xfs: prevent spoofing of rtbitmap blocks when recovering buffers
authorDarrick J. Wong <djwong@kernel.org>
Mon, 26 Jul 2021 23:43:17 +0000 (16:43 -0700)
committerDarrick J. Wong <djwong@kernel.org>
Thu, 29 Jul 2021 16:27:29 +0000 (09:27 -0700)
While reviewing the buffer item recovery code, the thought occurred to
me: in V5 filesystems we use log sequence number (LSN) tracking to avoid
replaying older metadata updates against newer log items.  However, we
use the magic number of the ondisk buffer to find the LSN of the ondisk
metadata, which means that if an attacker can control the layout of the
realtime device precisely enough that the start of an rt bitmap block
matches the magic and UUID of some other kind of block, they can control
the purported LSN of that spoofed block and thereby break log replay.

Since realtime bitmap and summary blocks don't have headers at all, we
have no way to tell if a block really should be replayed.  The best we
can do is replay unconditionally and hope for the best.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
fs/xfs/xfs_buf_item_recover.c

index 05fd816..4775485 100644 (file)
@@ -698,7 +698,8 @@ xlog_recover_do_inode_buffer(
 static xfs_lsn_t
 xlog_recover_get_buf_lsn(
        struct xfs_mount        *mp,
-       struct xfs_buf          *bp)
+       struct xfs_buf          *bp,
+       struct xfs_buf_log_format *buf_f)
 {
        uint32_t                magic32;
        uint16_t                magic16;
@@ -706,11 +707,20 @@ xlog_recover_get_buf_lsn(
        void                    *blk = bp->b_addr;
        uuid_t                  *uuid;
        xfs_lsn_t               lsn = -1;
+       uint16_t                blft;
 
        /* v4 filesystems always recover immediately */
        if (!xfs_sb_version_hascrc(&mp->m_sb))
                goto recover_immediately;
 
+       /*
+        * realtime bitmap and summary file blocks do not have magic numbers or
+        * UUIDs, so we must recover them immediately.
+        */
+       blft = xfs_blft_from_flags(buf_f);
+       if (blft == XFS_BLFT_RTBITMAP_BUF || blft == XFS_BLFT_RTSUMMARY_BUF)
+               goto recover_immediately;
+
        magic32 = be32_to_cpu(*(__be32 *)blk);
        switch (magic32) {
        case XFS_ABTB_CRC_MAGIC:
@@ -920,7 +930,7 @@ xlog_recover_buf_commit_pass2(
         * the verifier will be reset to match whatever recover turns that
         * buffer into.
         */
-       lsn = xlog_recover_get_buf_lsn(mp, bp);
+       lsn = xlog_recover_get_buf_lsn(mp, bp, buf_f);
        if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
                trace_xfs_log_recover_buf_skip(log, buf_f);
                xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN);