/*
* Return nonzero on i/o error.
* Return new buf to write out in *bufp.
*/
static int
dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype,
dnode_phys_t *dnp, const blkptr_t *bp,
const zbookmark_t *zb, dmu_tx_t *tx, arc_buf_t **bufp)
{
dsl_pool_t *dp = scn->scn_dp;
int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD;
int err;
if (BP_GET_LEVEL(bp) > 0) {
uint32_t flags = ARC_WAIT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
}
for (i = 0, cbp = (*bufp)->b_data; i < epb; i++, cbp++) {
dsl_scan_prefetch(scn, *bufp, cbp, zb->zb_objset,
zb->zb_object, zb->zb_blkid * epb + i);
}
for (i = 0, cbp = (*bufp)->b_data; i < epb; i++, cbp++) {
zbookmark_t czb;
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
dsl_scan_visitbp(cbp, &czb, dnp,
*bufp, ds, scn, ostype, tx);
}
} else if (BP_GET_TYPE(bp) == DMU_OT_USERGROUP_USED) {
/* ARGSUSED */
static void
ready(zio_t *zio, arc_buf_t *abuf, void *arg)
{
blkptr_t *bp = zio->io_bp;
blkptr_t *bp_orig = &zio->io_bp_orig;
objset_impl_t *os = arg;
dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
ASSERT(bp == os->os_rootbp);
ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
ASSERT(BP_GET_LEVEL(bp) == 0);
/*
* Update rootbp fill count.
*/
bp->blk_fill = 1; /* count the meta-dnode */
for (int i = 0; i < dnp->dn_nblkptr; i++)
bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
ASSERT(DVA_EQUAL(BP_IDENTITY(bp), BP_IDENTITY(bp_orig)));
} else {
if (zio->io_bp_orig.blk_birth == os->os_synctx->tx_txg)
(void) dsl_dataset_block_kill(os->os_dsl_dataset,
&zio->io_bp_orig, zio, os->os_synctx);
dsl_dataset_block_born(os->os_dsl_dataset, bp, os->os_synctx);
}
}
/* ARGSUSED */
static void
ready(zio_t *zio, arc_buf_t *abuf, void *arg)
{
int i;
blkptr_t *bp = zio->io_bp;
objset_impl_t *os = arg;
dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
ASSERTV(blkptr_t *bp_orig = &zio->io_bp_orig);
ASSERT(bp == os->os_rootbp);
ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
ASSERT(BP_GET_LEVEL(bp) == 0);
/*
* Update rootbp fill count: it should be the number of objects
* allocated in the object set (not counting the "special"
* objects that are stored in the objset_phys_t -- the meta
* dnode and user/group accounting objects).
*/
bp->blk_fill = 0;
for (i = 0; i < dnp->dn_nblkptr; i++)
bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
ASSERT(DVA_EQUAL(BP_IDENTITY(bp), BP_IDENTITY(bp_orig)));
} else {
if (zio->io_bp_orig.blk_birth == os->os_synctx->tx_txg)
(void) dsl_dataset_block_kill(os->os_dsl_dataset,
&zio->io_bp_orig, zio, os->os_synctx);
dsl_dataset_block_born(os->os_dsl_dataset, bp, os->os_synctx);
}
}
/* ARGSUSED */
static void
dsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp,
uint64_t objset, uint64_t object, uint64_t blkid)
{
zbookmark_t czb;
uint32_t flags = ARC_NOWAIT | ARC_PREFETCH;
if (zfs_no_scrub_prefetch)
return;
if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg ||
(BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE))
return;
SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid);
(void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp,
NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb);
}
static void
free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx)
{
dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
uint64_t bytesfreed = 0;
int i;
dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num);
for (i = 0; i < num; i++, bp++) {
uint64_t lsize, lvl;
dmu_object_type_t type;
if (BP_IS_HOLE(bp))
continue;
bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE);
ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys));
/*
* Save some useful information on the holes being
* punched, including logical size, type, and indirection
* level. Retaining birth time enables detection of when
* holes are punched for reducing the number of free
* records transmitted during a zfs send.
