xref: /linux/drivers/nvme/host/core.c (revision 6607aa6f6b68fc9b5955755f1b1be125cf2a9d03)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NVM Express device driver
4  * Copyright (c) 2011-2014, Intel Corporation.
5  */
6 
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/compat.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/hdreg.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/list_sort.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24 
25 #include "nvme.h"
26 #include "fabrics.h"
27 
28 #define CREATE_TRACE_POINTS
29 #include "trace.h"
30 
31 #define NVME_MINORS		(1U << MINORBITS)
32 
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
37 
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
42 
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
46 
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
50 
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 		 "max power saving latency for new devices; use PM QOS to change per device");
55 
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
59 
60 static bool streams;
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
63 
64 /*
65  * nvme_wq - hosts nvme related works that are not reset or delete
66  * nvme_reset_wq - hosts nvme reset works
67  * nvme_delete_wq - hosts nvme delete works
68  *
69  * nvme_wq will host works such as scan, aen handling, fw activation,
70  * keep-alive, periodic reconnects etc. nvme_reset_wq
71  * runs reset works which also flush works hosted on nvme_wq for
72  * serialization purposes. nvme_delete_wq host controller deletion
73  * works which flush reset works for serialization.
74  */
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
77 
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
80 
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
83 
84 static LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
86 
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_ctrl_base_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
91 
92 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
93 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
94 					   unsigned nsid);
95 
96 /*
97  * Prepare a queue for teardown.
98  *
99  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
100  * the capacity to 0 after that to avoid blocking dispatchers that may be
101  * holding bd_butex.  This will end buffered writers dirtying pages that can't
102  * be synced.
103  */
104 static void nvme_set_queue_dying(struct nvme_ns *ns)
105 {
106 	if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
107 		return;
108 
109 	blk_set_queue_dying(ns->queue);
110 	blk_mq_unquiesce_queue(ns->queue);
111 
112 	set_capacity_and_notify(ns->disk, 0);
113 }
114 
115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
116 {
117 	/*
118 	 * Only new queue scan work when admin and IO queues are both alive
119 	 */
120 	if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
121 		queue_work(nvme_wq, &ctrl->scan_work);
122 }
123 
124 /*
125  * Use this function to proceed with scheduling reset_work for a controller
126  * that had previously been set to the resetting state. This is intended for
127  * code paths that can't be interrupted by other reset attempts. A hot removal
128  * may prevent this from succeeding.
129  */
130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
131 {
132 	if (ctrl->state != NVME_CTRL_RESETTING)
133 		return -EBUSY;
134 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
135 		return -EBUSY;
136 	return 0;
137 }
138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
139 
140 static void nvme_failfast_work(struct work_struct *work)
141 {
142 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
143 			struct nvme_ctrl, failfast_work);
144 
145 	if (ctrl->state != NVME_CTRL_CONNECTING)
146 		return;
147 
148 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
149 	dev_info(ctrl->device, "failfast expired\n");
150 	nvme_kick_requeue_lists(ctrl);
151 }
152 
153 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
154 {
155 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
156 		return;
157 
158 	schedule_delayed_work(&ctrl->failfast_work,
159 			      ctrl->opts->fast_io_fail_tmo * HZ);
160 }
161 
162 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
163 {
164 	if (!ctrl->opts)
165 		return;
166 
167 	cancel_delayed_work_sync(&ctrl->failfast_work);
168 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
169 }
170 
171 
172 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
173 {
174 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
175 		return -EBUSY;
176 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
177 		return -EBUSY;
178 	return 0;
179 }
180 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
181 
182 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
183 {
184 	int ret;
185 
186 	ret = nvme_reset_ctrl(ctrl);
187 	if (!ret) {
188 		flush_work(&ctrl->reset_work);
189 		if (ctrl->state != NVME_CTRL_LIVE)
190 			ret = -ENETRESET;
191 	}
192 
193 	return ret;
194 }
195 
196 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
197 {
198 	dev_info(ctrl->device,
199 		 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
200 
201 	flush_work(&ctrl->reset_work);
202 	nvme_stop_ctrl(ctrl);
203 	nvme_remove_namespaces(ctrl);
204 	ctrl->ops->delete_ctrl(ctrl);
205 	nvme_uninit_ctrl(ctrl);
206 }
207 
208 static void nvme_delete_ctrl_work(struct work_struct *work)
209 {
210 	struct nvme_ctrl *ctrl =
211 		container_of(work, struct nvme_ctrl, delete_work);
212 
213 	nvme_do_delete_ctrl(ctrl);
214 }
215 
216 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
217 {
218 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
219 		return -EBUSY;
220 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
221 		return -EBUSY;
222 	return 0;
223 }
224 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
225 
226 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
227 {
228 	/*
229 	 * Keep a reference until nvme_do_delete_ctrl() complete,
230 	 * since ->delete_ctrl can free the controller.
231 	 */
232 	nvme_get_ctrl(ctrl);
233 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
234 		nvme_do_delete_ctrl(ctrl);
235 	nvme_put_ctrl(ctrl);
236 }
237 
238 static blk_status_t nvme_error_status(u16 status)
239 {
240 	switch (status & 0x7ff) {
241 	case NVME_SC_SUCCESS:
242 		return BLK_STS_OK;
243 	case NVME_SC_CAP_EXCEEDED:
244 		return BLK_STS_NOSPC;
245 	case NVME_SC_LBA_RANGE:
246 	case NVME_SC_CMD_INTERRUPTED:
247 	case NVME_SC_NS_NOT_READY:
248 		return BLK_STS_TARGET;
249 	case NVME_SC_BAD_ATTRIBUTES:
250 	case NVME_SC_ONCS_NOT_SUPPORTED:
251 	case NVME_SC_INVALID_OPCODE:
252 	case NVME_SC_INVALID_FIELD:
253 	case NVME_SC_INVALID_NS:
254 		return BLK_STS_NOTSUPP;
255 	case NVME_SC_WRITE_FAULT:
256 	case NVME_SC_READ_ERROR:
257 	case NVME_SC_UNWRITTEN_BLOCK:
258 	case NVME_SC_ACCESS_DENIED:
259 	case NVME_SC_READ_ONLY:
260 	case NVME_SC_COMPARE_FAILED:
261 		return BLK_STS_MEDIUM;
262 	case NVME_SC_GUARD_CHECK:
263 	case NVME_SC_APPTAG_CHECK:
264 	case NVME_SC_REFTAG_CHECK:
265 	case NVME_SC_INVALID_PI:
266 		return BLK_STS_PROTECTION;
267 	case NVME_SC_RESERVATION_CONFLICT:
268 		return BLK_STS_NEXUS;
269 	case NVME_SC_HOST_PATH_ERROR:
270 		return BLK_STS_TRANSPORT;
271 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
272 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
273 	case NVME_SC_ZONE_TOO_MANY_OPEN:
274 		return BLK_STS_ZONE_OPEN_RESOURCE;
275 	default:
276 		return BLK_STS_IOERR;
277 	}
278 }
279 
280 static void nvme_retry_req(struct request *req)
281 {
282 	struct nvme_ns *ns = req->q->queuedata;
283 	unsigned long delay = 0;
284 	u16 crd;
285 
286 	/* The mask and shift result must be <= 3 */
287 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
288 	if (ns && crd)
289 		delay = ns->ctrl->crdt[crd - 1] * 100;
290 
291 	nvme_req(req)->retries++;
292 	blk_mq_requeue_request(req, false);
293 	blk_mq_delay_kick_requeue_list(req->q, delay);
294 }
295 
296 enum nvme_disposition {
297 	COMPLETE,
298 	RETRY,
299 	FAILOVER,
300 };
301 
302 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
303 {
304 	if (likely(nvme_req(req)->status == 0))
305 		return COMPLETE;
306 
307 	if (blk_noretry_request(req) ||
308 	    (nvme_req(req)->status & NVME_SC_DNR) ||
309 	    nvme_req(req)->retries >= nvme_max_retries)
310 		return COMPLETE;
311 
312 	if (req->cmd_flags & REQ_NVME_MPATH) {
313 		if (nvme_is_path_error(nvme_req(req)->status) ||
314 		    blk_queue_dying(req->q))
315 			return FAILOVER;
316 	} else {
317 		if (blk_queue_dying(req->q))
318 			return COMPLETE;
319 	}
320 
321 	return RETRY;
322 }
323 
324 static inline void nvme_end_req(struct request *req)
325 {
326 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
327 
328 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
329 	    req_op(req) == REQ_OP_ZONE_APPEND)
330 		req->__sector = nvme_lba_to_sect(req->q->queuedata,
331 			le64_to_cpu(nvme_req(req)->result.u64));
332 
333 	nvme_trace_bio_complete(req);
334 	blk_mq_end_request(req, status);
335 }
336 
337 void nvme_complete_rq(struct request *req)
338 {
339 	trace_nvme_complete_rq(req);
340 	nvme_cleanup_cmd(req);
341 
342 	if (nvme_req(req)->ctrl->kas)
343 		nvme_req(req)->ctrl->comp_seen = true;
344 
345 	switch (nvme_decide_disposition(req)) {
346 	case COMPLETE:
347 		nvme_end_req(req);
348 		return;
349 	case RETRY:
350 		nvme_retry_req(req);
351 		return;
352 	case FAILOVER:
353 		nvme_failover_req(req);
354 		return;
355 	}
356 }
357 EXPORT_SYMBOL_GPL(nvme_complete_rq);
358 
359 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
360 {
361 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
362 				"Cancelling I/O %d", req->tag);
363 
364 	/* don't abort one completed request */
365 	if (blk_mq_request_completed(req))
366 		return true;
367 
368 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
369 	blk_mq_complete_request(req);
370 	return true;
371 }
372 EXPORT_SYMBOL_GPL(nvme_cancel_request);
373 
374 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
375 		enum nvme_ctrl_state new_state)
376 {
377 	enum nvme_ctrl_state old_state;
378 	unsigned long flags;
379 	bool changed = false;
380 
381 	spin_lock_irqsave(&ctrl->lock, flags);
382 
383 	old_state = ctrl->state;
384 	switch (new_state) {
385 	case NVME_CTRL_LIVE:
386 		switch (old_state) {
387 		case NVME_CTRL_NEW:
388 		case NVME_CTRL_RESETTING:
389 		case NVME_CTRL_CONNECTING:
390 			changed = true;
391 			fallthrough;
392 		default:
393 			break;
394 		}
395 		break;
396 	case NVME_CTRL_RESETTING:
397 		switch (old_state) {
398 		case NVME_CTRL_NEW:
399 		case NVME_CTRL_LIVE:
400 			changed = true;
401 			fallthrough;
402 		default:
403 			break;
404 		}
405 		break;
406 	case NVME_CTRL_CONNECTING:
407 		switch (old_state) {
408 		case NVME_CTRL_NEW:
409 		case NVME_CTRL_RESETTING:
410 			changed = true;
411 			fallthrough;
412 		default:
413 			break;
414 		}
415 		break;
416 	case NVME_CTRL_DELETING:
417 		switch (old_state) {
418 		case NVME_CTRL_LIVE:
419 		case NVME_CTRL_RESETTING:
420 		case NVME_CTRL_CONNECTING:
421 			changed = true;
422 			fallthrough;
423 		default:
424 			break;
425 		}
426 		break;
427 	case NVME_CTRL_DELETING_NOIO:
428 		switch (old_state) {
429 		case NVME_CTRL_DELETING:
430 		case NVME_CTRL_DEAD:
431 			changed = true;
432 			fallthrough;
433 		default:
434 			break;
435 		}
436 		break;
437 	case NVME_CTRL_DEAD:
438 		switch (old_state) {
439 		case NVME_CTRL_DELETING:
440 			changed = true;
441 			fallthrough;
442 		default:
443 			break;
444 		}
445 		break;
446 	default:
447 		break;
448 	}
449 
450 	if (changed) {
451 		ctrl->state = new_state;
452 		wake_up_all(&ctrl->state_wq);
453 	}
454 
455 	spin_unlock_irqrestore(&ctrl->lock, flags);
456 	if (!changed)
457 		return false;
458 
459 	if (ctrl->state == NVME_CTRL_LIVE) {
460 		if (old_state == NVME_CTRL_CONNECTING)
461 			nvme_stop_failfast_work(ctrl);
462 		nvme_kick_requeue_lists(ctrl);
463 	} else if (ctrl->state == NVME_CTRL_CONNECTING &&
464 		old_state == NVME_CTRL_RESETTING) {
465 		nvme_start_failfast_work(ctrl);
466 	}
467 	return changed;
468 }
469 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
470 
471 /*
472  * Returns true for sink states that can't ever transition back to live.
473  */
474 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
475 {
476 	switch (ctrl->state) {
477 	case NVME_CTRL_NEW:
478 	case NVME_CTRL_LIVE:
479 	case NVME_CTRL_RESETTING:
480 	case NVME_CTRL_CONNECTING:
481 		return false;
482 	case NVME_CTRL_DELETING:
483 	case NVME_CTRL_DELETING_NOIO:
484 	case NVME_CTRL_DEAD:
485 		return true;
486 	default:
487 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
488 		return true;
489 	}
490 }
491 
492 /*
493  * Waits for the controller state to be resetting, or returns false if it is
494  * not possible to ever transition to that state.
495  */
496 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
497 {
498 	wait_event(ctrl->state_wq,
499 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
500 		   nvme_state_terminal(ctrl));
501 	return ctrl->state == NVME_CTRL_RESETTING;
502 }
503 EXPORT_SYMBOL_GPL(nvme_wait_reset);
504 
505 static void nvme_free_ns_head(struct kref *ref)
506 {
507 	struct nvme_ns_head *head =
508 		container_of(ref, struct nvme_ns_head, ref);
509 
510 	nvme_mpath_remove_disk(head);
511 	ida_simple_remove(&head->subsys->ns_ida, head->instance);
512 	cleanup_srcu_struct(&head->srcu);
513 	nvme_put_subsystem(head->subsys);
514 	kfree(head);
515 }
516 
517 static void nvme_put_ns_head(struct nvme_ns_head *head)
518 {
519 	kref_put(&head->ref, nvme_free_ns_head);
520 }
521 
522 static void nvme_free_ns(struct kref *kref)
523 {
524 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
525 
526 	if (ns->ndev)
527 		nvme_nvm_unregister(ns);
528 
529 	put_disk(ns->disk);
530 	nvme_put_ns_head(ns->head);
531 	nvme_put_ctrl(ns->ctrl);
532 	kfree(ns);
533 }
534 
535 void nvme_put_ns(struct nvme_ns *ns)
536 {
537 	kref_put(&ns->kref, nvme_free_ns);
538 }
539 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
540 
541 static inline void nvme_clear_nvme_request(struct request *req)
542 {
543 	if (!(req->rq_flags & RQF_DONTPREP)) {
544 		nvme_req(req)->retries = 0;
545 		nvme_req(req)->flags = 0;
546 		req->rq_flags |= RQF_DONTPREP;
547 	}
548 }
549 
550 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
551 {
552 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
553 }
554 
555 static inline void nvme_init_request(struct request *req,
556 		struct nvme_command *cmd)
557 {
558 	if (req->q->queuedata)
559 		req->timeout = NVME_IO_TIMEOUT;
560 	else /* no queuedata implies admin queue */
561 		req->timeout = NVME_ADMIN_TIMEOUT;
562 
563 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
564 	nvme_clear_nvme_request(req);
565 	nvme_req(req)->cmd = cmd;
566 }
567 
568 struct request *nvme_alloc_request(struct request_queue *q,
569 		struct nvme_command *cmd, blk_mq_req_flags_t flags)
570 {
571 	struct request *req;
572 
573 	req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
574 	if (!IS_ERR(req))
575 		nvme_init_request(req, cmd);
576 	return req;
577 }
578 EXPORT_SYMBOL_GPL(nvme_alloc_request);
579 
580 static struct request *nvme_alloc_request_qid(struct request_queue *q,
581 		struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
582 {
583 	struct request *req;
584 
585 	req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
586 			qid ? qid - 1 : 0);
587 	if (!IS_ERR(req))
588 		nvme_init_request(req, cmd);
589 	return req;
590 }
591 
592 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
593 {
594 	struct nvme_command c;
595 
596 	memset(&c, 0, sizeof(c));
597 
598 	c.directive.opcode = nvme_admin_directive_send;
599 	c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
600 	c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
601 	c.directive.dtype = NVME_DIR_IDENTIFY;
602 	c.directive.tdtype = NVME_DIR_STREAMS;
603 	c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
604 
605 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
606 }
607 
608 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
609 {
610 	return nvme_toggle_streams(ctrl, false);
611 }
612 
613 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
614 {
615 	return nvme_toggle_streams(ctrl, true);
616 }
617 
618 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
619 				  struct streams_directive_params *s, u32 nsid)
620 {
621 	struct nvme_command c;
622 
623 	memset(&c, 0, sizeof(c));
624 	memset(s, 0, sizeof(*s));
625 
626 	c.directive.opcode = nvme_admin_directive_recv;
627 	c.directive.nsid = cpu_to_le32(nsid);
628 	c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
629 	c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
630 	c.directive.dtype = NVME_DIR_STREAMS;
631 
632 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
633 }
634 
635 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
636 {
637 	struct streams_directive_params s;
638 	int ret;
639 
640 	if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
641 		return 0;
642 	if (!streams)
643 		return 0;
644 
645 	ret = nvme_enable_streams(ctrl);
646 	if (ret)
647 		return ret;
648 
649 	ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
650 	if (ret)
651 		goto out_disable_stream;
652 
653 	ctrl->nssa = le16_to_cpu(s.nssa);
654 	if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
655 		dev_info(ctrl->device, "too few streams (%u) available\n",
656 					ctrl->nssa);
657 		goto out_disable_stream;
658 	}
659 
660 	ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
661 	dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
662 	return 0;
663 
664 out_disable_stream:
665 	nvme_disable_streams(ctrl);
666 	return ret;
667 }
668 
669 /*
670  * Check if 'req' has a write hint associated with it. If it does, assign
671  * a valid namespace stream to the write.
