nvme-fabrics: short-circuit connect retries
Subsystems (4)
v7 cover nvme-fabrics: short-circuit connect retries
I've splitted the last patch into the hopefully non controversial part 'do not retry when DNR is set' and the 'don't retry auth failures'. I hope we can get at least the first few patches in and have a lively discussion on the final patch at LSF.
changes:
v7:
- updated commit message patch #3
- split last patch so that only the DNR checks are added first and then the auth reconnect failure test
v6:
- reorder series
- extended nvmf_should_reconnect
- added auth retry logic
- https://lore.kernel.org/linux-nvme/[email protected]/
v5:
- nvme: do not mix kernel error code with nvme status
- nvmet: separate nvme status from dhchap status
- https://lore.kernel.org/linux-nvme/[email protected]/
v4:
- rebased
- added 'nvme: fixes for authentication errors' series
- https://lore.kernel.org/all/[email protected]/
v3:
- added my SOB tag
- fixed indention
- https://lore.kernel.org/linux-nvme/[email protected]/
v2:
- refresh/rebase on current head
- extended blktests (nvme/045) to cover this case
(see separate post)
- https://lore.kernel.org/linux-nvme/[email protected]/
v1:
- initial version
- https://lore.kernel.org/linux-nvme/[email protected]/
Daniel Wagner (1):
nvme: do not retry authentication failures
Hannes Reinecke (4):
nvmet: lock config semaphore when accessing DH-HMAC-CHAP key
nvmet: return DHCHAP status codes from nvmet_setup_auth()
nvme: return kernel error codes for admin queue connect
nvme-fabrics: short-circuit reconnect retries
drivers/nvme/host/auth.c | 6 +-- drivers/nvme/host/core.c | 6 +-- drivers/nvme/host/fabrics.c | 51 ++++++++++++++++---------- drivers/nvme/host/fabrics.h | 2 +- drivers/nvme/host/fc.c | 4 +- drivers/nvme/host/nvme.h | 2 +- drivers/nvme/host/rdma.c | 19 ++++++---- drivers/nvme/host/tcp.c | 22 +++++++---- drivers/nvme/target/auth.c | 22 +++++------ drivers/nvme/target/configfs.c | 22 ++++++++--- drivers/nvme/target/fabrics-cmd-auth.c | 49 +++++++++++++------------ drivers/nvme/target/fabrics-cmd.c | 11 +++--- drivers/nvme/target/nvmet.h | 8 ++-- 13 files changed, 129 insertions(+), 95 deletions(-)
Signature · 1 line
2.44.0
PATCH v7 1/5 nvmet: lock config semaphore when accessing DH-HMAC-CHAP key
From: Hannes Reinecke <[email protected]>
When the DH-HMAC-CHAP key is accessed via configfs we need to take the config semaphore as a reconnect might be running at the same time.
--- drivers/nvme/target/auth.c | 2 ++ drivers/nvme/target/configfs.c | 22 +++++++++++++++++----- 2 files changed, 19 insertions(+), 5 deletions(-)
drivers/nvme/target/auth.c +2 -0
@@ -44,6 +44,7 @@ int nvmet_auth_set_key(struct nvmet_host *host, const char *secret, dhchap_secret =kstrdup (secret,GFP_KERNEL );if (!dhchap_secret)return -ENOMEM ; +down_write (&nvmet_config_sem);if (set_ctrl) {kfree (host->dhchap_ctrl_secret); host->dhchap_ctrl_secret =strim (dhchap_secret);
@@ -53,6 +54,7 @@ int nvmet_auth_set_key(struct nvmet_host *host, const char *secret, host->dhchap_secret =strim (dhchap_secret); host->dhchap_key_hash = key_hash; } +up_write (&nvmet_config_sem);return 0 ; }
drivers/nvme/target/configfs.c +17 -5
@@ -1990,11 +1990,17 @@ static struct config_group nvmet_ports_group;static ssize_t nvmet_host_dhchap_key_show (struct config_item *item,char *page) { -u8 *dhchap_secret =to_host (item)->dhchap_secret; +u8 *dhchap_secret; +ssize_t ret; +down_read (&nvmet_config_sem); + dhchap_secret =to_host (item)->dhchap_secret;if (!dhchap_secret) -return sprintf (page,"\n" ); -return sprintf (page,"%s\n" , dhchap_secret); + ret =sprintf (page,"\n" ); +else + ret =sprintf (page,"%s\n" , dhchap_secret); +up_read (&nvmet_config_sem); +return ret; } static ssize_t nvmet_host_dhchap_key_store(struct config_item *item,
@@ -2018,10 +2024,16 @@ static ssize_t nvmet_host_dhchap_ctrl_key_show(struct config_item *item, char *page) {u8 *dhchap_secret =to_host (item)->dhchap_ctrl_secret; +ssize_t ret; +down_read (&nvmet_config_sem); + dhchap_secret =to_host (item)->dhchap_ctrl_secret;if (!dhchap_secret) -return sprintf (page,"\n" ); -return sprintf (page,"%s\n" , dhchap_secret); + ret =sprintf (page,"\n" ); +else + ret =sprintf (page,"%s\n" , dhchap_secret); +up_read (&nvmet_config_sem); +return ret; } static ssize_t nvmet_host_dhchap_ctrl_key_store(struct config_item *item,
Signature · 1 line
2.44.0
PATCH v7 2/5 nvmet: return DHCHAP status codes from nvmet_setup_auth()
In reply to Daniel WagnerFrom: Hannes Reinecke <[email protected]>
A failure in nvmet_setup_auth() does not mean that the NVMe authentication command failed, so we should rather return a protocol error with a 'failure1' response than an NVMe status.
Also update the type used for dhchap_step and dhchap_status to u8 to avoid confusions with nvme status. Furthermore, split dhchap_status and nvme status so we don't accidentally mix these return values.
[dwagner: - use u8 as type for dhchap_{step|status}
- separate nvme status from dhcap_status]
--- drivers/nvme/target/auth.c | 20 +++++------ drivers/nvme/target/fabrics-cmd-auth.c | 49 +++++++++++++------------- drivers/nvme/target/fabrics-cmd.c | 11 +++--- drivers/nvme/target/nvmet.h | 8 ++--- 4 files changed, 43 insertions(+), 45 deletions(-)
drivers/nvme/target/auth.c +8 -12
@@ -126,12 +126,11 @@ int nvmet_setup_dhgroup(struct nvmet_ctrl *ctrl, u8 dhgroup_id)return ret; } -int nvmet_setup_auth (struct nvmet_ctrl *ctrl) +u8 nvmet_setup_auth (struct nvmet_ctrl *ctrl) {int ret =0 ;struct nvmet_host_link *p;struct nvmet_host *host =NULL ; -const char *hash_name;down_read (&nvmet_config_sem);if (nvmet_is_disc_subsys (ctrl->subsys))
@@ -149,13 +148,16 @@ int nvmet_setup_auth(struct nvmet_ctrl *ctrl) }if (!host) {pr_debug ("host %s not found\n" , ctrl->hostnqn); - ret = -EPERM ; + ret =NVME_AUTH_DHCHAP_FAILURE_FAILED ; goto out_unlock; } ret =nvmet_setup_dhgroup (ctrl, host->dhchap_dhgroup_id); -if (ret <0 ) +if (ret <0 ) {pr_warn ("Failed to setup DH group" ); + ret =NVME_AUTH_DHCHAP_FAILURE_DHGROUP_UNUSABLE ; + goto out_unlock; + }if (!host->dhchap_secret) {pr_debug ("No authentication provided\n" );
@@ -166,12 +168,6 @@ int nvmet_setup_auth(struct nvmet_ctrl *ctrl)pr_debug ("Re-use existing hash ID %d\n" , ctrl->shash_id); }else { - hash_name =nvme_auth_hmac_name (host->dhchap_hash_id); -if (!hash_name) { -pr_warn ("Hash ID %d invalid\n" , host->dhchap_hash_id); - ret = -EINVAL ; - goto out_unlock; - } ctrl->shash_id = host->dhchap_hash_id; }
@@ -180,7 +176,7 @@ int nvmet_setup_auth(struct nvmet_ctrl *ctrl) ctrl->host_key =nvme_auth_extract_key (host->dhchap_secret +10 , host->dhchap_key_hash);if (IS_ERR (ctrl->host_key)) { - ret =PTR_ERR (ctrl->host_key); + ret =NVME_AUTH_DHCHAP_FAILURE_NOT_USABLE ; ctrl->host_key =NULL ; goto out_free_hash; }
@@ -198,7 +194,7 @@ int nvmet_setup_auth(struct nvmet_ctrl *ctrl) ctrl->ctrl_key =nvme_auth_extract_key (host->dhchap_ctrl_secret +10 , host->dhchap_ctrl_key_hash);if (IS_ERR (ctrl->ctrl_key)) { - ret =PTR_ERR (ctrl->ctrl_key); + ret =NVME_AUTH_DHCHAP_FAILURE_NOT_USABLE ; ctrl->ctrl_key =NULL ; goto out_free_hash; }
drivers/nvme/target/fabrics-cmd-auth.c +25 -24
110 lines not on this page · raw message
drivers/nvme/target/fabrics-cmd.c +6 -5
@@ -211,7 +211,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)struct nvmf_connect_data *d;struct nvmet_ctrl *ctrl =NULL ;u16 status; -int ret; +u8 dhchap_status;if (!nvmet_check_transfer_len (req,sizeof (struct nvmf_connect_data )))return ;
@@ -251,11 +251,12 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)uuid_copy (&ctrl->hostid, &d->hostid); - ret =nvmet_setup_auth (ctrl); -if (ret <0 ) { -pr_err ("Failed to setup authentication, error %d\n" , ret); + dhchap_status =nvmet_setup_auth (ctrl); +if (dhchap_status) { +pr_err ("Failed to setup authentication, dhchap status %u\n" , + dhchap_status);nvmet_ctrl_put (ctrl); -if (ret == -EPERM ) +if (dhchap_status ==NVME_AUTH_DHCHAP_FAILURE_FAILED ) status = (NVME_SC_CONNECT_INVALID_HOST |NVME_SC_DNR );else status =NVME_SC_INTERNAL ;
drivers/nvme/target/nvmet.h +4 -4
@@ -113,8 +113,8 @@ struct nvmet_sq {bool authenticated;struct delayed_work auth_expired_work;u16 dhchap_tid; -u16 dhchap_status; -int dhchap_step; +u8 dhchap_status; +u8 dhchap_step;u8 *dhchap_c1;u8 *dhchap_c2;u32 dhchap_s1;
@@ -721,7 +721,7 @@ void nvmet_execute_auth_receive(struct nvmet_req *req);int nvmet_auth_set_key (struct nvmet_host *host,const char *secret,bool set_ctrl);int nvmet_auth_set_host_hash (struct nvmet_host *host,const char *hash); -int nvmet_setup_auth (struct nvmet_ctrl *ctrl); +u8 nvmet_setup_auth (struct nvmet_ctrl *ctrl);void nvmet_auth_sq_init (struct nvmet_sq *sq);void nvmet_destroy_auth (struct nvmet_ctrl *ctrl);void nvmet_auth_sq_free (struct nvmet_sq *sq);
@@ -740,7 +740,7 @@ int nvmet_auth_ctrl_exponential(struct nvmet_req *req,int nvmet_auth_ctrl_sesskey (struct nvmet_req *req,u8 *buf,int buf_size);#else -static inline int nvmet_setup_auth (struct nvmet_ctrl *ctrl) +static inline u8 nvmet_setup_auth (struct nvmet_ctrl *ctrl) {return 0 ; }
Signature · 1 line
2.44.0
Re: [PATCH v7 2/5] nvmet: return DHCHAP status codes from nvmet_setup_auth()
Looks good:
PATCH v7 3/5 nvme: return kernel error codes for admin queue connect
In reply to Daniel WagnerFrom: Hannes Reinecke <[email protected]>
nvmf_connect_admin_queue returns NVMe error status codes and kernel error codes. This mixes the different domains which makes maintainability difficult.
