fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER START
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The contents of this file are subject to the terms of the
f65e61c04bc28ffd6bda04619c84330b420450b5ahrens * Common Development and Distribution License (the "License").
f65e61c04bc28ffd6bda04619c84330b420450b5ahrens * You may not use this file except in compliance with the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
fa9e4066f08beec538e775443c5be79dd423fcabahrens * See the License for the specific language governing permissions
fa9e4066f08beec538e775443c5be79dd423fcabahrens * and limitations under the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * When distributing Covered Code, include this CDDL HEADER in each
fa9e4066f08beec538e775443c5be79dd423fcabahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If applicable, add the following below this CDDL HEADER, with the
fa9e4066f08beec538e775443c5be79dd423fcabahrens * fields enclosed by brackets "[]" replaced with your own identifying
fa9e4066f08beec538e775443c5be79dd423fcabahrens * information: Portions Copyright [yyyy] [name of copyright owner]
fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER END
01025c89909911d8869b3c46753a2a76b9a2de8bJohn Harres * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
9dccfd2a04cd1645e2616b7307b3a88041aba991Albert Lee * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
79d728325e257b05859d2eef4a4dfbefdce05a76Matthew Ahrens * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
c3d26abc9ee97b4f60233556aadeb57e0bd30bb9Matthew Ahrens * Copyright (c) 2014 Integros [integros.com]
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/dsl_dataset.h> /* for dsl_dataset_block_freeable() */
fa9e4066f08beec538e775443c5be79dd423fcabahrens#include <sys/dsl_dir.h> /* for dsl_dir_tempreserve_*() */
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens#include <sys/zap_impl.h> /* for fzap_default_block_shift */
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrocktypedef void (*dmu_tx_hold_func_t)(dmu_tx_t *tx, struct dnode *dn,
fa9e4066f08beec538e775443c5be79dd423fcabahrens dmu_tx_t *tx = kmem_zalloc(sizeof (dmu_tx_t), KM_SLEEP);
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia list_create(&tx->tx_callbacks, sizeof (dmu_tx_callback_t),
503ad85c168c7992ccc310af845a581cff3c72b5Matthew Ahrens dmu_tx_t *tx = dmu_tx_create_dd(os->os_dsl_dataset->ds_dir);
503ad85c168c7992ccc310af845a581cff3c72b5Matthew Ahrens tx->tx_lastsnap_txg = dsl_dataset_prev_snap_txg(os->os_dsl_dataset);
fa9e4066f08beec538e775443c5be79dd423fcabahrensdmu_tx_create_assigned(struct dsl_pool *dp, uint64_t txg)
fa9e4066f08beec538e775443c5be79dd423fcabahrensdmu_tx_hold_object_impl(dmu_tx_t *tx, objset_t *os, uint64_t object,
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens enum dmu_tx_hold_type type, uint64_t arg1, uint64_t arg2)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * dn->dn_assigned_txg == tx->tx_txg doesn't pose a
fa9e4066f08beec538e775443c5be79dd423fcabahrens * problem, but there's no way for it to happen (for
fa9e4066f08beec538e775443c5be79dd423fcabahrens * now, at least).
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_create(&txh->txh_space_tooverwrite);
fa9e4066f08beec538e775443c5be79dd423fcabahrensdmu_tx_add_new_object(dmu_tx_t *tx, objset_t *os, uint64_t object)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If we're syncing, they can manipulate any object anyhow, and
fa9e4066f08beec538e775443c5be79dd423fcabahrens * the hold on the dnode_t can cause problems.
