xref: /openbmc/linux/drivers/nvme/host/fc.c (revision bbecb07f)
1 /*
2  * Copyright (c) 2016 Avago Technologies.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful.
9  * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND WARRANTIES,
10  * INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, FITNESS FOR A
11  * PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE DISCLAIMED, EXCEPT TO
12  * THE EXTENT THAT SUCH DISCLAIMERS ARE HELD TO BE LEGALLY INVALID.
13  * See the GNU General Public License for more details, a copy of which
14  * can be found in the file COPYING included with this package
15  *
16  */
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 #include <linux/module.h>
19 #include <linux/parser.h>
20 #include <uapi/scsi/fc/fc_fs.h>
21 #include <uapi/scsi/fc/fc_els.h>
22 #include <linux/delay.h>
23 
24 #include "nvme.h"
25 #include "fabrics.h"
26 #include <linux/nvme-fc-driver.h>
27 #include <linux/nvme-fc.h>
28 
29 
30 /* *************************** Data Structures/Defines ****************** */
31 
32 
33 enum nvme_fc_queue_flags {
34 	NVME_FC_Q_CONNECTED = 0,
35 	NVME_FC_Q_LIVE,
36 };
37 
38 #define NVMEFC_QUEUE_DELAY	3		/* ms units */
39 
40 #define NVME_FC_DEFAULT_DEV_LOSS_TMO	60	/* seconds */
41 
42 struct nvme_fc_queue {
43 	struct nvme_fc_ctrl	*ctrl;
44 	struct device		*dev;
45 	struct blk_mq_hw_ctx	*hctx;
46 	void			*lldd_handle;
47 	size_t			cmnd_capsule_len;
48 	u32			qnum;
49 	u32			rqcnt;
50 	u32			seqno;
51 
52 	u64			connection_id;
53 	atomic_t		csn;
54 
55 	unsigned long		flags;
56 } __aligned(sizeof(u64));	/* alignment for other things alloc'd with */
57 
58 enum nvme_fcop_flags {
59 	FCOP_FLAGS_TERMIO	= (1 << 0),
60 	FCOP_FLAGS_RELEASED	= (1 << 1),
61 	FCOP_FLAGS_COMPLETE	= (1 << 2),
62 	FCOP_FLAGS_AEN		= (1 << 3),
63 };
64 
65 struct nvmefc_ls_req_op {
66 	struct nvmefc_ls_req	ls_req;
67 
68 	struct nvme_fc_rport	*rport;
69 	struct nvme_fc_queue	*queue;
70 	struct request		*rq;
71 	u32			flags;
72 
73 	int			ls_error;
74 	struct completion	ls_done;
75 	struct list_head	lsreq_list;	/* rport->ls_req_list */
76 	bool			req_queued;
77 };
78 
79 enum nvme_fcpop_state {
80 	FCPOP_STATE_UNINIT	= 0,
81 	FCPOP_STATE_IDLE	= 1,
82 	FCPOP_STATE_ACTIVE	= 2,
83 	FCPOP_STATE_ABORTED	= 3,
84 	FCPOP_STATE_COMPLETE	= 4,
85 };
86 
87 struct nvme_fc_fcp_op {
88 	struct nvme_request	nreq;		/*
89 						 * nvme/host/core.c
90 						 * requires this to be
91 						 * the 1st element in the
92 						 * private structure
93 						 * associated with the
94 						 * request.
95 						 */
96 	struct nvmefc_fcp_req	fcp_req;
97 
98 	struct nvme_fc_ctrl	*ctrl;
99 	struct nvme_fc_queue	*queue;
100 	struct request		*rq;
101 
102 	atomic_t		state;
103 	u32			flags;
104 	u32			rqno;
105 	u32			nents;
106 
107 	struct nvme_fc_cmd_iu	cmd_iu;
108 	struct nvme_fc_ersp_iu	rsp_iu;
109 };
110 
111 struct nvme_fc_lport {
112 	struct nvme_fc_local_port	localport;
113 
114 	struct ida			endp_cnt;
115 	struct list_head		port_list;	/* nvme_fc_port_list */
116 	struct list_head		endp_list;
117 	struct device			*dev;	/* physical device for dma */
118 	struct nvme_fc_port_template	*ops;
119 	struct kref			ref;
120 	atomic_t                        act_rport_cnt;
121 } __aligned(sizeof(u64));	/* alignment for other things alloc'd with */
122 
123 struct nvme_fc_rport {
124 	struct nvme_fc_remote_port	remoteport;
125 
126 	struct list_head		endp_list; /* for lport->endp_list */
127 	struct list_head		ctrl_list;
128 	struct list_head		ls_req_list;
129 	struct device			*dev;	/* physical device for dma */
130 	struct nvme_fc_lport		*lport;
131 	spinlock_t			lock;
132 	struct kref			ref;
133 	atomic_t                        act_ctrl_cnt;
134 	unsigned long			dev_loss_end;
135 } __aligned(sizeof(u64));	/* alignment for other things alloc'd with */
136 
137 enum nvme_fcctrl_flags {
138 	FCCTRL_TERMIO		= (1 << 0),
139 };
140 
141 struct nvme_fc_ctrl {
142 	spinlock_t		lock;
143 	struct nvme_fc_queue	*queues;
144 	struct device		*dev;
145 	struct nvme_fc_lport	*lport;
146 	struct nvme_fc_rport	*rport;
147 	u32			cnum;
148 
149 	bool			assoc_active;
150 	u64			association_id;
151 
152 	struct list_head	ctrl_list;	/* rport->ctrl_list */
153 
154 	struct blk_mq_tag_set	admin_tag_set;
155 	struct blk_mq_tag_set	tag_set;
156 
157 	struct delayed_work	connect_work;
158 
159 	struct kref		ref;
160 	u32			flags;
161 	u32			iocnt;
162 	wait_queue_head_t	ioabort_wait;
163 
164 	struct nvme_fc_fcp_op	aen_ops[NVME_NR_AEN_COMMANDS];
165 
166 	struct nvme_ctrl	ctrl;
167 };
168 
169 static inline struct nvme_fc_ctrl *
170 to_fc_ctrl(struct nvme_ctrl *ctrl)
171 {
172 	return container_of(ctrl, struct nvme_fc_ctrl, ctrl);
173 }
174 
175 static inline struct nvme_fc_lport *
176 localport_to_lport(struct nvme_fc_local_port *portptr)
177 {
178 	return container_of(portptr, struct nvme_fc_lport, localport);
179 }
180 
181 static inline struct nvme_fc_rport *
182 remoteport_to_rport(struct nvme_fc_remote_port *portptr)
183 {
184 	return container_of(portptr, struct nvme_fc_rport, remoteport);
185 }
186 
187 static inline struct nvmefc_ls_req_op *
188 ls_req_to_lsop(struct nvmefc_ls_req *lsreq)
189 {
190 	return container_of(lsreq, struct nvmefc_ls_req_op, ls_req);
191 }
192 
193 static inline struct nvme_fc_fcp_op *
194 fcp_req_to_fcp_op(struct nvmefc_fcp_req *fcpreq)
195 {
196 	return container_of(fcpreq, struct nvme_fc_fcp_op, fcp_req);
197 }
198 
199 
200 
201 /* *************************** Globals **************************** */
202 
203 
204 static DEFINE_SPINLOCK(nvme_fc_lock);
205 
206 static LIST_HEAD(nvme_fc_lport_list);
207 static DEFINE_IDA(nvme_fc_local_port_cnt);
208 static DEFINE_IDA(nvme_fc_ctrl_cnt);
209 
210 
211 
212 /*
213  * These items are short-term. They will eventually be moved into
214  * a generic FC class. See comments in module init.
215  */
216 static struct class *fc_class;
217 static struct device *fc_udev_device;
218 
219 
220 /* *********************** FC-NVME Port Management ************************ */
221 
222 static void __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *,
223 			struct nvme_fc_queue *, unsigned int);
224 
225 static void
226 nvme_fc_free_lport(struct kref *ref)
227 {
228 	struct nvme_fc_lport *lport =
229 		container_of(ref, struct nvme_fc_lport, ref);
230 	unsigned long flags;
231 
232 	WARN_ON(lport->localport.port_state != FC_OBJSTATE_DELETED);
233 	WARN_ON(!list_empty(&lport->endp_list));
234 
235 	/* remove from transport list */
236 	spin_lock_irqsave(&nvme_fc_lock, flags);
237 	list_del(&lport->port_list);
238 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
239 
240 	ida_simple_remove(&nvme_fc_local_port_cnt, lport->localport.port_num);
241 	ida_destroy(&lport->endp_cnt);
242 
243 	put_device(lport->dev);
244 
245 	kfree(lport);
246 }
247 
248 static void
249 nvme_fc_lport_put(struct nvme_fc_lport *lport)
250 {
251 	kref_put(&lport->ref, nvme_fc_free_lport);
252 }
253 
254 static int
255 nvme_fc_lport_get(struct nvme_fc_lport *lport)
256 {
257 	return kref_get_unless_zero(&lport->ref);
258 }
259 
260 
261 static struct nvme_fc_lport *
262 nvme_fc_attach_to_unreg_lport(struct nvme_fc_port_info *pinfo,
263 			struct nvme_fc_port_template *ops,
264 			struct device *dev)
265 {
266 	struct nvme_fc_lport *lport;
267 	unsigned long flags;
268 
269 	spin_lock_irqsave(&nvme_fc_lock, flags);
270 
271 	list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
272 		if (lport->localport.node_name != pinfo->node_name ||
273 		    lport->localport.port_name != pinfo->port_name)
274 			continue;
275 
276 		if (lport->dev != dev) {
277 			lport = ERR_PTR(-EXDEV);
278 			goto out_done;
279 		}
280 
281 		if (lport->localport.port_state != FC_OBJSTATE_DELETED) {
282 			lport = ERR_PTR(-EEXIST);
283 			goto out_done;
284 		}
285 
286 		if (!nvme_fc_lport_get(lport)) {
287 			/*
288 			 * fails if ref cnt already 0. If so,
289 			 * act as if lport already deleted
290 			 */
291 			lport = NULL;
292 			goto out_done;
293 		}
294 
295 		/* resume the lport */
296 
297 		lport->ops = ops;
298 		lport->localport.port_role = pinfo->port_role;
299 		lport->localport.port_id = pinfo->port_id;
300 		lport->localport.port_state = FC_OBJSTATE_ONLINE;
301 
302 		spin_unlock_irqrestore(&nvme_fc_lock, flags);
303 
304 		return lport;
305 	}
306 
307 	lport = NULL;
308 
309 out_done:
310 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
311 
312 	return lport;
313 }
314 
315 /**
316  * nvme_fc_register_localport - transport entry point called by an
317  *                              LLDD to register the existence of a NVME
318  *                              host FC port.
319  * @pinfo:     pointer to information about the port to be registered
320  * @template:  LLDD entrypoints and operational parameters for the port
321  * @dev:       physical hardware device node port corresponds to. Will be
322  *             used for DMA mappings
323  * @lport_p:   pointer to a local port pointer. Upon success, the routine
324  *             will allocate a nvme_fc_local_port structure and place its
325  *             address in the local port pointer. Upon failure, local port
326  *             pointer will be set to 0.
327  *
328  * Returns:
329  * a completion status. Must be 0 upon success; a negative errno
330  * (ex: -ENXIO) upon failure.
331  */
332 int
333 nvme_fc_register_localport(struct nvme_fc_port_info *pinfo,
334 			struct nvme_fc_port_template *template,
335 			struct device *dev,
336 			struct nvme_fc_local_port **portptr)
337 {
338 	struct nvme_fc_lport *newrec;
339 	unsigned long flags;
340 	int ret, idx;
341 
342 	if (!template->localport_delete || !template->remoteport_delete ||
343 	    !template->ls_req || !template->fcp_io ||
344 	    !template->ls_abort || !template->fcp_abort ||
345 	    !template->max_hw_queues || !template->max_sgl_segments ||
346 	    !template->max_dif_sgl_segments || !template->dma_boundary) {
347 		ret = -EINVAL;
348 		goto out_reghost_failed;
349 	}
350 
351 	/*
352 	 * look to see if there is already a localport that had been
353 	 * deregistered and in the process of waiting for all the
354 	 * references to fully be removed.  If the references haven't
355 	 * expired, we can simply re-enable the localport. Remoteports
356 	 * and controller reconnections should resume naturally.
357 	 */
358 	newrec = nvme_fc_attach_to_unreg_lport(pinfo, template, dev);
359 
360 	/* found an lport, but something about its state is bad */
361 	if (IS_ERR(newrec)) {
362 		ret = PTR_ERR(newrec);
363 		goto out_reghost_failed;
364 
365 	/* found existing lport, which was resumed */
366 	} else if (newrec) {
367 		*portptr = &newrec->localport;
368 		return 0;
369 	}
370 
371 	/* nothing found - allocate a new localport struct */
372 
373 	newrec = kmalloc((sizeof(*newrec) + template->local_priv_sz),
374 			 GFP_KERNEL);
375 	if (!newrec) {
376 		ret = -ENOMEM;
377 		goto out_reghost_failed;
378 	}
379 
380 	idx = ida_simple_get(&nvme_fc_local_port_cnt, 0, 0, GFP_KERNEL);
381 	if (idx < 0) {
382 		ret = -ENOSPC;
383 		goto out_fail_kfree;
384 	}
385 
386 	if (!get_device(dev) && dev) {
387 		ret = -ENODEV;
388 		goto out_ida_put;
389 	}
390 
391 	INIT_LIST_HEAD(&newrec->port_list);
392 	INIT_LIST_HEAD(&newrec->endp_list);
393 	kref_init(&newrec->ref);
394 	atomic_set(&newrec->act_rport_cnt, 0);
395 	newrec->ops = template;
396 	newrec->dev = dev;
397 	ida_init(&newrec->endp_cnt);
398 	newrec->localport.private = &newrec[1];
399 	newrec->localport.node_name = pinfo->node_name;
400 	newrec->localport.port_name = pinfo->port_name;
401 	newrec->localport.port_role = pinfo->port_role;
402 	newrec->localport.port_id = pinfo->port_id;
403 	newrec->localport.port_state = FC_OBJSTATE_ONLINE;
404 	newrec->localport.port_num = idx;
405 
406 	spin_lock_irqsave(&nvme_fc_lock, flags);
407 	list_add_tail(&newrec->port_list, &nvme_fc_lport_list);
408 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
409 
410 	if (dev)
411 		dma_set_seg_boundary(dev, template->dma_boundary);
412 
413 	*portptr = &newrec->localport;
414 	return 0;
415 
416 out_ida_put:
417 	ida_simple_remove(&nvme_fc_local_port_cnt, idx);
418 out_fail_kfree:
419 	kfree(newrec);
420 out_reghost_failed:
421 	*portptr = NULL;
422 
423 	return ret;
424 }
425 EXPORT_SYMBOL_GPL(nvme_fc_register_localport);
426 
427 /**
428  * nvme_fc_unregister_localport - transport entry point called by an
429  *                              LLDD to deregister/remove a previously
430  *                              registered a NVME host FC port.
431  * @localport: pointer to the (registered) local port that is to be
432  *             deregistered.
433  *
434  * Returns:
435  * a completion status. Must be 0 upon success; a negative errno
436  * (ex: -ENXIO) upon failure.
