File Manager

003 File Manager

Current Path: /usr/src/sys/contrib/openzfs/module/zfs

usr / src / sys / contrib / openzfs / module / zfs /

📁 ..
📄 Makefile.in(4.57 KB)
📄 THIRDPARTYLICENSE.cityhash(1.03 KB)
📄 THIRDPARTYLICENSE.cityhash.descrip(39 B)
📄 abd.c(31.72 KB)
📄 aggsum.c(8.46 KB)
📄 arc.c(330.91 KB)
📄 blkptr.c(4.26 KB)
📄 bplist.c(2.38 KB)
📄 bpobj.c(27.82 KB)
📄 bptree.c(8.2 KB)
📄 bqueue.c(4.7 KB)
📄 btree.c(65.19 KB)
📄 dataset_kstats.c(6.32 KB)
📄 dbuf.c(141.84 KB)
📄 dbuf_stats.c(6.05 KB)
📄 ddt.c(28.01 KB)
📄 ddt_zap.c(4.38 KB)
📄 dmu.c(60.44 KB)
📄 dmu_diff.c(6.25 KB)
📄 dmu_object.c(14.73 KB)
📄 dmu_objset.c(79.38 KB)
📄 dmu_recv.c(98.69 KB)
📄 dmu_redact.c(37.09 KB)
📄 dmu_send.c(95.51 KB)
📄 dmu_traverse.c(21.98 KB)
📄 dmu_tx.c(40.13 KB)
📄 dmu_zfetch.c(13.56 KB)
📄 dnode.c(71.32 KB)
📄 dnode_sync.c(24.93 KB)
📄 dsl_bookmark.c(51.22 KB)
📄 dsl_crypt.c(78.06 KB)
📄 dsl_dataset.c(141.61 KB)
📄 dsl_deadlist.c(27.94 KB)
📄 dsl_deleg.c(19.92 KB)
📄 dsl_destroy.c(36.73 KB)
📄 dsl_dir.c(65.45 KB)
📄 dsl_pool.c(42.92 KB)
📄 dsl_prop.c(32.88 KB)
📄 dsl_scan.c(134.47 KB)
📄 dsl_synctask.c(7.94 KB)
📄 dsl_userhold.c(18.26 KB)
📄 edonr_zfs.c(3.21 KB)
📄 fm.c(39.75 KB)
📄 gzip.c(2.57 KB)
📄 hkdf.c(4.5 KB)
📄 lz4.c(26.46 KB)
📄 lzjb.c(3.85 KB)
📄 metaslab.c(190.62 KB)
📄 mmp.c(24.93 KB)
📄 multilist.c(12.21 KB)
📄 objlist.c(2.58 KB)
📄 pathname.c(2.52 KB)
📄 range_tree.c(24.9 KB)
📄 refcount.c(7.82 KB)
📄 rrwlock.c(10.82 KB)
📄 sa.c(59.23 KB)
📄 sha256.c(2.89 KB)
📄 skein_zfs.c(2.92 KB)
📄 spa.c(272.81 KB)
📄 spa_boot.c(1.28 KB)
📄 spa_checkpoint.c(21.54 KB)
📄 spa_config.c(16.97 KB)
📄 spa_errlog.c(11.35 KB)
📄 spa_history.c(17.75 KB)
📄 spa_log_spacemap.c(47.13 KB)
📄 spa_misc.c(75.4 KB)
📄 spa_stats.c(27.08 KB)
📄 space_map.c(31.45 KB)
📄 space_reftree.c(4.21 KB)
📄 txg.c(27.49 KB)
📄 uberblock.c(2.17 KB)
📄 unique.c(2.53 KB)
📄 vdev.c(143.91 KB)
📄 vdev_cache.c(11.71 KB)
📄 vdev_draid.c(96.86 KB)
📄 vdev_draid_rand.c(1.16 KB)
📄 vdev_indirect.c(60.83 KB)
📄 vdev_indirect_births.c(5.88 KB)
📄 vdev_indirect_mapping.c(18.06 KB)
📄 vdev_initialize.c(22.38 KB)
📄 vdev_label.c(57.52 KB)
📄 vdev_mirror.c(26.76 KB)
📄 vdev_missing.c(3.74 KB)
📄 vdev_queue.c(37.61 KB)
📄 vdev_raidz.c(76.09 KB)
📄 vdev_raidz_math.c(17.18 KB)
📄 vdev_raidz_math_aarch64_neon.c(112.13 KB)
📄 vdev_raidz_math_aarch64_neon_common.h(23.64 KB)
📄 vdev_raidz_math_aarch64_neonx2.c(5.39 KB)
📄 vdev_raidz_math_avx2.c(10.64 KB)
📄 vdev_raidz_math_avx512bw.c(10.93 KB)
📄 vdev_raidz_math_avx512f.c(17.9 KB)
📄 vdev_raidz_math_impl.h(34.78 KB)
📄 vdev_raidz_math_powerpc_altivec.c(213.46 KB)
📄 vdev_raidz_math_powerpc_altivec_common.h(22.62 KB)
📄 vdev_raidz_math_scalar.c(9.54 KB)
📄 vdev_raidz_math_sse2.c(25.84 KB)
📄 vdev_raidz_math_ssse3.c(118 KB)
📄 vdev_rebuild.c(34.65 KB)
📄 vdev_removal.c(72.08 KB)
📄 vdev_root.c(4.28 KB)
📄 vdev_trim.c(50.92 KB)
📄 zap.c(35.35 KB)
📄 zap_leaf.c(22.75 KB)
📄 zap_micro.c(42.18 KB)
📄 zcp.c(42.29 KB)
📄 zcp_get.c(21.01 KB)
📄 zcp_global.c(1.9 KB)
📄 zcp_iter.c(18.21 KB)
📄 zcp_set.c(2.37 KB)
📄 zcp_synctask.c(13.69 KB)
📄 zfeature.c(17.6 KB)
📄 zfs_byteswap.c(5.64 KB)
📄 zfs_fm.c(40.58 KB)
📄 zfs_fuid.c(19.69 KB)
📄 zfs_ioctl.c(195.48 KB)
📄 zfs_log.c(20.6 KB)
📄 zfs_onexit.c(5.09 KB)
📄 zfs_quota.c(12.93 KB)
📄 zfs_ratelimit.c(2.35 KB)
📄 zfs_replay.c(26.11 KB)
📄 zfs_rlock.c(20.38 KB)
📄 zfs_sa.c(13.09 KB)
📄 zfs_vnops.c(22 KB)
📄 zil.c(111.32 KB)
📄 zio.c(144.72 KB)
📄 zio_checksum.c(17.28 KB)
📄 zio_compress.c(5.92 KB)
📄 zio_inject.c(25.79 KB)
📄 zle.c(2.52 KB)
📄 zrlock.c(4.45 KB)
📄 zthr.c(16.41 KB)
📄 zvol.c(42.96 KB)

