*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
+ * Copyright (c) 2012 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_impl.h>
#include <sys/dbuf.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);
+
static void dbuf_destroy(dmu_buf_impl_t *db);
static int 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);
}
{
dbuf_hash_table_t *h = &dbuf_hash_table;
objset_t *os = dn->dn_objset;
- uint64_t obj = dn->dn_object;
- uint64_t hv = DBUF_HASH(os, obj, level, blkid);
- uint64_t idx = hv & h->hash_table_mask;
+ uint64_t obj;
+ uint64_t hv;
+ uint64_t idx;
dmu_buf_impl_t *db;
+ obj = dn->dn_object;
+ hv = DBUF_HASH(os, obj, level, blkid);
+ idx = hv & h->hash_table_mask;
+
mutex_enter(DBUF_HASH_MUTEX(h, idx));
for (db = h->hash_table[idx]; db != NULL; db = db->db_hash_next) {
if (DBUF_EQUAL(db, os, obj, level, blkid)) {
objset_t *os = db->db_objset;
uint64_t obj = db->db.db_object;
int level = db->db_level;
- uint64_t blkid = db->db_blkid;
- uint64_t hv = DBUF_HASH(os, obj, level, blkid);
- uint64_t idx = hv & h->hash_table_mask;
+ uint64_t blkid, hv, idx;
dmu_buf_impl_t *dbf;
+ blkid = db->db_blkid;
+ hv = DBUF_HASH(os, obj, level, blkid);
+ idx = hv & h->hash_table_mask;
+
mutex_enter(DBUF_HASH_MUTEX(h, idx));
for (dbf = h->hash_table[idx]; dbf != NULL; dbf = dbf->db_hash_next) {
if (DBUF_EQUAL(dbf, os, obj, level, blkid)) {
dbuf_hash_remove(dmu_buf_impl_t *db)
{
dbuf_hash_table_t *h = &dbuf_hash_table;
- uint64_t hv = DBUF_HASH(db->db_objset, db->db.db_object,
- db->db_level, db->db_blkid);
- uint64_t idx = hv & h->hash_table_mask;
+ uint64_t hv, idx;
dmu_buf_impl_t *dbf, **dbp;
+ hv = DBUF_HASH(db->db_objset, db->db.db_object,
+ db->db_level, db->db_blkid);
+ idx = hv & h->hash_table_mask;
+
/*
* We musn't hold db_mtx to maintin lock ordering:
* DBUF_HASH_MUTEX > db_mtx.
db->db_evict_func = NULL;
}
+boolean_t
+dbuf_is_metadata(dmu_buf_impl_t *db)
+{
+ if (db->db_level > 0) {
+ return (B_TRUE);
+ } else {
+ boolean_t is_metadata;
+
+ DB_DNODE_ENTER(db);
+ is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type);
+ DB_DNODE_EXIT(db);
+
+ return (is_metadata);
+ }
+}
+
void
dbuf_evict(dmu_buf_impl_t *db)
{
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_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);
}
static void
dbuf_verify(dmu_buf_impl_t *db)
{
- dnode_t *dn = db->db_dnode;
+ dnode_t *dn;
dbuf_dirty_record_t *dr;
ASSERT(MUTEX_HELD(&db->db_mtx));
return;
ASSERT(db->db_objset != NULL);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
if (dn == NULL) {
ASSERT(db->db_parent == NULL);
ASSERT(db->db_blkptr == NULL);
ASSERT3U(db->db.db_object, ==, dn->dn_object);
ASSERT3P(db->db_objset, ==, dn->dn_objset);
ASSERT3U(db->db_level, <, dn->dn_nlevels);
- ASSERT(db->db_blkid == DMU_BONUS_BLKID || db->db_blkid ==
- DMU_SPILL_BLKID || list_head(&dn->dn_dbufs));
+ ASSERT(db->db_blkid == DMU_BONUS_BLKID ||
+ db->db_blkid == DMU_SPILL_BLKID ||
+ !list_is_empty(&dn->dn_dbufs));
}
if (db->db_blkid == DMU_BONUS_BLKID) {
ASSERT(dn != NULL);
} 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);
* have the struct_rwlock. XXX indblksz no longer
* grows. safe to do this now?