437  */
438 int
439 nvme_fc_unregister_localport(struct nvme_fc_local_port *portptr)
440 {
441 	struct nvme_fc_lport *lport = localport_to_lport(portptr);
442 	unsigned long flags;
443 
444 	if (!portptr)
445 		return -EINVAL;
446 
447 	spin_lock_irqsave(&nvme_fc_lock, flags);
448 
449 	if (portptr->port_state != FC_OBJSTATE_ONLINE) {
450 		spin_unlock_irqrestore(&nvme_fc_lock, flags);
451 		return -EINVAL;
452 	}
453 	portptr->port_state = FC_OBJSTATE_DELETED;
454 
455 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
456 
457 	if (atomic_read(&lport->act_rport_cnt) == 0)
458 		lport->ops->localport_delete(&lport->localport);
459 
460 	nvme_fc_lport_put(lport);
461 
462 	return 0;
463 }
464 EXPORT_SYMBOL_GPL(nvme_fc_unregister_localport);
465 
466 /*
467  * TRADDR strings, per FC-NVME are fixed format:
468  *   "nn-0x<16hexdigits>:pn-0x<16hexdigits>" - 43 characters
469  * udev event will only differ by prefix of what field is
470  * being specified:
471  *    "NVMEFC_HOST_TRADDR=" or "NVMEFC_TRADDR=" - 19 max characters
472  *  19 + 43 + null_fudge = 64 characters
473  */
474 #define FCNVME_TRADDR_LENGTH		64
475 
476 static void
477 nvme_fc_signal_discovery_scan(struct nvme_fc_lport *lport,
478 		struct nvme_fc_rport *rport)
479 {
480 	char hostaddr[FCNVME_TRADDR_LENGTH];	/* NVMEFC_HOST_TRADDR=...*/
481 	char tgtaddr[FCNVME_TRADDR_LENGTH];	/* NVMEFC_TRADDR=...*/
482 	char *envp[4] = { "FC_EVENT=nvmediscovery", hostaddr, tgtaddr, NULL };
483 
484 	if (!(rport->remoteport.port_role & FC_PORT_ROLE_NVME_DISCOVERY))
485 		return;
486 
487 	snprintf(hostaddr, sizeof(hostaddr),
488 		"NVMEFC_HOST_TRADDR=nn-0x%016llx:pn-0x%016llx",
489 		lport->localport.node_name, lport->localport.port_name);
490 	snprintf(tgtaddr, sizeof(tgtaddr),
491 		"NVMEFC_TRADDR=nn-0x%016llx:pn-0x%016llx",
492 		rport->remoteport.node_name, rport->remoteport.port_name);
493 	kobject_uevent_env(&fc_udev_device->kobj, KOBJ_CHANGE, envp);
494 }
495 
496 static void
497 nvme_fc_free_rport(struct kref *ref)
498 {
499 	struct nvme_fc_rport *rport =
500 		container_of(ref, struct nvme_fc_rport, ref);
501 	struct nvme_fc_lport *lport =
502 			localport_to_lport(rport->remoteport.localport);
503 	unsigned long flags;
504 
505 	WARN_ON(rport->remoteport.port_state != FC_OBJSTATE_DELETED);
506 	WARN_ON(!list_empty(&rport->ctrl_list));
507 
508 	/* remove from lport list */
509 	spin_lock_irqsave(&nvme_fc_lock, flags);
510 	list_del(&rport->endp_list);
511 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
512 
513 	ida_simple_remove(&lport->endp_cnt, rport->remoteport.port_num);
514 
515 	kfree(rport);
516 
517 	nvme_fc_lport_put(lport);
518 }
519 
520 static void
521 nvme_fc_rport_put(struct nvme_fc_rport *rport)
522 {
523 	kref_put(&rport->ref, nvme_fc_free_rport);
524 }
525 
526 static int
527 nvme_fc_rport_get(struct nvme_fc_rport *rport)
528 {
529 	return kref_get_unless_zero(&rport->ref);
530 }
531 
532 static void
533 nvme_fc_resume_controller(struct nvme_fc_ctrl *ctrl)
534 {
535 	switch (ctrl->ctrl.state) {
536 	case NVME_CTRL_NEW:
537 	case NVME_CTRL_RECONNECTING:
538 		/*
539 		 * As all reconnects were suppressed, schedule a
540 		 * connect.
541 		 */
542 		dev_info(ctrl->ctrl.device,
543 			"NVME-FC{%d}: connectivity re-established. "
544 			"Attempting reconnect\n", ctrl->cnum);
545 
546 		queue_delayed_work(nvme_wq, &ctrl->connect_work, 0);
547 		break;
548 
549 	case NVME_CTRL_RESETTING:
550 		/*
551 		 * Controller is already in the process of terminating the
552 		 * association. No need to do anything further. The reconnect
553 		 * step will naturally occur after the reset completes.
554 		 */
555 		break;
556 
557 	default:
558 		/* no action to take - let it delete */
559 		break;
560 	}
561 }
562 
563 static struct nvme_fc_rport *
564 nvme_fc_attach_to_suspended_rport(struct nvme_fc_lport *lport,
565 				struct nvme_fc_port_info *pinfo)
566 {
567 	struct nvme_fc_rport *rport;
568 	struct nvme_fc_ctrl *ctrl;
569 	unsigned long flags;
570 
571 	spin_lock_irqsave(&nvme_fc_lock, flags);
572 
573 	list_for_each_entry(rport, &lport->endp_list, endp_list) {
574 		if (rport->remoteport.node_name != pinfo->node_name ||
575 		    rport->remoteport.port_name != pinfo->port_name)
576 			continue;
577 
578 		if (!nvme_fc_rport_get(rport)) {
579 			rport = ERR_PTR(-ENOLCK);
580 			goto out_done;
581 		}
582 
583 		spin_unlock_irqrestore(&nvme_fc_lock, flags);
584 
585 		spin_lock_irqsave(&rport->lock, flags);
586 
587 		/* has it been unregistered */
588 		if (rport->remoteport.port_state != FC_OBJSTATE_DELETED) {
589 			/* means lldd called us twice */
590 			spin_unlock_irqrestore(&rport->lock, flags);
591 			nvme_fc_rport_put(rport);
592 			return ERR_PTR(-ESTALE);
593 		}
594 
595 		rport->remoteport.port_state = FC_OBJSTATE_ONLINE;
596 		rport->dev_loss_end = 0;
597 
598 		/*
599 		 * kick off a reconnect attempt on all associations to the
600 		 * remote port. A successful reconnects will resume i/o.
601 		 */
602 		list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list)
603 			nvme_fc_resume_controller(ctrl);
604 
605 		spin_unlock_irqrestore(&rport->lock, flags);
606 
607 		return rport;
608 	}
609 
610 	rport = NULL;
611 
612 out_done:
613 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
614 
615 	return rport;
616 }
617 
618 static inline void
619 __nvme_fc_set_dev_loss_tmo(struct nvme_fc_rport *rport,
620 			struct nvme_fc_port_info *pinfo)
621 {
622 	if (pinfo->dev_loss_tmo)
623 		rport->remoteport.dev_loss_tmo = pinfo->dev_loss_tmo;
624 	else
625 		rport->remoteport.dev_loss_tmo = NVME_FC_DEFAULT_DEV_LOSS_TMO;
626 }
627 
628 /**
629  * nvme_fc_register_remoteport - transport entry point called by an
630  *                              LLDD to register the existence of a NVME
631  *                              subsystem FC port on its fabric.
632  * @localport: pointer to the (registered) local port that the remote
633  *             subsystem port is connected to.
634  * @pinfo:     pointer to information about the port to be registered
635  * @rport_p:   pointer to a remote port pointer. Upon success, the routine
636  *             will allocate a nvme_fc_remote_port structure and place its
637  *             address in the remote port pointer. Upon failure, remote port
638  *             pointer will be set to 0.
639  *
640  * Returns:
641  * a completion status. Must be 0 upon success; a negative errno
642  * (ex: -ENXIO) upon failure.
643  */
644 int
645 nvme_fc_register_remoteport(struct nvme_fc_local_port *localport,
646 				struct nvme_fc_port_info *pinfo,
647 				struct nvme_fc_remote_port **portptr)
648 {
649 	struct nvme_fc_lport *lport = localport_to_lport(localport);
650 	struct nvme_fc_rport *newrec;
651 	unsigned long flags;
652 	int ret, idx;
653 
654 	if (!nvme_fc_lport_get(lport)) {
655 		ret = -ESHUTDOWN;
656 		goto out_reghost_failed;
657 	}
658 
659 	/*
660 	 * look to see if there is already a remoteport that is waiting
661 	 * for a reconnect (within dev_loss_tmo) with the same WWN's.
662 	 * If so, transition to it and reconnect.
663 	 */
664 	newrec = nvme_fc_attach_to_suspended_rport(lport, pinfo);
665 
666 	/* found an rport, but something about its state is bad */
667 	if (IS_ERR(newrec)) {
668 		ret = PTR_ERR(newrec);
669 		goto out_lport_put;
670 
671 	/* found existing rport, which was resumed */
672 	} else if (newrec) {
673 		nvme_fc_lport_put(lport);
674 		__nvme_fc_set_dev_loss_tmo(newrec, pinfo);
675 		nvme_fc_signal_discovery_scan(lport, newrec);
676 		*portptr = &newrec->remoteport;
677 		return 0;
678 	}
679 
680 	/* nothing found - allocate a new remoteport struct */
681 
682 	newrec = kmalloc((sizeof(*newrec) + lport->ops->remote_priv_sz),
683 			 GFP_KERNEL);
684 	if (!newrec) {
685 		ret = -ENOMEM;
686 		goto out_lport_put;
687 	}
688 
689 	idx = ida_simple_get(&lport->endp_cnt, 0, 0, GFP_KERNEL);
690 	if (idx < 0) {
691 		ret = -ENOSPC;
692 		goto out_kfree_rport;
693 	}
694 
695 	INIT_LIST_HEAD(&newrec->endp_list);
696 	INIT_LIST_HEAD(&newrec->ctrl_list);
697 	INIT_LIST_HEAD(&newrec->ls_req_list);
698 	kref_init(&newrec->ref);
699 	atomic_set(&newrec->act_ctrl_cnt, 0);
700 	spin_lock_init(&newrec->lock);
701 	newrec->remoteport.localport = &lport->localport;
702 	newrec->dev = lport->dev;
703 	newrec->lport = lport;
704 	newrec->remoteport.private = &newrec[1];
705 	newrec->remoteport.port_role = pinfo->port_role;
706 	newrec->remoteport.node_name = pinfo->node_name;
707 	newrec->remoteport.port_name = pinfo->port_name;
708 	newrec->remoteport.port_id = pinfo->port_id;
709 	newrec->remoteport.port_state = FC_OBJSTATE_ONLINE;
710 	newrec->remoteport.port_num = idx;
711 	__nvme_fc_set_dev_loss_tmo(newrec, pinfo);
712 
713 	spin_lock_irqsave(&nvme_fc_lock, flags);
714 	list_add_tail(&newrec->endp_list, &lport->endp_list);
715 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
716 
717 	nvme_fc_signal_discovery_scan(lport, newrec);
718 
719 	*portptr = &newrec->remoteport;
720 	return 0;
721 
722 out_kfree_rport:
723 	kfree(newrec);
724 out_lport_put:
725 	nvme_fc_lport_put(lport);
726 out_reghost_failed:
727 	*portptr = NULL;
728 	return ret;
729 }
730 EXPORT_SYMBOL_GPL(nvme_fc_register_remoteport);
731 
732 static int
733 nvme_fc_abort_lsops(struct nvme_fc_rport *rport)
734 {
735 	struct nvmefc_ls_req_op *lsop;
736 	unsigned long flags;
737 
738 restart:
739 	spin_lock_irqsave(&rport->lock, flags);
740 
741 	list_for_each_entry(lsop, &rport->ls_req_list, lsreq_list) {
742 		if (!(lsop->flags & FCOP_FLAGS_TERMIO)) {
743 			lsop->flags |= FCOP_FLAGS_TERMIO;
744 			spin_unlock_irqrestore(&rport->lock, flags);
745 			rport->lport->ops->ls_abort(&rport->lport->localport,
746 						&rport->remoteport,
747 						&lsop->ls_req);
748 			goto restart;
749 		}
750 	}
751 	spin_unlock_irqrestore(&rport->lock, flags);
752 
753 	return 0;
754 }
755 
756 static void
757 nvme_fc_ctrl_connectivity_loss(struct nvme_fc_ctrl *ctrl)
758 {
759 	dev_info(ctrl->ctrl.device,
760 		"NVME-FC{%d}: controller connectivity lost. Awaiting "
761 		"Reconnect", ctrl->cnum);
762 
763 	switch (ctrl->ctrl.state) {
764 	case NVME_CTRL_NEW:
765 	case NVME_CTRL_LIVE:
766 		/*
767 		 * Schedule a controller reset. The reset will terminate the
768 		 * association and schedule the reconnect timer.  Reconnects
769 		 * will be attempted until either the ctlr_loss_tmo
770 		 * (max_retries * connect_delay) expires or the remoteport's
771 		 * dev_loss_tmo expires.
772 		 */
773 		if (nvme_reset_ctrl(&ctrl->ctrl)) {
774 			dev_warn(ctrl->ctrl.device,
775 				"NVME-FC{%d}: Couldn't schedule reset. "
776 				"Deleting controller.\n",
777 				ctrl->cnum);
778 			nvme_delete_ctrl(&ctrl->ctrl);
779 		}
780 		break;
781 
782 	case NVME_CTRL_RECONNECTING:
783 		/*
784 		 * The association has already been terminated and the
785 		 * controller is attempting reconnects.  No need to do anything
786 		 * futher.  Reconnects will be attempted until either the
787 		 * ctlr_loss_tmo (max_retries * connect_delay) expires or the
788 		 * remoteport's dev_loss_tmo expires.
789 		 */
790 		break;
791 
792 	case NVME_CTRL_RESETTING:
793 		/*
794 		 * Controller is already in the process of terminating the
795 		 * association.  No need to do anything further. The reconnect
796 		 * step will kick in naturally after the association is
797 		 * terminated.
798 		 */
799 		break;
800 
801 	case NVME_CTRL_DELETING:
802 	default:
803 		/* no action to take - let it delete */
804 		break;
805 	}
806 }
807 
808 /**
809  * nvme_fc_unregister_remoteport - transport entry point called by an
810  *                              LLDD to deregister/remove a previously
811  *                              registered a NVME subsystem FC port.
812  * @remoteport: pointer to the (registered) remote port that is to be
813  *              deregistered.
814  *
815  * Returns:
816  * a completion status. Must be 0 upon success; a negative errno
817  * (ex: -ENXIO) upon failure.
818  */
819 int
820 nvme_fc_unregister_remoteport(struct nvme_fc_remote_port *portptr)
821 {
822 	struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
823 	struct nvme_fc_ctrl *ctrl;
824 	unsigned long flags;
825 
826 	if (!portptr)
827 		return -EINVAL;
828 
829 	spin_lock_irqsave(&rport->lock, flags);
830 
831 	if (portptr->port_state != FC_OBJSTATE_ONLINE) {
832 		spin_unlock_irqrestore(&rport->lock, flags);
833 		return -EINVAL;
834 	}
835 	portptr->port_state = FC_OBJSTATE_DELETED;
836 
837 	rport->dev_loss_end = jiffies + (portptr->dev_loss_tmo * HZ);
838 
839 	list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list) {
840 		/* if dev_loss_tmo==0, dev loss is immediate */
841 		if (!portptr->dev_loss_tmo) {
842 			dev_warn(ctrl->ctrl.device,
843 				"NVME-FC{%d}: controller connectivity lost. "
844 				"Deleting controller.\n",
845 				ctrl->cnum);
846 			nvme_delete_ctrl(&ctrl->ctrl);
847 		} else
848 			nvme_fc_ctrl_connectivity_loss(ctrl);
849 	}
850 
851 	spin_unlock_irqrestore(&rport->lock, flags);
852 
853 	nvme_fc_abort_lsops(rport);
854 
855 	if (atomic_read(&rport->act_ctrl_cnt) == 0)
856 		rport->lport->ops->remoteport_delete(portptr);
857 
858 	/*
859 	 * release the reference, which will allow, if all controllers
860 	 * go away, which should only occur after dev_loss_tmo occurs,
861 	 * for the rport to be torn down.
862 	 */
863 	nvme_fc_rport_put(rport);
864 
865 	return 0;
866 }
867 EXPORT_SYMBOL_GPL(nvme_fc_unregister_remoteport);
868 
869 /**
870  * nvme_fc_rescan_remoteport - transport entry point called by an
871  *                              LLDD to request a nvme device rescan.
872  * @remoteport: pointer to the (registered) remote port that is to be
873  *              rescanned.
874  *
875  * Returns: N/A
876  */
877 void
878 nvme_fc_rescan_remoteport(struct nvme_fc_remote_port *remoteport)
879 {
880 	struct nvme_fc_rport *rport = remoteport_to_rport(remoteport);
881 
882 	nvme_fc_signal_discovery_scan(rport->lport, rport);
883 }
884 EXPORT_SYMBOL_GPL(nvme_fc_rescan_remoteport);
885 
886 int
887 nvme_fc_set_remoteport_devloss(struct nvme_fc_remote_port *portptr,
888 			u32 dev_loss_tmo)
889 {
890 	struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
891 	unsigned long flags;
892 
893 	spin_lock_irqsave(&rport->lock, flags);
894 
895 	if (portptr->port_state != FC_OBJSTATE_ONLINE) {
896 		spin_unlock_irqrestore(&rport->lock, flags);
897 		return -EINVAL;
898 	}
899 
900 	/* a dev_loss_tmo of 0 (immediate) is allowed to be set */
901 	rport->remoteport.dev_loss_tmo = dev_loss_tmo;
902 
903 	spin_unlock_irqrestore(&rport->lock, flags);
904 
905 	return 0;
906 }
907 EXPORT_SYMBOL_GPL(nvme_fc_set_remoteport_devloss);
908 
909 
910 /* *********************** FC-NVME DMA Handling **************************** */
911 
912 /*
913  * The fcloop device passes in a NULL device pointer. Real LLD's will
914  * pass in a valid device pointer. If NULL is passed to the dma mapping
915  * routines, depending on the platform, it may or may not succeed, and
916  * may crash.
917  *
918  * As such:
919  * Wrapper all the dma routines and check the dev pointer.
920  *
921  * If simple mappings (return just a dma address, we'll noop them,
922  * returning a dma address of 0.
923  *
924  * On more complex mappings (dma_map_sg), a pseudo routine fills
925  * in the scatter list, setting all dma addresses to 0.
