/* Done it all: now write 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
* entirely.
*
* Returns 1 if the journal needs to be aborted or 0 on success
*/
static int journal_write_commit_record(journal_t *journal,
transaction_t *commit_transaction)
{
struct journal_head *descriptor;
struct buffer_head *bh;
journal_header_t *header;
int ret;
if (is_journal_aborted(journal))
return 0;
descriptor = journal_get_descriptor_buffer(journal);
if (!descriptor)
return 1;
bh = jh2bh(descriptor);
header = (journal_header_t *)(bh->b_data);
header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
header->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
JBUFFER_TRACE(descriptor, "write commit block");
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);
return (ret == -EIO);
}
/* Done it all: now write 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
* entirely.
*
* Returns 1 if the journal needs to be aborted or 0 on success
*/
static int journal_write_commit_record(journal_t *journal,
transaction_t *commit_transaction)
{
struct journal_head *descriptor;
struct buffer_head *bh;
journal_header_t *header;
int ret;
if (is_journal_aborted(journal))
return 0;
descriptor = journal_get_descriptor_buffer(journal);
if (!descriptor)
return 1;
bh = jh2bh(descriptor);
header = (journal_header_t *)(bh->b_data);
header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
header->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
JBUFFER_TRACE(descriptor, "write commit block");
set_buffer_dirty(bh);
if (journal->j_flags & JFS_BARRIER) {
ret = __sync_dirty_buffer(bh, WRITE_SYNC | WRITE_BARRIER);
/*
* 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) {
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);
ret = sync_dirty_buffer(bh);
}
} else {
ret = sync_dirty_buffer(bh);
}
put_bh(bh); /* One for getblk() */
journal_put_journal_head(descriptor);
return (ret == -EIO);
}
int journal_get_write_access(handle_t *handle, struct buffer_head *bh)
{
struct journal_head *jh = journal_add_journal_head(bh);
int rc;
/* We do not want to get caught playing with fields which the
* log thread also manipulates. Make sure that the buffer
* completes any outstanding IO before proceeding. */
rc = do_get_write_access(handle, jh, 0);
journal_put_journal_head(jh);
return rc;
}
int journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
{
int err;
struct journal_head *jh = journal_add_journal_head(bh);
char *committed_data = NULL;
JBUFFER_TRACE(jh, "entry");
/*
* Do this first --- it can drop the journal lock, so we want to
* make sure that obtaining the committed_data is done
* atomically wrt. completion of any outstanding commits.
*/
err = do_get_write_access(handle, jh, 1);
if (err)
goto out;
repeat:
if (!jh->b_committed_data) {
committed_data = jbd_alloc(jh2bh(jh)->b_size, GFP_NOFS);
if (!committed_data) {
printk(KERN_EMERG "%s: No memory for committed data\n",
__func__);
err = -ENOMEM;
goto out;
}
}
jbd_lock_bh_state(bh);
if (!jh->b_committed_data) {
/* Copy out the current buffer contents into the
* preserved, committed copy. */
JBUFFER_TRACE(jh, "generate b_committed data");
if (!committed_data) {
jbd_unlock_bh_state(bh);
goto repeat;
}
jh->b_committed_data = committed_data;
committed_data = NULL;
memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
}
jbd_unlock_bh_state(bh);
out:
journal_put_journal_head(jh);
if (unlikely(committed_data))
jbd_free(committed_data, bh->b_size);
return err;
}
int journal_try_to_free_buffers(journal_t *journal,
struct page *page, gfp_t gfp_mask)
{
struct buffer_head *head;
struct buffer_head *bh;
int ret = 0;
J_ASSERT(PageLocked(page));
head = page_buffers(page);
bh = head;
do {
struct journal_head *jh;
/*
* We take our own ref against the journal_head here to avoid
* having to add tons of locking around each instance of
* journal_remove_journal_head() and journal_put_journal_head().
*/
jh = journal_grab_journal_head(bh);
if (!jh)
continue;
jbd_lock_bh_state(bh);
__journal_try_to_free_buffer(journal, bh);
journal_put_journal_head(jh);
jbd_unlock_bh_state(bh);
if (buffer_jbd(bh))
goto busy;
} while ((bh = bh->b_this_page) != head);
ret = try_to_free_buffers(page);
busy:
return ret;
}
/*
* journal_commit_transaction
*
* The primary function for committing a transaction to the log. This
* function is called by the journal thread to begin a complete commit.
*/
void journal_commit_transaction(journal_t *journal)
{
transaction_t *commit_transaction;
struct journal_head *jh, *new_jh, *descriptor;
struct buffer_head **wbuf = journal->j_wbuf;
int bufs;
int flags;
int err;
unsigned int 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;
struct blk_plug plug;
/*
* First job: lock down the current transaction and wait for
* all outstanding updates to complete.
*/
/* 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);
} else {
jbd_debug(3, "superblock not updated\n");
}
J_ASSERT(journal->j_running_transaction != NULL);
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);
commit_transaction->t_state = T_LOCKED;
trace_jbd_commit_locking(journal, commit_transaction);
spin_lock(&commit_transaction->t_handle_lock);
while (commit_transaction->t_updates) {
DEFINE_WAIT(wait);
prepare_to_wait(&journal->j_wait_updates, &wait,
TASK_UNINTERRUPTIBLE);
if (commit_transaction->t_updates) {
spin_unlock(&commit_transaction->t_handle_lock);
spin_unlock(&journal->j_state_lock);
schedule();
spin_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 <=
journal->j_max_transaction_buffers);
/*
* First thing we are allowed to do is to discard any remaining
* BJ_Reserved buffers. Note, it is _not_ permissible to assume
* that there are no such buffers: if a large filesystem
* operation like a truncate needs to split itself over multiple
* transactions, then it may try to do a journal_restart() while
* there are still BJ_Reserved buffers outstanding. These must
* be released cleanly from the current transaction.
*
* In this case, the filesystem must still reserve write access
* again before modifying the buffer in the new transaction, but
* 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 permissible.
*/
while (commit_transaction->t_reserved_list) {
jh = commit_transaction->t_reserved_list;
JBUFFER_TRACE(jh, "reserved, unused: refile");
/*
* A journal_get_undo_access()+journal_release_buffer() may
* leave undo-committed data.
*/
if (jh->b_committed_data) {
struct buffer_head *bh = jh2bh(jh);
jbd_lock_bh_state(bh);
jbd_free(jh->b_committed_data, bh->b_size);
jh->b_committed_data = NULL;
jbd_unlock_bh_state(bh);
}
journal_refile_buffer(journal, jh);
}
/*
* Now try to drop any written-back buffers from the journal's
* checkpoint lists. We do this *before* commit because it potentially
* frees some memory
*/
spin_lock(&journal->j_list_lock);
__journal_clean_checkpoint_list(journal);
spin_unlock(&journal->j_list_lock);
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;
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);
jbd_debug (3, "JBD: commit phase 2\n");
/*
* 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_SYNC);
blk_finish_plug(&plug);
/*
* Wait for all previously submitted IO to complete.
*/
spin_lock(&journal->j_list_lock);
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);
spin_lock(&journal->j_list_lock);
}
if (unlikely(!buffer_uptodate(bh))) {
if (!trylock_page(bh->b_page)) {
spin_unlock(&journal->j_list_lock);
lock_page(bh->b_page);
spin_lock(&journal->j_list_lock);
}
if (bh->b_page->mapping)
set_bit(AS_EIO, &bh->b_page->mapping->flags);
unlock_page(bh->b_page);
SetPageError(bh->b_page);
err = -EIO;
}
if (!inverted_lock(journal, bh)) {
put_bh(bh);
spin_lock(&journal->j_list_lock);
continue;
}
if (buffer_jbd(bh) && bh2jh(bh) == jh &&
jh->b_transaction == commit_transaction &&
jh->b_jlist == BJ_Locked)
__journal_unfile_buffer(jh);
jbd_unlock_bh_state(bh);
release_data_buffer(bh);
cond_resched_lock(&journal->j_list_lock);
}
spin_unlock(&journal->j_list_lock);
if (err) {
char b[BDEVNAME_SIZE];
printk(KERN_WARNING
"JBD: Detected IO errors while flushing file data "
"on %s\n", bdevname(journal->j_fs_dev, b));
if (journal->j_flags & JFS_ABORT_ON_SYNCDATA_ERR)
journal_abort(journal, err);
err = 0;
}
blk_start_plug(&plug);
journal_write_revoke_records(journal, commit_transaction, WRITE_SYNC);
/*
* 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:
*/
spin_lock(&journal->j_state_lock);
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);
descriptor = NULL;
bufs = 0;
while (commit_transaction->t_buffers) {
/* Find the next buffer to be journaled... */
jh = commit_transaction->t_buffers;
/* If we're in abort mode, we just un-journal the buffer and
release it. */
if (is_journal_aborted(journal)) {
clear_buffer_jbddirty(jh2bh(jh));
JBUFFER_TRACE(jh, "journal is aborting: refile");
journal_refile_buffer(journal, jh);
/* If that was the last one, we need to clean up
* any descriptor buffers which may have been
* already allocated, even if we are now
* aborting. */
if (!commit_transaction->t_buffers)
goto start_journal_io;
continue;
}
/* Make sure we have a descriptor block in which to
record the metadata buffer. */
if (!descriptor) {
struct buffer_head *bh;
J_ASSERT (bufs == 0);
jbd_debug(4, "JBD: get descriptor\n");
descriptor = journal_get_descriptor_buffer(journal);
if (!descriptor) {
journal_abort(journal, -EIO);
continue;
}
bh = jh2bh(descriptor);
jbd_debug(4, "JBD: got buffer %llu (%p)\n",
(unsigned long long)bh->b_blocknr, bh->b_data);
header = (journal_header_t *)&bh->b_data[0];
header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
header->h_blocktype = cpu_to_be32(JFS_DESCRIPTOR_BLOCK);
header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
tagp = &bh->b_data[sizeof(journal_header_t)];
space_left = bh->b_size - sizeof(journal_header_t);
first_tag = 1;
set_buffer_jwrite(bh);
set_buffer_dirty(bh);
wbuf[bufs++] = bh;
/* Record it so that we can wait for IO
completion later */
BUFFER_TRACE(bh, "ph3: file as descriptor");
journal_file_buffer(descriptor, commit_transaction,
BJ_LogCtl);
}
/* Where is the buffer to be written? */
err = journal_next_log_block(journal, &blocknr);
/* If the block mapping failed, just abandon the buffer
and repeat this loop: we'll fall into the
refile-on-abort condition above. */
if (err) {
journal_abort(journal, err);
continue;
}
/*
* start_this_handle() uses t_outstanding_credits to determine
* the free space in the log, but this counter is changed
* by journal_next_log_block() also.
