tls: fix use-after-free in tls_push_record
authorDaniel Borkmann <daniel@iogearbox.net>
Fri, 15 Jun 2018 01:07:45 +0000 (03:07 +0200)
committerDavid S. Miller <davem@davemloft.net>
Fri, 15 Jun 2018 16:14:30 +0000 (09:14 -0700)
commita447da7d00410278c90d3576782a43f8b675d7be
tree1ea7b2b56e13cae18573b37ab40bb134a4fcc8c4
parent695ad876d091530e3eb5def7827f8d0106ca3e9f
tls: fix use-after-free in tls_push_record

syzkaller managed to trigger a use-after-free in tls like the
following:

  BUG: KASAN: use-after-free in tls_push_record.constprop.15+0x6a2/0x810 [tls]
  Write of size 1 at addr ffff88037aa08000 by task a.out/2317

  CPU: 3 PID: 2317 Comm: a.out Not tainted 4.17.0+ #144
  Hardware name: LENOVO 20FBCTO1WW/20FBCTO1WW, BIOS N1FET47W (1.21 ) 11/28/2016
  Call Trace:
   dump_stack+0x71/0xab
   print_address_description+0x6a/0x280
   kasan_report+0x258/0x380
   ? tls_push_record.constprop.15+0x6a2/0x810 [tls]
   tls_push_record.constprop.15+0x6a2/0x810 [tls]
   tls_sw_push_pending_record+0x2e/0x40 [tls]
   tls_sk_proto_close+0x3fe/0x710 [tls]
   ? tcp_check_oom+0x4c0/0x4c0
   ? tls_write_space+0x260/0x260 [tls]
   ? kmem_cache_free+0x88/0x1f0
   inet_release+0xd6/0x1b0
   __sock_release+0xc0/0x240
   sock_close+0x11/0x20
   __fput+0x22d/0x660
   task_work_run+0x114/0x1a0
   do_exit+0x71a/0x2780
   ? mm_update_next_owner+0x650/0x650
   ? handle_mm_fault+0x2f5/0x5f0
   ? __do_page_fault+0x44f/0xa50
   ? mm_fault_error+0x2d0/0x2d0
   do_group_exit+0xde/0x300
   __x64_sys_exit_group+0x3a/0x50
   do_syscall_64+0x9a/0x300
   ? page_fault+0x8/0x30
   entry_SYSCALL_64_after_hwframe+0x44/0xa9

This happened through fault injection where aead_req allocation in
tls_do_encryption() eventually failed and we returned -ENOMEM from
the function. Turns out that the use-after-free is triggered from
tls_sw_sendmsg() in the second tls_push_record(). The error then
triggers a jump to waiting for memory in sk_stream_wait_memory()
resp. returning immediately in case of MSG_DONTWAIT. What follows is
the trim_both_sgl(sk, orig_size), which drops elements from the sg
list added via tls_sw_sendmsg(). Now the use-after-free gets triggered
when the socket is being closed, where tls_sk_proto_close() callback
is invoked. The tls_complete_pending_work() will figure that there's
a pending closed tls record to be flushed and thus calls into the
tls_push_pending_closed_record() from there. ctx->push_pending_record()
is called from the latter, which is the tls_sw_push_pending_record()
from sw path. This again calls into tls_push_record(). And here the
tls_fill_prepend() will panic since the buffer address has been freed
earlier via trim_both_sgl(). One way to fix it is to move the aead
request allocation out of tls_do_encryption() early into tls_push_record().
This means we don't prep the tls header and advance state to the
TLS_PENDING_CLOSED_RECORD before allocation which could potentially
fail happened. That fixes the issue on my side.

Fixes: 3c4d7559159b ("tls: kernel TLS support")
Reported-by: syzbot+5c74af81c547738e1684@syzkaller.appspotmail.com
Reported-by: syzbot+709f2810a6a05f11d4d3@syzkaller.appspotmail.com
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Dave Watson <davejwatson@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/tls/tls_sw.c