep meta data usage within limits, arc_shrink() is not + * used to avoid collapsing the arc_c value when only the + * arc_meta_limit is being exceeded. + */ + prune = (int64_t)arc_meta_used - (int64_t)arc_meta_limit; + if (prune > 0) + arc_adjust_meta(prune, B_TRUE); arc_adjust(); @@ -2399,16 +2467,27 @@ arc_get_data_buf(arc_buf_t *buf) state = (arc_mru->arcs_lsize[type] >= size && mfu_space > arc_mfu->arcs_size) ? arc_mru : arc_mfu; } + if ((buf->b_data = arc_evict(state, 0, size, TRUE, type)) == NULL) { if (type == ARC_BUFC_METADATA) { buf->b_data = zio_buf_alloc(size); arc_space_consume(size, ARC_SPACE_DATA); + + /* + * If we are unable to recycle an existing meta buffer + * signal the reclaim thread. It will notify users + * via the prune callback to drop references. The + * prune callback in run in the context of the reclaim + * thread to avoid deadlocking on the hash_lock. + */ + cv_signal(&arc_reclaim_thr_cv); } else { ASSERT(type == ARC_BUFC_DATA); buf->b_data = zio_data_buf_alloc(size); ARCSTAT_INCR(arcstat_data_size, size); atomic_add_64(&arc_size, size); } + ARCSTAT_BUMP(arcstat_recycle_miss); } ASSERT(buf->b_data != NULL); @@ -3021,6 +3100,37 @@ top: return (0); } +arc_prune_t * +arc_add_prune_callback(arc_prune_func_t *func, void *private) +{ + arc_prune_t *p; + + p = kmem_alloc(sizeof(*p), KM_SLEEP); + p->p_pfunc = func; + p->p_private = private; + list_link_init(&p->p_node); + refcount_create(&p->p_refcnt); + + mutex_enter(&arc_prune_mtx); + refcount_add(&p->p_refcnt, &arc_prune_list); + list_insert_head(&arc_prune_list, p); + mutex_exit(&arc_prune_mtx); + + return (p); +} + +void +arc_remove_prune_callback(arc_prune_t *p) +{ + mutex_enter(&arc_prune_mtx); + list_remove(&arc_prune_list, p); + if (refcount_remove(&p->p_refcnt, &arc_prune_list) == 0) { + refcount_destroy(&p->p_refcnt); + kmem_free(p, sizeof (*p)); + } + mutex_exit(&arc_prune_mtx); +} + void arc_set_callback(arc_buf_t *buf, arc_evict_func_t *func, void *private) { @@ -3598,8 +3708,8 @@ arc_init(void) if (zfs_arc_p_min_shift > 0) arc_p_min_shift = zfs_arc_p_min_shift; - if (zfs_arc_reduce_dnlc_percent > 0) - arc_reduce_dnlc_percent = zfs_arc_reduce_dnlc_percent; + if (zfs_arc_meta_prune > 0) + arc_meta_prune = zfs_arc_meta_prune; /* if kmem_flags are set, lets try to use less memory */ if (kmem_debugging()) @@ -3646,7 +3756,10 @@ arc_init(void) buf_init(); arc_thread_exit = 0; + list_create(&arc_prune_list, sizeof (arc_prune_t), + offsetof(arc_prune_t, p_node)); arc_eviction_list = NULL; + mutex_init(&arc_prune_mtx, NULL, MUTEX_DEFAULT, NULL); mutex_init(&arc_eviction_mtx, NULL, MUTEX_DEFAULT, NULL); bzero(&arc_eviction_hdr, sizeof (arc_buf_hdr_t)); @@ -3674,6 +3787,8 @@ arc_init(void) void arc_fini(void) { + arc_prune_t *p; + mutex_enter(&arc_reclaim_thr_lock); #ifdef _KERNEL spl_unregister_shrinker(&arc_shrinker); @@ -3693,6 +3808,17 @@ arc_fini(void) arc_ksp = NULL; } + mutex_enter(&arc_prune_mtx); + while ((p = list_head(&arc_prune_list)) != NULL) { + list_remove(&arc_prune_list, p); + refcount_remove(&p->p_refcnt, &arc_prune_list); + refcount_destroy(&p->p_refcnt); + kmem_free(p, sizeof (*p)); + } + mutex_exit(&arc_prune_mtx); + + list_destroy(&arc_prune_list); + mutex_destroy(&arc_prune_mtx); mutex_destroy(&arc_eviction_mtx); mutex_destroy(&arc_reclaim_thr_lock); cv_destroy(&arc_reclaim_thr_cv); @@ -4774,6 +4900,8 @@ l2arc_stop(void) EXPORT_SYMBOL(arc_read); EXPORT_SYMBOL(arc_buf_remove_ref); EXPORT_SYMBOL(arc_getbuf_func); +EXPORT_SYMBOL(arc_add_prune_callback); +EXPORT_SYMBOL(arc_remove_prune_callback); module_param(zfs_arc_min, ulong, 0444); MODULE_PARM_DESC(zfs_arc_min, "Min arc size"); @@ -4784,8 +4912,8 @@ MODULE_PARM_DESC(zfs_arc_max, "Max arc size"); module_param(zfs_arc_meta_limit, ulong, 0444); MODULE_PARM_DESC(zfs_arc_meta_limit, "Meta limit for arc size"); -module_param(zfs_arc_reduce_dnlc_percent, int, 0444); -MODULE_PARM_DESC(zfs_arc_reduce_dnlc_percent, "Meta reclaim percentage"); +module_param(zfs_arc_meta_prune, int, 0444); +MODULE_PARM_DESC(zfs_arc_meta_prune, "Bytes of meta data to prune"); module_param(zfs_arc_grow_retry, int, 0444); MODULE_PARM_DESC(zfs_arc_grow_retry, "Seconds before growing arc size"); diff --git a/module/zfs/zfs_vfsops.c b/module/zfs/zfs_vfsops.c index a0726e117..fb319a547 100644 --- a/module/zfs/zfs_vfsops.c +++ b/module/zfs/zfs_vfsops.c @@ -986,6 +986,26 @@ zfs_root(zfs_sb_t *zsb, struct inode **ipp) } EXPORT_SYMBOL(zfs_root); +#ifdef HAVE_SHRINK +int +zfs_sb_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects) +{ + zfs_sb_t *zsb = sb->s_fs_info; + struct shrinker *shrinker = &sb->s_shrink; + struct shrink_control sc = { + .nr_to_scan = nr_to_scan, + .gfp_mask = GFP_KERNEL, + }; + + ZFS_ENTER(zsb); + *objects = (*shrinker->shrink)(shrinker, &sc); + ZFS_EXIT(zsb); + + return (0); +} +EXPORT_SYMBOL(zfs_sb_prune); +#endif /* HAVE_SHRINK */ + /* * Teardown the zfs_sb_t::z_os. * @@ -1533,6 +1553,7 @@ zfs_init(void) zfs_znode_init(); dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); register_filesystem(&zpl_fs_type); + (void) arc_add_prune_callback(zpl_prune_sbs, NULL); } void diff --git a/module/zfs/zfs_znode.c b/module/zfs/zfs_znode.c index a35e3b5f2..709ae74f8 100644 --- a/module/zfs/zfs_znode.c +++ b/module/zfs/zfs_znode.c @@ -269,6 +269,7 @@ zfs_inode_destroy(struct inode *ip) mutex_enter(&zsb->z_znodes_lock); list_remove(&zsb->z_all_znodes, zp); + zsb->z_nr_znodes--; mutex_exit(&zsb->z_znodes_lock); if (zp->z_acl_cached) { @@ -401,6 +402,7 @@ zfs_znode_alloc(zfs_sb_t *zsb, dmu_buf_t *db, int blksz, mutex_enter(&zsb->z_znodes_lock); list_insert_tail(&zsb->z_all_znodes, zp); + zsb->z_nr_znodes++; membar_producer(); mutex_exit(&zsb->z_znodes_lock); diff --git a/module/zfs/zpl_super.c b/module/zfs/zpl_super.c index 650e9c0d3..3abb26a9e 100644 --- a/module/zfs/zpl_super.c +++ b/module/zfs/zpl_super.c @@ -199,34 +199,120 @@ zpl_kill_sb(struct super_block *sb) kill_anon_super(sb); } +#ifdef HAVE_SHRINK +/* + * Linux 3.1 - 3.x API + * + * The Linux 3.1 API introduced per-sb cache shrinkers to replace the + * global ones. This allows us a mechanism to cleanly target a specific + * zfs file system when the dnode and inode caches grow too large. + * + * In addition, the 3.0 kernel added the iterate_supers_type() helper + * function which is used to safely walk all of the zfs file systems. + */ +static void +zpl_prune_sb(struct super_block *sb, void *arg) +{ + int objects = 0; + int error; + + error = -zfs_sb_prune(sb, *(unsigned long *)arg, &objects); + ASSERT3S(error, <=, 0); + + return; +} + +void +zpl_prune_sbs(int64_t bytes_to_scan, void *private) +{ + unsigned long nr_to_scan = (bytes_to_scan / sizeof(znode_t)); + + iterate_supers_type(&zpl_fs_type, zpl_prune_sb, &nr_to_scan); + kmem_reap(); +} +#else +/* + * Linux 2.6.x - 3.0 API + * + * These are best effort interfaces are provided by the SPL to induce + * the Linux VM subsystem to reclaim a fraction of the both dnode and + * inode caches. Ideally, we want to just target the zfs file systems + * however our only option is to reclaim from them all. + */ +void +zpl_prune_sbs(int64_t bytes_to_scan, void *private) +{ + unsigned long nr_to_scan = (bytes_to_scan / sizeof(znode_t)); + + shrink_dcache_memory(nr_to_scan, GFP_KERNEL); + shrink_icache_memory(nr_to_scan, GFP_KERNEL); + kmem_reap(); +} +#endif /* HAVE_SHRINK */ + +#ifdef HAVE_NR_CACHED_OBJECTS +static int +zpl_nr_cached_objects(struct super_block *sb) +{ + zfs_sb_t *zsb = sb->s_fs_info; + int nr; + + mutex_enter(&zsb->z_znodes_lock); + nr = zsb->z_nr_znodes; + mutex_exit(&zsb->z_znodes_lock); + + return (nr); +} +#endif /* HAVE_NR_CACHED_OBJECTS */ + +#ifdef HAVE_FREE_CACHED_OBJECTS +/* + * Attempt to evict some meta data from the cache. The ARC operates in + * terms of bytes while the Linux VFS uses objects. Now because this is + * just a best effort eviction and the exact values aren't critical so we + * extrapolate from an object count to a byte size using the znode_t size. + */ +static void +zpl_free_cached_objects(struct super_block *sb, int nr_to_scan) +{ + arc_adjust_meta(nr_to_scan * sizeof(znode_t), B_FALSE); +} +#endif /* HAVE_FREE_CACHED_OBJECTS */ + const struct super_operations zpl_super_operations = { - .alloc_inode = zpl_inode_alloc, - .destroy_inode = zpl_inode_destroy, - .dirty_inode = NULL, - .write_inode = NULL, - .drop_inode = NULL, + .alloc_inode = zpl_inode_alloc, + .destroy_inode = zpl_inode_destroy, + .dirty_inode = NULL, + .write_inode = NULL, + .drop_inode = NULL, #ifdef HAVE_EVICT_INODE - .evict_inode = zpl_evict_inode, + .evict_inode = zpl_evict_inode, #else - .clear_inode = zpl_clear_inode, - .delete_inode = zpl_inode_delete, + .clear_inode = zpl_clear_inode, + .delete_inode = zpl_inode_delete, #endif /* HAVE_EVICT_INODE */ - .put_super = zpl_put_super, - .write_super = NULL, - .sync_fs = zpl_sync_fs, - .statfs = zpl_statfs, - .remount_fs = zpl_remount_fs, - .show_options = zpl_show_options, - .show_stats = NULL, + .put_super = zpl_put_super, + .write_super = NULL, + .sync_fs = zpl_sync_fs, + .statfs = zpl_statfs, + .remount_fs = zpl_remount_fs, + .show_options = zpl_show_options, + .show_stats = NULL, +#ifdef HAVE_NR_CACHED_OBJECTS + .nr_cached_objects = zpl_nr_cached_objects, +#endif /* HAVE_NR_CACHED_OBJECTS */ +#ifdef HAVE_FREE_CACHED_OBJECTS + .free_cached_objects = zpl_free_cached_objects, +#endif /* HAVE_FREE_CACHED_OBJECTS */ }; struct file_system_type zpl_fs_type = { - .owner = THIS_MODULE, - .name = ZFS_DRIVER, + .owner = THIS_MODULE, + .name = ZFS_DRIVER, #ifdef HAVE_MOUNT_NODEV - .mount = zpl_mount, + .mount = zpl_mount, #else - .get_sb = zpl_get_sb, + .get_sb = zpl_get_sb, #endif /* HAVE_MOUNT_NODEV */ - .kill_sb = zpl_kill_sb, + .kill_sb = zpl_kill_sb, }; |