bcachefs: Fix opts.recovery_pass_last

This was lost in the giant recovery pass rework - but it's used heavily
by bcachefs subcommand utilities.

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

bcachefs: Fix allocate -> self healing path

When we go to allocate and find taht a bucket in the freespace btree is
actually allocated, we're supposed to return nonzero to tell the
allocator to skip it.

This fixes an emergency read only due to a bucket/ptr gen mismatch - we
also don't return the correct bucket gen when this happens.

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

bcachefs: Fix endianness in casefold check/repair

Fixes: 010c89468134 ("bcachefs: Check for casefolded dirents in non casefolded dirs")
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Path must be locked if trans->locked && should_be_locked

If path->should_be_locked is true, that means user code (of the btree
API) has seen, in this transaction, something guarded by the node this
path has locked, and we have to keep it locked until the end of the
transaction.

Assert that we're not violating this; should_be_locked should also be
cleared only in _very_ special situations.

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

bcachefs: Simplify bch2_path_put()

Simplify the "do we need to keep this locked?" checks.

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

bcachefs: Plumb btree_trans for more locking asserts

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

bcachefs: Clear trans->locked before unlock

We're adding new should_be_locked assertions: it's going to be illegal
to unlock a should_be_locked path when trans->locked is true.

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

bcachefs: Clear should_be_locked before unlock in key_cache_drop()

We're adding new should_be_locked assertions, also add a comment
explaining why clearing should_be_locked is safe here.

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

bcachefs: bch2_path_get() reuses paths if upgrade_fails & !should_be_locked

Small additional optimization over the previous patch, bringing us
closer to the original behaviour, except when we need to clone to avoid
a transaction restart.

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

bcachefs: Give out new path if upgrade fails

Avoid transaction restarts due to failure to upgrade - we can traverse a
new iterator without a transaction restart.

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

bcachefs: Fix btree_path_get_locks when not doing trans restart

btree_path_get_locks, on failure, shouldn't unlock if we're not issuing
a transaction restart: we might drop locks we're not supposed to (if
path->should_be_locked is set).

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

bcachefs: btree_node_locked_type_nowrite()

Small helper to improve locking assertions.

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

bcachefs: Kill bch2_path_put_nokeep()

bch2_path_put_nokeep() was intended for paths we wouldn't need to
preserve for a transaction restart - it always frees them right away
when the ref hits 0.

But since paths are shared, freeing unconditionally is a bug, the path
might have been used elsewhere and have should_be_locked set, i.e. we
need to keep it locked until the end of the transaction.

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

bcachefs: bch2_journal_write_checksum()

We need to delay checksumming the journal write; we don't know the
blocksize until after we allocate the write.

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

bcachefs: Reduce stack usage in data_update_index_update()

Separate tracepoint message generation and other slowpath code into
non-inline functions, and use bch2_trans_log_str() instead of using a
printbuf for our journal message.

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

bcachefs: bch2_trans_log_str()

The data update path doesn't need a printbuf for its log message - this
will help reduce stack usage.

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

bcachefs: Kill bkey_buf usage in data_update_index_update()

Reduce stack usage - bkey_buf has a 96 byte buffer on the stack, but the
btree_trans bump allocator works just fine here.

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

bcachefs: Drop empty accounting updates

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

bcachefs: Improve trace_trans_restart_upgrade

- Convert to a 'fs_str' tracepoint that just emits as a string: this
  lets us build up the tracepoint with a printbuf, using our pretty
  printers, and they're much easier to manage

- Include locks_held, before and after

- Include the btree node pointer we failed on (error pointer, null, or
  real node)

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

bcachefs: fix bch2_inum_snapshot_to_path()

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

bcachefs: fix duplicate printk

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

bcachefs: BCH_INODE_has_case_insensitive

Add a flag for tracking whether a directory has case-insensitive
descendents - so that overlayfs can disallow mounting, even though the
filesystem supports case insensitivity.

This is a new on disk format version, with a (cheap) upgrade to ensure
the flag is correctly set on existing inodes.

Create, rename and fssetxattr are all plumbed to ensure the new flag is
set, and we've got new fsck code that hooks into check_inode(0.

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

bcachefs: bch2_inode_find_by_inum_snapshot()

Move a fsck.c helper into inode.c, eliminate some duplicate and organize
the inode lookup helpers.

