Skip to contentSkip to conversation

nvme-fabrics: reject I/O to offline device

PATCH Superseded Victor Gladkov 5 participants 14 messages Dec 1, 2019 – Dec 18, 2019 NVMe

v1 of 11 Superseded Latest is v11 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nvme-fabrics: reject I/O to offline device

Issue Description:
Commands get stuck while Host NVMe controller (TCP or RDMA) is in reconnect state.
NVMe controller enters into reconnect state when it loses connection with the target. It tries to reconnect every 10 seconds (default) until successful reconnection or until reconnect time-out is reached. The default reconnect time out is 10 minutes. This behavior is different than ISCSI where Commands during reconnect state returns with the following error: "rejecting I/O to offline device"

Fix Description:
Added a kernel module parameter "nvmef_reconnect_failfast" for nvme-fabrics module (default is true).
Interfere in the decision whether to queue IO command or retry IO command. The interface takes into account the controller reconnect state, in a way that during reconnect state, IO commands shall fail immediacy (default) or according to IO command timeout (depends on the module parameter value), and IO retry is prevented. As a result, commands do not get stuck in in reconnect state.

branch nvme-5.5

---

drivers/nvme/host/fabrics.c +5 -0
index 74b8818..ef89aff 100644
@@ -13,6 +13,10 @@
 #include "nvme.h"
 #include "fabrics.h"
 
+static bool nvmef_reconnect_failfast = 1;
+module_param_named(nvmef_reconnect_failfast, nvmef_reconnect_failfast, bool, S_IRUGO);
+MODULE_PARM_DESC(nvmef_reconnect_failfast, "failfast flag for I/O when controler is reconnecting, else use I/O command timeout (default true).");
+
 static LIST_HEAD(nvmf_transports);
 static DECLARE_RWSEM(nvmf_transports_rwsem);
 
@@ -549,6 +553,7 @@ blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
 {
        if (ctrl->state != NVME_CTRL_DELETING &&
            ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING && (((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout)) || nvmef_reconnect_failfast)) &&
            !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
                return BLK_STS_RESOURCE;
 

Regards,
Victor

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#1 lore raw
Quoted 4 lines · Victor Gladkov

Issue Description:
Commands get stuck while Host NVMe controller (TCP or RDMA) is in reconnect state.
NVMe controller enters into reconnect state when it loses connection with the target. It tries to reconnect every 10 seconds (default) until successful reconnection or until reconnect time-out is reached. The default reconnect time out is 10 minutes. This behavior is different than ISCSI where Commands during reconnect state returns with the following error: "rejecting I/O to offline device"

The values NVMF_DEF_RECONNECT_DELAY and NVMF_OPT_CTRL_LOSS_TMO are configurable using :-

  602         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
  603         { NVMF_OPT_CTRL_LOSS_TMO,       "ctrl_loss_tmo=%d"      }, 

Is it possible to get the same behavior with proper combination of above two values set ?

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#2 lore raw
In reply to Victor Gladkov
Quoted 8 lines · Victor Gladkov

Issue Description:
Commands get stuck while Host NVMe controller (TCP or RDMA) is in reconnect state.
NVMe controller enters into reconnect state when it loses connection with the target. It tries to reconnect every 10 seconds (default) until successful reconnection or until reconnect time-out is reached. The default reconnect time out is 10 minutes. This behavior is different than ISCSI where Commands during reconnect state returns with the following error: "rejecting I/O to offline device"

Fix Description:
Added a kernel module parameter "nvmef_reconnect_failfast" for nvme-fabrics module (default is true).
Interfere in the decision whether to queue IO command or retry IO command. The interface takes into account the controller reconnect state, in a way that during reconnect state, IO commands shall fail immediacy (default) or according to IO command timeout (depends on the module parameter value), and IO retry is prevented. As a result, commands do not get stuck in in reconnect state.

This the patch seems incorrect at least as described. Multipathing inherently will "fastfail" and send to other paths. So the only way something is "stuck" is if it's last path. If last path, we definitely don't want to prematurely release i/o before we've given the subsystem every opportunity to reconnect.

What I hear you saying is you don't like the kernel default controller-loss-timeout of 600s. What was designed, if you didn't like the kernel default, was to use the per-connection "--cltr-loss-tmo" option for "nvme connect".  The auto-connect scripts or the admin script that specifies the connection can set whatever value it likes.

If that seems hard to do, perhaps it's time to implement the options that allow for a config file to specify new values to be used on all connections, or on connections to specific subsystems, and so on. But I don't think the kernel needs to change.

-- james

Quoted 28 lines

branch nvme-5.5

---

drivers/nvme/host/fabrics.c +5 -0
index 74b8818..ef89aff 100644
@@ -13,6 +13,10 @@
  #include "nvme.h"
  #include "fabrics.h"
 
+static bool nvmef_reconnect_failfast = 1;
+module_param_named(nvmef_reconnect_failfast, nvmef_reconnect_failfast, bool, S_IRUGO);
+MODULE_PARM_DESC(nvmef_reconnect_failfast, "failfast flag for I/O when controler is reconnecting, else use I/O command timeout (default true).");
+
  static LIST_HEAD(nvmf_transports);
  static DECLARE_RWSEM(nvmf_transports_rwsem);
 
@@ -549,6 +553,7 @@ blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
  {
         if (ctrl->state != NVME_CTRL_DELETING &&
             ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING && (((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout)) || nvmef_reconnect_failfast)) &&
             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
                 return BLK_STS_RESOURCE;
 
Quoted 8 lines

Regards,
Victor

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#3 lore raw
Quoted 32 lines · James Smart
Quoted 2 lines · Victor Gladkov

Issue Description:
Commands get stuck while Host NVMe controller (TCP or RDMA) is in

reconnect state.

Quoted 1 line

NVMe controller enters into reconnect state when it loses connection with

the target. It tries to reconnect every 10 seconds (default) until successful reconnection or until reconnect time-out is reached. The default reconnect time out is 10 minutes.

Quoted 1 line

This behavior is different than ISCSI where Commands during reconnect

state returns with the following error: "rejecting I/O to offline device"

Quoted 3 lines

Fix Description:
Added a kernel module parameter "nvmef_reconnect_failfast" for nvme-

fabrics module (default is true).

Quoted 1 line

Interfere in the decision whether to queue IO command or retry IO

command. The interface takes into account the controller reconnect state, in a way that during reconnect state, IO commands shall fail immediacy (default) or according to IO command timeout (depends on the module parameter value), and IO retry is prevented. As a result, commands do not get stuck in in reconnect state.

This the patch seems incorrect at least as described. Multipathing inherently will "fastfail" and send to other paths. So the only way something is "stuck" is if it's last path. If last path, we definitely don't want to prematurely release i/o before we've given the subsystem every opportunity to reconnect.

What I hear you saying is you don't like the kernel default controller-loss- timeout of 600s. What was designed, if you didn't like the kernel default, was to use the per-connection "--cltr-loss-tmo"
option for "nvme connect".  The auto-connect scripts or the admin script that specifies the connection can set whatever value it likes.

Quoted 6 lines

If that seems hard to do, perhaps it's time to implement the options that allow for a config file to specify new values to be used on all connections, or on connections to specific subsystems, and so on. But I don't think the kernel needs to change.

-- james

James, thank you for the suggestion.
But let me explain it differently.

Applications are expecting commands to complete with success or error
within a certain timeout (30 seconds by default).
The NVMe host is enforcing that timeout while it is connected, never the less,
during reconnection, the timeout is not enforced and commands may get stuck for a long period or even forever.

The controller-loss-timeout should not affect IO timeout policy, these are two different policies.

Regards,
Victor
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#4 lore raw
Quoted 10 lines · Victor Gladkov
Quoted 9 lines · James Smart

O
James, thank you for the suggestion.
But let me explain it differently.

Applications are expecting commands to complete with success or error
within a certain timeout (30 seconds by default).
The NVMe host is enforcing that timeout while it is connected, never the less,
during reconnection, the timeout is not enforced and commands may get stuck for a long period or even forever.

Quoted 4 lines
Quoted 4 lines

The controller-loss-timeout should not affect IO timeout policy, these are two different policies.

Regards,
Victor

Ok - which says what does make sense to add is the portion:

  !(ctrl->state == NVME_CTRL_CONNECTING && ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout)))

But I don't think we need the failfast flag.

-- james

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#5 lore raw
Quoted 7 lines · James Smart
Quoted 3 lines · Victor Gladkov
Quoted 2 lines · James Smart

O
The controller-loss-timeout should not affect IO timeout policy, these are

two different policies.

Quoted 8 lines

Ok - which says what does make sense to add is the portion:

   !(ctrl->state == NVME_CTRL_CONNECTING && ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout)))

But I don't think we need the failfast flag.

-- james

OK. I think, it's good enough.

This is updated patch:

---

drivers/nvme/host/fabrics.c +2 -0
index 74b8818..b58abc1 100644
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
 {
        if (ctrl->state != NVME_CTRL_DELETING &&
            ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING &&
+            ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout))) &&
            !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
                return BLK_STS_RESOURCE;
 

---

Regards,
Victor
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#6 lore raw
Quoted 24 lines · Victor Gladkov
Quoted 7 lines · James Smart
Quoted 3 lines · Victor Gladkov
Quoted 2 lines · James Smart

O
The controller-loss-timeout should not affect IO timeout policy, these are

two different policies.
Ok - which says what does make sense to add is the portion:

Quoted 6 lines
    !(ctrl->state == NVME_CTRL_CONNECTING && ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout)))

But I don't think we need the failfast flag.

-- james

OK. I think, it's good enough.

