Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 318934) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 318935) @@ -1,3592 +1,3597 @@ /* * 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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include uint_t zfs_dbuf_evict_key; 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); #ifndef __lint extern inline void dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func_sync, dmu_buf_evict_func_t *evict_func_async, dmu_buf_t **clear_on_evict_dbufp); #endif /* ! __lint */ /* * Global data structures and functions for the dbuf cache. */ static kmem_cache_t *dbuf_kmem_cache; static taskq_t *dbu_evict_taskq; static kthread_t *dbuf_cache_evict_thread; static kmutex_t dbuf_evict_lock; static kcondvar_t dbuf_evict_cv; static boolean_t dbuf_evict_thread_exit; /* * LRU cache of dbufs. The dbuf cache maintains a list of dbufs that * are not currently held but have been recently released. These dbufs * are not eligible for arc eviction until they are aged out of the cache. * Dbufs are added to the dbuf cache once the last hold is released. If a * dbuf is later accessed and still exists in the dbuf cache, then it will * be removed from the cache and later re-added to the head of the cache. * Dbufs that are aged out of the cache will be immediately destroyed and * become eligible for arc eviction. */ static multilist_t *dbuf_cache; static refcount_t dbuf_cache_size; uint64_t dbuf_cache_max_bytes = 100 * 1024 * 1024; /* Cap the size of the dbuf cache to log2 fraction of arc size. */ int dbuf_cache_max_shift = 5; /* * The dbuf cache uses a three-stage eviction policy: * - A low water marker designates when the dbuf eviction thread * should stop evicting from the dbuf cache. * - When we reach the maximum size (aka mid water mark), we * signal the eviction thread to run. * - The high water mark indicates when the eviction thread * is unable to keep up with the incoming load and eviction must * happen in the context of the calling thread. * * The dbuf cache: * (max size) * low water mid water hi water * +----------------------------------------+----------+----------+ * | | | | * | | | | * | | | | * | | | | * +----------------------------------------+----------+----------+ * stop signal evict * evicting eviction directly * thread * * The high and low water marks indicate the operating range for the eviction * thread. The low water mark is, by default, 90% of the total size of the * cache and the high water mark is at 110% (both of these percentages can be * changed by setting dbuf_cache_lowater_pct and dbuf_cache_hiwater_pct, * respectively). The eviction thread will try to ensure that the cache remains * within this range by waking up every second and checking if the cache is * above the low water mark. The thread can also be woken up by callers adding * elements into the cache if the cache is larger than the mid water (i.e max * cache size). Once the eviction thread is woken up and eviction is required, * it will continue evicting buffers until it's able to reduce the cache size * to the low water mark. If the cache size continues to grow and hits the high * water mark, then callers adding elments to the cache will begin to evict * directly from the cache until the cache is no longer above the high water * mark. */ /* * The percentage above and below the maximum cache size. */ uint_t dbuf_cache_hiwater_pct = 10; uint_t dbuf_cache_lowater_pct = 10; /* ARGSUSED */ static int dbuf_cons(void *vdb, void *unused, int kmflag) { dmu_buf_impl_t *db = vdb; bzero(db, sizeof (dmu_buf_impl_t)); mutex_init(&db->db_mtx, NULL, MUTEX_DEFAULT, NULL); cv_init(&db->db_changed, NULL, CV_DEFAULT, NULL); multilist_link_init(&db->db_cache_link); refcount_create(&db->db_holds); return (0); } /* ARGSUSED */ static void dbuf_dest(void *vdb, void *unused) { dmu_buf_impl_t *db = vdb; mutex_destroy(&db->db_mtx); cv_destroy(&db->db_changed); ASSERT(!multilist_link_active(&db->db_cache_link)); refcount_destroy(&db->db_holds); } /* * dbuf hash table routines */ static dbuf_hash_table_t dbuf_hash_table; static uint64_t dbuf_hash_count; static uint64_t dbuf_hash(void *os, uint64_t obj, uint8_t lvl, uint64_t blkid) { uintptr_t osv = (uintptr_t)os; uint64_t crc = -1ULL; ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (lvl)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (blkid >> 0)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (blkid >> 8)) & 0xFF]; crc ^= (osv>>14) ^ (obj>>16) ^ (blkid>>16); return (crc); } #define DBUF_EQUAL(dbuf, os, obj, level, blkid) \ ((dbuf)->db.db_object == (obj) && \ (dbuf)->db_objset == (os) && \ (dbuf)->db_level == (level) && \ (dbuf)->db_blkid == (blkid)) dmu_buf_impl_t * dbuf_find(objset_t *os, uint64_t obj, uint8_t level, uint64_t blkid) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv = dbuf_hash(os, obj, level, blkid); uint64_t idx = hv & h->hash_table_mask; dmu_buf_impl_t *db; 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)) { mutex_enter(&db->db_mtx); if (db->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (db); } mutex_exit(&db->db_mtx); } } mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (NULL); } static dmu_buf_impl_t * dbuf_find_bonus(objset_t *os, uint64_t object) { dnode_t *dn; dmu_buf_impl_t *db = NULL; if (dnode_hold(os, object, FTAG, &dn) == 0) { rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_bonus != NULL) { db = dn->dn_bonus; mutex_enter(&db->db_mtx); } rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); } return (db); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. */ static dmu_buf_impl_t * dbuf_hash_insert(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; 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; dmu_buf_impl_t *dbf; 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)) { mutex_enter(&dbf->db_mtx); if (dbf->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (dbf); } mutex_exit(&dbf->db_mtx); } } mutex_enter(&db->db_mtx); db->db_hash_next = h->hash_table[idx]; h->hash_table[idx] = db; mutex_exit(DBUF_HASH_MUTEX(h, idx)); atomic_inc_64(&dbuf_hash_count); return (NULL); } /* * Remove an entry from the hash table. It must be in the EVICTING state. */ static void 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; dmu_buf_impl_t *dbf, **dbp; /* * We musn't hold db_mtx to maintain lock ordering: * DBUF_HASH_MUTEX > db_mtx. */ ASSERT(refcount_is_zero(&db->db_holds)); ASSERT(db->db_state == DB_EVICTING); ASSERT(!MUTEX_HELD(&db->db_mtx)); mutex_enter(DBUF_HASH_MUTEX(h, idx)); dbp = &h->hash_table[idx]; while ((dbf = *dbp) != db) { dbp = &dbf->db_hash_next; ASSERT(dbf != NULL); } *dbp = db->db_hash_next; db->db_hash_next = NULL; mutex_exit(DBUF_HASH_MUTEX(h, idx)); atomic_dec_64(&dbuf_hash_count); } typedef enum { DBVU_EVICTING, DBVU_NOT_EVICTING } dbvu_verify_type_t; static void dbuf_verify_user(dmu_buf_impl_t *db, dbvu_verify_type_t verify_type) { #ifdef ZFS_DEBUG int64_t holds; if (db->db_user == NULL) return; /* Only data blocks support the attachment of user data. */ ASSERT(db->db_level == 0); /* Clients must resolve a dbuf before attaching user data. */ ASSERT(db->db.db_data != NULL); ASSERT3U(db->db_state, ==, DB_CACHED); holds = refcount_count(&db->db_holds); if (verify_type == DBVU_EVICTING) { /* * Immediate eviction occurs when holds == dirtycnt. * For normal eviction buffers, holds is zero on * eviction, except when dbuf_fix_old_data() calls * dbuf_clear_data(). However, the hold count can grow * during eviction even though db_mtx is held (see * dmu_bonus_hold() for an example), so we can only * test the generic invariant that holds >= dirtycnt. */ ASSERT3U(holds, >=, db->db_dirtycnt); } else { if (db->db_user_immediate_evict == TRUE) ASSERT3U(holds, >=, db->db_dirtycnt); else ASSERT3U(holds, >, 0); } #endif } static void dbuf_evict_user(dmu_buf_impl_t *db) { dmu_buf_user_t *dbu = db->db_user; ASSERT(MUTEX_HELD(&db->db_mtx)); if (dbu == NULL) return; dbuf_verify_user(db, DBVU_EVICTING); db->db_user = NULL; #ifdef ZFS_DEBUG if (dbu->dbu_clear_on_evict_dbufp != NULL) *dbu->dbu_clear_on_evict_dbufp = NULL; #endif /* * There are two eviction callbacks - one that we call synchronously * and one that we invoke via a taskq. The async one is useful for * avoiding lock order reversals and limiting stack depth. * * Note that if we have a sync callback but no async callback, * it's likely that the sync callback will free the structure * containing the dbu. In that case we need to take care to not * dereference dbu after calling the sync evict func. */ boolean_t has_async = (dbu->dbu_evict_func_async != NULL); if (dbu->dbu_evict_func_sync != NULL) dbu->dbu_evict_func_sync(dbu); if (has_async) { taskq_dispatch_ent(dbu_evict_taskq, dbu->dbu_evict_func_async, dbu, 0, &dbu->dbu_tqent); } } 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); } } /* * This function *must* return indices evenly distributed between all * sublists of the multilist. This is needed due to how the dbuf eviction * code is laid out; dbuf_evict_thread() assumes dbufs are evenly * distributed between all sublists and uses this assumption when * deciding which sublist to evict from and how much to evict from it. */ unsigned int dbuf_cache_multilist_index_func(multilist_t *ml, void *obj) { dmu_buf_impl_t *db = obj; /* * The assumption here, is the hash value for a given * dmu_buf_impl_t will remain constant throughout it's lifetime * (i.e. it's objset, object, level and blkid fields don't change). * Thus, we don't need to store the dbuf's sublist index * on insertion, as this index can be recalculated on removal. * * Also, the low order bits of the hash value are thought to be * distributed evenly. Otherwise, in the case that the multilist * has a power of two number of sublists, each sublists' usage * would not be evenly distributed. */ return (dbuf_hash(db->db_objset, db->db.db_object, db->db_level, db->db_blkid) % multilist_get_num_sublists(ml)); } static inline boolean_t dbuf_cache_above_hiwater(void) { uint64_t dbuf_cache_hiwater_bytes = (dbuf_cache_max_bytes * dbuf_cache_hiwater_pct) / 100; return (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes + dbuf_cache_hiwater_bytes); } static inline boolean_t dbuf_cache_above_lowater(void) { uint64_t dbuf_cache_lowater_bytes = (dbuf_cache_max_bytes * dbuf_cache_lowater_pct) / 100; return (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes - dbuf_cache_lowater_bytes); } /* * Evict the oldest eligible dbuf from the dbuf cache. */ static void dbuf_evict_one(void) { int idx = multilist_get_random_index(dbuf_cache); multilist_sublist_t *mls = multilist_sublist_lock(dbuf_cache, idx); ASSERT(!MUTEX_HELD(&dbuf_evict_lock)); /* * Set the thread's tsd to indicate that it's processing evictions. * Once a thread stops evicting from the dbuf cache it will * reset its tsd to NULL. */ ASSERT3P(tsd_get(zfs_dbuf_evict_key), ==, NULL); (void) tsd_set(zfs_dbuf_evict_key, (void *)B_TRUE); dmu_buf_impl_t *db = multilist_sublist_tail(mls); while (db != NULL && mutex_tryenter(&db->db_mtx) == 0) { db = multilist_sublist_prev(mls, db); } DTRACE_PROBE2(dbuf__evict__one, dmu_buf_impl_t *, db, multilist_sublist_t *, mls); if (db != NULL) { multilist_sublist_remove(mls, db); multilist_sublist_unlock(mls); (void) refcount_remove_many(&dbuf_cache_size, db->db.db_size, db); dbuf_destroy(db); } else { multilist_sublist_unlock(mls); } (void) tsd_set(zfs_dbuf_evict_key, NULL); } /* * The dbuf evict thread is responsible for aging out dbufs from the * cache. Once the cache has reached it's maximum size, dbufs are removed * and destroyed. The eviction thread will continue running until the size * of the dbuf cache is at or below the maximum size. Once the dbuf is aged * out of the cache it is destroyed and becomes eligible for arc eviction. */ static void dbuf_evict_thread(void *dummy __unused) { callb_cpr_t cpr; CALLB_CPR_INIT(&cpr, &dbuf_evict_lock, callb_generic_cpr, FTAG); mutex_enter(&dbuf_evict_lock); while (!dbuf_evict_thread_exit) { while (!dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { CALLB_CPR_SAFE_BEGIN(&cpr); (void) cv_timedwait_hires(&dbuf_evict_cv, &dbuf_evict_lock, SEC2NSEC(1), MSEC2NSEC(1), 0); CALLB_CPR_SAFE_END(&cpr, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); /* * Keep evicting as long as we're above the low water mark * for the cache. We do this without holding the locks to * minimize lock contention. */ while (dbuf_cache_above_lowater() && !dbuf_evict_thread_exit) { dbuf_evict_one(); } mutex_enter(&dbuf_evict_lock); } dbuf_evict_thread_exit = B_FALSE; cv_broadcast(&dbuf_evict_cv); CALLB_CPR_EXIT(&cpr); /* drops dbuf_evict_lock */ thread_exit(); } /* * Wake up the dbuf eviction thread if the dbuf cache is at its max size. * If the dbuf cache is at its high water mark, then evict a dbuf from the * dbuf cache using the callers context. */ static void dbuf_evict_notify(void) { /* * We use thread specific data to track when a thread has * started processing evictions. This allows us to avoid deeply * nested stacks that would have a call flow similar to this: * * dbuf_rele()-->dbuf_rele_and_unlock()-->dbuf_evict_notify() * ^ | * | | * +-----dbuf_destroy()<--dbuf_evict_one()<--------+ * * The dbuf_eviction_thread will always have its tsd set until * that thread exits. All other threads will only set their tsd * if they are participating in the eviction process. This only * happens if the eviction thread is unable to process evictions * fast enough. To keep the dbuf cache size in check, other threads * can evict from the dbuf cache directly. Those threads will set * their tsd values so that we ensure that they only evict one dbuf * from the dbuf cache. */ if (tsd_get(zfs_dbuf_evict_key) != NULL) return; if (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes) { boolean_t evict_now = B_FALSE; mutex_enter(&dbuf_evict_lock); if (refcount_count(&dbuf_cache_size) > dbuf_cache_max_bytes) { evict_now = dbuf_cache_above_hiwater(); cv_signal(&dbuf_evict_cv); } mutex_exit(&dbuf_evict_lock); if (evict_now) { dbuf_evict_one(); } } } void dbuf_init(void) { uint64_t hsize = 1ULL << 16; dbuf_hash_table_t *h = &dbuf_hash_table; int i; /* * The hash table is big enough to fill all of physical memory * with an average 4K block size. The table will take up * totalmem*sizeof(void*)/4K (i.e. 2MB/GB with 8-byte pointers). */ while (hsize * 4096 < (uint64_t)physmem * PAGESIZE) hsize <<= 1; retry: h->hash_table_mask = hsize - 1; h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP); if (h->hash_table == NULL) { /* XXX - we should really return an error instead of assert */ ASSERT(hsize > (1ULL << 10)); hsize >>= 1; goto retry; } dbuf_kmem_cache = kmem_cache_create("dmu_buf_impl_t", sizeof (dmu_buf_impl_t), 0, dbuf_cons, dbuf_dest, NULL, NULL, NULL, 0); for (i = 0; i < DBUF_MUTEXES; i++) mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL); /* * Setup the parameters for the dbuf cache. We cap the size of the * dbuf cache to 1/32nd (default) of the size of the ARC. */ dbuf_cache_max_bytes = MIN(dbuf_cache_max_bytes, arc_max_bytes() >> dbuf_cache_max_shift); /* * All entries are queued via taskq_dispatch_ent(), so min/maxalloc * configuration is not required. */ dbu_evict_taskq = taskq_create("dbu_evict", 1, minclsyspri, 0, 0, 0); dbuf_cache = multilist_create(sizeof (dmu_buf_impl_t), offsetof(dmu_buf_impl_t, db_cache_link), dbuf_cache_multilist_index_func); refcount_create(&dbuf_cache_size); tsd_create(&zfs_dbuf_evict_key, NULL); dbuf_evict_thread_exit = B_FALSE; mutex_init(&dbuf_evict_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&dbuf_evict_cv, NULL, CV_DEFAULT, NULL); dbuf_cache_evict_thread = thread_create(NULL, 0, dbuf_evict_thread, NULL, 0, &p0, TS_RUN, minclsyspri); } void dbuf_fini(void) { dbuf_hash_table_t *h = &dbuf_hash_table; int i; for (i = 0; i < DBUF_MUTEXES; i++) mutex_destroy(&h->hash_mutexes[i]); kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *)); kmem_cache_destroy(dbuf_kmem_cache); taskq_destroy(dbu_evict_taskq); mutex_enter(&dbuf_evict_lock); dbuf_evict_thread_exit = B_TRUE; while (dbuf_evict_thread_exit) { cv_signal(&dbuf_evict_cv); cv_wait(&dbuf_evict_cv, &dbuf_evict_lock); } mutex_exit(&dbuf_evict_lock); tsd_destroy(&zfs_dbuf_evict_key); mutex_destroy(&dbuf_evict_lock); cv_destroy(&dbuf_evict_cv); refcount_destroy(&dbuf_cache_size); multilist_destroy(dbuf_cache); } /* * Other stuff. */ #ifdef ZFS_DEBUG static void dbuf_verify(dmu_buf_impl_t *db) { dnode_t *dn; dbuf_dirty_record_t *dr; ASSERT(MUTEX_HELD(&db->db_mtx)); if (!