Editing: vdev_cache.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */
/*
 * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
 */

#include <sys/zfs_context.h>
#include <sys/spa.h>
#include <sys/vdev_impl.h>
#include <sys/zio.h>
#include <sys/kstat.h>
#include <sys/abd.h>

/*
 * Virtual device read-ahead caching.
 *
 * This file implements a simple LRU read-ahead cache.  When the DMU reads
 * a given block, it will often want other, nearby blocks soon thereafter.
 * We take advantage of this by reading a larger disk region and caching
 * the result.  In the best case, this can turn 128 back-to-back 512-byte
 * reads into a single 64k read followed by 127 cache hits; this reduces
 * latency dramatically.  In the worst case, it can turn an isolated 512-byte
 * read into a 64k read, which doesn't affect latency all that much but is
 * terribly wasteful of bandwidth.  A more intelligent version of the cache
 * could keep track of access patterns and not do read-ahead unless it sees
 * at least two temporally close I/Os to the same region.  Currently, only
 * metadata I/O is inflated.  A further enhancement could take advantage of
 * more semantic information about the I/O.  And it could use something
 * faster than an AVL tree; that was chosen solely for convenience.
 *
 * There are five cache operations: allocate, fill, read, write, evict.
 *
 * (1) Allocate.  This reserves a cache entry for the specified region.
 *     We separate the allocate and fill operations so that multiple threads
 *     don't generate I/O for the same cache miss.
 *
 * (2) Fill.  When the I/O for a cache miss completes, the fill routine
 *     places the data in the previously allocated cache entry.
 *
 * (3) Read.  Read data from the cache.
 *
 * (4) Write.  Update cache contents after write completion.
 *
 * (5) Evict.  When allocating a new entry, we evict the oldest (LRU) entry
 *     if the total cache size exceeds zfs_vdev_cache_size.
 */

