Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 321553) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 321554) @@ -1,3589 +1,3590 @@ /* * 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 havepzio = (zio != NULL); 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; if (zio == NULL) zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); dbuf_read_impl(db, zio, flags); /* dbuf_read_impl has dropped db_mtx for us */ if (prefetch) dmu_zfetch(&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 (!havepzio) err = zio_wait(zio); } else { /* * Another reader came in while the dbuf was in flight * between UNCACHED and CACHED. Either a writer will finish * writing the buffer (sending the dbuf to CACHED) or the * first reader's request will reach the read_done callback * and send the dbuf to CACHED. Otherwise, a failure * occurred and the dbuf went to UNCACHED. */ mutex_exit(&db->db_mtx); if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db_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); } ASSERT(err || havepzio || db->db_state == DB_CACHED); 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)); 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: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c (revision 321553) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c (revision 321554) @@ -1,3489 +1,3501 @@ /* * 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 (c) 2011, 2015 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include SYSCTL_DECL(_vfs_zfs); SYSCTL_NODE(_vfs_zfs, OID_AUTO, metaslab, CTLFLAG_RW, 0, "ZFS metaslab"); #define GANG_ALLOCATION(flags) \ ((flags) & (METASLAB_GANG_CHILD | METASLAB_GANG_HEADER)) uint64_t metaslab_aliquot = 512ULL << 10; uint64_t metaslab_gang_bang = SPA_MAXBLOCKSIZE + 1; /* force gang blocks */ SYSCTL_QUAD(_vfs_zfs_metaslab, OID_AUTO, gang_bang, CTLFLAG_RWTUN, &metaslab_gang_bang, 0, "Force gang block allocation for blocks larger than or equal to this value"); /* * The in-core space map representation is more compact than its on-disk form. * The zfs_condense_pct determines how much more compact the in-core * space map representation must be before we compact it on-disk. * Values should be greater than or equal to 100. */ int zfs_condense_pct = 200; SYSCTL_INT(_vfs_zfs, OID_AUTO, condense_pct, CTLFLAG_RWTUN, &zfs_condense_pct, 0, "Condense on-disk spacemap when it is more than this many percents" " of in-memory counterpart"); /* * Condensing a metaslab is not guaranteed to actually reduce the amount of * space used on disk. In particular, a space map uses data in increments of * MAX(1 << ashift, space_map_blksize), so a metaslab might use the * same number of blocks after condensing. Since the goal of condensing is to * reduce the number of IOPs required to read the space map, we only want to * condense when we can be sure we will reduce the number of blocks used by the * space map. Unfortunately, we cannot precisely compute whether or not this is * the case in metaslab_should_condense since we are holding ms_lock. Instead, * we apply the following heuristic: do not condense a spacemap unless the * uncondensed size consumes greater than zfs_metaslab_condense_block_threshold * blocks. */ int zfs_metaslab_condense_block_threshold = 4; /* * The zfs_mg_noalloc_threshold defines which metaslab groups should * be eligible for allocation. The value is defined as a percentage of * free space. Metaslab groups that have more free space than * zfs_mg_noalloc_threshold are always eligible for allocations. Once * a metaslab group's free space is less than or equal to the * zfs_mg_noalloc_threshold the allocator will avoid allocating to that * group unless all groups in the pool have reached zfs_mg_noalloc_threshold. * Once all groups in the pool reach zfs_mg_noalloc_threshold then all * groups are allowed to accept allocations. Gang blocks are always * eligible to allocate on any metaslab group. The default value of 0 means * no metaslab group will be excluded based on this criterion. */ int zfs_mg_noalloc_threshold = 0; SYSCTL_INT(_vfs_zfs, OID_AUTO, mg_noalloc_threshold, CTLFLAG_RWTUN, &zfs_mg_noalloc_threshold, 0, "Percentage of metaslab group size that should be free" " to make it eligible for allocation"); /* * Metaslab groups are considered eligible for allocations if their * fragmenation metric (measured as a percentage) is less than or equal to * zfs_mg_fragmentation_threshold. If a metaslab group exceeds this threshold * then it will be skipped unless all metaslab groups within the metaslab * class have also crossed this threshold. */ int zfs_mg_fragmentation_threshold = 85; SYSCTL_INT(_vfs_zfs, OID_AUTO, mg_fragmentation_threshold, CTLFLAG_RWTUN, &zfs_mg_fragmentation_threshold, 0, "Percentage of metaslab group size that should be considered " "eligible for allocations unless all metaslab groups within the metaslab class " "have also crossed this threshold"); /* * Allow metaslabs to keep their active state as long as their fragmentation * percentage is less than or equal to zfs_metaslab_fragmentation_threshold. An * active metaslab that exceeds this threshold will no longer keep its active * status allowing better metaslabs to be selected. */ int zfs_metaslab_fragmentation_threshold = 70; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, fragmentation_threshold, CTLFLAG_RWTUN, &zfs_metaslab_fragmentation_threshold, 0, "Maximum percentage of metaslab fragmentation level to keep their active state"); /* * When set will load all metaslabs when pool is first opened. */ int metaslab_debug_load = 0; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, debug_load, CTLFLAG_RWTUN, &metaslab_debug_load, 0, "Load all metaslabs when pool is first opened"); /* * When set will prevent metaslabs from being unloaded. */ int metaslab_debug_unload = 0; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, debug_unload, CTLFLAG_RWTUN, &metaslab_debug_unload, 0, "Prevent metaslabs from being unloaded"); /* * Minimum size which forces the dynamic allocator to change * it's allocation strategy. Once the space map cannot satisfy * an allocation of this size then it switches to using more * aggressive strategy (i.e search by size rather than offset). */ uint64_t metaslab_df_alloc_threshold = SPA_OLD_MAXBLOCKSIZE; SYSCTL_QUAD(_vfs_zfs_metaslab, OID_AUTO, df_alloc_threshold, CTLFLAG_RWTUN, &metaslab_df_alloc_threshold, 0, "Minimum size which forces the dynamic allocator to change it's allocation strategy"); /* * The minimum free space, in percent, which must be available * in a space map to continue allocations in a first-fit fashion. * Once the space map's free space drops below this level we dynamically * switch to using best-fit allocations. */ int metaslab_df_free_pct = 4; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, df_free_pct, CTLFLAG_RWTUN, &metaslab_df_free_pct, 0, "The minimum free space, in percent, which must be available in a " "space map to continue allocations in a first-fit fashion"); /* * A metaslab is considered "free" if it contains a contiguous * segment which is greater than metaslab_min_alloc_size. */ uint64_t metaslab_min_alloc_size = DMU_MAX_ACCESS; SYSCTL_QUAD(_vfs_zfs_metaslab, OID_AUTO, min_alloc_size, CTLFLAG_RWTUN, &metaslab_min_alloc_size, 0, "A metaslab is considered \"free\" if it contains a contiguous " "segment which is greater than vfs.zfs.metaslab.min_alloc_size"); /* * Percentage of all cpus that can be used by the metaslab taskq. */ int metaslab_load_pct = 50; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, load_pct, CTLFLAG_RWTUN, &metaslab_load_pct, 0, "Percentage of cpus that can be used by the metaslab taskq"); /* * Determines how many txgs a metaslab may remain loaded without having any * allocations from it. As long as a metaslab continues to be used we will * keep it loaded. */ int metaslab_unload_delay = TXG_SIZE * 2; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, unload_delay, CTLFLAG_RWTUN, &metaslab_unload_delay, 0, "Number of TXGs that an unused metaslab can be kept in memory"); /* * Max number of metaslabs per group to preload. */ int metaslab_preload_limit = SPA_DVAS_PER_BP; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, preload_limit, CTLFLAG_RWTUN, &metaslab_preload_limit, 0, "Max number of metaslabs per group to preload"); /* * Enable/disable preloading of metaslab. */ boolean_t metaslab_preload_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, preload_enabled, CTLFLAG_RWTUN, &metaslab_preload_enabled, 0, "Max number of metaslabs per group to preload"); /* * Enable/disable fragmentation weighting on metaslabs. */ boolean_t metaslab_fragmentation_factor_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, fragmentation_factor_enabled, CTLFLAG_RWTUN, &metaslab_fragmentation_factor_enabled, 0, "Enable fragmentation weighting on metaslabs"); /* * Enable/disable lba weighting (i.e. outer tracks are given preference). */ boolean_t metaslab_lba_weighting_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, lba_weighting_enabled, CTLFLAG_RWTUN, &metaslab_lba_weighting_enabled, 0, "Enable LBA weighting (i.e. outer tracks are given preference)"); /* * Enable/disable metaslab group biasing. */ boolean_t metaslab_bias_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, bias_enabled, CTLFLAG_RWTUN, &metaslab_bias_enabled, 0, "Enable metaslab group biasing"); /* * Enable/disable segment-based metaslab selection. */ boolean_t zfs_metaslab_segment_weight_enabled = B_TRUE; /* * When using segment-based metaslab selection, we will continue * allocating from the active metaslab until we have exhausted * zfs_metaslab_switch_threshold of its buckets. */ int zfs_metaslab_switch_threshold = 2; /* * Internal switch to enable/disable the metaslab allocation tracing * facility. */ boolean_t metaslab_trace_enabled = B_TRUE; /* * Maximum entries that the metaslab allocation tracing facility will keep * in a given list when running in non-debug mode. We limit the number * of entries in non-debug mode to prevent us from using up too much memory. * The limit should be sufficiently large that we don't expect any allocation * to every exceed this value. In debug mode, the system will panic if this * limit is ever reached allowing for further investigation. */ uint64_t metaslab_trace_max_entries = 5000; static uint64_t metaslab_weight(metaslab_t *); static void metaslab_set_fragmentation(metaslab_t *); kmem_cache_t *metaslab_alloc_trace_cache; /* * ========================================================================== * Metaslab classes * ========================================================================== */ metaslab_class_t * metaslab_class_create(spa_t *spa, metaslab_ops_t *ops) { metaslab_class_t *mc; mc = kmem_zalloc(sizeof (metaslab_class_t), KM_SLEEP); mc->mc_spa = spa; mc->mc_rotor = NULL; mc->mc_ops = ops; mutex_init(&mc->mc_lock, NULL, MUTEX_DEFAULT, NULL); refcount_create_tracked(&mc->mc_alloc_slots); return (mc); } void metaslab_class_destroy(metaslab_class_t *mc) { ASSERT(mc->mc_rotor == NULL); ASSERT(mc->mc_alloc == 0); ASSERT(mc->mc_deferred == 0); ASSERT(mc->mc_space == 0); ASSERT(mc->mc_dspace == 0); refcount_destroy(&mc->mc_alloc_slots); mutex_destroy(&mc->mc_lock); kmem_free(mc, sizeof (metaslab_class_t)); } int metaslab_class_validate(metaslab_class_t *mc) { metaslab_group_t *mg; vdev_t *vd; /* * Must hold one of the spa_config locks. */ ASSERT(spa_config_held(mc->mc_spa, SCL_ALL, RW_READER) || spa_config_held(mc->mc_spa, SCL_ALL, RW_WRITER)); if ((mg = mc->mc_rotor) == NULL) return (0); do { vd = mg->mg_vd; ASSERT(vd->vdev_mg != NULL); ASSERT3P(vd->vdev_top, ==, vd); ASSERT3P(mg->mg_class, ==, mc); ASSERT3P(vd->vdev_ops, !=, &vdev_hole_ops); } while ((mg = mg->mg_next) != mc->mc_rotor); return (0); } void metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta, int64_t defer_delta, int64_t space_delta, int64_t dspace_delta) { atomic_add_64(&mc->mc_alloc, alloc_delta); atomic_add_64(&mc->mc_deferred, defer_delta); atomic_add_64(&mc->mc_space, space_delta); atomic_add_64(&mc->mc_dspace, dspace_delta); } void metaslab_class_minblocksize_update(metaslab_class_t *mc) { metaslab_group_t *mg; vdev_t *vd; uint64_t minashift = UINT64_MAX; if ((mg = mc->mc_rotor) == NULL) { mc->mc_minblocksize = SPA_MINBLOCKSIZE; return; } do { vd = mg->mg_vd; if (vd->vdev_ashift < minashift) minashift = vd->vdev_ashift; } while ((mg = mg->mg_next) != mc->mc_rotor); mc->mc_minblocksize = 1ULL << minashift; } uint64_t metaslab_class_get_alloc(metaslab_class_t *mc) { return (mc->mc_alloc); } uint64_t metaslab_class_get_deferred(metaslab_class_t *mc) { return (mc->mc_deferred); } uint64_t metaslab_class_get_space(metaslab_class_t *mc) { return (mc->mc_space); } uint64_t metaslab_class_get_dspace(metaslab_class_t *mc) { return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space); } uint64_t metaslab_class_get_minblocksize(metaslab_class_t *mc) { return (mc->mc_minblocksize); } void metaslab_class_histogram_verify(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t *mc_hist; int i; if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0) return; mc_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE, KM_SLEEP); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; /* * Skip any holes, uninitialized top-levels, or * vdevs that are not in this metalab class. */ if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) mc_hist[i] += mg->mg_histogram[i]; } for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) VERIFY3U(mc_hist[i], ==, mc->mc_histogram[i]); kmem_free(mc_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE); } /* * Calculate the metaslab class's fragmentation metric. The metric * is weighted based on the space contribution of each metaslab group. * The return value will be a number between 0 and 100 (inclusive), or * ZFS_FRAG_INVALID if the metric has not been set. See comment above the * zfs_frag_table for more information about the metric. */ uint64_t metaslab_class_fragmentation(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t fragmentation = 0; spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; /* * Skip any holes, uninitialized top-levels, or * vdevs that are not in this metalab class. */ if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } /* * If a metaslab group does not contain a fragmentation * metric then just bail out. */ if (mg->mg_fragmentation == ZFS_FRAG_INVALID) { spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (ZFS_FRAG_INVALID); } /* * Determine how much this metaslab_group is contributing * to the overall pool fragmentation metric. */ fragmentation += mg->mg_fragmentation * metaslab_group_get_space(mg); } fragmentation /= metaslab_class_get_space(mc); ASSERT3U(fragmentation, <=, 100); spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (fragmentation); } /* * Calculate the amount of expandable space that is available in * this metaslab class. If a device is expanded then its expandable * space will be the amount of allocatable space that is currently not * part of this metaslab class. */ uint64_t metaslab_class_expandable_space(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t space = 0; spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } /* * Calculate if we have enough space to add additional * metaslabs. We report the expandable space in terms * of the metaslab size since that's the unit of expansion. */ space += P2ALIGN(tvd->vdev_max_asize - tvd->vdev_asize, 1ULL << tvd->vdev_ms_shift); } spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (space); } static int metaslab_compare(const void *x1, const void *x2) { const metaslab_t *m1 = x1; const metaslab_t *m2 = x2; if (m1->ms_weight < m2->ms_weight) return (1); if (m1->ms_weight > m2->ms_weight) return (-1); /* * If the weights are identical, use the offset to force uniqueness. */ if (m1->ms_start < m2->ms_start) return (-1); if (m1->ms_start > m2->ms_start) return (1); ASSERT3P(m1, ==, m2); return (0); } /* * Verify that the space accounting on disk matches the in-core range_trees. */ void metaslab_verify_space(metaslab_t *msp, uint64_t txg) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; uint64_t allocated = 0; uint64_t sm_free_space, msp_free_space; ASSERT(MUTEX_HELD(&msp->ms_lock)); if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0) return; /* * We can only verify the metaslab space when we're called * from syncing context with a loaded metaslab that has an allocated * space map. Calling this in non-syncing context does not * provide a consistent view of the metaslab since we're performing * allocations in the future. */ if (txg != spa_syncing_txg(spa) || msp->ms_sm == NULL || !msp->ms_loaded) return; sm_free_space = msp->ms_size - space_map_allocated(msp->ms_sm) - space_map_alloc_delta(msp->ms_sm); /* * Account for future allocations since we would have already * deducted that space from the ms_freetree. */ for (int t = 0; t < TXG_CONCURRENT_STATES; t++) { allocated += range_tree_space(msp->ms_alloctree[(txg + t) & TXG_MASK]); } msp_free_space = range_tree_space(msp->ms_tree) + allocated + msp->ms_deferspace + range_tree_space(msp->ms_freedtree); VERIFY3U(sm_free_space, ==, msp_free_space); } /* * ========================================================================== * Metaslab groups * ========================================================================== */ /* * Update the allocatable flag and the metaslab group's capacity. * The allocatable flag is set to true if the capacity is below * the zfs_mg_noalloc_threshold or has a fragmentation value that is * greater than zfs_mg_fragmentation_threshold. If a metaslab group * transitions from allocatable to non-allocatable or vice versa then the * metaslab group's class is updated to reflect the transition. */ static void metaslab_group_alloc_update(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; metaslab_class_t *mc = mg->mg_class; vdev_stat_t *vs = &vd->vdev_stat; boolean_t was_allocatable; boolean_t was_initialized; ASSERT(vd == vd->vdev_top); mutex_enter(&mg->mg_lock); was_allocatable = mg->mg_allocatable; was_initialized = mg->mg_initialized; mg->mg_free_capacity = ((vs->vs_space - vs->vs_alloc) * 100) / (vs->vs_space + 1); mutex_enter(&mc->mc_lock); /* * If the metaslab group was just added then it won't * have any space until we finish syncing out this txg. * At that point we will consider it initialized and available * for allocations. We also don't consider non-activated * metaslab groups (e.g. vdevs that are in the middle of being removed) * to be initialized, because they can't be used for allocation. */ mg->mg_initialized = metaslab_group_initialized(mg); if (!was_initialized && mg->mg_initialized) { mc->mc_groups++; } else if (was_initialized && !mg->mg_initialized) { ASSERT3U(mc->mc_groups, >, 0); mc->mc_groups--; } if (mg->mg_initialized) mg->mg_no_free_space = B_FALSE; /* * A metaslab group is considered allocatable if it has plenty * of free space or is not heavily fragmented. We only take * fragmentation into account if the metaslab group has a valid * fragmentation metric (i.e. a value between 0 and 100). */ mg->mg_allocatable = (mg->mg_activation_count > 0 && mg->mg_free_capacity > zfs_mg_noalloc_threshold && (mg->mg_fragmentation == ZFS_FRAG_INVALID || mg->mg_fragmentation <= zfs_mg_fragmentation_threshold)); /* * The mc_alloc_groups maintains a count of the number of * groups in this metaslab class that are still above the * zfs_mg_noalloc_threshold. This is used by the allocating * threads to determine if they should avoid allocations to * a given group. The allocator will avoid allocations to a group * if that group has reached or is below the zfs_mg_noalloc_threshold * and there are still other groups that are above the threshold. * When a group transitions from allocatable to non-allocatable or * vice versa we update the metaslab class to reflect that change. * When the mc_alloc_groups value drops to 0 that means that all * groups have reached the zfs_mg_noalloc_threshold making all groups * eligible for allocations. This effectively means that all devices * are balanced again. */ if (was_allocatable && !mg->mg_allocatable) mc->mc_alloc_groups--; else if (!was_allocatable && mg->mg_allocatable) mc->mc_alloc_groups++; mutex_exit(&mc->mc_lock); mutex_exit(&mg->mg_lock); } metaslab_group_t * metaslab_group_create(metaslab_class_t *mc, vdev_t *vd) { metaslab_group_t *mg; mg = kmem_zalloc(sizeof (metaslab_group_t), KM_SLEEP); mutex_init(&mg->mg_lock, NULL, MUTEX_DEFAULT, NULL); avl_create(&mg->mg_metaslab_tree, metaslab_compare, sizeof (metaslab_t), offsetof(struct metaslab, ms_group_node)); mg->mg_vd = vd; mg->mg_class = mc; mg->mg_activation_count = 0; mg->mg_initialized = B_FALSE; mg->mg_no_free_space = B_TRUE; refcount_create_tracked(&mg->mg_alloc_queue_depth); mg->mg_taskq = taskq_create("metaslab_group_taskq", metaslab_load_pct, minclsyspri, 10, INT_MAX, TASKQ_THREADS_CPU_PCT); return (mg); } void metaslab_group_destroy(metaslab_group_t *mg) { ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); /* * We may have gone below zero with the activation count * either because we never activated in the first place or * because we're done, and possibly removing the vdev. */ ASSERT(mg->mg_activation_count <= 0); taskq_destroy(mg->mg_taskq); avl_destroy(&mg->mg_metaslab_tree); mutex_destroy(&mg->mg_lock); refcount_destroy(&mg->mg_alloc_queue_depth); kmem_free(mg, sizeof (metaslab_group_t)); } void metaslab_group_activate(metaslab_group_t *mg) { metaslab_class_t *mc = mg->mg_class; metaslab_group_t *mgprev, *mgnext; ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER)); ASSERT(mc->mc_rotor != mg); ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); ASSERT(mg->mg_activation_count <= 0); if (++mg->mg_activation_count <= 0) return; mg->mg_aliquot = metaslab_aliquot * MAX(1, mg->mg_vd->vdev_children); metaslab_group_alloc_update(mg); if ((mgprev = mc->mc_rotor) == NULL) { mg->mg_prev = mg; mg->mg_next = mg; } else { mgnext = mgprev->mg_next; mg->mg_prev = mgprev; mg->mg_next = mgnext; mgprev->mg_next = mg; mgnext->mg_prev = mg; } mc->mc_rotor = mg; metaslab_class_minblocksize_update(mc); } void metaslab_group_passivate(metaslab_group_t *mg) { metaslab_class_t *mc = mg->mg_class; metaslab_group_t *mgprev, *mgnext; ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER)); if (--mg->mg_activation_count != 0) { ASSERT(mc->mc_rotor != mg); ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); ASSERT(mg->mg_activation_count < 0); return; } taskq_wait(mg->mg_taskq); metaslab_group_alloc_update(mg); mgprev = mg->mg_prev; mgnext = mg->mg_next; if (mg == mgnext) { mc->mc_rotor = NULL; } else { mc->mc_rotor = mgnext; mgprev->mg_next = mgnext; mgnext->mg_prev = mgprev; } mg->mg_prev = NULL; mg->mg_next = NULL; metaslab_class_minblocksize_update(mc); } boolean_t metaslab_group_initialized(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; vdev_stat_t *vs = &vd->vdev_stat; return (vs->vs_space != 0 && mg->mg_activation_count > 0); } uint64_t metaslab_group_get_space(metaslab_group_t *mg) { return ((1ULL << mg->mg_vd->vdev_ms_shift) * mg->mg_vd->vdev_ms_count); } void metaslab_group_histogram_verify(metaslab_group_t *mg) { uint64_t *mg_hist; vdev_t *vd = mg->mg_vd; uint64_t ashift = vd->vdev_ashift; int i; if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0) return; mg_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE, KM_SLEEP); ASSERT3U(RANGE_TREE_HISTOGRAM_SIZE, >=, SPACE_MAP_HISTOGRAM_SIZE + ashift); for (int m = 0; m < vd->vdev_ms_count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp->ms_sm == NULL) continue; for (i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) mg_hist[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; } for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i ++) VERIFY3U(mg_hist[i], ==, mg->mg_histogram[i]); kmem_free(mg_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE); } static void metaslab_group_histogram_add(metaslab_group_t *mg, metaslab_t *msp) { metaslab_class_t *mc = mg->mg_class; uint64_t ashift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); if (msp->ms_sm == NULL) return; mutex_enter(&mg->mg_lock); for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { mg->mg_histogram[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; mc->mc_histogram[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; } mutex_exit(&mg->mg_lock); } void metaslab_group_histogram_remove(metaslab_group_t *mg, metaslab_t *msp) { metaslab_class_t *mc = mg->mg_class; uint64_t ashift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); if (msp->ms_sm == NULL) return; mutex_enter(&mg->mg_lock); for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { ASSERT3U(mg->mg_histogram[i + ashift], >=, msp->ms_sm->sm_phys->smp_histogram[i]); ASSERT3U(mc->mc_histogram[i + ashift], >=, msp->ms_sm->sm_phys->smp_histogram[i]); mg->mg_histogram[i + ashift] -= msp->ms_sm->sm_phys->smp_histogram[i]; mc->mc_histogram[i + ashift] -= msp->ms_sm->sm_phys->smp_histogram[i]; } mutex_exit(&mg->mg_lock); } static void metaslab_group_add(metaslab_group_t *mg, metaslab_t *msp) { ASSERT(msp->ms_group == NULL); mutex_enter(&mg->mg_lock); msp->ms_group = mg; msp->ms_weight = 0; avl_add(&mg->mg_metaslab_tree, msp); mutex_exit(&mg->mg_lock); mutex_enter(&msp->ms_lock); metaslab_group_histogram_add(mg, msp); mutex_exit(&msp->ms_lock); } static void metaslab_group_remove(metaslab_group_t *mg, metaslab_t *msp) { mutex_enter(&msp->ms_lock); metaslab_group_histogram_remove(mg, msp); mutex_exit(&msp->ms_lock); mutex_enter(&mg->mg_lock); ASSERT(msp->ms_group == mg); avl_remove(&mg->mg_metaslab_tree, msp); msp->ms_group = NULL; mutex_exit(&mg->mg_lock); } static void metaslab_group_sort(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight) { /* * Although in principle the weight can be any value, in * practice we do not use values in the range [1, 511]. */ ASSERT(weight >= SPA_MINBLOCKSIZE || weight == 0); ASSERT(MUTEX_HELD(&msp->ms_lock)); mutex_enter(&mg->mg_lock); ASSERT(msp->ms_group == mg); avl_remove(&mg->mg_metaslab_tree, msp); msp->ms_weight = weight; avl_add(&mg->mg_metaslab_tree, msp); mutex_exit(&mg->mg_lock); } /* * Calculate the fragmentation for a given metaslab group. We can use * a simple average here since all metaslabs within the group must have * the same size. The return value will be a value between 0 and 100 * (inclusive), or ZFS_FRAG_INVALID if less than half of the metaslab in this * group have a fragmentation metric. */ uint64_t metaslab_group_fragmentation(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; uint64_t fragmentation = 0; uint64_t valid_ms = 0; for (int m = 0; m < vd->vdev_ms_count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp->ms_fragmentation == ZFS_FRAG_INVALID) continue; valid_ms++; fragmentation += msp->ms_fragmentation; } if (valid_ms <= vd->vdev_ms_count / 2) return (ZFS_FRAG_INVALID); fragmentation /= valid_ms; ASSERT3U(fragmentation, <=, 100); return (fragmentation); } /* * Determine if a given metaslab group should skip allocations. A metaslab * group should avoid allocations if its free capacity is less than the * zfs_mg_noalloc_threshold or its fragmentation metric is greater than * zfs_mg_fragmentation_threshold and there is at least one metaslab group * that can still handle allocations. If the allocation throttle is enabled * then we skip allocations to devices that have reached their maximum * allocation queue depth unless the selected metaslab group is the only * eligible group remaining. */ static boolean_t metaslab_group_allocatable(metaslab_group_t *mg, metaslab_group_t *rotor, uint64_t psize) { spa_t *spa = mg->mg_vd->vdev_spa; metaslab_class_t *mc = mg->mg_class; /* * We can only consider skipping this metaslab group if it's * in the normal metaslab class and there are other metaslab * groups to select from. Otherwise, we always consider it eligible * for allocations. */ if (mc != spa_normal_class(spa) || mc->mc_groups <= 1) return (B_TRUE); /* * If the metaslab group's mg_allocatable flag is set (see comments * in metaslab_group_alloc_update() for more information) and * the allocation throttle is disabled then allow allocations to this * device. However, if the allocation throttle is enabled then * check if we have reached our allocation limit (mg_alloc_queue_depth) * to determine if we should allow allocations to this metaslab group. * If all metaslab groups are no longer considered allocatable * (mc_alloc_groups == 0) or we're trying to allocate the smallest * gang block size then we allow allocations on this metaslab group * regardless of the mg_allocatable or throttle settings. */ if (mg->mg_allocatable) { metaslab_group_t *mgp; int64_t qdepth; uint64_t qmax = mg->mg_max_alloc_queue_depth; if (!mc->mc_alloc_throttle_enabled) return (B_TRUE); /* * If this metaslab group does not have any free space, then * there is no point in looking further. */ if (mg->mg_no_free_space) return (B_FALSE); qdepth = refcount_count(&mg->mg_alloc_queue_depth); /* * If this metaslab group is below its qmax or it's * the only allocatable metasable group, then attempt * to allocate from it. */ if (qdepth < qmax || mc->mc_alloc_groups == 1) return (B_TRUE); ASSERT3U(mc->mc_alloc_groups, >, 1); /* * Since this metaslab group is at or over its qmax, we * need to determine if there are metaslab groups after this * one that might be able to handle this allocation. This is * racy since we can't hold the locks for all metaslab * groups at the same time when we make this check. */ for (mgp = mg->mg_next; mgp != rotor; mgp = mgp->mg_next) { qmax = mgp->mg_max_alloc_queue_depth; qdepth = refcount_count(&mgp->mg_alloc_queue_depth); /* * If there is another metaslab group that * might be able to handle the allocation, then * we return false so that we skip this group. */ if (qdepth < qmax && !mgp->mg_no_free_space) return (B_FALSE); } /* * We didn't find another group to handle the allocation * so we can't skip this metaslab group even though * we are at or over our qmax. */ return (B_TRUE); } else if (mc->mc_alloc_groups == 0 || psize == SPA_MINBLOCKSIZE) { return (B_TRUE); } return (B_FALSE); } /* * ========================================================================== * Range tree callbacks * ========================================================================== */ /* * Comparison function for the private size-ordered tree. Tree is sorted * by size, larger sizes at the end of the tree. */ static int metaslab_rangesize_compare(const void *x1, const void *x2) { const range_seg_t *r1 = x1; const range_seg_t *r2 = x2; uint64_t rs_size1 = r1->rs_end - r1->rs_start; uint64_t rs_size2 = r2->rs_end - r2->rs_start; if (rs_size1 < rs_size2) return (-1); if (rs_size1 > rs_size2) return (1); if (r1->rs_start < r2->rs_start) return (-1); if (r1->rs_start > r2->rs_start) return (1); return (0); } /* * Create any block allocator specific components. The current allocators * rely on using both a size-ordered range_tree_t and an array of uint64_t's. */ static void metaslab_rt_create(range_tree_t *rt, void *arg) { metaslab_t *msp = arg; ASSERT3P(rt->rt_arg, ==, msp); ASSERT(msp->ms_tree == NULL); avl_create(&msp->ms_size_tree, metaslab_rangesize_compare, sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node)); } /* * Destroy the block allocator specific components. */ static void metaslab_rt_destroy(range_tree_t *rt, void *arg) { metaslab_t *msp = arg; ASSERT3P(rt->rt_arg, ==, msp); ASSERT3P(msp->ms_tree, ==, rt); ASSERT0(avl_numnodes(&msp->ms_size_tree)); avl_destroy(&msp->ms_size_tree); } static void metaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg) { metaslab_t *msp = arg; ASSERT3P(rt->rt_arg, ==, msp); ASSERT3P(msp->ms_tree, ==, rt); VERIFY(!msp->ms_condensing); avl_add(&msp->ms_size_tree, rs); } static void metaslab_rt_remove(range_tree_t *rt, range_seg_t *rs, void *arg) { metaslab_t *msp = arg; ASSERT3P(rt->rt_arg, ==, msp); ASSERT3P(msp->ms_tree, ==, rt); VERIFY(!msp->ms_condensing); avl_remove(&msp->ms_size_tree, rs); } static void metaslab_rt_vacate(range_tree_t *rt, void *arg) { metaslab_t *msp = arg; ASSERT3P(rt->rt_arg, ==, msp); ASSERT3P(msp->ms_tree, ==, rt); /* * Normally one would walk the tree freeing nodes along the way. * Since the nodes are shared with the range trees we can avoid * walking all nodes and just reinitialize the avl tree. The nodes * will be freed by the range tree, so we don't want to free them here. */ avl_create(&msp->ms_size_tree, metaslab_rangesize_compare, sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node)); } static range_tree_ops_t metaslab_rt_ops = { metaslab_rt_create, metaslab_rt_destroy, metaslab_rt_add, metaslab_rt_remove, metaslab_rt_vacate }; /* * ========================================================================== * Common allocator routines * ========================================================================== */ /* * Return the maximum contiguous segment within the metaslab. */ uint64_t metaslab_block_maxsize(metaslab_t *msp) { avl_tree_t *t = &msp->ms_size_tree; range_seg_t *rs; if (t == NULL || (rs = avl_last(t)) == NULL) return (0ULL); return (rs->rs_end - rs->rs_start); } static range_seg_t * metaslab_block_find(avl_tree_t *t, uint64_t start, uint64_t size) { range_seg_t *rs, rsearch; avl_index_t where; rsearch.rs_start = start; rsearch.rs_end = start + size; rs = avl_find(t, &rsearch, &where); if (rs == NULL) { rs = avl_nearest(t, where, AVL_AFTER); } return (rs); } /* * This is a helper function that can be used by the allocator to find * a suitable block to allocate. This will search the specified AVL * tree looking for a block that matches the specified criteria. */ static uint64_t metaslab_block_picker(avl_tree_t *t, uint64_t *cursor, uint64_t size, uint64_t align) { range_seg_t *rs = metaslab_block_find(t, *cursor, size); while (rs != NULL) { uint64_t offset = P2ROUNDUP(rs->rs_start, align); if (offset + size <= rs->rs_end) { *cursor = offset + size; return (offset); } rs = AVL_NEXT(t, rs); } /* * If we know we've searched the whole map (*cursor == 0), give up. * Otherwise, reset the cursor to the beginning and try again. */ if (*cursor == 0) return (-1ULL); *cursor = 0; return (metaslab_block_picker(t, cursor, size, align)); } /* * ========================================================================== * The first-fit block allocator * ========================================================================== */ static uint64_t metaslab_ff_alloc(metaslab_t *msp, uint64_t size) { /* * Find the largest power of 2 block size that evenly divides the * requested size. This is used to try to allocate blocks with similar * alignment from the same area of the metaslab (i.e. same cursor * bucket) but it does not guarantee that other allocations sizes * may exist in the same region. */ uint64_t align = size & -size; uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1]; avl_tree_t *t = &msp->ms_tree->rt_root; return (metaslab_block_picker(t, cursor, size, align)); } static metaslab_ops_t metaslab_ff_ops = { metaslab_ff_alloc }; /* * ========================================================================== * Dynamic block allocator - * Uses the first fit allocation scheme until space get low and then * adjusts to a best fit allocation method. Uses metaslab_df_alloc_threshold * and metaslab_df_free_pct to determine when to switch the allocation scheme. * ========================================================================== */ static uint64_t metaslab_df_alloc(metaslab_t *msp, uint64_t size) { /* * Find the largest power of 2 block size that evenly divides the * requested size. This is used to try to allocate blocks with similar * alignment from the same area of the metaslab (i.e. same cursor * bucket) but it does not guarantee that other allocations sizes * may exist in the same region. */ uint64_t align = size & -size; uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1]; range_tree_t *rt = msp->ms_tree; avl_tree_t *t = &rt->rt_root; uint64_t max_size = metaslab_block_maxsize(msp); int free_pct = range_tree_space(rt) * 100 / msp->ms_size; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree)); if (max_size < size) return (-1ULL); /* * If we're running low on space switch to using the size * sorted AVL tree (best-fit). */ if (max_size < metaslab_df_alloc_threshold || free_pct < metaslab_df_free_pct) { t = &msp->ms_size_tree; *cursor = 0; } return (metaslab_block_picker(t, cursor, size, 1ULL)); } static metaslab_ops_t metaslab_df_ops = { metaslab_df_alloc }; /* * ========================================================================== * Cursor fit block allocator - * Select the largest region in the metaslab, set the cursor to the beginning * of the range and the cursor_end to the end of the range. As allocations * are made advance the cursor. Continue allocating from the cursor until * the range is exhausted and then find a new range. * ========================================================================== */ static uint64_t metaslab_cf_alloc(metaslab_t *msp, uint64_t size) { range_tree_t *rt = msp->ms_tree; avl_tree_t *t = &msp->ms_size_tree; uint64_t *cursor = &msp->ms_lbas[0]; uint64_t *cursor_end = &msp->ms_lbas[1]; uint64_t offset = 0; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&rt->rt_root)); ASSERT3U(*cursor_end, >=, *cursor); if ((*cursor + size) > *cursor_end) { range_seg_t *rs; rs = avl_last(&msp->ms_size_tree); if (rs == NULL || (rs->rs_end - rs->rs_start) < size) return (-1ULL); *cursor = rs->rs_start; *cursor_end = rs->rs_end; } offset = *cursor; *cursor += size; return (offset); } static metaslab_ops_t metaslab_cf_ops = { metaslab_cf_alloc }; /* * ========================================================================== * New dynamic fit allocator - * Select a region that is large enough to allocate 2^metaslab_ndf_clump_shift * contiguous blocks. If no region is found then just use the largest segment * that remains. * ========================================================================== */ /* * Determines desired number of contiguous blocks (2^metaslab_ndf_clump_shift) * to request from the allocator. */ uint64_t metaslab_ndf_clump_shift = 4; static uint64_t metaslab_ndf_alloc(metaslab_t *msp, uint64_t size) { avl_tree_t *t = &msp->ms_tree->rt_root; avl_index_t where; range_seg_t *rs, rsearch; uint64_t hbit = highbit64(size); uint64_t *cursor = &msp->ms_lbas[hbit - 1]; uint64_t max_size = metaslab_block_maxsize(msp); ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree)); if (max_size < size) return (-1ULL); rsearch.rs_start = *cursor; rsearch.rs_end = *cursor + size; rs = avl_find(t, &rsearch, &where); if (rs == NULL || (rs->rs_end - rs->rs_start) < size) { t = &msp->ms_size_tree; rsearch.rs_start = 0; rsearch.rs_end = MIN(max_size, 1ULL << (hbit + metaslab_ndf_clump_shift)); rs = avl_find(t, &rsearch, &where); if (rs == NULL) rs = avl_nearest(t, where, AVL_AFTER); ASSERT(rs != NULL); } if ((rs->rs_end - rs->rs_start) >= size) { *cursor = rs->rs_start + size; return (rs->rs_start); } return (-1ULL); } static metaslab_ops_t metaslab_ndf_ops = { metaslab_ndf_alloc }; metaslab_ops_t *zfs_metaslab_ops = &metaslab_df_ops; /* * ========================================================================== * Metaslabs * ========================================================================== */ /* * Wait for any in-progress metaslab loads to complete. */ void metaslab_load_wait(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); while (msp->ms_loading) { ASSERT(!msp->ms_loaded); cv_wait(&msp->ms_load_cv, &msp->ms_lock); } } int metaslab_load(metaslab_t *msp) { int error = 0; boolean_t success = B_FALSE; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(!msp->ms_loaded); ASSERT(!msp->ms_loading); msp->ms_loading = B_TRUE; /* * If the space map has not been allocated yet, then treat * all the space in the metaslab as free and add it to the * ms_tree. */ if (msp->ms_sm != NULL) error = space_map_load(msp->ms_sm, msp->ms_tree, SM_FREE); else range_tree_add(msp->ms_tree, msp->ms_start, msp->ms_size); success = (error == 0); msp->ms_loading = B_FALSE; if (success) { ASSERT3P(msp->ms_group, !=, NULL); msp->ms_loaded = B_TRUE; for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_walk(msp->ms_defertree[t], range_tree_remove, msp->ms_tree); } msp->ms_max_size = metaslab_block_maxsize(msp); } cv_broadcast(&msp->ms_load_cv); return (error); } void metaslab_unload(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); range_tree_vacate(msp->ms_tree, NULL, NULL); msp->ms_loaded = B_FALSE; msp->ms_weight &= ~METASLAB_ACTIVE_MASK; msp->ms_max_size = 0; } int metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg, metaslab_t **msp) { vdev_t *vd = mg->mg_vd; objset_t *mos = vd->vdev_spa->spa_meta_objset; metaslab_t *ms; int error; ms = kmem_zalloc(sizeof (metaslab_t), KM_SLEEP); mutex_init(&ms->ms_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&ms->ms_load_cv, NULL, CV_DEFAULT, NULL); ms->ms_id = id; ms->ms_start = id << vd->vdev_ms_shift; ms->ms_size = 1ULL << vd->vdev_ms_shift; /* * We only open space map objects that already exist. All others * will be opened when we finally allocate an object for it. */ if (object != 0) { error = space_map_open(&ms->ms_sm, mos, object, ms->ms_start, ms->ms_size, vd->vdev_ashift, &ms->ms_lock); if (error != 0) { kmem_free(ms, sizeof (metaslab_t)); return (error); } ASSERT(ms->ms_sm != NULL); } /* * We create the main range tree here, but we don't create the * other range trees until metaslab_sync_done(). This serves * two purposes: it allows metaslab_sync_done() to detect the * addition of new space; and for debugging, it ensures that we'd * data fault on any attempt to use this metaslab before it's ready. */ ms->ms_tree = range_tree_create(&metaslab_rt_ops, ms, &ms->ms_lock); metaslab_group_add(mg, ms); metaslab_set_fragmentation(ms); /* * If we're opening an existing pool (txg == 0) or creating * a new one (txg == TXG_INITIAL), all space is available now. * If we're adding space to an existing pool, the new space * does not become available until after this txg has synced. * The metaslab's weight will also be initialized when we sync * out this txg. This ensures that we don't attempt to allocate * from it before we have initialized it completely. */ if (txg <= TXG_INITIAL) metaslab_sync_done(ms, 0); /* * If metaslab_debug_load is set and we're initializing a metaslab * that has an allocated space map object then load the its space * map so that can verify frees. */ if (metaslab_debug_load && ms->ms_sm != NULL) { mutex_enter(&ms->ms_lock); VERIFY0(metaslab_load(ms)); mutex_exit(&ms->ms_lock); } if (txg != 0) { vdev_dirty(vd, 0, NULL, txg); vdev_dirty(vd, VDD_METASLAB, ms, txg); } *msp = ms; return (0); } void metaslab_fini(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; metaslab_group_remove(mg, msp); mutex_enter(&msp->ms_lock); VERIFY(msp->ms_group == NULL); vdev_space_update(mg->mg_vd, -space_map_allocated(msp->ms_sm), 0, -msp->ms_size); space_map_close(msp->ms_sm); metaslab_unload(msp); range_tree_destroy(msp->ms_tree); range_tree_destroy(msp->ms_freeingtree); range_tree_destroy(msp->ms_freedtree); for (int t = 0; t < TXG_SIZE; t++) { range_tree_destroy(msp->ms_alloctree[t]); } for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_destroy(msp->ms_defertree[t]); } ASSERT0(msp->ms_deferspace); mutex_exit(&msp->ms_lock); cv_destroy(&msp->ms_load_cv); mutex_destroy(&msp->ms_lock); kmem_free(msp, sizeof (metaslab_t)); } #define FRAGMENTATION_TABLE_SIZE 17 /* * This table defines a segment size based fragmentation metric that will * allow each metaslab to derive its own fragmentation value. This is done * by calculating the space in each bucket of the spacemap histogram and * multiplying that by the fragmetation metric in this table. Doing * this for all buckets and dividing it by the total amount of free * space in this metaslab (i.e. the total free space in all buckets) gives * us the fragmentation metric. This means that a high fragmentation metric * equates to most of the free space being comprised of small segments. * Conversely, if the metric is low, then most of the free space is in * large segments. A 10% change in fragmentation equates to approximately * double the number of segments. * * This table defines 0% fragmented space using 16MB segments. Testing has * shown that segments that are greater than or equal to 16MB do not suffer * from drastic performance problems. Using this value, we derive the rest * of the table. Since the fragmentation value is never stored on disk, it * is possible to change these calculations in the future. */ int zfs_frag_table[FRAGMENTATION_TABLE_SIZE] = { 100, /* 512B */ 100, /* 1K */ 98, /* 2K */ 95, /* 4K */ 90, /* 8K */ 80, /* 16K */ 70, /* 32K */ 60, /* 64K */ 50, /* 128K */ 40, /* 256K */ 30, /* 512K */ 20, /* 1M */ 15, /* 2M */ 10, /* 4M */ 5, /* 8M */ 0 /* 16M */ }; /* * Calclate the metaslab's fragmentation metric. A return value * of ZFS_FRAG_INVALID means that the metaslab has not been upgraded and does * not support this metric. Otherwise, the return value should be in the * range [0, 100]. */ static void metaslab_set_fragmentation(metaslab_t *msp) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; uint64_t fragmentation = 0; uint64_t total = 0; boolean_t feature_enabled = spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM); if (!feature_enabled) { msp->ms_fragmentation = ZFS_FRAG_INVALID; return; } /* * A null space map means that the entire metaslab is free * and thus is not fragmented. */ if (msp->ms_sm == NULL) { msp->ms_fragmentation = 0; return; } /* * If this metaslab's space map has not been upgraded, flag it * so that we upgrade next time we encounter it. */ if (msp->ms_sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) { uint64_t txg = spa_syncing_txg(spa); vdev_t *vd = msp->ms_group->mg_vd; - if (spa_writeable(spa)) { + /* + * If we've reached the final dirty txg, then we must + * be shutting down the pool. We don't want to dirty + * any data past this point so skip setting the condense + * flag. We can retry this action the next time the pool + * is imported. + */ + if (spa_writeable(spa) && txg < spa_final_dirty_txg(spa)) { msp->ms_condense_wanted = B_TRUE; vdev_dirty(vd, VDD_METASLAB, msp, txg + 1); spa_dbgmsg(spa, "txg %llu, requesting force condense: " - "msp %p, vd %p", txg, msp, vd); + "ms_id %llu, vdev_id %llu", txg, msp->ms_id, + vd->vdev_id); } msp->ms_fragmentation = ZFS_FRAG_INVALID; return; } for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { uint64_t space = 0; uint8_t shift = msp->ms_sm->sm_shift; int idx = MIN(shift - SPA_MINBLOCKSHIFT + i, FRAGMENTATION_TABLE_SIZE - 1); if (msp->ms_sm->sm_phys->smp_histogram[i] == 0) continue; space = msp->ms_sm->sm_phys->smp_histogram[i] << (i + shift); total += space; ASSERT3U(idx, <, FRAGMENTATION_TABLE_SIZE); fragmentation += space * zfs_frag_table[idx]; } if (total > 0) fragmentation /= total; ASSERT3U(fragmentation, <=, 100); msp->ms_fragmentation = fragmentation; } /* * Compute a weight -- a selection preference value -- for the given metaslab. * This is based on the amount of free space, the level of fragmentation, * the LBA range, and whether the metaslab is loaded. */ static uint64_t metaslab_space_weight(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; uint64_t weight, space; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(!vd->vdev_removing); /* * The baseline weight is the metaslab's free space. */ space = msp->ms_size - space_map_allocated(msp->ms_sm); if (metaslab_fragmentation_factor_enabled && msp->ms_fragmentation != ZFS_FRAG_INVALID) { /* * Use the fragmentation information to inversely scale * down the baseline weight. We need to ensure that we * don't exclude this metaslab completely when it's 100% * fragmented. To avoid this we reduce the fragmented value * by 1. */ space = (space * (100 - (msp->ms_fragmentation - 1))) / 100; /* * If space < SPA_MINBLOCKSIZE, then we will not allocate from * this metaslab again. The fragmentation metric may have * decreased the space to something smaller than * SPA_MINBLOCKSIZE, so reset the space to SPA_MINBLOCKSIZE * so that we can consume any remaining space. */ if (space > 0 && space < SPA_MINBLOCKSIZE) space = SPA_MINBLOCKSIZE; } weight = space; /* * Modern disks have uniform bit density and constant angular velocity. * Therefore, the outer recording zones are faster (higher bandwidth) * than the inner zones by the ratio of outer to inner track diameter, * which is typically around 2:1. We account for this by assigning * higher weight to lower metaslabs (multiplier ranging from 2x to 1x). * In effect, this means that we'll select the metaslab with the most * free bandwidth rather than simply the one with the most free space. */ if (metaslab_lba_weighting_enabled) { weight = 2 * weight - (msp->ms_id * weight) / vd->vdev_ms_count; ASSERT(weight >= space && weight <= 2 * space); } /* * If this metaslab is one we're actively using, adjust its * weight to make it preferable to any inactive metaslab so * we'll polish it off. If the fragmentation on this metaslab * has exceed our threshold, then don't mark it active. */ if (msp->ms_loaded && msp->ms_fragmentation != ZFS_FRAG_INVALID && msp->ms_fragmentation <= zfs_metaslab_fragmentation_threshold) { weight |= (msp->ms_weight & METASLAB_ACTIVE_MASK); } WEIGHT_SET_SPACEBASED(weight); return (weight); } /* * Return the weight of the specified metaslab, according to the segment-based * weighting algorithm. The metaslab must be loaded. This function can * be called within a sync pass since it relies only on the metaslab's * range tree which is always accurate when the metaslab is loaded. */ static uint64_t metaslab_weight_from_range_tree(metaslab_t *msp) { uint64_t weight = 0; uint32_t segments = 0; ASSERT(msp->ms_loaded); for (int i = RANGE_TREE_HISTOGRAM_SIZE - 1; i >= SPA_MINBLOCKSHIFT; i--) { uint8_t shift = msp->ms_group->mg_vd->vdev_ashift; int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1; segments <<= 1; segments += msp->ms_tree->rt_histogram[i]; /* * The range tree provides more precision than the space map * and must be downgraded so that all values fit within the * space map's histogram. This allows us to compare loaded * vs. unloaded metaslabs to determine which metaslab is * considered "best". */ if (i > max_idx) continue; if (segments != 0) { WEIGHT_SET_COUNT(weight, segments); WEIGHT_SET_INDEX(weight, i); WEIGHT_SET_ACTIVE(weight, 0); break; } } return (weight); } /* * Calculate the weight based on the on-disk histogram. This should only * be called after a sync pass has completely finished since the on-disk * information is updated in metaslab_sync(). */ static uint64_t metaslab_weight_from_spacemap(metaslab_t *msp) { uint64_t weight = 0; for (int i = SPACE_MAP_HISTOGRAM_SIZE - 1; i >= 0; i--) { if (msp->ms_sm->sm_phys->smp_histogram[i] != 0) { WEIGHT_SET_COUNT(weight, msp->ms_sm->sm_phys->smp_histogram[i]); WEIGHT_SET_INDEX(weight, i + msp->ms_sm->sm_shift); WEIGHT_SET_ACTIVE(weight, 0); break; } } return (weight); } /* * Compute a segment-based weight for the specified metaslab. The weight * is determined by highest bucket in the histogram. The information * for the highest bucket is encoded into the weight value. */ static uint64_t metaslab_segment_weight(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; uint64_t weight = 0; uint8_t shift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * The metaslab is completely free. */ if (space_map_allocated(msp->ms_sm) == 0) { int idx = highbit64(msp->ms_size) - 1; int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1; if (idx < max_idx) { WEIGHT_SET_COUNT(weight, 1ULL); WEIGHT_SET_INDEX(weight, idx); } else { WEIGHT_SET_COUNT(weight, 1ULL << (idx - max_idx)); WEIGHT_SET_INDEX(weight, max_idx); } WEIGHT_SET_ACTIVE(weight, 0); ASSERT(!WEIGHT_IS_SPACEBASED(weight)); return (weight); } ASSERT3U(msp->ms_sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t)); /* * If the metaslab is fully allocated then just make the weight 0. */ if (space_map_allocated(msp->ms_sm) == msp->ms_size) return (0); /* * If the metaslab is already loaded, then use the range tree to * determine the weight. Otherwise, we rely on the space map information * to generate the weight. */ if (msp->ms_loaded) { weight = metaslab_weight_from_range_tree(msp); } else { weight = metaslab_weight_from_spacemap(msp); } /* * If the metaslab was active the last time we calculated its weight * then keep it active. We want to consume the entire region that * is associated with this weight. */ if (msp->ms_activation_weight != 0 && weight != 0) WEIGHT_SET_ACTIVE(weight, WEIGHT_GET_ACTIVE(msp->ms_weight)); return (weight); } /* * Determine if we should attempt to allocate from this metaslab. If the * metaslab has a maximum size then we can quickly determine if the desired * allocation size can be satisfied. Otherwise, if we're using segment-based * weighting then we can determine the maximum allocation that this metaslab * can accommodate based on the index encoded in the weight. If we're using * space-based weights then rely on the entire weight (excluding the weight * type bit). */ boolean_t metaslab_should_allocate(metaslab_t *msp, uint64_t asize) { boolean_t should_allocate; if (msp->ms_max_size != 0) return (msp->ms_max_size >= asize); if (!WEIGHT_IS_SPACEBASED(msp->ms_weight)) { /* * The metaslab segment weight indicates segments in the * range [2^i, 2^(i+1)), where i is the index in the weight. * Since the asize might be in the middle of the range, we * should attempt the allocation if asize < 2^(i+1). */ should_allocate = (asize < 1ULL << (WEIGHT_GET_INDEX(msp->ms_weight) + 1)); } else { should_allocate = (asize <= (msp->ms_weight & ~METASLAB_WEIGHT_TYPE)); } return (should_allocate); } static uint64_t metaslab_weight(metaslab_t *msp) { vdev_t *vd = msp->ms_group->mg_vd; spa_t *spa = vd->vdev_spa; uint64_t weight; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * This vdev is in the process of being removed so there is nothing * for us to do here. */ if (vd->vdev_removing) { ASSERT0(space_map_allocated(msp->ms_sm)); ASSERT0(vd->vdev_ms_shift); return (0); } metaslab_set_fragmentation(msp); /* * Update the maximum size if the metaslab is loaded. This will * ensure that we get an accurate maximum size if newly freed space * has been added back into the free tree. */ if (msp->ms_loaded) msp->ms_max_size = metaslab_block_maxsize(msp); /* * Segment-based weighting requires space map histogram support. */ if (zfs_metaslab_segment_weight_enabled && spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM) && (msp->ms_sm == NULL || msp->ms_sm->sm_dbuf->db_size == sizeof (space_map_phys_t))) { weight = metaslab_segment_weight(msp); } else { weight = metaslab_space_weight(msp); } return (weight); } static int metaslab_activate(metaslab_t *msp, uint64_t activation_weight) { ASSERT(MUTEX_HELD(&msp->ms_lock)); if ((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0) { metaslab_load_wait(msp); if (!msp->ms_loaded) { int error = metaslab_load(msp); if (error) { metaslab_group_sort(msp->ms_group, msp, 0); return (error); } } msp->ms_activation_weight = msp->ms_weight; metaslab_group_sort(msp->ms_group, msp, msp->ms_weight | activation_weight); } ASSERT(msp->ms_loaded); ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK); return (0); } static void metaslab_passivate(metaslab_t *msp, uint64_t weight) { uint64_t size = weight & ~METASLAB_WEIGHT_TYPE; /* * If size < SPA_MINBLOCKSIZE, then we will not allocate from * this metaslab again. In that case, it had better be empty, * or we would be leaving space on the table. */ ASSERT(size >= SPA_MINBLOCKSIZE || range_tree_space(msp->ms_tree) == 0); ASSERT0(weight & METASLAB_ACTIVE_MASK); msp->ms_activation_weight = 0; metaslab_group_sort(msp->ms_group, msp, weight); ASSERT((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0); } /* * Segment-based metaslabs are activated once and remain active until * we either fail an allocation attempt (similar to space-based metaslabs) * or have exhausted the free space in zfs_metaslab_switch_threshold * buckets since the metaslab was activated. This function checks to see * if we've exhaused the zfs_metaslab_switch_threshold buckets in the * metaslab and passivates it proactively. This will allow us to select a * metaslabs with larger contiguous region if any remaining within this * metaslab group. If we're in sync pass > 1, then we continue using this * metaslab so that we don't dirty more block and cause more sync passes. */ void metaslab_segment_may_passivate(metaslab_t *msp) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; if (WEIGHT_IS_SPACEBASED(msp->ms_weight) || spa_sync_pass(spa) > 1) return; /* * Since we are in the middle of a sync pass, the most accurate * information that is accessible to us is the in-core range tree * histogram; calculate the new weight based on that information. */ uint64_t weight = metaslab_weight_from_range_tree(msp); int activation_idx = WEIGHT_GET_INDEX(msp->ms_activation_weight); int current_idx = WEIGHT_GET_INDEX(weight); if (current_idx <= activation_idx - zfs_metaslab_switch_threshold) metaslab_passivate(msp, weight); } static void metaslab_preload(void *arg) { metaslab_t *msp = arg; spa_t *spa = msp->ms_group->mg_vd->vdev_spa; ASSERT(!MUTEX_HELD(&msp->ms_group->mg_lock)); mutex_enter(&msp->ms_lock); metaslab_load_wait(msp); if (!msp->ms_loaded) (void) metaslab_load(msp); msp->ms_selected_txg = spa_syncing_txg(spa); mutex_exit(&msp->ms_lock); } static void metaslab_group_preload(metaslab_group_t *mg) { spa_t *spa = mg->mg_vd->vdev_spa; metaslab_t *msp; avl_tree_t *t = &mg->mg_metaslab_tree; int m = 0; if (spa_shutting_down(spa) || !metaslab_preload_enabled) { taskq_wait(mg->mg_taskq); return; } mutex_enter(&mg->mg_lock); /* * Load the next potential metaslabs */ for (msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) { /* * We preload only the maximum number of metaslabs specified * by metaslab_preload_limit. If a metaslab is being forced * to condense then we preload it too. This will ensure * that force condensing happens in the next txg. */ if (++m > metaslab_preload_limit && !msp->ms_condense_wanted) { continue; } VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload, msp, TQ_SLEEP) != 0); } mutex_exit(&mg->mg_lock); } /* * Determine if the space map's on-disk footprint is past our tolerance * for inefficiency. We would like to use the following criteria to make * our decision: * * 1. The size of the space map object should not dramatically increase as a * result of writing out the free space range tree. * * 2. The minimal on-disk space map representation is zfs_condense_pct/100 * times the size than the free space range tree representation * (i.e. zfs_condense_pct = 110 and in-core = 1MB, minimal = 1.1.MB). * * 3. The on-disk size of the space map should actually decrease. * * Checking the first condition is tricky since we don't want to walk * the entire AVL tree calculating the estimated on-disk size. Instead we * use the size-ordered range tree in the metaslab and calculate the * size required to write out the largest segment in our free tree. If the * size required to represent that segment on disk is larger than the space * map object then we avoid condensing this map. * * To determine the second criterion we use a best-case estimate and assume * each segment can be represented on-disk as a single 64-bit entry. We refer * to this best-case estimate as the space map's minimal form. * * Unfortunately, we cannot compute the on-disk size of the space map in this * context because we cannot accurately compute the effects of compression, etc. * Instead, we apply the heuristic described in the block comment for * zfs_metaslab_condense_block_threshold - we only condense if the space used * is greater than a threshold number of blocks. */ static boolean_t metaslab_should_condense(metaslab_t *msp) { space_map_t *sm = msp->ms_sm; range_seg_t *rs; uint64_t size, entries, segsz, object_size, optimal_size, record_size; dmu_object_info_t doi; uint64_t vdev_blocksize = 1 << msp->ms_group->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(msp->ms_loaded); /* * Use the ms_size_tree range tree, which is ordered by size, to * obtain the largest segment in the free tree. We always condense * metaslabs that are empty and metaslabs for which a condense * request has been made. */ rs = avl_last(&msp->ms_size_tree); if (rs == NULL || msp->ms_condense_wanted) return (B_TRUE); /* * Calculate the number of 64-bit entries this segment would * require when written to disk. If this single segment would be * larger on-disk than the entire current on-disk structure, then * clearly condensing will increase the on-disk structure size. */ size = (rs->rs_end - rs->rs_start) >> sm->sm_shift; entries = size / (MIN(size, SM_RUN_MAX)); segsz = entries * sizeof (uint64_t); optimal_size = sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root); object_size = space_map_length(msp->ms_sm); dmu_object_info_from_db(sm->sm_dbuf, &doi); record_size = MAX(doi.doi_data_block_size, vdev_blocksize); return (segsz <= object_size && object_size >= (optimal_size * zfs_condense_pct / 100) && object_size > zfs_metaslab_condense_block_threshold * record_size); } /* * Condense the on-disk space map representation to its minimized form. * The minimized form consists of a small number of allocations followed by * the entries of the free range tree. */ static void metaslab_condense(metaslab_t *msp, uint64_t txg, dmu_tx_t *tx) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; range_tree_t *condense_tree; space_map_t *sm = msp->ms_sm; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(spa_sync_pass(spa), ==, 1); ASSERT(msp->ms_loaded); spa_dbgmsg(spa, "condensing: txg %llu, msp[%llu] %p, vdev id %llu, " "spa %s, smp size %llu, segments %lu, forcing condense=%s", txg, msp->ms_id, msp, msp->ms_group->mg_vd->vdev_id, msp->ms_group->mg_vd->vdev_spa->spa_name, space_map_length(msp->ms_sm), avl_numnodes(&msp->ms_tree->rt_root), msp->ms_condense_wanted ? "TRUE" : "FALSE"); msp->ms_condense_wanted = B_FALSE; /* * Create an range tree that is 100% allocated. We remove segments * that have been freed in this txg, any deferred frees that exist, * and any allocation in the future. Removing segments should be * a relatively inexpensive operation since we expect these trees to * have a small number of nodes. */ condense_tree = range_tree_create(NULL, NULL, &msp->ms_lock); range_tree_add(condense_tree, msp->ms_start, msp->ms_size); /* * Remove what's been freed in this txg from the condense_tree. * Since we're in sync_pass 1, we know that all the frees from * this txg are in the freeingtree. */ range_tree_walk(msp->ms_freeingtree, range_tree_remove, condense_tree); for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_walk(msp->ms_defertree[t], range_tree_remove, condense_tree); } for (int t = 1; t < TXG_CONCURRENT_STATES; t++) { range_tree_walk(msp->ms_alloctree[(txg + t) & TXG_MASK], range_tree_remove, condense_tree); } /* * We're about to drop the metaslab's lock thus allowing * other consumers to change it's content. Set the * metaslab's ms_condensing flag to ensure that * allocations on this metaslab do not occur while we're * in the middle of committing it to disk. This is only critical * for the ms_tree as all other range trees use per txg * views of their content. */ msp->ms_condensing = B_TRUE; mutex_exit(&msp->ms_lock); space_map_truncate(sm, tx); mutex_enter(&msp->ms_lock); /* * While we would ideally like to create a space map representation * that consists only of allocation records, doing so can be * prohibitively expensive because the in-core free tree can be * large, and therefore computationally expensive to subtract * from the condense_tree. Instead we sync out two trees, a cheap * allocation only tree followed by the in-core free tree. While not * optimal, this is typically close to optimal, and much cheaper to * compute. */ space_map_write(sm, condense_tree, SM_ALLOC, tx); range_tree_vacate(condense_tree, NULL, NULL); range_tree_destroy(condense_tree); space_map_write(sm, msp->ms_tree, SM_FREE, tx); msp->ms_condensing = B_FALSE; } /* * Write a metaslab to disk in the context of the specified transaction group. */ void metaslab_sync(metaslab_t *msp, uint64_t txg) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; objset_t *mos = spa_meta_objset(spa); range_tree_t *alloctree = msp->ms_alloctree[txg & TXG_MASK]; dmu_tx_t *tx; uint64_t object = space_map_object(msp->ms_sm); ASSERT(!vd->vdev_ishole); /* * This metaslab has just been added so there's no work to do now. */ if (msp->ms_freeingtree == NULL) { ASSERT3P(alloctree, ==, NULL); return; } ASSERT3P(alloctree, !=, NULL); ASSERT3P(msp->ms_freeingtree, !=, NULL); ASSERT3P(msp->ms_freedtree, !=, NULL); /* * Normally, we don't want to process a metaslab if there * are no allocations or frees to perform. However, if the metaslab - * is being forced to condense we need to let it through. + * is being forced to condense and it's loaded, we need to let it + * through. */ if (range_tree_space(alloctree) == 0 && range_tree_space(msp->ms_freeingtree) == 0 && - !msp->ms_condense_wanted) + !(msp->ms_loaded && msp->ms_condense_wanted)) return; + + + VERIFY(txg <= spa_final_dirty_txg(spa)); /* * The only state that can actually be changing concurrently with * metaslab_sync() is the metaslab's ms_tree. No other thread can * be modifying this txg's alloctree, freeingtree, freedtree, or * space_map_phys_t. Therefore, we only hold ms_lock to satify * space map ASSERTs. We drop it whenever we call into the DMU, * because the DMU can call down to us (e.g. via zio_free()) at * any time. */ tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); if (msp->ms_sm == NULL) { uint64_t new_object; new_object = space_map_alloc(mos, tx); VERIFY3U(new_object, !=, 0); VERIFY0(space_map_open(&msp->ms_sm, mos, new_object, msp->ms_start, msp->ms_size, vd->vdev_ashift, &msp->ms_lock)); ASSERT(msp->ms_sm != NULL); } mutex_enter(&msp->ms_lock); /* * Note: metaslab_condense() clears the space map's histogram. * Therefore we must verify and remove this histogram before * condensing. */ metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); metaslab_group_histogram_remove(mg, msp); if (msp->ms_loaded && spa_sync_pass(spa) == 1 && metaslab_should_condense(msp)) { metaslab_condense(msp, txg, tx); } else { space_map_write(msp->ms_sm, alloctree, SM_ALLOC, tx); space_map_write(msp->ms_sm, msp->ms_freeingtree, SM_FREE, tx); } if (msp->ms_loaded) { /* * When the space map is loaded, we have an accruate * histogram in the range tree. This gives us an opportunity * to bring the space map's histogram up-to-date so we clear * it first before updating it. */ space_map_histogram_clear(msp->ms_sm); space_map_histogram_add(msp->ms_sm, msp->ms_tree, tx); /* * Since we've cleared the histogram we need to add back * any free space that has already been processed, plus * any deferred space. This allows the on-disk histogram * to accurately reflect all free space even if some space * is not yet available for allocation (i.e. deferred). */ space_map_histogram_add(msp->ms_sm, msp->ms_freedtree, tx); /* * Add back any deferred free space that has not been * added back into the in-core free tree yet. This will * ensure that we don't end up with a space map histogram * that is completely empty unless the metaslab is fully * allocated. */ for (int t = 0; t < TXG_DEFER_SIZE; t++) { space_map_histogram_add(msp->ms_sm, msp->ms_defertree[t], tx); } } /* * Always add the free space from this sync pass to the space * map histogram. We want to make sure that the on-disk histogram * accounts for all free space. If the space map is not loaded, * then we will lose some accuracy but will correct it the next * time we load the space map. */ space_map_histogram_add(msp->ms_sm, msp->ms_freeingtree, tx); metaslab_group_histogram_add(mg, msp); metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); /* * For sync pass 1, we avoid traversing this txg's free range tree * and instead will just swap the pointers for freeingtree and * freedtree. We can safely do this since the freed_tree is * guaranteed to be empty on the initial pass. */ if (spa_sync_pass(spa) == 1) { range_tree_swap(&msp->ms_freeingtree, &msp->ms_freedtree); } else { range_tree_vacate(msp->ms_freeingtree, range_tree_add, msp->ms_freedtree); } range_tree_vacate(alloctree, NULL, NULL); ASSERT0(range_tree_space(msp->ms_alloctree[txg & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_alloctree[TXG_CLEAN(txg) & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_freeingtree)); mutex_exit(&msp->ms_lock); if (object != space_map_object(msp->ms_sm)) { object = space_map_object(msp->ms_sm); dmu_write(mos, vd->vdev_ms_array, sizeof (uint64_t) * msp->ms_id, sizeof (uint64_t), &object, tx); } dmu_tx_commit(tx); } /* * Called after a transaction group has completely synced to mark * all of the metaslab's free space as usable. */ void metaslab_sync_done(metaslab_t *msp, uint64_t txg) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; range_tree_t **defer_tree; int64_t alloc_delta, defer_delta; boolean_t defer_allowed = B_TRUE; ASSERT(!vd->vdev_ishole); mutex_enter(&msp->ms_lock); /* * If this metaslab is just becoming available, initialize its * range trees and add its capacity to the vdev. */ if (msp->ms_freedtree == NULL) { for (int t = 0; t < TXG_SIZE; t++) { ASSERT(msp->ms_alloctree[t] == NULL); msp->ms_alloctree[t] = range_tree_create(NULL, msp, &msp->ms_lock); } ASSERT3P(msp->ms_freeingtree, ==, NULL); msp->ms_freeingtree = range_tree_create(NULL, msp, &msp->ms_lock); ASSERT3P(msp->ms_freedtree, ==, NULL); msp->ms_freedtree = range_tree_create(NULL, msp, &msp->ms_lock); for (int t = 0; t < TXG_DEFER_SIZE; t++) { ASSERT(msp->ms_defertree[t] == NULL); msp->ms_defertree[t] = range_tree_create(NULL, msp, &msp->ms_lock); } vdev_space_update(vd, 0, 0, msp->ms_size); } defer_tree = &msp->ms_defertree[txg % TXG_DEFER_SIZE]; uint64_t free_space = metaslab_class_get_space(spa_normal_class(spa)) - metaslab_class_get_alloc(spa_normal_class(spa)); if (free_space <= spa_get_slop_space(spa)) { defer_allowed = B_FALSE; } defer_delta = 0; alloc_delta = space_map_alloc_delta(msp->ms_sm); if (defer_allowed) { defer_delta = range_tree_space(msp->ms_freedtree) - range_tree_space(*defer_tree); } else { defer_delta -= range_tree_space(*defer_tree); } vdev_space_update(vd, alloc_delta + defer_delta, defer_delta, 0); /* * If there's a metaslab_load() in progress, wait for it to complete * so that we have a consistent view of the in-core space map. */ metaslab_load_wait(msp); /* * Move the frees from the defer_tree back to the free * range tree (if it's loaded). Swap the freed_tree and the * defer_tree -- this is safe to do because we've just emptied out * the defer_tree. */ range_tree_vacate(*defer_tree, msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree); if (defer_allowed) { range_tree_swap(&msp->ms_freedtree, defer_tree); } else { range_tree_vacate(msp->ms_freedtree, msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree); } space_map_update(msp->ms_sm); msp->ms_deferspace += defer_delta; ASSERT3S(msp->ms_deferspace, >=, 0); ASSERT3S(msp->ms_deferspace, <=, msp->ms_size); if (msp->ms_deferspace != 0) { /* * Keep syncing this metaslab until all deferred frees * are back in circulation. */ vdev_dirty(vd, VDD_METASLAB, msp, txg + 1); } /* * Calculate the new weights before unloading any metaslabs. * This will give us the most accurate weighting. */ metaslab_group_sort(mg, msp, metaslab_weight(msp)); /* * If the metaslab is loaded and we've not tried to load or allocate * from it in 'metaslab_unload_delay' txgs, then unload it. */ if (msp->ms_loaded && msp->ms_selected_txg + metaslab_unload_delay < txg) { for (int t = 1; t < TXG_CONCURRENT_STATES; t++) { VERIFY0(range_tree_space( msp->ms_alloctree[(txg + t) & TXG_MASK])); } if (!metaslab_debug_unload) metaslab_unload(msp); } mutex_exit(&msp->ms_lock); } void metaslab_sync_reassess(metaslab_group_t *mg) { metaslab_group_alloc_update(mg); mg->mg_fragmentation = metaslab_group_fragmentation(mg); /* * Preload the next potential metaslabs */ metaslab_group_preload(mg); } static