vdev_cache.c

Go to the documentation of this file.

/*
 * 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.
 */

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

/*
 * These tunables are for performance analysis.
 */
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 */

SYSCTL_DECL(_vfs_zfs_vdev);
SYSCTL_NODE(_vfs_zfs_vdev, OID_AUTO, cache, CTLFLAG_RW, 0, "ZFS VDEV Cache");
TUNABLE_INT("vfs.zfs.vdev.cache.max", &zfs_vdev_cache_max);
SYSCTL_INT(_vfs_zfs_vdev_cache, OID_AUTO, max, CTLFLAG_RDTUN,
    &zfs_vdev_cache_max, 0, "Maximum I/O request size that increase read size");
TUNABLE_INT("vfs.zfs.vdev.cache.size", &zfs_vdev_cache_size);
SYSCTL_INT(_vfs_zfs_vdev_cache, OID_AUTO, size, CTLFLAG_RDTUN,
    &zfs_vdev_cache_size, 0, "Size of VDEV cache");
TUNABLE_INT("vfs.zfs.vdev.cache.bshift", &zfs_vdev_cache_bshift);
SYSCTL_INT(_vfs_zfs_vdev_cache, OID_AUTO, bshift, CTLFLAG_RDTUN,
    &zfs_vdev_cache_bshift, 0, "Turn too small requests into 1 << this value");

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_add_64(&vdc_stats.stat.value.ui64, 1);

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

        if (ve1->ve_offset < ve2->ve_offset)
                return (-1);
        if (ve1->ve_offset > ve2->ve_offset)
                return (1);
        return (0);
}

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

        if (ve1->ve_lastused < ve2->ve_lastused)
                return (-1);
        if (ve1->ve_lastused > ve2->ve_lastused)
                return (1);

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

static void
vdev_cache_evict(vdev_cache_t *vc, vdev_cache_entry_t *ve)
{
        ASSERT(MUTEX_HELD(&vc->vc_lock));
        ASSERT(ve->ve_fill_io == NULL);
        ASSERT(ve->ve_data != NULL);

        avl_remove(&vc->vc_lastused_tree, ve);
        avl_remove(&vc->vc_offset_tree, ve);
        zio_buf_free(ve->ve_data, VCBS);
        kmem_free(ve, sizeof (vdev_cache_entry_t));
}

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);
                ASSERT(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_data = zio_buf_alloc(VCBS);

        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));
        ASSERT(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++;
        bcopy(ve->ve_data + cache_phase, zio->io_data, zio->io_size);
}

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;

        ASSERT(fio->io_size == VCBS);

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

        ASSERT(ve->ve_fill_io == fio);
        ASSERT(ve->ve_offset == fio->io_offset);
        ASSERT(ve->ve_data == fio->io_data);

        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.
         */
        while ((pio = zio_walk_parents(fio)) != 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);
}

int
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);
        uint64_t cache_phase = P2PHASE(zio->io_offset, VCBS);
        zio_t *fio;

        ASSERT(zio->io_type == ZIO_TYPE_READ);

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

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

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

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

        mutex_enter(&vc->vc_lock);

        ve_search.ve_offset = cache_offset;
        ve = avl_find(&vc->vc_offset_tree, &ve_search, NULL);

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

                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 (0);
                }

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

                mutex_exit(&vc->vc_lock);
                VDCSTAT_BUMP(vdc_stat_hits);
                return (0);
        }

        ve = vdev_cache_allocate(zio);

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

        fio = zio_vdev_delegated_io(zio->io_vd, cache_offset,
            ve->ve_data, VCBS, ZIO_TYPE_READ, ZIO_PRIORITY_CACHE_FILL,
            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 (0);
}

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;

        ASSERT(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 {
                        bcopy((char *)zio->io_data + start - io_start,
                            ve->ve_data + start - ve->ve_offset, 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;
        }
}