bcachefs: Drop a faulty assertion

the assertion was meant to check that bch2_journal_reclaim_fast() was
always being called, but since the atomic dec can happen outside of
j->lock the assertion itself can race

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

bcachefs: improved flush_held_btree_writes()

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

bcachefs: use correct wq for journal reclaim

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

bcachefs: Journal reclaim refactoring

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

bcachefs: Better journal debug

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

bcachefs: Don't get journal reservation until after we know insert will succeed

Checking if we can do the insert after getting the journal reservation
means potentially wasting space in the journal, which will break the new
pre reservation mechanism

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

bcachefs: fix integer underflow in journal code

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

bcachefs: Fix gc handling of bucket gens

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

bcachefs: fix a deadlock on startup

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

bcachefs: force str_hash code to be inlined

the btree trans iterator stuff relies on a dirty horrible hack that uses
a short backtrace to generate tokens, and gcc uninlining these functions
breaks that hack. ugh.

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

bcachefs: Assorted journal refactoring

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

bcachefs: fs_usage_u64s()

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

bcachefs: Add a mechanism for blocking the journal

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

bcachefs: Convert bucket invalidation to key marking path

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

bcachefs: fixes for cached data accounting

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

bcachefs: fix more locking bugs

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

bcachefs: refactor key marking code a bit

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

bcachefs: Fix double counting when gc is running

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

bcachefs: gc lock no longer needed for disk reservations

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

bcachefs: Fix oldest_gen handling

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

bcachefs: Fix a locking bug

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

bcachefs: Fix a lockdep splat

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

bcachefs: no need to run gc when initializing new fs

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

bcachefs: Don't need to walk inodes on clean shutdown

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

bcachefs: don't do initial gc if have alloc info feature

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

bcachefs: serialize persistent_reserved

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

bcachefs: initialize fs usage summary in recovery

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

bcachefs: journal_replay_early()

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

bcachefs: Write out fs usage

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

bcachefs: reserve space in journal for fs usage entries

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

bcachefs: fix bch2_sb_field_resize()

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

bcachefs: delete a debug printk

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

bcachefs: fix for unmount hang

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

bcachefs: Fix a bug when shutting down before allocator started

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

bcachefs: percpu utility code

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

bcachefs: sysfs trigger for bch2_alloc_write

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

bcachefs: New journal_entry_res mechanism

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

bcachefs: improve alloc_debug

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

bcachefs: fix inode counting

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

bcachefs: Persist stripe blocks_used

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

bcachefs: Improve c version of __bkey_cmp_bits

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

bcachefs: Persist alloc info on clean shutdown

 - Does not persist alloc info for stripes yet
 - Also does not yet include filesystem block/sector counts yet, from
struct fs_usage
 - Not made use of just yet

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

bcachefs: Fix fifo overflow in allocator startup

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

bcachefs: Fix check for if extent update is allocating

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

bcachefs: fix error message in device remove path

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

bcachefs: dio arithmetic improvements

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

bcachefs: Fix some reserve calculations

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

bcachefs: Add a workqueue for journal reclaim

journal reclaim writes btree nodes, which can end up waiting for in
flight btree writes to complete, and btree write completions run out of
workqueues - so we can't run out of the same workqueue or we risk
deadlock

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

bcachefs: Fix an allocator error path

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

bcachefs: fixes for getting stuck flushing journal pins

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

bcachefs: Fix a dio bug

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

bcachefs: More allocator startup improvements

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

bcachefs: correctly initialize bch_extent_ptr

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

bcachefs: use crc64 from lib/

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

bcachefs: Factor out acc_u64s()

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

bcachefs: Minor replicas.c refactoring

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

bcachefs: fix new reinherit_attrs ioctl

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

bcachefs: Lots of option handling improvements

Add helptext to option definitions - so  we can unify the option
handling with the format command

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

bcachefs: more project quota fixes

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

bcachefs: fix an rcu usage bug

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

bcachefs: fix ja->cur_idx use while reading journal

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

bcachefs: fix device remove error path

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

bcachefs: Fix duplicate ioctl nr

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

bcachefs: bch2_ioc_reinherit_attrs()