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockdmu_tx_check_ioerr(zio_t *zio, dnode_t *dn, int level, uint64_t blkid)
1ab7f2ded02e7a1bc3c73516eb27efa79bf2a2ffmaybee err = dbuf_read(db, zio, DB_RF_CANFAIL | DB_RF_NOPREFETCH);
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwickdmu_tx_count_twig(dmu_tx_hold_t *txh, dnode_t *dn, dmu_buf_impl_t *db,
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick int level, uint64_t blkid, boolean_t freeable, uint64_t *history)
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick if (level >= dn->dn_nlevels || history[level] == blkid)
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick space = (level == 0) ? dn->dn_datablksz : (1ULL << dn->dn_indblkshift);
c7cd242109c82107ec2e50013369e92be9d77702George Wilson dsl_dataset_block_freeable(ds, bp, bp->blk_birth)));
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tooverwrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tounref,
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* ARGSUSED */
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrensdmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len)
b515258426fed6c7311fd3f1dea697cfbd4085c6Matthew Ahrens max_bs = highbit64(txh->txh_tx->tx_objset->os_recordsize) - 1;
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * For i/o error checking, read the first and last level-0
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * blocks (if they are not aligned), and all the level-1 blocks.
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee if (start == 0 && (off > 0 || len < dn->dn_datablksz)) {
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock /* first level-0 block */
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock /* last level-0 block */
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock /* level-1 blocks */
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee int shft = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * The blocksize can't change,
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * so we can make a more precise estimate.
b515258426fed6c7311fd3f1dea697cfbd4085c6Matthew Ahrens * The blocksize can increase up to the recordsize,
b515258426fed6c7311fd3f1dea697cfbd4085c6Matthew Ahrens * or if it is already more than the recordsize,
b515258426fed6c7311fd3f1dea697cfbd4085c6Matthew Ahrens * up to the next power of 2.
b515258426fed6c7311fd3f1dea697cfbd4085c6Matthew Ahrens max_bs = MAX(max_bs, highbit64(dn->dn_datablksz - 1));
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * If this write is not off the end of the file
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * we need to account for overwrites/unref.
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick for (int l = 0; l < DN_MAX_LEVELS; l++)
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick dmu_tx_count_twig(txh, dn, db, 0, start, B_FALSE,
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * Account for new indirects appearing
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * before this IO gets assigned into a txg.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 'end' is the last thing we will access, not one past.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * This way we won't overflow when accessing the last byte.
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The object contains at most 2^(64 - min_bs) blocks,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * and each indirect level maps 2^epbs.
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * We also need a new blkid=0 indirect block
4a7f2a75ca428b8c3910f76ab6cea0dfadba8914Mark Maybee * to reference any existing file data.
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens if (refcount_count(&txh->txh_space_towrite) +
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_count(&txh->txh_space_tooverwrite) >
744947dc83c634d985ed3ad79ac9c5e28d1865fdTom Erickson dnode_t *mdn = DMU_META_DNODE(txh->txh_tx->tx_objset);
fa9e4066f08beec538e775443c5be79dd423fcabahrens dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset,
c7cd242109c82107ec2e50013369e92be9d77702George Wilson dn->dn_dbuf->db_blkptr, dn->dn_dbuf->db_blkptr->blk_birth)) {
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tooverwrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tounref, space, FTAG);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite, space, FTAG);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tounref,
fa9e4066f08beec538e775443c5be79dd423fcabahrensdmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len)
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrensdmu_tx_count_free(dmu_tx_hold_t *txh, uint64_t off, uint64_t len)
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee * The struct_rwlock protects us against dn_nlevels
c543ec060d1359f6c8a9507242521f344a2ac3efahrens * changing, in case (against all odds) we manage to dirty &
c543ec060d1359f6c8a9507242521f344a2ac3efahrens * sync out the changes after we check for being dirty.
01025c89909911d8869b3c46753a2a76b9a2de8bJohn Harres * Also, dbuf_hold_impl() wants us to have the struct_rwlock.
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee nblks = (len + dn->dn_datablksz - 1) >> dn->dn_datablkshift;
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen l0span = nblks; /* save for later use to calc level > 1 overhead */
fa9e4066f08beec538e775443c5be79dd423fcabahrens for (i = 0; i < nblks; i++) {
c7cd242109c82107ec2e50013369e92be9d77702George Wilson if (dsl_dataset_block_freeable(ds, bp, bp->blk_birth)) {
b7e50089acb2c25bc533ab07f5326ff48cfe3621maybee skipped += (new_blkid >> epbs) - (blkid >> epbs) - 1;
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_memory_tohold,
77179d12b73710da6830efdd32a38376a0af10beLori Alt * We don't check memory_tohold against DMU_MAX_ACCESS because
77179d12b73710da6830efdd32a38376a0af10beLori Alt * memory_tohold is an over-estimation (especially the >L1
77179d12b73710da6830efdd32a38376a0af10beLori Alt * indirect blocks), so it could fail. Callers should have
77179d12b73710da6830efdd32a38376a0af10beLori Alt * already verified that they will not be holding too much
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee err = dbuf_read(dbuf, NULL, DB_RF_HAVESTRUCT | DB_RF_CANFAIL);
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee if (err != 0) {
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee for (i = 0; i < tochk; i++) {
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen * Add in memory requirements of higher-level indirects.
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen * This assumes a worst-possible scenario for dn_nlevels and a
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen * worst-possible distribution of l1-blocks over the region to free.
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen uint64_t blkcnt = 1 + ((l0span >> epbs) >> epbs);
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen * Here we don't use DN_MAX_LEVEL, but calculate it with the
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen * given datablkshift and indblkshift. This makes the
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen * difference between 19 and 8 on large files.
31495a1e56860f4575614774a592fe33fc9c71f2Arne Jansen int maxlevel = 2 + (DN_MAX_OFFSET_SHIFT - dn->dn_datablkshift) /
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_memory_tohold,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens MAX(MIN(blkcnt, nl1blks), 1) << dn->dn_indblkshift,
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee /* account for new level 1 indirect blocks that might show up */
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee skipped = MIN(skipped, DMU_MAX_DELETEBLKCNT >> epbs);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_memory_tohold,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tofree, space, FTAG);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tounref, unref, FTAG);
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * This function marks the transaction as being a "net free". The end
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * result is that refquotas will be disabled for this transaction, and
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * this transaction will be able to use half of the pool space overhead
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * (see dsl_pool_adjustedsize()). Therefore this function should only
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * be called for transactions that we expect will not cause a net increase
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * in the amount of space used (but it's OK if that is occasionally not true).
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * Pretend that this operation will free 1GB of space. This
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * should be large enough to cancel out the largest write.
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * We don't want to use something like UINT64_MAX, because that would
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * cause overflows when doing math with these values (e.g. in
4bb73804952172060c9efb163b89c17f56804fe8Matthew Ahrens * dmu_tx_try_assign()).
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tofree,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tounref,
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrensdmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, uint64_t len)
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens * For i/o error checking, we read the first and last level-0
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens * blocks if they are not aligned, and all the level-1 blocks.
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens * Note: dbuf_free_range() assumes that we have not instantiated
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens * any level-0 dbufs that will be completely freed. Therefore we must
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens * exercise care to not read or count the first and last blocks
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens * if they are blocksize-aligned.
5253393b09789ec67bec153b866d7285a1cf1645Matthew Ahrens dmu_tx_count_write(txh, 0, dn->dn_datablksz);
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens /* first block will be modified if it is not aligned */
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens if (!IS_P2ALIGNED(off, 1 << dn->dn_datablkshift))
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens /* last block will be modified if it is not aligned */
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens if (!IS_P2ALIGNED(off + len, 1 << dn->dn_datablkshift))
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens * Check level-1 blocks.
2f3d878000c3b33cde13e16c4f0a1ab15d883a18Matthew Ahrens int shift = dn->dn_datablkshift + dn->dn_indblkshift -
bb411a08b05466bfe0c7095b6373bbc1587e259aMatthew Ahrens * dnode_reallocate() can result in an object with indirect
bb411a08b05466bfe0c7095b6373bbc1587e259aMatthew Ahrens * blocks having an odd data block size. In this case,
bb411a08b05466bfe0c7095b6373bbc1587e259aMatthew Ahrens * just check the single block.
