*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
+ * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
*/
#include <sys/zfs_context.h>
+#include <sys/arc.h>
#include <sys/dmu.h>
+#include <sys/dmu_send.h>
#include <sys/dmu_impl.h>
#include <sys/dbuf.h>
#include <sys/dmu_objset.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
+struct dbuf_hold_impl_data {
+ /* Function arguments */
+ dnode_t *dh_dn;
+ uint8_t dh_level;
+ uint64_t dh_blkid;
+ int dh_fail_sparse;
+ void *dh_tag;
+ dmu_buf_impl_t **dh_dbp;
+ /* Local variables */
+ dmu_buf_impl_t *dh_db;
+ dmu_buf_impl_t *dh_parent;
+ blkptr_t *dh_bp;
+ int dh_err;
+ dbuf_dirty_record_t *dh_dr;
+ arc_buf_contents_t dh_type;
+ int dh_depth;
+};
+
+static void __dbuf_hold_impl_init(struct dbuf_hold_impl_data *dh,
+ dnode_t *dn, uint8_t level, uint64_t blkid, int fail_sparse,
+ void *tag, dmu_buf_impl_t **dbp, int depth);
+static int __dbuf_hold_impl(struct dbuf_hold_impl_data *dh);
+
+/*
+ * Number of times that zfs_free_range() took the slow path while doing
+ * a zfs receive. A nonzero value indicates a potential performance problem.
+ */
+uint64_t zfs_free_range_recv_miss;
+
static void dbuf_destroy(dmu_buf_impl_t *db);
-static int dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
+static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx);
/*
mutex_init(&db->db_mtx, NULL, MUTEX_DEFAULT, NULL);
cv_init(&db->db_changed, NULL, CV_DEFAULT, NULL);
refcount_create(&db->db_holds);
+ list_link_init(&db->db_link);
return (0);
}
boolean_t is_metadata;
DB_DNODE_ENTER(db);
- is_metadata = dmu_ot[DB_DNODE(db)->dn_type].ot_metadata;
+ is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type);
DB_DNODE_EXIT(db);
return (is_metadata);
retry:
h->hash_table_mask = hsize - 1;
+#if defined(_KERNEL) && defined(HAVE_SPL)
+ /*
+ * Large allocations which do not require contiguous pages
+ * should be using vmem_alloc() in the linux kernel
+ */
+ h->hash_table = vmem_zalloc(hsize * sizeof (void *), KM_PUSHPAGE);
+#else
h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP);
+#endif
if (h->hash_table == NULL) {
/* XXX - we should really return an error instead of assert */
ASSERT(hsize > (1ULL << 10));
for (i = 0; i < DBUF_MUTEXES; i++)
mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL);
+
+ dbuf_stats_init(h);
}
void
dbuf_hash_table_t *h = &dbuf_hash_table;
int i;
+ dbuf_stats_destroy();
+
for (i = 0; i < DBUF_MUTEXES; i++)
mutex_destroy(&h->hash_mutexes[i]);
+#if defined(_KERNEL) && defined(HAVE_SPL)
+ /*
+ * Large allocations which do not require contiguous pages
+ * should be using vmem_free() in the linux kernel
+ */
+ vmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *));
+#else
kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *));
+#endif
kmem_cache_destroy(dbuf_cache);
}
} else if (db->db_blkid == DMU_SPILL_BLKID) {
ASSERT(dn != NULL);
ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen);
- ASSERT3U(db->db.db_offset, ==, 0);
+ ASSERT0(db->db.db_offset);
} else {
ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size);
}
&dn->dn_phys->dn_blkptr[db->db_blkid]);
} else {
/* db is pointed to by an indirect block */
- int epb = db->db_parent->db.db_size >> SPA_BLKPTRSHIFT;
+ ASSERTV(int epb = db->db_parent->db.db_size >>
+ SPA_BLKPTRSHIFT);
ASSERT3U(db->db_parent->db_level, ==, db->db_level+1);
ASSERT3U(db->db_parent->db.db_object, ==,
db->db.db_object);
* data when we evict this buffer.
*/
if (db->db_dirtycnt == 0) {
- uint64_t *buf = db->db.db_data;
+ ASSERTV(uint64_t *buf = db->db.db_data);
int i;
for (i = 0; i < db->db.db_size >> 3; i++) {
} else {
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
ASSERT3P(db->db_buf, ==, NULL);
- VERIFY(arc_buf_remove_ref(buf, db) == 1);
+ VERIFY(arc_buf_remove_ref(buf, db));
db->db_state = DB_UNCACHED;
}
cv_broadcast(&db->db_changed);
spa_t *spa;
zbookmark_t zb;
uint32_t aflags = ARC_NOWAIT;
- arc_buf_t *pbuf;
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
if (DBUF_IS_L2CACHEABLE(db))
aflags |= ARC_L2CACHE;
+ if (DBUF_IS_L2COMPRESSIBLE(db))
+ aflags |= ARC_L2COMPRESS;
SET_BOOKMARK(&zb, db->db_objset->os_dsl_dataset ?