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

bcachefs: rename keeps inheritable inode opts consistent

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

bcachefs: add bcachefs_effective xattrs

Allows seeing xattrs that were inherited, not explicitly set

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

bcachefs: Add flags to indicate if inode opts were inherited or explicitly set

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

bcachefs: bch2_fs_quota_transfer

improve quota transfer locking & make ei_qid usage more consistent

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

bcachefs: merge BCH_INODE_FIELDS_INHERIT/BCH_INODE_OPTS

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

bcachefs: use x-macros more consistently

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

bcachefs: Fix for building in userspace

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

bcachefs: Fix bkey_method compilation on gcc 7.3.0

Signed-off-by: Tim Schlueter <schlueter.tim@linux.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

bcachefs: Include summarized counts in fs_usage

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

bcachefs: refactor bch_fs_usage

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

bcachefs: Switch replicas to mark_lock

Prep work for upcoming disk accounting changes

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

bcachefs: fix an incorrect bkey_debugcheck() call

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

bcachefs: improve extent debugcheck fn

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

bcachefs: improve/clarify ptr_disk_sectors()

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

bcachefs: Fix for running in degraded mode

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

bcachefs: Compression fixes

regressions from switching disk space accounting to be in compressed
sectors

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

bcachefs: propagate BCH_WRITE_CACHED

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

bcachefs: s/usage_lock/mark_lock

better describes what it's for, and we're going to call a new lock
usage_lock

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

bcachefs: New blockcount field for bch_stripe

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

bcachefs: move dirty into bucket_mark

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

bcachefs: Add new alloc fields

prep work for persistent alloc info

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

bcachefs: Deferred btree updates

Will be used in the future for inode updates, which will be very helpful
for multithreaded workloads that have to update the inode with every
extent update (appends, or updates that change i_sectors)

Also will be used eventually for fully persistent alloc info

However - we still need a mechanism for reserving space in the journal
prior to getting a journal reservation, so it's not technically safe to
make use of this just yet, we could deadlock with the journal full
(although not likely to be an issue in practice)

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

bcachefs: drop bogus percpu_ref_tryget

caller should already be guarding against rw, and checking here breaks
when caller needs to finish updates for going RO

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

bcachefs: Track nr_inodes with the key marking machinery

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

bcachefs: Make bkey types globally unique

this lets us get rid of a lot of extra switch statements - in a lot of
places we dispatch on the btree node type, and then the key type, so
this is a nice cleanup across a lot of code.

Also improve the on disk format versioning stuff.

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

bcachefs: Fix a btree iter usage error

previously, if the code traversed to the next btree node, that could
return an error (due to lock restarts) - which was not being checked
for.

fix is to rework it so it never iterates past the current leaf node, and
pops an assertion if it ever sees an error.

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

bcachefs: Split out bkey_sort.c

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

bcachefs: fix waiting on an open journal entry

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

bcachefs: return errors correctly from gc

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

bcachefs: Hold usage_lock over mark_key and fs_usage_apply

Fixes an inconsistency at the end of gc

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

bcachefs: Stripes now properly subject to gc

gc now verifies the contents of the stripes radix tree, important for
persistent alloc info

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

bcachefs: Btree locking fix, refactoring

Hit an assertion, probably spurious, indicating an iterator was unlocked
when it shouldn't have been (spurious because it wasn't locked at all
when the caller called btree_insert_at()).

Add a flag, BTREE_ITER_NOUNLOCK, and tighten up the assertions

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

bcachefs: shim for userspace raid library

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

Revert "bcachefs: start erasure coding after journal replay"

This reverts commit 36f389604294dfc953e6f5624ceb683818d32f28.

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

bcachefs: Allow for new alloc fields

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

bcachefs: gc now operates on second set of bucket marks

This means we can now use gc to verify the allocation information -
important for testing persistant alloc info

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