*/
lsize = BP_GET_LSIZE(bp);
type = BP_GET_TYPE(bp);
lvl = BP_GET_LEVEL(bp);
bzero(bp, sizeof (blkptr_t));
if (spa_feature_is_active(dn->dn_objset->os_spa,
SPA_FEATURE_HOLE_BIRTH)) {
BP_SET_LSIZE(bp, lsize);
BP_SET_TYPE(bp, type);
BP_SET_LEVEL(bp, lvl);
BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0);
}
}
dnode_diduse_space(dn, -bytesfreed);
}
/* ARGSUSED */
static void
dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
{
blkptr_t *bp = zio->io_bp;
objset_t *os = arg;
dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
ASSERT3P(bp, ==, os->os_rootbp);
ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
ASSERT0(BP_GET_LEVEL(bp));
/*
* Update rootbp fill count: it should be the number of objects
* allocated in the object set (not counting the "special"
* objects that are stored in the objset_phys_t -- the meta
* dnode and user/group accounting objects).
*/
bp->blk_fill = 0;
for (int i = 0; i < dnp->dn_nblkptr; i++)
bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
}
/*
* Function called when a log block write completes
*/
static void
zil_lwb_write_done(zio_t *zio)
{
lwb_t *lwb = zio->io_private;
zilog_t *zilog = lwb->lwb_zilog;
ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
ASSERT(BP_GET_CHECKSUM(zio->io_bp) == ZIO_CHECKSUM_ZILOG);
ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG);
ASSERT(BP_GET_LEVEL(zio->io_bp) == 0);
ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER);
ASSERT(!BP_IS_GANG(zio->io_bp));
ASSERT(!BP_IS_HOLE(zio->io_bp));
ASSERT(zio->io_bp->blk_fill == 0);
/*
* Ensure the lwb buffer pointer is cleared before releasing
* the txg. If we have had an allocation failure and
* the txg is waiting to sync then we want want zil_sync()
* to remove the lwb so that it's not picked up as the next new
* one in zil_commit_writer(). zil_sync() will only remove
* the lwb if lwb_buf is null.
*/
zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
mutex_enter(&zilog->zl_lock);
lwb->lwb_buf = NULL;
if (zio->io_error)
zilog->zl_log_error = B_TRUE;
/*
* Now that we've written this log block, we have a stable pointer
* to the next block in the chain, so it's OK to let the txg in
* which we allocated the next block sync. We still have the
* zl_lock to ensure zil_sync doesn't kmem free the lwb.
*/
txg_rele_to_sync(&lwb->lwb_txgh);
mutex_exit(&zilog->zl_lock);
}
static void
traverse_prefetch_metadata(traverse_data_t *td,
const blkptr_t *bp, const zbookmark_t *zb)
{
uint32_t flags = ARC_NOWAIT | ARC_PREFETCH;
if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
return;
/*
* If we are in the process of resuming, don't prefetch, because
* some children will not be needed (and in fact may have already
* been freed).
*/
if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
return;
if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
return;
if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
return;
(void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
}
static int
traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
{
zbookmark_t czb;
int err = 0, lasterr = 0;
arc_buf_t *buf = NULL;
prefetch_data_t *pd = td->td_pfd;
boolean_t hard = td->td_flags & TRAVERSE_HARD;
if (bp->blk_birth == 0) {
err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp,
td->td_arg);
return (err);
}
if (bp->blk_birth <= td->td_min_txg)
return (0);
if (pd && !pd->pd_exited &&
((pd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
mutex_enter(&pd->pd_mtx);
ASSERT(pd->pd_blks_fetched >= 0);
while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
cv_wait(&pd->pd_cv, &pd->pd_mtx);
pd->pd_blks_fetched--;
cv_broadcast(&pd->pd_cv);
mutex_exit(&pd->pd_mtx);
}
if (td->td_flags & TRAVERSE_PRE) {
err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
td->td_arg);
if (err == TRAVERSE_VISIT_NO_CHILDREN)
return (0);
if (err)
return (err);
}
if (BP_GET_LEVEL(bp) > 0) {
uint32_t flags = ARC_WAIT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
err = dsl_read(NULL, td->td_spa, bp, pbuf,
arc_getbuf_func, &buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
if (err)
return (err);
/* recursively visitbp() blocks below this */
cbp = buf->b_data;
for (i = 0; i < epb; i++, cbp++) {
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
err = traverse_visitbp(td, dnp, buf, cbp, &czb);
if (err) {
if (!hard)
break;
lasterr = err;
}
}
} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
static int
traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
const blkptr_t *bp, const zbookmark_t *zb)
{
zbookmark_t czb;
int err = 0, lasterr = 0;
arc_buf_t *buf = NULL;
prefetch_data_t *pd = td->td_pfd;
boolean_t hard = td->td_flags & TRAVERSE_HARD;
boolean_t pause = B_FALSE;
switch (resume_skip_check(td, dnp, zb)) {
case RESUME_SKIP_ALL:
return (0);
case RESUME_SKIP_CHILDREN:
goto post;
case RESUME_SKIP_NONE:
break;
default:
ASSERT(0);
}
if (BP_IS_HOLE(bp)) {
err = td->td_func(td->td_spa, NULL, NULL, zb, dnp, td->td_arg);
return (err);
}
if (bp->blk_birth <= td->td_min_txg)
return (0);
if (pd && !pd->pd_exited &&
((pd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
mutex_enter(&pd->pd_mtx);
ASSERT(pd->pd_blks_fetched >= 0);
while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
cv_wait(&pd->pd_cv, &pd->pd_mtx);
pd->pd_blks_fetched--;
cv_broadcast(&pd->pd_cv);
mutex_exit(&pd->pd_mtx);
}
if (td->td_flags & TRAVERSE_PRE) {
err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
td->td_arg);
if (err == TRAVERSE_VISIT_NO_CHILDREN)
return (0);
if (err == ERESTART)
pause = B_TRUE; /* handle pausing at a common point */
if (err != 0)
goto post;
}
if (BP_GET_LEVEL(bp) > 0) {
uint32_t flags = ARC_WAIT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
if (err)
return (err);
cbp = buf->b_data;
for (i = 0; i < epb; i++) {
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
traverse_prefetch_metadata(td, &cbp[i], &czb);
}
/* recursively visitbp() blocks below this */
for (i = 0; i < epb; i++) {
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
err = traverse_visitbp(td, dnp, &cbp[i], &czb);
if (err) {
if (!hard)
break;
lasterr = err;
}
}
} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
static int
traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
const blkptr_t *bp, const zbookmark_phys_t *zb)
{
zbookmark_phys_t czb;
int err = 0;
arc_buf_t *buf = NULL;
prefetch_data_t *pd = td->td_pfd;
boolean_t hard = td->td_flags & TRAVERSE_HARD;
switch (resume_skip_check(td, dnp, zb)) {
case RESUME_SKIP_ALL:
return (0);
case RESUME_SKIP_CHILDREN:
goto post;
case RESUME_SKIP_NONE:
break;
default:
ASSERT(0);
}
if (bp->blk_birth == 0) {
/*
* Since this block has a birth time of 0 it must be one of
* two things: a hole created before the
* SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole
* which has always been a hole in an object.
*
* If a file is written sparsely, then the unwritten parts of
* the file were "always holes" -- that is, they have been
* holes since this object was allocated. However, we (and
* our callers) can not necessarily tell when an object was
* allocated. Therefore, if it's possible that this object
* was freed and then its object number reused, we need to
* visit all the holes with birth==0.
*
* If it isn't possible that the object number was reused,
* then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote
* all the blocks we will visit as part of this traversal,
* then this hole must have always existed, so we can skip
* it. We visit blocks born after (exclusive) td_min_txg.
*
* Note that the meta-dnode cannot be reallocated.