uint64_t metaslab_distance(metaslab_t *msp, dva_t *dva) { uint64_t ms_shift = msp->ms_group->mg_vd->vdev_ms_shift; uint64_t offset = DVA_GET_OFFSET(dva) >> ms_shift; uint64_t start = msp->ms_id; if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva)) return (1ULL << 63); if (offset < start) return ((start - offset) << ms_shift); if (offset > start) return ((offset - start) << ms_shift); return (0); } /* * ========================================================================== * Metaslab allocation tracing facility * ========================================================================== */ kstat_t *metaslab_trace_ksp; kstat_named_t metaslab_trace_over_limit; void metaslab_alloc_trace_init(void) { ASSERT(metaslab_alloc_trace_cache == NULL); metaslab_alloc_trace_cache = kmem_cache_create( "metaslab_alloc_trace_cache", sizeof (metaslab_alloc_trace_t), 0, NULL, NULL, NULL, NULL, NULL, 0); metaslab_trace_ksp = kstat_create("zfs", 0, "metaslab_trace_stats", "misc", KSTAT_TYPE_NAMED, 1, KSTAT_FLAG_VIRTUAL); if (metaslab_trace_ksp != NULL) { metaslab_trace_ksp->ks_data = &metaslab_trace_over_limit; kstat_named_init(&metaslab_trace_over_limit, "metaslab_trace_over_limit", KSTAT_DATA_UINT64); kstat_install(metaslab_trace_ksp); } } void metaslab_alloc_trace_fini(void) { if (metaslab_trace_ksp != NULL) { kstat_delete(metaslab_trace_ksp); metaslab_trace_ksp = NULL; } kmem_cache_destroy(metaslab_alloc_trace_cache); metaslab_alloc_trace_cache = NULL; } /* * Add an allocation trace element to the allocation tracing list. */ static void metaslab_trace_add(zio_alloc_list_t *zal, metaslab_group_t *mg, metaslab_t *msp, uint64_t psize, uint32_t dva_id, uint64_t offset) { if (!metaslab_trace_enabled) return; /* * When the tracing list reaches its maximum we remove * the second element in the list before adding a new one. * By removing the second element we preserve the original * entry as a clue to what allocations steps have already been * performed. */ if (zal->zal_size == metaslab_trace_max_entries) { metaslab_alloc_trace_t *mat_next; #ifdef DEBUG panic("too many entries in allocation list"); #endif atomic_inc_64(&metaslab_trace_over_limit.value.ui64); zal->zal_size--; mat_next = list_next(&zal->zal_list, list_head(&zal->zal_list)); list_remove(&zal->zal_list, mat_next); kmem_cache_free(metaslab_alloc_trace_cache, mat_next); } metaslab_alloc_trace_t *mat = kmem_cache_alloc(metaslab_alloc_trace_cache, KM_SLEEP); list_link_init(&mat->mat_list_node); mat->mat_mg = mg; mat->mat_msp = msp; mat->mat_size = psize; mat->mat_dva_id = dva_id; mat->mat_offset = offset; mat->mat_weight = 0; if (msp != NULL) mat->mat_weight = msp->ms_weight; /* * The list is part of the zio so locking is not required. Only * a single thread will perform allocations for a given zio. */ list_insert_tail(&zal->zal_list, mat); zal->zal_size++; ASSERT3U(zal->zal_size, <=, metaslab_trace_max_entries); } void metaslab_trace_init(zio_alloc_list_t *zal) { list_create(&zal->zal_list, sizeof (metaslab_alloc_trace_t), offsetof(metaslab_alloc_trace_t, mat_list_node)); zal->zal_size = 0; } void metaslab_trace_fini(zio_alloc_list_t *zal) { metaslab_alloc_trace_t *mat; while ((mat = list_remove_head(&zal->zal_list)) != NULL) kmem_cache_free(metaslab_alloc_trace_cache, mat); list_destroy(&zal->zal_list); zal->zal_size = 0; } /* * ========================================================================== * Metaslab block operations * ========================================================================== */ static void metaslab_group_alloc_increment(spa_t *spa, uint64_t vdev, void *tag, int flags) { if (!(flags & METASLAB_ASYNC_ALLOC) || flags & METASLAB_DONT_THROTTLE) return; metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; if (!mg->mg_class->mc_alloc_throttle_enabled) return; (void) refcount_add(&mg->mg_alloc_queue_depth, tag); } void metaslab_group_alloc_decrement(spa_t *spa, uint64_t vdev, void *tag, int flags) { if (!(flags & METASLAB_ASYNC_ALLOC) || flags & METASLAB_DONT_THROTTLE) return; metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; if (!mg->mg_class->mc_alloc_throttle_enabled) return; (void) refcount_remove(&mg->mg_alloc_queue_depth, tag); } void metaslab_group_alloc_verify(spa_t *spa, const blkptr_t *bp, void *tag) { #ifdef ZFS_DEBUG const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); for (int d = 0; d < ndvas; d++) { uint64_t vdev = DVA_GET_VDEV(&dva[d]); metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; VERIFY(refcount_not_held(&mg->mg_alloc_queue_depth, tag)); } #endif } static uint64_t metaslab_block_alloc(metaslab_t *msp, uint64_t size, uint64_t txg) { uint64_t start; range_tree_t *rt = msp->ms_tree; metaslab_class_t *mc = msp->ms_group->mg_class; VERIFY(!msp->ms_condensing); start = mc->mc_ops->msop_alloc(msp, size); if (start != -1ULL) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; VERIFY0(P2PHASE(start, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); VERIFY3U(range_tree_space(rt) - size, <=, msp->ms_size); range_tree_remove(rt, start, size); if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0) vdev_dirty(mg->mg_vd, VDD_METASLAB, msp, txg); range_tree_add(msp->ms_alloctree[txg & TXG_MASK], start, size); /* Track the last successful allocation */ msp->ms_alloc_txg = txg; metaslab_verify_space(msp, txg); } /* * Now that we've attempted the allocation we need to update the * metaslab's maximum block size since it may have changed. */ msp->ms_max_size = metaslab_block_maxsize(msp); return (start); } static uint64_t metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal, uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d) { metaslab_t *msp = NULL; uint64_t offset = -1ULL; uint64_t activation_weight; uint64_t target_distance; int i; activation_weight = METASLAB_WEIGHT_PRIMARY; for (i = 0; i < d; i++) { if (DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) { activation_weight = METASLAB_WEIGHT_SECONDARY; break; } } metaslab_t *search = kmem_alloc(sizeof (*search), KM_SLEEP); search->ms_weight = UINT64_MAX; search->ms_start = 0; for (;;) { boolean_t was_active; avl_tree_t *t = &mg->mg_metaslab_tree; avl_index_t idx; mutex_enter(&mg->mg_lock); /* * Find the metaslab with the highest weight that is less * than what we've already tried. In the common case, this * means that we will examine each metaslab at most once. * Note that concurrent callers could reorder metaslabs * by activation/passivation once we have dropped the mg_lock. * If a metaslab is activated by another thread, and we fail * to allocate from the metaslab we have selected, we may * not try the newly-activated metaslab, and instead activate * another metaslab. This is not optimal, but generally * does not cause any problems (a possible exception being * if every metaslab is completely full except for the * the newly-activated metaslab which we fail to examine). */ msp = avl_find(t, search, &idx); if (msp == NULL) msp = avl_nearest(t, idx, AVL_AFTER); for (; msp != NULL; msp = AVL_NEXT(t, msp)) { if (!metaslab_should_allocate(msp, asize)) { metaslab_trace_add(zal, mg, msp, asize, d, TRACE_TOO_SMALL); continue; } /* * If the selected metaslab is condensing, skip it. */ if (msp->ms_condensing) continue; was_active = msp->ms_weight & METASLAB_ACTIVE_MASK; if (activation_weight == METASLAB_WEIGHT_PRIMARY) break; target_distance = min_distance + (space_map_allocated(msp->ms_sm) != 0 ? 0 : min_distance >> 1); for (i = 0; i < d; i++) { if (metaslab_distance(msp, &dva[i]) < target_distance) break; } if (i == d) break; } mutex_exit(&mg->mg_lock); if (msp == NULL) { kmem_free(search, sizeof (*search)); return (-1ULL); } search->ms_weight = msp->ms_weight; search->ms_start = msp->ms_start + 1; mutex_enter(&msp->ms_lock); /* * Ensure that the metaslab we have selected is still * capable of handling our request. It's possible that * another thread may have changed the weight while we * were blocked on the metaslab lock. We check the * active status first to see if we need to reselect * a new metaslab. */ if (was_active && !(msp->ms_weight & METASLAB_ACTIVE_MASK)) { mutex_exit(&msp->ms_lock); continue; } if ((msp->ms_weight & METASLAB_WEIGHT_SECONDARY) && activation_weight == METASLAB_WEIGHT_PRIMARY) { metaslab_passivate(msp, msp->ms_weight & ~METASLAB_ACTIVE_MASK); mutex_exit(&msp->ms_lock); continue; } if (metaslab_activate(msp, activation_weight) != 0) { mutex_exit(&msp->ms_lock); continue; } msp->ms_selected_txg = txg; /* * Now that we have the lock, recheck to see if we should * continue to use this metaslab for this allocation. The * the metaslab is now loaded so metaslab_should_allocate() can * accurately determine if the allocation attempt should * proceed. */ if (!metaslab_should_allocate(msp, asize)) { /* Passivate this metaslab and select a new one. */ metaslab_trace_add(zal, mg, msp, asize, d, TRACE_TOO_SMALL); goto next; } /* * If this metaslab is currently condensing then pick again as * we can't manipulate this metaslab until it's committed * to disk. */ if (msp->ms_condensing) { metaslab_trace_add(zal, mg, msp, asize, d, TRACE_CONDENSING); mutex_exit(&msp->ms_lock); continue; } offset = metaslab_block_alloc(msp, asize, txg); metaslab_trace_add(zal, mg, msp, asize, d, offset); if (offset != -1ULL) { /* Proactively passivate the metaslab, if needed */ metaslab_segment_may_passivate(msp); break; } next: ASSERT(msp->ms_loaded); /* * We were unable to allocate from this metaslab so determine * a new weight for this metaslab. Now that we have loaded * the metaslab we can provide a better hint to the metaslab * selector. * * For space-based metaslabs, we use the maximum block size. * This information is only available when the metaslab * is loaded and is more accurate than the generic free * space weight that was calculated by metaslab_weight(). * This information allows us to quickly compare the maximum * available allocation in the metaslab to the allocation * size being requested. * * For segment-based metaslabs, determine the new weight * based on the highest bucket in the range tree. We * explicitly use the loaded segment weight (i.e. the range * tree histogram) since it contains the space that is * currently available for allocation and is accurate * even within a sync pass. */ if (WEIGHT_IS_SPACEBASED(msp->ms_weight)) { uint64_t weight = metaslab_block_maxsize(msp); WEIGHT_SET_SPACEBASED(weight); metaslab_passivate(msp, weight); } else { metaslab_passivate(msp, metaslab_weight_from_range_tree(msp)); } /* * We have just failed an allocation attempt, check * that metaslab_should_allocate() agrees. Otherwise, * we may end up in an infinite loop retrying the same * metaslab. */ ASSERT(!metaslab_should_allocate(msp, asize)); mutex_exit(&msp->ms_lock); } mutex_exit(&msp->ms_lock); kmem_free(search, sizeof (*search)); return (offset); } static uint64_t metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal, uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d) { uint64_t offset; ASSERT(mg->mg_initialized); offset = metaslab_group_alloc_normal(mg, zal, asize, txg, min_distance, dva, d); mutex_enter(&mg->mg_lock); if (offset == -1ULL) { mg->mg_failed_allocations++; metaslab_trace_add(zal, mg, NULL, asize, d, TRACE_GROUP_FAILURE); if (asize == SPA_GANGBLOCKSIZE) { /* * This metaslab group was unable to allocate * the minimum gang block size so it must be out of * space. We must notify the allocation throttle * to start skipping allocation attempts to this * metaslab group until more space becomes available. * Note: this failure cannot be caused by the * allocation throttle since the allocation throttle * is only responsible for skipping devices and * not failing block allocations. */ mg->mg_no_free_space = B_TRUE; } } mg->mg_allocations++; mutex_exit(&mg->mg_lock); return (offset); } /* * If we have to write a ditto block (i.e. more than one DVA for a given BP) * on the same vdev as an existing DVA of this BP, then try to allocate it * at least (vdev_asize / (2 ^ ditto_same_vdev_distance_shift)) away from the * existing DVAs. */ int ditto_same_vdev_distance_shift = 3; /* * Allocate a block for the specified i/o. */ static int metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize, dva_t *dva, int d, dva_t *hintdva, uint64_t txg, int flags, zio_alloc_list_t *zal) { metaslab_group_t *mg, *rotor; vdev_t *vd; boolean_t try_hard = B_FALSE; ASSERT(!DVA_IS_VALID(&dva[d])); /* * For testing, make some blocks above a certain size be gang blocks. */ if (psize >= metaslab_gang_bang && (ddi_get_lbolt() & 3) == 0) { metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG); return (SET_ERROR(ENOSPC)); } /* * Start at the rotor and loop through all mgs until we find something. * Note that there's no locking on mc_rotor or mc_aliquot because * nothing actually breaks if we miss a few updates -- we just won't * allocate quite as evenly. It all balances out over time. * * If we are doing ditto or log blocks, try to spread them across * consecutive vdevs. If we're forced to reuse a vdev before we've * allocated all of our ditto blocks, then try and spread them out on * that vdev as much as possible. If it turns out to not be possible, * gradually lower our standards until anything becomes acceptable. * Also, allocating on consecutive vdevs (as opposed to random vdevs) * gives us hope of containing our fault domains to something we're * able to reason about. Otherwise, any two top-level vdev failures * will guarantee the loss of data. With consecutive allocation, * only two adjacent top-level vdev failures will result in data loss. * * If we are doing gang blocks (hintdva is non-NULL), try to keep * ourselves on the same vdev as our gang block header. That * way, we can hope for locality in vdev_cache, plus it makes our * fault domains something tractable. */ if (hintdva) { vd = vdev_lookup_top(spa, DVA_GET_VDEV(&hintdva[d])); /* * It's possible the vdev we're using as the hint no * longer exists (i.e. removed). Consult the rotor when * all else fails. */ if (vd != NULL) { mg = vd->vdev_mg; if (flags & METASLAB_HINTBP_AVOID && mg->mg_next != NULL) mg = mg->mg_next; } else { mg = mc->mc_rotor; } } else if (d != 0) { vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d - 1])); mg = vd->vdev_mg->mg_next; } else { mg = mc->mc_rotor; } /* * If the hint put us into the wrong metaslab class, or into a * metaslab group that has been passivated, just follow the rotor. */ if (mg->mg_class != mc || mg->mg_activation_count <= 0) mg = mc->mc_rotor; rotor = mg; top: do { boolean_t allocatable; ASSERT(mg->mg_activation_count == 1); vd = mg->mg_vd; /* * Don't allocate from faulted devices. */ if (try_hard) { spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER); allocatable = vdev_allocatable(vd); spa_config_exit(spa, SCL_ZIO, FTAG); } else { allocatable = vdev_allocatable(vd); } /* * Determine if the selected metaslab group is eligible * for allocations. If we're ganging then don't allow * this metaslab group to skip allocations since that would * inadvertently return ENOSPC and suspend the pool * even though space is still available. */ if (allocatable && !GANG_ALLOCATION(flags) && !try_hard) { allocatable = metaslab_group_allocatable(mg, rotor, psize); } if (!allocatable) { metaslab_trace_add(zal, mg, NULL, psize, d, TRACE_NOT_ALLOCATABLE); goto next; } ASSERT(mg->mg_initialized); /* * Avoid writing single-copy data to a failing, * non-redundant vdev, unless we've already tried all * other vdevs. */ if ((vd->vdev_stat.vs_write_errors > 0 || vd->vdev_state < VDEV_STATE_HEALTHY) && d == 0 && !try_hard && vd->vdev_children == 0) { metaslab_trace_add(zal, mg, NULL, psize, d, TRACE_VDEV_ERROR); goto next; } ASSERT(mg->mg_class == mc); /* * If we don't need to try hard, then require that the * block be 1/8th of the device away from any other DVAs * in this BP. If we are trying hard, allow any offset * to be used (distance=0). */ uint64_t distance = 0; if (!try_hard) { distance = vd->vdev_asize >> ditto_same_vdev_distance_shift; if (distance <= (1ULL << vd->vdev_ms_shift)) distance = 0; } uint64_t asize = vdev_psize_to_asize(vd, psize); ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0); uint64_t offset = metaslab_group_alloc(mg, zal, asize, txg, distance, dva, d); if (offset != -1ULL) { /* * If we've just selected this metaslab group, * figure out whether the corresponding vdev is * over- or under-used relative to the pool, * and set an allocation bias to even it out. */ if (mc->mc_aliquot == 0 && metaslab_bias_enabled) { vdev_stat_t *vs = &vd->vdev_stat; int64_t vu, cu; vu = (vs->vs_alloc * 100) / (vs->vs_space + 1); cu = (mc->mc_alloc * 100) / (mc->mc_space + 1); /* * Calculate how much more or less we should * try to allocate from this device during * this iteration around the rotor. * For example, if a device is 80% full * and the pool is 20% full then we should * reduce allocations by 60% on this device. * * mg_bias = (20 - 80) * 512K / 100 = -307K * * This reduces allocations by 307K for this * iteration. */ mg->mg_bias = ((cu - vu) * (int64_t)mg->mg_aliquot) / 100; } else if (!metaslab_bias_enabled) { mg->mg_bias = 0; } if (atomic_add_64_nv(&mc->mc_aliquot, asize) >= mg->mg_aliquot + mg->mg_bias) { mc->mc_rotor = mg->mg_next; mc->mc_aliquot = 0; } DVA_SET_VDEV(&dva[d], vd->vdev_id); DVA_SET_OFFSET(&dva[d], offset); DVA_SET_GANG(&dva[d], !!(flags & METASLAB_GANG_HEADER)); DVA_SET_ASIZE(&dva[d], asize); return (0); } next: mc->mc_rotor = mg->mg_next; mc->mc_aliquot = 0; } while ((mg = mg->mg_next) != rotor); /* * If we haven't tried hard, do so now. */ if (!try_hard) { try_hard = B_TRUE; goto top; } bzero(&dva[d], sizeof (dva_t)); metaslab_trace_add(zal, rotor, NULL, psize, d, TRACE_ENOSPC); return (SET_ERROR(ENOSPC)); } /* * Free the block represented by DVA in the context of the specified * transaction group. */ static void metaslab_free_dva(spa_t *spa, const dva_t *dva, uint64_t txg, boolean_t now) { uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd; metaslab_t *msp; ASSERT(DVA_IS_VALID(dva)); if (txg > spa_freeze_txg(spa)) return; if ((vd = vdev_lookup_top(spa, vdev)) == NULL || (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) { cmn_err(CE_WARN, "metaslab_free_dva(): bad DVA %llu:%llu", (u_longlong_t)vdev, (u_longlong_t)offset); ASSERT(0); return; } msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; if (DVA_GET_GANG(dva)) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); mutex_enter(&msp->ms_lock); if (now) { range_tree_remove(msp->ms_alloctree[txg & TXG_MASK], offset, size); VERIFY(!msp->ms_condensing); VERIFY3U(offset, >=, msp->ms_start); VERIFY3U(offset + size, <=, msp->ms_start + msp->ms_size); VERIFY3U(range_tree_space(msp->ms_tree) + size, <=, msp->ms_size); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); range_tree_add(msp->ms_tree, offset, size); msp->ms_max_size = metaslab_block_maxsize(msp); } else { VERIFY3U(txg, ==, spa->spa_syncing_txg); if (range_tree_space(msp->ms_freeingtree) == 0) vdev_dirty(vd, VDD_METASLAB, msp, txg); range_tree_add(msp->ms_freeingtree, offset, size); } mutex_exit(&msp->ms_lock); } /* * Intent log support: upon opening the pool after a crash, notify the SPA * of blocks that the intent log has allocated for immediate write, but * which are still considered free by the SPA because the last transaction * group didn't commit yet. */ static int metaslab_claim_dva(spa_t *spa, const dva_t *dva, uint64_t txg) { uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd; metaslab_t *msp; int error = 0; ASSERT(DVA_IS_VALID(dva)); if ((vd = vdev_lookup_top(spa, vdev)) == NULL || (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) return (SET_ERROR(ENXIO)); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; if (DVA_GET_GANG(dva)) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); mutex_enter(&msp->ms_lock); if ((txg != 0 && spa_writeable(spa)) || !msp->ms_loaded) error = metaslab_activate(msp, METASLAB_WEIGHT_SECONDARY); if (error == 0 && !range_tree_contains(msp->ms_tree, offset, size)) error = SET_ERROR(ENOENT); if (error || txg == 0) { /* txg == 0 indicates dry run */ mutex_exit(&msp->ms_lock); return (error); } VERIFY(!msp->ms_condensing); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); VERIFY3U(range_tree_space(msp->ms_tree) - size, <=, msp->ms_size); range_tree_remove(msp->ms_tree, offset, size); if (spa_writeable(spa)) { /* don't dirty if we're zdb(1M) */ if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0) vdev_dirty(vd, VDD_METASLAB, msp, txg); range_tree_add(msp->ms_alloctree[txg & TXG_MASK], offset, size); } mutex_exit(&msp->ms_lock); return (0); } /* * Reserve some allocation slots. The reservation system must be called * before we call into the allocator. If there aren't any available slots * then the I/O will be throttled until an I/O completes and its slots are * freed up. The function returns true if it was successful in placing * the reservation. */ boolean_t metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, zio_t *zio, int flags) { uint64_t available_slots = 0; boolean_t slot_reserved = B_FALSE; ASSERT(mc->mc_alloc_throttle_enabled); mutex_enter(&mc->mc_lock); uint64_t reserved_slots = refcount_count(&mc->mc_alloc_slots); if (reserved_slots < mc->mc_alloc_max_slots) available_slots = mc->mc_alloc_max_slots - reserved_slots; if (slots <= available_slots || GANG_ALLOCATION(flags)) { /* * We reserve the slots individually so that we can unreserve * them individually when an I/O completes. */ for (int d = 0; d < slots; d++) { reserved_slots = refcount_add(&mc->mc_alloc_slots, zio); } zio->io_flags |= ZIO_FLAG_IO_ALLOCATING; slot_reserved = B_TRUE; } mutex_exit(&mc->mc_lock); return (slot_reserved); } void metaslab_class_throttle_unreserve(metaslab_class_t *mc, int slots, zio_t *zio) { ASSERT(mc->mc_alloc_throttle_enabled); mutex_enter(&mc->mc_lock); for (int d = 0; d < slots; d++) { (void) refcount_remove(&mc->mc_alloc_slots, zio); } mutex_exit(&mc->mc_lock); } int metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, blkptr_t *bp, int ndvas, uint64_t txg, blkptr_t *hintbp, int flags, zio_alloc_list_t *zal, zio_t *zio) { dva_t *dva = bp->blk_dva; dva_t *hintdva = hintbp->blk_dva; int error = 0; ASSERT(bp->blk_birth == 0); ASSERT(BP_PHYSICAL_BIRTH(bp) == 0); spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); if (mc->mc_rotor == NULL) { /* no vdevs in this class */ spa_config_exit(spa, SCL_ALLOC, FTAG); return (SET_ERROR(ENOSPC)); } ASSERT(ndvas > 0 && ndvas <= spa_max_replication(spa)); ASSERT(BP_GET_NDVAS(bp) == 0); ASSERT(hintbp == NULL || ndvas <= BP_GET_NDVAS(hintbp)); ASSERT3P(zal, !=, NULL); for (int d = 0; d < ndvas; d++) { error = metaslab_alloc_dva(spa, mc, psize, dva, d, hintdva, txg, flags, zal); if (error != 0) { for (d--; d >= 0; d--) { metaslab_free_dva(spa, &dva[d], txg, B_TRUE); metaslab_group_alloc_decrement(spa, DVA_GET_VDEV(&dva[d]), zio, flags); bzero(&dva[d], sizeof (dva_t)); } spa_config_exit(spa, SCL_ALLOC, FTAG); return (error); } else { /* * Update the metaslab group's queue depth * based on the newly allocated dva. */ metaslab_group_alloc_increment(spa, DVA_GET_VDEV(&dva[d]), zio, flags); } } ASSERT(error == 0); ASSERT(BP_GET_NDVAS(bp) == ndvas); spa_config_exit(spa, SCL_ALLOC, FTAG); BP_SET_BIRTH(bp, txg, txg); return (0); } void metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); ASSERT(!BP_IS_HOLE(bp)); ASSERT(!now || bp->blk_birth >= spa_syncing_txg(spa)); spa_config_enter(spa, SCL_FREE, FTAG, RW_READER); for (int d = 0; d < ndvas; d++) metaslab_free_dva(spa, &dva[d], txg, now); spa_config_exit(spa, SCL_FREE, FTAG); } int metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); int error = 0; ASSERT(!BP_IS_HOLE(bp)); if (txg != 0) { /* * First do a dry run to make sure all DVAs are claimable, * so we don't have to unwind from partial failures below. */ if ((error = metaslab_claim(spa, bp, 0)) != 0) return (error); } spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); for (int d = 0; d < ndvas; d++) if ((error = metaslab_claim_dva(spa, &dva[d], txg)) != 0) break; spa_config_exit(spa, SCL_ALLOC, FTAG); ASSERT(error == 0 || txg == 0); return (error); } void metaslab_check_free(spa_t *spa, const blkptr_t *bp) { if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0) return; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (int i = 0; i < BP_GET_NDVAS(bp); i++) { uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[i]); vdev_t *vd = vdev_lookup_top(spa, vdev); uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]); uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]); metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; if (msp->ms_loaded) range_tree_verify(msp->ms_tree, offset, size); range_tree_verify(msp->ms_freeingtree, offset, size); range_tree_verify(msp->ms_freedtree, offset, size); for (int j = 0; j < TXG_DEFER_SIZE; j++) range_tree_verify(msp->ms_defertree[j], offset, size); } spa_config_exit(spa, SCL_VDEV, FTAG); } Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c (revision 321553) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c (revision 321554) @@ -1,2186 +1,2196 @@ /* * 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 (c) 2011, 2017 by Delphix. All rights reserved. * Copyright 2015 Nexenta Systems, Inc. All rights reserved. * Copyright 2013 Martin Matuska . All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright 2013 Saso Kiselkov. 