*/
commit_transaction->t_outstanding_credits--;
/* 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. */
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_buffer_jwrite(jh2bh(jh));
/*
* akpm: journal_write_metadata_buffer() sets
* new_bh->b_transaction to commit_transaction.
* We need to clean this up before we release new_bh
* (which is of type BJ_IO)
*/
JBUFFER_TRACE(jh, "ph3: write metadata");
flags = journal_write_metadata_buffer(commit_transaction,
jh, &new_jh, blocknr);
set_buffer_jwrite(jh2bh(new_jh));
wbuf[bufs++] = jh2bh(new_jh);
/* Record the new block's tag in the current descriptor
buffer */
tag_flag = 0;
if (flags & 1)
tag_flag |= JFS_FLAG_ESCAPE;
if (!first_tag)
tag_flag |= JFS_FLAG_SAME_UUID;
tag = (journal_block_tag_t *) tagp;
tag->t_blocknr = cpu_to_be32(jh2bh(jh)->b_blocknr);
tag->t_flags = cpu_to_be32(tag_flag);
tagp += sizeof(journal_block_tag_t);
space_left -= sizeof(journal_block_tag_t);
if (first_tag) {
memcpy (tagp, journal->j_uuid, 16);
tagp += 16;
space_left -= 16;
first_tag = 0;
}
/* If there's no more to do, or if the descriptor is full,
let the IO rip! */
if (bufs == journal->j_wbufsize ||
commit_transaction->t_buffers == NULL ||
space_left < sizeof(journal_block_tag_t) + 16) {
jbd_debug(4, "JBD: Submit %d IOs\n", bufs);
/* Write an end-of-descriptor marker before
submitting the IOs. "tag" still points to
the last tag we set up. */
tag->t_flags |= cpu_to_be32(JFS_FLAG_LAST_TAG);
start_journal_io:
for (i = 0; i < bufs; i++) {
struct buffer_head *bh = wbuf[i];
lock_buffer(bh);
clear_buffer_dirty(bh);
set_buffer_uptodate(bh);
bh->b_end_io = journal_end_buffer_io_sync;
submit_bh(WRITE_SYNC, bh);
}
cond_resched();
/* Force a new descriptor to be generated next
time round the loop. */
descriptor = NULL;
bufs = 0;
}
}
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
transaction's t_log_list queue, and metadata buffers are on
the t_iobuf_list queue.
Wait for the buffers in reverse order. That way we are
less likely to be woken up until all IOs have completed, and
so we incur less scheduling load.
*/
jbd_debug(3, "JBD: commit phase 4\n");
/*
* akpm: these are BJ_IO, and j_list_lock is not needed.
* See __journal_try_to_free_buffer.
*/
wait_for_iobuf:
while (commit_transaction->t_iobuf_list != NULL) {
struct buffer_head *bh;
jh = commit_transaction->t_iobuf_list->b_tprev;
bh = jh2bh(jh);
if (buffer_locked(bh)) {
wait_on_buffer(bh);
goto wait_for_iobuf;
}
if (cond_resched())
goto wait_for_iobuf;
if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
clear_buffer_jwrite(bh);
JBUFFER_TRACE(jh, "ph4: unfile after journal write");
journal_unfile_buffer(journal, jh);
/*
* ->t_iobuf_list should contain only dummy buffer_heads
* which were created by journal_write_metadata_buffer().
*/
BUFFER_TRACE(bh, "dumping temporary bh");
journal_put_journal_head(jh);
__brelse(bh);
J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
free_buffer_head(bh);
/* We also have to unlock and free the corresponding
shadowed buffer */
jh = commit_transaction->t_shadow_list->b_tprev;
bh = jh2bh(jh);
clear_buffer_jwrite(bh);
J_ASSERT_BH(bh, buffer_jbddirty(bh));
/* The metadata is now released for reuse, but we need
to remember it against this transaction so that when
we finally commit, we can do any checkpointing
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. 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);
}
J_ASSERT (commit_transaction->t_shadow_list == NULL);
jbd_debug(3, "JBD: commit phase 5\n");
/* Here we wait for the revoke record and descriptor record buffers */
wait_for_ctlbuf:
while (commit_transaction->t_log_list != NULL) {
struct buffer_head *bh;
jh = commit_transaction->t_log_list->b_tprev;
bh = jh2bh(jh);
if (buffer_locked(bh)) {
wait_on_buffer(bh);
goto wait_for_ctlbuf;
}
if (cond_resched())
goto wait_for_ctlbuf;
if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
clear_buffer_jwrite(bh);
journal_unfile_buffer(journal, jh);
journal_put_journal_head(jh);
__brelse(bh); /* One for getblk */
/* AKPM: bforget here */
}
if (err)
journal_abort(journal, err);
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;
if (err)
journal_abort(journal, err);
/* End of a transaction! Finally, we can do checkpoint
processing: any buffers committed as a result of this
transaction can be removed from any checkpoint list it was on
before. */
jbd_debug(3, "JBD: commit phase 7\n");
J_ASSERT(commit_transaction->t_sync_datalist == NULL);
J_ASSERT(commit_transaction->t_buffers == NULL);
J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
J_ASSERT(commit_transaction->t_iobuf_list == NULL);
J_ASSERT(commit_transaction->t_shadow_list == NULL);
J_ASSERT(commit_transaction->t_log_list == NULL);
restart_loop:
/*
* As there are other places (journal_unmap_buffer()) adding buffers
* to this list we have to be careful and hold the j_list_lock.