(zfs_flags & ZFS_DEBUG_DBUF_VERIFY)) 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); } else { 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 || !avl_is_empty(&dn->dn_dbufs)); } if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT3U(db->db.db_offset, ==, DMU_BONUS_BLKID); } else if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT0(db->db.db_offset); } else { ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size); } for (dr = db->db_data_pending; dr != NULL; dr = dr->dr_next) ASSERT(dr->dr_dbuf == db); for (dr = db->db_last_dirty; dr != NULL; dr = dr->dr_next) ASSERT(dr->dr_dbuf == db); /* * We can't assert that db_size matches dn_datablksz because it * can be momentarily different when another thread is doing * dnode_set_blksz(). */ if (db->db_level == 0 && db->db.db_object == DMU_META_DNODE_OBJECT) { dr = db->db_data_pending; /* * It should only be modified in syncing context, so * make sure we only have one copy of the data. */ ASSERT(dr == NULL || dr->dt.dl.dr_data == db->db_buf); } /* verify db->db_blkptr */ if (db->db_blkptr) { if (db->db_parent == dn->dn_dbuf) { /* db is pointed to by the dnode */ /* ASSERT3U(db->db_blkid, <, dn->dn_nblkptr); */ if (DMU_OBJECT_IS_SPECIAL(db->db.db_object)) ASSERT(db->db_parent == NULL); else ASSERT(db->db_parent != NULL); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &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; ASSERT3U(db->db_parent->db_level, ==, db->db_level+1); ASSERT3U(db->db_parent->db.db_object, ==, db->db.db_object); /* * dnode_grow_indblksz() can make this fail if we don't * have the struct_rwlock. XXX indblksz no longer * grows. safe to do this now? */ if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) { ASSERT3P(db->db_blkptr, ==, ((blkptr_t *)db->db_parent->db.db_data + db->db_blkid % epb)); } } } if ((db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr)) && (db->db_buf == NULL || db->db_buf->b_data) && db->db.db_data && db->db_blkid != DMU_BONUS_BLKID && db->db_state != DB_FILL && !dn->dn_free_txg) { /* * If the blkptr isn't set but they have nonzero data, * it had better be dirty, otherwise we'll lose that * data when we evict this buffer. * * There is an exception to this rule for indirect blocks; in * this case, if the indirect block is a hole, we fill in a few * fields on each of the child blocks (importantly, birth time) * to prevent hole birth times from being lost when you * partially fill in a hole. */ if (db->db_dirtycnt == 0) { if (db->db_level == 0) { uint64_t *buf = db->db.db_data; int i; for (i = 0; i < db->db.db_size >> 3; i++) { ASSERT(buf[i] == 0); } } else { blkptr_t *bps = db->db.db_data; ASSERT3U(1 << DB_DNODE(db)->dn_indblkshift, ==, db->db.db_size); /* * We want to verify that all the blkptrs in the * indirect block are holes, but we may have * automatically set up a few fields for them. * We iterate through each blkptr and verify * they only have those fields set. */ for (int i = 0; i < db->db.db_size / sizeof (blkptr_t); i++) { blkptr_t *bp = &bps[i]; ASSERT(ZIO_CHECKSUM_IS_ZERO( &bp->blk_cksum)); ASSERT( DVA_IS_EMPTY(&bp->blk_dva[0]) && DVA_IS_EMPTY(&bp->blk_dva[1]) && DVA_IS_EMPTY(&bp->blk_dva[2])); ASSERT0(bp->blk_fill); ASSERT0(bp->blk_pad[0]); ASSERT0(bp->blk_pad[1]); ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT(BP_IS_HOLE(bp)); ASSERT0(bp->blk_phys_birth); } } } } DB_DNODE_EXIT(db); } #endif static void dbuf_clear_data(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); dbuf_evict_user(db); ASSERT3P(db->db_buf, ==, NULL); db->db.db_data = NULL; if (db->db_state != DB_NOFILL) db->db_state = DB_UNCACHED; } static void dbuf_set_data(dmu_buf_impl_t *db, arc_buf_t *buf) { ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(buf != NULL); db->db_buf = buf; ASSERT(buf->b_data != NULL); db->db.db_data = buf->b_data; } /* * Loan out an arc_buf for read. Return the loaned arc_buf. */ arc_buf_t * dbuf_loan_arcbuf(dmu_buf_impl_t *db) { arc_buf_t *abuf; ASSERT(db->db_blkid != DMU_BONUS_BLKID); 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 = db->db_objset->os_spa; mutex_exit(&db->db_mtx); abuf = arc_loan_buf(spa, B_FALSE, blksz); bcopy(db->db.db_data, abuf->b_data, blksz); } else { abuf = db->db_buf; arc_loan_inuse_buf(abuf, db); db->db_buf = NULL; dbuf_clear_data(db); mutex_exit(&db->db_mtx); } return (abuf); } /* * Calculate which level n block references the data at the level 0 offset * provided. */ uint64_t dbuf_whichblock(dnode_t *dn, int64_t level, uint64_t offset) { if (dn->dn_datablkshift != 0 && dn->dn_indblkshift != 0) { /* * The level n blkid is equal to the level 0 blkid divided by * the number of level 0s in a level n block. * * The level 0 blkid is offset >> datablkshift = * offset / 2^datablkshift. * * The number of level 0s in a level n is the number of block * pointers in an indirect block, raised to the power of level. * This is 2^(indblkshift - SPA_BLKPTRSHIFT)^level = * 2^(level*(indblkshift - SPA_BLKPTRSHIFT)). * * Thus, the level n blkid is: offset / * ((2^datablkshift)*(2^(level*(indblkshift - SPA_BLKPTRSHIFT))) * = offset / 2^(datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) * = offset >> (datablkshift + level * * (indblkshift - SPA_BLKPTRSHIFT)) */ return (offset >> (dn->dn_datablkshift + level * (dn->dn_indblkshift - SPA_BLKPTRSHIFT))); } else { ASSERT3U(offset, <, dn->dn_datablksz); return (0); } } static void dbuf_read_done(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; mutex_enter(&db->db_mtx); ASSERT3U(db->db_state, ==, DB_READ); /* * All reads are synchronous, so we must have a hold on the dbuf */ ASSERT(refcount_count(&db->db_holds) > 0); ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); if (db->db_level == 0 && db->db_freed_in_flight) { /* we were freed in flight; disregard any error */ arc_release(buf, db); bzero(buf->b_data, db->db.db_size); arc_buf_freeze(buf); db->db_freed_in_flight = FALSE; dbuf_set_data(db, buf); db->db_state = DB_CACHED; } else if (zio == NULL || zio->io_error == 0) { dbuf_set_data(db, buf); db->db_state = DB_CACHED; } else { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT3P(db->db_buf, ==, NULL); arc_buf_destroy(buf, db); db->db_state = DB_UNCACHED; } cv_broadcast(&db->db_changed); dbuf_rele_and_unlock(db, NULL); } static void dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags) { dnode_t *dn; zbookmark_phys_t zb; arc_flags_t aflags = ARC_FLAG_NOWAIT; 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)); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_state == DB_UNCACHED); ASSERT(db->db_buf == NULL); if (db->db_blkid == DMU_BONUS_BLKID) { int bonuslen = MIN(dn->dn_bonuslen, dn->dn_phys->dn_bonuslen); ASSERT3U(bonuslen, <=, db->db.db_size); db->db.db_data = zio_buf_alloc(DN_MAX_BONUSLEN); arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); if (bonuslen < DN_MAX_BONUSLEN) 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); db->db_state = DB_CACHED; mutex_exit(&db->db_mtx); return; } /* * Recheck BP_IS_HOLE() after dnode_block_freed() in case dnode_sync() * processes the delete record and clears the bp while we are waiting * for the dn_mtx (resulting in a "no" from block_freed). */ if (db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr) || (db->db_level == 0 && (dnode_block_freed(dn, db->db_blkid) || BP_IS_HOLE(db->db_blkptr)))) { arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dbuf_set_data(db, arc_alloc_buf(db->db_objset->os_spa, db, type, db->db.db_size)); bzero(db->db.db_data, db->db.db_size); if (db->db_blkptr != NULL && db->db_level > 0 && BP_IS_HOLE(db->db_blkptr) && db->db_blkptr->blk_birth != 0) { blkptr_t *bps = db->db.db_data; for (int i = 0; i < ((1 << DB_DNODE(db)->dn_indblkshift) / sizeof (blkptr_t)); i++) { blkptr_t *bp = &bps[i]; ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, 1 << dn->dn_indblkshift); BP_SET_LSIZE(bp, BP_GET_LEVEL(db->db_blkptr) == 1 ? dn->dn_datablksz : BP_GET_LSIZE(db->db_blkptr)); BP_SET_TYPE(bp, BP_GET_TYPE(db->db_blkptr)); BP_SET_LEVEL(bp, BP_GET_LEVEL(db->db_blkptr) - 1); BP_SET_BIRTH(bp, db->db_blkptr->blk_birth, 0); } } DB_DNODE_EXIT(db); db->db_state = DB_CACHED; mutex_exit(&db->db_mtx); return; } DB_DNODE_EXIT(db); db->db_state = DB_READ; mutex_exit(&db->db_mtx); if (DBUF_IS_L2CACHEABLE(db)) aflags |= ARC_FLAG_L2CACHE; 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); (void) arc_read(zio, db->db_objset->os_spa, db->db_blkptr, dbuf_read_done, db, ZIO_PRIORITY_SYNC_READ, (flags & DB_RF_CANFAIL) ? ZIO_FLAG_CANFAIL : ZIO_FLAG_MUSTSUCCEED, &aflags, &zb); } /* * This is our just-in-time copy function. It makes a copy of buffers that * have been modified in a previous transaction group before we access them in * the current active group. * * This function is used in three places: when we are dirtying a buffer for the * first time in a txg, when we are freeing a range in a dnode that includes * this buffer, and when we are accessing a buffer which was received compressed * and later referenced in a WRITE_BYREF record. * * Note that when we are called from dbuf_free_range() we do not put a hold on * the buffer, we just traverse the active dbuf list for the dnode. */ static void dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg) { dbuf_dirty_record_t *dr = db->db_last_dirty; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db.db_data != NULL); ASSERT(db->db_level == 0); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT); if (dr == NULL || (dr->dt.dl.dr_data != ((db->db_blkid == DMU_BONUS_BLKID) ? db->db.db_data : db->db_buf))) return; /* * 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, * or (if there a no active holders) * just null out the current db_data pointer. */ ASSERT(dr->dr_txg >= txg - 2); if (db->db_blkid == DMU_BONUS_BLKID) { /* Note that the data bufs here are zio_bufs */ dr->dt.dl.dr_data = zio_buf_alloc(DN_MAX_BONUSLEN); arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); bcopy(db->db.db_data, dr->dt.dl.dr_data, DN_MAX_BONUSLEN); } else if (refcount_count(&db->db_holds) > db->db_dirtycnt) { int size = arc_buf_size(db->db_buf); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); spa_t *spa = db->db_objset->os_spa; enum zio_compress compress_type = arc_get_compression(db->db_buf); if (compress_type == ZIO_COMPRESS_OFF) { dr->dt.dl.dr_data = arc_alloc_buf(spa, db, type, size); } else { ASSERT3U(type, ==, ARC_BUFC_DATA); dr->dt.dl.dr_data = arc_alloc_compressed_buf(spa, db, size, arc_buf_lsize(db->db_buf), compress_type); } bcopy(db->db.db_data, dr->dt.dl.dr_data->b_data, size); } else { db->db_buf = NULL; dbuf_clear_data(db); } } int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags) { int err = 0; boolean_t prefetch; dnode_t *dn; /* * We don't have to hold the mutex to check db_state because it * can't be freed while we have a hold on the buffer. */ ASSERT(!refcount_is_zero(&db->db_holds)); if (db->db_state == DB_NOFILL) return (SET_ERROR(EIO)); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_enter(&dn->dn_struct_rwlock, RW_READER); prefetch = db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && (flags & DB_RF_NOPREFETCH) == 0 && dn != NULL && DBUF_IS_CACHEABLE(db); mutex_enter(&db->db_mtx); if (db->db_state == DB_CACHED) { /* * If the arc buf is compressed, we need to decompress it to * read the data. This could happen during the "zfs receive" of * a stream which is compressed and deduplicated. */ if (db->db_buf != NULL && arc_get_compression(db->db_buf) != ZIO_COMPRESS_OFF) { dbuf_fix_old_data(db, spa_syncing_txg(dmu_objset_spa(db->db_objset))); err = arc_decompress(db->db_buf); dbuf_set_data(db, db->db_buf); } mutex_exit(&db->db_mtx); if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); } else if (db->db_state == DB_UNCACHED) { spa_t *spa = dn->dn_objset->os_spa; boolean_t need_wait = B_FALSE; if (zio == NULL && db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr)) { zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); need_wait = B_TRUE; } dbuf_read_impl(db, zio, flags); /* dbuf_read_impl has dropped db_mtx for us */ if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db_blkid, 1, B_TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); if (need_wait) 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_blkid, 1, B_TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) 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 || db->db_state == DB_FILL) { ASSERT(db->db_state == DB_READ || (flags & DB_RF_HAVESTRUCT) == 0); DTRACE_PROBE2(blocked__read, dmu_buf_impl_t *, db, zio_t *, zio); cv_wait(&db->db_changed, &db->db_mtx); } if (db->db_state == DB_UNCACHED) err = SET_ERROR(EIO); } mutex_exit(&db->db_mtx); } return (err); } static void dbuf_noread(dmu_buf_impl_t *db) { ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) 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 = db->db_objset->os_spa; ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); dbuf_set_data(db, arc_alloc_buf(spa, db, type, db->db.db_size)); db->db_state = DB_FILL; } else if (db->db_state == DB_NOFILL) { dbuf_clear_data(db); } else { ASSERT3U(db->db_state, ==, DB_CACHED); } mutex_exit(&db->db_mtx); } void dbuf_unoverride(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *bp = &dr->dt.dl.dr_overridden_by; uint64_t txg = dr->dr_txg; ASSERT(MUTEX_HELD(&db->db_mtx)); + /* + * This assert is valid because dmu_sync() expects to be called by + * a zilog's get_data while holding a range lock. This call only + * comes from dbuf_dirty() callers who must also hold a range lock. + */ ASSERT(dr->dt.dl.dr_override_state != DR_IN_DMU_SYNC); ASSERT(db->db_level == 0); if (db->db_blkid == DMU_BONUS_BLKID || dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN) return; ASSERT(db->db_data_pending != dr); /* free this block */ if (!BP_IS_HOLE(bp) && !dr->dt.dl.dr_nopwrite) zio_free(db->db_objset->os_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 * modifying the buffer, so they will immediately do * another (redundant) arc_release(). Therefore, leave * the buf thawed to save the effort of freezing & * immediately re-thawing it. */ arc_release(dr->dt.dl.dr_data, db); } /* * 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. */ void dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid, dmu_tx_t *tx) { dmu_buf_impl_t db_search; dmu_buf_impl_t *db, *db_next; uint64_t txg = tx->tx_txg; avl_index_t where; if (end_blkid > dn->dn_maxblkid && !