Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
[linux-2.6.git] / fs / jbd2 / commit.c
index da8d0eb..7f21cf3 100644 (file)
 #include <linux/pagemap.h>
 #include <linux/jiffies.h>
 #include <linux/crc32.h>
+#include <linux/writeback.h>
+#include <linux/backing-dev.h>
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/bitops.h>
+#include <trace/events/jbd2.h>
+#include <asm/system.h>
 
 /*
  * Default IO end handler for temporary BJ_IO buffer_heads.
@@ -37,8 +44,8 @@ static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 }
 
 /*
- * When an ext3-ordered file is truncated, it is possible that many pages are
- * not sucessfully freed, because they are attached to a committing transaction.
+ * When an ext4 file is truncated, it is possible that some pages are not
+ * successfully freed, because they are attached to a committing transaction.
  * After the transaction commits, these pages are left on the LRU, with no
  * ->mapping, and with attached buffers.  These pages are trivially reclaimable
  * by the VM, but their apparent absence upsets the VM accounting, and it makes
@@ -65,7 +72,7 @@ static void release_buffer_page(struct buffer_head *bh)
                goto nope;
 
        /* OK, it's a truncated page */
-       if (TestSetPageLocked(page))
+       if (!trylock_page(page))
                goto nope;
 
        page_cache_get(page);
@@ -80,21 +87,6 @@ nope:
 }
 
 /*
- * Try to acquire jbd_lock_bh_state() against the buffer, when j_list_lock is
- * held.  For ranking reasons we must trylock.  If we lose, schedule away and
- * return 0.  j_list_lock is dropped in this case.
- */
-static int inverted_lock(journal_t *journal, struct buffer_head *bh)
-{
-       if (!jbd_trylock_bh_state(bh)) {
-               spin_unlock(&journal->j_list_lock);
-               schedule();
-               return 0;
-       }
-       return 1;
-}
-
-/*
  * Done it all: now submit the commit record.  We should have
  * cleaned up our previous buffers by now, so if we are in abort
  * mode we can now just skip the rest of the journal write
@@ -111,7 +103,9 @@ static int journal_submit_commit_record(journal_t *journal,
        struct commit_header *tmp;
        struct buffer_head *bh;
        int ret;
-       int barrier_done = 0;
+       struct timespec now = current_kernel_time();
+
+       *cbh = NULL;
 
        if (is_journal_aborted(journal))
                return 0;
@@ -126,6 +120,8 @@ static int journal_submit_commit_record(journal_t *journal,
        tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
        tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
        tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
+       tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
+       tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
 
        if (JBD2_HAS_COMPAT_FEATURE(journal,
                                    JBD2_FEATURE_COMPAT_CHECKSUM)) {
@@ -136,41 +132,17 @@ static int journal_submit_commit_record(journal_t *journal,
 
        JBUFFER_TRACE(descriptor, "submit commit block");
        lock_buffer(bh);
-
-       set_buffer_dirty(bh);
+       clear_buffer_dirty(bh);
        set_buffer_uptodate(bh);
        bh->b_end_io = journal_end_buffer_io_sync;
 
        if (journal->j_flags & JBD2_BARRIER &&
-               !JBD2_HAS_COMPAT_FEATURE(journal,
-                                        JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
-               set_buffer_ordered(bh);
-               barrier_done = 1;
-       }
-       ret = submit_bh(WRITE, bh);
-
-       /* is it possible for another commit to fail at roughly
-        * the same time as this one?  If so, we don't want to
-        * trust the barrier flag in the super, but instead want
-        * to remember if we sent a barrier request
-        */
-       if (ret == -EOPNOTSUPP && barrier_done) {
-               char b[BDEVNAME_SIZE];
+           !JBD2_HAS_INCOMPAT_FEATURE(journal,
+                                      JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
+               ret = submit_bh(WRITE_SYNC | WRITE_FLUSH_FUA, bh);
+       else
+               ret = submit_bh(WRITE_SYNC, bh);
 
-               printk(KERN_WARNING
-                       "JBD: barrier-based sync failed on %s - "
-                       "disabling barriers\n",
-                       bdevname(journal->j_dev, b));
-               spin_lock(&journal->j_state_lock);
-               journal->j_flags &= ~JBD2_BARRIER;
-               spin_unlock(&journal->j_state_lock);
-
-               /* And try again, without the barrier */
-               clear_buffer_ordered(bh);
-               set_buffer_uptodate(bh);
-               set_buffer_dirty(bh);
-               ret = submit_bh(WRITE, bh);
-       }
        *cbh = bh;
        return ret;
 }
@@ -179,7 +151,8 @@ static int journal_submit_commit_record(journal_t *journal,
  * This function along with journal_submit_commit_record
  * allows to write the commit record asynchronously.
  */
-static int journal_wait_on_commit_record(struct buffer_head *bh)
+static int journal_wait_on_commit_record(journal_t *journal,
+                                        struct buffer_head *bh)
 {
        int ret = 0;
 
@@ -195,159 +168,116 @@ static int journal_wait_on_commit_record(struct buffer_head *bh)
 }
 
 /*
- * Wait for all submitted IO to complete.
+ * write the filemap data using writepage() address_space_operations.
+ * We don't do block allocation here even for delalloc. We don't
+ * use writepages() because with dealyed allocation we may be doing
+ * block allocation in writepages().
  */
-static int journal_wait_on_locked_list(journal_t *journal,
-                                      transaction_t *commit_transaction)
+static int journal_submit_inode_data_buffers(struct address_space *mapping)
 {
-       int ret = 0;
-       struct journal_head *jh;
-
-       while (commit_transaction->t_locked_list) {
-               struct buffer_head *bh;
-
-               jh = commit_transaction->t_locked_list->b_tprev;
-               bh = jh2bh(jh);
-               get_bh(bh);
-               if (buffer_locked(bh)) {
-                       spin_unlock(&journal->j_list_lock);
-                       wait_on_buffer(bh);
-                       if (unlikely(!buffer_uptodate(bh)))
-                               ret = -EIO;
-                       spin_lock(&journal->j_list_lock);
-               }
-               if (!inverted_lock(journal, bh)) {
-                       put_bh(bh);
-                       spin_lock(&journal->j_list_lock);
-                       continue;
-               }
-               if (buffer_jbd(bh) && jh->b_jlist == BJ_Locked) {
-                       __jbd2_journal_unfile_buffer(jh);
-                       jbd_unlock_bh_state(bh);
-                       jbd2_journal_remove_journal_head(bh);
-                       put_bh(bh);
-               } else {
-                       jbd_unlock_bh_state(bh);
-               }
-               put_bh(bh);
-               cond_resched_lock(&journal->j_list_lock);
-       }
+       int ret;
+       struct writeback_control wbc = {
+               .sync_mode =  WB_SYNC_ALL,
+               .nr_to_write = mapping->nrpages * 2,
+               .range_start = 0,
+               .range_end = i_size_read(mapping->host),
+       };
+
+       ret = generic_writepages(mapping, &wbc);
        return ret;
-  }
+}
 
-static void journal_do_submit_data(struct buffer_head **wbuf, int bufs)
+/*
+ * Submit all the data buffers of inode associated with the transaction to
+ * disk.
+ *
+ * We are in a committing transaction. Therefore no new inode can be added to
+ * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
+ * operate on from being released while we write out pages.
+ */
+static int journal_submit_data_buffers(journal_t *journal,
+               transaction_t *commit_transaction)
 {
-       int i;
+       struct jbd2_inode *jinode;
+       int err, ret = 0;
+       struct address_space *mapping;
 
-       for (i = 0; i < bufs; i++) {
-               wbuf[i]->b_end_io = end_buffer_write_sync;
-               /* We use-up our safety reference in submit_bh() */
-               submit_bh(WRITE, wbuf[i]);
+       spin_lock(&journal->j_list_lock);
+       list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
+               mapping = jinode->i_vfs_inode->i_mapping;
+               set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
+               spin_unlock(&journal->j_list_lock);
+               /*
+                * submit the inode data buffers. We use writepage
+                * instead of writepages. Because writepages can do
+                * block allocation  with delalloc. We need to write
+                * only allocated blocks here.
+                */
+               trace_jbd2_submit_inode_data(jinode->i_vfs_inode);
+               err = journal_submit_inode_data_buffers(mapping);
+               if (!ret)
+                       ret = err;
+               spin_lock(&journal->j_list_lock);
+               J_ASSERT(jinode->i_transaction == commit_transaction);
+               clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
+               smp_mb__after_clear_bit();
+               wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
        }
+       spin_unlock(&journal->j_list_lock);
+       return ret;
 }
 
 /*
- *  Submit all the data buffers to disk
+ * Wait for data submitted for writeout, refile inodes to proper
+ * transaction if needed.
+ *
  */
-static void journal_submit_data_buffers(journal_t *journal,
-                               transaction_t *commit_transaction)
+static int journal_finish_inode_data_buffers(journal_t *journal,
+               transaction_t *commit_transaction)
 {
-       struct journal_head *jh;
-       struct buffer_head *bh;
-       int locked;
-       int bufs = 0;
-       struct buffer_head **wbuf = journal->j_wbuf;
+       struct jbd2_inode *jinode, *next_i;
+       int err, ret = 0;
 