*/
spin_lock(&journal->j_list_lock);
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);
/*
* If there is undo-protected committed data against
* this buffer, then we can remove it now. If it is a
* buffer needing such protection, the old frozen_data
* field now points to a committed version of the
* buffer, so rotate that field to the new committed
* data.
*
* Otherwise, we can just throw away the frozen data now.
*/
if (jh->b_committed_data) {
jbd_free(jh->b_committed_data, bh->b_size);
jh->b_committed_data = NULL;
if (jh->b_frozen_data) {
jh->b_committed_data = jh->b_frozen_data;
jh->b_frozen_data = NULL;
}
} else if (jh->b_frozen_data) {
jbd_free(jh->b_frozen_data, bh->b_size);
jh->b_frozen_data = NULL;
}
spin_lock(&journal->j_list_lock);
cp_transaction = jh->b_cp_transaction;
if (cp_transaction) {
JBUFFER_TRACE(jh, "remove from old cp transaction");
__journal_remove_checkpoint(jh);
}
/* Only re-checkpoint the buffer_head if it is marked
* dirty. If the buffer was added to the BJ_Forget list
* by journal_forget, it may no longer be dirty and
* 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);
}
if (buffer_jbddirty(bh)) {
JBUFFER_TRACE(jh, "add to new checkpointing trans");
__journal_insert_checkpoint(jh, commit_transaction);
if (is_journal_aborted(journal))
clear_buffer_jbddirty(bh);
} else {
J_ASSERT_BH(bh, !buffer_dirty(bh));
/*
* 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.
*/
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);
/*
* This is a bit sleazy. We use j_list_lock to protect transition
* of a transaction into T_FINISHED state and calling
* __journal_drop_transaction(). Otherwise we could race with
* other checkpointing code processing the transaction...
*/
spin_lock(&journal->j_state_lock);
spin_lock(&journal->j_list_lock);
/*
* Now recheck if some buffers did not get attached to the transaction
* while the lock was dropped...
*/
if (commit_transaction->t_forget) {
spin_unlock(&journal->j_list_lock);
spin_unlock(&journal->j_state_lock);
goto restart_loop;
}
/* Done with this transaction! */
jbd_debug(3, "JBD: commit phase 8\n");
J_ASSERT(commit_transaction->t_state == T_COMMIT_RECORD);
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;
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 commit time
*/
if (likely(journal->j_average_commit_time))
journal->j_average_commit_time = (commit_time*3 +
journal->j_average_commit_time) / 4;
else
journal->j_average_commit_time = commit_time;
spin_unlock(&journal->j_state_lock);
if (commit_transaction->t_checkpoint_list == NULL &&
commit_transaction->t_checkpoint_io_list == NULL) {
__journal_drop_transaction(journal, commit_transaction);
} else {
if (journal->j_checkpoint_transactions == NULL) {
journal->j_checkpoint_transactions = commit_transaction;
commit_transaction->t_cpnext = commit_transaction;
commit_transaction->t_cpprev = commit_transaction;
} else {
commit_transaction->t_cpnext =
journal->j_checkpoint_transactions;
commit_transaction->t_cpprev =
commit_transaction->t_cpnext->t_cpprev;
commit_transaction->t_cpnext->t_cpprev =
commit_transaction;
commit_transaction->t_cpprev->t_cpnext =
commit_transaction;
}
}
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);
wake_up(&journal->j_wait_done_commit);
}
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 journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
{
transaction_t *transaction;
struct journal_head *jh;
int may_free = 1;
int ret;
BUFFER_TRACE(bh, "entry");
/*
* It is safe to proceed here without the j_list_lock because the
* buffers cannot be stolen by try_to_free_buffers as long as we are
* holding the page lock. --sct
*/
if (!buffer_jbd(bh))
goto zap_buffer_unlocked;
spin_lock(&journal->j_state_lock);
jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
jh = journal_grab_journal_head(bh);
if (!jh)
goto zap_buffer_no_jh;
/*
* We cannot remove the buffer from checkpoint lists until the
* transaction adding inode to orphan list (let's call it T)
* is committed. Otherwise if the transaction changing the
* buffer would be cleaned from the journal before T is
* committed, a crash will cause that the correct contents of
* the buffer will be lost. On the other hand we have to
* clear the buffer dirty bit at latest at the moment when the
* transaction marking the buffer as freed in the filesystem
* structures is committed because from that moment on the
* buffer can be reallocated and used by a different page.
* Since the block hasn't been freed yet but the inode has
* already been added to orphan list, it is safe for us to add
* the buffer to BJ_Forget list of the newest transaction.