This is updated patch:

---

drivers/nvme/host/fabrics.c +0 -0
index 74b8818..b58abc1 100644

drivers/nvme/host/fabrics.c blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl, · line 556

Quoted 8 lines
drivers/nvme/host/fabrics.c +2 -0
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
  {
         if (ctrl->state != NVME_CTRL_DELETING &&
             ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING &&
+            ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout))) &&
             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
                 return BLK_STS_RESOURCE;
 

Did you test this to ensure it's doing what you expect. I'm not sure that all the timers are set right at this point. Most I/O's timeout from a deadline time stamped at blk_mq_start_request(). But that routine is actually called by the transports post the nvmf_check_ready/fail_nonready calls.  E.g. the io is not yet in flight, thus queued, and the blk-mq internal queuing doesn't count against the io timeout.  I can't see anything that guarantees start_time_ns is set.

-- james

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

maintainer
#7 lore raw
Quoted 11 lines
Quoted 11 lines

---

drivers/nvme/host/fabrics.c +0 -0
index 74b8818..b58abc1 100644
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct 
 nvme_ctrl *ctrl,
   {

         if (ctrl->state != NVME_CTRL_DELETING &&
             ctrl->state != NVME_CTRL_DEAD &&

Quoted 8 lines
Quoted 6 lines

+           !(ctrl->state == NVME_CTRL_CONNECTING &&
+            ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout))) &&
             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH)) return BLK_STS_RESOURCE;

Did you test this to ensure it's doing what you expect. I'm not sure that all the timers are set right at this point. Most I/O's timeout from a deadline time stamped at blk_mq_start_request(). But that routine is actually called by the transports post the nvmf_check_ready/fail_nonready calls.  E.g. the io is not yet in flight, thus queued, and the blk-mq internal queuing doesn't count against the io timeout.  I can't see anything that guarantees start_time_ns is set.

I'm not sure this behavior for failing I/O always desired? some consumers would actually not want the I/O to fail prematurely if we are not multipathing...

I think we need a fail_fast_tmo set in when establishing the controller to get it right.

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#8 lore raw
Quoted 20 lines · Sagi Grimberg
Quoted 11 lines
Quoted 11 lines

---

drivers/nvme/host/fabrics.c +0 -0
index 74b8818..b58abc1 100644
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct 
 nvme_ctrl *ctrl,
   {

         if (ctrl->state != NVME_CTRL_DELETING &&
             ctrl->state != NVME_CTRL_DEAD &&

Quoted 8 lines
Quoted 6 lines

+           !(ctrl->state == NVME_CTRL_CONNECTING &&
+            ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout))) &&
             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH)) return BLK_STS_RESOURCE;

Did you test this to ensure it's doing what you expect. I'm not sure that all the timers are set right at this point. Most I/O's timeout from a deadline time stamped at blk_mq_start_request(). But that routine is actually called by the transports post the nvmf_check_ready/fail_nonready calls.  E.g. the io is not yet in flight, thus queued, and the blk-mq internal queuing doesn't count against the io timeout.  I can't see anything that guarantees start_time_ns is set.

Quoted 7 lines

I'm not sure this behavior for failing I/O always desired? some consumers would actually not want the I/O to fail prematurely if we are not multipathing...

I think we need a fail_fast_tmo set in when establishing the controller to get it right.

Agreed. This whole patch looks like someone is trying to reimplement fast_io_fail_tmo / dev_loss_tmo. As we're moving into unreliable fabrics I guess we'll need a similar mechanism.

Cheers,

Hannes

Signature · 9 lines
Dr. Hannes Reinecke            Teamlead Storage & Networking
[email protected]                               +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#9 lore raw
Quoted 37 lines · Hannes Reinecke
Quoted 28 lines · Sagi Grimberg
Quoted 19 lines
Quoted 9 lines

---

(quoted diff) +0 -0

b/drivers/nvme/host/fabrics.c index 74b8818..b58abc1 100644

drivers/nvme/host/fabrics.c +0 -0
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct
 nvme_ctrl *ctrl,
   {
Quoted 9 lines

         if (ctrl->state != NVME_CTRL_DELETING && ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING && +            ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout))) && !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
                 return BLK_STS_RESOURCE;

Quoted 8 lines

Did you test this to ensure it's doing what you expect. I'm not sure that all the timers are set right at this point. Most I/O's timeout from a deadline time stamped at blk_mq_start_request(). But that routine is actually called by the transports post the nvmf_check_ready/fail_nonready calls.  E.g. the io is not yet in flight, thus queued, and the blk-mq internal queuing doesn't count against the io timeout.  I can't see anything that guarantees start_time_ns is set.

Quoted 7 lines

I'm not sure this behavior for failing I/O always desired? some consumers would actually not want the I/O to fail prematurely if we are not multipathing...

I think we need a fail_fast_tmo set in when establishing the controller to get it right.

Quoted 7 lines

Agreed. This whole patch looks like someone is trying to reimplement fast_io_fail_tmo / dev_loss_tmo.
As we're moving into unreliable fabrics I guess we'll need a similar mechanism.

Cheers,

Hannes

Following your suggestions, I added a new session parameter called "fast_fail_tmo".
The timeout is measured in seconds from the controller reconnect, any command beyond that timeout is rejected. The new parameter value may be passed during ‘connect’, and its default value is 30 seconds. A value of -1 means no timeout (in similar to current behavior).

---

drivers/nvme/host/fabrics.c +26 -3
index 74b8818..ed6b911 100644
@@ -406,6 +406,7 @@
 	}
 
 	ctrl->cntlid = le16_to_cpu(res.u16);
+	ctrl->start_reconnect_ns = ktime_get_ns();
 
 out_free_data:
 	kfree(data);
@@ -474,8 +475,12 @@
 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
 {
 	if (ctrl->opts->max_reconnects == -1 ||
-	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
+	    ctrl->nr_reconnects < ctrl->opts->max_reconnects){
+		if(ctrl->nr_reconnects == 0)
+			ctrl->start_reconnect_ns = ktime_get_ns();
+
 		return true;
+	}
 
 	return false;
 }
@@ -549,6 +554,8 @@
 {
 	if (ctrl->state != NVME_CTRL_DELETING &&
 	    ctrl->state != NVME_CTRL_DEAD &&
+            !(ctrl->state == NVME_CTRL_CONNECTING && ctrl->opts->fail_fast_tmo_ns >= 0 &&
+            ((ktime_get_ns() - ctrl->start_reconnect_ns) >  ctrl->opts->fail_fast_tmo_ns)) &&
 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 		return BLK_STS_RESOURCE;
 
@@ -612,6 +619,7 @@
 	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
 	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
 	{ NVMF_OPT_TOS,			"tos=%d"		},
+	{ NVMF_OPT_FAIL_FAST_TMO,	"fail_fast_tmo=%d"	},
 	{ NVMF_OPT_ERR,			NULL			}
 };
 
@@ -623,6 +631,7 @@
 	int token, ret = 0;
 	size_t nqnlen  = 0;
 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
+	int fail_fast_tmo = NVMF_DEF_FAIL_FAST_TMO;
 	uuid_t hostid;
 
 	/* Set defaults */
@@ -751,6 +760,16 @@
 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
 			ctrl_loss_tmo = token;
 			break;
+		case NVMF_OPT_FAIL_FAST_TMO:
+			if (match_int(args, &token)) {
+				ret = -EINVAL;
+				goto out;
+			}
+
+			if (token < 0)
+				pr_warn("fail_fast_tmo < 0, I/O will wait for reconnect/loss controller\n");
+			fail_fast_tmo = token;
+			break;
 		case NVMF_OPT_HOSTNQN:
 			if (opts->host) {
 				pr_err("hostnqn already user-assigned: %s\n",
@@ -887,6 +906,11 @@
 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
 						opts->reconnect_delay);
 
+	if (fail_fast_tmo < 0)
+		opts->fail_fast_tmo_ns = -1;
+	else
+		opts->fail_fast_tmo_ns = fail_fast_tmo * NSEC_PER_SEC;
+
 	if (!opts->host) {
 		kref_get(&nvmf_default_host->ref);
 		opts->host = nvmf_default_host;
@@ -985,7 +1009,7 @@
 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
 				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
-				 NVMF_OPT_DISABLE_SQFLOW)
+				 NVMF_OPT_DISABLE_SQFLOW|NVMF_OPT_FAIL_FAST_TMO)
 
 static struct nvme_ctrl *
 nvmf_create_ctrl(struct device *dev, const char *buf)
@@ -1198,7 +1222,6 @@
 	class_destroy(nvmf_class);
 	nvmf_host_put(nvmf_default_host);
 
-	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
drivers/nvme/host/fabrics.h +5 -0
index 93f08d7..235c767 100644
@@ -15,6 +15,8 @@
 #define NVMF_DEF_RECONNECT_DELAY	10
 /* default to 600 seconds of reconnect attempts before giving up */
 #define NVMF_DEF_CTRL_LOSS_TMO		600
+/* default to 30 seconds of fail fast IO commands  */
+#define NVMF_DEF_FAIL_FAST_TMO		30
 
 /*
  * Define a host as seen by the target.  We allocate one at boot, but also
@@ -56,6 +58,7 @@
 	NVMF_OPT_NR_WRITE_QUEUES = 1 << 17,
 	NVMF_OPT_NR_POLL_QUEUES = 1 << 18,
 	NVMF_OPT_TOS		= 1 << 19,
+	NVMF_OPT_FAIL_FAST_TMO	= 1 << 20,
 };
 