926  */
927 
928 static inline dma_addr_t
929 fc_dma_map_single(struct device *dev, void *ptr, size_t size,
930 		enum dma_data_direction dir)
931 {
932 	return dev ? dma_map_single(dev, ptr, size, dir) : (dma_addr_t)0L;
933 }
934 
935 static inline int
936 fc_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
937 {
938 	return dev ? dma_mapping_error(dev, dma_addr) : 0;
939 }
940 
941 static inline void
942 fc_dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
943 	enum dma_data_direction dir)
944 {
945 	if (dev)
946 		dma_unmap_single(dev, addr, size, dir);
947 }
948 
949 static inline void
950 fc_dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr, size_t size,
951 		enum dma_data_direction dir)
952 {
953 	if (dev)
954 		dma_sync_single_for_cpu(dev, addr, size, dir);
955 }
956 
957 static inline void
958 fc_dma_sync_single_for_device(struct device *dev, dma_addr_t addr, size_t size,
959 		enum dma_data_direction dir)
960 {
961 	if (dev)
962 		dma_sync_single_for_device(dev, addr, size, dir);
963 }
964 
965 /* pseudo dma_map_sg call */
966 static int
967 fc_map_sg(struct scatterlist *sg, int nents)
968 {
969 	struct scatterlist *s;
970 	int i;
971 
972 	WARN_ON(nents == 0 || sg[0].length == 0);
973 
974 	for_each_sg(sg, s, nents, i) {
975 		s->dma_address = 0L;
976 #ifdef CONFIG_NEED_SG_DMA_LENGTH
977 		s->dma_length = s->length;
978 #endif
979 	}
980 	return nents;
981 }
982 
983 static inline int
984 fc_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
985 		enum dma_data_direction dir)
986 {
987 	return dev ? dma_map_sg(dev, sg, nents, dir) : fc_map_sg(sg, nents);
988 }
989 
990 static inline void
991 fc_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
992 		enum dma_data_direction dir)
993 {
994 	if (dev)
995 		dma_unmap_sg(dev, sg, nents, dir);
996 }
997 
998 /* *********************** FC-NVME LS Handling **************************** */
999 
1000 static void nvme_fc_ctrl_put(struct nvme_fc_ctrl *);
1001 static int nvme_fc_ctrl_get(struct nvme_fc_ctrl *);
1002 
1003 
1004 static void
1005 __nvme_fc_finish_ls_req(struct nvmefc_ls_req_op *lsop)
1006 {
1007 	struct nvme_fc_rport *rport = lsop->rport;
1008 	struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1009 	unsigned long flags;
1010 
1011 	spin_lock_irqsave(&rport->lock, flags);
1012 
1013 	if (!lsop->req_queued) {
1014 		spin_unlock_irqrestore(&rport->lock, flags);
1015 		return;
1016 	}
1017 
1018 	list_del(&lsop->lsreq_list);
1019 
1020 	lsop->req_queued = false;
1021 
1022 	spin_unlock_irqrestore(&rport->lock, flags);
1023 
1024 	fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
1025 				  (lsreq->rqstlen + lsreq->rsplen),
1026 				  DMA_BIDIRECTIONAL);
1027 
1028 	nvme_fc_rport_put(rport);
1029 }
1030 
1031 static int
1032 __nvme_fc_send_ls_req(struct nvme_fc_rport *rport,
1033 		struct nvmefc_ls_req_op *lsop,
1034 		void (*done)(struct nvmefc_ls_req *req, int status))
1035 {
1036 	struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1037 	unsigned long flags;
1038 	int ret = 0;
1039 
1040 	if (rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
1041 		return -ECONNREFUSED;
1042 
1043 	if (!nvme_fc_rport_get(rport))
1044 		return -ESHUTDOWN;
1045 
1046 	lsreq->done = done;
1047 	lsop->rport = rport;
1048 	lsop->req_queued = false;
1049 	INIT_LIST_HEAD(&lsop->lsreq_list);
1050 	init_completion(&lsop->ls_done);
1051 
1052 	lsreq->rqstdma = fc_dma_map_single(rport->dev, lsreq->rqstaddr,
1053 				  lsreq->rqstlen + lsreq->rsplen,
1054 				  DMA_BIDIRECTIONAL);
1055 	if (fc_dma_mapping_error(rport->dev, lsreq->rqstdma)) {
1056 		ret = -EFAULT;
1057 		goto out_putrport;
1058 	}
1059 	lsreq->rspdma = lsreq->rqstdma + lsreq->rqstlen;
1060 
1061 	spin_lock_irqsave(&rport->lock, flags);
1062 
1063 	list_add_tail(&lsop->lsreq_list, &rport->ls_req_list);
1064 
1065 	lsop->req_queued = true;
1066 
1067 	spin_unlock_irqrestore(&rport->lock, flags);
1068 
1069 	ret = rport->lport->ops->ls_req(&rport->lport->localport,
1070 					&rport->remoteport, lsreq);
1071 	if (ret)
1072 		goto out_unlink;
1073 
1074 	return 0;
1075 
1076 out_unlink:
1077 	lsop->ls_error = ret;
1078 	spin_lock_irqsave(&rport->lock, flags);
1079 	lsop->req_queued = false;
1080 	list_del(&lsop->lsreq_list);
1081 	spin_unlock_irqrestore(&rport->lock, flags);
1082 	fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
1083 				  (lsreq->rqstlen + lsreq->rsplen),
1084 				  DMA_BIDIRECTIONAL);
1085 out_putrport:
1086 	nvme_fc_rport_put(rport);
1087 
1088 	return ret;
1089 }
1090 
1091 static void
1092 nvme_fc_send_ls_req_done(struct nvmefc_ls_req *lsreq, int status)
1093 {
1094 	struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
1095 
1096 	lsop->ls_error = status;
1097 	complete(&lsop->ls_done);
1098 }
1099 
1100 static int
1101 nvme_fc_send_ls_req(struct nvme_fc_rport *rport, struct nvmefc_ls_req_op *lsop)
1102 {
1103 	struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1104 	struct fcnvme_ls_rjt *rjt = lsreq->rspaddr;
1105 	int ret;
1106 
1107 	ret = __nvme_fc_send_ls_req(rport, lsop, nvme_fc_send_ls_req_done);
1108 
1109 	if (!ret) {
1110 		/*
1111 		 * No timeout/not interruptible as we need the struct
1112 		 * to exist until the lldd calls us back. Thus mandate
1113 		 * wait until driver calls back. lldd responsible for
1114 		 * the timeout action
1115 		 */
1116 		wait_for_completion(&lsop->ls_done);
1117 
1118 		__nvme_fc_finish_ls_req(lsop);
1119 
1120 		ret = lsop->ls_error;
1121 	}
1122 
1123 	if (ret)
1124 		return ret;
1125 
1126 	/* ACC or RJT payload ? */
1127 	if (rjt->w0.ls_cmd == FCNVME_LS_RJT)
1128 		return -ENXIO;
1129 
1130 	return 0;
1131 }
1132 
1133 static int
1134 nvme_fc_send_ls_req_async(struct nvme_fc_rport *rport,
1135 		struct nvmefc_ls_req_op *lsop,
1136 		void (*done)(struct nvmefc_ls_req *req, int status))
1137 {
1138 	/* don't wait for completion */
1139 
1140 	return __nvme_fc_send_ls_req(rport, lsop, done);
1141 }
1142 
1143 /* Validation Error indexes into the string table below */
1144 enum {
1145 	VERR_NO_ERROR		= 0,
1146 	VERR_LSACC		= 1,
1147 	VERR_LSDESC_RQST	= 2,
1148 	VERR_LSDESC_RQST_LEN	= 3,
1149 	VERR_ASSOC_ID		= 4,
1150 	VERR_ASSOC_ID_LEN	= 5,
1151 	VERR_CONN_ID		= 6,
1152 	VERR_CONN_ID_LEN	= 7,
1153 	VERR_CR_ASSOC		= 8,
1154 	VERR_CR_ASSOC_ACC_LEN	= 9,
1155 	VERR_CR_CONN		= 10,
1156 	VERR_CR_CONN_ACC_LEN	= 11,
1157 	VERR_DISCONN		= 12,
1158 	VERR_DISCONN_ACC_LEN	= 13,
1159 };
1160 
1161 static char *validation_errors[] = {
1162 	"OK",
1163 	"Not LS_ACC",
1164 	"Not LSDESC_RQST",
1165 	"Bad LSDESC_RQST Length",
1166 	"Not Association ID",
1167 	"Bad Association ID Length",
1168 	"Not Connection ID",
1169 	"Bad Connection ID Length",
1170 	"Not CR_ASSOC Rqst",
1171 	"Bad CR_ASSOC ACC Length",
1172 	"Not CR_CONN Rqst",
1173 	"Bad CR_CONN ACC Length",
1174 	"Not Disconnect Rqst",
1175 	"Bad Disconnect ACC Length",
1176 };
1177 
1178 static int
1179 nvme_fc_connect_admin_queue(struct nvme_fc_ctrl *ctrl,
1180 	struct nvme_fc_queue *queue, u16 qsize, u16 ersp_ratio)
1181 {
1182 	struct nvmefc_ls_req_op *lsop;
1183 	struct nvmefc_ls_req *lsreq;
1184 	struct fcnvme_ls_cr_assoc_rqst *assoc_rqst;
1185 	struct fcnvme_ls_cr_assoc_acc *assoc_acc;
1186 	int ret, fcret = 0;
1187 
1188 	lsop = kzalloc((sizeof(*lsop) +
1189 			 ctrl->lport->ops->lsrqst_priv_sz +
1190 			 sizeof(*assoc_rqst) + sizeof(*assoc_acc)), GFP_KERNEL);
1191 	if (!lsop) {
1192 		ret = -ENOMEM;
1193 		goto out_no_memory;
1194 	}
1195 	lsreq = &lsop->ls_req;
1196 
1197 	lsreq->private = (void *)&lsop[1];
1198 	assoc_rqst = (struct fcnvme_ls_cr_assoc_rqst *)
1199 			(lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1200 	assoc_acc = (struct fcnvme_ls_cr_assoc_acc *)&assoc_rqst[1];
1201 
1202 	assoc_rqst->w0.ls_cmd = FCNVME_LS_CREATE_ASSOCIATION;
1203 	assoc_rqst->desc_list_len =
1204 			cpu_to_be32(sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
1205 
1206 	assoc_rqst->assoc_cmd.desc_tag =
1207 			cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD);
1208 	assoc_rqst->assoc_cmd.desc_len =
1209 			fcnvme_lsdesc_len(
1210 				sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
1211 
1212 	assoc_rqst->assoc_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
1213 	assoc_rqst->assoc_cmd.sqsize = cpu_to_be16(qsize);
1214 	/* Linux supports only Dynamic controllers */
1215 	assoc_rqst->assoc_cmd.cntlid = cpu_to_be16(0xffff);
1216 	uuid_copy(&assoc_rqst->assoc_cmd.hostid, &ctrl->ctrl.opts->host->id);
1217 	strncpy(assoc_rqst->assoc_cmd.hostnqn, ctrl->ctrl.opts->host->nqn,
1218 		min(FCNVME_ASSOC_HOSTNQN_LEN, NVMF_NQN_SIZE));
1219 	strncpy(assoc_rqst->assoc_cmd.subnqn, ctrl->ctrl.opts->subsysnqn,
1220 		min(FCNVME_ASSOC_SUBNQN_LEN, NVMF_NQN_SIZE));
1221 
1222 	lsop->queue = queue;
1223 	lsreq->rqstaddr = assoc_rqst;
1224 	lsreq->rqstlen = sizeof(*assoc_rqst);
1225 	lsreq->rspaddr = assoc_acc;
1226 	lsreq->rsplen = sizeof(*assoc_acc);
1227 	lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1228 
1229 	ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
1230 	if (ret)
1231 		goto out_free_buffer;
1232 
1233 	/* process connect LS completion */
1234 
1235 	/* validate the ACC response */
1236 	if (assoc_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1237 		fcret = VERR_LSACC;
1238 	else if (assoc_acc->hdr.desc_list_len !=
1239 			fcnvme_lsdesc_len(
1240 				sizeof(struct fcnvme_ls_cr_assoc_acc)))
1241 		fcret = VERR_CR_ASSOC_ACC_LEN;
1242 	else if (assoc_acc->hdr.rqst.desc_tag !=
1243 			cpu_to_be32(FCNVME_LSDESC_RQST))
1244 		fcret = VERR_LSDESC_RQST;
1245 	else if (assoc_acc->hdr.rqst.desc_len !=
1246 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1247 		fcret = VERR_LSDESC_RQST_LEN;
1248 	else if (assoc_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_ASSOCIATION)
1249 		fcret = VERR_CR_ASSOC;
1250 	else if (assoc_acc->associd.desc_tag !=
1251 			cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
1252 		fcret = VERR_ASSOC_ID;
1253 	else if (assoc_acc->associd.desc_len !=
1254 			fcnvme_lsdesc_len(
1255 				sizeof(struct fcnvme_lsdesc_assoc_id)))
1256 		fcret = VERR_ASSOC_ID_LEN;
1257 	else if (assoc_acc->connectid.desc_tag !=
1258 			cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1259 		fcret = VERR_CONN_ID;
1260 	else if (assoc_acc->connectid.desc_len !=
1261 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1262 		fcret = VERR_CONN_ID_LEN;
1263 
1264 	if (fcret) {
1265 		ret = -EBADF;
1266 		dev_err(ctrl->dev,
1267 			"q %d connect failed: %s\n",
1268 			queue->qnum, validation_errors[fcret]);
1269 	} else {
1270 		ctrl->association_id =
1271 			be64_to_cpu(assoc_acc->associd.association_id);
1272 		queue->connection_id =
1273 			be64_to_cpu(assoc_acc->connectid.connection_id);
1274 		set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1275 	}
1276 
1277 out_free_buffer:
1278 	kfree(lsop);
1279 out_no_memory:
1280 	if (ret)
1281 		dev_err(ctrl->dev,
1282 			"queue %d connect admin queue failed (%d).\n",
1283 			queue->qnum, ret);
1284 	return ret;
1285 }
1286 
1287 static int
1288 nvme_fc_connect_queue(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
1289 			u16 qsize, u16 ersp_ratio)
1290 {
1291 	struct nvmefc_ls_req_op *lsop;
1292 	struct nvmefc_ls_req *lsreq;
1293 	struct fcnvme_ls_cr_conn_rqst *conn_rqst;
1294 	struct fcnvme_ls_cr_conn_acc *conn_acc;
1295 	int ret, fcret = 0;
1296 
1297 	lsop = kzalloc((sizeof(*lsop) +
1298 			 ctrl->lport->ops->lsrqst_priv_sz +
1299 			 sizeof(*conn_rqst) + sizeof(*conn_acc)), GFP_KERNEL);
1300 	if (!lsop) {
1301 		ret = -ENOMEM;
1302 		goto out_no_memory;
1303 	}
1304 	lsreq = &lsop->ls_req;
1305 
1306 	lsreq->private = (void *)&lsop[1];
1307 	conn_rqst = (struct fcnvme_ls_cr_conn_rqst *)
1308 			(lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1309 	conn_acc = (struct fcnvme_ls_cr_conn_acc *)&conn_rqst[1];
1310 
1311 	conn_rqst->w0.ls_cmd = FCNVME_LS_CREATE_CONNECTION;
1312 	conn_rqst->desc_list_len = cpu_to_be32(
1313 				sizeof(struct fcnvme_lsdesc_assoc_id) +
1314 				sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1315 
1316 	conn_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1317 	conn_rqst->associd.desc_len =
1318 			fcnvme_lsdesc_len(
1319 				sizeof(struct fcnvme_lsdesc_assoc_id));
1320 	conn_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1321 	conn_rqst->connect_cmd.desc_tag =
1322 			cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD);
1323 	conn_rqst->connect_cmd.desc_len =
1324 			fcnvme_lsdesc_len(
1325 				sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1326 	conn_rqst->connect_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
1327 	conn_rqst->connect_cmd.qid  = cpu_to_be16(queue->qnum);
1328 	conn_rqst->connect_cmd.sqsize = cpu_to_be16(qsize);
1329 
1330 	lsop->queue = queue;
1331 	lsreq->rqstaddr = conn_rqst;
1332 	lsreq->rqstlen = sizeof(*conn_rqst);
1333 	lsreq->rspaddr = conn_acc;
1334 	lsreq->rsplen = sizeof(*conn_acc);
1335 	lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1336 
1337 	ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
1338 	if (ret)
1339 		goto out_free_buffer;
1340 
1341 	/* process connect LS completion */
1342 
1343 	/* validate the ACC response */
1344 	if (conn_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1345 		fcret = VERR_LSACC;
1346 	else if (conn_acc->hdr.desc_list_len !=
1347 			fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)))
1348 		fcret = VERR_CR_CONN_ACC_LEN;
1349 	else if (conn_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
1350 		fcret = VERR_LSDESC_RQST;
1351 	else if (conn_acc->hdr.rqst.desc_len !=
1352 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1353 		fcret = VERR_LSDESC_RQST_LEN;
1354 	else if (conn_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_CONNECTION)
1355 		fcret = VERR_CR_CONN;
1356 	else if (conn_acc->connectid.desc_tag !=
1357 			cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1358 		fcret = VERR_CONN_ID;
1359 	else if (conn_acc->connectid.desc_len !=
1360 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1361 		fcret = VERR_CONN_ID_LEN;
1362 
1363 	if (fcret) {
1364 		ret = -EBADF;
1365 		dev_err(ctrl->dev,
1366 			"q %d connect failed: %s\n",
1367 			queue->qnum, validation_errors[fcret]);
1368 	} else {
1369 		queue->connection_id =
1370 			be64_to_cpu(conn_acc->connectid.connection_id);
1371 		set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1372 	}
1373 
1374 out_free_buffer:
1375 	kfree(lsop);
1376 out_no_memory:
1377 	if (ret)
1378 		dev_err(ctrl->dev,
1379 			"queue %d connect command failed (%d).\n",
1380 			queue->qnum, ret);
1381 	return ret;
1382 }
1383 
1384 static void
1385 nvme_fc_disconnect_assoc_done(struct nvmefc_ls_req *lsreq, int status)
1386 {
1387 	struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
1388 
1389 	__nvme_fc_finish_ls_req(lsop);
1390 
1391 	/* fc-nvme iniator doesn't care about success or failure of cmd */
1392 
1393 	kfree(lsop);
1394 }
1395 
1396 /*
1397  * This routine sends a FC-NVME LS to disconnect (aka terminate)
1398  * the FC-NVME Association.  Terminating the association also
1399  * terminates the FC-NVME connections (per queue, both admin and io
1400  * queues) that are part of the association. E.g. things are torn
1401  * down, and the related FC-NVME Association ID and Connection IDs
1402  * become invalid.
