[SECURITY PATCH 036/117] zfs: Fix possible negative shift operation

[Daniel Kiper]

From: Darren Kenny <darren.kenny@oracle.com>

While it is possible for the return value from zfs_log2() to be zero
(0), it is quite unlikely, given that the previous assignment to blksz
is shifted up by SPA_MINBLOCKSHIFT (9) before 9 is subtracted at the
assignment to epbs.

But, while unlikely during a normal operation, it may be that a carefully
crafted ZFS filesystem could result in a zero (0) value to the
dn_datalbkszsec field, which means that the shift left does nothing
and assigns zero (0) to blksz, resulting in a negative epbs value.

Fixes: CID 73608

Signed-off-by: Darren Kenny <darren.kenny@oracle.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
---
 grub-core/fs/zfs/zfs.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/grub-core/fs/zfs/zfs.c b/grub-core/fs/zfs/zfs.c
index b6e1e178d..695e6ea30 100644
--- a/grub-core/fs/zfs/zfs.c
+++ b/grub-core/fs/zfs/zfs.c
@@ -2670,6 +2670,11 @@ dnode_get (dnode_end_t * mdn, grub_uint64_t objnum, 
grub_uint8_t type,
   blksz = grub_zfs_to_cpu16 (mdn->dn.dn_datablkszsec, 
                             mdn->endian) << SPA_MINBLOCKSHIFT;
   epbs = zfs_log2 (blksz) - DNODE_SHIFT;
+
+  /* While this should never happen, we should check that epbs is not 
negative. */
+  if (epbs < 0)
+    epbs = 0;
+
   blkid = objnum >> epbs;
   idx = objnum & ((1 << epbs) - 1);
 
-- 
2.11.0