xref: /openbmc/linux/drivers/nvme/host/core.c (revision a33ca174)
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 	if (req->cmd_flags & REQ_NVME_MPATH)
388 		nvme_mpath_end_request(req);
389 	blk_mq_end_request(req, status);
390 }
391 
392 void nvme_complete_rq(struct request *req)
393 {
394 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
395 
396 	trace_nvme_complete_rq(req);
397 	nvme_cleanup_cmd(req);
398 
399 	if (ctrl->kas)
400 		ctrl->comp_seen = true;
401 
402 	switch (nvme_decide_disposition(req)) {
403 	case COMPLETE:
404 		nvme_end_req(req);
405 		return;
406 	case RETRY:
407 		nvme_retry_req(req);
408 		return;
409 	case FAILOVER:
410 		nvme_failover_req(req);
411 		return;
412 	case AUTHENTICATE:
413 #ifdef CONFIG_NVME_AUTH
414 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
415 		nvme_retry_req(req);
416 #else
417 		nvme_end_req(req);
418 #endif
419 		return;
420 	}
421 }
422 EXPORT_SYMBOL_GPL(nvme_complete_rq);
423 
424 void nvme_complete_batch_req(struct request *req)
425 {
426 	trace_nvme_complete_rq(req);
427 	nvme_cleanup_cmd(req);
428 	nvme_end_req_zoned(req);
429 }
430 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
431 
432 /*
433  * Called to unwind from ->queue_rq on a failed command submission so that the
434  * multipathing code gets called to potentially failover to another path.
435  * The caller needs to unwind all transport specific resource allocations and
436  * must return propagate the return value.
437  */
438 blk_status_t nvme_host_path_error(struct request *req)
439 {
440 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
441 	blk_mq_set_request_complete(req);
442 	nvme_complete_rq(req);
443 	return BLK_STS_OK;
444 }
445 EXPORT_SYMBOL_GPL(nvme_host_path_error);
446 
447 bool nvme_cancel_request(struct request *req, void *data)
448 {
449 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
450 				"Cancelling I/O %d", req->tag);
451 
452 	/* don't abort one completed request */
453 	if (blk_mq_request_completed(req))
454 		return true;
455 
456 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
457 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
458 	blk_mq_complete_request(req);
459 	return true;
460 }
461 EXPORT_SYMBOL_GPL(nvme_cancel_request);
462 
463 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
464 {
465 	if (ctrl->tagset) {
466 		blk_mq_tagset_busy_iter(ctrl->tagset,
467 				nvme_cancel_request, ctrl);
468 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
469 	}
470 }
471 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
472 
473 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
474 {
475 	if (ctrl->admin_tagset) {
476 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
477 				nvme_cancel_request, ctrl);
478 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
479 	}
480 }
481 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
482 
483 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
484 		enum nvme_ctrl_state new_state)
485 {
486 	enum nvme_ctrl_state old_state;
487 	unsigned long flags;
488 	bool changed = false;
489 
490 	spin_lock_irqsave(&ctrl->lock, flags);
491 
492 	old_state = ctrl->state;
493 	switch (new_state) {
494 	case NVME_CTRL_LIVE:
495 		switch (old_state) {
496 		case NVME_CTRL_NEW:
497 		case NVME_CTRL_RESETTING:
498 		case NVME_CTRL_CONNECTING:
499 			changed = true;
500 			fallthrough;
501 		default:
502 			break;
503 		}
504 		break;
505 	case NVME_CTRL_RESETTING:
506 		switch (old_state) {
507 		case NVME_CTRL_NEW:
508 		case NVME_CTRL_LIVE:
509 			changed = true;
510 			fallthrough;
511 		default:
512 			break;
513 		}
514 		break;
515 	case NVME_CTRL_CONNECTING:
516 		switch (old_state) {
517 		case NVME_CTRL_NEW:
518 		case NVME_CTRL_RESETTING:
519 			changed = true;
520 			fallthrough;
521 		default:
522 			break;
523 		}
524 		break;
525 	case NVME_CTRL_DELETING:
526 		switch (old_state) {
527 		case NVME_CTRL_LIVE:
528 		case NVME_CTRL_RESETTING:
529 		case NVME_CTRL_CONNECTING:
530 			changed = true;
531 			fallthrough;
532 		default:
533 			break;
534 		}
535 		break;
536 	case NVME_CTRL_DELETING_NOIO:
537 		switch (old_state) {
538 		case NVME_CTRL_DELETING:
539 		case NVME_CTRL_DEAD:
540 			changed = true;
541 			fallthrough;
542 		default:
543 			break;
544 		}
545 		break;
546 	case NVME_CTRL_DEAD:
547 		switch (old_state) {
548 		case NVME_CTRL_DELETING:
549 			changed = true;
550 			fallthrough;
551 		default:
552 			break;
553 		}
554 		break;
555 	default:
556 		break;
557 	}
558 
559 	if (changed) {
560 		ctrl->state = new_state;
561 		wake_up_all(&ctrl->state_wq);
562 	}
563 
564 	spin_unlock_irqrestore(&ctrl->lock, flags);
565 	if (!changed)
566 		return false;
567 
568 	if (ctrl->state == NVME_CTRL_LIVE) {
569 		if (old_state == NVME_CTRL_CONNECTING)
570 			nvme_stop_failfast_work(ctrl);
571 		nvme_kick_requeue_lists(ctrl);
572 	} else if (ctrl->state == NVME_CTRL_CONNECTING &&
573 		old_state == NVME_CTRL_RESETTING) {
574 		nvme_start_failfast_work(ctrl);
575 	}
576 	return changed;
577 }
578 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
579 
580 /*
581  * Returns true for sink states that can't ever transition back to live.
582  */
583 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
584 {
585 	switch (ctrl->state) {
586 	case NVME_CTRL_NEW:
587 	case NVME_CTRL_LIVE:
588 	case NVME_CTRL_RESETTING:
589 	case NVME_CTRL_CONNECTING:
590 		return false;
591 	case NVME_CTRL_DELETING:
592 	case NVME_CTRL_DELETING_NOIO:
593 	case NVME_CTRL_DEAD:
594 		return true;
595 	default:
596 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
597 		return true;
598 	}
599 }
600 
601 /*
602  * Waits for the controller state to be resetting, or returns false if it is
603  * not possible to ever transition to that state.
604  */
605 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
606 {
607 	wait_event(ctrl->state_wq,
608 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
609 		   nvme_state_terminal(ctrl));
610 	return ctrl->state == NVME_CTRL_RESETTING;
611 }
612 EXPORT_SYMBOL_GPL(nvme_wait_reset);
613 
614 static void nvme_free_ns_head(struct kref *ref)
615 {
616 	struct nvme_ns_head *head =
617 		container_of(ref, struct nvme_ns_head, ref);
618 
619 	nvme_mpath_remove_disk(head);
620 	ida_free(&head->subsys->ns_ida, head->instance);
621 	cleanup_srcu_struct(&head->srcu);
622 	nvme_put_subsystem(head->subsys);
623 	kfree(head);
624 }
625 
626 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
627 {
628 	return kref_get_unless_zero(&head->ref);
629 }
630 
631 void nvme_put_ns_head(struct nvme_ns_head *head)
632 {
633 	kref_put(&head->ref, nvme_free_ns_head);
634 }
635 
636 static void nvme_free_ns(struct kref *kref)
637 {
638 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
639 
640 	put_disk(ns->disk);
641 	nvme_put_ns_head(ns->head);
642 	nvme_put_ctrl(ns->ctrl);
643 	kfree(ns);
644 }
645 
646 static inline bool nvme_get_ns(struct nvme_ns *ns)
647 {
648 	return kref_get_unless_zero(&ns->kref);
649 }
650 
651 void nvme_put_ns(struct nvme_ns *ns)
652 {
653 	kref_put(&ns->kref, nvme_free_ns);
654 }
655 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
656 
657 static inline void nvme_clear_nvme_request(struct request *req)
658 {
659 	nvme_req(req)->status = 0;
660 	nvme_req(req)->retries = 0;
661 	nvme_req(req)->flags = 0;
662 	req->rq_flags |= RQF_DONTPREP;
663 }
664 
665 /* initialize a passthrough request */
666 void nvme_init_request(struct request *req, struct nvme_command *cmd)
667 {
668 	if (req->q->queuedata)
669 		req->timeout = NVME_IO_TIMEOUT;
670 	else /* no queuedata implies admin queue */
671 		req->timeout = NVME_ADMIN_TIMEOUT;
672 
673 	/* passthru commands should let the driver set the SGL flags */
674 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
675 
676 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
677 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
678 		req->cmd_flags |= REQ_POLLED;
679 	nvme_clear_nvme_request(req);
680 	req->rq_flags |= RQF_QUIET;
681 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
682 }
683 EXPORT_SYMBOL_GPL(nvme_init_request);
684 
685 /*
686  * For something we're not in a state to send to the device the default action
687  * is to busy it and retry it after the controller state is recovered.  However,
688  * if the controller is deleting or if anything is marked for failfast or
689  * nvme multipath it is immediately failed.
690  *
691  * Note: commands used to initialize the controller will be marked for failfast.
692  * Note: nvme cli/ioctl commands are marked for failfast.
693  */
694 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
695 		struct request *rq)
696 {
697 	if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
698 	    ctrl->state != NVME_CTRL_DELETING &&
699 	    ctrl->state != NVME_CTRL_DEAD &&
700 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
701 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
702 		return BLK_STS_RESOURCE;
703 	return nvme_host_path_error(rq);
704 }
705 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
706 
707 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
708 		bool queue_live)
709 {
710 	struct nvme_request *req = nvme_req(rq);
711 
712 	/*
713 	 * currently we have a problem sending passthru commands
714 	 * on the admin_q if the controller is not LIVE because we can't
715 	 * make sure that they are going out after the admin connect,
716 	 * controller enable and/or other commands in the initialization
717 	 * sequence. until the controller will be LIVE, fail with
718 	 * BLK_STS_RESOURCE so that they will be rescheduled.
719 	 */
720 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
721 		return false;
722 
723 	if (ctrl->ops->flags & NVME_F_FABRICS) {
724 		/*
725 		 * Only allow commands on a live queue, except for the connect
726 		 * command, which is require to set the queue live in the
727 		 * appropinquate states.
728 		 */
729 		switch (ctrl->state) {
730 		case NVME_CTRL_CONNECTING:
731 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
732 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
733 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
734 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
735 				return true;
736 			break;
737 		default:
738 			break;
739 		case NVME_CTRL_DEAD:
740 			return false;
741 		}
742 	}
743 
744 	return queue_live;
745 }
746 EXPORT_SYMBOL_GPL(__nvme_check_ready);
747 
748 static inline void nvme_setup_flush(struct nvme_ns *ns,
749 		struct nvme_command *cmnd)
750 {
751 	memset(cmnd, 0, sizeof(*cmnd));
752 	cmnd->common.opcode = nvme_cmd_flush;
753 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
754 }
755 
756 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
757 		struct nvme_command *cmnd)
758 {
759 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
760 	struct nvme_dsm_range *range;
761 	struct bio *bio;
762 
763 	/*
764 	 * Some devices do not consider the DSM 'Number of Ranges' field when
765 	 * determining how much data to DMA. Always allocate memory for maximum
766 	 * number of segments to prevent device reading beyond end of buffer.
767 	 */
768 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
769 
770 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
771 	if (!range) {
772 		/*
773 		 * If we fail allocation our range, fallback to the controller
774 		 * discard page. If that's also busy, it's safe to return
775 		 * busy, as we know we can make progress once that's freed.
776 		 */
777 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
778 			return BLK_STS_RESOURCE;
779 
780 		range = page_address(ns->ctrl->discard_page);
781 	}
782 
783 	__rq_for_each_bio(bio, req) {
784 		u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
785 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
786 
787 		if (n < segments) {
788 			range[n].cattr = cpu_to_le32(0);
789 			range[n].nlb = cpu_to_le32(nlb);
790 			range[n].slba = cpu_to_le64(slba);
791 		}
792 		n++;
793 	}
794 
795 	if (WARN_ON_ONCE(n != segments)) {
796 		if (virt_to_page(range) == ns->ctrl->discard_page)
797 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
798 		else
799 			kfree(range);
800 		return BLK_STS_IOERR;
801 	}
802 
803 	memset(cmnd, 0, sizeof(*cmnd));
804 	cmnd->dsm.opcode = nvme_cmd_dsm;
805 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
806 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
807 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
808 
809 	req->special_vec.bv_page = virt_to_page(range);
810 	req->special_vec.bv_offset = offset_in_page(range);
811 	req->special_vec.bv_len = alloc_size;
812 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
813 
814 	return BLK_STS_OK;
815 }
816 
817 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
818 			      struct request *req)
819 {
820 	u32 upper, lower;
821 	u64 ref48;
822 
823 	/* both rw and write zeroes share the same reftag format */
824 	switch (ns->guard_type) {
825 	case NVME_NVM_NS_16B_GUARD:
826 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
827 		break;
828 	case NVME_NVM_NS_64B_GUARD:
829 		ref48 = ext_pi_ref_tag(req);
830 		lower = lower_32_bits(ref48);
831 		upper = upper_32_bits(ref48);
832 
833 		cmnd->rw.reftag = cpu_to_le32(lower);
834 		cmnd->rw.cdw3 = cpu_to_le32(upper);
835 		break;
836 	default:
837 		break;
838 	}
839 }
840 
841 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
842 		struct request *req, struct nvme_command *cmnd)
843 {
844 	memset(cmnd, 0, sizeof(*cmnd));
845 
846 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
847 		return nvme_setup_discard(ns, req, cmnd);
848 
849 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
850 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
851 	cmnd->write_zeroes.slba =
852 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
853 	cmnd->write_zeroes.length =
854 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
855 
856 	if (!(req->cmd_flags & REQ_NOUNMAP) && (ns->features & NVME_NS_DEAC))
857 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
858 
859 	if (nvme_ns_has_pi(ns)) {
860 		cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
861 
862 		switch (ns->pi_type) {
863 		case NVME_NS_DPS_PI_TYPE1:
864 		case NVME_NS_DPS_PI_TYPE2:
865 			nvme_set_ref_tag(ns, cmnd, req);
866 			break;
867 		}
868 	}
869 
870 	return BLK_STS_OK;
871 }
872 
873 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
874 		struct request *req, struct nvme_command *cmnd,
875 		enum nvme_opcode op)
876 {
877 	u16 control = 0;
878 	u32 dsmgmt = 0;
879 
880 	if (req->cmd_flags & REQ_FUA)
881 		control |= NVME_RW_FUA;
882 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
883 		control |= NVME_RW_LR;
884 
885 	if (req->cmd_flags & REQ_RAHEAD)
886 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
887 
888 	cmnd->rw.opcode = op;
889 	cmnd->rw.flags = 0;
890 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
891 	cmnd->rw.cdw2 = 0;
892 	cmnd->rw.cdw3 = 0;
893 	cmnd->rw.metadata = 0;
894 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
895 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
896 	cmnd->rw.reftag = 0;
897 	cmnd->rw.apptag = 0;
898 	cmnd->rw.appmask = 0;
899 
900 	if (ns->ms) {
901 		/*
902 		 * If formated with metadata, the block layer always provides a
903 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
904 		 * we enable the PRACT bit for protection information or set the
905 		 * namespace capacity to zero to prevent any I/O.
