[net,0/5] tls: fix some issues with async encryption

Message

Sabrina Dubroca Sept. 6, 2023, 5:08 p.m. UTC

I've been playing with a few hacks in the crypto code (forcing async
crypto for every request, smaller cryptd queue), and that has revealed
some bugs while running the selftests.

With this setup and those patches applied, the bad_in_large_read test
case still fails. With all-async crypto, we don't know which record
threw the EBADMSG, so we can't keep the first couple of records that
were decrypted correctly. We have to throw away the whole batch.

Liu Jian has also found a bug with async crypto and wrapping record
numbers:
https://lore.kernel.org/netdev/[e-mail]/
It will get fixed separately and I'll submit a selftest for this case.

Sabrina Dubroca (5):
  net: tls: handle -EBUSY on async encrypt/decrypt requests
  tls: fix use-after-free with partial reads and async decrypt
  tls: fix returned read length with async !zc decrypt
  tls: fix race condition in async decryption of corrupted records
  tls: don't decrypt the next record if it's of a different type

 net/tls/tls_sw.c | 45 +++++++++++++++++++++++++++++++++++----------
 1 file changed, 35 insertions(+), 10 deletions(-)