/*
* Clean up transaction's list of buffers submitted for io.
* We wait for any pending IO to complete and remove any clean
* buffers. Note that we take the buffers in the opposite ordering
* from the one in which they were submitted for IO.
*
* Called with j_list_lock held.
*/
static void __wait_cp_io(journal_t *journal, transaction_t *transaction)
{
struct journal_head *jh;
struct buffer_head *bh;
tid_t this_tid;
int released = 0;
this_tid = transaction->t_tid;
restart:
/* Did somebody clean up the transaction in the meanwhile? */
if (journal->j_checkpoint_transactions != transaction ||
transaction->t_tid != this_tid)
return;
while (!released && transaction->t_checkpoint_io_list) {
jh = transaction->t_checkpoint_io_list;
bh = jh2bh(jh);
if (!jbd_trylock_bh_state(bh)) {
jbd_sync_bh(journal, bh);
spin_lock(&journal->j_list_lock);
goto restart;
}
if (buffer_locked(bh)) {
atomic_inc(&bh->b_count);
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
wait_on_buffer(bh);
/* the journal_head may have gone by now */
BUFFER_TRACE(bh, "brelse");
__brelse(bh);
spin_lock(&journal->j_list_lock);
goto restart;
}
/*
* Now in whatever state the buffer currently is, we know that
* it has been written out and so we can drop it from the list
*/
released = __journal_remove_checkpoint(jh);
jbd_unlock_bh_state(bh);
journal_remove_journal_head(bh);
__brelse(bh);
}
}
/*
* Perform an actual checkpoint. We don't write out only enough to
* satisfy the current blocked requests: rather we submit a reasonably
* sized chunk of the outstanding data to disk at once for
* efficiency. __log_wait_for_space() will retry if we didn't free enough.
*
* However, we _do_ take into account the amount requested so that once
* the IO has been queued, we can return as soon as enough of it has
* completed to disk.
*
* The journal should be locked before calling this function.
*/
int log_do_checkpoint(journal_t *journal)
{
int result;
int batch_count = 0;
struct buffer_head *bhs[NR_BATCH];
jbd_debug(1, "Start checkpoint\n");
/*
* First thing: if there are any transactions in the log which
* don't need checkpointing, just eliminate them from the
* journal straight away.
*/
result = cleanup_journal_tail(journal);
jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
if (result <= 0)
return result;
/*
* OK, we need to start writing disk blocks. Try to free up a
* quarter of the log in a single checkpoint if we can.
*/
/*
* AKPM: check this code. I had a feeling a while back that it
* degenerates into a busy loop at unmount time.
*/
spin_lock(&journal->j_list_lock);
while (journal->j_checkpoint_transactions) {
transaction_t *transaction;
struct journal_head *jh, *last_jh, *next_jh;
int drop_count = 0;
int cleanup_ret, retry = 0;
tid_t this_tid;
transaction = journal->j_checkpoint_transactions;
this_tid = transaction->t_tid;
jh = transaction->t_checkpoint_list;
last_jh = jh->b_cpprev;
next_jh = jh;
do {
struct buffer_head *bh;
jh = next_jh;
next_jh = jh->b_cpnext;
bh = jh2bh(jh);
if (!jbd_trylock_bh_state(bh)) {
jbd_sync_bh(journal, bh);
spin_lock(&journal->j_list_lock);
retry = 1;
break;
}
retry = __flush_buffer(journal, jh, bhs, &batch_count, &drop_count);
if (cond_resched_lock(&journal->j_list_lock)) {
retry = 1;
break;
}
} while (jh != last_jh && !retry);
if (batch_count) {
__flush_batch(journal, bhs, &batch_count);
retry = 1;
}
/*
* If someone cleaned up this transaction while we slept, we're
* done
*/
if (journal->j_checkpoint_transactions != transaction)
break;
if (retry)
continue;
/*
* Maybe it's a new transaction, but it fell at the same
* address
*/
if (transaction->t_tid != this_tid)
continue;
/*
* We have walked the whole transaction list without
* finding anything to write to disk. We had better be
* able to make some progress or we are in trouble.
*/
cleanup_ret = __cleanup_transaction(journal, transaction);
J_ASSERT(drop_count != 0 || cleanup_ret != 0);
if (journal->j_checkpoint_transactions != transaction)
break;
}
spin_unlock(&journal->j_list_lock);
result = cleanup_journal_tail(journal);
if (result < 0)
return result;
return 0;
}
/*
* Clean up a transaction's checkpoint list.
