int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
tid_t tid;
int need_to_start, ret;
if (is_handle_aborted(handle))
return 0;
J_ASSERT(atomic_read(&transaction->t_updates) > 0);
J_ASSERT(journal_current_handle() == handle);
read_lock(&journal->j_state_lock);
spin_lock(&transaction->t_handle_lock);
atomic_sub(handle->h_buffer_credits,
&transaction->t_outstanding_credits);
if (atomic_dec_and_test(&transaction->t_updates))
wake_up(&journal->j_wait_updates);
spin_unlock(&transaction->t_handle_lock);
jbd_debug(2, "restarting handle %p\n", handle);
tid = transaction->t_tid;
need_to_start = !tid_geq(journal->j_commit_request, tid);
read_unlock(&journal->j_state_lock);
if (need_to_start)
jbd2_log_start_commit(journal, tid);
lock_map_release(&handle->h_lockdep_map);
handle->h_buffer_credits = nblocks;
ret = start_this_handle(journal, handle, gfp_mask);
return ret;
}
int journal_extend(handle_t *handle, int nblocks)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
int result;
int wanted;
result = -EIO;
if (is_handle_aborted(handle))
goto out;
result = 1;
spin_lock(&journal->j_state_lock);
/* Don't extend a locked-down transaction! */
if (handle->h_transaction->t_state != T_RUNNING) {
jbd_debug(3, "denied handle %p %d blocks: "
"transaction not running\n", handle, nblocks);
goto error_out;
}
spin_lock(&transaction->t_handle_lock);
wanted = transaction->t_outstanding_credits + nblocks;
if (wanted > journal->j_max_transaction_buffers) {
jbd_debug(3, "denied handle %p %d blocks: "
"transaction too large\n", handle, nblocks);
goto unlock;
}
if (wanted > __log_space_left(journal)) {
jbd_debug(3, "denied handle %p %d blocks: "
"insufficient log space\n", handle, nblocks);
goto unlock;
}
handle->h_buffer_credits += nblocks;
transaction->t_outstanding_credits += nblocks;
result = 0;
jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
unlock:
spin_unlock(&transaction->t_handle_lock);
error_out:
spin_unlock(&journal->j_state_lock);
out:
return result;
}
int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
struct journal_head *jh = jbd2_journal_add_journal_head(bh);
int err;
jbd_debug(5, "journal_head %p\n", jh);
err = -EROFS;
if (is_handle_aborted(handle))
goto out;
err = 0;
JBUFFER_TRACE(jh, "entry");
jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
jh->b_transaction == NULL ||
(jh->b_transaction == journal->j_committing_transaction &&
jh->b_jlist == BJ_Forget)));
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
if (jh->b_transaction == NULL) {
clear_buffer_dirty(jh2bh(jh));
jh->b_modified = 0;
JBUFFER_TRACE(jh, "file as BJ_Reserved");
__jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
} else if (jh->b_transaction == journal->j_committing_transaction) {
jh->b_modified = 0;
JBUFFER_TRACE(jh, "set next transaction");
jh->b_next_transaction = transaction;
}
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
JBUFFER_TRACE(jh, "cancelling revoke");
jbd2_journal_cancel_revoke(handle, jh);
out:
jbd2_journal_put_journal_head(jh);
return err;
}
int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
if (is_handle_aborted(handle))
return -EIO;
jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
transaction->t_tid);
if (jinode->i_transaction == transaction ||
jinode->i_next_transaction == transaction)
return 0;
spin_lock(&journal->j_list_lock);
if (jinode->i_transaction == transaction ||
jinode->i_next_transaction == transaction)
goto done;
if (!transaction->t_need_data_flush)
transaction->t_need_data_flush = 1;
if (jinode->i_transaction) {
J_ASSERT(jinode->i_next_transaction == NULL);
J_ASSERT(jinode->i_transaction ==
journal->j_committing_transaction);
jinode->i_next_transaction = transaction;
goto done;
}
J_ASSERT(!jinode->i_next_transaction);
jinode->i_transaction = transaction;
list_add(&jinode->i_list, &transaction->t_inode_list);
done:
spin_unlock(&journal->j_list_lock);
return 0;
}
void ext4_journal_abort_handle(const char *caller, unsigned int line,
const char *err_fn, struct buffer_head *bh,
handle_t *handle, int err)
{
char nbuf[16];
const char *errstr = ext4_decode_error(NULL, err, nbuf);
BUG_ON(!ext4_handle_valid(handle));
if (bh)
BUFFER_TRACE(bh, "abort");
if (!handle->h_err)
handle->h_err = err;
if (is_handle_aborted(handle))
return;
printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
caller, line, errstr, err_fn);
jbd2_journal_abort_handle(handle);
}
int journal_restart(handle_t *handle, int nblocks)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
int ret;
/* If we've had an abort of any type, don't even think about
* actually doing the restart! */
if (is_handle_aborted(handle))
return 0;
/*
* First unlink the handle from its current transaction, and start the
* commit on that.
*/
J_ASSERT(transaction->t_updates > 0);
J_ASSERT(journal_current_handle() == handle);
spin_lock(&journal->j_state_lock);
spin_lock(&transaction->t_handle_lock);
transaction->t_outstanding_credits -= handle->h_buffer_credits;
transaction->t_updates--;
if (!transaction->t_updates)
wake_up(&journal->j_wait_updates);
spin_unlock(&transaction->t_handle_lock);
jbd_debug(2, "restarting handle %p\n", handle);
__log_start_commit(journal, transaction->t_tid);
spin_unlock(&journal->j_state_lock);
lock_map_release(&handle->h_lockdep_map);
handle->h_buffer_credits = nblocks;
ret = start_this_handle(journal, handle);
return ret;
}
int journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
struct journal_head *jh = bh2jh(bh);
jbd_debug(5, "journal_head %p\n", jh);
JBUFFER_TRACE(jh, "entry");
if (is_handle_aborted(handle))
goto out;
jbd_lock_bh_state(bh);
if (jh->b_modified == 0) {
/*
* This buffer's got modified and becoming part
* of the transaction. This needs to be done
* once a transaction -bzzz
*/
jh->b_modified = 1;
J_ASSERT_JH(jh, handle->h_buffer_credits > 0);
handle->h_buffer_credits--;
}
/*
* fastpath, to avoid expensive locking. If this buffer is already
* on the running transaction's metadata list there is nothing to do.
* Nobody can take it off again because there is a handle open.
* I _think_ we're OK here with SMP barriers - a mistaken decision will
* result in this test being false, so we go in and take the locks.
*/
if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
JBUFFER_TRACE(jh, "fastpath");
J_ASSERT_JH(jh, jh->b_transaction ==
journal->j_running_transaction);
goto out_unlock_bh;
}
set_buffer_jbddirty(bh);
/*
* Metadata already on the current transaction list doesn't
* need to be filed. Metadata on another transaction's list must
* be committing, and will be refiled once the commit completes:
* leave it alone for now.
*/
if (jh->b_transaction != transaction) {
JBUFFER_TRACE(jh, "already on other transaction");
J_ASSERT_JH(jh, jh->b_transaction ==
journal->j_committing_transaction);
J_ASSERT_JH(jh, jh->b_next_transaction == transaction);
/* And this case is illegal: we can't reuse another
* transaction's data buffer, ever. */
goto out_unlock_bh;
}
/* That test should have eliminated the following case: */
J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
JBUFFER_TRACE(jh, "file as BJ_Metadata");
spin_lock(&journal->j_list_lock);
__journal_file_buffer(jh, handle->h_transaction, BJ_Metadata);
spin_unlock(&journal->j_list_lock);
out_unlock_bh:
jbd_unlock_bh_state(bh);
out:
JBUFFER_TRACE(jh, "exit");
return 0;
}
int journal_dirty_data(handle_t *handle, struct buffer_head *bh)
{
journal_t *journal = handle->h_transaction->t_journal;
int need_brelse = 0;
struct journal_head *jh;
int ret = 0;
if (is_handle_aborted(handle))
return ret;
jh = journal_add_journal_head(bh);
JBUFFER_TRACE(jh, "entry");
/*
* The buffer could *already* be dirty. Writeout can start
* at any time.
*/
jbd_debug(4, "jh: %p, tid:%d\n", jh, handle->h_transaction->t_tid);
/*
* What if the buffer is already part of a running transaction?
*
* There are two cases:
* 1) It is part of the current running transaction. Refile it,
* just in case we have allocated it as metadata, deallocated
* it, then reallocated it as data.
* 2) It is part of the previous, still-committing transaction.
* If all we want to do is to guarantee that the buffer will be
* written to disk before this new transaction commits, then
* being sure that the *previous* transaction has this same
* property is sufficient for us! Just leave it on its old
* transaction.
*
* In case (2), the buffer must not already exist as metadata
* --- that would violate write ordering (a transaction is free
* to write its data at any point, even before the previous
* committing transaction has committed). The caller must
* never, ever allow this to happen: there's nothing we can do
* about it in this layer.
*/
jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
/* Now that we have bh_state locked, are we really still mapped? */
if (!buffer_mapped(bh)) {
JBUFFER_TRACE(jh, "unmapped buffer, bailing out");
goto no_journal;
}
if (jh->b_transaction) {
JBUFFER_TRACE(jh, "has transaction");
if (jh->b_transaction != handle->h_transaction) {
JBUFFER_TRACE(jh, "belongs to older transaction");
J_ASSERT_JH(jh, jh->b_transaction ==
journal->j_committing_transaction);
/* @@@ IS THIS TRUE ? */
/*
* Not any more. Scenario: someone does a write()
* in data=journal mode. The buffer's transaction has
* moved into commit. Then someone does another
* write() to the file. We do the frozen data copyout
* and set b_next_transaction to point to j_running_t.
* And while we're in that state, someone does a
* writepage() in an attempt to pageout the same area
* of the file via a shared mapping. At present that
* calls journal_dirty_data(), and we get right here.
* It may be too late to journal the data. Simply
* falling through to the next test will suffice: the
* data will be dirty and wil be checkpointed. The
* ordering comments in the next comment block still
* apply.
*/
//J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
/*
* If we're journalling data, and this buffer was
* subject to a write(), it could be metadata, forget
* or shadow against the committing transaction. Now,
* someone has dirtied the same darn page via a mapping
* and it is being writepage()'d.
* We *could* just steal the page from commit, with some
* fancy locking there. Instead, we just skip it -
* don't tie the page's buffers to the new transaction
* at all.
* Implication: if we crash before the writepage() data
* is written into the filesystem, recovery will replay
* the write() data.
*/
if (jh->b_jlist != BJ_None &&
jh->b_jlist != BJ_SyncData &&
jh->b_jlist != BJ_Locked) {
JBUFFER_TRACE(jh, "Not stealing");
goto no_journal;
}
/*
* This buffer may be undergoing writeout in commit. We
* can't return from here and let the caller dirty it
* again because that can cause the write-out loop in
* commit to never terminate.
*/
if (buffer_dirty(bh)) {
get_bh(bh);
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
need_brelse = 1;
sync_dirty_buffer(bh);
jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
/* Since we dropped the lock... */
if (!buffer_mapped(bh)) {
JBUFFER_TRACE(jh, "buffer got unmapped");
goto no_journal;
}
/* The buffer may become locked again at any
time if it is redirtied */
}
/*
* We cannot remove the buffer with io error from the
* committing transaction, because otherwise it would
* miss the error and the commit would not abort.
*/
if (unlikely(!buffer_uptodate(bh))) {
ret = -EIO;
goto no_journal;
}
if (jh->b_transaction != NULL) {
JBUFFER_TRACE(jh, "unfile from commit");
__journal_temp_unlink_buffer(jh);
/* It still points to the committing
* transaction; move it to this one so
* that the refile assert checks are
* happy. */
jh->b_transaction = handle->h_transaction;
}
/* The buffer will be refiled below */
}
/*
* Special case --- the buffer might actually have been
* allocated and then immediately deallocated in the previous,
* committing transaction, so might still be left on that
* transaction's metadata lists.
*/
if (jh->b_jlist != BJ_SyncData && jh->b_jlist != BJ_Locked) {
JBUFFER_TRACE(jh, "not on correct data list: unfile");
J_ASSERT_JH(jh, jh->b_jlist != BJ_Shadow);
__journal_temp_unlink_buffer(jh);
jh->b_transaction = handle->h_transaction;
JBUFFER_TRACE(jh, "file as data");
__journal_file_buffer(jh, handle->h_transaction,
BJ_SyncData);
}
} else {
JBUFFER_TRACE(jh, "not on a transaction");
__journal_file_buffer(jh, handle->h_transaction, BJ_SyncData);
}
no_journal:
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
if (need_brelse) {
BUFFER_TRACE(bh, "brelse");
__brelse(bh);
}
JBUFFER_TRACE(jh, "exit");
journal_put_journal_head(jh);
return ret;
}
int journal_get_create_access(handle_t *handle, struct buffer_head *bh)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
struct journal_head *jh = journal_add_journal_head(bh);
int err;
jbd_debug(5, "journal_head %p\n", jh);
err = -EROFS;
if (is_handle_aborted(handle))
goto out;
err = 0;
JBUFFER_TRACE(jh, "entry");
/*
* The buffer may already belong to this transaction due to pre-zeroing
* in the filesystem's new_block code. It may also be on the previous,
* committing transaction's lists, but it HAS to be in Forget state in
* that case: the transaction must have deleted the buffer for it to be
* reused here.
*/
jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
jh->b_transaction == NULL ||
(jh->b_transaction == journal->j_committing_transaction &&
jh->b_jlist == BJ_Forget)));
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
if (jh->b_transaction == NULL) {
/*
* Previous journal_forget() could have left the buffer
* with jbddirty bit set because it was being committed. When
* the commit finished, we've filed the buffer for
* checkpointing and marked it dirty. Now we are reallocating
* the buffer so the transaction freeing it must have
* committed and so it's safe to clear the dirty bit.
*/
clear_buffer_dirty(jh2bh(jh));
jh->b_transaction = transaction;
/* first access by this transaction */
jh->b_modified = 0;
JBUFFER_TRACE(jh, "file as BJ_Reserved");
__journal_file_buffer(jh, transaction, BJ_Reserved);
} else if (jh->b_transaction == journal->j_committing_transaction) {
/* first access by this transaction */
jh->b_modified = 0;
JBUFFER_TRACE(jh, "set next transaction");
jh->b_next_transaction = transaction;
}
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
/*
* akpm: I added this. ext3_alloc_branch can pick up new indirect
* blocks which contain freed but then revoked metadata. We need
* to cancel the revoke in case we end up freeing it yet again
* and the reallocating as data - this would cause a second revoke,
* which hits an assertion error.
*/
JBUFFER_TRACE(jh, "cancelling revoke");
journal_cancel_revoke(handle, jh);
journal_put_journal_head(jh);
out:
return err;
}
static int
do_get_write_access(handle_t *handle, struct journal_head *jh,
int force_copy)
{
struct buffer_head *bh;
transaction_t *transaction;
journal_t *journal;
int error;
char *frozen_buffer = NULL;
int need_copy = 0;
if (is_handle_aborted(handle))
return -EROFS;
transaction = handle->h_transaction;
journal = transaction->t_journal;
jbd_debug(5, "buffer_head %p, force_copy %d\n", jh, force_copy);
JBUFFER_TRACE(jh, "entry");
repeat:
bh = jh2bh(jh);
/* @@@ Need to check for errors here at some point. */
lock_buffer(bh);
jbd_lock_bh_state(bh);
/* We now hold the buffer lock so it is safe to query the buffer
* state. Is the buffer dirty?
*
* If so, there are two possibilities. The buffer may be
* non-journaled, and undergoing a quite legitimate writeback.
* Otherwise, it is journaled, and we don't expect dirty buffers
* in that state (the buffers should be marked JBD_Dirty
* instead.) So either the IO is being done under our own
* control and this is a bug, or it's a third party IO such as
* dump(8) (which may leave the buffer scheduled for read ---
* ie. locked but not dirty) or tune2fs (which may actually have
* the buffer dirtied, ugh.) */
if (buffer_dirty(bh)) {
/*
* First question: is this buffer already part of the current
* transaction or the existing committing transaction?
*/
if (jh->b_transaction) {
J_ASSERT_JH(jh,
jh->b_transaction == transaction ||
jh->b_transaction ==
journal->j_committing_transaction);
if (jh->b_next_transaction)
J_ASSERT_JH(jh, jh->b_next_transaction ==
transaction);
warn_dirty_buffer(bh);
}
/*
* In any case we need to clean the dirty flag and we must
* do it under the buffer lock to be sure we don't race
* with running write-out.
*/
JBUFFER_TRACE(jh, "Journalling dirty buffer");
clear_buffer_dirty(bh);
set_buffer_jbddirty(bh);
}
unlock_buffer(bh);
error = -EROFS;
if (is_handle_aborted(handle)) {
jbd_unlock_bh_state(bh);
goto out;
}
error = 0;
/*
* The buffer is already part of this transaction if b_transaction or
* b_next_transaction points to it
*/
if (jh->b_transaction == transaction ||
jh->b_next_transaction == transaction)
goto done;
/*
* this is the first time this transaction is touching this buffer,
* reset the modified flag
*/
jh->b_modified = 0;
/*
* If there is already a copy-out version of this buffer, then we don't
* need to make another one
*/
if (jh->b_frozen_data) {
JBUFFER_TRACE(jh, "has frozen data");
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
jh->b_next_transaction = transaction;
goto done;
}
/* Is there data here we need to preserve? */
if (jh->b_transaction && jh->b_transaction != transaction) {
JBUFFER_TRACE(jh, "owned by older transaction");
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
J_ASSERT_JH(jh, jh->b_transaction ==
journal->j_committing_transaction);
/* There is one case we have to be very careful about.
* If the committing transaction is currently writing
* this buffer out to disk and has NOT made a copy-out,
* then we cannot modify the buffer contents at all
* right now. The essence of copy-out is that it is the
* extra copy, not the primary copy, which gets
* 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);
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);
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.
*
* 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) {
JBUFFER_TRACE(jh, "generate frozen data");
if (!frozen_buffer) {
JBUFFER_TRACE(jh, "allocate memory for buffer");
jbd_unlock_bh_state(bh);
frozen_buffer =
jbd_alloc(jh2bh(jh)->b_size,
GFP_NOFS);
if (!frozen_buffer) {
printk(KERN_EMERG
"%s: OOM for frozen_buffer\n",
__func__);
JBUFFER_TRACE(jh, "oom!");
error = -ENOMEM;
jbd_lock_bh_state(bh);
goto done;
}
goto repeat;
}
jh->b_frozen_data = frozen_buffer;
frozen_buffer = NULL;
need_copy = 1;
}
jh->b_next_transaction = transaction;
}
/*
* Finally, if the buffer is not journaled right now, we need to make
* sure it doesn't get written to disk before the caller actually
* commits the new data
*/
if (!jh->b_transaction) {
JBUFFER_TRACE(jh, "no transaction");
J_ASSERT_JH(jh, !jh->b_next_transaction);
jh->b_transaction = transaction;
JBUFFER_TRACE(jh, "file as BJ_Reserved");
spin_lock(&journal->j_list_lock);
__journal_file_buffer(jh, transaction, BJ_Reserved);
spin_unlock(&journal->j_list_lock);
}
done:
if (need_copy) {
struct page *page;
int offset;
char *source;
J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
"Possible IO failure.\n");
page = jh2bh(jh)->b_page;
offset = ((unsigned long) jh2bh(jh)->b_data) & ~PAGE_MASK;
source = kmap_atomic(page, KM_USER0);
memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
kunmap_atomic(source, KM_USER0);
}
jbd_unlock_bh_state(bh);
/*
* If we are about to journal a buffer, then any revoke pending on it is
* no longer valid
*/
journal_cancel_revoke(handle, jh);
out:
if (unlikely(frozen_buffer)) /* It's usually NULL */
jbd_free(frozen_buffer, bh->b_size);
JBUFFER_TRACE(jh, "exit");
return error;
}
int journal_stop(handle_t *handle)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
int err;
pid_t pid;
J_ASSERT(journal_current_handle() == handle);
if (is_handle_aborted(handle))
err = -EIO;
else {
J_ASSERT(transaction->t_updates > 0);
err = 0;
}
if (--handle->h_ref > 0) {
jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
handle->h_ref);
return err;
}
jbd_debug(4, "Handle %p going down\n", handle);
/*
* Implement synchronous transaction batching. If the handle
* was synchronous, don't force a commit immediately. Let's
* yield and let another thread piggyback onto this transaction.
