iomap: fix WARN_ON_ONCE() from unprivileged users
authorQian Cai <cai@lca.pw>
Thu, 10 Sep 2020 15:26:15 +0000 (08:26 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 10 Sep 2020 15:26:15 +0000 (08:26 -0700)
It is trivial to trigger a WARN_ON_ONCE(1) in iomap_dio_actor() by
unprivileged users which would taint the kernel, or worse - panic if
panic_on_warn or panic_on_taint is set. Hence, just convert it to
pr_warn_ratelimited() to let users know their workloads are racing.
Thank Dave Chinner for the initial analysis of the racing reproducers.

Signed-off-by: Qian Cai <cai@lca.pw>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
fs/iomap/direct-io.c

index c1aafb2..9519113 100644 (file)
@@ -388,6 +388,16 @@ iomap_dio_actor(struct inode *inode, loff_t pos, loff_t length,
                return iomap_dio_bio_actor(inode, pos, length, dio, iomap);
        case IOMAP_INLINE:
                return iomap_dio_inline_actor(inode, pos, length, dio, iomap);
+       case IOMAP_DELALLOC:
+               /*
+                * DIO is not serialised against mmap() access at all, and so
+                * if the page_mkwrite occurs between the writeback and the
+                * iomap_apply() call in the DIO path, then it will see the
+                * DELALLOC block that the page-mkwrite allocated.
+                */
+               pr_warn_ratelimited("Direct I/O collision with buffered writes! File: %pD4 Comm: %.20s\n",
+                                   dio->iocb->ki_filp, current->comm);
+               return -EIO;
        default:
                WARN_ON_ONCE(1);
                return -EIO;