1403  *
1404  * The behavior of the fc-nvme initiator is such that it's
1405  * understanding of the association and connections will implicitly
1406  * be torn down. The action is implicit as it may be due to a loss of
1407  * connectivity with the fc-nvme target, so you may never get a
1408  * response even if you tried.  As such, the action of this routine
1409  * is to asynchronously send the LS, ignore any results of the LS, and
1410  * continue on with terminating the association. If the fc-nvme target
1411  * is present and receives the LS, it too can tear down.
1412  */
1413 static void
1414 nvme_fc_xmt_disconnect_assoc(struct nvme_fc_ctrl *ctrl)
1415 {
1416 	struct fcnvme_ls_disconnect_rqst *discon_rqst;
1417 	struct fcnvme_ls_disconnect_acc *discon_acc;
1418 	struct nvmefc_ls_req_op *lsop;
1419 	struct nvmefc_ls_req *lsreq;
1420 	int ret;
1421 
1422 	lsop = kzalloc((sizeof(*lsop) +
1423 			 ctrl->lport->ops->lsrqst_priv_sz +
1424 			 sizeof(*discon_rqst) + sizeof(*discon_acc)),
1425 			GFP_KERNEL);
1426 	if (!lsop)
1427 		/* couldn't sent it... too bad */
1428 		return;
1429 
1430 	lsreq = &lsop->ls_req;
1431 
1432 	lsreq->private = (void *)&lsop[1];
1433 	discon_rqst = (struct fcnvme_ls_disconnect_rqst *)
1434 			(lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1435 	discon_acc = (struct fcnvme_ls_disconnect_acc *)&discon_rqst[1];
1436 
1437 	discon_rqst->w0.ls_cmd = FCNVME_LS_DISCONNECT;
1438 	discon_rqst->desc_list_len = cpu_to_be32(
1439 				sizeof(struct fcnvme_lsdesc_assoc_id) +
1440 				sizeof(struct fcnvme_lsdesc_disconn_cmd));
1441 
1442 	discon_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1443 	discon_rqst->associd.desc_len =
1444 			fcnvme_lsdesc_len(
1445 				sizeof(struct fcnvme_lsdesc_assoc_id));
1446 
1447 	discon_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1448 
1449 	discon_rqst->discon_cmd.desc_tag = cpu_to_be32(
1450 						FCNVME_LSDESC_DISCONN_CMD);
1451 	discon_rqst->discon_cmd.desc_len =
1452 			fcnvme_lsdesc_len(
1453 				sizeof(struct fcnvme_lsdesc_disconn_cmd));
1454 	discon_rqst->discon_cmd.scope = FCNVME_DISCONN_ASSOCIATION;
1455 	discon_rqst->discon_cmd.id = cpu_to_be64(ctrl->association_id);
1456 
1457 	lsreq->rqstaddr = discon_rqst;
1458 	lsreq->rqstlen = sizeof(*discon_rqst);
1459 	lsreq->rspaddr = discon_acc;
1460 	lsreq->rsplen = sizeof(*discon_acc);
1461 	lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1462 
1463 	ret = nvme_fc_send_ls_req_async(ctrl->rport, lsop,
1464 				nvme_fc_disconnect_assoc_done);
1465 	if (ret)
1466 		kfree(lsop);
1467 
1468 	/* only meaningful part to terminating the association */
1469 	ctrl->association_id = 0;
1470 }
1471 
1472 
1473 /* *********************** NVME Ctrl Routines **************************** */
1474 
1475 static void __nvme_fc_final_op_cleanup(struct request *rq);
1476 static void nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg);
1477 
1478 static int
1479 nvme_fc_reinit_request(void *data, struct request *rq)
1480 {
1481 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1482 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1483 
1484 	memset(cmdiu, 0, sizeof(*cmdiu));
1485 	cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1486 	cmdiu->fc_id = NVME_CMD_FC_ID;
1487 	cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1488 	memset(&op->rsp_iu, 0, sizeof(op->rsp_iu));
1489 
1490 	return 0;
1491 }
1492 
1493 static void
1494 __nvme_fc_exit_request(struct nvme_fc_ctrl *ctrl,
1495 		struct nvme_fc_fcp_op *op)
1496 {
1497 	fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.rspdma,
1498 				sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1499 	fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.cmddma,
1500 				sizeof(op->cmd_iu), DMA_TO_DEVICE);
1501 
1502 	atomic_set(&op->state, FCPOP_STATE_UNINIT);
1503 }
1504 
1505 static void
1506 nvme_fc_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1507 		unsigned int hctx_idx)
1508 {
1509 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1510 
1511 	return __nvme_fc_exit_request(set->driver_data, op);
1512 }
1513 
1514 static int
1515 __nvme_fc_abort_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_fcp_op *op)
1516 {
1517 	int state;
1518 
1519 	state = atomic_xchg(&op->state, FCPOP_STATE_ABORTED);
1520 	if (state != FCPOP_STATE_ACTIVE) {
1521 		atomic_set(&op->state, state);
1522 		return -ECANCELED;
1523 	}
1524 
1525 	ctrl->lport->ops->fcp_abort(&ctrl->lport->localport,
1526 					&ctrl->rport->remoteport,
1527 					op->queue->lldd_handle,
1528 					&op->fcp_req);
1529 
1530 	return 0;
1531 }
1532 
1533 static void
1534 nvme_fc_abort_aen_ops(struct nvme_fc_ctrl *ctrl)
1535 {
1536 	struct nvme_fc_fcp_op *aen_op = ctrl->aen_ops;
1537 	unsigned long flags;
1538 	int i, ret;
1539 
1540 	for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++) {
1541 		if (atomic_read(&aen_op->state) != FCPOP_STATE_ACTIVE)
1542 			continue;
1543 
1544 		spin_lock_irqsave(&ctrl->lock, flags);
1545 		if (ctrl->flags & FCCTRL_TERMIO) {
1546 			ctrl->iocnt++;
1547 			aen_op->flags |= FCOP_FLAGS_TERMIO;
1548 		}
1549 		spin_unlock_irqrestore(&ctrl->lock, flags);
1550 
1551 		ret = __nvme_fc_abort_op(ctrl, aen_op);
1552 		if (ret) {
1553 			/*
1554 			 * if __nvme_fc_abort_op failed the io wasn't
1555 			 * active. Thus this call path is running in
1556 			 * parallel to the io complete. Treat as non-error.
1557 			 */
1558 
1559 			/* back out the flags/counters */
1560 			spin_lock_irqsave(&ctrl->lock, flags);
1561 			if (ctrl->flags & FCCTRL_TERMIO)
1562 				ctrl->iocnt--;
1563 			aen_op->flags &= ~FCOP_FLAGS_TERMIO;
1564 			spin_unlock_irqrestore(&ctrl->lock, flags);
1565 			return;
1566 		}
1567 	}
1568 }
1569 
1570 static inline int
1571 __nvme_fc_fcpop_chk_teardowns(struct nvme_fc_ctrl *ctrl,
1572 		struct nvme_fc_fcp_op *op)
1573 {
1574 	unsigned long flags;
1575 	bool complete_rq = false;
1576 
1577 	spin_lock_irqsave(&ctrl->lock, flags);
1578 	if (unlikely(op->flags & FCOP_FLAGS_TERMIO)) {
1579 		if (ctrl->flags & FCCTRL_TERMIO) {
1580 			if (!--ctrl->iocnt)
1581 				wake_up(&ctrl->ioabort_wait);
1582 		}
1583 	}
1584 	if (op->flags & FCOP_FLAGS_RELEASED)
1585 		complete_rq = true;
1586 	else
1587 		op->flags |= FCOP_FLAGS_COMPLETE;
1588 	spin_unlock_irqrestore(&ctrl->lock, flags);
1589 
1590 	return complete_rq;
1591 }
1592 
1593 static void
1594 nvme_fc_fcpio_done(struct nvmefc_fcp_req *req)
1595 {
1596 	struct nvme_fc_fcp_op *op = fcp_req_to_fcp_op(req);
1597 	struct request *rq = op->rq;
1598 	struct nvmefc_fcp_req *freq = &op->fcp_req;
1599 	struct nvme_fc_ctrl *ctrl = op->ctrl;
1600 	struct nvme_fc_queue *queue = op->queue;
1601 	struct nvme_completion *cqe = &op->rsp_iu.cqe;
1602 	struct nvme_command *sqe = &op->cmd_iu.sqe;
1603 	__le16 status = cpu_to_le16(NVME_SC_SUCCESS << 1);
1604 	union nvme_result result;
1605 	bool terminate_assoc = true;
1606 
1607 	/*
1608 	 * WARNING:
1609 	 * The current linux implementation of a nvme controller
1610 	 * allocates a single tag set for all io queues and sizes
1611 	 * the io queues to fully hold all possible tags. Thus, the
1612 	 * implementation does not reference or care about the sqhd
1613 	 * value as it never needs to use the sqhd/sqtail pointers
1614 	 * for submission pacing.
1615 	 *
1616 	 * This affects the FC-NVME implementation in two ways:
1617 	 * 1) As the value doesn't matter, we don't need to waste
1618 	 *    cycles extracting it from ERSPs and stamping it in the
1619 	 *    cases where the transport fabricates CQEs on successful
1620 	 *    completions.
1621 	 * 2) The FC-NVME implementation requires that delivery of
1622 	 *    ERSP completions are to go back to the nvme layer in order
1623 	 *    relative to the rsn, such that the sqhd value will always
1624 	 *    be "in order" for the nvme layer. As the nvme layer in
1625 	 *    linux doesn't care about sqhd, there's no need to return
1626 	 *    them in order.
1627 	 *
1628 	 * Additionally:
1629 	 * As the core nvme layer in linux currently does not look at
1630 	 * every field in the cqe - in cases where the FC transport must
1631 	 * fabricate a CQE, the following fields will not be set as they
1632 	 * are not referenced:
1633 	 *      cqe.sqid,  cqe.sqhd,  cqe.command_id
1634 	 *
1635 	 * Failure or error of an individual i/o, in a transport
1636 	 * detected fashion unrelated to the nvme completion status,
1637 	 * potentially cause the initiator and target sides to get out
1638 	 * of sync on SQ head/tail (aka outstanding io count allowed).
1639 	 * Per FC-NVME spec, failure of an individual command requires
1640 	 * the connection to be terminated, which in turn requires the
1641 	 * association to be terminated.
1642 	 */
1643 
1644 	fc_dma_sync_single_for_cpu(ctrl->lport->dev, op->fcp_req.rspdma,
1645 				sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1646 
1647 	if (atomic_read(&op->state) == FCPOP_STATE_ABORTED ||
1648 			op->flags & FCOP_FLAGS_TERMIO)
1649 		status = cpu_to_le16(NVME_SC_ABORT_REQ << 1);
1650 	else if (freq->status)
1651 		status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1652 
1653 	/*
1654 	 * For the linux implementation, if we have an unsuccesful
1655 	 * status, they blk-mq layer can typically be called with the
1656 	 * non-zero status and the content of the cqe isn't important.
1657 	 */
1658 	if (status)
1659 		goto done;
1660 
1661 	/*
1662 	 * command completed successfully relative to the wire
1663 	 * protocol. However, validate anything received and
1664 	 * extract the status and result from the cqe (create it
1665 	 * where necessary).
1666 	 */
1667 
1668 	switch (freq->rcv_rsplen) {
1669 
1670 	case 0:
1671 	case NVME_FC_SIZEOF_ZEROS_RSP:
1672 		/*
1673 		 * No response payload or 12 bytes of payload (which
1674 		 * should all be zeros) are considered successful and
1675 		 * no payload in the CQE by the transport.
1676 		 */
1677 		if (freq->transferred_length !=
1678 			be32_to_cpu(op->cmd_iu.data_len)) {
1679 			status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1680 			goto done;
1681 		}
1682 		result.u64 = 0;
1683 		break;
1684 
1685 	case sizeof(struct nvme_fc_ersp_iu):
1686 		/*
1687 		 * The ERSP IU contains a full completion with CQE.
1688 		 * Validate ERSP IU and look at cqe.