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 #include #include #include #include #include #include #include "zfs_prop.h" #include #if defined(__FreeBSD__) && defined(_KERNEL) #include #include #endif /* * SPA locking * * There are four basic locks for managing spa_t structures: * * spa_namespace_lock (global mutex) * * This lock must be acquired to do any of the following: * * - Lookup a spa_t by name * - Add or remove a spa_t from the namespace * - Increase spa_refcount from non-zero * - Check if spa_refcount is zero * - Rename a spa_t * - add/remove/attach/detach devices * - Held for the duration of create/destroy/import/export * * It does not need to handle recursion. A create or destroy may * reference objects (files or zvols) in other pools, but by * definition they must have an existing reference, and will never need * to lookup a spa_t by name. * * spa_refcount (per-spa refcount_t protected by mutex) * * This reference count keep track of any active users of the spa_t. The * spa_t cannot be destroyed or freed while this is non-zero. Internally, * the refcount is never really 'zero' - opening a pool implicitly keeps * some references in the DMU. Internally we check against spa_minref, but * present the image of a zero/non-zero value to consumers. * * spa_config_lock[] (per-spa array of rwlocks) * * This protects the spa_t from config changes, and must be held in * the following circumstances: * * - RW_READER to perform I/O to the spa * - RW_WRITER to change the vdev config * * The locking order is fairly straightforward: * * spa_namespace_lock -> spa_refcount * * The namespace lock must be acquired to increase the refcount from 0 * or to check if it is zero. * * spa_refcount -> spa_config_lock[] * * There must be at least one valid reference on the spa_t to acquire * the config lock. * * spa_namespace_lock -> spa_config_lock[] * * The namespace lock must always be taken before the config lock. * * * The spa_namespace_lock can be acquired directly and is globally visible. * * The namespace is manipulated using the following functions, all of which * require the spa_namespace_lock to be held. * * spa_lookup() Lookup a spa_t by name. * * spa_add() Create a new spa_t in the namespace. * * spa_remove() Remove a spa_t from the namespace. This also * frees up any memory associated with the spa_t. * * spa_next() Returns the next spa_t in the system, or the * first if NULL is passed. * * spa_evict_all() Shutdown and remove all spa_t structures in * the system. * * spa_guid_exists() Determine whether a pool/device guid exists. * * The spa_refcount is manipulated using the following functions: * * spa_open_ref() Adds a reference to the given spa_t. Must be * called with spa_namespace_lock held if the * refcount is currently zero. * * spa_close() Remove a reference from the spa_t. This will * not free the spa_t or remove it from the * namespace. No locking is required. * * spa_refcount_zero() Returns true if the refcount is currently * zero. Must be called with spa_namespace_lock * held. * * The spa_config_lock[] is an array of rwlocks, ordered as follows: * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). * * To read the configuration, it suffices to hold one of these locks as reader. * To modify the configuration, you must hold all locks as writer. To modify * vdev state without altering the vdev tree's topology (e.g. online/offline), * you must hold SCL_STATE and SCL_ZIO as writer. * * We use these distinct config locks to avoid recursive lock entry. * For example, spa_sync() (which holds SCL_CONFIG as reader) induces * block allocations (SCL_ALLOC), which may require reading space maps * from disk (dmu_read() -> zio_read() -> SCL_ZIO). * * The spa config locks cannot be normal rwlocks because we need the * ability to hand off ownership. For example, SCL_ZIO is acquired * by the issuing thread and later released by an interrupt thread. * They do, however, obey the usual write-wanted semantics to prevent * writer (i.e. system administrator) starvation. * * The lock acquisition rules are as follows: * * SCL_CONFIG * Protects changes to the vdev tree topology, such as vdev * add/remove/attach/detach. Protects the dirty config list * (spa_config_dirty_list) and the set of spares and l2arc devices. * * SCL_STATE * Protects changes to pool state and vdev state, such as vdev * online/offline/fault/degrade/clear. Protects the dirty state list * (spa_state_dirty_list) and global pool state (spa_state). * * SCL_ALLOC * Protects changes to metaslab groups and classes. * Held as reader by metaslab_alloc() and metaslab_claim(). * * SCL_ZIO * Held by bp-level zios (those which have no io_vd upon entry) * to prevent changes to the vdev tree. The bp-level zio implicitly * protects all of its vdev child zios, which do not hold SCL_ZIO. * * SCL_FREE * Protects changes to metaslab groups and classes. * Held as reader by metaslab_free(). SCL_FREE is distinct from * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free * blocks in zio_done() while another i/o that holds either * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. * * SCL_VDEV * Held as reader to prevent changes to the vdev tree during trivial * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the * other locks, and lower than all of them, to ensure that it's safe * to acquire regardless of caller context. * * In addition, the following rules apply: * * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. * The lock ordering is SCL_CONFIG > spa_props_lock. * * (b) I/O operations on leaf vdevs. For any zio operation that takes * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), * or zio_write_phys() -- the caller must ensure that the config cannot * cannot change in the interim, and that the vdev cannot be reopened. * SCL_STATE as reader suffices for both. * * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). * * spa_vdev_enter() Acquire the namespace lock and the config lock * for writing. * * spa_vdev_exit() Release the config lock, wait for all I/O * to complete, sync the updated configs to the * cache, and release the namespace lock. * * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual * locking is, always, based on spa_namespace_lock and spa_config_lock[]. * * spa_rename() is also implemented within this file since it requires * manipulation of the namespace. */ static avl_tree_t spa_namespace_avl; kmutex_t spa_namespace_lock; static kcondvar_t spa_namespace_cv; static int spa_active_count; int spa_max_replication_override = SPA_DVAS_PER_BP; static kmutex_t spa_spare_lock; static avl_tree_t spa_spare_avl; static kmutex_t spa_l2cache_lock; static avl_tree_t spa_l2cache_avl; kmem_cache_t *spa_buffer_pool; int spa_mode_global; #ifdef ZFS_DEBUG /* Everything except dprintf and spa is on by default in debug builds */ int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA); #else int zfs_flags = 0; #endif /* * zfs_recover can be set to nonzero to attempt to recover from * otherwise-fatal errors, typically caused by on-disk corruption. When * set, calls to zfs_panic_recover() will turn into warning messages. * This should only be used as a last resort, as it typically results * in leaked space, or worse. */ boolean_t zfs_recover = B_FALSE; /* * If destroy encounters an EIO while reading metadata (e.g. indirect * blocks), space referenced by the missing metadata can not be freed. * Normally this causes the background destroy to become "stalled", as * it is unable to make forward progress. While in this stalled state, * all remaining space to free from the error-encountering filesystem is * "temporarily leaked". Set this flag to cause it to ignore the EIO, * permanently leak the space from indirect blocks that can not be read, * and continue to free everything else that it can. * * The default, "stalling" behavior is useful if the storage partially * fails (i.e. some but not all i/os fail), and then later recovers. In * this case, we will be able to continue pool operations while it is * partially failed, and when it recovers, we can continue to free the * space, with no leaks. However, note that this case is actually * fairly rare. * * Typically pools either (a) fail completely (but perhaps temporarily, * e.g. a top-level vdev going offline), or (b) have localized, * permanent errors (e.g. disk returns the wrong data due to bit flip or * firmware bug). In case (a), this setting does not matter because the * pool will be suspended and the sync thread will not be able to make * forward progress regardless. In case (b), because the error is * permanent, the best we can do is leak the minimum amount of space, * which is what setting this flag will do. Therefore, it is reasonable * for this flag to normally be set, but we chose the more conservative * approach of not setting it, so that there is no possibility of * leaking space in the "partial temporary" failure case. */ boolean_t zfs_free_leak_on_eio = B_FALSE; /* * Expiration time in milliseconds. This value has two meanings. First it is * used to determine when the spa_deadman() logic should fire. By default the * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. * Secondly, the value determines if an I/O is considered "hung". Any I/O that * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting * in a system panic. */ uint64_t zfs_deadman_synctime_ms = 1000000ULL; /* * Check time in milliseconds. This defines the frequency at which we check * for hung I/O. */ uint64_t zfs_deadman_checktime_ms = 5000ULL; /* * Default value of -1 for zfs_deadman_enabled is resolved in * zfs_deadman_init() */ int zfs_deadman_enabled = -1; /* * The worst case is single-sector max-parity RAID-Z blocks, in which * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) * times the size; so just assume that. Add to this the fact that * we can have up to 3 DVAs per bp, and one more factor of 2 because * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, * the worst case is: * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 */ int spa_asize_inflation = 24; #if defined(__FreeBSD__) && defined(_KERNEL) SYSCTL_DECL(_vfs_zfs); SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0, "Try to recover from otherwise-fatal errors."); static int sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS) { int err, val; val = zfs_flags; err = sysctl_handle_int(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); /* * ZFS_DEBUG_MODIFY must be enabled prior to boot so all * arc buffers in the system have the necessary additional * checksum data. However, it is safe to disable at any * time. */ if (!(zfs_flags & ZFS_DEBUG_MODIFY)) val &= ~ZFS_DEBUG_MODIFY; zfs_flags = val; return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, debugflags, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int), sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing."); SYSCTL_PROC(_vfs_zfs, OID_AUTO, debug_flags, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int), sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing (deprecated, see vfs.zfs.debugflags)."); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, &zfs_deadman_synctime_ms, 0, "Stalled ZFS I/O expiration time in milliseconds"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, &zfs_deadman_checktime_ms, 0, "Period of checks for stalled ZFS I/O in milliseconds"); SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN, &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes"); #endif #ifndef illumos #ifdef _KERNEL static void zfs_deadman_init() { /* * If we are not i386 or amd64 or in a virtual machine, * disable ZFS deadman thread by default */ if (zfs_deadman_enabled == -1) { #if defined(__amd64__) || defined(__i386__) zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; #else zfs_deadman_enabled = 0; #endif } } #endif /* _KERNEL */ #endif /* !illumos */ /* * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in * the pool to be consumed. This ensures that we don't run the pool * completely out of space, due to unaccounted changes (e.g. to the MOS). * It also limits the worst-case time to allocate space. If we have * less than this amount of free space, most ZPL operations (e.g. write, * create) will return ENOSPC. * * Certain operations (e.g. file removal, most administrative actions) can * use half the slop space. They will only return ENOSPC if less than half * the slop space is free. Typically, once the pool has less than the slop * space free, the user will use these operations to free up space in the pool. * These are the operations that call dsl_pool_adjustedsize() with the netfree * argument set to TRUE. * * A very restricted set of operations are always permitted, regardless of * the amount of free space. These are the operations that call * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these * operations result in a net increase in the amount of space used, * it is possible to run the pool completely out of space, causing it to * be permanently read-only. * * Note that on very small pools, the slop space will be larger than * 3.2%, in an effort to have it be at least spa_min_slop (128MB), * but we never allow it to be more than half the pool size. * * See also the comments in zfs_space_check_t. */ int spa_slop_shift = 5; SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN, &spa_slop_shift, 0, "Shift value of reserved space (1/(2^spa_slop_shift))."); uint64_t spa_min_slop = 128 * 1024 * 1024; SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, spa_min_slop, CTLFLAG_RWTUN, &spa_min_slop, 0, "Minimal value of reserved space"); /* * ========================================================================== * SPA config locking * ========================================================================== */ static void spa_config_lock_init(spa_t *spa) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); refcount_create_untracked(&scl->scl_count); scl->scl_writer = NULL; scl->scl_write_wanted = 0; } } static void spa_config_lock_destroy(spa_t *spa) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; mutex_destroy(&scl->scl_lock); cv_destroy(&scl->scl_cv); refcount_destroy(&scl->scl_count); ASSERT(scl->scl_writer == NULL); ASSERT(scl->scl_write_wanted == 0); } } int spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); if (rw == RW_READER) { if (scl->scl_writer || scl->scl_write_wanted) { mutex_exit(&scl->scl_lock); spa_config_exit(spa, locks & ((1 << i) - 1), tag); return (0); } } else { ASSERT(scl->scl_writer != curthread); if (!refcount_is_zero(&scl->scl_count)) { mutex_exit(&scl->scl_lock); spa_config_exit(spa, locks & ((1 << i) - 1), tag); return (0); } scl->scl_writer = curthread; } (void) refcount_add(&scl->scl_count, tag); mutex_exit(&scl->scl_lock); } return (1); } void spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) { int wlocks_held = 0; ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (scl->scl_writer == curthread) wlocks_held |= (1 << i); if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); if (rw == RW_READER) { while (scl->scl_writer || scl->scl_write_wanted) { cv_wait(&scl->scl_cv, &scl->scl_lock); } } else { ASSERT(scl->scl_writer != curthread); while (!refcount_is_zero(&scl->scl_count)) { scl->scl_write_wanted++; cv_wait(&scl->scl_cv, &scl->scl_lock); scl->scl_write_wanted--; } scl->scl_writer = curthread; } (void) refcount_add(&scl->scl_count, tag); mutex_exit(&scl->scl_lock); } ASSERT(wlocks_held <= locks); } void spa_config_exit(spa_t *spa, int locks, void *tag) { for (int i = SCL_LOCKS - 1; i >= 0; i--) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); ASSERT(!refcount_is_zero(&scl->scl_count)); if (refcount_remove(&scl->scl_count, tag) == 0) { ASSERT(scl->scl_writer == NULL || scl->scl_writer == curthread); scl->scl_writer = NULL; /* OK in either case */ cv_broadcast(&scl->scl_cv); } mutex_exit(&scl->scl_lock); } } int spa_config_held(spa_t *spa, int locks, krw_t rw) { int locks_held = 0; for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || (rw == RW_WRITER && scl->scl_writer == curthread)) locks_held |= 1 << i; } return (locks_held); } /* * ========================================================================== * SPA namespace functions * ========================================================================== */ /* * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. * Returns NULL if no matching spa_t is found. */ spa_t * spa_lookup(const char *name) { static spa_t search; /* spa_t is large; don't allocate on stack */ spa_t *spa; avl_index_t where; char *cp; ASSERT(MUTEX_HELD(&spa_namespace_lock)); (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); /* * If it's a full dataset name, figure out the pool name and * just use that. */ cp = strpbrk(search.spa_name, "/@#"); if (cp != NULL) *cp = '\0'; spa = avl_find(&spa_namespace_avl, &search, &where); return (spa); } /* * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. * If the zfs_deadman_enabled flag is set then it inspects all vdev queues * looking for potentially hung I/Os. */ static void spa_deadman(void *arg, int pending) { spa_t *spa = arg; /* * Disable the deadman timer if the pool is suspended. */ if (spa_suspended(spa)) { #ifdef illumos VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); #else /* Nothing. just don't schedule any future callouts. */ #endif return; } zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", (gethrtime() - spa->spa_sync_starttime) / NANOSEC, ++spa->spa_deadman_calls); if (zfs_deadman_enabled) vdev_deadman(spa->spa_root_vdev); #ifdef __FreeBSD__ #ifdef _KERNEL callout_schedule(&spa->spa_deadman_cycid, hz * zfs_deadman_checktime_ms / MILLISEC); #endif #endif } #if defined(__FreeBSD__) && defined(_KERNEL) static void spa_deadman_timeout(void *arg) { spa_t *spa = arg; taskqueue_enqueue(taskqueue_thread, &spa->spa_deadman_task); } #endif /* * Create an uninitialized spa_t with the given name. Requires * spa_namespace_lock. The caller must ensure that the spa_t doesn't already * exist by calling spa_lookup() first. */ spa_t * spa_add(const char *name, nvlist_t *config, const char *altroot) { spa_t *spa; spa_config_dirent_t *dp; #ifdef illumos cyc_handler_t hdlr; cyc_time_t when; #endif ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); for (int t = 0; t < TXG_SIZE; t++) bplist_create(&spa->spa_free_bplist[t]); (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); spa->spa_state = POOL_STATE_UNINITIALIZED; spa->spa_freeze_txg = UINT64_MAX; spa->spa_final_txg = UINT64_MAX; spa->spa_load_max_txg = UINT64_MAX; spa->spa_proc = &p0; spa->spa_proc_state = SPA_PROC_NONE; #ifdef illumos hdlr.cyh_func = spa_deadman; hdlr.cyh_arg = spa; hdlr.cyh_level = CY_LOW_LEVEL; #endif spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); #ifdef illumos /* * This determines how often we need to check for hung I/Os after * the cyclic has already fired. Since checking for hung I/Os is * an expensive operation we don't want to check too frequently. * Instead wait for 5 seconds before checking again. */ when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); when.cyt_when = CY_INFINITY; mutex_enter(&cpu_lock); spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); mutex_exit(&cpu_lock); #else /* !illumos */ #ifdef _KERNEL /* * callout(9) does not provide a way to initialize a callout with * a function and an argument, so we use callout_reset() to schedule * the callout in the very distant future. Even if that event ever * fires, it should be okayas we won't have any active zio-s. * But normally spa_sync() will reschedule the callout with a proper * timeout. * callout(9) does not allow the callback function to sleep but * vdev_deadman() needs to acquire vq_lock and illumos mutexes are * emulated using sx(9). For this reason spa_deadman_timeout() * will schedule spa_deadman() as task on a taskqueue that allows * sleeping. */ TASK_INIT(&spa->spa_deadman_task, 0, spa_deadman, spa); callout_init(&spa->spa_deadman_cycid, 1); callout_reset_sbt(&spa->spa_deadman_cycid, SBT_MAX, 0, spa_deadman_timeout, spa, 0); #endif #endif refcount_create(&spa->spa_refcount); spa_config_lock_init(spa); avl_add(&spa_namespace_avl, spa); /* * Set the alternate root, if there is one. */ if (altroot) { spa->spa_root = spa_strdup(altroot); spa_active_count++; } avl_create(&spa->spa_alloc_tree, zio_bookmark_compare, sizeof (zio_t), offsetof(zio_t, io_alloc_node)); /* * Every pool starts with the default cachefile */ list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), offsetof(spa_config_dirent_t, scd_link)); dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); list_insert_head(&spa->spa_config_list, dp); VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, KM_SLEEP) == 0); if (config != NULL) { nvlist_t *features; if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, &features) == 0) { VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); } VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); } if (spa->spa_label_features == NULL) { VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, KM_SLEEP) == 0); } spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); spa->spa_min_ashift = INT_MAX; spa->spa_max_ashift = 0; /* * As a pool is being created, treat all features as disabled by * setting SPA_FEATURE_DISABLED for all entries in the feature * refcount cache. */ for (int i = 0; i < SPA_FEATURES; i++) { spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; } return (spa); } /* * Removes a spa_t from the namespace, freeing up any memory used. Requires * spa_namespace_lock. This is called only after the spa_t has been closed and * deactivated. */ void spa_remove(spa_t *spa) { spa_config_dirent_t *dp; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0); nvlist_free(spa->spa_config_splitting); avl_remove(&spa_namespace_avl, spa); cv_broadcast(&spa_namespace_cv); if (spa->spa_root) { spa_strfree(spa->spa_root); spa_active_count--; } while ((dp = list_head(&spa->spa_config_list)) != NULL) { list_remove(&spa->spa_config_list, dp); if (dp->scd_path != NULL) spa_strfree(dp->scd_path); kmem_free(dp, sizeof (spa_config_dirent_t)); } avl_destroy(&spa->spa_alloc_tree); list_destroy(&spa->spa_config_list); nvlist_free(spa->spa_label_features); nvlist_free(spa->spa_load_info); spa_config_set(spa, NULL); #ifdef illumos mutex_enter(&cpu_lock); if (spa->spa_deadman_cycid != CYCLIC_NONE) cyclic_remove(spa->spa_deadman_cycid); mutex_exit(&cpu_lock); spa->spa_deadman_cycid = CYCLIC_NONE; #else /* !illumos */ #ifdef _KERNEL callout_drain(&spa->spa_deadman_cycid); taskqueue_drain(taskqueue_thread, &spa->spa_deadman_task); #endif #endif refcount_destroy(&spa->spa_refcount); spa_config_lock_destroy(spa); for (int t = 0; t < TXG_SIZE; t++) bplist_destroy(&spa->spa_free_bplist[t]); zio_checksum_templates_free(spa); cv_destroy(&spa->spa_async_cv); cv_destroy(&spa->spa_evicting_os_cv); cv_destroy(&spa->spa_proc_cv); cv_destroy(&spa->spa_scrub_io_cv); cv_destroy(&spa->spa_suspend_cv); mutex_destroy(&spa->spa_alloc_lock); mutex_destroy(&spa->spa_async_lock); mutex_destroy(&spa->spa_errlist_lock); mutex_destroy(&spa->spa_errlog_lock); mutex_destroy(&spa->spa_evicting_os_lock); mutex_destroy(&spa->spa_history_lock); mutex_destroy(&spa->spa_proc_lock); mutex_destroy(&spa->spa_props_lock); mutex_destroy(&spa->spa_cksum_tmpls_lock); mutex_destroy(&spa->spa_scrub_lock); mutex_destroy(&spa->spa_suspend_lock); mutex_destroy(&spa->spa_vdev_top_lock); kmem_free(spa, sizeof (spa_t)); } /* * Given a pool, return the next pool in the namespace, or NULL if there is * none. If 'prev' is NULL, return the first pool. */ spa_t * spa_next(spa_t *prev) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (prev) return (AVL_NEXT(&spa_namespace_avl, prev)); else return (avl_first(&spa_namespace_avl)); } /* * ========================================================================== * SPA refcount functions * ========================================================================== */ /* * Add a reference to the given spa_t. Must have at least one reference, or * have the namespace lock held. */ void spa_open_ref(spa_t *spa, void *tag) { ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || MUTEX_HELD(&spa_namespace_lock)); (void) refcount_add(&spa->spa_refcount, tag); } /* * Remove a reference to the given spa_t. Must have at least one reference, or * have the namespace lock held. */ void spa_close(spa_t *spa, void *tag) { ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || MUTEX_HELD(&spa_namespace_lock)); (void) refcount_remove(&spa->spa_refcount, tag); } /* * Remove a reference to the given spa_t held by a dsl dir that is * being asynchronously released. Async releases occur from a taskq * performing eviction of dsl datasets and dirs. The namespace lock * isn't held and the hold by the object being evicted may contribute to * spa_minref (e.g. dataset or directory released during pool export), * so the asserts in spa_close() do not apply. */ void spa_async_close(spa_t *spa, void *tag) { (void) refcount_remove(&spa->spa_refcount, tag); } /* * Check to see if the spa refcount is zero. Must be called with * spa_namespace_lock held. We really compare against spa_minref, which is the * number of references acquired when opening a pool */ boolean_t spa_refcount_zero(spa_t *spa) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); return (refcount_count(&spa->spa_refcount) == spa->spa_minref); } /* * ========================================================================== * SPA spare and l2cache tracking * ========================================================================== */ /* * Hot spares and cache devices are tracked using the same code below, * for 'auxiliary' devices. */ typedef struct spa_aux { uint64_t aux_guid; uint64_t aux_pool; avl_node_t aux_avl; int aux_count; } spa_aux_t; static int spa_aux_compare(const void *a, const void *b) { const spa_aux_t *sa = a; const spa_aux_t *sb = b; if (sa->aux_guid < sb->aux_guid) return (-1); else if (sa->aux_guid > sb->aux_guid) return (1); else return (0); } void spa_aux_add(vdev_t *vd, avl_tree_t *avl) { avl_index_t where; spa_aux_t search; spa_aux_t *aux; search.aux_guid = vd->vdev_guid; if ((aux = avl_find(avl, &search, &where)) != NULL) { aux->aux_count++; } else { aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); aux->aux_guid = vd->vdev_guid; aux->aux_count = 1; avl_insert(avl, aux, where); } } void spa_aux_remove(vdev_t *vd, avl_tree_t *avl) { spa_aux_t search; spa_aux_t *aux; avl_index_t where; search.aux_guid = vd->vdev_guid; aux = avl_find(avl, &search, &where); ASSERT(aux != NULL); if (--aux->aux_count == 0) { avl_remove(avl, aux); kmem_free(aux, sizeof (spa_aux_t)); } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { aux->aux_pool = 0ULL; } } boolean_t spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) { spa_aux_t search, *found; search.aux_guid = guid; found = avl_find(avl, &search, NULL); if (pool) { if (found) *pool = found->aux_pool; else *pool = 0ULL; } if (refcnt) { if (found) *refcnt = found->aux_count; else *refcnt = 0; } return (found != NULL); } void spa_aux_activate(vdev_t *vd, avl_tree_t *avl) { spa_aux_t search, *found; avl_index_t where; search.aux_guid = vd->vdev_guid; found = avl_find(avl, &search, &where); ASSERT(found != NULL); ASSERT(found->aux_pool == 0ULL); found->aux_pool = spa_guid(vd->vdev_spa); } /* * Spares are tracked globally due to the following constraints: * * - A spare may be part of multiple pools. * - A spare may be added to a pool even if it's actively in use within * another pool. * - A spare in use in any pool can only be the source of a replacement if * the target is a spare in the same pool. * * We keep track of all spares on the system through the use of a reference * counted AVL tree. When a vdev is added as a spare, or used as a replacement * spare, then we bump the reference count in the AVL tree. In addition, we set * the 'vdev_isspare' member to indicate that the device is a spare (active or * inactive). When a spare is made active (used to replace a device in the * pool), we also keep track of which pool its been made a part of. * * The 'spa_spare_lock' protects the AVL tree. These functions are normally * called under the spa_namespace lock as part of vdev reconfiguration. The * separate spare lock exists for the status query path, which does not need to * be completely consistent with respect to other vdev configuration changes. */ static int spa_spare_compare(const void *a, const void *b) { return (spa_aux_compare(a, b)); } void spa_spare_add(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(!vd->vdev_isspare); spa_aux_add(vd, &spa_spare_avl); vd->vdev_isspare = B_TRUE; mutex_exit(&spa_spare_lock); } void spa_spare_remove(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(vd->vdev_isspare); spa_aux_remove(vd, &spa_spare_avl); vd->vdev_isspare = B_FALSE; mutex_exit(&spa_spare_lock); } boolean_t spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) { boolean_t found; mutex_enter(&spa_spare_lock); found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); mutex_exit(&spa_spare_lock); return (found); } void spa_spare_activate(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(vd->vdev_isspare); spa_aux_activate(vd, &spa_spare_avl); mutex_exit(&spa_spare_lock); } /* * Level 2 ARC devices are tracked globally for the same reasons as spares. * Cache devices currently only support one pool per cache device, and so * for these devices the aux reference count is currently unused beyond 1. */ static int spa_l2cache_compare(const void *a, const void *b) { return (spa_aux_compare(a, b)); } void spa_l2cache_add(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(!vd->vdev_isl2cache); spa_aux_add(vd, &spa_l2cache_avl); vd->vdev_isl2cache = B_TRUE; mutex_exit(&spa_l2cache_lock); } void spa_l2cache_remove(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(vd->vdev_isl2cache); spa_aux_remove(vd, &spa_l2cache_avl); vd->vdev_isl2cache = B_FALSE; mutex_exit(&spa_l2cache_lock); } boolean_t spa_l2cache_exists(uint64_t guid, uint64_t *pool) { boolean_t found; mutex_enter(&spa_l2cache_lock); found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); mutex_exit(&spa_l2cache_lock); return (found); } void spa_l2cache_activate(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(vd->vdev_isl2cache); spa_aux_activate(vd, &spa_l2cache_avl); mutex_exit(&spa_l2cache_lock); } /* * ========================================================================== * SPA vdev locking * ========================================================================== */ /* * Lock the given spa_t for the purpose of adding or removing a vdev. * Grabs the global spa_namespace_lock plus the spa config lock for writing. * It returns the next transaction group for the spa_t. */ uint64_t spa_vdev_enter(spa_t *spa) { mutex_enter(&spa->spa_vdev_top_lock); mutex_enter(&spa_namespace_lock); return (spa_vdev_config_enter(spa)); } /* * Internal implementation for spa_vdev_enter(). Used when a vdev * operation requires multiple syncs (i.e. removing a device) while * keeping the spa_namespace_lock held. */ uint64_t spa_vdev_config_enter(spa_t *spa) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); return (spa_last_synced_txg(spa) + 1); } /* * Used in combination with spa_vdev_config_enter() to allow the syncing * of multiple transactions without releasing the spa_namespace_lock. */ void spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); int config_changed = B_FALSE; ASSERT(txg > spa_last_synced_txg(spa)); spa->spa_pending_vdev = NULL; /* * Reassess the DTLs. */ vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { config_changed = B_TRUE; spa->spa_config_generation++; } /* * Verify the metaslab classes. */ ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); spa_config_exit(spa, SCL_ALL, spa); /* * Panic the system if the specified tag requires it. This * is useful for ensuring that configurations are updated * transactionally. */ if (zio_injection_enabled) zio_handle_panic_injection(spa, tag, 0); /* * Note: this txg_wait_synced() is important because it ensures * that there won't be more than one config change per txg. * This allows us to use the txg as the generation number. */ if (error == 0) txg_wait_synced(spa->spa_dsl_pool, txg); if (vd != NULL) { ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); vdev_free(vd); spa_config_exit(spa, SCL_ALL, spa); } /* * If the config changed, update the config cache. */ if (config_changed) spa_config_sync(spa, B_FALSE, B_TRUE); } /* * Unlock the spa_t after adding or removing a vdev. Besides undoing the * locking of spa_vdev_enter(), we also want make sure the transactions have * synced to disk, and then update the global configuration cache with the new * information. */ int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) { spa_vdev_config_exit(spa, vd, txg, error, FTAG); mutex_exit(&spa_namespace_lock); mutex_exit(&spa->spa_vdev_top_lock); return (error); } /* * Lock the given spa_t for the purpose of changing vdev state. */ void spa_vdev_state_enter(spa_t *spa, int oplocks) { int locks = SCL_STATE_ALL | oplocks; /* * Root pools may need to read of the underlying devfs filesystem * when opening up a vdev. Unfortunately if we're holding the * SCL_ZIO lock it will result in a deadlock when we try to issue * the read from the root filesystem. Instead we "prefetch" * the associated vnodes that we need prior to opening the * underlying devices and cache them so that we can prevent * any I/O when we are doing the actual open. */ if (spa_is_root(spa)) { int low = locks & ~(SCL_ZIO - 1); int high = locks & ~low; spa_config_enter(spa, high, spa, RW_WRITER); vdev_hold(spa->spa_root_vdev); spa_config_enter(spa, low, spa, RW_WRITER); } else { spa_config_enter(spa, locks, spa, RW_WRITER); } spa->spa_vdev_locks = locks; } int spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) { boolean_t config_changed = B_FALSE; if (vd != NULL || error == 0) vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 0, 0, B_FALSE); if (vd != NULL) { vdev_state_dirty(vd->vdev_top); config_changed = B_TRUE; spa->spa_config_generation++; } if (spa_is_root(spa)) vdev_rele(spa->spa_root_vdev); ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); spa_config_exit(spa, spa->spa_vdev_locks, spa); /* * If anything changed, wait for it to sync. This ensures that, * from the system administrator's perspective, zpool(1M) commands * are synchronous. This is important for things like zpool offline: * when the command completes, you expect no further I/O from ZFS. */ if (vd != NULL) txg_wait_synced(spa->spa_dsl_pool, 0); /* * If the config changed, update the config cache. */ if (config_changed) { mutex_enter(&spa_namespace_lock); spa_config_sync(spa, B_FALSE, B_TRUE); mutex_exit(&spa_namespace_lock); } return (error); } /* * ========================================================================== * Miscellaneous functions * ========================================================================== */ void spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) { if (!nvlist_exists(spa->spa_label_features, feature)) { fnvlist_add_boolean(spa->spa_label_features, feature); /* * When we are creating the pool (tx_txg==TXG_INITIAL), we can't * dirty the vdev config because lock SCL_CONFIG is not held. * Thankfully, in this case we don't need to dirty the config * because it will be written out anyway when we finish * creating the pool. */ if (tx->tx_txg != TXG_INITIAL) vdev_config_dirty(spa->spa_root_vdev); } } void spa_deactivate_mos_feature(spa_t *spa, const char *feature) { if (nvlist_remove_all(spa->spa_label_features, feature) == 0) vdev_config_dirty(spa->spa_root_vdev); } /* * Rename a spa_t. */ int spa_rename(const char *name, const char *newname) { spa_t *spa; int err; /* * Lookup the spa_t and grab the config lock for writing. We need to * actually open the pool so that we can sync out the necessary labels. * It's OK to call spa_open() with the namespace lock held because we * allow recursive calls for other reasons. */ mutex_enter(&spa_namespace_lock); if ((err = spa_open(name, &spa, FTAG)) != 0) { mutex_exit(&spa_namespace_lock); return (err); } spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); avl_remove(&spa_namespace_avl, spa); (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); avl_add(&spa_namespace_avl, spa); /* * Sync all labels to disk with the new names by marking the root vdev * dirty and waiting for it to sync. It will pick up the new pool name * during the sync. */ vdev_config_dirty(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); txg_wait_synced(spa->spa_dsl_pool, 0); /* * Sync the updated config cache. */ spa_config_sync(spa, B_FALSE, B_TRUE); spa_close(spa, FTAG); mutex_exit(&spa_namespace_lock); return (0); } /* * Return the spa_t associated with given pool_guid, if it exists. If * device_guid is non-zero, determine whether the pool exists *and* contains * a device with the specified device_guid. */ spa_t * spa_by_guid(uint64_t pool_guid, uint64_t device_guid) { spa_t *spa; avl_tree_t *t = &spa_namespace_avl; ASSERT(MUTEX_HELD(&spa_namespace_lock)); for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { if (spa->spa_state == POOL_STATE_UNINITIALIZED) continue; if (spa->spa_root_vdev == NULL) continue; if (spa_guid(spa) == pool_guid) { if (device_guid == 0) break; if (vdev_lookup_by_guid(spa->spa_root_vdev, device_guid) != NULL) break; /* * Check any devices we may be in the process of adding. */ if (spa->spa_pending_vdev) { if (vdev_lookup_by_guid(spa->spa_pending_vdev, device_guid) != NULL) break; } } } return (spa); } /* * Determine whether a pool with the given pool_guid exists. */ boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) { return (spa_by_guid(pool_guid, device_guid) != NULL); } char * spa_strdup(const char *s) { size_t len; char *new; len = strlen(s); new = kmem_alloc(len + 1, KM_SLEEP); bcopy(s, new, len); new[len] = '\0'; return (new); } void spa_strfree(char *s) { kmem_free(s, strlen(s) + 1); } uint64_t spa_get_random(uint64_t range) { uint64_t r; ASSERT(range != 0); (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); return (r % range); } uint64_t spa_generate_guid(spa_t *spa) { uint64_t guid = spa_get_random(-1ULL); if (spa != NULL) { while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) guid = spa_get_random(-1ULL); } else { while (guid == 0 || spa_guid_exists(guid, 0)) guid = spa_get_random(-1ULL); } return (guid); } void snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) { char type[256]; char *checksum = NULL; char *compress = NULL; if (bp != NULL) { if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { dmu_object_byteswap_t bswap = DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); (void) snprintf(type, sizeof (type), "bswap %s %s", DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? "metadata" : "data", dmu_ot_byteswap[bswap].ob_name); } else { (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, sizeof (type)); } if (!BP_IS_EMBEDDED(bp)) { checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; } compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; } SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, compress); } void spa_freeze(spa_t *spa) { uint64_t freeze_txg = 0; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); if (spa->spa_freeze_txg == UINT64_MAX) { freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; spa->spa_freeze_txg = freeze_txg; } spa_config_exit(spa, SCL_ALL, FTAG); if (freeze_txg != 0) txg_wait_synced(spa_get_dsl(spa), freeze_txg); } void zfs_panic_recover(const char *fmt, ...) { va_list adx; va_start(adx, fmt); vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); va_end(adx); } /* * This is a stripped-down version of strtoull, suitable only for converting * lowercase hexadecimal numbers that don't overflow. */ uint64_t zfs_strtonum(const char *str, char **nptr) { uint64_t val = 0; char c; int digit; while ((c = *str) != '\0') { if (c >= '0' && c <= '9') digit = c - '0'; else if (c >= 'a' && c <= 'f') digit = 10 + c - 'a'; else break; val *= 16; val += digit; str++; } if (nptr) *nptr = (char *)str; return (val); } /* * ========================================================================== * Accessor functions * ========================================================================== */ boolean_t spa_shutting_down(spa_t *spa) { return (spa->spa_async_suspended); } dsl_pool_t * spa_get_dsl(spa_t *spa) { return (spa->spa_dsl_pool); } boolean_t spa_is_initializing(spa_t *spa) { return (spa->spa_is_initializing); } blkptr_t * spa_get_rootblkptr(spa_t *spa) { return (&spa->spa_ubsync.ub_rootbp); } void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) { spa->spa_uberblock.ub_rootbp = *bp; } void spa_altroot(spa_t *spa, char *buf, size_t buflen) { if (spa->spa_root == NULL) buf[0] = '\0'; else (void) strncpy(buf, spa->spa_root, buflen); } int spa_sync_pass(spa_t *spa) { return (spa->spa_sync_pass); } char * spa_name(spa_t *spa) { return (spa->spa_name); } uint64_t spa_guid(spa_t *spa) { dsl_pool_t *dp = spa_get_dsl(spa); uint64_t guid; /* * If we fail to parse the config during spa_load(), we can go through * the error path (which posts an ereport) and end up here with no root * vdev. We stash the original pool guid in 'spa_config_guid' to handle * this case. */ if (spa->spa_root_vdev == NULL) return (spa->spa_config_guid); guid = spa->spa_last_synced_guid != 0 ? spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; /* * Return the most recently synced out guid unless we're * in syncing context. */ if (dp && dsl_pool_sync_context(dp)) return (spa->spa_root_vdev->vdev_guid); else return (guid); } uint64_t spa_load_guid(spa_t *spa) { /* * This is a GUID that exists solely as a reference for the * purposes of the arc. It is generated at load time, and * is never written to persistent storage. */ return (spa->spa_load_guid); } uint64_t spa_last_synced_txg(spa_t *spa) { return (spa->spa_ubsync.ub_txg); } uint64_t spa_first_txg(spa_t *spa) { return (spa->spa_first_txg); } uint64_t spa_syncing_txg(spa_t *spa) { return (spa->spa_syncing_txg); } +/* + * Return the last txg where data can be dirtied. The final txgs + * will be used to just clear out any deferred frees that remain. + */ +uint64_t +spa_final_dirty_txg(spa_t *spa) +{ + return (spa->spa_final_txg - TXG_DEFER_SIZE); +} + pool_state_t spa_state(spa_t *spa) { return (spa->spa_state); } spa_load_state_t spa_load_state(spa_t *spa) { return (spa->spa_load_state); } uint64_t spa_freeze_txg(spa_t *spa) { return (spa->spa_freeze_txg); } /* ARGSUSED */ uint64_t spa_get_worst_case_asize(spa_t *spa, uint64_t lsize) { return (lsize * spa_asize_inflation); } /* * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), * or at least 128MB, unless that would cause it to be more than half the * pool size. * * See the comment above spa_slop_shift for details. */ uint64_t spa_get_slop_space(spa_t *spa) { uint64_t space = spa_get_dspace(spa); return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop))); } uint64_t spa_get_dspace(spa_t *spa) { return (spa->spa_dspace); } void spa_update_dspace(spa_t *spa) { spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + ddt_get_dedup_dspace(spa); } /* * Return the failure mode that has been set to this pool. The default * behavior will be to block all I/Os when a complete failure occurs. */ uint8_t spa_get_failmode(spa_t *spa) { return (spa->spa_failmode); } boolean_t spa_suspended(spa_t *spa) { return (spa->spa_suspended); } uint64_t spa_version(spa_t *spa) { return (spa->spa_ubsync.ub_version); } boolean_t spa_deflate(spa_t *spa) { return (spa->spa_deflate); } metaslab_class_t * spa_normal_class(spa_t *spa) { return (spa->spa_normal_class); } metaslab_class_t * spa_log_class(spa_t *spa) { return (spa->spa_log_class); } void spa_evicting_os_register(spa_t *spa, objset_t *os) { mutex_enter(&spa->spa_evicting_os_lock); list_insert_head(&spa->spa_evicting_os_list, os); mutex_exit(&spa->spa_evicting_os_lock); } void spa_evicting_os_deregister(spa_t *spa, objset_t *os) { mutex_enter(&spa->spa_evicting_os_lock); list_remove(&spa->spa_evicting_os_list, os); cv_broadcast(&spa->spa_evicting_os_cv); mutex_exit(&spa->spa_evicting_os_lock); } void spa_evicting_os_wait(spa_t *spa) { mutex_enter(&spa->spa_evicting_os_lock); while (!list_is_empty(&spa->spa_evicting_os_list)) cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock); mutex_exit(&spa->spa_evicting_os_lock); dmu_buf_user_evict_wait(); } int spa_max_replication(spa_t *spa) { /* * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to * handle BPs with more than one DVA allocated. Set our max * replication level accordingly. */ if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) return (1); return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); } int spa_prev_software_version(spa_t *spa) { return (spa->spa_prev_software_version); } uint64_t spa_deadman_synctime(spa_t *spa) { return (spa->spa_deadman_synctime); } uint64_t dva_get_dsize_sync(spa_t *spa, const dva_t *dva) { uint64_t asize = DVA_GET_ASIZE(dva); uint64_t dsize = asize; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); if (asize != 0 && spa->spa_deflate) { uint64_t vdev = DVA_GET_VDEV(dva); vdev_t *vd = vdev_lookup_top(spa, vdev); if (vd == NULL) { panic( "dva_get_dsize_sync(): bad DVA %llu:%llu", (u_longlong_t)vdev, (u_longlong_t)asize); } dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; } return (dsize); } uint64_t bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) { uint64_t dsize = 0; for (int d = 0; d < BP_GET_NDVAS(bp); d++) dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); return (dsize); } uint64_t bp_get_dsize(spa_t *spa, const blkptr_t *bp) { uint64_t dsize = 0; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (int d = 0; d < BP_GET_NDVAS(bp); d++) dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); spa_config_exit(spa, SCL_VDEV, FTAG); return (dsize); } /* * ========================================================================== * Initialization and Termination * ========================================================================== */ static int spa_name_compare(const void *a1, const void *a2) { const spa_t *s1 = a1; const spa_t *s2 = a2; int s; s = strcmp(s1->spa_name, s2->spa_name); if (s > 0) return (1); if (s < 0) return (-1); return (0); } int spa_busy(void) { return (spa_active_count); } void spa_boot_init() { spa_config_load(); } #ifdef _KERNEL EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); #endif void spa_init(int mode) { mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), offsetof(spa_t, spa_avl)); avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), offsetof(spa_aux_t, aux_avl)); avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), offsetof(spa_aux_t, aux_avl)); spa_mode_global = mode; #ifdef illumos #ifdef _KERNEL spa_arch_init(); #else if (spa_mode_global != FREAD && dprintf_find_string("watch")) { arc_procfd = open("/proc/self/ctl", O_WRONLY); if (arc_procfd == -1) { perror("could not enable watchpoints: " "opening /proc/self/ctl failed: "); } else { arc_watch = B_TRUE; } } #endif #endif /* illumos */ refcount_sysinit(); unique_init(); range_tree_init(); metaslab_alloc_trace_init(); zio_init(); lz4_init(); dmu_init(); zil_init(); vdev_cache_stat_init(); zfs_prop_init(); zpool_prop_init(); zpool_feature_init(); spa_config_load(); l2arc_start(); #ifndef illumos #ifdef _KERNEL zfs_deadman_init(); #endif #endif /* !illumos */ } void spa_fini(void) { l2arc_stop(); spa_evict_all(); vdev_cache_stat_fini(); zil_fini(); dmu_fini(); lz4_fini(); zio_fini(); metaslab_alloc_trace_fini(); range_tree_fini(); unique_fini(); refcount_fini(); avl_destroy(&spa_namespace_avl); avl_destroy(&spa_spare_avl); avl_destroy(&spa_l2cache_avl); cv_destroy(&spa_namespace_cv); mutex_destroy(&spa_namespace_lock); mutex_destroy(&spa_spare_lock); mutex_destroy(&spa_l2cache_lock); } /* * Return whether this pool has slogs. No locking needed. * It's not a problem if the wrong answer is returned as it's only for * performance and not correctness */ boolean_t spa_has_slogs(spa_t *spa) { return (spa->spa_log_class->mc_rotor != NULL); } spa_log_state_t spa_get_log_state(spa_t *spa) { return (spa->spa_log_state); } void spa_set_log_state(spa_t *spa, spa_log_state_t state) { spa->spa_log_state = state; } boolean_t spa_is_root(spa_t *spa) { return (spa->spa_is_root); } boolean_t spa_writeable(spa_t *spa) { return (!!