aboutsummaryrefslogtreecommitdiff
path: root/fs/jbd/transaction.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/jbd/transaction.c')
-rw-r--r--fs/jbd/transaction.c214
1 files changed, 127 insertions, 87 deletions
diff --git a/fs/jbd/transaction.c b/fs/jbd/transaction.c
index 846a3f31411..1695ba8334a 100644
--- a/fs/jbd/transaction.c
+++ b/fs/jbd/transaction.c
@@ -207,7 +207,7 @@ repeat_locked:
* the committing transaction. Really, we only need to give it
* committing_transaction->t_outstanding_credits plus "enough" for
* the log control blocks.
- * Also, this test is inconsitent with the matching one in
+ * Also, this test is inconsistent with the matching one in
* journal_extend().
*/
if (__log_space_left(journal) < jbd_space_needed(journal)) {
@@ -245,7 +245,6 @@ static handle_t *new_handle(int nblocks)
handle_t *handle = jbd_alloc_handle(GFP_NOFS);
if (!handle)
return NULL;
- memset(handle, 0, sizeof(*handle));
handle->h_buffer_credits = nblocks;
handle->h_ref = 1;
@@ -266,7 +265,8 @@ static handle_t *new_handle(int nblocks)
* This function is visible to journal users (like ext3fs), so is not
* called with the journal already locked.
*
- * Return a pointer to a newly allocated handle, or NULL on failure
+ * Return a pointer to a newly allocated handle, or an ERR_PTR() value
+ * on failure.
*/
handle_t *journal_start(journal_t *journal, int nblocks)
{
@@ -425,17 +425,34 @@ int journal_restart(handle_t *handle, int nblocks)
* void journal_lock_updates () - establish a transaction barrier.
* @journal: Journal to establish a barrier on.
*
- * This locks out any further updates from being started, and blocks
- * until all existing updates have completed, returning only once the
- * journal is in a quiescent state with no updates running.
+ * This locks out any further updates from being started, and blocks until all
+ * existing updates have completed, returning only once the journal is in a
+ * quiescent state with no updates running.
*
- * The journal lock should not be held on entry.
+ * We do not use simple mutex for synchronization as there are syscalls which
+ * want to return with filesystem locked and that trips up lockdep. Also
+ * hibernate needs to lock filesystem but locked mutex then blocks hibernation.
+ * Since locking filesystem is rare operation, we use simple counter and
+ * waitqueue for locking.
*/
void journal_lock_updates(journal_t *journal)
{
DEFINE_WAIT(wait);
+wait:
+ /* Wait for previous locked operation to finish */
+ wait_event(journal->j_wait_transaction_locked,
+ journal->j_barrier_count == 0);
+
spin_lock(&journal->j_state_lock);
+ /*
+ * Check reliably under the lock whether we are the ones winning the race
+ * and locking the journal
+ */
+ if (journal->j_barrier_count > 0) {
+ spin_unlock(&journal->j_state_lock);
+ goto wait;
+ }
++journal->j_barrier_count;
/* Wait until there are no running updates */
@@ -459,14 +476,6 @@ void journal_lock_updates(journal_t *journal)
spin_lock(&journal->j_state_lock);
}
spin_unlock(&journal->j_state_lock);
-
- /*
- * We have now established a barrier against other normal updates, but
- * we also need to barrier against other journal_lock_updates() calls
- * to make sure that we serialise special journal-locked operations
- * too.
- */
- mutex_lock(&journal->j_barrier);
}
/**
@@ -474,14 +483,11 @@ void journal_lock_updates(journal_t *journal)
* @journal: Journal to release the barrier on.
*
* Release a transaction barrier obtained with journal_lock_updates().
- *
- * Should be called without the journal lock held.
