xref: /openbmc/linux/drivers/nvme/host/fc.c (revision ddc141e5)
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 NVME_FC_DEFAULT_DEV_LOSS_TMO	60	/* seconds */
39 
40 struct nvme_fc_queue {
41 	struct nvme_fc_ctrl	*ctrl;
42 	struct device		*dev;
43 	struct blk_mq_hw_ctx	*hctx;
44 	void			*lldd_handle;
45 	size_t			cmnd_capsule_len;
46 	u32			qnum;
47 	u32			rqcnt;
48 	u32			seqno;
49 
50 	u64			connection_id;
51 	atomic_t		csn;
52 
53 	unsigned long		flags;
54 } __aligned(sizeof(u64));	/* alignment for other things alloc'd with */
55 
56 enum nvme_fcop_flags {
57 	FCOP_FLAGS_TERMIO	= (1 << 0),
58 	FCOP_FLAGS_AEN		= (1 << 1),
59 };
60 
61 struct nvmefc_ls_req_op {
62 	struct nvmefc_ls_req	ls_req;
63 
64 	struct nvme_fc_rport	*rport;
65 	struct nvme_fc_queue	*queue;
66 	struct request		*rq;
67 	u32			flags;
68 
69 	int			ls_error;
70 	struct completion	ls_done;
71 	struct list_head	lsreq_list;	/* rport->ls_req_list */
72 	bool			req_queued;
73 };
74 
75 enum nvme_fcpop_state {
76 	FCPOP_STATE_UNINIT	= 0,
77 	FCPOP_STATE_IDLE	= 1,
78 	FCPOP_STATE_ACTIVE	= 2,
79 	FCPOP_STATE_ABORTED	= 3,
80 	FCPOP_STATE_COMPLETE	= 4,
81 };
82 
83 struct nvme_fc_fcp_op {
84 	struct nvme_request	nreq;		/*
85 						 * nvme/host/core.c
86 						 * requires this to be
87 						 * the 1st element in the
88 						 * private structure
89 						 * associated with the
90 						 * request.
91 						 */
92 	struct nvmefc_fcp_req	fcp_req;
93 
94 	struct nvme_fc_ctrl	*ctrl;
95 	struct nvme_fc_queue	*queue;
96 	struct request		*rq;
97 
98 	atomic_t		state;
99 	u32			flags;
100 	u32			rqno;
101 	u32			nents;
102 
103 	struct nvme_fc_cmd_iu	cmd_iu;
104 	struct nvme_fc_ersp_iu	rsp_iu;
105 };
106 
107 struct nvme_fc_lport {
108 	struct nvme_fc_local_port	localport;
109 
110 	struct ida			endp_cnt;
111 	struct list_head		port_list;	/* nvme_fc_port_list */
112 	struct list_head		endp_list;
113 	struct device			*dev;	/* physical device for dma */
114 	struct nvme_fc_port_template	*ops;
115 	struct kref			ref;
116 	atomic_t                        act_rport_cnt;
117 } __aligned(sizeof(u64));	/* alignment for other things alloc'd with */
118 
119 struct nvme_fc_rport {
120 	struct nvme_fc_remote_port	remoteport;
121 
122 	struct list_head		endp_list; /* for lport->endp_list */
123 	struct list_head		ctrl_list;
124 	struct list_head		ls_req_list;
125 	struct device			*dev;	/* physical device for dma */
126 	struct nvme_fc_lport		*lport;
127 	spinlock_t			lock;
128 	struct kref			ref;
129 	atomic_t                        act_ctrl_cnt;
130 	unsigned long			dev_loss_end;
131 } __aligned(sizeof(u64));	/* alignment for other things alloc'd with */
132 
133 enum nvme_fcctrl_flags {
134 	FCCTRL_TERMIO		= (1 << 0),
135 };
136 
137 struct nvme_fc_ctrl {
138 	spinlock_t		lock;
139 	struct nvme_fc_queue	*queues;
140 	struct device		*dev;
141 	struct nvme_fc_lport	*lport;
142 	struct nvme_fc_rport	*rport;
143 	u32			cnum;
144 
145 	bool			assoc_active;
146 	u64			association_id;
147 
148 	struct list_head	ctrl_list;	/* rport->ctrl_list */
149 
150 	struct blk_mq_tag_set	admin_tag_set;
151 	struct blk_mq_tag_set	tag_set;
152 
153 	struct delayed_work	connect_work;
154 
155 	struct kref		ref;
156 	u32			flags;
157 	u32			iocnt;
158 	wait_queue_head_t	ioabort_wait;
159 
160 	struct nvme_fc_fcp_op	aen_ops[NVME_NR_AEN_COMMANDS];
161 
162 	struct nvme_ctrl	ctrl;
163 };
164 
165 static inline struct nvme_fc_ctrl *
166 to_fc_ctrl(struct nvme_ctrl *ctrl)
167 {
168 	return container_of(ctrl, struct nvme_fc_ctrl, ctrl);
169 }
170 
171 static inline struct nvme_fc_lport *
172 localport_to_lport(struct nvme_fc_local_port *portptr)
173 {
174 	return container_of(portptr, struct nvme_fc_lport, localport);
175 }
176 
177 static inline struct nvme_fc_rport *
178 remoteport_to_rport(struct nvme_fc_remote_port *portptr)
179 {
180 	return container_of(portptr, struct nvme_fc_rport, remoteport);
181 }
182 
183 static inline struct nvmefc_ls_req_op *
184 ls_req_to_lsop(struct nvmefc_ls_req *lsreq)
185 {
186 	return container_of(lsreq, struct nvmefc_ls_req_op, ls_req);
187 }
188 
189 static inline struct nvme_fc_fcp_op *
190 fcp_req_to_fcp_op(struct nvmefc_fcp_req *fcpreq)
191 {
192 	return container_of(fcpreq, struct nvme_fc_fcp_op, fcp_req);
193 }
194 
195 
196 
197 /* *************************** Globals **************************** */
198 
199 
200 static DEFINE_SPINLOCK(nvme_fc_lock);
201 
202 static LIST_HEAD(nvme_fc_lport_list);
203 static DEFINE_IDA(nvme_fc_local_port_cnt);
204 static DEFINE_IDA(nvme_fc_ctrl_cnt);
205 
206 
207 
208 /*
209  * These items are short-term. They will eventually be moved into
210  * a generic FC class. See comments in module init.
211  */
212 static struct class *fc_class;
213 static struct device *fc_udev_device;
214 
215 
216 /* *********************** FC-NVME Port Management ************************ */
217 
218 static void __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *,
219 			struct nvme_fc_queue *, unsigned int);
220 
221 static void
222 nvme_fc_free_lport(struct kref *ref)
223 {
224 	struct nvme_fc_lport *lport =
225 		container_of(ref, struct nvme_fc_lport, ref);
226 	unsigned long flags;
227 
228 	WARN_ON(lport->localport.port_state != FC_OBJSTATE_DELETED);
229 	WARN_ON(!list_empty(&lport->endp_list));
230 
231 	/* remove from transport list */
232 	spin_lock_irqsave(&nvme_fc_lock, flags);
233 	list_del(&lport->port_list);
234 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
235 
236 	ida_simple_remove(&nvme_fc_local_port_cnt, lport->localport.port_num);
237 	ida_destroy(&lport->endp_cnt);
238 
239 	put_device(lport->dev);
240 
241 	kfree(lport);
242 }
243 
244 static void
245 nvme_fc_lport_put(struct nvme_fc_lport *lport)
246 {
247 	kref_put(&lport->ref, nvme_fc_free_lport);
248 }
249 
250 static int
251 nvme_fc_lport_get(struct nvme_fc_lport *lport)
252 {
253 	return kref_get_unless_zero(&lport->ref);
254 }
255 
256 
257 static struct nvme_fc_lport *
258 nvme_fc_attach_to_unreg_lport(struct nvme_fc_port_info *pinfo,
259 			struct nvme_fc_port_template *ops,
260 			struct device *dev)
261 {
262 	struct nvme_fc_lport *lport;
263 	unsigned long flags;
264 
265 	spin_lock_irqsave(&nvme_fc_lock, flags);
266 
267 	list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
268 		if (lport->localport.node_name != pinfo->node_name ||
269 		    lport->localport.port_name != pinfo->port_name)
270 			continue;
271 
272 		if (lport->dev != dev) {
273 			lport = ERR_PTR(-EXDEV);
274 			goto out_done;
275 		}
276 
277 		if (lport->localport.port_state != FC_OBJSTATE_DELETED) {
278 			lport = ERR_PTR(-EEXIST);
279 			goto out_done;
280 		}
281 
282 		if (!nvme_fc_lport_get(lport)) {
283 			/*
284 			 * fails if ref cnt already 0. If so,
285 			 * act as if lport already deleted
286 			 */
287 			lport = NULL;
288 			goto out_done;
289 		}
290 
291 		/* resume the lport */
292 
293 		lport->ops = ops;
294 		lport->localport.port_role = pinfo->port_role;
295 		lport->localport.port_id = pinfo->port_id;
296 		lport->localport.port_state = FC_OBJSTATE_ONLINE;
297 
298 		spin_unlock_irqrestore(&nvme_fc_lock, flags);
299 
300 		return lport;
301 	}
302 
303 	lport = NULL;
304 
305 out_done:
306 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
307 
308 	return lport;
309 }
310 
311 /**
312  * nvme_fc_register_localport - transport entry point called by an
313  *                              LLDD to register the existence of a NVME
314  *                              host FC port.
315  * @pinfo:     pointer to information about the port to be registered
316  * @template:  LLDD entrypoints and operational parameters for the port
317  * @dev:       physical hardware device node port corresponds to. Will be
318  *             used for DMA mappings
319  * @lport_p:   pointer to a local port pointer. Upon success, the routine
320  *             will allocate a nvme_fc_local_port structure and place its
321  *             address in the local port pointer. Upon failure, local port
322  *             pointer will be set to 0.
323  *
324  * Returns:
325  * a completion status. Must be 0 upon success; a negative errno
326  * (ex: -ENXIO) upon failure.
327  */
328 int
329 nvme_fc_register_localport(struct nvme_fc_port_info *pinfo,
330 			struct nvme_fc_port_template *template,
331 			struct device *dev,
332 			struct nvme_fc_local_port **portptr)
333 {
334 	struct nvme_fc_lport *newrec;
335 	unsigned long flags;
336 	int ret, idx;
337 
338 	if (!template->localport_delete || !template->remoteport_delete ||
339 	    !template->ls_req || !template->fcp_io ||
340 	    !template->ls_abort || !template->fcp_abort ||
341 	    !template->max_hw_queues || !template->max_sgl_segments ||
342 	    !template->max_dif_sgl_segments || !template->dma_boundary) {
343 		ret = -EINVAL;
344 		goto out_reghost_failed;
345 	}
346 
347 	/*
348 	 * look to see if there is already a localport that had been
349 	 * deregistered and in the process of waiting for all the
350 	 * references to fully be removed.  If the references haven't
351 	 * expired, we can simply re-enable the localport. Remoteports
352 	 * and controller reconnections should resume naturally.
353 	 */
354 	newrec = nvme_fc_attach_to_unreg_lport(pinfo, template, dev);
355 
356 	/* found an lport, but something about its state is bad */
357 	if (IS_ERR(newrec)) {
358 		ret = PTR_ERR(newrec);
359 		goto out_reghost_failed;
360 
361 	/* found existing lport, which was resumed */
362 	} else if (newrec) {
363 		*portptr = &newrec->localport;
364 		return 0;
365 	}
366 
367 	/* nothing found - allocate a new localport struct */
368 
369 	newrec = kmalloc((sizeof(*newrec) + template->local_priv_sz),
370 			 GFP_KERNEL);
371 	if (!newrec) {
372 		ret = -ENOMEM;
373 		goto out_reghost_failed;
374 	}
375 
376 	idx = ida_simple_get(&nvme_fc_local_port_cnt, 0, 0, GFP_KERNEL);
377 	if (idx < 0) {
378 		ret = -ENOSPC;
379 		goto out_fail_kfree;
380 	}
381 
382 	if (!get_device(dev) && dev) {
383 		ret = -ENODEV;
384 		goto out_ida_put;
385 	}
386 
387 	INIT_LIST_HEAD(&newrec->port_list);
388 	INIT_LIST_HEAD(&newrec->endp_list);
389 	kref_init(&newrec->ref);
390 	atomic_set(&newrec->act_rport_cnt, 0);
391 	newrec->ops = template;
392 	newrec->dev = dev;
393 	ida_init(&newrec->endp_cnt);
394 	newrec->localport.private = &newrec[1];
395 	newrec->localport.node_name = pinfo->node_name;
396 	newrec->localport.port_name = pinfo->port_name;
397 	newrec->localport.port_role = pinfo->port_role;
398 	newrec->localport.port_id = pinfo->port_id;
399 	newrec->localport.port_state = FC_OBJSTATE_ONLINE;
400 	newrec->localport.port_num = idx;
401 
402 	spin_lock_irqsave(&nvme_fc_lock, flags);
403 	list_add_tail(&newrec->port_list, &nvme_fc_lport_list);
404 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
405 
406 	if (dev)
407 		dma_set_seg_boundary(dev, template->dma_boundary);
408 
409 	*portptr = &newrec->localport;
410 	return 0;
411 
412 out_ida_put:
413 	ida_simple_remove(&nvme_fc_local_port_cnt, idx);
414 out_fail_kfree:
415 	kfree(newrec);
416 out_reghost_failed:
417 	*portptr = NULL;
418 
419 	return ret;
420 }
421 EXPORT_SYMBOL_GPL(nvme_fc_register_localport);
422 
423 /**
424  * nvme_fc_unregister_localport - transport entry point called by an
425  *                              LLDD to deregister/remove a previously
426  *                              registered a NVME host FC port.
427  * @localport: pointer to the (registered) local port that is to be
428  *             deregistered.
429  *
430  * Returns:
431  * a completion status. Must be 0 upon success; a negative errno
432  * (ex: -ENXIO) upon failure.
433  */
434 int
435 nvme_fc_unregister_localport(struct nvme_fc_local_port *portptr)
436 {
437 	struct nvme_fc_lport *lport = localport_to_lport(portptr);
438 	unsigned long flags;
439 
440 	if (!portptr)
441 		return -EINVAL;
442 
443 	spin_lock_irqsave(&nvme_fc_lock, flags);
444 
445 	if (portptr->port_state != FC_OBJSTATE_ONLINE) {
446 		spin_unlock_irqrestore(&nvme_fc_lock, flags);
447 		return -EINVAL;
448 	}
449 	portptr->port_state = FC_OBJSTATE_DELETED;
450 
451 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
452 
453 	if (atomic_read(&lport->act_rport_cnt) == 0)
454 		lport->ops->localport_delete(&lport->localport);
455 
456 	nvme_fc_lport_put(lport);
457 
458 	return 0;
459 }
460 EXPORT_SYMBOL_GPL(nvme_fc_unregister_localport);
461 
462 /*
463  * TRADDR strings, per FC-NVME are fixed format:
464  *   "nn-0x<16hexdigits>:pn-0x<16hexdigits>" - 43 characters
465  * udev event will only differ by prefix of what field is
466  * being specified:
467  *    "NVMEFC_HOST_TRADDR=" or "NVMEFC_TRADDR=" - 19 max characters
468  *  19 + 43 + null_fudge = 64 characters
469  */
470 #define FCNVME_TRADDR_LENGTH		64
471 
472 static void
473 nvme_fc_signal_discovery_scan(struct nvme_fc_lport *lport,
474 		struct nvme_fc_rport *rport)
475 {
476 	char hostaddr[FCNVME_TRADDR_LENGTH];	/* NVMEFC_HOST_TRADDR=...*/
477 	char tgtaddr[FCNVME_TRADDR_LENGTH];	/* NVMEFC_TRADDR=...*/
478 	char *envp[4] = { "FC_EVENT=nvmediscovery", hostaddr, tgtaddr, NULL };
479 
480 	if (!(rport->remoteport.port_role & FC_PORT_ROLE_NVME_DISCOVERY))
481 		return;
482 
483 	snprintf(hostaddr, sizeof(hostaddr),
484 		"NVMEFC_HOST_TRADDR=nn-0x%016llx:pn-0x%016llx",
485 		lport->localport.node_name, lport->localport.port_name);
486 	snprintf(tgtaddr, sizeof(tgtaddr),
487 		"NVMEFC_TRADDR=nn-0x%016llx:pn-0x%016llx",
488 		rport->remoteport.node_name, rport->remoteport.port_name);
489 	kobject_uevent_env(&fc_udev_device->kobj, KOBJ_CHANGE, envp);
490 }
491 
492 static void
493 nvme_fc_free_rport(struct kref *ref)
494 {
495 	struct nvme_fc_rport *rport =
496 		container_of(ref, struct nvme_fc_rport, ref);
497 	struct nvme_fc_lport *lport =
498 			localport_to_lport(rport->remoteport.localport);
499 	unsigned long flags;
500 
501 	WARN_ON(rport->remoteport.port_state != FC_OBJSTATE_DELETED);
502 	WARN_ON(!list_empty(&rport->ctrl_list));
503 
504 	/* remove from lport list */
505 	spin_lock_irqsave(&nvme_fc_lock, flags);
506 	list_del(&rport->endp_list);
507 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
508 
509 	ida_simple_remove(&lport->endp_cnt, rport->remoteport.port_num);
510 
511 	kfree(rport);
512 
513 	nvme_fc_lport_put(lport);
514 }
515 
516 static void
517 nvme_fc_rport_put(struct nvme_fc_rport *rport)
518 {
519 	kref_put(&rport->ref, nvme_fc_free_rport);
520 }
521 
522 static int
523 nvme_fc_rport_get(struct nvme_fc_rport *rport)
524 {
525 	return kref_get_unless_zero(&rport->ref);
526 }
527 
528 static void
529 nvme_fc_resume_controller(struct nvme_fc_ctrl *ctrl)
530 {
531 	switch (ctrl->ctrl.state) {
532 	case NVME_CTRL_NEW:
533 	case NVME_CTRL_CONNECTING:
534 		/*
535 		 * As all reconnects were suppressed, schedule a
536 		 * connect.
537 		 */
538 		dev_info(ctrl->ctrl.device,
539 			"NVME-FC{%d}: connectivity re-established. "
540 			"Attempting reconnect\n", ctrl->cnum);
541 
542 		queue_delayed_work(nvme_wq, &ctrl->connect_work, 0);
543 		break;
544 
545 	case NVME_CTRL_RESETTING:
546 		/*
547 		 * Controller is already in the process of terminating the
548 		 * association. No need to do anything further. The reconnect
549 		 * step will naturally occur after the reset completes.
550 		 */
551 		break;
552 
553 	default:
554 		/* no action to take - let it delete */
555 		break;
556 	}
557 }
558 
559 static struct nvme_fc_rport *
560 nvme_fc_attach_to_suspended_rport(struct nvme_fc_lport *lport,
561 				struct nvme_fc_port_info *pinfo)
562 {
563 	struct nvme_fc_rport *rport;
564 	struct nvme_fc_ctrl *ctrl;
565 	unsigned long flags;
566 
567 	spin_lock_irqsave(&nvme_fc_lock, flags);
568 
569 	list_for_each_entry(rport, &lport->endp_list, endp_list) {
570 		if (rport->remoteport.node_name != pinfo->node_name ||
571 		    rport->remoteport.port_name != pinfo->port_name)
572 			continue;
573 
574 		if (!nvme_fc_rport_get(rport)) {
575 			rport = ERR_PTR(-ENOLCK);
576 			goto out_done;
577 		}
578 
579 		spin_unlock_irqrestore(&nvme_fc_lock, flags);
580 
581 		spin_lock_irqsave(&rport->lock, flags);
582 
583 		/* has it been unregistered */
584 		if (rport->remoteport.port_state != FC_OBJSTATE_DELETED) {
585 			/* means lldd called us twice */
586 			spin_unlock_irqrestore(&rport->lock, flags);
587 			nvme_fc_rport_put(rport);
588 			return ERR_PTR(-ESTALE);
589 		}
590 
591 		rport->remoteport.port_state = FC_OBJSTATE_ONLINE;
592 		rport->dev_loss_end = 0;
593 
594 		/*
595 		 * kick off a reconnect attempt on all associations to the
596 		 * remote port. A successful reconnects will resume i/o.
597 		 */
598 		list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list)
599 			nvme_fc_resume_controller(ctrl);
600 
601 		spin_unlock_irqrestore(&rport->lock, flags);
602 
603 		return rport;
604 	}
605 
606 	rport = NULL;
607 
608 out_done:
609 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
610 
611 	return rport;
612 }
613 
614 static inline void
615 __nvme_fc_set_dev_loss_tmo(struct nvme_fc_rport *rport,
616 			struct nvme_fc_port_info *pinfo)
617 {
618 	if (pinfo->dev_loss_tmo)
619 		rport->remoteport.dev_loss_tmo = pinfo->dev_loss_tmo;
620 	else
621 		rport->remoteport.dev_loss_tmo = NVME_FC_DEFAULT_DEV_LOSS_TMO;
622 }
623 
624 /**
625  * nvme_fc_register_remoteport - transport entry point called by an
626  *                              LLDD to register the existence of a NVME
627  *                              subsystem FC port on its fabric.
628  * @localport: pointer to the (registered) local port that the remote
629  *             subsystem port is connected to.
630  * @pinfo:     pointer to information about the port to be registered
631  * @rport_p:   pointer to a remote port pointer. Upon success, the routine
632  *             will allocate a nvme_fc_remote_port structure and place its
633  *             address in the remote port pointer. Upon failure, remote port
634  *             pointer will be set to 0.
635  *
636  * Returns:
637  * a completion status. Must be 0 upon success; a negative errno
638  * (ex: -ENXIO) upon failure.
639  */
640 int
641 nvme_fc_register_remoteport(struct nvme_fc_local_port *localport,
642 				struct nvme_fc_port_info *pinfo,
643 				struct nvme_fc_remote_port **portptr)
644 {
645 	struct nvme_fc_lport *lport = localport_to_lport(localport);
646 	struct nvme_fc_rport *newrec;
647 	unsigned long flags;
648 	int ret, idx;
649 
650 	if (!nvme_fc_lport_get(lport)) {
651 		ret = -ESHUTDOWN;
652 		goto out_reghost_failed;
653 	}
654 
655 	/*
656 	 * look to see if there is already a remoteport that is waiting
657 	 * for a reconnect (within dev_loss_tmo) with the same WWN's.
658 	 * If so, transition to it and reconnect.
659 	 */
660 	newrec = nvme_fc_attach_to_suspended_rport(lport, pinfo);
661 
662 	/* found an rport, but something about its state is bad */
663 	if (IS_ERR(newrec)) {
664 		ret = PTR_ERR(newrec);
665 		goto out_lport_put;
666 
667 	/* found existing rport, which was resumed */
668 	} else if (newrec) {
669 		nvme_fc_lport_put(lport);
670 		__nvme_fc_set_dev_loss_tmo(newrec, pinfo);
671 		nvme_fc_signal_discovery_scan(lport, newrec);
672 		*portptr = &newrec->remoteport;
673 		return 0;
674 	}
675 
676 	/* nothing found - allocate a new remoteport struct */
677 
678 	newrec = kmalloc((sizeof(*newrec) + lport->ops->remote_priv_sz),
679 			 GFP_KERNEL);
680 	if (!newrec) {
681 		ret = -ENOMEM;
682 		goto out_lport_put;
683 	}
684 
685 	idx = ida_simple_get(&lport->endp_cnt, 0, 0, GFP_KERNEL);
686 	if (idx < 0) {
687 		ret = -ENOSPC;
688 		goto out_kfree_rport;
689 	}
690 
691 	INIT_LIST_HEAD(&newrec->endp_list);
692 	INIT_LIST_HEAD(&newrec->ctrl_list);
693 	INIT_LIST_HEAD(&newrec->ls_req_list);
694 	kref_init(&newrec->ref);
695 	atomic_set(&newrec->act_ctrl_cnt, 0);
696 	spin_lock_init(&newrec->lock);
697 	newrec->remoteport.localport = &lport->localport;
698 	newrec->dev = lport->dev;
699 	newrec->lport = lport;
700 	newrec->remoteport.private = &newrec[1];
701 	newrec->remoteport.port_role = pinfo->port_role;
702 	newrec->remoteport.node_name = pinfo->node_name;
703 	newrec->remoteport.port_name = pinfo->port_name;
704 	newrec->remoteport.port_id = pinfo->port_id;
705 	newrec->remoteport.port_state = FC_OBJSTATE_ONLINE;
706 	newrec->remoteport.port_num = idx;
707 	__nvme_fc_set_dev_loss_tmo(newrec, pinfo);
708 
709 	spin_lock_irqsave(&nvme_fc_lock, flags);
710 	list_add_tail(&newrec->endp_list, &lport->endp_list);
711 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
712 
713 	nvme_fc_signal_discovery_scan(lport, newrec);
714 
715 	*portptr = &newrec->remoteport;
716 	return 0;
717 
718 out_kfree_rport:
719 	kfree(newrec);
720 out_lport_put:
721 	nvme_fc_lport_put(lport);
722 out_reghost_failed:
723 	*portptr = NULL;
724 	return ret;
725 }
726 EXPORT_SYMBOL_GPL(nvme_fc_register_remoteport);
727 
728 static int
729 nvme_fc_abort_lsops(struct nvme_fc_rport *rport)
730 {
731 	struct nvmefc_ls_req_op *lsop;
732 	unsigned long flags;
733 
734 restart:
735 	spin_lock_irqsave(&rport->lock, flags);
736 
737 	list_for_each_entry(lsop, &rport->ls_req_list, lsreq_list) {
738 		if (!(lsop->flags & FCOP_FLAGS_TERMIO)) {
739 			lsop->flags |= FCOP_FLAGS_TERMIO;
740 			spin_unlock_irqrestore(&rport->lock, flags);
741 			rport->lport->ops->ls_abort(&rport->lport->localport,
742 						&rport->remoteport,
743 						&lsop->ls_req);
744 			goto restart;
745 		}
746 	}
747 	spin_unlock_irqrestore(&rport->lock, flags);
748 
749 	return 0;
750 }
751 
752 static void
753 nvme_fc_ctrl_connectivity_loss(struct nvme_fc_ctrl *ctrl)
754 {
755 	dev_info(ctrl->ctrl.device,
756 		"NVME-FC{%d}: controller connectivity lost. Awaiting "
757 		"Reconnect", ctrl->cnum);
758 
759 	switch (ctrl->ctrl.state) {
760 	case NVME_CTRL_NEW:
761 	case NVME_CTRL_LIVE:
762 		/*
763 		 * Schedule a controller reset. The reset will terminate the
764 		 * association and schedule the reconnect timer.  Reconnects
765 		 * will be attempted until either the ctlr_loss_tmo
766 		 * (max_retries * connect_delay) expires or the remoteport's
767 		 * dev_loss_tmo expires.
768 		 */
769 		if (nvme_reset_ctrl(&ctrl->ctrl)) {
770 			dev_warn(ctrl->ctrl.device,
771 				"NVME-FC{%d}: Couldn't schedule reset. "
772 				"Deleting controller.\n",
773 				ctrl->cnum);
774 			nvme_delete_ctrl(&ctrl->ctrl);
775 		}
776 		break;
777 
778 	case NVME_CTRL_CONNECTING:
779 		/*
780 		 * The association has already been terminated and the
781 		 * controller is attempting reconnects.  No need to do anything
782 		 * futher.  Reconnects will be attempted until either the
783 		 * ctlr_loss_tmo (max_retries * connect_delay) expires or the
784 		 * remoteport's dev_loss_tmo expires.
785 		 */
786 		break;
787 
788 	case NVME_CTRL_RESETTING:
789 		/*
790 		 * Controller is already in the process of terminating the
791 		 * association.  No need to do anything further. The reconnect
792 		 * step will kick in naturally after the association is
793 		 * terminated.
794 		 */
795 		break;
796 
797 	case NVME_CTRL_DELETING:
798 	default:
799 		/* no action to take - let it delete */
800 		break;
801 	}
802 }
803 
804 /**
805  * nvme_fc_unregister_remoteport - transport entry point called by an
806  *                              LLDD to deregister/remove a previously
807  *                              registered a NVME subsystem FC port.
808  * @remoteport: pointer to the (registered) remote port that is to be
809  *              deregistered.
810  *
811  * Returns:
812  * a completion status. Must be 0 upon success; a negative errno
813  * (ex: -ENXIO) upon failure.
814  */
815 int
816 nvme_fc_unregister_remoteport(struct nvme_fc_remote_port *portptr)
817 {
818 	struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
819 	struct nvme_fc_ctrl *ctrl;
820 	unsigned long flags;
821 
822 	if (!portptr)
823 		return -EINVAL;
824 
825 	spin_lock_irqsave(&rport->lock, flags);
826 
827 	if (portptr->port_state != FC_OBJSTATE_ONLINE) {
828 		spin_unlock_irqrestore(&rport->lock, flags);
829 		return -EINVAL;
830 	}
831 	portptr->port_state = FC_OBJSTATE_DELETED;
832 
833 	rport->dev_loss_end = jiffies + (portptr->dev_loss_tmo * HZ);
834 
835 	list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list) {
836 		/* if dev_loss_tmo==0, dev loss is immediate */
837 		if (!portptr->dev_loss_tmo) {
838 			dev_warn(ctrl->ctrl.device,
839 				"NVME-FC{%d}: controller connectivity lost. "
840 				"Deleting controller.\n",
841 				ctrl->cnum);
842 			nvme_delete_ctrl(&ctrl->ctrl);
843 		} else
844 			nvme_fc_ctrl_connectivity_loss(ctrl);
845 	}
846 
847 	spin_unlock_irqrestore(&rport->lock, flags);
848 
849 	nvme_fc_abort_lsops(rport);
850 
851 	if (atomic_read(&rport->act_ctrl_cnt) == 0)
852 		rport->lport->ops->remoteport_delete(portptr);
853 
854 	/*
855 	 * release the reference, which will allow, if all controllers
856 	 * go away, which should only occur after dev_loss_tmo occurs,
857 	 * for the rport to be torn down.
858 	 */
859 	nvme_fc_rport_put(rport);
860 
861 	return 0;
862 }
863 EXPORT_SYMBOL_GPL(nvme_fc_unregister_remoteport);
864 
865 /**
866  * nvme_fc_rescan_remoteport - transport entry point called by an
867  *                              LLDD to request a nvme device rescan.
868  * @remoteport: pointer to the (registered) remote port that is to be
869  *              rescanned.
870  *
871  * Returns: N/A
872  */
873 void
874 nvme_fc_rescan_remoteport(struct nvme_fc_remote_port *remoteport)
875 {
876 	struct nvme_fc_rport *rport = remoteport_to_rport(remoteport);
877 
878 	nvme_fc_signal_discovery_scan(rport->lport, rport);
879 }
880 EXPORT_SYMBOL_GPL(nvme_fc_rescan_remoteport);
881 
882 int
883 nvme_fc_set_remoteport_devloss(struct nvme_fc_remote_port *portptr,
884 			u32 dev_loss_tmo)
885 {
886 	struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
887 	unsigned long flags;
888 
889 	spin_lock_irqsave(&rport->lock, flags);
890 
891 	if (portptr->port_state != FC_OBJSTATE_ONLINE) {
892 		spin_unlock_irqrestore(&rport->lock, flags);
893 		return -EINVAL;
894 	}
895 
896 	/* a dev_loss_tmo of 0 (immediate) is allowed to be set */
897 	rport->remoteport.dev_loss_tmo = dev_loss_tmo;
898 
899 	spin_unlock_irqrestore(&rport->lock, flags);
900 
901 	return 0;
902 }
903 EXPORT_SYMBOL_GPL(nvme_fc_set_remoteport_devloss);
904 
905 
906 /* *********************** FC-NVME DMA Handling **************************** */
907 
908 /*
909  * The fcloop device passes in a NULL device pointer. Real LLD's will
910  * pass in a valid device pointer. If NULL is passed to the dma mapping
911  * routines, depending on the platform, it may or may not succeed, and
912  * may crash.
913  *
914  * As such:
915  * Wrapper all the dma routines and check the dev pointer.
916  *
917  * If simple mappings (return just a dma address, we'll noop them,
918  * returning a dma address of 0.
919  *
920  * On more complex mappings (dma_map_sg), a pseudo routine fills
921  * in the scatter list, setting all dma addresses to 0.
