The sort in the btree write buffer flush path is a very hot path, and
it's particularly performance sensitive since it's single threaded and
can block every other thread on a multithreaded write workload.

It's well worth doing a sort with inlined cmp and swap functions.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Previosuly, the transaction commit path would have to add keys to the
btree write buffer as a separate operation, requiring additional global
synchronization.

This patch introduces a new journal entry type, which indicates that the
keys need to be copied into the btree write buffer prior to being
written out. We switch the journal entry type back to
JSET_ENTRY_btree_keys prior to write, so this is not an on disk format
change.

Flushing the btree write buffer may require pulling keys out of journal
entries yet to be written, and quiescing outstanding journal
reservations; we previously added journal->buf_lock for synchronization
with the journal write path.

We also can't put strict bounds on the number of keys in the journal
destined for the write buffer, which means we might overflow the size of
the preallocated buffer and have to reallocate - this introduces a
potentially fatal memory allocation failure. This is something we'll
have to watch for, if it becomes an issue in practice we can do
additional mitigation.

The transaction commit path no longer has to explicitly check if the
write buffer is full and wait on flushing; this is another performance
optimization. Instead, when the btree write buffer is close to full we
change the journal watermark, so that only reservations for journal
reclaim are allowed.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Add a new lock for synchronizing between journal IO path and btree write
buffer flush.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Ensure that journal bufs that haven't been written can't be reclaimed
from the journal pin fifo, and can thus have new pins taken.

Prep work for changing the btree write buffer to pull keys from the
journal directly.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

In the future we'll be making trans->paths resizable and potentially
having _many_ more paths (for fsck); we need to start fixing algorithms
that walk each path in a transaction where possible.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Instead of using a darray, we now allocate journal entries for the
transaction commit path with our normal bump allocator - with an inlined
fastpath, and using btree_transaction_stats to remember how much to
initially allocate so as to avoid transaction restarts.

This is prep work for converting write buffer updates to use this
mechanism.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

These were for extra info in tracepoints for debugging a specialized
issue - we do not want to bloat btree_path for this, at least in release
builds.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

c->curr_recovery_pass can go backwards; this adds a non rewinding
version, c->recovery_pass_done.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Fix a few typos in the six.h header file.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Brian Foster <bfoster@redhat.com>
Cc: linux-bcachefs@vger.kernel.org
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

for_each_btree_key() handles transaction restarts, like
for_each_btree_key2(), but only calls bch2_trans_begin() after a
transaction restart - for_each_btree_key2() wraps every loop iteration
in a transaction.

The for_each_btree_key() behaviour is problematic when it leads to
holding the SRCU lock that prevents key cache reclaim for an unbounded
amount of time - there's no real need to keep it around.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

continue now works as in any other loop
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

In the debugfs code, we had an incorrect use of drop_locks_do(); on
transaction restart we don't want to restart the current loop iteration,
since we've already emitted the current key to the buffer for userspace.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This adds a new ioctl for running fsck on a mounted, in use filesystem.

This reuses the fsck_thread code from the previous patch for running
fsck on an offline, unmounted filesystem, so that log messages for the
fsck thread are redirected to userspace.

Only one running fsck instance is allowed at a time; a new semaphore
(since the lock will be taken by one thread and released by another) is
added for this.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This adds a new ioctl for running fsck on a list of devices.

Normally, if we wish to use the kernel's implementation of fsck we'd run
it at mount time with -o fsck. This ioctl lets us run fsck without
mounting, so that userspace bcachefs-tools can transparently switch to
the kernel's implementation of fsck when appropriate - primarily if the
kernel version of bcachefs better matches the filesystem on disk.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Add a new helper for running online recovery passes - i.e. online fsck.
This is a subset of our normal recovery passes, and does not - for now -
use or follow c->curr_recovery_pass.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Online fsck is coming, and many of our recovery/fsck passes are already
safe to run while the filesystem is in use - mark which ones.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Upcoming patches are going to add two new ioctls for running fsck in the
kernel, but pretending that we're running our normal userspace fsck.