 /**
@@ -89,6 +92,7 @@
  * @nr_write_queues: number of queues for write I/O
  * @nr_poll_queues: number of queues for polling I/O
  * @tos: type of service
+ * @fast_fail_tmo_ns: Fast I/O fail timeout in nanoseconds;
  */
 struct nvmf_ctrl_options {
 	unsigned		mask;
@@ -111,6 +115,7 @@
 	unsigned int		nr_write_queues;
 	unsigned int		nr_poll_queues;
 	int			tos;
+	u64			fail_fast_tmo_ns;
 };
 
 /*
drivers/nvme/host/nvme.h +1 -0
index 34ac79c..9b5f6d9 100644
@@ -282,6 +282,7 @@
 	u16 icdoff;
 	u16 maxcmd;
 	int nr_reconnects;
+	u64 start_reconnect_ns;
 	struct nvmf_ctrl_options *opts;
 
 	struct page *discard_page;

---
Regards,
Victor
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#10 lore raw
Quoted 53 lines · Victor Gladkov
Quoted 33 lines · Hannes Reinecke
Quoted 25 lines · Sagi Grimberg
Quoted 17 lines
Quoted 12 lines

---

(quoted diff) +0 -0

b/drivers/nvme/host/fabrics.c index 74b8818..b58abc1 100644

drivers/nvme/host/fabrics.c +2 -0
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct
 nvme_ctrl *ctrl,
   {
          if (ctrl->state != NVME_CTRL_DELETING &&
              ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING &&
+            ((ktime_get_ns() - rq->start_time_ns) >
Quoted 5 lines

jiffies_to_nsecs(rq->timeout))) &&
             !blk_noretry_request(rq) && !(rq->cmd_flags &
REQ_NVME_MPATH))
                 return BLK_STS_RESOURCE;

Quoted 8 lines

Did you test this to ensure it's doing what you expect. I'm not sure that all the timers are set right at this point. Most I/O's timeout from a deadline time stamped at blk_mq_start_request(). But that routine is actually called by the transports post the nvmf_check_ready/fail_nonready calls.  E.g. the io is not yet in flight, thus queued, and the blk-mq internal queuing doesn't count against the io timeout.  I can't see anything that guarantees start_time_ns is set.

Quoted 7 lines

I'm not sure this behavior for failing I/O always desired? some consumers would actually not want the I/O to fail prematurely if we are not multipathing...

I think we need a fail_fast_tmo set in when establishing the controller to get it right.

Quoted 7 lines

Agreed. This whole patch looks like someone is trying to reimplement fast_io_fail_tmo / dev_loss_tmo.
As we're moving into unreliable fabrics I guess we'll need a similar mechanism.

Cheers,

Hannes

Following your suggestions, I added a new session parameter called "fast_fail_tmo".
The timeout is measured in seconds from the controller reconnect, any command beyond that timeout is rejected. The new parameter value may be passed during ‘connect’, and its default value is 30 seconds. A value of -1 means no timeout (in similar to current behavior).

---

drivers/nvme/host/fabrics.c +0 -0
index 74b8818..ed6b911 100644

drivers/nvme/host/fabrics.c line 409

Quoted 4 lines
drivers/nvme/host/fabrics.c +1 -0
@@ -406,6 +406,7 @@
  	}
 
  	ctrl->cntlid = le16_to_cpu(res.u16);
+	ctrl->start_reconnect_ns = ktime_get_ns();

It doesn't seem correct to start the reconnect timer once a admin queue connect has succeeded. There may be lots of time prior when there is no connectivity and the timer hasn't been set and thus isn't running.

ctrl->start_reconnect_ns should be set when the ctrl transitions to NVME_CTRL_CONNECTING. Would have been nice to set the value in , but as that's in nvme_change_ctrl_state() in core, and as it's more than just changing state, I guess that means it has to be in each of the fabric transports as they make the call to change to the state.

note: if posting a new patch. Start it as a new patch thread, not a reply on this thread.

Quoted 4 lines
drivers/nvme/host/fabrics.c +0 -0
  out_free_data:
  	kfree(data);

drivers/nvme/host/fabrics.c line 483

Quoted 10 lines
drivers/nvme/host/fabrics.c +5 -1
@@ -474,8 +475,12 @@
  bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
  {
  	if (ctrl->opts->max_reconnects == -1 ||
-	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
+	    ctrl->nr_reconnects < ctrl->opts->max_reconnects){
+		if(ctrl->nr_reconnects == 0)
+			ctrl->start_reconnect_ns = ktime_get_ns();
+
  		return true;
+	}

This is making up for what I described above, but means start_reconnect_ns won't be set until after the first expiration of the reconnect attempt. So there's at least 1 window skipped with no timer.  If the transports just start the timer after the state transition, there is no need for this.

drivers/nvme/host/fabrics.c line 558

Quoted 9 lines
drivers/nvme/host/fabrics.c +2 -0
  	return false;
  }
@@ -549,6 +554,8 @@
  {
  	if (ctrl->state != NVME_CTRL_DELETING &&
  	    ctrl->state != NVME_CTRL_DEAD &&
+            !(ctrl->state == NVME_CTRL_CONNECTING && ctrl->opts->fail_fast_tmo_ns >= 0 &&
+            ((ktime_get_ns() - ctrl->start_reconnect_ns) >  ctrl->opts->fail_fast_tmo_ns)) &&

This is hard to read - I'd prefer just making a small inline helper that is called.
As for the logic: looks good. I'm a little concerned about a race on CONNECTING and start_reconnect_ns being set right as there's no locks (and we don't want to add them), but it's very very very small.

prefer it's just fail_fast_tmo_ns be unsigned, with 0 being disabled/no timeout.