*/
void journal_unlock_updates (journal_t *journal)
{
J_ASSERT(journal->j_barrier_count != 0);
- mutex_unlock(&journal->j_barrier);
spin_lock(&journal->j_state_lock);
--journal->j_barrier_count;
spin_unlock(&journal->j_state_lock);
@@ -669,7 +675,7 @@ repeat:
jbd_alloc(jh2bh(jh)->b_size,
GFP_NOFS);
if (!frozen_buffer) {
- printk(KERN_EMERG
+ printk(KERN_ERR
"%s: OOM for frozen_buffer\n",
__func__);
JBUFFER_TRACE(jh, "oom!");
@@ -695,7 +701,6 @@ repeat:
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);
@@ -712,9 +717,9 @@ done:
"Possible IO failure.\n");
page = jh2bh(jh)->b_page;
offset = offset_in_page(jh2bh(jh)->b_data);
- source = kmap_atomic(page, KM_USER0);
+ source = kmap_atomic(page);
memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
- kunmap_atomic(source, KM_USER0);
+ kunmap_atomic(source);
}
jbd_unlock_bh_state(bh);
@@ -817,7 +822,6 @@ int journal_get_create_access(handle_t *handle, struct buffer_head *bh)
* 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;
@@ -843,8 +847,8 @@ int journal_get_create_access(handle_t *handle, struct buffer_head *bh)
*/
JBUFFER_TRACE(jh, "cancelling revoke");
journal_cancel_revoke(handle, jh);
- journal_put_journal_head(jh);
out:
+ journal_put_journal_head(jh);
return err;
}
@@ -894,7 +898,7 @@ 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",
+ printk(KERN_ERR "%s: No memory for committed data\n",
__func__);
err = -ENOMEM;
goto out;
@@ -1068,8 +1072,9 @@ int journal_dirty_data(handle_t *handle, struct buffer_head *bh)
ret = -EIO;
goto no_journal;
}
-
- if (jh->b_transaction != NULL) {
+ /* We might have slept so buffer could be refiled now */
+ if (jh->b_transaction != NULL &&
+ jh->b_transaction != handle->h_transaction) {
JBUFFER_TRACE(jh, "unfile from commit");
__journal_temp_unlink_buffer(jh);
/* It still points to the committing
@@ -1090,8 +1095,6 @@ int journal_dirty_data(handle_t *handle, struct buffer_head *bh)
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);
@@ -1255,7 +1258,7 @@ int journal_forget (handle_t *handle, struct buffer_head *bh)
goto not_jbd;
}
- /* keep track of wether or not this transaction modified us */
+ /* keep track of whether or not this transaction modified us */
was_modified = jh->b_modified;
/*
@@ -1299,8 +1302,6 @@ int journal_forget (handle_t *handle, struct buffer_head *bh)
__journal_file_buffer(jh, transaction, BJ_Forget);
} else {
__journal_unfile_buffer(jh);
- journal_remove_journal_head(bh);
- __brelse(bh);
if (!buffer_jbd(bh)) {
spin_unlock(&journal->j_list_lock);
jbd_unlock_bh_state(bh);
@@ -1392,7 +1393,7 @@ int journal_stop(handle_t *handle)
* 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
+ * 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
@@ -1431,8 +1432,6 @@ int journal_stop(handle_t *handle)
}
}
- 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);
@@ -1621,19 +1620,32 @@ static void __journal_temp_unlink_buffer(struct journal_head *jh)
mark_buffer_dirty(bh); /* Expose it to the VM */
}
+/*
+ * Remove buffer from all transactions.
+ *
+ * Called with bh_state lock and j_list_lock
+ *
+ * jh and bh may be already freed when this function returns.
+ */
void __journal_unfile_buffer(struct journal_head *jh)
{
__journal_temp_unlink_buffer(jh);
jh->b_transaction = NULL;
+ journal_put_journal_head(jh);
}
void journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
{
- jbd_lock_bh_state(jh2bh(jh));
+ struct buffer_head *bh = jh2bh(jh);
+
+ /* Get reference so that buffer cannot be freed before we unlock it */
+ get_bh(bh);
+ jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
__journal_unfile_buffer(jh);
spin_unlock(&journal->j_list_lock);
- jbd_unlock_bh_state(jh2bh(jh));
+ jbd_unlock_bh_state(bh);
+ __brelse(bh);
}
/*
@@ -1660,16 +1672,12 @@ __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
/* A written-back ordered data buffer */
JBUFFER_TRACE(jh, "release data");
__journal_unfile_buffer(jh);
- journal_remove_journal_head(bh);
- __brelse(bh);
}
} else 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");
__journal_remove_checkpoint(jh);
- journal_remove_journal_head(bh);
- __brelse(bh);
}
}
spin_unlock(&journal->j_list_lock);
@@ -1732,7 +1740,7 @@ int journal_try_to_free_buffers(journal_t *journal,
/*
* 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().
