Re: [PATCH v2] jbd2: Fix data missing when reusing bh which is ready to be checkpointed

From: Zhihao Cheng
Date: Sat Jan 07 2023 - 04:16:22 EST


在 2023/1/6 22:22, Jan Kara 写道:

Hi Jan, thanks for reviewing.Some discussions below:


diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
index 6a404ac1c178..06a5e7961ef2 100644
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -1010,36 +1010,37 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
* ie. locked but not dirty) or tune2fs (which may actually have
* the buffer dirtied, ugh.) */
- if (buffer_dirty(bh)) {
+ if (buffer_dirty(bh) && jh->b_transaction) {
/*
* First question: is this buffer already part of the current
* transaction or the existing committing transaction?
*/
- if (jh->b_transaction) {
- J_ASSERT_JH(jh,
- jh->b_transaction == transaction ||
- jh->b_transaction ==
- journal->j_committing_transaction);
- if (jh->b_next_transaction)
- J_ASSERT_JH(jh, jh->b_next_transaction ==
- transaction);
- warn_dirty_buffer(bh);
- }
+ J_ASSERT_JH(jh, jh->b_transaction == transaction ||
+ jh->b_transaction == journal->j_committing_transaction);
+ 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.
+ * 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, "Journalling dirty buffer");
clear_buffer_dirty(bh);
+ /*
+ * Setting jbddirty after clearing buffer dirty is necessary.
+ * Function jbd2_journal_restart() could keep buffer on
+ * BJ_Reserved list until the transaction committing, then the
+ * buffer won't be dirtied by jbd2_journal_refile_buffer()
+ * after committing, the buffer couldn't fall on disk even
+ * last checkpoint finished, which may corrupt filesystem.
+ */
set_buffer_jbddirty(bh);
}

So I think the sequence:

if (buffer_dirty(bh)) {
warn_dirty_buffer(bh);
JBUFFER_TRACE(jh, "Journalling dirty buffer");
clear_buffer_dirty(bh);
set_buffer_jbddirty(bh);
}

can be moved into the branch

if (jh->b_transaction == transaction ||
jh->b_next_transaction == transaction) {

below. That way you can drop the assertions as well because they happen
later in do_get_write_access() again anyway.

1. If we move the squence:
if (buffer_dirty(bh)) {
warn_dirty_buffer(bh);
JBUFFER_TRACE(jh, "Journalling dirty buffer");
clear_buffer_dirty(bh);
set_buffer_jbddirty(bh);
}

into the branch

if (jh->b_transaction == transaction ||
jh->b_next_transaction == transaction) {

, then we have a new situation(jh->b_transaction == journal->j_committing_transaction) to clear buffer dirty, so we need to add an else-branch like(based on v2 patch):
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -1092,6 +1092,10 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
spin_unlock(&journal->j_list_lock);
unlock_buffer(bh);
goto done;
+ } else if (test_clear_buffer_dirty(bh)) {
+ warn_dirty_buffer(bh);
+ JBUFFER_TRACE(jh, "Journalling dirty buffer");
+ set_buffer_jbddirty(bh);
}
unlock_buffer(bh);

I think we'd better not to move the sequence?

2. I agree that the assertions in branch 'if (jh->b_transaction)' are redundant, I will remove them in v3. Thanks for pointing that.

Also I don't quite understand the new comment you have added. Do you mean
we need to not only clear BH_Dirty bit but also set BH_JBDdirty as dirtying
(through jbd2_journal_dirty_metadata()) does not have to follow after
do_get_write_access()?


Yes.
I think one reason we have non-empty commit_transaction->t_reserved_list is that: jbd2_journal_restart() could let jh attach to transaction_1 and dirty jh in transaction_2.

buffer is dirty after trans_0 committed
do_get_write_access() => jh->trans = old_handle->trans_1, clear buffer dirty & set jbddirty, BJ_Reserved
jbd2_journal_restart() => stop old_handle && may jbd2_log_start_commit && start new_handle with trans_2
jbd2_journal_commit_transaction() => clear jbddirty & set buffer dirty & set jh->b_transaction = NULL
do_checkpoint => buffer is fell on disk. If do_get_write_access() not mark jbddirty, buffer won't be fell on disk after checkpoint, which could corrupt filesystem.

I'm not sure whether we have the above path in realworld. I guess it exists in theory according to the comments:
/*
* 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 jbd2_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 jbd2_journal_get_write_access() calls to the same
* buffer are perfectly permissible.
* We use journal->j_state_lock here to serialize processing of
* t_reserved_list with eviction of buffers from journal_unmap_buffer().
*/
while (commit_transaction->t_reserved_list) { [...]