kernel/drivers/net/ethernet/chelsio/inline_crypto, branch linux-5.14.y

tcp: switch orphan_count to bare per-cpu counters

2021-11-17T10:04:07Z

[ Upstream commit 19757cebf0c5016a1f36f7fe9810a9f0b33c0832 ] Use of percpu_counter structure to track count of orphaned sockets is causing problems on modern hosts with 256 cpus or more. Stefan Bach reported a serious spinlock contention in real workloads, that I was able to reproduce with a netfilter rule dropping incoming FIN packets. 53.56% server [kernel.kallsyms] [k] queued_spin_lock_slowpath | ---queued_spin_lock_slowpath | --53.51%--_raw_spin_lock_irqsave | --53.51%--__percpu_counter_sum tcp_check_oom | |--39.03%--__tcp_close | tcp_close | inet_release | inet6_release | sock_close | __fput | ____fput | task_work_run | exit_to_usermode_loop | do_syscall_64 | entry_SYSCALL_64_after_hwframe | __GI___libc_close | --14.48%--tcp_out_of_resources tcp_write_timeout tcp_retransmit_timer tcp_write_timer_handler tcp_write_timer call_timer_fn expire_timers __run_timers run_timer_softirq __softirqentry_text_start As explained in commit cf86a086a180 ("net/dst: use a smaller percpu_counter batch for dst entries accounting"), default batch size is too big for the default value of tcp_max_orphans (262144). But even if we reduce batch sizes, there would still be cases where the estimated count of orphans is beyond the limit, and where tcp_too_many_orphans() has to call the expensive percpu_counter_sum_positive(). One solution is to use plain per-cpu counters, and have a timer to periodically refresh this cache. Updating this cache every 100ms seems about right, tcp pressure state is not radically changing over shorter periods. percpu_counter was nice 15 years ago while hosts had less than 16 cpus, not anymore by current standards. v2: Fix the build issue for CONFIG_CRYPTO_DEV_CHELSIO_TLS=m, reported by kernel test robot Remove unused socket argument from tcp_too_many_orphans() Fixes: dd24c00191d5 ("net: Use a percpu_counter for orphan_count") Signed-off-by: Eric Dumazet Reported-by: Stefan Bach Cc: Neal Cardwell Acked-by: Neal Cardwell Signed-off-by: David S. Miller Signed-off-by: Sasha Levin

net: sock: introduce sk_error_report

2021-06-29T18:28:21Z

This patch introduces a function wrapper to call the sk_error_report callback. That will prepare to add additional handling whenever sk_error_report is called, for example to trace socket errors. Signed-off-by: Alexander Aring Signed-off-by: David S. Miller

cxgb4/ch_ktls: Clear resources when pf4 device is removed

2021-05-13T22:44:54Z

This patch maintain the list of active tids and clear all the active connection resources when DETACH notification comes. Fixes: a8c16e8ed624f ("crypto/chcr: move nic TLS functionality to drivers/net") Signed-off-by: Ayush Sawal Signed-off-by: David S. Miller

chelsio/chtls: unlock on error in chtls_pt_recvmsg()

2021-05-12T21:30:00Z

This error path needs to release some memory and call release_sock(sk); before returning. Fixes: 6919a8264a32 ("Crypto/chtls: add/delete TLS header in driver") Signed-off-by: Dan Carpenter Signed-off-by: David S. Miller

ch_ktls: Remove redundant variable result

2021-04-26T01:07:11Z

Variable result is being assigned a value from a calculation however the variable is never read, so this redundant variable can be removed. Cleans up the following clang-analyzer warning: drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c:1488:2: warning: Value stored to 'pos' is never read [clang-analyzer-deadcode.DeadStores]. drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c:876:3: warning: Value stored to 'pos' is never read [clang-analyzer-deadcode.DeadStores]. drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c:36:3: warning: Value stored to 'start' is never read [clang-analyzer-deadcode.DeadStores]. Reported-by: Abaci Robot Signed-off-by: Jiapeng Chong Signed-off-by: David S. Miller

ch_ktls: do not send snd_una update to TCB in middle

2021-04-15T23:55:49Z

snd_una update should not be done when the same skb is being sent out.chcr_short_record_handler() sends it again even though SND_UNA update is already sent for the skb in chcr_ktls_xmit(), which causes mismatch in un-acked TCP seq number, later causes problem in sending out complete record. Fixes: 429765a149f1 ("chcr: handle partial end part of a record") Signed-off-by: Vinay Kumar Yadav Signed-off-by: Rohit Maheshwari Signed-off-by: David S. Miller

ch_ktls: tcb close causes tls connection failure

2021-04-15T23:55:49Z

HW doesn't need marking TCB closed. This TCB state change sometimes causes problem to the new connection which gets the same tid. Fixes: 34aba2c45024 ("cxgb4/chcr : Register to tls add and del callback") Signed-off-by: Vinay Kumar Yadav Signed-off-by: Rohit Maheshwari Signed-off-by: David S. Miller

ch_ktls: fix device connection close

2021-04-15T23:55:49Z

When sge queue is full and chcr_ktls_xmit_wr_complete() returns failure, skb is not freed if it is not the last tls record in this skb, causes refcount never gets freed and tls_dev_del() never gets called on this connection. Fixes: 5a4b9fe7fece ("cxgb4/chcr: complete record tx handling") Signed-off-by: Vinay Kumar Yadav Signed-off-by: Rohit Maheshwari Signed-off-by: David S. Miller

ch_ktls: Fix kernel panic

2021-04-15T23:55:49Z

Taking page refcount is not ideal and causes kernel panic sometimes. It's better to take tx_ctx lock for the complete skb transmit, to avoid page cleanup if ACK received in middle. Fixes: 5a4b9fe7fece ("cxgb4/chcr: complete record tx handling") Signed-off-by: Vinay Kumar Yadav Signed-off-by: Rohit Maheshwari Signed-off-by: David S. Miller

ch_ktls: fix enum-conversion warning

2021-03-24T19:35:39Z

gcc points out an incorrect enum assignment: drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c: In function 'chcr_ktls_cpl_set_tcb_rpl': drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c:684:22: warning: implicit conversion from 'enum ' to 'enum ch_ktls_open_state' [-Wenum-conversion] This appears harmless, and should apparently use 'CH_KTLS_OPEN_SUCCESS' instead of 'false', with the same value '0'. Fixes: efca3878a5fb ("ch_ktls: Issue if connection offload fails") Reviewed-by: Andrew Lunn Signed-off-by: Arnd Bergmann Signed-off-by: David S. Miller