bpf, cgroup: Fix kernel BUG in purge_effective_progs

Syzkaller reported a triggered kernel BUG as follows:

  ------------[ cut here ]------------
  kernel BUG at kernel/bpf/cgroup.c:925!
  invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
  CPU: 1 PID: 194 Comm: detach Not tainted 5.19.0-14184-g69dac8e4 #8
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
  rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
  RIP: 0010:__cgroup_bpf_detach+0x1f2/0x2a0
  Code: 00 e8 92 60 30 00 84 c0 75 d8 4c 89 e0 31 f6 85 f6 74 19 42 f6 84
  28 48 05 00 00 02 75 0e 48 8b 80 c0 00 00 00 48 85 c0 75 e5 <0f> 0b 48
  8b 0c5
  RSP: 0018:ffffc9000055bdb0 EFLAGS: 00000246
  RAX: 0000000000000000 RBX: ffff888100ec0800 RCX: ffffc900000f1000
  RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffff888100ec4578
  RBP: 0000000000000000 R08: ffff888100ec0800 R09: 0000000000000040
  R10: 0000000000000000 R11: 0000000000000000 R12: ffff888100ec4000
  R13: 000000000000000d R14: ffffc90000199000 R15: ffff888100effb00
  FS:  00007f68213d2b80(0000) GS:ffff88813bc80000(0000)
  knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 000055f74a0e5850 CR3: 0000000102836000 CR4: 00000000000006e0
  Call Trace:
   <TASK>
   cgroup_bpf_prog_detach+0xcc/0x100
   __sys_bpf+0x2273/0x2a00
   __x64_sys_bpf+0x17/0x20
   do_syscall_64+0x3b/0x90
   entry_SYSCALL_64_after_hwframe+0x63/0xcd
  RIP: 0033:0x7f68214dbcb9
  Code: 08 44 89 e0 5b 41 5c c3 66 0f 1f 84 00 00 00 00 00 48 89 f8 48 89
  f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01
  f0 ff8
  RSP: 002b:00007ffeb487db68 EFLAGS: 00000246 ORIG_RAX: 0000000000000141
  RAX: ffffffffffffffda RBX: 000000000000000b RCX: 00007f68214dbcb9
  RDX: 0000000000000090 RSI: 00007ffeb487db70 RDI: 0000000000000009
  RBP: 0000000000000003 R08: 0000000000000012 R09: 0000000b00000003
  R10: 00007ffeb487db70 R11: 0000000000000246 R12: 00007ffeb487dc20
  R13: 0000000000000004 R14: 0000000000000001 R15: 000055f74a1011b0
   </TASK>
  Modules linked in:
  ---[ end trace 0000000000000000 ]---

Repetition steps:

For the following cgroup tree,

  root
   |
  cg1
   |
  cg2

  1. attach prog2 to cg2, and then attach prog1 to cg1, both bpf progs
     attach type is NONE or OVERRIDE.
  2. write 1 to /proc/thread-self/fail-nth for failslab.
  3. detach prog1 for cg1, and then kernel BUG occur.

Failslab injection will cause kmalloc fail and fall back to
purge_effective_progs. The problem is that cg2 have attached another prog,
so when go through cg2 layer, iteration will add pos to 1, and subsequent
operations will be skipped by the following condition, and cg will meet
NULL in the end.

  `if (pos && !(cg->bpf.flags[atype] & BPF_F_ALLOW_MULTI))`

The NULL cg means no link or prog match, this is as expected, and it's not
a bug. So here just skip the no match situation.

Fixes: 4c46091e ("bpf: Fix KASAN use-after-free Read in compute_effective_progs")
Signed-off-by: Pu Lehui <pulehui@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220813134030.1972696-1-pulehui@huawei.com

kernel/bpf/cgroup.c

@@ -921,8 +921,10 @@ static void purge_effective_progs(struct cgroup *cgrp, struct bpf_prog *prog,
 				pos++;
 			}
 		}
+
+		/* no link or prog match, skip the cgroup of this layer */
+		continue;
 found:
-		BUG_ON(!cg);
 		progs = rcu_dereference_protected(
 				desc->bpf.effective[atype],
 				lockdep_is_held(&cgroup_mutex));