Quoted 21 lines
drivers/nvme/host/fabrics.c +2 -0
  	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
  		return BLK_STS_RESOURCE;
 
@@ -612,6 +619,7 @@
  	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
  	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
  	{ NVMF_OPT_TOS,			"tos=%d"		},
+	{ NVMF_OPT_FAIL_FAST_TMO,	"fail_fast_tmo=%d"	},
  	{ NVMF_OPT_ERR,			NULL			}
  };
 
@@ -623,6 +631,7 @@
  	int token, ret = 0;
  	size_t nqnlen  = 0;
  	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
+	int fail_fast_tmo = NVMF_DEF_FAIL_FAST_TMO;
  	uuid_t hostid;
 
  	/* Set defaults */
@@ -751,6 +760,16 @@
  				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");

drivers/nvme/host/fabrics.c line 770

Quoted 10 lines
drivers/nvme/host/fabrics.c +8 -0
  			ctrl_loss_tmo = token;
  			break;
+		case NVMF_OPT_FAIL_FAST_TMO:
+			if (match_int(args, &token)) {
+				ret = -EINVAL;
+				goto out;
+			}
+
+			if (token < 0)
+				pr_warn("fail_fast_tmo < 0, I/O will wait for reconnect/loss controller\n");

see above. Prefer <=0 be no timeout.
do you want a warning if larger than controller loss tmo ?

drivers/nvme/host/fabrics.c line 910

Quoted 11 lines
drivers/nvme/host/fabrics.c +4 -0
+			fail_fast_tmo = token;
+			break;
  		case NVMF_OPT_HOSTNQN:
  			if (opts->host) {
  				pr_err("hostnqn already user-assigned: %s\n",
@@ -887,6 +906,11 @@
  		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
  						opts->reconnect_delay);
 
+	if (fail_fast_tmo < 0)
+		opts->fail_fast_tmo_ns = -1;

would prefer it be set to 0 to mean disabled.

Quoted 16 lines
drivers/nvme/host/fabrics.c +4 -1
+	else
+		opts->fail_fast_tmo_ns = fail_fast_tmo * NSEC_PER_SEC;
+
  	if (!opts->host) {
  		kref_get(&nvmf_default_host->ref);
  		opts->host = nvmf_default_host;
@@ -985,7 +1009,7 @@
  #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
  				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
  				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
-				 NVMF_OPT_DISABLE_SQFLOW)
+				 NVMF_OPT_DISABLE_SQFLOW|NVMF_OPT_FAIL_FAST_TMO)
 
  static struct nvme_ctrl *
  nvmf_create_ctrl(struct device *dev, const char *buf)

drivers/nvme/host/fabrics.c line 1201

Quoted 4 lines
drivers/nvme/host/fabrics.c +0 -1
@@ -1198,7 +1222,6 @@
  	class_destroy(nvmf_class);
  	nvmf_host_put(nvmf_default_host);
 
-	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);

Odd line to have in this patch.

Quoted 56 lines
drivers/nvme/host/fabrics.c +0 -0
  	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
  	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
  	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
drivers/nvme/host/fabrics.h +5 -0
index 93f08d7..235c767 100644
@@ -15,6 +15,8 @@
  #define NVMF_DEF_RECONNECT_DELAY	10
  /* default to 600 seconds of reconnect attempts before giving up */
  #define NVMF_DEF_CTRL_LOSS_TMO		600
+/* default to 30 seconds of fail fast IO commands  */
+#define NVMF_DEF_FAIL_FAST_TMO		30
 
  /*
   * Define a host as seen by the target.  We allocate one at boot, but also
@@ -56,6 +58,7 @@
  	NVMF_OPT_NR_WRITE_QUEUES = 1 << 17,
  	NVMF_OPT_NR_POLL_QUEUES = 1 << 18,
  	NVMF_OPT_TOS		= 1 << 19,
+	NVMF_OPT_FAIL_FAST_TMO	= 1 << 20,
  };
 
  /**
@@ -89,6 +92,7 @@
   * @nr_write_queues: number of queues for write I/O
   * @nr_poll_queues: number of queues for polling I/O
   * @tos: type of service
+ * @fast_fail_tmo_ns: Fast I/O fail timeout in nanoseconds;
   */
  struct nvmf_ctrl_options {
  	unsigned		mask;
@@ -111,6 +115,7 @@
  	unsigned int		nr_write_queues;
  	unsigned int		nr_poll_queues;
  	int			tos;
+	u64			fail_fast_tmo_ns;
  };
 
