[PATCH net-next 0/4] net/rds: own the fastpath locks across connection teardown

Hi all,

This is a follow-up set to the recent "net/rds: Bug fix ports, part 2"
series [1]. During review of v1, the later half of the set needed more
work than just a respin, so it was split off into a separate set here
along with a few other companion fixes identified in the reivew.

RDS connection teardown quiesces the transmit and receive-refill fast
paths by waiting for the RDS_IN_XMIT/RDS_RECV_REFILL bits to be
sampled clear.  Sampling a bit clear is not owning it: the fast path
can re-take its bit right after the wait returns and then run
concurrently with the transport shutdown and the send-state reset.
Oracle UEK closed this by making teardown acquire the bits as locks
("rds: Make sure transmit path and connection tear-down does not run
concurrently"); patches 3 and 4 do the same for the two
rds_send_path_reset() call sites upstream.  These pacthes are
effectively v2 of patches 3 and 4 from the aforementioned
"net/rds: Bug fix ports, part 2" [1] set.

Making teardown block on the bits as locks promotes two latent
ordering bugs from rare to load-bearing, so they are fixed first:

  Patch 1: release_in_xmit() checks waitqueue_active() after
  clear_bit_unlock(), which does not order that read; the wake-up of
  the (now uninterruptible, untimed) teardown wait can be lost.  Use
  wq_has_sleeper().

  Patch 2: rds_conn_path_reset() wipes the whole cp_flags word with a
  plain store.  Once teardown owns bits in that word across the
  reset, a blanket store would end lock ownership early - and it
  already races atomic RMWs on the same word today.  Clear the bits
  the reset is responsible for individually, as Oracle UEK also does.

With those in place, patch 3 has rds_conn_shutdown() hold both bit
locks across the transport shutdown and path reset, and patch 4
converts rds_tcp_reset_callbacks() from waiting on RDS_IN_XMIT to
acquiring it, which also serializes the duelling-SYN accept path
against a concurrent teardown of the same path (the old code could
resume with a socket pointer the teardown had already released).

[PATCH net-next 1/4] net/rds: use wq_has_sleeper() in release_in_xmit()
  Restore full barrier before wake-up checks in release_in_xmit()

[PATCH net-next 2/4] net/rds: clear cp_flags bits individually in rds_conn_path_reset()
  Partial port of commit d04896037223 ("net/rds: Preserve essential connection state flags")
  https://github.com/oracle/linux-uek/commit/d04896037223

[PATCH net 3/4] net/rds: acquire the fastpath locks in rds_conn_shutdown()
  Port commit 2b8aaa4f163b ("rds: Make sure transmit path and connection tear-down does not run concurrently")
  https://github.com/oracle/linux-uek/commit/2b8aaa4f163b 

[PATCH net 4/4] net/rds: acquire RDS_IN_XMIT in rds_tcp_reset_callbacks()
  Extend the last port to cover extra call sites in rds_tcp_reset_callbacks()

Questions and comments appreciated!

Thanks,
Allison

[1] https://lore.kernel.org/netdev/[email protected]/

Allison Henderson (3):
  net/rds: use wq_has_sleeper() in release_in_xmit()
  net/rds: clear cp_flags bits individually in rds_conn_path_reset()
  net/rds: acquire RDS_IN_XMIT in rds_tcp_reset_callbacks()

Håkon Bugge (1):
  net/rds: acquire the fastpath locks in rds_conn_shutdown()

 net/rds/connection.c | 32 +++++++++++++++++++++++++---
 net/rds/send.c       | 12 ++++++++---
 net/rds/tcp.c        | 50 +++++++++++++++++++++++++++-----------------
 3 files changed, 69 insertions(+), 25 deletions(-)


base-commit: 3da8c3c8b8fa99505624b65ef590482f48e766b6
-- 
2.25.1