#include <linux/timer.h>
#include <linux/mm.h>
#include <linux/highmem.h>
+#include <linux/hrtimer.h>
static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
{
transaction->t_journal = journal;
transaction->t_state = T_RUNNING;
+ transaction->t_start_time = ktime_get();
transaction->t_tid = journal->j_transaction_sequence++;
transaction->t_expires = jiffies + journal->j_commit_interval;
spin_lock_init(&transaction->t_handle_lock);
INIT_LIST_HEAD(&transaction->t_inode_list);
+ INIT_LIST_HEAD(&transaction->t_private_list);
/* Set up the commit timer for the new transaction. */
- journal->j_commit_timer.expires = round_jiffies(transaction->t_expires);
+ journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
add_timer(&journal->j_commit_timer);
J_ASSERT(journal->j_running_transaction == NULL);
__jbd2_log_space_left(journal));
spin_unlock(&transaction->t_handle_lock);
spin_unlock(&journal->j_state_lock);
+
+ lock_map_acquire(&handle->h_lockdep_map);
out:
if (unlikely(new_transaction)) /* It's usually NULL */
kfree(new_transaction);
handle = ERR_PTR(err);
goto out;
}
-
- lock_acquire(&handle->h_lockdep_map, 0, 0, 0, 2, _THIS_IP_);
out:
return handle;
}
__jbd2_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;
wake_up(&journal->j_wait_transaction_locked);
}
-/*
- * Report any unexpected dirty buffers which turn up. Normally those
- * indicate an error, but they can occur if the user is running (say)
- * tune2fs to modify the live filesystem, so we need the option of
- * continuing as gracefully as possible. #
- *
- * The caller should already hold the journal lock and
- * j_list_lock spinlock: most callers will need those anyway
- * in order to probe the buffer's journaling state safely.
- */
-static void jbd_unexpected_dirty_buffer(struct journal_head *jh)
+static void warn_dirty_buffer(struct buffer_head *bh)
{
- int jlist;
-
- /* If this buffer is one which might reasonably be dirty
- * --- ie. data, or not part of this journal --- then
- * we're OK to leave it alone, but otherwise we need to
- * move the dirty bit to the journal's own internal
- * JBDDirty bit. */
- jlist = jh->b_jlist;
-
- if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
- jlist == BJ_Shadow || jlist == BJ_Forget) {
- struct buffer_head *bh = jh2bh(jh);
+ char b[BDEVNAME_SIZE];
- if (test_clear_buffer_dirty(bh))
- set_buffer_jbddirty(bh);
- }
+ printk(KERN_WARNING
+ "JBD: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
+ "There's a risk of filesystem corruption in case of system "
+ "crash.\n",
+ bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
}
/*
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, "Unexpected dirty buffer");
- jbd_unexpected_dirty_buffer(jh);
+ JBUFFER_TRACE(jh, "Journalling dirty buffer");
+ clear_buffer_dirty(bh);
+ set_buffer_jbddirty(bh);
}
unlock_buffer(bh);
source = kmap_atomic(page, KM_USER0);
memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
kunmap_atomic(source, KM_USER0);
+
+ /*
+ * Now that the frozen data is saved off, we need to store
+ * any matching triggers.
+ */
+ jh->b_frozen_triggers = jh->b_triggers;
}
jbd_unlock_bh_state(bh);
J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
if (jh->b_transaction == NULL) {
+ /*
+ * Previous jbd2_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 */
}
/**
- * int jbd2_journal_dirty_data() - mark a buffer as containing dirty data which
- * needs to be flushed before we can commit the
- * current transaction.
- * @handle: transaction
- * @bh: bufferhead to mark
- *
- * The buffer is placed on the transaction's data list and is marked as
- * belonging to the transaction.
+ * void jbd2_journal_set_triggers() - Add triggers for commit writeout
+ * @bh: buffer to trigger on
+ * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
*
- * Returns error number or 0 on success.
