bcachefs: Fix bch2_sort_repack_merge()

bch2_bkey_normalize() modifies the value, and we were modifying the
original value in the src btree node - but, we're called without a write
lock held on the src node. Oops...

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

bcachefs: Reflink

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

bcachefs: Refactor bch2_extent_trim_atomic() for reflink

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

bcachefs: Move node iterator fixup to extent_bset_insert()

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

bcachefs: Fix bch2_btree_node_iter_fix()

bch2_btree_node_iter_prev_all() depends on an invariant that wasn't
being maintained for extent leaf nodes - specifically, the node iterator
may not have advanced past any keys that compare after the key the node
iterator points to.

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

bcachefs: Fix bch2_btree_node_iter_prev_filter()

bch2_btree_node_iter_prev_filter() tried to be smart about iterating
backwards when skipping over whiteouts/discards - but unfortunately,
doing so can leave the node iterator in an inconsistent state; the sane
solution is to just always iterate backwards one key at a time.

But we compact btree nodes when more than a quarter of the keys are
whiteouts/discards, so the optimization wasn't buying us that much
anyways.

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

bcachefs: Fix __bch2_btree_iter_peek_slot_extents()

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

bcachefs: Improved debug checks

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

bcachefs: Rework calling convention for marking overwrites

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

bcachefs: Mark space as unallocated on write failure

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

bcachefs: Ensure bch2_trans_get_iter() returns iters with correct locks

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

bcachefs: fix for_each_btree_key()

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

bcachefs: trans_get_key() now works correctly for extents

More prep work for reflink: for extents, we're not looking for an exact
mach on pos, rather that the pos is within the range of the key the
iterator points to.

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

bcachefs: Export correct blocksize to vfs

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

bcachefs: Truncate/fpunch now works on block boundaries, not page

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

bcachefs: Count reserved extents as holes

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

bcachefs: Handle partial pages in seek data/hole

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

bcachefs: Change buffered write path to write to partial pages

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

bcachefs: Change __bch2_writepage() to not write to holes

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

bcachefs: Fix bch2_seek_data()

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

bcachefs: Refactor various code to not be extent specific

With reflink, various code now has to handle both KEY_TYPE_extent
or KEY_TYPE_reflink_v - so, convert it to be generic across all keys
with pointers.

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

bcachefs: Refactor __bch2_cut_front()

Minor cleanup - prep work for new key types for reflink

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

bcachefs: Dont't call bch2_trans_begin_updates() in bch2_extent_update()

Prep work for reflink - for reflink, we're going to be using
bch2_extent_update() with other updates in the same transaction.

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

bcachefs: Don't unlink iters on unsuccessful commit

Where unlink_on_commit is used, on unsuccessfull commit we're likely
retrying the whole update and were going to be using the same iterators
again.

The management of multiple iterators needs to be gone over a fair bit
more at some point...

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

bcachefs: add missing bch2_trans_begin() call

for_each_btree_key() calls bch2_trans_get_iter() - we have to reset the
transaction state before getting the iterator again, in the retry path

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

bcachefs: Add offset_into_extent param to bch2_read_extent()

With reflink, we'll no longer be able to calculate the offset of the
data we want into the extent we're reading from from the extent pos and
the iter pos - we'll have to pass it in separately.

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

bcachefs: Print out name of bkey type

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

bcachefs: Don't overflow trans with iters from triggers

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

bcachefs: Convert some assertions to fsck errors

Actual repair code will come later, but this is a start

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

bcachefs: Fix ec_stripes_read()

Change it to not mark keys that will be overwritten by keys in the
journal - this fixes a bug where we pop an assertion in
bucket_set_stripe() because of a stale pointer - because the stripe that
has the stale pointer has been deleted.

This code could be factored out and used elsewhere, at some point.

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

bcachefs: Fix stripe_idx_to_delete()

There was a null ptr deref when there wasn't a stripes heap allocated

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

bcachefs: Don't try to delete stripes when RO

We weren't checking for errors when trying to delet stripes, which meant
ec_stripe_delete_work() would spin trying to delete the same stripe over
and over.

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

bcachefs: Track dirtyness at sector level, not page

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

bcachefs: Kill page_state_cmpxchg

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

bcachefs: Always touch page state with page locked

This will mean we don't have to use cmpxchg for modifying page state,
which will simplify a fair amount of code

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

bcachefs: Fix moving compressed data

bio_uncompress_inplace() used to potentially need to extend the bio to
be big enough for the uncompressed data, which has become problematic
with multipage bvecs - but, the move extent path actually already
allocated the bios to be big enough for the uncompressed data.

The promote path needed to be fixed, though.

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

bcachefs: kill bio_for_each_contig_segment()

With multipage bvecs it's not needed anymore

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

bcachefs: Kill direct access to bi_io_vec

Switch to always using bio_add_page(), which merges contiguous pages now
that we have multipage bvecs.

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

bcachefs: kill bch2_crc64_update

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

bcachefs: Fix for building with old gcc

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

bcachefs: Improve bch2_lock_inodes()

Can now be used for the two different types of locks we have so far

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

bcachefs: Check for key size > offset

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

bcachefs: Refactor trans_(get|update)_key

these are still pretty ugly...

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

bcachefs: Update path now handles triggers that generate more triggers

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

bcachefs: add inode_generation_to_text method

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

bcachefs: Fix array overrun with unknown btree roots

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

bcachefs: fix kasan splat

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

bcachefs: Delete a spurious assertion

bch_write_op->written used to be a u16, but it's not so the assertion
isn't needed anymore - and 5.1 can send larger bios.

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

bcachefs: fix __bch2_xattr_bcachefs_get()

We were returning -ERANGE when the size of the buffer passed in was
exactly the size of the xattr val

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

bcachefs: use memalloc_nofs_save() for vmalloc allocation

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

bcachefs: Fix promoting to cache devices (durability = 0)

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

bcachefs: Don't allow bkey vals that are too big in extents btree

Make sure we don't overflow BKEY_PADDED keys

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

bcachefs: Reduce BKEY_PADDED usage

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

bcachefs: bkey_merge() now takes bkey_s

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

bcachefs: Don't overflow stack in bch2_extent_merge_inline()

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

bcachefs: Fix starting copygc when already started

We can sometimes call bch2_dev_read_write() when the device is already
RW (in error paths).

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

bcachefs: Fix an error path in bch2_btree_iter_traverse()

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

bcachefs: Improve key marking interface

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

bcachefs: Fix cached sectors not being updated on invalidate

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

bcachefs: better BTREE_INSERT_NO_CLEAR_REPLICAS

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

bcachefs: fix a mount error path

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

bcachefs: fix bch2_extent_merge()

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

bcachefs: Don't use a fixed size buffer for fs_usage_deltas

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

bcachefs: Rip out old hacky transaction restart tracing

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

bcachefs: More work to avoid transaction restarts

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

bcachefs: Avoid spurious transaction restarts

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

bcachefs: Ensure bch2_btree_iter_next() always advances

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

bcachefs: Fix a bug with spinning on the journal

Transactional triggers meant that when we failed to get a journal
reservation, then bailed out into the error path to block on a journal
reservation, the second blocking call into the journal code was asking
for less space, which is not what we want.

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

bcachefs: Delete duplicate code

Also rename for consistency

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

bcachefs: Fix spurious inconsistency in recovery

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

bcachefs: improved btree locking tracepoints

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

bcachefs: bch2_btree_delete_at_range()

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

bcachefs: stripe creation fixes

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

bcachefs: Merge extents with checksums

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

bcachefs: kill BTREE_ITER_NOUNLOCK

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

bcachefs: Don't pass around may_drop_locks

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

bcachefs: Kill remaining bch2_btree_iter_unlock() uses

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

bcachefs: Fix for the stripes mark path and gc

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

bcachefs: Redo replicas gc mechanism

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

bcachefs: Fix return code from bch2_fs_start()

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

bcachefs: Don't run fsck by default at mount time

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

bcachefs: some improvements to startup messages and options

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

bcachefs: Various improvements to bch2_alloc_write()

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

bcachefs: bch2_trans_mark_update()

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

bcachefs: btree_bkey_cached_common

This is prep work for the btree key cache: btree iterators will point to
either struct btree, or a new struct bkey_cached.

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

bcachefs: Write out fs usage consistently

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

bcachefs: Avoid write lock on mark_lock

mark_lock is a frequently taken lock, and there's also potential for
deadlocks since currently bch2_clear_page_bits which is called from
memory reclaim has to take it to drop disk reservations.

The disk reservation get path takes it when it recalculates the number
of sectors known to be available, but it's not really needed for
consistency.  We just want to make sure we only have one thread updating
the sectors_available count, which we can do with a dedicated mutex.

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

bcachefs: Fix a bug with multiple iterators being traversed

If upgrade fails on one iterator, but it was copied from another
iterator and will be freed before transaction restart, then the original
iterator will get traversed first, so we need to make required btree
nodes on the original iterator will be traversed too.

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

bcachefs: fix bch2_rbio_narrow_crcs()

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

bcachefs: Fix journal shutdown path

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

bcachefs: Fix fsync error reporting

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

bcachefs: bch2_bkey_ptrs_invalid()

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

bcachefs: Don't hardcode BTREE_ID_EXTENTS

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

bcachefs: fix bch2_flags_to_text()

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

bcachefs: Add actual tracepoints for transaction restarts

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

bcachefs: copy correct journal_seq to dir in create

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

bcachefs: Return errors from for_each_btree_key()

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

bcachefs: fix triggers for stripes btree

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

bcachefs: use same timesource as current_time()

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

bcachefs: Fix hang while shutting down

If the allocator thread exited before bch2_dev_allocator_stop() was
called (because of an error), bch2_dev_allocator_quiesce() could hang.

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