diff mbox series

fs/ceph/addr: pass using_pgpriv2=false to fscache_write_to_cache()

Message ID 20240729090639.852732-1-max.kellermann@ionos.com (mailing list archive)
State New
Headers show
Series fs/ceph/addr: pass using_pgpriv2=false to fscache_write_to_cache() | expand

Commit Message

Max Kellermann July 29, 2024, 9:06 a.m. UTC
This piece was missing in commit ae678317b95e ("netfs: Remove
deprecated use of PG_private_2 as a second writeback flag").

There is one remaining use of PG_private_2: the function
__fscache_clear_page_bits(), whose only purpose is to clear
PG_private_2.  This is done via folio_end_private_2() which also
releases the folio reference which was supposed to be taken by
folio_start_private_2() (via ceph_set_page_fscache()).

__fscache_clear_page_bits() is called by __fscache_write_to_cache(),
but only if the parameter using_pgpriv2 is true; the only caller of
that function is ceph_fscache_write_to_cache() which still passes
true.

By calling folio_end_private_2() without folio_start_private_2(), the
folio refcounter breaks and causes trouble like RCU stalls and general
protection faults.

Cc: stable@vger.kernel.org
Fixes: ae678317b95e ("netfs: Remove deprecated use of PG_private_2 as a second writeback flag")
Link: https://lore.kernel.org/ceph-devel/CAKPOu+_DA8XiMAA2ApMj7Pyshve_YWknw8Hdt1=zCy9Y87R1qw@mail.gmail.com/
Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
---
 fs/ceph/addr.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

David Howells July 30, 2024, 2:26 p.m. UTC | #1
For the moment, ceph has to continue using PG_private_2.  It doesn't use
netfs_writepages().  I have mostly complete patches to fix that, but they got
popped onto the back burner for a bit.

I've finally managed to get cephfs set up and can now reproduce the hang
you're seeing.

David
David Howells July 30, 2024, 3:41 p.m. UTC | #2
I think the right thing to do is probably to at least partially revert:

	ae678317b95e760607c7b20b97c9cd4ca9ed6e1a
	netfs: Remove deprecated use of PG_private_2 as a second writeback flag

for the moment.  That removed the bit that actually did the write to the
cache on behalf of ceph.

David
diff mbox series

Patch

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index 8c16bc5250ef..aacea3e8fd6d 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -512,7 +512,7 @@  static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, b
 	struct fscache_cookie *cookie = ceph_fscache_cookie(ci);
 
 	fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode),
-			       ceph_fscache_write_terminated, inode, true, caching);
+			       ceph_fscache_write_terminated, inode, false, caching);
 }
 #else
 static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)