+ * Set any triggers on this journal_head. This is always safe, because
+ * triggers for a committing buffer will be saved off, and triggers for
+ * a running transaction will match the buffer in that transaction.
*
- * jbd2_journal_dirty_data() can be called via page_launder->ext3_writepage
- * by kswapd.
+ * Call with NULL to clear the triggers.
*/
-int jbd2_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
+void jbd2_journal_set_triggers(struct buffer_head *bh,
+ struct jbd2_buffer_trigger_type *type)
{
- journal_t *journal = handle->h_transaction->t_journal;
- int need_brelse = 0;
- struct journal_head *jh;
-
- if (is_handle_aborted(handle))
- return 0;
-
- jh = jbd2_journal_add_journal_head(bh);
- JBUFFER_TRACE(jh, "entry");
+ struct journal_head *jh = bh2jh(bh);
- /*
- * 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);
+ jh->b_triggers = type;
+}
- /*
- * 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);
+void jbd2_buffer_commit_trigger(struct journal_head *jh, void *mapped_data,
+ struct jbd2_buffer_trigger_type *triggers)
+{
+ struct buffer_head *bh = jh2bh(jh);
- /* 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 (!triggers || !triggers->t_commit)
+ return;
- 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);
+ triggers->t_commit(triggers, bh, mapped_data, bh->b_size);
+}
- /* @@@ 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 jbd2_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);
+void jbd2_buffer_abort_trigger(struct journal_head *jh,
+ struct jbd2_buffer_trigger_type *triggers)
+{
+ if (!triggers || !triggers->t_abort)
+ return;
- /*
- * 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;
- }
+ triggers->t_abort(triggers, jh2bh(jh));
+}
- /*
- * 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 */
- }
- /* journal_clean_data_list() may have got there first */
- if (jh->b_transaction != NULL) {
- JBUFFER_TRACE(jh, "unfile from commit");
- __jbd2_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);
- __jbd2_journal_temp_unlink_buffer(jh);
- jh->b_transaction = handle->h_transaction;
- JBUFFER_TRACE(jh, "file as data");
- __jbd2_journal_file_buffer(jh, handle->h_transaction,
- BJ_SyncData);
- }
- } else {
- JBUFFER_TRACE(jh, "not on a transaction");
- __jbd2_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");
- jbd2_journal_put_journal_head(jh);
- return 0;
-}
/**
* int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
{
transaction_t *transaction = handle->h_transaction;
journal_t *journal = transaction->t_journal;
- int old_handle_count, err;
+ int err;
pid_t pid;
J_ASSERT(journal_current_handle() == 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...
+ * 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...
*
- * 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.
+ * We try and optimize the sleep time against what the
+ * underlying disk can do, instead of having a static sleep
+ * time. This is useful 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;
- do {
- old_handle_count = transaction->t_handle_count;
- schedule_timeout_uninterruptible(1);
- } while (old_handle_count != transaction->t_handle_count);
+
+ 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 = 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;
spin_lock(&journal->j_state_lock);
spin_lock(&transaction->t_handle_lock);
spin_unlock(&journal->j_state_lock);
}
- lock_release(&handle->h_lockdep_map, 1, _THIS_IP_);
+ lock_map_release(&handle->h_lockdep_map);
jbd2_free_handle(handle);
return err;
* Remove a buffer from the appropriate transaction list.
*
* Note that this function can *change* the value of
- * bh->b_transaction->t_sync_datalist, t_buffers, t_forget,
- * t_iobuf_list, t_shadow_list, t_log_list or t_reserved_list. If the caller
- * is holding onto a copy of one of thee pointers, it could go bad.
- * Generally the caller needs to re-read the pointer from the transaction_t.
+ * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list,
+ * t_log_list or t_reserved_list. If the caller is holding onto a copy of one
+ * of these pointers, it could go bad. Generally the caller needs to re-read
+ * the pointer from the transaction_t.