*/
if (!send_holes_without_birth_time &&
(!td->td_realloc_possible ||
zb->zb_object == DMU_META_DNODE_OBJECT) &&
td->td_hole_birth_enabled_txg <= td->td_min_txg)
return (0);
} else if (bp->blk_birth <= td->td_min_txg) {
return (0);
}
if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) {
uint64_t size = BP_GET_LSIZE(bp);
mutex_enter(&pd->pd_mtx);
ASSERT(pd->pd_bytes_fetched >= 0);
while (pd->pd_bytes_fetched < size && !pd->pd_exited)
cv_wait(&pd->pd_cv, &pd->pd_mtx);
pd->pd_bytes_fetched -= size;
cv_broadcast(&pd->pd_cv);
mutex_exit(&pd->pd_mtx);
}
if (BP_IS_HOLE(bp)) {
err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
if (err != 0)
goto post;
return (0);
}
if (td->td_flags & TRAVERSE_PRE) {
err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
td->td_arg);
if (err == TRAVERSE_VISIT_NO_CHILDREN)
return (0);
if (err != 0)
goto post;
}
if (BP_GET_LEVEL(bp) > 0) {
arc_flags_t flags = ARC_FLAG_WAIT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
if (err != 0)
goto post;
cbp = buf->b_data;
for (i = 0; i < epb; i++) {
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
traverse_prefetch_metadata(td, &cbp[i], &czb);
}
/* recursively visitbp() blocks below this */
for (i = 0; i < epb; i++) {
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
err = traverse_visitbp(td, dnp, &cbp[i], &czb);
if (err != 0)
break;
}
} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
static int
traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
const blkptr_t *bp, const zbookmark_phys_t *zb)
{
zbookmark_phys_t czb;
int err = 0;
arc_buf_t *buf = NULL;
prefetch_data_t *pd = td->td_pfd;
boolean_t hard = td->td_flags & TRAVERSE_HARD;
switch (resume_skip_check(td, dnp, zb)) {
case RESUME_SKIP_ALL:
return (0);
case RESUME_SKIP_CHILDREN:
goto post;
case RESUME_SKIP_NONE:
break;
default:
ASSERT(0);
}
if (bp->blk_birth == 0) {
/*
* Since this block has a birth time of 0 it must be a
* hole created before the SPA_FEATURE_HOLE_BIRTH
* feature was enabled. If SPA_FEATURE_HOLE_BIRTH
* was enabled before the min_txg for this traveral we
* know the hole must have been created before the
* min_txg for this traveral, so we can skip it. If
* SPA_FEATURE_HOLE_BIRTH was enabled after the min_txg
* for this traveral we cannot tell if the hole was
* created before or after the min_txg for this
* traversal, so we cannot skip it.
*/
if (td->td_hole_birth_enabled_txg < td->td_min_txg)
return (0);
} else if (bp->blk_birth <= td->td_min_txg) {
return (0);
}
if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) {
mutex_enter(&pd->pd_mtx);
ASSERT(pd->pd_blks_fetched >= 0);
while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
cv_wait(&pd->pd_cv, &pd->pd_mtx);
pd->pd_blks_fetched--;
cv_broadcast(&pd->pd_cv);
mutex_exit(&pd->pd_mtx);
}
if (BP_IS_HOLE(bp)) {
err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
if (err != 0)
goto post;
return (0);
}
if (td->td_flags & TRAVERSE_PRE) {
err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
td->td_arg);
if (err == TRAVERSE_VISIT_NO_CHILDREN)
return (0);
if (err != 0)
goto post;
}
if (BP_GET_LEVEL(bp) > 0) {
arc_flags_t flags = ARC_FLAG_WAIT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
if (err != 0)
goto post;
cbp = buf->b_data;
for (i = 0; i < epb; i++) {
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
traverse_prefetch_metadata(td, &cbp[i], &czb);
}
/* recursively visitbp() blocks below this */
for (i = 0; i < epb; i++) {
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
err = traverse_visitbp(td, dnp, &cbp[i], &czb);
if (err != 0)
break;
}
} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