906 		 */
907 		if (!blk_integrity_rq(req)) {
908 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
909 				return BLK_STS_NOTSUPP;
910 			control |= NVME_RW_PRINFO_PRACT;
911 		}
912 
913 		switch (ns->pi_type) {
914 		case NVME_NS_DPS_PI_TYPE3:
915 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
916 			break;
917 		case NVME_NS_DPS_PI_TYPE1:
918 		case NVME_NS_DPS_PI_TYPE2:
919 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
920 					NVME_RW_PRINFO_PRCHK_REF;
921 			if (op == nvme_cmd_zone_append)
922 				control |= NVME_RW_APPEND_PIREMAP;
923 			nvme_set_ref_tag(ns, cmnd, req);
924 			break;
925 		}
926 	}
927 
928 	cmnd->rw.control = cpu_to_le16(control);
929 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
930 	return 0;
931 }
932 
933 void nvme_cleanup_cmd(struct request *req)
934 {
935 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
936 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
937 
938 		if (req->special_vec.bv_page == ctrl->discard_page)
939 			clear_bit_unlock(0, &ctrl->discard_page_busy);
940 		else
941 			kfree(bvec_virt(&req->special_vec));
942 	}
943 }
944 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
945 
946 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
947 {
948 	struct nvme_command *cmd = nvme_req(req)->cmd;
949 	blk_status_t ret = BLK_STS_OK;
950 
951 	if (!(req->rq_flags & RQF_DONTPREP))
952 		nvme_clear_nvme_request(req);
953 
954 	switch (req_op(req)) {
955 	case REQ_OP_DRV_IN:
956 	case REQ_OP_DRV_OUT:
957 		/* these are setup prior to execution in nvme_init_request() */
958 		break;
959 	case REQ_OP_FLUSH:
960 		nvme_setup_flush(ns, cmd);
961 		break;
962 	case REQ_OP_ZONE_RESET_ALL:
963 	case REQ_OP_ZONE_RESET:
964 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
965 		break;
966 	case REQ_OP_ZONE_OPEN:
967 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
968 		break;
969 	case REQ_OP_ZONE_CLOSE:
970 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
971 		break;
972 	case REQ_OP_ZONE_FINISH:
973 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
974 		break;
975 	case REQ_OP_WRITE_ZEROES:
976 		ret = nvme_setup_write_zeroes(ns, req, cmd);
977 		break;
978 	case REQ_OP_DISCARD:
979 		ret = nvme_setup_discard(ns, req, cmd);
980 		break;
981 	case REQ_OP_READ:
982 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
983 		break;
984 	case REQ_OP_WRITE:
985 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
986 		break;
987 	case REQ_OP_ZONE_APPEND:
988 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
989 		break;
990 	default:
991 		WARN_ON_ONCE(1);
992 		return BLK_STS_IOERR;
993 	}
994 
995 	cmd->common.command_id = nvme_cid(req);
996 	trace_nvme_setup_cmd(req, cmd);
997 	return ret;
998 }
999 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1000 
1001 /*
1002  * Return values:
1003  * 0:  success
1004  * >0: nvme controller's cqe status response
1005  * <0: kernel error in lieu of controller response
1006  */
1007 static int nvme_execute_rq(struct request *rq, bool at_head)
1008 {
1009 	blk_status_t status;
1010 
1011 	status = blk_execute_rq(rq, at_head);
1012 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1013 		return -EINTR;
1014 	if (nvme_req(rq)->status)
1015 		return nvme_req(rq)->status;
1016 	return blk_status_to_errno(status);
1017 }
1018 
1019 /*
1020  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1021  * if the result is positive, it's an NVM Express status code
1022  */
1023 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1024 		union nvme_result *result, void *buffer, unsigned bufflen,
1025 		int qid, int at_head, blk_mq_req_flags_t flags)
1026 {
1027 	struct request *req;
1028 	int ret;
1029 
1030 	if (qid == NVME_QID_ANY)
1031 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1032 	else
1033 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1034 						qid - 1);
1035 
1036 	if (IS_ERR(req))
1037 		return PTR_ERR(req);
1038 	nvme_init_request(req, cmd);
1039 
1040 	if (buffer && bufflen) {
1041 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1042 		if (ret)
1043 			goto out;
1044 	}
1045 
1046 	ret = nvme_execute_rq(req, at_head);
1047 	if (result && ret >= 0)
1048 		*result = nvme_req(req)->result;
1049  out:
1050 	blk_mq_free_request(req);
1051 	return ret;
1052 }
1053 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1054 
1055 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1056 		void *buffer, unsigned bufflen)
1057 {
1058 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1059 			NVME_QID_ANY, 0, 0);
1060 }
1061 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1062 
1063 static u32 nvme_known_admin_effects(u8 opcode)
1064 {
1065 	switch (opcode) {
1066 	case nvme_admin_format_nvm:
1067 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1068 			NVME_CMD_EFFECTS_CSE_MASK;
1069 	case nvme_admin_sanitize_nvm:
1070 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1071 	default:
1072 		break;
1073 	}
1074 	return 0;
1075 }
1076 
1077 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1078 {
1079 	u32 effects = 0;
1080 
1081 	if (ns) {
1082 		if (ns->head->effects)
1083 			effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1084 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1085 			dev_warn_once(ctrl->device,
1086 				"IO command:%02x has unhandled effects:%08x\n",
1087 				opcode, effects);
1088 		return 0;
1089 	}
1090 
1091 	if (ctrl->effects)
1092 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1093 	effects |= nvme_known_admin_effects(opcode);
1094 
1095 	return effects;
1096 }
1097 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1098 
1099 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1100 			       u8 opcode)
1101 {
1102 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1103 
1104 	/*
1105 	 * For simplicity, IO to all namespaces is quiesced even if the command
1106 	 * effects say only one namespace is affected.
1107 	 */
1108 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1109 		mutex_lock(&ctrl->scan_lock);
1110 		mutex_lock(&ctrl->subsys->lock);
1111 		nvme_mpath_start_freeze(ctrl->subsys);
1112 		nvme_mpath_wait_freeze(ctrl->subsys);
1113 		nvme_start_freeze(ctrl);
1114 		nvme_wait_freeze(ctrl);
1115 	}
1116 	return effects;
1117 }
1118 
1119 void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1120 		       struct nvme_command *cmd, int status)
1121 {
1122 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1123 		nvme_unfreeze(ctrl);
1124 		nvme_mpath_unfreeze(ctrl->subsys);
1125 		mutex_unlock(&ctrl->subsys->lock);
1126 		mutex_unlock(&ctrl->scan_lock);
1127 	}
1128 	if (effects & NVME_CMD_EFFECTS_CCC) {
1129 		dev_info(ctrl->device,
1130 "controller capabilities changed, reset may be required to take effect.\n");
1131 	}
1132 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1133 		nvme_queue_scan(ctrl);
1134 		flush_work(&ctrl->scan_work);
1135 	}
1136 
1137 	switch (cmd->common.opcode) {
1138 	case nvme_admin_set_features:
1139 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1140 		case NVME_FEAT_KATO:
1141 			/*
1142 			 * Keep alive commands interval on the host should be
1143 			 * updated when KATO is modified by Set Features
1144 			 * commands.
1145 			 */
1146 			if (!status)
1147 				nvme_update_keep_alive(ctrl, cmd);
1148 			break;
1149 		default:
1150 			break;
1151 		}
1152 		break;
1153 	default:
1154 		break;
1155 	}
1156 }
1157 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1158 
1159 int nvme_execute_passthru_rq(struct request *rq, u32 *effects)
1160 {
1161 	struct nvme_command *cmd = nvme_req(rq)->cmd;
1162 	struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1163 	struct nvme_ns *ns = rq->q->queuedata;
1164 
1165 	*effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1166 	return nvme_execute_rq(rq, false);
1167 }
1168 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1169 
1170 /*
1171  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1172  *
1173  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1174  *   accounting for transport roundtrip times [..].
1175  */
1176 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1177 {
1178 	queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1179 }
1180 
1181 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1182 						 blk_status_t status)
1183 {
1184 	struct nvme_ctrl *ctrl = rq->end_io_data;
1185 	unsigned long flags;
1186 	bool startka = false;
1187 
1188 	blk_mq_free_request(rq);
1189 
1190 	if (status) {
1191 		dev_err(ctrl->device,
1192 			"failed nvme_keep_alive_end_io error=%d\n",
1193 				status);
1194 		return RQ_END_IO_NONE;
1195 	}
1196 
1197 	ctrl->comp_seen = false;
1198 	spin_lock_irqsave(&ctrl->lock, flags);
1199 	if (ctrl->state == NVME_CTRL_LIVE ||
1200 	    ctrl->state == NVME_CTRL_CONNECTING)
1201 		startka = true;
1202 	spin_unlock_irqrestore(&ctrl->lock, flags);
1203 	if (startka)
1204 		nvme_queue_keep_alive_work(ctrl);
1205 	return RQ_END_IO_NONE;
1206 }
1207 
1208 static void nvme_keep_alive_work(struct work_struct *work)
1209 {
1210 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1211 			struct nvme_ctrl, ka_work);
1212 	bool comp_seen = ctrl->comp_seen;
1213 	struct request *rq;
1214 
1215 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1216 		dev_dbg(ctrl->device,
1217 			"reschedule traffic based keep-alive timer\n");
1218 		ctrl->comp_seen = false;
1219 		nvme_queue_keep_alive_work(ctrl);
1220 		return;
1221 	}
1222 
1223 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1224 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1225 	if (IS_ERR(rq)) {
1226 		/* allocation failure, reset the controller */
1227 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1228 		nvme_reset_ctrl(ctrl);
1229 		return;
1230 	}
1231 	nvme_init_request(rq, &ctrl->ka_cmd);
1232 
1233 	rq->timeout = ctrl->kato * HZ;
1234 	rq->end_io = nvme_keep_alive_end_io;
1235 	rq->end_io_data = ctrl;
1236 	blk_execute_rq_nowait(rq, false);
1237 }
1238 
1239 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1240 {
1241 	if (unlikely(ctrl->kato == 0))
1242 		return;
1243 
1244 	nvme_queue_keep_alive_work(ctrl);
1245 }
1246 
1247 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1248 {
1249 	if (unlikely(ctrl->kato == 0))
1250 		return;
1251 
1252 	cancel_delayed_work_sync(&ctrl->ka_work);
1253 }
1254 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1255 
1256 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1257 				   struct nvme_command *cmd)
1258 {
1259 	unsigned int new_kato =
1260 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1261 
1262 	dev_info(ctrl->device,
1263 		 "keep alive interval updated from %u ms to %u ms\n",
1264 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1265 
1266 	nvme_stop_keep_alive(ctrl);
1267 	ctrl->kato = new_kato;
1268 	nvme_start_keep_alive(ctrl);
1269 }
1270 
1271 /*
1272  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1273  * flag, thus sending any new CNS opcodes has a big chance of not working.
1274  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1275  * (but not for any later version).