*/
transaction = jh->b_transaction;
if (transaction == NULL) {
/* First case: not on any transaction. If it
* has no checkpoint link, then we can zap it:
* it's a writeback-mode buffer so we don't care
* if it hits disk safely. */
if (!jh->b_cp_transaction) {
JBUFFER_TRACE(jh, "not on any transaction: zap");
goto zap_buffer;
}
if (!buffer_dirty(bh)) {
/* bdflush has written it. We can drop it now */
goto zap_buffer;
}
/* OK, it must be in the journal but still not
* written fully to disk: it's metadata or
* journaled data... */
if (journal->j_running_transaction) {
/* ... and once the current transaction has
* committed, the buffer won't be needed any
* longer. */
JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
ret = __dispose_buffer(jh,
journal->j_running_transaction);
journal_put_journal_head(jh);
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
spin_unlock(&journal->j_state_lock);
return ret;
} else {
/* There is no currently-running transaction. So the
* orphan record which we wrote for this file must have
* passed into commit. We must attach this buffer to
* the committing transaction, if it exists. */
if (journal->j_committing_transaction) {
JBUFFER_TRACE(jh, "give to committing trans");
ret = __dispose_buffer(jh,
journal->j_committing_transaction);
journal_put_journal_head(jh);
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
spin_unlock(&journal->j_state_lock);
return ret;
} else {
/* The orphan record's transaction has
* committed. We can cleanse this buffer */
clear_buffer_jbddirty(bh);
goto zap_buffer;
}
}
} else if (transaction == journal->j_committing_transaction) {
JBUFFER_TRACE(jh, "on committing transaction");
if (jh->b_jlist == BJ_Locked) {
/*
* The buffer is on the committing transaction's locked
* list. We have the buffer locked, so I/O has
* completed. So we can nail the buffer now.
*/
may_free = __dispose_buffer(jh, transaction);
goto zap_buffer;
}
/*
* The buffer is committing, we simply cannot touch
* it. So we just set j_next_transaction to the
* running transaction (if there is one) and mark
* buffer as freed so that commit code knows it should
* clear dirty bits when it is done with the buffer.
*/
set_buffer_freed(bh);
if (journal->j_running_transaction && buffer_jbddirty(bh))
jh->b_next_transaction = journal->j_running_transaction;
journal_put_journal_head(jh);
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
spin_unlock(&journal->j_state_lock);
return 0;
} else {
/* Good, the buffer belongs to the running transaction.
* We are writing our own transaction's data, not any
* previous one's, so it is safe to throw it away
* (remember that we expect the filesystem to have set
* i_size already for this truncate so recovery will not
* expose the disk blocks we are discarding here.) */
J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
JBUFFER_TRACE(jh, "on running transaction");
may_free = __dispose_buffer(jh, transaction);
}
zap_buffer:
journal_put_journal_head(jh);
zap_buffer_no_jh:
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
spin_unlock(&journal->j_state_lock);
zap_buffer_unlocked:
clear_buffer_dirty(bh);
J_ASSERT_BH(bh, !buffer_jbddirty(bh));
clear_buffer_mapped(bh);
clear_buffer_req(bh);
clear_buffer_new(bh);
bh->b_bdev = NULL;
return may_free;
}
/*
* journal_commit_transaction
*
* The primary function for committing a transaction to the log. This
* function is called by the journal thread to begin a complete commit.
*/
void journal_commit_transaction(journal_t *journal)
{
transaction_t *commit_transaction;
struct journal_head *jh, *new_jh, *descriptor;
struct buffer_head *wbuf[64];
int bufs;
int flags;
int err;
unsigned long blocknr;
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;
/*
* 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 journal_flush? */
if (journal->j_flags & JFS_FLUSHED) {
jbd_debug(3, "super block updated\n");
journal_update_superblock(journal, 1);
} else {
jbd_debug(3, "superblock not updated\n");
}
J_ASSERT(journal->j_running_transaction != NULL);
J_ASSERT(journal->j_committing_transaction == NULL);
commit_transaction = journal->j_running_transaction;
J_ASSERT(commit_transaction->t_state == T_RUNNING);
jbd_debug(1, "JBD: starting commit of transaction %d\n",
commit_transaction->t_tid);
spin_lock(&journal->j_state_lock);
commit_transaction->t_state = T_LOCKED;
spin_lock(&commit_transaction->t_handle_lock);
while (commit_transaction->t_updates) {
DEFINE_WAIT(wait);
prepare_to_wait(&journal->j_wait_updates, &wait,
TASK_UNINTERRUPTIBLE);
if (commit_transaction->t_updates) {
spin_unlock(&commit_transaction->t_handle_lock);
spin_unlock(&journal->j_state_lock);
schedule();
spin_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 <=
journal->j_max_transaction_buffers);
/*
* First thing we are allowed to do is to discard any remaining
* BJ_Reserved buffers. Note, it is _not_ permissible to assume
* that there are no such buffers: if a large filesystem
* operation like a truncate needs to split itself over multiple
* transactions, then it may try to do a journal_restart() while
* there are still BJ_Reserved buffers outstanding. These must
* be released cleanly from the current transaction.
*
* In this case, the filesystem must still reserve write access
* again before modifying the buffer in the new transaction, but
* 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.
*/
while (commit_transaction->t_reserved_list) {
jh = commit_transaction->t_reserved_list;
JBUFFER_TRACE(jh, "reserved, unused: refile");
/*
* A journal_get_undo_access()+journal_release_buffer() may
* leave undo-committed data.