  /*
drivers/nvme/host/nvme.h +1 -1
index 34ac79c..9b5f6d9 100644
@@ -282,6 +282,7 @@
  	u16 icdoff;
  	u16 maxcmd;
  	int nr_reconnects;
+	u64 start_reconnect_ns;
  	struct nvmf_ctrl_options *opts;
 
  	struct page *discard_page;
 
---
Quoted 2 lines

Regards,
Victor

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

maintainer
#11 lore raw
In reply to Victor Gladkov
Quoted 42 lines · Victor Gladkov
Quoted 41 lines · Hannes Reinecke
Quoted 28 lines · Sagi Grimberg
Quoted 19 lines
Quoted 13 lines

---

(quoted diff) +0 -0

b/drivers/nvme/host/fabrics.c index 74b8818..b58abc1 100644

drivers/nvme/host/fabrics.c +2 -0
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct
 nvme_ctrl *ctrl,
   {
          if (ctrl->state != NVME_CTRL_DELETING &&
              ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING &&
+            ((ktime_get_ns() - rq->start_time_ns) >
Quoted 5 lines

jiffies_to_nsecs(rq->timeout))) &&
             !blk_noretry_request(rq) && !(rq->cmd_flags &
REQ_NVME_MPATH))
                 return BLK_STS_RESOURCE;

Quoted 8 lines

Did you test this to ensure it's doing what you expect. I'm not sure that all the timers are set right at this point. Most I/O's timeout from a deadline time stamped at blk_mq_start_request(). But that routine is actually called by the transports post the nvmf_check_ready/fail_nonready calls.  E.g. the io is not yet in flight, thus queued, and the blk-mq internal queuing doesn't count against the io timeout.  I can't see anything that guarantees start_time_ns is set.

Quoted 7 lines

I'm not sure this behavior for failing I/O always desired? some consumers would actually not want the I/O to fail prematurely if we are not multipathing...

I think we need a fail_fast_tmo set in when establishing the controller to get it right.

Agreed. This whole patch looks like someone is trying to reimplement fast_io_fail_tmo / dev_loss_tmo.
As we're moving into unreliable fabrics I guess we'll need a similar mechanism.

Quoted 8 lines
Quoted 3 lines

Cheers,

Hannes

Following your suggestions, I added a new session parameter called "fast_fail_tmo".
The timeout is measured in seconds from the controller reconnect, any command beyond that timeout is rejected. The new parameter value may be passed during ‘connect’, and its default value is 30 seconds.

The default should be consistent with the existing behavior.

Quoted 30 lines

A value of -1 means no timeout (in similar to current behavior).

---

drivers/nvme/host/fabrics.c +6 -1
index 74b8818..ed6b911 100644
@@ -406,6 +406,7 @@
  	}
 
  	ctrl->cntlid = le16_to_cpu(res.u16);
+	ctrl->start_reconnect_ns = ktime_get_ns();
 
  out_free_data:
  	kfree(data);
@@ -474,8 +475,12 @@
  bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
  {
  	if (ctrl->opts->max_reconnects == -1 ||
-	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
+	    ctrl->nr_reconnects < ctrl->opts->max_reconnects){
+		if(ctrl->nr_reconnects == 0)
+			ctrl->start_reconnect_ns = ktime_get_ns();
+
  		return true;
+	}
 
  	return false;
  }

drivers/nvme/host/fabrics.c line 560

Quoted 7 lines
drivers/nvme/host/fabrics.c +2 -0
@@ -549,6 +554,8 @@
  {
  	if (ctrl->state != NVME_CTRL_DELETING &&
  	    ctrl->state != NVME_CTRL_DEAD &&
+            !(ctrl->state == NVME_CTRL_CONNECTING && ctrl->opts->fail_fast_tmo_ns >= 0 &&
+            ((ktime_get_ns() - ctrl->start_reconnect_ns) >  ctrl->opts->fail_fast_tmo_ns)) &&
  	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
  		return BLK_STS_RESOURCE;

I cannot comprehend what is going on here...

We should have a dedicated delayed_work that transitions the controller to a FAIL_FAST state and cancels the inflight requests again. This work should be triggered when the error is detected.