*/
- if (RW_WRITE_HELD(&db->db_dnode->dn_struct_rwlock)) {
+ if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) {
ASSERT3P(db->db_blkptr, ==,
((blkptr_t *)db->db_parent->db.db_data +
db->db_blkid % epb));
* 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++) {
}
}
}
+ DB_DNODE_EXIT(db);
}
#endif
mutex_enter(&db->db_mtx);
if (arc_released(db->db_buf) || refcount_count(&db->db_holds) > 1) {
int blksz = db->db.db_size;
+ spa_t *spa;
+
mutex_exit(&db->db_mtx);
- abuf = arc_loan_buf(db->db_dnode->dn_objset->os_spa, blksz);
+ DB_GET_SPA(&spa, db);
+ abuf = arc_loan_buf(spa, blksz);
bcopy(db->db.db_data, abuf->b_data, blksz);
} else {
abuf = db->db_buf;
static void
dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t *flags)
{
- dnode_t *dn = db->db_dnode;
+ dnode_t *dn;
+ spa_t *spa;
zbookmark_t zb;
uint32_t aflags = ARC_NOWAIT;
- arc_buf_t *pbuf;
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
ASSERT(!refcount_is_zero(&db->db_holds));
/* We need the struct_rwlock to prevent db_blkptr from changing. */
ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock));
bzero(db->db.db_data, DN_MAX_BONUSLEN);
if (bonuslen)
bcopy(DN_BONUS(dn->dn_phys), db->db.db_data, bonuslen);
+ DB_DNODE_EXIT(db);
dbuf_update_data(db);
db->db_state = DB_CACHED;
mutex_exit(&db->db_mtx);
dbuf_set_data(db, arc_buf_alloc(dn->dn_objset->os_spa,
db->db.db_size, db, type));
+ DB_DNODE_EXIT(db);
bzero(db->db.db_data, db->db.db_size);
db->db_state = DB_CACHED;
*flags |= DB_RF_CACHED;
return;
}
+ spa = dn->dn_objset->os_spa;
+ DB_DNODE_EXIT(db);
+
db->db_state = DB_READ;
mutex_exit(&db->db_mtx);
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, dn->dn_objset->os_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);
int err = 0;
int havepzio = (zio != NULL);
int prefetch;
+ dnode_t *dn;
/*
* We don't have to hold the mutex to check db_state because it
if (db->db_state == DB_NOFILL)
return (EIO);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
if ((flags & DB_RF_HAVESTRUCT) == 0)
- rw_enter(&db->db_dnode->dn_struct_rwlock, RW_READER);
+ rw_enter(&dn->dn_struct_rwlock, RW_READER);
prefetch = db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID &&
- (flags & DB_RF_NOPREFETCH) == 0 && db->db_dnode != NULL &&
+ (flags & DB_RF_NOPREFETCH) == 0 && dn != NULL &&
DBUF_IS_CACHEABLE(db);
mutex_enter(&db->db_mtx);
if (db->db_state == DB_CACHED) {
mutex_exit(&db->db_mtx);
if (prefetch)
- dmu_zfetch(&db->db_dnode->dn_zfetch, db->db.db_offset,
+ dmu_zfetch(&dn->dn_zfetch, db->db.db_offset,
db->db.db_size, TRUE);
if ((flags & DB_RF_HAVESTRUCT) == 0)
- rw_exit(&db->db_dnode->dn_struct_rwlock);
+ rw_exit(&dn->dn_struct_rwlock);
+ DB_DNODE_EXIT(db);
} else if (db->db_state == DB_UNCACHED) {
- if (zio == NULL) {
- zio = zio_root(db->db_dnode->dn_objset->os_spa,
- NULL, NULL, ZIO_FLAG_CANFAIL);
- }
+ spa_t *spa = dn->dn_objset->os_spa;
+
+ if (zio == NULL)
+ zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
dbuf_read_impl(db, zio, &flags);
/* dbuf_read_impl has dropped db_mtx for us */
if (prefetch)
- dmu_zfetch(&db->db_dnode->dn_zfetch, db->db.db_offset,
+ dmu_zfetch(&dn->dn_zfetch, db->db.db_offset,
db->db.db_size, flags & DB_RF_CACHED);
if ((flags & DB_RF_HAVESTRUCT) == 0)
- rw_exit(&db->db_dnode->dn_struct_rwlock);
+ rw_exit(&dn->dn_struct_rwlock);
+ DB_DNODE_EXIT(db);
if (!havepzio)
err = zio_wait(zio);
} else {
mutex_exit(&db->db_mtx);
if (prefetch)
- dmu_zfetch(&db->db_dnode->dn_zfetch, db->db.db_offset,
+ dmu_zfetch(&dn->dn_zfetch, db->db.db_offset,
db->db.db_size, TRUE);
if ((flags & DB_RF_HAVESTRUCT) == 0)
- rw_exit(&db->db_dnode->dn_struct_rwlock);
+ rw_exit(&dn->dn_struct_rwlock);
+ DB_DNODE_EXIT(db);
mutex_enter(&db->db_mtx);
if ((flags & DB_RF_NEVERWAIT) == 0) {
cv_wait(&db->db_changed, &db->db_mtx);
if (db->db_state == DB_UNCACHED) {
arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
+ spa_t *spa;
ASSERT(db->db_buf == NULL);