1689 		 */
1690 		if (unlikely(be16_to_cpu(op->rsp_iu.iu_len) !=
1691 					(freq->rcv_rsplen / 4) ||
1692 			     be32_to_cpu(op->rsp_iu.xfrd_len) !=
1693 					freq->transferred_length ||
1694 			     op->rsp_iu.status_code ||
1695 			     sqe->common.command_id != cqe->command_id)) {
1696 			status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1697 			goto done;
1698 		}
1699 		result = cqe->result;
1700 		status = cqe->status;
1701 		break;
1702 
1703 	default:
1704 		status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1705 		goto done;
1706 	}
1707 
1708 	terminate_assoc = false;
1709 
1710 done:
1711 	if (op->flags & FCOP_FLAGS_AEN) {
1712 		nvme_complete_async_event(&queue->ctrl->ctrl, status, &result);
1713 		__nvme_fc_fcpop_chk_teardowns(ctrl, op);
1714 		atomic_set(&op->state, FCPOP_STATE_IDLE);
1715 		op->flags = FCOP_FLAGS_AEN;	/* clear other flags */
1716 		nvme_fc_ctrl_put(ctrl);
1717 		goto check_error;
1718 	}
1719 
1720 	/*
1721 	 * Force failures of commands if we're killing the controller
1722 	 * or have an error on a command used to create an new association
1723 	 */
1724 	if (status &&
1725 	    (blk_queue_dying(rq->q) ||
1726 	     ctrl->ctrl.state == NVME_CTRL_NEW ||
1727 	     ctrl->ctrl.state == NVME_CTRL_RECONNECTING))
1728 		status |= cpu_to_le16(NVME_SC_DNR << 1);
1729 
1730 	if (__nvme_fc_fcpop_chk_teardowns(ctrl, op))
1731 		__nvme_fc_final_op_cleanup(rq);
1732 	else
1733 		nvme_end_request(rq, status, result);
1734 
1735 check_error:
1736 	if (terminate_assoc)
1737 		nvme_fc_error_recovery(ctrl, "transport detected io error");
1738 }
1739 
1740 static int
1741 __nvme_fc_init_request(struct nvme_fc_ctrl *ctrl,
1742 		struct nvme_fc_queue *queue, struct nvme_fc_fcp_op *op,
1743 		struct request *rq, u32 rqno)
1744 {
1745 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1746 	int ret = 0;
1747 
1748 	memset(op, 0, sizeof(*op));
1749 	op->fcp_req.cmdaddr = &op->cmd_iu;
1750 	op->fcp_req.cmdlen = sizeof(op->cmd_iu);
1751 	op->fcp_req.rspaddr = &op->rsp_iu;
1752 	op->fcp_req.rsplen = sizeof(op->rsp_iu);
1753 	op->fcp_req.done = nvme_fc_fcpio_done;
1754 	op->fcp_req.first_sgl = (struct scatterlist *)&op[1];
1755 	op->fcp_req.private = &op->fcp_req.first_sgl[SG_CHUNK_SIZE];
1756 	op->ctrl = ctrl;
1757 	op->queue = queue;
1758 	op->rq = rq;
1759 	op->rqno = rqno;
1760 
1761 	cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1762 	cmdiu->fc_id = NVME_CMD_FC_ID;
1763 	cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1764 
1765 	op->fcp_req.cmddma = fc_dma_map_single(ctrl->lport->dev,
1766 				&op->cmd_iu, sizeof(op->cmd_iu), DMA_TO_DEVICE);
1767 	if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.cmddma)) {
1768 		dev_err(ctrl->dev,
1769 			"FCP Op failed - cmdiu dma mapping failed.\n");
1770 		ret = EFAULT;
1771 		goto out_on_error;
1772 	}
1773 
1774 	op->fcp_req.rspdma = fc_dma_map_single(ctrl->lport->dev,
1775 				&op->rsp_iu, sizeof(op->rsp_iu),
1776 				DMA_FROM_DEVICE);
1777 	if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.rspdma)) {
1778 		dev_err(ctrl->dev,
1779 			"FCP Op failed - rspiu dma mapping failed.\n");
1780 		ret = EFAULT;
1781 	}
1782 
1783 	atomic_set(&op->state, FCPOP_STATE_IDLE);
1784 out_on_error:
1785 	return ret;
1786 }
1787 
1788 static int
1789 nvme_fc_init_request(struct blk_mq_tag_set *set, struct request *rq,
1790 		unsigned int hctx_idx, unsigned int numa_node)
1791 {
1792 	struct nvme_fc_ctrl *ctrl = set->driver_data;
1793 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1794 	int queue_idx = (set == &ctrl->tag_set) ? hctx_idx + 1 : 0;
1795 	struct nvme_fc_queue *queue = &ctrl->queues[queue_idx];
1796 
1797 	return __nvme_fc_init_request(ctrl, queue, op, rq, queue->rqcnt++);
1798 }
1799 
1800 static int
1801 nvme_fc_init_aen_ops(struct nvme_fc_ctrl *ctrl)
1802 {
1803 	struct nvme_fc_fcp_op *aen_op;
1804 	struct nvme_fc_cmd_iu *cmdiu;
1805 	struct nvme_command *sqe;
1806 	void *private;
1807 	int i, ret;
1808 
1809 	aen_op = ctrl->aen_ops;
1810 	for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++) {
1811 		private = kzalloc(ctrl->lport->ops->fcprqst_priv_sz,
1812 						GFP_KERNEL);
1813 		if (!private)
1814 			return -ENOMEM;
1815 
1816 		cmdiu = &aen_op->cmd_iu;
1817 		sqe = &cmdiu->sqe;
1818 		ret = __nvme_fc_init_request(ctrl, &ctrl->queues[0],
1819 				aen_op, (struct request *)NULL,
1820 				(NVME_AQ_BLK_MQ_DEPTH + i));
1821 		if (ret) {
1822 			kfree(private);
1823 			return ret;
1824 		}
1825 
1826 		aen_op->flags = FCOP_FLAGS_AEN;
1827 		aen_op->fcp_req.first_sgl = NULL; /* no sg list */
1828 		aen_op->fcp_req.private = private;
1829 
1830 		memset(sqe, 0, sizeof(*sqe));
1831 		sqe->common.opcode = nvme_admin_async_event;
1832 		/* Note: core layer may overwrite the sqe.command_id value */
1833 		sqe->common.command_id = NVME_AQ_BLK_MQ_DEPTH + i;
1834 	}
1835 	return 0;
1836 }
1837 
1838 static void
1839 nvme_fc_term_aen_ops(struct nvme_fc_ctrl *ctrl)
1840 {
1841 	struct nvme_fc_fcp_op *aen_op;
1842 	int i;
1843 
1844 	aen_op = ctrl->aen_ops;
1845 	for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++) {
1846 		if (!aen_op->fcp_req.private)
1847 			continue;
1848 
1849 		__nvme_fc_exit_request(ctrl, aen_op);
1850 
1851 		kfree(aen_op->fcp_req.private);
1852 		aen_op->fcp_req.private = NULL;
1853 	}
1854 }
1855 
1856 static inline void
1857 __nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, struct nvme_fc_ctrl *ctrl,
1858 		unsigned int qidx)
1859 {
1860 	struct nvme_fc_queue *queue = &ctrl->queues[qidx];
1861 
1862 	hctx->driver_data = queue;
1863 	queue->hctx = hctx;
1864 }
1865 
1866 static int
1867 nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1868 		unsigned int hctx_idx)
1869 {
1870 	struct nvme_fc_ctrl *ctrl = data;
1871 
1872 	__nvme_fc_init_hctx(hctx, ctrl, hctx_idx + 1);
1873 
1874 	return 0;
1875 }
1876 
1877 static int
1878 nvme_fc_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1879 		unsigned int hctx_idx)
1880 {
1881 	struct nvme_fc_ctrl *ctrl = data;
1882 
1883 	__nvme_fc_init_hctx(hctx, ctrl, hctx_idx);
1884 
1885 	return 0;
1886 }
1887 
1888 static void
1889 nvme_fc_init_queue(struct nvme_fc_ctrl *ctrl, int idx)
1890 {
1891 	struct nvme_fc_queue *queue;
1892 
1893 	queue = &ctrl->queues[idx];
1894 	memset(queue, 0, sizeof(*queue));
1895 	queue->ctrl = ctrl;
1896 	queue->qnum = idx;
1897 	atomic_set(&queue->csn, 1);
1898 	queue->dev = ctrl->dev;
1899 
1900 	if (idx > 0)
1901 		queue->cmnd_capsule_len = ctrl->ctrl.ioccsz * 16;
1902 	else
1903 		queue->cmnd_capsule_len = sizeof(struct nvme_command);
1904 
1905 	/*
1906 	 * Considered whether we should allocate buffers for all SQEs
1907 	 * and CQEs and dma map them - mapping their respective entries
1908 	 * into the request structures (kernel vm addr and dma address)
1909 	 * thus the driver could use the buffers/mappings directly.
1910 	 * It only makes sense if the LLDD would use them for its
1911 	 * messaging api. It's very unlikely most adapter api's would use
1912 	 * a native NVME sqe/cqe. More reasonable if FC-NVME IU payload
1913 	 * structures were used instead.
1914 	 */
1915 }
1916 
1917 /*
1918  * This routine terminates a queue at the transport level.
1919  * The transport has already ensured that all outstanding ios on
1920  * the queue have been terminated.
1921  * The transport will send a Disconnect LS request to terminate
1922  * the queue's connection. Termination of the admin queue will also
1923  * terminate the association at the target.
1924  */
1925 static void
1926 nvme_fc_free_queue(struct nvme_fc_queue *queue)
1927 {
1928 	if (!test_and_clear_bit(NVME_FC_Q_CONNECTED, &queue->flags))
1929 		return;
1930 
1931 	clear_bit(NVME_FC_Q_LIVE, &queue->flags);
1932 	/*
1933 	 * Current implementation never disconnects a single queue.
1934 	 * It always terminates a whole association. So there is never
1935 	 * a disconnect(queue) LS sent to the target.
1936 	 */
1937 
1938 	queue->connection_id = 0;
1939 }
1940 
1941 static void
1942 __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *ctrl,
1943 	struct nvme_fc_queue *queue, unsigned int qidx)
1944 {
1945 	if (ctrl->lport->ops->delete_queue)
1946 		ctrl->lport->ops->delete_queue(&ctrl->lport->localport, qidx,
1947 				queue->lldd_handle);
1948 	queue->lldd_handle = NULL;
1949 }
1950 
1951 static void
1952 nvme_fc_free_io_queues(struct nvme_fc_ctrl *ctrl)
1953 {
1954 	int i;
1955 
1956 	for (i = 1; i < ctrl->ctrl.queue_count; i++)
1957 		nvme_fc_free_queue(&ctrl->queues[i]);
1958 }
1959 
1960 static int
1961 __nvme_fc_create_hw_queue(struct nvme_fc_ctrl *ctrl,
1962 	struct nvme_fc_queue *queue, unsigned int qidx, u16 qsize)
1963 {
1964 	int ret = 0;
1965 
1966 	queue->lldd_handle = NULL;
1967 	if (ctrl->lport->ops->create_queue)
1968 		ret = ctrl->lport->ops->create_queue(&ctrl->lport->localport,
1969 				qidx, qsize, &queue->lldd_handle);
1970 
1971 	return ret;
1972 }
1973 
1974 static void
1975 nvme_fc_delete_hw_io_queues(struct nvme_fc_ctrl *ctrl)
1976 {
1977 	struct nvme_fc_queue *queue = &ctrl->queues[ctrl->ctrl.queue_count - 1];
1978 	int i;
1979 
1980 	for (i = ctrl->ctrl.queue_count - 1; i >= 1; i--, queue--)
1981 		__nvme_fc_delete_hw_queue(ctrl, queue, i);
1982 }
1983 
1984 static int
1985 nvme_fc_create_hw_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1986 {
1987 	struct nvme_fc_queue *queue = &ctrl->queues[1];
1988 	int i, ret;
1989 
1990 	for (i = 1; i < ctrl->ctrl.queue_count; i++, queue++) {
1991 		ret = __nvme_fc_create_hw_queue(ctrl, queue, i, qsize);
1992 		if (ret)
1993 			goto delete_queues;
1994 	}
1995 
1996 	return 0;
1997 
1998 delete_queues:
1999 	for (; i >= 0; i--)
2000 		__nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[i], i);
2001 	return ret;
2002 }
2003 
2004 static int
2005 nvme_fc_connect_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
2006 {
2007 	int i, ret = 0;
2008 
2009 	for (i = 1; i < ctrl->ctrl.queue_count; i++) {
2010 		ret = nvme_fc_connect_queue(ctrl, &ctrl->queues[i], qsize,
2011 					(qsize / 5));
2012 		if (ret)
2013 			break;
2014 		ret = nvmf_connect_io_queue(&ctrl->ctrl, i);
2015 		if (ret)
2016 			break;
2017 
2018 		set_bit(NVME_FC_Q_LIVE, &ctrl->queues[i].flags);
2019 	}
2020 
2021 	return ret;
2022 }
2023 
2024 static void
2025 nvme_fc_init_io_queues(struct nvme_fc_ctrl *ctrl)
2026 {
2027 	int i;
2028 
2029 	for (i = 1; i < ctrl->ctrl.queue_count; i++)
2030 		nvme_fc_init_queue(ctrl, i);
2031 }
2032 
2033 static void
2034 nvme_fc_ctrl_free(struct kref *ref)
2035 {
2036 	struct nvme_fc_ctrl *ctrl =
2037 		container_of(ref, struct nvme_fc_ctrl, ref);
2038 	unsigned long flags;
2039 
2040 	if (ctrl->ctrl.tagset) {
2041 		blk_cleanup_queue(ctrl->ctrl.connect_q);
2042 		blk_mq_free_tag_set(&ctrl->tag_set);
2043 	}
2044 
2045 	/* remove from rport list */
2046 	spin_lock_irqsave(&ctrl->rport->lock, flags);
2047 	list_del(&ctrl->ctrl_list);
2048 	spin_unlock_irqrestore(&ctrl->rport->lock, flags);
2049 
2050 	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2051 	blk_cleanup_queue(ctrl->ctrl.admin_q);
2052 	blk_mq_free_tag_set(&ctrl->admin_tag_set);
2053 
2054 	kfree(ctrl->queues);
2055 
2056 	put_device(ctrl->dev);
2057 	nvme_fc_rport_put(ctrl->rport);
2058 
2059 	ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
2060 	if (ctrl->ctrl.opts)
2061 		nvmf_free_options(ctrl->ctrl.opts);
2062 	kfree(ctrl);
2063 }
2064 
2065 static void
2066 nvme_fc_ctrl_put(struct nvme_fc_ctrl *ctrl)
2067 {
2068 	kref_put(&ctrl->ref, nvme_fc_ctrl_free);
2069 }
2070 
2071 static int
2072 nvme_fc_ctrl_get(struct nvme_fc_ctrl *ctrl)
2073 {
2074 	return kref_get_unless_zero(&ctrl->ref);
2075 }
2076 
2077 /*
2078  * All accesses from nvme core layer done - can now free the
2079  * controller. Called after last nvme_put_ctrl() call
2080  */
2081 static void
2082 nvme_fc_nvme_ctrl_freed(struct nvme_ctrl *nctrl)
2083 {
2084 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2085 
2086 	WARN_ON(nctrl != &ctrl->ctrl);
2087 
2088 	nvme_fc_ctrl_put(ctrl);
2089 }
2090 
2091 static void
2092 nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg)
2093 {
2094 	/* only proceed if in LIVE state - e.g. on first error */
2095 	if (ctrl->ctrl.state != NVME_CTRL_LIVE)
2096 		return;
2097 
2098 	dev_warn(ctrl->ctrl.device,
2099 		"NVME-FC{%d}: transport association error detected: %s\n",
2100 		ctrl->cnum, errmsg);
2101 	dev_warn(ctrl->ctrl.device,
2102 		"NVME-FC{%d}: resetting controller\n", ctrl->cnum);
2103 
2104 	nvme_reset_ctrl(&ctrl->ctrl);
2105 }
2106 
2107 static enum blk_eh_timer_return
2108 nvme_fc_timeout(struct request *rq, bool reserved)
2109 {
2110 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2111 	struct nvme_fc_ctrl *ctrl = op->ctrl;
2112 	int ret;
2113 
2114 	if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE ||
2115 			atomic_read(&op->state) == FCPOP_STATE_ABORTED)
2116 		return BLK_EH_RESET_TIMER;
2117 
2118 	ret = __nvme_fc_abort_op(ctrl, op);
2119 	if (ret)
2120 		/* io wasn't active to abort */
2121 		return BLK_EH_NOT_HANDLED;
2122 
2123 	/*
2124 	 * we can't individually ABTS an io without affecting the queue,
2125 	 * thus killing the queue, adn thus the association.
2126 	 * So resolve by performing a controller reset, which will stop
2127 	 * the host/io stack, terminate the association on the link,
2128 	 * and recreate an association on the link.
2129 	 */
2130 	nvme_fc_error_recovery(ctrl, "io timeout error");
2131 
2132 	/*
2133 	 * the io abort has been initiated. Have the reset timer
2134 	 * restarted and the abort completion will complete the io
2135 	 * shortly. Avoids a synchronous wait while the abort finishes.
2136 	 */
2137 	return BLK_EH_RESET_TIMER;
2138 }
2139 
2140 static int
2141 nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
2142 		struct nvme_fc_fcp_op *op)
2143 {
2144 	struct nvmefc_fcp_req *freq = &op->fcp_req;
2145 	enum dma_data_direction dir;
2146 	int ret;
2147 
2148 	freq->sg_cnt = 0;
2149 
2150 	if (!blk_rq_payload_bytes(rq))
2151 		return 0;
2152 
2153 	freq->sg_table.sgl = freq->first_sgl;
2154 	ret = sg_alloc_table_chained(&freq->sg_table,
2155 			blk_rq_nr_phys_segments(rq), freq->sg_table.sgl);
2156 	if (ret)
2157 		return -ENOMEM;
2158 
2159 	op->nents = blk_rq_map_sg(rq->q, rq, freq->sg_table.sgl);
2160 	WARN_ON(op->nents > blk_rq_nr_phys_segments(rq));
2161 	dir = (rq_data_dir(rq) == WRITE) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
2162 	freq->sg_cnt = fc_dma_map_sg(ctrl->lport->dev, freq->sg_table.sgl,
2163 				op->nents, dir);
2164 	if (unlikely(freq->sg_cnt <= 0)) {
2165 		sg_free_table_chained(&freq->sg_table, true);
2166 		freq->sg_cnt = 0;
2167 		return -EFAULT;
2168 	}
2169 
2170 	/*
2171 	 * TODO: blk_integrity_rq(rq)  for DIF
2172 	 */
2173 	return 0;
2174 }
2175 
2176 static void
2177 nvme_fc_unmap_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
2178 		struct nvme_fc_fcp_op *op)
2179 {
2180 	struct nvmefc_fcp_req *freq = &op->fcp_req;
2181 
2182 	if (!freq->sg_cnt)
2183 		return;
2184 
2185 	fc_dma_unmap_sg(ctrl->lport->dev, freq->sg_table.sgl, op->nents,
2186 				((rq_data_dir(rq) == WRITE) ?
2187 					DMA_TO_DEVICE : DMA_FROM_DEVICE));
2188 
2189 	nvme_cleanup_cmd(rq);
2190 
2191 	sg_free_table_chained(&freq->sg_table, true);
2192 
2193 	freq->sg_cnt = 0;
2194 }
2195 
2196 /*
2197  * In FC, the queue is a logical thing. At transport connect, the target
2198  * creates its "queue" and returns a handle that is to be given to the
2199  * target whenever it posts something to the corresponding SQ.  When an
2200  * SQE is sent on a SQ, FC effectively considers the SQE, or rather the
2201  * command contained within the SQE, an io, and assigns a FC exchange
2202  * to it. The SQE and the associated SQ handle are sent in the initial
2203  * CMD IU sents on the exchange. All transfers relative to the io occur
2204  * as part of the exchange.  The CQE is the last thing for the io,
2205  * which is transferred (explicitly or implicitly) with the RSP IU
2206  * sent on the exchange. After the CQE is received, the FC exchange is
2207  * terminaed and the Exchange may be used on a different io.
2208  *
2209  * The transport to LLDD api has the transport making a request for a
2210  * new fcp io request to the LLDD. The LLDD then allocates a FC exchange
2211  * resource and transfers the command. The LLDD will then process all
2212  * steps to complete the io. Upon completion, the transport done routine
2213  * is called.
2214  *
2215  * So - while the operation is outstanding to the LLDD, there is a link
2216  * level FC exchange resource that is also outstanding. This must be
2217  * considered in all cleanup operations.