* Keep doing that while new threads continue to arrive.
* It doesn't cost much - we're about to run a commit and sleep
* on IO anyway. Speeds up many-threaded, many-dir operations
* by 30x or more...
*
* We try and optimize the sleep time against what the underlying disk
* can do, instead of having a static sleep time. This is usefull for
* the case where our storage is so fast that it is more optimal to go
* ahead and force a flush and wait for the transaction to be committed
* than it is to wait for an arbitrary amount of time for new writers to
* join the transaction. We achieve this by measuring how long it takes
* to commit a transaction, and compare it with how long this
* transaction has been running, and if run time < commit time then we
* sleep for the delta and commit. This greatly helps super fast disks
* that would see slowdowns as more threads started doing fsyncs.
*
* But don't do this if this process was the most recent one to
* perform a synchronous write. We do this to detect the case where a
* single process is doing a stream of sync writes. No point in waiting
* for joiners in that case.
*/
pid = current->pid;
if (handle->h_sync && journal->j_last_sync_writer != pid) {
u64 commit_time, trans_time;
journal->j_last_sync_writer = pid;
spin_lock(&journal->j_state_lock);
commit_time = journal->j_average_commit_time;
spin_unlock(&journal->j_state_lock);
trans_time = ktime_to_ns(ktime_sub(ktime_get(),
transaction->t_start_time));
commit_time = min_t(u64, commit_time,
1000*jiffies_to_usecs(1));
if (trans_time < commit_time) {
ktime_t expires = ktime_add_ns(ktime_get(),
commit_time);
set_current_state(TASK_UNINTERRUPTIBLE);
schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
}
}
if (handle->h_sync)
transaction->t_synchronous_commit = 1;
current->journal_info = NULL;
spin_lock(&journal->j_state_lock);
spin_lock(&transaction->t_handle_lock);
transaction->t_outstanding_credits -= handle->h_buffer_credits;
transaction->t_updates--;
if (!transaction->t_updates) {
wake_up(&journal->j_wait_updates);
if (journal->j_barrier_count)
wake_up(&journal->j_wait_transaction_locked);
}
/*
* If the handle is marked SYNC, we need to set another commit
* going! We also want to force a commit if the current
* transaction is occupying too much of the log, or if the
* transaction is too old now.
*/
if (handle->h_sync ||
transaction->t_outstanding_credits >
journal->j_max_transaction_buffers ||
time_after_eq(jiffies, transaction->t_expires)) {
/* Do this even for aborted journals: an abort still
* completes the commit thread, it just doesn't write
* anything to disk. */
tid_t tid = transaction->t_tid;
spin_unlock(&transaction->t_handle_lock);
jbd_debug(2, "transaction too old, requesting commit for "
"handle %p\n", handle);
/* This is non-blocking */
__log_start_commit(journal, transaction->t_tid);
spin_unlock(&journal->j_state_lock);
/*
* Special case: JFS_SYNC synchronous updates require us
* to wait for the commit to complete.
