/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright 2011 Nexenta Systems, Inc. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
*/
#include <sys/dmu.h>
#include <sys/dbuf.h>
#include <sys/dmu_tx.h>
#include <sys/dmu_objset.h>
-#include <sys/dsl_dataset.h> /* for dsl_dataset_block_freeable() */
-#include <sys/dsl_dir.h> /* for dsl_dir_tempreserve_*() */
+#include <sys/dsl_dataset.h>
+#include <sys/dsl_dir.h>
#include <sys/dsl_pool.h>
-#include <sys/zap_impl.h> /* for fzap_default_block_shift */
+#include <sys/zap_impl.h>
#include <sys/spa.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
#include <sys/zfs_context.h>
#include <sys/varargs.h>
+#include <sys/trace_dmu.h>
typedef void (*dmu_tx_hold_func_t)(dmu_tx_t *tx, struct dnode *dn,
uint64_t arg1, uint64_t arg2);
{ "dmu_tx_error", KSTAT_DATA_UINT64 },
{ "dmu_tx_suspended", KSTAT_DATA_UINT64 },
{ "dmu_tx_group", KSTAT_DATA_UINT64 },
- { "dmu_tx_how", KSTAT_DATA_UINT64 },
{ "dmu_tx_memory_reserve", KSTAT_DATA_UINT64 },
{ "dmu_tx_memory_reclaim", KSTAT_DATA_UINT64 },
- { "dmu_tx_memory_inflight", KSTAT_DATA_UINT64 },
{ "dmu_tx_dirty_throttle", KSTAT_DATA_UINT64 },
- { "dmu_tx_write_limit", KSTAT_DATA_UINT64 },
+ { "dmu_tx_dirty_delay", KSTAT_DATA_UINT64 },
+ { "dmu_tx_dirty_over_max", KSTAT_DATA_UINT64 },
{ "dmu_tx_quota", KSTAT_DATA_UINT64 },
};
dmu_tx_t *
dmu_tx_create_dd(dsl_dir_t *dd)
{
- dmu_tx_t *tx = kmem_zalloc(sizeof (dmu_tx_t), KM_PUSHPAGE);
+ dmu_tx_t *tx = kmem_zalloc(sizeof (dmu_tx_t), KM_SLEEP);
tx->tx_dir = dd;
if (dd != NULL)
tx->tx_pool = dd->dd_pool;
offsetof(dmu_tx_hold_t, txh_node));
list_create(&tx->tx_callbacks, sizeof (dmu_tx_callback_t),
offsetof(dmu_tx_callback_t, dcb_node));
-#ifdef DEBUG_DMU_TX
- refcount_create(&tx->tx_space_written);
- refcount_create(&tx->tx_space_freed);
-#endif
+ tx->tx_start = gethrtime();
return (tx);
}
{
dmu_tx_t *tx = dmu_tx_create_dd(os->os_dsl_dataset->ds_dir);
tx->tx_objset = os;
- tx->tx_lastsnap_txg = dsl_dataset_prev_snap_txg(os->os_dsl_dataset);
return (tx);
}
{
dmu_tx_t *tx = dmu_tx_create_dd(NULL);
- ASSERT3U(txg, <=, dp->dp_tx.tx_open_txg);
+ txg_verify(dp->dp_spa, txg);
tx->tx_pool = dp;
tx->tx_txg = txg;
tx->tx_anyobj = TRUE;
}
static dmu_tx_hold_t *
-dmu_tx_hold_object_impl(dmu_tx_t *tx, objset_t *os, uint64_t object,
- enum dmu_tx_hold_type type, uint64_t arg1, uint64_t arg2)
+dmu_tx_hold_dnode_impl(dmu_tx_t *tx, dnode_t *dn, enum dmu_tx_hold_type type,
+ uint64_t arg1, uint64_t arg2)
{
dmu_tx_hold_t *txh;
- dnode_t *dn = NULL;
- int err;
- if (object != DMU_NEW_OBJECT) {
- err = dnode_hold(os, object, tx, &dn);
- if (err) {
- tx->tx_err = err;
- return (NULL);
- }
-
- if (err == 0 && tx->tx_txg != 0) {
+ if (dn != NULL) {
+ (void) refcount_add(&dn->dn_holds, tx);
+ if (tx->tx_txg != 0) {
mutex_enter(&dn->dn_mtx);
/*
* dn->dn_assigned_txg == tx->tx_txg doesn't pose a
}
}
- txh = kmem_zalloc(sizeof (dmu_tx_hold_t), KM_PUSHPAGE);
+ txh = kmem_zalloc(sizeof (dmu_tx_hold_t), KM_SLEEP);
txh->txh_tx = tx;
txh->txh_dnode = dn;
-#ifdef DEBUG_DMU_TX
+ refcount_create(&txh->txh_space_towrite);
+ refcount_create(&txh->txh_memory_tohold);
txh->txh_type = type;
txh->txh_arg1 = arg1;
txh->txh_arg2 = arg2;
-#endif
list_insert_tail(&tx->tx_holds, txh);
return (txh);
}
+static dmu_tx_hold_t *
+dmu_tx_hold_object_impl(dmu_tx_t *tx, objset_t *os, uint64_t object,
+ enum dmu_tx_hold_type type, uint64_t arg1, uint64_t arg2)
+{
+ dnode_t *dn = NULL;
+ dmu_tx_hold_t *txh;
+ int err;
+
+ if (object != DMU_NEW_OBJECT) {
+ err = dnode_hold(os, object, FTAG, &dn);
+ if (err != 0) {
+ tx->tx_err = err;
+ return (NULL);
+ }
+ }
+ txh = dmu_tx_hold_dnode_impl(tx, dn, type, arg1, arg2);
+ if (dn != NULL)
+ dnode_rele(dn, FTAG);
+ return (txh);
+}
+
void
-dmu_tx_add_new_object(dmu_tx_t *tx, objset_t *os, uint64_t object)
+dmu_tx_add_new_object(dmu_tx_t *tx, dnode_t *dn)
{
/*
* If we're syncing, they can manipulate any object anyhow, and
* the hold on the dnode_t can cause problems.
*/
- if (!dmu_tx_is_syncing(tx)) {
- (void) dmu_tx_hold_object_impl(tx, os,
- object, THT_NEWOBJECT, 0, 0);
- }
+ if (!dmu_tx_is_syncing(tx))
+ (void) dmu_tx_hold_dnode_impl(tx, dn, THT_NEWOBJECT, 0, 0);
}
+/*
+ * This function reads specified data from disk. The specified data will
+ * be needed to perform the transaction -- i.e, it will be read after
+ * we do dmu_tx_assign(). There are two reasons that we read the data now
+ * (before dmu_tx_assign()):
+ *
+ * 1. Reading it now has potentially better performance. The transaction
+ * has not yet been assigned, so the TXG is not held open, and also the
+ * caller typically has less locks held when calling dmu_tx_hold_*() than
+ * after the transaction has been assigned. This reduces the lock (and txg)
+ * hold times, thus reducing lock contention.
+ *
+ * 2. It is easier for callers (primarily the ZPL) to handle i/o errors
+ * that are detected before they start making changes to the DMU state
+ * (i.e. now). Once the transaction has been assigned, and some DMU
+ * state has been changed, it can be difficult to recover from an i/o
+ * error (e.g. to undo the changes already made in memory at the DMU
+ * layer). Typically code to do so does not exist in the caller -- it
+ * assumes that the data has already been cached and thus i/o errors are
+ * not possible.
+ *
+ * It has been observed that the i/o initiated here can be a performance
+ * problem, and it appears to be optional, because we don't look at the
+ * data which is read. However, removing this read would only serve to
+ * move the work elsewhere (after the dmu_tx_assign()), where it may
+ * have a greater impact on performance (in addition to the impact on
+ * fault tolerance noted above).