/*
 * These tunables are for performance analysis.
 */
/*
 * All i/os smaller than zfs_vdev_cache_max will be turned into
 * 1<<zfs_vdev_cache_bshift byte reads by the vdev_cache (aka software
 * track buffer).  At most zfs_vdev_cache_size bytes will be kept in each
 * vdev's vdev_cache.
 *
 * TODO: Note that with the current ZFS code, it turns out that the
 * vdev cache is not helpful, and in some cases actually harmful.  It
 * is better if we disable this.  Once some time has passed, we should
 * actually remove this to simplify the code.  For now we just disable
 * it by setting the zfs_vdev_cache_size to zero.  Note that Solaris 11
 * has made these same changes.
 */
int zfs_vdev_cache_max = 1<<14;			/* 16KB */
int zfs_vdev_cache_size = 0;
int zfs_vdev_cache_bshift = 16;

#define	VCBS (1 << zfs_vdev_cache_bshift)	/* 64KB */

kstat_t	*vdc_ksp = NULL;

typedef struct vdc_stats {
	kstat_named_t vdc_stat_delegations;
	kstat_named_t vdc_stat_hits;
	kstat_named_t vdc_stat_misses;
} vdc_stats_t;

static vdc_stats_t vdc_stats = {
	{ "delegations",	KSTAT_DATA_UINT64 },
	{ "hits",		KSTAT_DATA_UINT64 },
	{ "misses",		KSTAT_DATA_UINT64 }
};

#define	VDCSTAT_BUMP(stat)	atomic_inc_64(&vdc_stats.stat.value.ui64);

static inline int
vdev_cache_offset_compare(const void *a1, const void *a2)
{
	const vdev_cache_entry_t *ve1 = (const vdev_cache_entry_t *)a1;
	const vdev_cache_entry_t *ve2 = (const vdev_cache_entry_t *)a2;

return (TREE_CMP(ve1->ve_offset, ve2->ve_offset));
}

static int
vdev_cache_lastused_compare(const void *a1, const void *a2)
{
	const vdev_cache_entry_t *ve1 = (const vdev_cache_entry_t *)a1;
	const vdev_cache_entry_t *ve2 = (const vdev_cache_entry_t *)a2;

int cmp = TREE_CMP(ve1->ve_lastused, ve2->ve_lastused);
	if (likely(cmp))
		return (cmp);

/*
	 * Among equally old entries, sort by offset to ensure uniqueness.
	 */
	return (vdev_cache_offset_compare(a1, a2));
}

/*
 * Evict the specified entry from the cache.
 */
static void
vdev_cache_evict(vdev_cache_t *vc, vdev_cache_entry_t *ve)
{
	ASSERT(MUTEX_HELD(&vc->vc_lock));
	ASSERT3P(ve->ve_fill_io, ==, NULL);
	ASSERT3P(ve->ve_abd, !=, NULL);

avl_remove(&vc->vc_lastused_tree, ve);
	avl_remove(&vc->vc_offset_tree, ve);
	abd_free(ve->ve_abd);
	kmem_free(ve, sizeof (vdev_cache_entry_t));
}

/*
 * Allocate an entry in the cache.  At the point we don't have the data,
 * we're just creating a placeholder so that multiple threads don't all
 * go off and read the same blocks.
 */
static vdev_cache_entry_t *
vdev_cache_allocate(zio_t *zio)
{
	vdev_cache_t *vc = &zio->io_vd->vdev_cache;
	uint64_t offset = P2ALIGN(zio->io_offset, VCBS);
	vdev_cache_entry_t *ve;

ASSERT(MUTEX_HELD(&vc->vc_lock));

if (zfs_vdev_cache_size == 0)
		return (NULL);

/*
	 * If adding a new entry would exceed the cache size,
	 * evict the oldest entry (LRU).
	 */
	if ((avl_numnodes(&vc->vc_lastused_tree) << zfs_vdev_cache_bshift) >
	    zfs_vdev_cache_size) {
		ve = avl_first(&vc->vc_lastused_tree);
		if (ve->ve_fill_io != NULL)
			return (NULL);
		ASSERT3U(ve->ve_hits, !=, 0);
		vdev_cache_evict(vc, ve);
	}

ve = kmem_zalloc(sizeof (vdev_cache_entry_t), KM_SLEEP);
	ve->ve_offset = offset;
	ve->ve_lastused = ddi_get_lbolt();
	ve->ve_abd = abd_alloc_for_io(VCBS, B_TRUE);

avl_add(&vc->vc_offset_tree, ve);
	avl_add(&vc->vc_lastused_tree, ve);

return (ve);
}

static void
vdev_cache_hit(vdev_cache_t *vc, vdev_cache_entry_t *ve, zio_t *zio)
{
	uint64_t cache_phase = P2PHASE(zio->io_offset, VCBS);

