projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
nvme-fc: create fc class and transport device
[linux-2.6-block.git] / drivers / nvme / host / fc.c
CommitLineData
e399441d
JS		 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>
61bff8ef 22#include <linux/delay.h>
e399441d
JS		 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/*
34 * We handle AEN commands ourselves and don't even let the
35 * block layer know about them.
36 */
37#define NVME_FC_NR_AEN_COMMANDS 1
38#define NVME_FC_AQ_BLKMQ_DEPTH \
7aa1f427 39 (NVME_AQ_DEPTH - NVME_FC_NR_AEN_COMMANDS)
e399441d
JS		 40#define AEN_CMDID_BASE (NVME_FC_AQ_BLKMQ_DEPTH + 1)
41
42enum nvme_fc_queue_flags {
43 NVME_FC_Q_CONNECTED = (1 << 0),
44};
45
46#define NVMEFC_QUEUE_DELAY 3 /* ms units */
47
48struct nvme_fc_queue {
49 struct nvme_fc_ctrl *ctrl;
50 struct device *dev;
51 struct blk_mq_hw_ctx *hctx;
52 void *lldd_handle;
53 int queue_size;
54 size_t cmnd_capsule_len;
55 u32 qnum;
56 u32 rqcnt;
57 u32 seqno;
58
59 u64 connection_id;
60 atomic_t csn;
61
62 unsigned long flags;
63} __aligned(sizeof(u64)); /* alignment for other things alloc'd with */
64
8d64daf7
JS		 65enum nvme_fcop_flags {
66 FCOP_FLAGS_TERMIO = (1 << 0),
67 FCOP_FLAGS_RELEASED = (1 << 1),
68 FCOP_FLAGS_COMPLETE = (1 << 2),
78a7ac26 69 FCOP_FLAGS_AEN = (1 << 3),
8d64daf7
JS		 70};
71
e399441d
JS		 72struct nvmefc_ls_req_op {
73 struct nvmefc_ls_req ls_req;
74
c913a8b0 75 struct nvme_fc_rport *rport;
e399441d
JS		 76 struct nvme_fc_queue *queue;
77 struct request *rq;
8d64daf7 78 u32 flags;
e399441d
JS		 79
80 int ls_error;
81 struct completion ls_done;
c913a8b0 82 struct list_head lsreq_list; /* rport->ls_req_list */
e399441d
JS		 83 bool req_queued;
84};
85
86enum nvme_fcpop_state {
87 FCPOP_STATE_UNINIT = 0,
88 FCPOP_STATE_IDLE = 1,
89 FCPOP_STATE_ACTIVE = 2,
90 FCPOP_STATE_ABORTED = 3,
78a7ac26 91 FCPOP_STATE_COMPLETE = 4,
e399441d
JS		 92};
93
94struct nvme_fc_fcp_op {
95 struct nvme_request nreq; /*
96 * nvme/host/core.c
97 * requires this to be
98 * the 1st element in the
99 * private structure
100 * associated with the
101 * request.
102 */
103 struct nvmefc_fcp_req fcp_req;
104
105 struct nvme_fc_ctrl *ctrl;
106 struct nvme_fc_queue *queue;
107 struct request *rq;
108
109 atomic_t state;
78a7ac26 110 u32 flags;
e399441d
JS		 111 u32 rqno;
112 u32 nents;
113
114 struct nvme_fc_cmd_iu cmd_iu;
115 struct nvme_fc_ersp_iu rsp_iu;
116};
117
118struct nvme_fc_lport {
119 struct nvme_fc_local_port localport;
120
121 struct ida endp_cnt;
122 struct list_head port_list; /* nvme_fc_port_list */
123 struct list_head endp_list;
124 struct device *dev; /* physical device for dma */
125 struct nvme_fc_port_template *ops;
126 struct kref ref;
127} __aligned(sizeof(u64)); /* alignment for other things alloc'd with */
128
129struct nvme_fc_rport {
130 struct nvme_fc_remote_port remoteport;
131
132 struct list_head endp_list; /* for lport->endp_list */
133 struct list_head ctrl_list;
c913a8b0
JS		 134 struct list_head ls_req_list;
135 struct device *dev; /* physical device for dma */
136 struct nvme_fc_lport *lport;
e399441d
JS		 137 spinlock_t lock;
138 struct kref ref;
139} __aligned(sizeof(u64)); /* alignment for other things alloc'd with */
140
61bff8ef
JS		 141enum nvme_fcctrl_flags {
142 FCCTRL_TERMIO = (1 << 0),
e399441d
JS		 143};
144
145struct nvme_fc_ctrl {
146 spinlock_t lock;
147 struct nvme_fc_queue *queues;
e399441d
JS		 148 struct device *dev;
149 struct nvme_fc_lport *lport;
150 struct nvme_fc_rport *rport;
151 u32 cnum;
152
153 u64 association_id;
154
e399441d 155 struct list_head ctrl_list; /* rport->ctrl_list */
e399441d
JS		 156
157 struct blk_mq_tag_set admin_tag_set;
158 struct blk_mq_tag_set tag_set;
159
160 struct work_struct delete_work;
61bff8ef 161 struct delayed_work connect_work;
61bff8ef 162
e399441d 163 struct kref ref;
61bff8ef
JS		 164 u32 flags;
165 u32 iocnt;
36715cf4 166 wait_queue_head_t ioabort_wait;
e399441d
JS		 167
168 struct nvme_fc_fcp_op aen_ops[NVME_FC_NR_AEN_COMMANDS];
169
170 struct nvme_ctrl ctrl;
171};
172
173static inline struct nvme_fc_ctrl *
174to_fc_ctrl(struct nvme_ctrl *ctrl)
175{
176 return container_of(ctrl, struct nvme_fc_ctrl, ctrl);
177}
178
179static inline struct nvme_fc_lport *
180localport_to_lport(struct nvme_fc_local_port *portptr)
181{
182 return container_of(portptr, struct nvme_fc_lport, localport);
183}
184
185static inline struct nvme_fc_rport *
186remoteport_to_rport(struct nvme_fc_remote_port *portptr)
187{
188 return container_of(portptr, struct nvme_fc_rport, remoteport);
189}
190
191static inline struct nvmefc_ls_req_op *
192ls_req_to_lsop(struct nvmefc_ls_req *lsreq)
193{
194 return container_of(lsreq, struct nvmefc_ls_req_op, ls_req);
195}
196
197static inline struct nvme_fc_fcp_op *
198fcp_req_to_fcp_op(struct nvmefc_fcp_req *fcpreq)
199{
200 return container_of(fcpreq, struct nvme_fc_fcp_op, fcp_req);
201}
202
203
204
205/* *************************** Globals **************************** */
206
207
208static DEFINE_SPINLOCK(nvme_fc_lock);
209
210static LIST_HEAD(nvme_fc_lport_list);
211static DEFINE_IDA(nvme_fc_local_port_cnt);
212static DEFINE_IDA(nvme_fc_ctrl_cnt);
213
e399441d
JS		 214
215
5f568556
JS		 216/*
217 * These items are short-term. They will eventually be moved into
218 * a generic FC class. See comments in module init.
219 */
220static struct class *fc_class;
221static struct device *fc_udev_device;
222
e399441d
JS		 223
224/* *********************** FC-NVME Port Management ************************ */
225
226static int __nvme_fc_del_ctrl(struct nvme_fc_ctrl *);
227static void __nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *,
228 struct nvme_fc_queue *, unsigned int);
229
5533d424
JS		 230static void
231nvme_fc_free_lport(struct kref *ref)
232{
233 struct nvme_fc_lport *lport =
234 container_of(ref, struct nvme_fc_lport, ref);
235 unsigned long flags;
236
237 WARN_ON(lport->localport.port_state != FC_OBJSTATE_DELETED);
238 WARN_ON(!list_empty(&lport->endp_list));
239
240 /* remove from transport list */
241 spin_lock_irqsave(&nvme_fc_lock, flags);
242 list_del(&lport->port_list);
243 spin_unlock_irqrestore(&nvme_fc_lock, flags);
244
245 /* let the LLDD know we've finished tearing it down */
246 lport->ops->localport_delete(&lport->localport);
247
248 ida_simple_remove(&nvme_fc_local_port_cnt, lport->localport.port_num);
249 ida_destroy(&lport->endp_cnt);
250
251 put_device(lport->dev);
252
253 kfree(lport);
254}
255
256static void
257nvme_fc_lport_put(struct nvme_fc_lport *lport)
258{
259 kref_put(&lport->ref, nvme_fc_free_lport);
260}
261
262static int
263nvme_fc_lport_get(struct nvme_fc_lport *lport)
264{
265 return kref_get_unless_zero(&lport->ref);
266}
267
268
269static struct nvme_fc_lport *
270nvme_fc_attach_to_unreg_lport(struct nvme_fc_port_info *pinfo)
271{
272 struct nvme_fc_lport *lport;
273 unsigned long flags;
274
275 spin_lock_irqsave(&nvme_fc_lock, flags);
276
277 list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
278 if (lport->localport.node_name != pinfo->node_name ||
279 lport->localport.port_name != pinfo->port_name)
280 continue;
281
282 if (lport->localport.port_state != FC_OBJSTATE_DELETED) {
283 lport = ERR_PTR(-EEXIST);
284 goto out_done;
285 }
286
287 if (!nvme_fc_lport_get(lport)) {
288 /*
289 * fails if ref cnt already 0. If so,
290 * act as if lport already deleted
291 */
292 lport = NULL;
293 goto out_done;
294 }
295
296 /* resume the lport */
297
298 lport->localport.port_role = pinfo->port_role;
299 lport->localport.port_id = pinfo->port_id;
300 lport->localport.port_state = FC_OBJSTATE_ONLINE;
301
302 spin_unlock_irqrestore(&nvme_fc_lock, flags);
303
304 return lport;
305 }
306
307 lport = NULL;
308
309out_done:
310 spin_unlock_irqrestore(&nvme_fc_lock, flags);
311
312 return lport;
313}
e399441d
JS		 314
315/**
316 * nvme_fc_register_localport - transport entry point called by an
317 * LLDD to register the existence of a NVME
318 * host FC port.
319 * @pinfo: pointer to information about the port to be registered
320 * @template: LLDD entrypoints and operational parameters for the port
321 * @dev: physical hardware device node port corresponds to. Will be
322 * used for DMA mappings
323 * @lport_p: pointer to a local port pointer. Upon success, the routine
324 * will allocate a nvme_fc_local_port structure and place its
325 * address in the local port pointer. Upon failure, local port
326 * pointer will be set to 0.
327 *
328 * Returns:
329 * a completion status. Must be 0 upon success; a negative errno
330 * (ex: -ENXIO) upon failure.
331 */
332int
333nvme_fc_register_localport(struct nvme_fc_port_info *pinfo,
334 struct nvme_fc_port_template *template,
335 struct device *dev,
336 struct nvme_fc_local_port **portptr)
337{
338 struct nvme_fc_lport *newrec;
339 unsigned long flags;
340 int ret, idx;
341
342 if (!template->localport_delete || !template->remoteport_delete ||
343 !template->ls_req || !template->fcp_io ||
344 !template->ls_abort || !template->fcp_abort ||
345 !template->max_hw_queues || !template->max_sgl_segments ||
346 !template->max_dif_sgl_segments || !template->dma_boundary) {
347 ret = -EINVAL;
348 goto out_reghost_failed;
349 }
350
5533d424
JS		 351 /*
352 * look to see if there is already a localport that had been
353 * deregistered and in the process of waiting for all the
354 * references to fully be removed. If the references haven't
355 * expired, we can simply re-enable the localport. Remoteports
356 * and controller reconnections should resume naturally.
357 */
358 newrec = nvme_fc_attach_to_unreg_lport(pinfo);
359
360 /* found an lport, but something about its state is bad */
361 if (IS_ERR(newrec)) {
362 ret = PTR_ERR(newrec);
363 goto out_reghost_failed;
364
365 /* found existing lport, which was resumed */
366 } else if (newrec) {
367 *portptr = &newrec->localport;
368 return 0;
369 }
370
371 /* nothing found - allocate a new localport struct */
372
e399441d
JS		 373 newrec = kmalloc((sizeof(*newrec) + template->local_priv_sz),
374 GFP_KERNEL);
375 if (!newrec) {
376 ret = -ENOMEM;
377 goto out_reghost_failed;
378 }
379
380 idx = ida_simple_get(&nvme_fc_local_port_cnt, 0, 0, GFP_KERNEL);
381 if (idx < 0) {
382 ret = -ENOSPC;
383 goto out_fail_kfree;
384 }
385
386 if (!get_device(dev) && dev) {
387 ret = -ENODEV;
388 goto out_ida_put;
389 }
390
391 INIT_LIST_HEAD(&newrec->port_list);
392 INIT_LIST_HEAD(&newrec->endp_list);
393 kref_init(&newrec->ref);
394 newrec->ops = template;
395 newrec->dev = dev;
396 ida_init(&newrec->endp_cnt);
397 newrec->localport.private = &newrec[1];
398 newrec->localport.node_name = pinfo->node_name;
399 newrec->localport.port_name = pinfo->port_name;
400 newrec->localport.port_role = pinfo->port_role;
401 newrec->localport.port_id = pinfo->port_id;
402 newrec->localport.port_state = FC_OBJSTATE_ONLINE;
403 newrec->localport.port_num = idx;
404
405 spin_lock_irqsave(&nvme_fc_lock, flags);
406 list_add_tail(&newrec->port_list, &nvme_fc_lport_list);
407 spin_unlock_irqrestore(&nvme_fc_lock, flags);
408
409 if (dev)
410 dma_set_seg_boundary(dev, template->dma_boundary);
411
412 *portptr = &newrec->localport;
413 return 0;
414
415out_ida_put:
416 ida_simple_remove(&nvme_fc_local_port_cnt, idx);
417out_fail_kfree:
418 kfree(newrec);
419out_reghost_failed:
420 *portptr = NULL;
421
422 return ret;
423}
424EXPORT_SYMBOL_GPL(nvme_fc_register_localport);
425
e399441d
JS		 426/**
427 * nvme_fc_unregister_localport - transport entry point called by an
428 * LLDD to deregister/remove a previously
429 * registered a NVME host FC port.
430 * @localport: pointer to the (registered) local port that is to be
431 * deregistered.
432 *
433 * Returns:
434 * a completion status. Must be 0 upon success; a negative errno
435 * (ex: -ENXIO) upon failure.
436 */
437int
438nvme_fc_unregister_localport(struct nvme_fc_local_port *portptr)
439{
440 struct nvme_fc_lport *lport = localport_to_lport(portptr);
441 unsigned long flags;
442
443 if (!portptr)
444 return -EINVAL;
445
446 spin_lock_irqsave(&nvme_fc_lock, flags);
447
448 if (portptr->port_state != FC_OBJSTATE_ONLINE) {
449 spin_unlock_irqrestore(&nvme_fc_lock, flags);
450 return -EINVAL;
451 }
452 portptr->port_state = FC_OBJSTATE_DELETED;
453
454 spin_unlock_irqrestore(&nvme_fc_lock, flags);
455
456 nvme_fc_lport_put(lport);
457
458 return 0;
459}
460EXPORT_SYMBOL_GPL(nvme_fc_unregister_localport);
461
eaefd5ab
JS		 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
472static void
473nvme_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
e399441d
JS		 492/**
493 * nvme_fc_register_remoteport - transport entry point called by an
494 * LLDD to register the existence of a NVME
495 * subsystem FC port on its fabric.
496 * @localport: pointer to the (registered) local port that the remote
497 * subsystem port is connected to.
498 * @pinfo: pointer to information about the port to be registered
499 * @rport_p: pointer to a remote port pointer. Upon success, the routine
500 * will allocate a nvme_fc_remote_port structure and place its
501 * address in the remote port pointer. Upon failure, remote port
502 * pointer will be set to 0.
503 *
504 * Returns:
505 * a completion status. Must be 0 upon success; a negative errno
506 * (ex: -ENXIO) upon failure.
