This adds support for nocow mode, where we do writes in-place when
possible. Patch components:
- New boolean filesystem and inode option, nocow: note that when nocow
is enabled, data checksumming and compression are implicitly disabled
- To prevent in-place writes from racing with data moves
(data_update.c) or bucket reuse (i.e. a bucket being reused and
re-allocated while a nocow write is in flight, we have a new locking
mechanism.
Buckets can be locked for either data update or data move, using a
fixed size hash table of two_state_shared locks. We don't have any
chaining, meaning updates and moves to different buckets that hash to
the same lock will wait unnecessarily - we'll want to watch for this
becoming an issue.
- The allocator path also needs to check for in-place writes in flight
to a given bucket before giving it out: thus we add another counter
to bucket_alloc_state so we can track this.
- Fsync now may need to issue cache flushes to block devices instead of
flushing the journal. We add a device bitmask to bch_inode_info,
ei_devs_need_flush, which tracks devices that need to have flushes
issued - note that this will lead to unnecessary flushes when other
codepaths have already issued flushes, we may want to replace this with
a sequence number.
- New nocow write path: look up extents, and if they're writable write
to them - otherwise fall back to the normal COW write path.
XXX: switch to sequence numbers instead of bitmask for devs needing
journal flush
XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to
run in process context - see if we can improve this
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The data update path requires special support for unwritten extents - we
still need to be able to move them, but there's no need to read or write
anything.
This patch adds a new error code to tell bch2_move_extent() that we're
short circuiting the read, and adds bch2_update_unwritten_extent() to
create a reservation then call __bch2_data_update_index_update().
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
The btree key cache mainly helps with lock contention, at the cost of
additional memory overhead. During some fsck passes the memory overhead
really matters, but fsck is single threaded so lock contention is an
issue - so skipping the key cache during fsck will help with
performance.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Previously, copygc needed to walk the entire extents & reflink btrees to
find extents that needed to be moved.
Now that we have backpointers, this patch implements
bch2_evacuate_bucket() in the move code, which copygc now uses for
evacuating mostly empty buckets.
Also, thanks to the new backpointers code, copygc can now move btree
nodes.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds a debug mode where we split up the c->writes refcount into
distinct refcounts for every codepath that takes a reference, and adds
sysfs code to print the value of each ref.
This will make it easier to debug shutdown hangs due to refcount leaks.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This improves io_opts() and makes it a non-inline function - it's big
enough that it probably shouldn't be.
Also, bch_io_opts no longer needs fields for whether options are
defined, so we can slim it down a bit.
We'd like to stop passing around the full bch_io_opts, but that'll be
tricky because of bch2_rebalance_add_key().
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This introduces some new conveniences, to help cut down on boilerplate:
- bch2_trans_kmalloc_nomemzero() - performance optimiation
- bch2_bkey_make_mut()
- bch2_bkey_get_mut()
- bch2_bkey_get_mut_typed()
- bch2_bkey_alloc()
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This patch introduces
- bpos_eq()
- bpos_lt()
- bpos_le()
- bpos_gt()
- bpos_ge()
and equivalent replacements for bkey_cmp().
Looking at the generated assembly these could probably be improved
further, but we already see a significant code size improvement with
this patch.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
- Marking a non-static function as inline doesn't actually work and is
now causing problems - drop that
- Introduce BCACHEFS_LOG_PREFIX for when we want to prefix log messages
with bcachefs (filesystem name)
- Userspace doesn't have real percpu variables (maybe we can get this
fixed someday), put an #ifdef around bch2_disk_reservation_add()
fastpath
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
checkpatch.pl gives lots of warnings that we don't want - suggested
ignore list:
ASSIGN_IN_IF
UNSPECIFIED_INT - bcachefs coding style prefers single token type names
NEW_TYPEDEFS - typedefs are occasionally good
FUNCTION_ARGUMENTS - we prefer to look at functions in .c files
(hopefully with docbook documentation), not .h
file prototypes
MULTISTATEMENT_MACRO_USE_DO_WHILE
- we have _many_ x-macros and other macros where
we can't do this
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Cached pointers are generally dropped, not moved: this led to an
assertion firing in the data update path when there were no new replicas
being written.