ASSERT(MUTEX_HELD(&vc->vc_lock));
	ASSERT3P(ve->ve_fill_io, ==, NULL);

if (ve->ve_lastused != ddi_get_lbolt()) {
		avl_remove(&vc->vc_lastused_tree, ve);
		ve->ve_lastused = ddi_get_lbolt();
		avl_add(&vc->vc_lastused_tree, ve);
	}

ve->ve_hits++;
	abd_copy_off(zio->io_abd, ve->ve_abd, 0, cache_phase, zio->io_size);
}

/*
 * Fill a previously allocated cache entry with data.
 */
static void
vdev_cache_fill(zio_t *fio)
{
	vdev_t *vd = fio->io_vd;
	vdev_cache_t *vc = &vd->vdev_cache;
	vdev_cache_entry_t *ve = fio->io_private;
	zio_t *pio;

ASSERT3U(fio->io_size, ==, VCBS);

/*
	 * Add data to the cache.
	 */
	mutex_enter(&vc->vc_lock);

ASSERT3P(ve->ve_fill_io, ==, fio);
	ASSERT3U(ve->ve_offset, ==, fio->io_offset);
	ASSERT3P(ve->ve_abd, ==, fio->io_abd);

ve->ve_fill_io = NULL;

/*
	 * Even if this cache line was invalidated by a missed write update,
	 * any reads that were queued up before the missed update are still
	 * valid, so we can satisfy them from this line before we evict it.
	 */
	zio_link_t *zl = NULL;
	while ((pio = zio_walk_parents(fio, &zl)) != NULL)
		vdev_cache_hit(vc, ve, pio);

if (fio->io_error || ve->ve_missed_update)
		vdev_cache_evict(vc, ve);

mutex_exit(&vc->vc_lock);
}

/*
 * Read data from the cache.  Returns B_TRUE cache hit, B_FALSE on miss.
 */
boolean_t
vdev_cache_read(zio_t *zio)
{
	vdev_cache_t *vc = &zio->io_vd->vdev_cache;
	vdev_cache_entry_t *ve, *ve_search;
	uint64_t cache_offset = P2ALIGN(zio->io_offset, VCBS);
	zio_t *fio;
	uint64_t cache_phase __maybe_unused = P2PHASE(zio->io_offset, VCBS);

ASSERT3U(zio->io_type, ==, ZIO_TYPE_READ);

if (zio->io_flags & ZIO_FLAG_DONT_CACHE)
		return (B_FALSE);

if (zio->io_size > zfs_vdev_cache_max)
		return (B_FALSE);

/*
	 * If the I/O straddles two or more cache blocks, don't cache it.
	 */
	if (P2BOUNDARY(zio->io_offset, zio->io_size, VCBS))
		return (B_FALSE);

ASSERT3U(cache_phase + zio->io_size, <=, VCBS);

mutex_enter(&vc->vc_lock);

ve_search = kmem_alloc(sizeof (vdev_cache_entry_t), KM_SLEEP);
	ve_search->ve_offset = cache_offset;
	ve = avl_find(&vc->vc_offset_tree, ve_search, NULL);
	kmem_free(ve_search, sizeof (vdev_cache_entry_t));

if (ve != NULL) {
		if (ve->ve_missed_update) {
			mutex_exit(&vc->vc_lock);
			return (B_FALSE);
		}

if ((fio = ve->ve_fill_io) != NULL) {
			zio_vdev_io_bypass(zio);
			zio_add_child(zio, fio);
			mutex_exit(&vc->vc_lock);
			VDCSTAT_BUMP(vdc_stat_delegations);
			return (B_TRUE);
		}

vdev_cache_hit(vc, ve, zio);
		zio_vdev_io_bypass(zio);

mutex_exit(&vc->vc_lock);
		VDCSTAT_BUMP(vdc_stat_hits);
		return (B_TRUE);
	}

ve = vdev_cache_allocate(zio);