1276  */
1277 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1278 {
1279 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1280 		return ctrl->vs < NVME_VS(1, 2, 0);
1281 	return ctrl->vs < NVME_VS(1, 1, 0);
1282 }
1283 
1284 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1285 {
1286 	struct nvme_command c = { };
1287 	int error;
1288 
1289 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1290 	c.identify.opcode = nvme_admin_identify;
1291 	c.identify.cns = NVME_ID_CNS_CTRL;
1292 
1293 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1294 	if (!*id)
1295 		return -ENOMEM;
1296 
1297 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1298 			sizeof(struct nvme_id_ctrl));
1299 	if (error)
1300 		kfree(*id);
1301 	return error;
1302 }
1303 
1304 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1305 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1306 {
1307 	const char *warn_str = "ctrl returned bogus length:";
1308 	void *data = cur;
1309 
1310 	switch (cur->nidt) {
1311 	case NVME_NIDT_EUI64:
1312 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1313 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1314 				 warn_str, cur->nidl);
1315 			return -1;
1316 		}
1317 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1318 			return NVME_NIDT_EUI64_LEN;
1319 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1320 		return NVME_NIDT_EUI64_LEN;
1321 	case NVME_NIDT_NGUID:
1322 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1323 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1324 				 warn_str, cur->nidl);
1325 			return -1;
1326 		}
1327 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1328 			return NVME_NIDT_NGUID_LEN;
1329 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1330 		return NVME_NIDT_NGUID_LEN;
1331 	case NVME_NIDT_UUID:
1332 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1333 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1334 				 warn_str, cur->nidl);
1335 			return -1;
1336 		}
1337 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1338 			return NVME_NIDT_UUID_LEN;
1339 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1340 		return NVME_NIDT_UUID_LEN;
1341 	case NVME_NIDT_CSI:
1342 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1343 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1344 				 warn_str, cur->nidl);
1345 			return -1;
1346 		}
1347 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1348 		*csi_seen = true;
1349 		return NVME_NIDT_CSI_LEN;
1350 	default:
1351 		/* Skip unknown types */
1352 		return cur->nidl;
1353 	}
1354 }
1355 
1356 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1357 		struct nvme_ns_info *info)
1358 {
1359 	struct nvme_command c = { };
1360 	bool csi_seen = false;
1361 	int status, pos, len;
1362 	void *data;
1363 
1364 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1365 		return 0;
1366 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1367 		return 0;
1368 
1369 	c.identify.opcode = nvme_admin_identify;
1370 	c.identify.nsid = cpu_to_le32(info->nsid);
1371 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1372 
1373 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1374 	if (!data)
1375 		return -ENOMEM;
1376 
1377 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1378 				      NVME_IDENTIFY_DATA_SIZE);
1379 	if (status) {
1380 		dev_warn(ctrl->device,
1381 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1382 			info->nsid, status);
1383 		goto free_data;
1384 	}
1385 
1386 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1387 		struct nvme_ns_id_desc *cur = data + pos;
1388 
1389 		if (cur->nidl == 0)
1390 			break;
1391 
1392 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1393 		if (len < 0)
1394 			break;
1395 
1396 		len += sizeof(*cur);
1397 	}
1398 
1399 	if (nvme_multi_css(ctrl) && !csi_seen) {
1400 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1401 			 info->nsid);
1402 		status = -EINVAL;
1403 	}
1404 
1405 free_data:
1406 	kfree(data);
1407 	return status;
1408 }
1409 
1410 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1411 			struct nvme_id_ns **id)
1412 {
1413 	struct nvme_command c = { };
1414 	int error;
1415 
1416 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1417 	c.identify.opcode = nvme_admin_identify;
1418 	c.identify.nsid = cpu_to_le32(nsid);
1419 	c.identify.cns = NVME_ID_CNS_NS;
1420 
1421 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1422 	if (!*id)
1423 		return -ENOMEM;
1424 
1425 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1426 	if (error) {
1427 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1428 		goto out_free_id;
1429 	}
1430 
1431 	error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1432 	if ((*id)->ncap == 0) /* namespace not allocated or attached */
1433 		goto out_free_id;
1434 	return 0;
1435 
1436 out_free_id:
1437 	kfree(*id);
1438 	return error;
1439 }
1440 
1441 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1442 		struct nvme_ns_info *info)
1443 {
1444 	struct nvme_ns_ids *ids = &info->ids;
1445 	struct nvme_id_ns *id;
1446 	int ret;
1447 
1448 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1449 	if (ret)
1450 		return ret;
1451 	info->anagrpid = id->anagrpid;
1452 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1453 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1454 	info->is_ready = true;
1455 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1456 		dev_info(ctrl->device,
1457 			 "Ignoring bogus Namespace Identifiers\n");
1458 	} else {
1459 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1460 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1461 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1462 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1463 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1464 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1465 	}
1466 	kfree(id);
1467 	return 0;
1468 }
1469 
1470 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1471 		struct nvme_ns_info *info)
1472 {
1473 	struct nvme_id_ns_cs_indep *id;
1474 	struct nvme_command c = {
1475 		.identify.opcode	= nvme_admin_identify,
1476 		.identify.nsid		= cpu_to_le32(info->nsid),
1477 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1478 	};
1479 	int ret;
1480 
1481 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1482 	if (!id)
1483 		return -ENOMEM;
1484 
1485 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1486 	if (!ret) {
1487 		info->anagrpid = id->anagrpid;
1488 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1489 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1490 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1491 	}
1492 	kfree(id);
1493 	return ret;
1494 }
1495 
1496 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1497 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1498 {
1499 	union nvme_result res = { 0 };
1500 	struct nvme_command c = { };
1501 	int ret;
1502 
1503 	c.features.opcode = op;
1504 	c.features.fid = cpu_to_le32(fid);
1505 	c.features.dword11 = cpu_to_le32(dword11);
1506 
1507 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1508 			buffer, buflen, NVME_QID_ANY, 0, 0);
1509 	if (ret >= 0 && result)
1510 		*result = le32_to_cpu(res.u32);
1511 	return ret;
1512 }
1513 
1514 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1515 		      unsigned int dword11, void *buffer, size_t buflen,
1516 		      u32 *result)
1517 {
1518 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1519 			     buflen, result);
1520 }
1521 EXPORT_SYMBOL_GPL(nvme_set_features);
1522 
1523 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1524 		      unsigned int dword11, void *buffer, size_t buflen,
1525 		      u32 *result)
1526 {
1527 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1528 			     buflen, result);
1529 }
1530 EXPORT_SYMBOL_GPL(nvme_get_features);
1531 
1532 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1533 {
1534 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1535 	u32 result;
1536 	int status, nr_io_queues;
1537 
1538 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1539 			&result);
1540 	if (status < 0)
1541 		return status;
1542 
1543 	/*
1544 	 * Degraded controllers might return an error when setting the queue
1545 	 * count.  We still want to be able to bring them online and offer
1546 	 * access to the admin queue, as that might be only way to fix them up.
1547 	 */
1548 	if (status > 0) {
1549 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1550 		*count = 0;
1551 	} else {
1552 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1553 		*count = min(*count, nr_io_queues);
1554 	}
1555 
1556 	return 0;
1557 }
1558 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1559 
1560 #define NVME_AEN_SUPPORTED \
1561 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1562 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1563 
1564 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1565 {
1566 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1567 	int status;
1568 
1569 	if (!supported_aens)
1570 		return;
1571 
1572 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1573 			NULL, 0, &result);
1574 	if (status)
1575 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1576 			 supported_aens);
1577 
1578 	queue_work(nvme_wq, &ctrl->async_event_work);
1579 }
1580 
1581 static int nvme_ns_open(struct nvme_ns *ns)
1582 {
1583 
1584 	/* should never be called due to GENHD_FL_HIDDEN */
1585 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1586 		goto fail;
1587 	if (!nvme_get_ns(ns))
1588 		goto fail;
1589 	if (!try_module_get(ns->ctrl->ops->module))
1590 		goto fail_put_ns;
1591 
1592 	return 0;
1593 
1594 fail_put_ns:
1595 	nvme_put_ns(ns);
1596 fail:
1597 	return -ENXIO;
1598 }
1599 
1600 static void nvme_ns_release(struct nvme_ns *ns)
1601 {
1602 
1603 	module_put(ns->ctrl->ops->module);
1604 	nvme_put_ns(ns);
1605 }
1606 
1607 static int nvme_open(struct block_device *bdev, fmode_t mode)
1608 {
1609 	return nvme_ns_open(bdev->bd_disk->private_data);
1610 }
1611 
1612 static void nvme_release(struct gendisk *disk, fmode_t mode)
1613 {
1614 	nvme_ns_release(disk->private_data);
1615 }
1616 
1617 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1618 {
1619 	/* some standard values */
1620 	geo->heads = 1 << 6;
1621 	geo->sectors = 1 << 5;
1622 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1623 	return 0;
1624 }
1625 
1626 #ifdef CONFIG_BLK_DEV_INTEGRITY
1627 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1628 				u32 max_integrity_segments)
1629 {
1630 	struct blk_integrity integrity = { };
1631 
1632 	switch (ns->pi_type) {
1633 	case NVME_NS_DPS_PI_TYPE3:
1634 		switch (ns->guard_type) {
1635 		case NVME_NVM_NS_16B_GUARD:
1636 			integrity.profile = &t10_pi_type3_crc;
1637 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1638 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1639 			break;
1640 		case NVME_NVM_NS_64B_GUARD:
1641 			integrity.profile = &ext_pi_type3_crc64;
1642 			integrity.tag_size = sizeof(u16) + 6;
1643 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1644 			break;
1645 		default:
1646 			integrity.profile = NULL;
1647 			break;
1648 		}
1649 		break;
1650 	case NVME_NS_DPS_PI_TYPE1:
1651 	case NVME_NS_DPS_PI_TYPE2:
1652 		switch (ns->guard_type) {
1653 		case NVME_NVM_NS_16B_GUARD:
1654 			integrity.profile = &t10_pi_type1_crc;
1655 			integrity.tag_size = sizeof(u16);
1656 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1657 			break;
1658 		case NVME_NVM_NS_64B_GUARD:
1659 			integrity.profile = &ext_pi_type1_crc64;
1660 			integrity.tag_size = sizeof(u16);
1661 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1662 			break;
1663 		default:
1664 			integrity.profile = NULL;
1665 			break;
1666 		}
1667 		break;
1668 	default:
1669 		integrity.profile = NULL;
1670 		break;
1671 	}
1672 
1673 	integrity.tuple_size = ns->ms;
1674 	blk_integrity_register(disk, &integrity);
1675 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1676 }
1677 #else
1678 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1679 				u32 max_integrity_segments)
1680 {
1681 }
1682 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1683 
1684 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1685 {
1686 	struct nvme_ctrl *ctrl = ns->ctrl;
1687 	struct request_queue *queue = disk->queue;
1688 	u32 size = queue_logical_block_size(queue);
1689 
1690 	if (ctrl->max_discard_sectors == 0) {
1691 		blk_queue_max_discard_sectors(queue, 0);
1692 		return;
1693 	}
1694 
1695 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1696 			NVME_DSM_MAX_RANGES);
1697 
1698 	queue->limits.discard_granularity = size;
1699 
1700 	/* If discard is already enabled, don't reset queue limits */
1701 	if (queue->limits.max_discard_sectors)
1702 		return;
1703 
1704 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1705 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1706 
1707 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1708 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1709 
1710 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1711 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1712 }
1713 
1714 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1715 {
1716 	return uuid_equal(&a->uuid, &b->uuid) &&
1717 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1718 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1719 		a->csi == b->csi;
1720 }
1721 
1722 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1723 {
1724 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1725 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1726 	struct nvme_ctrl *ctrl = ns->ctrl;
1727 	struct nvme_command c = { };
1728 	struct nvme_id_ns_nvm *nvm;
1729 	int ret = 0;
1730 	u32 elbaf;
1731 
1732 	ns->pi_size = 0;
1733 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1734 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1735 		ns->pi_size = sizeof(struct t10_pi_tuple);
1736 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1737 		goto set_pi;
1738 	}
1739 
1740 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1741 	if (!nvm)
1742 		return -ENOMEM;
1743 
1744 	c.identify.opcode = nvme_admin_identify;
1745 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1746 	c.identify.cns = NVME_ID_CNS_CS_NS;
1747 	c.identify.csi = NVME_CSI_NVM;
1748 
1749 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1750 	if (ret)
1751 		goto free_data;
1752 
1753 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1754 
1755 	/* no support for storage tag formats right now */
1756 	if (nvme_elbaf_sts(elbaf))
1757 		goto free_data;
1758 
1759 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1760 	switch (ns->guard_type) {
1761 	case NVME_NVM_NS_64B_GUARD:
1762 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1763 		break;
1764 	case NVME_NVM_NS_16B_GUARD:
1765 		ns->pi_size = sizeof(struct t10_pi_tuple);
1766 		break;
1767 	default:
1768 		break;
1769 	}
1770 
1771 free_data:
1772 	kfree(nvm);
1773 set_pi:
1774 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1775 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1776 	else
1777 		ns->pi_type = 0;
1778 
1779 	return ret;
1780 }
1781 
1782 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1783 {
1784 	struct nvme_ctrl *ctrl = ns->ctrl;
1785 
1786 	if (nvme_init_ms(ns, id))
1787 		return;
1788 
1789 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1790 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1791 		return;
1792 
1793 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1794 		/*
1795 		 * The NVMe over Fabrics specification only supports metadata as
1796 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1797 		 * remap the separate metadata buffer from the block layer.
1798 		 */
1799 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1800 			return;
1801 
1802 		ns->features |= NVME_NS_EXT_LBAS;
1803 
1804 		/*
1805 		 * The current fabrics transport drivers support namespace
1806 		 * metadata formats only if nvme_ns_has_pi() returns true.
1807 		 * Suppress support for all other formats so the namespace will
1808 		 * have a 0 capacity and not be usable through the block stack.
1809 		 *
1810 		 * Note, this check will need to be modified if any drivers
1811 		 * gain the ability to use other metadata formats.
1812 		 */
1813 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1814 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1815 	} else {
1816 		/*
1817 		 * For PCIe controllers, we can't easily remap the separate
1818 		 * metadata buffer from the block layer and thus require a
1819 		 * separate metadata buffer for block layer metadata/PI support.
1820 		 * We allow extended LBAs for the passthrough interface, though.
1821 		 */
1822 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1823 			ns->features |= NVME_NS_EXT_LBAS;
1824 		else
1825 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1826 	}
1827 }
1828 
1829 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1830 		struct request_queue *q)
1831 {
1832 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1833 
1834 	if (ctrl->max_hw_sectors) {
1835 		u32 max_segments =
1836 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1837 
1838 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1839 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1840 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1841 	}
1842 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1843 	blk_queue_dma_alignment(q, 3);
1844 	blk_queue_write_cache(q, vwc, vwc);
1845 }
1846 
1847 static void nvme_update_disk_info(struct gendisk *disk,
1848 		struct nvme_ns *ns, struct nvme_id_ns *id)
1849 {
1850 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1851 	unsigned short bs = 1 << ns->lba_shift;
1852 	u32 atomic_bs, phys_bs, io_opt = 0;
1853 
1854 	/*
1855 	 * The block layer can't support LBA sizes larger than the page size
1856 	 * yet, so catch this early and don't allow block I/O.
1857 	 */
1858 	if (ns->lba_shift > PAGE_SHIFT) {
1859 		capacity = 0;
1860 		bs = (1 << 9);
1861 	}
1862 
1863 	blk_integrity_unregister(disk);
1864 
1865 	atomic_bs = phys_bs = bs;
1866 	if (id->nabo == 0) {
1867 		/*
1868 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1869 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1870 		 * 0 then AWUPF must be used instead.
1871 		 */
1872 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1873 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1874 		else
1875 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1876 	}
1877 
1878 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1879 		/* NPWG = Namespace Preferred Write Granularity */
1880 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1881 		/* NOWS = Namespace Optimal Write Size */
1882 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1883 	}
1884 
1885 	blk_queue_logical_block_size(disk->queue, bs);
1886 	/*
1887 	 * Linux filesystems assume writing a single physical block is
1888 	 * an atomic operation. Hence limit the physical block size to the
1889 	 * value of the Atomic Write Unit Power Fail parameter.
1890 	 */
1891 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1892 	blk_queue_io_min(disk->queue, phys_bs);
1893 	blk_queue_io_opt(disk->queue, io_opt);
1894 
1895 	/*
1896 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1897 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1898 	 * I/O to namespaces with metadata except when the namespace supports
1899 	 * PI, as it can strip/insert in that case.
1900 	 */
1901 	if (ns->ms) {
1902 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1903 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1904 			nvme_init_integrity(disk, ns,
1905 					    ns->ctrl->max_integrity_segments);
1906 		else if (!nvme_ns_has_pi(ns))
1907 			capacity = 0;
1908 	}
1909 
1910 	set_capacity_and_notify(disk, capacity);
1911 
1912 	nvme_config_discard(disk, ns);
1913 	blk_queue_max_write_zeroes_sectors(disk->queue,
1914 					   ns->ctrl->max_zeroes_sectors);
1915 }
1916 
1917 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1918 {
1919 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1920 }
1921 
1922 static inline bool nvme_first_scan(struct gendisk *disk)
1923 {
1924 	/* nvme_alloc_ns() scans the disk prior to adding it */
1925 	return !disk_live(disk);
1926 }
1927 
1928 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1929 {
1930 	struct nvme_ctrl *ctrl = ns->ctrl;
1931 	u32 iob;
1932 
1933 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1934 	    is_power_of_2(ctrl->max_hw_sectors))
1935 		iob = ctrl->max_hw_sectors;
1936 	else
1937 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1938 
1939 	if (!iob)
1940 		return;
1941 
1942 	if (!is_power_of_2(iob)) {
1943 		if (nvme_first_scan(ns->disk))
1944 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1945 				ns->disk->disk_name, iob);
1946 		return;
1947 	}
1948 
1949 	if (blk_queue_is_zoned(ns->disk->queue)) {
1950 		if (nvme_first_scan(ns->disk))
1951 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1952 				ns->disk->disk_name);
1953 		return;
1954 	}
1955 
1956 	blk_queue_chunk_sectors(ns->queue, iob);
1957 }
1958 
1959 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1960 		struct nvme_ns_info *info)
1961 {
1962 	blk_mq_freeze_queue(ns->disk->queue);
1963 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1964 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1965 	blk_mq_unfreeze_queue(ns->disk->queue);
1966 
1967 	if (nvme_ns_head_multipath(ns->head)) {
1968 		blk_mq_freeze_queue(ns->head->disk->queue);
1969 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1970 		nvme_mpath_revalidate_paths(ns);
1971 		blk_stack_limits(&ns->head->disk->queue->limits,
1972 				 &ns->queue->limits, 0);
1973 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
1974 		blk_mq_unfreeze_queue(ns->head->disk->queue);
1975 	}
1976 
1977 	/* Hide the block-interface for these devices */
1978 	ns->disk->flags |= GENHD_FL_HIDDEN;
1979 	set_bit(NVME_NS_READY, &ns->flags);
1980 
1981 	return 0;
1982 }
1983 
1984 static int nvme_update_ns_info_block(struct nvme_ns *ns,
1985 		struct nvme_ns_info *info)
1986 {
1987 	struct nvme_id_ns *id;
1988 	unsigned lbaf;
1989 	int ret;
1990 
1991 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
1992 	if (ret)
1993 		return ret;
1994 
1995 	blk_mq_freeze_queue(ns->disk->queue);
1996 	lbaf = nvme_lbaf_index(id->flbas);
1997 	ns->lba_shift = id->lbaf[lbaf].ds;
1998 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1999 
2000 	nvme_configure_metadata(ns, id);
2001 	nvme_set_chunk_sectors(ns, id);
2002 	nvme_update_disk_info(ns->disk, ns, id);
2003 
2004 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2005 		ret = nvme_update_zone_info(ns, lbaf);
2006 		if (ret) {
2007 			blk_mq_unfreeze_queue(ns->disk->queue);
2008 			goto out;
2009 		}
2010 	}
2011 
2012 	/*
2013 	 * Only set the DEAC bit if the device guarantees that reads from
2014 	 * deallocated data return zeroes.  While the DEAC bit does not
2015 	 * require that, it must be a no-op if reads from deallocated data
2016 	 * do not return zeroes.
2017 	 */
2018 	if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2019 		ns->features |= NVME_NS_DEAC;
2020 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2021 	set_bit(NVME_NS_READY, &ns->flags);
2022 	blk_mq_unfreeze_queue(ns->disk->queue);
2023 
2024 	if (blk_queue_is_zoned(ns->queue)) {
2025 		ret = nvme_revalidate_zones(ns);
2026 		if (ret && !nvme_first_scan(ns->disk))
2027 			goto out;
2028 	}
2029 
2030 	if (nvme_ns_head_multipath(ns->head)) {
2031 		blk_mq_freeze_queue(ns->head->disk->queue);
2032 		nvme_update_disk_info(ns->head->disk, ns, id);
2033 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2034 		nvme_mpath_revalidate_paths(ns);
2035 		blk_stack_limits(&ns->head->disk->queue->limits,
2036 				 &ns->queue->limits, 0);
2037 		disk_update_readahead(ns->head->disk);
2038 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2039 	}
2040 
2041 	ret = 0;
2042 out:
2043 	/*
2044 	 * If probing fails due an unsupported feature, hide the block device,
2045 	 * but still allow other access.
2046 	 */
2047 	if (ret == -ENODEV) {
2048 		ns->disk->flags |= GENHD_FL_HIDDEN;
2049 		set_bit(NVME_NS_READY, &ns->flags);
2050 		ret = 0;
2051 	}
2052 	kfree(id);
2053 	return ret;
2054 }
2055 
2056 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2057 {
2058 	switch (info->ids.csi) {
2059 	case NVME_CSI_ZNS:
2060 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2061 			dev_info(ns->ctrl->device,
2062 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2063 				info->nsid);
2064 			return nvme_update_ns_info_generic(ns, info);
2065 		}
2066 		return nvme_update_ns_info_block(ns, info);
2067 	case NVME_CSI_NVM:
2068 		return nvme_update_ns_info_block(ns, info);
2069 	default:
2070 		dev_info(ns->ctrl->device,
2071 			"block device for nsid %u not supported (csi %u)\n",
2072 			info->nsid, info->ids.csi);
2073 		return nvme_update_ns_info_generic(ns, info);
2074 	}
2075 }
2076 
2077 static char nvme_pr_type(enum pr_type type)
2078 {
2079 	switch (type) {
2080 	case PR_WRITE_EXCLUSIVE:
2081 		return 1;
2082 	case PR_EXCLUSIVE_ACCESS:
2083 		return 2;
2084 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
2085 		return 3;
2086 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2087 		return 4;
2088 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
2089 		return 5;
2090 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2091 		return 6;
2092 	default:
2093 		return 0;
2094 	}
2095 }
2096 
2097 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
2098 		struct nvme_command *c, u8 data[16])
2099 {
2100 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
2101 	int srcu_idx = srcu_read_lock(&head->srcu);
2102 	struct nvme_ns *ns = nvme_find_path(head);
2103 	int ret = -EWOULDBLOCK;
2104 
2105 	if (ns) {
2106 		c->common.nsid = cpu_to_le32(ns->head->ns_id);
2107 		ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2108 	}
2109 	srcu_read_unlock(&head->srcu, srcu_idx);
2110 	return ret;
2111 }
2112 
2113 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2114 		u8 data[16])
2115 {
2116 	c->common.nsid = cpu_to_le32(ns->head->ns_id);
2117 	return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2118 }
2119 
2120 static int nvme_sc_to_pr_err(int nvme_sc)
2121 {
2122 	if (nvme_is_path_error(nvme_sc))
2123 		return PR_STS_PATH_FAILED;
2124 
2125 	switch (nvme_sc) {
2126 	case NVME_SC_SUCCESS:
2127 		return PR_STS_SUCCESS;
2128 	case NVME_SC_RESERVATION_CONFLICT:
2129 		return PR_STS_RESERVATION_CONFLICT;
2130 	case NVME_SC_ONCS_NOT_SUPPORTED:
2131 		return -EOPNOTSUPP;
2132 	case NVME_SC_BAD_ATTRIBUTES:
2133 	case NVME_SC_INVALID_OPCODE:
2134 	case NVME_SC_INVALID_FIELD:
2135 	case NVME_SC_INVALID_NS:
2136 		return -EINVAL;
2137 	default:
2138 		return PR_STS_IOERR;
2139 	}
2140 }
2141 
2142 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2143 				u64 key, u64 sa_key, u8 op)
2144 {
2145 	struct nvme_command c = { };
2146 	u8 data[16] = { 0, };
2147 	int ret;
2148 
2149 	put_unaligned_le64(key, &data[0]);
2150 	put_unaligned_le64(sa_key, &data[8]);
2151 
2152 	c.common.opcode = op;
2153 	c.common.cdw10 = cpu_to_le32(cdw10);
2154 
2155 	if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2156 	    bdev->bd_disk->fops == &nvme_ns_head_ops)
2157 		ret = nvme_send_ns_head_pr_command(bdev, &c, data);
2158 	else
2159 		ret = nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c,
2160 					      data);
2161 	if (ret < 0)
2162 		return ret;
2163 
2164 	return nvme_sc_to_pr_err(ret);
2165 }
2166 
2167 static int nvme_pr_register(struct block_device *bdev, u64 old,
2168 		u64 new, unsigned flags)
2169 {
2170 	u32 cdw10;
2171 
2172 	if (flags & ~PR_FL_IGNORE_KEY)
2173 		return -EOPNOTSUPP;
2174 
2175 	cdw10 = old ? 2 : 0;
2176 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2177 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2178 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2179 }
2180 
2181 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2182 		enum pr_type type, unsigned flags)
2183 {
2184 	u32 cdw10;
2185 
2186 	if (flags & ~PR_FL_IGNORE_KEY)
2187 		return -EOPNOTSUPP;
2188 
2189 	cdw10 = nvme_pr_type(type) << 8;
2190 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2191 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2192 }
2193 
2194 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2195 		enum pr_type type, bool abort)
2196 {
2197 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2198 
2199 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2200 }
2201 
2202 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2203 {
2204 	u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2205 
2206 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2207 }
2208 
2209 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2210 {
2211 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2212 
2213 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2214 }
2215 
2216 const struct pr_ops nvme_pr_ops = {
2217 	.pr_register	= nvme_pr_register,
2218 	.pr_reserve	= nvme_pr_reserve,
2219 	.pr_release	= nvme_pr_release,
2220 	.pr_preempt	= nvme_pr_preempt,
2221 	.pr_clear	= nvme_pr_clear,
2222 };
2223 
2224 #ifdef CONFIG_BLK_SED_OPAL
2225 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2226 		bool send)
2227 {
2228 	struct nvme_ctrl *ctrl = data;
2229 	struct nvme_command cmd = { };
2230 
2231 	if (send)
2232 		cmd.common.opcode = nvme_admin_security_send;
2233 	else
2234 		cmd.common.opcode = nvme_admin_security_recv;
2235 	cmd.common.nsid = 0;
2236 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2237 	cmd.common.cdw11 = cpu_to_le32(len);
2238 
2239 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2240 			NVME_QID_ANY, 1, 0);
2241 }
2242 
2243 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2244 {
2245 	if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2246 		if (!ctrl->opal_dev)
2247 			ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2248 		else if (was_suspended)
2249 			opal_unlock_from_suspend(ctrl->opal_dev);
2250 	} else {
2251 		free_opal_dev(ctrl->opal_dev);
2252 		ctrl->opal_dev = NULL;
2253 	}
2254 }
2255 #else
2256 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2257 {
2258 }
2259 #endif /* CONFIG_BLK_SED_OPAL */
2260 
2261 #ifdef CONFIG_BLK_DEV_ZONED
2262 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2263 		unsigned int nr_zones, report_zones_cb cb, void *data)
2264 {
2265 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2266 			data);
2267 }
2268 #else
2269 #define nvme_report_zones	NULL
2270 #endif /* CONFIG_BLK_DEV_ZONED */
2271 
2272 static const struct block_device_operations nvme_bdev_ops = {
2273 	.owner		= THIS_MODULE,
2274 	.ioctl		= nvme_ioctl,
2275 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2276 	.open		= nvme_open,
2277 	.release	= nvme_release,
2278 	.getgeo		= nvme_getgeo,
2279 	.report_zones	= nvme_report_zones,
2280 	.pr_ops		= &nvme_pr_ops,
2281 };
2282 
2283 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2284 		u32 timeout, const char *op)
2285 {
2286 	unsigned long timeout_jiffies = jiffies + timeout * HZ;
2287 	u32 csts;
2288 	int ret;
2289 
2290 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2291 		if (csts == ~0)
2292 			return -ENODEV;
2293 		if ((csts & mask) == val)
2294 			break;
2295 
2296 		usleep_range(1000, 2000);
2297 		if (fatal_signal_pending(current))
2298 			return -EINTR;
2299 		if (time_after(jiffies, timeout_jiffies)) {
2300 			dev_err(ctrl->device,
2301 				"Device not ready; aborting %s, CSTS=0x%x\n",
2302 				op, csts);
2303 			return -ENODEV;
2304 		}
2305 	}
2306 
2307 	return ret;
2308 }
2309 
2310 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2311 {
2312 	int ret;
2313 
2314 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2315 	if (shutdown)
2316 		ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2317 	else
2318 		ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2319 
2320 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2321 	if (ret)
2322 		return ret;
2323 
2324 	if (shutdown) {
2325 		return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2326 				       NVME_CSTS_SHST_CMPLT,
2327 				       ctrl->shutdown_timeout, "shutdown");
2328 	}
2329 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2330 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2331 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2332 			       (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2333 }
2334 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2335 
2336 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2337 {
2338 	unsigned dev_page_min;
2339 	u32 timeout;
2340 	int ret;
2341 
2342 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2343 	if (ret) {
2344 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2345 		return ret;
2346 	}
2347 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2348 
2349 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2350 		dev_err(ctrl->device,
2351 			"Minimum device page size %u too large for host (%u)\n",
2352 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2353 		return -ENODEV;
2354 	}
2355 
2356 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2357 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2358 	else
2359 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2360 
2361 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2362 		u32 crto;
2363 
2364 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2365 		if (ret) {
2366 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2367 				ret);
2368 			return ret;
2369 		}
2370 
2371 		if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2372 			ctrl->ctrl_config |= NVME_CC_CRIME;
2373 			timeout = NVME_CRTO_CRIMT(crto);
2374 		} else {
2375 			timeout = NVME_CRTO_CRWMT(crto);
2376 		}
2377 	} else {
2378 		timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2379 	}
2380 
2381 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2382 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2383 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2384 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2385 	if (ret)
2386 		return ret;
2387 
2388 	/* Flush write to device (required if transport is PCI) */
2389 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2390 	if (ret)
2391 		return ret;
2392 
2393 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2394 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2395 	if (ret)
2396 		return ret;
2397 	return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2398 			       (timeout + 1) / 2, "initialisation");
2399 }
2400 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2401 
2402 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2403 {
2404 	__le64 ts;
2405 	int ret;
2406 
2407 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2408 		return 0;
2409 
2410 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2411 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2412 			NULL);
2413 	if (ret)
2414 		dev_warn_once(ctrl->device,
2415 			"could not set timestamp (%d)\n", ret);
2416 	return ret;
2417 }
2418 
2419 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2420 {
2421 	struct nvme_feat_host_behavior *host;
2422 	u8 acre = 0, lbafee = 0;
2423 	int ret;
2424 
2425 	/* Don't bother enabling the feature if retry delay is not reported */
2426 	if (ctrl->crdt[0])
2427 		acre = NVME_ENABLE_ACRE;
2428 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2429 		lbafee = NVME_ENABLE_LBAFEE;
2430 
2431 	if (!acre && !lbafee)
2432 		return 0;
2433 
2434 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2435 	if (!host)
2436 		return 0;
2437 
2438 	host->acre = acre;
2439 	host->lbafee = lbafee;
2440 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2441 				host, sizeof(*host), NULL);
2442 	kfree(host);
2443 	return ret;
2444 }
2445 
2446 /*
2447  * The function checks whether the given total (exlat + enlat) latency of
2448  * a power state allows the latter to be used as an APST transition target.
2449  * It does so by comparing the latency to the primary and secondary latency
2450  * tolerances defined by module params. If there's a match, the corresponding
2451  * timeout value is returned and the matching tolerance index (1 or 2) is
2452  * reported.
2453  */
2454 static bool nvme_apst_get_transition_time(u64 total_latency,
2455 		u64 *transition_time, unsigned *last_index)
2456 {
2457 	if (total_latency <= apst_primary_latency_tol_us) {
2458 		if (*last_index == 1)
2459 			return false;
2460 		*last_index = 1;
2461 		*transition_time = apst_primary_timeout_ms;
2462 		return true;
2463 	}
2464 	if (apst_secondary_timeout_ms &&
2465 		total_latency <= apst_secondary_latency_tol_us) {
2466 		if (*last_index <= 2)
2467 			return false;
2468 		*last_index = 2;
2469 		*transition_time = apst_secondary_timeout_ms;
2470 		return true;
2471 	}
2472 	return false;
2473 }
2474 
2475 /*
2476  * APST (Autonomous Power State Transition) lets us program a table of power
2477  * state transitions that the controller will perform automatically.
2478  *
2479  * Depending on module params, one of the two supported techniques will be used:
2480  *
2481  * - If the parameters provide explicit timeouts and tolerances, they will be
2482  *   used to build a table with up to 2 non-operational states to transition to.
2483  *   The default parameter values were selected based on the values used by
2484  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2485  *   regeneration of the APST table in the event of switching between external
2486  *   and battery power, the timeouts and tolerances reflect a compromise
2487  *   between values used by Microsoft for AC and battery scenarios.
2488  * - If not, we'll configure the table with a simple heuristic: we are willing
2489  *   to spend at most 2% of the time transitioning between power states.
2490  *   Therefore, when running in any given state, we will enter the next
2491  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2492  *   microseconds, as long as that state's exit latency is under the requested
2493  *   maximum latency.
2494  *
2495  * We will not autonomously enter any non-operational state for which the total
2496  * latency exceeds ps_max_latency_us.
2497  *
2498  * Users can set ps_max_latency_us to zero to turn off APST.
2499  */
2500 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2501 {
2502 	struct nvme_feat_auto_pst *table;
2503 	unsigned apste = 0;
2504 	u64 max_lat_us = 0;
2505 	__le64 target = 0;
2506 	int max_ps = -1;
2507 	int state;
2508 	int ret;
2509 	unsigned last_lt_index = UINT_MAX;
2510 
2511 	/*
2512 	 * If APST isn't supported or if we haven't been initialized yet,
2513 	 * then don't do anything.
2514 	 */
2515 	if (!ctrl->apsta)
2516 		return 0;
2517 
2518 	if (ctrl->npss > 31) {
2519 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2520 		return 0;
2521 	}
2522 
2523 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2524 	if (!table)
2525 		return 0;
2526 
2527 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2528 		/* Turn off APST. */
2529 		dev_dbg(ctrl->device, "APST disabled\n");
2530 		goto done;
2531 	}
2532 
2533 	/*
2534 	 * Walk through all states from lowest- to highest-power.
2535 	 * According to the spec, lower-numbered states use more power.  NPSS,
2536 	 * despite the name, is the index of the lowest-power state, not the
2537 	 * number of states.
2538 	 */
2539 	for (state = (int)ctrl->npss; state >= 0; state--) {
2540 		u64 total_latency_us, exit_latency_us, transition_ms;
2541 
2542 		if (target)
2543 			table->entries[state] = target;
2544 
2545 		/*
2546 		 * Don't allow transitions to the deepest state if it's quirked
2547 		 * off.
2548 		 */
2549 		if (state == ctrl->npss &&
2550 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2551 			continue;
2552 
2553 		/*
2554 		 * Is this state a useful non-operational state for higher-power
2555 		 * states to autonomously transition to?
2556 		 */
2557 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2558 			continue;
2559 
2560 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2561 		if (exit_latency_us > ctrl->ps_max_latency_us)
2562 			continue;
2563 
2564 		total_latency_us = exit_latency_us +
2565 			le32_to_cpu(ctrl->psd[state].entry_lat);
2566 
2567 		/*
2568 		 * This state is good. It can be used as the APST idle target
2569 		 * for higher power states.
2570 		 */
2571 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2572 			if (!nvme_apst_get_transition_time(total_latency_us,
2573 					&transition_ms, &last_lt_index))
2574 				continue;
2575 		} else {
2576 			transition_ms = total_latency_us + 19;
2577 			do_div(transition_ms, 20);
2578 			if (transition_ms > (1 << 24) - 1)
2579 				transition_ms = (1 << 24) - 1;
2580 		}
2581 
2582 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2583 		if (max_ps == -1)
2584 			max_ps = state;
2585 		if (total_latency_us > max_lat_us)
2586 			max_lat_us = total_latency_us;
2587 	}
2588 
2589 	if (max_ps == -1)
2590 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2591 	else
2592 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2593 			max_ps, max_lat_us, (int)sizeof(*table), table);
2594 	apste = 1;
2595 
2596 done:
2597 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2598 				table, sizeof(*table), NULL);
2599 	if (ret)
2600 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2601 	kfree(table);
2602 	return ret;
2603 }
2604 
2605 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2606 {
2607 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2608 	u64 latency;
2609 
2610 	switch (val) {
2611 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2612 	case PM_QOS_LATENCY_ANY:
2613 		latency = U64_MAX;
2614 		break;
2615 
2616 	default:
2617 		latency = val;
2618 	}
2619 
2620 	if (ctrl->ps_max_latency_us != latency) {
2621 		ctrl->ps_max_latency_us = latency;
2622 		if (ctrl->state == NVME_CTRL_LIVE)
2623 			nvme_configure_apst(ctrl);
2624 	}
2625 }
2626 
2627 struct nvme_core_quirk_entry {
2628 	/*
2629 	 * NVMe model and firmware strings are padded with spaces.  For
2630 	 * simplicity, strings in the quirk table are padded with NULLs
2631 	 * instead.
2632 	 */
2633 	u16 vid;
2634 	const char *mn;
2635 	const char *fr;
2636 	unsigned long quirks;
2637 };
2638 
2639 static const struct nvme_core_quirk_entry core_quirks[] = {
2640 	{
2641 		/*
2642 		 * This Toshiba device seems to die using any APST states.  See:
2643 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2644 		 */
2645 		.vid = 0x1179,
2646 		.mn = "THNSF5256GPUK TOSHIBA",
2647 		.quirks = NVME_QUIRK_NO_APST,
2648 	},
2649 	{
2650 		/*
2651 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2652 		 * condition associated with actions related to suspend to idle
2653 		 * LiteON has resolved the problem in future firmware
2654 		 */
2655 		.vid = 0x14a4,
2656 		.fr = "22301111",
2657 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2658 	},
2659 	{
2660 		/*
2661 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2662 		 * aborts I/O during any load, but more easily reproducible
2663 		 * with discards (fstrim).
2664 		 *
2665 		 * The device is left in a state where it is also not possible
2666 		 * to use "nvme set-feature" to disable APST, but booting with
2667 		 * nvme_core.default_ps_max_latency=0 works.
2668 		 */
2669 		.vid = 0x1e0f,
2670 		.mn = "KCD6XVUL6T40",
2671 		.quirks = NVME_QUIRK_NO_APST,
2672 	},
2673 	{
2674 		/*
2675 		 * The external Samsung X5 SSD fails initialization without a
2676 		 * delay before checking if it is ready and has a whole set of
2677 		 * other problems.  To make this even more interesting, it
2678 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2679 		 * does not need or want these quirks.
2680 		 */
2681 		.vid = 0x144d,
2682 		.mn = "Samsung Portable SSD X5",
2683 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2684 			  NVME_QUIRK_NO_DEEPEST_PS |
2685 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2686 	}
2687 };
2688 
2689 /* match is null-terminated but idstr is space-padded. */
2690 static bool string_matches(const char *idstr, const char *match, size_t len)
2691 {
2692 	size_t matchlen;
2693 
2694 	if (!match)
2695 		return true;
2696 
2697 	matchlen = strlen(match);
2698 	WARN_ON_ONCE(matchlen > len);
2699 
2700 	if (memcmp(idstr, match, matchlen))
2701 		return false;
2702 
2703 	for (; matchlen < len; matchlen++)
2704 		if (idstr[matchlen] != ' ')
2705 			return false;
2706 
2707 	return true;
2708 }
2709 
2710 static bool quirk_matches(const struct nvme_id_ctrl *id,
2711 			  const struct nvme_core_quirk_entry *q)
2712 {
2713 	return q->vid == le16_to_cpu(id->vid) &&
2714 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2715 		string_matches(id->fr, q->fr, sizeof(id->fr));
2716 }
2717 
2718 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2719 		struct nvme_id_ctrl *id)
2720 {
2721 	size_t nqnlen;
2722 	int off;
2723 
2724 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2725 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2726 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2727 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2728 			return;
2729 		}
2730 
2731 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2732 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2733 	}
2734 
2735 	/*
2736 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2737 	 * Base Specification 2.0.  It is slightly different from the format
2738 	 * specified there due to historic reasons, and we can't change it now.
2739 	 */
2740 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2741 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2742 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2743 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2744 	off += sizeof(id->sn);
2745 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2746 	off += sizeof(id->mn);
2747 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2748 }
2749 
2750 static void nvme_release_subsystem(struct device *dev)
2751 {
2752 	struct nvme_subsystem *subsys =
2753 		container_of(dev, struct nvme_subsystem, dev);
2754 
2755 	if (subsys->instance >= 0)
2756 		ida_free(&nvme_instance_ida, subsys->instance);
2757 	kfree(subsys);
2758 }
2759 
2760 static void nvme_destroy_subsystem(struct kref *ref)
2761 {
2762 	struct nvme_subsystem *subsys =
2763 			container_of(ref, struct nvme_subsystem, ref);
2764 
2765 	mutex_lock(&nvme_subsystems_lock);
2766 	list_del(&subsys->entry);
2767 	mutex_unlock(&nvme_subsystems_lock);
2768 
2769 	ida_destroy(&subsys->ns_ida);
2770 	device_del(&subsys->dev);
2771 	put_device(&subsys->dev);
2772 }
2773 
2774 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2775 {
2776 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2777 }
2778 
2779 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2780 {
2781 	struct nvme_subsystem *subsys;
2782 
2783 	lockdep_assert_held(&nvme_subsystems_lock);
2784 
2785 	/*
2786 	 * Fail matches for discovery subsystems. This results
2787 	 * in each discovery controller bound to a unique subsystem.
2788 	 * This avoids issues with validating controller values
2789 	 * that can only be true when there is a single unique subsystem.
2790 	 * There may be multiple and completely independent entities
2791 	 * that provide discovery controllers.
2792 	 */
2793 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2794 		return NULL;
2795 
2796 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2797 		if (strcmp(subsys->subnqn, subsysnqn))
2798 			continue;
2799 		if (!kref_get_unless_zero(&subsys->ref))
2800 			continue;
2801 		return subsys;
2802 	}
2803 
2804 	return NULL;
2805 }
2806 
2807 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2808 	struct device_attribute subsys_attr_##_name = \
2809 		__ATTR(_name, _mode, _show, NULL)
2810 
2811 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2812 				    struct device_attribute *attr,
2813 				    char *buf)
2814 {
2815 	struct nvme_subsystem *subsys =
2816 		container_of(dev, struct nvme_subsystem, dev);
2817 
2818 	return sysfs_emit(buf, "%s\n", subsys->subnqn);
2819 }
2820 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2821 
2822 static ssize_t nvme_subsys_show_type(struct device *dev,
2823 				    struct device_attribute *attr,
2824 				    char *buf)
2825 {
2826 	struct nvme_subsystem *subsys =
2827 		container_of(dev, struct nvme_subsystem, dev);
2828 
2829 	switch (subsys->subtype) {
2830 	case NVME_NQN_DISC:
2831 		return sysfs_emit(buf, "discovery\n");
2832 	case NVME_NQN_NVME:
2833 		return sysfs_emit(buf, "nvm\n");
2834 	default:
2835 		return sysfs_emit(buf, "reserved\n");
2836 	}
2837 }
2838 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2839 
2840 #define nvme_subsys_show_str_function(field)				\
2841 static ssize_t subsys_##field##_show(struct device *dev,		\
2842 			    struct device_attribute *attr, char *buf)	\
2843 {									\
2844 	struct nvme_subsystem *subsys =					\
2845 		container_of(dev, struct nvme_subsystem, dev);		\
2846 	return sysfs_emit(buf, "%.*s\n",				\
2847 			   (int)sizeof(subsys->field), subsys->field);	\
2848 }									\
2849 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2850 
2851 nvme_subsys_show_str_function(model);
2852 nvme_subsys_show_str_function(serial);
2853 nvme_subsys_show_str_function(firmware_rev);
2854 
2855 static struct attribute *nvme_subsys_attrs[] = {
2856 	&subsys_attr_model.attr,
2857 	&subsys_attr_serial.attr,
2858 	&subsys_attr_firmware_rev.attr,
2859 	&subsys_attr_subsysnqn.attr,
2860 	&subsys_attr_subsystype.attr,
2861 #ifdef CONFIG_NVME_MULTIPATH
2862 	&subsys_attr_iopolicy.attr,
2863 #endif
2864 	NULL,
2865 };
2866 
2867 static const struct attribute_group nvme_subsys_attrs_group = {
2868 	.attrs = nvme_subsys_attrs,
2869 };
2870 
2871 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2872 	&nvme_subsys_attrs_group,
2873 	NULL,
2874 };
2875 
2876 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2877 {
2878 	return ctrl->opts && ctrl->opts->discovery_nqn;
2879 }
2880 
2881 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2882 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2883 {
2884 	struct nvme_ctrl *tmp;
2885 
2886 	lockdep_assert_held(&nvme_subsystems_lock);
2887 
2888 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2889 		if (nvme_state_terminal(tmp))
2890 			continue;
2891 
2892 		if (tmp->cntlid == ctrl->cntlid) {
2893 			dev_err(ctrl->device,
2894 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2895 				ctrl->cntlid, dev_name(tmp->device),
2896 				subsys->subnqn);
2897 			return false;
2898 		}
2899 
2900 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2901 		    nvme_discovery_ctrl(ctrl))
2902 			continue;
2903 
2904 		dev_err(ctrl->device,
2905 			"Subsystem does not support multiple controllers\n");
2906 		return false;
2907 	}
2908 
2909 	return true;
2910 }
2911 
2912 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2913 {
2914 	struct nvme_subsystem *subsys, *found;
2915 	int ret;
2916 
2917 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2918 	if (!subsys)
2919 		return -ENOMEM;
2920 
2921 	subsys->instance = -1;
2922 	mutex_init(&subsys->lock);
2923 	kref_init(&subsys->ref);
2924 	INIT_LIST_HEAD(&subsys->ctrls);
2925 	INIT_LIST_HEAD(&subsys->nsheads);
2926 	nvme_init_subnqn(subsys, ctrl, id);
2927 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2928 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2929 	subsys->vendor_id = le16_to_cpu(id->vid);
2930 	subsys->cmic = id->cmic;
2931 
2932 	/* Versions prior to 1.4 don't necessarily report a valid type */
2933 	if (id->cntrltype == NVME_CTRL_DISC ||
2934 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2935 		subsys->subtype = NVME_NQN_DISC;
2936 	else
2937 		subsys->subtype = NVME_NQN_NVME;
2938 
2939 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2940 		dev_err(ctrl->device,
2941 			"Subsystem %s is not a discovery controller",
2942 			subsys->subnqn);
2943 		kfree(subsys);
2944 		return -EINVAL;
2945 	}
2946 	subsys->awupf = le16_to_cpu(id->awupf);
2947 	nvme_mpath_default_iopolicy(subsys);
2948 
2949 	subsys->dev.class = nvme_subsys_class;
2950 	subsys->dev.release = nvme_release_subsystem;
2951 	subsys->dev.groups = nvme_subsys_attrs_groups;
2952 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2953 	device_initialize(&subsys->dev);
2954 
2955 	mutex_lock(&nvme_subsystems_lock);
2956 	found = __nvme_find_get_subsystem(subsys->subnqn);
2957 	if (found) {
2958 		put_device(&subsys->dev);
2959 		subsys = found;
2960 
2961 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2962 			ret = -EINVAL;
2963 			goto out_put_subsystem;
2964 		}
2965 	} else {
2966 		ret = device_add(&subsys->dev);
2967 		if (ret) {
2968 			dev_err(ctrl->device,
2969 				"failed to register subsystem device.\n");
2970 			put_device(&subsys->dev);
2971 			goto out_unlock;
2972 		}
2973 		ida_init(&subsys->ns_ida);
2974 		list_add_tail(&subsys->entry, &nvme_subsystems);
2975 	}
2976 
2977 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2978 				dev_name(ctrl->device));
2979 	if (ret) {
2980 		dev_err(ctrl->device,
2981 			"failed to create sysfs link from subsystem.\n");
2982 		goto out_put_subsystem;
2983 	}
2984 
2985 	if (!found)
2986 		subsys->instance = ctrl->instance;
2987 	ctrl->subsys = subsys;
2988 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2989 	mutex_unlock(&nvme_subsystems_lock);
2990 	return 0;
2991 
2992 out_put_subsystem:
2993 	nvme_put_subsystem(subsys);
2994 out_unlock:
2995 	mutex_unlock(&nvme_subsystems_lock);
2996 	return ret;
2997 }
2998 
2999 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3000 		void *log, size_t size, u64 offset)
3001 {
3002 	struct nvme_command c = { };
3003 	u32 dwlen = nvme_bytes_to_numd(size);
3004 
3005 	c.get_log_page.opcode = nvme_admin_get_log_page;
3006 	c.get_log_page.nsid = cpu_to_le32(nsid);
3007 	c.get_log_page.lid = log_page;
3008 	c.get_log_page.lsp = lsp;
3009 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3010 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3011 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3012 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3013 	c.get_log_page.csi = csi;
3014 
3015 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3016 }
3017 
3018 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3019 				struct nvme_effects_log **log)
3020 {
3021 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
3022 	int ret;
3023 
3024 	if (cel)
3025 		goto out;
3026 
3027 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3028 	if (!cel)
3029 		return -ENOMEM;
3030 
3031 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3032 			cel, sizeof(*cel), 0);
3033 	if (ret) {
3034 		kfree(cel);
3035 		return ret;
3036 	}
3037 
3038 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3039 out:
3040 	*log = cel;
3041 	return 0;
3042 }
3043 
3044 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3045 {
3046 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3047 
3048 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
3049 		return UINT_MAX;
3050 	return val;
3051 }
3052 
3053 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3054 {
3055 	struct nvme_command c = { };
3056 	struct nvme_id_ctrl_nvm *id;
3057 	int ret;
3058 
3059 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3060 		ctrl->max_discard_sectors = UINT_MAX;
3061 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3062 	} else {
3063 		ctrl->max_discard_sectors = 0;
3064 		ctrl->max_discard_segments = 0;
3065 	}
3066 
3067 	/*
3068 	 * Even though NVMe spec explicitly states that MDTS is not applicable
3069 	 * to the write-zeroes, we are cautious and limit the size to the
3070 	 * controllers max_hw_sectors value, which is based on the MDTS field
3071 	 * and possibly other limiting factors.
3072 	 */
3073 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3074 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3075 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3076 	else
3077 		ctrl->max_zeroes_sectors = 0;
3078 
3079 	if (nvme_ctrl_limited_cns(ctrl))
3080 		return 0;
3081 
3082 	id = kzalloc(sizeof(*id), GFP_KERNEL);
3083 	if (!id)
3084 		return -ENOMEM;
3085 
3086 	c.identify.opcode = nvme_admin_identify;
3087 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
3088 	c.identify.csi = NVME_CSI_NVM;
3089 
3090 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3091 	if (ret)
3092 		goto free_data;
3093 
3094 	if (id->dmrl)
3095 		ctrl->max_discard_segments = id->dmrl;
3096 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3097 	if (id->wzsl)
3098 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3099 
3100 free_data:
3101 	kfree(id);
3102 	return ret;
3103 }
3104 
3105 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3106 {
3107 	struct nvme_id_ctrl *id;
3108 	u32 max_hw_sectors;
3109 	bool prev_apst_enabled;
3110 	int ret;
3111 
3112 	ret = nvme_identify_ctrl(ctrl, &id);
3113 	if (ret) {
3114 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3115 		return -EIO;
3116 	}
3117 
3118 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3119 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3120 		if (ret < 0)
3121 			goto out_free;
3122 	}
3123 
3124 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3125 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3126 
3127 	if (!ctrl->identified) {
3128 		unsigned int i;
3129 
3130 		/*
3131 		 * Check for quirks.  Quirk can depend on firmware version,
3132 		 * so, in principle, the set of quirks present can change
3133 		 * across a reset.  As a possible future enhancement, we
3134 		 * could re-scan for quirks every time we reinitialize
3135 		 * the device, but we'd have to make sure that the driver
3136 		 * behaves intelligently if the quirks change.
3137 		 */
3138 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3139 			if (quirk_matches(id, &core_quirks[i]))
3140 				ctrl->quirks |= core_quirks[i].quirks;
3141 		}
3142 
3143 		ret = nvme_init_subsystem(ctrl, id);
3144 		if (ret)
3145 			goto out_free;
3146 	}
3147 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3148 	       sizeof(ctrl->subsys->firmware_rev));
3149 
3150 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3151 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3152 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3153 	}
3154 
3155 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3156 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3157 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3158 
3159 	ctrl->oacs = le16_to_cpu(id->oacs);
3160 	ctrl->oncs = le16_to_cpu(id->oncs);
3161 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3162 	ctrl->oaes = le32_to_cpu(id->oaes);
3163 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3164 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3165 
3166 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3167 	ctrl->vwc = id->vwc;
3168 	if (id->mdts)
3169 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3170 	else
3171 		max_hw_sectors = UINT_MAX;
3172 	ctrl->max_hw_sectors =
3173 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3174 
3175 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3176 	ctrl->sgls = le32_to_cpu(id->sgls);
3177 	ctrl->kas = le16_to_cpu(id->kas);
3178 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3179 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3180 
3181 	ctrl->cntrltype = id->cntrltype;
3182 	ctrl->dctype = id->dctype;
3183 
3184 	if (id->rtd3e) {
3185 		/* us -> s */
3186 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3187 
3188 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3189 						 shutdown_timeout, 60);
3190 
3191 		if (ctrl->shutdown_timeout != shutdown_timeout)
3192 			dev_info(ctrl->device,
3193 				 "Shutdown timeout set to %u seconds\n",
3194 				 ctrl->shutdown_timeout);
3195 	} else
3196 		ctrl->shutdown_timeout = shutdown_timeout;
3197 
3198 	ctrl->npss = id->npss;
3199 	ctrl->apsta = id->apsta;
3200 	prev_apst_enabled = ctrl->apst_enabled;
3201 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3202 		if (force_apst && id->apsta) {
3203 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3204 			ctrl->apst_enabled = true;
3205 		} else {
3206 			ctrl->apst_enabled = false;
3207 		}
3208 	} else {
3209 		ctrl->apst_enabled = id->apsta;
3210 	}
3211 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3212 
3213 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3214 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3215 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3216 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3217 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3218 
3219 		/*
3220 		 * In fabrics we need to verify the cntlid matches the
3221 		 * admin connect
3222 		 */
3223 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3224 			dev_err(ctrl->device,
3225 				"Mismatching cntlid: Connect %u vs Identify "
3226 				"%u, rejecting\n",
3227 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3228 			ret = -EINVAL;
3229 			goto out_free;
3230 		}
3231 
3232 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3233 			dev_err(ctrl->device,
3234 				"keep-alive support is mandatory for fabrics\n");
3235 			ret = -EINVAL;
3236 			goto out_free;
3237 		}
3238 	} else {
3239 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3240 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3241 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3242 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3243 	}
3244 
3245 	ret = nvme_mpath_init_identify(ctrl, id);
3246 	if (ret < 0)
3247 		goto out_free;
3248 
3249 	if (ctrl->apst_enabled && !prev_apst_enabled)
3250 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3251 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3252 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3253 
3254 out_free:
3255 	kfree(id);
3256 	return ret;
3257 }
3258 
3259 /*
3260  * Initialize the cached copies of the Identify data and various controller
3261  * register in our nvme_ctrl structure.  This should be called as soon as
3262  * the admin queue is fully up and running.
3263  */
3264 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3265 {
3266 	int ret;
3267 
3268 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3269 	if (ret) {
3270 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3271 		return ret;
3272 	}
3273 
3274 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3275 
3276 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3277 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3278 
3279 	ret = nvme_init_identify(ctrl);
3280 	if (ret)
3281 		return ret;
3282 
3283 	ret = nvme_configure_apst(ctrl);
3284 	if (ret < 0)
3285 		return ret;
3286 
3287 	ret = nvme_configure_timestamp(ctrl);
3288 	if (ret < 0)
3289 		return ret;
3290 
3291 	ret = nvme_configure_host_options(ctrl);
3292 	if (ret < 0)
3293 		return ret;
3294 
3295 	nvme_configure_opal(ctrl, was_suspended);
3296 
3297 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3298 		/*
3299 		 * Do not return errors unless we are in a controller reset,
3300 		 * the controller works perfectly fine without hwmon.