148434217c040ea38dc844384f6ba68d9b325906Matthew Ahrensdmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name)
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * We will be able to fit a new object's entries into one leaf
fa9e4066f08beec538e775443c5be79dd423fcabahrens * block. So there will be at most 2 blocks total,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * including the header block.
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens dmu_tx_count_write(txh, 0, 2 << fzap_default_block_shift);
ad135b5d644628e791c3188a6ecbd9c257961ef8Christopher Siden ASSERT3P(DMU_OT_BYTESWAP(dn->dn_type), ==, DMU_BSWAP_ZAP);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If there is only one block (i.e. this is a micro-zap)
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * and we are not adding anything, the accounting is simple.
b6130ead041e56a0331b89b333bdddab8afb9b1cmaybee * Use max block size here, since we don't know how much
b6130ead041e56a0331b89b333bdddab8afb9b1cmaybee * the size will change between now and the dbuf dirty call.
fa9e4066f08beec538e775443c5be79dd423fcabahrens if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tooverwrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tounref,
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * access the name in this fat-zap so that we'll check
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * for i/o errors to the leaf blocks, etc.
79d728325e257b05859d2eef4a4dfbefdce05a76Matthew Ahrens err = zap_lookup_by_dnode(dn, name, 8, 0, NULL);
79d728325e257b05859d2eef4a4dfbefdce05a76Matthew Ahrens err = zap_count_write_by_dnode(dn, name, add,
720d1aa11ffe42c822fb6151fb4ae83194389942Sanjeev Bagewadi &txh->txh_space_towrite, &txh->txh_space_tooverwrite);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If the modified blocks are scattered to the four winds,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * we'll have to modify an indirect twig for each. We can make
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * modifications at up to 3 locations:
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * - header block at the beginning of the object
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * - target leaf block
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * - end of the object, where we might need to write:
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * - a new leaf block if the target block needs to be split
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * - the new pointer table, if it is growing
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens * - the new cookie table, if it is growing
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens dsl_dataset_phys(dn->dn_objset->os_dsl_dataset);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens for (int lvl = 1; lvl < dn->dn_nlevels; lvl++) {
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens uint64_t num_indirects = 1 + (dn->dn_maxblkid >> (epbs * lvl));
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens uint64_t spc = MIN(3, num_indirects) << dn->dn_indblkshift;
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tooverwrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite, space, FTAG);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * By asserting that the tx is assigned, we're counting the
fa9e4066f08beec538e775443c5be79dd423fcabahrens * number of dn_tx_holds, which is the same as the number of
fa9e4066f08beec538e775443c5be79dd423fcabahrens * dn_holds. Otherwise, we'd be counting dn_holds, but
fa9e4066f08beec538e775443c5be79dd423fcabahrens * dn_tx_holds could be 0.
fa9e4066f08beec538e775443c5be79dd423fcabahrens /* if (tx->tx_anyobj == TRUE) */
fa9e4066f08beec538e775443c5be79dd423fcabahrens /* return (0); */
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens if (txh->txh_dnode && txh->txh_dnode->dn_object == object)
503ad85c168c7992ccc310af845a581cff3c72b5Matthew Ahrens ASSERT(tx->tx_objset == NULL || dn->dn_objset == tx->tx_objset);
fa9e4066f08beec538e775443c5be79dd423fcabahrens /* XXX No checking on the meta dnode for now */
744947dc83c634d985ed3ad79ac9c5e28d1865fdTom Erickson if (db->db.db_object == DMU_META_DNODE_OBJECT) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens ASSERT(dn == NULL || dn->dn_assigned_txg == tx->tx_txg);
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens if (txh->txh_dnode == dn && txh->txh_type != THT_NEWOBJECT)
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens if (txh->txh_dnode == NULL || txh->txh_dnode == dn) {
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens /* XXX txh_arg2 better not be zero... */
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens dprintf("found txh type %x beginblk=%llx endblk=%llx\n",
fa9e4066f08beec538e775443c5be79dd423fcabahrens * We will let this hold work for the bonus
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * or spill buffer so that we don't need to
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * hold it when creating a new object.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * They might have to increase nlevels,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * thus dirtying the new TLIBs. Or the
fa9e4066f08beec538e775443c5be79dd423fcabahrens * might have to change the block size,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * thus dirying the new lvl=0 blk=0.