+ */
static int
dmu_tx_check_ioerr(zio_t *zio, dnode_t *dn, int level, uint64_t blkid)
{
return (err);
}
-static void
-dmu_tx_count_twig(dmu_tx_hold_t *txh, dnode_t *dn, dmu_buf_impl_t *db,
- int level, uint64_t blkid, boolean_t freeable, uint64_t *history)
-{
- objset_t *os = dn->dn_objset;
- dsl_dataset_t *ds = os->os_dsl_dataset;
- int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
- dmu_buf_impl_t *parent = NULL;
- blkptr_t *bp = NULL;
- uint64_t space;
-
- if (level >= dn->dn_nlevels || history[level] == blkid)
- return;
-
- history[level] = blkid;
-
- space = (level == 0) ? dn->dn_datablksz : (1ULL << dn->dn_indblkshift);
-
- if (db == NULL || db == dn->dn_dbuf) {
- ASSERT(level != 0);
- db = NULL;
- } else {
- ASSERT(DB_DNODE(db) == dn);
- ASSERT(db->db_level == level);
- ASSERT(db->db.db_size == space);
- ASSERT(db->db_blkid == blkid);
- bp = db->db_blkptr;
- parent = db->db_parent;
- }
-
- freeable = (bp && (freeable ||
- dsl_dataset_block_freeable(ds, bp, bp->blk_birth)));
-
- if (freeable)
- txh->txh_space_tooverwrite += space;
- else
- txh->txh_space_towrite += space;
- if (bp)
- txh->txh_space_tounref += bp_get_dsize(os->os_spa, bp);
-
- dmu_tx_count_twig(txh, dn, parent, level + 1,
- blkid >> epbs, freeable, history);
-}
-
/* ARGSUSED */
static void
dmu_tx_count_write(dmu_tx_hold_t *txh, uint64_t off, uint64_t len)
{
dnode_t *dn = txh->txh_dnode;
- uint64_t start, end, i;
- int min_bs, max_bs, min_ibs, max_ibs, epbs, bits;
int err = 0;
- int l;
if (len == 0)
return;
- min_bs = SPA_MINBLOCKSHIFT;
- max_bs = SPA_MAXBLOCKSHIFT;
- min_ibs = DN_MIN_INDBLKSHIFT;
- max_ibs = DN_MAX_INDBLKSHIFT;
-
- if (dn) {
- uint64_t history[DN_MAX_LEVELS];
- int nlvls = dn->dn_nlevels;
- int delta;
+ (void) refcount_add_many(&txh->txh_space_towrite, len, FTAG);
- /*
- * For i/o error checking, read the first and last level-0
- * blocks (if they are not aligned), and all the level-1 blocks.
- */
- if (dn->dn_maxblkid == 0) {
- delta = dn->dn_datablksz;
- start = (off < dn->dn_datablksz) ? 0 : 1;
- end = (off+len <= dn->dn_datablksz) ? 0 : 1;
- if (start == 0 && (off > 0 || len < dn->dn_datablksz)) {
- err = dmu_tx_check_ioerr(NULL, dn, 0, 0);
- if (err)
- goto out;
- delta -= off;
- }
- } else {
- zio_t *zio = zio_root(dn->dn_objset->os_spa,
- NULL, NULL, ZIO_FLAG_CANFAIL);
-
- /* first level-0 block */
- 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)
- goto out;
- }
+ if (refcount_count(&txh->txh_space_towrite) > 2 * DMU_MAX_ACCESS)
+ err = SET_ERROR(EFBIG);
- /* last level-0 block */
- end = (off+len-1) >> dn->dn_datablkshift;
- if (end != start && end <= dn->dn_maxblkid &&
- P2PHASE(off+len, dn->dn_datablksz)) {
- err = dmu_tx_check_ioerr(zio, dn, 0, end);
- if (err)
- goto out;
- }
+ if (dn == NULL)
+ return;
- /* level-1 blocks */
- if (nlvls > 1) {
- int shft = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
- for (i = (start>>shft)+1; i < end>>shft; i++) {
- err = dmu_tx_check_ioerr(zio, dn, 1, i);
- if (err)
- goto out;
- }
+ /*
+ * For i/o error checking, read the blocks that will be needed
+ * to perform the write: the first and last level-0 blocks (if
+ * they are not aligned, i.e. if they are partial-block writes),
+ * and all the level-1 blocks.
+ */
+ 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;
}
-
- err = zio_wait(zio);
- if (err)
- goto out;
- delta = P2NPHASE(off, dn->dn_datablksz);
- }
-
- min_ibs = max_ibs = dn->dn_indblkshift;
- if (dn->dn_maxblkid > 0) {
- /*
- * The blocksize can't change,
- * so we can make a more precise estimate.
- */
- ASSERT(dn->dn_datablkshift != 0);
- min_bs = max_bs = dn->dn_datablkshift;
}
+ } else {
+ zio_t *zio = zio_root(dn->dn_objset->os_spa,
+ NULL, NULL, ZIO_FLAG_CANFAIL);
- /*
- * If this write is not off the end of the file
- * we need to account for overwrites/unref.
- */
- if (start <= dn->dn_maxblkid) {
- for (l = 0; l < DN_MAX_LEVELS; l++)
- history[l] = -1ULL;
+ /* 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;
+ }
}
- while (start <= dn->dn_maxblkid) {
- dmu_buf_impl_t *db;
-
- rw_enter(&dn->dn_struct_rwlock, RW_READER);
- err = dbuf_hold_impl(dn, 0, start, FALSE, FTAG, &db);
- rw_exit(&dn->dn_struct_rwlock);
- if (err) {
+ /* last level-0 block */
+ uint64_t end = (off + len - 1) >> dn->dn_datablkshift;
+ if (end != start && end <= dn->dn_maxblkid &&
+ P2PHASE(off + len, dn->dn_datablksz)) {
+ err = dmu_tx_check_ioerr(zio, dn, 0, end);
+ if (err != 0) {
txh->txh_tx->tx_err = err;
- return;
}
+ }
- dmu_tx_count_twig(txh, dn, db, 0, start, B_FALSE,
- history);
- dbuf_rele(db, FTAG);
- if (++start > end) {
- /*
- * Account for new indirects appearing
- * before this IO gets assigned into a txg.
- */
- bits = 64 - min_bs;
- epbs = min_ibs - SPA_BLKPTRSHIFT;
- for (bits -= epbs * (nlvls - 1);
- bits >= 0; bits -= epbs)
- txh->txh_fudge += 1ULL << max_ibs;
- goto out;
+ /* level-1 blocks */
+ if (dn->dn_nlevels > 1) {
+ int shft = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
+ for (uint64_t i = (start >> shft) + 1;
+ i < end >> shft; i++) {
+ err = dmu_tx_check_ioerr(zio, dn, 1, i);
+ if (err != 0) {
+ txh->txh_tx->tx_err = err;
+ }
}
- off += delta;
- if (len >= delta)
- len -= delta;
- delta = dn->dn_datablksz;
}
- }
- /*
- * 'end' is the last thing we will access, not one past.
- * This way we won't overflow when accessing the last byte.
- */
- start = P2ALIGN(off, 1ULL << max_bs);
- end = P2ROUNDUP(off + len, 1ULL << max_bs) - 1;
- txh->txh_space_towrite += end - start + 1;
-
- start >>= min_bs;
- end >>= min_bs;
-
- epbs = min_ibs - SPA_BLKPTRSHIFT;
-
- /*
- * The object contains at most 2^(64 - min_bs) blocks,
- * and each indirect level maps 2^epbs.
- */
- for (bits = 64 - min_bs; bits >= 0; bits -= epbs) {
- start >>= epbs;
- end >>= epbs;
- ASSERT3U(end, >=, start);
- txh->txh_space_towrite += (end - start + 1) << max_ibs;
- if (start != 0) {
- /*
- * We also need a new blkid=0 indirect block
- * to reference any existing file data.