(start_blkid == DMU_SPILL_BLKID || end_blkid == DMU_SPILL_BLKID)) end_blkid = dn->dn_maxblkid; dprintf_dnode(dn, "start=%llu end=%llu\n", start_blkid, end_blkid); db_search.db_level = 0; db_search.db_blkid = start_blkid; db_search.db_state = DB_SEARCH; mutex_enter(&dn->dn_dbufs_mtx); db = avl_find(&dn->dn_dbufs, &db_search, &where); ASSERT3P(db, ==, NULL); db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); for (; db != NULL; db = db_next) { db_next = AVL_NEXT(&dn->dn_dbufs, db); ASSERT(db->db_blkid != DMU_BONUS_BLKID); if (db->db_level != 0 || db->db_blkid > end_blkid) { break; } ASSERT3U(db->db_blkid, >=, start_blkid); /* found a level 0 buffer in the range */ mutex_enter(&db->db_mtx); if (dbuf_undirty(db, tx)) { /* mutex has been dropped and dbuf destroyed */ continue; } if (db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL || db->db_state == DB_EVICTING) { ASSERT(db->db.db_data == NULL); mutex_exit(&db->db_mtx); continue; } if (db->db_state == DB_READ || db->db_state == DB_FILL) { /* will be handled in dbuf_read_done or dbuf_rele */ db->db_freed_in_flight = TRUE; mutex_exit(&db->db_mtx); continue; } if (refcount_count(&db->db_holds) == 0) { ASSERT(db->db_buf); dbuf_destroy(db); continue; } /* The dbuf is referenced */ if (db->db_last_dirty != NULL) { dbuf_dirty_record_t *dr = db->db_last_dirty; if (dr->dr_txg == txg) { /* * This buffer is "in-use", re-adjust the file * size to reflect that this buffer may * contain new data when we sync. */ if (db->db_blkid != DMU_SPILL_BLKID && db->db_blkid > dn->dn_maxblkid) dn->dn_maxblkid = db->db_blkid; dbuf_unoverride(dr); } else { /* * This dbuf is not dirty in the open context. * Either uncache it (if its not referenced in * the open context) or reset its contents to * empty. */ dbuf_fix_old_data(db, txg); } } /* clear the contents if its cached */ if (db->db_state == DB_CACHED) { ASSERT(db->db.db_data != NULL); arc_release(db->db_buf, db); bzero(db->db.db_data, db->db.db_size); arc_buf_freeze(db->db_buf); } mutex_exit(&db->db_mtx); } mutex_exit(&dn->dn_dbufs_mtx); } void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx) { 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(&dn->dn_struct_rwlock)); /* * This call to dmu_buf_will_dirty() with the dn_struct_rwlock held * is OK, because there can be no other references to the db * when we are changing its size, so no concurrent DB_FILL can * be happening. */ /* * XXX we should be doing a dbuf_read, checking the return * value and returning that up to our callers */ dmu_buf_will_dirty(&db->db, tx); /* create the data buffer for the new block */ buf = arc_alloc_buf(dn->dn_objset->os_spa, db, type, size); /* copy old block data to the new block */ obuf = db->db_buf; bcopy(obuf->b_data, buf->b_data, MIN(osize, size)); /* zero the remainder */ if (size > osize) bzero((uint8_t *)buf->b_data + osize, size - osize); mutex_enter(&db->db_mtx); dbuf_set_data(db, buf); arc_buf_destroy(obuf, db); db->db.db_size = size; if (db->db_level == 0) { ASSERT3U(db->db_last_dirty->dr_txg, ==, tx->tx_txg); db->db_last_dirty->dt.dl.dr_data = buf; } mutex_exit(&db->db_mtx); dmu_objset_willuse_space(dn->dn_objset, size - osize, tx); DB_DNODE_EXIT(db); } void dbuf_release_bp(dmu_buf_impl_t *db) { objset_t *os = db->db_objset; 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)); (void) arc_release(db->db_buf, db); } /* * We already have a dirty record for this TXG, and we are being * dirtied again. */ static void dbuf_redirty(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID) { /* * If this buffer has already been written out, * we now need to reset its state. */ dbuf_unoverride(dr); if (db->db.db_object != DMU_META_DNODE_OBJECT && db->db_state != DB_NOFILL) { /* Already released on initial dirty, so just thaw. */ ASSERT(arc_released(db->db_buf)); arc_buf_thaw(db->db_buf); } } } dbuf_dirty_record_t * dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; objset_t *os; dbuf_dirty_record_t **drp, *dr; int drop_struct_lock = FALSE; int txgoff = tx->tx_txg & TXG_MASK; ASSERT(tx->tx_txg != 0); 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 * were already pre-dirtied in open context. */ #ifdef DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_enter(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); } ASSERT(!dmu_tx_is_syncing(tx) || BP_IS_HOLE(dn->dn_objset->os_rootbp) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_objset->os_dsl_dataset == NULL); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif /* * We make this assert for private objects as well, but after we * check if we're already dirty. They are allowed to re-dirty * in syncing context. */ ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); mutex_enter(&db->db_mtx); /* * XXX make this true for indirects too? The problem is that * transactions created with dmu_tx_create_assigned() from * syncing context don't bother holding ahead. */ ASSERT(db->db_level != 0 || db->db_state == DB_CACHED || db->db_state == DB_FILL || db->db_state == DB_NOFILL); mutex_enter(&dn->dn_mtx); /* * Don't set dirtyctx to SYNC if we're just modifying this as we * initialize the objset. */ if (dn->dn_dirtyctx == DN_UNDIRTIED) { if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_enter(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); } if (!BP_IS_HOLE(dn->dn_objset->os_rootbp)) { 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); } if (dn->dn_objset->os_dsl_dataset != NULL) { rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG); } } mutex_exit(&dn->dn_mtx); if (db->db_blkid == DMU_SPILL_BLKID) dn->dn_have_spill = B_TRUE; /* * If this buffer is already dirty, we're done. */ drp = &db->db_last_dirty; ASSERT(*drp == NULL || (*drp)->dr_txg <= tx->tx_txg || db->db.db_object == DMU_META_DNODE_OBJECT); 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); dbuf_redirty(dr); mutex_exit(&db->db_mtx); return (dr); } /* * Only valid if not already dirty. */ ASSERT(dn->dn_object == 0 || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); ASSERT3U(dn->dn_nlevels, >, db->db_level); ASSERT((dn->dn_phys->dn_nlevels == 0 && db->db_level == 0) || dn->dn_phys->dn_nlevels > db->db_level || dn->dn_next_nlevels[txgoff] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-1) & TXG_MASK] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-2) & TXG_MASK] > db->db_level); /* * We should only be dirtying in syncing context if it's the * mos or we're initializing the os or it's a special object. * However, we are allowed to dirty in syncing context provided * we already dirtied it in open context. Hence we must make * this assertion only if we're not already dirty. */ os = dn->dn_objset; VERIFY3U(tx->tx_txg, <=, spa_final_dirty_txg(os->os_spa)); #ifdef DEBUG if (dn->dn_objset->os_dsl_dataset != NULL) rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_READER, FTAG); ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp)); if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif ASSERT(db->db.db_size != 0); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); if (db->db_blkid != DMU_BONUS_BLKID) { dmu_objset_willuse_space(os, db->db.db_size, tx); } /* * If this buffer is dirty in an old transaction group we need * 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); if (db->db_level == 0) { void *data_old = db->db_buf; if (db->db_state != DB_NOFILL) { if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db.db_data; } else if (db->db.db_object != DMU_META_DNODE_OBJECT) { /* * Release the data buffer from the cache so * that we can modify it without impacting * possible other users of this cached data * block. Note that indirect blocks and * private objects are not released until the * syncing state (since they are only modified * then). */ arc_release(db->db_buf, db); dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db_buf; } ASSERT(data_old != NULL); } dr->dt.dl.dr_data = data_old; } else { mutex_init(&dr->dt.di.dr_mtx, NULL, MUTEX_DEFAULT, NULL); list_create(&dr->dt.di.dr_children, 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; *drp = dr; /* * We could have been freed_in_flight between the dbuf_noread * and dbuf_dirty. We win, as though the dbuf_noread() had * happened after the free. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && db->db_blkid != DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_free_ranges[txgoff] != NULL) { range_tree_clear(dn->dn_free_ranges[txgoff], db->db_blkid, 1); } mutex_exit(&dn->dn_mtx); db->db_freed_in_flight = FALSE; } /* * This buffer is now part of this txg */ dbuf_add_ref(db, (void *)(uintptr_t)tx->tx_txg); db->db_dirtycnt += 1; ASSERT3U(db->db_dirtycnt, <=, 3); mutex_exit(&db->db_mtx); if (db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } /* * The dn_struct_rwlock prevents db_blkptr from changing * due to a write from syncing context completing * while we are running, so we want to acquire it before * looking at db_blkptr. */ if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { rw_enter(&dn->dn_struct_rwlock, RW_READER); drop_struct_lock = TRUE; } /* * If we are overwriting a dedup BP, then unless it is snapshotted, * when we get to syncing context we will need to decrement its * refcount in the DDT. Prefetch the relevant DDT block so that * syncing context won't have to wait for the i/o. */ ddt_prefetch(os->os_spa, db->db_blkptr); if (db->db_level == 0) { dnode_new_blkid(dn, db->db_blkid, tx, drop_struct_lock); ASSERT(dn->dn_maxblkid >= db->db_blkid); } if (db->db_level+1 < dn->dn_nlevels) { dmu_buf_impl_t *parent = db->db_parent; dbuf_dirty_record_t *di; int parent_held = FALSE; if (db->db_parent == NULL || db->db_parent == dn->dn_dbuf) { int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; parent = dbuf_hold_level(dn, db->db_level+1, db->db_blkid >> epbs, FTAG); ASSERT(parent != NULL); parent_held = TRUE; } if (drop_struct_lock) rw_exit(&dn->dn_struct_rwlock); ASSERT3U(db->db_level+1, ==, parent->db_level); di = dbuf_dirty(parent, tx); if (parent_held) dbuf_rele(parent, FTAG); mutex_enter(&db->db_mtx); /* * 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); ASSERT3U(di->dr_txg, ==, tx->tx_txg); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&di->dt.di.dr_children, dr); mutex_exit(&di->dt.di.dr_mtx); dr->dr_parent = di; } mutex_exit(&db->db_mtx); } else { ASSERT(db->db_level+1 == dn->dn_nlevels); ASSERT(db->db_blkid < dn->dn_nblkptr); 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); mutex_exit(&dn->dn_mtx); if (drop_struct_lock) rw_exit(&dn->dn_struct_rwlock); } dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } /* * 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; uint64_t txg = tx->tx_txg; dbuf_dirty_record_t *dr, **drp; ASSERT(txg != 0); /* * Due to our use of dn_nlevels below, this can only be called * in open context, unless we are operating on the MOS. * From syncing context, dn_nlevels may be different from the * dn_nlevels used when dbuf was dirtied. */ ASSERT(db->db_objset == dmu_objset_pool(db->db_objset)->dp_meta_objset || txg != spa_syncing_txg(dmu_objset_spa(db->db_objset))); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT0(db->db_level); ASSERT(MUTEX_HELD(&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) return (B_FALSE); ASSERT(dr->dr_txg == txg); ASSERT(dr->dr_dbuf == db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); ASSERT(db->db.db_size != 0); dsl_pool_undirty_space(dmu_objset_pool(dn->dn_objset), dr->dr_accounted, txg); *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_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_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) arc_buf_destroy(dr->dt.dl.dr_data, db); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; if (refcount_remove(&db->db_holds, (void *)(uintptr_t)txg) == 0) { ASSERT(db->db_state == DB_NOFILL || arc_released(db->db_buf)); dbuf_destroy(db); return (B_TRUE); } return (B_FALSE); } void dmu_buf_will_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; int rf = DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH; ASSERT(tx->tx_txg != 0); ASSERT(!refcount_is_zero(&db->db_holds)); /* * Quick check for dirtyness. For already dirty blocks, this * reduces runtime of this function by >90%, and overall performance * by 50% for some workloads (e.g. file deletion with indirect blocks * cached). */ mutex_enter(&db->db_mtx); dbuf_dirty_record_t *dr; for (dr = db->db_last_dirty; dr != NULL && dr->dr_txg >= tx->tx_txg; dr = dr->dr_next) { /* * It's possible that it is already dirty but not cached, * because there are some calls to dbuf_dirty() that don't * go through dmu_buf_will_dirty(). */ if (dr->dr_txg == tx->tx_txg && db->db_state == DB_CACHED) { /* This dbuf is already dirty and cached. */ dbuf_redirty(dr); mutex_exit(&db->db_mtx); return; } } mutex_exit(&db->db_mtx); 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); } void dmu_buf_will_not_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_state = DB_NOFILL; dmu_buf_will_fill(db_fake, tx); } void dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(tx->tx_txg != 0); ASSERT(db->db_level == 0); ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); dbuf_noread(db); (void) dbuf_dirty(db, tx); } #pragma weak dmu_buf_fill_done = dbuf_fill_done /* ARGSUSED */ void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx) { mutex_enter(&db->db_mtx); DBUF_VERIFY(db); if (db->db_state == DB_FILL) { if (db->db_level == 0 && db->db_freed_in_flight) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); /* we were freed while filling */ /* XXX dbuf_undirty? */ bzero(db->db.db_data, db->db.db_size); db->db_freed_in_flight = FALSE; } db->db_state = DB_CACHED; cv_broadcast(&db->db_changed); } mutex_exit(&db->db_mtx); } void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, bp_embedded_type_t etype, enum zio_compress comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; struct dirty_leaf *dl; dmu_object_type_t type; if (etype == BP_EMBEDDED_TYPE_DATA) { ASSERT(spa_feature_is_active(dmu_objset_spa(db->db_objset), SPA_FEATURE_EMBEDDED_DATA)); } DB_DNODE_ENTER(db); type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); ASSERT0(db->db_level); ASSERT(db->db_blkid != DMU_BONUS_BLKID); dmu_buf_will_not_fill(dbuf, tx); ASSERT3U(db->db_last_dirty->dr_txg, ==, tx->tx_txg); dl = &db->db_last_dirty->dt.dl; encode_embedded_bp_compressed(&dl->dr_overridden_by, data, comp, uncompressed_size, compressed_size); BPE_SET_ETYPE(&dl->dr_overridden_by, etype); BP_SET_TYPE(&dl->dr_overridden_by, type); BP_SET_LEVEL(&dl->dr_overridden_by, 0); BP_SET_BYTEORDER(&dl->dr_overridden_by, byteorder); dl->dr_override_state = DR_OVERRIDDEN; dl->dr_overridden_by.blk_birth = db->db_last_dirty->dr_txg; } /* * Directly assign a provided arc buf to a given dbuf if it's not referenced * by anybody except our caller. Otherwise copy arcbuf's contents to dbuf. */ void 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_blkid != DMU_BONUS_BLKID); ASSERT(db->db_level == 0); ASSERT3U(dbuf_is_metadata(db), ==, arc_is_metadata(buf)); ASSERT(buf != NULL); ASSERT(arc_buf_lsize(buf) == db->db.db_size); ASSERT(tx->tx_txg != 0); arc_return_buf(buf, db); ASSERT(arc_released(buf)); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); ASSERT(db->db_state == DB_CACHED || db->db_state == DB_UNCACHED); if (db->db_state == DB_CACHED && refcount_count(&db->db_holds) - 1 > db->db_dirtycnt) { mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); bcopy(buf->b_data, db->db.db_data, db->db.db_size); arc_buf_destroy(buf, db); xuio_stat_wbuf_copied(); return; } xuio_stat_wbuf_nocopy(); if (db->db_state == DB_CACHED) { dbuf_dirty_record_t *dr = db->db_last_dirty; ASSERT(db->db_buf != NULL); if (dr != NULL && dr->dr_txg == tx->tx_txg) { ASSERT(dr->dt.dl.dr_data == db->db_buf); if (!arc_released(db->db_buf)) { ASSERT(dr->dt.dl.dr_override_state == DR_OVERRIDDEN); arc_release(db->db_buf, db); } dr->dt.dl.dr_data = buf; arc_buf_destroy(db->db_buf, db); } else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) { arc_release(db->db_buf, db); arc_buf_destroy(db->db_buf, db); } db->db_buf = NULL; } ASSERT(db->db_buf == NULL); dbuf_set_data(db, buf); db->db_state = DB_FILL; mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); dmu_buf_fill_done(&db->db, tx); } void dbuf_destroy(dmu_buf_impl_t *db) { dnode_t *dn; dmu_buf_impl_t *parent = db->db_parent; dmu_buf_impl_t *dndb; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(refcount_is_zero(&db->db_holds)); if (db->db_buf != NULL) { arc_buf_destroy(db->db_buf, db); db->db_buf = NULL; } if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(db->db.db_data != NULL); zio_buf_free(db->db.db_data, DN_MAX_BONUSLEN); arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); db->db_state = DB_UNCACHED; } dbuf_clear_data(db); if (multilist_link_active(&db->db_cache_link)) { multilist_remove(dbuf_cache, db); (void) refcount_remove_many(&dbuf_cache_size, db->db.db_size, db); } ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); ASSERT(db->db_data_pending == NULL); db->db_state = DB_EVICTING; db->db_blkptr = NULL; /* * Now that db_state is DB_EVICTING, nobody else can find this via * the hash table. We can now drop db_mtx, which allows us to * acquire the dn_dbufs_mtx. */ mutex_exit(&db->db_mtx); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dndb = dn->dn_dbuf; if (db->db_blkid != DMU_BONUS_BLKID) { boolean_t needlock = !MUTEX_HELD(&dn->dn_dbufs_mtx); if (needlock) mutex_enter(&dn->dn_dbufs_mtx); avl_remove(&dn->dn_dbufs, db); atomic_dec_32(&dn->dn_dbufs_count); membar_producer(); DB_DNODE_EXIT(db); if (needlock) mutex_exit(&dn->dn_dbufs_mtx); /* * 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_handle = NULL; dbuf_hash_remove(db); } else { DB_DNODE_EXIT(db); } ASSERT(refcount_is_zero(&db->db_holds)); db->db_parent = NULL; ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); ASSERT(db->db_hash_next == NULL); ASSERT(db->db_blkptr == NULL); ASSERT(db->db_data_pending == NULL); ASSERT(!multilist_link_active(&db->db_cache_link)); kmem_cache_free(dbuf_kmem_cache, db); arc_space_return(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); /* * 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); } /* * Note: While bpp will always be updated if the function returns success, * parentp will not be updated if the dnode does not have dn_dbuf filled in; * this happens when the dnode is the meta-dnode, or a userused or groupused * object. */ static int dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse, dmu_buf_impl_t **parentp, blkptr_t **bpp) { *parentp = NULL; *bpp = NULL; ASSERT(blkid != DMU_BONUS_BLKID); if (blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_have_spill && (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) *bpp = &dn->dn_phys->dn_spill; else *bpp = NULL; dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; mutex_exit(&dn->dn_mtx); return (0); } int nlevels = (dn->dn_phys->dn_nlevels == 0) ? 1 : dn->dn_phys->dn_nlevels; int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(level * epbs, <, 64); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); /* * This assertion shouldn't trip as long as the max indirect block size * is less than 1M. The reason for this is that up to that point, * the number of levels required to address an entire object with blocks * of size SPA_MINBLOCKSIZE satisfies nlevels * epbs + 1 <= 64. In * other words, if N * epbs + 1 > 64, then if (N-1) * epbs + 1 > 55 * (i.e. we can address the entire object), objects will all use at most * N-1 levels and the assertion won't overflow. However, once epbs is * 13, 4 * 13 + 1 = 53, but 5 * 13 + 1 = 66. Then, 4 levels will not be * enough to address an entire object, so objects will have 5 levels, * but then this assertion will overflow. * * All this is to say that if we ever increase DN_MAX_INDBLKSHIFT, we * need to redo this logic to handle overflows. */ ASSERT(level >= nlevels || ((nlevels - level - 1) * epbs) + highbit64(dn->dn_phys->dn_nblkptr) <= 64); if (level >= nlevels || blkid >= ((uint64_t)dn->dn_phys->dn_nblkptr << ((nlevels - level - 1) * epbs)) || (fail_sparse && blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) { /* the buffer has no parent yet */ 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, FALSE, NULL, parentp); if (err) return (err); err = dbuf_read(*parentp, NULL, (DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH | DB_RF_CANFAIL)); if (err) { dbuf_rele(*parentp, NULL); *parentp = NULL; return (err); } *bpp = ((blkptr_t *)(*parentp)->db.db_data) + (blkid & ((1ULL << epbs) - 1)); if (blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs))) ASSERT(BP_IS_HOLE(*bpp)); return (0); } else { /* the block is referenced from the dnode */ ASSERT3U(level, ==, nlevels-1); ASSERT(dn->dn_phys->dn_nblkptr == 0 || blkid < dn->dn_phys->dn_nblkptr); if (dn->dn_dbuf) { dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; } *bpp = &dn->dn_phys->dn_blkptr[blkid]; return (0); } } static dmu_buf_impl_t * dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid, dmu_buf_impl_t *parent, blkptr_t *blkptr) { objset_t *os = dn->dn_objset; dmu_buf_impl_t *db, *odb; ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_type != DMU_OT_NONE); db = kmem_cache_alloc(dbuf_kmem_cache, KM_SLEEP); db->db_objset = os; db->db.db_object = dn->dn_object; db->db_level = level; db->db_blkid = blkid; db->db_last_dirty = NULL; db->db_dirtycnt = 0; db->db_dnode_handle = dn->dn_handle; db->db_parent = parent; db->db_blkptr = blkptr; db->db_user = NULL; db->db_user_immediate_evict = FALSE; db->db_freed_in_flight = FALSE; db->db_pending_evict = FALSE; if (blkid == DMU_BONUS_BLKID) { ASSERT3P(parent, ==, dn->dn_dbuf); db->db.db_size = DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); db->db.db_offset = DMU_BONUS_BLKID; db->db_state = DB_UNCACHED; /* the bonus dbuf is not placed in the hash table */ arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); return (db); } else if (blkid == DMU_SPILL_BLKID) { db->db.db_size = (blkptr != NULL) ? BP_GET_LSIZE(blkptr) : SPA_MINBLOCKSIZE; db->db.db_offset = 0; } else { int blocksize = db->db_level ? 1 << dn->dn_indblkshift : dn->dn_datablksz; db->db.db_size = blocksize; db->db.db_offset = db->db_blkid * blocksize; } /* * Hold the dn_dbufs_mtx while we get the new dbuf * in the hash table *and* added to the dbufs list. * This prevents a possible deadlock with someone * trying to look up this dbuf before its added to the * dn_dbufs list. */ mutex_enter(&dn->dn_dbufs_mtx); db->db_state = DB_EVICTING; if ((odb = dbuf_hash_insert(db)) != NULL) { /* someone else inserted it first */ kmem_cache_free(dbuf_kmem_cache, db); mutex_exit(&dn->dn_dbufs_mtx); return (odb); } avl_add(&dn->dn_dbufs, db); db->db_state = DB_UNCACHED; mutex_exit(&dn->dn_dbufs_mtx); arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); if (parent && parent != dn->dn_dbuf) dbuf_add_ref(parent, db); ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || refcount_count(&dn->dn_holds) > 0); (void) refcount_add(&dn->dn_holds, db); atomic_inc_32(&dn->dn_dbufs_count); dprintf_dbuf(db, "db=%p\n", db); return (db); } typedef struct dbuf_prefetch_arg { spa_t *dpa_spa; /* The spa to issue the prefetch in. */ zbookmark_phys_t dpa_zb; /* The target block to prefetch. */ int dpa_epbs; /* Entries (blkptr_t's) Per Block Shift. */ int dpa_curlevel; /* The current level that we're reading */ dnode_t *dpa_dnode; /* The dnode associated with the prefetch */ zio_priority_t dpa_prio; /* The priority I/Os should be issued at. */ zio_t *dpa_zio; /* The parent zio_t for all prefetches. */ arc_flags_t dpa_aflags; /* Flags to pass to the final prefetch. */ } dbuf_prefetch_arg_t; /* * Actually issue the prefetch read for the block given. */ static void dbuf_issue_final_prefetch(dbuf_prefetch_arg_t *dpa, blkptr_t *bp) { if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) return; arc_flags_t aflags = dpa->dpa_aflags | ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); ASSERT3U(dpa->dpa_curlevel, ==, dpa->dpa_zb.zb_level); ASSERT(dpa->dpa_zio != NULL); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, NULL, NULL, dpa->dpa_prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, &dpa->dpa_zb); } /* * Called when an indirect block above our prefetch target is read in. This * will either read in the next indirect block down the tree or issue the actual * prefetch if the next block down is our target. */ static void dbuf_prefetch_indirect_done(zio_t *zio, arc_buf_t *abuf, void *private) { dbuf_prefetch_arg_t *dpa = private; ASSERT3S(dpa->dpa_zb.zb_level, <, dpa->dpa_curlevel); ASSERT3S(dpa->dpa_curlevel, >, 0); /* * The dpa_dnode is only valid if we are called with a NULL * zio. This indicates that the arc_read() returned without * first calling zio_read() to issue a physical read. Once * a physical read is made the dpa_dnode must be invalidated * as the locks guarding it may have been dropped. If the * dpa_dnode is still valid, then we want to add it to the dbuf * cache. To do so, we must hold the dbuf associated with the block * we just prefetched, read its contents so that we associate it * with an arc_buf_t, and then release it. */ if (zio != NULL) { ASSERT3S(BP_GET_LEVEL(zio->io_bp), ==, dpa->dpa_curlevel); if (zio->io_flags & ZIO_FLAG_RAW) { ASSERT3U(BP_GET_PSIZE(zio->io_bp), ==, zio->io_size); } else { ASSERT3U(BP_GET_LSIZE(zio->io_bp), ==, zio->io_size); } ASSERT3P(zio->io_spa, ==, dpa->dpa_spa); dpa->dpa_dnode = NULL; } else if (dpa->dpa_dnode != NULL) { uint64_t curblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); dmu_buf_impl_t *db = dbuf_hold_level(dpa->dpa_dnode, dpa->dpa_curlevel, curblkid, FTAG); (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_HAVESTRUCT); dbuf_rele(db, FTAG); } dpa->dpa_curlevel--; uint64_t nextblkid = dpa->dpa_zb.zb_blkid >> (dpa->dpa_epbs * (dpa->dpa_curlevel - dpa->dpa_zb.zb_level)); blkptr_t *bp = ((blkptr_t *)abuf->b_data) + P2PHASE(nextblkid, 1ULL << dpa->dpa_epbs); if (BP_IS_HOLE(bp) || (zio != NULL && zio->io_error != 0)) { kmem_free(dpa, sizeof (*dpa)); } else if (dpa->dpa_curlevel == dpa->dpa_zb.zb_level) { ASSERT3U(nextblkid, ==, dpa->dpa_zb.zb_blkid); dbuf_issue_final_prefetch(dpa, bp); kmem_free(dpa, sizeof (*dpa)); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; ASSERT3U(dpa->dpa_curlevel, ==, BP_GET_LEVEL(bp)); SET_BOOKMARK(&zb, dpa->dpa_zb.zb_objset, dpa->dpa_zb.zb_object, dpa->dpa_curlevel, nextblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, bp, dbuf_prefetch_indirect_done, dpa, dpa->dpa_prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } arc_buf_destroy(abuf, private); } /* * Issue prefetch reads for the given block on the given level. If the indirect * blocks above that block are not in memory, we will read them in * asynchronously. As a result, this call never blocks waiting for a read to * complete. */ void dbuf_prefetch(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio, arc_flags_t aflags) { blkptr_t bp; int epbs, nlevels, curlevel; uint64_t curblkid; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); if (blkid > dn->dn_maxblkid) return; if (dnode_block_freed(dn, blkid)) return; /* * This dnode hasn't been written to disk yet, so there's nothing to * prefetch. */ nlevels = dn->dn_phys->dn_nlevels; if (level >= nlevels || dn->dn_phys->dn_nblkptr == 0) return; epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; if (dn->dn_phys->dn_maxblkid < blkid << (epbs * level)) return; dmu_buf_impl_t *db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db != NULL) { mutex_exit(&db->db_mtx); /* * This dbuf already exists. It is either CACHED, or * (we assume) about to be read or filled. */ return; } /* * Find the closest ancestor (indirect block) of the target block * that is present in the cache. In this indirect block, we will * find the bp that is at curlevel, curblkid. */ curlevel = level; curblkid = blkid; while (curlevel < nlevels - 1) { int parent_level = curlevel + 1; uint64_t parent_blkid = curblkid >> epbs; dmu_buf_impl_t *db; if (dbuf_hold_impl(dn, parent_level, parent_blkid, FALSE, TRUE, FTAG, &db) == 0) { blkptr_t *bpp = db->db_buf->b_data; bp = bpp[P2PHASE(curblkid, 1 << epbs)]; dbuf_rele(db, FTAG); break; } curlevel = parent_level; curblkid = parent_blkid; } if (curlevel == nlevels - 1) { /* No cached indirect blocks found. */ ASSERT3U(curblkid, <, dn->dn_phys->dn_nblkptr); bp = dn->dn_phys->dn_blkptr[curblkid]; } if (BP_IS_HOLE(&bp)) return; ASSERT3U(curlevel, ==, BP_GET_LEVEL(&bp)); zio_t *pio = zio_root(dmu_objset_spa(dn->dn_objset), NULL, NULL, ZIO_FLAG_CANFAIL); dbuf_prefetch_arg_t *dpa = kmem_zalloc(sizeof (*dpa), KM_SLEEP); dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; SET_BOOKMARK(&dpa->dpa_zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, level, blkid); dpa->dpa_curlevel = curlevel; dpa->dpa_prio = prio; dpa->dpa_aflags = aflags; dpa->dpa_spa = dn->dn_objset->os_spa; dpa->dpa_dnode = dn; dpa->dpa_epbs = epbs; dpa->dpa_zio = pio; /* * If we have the indirect just above us, no need to do the asynchronous * prefetch chain; we'll just run the last step ourselves. If we're at * a higher level, though, we want to issue the prefetches for all the * indirect blocks asynchronously, so we can go on with whatever we were * doing. */ if (curlevel == level) { ASSERT3U(curblkid, ==, blkid); dbuf_issue_final_prefetch(dpa, &bp); kmem_free(dpa, sizeof (*dpa)); } else { arc_flags_t iter_aflags = ARC_FLAG_NOWAIT; zbookmark_phys_t zb; SET_BOOKMARK(&zb, ds != NULL ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, curlevel, curblkid); (void) arc_read(dpa->dpa_zio, dpa->dpa_spa, &bp, dbuf_prefetch_indirect_done, dpa, prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &iter_aflags, &zb); } /* * We use pio here instead of dpa_zio since it's possible that * dpa may have already been freed. */ zio_nowait(pio); } /* * 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, boolean_t fail_sparse, boolean_t fail_uncached, void *tag, dmu_buf_impl_t **dbp) { dmu_buf_impl_t *db, *parent = NULL; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT3U(dn->dn_nlevels, >, level); *dbp = NULL; top: /* dbuf_find() returns with db_mtx held */ db = dbuf_find(dn->dn_objset, dn->dn_object, level, blkid); if (db == NULL) { blkptr_t *bp = NULL; int err; if (fail_uncached) return (SET_ERROR(ENOENT)); 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 = SET_ERROR(ENOENT); if (err) { if (parent) dbuf_rele(parent, NULL); return (err); } } if (err && err != ENOENT) return (err); db = dbuf_create(dn, level, blkid, parent, bp); } if (fail_uncached && db->db_state != DB_CACHED) { mutex_exit(&db->db_mtx); return (SET_ERROR(ENOENT)); } if (db->db_buf != NULL) ASSERT3P(db->db.db_data, ==, db->db_buf->b_data); ASSERT(db->db_buf == NULL || arc_referenced(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_alloc_buf(dn->dn_objset->os_spa, db, type, db->db.db_size)); bcopy(dr->dt.dl.dr_data->b_data, db->db.db_data, db->db.db_size); } } if (multilist_link_active(&db->db_cache_link)) { ASSERT(refcount_is_zero(&db->db_holds)); multilist_remove(dbuf_cache, db); (void) refcount_remove_many(&dbuf_cache_size, db->db.db_size, db); } (void) refcount_add(&db->db_holds, tag); DBUF_VERIFY(db); mutex_exit(&db->db_mtx); /* NOTE: we can't rele the parent until after we drop the db_mtx */ if (parent) dbuf_rele(parent, NULL); ASSERT3P(DB_DNODE(db), ==, dn); ASSERT3U(db->db_blkid, ==, blkid); ASSERT3U(db->db_level, ==, level); *dbp = db; return (0); } dmu_buf_impl_t * dbuf_hold(dnode_t *dn, uint64_t blkid, void *tag) { return (dbuf_hold_level(dn, 0, blkid, tag)); } dmu_buf_impl_t * dbuf_hold_level(dnode_t *dn, int level, uint64_t blkid, void *tag) { dmu_buf_impl_t *db; int err = dbuf_hold_impl(dn, level, blkid, FALSE, FALSE, tag, &db); return (err ? NULL : db); } void dbuf_create_bonus(dnode_t *dn) { ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_bonus == NULL); dn->dn_bonus = dbuf_create(dn, 0, DMU_BONUS_BLKID, dn->dn_dbuf, NULL); } int 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 (SET_ERROR(ENOTSUP)); if (blksz == 0) blksz = SPA_MINBLOCKSIZE; ASSERT3U(blksz, <=, spa_maxblocksize(dmu_objset_spa(db->db_objset))); blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE); DB_DNODE_ENTER(db); dn = DB_DNODE(db); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); dbuf_new_size(db, blksz, tx); rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); return (0); } void dbuf_rm_spill(dnode_t *dn, dmu_tx_t *tx) { dbuf_free_range(dn, DMU_SPILL_BLKID, DMU_SPILL_BLKID, tx); } #pragma weak dmu_buf_add_ref = dbuf_add_ref void dbuf_add_ref(dmu_buf_impl_t *db, void *tag) { int64_t holds = refcount_add(&db->db_holds, tag); ASSERT3S(holds, >, 1); } #pragma weak dmu_buf_try_add_ref = dbuf_try_add_ref boolean_t dbuf_try_add_ref(dmu_buf_t *db_fake, objset_t *os, uint64_t obj, uint64_t blkid, void *tag) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dmu_buf_impl_t *found_db; boolean_t result = B_FALSE; if (db->db_blkid == DMU_BONUS_BLKID) found_db = dbuf_find_bonus(os, obj); else found_db = dbuf_find(os, obj, 0, blkid); if (found_db != NULL) { if (db == found_db && dbuf_refcount(db) > db->db_dirtycnt) { (void) refcount_add(&db->db_holds, tag); result = B_TRUE; } mutex_exit(&db->db_mtx); } return (result); } /* * 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. */ void dbuf_rele(dmu_buf_impl_t *db, void *tag) { mutex_enter(&db->db_mtx); dbuf_rele_and_unlock(db, tag); } void dmu_buf_rele(dmu_buf_t *db, void *tag) { dbuf_rele((dmu_buf_impl_t *)db, tag); } /* * dbuf_rele() for an already-locked dbuf. This is necessary to allow * db_dirtycnt and db_holds to be updated atomically. */ void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag) { int64_t holds; 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); /* * We can't freeze indirects if there is a possibility that they * may be modified in the current syncing context. */ if (db->db_buf != NULL && holds == (db->db_level == 0 ? db->db_dirtycnt : 0)) { arc_buf_freeze(db->db_buf); } if (holds == db->db_dirtycnt && db->db_level == 0 && db->db_user_immediate_evict) dbuf_evict_user(db); if (holds == 0) { if (db->db_blkid == DMU_BONUS_BLKID) { dnode_t *dn; boolean_t evict_dbuf = db->db_pending_evict; /* * If the dnode moves here, we cannot cross this * barrier until the move completes. */ DB_DNODE_ENTER(db); dn = DB_DNODE(db); atomic_dec_32(&dn->dn_dbufs_count); /* * Decrementing the dbuf count means that the bonus * buffer's dnode hold is no longer discounted in * dnode_move(). The dnode cannot move until after * the dnode_rele() below. */ DB_DNODE_EXIT(db); /* * Do not reference db after its lock is dropped. * Another thread may evict it. */ mutex_exit(&db->db_mtx); if (evict_dbuf) dnode_evict_bonus(dn); dnode_rele(dn, db); } else if (db->db_buf == NULL) { /* * This is a special case: we never associated this * dbuf with any data allocated from the ARC. */ ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); dbuf_destroy(db); } else if (arc_released(db->db_buf)) { /* * This dbuf has anonymous data associated with it. */ dbuf_destroy(db); } else { boolean_t do_arc_evict = B_FALSE; blkptr_t bp; spa_t *spa = dmu_objset_spa(db->db_objset); if (!DBUF_IS_CACHEABLE(db) && db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr) && !BP_IS_EMBEDDED(db->db_blkptr)) { do_arc_evict = B_TRUE; bp = *db->db_blkptr; } if (!DBUF_IS_CACHEABLE(db) || db->db_pending_evict) { dbuf_destroy(db); } else if (!multilist_link_active(&db->db_cache_link)) { multilist_insert(dbuf_cache, db); (void) refcount_add_many(&dbuf_cache_size, db->db.db_size, db); mutex_exit(&db->db_mtx); dbuf_evict_notify(); } if (do_arc_evict) arc_freed(spa, &bp); } } else { mutex_exit(&db->db_mtx); } } #pragma weak dmu_buf_refcount = dbuf_refcount uint64_t dbuf_refcount(dmu_buf_impl_t *db) { return (refcount_count(&db->db_holds)); } void * dmu_buf_replace_user(dmu_buf_t *db_fake, dmu_buf_user_t *old_user, dmu_buf_user_t *new_user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; mutex_enter(&db->db_mtx); dbuf_verify_user(db, DBVU_NOT_EVICTING); if (db->db_user == old_user) db->db_user = new_user; else old_user = db->db_user; dbuf_verify_user(db, DBVU_NOT_EVICTING); mutex_exit(&db->db_mtx); return (old_user); } void * dmu_buf_set_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, NULL, user)); } void * dmu_buf_set_user_ie(dmu_buf_t *db_fake, dmu_buf_user_t *user) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_user_immediate_evict = TRUE; return (dmu_buf_set_user(db_fake, user)); } void * dmu_buf_remove_user(dmu_buf_t *db_fake, dmu_buf_user_t *user) { return (dmu_buf_replace_user(db_fake, user, NULL)); } void * dmu_buf_get_user(dmu_buf_t *db_fake) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dbuf_verify_user(db, DBVU_NOT_EVICTING); return (db->db_user); } void dmu_buf_user_evict_wait() { taskq_wait(dbu_evict_taskq); } blkptr_t * dmu_buf_get_blkptr(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_blkptr); } objset_t * dmu_buf_get_objset(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_objset); } dnode_t * dmu_buf_dnode_enter(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_ENTER(dbi); return (DB_DNODE(dbi)); } void dmu_buf_dnode_exit(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; DB_DNODE_EXIT(dbi); } static void dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db) { /* ASSERT(dmu_tx_is_syncing(tx) */ ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_blkptr != NULL) return; if (db->db_blkid == DMU_SPILL_BLKID) { db->db_blkptr = &dn->dn_phys->dn_spill; BP_ZERO(db->db_blkptr); return; } if (db->db_level == dn->dn_phys->dn_nlevels-1) { /* * This buffer was allocated at a time when there was * no available blkptrs from the dnode, or it was * inappropriate to hook it in (i.e., nlevels mis-match). */ ASSERT(db->db_blkid < dn->dn_phys->dn_nblkptr); ASSERT(db->db_parent == NULL); db->db_parent = dn->dn_dbuf; db->db_blkptr = &dn->dn_phys->dn_blkptr[db->db_blkid]; DBUF_VERIFY(db); } else { dmu_buf_impl_t *parent = db->db_parent; int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT(dn->dn_phys->dn_nlevels > 1); if (parent == NULL) { mutex_exit(&db->db_mtx); rw_enter(&dn->dn_struct_rwlock, RW_READER); parent = dbuf_hold_level(dn, db->db_level + 1, db->db_blkid >> epbs, db); rw_exit(&dn->dn_struct_rwlock); mutex_enter(&db->db_mtx); db->db_parent = parent; } db->db_blkptr = (blkptr_t *)parent->db.db_data + (db->db_blkid & ((1ULL << epbs) - 1)); DBUF_VERIFY(db); } } 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; zio_t *zio; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); 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); mutex_enter(&db->db_mtx); } ASSERT3U(db->db_state, ==, DB_CACHED); ASSERT(db->db_buf != NULL); 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_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); dbuf_write(dr, db->db_buf, tx); zio = dr->dr_zio; mutex_enter(&dr->dt.di.dr_mtx); dbuf_sync_list(&dr->dt.di.dr_children, db->db_level - 1, tx); ASSERT(list_head(&dr->dt.di.dr_children) == NULL); mutex_exit(&dr->dt.di.dr_mtx); zio_nowait(zio); } 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; objset_t *os; uint64_t txg = tx->tx_txg; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); /* * To be synced, we must be dirtied. But we * might have been freed after the dirty. */ if (db->db_state == DB_UNCACHED) { /* This buffer has been freed since it was dirtied */ ASSERT(db->db.db_data == NULL); } else if (db->db_state == DB_FILL) { /* This buffer was freed and is now being re-filled */ ASSERT(db->db.db_data != dr->dt.dl.dr_data); } else { ASSERT(db->db_state == DB_CACHED || db->db_state == DB_NOFILL); } 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; mutex_exit(&dn->dn_mtx); } /* * If this is a bonus buffer, simply copy the bonus data into the * dnode. It will be written out when the dnode is synced (and it * will be synced, since it must have been dirty for dbuf_sync to * be called). */ if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_dirty_record_t **drp; ASSERT(*datap != NULL); 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); } db->db_data_pending = NULL; drp = &db->db_last_dirty; while (*drp != dr) drp = &(*drp)->dr_next; ASSERT(dr->dr_next == NULL); ASSERT(dr->dr_dbuf == db); *drp = dr->dr_next; if (dr->dr_dbuf->db_level != 0) { list_destroy(&dr->dt.di.dr_children); mutex_destroy(&dr->dt.di.dr_mtx); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); 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 * don't check the dr_override_state until we have returned from * dbuf_check_blkptr. */ dbuf_check_blkptr(dn, db); /* * 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) { ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); cv_wait(&db->db_changed, &db->db_mtx); ASSERT(dr->dt.dl.dr_override_state != DR_NOT_OVERRIDDEN); } if (db->db_state != DB_NOFILL && dn->dn_object != DMU_META_DNODE_OBJECT && refcount_count(&db->db_holds) > 1 && dr->dt.dl.dr_override_state != DR_OVERRIDDEN && *datap == db->db_buf) { /* * If this buffer is currently "in use" (i.e., there * are active holds and db_data still references it), * then make a copy before we start the write so that * any modifications from the open txg will not leak * into this write. * * NOTE: this copy does not need to be made for * objects only modified in the syncing context (e.g. * DNONE_DNODE blocks). */ int psize = arc_buf_size(*datap); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); enum zio_compress compress_type = arc_get_compression(*datap); if (compress_type == ZIO_COMPRESS_OFF) { *datap = arc_alloc_buf(os->os_spa, db, type, psize); } else { ASSERT3U(type, ==, ARC_BUFC_DATA); int lsize = arc_buf_lsize(*datap); *datap = arc_alloc_compressed_buf(os->os_spa, db, psize, lsize, compress_type); } bcopy(db->db.db_data, (*datap)->b_data, psize); } db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, *datap, tx); ASSERT(!list_link_active(&dr->dr_dirty_node)); if (dn->dn_object == DMU_META_DNODE_OBJECT) { list_insert_tail(&dn->dn_dirty_records[txg&TXG_MASK], dr); 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_sync_list(list_t *list, int level, dmu_tx_t *tx) { dbuf_dirty_record_t *dr; while (dr = list_head(list)) { if (dr->dr_zio != NULL) { /* * If we find an already initialized zio then we * are processing the meta-dnode, and we have finished. * The dbufs for all dnodes are put back on the list * during processing, so that we can zio_wait() * these IOs after initiating all child IOs. */ ASSERT3U(dr->dr_dbuf->db.db_object, ==, DMU_META_DNODE_OBJECT); break; } if (dr->dr_dbuf->db_blkid != DMU_BONUS_BLKID && dr->dr_dbuf->db_blkid != DMU_SPILL_BLKID) { VERIFY3U(dr->dr_dbuf->db_level, ==, level); } list_remove(list, dr); if (dr->dr_dbuf->db_level > 0) dbuf_sync_indirect(dr, tx); else dbuf_sync_leaf(dr, tx); } } /* ARGSUSED */ static void 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; spa_t *spa = zio->io_spa; int64_t delta; uint64_t fill = 0; int i; ASSERT3P(db->db_blkptr, !=, NULL); ASSERT3P(&db->db_data_pending->dr_bp_copy, ==, 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->blk_birth != 0) { ASSERT((db->db_blkid != DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_type) || (db->db_blkid == DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_bonustype) || BP_IS_EMBEDDED(bp)); ASSERT(BP_GET_LEVEL(bp) == db->db_level); } mutex_enter(&db->db_mtx); #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(bp)) && db->db_blkptr == &dn->dn_phys->dn_spill); } #endif if (db->db_level == 0) { mutex_enter(&dn->dn_mtx); if (db->db_blkid > dn->dn_phys->dn_maxblkid && db->db_blkid != DMU_SPILL_BLKID) dn->dn_phys->dn_maxblkid = db->db_blkid; mutex_exit(&dn->dn_mtx); if (dn->dn_type == DMU_OT_DNODE) { dnode_phys_t *dnp = db->db.db_data; for (i = db->db.db_size >> DNODE_SHIFT; i > 0; i--, dnp++) { if (dnp->dn_type != DMU_OT_NONE) fill++; } } else { if (BP_IS_HOLE(bp)) { fill = 0; } else { fill = 1; } } } else { blkptr_t *ibp = db->db.db_data; ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); for (i = db->db.db_size >> SPA_BLKPTRSHIFT; i > 0; i--, ibp++) { if (BP_IS_HOLE(ibp)) continue; fill += BP_GET_FILL(ibp); } } DB_DNODE_EXIT(db); if (!BP_IS_EMBEDDED(bp)) bp->blk_fill = fill; mutex_exit(&db->db_mtx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); *db->db_blkptr = *bp; rw_exit(&dn->dn_struct_rwlock); } /* ARGSUSED */ /* * This function gets called just prior to running through the compression * stage of the zio pipeline. If we're an indirect block comprised of only * holes, then we want this indirect to be compressed away to a hole. In * order to do that we must zero out any information about the holes that * this indirect points to prior to before we try to compress it. */ static void dbuf_write_children_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp; unsigned int epbs, i; ASSERT3U(db->db_level, >, 0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(epbs, <, 31); /* Determine if all our children are holes */ for (i = 0, bp = db->db.db_data; i < 1 << epbs; i++, bp++) { if (!BP_IS_HOLE(bp)) break; } /* * If all the children are holes, then zero them all out so that * we may get compressed away. */ if (i == 1 << epbs) { /* * We only found holes. Grab the rwlock to prevent * anybody from reading the blocks we're about to * zero out. */ rw_enter(&dn->dn_struct_rwlock, RW_WRITER); bzero(db->db.db_data, db->db.db_size); rw_exit(&dn->dn_struct_rwlock); } DB_DNODE_EXIT(db); } /* * 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) { dmu_buf_impl_t *db = vdb; blkptr_t *bp_orig = &zio->io_bp_orig; blkptr_t *bp = db->db_blkptr; objset_t *os = db->db_objset; dmu_tx_t *tx = os->os_synctx; dbuf_dirty_record_t **drp, *dr; ASSERT0(zio->io_error); ASSERT(db->db_blkptr == bp); /* * 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 { dsl_dataset_t *ds = os->os_dsl_dataset; (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, bp, tx); } mutex_enter(&db->db_mtx); DBUF_VERIFY(db); drp = &db->db_last_dirty; while ((dr = *drp) != db->db_data_pending) drp = &dr->dr_next; ASSERT(!list_link_active(&dr->dr_dirty_node)); ASSERT(dr->dr_dbuf == db); ASSERT(dr->dr_next == NULL); *drp = dr->dr_next; #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { 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 if (db->db_level == 0) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); if (db->db_state != DB_NOFILL) { if (dr->dt.dl.dr_data != db->db_buf) arc_buf_destroy(dr->dt.dl.dr_data, 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; ASSERT3U(db->db_blkid, <=, dn->dn_phys->dn_maxblkid >> (db->db_level * epbs)); ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, db->db.db_size); } DB_DNODE_EXIT(db); mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); cv_broadcast(&db->db_changed); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; db->db_data_pending = NULL; dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg); } static void dbuf_write_nofill_ready(zio_t *zio) { dbuf_write_ready(zio, NULL, zio->io_private); } static void dbuf_write_nofill_done(zio_t *zio) { dbuf_write_done(zio, NULL, zio->io_private); } static void dbuf_write_override_ready(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; dbuf_write_ready(zio, NULL, db); } static void dbuf_write_override_done(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *obp = &dr->dt.dl.dr_overridden_by; mutex_enter(&db->db_mtx); if (!BP_EQUAL(zio->io_bp, obp)) { if (!BP_IS_HOLE(obp)) dsl_free(spa_get_dsl(zio->io_spa), zio->io_txg, obp); arc_release(dr->dt.dl.dr_data, db); } mutex_exit(&db->db_mtx); 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) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn; objset_t *os; dmu_buf_impl_t *parent = db->db_parent; uint64_t txg = tx->tx_txg; zbookmark_phys_t zb; zio_prop_t zp; zio_t *zio; int wp_flag = 0; ASSERT(dmu_tx_is_syncing(tx)); 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) { /* * Private object buffers are released here rather * than in dbuf_dirty() since they are only modified * in the syncing context and we don't want the * overhead of making multiple copies of the data. */ if (BP_IS_HOLE(db->db_blkptr)) { arc_buf_thaw(data); } else { dbuf_release_bp(db); } } } 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)); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); zio = dn->dn_zio; } ASSERT(db->db_level == 0 || data == db->db_buf); ASSERT3U(db->db_blkptr->blk_birth, <=, txg); ASSERT(zio); SET_BOOKMARK(&zb, os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : DMU_META_OBJSET, db->db.db_object, db->db_level, db->db_blkid); if (db->db_blkid == DMU_SPILL_BLKID) wp_flag = WP_SPILL; wp_flag |= (db->db_state == DB_NOFILL) ? WP_NOFILL : 0; dmu_write_policy(os, dn, db->db_level, wp_flag, (data != NULL && arc_get_compression(data) != ZIO_COMPRESS_OFF) ? arc_get_compression(data) : ZIO_COMPRESS_INHERIT, &zp); DB_DNODE_EXIT(db); /* * We copy the blkptr now (rather than when we instantiate the dirty * record), because its value can change between open context and * syncing context. We do not need to hold dn_struct_rwlock to read * db_blkptr because we are in syncing context. */ dr->dr_bp_copy = *db->db_blkptr; if (db->db_level == 0 && dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { /* * The BP for this block has been provided by open context * (by dmu_sync() or dmu_buf_write_embedded()). */ void *contents = (data != NULL) ? data->b_data : NULL; dr->dr_zio = zio_write(zio, os->os_spa, txg, &dr->dr_bp_copy, contents, db->db.db_size, db->db.db_size, &zp, dbuf_write_override_ready, NULL, 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_nopwrite); mutex_exit(&db->db_mtx); } else if (db->db_state == DB_NOFILL) { ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF || zp.zp_checksum == ZIO_CHECKSUM_NOPARITY); dr->dr_zio = zio_write(zio, os->os_spa, txg, &dr->dr_bp_copy, NULL, db->db.db_size, db->db.db_size, &zp, dbuf_write_nofill_ready, NULL, NULL, dbuf_write_nofill_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb); } else { ASSERT(arc_released(data)); /* * For indirect blocks, we want to setup the children * ready callback so that we can properly handle an indirect * block that only contains holes. */ arc_done_func_t *children_ready_cb = NULL; if (db->db_level != 0) children_ready_cb = dbuf_write_children_ready; dr->dr_zio = arc_write(zio, os->os_spa, txg, &dr->dr_bp_copy, data, DBUF_IS_L2CACHEABLE(db), &zp, dbuf_write_ready, children_ready_cb, dbuf_write_physdone, dbuf_write_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); } } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_scan.