*
* Called under j_list_lock. The journal may not be locked.
*/
switch (jh->b_jlist) {
case BJ_None:
return;
- case BJ_SyncData:
- list = &transaction->t_sync_datalist;
- break;
case BJ_Metadata:
transaction->t_nr_buffers--;
J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
case BJ_Reserved:
list = &transaction->t_reserved_list;
break;
- case BJ_Locked:
- list = &transaction->t_locked_list;
- break;
}
__blist_del_buffer(list, jh);
goto out;
spin_lock(&journal->j_list_lock);
- if (jh->b_transaction != NULL && jh->b_cp_transaction == NULL) {
- if (jh->b_jlist == BJ_SyncData || jh->b_jlist == BJ_Locked) {
- /* A written-back ordered data buffer */
- JBUFFER_TRACE(jh, "release data");
- __jbd2_journal_unfile_buffer(jh);
- jbd2_journal_remove_journal_head(bh);
- __brelse(bh);
- }
- } else if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
+ if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
/* written-back checkpointed metadata buffer */
if (jh->b_jlist == BJ_None) {
JBUFFER_TRACE(jh, "remove from checkpoint list");
return;
}
-/*
- * jbd2_journal_try_to_free_buffers() could race with
- * jbd2_journal_commit_transaction(). The later might still hold the
- * reference count to the buffers when inspecting them on
- * t_syncdata_list or t_locked_list.
- *
- * jbd2_journal_try_to_free_buffers() will call this function to
- * wait for the current transaction to finish syncing data buffers, before
- * try to free that buffer.
- *
- * Called with journal->j_state_lock hold.
- */
-static void jbd2_journal_wait_for_transaction_sync_data(journal_t *journal)
-{
- transaction_t *transaction;
- tid_t tid;
-
- spin_lock(&journal->j_state_lock);
- transaction = journal->j_committing_transaction;
-
- if (!transaction) {
- spin_unlock(&journal->j_state_lock);
- return;
- }
-
- tid = transaction->t_tid;
- spin_unlock(&journal->j_state_lock);
- jbd2_log_wait_commit(journal, tid);
-}
-
/**
* int jbd2_journal_try_to_free_buffers() - try to free page buffers.
* @journal: journal for operation
ret = try_to_free_buffers(page);
- /*
- * There are a number of places where jbd2_journal_try_to_free_buffers()
- * could race with jbd2_journal_commit_transaction(), the later still
- * holds the reference to the buffers to free while processing them.
- * try_to_free_buffers() failed to free those buffers. Some of the
- * caller of releasepage() request page buffers to be dropped, otherwise
- * treat the fail-to-free as errors (such as generic_file_direct_IO())
- *
- * So, if the caller of try_to_release_page() wants the synchronous
- * behaviour(i.e make sure buffers are dropped upon return),
- * let's wait for the current transaction to finish flush of
- * dirty data buffers, then try to free those buffers again,
- * with the journal locked.
- */
- if (ret == 0 && (gfp_mask & __GFP_WAIT) && (gfp_mask & __GFP_FS)) {
- jbd2_journal_wait_for_transaction_sync_data(journal);
- ret = try_to_free_buffers(page);
- }
-
busy:
return ret;
}
if (jh->b_cp_transaction) {
JBUFFER_TRACE(jh, "on running+cp transaction");
+ /*
+ * We don't want to write the buffer anymore, clear the
+ * bit so that we don't confuse checks in
+ * __journal_file_buffer
+ */
+ clear_buffer_dirty(bh);
__jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
- clear_buffer_jbddirty(bh);
may_free = 0;
} else {
JBUFFER_TRACE(jh, "on running transaction");
if (!buffer_jbd(bh))
goto zap_buffer_unlocked;
+ /* OK, we have data buffer in journaled mode */
spin_lock(&journal->j_state_lock);
jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
}
} 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;
- }
/*
* If it is committing, we simply cannot touch it. We
* can remove it's next_transaction pointer from the
if (jh->b_transaction && jh->b_jlist == jlist)
return;
- /* The following list of buffer states needs to be consistent
- * with __jbd_unexpected_dirty_buffer()'s handling of dirty
- * state. */
-
if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
jlist == BJ_Shadow || jlist == BJ_Forget) {
+ /*
+ * For metadata buffers, we track dirty bit in buffer_jbddirty
+ * instead of buffer_dirty. We should not see a dirty bit set
+ * here because we clear it in do_get_write_access but e.g.
+ * tune2fs can modify the sb and set the dirty bit at any time
+ * so we try to gracefully handle that.
+ */
+ if (buffer_dirty(bh))
+ warn_dirty_buffer(bh);
if (test_clear_buffer_dirty(bh) ||
test_clear_buffer_jbddirty(bh))
was_dirty = 1;
J_ASSERT_JH(jh, !jh->b_committed_data);
J_ASSERT_JH(jh, !jh->b_frozen_data);
return;
- case BJ_SyncData:
- list = &transaction->t_sync_datalist;
- break;
case BJ_Metadata:
transaction->t_nr_buffers++;
list = &transaction->t_buffers;
case BJ_Reserved:
list = &transaction->t_reserved_list;
break;
- case BJ_Locked:
- list = &transaction->t_locked_list;
- break;
}
__blist_add_buffer(list, jh);
}
/*
- * This function must be called when inode is journaled in ordered mode
- * before truncation happens. It starts writeout of truncated part in
- * case it is in the committing transaction so that we stand to ordered
- * mode consistency guarantees.
+ * File truncate and transaction commit interact with each other in a
+ * non-trivial way. If a transaction writing data block A is
+ * committing, we cannot discard the data by truncate until we have
+ * written them. Otherwise if we crashed after the transaction with
+ * write has committed but before the transaction with truncate has
+ * committed, we could see stale data in block A. This function is a
+ * helper to solve this problem. It starts writeout of the truncated
+ * part in case it is in the committing transaction.
+ *
+ * Filesystem code must call this function when inode is journaled in
+ * ordered mode before truncation happens and after the inode has been
+ * placed on orphan list with the new inode size. The second condition
+ * avoids the race that someone writes new data and we start
+ * committing the transaction after this function has been called but
+ * before a transaction for truncate is started (and furthermore it
+ * allows us to optimize the case where the addition to orphan list
+ * happens in the same transaction as write --- we don't have to write
+ * any data in such case).
*/
-int jbd2_journal_begin_ordered_truncate(struct jbd2_inode *inode,
+int jbd2_journal_begin_ordered_truncate(journal_t *journal,
+ struct jbd2_inode *jinode,
loff_t new_size)
{
- journal_t *journal;
- transaction_t *commit_trans;
+ transaction_t *inode_trans, *commit_trans;
int ret = 0;
- if (!inode->i_transaction && !inode->i_next_transaction)
+ /* This is a quick check to avoid locking if not necessary */
+ if (!jinode->i_transaction)
goto out;
- journal = inode->i_transaction->t_journal;
+ /* Locks are here just to force reading of recent values, it is
+ * enough that the transaction was not committing before we started
+ * a transaction adding the inode to orphan list */
spin_lock(&journal->j_state_lock);
commit_trans = journal->j_committing_transaction;
spin_unlock(&journal->j_state_lock);
- if (inode->i_transaction == commit_trans) {
- ret = filemap_fdatawrite_range(inode->i_vfs_inode->i_mapping,
+ spin_lock(&journal->j_list_lock);
+ inode_trans = jinode->i_transaction;
+ spin_unlock(&journal->j_list_lock);
+ if (inode_trans == commit_trans) {
+ ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
new_size, LLONG_MAX);
if (ret)
jbd2_journal_abort(journal, ret);