db->db_objset->os_dsl_dataset->ds_object : DMU_META_OBJSET,
db->db.db_object, db->db_level, db->db_blkid);
dbuf_add_ref(db, NULL);
- /* ZIO_FLAG_CANFAIL callers have to check the parent zio's error */
-
- if (db->db_parent)
- pbuf = db->db_parent->db_buf;
- else
- pbuf = db->db_objset->os_phys_buf;
- (void) dsl_read(zio, spa, db->db_blkptr, pbuf,
+ (void) arc_read(zio, spa, db->db_blkptr,
dbuf_read_done, db, ZIO_PRIORITY_SYNC_READ,
(*flags & DB_RF_CANFAIL) ? ZIO_FLAG_CANFAIL : ZIO_FLAG_MUSTSUCCEED,
&aflags, &zb);
ASSERT(!refcount_is_zero(&db->db_holds));
if (db->db_state == DB_NOFILL)
- return (EIO);
+ return (SET_ERROR(EIO));
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
if (!havepzio)
err = zio_wait(zio);
} else {
+ /*
+ * Another reader came in while the dbuf was in flight
+ * between UNCACHED and CACHED. Either a writer will finish
+ * writing the buffer (sending the dbuf to CACHED) or the
+ * first reader's request will reach the read_done callback
+ * and send the dbuf to CACHED. Otherwise, a failure
+ * occurred and the dbuf went to UNCACHED.
+ */
mutex_exit(&db->db_mtx);
if (prefetch)
dmu_zfetch(&dn->dn_zfetch, db->db.db_offset,
rw_exit(&dn->dn_struct_rwlock);
DB_DNODE_EXIT(db);
+ /* Skip the wait per the caller's request. */
mutex_enter(&db->db_mtx);
if ((flags & DB_RF_NEVERWAIT) == 0) {
while (db->db_state == DB_READ ||
cv_wait(&db->db_changed, &db->db_mtx);
}
if (db->db_state == DB_UNCACHED)
- err = EIO;
+ err = SET_ERROR(EIO);
}
mutex_exit(&db->db_mtx);
}
ASSERT(db->db_data_pending != dr);
/* free this block */
- if (!BP_IS_HOLE(bp)) {
+ if (!BP_IS_HOLE(bp) && !dr->dt.dl.dr_nopwrite) {
spa_t *spa;
DB_GET_SPA(&spa, db);
zio_free(spa, txg, bp);
}
dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
+ dr->dt.dl.dr_nopwrite = B_FALSE;
+
/*
* Release the already-written buffer, so we leave it in
* a consistent dirty state. Note that all callers are
/*
* Evict (if its unreferenced) or clear (if its referenced) any level-0
* data blocks in the free range, so that any future readers will find
- * empty blocks. Also, if we happen accross any level-1 dbufs in the
+ * empty blocks. Also, if we happen across any level-1 dbufs in the
* range that have not already been marked dirty, mark them dirty so
* they stay in memory.
+ *
+ * This is a no-op if the dataset is in the middle of an incremental
+ * receive; see comment below for details.
*/
void
dbuf_free_range(dnode_t *dn, uint64_t start, uint64_t end, dmu_tx_t *tx)
last_l1 = end >> epbs;
}
dprintf_dnode(dn, "start=%llu end=%llu\n", start, end);
+
mutex_enter(&dn->dn_dbufs_mtx);
- for (db = list_head(&dn->dn_dbufs); db; db = db_next) {
+ if (start >= dn->dn_unlisted_l0_blkid * dn->dn_datablksz) {
+ /* There can't be any dbufs in this range; no need to search. */
+ mutex_exit(&dn->dn_dbufs_mtx);
+ return;
+ } else if (dmu_objset_is_receiving(dn->dn_objset)) {
+ /*
+ * If we are receiving, we expect there to be no dbufs in
+ * the range to be freed, because receive modifies each
+ * block at most once, and in offset order. If this is
+ * not the case, it can lead to performance problems,
+ * so note that we unexpectedly took the slow path.
+ */
+ atomic_inc_64(&zfs_free_range_recv_miss);
+ }
+
+ for (db = list_head(&dn->dn_dbufs); db != NULL; db = db_next) {
db_next = list_next(&dn->dn_dbufs, db);
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
continue;
/* found a level 0 buffer in the range */
- if (dbuf_undirty(db, tx))
+ mutex_enter(&db->db_mtx);
+ if (dbuf_undirty(db, tx)) {
+ /* mutex has been dropped and dbuf destroyed */
continue;
+ }
- mutex_enter(&db->db_mtx);
if (db->db_state == DB_UNCACHED ||
db->db_state == DB_NOFILL ||
db->db_state == DB_EVICTING) {
mutex_enter(&db->db_mtx);
dbuf_set_data(db, buf);
- VERIFY(arc_buf_remove_ref(obuf, db) == 1);
+ VERIFY(arc_buf_remove_ref(obuf, db));
db->db.db_size = size;
if (db->db_level == 0) {
dbuf_release_bp(dmu_buf_impl_t *db)
{
objset_t *os;
- zbookmark_t zb;
DB_GET_OBJSET(&os, db);
ASSERT(dsl_pool_sync_context(dmu_objset_pool(os)));
list_link_active(&os->os_dsl_dataset->ds_synced_link));
ASSERT(db->db_parent == NULL || arc_released(db->db_parent->db_buf));
- zb.zb_objset = os->os_dsl_dataset ?