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#12 lore raw
Quoted 98 lines · Sagi Grimberg
Quoted 42 lines · Victor Gladkov
Quoted 41 lines · Hannes Reinecke
Quoted 28 lines · Sagi Grimberg
Quoted 19 lines
Quoted 9 lines

---

(quoted diff) +0 -0

b/drivers/nvme/host/fabrics.c index 74b8818..b58abc1 100644

drivers/nvme/host/fabrics.c +0 -0
@@ -549,6 +549,8 @@ blk_status_t nvmf_fail_nonready_command(struct
 nvme_ctrl *ctrl,
    {
Quoted 9 lines

          if (ctrl->state != NVME_CTRL_DELETING && ctrl->state != NVME_CTRL_DEAD &&
+           !(ctrl->state == NVME_CTRL_CONNECTING && +            ((ktime_get_ns() - rq->start_time_ns) > jiffies_to_nsecs(rq->timeout))) && !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
                  return BLK_STS_RESOURCE;

Quoted 8 lines

Did you test this to ensure it's doing what you expect. I'm not sure that all the timers are set right at this point. Most I/O's timeout from a deadline time stamped at blk_mq_start_request(). But that routine is actually called by the transports post the nvmf_check_ready/fail_nonready calls.  E.g. the io is not yet in flight, thus queued, and the blk-mq internal queuing doesn't count against the io timeout.  I can't see anything that guarantees start_time_ns is set.

Quoted 7 lines

I'm not sure this behavior for failing I/O always desired? some consumers would actually not want the I/O to fail prematurely if we are not multipathing...

I think we need a fail_fast_tmo set in when establishing the controller to get it right.

Agreed. This whole patch looks like someone is trying to reimplement fast_io_fail_tmo / dev_loss_tmo.
As we're moving into unreliable fabrics I guess we'll need a similar mechanism.

Quoted 8 lines
Quoted 3 lines

Cheers,

Hannes

Following your suggestions, I added a new session parameter called "fast_fail_tmo".
The timeout is measured in seconds from the controller reconnect, any command beyond that timeout is rejected. The new parameter value may be passed during ‘connect’, and its default value is 30 seconds.

The default should be consistent with the existing behavior.

Quoted 31 lines

A value of -1 means no timeout (in similar to current behavior).

---

drivers/nvme/host/fabrics.c +1 -0
index 74b8818..ed6b911 100644
@@ -406,6 +406,7 @@
       }
 
       ctrl->cntlid = le16_to_cpu(res.u16);
+    ctrl->start_reconnect_ns = ktime_get_ns();
 

  out_free_data:
      kfree(data);

drivers/nvme/host/fabrics.c +0 -0
@@ -474,8 +475,12 @@
   bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
   {

      if (ctrl->opts->max_reconnects == -1 ||
-        ctrl->nr_reconnects < ctrl->opts->max_reconnects) +        ctrl->nr_reconnects < ctrl->opts->max_reconnects){ +        if(ctrl->nr_reconnects == 0)
+            ctrl->start_reconnect_ns = ktime_get_ns(); +
          return true;
+    }

      return false;
  }

drivers/nvme/host/fabrics.c +0 -0
@@ -549,6 +554,8 @@
   {
Quoted 8 lines

      if (ctrl->state != NVME_CTRL_DELETING &&
          ctrl->state != NVME_CTRL_DEAD &&
+            !(ctrl->state == NVME_CTRL_CONNECTING && ctrl->opts->fail_fast_tmo_ns >= 0 &&
+            ((ktime_get_ns() - ctrl->start_reconnect_ns) > ctrl->opts->fail_fast_tmo_ns)) &&
          !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH)) return BLK_STS_RESOURCE;

I cannot comprehend what is going on here...

Quoted 3 lines

We should have a dedicated delayed_work that transitions the controller to a FAIL_FAST state and cancels the inflight requests again. This work should be triggered when the error is detected.

I hope you're not suggesting a FAILFAST state.  No new controller state is needed.

I do agree that managing the time since transitioning to CONNECTING can be handled better and can address "abort all now" rather than waiting for retries to kick in.

In other words:
Add a controller flag of "failfast_expired"
When entering CONNECTING, schedule a delayed work item based on failfast timeout value.
If transition out of CONNECTING, terminate delayed work item and ensure failfast_expired is false. If delayed work item expires: set "failfast_expired" flag to true. Run through all inflight ios and cancel them. Update nvmf_fail_nonready_command() (above) per above, but with check on "!ctrl->failfast_expired".

-- james

_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#13 lore raw
In reply to Hannes Reinecke
Quoted 5 lines · Victor Gladkov
Quoted 3 lines · Hannes Reinecke
Quoted 1 line · Sagi Grimberg

...............................................

Quoted 21 lines
Quoted 6 lines
Quoted 3 lines

I think we need a fail_fast_tmo set in when establishing the controller to get it right.

Agreed. This whole patch looks like someone is trying to reimplement fast_io_fail_tmo / dev_loss_tmo.
As we're moving into unreliable fabrics I guess we'll need a similar

mechanism.

Quoted 4 lines

Cheers,

Hannes

Following your suggestions, I added a new session parameter called "fast_fail_tmo".
The timeout is measured in seconds from the controller reconnect, any command beyond that timeout is rejected.
The new parameter value may be passed during ‘connect’, and its default value is 30 seconds.
A value of -1 means no timeout (in similar to current behavior).

Quoted 3 lines

---
Regards,
Victor

James, Hannes & Sagi.
Can you please review last proposed patch?

Regards,
Victor
_______________________________________________
linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme