Author: mjg
Date: Wed Nov 23 19:50:12 2016
New Revision: 309067
URL: https://svnweb.freebsd.org/changeset/base/309067
Log:
cache: ensure that the number of bucket locks does not exceed hash size
The size can be changed by side effect of modifying kern.maxvnodes.
Since numbucketlocks was not modified, setting a sufficiently low value
would give more locks than actual buckets, which would then lead to
corruption.
Force the number of buckets to be not smaller.
Note this should not matter for real world cases.
Reported and tested by: pho
Modified:
head/sys/kern/vfs_cache.c
Modified: head/sys/kern/vfs_cache.c
==============================================================================
--- head/sys/kern/vfs_cache.c Wed Nov 23 19:19:11 2016 (r309066)
+++ head/sys/kern/vfs_cache.c Wed Nov 23 19:50:12 2016 (r309067)
@@ -1780,6 +1780,8 @@ nchinit(void *dummy __unused)
nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
numbucketlocks = cache_roundup_2(mp_ncpus * 64);
+ if (numbucketlocks > nchash + 1)
+ numbucketlocks = nchash + 1;
bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE,
M_WAITOK | M_ZERO);
for (i = 0; i < numbucketlocks; i++)
@@ -1828,7 +1830,11 @@ cache_changesize(int newmaxvnodes)
uint32_t hash;
int i;
- new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash);
+ newmaxvnodes = cache_roundup_2(newmaxvnodes * 2);
+ if (newmaxvnodes < numbucketlocks)
+ newmaxvnodes = numbucketlocks;
+
+ new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash);
/* If same hash table size, nothing to do */
if (nchash == new_nchash) {
free(new_nchashtbl, M_VFSCACHE);
_______________________________________________
[email protected] mailing list
https://lists.freebsd.org/mailman/listinfo/svn-src-all
To unsubscribe, send any mail to "[email protected]"