922  */
923 
924 static inline dma_addr_t
925 fc_dma_map_single(struct device *dev, void *ptr, size_t size,
926 		enum dma_data_direction dir)
927 {
928 	return dev ? dma_map_single(dev, ptr, size, dir) : (dma_addr_t)0L;
929 }
930 
931 static inline int
932 fc_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
933 {
934 	return dev ? dma_mapping_error(dev, dma_addr) : 0;
935 }
936 
937 static inline void
938 fc_dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
939 	enum dma_data_direction dir)
940 {
941 	if (dev)
942 		dma_unmap_single(dev, addr, size, dir);
943 }
944 
945 static inline void
946 fc_dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr, size_t size,
947 		enum dma_data_direction dir)
948 {
949 	if (dev)
950 		dma_sync_single_for_cpu(dev, addr, size, dir);
951 }
952 
953 static inline void
954 fc_dma_sync_single_for_device(struct device *dev, dma_addr_t addr, size_t size,
955 		enum dma_data_direction dir)
956 {
957 	if (dev)
958 		dma_sync_single_for_device(dev, addr, size, dir);
959 }
960 
961 /* pseudo dma_map_sg call */
962 static int
963 fc_map_sg(struct scatterlist *sg, int nents)
964 {
965 	struct scatterlist *s;
966 	int i;
967 
968 	WARN_ON(nents == 0 || sg[0].length == 0);
969 
970 	for_each_sg(sg, s, nents, i) {
971 		s->dma_address = 0L;
972 #ifdef CONFIG_NEED_SG_DMA_LENGTH
973 		s->dma_length = s->length;
974 #endif
975 	}
976 	return nents;
977 }
978 
979 static inline int
980 fc_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
981 		enum dma_data_direction dir)
982 {
983 	return dev ? dma_map_sg(dev, sg, nents, dir) : fc_map_sg(sg, nents);
984 }
985 
986 static inline void
987 fc_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
988 		enum dma_data_direction dir)
989 {
990 	if (dev)
991 		dma_unmap_sg(dev, sg, nents, dir);
992 }
993 
994 /* *********************** FC-NVME LS Handling **************************** */
995 
996 static void nvme_fc_ctrl_put(struct nvme_fc_ctrl *);
997 static int nvme_fc_ctrl_get(struct nvme_fc_ctrl *);
998 
999 
1000 static void
1001 __nvme_fc_finish_ls_req(struct nvmefc_ls_req_op *lsop)
1002 {
1003 	struct nvme_fc_rport *rport = lsop->rport;
1004 	struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1005 	unsigned long flags;
1006 
1007 	spin_lock_irqsave(&rport->lock, flags);
1008 
1009 	if (!lsop->req_queued) {
1010 		spin_unlock_irqrestore(&rport->lock, flags);
1011 		return;
1012 	}
1013 
1014 	list_del(&lsop->lsreq_list);
1015 
1016 	lsop->req_queued = false;
1017 
1018 	spin_unlock_irqrestore(&rport->lock, flags);
1019 
1020 	fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
1021 				  (lsreq->rqstlen + lsreq->rsplen),
1022 				  DMA_BIDIRECTIONAL);
1023 
1024 	nvme_fc_rport_put(rport);
1025 }
1026 
1027 static int
1028 __nvme_fc_send_ls_req(struct nvme_fc_rport *rport,
1029 		struct nvmefc_ls_req_op *lsop,
1030 		void (*done)(struct nvmefc_ls_req *req, int status))
1031 {
1032 	struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1033 	unsigned long flags;
1034 	int ret = 0;
1035 
1036 	if (rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
1037 		return -ECONNREFUSED;
1038 
1039 	if (!nvme_fc_rport_get(rport))
1040 		return -ESHUTDOWN;
1041 
1042 	lsreq->done = done;
1043 	lsop->rport = rport;
1044 	lsop->req_queued = false;
1045 	INIT_LIST_HEAD(&lsop->lsreq_list);
1046 	init_completion(&lsop->ls_done);
1047 
1048 	lsreq->rqstdma = fc_dma_map_single(rport->dev, lsreq->rqstaddr,
1049 				  lsreq->rqstlen + lsreq->rsplen,
1050 				  DMA_BIDIRECTIONAL);
1051 	if (fc_dma_mapping_error(rport->dev, lsreq->rqstdma)) {
1052 		ret = -EFAULT;
1053 		goto out_putrport;
1054 	}
1055 	lsreq->rspdma = lsreq->rqstdma + lsreq->rqstlen;
1056 
1057 	spin_lock_irqsave(&rport->lock, flags);
1058 
1059 	list_add_tail(&lsop->lsreq_list, &rport->ls_req_list);
1060 
1061 	lsop->req_queued = true;
1062 
1063 	spin_unlock_irqrestore(&rport->lock, flags);
1064 
1065 	ret = rport->lport->ops->ls_req(&rport->lport->localport,
1066 					&rport->remoteport, lsreq);
1067 	if (ret)
1068 		goto out_unlink;
1069 
1070 	return 0;
1071 
1072 out_unlink:
1073 	lsop->ls_error = ret;
1074 	spin_lock_irqsave(&rport->lock, flags);
1075 	lsop->req_queued = false;
1076 	list_del(&lsop->lsreq_list);
1077 	spin_unlock_irqrestore(&rport->lock, flags);
1078 	fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
1079 				  (lsreq->rqstlen + lsreq->rsplen),
1080 				  DMA_BIDIRECTIONAL);
1081 out_putrport:
1082 	nvme_fc_rport_put(rport);
1083 
1084 	return ret;
1085 }
1086 
1087 static void
1088 nvme_fc_send_ls_req_done(struct nvmefc_ls_req *lsreq, int status)
1089 {
1090 	struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
1091 
1092 	lsop->ls_error = status;
1093 	complete(&lsop->ls_done);
1094 }
1095 
1096 static int
1097 nvme_fc_send_ls_req(struct nvme_fc_rport *rport, struct nvmefc_ls_req_op *lsop)
1098 {
1099 	struct nvmefc_ls_req *lsreq = &lsop->ls_req;
1100 	struct fcnvme_ls_rjt *rjt = lsreq->rspaddr;
1101 	int ret;
1102 
1103 	ret = __nvme_fc_send_ls_req(rport, lsop, nvme_fc_send_ls_req_done);
1104 
1105 	if (!ret) {
1106 		/*
1107 		 * No timeout/not interruptible as we need the struct
1108 		 * to exist until the lldd calls us back. Thus mandate
1109 		 * wait until driver calls back. lldd responsible for
1110 		 * the timeout action
1111 		 */
1112 		wait_for_completion(&lsop->ls_done);
1113 
1114 		__nvme_fc_finish_ls_req(lsop);
1115 
1116 		ret = lsop->ls_error;
1117 	}
1118 
1119 	if (ret)
1120 		return ret;
1121 
1122 	/* ACC or RJT payload ? */
1123 	if (rjt->w0.ls_cmd == FCNVME_LS_RJT)
1124 		return -ENXIO;
1125 
1126 	return 0;
1127 }
1128 
1129 static int
1130 nvme_fc_send_ls_req_async(struct nvme_fc_rport *rport,
1131 		struct nvmefc_ls_req_op *lsop,
1132 		void (*done)(struct nvmefc_ls_req *req, int status))
1133 {
1134 	/* don't wait for completion */
1135 
1136 	return __nvme_fc_send_ls_req(rport, lsop, done);
1137 }
1138 
1139 /* Validation Error indexes into the string table below */
1140 enum {
1141 	VERR_NO_ERROR		= 0,
1142 	VERR_LSACC		= 1,
1143 	VERR_LSDESC_RQST	= 2,
1144 	VERR_LSDESC_RQST_LEN	= 3,
1145 	VERR_ASSOC_ID		= 4,
1146 	VERR_ASSOC_ID_LEN	= 5,
1147 	VERR_CONN_ID		= 6,
1148 	VERR_CONN_ID_LEN	= 7,
1149 	VERR_CR_ASSOC		= 8,
1150 	VERR_CR_ASSOC_ACC_LEN	= 9,
1151 	VERR_CR_CONN		= 10,
1152 	VERR_CR_CONN_ACC_LEN	= 11,
1153 	VERR_DISCONN		= 12,
1154 	VERR_DISCONN_ACC_LEN	= 13,
1155 };
1156 
1157 static char *validation_errors[] = {
1158 	"OK",
1159 	"Not LS_ACC",
1160 	"Not LSDESC_RQST",
1161 	"Bad LSDESC_RQST Length",
1162 	"Not Association ID",
1163 	"Bad Association ID Length",
1164 	"Not Connection ID",
1165 	"Bad Connection ID Length",
1166 	"Not CR_ASSOC Rqst",
1167 	"Bad CR_ASSOC ACC Length",
1168 	"Not CR_CONN Rqst",
1169 	"Bad CR_CONN ACC Length",
1170 	"Not Disconnect Rqst",
1171 	"Bad Disconnect ACC Length",
1172 };
1173 
1174 static int
1175 nvme_fc_connect_admin_queue(struct nvme_fc_ctrl *ctrl,
1176 	struct nvme_fc_queue *queue, u16 qsize, u16 ersp_ratio)
1177 {
1178 	struct nvmefc_ls_req_op *lsop;
1179 	struct nvmefc_ls_req *lsreq;
1180 	struct fcnvme_ls_cr_assoc_rqst *assoc_rqst;
1181 	struct fcnvme_ls_cr_assoc_acc *assoc_acc;
1182 	int ret, fcret = 0;
1183 
1184 	lsop = kzalloc((sizeof(*lsop) +
1185 			 ctrl->lport->ops->lsrqst_priv_sz +
1186 			 sizeof(*assoc_rqst) + sizeof(*assoc_acc)), GFP_KERNEL);
1187 	if (!lsop) {
1188 		ret = -ENOMEM;
1189 		goto out_no_memory;
1190 	}
1191 	lsreq = &lsop->ls_req;
1192 
1193 	lsreq->private = (void *)&lsop[1];
1194 	assoc_rqst = (struct fcnvme_ls_cr_assoc_rqst *)
1195 			(lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1196 	assoc_acc = (struct fcnvme_ls_cr_assoc_acc *)&assoc_rqst[1];
1197 
1198 	assoc_rqst->w0.ls_cmd = FCNVME_LS_CREATE_ASSOCIATION;
1199 	assoc_rqst->desc_list_len =
1200 			cpu_to_be32(sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
1201 
1202 	assoc_rqst->assoc_cmd.desc_tag =
1203 			cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD);
1204 	assoc_rqst->assoc_cmd.desc_len =
1205 			fcnvme_lsdesc_len(
1206 				sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
1207 
1208 	assoc_rqst->assoc_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
1209 	assoc_rqst->assoc_cmd.sqsize = cpu_to_be16(qsize);
1210 	/* Linux supports only Dynamic controllers */
1211 	assoc_rqst->assoc_cmd.cntlid = cpu_to_be16(0xffff);
1212 	uuid_copy(&assoc_rqst->assoc_cmd.hostid, &ctrl->ctrl.opts->host->id);
1213 	strncpy(assoc_rqst->assoc_cmd.hostnqn, ctrl->ctrl.opts->host->nqn,
1214 		min(FCNVME_ASSOC_HOSTNQN_LEN, NVMF_NQN_SIZE));
1215 	strncpy(assoc_rqst->assoc_cmd.subnqn, ctrl->ctrl.opts->subsysnqn,
1216 		min(FCNVME_ASSOC_SUBNQN_LEN, NVMF_NQN_SIZE));
1217 
1218 	lsop->queue = queue;
1219 	lsreq->rqstaddr = assoc_rqst;
1220 	lsreq->rqstlen = sizeof(*assoc_rqst);
1221 	lsreq->rspaddr = assoc_acc;
1222 	lsreq->rsplen = sizeof(*assoc_acc);
1223 	lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1224 
1225 	ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
1226 	if (ret)
1227 		goto out_free_buffer;
1228 
1229 	/* process connect LS completion */
1230 
1231 	/* validate the ACC response */
1232 	if (assoc_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1233 		fcret = VERR_LSACC;
1234 	else if (assoc_acc->hdr.desc_list_len !=
1235 			fcnvme_lsdesc_len(
1236 				sizeof(struct fcnvme_ls_cr_assoc_acc)))
1237 		fcret = VERR_CR_ASSOC_ACC_LEN;
1238 	else if (assoc_acc->hdr.rqst.desc_tag !=
1239 			cpu_to_be32(FCNVME_LSDESC_RQST))
1240 		fcret = VERR_LSDESC_RQST;
1241 	else if (assoc_acc->hdr.rqst.desc_len !=
1242 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1243 		fcret = VERR_LSDESC_RQST_LEN;
1244 	else if (assoc_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_ASSOCIATION)
1245 		fcret = VERR_CR_ASSOC;
1246 	else if (assoc_acc->associd.desc_tag !=
1247 			cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
1248 		fcret = VERR_ASSOC_ID;
1249 	else if (assoc_acc->associd.desc_len !=
1250 			fcnvme_lsdesc_len(
1251 				sizeof(struct fcnvme_lsdesc_assoc_id)))
1252 		fcret = VERR_ASSOC_ID_LEN;
1253 	else if (assoc_acc->connectid.desc_tag !=
1254 			cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1255 		fcret = VERR_CONN_ID;
1256 	else if (assoc_acc->connectid.desc_len !=
1257 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1258 		fcret = VERR_CONN_ID_LEN;
1259 
1260 	if (fcret) {
1261 		ret = -EBADF;
1262 		dev_err(ctrl->dev,
1263 			"q %d connect failed: %s\n",
1264 			queue->qnum, validation_errors[fcret]);
1265 	} else {
1266 		ctrl->association_id =
1267 			be64_to_cpu(assoc_acc->associd.association_id);
1268 		queue->connection_id =
1269 			be64_to_cpu(assoc_acc->connectid.connection_id);
1270 		set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1271 	}
1272 
1273 out_free_buffer:
1274 	kfree(lsop);
1275 out_no_memory:
1276 	if (ret)
1277 		dev_err(ctrl->dev,
1278 			"queue %d connect admin queue failed (%d).\n",
1279 			queue->qnum, ret);
1280 	return ret;
1281 }
1282 
1283 static int
1284 nvme_fc_connect_queue(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
1285 			u16 qsize, u16 ersp_ratio)
1286 {
1287 	struct nvmefc_ls_req_op *lsop;
1288 	struct nvmefc_ls_req *lsreq;
1289 	struct fcnvme_ls_cr_conn_rqst *conn_rqst;
1290 	struct fcnvme_ls_cr_conn_acc *conn_acc;
1291 	int ret, fcret = 0;
1292 
1293 	lsop = kzalloc((sizeof(*lsop) +
1294 			 ctrl->lport->ops->lsrqst_priv_sz +
1295 			 sizeof(*conn_rqst) + sizeof(*conn_acc)), GFP_KERNEL);
1296 	if (!lsop) {
1297 		ret = -ENOMEM;
1298 		goto out_no_memory;
1299 	}
1300 	lsreq = &lsop->ls_req;
1301 
1302 	lsreq->private = (void *)&lsop[1];
1303 	conn_rqst = (struct fcnvme_ls_cr_conn_rqst *)
1304 			(lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1305 	conn_acc = (struct fcnvme_ls_cr_conn_acc *)&conn_rqst[1];
1306 
1307 	conn_rqst->w0.ls_cmd = FCNVME_LS_CREATE_CONNECTION;
1308 	conn_rqst->desc_list_len = cpu_to_be32(
1309 				sizeof(struct fcnvme_lsdesc_assoc_id) +
1310 				sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1311 
1312 	conn_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1313 	conn_rqst->associd.desc_len =
1314 			fcnvme_lsdesc_len(
1315 				sizeof(struct fcnvme_lsdesc_assoc_id));
1316 	conn_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1317 	conn_rqst->connect_cmd.desc_tag =
1318 			cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD);
1319 	conn_rqst->connect_cmd.desc_len =
1320 			fcnvme_lsdesc_len(
1321 				sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1322 	conn_rqst->connect_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
1323 	conn_rqst->connect_cmd.qid  = cpu_to_be16(queue->qnum);
1324 	conn_rqst->connect_cmd.sqsize = cpu_to_be16(qsize);
1325 
1326 	lsop->queue = queue;
1327 	lsreq->rqstaddr = conn_rqst;
1328 	lsreq->rqstlen = sizeof(*conn_rqst);
1329 	lsreq->rspaddr = conn_acc;
1330 	lsreq->rsplen = sizeof(*conn_acc);
1331 	lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1332 
1333 	ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
1334 	if (ret)
1335 		goto out_free_buffer;
1336 
1337 	/* process connect LS completion */
1338 
1339 	/* validate the ACC response */
1340 	if (conn_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1341 		fcret = VERR_LSACC;
1342 	else if (conn_acc->hdr.desc_list_len !=
1343 			fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)))
1344 		fcret = VERR_CR_CONN_ACC_LEN;
1345 	else if (conn_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
1346 		fcret = VERR_LSDESC_RQST;
1347 	else if (conn_acc->hdr.rqst.desc_len !=
1348 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1349 		fcret = VERR_LSDESC_RQST_LEN;
1350 	else if (conn_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_CONNECTION)
1351 		fcret = VERR_CR_CONN;
1352 	else if (conn_acc->connectid.desc_tag !=
1353 			cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1354 		fcret = VERR_CONN_ID;
1355 	else if (conn_acc->connectid.desc_len !=
1356 			fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1357 		fcret = VERR_CONN_ID_LEN;
1358 
1359 	if (fcret) {
1360 		ret = -EBADF;
1361 		dev_err(ctrl->dev,
1362 			"q %d connect failed: %s\n",
1363 			queue->qnum, validation_errors[fcret]);
1364 	} else {
1365 		queue->connection_id =
1366 			be64_to_cpu(conn_acc->connectid.connection_id);
1367 		set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1368 	}
1369 
1370 out_free_buffer:
1371 	kfree(lsop);
1372 out_no_memory:
1373 	if (ret)
1374 		dev_err(ctrl->dev,
1375 			"queue %d connect command failed (%d).\n",
1376 			queue->qnum, ret);
1377 	return ret;
1378 }
1379 
1380 static void
1381 nvme_fc_disconnect_assoc_done(struct nvmefc_ls_req *lsreq, int status)
1382 {
1383 	struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
1384 
1385 	__nvme_fc_finish_ls_req(lsop);
1386 
1387 	/* fc-nvme iniator doesn't care about success or failure of cmd */
1388 
1389 	kfree(lsop);
1390 }
1391 
1392 /*
1393  * This routine sends a FC-NVME LS to disconnect (aka terminate)
1394  * the FC-NVME Association.  Terminating the association also
1395  * terminates the FC-NVME connections (per queue, both admin and io
1396  * queues) that are part of the association. E.g. things are torn
1397  * down, and the related FC-NVME Association ID and Connection IDs
1398  * become invalid.
1399  *
1400  * The behavior of the fc-nvme initiator is such that it's
1401  * understanding of the association and connections will implicitly
1402  * be torn down. The action is implicit as it may be due to a loss of
1403  * connectivity with the fc-nvme target, so you may never get a
1404  * response even if you tried.  As such, the action of this routine
1405  * is to asynchronously send the LS, ignore any results of the LS, and
1406  * continue on with terminating the association. If the fc-nvme target
1407  * is present and receives the LS, it too can tear down.
1408  */
1409 static void
1410 nvme_fc_xmt_disconnect_assoc(struct nvme_fc_ctrl *ctrl)
1411 {
1412 	struct fcnvme_ls_disconnect_rqst *discon_rqst;
1413 	struct fcnvme_ls_disconnect_acc *discon_acc;
1414 	struct nvmefc_ls_req_op *lsop;
1415 	struct nvmefc_ls_req *lsreq;
1416 	int ret;
1417 
1418 	lsop = kzalloc((sizeof(*lsop) +
1419 			 ctrl->lport->ops->lsrqst_priv_sz +
1420 			 sizeof(*discon_rqst) + sizeof(*discon_acc)),
1421 			GFP_KERNEL);
1422 	if (!lsop)
1423 		/* couldn't sent it... too bad */
1424 		return;
1425 
1426 	lsreq = &lsop->ls_req;
1427 
1428 	lsreq->private = (void *)&lsop[1];
1429 	discon_rqst = (struct fcnvme_ls_disconnect_rqst *)
1430 			(lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1431 	discon_acc = (struct fcnvme_ls_disconnect_acc *)&discon_rqst[1];
1432 
1433 	discon_rqst->w0.ls_cmd = FCNVME_LS_DISCONNECT;
1434 	discon_rqst->desc_list_len = cpu_to_be32(
1435 				sizeof(struct fcnvme_lsdesc_assoc_id) +
1436 				sizeof(struct fcnvme_lsdesc_disconn_cmd));
1437 
1438 	discon_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1439 	discon_rqst->associd.desc_len =
1440 			fcnvme_lsdesc_len(
1441 				sizeof(struct fcnvme_lsdesc_assoc_id));
1442 
1443 	discon_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1444 
1445 	discon_rqst->discon_cmd.desc_tag = cpu_to_be32(
1446 						FCNVME_LSDESC_DISCONN_CMD);
1447 	discon_rqst->discon_cmd.desc_len =
1448 			fcnvme_lsdesc_len(
1449 				sizeof(struct fcnvme_lsdesc_disconn_cmd));
1450 	discon_rqst->discon_cmd.scope = FCNVME_DISCONN_ASSOCIATION;
1451 	discon_rqst->discon_cmd.id = cpu_to_be64(ctrl->association_id);
1452 
1453 	lsreq->rqstaddr = discon_rqst;
1454 	lsreq->rqstlen = sizeof(*discon_rqst);
1455 	lsreq->rspaddr = discon_acc;
1456 	lsreq->rsplen = sizeof(*discon_acc);
1457 	lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1458 
1459 	ret = nvme_fc_send_ls_req_async(ctrl->rport, lsop,
1460 				nvme_fc_disconnect_assoc_done);
1461 	if (ret)
1462 		kfree(lsop);
1463 
1464 	/* only meaningful part to terminating the association */
1465 	ctrl->association_id = 0;
1466 }
1467 
1468 
1469 /* *********************** NVME Ctrl Routines **************************** */
1470 
1471 static void nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg);
1472 
1473 static int
1474 nvme_fc_reinit_request(void *data, struct request *rq)
1475 {
1476 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1477 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1478 
1479 	memset(cmdiu, 0, sizeof(*cmdiu));
1480 	cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1481 	cmdiu->fc_id = NVME_CMD_FC_ID;
1482 	cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1483 	memset(&op->rsp_iu, 0, sizeof(op->rsp_iu));
1484 
1485 	return 0;
1486 }
1487 
1488 static void
1489 __nvme_fc_exit_request(struct nvme_fc_ctrl *ctrl,
1490 		struct nvme_fc_fcp_op *op)
1491 {
1492 	fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.rspdma,
1493 				sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1494 	fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.cmddma,
1495 				sizeof(op->cmd_iu), DMA_TO_DEVICE);
1496 
1497 	atomic_set(&op->state, FCPOP_STATE_UNINIT);
1498 }
1499 
1500 static void
1501 nvme_fc_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1502 		unsigned int hctx_idx)
1503 {
1504 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1505 
1506 	return __nvme_fc_exit_request(set->driver_data, op);
1507 }
1508 
1509 static int
1510 __nvme_fc_abort_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_fcp_op *op)
1511 {
1512 	unsigned long flags;
1513 	int opstate;
1514 
1515 	spin_lock_irqsave(&ctrl->lock, flags);
1516 	opstate = atomic_xchg(&op->state, FCPOP_STATE_ABORTED);
1517 	if (opstate != FCPOP_STATE_ACTIVE)
1518 		atomic_set(&op->state, opstate);
1519 	else if (ctrl->flags & FCCTRL_TERMIO)
1520 		ctrl->iocnt++;
1521 	spin_unlock_irqrestore(&ctrl->lock, flags);
1522 
1523 	if (opstate != FCPOP_STATE_ACTIVE)
1524 		return -ECANCELED;
1525 
1526 	ctrl->lport->ops->fcp_abort(&ctrl->lport->localport,
1527 					&ctrl->rport->remoteport,
1528 					op->queue->lldd_handle,
1529 					&op->fcp_req);
1530 
1531 	return 0;
1532 }
1533 
1534 static void
1535 nvme_fc_abort_aen_ops(struct nvme_fc_ctrl *ctrl)
1536 {
1537 	struct nvme_fc_fcp_op *aen_op = ctrl->aen_ops;
1538 	int i;
1539 
1540 	for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++)
1541 		__nvme_fc_abort_op(ctrl, aen_op);
1542 }
1543 
1544 static inline void
1545 __nvme_fc_fcpop_chk_teardowns(struct nvme_fc_ctrl *ctrl,
1546 		struct nvme_fc_fcp_op *op, int opstate)
1547 {
1548 	unsigned long flags;
1549 
1550 	if (opstate == FCPOP_STATE_ABORTED) {
1551 		spin_lock_irqsave(&ctrl->lock, flags);
1552 		if (ctrl->flags & FCCTRL_TERMIO) {
1553 			if (!--ctrl->iocnt)
1554 				wake_up(&ctrl->ioabort_wait);
1555 		}
1556 		spin_unlock_irqrestore(&ctrl->lock, flags);
1557 	}
1558 }
1559 
1560 static void
1561 nvme_fc_fcpio_done(struct nvmefc_fcp_req *req)
1562 {
1563 	struct nvme_fc_fcp_op *op = fcp_req_to_fcp_op(req);
1564 	struct request *rq = op->rq;
1565 	struct nvmefc_fcp_req *freq = &op->fcp_req;
1566 	struct nvme_fc_ctrl *ctrl = op->ctrl;
1567 	struct nvme_fc_queue *queue = op->queue;
1568 	struct nvme_completion *cqe = &op->rsp_iu.cqe;
1569 	struct nvme_command *sqe = &op->cmd_iu.sqe;
1570 	__le16 status = cpu_to_le16(NVME_SC_SUCCESS << 1);
1571 	union nvme_result result;
1572 	bool terminate_assoc = true;
1573 	int opstate;
1574 
1575 	/*
1576 	 * WARNING:
1577 	 * The current linux implementation of a nvme controller
1578 	 * allocates a single tag set for all io queues and sizes
1579 	 * the io queues to fully hold all possible tags. Thus, the
1580 	 * implementation does not reference or care about the sqhd
1581 	 * value as it never needs to use the sqhd/sqtail pointers
1582 	 * for submission pacing.
1583 	 *
1584 	 * This affects the FC-NVME implementation in two ways:
1585 	 * 1) As the value doesn't matter, we don't need to waste
1586 	 *    cycles extracting it from ERSPs and stamping it in the
1587 	 *    cases where the transport fabricates CQEs on successful
1588 	 *    completions.
1589 	 * 2) The FC-NVME implementation requires that delivery of
1590 	 *    ERSP completions are to go back to the nvme layer in order
1591 	 *    relative to the rsn, such that the sqhd value will always
1592 	 *    be "in order" for the nvme layer. As the nvme layer in
1593 	 *    linux doesn't care about sqhd, there's no need to return
1594 	 *    them in order.
1595 	 *
1596 	 * Additionally:
1597 	 * As the core nvme layer in linux currently does not look at
1598 	 * every field in the cqe - in cases where the FC transport must
1599 	 * fabricate a CQE, the following fields will not be set as they
1600 	 * are not referenced:
1601 	 *      cqe.sqid,  cqe.sqhd,  cqe.command_id
1602 	 *
1603 	 * Failure or error of an individual i/o, in a transport
1604 	 * detected fashion unrelated to the nvme completion status,
1605 	 * potentially cause the initiator and target sides to get out
1606 	 * of sync on SQ head/tail (aka outstanding io count allowed).
1607 	 * Per FC-NVME spec, failure of an individual command requires
1608 	 * the connection to be terminated, which in turn requires the
1609 	 * association to be terminated.
1610 	 */
1611 
1612 	opstate = atomic_xchg(&op->state, FCPOP_STATE_COMPLETE);
1613 
1614 	fc_dma_sync_single_for_cpu(ctrl->lport->dev, op->fcp_req.rspdma,
1615 				sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1616 
1617 	if (opstate == FCPOP_STATE_ABORTED)
1618 		status = cpu_to_le16(NVME_SC_ABORT_REQ << 1);
1619 	else if (freq->status)
1620 		status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1621 
1622 	/*
1623 	 * For the linux implementation, if we have an unsuccesful
1624 	 * status, they blk-mq layer can typically be called with the
1625 	 * non-zero status and the content of the cqe isn't important.
1626 	 */
1627 	if (status)
1628 		goto done;
1629 
1630 	/*
1631 	 * command completed successfully relative to the wire
1632 	 * protocol. However, validate anything received and
1633 	 * extract the status and result from the cqe (create it
1634 	 * where necessary).
1635 	 */
1636 
1637 	switch (freq->rcv_rsplen) {
1638 
1639 	case 0:
1640 	case NVME_FC_SIZEOF_ZEROS_RSP:
1641 		/*
1642 		 * No response payload or 12 bytes of payload (which
1643 		 * should all be zeros) are considered successful and
1644 		 * no payload in the CQE by the transport.
1645 		 */
1646 		if (freq->transferred_length !=
1647 			be32_to_cpu(op->cmd_iu.data_len)) {
1648 			status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1649 			goto done;
1650 		}
1651 		result.u64 = 0;
1652 		break;
1653 
1654 	case sizeof(struct nvme_fc_ersp_iu):
1655 		/*
1656 		 * The ERSP IU contains a full completion with CQE.
1657 		 * Validate ERSP IU and look at cqe.
