ext4: check superblock mapped prior to committing

commit a17712c8e4be4fa5404d20e9cd3b2b21eae7bc56 upstream.

This patch attempts to close a hole leading to a BUG seen with hot
removals during writes [1].

A block device (NVME namespace in this test case) is formatted to EXT4
without partitions. It's mounted and write I/O is run to a file, then
the device is hot removed from the slot. The superblock attempts to be
written to the drive which is no longer present.

The typical chain of events leading to the BUG:
ext4_commit_super()
  __sync_dirty_buffer()
    submit_bh()
      submit_bh_wbc()
        BUG_ON(!buffer_mapped(bh));

This fix checks for the superblock's buffer head being mapped prior to
syncing.

[1] https://www.spinics.net/lists/linux-ext4/msg56527.html

Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index b5eded6280bed1dcd89364eec9b57e2f9c4de657..535bbf0b4fe47c4ed22d83d1a8fed278073c59a8 100644 (file)
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -4653,6 +4653,14 @@ static int ext4_commit_super(struct super_block *sb, int sync)
 
        if (!sbh || block_device_ejected(sb))
                return error;
+
+       /*
+        * The superblock bh should be mapped, but it might not be if the
+        * device was hot-removed. Not much we can do but fail the I/O.
+        */
+       if (!buffer_mapped(sbh))
+               return error;
+
        /*
         * If the file system is mounted read-only, don't update the
         * superblock write time.  This avoids updating the superblock