507 */
508int
509nvme_fc_register_remoteport(struct nvme_fc_local_port *localport,
510 struct nvme_fc_port_info *pinfo,
511 struct nvme_fc_remote_port **portptr)
512{
513 struct nvme_fc_lport *lport = localport_to_lport(localport);
514 struct nvme_fc_rport *newrec;
515 unsigned long flags;
516 int ret, idx;
517
518 newrec = kmalloc((sizeof(*newrec) + lport->ops->remote_priv_sz),
519 GFP_KERNEL);
520 if (!newrec) {
521 ret = -ENOMEM;
522 goto out_reghost_failed;
523 }
524
525 if (!nvme_fc_lport_get(lport)) {
526 ret = -ESHUTDOWN;
527 goto out_kfree_rport;
528 }
529
530 idx = ida_simple_get(&lport->endp_cnt, 0, 0, GFP_KERNEL);
531 if (idx < 0) {
532 ret = -ENOSPC;
533 goto out_lport_put;
534 }
535
536 INIT_LIST_HEAD(&newrec->endp_list);
537 INIT_LIST_HEAD(&newrec->ctrl_list);
c913a8b0 538 INIT_LIST_HEAD(&newrec->ls_req_list);
e399441d
JS		 539 kref_init(&newrec->ref);
540 spin_lock_init(&newrec->lock);
541 newrec->remoteport.localport = &lport->localport;
c913a8b0
JS		 542 newrec->dev = lport->dev;
543 newrec->lport = lport;
e399441d
JS		 544 newrec->remoteport.private = &newrec[1];
545 newrec->remoteport.port_role = pinfo->port_role;
546 newrec->remoteport.node_name = pinfo->node_name;
547 newrec->remoteport.port_name = pinfo->port_name;
548 newrec->remoteport.port_id = pinfo->port_id;
549 newrec->remoteport.port_state = FC_OBJSTATE_ONLINE;
550 newrec->remoteport.port_num = idx;
551
552 spin_lock_irqsave(&nvme_fc_lock, flags);
553 list_add_tail(&newrec->endp_list, &lport->endp_list);
554 spin_unlock_irqrestore(&nvme_fc_lock, flags);
555
eaefd5ab
JS		 556 nvme_fc_signal_discovery_scan(lport, newrec);
557
e399441d
JS		 558 *portptr = &newrec->remoteport;
559 return 0;
560
561out_lport_put:
562 nvme_fc_lport_put(lport);
563out_kfree_rport:
564 kfree(newrec);
565out_reghost_failed:
566 *portptr = NULL;
567 return ret;
e399441d
JS		 568}
569EXPORT_SYMBOL_GPL(nvme_fc_register_remoteport);
570
571static void
572nvme_fc_free_rport(struct kref *ref)
573{
574 struct nvme_fc_rport *rport =
575 container_of(ref, struct nvme_fc_rport, ref);
576 struct nvme_fc_lport *lport =
577 localport_to_lport(rport->remoteport.localport);
578 unsigned long flags;
579
580 WARN_ON(rport->remoteport.port_state != FC_OBJSTATE_DELETED);
581 WARN_ON(!list_empty(&rport->ctrl_list));
582
583 /* remove from lport list */
584 spin_lock_irqsave(&nvme_fc_lock, flags);
585 list_del(&rport->endp_list);
586 spin_unlock_irqrestore(&nvme_fc_lock, flags);
587
588 /* let the LLDD know we've finished tearing it down */
589 lport->ops->remoteport_delete(&rport->remoteport);
590
591 ida_simple_remove(&lport->endp_cnt, rport->remoteport.port_num);
592
593 kfree(rport);
594
595 nvme_fc_lport_put(lport);
596}
597
598static void
599nvme_fc_rport_put(struct nvme_fc_rport *rport)
600{
601 kref_put(&rport->ref, nvme_fc_free_rport);
602}
603
604static int
605nvme_fc_rport_get(struct nvme_fc_rport *rport)
606{
607 return kref_get_unless_zero(&rport->ref);
608}
609
8d64daf7
JS		 610static int
611nvme_fc_abort_lsops(struct nvme_fc_rport *rport)
612{
613 struct nvmefc_ls_req_op *lsop;
614 unsigned long flags;
615
616restart:
617 spin_lock_irqsave(&rport->lock, flags);
618
619 list_for_each_entry(lsop, &rport->ls_req_list, lsreq_list) {
620 if (!(lsop->flags & FCOP_FLAGS_TERMIO)) {
621 lsop->flags |= FCOP_FLAGS_TERMIO;
622 spin_unlock_irqrestore(&rport->lock, flags);
623 rport->lport->ops->ls_abort(&rport->lport->localport,
624 &rport->remoteport,
625 &lsop->ls_req);
626 goto restart;
627 }
628 }
629 spin_unlock_irqrestore(&rport->lock, flags);
630
631 return 0;
632}
633
e399441d
JS		 634/**
635 * nvme_fc_unregister_remoteport - transport entry point called by an
636 * LLDD to deregister/remove a previously
637 * registered a NVME subsystem FC port.
638 * @remoteport: pointer to the (registered) remote port that is to be
639 * deregistered.
640 *
641 * Returns:
642 * a completion status. Must be 0 upon success; a negative errno
643 * (ex: -ENXIO) upon failure.
644 */
645int
646nvme_fc_unregister_remoteport(struct nvme_fc_remote_port *portptr)
647{
648 struct nvme_fc_rport *rport = remoteport_to_rport(portptr);
649 struct nvme_fc_ctrl *ctrl;
650 unsigned long flags;
651
652 if (!portptr)
653 return -EINVAL;
654
655 spin_lock_irqsave(&rport->lock, flags);
656
657 if (portptr->port_state != FC_OBJSTATE_ONLINE) {
658 spin_unlock_irqrestore(&rport->lock, flags);
659 return -EINVAL;
660 }
661 portptr->port_state = FC_OBJSTATE_DELETED;
662
663 /* tear down all associations to the remote port */
664 list_for_each_entry(ctrl, &rport->ctrl_list, ctrl_list)
665 __nvme_fc_del_ctrl(ctrl);
666
667 spin_unlock_irqrestore(&rport->lock, flags);
668
8d64daf7
JS		 669 nvme_fc_abort_lsops(rport);
670
e399441d
JS		 671 nvme_fc_rport_put(rport);
672 return 0;
673}
674EXPORT_SYMBOL_GPL(nvme_fc_unregister_remoteport);
675
eaefd5ab
JS		 676/**
677 * nvme_fc_rescan_remoteport - transport entry point called by an
678 * LLDD to request a nvme device rescan.
679 * @remoteport: pointer to the (registered) remote port that is to be
680 * rescanned.
681 *
682 * Returns: N/A
683 */
684void
685nvme_fc_rescan_remoteport(struct nvme_fc_remote_port *remoteport)
686{
687 struct nvme_fc_rport *rport = remoteport_to_rport(remoteport);
688
689 nvme_fc_signal_discovery_scan(rport->lport, rport);
690}
691EXPORT_SYMBOL_GPL(nvme_fc_rescan_remoteport);
692
e399441d
JS		 693
694/* *********************** FC-NVME DMA Handling **************************** */
695
696/*
697 * The fcloop device passes in a NULL device pointer. Real LLD's will
698 * pass in a valid device pointer. If NULL is passed to the dma mapping
699 * routines, depending on the platform, it may or may not succeed, and
700 * may crash.
701 *
702 * As such:
703 * Wrapper all the dma routines and check the dev pointer.
704 *
705 * If simple mappings (return just a dma address, we'll noop them,
706 * returning a dma address of 0.
707 *
708 * On more complex mappings (dma_map_sg), a pseudo routine fills
709 * in the scatter list, setting all dma addresses to 0.
710 */
711
712static inline dma_addr_t
713fc_dma_map_single(struct device *dev, void *ptr, size_t size,
714 enum dma_data_direction dir)
715{
716 return dev ? dma_map_single(dev, ptr, size, dir) : (dma_addr_t)0L;
717}
718
719static inline int
720fc_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
721{
722 return dev ? dma_mapping_error(dev, dma_addr) : 0;
723}
724
725static inline void
726fc_dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
727 enum dma_data_direction dir)
728{
729 if (dev)
730 dma_unmap_single(dev, addr, size, dir);
731}
732
733static inline void
734fc_dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr, size_t size,
735 enum dma_data_direction dir)
736{
737 if (dev)
738 dma_sync_single_for_cpu(dev, addr, size, dir);
739}
740
741static inline void
742fc_dma_sync_single_for_device(struct device *dev, dma_addr_t addr, size_t size,
743 enum dma_data_direction dir)
744{
745 if (dev)
746 dma_sync_single_for_device(dev, addr, size, dir);
747}
748
749/* pseudo dma_map_sg call */
750static int
751fc_map_sg(struct scatterlist *sg, int nents)
752{
753 struct scatterlist *s;
754 int i;
755
756 WARN_ON(nents == 0 || sg[0].length == 0);
757
758 for_each_sg(sg, s, nents, i) {
759 s->dma_address = 0L;
760#ifdef CONFIG_NEED_SG_DMA_LENGTH
761 s->dma_length = s->length;
762#endif
763 }
764 return nents;
765}
766
767static inline int
768fc_dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
769 enum dma_data_direction dir)
770{
771 return dev ? dma_map_sg(dev, sg, nents, dir) : fc_map_sg(sg, nents);
772}
773
774static inline void
775fc_dma_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
776 enum dma_data_direction dir)
777{
778 if (dev)
779 dma_unmap_sg(dev, sg, nents, dir);
780}
781
782
783/* *********************** FC-NVME LS Handling **************************** */
784
785static void nvme_fc_ctrl_put(struct nvme_fc_ctrl *);
786static int nvme_fc_ctrl_get(struct nvme_fc_ctrl *);
787
788
789static void
c913a8b0 790__nvme_fc_finish_ls_req(struct nvmefc_ls_req_op *lsop)
e399441d 791{
c913a8b0 792 struct nvme_fc_rport *rport = lsop->rport;
e399441d
JS		 793 struct nvmefc_ls_req *lsreq = &lsop->ls_req;
794 unsigned long flags;
795
c913a8b0 796 spin_lock_irqsave(&rport->lock, flags);
e399441d
JS		 797
798 if (!lsop->req_queued) {
c913a8b0 799 spin_unlock_irqrestore(&rport->lock, flags);
e399441d
JS		 800 return;
801 }
802
803 list_del(&lsop->lsreq_list);
804
805 lsop->req_queued = false;
806
c913a8b0 807 spin_unlock_irqrestore(&rport->lock, flags);
e399441d 808
c913a8b0 809 fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
e399441d
JS		 810 (lsreq->rqstlen + lsreq->rsplen),
811 DMA_BIDIRECTIONAL);
812
c913a8b0 813 nvme_fc_rport_put(rport);
e399441d
JS		 814}
815
816static int
c913a8b0 817__nvme_fc_send_ls_req(struct nvme_fc_rport *rport,
e399441d
JS		 818 struct nvmefc_ls_req_op *lsop,
819 void (*done)(struct nvmefc_ls_req *req, int status))
820{
821 struct nvmefc_ls_req *lsreq = &lsop->ls_req;
822 unsigned long flags;
c913a8b0 823 int ret = 0;
e399441d 824
c913a8b0
JS		 825 if (rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
826 return -ECONNREFUSED;
827
828 if (!nvme_fc_rport_get(rport))
e399441d
JS		 829 return -ESHUTDOWN;
830
831 lsreq->done = done;
c913a8b0 832 lsop->rport = rport;
e399441d
JS		 833 lsop->req_queued = false;
834 INIT_LIST_HEAD(&lsop->lsreq_list);
835 init_completion(&lsop->ls_done);
836
c913a8b0 837 lsreq->rqstdma = fc_dma_map_single(rport->dev, lsreq->rqstaddr,
e399441d
JS		 838 lsreq->rqstlen + lsreq->rsplen,
839 DMA_BIDIRECTIONAL);
c913a8b0
JS		 840 if (fc_dma_mapping_error(rport->dev, lsreq->rqstdma)) {
841 ret = -EFAULT;
842 goto out_putrport;
e399441d
JS		 843 }
844 lsreq->rspdma = lsreq->rqstdma + lsreq->rqstlen;
845
c913a8b0 846 spin_lock_irqsave(&rport->lock, flags);
e399441d 847
c913a8b0 848 list_add_tail(&lsop->lsreq_list, &rport->ls_req_list);
e399441d
JS		 849
850 lsop->req_queued = true;
851
c913a8b0 852 spin_unlock_irqrestore(&rport->lock, flags);
e399441d 853
c913a8b0
JS		 854 ret = rport->lport->ops->ls_req(&rport->lport->localport,
855 &rport->remoteport, lsreq);
e399441d 856 if (ret)
c913a8b0
JS		 857 goto out_unlink;
858
859 return 0;
860
861out_unlink:
862 lsop->ls_error = ret;
863 spin_lock_irqsave(&rport->lock, flags);
864 lsop->req_queued = false;
865 list_del(&lsop->lsreq_list);
866 spin_unlock_irqrestore(&rport->lock, flags);
867 fc_dma_unmap_single(rport->dev, lsreq->rqstdma,
868 (lsreq->rqstlen + lsreq->rsplen),
869 DMA_BIDIRECTIONAL);
870out_putrport:
871 nvme_fc_rport_put(rport);
e399441d
JS		 872
873 return ret;
874}
875
876static void
877nvme_fc_send_ls_req_done(struct nvmefc_ls_req *lsreq, int status)
878{
879 struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
880
881 lsop->ls_error = status;
882 complete(&lsop->ls_done);
883}
884
885static int
c913a8b0 886nvme_fc_send_ls_req(struct nvme_fc_rport *rport, struct nvmefc_ls_req_op *lsop)
e399441d
JS		 887{
888 struct nvmefc_ls_req *lsreq = &lsop->ls_req;
889 struct fcnvme_ls_rjt *rjt = lsreq->rspaddr;
890 int ret;
891
c913a8b0 892 ret = __nvme_fc_send_ls_req(rport, lsop, nvme_fc_send_ls_req_done);
e399441d 893
c913a8b0 894 if (!ret) {
e399441d
JS		 895 /*
896 * No timeout/not interruptible as we need the struct
897 * to exist until the lldd calls us back. Thus mandate
898 * wait until driver calls back. lldd responsible for
899 * the timeout action
900 */
901 wait_for_completion(&lsop->ls_done);
902
c913a8b0 903 __nvme_fc_finish_ls_req(lsop);
e399441d 904
c913a8b0 905 ret = lsop->ls_error;
e399441d
JS		 906 }
907
c913a8b0
JS		 908 if (ret)
909 return ret;
910
e399441d
JS		 911 /* ACC or RJT payload ? */
912 if (rjt->w0.ls_cmd == FCNVME_LS_RJT)
913 return -ENXIO;
914
915 return 0;
916}
917
c913a8b0
JS		 918static int
919nvme_fc_send_ls_req_async(struct nvme_fc_rport *rport,
e399441d
JS		 920 struct nvmefc_ls_req_op *lsop,
921 void (*done)(struct nvmefc_ls_req *req, int status))
922{
e399441d
JS		 923 /* don't wait for completion */
924
c913a8b0 925 return __nvme_fc_send_ls_req(rport, lsop, done);
e399441d
JS		 926}
927
928/* Validation Error indexes into the string table below */
929enum {
930 VERR_NO_ERROR = 0,
931 VERR_LSACC = 1,
932 VERR_LSDESC_RQST = 2,
933 VERR_LSDESC_RQST_LEN = 3,
934 VERR_ASSOC_ID = 4,
935 VERR_ASSOC_ID_LEN = 5,
936 VERR_CONN_ID = 6,
937 VERR_CONN_ID_LEN = 7,
938 VERR_CR_ASSOC = 8,
939 VERR_CR_ASSOC_ACC_LEN = 9,
940 VERR_CR_CONN = 10,
941 VERR_CR_CONN_ACC_LEN = 11,
942 VERR_DISCONN = 12,
943 VERR_DISCONN_ACC_LEN = 13,
944};
945
946static char *validation_errors[] = {
947 "OK",
948 "Not LS_ACC",
949 "Not LSDESC_RQST",
950 "Bad LSDESC_RQST Length",
951 "Not Association ID",
952 "Bad Association ID Length",
953 "Not Connection ID",
954 "Bad Connection ID Length",
955 "Not CR_ASSOC Rqst",
956 "Bad CR_ASSOC ACC Length",
957 "Not CR_CONN Rqst",
958 "Bad CR_CONN ACC Length",
959 "Not Disconnect Rqst",
960 "Bad Disconnect ACC Length",
961};
962
963static int
964nvme_fc_connect_admin_queue(struct nvme_fc_ctrl *ctrl,
965 struct nvme_fc_queue *queue, u16 qsize, u16 ersp_ratio)
966{
967 struct nvmefc_ls_req_op *lsop;
968 struct nvmefc_ls_req *lsreq;
969 struct fcnvme_ls_cr_assoc_rqst *assoc_rqst;
970 struct fcnvme_ls_cr_assoc_acc *assoc_acc;
