c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-01-13 08:43:06 +03:00
Code Issues Packages Projects Releases Wiki Activity

dnode_next_offset: backtrack if lower level does not match

Browse Source 
This changes the basic search algorithm from a single search up and down
the tree to a full depth-first traversal to handle conditions where the
tree matches at a higher level but not a lower level.

Normally higher level blocks always point to matching blocks, but there
are cases where this does not happen:

1. Racing block pointer updates from dbuf_write_ready.

   Before f664f1ee7f (#8946), both dbuf_write_ready and
   dnode_next_offset held dn_struct_rwlock which protected against
   pointer writes from concurrent syncs.

   This no longer applies, so sync context can f.e. clear or fill all
   L1->L0 BPs before the L2->L1 BP and higher BP's are updated.

   dnode_free_range in particular can reach this case and skip over L1
   blocks that need to be dirtied. Later, sync will panic in
   free_children when trying to clear a non-dirty indirect block.

   This case was found with ztest.

2. txg > 0, non-hole case. This is #11196.

   Freeing blocks/dnodes breaks the assumption that a match at a higher
   level implies a match at a lower level when filtering txg > 0.

   Whenever some but not all L0 blocks are freed, the parent L1 block is
   rewritten. Its updated L2->L1 BP reflects a newer birth txg.

   Later when searching by txg, if the L1 block matches since the txg is
   newer, it is possible that none of the remaining L1->L0 BPs match if
   none have been updated.

   The same behavior is possible with dnode search at L0.

   This is reachable from dsl_destroy_head for synchronous freeing.
   When this happens open context fails to free objects leaving sync
   context stuck freeing potentially many objects.

   This is also reachable from traverse_pool for extreme rewind where it
   is theoretically possible that datasets not dirtied after txg are
   skipped if the MOS has high enough indirection to trigger this case.

In both of these cases, without backtracking the search ends prematurely
as ESRCH result implies no more matches in the entire object.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Akash B <akash-b@hpe.com>
Signed-off-by: Robert Evans <evansr@google.com>
Closes #16025
Closes #11196
This commit is contained in:
Robert Evans 2025-09-25 14:06:28 -04:00 committed by Tony Hutter
parent fb994a3cde
commit fdc683e863
1 changed files with 54 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
module/zfs/dnode.c
View File

@ -2615,6 +2615,32 @@ dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
return (error);
}
/*
* Adjust *offset to the next (or previous) block byte offset at lvl.
* Returns FALSE if *offset would overflow or underflow.
*/
static boolean_t
dnode_next_block(dnode_t *dn, int flags, uint64_t *offset, int lvl)
{
int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
int span = lvl * epbs + dn->dn_datablkshift;
uint64_t blkid, maxblkid;
if (span >= 8 * sizeof (uint64_t))
return (B_FALSE);
blkid = *offset >> span;
maxblkid = 1ULL << (8 * sizeof (*offset) - span);
if (!(flags & DNODE_FIND_BACKWARDS) && blkid + 1 < maxblkid)
*offset = (blkid + 1) << span;
else if ((flags & DNODE_FIND_BACKWARDS) && blkid > 0)
*offset = (blkid << span) - 1;
else
return (B_FALSE);
return (B_TRUE);
}
/*
* Find the next hole, data, or sparse region at or after *offset.
* The value 'blkfill' tells us how many items we expect to find
@ -2642,7 +2668,7 @@ int
dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset,
int minlvl, uint64_t blkfill, uint64_t txg)
{
uint64_t initial_offset = *offset;
uint64_t matched = *offset;
int lvl, maxlvl;
int error = 0;
@ -2666,16 +2692,36 @@ dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset,
maxlvl = dn->dn_phys->dn_nlevels;
for (lvl = minlvl; lvl <= maxlvl; lvl++) {
for (lvl = minlvl; lvl <= maxlvl; ) {
error = dnode_next_offset_level(dn,
flags, offset, lvl, blkfill, txg);
if (error != ESRCH)
if (error == 0 && lvl > minlvl) {
--lvl;
matched = *offset;
} else if (error == ESRCH && lvl < maxlvl &&
dnode_next_block(dn, flags, &matched, lvl)) {
/*
* Continue search at next/prev offset in lvl+1 block.
*
* Usually we only search upwards at the start of the
* search as higher level blocks point at a matching
* minlvl block in most cases, but we backtrack if not.
*
* This can happen for txg > 0 searches if the block
* contains only BPs/dnodes freed at that txg. It also
* happens if we are still syncing out the tree, and
* some BP's at higher levels are not updated yet.
*
* We must adjust offset to avoid coming back to the
* same offset and getting stuck looping forever. This
* also deals with the case where offset is already at
* the beginning or end of the object.
*/
++lvl;
*offset = matched;
} else {
break;
}
while (error == 0 && --lvl >= minlvl) {
error = dnode_next_offset_level(dn,
flags, offset, lvl, blkfill, txg);
}
}
/*
@ -2687,9 +2733,6 @@ dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset,
error = 0;
}
if (error == 0 && (flags & DNODE_FIND_BACKWARDS ?
initial_offset < *offset : initial_offset > *offset))
error = SET_ERROR(ESRCH);
out:
if (!(flags & DNODE_FIND_HAVELOCK))
rw_exit(&dn->dn_struct_rwlock);