releases/5.1.7/net-tls-fix-no-wakeup-on-partial-reads.patch

   1 From foo@baz Fri 31 May 2019 03:16:39 PM PDT
   2 From: Jakub Kicinski <jakub.kicinski@netronome.com>
   3 Date: Fri, 24 May 2019 10:34:32 -0700
   4 Subject: net/tls: fix no wakeup on partial reads
   5
   6 From: Jakub Kicinski <jakub.kicinski@netronome.com>
   7
   8 [ Upstream commit 04b25a5411f966c2e586909a8496553b71876fae ]
   9
  10 When tls_sw_recvmsg() partially copies a record it pops that
  11 record from ctx->recv_pkt and places it on rx_list.
  12
  13 Next iteration of tls_sw_recvmsg() reads from rx_list via
  14 process_rx_list() before it enters the decryption loop.
  15 If there is no more records to be read tls_wait_data()
  16 will put the process on the wait queue and got to sleep.
  17 This is incorrect, because some data was already copied
  18 in process_rx_list().
  19
  20 In case of RPC connections process may never get woken up,
  21 because peer also simply blocks in read().
  22
  23 I think this may also fix a similar issue when BPF is at
  24 play, because after __tcp_bpf_recvmsg() returns some data
  25 we subtract it from len and use continue to restart the
  26 loop, but len could have just reached 0, so again we'd
  27 sleep unnecessarily. That's added by:
  28 commit d3b18ad31f93 ("tls: add bpf support to sk_msg handling")
  29
  30 Fixes: 692d7b5d1f91 ("tls: Fix recvmsg() to be able to peek across multiple records")
  31 Reported-by: David Beckett <david.beckett@netronome.com>
  32 Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
  33 Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
  34 Tested-by: David Beckett <david.beckett@netronome.com>
  35 Signed-off-by: David S. Miller <davem@davemloft.net>
  36 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  37 ---
  38  net/tls/tls_sw.c |    8 ++------
  39  1 file changed, 2 insertions(+), 6 deletions(-)
  40
  41 --- a/net/tls/tls_sw.c
  42 +++ b/net/tls/tls_sw.c
  43 @@ -1692,7 +1692,7 @@ int tls_sw_recvmsg(struct sock *sk,
  44         len = len - copied;
  45         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
  46
  47 -       do {
  48 +       while (len && (decrypted + copied < target || ctx->recv_pkt)) {
  49                 bool retain_skb = false;
  50                 bool zc = false;
  51                 int to_decrypt;
  52 @@ -1823,11 +1823,7 @@ pick_next_record:
  53                 } else {
  54                         break;
  55                 }
  56 -
  57 -               /* If we have a new message from strparser, continue now. */
  58 -               if (decrypted + copied >= target && !ctx->recv_pkt)
  59 -                       break;
  60 -       } while (len);
  61 +       }
  62
  63  recv_end:
  64         if (num_async) {