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee * We will dirty all the level 1 blocks in
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee * the free range and perhaps the first and
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee * last level 0 block.
fa9e4066f08beec538e775443c5be79dd423fcabahrens panic("dirtying dbuf obj=%llx lvl=%u blkid=%llx but not tx_held\n",
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * If we can't do 10 iops, something is wrong. Let us go ahead
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * and hit zfs_dirty_data_max.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrensint zfs_delay_resolution_ns = 100 * 1000; /* 100 microseconds */
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * We delay transactions when we've determined that the backend storage
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * isn't able to accommodate the rate of incoming writes.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * If there is already a transaction waiting, we delay relative to when
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * that transaction finishes waiting. This way the calculated min_time
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * is independent of the number of threads concurrently executing
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * transactions.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * If we are the only waiter, wait relative to when the transaction
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * started, rather than the current time. This credits the transaction for
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * "time already served", e.g. reading indirect blocks.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * The minimum time for a transaction to take is calculated as:
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * min_time = scale * (dirty - min) / (max - dirty)
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * min_time is then capped at zfs_delay_max_ns.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * The delay has two degrees of freedom that can be adjusted via tunables.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * The percentage of dirty data at which we start to delay is defined by
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * zfs_delay_min_dirty_percent. This should typically be at or above
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * zfs_vdev_async_write_active_max_dirty_percent so that we only start to
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * delay after writing at full speed has failed to keep up with the incoming
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * write rate. The scale of the curve is defined by zfs_delay_scale. Roughly
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * speaking, this variable determines the amount of delay at the midpoint of
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 10ms +-------------------------------------------------------------*+
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 2ms + (midpoint) * +
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 1ms + v *** +
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * | zfs_delay_scale ----------> ******** |
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 0 +-------------------------------------*********----------------+
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 0% <- zfs_dirty_data_max -> 100%
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * Note that since the delay is added to the outstanding time remaining on the
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * most recent transaction, the delay is effectively the inverse of IOPS.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * Here the midpoint of 500us translates to 2000 IOPS. The shape of the curve
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * was chosen such that small changes in the amount of accumulated dirty data
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * in the first 3/4 of the curve yield relatively small differences in the
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * amount of delay.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * The effects can be easier to understand when the amount of delay is
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * represented on a log scale:
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 100ms +-------------------------------------------------------------++
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * | (midpoint) ** |
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 1ms + v **** +
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * + zfs_delay_scale ----------> ***** +
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 100us + ** +
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * +--------------------------------------------------------------+
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * 0% <- zfs_dirty_data_max -> 100%
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * Note here that only as the amount of dirty data approaches its limit does
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * the delay start to increase rapidly. The goal of a properly tuned system
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * should be to keep the amount of dirty data out of that range by first
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * ensuring that the appropriate limits are set for the I/O scheduler to reach
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * optimal throughput on the backend storage, and then by changing the value
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * of zfs_delay_scale to increase the steepness of the curve.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100;
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * The caller has already waited until we are under the max.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * We make them pass us the amount of dirty data so we don't
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * have to handle the case of it being >= the max, which could
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * cause a divide-by-zero if it's == the max.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens (dirty - delay_min_bytes) / (zfs_dirty_data_max - dirty);
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens min_tx_time = MIN(min_tx_time, zfs_delay_max_ns);
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens DTRACE_PROBE3(delay__mintime, dmu_tx_t *, tx, uint64_t, dirty,
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens while (cv_timedwait_hires(&curthread->t_delay_cv,
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens &curthread->t_delay_lock, wakeup, zfs_delay_resolution_ns,
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens CALLOUT_FLAG_ABSOLUTE | CALLOUT_FLAG_ROUNDUP) > 0)
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrensdmu_tx_try_assign(dmu_tx_t *tx, txg_how_t txg_how)
715614a4137a6d2100ff60821e21cabc22d06088maybee uint64_t towrite, tofree, tooverwrite, tounref, tohold, fudge;
0a4e9518a44f226be6d39383330b5b1792d2f184gw * If the user has indicated a blocking failure mode
0a4e9518a44f226be6d39383330b5b1792d2f184gw * then return ERESTART which will block in dmu_tx_wait().
0a4e9518a44f226be6d39383330b5b1792d2f184gw * Otherwise, return EIO so that an error can get
0a4e9518a44f226be6d39383330b5b1792d2f184gw * propagated back to the VOP calls.
0a4e9518a44f226be6d39383330b5b1792d2f184gw * Note that we always honor the txg_how flag regardless
0a4e9518a44f226be6d39383330b5b1792d2f184gw * of the failuremode setting.
0a4e9518a44f226be6d39383330b5b1792d2f184gw if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE &&
fa9e4066f08beec538e775443c5be79dd423fcabahrens tx->tx_txg = txg_hold_open(tx->tx_pool, &tx->tx_txgh);
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens * NB: No error returns are allowed after txg_hold_open, but
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens * before processing the dnode holds, due to the
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens * dmu_tx_unassign() logic.
715614a4137a6d2100ff60821e21cabc22d06088maybee towrite = tofree = tooverwrite = tounref = tohold = fudge = 0;
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens towrite += refcount_count(&txh->txh_space_towrite);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens tofree += refcount_count(&txh->txh_space_tofree);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens tooverwrite += refcount_count(&txh->txh_space_tooverwrite);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens tounref += refcount_count(&txh->txh_space_tounref);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens tohold += refcount_count(&txh->txh_memory_tohold);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * If a snapshot has been taken since we made our estimates,
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * assume that we won't be able to free or overwrite anything.
503ad85c168c7992ccc310af845a581cff3c72b5Matthew Ahrens dsl_dataset_prev_snap_txg(tx->tx_objset->os_dsl_dataset) >
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee /* needed allocation: worst-case estimate of write space */
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee asize = spa_get_asize(tx->tx_pool->dp_spa, towrite + tooverwrite);
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee /* freed space estimate: worst-case overwrite + free estimate */
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens fsize = spa_get_asize(tx->tx_pool->dp_spa, tooverwrite) + tofree;
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee /* convert unrefd space to worst-case estimate */
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee /* calculate memory footprint estimate */
715614a4137a6d2100ff60821e21cabc22d06088maybee * Add in 'tohold' to account for our dirty holds on this memory
715614a4137a6d2100ff60821e21cabc22d06088maybee * XXX - the "fudge" factor is to account for skipped blocks that
715614a4137a6d2100ff60821e21cabc22d06088maybee * we missed because dnode_next_offset() misses in-core-only blocks.
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee int err = dsl_dir_tempreserve_space(tx->tx_dir, memory,
cdb0ab79ea1af7b8fc339a04d4bf7426dc77ec4emaybee asize, fsize, usize, &tx->tx_tempreserve_cookie, tx);
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (0);
3e30c24aeefdee1631958ecf17f18da671781956Will Andrews * Walk the transaction's hold list, removing the hold on the
3e30c24aeefdee1631958ecf17f18da671781956Will Andrews * associated dnode, and notifying waiters if the refcount drops to 0.
