summaryrefslogtreecommitdiff
path: root/fs/jbd/commit.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/jbd/commit.c')
-rw-r--r--fs/jbd/commit.c228
1 files changed, 139 insertions, 89 deletions
diff --git a/fs/jbd/commit.c b/fs/jbd/commit.c
index ecb44c94ba8d..bb217dcb41af 100644
--- a/fs/jbd/commit.c
+++ b/fs/jbd/commit.c
@@ -20,6 +20,8 @@
#include <linux/mm.h>
#include <linux/pagemap.h>
#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <trace/events/jbd.h>
/*
* Default IO end handler for temporary BJ_IO buffer_heads.
@@ -84,7 +86,12 @@ nope:
static void release_data_buffer(struct buffer_head *bh)
{
if (buffer_freed(bh)) {
+ WARN_ON_ONCE(buffer_dirty(bh));
clear_buffer_freed(bh);
+ clear_buffer_mapped(bh);
+ clear_buffer_new(bh);
+ clear_buffer_req(bh);
+ bh->b_bdev = NULL;
release_buffer_page(bh);
} else
put_bh(bh);
@@ -119,7 +126,6 @@ static int journal_write_commit_record(journal_t *journal,
struct buffer_head *bh;
journal_header_t *header;
int ret;
- int barrier_done = 0;
if (is_journal_aborted(journal))
return 0;
@@ -137,34 +143,12 @@ static int journal_write_commit_record(journal_t *journal,
JBUFFER_TRACE(descriptor, "write commit block");
set_buffer_dirty(bh);
- if (journal->j_flags & JFS_BARRIER) {
- set_buffer_ordered(bh);
- barrier_done = 1;
- }
- ret = sync_dirty_buffer(bh);
- if (barrier_done)
- clear_buffer_ordered(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];
- 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 &= ~JFS_BARRIER;
- spin_unlock(&journal->j_state_lock);
-
- /* And try again, without the barrier */
- set_buffer_uptodate(bh);
- set_buffer_dirty(bh);
+ if (journal->j_flags & JFS_BARRIER)
+ ret = __sync_dirty_buffer(bh, WRITE_SYNC | WRITE_FLUSH_FUA);
+ else
ret = sync_dirty_buffer(bh);
- }
+
put_bh(bh); /* One for getblk() */
journal_put_journal_head(descriptor);
@@ -178,8 +162,17 @@ static void journal_do_submit_data(struct buffer_head **wbuf, int bufs,
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_op, wbuf[i]);
+ /*
+ * Here we write back pagecache data that may be mmaped. Since
+ * we cannot afford to clean the page and set PageWriteback
+ * here due to lock ordering (page lock ranks above transaction
+ * start), the data can change while IO is in flight. Tell the
+ * block layer it should bounce the bio pages if stable data
+ * during write is required.
+ *
+ * We use up our safety reference in submit_bh().
+ */
+ _submit_bh(write_op, wbuf[i], 1 << BIO_SNAP_STABLE);
}
}
@@ -226,6 +219,8 @@ write_out_data:
if (!trylock_buffer(bh)) {
BUFFER_TRACE(bh, "needs blocking lock");
spin_unlock(&journal->j_list_lock);
+ trace_jbd_do_submit_data(journal,
+ commit_transaction);
/* Write out all data to prevent deadlocks */
journal_do_submit_data(wbuf, bufs, write_op);
bufs = 0;
@@ -258,6 +253,8 @@ write_out_data:
jbd_unlock_bh_state(bh);
if (bufs == journal->j_wbufsize) {
spin_unlock(&journal->j_list_lock);
+ trace_jbd_do_submit_data(journal,
+ commit_transaction);
journal_do_submit_data(wbuf, bufs, write_op);
bufs = 0;
goto write_out_data;
@@ -275,10 +272,6 @@ write_out_data:
jbd_unlock_bh_state(bh);
if (locked)
unlock_buffer(bh);
- journal_remove_journal_head(bh);
- /* One for our safety reference, other for
- * journal_remove_journal_head() */
- put_bh(bh);
release_data_buffer(bh);
}
@@ -288,6 +281,7 @@ write_out_data:
}
}
spin_unlock(&journal->j_list_lock);
+ trace_jbd_do_submit_data(journal, commit_transaction);
journal_do_submit_data(wbuf, bufs, write_op);
return err;
@@ -317,6 +311,7 @@ void journal_commit_transaction(journal_t *journal)
int first_tag = 0;
int tag_flag;
int i;
+ struct blk_plug plug;
int write_op = WRITE;
/*
@@ -324,16 +319,19 @@ void journal_commit_transaction(journal_t *journal)
* 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 journal_flush? */
if (journal->j_flags & JFS_FLUSHED) {
jbd_debug(3, "super block updated\n");
- journal_update_superblock(journal, 1);
+ mutex_lock(&journal->j_checkpoint_mutex);
+ /*
+ * 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.