+ * journal_put_journal_head().
*/
jh = journal_grab_journal_head(bh);
if (!jh)
@@ -1769,10 +1777,9 @@ static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
int may_free = 1;
struct buffer_head *bh = jh2bh(jh);
- __journal_unfile_buffer(jh);
-
if (jh->b_cp_transaction) {
JBUFFER_TRACE(jh, "on running+cp transaction");
+ __journal_temp_unlink_buffer(jh);
/*
* We don't want to write the buffer anymore, clear the
* bit so that we don't confuse checks in
@@ -1783,8 +1790,7 @@ static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
may_free = 0;
} else {
JBUFFER_TRACE(jh, "on running transaction");
- journal_remove_journal_head(bh);
- __brelse(bh);
+ __journal_unfile_buffer(jh);
}
return may_free;
}
@@ -1836,15 +1842,16 @@ static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
* We're outside-transaction here. Either or both of j_running_transaction
* and j_committing_transaction may be NULL.
*/
-static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
+static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
+ int partial_page)
{
transaction_t *transaction;
struct journal_head *jh;
int may_free = 1;
- int ret;
BUFFER_TRACE(bh, "entry");
+retry:
/*
* 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
@@ -1872,10 +1879,18 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
* 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.
+ * block 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.
+ *
+ * Also we have to clear buffer_mapped flag of a truncated buffer
+ * because the buffer_head may be attached to the page straddling
+ * i_size (can happen only when blocksize < pagesize) and thus the
+ * buffer_head can be reused when the file is extended again. So we end
+ * up keeping around invalidated buffers attached to transactions'
+ * BJ_Forget list just to stop checkpointing code from cleaning up
+ * the transaction this buffer was modified in.
*/
transaction = jh->b_transaction;
if (transaction == NULL) {
@@ -1902,13 +1917,9 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
* committed, the buffer won't be needed any
* longer. */
JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
- ret = __dispose_buffer(jh,
+ may_free = __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;
+ goto zap_buffer;
} else {
/* There is no currently-running transaction. So the
* orphan record which we wrote for this file must have
@@ -1916,13 +1927,9 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
* the committing transaction, if it exists. */
if (journal->j_committing_transaction) {
JBUFFER_TRACE(jh, "give to committing trans");
- ret = __dispose_buffer(jh,
+ may_free = __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;
+ goto zap_buffer;
} else {
/* The orphan record's transaction has
* committed. We can cleanse this buffer */
@@ -1943,10 +1950,26 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
}
/*
* 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.
+ * it. If the page is straddling i_size we have to wait
+ * for commit and try again.
+ */
+ if (partial_page) {
+ tid_t tid = journal->j_committing_transaction->t_tid;
+
+ journal_put_journal_head(jh);
+ spin_unlock(&journal->j_list_lock);
+ jbd_unlock_bh_state(bh);
+ spin_unlock(&journal->j_state_lock);
+ unlock_buffer(bh);
+ log_wait_commit(journal, tid);
+ lock_buffer(bh);
+ goto retry;
+ }
+ /*
+ * OK, buffer won't be reachable after truncate. 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))
@@ -1969,6 +1992,14 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
}
zap_buffer:
+ /*
+ * This is tricky. Although the buffer is truncated, it may be reused
+ * if blocksize < pagesize and it is attached to the page straddling
+ * EOF. Since the buffer might have been added to BJ_Forget list of the
+ * running transaction, journal_get_write_access() won't clear
+ * b_modified and credit accounting gets confused. So clear b_modified
+ * here. */
+ jh->b_modified = 0;
journal_put_journal_head(jh);
zap_buffer_no_jh:
spin_unlock(&journal->j_list_lock);
@@ -1988,16 +2019,20 @@ zap_buffer_unlocked:
* void journal_invalidatepage() - invalidate a journal page
* @journal: journal to use for flush
* @page: page to flush
- * @offset: length of page to invalidate.
+ * @offset: offset of the range to invalidate
+ * @length: length of the range to invalidate
*
- * Reap page buffers containing data after offset in page.
+ * Reap page buffers containing data in specified range in page.
