Received-SPF: pass (google.com: domain of linux-ext4+bounces-1376-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Date: Fri, 23 Feb 2024 13:05:47 +0100
From: Jan Kara <jack@suse.cz>
To: Baokun Li <libaokun1@huawei.com>
Cc: Jan Kara <jack@suse.cz>, linux-ext4@vger.kernel.org, tytso@mit.edu,
	adilger.kernel@dilger.ca, ritesh.list@gmail.com,
	linux-kernel@vger.kernel.org, yi.zhang@huawei.com,
	yangerkun@huawei.com, chengzhihao1@huawei.com, yukuai3@huawei.com,
	stable@vger.kernel.org
Subject: Re: [PATCH 4/7] ext4: add positive int attr pointer to avoid sysfs
 variables overflow
Message-ID: <20240223120547.lojc4ccfewi6iotw@quack3>
References: <20240126085716.1363019-1-libaokun1@huawei.com>
 <20240126085716.1363019-5-libaokun1@huawei.com>
 <20240213165810.3k4lnxaqzdwrdj35@quack3>
 <83c16b1a-832d-2ffd-6100-1f2b80ca2f35@huawei.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <83c16b1a-832d-2ffd-6100-1f2b80ca2f35@huawei.com>

On Sat 17-02-24 15:41:43, Baokun Li wrote:
> On 2024/2/14 0:58, Jan Kara wrote:
> > On Fri 26-01-24 16:57:13, Baokun Li wrote:
> > > We can easily trigger a BUG_ON by using the following commands:
> > > 
> > >      mount /dev/$disk /tmp/test
> > >      echo 2147483650 > /sys/fs/ext4/$disk/mb_group_prealloc
> > >      echo test > /tmp/test/file && sync
> > > 
> > > ==================================================================
> > > kernel BUG at fs/ext4/mballoc.c:2029!
> > > invalid opcode: 0000 [#1] PREEMPT SMP PTI
> > > CPU: 3 PID: 320 Comm: kworker/u36:1 Not tainted 6.8.0-rc1 #462
> > > RIP: 0010:mb_mark_used+0x358/0x370
> > > [...]
> > > Call Trace:
> > >   ext4_mb_use_best_found+0x56/0x140
> > >   ext4_mb_complex_scan_group+0x196/0x2f0
> > >   ext4_mb_regular_allocator+0xa92/0xf00
> > >   ext4_mb_new_blocks+0x302/0xbc0
> > >   ext4_ext_map_blocks+0x95a/0xef0
> > >   ext4_map_blocks+0x2b1/0x680
> > >   ext4_do_writepages+0x733/0xbd0
> > > [...]
> > > ==================================================================
> > > 
> > > In ext4_mb_normalize_group_request():
> > >      ac->ac_g_ex.fe_len = EXT4_SB(sb)->s_mb_group_prealloc;
> > > 
> > > Here fe_len is of type int, but s_mb_group_prealloc is of type unsigned
> > > int, so setting s_mb_group_prealloc to 2147483650 overflows fe_len to a
> > > negative number, which ultimately triggers a BUG_ON() in mb_mark_used().
> > > 
> > > Therefore, we add attr_pointer_pi (aka positive int attr pointer) with a
> > > value range of 0-INT_MAX to avoid the above problem. In addition to the
> > > mb_group_prealloc sysfs interface, the following interfaces also have uint
> > > to int conversions that result in overflows, and are also fixed.
> > > 
> > >    err_ratelimit_burst
> > >    msg_ratelimit_burst
> > >    warning_ratelimit_burst
> > >    err_ratelimit_interval_ms
> > >    msg_ratelimit_interval_ms
> > >    warning_ratelimit_interval_ms
> > >    mb_best_avail_max_trim_order
> > > 
> > > CC: stable@vger.kernel.org
> > > Signed-off-by: Baokun Li <libaokun1@huawei.com>
> > I don't think you need to change s_mb_group_prealloc here and then restrict
> > it even further in the next patch. I'd just leave it alone here.
> Yes, we could put the next patch before this one, but using
> s_mb_group_prealloc as an example makes it easier to understand
> why the attr_pointer_pi case is added here.There are several other
> variables that don't have more convincing examples.

Yes, I think reordering would be good. Because I've read the convertion and
started wondering: "is this enough?"

> > Also I think that limiting mb_best_avail_max_trim_order to 64 instead of
> > INT_MAX will make us more resilient to surprises in the future :) But I
> > don't really insist.
> > 
> > 								Honza
> I think it's enough here to make sure that mb_best_avail_max_trim_order
> is a positive number, since we always make sure that min_order
> is not less than 0, as follows:
> 
>          order = fls(ac->ac_g_ex.fe_len) - 1;
>          min_order = order - sbi->s_mb_best_avail_max_trim_order;
>          if (min_order < 0)
>                  min_order = 0;
> 
> An oversized mb_best_avail_max_trim_order can be interpreted as
> always being CR_ANY_FREE. ????

Well, s_mb_best_avail_max_trim_order is not about allocation passes but
about how many times are we willing to shorten the goal extent to half and
still use the advanced free blocks search. And I agree that the mballoc
code is careful enough that large numbers don't matter there but still why
allowing storing garbage values? It is nicer to tell sysadmin he did
something wrong right away.

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR