On Wed, Jul 22, 2020 at 11:22:23AM -0700, syzbot wrote:

Hello,

syzbot found the following issue on:

HEAD commit:    4f1b4da5 Merge branch 'net-atlantic-various-features'
git tree:       net-next
console output: https://syzkaller.appspot.com/x/log.txt?x=14b3a040900000
kernel config:  https://syzkaller.appspot.com/x/.config?x=2b7b67c0c1819c87
dashboard link: https://syzkaller.appspot.com/bug?extid=0e4699d000d8b874d8dc
compiler:       gcc (GCC) 10.1.0-syz 20200507
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=14c93358900000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=14ab61f0900000

The syz repo has:
setsockopt$inet_sctp6_SCTP_MAX_BURST(r0, 0x84, 0x10, &(0x7f0000000100)=@assoc_value, 0x8)
                      ^^^^^^^^^^^^^^           ^^^^

#define SCTP_DELAYED_ACK_TIME   16
#define SCTP_DELAYED_ACK SCTP_DELAYED_ACK_TIME
#define SCTP_DELAYED_SACK SCTP_DELAYED_ACK_TIME
#define SCTP_MAX_BURST  20              /* Set/Get max burst */

C repro has:
  syscall(__NR_setsockopt, r[0], 0x84, 0x10, 0x20000100ul, 8ul);
                                       ^^^^

So I'm wondering, what was the real intention of the call?


Anyhow, the issue is real, introduced by ebb25defdc17 ("sctp: pass a
kernel pointer to sctp_setsockopt_delayed_ack"). It used to use a
local storage bigger than the data provided by the user and used
one struct to read another's content on top of it. Quite masked.
I'll cook a fix.

  Marcelo