This path adds a data_options field for pointers to be dropped, and
tweaks move_extent() to check if we're only dropping pointers, not
writing new ones, before kicking off a data update operation.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Also, do some reorganizing/renaming, convert atomic counters in bch_fs
to persistent counters, and add a few missing counters.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Now that we have error codes, with subtypes, we can switch to our own
error code for transaction restarts - and even better, a distinct error
code for each transaction restart reason: clearer code and better
debugging.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
There's no point reading an extent in order to move it if the write is
going to fail because we're shutting down. This patch changes the move
path so that moving_io now owns a ref on c->writes - as a bonus,
rebalance and copygc will now notice that we're shutting down and exit
quicker.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
- add bch2_moving_ctxt_(init|exit)
- split out __bch2_evacutae_bucket() which takes an existing
moving_ctxt, this will be used for improving copygc performance by
pipelining across multiple buckets
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
move_ratelimit() now has a bool that specifies whether we want to
wait for copygc to finish.
When copygc is running, we're probably low on free buckets instead
of consuming the remaining buckets, we want to wait for copygc to
finish.
This should help with performance, and run away bucket fragmentation.
Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This patch significantly cleans up and simplifies the data_update
interface. Instead of only being able to specify a single pointer by
device to rewrite, we're now able to specify any or all of the pointers
in the original extent to be rewrited, as a bitmask.
data_cmd is no more: the various pred functions now just return true if
the extent should be moved/updated. All the data_update path does is
rewrite existing replicas, or add new ones.
This fixes a bug where with background compression on replicated
filesystems, where rebalance -> data_update would incorrectly drop the
wrong old replica, and keep trying to recompress an extent pointer and
each time failing to drop the right replica. Oops.
Now, the data update path doesn't look at the io options to decide which
pointers to keep and which to drop - it only goes off of the
data_update_options passed to it.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This is the start of reorganizing the data IO paths. The plan is to also
break apart io.c into data_read.c and data_write.c, and migrate_write
will be renamed to the data_update path.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This adds a new option, move_bytes_in_flight, for configuring the amount
of IO in flight by copygc/rebalance - users with many devices in their
filesystem will want to increase this.
In the future we should be smarter about this, but this is an easy
improvement.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This adds a new superblock field for persisting counters
and adds a sysfs interface in counters/ exposing these counters.
The superblock field is ignored by older versions letting us avoid
an on disk version bump.
Each sysfs file outputs a counter that tracks since filesystem
creation and a counter for the current mount session.
Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Delete some obsolete tracepoints, organize alloc tracepoints better,
make a few tracepoints more consistent.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We're seeing occasional firings of the assertion in the key cache
shutdown code that nr_dirty == 0, which means we must sometimes be doing
transaction commits after we've gone read only.
Cleanups & changes:
- BCH_FS_ALLOC_CLEAN renamed to BCH_FS_CLEAN_SHUTDOWN
- new helper bch2_btree_interior_updates_flush(), which returns true if
it had to wait
- bch2_btree_flush_writes() now also returns true if there were btree
writes in flight
- __bch2_fs_read_only now checks if btree writes were in flight in the
shutdown loop: btree write completion does a transaction update, to
update the pointer in the parent node
- assert that !BCH_FS_CLEAN_SHUTDOWN in __bch2_trans_commit
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We don't actually want copygc allocations to be nowait - an allocation
for copygc might fail and then later succeed due to a bucket needing to
wait on journal commit, or to be discarded.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
In move_read_endio, we were checking if the next pending write has its
read completed - but this can turn after a use after free (and we were
accessing the list without a lock), so instead just better to just
unconditionally do the wakeup.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Since we retry reads when we discover we read from a pointer that went
stale, if a dirty pointer is erroniously stale it would cause us to loop
retrying that read forever - unless we check before issuing the read,
while the btree is still locked, when we know that a dirty pointer
should never be stale.