-       /*
-        * Whenever we unlock the journal and sleep, things can get added
-        * onto ->t_sync_datalist, so we have to keep looping back to
-        * write_out_data until we *know* that the list is empty.
-        *
-        * Cleanup any flushed data buffers from the data list.  Even in
-        * abort mode, we want to flush this out as soon as possible.
-        */
-write_out_data:
-       cond_resched();
+       /* For locking, see the comment in journal_submit_data_buffers() */
        spin_lock(&journal->j_list_lock);
-
-       while (commit_transaction->t_sync_datalist) {
-               jh = commit_transaction->t_sync_datalist;
-               bh = jh2bh(jh);
-               locked = 0;
-
-               /* Get reference just to make sure buffer does not disappear
-                * when we are forced to drop various locks */
-               get_bh(bh);
-               /* If the buffer is dirty, we need to submit IO and hence
-                * we need the buffer lock. We try to lock the buffer without
-                * blocking. If we fail, we need to drop j_list_lock and do
-                * blocking lock_buffer().
-                */
-               if (buffer_dirty(bh)) {
-                       if (test_set_buffer_locked(bh)) {
-                               BUFFER_TRACE(bh, "needs blocking lock");
-                               spin_unlock(&journal->j_list_lock);
-                               /* Write out all data to prevent deadlocks */
-                               journal_do_submit_data(wbuf, bufs);
-                               bufs = 0;
-                               lock_buffer(bh);
-                               spin_lock(&journal->j_list_lock);
-                       }
-                       locked = 1;
-               }
-               /* We have to get bh_state lock. Again out of order, sigh. */
-               if (!inverted_lock(journal, bh)) {
-                       jbd_lock_bh_state(bh);
-                       spin_lock(&journal->j_list_lock);
-               }
-               /* Someone already cleaned up the buffer? */
-               if (!buffer_jbd(bh)
-                       || jh->b_transaction != commit_transaction
-                       || jh->b_jlist != BJ_SyncData) {
-                       jbd_unlock_bh_state(bh);
-                       if (locked)
-                               unlock_buffer(bh);
-                       BUFFER_TRACE(bh, "already cleaned up");
-                       put_bh(bh);
-                       continue;
-               }
-               if (locked && test_clear_buffer_dirty(bh)) {
-                       BUFFER_TRACE(bh, "needs writeout, adding to array");
-                       wbuf[bufs++] = bh;
-                       __jbd2_journal_file_buffer(jh, commit_transaction,
-                                               BJ_Locked);
-                       jbd_unlock_bh_state(bh);
-                       if (bufs == journal->j_wbufsize) {
-                               spin_unlock(&journal->j_list_lock);
-                               journal_do_submit_data(wbuf, bufs);
-                               bufs = 0;
-                               goto write_out_data;
-                       }
-               } else if (!locked && buffer_locked(bh)) {
-                       __jbd2_journal_file_buffer(jh, commit_transaction,
-                                               BJ_Locked);
-                       jbd_unlock_bh_state(bh);
-                       put_bh(bh);
-               } else {
-                       BUFFER_TRACE(bh, "writeout complete: unfile");
-                       __jbd2_journal_unfile_buffer(jh);
-                       jbd_unlock_bh_state(bh);
-                       if (locked)
-                               unlock_buffer(bh);
-                       jbd2_journal_remove_journal_head(bh);
-                       /* Once for our safety reference, once for
-                        * jbd2_journal_remove_journal_head() */
-                       put_bh(bh);
-                       put_bh(bh);
+       list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
+               set_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
+               spin_unlock(&journal->j_list_lock);
+               err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
+               if (err) {
+                       /*
+                        * Because AS_EIO is cleared by
+                        * filemap_fdatawait_range(), set it again so
+                        * that user process can get -EIO from fsync().
+                        */
+                       set_bit(AS_EIO,
+                               &jinode->i_vfs_inode->i_mapping->flags);
+
+                       if (!ret)
+                               ret = err;
                }
+               spin_lock(&journal->j_list_lock);
+               clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
+               smp_mb__after_clear_bit();
+               wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
+       }
 