*/
if (jh->b_committed_data) {
struct buffer_head *bh = jh2bh(jh);
jbd_lock_bh_state(bh);
if (jh->b_committed_data) {
kfree(jh->b_committed_data);
jh->b_committed_data = NULL;
}
jbd_unlock_bh_state(bh);
}
journal_refile_buffer(journal, jh);
}
/*
* Now try to drop any written-back buffers from the journal's
* checkpoint lists. We do this *before* commit because it potentially
* frees some memory
*/
spin_lock(&journal->j_list_lock);
__journal_clean_checkpoint_list(journal);
spin_unlock(&journal->j_list_lock);
jbd_debug (3, "JBD: commit phase 1\n");
/*
* Switch to a new revoke table.
*/
journal_switch_revoke_table(journal);
commit_transaction->t_state = T_FLUSH;
journal->j_committing_transaction = commit_transaction;
journal->j_running_transaction = NULL;
commit_transaction->t_log_start = journal->j_head;
wake_up(&journal->j_wait_transaction_locked);
spin_unlock(&journal->j_state_lock);
jbd_debug (3, "JBD: commit phase 2\n");
/*
* Now start flushing things to disk, in the order they appear
* on the transaction lists. Data blocks go first.
*/
err = 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.
*/
bufs = 0;
/*
* 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();
spin_lock(&journal->j_list_lock);
while (commit_transaction->t_sync_datalist) {
struct buffer_head *bh;
jh = commit_transaction->t_sync_datalist;
commit_transaction->t_sync_datalist = jh->b_tnext;
bh = jh2bh(jh);
if (buffer_locked(bh)) {
BUFFER_TRACE(bh, "locked");
if (!inverted_lock(journal, bh))
goto write_out_data;
__journal_unfile_buffer(jh);
__journal_file_buffer(jh, commit_transaction,
BJ_Locked);
jbd_unlock_bh_state(bh);
if (lock_need_resched(&journal->j_list_lock)) {
spin_unlock(&journal->j_list_lock);
goto write_out_data;
}
} else {
if (buffer_dirty(bh)) {
BUFFER_TRACE(bh, "start journal writeout");
get_bh(bh);
wbuf[bufs++] = bh;
if (bufs == ARRAY_SIZE(wbuf)) {
jbd_debug(2, "submit %d writes\n",
bufs);
spin_unlock(&journal->j_list_lock);
ll_rw_block(WRITE, bufs, wbuf);
journal_brelse_array(wbuf, bufs);
bufs = 0;
goto write_out_data;
}
} else {
BUFFER_TRACE(bh, "writeout complete: unfile");
if (!inverted_lock(journal, bh))
goto write_out_data;
__journal_unfile_buffer(jh);
jbd_unlock_bh_state(bh);
journal_remove_journal_head(bh);
put_bh(bh);
if (lock_need_resched(&journal->j_list_lock)) {
spin_unlock(&journal->j_list_lock);
goto write_out_data;
}
}
}
}
if (bufs) {
spin_unlock(&journal->j_list_lock);
ll_rw_block(WRITE, bufs, wbuf);
journal_brelse_array(wbuf, bufs);
spin_lock(&journal->j_list_lock);
}
/*
* Wait for all previously submitted IO to complete.
*/
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)))
err = -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) {
__journal_unfile_buffer(jh);
jbd_unlock_bh_state(bh);
journal_remove_journal_head(bh);
put_bh(bh);
} else {
jbd_unlock_bh_state(bh);
}
put_bh(bh);
// cond_resched_lock(&journal->j_list_lock);
}
spin_unlock(&journal->j_list_lock);
journal_write_revoke_records(journal, commit_transaction);
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:
*/
commit_transaction->t_state = T_COMMIT;
descriptor = NULL;
bufs = 0;
while (commit_transaction->t_buffers) {
/* Find the next buffer to be journaled... */
jh = commit_transaction->t_buffers;
/* If we're in abort mode, we just un-journal the buffer and
release it for background writing. */
if (is_journal_aborted(journal)) {
JBUFFER_TRACE(jh, "journal is aborting: refile");
journal_refile_buffer(journal, jh);
/* If that was the last one, we need to clean up
* any descriptor buffers which may have been
* already allocated, even if we are now
* aborting. */
if (!commit_transaction->t_buffers)
goto start_journal_io;
continue;
}
/* Make sure we have a descriptor block in which to
record the metadata buffer. */
if (!descriptor) {
struct buffer_head *bh;
J_ASSERT (bufs == 0);
jbd_debug(4, "JBD: get descriptor\n");
descriptor = journal_get_descriptor_buffer(journal);
if (!descriptor) {
__journal_abort_hard(journal);
continue;
}
bh = jh2bh(descriptor);
jbd_debug(4, "JBD: got buffer %llu (%p)\n",
(unsigned long long)bh->b_blocknr, bh->b_data);
header = (journal_header_t *)&bh->b_data[0];
header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
header->h_blocktype = cpu_to_be32(JFS_DESCRIPTOR_BLOCK);
header->h_sequence = cpu_to_be32(commit_transaction->t_tid);
tagp = &bh->b_data[sizeof(journal_header_t)];
space_left = bh->b_size - sizeof(journal_header_t);
first_tag = 1;
set_buffer_jwrite(bh);
set_buffer_dirty(bh);
wbuf[bufs++] = bh;
/* Record it so that we can wait for IO
completion later */
BUFFER_TRACE(bh, "ph3: file as descriptor");
journal_file_buffer(descriptor, commit_transaction,
BJ_LogCtl);
}
/* Where is the buffer to be written? */
err = journal_next_log_block(journal, &blocknr);
/* If the block mapping failed, just abandon the buffer
and repeat this loop: we'll fall into the
refile-on-abort condition above. */
if (err) {
__journal_abort_hard(journal);
continue;
}
/*
* start_this_handle() uses t_outstanding_credits to determine
* the free space in the log, but this counter is changed
* by journal_next_log_block() also.