3301 		 */
3302 		ret = nvme_hwmon_init(ctrl);
3303 		if (ret == -EINTR)
3304 			return ret;
3305 	}
3306 
3307 	ctrl->identified = true;
3308 
3309 	return 0;
3310 }
3311 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3312 
3313 static int nvme_dev_open(struct inode *inode, struct file *file)
3314 {
3315 	struct nvme_ctrl *ctrl =
3316 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3317 
3318 	switch (ctrl->state) {
3319 	case NVME_CTRL_LIVE:
3320 		break;
3321 	default:
3322 		return -EWOULDBLOCK;
3323 	}
3324 
3325 	nvme_get_ctrl(ctrl);
3326 	if (!try_module_get(ctrl->ops->module)) {
3327 		nvme_put_ctrl(ctrl);
3328 		return -EINVAL;
3329 	}
3330 
3331 	file->private_data = ctrl;
3332 	return 0;
3333 }
3334 
3335 static int nvme_dev_release(struct inode *inode, struct file *file)
3336 {
3337 	struct nvme_ctrl *ctrl =
3338 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3339 
3340 	module_put(ctrl->ops->module);
3341 	nvme_put_ctrl(ctrl);
3342 	return 0;
3343 }
3344 
3345 static const struct file_operations nvme_dev_fops = {
3346 	.owner		= THIS_MODULE,
3347 	.open		= nvme_dev_open,
3348 	.release	= nvme_dev_release,
3349 	.unlocked_ioctl	= nvme_dev_ioctl,
3350 	.compat_ioctl	= compat_ptr_ioctl,
3351 	.uring_cmd	= nvme_dev_uring_cmd,
3352 };
3353 
3354 static ssize_t nvme_sysfs_reset(struct device *dev,
3355 				struct device_attribute *attr, const char *buf,
3356 				size_t count)
3357 {
3358 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3359 	int ret;
3360 
3361 	ret = nvme_reset_ctrl_sync(ctrl);
3362 	if (ret < 0)
3363 		return ret;
3364 	return count;
3365 }
3366 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3367 
3368 static ssize_t nvme_sysfs_rescan(struct device *dev,
3369 				struct device_attribute *attr, const char *buf,
3370 				size_t count)
3371 {
3372 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3373 
3374 	nvme_queue_scan(ctrl);
3375 	return count;
3376 }
3377 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3378 
3379 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3380 {
3381 	struct gendisk *disk = dev_to_disk(dev);
3382 
3383 	if (disk->fops == &nvme_bdev_ops)
3384 		return nvme_get_ns_from_dev(dev)->head;
3385 	else
3386 		return disk->private_data;
3387 }
3388 
3389 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3390 		char *buf)
3391 {
3392 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3393 	struct nvme_ns_ids *ids = &head->ids;
3394 	struct nvme_subsystem *subsys = head->subsys;
3395 	int serial_len = sizeof(subsys->serial);
3396 	int model_len = sizeof(subsys->model);
3397 
3398 	if (!uuid_is_null(&ids->uuid))
3399 		return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3400 
3401 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3402 		return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3403 
3404 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3405 		return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3406 
3407 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3408 				  subsys->serial[serial_len - 1] == '\0'))
3409 		serial_len--;
3410 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3411 				 subsys->model[model_len - 1] == '\0'))
3412 		model_len--;
3413 
3414 	return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3415 		serial_len, subsys->serial, model_len, subsys->model,
3416 		head->ns_id);
3417 }
3418 static DEVICE_ATTR_RO(wwid);
3419 
3420 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3421 		char *buf)
3422 {
3423 	return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3424 }
3425 static DEVICE_ATTR_RO(nguid);
3426 
3427 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3428 		char *buf)
3429 {
3430 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3431 
3432 	/* For backward compatibility expose the NGUID to userspace if
3433 	 * we have no UUID set
3434 	 */
3435 	if (uuid_is_null(&ids->uuid)) {
3436 		dev_warn_ratelimited(dev,
3437 			"No UUID available providing old NGUID\n");
3438 		return sysfs_emit(buf, "%pU\n", ids->nguid);
3439 	}
3440 	return sysfs_emit(buf, "%pU\n", &ids->uuid);
3441 }
3442 static DEVICE_ATTR_RO(uuid);
3443 
3444 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3445 		char *buf)
3446 {
3447 	return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3448 }
3449 static DEVICE_ATTR_RO(eui);
3450 
3451 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3452 		char *buf)
3453 {
3454 	return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3455 }
3456 static DEVICE_ATTR_RO(nsid);
3457 
3458 static struct attribute *nvme_ns_id_attrs[] = {
3459 	&dev_attr_wwid.attr,
3460 	&dev_attr_uuid.attr,
3461 	&dev_attr_nguid.attr,
3462 	&dev_attr_eui.attr,
3463 	&dev_attr_nsid.attr,
3464 #ifdef CONFIG_NVME_MULTIPATH
3465 	&dev_attr_ana_grpid.attr,
3466 	&dev_attr_ana_state.attr,
3467 #endif
3468 	NULL,
3469 };
3470 
3471 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3472 		struct attribute *a, int n)
3473 {
3474 	struct device *dev = container_of(kobj, struct device, kobj);
3475 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3476 
3477 	if (a == &dev_attr_uuid.attr) {
3478 		if (uuid_is_null(&ids->uuid) &&
3479 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3480 			return 0;
3481 	}
3482 	if (a == &dev_attr_nguid.attr) {
3483 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3484 			return 0;
3485 	}
3486 	if (a == &dev_attr_eui.attr) {
3487 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3488 			return 0;
3489 	}
3490 #ifdef CONFIG_NVME_MULTIPATH
3491 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3492 		if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3493 			return 0;
3494 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3495 			return 0;
3496 	}
3497 #endif
3498 	return a->mode;
3499 }
3500 
3501 static const struct attribute_group nvme_ns_id_attr_group = {
3502 	.attrs		= nvme_ns_id_attrs,
3503 	.is_visible	= nvme_ns_id_attrs_are_visible,
3504 };
3505 
3506 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3507 	&nvme_ns_id_attr_group,
3508 	NULL,
3509 };
3510 
3511 #define nvme_show_str_function(field)						\
3512 static ssize_t  field##_show(struct device *dev,				\
3513 			    struct device_attribute *attr, char *buf)		\
3514 {										\
3515         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3516         return sysfs_emit(buf, "%.*s\n",					\
3517 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3518 }										\
3519 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3520 
3521 nvme_show_str_function(model);
3522 nvme_show_str_function(serial);
3523 nvme_show_str_function(firmware_rev);
3524 
3525 #define nvme_show_int_function(field)						\
3526 static ssize_t  field##_show(struct device *dev,				\
3527 			    struct device_attribute *attr, char *buf)		\
3528 {										\
3529         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3530         return sysfs_emit(buf, "%d\n", ctrl->field);				\
3531 }										\
3532 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3533 
3534 nvme_show_int_function(cntlid);
3535 nvme_show_int_function(numa_node);
3536 nvme_show_int_function(queue_count);
3537 nvme_show_int_function(sqsize);
3538 nvme_show_int_function(kato);
3539 
3540 static ssize_t nvme_sysfs_delete(struct device *dev,
3541 				struct device_attribute *attr, const char *buf,
3542 				size_t count)
3543 {
3544 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3545 
3546 	if (device_remove_file_self(dev, attr))
3547 		nvme_delete_ctrl_sync(ctrl);
3548 	return count;
3549 }
3550 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3551 
3552 static ssize_t nvme_sysfs_show_transport(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->ops->name);
3559 }
3560 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3561 
3562 static ssize_t nvme_sysfs_show_state(struct device *dev,
3563 				     struct device_attribute *attr,
3564 				     char *buf)
3565 {
3566 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3567 	static const char *const state_name[] = {
3568 		[NVME_CTRL_NEW]		= "new",
3569 		[NVME_CTRL_LIVE]	= "live",
3570 		[NVME_CTRL_RESETTING]	= "resetting",
3571 		[NVME_CTRL_CONNECTING]	= "connecting",
3572 		[NVME_CTRL_DELETING]	= "deleting",
3573 		[NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3574 		[NVME_CTRL_DEAD]	= "dead",
3575 	};
3576 
3577 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3578 	    state_name[ctrl->state])
3579 		return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3580 
3581 	return sysfs_emit(buf, "unknown state\n");
3582 }
3583 
3584 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3585 
3586 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3587 					 struct device_attribute *attr,
3588 					 char *buf)
3589 {
3590 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3591 
3592 	return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3593 }
3594 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3595 
3596 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3597 					struct device_attribute *attr,
3598 					char *buf)
3599 {
3600 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3601 
3602 	return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3603 }
3604 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3605 
3606 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3607 					struct device_attribute *attr,
3608 					char *buf)
3609 {
3610 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3611 
3612 	return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3613 }
3614 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3615 
3616 static ssize_t nvme_sysfs_show_address(struct device *dev,
3617 					 struct device_attribute *attr,
3618 					 char *buf)
3619 {
3620 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3621 
3622 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3623 }
3624 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3625 
3626 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3627 		struct device_attribute *attr, char *buf)
3628 {
3629 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3630 	struct nvmf_ctrl_options *opts = ctrl->opts;
3631 
3632 	if (ctrl->opts->max_reconnects == -1)
3633 		return sysfs_emit(buf, "off\n");
3634 	return sysfs_emit(buf, "%d\n",
3635 			  opts->max_reconnects * opts->reconnect_delay);
3636 }
3637 
3638 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3639 		struct device_attribute *attr, const char *buf, size_t count)
3640 {
3641 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3642 	struct nvmf_ctrl_options *opts = ctrl->opts;
3643 	int ctrl_loss_tmo, err;
3644 
3645 	err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3646 	if (err)
3647 		return -EINVAL;
3648 
3649 	if (ctrl_loss_tmo < 0)
3650 		opts->max_reconnects = -1;
3651 	else
3652 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3653 						opts->reconnect_delay);
3654 	return count;
3655 }
3656 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3657 	nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3658 
3659 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3660 		struct device_attribute *attr, char *buf)
3661 {
3662 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3663 
3664 	if (ctrl->opts->reconnect_delay == -1)
3665 		return sysfs_emit(buf, "off\n");
3666 	return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3667 }
3668 
3669 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3670 		struct device_attribute *attr, const char *buf, size_t count)
3671 {
3672 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3673 	unsigned int v;
3674 	int err;
3675 
3676 	err = kstrtou32(buf, 10, &v);
3677 	if (err)
3678 		return err;
3679 
3680 	ctrl->opts->reconnect_delay = v;
3681 	return count;
3682 }
3683 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3684 	nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3685 
3686 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3687 		struct device_attribute *attr, char *buf)
3688 {
3689 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3690 
3691 	if (ctrl->opts->fast_io_fail_tmo == -1)
3692 		return sysfs_emit(buf, "off\n");
3693 	return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3694 }
3695 
3696 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3697 		struct device_attribute *attr, const char *buf, size_t count)
3698 {
3699 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3700 	struct nvmf_ctrl_options *opts = ctrl->opts;
3701 	int fast_io_fail_tmo, err;
3702 
3703 	err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3704 	if (err)
3705 		return -EINVAL;
3706 
3707 	if (fast_io_fail_tmo < 0)
3708 		opts->fast_io_fail_tmo = -1;
3709 	else
3710 		opts->fast_io_fail_tmo = fast_io_fail_tmo;
3711 	return count;
3712 }
3713 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3714 	nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3715 
3716 static ssize_t cntrltype_show(struct device *dev,
3717 			      struct device_attribute *attr, char *buf)
3718 {
3719 	static const char * const type[] = {
3720 		[NVME_CTRL_IO] = "io\n",
3721 		[NVME_CTRL_DISC] = "discovery\n",
3722 		[NVME_CTRL_ADMIN] = "admin\n",
3723 	};
3724 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3725 
3726 	if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3727 		return sysfs_emit(buf, "reserved\n");
3728 
3729 	return sysfs_emit(buf, type[ctrl->cntrltype]);
3730 }
3731 static DEVICE_ATTR_RO(cntrltype);
3732 
3733 static ssize_t dctype_show(struct device *dev,
3734 			   struct device_attribute *attr, char *buf)
3735 {
3736 	static const char * const type[] = {
3737 		[NVME_DCTYPE_NOT_REPORTED] = "none\n",
3738 		[NVME_DCTYPE_DDC] = "ddc\n",
3739 		[NVME_DCTYPE_CDC] = "cdc\n",
3740 	};
3741 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3742 
3743 	if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3744 		return sysfs_emit(buf, "reserved\n");
3745 
3746 	return sysfs_emit(buf, type[ctrl->dctype]);
3747 }
3748 static DEVICE_ATTR_RO(dctype);
3749 
3750 #ifdef CONFIG_NVME_AUTH
3751 static ssize_t nvme_ctrl_dhchap_secret_show(struct device *dev,
3752 		struct device_attribute *attr, char *buf)
3753 {
3754 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3755 	struct nvmf_ctrl_options *opts = ctrl->opts;
3756 
3757 	if (!opts->dhchap_secret)
3758 		return sysfs_emit(buf, "none\n");
3759 	return sysfs_emit(buf, "%s\n", opts->dhchap_secret);
3760 }
3761 
3762 static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
3763 		struct device_attribute *attr, const char *buf, size_t count)
3764 {
3765 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3766 	struct nvmf_ctrl_options *opts = ctrl->opts;
3767 	char *dhchap_secret;
3768 
3769 	if (!ctrl->opts->dhchap_secret)
3770 		return -EINVAL;
3771 	if (count < 7)
3772 		return -EINVAL;
3773 	if (memcmp(buf, "DHHC-1:", 7))
3774 		return -EINVAL;
3775 
3776 	dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3777 	if (!dhchap_secret)
3778 		return -ENOMEM;
3779 	memcpy(dhchap_secret, buf, count);
3780 	nvme_auth_stop(ctrl);
3781 	if (strcmp(dhchap_secret, opts->dhchap_secret)) {
3782 		struct nvme_dhchap_key *key, *host_key;
3783 		int ret;
3784 
3785 		ret = nvme_auth_generate_key(dhchap_secret, &key);
3786 		if (ret)
3787 			return ret;
3788 		kfree(opts->dhchap_secret);
3789 		opts->dhchap_secret = dhchap_secret;
3790 		host_key = ctrl->host_key;
3791 		mutex_lock(&ctrl->dhchap_auth_mutex);
3792 		ctrl->host_key = key;
3793 		mutex_unlock(&ctrl->dhchap_auth_mutex);
3794 		nvme_auth_free_key(host_key);
3795 	}
3796 	/* Start re-authentication */
3797 	dev_info(ctrl->device, "re-authenticating controller\n");
3798 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3799 
3800 	return count;
3801 }
3802 static DEVICE_ATTR(dhchap_secret, S_IRUGO | S_IWUSR,
3803 	nvme_ctrl_dhchap_secret_show, nvme_ctrl_dhchap_secret_store);
3804 
3805 static ssize_t nvme_ctrl_dhchap_ctrl_secret_show(struct device *dev,
3806 		struct device_attribute *attr, char *buf)
3807 {
3808 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3809 	struct nvmf_ctrl_options *opts = ctrl->opts;
3810 
3811 	if (!opts->dhchap_ctrl_secret)
3812 		return sysfs_emit(buf, "none\n");
3813 	return sysfs_emit(buf, "%s\n", opts->dhchap_ctrl_secret);
3814 }
3815 
3816 static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
3817 		struct device_attribute *attr, const char *buf, size_t count)
3818 {
3819 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3820 	struct nvmf_ctrl_options *opts = ctrl->opts;
3821 	char *dhchap_secret;
3822 
3823 	if (!ctrl->opts->dhchap_ctrl_secret)
3824 		return -EINVAL;
3825 	if (count < 7)
3826 		return -EINVAL;
3827 	if (memcmp(buf, "DHHC-1:", 7))
3828 		return -EINVAL;
3829 
3830 	dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3831 	if (!dhchap_secret)
3832 		return -ENOMEM;
3833 	memcpy(dhchap_secret, buf, count);
3834 	nvme_auth_stop(ctrl);
3835 	if (strcmp(dhchap_secret, opts->dhchap_ctrl_secret)) {
3836 		struct nvme_dhchap_key *key, *ctrl_key;
3837 		int ret;
3838 
3839 		ret = nvme_auth_generate_key(dhchap_secret, &key);
3840 		if (ret)
3841 			return ret;
3842 		kfree(opts->dhchap_ctrl_secret);
3843 		opts->dhchap_ctrl_secret = dhchap_secret;
3844 		ctrl_key = ctrl->ctrl_key;
3845 		mutex_lock(&ctrl->dhchap_auth_mutex);
3846 		ctrl->ctrl_key = key;
3847 		mutex_unlock(&ctrl->dhchap_auth_mutex);
3848 		nvme_auth_free_key(ctrl_key);
3849 	}
3850 	/* Start re-authentication */
3851 	dev_info(ctrl->device, "re-authenticating controller\n");
3852 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3853 
3854 	return count;
3855 }
3856 static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR,
3857 	nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);
3858 #endif
3859 
3860 static struct attribute *nvme_dev_attrs[] = {
3861 	&dev_attr_reset_controller.attr,
3862 	&dev_attr_rescan_controller.attr,
3863 	&dev_attr_model.attr,
3864 	&dev_attr_serial.attr,
3865 	&dev_attr_firmware_rev.attr,
3866 	&dev_attr_cntlid.attr,
3867 	&dev_attr_delete_controller.attr,
3868 	&dev_attr_transport.attr,
3869 	&dev_attr_subsysnqn.attr,
3870 	&dev_attr_address.attr,
3871 	&dev_attr_state.attr,
3872 	&dev_attr_numa_node.attr,
3873 	&dev_attr_queue_count.attr,
3874 	&dev_attr_sqsize.attr,
3875 	&dev_attr_hostnqn.attr,
3876 	&dev_attr_hostid.attr,
3877 	&dev_attr_ctrl_loss_tmo.attr,
3878 	&dev_attr_reconnect_delay.attr,
3879 	&dev_attr_fast_io_fail_tmo.attr,
3880 	&dev_attr_kato.attr,
3881 	&dev_attr_cntrltype.attr,
3882 	&dev_attr_dctype.attr,
3883 #ifdef CONFIG_NVME_AUTH
3884 	&dev_attr_dhchap_secret.attr,
3885 	&dev_attr_dhchap_ctrl_secret.attr,
3886 #endif
3887 	NULL
3888 };
3889 
3890 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3891 		struct attribute *a, int n)
3892 {
3893 	struct device *dev = container_of(kobj, struct device, kobj);
3894 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3895 
3896 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3897 		return 0;
3898 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3899 		return 0;
3900 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3901 		return 0;
3902 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3903 		return 0;
3904 	if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3905 		return 0;
3906 	if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3907 		return 0;
3908 	if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3909 		return 0;
3910 #ifdef CONFIG_NVME_AUTH
3911 	if (a == &dev_attr_dhchap_secret.attr && !ctrl->opts)
3912 		return 0;
3913 	if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)
3914 		return 0;
3915 #endif
3916 
3917 	return a->mode;
3918 }
3919 
3920 const struct attribute_group nvme_dev_attrs_group = {
3921 	.attrs		= nvme_dev_attrs,
3922 	.is_visible	= nvme_dev_attrs_are_visible,
3923 };
3924 EXPORT_SYMBOL_GPL(nvme_dev_attrs_group);
3925 
3926 static const struct attribute_group *nvme_dev_attr_groups[] = {
3927 	&nvme_dev_attrs_group,
3928 	NULL,
3929 };
3930 
3931 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3932 		unsigned nsid)
3933 {
3934 	struct nvme_ns_head *h;
3935 
3936 	lockdep_assert_held(&ctrl->subsys->lock);
3937 
3938 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3939 		/*
3940 		 * Private namespaces can share NSIDs under some conditions.