*/
if (handle->h_sync && !(current->flags & PF_MEMALLOC))
err = log_wait_commit(journal, tid);
} else {
spin_unlock(&transaction->t_handle_lock);
spin_unlock(&journal->j_state_lock);
}
lock_map_release(&handle->h_lockdep_map);
jbd_free_handle(handle);
return err;
}
int jbd2_journal_stop(handle_t *handle)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
int err, wait_for_commit = 0;
tid_t tid;
pid_t pid;
J_ASSERT(journal_current_handle() == handle);
if (is_handle_aborted(handle))
err = -EIO;
else {
J_ASSERT(atomic_read(&transaction->t_updates) > 0);
err = 0;
}
if (--handle->h_ref > 0) {
jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
handle->h_ref);
return err;
}
jbd_debug(4, "Handle %p going down\n", handle);
pid = current->pid;
if (handle->h_sync && journal->j_last_sync_writer != pid) {
u64 commit_time, trans_time;
journal->j_last_sync_writer = pid;
read_lock(&journal->j_state_lock);
commit_time = journal->j_average_commit_time;
read_unlock(&journal->j_state_lock);
trans_time = ktime_to_ns(ktime_sub(ktime_get(),
transaction->t_start_time));
commit_time = max_t(u64, commit_time,
1000*journal->j_min_batch_time);
commit_time = min_t(u64, commit_time,
1000*journal->j_max_batch_time);
if (trans_time < commit_time) {
ktime_t expires = ktime_add_ns(ktime_get(),
commit_time);
set_current_state(TASK_UNINTERRUPTIBLE);
schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
}
}
if (handle->h_sync)
transaction->t_synchronous_commit = 1;
current->journal_info = NULL;
atomic_sub(handle->h_buffer_credits,
&transaction->t_outstanding_credits);
if (handle->h_sync ||
(atomic_read(&transaction->t_outstanding_credits) >
journal->j_max_transaction_buffers) ||
time_after_eq(jiffies, transaction->t_expires)) {
jbd_debug(2, "transaction too old, requesting commit for "
"handle %p\n", handle);
jbd2_log_start_commit(journal, transaction->t_tid);
if (handle->h_sync && !(current->flags & PF_MEMALLOC))
wait_for_commit = 1;
}
tid = transaction->t_tid;
if (atomic_dec_and_test(&transaction->t_updates)) {
wake_up(&journal->j_wait_updates);
if (journal->j_barrier_count)
wake_up(&journal->j_wait_transaction_locked);
}
if (wait_for_commit)
err = jbd2_log_wait_commit(journal, tid);
lock_map_release(&handle->h_lockdep_map);
jbd2_free_handle(handle);
return err;
}
int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
struct journal_head *jh = bh2jh(bh);
int ret = 0;
jbd_debug(5, "journal_head %p\n", jh);
JBUFFER_TRACE(jh, "entry");
if (is_handle_aborted(handle))
goto out;
if (!buffer_jbd(bh)) {
ret = -EUCLEAN;
goto out;
}
jbd_lock_bh_state(bh);
if (jh->b_modified == 0) {
jh->b_modified = 1;
J_ASSERT_JH(jh, handle->h_buffer_credits > 0);
handle->h_buffer_credits--;
}
if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
JBUFFER_TRACE(jh, "fastpath");
if (unlikely(jh->b_transaction !=
journal->j_running_transaction)) {
printk(KERN_EMERG "JBD: %s: "
"jh->b_transaction (%llu, %p, %u) != "
"journal->j_running_transaction (%p, %u)",
journal->j_devname,
(unsigned long long) bh->b_blocknr,
jh->b_transaction,
jh->b_transaction ? jh->b_transaction->t_tid : 0,
journal->j_running_transaction,
journal->j_running_transaction ?
journal->j_running_transaction->t_tid : 0);
ret = -EINVAL;
}
goto out_unlock_bh;
}
set_buffer_jbddirty(bh);
if (jh->b_transaction != transaction) {
JBUFFER_TRACE(jh, "already on other transaction");
if (unlikely(jh->b_transaction !=
journal->j_committing_transaction)) {
printk(KERN_EMERG "JBD: %s: "
"jh->b_transaction (%llu, %p, %u) != "
"journal->j_committing_transaction (%p, %u)",
journal->j_devname,
(unsigned long long) bh->b_blocknr,
jh->b_transaction,
jh->b_transaction ? jh->b_transaction->t_tid : 0,
journal->j_committing_transaction,
journal->j_committing_transaction ?
journal->j_committing_transaction->t_tid : 0);
ret = -EINVAL;
}
if (unlikely(jh->b_next_transaction != transaction)) {
printk(KERN_EMERG "JBD: %s: "
"jh->b_next_transaction (%llu, %p, %u) != "
"transaction (%p, %u)",
journal->j_devname,
(unsigned long long) bh->b_blocknr,
jh->b_next_transaction,
jh->b_next_transaction ?