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_scan.c (revision 318934) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_scan.c (revision 318935) @@ -1,1922 +1,1924 @@ /* * 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 (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2015 by Delphix. All rights reserved. * Copyright 2016 Gary Mills */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #endif typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_phys_t *); static scan_cb_t dsl_scan_scrub_cb; static void dsl_scan_cancel_sync(void *, dmu_tx_t *); static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *); static boolean_t dsl_scan_restarting(dsl_scan_t *, dmu_tx_t *); unsigned int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */ unsigned int zfs_resilver_delay = 2; /* number of ticks to delay resilver */ unsigned int zfs_scrub_delay = 4; /* number of ticks to delay scrub */ unsigned int zfs_scan_idle = 50; /* idle window in clock ticks */ unsigned int zfs_scan_min_time_ms = 1000; /* min millisecs to scrub per txg */ unsigned int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */ unsigned int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */ boolean_t zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */ boolean_t zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */ SYSCTL_DECL(_vfs_zfs); SYSCTL_UINT(_vfs_zfs, OID_AUTO, top_maxinflight, CTLFLAG_RWTUN, &zfs_top_maxinflight, 0, "Maximum I/Os per top-level vdev"); SYSCTL_UINT(_vfs_zfs, OID_AUTO, resilver_delay, CTLFLAG_RWTUN, &zfs_resilver_delay, 0, "Number of ticks to delay resilver"); SYSCTL_UINT(_vfs_zfs, OID_AUTO, scrub_delay, CTLFLAG_RWTUN, &zfs_scrub_delay, 0, "Number of ticks to delay scrub"); SYSCTL_UINT(_vfs_zfs, OID_AUTO, scan_idle, CTLFLAG_RWTUN, &zfs_scan_idle, 0, "Idle scan window in clock ticks"); SYSCTL_UINT(_vfs_zfs, OID_AUTO, scan_min_time_ms, CTLFLAG_RWTUN, &zfs_scan_min_time_ms, 0, "Min millisecs to scrub per txg"); SYSCTL_UINT(_vfs_zfs, OID_AUTO, free_min_time_ms, CTLFLAG_RWTUN, &zfs_free_min_time_ms, 0, "Min millisecs to free per txg"); SYSCTL_UINT(_vfs_zfs, OID_AUTO, resilver_min_time_ms, CTLFLAG_RWTUN, &zfs_resilver_min_time_ms, 0, "Min millisecs to resilver per txg"); SYSCTL_INT(_vfs_zfs, OID_AUTO, no_scrub_io, CTLFLAG_RWTUN, &zfs_no_scrub_io, 0, "Disable scrub I/O"); SYSCTL_INT(_vfs_zfs, OID_AUTO, no_scrub_prefetch, CTLFLAG_RWTUN, &zfs_no_scrub_prefetch, 0, "Disable scrub prefetching"); enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE; /* max number of blocks to free in a single TXG */ uint64_t zfs_free_max_blocks = UINT64_MAX; SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, free_max_blocks, CTLFLAG_RWTUN, &zfs_free_max_blocks, 0, "Maximum number of blocks to free in one TXG"); #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \ ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \ (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER) extern int zfs_txg_timeout; /* * Enable/disable the processing of the free_bpobj object. */ boolean_t zfs_free_bpobj_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs, OID_AUTO, free_bpobj_enabled, CTLFLAG_RWTUN, &zfs_free_bpobj_enabled, 0, "Enable free_bpobj processing"); /* the order has to match pool_scan_type */ static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = { NULL, dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */ dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */ }; int dsl_scan_init(dsl_pool_t *dp, uint64_t txg) { int err; dsl_scan_t *scn; spa_t *spa = dp->dp_spa; uint64_t f; scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP); scn->scn_dp = dp; /* * It's possible that we're resuming a scan after a reboot so * make sure that the scan_async_destroying flag is initialized * appropriately. */ ASSERT(!scn->scn_async_destroying); scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY); err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_func", sizeof (uint64_t), 1, &f); if (err == 0) { /* * There was an old-style scrub in progress. Restart a * new-style scrub from the beginning. */ scn->scn_restart_txg = txg; zfs_dbgmsg("old-style scrub was in progress; " "restarting new-style scrub in txg %llu", scn->scn_restart_txg); /* * Load the queue obj from the old location so that it * can be freed by dsl_scan_done(). */ (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_queue", sizeof (uint64_t), 1, &scn->scn_phys.scn_queue_obj); } else { err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys); if (err == ENOENT) return (0); else if (err) return (err); if (scn->scn_phys.scn_state == DSS_SCANNING && spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) { /* * A new-type scrub was in progress on an old * pool, and the pool was accessed by old * software. Restart from the beginning, since * the old software may have changed the pool in * the meantime. */ scn->scn_restart_txg = txg; zfs_dbgmsg("new-style scrub was modified " "by old software; restarting in txg %llu", scn->scn_restart_txg); } } spa_scan_stat_init(spa); return (0); } void dsl_scan_fini(dsl_pool_t *dp) { if (dp->dp_scan) { kmem_free(dp->dp_scan, sizeof (dsl_scan_t)); dp->dp_scan = NULL; } } /* ARGSUSED */ static int dsl_scan_setup_check(void *arg, dmu_tx_t *tx) { dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; if (scn->scn_phys.scn_state == DSS_SCANNING) return (SET_ERROR(EBUSY)); return (0); } static void dsl_scan_setup_sync(void *arg, dmu_tx_t *tx) { dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; pool_scan_func_t *funcp = arg; dmu_object_type_t ot = 0; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; ASSERT(scn->scn_phys.scn_state != DSS_SCANNING); ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS); bzero(&scn->scn_phys, sizeof (scn->scn_phys)); scn->scn_phys.scn_func = *funcp; scn->scn_phys.scn_state = DSS_SCANNING; scn->scn_phys.scn_min_txg = 0; scn->scn_phys.scn_max_txg = tx->tx_txg; scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */ scn->scn_phys.scn_start_time = gethrestime_sec(); scn->scn_phys.scn_errors = 0; scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc; scn->scn_restart_txg = 0; scn->scn_done_txg = 0; spa_scan_stat_init(spa); if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max; /* rewrite all disk labels */ vdev_config_dirty(spa->spa_root_vdev); if (vdev_resilver_needed(spa->spa_root_vdev, &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) { spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START); } else { spa_event_notify(spa, NULL, ESC_ZFS_SCRUB_START); } spa->spa_scrub_started = B_TRUE; /* * If this is an incremental scrub, limit the DDT scrub phase * to just the auto-ditto class (for correctness); the rest * of the scrub should go faster using top-down pruning. */ if (scn->scn_phys.scn_min_txg > TXG_INITIAL) scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO; } /* back to the generic stuff */ if (dp->dp_blkstats == NULL) { dp->dp_blkstats = kmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP); } bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) ot = DMU_OT_ZAP_OTHER; scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset, ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx); dsl_scan_sync_state(scn, tx); spa_history_log_internal(spa, "scan setup", tx, "func=%u mintxg=%llu maxtxg=%llu", *funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg); } /* ARGSUSED */ static void dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx) { static const char *old_names[] = { "scrub_bookmark", "scrub_ddt_bookmark", "scrub_ddt_class_max", "scrub_queue", "scrub_min_txg", "scrub_max_txg", "scrub_func", "scrub_errors", NULL }; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; int i; /* Remove any remnants of an old-style scrub. */ for (i = 0; old_names[i]; i++) { (void) zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx); } if (scn->scn_phys.scn_queue_obj != 0) { VERIFY(0 == dmu_object_free(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, tx)); scn->scn_phys.scn_queue_obj = 0; } /* * If we were "restarted" from a stopped state, don't bother * with anything else. */ if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (complete) scn->scn_phys.scn_state = DSS_FINISHED; else scn->scn_phys.scn_state = DSS_CANCELED; if (dsl_scan_restarting(scn, tx)) spa_history_log_internal(spa, "scan aborted, restarting", tx, "errors=%llu", spa_get_errlog_size(spa)); else if (!complete) spa_history_log_internal(spa, "scan cancelled", tx, "errors=%llu", spa_get_errlog_size(spa)); else spa_history_log_internal(spa, "scan done", tx, "errors=%llu", spa_get_errlog_size(spa)); if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight > 0) { cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); } mutex_exit(&spa->spa_scrub_lock); spa->spa_scrub_started = B_FALSE; spa->spa_scrub_active = B_FALSE; /* * If the scrub/resilver completed, update all DTLs to * reflect this. Whether it succeeded or not, vacate * all temporary scrub DTLs. */ vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg, complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE); if (complete) { spa_event_notify(spa, NULL, scn->scn_phys.scn_min_txg ? ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH); } spa_errlog_rotate(spa); /* * We may have finished replacing a device. * Let the async thread assess this and handle the detach. */ spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); } scn->scn_phys.scn_end_time = gethrestime_sec(); } /* ARGSUSED */ static int dsl_scan_cancel_check(void *arg, dmu_tx_t *tx) { dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; if (scn->scn_phys.scn_state != DSS_SCANNING) return (SET_ERROR(ENOENT)); return (0); } /* ARGSUSED */ static void dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx) { dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; dsl_scan_done(scn, B_FALSE, tx); dsl_scan_sync_state(scn, tx); } int dsl_scan_cancel(dsl_pool_t *dp) { return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check, dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED)); } static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb, dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx); static void dsl_scan_visitdnode(dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx); void dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp) { zio_free(dp->dp_spa, txg, bp); } void dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp) { ASSERT(dsl_pool_sync_context(dp)); zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, BP_GET_PSIZE(bpp), pio->io_flags)); } static uint64_t dsl_scan_ds_maxtxg(dsl_dataset_t *ds) { uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg; if (ds->ds_is_snapshot) return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg)); return (smt); } static void dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx) { VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys, tx)); } extern int zfs_vdev_async_write_active_min_dirty_percent; static boolean_t dsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_phys_t *zb) { /* we never skip user/group accounting objects */ if (zb && (int64_t)zb->zb_object < 0) return (B_FALSE); if (scn->scn_pausing) return (B_TRUE); /* we're already pausing */ if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) return (B_FALSE); /* we're resuming */ /* We only know how to resume from level-0 blocks. */ if (zb && zb->zb_level != 0) return (B_FALSE); /* * We pause if: * - we have scanned for the maximum time: an entire txg * timeout (default 5 sec) * or * - we have scanned for at least the minimum time (default 1 sec * for scrub, 3 sec for resilver), and either we have sufficient * dirty data that we are starting to write more quickly * (default 30%), or someone is explicitly waiting for this txg * to complete. * or * - the spa is shutting down because this pool is being exported * or the machine is rebooting. */ int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scan_min_time_ms; uint64_t elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; int dirty_pct = scn->scn_dp->dp_dirty_total * 100 / zfs_dirty_data_max; if (elapsed_nanosecs / NANOSEC >= zfs_txg_timeout || (NSEC2MSEC(elapsed_nanosecs) > mintime && (txg_sync_waiting(scn->scn_dp) || dirty_pct >= zfs_vdev_async_write_active_min_dirty_percent)) || spa_shutting_down(scn->scn_dp->dp_spa)) { if (zb) { dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); scn->scn_phys.scn_bookmark = *zb; } dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n", (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor); scn->scn_pausing = B_TRUE; return (B_TRUE); } return (B_FALSE); } typedef struct zil_scan_arg { dsl_pool_t *zsa_dp; zil_header_t *zsa_zh; } zil_scan_arg_t; /* ARGSUSED */ static int dsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) { zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; zbookmark_phys_t zb; if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) return (0); /* * One block ("stubby") can be allocated a long time ago; we * want to visit that one because it has been allocated * (on-disk) even if it hasn't been claimed (even though for * scrub there's nothing to do to it). */ if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa)) return (0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); return (0); } /* ARGSUSED */ static int dsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) { if (lrc->lrc_txtype == TX_WRITE) { zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; lr_write_t *lr = (lr_write_t *)lrc; blkptr_t *bp = &lr->lr_blkptr; zbookmark_phys_t zb; if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) return (0); /* * birth can be < claim_txg if this record's txg is * already txg sync'ed (but this log block contains * other records that are not synced) */ if (claim_txg == 0 || bp->blk_birth < claim_txg) return (0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], lr->lr_foid, ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); } return (0); } static void dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh) { uint64_t claim_txg = zh->zh_claim_txg; zil_scan_arg_t zsa = { dp, zh }; zilog_t *zilog; /* * We only want to visit blocks that have been claimed but not yet * replayed (or, in read-only mode, blocks that *would* be claimed). */ if (claim_txg == 0 && spa_writeable(dp->dp_spa)) return; zilog = zil_alloc(dp->dp_meta_objset, zh); (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa, claim_txg); zil_free(zilog); } /* ARGSUSED */ static void dsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp, uint64_t objset, uint64_t object, uint64_t blkid) { zbookmark_phys_t czb; arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; if (zfs_no_scrub_prefetch) return; if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg || (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)) return; SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid); (void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb); } static boolean_t dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp, const zbookmark_phys_t *zb) { /* * We never skip over user/group accounting objects (obj<0) */ if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) && (int64_t)zb->zb_object >= 0) { /* * If we already visited this bp & everything below (in * a prior txg sync), don't bother doing it again. */ if (zbookmark_subtree_completed(dnp, zb, &scn->scn_phys.scn_bookmark)) return (B_TRUE); /* * If we found the block we're trying to resume from, or * we went past it to a different object, zero it out to * indicate that it's OK to start checking for pausing * again. */ if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 || zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) { dprintf("resuming at %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb)); } } return (B_FALSE); } /* * Return nonzero on i/o error. * Return new buf to write out in *bufp. */ static int dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, const blkptr_t *bp, const zbookmark_phys_t *zb, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; int err; if (BP_GET_LEVEL(bp) > 0) { arc_flags_t flags = ARC_FLAG_WAIT; int i; blkptr_t *cbp; int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; arc_buf_t *buf; err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { dsl_scan_prefetch(scn, buf, cbp, zb->zb_objset, zb->zb_object, zb->zb_blkid * epb + i); } for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, zb->zb_level - 1, zb->zb_blkid * epb + i); dsl_scan_visitbp(cbp, &czb, dnp, ds, scn, ostype, tx); } arc_buf_destroy(buf, &buf); } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { arc_flags_t flags = ARC_FLAG_WAIT; dnode_phys_t *cdnp; int i, j; int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; arc_buf_t *buf; err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) { for (j = 0; j < cdnp->dn_nblkptr; j++) { blkptr_t *cbp = &cdnp->dn_blkptr[j]; dsl_scan_prefetch(scn, buf, cbp, zb->zb_objset, zb->zb_blkid * epb + i, j); } } for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) { dsl_scan_visitdnode(scn, ds, ostype, cdnp, zb->zb_blkid * epb + i, tx); } arc_buf_destroy(buf, &buf); } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { arc_flags_t flags = ARC_FLAG_WAIT; objset_phys_t *osp; arc_buf_t *buf; err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } osp = buf->b_data; dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_meta_dnode, DMU_META_DNODE_OBJECT, tx); if (OBJSET_BUF_HAS_USERUSED(buf)) { /* * We also always visit user/group accounting * objects, and never skip them, even if we are * pausing. This is necessary so that the space * deltas from this txg get integrated. */ dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_groupused_dnode, DMU_GROUPUSED_OBJECT, tx); dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_userused_dnode, DMU_USERUSED_OBJECT, tx); } arc_buf_destroy(buf, &buf); } return (0); } static void dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx) { int j; for (j = 0; j < dnp->dn_nblkptr; j++) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, dnp->dn_nlevels - 1, j); dsl_scan_visitbp(&dnp->dn_blkptr[j], &czb, dnp, ds, scn, ostype, tx); } if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { zbookmark_phys_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, 0, DMU_SPILL_BLKID); dsl_scan_visitbp(&dnp->dn_spill, &czb, dnp, ds, scn, ostype, tx); } } /* * The arguments are in this order because mdb can only print the * first 5; we want them to be useful. */ static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb, dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; arc_buf_t *buf = NULL; blkptr_t bp_toread = *bp; /* ASSERT(pbuf == NULL || arc_released(pbuf)); */ if (dsl_scan_check_pause(scn, zb)) return; if (dsl_scan_check_resume(scn, dnp, zb)) return; if (BP_IS_HOLE(bp)) return; scn->scn_visited_this_txg++; dprintf_bp(bp, "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p", ds, ds ? ds->ds_object : 0, zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid, bp); if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) return; if (dsl_scan_recurse(scn, ds, ostype, dnp, &bp_toread, zb, tx) != 0) return; /* - * If dsl_scan_ddt() has aready visited this block, it will have + * If dsl_scan_ddt() has already visited this block, it will have * already done any translations or scrubbing, so don't call the * callback again. */ if (ddt_class_contains(dp->dp_spa, scn->scn_phys.scn_ddt_class_max, bp)) { ASSERT(buf == NULL); return; } /* * If this block is from the future (after cur_max_txg), then we * are doing this on behalf of a deleted snapshot, and we will * revisit the future block on the next pass of this dataset. * Don't scan it now unless we need to because something * under it was modified. */ if (BP_PHYSICAL_BIRTH(bp) <= scn->scn_phys.scn_cur_max_txg) { scan_funcs[scn->scn_phys.scn_func](dp, bp, zb); } } static void dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) { zbookmark_phys_t zb; SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); dsl_scan_visitbp(bp, &zb, NULL, ds, scn, DMU_OST_NONE, tx); dprintf_ds(ds, "finished scan%s", ""); } void dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) { if (ds->ds_is_snapshot) { /* * Note: * - scn_cur_{min,max}_txg stays the same. * - Setting the flag is not really necessary if * scn_cur_max_txg == scn_max_txg, because there * is nothing after this snapshot that we care * about. However, we set it anyway and then * ignore it when we retraverse it in * dsl_scan_visitds(). */ scn->scn_phys.scn_bookmark.zb_objset = dsl_dataset_phys(ds)->ds_next_snap_obj; zfs_dbgmsg("destroying ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)-> ds_next_snap_obj); scn->scn_phys.scn_flags |= DSF_VISIT_DS_AGAIN; } else { SET_BOOKMARK(&scn->scn_phys.scn_bookmark, ZB_DESTROYED_OBJSET, 0, 0, 0); zfs_dbgmsg("destroying ds %llu; currently traversing; " "reset bookmark to -1,0,0,0", (u_longlong_t)ds->ds_object); } } else if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); if (ds->ds_is_snapshot) { /* * We keep the same mintxg; it could be > * ds_creation_txg if the previous snapshot was * deleted too. */ VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_next_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("destroying ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)-> ds_next_snap_obj); } else { zfs_dbgmsg("destroying ds %llu; in queue; removing", (u_longlong_t)ds->ds_object); } } /* * dsl_scan_sync() should be called after this, and should sync * out our changed state, but just to be safe, do it here. */ dsl_scan_sync_state(scn, tx); } void dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (scn->scn_phys.scn_state != DSS_SCANNING) return; ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) { scn->scn_phys.scn_bookmark.zb_objset = dsl_dataset_phys(ds)->ds_prev_snap_obj; zfs_dbgmsg("snapshotting ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); } else if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("snapshotting ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); } dsl_scan_sync_state(scn, tx); } void dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx) { dsl_pool_t *dp = ds1->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (scn->scn_phys.scn_bookmark.zb_objset == ds1->ds_object) { scn->scn_phys.scn_bookmark.zb_objset = ds2->ds_object; zfs_dbgmsg("clone_swap ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds1->ds_object, (u_longlong_t)ds2->ds_object); } else if (scn->scn_phys.scn_bookmark.zb_objset == ds2->ds_object) { scn->scn_phys.scn_bookmark.zb_objset = ds1->ds_object; zfs_dbgmsg("clone_swap ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds2->ds_object, (u_longlong_t)ds1->ds_object); } if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, &mintxg) == 0) { int err; ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, tx)); err = zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx); VERIFY(err == 0 || err == EEXIST); if (err == EEXIST) { /* Both were there to begin with */ VERIFY(0 == zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx)); } zfs_dbgmsg("clone_swap ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds1->ds_object, (u_longlong_t)ds2->ds_object); } else if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg) == 0) { ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, tx)); VERIFY(0 == zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx)); zfs_dbgmsg("clone_swap ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds2->ds_object, (u_longlong_t)ds1->ds_object); } dsl_scan_sync_state(scn, tx); } struct enqueue_clones_arg { dmu_tx_t *tx; uint64_t originobj; }; /* ARGSUSED */ static int enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { struct enqueue_clones_arg *eca = arg; dsl_dataset_t *ds; int err; dsl_scan_t *scn = dp->dp_scan; if (dsl_dir_phys(hds->ds_dir)->dd_origin_obj != eca->originobj) return (0); err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); while (dsl_dataset_phys(ds)->ds_prev_snap_obj != eca->originobj) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); dsl_dataset_rele(ds, FTAG); if (err) return (err); ds = prev; } VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg, eca->tx) == 0); dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; dsl_dataset_t *ds; objset_t *os; VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (scn->scn_phys.scn_cur_min_txg >= scn->scn_phys.scn_max_txg) { /* * This can happen if this snapshot was created after the * scan started, and we already completed a previous snapshot * that was created after the scan started. This snapshot * only references blocks with: * * birth < our ds_creation_txg * cur_min_txg is no less than ds_creation_txg. * We have already visited these blocks. * or * birth > scn_max_txg * The scan requested not to visit these blocks. * * Subsequent snapshots (and clones) can reference our * blocks, or blocks with even higher birth times. * Therefore we do not need to visit them either, * so we do not add them to the work queue. * * Note that checking for cur_min_txg >= cur_max_txg * is not sufficient, because in that case we may need to * visit subsequent snapshots. This happens when min_txg > 0, * which raises cur_min_txg. In this case we will visit * this dataset but skip all of its blocks, because the * rootbp's birth time is < cur_min_txg. Then we will * add the next snapshots/clones to the work queue. */ char *dsname = kmem_alloc(MAXNAMELEN, KM_SLEEP); dsl_dataset_name(ds, dsname); zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because " "cur_min_txg (%llu) >= max_txg (%llu)", dsobj, dsname, scn->scn_phys.scn_cur_min_txg, scn->scn_phys.scn_max_txg); kmem_free(dsname, MAXNAMELEN); goto out; } if (dmu_objset_from_ds(ds, &os)) goto out; /* * Only the ZIL in the head (non-snapshot) is valid. Even though * snapshots can have ZIL block pointers (which may be the same * BP as in the head), they must be ignored. So we traverse the * ZIL here, rather than in scan_recurse(), because the regular * snapshot block-sharing rules don't apply to it. */ if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !ds->ds_is_snapshot) dsl_scan_zil(dp, &os->os_zil_header); /* * Iterate over the bps in this ds. */ dmu_buf_will_dirty(ds->ds_dbuf, tx); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); dsl_scan_visit_rootbp(scn, ds, &dsl_dataset_phys(ds)->ds_bp, tx); rrw_exit(&ds->ds_bp_rwlock, FTAG); char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); dsl_dataset_name(ds, dsname); zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; " "pausing=%u", (longlong_t)dsobj, dsname, (longlong_t)scn->scn_phys.scn_cur_min_txg, (longlong_t)scn->scn_phys.scn_cur_max_txg, (int)scn->scn_pausing); kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN); if (scn->scn_pausing) goto out; /* * We've finished this pass over this dataset. */ /* * If we did not completely visit this dataset, do another pass. */ if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) { zfs_dbgmsg("incomplete pass; visiting again"); scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN; VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, scn->scn_phys.scn_cur_max_txg, tx) == 0); goto out; } /* * Add descendent datasets to work queue. */ if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) { VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_next_snap_obj, dsl_dataset_phys(ds)->ds_creation_txg, tx) == 0); } if (dsl_dataset_phys(ds)->ds_num_children > 1) { boolean_t usenext = B_FALSE; if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { uint64_t count; /* * A bug in a previous version of the code could * cause upgrade_clones_cb() to not set * ds_next_snap_obj when it should, leading to a * missing entry. Therefore we can only use the * next_clones_obj when its count is correct. */ int err = zap_count(dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj, &count); if (err == 0 && count == dsl_dataset_phys(ds)->ds_num_children - 1) usenext = B_TRUE; } if (usenext) { VERIFY0(zap_join_key(dp->dp_meta_objset, dsl_dataset_phys(ds)->ds_next_clones_obj, scn->scn_phys.scn_queue_obj, dsl_dataset_phys(ds)->ds_creation_txg, tx)); } else { struct enqueue_clones_arg eca; eca.tx = tx; eca.originobj = ds->ds_object; VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, enqueue_clones_cb, &eca, DS_FIND_CHILDREN)); } } out: dsl_dataset_rele(ds, FTAG); } /* ARGSUSED */ static int enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) { dmu_tx_t *tx = arg; dsl_dataset_t *ds; int err; dsl_scan_t *scn = dp->dp_scan; err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); if (err) return (err); while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); if (err) { dsl_dataset_rele(ds, FTAG); return (err); } /* * If this is a clone, we don't need to worry about it for now. */ if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) { dsl_dataset_rele(ds, FTAG); dsl_dataset_rele(prev, FTAG); return (0); } dsl_dataset_rele(ds, FTAG); ds = prev; } VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg, tx) == 0); dsl_dataset_rele(ds, FTAG); return (0); } /* * Scrub/dedup interaction. * * If there are N references to a deduped block, we don't want to scrub it * N times -- ideally, we should scrub it exactly once. * * We leverage the fact that the dde's replication class (enum ddt_class) * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order. * * To prevent excess scrubbing, the scrub begins by walking the DDT * to find all blocks with refcnt > 1, and scrubs each of these once. * Since there are two replication classes which contain blocks with * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first. * Finally the top-down scrub begins, only visiting blocks with refcnt == 1. * * There would be nothing more to say if a block's refcnt couldn't change * during a scrub, but of course it can so we must account for changes * in a block's replication class. * * Here's an example of what can occur: * * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1 * when visited during the top-down scrub phase, it will be scrubbed twice. * This negates our scrub optimization, but is otherwise harmless. * * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1 * on each visit during the top-down scrub phase, it will never be scrubbed. * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1 * while a scrub is in progress, it scrubs the block right then. */ static void dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx) { ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark; ddt_entry_t dde = { 0 }; int error; uint64_t n = 0; while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) { ddt_t *ddt; if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max) break; dprintf("visiting ddb=%llu/%llu/%llu/%llx\n", (longlong_t)ddb->ddb_class, (longlong_t)ddb->ddb_type, (longlong_t)ddb->ddb_checksum, (longlong_t)ddb->ddb_cursor); /* There should be no pending changes to the dedup table */ ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum]; ASSERT(avl_first(&ddt->ddt_tree) == NULL); dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx); n++; if (dsl_scan_check_pause(scn, NULL)) break; } zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u", (longlong_t)n, (int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_pausing); ASSERT(error == 0 || error == ENOENT); ASSERT(error != ENOENT || ddb->ddb_class > scn->scn_phys.scn_ddt_class_max); } /* ARGSUSED */ void dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum, ddt_entry_t *dde, dmu_tx_t *tx) { const ddt_key_t *ddk = &dde->dde_key; ddt_phys_t *ddp = dde->dde_phys; blkptr_t bp; zbookmark_phys_t zb = { 0 }; if (scn->scn_phys.scn_state != DSS_SCANNING) return; for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { if (ddp->ddp_phys_birth == 0 || ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg) continue; ddt_bp_create(checksum, ddk, ddp, &bp); scn->scn_visited_this_txg++; scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb); } } static void dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; zap_cursor_t zc; zap_attribute_t za; if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= scn->scn_phys.scn_ddt_class_max) { scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_ddt(scn, tx); if (scn->scn_pausing) return; } if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) { /* First do the MOS & ORIGIN */ scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_visit_rootbp(scn, NULL, &dp->dp_meta_rootbp, tx); spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); if (scn->scn_pausing) return; if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) { VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, enqueue_cb, tx, DS_FIND_CHILDREN)); } else { dsl_scan_visitds(scn, dp->dp_origin_snap->ds_object, tx); } ASSERT(!scn->scn_pausing); } else if (scn->scn_phys.scn_bookmark.zb_objset != ZB_DESTROYED_OBJSET) { /* * If we were paused, continue from here. Note if the * ds we were paused on was deleted, the zb_objset may * be -1, so we will skip this and find a new objset * below. */ dsl_scan_visitds(scn, scn->scn_phys.scn_bookmark.zb_objset, tx); if (scn->scn_pausing) return; } /* * In case we were paused right at the end of the ds, zero the * bookmark so we don't think that we're still trying to resume. */ bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_phys_t)); /* keep pulling things out of the zap-object-as-queue */ while (zap_cursor_init(&zc, dp->dp_meta_objset, scn->scn_phys.scn_queue_obj), zap_cursor_retrieve(&zc, &za) == 0) { dsl_dataset_t *ds; uint64_t dsobj; dsobj = strtonum(za.za_name, NULL); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsobj, tx)); /* Set up min/max txg */ VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (za.za_first_integer != 0) { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, za.za_first_integer); } else { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, dsl_dataset_phys(ds)->ds_prev_snap_txg); } scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds); dsl_dataset_rele(ds, FTAG); dsl_scan_visitds(scn, dsobj, tx); zap_cursor_fini(&zc); if (scn->scn_pausing) return; } zap_cursor_fini(&zc); } static boolean_t dsl_scan_free_should_pause(dsl_scan_t *scn) { uint64_t elapsed_nanosecs; if (zfs_recover) return (B_FALSE); if (scn->scn_visited_this_txg >= zfs_free_max_blocks) return (B_TRUE); elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms && txg_sync_waiting(scn->scn_dp)) || spa_shutting_down(scn->scn_dp->dp_spa)); } static int dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { dsl_scan_t *scn = arg; if (!scn->scn_is_bptree || (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) { if (dsl_scan_free_should_pause(scn)) return (SET_ERROR(ERESTART)); } zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa, dmu_tx_get_txg(tx), bp, BP_GET_PSIZE(bp), 0)); dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp), -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); scn->scn_visited_this_txg++; return (0); } boolean_t dsl_scan_active(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; uint64_t used = 0, comp, uncomp; if (spa->spa_load_state != SPA_LOAD_NONE) return (B_FALSE); if (spa_shutting_down(spa)) return (B_FALSE); if (scn->scn_phys.scn_state == DSS_SCANNING || (scn->scn_async_destroying && !scn->scn_async_stalled)) return (B_TRUE); if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) { (void) bpobj_space(&scn->scn_dp->dp_free_bpobj, &used, &comp, &uncomp); } return (used != 0); } +/* Called whenever a txg syncs. */ void dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) { dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; int err = 0; /* * Check for scn_restart_txg before checking spa_load_state, so * that we can restart an old-style scan while the pool is being * imported (see dsl_scan_init). */ if (dsl_scan_restarting(scn, tx)) { pool_scan_func_t func = POOL_SCAN_SCRUB; dsl_scan_done(scn, B_FALSE, tx); if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) func = POOL_SCAN_RESILVER; zfs_dbgmsg("restarting scan func=%u txg=%llu", func, tx->tx_txg); dsl_scan_setup_sync(&func, tx); } /* * Only process scans in sync pass 1. */ if (spa_sync_pass(dp->dp_spa) > 1) return; /* * If the spa is shutting down, then stop scanning. This will * ensure that the scan does not dirty any new data during the * shutdown phase. */ if (spa_shutting_down(spa)) return; /* * If the scan is inactive due to a stalled async destroy, try again. */ if (!scn->scn_async_stalled && !dsl_scan_active(scn)) return; scn->scn_visited_this_txg = 0; scn->scn_pausing = B_FALSE; scn->scn_sync_start_time = gethrtime(); spa->spa_scrub_active = B_TRUE; /* * First process the async destroys. If we pause, don't do * any scrubbing or resilvering. This ensures that there are no * async destroys while we are scanning, so the scan code doesn't * have to worry about traversing it. It is also faster to free the * blocks than to scrub them. */ if (zfs_free_bpobj_enabled && spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { scn->scn_is_bptree = B_FALSE; scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bpobj_iterate(&dp->dp_free_bpobj, dsl_scan_free_block_cb, scn, tx); VERIFY3U(0, ==, zio_wait(scn->scn_zio_root)); if (err != 0 && err != ERESTART) zfs_panic_recover("error %u from bpobj_iterate()", err); } if (err == 0 && spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) { ASSERT(scn->scn_async_destroying); scn->scn_is_bptree = B_TRUE; scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bptree_iterate(dp->dp_meta_objset, dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx); VERIFY0(zio_wait(scn->scn_zio_root)); if (err == EIO || err == ECKSUM) { err = 0; } else if (err != 0 && err != ERESTART) { zfs_panic_recover("error %u from " "traverse_dataset_destroyed()", err); } if (bptree_is_empty(dp->dp_meta_objset, dp->dp_bptree_obj)) { /* finished; deactivate async destroy feature */ spa_feature_decr(spa, SPA_FEATURE_ASYNC_DESTROY, tx); ASSERT(!spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)); VERIFY0(zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_BPTREE_OBJ, tx)); VERIFY0(bptree_free(dp->dp_meta_objset, dp->dp_bptree_obj, tx)); dp->dp_bptree_obj = 0; scn->scn_async_destroying = B_FALSE; scn->scn_async_stalled = B_FALSE; } else { /* * If we didn't make progress, mark the async * destroy as stalled, so that we will not initiate * a spa_sync() on its behalf. Note that we only * check this if we are not finished, because if the * bptree had no blocks for us to visit, we can * finish without "making progress". */ scn->scn_async_stalled = (scn->scn_visited_this_txg == 0); } } if (scn->scn_visited_this_txg) { zfs_dbgmsg("freed %llu blocks in %llums from " "free_bpobj/bptree txg %llu; err=%d", (longlong_t)scn->scn_visited_this_txg, (longlong_t) NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), (longlong_t)tx->tx_txg, err); scn->scn_visited_this_txg = 0; /* * Write out changes to the DDT that may be required as a * result of the blocks freed. This ensures that the DDT * is clean when a scrub/resilver runs. */ ddt_sync(spa, tx->tx_txg); } if (err != 0) return; if (dp->dp_free_dir != NULL && !scn->scn_async_destroying && zfs_free_leak_on_eio && (dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes != 0 || dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes != 0 || dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes != 0)) { /* * We have finished background destroying, but there is still * some space left in the dp_free_dir. Transfer this leaked * space to the dp_leak_dir. */ if (dp->dp_leak_dir == NULL) { rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); (void) dsl_dir_create_sync(dp, dp->dp_root_dir, LEAK_DIR_NAME, tx); VERIFY0(dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, &dp->dp_leak_dir)); rrw_exit(&dp->dp_config_rwlock, FTAG); } dsl_dir_diduse_space(dp->dp_leak_dir, DD_USED_HEAD, dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, -dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, -dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, -dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); } if (dp->dp_free_dir != NULL && !scn->scn_async_destroying) { /* finished; verify that space accounting went to zero */ ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes); ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes); ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes); } if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (scn->scn_done_txg == tx->tx_txg) { ASSERT(!scn->scn_pausing); /* finished with scan. */ zfs_dbgmsg("txg %llu scan complete", tx->tx_txg); dsl_scan_done(scn, B_TRUE, tx); ASSERT3U(spa->spa_scrub_inflight, ==, 0); dsl_scan_sync_state(scn, tx); return; } if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= scn->scn_phys.scn_ddt_class_max) { zfs_dbgmsg("doing scan sync txg %llu; " "ddt bm=%llu/%llu/%llu/%llx", (longlong_t)tx->tx_txg, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor); ASSERT(scn->scn_phys.scn_bookmark.zb_objset == 0); ASSERT(scn->scn_phys.scn_bookmark.zb_object == 0); ASSERT(scn->scn_phys.scn_bookmark.zb_level == 0); ASSERT(scn->scn_phys.scn_bookmark.zb_blkid == 0); } else { zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu", (longlong_t)tx->tx_txg, (longlong_t)scn->scn_phys.scn_bookmark.zb_objset, (longlong_t)scn->scn_phys.scn_bookmark.zb_object, (longlong_t)scn->scn_phys.scn_bookmark.zb_level, (longlong_t)scn->scn_phys.scn_bookmark.zb_blkid); } scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); dsl_pool_config_enter(dp, FTAG); dsl_scan_visit(scn, tx); dsl_pool_config_exit(dp, FTAG); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; zfs_dbgmsg("visited %llu blocks in %llums", (longlong_t)scn->scn_visited_this_txg, (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time)); if (!scn->scn_pausing) { scn->scn_done_txg = tx->tx_txg + 1; zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu", tx->tx_txg, scn->scn_done_txg); } if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight > 0) { cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); } mutex_exit(&spa->spa_scrub_lock); } dsl_scan_sync_state(scn, tx); } /* * This will start a new scan, or restart an existing one. */ void dsl_resilver_restart(dsl_pool_t *dp, uint64_t txg) { if (txg == 0) { dmu_tx_t *tx; tx = dmu_tx_create_dd(dp->dp_mos_dir); VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT)); txg = dmu_tx_get_txg(tx); dp->dp_scan->scn_restart_txg = txg; dmu_tx_commit(tx); } else { dp->dp_scan->scn_restart_txg = txg; } zfs_dbgmsg("restarting resilver txg=%llu", txg); } boolean_t dsl_scan_resilvering(dsl_pool_t *dp) { return (dp->dp_scan->scn_phys.scn_state == DSS_SCANNING && dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER); } /* * scrub consumers */ static void count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp) { int i; /* * If we resume after a reboot, zab will be NULL; don't record * incomplete stats in that case. */ if (zab == NULL) return; for (i = 0; i < 4; i++) { int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS; int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL; if (t & DMU_OT_NEWTYPE) t = DMU_OT_OTHER; zfs_blkstat_t *zb = &zab->zab_type[l][t]; int equal; zb->zb_count++; zb->zb_asize += BP_GET_ASIZE(bp); zb->zb_lsize += BP_GET_LSIZE(bp); zb->zb_psize += BP_GET_PSIZE(bp); zb->zb_gangs += BP_COUNT_GANG(bp); switch (BP_GET_NDVAS(bp)) { case 2: if (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) zb->zb_ditto_2_of_2_samevdev++; break; case 3: equal = (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) + (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[2])) + (DVA_GET_VDEV(&bp->blk_dva[1]) == DVA_GET_VDEV(&bp->blk_dva[2])); if (equal == 1) zb->zb_ditto_2_of_3_samevdev++; else if (equal == 3) zb->zb_ditto_3_of_3_samevdev++; break; } } } static void dsl_scan_scrub_done(zio_t *zio) { spa_t *spa = zio->io_spa; zio_data_buf_free(zio->io_data, zio->io_size); mutex_enter(&spa->spa_scrub_lock); spa->spa_scrub_inflight--; cv_broadcast(&spa->spa_scrub_io_cv); if (zio->io_error && (zio->io_error != ECKSUM || !(zio->io_flags & ZIO_FLAG_SPECULATIVE))) { spa->spa_dsl_pool->dp_scan->scn_phys.scn_errors++; } mutex_exit(&spa->spa_scrub_lock); } static int dsl_scan_scrub_cb(dsl_pool_t *dp, const blkptr_t *bp, const zbookmark_phys_t *zb) { dsl_scan_t *scn = dp->dp_scan; size_t size = BP_GET_PSIZE(bp); spa_t *spa = dp->dp_spa; uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp); boolean_t needs_io; int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL; unsigned int scan_delay = 0; if (phys_birth <= scn->scn_phys.scn_min_txg || phys_birth >= scn->scn_phys.scn_max_txg) return (0); count_block(dp->dp_blkstats, bp); if (BP_IS_EMBEDDED(bp)) return (0); ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn)); if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) { zio_flags |= ZIO_FLAG_SCRUB; needs_io = B_TRUE; scan_delay = zfs_scrub_delay; } else { ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER); zio_flags |= ZIO_FLAG_RESILVER; needs_io = B_FALSE; scan_delay = zfs_resilver_delay; } /* If it's an intent log block, failure is expected. */ if (zb->zb_level == ZB_ZIL_LEVEL) zio_flags |= ZIO_FLAG_SPECULATIVE; for (int d = 0; d < BP_GET_NDVAS(bp); d++) { vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[d])); /* * Keep track of how much data we've examined so that * zpool(1M) status can make useful progress reports. */ scn->scn_phys.scn_examined += DVA_GET_ASIZE(&bp->blk_dva[d]); spa->spa_scan_pass_exam += DVA_GET_ASIZE(&bp->blk_dva[d]); /* if it's a resilver, this may not be in the target range */ if (!needs_io) { if (DVA_GET_GANG(&bp->blk_dva[d])) { /* * Gang members may be spread across multiple * vdevs, so the best estimate we have is the * scrub range, which has already been checked. * XXX -- it would be better to change our * allocation policy to ensure that all * gang members reside on the same vdev. */ needs_io = B_TRUE; } else { needs_io = vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1); } } } if (needs_io && !zfs_no_scrub_io) { vdev_t *rvd = spa->spa_root_vdev; uint64_t maxinflight = rvd->vdev_children * MAX(zfs_top_maxinflight, 1); void *data = zio_data_buf_alloc(size); mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight >= maxinflight) cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); spa->spa_scrub_inflight++; mutex_exit(&spa->spa_scrub_lock); /* * If we're seeing recent (zfs_scan_idle) "important" I/Os * then throttle our workload to limit the impact of a scan. */ if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle) delay(MAX((int)scan_delay, 0)); zio_nowait(zio_read(NULL, spa, bp, data, size, dsl_scan_scrub_done, NULL, ZIO_PRIORITY_SCRUB, zio_flags, zb)); } /* do not relocate this block */ return (0); } +/* Called by the ZFS_IOC_POOL_SCAN ioctl to start a scrub or resilver */ int dsl_scan(dsl_pool_t *dp, pool_scan_func_t func) { spa_t *spa = dp->dp_spa; /* * Purge all vdev caches and probe all devices. We do this here * rather than in sync context because this requires a writer lock * on the spa_config lock, which we can't do from sync context. The * spa_scrub_reopen flag indicates that vdev_open() should not * attempt to start another scrub. */ spa_vdev_state_enter(spa, SCL_NONE); spa->spa_scrub_reopen = B_TRUE; vdev_reopen(spa->spa_root_vdev); spa->spa_scrub_reopen = B_FALSE; (void) spa_vdev_state_exit(spa, NULL, 0); return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check, dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_NONE)); } static boolean_t dsl_scan_restarting(dsl_scan_t *scn, dmu_tx_t *tx) { return (scn->scn_restart_txg != 0 && scn->scn_restart_txg <= tx->tx_txg); } Index: head/sys/cddl/contrib/opensolaris =================================================================== --- head/sys/cddl/contrib/opensolaris (revision 318934) +++ head/sys/cddl/contrib/opensolaris (revision 318935) Property changes on: head/sys/cddl/contrib/opensolaris ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /vendor-sys/illumos/dist:r318934