This patch adds that check, along with printing some helpful debug info.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We need to ensure we don't have any btree locks held when calling
do_pending_writes() - besides issuing IOs, upcoming allocator changes
will have allocations doing btree lookups directly.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We weren't checking for -EINTR in the main loop in __bch2_move_data -
this code predates modern transaction restarts.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
It doesn't make much sense to be erasure coding cached pointers, we
should be erasure coding one of the dirty pointers in an extent. This
patch makes sure we're passing BCH_WRITE_CACHED when we expect the new
pointer to be a cached pointer, and tweaks the write path to not
allocate from a stripe when BCH_WRITE_CACHED is set - and fixes an
assertion we were hitting in the ec path where when adding the stripe to
an extent and deleting the other pointers the pointer to the stripe
didn't exist (because dropping all dirty pointers from an extent turns
it into a KEY_TYPE_error key).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
With erasure coding, copygc's count of sectors to move was off, which
matters for the debug statement it prints out when it's not able to move
all the data it tried to.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Add fields to inode & alloc keys that record the journal sequence number
when they were most recently modified.
For alloc keys, this is needed to know what journal sequence number we
have to flush before the bucket can be reused. Currently this is tracked
in memory, but we'll be getting rid of the in memory bucket array.
For inodes, this is needed for fsync when the inode has been evicted
from the vfs cache. Currently we use a bloom filter per outstanding
journal buf - but that mechanism has been broken since we added the
ability to not issue a flush/fua for every journal write.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
- The backpointer that ec_stripe_update_ptrs() uses now needs to include
the snapshot ID, which means we have to change where we add the
backpointer to after getting the snapshot ID for the new extents
- ec_stripe_update_ptrs() needs to be calling bch2_trans_begin()
- improve error message in bch2_mark_stripe()
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This fixes a possible race where we fail to remove a device because of
btree nodes still on it, that are being deleted by in flight btree
updates.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
We have been getting away from handling transaction restarts locally -
convert bch2_btree_node_rewrite() to the newer style.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
But we don't need to call it from outside the btree iterator code
anymore, since it's called by bch2_trans_begin() and
bch2_btree_path_traverse().
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
It shouldn't be necessary when we're only using a single iterator and
not doing updates, but that's harder to debug at the moment.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Now that peek_node()/next_node() are converted to return errors
directly, we don't need bch2_trans_exit() to return errors - it's
cleaner this way and wasn't used much anymore.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This changes for_each_btree_node() to work like for_each_btree_key(),
and to that end bch2_btree_iter_peek_node() and next_node() also return
error ptrs.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
When force-removing a device, we were silently dropping extents that we
no longer had pointers for - we should have been switching them to
KEY_TYPE_error, so that reads for data that was lost return errors.
This patch adds the logic for switching a key to KEY_TYPE_error to
bch2_bkey_drop_ptr(), and improves the logic somewhat.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
It was switching off of the key type incorrectly - this code must've
been quite old, and not rereplicating anything that wasn't a
btree_ptr_v1 or a plain old extent.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
The data move path operates on existing extents, and not within a
subvolume as the regular IO paths do. It needs to change because it may
cause existing extents to be split, and when splitting an existing
extent in an ancestor snapshot we need to make sure the new split has
the same visibility in child snapshots as the existing extent.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
To implement snapshots, we need every filesystem btree operation (every
btree operation without a subvolume) to start by looking up the
subvolume and getting the current snapshot ID, with
bch2_subvolume_get_snapshot() - then, that snapshot ID is used for doing
btree lookups in BTREE_ITER_FILTER_SNAPSHOTS mode.
This patch adds those bch2_subvolume_get_snapshot() calls, and also
switches to passing around a subvol_inum instead of just an inode
number.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
This splits btree_iter into two components: btree_iter is now the
externally visible componont, and it points to a btree_path which is now
reference counted.
This means we no longer have to clone iterators up front if they might
be mutated - btree_path can be shared by multiple iterators, and cloned
if an iterator would mutate a shared btree_path. This will help us use
iterators more efficiently, as well as slimming down the main long lived
state in btree_trans, and significantly cleans up the logic for iterator
lifetimes.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
If an extent ends up with a replica that is encrypted an a replica that
isn't encrypted (due the user changing options), and then
copygc/rebalance moves one of the replicas by reading from the
unencrypted replica, we had a bug where we wouldn't correctly initialize
op->nonce - for each crc field in an extent, crc.offset + crc.nonce must
be equal.
This patch fixes that by moving op.nonce initialization to
bch2_migrate_write_init.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
BTREE_ITER_SET_POS_AFTER_COMMIT is used internally to automagically
advance extent btree iterators on sucessful commit.
But with the upcomnig btree_path patch it's getting more awkward to
support, and it adds overhead to core data structures that's only used
in a few places, and can be easily done by the caller instead.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>