2218  */
2219 static blk_status_t
2220 nvme_fc_start_fcp_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
2221 	struct nvme_fc_fcp_op *op, u32 data_len,
2222 	enum nvmefc_fcp_datadir	io_dir)
2223 {
2224 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2225 	struct nvme_command *sqe = &cmdiu->sqe;
2226 	u32 csn;
2227 	int ret;
2228 
2229 	/*
2230 	 * before attempting to send the io, check to see if we believe
2231 	 * the target device is present
2232 	 */
2233 	if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
2234 		goto busy;
2235 
2236 	if (!nvme_fc_ctrl_get(ctrl))
2237 		return BLK_STS_IOERR;
2238 
2239 	/* format the FC-NVME CMD IU and fcp_req */
2240 	cmdiu->connection_id = cpu_to_be64(queue->connection_id);
2241 	csn = atomic_inc_return(&queue->csn);
2242 	cmdiu->csn = cpu_to_be32(csn);
2243 	cmdiu->data_len = cpu_to_be32(data_len);
2244 	switch (io_dir) {
2245 	case NVMEFC_FCP_WRITE:
2246 		cmdiu->flags = FCNVME_CMD_FLAGS_WRITE;
2247 		break;
2248 	case NVMEFC_FCP_READ:
2249 		cmdiu->flags = FCNVME_CMD_FLAGS_READ;
2250 		break;
2251 	case NVMEFC_FCP_NODATA:
2252 		cmdiu->flags = 0;
2253 		break;
2254 	}
2255 	op->fcp_req.payload_length = data_len;
2256 	op->fcp_req.io_dir = io_dir;
2257 	op->fcp_req.transferred_length = 0;
2258 	op->fcp_req.rcv_rsplen = 0;
2259 	op->fcp_req.status = NVME_SC_SUCCESS;
2260 	op->fcp_req.sqid = cpu_to_le16(queue->qnum);
2261 
2262 	/*
2263 	 * validate per fabric rules, set fields mandated by fabric spec
2264 	 * as well as those by FC-NVME spec.
2265 	 */
2266 	WARN_ON_ONCE(sqe->common.metadata);
2267 	sqe->common.flags |= NVME_CMD_SGL_METABUF;
2268 
2269 	/*
2270 	 * format SQE DPTR field per FC-NVME rules:
2271 	 *    type=0x5     Transport SGL Data Block Descriptor
2272 	 *    subtype=0xA  Transport-specific value
2273 	 *    address=0
2274 	 *    length=length of the data series
2275 	 */
2276 	sqe->rw.dptr.sgl.type = (NVME_TRANSPORT_SGL_DATA_DESC << 4) |
2277 					NVME_SGL_FMT_TRANSPORT_A;
2278 	sqe->rw.dptr.sgl.length = cpu_to_le32(data_len);
2279 	sqe->rw.dptr.sgl.addr = 0;
2280 
2281 	if (!(op->flags & FCOP_FLAGS_AEN)) {
2282 		ret = nvme_fc_map_data(ctrl, op->rq, op);
2283 		if (ret < 0) {
2284 			nvme_cleanup_cmd(op->rq);
2285 			nvme_fc_ctrl_put(ctrl);
2286 			if (ret == -ENOMEM || ret == -EAGAIN)
2287 				return BLK_STS_RESOURCE;
2288 			return BLK_STS_IOERR;
2289 		}
2290 	}
2291 
2292 	fc_dma_sync_single_for_device(ctrl->lport->dev, op->fcp_req.cmddma,
2293 				  sizeof(op->cmd_iu), DMA_TO_DEVICE);
2294 
2295 	atomic_set(&op->state, FCPOP_STATE_ACTIVE);
2296 
2297 	if (!(op->flags & FCOP_FLAGS_AEN))
2298 		blk_mq_start_request(op->rq);
2299 
2300 	ret = ctrl->lport->ops->fcp_io(&ctrl->lport->localport,
2301 					&ctrl->rport->remoteport,
2302 					queue->lldd_handle, &op->fcp_req);
2303 
2304 	if (ret) {
2305 		if (!(op->flags & FCOP_FLAGS_AEN))
2306 			nvme_fc_unmap_data(ctrl, op->rq, op);
2307 
2308 		nvme_fc_ctrl_put(ctrl);
2309 
2310 		if (ctrl->rport->remoteport.port_state == FC_OBJSTATE_ONLINE &&
2311 				ret != -EBUSY)
2312 			return BLK_STS_IOERR;
2313 
2314 		goto busy;
2315 	}
2316 
2317 	return BLK_STS_OK;
2318 
2319 busy:
2320 	if (!(op->flags & FCOP_FLAGS_AEN) && queue->hctx)
2321 		blk_mq_delay_run_hw_queue(queue->hctx, NVMEFC_QUEUE_DELAY);
2322 
2323 	return BLK_STS_RESOURCE;
2324 }
2325 
2326 static inline blk_status_t nvme_fc_is_ready(struct nvme_fc_queue *queue,
2327 		struct request *rq)
2328 {
2329 	if (unlikely(!test_bit(NVME_FC_Q_LIVE, &queue->flags)))
2330 		return nvmf_check_init_req(&queue->ctrl->ctrl, rq);
2331 	return BLK_STS_OK;
2332 }
2333 
2334 static blk_status_t
2335 nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,
2336 			const struct blk_mq_queue_data *bd)
2337 {
2338 	struct nvme_ns *ns = hctx->queue->queuedata;
2339 	struct nvme_fc_queue *queue = hctx->driver_data;
2340 	struct nvme_fc_ctrl *ctrl = queue->ctrl;
2341 	struct request *rq = bd->rq;
2342 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2343 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2344 	struct nvme_command *sqe = &cmdiu->sqe;
2345 	enum nvmefc_fcp_datadir	io_dir;
2346 	u32 data_len;
2347 	blk_status_t ret;
2348 
2349 	ret = nvme_fc_is_ready(queue, rq);
2350 	if (unlikely(ret))
2351 		return ret;
2352 
2353 	ret = nvme_setup_cmd(ns, rq, sqe);
2354 	if (ret)
2355 		return ret;
2356 
2357 	data_len = blk_rq_payload_bytes(rq);
2358 	if (data_len)
2359 		io_dir = ((rq_data_dir(rq) == WRITE) ?
2360 					NVMEFC_FCP_WRITE : NVMEFC_FCP_READ);
2361 	else
2362 		io_dir = NVMEFC_FCP_NODATA;
2363 
2364 	return nvme_fc_start_fcp_op(ctrl, queue, op, data_len, io_dir);
2365 }
2366 
2367 static struct blk_mq_tags *
2368 nvme_fc_tagset(struct nvme_fc_queue *queue)
2369 {
2370 	if (queue->qnum == 0)
2371 		return queue->ctrl->admin_tag_set.tags[queue->qnum];
2372 
2373 	return queue->ctrl->tag_set.tags[queue->qnum - 1];
2374 }
2375 
2376 static int
2377 nvme_fc_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
2378 
2379 {
2380 	struct nvme_fc_queue *queue = hctx->driver_data;
2381 	struct nvme_fc_ctrl *ctrl = queue->ctrl;
2382 	struct request *req;
2383 	struct nvme_fc_fcp_op *op;
2384 
2385 	req = blk_mq_tag_to_rq(nvme_fc_tagset(queue), tag);
2386 	if (!req)
2387 		return 0;
2388 
2389 	op = blk_mq_rq_to_pdu(req);
2390 
2391 	if ((atomic_read(&op->state) == FCPOP_STATE_ACTIVE) &&
2392 		 (ctrl->lport->ops->poll_queue))
2393 		ctrl->lport->ops->poll_queue(&ctrl->lport->localport,
2394 						 queue->lldd_handle);
2395 
2396 	return ((atomic_read(&op->state) != FCPOP_STATE_ACTIVE));
2397 }
2398 
2399 static void
2400 nvme_fc_submit_async_event(struct nvme_ctrl *arg)
2401 {
2402 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(arg);
2403 	struct nvme_fc_fcp_op *aen_op;
2404 	unsigned long flags;
2405 	bool terminating = false;
2406 	blk_status_t ret;
2407 
2408 	spin_lock_irqsave(&ctrl->lock, flags);
2409 	if (ctrl->flags & FCCTRL_TERMIO)
2410 		terminating = true;
2411 	spin_unlock_irqrestore(&ctrl->lock, flags);
2412 
2413 	if (terminating)
2414 		return;
2415 
2416 	aen_op = &ctrl->aen_ops[0];
2417 
2418 	ret = nvme_fc_start_fcp_op(ctrl, aen_op->queue, aen_op, 0,
2419 					NVMEFC_FCP_NODATA);
2420 	if (ret)
2421 		dev_err(ctrl->ctrl.device,
2422 			"failed async event work\n");
2423 }
2424 
2425 static void
2426 __nvme_fc_final_op_cleanup(struct request *rq)
2427 {
2428 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2429 	struct nvme_fc_ctrl *ctrl = op->ctrl;
2430 
2431 	atomic_set(&op->state, FCPOP_STATE_IDLE);
2432 	op->flags &= ~(FCOP_FLAGS_TERMIO | FCOP_FLAGS_RELEASED |
2433 			FCOP_FLAGS_COMPLETE);
2434 
2435 	nvme_fc_unmap_data(ctrl, rq, op);
2436 	nvme_complete_rq(rq);
2437 	nvme_fc_ctrl_put(ctrl);
2438 
2439 }
2440 
2441 static void
2442 nvme_fc_complete_rq(struct request *rq)
2443 {
2444 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2445 	struct nvme_fc_ctrl *ctrl = op->ctrl;
2446 	unsigned long flags;
2447 	bool completed = false;
2448 
2449 	/*
2450 	 * the core layer, on controller resets after calling
2451 	 * nvme_shutdown_ctrl(), calls complete_rq without our
2452 	 * calling blk_mq_complete_request(), thus there may still
2453 	 * be live i/o outstanding with the LLDD. Means transport has
2454 	 * to track complete calls vs fcpio_done calls to know what
2455 	 * path to take on completes and dones.
2456 	 */
2457 	spin_lock_irqsave(&ctrl->lock, flags);
2458 	if (op->flags & FCOP_FLAGS_COMPLETE)
2459 		completed = true;
2460 	else
2461 		op->flags |= FCOP_FLAGS_RELEASED;
2462 	spin_unlock_irqrestore(&ctrl->lock, flags);
2463 
2464 	if (completed)
2465 		__nvme_fc_final_op_cleanup(rq);
2466 }
2467 
2468 /*
2469  * This routine is used by the transport when it needs to find active
2470  * io on a queue that is to be terminated. The transport uses
2471  * blk_mq_tagset_busy_itr() to find the busy requests, which then invoke
2472  * this routine to kill them on a 1 by 1 basis.
2473  *
2474  * As FC allocates FC exchange for each io, the transport must contact
2475  * the LLDD to terminate the exchange, thus releasing the FC exchange.
2476  * After terminating the exchange the LLDD will call the transport's
2477  * normal io done path for the request, but it will have an aborted
2478  * status. The done path will return the io request back to the block
2479  * layer with an error status.
2480  */
2481 static void
2482 nvme_fc_terminate_exchange(struct request *req, void *data, bool reserved)
2483 {
2484 	struct nvme_ctrl *nctrl = data;
2485 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2486 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(req);
2487 	unsigned long flags;
2488 	int status;
2489 
2490 	if (!blk_mq_request_started(req))
2491 		return;
2492 
2493 	spin_lock_irqsave(&ctrl->lock, flags);
2494 	if (ctrl->flags & FCCTRL_TERMIO) {
2495 		ctrl->iocnt++;
2496 		op->flags |= FCOP_FLAGS_TERMIO;
2497 	}
2498 	spin_unlock_irqrestore(&ctrl->lock, flags);
2499 
2500 	status = __nvme_fc_abort_op(ctrl, op);
2501 	if (status) {
2502 		/*
2503 		 * if __nvme_fc_abort_op failed the io wasn't
2504 		 * active. Thus this call path is running in
2505 		 * parallel to the io complete. Treat as non-error.
2506 		 */
2507 
2508 		/* back out the flags/counters */
2509 		spin_lock_irqsave(&ctrl->lock, flags);
2510 		if (ctrl->flags & FCCTRL_TERMIO)
2511 			ctrl->iocnt--;
2512 		op->flags &= ~FCOP_FLAGS_TERMIO;
2513 		spin_unlock_irqrestore(&ctrl->lock, flags);
2514 		return;
2515 	}
2516 }
2517 
2518 
2519 static const struct blk_mq_ops nvme_fc_mq_ops = {
2520 	.queue_rq	= nvme_fc_queue_rq,
2521 	.complete	= nvme_fc_complete_rq,
2522 	.init_request	= nvme_fc_init_request,
2523 	.exit_request	= nvme_fc_exit_request,
2524 	.init_hctx	= nvme_fc_init_hctx,
2525 	.poll		= nvme_fc_poll,
2526 	.timeout	= nvme_fc_timeout,
2527 };
2528 
2529 static int
2530 nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)
2531 {
2532 	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2533 	unsigned int nr_io_queues;
2534 	int ret;
2535 
2536 	nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2537 				ctrl->lport->ops->max_hw_queues);
2538 	ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
2539 	if (ret) {
2540 		dev_info(ctrl->ctrl.device,
2541 			"set_queue_count failed: %d\n", ret);
2542 		return ret;
2543 	}
2544 
2545 	ctrl->ctrl.queue_count = nr_io_queues + 1;
2546 	if (!nr_io_queues)
2547 		return 0;
2548 
2549 	nvme_fc_init_io_queues(ctrl);
2550 
2551 	memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set));
2552 	ctrl->tag_set.ops = &nvme_fc_mq_ops;
2553 	ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size;
2554 	ctrl->tag_set.reserved_tags = 1; /* fabric connect */
2555 	ctrl->tag_set.numa_node = NUMA_NO_NODE;
2556 	ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2557 	ctrl->tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
2558 					(SG_CHUNK_SIZE *
2559 						sizeof(struct scatterlist)) +
2560 					ctrl->lport->ops->fcprqst_priv_sz;
2561 	ctrl->tag_set.driver_data = ctrl;
2562 	ctrl->tag_set.nr_hw_queues = ctrl->ctrl.queue_count - 1;
2563 	ctrl->tag_set.timeout = NVME_IO_TIMEOUT;
2564 
2565 	ret = blk_mq_alloc_tag_set(&ctrl->tag_set);
2566 	if (ret)
2567 		return ret;
2568 
2569 	ctrl->ctrl.tagset = &ctrl->tag_set;
2570 
2571 	ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set);
2572 	if (IS_ERR(ctrl->ctrl.connect_q)) {
2573 		ret = PTR_ERR(ctrl->ctrl.connect_q);
2574 		goto out_free_tag_set;
2575 	}
2576 
2577 	ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2578 	if (ret)
2579 		goto out_cleanup_blk_queue;
2580 
2581 	ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2582 	if (ret)
2583 		goto out_delete_hw_queues;
2584 
2585 	return 0;
2586 
2587 out_delete_hw_queues:
2588 	nvme_fc_delete_hw_io_queues(ctrl);
2589 out_cleanup_blk_queue:
2590 	blk_cleanup_queue(ctrl->ctrl.connect_q);
2591 out_free_tag_set:
2592 	blk_mq_free_tag_set(&ctrl->tag_set);
2593 	nvme_fc_free_io_queues(ctrl);
2594 
2595 	/* force put free routine to ignore io queues */
2596 	ctrl->ctrl.tagset = NULL;
2597 
2598 	return ret;
2599 }
2600 
2601 static int
2602 nvme_fc_reinit_io_queues(struct nvme_fc_ctrl *ctrl)
2603 {
2604 	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2605 	unsigned int nr_io_queues;
2606 	int ret;
2607 
2608 	nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2609 				ctrl->lport->ops->max_hw_queues);
2610 	ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
2611 	if (ret) {
2612 		dev_info(ctrl->ctrl.device,
2613 			"set_queue_count failed: %d\n", ret);
2614 		return ret;
2615 	}
2616 
2617 	ctrl->ctrl.queue_count = nr_io_queues + 1;
2618 	/* check for io queues existing */
2619 	if (ctrl->ctrl.queue_count == 1)
2620 		return 0;
2621 
2622 	nvme_fc_init_io_queues(ctrl);
2623 
2624 	ret = nvme_reinit_tagset(&ctrl->ctrl, ctrl->ctrl.tagset);
2625 	if (ret)
2626 		goto out_free_io_queues;
2627 
2628 	ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2629 	if (ret)
2630 		goto out_free_io_queues;
2631 
2632 	ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2633 	if (ret)
2634 		goto out_delete_hw_queues;
2635 
2636 	blk_mq_update_nr_hw_queues(&ctrl->tag_set, nr_io_queues);
2637 
2638 	return 0;
2639 
2640 out_delete_hw_queues:
2641 	nvme_fc_delete_hw_io_queues(ctrl);
2642 out_free_io_queues:
2643 	nvme_fc_free_io_queues(ctrl);
2644 	return ret;
2645 }
2646 
2647 static void
2648 nvme_fc_rport_active_on_lport(struct nvme_fc_rport *rport)
2649 {
2650 	struct nvme_fc_lport *lport = rport->lport;
2651 
2652 	atomic_inc(&lport->act_rport_cnt);
2653 }
2654 
2655 static void
2656 nvme_fc_rport_inactive_on_lport(struct nvme_fc_rport *rport)
2657 {
2658 	struct nvme_fc_lport *lport = rport->lport;
2659 	u32 cnt;
2660 
2661 	cnt = atomic_dec_return(&lport->act_rport_cnt);
2662 	if (cnt == 0 && lport->localport.port_state == FC_OBJSTATE_DELETED)
2663 		lport->ops->localport_delete(&lport->localport);
2664 }
2665 
2666 static int
2667 nvme_fc_ctlr_active_on_rport(struct nvme_fc_ctrl *ctrl)
2668 {
2669 	struct nvme_fc_rport *rport = ctrl->rport;
2670 	u32 cnt;
2671 
2672 	if (ctrl->assoc_active)
2673 		return 1;
2674 
2675 	ctrl->assoc_active = true;
2676 	cnt = atomic_inc_return(&rport->act_ctrl_cnt);
2677 	if (cnt == 1)
2678 		nvme_fc_rport_active_on_lport(rport);
2679 
2680 	return 0;
2681 }
2682 
2683 static int
2684 nvme_fc_ctlr_inactive_on_rport(struct nvme_fc_ctrl *ctrl)
2685 {
2686 	struct nvme_fc_rport *rport = ctrl->rport;
2687 	struct nvme_fc_lport *lport = rport->lport;
2688 	u32 cnt;
2689 
2690 	/* ctrl->assoc_active=false will be set independently */
2691 
2692 	cnt = atomic_dec_return(&rport->act_ctrl_cnt);
2693 	if (cnt == 0) {
2694 		if (rport->remoteport.port_state == FC_OBJSTATE_DELETED)
2695 			lport->ops->remoteport_delete(&rport->remoteport);
2696 		nvme_fc_rport_inactive_on_lport(rport);
2697 	}
2698 
2699 	return 0;
2700 }
2701 
2702 /*
2703  * This routine restarts the controller on the host side, and
2704  * on the link side, recreates the controller association.
