bpf: fix use after free in prog symbol exposure

syzkaller managed to trigger the warning in bpf_jit_free() which checks via
bpf_prog_kallsyms_verify_off() for potentially unlinked JITed BPF progs
in kallsyms, and subsequently trips over GPF when walking kallsyms entries:

  [...]
  8021q: adding VLAN 0 to HW filter on device batadv0
  8021q: adding VLAN 0 to HW filter on device batadv0
  WARNING: CPU: 0 PID: 9869 at kernel/bpf/core.c:810 bpf_jit_free+0x1e8/0x2a0
  Kernel panic - not syncing: panic_on_warn set ...
  CPU: 0 PID: 9869 Comm: kworker/0:7 Not tainted 5.0.0-rc8+ #1
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
  Workqueue: events bpf_prog_free_deferred
  Call Trace:
   __dump_stack lib/dump_stack.c:77 [inline]
   dump_stack+0x113/0x167 lib/dump_stack.c:113
   panic+0x212/0x40b kernel/panic.c:214
   __warn.cold.8+0x1b/0x38 kernel/panic.c:571
   report_bug+0x1a4/0x200 lib/bug.c:186
   fixup_bug arch/x86/kernel/traps.c:178 [inline]
   do_error_trap+0x11b/0x200 arch/x86/kernel/traps.c:271
   do_invalid_op+0x36/0x40 arch/x86/kernel/traps.c:290
   invalid_op+0x14/0x20 arch/x86/entry/entry_64.S:973
  RIP: 0010:bpf_jit_free+0x1e8/0x2a0
  Code: 02 4c 89 e2 83 e2 07 38 d0 7f 08 84 c0 0f 85 86 00 00 00 48 ba 00 02 00 00 00 00 ad de 0f b6 43 02 49 39 d6 0f 84 5f fe ff ff <0f> 0b e9 58 fe ff ff 48 b8 00 00 00 00 00 fc ff df 4c 89 e2 48 c1
  RSP: 0018:ffff888092f67cd8 EFLAGS: 00010202
  RAX: 0000000000000007 RBX: ffffc90001947000 RCX: ffffffff816e9d88
  RDX: dead000000000200 RSI: 0000000000000008 RDI: ffff88808769f7f0
  RBP: ffff888092f67d00 R08: fffffbfff1394059 R09: fffffbfff1394058
  R10: fffffbfff1394058 R11: ffffffff89ca02c7 R12: ffffc90001947002
  R13: ffffc90001947020 R14: ffffffff881eca80 R15: ffff88808769f7e8
  BUG: unable to handle kernel paging request at fffffbfff400d000
  #PF error: [normal kernel read fault]
  PGD 21ffee067 P4D 21ffee067 PUD 21ffed067 PMD 9f942067 PTE 0
  Oops: 0000 [#1] PREEMPT SMP KASAN
  CPU: 0 PID: 9869 Comm: kworker/0:7 Not tainted 5.0.0-rc8+ #1
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
  Workqueue: events bpf_prog_free_deferred
  RIP: 0010:bpf_get_prog_addr_region kernel/bpf/core.c:495 [inline]
  RIP: 0010:bpf_tree_comp kernel/bpf/core.c:558 [inline]
  RIP: 0010:__lt_find include/linux/rbtree_latch.h:115 [inline]
  RIP: 0010:latch_tree_find include/linux/rbtree_latch.h:208 [inline]
  RIP: 0010:bpf_prog_kallsyms_find+0x107/0x2e0 kernel/bpf/core.c:632
  Code: 00 f0 ff ff 44 38 c8 7f 08 84 c0 0f 85 fa 00 00 00 41 f6 45 02 01 75 02 0f 0b 48 39 da 0f 82 92 00 00 00 48 89 d8 48 c1 e8 03 <42> 0f b6 04 30 84 c0 74 08 3c 03 0f 8e 45 01 00 00 8b 03 48 c1 e0
  [...]

Upon further debugging, it turns out that whenever we trigger this
issue, the kallsyms removal in bpf_prog_ksym_node_del() was /skipped/
but yet bpf_jit_free() reported that the entry is /in use/.

Problem is that symbol exposure via bpf_prog_kallsyms_add() but also
perf_event_bpf_event() were done /after/ bpf_prog_new_fd(). Once the
fd is exposed to the public, a parallel close request came in right
before we attempted to do the bpf_prog_kallsyms_add().

Given at this time the prog reference count is one, we start to rip
everything underneath us via bpf_prog_release() -> bpf_prog_put().
The memory is eventually released via deferred free, so we're seeing
that bpf_jit_free() has a kallsym entry because we added it from
bpf_prog_load() but /after/ bpf_prog_put() from the remote CPU.

Therefore, move both notifications /before/ we install the fd. The
issue was never seen between bpf_prog_alloc_id() and bpf_prog_new_fd()
because upon bpf_prog_get_fd_by_id() we'll take another reference to
the BPF prog, so we're still holding the original reference from the
bpf_prog_load().

Fixes: 6ee52e2a ("perf, bpf: Introduce PERF_RECORD_BPF_EVENT")
Fixes: 74451e66 ("bpf: make jited programs visible in traces")
Reported-by: syzbot+bd3bba6ff3fcea7a6ec6@syzkaller.appspotmail.com
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Song Liu <songliubraving@fb.com>

kernel/bpf/syscall.c

@@ -1707,20 +1707,26 @@ static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
 	if (err)
 		goto free_used_maps;
 
-	err = bpf_prog_new_fd(prog);
-	if (err < 0) {
-		/* failed to allocate fd.
-		 * bpf_prog_put() is needed because the above
-		 * bpf_prog_alloc_id() has published the prog
-		 * to the userspace and the userspace may
-		 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
-		 */
-		bpf_prog_put(prog);
-		return err;
-	}
-
+	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
+	 * effectively publicly exposed. However, retrieving via
+	 * bpf_prog_get_fd_by_id() will take another reference,
+	 * therefore it cannot be gone underneath us.
+	 *
+	 * Only for the time /after/ successful bpf_prog_new_fd()
+	 * and before returning to userspace, we might just hold
+	 * one reference and any parallel close on that fd could
+	 * rip everything out. Hence, below notifications must
+	 * happen before bpf_prog_new_fd().
+	 *
+	 * Also, any failure handling from this point onwards must
+	 * be using bpf_prog_put() given the program is exposed.
+	 */
 	bpf_prog_kallsyms_add(prog);
 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
+
+	err = bpf_prog_new_fd(prog);
+	if (err < 0)
+		bpf_prog_put(prog);
 	return err;
 
 free_used_maps: