summaryrefslogtreecommitdiffstats
path: root/fs/jbd2/transaction.c
diff options
context:
space:
mode:
authorJan Kara <jack@suse.cz>2013-06-04 12:08:56 -0400
committerTheodore Ts'o <tytso@mit.edu>2013-06-04 12:08:56 -0400
commitb34090e5e22a02fba0e4473056cce9420ad9dd0b (patch)
tree7ffb9ecd10ada2aefe9079c2df91405592132e47 /fs/jbd2/transaction.c
parente5a120aeb57f40ae568a5ca1dd6ace53d0213582 (diff)
jbd2: refine waiting for shadow buffers
Currently when we add a buffer to a transaction, we wait until the buffer is removed from BJ_Shadow list (so that we prevent any changes to the buffer that is just written to the journal). This can take unnecessarily long as a lot happens between the time the buffer is submitted to the journal and the time when we remove the buffer from BJ_Shadow list. (e.g. We wait for all data buffers in the transaction, we issue a cache flush, etc.) Also this creates a dependency of do_get_write_access() on transaction commit (namely waiting for data IO to complete) which we want to avoid when implementing transaction reservation. So we modify commit code to set new BH_Shadow flag when temporary shadowing buffer is created and we clear that flag once IO on that buffer is complete. This allows do_get_write_access() to wait only for BH_Shadow bit and thus removes the dependency on data IO completion. Reviewed-by: Zheng Liu <wenqing.lz@taobao.com> Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Diffstat (limited to 'fs/jbd2/transaction.c')
-rw-r--r--fs/jbd2/transaction.c44
1 files changed, 19 insertions, 25 deletions
diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
index f1c5392e62b..6f4248dd875 100644
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -619,6 +619,12 @@ static void warn_dirty_buffer(struct buffer_head *bh)
bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
}
+static int sleep_on_shadow_bh(void *word)
+{
+ io_schedule();
+ return 0;
+}
+
/*
* If the buffer is already part of the current transaction, then there
* is nothing we need to do. If it is already part of a prior
@@ -754,41 +760,29 @@ repeat:
* journaled. If the primary copy is already going to
* disk then we cannot do copy-out here. */
- if (jh->b_jlist == BJ_Shadow) {
- DEFINE_WAIT_BIT(wait, &bh->b_state, BH_Unshadow);
- wait_queue_head_t *wqh;
-
- wqh = bit_waitqueue(&bh->b_state, BH_Unshadow);
-
+ if (buffer_shadow(bh)) {
JBUFFER_TRACE(jh, "on shadow: sleep");
jbd_unlock_bh_state(bh);
- /* commit wakes up all shadow buffers after IO */
- for ( ; ; ) {
- prepare_to_wait(wqh, &wait.wait,
- TASK_UNINTERRUPTIBLE);
- if (jh->b_jlist != BJ_Shadow)
- break;
- schedule();
- }
- finish_wait(wqh, &wait.wait);
+ wait_on_bit(&bh->b_state, BH_Shadow,
+ sleep_on_shadow_bh, TASK_UNINTERRUPTIBLE);
goto repeat;
}
- /* Only do the copy if the currently-owning transaction
- * still needs it. If it is on the Forget list, the
- * committing transaction is past that stage. The
- * buffer had better remain locked during the kmalloc,
- * but that should be true --- we hold the journal lock
- * still and the buffer is already on the BUF_JOURNAL
- * list so won't be flushed.
+ /*
+ * Only do the copy if the currently-owning transaction still
+ * needs it. If buffer isn't on BJ_Metadata list, the
+ * committing transaction is past that stage (here we use the
+ * fact that BH_Shadow is set under bh_state lock together with
+ * refiling to BJ_Shadow list and at this point we know the
+ * buffer doesn't have BH_Shadow set).
*
* Subtle point, though: if this is a get_undo_access,
* then we will be relying on the frozen_data to contain
* the new value of the committed_data record after the
* transaction, so we HAVE to force the frozen_data copy
- * in that case. */
-
- if (jh->b_jlist != BJ_Forget || force_copy) {
+ * in that case.
+ */
+ if (jh->b_jlist == BJ_Metadata || force_copy) {
JBUFFER_TRACE(jh, "generate frozen data");
if (!frozen_buffer) {
JBUFFER_TRACE(jh, "allocate memory for buffer");