-               if (lock_need_resched(&journal->j_list_lock)) {
-                       spin_unlock(&journal->j_list_lock);
-                       goto write_out_data;
+       /* Now refile inode to proper lists */
+       list_for_each_entry_safe(jinode, next_i,
+                                &commit_transaction->t_inode_list, i_list) {
+               list_del(&jinode->i_list);
+               if (jinode->i_next_transaction) {
+                       jinode->i_transaction = jinode->i_next_transaction;
+                       jinode->i_next_transaction = NULL;
+                       list_add(&jinode->i_list,
+                               &jinode->i_transaction->t_inode_list);
+               } else {
+                       jinode->i_transaction = NULL;
                }
        }
        spin_unlock(&journal->j_list_lock);
-       journal_do_submit_data(wbuf, bufs);
+
+       return ret;
 }
 
 static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
@@ -388,28 +318,25 @@ void jbd2_journal_commit_transaction(journal_t *journal)
        int flags;
        int err;
        unsigned long long blocknr;
+       ktime_t start_time;
+       u64 commit_time;
        char *tagp = NULL;
        journal_header_t *header;
        journal_block_tag_t *tag = NULL;
        int space_left = 0;
        int first_tag = 0;
        int tag_flag;
-       int i;
+       int i, to_free = 0;
        int tag_bytes = journal_tag_bytes(journal);
        struct buffer_head *cbh = NULL; /* For transactional checksums */
        __u32 crc32_sum = ~0;
+       struct blk_plug plug;
 
        /*
         * First job: lock down the current transaction and wait for
         * all outstanding updates to complete.
         */
 
-#ifdef COMMIT_STATS
-       spin_lock(&journal->j_list_lock);
-       summarise_journal_usage(journal);
-       spin_unlock(&journal->j_list_lock);
-#endif
-
        /* 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");
@@ -424,35 +351,37 @@ void jbd2_journal_commit_transaction(journal_t *journal)
        commit_transaction = journal->j_running_transaction;
        J_ASSERT(commit_transaction->t_state == T_RUNNING);
 
+       trace_jbd2_start_commit(journal, commit_transaction);
        jbd_debug(1, "JBD: starting commit of transaction %d\n",
                        commit_transaction->t_tid);
 
-       spin_lock(&journal->j_state_lock);
+       write_lock(&journal->j_state_lock);
        commit_transaction->t_state = T_LOCKED;
 
-       stats.u.run.rs_wait = commit_transaction->t_max_wait;
-       stats.u.run.rs_locked = jiffies;
-       stats.u.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
-                                               stats.u.run.rs_locked);
+       trace_jbd2_commit_locking(journal, commit_transaction);
+       stats.run.rs_wait = commit_transaction->t_max_wait;
+       stats.run.rs_locked = jiffies;
+       stats.run.rs_running = jbd2_time_diff(commit_transaction->t_start,
+                                             stats.run.rs_locked);
 
        spin_lock(&commit_transaction->t_handle_lock);
-       while (commit_transaction->t_updates) {
+       while (atomic_read(&commit_transaction->t_updates)) {
                DEFINE_WAIT(wait);
 
                prepare_to_wait(&journal->j_wait_updates, &wait,
                                        TASK_UNINTERRUPTIBLE);
-               if (commit_transaction->t_updates) {
+               if (atomic_read(&commit_transaction->t_updates)) {
                        spin_unlock(&commit_transaction->t_handle_lock);
-                       spin_unlock(&journal->j_state_lock);
+                       write_unlock(&journal->j_state_lock);
                        schedule();
-                       spin_lock(&journal->j_state_lock);
+                       write_lock(&journal->j_state_lock);
                        spin_lock(&commit_transaction->t_handle_lock);
                }
                finish_wait(&journal->j_wait_updates, &wait);
        }
        spin_unlock(&commit_transaction->t_handle_lock);
 
-       J_ASSERT (commit_transaction->t_outstanding_credits <=
+       J_ASSERT (atomic_read(&commit_transaction->t_outstanding_credits) <=
                        journal->j_max_transaction_buffers);
 
        /*
@@ -469,7 +398,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
         * we do not require it to remember exactly which old buffers it
         * has reserved.  This is consistent with the existing behaviour
         * that multiple jbd2_journal_get_write_access() calls to the same
-        * buffer are perfectly permissable.
+        * buffer are perfectly permissible.
         */
        while (commit_transaction->t_reserved_list) {
                jh = commit_transaction->t_reserved_list;
@@ -505,86 +434,60 @@ void jbd2_journal_commit_transaction(journal_t *journal)
         */
        jbd2_journal_switch_revoke_table(journal);
 
-       stats.u.run.rs_flushing = jiffies;
-       stats.u.run.rs_locked = jbd2_time_diff(stats.u.run.rs_locked,
-                                              stats.u.run.rs_flushing);
+       trace_jbd2_commit_flushing(journal, commit_transaction);
+       stats.run.rs_flushing = jiffies;
+       stats.run.rs_locked = jbd2_time_diff(stats.run.rs_locked,
+                                            stats.run.rs_flushing);
 
        commit_transaction->t_state = T_FLUSH;
        journal->j_committing_transaction = commit_transaction;
        journal->j_running_transaction = NULL;
+       start_time = ktime_get();
        commit_transaction->t_log_start = journal->j_head;
        wake_up(&journal->j_wait_transaction_locked);
-       spin_unlock(&journal->j_state_lock);
+       write_unlock(&journal->j_state_lock);
 
        jbd_debug (3, "JBD: commit phase 2\n");
 
        /*
-        * First, drop modified flag: all accesses to the buffers
-        * will be tracked for a new trasaction only -bzzz
-        */
-       spin_lock(&journal->j_list_lock);
-       if (commit_transaction->t_buffers) {
-               new_jh = jh = commit_transaction->t_buffers->b_tnext;
-               do {
-                       J_ASSERT_JH(new_jh, new_jh->b_modified == 1 ||
-                                       new_jh->b_modified == 0);
-                       new_jh->b_modified = 0;
-                       new_jh = new_jh->b_tnext;
-               } while (new_jh != jh);
-       }
-       spin_unlock(&journal->j_list_lock);
-
-       /*
         * Now start flushing things to disk, in the order they appear
         * on the transaction lists.  Data blocks go first.
         */
-       err = 0;
-       journal_submit_data_buffers(journal, commit_transaction);
-
-       /*
-        * Wait for all previously submitted IO to complete if commit
-        * record is to be written synchronously.
-        */
-       spin_lock(&journal->j_list_lock);
-       if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
-               JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
-               err = journal_wait_on_locked_list(journal,
-                                               commit_transaction);
-
-       spin_unlock(&journal->j_list_lock);
-
+       err = journal_submit_data_buffers(journal, commit_transaction);
        if (err)
                jbd2_journal_abort(journal, err);
 
-       jbd2_journal_write_revoke_records(journal, commit_transaction);
+       blk_start_plug(&plug);
+       jbd2_journal_write_revoke_records(journal, commit_transaction,
+                                         WRITE_SYNC);
+       blk_finish_plug(&plug);
 
        jbd_debug(3, "JBD: commit phase 2\n");
 
        /*
-        * If we found any dirty or locked buffers, then we should have
-        * looped back up to the write_out_data label.  If there weren't
-        * any then journal_clean_data_list should have wiped the list
-        * clean by now, so check that it is in fact empty.
-        */
-       J_ASSERT (commit_transaction->t_sync_datalist == NULL);
-
-       jbd_debug (3, "JBD: commit phase 3\n");
-
-       /*
         * Way to go: we have now written out all of the data for a
         * transaction!  Now comes the tricky part: we need to write out
         * metadata.  Loop over the transaction's entire buffer list:
         */
+       write_lock(&journal->j_state_lock);
        commit_transaction->t_state = T_COMMIT;
+       write_unlock(&journal->j_state_lock);
+
+       trace_jbd2_commit_logging(journal, commit_transaction);
+       stats.run.rs_logging = jiffies;
+       stats.run.rs_flushing = jbd2_time_diff(stats.run.rs_flushing,
+                                              stats.run.rs_logging);
+       stats.run.rs_blocks =
+               atomic_read(&commit_transaction->t_outstanding_credits);
+       stats.run.rs_blocks_logged = 0;
 
-       stats.u.run.rs_logging = jiffies;
-       stats.u.run.rs_flushing = jbd2_time_diff(stats.u.run.rs_flushing,
-                                                stats.u.run.rs_logging);
-       stats.u.run.rs_blocks = commit_transaction->t_outstanding_credits;
-       stats.u.run.rs_blocks_logged = 0;
+       J_ASSERT(commit_transaction->t_nr_buffers <=
+                atomic_read(&commit_transaction->t_outstanding_credits));
 
+       err = 0;
        descriptor = NULL;
        bufs = 0;
+       blk_start_plug(&plug);
        while (commit_transaction->t_buffers) {
 
                /* Find the next buffer to be journaled... */
@@ -592,10 +495,15 @@ void jbd2_journal_commit_transaction(journal_t *journal)
                jh = commit_transaction->t_buffers;
 
                /* If we're in abort mode, we just un-journal the buffer and
-                  release it for background writing. */
+                  release it. */
 
                if (is_journal_aborted(journal)) {
+                       clear_buffer_jbddirty(jh2bh(jh));
                        JBUFFER_TRACE(jh, "journal is aborting: refile");
+                       jbd2_buffer_abort_trigger(jh,
+                                                 jh->b_frozen_data ?
+                                                 jh->b_frozen_triggers :
+                                                 jh->b_triggers);
                        jbd2_journal_refile_buffer(journal, jh);
                        /* If that was the last one, we need to clean up
                         * any descriptor buffers which may have been
@@ -660,7 +568,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
                 * the free space in the log, but this counter is changed
                 * by jbd2_journal_next_log_block() also.
                 */
-               commit_transaction->t_outstanding_credits--;
+               atomic_dec(&commit_transaction->t_outstanding_credits);
 