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens for (txh = list_head(&tx->tx_holds); txh != tx->tx_needassign_txh;
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Assign tx to a transaction group. txg_how can be one of:
fa9e4066f08beec538e775443c5be79dd423fcabahrens * (1) TXG_WAIT. If the current open txg is full, waits until there's
fa9e4066f08beec538e775443c5be79dd423fcabahrens * a new one. This should be used when you're not holding locks.
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens * It will only fail if we're truly out of space (or over quota).
fa9e4066f08beec538e775443c5be79dd423fcabahrens * (2) TXG_NOWAIT. If we can't assign into the current open txg without
fa9e4066f08beec538e775443c5be79dd423fcabahrens * blocking, returns immediately with ERESTART. This should be used
fa9e4066f08beec538e775443c5be79dd423fcabahrens * whenever you're holding locks. On an ERESTART error, the caller
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens * should drop locks, do a dmu_tx_wait(tx), and try again.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * (3) TXG_WAITED. Like TXG_NOWAIT, but indicates that dmu_tx_wait()
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * has already been called on behalf of this operation (though
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * most likely on a different tx).
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrensdmu_tx_assign(dmu_tx_t *tx, txg_how_t txg_how)
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens ASSERT(txg_how == TXG_WAIT || txg_how == TXG_NOWAIT ||
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens /* If we might wait, we must not hold the config lock. */
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens ASSERT(txg_how != TXG_WAIT || !dsl_pool_config_held(tx->tx_pool));
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens while ((err = dmu_tx_try_assign(tx, txg_how)) != 0) {
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (0);
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * dmu_tx_try_assign() has determined that we need to wait
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * because we've consumed much or all of the dirty buffer
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens while (dp->dp_dirty_total >= zfs_dirty_data_max)
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens cv_wait(&dp->dp_spaceavail_cv, &dp->dp_lock);
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * Note: setting tx_waited only has effect if the caller
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * used TX_WAIT. Otherwise they are going to destroy
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * this tx and try again. The common case, zfs_write(),
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * uses TX_WAIT.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens } else if (spa_suspended(spa) || tx->tx_lasttried_txg == 0) {
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * If the pool is suspended we need to wait until it
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * is resumed. Note that it's possible that the pool
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * has become active after this thread has tried to
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * obtain a tx. If that's the case then tx_lasttried_txg
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * would not have been set.
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens txg_wait_synced(dp, spa_last_synced_txg(spa) + 1);
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * A dnode is assigned to the quiescing txg. Wait for its
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens * transaction to complete.
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens while (dn->dn_assigned_txg == tx->tx_lasttried_txg - 1)
8a2f1b9190d1dc288470a1fd2776d79ce82cb129ahrens txg_wait_open(tx->tx_pool, tx->tx_lasttried_txg + 1);
fa9e4066f08beec538e775443c5be79dd423fcabahrens ASSERT3U(refcount_count(&tx->tx_space_written) + delta, <=,
fa9e4066f08beec538e775443c5be79dd423fcabahrens (void) refcount_add_many(&tx->tx_space_written, delta, NULL);
fa9e4066f08beec538e775443c5be79dd423fcabahrens (void) refcount_add_many(&tx->tx_space_freed, -delta, NULL);
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens while ((txh = list_head(&tx->tx_holds)) != NULL) {
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_destroy_many(&txh->txh_space_towrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_destroy_many(&txh->txh_space_tofree,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_destroy_many(&txh->txh_space_tooverwrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_count(&txh->txh_space_tooverwrite));
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_destroy_many(&txh->txh_space_tounref,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens refcount_destroy_many(&txh->txh_memory_tohold,
3e30c24aeefdee1631958ecf17f18da671781956Will Andrews * Go through the transaction's hold list and remove holds on
3e30c24aeefdee1631958ecf17f18da671781956Will Andrews * associated dnodes, notifying waiters if no holds remain.
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens for (dmu_tx_hold_t *txh = list_head(&tx->tx_holds); txh != NULL;
fa9e4066f08beec538e775443c5be79dd423fcabahrens dsl_dir_tempreserve_clear(tx->tx_tempreserve_cookie, tx);
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia txg_register_callbacks(&tx->tx_txgh, &tx->tx_callbacks);
fa9e4066f08beec538e775443c5be79dd423fcabahrens dprintf("towrite=%llu written=%llu tofree=%llu freed=%llu\n",
fa9e4066f08beec538e775443c5be79dd423fcabahrens tx->tx_space_towrite, refcount_count(&tx->tx_space_written),
fa9e4066f08beec538e775443c5be79dd423fcabahrens tx->tx_space_tofree, refcount_count(&tx->tx_space_freed));
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia * Call any registered callbacks with an error code.
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia dmu_tx_do_callbacks(&tx->tx_callbacks, ECANCELED);
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correiadmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *func, void *data)
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia dcb = kmem_alloc(sizeof (dmu_tx_callback_t), KM_SLEEP);
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia list_insert_tail(&tx->tx_callbacks, dcb);
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia * Call all the commit callbacks on a list, with a given error code.
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correiadmu_tx_do_callbacks(list_t *cb_list, int error)
d20e665c84abf083a9e8b62cca93383ecb55afdfRicardo M. Correia kmem_free(dcb, sizeof (dmu_tx_callback_t));
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * Interface to hold a bunch of attributes.
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * used for creating new files.
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * attrsize is the total size of all attributes
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * to be added during object creation
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * For updating/adding a single attribute dmu_tx_hold_sa() should be used.
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * hold necessary attribute name for attribute registration.
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * should be a very rare case where this is needed. If it does
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * happen it would only happen on the first write to the file system.
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaumdmu_tx_sa_registration_hold(sa_os_t *sa, dmu_tx_t *tx)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaumdmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, object,
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum /* If blkptr doesn't exist then add space to towrite */
9dccfd2a04cd1645e2616b7307b3a88041aba991Albert Lee if (!(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tooverwrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_towrite,
0c779ad424a92a84d1e07d47cab7f8009189202bMatthew Ahrens (void) refcount_add_many(&txh->txh_space_tounref,
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaumdmu_tx_hold_sa_create(dmu_tx_t *tx, int attrsize)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum if (tx->tx_objset->os_sa->sa_master_obj == 0)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum if (tx->tx_objset->os_sa->sa_layout_attr_obj)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, sa->sa_layout_attr_obj, B_TRUE, NULL);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_LAYOUTS);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_REGISTRY);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum if (attrsize <= DN_MAX_BONUSLEN && !sa->sa_force_spill)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum (void) dmu_tx_hold_object_impl(tx, tx->tx_objset, DMU_NEW_OBJECT,
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * Hold SA attribute
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * dmu_tx_hold_sa(dmu_tx_t *tx, sa_handle_t *, attribute, add, size)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * variable_size is the total size of all variable sized attributes
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * passed to this function. It is not the total size of all
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum * variable size attributes that *may* exist on this object.
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaumdmu_tx_hold_sa(dmu_tx_t *tx, sa_handle_t *hdl, boolean_t may_grow)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum if (tx->tx_objset->os_sa->sa_master_obj == 0)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum if (tx->tx_objset->os_sa->sa_reg_attr_obj == 0 ||
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum tx->tx_objset->os_sa->sa_layout_attr_obj == 0) {
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_LAYOUTS);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_REGISTRY);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum if (may_grow && tx->tx_objset->os_sa->sa_layout_attr_obj)
0a586cea3ceec7e5e50e7e54c745082a7a333ac2Mark Shellenbaum dmu_tx_hold_zap(tx, sa->sa_layout_attr_obj, B_TRUE, NULL);
744947dc83c634d985ed3ad79ac9c5e28d1865fdTom Erickson if (sa->sa_force_spill || may_grow || hdl->sa_spill) {