672  */
673 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
674 				     struct request *req, u16 *control,
675 				     u32 *dsmgmt)
676 {
677 	enum rw_hint streamid = req->write_hint;
678 
679 	if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
680 		streamid = 0;
681 	else {
682 		streamid--;
683 		if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
684 			return;
685 
686 		*control |= NVME_RW_DTYPE_STREAMS;
687 		*dsmgmt |= streamid << 16;
688 	}
689 
690 	if (streamid < ARRAY_SIZE(req->q->write_hints))
691 		req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
692 }
693 
694 static void nvme_setup_passthrough(struct request *req,
695 		struct nvme_command *cmd)
696 {
697 	memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
698 	/* passthru commands should let the driver set the SGL flags */
699 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
700 }
701 
702 static inline void nvme_setup_flush(struct nvme_ns *ns,
703 		struct nvme_command *cmnd)
704 {
705 	cmnd->common.opcode = nvme_cmd_flush;
706 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
707 }
708 
709 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
710 		struct nvme_command *cmnd)
711 {
712 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
713 	struct nvme_dsm_range *range;
714 	struct bio *bio;
715 
716 	/*
717 	 * Some devices do not consider the DSM 'Number of Ranges' field when
718 	 * determining how much data to DMA. Always allocate memory for maximum
719 	 * number of segments to prevent device reading beyond end of buffer.
720 	 */
721 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
722 
723 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
724 	if (!range) {
725 		/*
726 		 * If we fail allocation our range, fallback to the controller
727 		 * discard page. If that's also busy, it's safe to return
728 		 * busy, as we know we can make progress once that's freed.
729 		 */
730 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
731 			return BLK_STS_RESOURCE;
732 
733 		range = page_address(ns->ctrl->discard_page);
734 	}
735 
736 	__rq_for_each_bio(bio, req) {
737 		u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
738 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
739 
740 		if (n < segments) {
741 			range[n].cattr = cpu_to_le32(0);
742 			range[n].nlb = cpu_to_le32(nlb);
743 			range[n].slba = cpu_to_le64(slba);
744 		}
745 		n++;
746 	}
747 
748 	if (WARN_ON_ONCE(n != segments)) {
749 		if (virt_to_page(range) == ns->ctrl->discard_page)
750 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
751 		else
752 			kfree(range);
753 		return BLK_STS_IOERR;
754 	}
755 
756 	cmnd->dsm.opcode = nvme_cmd_dsm;
757 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
758 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
759 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
760 
761 	req->special_vec.bv_page = virt_to_page(range);
762 	req->special_vec.bv_offset = offset_in_page(range);
763 	req->special_vec.bv_len = alloc_size;
764 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
765 
766 	return BLK_STS_OK;
767 }
768 
769 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
770 		struct request *req, struct nvme_command *cmnd)
771 {
772 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
773 		return nvme_setup_discard(ns, req, cmnd);
774 
775 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
776 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
777 	cmnd->write_zeroes.slba =
778 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
779 	cmnd->write_zeroes.length =
780 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
781 	cmnd->write_zeroes.control = 0;
782 	return BLK_STS_OK;
783 }
784 
785 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
786 		struct request *req, struct nvme_command *cmnd,
787 		enum nvme_opcode op)
788 {
789 	struct nvme_ctrl *ctrl = ns->ctrl;
790 	u16 control = 0;
791 	u32 dsmgmt = 0;
792 
793 	if (req->cmd_flags & REQ_FUA)
794 		control |= NVME_RW_FUA;
795 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
796 		control |= NVME_RW_LR;
797 
798 	if (req->cmd_flags & REQ_RAHEAD)
799 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
800 
801 	cmnd->rw.opcode = op;
802 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
803 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
804 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
805 
806 	if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
807 		nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
808 
809 	if (ns->ms) {
810 		/*
811 		 * If formated with metadata, the block layer always provides a
812 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
813 		 * we enable the PRACT bit for protection information or set the
814 		 * namespace capacity to zero to prevent any I/O.
815 		 */
816 		if (!blk_integrity_rq(req)) {
817 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
818 				return BLK_STS_NOTSUPP;
819 			control |= NVME_RW_PRINFO_PRACT;
820 		}
821 
822 		switch (ns->pi_type) {
823 		case NVME_NS_DPS_PI_TYPE3:
824 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
825 			break;
826 		case NVME_NS_DPS_PI_TYPE1:
827 		case NVME_NS_DPS_PI_TYPE2:
828 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
829 					NVME_RW_PRINFO_PRCHK_REF;
830 			if (op == nvme_cmd_zone_append)
831 				control |= NVME_RW_APPEND_PIREMAP;
832 			cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
833 			break;
834 		}
835 	}
836 
837 	cmnd->rw.control = cpu_to_le16(control);
838 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
839 	return 0;
840 }
841 
842 void nvme_cleanup_cmd(struct request *req)
843 {
844 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
845 		struct nvme_ns *ns = req->rq_disk->private_data;
846 		struct page *page = req->special_vec.bv_page;
847 
848 		if (page == ns->ctrl->discard_page)
849 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
850 		else
851 			kfree(page_address(page) + req->special_vec.bv_offset);
852 	}
853 }
854 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
855 
856 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
857 		struct nvme_command *cmd)
858 {
859 	blk_status_t ret = BLK_STS_OK;
860 
861 	nvme_clear_nvme_request(req);
862 
863 	memset(cmd, 0, sizeof(*cmd));
864 	switch (req_op(req)) {
865 	case REQ_OP_DRV_IN:
866 	case REQ_OP_DRV_OUT:
867 		nvme_setup_passthrough(req, cmd);
868 		break;
869 	case REQ_OP_FLUSH:
870 		nvme_setup_flush(ns, cmd);
871 		break;
872 	case REQ_OP_ZONE_RESET_ALL:
873 	case REQ_OP_ZONE_RESET:
874 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
875 		break;
876 	case REQ_OP_ZONE_OPEN:
877 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
878 		break;
879 	case REQ_OP_ZONE_CLOSE:
880 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
881 		break;
882 	case REQ_OP_ZONE_FINISH:
883 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
884 		break;
885 	case REQ_OP_WRITE_ZEROES:
886 		ret = nvme_setup_write_zeroes(ns, req, cmd);
887 		break;
888 	case REQ_OP_DISCARD:
889 		ret = nvme_setup_discard(ns, req, cmd);
890 		break;
891 	case REQ_OP_READ:
892 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
893 		break;
894 	case REQ_OP_WRITE:
895 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
896 		break;
897 	case REQ_OP_ZONE_APPEND:
898 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
899 		break;
900 	default:
901 		WARN_ON_ONCE(1);
902 		return BLK_STS_IOERR;
903 	}
904 
905 	cmd->common.command_id = req->tag;
906 	trace_nvme_setup_cmd(req, cmd);
907 	return ret;
908 }
909 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
910 
911 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
912 {
913 	struct completion *waiting = rq->end_io_data;
914 
915 	rq->end_io_data = NULL;
916 	complete(waiting);
917 }
918 
919 static void nvme_execute_rq_polled(struct request_queue *q,
920 		struct gendisk *bd_disk, struct request *rq, int at_head)
921 {
922 	DECLARE_COMPLETION_ONSTACK(wait);
923 
924 	WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
925 
926 	rq->cmd_flags |= REQ_HIPRI;
927 	rq->end_io_data = &wait;
928 	blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
929 
930 	while (!completion_done(&wait)) {
931 		blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
932 		cond_resched();
933 	}
934 }
935 
936 /*
937  * Returns 0 on success.  If the result is negative, it's a Linux error code;
938  * if the result is positive, it's an NVM Express status code
939  */
940 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
941 		union nvme_result *result, void *buffer, unsigned bufflen,
942 		unsigned timeout, int qid, int at_head,
943 		blk_mq_req_flags_t flags, bool poll)
944 {
945 	struct request *req;
946 	int ret;
947 
948 	if (qid == NVME_QID_ANY)
949 		req = nvme_alloc_request(q, cmd, flags);
950 	else
951 		req = nvme_alloc_request_qid(q, cmd, flags, qid);
952 	if (IS_ERR(req))
953 		return PTR_ERR(req);
954 
955 	if (timeout)
956 		req->timeout = timeout;
957 
958 	if (buffer && bufflen) {
959 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
960 		if (ret)
961 			goto out;
962 	}
963 
964 	if (poll)
965 		nvme_execute_rq_polled(req->q, NULL, req, at_head);
966 	else
967 		blk_execute_rq(req->q, NULL, req, at_head);
968 	if (result)
969 		*result = nvme_req(req)->result;
970 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
971 		ret = -EINTR;
972 	else
973 		ret = nvme_req(req)->status;
974  out:
975 	blk_mq_free_request(req);
976 	return ret;
977 }
978 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
979 
980 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
981 		void *buffer, unsigned bufflen)
982 {
983 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
984 			NVME_QID_ANY, 0, 0, false);
985 }
986 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
987 
988 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
989 		unsigned len, u32 seed, bool write)
990 {
991 	struct bio_integrity_payload *bip;
992 	int ret = -ENOMEM;
993 	void *buf;
994 
995 	buf = kmalloc(len, GFP_KERNEL);
996 	if (!buf)
997 		goto out;
998 
999 	ret = -EFAULT;
1000 	if (write && copy_from_user(buf, ubuf, len))
1001 		goto out_free_meta;
1002 
1003 	bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1004 	if (IS_ERR(bip)) {
1005 		ret = PTR_ERR(bip);
1006 		goto out_free_meta;
1007 	}
1008 
1009 	bip->bip_iter.bi_size = len;
1010 	bip->bip_iter.bi_sector = seed;
1011 	ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1012 			offset_in_page(buf));
1013 	if (ret == len)
1014 		return buf;
1015 	ret = -ENOMEM;
1016 out_free_meta:
1017 	kfree(buf);
1018 out:
1019 	return ERR_PTR(ret);
1020 }
1021 
1022 static u32 nvme_known_admin_effects(u8 opcode)
1023 {
1024 	switch (opcode) {
1025 	case nvme_admin_format_nvm:
1026 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1027 			NVME_CMD_EFFECTS_CSE_MASK;
1028 	case nvme_admin_sanitize_nvm:
1029 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1030 	default:
1031 		break;
1032 	}
1033 	return 0;
1034 }
1035 
1036 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1037 {
1038 	u32 effects = 0;
1039 
1040 	if (ns) {
1041 		if (ns->head->effects)
1042 			effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1043 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1044 			dev_warn(ctrl->device,
1045 				 "IO command:%02x has unhandled effects:%08x\n",
1046 				 opcode, effects);
1047 		return 0;
1048 	}
1049 
1050 	if (ctrl->effects)
1051 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1052 	effects |= nvme_known_admin_effects(opcode);
1053 
1054 	return effects;
1055 }
1056 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1057 
1058 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1059 			       u8 opcode)
1060 {
1061 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1062 
1063 	/*
1064 	 * For simplicity, IO to all namespaces is quiesced even if the command
1065 	 * effects say only one namespace is affected.
1066 	 */
1067 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1068 		mutex_lock(&ctrl->scan_lock);
1069 		mutex_lock(&ctrl->subsys->lock);
1070 		nvme_mpath_start_freeze(ctrl->subsys);
1071 		nvme_mpath_wait_freeze(ctrl->subsys);
1072 		nvme_start_freeze(ctrl);
1073 		nvme_wait_freeze(ctrl);
1074 	}
1075 	return effects;
1076 }
1077 
1078 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1079 {
1080 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1081 		nvme_unfreeze(ctrl);
1082 		nvme_mpath_unfreeze(ctrl->subsys);
1083 		mutex_unlock(&ctrl->subsys->lock);
1084 		nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1085 		mutex_unlock(&ctrl->scan_lock);
1086 	}
1087 	if (effects & NVME_CMD_EFFECTS_CCC)
1088 		nvme_init_identify(ctrl);
1089 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1090 		nvme_queue_scan(ctrl);
1091 		flush_work(&ctrl->scan_work);
1092 	}
1093 }
1094 
1095 void nvme_execute_passthru_rq(struct request *rq)
1096 {
1097 	struct nvme_command *cmd = nvme_req(rq)->cmd;
1098 	struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1099 	struct nvme_ns *ns = rq->q->queuedata;
1100 	struct gendisk *disk = ns ? ns->disk : NULL;
1101 	u32 effects;
1102 
1103 	effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1104 	blk_execute_rq(rq->q, disk, rq, 0);
1105 	nvme_passthru_end(ctrl, effects);
1106 }
1107 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1108 
1109 static int nvme_submit_user_cmd(struct request_queue *q,
1110 		struct nvme_command *cmd, void __user *ubuffer,
1111 		unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1112 		u32 meta_seed, u64 *result, unsigned timeout)
1113 {
1114 	bool write = nvme_is_write(cmd);
1115 	struct nvme_ns *ns = q->queuedata;
1116 	struct gendisk *disk = ns ? ns->disk : NULL;
1117 	struct request *req;
1118 	struct bio *bio = NULL;
1119 	void *meta = NULL;
1120 	int ret;
1121 
1122 	req = nvme_alloc_request(q, cmd, 0);
1123 	if (IS_ERR(req))
1124 		return PTR_ERR(req);
1125 
1126 	if (timeout)
1127 		req->timeout = timeout;
1128 	nvme_req(req)->flags |= NVME_REQ_USERCMD;
1129 
1130 	if (ubuffer && bufflen) {
1131 		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1132 				GFP_KERNEL);
1133 		if (ret)
1134 			goto out;
1135 		bio = req->bio;
1136 		bio->bi_disk = disk;
1137 		if (disk && meta_buffer && meta_len) {
1138 			meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1139 					meta_seed, write);
1140 			if (IS_ERR(meta)) {
1141 				ret = PTR_ERR(meta);
1142 				goto out_unmap;
1143 			}
1144 			req->cmd_flags |= REQ_INTEGRITY;
1145 		}
1146 	}
1147 
1148 	nvme_execute_passthru_rq(req);
1149 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1150 		ret = -EINTR;
1151 	else
1152 		ret = nvme_req(req)->status;
1153 	if (result)
1154 		*result = le64_to_cpu(nvme_req(req)->result.u64);
1155 	if (meta && !ret && !write) {
1156 		if (copy_to_user(meta_buffer, meta, meta_len))
1157 			ret = -EFAULT;
1158 	}
1159 	kfree(meta);
1160  out_unmap:
1161 	if (bio)
1162 		blk_rq_unmap_user(bio);
1163  out:
1164 	blk_mq_free_request(req);
1165 	return ret;
1166 }
1167 
1168 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1169 {
1170 	struct nvme_ctrl *ctrl = rq->end_io_data;
1171 	unsigned long flags;
1172 	bool startka = false;
1173 
1174 	blk_mq_free_request(rq);
1175 
1176 	if (status) {
1177 		dev_err(ctrl->device,
1178 			"failed nvme_keep_alive_end_io error=%d\n",
1179 				status);
1180 		return;
1181 	}
1182 
1183 	ctrl->comp_seen = false;
1184 	spin_lock_irqsave(&ctrl->lock, flags);
1185 	if (ctrl->state == NVME_CTRL_LIVE ||
1186 	    ctrl->state == NVME_CTRL_CONNECTING)
1187 		startka = true;
1188 	spin_unlock_irqrestore(&ctrl->lock, flags);
1189 	if (startka)
1190 		queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1191 }
1192 
1193 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1194 {
1195 	struct request *rq;
1196 
1197 	rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1198 			BLK_MQ_REQ_RESERVED);
1199 	if (IS_ERR(rq))
1200 		return PTR_ERR(rq);
1201 
1202 	rq->timeout = ctrl->kato * HZ;
1203 	rq->end_io_data = ctrl;
1204 
1205 	blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1206 
1207 	return 0;
1208 }
1209 
1210 static void nvme_keep_alive_work(struct work_struct *work)
1211 {
1212 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1213 			struct nvme_ctrl, ka_work);
1214 	bool comp_seen = ctrl->comp_seen;
1215 
1216 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1217 		dev_dbg(ctrl->device,
1218 			"reschedule traffic based keep-alive timer\n");
1219 		ctrl->comp_seen = false;
1220 		queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1221 		return;
1222 	}
1223 
1224 	if (nvme_keep_alive(ctrl)) {
1225 		/* allocation failure, reset the controller */
1226 		dev_err(ctrl->device, "keep-alive failed\n");
1227 		nvme_reset_ctrl(ctrl);
1228 		return;
1229 	}
1230 }
1231 
1232 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1233 {
1234 	if (unlikely(ctrl->kato == 0))
1235 		return;
1236 
1237 	queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1238 }
1239 
1240 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1241 {
1242 	if (unlikely(ctrl->kato == 0))
1243 		return;
1244 
1245 	cancel_delayed_work_sync(&ctrl->ka_work);
1246 }
1247 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1248 
1249 /*
1250  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1251  * flag, thus sending any new CNS opcodes has a big chance of not working.
1252  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1253  * (but not for any later version).
1254  */
1255 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1256 {
1257 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1258 		return ctrl->vs < NVME_VS(1, 2, 0);
1259 	return ctrl->vs < NVME_VS(1, 1, 0);
1260 }
1261 
1262 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1263 {
1264 	struct nvme_command c = { };
1265 	int error;
1266 
1267 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1268 	c.identify.opcode = nvme_admin_identify;
1269 	c.identify.cns = NVME_ID_CNS_CTRL;
1270 
1271 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1272 	if (!*id)
1273 		return -ENOMEM;
1274 
1275 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1276 			sizeof(struct nvme_id_ctrl));
1277 	if (error)
1278 		kfree(*id);
1279 	return error;
1280 }
1281 
1282 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1283 {
1284 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1285 }
1286 
1287 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1288 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1289 {
1290 	const char *warn_str = "ctrl returned bogus length:";
1291 	void *data = cur;
1292 
1293 	switch (cur->nidt) {
1294 	case NVME_NIDT_EUI64:
1295 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1296 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1297 				 warn_str, cur->nidl);
1298 			return -1;
1299 		}
1300 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1301 		return NVME_NIDT_EUI64_LEN;
1302 	case NVME_NIDT_NGUID:
1303 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1304 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1305 				 warn_str, cur->nidl);
1306 			return -1;
1307 		}
1308 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1309 		return NVME_NIDT_NGUID_LEN;
1310 	case NVME_NIDT_UUID:
1311 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1312 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1313 				 warn_str, cur->nidl);
1314 			return -1;
1315 		}
1316 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1317 		return NVME_NIDT_UUID_LEN;
1318 	case NVME_NIDT_CSI:
1319 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1320 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1321 				 warn_str, cur->nidl);
1322 			return -1;
1323 		}
1324 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1325 		*csi_seen = true;
1326 		return NVME_NIDT_CSI_LEN;
1327 	default:
1328 		/* Skip unknown types */
1329 		return cur->nidl;
1330 	}
1331 }
1332 
1333 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1334 		struct nvme_ns_ids *ids)
1335 {
1336 	struct nvme_command c = { };
1337 	bool csi_seen = false;
1338 	int status, pos, len;
1339 	void *data;
1340 
1341 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1342 		return 0;
1343 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1344 		return 0;
1345 
1346 	c.identify.opcode = nvme_admin_identify;
1347 	c.identify.nsid = cpu_to_le32(nsid);
1348 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1349 
1350 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1351 	if (!data)
1352 		return -ENOMEM;
1353 
1354 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1355 				      NVME_IDENTIFY_DATA_SIZE);
1356 	if (status) {
1357 		dev_warn(ctrl->device,
1358 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1359 			nsid, status);
1360 		goto free_data;
1361 	}
1362 
1363 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1364 		struct nvme_ns_id_desc *cur = data + pos;
1365 
1366 		if (cur->nidl == 0)
1367 			break;
1368 
1369 		len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1370 		if (len < 0)
1371 			break;
1372 
1373 		len += sizeof(*cur);
1374 	}
1375 
1376 	if (nvme_multi_css(ctrl) && !csi_seen) {
1377 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1378 			 nsid);
1379 		status = -EINVAL;
1380 	}
1381 
1382 free_data:
1383 	kfree(data);
1384 	return status;
1385 }
1386 
1387 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1388 			struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1389 {
1390 	struct nvme_command c = { };
1391 	int error;
1392 
1393 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1394 	c.identify.opcode = nvme_admin_identify;
1395 	c.identify.nsid = cpu_to_le32(nsid);
1396 	c.identify.cns = NVME_ID_CNS_NS;
1397 
1398 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1399 	if (!*id)
1400 		return -ENOMEM;
1401 
1402 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1403 	if (error) {
1404 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1405 		goto out_free_id;
1406 	}
1407 
1408 	error = -ENODEV;
1409 	if ((*id)->ncap == 0) /* namespace not allocated or attached */
1410 		goto out_free_id;
1411 
1412 	if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1413 	    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1414 		memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1415 	if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1416 	    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1417 		memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1418 
1419 	return 0;
1420 
1421 out_free_id:
1422 	kfree(*id);
1423 	return error;
1424 }
1425 
1426 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1427 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1428 {
1429 	union nvme_result res = { 0 };
1430 	struct nvme_command c;
1431 	int ret;
1432 
1433 	memset(&c, 0, sizeof(c));
1434 	c.features.opcode = op;
1435 	c.features.fid = cpu_to_le32(fid);
1436 	c.features.dword11 = cpu_to_le32(dword11);
1437 
1438 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1439 			buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1440 	if (ret >= 0 && result)
1441 		*result = le32_to_cpu(res.u32);
1442 	return ret;
1443 }
1444 
1445 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1446 		      unsigned int dword11, void *buffer, size_t buflen,
1447 		      u32 *result)
1448 {
1449 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1450 			     buflen, result);
1451 }
1452 EXPORT_SYMBOL_GPL(nvme_set_features);
1453 
1454 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1455 		      unsigned int dword11, void *buffer, size_t buflen,
1456 		      u32 *result)
1457 {
1458 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1459 			     buflen, result);
1460 }
1461 EXPORT_SYMBOL_GPL(nvme_get_features);
1462 
1463 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1464 {
1465 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1466 	u32 result;
1467 	int status, nr_io_queues;
1468 
1469 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1470 			&result);
1471 	if (status < 0)
1472 		return status;
1473 
1474 	/*
1475 	 * Degraded controllers might return an error when setting the queue
1476 	 * count.  We still want to be able to bring them online and offer
1477 	 * access to the admin queue, as that might be only way to fix them up.
1478 	 */
1479 	if (status > 0) {
1480 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1481 		*count = 0;
1482 	} else {
1483 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1484 		*count = min(*count, nr_io_queues);
1485 	}
1486 
1487 	return 0;
1488 }
1489 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1490 
1491 #define NVME_AEN_SUPPORTED \
1492 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1493 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1494 
1495 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1496 {
1497 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1498 	int status;
1499 
1500 	if (!supported_aens)
1501 		return;
1502 
1503 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1504 			NULL, 0, &result);
1505 	if (status)
1506 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1507 			 supported_aens);
1508 
1509 	queue_work(nvme_wq, &ctrl->async_event_work);
1510 }
1511 
1512 /*
1513  * Convert integer values from ioctl structures to user pointers, silently
1514  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1515  * kernels.
1516  */
1517 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1518 {
1519 	if (in_compat_syscall())
1520 		ptrval = (compat_uptr_t)ptrval;
1521 	return (void __user *)ptrval;
1522 }
1523 
1524 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1525 {
1526 	struct nvme_user_io io;
1527 	struct nvme_command c;
1528 	unsigned length, meta_len;
1529 	void __user *metadata;
1530 
1531 	if (copy_from_user(&io, uio, sizeof(io)))
1532 		return -EFAULT;
1533 	if (io.flags)
1534 		return -EINVAL;
1535 
1536 	switch (io.opcode) {
1537 	case nvme_cmd_write:
1538 	case nvme_cmd_read:
1539 	case nvme_cmd_compare:
1540 		break;
1541 	default:
1542 		return -EINVAL;
1543 	}
1544 
1545 	length = (io.nblocks + 1) << ns->lba_shift;
1546 	meta_len = (io.nblocks + 1) * ns->ms;
1547 	metadata = nvme_to_user_ptr(io.metadata);
1548 
1549 	if (ns->features & NVME_NS_EXT_LBAS) {
1550 		length += meta_len;
1551 		meta_len = 0;
1552 	} else if (meta_len) {
1553 		if ((io.metadata & 3) || !io.metadata)
1554 			return -EINVAL;
1555 	}
1556 
1557 	memset(&c, 0, sizeof(c));
1558 	c.rw.opcode = io.opcode;
1559 	c.rw.flags = io.flags;
1560 	c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1561 	c.rw.slba = cpu_to_le64(io.slba);
1562 	c.rw.length = cpu_to_le16(io.nblocks);
1563 	c.rw.control = cpu_to_le16(io.control);
1564 	c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1565 	c.rw.reftag = cpu_to_le32(io.reftag);
1566 	c.rw.apptag = cpu_to_le16(io.apptag);
1567 	c.rw.appmask = cpu_to_le16(io.appmask);
1568 
1569 	return nvme_submit_user_cmd(ns->queue, &c,
1570 			nvme_to_user_ptr(io.addr), length,
1571 			metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1572 }
1573 
1574 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1575 			struct nvme_passthru_cmd __user *ucmd)
1576 {
1577 	struct nvme_passthru_cmd cmd;
1578 	struct nvme_command c;
1579 	unsigned timeout = 0;
1580 	u64 result;
1581 	int status;
1582 
1583 	if (!capable(CAP_SYS_ADMIN))
1584 		return -EACCES;
1585 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1586 		return -EFAULT;
1587 	if (cmd.flags)
1588 		return -EINVAL;
1589 
1590 	memset(&c, 0, sizeof(c));
1591 	c.common.opcode = cmd.opcode;
1592 	c.common.flags = cmd.flags;
1593 	c.common.nsid = cpu_to_le32(cmd.nsid);
1594 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1595 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1596 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1597 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1598 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1599 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1600 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1601 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1602 
1603 	if (cmd.timeout_ms)
1604 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1605 
1606 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1607 			nvme_to_user_ptr(cmd.addr), cmd.data_len,
1608 			nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1609 			0, &result, timeout);
1610 
1611 	if (status >= 0) {
1612 		if (put_user(result, &ucmd->result))
1613 			return -EFAULT;
1614 	}
1615 
1616 	return status;
1617 }
1618 
1619 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1620 			struct nvme_passthru_cmd64 __user *ucmd)
1621 {
1622 	struct nvme_passthru_cmd64 cmd;
1623 	struct nvme_command c;
1624 	unsigned timeout = 0;
1625 	int status;
1626 
1627 	if (!capable(CAP_SYS_ADMIN))
1628 		return -EACCES;
1629 	if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1630 		return -EFAULT;
1631 	if (cmd.flags)
1632 		return -EINVAL;
1633 
1634 	memset(&c, 0, sizeof(c));
1635 	c.common.opcode = cmd.opcode;
1636 	c.common.flags = cmd.flags;
1637 	c.common.nsid = cpu_to_le32(cmd.nsid);
1638 	c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1639 	c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1640 	c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1641 	c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1642 	c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1643 	c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1644 	c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1645 	c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1646 
1647 	if (cmd.timeout_ms)
1648 		timeout = msecs_to_jiffies(cmd.timeout_ms);
1649 
1650 	status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1651 			nvme_to_user_ptr(cmd.addr), cmd.data_len,
1652 			nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1653 			0, &cmd.result, timeout);
1654 
1655 	if (status >= 0) {
1656 		if (put_user(cmd.result, &ucmd->result))
1657 			return -EFAULT;
1658 	}
1659 
1660 	return status;
1661 }
1662 
1663 /*
1664  * Issue ioctl requests on the first available path.  Note that unlike normal
1665  * block layer requests we will not retry failed request on another controller.
1666  */
1667 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1668 		struct nvme_ns_head **head, int *srcu_idx)
1669 {
1670 #ifdef CONFIG_NVME_MULTIPATH
1671 	if (disk->fops == &nvme_ns_head_ops) {
1672 		struct nvme_ns *ns;
1673 
1674 		*head = disk->private_data;
1675 		*srcu_idx = srcu_read_lock(&(*head)->srcu);
1676 		ns = nvme_find_path(*head);
1677 		if (!ns)
1678 			srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1679 		return ns;
1680 	}
1681 #endif
1682 	*head = NULL;
1683 	*srcu_idx = -1;
1684 	return disk->private_data;
1685 }
1686 
1687 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1688 {
1689 	if (head)
1690 		srcu_read_unlock(&head->srcu, idx);
1691 }
1692 
1693 static bool is_ctrl_ioctl(unsigned int cmd)
1694 {
1695 	if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1696 		return true;
1697 	if (is_sed_ioctl(cmd))
1698 		return true;
1699 	return false;
1700 }
1701 
1702 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1703 				  void __user *argp,
1704 				  struct nvme_ns_head *head,
1705 				  int srcu_idx)
1706 {
1707 	struct nvme_ctrl *ctrl = ns->ctrl;
1708 	int ret;
1709 
1710 	nvme_get_ctrl(ns->ctrl);
1711 	nvme_put_ns_from_disk(head, srcu_idx);
1712 
1713 	switch (cmd) {
1714 	case NVME_IOCTL_ADMIN_CMD:
1715 		ret = nvme_user_cmd(ctrl, NULL, argp);
1716 		break;
1717 	case NVME_IOCTL_ADMIN64_CMD:
1718 		ret = nvme_user_cmd64(ctrl, NULL, argp);
1719 		break;
1720 	default:
1721 		ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1722 		break;
1723 	}
1724 	nvme_put_ctrl(ctrl);
1725 	return ret;
1726 }
1727 
1728 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1729 		unsigned int cmd, unsigned long arg)
1730 {
1731 	struct nvme_ns_head *head = NULL;
1732 	void __user *argp = (void __user *)arg;
1733 	struct nvme_ns *ns;
1734 	int srcu_idx, ret;
1735 
1736 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1737 	if (unlikely(!ns))
1738 		return -EWOULDBLOCK;
1739 
1740 	/*
1741 	 * Handle ioctls that apply to the controller instead of the namespace
1742 	 * seperately and drop the ns SRCU reference early.  This avoids a
1743 	 * deadlock when deleting namespaces using the passthrough interface.
1744 	 */
1745 	if (is_ctrl_ioctl(cmd))
1746 		return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1747 
1748 	switch (cmd) {
1749 	case NVME_IOCTL_ID:
1750 		force_successful_syscall_return();
1751 		ret = ns->head->ns_id;
1752 		break;
1753 	case NVME_IOCTL_IO_CMD:
1754 		ret = nvme_user_cmd(ns->ctrl, ns, argp);
1755 		break;
1756 	case NVME_IOCTL_SUBMIT_IO:
1757 		ret = nvme_submit_io(ns, argp);
1758 		break;
1759 	case NVME_IOCTL_IO64_CMD:
1760 		ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1761 		break;
1762 	default:
1763 		if (ns->ndev)
1764 			ret = nvme_nvm_ioctl(ns, cmd, arg);
1765 		else
1766 			ret = -ENOTTY;
1767 	}
1768 
1769 	nvme_put_ns_from_disk(head, srcu_idx);
1770 	return ret;
1771 }
1772 
1773 #ifdef CONFIG_COMPAT
1774 struct nvme_user_io32 {
1775 	__u8	opcode;
1776 	__u8	flags;
1777 	__u16	control;
1778 	__u16	nblocks;
1779 	__u16	rsvd;
1780 	__u64	metadata;
1781 	__u64	addr;
1782 	__u64	slba;
1783 	__u32	dsmgmt;
1784 	__u32	reftag;
1785 	__u16	apptag;
1786 	__u16	appmask;
1787 } __attribute__((__packed__));
1788 
1789 #define NVME_IOCTL_SUBMIT_IO32	_IOW('N', 0x42, struct nvme_user_io32)
1790 
1791 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1792 		unsigned int cmd, unsigned long arg)
1793 {
1794 	/*
1795 	 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1796 	 * between 32 bit programs and 64 bit kernel.
1797 	 * The cause is that the results of sizeof(struct nvme_user_io),
1798 	 * which is used to define NVME_IOCTL_SUBMIT_IO,
1799 	 * are not same between 32 bit compiler and 64 bit compiler.
1800 	 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1801 	 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1802 	 * Other IOCTL numbers are same between 32 bit and 64 bit.
1803 	 * So there is nothing to do regarding to other IOCTL numbers.
1804 	 */
1805 	if (cmd == NVME_IOCTL_SUBMIT_IO32)
1806 		return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1807 
1808 	return nvme_ioctl(bdev, mode, cmd, arg);
1809 }
1810 #else
1811 #define nvme_compat_ioctl	NULL
1812 #endif /* CONFIG_COMPAT */
1813 
1814 static int nvme_open(struct block_device *bdev, fmode_t mode)
1815 {
1816 	struct nvme_ns *ns = bdev->bd_disk->private_data;
1817 
1818 #ifdef CONFIG_NVME_MULTIPATH
1819 	/* should never be called due to GENHD_FL_HIDDEN */
1820 	if (WARN_ON_ONCE(ns->head->disk))
1821 		goto fail;
1822 #endif
1823 	if (!kref_get_unless_zero(&ns->kref))
1824 		goto fail;
1825 	if (!try_module_get(ns->ctrl->ops->module))
1826 		goto fail_put_ns;
1827 
1828 	return 0;
1829 
1830 fail_put_ns:
1831 	nvme_put_ns(ns);
1832 fail:
1833 	return -ENXIO;
1834 }
1835 
1836 static void nvme_release(struct gendisk *disk, fmode_t mode)
1837 {
1838 	struct nvme_ns *ns = disk->private_data;
1839 
1840 	module_put(ns->ctrl->ops->module);
1841 	nvme_put_ns(ns);
1842 }
1843 
1844 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1845 {
1846 	/* some standard values */
1847 	geo->heads = 1 << 6;
1848 	geo->sectors = 1 << 5;
1849 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1850 	return 0;
1851 }
1852 
1853 #ifdef CONFIG_BLK_DEV_INTEGRITY
1854 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1855 				u32 max_integrity_segments)
1856 {
1857 	struct blk_integrity integrity;
1858 
1859 	memset(&integrity, 0, sizeof(integrity));
1860 	switch (pi_type) {
1861 	case NVME_NS_DPS_PI_TYPE3:
1862 		integrity.profile = &t10_pi_type3_crc;
1863 		integrity.tag_size = sizeof(u16) + sizeof(u32);
1864 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1865 		break;
1866 	case NVME_NS_DPS_PI_TYPE1:
1867 	case NVME_NS_DPS_PI_TYPE2:
1868 		integrity.profile = &t10_pi_type1_crc;
1869 		integrity.tag_size = sizeof(u16);
1870 		integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1871 		break;
1872 	default:
1873 		integrity.profile = NULL;
1874 		break;
1875 	}
1876 	integrity.tuple_size = ms;
1877 	blk_integrity_register(disk, &integrity);
1878 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1879 }
1880 #else
1881 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1882 				u32 max_integrity_segments)
1883 {
1884 }
1885 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1886 
1887 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1888 {
1889 	struct nvme_ctrl *ctrl = ns->ctrl;
1890 	struct request_queue *queue = disk->queue;
1891 	u32 size = queue_logical_block_size(queue);
1892 
1893 	if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1894 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1895 		return;
1896 	}
1897 
1898 	if (ctrl->nr_streams && ns->sws && ns->sgs)
1899 		size *= ns->sws * ns->sgs;
1900 
1901 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1902 			NVME_DSM_MAX_RANGES);
1903 
1904 	queue->limits.discard_alignment = 0;
1905 	queue->limits.discard_granularity = size;
1906 
1907 	/* If discard is already enabled, don't reset queue limits */
1908 	if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1909 		return;
1910 
1911 	blk_queue_max_discard_sectors(queue, UINT_MAX);
1912 	blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1913 
1914 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1915 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1916 }
1917 
1918 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1919 {
1920 	u64 max_blocks;
1921 
1922 	if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1923 	    (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1924 		return;
1925 	/*
1926 	 * Even though NVMe spec explicitly states that MDTS is not
1927 	 * applicable to the write-zeroes:- "The restriction does not apply to
1928 	 * commands that do not transfer data between the host and the
1929 	 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1930 	 * In order to be more cautious use controller's max_hw_sectors value
1931 	 * to configure the maximum sectors for the write-zeroes which is
1932 	 * configured based on the controller's MDTS field in the
1933 	 * nvme_init_identify() if available.
1934 	 */
1935 	if (ns->ctrl->max_hw_sectors == UINT_MAX)
1936 		max_blocks = (u64)USHRT_MAX + 1;
1937 	else
1938 		max_blocks = ns->ctrl->max_hw_sectors + 1;
1939 
1940 	blk_queue_max_write_zeroes_sectors(disk->queue,
1941 					   nvme_lba_to_sect(ns, max_blocks));
1942 }
1943 
1944 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1945 {
1946 	return !uuid_is_null(&ids->uuid) ||
1947 		memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1948 		memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1949 }
1950 
1951 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1952 {
1953 	return uuid_equal(&a->uuid, &b->uuid) &&
1954 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1955 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1956 		a->csi == b->csi;
1957 }
1958 
1959 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1960 				 u32 *phys_bs, u32 *io_opt)
1961 {
1962 	struct streams_directive_params s;
1963 	int ret;
1964 
1965 	if (!ctrl->nr_streams)
1966 		return 0;
1967 
1968 	ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1969 	if (ret)
1970 		return ret;
1971 
1972 	ns->sws = le32_to_cpu(s.sws);
1973 	ns->sgs = le16_to_cpu(s.sgs);
1974 
1975 	if (ns->sws) {
1976 		*phys_bs = ns->sws * (1 << ns->lba_shift);
1977 		if (ns->sgs)
1978 			*io_opt = *phys_bs * ns->sgs;
1979 	}
1980 
1981 	return 0;
1982 }
1983 
1984 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1985 {
1986 	struct nvme_ctrl *ctrl = ns->ctrl;
1987 
1988 	/*
1989 	 * The PI implementation requires the metadata size to be equal to the
1990 	 * t10 pi tuple size.
1991 	 */
1992 	ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1993 	if (ns->ms == sizeof(struct t10_pi_tuple))
1994 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1995 	else
1996 		ns->pi_type = 0;
1997 
1998 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1999 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2000 		return 0;
2001 	if (ctrl->ops->flags & NVME_F_FABRICS) {
2002 		/*
2003 		 * The NVMe over Fabrics specification only supports metadata as
2004 		 * part of the extended data LBA.  We rely on HCA/HBA support to
2005 		 * remap the separate metadata buffer from the block layer.
2006 		 */
2007 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2008 			return -EINVAL;
2009 		if (ctrl->max_integrity_segments)
2010 			ns->features |=
2011 				(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2012 	} else {
2013 		/*
2014 		 * For PCIe controllers, we can't easily remap the separate
2015 		 * metadata buffer from the block layer and thus require a
2016 		 * separate metadata buffer for block layer metadata/PI support.
2017 		 * We allow extended LBAs for the passthrough interface, though.
2018 		 */
2019 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
2020 			ns->features |= NVME_NS_EXT_LBAS;
2021 		else
2022 			ns->features |= NVME_NS_METADATA_SUPPORTED;
2023 	}
2024 
2025 	return 0;
2026 }
2027 
2028 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2029 		struct request_queue *q)
2030 {
2031 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2032 
2033 	if (ctrl->max_hw_sectors) {
2034 		u32 max_segments =
2035 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2036 
2037 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
2038 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2039 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2040 	}
2041 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2042 	blk_queue_dma_alignment(q, 7);
2043 	blk_queue_write_cache(q, vwc, vwc);
2044 }
2045 
2046 static void nvme_update_disk_info(struct gendisk *disk,
2047 		struct nvme_ns *ns, struct nvme_id_ns *id)
2048 {
2049 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2050 	unsigned short bs = 1 << ns->lba_shift;
2051 	u32 atomic_bs, phys_bs, io_opt = 0;
2052 
2053 	/*
2054 	 * The block layer can't support LBA sizes larger than the page size
2055 	 * yet, so catch this early and don't allow block I/O.
2056 	 */
2057 	if (ns->lba_shift > PAGE_SHIFT) {
2058 		capacity = 0;
2059 		bs = (1 << 9);
2060 	}
2061 
2062 	blk_integrity_unregister(disk);
2063 
2064 	atomic_bs = phys_bs = bs;
2065 	nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2066 	if (id->nabo == 0) {
2067 		/*
2068 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
2069 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
2070 		 * 0 then AWUPF must be used instead.
2071 		 */
2072 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2073 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2074 		else
2075 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2076 	}
2077 
2078 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2079 		/* NPWG = Namespace Preferred Write Granularity */
2080 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2081 		/* NOWS = Namespace Optimal Write Size */
2082 		io_opt = bs * (1 + le16_to_cpu(id->nows));
2083 	}
2084 
2085 	blk_queue_logical_block_size(disk->queue, bs);
2086 	/*
2087 	 * Linux filesystems assume writing a single physical block is
2088 	 * an atomic operation. Hence limit the physical block size to the
2089 	 * value of the Atomic Write Unit Power Fail parameter.
2090 	 */
2091 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2092 	blk_queue_io_min(disk->queue, phys_bs);
2093 	blk_queue_io_opt(disk->queue, io_opt);
2094 
2095 	/*
2096 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
2097 	 * metadata masquerading as Type 0 if supported, otherwise reject block
2098 	 * I/O to namespaces with metadata except when the namespace supports
2099 	 * PI, as it can strip/insert in that case.
2100 	 */
2101 	if (ns->ms) {
2102 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2103 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
2104 			nvme_init_integrity(disk, ns->ms, ns->pi_type,
2105 					    ns->ctrl->max_integrity_segments);
2106 		else if (!nvme_ns_has_pi(ns))
2107 			capacity = 0;
2108 	}
2109 
2110 	set_capacity_and_notify(disk, capacity);
2111 
2112 	nvme_config_discard(disk, ns);
2113 	nvme_config_write_zeroes(disk, ns);
2114 
2115 	if ((id->nsattr & NVME_NS_ATTR_RO) ||
2116 	    test_bit(NVME_NS_FORCE_RO, &ns->flags))
2117 		set_disk_ro(disk, true);
2118 }
2119 
2120 static inline bool nvme_first_scan(struct gendisk *disk)
2121 {
2122 	/* nvme_alloc_ns() scans the disk prior to adding it */
2123 	return !(disk->flags & GENHD_FL_UP);
2124 }
2125 
2126 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2127 {
2128 	struct nvme_ctrl *ctrl = ns->ctrl;
2129 	u32 iob;
2130 
2131 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2132 	    is_power_of_2(ctrl->max_hw_sectors))
2133 		iob = ctrl->max_hw_sectors;
2134 	else
2135 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2136 
2137 	if (!iob)
2138 		return;
2139 
2140 	if (!is_power_of_2(iob)) {
2141 		if (nvme_first_scan(ns->disk))
2142 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2143 				ns->disk->disk_name, iob);
2144 		return;
2145 	}
2146 
2147 	if (blk_queue_is_zoned(ns->disk->queue)) {
2148 		if (nvme_first_scan(ns->disk))
2149 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
2150 				ns->disk->disk_name);
2151 		return;
2152 	}
2153 
2154 	blk_queue_chunk_sectors(ns->queue, iob);
2155 }
2156 
2157 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2158 {
2159 	unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2160 	int ret;
2161 
2162 	blk_mq_freeze_queue(ns->disk->queue);
2163 	ns->lba_shift = id->lbaf[lbaf].ds;
2164 	nvme_set_queue_limits(ns->ctrl, ns->queue);
2165 
2166 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2167 		ret = nvme_update_zone_info(ns, lbaf);
2168 		if (ret)
2169 			goto out_unfreeze;
2170 	}
2171 
2172 	ret = nvme_configure_metadata(ns, id);
2173 	if (ret)
2174 		goto out_unfreeze;
2175 	nvme_set_chunk_sectors(ns, id);
2176 	nvme_update_disk_info(ns->disk, ns, id);
2177 	blk_mq_unfreeze_queue(ns->disk->queue);
2178 
2179 	if (blk_queue_is_zoned(ns->queue)) {
2180 		ret = nvme_revalidate_zones(ns);
2181 		if (ret && !nvme_first_scan(ns->disk))
2182 			return ret;
2183 	}
2184 
2185 #ifdef CONFIG_NVME_MULTIPATH
2186 	if (ns->head->disk) {
2187 		blk_mq_freeze_queue(ns->head->disk->queue);
2188 		nvme_update_disk_info(ns->head->disk, ns, id);
2189 		blk_stack_limits(&ns->head->disk->queue->limits,
2190 				 &ns->queue->limits, 0);
2191 		blk_queue_update_readahead(ns->head->disk->queue);
2192 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2193 	}
2194 #endif
2195 	return 0;
2196 
2197 out_unfreeze:
2198 	blk_mq_unfreeze_queue(ns->disk->queue);
2199 	return ret;
2200 }
2201 
2202 static char nvme_pr_type(enum pr_type type)
2203 {
2204 	switch (type) {
2205 	case PR_WRITE_EXCLUSIVE:
2206 		return 1;
2207 	case PR_EXCLUSIVE_ACCESS:
2208 		return 2;
2209 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
2210 		return 3;
2211 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2212 		return 4;
2213 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
2214 		return 5;
2215 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2216 		return 6;
2217 	default:
2218 		return 0;
2219 	}
2220 };
2221 
2222 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2223 				u64 key, u64 sa_key, u8 op)
2224 {
2225 	struct nvme_ns_head *head = NULL;
2226 	struct nvme_ns *ns;
2227 	struct nvme_command c;
2228 	int srcu_idx, ret;
2229 	u8 data[16] = { 0, };
2230 
2231 	ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2232 	if (unlikely(!ns))
2233 		return -EWOULDBLOCK;
2234 
2235 	put_unaligned_le64(key, &data[0]);
2236 	put_unaligned_le64(sa_key, &data[8]);
2237 
2238 	memset(&c, 0, sizeof(c));
2239 	c.common.opcode = op;
2240 	c.common.nsid = cpu_to_le32(ns->head->ns_id);
2241 	c.common.cdw10 = cpu_to_le32(cdw10);
2242 
2243 	ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2244 	nvme_put_ns_from_disk(head, srcu_idx);
2245 	return ret;
2246 }
2247 
2248 static int nvme_pr_register(struct block_device *bdev, u64 old,
2249 		u64 new, unsigned flags)
2250 {
2251 	u32 cdw10;
2252 
2253 	if (flags & ~PR_FL_IGNORE_KEY)
2254 		return -EOPNOTSUPP;
2255 
2256 	cdw10 = old ? 2 : 0;
2257 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2258 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2259 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2260 }
2261 
2262 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2263 		enum pr_type type, unsigned flags)
2264 {
2265 	u32 cdw10;
2266 
2267 	if (flags & ~PR_FL_IGNORE_KEY)
2268 		return -EOPNOTSUPP;
2269 
2270 	cdw10 = nvme_pr_type(type) << 8;
2271 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2272 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2273 }
2274 
2275 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2276 		enum pr_type type, bool abort)
2277 {
2278 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2279 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2280 }
2281 
2282 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2283 {
2284 	u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2285 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2286 }
2287 
2288 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2289 {
2290 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2291 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2292 }
2293 
2294 static const struct pr_ops nvme_pr_ops = {
2295 	.pr_register	= nvme_pr_register,
2296 	.pr_reserve	= nvme_pr_reserve,
2297 	.pr_release	= nvme_pr_release,
2298 	.pr_preempt	= nvme_pr_preempt,
2299 	.pr_clear	= nvme_pr_clear,
2300 };
2301 
2302 #ifdef CONFIG_BLK_SED_OPAL
2303 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2304 		bool send)
2305 {
2306 	struct nvme_ctrl *ctrl = data;
2307 	struct nvme_command cmd;
2308 
2309 	memset(&cmd, 0, sizeof(cmd));
2310 	if (send)
2311 		cmd.common.opcode = nvme_admin_security_send;
2312 	else
2313 		cmd.common.opcode = nvme_admin_security_recv;
2314 	cmd.common.nsid = 0;
2315 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2316 	cmd.common.cdw11 = cpu_to_le32(len);
2317 
2318 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2319 			NVME_QID_ANY, 1, 0, false);
2320 }
2321 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2322 #endif /* CONFIG_BLK_SED_OPAL */
2323 
2324 static const struct block_device_operations nvme_bdev_ops = {
2325 	.owner		= THIS_MODULE,
2326 	.ioctl		= nvme_ioctl,
2327 	.compat_ioctl	= nvme_compat_ioctl,
2328 	.open		= nvme_open,
2329 	.release	= nvme_release,
2330 	.getgeo		= nvme_getgeo,
2331 	.report_zones	= nvme_report_zones,
2332 	.pr_ops		= &nvme_pr_ops,
2333 };
2334 
2335 #ifdef CONFIG_NVME_MULTIPATH
2336 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2337 {
2338 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
2339 
2340 	if (!kref_get_unless_zero(&head->ref))
2341 		return -ENXIO;
2342 	return 0;
2343 }
2344 
2345 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2346 {
2347 	nvme_put_ns_head(disk->private_data);
2348 }
2349 
2350 const struct block_device_operations nvme_ns_head_ops = {
2351 	.owner		= THIS_MODULE,
2352 	.submit_bio	= nvme_ns_head_submit_bio,
2353 	.open		= nvme_ns_head_open,
2354 	.release	= nvme_ns_head_release,
2355 	.ioctl		= nvme_ioctl,
2356 	.compat_ioctl	= nvme_compat_ioctl,
2357 	.getgeo		= nvme_getgeo,
2358 	.report_zones	= nvme_report_zones,
2359 	.pr_ops		= &nvme_pr_ops,
2360 };
2361 #endif /* CONFIG_NVME_MULTIPATH */
2362 
2363 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2364 {
2365 	unsigned long timeout =
2366 		((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2367 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2368 	int ret;
2369 
2370 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2371 		if (csts == ~0)
2372 			return -ENODEV;
2373 		if ((csts & NVME_CSTS_RDY) == bit)
2374 			break;
2375 
2376 		usleep_range(1000, 2000);
2377 		if (fatal_signal_pending(current))
2378 			return -EINTR;
2379 		if (time_after(jiffies, timeout)) {
2380 			dev_err(ctrl->device,
2381 				"Device not ready; aborting %s, CSTS=0x%x\n",
2382 				enabled ? "initialisation" : "reset", csts);
2383 			return -ENODEV;
2384 		}
2385 	}
2386 
2387 	return ret;
2388 }
2389 
2390 /*
2391  * If the device has been passed off to us in an enabled state, just clear
2392  * the enabled bit.  The spec says we should set the 'shutdown notification
2393  * bits', but doing so may cause the device to complete commands to the
2394  * admin queue ... and we don't know what memory that might be pointing at!
2395  */
2396 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2397 {
2398 	int ret;
2399 
2400 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2401 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2402 
2403 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2404 	if (ret)
2405 		return ret;
2406 
2407 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2408 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2409 
2410 	return nvme_wait_ready(ctrl, ctrl->cap, false);
2411 }
2412 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2413 
2414 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2415 {
2416 	unsigned dev_page_min;
2417 	int ret;
2418 
2419 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2420 	if (ret) {
2421 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2422 		return ret;
2423 	}
2424 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2425 
2426 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2427 		dev_err(ctrl->device,
2428 			"Minimum device page size %u too large for host (%u)\n",
2429 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2430 		return -ENODEV;
2431 	}
2432 
2433 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2434 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2435 	else
2436 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2437 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2438 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2439 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2440 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2441 
2442 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2443 	if (ret)
2444 		return ret;
2445 	return nvme_wait_ready(ctrl, ctrl->cap, true);
2446 }
2447 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2448 
2449 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2450 {
2451 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2452 	u32 csts;
2453 	int ret;
2454 
2455 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2456 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2457 
2458 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2459 	if (ret)
2460 		return ret;
2461 
2462 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2463 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2464 			break;
2465 
2466 		msleep(100);
2467 		if (fatal_signal_pending(current))
2468 			return -EINTR;
2469 		if (time_after(jiffies, timeout)) {
2470 			dev_err(ctrl->device,
2471 				"Device shutdown incomplete; abort shutdown\n");
2472 			return -ENODEV;
2473 		}
2474 	}
2475 
2476 	return ret;
2477 }
2478 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2479 
2480 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2481 {
2482 	__le64 ts;
2483 	int ret;
2484 
2485 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2486 		return 0;
2487 
2488 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2489 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2490 			NULL);
2491 	if (ret)
2492 		dev_warn_once(ctrl->device,
2493 			"could not set timestamp (%d)\n", ret);
2494 	return ret;
2495 }
2496 
2497 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2498 {
2499 	struct nvme_feat_host_behavior *host;
2500 	int ret;
2501 
2502 	/* Don't bother enabling the feature if retry delay is not reported */
2503 	if (!ctrl->crdt[0])
2504 		return 0;
2505 
2506 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2507 	if (!host)
2508 		return 0;
2509 
2510 	host->acre = NVME_ENABLE_ACRE;
2511 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2512 				host, sizeof(*host), NULL);
2513 	kfree(host);
2514 	return ret;
2515 }
2516 
2517 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2518 {
2519 	/*
2520 	 * APST (Autonomous Power State Transition) lets us program a
2521 	 * table of power state transitions that the controller will
2522 	 * perform automatically.  We configure it with a simple
2523 	 * heuristic: we are willing to spend at most 2% of the time
2524 	 * transitioning between power states.  Therefore, when running
2525 	 * in any given state, we will enter the next lower-power
2526 	 * non-operational state after waiting 50 * (enlat + exlat)
2527 	 * microseconds, as long as that state's exit latency is under
2528 	 * the requested maximum latency.
2529 	 *
2530 	 * We will not autonomously enter any non-operational state for
2531 	 * which the total latency exceeds ps_max_latency_us.  Users
2532 	 * can set ps_max_latency_us to zero to turn off APST.
2533 	 */
2534 
2535 	unsigned apste;
2536 	struct nvme_feat_auto_pst *table;
2537 	u64 max_lat_us = 0;
2538 	int max_ps = -1;
2539 	int ret;
2540 
2541 	/*
2542 	 * If APST isn't supported or if we haven't been initialized yet,
2543 	 * then don't do anything.
2544 	 */
2545 	if (!ctrl->apsta)
2546 		return 0;
2547 
2548 	if (ctrl->npss > 31) {
2549 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2550 		return 0;
2551 	}
2552 
2553 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2554 	if (!table)
2555 		return 0;
2556 
2557 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2558 		/* Turn off APST. */
2559 		apste = 0;
2560 		dev_dbg(ctrl->device, "APST disabled\n");
2561 	} else {
2562 		__le64 target = cpu_to_le64(0);
2563 		int state;
2564 
2565 		/*
2566 		 * Walk through all states from lowest- to highest-power.
2567 		 * According to the spec, lower-numbered states use more
2568 		 * power.  NPSS, despite the name, is the index of the
2569 		 * lowest-power state, not the number of states.
2570 		 */
2571 		for (state = (int)ctrl->npss; state >= 0; state--) {
2572 			u64 total_latency_us, exit_latency_us, transition_ms;
2573 
2574 			if (target)
2575 				table->entries[state] = target;
2576 
2577 			/*
2578 			 * Don't allow transitions to the deepest state
2579 			 * if it's quirked off.
2580 			 */
2581 			if (state == ctrl->npss &&
2582 			    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2583 				continue;
2584 
2585 			/*
2586 			 * Is this state a useful non-operational state for
2587 			 * higher-power states to autonomously transition to?
2588 			 */
2589 			if (!(ctrl->psd[state].flags &
2590 			      NVME_PS_FLAGS_NON_OP_STATE))
2591 				continue;
2592 
2593 			exit_latency_us =
2594 				(u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2595 			if (exit_latency_us > ctrl->ps_max_latency_us)
2596 				continue;
2597 
2598 			total_latency_us =
2599 				exit_latency_us +
2600 				le32_to_cpu(ctrl->psd[state].entry_lat);
2601 
2602 			/*
2603 			 * This state is good.  Use it as the APST idle
2604 			 * target for higher power states.
2605 			 */
2606 			transition_ms = total_latency_us + 19;
2607 			do_div(transition_ms, 20);
2608 			if (transition_ms > (1 << 24) - 1)
2609 				transition_ms = (1 << 24) - 1;
2610 
2611 			target = cpu_to_le64((state << 3) |
2612 					     (transition_ms << 8));
2613 
2614 			if (max_ps == -1)
2615 				max_ps = state;
2616 
2617 			if (total_latency_us > max_lat_us)
2618 				max_lat_us = total_latency_us;
2619 		}
2620 
2621 		apste = 1;
2622 
2623 		if (max_ps == -1) {
2624 			dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2625 		} else {
2626 			dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2627 				max_ps, max_lat_us, (int)sizeof(*table), table);
2628 		}
2629 	}
2630 
2631 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2632 				table, sizeof(*table), NULL);
2633 	if (ret)
2634 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2635 
2636 	kfree(table);
2637 	return ret;
2638 }
2639 
2640 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2641 {
2642 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2643 	u64 latency;
2644 
2645 	switch (val) {
2646 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2647 	case PM_QOS_LATENCY_ANY:
2648 		latency = U64_MAX;
2649 		break;
2650 
2651 	default:
2652 		latency = val;
2653 	}
2654 
2655 	if (ctrl->ps_max_latency_us != latency) {
2656 		ctrl->ps_max_latency_us = latency;
2657 		nvme_configure_apst(ctrl);
2658 	}
2659 }
2660 
2661 struct nvme_core_quirk_entry {
2662 	/*
2663 	 * NVMe model and firmware strings are padded with spaces.  For
2664 	 * simplicity, strings in the quirk table are padded with NULLs
2665 	 * instead.
2666 	 */
2667 	u16 vid;
2668 	const char *mn;
2669 	const char *fr;
2670 	unsigned long quirks;
2671 };
2672 
2673 static const struct nvme_core_quirk_entry core_quirks[] = {
2674 	{
2675 		/*
2676 		 * This Toshiba device seems to die using any APST states.  See:
2677 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2678 		 */
2679 		.vid = 0x1179,
2680 		.mn = "THNSF5256GPUK TOSHIBA",
2681 		.quirks = NVME_QUIRK_NO_APST,
2682 	},
2683 	{
2684 		/*
2685 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2686 		 * condition associated with actions related to suspend to idle
2687 		 * LiteON has resolved the problem in future firmware
2688 		 */
2689 		.vid = 0x14a4,
2690 		.fr = "22301111",
2691 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2692 	}
2693 };
2694 
2695 /* match is null-terminated but idstr is space-padded. */
2696 static bool string_matches(const char *idstr, const char *match, size_t len)
2697 {
2698 	size_t matchlen;
2699 
2700 	if (!match)
2701 		return true;
2702 
2703 	matchlen = strlen(match);
2704 	WARN_ON_ONCE(matchlen > len);
2705 
2706 	if (memcmp(idstr, match, matchlen))
2707 		return false;
2708 
2709 	for (; matchlen < len; matchlen++)
2710 		if (idstr[matchlen] != ' ')
2711 			return false;
2712 
2713 	return true;
2714 }
2715 
2716 static bool quirk_matches(const struct nvme_id_ctrl *id,
2717 			  const struct nvme_core_quirk_entry *q)
2718 {
2719 	return q->vid == le16_to_cpu(id->vid) &&
2720 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2721 		string_matches(id->fr, q->fr, sizeof(id->fr));
2722 }
2723 
2724 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2725 		struct nvme_id_ctrl *id)
2726 {
2727 	size_t nqnlen;
2728 	int off;
2729 
2730 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2731 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2732 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2733 			strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2734 			return;
2735 		}
2736 
2737 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2738 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2739 	}
2740 
2741 	/* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2742 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2743 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2744 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2745 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2746 	off += sizeof(id->sn);
2747 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2748 	off += sizeof(id->mn);
2749 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2750 }
2751 
2752 static void nvme_release_subsystem(struct device *dev)
2753 {
2754 	struct nvme_subsystem *subsys =
2755 		container_of(dev, struct nvme_subsystem, dev);
2756 
2757 	if (subsys->instance >= 0)
2758 		ida_simple_remove(&nvme_instance_ida, subsys->instance);
2759 	kfree(subsys);
2760 }
2761 
2762 static void nvme_destroy_subsystem(struct kref *ref)
2763 {
2764 	struct nvme_subsystem *subsys =
2765 			container_of(ref, struct nvme_subsystem, ref);
2766 
2767 	mutex_lock(&nvme_subsystems_lock);
2768 	list_del(&subsys->entry);
2769 	mutex_unlock(&nvme_subsystems_lock);
2770 
2771 	ida_destroy(&subsys->ns_ida);
2772 	device_del(&subsys->dev);
2773 	put_device(&subsys->dev);
2774 }
2775 
2776 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2777 {
2778 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2779 }
2780 
2781 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2782 {
2783 	struct nvme_subsystem *subsys;
2784 
2785 	lockdep_assert_held(&nvme_subsystems_lock);
2786 
2787 	/*
2788 	 * Fail matches for discovery subsystems. This results
2789 	 * in each discovery controller bound to a unique subsystem.
2790 	 * This avoids issues with validating controller values
2791 	 * that can only be true when there is a single unique subsystem.
2792 	 * There may be multiple and completely independent entities
2793 	 * that provide discovery controllers.
2794 	 */
2795 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2796 		return NULL;
2797 
2798 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2799 		if (strcmp(subsys->subnqn, subsysnqn))
2800 			continue;
2801 		if (!kref_get_unless_zero(&subsys->ref))
2802 			continue;
2803 		return subsys;
2804 	}
2805 
2806 	return NULL;
2807 }
2808 
2809 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2810 	struct device_attribute subsys_attr_##_name = \
2811 		__ATTR(_name, _mode, _show, NULL)
2812 
2813 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2814 				    struct device_attribute *attr,
2815 				    char *buf)
2816 {
2817 	struct nvme_subsystem *subsys =
2818 		container_of(dev, struct nvme_subsystem, dev);
2819 
2820 	return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2821 }
2822 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2823 
2824 #define nvme_subsys_show_str_function(field)				\
2825 static ssize_t subsys_##field##_show(struct device *dev,		\
2826 			    struct device_attribute *attr, char *buf)	\
2827 {									\
2828 	struct nvme_subsystem *subsys =					\
2829 		container_of(dev, struct nvme_subsystem, dev);		\
2830 	return sprintf(buf, "%.*s\n",					\
2831 		       (int)sizeof(subsys->field), subsys->field);	\
2832 }									\
2833 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2834 
2835 nvme_subsys_show_str_function(model);
2836 nvme_subsys_show_str_function(serial);
2837 nvme_subsys_show_str_function(firmware_rev);
2838 
2839 static struct attribute *nvme_subsys_attrs[] = {
2840 	&subsys_attr_model.attr,
2841 	&subsys_attr_serial.attr,
2842 	&subsys_attr_firmware_rev.attr,
2843 	&subsys_attr_subsysnqn.attr,
2844 #ifdef CONFIG_NVME_MULTIPATH
2845 	&subsys_attr_iopolicy.attr,
2846 #endif
2847 	NULL,
2848 };
2849 
2850 static struct attribute_group nvme_subsys_attrs_group = {
2851 	.attrs = nvme_subsys_attrs,
2852 };
2853 
2854 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2855 	&nvme_subsys_attrs_group,
2856 	NULL,
2857 };
2858 
2859 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2860 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2861 {
2862 	struct nvme_ctrl *tmp;
2863 
2864 	lockdep_assert_held(&nvme_subsystems_lock);
2865 
2866 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2867 		if (nvme_state_terminal(tmp))
2868 			continue;
2869 
2870 		if (tmp->cntlid == ctrl->cntlid) {
2871 			dev_err(ctrl->device,
2872 				"Duplicate cntlid %u with %s, rejecting\n",
2873 				ctrl->cntlid, dev_name(tmp->device));
2874 			return false;
2875 		}
2876 
2877 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2878 		    (ctrl->opts && ctrl->opts->discovery_nqn))
2879 			continue;
2880 
2881 		dev_err(ctrl->device,
2882 			"Subsystem does not support multiple controllers\n");
2883 		return false;
2884 	}
2885 
2886 	return true;
2887 }
2888 
2889 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2890 {
2891 	struct nvme_subsystem *subsys, *found;
2892 	int ret;
2893 
2894 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2895 	if (!subsys)
2896 		return -ENOMEM;
2897 
2898 	subsys->instance = -1;
2899 	mutex_init(&subsys->lock);
2900 	kref_init(&subsys->ref);
2901 	INIT_LIST_HEAD(&subsys->ctrls);
2902 	INIT_LIST_HEAD(&subsys->nsheads);
2903 	nvme_init_subnqn(subsys, ctrl, id);
2904 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2905 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2906 	memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2907 	subsys->vendor_id = le16_to_cpu(id->vid);
2908 	subsys->cmic = id->cmic;
2909 	subsys->awupf = le16_to_cpu(id->awupf);
2910 #ifdef CONFIG_NVME_MULTIPATH
2911 	subsys->iopolicy = NVME_IOPOLICY_NUMA;
2912 #endif
2913 
2914 	subsys->dev.class = nvme_subsys_class;
2915 	subsys->dev.release = nvme_release_subsystem;
2916 	subsys->dev.groups = nvme_subsys_attrs_groups;
2917 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2918 	device_initialize(&subsys->dev);
2919 
2920 	mutex_lock(&nvme_subsystems_lock);
2921 	found = __nvme_find_get_subsystem(subsys->subnqn);
2922 	if (found) {
2923 		put_device(&subsys->dev);
2924 		subsys = found;
2925 
2926 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2927 			ret = -EINVAL;
2928 			goto out_put_subsystem;
2929 		}
2930 	} else {
2931 		ret = device_add(&subsys->dev);
2932 		if (ret) {
2933 			dev_err(ctrl->device,
2934 				"failed to register subsystem device.\n");
2935 			put_device(&subsys->dev);
2936 			goto out_unlock;
2937 		}
2938 		ida_init(&subsys->ns_ida);
2939 		list_add_tail(&subsys->entry, &nvme_subsystems);
2940 	}
2941 
2942 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2943 				dev_name(ctrl->device));
2944 	if (ret) {
2945 		dev_err(ctrl->device,
2946 			"failed to create sysfs link from subsystem.\n");
2947 		goto out_put_subsystem;
2948 	}
2949 
2950 	if (!found)
2951 		subsys->instance = ctrl->instance;
2952 	ctrl->subsys = subsys;
2953 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2954 	mutex_unlock(&nvme_subsystems_lock);
2955 	return 0;
2956 
2957 out_put_subsystem:
2958 	nvme_put_subsystem(subsys);
2959 out_unlock:
2960 	mutex_unlock(&nvme_subsystems_lock);
2961 	return ret;
2962 }
2963 
2964 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2965 		void *log, size_t size, u64 offset)
2966 {
2967 	struct nvme_command c = { };
2968 	u32 dwlen = nvme_bytes_to_numd(size);
2969 
2970 	c.get_log_page.opcode = nvme_admin_get_log_page;
2971 	c.get_log_page.nsid = cpu_to_le32(nsid);
2972 	c.get_log_page.lid = log_page;
2973 	c.get_log_page.lsp = lsp;
2974 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2975 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2976 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2977 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2978 	c.get_log_page.csi = csi;
2979 
2980 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2981 }
2982 
2983 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2984 				struct nvme_effects_log **log)
2985 {
2986 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2987 	int ret;
2988 
2989 	if (cel)
2990 		goto out;
2991 
2992 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2993 	if (!cel)
2994 		return -ENOMEM;
2995 
2996 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2997 			cel, sizeof(*cel), 0);
2998 	if (ret) {
2999 		kfree(cel);
3000 		return ret;
3001 	}
3002 
3003 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3004 out:
3005 	*log = cel;
3006 	return 0;
3007 }
3008 
3009 /*
3010  * Initialize the cached copies of the Identify data and various controller
3011  * register in our nvme_ctrl structure.  This should be called as soon as
3012  * the admin queue is fully up and running.
3013  */
3014 int nvme_init_identify(struct nvme_ctrl *ctrl)
3015 {
3016 	struct nvme_id_ctrl *id;
3017 	int ret, page_shift;
3018 	u32 max_hw_sectors;
3019 	bool prev_apst_enabled;
3020 
3021 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3022 	if (ret) {
3023 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3024 		return ret;
3025 	}
3026 	page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3027 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3028 
3029 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3030 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3031 
3032 	ret = nvme_identify_ctrl(ctrl, &id);
3033 	if (ret) {
3034 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3035 		return -EIO;
3036 	}
3037 
3038 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3039 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3040 		if (ret < 0)
3041 			goto out_free;
3042 	}
3043 
3044 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3045 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3046 
3047 	if (!ctrl->identified) {
3048 		int i;
3049 
3050 		ret = nvme_init_subsystem(ctrl, id);
3051 		if (ret)
3052 			goto out_free;
3053 
3054 		/*
3055 		 * Check for quirks.  Quirk can depend on firmware version,
3056 		 * so, in principle, the set of quirks present can change
3057 		 * across a reset.  As a possible future enhancement, we
3058 		 * could re-scan for quirks every time we reinitialize
3059 		 * the device, but we'd have to make sure that the driver
3060 		 * behaves intelligently if the quirks change.
3061 		 */
3062 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3063 			if (quirk_matches(id, &core_quirks[i]))
3064 				ctrl->quirks |= core_quirks[i].quirks;
3065 		}
3066 	}
3067 
3068 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3069 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3070 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3071 	}
3072 
3073 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3074 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3075 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3076 
3077 	ctrl->oacs = le16_to_cpu(id->oacs);
3078 	ctrl->oncs = le16_to_cpu(id->oncs);
3079 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3080 	ctrl->oaes = le32_to_cpu(id->oaes);
3081 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3082 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3083 
3084 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3085 	ctrl->vwc = id->vwc;
3086 	if (id->mdts)
3087 		max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3088 	else
3089 		max_hw_sectors = UINT_MAX;
3090 	ctrl->max_hw_sectors =
3091 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3092 
3093 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3094 	ctrl->sgls = le32_to_cpu(id->sgls);
3095 	ctrl->kas = le16_to_cpu(id->kas);
3096 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3097 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3098 
3099 	if (id->rtd3e) {
3100 		/* us -> s */
3101 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3102 
3103 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3104 						 shutdown_timeout, 60);
3105 
3106 		if (ctrl->shutdown_timeout != shutdown_timeout)
3107 			dev_info(ctrl->device,
3108 				 "Shutdown timeout set to %u seconds\n",
3109 				 ctrl->shutdown_timeout);
3110 	} else
3111 		ctrl->shutdown_timeout = shutdown_timeout;
3112 
3113 	ctrl->npss = id->npss;
3114 	ctrl->apsta = id->apsta;
3115 	prev_apst_enabled = ctrl->apst_enabled;
3116 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3117 		if (force_apst && id->apsta) {
3118 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3119 			ctrl->apst_enabled = true;
3120 		} else {
3121 			ctrl->apst_enabled = false;
3122 		}
3123 	} else {
3124 		ctrl->apst_enabled = id->apsta;
3125 	}
3126 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3127 
3128 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3129 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3130 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3131 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3132 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3133 
3134 		/*
3135 		 * In fabrics we need to verify the cntlid matches the
3136 		 * admin connect
3137 		 */
3138 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3139 			dev_err(ctrl->device,
3140 				"Mismatching cntlid: Connect %u vs Identify "
3141 				"%u, rejecting\n",
3142 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3143 			ret = -EINVAL;
3144 			goto out_free;
3145 		}
3146 
3147 		if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
3148 			dev_err(ctrl->device,
3149 				"keep-alive support is mandatory for fabrics\n");
3150 			ret = -EINVAL;
3151 			goto out_free;
3152 		}
3153 	} else {
3154 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3155 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3156 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3157 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3158 	}
3159 
3160 	ret = nvme_mpath_init(ctrl, id);
3161 	kfree(id);
3162 
3163 	if (ret < 0)
3164 		return ret;
3165 
3166 	if (ctrl->apst_enabled && !prev_apst_enabled)
3167 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3168 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3169 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3170 
3171 	ret = nvme_configure_apst(ctrl);
3172 	if (ret < 0)
3173 		return ret;
3174 
3175 	ret = nvme_configure_timestamp(ctrl);
3176 	if (ret < 0)
3177 		return ret;
3178 
3179 	ret = nvme_configure_directives(ctrl);
3180 	if (ret < 0)
3181 		return ret;
3182 
3183 	ret = nvme_configure_acre(ctrl);
3184 	if (ret < 0)
3185 		return ret;
3186 
3187 	if (!ctrl->identified) {
3188 		ret = nvme_hwmon_init(ctrl);
3189 		if (ret < 0)
3190 			return ret;
3191 	}
3192 
3193 	ctrl->identified = true;
3194 
3195 	return 0;
3196 
3197 out_free:
3198 	kfree(id);
3199 	return ret;
3200 }
3201 EXPORT_SYMBOL_GPL(nvme_init_identify);
3202 
3203 static int nvme_dev_open(struct inode *inode, struct file *file)
3204 {
3205 	struct nvme_ctrl *ctrl =
3206 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3207 
3208 	switch (ctrl->state) {
3209 	case NVME_CTRL_LIVE:
3210 		break;
3211 	default:
3212 		return -EWOULDBLOCK;
3213 	}
3214 
3215 	nvme_get_ctrl(ctrl);
3216 	if (!try_module_get(ctrl->ops->module)) {
3217 		nvme_put_ctrl(ctrl);
3218 		return -EINVAL;
3219 	}
3220 
3221 	file->private_data = ctrl;
3222 	return 0;
3223 }
3224 
3225 static int nvme_dev_release(struct inode *inode, struct file *file)
3226 {
3227 	struct nvme_ctrl *ctrl =
3228 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3229 
3230 	module_put(ctrl->ops->module);
3231 	nvme_put_ctrl(ctrl);
3232 	return 0;
3233 }
3234 
3235 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3236 {
3237 	struct nvme_ns *ns;
3238 	int ret;
3239 
3240 	down_read(&ctrl->namespaces_rwsem);
3241 	if (list_empty(&ctrl->namespaces)) {
3242 		ret = -ENOTTY;
3243 		goto out_unlock;
3244 	}
3245 
3246 	ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3247 	if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3248 		dev_warn(ctrl->device,
3249 			"NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3250 		ret = -EINVAL;
3251 		goto out_unlock;
3252 	}
3253 
3254 	dev_warn(ctrl->device,
3255 		"using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3256 	kref_get(&ns->kref);
3257 	up_read(&ctrl->namespaces_rwsem);
3258 
3259 	ret = nvme_user_cmd(ctrl, ns, argp);
3260 	nvme_put_ns(ns);
3261 	return ret;
3262 
3263 out_unlock:
3264 	up_read(&ctrl->namespaces_rwsem);
3265 	return ret;
3266 }
3267 
3268 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3269 		unsigned long arg)
3270 {
3271 	struct nvme_ctrl *ctrl = file->private_data;
3272 	void __user *argp = (void __user *)arg;
3273 
3274 	switch (cmd) {
3275 	case NVME_IOCTL_ADMIN_CMD:
3276 		return nvme_user_cmd(ctrl, NULL, argp);
3277 	case NVME_IOCTL_ADMIN64_CMD:
3278 		return nvme_user_cmd64(ctrl, NULL, argp);
3279 	case NVME_IOCTL_IO_CMD:
3280 		return nvme_dev_user_cmd(ctrl, argp);
3281 	case NVME_IOCTL_RESET:
3282 		dev_warn(ctrl->device, "resetting controller\n");
3283 		return nvme_reset_ctrl_sync(ctrl);
3284 	case NVME_IOCTL_SUBSYS_RESET:
3285 		return nvme_reset_subsystem(ctrl);
3286 	case NVME_IOCTL_RESCAN:
3287 		nvme_queue_scan(ctrl);
3288 		return 0;
3289 	default:
3290 		return -ENOTTY;
3291 	}
3292 }
3293 
3294 static const struct file_operations nvme_dev_fops = {
3295 	.owner		= THIS_MODULE,
3296 	.open		= nvme_dev_open,
3297 	.release	= nvme_dev_release,
3298 	.unlocked_ioctl	= nvme_dev_ioctl,
3299 	.compat_ioctl	= compat_ptr_ioctl,
3300 };
3301 
3302 static ssize_t nvme_sysfs_reset(struct device *dev,
3303 				struct device_attribute *attr, const char *buf,
3304 				size_t count)
3305 {
3306 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3307 	int ret;
3308 
3309 	ret = nvme_reset_ctrl_sync(ctrl);
3310 	if (ret < 0)
3311 		return ret;
3312 	return count;
3313 }
3314 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3315 
3316 static ssize_t nvme_sysfs_rescan(struct device *dev,
3317 				struct device_attribute *attr, const char *buf,
3318 				size_t count)
3319 {
3320 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3321 
3322 	nvme_queue_scan(ctrl);
3323 	return count;
3324 }
3325 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3326 
3327 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3328 {
3329 	struct gendisk *disk = dev_to_disk(dev);
3330 
3331 	if (disk->fops == &nvme_bdev_ops)
3332 		return nvme_get_ns_from_dev(dev)->head;
3333 	else
3334 		return disk->private_data;
3335 }
3336 
3337 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3338 		char *buf)
3339 {
3340 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3341 	struct nvme_ns_ids *ids = &head->ids;
3342 	struct nvme_subsystem *subsys = head->subsys;
3343 	int serial_len = sizeof(subsys->serial);
3344 	int model_len = sizeof(subsys->model);
3345 
3346 	if (!uuid_is_null(&ids->uuid))
3347 		return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3348 
3349 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3350 		return sprintf(buf, "eui.%16phN\n", ids->nguid);
3351 
3352 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3353 		return sprintf(buf, "eui.%8phN\n", ids->eui64);
3354 
3355 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3356 				  subsys->serial[serial_len - 1] == '\0'))
3357 		serial_len--;
3358 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3359 				 subsys->model[model_len - 1] == '\0'))
3360 		model_len--;
3361 
3362 	return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3363 		serial_len, subsys->serial, model_len, subsys->model,
3364 		head->ns_id);
3365 }
3366 static DEVICE_ATTR_RO(wwid);
3367 
3368 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3369 		char *buf)
3370 {
3371 	return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3372 }
3373 static DEVICE_ATTR_RO(nguid);
3374 
3375 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3376 		char *buf)
3377 {
3378 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3379 
3380 	/* For backward compatibility expose the NGUID to userspace if
3381 	 * we have no UUID set
3382 	 */
3383 	if (uuid_is_null(&ids->uuid)) {
3384 		printk_ratelimited(KERN_WARNING
3385 				   "No UUID available providing old NGUID\n");
3386 		return sprintf(buf, "%pU\n", ids->nguid);
3387 	}
3388 	return sprintf(buf, "%pU\n", &ids->uuid);
3389 }
3390 static DEVICE_ATTR_RO(uuid);
3391 
3392 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3393 		char *buf)
3394 {
3395 	return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3396 }
3397 static DEVICE_ATTR_RO(eui);
3398 
3399 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3400 		char *buf)
3401 {
3402 	return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3403 }
3404 static DEVICE_ATTR_RO(nsid);
3405 
3406 static struct attribute *nvme_ns_id_attrs[] = {
3407 	&dev_attr_wwid.attr,
3408 	&dev_attr_uuid.attr,
3409 	&dev_attr_nguid.attr,
3410 	&dev_attr_eui.attr,
3411 	&dev_attr_nsid.attr,
3412 #ifdef CONFIG_NVME_MULTIPATH
3413 	&dev_attr_ana_grpid.attr,
3414 	&dev_attr_ana_state.attr,
3415 #endif
3416 	NULL,
3417 };
3418 
3419 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3420 		struct attribute *a, int n)
3421 {
3422 	struct device *dev = container_of(kobj, struct device, kobj);
3423 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3424 
3425 	if (a == &dev_attr_uuid.attr) {
3426 		if (uuid_is_null(&ids->uuid) &&
3427 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3428 			return 0;
3429 	}
3430 	if (a == &dev_attr_nguid.attr) {
3431 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3432 			return 0;
3433 	}
3434 	if (a == &dev_attr_eui.attr) {
3435 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3436 			return 0;
3437 	}
3438 #ifdef CONFIG_NVME_MULTIPATH
3439 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3440 		if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3441 			return 0;
3442 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3443 			return 0;
3444 	}
3445 #endif
3446 	return a->mode;
3447 }
3448 
3449 static const struct attribute_group nvme_ns_id_attr_group = {
3450 	.attrs		= nvme_ns_id_attrs,
3451 	.is_visible	= nvme_ns_id_attrs_are_visible,
3452 };
3453 
3454 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3455 	&nvme_ns_id_attr_group,
3456 #ifdef CONFIG_NVM
3457 	&nvme_nvm_attr_group,
3458 #endif
3459 	NULL,
3460 };
3461 
3462 #define nvme_show_str_function(field)						\
3463 static ssize_t  field##_show(struct device *dev,				\
3464 			    struct device_attribute *attr, char *buf)		\
3465 {										\
3466         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3467         return sprintf(buf, "%.*s\n",						\
3468 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3469 }										\
3470 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3471 
3472 nvme_show_str_function(model);
3473 nvme_show_str_function(serial);
3474 nvme_show_str_function(firmware_rev);
3475 
3476 #define nvme_show_int_function(field)						\
3477 static ssize_t  field##_show(struct device *dev,				\
3478 			    struct device_attribute *attr, char *buf)		\
3479 {										\
3480         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3481         return sprintf(buf, "%d\n", ctrl->field);	\
3482 }										\
3483 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3484 
3485 nvme_show_int_function(cntlid);
3486 nvme_show_int_function(numa_node);
3487 nvme_show_int_function(queue_count);
3488 nvme_show_int_function(sqsize);
3489 
3490 static ssize_t nvme_sysfs_delete(struct device *dev,
3491 				struct device_attribute *attr, const char *buf,
3492 				size_t count)
3493 {
3494 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3495 
3496 	if (device_remove_file_self(dev, attr))
3497 		nvme_delete_ctrl_sync(ctrl);
3498 	return count;
3499 }
3500 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3501 
3502 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3503 					 struct device_attribute *attr,
3504 					 char *buf)
3505 {
3506 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3507 
3508 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3509 }
3510 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3511 
3512 static ssize_t nvme_sysfs_show_state(struct device *dev,
3513 				     struct device_attribute *attr,
3514 				     char *buf)
3515 {
3516 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3517 	static const char *const state_name[] = {
3518 		[NVME_CTRL_NEW]		= "new",
3519 		[NVME_CTRL_LIVE]	= "live",
3520 		[NVME_CTRL_RESETTING]	= "resetting",
3521 		[NVME_CTRL_CONNECTING]	= "connecting",
3522 		[NVME_CTRL_DELETING]	= "deleting",
3523 		[NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3524 		[NVME_CTRL_DEAD]	= "dead",
3525 	};
3526 
3527 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3528 	    state_name[ctrl->state])
3529 		return sprintf(buf, "%s\n", state_name[ctrl->state]);
3530 
3531 	return sprintf(buf, "unknown state\n");
3532 }
3533 
3534 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3535 
3536 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3537 					 struct device_attribute *attr,
3538 					 char *buf)
3539 {
3540 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3541 
3542 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3543 }
3544 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3545 
3546 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3547 					struct device_attribute *attr,
3548 					char *buf)
3549 {
3550 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3551 
3552 	return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3553 }
3554 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3555 
3556 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3557 					struct device_attribute *attr,
3558 					char *buf)
3559 {
3560 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3561 
3562 	return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3563 }
3564 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3565 
3566 static ssize_t nvme_sysfs_show_address(struct device *dev,
3567 					 struct device_attribute *attr,
3568 					 char *buf)
3569 {
3570 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3571 
3572 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3573 }
3574 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3575 
3576 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3577 		struct device_attribute *attr, char *buf)
3578 {
3579 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3580 	struct nvmf_ctrl_options *opts = ctrl->opts;
3581 
3582 	if (ctrl->opts->max_reconnects == -1)
3583 		return sprintf(buf, "off\n");
3584 	return sprintf(buf, "%d\n",
3585 			opts->max_reconnects * opts->reconnect_delay);
3586 }
3587 
3588 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3589 		struct device_attribute *attr, const char *buf, size_t count)
3590 {
3591 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3592 	struct nvmf_ctrl_options *opts = ctrl->opts;
3593 	int ctrl_loss_tmo, err;
3594 
3595 	err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3596 	if (err)
3597 		return -EINVAL;
3598 
3599 	else if (ctrl_loss_tmo < 0)
3600 		opts->max_reconnects = -1;
3601 	else
3602 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3603 						opts->reconnect_delay);
3604 	return count;
3605 }
3606 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3607 	nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3608 
3609 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3610 		struct device_attribute *attr, char *buf)
3611 {
3612 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3613 
3614 	if (ctrl->opts->reconnect_delay == -1)
3615 		return sprintf(buf, "off\n");
3616 	return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3617 }
3618 
3619 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3620 		struct device_attribute *attr, const char *buf, size_t count)
3621 {
3622 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3623 	unsigned int v;
3624 	int err;
3625 
3626 	err = kstrtou32(buf, 10, &v);
3627 	if (err)
3628 		return err;
3629 
3630 	ctrl->opts->reconnect_delay = v;
3631 	return count;
3632 }
3633 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3634 	nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3635 
3636 static struct attribute *nvme_dev_attrs[] = {
3637 	&dev_attr_reset_controller.attr,
3638 	&dev_attr_rescan_controller.attr,
3639 	&dev_attr_model.attr,
3640 	&dev_attr_serial.attr,
3641 	&dev_attr_firmware_rev.attr,
3642 	&dev_attr_cntlid.attr,
3643 	&dev_attr_delete_controller.attr,
3644 	&dev_attr_transport.attr,
3645 	&dev_attr_subsysnqn.attr,
3646 	&dev_attr_address.attr,
3647 	&dev_attr_state.attr,
3648 	&dev_attr_numa_node.attr,
3649 	&dev_attr_queue_count.attr,
3650 	&dev_attr_sqsize.attr,
3651 	&dev_attr_hostnqn.attr,
3652 	&dev_attr_hostid.attr,
3653 	&dev_attr_ctrl_loss_tmo.attr,
3654 	&dev_attr_reconnect_delay.attr,
3655 	NULL
3656 };
3657 
3658 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3659 		struct attribute *a, int n)
3660 {
3661 	struct device *dev = container_of(kobj, struct device, kobj);
3662 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3663 
3664 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3665 		return 0;
3666 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3667 		return 0;
3668 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3669 		return 0;
3670 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3671 		return 0;
3672 	if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3673 		return 0;
3674 	if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3675 		return 0;
3676 
3677 	return a->mode;
3678 }
3679 
3680 static struct attribute_group nvme_dev_attrs_group = {
3681 	.attrs		= nvme_dev_attrs,
3682 	.is_visible	= nvme_dev_attrs_are_visible,
3683 };
3684 
3685 static const struct attribute_group *nvme_dev_attr_groups[] = {
3686 	&nvme_dev_attrs_group,
3687 	NULL,
3688 };
3689 
3690 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3691 		unsigned nsid)
3692 {
3693 	struct nvme_ns_head *h;
3694 
3695 	lockdep_assert_held(&subsys->lock);
3696 
3697 	list_for_each_entry(h, &subsys->nsheads, entry) {
3698 		if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3699 			return h;
3700 	}
3701 
3702 	return NULL;
3703 }
3704 
3705 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3706 		struct nvme_ns_head *new)
3707 {
3708 	struct nvme_ns_head *h;
3709 
3710 	lockdep_assert_held(&subsys->lock);
3711 
3712 	list_for_each_entry(h, &subsys->nsheads, entry) {
3713 		if (nvme_ns_ids_valid(&new->ids) &&
3714 		    nvme_ns_ids_equal(&new->ids, &h->ids))
3715 			return -EINVAL;
3716 	}
3717 
3718 	return 0;
3719 }
3720 
3721 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3722 		unsigned nsid, struct nvme_ns_ids *ids)
3723 {
3724 	struct nvme_ns_head *head;
3725 	size_t size = sizeof(*head);
3726 	int ret = -ENOMEM;
3727 
3728 #ifdef CONFIG_NVME_MULTIPATH
3729 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
3730 #endif
3731 
3732 	head = kzalloc(size, GFP_KERNEL);
3733 	if (!head)
3734 		goto out;
3735 	ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3736 	if (ret < 0)
3737 		goto out_free_head;
3738 	head->instance = ret;
3739 	INIT_LIST_HEAD(&head->list);
3740 	ret = init_srcu_struct(&head->srcu);
3741 	if (ret)
3742 		goto out_ida_remove;
3743 	head->subsys = ctrl->subsys;
3744 	head->ns_id = nsid;
3745 	head->ids = *ids;
3746 	kref_init(&head->ref);
3747 
3748 	ret = __nvme_check_ids(ctrl->subsys, head);
3749 	if (ret) {
3750 		dev_err(ctrl->device,
3751 			"duplicate IDs for nsid %d\n", nsid);
3752 		goto out_cleanup_srcu;
3753 	}
3754 
3755 	if (head->ids.csi) {
3756 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3757 		if (ret)
3758 			goto out_cleanup_srcu;
3759 	} else
3760 		head->effects = ctrl->effects;
3761 
3762 	ret = nvme_mpath_alloc_disk(ctrl, head);
3763 	if (ret)
3764 		goto out_cleanup_srcu;
3765 
3766 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3767 
3768 	kref_get(&ctrl->subsys->ref);
3769 
3770 	return head;
3771 out_cleanup_srcu:
3772 	cleanup_srcu_struct(&head->srcu);
3773 out_ida_remove:
3774 	ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3775 out_free_head:
3776 	kfree(head);
3777 out:
3778 	if (ret > 0)
3779 		ret = blk_status_to_errno(nvme_error_status(ret));
3780 	return ERR_PTR(ret);
3781 }
3782 
3783 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3784 		struct nvme_ns_ids *ids, bool is_shared)
3785 {
3786 	struct nvme_ctrl *ctrl = ns->ctrl;
3787 	struct nvme_ns_head *head = NULL;
3788 	int ret = 0;
3789 
3790 	mutex_lock(&ctrl->subsys->lock);
3791 	head = nvme_find_ns_head(ctrl->subsys, nsid);
3792 	if (!head) {
3793 		head = nvme_alloc_ns_head(ctrl, nsid, ids);
3794 		if (IS_ERR(head)) {
3795 			ret = PTR_ERR(head);
3796 			goto out_unlock;
3797 		}
3798 		head->shared = is_shared;
3799 	} else {
3800 		ret = -EINVAL;
3801 		if (!is_shared || !head->shared) {
3802 			dev_err(ctrl->device,
3803 				"Duplicate unshared namespace %d\n", nsid);
3804 			goto out_put_ns_head;
3805 		}
3806 		if (!nvme_ns_ids_equal(&head->ids, ids)) {
3807 			dev_err(ctrl->device,
3808 				"IDs don't match for shared namespace %d\n",
3809 					nsid);
3810 			goto out_put_ns_head;
3811 		}
3812 	}
3813 
3814 	list_add_tail(&ns->siblings, &head->list);
3815 	ns->head = head;
3816 	mutex_unlock(&ctrl->subsys->lock);
3817 	return 0;
3818 
3819 out_put_ns_head:
3820 	nvme_put_ns_head(head);
3821 out_unlock:
3822 	mutex_unlock(&ctrl->subsys->lock);
3823 	return ret;
3824 }
3825 
3826 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3827 {
3828 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3829 	struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3830 
3831 	return nsa->head->ns_id - nsb->head->ns_id;
3832 }
3833 
3834 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3835 {
3836 	struct nvme_ns *ns, *ret = NULL;
3837 
3838 	down_read(&ctrl->namespaces_rwsem);
3839 	list_for_each_entry(ns, &ctrl->namespaces, list) {
3840 		if (ns->head->ns_id == nsid) {
3841 			if (!kref_get_unless_zero(&ns->kref))
3842 				continue;
3843 			ret = ns;
3844 			break;
3845 		}
3846 		if (ns->head->ns_id > nsid)
3847 			break;
3848 	}
3849 	up_read(&ctrl->namespaces_rwsem);
3850 	return ret;
3851 }
3852 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3853 
3854 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3855 		struct nvme_ns_ids *ids)
3856 {
3857 	struct nvme_ns *ns;
3858 	struct gendisk *disk;
3859 	struct nvme_id_ns *id;
3860 	char disk_name[DISK_NAME_LEN];
3861 	int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3862 
3863 	if (nvme_identify_ns(ctrl, nsid, ids, &id))
3864 		return;
3865 
3866 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3867 	if (!ns)
3868 		goto out_free_id;
3869 
3870 	ns->queue = blk_mq_init_queue(ctrl->tagset);
3871 	if (IS_ERR(ns->queue))
3872 		goto out_free_ns;
3873 
3874 	if (ctrl->opts && ctrl->opts->data_digest)
3875 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3876 
3877 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3878 	if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3879 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3880 
3881 	ns->queue->queuedata = ns;
3882 	ns->ctrl = ctrl;
3883 	kref_init(&ns->kref);
3884 
3885 	if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3886 		goto out_free_queue;
3887 	nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3888 
3889 	disk = alloc_disk_node(0, node);
3890 	if (!disk)
3891 		goto out_unlink_ns;
3892 
3893 	disk->fops = &nvme_bdev_ops;
3894 	disk->private_data = ns;
3895 	disk->queue = ns->queue;
3896 	disk->flags = flags;
3897 	memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3898 	ns->disk = disk;
3899 
3900 	if (nvme_update_ns_info(ns, id))
3901 		goto out_put_disk;
3902 
3903 	if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3904 		if (nvme_nvm_register(ns, disk_name, node)) {
3905 			dev_warn(ctrl->device, "LightNVM init failure\n");
3906 			goto out_put_disk;
3907 		}
3908 	}
3909 
3910 	down_write(&ctrl->namespaces_rwsem);
3911 	list_add_tail(&ns->list, &ctrl->namespaces);
3912 	up_write(&ctrl->namespaces_rwsem);
3913 
3914 	nvme_get_ctrl(ctrl);
3915 
3916 	device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3917 
3918 	nvme_mpath_add_disk(ns, id);
3919 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3920 	kfree(id);
3921 
3922 	return;
3923  out_put_disk:
3924 	/* prevent double queue cleanup */
3925 	ns->disk->queue = NULL;
3926 	put_disk(ns->disk);
3927  out_unlink_ns:
3928 	mutex_lock(&ctrl->subsys->lock);
3929 	list_del_rcu(&ns->siblings);
3930 	if (list_empty(&ns->head->list))
3931 		list_del_init(&ns->head->entry);
3932 	mutex_unlock(&ctrl->subsys->lock);
3933 	nvme_put_ns_head(ns->head);
3934  out_free_queue:
3935 	blk_cleanup_queue(ns->queue);
3936  out_free_ns:
3937 	kfree(ns);
3938  out_free_id:
3939 	kfree(id);
3940 }
3941 
3942 static void nvme_ns_remove(struct nvme_ns *ns)
3943 {
3944 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3945 		return;
3946 
3947 	set_capacity(ns->disk, 0);
3948 	nvme_fault_inject_fini(&ns->fault_inject);
3949 
3950 	mutex_lock(&ns->ctrl->subsys->lock);
3951 	list_del_rcu(&ns->siblings);
3952 	if (list_empty(&ns->head->list))
3953 		list_del_init(&ns->head->entry);
3954 	mutex_unlock(&ns->ctrl->subsys->lock);
3955 
3956 	synchronize_rcu(); /* guarantee not available in head->list */
3957 	nvme_mpath_clear_current_path(ns);
3958 	synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3959 
3960 	if (ns->disk->flags & GENHD_FL_UP) {
3961 		del_gendisk(ns->disk);
3962 		blk_cleanup_queue(ns->queue);
3963 		if (blk_get_integrity(ns->disk))
3964 			blk_integrity_unregister(ns->disk);
3965 	}
3966 
3967 	down_write(&ns->ctrl->namespaces_rwsem);
3968 	list_del_init(&ns->list);
3969 	up_write(&ns->ctrl->namespaces_rwsem);
3970 
3971 	nvme_mpath_check_last_path(ns);
3972 	nvme_put_ns(ns);
3973 }
3974 
3975 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3976 {
3977 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3978 
3979 	if (ns) {
3980 		nvme_ns_remove(ns);
3981 		nvme_put_ns(ns);
3982 	}
3983 }
3984 
3985 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3986 {
3987 	struct nvme_id_ns *id;
3988 	int ret = -ENODEV;
3989 
3990 	if (test_bit(NVME_NS_DEAD, &ns->flags))
3991 		goto out;
3992 
3993 	ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3994 	if (ret)
3995 		goto out;
3996 
3997 	ret = -ENODEV;
3998 	if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3999 		dev_err(ns->ctrl->device,
4000 			"identifiers changed for nsid %d\n", ns->head->ns_id);
4001 		goto out_free_id;
4002 	}
4003 
4004 	ret = nvme_update_ns_info(ns, id);
4005 
4006 out_free_id:
4007 	kfree(id);
4008 out:
4009 	/*
4010 	 * Only remove the namespace if we got a fatal error back from the
4011 	 * device, otherwise ignore the error and just move on.
4012 	 *
4013 	 * TODO: we should probably schedule a delayed retry here.
4014 	 */
4015 	if (ret && ret != -ENOMEM && !(ret > 0 && !(ret & NVME_SC_DNR)))
4016 		nvme_ns_remove(ns);
4017 }
4018 
4019 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4020 {
4021 	struct nvme_ns_ids ids = { };
4022 	struct nvme_ns *ns;
4023 
4024 	if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4025 		return;
4026 
4027 	ns = nvme_find_get_ns(ctrl, nsid);
4028 	if (ns) {
4029 		nvme_validate_ns(ns, &ids);
4030 		nvme_put_ns(ns);
4031 		return;
4032 	}
4033 
4034 	switch (ids.csi) {
4035 	case NVME_CSI_NVM:
4036 		nvme_alloc_ns(ctrl, nsid, &ids);
4037 		break;
4038 	case NVME_CSI_ZNS:
4039 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4040 			dev_warn(ctrl->device,
4041 				"nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4042 				nsid);
4043 			break;
4044 		}
4045 		nvme_alloc_ns(ctrl, nsid, &ids);
4046 		break;
4047 	default:
4048 		dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4049 			ids.csi, nsid);
4050 		break;
4051 	}
4052 }
4053 
4054 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4055 					unsigned nsid)
4056 {
4057 	struct nvme_ns *ns, *next;
4058 	LIST_HEAD(rm_list);
4059 
4060 	down_write(&ctrl->namespaces_rwsem);
4061 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4062 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4063 			list_move_tail(&ns->list, &rm_list);
4064 	}
4065 	up_write(&ctrl->namespaces_rwsem);
4066 
4067 	list_for_each_entry_safe(ns, next, &rm_list, list)
4068 		nvme_ns_remove(ns);
4069 
4070 }
4071 
4072 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4073 {
4074 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4075 	__le32 *ns_list;
4076 	u32 prev = 0;
4077 	int ret = 0, i;
4078 
4079 	if (nvme_ctrl_limited_cns(ctrl))
4080 		return -EOPNOTSUPP;
4081 
4082 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4083 	if (!ns_list)
4084 		return -ENOMEM;
4085 
4086 	for (;;) {
4087 		struct nvme_command cmd = {
4088 			.identify.opcode	= nvme_admin_identify,
4089 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4090 			.identify.nsid		= cpu_to_le32(prev),
4091 		};
4092 
4093 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4094 					    NVME_IDENTIFY_DATA_SIZE);
4095 		if (ret) {
4096 			dev_warn(ctrl->device,
4097 				"Identify NS List failed (status=0x%x)\n", ret);
4098 			goto free;
4099 		}
4100 
4101 		for (i = 0; i < nr_entries; i++) {
4102 			u32 nsid = le32_to_cpu(ns_list[i]);
4103 
4104 			if (!nsid)	/* end of the list? */
4105 				goto out;
4106 			nvme_validate_or_alloc_ns(ctrl, nsid);
4107 			while (++prev < nsid)
4108 				nvme_ns_remove_by_nsid(ctrl, prev);
4109 		}
4110 	}
4111  out:
4112 	nvme_remove_invalid_namespaces(ctrl, prev);
4113  free:
4114 	kfree(ns_list);
4115 	return ret;
4116 }
4117 
4118 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4119 {
4120 	struct nvme_id_ctrl *id;
4121 	u32 nn, i;
4122 
4123 	if (nvme_identify_ctrl(ctrl, &id))
4124 		return;
4125 	nn = le32_to_cpu(id->nn);
4126 	kfree(id);
4127 
4128 	for (i = 1; i <= nn; i++)
4129 		nvme_validate_or_alloc_ns(ctrl, i);
4130 
4131 	nvme_remove_invalid_namespaces(ctrl, nn);
4132 }
4133 
4134 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4135 {
4136 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4137 	__le32 *log;
4138 	int error;
4139 
4140 	log = kzalloc(log_size, GFP_KERNEL);
4141 	if (!log)
4142 		return;
4143 
4144 	/*
4145 	 * We need to read the log to clear the AEN, but we don't want to rely
4146 	 * on it for the changed namespace information as userspace could have
4147 	 * raced with us in reading the log page, which could cause us to miss
4148 	 * updates.
4149 	 */
4150 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4151 			NVME_CSI_NVM, log, log_size, 0);
4152 	if (error)
4153 		dev_warn(ctrl->device,
4154 			"reading changed ns log failed: %d\n", error);
4155 
4156 	kfree(log);
4157 }
4158 
4159 static void nvme_scan_work(struct work_struct *work)
4160 {
4161 	struct nvme_ctrl *ctrl =
4162 		container_of(work, struct nvme_ctrl, scan_work);
4163 
4164 	/* No tagset on a live ctrl means IO queues could not created */
4165 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4166 		return;
4167 
4168 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4169 		dev_info(ctrl->device, "rescanning namespaces.\n");
4170 		nvme_clear_changed_ns_log(ctrl);
4171 	}
4172 
4173 	mutex_lock(&ctrl->scan_lock);
4174 	if (nvme_scan_ns_list(ctrl) != 0)
4175 		nvme_scan_ns_sequential(ctrl);
4176 	mutex_unlock(&ctrl->scan_lock);
4177 
4178 	down_write(&ctrl->namespaces_rwsem);
4179 	list_sort(NULL, &ctrl->namespaces, ns_cmp);
4180 	up_write(&ctrl->namespaces_rwsem);
4181 }
4182 
4183 /*
4184  * This function iterates the namespace list unlocked to allow recovery from
4185  * controller failure. It is up to the caller to ensure the namespace list is
4186  * not modified by scan work while this function is executing.
4187  */
4188 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4189 {
4190 	struct nvme_ns *ns, *next;
4191 	LIST_HEAD(ns_list);
4192 
4193 	/*
4194 	 * make sure to requeue I/O to all namespaces as these
4195 	 * might result from the scan itself and must complete
4196 	 * for the scan_work to make progress
4197 	 */
4198 	nvme_mpath_clear_ctrl_paths(ctrl);
4199 
4200 	/* prevent racing with ns scanning */
4201 	flush_work(&ctrl->scan_work);
4202 
4203 	/*
4204 	 * The dead states indicates the controller was not gracefully
4205 	 * disconnected. In that case, we won't be able to flush any data while
4206 	 * removing the namespaces' disks; fail all the queues now to avoid
4207 	 * potentially having to clean up the failed sync later.
4208 	 */
4209 	if (ctrl->state == NVME_CTRL_DEAD)
4210 		nvme_kill_queues(ctrl);
4211 
4212 	/* this is a no-op when called from the controller reset handler */
4213 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4214 
4215 	down_write(&ctrl->namespaces_rwsem);
4216 	list_splice_init(&ctrl->namespaces, &ns_list);
4217 	up_write(&ctrl->namespaces_rwsem);
4218 
4219 	list_for_each_entry_safe(ns, next, &ns_list, list)
4220 		nvme_ns_remove(ns);
4221 }
4222 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4223 
4224 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4225 {
4226 	struct nvme_ctrl *ctrl =
4227 		container_of(dev, struct nvme_ctrl, ctrl_device);
4228 	struct nvmf_ctrl_options *opts = ctrl->opts;
4229 	int ret;
4230 
4231 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4232 	if (ret)
4233 		return ret;
4234 
4235 	if (opts) {
4236 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4237 		if (ret)
4238 			return ret;
4239 
4240 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4241 				opts->trsvcid ?: "none");
4242 		if (ret)
4243 			return ret;
4244 
4245 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4246 				opts->host_traddr ?: "none");
4247 	}
4248 	return ret;
4249 }
4250 
4251 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4252 {
4253 	char *envp[2] = { NULL, NULL };
4254 	u32 aen_result = ctrl->aen_result;
4255 
4256 	ctrl->aen_result = 0;
4257 	if (!aen_result)
4258 		return;
4259 
4260 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4261 	if (!envp[0])
4262 		return;
4263 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4264 	kfree(envp[0]);
4265 }
4266 
4267 static void nvme_async_event_work(struct work_struct *work)
4268 {
4269 	struct nvme_ctrl *ctrl =
4270 		container_of(work, struct nvme_ctrl, async_event_work);
4271 
4272 	nvme_aen_uevent(ctrl);
4273 	ctrl->ops->submit_async_event(ctrl);
4274 }
4275 
4276 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4277 {
4278 
4279 	u32 csts;
4280 
4281 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4282 		return false;
4283 
4284 	if (csts == ~0)
4285 		return false;
4286 
4287 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4288 }
4289 
4290 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4291 {
4292 	struct nvme_fw_slot_info_log *log;
4293 
4294 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4295 	if (!log)
4296 		return;
4297 
4298 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4299 			log, sizeof(*log), 0))
4300 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4301 	kfree(log);
4302 }
4303 
4304 static void nvme_fw_act_work(struct work_struct *work)
4305 {
4306 	struct nvme_ctrl *ctrl = container_of(work,
4307 				struct nvme_ctrl, fw_act_work);
4308 	unsigned long fw_act_timeout;
4309 
4310 	if (ctrl->mtfa)
4311 		fw_act_timeout = jiffies +
4312 				msecs_to_jiffies(ctrl->mtfa * 100);
4313 	else
4314 		fw_act_timeout = jiffies +
4315 				msecs_to_jiffies(admin_timeout * 1000);
4316 
4317 	nvme_stop_queues(ctrl);
4318 	while (nvme_ctrl_pp_status(ctrl)) {
4319 		if (time_after(jiffies, fw_act_timeout)) {
4320 			dev_warn(ctrl->device,
4321 				"Fw activation timeout, reset controller\n");
4322 			nvme_try_sched_reset(ctrl);
4323 			return;
4324 		}
4325 		msleep(100);
4326 	}
4327 
4328 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4329 		return;
4330 
4331 	nvme_start_queues(ctrl);
4332 	/* read FW slot information to clear the AER */
4333 	nvme_get_fw_slot_info(ctrl);
4334 }
4335 
4336 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4337 {
4338 	u32 aer_notice_type = (result & 0xff00) >> 8;
4339 
4340 	trace_nvme_async_event(ctrl, aer_notice_type);
4341 
4342 	switch (aer_notice_type) {
4343 	case NVME_AER_NOTICE_NS_CHANGED:
4344 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4345 		nvme_queue_scan(ctrl);
4346 		break;
4347 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4348 		/*
4349 		 * We are (ab)using the RESETTING state to prevent subsequent
4350 		 * recovery actions from interfering with the controller's
4351 		 * firmware activation.
4352 		 */
4353 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4354 			queue_work(nvme_wq, &ctrl->fw_act_work);
4355 		break;
4356 #ifdef CONFIG_NVME_MULTIPATH
4357 	case NVME_AER_NOTICE_ANA:
4358 		if (!ctrl->ana_log_buf)
4359 			break;
4360 		queue_work(nvme_wq, &ctrl->ana_work);
4361 		break;
4362 #endif
4363 	case NVME_AER_NOTICE_DISC_CHANGED:
4364 		ctrl->aen_result = result;
4365 		break;
4366 	default:
4367 		dev_warn(ctrl->device, "async event result %08x\n", result);
4368 	}
4369 }
4370 
4371 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4372 		volatile union nvme_result *res)
4373 {
4374 	u32 result = le32_to_cpu(res->u32);
4375 	u32 aer_type = result & 0x07;
4376 
4377 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4378 		return;
4379 
4380 	switch (aer_type) {
4381 	case NVME_AER_NOTICE:
4382 		nvme_handle_aen_notice(ctrl, result);
4383 		break;
4384 	case NVME_AER_ERROR:
4385 	case NVME_AER_SMART:
4386 	case NVME_AER_CSS:
4387 	case NVME_AER_VS:
4388 		trace_nvme_async_event(ctrl, aer_type);
4389 		ctrl->aen_result = result;
4390 		break;
4391 	default:
4392 		break;
4393 	}
4394 	queue_work(nvme_wq, &ctrl->async_event_work);
4395 }
4396 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4397 
4398 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4399 {
4400 	nvme_mpath_stop(ctrl);
4401 	nvme_stop_keep_alive(ctrl);
4402 	nvme_stop_failfast_work(ctrl);
4403 	flush_work(&ctrl->async_event_work);
4404 	cancel_work_sync(&ctrl->fw_act_work);
4405 }
4406 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4407 
4408 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4409 {
4410 	nvme_start_keep_alive(ctrl);
4411 
4412 	nvme_enable_aen(ctrl);
4413 
4414 	if (ctrl->queue_count > 1) {
4415 		nvme_queue_scan(ctrl);
4416 		nvme_start_queues(ctrl);
4417 	}
4418 }
4419 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4420 
4421 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4422 {
4423 	nvme_fault_inject_fini(&ctrl->fault_inject);
4424 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4425 	cdev_device_del(&ctrl->cdev, ctrl->device);
4426 	nvme_put_ctrl(ctrl);
4427 }
4428 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4429 
4430 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4431 {
4432 	struct nvme_effects_log	*cel;
4433 	unsigned long i;
4434 
4435 	xa_for_each (&ctrl->cels, i, cel) {
4436 		xa_erase(&ctrl->cels, i);
4437 		kfree(cel);
4438 	}
4439 
4440 	xa_destroy(&ctrl->cels);
4441 }
4442 
4443 static void nvme_free_ctrl(struct device *dev)
4444 {
4445 	struct nvme_ctrl *ctrl =
4446 		container_of(dev, struct nvme_ctrl, ctrl_device);
4447 	struct nvme_subsystem *subsys = ctrl->subsys;
4448 
4449 	if (!subsys || ctrl->instance != subsys->instance)
4450 		ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4451 
4452 	nvme_free_cels(ctrl);
4453 	nvme_mpath_uninit(ctrl);
4454 	__free_page(ctrl->discard_page);
4455 
4456 	if (subsys) {
4457 		mutex_lock(&nvme_subsystems_lock);
4458 		list_del(&ctrl->subsys_entry);
4459 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4460 		mutex_unlock(&nvme_subsystems_lock);
4461 	}
4462 
4463 	ctrl->ops->free_ctrl(ctrl);
4464 
4465 	if (subsys)
4466 		nvme_put_subsystem(subsys);
4467 }
4468 
4469 /*
4470  * Initialize a NVMe controller structures.  This needs to be called during
4471  * earliest initialization so that we have the initialized structured around
4472  * during probing.
4473  */
4474 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4475 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
4476 {
4477 	int ret;
4478 
4479 	ctrl->state = NVME_CTRL_NEW;
4480 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4481 	spin_lock_init(&ctrl->lock);
4482 	mutex_init(&ctrl->scan_lock);
4483 	INIT_LIST_HEAD(&ctrl->namespaces);
4484 	xa_init(&ctrl->cels);
4485 	init_rwsem(&ctrl->namespaces_rwsem);
4486 	ctrl->dev = dev;
4487 	ctrl->ops = ops;
4488 	ctrl->quirks = quirks;
4489 	ctrl->numa_node = NUMA_NO_NODE;
4490 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4491 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4492 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4493 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4494 	init_waitqueue_head(&ctrl->state_wq);
4495 
4496 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4497 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4498 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4499 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4500 
4501 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4502 			PAGE_SIZE);
4503 	ctrl->discard_page = alloc_page(GFP_KERNEL);
4504 	if (!ctrl->discard_page) {
4505 		ret = -ENOMEM;
4506 		goto out;
4507 	}
4508 
4509 	ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4510 	if (ret < 0)
4511 		goto out;
4512 	ctrl->instance = ret;
4513 
4514 	device_initialize(&ctrl->ctrl_device);
4515 	ctrl->device = &ctrl->ctrl_device;
4516 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4517 			ctrl->instance);
4518 	ctrl->device->class = nvme_class;
4519 	ctrl->device->parent = ctrl->dev;
4520 	ctrl->device->groups = nvme_dev_attr_groups;
4521 	ctrl->device->release = nvme_free_ctrl;
4522 	dev_set_drvdata(ctrl->device, ctrl);
4523 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4524 	if (ret)
4525 		goto out_release_instance;
4526 
4527 	nvme_get_ctrl(ctrl);
4528 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
4529 	ctrl->cdev.owner = ops->module;
4530 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4531 	if (ret)
4532 		goto out_free_name;
4533 
4534 	/*
4535 	 * Initialize latency tolerance controls.  The sysfs files won't
4536 	 * be visible to userspace unless the device actually supports APST.
4537 	 */
4538 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4539 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4540 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4541 
4542 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4543 
4544 	return 0;
4545 out_free_name:
4546 	nvme_put_ctrl(ctrl);
4547 	kfree_const(ctrl->device->kobj.name);
4548 out_release_instance:
4549 	ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4550 out:
4551 	if (ctrl->discard_page)
4552 		__free_page(ctrl->discard_page);
4553 	return ret;
4554 }
4555 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4556 
4557 /**
4558  * nvme_kill_queues(): Ends all namespace queues
4559  * @ctrl: the dead controller that needs to end
4560  *
4561  * Call this function when the driver determines it is unable to get the
4562  * controller in a state capable of servicing IO.
4563  */
4564 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4565 {
4566 	struct nvme_ns *ns;
4567 
4568 	down_read(&ctrl->namespaces_rwsem);
4569 
4570 	/* Forcibly unquiesce queues to avoid blocking dispatch */
4571 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4572 		blk_mq_unquiesce_queue(ctrl->admin_q);
4573 
4574 	list_for_each_entry(ns, &ctrl->namespaces, list)
4575 		nvme_set_queue_dying(ns);
4576 
4577 	up_read(&ctrl->namespaces_rwsem);
4578 }
4579 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4580 
4581 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4582 {
4583 	struct nvme_ns *ns;
4584 
4585 	down_read(&ctrl->namespaces_rwsem);
4586 	list_for_each_entry(ns, &ctrl->namespaces, list)
4587 		blk_mq_unfreeze_queue(ns->queue);
4588 	up_read(&ctrl->namespaces_rwsem);
4589 }
4590 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4591 
4592 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4593 {
4594 	struct nvme_ns *ns;
4595 
4596 	down_read(&ctrl->namespaces_rwsem);
4597 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4598 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4599 		if (timeout <= 0)
4600 			break;
4601 	}
4602 	up_read(&ctrl->namespaces_rwsem);
4603 	return timeout;
4604 }
4605 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4606 
4607 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4608 {
4609 	struct nvme_ns *ns;
4610 
4611 	down_read(&ctrl->namespaces_rwsem);
4612 	list_for_each_entry(ns, &ctrl->namespaces, list)
4613 		blk_mq_freeze_queue_wait(ns->queue);
4614 	up_read(&ctrl->namespaces_rwsem);
4615 }
4616 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4617 
4618 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4619 {
4620 	struct nvme_ns *ns;
4621 
4622 	down_read(&ctrl->namespaces_rwsem);
4623 	list_for_each_entry(ns, &ctrl->namespaces, list)
4624 		blk_freeze_queue_start(ns->queue);
4625 	up_read(&ctrl->namespaces_rwsem);
4626 }
4627 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4628 
4629 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4630 {
4631 	struct nvme_ns *ns;
4632 
4633 	down_read(&ctrl->namespaces_rwsem);
4634 	list_for_each_entry(ns, &ctrl->namespaces, list)
4635 		blk_mq_quiesce_queue(ns->queue);
4636 	up_read(&ctrl->namespaces_rwsem);
4637 }
4638 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4639 
4640 void nvme_start_queues(struct nvme_ctrl *ctrl)
4641 {
4642 	struct nvme_ns *ns;
4643 
4644 	down_read(&ctrl->namespaces_rwsem);
4645 	list_for_each_entry(ns, &ctrl->namespaces, list)
4646 		blk_mq_unquiesce_queue(ns->queue);
4647 	up_read(&ctrl->namespaces_rwsem);
4648 }
4649 EXPORT_SYMBOL_GPL(nvme_start_queues);
4650 
4651 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4652 {
4653 	struct nvme_ns *ns;
4654 
4655 	down_read(&ctrl->namespaces_rwsem);
4656 	list_for_each_entry(ns, &ctrl->namespaces, list)
4657 		blk_sync_queue(ns->queue);
4658 	up_read(&ctrl->namespaces_rwsem);
4659 }
4660 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4661 
4662 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4663 {
4664 	nvme_sync_io_queues(ctrl);
4665 	if (ctrl->admin_q)
4666 		blk_sync_queue(ctrl->admin_q);
4667 }
4668 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4669 
4670 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4671 {
4672 	if (file->f_op != &nvme_dev_fops)
4673 		return NULL;
4674 	return file->private_data;
4675 }
4676 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4677 
4678 /*
4679  * Check we didn't inadvertently grow the command structure sizes:
4680  */
4681 static inline void _nvme_check_size(void)
4682 {
4683 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4684 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4685 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4686 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4687 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4688 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4689 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4690 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4691 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4692 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4693 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4694 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4695 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4696 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4697 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4698 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4699 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4700 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4701 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4702 }
4703 
4704 
4705 static int __init nvme_core_init(void)
4706 {
4707 	int result = -ENOMEM;
4708 
4709 	_nvme_check_size();
4710 
4711 	nvme_wq = alloc_workqueue("nvme-wq",
4712 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4713 	if (!nvme_wq)
4714 		goto out;
4715 
4716 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4717 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4718 	if (!nvme_reset_wq)
4719 		goto destroy_wq;
4720 
4721 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4722 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4723 	if (!nvme_delete_wq)
4724 		goto destroy_reset_wq;
4725 
4726 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4727 			NVME_MINORS, "nvme");
4728 	if (result < 0)
4729 		goto destroy_delete_wq;
4730 
4731 	nvme_class = class_create(THIS_MODULE, "nvme");
4732 	if (IS_ERR(nvme_class)) {
4733 		result = PTR_ERR(nvme_class);
4734 		goto unregister_chrdev;
4735 	}
4736 	nvme_class->dev_uevent = nvme_class_uevent;
4737 
4738 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4739 	if (IS_ERR(nvme_subsys_class)) {
4740 		result = PTR_ERR(nvme_subsys_class);
4741 		goto destroy_class;
4742 	}
4743 	return 0;
4744 
4745 destroy_class:
4746 	class_destroy(nvme_class);
4747 unregister_chrdev:
4748 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4749 destroy_delete_wq:
4750 	destroy_workqueue(nvme_delete_wq);
4751 destroy_reset_wq:
4752 	destroy_workqueue(nvme_reset_wq);
4753 destroy_wq:
4754 	destroy_workqueue(nvme_wq);
4755 out:
4756 	return result;
4757 }
4758 
4759 static void __exit nvme_core_exit(void)
4760 {
4761 	class_destroy(nvme_subsys_class);
4762 	class_destroy(nvme_class);
4763 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4764 	destroy_workqueue(nvme_delete_wq);
4765 	destroy_workqueue(nvme_reset_wq);
4766 	destroy_workqueue(nvme_wq);
4767 	ida_destroy(&nvme_instance_ida);
4768 }
4769 
4770 MODULE_LICENSE("GPL");
4771 MODULE_VERSION("1.0");
4772 module_init(nvme_core_init);
4773 module_exit(nvme_core_exit);
4774