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

bcachefs: bch2_inum_snapshot_to_path()

Add a better helper for printing out paths of inodes when we don't know
the subvolume, for fsck.

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

bcachefs: bch2_rename_trans() only runs rename-to-dir code if needed

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

bcachefs: subvol_inum_eq()

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

bcachefs: Don't set bi_casefold on non directories

bi_casefold only makes sense for directories, and since it's one of the
variable length fields setting it unnecessarily wastes space.

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

bcachefs: Remove duplicate call to bch2_trans_begin()

There is one in for_each_btree_key_max().

Signed-off-by: Alan Huang <mmpgouride@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Call bch2_bkey_set_needs_rebalance() earlier in write path

There's no reason to be running this inside our transaction; it forces
us to copy the key we're updating to a temporary, which we'd like to
skip.

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

bcachefs: Simplify bch2_extent_atomic_end()

It used to be that we had a fixed maximum number of btree paths to work
with - 64.

That's no longer the case, so bch2_extent_atomic_end() doesn't have to
be as strict.

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

bcachefs: Coalesce accounting in trans commit

Accounting has gotten quite heavy, and there's lots of redundancy in
accounting updates within a transaction, as we often add/delete multiple
extents that touch the same accountign counters.

This will reduce the amount of data that we journal, and reduce pressure
downstream on the btree write buffer.

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

bcachefs: Split out accounting in transaction commit

There can be a lot of rendundancy in accounting updates within a single
btree transaction.

Split out accounting updates so that they can be deduped, in the next
commit.

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

bcachefs: btree_trans_subbuf

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

bcachefs: Make accounting mismatch errors more readable

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

bcachefs: async objs now support bch_write_ops

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

bcachefs: fix bch2_debugfs_flush_buf() when tabstops are in use

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

bcachefs: fsck: Include loops in error messages

This fixes the subvol loop checking and directory loop checking to print
the loop.

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

bcachefs: bch2_check_bucket_backpointer_mismatch()

Detect buckets with missing backpointers, and run repair on demand.

__bch2_move_data_phys() now calls
bch2_check_bucket_backpointer_mismatch() as it walks buckets, which
checks for missing backpointers by comparing backpointers against bucket
sector counts.

When missing backpointers are detected, we kick off
bch2_check_extents_to_backpointers() asynchronously - right away if
we're trying to evacuate, or with a threshold if we're just running
copygc.

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

bcachefs: Improve bucket_bitmap code

Add some more helpers, and mismatches is now a superset of the empty
bitmap - simplifies most checks.

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

bcachefs: Run recovery passes asynchronously

When we request a recovery pass to be run online, i.e. not during
recovery, if it's an online pass it'll now be run in the background,
instead of waiting for the next mount.

To avoid situations where recovery passes are running continuously, this
also includes ratelimiting: if the RUN_RECOVERY_PASS_ratelimit flag is
passed, the pass may be deferred until later - depending on the runtime
and last run stats in the recovery_passes superblock section.

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

bcachefs: bch2_run_explicit_recovery_pass() cleanup

Consolidate the run_explicit_recovery_pass() interfaces by adding a
flags parameter; this will also let us add a RUN_RECOVERY_PASS_ratelimit
flag.

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

bcachefs: bch2_recovery_pass_status_to_text()

Show recovery pass status in sysfs - important now that we're running
them automatically in the background.

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

bcachefs: Reduce usage of recovery.curr_pass

We want recovery.curr_pass to be private to the recovery passes code,
for better showing recovery pass status; also, it may rewind and is
generally not the correct member to use.

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

bcachefs: __bch2_run_recovery_passes()

Consolidate bch2_run_recovery_passes() and
bch2_run_online_recovery_passes(), prep work for automatically
scheduling and running recovery passes in the background.

- Now takes a mask of which passes to run, automatic background repair
  will pass in sb.recovery_passes_required.

- Skips passes that are failing: a pass that failed may be reattempted
  after another pass succeeds (some passes depend on repair done by
  other passes for successful completion).

- bch2_recovery_passes_match() helper to skip alloc passes on a
  filesystem without alloc info.