(spa->spa_mode & FWRITE)); } /* * Returns true if there is a pending sync task in any of the current * syncing txg, the current quiescing txg, or the current open txg. */ boolean_t spa_has_pending_synctask(spa_t *spa) { return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks)); } int spa_mode(spa_t *spa) { return (spa->spa_mode); } uint64_t spa_bootfs(spa_t *spa) { return (spa->spa_bootfs); } uint64_t spa_delegation(spa_t *spa) { return (spa->spa_delegation); } objset_t * spa_meta_objset(spa_t *spa) { return (spa->spa_meta_objset); } enum zio_checksum spa_dedup_checksum(spa_t *spa) { return (spa->spa_dedup_checksum); } /* * Reset pool scan stat per scan pass (or reboot). */ void spa_scan_stat_init(spa_t *spa) { /* data not stored on disk */ spa->spa_scan_pass_start = gethrestime_sec(); spa->spa_scan_pass_exam = 0; vdev_scan_stat_init(spa->spa_root_vdev); } /* * Get scan stats for zpool status reports */ int spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) { dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) return (SET_ERROR(ENOENT)); bzero(ps, sizeof (pool_scan_stat_t)); /* data stored on disk */ ps->pss_func = scn->scn_phys.scn_func; ps->pss_start_time = scn->scn_phys.scn_start_time; ps->pss_end_time = scn->scn_phys.scn_end_time; ps->pss_to_examine = scn->scn_phys.scn_to_examine; ps->pss_examined = scn->scn_phys.scn_examined; ps->pss_to_process = scn->scn_phys.scn_to_process; ps->pss_processed = scn->scn_phys.scn_processed; ps->pss_errors = scn->scn_phys.scn_errors; ps->pss_state = scn->scn_phys.scn_state; /* data not stored on disk */ ps->pss_pass_start = spa->spa_scan_pass_start; ps->pss_pass_exam = spa->spa_scan_pass_exam; return (0); } boolean_t spa_debug_enabled(spa_t *spa) { return (spa->spa_debug); } int spa_maxblocksize(spa_t *spa) { if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) return (SPA_MAXBLOCKSIZE); else return (SPA_OLD_MAXBLOCKSIZE); } Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/space_map.c =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/space_map.c (revision 321553) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/space_map.c (revision 321554) @@ -1,546 +1,548 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* - * Copyright (c) 2012, 2014 by Delphix. All rights reserved. + * Copyright (c) 2012, 2016 by Delphix. All rights reserved. */ #include #include #include #include #include #include #include #include #include #include SYSCTL_DECL(_vfs_zfs); /* * The data for a given space map can be kept on blocks of any size. * Larger blocks entail fewer i/o operations, but they also cause the * DMU to keep more data in-core, and also to waste more i/o bandwidth * when only a few blocks have changed since the last transaction group. */ int space_map_blksz = (1 << 12); SYSCTL_INT(_vfs_zfs, OID_AUTO, space_map_blksz, CTLFLAG_RDTUN, &space_map_blksz, 0, "Maximum block size for space map. Must be power of 2 and greater than 4096."); /* * Load the space map disk into the specified range tree. Segments of maptype * are added to the range tree, other segment types are removed. * * Note: space_map_load() will drop sm_lock across dmu_read() calls. * The caller must be OK with this. */ int space_map_load(space_map_t *sm, range_tree_t *rt, maptype_t maptype) { uint64_t *entry, *entry_map, *entry_map_end; uint64_t bufsize, size, offset, end, space; int error = 0; ASSERT(MUTEX_HELD(sm->sm_lock)); end = space_map_length(sm); space = space_map_allocated(sm); VERIFY0(range_tree_space(rt)); if (maptype == SM_FREE) { range_tree_add(rt, sm->sm_start, sm->sm_size); space = sm->sm_size - space; } bufsize = MAX(sm->sm_blksz, SPA_MINBLOCKSIZE); entry_map = zio_buf_alloc(bufsize); mutex_exit(sm->sm_lock); if (end > bufsize) { dmu_prefetch(sm->sm_os, space_map_object(sm), 0, bufsize, end - bufsize, ZIO_PRIORITY_SYNC_READ); } mutex_enter(sm->sm_lock); for (offset = 0; offset < end; offset += bufsize) { size = MIN(end - offset, bufsize); VERIFY(P2PHASE(size, sizeof (uint64_t)) == 0); VERIFY(size != 0); ASSERT3U(sm->sm_blksz, !=, 0); dprintf("object=%llu offset=%llx size=%llx\n", space_map_object(sm), offset, size); mutex_exit(sm->sm_lock); error = dmu_read(sm->sm_os, space_map_object(sm), offset, size, entry_map, DMU_READ_PREFETCH); mutex_enter(sm->sm_lock); if (error != 0) break; entry_map_end = entry_map + (size / sizeof (uint64_t)); for (entry = entry_map; entry < entry_map_end; entry++) { uint64_t e = *entry; uint64_t offset, size; if (SM_DEBUG_DECODE(e)) /* Skip debug entries */ continue; offset = (SM_OFFSET_DECODE(e) << sm->sm_shift) + sm->sm_start; size = SM_RUN_DECODE(e) << sm->sm_shift; VERIFY0(P2PHASE(offset, 1ULL << sm->sm_shift)); VERIFY0(P2PHASE(size, 1ULL << sm->sm_shift)); VERIFY3U(offset, >=, sm->sm_start); VERIFY3U(offset + size, <=, sm->sm_start + sm->sm_size); if (SM_TYPE_DECODE(e) == maptype) { VERIFY3U(range_tree_space(rt) + size, <=, sm->sm_size); range_tree_add(rt, offset, size); } else { range_tree_remove(rt, offset, size); } } } if (error == 0) VERIFY3U(range_tree_space(rt), ==, space); else range_tree_vacate(rt, NULL, NULL); zio_buf_free(entry_map, bufsize); return (error); } void space_map_histogram_clear(space_map_t *sm) { if (sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) return; bzero(sm->sm_phys->smp_histogram, sizeof (sm->sm_phys->smp_histogram)); } boolean_t space_map_histogram_verify(space_map_t *sm, range_tree_t *rt) { /* * Verify that the in-core range tree does not have any * ranges smaller than our sm_shift size. */ for (int i = 0; i < sm->sm_shift; i++) { if (rt->rt_histogram[i] != 0) return (B_FALSE); } return (B_TRUE); } void space_map_histogram_add(space_map_t *sm, range_tree_t *rt, dmu_tx_t *tx) { int idx = 0; ASSERT(MUTEX_HELD(rt->rt_lock)); ASSERT(dmu_tx_is_syncing(tx)); VERIFY3U(space_map_object(sm), !=, 0); if (sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) return; dmu_buf_will_dirty(sm->sm_dbuf, tx); ASSERT(space_map_histogram_verify(sm, rt)); /* * Transfer the content of the range tree histogram to the space * map histogram. The space map histogram contains 32 buckets ranging * between 2^sm_shift to 2^(32+sm_shift-1). The range tree, * however, can represent ranges from 2^0 to 2^63. Since the space * map only cares about allocatable blocks (minimum of sm_shift) we * can safely ignore all ranges in the range tree smaller than sm_shift. */ for (int i = sm->sm_shift; i < RANGE_TREE_HISTOGRAM_SIZE; i++) { /* * Since the largest histogram bucket in the space map is * 2^(32+sm_shift-1), we need to normalize the values in * the range tree for any bucket larger than that size. For * example given an sm_shift of 9, ranges larger than 2^40 * would get normalized as if they were 1TB ranges. Assume * the range tree had a count of 5 in the 2^44 (16TB) bucket, * the calculation below would normalize this to 5 * 2^4 (16). */ ASSERT3U(i, >=, idx + sm->sm_shift); sm->sm_phys->smp_histogram[idx] += rt->rt_histogram[i] << (i - idx - sm->sm_shift); /* * Increment the space map's index as long as we haven't * reached the maximum bucket size. Accumulate all ranges * larger than the max bucket size into the last bucket. */ if (idx < SPACE_MAP_HISTOGRAM_SIZE - 1) { ASSERT3U(idx + sm->sm_shift, ==, i); idx++; ASSERT3U(idx, <, SPACE_MAP_HISTOGRAM_SIZE); } } } uint64_t space_map_entries(space_map_t *sm, range_tree_t *rt) { avl_tree_t *t = &rt->rt_root; range_seg_t *rs; uint64_t size, entries; /* * All space_maps always have a debug entry so account for it here. */ entries = 1; /* * Traverse the range tree and calculate the number of space map * entries that would be required to write out the range tree. */ for (rs = avl_first(t); rs != NULL; rs = AVL_NEXT(t, rs)) { size = (rs->rs_end - rs->rs_start) >> sm->sm_shift; entries += howmany(size, SM_RUN_MAX); } return (entries); } /* * Note: space_map_write() will drop sm_lock across dmu_write() calls. */ void space_map_write(space_map_t *sm, range_tree_t *rt, maptype_t maptype, dmu_tx_t *tx) { objset_t *os = sm->sm_os; spa_t *spa = dmu_objset_spa(os); avl_tree_t *t = &rt->rt_root; range_seg_t *rs; uint64_t size, total, rt_space, nodes; uint64_t *entry, *entry_map, *entry_map_end; uint64_t expected_entries, actual_entries = 1; ASSERT(MUTEX_HELD(rt->rt_lock)); ASSERT(dsl_pool_sync_context(dmu_objset_pool(os))); VERIFY3U(space_map_object(sm), !=, 0); dmu_buf_will_dirty(sm->sm_dbuf, tx); /* * This field is no longer necessary since the in-core space map * now contains the object number but is maintained for backwards * compatibility. */ sm->sm_phys->smp_object = sm->sm_object; if (range_tree_space(rt) == 0) { VERIFY3U(sm->sm_object, ==, sm->sm_phys->smp_object); return; } if (maptype == SM_ALLOC) sm->sm_phys->smp_alloc += range_tree_space(rt); else sm->sm_phys->smp_alloc -= range_tree_space(rt); expected_entries = space_map_entries(sm, rt); entry_map = zio_buf_alloc(sm->sm_blksz); entry_map_end = entry_map + (sm->sm_blksz / sizeof (uint64_t)); entry = entry_map; *entry++ = SM_DEBUG_ENCODE(1) | SM_DEBUG_ACTION_ENCODE(maptype) | SM_DEBUG_SYNCPASS_ENCODE(spa_sync_pass(spa)) | SM_DEBUG_TXG_ENCODE(dmu_tx_get_txg(tx)); total = 0; nodes = avl_numnodes(&rt->rt_root); rt_space = range_tree_space(rt); for (rs = avl_first(t); rs != NULL; rs = AVL_NEXT(t, rs)) { uint64_t start; size = (rs->rs_end - rs->rs_start) >> sm->sm_shift; start = (rs->rs_start - sm->sm_start) >> sm->sm_shift; total += size << sm->sm_shift; while (size != 0) { uint64_t run_len; run_len = MIN(size, SM_RUN_MAX); if (entry == entry_map_end) { mutex_exit(rt->rt_lock); dmu_write(os, space_map_object(sm), sm->sm_phys->smp_objsize, sm->sm_blksz, entry_map, tx); mutex_enter(rt->rt_lock); sm->sm_phys->smp_objsize += sm->sm_blksz; entry = entry_map; } *entry++ = SM_OFFSET_ENCODE(start) | SM_TYPE_ENCODE(maptype) | SM_RUN_ENCODE(run_len); start += run_len; size -= run_len; actual_entries++; } } if (entry != entry_map) { size = (entry - entry_map) * sizeof (uint64_t); mutex_exit(rt->rt_lock); dmu_write(os, space_map_object(sm), sm->sm_phys->smp_objsize, size, entry_map, tx); mutex_enter(rt->rt_lock); sm->sm_phys->smp_objsize += size; } ASSERT3U(expected_entries, ==, actual_entries); /* * Ensure that the space_map's accounting wasn't changed * while we were in the middle of writing it out. */ VERIFY3U(nodes, ==, avl_numnodes(&rt->rt_root)); VERIFY3U(range_tree_space(rt), ==, rt_space); VERIFY3U(range_tree_space(rt), ==, total); zio_buf_free(entry_map, sm->sm_blksz); } static int space_map_open_impl(space_map_t *sm) { int error; u_longlong_t blocks; error = dmu_bonus_hold(sm->sm_os, sm->sm_object, sm, &sm->sm_dbuf); if (error) return (error); dmu_object_size_from_db(sm->sm_dbuf, &sm->sm_blksz, &blocks); sm->sm_phys = sm->sm_dbuf->db_data; return (0); } int space_map_open(space_map_t **smp, objset_t *os, uint64_t object, uint64_t start, uint64_t size, uint8_t shift, kmutex_t *lp) { space_map_t *sm; int error; ASSERT(*smp == NULL); ASSERT(os != NULL); ASSERT(object != 0); sm = kmem_zalloc(sizeof (space_map_t), KM_SLEEP); sm->sm_start = start; sm->sm_size = size; sm->sm_shift = shift; sm->sm_lock = lp; sm->sm_os = os; sm->sm_object = object; error = space_map_open_impl(sm); if (error != 0) { space_map_close(sm); return (error); } *smp = sm; return (0); } void space_map_close(space_map_t *sm) { if (sm == NULL) return; if (sm->sm_dbuf != NULL) dmu_buf_rele(sm->sm_dbuf, sm); sm->sm_dbuf = NULL; sm->sm_phys = NULL; kmem_free(sm, sizeof (*sm)); } void space_map_truncate(space_map_t *sm, dmu_tx_t *tx) { objset_t *os = sm->sm_os; spa_t *spa = dmu_objset_spa(os); dmu_object_info_t doi; ASSERT(dsl_pool_sync_context(dmu_objset_pool(os))); ASSERT(dmu_tx_is_syncing(tx)); + VERIFY3U(dmu_tx_get_txg(tx), <=, spa_final_dirty_txg(spa)); dmu_object_info_from_db(sm->sm_dbuf, &doi); /* * If the space map has the wrong bonus size (because * SPA_FEATURE_SPACEMAP_HISTOGRAM has recently been enabled), or * the wrong block size (because space_map_blksz has changed), * free and re-allocate its object with the updated sizes. * * Otherwise, just truncate the current object. */ if ((spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM) && doi.doi_bonus_size != sizeof (space_map_phys_t)) || doi.doi_data_block_size != space_map_blksz) { - zfs_dbgmsg("txg %llu, spa %s, reallocating: " - "old bonus %u, old blocksz %u", dmu_tx_get_txg(tx), - spa_name(spa), doi.doi_bonus_size, doi.doi_data_block_size); + zfs_dbgmsg("txg %llu, spa %s, sm %p, reallocating " + "object[%llu]: old bonus %u, old blocksz %u", + dmu_tx_get_txg(tx), spa_name(spa), sm, sm->sm_object, + doi.doi_bonus_size, doi.doi_data_block_size); space_map_free(sm, tx); dmu_buf_rele(sm->sm_dbuf, sm); sm->sm_object = space_map_alloc(sm->sm_os, tx); VERIFY0(space_map_open_impl(sm)); } else { VERIFY0(dmu_free_range(os, space_map_object(sm), 0, -1ULL, tx)); /* * If the spacemap is reallocated, its histogram * will be reset. Do the same in the common case so that * bugs related to the uncommon case do not go unnoticed. */ bzero(sm->sm_phys->smp_histogram, sizeof (sm->sm_phys->smp_histogram)); } dmu_buf_will_dirty(sm->sm_dbuf, tx); sm->sm_phys->smp_objsize = 0; sm->sm_phys->smp_alloc = 0; } /* * Update the in-core space_map allocation and length values. */ void space_map_update(space_map_t *sm) { if (sm == NULL) return; ASSERT(MUTEX_HELD(sm->sm_lock)); sm->sm_alloc = sm->sm_phys->smp_alloc; sm->sm_length = sm->sm_phys->smp_objsize; } uint64_t space_map_alloc(objset_t *os, dmu_tx_t *tx) { spa_t *spa = dmu_objset_spa(os); uint64_t object; int bonuslen; if (spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) { spa_feature_incr(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM, tx); bonuslen = sizeof (space_map_phys_t); ASSERT3U(bonuslen, <=, dmu_bonus_max()); } else { bonuslen = SPACE_MAP_SIZE_V0; } object = dmu_object_alloc(os, DMU_OT_SPACE_MAP, space_map_blksz, DMU_OT_SPACE_MAP_HEADER, bonuslen, tx); return (object); } void space_map_free(space_map_t *sm, dmu_tx_t *tx) { spa_t *spa; if (sm == NULL) return; spa = dmu_objset_spa(sm->sm_os); if (spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) { dmu_object_info_t doi; dmu_object_info_from_db(sm->sm_dbuf, &doi); if (doi.doi_bonus_size != SPACE_MAP_SIZE_V0) { VERIFY(spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)); spa_feature_decr(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM, tx); } } VERIFY3U(dmu_object_free(sm->sm_os, space_map_object(sm), tx), ==, 0); sm->sm_object = 0; } uint64_t space_map_object(space_map_t *sm) { return (sm != NULL ? sm->sm_object : 0); } /* * Returns the already synced, on-disk allocated space. */ uint64_t space_map_allocated(space_map_t *sm) { return (sm != NULL ? sm->sm_alloc : 0); } /* * Returns the already synced, on-disk length; */ uint64_t space_map_length(space_map_t *sm) { return (sm != NULL ? sm->sm_length : 0); } /* * Returns the allocated space that is currently syncing. */ int64_t space_map_alloc_delta(space_map_t *sm) { if (sm == NULL) return (0); ASSERT(sm->sm_dbuf != NULL); return (sm->sm_phys->smp_alloc - space_map_allocated(sm)); } Index: stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa.h =================================================================== --- stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa.h (revision 321553) +++ stable/11/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa.h (revision 321554) @@ -1,926 +1,927 @@ /* * 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 (c) 2011, 2016 by Delphix. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright 2013 Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #ifndef _SYS_SPA_H #define _SYS_SPA_H #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* * Forward references that lots of things need. */ typedef struct spa spa_t; typedef struct vdev vdev_t; typedef struct metaslab metaslab_t; typedef struct metaslab_group metaslab_group_t; typedef struct metaslab_class metaslab_class_t; typedef struct zio zio_t; typedef struct zilog zilog_t; typedef struct spa_aux_vdev spa_aux_vdev_t; typedef struct ddt ddt_t; typedef struct ddt_entry ddt_entry_t; struct dsl_pool; struct dsl_dataset; /* * General-purpose 32-bit and 64-bit bitfield encodings. */ #define BF32_DECODE(x, low, len) P2PHASE((x) >> (low), 1U << (len)) #define BF64_DECODE(x, low, len) P2PHASE((x) >> (low), 1ULL << (len)) #define BF32_ENCODE(x, low, len) (P2PHASE((x), 1U << (len)) << (low)) #define BF64_ENCODE(x, low, len) (P2PHASE((x), 1ULL << (len)) << (low)) #define BF32_GET(x, low, len) BF32_DECODE(x, low, len) #define BF64_GET(x, low, len) BF64_DECODE(x, low, len) #define BF32_SET(x, low, len, val) do { \ ASSERT3U(val, <, 1U << (len)); \ ASSERT3U(low + len, <=, 32); \ (x) ^= BF32_ENCODE((x >> low) ^ (val), low, len); \ _NOTE(CONSTCOND) } while (0) #define BF64_SET(x, low, len, val) do { \ ASSERT3U(val, <, 1ULL << (len)); \ ASSERT3U(low + len, <=, 64); \ ((x) ^= BF64_ENCODE((x >> low) ^ (val), low, len)); \ _NOTE(CONSTCOND) } while (0) #define BF32_GET_SB(x, low, len, shift, bias) \ ((BF32_GET(x, low, len) + (bias)) << (shift)) #define BF64_GET_SB(x, low, len, shift, bias) \ ((BF64_GET(x, low, len) + (bias)) << (shift)) #define BF32_SET_SB(x, low, len, shift, bias, val) do { \ ASSERT(IS_P2ALIGNED(val, 1U << shift)); \ ASSERT3S((val) >> (shift), >=, bias); \ BF32_SET(x, low, len, ((val) >> (shift)) - (bias)); \ _NOTE(CONSTCOND) } while (0) #define BF64_SET_SB(x, low, len, shift, bias, val) do { \ ASSERT(IS_P2ALIGNED(val, 1ULL << shift)); \ ASSERT3S((val) >> (shift), >=, bias); \ BF64_SET(x, low, len, ((val) >> (shift)) - (bias)); \ _NOTE(CONSTCOND) } while (0) /* * We currently support block sizes from 512 bytes to 16MB. * The benefits of larger blocks, and thus larger IO, need to be weighed * against the cost of COWing a giant block to modify one byte, and the * large latency of reading or writing a large block. * * Note that although blocks up to 16MB are supported, the recordsize * property can not be set larger than zfs_max_recordsize (default 1MB). * See the comment near zfs_max_recordsize in dsl_dataset.c for details. * * Note that although the LSIZE field of the blkptr_t can store sizes up * to 32MB, the dnode's dn_datablkszsec can only store sizes up to * 32MB - 512 bytes. Therefore, we limit SPA_MAXBLOCKSIZE to 16MB. */ #define SPA_MINBLOCKSHIFT 9 #define SPA_OLD_MAXBLOCKSHIFT 17 #define SPA_MAXBLOCKSHIFT 24 #define SPA_MINBLOCKSIZE (1ULL << SPA_MINBLOCKSHIFT) #define SPA_OLD_MAXBLOCKSIZE (1ULL << SPA_OLD_MAXBLOCKSHIFT) #define SPA_MAXBLOCKSIZE (1ULL << SPA_MAXBLOCKSHIFT) /* * Default maximum supported logical ashift. * * The current 8k allocation block size limit is due to the 8k * aligned/sized operations performed by vdev_probe() on * vdev_label->vl_pad2. Using another "safe region" for these tests * would allow the limit to be raised to 16k, at the expense of * only having 8 available uberblocks in the label area. */ #define SPA_MAXASHIFT 13 /* * Default minimum supported logical ashift. */ #define SPA_MINASHIFT SPA_MINBLOCKSHIFT /* * Size of block to hold the configuration data (a packed nvlist) */ #define SPA_CONFIG_BLOCKSIZE (1ULL << 14) /* * The DVA size encodings for LSIZE and PSIZE support blocks up to 32MB. * The ASIZE encoding should be at least 64 times larger (6 more bits) * to support up to 4-way RAID-Z mirror mode with worst-case gang block * overhead, three DVAs per bp, plus one more bit in case we do anything * else that expands the ASIZE. */ #define SPA_LSIZEBITS 16 /* LSIZE up to 32M (2^16 * 512) */ #define SPA_PSIZEBITS 16 /* PSIZE up to 32M (2^16 * 512) */ #define SPA_ASIZEBITS 24 /* ASIZE up to 64 times larger */ #define SPA_COMPRESSBITS 7 /* * All SPA data is represented by 128-bit data virtual addresses (DVAs). * The members of the dva_t should be considered opaque outside the SPA. */ typedef struct dva { uint64_t dva_word[2]; } dva_t; /* * Each block has a 256-bit checksum -- strong enough for cryptographic hashes. */ typedef struct zio_cksum { uint64_t zc_word[4]; } zio_cksum_t; /* * Some checksums/hashes need a 256-bit initialization salt. This salt is kept * secret and is suitable for use in MAC algorithms as the key. */ typedef struct zio_cksum_salt { uint8_t zcs_bytes[32]; } zio_cksum_salt_t; /* * Each block is described by its DVAs, time of birth, checksum, etc. * The word-by-word, bit-by-bit layout of the blkptr is as follows: * * 64 56 48 40 32 24 16 8 0 * +-------+-------+-------+-------+-------+-------+-------+-------+ * 0 | vdev1 | GRID | ASIZE | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 1 |G| offset1 | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 2 | vdev2 | GRID | ASIZE | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 3 |G| offset2 | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 4 | vdev3 | GRID | ASIZE | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 5 |G| offset3 | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 6 |BDX|lvl| type | cksum |E| comp| PSIZE | LSIZE | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 7 | padding | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 8 | padding | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 9 | physical birth txg | * +-------+-------+-------+-------+-------+-------+-------+-------+ * a | logical birth txg | * +-------+-------+-------+-------+-------+-------+-------+-------+ * b | fill count | * +-------+-------+-------+-------+-------+-------+-------+-------+ * c | checksum[0] | * +-------+-------+-------+-------+-------+-------+-------+-------+ * d | checksum[1] | * +-------+-------+-------+-------+-------+-------+-------+-------+ * e | checksum[2] | * +-------+-------+-------+-------+-------+-------+-------+-------+ * f | checksum[3] | * +-------+-------+-------+-------+-------+-------+-------+-------+ * * Legend: * * vdev virtual device ID * offset offset into virtual device * LSIZE logical size * PSIZE physical size (after compression) * ASIZE allocated size (including RAID-Z parity and gang block headers) * GRID RAID-Z layout information (reserved for future use) * cksum checksum function * comp compression function * G gang block indicator * B byteorder (endianness) * D dedup * X encryption (on version 30, which is not supported) * E blkptr_t contains embedded data (see below) * lvl level of indirection * type DMU object type * phys birth txg of block allocation; zero if same as logical birth txg * log. birth transaction group in which the block was logically born * fill count number of non-zero blocks under this bp * checksum[4] 256-bit checksum of the data this bp describes */ /* * "Embedded" blkptr_t's don't actually point to a block, instead they * have a data payload embedded in the blkptr_t itself. See the comment * in blkptr.c for more details. * * The blkptr_t is laid out as follows: * * 64 56 48 40 32 24 16 8 0 * +-------+-------+-------+-------+-------+-------+-------+-------+ * 0 | payload | * 1 | payload | * 2 | payload | * 3 | payload | * 4 | payload | * 5 | payload | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 6 |BDX|lvl| type | etype |E| comp| PSIZE| LSIZE | * +-------+-------+-------+-------+-------+-------+-------+-------+ * 7 | payload | * 8 | payload | * 9 | payload | * +-------+-------+-------+-------+-------+-------+-------+-------+ * a | logical birth txg | * +-------+-------+-------+-------+-------+-------+-------+-------+ * b | payload | * c | payload | * d | payload | * e | payload | * f | payload | * +-------+-------+-------+-------+-------+-------+-------+-------+ * * Legend: * * payload contains the embedded data * B (byteorder) byteorder (endianness) * D (dedup) padding (set to zero) * X encryption (set to zero; see above) * E (embedded) set to one * lvl indirection level * type DMU object type * etype how to interpret embedded data (BP_EMBEDDED_TYPE_*) * comp compression function of payload * PSIZE size of payload after compression, in bytes * LSIZE logical size of payload, in bytes * note that 25 bits is enough to store the largest * "normal" BP's LSIZE (2^16 * 2^9) in bytes * log. birth transaction group in which the block was logically born * * Note that LSIZE and PSIZE are stored in bytes, whereas for non-embedded * bp's they are stored in units of SPA_MINBLOCKSHIFT. * Generally, the generic BP_GET_*() macros can be used on embedded BP's. * The B, D, X, lvl, type, and comp fields are stored the same as with normal * BP's so the BP_SET_* macros can be used with them. etype, PSIZE, LSIZE must * be set with the BPE_SET_* macros. BP_SET_EMBEDDED() should be called before * other macros, as they assert that they are only used on BP's of the correct * "embedded-ness". */ #define BPE_GET_ETYPE(bp) \ (ASSERT(BP_IS_EMBEDDED(bp)), \ BF64_GET((bp)->blk_prop, 40, 8)) #define BPE_SET_ETYPE(bp, t) do { \ ASSERT(BP_IS_EMBEDDED(bp)); \ BF64_SET((bp)->blk_prop, 40, 8, t); \ _NOTE(CONSTCOND) } while (0) #define BPE_GET_LSIZE(bp) \ (ASSERT(BP_IS_EMBEDDED(bp)), \ BF64_GET_SB((bp)->blk_prop, 0, 25, 0, 1)) #define BPE_SET_LSIZE(bp, x) do { \ ASSERT(BP_IS_EMBEDDED(bp)); \ BF64_SET_SB((bp)->blk_prop, 0, 25, 0, 1, x); \ _NOTE(CONSTCOND) } while (0) #define BPE_GET_PSIZE(bp) \ (ASSERT(BP_IS_EMBEDDED(bp)), \ BF64_GET_SB((bp)->blk_prop, 25, 7, 0, 1)) #define BPE_SET_PSIZE(bp, x) do { \ ASSERT(BP_IS_EMBEDDED(bp)); \ BF64_SET_SB((bp)->blk_prop, 25, 7, 0, 1, x); \ _NOTE(CONSTCOND) } while (0) typedef enum bp_embedded_type { BP_EMBEDDED_TYPE_DATA, BP_EMBEDDED_TYPE_RESERVED, /* Reserved for an unintegrated feature. */ NUM_BP_EMBEDDED_TYPES = BP_EMBEDDED_TYPE_RESERVED } bp_embedded_type_t; #define BPE_NUM_WORDS 14 #define BPE_PAYLOAD_SIZE (BPE_NUM_WORDS * sizeof (uint64_t)) #define BPE_IS_PAYLOADWORD(bp, wp) \ ((wp) != &(bp)->blk_prop && (wp) != &(bp)->blk_birth) #define SPA_BLKPTRSHIFT 7 /* blkptr_t is 128 bytes */ #define SPA_DVAS_PER_BP 3 /* Number of DVAs in a bp */ /* * A block is a hole when it has either 1) never been written to, or * 2) is zero-filled. In both cases, ZFS can return all zeroes for all reads * without physically allocating disk space. Holes are represented in the * blkptr_t structure by zeroed blk_dva. Correct checking for holes is * done through the BP_IS_HOLE macro. For holes, the logical size, level, * DMU object type, and birth times are all also stored for holes that * were written to at some point (i.e. were punched after having been filled). */ typedef struct blkptr { dva_t blk_dva[SPA_DVAS_PER_BP]; /* Data Virtual Addresses */ uint64_t blk_prop; /* size, compression, type, etc */ uint64_t blk_pad[2]; /* Extra space for the future */ uint64_t blk_phys_birth; /* txg when block was allocated */ uint64_t blk_birth; /* transaction group at birth */ uint64_t blk_fill; /* fill count */ zio_cksum_t blk_cksum; /* 256-bit checksum */ } blkptr_t; /* * Macros to get and set fields in a bp or DVA. */ #define DVA_GET_ASIZE(dva) \ BF64_GET_SB((dva)->dva_word[0], 0, SPA_ASIZEBITS, SPA_MINBLOCKSHIFT, 0) #define DVA_SET_ASIZE(dva, x) \ BF64_SET_SB((dva)->dva_word[0], 0, SPA_ASIZEBITS, \ SPA_MINBLOCKSHIFT, 0, x) #define DVA_GET_GRID(dva) BF64_GET((dva)->dva_word[0], 24, 8) #define DVA_SET_GRID(dva, x) BF64_SET((dva)->dva_word[0], 24, 8, x) #define DVA_GET_VDEV(dva) BF64_GET((dva)->dva_word[0], 32, 32) #define DVA_SET_VDEV(dva, x) BF64_SET((dva)->dva_word[0], 32, 32, x) #define DVA_GET_OFFSET(dva) \ BF64_GET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0) #define DVA_SET_OFFSET(dva, x) \ BF64_SET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0, x) #define DVA_GET_GANG(dva) BF64_GET((dva)->dva_word[1], 63, 1) #define DVA_SET_GANG(dva, x) BF64_SET((dva)->dva_word[1], 63, 1, x) #define BP_GET_LSIZE(bp) \ (BP_IS_EMBEDDED(bp) ? \ (BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA ? BPE_GET_LSIZE(bp) : 0): \ BF64_GET_SB((bp)->blk_prop, 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1)) #define BP_SET_LSIZE(bp, x) do { \ ASSERT(!BP_IS_EMBEDDED(bp)); \ BF64_SET_SB((bp)->blk_prop, \ 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \ _NOTE(CONSTCOND) } while (0) #define BP_GET_PSIZE(bp) \ (BP_IS_EMBEDDED(bp) ? 0 : \ BF64_GET_SB((bp)->blk_prop, 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1)) #define BP_SET_PSIZE(bp, x) do { \ ASSERT(!BP_IS_EMBEDDED(bp)); \ BF64_SET_SB((bp)->blk_prop, \ 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \ _NOTE(CONSTCOND) } while (0) #define BP_GET_COMPRESS(bp) \ BF64_GET((bp)->blk_prop, 32, SPA_COMPRESSBITS) #define BP_SET_COMPRESS(bp, x) \ BF64_SET((bp)->blk_prop, 32, SPA_COMPRESSBITS, x) #define BP_IS_EMBEDDED(bp) BF64_GET((bp)->blk_prop, 39, 1) #define BP_SET_EMBEDDED(bp, x) BF64_SET((bp)->blk_prop, 39, 1, x) #define BP_GET_CHECKSUM(bp) \ (BP_IS_EMBEDDED(bp) ? ZIO_CHECKSUM_OFF : \ BF64_GET((bp)->blk_prop, 40, 8)) #define BP_SET_CHECKSUM(bp, x) do { \ ASSERT(!BP_IS_EMBEDDED(bp)); \ BF64_SET((bp)->blk_prop, 40, 8, x); \ _NOTE(CONSTCOND) } while (0) #define BP_GET_TYPE(bp) BF64_GET((bp)->blk_prop, 48, 8) #define BP_SET_TYPE(bp, x) BF64_SET((bp)->blk_prop, 48, 8, x) #define BP_GET_LEVEL(bp) BF64_GET((bp)->blk_prop, 56, 5) #define BP_SET_LEVEL(bp, x) BF64_SET((bp)->blk_prop, 56, 5, x) #define BP_GET_DEDUP(bp) BF64_GET((bp)->blk_prop, 62, 1) #define BP_SET_DEDUP(bp, x) BF64_SET((bp)->blk_prop, 62, 1, x) #define BP_GET_BYTEORDER(bp) BF64_GET((bp)->blk_prop, 63, 1) #define BP_SET_BYTEORDER(bp, x) BF64_SET((bp)->blk_prop, 63, 1, x) #define BP_PHYSICAL_BIRTH(bp) \ (BP_IS_EMBEDDED(bp) ? 0 : \ (bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth) #define BP_SET_BIRTH(bp, logical, physical) \ { \ ASSERT(!BP_IS_EMBEDDED(bp)); \ (bp)->blk_birth = (logical); \ (bp)->blk_phys_birth = ((logical) == (physical) ? 0 : (physical)); \ } #define BP_GET_FILL(bp) (BP_IS_EMBEDDED(bp) ? 1 : (bp)->blk_fill) #define BP_GET_ASIZE(bp) \ (BP_IS_EMBEDDED(bp) ? 0 : \ DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \ DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \ DVA_GET_ASIZE(&(bp)->blk_dva[2])) #define BP_GET_UCSIZE(bp) \ ((BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp))) ? \ BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp)) #define BP_GET_NDVAS(bp) \ (BP_IS_EMBEDDED(bp) ? 0 : \ !!DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \ !!DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \ !!DVA_GET_ASIZE(&(bp)->blk_dva[2])) #define BP_COUNT_GANG(bp) \ (BP_IS_EMBEDDED(bp) ? 0 : \ (DVA_GET_GANG(&(bp)->blk_dva[0]) + \ DVA_GET_GANG(&(bp)->blk_dva[1]) + \ DVA_GET_GANG(&(bp)->blk_dva[2]))) #define DVA_EQUAL(dva1, dva2) \ ((dva1)->dva_word[1] == (dva2)->dva_word[1] && \ (dva1)->dva_word[0] == (dva2)->dva_word[0]) #define BP_EQUAL(bp1, bp2) \ (BP_PHYSICAL_BIRTH(bp1) == BP_PHYSICAL_BIRTH(bp2) && \ (bp1)->blk_birth == (bp2)->blk_birth && \ DVA_EQUAL(&(bp1)->blk_dva[0], &(bp2)->blk_dva[0]) && \ DVA_EQUAL(&(bp1)->blk_dva[1], &(bp2)->blk_dva[1]) && \ DVA_EQUAL(&(bp1)->blk_dva[2], &(bp2)->blk_dva[2])) #define ZIO_CHECKSUM_EQUAL(zc1, zc2) \ (0 == (((zc1).zc_word[0] - (zc2).zc_word[0]) | \ ((zc1).zc_word[1] - (zc2).zc_word[1]) | \ ((zc1).zc_word[2] - (zc2).zc_word[2]) | \ ((zc1).zc_word[3] - (zc2).zc_word[3]))) #define ZIO_CHECKSUM_IS_ZERO(zc) \ (0 == ((zc)->zc_word[0] | (zc)->zc_word[1] | \ (zc)->zc_word[2] | (zc)->zc_word[3])) #define ZIO_CHECKSUM_BSWAP(zcp) \ { \ (zcp)->zc_word[0] = BSWAP_64((zcp)->zc_word[0]); \ (zcp)->zc_word[1] = BSWAP_64((zcp)->zc_word[1]); \ (zcp)->zc_word[2] = BSWAP_64((zcp)->zc_word[2]); \ (zcp)->zc_word[3] = BSWAP_64((zcp)->zc_word[3]); \ } #define DVA_IS_VALID(dva) (DVA_GET_ASIZE(dva) != 0) #define ZIO_SET_CHECKSUM(zcp, w0, w1, w2, w3) \ { \ (zcp)->zc_word[0] = w0; \ (zcp)->zc_word[1] = w1; \ (zcp)->zc_word[2] = w2; \ (zcp)->zc_word[3] = w3; \ } #define BP_IDENTITY(bp) (ASSERT(!BP_IS_EMBEDDED(bp)), &(bp)->blk_dva[0]) #define BP_IS_GANG(bp) \ (BP_IS_EMBEDDED(bp) ? B_FALSE : DVA_GET_GANG(BP_IDENTITY(bp))) #define DVA_IS_EMPTY(dva) ((dva)->dva_word[0] == 0ULL && \ (dva)->dva_word[1] == 0ULL) #define BP_IS_HOLE(bp) \ (!BP_IS_EMBEDDED(bp) && DVA_IS_EMPTY(BP_IDENTITY(bp))) /* BP_IS_RAIDZ(bp) assumes no block compression */ #define BP_IS_RAIDZ(bp) (DVA_GET_ASIZE(&(bp)->blk_dva[0]) > \ BP_GET_PSIZE(bp)) #define BP_ZERO(bp) \ { \ (bp)->blk_dva[0].dva_word[0] = 0; \ (bp)->blk_dva[0].dva_word[1] = 0; \ (bp)->blk_dva[1].dva_word[0] = 0; \ (bp)->blk_dva[1].dva_word[1] = 0; \ (bp)->blk_dva[2].dva_word[0] = 0; \ (bp)->blk_dva[2].dva_word[1] = 0; \ (bp)->blk_prop = 0; \ (bp)->blk_pad[0] = 0; \ (bp)->blk_pad[1] = 0; \ (bp)->blk_phys_birth = 0; \ (bp)->blk_birth = 0; \ (bp)->blk_fill = 0; \ ZIO_SET_CHECKSUM(&(bp)->blk_cksum, 0, 0, 0, 0); \ } #if BYTE_ORDER == _BIG_ENDIAN #define ZFS_HOST_BYTEORDER (0ULL) #else #define ZFS_HOST_BYTEORDER (1ULL) #endif #define BP_SHOULD_BYTESWAP(bp) (BP_GET_BYTEORDER(bp) != ZFS_HOST_BYTEORDER) #define BP_SPRINTF_LEN 320 /* * This macro allows code sharing between zfs, libzpool, and mdb. * 'func' is either snprintf() or mdb_snprintf(). * 'ws' (whitespace) can be ' ' for single-line format, '\n' for multi-line. */ #define SNPRINTF_BLKPTR(func, ws, buf, size, bp, type, checksum, compress) \ { \ static const char *copyname[] = \ { "zero", "single", "double", "triple" }; \ int len = 0; \ int copies = 0; \ \ if (bp == NULL) { \ len += func(buf + len, size - len, ""); \ } else if (BP_IS_HOLE(bp)) { \ len += func(buf + len, size - len, \ "HOLE [L%llu %s] " \ "size=%llxL birth=%lluL", \ (u_longlong_t)BP_GET_LEVEL(bp), \ type, \ (u_longlong_t)BP_GET_LSIZE(bp), \ (u_longlong_t)bp->blk_birth); \ } else if (BP_IS_EMBEDDED(bp)) { \ len = func(buf + len, size - len, \ "EMBEDDED [L%llu %s] et=%u %s " \ "size=%llxL/%llxP birth=%lluL", \ (u_longlong_t)BP_GET_LEVEL(bp), \ type, \ (int)BPE_GET_ETYPE(bp), \ compress, \ (u_longlong_t)BPE_GET_LSIZE(bp), \ (u_longlong_t)BPE_GET_PSIZE(bp), \ (u_longlong_t)bp->blk_birth); \ } else { \ for (int d = 0; d < BP_GET_NDVAS(bp); d++) { \ const dva_t *dva = &bp->blk_dva[d]; \ if (DVA_IS_VALID(dva)) \ copies++; \ len += func(buf + len, size - len, \ "DVA[%d]=<%llu:%llx:%llx>%c", d, \ (u_longlong_t)DVA_GET_VDEV(dva), \ (u_longlong_t)DVA_GET_OFFSET(dva), \ (u_longlong_t)DVA_GET_ASIZE(dva), \ ws); \ } \ if (BP_IS_GANG(bp) && \ DVA_GET_ASIZE(&bp->blk_dva[2]) <= \ DVA_GET_ASIZE(&bp->blk_dva[1]) / 2) \ copies--; \ len += func(buf + len, size - len, \ "[L%llu %s] %s %s %s %s %s %s%c" \ "size=%llxL/%llxP birth=%lluL/%lluP fill=%llu%c" \ "cksum=%llx:%llx:%llx:%llx", \ (u_longlong_t)BP_GET_LEVEL(bp), \ type, \ checksum, \ compress, \ BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE", \ BP_IS_GANG(bp) ? "gang" : "contiguous", \ BP_GET_DEDUP(bp) ? "dedup" : "unique", \ copyname[copies], \ ws, \ (u_longlong_t)BP_GET_LSIZE(bp), \ (u_longlong_t)BP_GET_PSIZE(bp), \ (u_longlong_t)bp->blk_birth, \ (u_longlong_t)BP_PHYSICAL_BIRTH(bp), \ (u_longlong_t)BP_GET_FILL(bp), \ ws, \ (u_longlong_t)bp->blk_cksum.zc_word[0], \ (u_longlong_t)bp->blk_cksum.zc_word[1], \ (u_longlong_t)bp->blk_cksum.zc_word[2], \ (u_longlong_t)bp->blk_cksum.zc_word[3]); \ } \ ASSERT(len < size); \ } #define BP_GET_BUFC_TYPE(bp) \ (((BP_GET_LEVEL(bp) > 0) || (DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))) ? \ ARC_BUFC_METADATA : ARC_BUFC_DATA) typedef enum spa_import_type { SPA_IMPORT_EXISTING, SPA_IMPORT_ASSEMBLE } spa_import_type_t; /* state manipulation functions */ extern int spa_open(const char *pool, spa_t **, void *tag); extern int spa_open_rewind(const char *pool, spa_t **, void *tag, nvlist_t *policy, nvlist_t **config); extern int spa_get_stats(const char *pool, nvlist_t **config, char *altroot, size_t buflen); extern int spa_create(const char *pool, nvlist_t *config, nvlist_t *props, nvlist_t *zplprops); #ifdef illumos extern int spa_import_rootpool(char *devpath, char *devid); #else extern int spa_import_rootpool(const char *name); #endif extern int spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags); extern nvlist_t *spa_tryimport(nvlist_t *tryconfig); extern int spa_destroy(char *pool); extern int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, boolean_t hardforce); extern int spa_reset(char *pool); extern void spa_async_request(spa_t *spa, int flag); extern void spa_async_unrequest(spa_t *spa, int flag); extern void spa_async_suspend(spa_t *spa); extern void spa_async_resume(spa_t *spa); extern spa_t *spa_inject_addref(char *pool); extern void spa_inject_delref(spa_t *spa); extern void spa_scan_stat_init(spa_t *spa); extern int spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps); #define SPA_ASYNC_CONFIG_UPDATE 0x01 #define SPA_ASYNC_REMOVE 0x02 #define SPA_ASYNC_PROBE 0x04 #define SPA_ASYNC_RESILVER_DONE 0x08 #define SPA_ASYNC_RESILVER 0x10 #define SPA_ASYNC_AUTOEXPAND 0x20 #define SPA_ASYNC_REMOVE_DONE 0x40 #define SPA_ASYNC_REMOVE_STOP 0x80 /* * Controls the behavior of spa_vdev_remove(). */ #define SPA_REMOVE_UNSPARE 0x01 #define SPA_REMOVE_DONE 0x02 /* device manipulation */ extern int spa_vdev_add(spa_t *spa, nvlist_t *nvroot); extern int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing); extern int spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done); extern int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare); extern boolean_t spa_vdev_remove_active(spa_t *spa); extern int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath); extern int spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru); extern int spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, nvlist_t *props, boolean_t exp); /* spare state (which is global across all pools) */ extern void spa_spare_add(vdev_t *vd); extern void spa_spare_remove(vdev_t *vd); extern boolean_t spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt); extern void spa_spare_activate(vdev_t *vd); /* L2ARC state (which is global across all pools) */ extern void spa_l2cache_add(vdev_t *vd); extern void spa_l2cache_remove(vdev_t *vd); extern boolean_t spa_l2cache_exists(uint64_t guid, uint64_t *pool); extern void spa_l2cache_activate(vdev_t *vd); extern void spa_l2cache_drop(spa_t *spa); /* scanning */ extern int spa_scan(spa_t *spa, pool_scan_func_t func); extern int spa_scan_stop(spa_t *spa); /* spa syncing */ extern void spa_sync(spa_t *spa, uint64_t txg); /* only for DMU use */ extern void spa_sync_allpools(void); /* spa namespace global mutex */ extern kmutex_t spa_namespace_lock; /* * SPA configuration functions in spa_config.c */ #define SPA_CONFIG_UPDATE_POOL 0 #define SPA_CONFIG_UPDATE_VDEVS 1 extern void spa_config_sync(spa_t *, boolean_t, boolean_t); extern void spa_config_load(void); extern nvlist_t *spa_all_configs(uint64_t *); extern void spa_config_set(spa_t *spa, nvlist_t *config); extern nvlist_t *spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats); extern void spa_config_update(spa_t *spa, int what); /* * Miscellaneous SPA routines in spa_misc.c */ /* Namespace manipulation */ extern spa_t *spa_lookup(const char *name); extern spa_t *spa_add(const char *name, nvlist_t *config, const char *altroot); extern void spa_remove(spa_t *spa); extern spa_t *spa_next(spa_t *prev); /* Refcount functions */ extern void spa_open_ref(spa_t *spa, void *tag); extern void spa_close(spa_t *spa, void *tag); extern void spa_async_close(spa_t *spa, void *tag); extern boolean_t spa_refcount_zero(spa_t *spa); #define SCL_NONE 0x00 #define SCL_CONFIG 0x01 #define SCL_STATE 0x02 #define SCL_L2ARC 0x04 /* hack until L2ARC 2.0 */ #define SCL_ALLOC 0x08 #define SCL_ZIO 0x10 #define SCL_FREE 0x20 #define SCL_VDEV 0x40 #define SCL_LOCKS 7 #define SCL_ALL ((1 << SCL_LOCKS) - 1) #define SCL_STATE_ALL (SCL_STATE | SCL_L2ARC | SCL_ZIO) /* Pool configuration locks */ extern int spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw); extern void spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw); extern void spa_config_exit(spa_t *spa, int locks, void *tag); extern int spa_config_held(spa_t *spa, int locks, krw_t rw); /* Pool vdev add/remove lock */ extern uint64_t spa_vdev_enter(spa_t *spa); extern uint64_t spa_vdev_config_enter(spa_t *spa); extern void spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag); extern int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error); /* Pool vdev state change lock */ extern void spa_vdev_state_enter(spa_t *spa, int oplock); extern int spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error); /* Log state */ typedef enum spa_log_state { SPA_LOG_UNKNOWN = 0, /* unknown log state */ SPA_LOG_MISSING, /* missing log(s) */ SPA_LOG_CLEAR, /* clear the log(s) */ SPA_LOG_GOOD, /* log(s) are good */ } spa_log_state_t; extern spa_log_state_t spa_get_log_state(spa_t *spa); extern void spa_set_log_state(spa_t *spa, spa_log_state_t state); extern int spa_offline_log(spa_t *spa); /* Log claim callback */ extern void spa_claim_notify(zio_t *zio); /* Accessor functions */ extern boolean_t spa_shutting_down(spa_t *spa); extern struct dsl_pool *spa_get_dsl(spa_t *spa); extern boolean_t spa_is_initializing(spa_t *spa); extern blkptr_t *spa_get_rootblkptr(spa_t *spa); extern void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp); extern void spa_altroot(spa_t *, char *, size_t); extern int spa_sync_pass(spa_t *spa); extern char *spa_name(spa_t *spa); extern uint64_t spa_guid(spa_t *spa); extern uint64_t spa_load_guid(spa_t *spa); extern uint64_t spa_last_synced_txg(spa_t *spa); extern uint64_t spa_first_txg(spa_t *spa); extern uint64_t spa_syncing_txg(spa_t *spa); +extern uint64_t spa_final_dirty_txg(spa_t *spa); extern uint64_t spa_version(spa_t *spa); extern pool_state_t spa_state(spa_t *spa); extern spa_load_state_t spa_load_state(spa_t *spa); extern uint64_t spa_freeze_txg(spa_t *spa); extern uint64_t spa_get_worst_case_asize(spa_t *spa, uint64_t lsize); extern uint64_t spa_get_dspace(spa_t *spa); extern uint64_t spa_get_slop_space(spa_t *spa); extern void spa_update_dspace(spa_t *spa); extern uint64_t spa_version(spa_t *spa); extern boolean_t spa_deflate(spa_t *spa); extern metaslab_class_t *spa_normal_class(spa_t *spa); extern metaslab_class_t *spa_log_class(spa_t *spa); extern void spa_evicting_os_register(spa_t *, objset_t *os); extern void spa_evicting_os_deregister(spa_t *, objset_t *os); extern void spa_evicting_os_wait(spa_t *spa); extern int spa_max_replication(spa_t *spa); extern int spa_prev_software_version(spa_t *spa); extern int spa_busy(void); extern uint8_t spa_get_failmode(spa_t *spa); extern boolean_t spa_suspended(spa_t *spa); extern uint64_t spa_bootfs(spa_t *spa); extern uint64_t spa_delegation(spa_t *spa); extern objset_t *spa_meta_objset(spa_t *spa); extern uint64_t spa_deadman_synctime(spa_t *spa); /* Miscellaneous support routines */ extern void spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx); extern void spa_deactivate_mos_feature(spa_t *spa, const char *feature); extern int spa_rename(const char *oldname, const char *newname); extern spa_t *spa_by_guid(uint64_t pool_guid, uint64_t device_guid); extern boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid); extern char *spa_strdup(const char *); extern void spa_strfree(char *); extern uint64_t spa_get_random(uint64_t range); extern uint64_t spa_generate_guid(spa_t *spa); extern void snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp); extern void spa_freeze(spa_t *spa); extern int spa_change_guid(spa_t *spa); extern void spa_upgrade(spa_t *spa, uint64_t version); extern void spa_evict_all(void); extern vdev_t *spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t l2cache); extern boolean_t spa_has_spare(spa_t *, uint64_t guid); extern uint64_t dva_get_dsize_sync(spa_t *spa, const dva_t *dva); extern uint64_t bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp); extern uint64_t bp_get_dsize(spa_t *spa, const blkptr_t *bp); extern boolean_t spa_has_slogs(spa_t *spa); extern boolean_t spa_is_root(spa_t *spa); extern boolean_t spa_writeable(spa_t *spa); extern boolean_t spa_has_pending_synctask(spa_t *spa); extern int spa_maxblocksize(spa_t *spa); extern void zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp); extern int spa_mode(spa_t *spa); extern uint64_t zfs_strtonum(const char *str, char **nptr); #define strtonum(str, nptr) zfs_strtonum((str), (nptr)) extern char *spa_his_ievent_table[]; extern void spa_history_create_obj(spa_t *spa, dmu_tx_t *tx); extern int spa_history_get(spa_t *spa, uint64_t *offset, uint64_t *len_read, char *his_buf); extern int spa_history_log(spa_t *spa, const char *his_buf); extern int spa_history_log_nvl(spa_t *spa, nvlist_t *nvl); extern void spa_history_log_version(spa_t *spa, const char *operation); extern void spa_history_log_internal(spa_t *spa, const char *operation, dmu_tx_t *tx, const char *fmt, ...); extern void spa_history_log_internal_ds(struct dsl_dataset *ds, const char *op, dmu_tx_t *tx, const char *fmt, ...); extern void spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation, dmu_tx_t *tx, const char *fmt, ...); /* error handling */ struct zbookmark_phys; extern void spa_log_error(spa_t *spa, zio_t *zio); extern void zfs_ereport_post(const char *cls, spa_t *spa, vdev_t *vd, zio_t *zio, uint64_t stateoroffset, uint64_t length); extern void zfs_post_remove(spa_t *spa, vdev_t *vd); extern void zfs_post_state_change(spa_t *spa, vdev_t *vd); extern void zfs_post_autoreplace(spa_t *spa, vdev_t *vd); extern uint64_t spa_get_errlog_size(spa_t *spa); extern int spa_get_errlog(spa_t *spa, void *uaddr, size_t *count); extern void spa_errlog_rotate(spa_t *spa); extern void spa_errlog_drain(spa_t *spa); extern void spa_errlog_sync(spa_t *spa, uint64_t txg); extern void spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub); /* vdev cache */ extern void vdev_cache_stat_init(void); extern void vdev_cache_stat_fini(void); /* Initialization and termination */ extern void spa_init(int flags); extern void spa_fini(void); extern void spa_boot_init(); /* properties */ extern int spa_prop_set(spa_t *spa, nvlist_t *nvp); extern int spa_prop_get(spa_t *spa, nvlist_t **nvp); extern void spa_prop_clear_bootfs(spa_t *spa, uint64_t obj, dmu_tx_t *tx); extern void spa_configfile_set(spa_t *, nvlist_t *, boolean_t); /* asynchronous event notification */ extern void spa_event_notify(spa_t *spa, vdev_t *vdev, const char *name); #ifdef ZFS_DEBUG #define dprintf_bp(bp, fmt, ...) do { \ if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, (bp)); \ dprintf(fmt " %s\n", __VA_ARGS__, __blkbuf); \ kmem_free(__blkbuf, BP_SPRINTF_LEN); \ } \ _NOTE(CONSTCOND) } while (0) #else #define dprintf_bp(bp, fmt, ...) #endif extern boolean_t spa_debug_enabled(spa_t *spa); #define spa_dbgmsg(spa, ...) \ { \ if (spa_debug_enabled(spa)) \ zfs_dbgmsg(__VA_ARGS__); \ } extern int spa_mode_global; /* mode, e.g. FREAD | FWRITE */ #ifdef __cplusplus } #endif #endif /* _SYS_SPA_H */ Index: stable/11 =================================================================== --- stable/11 (revision 321553) +++ stable/11 (revision 321554) Property changes on: stable/11 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r318829