Re: [PATCH v2 4/4] btrfs: dump all space infos if we abort transaction due to ENOSPC

[Sweet Tea Dorminy <sweettea-kernel@dorminy.me>]

On 7/26/22 14:20, David Sterba wrote:
> On Wed, Jul 20, 2022 at 09:03:33AM +0800, Qu Wenruo wrote:
>>>> @@ -346,12 +346,14 @@ void __cold btrfs_err_32bit_limit(struct
>>>> btrfs_fs_info *fs_info)
>>>>    __cold
>>>>    void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
>>>>                       const char *function,
>>>> -                   unsigned int line, int errno)
>>>> +                   unsigned int line, int errno, bool first_hit)
>>>>    {
>>>>        struct btrfs_fs_info *fs_info = trans->fs_info;
>>>>        WRITE_ONCE(trans->aborted, errno);
>>>>        WRITE_ONCE(trans->transaction->aborted, errno);
>>>> +    if (first_hit && errno == -ENOSPC)
>>>> +        btrfs_dump_fs_space_info(fs_info);
>>>>        /* Wake up anybody who may be waiting on this transaction */
>>>>        wake_up(&fs_info->transaction_wait);
>>>>        wake_up(&fs_info->transaction_blocked_wait);
>>> DO_ONCE_LITE(btrfs_dump_fs_space_info, fs_info) from <linux/once_lite.h>
>>> seems like a more lightweight way to dump the space infos once upon
>>> first transaction abort. Then you don't have to plumb through the
>>> 'first_hit' parameter from btrfs_abort_transaction(), and this change
>>> becomes even more minimal than it already is.
>>
>> Sounds pretty awesome!
> 
> But DO_ONCE_LITE stores the status in one static variable, this cant' be
> used because we want to track the status per filesystem, and also per
> mount. Ie. repeated ENOSPC after umount/mount cycle won't be reported,
> also another filesystem hitting abort due to ENOSPC.

That's a great point, I agree it would be better to print the first 
abort per filesystem/mount instead of only once per reboot now that it's 
noted.