- */
- txh->txh_space_towrite += 1ULL << max_ibs;
+ err = zio_wait(zio);
+ if (err != 0) {
+ txh->txh_tx->tx_err = err;
}
}
-
-out:
- if (txh->txh_space_towrite + txh->txh_space_tooverwrite >
- 2 * DMU_MAX_ACCESS)
- err = SET_ERROR(EFBIG);
-
- if (err)
- txh->txh_tx->tx_err = err;
}
static void
dmu_tx_count_dnode(dmu_tx_hold_t *txh)
{
- dnode_t *dn = txh->txh_dnode;
- dnode_t *mdn = DMU_META_DNODE(txh->txh_tx->tx_objset);
- uint64_t space = mdn->dn_datablksz +
- ((mdn->dn_nlevels-1) << mdn->dn_indblkshift);
-
- if (dn && dn->dn_dbuf->db_blkptr &&
- dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset,
- dn->dn_dbuf->db_blkptr, dn->dn_dbuf->db_blkptr->blk_birth)) {
- txh->txh_space_tooverwrite += space;
- txh->txh_space_tounref += space;
- } else {
- txh->txh_space_towrite += space;
- if (dn && dn->dn_dbuf->db_blkptr)
- txh->txh_space_tounref += space;
- }
+ (void) refcount_add_many(&txh->txh_space_towrite, DNODE_MIN_SIZE, FTAG);
}
void
{
dmu_tx_hold_t *txh;
- ASSERT(tx->tx_txg == 0);
- ASSERT(len < DMU_MAX_ACCESS);
+ ASSERT0(tx->tx_txg);
+ ASSERT3U(len, <=, DMU_MAX_ACCESS);
ASSERT(len == 0 || UINT64_MAX - off >= len - 1);
txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
object, THT_WRITE, off, len);
- if (txh == NULL)
- return;
-
- dmu_tx_count_write(txh, off, len);
- dmu_tx_count_dnode(txh);
+ if (txh != NULL) {
+ dmu_tx_count_write(txh, off, len);
+ dmu_tx_count_dnode(txh);
+ }
}
-static void
-dmu_tx_count_free(dmu_tx_hold_t *txh, uint64_t off, uint64_t len)
+void
+dmu_tx_hold_write_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off, int len)
{
- uint64_t blkid, nblks, lastblk;
- uint64_t space = 0, unref = 0, skipped = 0;
- dnode_t *dn = txh->txh_dnode;
- dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
- spa_t *spa = txh->txh_tx->tx_pool->dp_spa;
- int epbs;
- uint64_t l0span = 0, nl1blks = 0;
-
- if (dn->dn_nlevels == 0)
- return;
-
- /*
- * The struct_rwlock protects us against dn_nlevels
- * changing, in case (against all odds) we manage to dirty &
- * sync out the changes after we check for being dirty.
- * Also, dbuf_hold_impl() wants us to have the struct_rwlock.
- */
- rw_enter(&dn->dn_struct_rwlock, RW_READER);
- epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
- if (dn->dn_maxblkid == 0) {
- if (off == 0 && len >= dn->dn_datablksz) {
- blkid = 0;
- nblks = 1;
- } else {
- rw_exit(&dn->dn_struct_rwlock);
- return;
- }
- } else {
- blkid = off >> dn->dn_datablkshift;
- nblks = (len + dn->dn_datablksz - 1) >> dn->dn_datablkshift;
-
- if (blkid > dn->dn_maxblkid) {
- rw_exit(&dn->dn_struct_rwlock);
- return;
- }
- if (blkid + nblks > dn->dn_maxblkid)
- nblks = dn->dn_maxblkid - blkid + 1;
-
- }
- l0span = nblks; /* save for later use to calc level > 1 overhead */
- if (dn->dn_nlevels == 1) {
- int i;
- for (i = 0; i < nblks; i++) {
- blkptr_t *bp = dn->dn_phys->dn_blkptr;
- ASSERT3U(blkid + i, <, dn->dn_nblkptr);
- bp += blkid + i;
- if (dsl_dataset_block_freeable(ds, bp, bp->blk_birth)) {
- dprintf_bp(bp, "can free old%s", "");
- space += bp_get_dsize(spa, bp);
- }
- unref += BP_GET_ASIZE(bp);
- }
- nl1blks = 1;
- nblks = 0;
- }
-
- lastblk = blkid + nblks - 1;
- while (nblks) {
- dmu_buf_impl_t *dbuf;
- uint64_t ibyte, new_blkid;
- int epb = 1 << epbs;
- int err, i, blkoff, tochk;
- blkptr_t *bp;
-
- ibyte = blkid << dn->dn_datablkshift;
- err = dnode_next_offset(dn,
- DNODE_FIND_HAVELOCK, &ibyte, 2, 1, 0);
- new_blkid = ibyte >> dn->dn_datablkshift;
- if (err == ESRCH) {
- skipped += (lastblk >> epbs) - (blkid >> epbs) + 1;
- break;
- }
- if (err) {
- txh->txh_tx->tx_err = err;
- break;
- }
- if (new_blkid > lastblk) {
- skipped += (lastblk >> epbs) - (blkid >> epbs) + 1;
- break;
- }
-
- if (new_blkid > blkid) {
- ASSERT((new_blkid >> epbs) > (blkid >> epbs));
- skipped += (new_blkid >> epbs) - (blkid >> epbs) - 1;
- nblks -= new_blkid - blkid;
- blkid = new_blkid;
- }
- blkoff = P2PHASE(blkid, epb);
- tochk = MIN(epb - blkoff, nblks);
-
- err = dbuf_hold_impl(dn, 1, blkid >> epbs, FALSE, FTAG, &dbuf);
- if (err) {
- txh->txh_tx->tx_err = err;
- break;
- }
-
- txh->txh_memory_tohold += dbuf->db.db_size;
-
- /*
- * We don't check memory_tohold against DMU_MAX_ACCESS because
- * memory_tohold is an over-estimation (especially the >L1
- * indirect blocks), so it could fail. Callers should have
- * already verified that they will not be holding too much
- * memory.
- */
-
- err = dbuf_read(dbuf, NULL, DB_RF_HAVESTRUCT | DB_RF_CANFAIL);
- if (err != 0) {
- txh->txh_tx->tx_err = err;
- dbuf_rele(dbuf, FTAG);
- break;
- }
-
- bp = dbuf->db.db_data;
- bp += blkoff;
-
- for (i = 0; i < tochk; i++) {
- if (dsl_dataset_block_freeable(ds, &bp[i],
- bp[i].blk_birth)) {
- dprintf_bp(&bp[i], "can free old%s", "");
- space += bp_get_dsize(spa, &bp[i]);
- }
- unref += BP_GET_ASIZE(bp);
- }
- dbuf_rele(dbuf, FTAG);
-
- ++nl1blks;
- blkid += tochk;
- nblks -= tochk;
- }
- rw_exit(&dn->dn_struct_rwlock);
-
- /*
- * Add in memory requirements of higher-level indirects.
- * This assumes a worst-possible scenario for dn_nlevels and a
- * worst-possible distribution of l1-blocks over the region to free.
- */
- {
- uint64_t blkcnt = 1 + ((l0span >> epbs) >> epbs);
- int level = 2;
- /*
- * Here we don't use DN_MAX_LEVEL, but calculate it with the
- * given datablkshift and indblkshift. This makes the
- * difference between 19 and 8 on large files.
