jbd: Fix oops in journal_remove_journal_head()

journal_remove_journal_head() can oops when trying to access journal_head
returned by bh2jh(). This is caused for example by the following race:

	TASK1					TASK2
  journal_commit_transaction()
    ...
    processing t_forget list
      __journal_refile_buffer(jh);
      if (!jh->b_transaction) {
        jbd_unlock_bh_state(bh);
					journal_try_to_free_buffers()
					  journal_grab_journal_head(bh)
					  jbd_lock_bh_state(bh)
					  __journal_try_to_free_buffer()
					  journal_put_journal_head(jh)
        journal_remove_journal_head(bh);

journal_put_journal_head() in TASK2 sees that b_jcount == 0 and buffer is not
part of any transaction and thus frees journal_head before TASK1 gets to doing
so. Note that even buffer_head can be released by try_to_free_buffers() after
journal_put_journal_head() which adds even larger opportunity for oops (but I
didn't see this happen in reality).

Fix the problem by making transactions hold their own journal_head reference
(in b_jcount). That way we don't have to remove journal_head explicitely via
journal_remove_journal_head() and instead just remove journal_head when
b_jcount drops to zero. The result of this is that [__]journal_refile_buffer(),
[__]journal_unfile_buffer(), and __journal_remove_checkpoint() can free
journal_head which needs modification of a few callers. Also we have to be
careful because once journal_head is removed, buffer_head might be freed as
well. So we have to get our own buffer_head reference where it matters.

Signed-off-by: Jan Kara <jack@suse.cz>
This commit is contained in:
Jan Kara
2011-06-24 23:11:59 +02:00
parent 2c2ea9451f
commit bb189247f3
5 changed files with 103 additions and 137 deletions

View File

@@ -1803,10 +1803,9 @@ static void journal_free_journal_head(struct journal_head *jh)
* When a buffer has its BH_JBD bit set it is immune from being released by
* core kernel code, mainly via ->b_count.
*
* A journal_head may be detached from its buffer_head when the journal_head's
* b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
* Various places in JBD call journal_remove_journal_head() to indicate that the
* journal_head can be dropped if needed.
* A journal_head is detached from its buffer_head when the journal_head's
* b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
* transaction (b_cp_transaction) hold their references to b_jcount.
*
* Various places in the kernel want to attach a journal_head to a buffer_head
* _before_ attaching the journal_head to a transaction. To protect the
@@ -1819,17 +1818,16 @@ static void journal_free_journal_head(struct journal_head *jh)
* (Attach a journal_head if needed. Increments b_jcount)
* struct journal_head *jh = journal_add_journal_head(bh);
* ...
* jh->b_transaction = xxx;
* journal_put_journal_head(jh);
*
* Now, the journal_head's b_jcount is zero, but it is safe from being released
* because it has a non-zero b_transaction.
* (Get another reference for transaction)
* journal_grab_journal_head(bh);
* jh->b_transaction = xxx;
* (Put original reference)
* journal_put_journal_head(jh);
*/
/*
* Give a buffer_head a journal_head.
*
* Doesn't need the journal lock.
* May sleep.
*/
struct journal_head *journal_add_journal_head(struct buffer_head *bh)
@@ -1893,61 +1891,29 @@ static void __journal_remove_journal_head(struct buffer_head *bh)
struct journal_head *jh = bh2jh(bh);
J_ASSERT_JH(jh, jh->b_jcount >= 0);
get_bh(bh);
if (jh->b_jcount == 0) {
if (jh->b_transaction == NULL &&
jh->b_next_transaction == NULL &&
jh->b_cp_transaction == NULL) {
J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
J_ASSERT_BH(bh, buffer_jbd(bh));
J_ASSERT_BH(bh, jh2bh(jh) == bh);
BUFFER_TRACE(bh, "remove journal_head");
if (jh->b_frozen_data) {
printk(KERN_WARNING "%s: freeing "
"b_frozen_data\n",
__func__);
jbd_free(jh->b_frozen_data, bh->b_size);
}
if (jh->b_committed_data) {
printk(KERN_WARNING "%s: freeing "
"b_committed_data\n",
__func__);
jbd_free(jh->b_committed_data, bh->b_size);
}
bh->b_private = NULL;
jh->b_bh = NULL; /* debug, really */
clear_buffer_jbd(bh);
__brelse(bh);
journal_free_journal_head(jh);
} else {
BUFFER_TRACE(bh, "journal_head was locked");
}
J_ASSERT_JH(jh, jh->b_transaction == NULL);
J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
J_ASSERT_BH(bh, buffer_jbd(bh));
J_ASSERT_BH(bh, jh2bh(jh) == bh);
BUFFER_TRACE(bh, "remove journal_head");
if (jh->b_frozen_data) {
printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
jbd_free(jh->b_frozen_data, bh->b_size);
}
if (jh->b_committed_data) {
printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
jbd_free(jh->b_committed_data, bh->b_size);
}
bh->b_private = NULL;
jh->b_bh = NULL; /* debug, really */
clear_buffer_jbd(bh);
journal_free_journal_head(jh);
}
/*
* journal_remove_journal_head(): if the buffer isn't attached to a transaction
* and has a zero b_jcount then remove and release its journal_head. If we did
* see that the buffer is not used by any transaction we also "logically"
* decrement ->b_count.
*
* We in fact take an additional increment on ->b_count as a convenience,
* because the caller usually wants to do additional things with the bh
* after calling here.
* The caller of journal_remove_journal_head() *must* run __brelse(bh) at some
* time. Once the caller has run __brelse(), the buffer is eligible for
* reaping by try_to_free_buffers().
*/
void journal_remove_journal_head(struct buffer_head *bh)
{
jbd_lock_bh_journal_head(bh);
__journal_remove_journal_head(bh);
jbd_unlock_bh_journal_head(bh);
}
/*
* Drop a reference on the passed journal_head. If it fell to zero then try to
* Drop a reference on the passed journal_head. If it fell to zero then
* release the journal_head from the buffer_head.
*/
void journal_put_journal_head(struct journal_head *jh)
@@ -1957,11 +1923,12 @@ void journal_put_journal_head(struct journal_head *jh)
jbd_lock_bh_journal_head(bh);
J_ASSERT_JH(jh, jh->b_jcount > 0);
--jh->b_jcount;
if (!jh->b_jcount && !jh->b_transaction) {
if (!jh->b_jcount) {
__journal_remove_journal_head(bh);
jbd_unlock_bh_journal_head(bh);
__brelse(bh);
}
jbd_unlock_bh_journal_head(bh);
} else
jbd_unlock_bh_journal_head(bh);
}
/*