Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934928AbbHJKdf (ORCPT ); Mon, 10 Aug 2015 06:33:35 -0400 Received: from shadbolt.e.decadent.org.uk ([88.96.1.126]:44069 "EHLO shadbolt.e.decadent.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933944AbbHJKQv (ORCPT ); Mon, 10 Aug 2015 06:16:51 -0400 Content-Type: text/plain; charset="UTF-8" Content-Disposition: inline Content-Transfer-Encoding: 8bit MIME-Version: 1.0 From: Ben Hutchings To: linux-kernel@vger.kernel.org, stable@vger.kernel.org CC: akpm@linux-foundation.org, "Jan Kara" , "Theodore Ts'o" Date: Mon, 10 Aug 2015 12:12:31 +0200 Message-ID: X-Mailer: LinuxStableQueue (scripts by bwh) Subject: [PATCH 3.2 032/110] jbd2: issue cache flush after checkpointing even with internal journal In-Reply-To: X-SA-Exim-Connect-IP: 2.173.94.72 X-SA-Exim-Mail-From: ben@decadent.org.uk X-SA-Exim-Scanned: No (on shadbolt.decadent.org.uk); SAEximRunCond expanded to false Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 14575 Lines: 406 3.2.71-rc1 review patch. If anyone has any objections, please let me know. ------------------ From: Jan Kara commit 79feb521a44705262d15cc819a4117a447b11ea7 upstream. When we reach jbd2_cleanup_journal_tail(), there is no guarantee that checkpointed buffers are on a stable storage - especially if buffers were written out by jbd2_log_do_checkpoint(), they are likely to be only in disk's caches. Thus when we update journal superblock effectively removing old transaction from journal, this write of superblock can get to stable storage before those checkpointed buffers which can result in filesystem corruption after a crash. Thus we must unconditionally issue a cache flush before we update journal superblock in these cases. A similar problem can also occur if journal superblock is written only in disk's caches, other transaction starts reusing space of the transaction cleaned from the log and power failure happens. Subsequent journal replay would still try to replay the old transaction but some of it's blocks may be already overwritten by the new transaction. For this reason we must use WRITE_FUA when updating log tail and we must first write new log tail to disk and update in-memory information only after that. Signed-off-by: Jan Kara Signed-off-by: "Theodore Ts'o" [bwh: Prerequisite for "jbd2: fix ocfs2 corrupt when updating journal superblock fails". Backported to 3.2: - Adjust context - Drop changes to jbd2_journal_update_sb_log_tail trace event] Signed-off-by: Ben Hutchings --- --- a/fs/jbd2/checkpoint.c +++ b/fs/jbd2/checkpoint.c @@ -478,79 +478,28 @@ out: int jbd2_cleanup_journal_tail(journal_t *journal) { - transaction_t * transaction; tid_t first_tid; - unsigned long blocknr, freed; + unsigned long blocknr; if (is_journal_aborted(journal)) return 1; - /* OK, work out the oldest transaction remaining in the log, and - * the log block it starts at. - * - * If the log is now empty, we need to work out which is the - * next transaction ID we will write, and where it will - * start. */ - - write_lock(&journal->j_state_lock); - spin_lock(&journal->j_list_lock); - transaction = journal->j_checkpoint_transactions; - if (transaction) { - first_tid = transaction->t_tid; - blocknr = transaction->t_log_start; - } else if ((transaction = journal->j_committing_transaction) != NULL) { - first_tid = transaction->t_tid; - blocknr = transaction->t_log_start; - } else if ((transaction = journal->j_running_transaction) != NULL) { - first_tid = transaction->t_tid; - blocknr = journal->j_head; - } else { - first_tid = journal->j_transaction_sequence; - blocknr = journal->j_head; - } - spin_unlock(&journal->j_list_lock); - J_ASSERT(blocknr != 0); - - /* If the oldest pinned transaction is at the tail of the log - already then there's not much we can do right now. */ - if (journal->j_tail_sequence == first_tid) { - write_unlock(&journal->j_state_lock); + if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr)) return 1; - } - - /* OK, update the superblock to recover the freed space. - * Physical blocks come first: have we wrapped beyond the end of - * the log? */ - freed = blocknr - journal->j_tail; - if (blocknr < journal->j_tail) - freed = freed + journal->j_last - journal->j_first; - - trace_jbd2_cleanup_journal_tail(journal, first_tid, blocknr, freed); - jbd_debug(1, - "Cleaning journal tail from %d to %d (offset %lu), " - "freeing %lu\n", - journal->j_tail_sequence, first_tid, blocknr, freed); - - journal->j_free += freed; - journal->j_tail_sequence = first_tid; - journal->j_tail = blocknr; - write_unlock(&journal->j_state_lock); + J_ASSERT(blocknr != 0); /* - * If there is an external journal, we need to make sure that - * any data blocks that were recently written out --- perhaps - * by jbd2_log_do_checkpoint() --- are flushed out before we - * drop the transactions from the external journal. It's - * unlikely this will be necessary, especially with a - * appropriately sized journal, but we need this to guarantee - * correctness. Fortunately jbd2_cleanup_journal_tail() - * doesn't get called all that often. + * We need to make sure that any blocks that were recently written out + * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before + * we drop the transactions from the journal. It's unlikely this will + * be necessary, especially with an appropriately sized journal, but we + * need this to guarantee correctness. Fortunately + * jbd2_cleanup_journal_tail() doesn't get called all that often. */ - if ((journal->j_fs_dev != journal->j_dev) && - (journal->j_flags & JBD2_BARRIER)) + if (journal->j_flags & JBD2_BARRIER) blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); - if (!(journal->j_flags & JBD2_ABORT)) - jbd2_journal_update_sb_log_tail(journal); + + __jbd2_update_log_tail(journal, first_tid, blocknr); return 0; } --- a/fs/jbd2/commit.c +++ b/fs/jbd2/commit.c @@ -340,7 +340,16 @@ void jbd2_journal_commit_transaction(jou /* Do we need to erase the effects of a prior jbd2_journal_flush? */ if (journal->j_flags & JBD2_FLUSHED) { jbd_debug(3, "super block updated\n"); - jbd2_journal_update_sb_log_tail(journal); + /* + * We hold j_checkpoint_mutex so tail cannot change under us. + * We don't need any special data guarantees for writing sb + * since journal is empty and it is ok for write to be + * flushed only with transaction commit. + */ + jbd2_journal_update_sb_log_tail(journal, + journal->j_tail_sequence, + journal->j_tail, + WRITE_SYNC); } else { jbd_debug(3, "superblock not updated\n"); } --- a/fs/jbd2/journal.c +++ b/fs/jbd2/journal.c @@ -775,6 +775,85 @@ struct journal_head *jbd2_journal_get_de return jbd2_journal_add_journal_head(bh); } +/* + * Return tid of the oldest transaction in the journal and block in the journal + * where the transaction starts. + * + * If the journal is now empty, return which will be the next transaction ID + * we will write and where will that transaction start. + * + * The return value is 0 if journal tail cannot be pushed any further, 1 if + * it can. + */ +int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, + unsigned long *block) +{ + transaction_t *transaction; + int ret; + + read_lock(&journal->j_state_lock); + spin_lock(&journal->j_list_lock); + transaction = journal->j_checkpoint_transactions; + if (transaction) { + *tid = transaction->t_tid; + *block = transaction->t_log_start; + } else if ((transaction = journal->j_committing_transaction) != NULL) { + *tid = transaction->t_tid; + *block = transaction->t_log_start; + } else if ((transaction = journal->j_running_transaction) != NULL) { + *tid = transaction->t_tid; + *block = journal->j_head; + } else { + *tid = journal->j_transaction_sequence; + *block = journal->j_head; + } + ret = tid_gt(*tid, journal->j_tail_sequence); + spin_unlock(&journal->j_list_lock); + read_unlock(&journal->j_state_lock); + + return ret; +} + +/* + * Update information in journal structure and in on disk journal superblock + * about log tail. This function does not check whether information passed in + * really pushes log tail further. It's responsibility of the caller to make + * sure provided log tail information is valid (e.g. by holding + * j_checkpoint_mutex all the time between computing log tail and calling this + * function as is the case with jbd2_cleanup_journal_tail()). + * + * Requires j_checkpoint_mutex + */ +void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) +{ + unsigned long freed; + + BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); + + /* + * We cannot afford for write to remain in drive's caches since as + * soon as we update j_tail, next transaction can start reusing journal + * space and if we lose sb update during power failure we'd replay + * old transaction with possibly newly overwritten data. + */ + jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); + write_lock(&journal->j_state_lock); + freed = block - journal->j_tail; + if (block < journal->j_tail) + freed += journal->j_last - journal->j_first; + + trace_jbd2_update_log_tail(journal, tid, block, freed); + jbd_debug(1, + "Cleaning journal tail from %d to %d (offset %lu), " + "freeing %lu\n", + journal->j_tail_sequence, tid, block, freed); + + journal->j_free += freed; + journal->j_tail_sequence = tid; + journal->j_tail = block; + write_unlock(&journal->j_state_lock); +} + struct jbd2_stats_proc_session { journal_t *journal; struct transaction_stats_s *stats; @@ -1156,16 +1235,28 @@ static int journal_reset(journal_t *jour journal->j_errno); journal->j_flags |= JBD2_FLUSHED; } else { - /* Add the dynamic fields and write it to disk. */ - jbd2_journal_update_sb_log_tail(journal); + /* + * Update log tail information. We use WRITE_FUA since new + * transaction will start reusing journal space and so we + * must make sure information about current log tail is on + * disk before that. + */ + jbd2_journal_update_sb_log_tail(journal, + journal->j_tail_sequence, + journal->j_tail, + WRITE_FUA); } return jbd2_journal_start_thread(journal); } -static void jbd2_write_superblock(journal_t *journal) +static void jbd2_write_superblock(journal_t *journal, int write_op) { struct buffer_head *bh = journal->j_sb_buffer; + int ret; + if (!(journal->j_flags & JBD2_BARRIER)) + write_op &= ~(REQ_FUA | REQ_FLUSH); + lock_buffer(bh); if (buffer_write_io_error(bh)) { /* * Oh, dear. A previous attempt to write the journal @@ -1181,39 +1272,44 @@ static void jbd2_write_superblock(journa clear_buffer_write_io_error(bh); set_buffer_uptodate(bh); } - - BUFFER_TRACE(bh, "marking dirty"); - mark_buffer_dirty(bh); - sync_dirty_buffer(bh); + get_bh(bh); + bh->b_end_io = end_buffer_write_sync; + ret = submit_bh(write_op, bh); + wait_on_buffer(bh); if (buffer_write_io_error(bh)) { - printk(KERN_ERR "JBD2: I/O error detected " - "when updating journal superblock for %s.\n", - journal->j_devname); clear_buffer_write_io_error(bh); set_buffer_uptodate(bh); + ret = -EIO; + } + if (ret) { + printk(KERN_ERR "JBD2: Error %d detected when updating " + "journal superblock for %s.\n", ret, + journal->j_devname); } } /** * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. * @journal: The journal to update. + * @tail_tid: TID of the new transaction at the tail of the log + * @tail_block: The first block of the transaction at the tail of the log + * @write_op: With which operation should we write the journal sb * * Update a journal's superblock information about log tail and write it to * disk, waiting for the IO to complete. */ -void jbd2_journal_update_sb_log_tail(journal_t *journal) +void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, + unsigned long tail_block, int write_op) { journal_superblock_t *sb = journal->j_superblock; - read_lock(&journal->j_state_lock); - jbd_debug(1, "JBD2: updating superblock (start %ld, seq %d)\n", - journal->j_tail, journal->j_tail_sequence); + jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", + tail_block, tail_tid); - sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); - sb->s_start = cpu_to_be32(journal->j_tail); - read_unlock(&journal->j_state_lock); + sb->s_sequence = cpu_to_be32(tail_tid); + sb->s_start = cpu_to_be32(tail_block); - jbd2_write_superblock(journal); + jbd2_write_superblock(journal, write_op); /* Log is no longer empty */ write_lock(&journal->j_state_lock); WARN_ON(!sb->s_sequence); @@ -1240,7 +1336,7 @@ static void jbd2_mark_journal_empty(jour sb->s_start = cpu_to_be32(0); read_unlock(&journal->j_state_lock); - jbd2_write_superblock(journal); + jbd2_write_superblock(journal, WRITE_FUA); /* Log is no longer empty */ write_lock(&journal->j_state_lock); @@ -1266,7 +1362,7 @@ static void jbd2_journal_update_sb_errno sb->s_errno = cpu_to_be32(journal->j_errno); read_unlock(&journal->j_state_lock); - jbd2_write_superblock(journal); + jbd2_write_superblock(journal, WRITE_SYNC); } /* --- a/fs/jbd2/recovery.c +++ b/fs/jbd2/recovery.c @@ -21,6 +21,7 @@ #include #include #include +#include #endif /* @@ -265,7 +266,9 @@ int jbd2_journal_recover(journal_t *jour err2 = sync_blockdev(journal->j_fs_dev); if (!err) err = err2; - + /* Make sure all replayed data is on permanent storage */ + if (journal->j_flags & JBD2_BARRIER) + blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); return err; } --- a/include/linux/jbd2.h +++ b/include/linux/jbd2.h @@ -972,6 +972,9 @@ extern void __journal_clean_data_list(tr /* Log buffer allocation */ extern struct journal_head * jbd2_journal_get_descriptor_buffer(journal_t *); int jbd2_journal_next_log_block(journal_t *, unsigned long long *); +int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, + unsigned long *block); +void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block); /* Commit management */ extern void jbd2_journal_commit_transaction(journal_t *); @@ -1083,7 +1086,8 @@ extern int jbd2_journal_destroy (j extern int jbd2_journal_recover (journal_t *journal); extern int jbd2_journal_wipe (journal_t *, int); extern int jbd2_journal_skip_recovery (journal_t *); -extern void jbd2_journal_update_sb_log_tail (journal_t *); +extern void jbd2_journal_update_sb_log_tail (journal_t *, tid_t, + unsigned long, int); extern void __jbd2_journal_abort_hard (journal_t *); extern void jbd2_journal_abort (journal_t *, int); extern int jbd2_journal_errno (journal_t *); --- a/include/trace/events/jbd2.h +++ b/include/trace/events/jbd2.h @@ -200,7 +200,7 @@ TRACE_EVENT(jbd2_checkpoint_stats, __entry->forced_to_close, __entry->written, __entry->dropped) ); -TRACE_EVENT(jbd2_cleanup_journal_tail, +TRACE_EVENT(jbd2_update_log_tail, TP_PROTO(journal_t *journal, tid_t first_tid, unsigned long block_nr, unsigned long freed), -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/