- os->os_dsl_dataset->ds_object : 0;
- zb.zb_object = db->db.db_object;
- zb.zb_level = db->db_level;
- zb.zb_blkid = db->db_blkid;
- (void) arc_release_bp(db->db_buf, db,
- db->db_blkptr, os->os_spa, &zb);
+ (void) arc_release(db->db_buf, db);
}
dbuf_dirty_record_t *
dn->dn_dirtyctx =
(dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN);
ASSERT(dn->dn_dirtyctx_firstset == NULL);
- dn->dn_dirtyctx_firstset = kmem_alloc(1, KM_SLEEP);
+ dn->dn_dirtyctx_firstset = kmem_alloc(1, KM_PUSHPAGE);
}
mutex_exit(&dn->dn_mtx);
* to make a copy of it so that the changes we make in this
* transaction group won't leak out when we sync the older txg.
*/
- dr = kmem_zalloc(sizeof (dbuf_dirty_record_t), KM_SLEEP);
+ dr = kmem_zalloc(sizeof (dbuf_dirty_record_t), KM_PUSHPAGE);
+ list_link_init(&dr->dr_dirty_node);
if (db->db_level == 0) {
void *data_old = db->db_buf;
sizeof (dbuf_dirty_record_t),
offsetof(dbuf_dirty_record_t, dr_dirty_node));
}
+ if (db->db_blkid != DMU_BONUS_BLKID && os->os_dsl_dataset != NULL)
+ dr->dr_accounted = db->db.db_size;
dr->dr_dbuf = db;
dr->dr_txg = tx->tx_txg;
dr->dr_next = *drp;
dbuf_rele(parent, FTAG);
mutex_enter(&db->db_mtx);
- /* possible race with dbuf_undirty() */
+ /*
+ * Since we've dropped the mutex, it's possible that
+ * dbuf_undirty() might have changed this out from under us.
+ */
if (db->db_last_dirty == dr ||
dn->dn_object == DMU_META_DNODE_OBJECT) {
mutex_enter(&di->dt.di.dr_mtx);
return (dr);
}
-static int
+/*
+ * Undirty a buffer in the transaction group referenced by the given
+ * transaction. Return whether this evicted the dbuf.
+ */
+static boolean_t
dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
{
dnode_t *dn;
ASSERT(txg != 0);
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
+ ASSERT0(db->db_level);
+ ASSERT(MUTEX_HELD(&db->db_mtx));
- mutex_enter(&db->db_mtx);
/*
* If this buffer is not dirty, we're done.
*/
for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
if (dr->dr_txg <= txg)
break;
- if (dr == NULL || dr->dr_txg < txg) {
- mutex_exit(&db->db_mtx);
- return (0);
- }
+ if (dr == NULL || dr->dr_txg < txg)
+ return (B_FALSE);
ASSERT(dr->dr_txg == txg);
ASSERT(dr->dr_dbuf == db);
dn = DB_DNODE(db);
/*
- * If this buffer is currently held, we cannot undirty
- * it, since one of the current holders may be in the
- * middle of an update. Note that users of dbuf_undirty()
- * should not place a hold on the dbuf before the call.
+ * Note: This code will probably work even if there are concurrent
+ * holders, but it is untested in that scenerio, as the ZPL and
+ * ztest have additional locking (the range locks) that prevents
+ * that type of concurrent access.
*/
- if (refcount_count(&db->db_holds) > db->db_dirtycnt) {
- mutex_exit(&db->db_mtx);
- /* Make sure we don't toss this buffer at sync phase */
- mutex_enter(&dn->dn_mtx);
- dnode_clear_range(dn, db->db_blkid, 1, tx);
- mutex_exit(&dn->dn_mtx);
- DB_DNODE_EXIT(db);
- return (0);
- }
+ ASSERT3U(refcount_count(&db->db_holds), ==, db->db_dirtycnt);
dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size);
ASSERT(db->db.db_size != 0);
- /* XXX would be nice to fix up dn_towrite_space[] */
+ /*
+ * Any space we accounted for in dp_dirty_* will be cleaned up by
+ * dsl_pool_sync(). This is relatively rare so the discrepancy
+ * is not a big deal.
+ */
*drp = dr->dr_next;
+ /*
+ * Note that there are three places in dbuf_dirty()
+ * where this dirty record may be put on a list.
+ * Make sure to do a list_remove corresponding to
+ * every one of those list_insert calls.
+ */
if (dr->dr_parent) {
mutex_enter(&dr->dr_parent->dt.di.dr_mtx);
list_remove(&dr->dr_parent->dt.di.dr_children, dr);
mutex_exit(&dr->dr_parent->dt.di.dr_mtx);
- } else if (db->db_level+1 == dn->dn_nlevels) {
+ } else if (db->db_blkid == DMU_SPILL_BLKID ||
+ db->db_level+1 == dn->dn_nlevels) {
ASSERT(db->db_blkptr == NULL || db->db_parent == dn->dn_dbuf);
mutex_enter(&dn->dn_mtx);
list_remove(&dn->dn_dirty_records[txg & TXG_MASK], dr);
}
DB_DNODE_EXIT(db);
- if (db->db_level == 0) {
- if (db->db_state != DB_NOFILL) {
- dbuf_unoverride(dr);
+ if (db->db_state != DB_NOFILL) {
+ dbuf_unoverride(dr);
- ASSERT(db->db_buf != NULL);
- ASSERT(dr->dt.dl.dr_data != NULL);
- if (dr->dt.dl.dr_data != db->db_buf)
- VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data,
- db) == 1);
- }
- } else {
ASSERT(db->db_buf != NULL);
- ASSERT(list_head(&dr->dt.di.dr_children) == NULL);
- mutex_destroy(&dr->dt.di.dr_mtx);
- list_destroy(&dr->dt.di.dr_children);
+ ASSERT(dr->dt.dl.dr_data != NULL);
+ if (dr->dt.dl.dr_data != db->db_buf)
+ VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, db));
}
kmem_free(dr, sizeof (dbuf_dirty_record_t));
ASSERT(db->db_state == DB_NOFILL || arc_released(buf));
dbuf_set_data(db, NULL);
- VERIFY(arc_buf_remove_ref(buf, db) == 1);
+ VERIFY(arc_buf_remove_ref(buf, db));
dbuf_evict(db);
- return (1);
+ return (B_TRUE);
}
- mutex_exit(&db->db_mtx);
- return (0);
+ return (B_FALSE);
}
#pragma weak dmu_buf_will_dirty = dbuf_will_dirty
mutex_exit(&db->db_mtx);
(void) dbuf_dirty(db, tx);
bcopy(buf->b_data, db->db.db_data, db->db.db_size);
- VERIFY(arc_buf_remove_ref(buf, db) == 1);
+ VERIFY(arc_buf_remove_ref(buf, db));
xuio_stat_wbuf_copied();
return;
}
arc_release(db->db_buf, db);
}
dr->dt.dl.dr_data = buf;
- VERIFY(arc_buf_remove_ref(db->db_buf, db) == 1);
+ VERIFY(arc_buf_remove_ref(db->db_buf, db));
} else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) {
arc_release(db->db_buf, db);
- VERIFY(arc_buf_remove_ref(db->db_buf, db) == 1);
+ VERIFY(arc_buf_remove_ref(db->db_buf, db));
}
db->db_buf = NULL;
}
/*
* "Clear" the contents of this dbuf. This will mark the dbuf
- * EVICTING and clear *most* of its references. Unfortunetely,
+ * EVICTING and clear *most* of its references. Unfortunately,
* when we are not holding the dn_dbufs_mtx, we can't clear the
* entry in the dn_dbufs list. We have to wait until dbuf_destroy()
* in this case. For callers from the DMU we will usually see:
dbuf_rele(parent, db);
}
-static int
+__attribute__((always_inline))
+static inline int
dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse,
- dmu_buf_impl_t **parentp, blkptr_t **bpp)
+ dmu_buf_impl_t **parentp, blkptr_t **bpp, struct dbuf_hold_impl_data *dh)
{
int nlevels, epbs;
if (level >= nlevels ||
(blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) {
/* the buffer has no parent yet */
- return (ENOENT);
+ return (SET_ERROR(ENOENT));
} else if (level < nlevels-1) {
/* this block is referenced from an indirect block */
- int err = dbuf_hold_impl(dn, level+1,
- blkid >> epbs, fail_sparse, NULL, parentp);
+ int err;
+ if (dh == NULL) {
+ err = dbuf_hold_impl(dn, level+1, blkid >> epbs,
+ fail_sparse, NULL, parentp);
+ } else {
+ __dbuf_hold_impl_init(dh + 1, dn, dh->dh_level + 1,
+ blkid >> epbs, fail_sparse, NULL,
+ parentp, dh->dh_depth + 1);
+ err = __dbuf_hold_impl(dh + 1);
+ }
if (err)
return (err);
err = dbuf_read(*parentp, NULL,
ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
ASSERT(dn->dn_type != DMU_OT_NONE);
- db = kmem_cache_alloc(dbuf_cache, KM_SLEEP);
+ db = kmem_cache_alloc(dbuf_cache, KM_PUSHPAGE);
db->db_objset = os;
db->db.db_object = dn->dn_object;
db->db.db_offset = 0;
} else {
int blocksize =
- db->db_level ? 1<<dn->dn_indblkshift : dn->dn_datablksz;
+ db->db_level ? 1 << dn->dn_indblkshift : dn->dn_datablksz;
db->db.db_size = blocksize;
db->db.db_offset = db->db_blkid * blocksize;
}
return (odb);
}
list_insert_head(&dn->dn_dbufs, db);
+ if (db->db_level == 0 && db->db_blkid >=
+ dn->dn_unlisted_l0_blkid)
+ dn->dn_unlisted_l0_blkid = db->db_blkid + 1;
db->db_state = DB_UNCACHED;
mutex_exit(&dn->dn_dbufs_mtx);
arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER);
}
void
-dbuf_prefetch(dnode_t *dn, uint64_t blkid)
+dbuf_prefetch(dnode_t *dn, uint64_t blkid, zio_priority_t prio)
{
dmu_buf_impl_t *db = NULL;
blkptr_t *bp = NULL;
return;
/* dbuf_find() returns with db_mtx held */
- if (db = dbuf_find(dn, 0, blkid)) {
+ if ((db = dbuf_find(dn, 0, blkid))) {
/*
* This dbuf is already in the cache. We assume that
* it is already CACHED, or else about to be either
return;
}
- if (dbuf_findbp(dn, 0, blkid, TRUE, &db, &bp) == 0) {
+ if (dbuf_findbp(dn, 0, blkid, TRUE, &db, &bp, NULL) == 0) {
if (bp && !BP_IS_HOLE(bp)) {
- int priority = dn->dn_type == DMU_OT_DDT_ZAP ?
- ZIO_PRIORITY_DDT_PREFETCH : ZIO_PRIORITY_ASYNC_READ;
- arc_buf_t *pbuf;
dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
zbookmark_t zb;
SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
dn->dn_object, 0, blkid);
- if (db)
- pbuf = db->db_buf;
- else
- pbuf = dn->dn_objset->os_phys_buf;
-
- (void) dsl_read(NULL, dn->dn_objset->os_spa,
- bp, pbuf, NULL, NULL, priority,
+ (void) arc_read(NULL, dn->dn_objset->os_spa,
+ bp, NULL, NULL, prio,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
&aflags, &zb);
}
}
}
+#define DBUF_HOLD_IMPL_MAX_DEPTH 20
+
/*
* Returns with db_holds incremented, and db_mtx not held.
* Note: dn_struct_rwlock must be held.
*/
-int
-dbuf_hold_impl(dnode_t *dn, uint8_t level, uint64_t blkid, int fail_sparse,
- void *tag, dmu_buf_impl_t **dbp)
+static int
+__dbuf_hold_impl(struct dbuf_hold_impl_data *dh)
{
- dmu_buf_impl_t *db, *parent = NULL;
+ ASSERT3S(dh->dh_depth, <, DBUF_HOLD_IMPL_MAX_DEPTH);
+ dh->dh_parent = NULL;
- ASSERT(blkid != DMU_BONUS_BLKID);
- ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
- ASSERT3U(dn->dn_nlevels, >, level);
+ ASSERT(dh->dh_blkid != DMU_BONUS_BLKID);
+ ASSERT(RW_LOCK_HELD(&dh->dh_dn->dn_struct_rwlock));
+ ASSERT3U(dh->dh_dn->dn_nlevels, >, dh->dh_level);
- *dbp = NULL;
+ *(dh->dh_dbp) = NULL;
top:
/* dbuf_find() returns with db_mtx held */
- db = dbuf_find(dn, level, blkid);
-
- if (db == NULL) {
- blkptr_t *bp = NULL;
- int err;
-
- ASSERT3P(parent, ==, NULL);
- err = dbuf_findbp(dn, level, blkid, fail_sparse, &parent, &bp);
- if (fail_sparse) {
- if (err == 0 && bp && BP_IS_HOLE(bp))
- err = ENOENT;
- if (err) {
- if (parent)
- dbuf_rele(parent, NULL);
- return (err);
+ dh->dh_db = dbuf_find(dh->dh_dn, dh->dh_level, dh->dh_blkid);
+
+ if (dh->dh_db == NULL) {
+ dh->dh_bp = NULL;
+
+ ASSERT3P(dh->dh_parent, ==, NULL);
+ dh->dh_err = dbuf_findbp(dh->dh_dn, dh->dh_level, dh->dh_blkid,
+ dh->dh_fail_sparse, &dh->dh_parent,
+ &dh->dh_bp, dh);
+ if (dh->dh_fail_sparse) {
+ if (dh->dh_err == 0 &&
+ dh->dh_bp && BP_IS_HOLE(dh->dh_bp))
+ dh->dh_err = SET_ERROR(ENOENT);
+ if (dh->dh_err) {
+ if (dh->dh_parent)
+ dbuf_rele(dh->dh_parent, NULL);
+ return (dh->dh_err);
}
}
- if (err && err != ENOENT)
- return (err);
- db = dbuf_create(dn, level, blkid, parent, bp);
- }
-
- if (db->db_buf && refcount_is_zero(&db->db_holds)) {
- arc_buf_add_ref(db->db_buf, db);
- if (db->db_buf->b_data == NULL) {
- dbuf_clear(db);
- if (parent) {
- dbuf_rele(parent, NULL);
- parent = NULL;
+ if (dh->dh_err && dh->dh_err != ENOENT)
+ return (dh->dh_err);
+ dh->dh_db = dbuf_create(dh->dh_dn, dh->dh_level, dh->dh_blkid,
+ dh->dh_parent, dh->dh_bp);
+ }
+
+ if (dh->dh_db->db_buf && refcount_is_zero(&dh->dh_db->db_holds)) {
+ arc_buf_add_ref(dh->dh_db->db_buf, dh->dh_db);
+ if (dh->dh_db->db_buf->b_data == NULL) {
+ dbuf_clear(dh->dh_db);
+ if (dh->dh_parent) {
+ dbuf_rele(dh->dh_parent, NULL);
+ dh->dh_parent = NULL;
}
goto top;
}
- ASSERT3P(db->db.db_data, ==, db->db_buf->b_data);
+ ASSERT3P(dh->dh_db->db.db_data, ==, dh->dh_db->db_buf->b_data);
}
- ASSERT(db->db_buf == NULL || arc_referenced(db->db_buf));
+ ASSERT(dh->dh_db->db_buf == NULL || arc_referenced(dh->dh_db->db_buf));
/*
* If this buffer is currently syncing out, and we are are
* still referencing it from db_data, we need to make a copy
* of it in case we decide we want to dirty it again in this txg.