ASSERT(db->db.db_data == NULL);
- dbuf_set_data(db, arc_buf_alloc(db->db_dnode->dn_objset->os_spa,
- db->db.db_size, db, type));
+ DB_GET_SPA(&spa, db);
+ dbuf_set_data(db, arc_buf_alloc(spa, db->db.db_size, db, type));
db->db_state = DB_FILL;
} else if (db->db_state == DB_NOFILL) {
dbuf_set_data(db, NULL);
/*
* If the last dirty record for this dbuf has not yet synced
* and its referencing the dbuf data, either:
- * reset the reference to point to a new copy,
+ * reset the reference to point to a new copy,
* or (if there a no active holders)
* just null out the current db_data pointer.
*/
} else if (refcount_count(&db->db_holds) > db->db_dirtycnt) {
int size = db->db.db_size;
arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
- dr->dt.dl.dr_data = arc_buf_alloc(
- db->db_dnode->dn_objset->os_spa, size, db, type);
+ spa_t *spa;
+
+ DB_GET_SPA(&spa, db);
+ dr->dt.dl.dr_data = arc_buf_alloc(spa, size, db, type);
bcopy(db->db.db_data, dr->dt.dl.dr_data->b_data, size);
} else {
dbuf_set_data(db, NULL);
ASSERT(db->db_data_pending != dr);
/* free this block */
- if (!BP_IS_HOLE(bp))
- zio_free(db->db_dnode->dn_objset->os_spa, txg, bp);
+ if (!BP_IS_HOLE(bp)) {
+ spa_t *spa;
+ DB_GET_SPA(&spa, db);
+ zio_free(spa, txg, bp);
+ }
dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
/*
* Release the already-written buffer, so we leave it in
else if (db->db_blkptr)
birth_txg = db->db_blkptr->blk_birth;
- /* If we don't exist or are in a snapshot, we can't be freed */
+ /*
+ * If we don't exist or are in a snapshot, we can't be freed.
+ * Don't pass the bp to dsl_dataset_block_freeable() since we
+ * are holding the db_mtx lock and might deadlock if we are
+ * prefetching a dedup-ed block.
+ */
if (birth_txg)
return (ds == NULL ||
- dsl_dataset_block_freeable(ds, db->db_blkptr, birth_txg));
+ dsl_dataset_block_freeable(ds, NULL, birth_txg));
else
return (FALSE);
}
arc_buf_t *buf, *obuf;
int osize = db->db.db_size;
arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
+ dnode_t *dn;
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
+
/* XXX does *this* func really need the lock? */
- ASSERT(RW_WRITE_HELD(&db->db_dnode->dn_struct_rwlock));
+ ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock));
/*
* This call to dbuf_will_dirty() with the dn_struct_rwlock held
dbuf_will_dirty(db, tx);
/* create the data buffer for the new block */
- buf = arc_buf_alloc(db->db_dnode->dn_objset->os_spa, size, db, type);
+ buf = arc_buf_alloc(dn->dn_objset->os_spa, size, db, type);
/* copy old block data to the new block */
obuf = db->db_buf;
}
mutex_exit(&db->db_mtx);
- dnode_willuse_space(db->db_dnode, size-osize, tx);
+ dnode_willuse_space(dn, size-osize, tx);
+ DB_DNODE_EXIT(db);
}
void
dbuf_release_bp(dmu_buf_impl_t *db)
{
- objset_t *os = db->db_dnode->dn_objset;
- zbookmark_t zb;
+ objset_t *os;
+ DB_GET_OBJSET(&os, db);
ASSERT(dsl_pool_sync_context(dmu_objset_pool(os)));
ASSERT(arc_released(os->os_phys_buf) ||
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 *
dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
{
- dnode_t *dn = db->db_dnode;
- objset_t *os = dn->dn_objset;
+ dnode_t *dn;
+ objset_t *os;
dbuf_dirty_record_t **drp, *dr;
int drop_struct_lock = FALSE;
boolean_t do_free_accounting = B_FALSE;
ASSERT(!refcount_is_zero(&db->db_holds));
DMU_TX_DIRTY_BUF(tx, db);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
/*
* Shouldn't dirty a regular buffer in syncing context. Private
* objects may be dirtied in syncing context, but only if they
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);
while ((dr = *drp) != NULL && dr->dr_txg > tx->tx_txg)
drp = &dr->dr_next;
if (dr && dr->dr_txg == tx->tx_txg) {
+ DB_DNODE_EXIT(db);
+
if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID) {
/*
* If this buffer has already been written out,
* we already dirtied it in open context. Hence we must make
* this assertion only if we're not already dirty.
*/
+ os = dn->dn_objset;
ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) ||
os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp));
ASSERT(db->db.db_size != 0);
* 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;
list_insert_tail(&dn->dn_dirty_records[txgoff], dr);
mutex_exit(&dn->dn_mtx);
dnode_setdirty(dn, tx);
+ DB_DNODE_EXIT(db);
return (dr);
} else if (do_free_accounting) {
blkptr_t *bp = db->db_blkptr;
* db_blkptr, but since this is just a guess,
* it's OK if we get an odd answer.
*/
+ ddt_prefetch(os->os_spa, bp);
dnode_willuse_space(dn, -willfree, tx);
}
} else {
ASSERT(db->db_level+1 == dn->dn_nlevels);
ASSERT(db->db_blkid < dn->dn_nblkptr);
- ASSERT(db->db_parent == NULL ||
- db->db_parent == db->db_dnode->dn_dbuf);
+ ASSERT(db->db_parent == NULL || db->db_parent == dn->dn_dbuf);
mutex_enter(&dn->dn_mtx);
ASSERT(!list_link_active(&dr->dr_dirty_node));
list_insert_tail(&dn->dn_dirty_records[txgoff], dr);
}
dnode_setdirty(dn, tx);
+ DB_DNODE_EXIT(db);
return (dr);
}
static int
dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
{
- dnode_t *dn = db->db_dnode;
+ dnode_t *dn;
uint64_t txg = tx->tx_txg;
dbuf_dirty_record_t *dr, **drp;
ASSERT(dr->dr_txg == txg);
ASSERT(dr->dr_dbuf == db);
+ DB_DNODE_ENTER(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.
+ * Also note: we can get here with a spill block, so
+ * test for that similar to how dbuf_dirty does.
*/
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);
+ if (db->db_blkid != DMU_SPILL_BLKID) {
+ 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);
}
*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);
mutex_exit(&dn->dn_mtx);
}
+ DB_DNODE_EXIT(db);
if (db->db_level == 0) {
if (db->db_state != DB_NOFILL) {
ASSERT(tx->tx_txg != 0);
ASSERT(!refcount_is_zero(&db->db_holds));
- if (RW_WRITE_HELD(&db->db_dnode->dn_struct_rwlock))
+ DB_DNODE_ENTER(db);
+ if (RW_WRITE_HELD(&DB_DNODE(db)->dn_struct_rwlock))
rf |= DB_RF_HAVESTRUCT;
+ DB_DNODE_EXIT(db);
(void) dbuf_read(db, NULL, rf);
(void) dbuf_dirty(db, tx);
}
dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx)
{
ASSERT(!refcount_is_zero(&db->db_holds));
- ASSERT(db->db_dnode->dn_object != DMU_META_DNODE_OBJECT);
ASSERT(db->db_blkid != DMU_BONUS_BLKID);
ASSERT(db->db_level == 0);
ASSERT(DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA);
* in this case. For callers from the DMU we will usually see:
* dbuf_clear()->arc_buf_evict()->dbuf_do_evict()->dbuf_destroy()
* For the arc callback, we will usually see:
- * dbuf_do_evict()->dbuf_clear();dbuf_destroy()
+ * dbuf_do_evict()->dbuf_clear();dbuf_destroy()
* Sometimes, though, we will get a mix of these two:
* DMU: dbuf_clear()->arc_buf_evict()
* ARC: dbuf_do_evict()->dbuf_destroy()
void
dbuf_clear(dmu_buf_impl_t *db)
{
- dnode_t *dn = db->db_dnode;
+ dnode_t *dn;
dmu_buf_impl_t *parent = db->db_parent;
- dmu_buf_impl_t *dndb = dn->dn_dbuf;
+ dmu_buf_impl_t *dndb;
int dbuf_gone = FALSE;
ASSERT(MUTEX_HELD(&db->db_mtx));
db->db_state = DB_EVICTING;
db->db_blkptr = NULL;
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
+ dndb = dn->dn_dbuf;
if (db->db_blkid != DMU_BONUS_BLKID && MUTEX_HELD(&dn->dn_dbufs_mtx)) {
list_remove(&dn->dn_dbufs, db);
+ (void) atomic_dec_32_nv(&dn->dn_dbufs_count);
+ membar_producer();
+ DB_DNODE_EXIT(db);
+ /*
+ * Decrementing the dbuf count means that the hold corresponding
+ * to the removed dbuf is no longer discounted in dnode_move(),
+ * so the dnode cannot be moved until after we release the hold.