                /* Bump b_count to prevent truncate from stumbling over
                    the shadowed buffer!  @@@ This can go if we ever get
@@ -681,6 +589,10 @@ void jbd2_journal_commit_transaction(journal_t *journal)
                JBUFFER_TRACE(jh, "ph3: write metadata");
                flags = jbd2_journal_write_metadata_buffer(commit_transaction,
                                                      jh, &new_jh, blocknr);
+               if (flags < 0) {
+                       jbd2_journal_abort(journal, flags);
+                       continue;
+               }
                set_bit(BH_JWrite, &jh2bh(new_jh)->b_state);
                wbuf[bufs++] = jh2bh(new_jh);
 
@@ -737,10 +649,10 @@ start_journal_io:
                                clear_buffer_dirty(bh);
                                set_buffer_uptodate(bh);
                                bh->b_end_io = journal_end_buffer_io_sync;
-                               submit_bh(WRITE, bh);
+                               submit_bh(WRITE_SYNC, bh);
                        }
                        cond_resched();
-                       stats.u.run.rs_blocks_logged += bufs;
+                       stats.run.rs_blocks_logged += bufs;
 
                        /* Force a new descriptor to be generated next
                            time round the loop. */
@@ -749,23 +661,41 @@ start_journal_io:
                }
        }
 
-       /* Done it all: now write the commit record asynchronously. */
+       err = journal_finish_inode_data_buffers(journal, commit_transaction);
+       if (err) {
+               printk(KERN_WARNING
+                       "JBD2: Detected IO errors while flushing file data "
+                      "on %s\n", journal->j_devname);
+               if (journal->j_flags & JBD2_ABORT_ON_SYNCDATA_ERR)
+                       jbd2_journal_abort(journal, err);
+               err = 0;
+       }
 
+       write_lock(&journal->j_state_lock);
+       J_ASSERT(commit_transaction->t_state == T_COMMIT);
+       commit_transaction->t_state = T_COMMIT_DFLUSH;
+       write_unlock(&journal->j_state_lock);
+       /* 
+        * If the journal is not located on the file system device,
+        * then we must flush the file system device before we issue
+        * the commit record
+        */
+       if (commit_transaction->t_need_data_flush &&
+           (journal->j_fs_dev != journal->j_dev) &&
+           (journal->j_flags & JBD2_BARRIER))
+               blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
+
+       /* Done it all: now write the commit record asynchronously. */
        if (JBD2_HAS_INCOMPAT_FEATURE(journal,
-               JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
+                                     JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
                err = journal_submit_commit_record(journal, commit_transaction,
                                                 &cbh, crc32_sum);
                if (err)
                        __jbd2_journal_abort_hard(journal);
-
-               spin_lock(&journal->j_list_lock);
-               err = journal_wait_on_locked_list(journal,
-                                               commit_transaction);
-               spin_unlock(&journal->j_list_lock);
-               if (err)
-                       __jbd2_journal_abort_hard(journal);
        }
 
+       blk_finish_plug(&plug);
+
        /* Lo and behold: we have just managed to send a transaction to
            the log.  Before we can commit it, wait for the IO so far to
            complete.  Control buffers being written are on the
@@ -777,7 +707,7 @@ start_journal_io:
           so we incur less scheduling load.
        */
 
-       jbd_debug(3, "JBD: commit phase 4\n");
+       jbd_debug(3, "JBD: commit phase 3\n");
 
        /*
         * akpm: these are BJ_IO, and j_list_lock is not needed.
@@ -827,8 +757,13 @@ wait_for_iobuf:
                    required. */
                JBUFFER_TRACE(jh, "file as BJ_Forget");
                jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
-               /* Wake up any transactions which were waiting for this
-                  IO to complete */
+               /*
+                * Wake up any transactions which were waiting for this IO to
+                * complete. The barrier must be here so that changes by
+                * jbd2_journal_file_buffer() take effect before wake_up_bit()
+                * does the waitqueue check.
+                */
+               smp_mb();
                wake_up_bit(&bh->b_state, BH_Unshadow);
                JBUFFER_TRACE(jh, "brelse shadowed buffer");
                __brelse(bh);
@@ -836,7 +771,7 @@ wait_for_iobuf:
 
        J_ASSERT (commit_transaction->t_shadow_list == NULL);
 
-       jbd_debug(3, "JBD: commit phase 5\n");
+       jbd_debug(3, "JBD: commit phase 4\n");
 
        /* Here we wait for the revoke record and descriptor record buffers */
  wait_for_ctlbuf:
@@ -863,16 +798,29 @@ wait_for_iobuf:
                /* AKPM: bforget here */
        }
 
