btrfs: fix wrong address when faulting in pages in the search ioctl

When faulting in the pages for the user supplied buffer for the search
ioctl, we are passing only the base address of the buffer to the function
fault_in_pages_writeable(). This means that after the first iteration of
the while loop that searches for leaves, when we have a non-zero offset,
stored in 'sk_offset', we try to fault in a wrong page range.

So fix this by adding the offset in 'sk_offset' to the base address of the
user supplied buffer when calling fault_in_pages_writeable().

Several users have reported that the applications compsize and bees have
started to operate incorrectly since commit a48b73eca4ceb9 ("btrfs: fix
potential deadlock in the search ioctl") was added to stable trees, and
these applications make heavy use of the search ioctls. This fixes their
issues.

Link: https://lore.kernel.org/linux-btrfs/[email protected]/
Link: https://github.com/kilobyte/compsize/issues/34
Fixes: a48b73eca4ceb9 ("btrfs: fix potential deadlock in the search ioctl")
CC: [email protected] # 4.4+
Tested-by: A L <[email protected]>
Reviewed-by: Josef Bacik <[email protected]>
Signed-off-by: Filipe Manana <[email protected]>
Reviewed-by: David Sterba <[email protected]>
Signed-off-by: David Sterba <[email protected]>

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index ac45f022b495886a995c06f9722ec3d71152e436..2d9109d9e98f9dae41b29d2969cb9cc1345e26e5 100644 (file)
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -2193,7 +2193,8 @@ static noinline int search_ioctl(struct inode *inode,
        key.offset = sk->min_offset;
 
        while (1) {
-               ret = fault_in_pages_writeable(ubuf, *buf_size - sk_offset);
+               ret = fault_in_pages_writeable(ubuf + sk_offset,
+                                              *buf_size - sk_offset);
                if (ret)
                        break;