+ * The membar_producer() ensures visibility of the decremented
+ * value in dnode_move(), since DB_DNODE_EXIT doesn't actually
+ * release any lock.
+ */
dnode_rele(dn, db);
- db->db_dnode = NULL;
+ db->db_dnode_handle = NULL;
+ } else {
+ DB_DNODE_EXIT(db);
}
if (db->db_buf)
mutex_exit(&db->db_mtx);
/*
- * If this dbuf is referened from an indirect dbuf,
+ * If this dbuf is referenced from an indirect dbuf,
* decrement the ref count on the indirect dbuf.
*/
if (parent && parent != dndb)
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;
return (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_blkid = blkid;
db->db_last_dirty = NULL;
db->db_dirtycnt = 0;
- db->db_dnode = dn;
+ db->db_dnode_handle = dn->dn_handle;
db->db_parent = parent;
db->db_blkptr = blkptr;
ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT ||
refcount_count(&dn->dn_holds) > 0);
(void) refcount_add(&dn->dn_holds, db);
+ (void) atomic_inc_32_nv(&dn->dn_dbufs_count);
dprintf_dbuf(db, "db=%p\n", db);
* If this dbuf is still on the dn_dbufs list,
* remove it from that list.
*/
- if (db->db_dnode) {
- dnode_t *dn = db->db_dnode;
+ if (db->db_dnode_handle != NULL) {
+ dnode_t *dn;
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
mutex_enter(&dn->dn_dbufs_mtx);
list_remove(&dn->dn_dbufs, db);
+ (void) atomic_dec_32_nv(&dn->dn_dbufs_count);
mutex_exit(&dn->dn_dbufs_mtx);
-
+ DB_DNODE_EXIT(db);
+ /*
+ * Decrementing the dbuf count means that the hold
+ * corresponding to the removed dbuf is no longer
+ * discounted in dnode_move(), so the dnode cannot be
+ * moved until after we release the hold.
+ */
dnode_rele(dn, db);
- db->db_dnode = NULL;
+ db->db_dnode_handle = NULL;
}
dbuf_hash_remove(db);
}
return;
/* dbuf_find() returns with db_mtx held */
- if (db = dbuf_find(dn, 0, blkid)) {
- if (refcount_count(&db->db_holds) > 0) {
- /*
- * This dbuf is active. We assume that it is
- * already CACHED, or else about to be either
- * read or filled.
- */
- mutex_exit(&db->db_mtx);
- return;
- }
+ 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
+ * read or filled.
+ */
mutex_exit(&db->db_mtx);
- db = NULL;
+ 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, priority,
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 = 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(db->db_dnode->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->db_dnode, ==, 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)
{
dbuf_spill_set_blksz(dmu_buf_t *db_fake, uint64_t blksz, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
+ dnode_t *dn;
+
if (db->db_blkid != DMU_SPILL_BLKID)
return (ENOTSUP);
if (blksz == 0)
else
blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE);
- rw_enter(&db->db_dnode->dn_struct_rwlock, RW_WRITER);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
+ rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
dbuf_new_size(db, blksz, tx);
- rw_exit(&db->db_dnode->dn_struct_rwlock);
+ rw_exit(&dn->dn_struct_rwlock);
+ DB_DNODE_EXIT(db);
return (0);
}
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);
}
+/*
+ * If you call dbuf_rele() you had better not be referencing the dnode handle
+ * unless you have some other direct or indirect hold on the dnode. (An indirect
+ * hold is a hold on one of the dnode's dbufs, including the bonus buffer.)