if (ve == NULL) {
		mutex_exit(&vc->vc_lock);
		return (B_FALSE);
	}

fio = zio_vdev_delegated_io(zio->io_vd, cache_offset,
	    ve->ve_abd, VCBS, ZIO_TYPE_READ, ZIO_PRIORITY_NOW,
	    ZIO_FLAG_DONT_CACHE, vdev_cache_fill, ve);

ve->ve_fill_io = fio;
	zio_vdev_io_bypass(zio);
	zio_add_child(zio, fio);

mutex_exit(&vc->vc_lock);
	zio_nowait(fio);
	VDCSTAT_BUMP(vdc_stat_misses);

return (B_TRUE);
}

/*
 * Update cache contents upon write completion.
 */
void
vdev_cache_write(zio_t *zio)
{
	vdev_cache_t *vc = &zio->io_vd->vdev_cache;
	vdev_cache_entry_t *ve, ve_search;
	uint64_t io_start = zio->io_offset;
	uint64_t io_end = io_start + zio->io_size;
	uint64_t min_offset = P2ALIGN(io_start, VCBS);
	uint64_t max_offset = P2ROUNDUP(io_end, VCBS);
	avl_index_t where;

ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE);

mutex_enter(&vc->vc_lock);

ve_search.ve_offset = min_offset;
	ve = avl_find(&vc->vc_offset_tree, &ve_search, &where);

if (ve == NULL)
		ve = avl_nearest(&vc->vc_offset_tree, where, AVL_AFTER);

while (ve != NULL && ve->ve_offset < max_offset) {
		uint64_t start = MAX(ve->ve_offset, io_start);
		uint64_t end = MIN(ve->ve_offset + VCBS, io_end);

if (ve->ve_fill_io != NULL) {
			ve->ve_missed_update = 1;
		} else {
			abd_copy_off(ve->ve_abd, zio->io_abd,
			    start - ve->ve_offset, start - io_start,
			    end - start);
		}
		ve = AVL_NEXT(&vc->vc_offset_tree, ve);
	}
	mutex_exit(&vc->vc_lock);
}

void
vdev_cache_purge(vdev_t *vd)
{
	vdev_cache_t *vc = &vd->vdev_cache;
	vdev_cache_entry_t *ve;

mutex_enter(&vc->vc_lock);
	while ((ve = avl_first(&vc->vc_offset_tree)) != NULL)
		vdev_cache_evict(vc, ve);
	mutex_exit(&vc->vc_lock);
}

void
vdev_cache_init(vdev_t *vd)
{
	vdev_cache_t *vc = &vd->vdev_cache;

mutex_init(&vc->vc_lock, NULL, MUTEX_DEFAULT, NULL);

avl_create(&vc->vc_offset_tree, vdev_cache_offset_compare,
	    sizeof (vdev_cache_entry_t),
	    offsetof(struct vdev_cache_entry, ve_offset_node));

avl_create(&vc->vc_lastused_tree, vdev_cache_lastused_compare,
	    sizeof (vdev_cache_entry_t),
	    offsetof(struct vdev_cache_entry, ve_lastused_node));
}

void
vdev_cache_fini(vdev_t *vd)
{
	vdev_cache_t *vc = &vd->vdev_cache;

vdev_cache_purge(vd);

avl_destroy(&vc->vc_offset_tree);
	avl_destroy(&vc->vc_lastused_tree);

mutex_destroy(&vc->vc_lock);
}

void
vdev_cache_stat_init(void)
{
	vdc_ksp = kstat_create("zfs", 0, "vdev_cache_stats", "misc",
	    KSTAT_TYPE_NAMED, sizeof (vdc_stats) / sizeof (kstat_named_t),
	    KSTAT_FLAG_VIRTUAL);
	if (vdc_ksp != NULL) {
		vdc_ksp->ks_data = &vdc_stats;
		kstat_install(vdc_ksp);
	}
}

void
vdev_cache_stat_fini(void)
{
	if (vdc_ksp != NULL) {
		kstat_delete(vdc_ksp);
		vdc_ksp = NULL;
	}
}

/* BEGIN CSTYLED */
ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, cache_max, INT, ZMOD_RW,
	"Inflate reads small than max");

ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, cache_size, INT, ZMOD_RD,
	"Total size of the per-disk cache");

ZFS_MODULE_PARAM(zfs_vdev, zfs_vdev_, cache_bshift, INT, ZMOD_RW,
	"Shift size to inflate reads too");
/* END CSTYLED */

003 File Manager

Editing: vdev_cache.c

Upload File

Create Folder