*/
- if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID &&
- dn->dn_object != DMU_META_DNODE_OBJECT &&
- db->db_state == DB_CACHED && db->db_data_pending) {
- dbuf_dirty_record_t *dr = db->db_data_pending;
-
- if (dr->dt.dl.dr_data == db->db_buf) {
- arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
-
- dbuf_set_data(db,
- arc_buf_alloc(dn->dn_objset->os_spa,
- db->db.db_size, db, type));
- bcopy(dr->dt.dl.dr_data->b_data, db->db.db_data,
- db->db.db_size);
+ if (dh->dh_db->db_level == 0 &&
+ dh->dh_db->db_blkid != DMU_BONUS_BLKID &&
+ dh->dh_dn->dn_object != DMU_META_DNODE_OBJECT &&
+ dh->dh_db->db_state == DB_CACHED && dh->dh_db->db_data_pending) {
+ dh->dh_dr = dh->dh_db->db_data_pending;
+
+ if (dh->dh_dr->dt.dl.dr_data == dh->dh_db->db_buf) {
+ dh->dh_type = DBUF_GET_BUFC_TYPE(dh->dh_db);
+
+ dbuf_set_data(dh->dh_db,
+ arc_buf_alloc(dh->dh_dn->dn_objset->os_spa,
+ dh->dh_db->db.db_size, dh->dh_db, dh->dh_type));
+ bcopy(dh->dh_dr->dt.dl.dr_data->b_data,
+ dh->dh_db->db.db_data, dh->dh_db->db.db_size);
}
}
- (void) refcount_add(&db->db_holds, tag);
- dbuf_update_data(db);
- DBUF_VERIFY(db);
- mutex_exit(&db->db_mtx);
+ (void) refcount_add(&dh->dh_db->db_holds, dh->dh_tag);
+ dbuf_update_data(dh->dh_db);
+ DBUF_VERIFY(dh->dh_db);
+ mutex_exit(&dh->dh_db->db_mtx);
/* NOTE: we can't rele the parent until after we drop the db_mtx */
- if (parent)
- dbuf_rele(parent, NULL);
+ if (dh->dh_parent)
+ dbuf_rele(dh->dh_parent, NULL);
- ASSERT3P(DB_DNODE(db), ==, dn);
- ASSERT3U(db->db_blkid, ==, blkid);
- ASSERT3U(db->db_level, ==, level);
- *dbp = db;
+ ASSERT3P(DB_DNODE(dh->dh_db), ==, dh->dh_dn);
+ ASSERT3U(dh->dh_db->db_blkid, ==, dh->dh_blkid);
+ ASSERT3U(dh->dh_db->db_level, ==, dh->dh_level);
+ *(dh->dh_dbp) = dh->dh_db;
return (0);
}
+/*
+ * The following code preserves the recursive function dbuf_hold_impl()
+ * but moves the local variables AND function arguments to the heap to
+ * minimize the stack frame size. Enough space is initially allocated
+ * on the stack for 20 levels of recursion.
+ */
+int
+dbuf_hold_impl(dnode_t *dn, uint8_t level, uint64_t blkid, int fail_sparse,
+ void *tag, dmu_buf_impl_t **dbp)
+{
+ struct dbuf_hold_impl_data *dh;
+ int error;
+
+ dh = kmem_zalloc(sizeof (struct dbuf_hold_impl_data) *
+ DBUF_HOLD_IMPL_MAX_DEPTH, KM_PUSHPAGE);
+ __dbuf_hold_impl_init(dh, dn, level, blkid, fail_sparse, tag, dbp, 0);
+
+ error = __dbuf_hold_impl(dh);
+
+ kmem_free(dh, sizeof (struct dbuf_hold_impl_data) *
+ DBUF_HOLD_IMPL_MAX_DEPTH);
+
+ return (error);
+}
+
+static void
+__dbuf_hold_impl_init(struct dbuf_hold_impl_data *dh,
+ dnode_t *dn, uint8_t level, uint64_t blkid, int fail_sparse,
+ void *tag, dmu_buf_impl_t **dbp, int depth)
+{
+ dh->dh_dn = dn;
+ dh->dh_level = level;
+ dh->dh_blkid = blkid;
+ dh->dh_fail_sparse = fail_sparse;
+ dh->dh_tag = tag;
+ dh->dh_dbp = dbp;
+ dh->dh_depth = depth;
+}
+
dmu_buf_impl_t *
dbuf_hold(dnode_t *dn, uint64_t blkid, void *tag)
{
dnode_t *dn;
if (db->db_blkid != DMU_SPILL_BLKID)
- return (ENOTSUP);
+ return (SET_ERROR(ENOTSUP));
if (blksz == 0)
blksz = SPA_MINBLOCKSIZE;
if (blksz > SPA_MAXBLOCKSIZE)
void
dbuf_add_ref(dmu_buf_impl_t *db, void *tag)
{
- int64_t holds = refcount_add(&db->db_holds, tag);
- ASSERT(holds > 1);
+ VERIFY(refcount_add(&db->db_holds, tag) > 1);
}
/*
* This dbuf has anonymous data associated with it.
*/
dbuf_set_data(db, NULL);
- VERIFY(arc_buf_remove_ref(buf, db) == 1);
+ VERIFY(arc_buf_remove_ref(buf, db));
dbuf_evict(db);
} else {
- VERIFY(arc_buf_remove_ref(db->db_buf, db) == 0);
- if (!DBUF_IS_CACHEABLE(db))
+ VERIFY(!arc_buf_remove_ref(db->db_buf, db));
+
+ /*
+ * A dbuf will be eligible for eviction if either the
+ * 'primarycache' property is set or a duplicate
+ * copy of this buffer is already cached in the arc.
+ *
+ * In the case of the 'primarycache' a buffer
+ * is considered for eviction if it matches the
+ * criteria set in the property.
+ *
+ * To decide if our buffer is considered a
+ * duplicate, we must call into the arc to determine
+ * if multiple buffers are referencing the same
+ * block on-disk. If so, then we simply evict
+ * ourselves.
+ */
+ if (!DBUF_IS_CACHEABLE(db) ||
+ arc_buf_eviction_needed(db->db_buf))
dbuf_clear(db);
else
mutex_exit(&db->db_mtx);
return (res);
}
+blkptr_t *
+dmu_buf_get_blkptr(dmu_buf_t *db)
+{
+ dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
+ return (dbi->db_blkptr);
+}
+
static void
dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db)
{
}
}
-static void
+/*
+ * dbuf_sync_indirect() is called recursively from dbuf_sync_list() so it
+ * is critical the we not allow the compiler to inline this function in to
+ * dbuf_sync_list() thereby drastically bloating the stack usage.
+ */
+noinline static void
dbuf_sync_indirect(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
ASSERT(db->db_level > 0);
DBUF_VERIFY(db);
+ /* Read the block if it hasn't been read yet. */
if (db->db_buf == NULL) {
mutex_exit(&db->db_mtx);
(void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED);
DB_DNODE_ENTER(db);
dn = DB_DNODE(db);
+ /* Indirect block size must match what the dnode thinks it is. */
ASSERT3U(db->db.db_size, ==, 1<<dn->dn_phys->dn_indblkshift);
dbuf_check_blkptr(dn, db);
DB_DNODE_EXIT(db);
+ /* Provide the pending dirty record to child dbufs */
db->db_data_pending = dr;
mutex_exit(&db->db_mtx);
zio_nowait(zio);
}
-static void
+/*
+ * dbuf_sync_leaf() is called recursively from dbuf_sync_list() so it is
+ * critical the we not allow the compiler to inline this function in to
+ * dbuf_sync_list() thereby drastically bloating the stack usage.
+ */
+noinline static void
dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
{
arc_buf_t **datap = &dr->dt.dl.dr_data;
dbuf_dirty_record_t **drp;
ASSERT(*datap != NULL);
- ASSERT3U(db->db_level, ==, 0);
+ ASSERT0(db->db_level);
ASSERT3U(dn->dn_phys->dn_bonuslen, <=, DN_MAX_BONUSLEN);
bcopy(*datap, DN_BONUS(dn->dn_phys), dn->dn_phys->dn_bonuslen);
DB_DNODE_EXIT(db);
ASSERT(dr->dr_next == NULL);
ASSERT(dr->dr_dbuf == db);
*drp = dr->dr_next;
+ if (dr->dr_dbuf->db_level != 0) {
+ mutex_destroy(&dr->dt.di.dr_mtx);
+ list_destroy(&dr->dt.di.dr_children);
+ }
kmem_free(dr, sizeof (dbuf_dirty_record_t));
ASSERT(db->db_dirtycnt > 0);
db->db_dirtycnt -= 1;
{
dbuf_dirty_record_t *dr;
- while (dr = list_head(list)) {
+ while ((dr = list_head(list))) {
if (dr->dr_zio != NULL) {
/*
* If we find an already initialized zio then we
mutex_exit(&db->db_mtx);
}
+/*
+ * The SPA will call this callback several times for each zio - once
+ * for every physical child i/o (zio->io_phys_children times). This
+ * allows the DMU to monitor the progress of each logical i/o. For example,
+ * there may be 2 copies of an indirect block, or many fragments of a RAID-Z
+ * block. There may be a long delay before all copies/fragments are completed,
+ * so this callback allows us to retire dirty space gradually, as the physical
+ * i/os complete.
+ */
+/* ARGSUSED */
+static void
+dbuf_write_physdone(zio_t *zio, arc_buf_t *buf, void *arg)
+{
+ dmu_buf_impl_t *db = arg;
+ objset_t *os = db->db_objset;
+ dsl_pool_t *dp = dmu_objset_pool(os);
+ dbuf_dirty_record_t *dr;
+ int delta = 0;
+
+ dr = db->db_data_pending;
+ ASSERT3U(dr->dr_txg, ==, zio->io_txg);
+
+ /*
+ * The callback will be called io_phys_children times. Retire one
+ * portion of our dirty space each time we are called. Any rounding
+ * error will be cleaned up by dsl_pool_sync()'s call to
+ * dsl_pool_undirty_space().
+ */
+ delta = dr->dr_accounted / zio->io_phys_children;
+ dsl_pool_undirty_space(dp, delta, zio->io_txg);
+}
+
/* ARGSUSED */
static void
dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb)
uint64_t txg = zio->io_txg;
dbuf_dirty_record_t **drp, *dr;
- ASSERT3U(zio->io_error, ==, 0);
+ ASSERT0(zio->io_error);
ASSERT(db->db_blkptr == bp);
- if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
+ /*
+ * For nopwrites and rewrites we ensure that the bp matches our
+ * original and bypass all the accounting.