+ * Without that, the dbuf_rele() could lead to a dnode_rele() followed by the
+ * dnode's parent dbuf evicting its dnode handles.
+ */
#pragma weak dmu_buf_rele = dbuf_rele
void
dbuf_rele(dmu_buf_impl_t *db, void *tag)
ASSERT(MUTEX_HELD(&db->db_mtx));
DBUF_VERIFY(db);
+ /*
+ * Remove the reference to the dbuf before removing its hold on the
+ * dnode so we can guarantee in dnode_move() that a referenced bonus
+ * buffer has a corresponding dnode hold.
+ */
holds = refcount_remove(&db->db_holds, tag);
ASSERT(holds >= 0);
if (holds == 0) {
if (db->db_blkid == DMU_BONUS_BLKID) {
mutex_exit(&db->db_mtx);
- dnode_rele(db->db_dnode, db);
+
+ /*
+ * If the dnode moves here, we cannot cross this barrier
+ * until the move completes.
+ */
+ DB_DNODE_ENTER(db);
+ (void) atomic_dec_32_nv(&DB_DNODE(db)->dn_dbufs_count);
+ DB_DNODE_EXIT(db);
+ /*
+ * The bonus buffer's dnode hold is no longer discounted
+ * in dnode_move(). The dnode cannot move until after
+ * the dnode_rele().
+ */
+ dnode_rele(DB_DNODE(db), db);
} else if (db->db_buf == NULL) {
/*
* This is a special case: we never associated this
dbuf_evict(db);
} else {
VERIFY(arc_buf_remove_ref(db->db_buf, db) == 0);
- if (!DBUF_IS_CACHEABLE(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);
}
}
-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;
- dnode_t *dn = db->db_dnode;
+ dnode_t *dn;
zio_t *zio;
ASSERT(dmu_tx_is_syncing(tx));
mutex_enter(&db->db_mtx);
}
ASSERT3U(db->db_state, ==, DB_CACHED);
- ASSERT3U(db->db.db_size, ==, 1<<dn->dn_phys->dn_indblkshift);
ASSERT(db->db_buf != NULL);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
+ ASSERT3U(db->db.db_size, ==, 1<<dn->dn_phys->dn_indblkshift);
dbuf_check_blkptr(dn, db);
+ DB_DNODE_EXIT(db);
db->db_data_pending = dr;
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;
dmu_buf_impl_t *db = dr->dr_dbuf;
- dnode_t *dn = db->db_dnode;
- objset_t *os = dn->dn_objset;
+ dnode_t *dn;
+ objset_t *os;
uint64_t txg = tx->tx_txg;
ASSERT(dmu_tx_is_syncing(tx));
}
DBUF_VERIFY(db);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
+
if (db->db_blkid == DMU_SPILL_BLKID) {
mutex_enter(&dn->dn_mtx);
dn->dn_phys->dn_flags |= DNODE_FLAG_SPILL_BLKPTR;
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);
+
if (*datap != db->db.db_data) {
zio_buf_free(*datap, DN_MAX_BONUSLEN);
arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
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;
return;
}
+ os = dn->dn_objset;
+
/*
* This function may have dropped the db_mtx lock allowing a dmu_sync
* operation to sneak in. As a result, we need to ensure that we
dbuf_check_blkptr(dn, db);
/*
- * If this buffer is in the middle of an immdiate write,
+ * If this buffer is in the middle of an immediate write,
* wait for the synchronous IO to complete.
*/
while (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) {
dbuf_write(dr, *datap, tx);
ASSERT(!list_link_active(&dr->dr_dirty_node));
- if (dn->dn_object == DMU_META_DNODE_OBJECT)
+ if (dn->dn_object == DMU_META_DNODE_OBJECT) {
list_insert_tail(&dn->dn_dirty_records[txg&TXG_MASK], dr);
- else
+ DB_DNODE_EXIT(db);
+ } else {
+ /*
+ * Although zio_nowait() does not "wait for an IO", it does
+ * initiate the IO. If this is an empty write it seems plausible
+ * that the IO could actually be completed before the nowait
+ * returns. We need to DB_DNODE_EXIT() first in case
+ * zio_nowait() invalidates the dbuf.
