[PATCH] files-backend: unlock packed store only if locked

[PATCH] files-backend: unlock packed store only if locked

[Jonathan Tan]

In commit 42c7f7ff9685 ("commit_packed_refs(): remove call to
`packed_refs_unlock()`", 2017-06-23), a call to packed_refs_unlock() was
added to files_initial_transaction_commit() in order to compensate for
removing that call from commit_packed_refs(). However, that call was
added in the cleanup section, which is run even if the packed_ref_store
was never locked (which happens if an error occurs earlier in the
function).

Create a new cleanup goto target which runs packed_refs_unlock(), and
ensure that only goto statements after a successful invocation of
packed_refs_lock() jump there.

Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
---
I noticed this when one of our servers sent duplicate refs in its ref
advertisement (noticed through GIT_TRACE_PACKET). With this change (and
before the aforementioned commit 42c7f7ff9685), the error message is
"fatal: multiple updates for ref '<ref>' not allowed", which gives a
bigger clue to the problem. Currently, it is "fatal: BUG:
packed_refs_unlock() called when not locked".

(I couldn't replicate this problem in C Git.)
---
 refs/files-backend.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/refs/files-backend.c b/refs/files-backend.c
index f75d960e1..89bc5584a 100644
--- a/refs/files-backend.c
+++ b/refs/files-backend.c
@@ -2931,13 +2931,14 @@ static int files_initial_transaction_commit(struct ref_store *ref_store,
 
 	if (initial_ref_transaction_commit(packed_transaction, err)) {
 		ret = TRANSACTION_GENERIC_ERROR;
-		goto cleanup;
+		goto locked_cleanup;
 	}
 
+locked_cleanup:
+	packed_refs_unlock(refs->packed_ref_store);
 cleanup:
 	if (packed_transaction)
 		ref_transaction_free(packed_transaction);
-	packed_refs_unlock(refs->packed_ref_store);
 	transaction->state = REF_TRANSACTION_CLOSED;
 	string_list_clear(&affected_refnames, 0);
 	return ret;
-- 
2.16.0.rc1.238.g530d649a79-goog

Re: [PATCH] files-backend: unlock packed store only if locked

[Jeff King]

On Tue, Feb 06, 2018 at 12:36:15PM -0800, Jonathan Tan wrote:

> In commit 42c7f7ff9685 ("commit_packed_refs(): remove call to
> `packed_refs_unlock()`", 2017-06-23), a call to packed_refs_unlock() was
> added to files_initial_transaction_commit() in order to compensate for
> removing that call from commit_packed_refs(). However, that call was
> added in the cleanup section, which is run even if the packed_ref_store
> was never locked (which happens if an error occurs earlier in the
> function).
> 
> Create a new cleanup goto target which runs packed_refs_unlock(), and
> ensure that only goto statements after a successful invocation of
> packed_refs_lock() jump there.

The explanation and patch look good to me.

But this all seemed strangely familiar... I think this is the same bug
as:

  https://public-inbox.org/git/20180118143841.1a4c674d@novascotia/

which is queued as mr/packed-ref-store-fix. It's listed as "will merge
to next" in the "what's cooking" from Jan 31st.

> diff --git a/refs/files-backend.c b/refs/files-backend.c
> index f75d960e1..89bc5584a 100644
> --- a/refs/files-backend.c
> +++ b/refs/files-backend.c
> @@ -2931,13 +2931,14 @@ static int files_initial_transaction_commit(struct ref_store *ref_store,
>  
>  	if (initial_ref_transaction_commit(packed_transaction, err)) {
>  		ret = TRANSACTION_GENERIC_ERROR;
> -		goto cleanup;
> +		goto locked_cleanup;
>  	}
>  
> +locked_cleanup:
> +	packed_refs_unlock(refs->packed_ref_store);
>  cleanup:
>  	if (packed_transaction)
>  		ref_transaction_free(packed_transaction);
> -	packed_refs_unlock(refs->packed_ref_store);

I actually like this double-label a bit more than what is queued on
mr/packed-ref-store-fix, though I am OK with either solution.

-Peff

Re: [PATCH] files-backend: unlock packed store only if locked

[Jonathan Tan]

On Wed, 7 Feb 2018 09:42:51 -0500
Jeff King <peff@peff.net> wrote:

> But this all seemed strangely familiar... I think this is the same bug
> as:
> 
>   https://public-inbox.org/git/20180118143841.1a4c674d@novascotia/
> 
> which is queued as mr/packed-ref-store-fix. It's listed as "will merge
> to next" in the "what's cooking" from Jan 31st.

Ah...thanks, I didn't notice that.

> I actually like this double-label a bit more than what is queued on
> mr/packed-ref-store-fix, though I am OK with either solution.

Same here.

Re: [PATCH] files-backend: unlock packed store only if locked

[Junio C Hamano]

Jonathan Tan <jonathantanmy@google.com> writes:

> On Wed, 7 Feb 2018 09:42:51 -0500
> Jeff King <peff@peff.net> wrote:
>
>> But this all seemed strangely familiar... I think this is the same bug
>> as:
>> 
>>   https://public-inbox.org/git/20180118143841.1a4c674d@novascotia/
>> 
>> which is queued as mr/packed-ref-store-fix. It's listed as "will merge
>> to next" in the "what's cooking" from Jan 31st.
>
> Ah...thanks, I didn't notice that.
>
>> I actually like this double-label a bit more than what is queued on
>> mr/packed-ref-store-fix, though I am OK with either solution.
>
> Same here.

I do agree that the double-label approach is more future-proof way,
especially if we anticipate that there will be more code after the
"attempt initial ref transaction commit" block before the
packed-ref-store is unlocked.  On the other hand, introduction of
the locked_cleanup label can be done as part of such a change that
adds new code that needs to be skipped, so I am OK with what is
queued there.

Thanks, all.