--- drivers/nvme/host/core.c | 6 +++--- drivers/nvme/host/fabrics.c | 31 +++++++++++++------------------ drivers/nvme/host/nvme.h | 2 +- 3 files changed, 17 insertions(+), 22 deletions(-)
drivers/nvme/host/core.c +3 -3
@@ -383,14 +383,14 @@ static inline enum nvme_disposition nvme_decide_disposition(struct request *req)if (likely (nvme_req (req)->status ==0 ))return COMPLETE ; -if ((nvme_req (req)->status &0x7ff ) ==NVME_SC_AUTH_REQUIRED ) -return AUTHENTICATE ; -if (blk_noretry_request (req) || (nvme_req (req)->status &NVME_SC_DNR ) ||nvme_req (req)->retries >= nvme_max_retries)return COMPLETE ; +if ((nvme_req (req)->status &0x7ff ) ==NVME_SC_AUTH_REQUIRED ) +return AUTHENTICATE ; +if (req->cmd_flags &REQ_NVME_MPATH ) {if (nvme_is_path_error (nvme_req (req)->status) ||blk_queue_dying (req->q))
drivers/nvme/host/fabrics.c +13 -18
@@ -428,12 +428,6 @@ static void nvmf_connect_cmd_prep(struct nvme_ctrl *ctrl, u16 qid,* fabrics-protocol connection of the NVMe Admin queue between the * host system device and the allocated NVMe controller on the * target system via a NVMe Fabrics "Connect" command. -* -* Return: -* 0: success -* > 0: NVMe error status code -* < 0: Linux errno error code -* */ int nvmf_connect_admin_queue (struct nvme_ctrl *ctrl) {
@@ -467,7 +461,7 @@ int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)if (result &NVME_CONNECT_AUTHREQ_ASCR ) {dev_warn (ctrl->device,"qid 0: secure concatenation is not supported\n" ); - ret =NVME_SC_AUTH_REQUIRED ; + ret = -EOPNOTSUPP ; goto out_free_data; }/* Authentication required */
@@ -475,14 +469,14 @@ int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)if (ret) {dev_warn (ctrl->device,"qid 0: authentication setup failed\n" ); - ret =NVME_SC_AUTH_REQUIRED ; goto out_free_data; } ret =nvme_auth_wait (ctrl,0 ); -if (ret) +if (ret) {dev_warn (ctrl->device, -"qid 0: authentication failed\n" ); -else +"qid 0: authentication failed, error %d\n" , + ret); + }else dev_info (ctrl->device,"qid 0: authenticated\n" ); }
@@ -542,7 +536,7 @@ int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)if (result &NVME_CONNECT_AUTHREQ_ASCR ) {dev_warn (ctrl->device,"qid 0: secure concatenation is not supported\n" ); - ret =NVME_SC_AUTH_REQUIRED ; + ret = -EOPNOTSUPP ; goto out_free_data; }/* Authentication required */
@@ -550,12 +544,13 @@ int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)if (ret) {dev_warn (ctrl->device,"qid %d: authentication setup failed\n" , qid); - ret =NVME_SC_AUTH_REQUIRED ; - }else { - ret =nvme_auth_wait (ctrl, qid); -if (ret) -dev_warn (ctrl->device, -"qid %u: authentication failed\n" , qid); + goto out_free_data; + } + ret =nvme_auth_wait (ctrl, qid); +if (ret) { +dev_warn (ctrl->device, +"qid %u: authentication failed, error %d\n" , + qid, ret); } } out_free_data:
drivers/nvme/host/nvme.h +1 -1
@@ -1122,7 +1122,7 @@ static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid) }static inline int nvme_auth_wait (struct nvme_ctrl *ctrl,int qid) { -return NVME_SC_AUTH_REQUIRED ; +return -EPROTONOSUPPORT ; }static inline void nvme_auth_free (struct nvme_ctrl *ctrl) {};#endif
Signature · 1 line
2.44.0
PATCH v7 4/5 nvme-fabrics: short-circuit reconnect retries
In reply to Daniel WagnerFrom: Hannes Reinecke <[email protected]>
Returning a nvme status from nvme_tcp_setup_ctrl() indicates that the association was established and we have received a status from the controller; consequently we should honour the DNR bit. If not any future reconnect attempts will just return the same error, so we can short-circuit the reconnect attempts and fail the connection directly.