- */
- int maxlevel = 2 + (DN_MAX_OFFSET_SHIFT - dn->dn_datablkshift) /
- (dn->dn_indblkshift - SPA_BLKPTRSHIFT);
+ dmu_tx_hold_t *txh;
- while (level++ < maxlevel) {
- txh->txh_memory_tohold += MAX(MIN(blkcnt, nl1blks), 1)
- << dn->dn_indblkshift;
- blkcnt = 1 + (blkcnt >> epbs);
- }
- }
+ ASSERT0(tx->tx_txg);
+ ASSERT3U(len, <=, DMU_MAX_ACCESS);
+ ASSERT(len == 0 || UINT64_MAX - off >= len - 1);
- /* account for new level 1 indirect blocks that might show up */
- if (skipped > 0) {
- txh->txh_fudge += skipped << dn->dn_indblkshift;
- skipped = MIN(skipped, DMU_MAX_DELETEBLKCNT >> epbs);
- txh->txh_memory_tohold += skipped << dn->dn_indblkshift;
+ txh = dmu_tx_hold_dnode_impl(tx, dn, THT_WRITE, off, len);
+ if (txh != NULL) {
+ dmu_tx_count_write(txh, off, len);
+ dmu_tx_count_dnode(txh);
}
- txh->txh_space_tofree += space;
- txh->txh_space_tounref += unref;
}
+/*
+ * This function marks the transaction as being a "net free". The end
+ * result is that refquotas will be disabled for this transaction, and
+ * this transaction will be able to use half of the pool space overhead
+ * (see dsl_pool_adjustedsize()). Therefore this function should only
+ * be called for transactions that we expect will not cause a net increase
+ * in the amount of space used (but it's OK if that is occasionally not true).
+ */
void
-dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, uint64_t len)
+dmu_tx_mark_netfree(dmu_tx_t *tx)
{
- dmu_tx_hold_t *txh;
- dnode_t *dn;
+ tx->tx_netfree = B_TRUE;
+}
+
+static void
+dmu_tx_hold_free_impl(dmu_tx_hold_t *txh, uint64_t off, uint64_t len)
+{
+ dmu_tx_t *tx = txh->txh_tx;
+ dnode_t *dn = txh->txh_dnode;
int err;
- zio_t *zio;
ASSERT(tx->tx_txg == 0);
- txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
- object, THT_FREE, off, len);
- if (txh == NULL)
- return;
- dn = txh->txh_dnode;
+ dmu_tx_count_dnode(txh);
- if (off >= (dn->dn_maxblkid+1) * dn->dn_datablksz)
+ if (off >= (dn->dn_maxblkid + 1) * dn->dn_datablksz)
return;
if (len == DMU_OBJECT_END)
- len = (dn->dn_maxblkid+1) * dn->dn_datablksz - off;
+ len = (dn->dn_maxblkid + 1) * dn->dn_datablksz - off;
dmu_tx_count_dnode(txh);
*/
if (dn->dn_datablkshift == 0) {
if (off != 0 || len < dn->dn_datablksz)
- dmu_tx_count_write(txh, off, len);
+ dmu_tx_count_write(txh, 0, dn->dn_datablksz);
} else {
/* first block will be modified if it is not aligned */
if (!IS_P2ALIGNED(off, 1 << dn->dn_datablkshift))
dmu_tx_count_write(txh, off, 1);
/* last block will be modified if it is not aligned */
if (!IS_P2ALIGNED(off + len, 1 << dn->dn_datablkshift))
- dmu_tx_count_write(txh, off+len, 1);
+ dmu_tx_count_write(txh, off + len, 1);
}
/*
uint64_t end = (off + len) >> shift;
uint64_t i;
- ASSERT(dn->dn_datablkshift != 0);
ASSERT(dn->dn_indblkshift != 0);
- zio = zio_root(tx->tx_pool->dp_spa,
+ /*
+ * dnode_reallocate() can result in an object with indirect
+ * blocks having an odd data block size. In this case,
+ * just check the single block.
+ */
+ if (dn->dn_datablkshift == 0)
+ start = end = 0;
+
+ zio_t *zio = zio_root(tx->tx_pool->dp_spa,
NULL, NULL, ZIO_FLAG_CANFAIL);
for (i = start; i <= end; i++) {
uint64_t ibyte = i << shift;
err = dnode_next_offset(dn, 0, &ibyte, 2, 1, 0);
i = ibyte >> shift;
- if (err == ESRCH)
+ if (err == ESRCH || i > end)
break;
- if (err) {
+ if (err != 0) {
tx->tx_err = err;
+ (void) zio_wait(zio);
return;
}
+ (void) refcount_add_many(&txh->txh_memory_tohold,
+ 1 << dn->dn_indblkshift, FTAG);
+
err = dmu_tx_check_ioerr(zio, dn, 1, i);
- if (err) {
+ if (err != 0) {
tx->tx_err = err;
+ (void) zio_wait(zio);
return;
}
}
err = zio_wait(zio);
- if (err) {
+ if (err != 0) {
tx->tx_err = err;
return;
}
}
+}
+
+void
+dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, uint64_t len)
+{
+ dmu_tx_hold_t *txh;
- dmu_tx_count_free(txh, off, len);
+ txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
+ object, THT_FREE, off, len);
+ if (txh != NULL)
+ (void) dmu_tx_hold_free_impl(txh, off, len);
}
void
-dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name)
+dmu_tx_hold_free_by_dnode(dmu_tx_t *tx, dnode_t *dn, uint64_t off, uint64_t len)
{
dmu_tx_hold_t *txh;
- dnode_t *dn;
- uint64_t nblocks;
- int epbs, err;
- ASSERT(tx->tx_txg == 0);
+ txh = dmu_tx_hold_dnode_impl(tx, dn, THT_FREE, off, len);
+ if (txh != NULL)
+ (void) dmu_tx_hold_free_impl(txh, off, len);
+}
- txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
- object, THT_ZAP, add, (uintptr_t)name);
- if (txh == NULL)
- return;
- dn = txh->txh_dnode;
+static void
+dmu_tx_hold_zap_impl(dmu_tx_hold_t *txh, const char *name)
+{
+ dmu_tx_t *tx = txh->txh_tx;
+ dnode_t *dn = txh->txh_dnode;
+ int err;
+
+ ASSERT(tx->tx_txg == 0);
dmu_tx_count_dnode(txh);
- if (dn == NULL) {
- /*
- * We will be able to fit a new object's entries into one leaf
- * block. So there will be at most 2 blocks total,
- * including the header block.
- */
- dmu_tx_count_write(txh, 0, 2 << fzap_default_block_shift);
+ /*
+ * Modifying a almost-full microzap is around the worst case (128KB)
+ *
+ * If it is a fat zap, the worst case would be 7*16KB=112KB:
+ * - 3 blocks overwritten: target leaf, ptrtbl block, header block
+ * - 4 new blocks written if adding:
+ * - 2 blocks for possibly split leaves,
+ * - 2 grown ptrtbl blocks
+ */
+ (void) refcount_add_many(&txh->txh_space_towrite,
+ MZAP_MAX_BLKSZ, FTAG);
+
+ if (dn == NULL)
return;
- }
ASSERT3U(DMU_OT_BYTESWAP(dn->dn_type), ==, DMU_BSWAP_ZAP);
- if (dn->dn_maxblkid == 0 && !add) {
- blkptr_t *bp;
-
+ if (dn->dn_maxblkid == 0 || name == NULL) {
/*
- * If there is only one block (i.e. this is a micro-zap)
- * and we are not adding anything, the accounting is simple.
+ * This is a microzap (only one block), or we don't know
+ * the name. Check the first block for i/o errors.
*/
err = dmu_tx_check_ioerr(NULL, dn, 0, 0);
- if (err) {
+ if (err != 0) {
tx->tx_err = err;
- return;
}
-
- /*
- * Use max block size here, since we don't know how much
- * the size will change between now and the dbuf dirty call.
- */
- bp = &dn->dn_phys->dn_blkptr[0];
- if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset,
- bp, bp->blk_birth))
- txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE;
- else
- txh->txh_space_towrite += SPA_MAXBLOCKSIZE;
- if (!BP_IS_HOLE(bp))
- txh->txh_space_tounref += SPA_MAXBLOCKSIZE;
- return;
- }
-
- if (dn->dn_maxblkid > 0 && name) {
+ } else {
/*
- * access the name in this fat-zap so that we'll check
- * for i/o errors to the leaf blocks, etc.
+ * Access the name so that we'll check for i/o errors to
+ * the leaf blocks, etc. We ignore ENOENT, as this name
+ * may not yet exist.
*/
- err = zap_lookup(dn->dn_objset, dn->dn_object, name,
- 8, 0, NULL);
- if (err == EIO) {
+ err = zap_lookup_by_dnode(dn, name, 8, 0, NULL);
+ if (err == EIO || err == ECKSUM || err == ENXIO) {
tx->tx_err = err;
- return;
}
}
+}
+
+void
+dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name)
+{
+ dmu_tx_hold_t *txh;
- err = zap_count_write(dn->dn_objset, dn->dn_object, name, add,
- &txh->txh_space_towrite, &txh->txh_space_tooverwrite);
+ ASSERT0(tx->tx_txg);
- /*
- * If the modified blocks are scattered to the four winds,
- * we'll have to modify an indirect twig for each.
- */
- epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
- for (nblocks = dn->dn_maxblkid >> epbs; nblocks != 0; nblocks >>= epbs)
- if (dn->dn_objset->os_dsl_dataset->ds_phys->ds_prev_snap_obj)
- txh->txh_space_towrite += 3 << dn->dn_indblkshift;
- else
- txh->txh_space_tooverwrite += 3 << dn->dn_indblkshift;
+ txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
+ object, THT_ZAP, add, (uintptr_t)name);
+ if (txh != NULL)
+ dmu_tx_hold_zap_impl(txh, name);
+}
+
+void
+dmu_tx_hold_zap_by_dnode(dmu_tx_t *tx, dnode_t *dn, int add, const char *name)
+{
+ dmu_tx_hold_t *txh;
+
+ ASSERT0(tx->tx_txg);
+ ASSERT(dn != NULL);
+
+ txh = dmu_tx_hold_dnode_impl(tx, dn, THT_ZAP, add, (uintptr_t)name);
+ if (txh != NULL)
+ dmu_tx_hold_zap_impl(txh, name);
}
void
}
void
-dmu_tx_hold_space(dmu_tx_t *tx, uint64_t space)
+dmu_tx_hold_bonus_by_dnode(dmu_tx_t *tx, dnode_t *dn)
{
dmu_tx_hold_t *txh;
- ASSERT(tx->tx_txg == 0);
+ ASSERT0(tx->tx_txg);
- txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
- DMU_NEW_OBJECT, THT_SPACE, space, 0);
+ txh = dmu_tx_hold_dnode_impl(tx, dn, THT_BONUS, 0, 0);
if (txh)
- txh->txh_space_towrite += space;
+ dmu_tx_count_dnode(txh);
}
-int
-dmu_tx_holds(dmu_tx_t *tx, uint64_t object)
+void
+dmu_tx_hold_space(dmu_tx_t *tx, uint64_t space)
{
dmu_tx_hold_t *txh;
- int holds = 0;
- /*
- * By asserting that the tx is assigned, we're counting the
- * number of dn_tx_holds, which is the same as the number of
- * dn_holds. Otherwise, we'd be counting dn_holds, but
- * dn_tx_holds could be 0.
- */
- ASSERT(tx->tx_txg != 0);
-
- /* if (tx->tx_anyobj == TRUE) */
- /* return (0); */
-
- for (txh = list_head(&tx->tx_holds); txh;
- txh = list_next(&tx->tx_holds, txh)) {
- if (txh->txh_dnode && txh->txh_dnode->dn_object == object)
- holds++;
- }
+ ASSERT(tx->tx_txg == 0);
- return (holds);
+ txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
+ DMU_NEW_OBJECT, THT_SPACE, space, 0);
+ if (txh)
+ (void) refcount_add_many(&txh->txh_space_towrite, space, FTAG);
}
-#ifdef DEBUG_DMU_TX
+#ifdef ZFS_DEBUG
void
dmu_tx_dirty_buf(dmu_tx_t *tx, dmu_buf_impl_t *db)
{
- dmu_tx_hold_t *txh;
- int match_object = FALSE, match_offset = FALSE;
- dnode_t *dn;
+ boolean_t match_object = B_FALSE;
+ boolean_t match_offset = B_FALSE;
DB_DNODE_ENTER(db);
- dn = DB_DNODE(db);
- ASSERT(dn != NULL);
+ dnode_t *dn = DB_DNODE(db);
ASSERT(tx->tx_txg != 0);
ASSERT(tx->tx_objset == NULL || dn->dn_objset == tx->tx_objset);
ASSERT3U(dn->dn_object, ==, db->db.db_object);
return;
}
- for (txh = list_head(&tx->tx_holds); txh;
+ for (dmu_tx_hold_t *txh = list_head(&tx->tx_holds); txh != NULL;
txh = list_next(&tx->tx_holds, txh)) {
ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg);
if (txh->txh_dnode == dn && txh->txh_type != THT_NEWOBJECT)
match_object = TRUE;
break;
default:
- ASSERT(!"bad txh_type");
+ cmn_err(CE_PANIC, "bad txh_type %d",
+ txh->txh_type);
}
}
if (match_object && match_offset) {
}
#endif
+/*
+ * If we can't do 10 iops, something is wrong. Let us go ahead
+ * and hit zfs_dirty_data_max.
+ */
+hrtime_t zfs_delay_max_ns = 100 * MICROSEC; /* 100 milliseconds */
+int zfs_delay_resolution_ns = 100 * 1000; /* 100 microseconds */
+
+/*
+ * We delay transactions when we've determined that the backend storage
+ * isn't able to accommodate the rate of incoming writes.
+ *
+ * If there is already a transaction waiting, we delay relative to when
+ * that transaction finishes waiting. This way the calculated min_time
+ * is independent of the number of threads concurrently executing
+ * transactions.
+ *
+ * If we are the only waiter, wait relative to when the transaction
+ * started, rather than the current time. This credits the transaction for
+ * "time already served", e.g. reading indirect blocks.
+ *
+ * The minimum time for a transaction to take is calculated as:
+ * min_time = scale * (dirty - min) / (max - dirty)
+ * min_time is then capped at zfs_delay_max_ns.
+ *
+ * The delay has two degrees of freedom that can be adjusted via tunables.
+ * The percentage of dirty data at which we start to delay is defined by
+ * zfs_delay_min_dirty_percent. This should typically be at or above
+ * zfs_vdev_async_write_active_max_dirty_percent so that we only start to
+ * delay after writing at full speed has failed to keep up with the incoming
+ * write rate. The scale of the curve is defined by zfs_delay_scale. Roughly
+ * speaking, this variable determines the amount of delay at the midpoint of
+ * the curve.
+ *
+ * delay
+ * 10ms +-------------------------------------------------------------*+
+ * | *|
+ * 9ms + *+
+ * | *|
+ * 8ms + *+
+ * | * |
+ * 7ms + * +
+ * | * |
+ * 6ms + * +
+ * | * |
+ * 5ms + * +
+ * | * |
+ * 4ms + * +
+ * | * |
+ * 3ms + * +
+ * | * |
+ * 2ms + (midpoint) * +
+ * | | ** |
+ * 1ms + v *** +
+ * | zfs_delay_scale ----------> ******** |
+ * 0 +-------------------------------------*********----------------+
+ * 0% <- zfs_dirty_data_max -> 100%
+ *
+ * Note that since the delay is added to the outstanding time remaining on the
+ * most recent transaction, the delay is effectively the inverse of IOPS.
+ * Here the midpoint of 500us translates to 2000 IOPS. The shape of the curve
+ * was chosen such that small changes in the amount of accumulated dirty data
+ * in the first 3/4 of the curve yield relatively small differences in the
+ * amount of delay.
+ *
+ * The effects can be easier to understand when the amount of delay is
+ * represented on a log scale:
+ *
+ * delay
+ * 100ms +-------------------------------------------------------------++
+ * + +
+ * | |
+ * + *+
+ * 10ms + *+
+ * + ** +
+ * | (midpoint) ** |
+ * + | ** +
+ * 1ms + v **** +
+ * + zfs_delay_scale ----------> ***** +
+ * | **** |
+ * + **** +
+ * 100us + ** +
+ * + * +
+ * | * |
+ * + * +
+ * 10us + * +
+ * + +
+ * | |
+ * + +
+ * +--------------------------------------------------------------+
+ * 0% <- zfs_dirty_data_max -> 100%
+ *
+ * Note here that only as the amount of dirty data approaches its limit does
+ * the delay start to increase rapidly. The goal of a properly tuned system
+ * should be to keep the amount of dirty data out of that range by first
+ * ensuring that the appropriate limits are set for the I/O scheduler to reach
+ * optimal throughput on the backend storage, and then by changing the value
+ * of zfs_delay_scale to increase the steepness of the curve.
+ */
+static void
+dmu_tx_delay(dmu_tx_t *tx, uint64_t dirty)
+{
+ dsl_pool_t *dp = tx->tx_pool;
+ uint64_t delay_min_bytes =
+ zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100;
+ hrtime_t wakeup, min_tx_time, now;
+
+ if (dirty <= delay_min_bytes)
+ return;
+
+ /*
+ * The caller has already waited until we are under the max.
+ * We make them pass us the amount of dirty data so we don't
+ * have to handle the case of it being >= the max, which could
+ * cause a divide-by-zero if it's == the max.
+ */
+ ASSERT3U(dirty, <, zfs_dirty_data_max);
+
+ now = gethrtime();
+ min_tx_time = zfs_delay_scale *
+ (dirty - delay_min_bytes) / (zfs_dirty_data_max - dirty);
+ min_tx_time = MIN(min_tx_time, zfs_delay_max_ns);
+ if (now > tx->tx_start + min_tx_time)
+ return;
+
+ DTRACE_PROBE3(delay__mintime, dmu_tx_t *, tx, uint64_t, dirty,
+ uint64_t, min_tx_time);
+
+ mutex_enter(&dp->dp_lock);
+ wakeup = MAX(tx->tx_start + min_tx_time,
+ dp->dp_last_wakeup + min_tx_time);
+ dp->dp_last_wakeup = wakeup;
+ mutex_exit(&dp->dp_lock);
+
+ zfs_sleep_until(wakeup);
+}
+
+/*
+ * This routine attempts to assign the transaction to a transaction group.
+ * To do so, we must determine if there is sufficient free space on disk.
+ *
+ * If this is a "netfree" transaction (i.e. we called dmu_tx_mark_netfree()
+ * on it), then it is assumed that there is sufficient free space,
+ * unless there's insufficient slop space in the pool (see the comment
+ * above spa_slop_shift in spa_misc.c).
+ *
+ * If it is not a "netfree" transaction, then if the data already on disk
+ * is over the allowed usage (e.g. quota), this will fail with EDQUOT or
+ * ENOSPC. Otherwise, if the current rough estimate of pending changes,
+ * plus the rough estimate of this transaction's changes, may exceed the
+ * allowed usage, then this will fail with ERESTART, which will cause the
+ * caller to wait for the pending changes to be written to disk (by waiting
+ * for the next TXG to open), and then check the space usage again.
+ *
+ * The rough estimate of pending changes is comprised of the sum of:
+ *
+ * - this transaction's holds' txh_space_towrite
+ *
+ * - dd_tempreserved[], which is the sum of in-flight transactions'
+ * holds' txh_space_towrite (i.e. those transactions that have called
+ * dmu_tx_assign() but not yet called dmu_tx_commit()).
+ *
+ * - dd_space_towrite[], which is the amount of dirtied dbufs.
+ *
+ * Note that all of these values are inflated by spa_get_worst_case_asize(),
+ * which means that we may get ERESTART well before we are actually in danger
+ * of running out of space, but this also mitigates any small inaccuracies
+ * in the rough estimate (e.g. txh_space_towrite doesn't take into account
+ * indirect blocks, and dd_space_towrite[] doesn't take into account changes
+ * to the MOS).
+ *
+ * Note that due to this algorithm, it is possible to exceed the allowed
+ * usage by one transaction. Also, as we approach the allowed usage,
+ * we will allow a very limited amount of changes into each TXG, thus
+ * decreasing performance.
+ */
static int
-dmu_tx_try_assign(dmu_tx_t *tx, txg_how_t txg_how)
+dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how)
{
- dmu_tx_hold_t *txh;
spa_t *spa = tx->tx_pool->dp_spa;
- uint64_t memory, asize, fsize, usize;
- uint64_t towrite, tofree, tooverwrite, tounref, tohold, fudge;
ASSERT0(tx->tx_txg);
* of the failuremode setting.
*/
if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE &&
- txg_how != TXG_WAIT)
+ !(txg_how & TXG_WAIT))
return (SET_ERROR(EIO));
return (SET_ERROR(ERESTART));
}
+ if (!tx->tx_dirty_delayed &&
+ dsl_pool_need_dirty_delay(tx->tx_pool)) {
+ tx->tx_wait_dirty = B_TRUE;
+ DMU_TX_STAT_BUMP(dmu_tx_dirty_delay);
+ return (ERESTART);
+ }
+
tx->tx_txg = txg_hold_open(tx->tx_pool, &tx->tx_txgh);
tx->tx_needassign_txh = NULL;
* dmu_tx_unassign() logic.
*/
- towrite = tofree = tooverwrite = tounref = tohold = fudge = 0;
- for (txh = list_head(&tx->tx_holds); txh;
+ uint64_t towrite = 0;
+ uint64_t tohold = 0;
+ for (dmu_tx_hold_t *txh = list_head(&tx->tx_holds); txh != NULL;
txh = list_next(&tx->tx_holds, txh)) {
dnode_t *dn = txh->txh_dnode;
if (dn != NULL) {
(void) refcount_add(&dn->dn_tx_holds, tx);
mutex_exit(&dn->dn_mtx);
}
- towrite += txh->txh_space_towrite;
- tofree += txh->txh_space_tofree;
- tooverwrite += txh->txh_space_tooverwrite;
- tounref += txh->txh_space_tounref;
- tohold += txh->txh_memory_tohold;
- fudge += txh->txh_fudge;
- }
-
- /*
- * If a snapshot has been taken since we made our estimates,
- * assume that we won't be able to free or overwrite anything.
- */
- if (tx->tx_objset &&
- dsl_dataset_prev_snap_txg(tx->tx_objset->os_dsl_dataset) >
- tx->tx_lastsnap_txg) {
- towrite += tooverwrite;
- tooverwrite = tofree = 0;
+ towrite += refcount_count(&txh->txh_space_towrite);
+ tohold += refcount_count(&txh->txh_memory_tohold);
}
/* needed allocation: worst-case estimate of write space */
- asize = spa_get_asize(tx->tx_pool->dp_spa, towrite + tooverwrite);
- /* freed space estimate: worst-case overwrite + free estimate */
- fsize = spa_get_asize(tx->tx_pool->dp_spa, tooverwrite) + tofree;
- /* convert unrefd space to worst-case estimate */
- usize = spa_get_asize(tx->tx_pool->dp_spa, tounref);
+ uint64_t asize = spa_get_worst_case_asize(tx->tx_pool->dp_spa, towrite);
/* calculate memory footprint estimate */
- memory = towrite + tooverwrite + tohold;
-
-#ifdef DEBUG_DMU_TX
- /*
- * Add in 'tohold' to account for our dirty holds on this memory
- * XXX - the "fudge" factor is to account for skipped blocks that
- * we missed because dnode_next_offset() misses in-core-only blocks.