+ */
+ journal_update_sb_log_tail(journal, journal->j_tail_sequence,
+ journal->j_tail, WRITE_SYNC);
+ mutex_unlock(&journal->j_checkpoint_mutex);
} else {
jbd_debug(3, "superblock not updated\n");
}
@@ -342,21 +340,16 @@ void journal_commit_transaction(journal_t *journal)
J_ASSERT(journal->j_committing_transaction == NULL);
commit_transaction = journal->j_running_transaction;
- J_ASSERT(commit_transaction->t_state == T_RUNNING);
+ trace_jbd_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);
+ J_ASSERT(commit_transaction->t_state == T_RUNNING);
commit_transaction->t_state = T_LOCKED;
- /*
- * Use plugged writes here, since we want to submit several before
- * we unplug the device. We don't do explicit unplugging in here,
- * instead we rely on sync_buffer() doing the unplug for us.
- */
- if (commit_transaction->t_synchronous_commit)
- write_op = WRITE_SYNC_PLUG;
+ trace_jbd_commit_locking(journal, commit_transaction);
spin_lock(&commit_transaction->t_handle_lock);
while (commit_transaction->t_updates) {
DEFINE_WAIT(wait);
@@ -391,7 +384,7 @@ void 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 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;
@@ -423,10 +416,17 @@ void journal_commit_transaction(journal_t *journal)
jbd_debug (3, "JBD: commit phase 1\n");
/*
+ * Clear revoked flag to reflect there is no revoked buffers
+ * in the next transaction which is going to be started.
+ */
+ journal_clear_buffer_revoked_flags(journal);
+
+ /*
* Switch to a new revoke table.
*/
journal_switch_revoke_table(journal);
+ trace_jbd_commit_flushing(journal, commit_transaction);
commit_transaction->t_state = T_FLUSH;
journal->j_committing_transaction = commit_transaction;
journal->j_running_transaction = NULL;
@@ -437,12 +437,17 @@ void journal_commit_transaction(journal_t *journal)
jbd_debug (3, "JBD: commit phase 2\n");
+ if (tid_geq(journal->j_commit_waited, commit_transaction->t_tid))
+ write_op = WRITE_SYNC;
+
/*
* Now start flushing things to disk, in the order they appear
* on the transaction lists. Data blocks go first.
*/
+ blk_start_plug(&plug);
err = journal_submit_data_buffers(journal, commit_transaction,
write_op);
+ blk_finish_plug(&plug);
/*
* Wait for all previously submitted IO to complete.
@@ -479,14 +484,9 @@ void journal_commit_transaction(journal_t *journal)
}
if (buffer_jbd(bh) && bh2jh(bh) == jh &&
jh->b_transaction == commit_transaction &&
- jh->b_jlist == BJ_Locked) {
+ jh->b_jlist == BJ_Locked)
__journal_unfile_buffer(jh);
- jbd_unlock_bh_state(bh);
- journal_remove_journal_head(bh);
- put_bh(bh);
- } else {
- jbd_unlock_bh_state(bh);
- }
+ jbd_unlock_bh_state(bh);
release_data_buffer(bh);
cond_resched_lock(&journal->j_list_lock);
}
@@ -503,6 +503,8 @@ void journal_commit_transaction(journal_t *journal)
err = 0;
}
+ blk_start_plug(&plug);
+
journal_write_revoke_records(journal, commit_transaction, write_op);
/*
@@ -524,6 +526,7 @@ void journal_commit_transaction(journal_t *journal)
commit_transaction->t_state = T_COMMIT;
spin_unlock(&journal->j_state_lock);
+ trace_jbd_commit_logging(journal, commit_transaction);
J_ASSERT(commit_transaction->t_nr_buffers <=
commit_transaction->t_outstanding_credits);
@@ -610,13 +613,13 @@ void journal_commit_transaction(journal_t *journal)
/* Bump b_count to prevent truncate from stumbling over
the shadowed buffer! @@@ This can go if we ever get
rid of the BJ_IO/BJ_Shadow pairing of buffers. */
- atomic_inc(&jh2bh(jh)->b_count);
+ get_bh(jh2bh(jh));
/* Make a temporary IO buffer with which to write it out
(this will requeue both the metadata buffer and the
temporary IO buffer). new_bh goes on BJ_IO*/
- set_bit(BH_JWrite, &jh2bh(jh)->b_state);
+ set_buffer_jwrite(jh2bh(jh));
/*
* akpm: journal_write_metadata_buffer() sets
* new_bh->b_transaction to commit_transaction.
@@ -626,7 +629,7 @@ void journal_commit_transaction(journal_t *journal)
JBUFFER_TRACE(jh, "ph3: write metadata");
flags = journal_write_metadata_buffer(commit_transaction,
jh, &new_jh, blocknr);
- set_bit(BH_JWrite, &jh2bh(new_jh)->b_state);
+ set_buffer_jwrite(jh2bh(new_jh));
wbuf[bufs++] = jh2bh(new_jh);
/* Record the new block's tag in the current descriptor
@@ -673,7 +676,17 @@ start_journal_io:
clear_buffer_dirty(bh);
set_buffer_uptodate(bh);
bh->b_end_io = journal_end_buffer_io_sync;
- submit_bh(write_op, bh);
+ /*
+ * In data=journal mode, here we can end up
+ * writing pagecache data that might be
+ * mmapped. Since we can't afford to clean the
+ * page and set PageWriteback (see the comment
+ * near the other use of _submit_bh()), the
+ * data can change while the write is in
+ * flight. Tell the block layer to bounce the
+ * bio pages if stable pages are required.
+ */
+ _submit_bh(write_op, bh, 1 << BIO_SNAP_STABLE);
}
cond_resched();
@@ -684,6 +697,8 @@ start_journal_io:
}
}
+ 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
@@ -736,7 +751,7 @@ wait_for_iobuf:
shadowed buffer */
jh = commit_transaction->t_shadow_list->b_tprev;
bh = jh2bh(jh);
- clear_bit(BH_JWrite, &bh->b_state);
+ clear_buffer_jwrite(bh);
J_ASSERT_BH(bh, buffer_jbddirty(bh));
/* The metadata is now released for reuse, but we need
@@ -745,8 +760,13 @@ wait_for_iobuf:
required. */
JBUFFER_TRACE(jh, "file as BJ_Forget");
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 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);
@@ -786,6 +806,12 @@ wait_for_iobuf:
jbd_debug(3, "JBD: commit phase 6\n");
+ /* All metadata is written, now write commit record and do cleanup */
+ spin_lock(&journal->j_state_lock);
+ J_ASSERT(commit_transaction->t_state == T_COMMIT);
+ commit_transaction->t_state = T_COMMIT_RECORD;
+ spin_unlock(&journal->j_state_lock);
+
if (journal_write_commit_record(journal, commit_transaction))
err = -EIO;
@@ -815,10 +841,16 @@ restart_loop:
while (commit_transaction->t_forget) {
transaction_t *cp_transaction;
struct buffer_head *bh;
+ int try_to_free = 0;
jh = commit_transaction->t_forget;
spin_unlock(&journal->j_list_lock);
bh = jh2bh(jh);
+ /*
+ * Get a reference so that bh cannot be freed before we are
+ * done with it.
+ */
+ get_bh(bh);
jbd_lock_bh_state(bh);
J_ASSERT_JH(jh, jh->b_transaction == commit_transaction ||
jh->b_transaction == journal->j_running_transaction);
@@ -858,17 +890,35 @@ restart_loop:
* there's no point in keeping a checkpoint record for
* it. */
- /* 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 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) && !jh->b_next_transaction) {
- clear_buffer_freed(bh);
- clear_buffer_jbddirty(bh);
+ /*
+ * A buffer which has been freed while still being journaled by
+ * a previous transaction.
+ */
+ if (buffer_freed(bh)) {
+ /*
+ * If the running transaction is the one containing
+ * "add to orphan" operation (b_next_transaction !=
+ * NULL), we have to wait for that transaction to
+ * commit before we can really get rid of the buffer.
+ * So just clear b_modified to not confuse transaction
+ * credit accounting and refile the buffer to
+ * BJ_Forget of the running transaction. If the just
+ * committed transaction contains "add to orphan"
+ * operation, we can completely invalidate the buffer
+ * now. We are rather throughout in that since the
+ * buffer may be still accessible when blocksize <
+ * pagesize and it is attached to the last partial
+ * page.
+ */
+ jh->b_modified = 0;
+ if (!jh->b_next_transaction) {
+ clear_buffer_freed(bh);
+ clear_buffer_jbddirty(bh);
+ clear_buffer_mapped(bh);
+ clear_buffer_new(bh);
+ clear_buffer_req(bh);
+ bh->b_bdev = NULL;
+ }
}
if (buffer_jbddirty(bh)) {
@@ -876,28 +926,27 @@ restart_loop:
__journal_insert_checkpoint(jh, commit_transaction);
if (is_journal_aborted(journal))
clear_buffer_jbddirty(bh);
- JBUFFER_TRACE(jh, "refile for checkpoint writeback");
- __journal_refile_buffer(jh);
- jbd_unlock_bh_state(bh);
} else {
J_ASSERT_BH(bh, !buffer_dirty(bh));
- /* The buffer on BJ_Forget list and not jbddirty means
+ /*
+ * The buffer on BJ_Forget list and not jbddirty means
* it has been freed by this transaction and hence it
* could not have been reallocated until this
* transaction has committed. *BUT* it could be
* reallocated once we have written all the data to
* disk and before we process the buffer on BJ_Forget
- * list. */
- JBUFFER_TRACE(jh, "refile or unfile freed buffer");
- __journal_refile_buffer(jh);
- if (!jh->b_transaction) {
- jbd_unlock_bh_state(bh);
- /* needs a brelse */
- journal_remove_journal_head(bh);
- release_buffer_page(bh);
- } else
- jbd_unlock_bh_state(bh);
+ * list.
+ */
+ if (!jh->b_next_transaction)
+ try_to_free = 1;
}
+ JBUFFER_TRACE(jh, "refile or unfile freed buffer");
+ __journal_refile_buffer(jh);
+ jbd_unlock_bh_state(bh);
+ if (try_to_free)
+ release_buffer_page(bh);
+ else
+ __brelse(bh);
cond_resched_lock(&journal->j_list_lock);
}
spin_unlock(&journal->j_list_lock);
@@ -923,7 +972,7 @@ restart_loop:
jbd_debug(3, "JBD: commit phase 8\n");
- J_ASSERT(commit_transaction->t_state == T_COMMIT);
+ J_ASSERT(commit_transaction->t_state == T_COMMIT_RECORD);
commit_transaction->t_state = T_FINISHED;
J_ASSERT(commit_transaction == journal->j_committing_transaction);
@@ -964,6 +1013,7 @@ restart_loop:
}
spin_unlock(&journal->j_list_lock);
+ trace_jbd_end_commit(journal, commit_transaction);
jbd_debug(1, "JBD: commit %d complete, head %d\n",
journal->j_commit_sequence, journal->j_tail_sequence);
generated by cgit (git 2.34.1) at 2025-08-10 21:34:59 +0000