--- drivers/nvme/host/fabrics.c | 14 +++++++++++++- drivers/nvme/host/fabrics.h | 2 +- drivers/nvme/host/fc.c | 4 +--- drivers/nvme/host/rdma.c | 19 ++++++++++++------- drivers/nvme/host/tcp.c | 22 ++++++++++++++-------- 5 files changed, 41 insertions(+), 20 deletions(-)
drivers/nvme/host/fabrics.c +13 -1
@@ -559,8 +559,20 @@ int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid) }EXPORT_SYMBOL_GPL (nvmf_connect_io_queue); -bool nvmf_should_reconnect (struct nvme_ctrl *ctrl) +/* +* Evaluate the status information returned by the transport in order to decided +* if a reconnect attempt should be scheduled. +* +* Do not retry when: +* +* - the DNR bit is set and the specification states no further connect +* attempts with the same set of paramenters should be attempted. +*/ +bool nvmf_should_reconnect (struct nvme_ctrl *ctrl,int status) { +if (status >0 && (status &NVME_SC_DNR )) +return false; +if (ctrl->opts->max_reconnects ==-1 || ctrl->nr_reconnects < ctrl->opts->max_reconnects)return true;
drivers/nvme/host/fabrics.h +1 -1
@@ -223,7 +223,7 @@ int nvmf_register_transport(struct nvmf_transport_ops *ops);void nvmf_unregister_transport (struct nvmf_transport_ops *ops);void nvmf_free_options (struct nvmf_ctrl_options *opts);int nvmf_get_address (struct nvme_ctrl *ctrl,char *buf,int size); -bool nvmf_should_reconnect (struct nvme_ctrl *ctrl); +bool nvmf_should_reconnect (struct nvme_ctrl *ctrl,int status);bool nvmf_ip_options_match (struct nvme_ctrl *ctrl,struct nvmf_ctrl_options *opts); void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
drivers/nvme/host/fc.c +1 -3
@@ -3310,12 +3310,10 @@ nvme_fc_reconnect_or_delete(struct nvme_fc_ctrl *ctrl, int status)dev_info (ctrl->ctrl.device,"NVME-FC{%d}: reset: Reconnect attempt failed (%d)\n" , ctrl->cnum, status); -if (status >0 && (status &NVME_SC_DNR )) - recon = false; }else if (time_after_eq (jiffies, rport->dev_loss_end)) recon = false; -if (recon &&nvmf_should_reconnect (&ctrl->ctrl)) { +if (recon &&nvmf_should_reconnect (&ctrl->ctrl, status)) {if (portptr->port_state ==FC_OBJSTATE_ONLINE )dev_info (ctrl->ctrl.device, "NVME-FC{%d}: Reconnect attempt in %ld "
drivers/nvme/host/rdma.c +12 -7
@@ -982,7 +982,8 @@ static void nvme_rdma_free_ctrl(struct nvme_ctrl *nctrl)kfree (ctrl); } -static void nvme_rdma_reconnect_or_remove (struct nvme_rdma_ctrl *ctrl) +static void nvme_rdma_reconnect_or_remove (struct nvme_rdma_ctrl *ctrl, +int status) {enum nvme_ctrl_state state =nvme_ctrl_state (&ctrl->ctrl);
@@ -992,7 +993,7 @@ static void nvme_rdma_reconnect_or_remove(struct nvme_rdma_ctrl *ctrl)return ; } -if (nvmf_should_reconnect (&ctrl->ctrl)) { +if (nvmf_should_reconnect (&ctrl->ctrl, status)) {dev_info (ctrl->ctrl.device,"Reconnecting in %d seconds...\n" , ctrl->ctrl.opts->reconnect_delay); queue_delayed_work(nvme_wq, &ctrl->reconnect_work,
@@ -1104,10 +1105,12 @@ static void nvme_rdma_reconnect_ctrl_work(struct work_struct *work) {struct nvme_rdma_ctrl *ctrl =container_of (to_delayed_work (work), struct nvme_rdma_ctrl, reconnect_work); +int ret; ++ctrl->ctrl.nr_reconnects; -if (nvme_rdma_setup_ctrl (ctrl, false)) + ret =nvme_rdma_setup_ctrl (ctrl, false); +if (ret) goto requeue; dev_info(ctrl->ctrl.device, "Successfully reconnected (%d attempts)\n",
@@ -1120,7 +1123,7 @@ static void nvme_rdma_reconnect_ctrl_work(struct work_struct *work) requeue:dev_info (ctrl->ctrl.device,"Failed reconnect attempt %d\n" , ctrl->ctrl.nr_reconnects); -nvme_rdma_reconnect_or_remove (ctrl); +nvme_rdma_reconnect_or_remove (ctrl, ret); }static void nvme_rdma_error_recovery_work (struct work_struct *work)
@@ -1145,7 +1148,7 @@ static void nvme_rdma_error_recovery_work(struct work_struct *work)return ; } -nvme_rdma_reconnect_or_remove (ctrl); +nvme_rdma_reconnect_or_remove (ctrl,0 ); }static void nvme_rdma_error_recovery (struct nvme_rdma_ctrl *ctrl)
@@ -2169,6 +2172,7 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work) {struct nvme_rdma_ctrl *ctrl = container_of(work, struct nvme_rdma_ctrl, ctrl.reset_work); +int ret;nvme_stop_ctrl (&ctrl->ctrl);nvme_rdma_shutdown_ctrl (ctrl, false);
@@ -2179,14 +2183,15 @@ static void nvme_rdma_reset_ctrl_work(struct work_struct *work)return ; } -if (nvme_rdma_setup_ctrl (ctrl, false)) + ret =nvme_rdma_setup_ctrl (ctrl, false); +if (ret) goto out_fail;return ; out_fail: ++ctrl->ctrl.nr_reconnects; -nvme_rdma_reconnect_or_remove (ctrl); +nvme_rdma_reconnect_or_remove (ctrl, ret); }static const struct nvme_ctrl_ops nvme_rdma_ctrl_ops = {
drivers/nvme/host/tcp.c +14 -8
@@ -2155,7 +2155,8 @@ static void nvme_tcp_teardown_io_queues(struct nvme_ctrl *ctrl,nvme_tcp_destroy_io_queues (ctrl, remove); } -static void nvme_tcp_reconnect_or_remove (struct nvme_ctrl *ctrl) +static void nvme_tcp_reconnect_or_remove (struct nvme_ctrl *ctrl, +int status) {enum nvme_ctrl_state state =nvme_ctrl_state (ctrl);
@@ -2165,13 +2166,14 @@ static void nvme_tcp_reconnect_or_remove(struct nvme_ctrl *ctrl)return ; } -if (nvmf_should_reconnect (ctrl)) { +if (nvmf_should_reconnect (ctrl, status)) {dev_info (ctrl->device,"Reconnecting in %d seconds...\n" , ctrl->opts->reconnect_delay);queue_delayed_work (nvme_wq, &to_tcp_ctrl (ctrl)->connect_work, ctrl->opts->reconnect_delay *HZ ); }else { -dev_info (ctrl->device,"Removing controller...\n" ); +dev_info (ctrl->device,"Removing controller (%d)...\n" , + status);nvme_delete_ctrl (ctrl); } }
@@ -2252,10 +2254,12 @@ static void nvme_tcp_reconnect_ctrl_work(struct work_struct *work)struct nvme_tcp_ctrl *tcp_ctrl =container_of (to_delayed_work (work), struct nvme_tcp_ctrl, connect_work);struct nvme_ctrl *ctrl = &tcp_ctrl->ctrl; +int ret; ++ctrl->nr_reconnects; -if (nvme_tcp_setup_ctrl (ctrl, false)) + ret =nvme_tcp_setup_ctrl (ctrl, false); +if (ret) goto requeue; dev_info(ctrl->device, "Successfully reconnected (%d attempt)\n",
@@ -2268,7 +2272,7 @@ static void nvme_tcp_reconnect_ctrl_work(struct work_struct *work) requeue:dev_info (ctrl->device,"Failed reconnect attempt %d\n" , ctrl->nr_reconnects); -nvme_tcp_reconnect_or_remove (ctrl); +nvme_tcp_reconnect_or_remove (ctrl, ret); }static void nvme_tcp_error_recovery_work (struct work_struct *work)
@@ -2295,7 +2299,7 @@ static void nvme_tcp_error_recovery_work(struct work_struct *work)return ; } -nvme_tcp_reconnect_or_remove (ctrl); +nvme_tcp_reconnect_or_remove (ctrl,0 ); }static void nvme_tcp_teardown_ctrl (struct nvme_ctrl *ctrl,bool shutdown)
