Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 31 Oct 2022 11:11:58 -0400
From:   Josef Bacik <josef@toxicpanda.com>
To:     Hawkins Jiawei <yin31149@gmail.com>
Cc:     Chris Mason <clm@fb.com>, David Sterba <dsterba@suse.com>,
        18801353760@163.com, linux-btrfs@vger.kernel.org,
        linux-kernel@vger.kernel.org, llvm@lists.linux.dev,
        nathan@kernel.org, ndesaulniers@google.com,
        syzbot+dde7e853812ed57835ea@syzkaller.appspotmail.com,
        syzkaller-bugs@googlegroups.com, trix@redhat.com
Subject: Re: [PATCH] btrfs: update bytes_may_use in
 btrfs_free_reserved_bytes()
Message-ID: <Y1/lvgHE4JKvxsi8@localhost.localdomain>
References: <20221030162223.25970-1-yin31149@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20221030162223.25970-1-yin31149@gmail.com>
Precedence: bulk

On Mon, Oct 31, 2022 at 12:22:24AM +0800, Hawkins Jiawei wrote:
> Syzkaller reports warning as follows:
> =====================================
> WARNING: CPU: 0 PID: 3612 at fs/btrfs/space-info.h:122
>   btrfs_space_info_free_bytes_may_use fs/btrfs/space-info.h:154 [inline]
> WARNING: CPU: 0 PID: 3612 at fs/btrfs/space-info.h:122
>   block_rsv_release_bytes fs/btrfs/block-rsv.c:151 [inline]
> WARNING: CPU: 0 PID: 3612 at fs/btrfs/space-info.h:122
>   btrfs_block_rsv_release+0x5d1/0x730 fs/btrfs/block-rsv.c:295
> Modules linked in:
> [...]
> RIP: 0010:btrfs_space_info_update_bytes_may_use
>   fs/btrfs/space-info.h:122 [inline]
> RIP: 0010:btrfs_space_info_free_bytes_may_use
>   fs/btrfs/space-info.h:154 [inline]
> RIP: 0010:block_rsv_release_bytes
>   fs/btrfs/block-rsv.c:151 [inline]
> RIP: 0010:btrfs_block_rsv_release+0x5d1/0x730
>   fs/btrfs/block-rsv.c:295
> [...]
> Call Trace:
>  <TASK>
>  btrfs_release_global_block_rsv+0x2f/0x250 fs/btrfs/block-rsv.c:463
>  btrfs_free_block_groups+0xb67/0xfd0 fs/btrfs/block-group.c:4053
>  close_ctree+0x6c5/0xbde fs/btrfs/disk-io.c:4710
>  generic_shutdown_super+0x130/0x310 fs/super.c:491
>  kill_anon_super+0x36/0x60 fs/super.c:1085
>  btrfs_kill_super+0x3d/0x50 fs/btrfs/super.c:2441
>  deactivate_locked_super+0xa7/0xf0 fs/super.c:331
>  cleanup_mnt+0x4ce/0x560 fs/namespace.c:1186
>  task_work_run+0x146/0x1c0 kernel/task_work.c:177
>  ptrace_notify+0x29a/0x340 kernel/signal.c:2354
>  ptrace_report_syscall include/linux/ptrace.h:420 [inline]
>  ptrace_report_syscall_exit include/linux/ptrace.h:482 [inline]
>  syscall_exit_work+0x8c/0xe0 kernel/entry/common.c:249
>  syscall_exit_to_user_mode_prepare+0x63/0xc0 kernel/entry/common.c:276
>  __syscall_exit_to_user_mode_work kernel/entry/common.c:281 [inline]
>  syscall_exit_to_user_mode+0xa/0x60 kernel/entry/common.c:294
>  do_syscall_64+0x49/0xb0 arch/x86/entry/common.c:86
>  entry_SYSCALL_64_after_hwframe+0x63/0xcd
>  [...]
>  </TASK>
> =====================================
> 
> In btrfs_new_extent_direct(), kernel will reserves space for extent
> by btrfs_reserve_extent(), and frees those space by
> btrfs_free_reserved_extent() if btrfs_create_dio_extent() fails.
> 
> Yet the problem is that, it may not update the space
> info correctly. To be more specific, kernel will
> converts space from ->bytes_may_use to ->bytes_reserved, in
> btrfs_add_reserved_bytes() when reserving space.
> But when freeing those space in btrfs_free_reserved_bytes(),
> kernel does not convert space from ->bytes_reserved back to
> ->bytes_may_use, which triggers the above warning.
> 
> This patch solves it by converting space from ->bytes_reserved
> back to ->bytes_may_use in btrfs_free_reserved_bytes().
> 

This isn't correct.  I haven't looked at the code yet, but reservations go into
->bytes_may_use, and then when we reserve the space we subtract the reservation
from ->bytes_may_use and add it to ->bytes_reserved.  If we free the reserved
extent we only have to update ->bytes_reserved.  What may be happening here is
we're failing to free the rest of our ->bytes_may_use resrvation, and that part
needs to be addressed.  This fix as it stands however is incorrect.  Thanks,

Josef