This was planned to be done ages ago, now finally completed; there are
places where we have quite a few btree_trans objects on the stack, so
this reduces stack usage somewhat.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
There's something going on with the data move path; log the original key
being moved for debugging.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Replace these with proper private error codes, so that when we get an
error message we're not sifting through the entire codebase to see where
it came from.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This fixes another "rebalance spinning and doing no work" issue;
rebalance was reading extents it wanted to move, but then failing in
bch2_write() -> bch2_alloc_sectors_start() due to being unable to
allocate sufficient replicas.
This was triggered by a user playing with the durability settings, the
foreground device was an NVME device with durability=2, and originally
he'd set the background device to durability=2 as well, but changed it
back to 1 (the default) after seeing IO errors.
That meant that with replicas=2, we want to move data off the NVME
device which satisfies that constraint, but with a single durability=1
device on the background target there's no way to move the extent to
that target while satisfiying the "required replicas" constraint.
The solution for now is for bch2_data_update_init() to check for this,
and return an error - before kicking off the read.
bch2_data_update_init() already had two different checks for "will we be
able to write this extent", with partially duplicated code, so this
patch combines and improves that logic.
Additionally, we now always bail out and return an error if there's
insufficient space on the destination target. Previously, we only did
this for BCH_WRITE_alloc_nowait moves, because it might be the case that
copygc just needs to free up space on the destination target.
But we really shouldn't kick off a move if the destination is full, we
can't currently distinguish between "really full" and "just need to wait
for copygc", and if we are going to wait on copygc it'd be better to do
that before kicking off the move.
This will additionally fix "rebalance spinning" issues caused by a
filesystem that has more data than can fit in background_target - which
is a valid scenario, since we don't exclude foreground/cache devices
when calculating filesystem capacity.
Reported-by: Maël Kerbiriou <mael.kerbiriou@free.fr>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Read flags are codepath dependent and change as they're passed around,
while the fields in rbio._state are mostly fixed properties of that
particular object.
Losing track of BCH_READ_data_update would be bad, and previously it was
not obvious if it was always correctly set in the rbio, so this is a
safety cleanup.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
If a device is ro or failed, we might not have anywhere to move a
replica.
Check for this early, before doing the read and attempting to write.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Reorganize counters a bit, grouping related counters together.
New counters:
- io_read_inline
- io_read_hole
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new data op to walk all data and metadata in a filesystem,
checking if it can be read successfully, and on error repairing from
another copy if possible.
- New helper: bch2_dev_idx_is_online(), so that we can bail out and
report to userspace when we're unable to scrub because the device is
offline
- data_update_opts, which controls the data move path, now understands
scrub: data is only read, not written. The read path is responsible
for rewriting on read error, as with other reads.
- scrub_pred skips data extents that don't have checksums
- bch_ioctl_data has a new scrub member, which has a data_types field
for data types to check - i.e. all data types, or only metadata.
- Add new entries to bch_move_stats so that we can report numbers for
corrected and uncorrected errors
- Add a new enum to bch_ioctl_data_event for explicitly reporting
completion and return code (i.e. device offline)
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
If a data update doesn't want to block on allocations (promotes, self
healing on read error) - check if the allocation would fail before
kicking off the data update and calling into the write path.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Initialize the write op first, so that in the next patch we can check if
the allocator would block (for BCH_WRITE_alloc_nowait ops) and bail out
before taking nocow locks/dev refs.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new helper for bch2_moving_ctxt_to_text(), which may be used to
debug if moving_ios are getting stuck.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
"nonce inconstancy" is popping up again, causing us to go emergency
read-only.
This one looks less serious, i.e. specific to the encryption path and
not indicative of a data corruption bug. But we'll need more info to
track it down.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We're currently debugging issues with rebalance, where it's not making
progress as quickly as it should be (or sometimes not at all).
Add the full data_update to the move_extent_finish tracepoint, so we can
check that the replicas we wrote match what we were supposed to do.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
bch2_snapshot_equiv() is going away; convert users that just wanted to
know if the snapshot exists to something better
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Add a new parameter to bkey validate functions, and use it to improve
invalid bkey error messages: we can now print the btree and depth it
came from, or if it came from the journal, or is a btree root.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We had a bug report where the data update path was creating an extent
that failed to validate because it had too many pointers; almost all of
them were cached.
To fix this, we have:
- want_cached_ptr(), a new helper that checks if we even want a cached
pointer (is on appropriate target, device is readable).
- bch2_extent_set_ptr_cached() now only sets a pointer cached if we want
it.
- bch2_extent_normalize_by_opts() now ensures that we only have a single
cached pointer that we want.
While working on this, it was noticed that this doesn't work well with
reflinked data and per-file options. Another patch series is coming that
plumbs through additional io path options through bch_extent_rebalance,
with improved option handling.
Reported-by: Reed Riley <reed@riley.engineer>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This fixes an assertion pop in io_write.c - if we don't return an error
we're supposed to have completed all the btree updates.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
data_update_init() does a bunch of complicated stuff to decide how many
replicas to add, since we only want to increase an extent's durability
on an explicit rereplicate, but extent pointers may be on devices with
different durability settings.
There was a corner case when evacuating a device that had been set to
durability=0 after data had been written to it, and extents on that
device had already been rereplicated - then evacuate only needs to drop
pointers on that device, not move them.
So the assert for !m->op.nr_replicas was spurious; this was a perfectly
legitimate case that needed to be handled.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
bkey_fsck_err() was added as an interface that looks like fsck_err(),
but previously all it did was ensure that the appropriate error counter
was incremented in the superblock.
This is a cleanup and bugfix patch that converts it to a wrapper around
fsck_err(). This is needed to fix an issue with the upgrade path to
disk_accounting_v3, where the "silent fix" error list now includes
bkey_fsck errors; fsck_err() handles this in a unified way, and since we
need to change printing of bkey fsck errors from the caller to the inner
bkey_fsck_err() calls, this ends up being a pretty big change.
Als,, rename .invalid() methods to .validate(), for clarity, while we're
changing the function signature anyways (to drop the printbuf argument).
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
bch2_dev_bkey_exists() is going away; bch2_dev_have_ref() documents that
we're looking up a device without checking if it's present because we
have a reference to it already.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
We're about to add new asserts for btree_trans locking consistency, and
part of that requires that aren't using the btree_trans while it's
unlocked.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Combine iter/update/trigger/str_hash flags into a single enum, and
x-macroize them for a to_text() function later.
These flags are all for a specific iter/key/update context, so it makes
sense to group them together - iter/update/trigger flags were already
given distinct bits, this cleans up and unifies that handling.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Previously, we assumed that keys were consistent with the snapshots
btree - but that's not correct as fsck may not have been run or may not
be complete.
This adds checks and error handling when using the in-memory snapshots
table (that mirrors the snapshots btree).
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The "apply this compression method in the background" paths now use the
compression option if background_compression is not set; this means that
setting or changing the compression option will cause existing data to
be compressed accordingly in the background.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
