| Age | Commit message (Collapse) | Author |
|---|
|
In the buffered write path, we have to check for short writes that write
to the full page, where the page wasn't UpToDate; when this happens, the
page is partly garbage, so we have to zero it out and revert that part
of the write.
This check was wrong - we reverted total from copied, but didn't revert
the iov_iter, probably also leading to corrupted writes.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This solves internal fragmentation within stripes. We already have
copygc, which evacuates buckets that are partially or mostly empty, but
it's up to the ec code that manages stripes to deal with stripes that
have empty buckets in them.
This patch changes the path for creating new stripes to check if there's
existing stripes with empty buckets - and if so, update them with new
data buckets instead of creating new stripes.
TODO: improve the disk space accounting so that we can only use this
(more expensive path) when we have too much fragmentation in existing
stripes.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Prep work for the patch to update existing stripes with new data blocks.
This moves allocating new stripes into ec.c, and also sets up the data
structures so that we can handly only allocating some of the blocks in a
stripe.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is mainly to solve a lock ordering issue, and also simplifies the
code a bit.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Soon we'll be able to modify existing stripes - replacing empty blocks
with new blocks and new p/q blocks. This patch updates the trigger code
to handle pointers changing in an existing stripe; also, it
significantly improves how the stripes heap works, which means we can
get rid of the stripe creation/deletion lock.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
The trigger for stripe keys is shortly going to need both the old and
the new key passed to the trigger - this patch does that rework.
For now, this just changes the in memory triggers, and this doesn't
change how extent triggers work.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is prep work for reworking the triggers machinery - we have
triggers that need to know both the old and the new key.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This fixes a bug where recovery fails when one of the devices is read
only.
Also - consolidate the "must rewrite this node to insert it" behind a
new btree node flag.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Improved error messages are always a good thing
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
If we're going to the trouble of having these nice error strings, let's
make them available.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Journal replay just treats extents like regular keys - it doesn't split
existing extents, so bch2_extent_can_insert() shouldn't be called from
that path.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
It appears this was erronious, a different bug was responsible
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This fixes a bug where the BCH_WRITE_SKIP_CLOSURE_PUT was set
incorrectly, causing the completion to be delivered multiple times.
oops.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
It's supposed to check whether we're splitting a compressed extent and
if so get a bigger disk reservation - hence this fixes a "disk usage
increased by x without a reservaiton" bug.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We use sentinal values that aren't NULL to indicate there's a btree node
at a higher level; occasionally, this may result in
btree_iter_up_until_good_node() stopping at one of those sentinal
values.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Whenever we're doing an update that has pointers, that generally means
we need to do the update in order to release open bucket references - so
we should be using the btree open bucket reserve.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Turns out it's possible during shutdown for the allocator to get stuck
spinning on bch2_invalidate_buckets() without hitting any of the other
checks.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
The code that checks lock ordering was recently changed to go off of the
pos of the btree node, rather than the iterator, but the btree cache
code didn't update to handle iterators that point to cached bkeys. Oops
Also, update various debug code.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
state lock isn't held at startup
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Also tweak the allocator to be more aggressive about keeping it full.
The recent changes to make updates to interior nodes transactional (and
thus generate updates to the alloc btree) all put more stress on the
btree node reserves.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This introduces a new kind of btree iterator, cached iterators, which
point to keys cached in a hash table. The cache also acts as a write
cache - in the update path, we journal the update but defer updating the
btree until the cached entry is flushed by journal reclaim.
Cache coherency is for now up to the users to handle, which isn't ideal
but should be good enough for now.
These new iterators will be used for updating inodes and alloc info (the
alloc and stripes btrees).
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
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@gmail.com>
|
|
Full mark and sweep gc doesn't (yet?) work with the new btree key cache
code, but it also blocks updates to interior btree nodes for the
duration and isn't really necessary in practice; we aren't currently
attempting to repair errors in allocation info at runtime.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
|
|
|
|
To be used the debug tool that dumps the contents of the journal.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Btree node lock ordering is based on the logical key. However, 'struct
btree' may be reused for a different btree node under memory pressure.
This patch uses the new six lock callback to check if a btree node is no
longer the node we wanted to lock before blocking.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This allows for functionality like mutex_lock_killable(), but in a
generic manner: to be used for deadlock avoidance in bcachefs.
Also add a function for waking up all waiters on a lock, causing them to
recheck if they should bail out.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
__bch2_btree_node_lock() was incorrectly using iter->pos as a proxy for
btree node lock ordering, this caused an off by one error that was
triggered by bch2_btree_node_get_sibling() getting the previous node.
This refactors the code to compare against btree node keys directly.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
|
|
We've been seeing btree updates get stuck, due to some sort of bug; when
this happens, buffered writeback will keep queueing up writes that lead
to the system running out of memory.
Not sure if this kind of throttling is something we'll want to keep and
improve, or get rid of when the bug with btree updates getting stuck is
fixed. For now it should make debugging easier.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This splits out the journalling code from the btree update code; prep
work for the btree key cache.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This is better than skipping the journal pre-reservation if we already
have one - we should still acount for the journal reservation we're
going to have to get.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
We need a larger open bucket reserve now that the btree interior update
path holds onto open bucket references; filesystems with many high
through devices may need more open buckets now.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
The btree cache lock is needed for reclaiming from the btree node cache,
and memory allocation can potentially spin and sleep (for 100 ms at a
time), so.. don't do that.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Also, in the btree_update_start() path, if we already have a journal
pre-reservation we don't want to take another - that's a deadlock.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
bch2_btree_node_mem_alloc() uses memalloc_nofs_save()/GFP_NOFS, but
GFP_NOFS does include __GFP_IO - oops. We used to use GFP_NOIO, but as
we're a filesystem now GFP_NOFS makes more sense now and is looser.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
bch2_btree_iter_downgrade() was looping over all iterators in a
transaction; bch2_trans_downgrade() should be doing that.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
|
This will help narrow down which code is at fault when this happens.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
|