+ */
+ DB_DNODE_EXIT(db);
zio_nowait(dr->dr_zio);
+ }
}
void
{
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
dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
{
dmu_buf_impl_t *db = vdb;
+ dnode_t *dn;
blkptr_t *bp = zio->io_bp;
blkptr_t *bp_orig = &zio->io_bp_orig;
- dnode_t *dn = db->db_dnode;
spa_t *spa = zio->io_spa;
int64_t delta;
uint64_t fill = 0;
ASSERT(db->db_blkptr == bp);
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig);
dnode_diduse_space(dn, delta - zio->io_prev_space_delta);
zio->io_prev_space_delta = delta;
if (BP_IS_HOLE(bp)) {
ASSERT(bp->blk_fill == 0);
+ DB_DNODE_EXIT(db);
return;
}
#ifdef ZFS_DEBUG
if (db->db_blkid == DMU_SPILL_BLKID) {
- dnode_t *dn = db->db_dnode;
ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR);
ASSERT(!(BP_IS_HOLE(db->db_blkptr)) &&
db->db_blkptr == &dn->dn_phys->dn_spill);
fill += ibp->blk_fill;
}
}
+ DB_DNODE_EXIT(db);
bp->blk_fill = fill;
dmu_buf_impl_t *db = vdb;
blkptr_t *bp = zio->io_bp;
blkptr_t *bp_orig = &zio->io_bp_orig;
- dnode_t *dn = db->db_dnode;
- objset_t *os = dn->dn_objset;
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) {
ASSERT(BP_EQUAL(bp, bp_orig));
} else {
- dsl_dataset_t *ds = os->os_dsl_dataset;
- dmu_tx_t *tx = os->os_synctx;
+ objset_t *os;
+ dsl_dataset_t *ds;
+ dmu_tx_t *tx;
+
+ DB_GET_OBJSET(&os, db);
+ ds = os->os_dsl_dataset;
+ tx = os->os_synctx;
(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
dsl_dataset_block_born(ds, bp, tx);
#ifdef ZFS_DEBUG
if (db->db_blkid == DMU_SPILL_BLKID) {
- dnode_t *dn = db->db_dnode;
+ dnode_t *dn;
+
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR);
ASSERT(!(BP_IS_HOLE(db->db_blkptr)) &&
db->db_blkptr == &dn->dn_phys->dn_spill);
+ DB_DNODE_EXIT(db);
}
#endif
arc_set_callback(db->db_buf, dbuf_do_evict, db);
}
} else {
+ dnode_t *dn;
+
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(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
>> (db->db_level * epbs), >=, db->db_blkid);
arc_set_callback(db->db_buf, dbuf_do_evict, db);
}
+ DB_DNODE_EXIT(db);
mutex_destroy(&dr->dt.di.dr_mtx);
list_destroy(&dr->dt.di.dr_children);
}
dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx)
{
dmu_buf_impl_t *db = dr->dr_dbuf;
- dnode_t *dn = db->db_dnode;
- objset_t *os = dn->dn_objset;
+ dnode_t *dn;
+ objset_t *os;
dmu_buf_impl_t *parent = db->db_parent;
uint64_t txg = tx->tx_txg;
zbookmark_t zb;
zio_t *zio;
int wp_flag = 0;
+ DB_DNODE_ENTER(db);
+ dn = DB_DNODE(db);
+ os = dn->dn_objset;
+
if (db->db_state != DB_NOFILL) {
if (db->db_level > 0 || dn->dn_type == DMU_OT_DNODE) {
/*
wp_flag |= (db->db_state == DB_NOFILL) ? WP_NOFILL : 0;
dmu_write_policy(os, dn, db->db_level, wp_flag, &zp);
+ DB_DNODE_EXIT(db);
if (db->db_level == 0 && dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
ASSERT(db->db_state != DB_NOFILL);
} 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,
- ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
- }
-}
+ db->db_blkptr, data, DBUF_IS_L2CACHEABLE(db),
+ DBUF_IS_L2COMPRESSIBLE(db), &zp, dbuf_write_ready,
+ 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
git://git.camperquake.de / zfs.git / blobdiff