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

bcachefs: struct bch_fs_recovery

bch_fs has gotten obnoxiously big, let's start organizing thins a bit
better.

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

bcachefs: kill copy in bch2_disk_accounting_mod()

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

bcachefs: Optimize bch2_trans_start_alloc_update()

Avoid doing more updates if we already have one.

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

bcachefs: btree key cache asserts

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

bcachefs: journal path now uses discard_opt_enabled()

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

bcachefs: relock_fail tracepoint now includes btree

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

bcachefs: do_rebalance_scan() now only updates bch_extent_rebalance

This ensures that our pending rebalance work accounting is accurate
quickly.

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

bcachefs: better error message for subvol_fs_path_parent_wrong

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

bcachefs: Improve bch2_repair_inode_hash_info()

Improve this so it can be used by fsck.c check_inode(); it provides a
much better error message than the check_inode() version.

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

bcachefs: bch2_inode_find_snapshot_root()

Factor out a small common helper.

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

bcachefs: Early return to avoid unnecessary lock

Signed-off-by: Alan Huang <mmpgouride@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Kill BTREE_TRIGGER_bucket_invalidate

Signed-off-by: Alan Huang <mmpgouride@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Fix opt hooks in sysfs for non sb option

We weren't checking if the option changed for non-superblock options -
this led to rebalance not waking up when enabling the
"rebalance_enabled" option.

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

bcachefs: fix can_write_extent()

Failing to check the return value of bch2_dev_rcu(): we could
(technically) race with device removal.

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

bcachefs: Add tracepoint, counter for io_move_created_rebalance

Internal moves shouldn't add new rebalance_work, but it's been reported
that this seems to be happening. Add a tracepoint and counter so we can
see what's going on.

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

bcachefs: move_buckets in rhashtable when allocated

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

bcachefs: Move pending buckets queue to buckets_in_flight

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

bcachefs: kill move_bucket_in_flight

Small cleanup/simplification, and prep work for the next patch, which
will add checking if buckets don't get evacuated because they're missing
backpointers.

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

bcachefs: bch2_fs_emergency_read_only2()

More error message cleanup: instead of multiple printk()s per error, we
want to be building up a single error message in a printbuf, so that it
can be printed with indenting that shows grouping and avoid errors
getting interspersed or lost in the log.

This gets rid of most calls to bch2_fs_emergency_read_only(). We still
have calls to
 - bch2_fatal_error()
 - bch2_fs_fatal_error()
 - bch2_fs_fatal_err_on()

that need work.

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

bcachefs: Extra write buffer asserts

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

bcachefs: add missing locking in bch2_write_point_to_text()

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

bcachefs: Don't rewind recovery if not in recovery

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

bcachefs: Rename fsck_running, recovery_running flags

Slightly more readable.

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

bcachefs: debug_check_bkey_unpack

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

bcachefs: debug_check_bset_lookups

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

bcachefs: debug_check_iterators no longer requires BCACHEFS_DEBUG

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

bcachefs: debug_check_btree_locking modparam

Don't put btree locking asserts behind CONFIG_BCACHEFS_DEBUG, put them
behind a module parameter.

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

bcachefs: Debug params are now static_keys

We'd like users to be able to debug without building custom kernels, so
this will help us get rid of CONFIG_BCACHEFS_DEBUG, at least for most
things.

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

bcachefs: Slim down inlined part of bch2_btree_path_upgrade()

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

bcachefs: online_fsck_mutex -> run_recovery_passes_lock

Prep work for automatically running recovery passes asynchronously.

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

bcachefs: bch_sb_field_recovery_passes

New superblock section for statistics on recovery passes - last time
ran (successfully), last runtime.

This will be used by self healing code to determine when to kick off
potentially expensive recovery passes.

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

bcachefs: recovery_passes_types.h -> recovery_passes_format.h

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

bcachefs: print label correctly in sb_member_to_text()

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

bcachefs: "buckets with backpointer mismatches" now allocated on demand

More self healing work: we're going to be calling
check_bucket_backpointer_mismatch() at runtime, outside of fsck.

Then when we need to we'll kick off the full
check_extents_to_backpointers recovery pass.

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

bcachefs: delete dead items in bch_dev

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

bcachefs: kill dead code in move_data_phys()

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

bcachefs: buckets_in_flight on stack

copygc runs with a full stack available, there's no reason to
dynamically allocate this.

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

bcachefs: bch2_copygc_dev_wait_amount()

Factor out the per-device calculations, for better introspection.

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

bcachefs: Add missing include

fix debug build in userspace

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

bcachefs: Knob for manual snapshot deletion

Add 'opts.snapshot_deletion_enabled', enabled by default.

This may be turned off so that the new sysfs knob,
'internal/trigger_delete_dead_snapshots', may be used instead - this
will allow snapshot deletion to be profiled more easily.

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

bcachefs: bcachefs_metadata_version_fast_device_removal

Fast device removal, that uses backpointers to find pointers to the
device being removed instead of a full metadata scan.

This requires BCH_SB_MEMBER_DELETED_UUID, which is an incompatible
change - hence the version number bump. We don't fully trust
backpointers, so we don't want to reuse device indexes until after a
fsck has verified that there aren't any pointers to removed devices.

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

bcachefs: bch2_dev_data_drop_by_backpointers()

Currently, device removal has to scan all metadata for pointers to the
device being removed.

Add a new method, with the same interface as bch2_dev_data_drop(), that
scans by backpointers instead - this will drastically speed up device
removal.

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

bcachefs: BCH_SB_MEMBER_DELETED_UUID

Add a sentinal value for devices that have been removed, but don't want
to reuse their index until a fsck has completed.

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

bcachefs: bch2_dev_remove_stripes() respects degraded flags

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

bcachefs: opts.rebalance_on_ac_only

Add an option for setting rebalance to only run when connected to mains
power.

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

bcachefs: __bch2_fs_free() cleanup

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

bcachefs: Improve bch2_extent_ptr_set_cached()

Preferentially keep existing cached pointers instead of adding new ones.

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

bcachefs: improve check_inode_hash_info_matches_root() error message

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

bcachefs: inline bch2_ob_ptr()

This was an oversight, we want bch2_alloc_sectors_append_ptrs_inlined()
fully inlined.

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

bcachefs: bch2_dev_in_target() no longer takes rcu_read_lock()

Minor optimization, the caller generally has it already.

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

bcachefs: bch2_journal_write() refactoring

Make the locking easier to follow; also take io_refs earlier, in
__journal_write_alloc().

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

bcachefs: delete_dead_snapshot_keys_v2()

Since extents, dirents and xattrs require an inode with the
corresponding snapshot ID to exists, we can avoid a lot of scanning by
only scanning those trees for keys to process if the correspending inode
exists.

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

bcachefs: bcachefs_metadata_version_snapshot_deletion_v2

We're going to be speeding up snapshot deletion, by only having it
process the extents/dirents/xattrs btrees if an inode of a given
snapshot ID was present.

This raises the possibility of 'bkey_in_missing_snapshot' errors popping
up, if we ever accidentally don't do the corresponding inode update, or
if the new algorithm has bugs.

So instead of deleting snapshot IDs, add a new deleted flag, so that
'key in missing snapshot' errors can more definitively tell what
happened and automatically repair.

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

bcachefs: BCH_SNAPSHOT_DELETED -> BCH_SNAPSHOT_WILL_DELETE

We're going to be speeding up snapshot deletion, by only having it
process the extents/dirents/xattrs btrees if an inode of a given
snapshot ID was present.

This raises the possibility of 'bkey_in_missing_snapshot' errors popping
up, if we ever accidentally don't do the corresponding inode update, or
if the new algorithm has bugs.

So we'll want to be able to differentiate more definitively between
'snapshot went missing' (and perhaps needs to be reconstructed), and
'key in snapshot that was deleted'.

So instead of deleting snapshot IDs, we'll be adding a new deleted flag
and leaving them permanently.

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

bcachefs: Skip unrelated snapshot trees in snapshot deletion

Don't scan keys in inodes for which the snapshot tree doesn't match any
we're deleting from.

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

bcachefs: BCH_FSCK_ERR_snapshot_key_missing_inode_snapshot

We're going to be doing some snapshot deletion performance improvements,
and those will strictly require that if an extent/dirent/xattr is
present, an inode is present in that snapshot ID.

We already check for this, but we don't repair it on disk: this patch
adds that repair and turns it into a real fsck_err().

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