jh->b_next_transaction->t_tid : 0,
transaction, transaction->t_tid);
ret = -EINVAL;
}
goto out_unlock_bh;
}
J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
JBUFFER_TRACE(jh, "file as BJ_Metadata");
spin_lock(&journal->j_list_lock);
__jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_Metadata);
spin_unlock(&journal->j_list_lock);
out_unlock_bh:
jbd_unlock_bh_state(bh);
out:
JBUFFER_TRACE(jh, "exit");
WARN_ON(ret);
return ret;
}
static int
do_get_write_access(handle_t *handle, struct journal_head *jh,
int force_copy)
{
struct buffer_head *bh;
transaction_t *transaction;
journal_t *journal;
int error;
char *frozen_buffer = NULL;
int need_copy = 0;
if (is_handle_aborted(handle))
return -EROFS;
transaction = handle->h_transaction;
journal = transaction->t_journal;
jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
JBUFFER_TRACE(jh, "entry");
repeat:
bh = jh2bh(jh);
lock_buffer(bh);
jbd_lock_bh_state(bh);
if (buffer_dirty(bh)) {
if (jh->b_transaction) {
J_ASSERT_JH(jh,
jh->b_transaction == transaction ||
jh->b_transaction ==
journal->j_committing_transaction);
if (jh->b_next_transaction)
J_ASSERT_JH(jh, jh->b_next_transaction ==
transaction);
warn_dirty_buffer(bh);
}
JBUFFER_TRACE(jh, "Journalling dirty buffer");
clear_buffer_dirty(bh);
set_buffer_jbddirty(bh);
}
unlock_buffer(bh);
error = -EROFS;
if (is_handle_aborted(handle)) {
jbd_unlock_bh_state(bh);
goto out;
}
error = 0;
if (jh->b_transaction == transaction ||
jh->b_next_transaction == transaction)
goto done;
jh->b_modified = 0;
if (jh->b_frozen_data) {
JBUFFER_TRACE(jh, "has frozen data");
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
jh->b_next_transaction = transaction;
goto done;
}
if (jh->b_transaction && jh->b_transaction != transaction) {
JBUFFER_TRACE(jh, "owned by older transaction");
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
J_ASSERT_JH(jh, jh->b_transaction ==
journal->j_committing_transaction);
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);
JBUFFER_TRACE(jh, "on shadow: sleep");
jbd_unlock_bh_state(bh);
for ( ; ; ) {
prepare_to_wait(wqh, &wait.wait,
TASK_UNINTERRUPTIBLE);
if (jh->b_jlist != BJ_Shadow)
break;
schedule();
}
finish_wait(wqh, &wait.wait);
goto repeat;
}
if (jh->b_jlist != BJ_Forget || force_copy) {
JBUFFER_TRACE(jh, "generate frozen data");
if (!frozen_buffer) {
JBUFFER_TRACE(jh, "allocate memory for buffer");
jbd_unlock_bh_state(bh);
frozen_buffer =
jbd2_alloc(jh2bh(jh)->b_size,
GFP_NOFS);
if (!frozen_buffer) {
printk(KERN_EMERG
"%s: OOM for frozen_buffer\n",
__func__);
JBUFFER_TRACE(jh, "oom!");
error = -ENOMEM;
jbd_lock_bh_state(bh);
goto done;
}
goto repeat;
}
jh->b_frozen_data = frozen_buffer;
frozen_buffer = NULL;
need_copy = 1;
}
jh->b_next_transaction = transaction;
}
if (!jh->b_transaction) {
JBUFFER_TRACE(jh, "no transaction");
J_ASSERT_JH(jh, !jh->b_next_transaction);
JBUFFER_TRACE(jh, "file as BJ_Reserved");
spin_lock(&journal->j_list_lock);
__jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
spin_unlock(&journal->j_list_lock);
}
done:
if (need_copy) {
struct page *page;
int offset;
char *source;
J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
"Possible IO failure.\n");
page = jh2bh(jh)->b_page;
offset = offset_in_page(jh2bh(jh)->b_data);
source = kmap_atomic(page);
jbd2_buffer_frozen_trigger(jh, source + offset,
jh->b_triggers);
memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
kunmap_atomic(source);
jh->b_frozen_triggers = jh->b_triggers;
}
jbd_unlock_bh_state(bh);
jbd2_journal_cancel_revoke(handle, jh);
out:
if (unlikely(frozen_buffer))
jbd2_free(frozen_buffer, bh->b_size);
JBUFFER_TRACE(jh, "exit");
return error;
}