This patch adds some plumbing for redirecting our normal log messages
away from the dmesg log to a thread_with_file file descriptor - via a
struct log_output, which will be consumed by the fsck f_op's read method.

The new ioctls will allow for running fsck in the kernel against an
offline filesystem (without mounting it), and an online filesystem. For
an offline filesystem we need a way to pass in a pointer to the
log_output, which is done via a new hidden opts.h option.

For online fsck, we can set c->output directly, but only want to
redirect log messages from the thread running fsck - hence the new
c->output_filter method.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Abstract out a new helper from the data job code, for connecting a
kthread to a file descriptor.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Add a new refcount for async ops that don't necessarily need the fs to
be RW, with similar lifetime/rules otherwise as c->writes.

To be used by online fsck.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

single_device.merge_torture_flakey is, very rarely, finding a btree node
that doesn't match the key that points to it: this patch improves the
error message to print out more fields from the btree node header, so
that we can see what else does or does not match the key.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

When investigating transient failures of generic/441 on bcachefs, it
was determined that the cause of the failure was a combination of
unconditional emergency shutdown and racing between background
journal activity and the test switchover from a working device
mapper table to an error injecting table.

Part of the reason for this sequence of events is that bcachefs
aggressively flushes as much as possible during fsync(), regardless
of errors. While this is reasonable behavior, it is technically
unnecessary because once an error is returned from fsync(), the
caller cannot make any assumptions about the resilience of data.

Tweak the bch2_fsync() logic to return an error on failure of any of
the steps involved in the flush. Note that this change alone does
not prevent generic/441 failure, but in combination with a test
tweak to avoid racing during the dm-error table switchover it avoids
the unnecessary shutdowns and allows the test to pass reliably on
bcachefs.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Continuing the project of replacing generic error codes with more
specific ones.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Remove obsolete comment about zstd, since approach changed during
development of commit bbc3a460Signed-off-by: Richard Davies <richard@arachsys.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This gives us more context information - e.g. which codepath is invoking
btree node reads.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs grabs s_umount and sets SB_RDONLY when the fs is shutdown
via the ioctl() interface. This has a couple issues related to
interactions between shutdown and freeze:

1. The flags == FSOP_GOING_FLAGS_DEFAULT case is a deadlock vector
   because freeze_bdev() calls into freeze_super(), which also
   acquires s_umount.

2. If an explicit shutdown occurs while the sb is frozen, SB_RDONLY
   alters the thaw path as if the sb was read-only at freeze time.
   This effectively leaks the frozen state and leaves the sb frozen
   indefinitely.

The usage of SB_RDONLY here goes back to the initial bcachefs commit
and AFAICT is simply historical behavior. This behavior is unique to
bcachefs relative to the handful of other filesystems that support
the shutdown ioctl(). Typically, SB_RDONLY is reserved for the
proper remount path, which itself is restricted from modifying
frozen superblocks in reconfigure_super(). Drop the unnecessary sb
lock and flags update bch2_ioc_goingdown() to address both of these
issues.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

backpointers fsck now always runs in rw mode - the btree is being
modified while it runs, by e.g. copygc, rebalance, the discard worker,
the invalidate worker.

We could find a missing backpointer, flush the btree write buffer, and
then on the next iteration find a new key at the exact same position -
which will most likely need another write buffer flush.

Hence, we have to check for an exact match on last_flushed, not just the
pos.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Co-developed-by: Kent Overstreet <kent.overstreet@linux.dev>
Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

This eliminates a lot of BCH_TRANS_COMMIT_lazy_rw flags, and is less
error prone.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Signed-off-by: Daniel Hill <daniel@gluo.nz>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Fake flexible arrays (zero-length and one-element arrays) are
deprecated, and should be replaced by flexible-array members.

So, replace zero-length arrays with flexible-array members
in multiple structures.
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>