On 2023/11/17 12:59, D. Wythe wrote:
From: "D. Wythe" <alibuda@xxxxxxxxxxxxxxxxx> We found a data corruption issue during testing of SMC-R on Redis applications. The benchmark has a low probability of reporting a strange error as shown below. "Error: Protocol error, got "\xe2" as reply type byte" Finally, we found that the retrieved error data was as follows: 0xE2 0xD4 0xC3 0xD9 0x04 0x00 0x2C 0x20 0xA6 0x56 0x00 0x16 0x3E 0x0C 0xCB 0x04 0x02 0x01 0x00 0x00 0x20 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xE2 It is quite obvious that this is a SMC DECLINE message, which means that the applications received SMC protocol message. We found that this was caused by the following situations: client server proposal -------------> accept <------------- confirm -------------> wait confirm failed llc confirm x------ (after 2s)timeout wait rsp wait decline (after 1s) timeout (after 2s) timeout decline --------------> decline <-------------- As a result, a decline message was sent in the implementation, and this message was read from TCP by the already-fallback connection. This patch double the client timeout as 2x of the server value,
Is the client's timeout doubled? From the code below, it is server's timeout that has been doubled.
With this simple change, the Decline messages should never cross or collide (during Confirm link timeout). This issue requires an immediate solution, since the protocol updates involve a more long-term solution. Fixes: 0fb0b02bd6fd ("net/smc: adapt SMC client code to use the LLC flow") Signed-off-by: D. Wythe <alibuda@xxxxxxxxxxxxxxxxx> --- include/net/netns/smc.h | 2 ++ net/smc/af_smc.c | 3 ++- net/smc/smc_sysctl.c | 12 ++++++++++++ 3 files changed, 16 insertions(+), 1 deletion(-) diff --git a/include/net/netns/smc.h b/include/net/netns/smc.h index 582212a..5198896 100644 --- a/include/net/netns/smc.h +++ b/include/net/netns/smc.h @@ -22,5 +22,7 @@ struct netns_smc { int sysctl_smcr_testlink_time; int sysctl_wmem; int sysctl_rmem; + /* server's Confirm Link timeout in seconds */ + int sysctl_smcr_srv_confirm_link_timeout; }; #endif diff --git a/net/smc/af_smc.c b/net/smc/af_smc.c index abd2667..b86ad30 100644 --- a/net/smc/af_smc.c +++ b/net/smc/af_smc.c @@ -1870,7 +1870,8 @@ static int smcr_serv_conf_first_link(struct smc_sock *smc) return SMC_CLC_DECL_TIMEOUT_CL;/* receive CONFIRM LINK response from client over the RoCE fabric */- qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, + qentry = smc_llc_wait(link->lgr, link, + sock_net(&smc->sk)->smc.sysctl_smcr_srv_confirm_link_timeout, SMC_LLC_CONFIRM_LINK); if (!qentry) { struct smc_clc_msg_decline dclc; diff --git a/net/smc/smc_sysctl.c b/net/smc/smc_sysctl.c index 5cbc18c..919f3f7 100644 --- a/net/smc/smc_sysctl.c +++ b/net/smc/smc_sysctl.c @@ -51,6 +51,13 @@ .proc_handler = proc_dointvec_jiffies, }, { + .procname = "smcr_srv_confirm_link_timeout", + .data = &init_net.smc.sysctl_smcr_srv_confirm_link_timeout, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = proc_dointvec_jiffies, + }, + { .procname = "wmem", .data = &init_net.smc.sysctl_wmem, .maxlen = sizeof(int), @@ -95,6 +102,11 @@ int __net_init smc_sysctl_net_init(struct net *net) net->smc.sysctl_autocorking_size = SMC_AUTOCORKING_DEFAULT_SIZE; net->smc.sysctl_smcr_buf_type = SMCR_PHYS_CONT_BUFS; net->smc.sysctl_smcr_testlink_time = SMC_LLC_TESTLINK_DEFAULT_TIME; + /* Increasing the server's timeout by twice as much as the client's + * timeout by default can temporarily avoid decline messages of + * both side been crossed or collided.
'both sides' or maybe better for '..avoid decline messages of both sides crossing or colliding.' Thanks, Wen Gu
+ */ + net->smc.sysctl_smcr_srv_confirm_link_timeout = 2 * SMC_LLC_WAIT_TIME; WRITE_ONCE(net->smc.sysctl_wmem, net_smc_wmem_init); WRITE_ONCE(net->smc.sysctl_rmem, net_smc_rmem_init);