*/
commit_transaction->t_outstanding_credits--;
/* 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);
/* 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);
/*
* akpm: journal_write_metadata_buffer() sets
* new_bh->b_transaction to commit_transaction.
* We need to clean this up before we release new_bh
* (which is of type BJ_IO)
*/
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);
wbuf[bufs++] = jh2bh(new_jh);
/* Record the new block's tag in the current descriptor
buffer */
tag_flag = 0;
if (flags & 1)
tag_flag |= JFS_FLAG_ESCAPE;
if (!first_tag)
tag_flag |= JFS_FLAG_SAME_UUID;
tag = (journal_block_tag_t *) tagp;
tag->t_blocknr = cpu_to_be32(jh2bh(jh)->b_blocknr);
tag->t_flags = cpu_to_be32(tag_flag);
tagp += sizeof(journal_block_tag_t);
space_left -= sizeof(journal_block_tag_t);
if (first_tag) {
memcpy (tagp, journal->j_uuid, 16);
tagp += 16;
space_left -= 16;
first_tag = 0;
}
/* If there's no more to do, or if the descriptor is full,
let the IO rip! */
if (bufs == ARRAY_SIZE(wbuf) ||
commit_transaction->t_buffers == NULL ||
space_left < sizeof(journal_block_tag_t) + 16) {
jbd_debug(4, "JBD: Submit %d IOs\n", bufs);
/* Write an end-of-descriptor marker before
submitting the IOs. "tag" still points to
the last tag we set up. */
tag->t_flags |= cpu_to_be32(JFS_FLAG_LAST_TAG);
start_journal_io:
for (i = 0; i < bufs; i++) {
struct buffer_head *bh = wbuf[i];
lock_buffer(bh);
clear_buffer_dirty(bh);
set_buffer_uptodate(bh);
bh->b_end_io = journal_end_buffer_io_sync;
submit_bh(WRITE, bh);
}
cond_resched();
/* Force a new descriptor to be generated next
time round the loop. */
descriptor = NULL;
bufs = 0;
}
}
/* 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
transaction's t_log_list queue, and metadata buffers are on
the t_iobuf_list queue.
Wait for the buffers in reverse order. That way we are
less likely to be woken up until all IOs have completed, and
so we incur less scheduling load.
*/
jbd_debug(3, "JBD: commit phase 4\n");
/*
* akpm: these are BJ_IO, and j_list_lock is not needed.
* See __journal_try_to_free_buffer.
*/
wait_for_iobuf:
while (commit_transaction->t_iobuf_list != NULL) {
struct buffer_head *bh;
jh = commit_transaction->t_iobuf_list->b_tprev;
bh = jh2bh(jh);
if (buffer_locked(bh)) {
wait_on_buffer(bh);
goto wait_for_iobuf;
}
if (cond_resched())
goto wait_for_iobuf;
if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
clear_buffer_jwrite(bh);
JBUFFER_TRACE(jh, "ph4: unfile after journal write");
journal_unfile_buffer(journal, jh);
/*
* ->t_iobuf_list should contain only dummy buffer_heads
* which were created by journal_write_metadata_buffer().
*/
BUFFER_TRACE(bh, "dumping temporary bh");
journal_put_journal_head(jh);
__brelse(bh);
J_ASSERT_BH(bh, atomic_read(&bh->b_count) == 0);
free_buffer_head(bh);
/* We also have to unlock and free the corresponding
shadowed buffer */
jh = commit_transaction->t_shadow_list->b_tprev;
bh = jh2bh(jh);
clear_bit(BH_JWrite, &bh->b_state);
J_ASSERT_BH(bh, buffer_jbddirty(bh));
/* The metadata is now released for reuse, but we need
to remember it against this transaction so that when
we finally commit, we can do any checkpointing
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_bit(&bh->b_state, BH_Unshadow);
JBUFFER_TRACE(jh, "brelse shadowed buffer");
__brelse(bh);
}
J_ASSERT (commit_transaction->t_shadow_list == NULL);
jbd_debug(3, "JBD: commit phase 5\n");
/* Here we wait for the revoke record and descriptor record buffers */
wait_for_ctlbuf:
while (commit_transaction->t_log_list != NULL) {
struct buffer_head *bh;
jh = commit_transaction->t_log_list->b_tprev;
bh = jh2bh(jh);
if (buffer_locked(bh)) {
wait_on_buffer(bh);
goto wait_for_ctlbuf;
}
if (cond_resched())
goto wait_for_ctlbuf;
if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
BUFFER_TRACE(bh, "ph5: control buffer writeout done: unfile");
clear_buffer_jwrite(bh);
journal_unfile_buffer(journal, jh);
journal_put_journal_head(jh);
__brelse(bh); /* One for getblk */
/* AKPM: bforget here */
}
jbd_debug(3, "JBD: commit phase 6\n");
if (is_journal_aborted(journal))
goto skip_commit;
/* Done it all: now write 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
* entirely. */
descriptor = journal_get_descriptor_buffer(journal);
if (!descriptor) {
__journal_abort_hard(journal);
goto skip_commit;
}
/* AKPM: buglet - add `i' to tmp! */
for (i = 0; i < jh2bh(descriptor)->b_size; i += 512) {
journal_header_t *tmp =
(journal_header_t*)jh2bh(descriptor)->b_data;
tmp->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER);
