diff options
| author | Mark Johnston <[email protected]> | 2020-06-30 15:54:42 -0400 |
|---|---|---|
| committer | Brian Behlendorf <[email protected]> | 2020-07-06 11:53:57 -0700 |
| commit | cd32b4f5b79c97b293f7be3fe9ddfc9024f7d734 (patch) | |
| tree | 3b2eb31b6c5b0dc21d3848a194e50fe277d4f493 /module | |
| parent | 6e0056171234b84450af2afbb6594bd3b09422b5 (diff) | |
Fix a deadlock in the FreeBSD getpages VOP
FreeBSD has a per-page "busy" lock which is held when handling a page
fault on a mapped file. This lock is also acquired when copying data
from the DMU to the page cache in zfs_write(). File range locks are
also acquired in both of these paths, in the opposite order with respect
to the busy lock.
In the getpages VOP, the range lock is only used to determine the extent
of optional read-ahead and read-behind operations. To resolve the lock
order reversal, modify the getpages VOP to avoid blocking on the range
lock.
Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Ryan Moeller <[email protected]>
Signed-off-by: Mark Johnston <[email protected]>
Closes #10519
Diffstat (limited to 'module')
| -rw-r--r-- | module/os/freebsd/zfs/zfs_vnops.c | 13 |
1 files changed, 12 insertions, 1 deletions
diff --git a/module/os/freebsd/zfs/zfs_vnops.c b/module/os/freebsd/zfs/zfs_vnops.c index 6698e3655..1f1c0cb93 100644 --- a/module/os/freebsd/zfs/zfs_vnops.c +++ b/module/os/freebsd/zfs/zfs_vnops.c @@ -4809,9 +4809,20 @@ zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, */ for (;;) { blksz = zp->z_blksz; - lr = zfs_rangelock_enter(&zp->z_rangelock, + lr = zfs_rangelock_tryenter(&zp->z_rangelock, rounddown(start, blksz), roundup(end, blksz) - rounddown(start, blksz), RL_READER); + if (lr == NULL) { + if (rahead != NULL) { + *rahead = 0; + rahead = NULL; + } + if (rbehind != NULL) { + *rbehind = 0; + rbehind = NULL; + } + break; + } if (blksz == zp->z_blksz) break; zfs_rangelock_exit(lr); |