*/
void journal_invalidatepage(journal_t *journal,
struct page *page,
- unsigned long offset)
+ unsigned int offset,
+ unsigned int length)
{
struct buffer_head *head, *bh, *next;
+ unsigned int stop = offset + length;
unsigned int curr_off = 0;
+ int partial_page = (offset || length < PAGE_CACHE_SIZE);
int may_free = 1;
if (!PageLocked(page))
@@ -2005,6 +2040,8 @@ void journal_invalidatepage(journal_t *journal,
if (!page_has_buffers(page))
return;
+ BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
+
/* We will potentially be playing with lists other than just the
* data lists (especially for journaled data mode), so be
* cautious in our locking. */
@@ -2014,10 +2051,14 @@ void journal_invalidatepage(journal_t *journal,
unsigned int next_off = curr_off + bh->b_size;
next = bh->b_this_page;
+ if (next_off > stop)
+ return;
+
if (offset <= curr_off) {
/* This block is wholly outside the truncation point */
lock_buffer(bh);
- may_free &= journal_unmap_buffer(journal, bh);
+ may_free &= journal_unmap_buffer(journal, bh,
+ partial_page);
unlock_buffer(bh);
}
curr_off = next_off;
@@ -2025,7 +2066,7 @@ void journal_invalidatepage(journal_t *journal,
} while (bh != head);
- if (!offset) {
+ if (!partial_page) {
if (may_free && try_to_free_buffers(page))
J_ASSERT(!page_has_buffers(page));
}
@@ -2069,6 +2110,8 @@ void __journal_file_buffer(struct journal_head *jh,
if (jh->b_transaction)
__journal_temp_unlink_buffer(jh);
+ else
+ journal_grab_journal_head(bh);
jh->b_transaction = transaction;
switch (jlist) {
@@ -2126,9 +2169,10 @@ void journal_file_buffer(struct journal_head *jh,
* already started to be used by a subsequent transaction, refile the
* buffer on that transaction's metadata list.
*
- * Called under journal->j_list_lock
- *
+ * Called under j_list_lock
* Called under jbd_lock_bh_state(jh2bh(jh))
+ *
+ * jh and bh may be already free when this function returns
*/
void __journal_refile_buffer(struct journal_head *jh)
{
@@ -2152,6 +2196,11 @@ void __journal_refile_buffer(struct journal_head *jh)
was_dirty = test_clear_buffer_jbddirty(bh);
__journal_temp_unlink_buffer(jh);
+ /*
+ * We set b_transaction here because b_next_transaction will inherit
+ * our jh reference and thus __journal_file_buffer() must not take a
+ * new one.
+ */
jh->b_transaction = jh->b_next_transaction;
jh->b_next_transaction = NULL;
if (buffer_freed(bh))
@@ -2168,30 +2217,21 @@ void __journal_refile_buffer(struct journal_head *jh)
}
/*
- * For the unlocked version of this call, also make sure that any
- * hanging journal_head is cleaned up if necessary.
- *
- * __journal_refile_buffer is usually called as part of a single locked
- * operation on a buffer_head, in which the caller is probably going to
- * be hooking the journal_head onto other lists. In that case it is up
- * to the caller to remove the journal_head if necessary. For the
- * unlocked journal_refile_buffer call, the caller isn't going to be
- * doing anything else to the buffer so we need to do the cleanup
- * ourselves to avoid a jh leak.
- *
- * *** The journal_head may be freed by this call! ***
+ * __journal_refile_buffer() with necessary locking added. We take our bh
+ * reference so that we can safely unlock bh.
+ *
+ * The jh and bh may be freed by this call.
*/
void journal_refile_buffer(journal_t *journal, struct journal_head *jh)
{
struct buffer_head *bh = jh2bh(jh);
+ /* Get reference so that buffer cannot be freed before we unlock it */
+ get_bh(bh);
jbd_lock_bh_state(bh);
spin_lock(&journal->j_list_lock);
-
__journal_refile_buffer(jh);
jbd_unlock_bh_state(bh);
- journal_remove_journal_head(bh);
-
spin_unlock(&journal->j_list_lock);
__brelse(bh);
}
generated by cgit v1.2.3-70-g09d2 (git 2.46.0) at 2025-11-05 09:22:04 +0000