ext4, jbd2: ensure entering into panic after recording an error in superblock
authorDaeho Jeong <daeho.jeong@samsung.com>
Sun, 18 Oct 2015 21:02:56 +0000 (17:02 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 15 Dec 2015 05:24:35 +0000 (21:24 -0800)
commit 4327ba52afd03fc4b5afa0ee1d774c9c5b0e85c5 upstream.

If a EXT4 filesystem utilizes JBD2 journaling and an error occurs, the
journaling will be aborted first and the error number will be recorded
into JBD2 superblock and, finally, the system will enter into the
panic state in "errors=panic" option.  But, in the rare case, this
sequence is little twisted like the below figure and it will happen
that the system enters into panic state, which means the system reset
in mobile environment, before completion of recording an error in the
journal superblock. In this case, e2fsck cannot recognize that the
filesystem failure occurred in the previous run and the corruption
wouldn't be fixed.

Task A                        Task B
ext4_handle_error()
-> jbd2_journal_abort()
  -> __journal_abort_soft()
    -> __jbd2_journal_abort_hard()
    | -> journal->j_flags |= JBD2_ABORT;
    |
    |                         __ext4_abort()
    |                         -> jbd2_journal_abort()
    |                         | -> __journal_abort_soft()
    |                         |   -> if (journal->j_flags & JBD2_ABORT)
    |                         |           return;
    |                         -> panic()
    |
    -> jbd2_journal_update_sb_errno()

Tested-by: Hobin Woo <hobin.woo@samsung.com>
Signed-off-by: Daeho Jeong <daeho.jeong@samsung.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/ext4/super.c
fs/jbd2/journal.c
include/linux/jbd2.h

index ff89971e3ee0fdc1b27a75ce54c9ca1c5ce81234..8a3b9f14d198831dc76dabccfc97134e9c609da0 100644 (file)
@@ -396,9 +396,13 @@ static void ext4_handle_error(struct super_block *sb)
                smp_wmb();
                sb->s_flags |= MS_RDONLY;
        }
-       if (test_opt(sb, ERRORS_PANIC))
+       if (test_opt(sb, ERRORS_PANIC)) {
+               if (EXT4_SB(sb)->s_journal &&
+                 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
+                       return;
                panic("EXT4-fs (device %s): panic forced after error\n",
                        sb->s_id);
+       }
 }
 
 #define ext4_error_ratelimit(sb)                                       \
@@ -587,8 +591,12 @@ void __ext4_abort(struct super_block *sb, const char *function,
                        jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
                save_error_info(sb, function, line);
        }
-       if (test_opt(sb, ERRORS_PANIC))
+       if (test_opt(sb, ERRORS_PANIC)) {
+               if (EXT4_SB(sb)->s_journal &&
+                 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
+                       return;
                panic("EXT4-fs panic from previous error\n");
+       }
 }
 
 void __ext4_msg(struct super_block *sb,
index 7003c09257601c78c7ebeac08555a13dd11ba8c2..0469f32918a5a78e0a2dd2a66fa187c882cc5dae 100644 (file)
@@ -2086,8 +2086,12 @@ static void __journal_abort_soft (journal_t *journal, int errno)
 
        __jbd2_journal_abort_hard(journal);
 
-       if (errno)
+       if (errno) {
                jbd2_journal_update_sb_errno(journal);
+               write_lock(&journal->j_state_lock);
+               journal->j_flags |= JBD2_REC_ERR;
+               write_unlock(&journal->j_state_lock);
+       }
 }
 
 /**
index eb1cebed3f36e84a6d93791670797749f65a73fa..c90c9b70e568589bbdfa1449f8226c4e2134525d 100644 (file)
@@ -1007,6 +1007,7 @@ struct journal_s
 #define JBD2_ABORT_ON_SYNCDATA_ERR     0x040   /* Abort the journal on file
                                                 * data write error in ordered
                                                 * mode */
+#define JBD2_REC_ERR   0x080   /* The errno in the sb has been recorded */
 
 /*
  * Function declarations for the journaling transaction and buffer