On 1/6/21 8:57 AM, Brian Foster wrote:
> On Tue, Jan 05, 2021 at 04:02:18PM -0600, Eric Sandeen wrote:
>> This pre-patch helps make the next libxfs-sync for 5.11 a bit
>> more clear.
>>
>> In reality, the libxfs_inode_alloc function matches the kernel's
>> xfs_dir_ialloc so rename it for clarity before the rest of the
>> sync, and change several variable names for the same reason.
>>
>> Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx>
>> ---
>>
> ...
>> diff --git a/libxfs/util.c b/libxfs/util.c
>> index 252cf91e..62eadaea 100644
>> --- a/libxfs/util.c
>> +++ b/libxfs/util.c
> ...
>> @@ -559,25 +561,25 @@ libxfs_inode_alloc(
>>
>> if (ialloc_context) {
>>
>> - xfs_trans_bhold(*tp, ialloc_context);
>> + xfs_trans_bhold(tp, ialloc_context);
>>
>> - error = xfs_trans_roll(tp);
>> - if (error) {
>> + code = xfs_trans_roll(tpp);
>
> The subsequent uses of tp no longer refer to the right transaction after
> this roll. FWIW, there is a subtle difference with the kernel code where
> this call passes &tp and then updates tpp on return, but we could also
> just update tp here if that's still good enough to facilitate the libxfs
> sync.
Whoops, now I remember meaning to go back and check that, thanks, I'll
take a closer look.
-Eric
