tls: fix peeking with sync+async decryption
authorSabrina Dubroca <sd@queasysnail.net>
Wed, 28 Feb 2024 22:43:58 +0000 (23:43 +0100)
committerJakub Kicinski <kuba@kernel.org>
Thu, 29 Feb 2024 17:07:16 +0000 (09:07 -0800)
If we peek from 2 records with a currently empty rx_list, and the
first record is decrypted synchronously but the second record is
decrypted async, the following happens:
  1. decrypt record 1 (sync)
  2. copy from record 1 to the userspace's msg
  3. queue the decrypted record to rx_list for future read(!PEEK)
  4. decrypt record 2 (async)
  5. queue record 2 to rx_list
  6. call process_rx_list to copy data from the 2nd record

We currently pass copied=0 as skip offset to process_rx_list, so we
end up copying once again from the first record. We should skip over
the data we've already copied.

Seen with selftest tls.12_aes_gcm.recv_peek_large_buf_mult_recs

Fixes: 692d7b5d1f91 ("tls: Fix recvmsg() to be able to peek across multiple records")
Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/1b132d2b2b99296bfde54e8a67672d90d6d16e71.1709132643.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/tls/tls_sw.c

index 9f23ba321efe6c48721be793d848e148c58b7b4c..1394fc44f378899cd04a4d0cae9304a581416dcd 100644 (file)
@@ -1950,6 +1950,7 @@ int tls_sw_recvmsg(struct sock *sk,
        struct strp_msg *rxm;
        struct tls_msg *tlm;
        ssize_t copied = 0;
+       ssize_t peeked = 0;
        bool async = false;
        int target, err;
        bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
@@ -2097,8 +2098,10 @@ put_on_rx_list:
                        if (err < 0)
                                goto put_on_rx_list_err;
 
-                       if (is_peek)
+                       if (is_peek) {
+                               peeked += chunk;
                                goto put_on_rx_list;
+                       }
 
                        if (partially_consumed) {
                                rxm->offset += chunk;
@@ -2137,8 +2140,8 @@ recv_end:
 
                /* Drain records from the rx_list & copy if required */
                if (is_peek || is_kvec)
-                       err = process_rx_list(ctx, msg, &control, copied,
-                                             decrypted, is_peek, NULL);
+                       err = process_rx_list(ctx, msg, &control, copied + peeked,
+                                             decrypted - peeked, is_peek, NULL);
                else
                        err = process_rx_list(ctx, msg, &control, 0,
                                              async_copy_bytes, is_peek, NULL);