/*
* 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;
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);
}
/*
* Fill a previously allocated cache entry with data.
*/
static void
vdev_cache_fill(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
vdev_cache_t *vc = &vd->vdev_cache;
vdev_cache_entry_t *ve = zio->io_private;
zio_t *dio;
ASSERT(zio->io_size == VCBS);
/*
* Add data to the cache.
*/
mutex_enter(&vc->vc_lock);
ASSERT(ve->ve_fill_io == zio);
ASSERT(ve->ve_offset == zio->io_offset);
ASSERT(ve->ve_data == zio->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.
*/
for (dio = zio->io_delegate_list; dio; dio = dio->io_delegate_next)
vdev_cache_hit(vc, ve, dio);
if (zio->io_error || ve->ve_missed_update)
vdev_cache_evict(vc, ve);
mutex_exit(&vc->vc_lock);
while ((dio = zio->io_delegate_list) != NULL) {
zio->io_delegate_list = dio->io_delegate_next;
dio->io_delegate_next = NULL;
dio->io_error = zio->io_error;
zio_next_stage(dio);
}
}
/*
* Read data from the cache. Returns 0 on cache hit, errno on a miss.
*/
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);
}
/*
* Read data from the cache. Returns 0 on cache hit, errno on a miss.
*/
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 (P2CROSS(zio->io_offset, zio->io_offset + zio->io_size - 1, 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->io_delegate_next = fio->io_delegate_list;
fio->io_delegate_list = zio;
zio_vdev_io_bypass(zio);
mutex_exit(&vc->vc_lock);
return (0);
}
vdev_cache_hit(vc, ve, zio);
zio_vdev_io_bypass(zio);
mutex_exit(&vc->vc_lock);
zio_next_stage(zio);
return (0);
}
if (!(zio->io_flags & ZIO_FLAG_METADATA)) {
mutex_exit(&vc->vc_lock);
return (EINVAL);
}
ve = vdev_cache_allocate(zio);
if (ve == NULL) {
mutex_exit(&vc->vc_lock);
return (ENOMEM);
}
fio = zio_vdev_child_io(zio, NULL, zio->io_vd, cache_offset,
ve->ve_data, VCBS, ZIO_TYPE_READ, ZIO_PRIORITY_CACHE_FILL,
ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_PROPAGATE |
ZIO_FLAG_DONT_RETRY | ZIO_FLAG_NOBOOKMARK,
vdev_cache_fill, ve);
ve->ve_fill_io = fio;
fio->io_delegate_list = zio;
zio_vdev_io_bypass(zio);
mutex_exit(&vc->vc_lock);
zio_nowait(fio);
return (0);
}