2705  */
2706 static int
2707 nvme_fc_create_association(struct nvme_fc_ctrl *ctrl)
2708 {
2709 	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2710 	int ret;
2711 	bool changed;
2712 
2713 	++ctrl->ctrl.nr_reconnects;
2714 
2715 	if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
2716 		return -ENODEV;
2717 
2718 	if (nvme_fc_ctlr_active_on_rport(ctrl))
2719 		return -ENOTUNIQ;
2720 
2721 	/*
2722 	 * Create the admin queue
2723 	 */
2724 
2725 	nvme_fc_init_queue(ctrl, 0);
2726 
2727 	ret = __nvme_fc_create_hw_queue(ctrl, &ctrl->queues[0], 0,
2728 				NVME_AQ_BLK_MQ_DEPTH);
2729 	if (ret)
2730 		goto out_free_queue;
2731 
2732 	ret = nvme_fc_connect_admin_queue(ctrl, &ctrl->queues[0],
2733 				NVME_AQ_BLK_MQ_DEPTH,
2734 				(NVME_AQ_BLK_MQ_DEPTH / 4));
2735 	if (ret)
2736 		goto out_delete_hw_queue;
2737 
2738 	if (ctrl->ctrl.state != NVME_CTRL_NEW)
2739 		blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2740 
2741 	ret = nvmf_connect_admin_queue(&ctrl->ctrl);
2742 	if (ret)
2743 		goto out_disconnect_admin_queue;
2744 
2745 	set_bit(NVME_FC_Q_LIVE, &ctrl->queues[0].flags);
2746 
2747 	/*
2748 	 * Check controller capabilities
2749 	 *
2750 	 * todo:- add code to check if ctrl attributes changed from
2751 	 * prior connection values
2752 	 */
2753 
2754 	ret = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->ctrl.cap);
2755 	if (ret) {
2756 		dev_err(ctrl->ctrl.device,
2757 			"prop_get NVME_REG_CAP failed\n");
2758 		goto out_disconnect_admin_queue;
2759 	}
2760 
2761 	ctrl->ctrl.sqsize =
2762 		min_t(int, NVME_CAP_MQES(ctrl->ctrl.cap) + 1, ctrl->ctrl.sqsize);
2763 
2764 	ret = nvme_enable_ctrl(&ctrl->ctrl, ctrl->ctrl.cap);
2765 	if (ret)
2766 		goto out_disconnect_admin_queue;
2767 
2768 	ctrl->ctrl.max_hw_sectors =
2769 		(ctrl->lport->ops->max_sgl_segments - 1) << (PAGE_SHIFT - 9);
2770 
2771 	ret = nvme_init_identify(&ctrl->ctrl);
2772 	if (ret)
2773 		goto out_disconnect_admin_queue;
2774 
2775 	/* sanity checks */
2776 
2777 	/* FC-NVME does not have other data in the capsule */
2778 	if (ctrl->ctrl.icdoff) {
2779 		dev_err(ctrl->ctrl.device, "icdoff %d is not supported!\n",
2780 				ctrl->ctrl.icdoff);
2781 		goto out_disconnect_admin_queue;
2782 	}
2783 
2784 	/* FC-NVME supports normal SGL Data Block Descriptors */
2785 
2786 	if (opts->queue_size > ctrl->ctrl.maxcmd) {
2787 		/* warn if maxcmd is lower than queue_size */
2788 		dev_warn(ctrl->ctrl.device,
2789 			"queue_size %zu > ctrl maxcmd %u, reducing "
2790 			"to queue_size\n",
2791 			opts->queue_size, ctrl->ctrl.maxcmd);
2792 		opts->queue_size = ctrl->ctrl.maxcmd;
2793 	}
2794 
2795 	ret = nvme_fc_init_aen_ops(ctrl);
2796 	if (ret)
2797 		goto out_term_aen_ops;
2798 
2799 	/*
2800 	 * Create the io queues
2801 	 */
2802 
2803 	if (ctrl->ctrl.queue_count > 1) {
2804 		if (ctrl->ctrl.state == NVME_CTRL_NEW)
2805 			ret = nvme_fc_create_io_queues(ctrl);
2806 		else
2807 			ret = nvme_fc_reinit_io_queues(ctrl);
2808 		if (ret)
2809 			goto out_term_aen_ops;
2810 	}
2811 
2812 	changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE);
2813 
2814 	ctrl->ctrl.nr_reconnects = 0;
2815 
2816 	if (changed)
2817 		nvme_start_ctrl(&ctrl->ctrl);
2818 
2819 	return 0;	/* Success */
2820 
2821 out_term_aen_ops:
2822 	nvme_fc_term_aen_ops(ctrl);
2823 out_disconnect_admin_queue:
2824 	/* send a Disconnect(association) LS to fc-nvme target */
2825 	nvme_fc_xmt_disconnect_assoc(ctrl);
2826 out_delete_hw_queue:
2827 	__nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2828 out_free_queue:
2829 	nvme_fc_free_queue(&ctrl->queues[0]);
2830 	ctrl->assoc_active = false;
2831 	nvme_fc_ctlr_inactive_on_rport(ctrl);
2832 
2833 	return ret;
2834 }
2835 
2836 /*
2837  * This routine stops operation of the controller on the host side.
2838  * On the host os stack side: Admin and IO queues are stopped,
2839  *   outstanding ios on them terminated via FC ABTS.
2840  * On the link side: the association is terminated.
2841  */
2842 static void
2843 nvme_fc_delete_association(struct nvme_fc_ctrl *ctrl)
2844 {
2845 	unsigned long flags;
2846 
2847 	if (!ctrl->assoc_active)
2848 		return;
2849 	ctrl->assoc_active = false;
2850 
2851 	spin_lock_irqsave(&ctrl->lock, flags);
2852 	ctrl->flags |= FCCTRL_TERMIO;
2853 	ctrl->iocnt = 0;
2854 	spin_unlock_irqrestore(&ctrl->lock, flags);
2855 
2856 	/*
2857 	 * If io queues are present, stop them and terminate all outstanding
2858 	 * ios on them. As FC allocates FC exchange for each io, the
2859 	 * transport must contact the LLDD to terminate the exchange,
2860 	 * thus releasing the FC exchange. We use blk_mq_tagset_busy_itr()
2861 	 * to tell us what io's are busy and invoke a transport routine
2862 	 * to kill them with the LLDD.  After terminating the exchange
2863 	 * the LLDD will call the transport's normal io done path, but it
2864 	 * will have an aborted status. The done path will return the
2865 	 * io requests back to the block layer as part of normal completions
2866 	 * (but with error status).
2867 	 */
2868 	if (ctrl->ctrl.queue_count > 1) {
2869 		nvme_stop_queues(&ctrl->ctrl);
2870 		blk_mq_tagset_busy_iter(&ctrl->tag_set,
2871 				nvme_fc_terminate_exchange, &ctrl->ctrl);
2872 	}
2873 
2874 	/*
2875 	 * Other transports, which don't have link-level contexts bound
2876 	 * to sqe's, would try to gracefully shutdown the controller by
2877 	 * writing the registers for shutdown and polling (call
2878 	 * nvme_shutdown_ctrl()). Given a bunch of i/o was potentially
2879 	 * just aborted and we will wait on those contexts, and given
2880 	 * there was no indication of how live the controlelr is on the
2881 	 * link, don't send more io to create more contexts for the
2882 	 * shutdown. Let the controller fail via keepalive failure if
2883 	 * its still present.
2884 	 */
2885 
2886 	/*
2887 	 * clean up the admin queue. Same thing as above.
2888 	 * use blk_mq_tagset_busy_itr() and the transport routine to
2889 	 * terminate the exchanges.
2890 	 */
2891 	if (ctrl->ctrl.state != NVME_CTRL_NEW)
2892 		blk_mq_quiesce_queue(ctrl->ctrl.admin_q);
2893 	blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
2894 				nvme_fc_terminate_exchange, &ctrl->ctrl);
2895 
2896 	/* kill the aens as they are a separate path */
2897 	nvme_fc_abort_aen_ops(ctrl);
2898 
2899 	/* wait for all io that had to be aborted */
2900 	spin_lock_irq(&ctrl->lock);
2901 	wait_event_lock_irq(ctrl->ioabort_wait, ctrl->iocnt == 0, ctrl->lock);
2902 	ctrl->flags &= ~FCCTRL_TERMIO;
2903 	spin_unlock_irq(&ctrl->lock);
2904 
2905 	nvme_fc_term_aen_ops(ctrl);
2906 
2907 	/*
2908 	 * send a Disconnect(association) LS to fc-nvme target
2909 	 * Note: could have been sent at top of process, but
2910 	 * cleaner on link traffic if after the aborts complete.
2911 	 * Note: if association doesn't exist, association_id will be 0
2912 	 */
2913 	if (ctrl->association_id)
2914 		nvme_fc_xmt_disconnect_assoc(ctrl);
2915 
2916 	if (ctrl->ctrl.tagset) {
2917 		nvme_fc_delete_hw_io_queues(ctrl);
2918 		nvme_fc_free_io_queues(ctrl);
2919 	}
2920 
2921 	__nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2922 	nvme_fc_free_queue(&ctrl->queues[0]);
2923 
2924 	nvme_fc_ctlr_inactive_on_rport(ctrl);
2925 }
2926 
2927 static void
2928 nvme_fc_delete_ctrl(struct nvme_ctrl *nctrl)
2929 {
2930 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2931 
2932 	cancel_delayed_work_sync(&ctrl->connect_work);
2933 	/*
2934 	 * kill the association on the link side.  this will block
2935 	 * waiting for io to terminate
2936 	 */
2937 	nvme_fc_delete_association(ctrl);
2938 }
2939 
2940 static void
2941 nvme_fc_reconnect_or_delete(struct nvme_fc_ctrl *ctrl, int status)
2942 {
2943 	struct nvme_fc_rport *rport = ctrl->rport;
2944 	struct nvme_fc_remote_port *portptr = &rport->remoteport;
2945 	unsigned long recon_delay = ctrl->ctrl.opts->reconnect_delay * HZ;
2946 	bool recon = true;
2947 
2948 	if (ctrl->ctrl.state != NVME_CTRL_RECONNECTING)
2949 		return;
2950 
2951 	if (portptr->port_state == FC_OBJSTATE_ONLINE)
2952 		dev_info(ctrl->ctrl.device,
2953 			"NVME-FC{%d}: reset: Reconnect attempt failed (%d)\n",
2954 			ctrl->cnum, status);
2955 	else if (time_after_eq(jiffies, rport->dev_loss_end))
2956 		recon = false;
2957 
2958 	if (recon && nvmf_should_reconnect(&ctrl->ctrl)) {
2959 		if (portptr->port_state == FC_OBJSTATE_ONLINE)
2960 			dev_info(ctrl->ctrl.device,
2961 				"NVME-FC{%d}: Reconnect attempt in %ld "
2962 				"seconds\n",
2963 				ctrl->cnum, recon_delay / HZ);
2964 		else if (time_after(jiffies + recon_delay, rport->dev_loss_end))
2965 			recon_delay = rport->dev_loss_end - jiffies;
2966 
2967 		queue_delayed_work(nvme_wq, &ctrl->connect_work, recon_delay);
2968 	} else {
2969 		if (portptr->port_state == FC_OBJSTATE_ONLINE)
2970 			dev_warn(ctrl->ctrl.device,
2971 				"NVME-FC{%d}: Max reconnect attempts (%d) "
2972 				"reached. Removing controller\n",
2973 				ctrl->cnum, ctrl->ctrl.nr_reconnects);
2974 		else
2975 			dev_warn(ctrl->ctrl.device,
2976 				"NVME-FC{%d}: dev_loss_tmo (%d) expired "
2977 				"while waiting for remoteport connectivity. "
2978 				"Removing controller\n", ctrl->cnum,
2979 				portptr->dev_loss_tmo);
2980 		WARN_ON(nvme_delete_ctrl(&ctrl->ctrl));
2981 	}
2982 }
2983 
2984 static void
2985 nvme_fc_reset_ctrl_work(struct work_struct *work)
2986 {
2987 	struct nvme_fc_ctrl *ctrl =
2988 		container_of(work, struct nvme_fc_ctrl, ctrl.reset_work);
2989 	int ret;
2990 
2991 	nvme_stop_ctrl(&ctrl->ctrl);
2992 
2993 	/* will block will waiting for io to terminate */
2994 	nvme_fc_delete_association(ctrl);
2995 
2996 	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RECONNECTING)) {
2997 		dev_err(ctrl->ctrl.device,
2998 			"NVME-FC{%d}: error_recovery: Couldn't change state "
2999 			"to RECONNECTING\n", ctrl->cnum);
3000 		return;
3001 	}
3002 
3003 	if (ctrl->rport->remoteport.port_state == FC_OBJSTATE_ONLINE)
3004 		ret = nvme_fc_create_association(ctrl);
3005 	else
3006 		ret = -ENOTCONN;
3007 
3008 	if (ret)
3009 		nvme_fc_reconnect_or_delete(ctrl, ret);
3010 	else
3011 		dev_info(ctrl->ctrl.device,
3012 			"NVME-FC{%d}: controller reset complete\n",
3013 			ctrl->cnum);
3014 }
3015 
3016 static const struct nvme_ctrl_ops nvme_fc_ctrl_ops = {
3017 	.name			= "fc",
3018 	.module			= THIS_MODULE,
3019 	.flags			= NVME_F_FABRICS,
3020 	.reg_read32		= nvmf_reg_read32,
3021 	.reg_read64		= nvmf_reg_read64,
3022 	.reg_write32		= nvmf_reg_write32,
3023 	.free_ctrl		= nvme_fc_nvme_ctrl_freed,
3024 	.submit_async_event	= nvme_fc_submit_async_event,
3025 	.delete_ctrl		= nvme_fc_delete_ctrl,
3026 	.get_address		= nvmf_get_address,
3027 	.reinit_request		= nvme_fc_reinit_request,
3028 };
3029 
3030 static void
3031 nvme_fc_connect_ctrl_work(struct work_struct *work)
3032 {
3033 	int ret;
3034 
3035 	struct nvme_fc_ctrl *ctrl =
3036 			container_of(to_delayed_work(work),
3037 				struct nvme_fc_ctrl, connect_work);
3038 
3039 	ret = nvme_fc_create_association(ctrl);
3040 	if (ret)
3041 		nvme_fc_reconnect_or_delete(ctrl, ret);
3042 	else
3043 		dev_info(ctrl->ctrl.device,
3044 			"NVME-FC{%d}: controller reconnect complete\n",
3045 			ctrl->cnum);
3046 }
3047 
3048 
3049 static const struct blk_mq_ops nvme_fc_admin_mq_ops = {
3050 	.queue_rq	= nvme_fc_queue_rq,
3051 	.complete	= nvme_fc_complete_rq,
3052 	.init_request	= nvme_fc_init_request,
3053 	.exit_request	= nvme_fc_exit_request,
3054 	.init_hctx	= nvme_fc_init_admin_hctx,
3055 	.timeout	= nvme_fc_timeout,
3056 };
3057 
3058 
3059 /*
3060  * Fails a controller request if it matches an existing controller
3061  * (association) with the same tuple:
3062  * <Host NQN, Host ID, local FC port, remote FC port, SUBSYS NQN>
3063  *
3064  * The ports don't need to be compared as they are intrinsically
3065  * already matched by the port pointers supplied.
3066  */
3067 static bool
3068 nvme_fc_existing_controller(struct nvme_fc_rport *rport,
3069 		struct nvmf_ctrl_options *opts)
3070 {
3071 	struct nvme_fc_ctrl *ctrl;
3072 	unsigned long flags;
3073 	bool found = false;
3074 
3075 	spin_lock_irqsave(&rport->lock, flags);
3076 	list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list) {
3077 		found = nvmf_ctlr_matches_baseopts(&ctrl->ctrl, opts);
3078 		if (found)
3079 			break;
3080 	}
3081 	spin_unlock_irqrestore(&rport->lock, flags);
3082 
3083 	return found;
3084 }
3085 
3086 static struct nvme_ctrl *
3087 nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,
3088 	struct nvme_fc_lport *lport, struct nvme_fc_rport *rport)
3089 {
3090 	struct nvme_fc_ctrl *ctrl;
3091 	unsigned long flags;
3092 	int ret, idx, retry;
3093 
3094 	if (!(rport->remoteport.port_role &
3095 	    (FC_PORT_ROLE_NVME_DISCOVERY | FC_PORT_ROLE_NVME_TARGET))) {
3096 		ret = -EBADR;
3097 		goto out_fail;
3098 	}
3099 
3100 	if (!opts->duplicate_connect &&
3101 	    nvme_fc_existing_controller(rport, opts)) {
3102 		ret = -EALREADY;
3103 		goto out_fail;
3104 	}
3105 
3106 	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
3107 	if (!ctrl) {
3108 		ret = -ENOMEM;
3109 		goto out_fail;
3110 	}
3111 
3112 	idx = ida_simple_get(&nvme_fc_ctrl_cnt, 0, 0, GFP_KERNEL);
3113 	if (idx < 0) {
3114 		ret = -ENOSPC;
3115 		goto out_free_ctrl;
3116 	}
3117 
3118 	ctrl->ctrl.opts = opts;
3119 	INIT_LIST_HEAD(&ctrl->ctrl_list);
3120 	ctrl->lport = lport;
3121 	ctrl->rport = rport;
3122 	ctrl->dev = lport->dev;
3123 	ctrl->cnum = idx;
3124 	ctrl->assoc_active = false;
3125 	init_waitqueue_head(&ctrl->ioabort_wait);
3126 
3127 	get_device(ctrl->dev);
3128 	kref_init(&ctrl->ref);
3129 
3130 	INIT_WORK(&ctrl->ctrl.reset_work, nvme_fc_reset_ctrl_work);
3131 	INIT_DELAYED_WORK(&ctrl->connect_work, nvme_fc_connect_ctrl_work);
3132 	spin_lock_init(&ctrl->lock);
3133 
3134 	/* io queue count */
3135 	ctrl->ctrl.queue_count = min_t(unsigned int,
3136 				opts->nr_io_queues,
3137 				lport->ops->max_hw_queues);
3138 	ctrl->ctrl.queue_count++;	/* +1 for admin queue */
3139 
3140 	ctrl->ctrl.sqsize = opts->queue_size - 1;
3141 	ctrl->ctrl.kato = opts->kato;
3142 
3143 	ret = -ENOMEM;
3144 	ctrl->queues = kcalloc(ctrl->ctrl.queue_count,
3145 				sizeof(struct nvme_fc_queue), GFP_KERNEL);
3146 	if (!ctrl->queues)
3147 		goto out_free_ida;
3148 
3149 	memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set));
3150 	ctrl->admin_tag_set.ops = &nvme_fc_admin_mq_ops;
3151 	ctrl->admin_tag_set.queue_depth = NVME_AQ_MQ_TAG_DEPTH;
3152 	ctrl->admin_tag_set.reserved_tags = 2; /* fabric connect + Keep-Alive */
3153 	ctrl->admin_tag_set.numa_node = NUMA_NO_NODE;
3154 	ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
3155 					(SG_CHUNK_SIZE *
3156 						sizeof(struct scatterlist)) +
3157 					ctrl->lport->ops->fcprqst_priv_sz;
3158 	ctrl->admin_tag_set.driver_data = ctrl;
3159 	ctrl->admin_tag_set.nr_hw_queues = 1;
3160 	ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT;
3161 	ctrl->admin_tag_set.flags = BLK_MQ_F_NO_SCHED;
3162 
3163 	ret = blk_mq_alloc_tag_set(&ctrl->admin_tag_set);
3164 	if (ret)
3165 		goto out_free_queues;
3166 	ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set;
3167 
3168 	ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);
3169 	if (IS_ERR(ctrl->ctrl.admin_q)) {
3170 		ret = PTR_ERR(ctrl->ctrl.admin_q);
3171 		goto out_free_admin_tag_set;
3172 	}
3173 
3174 	/*
3175 	 * Would have been nice to init io queues tag set as well.
3176 	 * However, we require interaction from the controller
3177 	 * for max io queue count before we can do so.
3178 	 * Defer this to the connect path.
3179 	 */
3180 
3181 	ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_fc_ctrl_ops, 0);
3182 	if (ret)
3183 		goto out_cleanup_admin_q;
3184 
3185 	/* at this point, teardown path changes to ref counting on nvme ctrl */
3186 
3187 	spin_lock_irqsave(&rport->lock, flags);
3188 	list_add_tail(&ctrl->ctrl_list, &rport->ctrl_list);
3189 	spin_unlock_irqrestore(&rport->lock, flags);
3190 
3191 	/*
3192 	 * It's possible that transactions used to create the association
3193 	 * may fail. Examples: CreateAssociation LS or CreateIOConnection
3194 	 * LS gets dropped/corrupted/fails; or a frame gets dropped or a
3195 	 * command times out for one of the actions to init the controller
3196 	 * (Connect, Get/Set_Property, Set_Features, etc). Many of these
3197 	 * transport errors (frame drop, LS failure) inherently must kill
3198 	 * the association. The transport is coded so that any command used
3199 	 * to create the association (prior to a LIVE state transition
3200 	 * while NEW or RECONNECTING) will fail if it completes in error or
3201 	 * times out.
3202 	 *
3203 	 * As such: as the connect request was mostly likely due to a
3204 	 * udev event that discovered the remote port, meaning there is
3205 	 * not an admin or script there to restart if the connect
3206 	 * request fails, retry the initial connection creation up to
3207 	 * three times before giving up and declaring failure.
3208 	 */
3209 	for (retry = 0; retry < 3; retry++) {
3210 		ret = nvme_fc_create_association(ctrl);
3211 		if (!ret)
3212 			break;
3213 	}
3214 
3215 	if (ret) {
3216 		/* couldn't schedule retry - fail out */
3217 		dev_err(ctrl->ctrl.device,
3218 			"NVME-FC{%d}: Connect retry failed\n", ctrl->cnum);
3219 
3220 		ctrl->ctrl.opts = NULL;
3221 
3222 		/* initiate nvme ctrl ref counting teardown */
3223 		nvme_uninit_ctrl(&ctrl->ctrl);
3224 		nvme_put_ctrl(&ctrl->ctrl);
3225 
3226 		/* Remove core ctrl ref. */
3227 		nvme_put_ctrl(&ctrl->ctrl);
3228 
3229 		/* as we're past the point where we transition to the ref
3230 		 * counting teardown path, if we return a bad pointer here,
3231 		 * the calling routine, thinking it's prior to the
3232 		 * transition, will do an rport put. Since the teardown
3233 		 * path also does a rport put, we do an extra get here to
3234 		 * so proper order/teardown happens.
3235 		 */
3236 		nvme_fc_rport_get(rport);
3237 
3238 		if (ret > 0)
3239 			ret = -EIO;
3240 		return ERR_PTR(ret);
3241 	}
3242 
3243 	nvme_get_ctrl(&ctrl->ctrl);
3244 
3245 	dev_info(ctrl->ctrl.device,
3246 		"NVME-FC{%d}: new ctrl: NQN \"%s\"\n",
3247 		ctrl->cnum, ctrl->ctrl.opts->subsysnqn);
3248 
3249 	return &ctrl->ctrl;
3250 
3251 out_cleanup_admin_q:
3252 	blk_cleanup_queue(ctrl->ctrl.admin_q);
3253 out_free_admin_tag_set:
3254 	blk_mq_free_tag_set(&ctrl->admin_tag_set);
3255 out_free_queues:
3256 	kfree(ctrl->queues);
3257 out_free_ida:
3258 	put_device(ctrl->dev);
3259 	ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
3260 out_free_ctrl:
3261 	kfree(ctrl);
3262 out_fail:
3263 	/* exit via here doesn't follow ctlr ref points */
3264 	return ERR_PTR(ret);
3265 }
3266 
3267 
3268 struct nvmet_fc_traddr {
3269 	u64	nn;
3270 	u64	pn;
3271 };
3272 
3273 static int
3274 __nvme_fc_parse_u64(substring_t *sstr, u64 *val)
3275 {
3276 	u64 token64;
3277 
3278 	if (match_u64(sstr, &token64))
3279 		return -EINVAL;
3280 	*val = token64;
3281 
3282 	return 0;
3283 }
3284 
3285 /*
3286  * This routine validates and extracts the WWN's from the TRADDR string.
3287  * As kernel parsers need the 0x to determine number base, universally
3288  * build string to parse with 0x prefix before parsing name strings.
3289  */
3290 static int
3291 nvme_fc_parse_traddr(struct nvmet_fc_traddr *traddr, char *buf, size_t blen)
3292 {
3293 	char name[2 + NVME_FC_TRADDR_HEXNAMELEN + 1];
3294 	substring_t wwn = { name, &name[sizeof(name)-1] };
3295 	int nnoffset, pnoffset;
3296 
3297 	/* validate it string one of the 2 allowed formats */
3298 	if (strnlen(buf, blen) == NVME_FC_TRADDR_MAXLENGTH &&
3299 			!strncmp(buf, "nn-0x", NVME_FC_TRADDR_OXNNLEN) &&
3300 			!strncmp(&buf[NVME_FC_TRADDR_MAX_PN_OFFSET],
3301 				"pn-0x", NVME_FC_TRADDR_OXNNLEN)) {
3302 		nnoffset = NVME_FC_TRADDR_OXNNLEN;
3303 		pnoffset = NVME_FC_TRADDR_MAX_PN_OFFSET +
3304 						NVME_FC_TRADDR_OXNNLEN;
3305 	} else if ((strnlen(buf, blen) == NVME_FC_TRADDR_MINLENGTH &&
3306 			!strncmp(buf, "nn-", NVME_FC_TRADDR_NNLEN) &&
3307 			!strncmp(&buf[NVME_FC_TRADDR_MIN_PN_OFFSET],
3308 				"pn-", NVME_FC_TRADDR_NNLEN))) {
3309 		nnoffset = NVME_FC_TRADDR_NNLEN;
3310 		pnoffset = NVME_FC_TRADDR_MIN_PN_OFFSET + NVME_FC_TRADDR_NNLEN;
3311 	} else
3312 		goto out_einval;
3313 
3314 	name[0] = '0';
3315 	name[1] = 'x';
3316 	name[2 + NVME_FC_TRADDR_HEXNAMELEN] = 0;
3317 
3318 	memcpy(&name[2], &buf[nnoffset], NVME_FC_TRADDR_HEXNAMELEN);
3319 	if (__nvme_fc_parse_u64(&wwn, &traddr->nn))
3320 		goto out_einval;
3321 
3322 	memcpy(&name[2], &buf[pnoffset], NVME_FC_TRADDR_HEXNAMELEN);
3323 	if (__nvme_fc_parse_u64(&wwn, &traddr->pn))
3324 		goto out_einval;
3325 
3326 	return 0;
3327 
3328 out_einval:
3329 	pr_warn("%s: bad traddr string\n", __func__);
3330 	return -EINVAL;
3331 }
3332 
3333 static struct nvme_ctrl *
3334 nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts)
3335 {
3336 	struct nvme_fc_lport *lport;
3337 	struct nvme_fc_rport *rport;
3338 	struct nvme_ctrl *ctrl;
3339 	struct nvmet_fc_traddr laddr = { 0L, 0L };
3340 	struct nvmet_fc_traddr raddr = { 0L, 0L };
3341 	unsigned long flags;
3342 	int ret;
3343 
3344 	ret = nvme_fc_parse_traddr(&raddr, opts->traddr, NVMF_TRADDR_SIZE);
3345 	if (ret || !raddr.nn || !raddr.pn)
3346 		return ERR_PTR(-EINVAL);
3347 
3348 	ret = nvme_fc_parse_traddr(&laddr, opts->host_traddr, NVMF_TRADDR_SIZE);
3349 	if (ret || !laddr.nn || !laddr.pn)
3350 		return ERR_PTR(-EINVAL);
3351 
3352 	/* find the host and remote ports to connect together */
3353 	spin_lock_irqsave(&nvme_fc_lock, flags);
3354 	list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
3355 		if (lport->localport.node_name != laddr.nn ||
3356 		    lport->localport.port_name != laddr.pn)
3357 			continue;
3358 
3359 		list_for_each_entry(rport, &lport->endp_list, endp_list) {
3360 			if (rport->remoteport.node_name != raddr.nn ||
3361 			    rport->remoteport.port_name != raddr.pn)
3362 				continue;
3363 
3364 			/* if fail to get reference fall through. Will error */
3365 			if (!nvme_fc_rport_get(rport))
3366 				break;
3367 
3368 			spin_unlock_irqrestore(&nvme_fc_lock, flags);
3369 
3370 			ctrl = nvme_fc_init_ctrl(dev, opts, lport, rport);
3371 			if (IS_ERR(ctrl))
3372 				nvme_fc_rport_put(rport);
3373 			return ctrl;
3374 		}
3375 	}
3376 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
3377 
3378 	return ERR_PTR(-ENOENT);
3379 }
3380 
3381 
3382 static struct nvmf_transport_ops nvme_fc_transport = {
3383 	.name		= "fc",
3384 	.required_opts	= NVMF_OPT_TRADDR | NVMF_OPT_HOST_TRADDR,
3385 	.allowed_opts	= NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_CTRL_LOSS_TMO,
3386 	.create_ctrl	= nvme_fc_create_ctrl,
3387 };
3388 
3389 static int __init nvme_fc_init_module(void)
3390 {
3391 	int ret;
3392 
3393 	/*
3394 	 * NOTE:
3395 	 * It is expected that in the future the kernel will combine
3396 	 * the FC-isms that are currently under scsi and now being
3397 	 * added to by NVME into a new standalone FC class. The SCSI
3398 	 * and NVME protocols and their devices would be under this
3399 	 * new FC class.
3400 	 *
3401 	 * As we need something to post FC-specific udev events to,
3402 	 * specifically for nvme probe events, start by creating the
3403 	 * new device class.  When the new standalone FC class is
3404 	 * put in place, this code will move to a more generic
3405 	 * location for the class.
3406 	 */
3407 	fc_class = class_create(THIS_MODULE, "fc");
3408 	if (IS_ERR(fc_class)) {
3409 		pr_err("couldn't register class fc\n");
3410 		return PTR_ERR(fc_class);
3411 	}
3412 
3413 	/*
3414 	 * Create a device for the FC-centric udev events
3415 	 */
3416 	fc_udev_device = device_create(fc_class, NULL, MKDEV(0, 0), NULL,
3417 				"fc_udev_device");
3418 	if (IS_ERR(fc_udev_device)) {
3419 		pr_err("couldn't create fc_udev device!\n");
3420 		ret = PTR_ERR(fc_udev_device);
3421 		goto out_destroy_class;
3422 	}
3423 
3424 	ret = nvmf_register_transport(&nvme_fc_transport);
3425 	if (ret)
3426 		goto out_destroy_device;
3427 
3428 	return 0;
3429 
3430 out_destroy_device:
3431 	device_destroy(fc_class, MKDEV(0, 0));
3432 out_destroy_class:
3433 	class_destroy(fc_class);
3434 	return ret;
3435 }
3436 
3437 static void __exit nvme_fc_exit_module(void)
3438 {
3439 	/* sanity check - all lports should be removed */
3440 	if (!list_empty(&nvme_fc_lport_list))
3441 		pr_warn("%s: localport list not empty\n", __func__);
3442 
3443 	nvmf_unregister_transport(&nvme_fc_transport);
3444 
3445 	ida_destroy(&nvme_fc_local_port_cnt);
3446 	ida_destroy(&nvme_fc_ctrl_cnt);
3447 
3448 	device_destroy(fc_class, MKDEV(0, 0));
3449 	class_destroy(fc_class);
3450 }
3451 
3452 module_init(nvme_fc_init_module);
3453 module_exit(nvme_fc_exit_module);
3454 
3455 MODULE_LICENSE("GPL v2");
3456