-       jbd_debug(3, "JBD: commit phase 6\n");
+       if (err)
+               jbd2_journal_abort(journal, err);
+
+       jbd_debug(3, "JBD: commit phase 5\n");
+       write_lock(&journal->j_state_lock);
+       J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
+       commit_transaction->t_state = T_COMMIT_JFLUSH;
+       write_unlock(&journal->j_state_lock);
 
        if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
-               JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
+                                      JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
                err = journal_submit_commit_record(journal, commit_transaction,
                                                &cbh, crc32_sum);
                if (err)
                        __jbd2_journal_abort_hard(journal);
        }
-       err = journal_wait_on_commit_record(cbh);
+       if (cbh)
+               err = journal_wait_on_commit_record(journal, cbh);
+       if (JBD2_HAS_INCOMPAT_FEATURE(journal,
+                                     JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) &&
+           journal->j_flags & JBD2_BARRIER) {
+               blkdev_issue_flush(journal->j_dev, GFP_KERNEL, NULL);
+       }
 
        if (err)
                jbd2_journal_abort(journal, err);
@@ -882,9 +830,9 @@ wait_for_iobuf:
            transaction can be removed from any checkpoint list it was on
            before. */
 
-       jbd_debug(3, "JBD: commit phase 7\n");
+       jbd_debug(3, "JBD: commit phase 6\n");
 
-       J_ASSERT(commit_transaction->t_sync_datalist == NULL);
+       J_ASSERT(list_empty(&commit_transaction->t_inode_list));
        J_ASSERT(commit_transaction->t_buffers == NULL);
        J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
        J_ASSERT(commit_transaction->t_iobuf_list == NULL);
@@ -905,8 +853,7 @@ restart_loop:
                spin_unlock(&journal->j_list_lock);
                bh = jh2bh(jh);
                jbd_lock_bh_state(bh);
-               J_ASSERT_JH(jh, jh->b_transaction == commit_transaction ||
-                       jh->b_transaction == journal->j_running_transaction);
+               J_ASSERT_JH(jh, jh->b_transaction == commit_transaction);
 
                /*
                 * If there is undo-protected committed data against
@@ -917,6 +864,9 @@ restart_loop:
                 * data.
                 *
                 * Otherwise, we can just throw away the frozen data now.
+                *
+                * We also know that the frozen data has already fired
+                * its triggers if they exist, so we can clear that too.
                 */
                if (jh->b_committed_data) {
                        jbd2_free(jh->b_committed_data, bh->b_size);
@@ -924,10 +874,12 @@ restart_loop:
                        if (jh->b_frozen_data) {
                                jh->b_committed_data = jh->b_frozen_data;
                                jh->b_frozen_data = NULL;
+                               jh->b_frozen_triggers = NULL;
                        }
                } else if (jh->b_frozen_data) {
                        jbd2_free(jh->b_frozen_data, bh->b_size);
                        jh->b_frozen_data = NULL;
+                       jh->b_frozen_triggers = NULL;
                }
 
                spin_lock(&journal->j_list_lock);
@@ -947,12 +899,12 @@ restart_loop:
                /* A buffer which has been freed while still being
                 * journaled by a previous transaction may end up still
                 * being dirty here, but we want to avoid writing back
-                * that buffer in the future now that the last use has
-                * been committed.  That's not only a performance gain,
-                * it also stops aliasing problems if the buffer is left
-                * behind for writeback and gets reallocated for another
+                * that buffer in the future after the "add to orphan"
+                * operation been committed,  That's not only a performance
+                * gain, it also stops aliasing problems if the buffer is
+                * left behind for writeback and gets reallocated for another
                 * use in a different page. */
-               if (buffer_freed(bh)) {
+               if (buffer_freed(bh) && !jh->b_next_transaction) {
                        clear_buffer_freed(bh);
                        clear_buffer_jbddirty(bh);
                }
@@ -960,6 +912,8 @@ restart_loop:
                if (buffer_jbddirty(bh)) {
                        JBUFFER_TRACE(jh, "add to new checkpointing trans");
                        __jbd2_journal_insert_checkpoint(jh, commit_transaction);
+                       if (is_journal_aborted(journal))
+                               clear_buffer_jbddirty(bh);
                        JBUFFER_TRACE(jh, "refile for checkpoint writeback");
                        __jbd2_journal_refile_buffer(jh);
                        jbd_unlock_bh_state(bh);
@@ -991,7 +945,7 @@ restart_loop:
         * __jbd2_journal_drop_transaction(). Otherwise we could race with
         * other checkpointing code processing the transaction...
         */
-       spin_lock(&journal->j_state_lock);
+       write_lock(&journal->j_state_lock);
        spin_lock(&journal->j_list_lock);
        /*
         * Now recheck if some buffers did not get attached to the transaction
@@ -999,55 +953,65 @@ restart_loop:
         */
        if (commit_transaction->t_forget) {
                spin_unlock(&journal->j_list_lock);
-               spin_unlock(&journal->j_state_lock);
+               write_unlock(&journal->j_state_lock);
                goto restart_loop;
        }
 
        /* Done with this transaction! */
 
-       jbd_debug(3, "JBD: commit phase 8\n");
+       jbd_debug(3, "JBD: commit phase 7\n");
 
-       J_ASSERT(commit_transaction->t_state == T_COMMIT);
+       J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
 
        commit_transaction->t_start = jiffies;
-       stats.u.run.rs_logging = jbd2_time_diff(stats.u.run.rs_logging,
-                                               commit_transaction->t_start);
+       stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
+                                             commit_transaction->t_start);
 
        /*
-        * File the transaction for history
+        * File the transaction statistics
         */
-       stats.ts_type = JBD2_STATS_RUN;
        stats.ts_tid = commit_transaction->t_tid;
-       stats.u.run.rs_handle_count = commit_transaction->t_handle_count;
-       spin_lock(&journal->j_history_lock);
-       memcpy(journal->j_history + journal->j_history_cur, &stats,
-                       sizeof(stats));
-       if (++journal->j_history_cur == journal->j_history_max)
-               journal->j_history_cur = 0;
+       stats.run.rs_handle_count =
+               atomic_read(&commit_transaction->t_handle_count);
+       trace_jbd2_run_stats(journal->j_fs_dev->bd_dev,
+                            commit_transaction->t_tid, &stats.run);
 
        /*
         * Calculate overall stats
         */
+       spin_lock(&journal->j_history_lock);
        journal->j_stats.ts_tid++;
-       journal->j_stats.u.run.rs_wait += stats.u.run.rs_wait;
-       journal->j_stats.u.run.rs_running += stats.u.run.rs_running;
-       journal->j_stats.u.run.rs_locked += stats.u.run.rs_locked;
-       journal->j_stats.u.run.rs_flushing += stats.u.run.rs_flushing;
-       journal->j_stats.u.run.rs_logging += stats.u.run.rs_logging;
-       journal->j_stats.u.run.rs_handle_count += stats.u.run.rs_handle_count;
-       journal->j_stats.u.run.rs_blocks += stats.u.run.rs_blocks;
-       journal->j_stats.u.run.rs_blocks_logged += stats.u.run.rs_blocks_logged;
+       journal->j_stats.run.rs_wait += stats.run.rs_wait;
+       journal->j_stats.run.rs_running += stats.run.rs_running;
+       journal->j_stats.run.rs_locked += stats.run.rs_locked;
+       journal->j_stats.run.rs_flushing += stats.run.rs_flushing;
+       journal->j_stats.run.rs_logging += stats.run.rs_logging;
+       journal->j_stats.run.rs_handle_count += stats.run.rs_handle_count;
+       journal->j_stats.run.rs_blocks += stats.run.rs_blocks;
+       journal->j_stats.run.rs_blocks_logged += stats.run.rs_blocks_logged;
        spin_unlock(&journal->j_history_lock);
 
        commit_transaction->t_state = T_FINISHED;
        J_ASSERT(commit_transaction == journal->j_committing_transaction);
        journal->j_commit_sequence = commit_transaction->t_tid;
        journal->j_committing_transaction = NULL;
-       spin_unlock(&journal->j_state_lock);
+       commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
+
+       /*
+        * weight the commit time higher than the average time so we don't
+        * react too strongly to vast changes in the commit time
+        */
+       if (likely(journal->j_average_commit_time))
+               journal->j_average_commit_time = (commit_time +
+                               journal->j_average_commit_time*3) / 4;
+       else
+               journal->j_average_commit_time = commit_time;
+       write_unlock(&journal->j_state_lock);
 
        if (commit_transaction->t_checkpoint_list == NULL &&
            commit_transaction->t_checkpoint_io_list == NULL) {
                __jbd2_journal_drop_transaction(journal, commit_transaction);
+               to_free = 1;
        } else {
                if (journal->j_checkpoint_transactions == NULL) {
                        journal->j_checkpoint_transactions = commit_transaction;
@@ -1066,8 +1030,14 @@ restart_loop:
        }
        spin_unlock(&journal->j_list_lock);
 
+       if (journal->j_commit_callback)
+               journal->j_commit_callback(journal, commit_transaction);
+
+       trace_jbd2_end_commit(journal, commit_transaction);
        jbd_debug(1, "JBD: commit %d complete, head %d\n",
                  journal->j_commit_sequence, journal->j_tail_sequence);
+       if (to_free)
+               kfree(commit_transaction);
 
        wake_up(&journal->j_wait_done_commit);
 }