tmp->h_blocktype = cpu_to_be32(JFS_COMMIT_BLOCK);
tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
}
JBUFFER_TRACE(descriptor, "write commit block");
{
struct buffer_head *bh = jh2bh(descriptor);
int ret;
int barrier_done = 0;
set_buffer_dirty(bh);
if (journal->j_flags & JFS_BARRIER) {
set_buffer_ordered(bh);
barrier_done = 1;
}
ret = sync_dirty_buffer(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 */
clear_buffer_ordered(bh);
set_buffer_uptodate(bh);
set_buffer_dirty(bh);
ret = sync_dirty_buffer(bh);
}
if (unlikely(ret == -EIO))
err = -EIO;
put_bh(bh); /* One for getblk() */
journal_put_journal_head(descriptor);
}
/* End of a transaction! Finally, we can do checkpoint
processing: any buffers committed as a result of this
transaction can be removed from any checkpoint list it was on
before. */
skip_commit: /* The journal should be unlocked by now. */
if (err)
__journal_abort_hard(journal);
jbd_debug(3, "JBD: commit phase 7\n");
J_ASSERT(commit_transaction->t_sync_datalist == NULL);
J_ASSERT(commit_transaction->t_buffers == NULL);
J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
J_ASSERT(commit_transaction->t_iobuf_list == NULL);
J_ASSERT(commit_transaction->t_shadow_list == NULL);
J_ASSERT(commit_transaction->t_log_list == NULL);
restart_loop:
while (commit_transaction->t_forget) {
transaction_t *cp_transaction;
struct buffer_head *bh;
jh = commit_transaction->t_forget;
bh = jh2bh(jh);
jbd_lock_bh_state(bh);
J_ASSERT_JH(jh, jh->b_transaction == commit_transaction ||
jh->b_transaction == journal->j_running_transaction);
/*
* If there is undo-protected committed data against
* this buffer, then we can remove it now. If it is a
* buffer needing such protection, the old frozen_data
* field now points to a committed version of the
* buffer, so rotate that field to the new committed
* data.
*
* Otherwise, we can just throw away the frozen data now.
*/
if (jh->b_committed_data) {
kfree(jh->b_committed_data);
jh->b_committed_data = NULL;
if (jh->b_frozen_data) {
jh->b_committed_data = jh->b_frozen_data;
jh->b_frozen_data = NULL;
}
} else if (jh->b_frozen_data) {
kfree(jh->b_frozen_data);
jh->b_frozen_data = NULL;
}
spin_lock(&journal->j_list_lock);
cp_transaction = jh->b_cp_transaction;
if (cp_transaction) {
JBUFFER_TRACE(jh, "remove from old cp transaction");
__journal_remove_checkpoint(jh);
}
/* Only re-checkpoint the buffer_head if it is marked
* dirty. If the buffer was added to the BJ_Forget list
* by journal_forget, it may no longer be dirty and
* 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 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
* use in a different page. */
if (buffer_freed(bh)) {
clear_buffer_freed(bh);
clear_buffer_jbddirty(bh);
}
if (buffer_jbddirty(bh)) {
JBUFFER_TRACE(jh, "add to new checkpointing trans");
__journal_insert_checkpoint(jh, commit_transaction);
JBUFFER_TRACE(jh, "refile for checkpoint writeback");
__journal_refile_buffer(jh);
jbd_unlock_bh_state(bh);
} else {
J_ASSERT_BH(bh, !buffer_dirty(bh));
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
__journal_unfile_buffer(jh);
jbd_unlock_bh_state(bh);
journal_remove_journal_head(bh); /* needs a brelse */
release_buffer_page(bh);
}
spin_unlock(&journal->j_list_lock);
if (cond_resched())
goto restart_loop;
}
/* Done with this transaction! */
jbd_debug(3, "JBD: commit phase 8\n");
J_ASSERT(commit_transaction->t_state == T_COMMIT);
/*
* This is a bit sleazy. We borrow j_list_lock to protect
* journal->j_committing_transaction in __journal_remove_checkpoint.
* Really, __jornal_remove_checkpoint should be using j_state_lock but
* it's a bit hassle to hold that across __journal_remove_checkpoint
*/
spin_lock(&journal->j_state_lock);
spin_lock(&journal->j_list_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);
if (commit_transaction->t_checkpoint_list == NULL) {
__journal_drop_transaction(journal, commit_transaction);
} else {
if (journal->j_checkpoint_transactions == NULL) {
journal->j_checkpoint_transactions = commit_transaction;
commit_transaction->t_cpnext = commit_transaction;
commit_transaction->t_cpprev = commit_transaction;
} else {
commit_transaction->t_cpnext =
journal->j_checkpoint_transactions;
commit_transaction->t_cpprev =
commit_transaction->t_cpnext->t_cpprev;
commit_transaction->t_cpnext->t_cpprev =
commit_transaction;
commit_transaction->t_cpprev->t_cpnext =
commit_transaction;
}
}
spin_unlock(&journal->j_list_lock);
jbd_debug(1, "JBD: commit %d complete, head %d\n",
journal->j_commit_sequence, journal->j_tail_sequence);
wake_up(&journal->j_wait_done_commit);
}