+ */
+ if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) {
ASSERT(BP_EQUAL(bp, bp_orig));
} else {
objset_t *os;
if (db->db_state != DB_NOFILL) {
if (dr->dt.dl.dr_data != db->db_buf)
VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data,
- db) == 1);
+ db));
else if (!arc_released(db->db_buf))
arc_set_callback(db->db_buf, dbuf_do_evict, db);
}
ASSERT(list_head(&dr->dt.di.dr_children) == NULL);
ASSERT3U(db->db.db_size, ==, 1<<dn->dn_phys->dn_indblkshift);
if (!BP_IS_HOLE(db->db_blkptr)) {
- int epbs =
- dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT;
+ ASSERTV(int epbs = dn->dn_phys->dn_indblkshift -
+ SPA_BLKPTRSHIFT);
ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==,
db->db.db_size);
ASSERT3U(dn->dn_phys->dn_maxblkid
ASSERT(db->db_dirtycnt > 0);
db->db_dirtycnt -= 1;
db->db_data_pending = NULL;
+
dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg);
}
dbuf_write_done(zio, NULL, db);
}
+/* Issue I/O to commit a dirty buffer to disk. */
static void
dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx)
{
}
if (parent != dn->dn_dbuf) {
+ /* Our parent is an indirect block. */
+ /* We have a dirty parent that has been scheduled for write. */
ASSERT(parent && parent->db_data_pending);
+ /* Our parent's buffer is one level closer to the dnode. */
ASSERT(db->db_level == parent->db_level-1);
+ /*
+ * We're about to modify our parent's db_data by modifying
+ * our block pointer, so the parent must be released.
+ */
ASSERT(arc_released(parent->db_buf));
zio = parent->db_data_pending->dr_zio;
} else {
+ /* Our parent is the dnode itself. */
ASSERT((db->db_level == dn->dn_phys->dn_nlevels-1 &&
db->db_blkid != DMU_SPILL_BLKID) ||
(db->db_blkid == DMU_SPILL_BLKID && db->db_level == 0));
ASSERT(db->db_state != DB_NOFILL);
dr->dr_zio = zio_write(zio, os->os_spa, txg,
db->db_blkptr, data->b_data, arc_buf_size(data), &zp,
- dbuf_write_override_ready, dbuf_write_override_done, dr,
- ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
+ dbuf_write_override_ready, NULL, dbuf_write_override_done,
+ dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
mutex_enter(&db->db_mtx);
dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
zio_write_override(dr->dr_zio, &dr->dt.dl.dr_overridden_by,
- dr->dt.dl.dr_copies);
+ dr->dt.dl.dr_copies, dr->dt.dl.dr_nopwrite);
mutex_exit(&db->db_mtx);
} else if (db->db_state == DB_NOFILL) {
ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF);
dr->dr_zio = zio_write(zio, os->os_spa, txg,
db->db_blkptr, NULL, db->db.db_size, &zp,
- dbuf_write_nofill_ready, dbuf_write_nofill_done, db,
+ dbuf_write_nofill_ready, NULL, dbuf_write_nofill_done, db,
ZIO_PRIORITY_ASYNC_WRITE,
ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb);
} else {
ASSERT(arc_released(data));
dr->dr_zio = arc_write(zio, os->os_spa, txg,
- db->db_blkptr, data, DBUF_IS_L2CACHEABLE(db), &zp,
- dbuf_write_ready, dbuf_write_done, db,
+ db->db_blkptr, data, DBUF_IS_L2CACHEABLE(db),
+ DBUF_IS_L2COMPRESSIBLE(db), &zp, dbuf_write_ready,
+ dbuf_write_physdone, dbuf_write_done, db,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
}
}
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+EXPORT_SYMBOL(dbuf_find);
+EXPORT_SYMBOL(dbuf_is_metadata);
+EXPORT_SYMBOL(dbuf_evict);
+EXPORT_SYMBOL(dbuf_loan_arcbuf);
+EXPORT_SYMBOL(dbuf_whichblock);
+EXPORT_SYMBOL(dbuf_read);
+EXPORT_SYMBOL(dbuf_unoverride);
+EXPORT_SYMBOL(dbuf_free_range);
+EXPORT_SYMBOL(dbuf_new_size);
+EXPORT_SYMBOL(dbuf_release_bp);
+EXPORT_SYMBOL(dbuf_dirty);
+EXPORT_SYMBOL(dmu_buf_will_dirty);
+EXPORT_SYMBOL(dmu_buf_will_not_fill);
+EXPORT_SYMBOL(dmu_buf_will_fill);
+EXPORT_SYMBOL(dmu_buf_fill_done);
+EXPORT_SYMBOL(dmu_buf_rele);
+EXPORT_SYMBOL(dbuf_assign_arcbuf);
+EXPORT_SYMBOL(dbuf_clear);
+EXPORT_SYMBOL(dbuf_prefetch);
+EXPORT_SYMBOL(dbuf_hold_impl);
+EXPORT_SYMBOL(dbuf_hold);
+EXPORT_SYMBOL(dbuf_hold_level);
+EXPORT_SYMBOL(dbuf_create_bonus);
+EXPORT_SYMBOL(dbuf_spill_set_blksz);
+EXPORT_SYMBOL(dbuf_rm_spill);
+EXPORT_SYMBOL(dbuf_add_ref);
+EXPORT_SYMBOL(dbuf_rele);
+EXPORT_SYMBOL(dbuf_rele_and_unlock);
+EXPORT_SYMBOL(dbuf_refcount);
+EXPORT_SYMBOL(dbuf_sync_list);
+EXPORT_SYMBOL(dmu_buf_set_user);
+EXPORT_SYMBOL(dmu_buf_set_user_ie);
+EXPORT_SYMBOL(dmu_buf_update_user);
+EXPORT_SYMBOL(dmu_buf_get_user);
+EXPORT_SYMBOL(dmu_buf_freeable);
+#endif