xref: /openbmc/linux/drivers/nvme/host/core.c (revision 7cffcade)
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/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.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 #include <linux/nvme-auth.h>
28 
29 #define CREATE_TRACE_POINTS
30 #include "trace.h"
31 
32 #define NVME_MINORS		(1U << MINORBITS)
33 
34 struct nvme_ns_info {
35 	struct nvme_ns_ids ids;
36 	u32 nsid;
37 	__le32 anagrpid;
38 	bool is_shared;
39 	bool is_readonly;
40 	bool is_ready;
41 };
42 
43 unsigned int admin_timeout = 60;
44 module_param(admin_timeout, uint, 0644);
45 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
46 EXPORT_SYMBOL_GPL(admin_timeout);
47 
48 unsigned int nvme_io_timeout = 30;
49 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
50 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
51 EXPORT_SYMBOL_GPL(nvme_io_timeout);
52 
53 static unsigned char shutdown_timeout = 5;
54 module_param(shutdown_timeout, byte, 0644);
55 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
56 
57 static u8 nvme_max_retries = 5;
58 module_param_named(max_retries, nvme_max_retries, byte, 0644);
59 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
60 
61 static unsigned long default_ps_max_latency_us = 100000;
62 module_param(default_ps_max_latency_us, ulong, 0644);
63 MODULE_PARM_DESC(default_ps_max_latency_us,
64 		 "max power saving latency for new devices; use PM QOS to change per device");
65 
66 static bool force_apst;
67 module_param(force_apst, bool, 0644);
68 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
69 
70 static unsigned long apst_primary_timeout_ms = 100;
71 module_param(apst_primary_timeout_ms, ulong, 0644);
72 MODULE_PARM_DESC(apst_primary_timeout_ms,
73 	"primary APST timeout in ms");
74 
75 static unsigned long apst_secondary_timeout_ms = 2000;
76 module_param(apst_secondary_timeout_ms, ulong, 0644);
77 MODULE_PARM_DESC(apst_secondary_timeout_ms,
78 	"secondary APST timeout in ms");
79 
80 static unsigned long apst_primary_latency_tol_us = 15000;
81 module_param(apst_primary_latency_tol_us, ulong, 0644);
82 MODULE_PARM_DESC(apst_primary_latency_tol_us,
83 	"primary APST latency tolerance in us");
84 
85 static unsigned long apst_secondary_latency_tol_us = 100000;
86 module_param(apst_secondary_latency_tol_us, ulong, 0644);
87 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
88 	"secondary APST latency tolerance in us");
89 
90 /*
91  * nvme_wq - hosts nvme related works that are not reset or delete
92  * nvme_reset_wq - hosts nvme reset works
93  * nvme_delete_wq - hosts nvme delete works
94  *
95  * nvme_wq will host works such as scan, aen handling, fw activation,
96  * keep-alive, periodic reconnects etc. nvme_reset_wq
97  * runs reset works which also flush works hosted on nvme_wq for
98  * serialization purposes. nvme_delete_wq host controller deletion
99  * works which flush reset works for serialization.
100  */
101 struct workqueue_struct *nvme_wq;
102 EXPORT_SYMBOL_GPL(nvme_wq);
103 
104 struct workqueue_struct *nvme_reset_wq;
105 EXPORT_SYMBOL_GPL(nvme_reset_wq);
106 
107 struct workqueue_struct *nvme_delete_wq;
108 EXPORT_SYMBOL_GPL(nvme_delete_wq);
109 
110 static LIST_HEAD(nvme_subsystems);
111 static DEFINE_MUTEX(nvme_subsystems_lock);
112 
113 static DEFINE_IDA(nvme_instance_ida);
114 static dev_t nvme_ctrl_base_chr_devt;
115 static struct class *nvme_class;
116 static struct class *nvme_subsys_class;
117 
118 static DEFINE_IDA(nvme_ns_chr_minor_ida);
119 static dev_t nvme_ns_chr_devt;
120 static struct class *nvme_ns_chr_class;
121 
122 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
123 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
124 					   unsigned nsid);
125 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
126 				   struct nvme_command *cmd);
127 
128 void nvme_queue_scan(struct nvme_ctrl *ctrl)
129 {
130 	/*
131 	 * Only new queue scan work when admin and IO queues are both alive
132 	 */
133 	if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
134 		queue_work(nvme_wq, &ctrl->scan_work);
135 }
136 
137 /*
138  * Use this function to proceed with scheduling reset_work for a controller
139  * that had previously been set to the resetting state. This is intended for
140  * code paths that can't be interrupted by other reset attempts. A hot removal
141  * may prevent this from succeeding.
142  */
143 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
144 {
145 	if (ctrl->state != NVME_CTRL_RESETTING)
146 		return -EBUSY;
147 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
148 		return -EBUSY;
149 	return 0;
150 }
151 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
152 
153 static void nvme_failfast_work(struct work_struct *work)
154 {
155 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
156 			struct nvme_ctrl, failfast_work);
157 
158 	if (ctrl->state != NVME_CTRL_CONNECTING)
159 		return;
160 
161 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
162 	dev_info(ctrl->device, "failfast expired\n");
163 	nvme_kick_requeue_lists(ctrl);
164 }
165 
166 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
167 {
168 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
169 		return;
170 
171 	schedule_delayed_work(&ctrl->failfast_work,
172 			      ctrl->opts->fast_io_fail_tmo * HZ);
173 }
174 
175 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
176 {
177 	if (!ctrl->opts)
178 		return;
179 
180 	cancel_delayed_work_sync(&ctrl->failfast_work);
181 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
182 }
183 
184 
185 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
186 {
187 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
188 		return -EBUSY;
189 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
190 		return -EBUSY;
191 	return 0;
192 }
193 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
194 
195 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
196 {
197 	int ret;
198 
199 	ret = nvme_reset_ctrl(ctrl);
200 	if (!ret) {
201 		flush_work(&ctrl->reset_work);
202 		if (ctrl->state != NVME_CTRL_LIVE)
203 			ret = -ENETRESET;
204 	}
205 
206 	return ret;
207 }
208 
209 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
210 {
211 	dev_info(ctrl->device,
212 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
213 
214 	flush_work(&ctrl->reset_work);
215 	nvme_stop_ctrl(ctrl);
216 	nvme_remove_namespaces(ctrl);
217 	ctrl->ops->delete_ctrl(ctrl);
218 	nvme_uninit_ctrl(ctrl);
219 }
220 
221 static void nvme_delete_ctrl_work(struct work_struct *work)
222 {
223 	struct nvme_ctrl *ctrl =
224 		container_of(work, struct nvme_ctrl, delete_work);
225 
226 	nvme_do_delete_ctrl(ctrl);
227 }
228 
229 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
230 {
231 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
232 		return -EBUSY;
233 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
234 		return -EBUSY;
235 	return 0;
236 }
237 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
238 
239 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
240 {
241 	/*
242 	 * Keep a reference until nvme_do_delete_ctrl() complete,
243 	 * since ->delete_ctrl can free the controller.
244 	 */
245 	nvme_get_ctrl(ctrl);
246 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
247 		nvme_do_delete_ctrl(ctrl);
248 	nvme_put_ctrl(ctrl);
249 }
250 
251 static blk_status_t nvme_error_status(u16 status)
252 {
253 	switch (status & 0x7ff) {
254 	case NVME_SC_SUCCESS:
255 		return BLK_STS_OK;
256 	case NVME_SC_CAP_EXCEEDED:
257 		return BLK_STS_NOSPC;
258 	case NVME_SC_LBA_RANGE:
259 	case NVME_SC_CMD_INTERRUPTED:
260 	case NVME_SC_NS_NOT_READY:
261 		return BLK_STS_TARGET;
262 	case NVME_SC_BAD_ATTRIBUTES:
263 	case NVME_SC_ONCS_NOT_SUPPORTED:
264 	case NVME_SC_INVALID_OPCODE:
265 	case NVME_SC_INVALID_FIELD:
266 	case NVME_SC_INVALID_NS:
267 		return BLK_STS_NOTSUPP;
268 	case NVME_SC_WRITE_FAULT:
269 	case NVME_SC_READ_ERROR:
270 	case NVME_SC_UNWRITTEN_BLOCK:
271 	case NVME_SC_ACCESS_DENIED:
272 	case NVME_SC_READ_ONLY:
273 	case NVME_SC_COMPARE_FAILED:
274 		return BLK_STS_MEDIUM;
275 	case NVME_SC_GUARD_CHECK:
276 	case NVME_SC_APPTAG_CHECK:
277 	case NVME_SC_REFTAG_CHECK:
278 	case NVME_SC_INVALID_PI:
279 		return BLK_STS_PROTECTION;
280 	case NVME_SC_RESERVATION_CONFLICT:
281 		return BLK_STS_NEXUS;
282 	case NVME_SC_HOST_PATH_ERROR:
283 		return BLK_STS_TRANSPORT;
284 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
285 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
286 	case NVME_SC_ZONE_TOO_MANY_OPEN:
287 		return BLK_STS_ZONE_OPEN_RESOURCE;
288 	default:
289 		return BLK_STS_IOERR;
290 	}
291 }
292 
293 static void nvme_retry_req(struct request *req)
294 {
295 	unsigned long delay = 0;
296 	u16 crd;
297 
298 	/* The mask and shift result must be <= 3 */
299 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
300 	if (crd)
301 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
302 
303 	nvme_req(req)->retries++;
304 	blk_mq_requeue_request(req, false);
305 	blk_mq_delay_kick_requeue_list(req->q, delay);
306 }
307 
308 static void nvme_log_error(struct request *req)
309 {
310 	struct nvme_ns *ns = req->q->queuedata;
311 	struct nvme_request *nr = nvme_req(req);
312 
313 	if (ns) {
314 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
315 		       ns->disk ? ns->disk->disk_name : "?",
316 		       nvme_get_opcode_str(nr->cmd->common.opcode),
317 		       nr->cmd->common.opcode,
318 		       (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
319 		       (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
320 		       nvme_get_error_status_str(nr->status),
321 		       nr->status >> 8 & 7,	/* Status Code Type */
322 		       nr->status & 0xff,	/* Status Code */
323 		       nr->status & NVME_SC_MORE ? "MORE " : "",
324 		       nr->status & NVME_SC_DNR  ? "DNR "  : "");
325 		return;
326 	}
327 
328 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
329 			   dev_name(nr->ctrl->device),
330 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
331 			   nr->cmd->common.opcode,
332 			   nvme_get_error_status_str(nr->status),
333 			   nr->status >> 8 & 7,	/* Status Code Type */
334 			   nr->status & 0xff,	/* Status Code */
335 			   nr->status & NVME_SC_MORE ? "MORE " : "",
336 			   nr->status & NVME_SC_DNR  ? "DNR "  : "");
337 }
338 
339 enum nvme_disposition {
340 	COMPLETE,
341 	RETRY,
342 	FAILOVER,
343 	AUTHENTICATE,
344 };
345 
346 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
347 {
348 	if (likely(nvme_req(req)->status == 0))
349 		return COMPLETE;
350 
351 	if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
352 		return AUTHENTICATE;
353 
354 	if (blk_noretry_request(req) ||
355 	    (nvme_req(req)->status & NVME_SC_DNR) ||
356 	    nvme_req(req)->retries >= nvme_max_retries)
357 		return COMPLETE;
358 
359 	if (req->cmd_flags & REQ_NVME_MPATH) {
360 		if (nvme_is_path_error(nvme_req(req)->status) ||
361 		    blk_queue_dying(req->q))
362 			return FAILOVER;
363 	} else {
364 		if (blk_queue_dying(req->q))
365 			return COMPLETE;
366 	}
367 
368 	return RETRY;
369 }
370 
371 static inline void nvme_end_req_zoned(struct request *req)
372 {
373 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
374 	    req_op(req) == REQ_OP_ZONE_APPEND)
375 		req->__sector = nvme_lba_to_sect(req->q->queuedata,
376 			le64_to_cpu(nvme_req(req)->result.u64));
377 }
378 
379 static inline void nvme_end_req(struct request *req)
380 {
381 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
382 
383 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
384 		nvme_log_error(req);
385 	nvme_end_req_zoned(req);
386 	nvme_trace_bio_complete(req);
387 	blk_mq_end_request(req, status);
388 }
389 
390 void nvme_complete_rq(struct request *req)
391 {
392 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
393 
394 	trace_nvme_complete_rq(req);
395 	nvme_cleanup_cmd(req);
396 
397 	if (ctrl->kas)
398 		ctrl->comp_seen = true;
399 
400 	switch (nvme_decide_disposition(req)) {
401 	case COMPLETE:
402 		nvme_end_req(req);
403 		return;
404 	case RETRY:
405 		nvme_retry_req(req);
406 		return;
407 	case FAILOVER:
408 		nvme_failover_req(req);
409 		return;
410 	case AUTHENTICATE:
411 #ifdef CONFIG_NVME_AUTH
412 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
413 		nvme_retry_req(req);
414 #else
415 		nvme_end_req(req);
416 #endif
417 		return;
418 	}
419 }
420 EXPORT_SYMBOL_GPL(nvme_complete_rq);
421 
422 void nvme_complete_batch_req(struct request *req)
423 {
424 	trace_nvme_complete_rq(req);
425 	nvme_cleanup_cmd(req);
426 	nvme_end_req_zoned(req);
427 }
428 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
429 
430 /*
431  * Called to unwind from ->queue_rq on a failed command submission so that the
432  * multipathing code gets called to potentially failover to another path.
433  * The caller needs to unwind all transport specific resource allocations and
434  * must return propagate the return value.
435  */
436 blk_status_t nvme_host_path_error(struct request *req)
437 {
438 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
439 	blk_mq_set_request_complete(req);
440 	nvme_complete_rq(req);
441 	return BLK_STS_OK;
442 }
443 EXPORT_SYMBOL_GPL(nvme_host_path_error);
444 
445 bool nvme_cancel_request(struct request *req, void *data)
446 {
447 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
448 				"Cancelling I/O %d", req->tag);
449 
450 	/* don't abort one completed request */
451 	if (blk_mq_request_completed(req))
452 		return true;
453 
454 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
455 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
456 	blk_mq_complete_request(req);
457 	return true;
458 }
459 EXPORT_SYMBOL_GPL(nvme_cancel_request);
460 
461 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
462 {
463 	if (ctrl->tagset) {
464 		blk_mq_tagset_busy_iter(ctrl->tagset,
465 				nvme_cancel_request, ctrl);
466 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
467 	}
468 }
469 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
470 
471 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
472 {
473 	if (ctrl->admin_tagset) {
474 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
475 				nvme_cancel_request, ctrl);
476 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
477 	}
478 }
479 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
480 
481 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
482 		enum nvme_ctrl_state new_state)
483 {
484 	enum nvme_ctrl_state old_state;
485 	unsigned long flags;
486 	bool changed = false;
487 
488 	spin_lock_irqsave(&ctrl->lock, flags);
489 
490 	old_state = ctrl->state;
491 	switch (new_state) {
492 	case NVME_CTRL_LIVE:
493 		switch (old_state) {
494 		case NVME_CTRL_NEW:
495 		case NVME_CTRL_RESETTING:
496 		case NVME_CTRL_CONNECTING:
497 			changed = true;
498 			fallthrough;
499 		default:
500 			break;
501 		}
502 		break;
503 	case NVME_CTRL_RESETTING:
504 		switch (old_state) {
505 		case NVME_CTRL_NEW:
506 		case NVME_CTRL_LIVE:
507 			changed = true;
508 			fallthrough;
509 		default:
510 			break;
511 		}
512 		break;
513 	case NVME_CTRL_CONNECTING:
514 		switch (old_state) {
515 		case NVME_CTRL_NEW:
516 		case NVME_CTRL_RESETTING:
517 			changed = true;
518 			fallthrough;
519 		default:
520 			break;
521 		}
522 		break;
523 	case NVME_CTRL_DELETING:
524 		switch (old_state) {
525 		case NVME_CTRL_LIVE:
526 		case NVME_CTRL_RESETTING:
527 		case NVME_CTRL_CONNECTING:
528 			changed = true;
529 			fallthrough;
530 		default:
531 			break;
532 		}
533 		break;
534 	case NVME_CTRL_DELETING_NOIO:
535 		switch (old_state) {
536 		case NVME_CTRL_DELETING:
537 		case NVME_CTRL_DEAD:
538 			changed = true;
539 			fallthrough;
540 		default:
541 			break;
542 		}
543 		break;
544 	case NVME_CTRL_DEAD:
545 		switch (old_state) {
546 		case NVME_CTRL_DELETING:
547 			changed = true;
548 			fallthrough;
549 		default:
550 			break;
551 		}
552 		break;
553 	default:
554 		break;
555 	}
556 
557 	if (changed) {
558 		ctrl->state = new_state;
559 		wake_up_all(&ctrl->state_wq);
560 	}
561 
562 	spin_unlock_irqrestore(&ctrl->lock, flags);
563 	if (!changed)
564 		return false;
565 
566 	if (ctrl->state == NVME_CTRL_LIVE) {
567 		if (old_state == NVME_CTRL_CONNECTING)
568 			nvme_stop_failfast_work(ctrl);
569 		nvme_kick_requeue_lists(ctrl);
570 	} else if (ctrl->state == NVME_CTRL_CONNECTING &&
571 		old_state == NVME_CTRL_RESETTING) {
572 		nvme_start_failfast_work(ctrl);
573 	}
574 	return changed;
575 }
576 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
577 
578 /*
579  * Returns true for sink states that can't ever transition back to live.
580  */
581 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
582 {
583 	switch (ctrl->state) {
584 	case NVME_CTRL_NEW:
585 	case NVME_CTRL_LIVE:
586 	case NVME_CTRL_RESETTING:
587 	case NVME_CTRL_CONNECTING:
588 		return false;
589 	case NVME_CTRL_DELETING:
590 	case NVME_CTRL_DELETING_NOIO:
591 	case NVME_CTRL_DEAD:
592 		return true;
593 	default:
594 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
595 		return true;
596 	}
597 }
598 
599 /*
600  * Waits for the controller state to be resetting, or returns false if it is
601  * not possible to ever transition to that state.
602  */
603 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
604 {
605 	wait_event(ctrl->state_wq,
606 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
607 		   nvme_state_terminal(ctrl));
608 	return ctrl->state == NVME_CTRL_RESETTING;
609 }
610 EXPORT_SYMBOL_GPL(nvme_wait_reset);
611 
612 static void nvme_free_ns_head(struct kref *ref)
613 {
614 	struct nvme_ns_head *head =
615 		container_of(ref, struct nvme_ns_head, ref);
616 
617 	nvme_mpath_remove_disk(head);
618 	ida_free(&head->subsys->ns_ida, head->instance);
619 	cleanup_srcu_struct(&head->srcu);
620 	nvme_put_subsystem(head->subsys);
621 	kfree(head);
622 }
623 
624 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
625 {
626 	return kref_get_unless_zero(&head->ref);
627 }
628 
629 void nvme_put_ns_head(struct nvme_ns_head *head)
630 {
631 	kref_put(&head->ref, nvme_free_ns_head);
632 }
633 
634 static void nvme_free_ns(struct kref *kref)
635 {
636 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
637 
638 	put_disk(ns->disk);
639 	nvme_put_ns_head(ns->head);
640 	nvme_put_ctrl(ns->ctrl);
641 	kfree(ns);
642 }
643 
644 static inline bool nvme_get_ns(struct nvme_ns *ns)
645 {
646 	return kref_get_unless_zero(&ns->kref);
647 }
648 
649 void nvme_put_ns(struct nvme_ns *ns)
650 {
651 	kref_put(&ns->kref, nvme_free_ns);
652 }
653 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
654 
655 static inline void nvme_clear_nvme_request(struct request *req)
656 {
657 	nvme_req(req)->status = 0;
658 	nvme_req(req)->retries = 0;
659 	nvme_req(req)->flags = 0;
660 	req->rq_flags |= RQF_DONTPREP;
661 }
662 
663 /* initialize a passthrough request */
664 void nvme_init_request(struct request *req, struct nvme_command *cmd)
665 {
666 	if (req->q->queuedata)
667 		req->timeout = NVME_IO_TIMEOUT;
668 	else /* no queuedata implies admin queue */
669 		req->timeout = NVME_ADMIN_TIMEOUT;
670 
671 	/* passthru commands should let the driver set the SGL flags */
672 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
673 
674 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
675 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
676 		req->cmd_flags |= REQ_POLLED;
677 	nvme_clear_nvme_request(req);
678 	req->rq_flags |= RQF_QUIET;
679 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
680 }
681 EXPORT_SYMBOL_GPL(nvme_init_request);
682 
683 /*
684  * For something we're not in a state to send to the device the default action
685  * is to busy it and retry it after the controller state is recovered.  However,
686  * if the controller is deleting or if anything is marked for failfast or
687  * nvme multipath it is immediately failed.
688  *
689  * Note: commands used to initialize the controller will be marked for failfast.
690  * Note: nvme cli/ioctl commands are marked for failfast.
691  */
692 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
693 		struct request *rq)
694 {
695 	if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
696 	    ctrl->state != NVME_CTRL_DELETING &&
697 	    ctrl->state != NVME_CTRL_DEAD &&
698 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
699 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
700 		return BLK_STS_RESOURCE;
701 	return nvme_host_path_error(rq);
702 }
703 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
704 
705 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
706 		bool queue_live)
707 {
708 	struct nvme_request *req = nvme_req(rq);
709 
710 	/*
711 	 * currently we have a problem sending passthru commands
712 	 * on the admin_q if the controller is not LIVE because we can't
713 	 * make sure that they are going out after the admin connect,
714 	 * controller enable and/or other commands in the initialization
715 	 * sequence. until the controller will be LIVE, fail with
716 	 * BLK_STS_RESOURCE so that they will be rescheduled.
717 	 */
718 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
719 		return false;
720 
721 	if (ctrl->ops->flags & NVME_F_FABRICS) {
722 		/*
723 		 * Only allow commands on a live queue, except for the connect
724 		 * command, which is require to set the queue live in the
725 		 * appropinquate states.
726 		 */
727 		switch (ctrl->state) {
728 		case NVME_CTRL_CONNECTING:
729 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
730 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
731 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
732 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
733 				return true;
734 			break;
735 		default:
736 			break;
737 		case NVME_CTRL_DEAD:
738 			return false;
739 		}
740 	}
741 
742 	return queue_live;
743 }
744 EXPORT_SYMBOL_GPL(__nvme_check_ready);
745 
746 static inline void nvme_setup_flush(struct nvme_ns *ns,
747 		struct nvme_command *cmnd)
748 {
749 	memset(cmnd, 0, sizeof(*cmnd));
750 	cmnd->common.opcode = nvme_cmd_flush;
751 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
752 }
753 
754 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
755 		struct nvme_command *cmnd)
756 {
757 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
758 	struct nvme_dsm_range *range;
759 	struct bio *bio;
760 
761 	/*
762 	 * Some devices do not consider the DSM 'Number of Ranges' field when
763 	 * determining how much data to DMA. Always allocate memory for maximum
764 	 * number of segments to prevent device reading beyond end of buffer.
765 	 */
766 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
767 
768 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
769 	if (!range) {
770 		/*
771 		 * If we fail allocation our range, fallback to the controller
772 		 * discard page. If that's also busy, it's safe to return
773 		 * busy, as we know we can make progress once that's freed.
774 		 */
775 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
776 			return BLK_STS_RESOURCE;
777 
778 		range = page_address(ns->ctrl->discard_page);
779 	}
780 
781 	__rq_for_each_bio(bio, req) {
782 		u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
783 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
784 
785 		if (n < segments) {
786 			range[n].cattr = cpu_to_le32(0);
787 			range[n].nlb = cpu_to_le32(nlb);
788 			range[n].slba = cpu_to_le64(slba);
789 		}
790 		n++;
791 	}
792 
793 	if (WARN_ON_ONCE(n != segments)) {
794 		if (virt_to_page(range) == ns->ctrl->discard_page)
795 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
796 		else
797 			kfree(range);
798 		return BLK_STS_IOERR;
799 	}
800 
801 	memset(cmnd, 0, sizeof(*cmnd));
802 	cmnd->dsm.opcode = nvme_cmd_dsm;
803 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
804 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
805 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
806 
807 	req->special_vec.bv_page = virt_to_page(range);
808 	req->special_vec.bv_offset = offset_in_page(range);
809 	req->special_vec.bv_len = alloc_size;
810 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
811 
812 	return BLK_STS_OK;
813 }
814 
815 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
816 			      struct request *req)
817 {
818 	u32 upper, lower;
819 	u64 ref48;
820 
821 	/* both rw and write zeroes share the same reftag format */
822 	switch (ns->guard_type) {
823 	case NVME_NVM_NS_16B_GUARD:
824 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
825 		break;
826 	case NVME_NVM_NS_64B_GUARD:
827 		ref48 = ext_pi_ref_tag(req);
828 		lower = lower_32_bits(ref48);
829 		upper = upper_32_bits(ref48);
830 
831 		cmnd->rw.reftag = cpu_to_le32(lower);
832 		cmnd->rw.cdw3 = cpu_to_le32(upper);
833 		break;
834 	default:
835 		break;
836 	}
837 }
838 
839 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
840 		struct request *req, struct nvme_command *cmnd)
841 {
842 	memset(cmnd, 0, sizeof(*cmnd));
843 
844 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
845 		return nvme_setup_discard(ns, req, cmnd);
846 
847 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
848 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
849 	cmnd->write_zeroes.slba =
850 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
851 	cmnd->write_zeroes.length =
852 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
853 
854 	if (nvme_ns_has_pi(ns)) {
855 		cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
856 
857 		switch (ns->pi_type) {
858 		case NVME_NS_DPS_PI_TYPE1:
859 		case NVME_NS_DPS_PI_TYPE2:
860 			nvme_set_ref_tag(ns, cmnd, req);
861 			break;
862 		}
863 	}
864 
865 	return BLK_STS_OK;
866 }
867 
868 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
869 		struct request *req, struct nvme_command *cmnd,
870 		enum nvme_opcode op)
871 {
872 	u16 control = 0;
873 	u32 dsmgmt = 0;
874 
875 	if (req->cmd_flags & REQ_FUA)
876 		control |= NVME_RW_FUA;
877 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
878 		control |= NVME_RW_LR;
879 
880 	if (req->cmd_flags & REQ_RAHEAD)
881 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
882 
883 	cmnd->rw.opcode = op;
884 	cmnd->rw.flags = 0;
885 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
886 	cmnd->rw.cdw2 = 0;
887 	cmnd->rw.cdw3 = 0;
888 	cmnd->rw.metadata = 0;
889 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
890 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
891 	cmnd->rw.reftag = 0;
892 	cmnd->rw.apptag = 0;
893 	cmnd->rw.appmask = 0;
894 
895 	if (ns->ms) {
896 		/*
897 		 * If formated with metadata, the block layer always provides a
898 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
899 		 * we enable the PRACT bit for protection information or set the
900 		 * namespace capacity to zero to prevent any I/O.
901 		 */
902 		if (!blk_integrity_rq(req)) {
903 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
904 				return BLK_STS_NOTSUPP;
905 			control |= NVME_RW_PRINFO_PRACT;
906 		}
907 
908 		switch (ns->pi_type) {
909 		case NVME_NS_DPS_PI_TYPE3:
910 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
911 			break;
912 		case NVME_NS_DPS_PI_TYPE1:
913 		case NVME_NS_DPS_PI_TYPE2:
914 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
915 					NVME_RW_PRINFO_PRCHK_REF;
916 			if (op == nvme_cmd_zone_append)
917 				control |= NVME_RW_APPEND_PIREMAP;
918 			nvme_set_ref_tag(ns, cmnd, req);
919 			break;
920 		}
921 	}
922 
923 	cmnd->rw.control = cpu_to_le16(control);
924 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
925 	return 0;
926 }
927 
928 void nvme_cleanup_cmd(struct request *req)
929 {
930 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
931 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
932 
933 		if (req->special_vec.bv_page == ctrl->discard_page)
934 			clear_bit_unlock(0, &ctrl->discard_page_busy);
935 		else
936 			kfree(bvec_virt(&req->special_vec));
937 	}
938 }
939 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
940 
941 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
942 {
943 	struct nvme_command *cmd = nvme_req(req)->cmd;
944 	blk_status_t ret = BLK_STS_OK;
945 
946 	if (!(req->rq_flags & RQF_DONTPREP))
947 		nvme_clear_nvme_request(req);
948 
949 	switch (req_op(req)) {
950 	case REQ_OP_DRV_IN:
951 	case REQ_OP_DRV_OUT:
952 		/* these are setup prior to execution in nvme_init_request() */
953 		break;
954 	case REQ_OP_FLUSH:
955 		nvme_setup_flush(ns, cmd);
956 		break;
957 	case REQ_OP_ZONE_RESET_ALL:
958 	case REQ_OP_ZONE_RESET:
959 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
960 		break;
961 	case REQ_OP_ZONE_OPEN:
962 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
963 		break;
964 	case REQ_OP_ZONE_CLOSE:
965 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
966 		break;
967 	case REQ_OP_ZONE_FINISH:
968 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
969 		break;
970 	case REQ_OP_WRITE_ZEROES:
971 		ret = nvme_setup_write_zeroes(ns, req, cmd);
972 		break;
973 	case REQ_OP_DISCARD:
974 		ret = nvme_setup_discard(ns, req, cmd);
975 		break;
976 	case REQ_OP_READ:
977 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
978 		break;
979 	case REQ_OP_WRITE:
980 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
981 		break;
982 	case REQ_OP_ZONE_APPEND:
983 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
984 		break;
985 	default:
986 		WARN_ON_ONCE(1);
987 		return BLK_STS_IOERR;
988 	}
989 
990 	cmd->common.command_id = nvme_cid(req);
991 	trace_nvme_setup_cmd(req, cmd);
992 	return ret;
993 }
994 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
995 
996 /*
997  * Return values:
998  * 0:  success
999  * >0: nvme controller's cqe status response
1000  * <0: kernel error in lieu of controller response
1001  */
1002 static int nvme_execute_rq(struct request *rq, bool at_head)
1003 {
1004 	blk_status_t status;
1005 
1006 	status = blk_execute_rq(rq, at_head);
1007 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1008 		return -EINTR;
1009 	if (nvme_req(rq)->status)
1010 		return nvme_req(rq)->status;
1011 	return blk_status_to_errno(status);
1012 }
1013 
1014 /*
1015  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1016  * if the result is positive, it's an NVM Express status code
1017  */
1018 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1019 		union nvme_result *result, void *buffer, unsigned bufflen,
1020 		int qid, int at_head, blk_mq_req_flags_t flags)
1021 {
1022 	struct request *req;
1023 	int ret;
1024 
1025 	if (qid == NVME_QID_ANY)
1026 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1027 	else
1028 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1029 						qid - 1);
1030 
1031 	if (IS_ERR(req))
1032 		return PTR_ERR(req);
1033 	nvme_init_request(req, cmd);
1034 
1035 	if (buffer && bufflen) {
1036 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1037 		if (ret)
1038 			goto out;
1039 	}
1040 
1041 	ret = nvme_execute_rq(req, at_head);
1042 	if (result && ret >= 0)
1043 		*result = nvme_req(req)->result;
1044  out:
1045 	blk_mq_free_request(req);
1046 	return ret;
1047 }
1048 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1049 
1050 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1051 		void *buffer, unsigned bufflen)
1052 {
1053 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1054 			NVME_QID_ANY, 0, 0);
1055 }
1056 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1057 
1058 static u32 nvme_known_admin_effects(u8 opcode)
1059 {
1060 	switch (opcode) {
1061 	case nvme_admin_format_nvm:
1062 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1063 			NVME_CMD_EFFECTS_CSE_MASK;
1064 	case nvme_admin_sanitize_nvm:
1065 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1066 	default:
1067 		break;
1068 	}
1069 	return 0;
1070 }
1071 
1072 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1073 {
1074 	u32 effects = 0;
1075 
1076 	if (ns) {
1077 		if (ns->head->effects)
1078 			effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1079 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1080 			dev_warn_once(ctrl->device,
1081 				"IO command:%02x has unhandled effects:%08x\n",
1082 				opcode, effects);
1083 		return 0;
1084 	}
1085 
1086 	if (ctrl->effects)
1087 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1088 	effects |= nvme_known_admin_effects(opcode);
1089 
1090 	return effects;
1091 }
1092 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1093 
1094 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1095 			       u8 opcode)
1096 {
1097 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1098 
1099 	/*
1100 	 * For simplicity, IO to all namespaces is quiesced even if the command
1101 	 * effects say only one namespace is affected.
1102 	 */
1103 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1104 		mutex_lock(&ctrl->scan_lock);
1105 		mutex_lock(&ctrl->subsys->lock);
1106 		nvme_mpath_start_freeze(ctrl->subsys);
1107 		nvme_mpath_wait_freeze(ctrl->subsys);
1108 		nvme_start_freeze(ctrl);
1109 		nvme_wait_freeze(ctrl);
1110 	}
1111 	return effects;
1112 }
1113 
1114 void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1115 		       struct nvme_command *cmd, int status)
1116 {
1117 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1118 		nvme_unfreeze(ctrl);
1119 		nvme_mpath_unfreeze(ctrl->subsys);
1120 		mutex_unlock(&ctrl->subsys->lock);
1121 		nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1122 		mutex_unlock(&ctrl->scan_lock);
1123 	}
1124 	if (effects & NVME_CMD_EFFECTS_CCC)
1125 		nvme_init_ctrl_finish(ctrl);
1126 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1127 		nvme_queue_scan(ctrl);
1128 		flush_work(&ctrl->scan_work);
1129 	}
1130 
1131 	switch (cmd->common.opcode) {
1132 	case nvme_admin_set_features:
1133 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1134 		case NVME_FEAT_KATO:
1135 			/*
1136 			 * Keep alive commands interval on the host should be
1137 			 * updated when KATO is modified by Set Features
1138 			 * commands.
1139 			 */
1140 			if (!status)
1141 				nvme_update_keep_alive(ctrl, cmd);
1142 			break;
1143 		default:
1144 			break;
1145 		}
1146 		break;
1147 	default:
1148 		break;
1149 	}
1150 }
1151 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1152 
1153 int nvme_execute_passthru_rq(struct request *rq, u32 *effects)
1154 {
1155 	struct nvme_command *cmd = nvme_req(rq)->cmd;
1156 	struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1157 	struct nvme_ns *ns = rq->q->queuedata;
1158 
1159 	*effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1160 	return nvme_execute_rq(rq, false);
1161 }
1162 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1163 
1164 /*
1165  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1166  *
1167  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1168  *   accounting for transport roundtrip times [..].
1169  */
1170 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1171 {
1172 	queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1173 }
1174 
1175 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1176 						 blk_status_t status)
1177 {
1178 	struct nvme_ctrl *ctrl = rq->end_io_data;
1179 	unsigned long flags;
1180 	bool startka = false;
1181 
1182 	blk_mq_free_request(rq);
1183 
1184 	if (status) {
1185 		dev_err(ctrl->device,
1186 			"failed nvme_keep_alive_end_io error=%d\n",
1187 				status);
1188 		return RQ_END_IO_NONE;
1189 	}
1190 
1191 	ctrl->comp_seen = false;
1192 	spin_lock_irqsave(&ctrl->lock, flags);
1193 	if (ctrl->state == NVME_CTRL_LIVE ||
1194 	    ctrl->state == NVME_CTRL_CONNECTING)
1195 		startka = true;
1196 	spin_unlock_irqrestore(&ctrl->lock, flags);
1197 	if (startka)
1198 		nvme_queue_keep_alive_work(ctrl);
1199 	return RQ_END_IO_NONE;
1200 }
1201 
1202 static void nvme_keep_alive_work(struct work_struct *work)
1203 {
1204 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1205 			struct nvme_ctrl, ka_work);
1206 	bool comp_seen = ctrl->comp_seen;
1207 	struct request *rq;
1208 
1209 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1210 		dev_dbg(ctrl->device,
1211 			"reschedule traffic based keep-alive timer\n");
1212 		ctrl->comp_seen = false;
1213 		nvme_queue_keep_alive_work(ctrl);
1214 		return;
1215 	}
1216 
1217 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1218 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1219 	if (IS_ERR(rq)) {
1220 		/* allocation failure, reset the controller */
1221 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1222 		nvme_reset_ctrl(ctrl);
1223 		return;
1224 	}
1225 	nvme_init_request(rq, &ctrl->ka_cmd);
1226 
1227 	rq->timeout = ctrl->kato * HZ;
1228 	rq->end_io = nvme_keep_alive_end_io;
1229 	rq->end_io_data = ctrl;
1230 	blk_execute_rq_nowait(rq, false);
1231 }
1232 
1233 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1234 {
1235 	if (unlikely(ctrl->kato == 0))
1236 		return;
1237 
1238 	nvme_queue_keep_alive_work(ctrl);
1239 }
1240 
1241 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1242 {
1243 	if (unlikely(ctrl->kato == 0))
1244 		return;
1245 
1246 	cancel_delayed_work_sync(&ctrl->ka_work);
1247 }
1248 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1249 
1250 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1251 				   struct nvme_command *cmd)
1252 {
1253 	unsigned int new_kato =
1254 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1255 
1256 	dev_info(ctrl->device,
1257 		 "keep alive interval updated from %u ms to %u ms\n",
1258 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1259 
1260 	nvme_stop_keep_alive(ctrl);
1261 	ctrl->kato = new_kato;
1262 	nvme_start_keep_alive(ctrl);
1263 }
1264 
1265 /*
1266  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1267  * flag, thus sending any new CNS opcodes has a big chance of not working.
1268  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1269  * (but not for any later version).
1270  */
1271 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1272 {
1273 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1274 		return ctrl->vs < NVME_VS(1, 2, 0);
1275 	return ctrl->vs < NVME_VS(1, 1, 0);
1276 }
1277 
1278 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1279 {
1280 	struct nvme_command c = { };
1281 	int error;
1282 
1283 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1284 	c.identify.opcode = nvme_admin_identify;
1285 	c.identify.cns = NVME_ID_CNS_CTRL;
1286 
1287 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1288 	if (!*id)
1289 		return -ENOMEM;
1290 
1291 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1292 			sizeof(struct nvme_id_ctrl));
1293 	if (error)
1294 		kfree(*id);
1295 	return error;
1296 }
1297 
1298 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1299 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1300 {
1301 	const char *warn_str = "ctrl returned bogus length:";
1302 	void *data = cur;
1303 
1304 	switch (cur->nidt) {
1305 	case NVME_NIDT_EUI64:
1306 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1307 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1308 				 warn_str, cur->nidl);
1309 			return -1;
1310 		}
1311 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1312 			return NVME_NIDT_EUI64_LEN;
1313 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1314 		return NVME_NIDT_EUI64_LEN;
1315 	case NVME_NIDT_NGUID:
1316 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1317 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1318 				 warn_str, cur->nidl);
1319 			return -1;
1320 		}
1321 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1322 			return NVME_NIDT_NGUID_LEN;
1323 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1324 		return NVME_NIDT_NGUID_LEN;
1325 	case NVME_NIDT_UUID:
1326 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1327 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1328 				 warn_str, cur->nidl);
1329 			return -1;
1330 		}
1331 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1332 			return NVME_NIDT_UUID_LEN;
1333 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1334 		return NVME_NIDT_UUID_LEN;
1335 	case NVME_NIDT_CSI:
1336 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1337 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1338 				 warn_str, cur->nidl);
1339 			return -1;
1340 		}
1341 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1342 		*csi_seen = true;
1343 		return NVME_NIDT_CSI_LEN;
1344 	default:
1345 		/* Skip unknown types */
1346 		return cur->nidl;
1347 	}
1348 }
1349 
1350 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1351 		struct nvme_ns_info *info)
1352 {
1353 	struct nvme_command c = { };
1354 	bool csi_seen = false;
1355 	int status, pos, len;
1356 	void *data;
1357 
1358 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1359 		return 0;
1360 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1361 		return 0;
1362 
1363 	c.identify.opcode = nvme_admin_identify;
1364 	c.identify.nsid = cpu_to_le32(info->nsid);
1365 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1366 
1367 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1368 	if (!data)
1369 		return -ENOMEM;
1370 
1371 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1372 				      NVME_IDENTIFY_DATA_SIZE);
1373 	if (status) {
1374 		dev_warn(ctrl->device,
1375 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1376 			info->nsid, status);
1377 		goto free_data;
1378 	}
1379 
1380 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1381 		struct nvme_ns_id_desc *cur = data + pos;
1382 
1383 		if (cur->nidl == 0)
1384 			break;
1385 
1386 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1387 		if (len < 0)
1388 			break;
1389 
1390 		len += sizeof(*cur);
1391 	}
1392 
1393 	if (nvme_multi_css(ctrl) && !csi_seen) {
1394 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1395 			 info->nsid);
1396 		status = -EINVAL;
1397 	}
1398 
1399 free_data:
1400 	kfree(data);
1401 	return status;
1402 }
1403 
1404 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1405 			struct nvme_id_ns **id)
1406 {
1407 	struct nvme_command c = { };
1408 	int error;
1409 
1410 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1411 	c.identify.opcode = nvme_admin_identify;
1412 	c.identify.nsid = cpu_to_le32(nsid);
1413 	c.identify.cns = NVME_ID_CNS_NS;
1414 
1415 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1416 	if (!*id)
1417 		return -ENOMEM;
1418 
1419 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1420 	if (error) {
1421 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1422 		goto out_free_id;
1423 	}
1424 
1425 	error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1426 	if ((*id)->ncap == 0) /* namespace not allocated or attached */
1427 		goto out_free_id;
1428 	return 0;
1429 
1430 out_free_id:
1431 	kfree(*id);
1432 	return error;
1433 }
1434 
1435 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1436 		struct nvme_ns_info *info)
1437 {
1438 	struct nvme_ns_ids *ids = &info->ids;
1439 	struct nvme_id_ns *id;
1440 	int ret;
1441 
1442 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1443 	if (ret)
1444 		return ret;
1445 	info->anagrpid = id->anagrpid;
1446 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1447 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1448 	info->is_ready = true;
1449 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1450 		dev_info(ctrl->device,
1451 			 "Ignoring bogus Namespace Identifiers\n");
1452 	} else {
1453 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1454 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1455 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1456 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1457 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1458 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1459 	}
1460 	kfree(id);
1461 	return 0;
1462 }
1463 
1464 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1465 		struct nvme_ns_info *info)
1466 {
1467 	struct nvme_id_ns_cs_indep *id;
1468 	struct nvme_command c = {
1469 		.identify.opcode	= nvme_admin_identify,
1470 		.identify.nsid		= cpu_to_le32(info->nsid),
1471 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1472 	};
1473 	int ret;
1474 
1475 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1476 	if (!id)
1477 		return -ENOMEM;
1478 
1479 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1480 	if (!ret) {
1481 		info->anagrpid = id->anagrpid;
1482 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1483 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1484 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1485 	}
1486 	kfree(id);
1487 	return ret;
1488 }
1489 
1490 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1491 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1492 {
1493 	union nvme_result res = { 0 };
1494 	struct nvme_command c = { };
1495 	int ret;
1496 
1497 	c.features.opcode = op;
1498 	c.features.fid = cpu_to_le32(fid);
1499 	c.features.dword11 = cpu_to_le32(dword11);
1500 
1501 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1502 			buffer, buflen, NVME_QID_ANY, 0, 0);
1503 	if (ret >= 0 && result)
1504 		*result = le32_to_cpu(res.u32);
1505 	return ret;
1506 }
1507 
1508 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1509 		      unsigned int dword11, void *buffer, size_t buflen,
1510 		      u32 *result)
1511 {
1512 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1513 			     buflen, result);
1514 }
1515 EXPORT_SYMBOL_GPL(nvme_set_features);
1516 
1517 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1518 		      unsigned int dword11, void *buffer, size_t buflen,
1519 		      u32 *result)
1520 {
1521 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1522 			     buflen, result);
1523 }
1524 EXPORT_SYMBOL_GPL(nvme_get_features);
1525 
1526 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1527 {
1528 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1529 	u32 result;
1530 	int status, nr_io_queues;
1531 
1532 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1533 			&result);
1534 	if (status < 0)
1535 		return status;
1536 
1537 	/*
1538 	 * Degraded controllers might return an error when setting the queue
1539 	 * count.  We still want to be able to bring them online and offer
1540 	 * access to the admin queue, as that might be only way to fix them up.
1541 	 */
1542 	if (status > 0) {
1543 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1544 		*count = 0;
1545 	} else {
1546 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1547 		*count = min(*count, nr_io_queues);
1548 	}
1549 
1550 	return 0;
1551 }
1552 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1553 
1554 #define NVME_AEN_SUPPORTED \
1555 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1556 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1557 
1558 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1559 {
1560 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1561 	int status;
1562 
1563 	if (!supported_aens)
1564 		return;
1565 
1566 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1567 			NULL, 0, &result);
1568 	if (status)
1569 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1570 			 supported_aens);
1571 
1572 	queue_work(nvme_wq, &ctrl->async_event_work);
1573 }
1574 
1575 static int nvme_ns_open(struct nvme_ns *ns)
1576 {
1577 
1578 	/* should never be called due to GENHD_FL_HIDDEN */
1579 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1580 		goto fail;
1581 	if (!nvme_get_ns(ns))
1582 		goto fail;
1583 	if (!try_module_get(ns->ctrl->ops->module))
1584 		goto fail_put_ns;
1585 
1586 	return 0;
1587 
1588 fail_put_ns:
1589 	nvme_put_ns(ns);
1590 fail:
1591 	return -ENXIO;
1592 }
1593 
1594 static void nvme_ns_release(struct nvme_ns *ns)
1595 {
1596 
1597 	module_put(ns->ctrl->ops->module);
1598 	nvme_put_ns(ns);
1599 }
1600 
1601 static int nvme_open(struct block_device *bdev, fmode_t mode)
1602 {
1603 	return nvme_ns_open(bdev->bd_disk->private_data);
1604 }
1605 
1606 static void nvme_release(struct gendisk *disk, fmode_t mode)
1607 {
1608 	nvme_ns_release(disk->private_data);
1609 }
1610 
1611 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1612 {
1613 	/* some standard values */
1614 	geo->heads = 1 << 6;
1615 	geo->sectors = 1 << 5;
1616 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1617 	return 0;
1618 }
1619 
1620 #ifdef CONFIG_BLK_DEV_INTEGRITY
1621 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1622 				u32 max_integrity_segments)
1623 {
1624 	struct blk_integrity integrity = { };
1625 
1626 	switch (ns->pi_type) {
1627 	case NVME_NS_DPS_PI_TYPE3:
1628 		switch (ns->guard_type) {
1629 		case NVME_NVM_NS_16B_GUARD:
1630 			integrity.profile = &t10_pi_type3_crc;
1631 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1632 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1633 			break;
1634 		case NVME_NVM_NS_64B_GUARD:
1635 			integrity.profile = &ext_pi_type3_crc64;
1636 			integrity.tag_size = sizeof(u16) + 6;
1637 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1638 			break;
1639 		default:
1640 			integrity.profile = NULL;
1641 			break;
1642 		}
1643 		break;
1644 	case NVME_NS_DPS_PI_TYPE1:
1645 	case NVME_NS_DPS_PI_TYPE2:
1646 		switch (ns->guard_type) {
1647 		case NVME_NVM_NS_16B_GUARD:
1648 			integrity.profile = &t10_pi_type1_crc;
1649 			integrity.tag_size = sizeof(u16);
1650 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1651 			break;
1652 		case NVME_NVM_NS_64B_GUARD:
1653 			integrity.profile = &ext_pi_type1_crc64;
1654 			integrity.tag_size = sizeof(u16);
1655 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1656 			break;
1657 		default:
1658 			integrity.profile = NULL;
1659 			break;
1660 		}
1661 		break;
1662 	default:
1663 		integrity.profile = NULL;
1664 		break;
1665 	}
1666 
1667 	integrity.tuple_size = ns->ms;
1668 	blk_integrity_register(disk, &integrity);
1669 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1670 }
1671 #else
1672 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1673 				u32 max_integrity_segments)
1674 {
1675 }
1676 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1677 
1678 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1679 {
1680 	struct nvme_ctrl *ctrl = ns->ctrl;
1681 	struct request_queue *queue = disk->queue;
1682 	u32 size = queue_logical_block_size(queue);
1683 
1684 	if (ctrl->max_discard_sectors == 0) {
1685 		blk_queue_max_discard_sectors(queue, 0);
1686 		return;
1687 	}
1688 
1689 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1690 			NVME_DSM_MAX_RANGES);
1691 
1692 	queue->limits.discard_granularity = size;
1693 
1694 	/* If discard is already enabled, don't reset queue limits */
1695 	if (queue->limits.max_discard_sectors)
1696 		return;
1697 
1698 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1699 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1700 
1701 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1702 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1703 
1704 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1705 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1706 }
1707 
1708 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1709 {
1710 	return uuid_equal(&a->uuid, &b->uuid) &&
1711 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1712 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1713 		a->csi == b->csi;
1714 }
1715 
1716 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1717 {
1718 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1719 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1720 	struct nvme_ctrl *ctrl = ns->ctrl;
1721 	struct nvme_command c = { };
1722 	struct nvme_id_ns_nvm *nvm;
1723 	int ret = 0;
1724 	u32 elbaf;
1725 
1726 	ns->pi_size = 0;
1727 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1728 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1729 		ns->pi_size = sizeof(struct t10_pi_tuple);
1730 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1731 		goto set_pi;
1732 	}
1733 
1734 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1735 	if (!nvm)
1736 		return -ENOMEM;
1737 
1738 	c.identify.opcode = nvme_admin_identify;
1739 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1740 	c.identify.cns = NVME_ID_CNS_CS_NS;
1741 	c.identify.csi = NVME_CSI_NVM;
1742 
1743 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1744 	if (ret)
1745 		goto free_data;
1746 
1747 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1748 
1749 	/* no support for storage tag formats right now */
1750 	if (nvme_elbaf_sts(elbaf))
1751 		goto free_data;
1752 
1753 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1754 	switch (ns->guard_type) {
1755 	case NVME_NVM_NS_64B_GUARD:
1756 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1757 		break;
1758 	case NVME_NVM_NS_16B_GUARD:
1759 		ns->pi_size = sizeof(struct t10_pi_tuple);
1760 		break;
1761 	default:
1762 		break;
1763 	}
1764 
1765 free_data:
1766 	kfree(nvm);
1767 set_pi:
1768 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1769 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1770 	else
1771 		ns->pi_type = 0;
1772 
1773 	return ret;
1774 }
1775 
1776 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1777 {
1778 	struct nvme_ctrl *ctrl = ns->ctrl;
1779 
1780 	if (nvme_init_ms(ns, id))
1781 		return;
1782 
1783 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1784 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1785 		return;
1786 
1787 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1788 		/*
1789 		 * The NVMe over Fabrics specification only supports metadata as
1790 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1791 		 * remap the separate metadata buffer from the block layer.
1792 		 */
1793 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1794 			return;
1795 
1796 		ns->features |= NVME_NS_EXT_LBAS;
1797 
1798 		/*
1799 		 * The current fabrics transport drivers support namespace
1800 		 * metadata formats only if nvme_ns_has_pi() returns true.
1801 		 * Suppress support for all other formats so the namespace will
1802 		 * have a 0 capacity and not be usable through the block stack.
1803 		 *
1804 		 * Note, this check will need to be modified if any drivers
1805 		 * gain the ability to use other metadata formats.
1806 		 */
1807 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1808 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1809 	} else {
1810 		/*
1811 		 * For PCIe controllers, we can't easily remap the separate
1812 		 * metadata buffer from the block layer and thus require a
1813 		 * separate metadata buffer for block layer metadata/PI support.
1814 		 * We allow extended LBAs for the passthrough interface, though.
1815 		 */
1816 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1817 			ns->features |= NVME_NS_EXT_LBAS;
1818 		else
1819 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1820 	}
1821 }
1822 
1823 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1824 		struct request_queue *q)
1825 {
1826 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1827 
1828 	if (ctrl->max_hw_sectors) {
1829 		u32 max_segments =
1830 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1831 
1832 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1833 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1834 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1835 	}
1836 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1837 	blk_queue_dma_alignment(q, 3);
1838 	blk_queue_write_cache(q, vwc, vwc);
1839 }
1840 
1841 static void nvme_update_disk_info(struct gendisk *disk,
1842 		struct nvme_ns *ns, struct nvme_id_ns *id)
1843 {
1844 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1845 	unsigned short bs = 1 << ns->lba_shift;
1846 	u32 atomic_bs, phys_bs, io_opt = 0;
1847 
1848 	/*
1849 	 * The block layer can't support LBA sizes larger than the page size
1850 	 * yet, so catch this early and don't allow block I/O.
1851 	 */
1852 	if (ns->lba_shift > PAGE_SHIFT) {
1853 		capacity = 0;
1854 		bs = (1 << 9);
1855 	}
1856 
1857 	blk_integrity_unregister(disk);
1858 
1859 	atomic_bs = phys_bs = bs;
1860 	if (id->nabo == 0) {
1861 		/*
1862 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1863 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1864 		 * 0 then AWUPF must be used instead.
1865 		 */
1866 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1867 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1868 		else
1869 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1870 	}
1871 
1872 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1873 		/* NPWG = Namespace Preferred Write Granularity */
1874 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1875 		/* NOWS = Namespace Optimal Write Size */
1876 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1877 	}
1878 
1879 	blk_queue_logical_block_size(disk->queue, bs);
1880 	/*
1881 	 * Linux filesystems assume writing a single physical block is
1882 	 * an atomic operation. Hence limit the physical block size to the
1883 	 * value of the Atomic Write Unit Power Fail parameter.
1884 	 */
1885 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1886 	blk_queue_io_min(disk->queue, phys_bs);
1887 	blk_queue_io_opt(disk->queue, io_opt);
1888 
1889 	/*
1890 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1891 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1892 	 * I/O to namespaces with metadata except when the namespace supports
1893 	 * PI, as it can strip/insert in that case.
1894 	 */
1895 	if (ns->ms) {
1896 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1897 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1898 			nvme_init_integrity(disk, ns,
1899 					    ns->ctrl->max_integrity_segments);
1900 		else if (!nvme_ns_has_pi(ns))
1901 			capacity = 0;
1902 	}
1903 
1904 	set_capacity_and_notify(disk, capacity);
1905 
1906 	nvme_config_discard(disk, ns);
1907 	blk_queue_max_write_zeroes_sectors(disk->queue,
1908 					   ns->ctrl->max_zeroes_sectors);
1909 }
1910 
1911 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1912 {
1913 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1914 }
1915 
1916 static inline bool nvme_first_scan(struct gendisk *disk)
1917 {
1918 	/* nvme_alloc_ns() scans the disk prior to adding it */
1919 	return !disk_live(disk);
1920 }
1921 
1922 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1923 {
1924 	struct nvme_ctrl *ctrl = ns->ctrl;
1925 	u32 iob;
1926 
1927 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1928 	    is_power_of_2(ctrl->max_hw_sectors))
1929 		iob = ctrl->max_hw_sectors;
1930 	else
1931 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1932 
1933 	if (!iob)
1934 		return;
1935 
1936 	if (!is_power_of_2(iob)) {
1937 		if (nvme_first_scan(ns->disk))
1938 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1939 				ns->disk->disk_name, iob);
1940 		return;
1941 	}
1942 
1943 	if (blk_queue_is_zoned(ns->disk->queue)) {
1944 		if (nvme_first_scan(ns->disk))
1945 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1946 				ns->disk->disk_name);
1947 		return;
1948 	}
1949 
1950 	blk_queue_chunk_sectors(ns->queue, iob);
1951 }
1952 
1953 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1954 		struct nvme_ns_info *info)
1955 {
1956 	blk_mq_freeze_queue(ns->disk->queue);
1957 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1958 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1959 	blk_mq_unfreeze_queue(ns->disk->queue);
1960 
1961 	if (nvme_ns_head_multipath(ns->head)) {
1962 		blk_mq_freeze_queue(ns->head->disk->queue);
1963 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1964 		nvme_mpath_revalidate_paths(ns);
1965 		blk_stack_limits(&ns->head->disk->queue->limits,
1966 				 &ns->queue->limits, 0);
1967 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
1968 		blk_mq_unfreeze_queue(ns->head->disk->queue);
1969 	}
1970 
1971 	/* Hide the block-interface for these devices */
1972 	ns->disk->flags |= GENHD_FL_HIDDEN;
1973 	set_bit(NVME_NS_READY, &ns->flags);
1974 
1975 	return 0;
1976 }
1977 
1978 static int nvme_update_ns_info_block(struct nvme_ns *ns,
1979 		struct nvme_ns_info *info)
1980 {
1981 	struct nvme_id_ns *id;
1982 	unsigned lbaf;
1983 	int ret;
1984 
1985 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
1986 	if (ret)
1987 		return ret;
1988 
1989 	blk_mq_freeze_queue(ns->disk->queue);
1990 	lbaf = nvme_lbaf_index(id->flbas);
1991 	ns->lba_shift = id->lbaf[lbaf].ds;
1992 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1993 
1994 	nvme_configure_metadata(ns, id);
1995 	nvme_set_chunk_sectors(ns, id);
1996 	nvme_update_disk_info(ns->disk, ns, id);
1997 
1998 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
1999 		ret = nvme_update_zone_info(ns, lbaf);
2000 		if (ret) {
2001 			blk_mq_unfreeze_queue(ns->disk->queue);
2002 			goto out;
2003 		}
2004 	}
2005 
2006 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2007 	set_bit(NVME_NS_READY, &ns->flags);
2008 	blk_mq_unfreeze_queue(ns->disk->queue);
2009 
2010 	if (blk_queue_is_zoned(ns->queue)) {
2011 		ret = nvme_revalidate_zones(ns);
2012 		if (ret && !nvme_first_scan(ns->disk))
2013 			goto out;
2014 	}
2015 
2016 	if (nvme_ns_head_multipath(ns->head)) {
2017 		blk_mq_freeze_queue(ns->head->disk->queue);
2018 		nvme_update_disk_info(ns->head->disk, ns, id);
2019 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2020 		nvme_mpath_revalidate_paths(ns);
2021 		blk_stack_limits(&ns->head->disk->queue->limits,
2022 				 &ns->queue->limits, 0);
2023 		disk_update_readahead(ns->head->disk);
2024 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2025 	}
2026 
2027 	ret = 0;
2028 out:
2029 	/*
2030 	 * If probing fails due an unsupported feature, hide the block device,
2031 	 * but still allow other access.
2032 	 */
2033 	if (ret == -ENODEV) {
2034 		ns->disk->flags |= GENHD_FL_HIDDEN;
2035 		set_bit(NVME_NS_READY, &ns->flags);
2036 		ret = 0;
2037 	}
2038 	kfree(id);
2039 	return ret;
2040 }
2041 
2042 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2043 {
2044 	switch (info->ids.csi) {
2045 	case NVME_CSI_ZNS:
2046 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2047 			dev_info(ns->ctrl->device,
2048 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2049 				info->nsid);
2050 			return nvme_update_ns_info_generic(ns, info);
2051 		}
2052 		return nvme_update_ns_info_block(ns, info);
2053 	case NVME_CSI_NVM:
2054 		return nvme_update_ns_info_block(ns, info);
2055 	default:
2056 		dev_info(ns->ctrl->device,
2057 			"block device for nsid %u not supported (csi %u)\n",
2058 			info->nsid, info->ids.csi);
2059 		return nvme_update_ns_info_generic(ns, info);
2060 	}
2061 }
2062 
2063 static char nvme_pr_type(enum pr_type type)
2064 {
2065 	switch (type) {
2066 	case PR_WRITE_EXCLUSIVE:
2067 		return 1;
2068 	case PR_EXCLUSIVE_ACCESS:
2069 		return 2;
2070 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
2071 		return 3;
2072 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2073 		return 4;
2074 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
2075 		return 5;
2076 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2077 		return 6;
2078 	default:
2079 		return 0;
2080 	}
2081 }
2082 
2083 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
2084 		struct nvme_command *c, u8 data[16])
2085 {
2086 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
2087 	int srcu_idx = srcu_read_lock(&head->srcu);
2088 	struct nvme_ns *ns = nvme_find_path(head);
2089 	int ret = -EWOULDBLOCK;
2090 
2091 	if (ns) {
2092 		c->common.nsid = cpu_to_le32(ns->head->ns_id);
2093 		ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2094 	}
2095 	srcu_read_unlock(&head->srcu, srcu_idx);
2096 	return ret;
2097 }
2098 
2099 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2100 		u8 data[16])
2101 {
2102 	c->common.nsid = cpu_to_le32(ns->head->ns_id);
2103 	return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2104 }
2105 
2106 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2107 				u64 key, u64 sa_key, u8 op)
2108 {
2109 	struct nvme_command c = { };
2110 	u8 data[16] = { 0, };
2111 
2112 	put_unaligned_le64(key, &data[0]);
2113 	put_unaligned_le64(sa_key, &data[8]);
2114 
2115 	c.common.opcode = op;
2116 	c.common.cdw10 = cpu_to_le32(cdw10);
2117 
2118 	if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2119 	    bdev->bd_disk->fops == &nvme_ns_head_ops)
2120 		return nvme_send_ns_head_pr_command(bdev, &c, data);
2121 	return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2122 }
2123 
2124 static int nvme_pr_register(struct block_device *bdev, u64 old,
2125 		u64 new, unsigned flags)
2126 {
2127 	u32 cdw10;
2128 
2129 	if (flags & ~PR_FL_IGNORE_KEY)
2130 		return -EOPNOTSUPP;
2131 
2132 	cdw10 = old ? 2 : 0;
2133 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2134 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2135 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2136 }
2137 
2138 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2139 		enum pr_type type, unsigned flags)
2140 {
2141 	u32 cdw10;
2142 
2143 	if (flags & ~PR_FL_IGNORE_KEY)
2144 		return -EOPNOTSUPP;
2145 
2146 	cdw10 = nvme_pr_type(type) << 8;
2147 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2148 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2149 }
2150 
2151 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2152 		enum pr_type type, bool abort)
2153 {
2154 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2155 
2156 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2157 }
2158 
2159 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2160 {
2161 	u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2162 
2163 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2164 }
2165 
2166 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2167 {
2168 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2169 
2170 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2171 }
2172 
2173 const struct pr_ops nvme_pr_ops = {
2174 	.pr_register	= nvme_pr_register,
2175 	.pr_reserve	= nvme_pr_reserve,
2176 	.pr_release	= nvme_pr_release,
2177 	.pr_preempt	= nvme_pr_preempt,
2178 	.pr_clear	= nvme_pr_clear,
2179 };
2180 
2181 #ifdef CONFIG_BLK_SED_OPAL
2182 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2183 		bool send)
2184 {
2185 	struct nvme_ctrl *ctrl = data;
2186 	struct nvme_command cmd = { };
2187 
2188 	if (send)
2189 		cmd.common.opcode = nvme_admin_security_send;
2190 	else
2191 		cmd.common.opcode = nvme_admin_security_recv;
2192 	cmd.common.nsid = 0;
2193 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2194 	cmd.common.cdw11 = cpu_to_le32(len);
2195 
2196 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2197 			NVME_QID_ANY, 1, 0);
2198 }
2199 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2200 #endif /* CONFIG_BLK_SED_OPAL */
2201 
2202 #ifdef CONFIG_BLK_DEV_ZONED
2203 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2204 		unsigned int nr_zones, report_zones_cb cb, void *data)
2205 {
2206 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2207 			data);
2208 }
2209 #else
2210 #define nvme_report_zones	NULL
2211 #endif /* CONFIG_BLK_DEV_ZONED */
2212 
2213 static const struct block_device_operations nvme_bdev_ops = {
2214 	.owner		= THIS_MODULE,
2215 	.ioctl		= nvme_ioctl,
2216 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2217 	.open		= nvme_open,
2218 	.release	= nvme_release,
2219 	.getgeo		= nvme_getgeo,
2220 	.report_zones	= nvme_report_zones,
2221 	.pr_ops		= &nvme_pr_ops,
2222 };
2223 
2224 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 timeout, bool enabled)
2225 {
2226 	unsigned long timeout_jiffies = ((timeout + 1) * HZ / 2) + jiffies;
2227 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2228 	int ret;
2229 
2230 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2231 		if (csts == ~0)
2232 			return -ENODEV;
2233 		if ((csts & NVME_CSTS_RDY) == bit)
2234 			break;
2235 
2236 		usleep_range(1000, 2000);
2237 		if (fatal_signal_pending(current))
2238 			return -EINTR;
2239 		if (time_after(jiffies, timeout_jiffies)) {
2240 			dev_err(ctrl->device,
2241 				"Device not ready; aborting %s, CSTS=0x%x\n",
2242 				enabled ? "initialisation" : "reset", csts);
2243 			return -ENODEV;
2244 		}
2245 	}
2246 
2247 	return ret;
2248 }
2249 
2250 /*
2251  * If the device has been passed off to us in an enabled state, just clear
2252  * the enabled bit.  The spec says we should set the 'shutdown notification
2253  * bits', but doing so may cause the device to complete commands to the
2254  * admin queue ... and we don't know what memory that might be pointing at!
2255  */
2256 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2257 {
2258 	int ret;
2259 
2260 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2261 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2262 
2263 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2264 	if (ret)
2265 		return ret;
2266 
2267 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2268 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2269 
2270 	return nvme_wait_ready(ctrl, NVME_CAP_TIMEOUT(ctrl->cap), false);
2271 }
2272 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2273 
2274 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2275 {
2276 	unsigned dev_page_min;
2277 	u32 timeout;
2278 	int ret;
2279 
2280 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2281 	if (ret) {
2282 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2283 		return ret;
2284 	}
2285 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2286 
2287 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2288 		dev_err(ctrl->device,
2289 			"Minimum device page size %u too large for host (%u)\n",
2290 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2291 		return -ENODEV;
2292 	}
2293 
2294 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2295 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2296 	else
2297 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2298 
2299 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2300 		u32 crto;
2301 
2302 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2303 		if (ret) {
2304 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2305 				ret);
2306 			return ret;
2307 		}
2308 
2309 		if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2310 			ctrl->ctrl_config |= NVME_CC_CRIME;
2311 			timeout = NVME_CRTO_CRIMT(crto);
2312 		} else {
2313 			timeout = NVME_CRTO_CRWMT(crto);
2314 		}
2315 	} else {
2316 		timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2317 	}
2318 
2319 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2320 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2321 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2322 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2323 	if (ret)
2324 		return ret;
2325 
2326 	/* Flush write to device (required if transport is PCI) */
2327 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2328 	if (ret)
2329 		return ret;
2330 
2331 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2332 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2333 	if (ret)
2334 		return ret;
2335 	return nvme_wait_ready(ctrl, timeout, true);
2336 }
2337 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2338 
2339 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2340 {
2341 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2342 	u32 csts;
2343 	int ret;
2344 
2345 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2346 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2347 
2348 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2349 	if (ret)
2350 		return ret;
2351 
2352 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2353 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2354 			break;
2355 
2356 		msleep(100);
2357 		if (fatal_signal_pending(current))
2358 			return -EINTR;
2359 		if (time_after(jiffies, timeout)) {
2360 			dev_err(ctrl->device,
2361 				"Device shutdown incomplete; abort shutdown\n");
2362 			return -ENODEV;
2363 		}
2364 	}
2365 
2366 	return ret;
2367 }
2368 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2369 
2370 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2371 {
2372 	__le64 ts;
2373 	int ret;
2374 
2375 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2376 		return 0;
2377 
2378 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2379 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2380 			NULL);
2381 	if (ret)
2382 		dev_warn_once(ctrl->device,
2383 			"could not set timestamp (%d)\n", ret);
2384 	return ret;
2385 }
2386 
2387 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2388 {
2389 	struct nvme_feat_host_behavior *host;
2390 	u8 acre = 0, lbafee = 0;
2391 	int ret;
2392 
2393 	/* Don't bother enabling the feature if retry delay is not reported */
2394 	if (ctrl->crdt[0])
2395 		acre = NVME_ENABLE_ACRE;
2396 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2397 		lbafee = NVME_ENABLE_LBAFEE;
2398 
2399 	if (!acre && !lbafee)
2400 		return 0;
2401 
2402 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2403 	if (!host)
2404 		return 0;
2405 
2406 	host->acre = acre;
2407 	host->lbafee = lbafee;
2408 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2409 				host, sizeof(*host), NULL);
2410 	kfree(host);
2411 	return ret;
2412 }
2413 
2414 /*
2415  * The function checks whether the given total (exlat + enlat) latency of
2416  * a power state allows the latter to be used as an APST transition target.
2417  * It does so by comparing the latency to the primary and secondary latency
2418  * tolerances defined by module params. If there's a match, the corresponding
2419  * timeout value is returned and the matching tolerance index (1 or 2) is
2420  * reported.
2421  */
2422 static bool nvme_apst_get_transition_time(u64 total_latency,
2423 		u64 *transition_time, unsigned *last_index)
2424 {
2425 	if (total_latency <= apst_primary_latency_tol_us) {
2426 		if (*last_index == 1)
2427 			return false;
2428 		*last_index = 1;
2429 		*transition_time = apst_primary_timeout_ms;
2430 		return true;
2431 	}
2432 	if (apst_secondary_timeout_ms &&
2433 		total_latency <= apst_secondary_latency_tol_us) {
2434 		if (*last_index <= 2)
2435 			return false;
2436 		*last_index = 2;
2437 		*transition_time = apst_secondary_timeout_ms;
2438 		return true;
2439 	}
2440 	return false;
2441 }
2442 
2443 /*
2444  * APST (Autonomous Power State Transition) lets us program a table of power
2445  * state transitions that the controller will perform automatically.
2446  *
2447  * Depending on module params, one of the two supported techniques will be used:
2448  *
2449  * - If the parameters provide explicit timeouts and tolerances, they will be
2450  *   used to build a table with up to 2 non-operational states to transition to.
2451  *   The default parameter values were selected based on the values used by
2452  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2453  *   regeneration of the APST table in the event of switching between external
2454  *   and battery power, the timeouts and tolerances reflect a compromise
2455  *   between values used by Microsoft for AC and battery scenarios.
2456  * - If not, we'll configure the table with a simple heuristic: we are willing
2457  *   to spend at most 2% of the time transitioning between power states.
2458  *   Therefore, when running in any given state, we will enter the next
2459  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2460  *   microseconds, as long as that state's exit latency is under the requested
2461  *   maximum latency.
2462  *
2463  * We will not autonomously enter any non-operational state for which the total
2464  * latency exceeds ps_max_latency_us.
2465  *
2466  * Users can set ps_max_latency_us to zero to turn off APST.
2467  */
2468 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2469 {
2470 	struct nvme_feat_auto_pst *table;
2471 	unsigned apste = 0;
2472 	u64 max_lat_us = 0;
2473 	__le64 target = 0;
2474 	int max_ps = -1;
2475 	int state;
2476 	int ret;
2477 	unsigned last_lt_index = UINT_MAX;
2478 
2479 	/*
2480 	 * If APST isn't supported or if we haven't been initialized yet,
2481 	 * then don't do anything.
2482 	 */
2483 	if (!ctrl->apsta)
2484 		return 0;
2485 
2486 	if (ctrl->npss > 31) {
2487 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2488 		return 0;
2489 	}
2490 
2491 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2492 	if (!table)
2493 		return 0;
2494 
2495 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2496 		/* Turn off APST. */
2497 		dev_dbg(ctrl->device, "APST disabled\n");
2498 		goto done;
2499 	}
2500 
2501 	/*
2502 	 * Walk through all states from lowest- to highest-power.
2503 	 * According to the spec, lower-numbered states use more power.  NPSS,
2504 	 * despite the name, is the index of the lowest-power state, not the
2505 	 * number of states.
2506 	 */
2507 	for (state = (int)ctrl->npss; state >= 0; state--) {
2508 		u64 total_latency_us, exit_latency_us, transition_ms;
2509 
2510 		if (target)
2511 			table->entries[state] = target;
2512 
2513 		/*
2514 		 * Don't allow transitions to the deepest state if it's quirked
2515 		 * off.
2516 		 */
2517 		if (state == ctrl->npss &&
2518 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2519 			continue;
2520 
2521 		/*
2522 		 * Is this state a useful non-operational state for higher-power
2523 		 * states to autonomously transition to?
2524 		 */
2525 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2526 			continue;
2527 
2528 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2529 		if (exit_latency_us > ctrl->ps_max_latency_us)
2530 			continue;
2531 
2532 		total_latency_us = exit_latency_us +
2533 			le32_to_cpu(ctrl->psd[state].entry_lat);
2534 
2535 		/*
2536 		 * This state is good. It can be used as the APST idle target
2537 		 * for higher power states.
2538 		 */
2539 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2540 			if (!nvme_apst_get_transition_time(total_latency_us,
2541 					&transition_ms, &last_lt_index))
2542 				continue;
2543 		} else {
2544 			transition_ms = total_latency_us + 19;
2545 			do_div(transition_ms, 20);
2546 			if (transition_ms > (1 << 24) - 1)
2547 				transition_ms = (1 << 24) - 1;
2548 		}
2549 
2550 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2551 		if (max_ps == -1)
2552 			max_ps = state;
2553 		if (total_latency_us > max_lat_us)
2554 			max_lat_us = total_latency_us;
2555 	}
2556 
2557 	if (max_ps == -1)
2558 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2559 	else
2560 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2561 			max_ps, max_lat_us, (int)sizeof(*table), table);
2562 	apste = 1;
2563 
2564 done:
2565 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2566 				table, sizeof(*table), NULL);
2567 	if (ret)
2568 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2569 	kfree(table);
2570 	return ret;
2571 }
2572 
2573 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2574 {
2575 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2576 	u64 latency;
2577 
2578 	switch (val) {
2579 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2580 	case PM_QOS_LATENCY_ANY:
2581 		latency = U64_MAX;
2582 		break;
2583 
2584 	default:
2585 		latency = val;
2586 	}
2587 
2588 	if (ctrl->ps_max_latency_us != latency) {
2589 		ctrl->ps_max_latency_us = latency;
2590 		if (ctrl->state == NVME_CTRL_LIVE)
2591 			nvme_configure_apst(ctrl);
2592 	}
2593 }
2594 
2595 struct nvme_core_quirk_entry {
2596 	/*
2597 	 * NVMe model and firmware strings are padded with spaces.  For
2598 	 * simplicity, strings in the quirk table are padded with NULLs
2599 	 * instead.
2600 	 */
2601 	u16 vid;
2602 	const char *mn;
2603 	const char *fr;
2604 	unsigned long quirks;
2605 };
2606 
2607 static const struct nvme_core_quirk_entry core_quirks[] = {
2608 	{
2609 		/*
2610 		 * This Toshiba device seems to die using any APST states.  See:
2611 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2612 		 */
2613 		.vid = 0x1179,
2614 		.mn = "THNSF5256GPUK TOSHIBA",
2615 		.quirks = NVME_QUIRK_NO_APST,
2616 	},
2617 	{
2618 		/*
2619 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2620 		 * condition associated with actions related to suspend to idle
2621 		 * LiteON has resolved the problem in future firmware
2622 		 */
2623 		.vid = 0x14a4,
2624 		.fr = "22301111",
2625 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2626 	},
2627 	{
2628 		/*
2629 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2630 		 * aborts I/O during any load, but more easily reproducible
2631 		 * with discards (fstrim).
2632 		 *
2633 		 * The device is left in a state where it is also not possible
2634 		 * to use "nvme set-feature" to disable APST, but booting with
2635 		 * nvme_core.default_ps_max_latency=0 works.
2636 		 */
2637 		.vid = 0x1e0f,
2638 		.mn = "KCD6XVUL6T40",
2639 		.quirks = NVME_QUIRK_NO_APST,
2640 	},
2641 	{
2642 		/*
2643 		 * The external Samsung X5 SSD fails initialization without a
2644 		 * delay before checking if it is ready and has a whole set of
2645 		 * other problems.  To make this even more interesting, it
2646 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2647 		 * does not need or want these quirks.
2648 		 */
2649 		.vid = 0x144d,
2650 		.mn = "Samsung Portable SSD X5",
2651 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2652 			  NVME_QUIRK_NO_DEEPEST_PS |
2653 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2654 	}
2655 };
2656 
2657 /* match is null-terminated but idstr is space-padded. */
2658 static bool string_matches(const char *idstr, const char *match, size_t len)
2659 {
2660 	size_t matchlen;
2661 
2662 	if (!match)
2663 		return true;
2664 
2665 	matchlen = strlen(match);
2666 	WARN_ON_ONCE(matchlen > len);
2667 
2668 	if (memcmp(idstr, match, matchlen))
2669 		return false;
2670 
2671 	for (; matchlen < len; matchlen++)
2672 		if (idstr[matchlen] != ' ')
2673 			return false;
2674 
2675 	return true;
2676 }
2677 
2678 static bool quirk_matches(const struct nvme_id_ctrl *id,
2679 			  const struct nvme_core_quirk_entry *q)
2680 {
2681 	return q->vid == le16_to_cpu(id->vid) &&
2682 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2683 		string_matches(id->fr, q->fr, sizeof(id->fr));
2684 }
2685 
2686 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2687 		struct nvme_id_ctrl *id)
2688 {
2689 	size_t nqnlen;
2690 	int off;
2691 
2692 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2693 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2694 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2695 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2696 			return;
2697 		}
2698 
2699 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2700 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2701 	}
2702 
2703 	/*
2704 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2705 	 * Base Specification 2.0.  It is slightly different from the format
2706 	 * specified there due to historic reasons, and we can't change it now.
2707 	 */
2708 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2709 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2710 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2711 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2712 	off += sizeof(id->sn);
2713 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2714 	off += sizeof(id->mn);
2715 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2716 }
2717 
2718 static void nvme_release_subsystem(struct device *dev)
2719 {
2720 	struct nvme_subsystem *subsys =
2721 		container_of(dev, struct nvme_subsystem, dev);
2722 
2723 	if (subsys->instance >= 0)
2724 		ida_free(&nvme_instance_ida, subsys->instance);
2725 	kfree(subsys);
2726 }
2727 
2728 static void nvme_destroy_subsystem(struct kref *ref)
2729 {
2730 	struct nvme_subsystem *subsys =
2731 			container_of(ref, struct nvme_subsystem, ref);
2732 
2733 	mutex_lock(&nvme_subsystems_lock);
2734 	list_del(&subsys->entry);
2735 	mutex_unlock(&nvme_subsystems_lock);
2736 
2737 	ida_destroy(&subsys->ns_ida);
2738 	device_del(&subsys->dev);
2739 	put_device(&subsys->dev);
2740 }
2741 
2742 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2743 {
2744 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2745 }
2746 
2747 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2748 {
2749 	struct nvme_subsystem *subsys;
2750 
2751 	lockdep_assert_held(&nvme_subsystems_lock);
2752 
2753 	/*
2754 	 * Fail matches for discovery subsystems. This results
2755 	 * in each discovery controller bound to a unique subsystem.
2756 	 * This avoids issues with validating controller values
2757 	 * that can only be true when there is a single unique subsystem.
2758 	 * There may be multiple and completely independent entities
2759 	 * that provide discovery controllers.
2760 	 */
2761 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2762 		return NULL;
2763 
2764 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2765 		if (strcmp(subsys->subnqn, subsysnqn))
2766 			continue;
2767 		if (!kref_get_unless_zero(&subsys->ref))
2768 			continue;
2769 		return subsys;
2770 	}
2771 
2772 	return NULL;
2773 }
2774 
2775 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2776 	struct device_attribute subsys_attr_##_name = \
2777 		__ATTR(_name, _mode, _show, NULL)
2778 
2779 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2780 				    struct device_attribute *attr,
2781 				    char *buf)
2782 {
2783 	struct nvme_subsystem *subsys =
2784 		container_of(dev, struct nvme_subsystem, dev);
2785 
2786 	return sysfs_emit(buf, "%s\n", subsys->subnqn);
2787 }
2788 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2789 
2790 static ssize_t nvme_subsys_show_type(struct device *dev,
2791 				    struct device_attribute *attr,
2792 				    char *buf)
2793 {
2794 	struct nvme_subsystem *subsys =
2795 		container_of(dev, struct nvme_subsystem, dev);
2796 
2797 	switch (subsys->subtype) {
2798 	case NVME_NQN_DISC:
2799 		return sysfs_emit(buf, "discovery\n");
2800 	case NVME_NQN_NVME:
2801 		return sysfs_emit(buf, "nvm\n");
2802 	default:
2803 		return sysfs_emit(buf, "reserved\n");
2804 	}
2805 }
2806 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2807 
2808 #define nvme_subsys_show_str_function(field)				\
2809 static ssize_t subsys_##field##_show(struct device *dev,		\
2810 			    struct device_attribute *attr, char *buf)	\
2811 {									\
2812 	struct nvme_subsystem *subsys =					\
2813 		container_of(dev, struct nvme_subsystem, dev);		\
2814 	return sysfs_emit(buf, "%.*s\n",				\
2815 			   (int)sizeof(subsys->field), subsys->field);	\
2816 }									\
2817 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2818 
2819 nvme_subsys_show_str_function(model);
2820 nvme_subsys_show_str_function(serial);
2821 nvme_subsys_show_str_function(firmware_rev);
2822 
2823 static struct attribute *nvme_subsys_attrs[] = {
2824 	&subsys_attr_model.attr,
2825 	&subsys_attr_serial.attr,
2826 	&subsys_attr_firmware_rev.attr,
2827 	&subsys_attr_subsysnqn.attr,
2828 	&subsys_attr_subsystype.attr,
2829 #ifdef CONFIG_NVME_MULTIPATH
2830 	&subsys_attr_iopolicy.attr,
2831 #endif
2832 	NULL,
2833 };
2834 
2835 static const struct attribute_group nvme_subsys_attrs_group = {
2836 	.attrs = nvme_subsys_attrs,
2837 };
2838 
2839 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2840 	&nvme_subsys_attrs_group,
2841 	NULL,
2842 };
2843 
2844 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2845 {
2846 	return ctrl->opts && ctrl->opts->discovery_nqn;
2847 }
2848 
2849 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2850 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2851 {
2852 	struct nvme_ctrl *tmp;
2853 
2854 	lockdep_assert_held(&nvme_subsystems_lock);
2855 
2856 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2857 		if (nvme_state_terminal(tmp))
2858 			continue;
2859 
2860 		if (tmp->cntlid == ctrl->cntlid) {
2861 			dev_err(ctrl->device,
2862 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2863 				ctrl->cntlid, dev_name(tmp->device),
2864 				subsys->subnqn);
2865 			return false;
2866 		}
2867 
2868 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2869 		    nvme_discovery_ctrl(ctrl))
2870 			continue;
2871 
2872 		dev_err(ctrl->device,
2873 			"Subsystem does not support multiple controllers\n");
2874 		return false;
2875 	}
2876 
2877 	return true;
2878 }
2879 
2880 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2881 {
2882 	struct nvme_subsystem *subsys, *found;
2883 	int ret;
2884 
2885 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2886 	if (!subsys)
2887 		return -ENOMEM;
2888 
2889 	subsys->instance = -1;
2890 	mutex_init(&subsys->lock);
2891 	kref_init(&subsys->ref);
2892 	INIT_LIST_HEAD(&subsys->ctrls);
2893 	INIT_LIST_HEAD(&subsys->nsheads);
2894 	nvme_init_subnqn(subsys, ctrl, id);
2895 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2896 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2897 	subsys->vendor_id = le16_to_cpu(id->vid);
2898 	subsys->cmic = id->cmic;
2899 
2900 	/* Versions prior to 1.4 don't necessarily report a valid type */
2901 	if (id->cntrltype == NVME_CTRL_DISC ||
2902 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2903 		subsys->subtype = NVME_NQN_DISC;
2904 	else
2905 		subsys->subtype = NVME_NQN_NVME;
2906 
2907 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2908 		dev_err(ctrl->device,
2909 			"Subsystem %s is not a discovery controller",
2910 			subsys->subnqn);
2911 		kfree(subsys);
2912 		return -EINVAL;
2913 	}
2914 	subsys->awupf = le16_to_cpu(id->awupf);
2915 	nvme_mpath_default_iopolicy(subsys);
2916 
2917 	subsys->dev.class = nvme_subsys_class;
2918 	subsys->dev.release = nvme_release_subsystem;
2919 	subsys->dev.groups = nvme_subsys_attrs_groups;
2920 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2921 	device_initialize(&subsys->dev);
2922 
2923 	mutex_lock(&nvme_subsystems_lock);
2924 	found = __nvme_find_get_subsystem(subsys->subnqn);
2925 	if (found) {
2926 		put_device(&subsys->dev);
2927 		subsys = found;
2928 
2929 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2930 			ret = -EINVAL;
2931 			goto out_put_subsystem;
2932 		}
2933 	} else {
2934 		ret = device_add(&subsys->dev);
2935 		if (ret) {
2936 			dev_err(ctrl->device,
2937 				"failed to register subsystem device.\n");
2938 			put_device(&subsys->dev);
2939 			goto out_unlock;
2940 		}
2941 		ida_init(&subsys->ns_ida);
2942 		list_add_tail(&subsys->entry, &nvme_subsystems);
2943 	}
2944 
2945 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2946 				dev_name(ctrl->device));
2947 	if (ret) {
2948 		dev_err(ctrl->device,
2949 			"failed to create sysfs link from subsystem.\n");
2950 		goto out_put_subsystem;
2951 	}
2952 
2953 	if (!found)
2954 		subsys->instance = ctrl->instance;
2955 	ctrl->subsys = subsys;
2956 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2957 	mutex_unlock(&nvme_subsystems_lock);
2958 	return 0;
2959 
2960 out_put_subsystem:
2961 	nvme_put_subsystem(subsys);
2962 out_unlock:
2963 	mutex_unlock(&nvme_subsystems_lock);
2964 	return ret;
2965 }
2966 
2967 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2968 		void *log, size_t size, u64 offset)
2969 {
2970 	struct nvme_command c = { };
2971 	u32 dwlen = nvme_bytes_to_numd(size);
2972 
2973 	c.get_log_page.opcode = nvme_admin_get_log_page;
2974 	c.get_log_page.nsid = cpu_to_le32(nsid);
2975 	c.get_log_page.lid = log_page;
2976 	c.get_log_page.lsp = lsp;
2977 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2978 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2979 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2980 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2981 	c.get_log_page.csi = csi;
2982 
2983 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2984 }
2985 
2986 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2987 				struct nvme_effects_log **log)
2988 {
2989 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2990 	int ret;
2991 
2992 	if (cel)
2993 		goto out;
2994 
2995 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2996 	if (!cel)
2997 		return -ENOMEM;
2998 
2999 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3000 			cel, sizeof(*cel), 0);
3001 	if (ret) {
3002 		kfree(cel);
3003 		return ret;
3004 	}
3005 
3006 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3007 out:
3008 	*log = cel;
3009 	return 0;
3010 }
3011 
3012 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3013 {
3014 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3015 
3016 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
3017 		return UINT_MAX;
3018 	return val;
3019 }
3020 
3021 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3022 {
3023 	struct nvme_command c = { };
3024 	struct nvme_id_ctrl_nvm *id;
3025 	int ret;
3026 
3027 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3028 		ctrl->max_discard_sectors = UINT_MAX;
3029 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3030 	} else {
3031 		ctrl->max_discard_sectors = 0;
3032 		ctrl->max_discard_segments = 0;
3033 	}
3034 
3035 	/*
3036 	 * Even though NVMe spec explicitly states that MDTS is not applicable
3037 	 * to the write-zeroes, we are cautious and limit the size to the
3038 	 * controllers max_hw_sectors value, which is based on the MDTS field
3039 	 * and possibly other limiting factors.
3040 	 */
3041 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3042 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3043 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3044 	else
3045 		ctrl->max_zeroes_sectors = 0;
3046 
3047 	if (nvme_ctrl_limited_cns(ctrl))
3048 		return 0;
3049 
3050 	id = kzalloc(sizeof(*id), GFP_KERNEL);
3051 	if (!id)
3052 		return 0;
3053 
3054 	c.identify.opcode = nvme_admin_identify;
3055 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
3056 	c.identify.csi = NVME_CSI_NVM;
3057 
3058 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3059 	if (ret)
3060 		goto free_data;
3061 
3062 	if (id->dmrl)
3063 		ctrl->max_discard_segments = id->dmrl;
3064 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3065 	if (id->wzsl)
3066 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3067 
3068 free_data:
3069 	kfree(id);
3070 	return ret;
3071 }
3072 
3073 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3074 {
3075 	struct nvme_id_ctrl *id;
3076 	u32 max_hw_sectors;
3077 	bool prev_apst_enabled;
3078 	int ret;
3079 
3080 	ret = nvme_identify_ctrl(ctrl, &id);
3081 	if (ret) {
3082 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3083 		return -EIO;
3084 	}
3085 
3086 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3087 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3088 		if (ret < 0)
3089 			goto out_free;
3090 	}
3091 
3092 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3093 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3094 
3095 	if (!ctrl->identified) {
3096 		unsigned int i;
3097 
3098 		ret = nvme_init_subsystem(ctrl, id);
3099 		if (ret)
3100 			goto out_free;
3101 
3102 		/*
3103 		 * Check for quirks.  Quirk can depend on firmware version,
3104 		 * so, in principle, the set of quirks present can change
3105 		 * across a reset.  As a possible future enhancement, we
3106 		 * could re-scan for quirks every time we reinitialize
3107 		 * the device, but we'd have to make sure that the driver
3108 		 * behaves intelligently if the quirks change.
3109 		 */
3110 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3111 			if (quirk_matches(id, &core_quirks[i]))
3112 				ctrl->quirks |= core_quirks[i].quirks;
3113 		}
3114 	}
3115 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3116 	       sizeof(ctrl->subsys->firmware_rev));
3117 
3118 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3119 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3120 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3121 	}
3122 
3123 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3124 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3125 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3126 
3127 	ctrl->oacs = le16_to_cpu(id->oacs);
3128 	ctrl->oncs = le16_to_cpu(id->oncs);
3129 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3130 	ctrl->oaes = le32_to_cpu(id->oaes);
3131 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3132 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3133 
3134 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3135 	ctrl->vwc = id->vwc;
3136 	if (id->mdts)
3137 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3138 	else
3139 		max_hw_sectors = UINT_MAX;
3140 	ctrl->max_hw_sectors =
3141 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3142 
3143 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3144 	ctrl->sgls = le32_to_cpu(id->sgls);
3145 	ctrl->kas = le16_to_cpu(id->kas);
3146 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3147 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3148 
3149 	ctrl->cntrltype = id->cntrltype;
3150 	ctrl->dctype = id->dctype;
3151 
3152 	if (id->rtd3e) {
3153 		/* us -> s */
3154 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3155 
3156 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3157 						 shutdown_timeout, 60);
3158 
3159 		if (ctrl->shutdown_timeout != shutdown_timeout)
3160 			dev_info(ctrl->device,
3161 				 "Shutdown timeout set to %u seconds\n",
3162 				 ctrl->shutdown_timeout);
3163 	} else
3164 		ctrl->shutdown_timeout = shutdown_timeout;
3165 
3166 	ctrl->npss = id->npss;
3167 	ctrl->apsta = id->apsta;
3168 	prev_apst_enabled = ctrl->apst_enabled;
3169 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3170 		if (force_apst && id->apsta) {
3171 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3172 			ctrl->apst_enabled = true;
3173 		} else {
3174 			ctrl->apst_enabled = false;
3175 		}
3176 	} else {
3177 		ctrl->apst_enabled = id->apsta;
3178 	}
3179 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3180 
3181 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3182 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3183 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3184 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3185 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3186 
3187 		/*
3188 		 * In fabrics we need to verify the cntlid matches the
3189 		 * admin connect
3190 		 */
3191 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3192 			dev_err(ctrl->device,
3193 				"Mismatching cntlid: Connect %u vs Identify "
3194 				"%u, rejecting\n",
3195 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3196 			ret = -EINVAL;
3197 			goto out_free;
3198 		}
3199 
3200 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3201 			dev_err(ctrl->device,
3202 				"keep-alive support is mandatory for fabrics\n");
3203 			ret = -EINVAL;
3204 			goto out_free;
3205 		}
3206 	} else {
3207 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3208 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3209 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3210 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3211 	}
3212 
3213 	ret = nvme_mpath_init_identify(ctrl, id);
3214 	if (ret < 0)
3215 		goto out_free;
3216 
3217 	if (ctrl->apst_enabled && !prev_apst_enabled)
3218 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3219 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3220 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3221 
3222 out_free:
3223 	kfree(id);
3224 	return ret;
3225 }
3226 
3227 /*
3228  * Initialize the cached copies of the Identify data and various controller
3229  * register in our nvme_ctrl structure.  This should be called as soon as
3230  * the admin queue is fully up and running.
3231  */
3232 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3233 {
3234 	int ret;
3235 
3236 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3237 	if (ret) {
3238 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3239 		return ret;
3240 	}
3241 
3242 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3243 
3244 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3245 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3246 
3247 	ret = nvme_init_identify(ctrl);
3248 	if (ret)
3249 		return ret;
3250 
3251 	ret = nvme_configure_apst(ctrl);
3252 	if (ret < 0)
3253 		return ret;
3254 
3255 	ret = nvme_configure_timestamp(ctrl);
3256 	if (ret < 0)
3257 		return ret;
3258 
3259 	ret = nvme_configure_host_options(ctrl);
3260 	if (ret < 0)
3261 		return ret;
3262 
3263 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3264 		/*
3265 		 * Do not return errors unless we are in a controller reset,
3266 		 * the controller works perfectly fine without hwmon.
3267 		 */
3268 		ret = nvme_hwmon_init(ctrl);
3269 		if (ret == -EINTR)
3270 			return ret;
3271 	}
3272 
3273 	ctrl->identified = true;
3274 
3275 	return 0;
3276 }
3277 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3278 
3279 static int nvme_dev_open(struct inode *inode, struct file *file)
3280 {
3281 	struct nvme_ctrl *ctrl =
3282 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3283 
3284 	switch (ctrl->state) {
3285 	case NVME_CTRL_LIVE:
3286 		break;
3287 	default:
3288 		return -EWOULDBLOCK;
3289 	}
3290 
3291 	nvme_get_ctrl(ctrl);
3292 	if (!try_module_get(ctrl->ops->module)) {
3293 		nvme_put_ctrl(ctrl);
3294 		return -EINVAL;
3295 	}
3296 
3297 	file->private_data = ctrl;
3298 	return 0;
3299 }
3300 
3301 static int nvme_dev_release(struct inode *inode, struct file *file)
3302 {
3303 	struct nvme_ctrl *ctrl =
3304 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3305 
3306 	module_put(ctrl->ops->module);
3307 	nvme_put_ctrl(ctrl);
3308 	return 0;
3309 }
3310 
3311 static const struct file_operations nvme_dev_fops = {
3312 	.owner		= THIS_MODULE,
3313 	.open		= nvme_dev_open,
3314 	.release	= nvme_dev_release,
3315 	.unlocked_ioctl	= nvme_dev_ioctl,
3316 	.compat_ioctl	= compat_ptr_ioctl,
3317 	.uring_cmd	= nvme_dev_uring_cmd,
3318 };
3319 
3320 static ssize_t nvme_sysfs_reset(struct device *dev,
3321 				struct device_attribute *attr, const char *buf,
3322 				size_t count)
3323 {
3324 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3325 	int ret;
3326 
3327 	ret = nvme_reset_ctrl_sync(ctrl);
3328 	if (ret < 0)
3329 		return ret;
3330 	return count;
3331 }
3332 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3333 
3334 static ssize_t nvme_sysfs_rescan(struct device *dev,
3335 				struct device_attribute *attr, const char *buf,
3336 				size_t count)
3337 {
3338 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3339 
3340 	nvme_queue_scan(ctrl);
3341 	return count;
3342 }
3343 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3344 
3345 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3346 {
3347 	struct gendisk *disk = dev_to_disk(dev);
3348 
3349 	if (disk->fops == &nvme_bdev_ops)
3350 		return nvme_get_ns_from_dev(dev)->head;
3351 	else
3352 		return disk->private_data;
3353 }
3354 
3355 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3356 		char *buf)
3357 {
3358 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3359 	struct nvme_ns_ids *ids = &head->ids;
3360 	struct nvme_subsystem *subsys = head->subsys;
3361 	int serial_len = sizeof(subsys->serial);
3362 	int model_len = sizeof(subsys->model);
3363 
3364 	if (!uuid_is_null(&ids->uuid))
3365 		return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3366 
3367 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3368 		return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3369 
3370 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3371 		return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3372 
3373 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3374 				  subsys->serial[serial_len - 1] == '\0'))
3375 		serial_len--;
3376 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3377 				 subsys->model[model_len - 1] == '\0'))
3378 		model_len--;
3379 
3380 	return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3381 		serial_len, subsys->serial, model_len, subsys->model,
3382 		head->ns_id);
3383 }
3384 static DEVICE_ATTR_RO(wwid);
3385 
3386 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3387 		char *buf)
3388 {
3389 	return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3390 }
3391 static DEVICE_ATTR_RO(nguid);
3392 
3393 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3394 		char *buf)
3395 {
3396 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3397 
3398 	/* For backward compatibility expose the NGUID to userspace if
3399 	 * we have no UUID set
3400 	 */
3401 	if (uuid_is_null(&ids->uuid)) {
3402 		dev_warn_ratelimited(dev,
3403 			"No UUID available providing old NGUID\n");
3404 		return sysfs_emit(buf, "%pU\n", ids->nguid);
3405 	}
3406 	return sysfs_emit(buf, "%pU\n", &ids->uuid);
3407 }
3408 static DEVICE_ATTR_RO(uuid);
3409 
3410 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3411 		char *buf)
3412 {
3413 	return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3414 }
3415 static DEVICE_ATTR_RO(eui);
3416 
3417 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3418 		char *buf)
3419 {
3420 	return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3421 }
3422 static DEVICE_ATTR_RO(nsid);
3423 
3424 static struct attribute *nvme_ns_id_attrs[] = {
3425 	&dev_attr_wwid.attr,
3426 	&dev_attr_uuid.attr,
3427 	&dev_attr_nguid.attr,
3428 	&dev_attr_eui.attr,
3429 	&dev_attr_nsid.attr,
3430 #ifdef CONFIG_NVME_MULTIPATH
3431 	&dev_attr_ana_grpid.attr,
3432 	&dev_attr_ana_state.attr,
3433 #endif
3434 	NULL,
3435 };
3436 
3437 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3438 		struct attribute *a, int n)
3439 {
3440 	struct device *dev = container_of(kobj, struct device, kobj);
3441 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3442 
3443 	if (a == &dev_attr_uuid.attr) {
3444 		if (uuid_is_null(&ids->uuid) &&
3445 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3446 			return 0;
3447 	}
3448 	if (a == &dev_attr_nguid.attr) {
3449 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3450 			return 0;
3451 	}
3452 	if (a == &dev_attr_eui.attr) {
3453 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3454 			return 0;
3455 	}
3456 #ifdef CONFIG_NVME_MULTIPATH
3457 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3458 		if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3459 			return 0;
3460 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3461 			return 0;
3462 	}
3463 #endif
3464 	return a->mode;
3465 }
3466 
3467 static const struct attribute_group nvme_ns_id_attr_group = {
3468 	.attrs		= nvme_ns_id_attrs,
3469 	.is_visible	= nvme_ns_id_attrs_are_visible,
3470 };
3471 
3472 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3473 	&nvme_ns_id_attr_group,
3474 	NULL,
3475 };
3476 
3477 #define nvme_show_str_function(field)						\
3478 static ssize_t  field##_show(struct device *dev,				\
3479 			    struct device_attribute *attr, char *buf)		\
3480 {										\
3481         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3482         return sysfs_emit(buf, "%.*s\n",					\
3483 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3484 }										\
3485 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3486 
3487 nvme_show_str_function(model);
3488 nvme_show_str_function(serial);
3489 nvme_show_str_function(firmware_rev);
3490 
3491 #define nvme_show_int_function(field)						\
3492 static ssize_t  field##_show(struct device *dev,				\
3493 			    struct device_attribute *attr, char *buf)		\
3494 {										\
3495         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3496         return sysfs_emit(buf, "%d\n", ctrl->field);				\
3497 }										\
3498 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3499 
3500 nvme_show_int_function(cntlid);
3501 nvme_show_int_function(numa_node);
3502 nvme_show_int_function(queue_count);
3503 nvme_show_int_function(sqsize);
3504 nvme_show_int_function(kato);
3505 
3506 static ssize_t nvme_sysfs_delete(struct device *dev,
3507 				struct device_attribute *attr, const char *buf,
3508 				size_t count)
3509 {
3510 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3511 
3512 	if (device_remove_file_self(dev, attr))
3513 		nvme_delete_ctrl_sync(ctrl);
3514 	return count;
3515 }
3516 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3517 
3518 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3519 					 struct device_attribute *attr,
3520 					 char *buf)
3521 {
3522 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3523 
3524 	return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3525 }
3526 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3527 
3528 static ssize_t nvme_sysfs_show_state(struct device *dev,
3529 				     struct device_attribute *attr,
3530 				     char *buf)
3531 {
3532 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3533 	static const char *const state_name[] = {
3534 		[NVME_CTRL_NEW]		= "new",
3535 		[NVME_CTRL_LIVE]	= "live",
3536 		[NVME_CTRL_RESETTING]	= "resetting",
3537 		[NVME_CTRL_CONNECTING]	= "connecting",
3538 		[NVME_CTRL_DELETING]	= "deleting",
3539 		[NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3540 		[NVME_CTRL_DEAD]	= "dead",
3541 	};
3542 
3543 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3544 	    state_name[ctrl->state])
3545 		return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3546 
3547 	return sysfs_emit(buf, "unknown state\n");
3548 }
3549 
3550 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3551 
3552 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3553 					 struct device_attribute *attr,
3554 					 char *buf)
3555 {
3556 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3557 
3558 	return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3559 }
3560 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3561 
3562 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3563 					struct device_attribute *attr,
3564 					char *buf)
3565 {
3566 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3567 
3568 	return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3569 }
3570 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3571 
3572 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3573 					struct device_attribute *attr,
3574 					char *buf)
3575 {
3576 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3577 
3578 	return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3579 }
3580 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3581 
3582 static ssize_t nvme_sysfs_show_address(struct device *dev,
3583 					 struct device_attribute *attr,
3584 					 char *buf)
3585 {
3586 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3587 
3588 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3589 }
3590 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3591 
3592 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3593 		struct device_attribute *attr, char *buf)
3594 {
3595 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3596 	struct nvmf_ctrl_options *opts = ctrl->opts;
3597 
3598 	if (ctrl->opts->max_reconnects == -1)
3599 		return sysfs_emit(buf, "off\n");
3600 	return sysfs_emit(buf, "%d\n",
3601 			  opts->max_reconnects * opts->reconnect_delay);
3602 }
3603 
3604 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3605 		struct device_attribute *attr, const char *buf, size_t count)
3606 {
3607 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3608 	struct nvmf_ctrl_options *opts = ctrl->opts;
3609 	int ctrl_loss_tmo, err;
3610 
3611 	err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3612 	if (err)
3613 		return -EINVAL;
3614 
3615 	if (ctrl_loss_tmo < 0)
3616 		opts->max_reconnects = -1;
3617 	else
3618 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3619 						opts->reconnect_delay);
3620 	return count;
3621 }
3622 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3623 	nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3624 
3625 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3626 		struct device_attribute *attr, char *buf)
3627 {
3628 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3629 
3630 	if (ctrl->opts->reconnect_delay == -1)
3631 		return sysfs_emit(buf, "off\n");
3632 	return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3633 }
3634 
3635 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3636 		struct device_attribute *attr, const char *buf, size_t count)
3637 {
3638 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3639 	unsigned int v;
3640 	int err;
3641 
3642 	err = kstrtou32(buf, 10, &v);
3643 	if (err)
3644 		return err;
3645 
3646 	ctrl->opts->reconnect_delay = v;
3647 	return count;
3648 }
3649 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3650 	nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3651 
3652 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3653 		struct device_attribute *attr, char *buf)
3654 {
3655 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3656 
3657 	if (ctrl->opts->fast_io_fail_tmo == -1)
3658 		return sysfs_emit(buf, "off\n");
3659 	return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3660 }
3661 
3662 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3663 		struct device_attribute *attr, const char *buf, size_t count)
3664 {
3665 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3666 	struct nvmf_ctrl_options *opts = ctrl->opts;
3667 	int fast_io_fail_tmo, err;
3668 
3669 	err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3670 	if (err)
3671 		return -EINVAL;
3672 
3673 	if (fast_io_fail_tmo < 0)
3674 		opts->fast_io_fail_tmo = -1;
3675 	else
3676 		opts->fast_io_fail_tmo = fast_io_fail_tmo;
3677 	return count;
3678 }
3679 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3680 	nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3681 
3682 static ssize_t cntrltype_show(struct device *dev,
3683 			      struct device_attribute *attr, char *buf)
3684 {
3685 	static const char * const type[] = {
3686 		[NVME_CTRL_IO] = "io\n",
3687 		[NVME_CTRL_DISC] = "discovery\n",
3688 		[NVME_CTRL_ADMIN] = "admin\n",
3689 	};
3690 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3691 
3692 	if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3693 		return sysfs_emit(buf, "reserved\n");
3694 
3695 	return sysfs_emit(buf, type[ctrl->cntrltype]);
3696 }
3697 static DEVICE_ATTR_RO(cntrltype);
3698 
3699 static ssize_t dctype_show(struct device *dev,
3700 			   struct device_attribute *attr, char *buf)
3701 {
3702 	static const char * const type[] = {
3703 		[NVME_DCTYPE_NOT_REPORTED] = "none\n",
3704 		[NVME_DCTYPE_DDC] = "ddc\n",
3705 		[NVME_DCTYPE_CDC] = "cdc\n",
3706 	};
3707 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3708 
3709 	if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3710 		return sysfs_emit(buf, "reserved\n");
3711 
3712 	return sysfs_emit(buf, type[ctrl->dctype]);
3713 }
3714 static DEVICE_ATTR_RO(dctype);
3715 
3716 #ifdef CONFIG_NVME_AUTH
3717 static ssize_t nvme_ctrl_dhchap_secret_show(struct device *dev,
3718 		struct device_attribute *attr, char *buf)
3719 {
3720 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3721 	struct nvmf_ctrl_options *opts = ctrl->opts;
3722 
3723 	if (!opts->dhchap_secret)
3724 		return sysfs_emit(buf, "none\n");
3725 	return sysfs_emit(buf, "%s\n", opts->dhchap_secret);
3726 }
3727 
3728 static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
3729 		struct device_attribute *attr, const char *buf, size_t count)
3730 {
3731 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3732 	struct nvmf_ctrl_options *opts = ctrl->opts;
3733 	char *dhchap_secret;
3734 
3735 	if (!ctrl->opts->dhchap_secret)
3736 		return -EINVAL;
3737 	if (count < 7)
3738 		return -EINVAL;
3739 	if (memcmp(buf, "DHHC-1:", 7))
3740 		return -EINVAL;
3741 
3742 	dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3743 	if (!dhchap_secret)
3744 		return -ENOMEM;
3745 	memcpy(dhchap_secret, buf, count);
3746 	nvme_auth_stop(ctrl);
3747 	if (strcmp(dhchap_secret, opts->dhchap_secret)) {
3748 		int ret;
3749 
3750 		ret = nvme_auth_generate_key(dhchap_secret, &ctrl->host_key);
3751 		if (ret)
3752 			return ret;
3753 		kfree(opts->dhchap_secret);
3754 		opts->dhchap_secret = dhchap_secret;
3755 		/* Key has changed; re-authentication with new key */
3756 		nvme_auth_reset(ctrl);
3757 	}
3758 	/* Start re-authentication */
3759 	dev_info(ctrl->device, "re-authenticating controller\n");
3760 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3761 
3762 	return count;
3763 }
3764 static DEVICE_ATTR(dhchap_secret, S_IRUGO | S_IWUSR,
3765 	nvme_ctrl_dhchap_secret_show, nvme_ctrl_dhchap_secret_store);
3766 
3767 static ssize_t nvme_ctrl_dhchap_ctrl_secret_show(struct device *dev,
3768 		struct device_attribute *attr, char *buf)
3769 {
3770 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3771 	struct nvmf_ctrl_options *opts = ctrl->opts;
3772 
3773 	if (!opts->dhchap_ctrl_secret)
3774 		return sysfs_emit(buf, "none\n");
3775 	return sysfs_emit(buf, "%s\n", opts->dhchap_ctrl_secret);
3776 }
3777 
3778 static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
3779 		struct device_attribute *attr, const char *buf, size_t count)
3780 {
3781 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3782 	struct nvmf_ctrl_options *opts = ctrl->opts;
3783 	char *dhchap_secret;
3784 
3785 	if (!ctrl->opts->dhchap_ctrl_secret)
3786 		return -EINVAL;
3787 	if (count < 7)
3788 		return -EINVAL;
3789 	if (memcmp(buf, "DHHC-1:", 7))
3790 		return -EINVAL;
3791 
3792 	dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3793 	if (!dhchap_secret)
3794 		return -ENOMEM;
3795 	memcpy(dhchap_secret, buf, count);
3796 	nvme_auth_stop(ctrl);
3797 	if (strcmp(dhchap_secret, opts->dhchap_ctrl_secret)) {
3798 		int ret;
3799 
3800 		ret = nvme_auth_generate_key(dhchap_secret, &ctrl->ctrl_key);
3801 		if (ret)
3802 			return ret;
3803 		kfree(opts->dhchap_ctrl_secret);
3804 		opts->dhchap_ctrl_secret = dhchap_secret;
3805 		/* Key has changed; re-authentication with new key */
3806 		nvme_auth_reset(ctrl);
3807 	}
3808 	/* Start re-authentication */
3809 	dev_info(ctrl->device, "re-authenticating controller\n");
3810 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3811 
3812 	return count;
3813 }
3814 static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR,
3815 	nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);
3816 #endif
3817 
3818 static struct attribute *nvme_dev_attrs[] = {
3819 	&dev_attr_reset_controller.attr,
3820 	&dev_attr_rescan_controller.attr,
3821 	&dev_attr_model.attr,
3822 	&dev_attr_serial.attr,
3823 	&dev_attr_firmware_rev.attr,
3824 	&dev_attr_cntlid.attr,
3825 	&dev_attr_delete_controller.attr,
3826 	&dev_attr_transport.attr,
3827 	&dev_attr_subsysnqn.attr,
3828 	&dev_attr_address.attr,
3829 	&dev_attr_state.attr,
3830 	&dev_attr_numa_node.attr,
3831 	&dev_attr_queue_count.attr,
3832 	&dev_attr_sqsize.attr,
3833 	&dev_attr_hostnqn.attr,
3834 	&dev_attr_hostid.attr,
3835 	&dev_attr_ctrl_loss_tmo.attr,
3836 	&dev_attr_reconnect_delay.attr,
3837 	&dev_attr_fast_io_fail_tmo.attr,
3838 	&dev_attr_kato.attr,
3839 	&dev_attr_cntrltype.attr,
3840 	&dev_attr_dctype.attr,
3841 #ifdef CONFIG_NVME_AUTH
3842 	&dev_attr_dhchap_secret.attr,
3843 	&dev_attr_dhchap_ctrl_secret.attr,
3844 #endif
3845 	NULL
3846 };
3847 
3848 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3849 		struct attribute *a, int n)
3850 {
3851 	struct device *dev = container_of(kobj, struct device, kobj);
3852 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3853 
3854 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3855 		return 0;
3856 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3857 		return 0;
3858 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3859 		return 0;
3860 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3861 		return 0;
3862 	if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3863 		return 0;
3864 	if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3865 		return 0;
3866 	if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3867 		return 0;
3868 #ifdef CONFIG_NVME_AUTH
3869 	if (a == &dev_attr_dhchap_secret.attr && !ctrl->opts)
3870 		return 0;
3871 	if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)
3872 		return 0;
3873 #endif
3874 
3875 	return a->mode;
3876 }
3877 
3878 static const struct attribute_group nvme_dev_attrs_group = {
3879 	.attrs		= nvme_dev_attrs,
3880 	.is_visible	= nvme_dev_attrs_are_visible,
3881 };
3882 
3883 static const struct attribute_group *nvme_dev_attr_groups[] = {
3884 	&nvme_dev_attrs_group,
3885 	NULL,
3886 };
3887 
3888 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3889 		unsigned nsid)
3890 {
3891 	struct nvme_ns_head *h;
3892 
3893 	lockdep_assert_held(&ctrl->subsys->lock);
3894 
3895 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3896 		/*
3897 		 * Private namespaces can share NSIDs under some conditions.
3898 		 * In that case we can't use the same ns_head for namespaces
3899 		 * with the same NSID.
3900 		 */
3901 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3902 			continue;
3903 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3904 			return h;
3905 	}
3906 
3907 	return NULL;
3908 }
3909 
3910 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3911 		struct nvme_ns_ids *ids)
3912 {
3913 	bool has_uuid = !uuid_is_null(&ids->uuid);
3914 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3915 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3916 	struct nvme_ns_head *h;
3917 
3918 	lockdep_assert_held(&subsys->lock);
3919 
3920 	list_for_each_entry(h, &subsys->nsheads, entry) {
3921 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3922 			return -EINVAL;
3923 		if (has_nguid &&
3924 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3925 			return -EINVAL;
3926 		if (has_eui64 &&
3927 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3928 			return -EINVAL;
3929 	}
3930 
3931 	return 0;
3932 }
3933 
3934 static void nvme_cdev_rel(struct device *dev)
3935 {
3936 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3937 }
3938 
3939 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3940 {
3941 	cdev_device_del(cdev, cdev_device);
3942 	put_device(cdev_device);
3943 }
3944 
3945 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3946 		const struct file_operations *fops, struct module *owner)
3947 {
3948 	int minor, ret;
3949 
3950 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3951 	if (minor < 0)
3952 		return minor;
3953 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3954 	cdev_device->class = nvme_ns_chr_class;
3955 	cdev_device->release = nvme_cdev_rel;
3956 	device_initialize(cdev_device);
3957 	cdev_init(cdev, fops);
3958 	cdev->owner = owner;
3959 	ret = cdev_device_add(cdev, cdev_device);
3960 	if (ret)
3961 		put_device(cdev_device);
3962 
3963 	return ret;
3964 }
3965 
3966 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3967 {
3968 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3969 }
3970 
3971 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3972 {
3973 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3974 	return 0;
3975 }
3976 
3977 static const struct file_operations nvme_ns_chr_fops = {
3978 	.owner		= THIS_MODULE,
3979 	.open		= nvme_ns_chr_open,
3980 	.release	= nvme_ns_chr_release,
3981 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3982 	.compat_ioctl	= compat_ptr_ioctl,
3983 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3984 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3985 };
3986 
3987 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3988 {
3989 	int ret;
3990 
3991 	ns->cdev_device.parent = ns->ctrl->device;
3992 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3993 			   ns->ctrl->instance, ns->head->instance);
3994 	if (ret)
3995 		return ret;
3996 
3997 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3998 			     ns->ctrl->ops->module);
3999 }
4000 
4001 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
4002 		struct nvme_ns_info *info)
4003 {
4004 	struct nvme_ns_head *head;
4005 	size_t size = sizeof(*head);
4006 	int ret = -ENOMEM;
4007 
4008 #ifdef CONFIG_NVME_MULTIPATH
4009 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
4010 #endif
4011 
4012 	head = kzalloc(size, GFP_KERNEL);
4013 	if (!head)
4014 		goto out;
4015 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
4016 	if (ret < 0)
4017 		goto out_free_head;
4018 	head->instance = ret;
4019 	INIT_LIST_HEAD(&head->list);
4020 	ret = init_srcu_struct(&head->srcu);
4021 	if (ret)
4022 		goto out_ida_remove;
4023 	head->subsys = ctrl->subsys;
4024 	head->ns_id = info->nsid;
4025 	head->ids = info->ids;
4026 	head->shared = info->is_shared;
4027 	kref_init(&head->ref);
4028 
4029 	if (head->ids.csi) {
4030 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
4031 		if (ret)
4032 			goto out_cleanup_srcu;
4033 	} else
4034 		head->effects = ctrl->effects;
4035 
4036 	ret = nvme_mpath_alloc_disk(ctrl, head);
4037 	if (ret)
4038 		goto out_cleanup_srcu;
4039 
4040 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
4041 
4042 	kref_get(&ctrl->subsys->ref);
4043 
4044 	return head;
4045 out_cleanup_srcu:
4046 	cleanup_srcu_struct(&head->srcu);
4047 out_ida_remove:
4048 	ida_free(&ctrl->subsys->ns_ida, head->instance);
4049 out_free_head:
4050 	kfree(head);
4051 out:
4052 	if (ret > 0)
4053 		ret = blk_status_to_errno(nvme_error_status(ret));
4054 	return ERR_PTR(ret);
4055 }
4056 
4057 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
4058 		struct nvme_ns_ids *ids)
4059 {
4060 	struct nvme_subsystem *s;
4061 	int ret = 0;
4062 
4063 	/*
4064 	 * Note that this check is racy as we try to avoid holding the global
4065 	 * lock over the whole ns_head creation.  But it is only intended as
4066 	 * a sanity check anyway.
4067 	 */
4068 	mutex_lock(&nvme_subsystems_lock);
4069 	list_for_each_entry(s, &nvme_subsystems, entry) {
4070 		if (s == this)
4071 			continue;
4072 		mutex_lock(&s->lock);
4073 		ret = nvme_subsys_check_duplicate_ids(s, ids);
4074 		mutex_unlock(&s->lock);
4075 		if (ret)
4076 			break;
4077 	}
4078 	mutex_unlock(&nvme_subsystems_lock);
4079 
4080 	return ret;
4081 }
4082 
4083 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
4084 {
4085 	struct nvme_ctrl *ctrl = ns->ctrl;
4086 	struct nvme_ns_head *head = NULL;
4087 	int ret;
4088 
4089 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
4090 	if (ret) {
4091 		dev_err(ctrl->device,
4092 			"globally duplicate IDs for nsid %d\n", info->nsid);
4093 		nvme_print_device_info(ctrl);
4094 		return ret;
4095 	}
4096 
4097 	mutex_lock(&ctrl->subsys->lock);
4098 	head = nvme_find_ns_head(ctrl, info->nsid);
4099 	if (!head) {
4100 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4101 		if (ret) {
4102 			dev_err(ctrl->device,
4103 				"duplicate IDs in subsystem for nsid %d\n",
4104 				info->nsid);
4105 			goto out_unlock;
4106 		}
4107 		head = nvme_alloc_ns_head(ctrl, info);
4108 		if (IS_ERR(head)) {
4109 			ret = PTR_ERR(head);
4110 			goto out_unlock;
4111 		}
4112 	} else {
4113 		ret = -EINVAL;
4114 		if (!info->is_shared || !head->shared) {
4115 			dev_err(ctrl->device,
4116 				"Duplicate unshared namespace %d\n",
4117 				info->nsid);
4118 			goto out_put_ns_head;
4119 		}
4120 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4121 			dev_err(ctrl->device,
4122 				"IDs don't match for shared namespace %d\n",
4123 					info->nsid);
4124 			goto out_put_ns_head;
4125 		}
4126 
4127 		if (!multipath && !list_empty(&head->list)) {
4128 			dev_warn(ctrl->device,
4129 				"Found shared namespace %d, but multipathing not supported.\n",
4130 				info->nsid);
4131 			dev_warn_once(ctrl->device,
4132 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
4133 		}
4134 	}
4135 
4136 	list_add_tail_rcu(&ns->siblings, &head->list);
4137 	ns->head = head;
4138 	mutex_unlock(&ctrl->subsys->lock);
4139 	return 0;
4140 
4141 out_put_ns_head:
4142 	nvme_put_ns_head(head);
4143 out_unlock:
4144 	mutex_unlock(&ctrl->subsys->lock);
4145 	return ret;
4146 }
4147 
4148 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4149 {
4150 	struct nvme_ns *ns, *ret = NULL;
4151 
4152 	down_read(&ctrl->namespaces_rwsem);
4153 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4154 		if (ns->head->ns_id == nsid) {
4155 			if (!nvme_get_ns(ns))
4156 				continue;
4157 			ret = ns;
4158 			break;
4159 		}
4160 		if (ns->head->ns_id > nsid)
4161 			break;
4162 	}
4163 	up_read(&ctrl->namespaces_rwsem);
4164 	return ret;
4165 }
4166 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
4167 
4168 /*
4169  * Add the namespace to the controller list while keeping the list ordered.
4170  */
4171 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4172 {
4173 	struct nvme_ns *tmp;
4174 
4175 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4176 		if (tmp->head->ns_id < ns->head->ns_id) {
4177 			list_add(&ns->list, &tmp->list);
4178 			return;
4179 		}
4180 	}
4181 	list_add(&ns->list, &ns->ctrl->namespaces);
4182 }
4183 
4184 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4185 {
4186 	struct nvme_ns *ns;
4187 	struct gendisk *disk;
4188 	int node = ctrl->numa_node;
4189 
4190 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4191 	if (!ns)
4192 		return;
4193 
4194 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
4195 	if (IS_ERR(disk))
4196 		goto out_free_ns;
4197 	disk->fops = &nvme_bdev_ops;
4198 	disk->private_data = ns;
4199 
4200 	ns->disk = disk;
4201 	ns->queue = disk->queue;
4202 
4203 	if (ctrl->opts && ctrl->opts->data_digest)
4204 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4205 
4206 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4207 	if (ctrl->ops->supports_pci_p2pdma &&
4208 	    ctrl->ops->supports_pci_p2pdma(ctrl))
4209 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4210 
4211 	ns->ctrl = ctrl;
4212 	kref_init(&ns->kref);
4213 
4214 	if (nvme_init_ns_head(ns, info))
4215 		goto out_cleanup_disk;
4216 
4217 	/*
4218 	 * If multipathing is enabled, the device name for all disks and not
4219 	 * just those that represent shared namespaces needs to be based on the
4220 	 * subsystem instance.  Using the controller instance for private
4221 	 * namespaces could lead to naming collisions between shared and private
4222 	 * namespaces if they don't use a common numbering scheme.
4223 	 *
4224 	 * If multipathing is not enabled, disk names must use the controller
4225 	 * instance as shared namespaces will show up as multiple block
4226 	 * devices.
4227 	 */
4228 	if (ns->head->disk) {
4229 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4230 			ctrl->instance, ns->head->instance);
4231 		disk->flags |= GENHD_FL_HIDDEN;
4232 	} else if (multipath) {
4233 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4234 			ns->head->instance);
4235 	} else {
4236 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4237 			ns->head->instance);
4238 	}
4239 
4240 	if (nvme_update_ns_info(ns, info))
4241 		goto out_unlink_ns;
4242 
4243 	down_write(&ctrl->namespaces_rwsem);
4244 	nvme_ns_add_to_ctrl_list(ns);
4245 	up_write(&ctrl->namespaces_rwsem);
4246 	nvme_get_ctrl(ctrl);
4247 
4248 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4249 		goto out_cleanup_ns_from_list;
4250 
4251 	if (!nvme_ns_head_multipath(ns->head))
4252 		nvme_add_ns_cdev(ns);
4253 
4254 	nvme_mpath_add_disk(ns, info->anagrpid);
4255 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4256 
4257 	return;
4258 
4259  out_cleanup_ns_from_list:
4260 	nvme_put_ctrl(ctrl);
4261 	down_write(&ctrl->namespaces_rwsem);
4262 	list_del_init(&ns->list);
4263 	up_write(&ctrl->namespaces_rwsem);
4264  out_unlink_ns:
4265 	mutex_lock(&ctrl->subsys->lock);
4266 	list_del_rcu(&ns->siblings);
4267 	if (list_empty(&ns->head->list))
4268 		list_del_init(&ns->head->entry);
4269 	mutex_unlock(&ctrl->subsys->lock);
4270 	nvme_put_ns_head(ns->head);
4271  out_cleanup_disk:
4272 	put_disk(disk);
4273  out_free_ns:
4274 	kfree(ns);
4275 }
4276 
4277 static void nvme_ns_remove(struct nvme_ns *ns)
4278 {
4279 	bool last_path = false;
4280 
4281 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4282 		return;
4283 
4284 	clear_bit(NVME_NS_READY, &ns->flags);
4285 	set_capacity(ns->disk, 0);
4286 	nvme_fault_inject_fini(&ns->fault_inject);
4287 
4288 	/*
4289 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4290 	 * this ns going back into current_path.
4291 	 */
4292 	synchronize_srcu(&ns->head->srcu);
4293 
4294 	/* wait for concurrent submissions */
4295 	if (nvme_mpath_clear_current_path(ns))
4296 		synchronize_srcu(&ns->head->srcu);
4297 
4298 	mutex_lock(&ns->ctrl->subsys->lock);
4299 	list_del_rcu(&ns->siblings);
4300 	if (list_empty(&ns->head->list)) {
4301 		list_del_init(&ns->head->entry);
4302 		last_path = true;
4303 	}
4304 	mutex_unlock(&ns->ctrl->subsys->lock);
4305 
4306 	/* guarantee not available in head->list */
4307 	synchronize_srcu(&ns->head->srcu);
4308 
4309 	if (!nvme_ns_head_multipath(ns->head))
4310 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4311 	del_gendisk(ns->disk);
4312 
4313 	down_write(&ns->ctrl->namespaces_rwsem);
4314 	list_del_init(&ns->list);
4315 	up_write(&ns->ctrl->namespaces_rwsem);
4316 
4317 	if (last_path)
4318 		nvme_mpath_shutdown_disk(ns->head);
4319 	nvme_put_ns(ns);
4320 }
4321 
4322 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4323 {
4324 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4325 
4326 	if (ns) {
4327 		nvme_ns_remove(ns);
4328 		nvme_put_ns(ns);
4329 	}
4330 }
4331 
4332 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4333 {
4334 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4335 
4336 	if (test_bit(NVME_NS_DEAD, &ns->flags))
4337 		goto out;
4338 
4339 	ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4340 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4341 		dev_err(ns->ctrl->device,
4342 			"identifiers changed for nsid %d\n", ns->head->ns_id);
4343 		goto out;
4344 	}
4345 
4346 	ret = nvme_update_ns_info(ns, info);
4347 out:
4348 	/*
4349 	 * Only remove the namespace if we got a fatal error back from the
4350 	 * device, otherwise ignore the error and just move on.
4351 	 *
4352 	 * TODO: we should probably schedule a delayed retry here.
4353 	 */
4354 	if (ret > 0 && (ret & NVME_SC_DNR))
4355 		nvme_ns_remove(ns);
4356 }
4357 
4358 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4359 {
4360 	struct nvme_ns_info info = { .nsid = nsid };
4361 	struct nvme_ns *ns;
4362 
4363 	if (nvme_identify_ns_descs(ctrl, &info))
4364 		return;
4365 
4366 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4367 		dev_warn(ctrl->device,
4368 			"command set not reported for nsid: %d\n", nsid);
4369 		return;
4370 	}
4371 
4372 	/*
4373 	 * If available try to use the Command Set Idependent Identify Namespace
4374 	 * data structure to find all the generic information that is needed to
4375 	 * set up a namespace.  If not fall back to the legacy version.
4376 	 */
4377 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4378 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS)) {
4379 		if (nvme_ns_info_from_id_cs_indep(ctrl, &info))
4380 			return;
4381 	} else {
4382 		if (nvme_ns_info_from_identify(ctrl, &info))
4383 			return;
4384 	}
4385 
4386 	/*
4387 	 * Ignore the namespace if it is not ready. We will get an AEN once it
4388 	 * becomes ready and restart the scan.
4389 	 */
4390 	if (!info.is_ready)
4391 		return;
4392 
4393 	ns = nvme_find_get_ns(ctrl, nsid);
4394 	if (ns) {
4395 		nvme_validate_ns(ns, &info);
4396 		nvme_put_ns(ns);
4397 	} else {
4398 		nvme_alloc_ns(ctrl, &info);
4399 	}
4400 }
4401 
4402 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4403 					unsigned nsid)
4404 {
4405 	struct nvme_ns *ns, *next;
4406 	LIST_HEAD(rm_list);
4407 
4408 	down_write(&ctrl->namespaces_rwsem);
4409 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4410 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4411 			list_move_tail(&ns->list, &rm_list);
4412 	}
4413 	up_write(&ctrl->namespaces_rwsem);
4414 
4415 	list_for_each_entry_safe(ns, next, &rm_list, list)
4416 		nvme_ns_remove(ns);
4417 
4418 }
4419 
4420 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4421 {
4422 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4423 	__le32 *ns_list;
4424 	u32 prev = 0;
4425 	int ret = 0, i;
4426 
4427 	if (nvme_ctrl_limited_cns(ctrl))
4428 		return -EOPNOTSUPP;
4429 
4430 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4431 	if (!ns_list)
4432 		return -ENOMEM;
4433 
4434 	for (;;) {
4435 		struct nvme_command cmd = {
4436 			.identify.opcode	= nvme_admin_identify,
4437 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4438 			.identify.nsid		= cpu_to_le32(prev),
4439 		};
4440 
4441 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4442 					    NVME_IDENTIFY_DATA_SIZE);
4443 		if (ret) {
4444 			dev_warn(ctrl->device,
4445 				"Identify NS List failed (status=0x%x)\n", ret);
4446 			goto free;
4447 		}
4448 
4449 		for (i = 0; i < nr_entries; i++) {
4450 			u32 nsid = le32_to_cpu(ns_list[i]);
4451 
4452 			if (!nsid)	/* end of the list? */
4453 				goto out;
4454 			nvme_scan_ns(ctrl, nsid);
4455 			while (++prev < nsid)
4456 				nvme_ns_remove_by_nsid(ctrl, prev);
4457 		}
4458 	}
4459  out:
4460 	nvme_remove_invalid_namespaces(ctrl, prev);
4461  free:
4462 	kfree(ns_list);
4463 	return ret;
4464 }
4465 
4466 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4467 {
4468 	struct nvme_id_ctrl *id;
4469 	u32 nn, i;
4470 
4471 	if (nvme_identify_ctrl(ctrl, &id))
4472 		return;
4473 	nn = le32_to_cpu(id->nn);
4474 	kfree(id);
4475 
4476 	for (i = 1; i <= nn; i++)
4477 		nvme_scan_ns(ctrl, i);
4478 
4479 	nvme_remove_invalid_namespaces(ctrl, nn);
4480 }
4481 
4482 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4483 {
4484 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4485 	__le32 *log;
4486 	int error;
4487 
4488 	log = kzalloc(log_size, GFP_KERNEL);
4489 	if (!log)
4490 		return;
4491 
4492 	/*
4493 	 * We need to read the log to clear the AEN, but we don't want to rely
4494 	 * on it for the changed namespace information as userspace could have
4495 	 * raced with us in reading the log page, which could cause us to miss
4496 	 * updates.
4497 	 */
4498 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4499 			NVME_CSI_NVM, log, log_size, 0);
4500 	if (error)
4501 		dev_warn(ctrl->device,
4502 			"reading changed ns log failed: %d\n", error);
4503 
4504 	kfree(log);
4505 }
4506 
4507 static void nvme_scan_work(struct work_struct *work)
4508 {
4509 	struct nvme_ctrl *ctrl =
4510 		container_of(work, struct nvme_ctrl, scan_work);
4511 	int ret;
4512 
4513 	/* No tagset on a live ctrl means IO queues could not created */
4514 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4515 		return;
4516 
4517 	/*
4518 	 * Identify controller limits can change at controller reset due to
4519 	 * new firmware download, even though it is not common we cannot ignore
4520 	 * such scenario. Controller's non-mdts limits are reported in the unit
4521 	 * of logical blocks that is dependent on the format of attached
4522 	 * namespace. Hence re-read the limits at the time of ns allocation.
4523 	 */
4524 	ret = nvme_init_non_mdts_limits(ctrl);
4525 	if (ret < 0) {
4526 		dev_warn(ctrl->device,
4527 			"reading non-mdts-limits failed: %d\n", ret);
4528 		return;
4529 	}
4530 
4531 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4532 		dev_info(ctrl->device, "rescanning namespaces.\n");
4533 		nvme_clear_changed_ns_log(ctrl);
4534 	}
4535 
4536 	mutex_lock(&ctrl->scan_lock);
4537 	if (nvme_scan_ns_list(ctrl) != 0)
4538 		nvme_scan_ns_sequential(ctrl);
4539 	mutex_unlock(&ctrl->scan_lock);
4540 }
4541 
4542 /*
4543  * This function iterates the namespace list unlocked to allow recovery from
4544  * controller failure. It is up to the caller to ensure the namespace list is
4545  * not modified by scan work while this function is executing.
4546  */
4547 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4548 {
4549 	struct nvme_ns *ns, *next;
4550 	LIST_HEAD(ns_list);
4551 
4552 	/*
4553 	 * make sure to requeue I/O to all namespaces as these
4554 	 * might result from the scan itself and must complete
4555 	 * for the scan_work to make progress
4556 	 */
4557 	nvme_mpath_clear_ctrl_paths(ctrl);
4558 
4559 	/* prevent racing with ns scanning */
4560 	flush_work(&ctrl->scan_work);
4561 
4562 	/*
4563 	 * The dead states indicates the controller was not gracefully
4564 	 * disconnected. In that case, we won't be able to flush any data while
4565 	 * removing the namespaces' disks; fail all the queues now to avoid
4566 	 * potentially having to clean up the failed sync later.
4567 	 */
4568 	if (ctrl->state == NVME_CTRL_DEAD)
4569 		nvme_kill_queues(ctrl);
4570 
4571 	/* this is a no-op when called from the controller reset handler */
4572 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4573 
4574 	down_write(&ctrl->namespaces_rwsem);
4575 	list_splice_init(&ctrl->namespaces, &ns_list);
4576 	up_write(&ctrl->namespaces_rwsem);
4577 
4578 	list_for_each_entry_safe(ns, next, &ns_list, list)
4579 		nvme_ns_remove(ns);
4580 }
4581 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4582 
4583 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4584 {
4585 	struct nvme_ctrl *ctrl =
4586 		container_of(dev, struct nvme_ctrl, ctrl_device);
4587 	struct nvmf_ctrl_options *opts = ctrl->opts;
4588 	int ret;
4589 
4590 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4591 	if (ret)
4592 		return ret;
4593 
4594 	if (opts) {
4595 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4596 		if (ret)
4597 			return ret;
4598 
4599 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4600 				opts->trsvcid ?: "none");
4601 		if (ret)
4602 			return ret;
4603 
4604 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4605 				opts->host_traddr ?: "none");
4606 		if (ret)
4607 			return ret;
4608 
4609 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4610 				opts->host_iface ?: "none");
4611 	}
4612 	return ret;
4613 }
4614 
4615 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4616 {
4617 	char *envp[2] = { envdata, NULL };
4618 
4619 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4620 }
4621 
4622 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4623 {
4624 	char *envp[2] = { NULL, NULL };
4625 	u32 aen_result = ctrl->aen_result;
4626 
4627 	ctrl->aen_result = 0;
4628 	if (!aen_result)
4629 		return;
4630 
4631 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4632 	if (!envp[0])
4633 		return;
4634 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4635 	kfree(envp[0]);
4636 }
4637 
4638 static void nvme_async_event_work(struct work_struct *work)
4639 {
4640 	struct nvme_ctrl *ctrl =
4641 		container_of(work, struct nvme_ctrl, async_event_work);
4642 
4643 	nvme_aen_uevent(ctrl);
4644 
4645 	/*
4646 	 * The transport drivers must guarantee AER submission here is safe by
4647 	 * flushing ctrl async_event_work after changing the controller state
4648 	 * from LIVE and before freeing the admin queue.
4649 	*/
4650 	if (ctrl->state == NVME_CTRL_LIVE)
4651 		ctrl->ops->submit_async_event(ctrl);
4652 }
4653 
4654 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4655 {
4656 
4657 	u32 csts;
4658 
4659 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4660 		return false;
4661 
4662 	if (csts == ~0)
4663 		return false;
4664 
4665 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4666 }
4667 
4668 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4669 {
4670 	struct nvme_fw_slot_info_log *log;
4671 
4672 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4673 	if (!log)
4674 		return;
4675 
4676 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4677 			log, sizeof(*log), 0))
4678 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4679 	kfree(log);
4680 }
4681 
4682 static void nvme_fw_act_work(struct work_struct *work)
4683 {
4684 	struct nvme_ctrl *ctrl = container_of(work,
4685 				struct nvme_ctrl, fw_act_work);
4686 	unsigned long fw_act_timeout;
4687 
4688 	if (ctrl->mtfa)
4689 		fw_act_timeout = jiffies +
4690 				msecs_to_jiffies(ctrl->mtfa * 100);
4691 	else
4692 		fw_act_timeout = jiffies +
4693 				msecs_to_jiffies(admin_timeout * 1000);
4694 
4695 	nvme_stop_queues(ctrl);
4696 	while (nvme_ctrl_pp_status(ctrl)) {
4697 		if (time_after(jiffies, fw_act_timeout)) {
4698 			dev_warn(ctrl->device,
4699 				"Fw activation timeout, reset controller\n");
4700 			nvme_try_sched_reset(ctrl);
4701 			return;
4702 		}
4703 		msleep(100);
4704 	}
4705 
4706 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4707 		return;
4708 
4709 	nvme_start_queues(ctrl);
4710 	/* read FW slot information to clear the AER */
4711 	nvme_get_fw_slot_info(ctrl);
4712 
4713 	queue_work(nvme_wq, &ctrl->async_event_work);
4714 }
4715 
4716 static u32 nvme_aer_type(u32 result)
4717 {
4718 	return result & 0x7;
4719 }
4720 
4721 static u32 nvme_aer_subtype(u32 result)
4722 {
4723 	return (result & 0xff00) >> 8;
4724 }
4725 
4726 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4727 {
4728 	u32 aer_notice_type = nvme_aer_subtype(result);
4729 	bool requeue = true;
4730 
4731 	trace_nvme_async_event(ctrl, aer_notice_type);
4732 
4733 	switch (aer_notice_type) {
4734 	case NVME_AER_NOTICE_NS_CHANGED:
4735 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4736 		nvme_queue_scan(ctrl);
4737 		break;
4738 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4739 		/*
4740 		 * We are (ab)using the RESETTING state to prevent subsequent
4741 		 * recovery actions from interfering with the controller's
4742 		 * firmware activation.
4743 		 */
4744 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4745 			nvme_auth_stop(ctrl);
4746 			requeue = false;
4747 			queue_work(nvme_wq, &ctrl->fw_act_work);
4748 		}
4749 		break;
4750 #ifdef CONFIG_NVME_MULTIPATH
4751 	case NVME_AER_NOTICE_ANA:
4752 		if (!ctrl->ana_log_buf)
4753 			break;
4754 		queue_work(nvme_wq, &ctrl->ana_work);
4755 		break;
4756 #endif
4757 	case NVME_AER_NOTICE_DISC_CHANGED:
4758 		ctrl->aen_result = result;
4759 		break;
4760 	default:
4761 		dev_warn(ctrl->device, "async event result %08x\n", result);
4762 	}
4763 	return requeue;
4764 }
4765 
4766 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4767 {
4768 	trace_nvme_async_event(ctrl, NVME_AER_ERROR);
4769 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4770 	nvme_reset_ctrl(ctrl);
4771 }
4772 
4773 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4774 		volatile union nvme_result *res)
4775 {
4776 	u32 result = le32_to_cpu(res->u32);
4777 	u32 aer_type = nvme_aer_type(result);
4778 	u32 aer_subtype = nvme_aer_subtype(result);
4779 	bool requeue = true;
4780 
4781 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4782 		return;
4783 
4784 	switch (aer_type) {
4785 	case NVME_AER_NOTICE:
4786 		requeue = nvme_handle_aen_notice(ctrl, result);
4787 		break;
4788 	case NVME_AER_ERROR:
4789 		/*
4790 		 * For a persistent internal error, don't run async_event_work
4791 		 * to submit a new AER. The controller reset will do it.
4792 		 */
4793 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4794 			nvme_handle_aer_persistent_error(ctrl);
4795 			return;
4796 		}
4797 		fallthrough;
4798 	case NVME_AER_SMART:
4799 	case NVME_AER_CSS:
4800 	case NVME_AER_VS:
4801 		trace_nvme_async_event(ctrl, aer_type);
4802 		ctrl->aen_result = result;
4803 		break;
4804 	default:
4805 		break;
4806 	}
4807 
4808 	if (requeue)
4809 		queue_work(nvme_wq, &ctrl->async_event_work);
4810 }
4811 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4812 
4813 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4814 		const struct blk_mq_ops *ops, unsigned int flags,
4815 		unsigned int cmd_size)
4816 {
4817 	int ret;
4818 
4819 	memset(set, 0, sizeof(*set));
4820 	set->ops = ops;
4821 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4822 	if (ctrl->ops->flags & NVME_F_FABRICS)
4823 		set->reserved_tags = NVMF_RESERVED_TAGS;
4824 	set->numa_node = ctrl->numa_node;
4825 	set->flags = flags;
4826 	set->cmd_size = cmd_size;
4827 	set->driver_data = ctrl;
4828 	set->nr_hw_queues = 1;
4829 	set->timeout = NVME_ADMIN_TIMEOUT;
4830 	ret = blk_mq_alloc_tag_set(set);
4831 	if (ret)
4832 		return ret;
4833 
4834 	ctrl->admin_q = blk_mq_init_queue(set);
4835 	if (IS_ERR(ctrl->admin_q)) {
4836 		ret = PTR_ERR(ctrl->admin_q);
4837 		goto out_free_tagset;
4838 	}
4839 
4840 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4841 		ctrl->fabrics_q = blk_mq_init_queue(set);
4842 		if (IS_ERR(ctrl->fabrics_q)) {
4843 			ret = PTR_ERR(ctrl->fabrics_q);
4844 			goto out_cleanup_admin_q;
4845 		}
4846 	}
4847 
4848 	ctrl->admin_tagset = set;
4849 	return 0;
4850 
4851 out_cleanup_admin_q:
4852 	blk_mq_destroy_queue(ctrl->admin_q);
4853 out_free_tagset:
4854 	blk_mq_free_tag_set(ctrl->admin_tagset);
4855 	return ret;
4856 }
4857 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4858 
4859 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4860 {
4861 	blk_mq_destroy_queue(ctrl->admin_q);
4862 	if (ctrl->ops->flags & NVME_F_FABRICS)
4863 		blk_mq_destroy_queue(ctrl->fabrics_q);
4864 	blk_mq_free_tag_set(ctrl->admin_tagset);
4865 }
4866 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4867 
4868 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4869 		const struct blk_mq_ops *ops, unsigned int flags,
4870 		unsigned int cmd_size)
4871 {
4872 	int ret;
4873 
4874 	memset(set, 0, sizeof(*set));
4875 	set->ops = ops;
4876 	set->queue_depth = ctrl->sqsize + 1;
4877 	set->reserved_tags = NVMF_RESERVED_TAGS;
4878 	set->numa_node = ctrl->numa_node;
4879 	set->flags = flags;
4880 	set->cmd_size = cmd_size,
4881 	set->driver_data = ctrl;
4882 	set->nr_hw_queues = ctrl->queue_count - 1;
4883 	set->timeout = NVME_IO_TIMEOUT;
4884 	if (ops->map_queues)
4885 		set->nr_maps = ctrl->opts->nr_poll_queues ? HCTX_MAX_TYPES : 2;
4886 	ret = blk_mq_alloc_tag_set(set);
4887 	if (ret)
4888 		return ret;
4889 
4890 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4891 		ctrl->connect_q = blk_mq_init_queue(set);
4892         	if (IS_ERR(ctrl->connect_q)) {
4893 			ret = PTR_ERR(ctrl->connect_q);
4894 			goto out_free_tag_set;
4895 		}
4896 	}
4897 
4898 	ctrl->tagset = set;
4899 	return 0;
4900 
4901 out_free_tag_set:
4902 	blk_mq_free_tag_set(set);
4903 	return ret;
4904 }
4905 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4906 
4907 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4908 {
4909 	if (ctrl->ops->flags & NVME_F_FABRICS)
4910 		blk_mq_destroy_queue(ctrl->connect_q);
4911 	blk_mq_free_tag_set(ctrl->tagset);
4912 }
4913 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4914 
4915 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4916 {
4917 	nvme_mpath_stop(ctrl);
4918 	nvme_auth_stop(ctrl);
4919 	nvme_stop_keep_alive(ctrl);
4920 	nvme_stop_failfast_work(ctrl);
4921 	flush_work(&ctrl->async_event_work);
4922 	cancel_work_sync(&ctrl->fw_act_work);
4923 	if (ctrl->ops->stop_ctrl)
4924 		ctrl->ops->stop_ctrl(ctrl);
4925 }
4926 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4927 
4928 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4929 {
4930 	nvme_start_keep_alive(ctrl);
4931 
4932 	nvme_enable_aen(ctrl);
4933 
4934 	/*
4935 	 * persistent discovery controllers need to send indication to userspace
4936 	 * to re-read the discovery log page to learn about possible changes
4937 	 * that were missed. We identify persistent discovery controllers by
4938 	 * checking that they started once before, hence are reconnecting back.
4939 	 */
4940 	if (test_and_set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4941 	    nvme_discovery_ctrl(ctrl))
4942 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4943 
4944 	if (ctrl->queue_count > 1) {
4945 		nvme_queue_scan(ctrl);
4946 		nvme_start_queues(ctrl);
4947 		nvme_mpath_update(ctrl);
4948 	}
4949 
4950 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4951 }
4952 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4953 
4954 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4955 {
4956 	nvme_hwmon_exit(ctrl);
4957 	nvme_fault_inject_fini(&ctrl->fault_inject);
4958 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4959 	cdev_device_del(&ctrl->cdev, ctrl->device);
4960 	nvme_put_ctrl(ctrl);
4961 }
4962 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4963 
4964 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4965 {
4966 	struct nvme_effects_log	*cel;
4967 	unsigned long i;
4968 
4969 	xa_for_each(&ctrl->cels, i, cel) {
4970 		xa_erase(&ctrl->cels, i);
4971 		kfree(cel);
4972 	}
4973 
4974 	xa_destroy(&ctrl->cels);
4975 }
4976 
4977 static void nvme_free_ctrl(struct device *dev)
4978 {
4979 	struct nvme_ctrl *ctrl =
4980 		container_of(dev, struct nvme_ctrl, ctrl_device);
4981 	struct nvme_subsystem *subsys = ctrl->subsys;
4982 
4983 	if (!subsys || ctrl->instance != subsys->instance)
4984 		ida_free(&nvme_instance_ida, ctrl->instance);
4985 
4986 	nvme_free_cels(ctrl);
4987 	nvme_mpath_uninit(ctrl);
4988 	nvme_auth_stop(ctrl);
4989 	nvme_auth_free(ctrl);
4990 	__free_page(ctrl->discard_page);
4991 
4992 	if (subsys) {
4993 		mutex_lock(&nvme_subsystems_lock);
4994 		list_del(&ctrl->subsys_entry);
4995 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4996 		mutex_unlock(&nvme_subsystems_lock);
4997 	}
4998 
4999 	ctrl->ops->free_ctrl(ctrl);
5000 
5001 	if (subsys)
5002 		nvme_put_subsystem(subsys);
5003 }
5004 
5005 /*
5006  * Initialize a NVMe controller structures.  This needs to be called during
5007  * earliest initialization so that we have the initialized structured around
5008  * during probing.
5009  */
5010 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
5011 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
5012 {
5013 	int ret;
5014 
5015 	ctrl->state = NVME_CTRL_NEW;
5016 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
5017 	spin_lock_init(&ctrl->lock);
5018 	mutex_init(&ctrl->scan_lock);
5019 	INIT_LIST_HEAD(&ctrl->namespaces);
5020 	xa_init(&ctrl->cels);
5021 	init_rwsem(&ctrl->namespaces_rwsem);
5022 	ctrl->dev = dev;
5023 	ctrl->ops = ops;
5024 	ctrl->quirks = quirks;
5025 	ctrl->numa_node = NUMA_NO_NODE;
5026 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
5027 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
5028 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
5029 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
5030 	init_waitqueue_head(&ctrl->state_wq);
5031 
5032 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
5033 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
5034 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
5035 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
5036 
5037 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
5038 			PAGE_SIZE);
5039 	ctrl->discard_page = alloc_page(GFP_KERNEL);
5040 	if (!ctrl->discard_page) {
5041 		ret = -ENOMEM;
5042 		goto out;
5043 	}
5044 
5045 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
5046 	if (ret < 0)
5047 		goto out;
5048 	ctrl->instance = ret;
5049 
5050 	device_initialize(&ctrl->ctrl_device);
5051 	ctrl->device = &ctrl->ctrl_device;
5052 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
5053 			ctrl->instance);
5054 	ctrl->device->class = nvme_class;
5055 	ctrl->device->parent = ctrl->dev;
5056 	ctrl->device->groups = nvme_dev_attr_groups;
5057 	ctrl->device->release = nvme_free_ctrl;
5058 	dev_set_drvdata(ctrl->device, ctrl);
5059 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
5060 	if (ret)
5061 		goto out_release_instance;
5062 
5063 	nvme_get_ctrl(ctrl);
5064 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
5065 	ctrl->cdev.owner = ops->module;
5066 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
5067 	if (ret)
5068 		goto out_free_name;
5069 
5070 	/*
5071 	 * Initialize latency tolerance controls.  The sysfs files won't
5072 	 * be visible to userspace unless the device actually supports APST.
5073 	 */
5074 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
5075 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
5076 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
5077 
5078 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
5079 	nvme_mpath_init_ctrl(ctrl);
5080 	nvme_auth_init_ctrl(ctrl);
5081 
5082 	return 0;
5083 out_free_name:
5084 	nvme_put_ctrl(ctrl);
5085 	kfree_const(ctrl->device->kobj.name);
5086 out_release_instance:
5087 	ida_free(&nvme_instance_ida, ctrl->instance);
5088 out:
5089 	if (ctrl->discard_page)
5090 		__free_page(ctrl->discard_page);
5091 	return ret;
5092 }
5093 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
5094 
5095 static void nvme_start_ns_queue(struct nvme_ns *ns)
5096 {
5097 	if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
5098 		blk_mq_unquiesce_queue(ns->queue);
5099 }
5100 
5101 static void nvme_stop_ns_queue(struct nvme_ns *ns)
5102 {
5103 	if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
5104 		blk_mq_quiesce_queue(ns->queue);
5105 	else
5106 		blk_mq_wait_quiesce_done(ns->queue);
5107 }
5108 
5109 /*
5110  * Prepare a queue for teardown.
5111  *
5112  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
5113  * the capacity to 0 after that to avoid blocking dispatchers that may be
5114  * holding bd_butex.  This will end buffered writers dirtying pages that can't
5115  * be synced.
5116  */
5117 static void nvme_set_queue_dying(struct nvme_ns *ns)
5118 {
5119 	if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
5120 		return;
5121 
5122 	blk_mark_disk_dead(ns->disk);
5123 	nvme_start_ns_queue(ns);
5124 
5125 	set_capacity_and_notify(ns->disk, 0);
5126 }
5127 
5128 /**
5129  * nvme_kill_queues(): Ends all namespace queues
5130  * @ctrl: the dead controller that needs to end
5131  *
5132  * Call this function when the driver determines it is unable to get the
5133  * controller in a state capable of servicing IO.
5134  */
5135 void nvme_kill_queues(struct nvme_ctrl *ctrl)
5136 {
5137 	struct nvme_ns *ns;
5138 
5139 	down_read(&ctrl->namespaces_rwsem);
5140 
5141 	/* Forcibly unquiesce queues to avoid blocking dispatch */
5142 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
5143 		nvme_start_admin_queue(ctrl);
5144 
5145 	list_for_each_entry(ns, &ctrl->namespaces, list)
5146 		nvme_set_queue_dying(ns);
5147 
5148 	up_read(&ctrl->namespaces_rwsem);
5149 }
5150 EXPORT_SYMBOL_GPL(nvme_kill_queues);
5151 
5152 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5153 {
5154 	struct nvme_ns *ns;
5155 
5156 	down_read(&ctrl->namespaces_rwsem);
5157 	list_for_each_entry(ns, &ctrl->namespaces, list)
5158 		blk_mq_unfreeze_queue(ns->queue);
5159 	up_read(&ctrl->namespaces_rwsem);
5160 }
5161 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5162 
5163 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5164 {
5165 	struct nvme_ns *ns;
5166 
5167 	down_read(&ctrl->namespaces_rwsem);
5168 	list_for_each_entry(ns, &ctrl->namespaces, list) {
5169 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5170 		if (timeout <= 0)
5171 			break;
5172 	}
5173 	up_read(&ctrl->namespaces_rwsem);
5174 	return timeout;
5175 }
5176 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5177 
5178 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5179 {
5180 	struct nvme_ns *ns;
5181 
5182 	down_read(&ctrl->namespaces_rwsem);
5183 	list_for_each_entry(ns, &ctrl->namespaces, list)
5184 		blk_mq_freeze_queue_wait(ns->queue);
5185 	up_read(&ctrl->namespaces_rwsem);
5186 }
5187 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5188 
5189 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5190 {
5191 	struct nvme_ns *ns;
5192 
5193 	down_read(&ctrl->namespaces_rwsem);
5194 	list_for_each_entry(ns, &ctrl->namespaces, list)
5195 		blk_freeze_queue_start(ns->queue);
5196 	up_read(&ctrl->namespaces_rwsem);
5197 }
5198 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5199 
5200 void nvme_stop_queues(struct nvme_ctrl *ctrl)
5201 {
5202 	struct nvme_ns *ns;
5203 
5204 	down_read(&ctrl->namespaces_rwsem);
5205 	list_for_each_entry(ns, &ctrl->namespaces, list)
5206 		nvme_stop_ns_queue(ns);
5207 	up_read(&ctrl->namespaces_rwsem);
5208 }
5209 EXPORT_SYMBOL_GPL(nvme_stop_queues);
5210 
5211 void nvme_start_queues(struct nvme_ctrl *ctrl)
5212 {
5213 	struct nvme_ns *ns;
5214 
5215 	down_read(&ctrl->namespaces_rwsem);
5216 	list_for_each_entry(ns, &ctrl->namespaces, list)
5217 		nvme_start_ns_queue(ns);
5218 	up_read(&ctrl->namespaces_rwsem);
5219 }
5220 EXPORT_SYMBOL_GPL(nvme_start_queues);
5221 
5222 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
5223 {
5224 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5225 		blk_mq_quiesce_queue(ctrl->admin_q);
5226 	else
5227 		blk_mq_wait_quiesce_done(ctrl->admin_q);
5228 }
5229 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
5230 
5231 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
5232 {
5233 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5234 		blk_mq_unquiesce_queue(ctrl->admin_q);
5235 }
5236 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
5237 
5238 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5239 {
5240 	struct nvme_ns *ns;
5241 
5242 	down_read(&ctrl->namespaces_rwsem);
5243 	list_for_each_entry(ns, &ctrl->namespaces, list)
5244 		blk_sync_queue(ns->queue);
5245 	up_read(&ctrl->namespaces_rwsem);
5246 }
5247 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5248 
5249 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5250 {
5251 	nvme_sync_io_queues(ctrl);
5252 	if (ctrl->admin_q)
5253 		blk_sync_queue(ctrl->admin_q);
5254 }
5255 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5256 
5257 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5258 {
5259 	if (file->f_op != &nvme_dev_fops)
5260 		return NULL;
5261 	return file->private_data;
5262 }
5263 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
5264 
5265 /*
5266  * Check we didn't inadvertently grow the command structure sizes:
5267  */
5268 static inline void _nvme_check_size(void)
5269 {
5270 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5271 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5272 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5273 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5274 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5275 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5276 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5277 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5278 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5279 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5280 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5281 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5282 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5283 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5284 			NVME_IDENTIFY_DATA_SIZE);
5285 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5286 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5287 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5288 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5289 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5290 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5291 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5292 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5293 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5294 }
5295 
5296 
5297 static int __init nvme_core_init(void)
5298 {
5299 	int result = -ENOMEM;
5300 
5301 	_nvme_check_size();
5302 
5303 	nvme_wq = alloc_workqueue("nvme-wq",
5304 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5305 	if (!nvme_wq)
5306 		goto out;
5307 
5308 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
5309 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5310 	if (!nvme_reset_wq)
5311 		goto destroy_wq;
5312 
5313 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
5314 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5315 	if (!nvme_delete_wq)
5316 		goto destroy_reset_wq;
5317 
5318 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5319 			NVME_MINORS, "nvme");
5320 	if (result < 0)
5321 		goto destroy_delete_wq;
5322 
5323 	nvme_class = class_create(THIS_MODULE, "nvme");
5324 	if (IS_ERR(nvme_class)) {
5325 		result = PTR_ERR(nvme_class);
5326 		goto unregister_chrdev;
5327 	}
5328 	nvme_class->dev_uevent = nvme_class_uevent;
5329 
5330 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
5331 	if (IS_ERR(nvme_subsys_class)) {
5332 		result = PTR_ERR(nvme_subsys_class);
5333 		goto destroy_class;
5334 	}
5335 
5336 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5337 				     "nvme-generic");
5338 	if (result < 0)
5339 		goto destroy_subsys_class;
5340 
5341 	nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
5342 	if (IS_ERR(nvme_ns_chr_class)) {
5343 		result = PTR_ERR(nvme_ns_chr_class);
5344 		goto unregister_generic_ns;
5345 	}
5346 
5347 	return 0;
5348 
5349 unregister_generic_ns:
5350 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5351 destroy_subsys_class:
5352 	class_destroy(nvme_subsys_class);
5353 destroy_class:
5354 	class_destroy(nvme_class);
5355 unregister_chrdev:
5356 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5357 destroy_delete_wq:
5358 	destroy_workqueue(nvme_delete_wq);
5359 destroy_reset_wq:
5360 	destroy_workqueue(nvme_reset_wq);
5361 destroy_wq:
5362 	destroy_workqueue(nvme_wq);
5363 out:
5364 	return result;
5365 }
5366 
5367 static void __exit nvme_core_exit(void)
5368 {
5369 	class_destroy(nvme_ns_chr_class);
5370 	class_destroy(nvme_subsys_class);
5371 	class_destroy(nvme_class);
5372 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5373 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5374 	destroy_workqueue(nvme_delete_wq);
5375 	destroy_workqueue(nvme_reset_wq);
5376 	destroy_workqueue(nvme_wq);
5377 	ida_destroy(&nvme_ns_chr_minor_ida);
5378 	ida_destroy(&nvme_instance_ida);
5379 }
5380 
5381 MODULE_LICENSE("GPL");
5382 MODULE_VERSION("1.0");
5383 module_init(nvme_core_init);
5384 module_exit(nvme_core_exit);
5385