On 8/31/21 12:05 PM, Hao Xu wrote:

在 2021/8/31 下午5:42, Pavel Begunkov 写道:

quoted

On 8/31/21 10:19 AM, Hao Xu wrote:

quoted

在 2021/8/31 上午10:14, Jens Axboe 写道:

quoted

On 8/30/21 2:45 PM, syzbot wrote:

quoted

syzbot has found a reproducer for the following issue on:

HEAD commit:    93717cde744f Add linux-next specific files for 20210830
git tree:       linux-next
console output: https://syzkaller.appspot.com/x/log.txt?x=15200fad300000
kernel config:  https://syzkaller.appspot.com/x/.config?x=c643ef5289990dd1
dashboard link: https://syzkaller.appspot.com/bug?extid=f9704d1878e290eddf73
compiler:       gcc (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils for Debian) 2.35.1
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=111f5f9d300000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=1651a415300000

IMPORTANT: if you fix the issue, please add the following tag to the commit:
Reported-by: syzbot+f9704d1878e290eddf73@syzkaller.appspotmail.com

general protection fault, probably for non-canonical address 0xdffffc0000000005: 0000 [#1] PREEMPT SMP KASAN
KASAN: null-ptr-deref in range [0x0000000000000028-0x000000000000002f]
CPU: 0 PID: 6548 Comm: syz-executor433 Not tainted 5.14.0-next-20210830-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:sock_from_file+0x20/0x90 net/socket.c:505
Code: f5 ff ff ff c3 0f 1f 44 00 00 41 54 53 48 89 fb e8 85 e9 62 fa 48 8d 7b 28 48 b8 00 00 00 00 00 fc ff df 48 89 fa 48 c1 ea 03 <80> 3c 02 00 75 4f 45 31 e4 48 81 7b 28 80 f1 8a 8a 74 0c e8 58 e9
RSP: 0018:ffffc90002caf8e8 EFLAGS: 00010206
RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000000
RDX: 0000000000000005 RSI: ffffffff8713203b RDI: 0000000000000028
RBP: ffff888019fc0780 R08: ffffffff899aee40 R09: ffffffff81e21978
R10: 0000000000000027 R11: 0000000000000009 R12: dffffc0000000000
R13: 1ffff110033f80f9 R14: 0000000000000003 R15: ffff888019fc0780
FS:  00000000013b5300(0000) GS:ffff8880b9c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000004ae0f0 CR3: 000000001d355000 CR4: 00000000001506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
   io_sendmsg+0x98/0x640 fs/io_uring.c:4681
   io_issue_sqe+0x14de/0x6ba0 fs/io_uring.c:6578
   __io_queue_sqe+0x90/0xb50 fs/io_uring.c:6864
   io_req_task_submit+0xbf/0x1b0 fs/io_uring.c:2218
   tctx_task_work+0x166/0x610 fs/io_uring.c:2143
   task_work_run+0xdd/0x1a0 kernel/task_work.c:164
   tracehook_notify_signal include/linux/tracehook.h:212 [inline]
   handle_signal_work kernel/entry/common.c:146 [inline]
   exit_to_user_mode_loop kernel/entry/common.c:172 [inline]
   exit_to_user_mode_prepare+0x256/0x290 kernel/entry/common.c:209
   __syscall_exit_to_user_mode_work kernel/entry/common.c:291 [inline]
   syscall_exit_to_user_mode+0x19/0x60 kernel/entry/common.c:302
   do_syscall_64+0x42/0xb0 arch/x86/entry/common.c:86
   entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x43fd49

Hao, this is due to:

commit a8295b982c46d4a7c259a4cdd58a2681929068a9
Author: Hao Xu [off-list ref]
Date:   Fri Aug 27 17:46:09 2021 +0800

      io_uring: fix failed linkchain code logic

which causes some weirdly super long chains from that single sqe.
Can you take a look, please?

Sure, I'm working on this.

Ah, saw it after sending a patch. It's nothing too curious, just
a small error in logic. More interesting that we don't have a
test case covering it, we should definitely add something.

Saw your patch after coding my fix..😂
Since my email client doesn't receive your patch(only saw it in
webpage https://lore.kernel.org/), I put my comment here:

Hmm, does it happen often? I'll CC you

quoted

 fs/io_uring.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 473a977c7979..a531c7324ea8 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c

@@ -6717,6 +6717,8 @@ static inline void io_queue_sqe(struct io_kiocb *req)

     if (likely(!(req->flags & (REQ_F_FORCE_ASYNC | REQ_F_FAIL)))) {
         __io_queue_sqe(req);
     } else if (req->flags & REQ_F_FAIL) {
+        /* fail all, we don't submit */
+        req->flags &= ~REQ_F_HARDLINK;

maybe set REQ_F_LINK here?

if (unlikely((req->flags & REQ_F_FAIL) &&
	     !(req->flags & REQ_F_HARDLINK))) {
	posted |= (req->link != NULL);
	io_fail_links(req);
}

The problem is hardlink, normal will be failed. But there is indeed
a problem with both patches, 

if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK))
	// kill linked

Will resend with some tests on top

quoted

         io_req_complete_failed(req, req->result);
     } else {
         int ret = io_req_prep_async(req);

-- 
Pavel Begunkov