- */
- tx->tx_space_towrite = asize +
- spa_get_asize(tx->tx_pool->dp_spa, tohold + fudge);
- tx->tx_space_tofree = tofree;
- tx->tx_space_tooverwrite = tooverwrite;
- tx->tx_space_tounref = tounref;
-#endif
+ uint64_t memory = towrite + tohold;
- if (tx->tx_dir && asize != 0) {
+ if (tx->tx_dir != NULL && asize != 0) {
int err = dsl_dir_tempreserve_space(tx->tx_dir, memory,
- asize, fsize, usize, &tx->tx_tempreserve_cookie, tx);
- if (err)
+ asize, tx->tx_netfree, &tx->tx_tempreserve_cookie, tx);
+ if (err != 0)
return (err);
}
static void
dmu_tx_unassign(dmu_tx_t *tx)
{
- dmu_tx_hold_t *txh;
-
if (tx->tx_txg == 0)
return;
* Walk the transaction's hold list, removing the hold on the
* associated dnode, and notifying waiters if the refcount drops to 0.
*/
- for (txh = list_head(&tx->tx_holds); txh != tx->tx_needassign_txh;
+ for (dmu_tx_hold_t *txh = list_head(&tx->tx_holds);
+ txh && txh != tx->tx_needassign_txh;
txh = list_next(&tx->tx_holds, txh)) {
dnode_t *dn = txh->txh_dnode;
}
/*
- * Assign tx to a transaction group. txg_how can be one of:
+ * Assign tx to a transaction group; txg_how is a bitmask:
+ *
+ * If TXG_WAIT is set and the currently open txg is full, this function
+ * will wait until there's a new txg. This should be used when no locks
+ * are being held. With this bit set, this function will only fail if
+ * we're truly out of space (or over quota).
*
- * (1) TXG_WAIT. If the current open txg is full, waits until there's
- * a new one. This should be used when you're not holding locks.
- * It will only fail if we're truly out of space (or over quota).
+ * If TXG_WAIT is *not* set and we can't assign into the currently open
+ * txg without blocking, this function will return immediately with
+ * ERESTART. This should be used whenever locks are being held. On an
+ * ERESTART error, the caller should drop all locks, call dmu_tx_wait(),
+ * and try again.
*
- * (2) TXG_NOWAIT. If we can't assign into the current open txg without
- * blocking, returns immediately with ERESTART. This should be used
- * whenever you're holding locks. On an ERESTART error, the caller
- * should drop locks, do a dmu_tx_wait(tx), and try again.
+ * If TXG_NOTHROTTLE is set, this indicates that this tx should not be
+ * delayed due on the ZFS Write Throttle (see comments in dsl_pool.c for
+ * details on the throttle). This is used by the VFS operations, after
+ * they have already called dmu_tx_wait() (though most likely on a
+ * different tx).
*/
int
-dmu_tx_assign(dmu_tx_t *tx, txg_how_t txg_how)
+dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how)
{
- hrtime_t before;
int err;
ASSERT(tx->tx_txg == 0);
- ASSERT(txg_how == TXG_WAIT || txg_how == TXG_NOWAIT);
+ ASSERT0(txg_how & ~(TXG_WAIT | TXG_NOTHROTTLE));
ASSERT(!dsl_pool_sync_context(tx->tx_pool));
- before = gethrtime();
-
/* If we might wait, we must not hold the config lock. */
- ASSERT(txg_how != TXG_WAIT || !dsl_pool_config_held(tx->tx_pool));
+ IMPLY((txg_how & TXG_WAIT), !dsl_pool_config_held(tx->tx_pool));
+
+ if ((txg_how & TXG_NOTHROTTLE))
+ tx->tx_dirty_delayed = B_TRUE;
while ((err = dmu_tx_try_assign(tx, txg_how)) != 0) {
dmu_tx_unassign(tx);
- if (err != ERESTART || txg_how != TXG_WAIT)
+ if (err != ERESTART || !(txg_how & TXG_WAIT))
return (err);
dmu_tx_wait(tx);
txg_rele_to_quiesce(&tx->tx_txgh);
- spa_tx_assign_add_nsecs(tx->tx_pool->dp_spa, gethrtime() - before);
-
return (0);
}
dmu_tx_wait(dmu_tx_t *tx)
{
spa_t *spa = tx->tx_pool->dp_spa;
+ dsl_pool_t *dp = tx->tx_pool;
+ hrtime_t before;
ASSERT(tx->tx_txg == 0);
ASSERT(!dsl_pool_config_held(tx->tx_pool));
- /*
- * It's possible that the pool has become active after this thread
- * has tried to obtain a tx. If that's the case then his
- * tx_lasttried_txg would not have been assigned.
- */
- if (spa_suspended(spa) || tx->tx_lasttried_txg == 0) {
- txg_wait_synced(tx->tx_pool, spa_last_synced_txg(spa) + 1);
+ before = gethrtime();
+
+ if (tx->tx_wait_dirty) {
+ uint64_t dirty;
+
+ /*
+ * dmu_tx_try_assign() has determined that we need to wait
+ * because we've consumed much or all of the dirty buffer
+ * space.
+ */
+ mutex_enter(&dp->dp_lock);
+ if (dp->dp_dirty_total >= zfs_dirty_data_max)
+ DMU_TX_STAT_BUMP(dmu_tx_dirty_over_max);
+ while (dp->dp_dirty_total >= zfs_dirty_data_max)
+ cv_wait(&dp->dp_spaceavail_cv, &dp->dp_lock);
+ dirty = dp->dp_dirty_total;
+ mutex_exit(&dp->dp_lock);
+
+ dmu_tx_delay(tx, dirty);
+
+ tx->tx_wait_dirty = B_FALSE;
+
+ /*
+ * Note: setting tx_dirty_delayed only has effect if the
+ * caller used TX_WAIT. Otherwise they are going to
+ * destroy this tx and try again. The common case,
+ * zfs_write(), uses TX_WAIT.
+ */
+ tx->tx_dirty_delayed = B_TRUE;
+ } else if (spa_suspended(spa) || tx->tx_lasttried_txg == 0) {
+ /*
+ * If the pool is suspended we need to wait until it
+ * is resumed. Note that it's possible that the pool
+ * has become active after this thread has tried to
+ * obtain a tx. If that's the case then tx_lasttried_txg
+ * would not have been set.
+ */
+ txg_wait_synced(dp, spa_last_synced_txg(spa) + 1);
} else if (tx->tx_needassign_txh) {
dnode_t *dn = tx->tx_needassign_txh->txh_dnode;
mutex_exit(&dn->dn_mtx);
tx->tx_needassign_txh = NULL;
} else {
+ /*
+ * A dnode is assigned to the quiescing txg. Wait for its
+ * transaction to complete.
+ */
txg_wait_open(tx->tx_pool, tx->tx_lasttried_txg + 1);
}
+
+ spa_tx_assign_add_nsecs(spa, gethrtime() - before);
}
-void
-dmu_tx_willuse_space(dmu_tx_t *tx, int64_t delta)
+static void
+dmu_tx_destroy(dmu_tx_t *tx)
{
-#ifdef DEBUG_DMU_TX
- if (tx->tx_dir == NULL || delta == 0)
- return;
+ dmu_tx_hold_t *txh;
- if (delta > 0) {
- ASSERT3U(refcount_count(&tx->tx_space_written) + delta, <=,
- tx->tx_space_towrite);
- (void) refcount_add_many(&tx->tx_space_written, delta, NULL);
- } else {
- (void) refcount_add_many(&tx->tx_space_freed, -delta, NULL);
+ while ((txh = list_head(&tx->tx_holds)) != NULL) {
+ dnode_t *dn = txh->txh_dnode;
+
+ list_remove(&tx->tx_holds, txh);
+ refcount_destroy_many(&txh->txh_space_towrite,
+ refcount_count(&txh->txh_space_towrite));
+ refcount_destroy_many(&txh->txh_memory_tohold,
+ refcount_count(&txh->txh_memory_tohold));
+ kmem_free(txh, sizeof (dmu_tx_hold_t));
+ if (dn != NULL)
+ dnode_rele(dn, tx);
}
-#endif
+
+ list_destroy(&tx->tx_callbacks);
+ list_destroy(&tx->tx_holds);
+ kmem_free(tx, sizeof (dmu_tx_t));
}
void
* Go through the transaction's hold list and remove holds on
* associated dnodes, notifying waiters if no holds remain.
*/
- while ((txh = list_head(&tx->tx_holds))) {
+ for (txh = list_head(&tx->tx_holds); txh != NULL;
+ txh = list_next(&tx->tx_holds, txh)) {
dnode_t *dn = txh->txh_dnode;
- list_remove(&tx->tx_holds, txh);
- kmem_free(txh, sizeof (dmu_tx_hold_t));
if (dn == NULL)
continue;
+
mutex_enter(&dn->dn_mtx);
ASSERT3U(dn->dn_assigned_txg, ==, tx->tx_txg);
cv_broadcast(&dn->dn_notxholds);
}
mutex_exit(&dn->dn_mtx);
- dnode_rele(dn, tx);
}
if (tx->tx_tempreserve_cookie)
if (tx->tx_anyobj == FALSE)
txg_rele_to_sync(&tx->tx_txgh);
- list_destroy(&tx->tx_callbacks);
- list_destroy(&tx->tx_holds);
-#ifdef DEBUG_DMU_TX
- dprintf("towrite=%llu written=%llu tofree=%llu freed=%llu\n",
- tx->tx_space_towrite, refcount_count(&tx->tx_space_written),
- tx->tx_space_tofree, refcount_count(&tx->tx_space_freed));
- refcount_destroy_many(&tx->tx_space_written,
- refcount_count(&tx->tx_space_written));
- refcount_destroy_many(&tx->tx_space_freed,
- refcount_count(&tx->tx_space_freed));
-#endif
- kmem_free(tx, sizeof (dmu_tx_t));
+ dmu_tx_destroy(tx);
}
void
dmu_tx_abort(dmu_tx_t *tx)
{
- dmu_tx_hold_t *txh;
-
ASSERT(tx->tx_txg == 0);
- while ((txh = list_head(&tx->tx_holds))) {
- dnode_t *dn = txh->txh_dnode;
-
- list_remove(&tx->tx_holds, txh);
- kmem_free(txh, sizeof (dmu_tx_hold_t));
- if (dn != NULL)
- dnode_rele(dn, tx);
- }
-
/*
* Call any registered callbacks with an error code.
*/
if (!list_is_empty(&tx->tx_callbacks))
dmu_tx_do_callbacks(&tx->tx_callbacks, ECANCELED);
- list_destroy(&tx->tx_callbacks);
- list_destroy(&tx->tx_holds);
-#ifdef DEBUG_DMU_TX
- refcount_destroy_many(&tx->tx_space_written,
- refcount_count(&tx->tx_space_written));
- refcount_destroy_many(&tx->tx_space_freed,
- refcount_count(&tx->tx_space_freed));
-#endif
- kmem_free(tx, sizeof (dmu_tx_t));
+ dmu_tx_destroy(tx);
}
uint64_t
return (tx->tx_pool);
}
-
void
dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *func, void *data)
{
dmu_tx_callback_t *dcb;
- dcb = kmem_alloc(sizeof (dmu_tx_callback_t), KM_PUSHPAGE);
+ dcb = kmem_alloc(sizeof (dmu_tx_callback_t), KM_SLEEP);
dcb->dcb_func = func;
dcb->dcb_data = data;
{
dmu_tx_callback_t *dcb;
- while ((dcb = list_head(cb_list))) {
+ while ((dcb = list_tail(cb_list)) != NULL) {
list_remove(cb_list, dcb);
dcb->dcb_func(dcb->dcb_data, error);
kmem_free(dcb, sizeof (dmu_tx_callback_t));
static void
dmu_tx_sa_registration_hold(sa_os_t *sa, dmu_tx_t *tx)
{
- int i;
-
if (!sa->sa_need_attr_registration)
return;
- for (i = 0; i != sa->sa_num_attrs; i++) {
+ for (int i = 0; i != sa->sa_num_attrs; i++) {
if (!sa->sa_attr_table[i].sa_registered) {
if (sa->sa_reg_attr_obj)
dmu_tx_hold_zap(tx, sa->sa_reg_attr_obj,
}
}
-
void
dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object)
{
- dnode_t *dn;
dmu_tx_hold_t *txh;
txh = dmu_tx_hold_object_impl(tx, tx->tx_objset, object,
THT_SPILL, 0, 0);
- if (txh == NULL)
- return;
-
- dn = txh->txh_dnode;
-
- if (dn == NULL)
- return;
-
- /* If blkptr doesn't exist then add space to towrite */
- if (!(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
- txh->txh_space_towrite += SPA_MAXBLOCKSIZE;
- } else {
- blkptr_t *bp;
-
- bp = &dn->dn_phys->dn_spill;
- if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset,
- bp, bp->blk_birth))
- txh->txh_space_tooverwrite += SPA_MAXBLOCKSIZE;
- else
- txh->txh_space_towrite += SPA_MAXBLOCKSIZE;
- if (!BP_IS_HOLE(bp))
- txh->txh_space_tounref += SPA_MAXBLOCKSIZE;
- }
+ if (txh != NULL)
+ (void) refcount_add_many(&txh->txh_space_towrite,
+ SPA_OLD_MAXBLOCKSIZE, FTAG);
}
void
if (tx->tx_objset->os_sa->sa_master_obj == 0)
return;
- if (tx->tx_objset->os_sa->sa_layout_attr_obj)
+ if (tx->tx_objset->os_sa->sa_layout_attr_obj) {
dmu_tx_hold_zap(tx, sa->sa_layout_attr_obj, B_TRUE, NULL);
- else {
+ } else {
dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_LAYOUTS);
dmu_tx_hold_zap(tx, sa->sa_master_obj, B_TRUE, SA_REGISTRY);
dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
dmu_tx_sa_registration_hold(sa, tx);
- if (attrsize <= DN_MAX_BONUSLEN && !sa->sa_force_spill)
+ if (attrsize <= DN_OLD_MAX_BONUSLEN && !sa->sa_force_spill)
return;
(void) dmu_tx_hold_object_impl(tx, tx->tx_objset, DMU_NEW_OBJECT,
#if defined(_KERNEL) && defined(HAVE_SPL)
EXPORT_SYMBOL(dmu_tx_create);
EXPORT_SYMBOL(dmu_tx_hold_write);
+EXPORT_SYMBOL(dmu_tx_hold_write_by_dnode);
EXPORT_SYMBOL(dmu_tx_hold_free);
+EXPORT_SYMBOL(dmu_tx_hold_free_by_dnode);
EXPORT_SYMBOL(dmu_tx_hold_zap);
+EXPORT_SYMBOL(dmu_tx_hold_zap_by_dnode);
EXPORT_SYMBOL(dmu_tx_hold_bonus);
+EXPORT_SYMBOL(dmu_tx_hold_bonus_by_dnode);
EXPORT_SYMBOL(dmu_tx_abort);
EXPORT_SYMBOL(dmu_tx_assign);
EXPORT_SYMBOL(dmu_tx_wait);
EXPORT_SYMBOL(dmu_tx_commit);
+EXPORT_SYMBOL(dmu_tx_mark_netfree);
EXPORT_SYMBOL(dmu_tx_get_txg);
EXPORT_SYMBOL(dmu_tx_callback_register);
EXPORT_SYMBOL(dmu_tx_do_callbacks);