971 int ret, fcret = 0;
972
973 lsop = kzalloc((sizeof(*lsop) +
974 ctrl->lport->ops->lsrqst_priv_sz +
975 sizeof(*assoc_rqst) + sizeof(*assoc_acc)), GFP_KERNEL);
976 if (!lsop) {
977 ret = -ENOMEM;
978 goto out_no_memory;
979 }
980 lsreq = &lsop->ls_req;
981
982 lsreq->private = (void *)&lsop[1];
983 assoc_rqst = (struct fcnvme_ls_cr_assoc_rqst *)
984 (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
985 assoc_acc = (struct fcnvme_ls_cr_assoc_acc *)&assoc_rqst[1];
986
987 assoc_rqst->w0.ls_cmd = FCNVME_LS_CREATE_ASSOCIATION;
988 assoc_rqst->desc_list_len =
989 cpu_to_be32(sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
990
991 assoc_rqst->assoc_cmd.desc_tag =
992 cpu_to_be32(FCNVME_LSDESC_CREATE_ASSOC_CMD);
993 assoc_rqst->assoc_cmd.desc_len =
994 fcnvme_lsdesc_len(
995 sizeof(struct fcnvme_lsdesc_cr_assoc_cmd));
996
997 assoc_rqst->assoc_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
998 assoc_rqst->assoc_cmd.sqsize = cpu_to_be16(qsize);
999 /* Linux supports only Dynamic controllers */
1000 assoc_rqst->assoc_cmd.cntlid = cpu_to_be16(0xffff);
8e412263 1001 uuid_copy(&assoc_rqst->assoc_cmd.hostid, &ctrl->ctrl.opts->host->id);
e399441d
JS		 1002 strncpy(assoc_rqst->assoc_cmd.hostnqn, ctrl->ctrl.opts->host->nqn,
1003 min(FCNVME_ASSOC_HOSTNQN_LEN, NVMF_NQN_SIZE));
1004 strncpy(assoc_rqst->assoc_cmd.subnqn, ctrl->ctrl.opts->subsysnqn,
1005 min(FCNVME_ASSOC_SUBNQN_LEN, NVMF_NQN_SIZE));
1006
1007 lsop->queue = queue;
1008 lsreq->rqstaddr = assoc_rqst;
1009 lsreq->rqstlen = sizeof(*assoc_rqst);
1010 lsreq->rspaddr = assoc_acc;
1011 lsreq->rsplen = sizeof(*assoc_acc);
1012 lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1013
c913a8b0 1014 ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
e399441d
JS		 1015 if (ret)
1016 goto out_free_buffer;
1017
1018 /* process connect LS completion */
1019
1020 /* validate the ACC response */
1021 if (assoc_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1022 fcret = VERR_LSACC;
f77fc87c 1023 else if (assoc_acc->hdr.desc_list_len !=
e399441d
JS		 1024 fcnvme_lsdesc_len(
1025 sizeof(struct fcnvme_ls_cr_assoc_acc)))
1026 fcret = VERR_CR_ASSOC_ACC_LEN;
f77fc87c
JS		 1027 else if (assoc_acc->hdr.rqst.desc_tag !=
1028 cpu_to_be32(FCNVME_LSDESC_RQST))
e399441d
JS		 1029 fcret = VERR_LSDESC_RQST;
1030 else if (assoc_acc->hdr.rqst.desc_len !=
1031 fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1032 fcret = VERR_LSDESC_RQST_LEN;
1033 else if (assoc_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_ASSOCIATION)
1034 fcret = VERR_CR_ASSOC;
1035 else if (assoc_acc->associd.desc_tag !=
1036 cpu_to_be32(FCNVME_LSDESC_ASSOC_ID))
1037 fcret = VERR_ASSOC_ID;
1038 else if (assoc_acc->associd.desc_len !=
1039 fcnvme_lsdesc_len(
1040 sizeof(struct fcnvme_lsdesc_assoc_id)))
1041 fcret = VERR_ASSOC_ID_LEN;
1042 else if (assoc_acc->connectid.desc_tag !=
1043 cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1044 fcret = VERR_CONN_ID;
1045 else if (assoc_acc->connectid.desc_len !=
1046 fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1047 fcret = VERR_CONN_ID_LEN;
1048
1049 if (fcret) {
1050 ret = -EBADF;
1051 dev_err(ctrl->dev,
1052 "q %d connect failed: %s\n",
1053 queue->qnum, validation_errors[fcret]);
1054 } else {
1055 ctrl->association_id =
1056 be64_to_cpu(assoc_acc->associd.association_id);
1057 queue->connection_id =
1058 be64_to_cpu(assoc_acc->connectid.connection_id);
1059 set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1060 }
1061
1062out_free_buffer:
1063 kfree(lsop);
1064out_no_memory:
1065 if (ret)
1066 dev_err(ctrl->dev,
1067 "queue %d connect admin queue failed (%d).\n",
1068 queue->qnum, ret);
1069 return ret;
1070}
1071
1072static int
1073nvme_fc_connect_queue(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
1074 u16 qsize, u16 ersp_ratio)
1075{
1076 struct nvmefc_ls_req_op *lsop;
1077 struct nvmefc_ls_req *lsreq;
1078 struct fcnvme_ls_cr_conn_rqst *conn_rqst;
1079 struct fcnvme_ls_cr_conn_acc *conn_acc;
1080 int ret, fcret = 0;
1081
1082 lsop = kzalloc((sizeof(*lsop) +
1083 ctrl->lport->ops->lsrqst_priv_sz +
1084 sizeof(*conn_rqst) + sizeof(*conn_acc)), GFP_KERNEL);
1085 if (!lsop) {
1086 ret = -ENOMEM;
1087 goto out_no_memory;
1088 }
1089 lsreq = &lsop->ls_req;
1090
1091 lsreq->private = (void *)&lsop[1];
1092 conn_rqst = (struct fcnvme_ls_cr_conn_rqst *)
1093 (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1094 conn_acc = (struct fcnvme_ls_cr_conn_acc *)&conn_rqst[1];
1095
1096 conn_rqst->w0.ls_cmd = FCNVME_LS_CREATE_CONNECTION;
1097 conn_rqst->desc_list_len = cpu_to_be32(
1098 sizeof(struct fcnvme_lsdesc_assoc_id) +
1099 sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1100
1101 conn_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1102 conn_rqst->associd.desc_len =
1103 fcnvme_lsdesc_len(
1104 sizeof(struct fcnvme_lsdesc_assoc_id));
1105 conn_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1106 conn_rqst->connect_cmd.desc_tag =
1107 cpu_to_be32(FCNVME_LSDESC_CREATE_CONN_CMD);
1108 conn_rqst->connect_cmd.desc_len =
1109 fcnvme_lsdesc_len(
1110 sizeof(struct fcnvme_lsdesc_cr_conn_cmd));
1111 conn_rqst->connect_cmd.ersp_ratio = cpu_to_be16(ersp_ratio);
1112 conn_rqst->connect_cmd.qid = cpu_to_be16(queue->qnum);
1113 conn_rqst->connect_cmd.sqsize = cpu_to_be16(qsize);
1114
1115 lsop->queue = queue;
1116 lsreq->rqstaddr = conn_rqst;
1117 lsreq->rqstlen = sizeof(*conn_rqst);
1118 lsreq->rspaddr = conn_acc;
1119 lsreq->rsplen = sizeof(*conn_acc);
1120 lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1121
c913a8b0 1122 ret = nvme_fc_send_ls_req(ctrl->rport, lsop);
e399441d
JS		 1123 if (ret)
1124 goto out_free_buffer;
1125
1126 /* process connect LS completion */
1127
1128 /* validate the ACC response */
1129 if (conn_acc->hdr.w0.ls_cmd != FCNVME_LS_ACC)
1130 fcret = VERR_LSACC;
f77fc87c 1131 else if (conn_acc->hdr.desc_list_len !=
e399441d
JS		 1132 fcnvme_lsdesc_len(sizeof(struct fcnvme_ls_cr_conn_acc)))
1133 fcret = VERR_CR_CONN_ACC_LEN;
f77fc87c 1134 else if (conn_acc->hdr.rqst.desc_tag != cpu_to_be32(FCNVME_LSDESC_RQST))
e399441d
JS		 1135 fcret = VERR_LSDESC_RQST;
1136 else if (conn_acc->hdr.rqst.desc_len !=
1137 fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_rqst)))
1138 fcret = VERR_LSDESC_RQST_LEN;
1139 else if (conn_acc->hdr.rqst.w0.ls_cmd != FCNVME_LS_CREATE_CONNECTION)
1140 fcret = VERR_CR_CONN;
1141 else if (conn_acc->connectid.desc_tag !=
1142 cpu_to_be32(FCNVME_LSDESC_CONN_ID))
1143 fcret = VERR_CONN_ID;
1144 else if (conn_acc->connectid.desc_len !=
1145 fcnvme_lsdesc_len(sizeof(struct fcnvme_lsdesc_conn_id)))
1146 fcret = VERR_CONN_ID_LEN;
1147
1148 if (fcret) {
1149 ret = -EBADF;
1150 dev_err(ctrl->dev,
1151 "q %d connect failed: %s\n",
1152 queue->qnum, validation_errors[fcret]);
1153 } else {
1154 queue->connection_id =
1155 be64_to_cpu(conn_acc->connectid.connection_id);
1156 set_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1157 }
1158
1159out_free_buffer:
1160 kfree(lsop);
1161out_no_memory:
1162 if (ret)
1163 dev_err(ctrl->dev,
1164 "queue %d connect command failed (%d).\n",
1165 queue->qnum, ret);
1166 return ret;
1167}
1168
1169static void
1170nvme_fc_disconnect_assoc_done(struct nvmefc_ls_req *lsreq, int status)
1171{
1172 struct nvmefc_ls_req_op *lsop = ls_req_to_lsop(lsreq);
e399441d 1173
c913a8b0 1174 __nvme_fc_finish_ls_req(lsop);
e399441d
JS		 1175
1176 /* fc-nvme iniator doesn't care about success or failure of cmd */
1177
1178 kfree(lsop);
1179}
1180
1181/*
1182 * This routine sends a FC-NVME LS to disconnect (aka terminate)
1183 * the FC-NVME Association. Terminating the association also
1184 * terminates the FC-NVME connections (per queue, both admin and io
1185 * queues) that are part of the association. E.g. things are torn
1186 * down, and the related FC-NVME Association ID and Connection IDs
1187 * become invalid.
1188 *
1189 * The behavior of the fc-nvme initiator is such that it's
1190 * understanding of the association and connections will implicitly
1191 * be torn down. The action is implicit as it may be due to a loss of
1192 * connectivity with the fc-nvme target, so you may never get a
1193 * response even if you tried. As such, the action of this routine
1194 * is to asynchronously send the LS, ignore any results of the LS, and
1195 * continue on with terminating the association. If the fc-nvme target
1196 * is present and receives the LS, it too can tear down.
1197 */
1198static void
1199nvme_fc_xmt_disconnect_assoc(struct nvme_fc_ctrl *ctrl)
1200{
1201 struct fcnvme_ls_disconnect_rqst *discon_rqst;
1202 struct fcnvme_ls_disconnect_acc *discon_acc;
1203 struct nvmefc_ls_req_op *lsop;
1204 struct nvmefc_ls_req *lsreq;
c913a8b0 1205 int ret;
e399441d
JS		 1206
1207 lsop = kzalloc((sizeof(*lsop) +
1208 ctrl->lport->ops->lsrqst_priv_sz +
1209 sizeof(*discon_rqst) + sizeof(*discon_acc)),
1210 GFP_KERNEL);
1211 if (!lsop)
1212 /* couldn't sent it... too bad */
1213 return;
1214
1215 lsreq = &lsop->ls_req;
1216
1217 lsreq->private = (void *)&lsop[1];
1218 discon_rqst = (struct fcnvme_ls_disconnect_rqst *)
1219 (lsreq->private + ctrl->lport->ops->lsrqst_priv_sz);
1220 discon_acc = (struct fcnvme_ls_disconnect_acc *)&discon_rqst[1];
1221
1222 discon_rqst->w0.ls_cmd = FCNVME_LS_DISCONNECT;
1223 discon_rqst->desc_list_len = cpu_to_be32(
1224 sizeof(struct fcnvme_lsdesc_assoc_id) +
1225 sizeof(struct fcnvme_lsdesc_disconn_cmd));
1226
1227 discon_rqst->associd.desc_tag = cpu_to_be32(FCNVME_LSDESC_ASSOC_ID);
1228 discon_rqst->associd.desc_len =
1229 fcnvme_lsdesc_len(
1230 sizeof(struct fcnvme_lsdesc_assoc_id));
1231
1232 discon_rqst->associd.association_id = cpu_to_be64(ctrl->association_id);
1233
1234 discon_rqst->discon_cmd.desc_tag = cpu_to_be32(
1235 FCNVME_LSDESC_DISCONN_CMD);
1236 discon_rqst->discon_cmd.desc_len =
1237 fcnvme_lsdesc_len(
1238 sizeof(struct fcnvme_lsdesc_disconn_cmd));
1239 discon_rqst->discon_cmd.scope = FCNVME_DISCONN_ASSOCIATION;
1240 discon_rqst->discon_cmd.id = cpu_to_be64(ctrl->association_id);
1241
1242 lsreq->rqstaddr = discon_rqst;
1243 lsreq->rqstlen = sizeof(*discon_rqst);
1244 lsreq->rspaddr = discon_acc;
1245 lsreq->rsplen = sizeof(*discon_acc);
1246 lsreq->timeout = NVME_FC_CONNECT_TIMEOUT_SEC;
1247
c913a8b0
JS		 1248 ret = nvme_fc_send_ls_req_async(ctrl->rport, lsop,
1249 nvme_fc_disconnect_assoc_done);
1250 if (ret)
1251 kfree(lsop);
e399441d
JS		 1252
1253 /* only meaningful part to terminating the association */
1254 ctrl->association_id = 0;
1255}
1256
1257
1258/* *********************** NVME Ctrl Routines **************************** */
1259
78a7ac26 1260static void __nvme_fc_final_op_cleanup(struct request *rq);
f874d5d0 1261static void nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg);
e399441d
JS		 1262
1263static int
1264nvme_fc_reinit_request(void *data, struct request *rq)
1265{
1266 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1267 struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1268
1269 memset(cmdiu, 0, sizeof(*cmdiu));
1270 cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1271 cmdiu->fc_id = NVME_CMD_FC_ID;
1272 cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1273 memset(&op->rsp_iu, 0, sizeof(op->rsp_iu));
1274
1275 return 0;
1276}
1277
1278static void
1279__nvme_fc_exit_request(struct nvme_fc_ctrl *ctrl,
1280 struct nvme_fc_fcp_op *op)
1281{
1282 fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.rspdma,
1283 sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1284 fc_dma_unmap_single(ctrl->lport->dev, op->fcp_req.cmddma,
1285 sizeof(op->cmd_iu), DMA_TO_DEVICE);
1286
1287 atomic_set(&op->state, FCPOP_STATE_UNINIT);
1288}
1289
1290static void
d6296d39
CH		 1291nvme_fc_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1292 unsigned int hctx_idx)
e399441d
JS		 1293{
1294 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1295
d6296d39 1296 return __nvme_fc_exit_request(set->driver_data, op);
e399441d
JS		 1297}
1298
78a7ac26
JS		 1299static int
1300__nvme_fc_abort_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_fcp_op *op)
1301{
1302 int state;
1303
1304 state = atomic_xchg(&op->state, FCPOP_STATE_ABORTED);
1305 if (state != FCPOP_STATE_ACTIVE) {
1306 atomic_set(&op->state, state);
1307 return -ECANCELED;
1308 }
1309
1310 ctrl->lport->ops->fcp_abort(&ctrl->lport->localport,
1311 &ctrl->rport->remoteport,
1312 op->queue->lldd_handle,
1313 &op->fcp_req);
1314
1315 return 0;
1316}
1317
e399441d 1318static void
78a7ac26 1319nvme_fc_abort_aen_ops(struct nvme_fc_ctrl *ctrl)
e399441d
JS		 1320{
1321 struct nvme_fc_fcp_op *aen_op = ctrl->aen_ops;
78a7ac26
JS		 1322 unsigned long flags;
1323 int i, ret;
e399441d
JS		 1324
1325 for (i = 0; i < NVME_FC_NR_AEN_COMMANDS; i++, aen_op++) {
78a7ac26 1326 if (atomic_read(&aen_op->state) != FCPOP_STATE_ACTIVE)
e399441d 1327 continue;
78a7ac26
JS		 1328
1329 spin_lock_irqsave(&ctrl->lock, flags);
61bff8ef
JS		 1330 if (ctrl->flags & FCCTRL_TERMIO) {
1331 ctrl->iocnt++;
1332 aen_op->flags |= FCOP_FLAGS_TERMIO;
1333 }
78a7ac26
JS		 1334 spin_unlock_irqrestore(&ctrl->lock, flags);
1335
1336 ret = __nvme_fc_abort_op(ctrl, aen_op);
1337 if (ret) {
1338 /*
1339 * if __nvme_fc_abort_op failed the io wasn't
1340 * active. Thus this call path is running in
1341 * parallel to the io complete. Treat as non-error.
1342 */
1343
1344 /* back out the flags/counters */
1345 spin_lock_irqsave(&ctrl->lock, flags);
61bff8ef
JS		 1346 if (ctrl->flags & FCCTRL_TERMIO)
1347 ctrl->iocnt--;
78a7ac26
JS		 1348 aen_op->flags &= ~FCOP_FLAGS_TERMIO;
1349 spin_unlock_irqrestore(&ctrl->lock, flags);
1350 return;
1351 }
e399441d
JS		 1352 }
1353}
1354
78a7ac26
JS		 1355static inline int
1356__nvme_fc_fcpop_chk_teardowns(struct nvme_fc_ctrl *ctrl,
1357 struct nvme_fc_fcp_op *op)
1358{
1359 unsigned long flags;
1360 bool complete_rq = false;
1361
1362 spin_lock_irqsave(&ctrl->lock, flags);
61bff8ef 1363 if (unlikely(op->flags & FCOP_FLAGS_TERMIO)) {
36715cf4
JS		 1364 if (ctrl->flags & FCCTRL_TERMIO) {
1365 if (!--ctrl->iocnt)
1366 wake_up(&ctrl->ioabort_wait);
1367 }
61bff8ef 1368 }
78a7ac26
JS		 1369 if (op->flags & FCOP_FLAGS_RELEASED)
1370 complete_rq = true;
1371 else
1372 op->flags |= FCOP_FLAGS_COMPLETE;
1373 spin_unlock_irqrestore(&ctrl->lock, flags);
1374
1375 return complete_rq;
1376}
1377
baee29ac 1378static void
e399441d
JS		 1379nvme_fc_fcpio_done(struct nvmefc_fcp_req *req)
1380{
1381 struct nvme_fc_fcp_op *op = fcp_req_to_fcp_op(req);
1382 struct request *rq = op->rq;
1383 struct nvmefc_fcp_req *freq = &op->fcp_req;
1384 struct nvme_fc_ctrl *ctrl = op->ctrl;
1385 struct nvme_fc_queue *queue = op->queue;
1386 struct nvme_completion *cqe = &op->rsp_iu.cqe;
458f280d 1387 struct nvme_command *sqe = &op->cmd_iu.sqe;
d663b69f 1388 __le16 status = cpu_to_le16(NVME_SC_SUCCESS << 1);
27fa9bc5 1389 union nvme_result result;
f874d5d0 1390 bool complete_rq, terminate_assoc = true;
e399441d
JS		 1391
1392 /*
1393 * WARNING:
1394 * The current linux implementation of a nvme controller
1395 * allocates a single tag set for all io queues and sizes
1396 * the io queues to fully hold all possible tags. Thus, the
1397 * implementation does not reference or care about the sqhd
1398 * value as it never needs to use the sqhd/sqtail pointers
1399 * for submission pacing.
1400 *
1401 * This affects the FC-NVME implementation in two ways:
1402 * 1) As the value doesn't matter, we don't need to waste
1403 * cycles extracting it from ERSPs and stamping it in the
1404 * cases where the transport fabricates CQEs on successful
1405 * completions.
1406 * 2) The FC-NVME implementation requires that delivery of
1407 * ERSP completions are to go back to the nvme layer in order
1408 * relative to the rsn, such that the sqhd value will always
1409 * be "in order" for the nvme layer. As the nvme layer in
1410 * linux doesn't care about sqhd, there's no need to return
1411 * them in order.
1412 *
1413 * Additionally:
1414 * As the core nvme layer in linux currently does not look at
1415 * every field in the cqe - in cases where the FC transport must
1416 * fabricate a CQE, the following fields will not be set as they
1417 * are not referenced:
1418 * cqe.sqid, cqe.sqhd, cqe.command_id
f874d5d0
JS		 1419 *
1420 * Failure or error of an individual i/o, in a transport
1421 * detected fashion unrelated to the nvme completion status,
1422 * potentially cause the initiator and target sides to get out
1423 * of sync on SQ head/tail (aka outstanding io count allowed).
1424 * Per FC-NVME spec, failure of an individual command requires
1425 * the connection to be terminated, which in turn requires the
1426 * association to be terminated.
e399441d
JS		 1427 */
1428
1429 fc_dma_sync_single_for_cpu(ctrl->lport->dev, op->fcp_req.rspdma,
1430 sizeof(op->rsp_iu), DMA_FROM_DEVICE);
1431
1432 if (atomic_read(&op->state) == FCPOP_STATE_ABORTED)
d663b69f 1433 status = cpu_to_le16((NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1);
62eeacb0 1434 else if (freq->status)
56b7103a 1435 status = cpu_to_le16(NVME_SC_INTERNAL << 1);
e399441d
JS		 1436
1437 /*
1438 * For the linux implementation, if we have an unsuccesful
1439 * status, they blk-mq layer can typically be called with the
1440 * non-zero status and the content of the cqe isn't important.
1441 */
1442 if (status)
1443 goto done;
1444
1445 /*
1446 * command completed successfully relative to the wire
1447 * protocol. However, validate anything received and
1448 * extract the status and result from the cqe (create it
1449 * where necessary).
1450 */
1451
1452 switch (freq->rcv_rsplen) {
1453
1454 case 0:
1455 case NVME_FC_SIZEOF_ZEROS_RSP:
1456 /*
1457 * No response payload or 12 bytes of payload (which
1458 * should all be zeros) are considered successful and
1459 * no payload in the CQE by the transport.
1460 */
1461 if (freq->transferred_length !=
1462 be32_to_cpu(op->cmd_iu.data_len)) {
56b7103a 1463 status = cpu_to_le16(NVME_SC_INTERNAL << 1);
e399441d
JS		 1464 goto done;
1465 }
27fa9bc5 1466 result.u64 = 0;
e399441d
JS		 1467 break;
1468
1469 case sizeof(struct nvme_fc_ersp_iu):
1470 /*
1471 * The ERSP IU contains a full completion with CQE.
1472 * Validate ERSP IU and look at cqe.
1473 */
1474 if (unlikely(be16_to_cpu(op->rsp_iu.iu_len) !=
1475 (freq->rcv_rsplen / 4) ||
1476 be32_to_cpu(op->rsp_iu.xfrd_len) !=
1477 freq->transferred_length ||
726a1080 1478 op->rsp_iu.status_code ||
458f280d 1479 sqe->common.command_id != cqe->command_id)) {
56b7103a 1480 status = cpu_to_le16(NVME_SC_INTERNAL << 1);
e399441d
JS		 1481 goto done;
1482 }
27fa9bc5 1483 result = cqe->result;
d663b69f 1484 status = cqe->status;
e399441d
JS		 1485 break;
1486
1487 default:
56b7103a 1488 status = cpu_to_le16(NVME_SC_INTERNAL << 1);
e399441d
JS		 1489 goto done;
1490 }
1491
f874d5d0
JS		 1492 terminate_assoc = false;
1493
e399441d 1494done:
78a7ac26 1495 if (op->flags & FCOP_FLAGS_AEN) {
27fa9bc5 1496 nvme_complete_async_event(&queue->ctrl->ctrl, status, &result);
78a7ac26
JS		 1497 complete_rq = __nvme_fc_fcpop_chk_teardowns(ctrl, op);
1498 atomic_set(&op->state, FCPOP_STATE_IDLE);
1499 op->flags = FCOP_FLAGS_AEN; /* clear other flags */
e399441d 1500 nvme_fc_ctrl_put(ctrl);
f874d5d0 1501 goto check_error;
e399441d
JS		 1502 }
1503
78a7ac26
JS		 1504 complete_rq = __nvme_fc_fcpop_chk_teardowns(ctrl, op);
1505 if (!complete_rq) {
1506 if (unlikely(op->flags & FCOP_FLAGS_TERMIO)) {
e392e1f1 1507 status = cpu_to_le16(NVME_SC_ABORT_REQ << 1);
78a7ac26 1508 if (blk_queue_dying(rq->q))
e392e1f1 1509 status |= cpu_to_le16(NVME_SC_DNR << 1);
78a7ac26
JS		 1510 }
1511 nvme_end_request(rq, status, result);
1512 } else
1513 __nvme_fc_final_op_cleanup(rq);
f874d5d0
JS		 1514
1515check_error:
1516 if (terminate_assoc)
1517 nvme_fc_error_recovery(ctrl, "transport detected io error");
e399441d
JS		 1518}
1519
1520static int
1521__nvme_fc_init_request(struct nvme_fc_ctrl *ctrl,
1522 struct nvme_fc_queue *queue, struct nvme_fc_fcp_op *op,
1523 struct request *rq, u32 rqno)
1524{
1525 struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
1526 int ret = 0;
1527
1528 memset(op, 0, sizeof(*op));
1529 op->fcp_req.cmdaddr = &op->cmd_iu;
1530 op->fcp_req.cmdlen = sizeof(op->cmd_iu);
1531 op->fcp_req.rspaddr = &op->rsp_iu;
1532 op->fcp_req.rsplen = sizeof(op->rsp_iu);
1533 op->fcp_req.done = nvme_fc_fcpio_done;
1534 op->fcp_req.first_sgl = (struct scatterlist *)&op[1];
1535 op->fcp_req.private = &op->fcp_req.first_sgl[SG_CHUNK_SIZE];
1536 op->ctrl = ctrl;
1537 op->queue = queue;
1538 op->rq = rq;
1539 op->rqno = rqno;
1540
1541 cmdiu->scsi_id = NVME_CMD_SCSI_ID;
1542 cmdiu->fc_id = NVME_CMD_FC_ID;
1543 cmdiu->iu_len = cpu_to_be16(sizeof(*cmdiu) / sizeof(u32));
1544
1545 op->fcp_req.cmddma = fc_dma_map_single(ctrl->lport->dev,
1546 &op->cmd_iu, sizeof(op->cmd_iu), DMA_TO_DEVICE);
1547 if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.cmddma)) {
1548 dev_err(ctrl->dev,
1549 "FCP Op failed - cmdiu dma mapping failed.\n");
1550 ret = EFAULT;
1551 goto out_on_error;
1552 }
1553
1554 op->fcp_req.rspdma = fc_dma_map_single(ctrl->lport->dev,
1555 &op->rsp_iu, sizeof(op->rsp_iu),
1556 DMA_FROM_DEVICE);
1557 if (fc_dma_mapping_error(ctrl->lport->dev, op->fcp_req.rspdma)) {
1558 dev_err(ctrl->dev,
1559 "FCP Op failed - rspiu dma mapping failed.\n");
1560 ret = EFAULT;
1561 }
1562
1563 atomic_set(&op->state, FCPOP_STATE_IDLE);
1564out_on_error:
1565 return ret;
1566}
1567
1568static int
d6296d39
CH		 1569nvme_fc_init_request(struct blk_mq_tag_set *set, struct request *rq,
1570 unsigned int hctx_idx, unsigned int numa_node)
e399441d 1571{
d6296d39 1572 struct nvme_fc_ctrl *ctrl = set->driver_data;
e399441d 1573 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
76f983cb
CH		 1574 int queue_idx = (set == &ctrl->tag_set) ? hctx_idx + 1 : 0;
1575 struct nvme_fc_queue *queue = &ctrl->queues[queue_idx];
e399441d
JS		 1576
1577 return __nvme_fc_init_request(ctrl, queue, op, rq, queue->rqcnt++);
1578}
1579
1580static int
1581nvme_fc_init_aen_ops(struct nvme_fc_ctrl *ctrl)
1582{
1583 struct nvme_fc_fcp_op *aen_op;
1584 struct nvme_fc_cmd_iu *cmdiu;
1585 struct nvme_command *sqe;
61bff8ef 1586 void *private;
e399441d
JS		 1587 int i, ret;
1588
1589 aen_op = ctrl->aen_ops;
1590 for (i = 0; i < NVME_FC_NR_AEN_COMMANDS; i++, aen_op++) {
61bff8ef
JS		 1591 private = kzalloc(ctrl->lport->ops->fcprqst_priv_sz,
1592 GFP_KERNEL);
1593 if (!private)
1594 return -ENOMEM;
1595
e399441d
JS		 1596 cmdiu = &aen_op->cmd_iu;
1597 sqe = &cmdiu->sqe;
1598 ret = __nvme_fc_init_request(ctrl, &ctrl->queues[0],
1599 aen_op, (struct request *)NULL,
1600 (AEN_CMDID_BASE + i));
61bff8ef
JS		 1601 if (ret) {
1602 kfree(private);
e399441d 1603 return ret;
61bff8ef 1604 }
e399441d 1605
78a7ac26 1606 aen_op->flags = FCOP_FLAGS_AEN;
61bff8ef
JS		 1607 aen_op->fcp_req.first_sgl = NULL; /* no sg list */
1608 aen_op->fcp_req.private = private;
78a7ac26 1609
e399441d
JS		 1610 memset(sqe, 0, sizeof(*sqe));
1611 sqe->common.opcode = nvme_admin_async_event;
78a7ac26 1612 /* Note: core layer may overwrite the sqe.command_id value */
e399441d
JS		 1613 sqe->common.command_id = AEN_CMDID_BASE + i;
1614 }
1615 return 0;
1616}
1617
61bff8ef
JS		 1618static void
1619nvme_fc_term_aen_ops(struct nvme_fc_ctrl *ctrl)
1620{
1621 struct nvme_fc_fcp_op *aen_op;
1622 int i;
1623
1624 aen_op = ctrl->aen_ops;
1625 for (i = 0; i < NVME_FC_NR_AEN_COMMANDS; i++, aen_op++) {
1626 if (!aen_op->fcp_req.private)
1627 continue;
1628
1629 __nvme_fc_exit_request(ctrl, aen_op);
1630
1631 kfree(aen_op->fcp_req.private);
1632 aen_op->fcp_req.private = NULL;
1633 }
1634}
e399441d
JS		 1635
1636static inline void
1637__nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, struct nvme_fc_ctrl *ctrl,
1638 unsigned int qidx)
1639{
1640 struct nvme_fc_queue *queue = &ctrl->queues[qidx];
1641
1642 hctx->driver_data = queue;
1643 queue->hctx = hctx;
1644}
1645
1646static int
1647nvme_fc_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1648 unsigned int hctx_idx)
1649{
1650 struct nvme_fc_ctrl *ctrl = data;
1651
1652 __nvme_fc_init_hctx(hctx, ctrl, hctx_idx + 1);
1653
1654 return 0;
1655}
1656
1657static int
1658nvme_fc_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1659 unsigned int hctx_idx)
1660{
1661 struct nvme_fc_ctrl *ctrl = data;
1662
1663 __nvme_fc_init_hctx(hctx, ctrl, hctx_idx);
1664
1665 return 0;
1666}
1667
1668static void
1669nvme_fc_init_queue(struct nvme_fc_ctrl *ctrl, int idx, size_t queue_size)
1670{
1671 struct nvme_fc_queue *queue;
1672
1673 queue = &ctrl->queues[idx];
1674 memset(queue, 0, sizeof(*queue));
1675 queue->ctrl = ctrl;
1676 queue->qnum = idx;
1677 atomic_set(&queue->csn, 1);
1678 queue->dev = ctrl->dev;
1679
1680 if (idx > 0)
1681 queue->cmnd_capsule_len = ctrl->ctrl.ioccsz * 16;
1682 else
1683 queue->cmnd_capsule_len = sizeof(struct nvme_command);
1684
1685 queue->queue_size = queue_size;
1686
1687 /*
1688 * Considered whether we should allocate buffers for all SQEs
1689 * and CQEs and dma map them - mapping their respective entries
1690 * into the request structures (kernel vm addr and dma address)
1691 * thus the driver could use the buffers/mappings directly.
1692 * It only makes sense if the LLDD would use them for its
1693 * messaging api. It's very unlikely most adapter api's would use
1694 * a native NVME sqe/cqe. More reasonable if FC-NVME IU payload
1695 * structures were used instead.
1696 */
1697}
1698
1699/*
1700 * This routine terminates a queue at the transport level.
1701 * The transport has already ensured that all outstanding ios on
1702 * the queue have been terminated.
1703 * The transport will send a Disconnect LS request to terminate
1704 * the queue's connection. Termination of the admin queue will also
1705 * terminate the association at the target.
1706 */
1707static void
1708nvme_fc_free_queue(struct nvme_fc_queue *queue)
1709{
1710 if (!test_and_clear_bit(NVME_FC_Q_CONNECTED, &queue->flags))
1711 return;
1712
1713 /*
1714 * Current implementation never disconnects a single queue.
1715 * It always terminates a whole association. So there is never
1716 * a disconnect(queue) LS sent to the target.
1717 */
1718
1719 queue->connection_id = 0;
1720 clear_bit(NVME_FC_Q_CONNECTED, &queue->flags);
1721}
1722
1723static void
1724__nvme_fc_delete_hw_queue(struct nvme_fc_ctrl *ctrl,
1725 struct nvme_fc_queue *queue, unsigned int qidx)
1726{
1727 if (ctrl->lport->ops->delete_queue)
1728 ctrl->lport->ops->delete_queue(&ctrl->lport->localport, qidx,
1729 queue->lldd_handle);
1730 queue->lldd_handle = NULL;
1731}
1732
e399441d
JS		 1733static void
1734nvme_fc_free_io_queues(struct nvme_fc_ctrl *ctrl)
1735{
1736 int i;
1737
d858e5f0 1738 for (i = 1; i < ctrl->ctrl.queue_count; i++)
e399441d
JS		 1739 nvme_fc_free_queue(&ctrl->queues[i]);
1740}
1741
1742static int
1743__nvme_fc_create_hw_queue(struct nvme_fc_ctrl *ctrl,
1744 struct nvme_fc_queue *queue, unsigned int qidx, u16 qsize)
1745{
1746 int ret = 0;
1747
1748 queue->lldd_handle = NULL;
1749 if (ctrl->lport->ops->create_queue)
1750 ret = ctrl->lport->ops->create_queue(&ctrl->lport->localport,
1751 qidx, qsize, &queue->lldd_handle);
1752
1753 return ret;
1754}
1755
1756static void
1757nvme_fc_delete_hw_io_queues(struct nvme_fc_ctrl *ctrl)
1758{
d858e5f0 1759 struct nvme_fc_queue *queue = &ctrl->queues[ctrl->ctrl.queue_count - 1];
e399441d
JS		 1760 int i;
1761
d858e5f0 1762 for (i = ctrl->ctrl.queue_count - 1; i >= 1; i--, queue--)
e399441d
JS		 1763 __nvme_fc_delete_hw_queue(ctrl, queue, i);
1764}
1765
1766static int
1767nvme_fc_create_hw_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1768{
1769 struct nvme_fc_queue *queue = &ctrl->queues[1];
17a1ec08 1770 int i, ret;
e399441d 1771
d858e5f0 1772 for (i = 1; i < ctrl->ctrl.queue_count; i++, queue++) {
e399441d 1773 ret = __nvme_fc_create_hw_queue(ctrl, queue, i, qsize);
17a1ec08
JT		 1774 if (ret)
1775 goto delete_queues;
e399441d
JS		 1776 }
1777
1778 return 0;
17a1ec08
JT		 1779
1780delete_queues:
1781 for (; i >= 0; i--)
1782 __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[i], i);
1783 return ret;
e399441d
JS		 1784}
1785
1786static int
1787nvme_fc_connect_io_queues(struct nvme_fc_ctrl *ctrl, u16 qsize)
1788{
1789 int i, ret = 0;
1790
d858e5f0 1791 for (i = 1; i < ctrl->ctrl.queue_count; i++) {
e399441d
JS		 1792 ret = nvme_fc_connect_queue(ctrl, &ctrl->queues[i], qsize,
1793 (qsize / 5));
1794 if (ret)
1795 break;
1796 ret = nvmf_connect_io_queue(&ctrl->ctrl, i);
1797 if (ret)
1798 break;
1799 }
1800
1801 return ret;
1802}
1803
1804static void
1805nvme_fc_init_io_queues(struct nvme_fc_ctrl *ctrl)
1806{
1807 int i;
1808
d858e5f0 1809 for (i = 1; i < ctrl->ctrl.queue_count; i++)
e399441d
JS		 1810 nvme_fc_init_queue(ctrl, i, ctrl->ctrl.sqsize);
1811}
1812
1813static void
1814nvme_fc_ctrl_free(struct kref *ref)
1815{
1816 struct nvme_fc_ctrl *ctrl =
1817 container_of(ref, struct nvme_fc_ctrl, ref);
1818 unsigned long flags;
1819
61bff8ef
JS		 1820 if (ctrl->ctrl.tagset) {
1821 blk_cleanup_queue(ctrl->ctrl.connect_q);
1822 blk_mq_free_tag_set(&ctrl->tag_set);
e399441d
JS		 1823 }
1824
61bff8ef
JS		 1825 /* remove from rport list */
1826 spin_lock_irqsave(&ctrl->rport->lock, flags);
1827 list_del(&ctrl->ctrl_list);
1828 spin_unlock_irqrestore(&ctrl->rport->lock, flags);
1829
f9c5af5f 1830 blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
61bff8ef
JS		 1831 blk_cleanup_queue(ctrl->ctrl.admin_q);
1832 blk_mq_free_tag_set(&ctrl->admin_tag_set);
1833
1834 kfree(ctrl->queues);
1835
e399441d
JS		 1836 put_device(ctrl->dev);
1837 nvme_fc_rport_put(ctrl->rport);
1838
e399441d 1839 ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
de41447a
EM		 1840 if (ctrl->ctrl.opts)
1841 nvmf_free_options(ctrl->ctrl.opts);
e399441d
JS		 1842 kfree(ctrl);
1843}
1844
1845static void
1846nvme_fc_ctrl_put(struct nvme_fc_ctrl *ctrl)
1847{
1848 kref_put(&ctrl->ref, nvme_fc_ctrl_free);
1849}
1850
1851static int
1852nvme_fc_ctrl_get(struct nvme_fc_ctrl *ctrl)
1853{
1854 return kref_get_unless_zero(&ctrl->ref);
1855}
1856
1857/*
1858 * All accesses from nvme core layer done - can now free the
1859 * controller. Called after last nvme_put_ctrl() call
1860 */
1861static void
61bff8ef 1862nvme_fc_nvme_ctrl_freed(struct nvme_ctrl *nctrl)
e399441d
JS		 1863{
1864 struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
1865
1866 WARN_ON(nctrl != &ctrl->ctrl);
1867
61bff8ef
JS		 1868 nvme_fc_ctrl_put(ctrl);
1869}
e399441d 1870
61bff8ef
JS		 1871static void
1872nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg)
1873{
69fa9646
JS		 1874 /* only proceed if in LIVE state - e.g. on first error */
1875 if (ctrl->ctrl.state != NVME_CTRL_LIVE)
1876 return;
1877
61bff8ef
JS		 1878 dev_warn(ctrl->ctrl.device,
1879 "NVME-FC{%d}: transport association error detected: %s\n",
1880 ctrl->cnum, errmsg);
589ff775 1881 dev_warn(ctrl->ctrl.device,
61bff8ef 1882 "NVME-FC{%d}: resetting controller\n", ctrl->cnum);
e399441d 1883
61bff8ef
JS		 1884 if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RECONNECTING)) {
1885 dev_err(ctrl->ctrl.device,
1886 "NVME-FC{%d}: error_recovery: Couldn't change state "
1887 "to RECONNECTING\n", ctrl->cnum);
1888 return;
e399441d
JS		 1889 }
1890
d86c4d8e 1891 nvme_reset_ctrl(&ctrl->ctrl);
e399441d
JS		 1892}
1893
baee29ac 1894static enum blk_eh_timer_return
e399441d
JS		 1895nvme_fc_timeout(struct request *rq, bool reserved)
1896{
1897 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
1898 struct nvme_fc_ctrl *ctrl = op->ctrl;
1899 int ret;
1900
1901 if (reserved)
1902 return BLK_EH_RESET_TIMER;
1903
1904 ret = __nvme_fc_abort_op(ctrl, op);
1905 if (ret)
1906 /* io wasn't active to abort consider it done */
1907 return BLK_EH_HANDLED;
1908
1909 /*
61bff8ef
JS		 1910 * we can't individually ABTS an io without affecting the queue,
1911 * thus killing the queue, adn thus the association.
1912 * So resolve by performing a controller reset, which will stop
1913 * the host/io stack, terminate the association on the link,
1914 * and recreate an association on the link.
e399441d 1915 */
61bff8ef 1916 nvme_fc_error_recovery(ctrl, "io timeout error");
e399441d
JS		 1917
1918 return BLK_EH_HANDLED;
1919}
1920
1921static int
1922nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
1923 struct nvme_fc_fcp_op *op)
1924{
1925 struct nvmefc_fcp_req *freq = &op->fcp_req;
e399441d
JS		 1926 enum dma_data_direction dir;
1927 int ret;
1928
1929 freq->sg_cnt = 0;
1930
b131c61d 1931 if (!blk_rq_payload_bytes(rq))
e399441d
JS		 1932 return 0;
1933
1934 freq->sg_table.sgl = freq->first_sgl;
19e420bb
CH		 1935 ret = sg_alloc_table_chained(&freq->sg_table,
1936 blk_rq_nr_phys_segments(rq), freq->sg_table.sgl);
e399441d
JS		 1937 if (ret)
1938 return -ENOMEM;
1939
1940 op->nents = blk_rq_map_sg(rq->q, rq, freq->sg_table.sgl);
19e420bb 1941 WARN_ON(op->nents > blk_rq_nr_phys_segments(rq));
e399441d
JS		 1942 dir = (rq_data_dir(rq) == WRITE) ? DMA_TO_DEVICE : DMA_FROM_DEVICE;
1943 freq->sg_cnt = fc_dma_map_sg(ctrl->lport->dev, freq->sg_table.sgl,
1944 op->nents, dir);
1945 if (unlikely(freq->sg_cnt <= 0)) {
1946 sg_free_table_chained(&freq->sg_table, true);
1947 freq->sg_cnt = 0;
1948 return -EFAULT;
1949 }
1950
1951 /*
1952 * TODO: blk_integrity_rq(rq) for DIF
1953 */
1954 return 0;
1955}
1956
1957static void
1958nvme_fc_unmap_data(struct nvme_fc_ctrl *ctrl, struct request *rq,
1959 struct nvme_fc_fcp_op *op)
1960{
1961 struct nvmefc_fcp_req *freq = &op->fcp_req;
1962
1963 if (!freq->sg_cnt)
1964 return;
1965
1966 fc_dma_unmap_sg(ctrl->lport->dev, freq->sg_table.sgl, op->nents,
1967 ((rq_data_dir(rq) == WRITE) ?
1968 DMA_TO_DEVICE : DMA_FROM_DEVICE));
1969
1970 nvme_cleanup_cmd(rq);
1971
1972 sg_free_table_chained(&freq->sg_table, true);
1973
1974 freq->sg_cnt = 0;
1975}
1976
1977/*
1978 * In FC, the queue is a logical thing. At transport connect, the target
1979 * creates its "queue" and returns a handle that is to be given to the
1980 * target whenever it posts something to the corresponding SQ. When an
1981 * SQE is sent on a SQ, FC effectively considers the SQE, or rather the
1982 * command contained within the SQE, an io, and assigns a FC exchange
1983 * to it. The SQE and the associated SQ handle are sent in the initial
1984 * CMD IU sents on the exchange. All transfers relative to the io occur
1985 * as part of the exchange. The CQE is the last thing for the io,
1986 * which is transferred (explicitly or implicitly) with the RSP IU
1987 * sent on the exchange. After the CQE is received, the FC exchange is
1988 * terminaed and the Exchange may be used on a different io.
1989 *
1990 * The transport to LLDD api has the transport making a request for a
1991 * new fcp io request to the LLDD. The LLDD then allocates a FC exchange
1992 * resource and transfers the command. The LLDD will then process all
1993 * steps to complete the io. Upon completion, the transport done routine
1994 * is called.
1995 *
1996 * So - while the operation is outstanding to the LLDD, there is a link
1997 * level FC exchange resource that is also outstanding. This must be
1998 * considered in all cleanup operations.
1999 */
fc17b653 2000static blk_status_t
e399441d
JS		 2001nvme_fc_start_fcp_op(struct nvme_fc_ctrl *ctrl, struct nvme_fc_queue *queue,
2002 struct nvme_fc_fcp_op *op, u32 data_len,
2003 enum nvmefc_fcp_datadir io_dir)
2004{
2005 struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2006 struct nvme_command *sqe = &cmdiu->sqe;
2007 u32 csn;
2008 int ret;
2009
61bff8ef
JS		 2010 /*
2011 * before attempting to send the io, check to see if we believe
2012 * the target device is present
2013 */
2014 if (ctrl->rport->remoteport.port_state != FC_OBJSTATE_ONLINE)
8b25f351 2015 goto busy;
61bff8ef 2016
e399441d 2017 if (!nvme_fc_ctrl_get(ctrl))
fc17b653 2018 return BLK_STS_IOERR;
e399441d
JS		 2019
2020 /* format the FC-NVME CMD IU and fcp_req */
2021 cmdiu->connection_id = cpu_to_be64(queue->connection_id);
2022 csn = atomic_inc_return(&queue->csn);
2023 cmdiu->csn = cpu_to_be32(csn);
2024 cmdiu->data_len = cpu_to_be32(data_len);
2025 switch (io_dir) {
2026 case NVMEFC_FCP_WRITE:
2027 cmdiu->flags = FCNVME_CMD_FLAGS_WRITE;
2028 break;
2029 case NVMEFC_FCP_READ:
2030 cmdiu->flags = FCNVME_CMD_FLAGS_READ;
2031 break;
2032 case NVMEFC_FCP_NODATA:
2033 cmdiu->flags = 0;
2034 break;
2035 }
2036 op->fcp_req.payload_length = data_len;
2037 op->fcp_req.io_dir = io_dir;
2038 op->fcp_req.transferred_length = 0;
2039 op->fcp_req.rcv_rsplen = 0;
62eeacb0 2040 op->fcp_req.status = NVME_SC_SUCCESS;
e399441d
JS		 2041 op->fcp_req.sqid = cpu_to_le16(queue->qnum);
2042
2043 /*
2044 * validate per fabric rules, set fields mandated by fabric spec
2045 * as well as those by FC-NVME spec.
2046 */
2047 WARN_ON_ONCE(sqe->common.metadata);
e399441d
JS		 2048 sqe->common.flags |= NVME_CMD_SGL_METABUF;
2049
2050 /*
d9d34c0b
JS		 2051 * format SQE DPTR field per FC-NVME rules:
2052 * type=0x5 Transport SGL Data Block Descriptor
2053 * subtype=0xA Transport-specific value
2054 * address=0
2055 * length=length of the data series
e399441d 2056 */
d9d34c0b
JS		 2057 sqe->rw.dptr.sgl.type = (NVME_TRANSPORT_SGL_DATA_DESC << 4) |
2058 NVME_SGL_FMT_TRANSPORT_A;
e399441d
JS		 2059 sqe->rw.dptr.sgl.length = cpu_to_le32(data_len);
2060 sqe->rw.dptr.sgl.addr = 0;
2061
78a7ac26 2062 if (!(op->flags & FCOP_FLAGS_AEN)) {
e399441d
JS		 2063 ret = nvme_fc_map_data(ctrl, op->rq, op);
2064 if (ret < 0) {
e399441d
JS		 2065 nvme_cleanup_cmd(op->rq);
2066 nvme_fc_ctrl_put(ctrl);
fc17b653
CH		 2067 if (ret == -ENOMEM || ret == -EAGAIN)
2068 return BLK_STS_RESOURCE;
2069 return BLK_STS_IOERR;
e399441d
JS		 2070 }
2071 }
2072
2073 fc_dma_sync_single_for_device(ctrl->lport->dev, op->fcp_req.cmddma,
2074 sizeof(op->cmd_iu), DMA_TO_DEVICE);
2075
2076 atomic_set(&op->state, FCPOP_STATE_ACTIVE);
2077
78a7ac26 2078 if (!(op->flags & FCOP_FLAGS_AEN))
e399441d
JS		 2079 blk_mq_start_request(op->rq);
2080
2081 ret = ctrl->lport->ops->fcp_io(&ctrl->lport->localport,
2082 &ctrl->rport->remoteport,
2083 queue->lldd_handle, &op->fcp_req);
2084
2085 if (ret) {
8b25f351 2086 if (!(op->flags & FCOP_FLAGS_AEN))
e399441d 2087 nvme_fc_unmap_data(ctrl, op->rq, op);
e399441d
JS		 2088
2089 nvme_fc_ctrl_put(ctrl);
2090
8b25f351
JS		 2091 if (ctrl->rport->remoteport.port_state == FC_OBJSTATE_ONLINE &&
2092 ret != -EBUSY)
fc17b653 2093 return BLK_STS_IOERR;
e399441d 2094
8b25f351 2095 goto busy;
e399441d
JS		 2096 }
2097
fc17b653 2098 return BLK_STS_OK;
8b25f351
JS		 2099
2100busy:
2101 if (!(op->flags & FCOP_FLAGS_AEN) && queue->hctx)
2102 blk_mq_delay_run_hw_queue(queue->hctx, NVMEFC_QUEUE_DELAY);
2103
2104 return BLK_STS_RESOURCE;
e399441d
JS		 2105}
2106
fc17b653 2107static blk_status_t
e399441d
JS		 2108nvme_fc_queue_rq(struct blk_mq_hw_ctx *hctx,
2109 const struct blk_mq_queue_data *bd)
2110{
2111 struct nvme_ns *ns = hctx->queue->queuedata;
2112 struct nvme_fc_queue *queue = hctx->driver_data;
2113 struct nvme_fc_ctrl *ctrl = queue->ctrl;
2114 struct request *rq = bd->rq;
2115 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2116 struct nvme_fc_cmd_iu *cmdiu = &op->cmd_iu;
2117 struct nvme_command *sqe = &cmdiu->sqe;
2118 enum nvmefc_fcp_datadir io_dir;
2119 u32 data_len;
fc17b653 2120 blk_status_t ret;
e399441d
JS		 2121
2122 ret = nvme_setup_cmd(ns, rq, sqe);
2123 if (ret)
2124 return ret;
2125
b131c61d 2126 data_len = blk_rq_payload_bytes(rq);
e399441d
JS		 2127 if (data_len)
2128 io_dir = ((rq_data_dir(rq) == WRITE) ?
2129 NVMEFC_FCP_WRITE : NVMEFC_FCP_READ);
2130 else
2131 io_dir = NVMEFC_FCP_NODATA;
2132
2133 return nvme_fc_start_fcp_op(ctrl, queue, op, data_len, io_dir);
2134}
2135
2136static struct blk_mq_tags *
2137nvme_fc_tagset(struct nvme_fc_queue *queue)
2138{
2139 if (queue->qnum == 0)
2140 return queue->ctrl->admin_tag_set.tags[queue->qnum];
2141
2142 return queue->ctrl->tag_set.tags[queue->qnum - 1];
2143}
2144
2145static int
2146nvme_fc_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag)
2147
2148{
2149 struct nvme_fc_queue *queue = hctx->driver_data;
2150 struct nvme_fc_ctrl *ctrl = queue->ctrl;
2151 struct request *req;
2152 struct nvme_fc_fcp_op *op;
2153
2154 req = blk_mq_tag_to_rq(nvme_fc_tagset(queue), tag);
61bff8ef 2155 if (!req)
e399441d 2156 return 0;
e399441d
JS		 2157
2158 op = blk_mq_rq_to_pdu(req);
2159
2160 if ((atomic_read(&op->state) == FCPOP_STATE_ACTIVE) &&
2161 (ctrl->lport->ops->poll_queue))
2162 ctrl->lport->ops->poll_queue(&ctrl->lport->localport,
2163 queue->lldd_handle);
2164
2165 return ((atomic_read(&op->state) != FCPOP_STATE_ACTIVE));
2166}
2167
2168static void
2169nvme_fc_submit_async_event(struct nvme_ctrl *arg, int aer_idx)
2170{
2171 struct nvme_fc_ctrl *ctrl = to_fc_ctrl(arg);
2172 struct nvme_fc_fcp_op *aen_op;
61bff8ef
JS		 2173 unsigned long flags;
2174 bool terminating = false;
fc17b653 2175 blk_status_t ret;
e399441d
JS		 2176
2177 if (aer_idx > NVME_FC_NR_AEN_COMMANDS)
2178 return;
2179
61bff8ef
JS		 2180 spin_lock_irqsave(&ctrl->lock, flags);
2181 if (ctrl->flags & FCCTRL_TERMIO)
2182 terminating = true;
2183 spin_unlock_irqrestore(&ctrl->lock, flags);
2184
2185 if (terminating)
2186 return;
2187
e399441d
JS		 2188 aen_op = &ctrl->aen_ops[aer_idx];
2189
2190 ret = nvme_fc_start_fcp_op(ctrl, aen_op->queue, aen_op, 0,
2191 NVMEFC_FCP_NODATA);
2192 if (ret)
2193 dev_err(ctrl->ctrl.device,
2194 "failed async event work [%d]\n", aer_idx);
2195}
2196
2197static void
78a7ac26 2198__nvme_fc_final_op_cleanup(struct request *rq)
e399441d
JS		 2199{
2200 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2201 struct nvme_fc_ctrl *ctrl = op->ctrl;
e399441d 2202
78a7ac26
JS		 2203 atomic_set(&op->state, FCPOP_STATE_IDLE);
2204 op->flags &= ~(FCOP_FLAGS_TERMIO | FCOP_FLAGS_RELEASED |
2205 FCOP_FLAGS_COMPLETE);
e399441d 2206
e399441d 2207 nvme_fc_unmap_data(ctrl, rq, op);
77f02a7a 2208 nvme_complete_rq(rq);
e399441d
JS		 2209 nvme_fc_ctrl_put(ctrl);
2210
e399441d
JS		 2211}
2212
78a7ac26
JS		 2213static void
2214nvme_fc_complete_rq(struct request *rq)
2215{
2216 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(rq);
2217 struct nvme_fc_ctrl *ctrl = op->ctrl;
2218 unsigned long flags;
2219 bool completed = false;
2220
2221 /*
2222 * the core layer, on controller resets after calling
2223 * nvme_shutdown_ctrl(), calls complete_rq without our
2224 * calling blk_mq_complete_request(), thus there may still
2225 * be live i/o outstanding with the LLDD. Means transport has
2226 * to track complete calls vs fcpio_done calls to know what
2227 * path to take on completes and dones.
2228 */
2229 spin_lock_irqsave(&ctrl->lock, flags);
2230 if (op->flags & FCOP_FLAGS_COMPLETE)
2231 completed = true;
2232 else
2233 op->flags |= FCOP_FLAGS_RELEASED;
2234 spin_unlock_irqrestore(&ctrl->lock, flags);
2235
2236 if (completed)
2237 __nvme_fc_final_op_cleanup(rq);
2238}
2239
e399441d
JS		 2240/*
2241 * This routine is used by the transport when it needs to find active
2242 * io on a queue that is to be terminated. The transport uses
2243 * blk_mq_tagset_busy_itr() to find the busy requests, which then invoke
2244 * this routine to kill them on a 1 by 1 basis.
2245 *
2246 * As FC allocates FC exchange for each io, the transport must contact
2247 * the LLDD to terminate the exchange, thus releasing the FC exchange.
2248 * After terminating the exchange the LLDD will call the transport's
2249 * normal io done path for the request, but it will have an aborted
2250 * status. The done path will return the io request back to the block
2251 * layer with an error status.
2252 */
2253static void
2254nvme_fc_terminate_exchange(struct request *req, void *data, bool reserved)
2255{
2256 struct nvme_ctrl *nctrl = data;
2257 struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2258 struct nvme_fc_fcp_op *op = blk_mq_rq_to_pdu(req);
78a7ac26
JS		 2259 unsigned long flags;
2260 int status;
e399441d
JS		 2261
2262 if (!blk_mq_request_started(req))
2263 return;
2264
78a7ac26 2265 spin_lock_irqsave(&ctrl->lock, flags);
61bff8ef
JS		 2266 if (ctrl->flags & FCCTRL_TERMIO) {
2267 ctrl->iocnt++;
2268 op->flags |= FCOP_FLAGS_TERMIO;
2269 }
78a7ac26
JS		 2270 spin_unlock_irqrestore(&ctrl->lock, flags);
2271
e399441d 2272 status = __nvme_fc_abort_op(ctrl, op);
78a7ac26
JS		 2273 if (status) {
2274 /*
2275 * if __nvme_fc_abort_op failed the io wasn't
2276 * active. Thus this call path is running in
2277 * parallel to the io complete. Treat as non-error.
2278 */
2279
2280 /* back out the flags/counters */
2281 spin_lock_irqsave(&ctrl->lock, flags);
61bff8ef
JS		 2282 if (ctrl->flags & FCCTRL_TERMIO)
2283 ctrl->iocnt--;
78a7ac26
JS		 2284 op->flags &= ~FCOP_FLAGS_TERMIO;
2285 spin_unlock_irqrestore(&ctrl->lock, flags);
e399441d 2286 return;
78a7ac26 2287 }
e399441d
JS		 2288}
2289
78a7ac26 2290
61bff8ef
JS		 2291static const struct blk_mq_ops nvme_fc_mq_ops = {
2292 .queue_rq = nvme_fc_queue_rq,
2293 .complete = nvme_fc_complete_rq,
2294 .init_request = nvme_fc_init_request,
2295 .exit_request = nvme_fc_exit_request,
61bff8ef
JS		 2296 .init_hctx = nvme_fc_init_hctx,
2297 .poll = nvme_fc_poll,
2298 .timeout = nvme_fc_timeout,
2299};
e399441d 2300
61bff8ef
JS		 2301static int
2302nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)
e399441d 2303{
61bff8ef 2304 struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
7314183d 2305 unsigned int nr_io_queues;
61bff8ef 2306 int ret;
e399441d 2307
7314183d
SG		 2308 nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2309 ctrl->lport->ops->max_hw_queues);
2310 ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
61bff8ef
JS		 2311 if (ret) {
2312 dev_info(ctrl->ctrl.device,
2313 "set_queue_count failed: %d\n", ret);
2314 return ret;
2315 }
e399441d 2316
7314183d
SG		 2317 ctrl->ctrl.queue_count = nr_io_queues + 1;
2318 if (!nr_io_queues)
61bff8ef 2319 return 0;
e399441d 2320
61bff8ef 2321 nvme_fc_init_io_queues(ctrl);
e399441d 2322
61bff8ef
JS		 2323 memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set));
2324 ctrl->tag_set.ops = &nvme_fc_mq_ops;
2325 ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size;
2326 ctrl->tag_set.reserved_tags = 1; /* fabric connect */
2327 ctrl->tag_set.numa_node = NUMA_NO_NODE;
2328 ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2329 ctrl->tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
2330 (SG_CHUNK_SIZE *
2331 sizeof(struct scatterlist)) +
2332 ctrl->lport->ops->fcprqst_priv_sz;
2333 ctrl->tag_set.driver_data = ctrl;
d858e5f0 2334 ctrl->tag_set.nr_hw_queues = ctrl->ctrl.queue_count - 1;
61bff8ef 2335 ctrl->tag_set.timeout = NVME_IO_TIMEOUT;
e399441d 2336
61bff8ef
JS		 2337 ret = blk_mq_alloc_tag_set(&ctrl->tag_set);
2338 if (ret)
2339 return ret;
e399441d 2340
61bff8ef 2341 ctrl->ctrl.tagset = &ctrl->tag_set;
e399441d 2342
61bff8ef
JS		 2343 ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set);
2344 if (IS_ERR(ctrl->ctrl.connect_q)) {
2345 ret = PTR_ERR(ctrl->ctrl.connect_q);
2346 goto out_free_tag_set;
2347 }
e399441d 2348
61bff8ef 2349 ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
e399441d 2350 if (ret)
61bff8ef 2351 goto out_cleanup_blk_queue;
e399441d 2352
61bff8ef
JS		 2353 ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2354 if (ret)
2355 goto out_delete_hw_queues;
e399441d
JS		 2356
2357 return 0;
e399441d 2358
61bff8ef
JS		 2359out_delete_hw_queues:
2360 nvme_fc_delete_hw_io_queues(ctrl);
2361out_cleanup_blk_queue:
61bff8ef
JS		 2362 blk_cleanup_queue(ctrl->ctrl.connect_q);
2363out_free_tag_set:
2364 blk_mq_free_tag_set(&ctrl->tag_set);
2365 nvme_fc_free_io_queues(ctrl);
e399441d 2366
61bff8ef
JS		 2367 /* force put free routine to ignore io queues */
2368 ctrl->ctrl.tagset = NULL;
2369
2370 return ret;
2371}
e399441d
JS		 2372
2373static int
61bff8ef 2374nvme_fc_reinit_io_queues(struct nvme_fc_ctrl *ctrl)
e399441d
JS		 2375{
2376 struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
7314183d 2377 unsigned int nr_io_queues;
e399441d
JS		 2378 int ret;
2379
7314183d
SG		 2380 nr_io_queues = min(min(opts->nr_io_queues, num_online_cpus()),
2381 ctrl->lport->ops->max_hw_queues);
2382 ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues);
e399441d
JS		 2383 if (ret) {
2384 dev_info(ctrl->ctrl.device,
2385 "set_queue_count failed: %d\n", ret);
2386 return ret;
2387 }
2388
7314183d 2389 ctrl->ctrl.queue_count = nr_io_queues + 1;
61bff8ef 2390 /* check for io queues existing */
d858e5f0 2391 if (ctrl->ctrl.queue_count == 1)
e399441d
JS		 2392 return 0;
2393
e399441d
JS		 2394 nvme_fc_init_io_queues(ctrl);
2395
d352ae20 2396 ret = blk_mq_reinit_tagset(&ctrl->tag_set, nvme_fc_reinit_request);
e399441d 2397 if (ret)
61bff8ef 2398 goto out_free_io_queues;
e399441d
JS		 2399
2400 ret = nvme_fc_create_hw_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2401 if (ret)
61bff8ef 2402 goto out_free_io_queues;
e399441d
JS		 2403
2404 ret = nvme_fc_connect_io_queues(ctrl, ctrl->ctrl.opts->queue_size);
2405 if (ret)
2406 goto out_delete_hw_queues;
2407
cda5fd1a
SG		 2408 blk_mq_update_nr_hw_queues(&ctrl->tag_set, nr_io_queues);
2409
e399441d
JS		 2410 return 0;
2411
2412out_delete_hw_queues:
2413 nvme_fc_delete_hw_io_queues(ctrl);
61bff8ef 2414out_free_io_queues:
e399441d 2415 nvme_fc_free_io_queues(ctrl);
61bff8ef
JS		 2416 return ret;
2417}
e399441d 2418
61bff8ef
JS		 2419/*
2420 * This routine restarts the controller on the host side, and
2421 * on the link side, recreates the controller association.
2422 */
2423static int
2424nvme_fc_create_association(struct nvme_fc_ctrl *ctrl)
2425{
2426 struct nvmf_ctrl_options *opts = ctrl->ctrl.opts;
2427 u32 segs;
2428 int ret;
2429 bool changed;
2430
fdf9dfa8 2431 ++ctrl->ctrl.nr_reconnects;
61bff8ef
JS		 2432
2433 /*
2434 * Create the admin queue
2435 */
2436
2437 nvme_fc_init_queue(ctrl, 0, NVME_FC_AQ_BLKMQ_DEPTH);
2438
2439 ret = __nvme_fc_create_hw_queue(ctrl, &ctrl->queues[0], 0,
2440 NVME_FC_AQ_BLKMQ_DEPTH);
2441 if (ret)
2442 goto out_free_queue;
2443
2444 ret = nvme_fc_connect_admin_queue(ctrl, &ctrl->queues[0],
2445 NVME_FC_AQ_BLKMQ_DEPTH,
2446 (NVME_FC_AQ_BLKMQ_DEPTH / 4));
2447 if (ret)
2448 goto out_delete_hw_queue;
2449
2450 if (ctrl->ctrl.state != NVME_CTRL_NEW)
f9c5af5f 2451 blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);
61bff8ef
JS		 2452
2453 ret = nvmf_connect_admin_queue(&ctrl->ctrl);
2454 if (ret)
2455 goto out_disconnect_admin_queue;
2456
2457 /*
2458 * Check controller capabilities
2459 *
2460 * todo:- add code to check if ctrl attributes changed from
2461 * prior connection values
2462 */
2463
20d0dfe6 2464 ret = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->ctrl.cap);
61bff8ef
JS		 2465 if (ret) {
2466 dev_err(ctrl->ctrl.device,
2467 "prop_get NVME_REG_CAP failed\n");
2468 goto out_disconnect_admin_queue;
2469 }
2470
2471 ctrl->ctrl.sqsize =
20d0dfe6 2472 min_t(int, NVME_CAP_MQES(ctrl->ctrl.cap) + 1, ctrl->ctrl.sqsize);
61bff8ef 2473
20d0dfe6 2474 ret = nvme_enable_ctrl(&ctrl->ctrl, ctrl->ctrl.cap);
61bff8ef
JS		 2475 if (ret)
2476 goto out_disconnect_admin_queue;
2477
2478 segs = min_t(u32, NVME_FC_MAX_SEGMENTS,
2479 ctrl->lport->ops->max_sgl_segments);
2480 ctrl->ctrl.max_hw_sectors = (segs - 1) << (PAGE_SHIFT - 9);
2481
2482 ret = nvme_init_identify(&ctrl->ctrl);
2483 if (ret)
2484 goto out_disconnect_admin_queue;
2485
2486 /* sanity checks */
2487
2488 /* FC-NVME does not have other data in the capsule */
2489 if (ctrl->ctrl.icdoff) {
2490 dev_err(ctrl->ctrl.device, "icdoff %d is not supported!\n",
2491 ctrl->ctrl.icdoff);
2492 goto out_disconnect_admin_queue;
2493 }
2494
61bff8ef
JS		 2495 /* FC-NVME supports normal SGL Data Block Descriptors */
2496
2497 if (opts->queue_size > ctrl->ctrl.maxcmd) {
2498 /* warn if maxcmd is lower than queue_size */
2499 dev_warn(ctrl->ctrl.device,
2500 "queue_size %zu > ctrl maxcmd %u, reducing "
2501 "to queue_size\n",
2502 opts->queue_size, ctrl->ctrl.maxcmd);
2503 opts->queue_size = ctrl->ctrl.maxcmd;
2504 }
2505
2506 ret = nvme_fc_init_aen_ops(ctrl);
2507 if (ret)
2508 goto out_term_aen_ops;
2509
2510 /*
2511 * Create the io queues
2512 */
2513
d858e5f0 2514 if (ctrl->ctrl.queue_count > 1) {
61bff8ef
JS		 2515 if (ctrl->ctrl.state == NVME_CTRL_NEW)
2516 ret = nvme_fc_create_io_queues(ctrl);
2517 else
2518 ret = nvme_fc_reinit_io_queues(ctrl);
2519 if (ret)
2520 goto out_term_aen_ops;
2521 }
2522
2523 changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE);
2524 WARN_ON_ONCE(!changed);
2525
fdf9dfa8 2526 ctrl->ctrl.nr_reconnects = 0;
61bff8ef 2527
d09f2b45 2528 nvme_start_ctrl(&ctrl->ctrl);
61bff8ef
JS		 2529
2530 return 0; /* Success */
2531
2532out_term_aen_ops:
2533 nvme_fc_term_aen_ops(ctrl);
61bff8ef
JS		 2534out_disconnect_admin_queue:
2535 /* send a Disconnect(association) LS to fc-nvme target */
2536 nvme_fc_xmt_disconnect_assoc(ctrl);
2537out_delete_hw_queue:
2538 __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2539out_free_queue:
2540 nvme_fc_free_queue(&ctrl->queues[0]);
e399441d
JS		 2541
2542 return ret;
2543}
2544
61bff8ef
JS		 2545/*
2546 * This routine stops operation of the controller on the host side.
2547 * On the host os stack side: Admin and IO queues are stopped,
2548 * outstanding ios on them terminated via FC ABTS.
2549 * On the link side: the association is terminated.
2550 */
2551static void
2552nvme_fc_delete_association(struct nvme_fc_ctrl *ctrl)
2553{
2554 unsigned long flags;
2555
61bff8ef
JS		 2556 spin_lock_irqsave(&ctrl->lock, flags);
2557 ctrl->flags |= FCCTRL_TERMIO;
2558 ctrl->iocnt = 0;
2559 spin_unlock_irqrestore(&ctrl->lock, flags);
2560
2561 /*
2562 * If io queues are present, stop them and terminate all outstanding
2563 * ios on them. As FC allocates FC exchange for each io, the
2564 * transport must contact the LLDD to terminate the exchange,
2565 * thus releasing the FC exchange. We use blk_mq_tagset_busy_itr()
2566 * to tell us what io's are busy and invoke a transport routine
2567 * to kill them with the LLDD. After terminating the exchange
2568 * the LLDD will call the transport's normal io done path, but it
2569 * will have an aborted status. The done path will return the
2570 * io requests back to the block layer as part of normal completions
2571 * (but with error status).
2572 */
d858e5f0 2573 if (ctrl->ctrl.queue_count > 1) {
61bff8ef
JS		 2574 nvme_stop_queues(&ctrl->ctrl);
2575 blk_mq_tagset_busy_iter(&ctrl->tag_set,
2576 nvme_fc_terminate_exchange, &ctrl->ctrl);
2577 }
2578
2579 /*
2580 * Other transports, which don't have link-level contexts bound
2581 * to sqe's, would try to gracefully shutdown the controller by
2582 * writing the registers for shutdown and polling (call
2583 * nvme_shutdown_ctrl()). Given a bunch of i/o was potentially
2584 * just aborted and we will wait on those contexts, and given
2585 * there was no indication of how live the controlelr is on the
2586 * link, don't send more io to create more contexts for the
2587 * shutdown. Let the controller fail via keepalive failure if
2588 * its still present.
2589 */
2590
2591 /*
2592 * clean up the admin queue. Same thing as above.
2593 * use blk_mq_tagset_busy_itr() and the transport routine to
2594 * terminate the exchanges.
2595 */
f9c5af5f 2596 blk_mq_quiesce_queue(ctrl->ctrl.admin_q);
61bff8ef
JS		 2597 blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
2598 nvme_fc_terminate_exchange, &ctrl->ctrl);
2599
2600 /* kill the aens as they are a separate path */
2601 nvme_fc_abort_aen_ops(ctrl);
2602
2603 /* wait for all io that had to be aborted */
2604 spin_lock_irqsave(&ctrl->lock, flags);
36715cf4 2605 wait_event_lock_irq(ctrl->ioabort_wait, ctrl->iocnt == 0, ctrl->lock);
61bff8ef
JS		 2606 ctrl->flags &= ~FCCTRL_TERMIO;
2607 spin_unlock_irqrestore(&ctrl->lock, flags);
2608
2609 nvme_fc_term_aen_ops(ctrl);
2610
2611 /*
2612 * send a Disconnect(association) LS to fc-nvme target
2613 * Note: could have been sent at top of process, but
2614 * cleaner on link traffic if after the aborts complete.
2615 * Note: if association doesn't exist, association_id will be 0
2616 */
2617 if (ctrl->association_id)
2618 nvme_fc_xmt_disconnect_assoc(ctrl);
2619
2620 if (ctrl->ctrl.tagset) {
2621 nvme_fc_delete_hw_io_queues(ctrl);
2622 nvme_fc_free_io_queues(ctrl);
2623 }
2624
2625 __nvme_fc_delete_hw_queue(ctrl, &ctrl->queues[0], 0);
2626 nvme_fc_free_queue(&ctrl->queues[0]);
2627}
2628
2629static void
2630nvme_fc_delete_ctrl_work(struct work_struct *work)
2631{
2632 struct nvme_fc_ctrl *ctrl =
2633 container_of(work, struct nvme_fc_ctrl, delete_work);
2634
d86c4d8e 2635 cancel_work_sync(&ctrl->ctrl.reset_work);
61bff8ef 2636 cancel_delayed_work_sync(&ctrl->connect_work);
d09f2b45
SG		 2637 nvme_stop_ctrl(&ctrl->ctrl);
2638 nvme_remove_namespaces(&ctrl->ctrl);
61bff8ef
JS		 2639 /*
2640 * kill the association on the link side. this will block
2641 * waiting for io to terminate
2642 */
2643 nvme_fc_delete_association(ctrl);
2644
2645 /*
2646 * tear down the controller
a5321aa5
JS		 2647 * After the last reference on the nvme ctrl is removed,
2648 * the transport nvme_fc_nvme_ctrl_freed() callback will be
2649 * invoked. From there, the transport will tear down it's
2650 * logical queues and association.
61bff8ef
JS		 2651 */
2652 nvme_uninit_ctrl(&ctrl->ctrl);
2653
2654 nvme_put_ctrl(&ctrl->ctrl);
2655}
2656
5bbecdbc
JS		 2657static bool
2658__nvme_fc_schedule_delete_work(struct nvme_fc_ctrl *ctrl)
61bff8ef
JS		 2659{
2660 if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_DELETING))
5bbecdbc 2661 return true;
61bff8ef 2662
9a6327d2 2663 if (!queue_work(nvme_wq, &ctrl->delete_work))
5bbecdbc 2664 return true;
61bff8ef 2665
5bbecdbc
JS		 2666 return false;
2667}
2668
2669static int
2670__nvme_fc_del_ctrl(struct nvme_fc_ctrl *ctrl)
2671{
2672 return __nvme_fc_schedule_delete_work(ctrl) ? -EBUSY : 0;
61bff8ef
JS		 2673}
2674
2675/*
2676 * Request from nvme core layer to delete the controller
2677 */
2678static int
2679nvme_fc_del_nvme_ctrl(struct nvme_ctrl *nctrl)
2680{
2681 struct nvme_fc_ctrl *ctrl = to_fc_ctrl(nctrl);
2682 int ret;
2683
2684 if (!kref_get_unless_zero(&ctrl->ctrl.kref))
2685 return -EBUSY;
2686
2687 ret = __nvme_fc_del_ctrl(ctrl);
2688
2689 if (!ret)
9a6327d2 2690 flush_workqueue(nvme_wq);
61bff8ef
JS		 2691
2692 nvme_put_ctrl(&ctrl->ctrl);
2693
2694 return ret;
2695}
2696
5bbecdbc
JS		 2697static void
2698nvme_fc_reconnect_or_delete(struct nvme_fc_ctrl *ctrl, int status)
2699{
2700 /* If we are resetting/deleting then do nothing */
2701 if (ctrl->ctrl.state != NVME_CTRL_RECONNECTING) {
2702 WARN_ON_ONCE(ctrl->ctrl.state == NVME_CTRL_NEW ||
2703 ctrl->ctrl.state == NVME_CTRL_LIVE);
2704 return;
2705 }
2706
589ff775 2707 dev_info(ctrl->ctrl.device,
5bbecdbc
JS		 2708 "NVME-FC{%d}: reset: Reconnect attempt failed (%d)\n",
2709 ctrl->cnum, status);
2710
2711 if (nvmf_should_reconnect(&ctrl->ctrl)) {
2712 dev_info(ctrl->ctrl.device,
2713 "NVME-FC{%d}: Reconnect attempt in %d seconds.\n",
2714 ctrl->cnum, ctrl->ctrl.opts->reconnect_delay);
9a6327d2 2715 queue_delayed_work(nvme_wq, &ctrl->connect_work,
5bbecdbc
JS		 2716 ctrl->ctrl.opts->reconnect_delay * HZ);
2717 } else {
589ff775 2718 dev_warn(ctrl->ctrl.device,
5bbecdbc
JS		 2719 "NVME-FC{%d}: Max reconnect attempts (%d) "
2720 "reached. Removing controller\n",
fdf9dfa8 2721 ctrl->cnum, ctrl->ctrl.nr_reconnects);
5bbecdbc
JS		 2722 WARN_ON(__nvme_fc_schedule_delete_work(ctrl));
2723 }
2724}
2725
61bff8ef
JS		 2726static void
2727nvme_fc_reset_ctrl_work(struct work_struct *work)
2728{
2729 struct nvme_fc_ctrl *ctrl =
d86c4d8e 2730 container_of(work, struct nvme_fc_ctrl, ctrl.reset_work);
61bff8ef
JS		 2731 int ret;
2732
d09f2b45 2733 nvme_stop_ctrl(&ctrl->ctrl);
61bff8ef
JS		 2734 /* will block will waiting for io to terminate */
2735 nvme_fc_delete_association(ctrl);
2736
2737 ret = nvme_fc_create_association(ctrl);
5bbecdbc
JS		 2738 if (ret)
2739 nvme_fc_reconnect_or_delete(ctrl, ret);
2740 else
61bff8ef
JS		 2741 dev_info(ctrl->ctrl.device,
2742 "NVME-FC{%d}: controller reset complete\n", ctrl->cnum);
2743}
2744
61bff8ef
JS		 2745static const struct nvme_ctrl_ops nvme_fc_ctrl_ops = {
2746 .name = "fc",
2747 .module = THIS_MODULE,
d3d5b87d 2748 .flags = NVME_F_FABRICS,
61bff8ef
JS		 2749 .reg_read32 = nvmf_reg_read32,
2750 .reg_read64 = nvmf_reg_read64,
2751 .reg_write32 = nvmf_reg_write32,
61bff8ef
JS		 2752 .free_ctrl = nvme_fc_nvme_ctrl_freed,
2753 .submit_async_event = nvme_fc_submit_async_event,
2754 .delete_ctrl = nvme_fc_del_nvme_ctrl,
61bff8ef
JS		 2755 .get_address = nvmf_get_address,
2756};
2757
2758static void
2759nvme_fc_connect_ctrl_work(struct work_struct *work)
2760{
2761 int ret;
2762
2763 struct nvme_fc_ctrl *ctrl =
2764 container_of(to_delayed_work(work),
2765 struct nvme_fc_ctrl, connect_work);
2766
2767 ret = nvme_fc_create_association(ctrl);
5bbecdbc
JS		 2768 if (ret)
2769 nvme_fc_reconnect_or_delete(ctrl, ret);
2770 else
61bff8ef
JS		 2771 dev_info(ctrl->ctrl.device,
2772 "NVME-FC{%d}: controller reconnect complete\n",
2773 ctrl->cnum);
2774}
2775
2776
2777static const struct blk_mq_ops nvme_fc_admin_mq_ops = {
2778 .queue_rq = nvme_fc_queue_rq,
2779 .complete = nvme_fc_complete_rq,
76f983cb 2780 .init_request = nvme_fc_init_request,
61bff8ef 2781 .exit_request = nvme_fc_exit_request,
61bff8ef
JS		 2782 .init_hctx = nvme_fc_init_admin_hctx,
2783 .timeout = nvme_fc_timeout,
2784};
2785
e399441d
JS		 2786
2787static struct nvme_ctrl *
61bff8ef 2788nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,
e399441d
JS		 2789 struct nvme_fc_lport *lport, struct nvme_fc_rport *rport)
2790{
2791 struct nvme_fc_ctrl *ctrl;
2792 unsigned long flags;
2793 int ret, idx;
e399441d 2794
85e6a6ad
JS		 2795 if (!(rport->remoteport.port_role &
2796 (FC_PORT_ROLE_NVME_DISCOVERY | FC_PORT_ROLE_NVME_TARGET))) {
2797 ret = -EBADR;
2798 goto out_fail;
2799 }
2800
e399441d
JS		 2801 ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
2802 if (!ctrl) {
2803 ret = -ENOMEM;
2804 goto out_fail;
2805 }
2806
2807 idx = ida_simple_get(&nvme_fc_ctrl_cnt, 0, 0, GFP_KERNEL);
2808 if (idx < 0) {
2809 ret = -ENOSPC;
2810 goto out_free_ctrl;
2811 }
2812
2813 ctrl->ctrl.opts = opts;
2814 INIT_LIST_HEAD(&ctrl->ctrl_list);
e399441d
JS		 2815 ctrl->lport = lport;
2816 ctrl->rport = rport;
2817 ctrl->dev = lport->dev;
e399441d
JS		 2818 ctrl->cnum = idx;
2819
e399441d
JS		 2820 get_device(ctrl->dev);
2821 kref_init(&ctrl->ref);
2822
61bff8ef 2823 INIT_WORK(&ctrl->delete_work, nvme_fc_delete_ctrl_work);
d86c4d8e 2824 INIT_WORK(&ctrl->ctrl.reset_work, nvme_fc_reset_ctrl_work);
61bff8ef 2825 INIT_DELAYED_WORK(&ctrl->connect_work, nvme_fc_connect_ctrl_work);
e399441d
JS		 2826 spin_lock_init(&ctrl->lock);
2827
2828 /* io queue count */
d858e5f0 2829 ctrl->ctrl.queue_count = min_t(unsigned int,
e399441d
JS		 2830 opts->nr_io_queues,
2831 lport->ops->max_hw_queues);
d858e5f0 2832 ctrl->ctrl.queue_count++; /* +1 for admin queue */
e399441d
JS		 2833
2834 ctrl->ctrl.sqsize = opts->queue_size - 1;
2835 ctrl->ctrl.kato = opts->kato;
2836
2837 ret = -ENOMEM;
d858e5f0
SG		 2838 ctrl->queues = kcalloc(ctrl->ctrl.queue_count,
2839 sizeof(struct nvme_fc_queue), GFP_KERNEL);
e399441d 2840 if (!ctrl->queues)
61bff8ef 2841 goto out_free_ida;
e399441d 2842
61bff8ef
JS		 2843 memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set));
2844 ctrl->admin_tag_set.ops = &nvme_fc_admin_mq_ops;
2845 ctrl->admin_tag_set.queue_depth = NVME_FC_AQ_BLKMQ_DEPTH;
2846 ctrl->admin_tag_set.reserved_tags = 2; /* fabric connect + Keep-Alive */
2847 ctrl->admin_tag_set.numa_node = NUMA_NO_NODE;
2848 ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_fc_fcp_op) +
2849 (SG_CHUNK_SIZE *
2850 sizeof(struct scatterlist)) +
2851 ctrl->lport->ops->fcprqst_priv_sz;
2852 ctrl->admin_tag_set.driver_data = ctrl;
2853 ctrl->admin_tag_set.nr_hw_queues = 1;
2854 ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT;
e399441d 2855
61bff8ef 2856 ret = blk_mq_alloc_tag_set(&ctrl->admin_tag_set);
e399441d 2857 if (ret)
61bff8ef 2858 goto out_free_queues;
34b6c231 2859 ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set;
e399441d 2860
61bff8ef
JS		 2861 ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);
2862 if (IS_ERR(ctrl->ctrl.admin_q)) {
2863 ret = PTR_ERR(ctrl->ctrl.admin_q);
2864 goto out_free_admin_tag_set;
e399441d
JS		 2865 }
2866
61bff8ef
JS		 2867 /*
2868 * Would have been nice to init io queues tag set as well.
2869 * However, we require interaction from the controller
2870 * for max io queue count before we can do so.
2871 * Defer this to the connect path.
2872 */
e399441d 2873
61bff8ef
JS		 2874 ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_fc_ctrl_ops, 0);
2875 if (ret)
2876 goto out_cleanup_admin_q;
e399441d 2877
61bff8ef 2878 /* at this point, teardown path changes to ref counting on nvme ctrl */
e399441d
JS		 2879
2880 spin_lock_irqsave(&rport->lock, flags);
2881 list_add_tail(&ctrl->ctrl_list, &rport->ctrl_list);
2882 spin_unlock_irqrestore(&rport->lock, flags);
2883
61bff8ef
JS		 2884 ret = nvme_fc_create_association(ctrl);
2885 if (ret) {
de41447a 2886 ctrl->ctrl.opts = NULL;
61bff8ef
JS		 2887 /* initiate nvme ctrl ref counting teardown */
2888 nvme_uninit_ctrl(&ctrl->ctrl);
24b7f059 2889 nvme_put_ctrl(&ctrl->ctrl);
61bff8ef 2890
0b5a7669
JS		 2891 /* Remove core ctrl ref. */
2892 nvme_put_ctrl(&ctrl->ctrl);
2893
61bff8ef
JS		 2894 /* as we're past the point where we transition to the ref
2895 * counting teardown path, if we return a bad pointer here,
2896 * the calling routine, thinking it's prior to the
2897 * transition, will do an rport put. Since the teardown
2898 * path also does a rport put, we do an extra get here to
2899 * so proper order/teardown happens.
2900 */
2901 nvme_fc_rport_get(rport);
2902
2903 if (ret > 0)
2904 ret = -EIO;
2905 return ERR_PTR(ret);
e399441d
JS		 2906 }
2907
2cb657bc
JS		 2908 kref_get(&ctrl->ctrl.kref);
2909
61bff8ef
JS		 2910 dev_info(ctrl->ctrl.device,
2911 "NVME-FC{%d}: new ctrl: NQN \"%s\"\n",
2912 ctrl->cnum, ctrl->ctrl.opts->subsysnqn);
e399441d 2913
61bff8ef 2914 return &ctrl->ctrl;
e399441d 2915
61bff8ef
JS		 2916out_cleanup_admin_q:
2917 blk_cleanup_queue(ctrl->ctrl.admin_q);
2918out_free_admin_tag_set:
2919 blk_mq_free_tag_set(&ctrl->admin_tag_set);
2920out_free_queues:
2921 kfree(ctrl->queues);
e399441d 2922out_free_ida:
61bff8ef 2923 put_device(ctrl->dev);
e399441d
JS		 2924 ida_simple_remove(&nvme_fc_ctrl_cnt, ctrl->cnum);
2925out_free_ctrl:
2926 kfree(ctrl);
2927out_fail:
e399441d
JS		 2928 /* exit via here doesn't follow ctlr ref points */
2929 return ERR_PTR(ret);
2930}
2931
e399441d
JS		 2932
2933struct nvmet_fc_traddr {
2934 u64 nn;
2935 u64 pn;
2936};
2937
e399441d 2938static int
9c5358e1 2939__nvme_fc_parse_u64(substring_t *sstr, u64 *val)
e399441d 2940{
e399441d
JS		 2941 u64 token64;
2942
9c5358e1
JS		 2943 if (match_u64(sstr, &token64))
2944 return -EINVAL;
2945 *val = token64;
e399441d 2946
9c5358e1
JS		 2947 return 0;
2948}
e399441d 2949
9c5358e1
JS		 2950/*
2951 * This routine validates and extracts the WWN's from the TRADDR string.
2952 * As kernel parsers need the 0x to determine number base, universally
2953 * build string to parse with 0x prefix before parsing name strings.
2954 */
2955static int
2956nvme_fc_parse_traddr(struct nvmet_fc_traddr *traddr, char *buf, size_t blen)
2957{
2958 char name[2 + NVME_FC_TRADDR_HEXNAMELEN + 1];
2959 substring_t wwn = { name, &name[sizeof(name)-1] };
2960 int nnoffset, pnoffset;
2961
2962 /* validate it string one of the 2 allowed formats */
2963 if (strnlen(buf, blen) == NVME_FC_TRADDR_MAXLENGTH &&
2964 !strncmp(buf, "nn-0x", NVME_FC_TRADDR_OXNNLEN) &&
2965 !strncmp(&buf[NVME_FC_TRADDR_MAX_PN_OFFSET],
2966 "pn-0x", NVME_FC_TRADDR_OXNNLEN)) {
2967 nnoffset = NVME_FC_TRADDR_OXNNLEN;
2968 pnoffset = NVME_FC_TRADDR_MAX_PN_OFFSET +
2969 NVME_FC_TRADDR_OXNNLEN;
2970 } else if ((strnlen(buf, blen) == NVME_FC_TRADDR_MINLENGTH &&
2971 !strncmp(buf, "nn-", NVME_FC_TRADDR_NNLEN) &&
2972 !strncmp(&buf[NVME_FC_TRADDR_MIN_PN_OFFSET],
2973 "pn-", NVME_FC_TRADDR_NNLEN))) {
2974 nnoffset = NVME_FC_TRADDR_NNLEN;
2975 pnoffset = NVME_FC_TRADDR_MIN_PN_OFFSET + NVME_FC_TRADDR_NNLEN;
2976 } else
2977 goto out_einval;
e399441d 2978
9c5358e1
JS		 2979 name[0] = '0';
2980 name[1] = 'x';
2981 name[2 + NVME_FC_TRADDR_HEXNAMELEN] = 0;
2982
2983 memcpy(&name[2], &buf[nnoffset], NVME_FC_TRADDR_HEXNAMELEN);
2984 if (__nvme_fc_parse_u64(&wwn, &traddr->nn))
2985 goto out_einval;
2986
2987 memcpy(&name[2], &buf[pnoffset], NVME_FC_TRADDR_HEXNAMELEN);
2988 if (__nvme_fc_parse_u64(&wwn, &traddr->pn))
2989 goto out_einval;
2990
2991 return 0;
2992
2993out_einval:
2994 pr_warn("%s: bad traddr string\n", __func__);
2995 return -EINVAL;
e399441d
JS		 2996}
2997
2998static struct nvme_ctrl *
2999nvme_fc_create_ctrl(struct device *dev, struct nvmf_ctrl_options *opts)
3000{
3001 struct nvme_fc_lport *lport;
3002 struct nvme_fc_rport *rport;
61bff8ef 3003 struct nvme_ctrl *ctrl;
e399441d
JS		 3004 struct nvmet_fc_traddr laddr = { 0L, 0L };
3005 struct nvmet_fc_traddr raddr = { 0L, 0L };
3006 unsigned long flags;
3007 int ret;
3008
9c5358e1 3009 ret = nvme_fc_parse_traddr(&raddr, opts->traddr, NVMF_TRADDR_SIZE);
e399441d
JS		 3010 if (ret || !raddr.nn || !raddr.pn)
3011 return ERR_PTR(-EINVAL);
3012
9c5358e1 3013 ret = nvme_fc_parse_traddr(&laddr, opts->host_traddr, NVMF_TRADDR_SIZE);
e399441d
JS		 3014 if (ret || !laddr.nn || !laddr.pn)
3015 return ERR_PTR(-EINVAL);
3016
3017 /* find the host and remote ports to connect together */
3018 spin_lock_irqsave(&nvme_fc_lock, flags);
3019 list_for_each_entry(lport, &nvme_fc_lport_list, port_list) {
3020 if (lport->localport.node_name != laddr.nn ||
3021 lport->localport.port_name != laddr.pn)
3022 continue;
3023
3024 list_for_each_entry(rport, &lport->endp_list, endp_list) {
3025 if (rport->remoteport.node_name != raddr.nn ||
3026 rport->remoteport.port_name != raddr.pn)
3027 continue;
3028
3029 /* if fail to get reference fall through. Will error */
3030 if (!nvme_fc_rport_get(rport))
3031 break;
3032
3033 spin_unlock_irqrestore(&nvme_fc_lock, flags);
3034
61bff8ef
JS		 3035 ctrl = nvme_fc_init_ctrl(dev, opts, lport, rport);
3036 if (IS_ERR(ctrl))
3037 nvme_fc_rport_put(rport);
3038 return ctrl;
e399441d
JS		 3039 }
3040 }
3041 spin_unlock_irqrestore(&nvme_fc_lock, flags);
3042
3043 return ERR_PTR(-ENOENT);
3044}
3045
3046
3047static struct nvmf_transport_ops nvme_fc_transport = {
3048 .name = "fc",
3049 .required_opts = NVMF_OPT_TRADDR | NVMF_OPT_HOST_TRADDR,
5bbecdbc 3050 .allowed_opts = NVMF_OPT_RECONNECT_DELAY | NVMF_OPT_CTRL_LOSS_TMO,
e399441d
JS		 3051 .create_ctrl = nvme_fc_create_ctrl,
3052};
3053
3054static int __init nvme_fc_init_module(void)
3055{
5f568556
JS		 3056 int ret;
3057
3058 /*
3059 * NOTE:
3060 * It is expected that in the future the kernel will combine
3061 * the FC-isms that are currently under scsi and now being
3062 * added to by NVME into a new standalone FC class. The SCSI
3063 * and NVME protocols and their devices would be under this
3064 * new FC class.
3065 *
3066 * As we need something to post FC-specific udev events to,
3067 * specifically for nvme probe events, start by creating the
3068 * new device class. When the new standalone FC class is
3069 * put in place, this code will move to a more generic
3070 * location for the class.
3071 */
3072 fc_class = class_create(THIS_MODULE, "fc");
3073 if (IS_ERR(fc_class)) {
3074 pr_err("couldn't register class fc\n");
3075 return PTR_ERR(fc_class);
3076 }
3077
3078 /*
3079 * Create a device for the FC-centric udev events
3080 */
3081 fc_udev_device = device_create(fc_class, NULL, MKDEV(0, 0), NULL,
3082 "fc_udev_device");
3083 if (IS_ERR(fc_udev_device)) {
3084 pr_err("couldn't create fc_udev device!\n");
3085 ret = PTR_ERR(fc_udev_device);
3086 goto out_destroy_class;
3087 }
3088
3089 ret = nvmf_register_transport(&nvme_fc_transport);
3090 if (ret)
3091 goto out_destroy_device;
3092
3093 return 0;
3094
3095out_destroy_device:
3096 device_destroy(fc_class, MKDEV(0, 0));
3097out_destroy_class:
3098 class_destroy(fc_class);
3099 return ret;
e399441d
JS		 3100}
3101
3102static void __exit nvme_fc_exit_module(void)
3103{
3104 /* sanity check - all lports should be removed */
3105 if (!list_empty(&nvme_fc_lport_list))
3106 pr_warn("%s: localport list not empty\n", __func__);
3107
3108 nvmf_unregister_transport(&nvme_fc_transport);
3109
e399441d
JS		 3110 ida_destroy(&nvme_fc_local_port_cnt);
3111 ida_destroy(&nvme_fc_ctrl_cnt);
5f568556
JS		 3112
3113 device_destroy(fc_class, MKDEV(0, 0));
3114 class_destroy(fc_class);
e399441d
JS		 3115}
3116
3117module_init(nvme_fc_init_module);
3118module_exit(nvme_fc_exit_module);
3119
3120MODULE_LICENSE("GPL v2");
Linux 6.x block layer and io_uring tree(s)