From 133faca2752547a6e15cac7189698f877be86388 Mon Sep 17 00:00:00 2001 From: Brian Behlendorf Date: Tue, 9 May 2023 09:03:10 -0700 Subject: [PATCH] Add dmu_tx_hold_append() interface Provides an interface which callers can use to declare a write when the exact starting offset in not yet known. Since the full range being updated is not available only the first L0 block at the provided offset will be prefetched. Reviewed-by: Olaf Faaland Signed-off-by: Brian Behlendorf Closes #14819 --- include/sys/dmu.h | 3 ++ include/sys/dmu_tx.h | 1 + module/zfs/dmu_tx.c | 105 +++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 109 insertions(+) diff --git a/include/sys/dmu.h b/include/sys/dmu.h index 7bdd42e8be..12bd887200 100644 --- a/include/sys/dmu.h +++ b/include/sys/dmu.h @@ -778,6 +778,9 @@ dmu_tx_t *dmu_tx_create(objset_t *os); void dmu_tx_hold_write(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); void dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off, int len); +void dmu_tx_hold_append(dmu_tx_t *tx, uint64_t object, uint64_t off, int len); +void dmu_tx_hold_append_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off, + int len); void dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, uint64_t len); void dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off, diff --git a/include/sys/dmu_tx.h b/include/sys/dmu_tx.h index ad3f1b0e47..e8886fd4e6 100644 --- a/include/sys/dmu_tx.h +++ b/include/sys/dmu_tx.h @@ -90,6 +90,7 @@ enum dmu_tx_hold_type { THT_ZAP, THT_SPACE, THT_SPILL, + THT_APPEND, THT_NUMTYPES }; diff --git a/module/zfs/dmu_tx.c b/module/zfs/dmu_tx.c index 1eed0526b5..063934f394 100644 --- a/module/zfs/dmu_tx.c +++ b/module/zfs/dmu_tx.c @@ -290,6 +290,53 @@ dmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) } } +static void +dmu_tx_count_append(dmu_tx_hold_t *txh, uint64_t off, uint64_t len) +{ + dnode_t *dn = txh->txh_dnode; + int err = 0; + + if (len == 0) + return; + + (void) zfs_refcount_add_many(&txh->txh_space_towrite, len, FTAG); + + if (dn == NULL) + return; + + /* + * For i/o error checking, read the blocks that will be needed + * to perform the append; first level-0 block (if not aligned, i.e. + * if they are partial-block writes), no additional blocks are read. + */ + if (dn->dn_maxblkid == 0) { + if (off < dn->dn_datablksz && + (off > 0 || len < dn->dn_datablksz)) { + err = dmu_tx_check_ioerr(NULL, dn, 0, 0); + if (err != 0) { + txh->txh_tx->tx_err = err; + } + } + } else { + zio_t *zio = zio_root(dn->dn_objset->os_spa, + NULL, NULL, ZIO_FLAG_CANFAIL); + + /* first level-0 block */ + uint64_t start = off >> dn->dn_datablkshift; + if (P2PHASE(off, dn->dn_datablksz) || len < dn->dn_datablksz) { + err = dmu_tx_check_ioerr(zio, dn, 0, start); + if (err != 0) { + txh->txh_tx->tx_err = err; + } + } + + err = zio_wait(zio); + if (err != 0) { + txh->txh_tx->tx_err = err; + } + } +} + static void dmu_tx_count_dnode(dmu_tx_hold_t *txh) { @@ -330,6 +377,42 @@ dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off, int len) } } +/* + * Should be used when appending to an object and the exact offset is unknown. + * The write must occur at or beyond the specified offset. Only the L0 block + * at provided offset will be prefetched. + */ +void +dmu_tx_hold_append(dmu_tx_t *tx, uint64_t object, uint64_t off, int len) +{ + dmu_tx_hold_t *txh; + + ASSERT0(tx->tx_txg); + ASSERT3U(len, <=, DMU_MAX_ACCESS); + + txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, + object, THT_APPEND, off, DMU_OBJECT_END); + if (txh != NULL) { + dmu_tx_count_append(txh, off, len); + dmu_tx_count_dnode(txh); + } +} + +void +dmu_tx_hold_append_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off, int len) +{ + dmu_tx_hold_t *txh; + + ASSERT0(tx->tx_txg); + ASSERT3U(len, <=, DMU_MAX_ACCESS); + + txh = dmu_tx_hold_dnode_impl(tx, dn, THT_APPEND, off, DMU_OBJECT_END); + if (txh != NULL) { + dmu_tx_count_append(txh, off, len); + dmu_tx_count_dnode(txh); + } +} + /* * This function marks the transaction as being a "net free". The end * result is that refquotas will be disabled for this transaction, and @@ -638,6 +721,26 @@ dmu_tx_dirty_buf(dmu_tx_t *tx, dmu_buf_impl_t *db) if (blkid == 0) match_offset = TRUE; break; + case THT_APPEND: + if (blkid >= beginblk && (blkid <= endblk || + txh->txh_arg2 == DMU_OBJECT_END)) + match_offset = TRUE; + + /* + * THT_WRITE used for bonus and spill blocks. + */ + ASSERT(blkid != DMU_BONUS_BLKID && + blkid != DMU_SPILL_BLKID); + + /* + * They might have to increase nlevels, + * thus dirtying the new TLIBs. Or the + * might have to change the block size, + * thus dirying the new lvl=0 blk=0. + */ + if (blkid == 0) + match_offset = TRUE; + break; case THT_FREE: /* * We will dirty all the level 1 blocks in @@ -1421,6 +1524,8 @@ dmu_tx_fini(void) EXPORT_SYMBOL(dmu_tx_create); EXPORT_SYMBOL(dmu_tx_hold_write); EXPORT_SYMBOL(dmu_tx_hold_write_by_dnode); +EXPORT_SYMBOL(dmu_tx_hold_append); +EXPORT_SYMBOL(dmu_tx_hold_append_by_dnode); EXPORT_SYMBOL(dmu_tx_hold_free); EXPORT_SYMBOL(dmu_tx_hold_free_by_dnode); EXPORT_SYMBOL(dmu_tx_hold_zap);