1658 		 */
1659 		if (unlikely(be16_to_cpu(op->rsp_iu.iu_len) !=
1660 					(freq->rcv_rsplen / 4) ||
1661 			     be32_to_cpu(op->rsp_iu.xfrd_len) !=
1662 					freq->transferred_length ||
1663 			     op->rsp_iu.status_code ||
1664 			     sqe->common.command_id != cqe->command_id)) {
1665 			status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1666 			goto done;
1667 		}
1668 		result = cqe->result;
1669 		status = cqe->status;
1670 		break;
1671 
1672 	default:
1673 		status = cpu_to_le16(NVME_SC_INTERNAL << 1);
1674 		goto done;
1675 	}
1676 
1677 	terminate_assoc = false;
1678 
1679 done:
1680 	if (op->flags & FCOP_FLAGS_AEN) {
1681 		nvme_complete_async_event(&queue->ctrl->ctrl, status, &result);
1682 		__nvme_fc_fcpop_chk_teardowns(ctrl, op, opstate);
1683 		atomic_set(&op->state, FCPOP_STATE_IDLE);
1684 		op->flags = FCOP_FLAGS_AEN;	/* clear other flags */
1685 		nvme_fc_ctrl_put(ctrl);
1686 		goto check_error;
1687 	}
1688 
1689 	/*
1690 	 * Force failures of commands if we're killing the controller
1691 	 * or have an error on a command used to create an new association
1692 	 */
1693 	if (status &&
1694 	    (blk_queue_dying(rq->q) ||
1695 	     ctrl->ctrl.state == NVME_CTRL_NEW ||
1696 	     ctrl->ctrl.state == NVME_CTRL_CONNECTING))
1697 		status |= cpu_to_le16(NVME_SC_DNR << 1);
1698 
1699 	__nvme_fc_fcpop_chk_teardowns(ctrl, op, opstate);
1700 	nvme_end_request(rq, status, result);
1701 
1702 check_error:
1703 	if (terminate_assoc)
1704 		nvme_fc_error_recovery(ctrl, "transport detected io error");
1705 }
1706 
1707 static int
1708 __nvme_fc_init_request(struct nvme_fc_ctrl *ctrl,
1709 		struct nvme_fc_queue *queue, struct nvme_fc_fcp_op *op,
1710 		struct request *rq, u32 rqno)
1711 {
1712 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1713 	int ret = 0;
1714 
1715 	memset(op, 0, sizeof(*op));
1716 	op->fcp_req.cmdaddr = &op->cmd_iu;
1717 	op->fcp_req.cmdlen = sizeof(op->cmd_iu);
1718 	op->fcp_req.rspaddr = &op->rsp_iu;
1719 	op->fcp_req.rsplen = sizeof(op->rsp_iu);
1720 	op->fcp_req.done = nvme_fc_fcpio_done;
1721 	op->fcp_req.first_sgl = (struct scatterlist *)&op[1];
1722 	op->fcp_req.private = &op->fcp_req.first_sgl[SG_CHUNK_SIZE];
1723 	op->ctrl = ctrl;
1724 	op->queue = queue;
1725 	op->rq = rq;
1726 	op->rqno = rqno;
1727 
1728 	cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1729 	cmdiu->fc_id = NVME_CMD_FC_ID;
1730 	cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1731 
1732 	op->fcp_req.cmddma = fc_dma_map_single(ctrl->lport->dev,
1733 				&op->cmd_iu, sizeof(op->cmd_iu), DMA_TO_DEVICE);
1734 	if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.cmddma)) {
1735 		dev_err(ctrl->dev,
1736 			"FCP Op failed - cmdiu dma mapping failed.\n");
1737 		ret = EFAULT;
1738 		goto out_on_error;
1739 	}
1740 
1741 	op->fcp_req.rspdma = fc_dma_map_single(ctrl->lport->dev,
1742 				&op->rsp_iu, sizeof(op->rsp_iu),
1743 				DMA_FROM_DEVICE);
1744 	if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.rspdma)) {
1745 		dev_err(ctrl->dev,
1746 			"FCP Op failed - rspiu dma mapping failed.\n");
1747 		ret = EFAULT;
1748 	}
1749 
1750 	atomic_set(&op->state, FCPOP_STATE_IDLE);
1751 out_on_error:
1752 	return ret;
1753 }
1754 
1755 static int
1756 nvme_fc_init_request(struct blk_mq_tag_set *set, struct request *rq,
1757 		unsigned int hctx_idx, unsigned int numa_node)
1758 {
1759 	struct nvme_fc_ctrl *ctrl = set->driver_data;
1760 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1761 	int queue_idx = (set == &ctrl->tag_set) ? hctx_idx + 1 : 0;
1762 	struct nvme_fc_queue *queue = &ctrl->queues[queue_idx];
1763 
1764 	return __nvme_fc_init_request(ctrl, queue, op, rq, queue->rqcnt++);
1765 }
1766 
1767 static int
1768 nvme_fc_init_aen_ops(struct nvme_fc_ctrl *ctrl)
1769 {
1770 	struct nvme_fc_fcp_op *aen_op;
1771 	struct nvme_fc_cmd_iu *cmdiu;
1772 	struct nvme_command *sqe;
1773 	void *private;
1774 	int i, ret;
1775 
1776 	aen_op = ctrl->aen_ops;
1777 	for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++) {
1778 		private = kzalloc(ctrl->lport->ops->fcprqst_priv_sz,
1779 						GFP_KERNEL);
1780 		if (!private)
1781 			return -ENOMEM;
1782 
1783 		cmdiu = &aen_op->cmd_iu;
1784 		sqe = &cmdiu->sqe;
1785 		ret = __nvme_fc_init_request(ctrl, &ctrl->queues[0],
1786 				aen_op, (struct request *)NULL,
1787 				(NVME_AQ_BLK_MQ_DEPTH + i));
1788 		if (ret) {
1789 			kfree(private);
1790 			return ret;
1791 		}
1792 
1793 		aen_op->flags = FCOP_FLAGS_AEN;
1794 		aen_op->fcp_req.first_sgl = NULL; /* no sg list */
1795 		aen_op->fcp_req.private = private;
1796 
1797 		memset(sqe, 0, sizeof(*sqe));
1798 		sqe->common.opcode = nvme_admin_async_event;
1799 		/* Note: core layer may overwrite the sqe.command_id value */
1800 		sqe->common.command_id = NVME_AQ_BLK_MQ_DEPTH + i;
1801 	}
1802 	return 0;
1803 }
1804 
1805 static void
1806 nvme_fc_term_aen_ops(struct nvme_fc_ctrl *ctrl)
1807 {
1808 	struct nvme_fc_fcp_op *aen_op;
1809 	int i;
1810 
1811 	aen_op = ctrl->aen_ops;
1812 	for (i = 0; i < NVME_NR_AEN_COMMANDS; i++, aen_op++) {
1813 		if (!aen_op->fcp_req.private)
1814 			continue;
1815 
1816 		__nvme_fc_exit_request(ctrl, aen_op);
1817 
1818 		kfree(aen_op->fcp_req.private);
1819 		aen_op->fcp_req.private = NULL;
1820 	}
1821 }
1822 
1823 static inline void
1824 __nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, struct nvme_fc_ctrl *ctrl,
1825 		unsigned int qidx)
1826 {
1827 	struct nvme_fc_queue *queue = &ctrl->queues[qidx];
1828 
1829 	hctx->driver_data = queue;
1830 	queue->hctx = hctx;
1831 }
1832 
1833 static int
1834 nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1835 		unsigned int hctx_idx)
1836 {
1837 	struct nvme_fc_ctrl *ctrl = data;
1838 
1839 	__nvme_fc_init_hctx(hctx, ctrl, hctx_idx + 1);
1840 
1841 	return 0;
1842 }
1843 
1844 static int
1845 nvme_fc_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1846 		unsigned int hctx_idx)
1847 {
1848 	struct nvme_fc_ctrl *ctrl = data;
1849 
1850 	__nvme_fc_init_hctx(hctx, ctrl, hctx_idx);
1851 
1852 	return 0;
1853 }
1854 
1855 static void
1856 nvme_fc_init_queue(struct nvme_fc_ctrl *ctrl, int idx)
1857 {
1858 	struct nvme_fc_queue *queue;
1859 
1860 	queue = &ctrl->queues[idx];
1861 	memset(queue, 0, sizeof(*queue));
1862 	queue->ctrl = ctrl;
1863 	queue->qnum = idx;
1864 	atomic_set(&queue->csn, 1);
1865 	queue->dev = ctrl->dev;
1866 
1867 	if (idx > 0)
1868 		queue->cmnd_capsule_len = ctrl->ctrl.ioccsz * 16;
1869 	else
1870 		queue->cmnd_capsule_len = sizeof(struct nvme_command);
1871 
1872 	/*
1873 	 * Considered whether we should allocate buffers for all SQEs
1874 	 * and CQEs and dma map them - mapping their respective entries
1875 	 * into the request structures (kernel vm addr and dma address)
1876 	 * thus the driver could use the buffers/mappings directly.
1877 	 * It only makes sense if the LLDD would use them for its
1878 	 * messaging api. It's very unlikely most adapter api's would use
1879 	 * a native NVME sqe/cqe. More reasonable if FC-NVME IU payload
1880 	 * structures were used instead.
1881 	 */
1882 }
1883 
1884 /*
1885  * This routine terminates a queue at the transport level.
1886  * The transport has already ensured that all outstanding ios on
1887  * the queue have been terminated.
1888  * The transport will send a Disconnect LS request to terminate
1889  * the queue's connection. Termination of the admin queue will also
1890  * terminate the association at the target.
1891  */
1892 static void
1893 nvme_fc_free_queue(struct nvme_fc_queue *queue)
1894 {
1895 	if (!test_and_clear_bit(NVME_FC_Q_CONNECTED, &queue->flags))
1896 		return;
1897 
1898 	clear_bit(NVME_FC_Q_LIVE, &queue->flags);
1899 	/*
1900 	 * Current implementation never disconnects a single queue.
1901 	 * It always terminates a whole association. So there is never
1902 	 * a disconnect(queue) LS sent to the target.
1903 	 */
1904 
1905 	queue->connection_id = 0;
1906 }
1907 
1908 static void
1909 __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *ctrl,
1910 	struct nvme_fc_queue *queue, unsigned int qidx)
1911 {
1912 	if (ctrl->lport->ops->delete_queue)
1913 		ctrl->lport->ops->delete_queue(&ctrl->lport->localport, qidx,
1914 				queue->lldd_handle);
1915 	queue->lldd_handle = NULL;
1916 }
1917 
1918 static void
1919 nvme_fc_free_io_queues(struct nvme_fc_ctrl *ctrl)
1920 {
1921 	int i;
1922 
1923 	for (i = 1; i < ctrl->ctrl.queue_count; i++)
1924 		nvme_fc_free_queue(&ctrl->queues[i]);
1925 }
1926 
1927 static int
1928 __nvme_fc_create_hw_queue(struct nvme_fc_ctrl *ctrl,
1929 	struct nvme_fc_queue *queue, unsigned int qidx, u16 qsize)
1930 {
1931 	int ret = 0;
1932 
1933 	queue->lldd_handle = NULL;
1934 	if (ctrl->lport->ops->create_queue)
1935 		ret = ctrl->lport->ops->create_queue(&ctrl->lport->localport,
1936 				qidx, qsize, &queue->lldd_handle);
1937 
1938 	return ret;
1939 }
1940 
1941 static void
1942 nvme_fc_delete_hw_io_queues(struct nvme_fc_ctrl *ctrl)
1943 {
1944 	struct nvme_fc_queue *queue = &ctrl->queues[ctrl->ctrl.queue_count - 1];
1945 	int i;
1946 
1947 	for (i = ctrl->ctrl.queue_count - 1; i >= 1; i--, queue--)
1948 		__nvme_fc_delete_hw_queue(ctrl, queue, i);
1949 }
1950 
1951 static int
1952 nvme_fc_create_hw_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1953 {
1954 	struct nvme_fc_queue *queue = &ctrl->queues[1];
1955 	int i, ret;
1956 
1957 	for (i = 1; i < ctrl->ctrl.queue_count; i++, queue++) {
1958 		ret = __nvme_fc_create_hw_queue(ctrl, queue, i, qsize);
1959 		if (ret)
1960 			goto delete_queues;
1961 	}
1962 
1963 	return 0;
1964 
1965 delete_queues:
1966 	for (; i >= 0; i--)
1967 		__nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[i], i);
1968 	return ret;
1969 }
1970 
1971 static int
1972 nvme_fc_connect_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1973 {
1974 	int i, ret = 0;
1975 
1976 	for (i = 1; i < ctrl->ctrl.queue_count; i++) {
1977 		ret = nvme_fc_connect_queue(ctrl, &ctrl->queues[i], qsize,
1978 					(qsize / 5));
1979 		if (ret)
1980 			break;
1981 		ret = nvmf_connect_io_queue(&ctrl->ctrl, i);
1982 		if (ret)
1983 			break;
1984 
1985 		set_bit(NVME_FC_Q_LIVE, &ctrl->queues[i].flags);
1986 	}
1987 
1988 	return ret;
1989 }
1990 
1991 static void
1992 nvme_fc_init_io_queues(struct nvme_fc_ctrl *ctrl)
1993 {
1994 	int i;
1995 
1996 	for (i = 1; i < ctrl->ctrl.queue_count; i++)
1997 		nvme_fc_init_queue(ctrl, i);
1998 }
1999 
2000 static void
2001 nvme_fc_ctrl_free(struct kref *ref)
2002 {
2003 	struct nvme_fc_ctrl *ctrl =
2004 		container_of(ref, struct nvme_fc_ctrl, ref);
2005 	unsigned long flags;
2006 
2007 	if (ctrl->ctrl.tagset) {
2008 		blk_cleanup_queue(ctrl->ctrl.connect_q);
2009 		blk_mq_free_tag_set(&ctrl->tag_set);
2010 	}
2011 
2012 	/* remove from rport list */
2013 	spin_lock_irqsave(&ctrl->rport->lock, flags);
2014 	list_del(&ctrl->ctrl_list);
2015 	spin_unlock_irqrestore(&ctrl->rport->lock, flags);
2016 
2017 	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2018 	blk_cleanup_queue(ctrl->ctrl.admin_q);
2019 	blk_mq_free_tag_set(&ctrl->admin_tag_set);
2020 
2021 	kfree(ctrl->queues);
2022 
2023 	put_device(ctrl->dev);
2024 	nvme_fc_rport_put(ctrl->rport);
2025 
2026 	ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
2027 	if (ctrl->ctrl.opts)
2028 		nvmf_free_options(ctrl->ctrl.opts);
2029 	kfree(ctrl);
2030 }
2031 
2032 static void
2033 nvme_fc_ctrl_put(struct nvme_fc_ctrl *ctrl)
2034 {
2035 	kref_put(&ctrl->ref, nvme_fc_ctrl_free);
2036 }
2037 
2038 static int
2039 nvme_fc_ctrl_get(struct nvme_fc_ctrl *ctrl)
2040 {
2041 	return kref_get_unless_zero(&ctrl->ref);
2042 }
2043 
2044 /*
2045  * All accesses from nvme core layer done - can now free the
2046  * controller. Called after last nvme_put_ctrl() call
2047  */
2048 static void
2049 nvme_fc_nvme_ctrl_freed(struct nvme_ctrl *nctrl)
2050 {
2051 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2052 
2053 	WARN_ON(nctrl != &ctrl->ctrl);
2054 
2055 	nvme_fc_ctrl_put(ctrl);
2056 }
2057 
2058 static void
2059 nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg)
2060 {
2061 	/* only proceed if in LIVE state - e.g. on first error */
2062 	if (ctrl->ctrl.state != NVME_CTRL_LIVE)
2063 		return;
2064 
2065 	dev_warn(ctrl->ctrl.device,
2066 		"NVME-FC{%d}: transport association error detected: %s\n",
2067 		ctrl->cnum, errmsg);
2068 	dev_warn(ctrl->ctrl.device,
2069 		"NVME-FC{%d}: resetting controller\n", ctrl->cnum);
2070 
2071 	nvme_reset_ctrl(&ctrl->ctrl);
2072 }
2073 
2074 static enum blk_eh_timer_return
2075 nvme_fc_timeout(struct request *rq, bool reserved)
2076 {
2077 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2078 	struct nvme_fc_ctrl *ctrl = op->ctrl;
2079 	int ret;
2080 
2081 	if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE ||
2082 			atomic_read(&op->state) == FCPOP_STATE_ABORTED)
2083 		return BLK_EH_RESET_TIMER;
2084 
2085 	ret = __nvme_fc_abort_op(ctrl, op);
2086 	if (ret)
2087 		/* io wasn't active to abort */
2088 		return BLK_EH_NOT_HANDLED;
2089 
2090 	/*
2091 	 * we can't individually ABTS an io without affecting the queue,
2092 	 * thus killing the queue, adn thus the association.
2093 	 * So resolve by performing a controller reset, which will stop
2094 	 * the host/io stack, terminate the association on the link,
2095 	 * and recreate an association on the link.
2096 	 */
2097 	nvme_fc_error_recovery(ctrl, "io timeout error");
2098 
2099 	/*
2100 	 * the io abort has been initiated. Have the reset timer
2101 	 * restarted and the abort completion will complete the io
2102 	 * shortly. Avoids a synchronous wait while the abort finishes.
2103 	 */
2104 	return BLK_EH_RESET_TIMER;
2105 }
2106 
2107 static int
2108 nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
2109 		struct nvme_fc_fcp_op *op)
2110 {
2111 	struct nvmefc_fcp_req *freq = &op->fcp_req;
2112 	enum dma_data_direction dir;
2113 	int ret;
2114 
2115 	freq->sg_cnt = 0;
2116 
2117 	if (!blk_rq_payload_bytes(rq))
2118 		return 0;
2119 
2120 	freq->sg_table.sgl = freq->first_sgl;
2121 	ret = sg_alloc_table_chained(&freq->sg_table,
2122 			blk_rq_nr_phys_segments(rq), freq->sg_table.sgl);
2123 	if (ret)
2124 		return -ENOMEM;
2125 
2126 	op->nents = blk_rq_map_sg(rq->q, rq, freq->sg_table.sgl);
2127 	WARN_ON(op->nents > blk_rq_nr_phys_segments(rq));
2128 	dir = (rq_data_dir(rq) == WRITE) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
2129 	freq->sg_cnt = fc_dma_map_sg(ctrl->lport->dev, freq->sg_table.sgl,
2130 				op->nents, dir);
2131 	if (unlikely(freq->sg_cnt <= 0)) {
2132 		sg_free_table_chained(&freq->sg_table, true);
2133 		freq->sg_cnt = 0;
2134 		return -EFAULT;
2135 	}
2136 
2137 	/*
2138 	 * TODO: blk_integrity_rq(rq)  for DIF
2139 	 */
2140 	return 0;
2141 }
2142 
2143 static void
2144 nvme_fc_unmap_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
2145 		struct nvme_fc_fcp_op *op)
2146 {
2147 	struct nvmefc_fcp_req *freq = &op->fcp_req;
2148 
2149 	if (!freq->sg_cnt)
2150 		return;
2151 
2152 	fc_dma_unmap_sg(ctrl->lport->dev, freq->sg_table.sgl, op->nents,
2153 				((rq_data_dir(rq) == WRITE) ?
2154 					DMA_TO_DEVICE : DMA_FROM_DEVICE));
2155 
2156 	nvme_cleanup_cmd(rq);
2157 
2158 	sg_free_table_chained(&freq->sg_table, true);
2159 
2160 	freq->sg_cnt = 0;
2161 }
2162 
2163 /*
2164  * In FC, the queue is a logical thing. At transport connect, the target
2165  * creates its "queue" and returns a handle that is to be given to the
2166  * target whenever it posts something to the corresponding SQ.  When an
2167  * SQE is sent on a SQ, FC effectively considers the SQE, or rather the
2168  * command contained within the SQE, an io, and assigns a FC exchange
2169  * to it. The SQE and the associated SQ handle are sent in the initial
2170  * CMD IU sents on the exchange. All transfers relative to the io occur
2171  * as part of the exchange.  The CQE is the last thing for the io,
2172  * which is transferred (explicitly or implicitly) with the RSP IU
2173  * sent on the exchange. After the CQE is received, the FC exchange is
2174  * terminaed and the Exchange may be used on a different io.
2175  *
2176  * The transport to LLDD api has the transport making a request for a
2177  * new fcp io request to the LLDD. The LLDD then allocates a FC exchange
2178  * resource and transfers the command. The LLDD will then process all
2179  * steps to complete the io. Upon completion, the transport done routine
2180  * is called.
2181  *
2182  * So - while the operation is outstanding to the LLDD, there is a link
2183  * level FC exchange resource that is also outstanding. This must be
2184  * considered in all cleanup operations.
2185  */
2186 static blk_status_t
2187 nvme_fc_start_fcp_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
2188 	struct nvme_fc_fcp_op *op, u32 data_len,
2189 	enum nvmefc_fcp_datadir	io_dir)
2190 {
2191 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2192 	struct nvme_command *sqe = &cmdiu->sqe;
2193 	u32 csn;
2194 	int ret;
2195 
2196 	/*
2197 	 * before attempting to send the io, check to see if we believe
2198 	 * the target device is present
2199 	 */
2200 	if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
2201 		return BLK_STS_RESOURCE;
2202 
2203 	if (!nvme_fc_ctrl_get(ctrl))
2204 		return BLK_STS_IOERR;
2205 
2206 	/* format the FC-NVME CMD IU and fcp_req */
2207 	cmdiu->connection_id = cpu_to_be64(queue->connection_id);
2208 	csn = atomic_inc_return(&queue->csn);
2209 	cmdiu->csn = cpu_to_be32(csn);
2210 	cmdiu->data_len = cpu_to_be32(data_len);
2211 	switch (io_dir) {
2212 	case NVMEFC_FCP_WRITE:
2213 		cmdiu->flags = FCNVME_CMD_FLAGS_WRITE;
2214 		break;
2215 	case NVMEFC_FCP_READ:
2216 		cmdiu->flags = FCNVME_CMD_FLAGS_READ;
2217 		break;
2218 	case NVMEFC_FCP_NODATA:
2219 		cmdiu->flags = 0;
2220 		break;
2221 	}
2222 	op->fcp_req.payload_length = data_len;
2223 	op->fcp_req.io_dir = io_dir;
2224 	op->fcp_req.transferred_length = 0;
2225 	op->fcp_req.rcv_rsplen = 0;
2226 	op->fcp_req.status = NVME_SC_SUCCESS;
2227 	op->fcp_req.sqid = cpu_to_le16(queue->qnum);
2228 
2229 	/*
2230 	 * validate per fabric rules, set fields mandated by fabric spec
2231 	 * as well as those by FC-NVME spec.
2232 	 */
2233 	WARN_ON_ONCE(sqe->common.metadata);
2234 	sqe->common.flags |= NVME_CMD_SGL_METABUF;
2235 
2236 	/*
2237 	 * format SQE DPTR field per FC-NVME rules:
2238 	 *    type=0x5     Transport SGL Data Block Descriptor
2239 	 *    subtype=0xA  Transport-specific value
2240 	 *    address=0
2241 	 *    length=length of the data series
2242 	 */
2243 	sqe->rw.dptr.sgl.type = (NVME_TRANSPORT_SGL_DATA_DESC << 4) |
2244 					NVME_SGL_FMT_TRANSPORT_A;
2245 	sqe->rw.dptr.sgl.length = cpu_to_le32(data_len);
2246 	sqe->rw.dptr.sgl.addr = 0;
2247 
2248 	if (!(op->flags & FCOP_FLAGS_AEN)) {
2249 		ret = nvme_fc_map_data(ctrl, op->rq, op);
2250 		if (ret < 0) {
2251 			nvme_cleanup_cmd(op->rq);
2252 			nvme_fc_ctrl_put(ctrl);
2253 			if (ret == -ENOMEM || ret == -EAGAIN)
2254 				return BLK_STS_RESOURCE;
2255 			return BLK_STS_IOERR;
2256 		}
2257 	}
2258 
2259 	fc_dma_sync_single_for_device(ctrl->lport->dev, op->fcp_req.cmddma,
2260 				  sizeof(op->cmd_iu), DMA_TO_DEVICE);
2261 
2262 	atomic_set(&op->state, FCPOP_STATE_ACTIVE);
2263 
2264 	if (!(op->flags & FCOP_FLAGS_AEN))
2265 		blk_mq_start_request(op->rq);
2266 
2267 	ret = ctrl->lport->ops->fcp_io(&ctrl->lport->localport,
2268 					&ctrl->rport->remoteport,
2269 					queue->lldd_handle, &op->fcp_req);
2270 
2271 	if (ret) {
2272 		if (!(op->flags & FCOP_FLAGS_AEN))
2273 			nvme_fc_unmap_data(ctrl, op->rq, op);
2274 
2275 		nvme_fc_ctrl_put(ctrl);
2276 
2277 		if (ctrl->rport->remoteport.port_state == FC_OBJSTATE_ONLINE &&
2278 				ret != -EBUSY)
2279 			return BLK_STS_IOERR;
2280 
2281 		return BLK_STS_RESOURCE;
2282 	}
2283 
2284 	return BLK_STS_OK;
2285 }
2286 
2287 static inline blk_status_t nvme_fc_is_ready(struct nvme_fc_queue *queue,
2288 		struct request *rq)
2289 {
2290 	if (unlikely(!test_bit(NVME_FC_Q_LIVE, &queue->flags)))
2291 		return nvmf_check_init_req(&queue->ctrl->ctrl, rq);
2292 	return BLK_STS_OK;
2293 }
2294 
2295 static blk_status_t
2296 nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,
2297 			const struct blk_mq_queue_data *bd)
2298 {
2299 	struct nvme_ns *ns = hctx->queue->queuedata;
2300 	struct nvme_fc_queue *queue = hctx->driver_data;
2301 	struct nvme_fc_ctrl *ctrl = queue->ctrl;
2302 	struct request *rq = bd->rq;
2303 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2304 	struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2305 	struct nvme_command *sqe = &cmdiu->sqe;
2306 	enum nvmefc_fcp_datadir	io_dir;
2307 	u32 data_len;
2308 	blk_status_t ret;
2309 
2310 	ret = nvme_fc_is_ready(queue, rq);
2311 	if (unlikely(ret))
2312 		return ret;
2313 
2314 	ret = nvme_setup_cmd(ns, rq, sqe);
2315 	if (ret)
2316 		return ret;
2317 
2318 	data_len = blk_rq_payload_bytes(rq);
2319 	if (data_len)
2320 		io_dir = ((rq_data_dir(rq) == WRITE) ?
2321 					NVMEFC_FCP_WRITE : NVMEFC_FCP_READ);
2322 	else
2323 		io_dir = NVMEFC_FCP_NODATA;
2324 
2325 	return nvme_fc_start_fcp_op(ctrl, queue, op, data_len, io_dir);
2326 }
2327 
2328 static struct blk_mq_tags *
2329 nvme_fc_tagset(struct nvme_fc_queue *queue)
2330 {
2331 	if (queue->qnum == 0)
2332 		return queue->ctrl->admin_tag_set.tags[queue->qnum];
2333 
2334 	return queue->ctrl->tag_set.tags[queue->qnum - 1];
2335 }
2336 
2337 static int
2338 nvme_fc_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
2339 
2340 {
2341 	struct nvme_fc_queue *queue = hctx->driver_data;
2342 	struct nvme_fc_ctrl *ctrl = queue->ctrl;
2343 	struct request *req;
2344 	struct nvme_fc_fcp_op *op;
2345 
2346 	req = blk_mq_tag_to_rq(nvme_fc_tagset(queue), tag);
2347 	if (!req)
2348 		return 0;
2349 
2350 	op = blk_mq_rq_to_pdu(req);
2351 
2352 	if ((atomic_read(&op->state) == FCPOP_STATE_ACTIVE) &&
2353 		 (ctrl->lport->ops->poll_queue))
2354 		ctrl->lport->ops->poll_queue(&ctrl->lport->localport,
2355 						 queue->lldd_handle);
2356 
2357 	return ((atomic_read(&op->state) != FCPOP_STATE_ACTIVE));
2358 }
2359 
2360 static void
2361 nvme_fc_submit_async_event(struct nvme_ctrl *arg)
2362 {
2363 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(arg);
2364 	struct nvme_fc_fcp_op *aen_op;
2365 	unsigned long flags;
2366 	bool terminating = false;
2367 	blk_status_t ret;
2368 
2369 	spin_lock_irqsave(&ctrl->lock, flags);
2370 	if (ctrl->flags & FCCTRL_TERMIO)
2371 		terminating = true;
2372 	spin_unlock_irqrestore(&ctrl->lock, flags);
2373 
2374 	if (terminating)
2375 		return;
2376 
2377 	aen_op = &ctrl->aen_ops[0];
2378 
2379 	ret = nvme_fc_start_fcp_op(ctrl, aen_op->queue, aen_op, 0,
2380 					NVMEFC_FCP_NODATA);
2381 	if (ret)
2382 		dev_err(ctrl->ctrl.device,
2383 			"failed async event work\n");
2384 }
2385 
2386 static void
2387 nvme_fc_complete_rq(struct request *rq)
2388 {
2389 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2390 	struct nvme_fc_ctrl *ctrl = op->ctrl;
2391 
2392 	atomic_set(&op->state, FCPOP_STATE_IDLE);
2393 
2394 	nvme_fc_unmap_data(ctrl, rq, op);
2395 	nvme_complete_rq(rq);
2396 	nvme_fc_ctrl_put(ctrl);
2397 }
2398 
2399 /*
2400  * This routine is used by the transport when it needs to find active
2401  * io on a queue that is to be terminated. The transport uses
2402  * blk_mq_tagset_busy_itr() to find the busy requests, which then invoke
2403  * this routine to kill them on a 1 by 1 basis.
2404  *
2405  * As FC allocates FC exchange for each io, the transport must contact
2406  * the LLDD to terminate the exchange, thus releasing the FC exchange.
2407  * After terminating the exchange the LLDD will call the transport's
2408  * normal io done path for the request, but it will have an aborted
2409  * status. The done path will return the io request back to the block
2410  * layer with an error status.
2411  */
2412 static void
2413 nvme_fc_terminate_exchange(struct request *req, void *data, bool reserved)
2414 {
2415 	struct nvme_ctrl *nctrl = data;
2416 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2417 	struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(req);
2418 
2419 	if (!blk_mq_request_started(req))
2420 		return;
2421 
2422 	__nvme_fc_abort_op(ctrl, op);
2423 }
2424 
2425 
2426 static const struct blk_mq_ops nvme_fc_mq_ops = {
2427 	.queue_rq	= nvme_fc_queue_rq,
2428 	.complete	= nvme_fc_complete_rq,
2429 	.init_request	= nvme_fc_init_request,
2430 	.exit_request	= nvme_fc_exit_request,
2431 	.init_hctx	= nvme_fc_init_hctx,
2432 	.poll		= nvme_fc_poll,
2433 	.timeout	= nvme_fc_timeout,
2434 };
2435 
2436 static int
2437 nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)
2438 {
2439 	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2440 	unsigned int nr_io_queues;
2441 	int ret;
2442 
2443 	nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2444 				ctrl->lport->ops->max_hw_queues);
2445 	ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
2446 	if (ret) {
2447 		dev_info(ctrl->ctrl.device,
2448 			"set_queue_count failed: %d\n", ret);
2449 		return ret;
2450 	}
2451 
2452 	ctrl->ctrl.queue_count = nr_io_queues + 1;
2453 	if (!nr_io_queues)
2454 		return 0;
2455 
2456 	nvme_fc_init_io_queues(ctrl);
2457 
2458 	memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set));
2459 	ctrl->tag_set.ops = &nvme_fc_mq_ops;
2460 	ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size;
2461 	ctrl->tag_set.reserved_tags = 1; /* fabric connect */
2462 	ctrl->tag_set.numa_node = NUMA_NO_NODE;
2463 	ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2464 	ctrl->tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
2465 					(SG_CHUNK_SIZE *
2466 						sizeof(struct scatterlist)) +
2467 					ctrl->lport->ops->fcprqst_priv_sz;
2468 	ctrl->tag_set.driver_data = ctrl;
2469 	ctrl->tag_set.nr_hw_queues = ctrl->ctrl.queue_count - 1;
2470 	ctrl->tag_set.timeout = NVME_IO_TIMEOUT;
2471 
2472 	ret = blk_mq_alloc_tag_set(&ctrl->tag_set);
2473 	if (ret)
2474 		return ret;
2475 
2476 	ctrl->ctrl.tagset = &ctrl->tag_set;
2477 
2478 	ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set);
2479 	if (IS_ERR(ctrl->ctrl.connect_q)) {
2480 		ret = PTR_ERR(ctrl->ctrl.connect_q);
2481 		goto out_free_tag_set;
2482 	}
2483 
2484 	ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2485 	if (ret)
2486 		goto out_cleanup_blk_queue;
2487 
2488 	ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2489 	if (ret)
2490 		goto out_delete_hw_queues;
2491 
2492 	return 0;
2493 
2494 out_delete_hw_queues:
2495 	nvme_fc_delete_hw_io_queues(ctrl);
2496 out_cleanup_blk_queue:
2497 	blk_cleanup_queue(ctrl->ctrl.connect_q);
2498 out_free_tag_set:
2499 	blk_mq_free_tag_set(&ctrl->tag_set);
2500 	nvme_fc_free_io_queues(ctrl);
2501 
2502 	/* force put free routine to ignore io queues */
2503 	ctrl->ctrl.tagset = NULL;
2504 
2505 	return ret;
2506 }
2507 
2508 static int
2509 nvme_fc_reinit_io_queues(struct nvme_fc_ctrl *ctrl)
2510 {
2511 	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2512 	unsigned int nr_io_queues;
2513 	int ret;
2514 
2515 	nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2516 				ctrl->lport->ops->max_hw_queues);
2517 	ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
2518 	if (ret) {
2519 		dev_info(ctrl->ctrl.device,
2520 			"set_queue_count failed: %d\n", ret);
2521 		return ret;
2522 	}
2523 
2524 	ctrl->ctrl.queue_count = nr_io_queues + 1;
2525 	/* check for io queues existing */
2526 	if (ctrl->ctrl.queue_count == 1)
2527 		return 0;
2528 
2529 	nvme_fc_init_io_queues(ctrl);
2530 
2531 	ret = nvme_reinit_tagset(&ctrl->ctrl, ctrl->ctrl.tagset);
2532 	if (ret)
2533 		goto out_free_io_queues;
2534 
2535 	ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2536 	if (ret)
2537 		goto out_free_io_queues;
2538 
2539 	ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2540 	if (ret)
2541 		goto out_delete_hw_queues;
2542 
2543 	blk_mq_update_nr_hw_queues(&ctrl->tag_set, nr_io_queues);
2544 
2545 	return 0;
2546 
2547 out_delete_hw_queues:
2548 	nvme_fc_delete_hw_io_queues(ctrl);
2549 out_free_io_queues:
2550 	nvme_fc_free_io_queues(ctrl);
2551 	return ret;
2552 }
2553 
2554 static void
2555 nvme_fc_rport_active_on_lport(struct nvme_fc_rport *rport)
2556 {
2557 	struct nvme_fc_lport *lport = rport->lport;
2558 
2559 	atomic_inc(&lport->act_rport_cnt);
2560 }
2561 
2562 static void
2563 nvme_fc_rport_inactive_on_lport(struct nvme_fc_rport *rport)
2564 {
2565 	struct nvme_fc_lport *lport = rport->lport;
2566 	u32 cnt;
2567 
2568 	cnt = atomic_dec_return(&lport->act_rport_cnt);
2569 	if (cnt == 0 && lport->localport.port_state == FC_OBJSTATE_DELETED)
2570 		lport->ops->localport_delete(&lport->localport);
2571 }
2572 
2573 static int
2574 nvme_fc_ctlr_active_on_rport(struct nvme_fc_ctrl *ctrl)
2575 {
2576 	struct nvme_fc_rport *rport = ctrl->rport;
2577 	u32 cnt;
2578 
2579 	if (ctrl->assoc_active)
2580 		return 1;
2581 
2582 	ctrl->assoc_active = true;
2583 	cnt = atomic_inc_return(&rport->act_ctrl_cnt);
2584 	if (cnt == 1)
2585 		nvme_fc_rport_active_on_lport(rport);
2586 
2587 	return 0;
2588 }
2589 
2590 static int
2591 nvme_fc_ctlr_inactive_on_rport(struct nvme_fc_ctrl *ctrl)
2592 {
2593 	struct nvme_fc_rport *rport = ctrl->rport;
2594 	struct nvme_fc_lport *lport = rport->lport;
2595 	u32 cnt;
2596 
2597 	/* ctrl->assoc_active=false will be set independently */
2598 
2599 	cnt = atomic_dec_return(&rport->act_ctrl_cnt);
2600 	if (cnt == 0) {
2601 		if (rport->remoteport.port_state == FC_OBJSTATE_DELETED)
2602 			lport->ops->remoteport_delete(&rport->remoteport);
2603 		nvme_fc_rport_inactive_on_lport(rport);
2604 	}
2605 
2606 	return 0;
2607 }
2608 
2609 /*
2610  * This routine restarts the controller on the host side, and
2611  * on the link side, recreates the controller association.
2612  */
2613 static int
2614 nvme_fc_create_association(struct nvme_fc_ctrl *ctrl)
2615 {
2616 	struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2617 	int ret;
2618 	bool changed;
2619 
2620 	++ctrl->ctrl.nr_reconnects;
2621 
2622 	if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
2623 		return -ENODEV;
2624 
2625 	if (nvme_fc_ctlr_active_on_rport(ctrl))
2626 		return -ENOTUNIQ;
2627 
2628 	/*
2629 	 * Create the admin queue
2630 	 */
2631 
2632 	nvme_fc_init_queue(ctrl, 0);
2633 
2634 	ret = __nvme_fc_create_hw_queue(ctrl, &ctrl->queues[0], 0,
2635 				NVME_AQ_BLK_MQ_DEPTH);
2636 	if (ret)
2637 		goto out_free_queue;
2638 
2639 	ret = nvme_fc_connect_admin_queue(ctrl, &ctrl->queues[0],
2640 				NVME_AQ_BLK_MQ_DEPTH,
2641 				(NVME_AQ_BLK_MQ_DEPTH / 4));
2642 	if (ret)
2643 		goto out_delete_hw_queue;
2644 
2645 	if (ctrl->ctrl.state != NVME_CTRL_NEW)
2646 		blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2647 
2648 	ret = nvmf_connect_admin_queue(&ctrl->ctrl);
2649 	if (ret)
2650 		goto out_disconnect_admin_queue;
2651 
2652 	set_bit(NVME_FC_Q_LIVE, &ctrl->queues[0].flags);
2653 
2654 	/*
2655 	 * Check controller capabilities
2656 	 *
2657 	 * todo:- add code to check if ctrl attributes changed from
2658 	 * prior connection values
2659 	 */
2660 
2661 	ret = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->ctrl.cap);
2662 	if (ret) {
2663 		dev_err(ctrl->ctrl.device,
2664 			"prop_get NVME_REG_CAP failed\n");
2665 		goto out_disconnect_admin_queue;
2666 	}
2667 
2668 	ctrl->ctrl.sqsize =
2669 		min_t(int, NVME_CAP_MQES(ctrl->ctrl.cap) + 1, ctrl->ctrl.sqsize);
2670 
2671 	ret = nvme_enable_ctrl(&ctrl->ctrl, ctrl->ctrl.cap);
2672 	if (ret)
2673 		goto out_disconnect_admin_queue;
2674 
2675 	ctrl->ctrl.max_hw_sectors =
2676 		(ctrl->lport->ops->max_sgl_segments - 1) << (PAGE_SHIFT - 9);
2677 
2678 	ret = nvme_init_identify(&ctrl->ctrl);
2679 	if (ret)
2680 		goto out_disconnect_admin_queue;
2681 
2682 	/* sanity checks */
2683 
2684 	/* FC-NVME does not have other data in the capsule */
2685 	if (ctrl->ctrl.icdoff) {
2686 		dev_err(ctrl->ctrl.device, "icdoff %d is not supported!\n",
2687 				ctrl->ctrl.icdoff);
2688 		goto out_disconnect_admin_queue;
2689 	}
2690 
2691 	/* FC-NVME supports normal SGL Data Block Descriptors */
2692 
2693 	if (opts->queue_size > ctrl->ctrl.maxcmd) {
2694 		/* warn if maxcmd is lower than queue_size */
2695 		dev_warn(ctrl->ctrl.device,
2696 			"queue_size %zu > ctrl maxcmd %u, reducing "
2697 			"to queue_size\n",
2698 			opts->queue_size, ctrl->ctrl.maxcmd);
2699 		opts->queue_size = ctrl->ctrl.maxcmd;
2700 	}
2701 
2702 	ret = nvme_fc_init_aen_ops(ctrl);
2703 	if (ret)
2704 		goto out_term_aen_ops;
2705 
2706 	/*
2707 	 * Create the io queues
2708 	 */
2709 
2710 	if (ctrl->ctrl.queue_count > 1) {
2711 		if (ctrl->ctrl.state == NVME_CTRL_NEW)
2712 			ret = nvme_fc_create_io_queues(ctrl);
2713 		else
2714 			ret = nvme_fc_reinit_io_queues(ctrl);
2715 		if (ret)
2716 			goto out_term_aen_ops;
2717 	}
2718 
2719 	changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE);
2720 
2721 	ctrl->ctrl.nr_reconnects = 0;
2722 
2723 	if (changed)
2724 		nvme_start_ctrl(&ctrl->ctrl);
2725 
2726 	return 0;	/* Success */
2727 
2728 out_term_aen_ops:
2729 	nvme_fc_term_aen_ops(ctrl);
2730 out_disconnect_admin_queue:
2731 	/* send a Disconnect(association) LS to fc-nvme target */
2732 	nvme_fc_xmt_disconnect_assoc(ctrl);
2733 out_delete_hw_queue:
2734 	__nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2735 out_free_queue:
2736 	nvme_fc_free_queue(&ctrl->queues[0]);
2737 	ctrl->assoc_active = false;
2738 	nvme_fc_ctlr_inactive_on_rport(ctrl);
2739 
2740 	return ret;
2741 }
2742 
2743 /*
2744  * This routine stops operation of the controller on the host side.
2745  * On the host os stack side: Admin and IO queues are stopped,
2746  *   outstanding ios on them terminated via FC ABTS.
2747  * On the link side: the association is terminated.
2748  */
2749 static void
2750 nvme_fc_delete_association(struct nvme_fc_ctrl *ctrl)
2751 {
2752 	unsigned long flags;
2753 
2754 	if (!ctrl->assoc_active)
2755 		return;
2756 	ctrl->assoc_active = false;
2757 
2758 	spin_lock_irqsave(&ctrl->lock, flags);
2759 	ctrl->flags |= FCCTRL_TERMIO;
2760 	ctrl->iocnt = 0;
2761 	spin_unlock_irqrestore(&ctrl->lock, flags);
2762 
2763 	/*
2764 	 * If io queues are present, stop them and terminate all outstanding
2765 	 * ios on them. As FC allocates FC exchange for each io, the
2766 	 * transport must contact the LLDD to terminate the exchange,
2767 	 * thus releasing the FC exchange. We use blk_mq_tagset_busy_itr()
2768 	 * to tell us what io's are busy and invoke a transport routine
2769 	 * to kill them with the LLDD.  After terminating the exchange
2770 	 * the LLDD will call the transport's normal io done path, but it
2771 	 * will have an aborted status. The done path will return the
2772 	 * io requests back to the block layer as part of normal completions
2773 	 * (but with error status).
2774 	 */
2775 	if (ctrl->ctrl.queue_count > 1) {
2776 		nvme_stop_queues(&ctrl->ctrl);
2777 		blk_mq_tagset_busy_iter(&ctrl->tag_set,
2778 				nvme_fc_terminate_exchange, &ctrl->ctrl);
2779 	}
2780 
2781 	/*
2782 	 * Other transports, which don't have link-level contexts bound
2783 	 * to sqe's, would try to gracefully shutdown the controller by
2784 	 * writing the registers for shutdown and polling (call
2785 	 * nvme_shutdown_ctrl()). Given a bunch of i/o was potentially
2786 	 * just aborted and we will wait on those contexts, and given
2787 	 * there was no indication of how live the controlelr is on the
2788 	 * link, don't send more io to create more contexts for the
2789 	 * shutdown. Let the controller fail via keepalive failure if
2790 	 * its still present.
2791 	 */
2792 
2793 	/*
2794 	 * clean up the admin queue. Same thing as above.
2795 	 * use blk_mq_tagset_busy_itr() and the transport routine to
2796 	 * terminate the exchanges.
2797 	 */
2798 	if (ctrl->ctrl.state != NVME_CTRL_NEW)
2799 		blk_mq_quiesce_queue(ctrl->ctrl.admin_q);
2800 	blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
2801 				nvme_fc_terminate_exchange, &ctrl->ctrl);
2802 
2803 	/* kill the aens as they are a separate path */
2804 	nvme_fc_abort_aen_ops(ctrl);
2805 
2806 	/* wait for all io that had to be aborted */
2807 	spin_lock_irq(&ctrl->lock);
2808 	wait_event_lock_irq(ctrl->ioabort_wait, ctrl->iocnt == 0, ctrl->lock);
2809 	ctrl->flags &= ~FCCTRL_TERMIO;
2810 	spin_unlock_irq(&ctrl->lock);
2811 
2812 	nvme_fc_term_aen_ops(ctrl);
2813 
2814 	/*
2815 	 * send a Disconnect(association) LS to fc-nvme target
2816 	 * Note: could have been sent at top of process, but
2817 	 * cleaner on link traffic if after the aborts complete.
2818 	 * Note: if association doesn't exist, association_id will be 0
2819 	 */
2820 	if (ctrl->association_id)
2821 		nvme_fc_xmt_disconnect_assoc(ctrl);
2822 
2823 	if (ctrl->ctrl.tagset) {
2824 		nvme_fc_delete_hw_io_queues(ctrl);
2825 		nvme_fc_free_io_queues(ctrl);
2826 	}
2827 
2828 	__nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2829 	nvme_fc_free_queue(&ctrl->queues[0]);
2830 
2831 	/* re-enable the admin_q so anything new can fast fail */
2832 	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
2833 
2834 	nvme_fc_ctlr_inactive_on_rport(ctrl);
2835 }
2836 
2837 static void
2838 nvme_fc_delete_ctrl(struct nvme_ctrl *nctrl)
2839 {
2840 	struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2841 
2842 	cancel_delayed_work_sync(&ctrl->connect_work);
2843 	/*
2844 	 * kill the association on the link side.  this will block
2845 	 * waiting for io to terminate
2846 	 */
2847 	nvme_fc_delete_association(ctrl);
2848 
2849 	/* resume the io queues so that things will fast fail */
2850 	nvme_start_queues(nctrl);
2851 }
2852 
2853 static void
2854 nvme_fc_reconnect_or_delete(struct nvme_fc_ctrl *ctrl, int status)
2855 {
2856 	struct nvme_fc_rport *rport = ctrl->rport;
2857 	struct nvme_fc_remote_port *portptr = &rport->remoteport;
2858 	unsigned long recon_delay = ctrl->ctrl.opts->reconnect_delay * HZ;
2859 	bool recon = true;
2860 
2861 	if (ctrl->ctrl.state != NVME_CTRL_CONNECTING)
2862 		return;
2863 
2864 	if (portptr->port_state == FC_OBJSTATE_ONLINE)
2865 		dev_info(ctrl->ctrl.device,
2866 			"NVME-FC{%d}: reset: Reconnect attempt failed (%d)\n",
2867 			ctrl->cnum, status);
2868 	else if (time_after_eq(jiffies, rport->dev_loss_end))
2869 		recon = false;
2870 
2871 	if (recon && nvmf_should_reconnect(&ctrl->ctrl)) {
2872 		if (portptr->port_state == FC_OBJSTATE_ONLINE)
2873 			dev_info(ctrl->ctrl.device,
2874 				"NVME-FC{%d}: Reconnect attempt in %ld "
2875 				"seconds\n",
2876 				ctrl->cnum, recon_delay / HZ);
2877 		else if (time_after(jiffies + recon_delay, rport->dev_loss_end))
2878 			recon_delay = rport->dev_loss_end - jiffies;
2879 
2880 		queue_delayed_work(nvme_wq, &ctrl->connect_work, recon_delay);
2881 	} else {
2882 		if (portptr->port_state == FC_OBJSTATE_ONLINE)
2883 			dev_warn(ctrl->ctrl.device,
2884 				"NVME-FC{%d}: Max reconnect attempts (%d) "
2885 				"reached. Removing controller\n",
2886 				ctrl->cnum, ctrl->ctrl.nr_reconnects);
2887 		else
2888 			dev_warn(ctrl->ctrl.device,
2889 				"NVME-FC{%d}: dev_loss_tmo (%d) expired "
2890 				"while waiting for remoteport connectivity. "
2891 				"Removing controller\n", ctrl->cnum,
2892 				portptr->dev_loss_tmo);
2893 		WARN_ON(nvme_delete_ctrl(&ctrl->ctrl));
2894 	}
2895 }
2896 
2897 static void
2898 nvme_fc_reset_ctrl_work(struct work_struct *work)
2899 {
2900 	struct nvme_fc_ctrl *ctrl =
2901 		container_of(work, struct nvme_fc_ctrl, ctrl.reset_work);
2902 	int ret;
2903 
2904 	nvme_stop_ctrl(&ctrl->ctrl);
2905 
2906 	/* will block will waiting for io to terminate */
2907 	nvme_fc_delete_association(ctrl);
2908 
2909 	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_CONNECTING)) {
2910 		dev_err(ctrl->ctrl.device,
2911 			"NVME-FC{%d}: error_recovery: Couldn't change state "
2912 			"to CONNECTING\n", ctrl->cnum);
2913 		return;
2914 	}
2915 
2916 	if (ctrl->rport->remoteport.port_state == FC_OBJSTATE_ONLINE)
2917 		ret = nvme_fc_create_association(ctrl);
2918 	else
2919 		ret = -ENOTCONN;
2920 
2921 	if (ret)
2922 		nvme_fc_reconnect_or_delete(ctrl, ret);
2923 	else
2924 		dev_info(ctrl->ctrl.device,
2925 			"NVME-FC{%d}: controller reset complete\n",
2926 			ctrl->cnum);
2927 }
2928 
2929 static const struct nvme_ctrl_ops nvme_fc_ctrl_ops = {
2930 	.name			= "fc",
2931 	.module			= THIS_MODULE,
2932 	.flags			= NVME_F_FABRICS,
2933 	.reg_read32		= nvmf_reg_read32,
2934 	.reg_read64		= nvmf_reg_read64,
2935 	.reg_write32		= nvmf_reg_write32,
2936 	.free_ctrl		= nvme_fc_nvme_ctrl_freed,
2937 	.submit_async_event	= nvme_fc_submit_async_event,
2938 	.delete_ctrl		= nvme_fc_delete_ctrl,
2939 	.get_address		= nvmf_get_address,
2940 	.reinit_request		= nvme_fc_reinit_request,
2941 };
2942 
2943 static void
2944 nvme_fc_connect_ctrl_work(struct work_struct *work)
2945 {
2946 	int ret;
2947 
2948 	struct nvme_fc_ctrl *ctrl =
2949 			container_of(to_delayed_work(work),
2950 				struct nvme_fc_ctrl, connect_work);
2951 
2952 	ret = nvme_fc_create_association(ctrl);
2953 	if (ret)
2954 		nvme_fc_reconnect_or_delete(ctrl, ret);
2955 	else
2956 		dev_info(ctrl->ctrl.device,
2957 			"NVME-FC{%d}: controller reconnect complete\n",
2958 			ctrl->cnum);
2959 }
2960 
2961 
2962 static const struct blk_mq_ops nvme_fc_admin_mq_ops = {
2963 	.queue_rq	= nvme_fc_queue_rq,
2964 	.complete	= nvme_fc_complete_rq,
2965 	.init_request	= nvme_fc_init_request,
2966 	.exit_request	= nvme_fc_exit_request,
2967 	.init_hctx	= nvme_fc_init_admin_hctx,
2968 	.timeout	= nvme_fc_timeout,
2969 };
2970 
2971 
2972 /*
2973  * Fails a controller request if it matches an existing controller
2974  * (association) with the same tuple:
2975  * <Host NQN, Host ID, local FC port, remote FC port, SUBSYS NQN>
2976  *
2977  * The ports don't need to be compared as they are intrinsically
2978  * already matched by the port pointers supplied.
2979  */
2980 static bool
2981 nvme_fc_existing_controller(struct nvme_fc_rport *rport,
2982 		struct nvmf_ctrl_options *opts)
2983 {
2984 	struct nvme_fc_ctrl *ctrl;
2985 	unsigned long flags;
2986 	bool found = false;
2987 
2988 	spin_lock_irqsave(&rport->lock, flags);
2989 	list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list) {
2990 		found = nvmf_ctlr_matches_baseopts(&ctrl->ctrl, opts);
2991 		if (found)
2992 			break;
2993 	}
2994 	spin_unlock_irqrestore(&rport->lock, flags);
2995 
2996 	return found;
2997 }
2998 
2999 static struct nvme_ctrl *
3000 nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,
3001 	struct nvme_fc_lport *lport, struct nvme_fc_rport *rport)
3002 {
3003 	struct nvme_fc_ctrl *ctrl;
3004 	unsigned long flags;
3005 	int ret, idx, retry;
3006 
3007 	if (!(rport->remoteport.port_role &
3008 	    (FC_PORT_ROLE_NVME_DISCOVERY | FC_PORT_ROLE_NVME_TARGET))) {
3009 		ret = -EBADR;
3010 		goto out_fail;
3011 	}
3012 
3013 	if (!opts->duplicate_connect &&
3014 	    nvme_fc_existing_controller(rport, opts)) {
3015 		ret = -EALREADY;
3016 		goto out_fail;
3017 	}
3018 
3019 	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
3020 	if (!ctrl) {
3021 		ret = -ENOMEM;
3022 		goto out_fail;
3023 	}
3024 
3025 	idx = ida_simple_get(&nvme_fc_ctrl_cnt, 0, 0, GFP_KERNEL);
3026 	if (idx < 0) {
3027 		ret = -ENOSPC;
3028 		goto out_free_ctrl;
3029 	}
3030 
3031 	ctrl->ctrl.opts = opts;
3032 	INIT_LIST_HEAD(&ctrl->ctrl_list);
3033 	ctrl->lport = lport;
3034 	ctrl->rport = rport;
3035 	ctrl->dev = lport->dev;
3036 	ctrl->cnum = idx;
3037 	ctrl->assoc_active = false;
3038 	init_waitqueue_head(&ctrl->ioabort_wait);
3039 
3040 	get_device(ctrl->dev);
3041 	kref_init(&ctrl->ref);
3042 
3043 	INIT_WORK(&ctrl->ctrl.reset_work, nvme_fc_reset_ctrl_work);
3044 	INIT_DELAYED_WORK(&ctrl->connect_work, nvme_fc_connect_ctrl_work);
3045 	spin_lock_init(&ctrl->lock);
3046 
3047 	/* io queue count */
3048 	ctrl->ctrl.queue_count = min_t(unsigned int,
3049 				opts->nr_io_queues,
3050 				lport->ops->max_hw_queues);
3051 	ctrl->ctrl.queue_count++;	/* +1 for admin queue */
3052 
3053 	ctrl->ctrl.sqsize = opts->queue_size - 1;
3054 	ctrl->ctrl.kato = opts->kato;
3055 
3056 	ret = -ENOMEM;
3057 	ctrl->queues = kcalloc(ctrl->ctrl.queue_count,
3058 				sizeof(struct nvme_fc_queue), GFP_KERNEL);
3059 	if (!ctrl->queues)
3060 		goto out_free_ida;
3061 
3062 	memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set));
3063 	ctrl->admin_tag_set.ops = &nvme_fc_admin_mq_ops;
3064 	ctrl->admin_tag_set.queue_depth = NVME_AQ_MQ_TAG_DEPTH;
3065 	ctrl->admin_tag_set.reserved_tags = 2; /* fabric connect + Keep-Alive */
3066 	ctrl->admin_tag_set.numa_node = NUMA_NO_NODE;
3067 	ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
3068 					(SG_CHUNK_SIZE *
3069 						sizeof(struct scatterlist)) +
3070 					ctrl->lport->ops->fcprqst_priv_sz;
3071 	ctrl->admin_tag_set.driver_data = ctrl;
3072 	ctrl->admin_tag_set.nr_hw_queues = 1;
3073 	ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT;
3074 	ctrl->admin_tag_set.flags = BLK_MQ_F_NO_SCHED;
3075 
3076 	ret = blk_mq_alloc_tag_set(&ctrl->admin_tag_set);
3077 	if (ret)
3078 		goto out_free_queues;
3079 	ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set;
3080 
3081 	ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);
3082 	if (IS_ERR(ctrl->ctrl.admin_q)) {
3083 		ret = PTR_ERR(ctrl->ctrl.admin_q);
3084 		goto out_free_admin_tag_set;
3085 	}
3086 
3087 	/*
3088 	 * Would have been nice to init io queues tag set as well.
3089 	 * However, we require interaction from the controller
3090 	 * for max io queue count before we can do so.
3091 	 * Defer this to the connect path.
3092 	 */
3093 
3094 	ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_fc_ctrl_ops, 0);
3095 	if (ret)
3096 		goto out_cleanup_admin_q;
3097 
3098 	/* at this point, teardown path changes to ref counting on nvme ctrl */
3099 
3100 	spin_lock_irqsave(&rport->lock, flags);
3101 	list_add_tail(&ctrl->ctrl_list, &rport->ctrl_list);
3102 	spin_unlock_irqrestore(&rport->lock, flags);
3103 
3104 	/*
3105 	 * It's possible that transactions used to create the association
3106 	 * may fail. Examples: CreateAssociation LS or CreateIOConnection
3107 	 * LS gets dropped/corrupted/fails; or a frame gets dropped or a
3108 	 * command times out for one of the actions to init the controller
3109 	 * (Connect, Get/Set_Property, Set_Features, etc). Many of these
3110 	 * transport errors (frame drop, LS failure) inherently must kill
3111 	 * the association. The transport is coded so that any command used
3112 	 * to create the association (prior to a LIVE state transition
3113 	 * while NEW or CONNECTING) will fail if it completes in error or
3114 	 * times out.
3115 	 *
3116 	 * As such: as the connect request was mostly likely due to a
3117 	 * udev event that discovered the remote port, meaning there is
3118 	 * not an admin or script there to restart if the connect
3119 	 * request fails, retry the initial connection creation up to
3120 	 * three times before giving up and declaring failure.
3121 	 */
3122 	for (retry = 0; retry < 3; retry++) {
3123 		ret = nvme_fc_create_association(ctrl);
3124 		if (!ret)
3125 			break;
3126 	}
3127 
3128 	if (ret) {
3129 		/* couldn't schedule retry - fail out */
3130 		dev_err(ctrl->ctrl.device,
3131 			"NVME-FC{%d}: Connect retry failed\n", ctrl->cnum);
3132 
3133 		ctrl->ctrl.opts = NULL;
3134 
3135 		/* initiate nvme ctrl ref counting teardown */
3136 		nvme_uninit_ctrl(&ctrl->ctrl);
3137 
3138 		/* Remove core ctrl ref. */
3139 		nvme_put_ctrl(&ctrl->ctrl);
3140 
3141 		/* as we're past the point where we transition to the ref
3142 		 * counting teardown path, if we return a bad pointer here,
3143 		 * the calling routine, thinking it's prior to the
3144 		 * transition, will do an rport put. Since the teardown
3145 		 * path also does a rport put, we do an extra get here to
3146 		 * so proper order/teardown happens.
3147 		 */
3148 		nvme_fc_rport_get(rport);
3149 
3150 		if (ret > 0)
3151 			ret = -EIO;
3152 		return ERR_PTR(ret);
3153 	}
3154 
3155 	nvme_get_ctrl(&ctrl->ctrl);
3156 
3157 	dev_info(ctrl->ctrl.device,
3158 		"NVME-FC{%d}: new ctrl: NQN \"%s\"\n",
3159 		ctrl->cnum, ctrl->ctrl.opts->subsysnqn);
3160 
3161 	return &ctrl->ctrl;
3162 
3163 out_cleanup_admin_q:
3164 	blk_cleanup_queue(ctrl->ctrl.admin_q);
3165 out_free_admin_tag_set:
3166 	blk_mq_free_tag_set(&ctrl->admin_tag_set);
3167 out_free_queues:
3168 	kfree(ctrl->queues);
3169 out_free_ida:
3170 	put_device(ctrl->dev);
3171 	ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
3172 out_free_ctrl:
3173 	kfree(ctrl);
3174 out_fail:
3175 	/* exit via here doesn't follow ctlr ref points */
3176 	return ERR_PTR(ret);
3177 }
3178 
3179 
3180 struct nvmet_fc_traddr {
3181 	u64	nn;
3182 	u64	pn;
3183 };
3184 
3185 static int
3186 __nvme_fc_parse_u64(substring_t *sstr, u64 *val)
3187 {
3188 	u64 token64;
3189 
3190 	if (match_u64(sstr, &token64))
3191 		return -EINVAL;
3192 	*val = token64;
3193 
3194 	return 0;
3195 }
3196 
3197 /*
3198  * This routine validates and extracts the WWN's from the TRADDR string.
3199  * As kernel parsers need the 0x to determine number base, universally
3200  * build string to parse with 0x prefix before parsing name strings.
3201  */
3202 static int
3203 nvme_fc_parse_traddr(struct nvmet_fc_traddr *traddr, char *buf, size_t blen)
3204 {
3205 	char name[2 + NVME_FC_TRADDR_HEXNAMELEN + 1];
3206 	substring_t wwn = { name, &name[sizeof(name)-1] };
3207 	int nnoffset, pnoffset;
3208 
3209 	/* validate it string one of the 2 allowed formats */
3210 	if (strnlen(buf, blen) == NVME_FC_TRADDR_MAXLENGTH &&
3211 			!strncmp(buf, "nn-0x", NVME_FC_TRADDR_OXNNLEN) &&
3212 			!strncmp(&buf[NVME_FC_TRADDR_MAX_PN_OFFSET],
3213 				"pn-0x", NVME_FC_TRADDR_OXNNLEN)) {
3214 		nnoffset = NVME_FC_TRADDR_OXNNLEN;
3215 		pnoffset = NVME_FC_TRADDR_MAX_PN_OFFSET +
3216 						NVME_FC_TRADDR_OXNNLEN;
3217 	} else if ((strnlen(buf, blen) == NVME_FC_TRADDR_MINLENGTH &&
3218 			!strncmp(buf, "nn-", NVME_FC_TRADDR_NNLEN) &&
3219 			!strncmp(&buf[NVME_FC_TRADDR_MIN_PN_OFFSET],
3220 				"pn-", NVME_FC_TRADDR_NNLEN))) {
3221 		nnoffset = NVME_FC_TRADDR_NNLEN;
3222 		pnoffset = NVME_FC_TRADDR_MIN_PN_OFFSET + NVME_FC_TRADDR_NNLEN;
3223 	} else
3224 		goto out_einval;
3225 
3226 	name[0] = '0';
3227 	name[1] = 'x';
3228 	name[2 + NVME_FC_TRADDR_HEXNAMELEN] = 0;
3229 
3230 	memcpy(&name[2], &buf[nnoffset], NVME_FC_TRADDR_HEXNAMELEN);
3231 	if (__nvme_fc_parse_u64(&wwn, &traddr->nn))
3232 		goto out_einval;
3233 
3234 	memcpy(&name[2], &buf[pnoffset], NVME_FC_TRADDR_HEXNAMELEN);
3235 	if (__nvme_fc_parse_u64(&wwn, &traddr->pn))
3236 		goto out_einval;
3237 
3238 	return 0;
3239 
3240 out_einval:
3241 	pr_warn("%s: bad traddr string\n", __func__);
3242 	return -EINVAL;
3243 }
3244 
3245 static struct nvme_ctrl *
3246 nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts)
3247 {
3248 	struct nvme_fc_lport *lport;
3249 	struct nvme_fc_rport *rport;
3250 	struct nvme_ctrl *ctrl;
3251 	struct nvmet_fc_traddr laddr = { 0L, 0L };
3252 	struct nvmet_fc_traddr raddr = { 0L, 0L };
3253 	unsigned long flags;
3254 	int ret;
3255 
3256 	ret = nvme_fc_parse_traddr(&raddr, opts->traddr, NVMF_TRADDR_SIZE);
3257 	if (ret || !raddr.nn || !raddr.pn)
3258 		return ERR_PTR(-EINVAL);
3259 
3260 	ret = nvme_fc_parse_traddr(&laddr, opts->host_traddr, NVMF_TRADDR_SIZE);
3261 	if (ret || !laddr.nn || !laddr.pn)
3262 		return ERR_PTR(-EINVAL);
3263 
3264 	/* find the host and remote ports to connect together */
3265 	spin_lock_irqsave(&nvme_fc_lock, flags);
3266 	list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
3267 		if (lport->localport.node_name != laddr.nn ||
3268 		    lport->localport.port_name != laddr.pn)
3269 			continue;
3270 
3271 		list_for_each_entry(rport, &lport->endp_list, endp_list) {
3272 			if (rport->remoteport.node_name != raddr.nn ||
3273 			    rport->remoteport.port_name != raddr.pn)
3274 				continue;
3275 
3276 			/* if fail to get reference fall through. Will error */
3277 			if (!nvme_fc_rport_get(rport))
3278 				break;
3279 
3280 			spin_unlock_irqrestore(&nvme_fc_lock, flags);
3281 
3282 			ctrl = nvme_fc_init_ctrl(dev, opts, lport, rport);
3283 			if (IS_ERR(ctrl))
3284 				nvme_fc_rport_put(rport);
3285 			return ctrl;
3286 		}
3287 	}
3288 	spin_unlock_irqrestore(&nvme_fc_lock, flags);
3289 
3290 	return ERR_PTR(-ENOENT);
3291 }
3292 
3293 
3294 static struct nvmf_transport_ops nvme_fc_transport = {
3295 	.name		= "fc",
3296 	.module		= THIS_MODULE,
3297 	.required_opts	= NVMF_OPT_TRADDR | NVMF_OPT_HOST_TRADDR,
3298 	.allowed_opts	= NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_CTRL_LOSS_TMO,
3299 	.create_ctrl	= nvme_fc_create_ctrl,
3300 };
3301 
3302 static int __init nvme_fc_init_module(void)
3303 {
3304 	int ret;
3305 
3306 	/*
3307 	 * NOTE:
3308 	 * It is expected that in the future the kernel will combine
3309 	 * the FC-isms that are currently under scsi and now being
3310 	 * added to by NVME into a new standalone FC class. The SCSI
3311 	 * and NVME protocols and their devices would be under this
3312 	 * new FC class.
3313 	 *
3314 	 * As we need something to post FC-specific udev events to,
3315 	 * specifically for nvme probe events, start by creating the
3316 	 * new device class.  When the new standalone FC class is
3317 	 * put in place, this code will move to a more generic
3318 	 * location for the class.
3319 	 */
3320 	fc_class = class_create(THIS_MODULE, "fc");
3321 	if (IS_ERR(fc_class)) {
3322 		pr_err("couldn't register class fc\n");
3323 		return PTR_ERR(fc_class);
3324 	}
3325 
3326 	/*
3327 	 * Create a device for the FC-centric udev events
3328 	 */
3329 	fc_udev_device = device_create(fc_class, NULL, MKDEV(0, 0), NULL,
3330 				"fc_udev_device");
3331 	if (IS_ERR(fc_udev_device)) {
3332 		pr_err("couldn't create fc_udev device!\n");
3333 		ret = PTR_ERR(fc_udev_device);
3334 		goto out_destroy_class;
3335 	}
3336 
3337 	ret = nvmf_register_transport(&nvme_fc_transport);
3338 	if (ret)
3339 		goto out_destroy_device;
3340 
3341 	return 0;
3342 
3343 out_destroy_device:
3344 	device_destroy(fc_class, MKDEV(0, 0));
3345 out_destroy_class:
3346 	class_destroy(fc_class);
3347 	return ret;
3348 }
3349 
3350 static void __exit nvme_fc_exit_module(void)
3351 {
3352 	/* sanity check - all lports should be removed */
3353 	if (!list_empty(&nvme_fc_lport_list))
3354 		pr_warn("%s: localport list not empty\n", __func__);
3355 
3356 	nvmf_unregister_transport(&nvme_fc_transport);
3357 
3358 	ida_destroy(&nvme_fc_local_port_cnt);
3359 	ida_destroy(&nvme_fc_ctrl_cnt);
3360 
3361 	device_destroy(fc_class, MKDEV(0, 0));
3362 	class_destroy(fc_class);
3363 }
3364 
3365 module_init(nvme_fc_init_module);
3366 module_exit(nvme_fc_exit_module);
3367 
3368 MODULE_LICENSE("GPL v2");
3369