@@ -2315,6 +2319,7 @@ static void nvme_reset_ctrl_work(struct work_struct *work) {struct nvme_ctrl *ctrl = container_of(work, struct nvme_ctrl, reset_work); +int ret;nvme_stop_ctrl (ctrl);nvme_tcp_teardown_ctrl (ctrl, false);
@@ -2328,14 +2333,15 @@ static void nvme_reset_ctrl_work(struct work_struct *work)return ; } -if (nvme_tcp_setup_ctrl (ctrl, false)) + ret =nvme_tcp_setup_ctrl (ctrl, false); +if (ret) goto out_fail;return ; out_fail: ++ctrl->nr_reconnects; -nvme_tcp_reconnect_or_remove (ctrl); +nvme_tcp_reconnect_or_remove (ctrl, ret); }static void nvme_tcp_stop_ctrl (struct nvme_ctrl *ctrl)
Signature · 1 line
2.44.0
PATCH v7 5/5 nvme: do not retry authentication failures
In reply to Daniel WagnerWhen the key is invalid there is no point in retrying. Because the auth code returns kernel error codes only, we can't test on the DNR bit.
--- drivers/nvme/host/auth.c | 6 +++--- drivers/nvme/host/fabrics.c | 6 ++++++ 2 files changed, 9 insertions(+), 3 deletions(-)
drivers/nvme/host/auth.c +3 -3
@@ -730,7 +730,7 @@ static void nvme_queue_auth_work(struct work_struct *work) NVME_AUTH_DHCHAP_MESSAGE_CHALLENGE);if (ret) { chap->status = ret; - chap->error = -ECONNREFUSED ; + chap->error = -EKEYREJECTED ;return ; }
@@ -797,7 +797,7 @@ static void nvme_queue_auth_work(struct work_struct *work) NVME_AUTH_DHCHAP_MESSAGE_SUCCESS1);if (ret) { chap->status = ret; - chap->error = -ECONNREFUSED ; + chap->error = -EKEYREJECTED ;return ; }
@@ -818,7 +818,7 @@ static void nvme_queue_auth_work(struct work_struct *work) ret =nvme_auth_process_dhchap_success1 (ctrl, chap);if (ret) {/* Controller authentication failed */ - chap->error = -ECONNREFUSED ; + chap->error = -EKEYREJECTED ; goto fail2; }
drivers/nvme/host/fabrics.c +6 -0
@@ -567,12 +567,18 @@ EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);* * - the DNR bit is set and the specification states no further connect * attempts with the same set of paramenters should be attempted. +* +* - when the authentication attempt fails, because the key was invalid. +* This error code is set on the host side. */ bool nvmf_should_reconnect (struct nvme_ctrl *ctrl,int status) {if (status >0 && (status &NVME_SC_DNR ))return false; +if (status == -EKEYREJECTED ) +return false; +if (ctrl->opts->max_reconnects ==-1 || ctrl->nr_reconnects < ctrl->opts->max_reconnects)return true;
Signature · 1 line
2.44.0
Sagi Grimberg
maintainerRe: [PATCH v7 5/5] nvme: do not retry authentication failures
Quoted 2 lines · Daniel Wagner
When the key is invalid there is no point in retrying. Because the auth code returns kernel error codes only, we can't test on the DNR bit.
I still think that this patch has issues in the sense that in certain cases the controller can suddenly disappear.
Keith Busch
maintainerQuoted 4 lines · Daniel Wagner
I've splitted the last patch into the hopefully non controversial part 'do not retry when DNR is set' and the 'don't retry auth failures'. I hope we can get at least the first few patches in and have a lively discussion on the final patch at LSF.
I've dropped the previous series from the nvme-6.9 branch. This series has been added to the newly created nvme-6.10 branch.