folkslinux
/
zfs
mirror of https://github.com/openzfs/zfs.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

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 <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@ixsystems.com>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes #10519
This commit is contained in:
Mark Johnston 2020-06-30 15:54:42 -04:00 committed by Brian Behlendorf
parent 6e00561712
commit cd32b4f5b7
1 changed files with 12 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
module/os/freebsd/zfs/zfs_vnops.c
View File

@ -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);