On 7/11/20 7:01 AM, Miklos Szeredi wrote:
> On Thu, Jun 25, 2020 at 11:02 AM Vasily Averin <vvs@virtuozzo.com> wrote:
>>
>> In current implementation fuse_writepages_fill() tries to share the code:
>> for new wpa it calls tree_insert() with num_pages = 0
>> then switches to common code used non-modified num_pages
>> and increments it at the very end.
>>
>> Though it triggers WARN_ON(!wpa->ia.ap.num_pages) in tree_insert()
>>  WARNING: CPU: 1 PID: 17211 at fs/fuse/file.c:1728 tree_insert+0xab/0xc0 [fuse]
>>  RIP: 0010:tree_insert+0xab/0xc0 [fuse]
>>  Call Trace:
>>   fuse_writepages_fill+0x5da/0x6a0 [fuse]
>>   write_cache_pages+0x171/0x470
>>   fuse_writepages+0x8a/0x100 [fuse]
>>   do_writepages+0x43/0xe0
>>
>> This patch re-works fuse_writepages_fill() to call tree_insert()
>> with num_pages = 1 and avoids its subsequent increment and
>> an extra spin_lock(&fi->lock) for newly added wpa.
> 
> Looks good.  However, I don't like the way fuse_writepage_in_flight()
> is silently changed to insert page into the rb_tree.  Also the
> insertion can be merged with the search for in-flight and be done
> unconditionally to simplify the logic.  See attached patch.

Your patch looks correct for me except 2 things:
1) you have lost "data->wpa = NULL;" when fuse_writepage_add() returns false.
2) in the same case old code did not set data->orig_pages[ap->num_pages] = page;

I've lightly updated your patch to fix noticed problems, please see attached patch.

Thank you,
	Vasily Averin