*
* We wait for any pending IO to complete and make sure any clean
* buffers are removed from the transaction.
*
* Return 1 if we performed any actions which might have destroyed the
* checkpoint. (journal_remove_checkpoint() deletes the transaction when
* the last checkpoint buffer is cleansed)
*
* Called with j_list_lock held.
*/
static int __cleanup_transaction(journal_t *journal, transaction_t *transaction)
{
struct journal_head *jh, *next_jh, *last_jh;
struct buffer_head *bh;
int ret = 0;
assert_spin_locked(&journal->j_list_lock);
jh = transaction->t_checkpoint_list;
if (!jh)
return 0;
last_jh = jh->b_cpprev;
next_jh = jh;
do {
jh = next_jh;
bh = jh2bh(jh);
if (buffer_locked(bh)) {
atomic_inc(&bh->b_count);
spin_unlock(&journal->j_list_lock);
wait_on_buffer(bh);
/* the journal_head may have gone by now */
BUFFER_TRACE(bh, "brelse");
__brelse(bh);
goto out_return_1;
}
/*
* This is foul
*/
if (!jbd_trylock_bh_state(bh)) {
jbd_sync_bh(journal, bh);
goto out_return_1;
}
if (jh->b_transaction != NULL) {
transaction_t *t = jh->b_transaction;
tid_t tid = t->t_tid;
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
log_start_commit(journal, tid);
log_wait_commit(journal, tid);
goto out_return_1;
}
/*
* AKPM: I think the buffer_jbddirty test is redundant - it
* shouldn't have NULL b_transaction?
*/
next_jh = jh->b_cpnext;
if (!buffer_dirty(bh) && !buffer_jbddirty(bh)) {
BUFFER_TRACE(bh, "remove from checkpoint");
__journal_remove_checkpoint(jh);
jbd_unlock_bh_state(bh);
journal_remove_journal_head(bh);
__brelse(bh);
ret = 1;
} else {
jbd_unlock_bh_state(bh);
}
} while (jh != last_jh);
return ret;
out_return_1:
spin_lock(&journal->j_list_lock);
return 1;
}
/*
* Perform an actual checkpoint. We take the first transaction on the
* list of transactions to be checkpointed and send all its buffers
* to disk. We submit larger chunks of data at once.
*
* The journal should be locked before calling this function.
*/
int log_do_checkpoint(journal_t *journal)
{
transaction_t *transaction;
tid_t this_tid;
int result;
jbd_debug(1, "Start checkpoint\n");
/*
* First thing: if there are any transactions in the log which
* don't need checkpointing, just eliminate them from the
* journal straight away.
*/
result = cleanup_journal_tail(journal);
jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
if (result <= 0)
return result;
/*
* OK, we need to start writing disk blocks. Take one transaction
* and write it.
*/
spin_lock(&journal->j_list_lock);
if (!journal->j_checkpoint_transactions)
goto out;
transaction = journal->j_checkpoint_transactions;
this_tid = transaction->t_tid;
restart:
/*
* If someone cleaned up this transaction while we slept, we're
* done (maybe it's a new transaction, but it fell at the same
* address).
*/
if (journal->j_checkpoint_transactions == transaction &&
transaction->t_tid == this_tid) {
int batch_count = 0;
struct buffer_head *bhs[NR_BATCH];
struct journal_head *jh;
int retry = 0;
while (!retry && transaction->t_checkpoint_list) {
struct buffer_head *bh;
jh = transaction->t_checkpoint_list;
bh = jh2bh(jh);
if (!jbd_trylock_bh_state(bh)) {
jbd_sync_bh(journal, bh);
retry = 1;
break;
}
retry = __process_buffer(journal, jh, bhs,&batch_count);
if (!retry && lock_need_resched(&journal->j_list_lock)){
spin_unlock(&journal->j_list_lock);
retry = 1;
break;
}
}
if (batch_count) {
if (!retry) {
spin_unlock(&journal->j_list_lock);
retry = 1;
}
__flush_batch(journal, bhs, &batch_count);
}
if (retry) {
spin_lock(&journal->j_list_lock);
goto restart;
}
/*
* Now we have cleaned up the first transaction's checkpoint
* list. Let's clean up the second one
*/
__wait_cp_io(journal, transaction);
}
out:
spin_unlock(&journal->j_list_lock);
result = cleanup_journal_tail(journal);
if (result < 0)
return result;
return 0;
}