diff options
| author | Kent Overstreet <kent.overstreet@gmail.com> | 2020-07-22 22:40:32 -0400 |
|---|---|---|
| committer | Kent Overstreet <kent.overstreet@linux.dev> | 2023-10-22 17:08:43 -0400 |
| commit | 33e339619f7fda8c428daa5cb8fde7d68dad2edb (patch) | |
| tree | 11d3c064aeeb90f170dadd51451eb6e45d0809f6 /fs/bcachefs | |
| parent | 74ed7e560b794369adf87e0d310453bc78f4b273 (diff) | |
bcachefs: Fix a race with BCH_WRITE_SKIP_CLOSURE_PUT
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Diffstat (limited to 'fs/bcachefs')
| -rw-r--r-- | fs/bcachefs/io.c | 10 |
1 files changed, 10 insertions, 0 deletions
diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c index 7e57ca2e1071..d9e35329f707 100644 --- a/fs/bcachefs/io.c +++ b/fs/bcachefs/io.c @@ -1110,6 +1110,16 @@ again: goto flush_io; } + /* + * It's possible for the allocator to fail, put us on the + * freelist waitlist, and then succeed in one of various retry + * paths: if that happens, we need to disable the skip_put + * optimization because otherwise there won't necessarily be a + * barrier before we free the bch_write_op: + */ + if (atomic_read(&cl->remaining) & CLOSURE_WAITING) + skip_put = false; + bch2_open_bucket_get(c, wp, &op->open_buckets); ret = bch2_write_extent(op, wp, &bio); bch2_alloc_sectors_done(c, wp); |