3941 		 * In that case we can't use the same ns_head for namespaces
3942 		 * with the same NSID.
3943 		 */
3944 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3945 			continue;
3946 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3947 			return h;
3948 	}
3949 
3950 	return NULL;
3951 }
3952 
3953 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3954 		struct nvme_ns_ids *ids)
3955 {
3956 	bool has_uuid = !uuid_is_null(&ids->uuid);
3957 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3958 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3959 	struct nvme_ns_head *h;
3960 
3961 	lockdep_assert_held(&subsys->lock);
3962 
3963 	list_for_each_entry(h, &subsys->nsheads, entry) {
3964 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3965 			return -EINVAL;
3966 		if (has_nguid &&
3967 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3968 			return -EINVAL;
3969 		if (has_eui64 &&
3970 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3971 			return -EINVAL;
3972 	}
3973 
3974 	return 0;
3975 }
3976 
3977 static void nvme_cdev_rel(struct device *dev)
3978 {
3979 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3980 }
3981 
3982 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3983 {
3984 	cdev_device_del(cdev, cdev_device);
3985 	put_device(cdev_device);
3986 }
3987 
3988 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3989 		const struct file_operations *fops, struct module *owner)
3990 {
3991 	int minor, ret;
3992 
3993 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3994 	if (minor < 0)
3995 		return minor;
3996 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3997 	cdev_device->class = nvme_ns_chr_class;
3998 	cdev_device->release = nvme_cdev_rel;
3999 	device_initialize(cdev_device);
4000 	cdev_init(cdev, fops);
4001 	cdev->owner = owner;
4002 	ret = cdev_device_add(cdev, cdev_device);
4003 	if (ret)
4004 		put_device(cdev_device);
4005 
4006 	return ret;
4007 }
4008 
4009 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
4010 {
4011 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
4012 }
4013 
4014 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
4015 {
4016 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
4017 	return 0;
4018 }
4019 
4020 static const struct file_operations nvme_ns_chr_fops = {
4021 	.owner		= THIS_MODULE,
4022 	.open		= nvme_ns_chr_open,
4023 	.release	= nvme_ns_chr_release,
4024 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
4025 	.compat_ioctl	= compat_ptr_ioctl,
4026 	.uring_cmd	= nvme_ns_chr_uring_cmd,
4027 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
4028 };
4029 
4030 static int nvme_add_ns_cdev(struct nvme_ns *ns)
4031 {
4032 	int ret;
4033 
4034 	ns->cdev_device.parent = ns->ctrl->device;
4035 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
4036 			   ns->ctrl->instance, ns->head->instance);
4037 	if (ret)
4038 		return ret;
4039 
4040 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
4041 			     ns->ctrl->ops->module);
4042 }
4043 
4044 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
4045 		struct nvme_ns_info *info)
4046 {
4047 	struct nvme_ns_head *head;
4048 	size_t size = sizeof(*head);
4049 	int ret = -ENOMEM;
4050 
4051 #ifdef CONFIG_NVME_MULTIPATH
4052 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
4053 #endif
4054 
4055 	head = kzalloc(size, GFP_KERNEL);
4056 	if (!head)
4057 		goto out;
4058 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
4059 	if (ret < 0)
4060 		goto out_free_head;
4061 	head->instance = ret;
4062 	INIT_LIST_HEAD(&head->list);
4063 	ret = init_srcu_struct(&head->srcu);
4064 	if (ret)
4065 		goto out_ida_remove;
4066 	head->subsys = ctrl->subsys;
4067 	head->ns_id = info->nsid;
4068 	head->ids = info->ids;
4069 	head->shared = info->is_shared;
4070 	kref_init(&head->ref);
4071 
4072 	if (head->ids.csi) {
4073 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
4074 		if (ret)
4075 			goto out_cleanup_srcu;
4076 	} else
4077 		head->effects = ctrl->effects;
4078 
4079 	ret = nvme_mpath_alloc_disk(ctrl, head);
4080 	if (ret)
4081 		goto out_cleanup_srcu;
4082 
4083 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
4084 
4085 	kref_get(&ctrl->subsys->ref);
4086 
4087 	return head;
4088 out_cleanup_srcu:
4089 	cleanup_srcu_struct(&head->srcu);
4090 out_ida_remove:
4091 	ida_free(&ctrl->subsys->ns_ida, head->instance);
4092 out_free_head:
4093 	kfree(head);
4094 out:
4095 	if (ret > 0)
4096 		ret = blk_status_to_errno(nvme_error_status(ret));
4097 	return ERR_PTR(ret);
4098 }
4099 
4100 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
4101 		struct nvme_ns_ids *ids)
4102 {
4103 	struct nvme_subsystem *s;
4104 	int ret = 0;
4105 
4106 	/*
4107 	 * Note that this check is racy as we try to avoid holding the global
4108 	 * lock over the whole ns_head creation.  But it is only intended as
4109 	 * a sanity check anyway.
4110 	 */
4111 	mutex_lock(&nvme_subsystems_lock);
4112 	list_for_each_entry(s, &nvme_subsystems, entry) {
4113 		if (s == this)
4114 			continue;
4115 		mutex_lock(&s->lock);
4116 		ret = nvme_subsys_check_duplicate_ids(s, ids);
4117 		mutex_unlock(&s->lock);
4118 		if (ret)
4119 			break;
4120 	}
4121 	mutex_unlock(&nvme_subsystems_lock);
4122 
4123 	return ret;
4124 }
4125 
4126 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
4127 {
4128 	struct nvme_ctrl *ctrl = ns->ctrl;
4129 	struct nvme_ns_head *head = NULL;
4130 	int ret;
4131 
4132 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
4133 	if (ret) {
4134 		dev_err(ctrl->device,
4135 			"globally duplicate IDs for nsid %d\n", info->nsid);
4136 		nvme_print_device_info(ctrl);
4137 		return ret;
4138 	}
4139 
4140 	mutex_lock(&ctrl->subsys->lock);
4141 	head = nvme_find_ns_head(ctrl, info->nsid);
4142 	if (!head) {
4143 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4144 		if (ret) {
4145 			dev_err(ctrl->device,
4146 				"duplicate IDs in subsystem for nsid %d\n",
4147 				info->nsid);
4148 			goto out_unlock;
4149 		}
4150 		head = nvme_alloc_ns_head(ctrl, info);
4151 		if (IS_ERR(head)) {
4152 			ret = PTR_ERR(head);
4153 			goto out_unlock;
4154 		}
4155 	} else {
4156 		ret = -EINVAL;
4157 		if (!info->is_shared || !head->shared) {
4158 			dev_err(ctrl->device,
4159 				"Duplicate unshared namespace %d\n",
4160 				info->nsid);
4161 			goto out_put_ns_head;
4162 		}
4163 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4164 			dev_err(ctrl->device,
4165 				"IDs don't match for shared namespace %d\n",
4166 					info->nsid);
4167 			goto out_put_ns_head;
4168 		}
4169 
4170 		if (!multipath && !list_empty(&head->list)) {
4171 			dev_warn(ctrl->device,
4172 				"Found shared namespace %d, but multipathing not supported.\n",
4173 				info->nsid);
4174 			dev_warn_once(ctrl->device,
4175 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
4176 		}
4177 	}
4178 
4179 	list_add_tail_rcu(&ns->siblings, &head->list);
4180 	ns->head = head;
4181 	mutex_unlock(&ctrl->subsys->lock);
4182 	return 0;
4183 
4184 out_put_ns_head:
4185 	nvme_put_ns_head(head);
4186 out_unlock:
4187 	mutex_unlock(&ctrl->subsys->lock);
4188 	return ret;
4189 }
4190 
4191 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4192 {
4193 	struct nvme_ns *ns, *ret = NULL;
4194 
4195 	down_read(&ctrl->namespaces_rwsem);
4196 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4197 		if (ns->head->ns_id == nsid) {
4198 			if (!nvme_get_ns(ns))
4199 				continue;
4200 			ret = ns;
4201 			break;
4202 		}
4203 		if (ns->head->ns_id > nsid)
4204 			break;
4205 	}
4206 	up_read(&ctrl->namespaces_rwsem);
4207 	return ret;
4208 }
4209 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
4210 
4211 /*
4212  * Add the namespace to the controller list while keeping the list ordered.
4213  */
4214 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4215 {
4216 	struct nvme_ns *tmp;
4217 
4218 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4219 		if (tmp->head->ns_id < ns->head->ns_id) {
4220 			list_add(&ns->list, &tmp->list);
4221 			return;
4222 		}
4223 	}
4224 	list_add(&ns->list, &ns->ctrl->namespaces);
4225 }
4226 
4227 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4228 {
4229 	struct nvme_ns *ns;
4230 	struct gendisk *disk;
4231 	int node = ctrl->numa_node;
4232 
4233 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4234 	if (!ns)
4235 		return;
4236 
4237 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
4238 	if (IS_ERR(disk))
4239 		goto out_free_ns;
4240 	disk->fops = &nvme_bdev_ops;
4241 	disk->private_data = ns;
4242 
4243 	ns->disk = disk;
4244 	ns->queue = disk->queue;
4245 
4246 	if (ctrl->opts && ctrl->opts->data_digest)
4247 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4248 
4249 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4250 	if (ctrl->ops->supports_pci_p2pdma &&
4251 	    ctrl->ops->supports_pci_p2pdma(ctrl))
4252 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4253 
4254 	ns->ctrl = ctrl;
4255 	kref_init(&ns->kref);
4256 
4257 	if (nvme_init_ns_head(ns, info))
4258 		goto out_cleanup_disk;
4259 
4260 	/*
4261 	 * If multipathing is enabled, the device name for all disks and not
4262 	 * just those that represent shared namespaces needs to be based on the
4263 	 * subsystem instance.  Using the controller instance for private
4264 	 * namespaces could lead to naming collisions between shared and private
4265 	 * namespaces if they don't use a common numbering scheme.
4266 	 *
4267 	 * If multipathing is not enabled, disk names must use the controller
4268 	 * instance as shared namespaces will show up as multiple block
4269 	 * devices.
4270 	 */
4271 	if (ns->head->disk) {
4272 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4273 			ctrl->instance, ns->head->instance);
4274 		disk->flags |= GENHD_FL_HIDDEN;
4275 	} else if (multipath) {
4276 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4277 			ns->head->instance);
4278 	} else {
4279 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4280 			ns->head->instance);
4281 	}
4282 
4283 	if (nvme_update_ns_info(ns, info))
4284 		goto out_unlink_ns;
4285 
4286 	down_write(&ctrl->namespaces_rwsem);
4287 	nvme_ns_add_to_ctrl_list(ns);
4288 	up_write(&ctrl->namespaces_rwsem);
4289 	nvme_get_ctrl(ctrl);
4290 
4291 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4292 		goto out_cleanup_ns_from_list;
4293 
4294 	if (!nvme_ns_head_multipath(ns->head))
4295 		nvme_add_ns_cdev(ns);
4296 
4297 	nvme_mpath_add_disk(ns, info->anagrpid);
4298 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4299 
4300 	return;
4301 
4302  out_cleanup_ns_from_list:
4303 	nvme_put_ctrl(ctrl);
4304 	down_write(&ctrl->namespaces_rwsem);
4305 	list_del_init(&ns->list);
4306 	up_write(&ctrl->namespaces_rwsem);
4307  out_unlink_ns:
4308 	mutex_lock(&ctrl->subsys->lock);
4309 	list_del_rcu(&ns->siblings);
4310 	if (list_empty(&ns->head->list))
4311 		list_del_init(&ns->head->entry);
4312 	mutex_unlock(&ctrl->subsys->lock);
4313 	nvme_put_ns_head(ns->head);
4314  out_cleanup_disk:
4315 	put_disk(disk);
4316  out_free_ns:
4317 	kfree(ns);
4318 }
4319 
4320 static void nvme_ns_remove(struct nvme_ns *ns)
4321 {
4322 	bool last_path = false;
4323 
4324 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4325 		return;
4326 
4327 	clear_bit(NVME_NS_READY, &ns->flags);
4328 	set_capacity(ns->disk, 0);
4329 	nvme_fault_inject_fini(&ns->fault_inject);
4330 
4331 	/*
4332 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4333 	 * this ns going back into current_path.
4334 	 */
4335 	synchronize_srcu(&ns->head->srcu);
4336 
4337 	/* wait for concurrent submissions */
4338 	if (nvme_mpath_clear_current_path(ns))
4339 		synchronize_srcu(&ns->head->srcu);
4340 
4341 	mutex_lock(&ns->ctrl->subsys->lock);
4342 	list_del_rcu(&ns->siblings);
4343 	if (list_empty(&ns->head->list)) {
4344 		list_del_init(&ns->head->entry);
4345 		last_path = true;
4346 	}
4347 	mutex_unlock(&ns->ctrl->subsys->lock);
4348 
4349 	/* guarantee not available in head->list */
4350 	synchronize_srcu(&ns->head->srcu);
4351 
4352 	if (!nvme_ns_head_multipath(ns->head))
4353 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4354 	del_gendisk(ns->disk);
4355 
4356 	down_write(&ns->ctrl->namespaces_rwsem);
4357 	list_del_init(&ns->list);
4358 	up_write(&ns->ctrl->namespaces_rwsem);
4359 
4360 	if (last_path)
4361 		nvme_mpath_shutdown_disk(ns->head);
4362 	nvme_put_ns(ns);
4363 }
4364 
4365 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4366 {
4367 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4368 
4369 	if (ns) {
4370 		nvme_ns_remove(ns);
4371 		nvme_put_ns(ns);
4372 	}
4373 }
4374 
4375 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4376 {
4377 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4378 
4379 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4380 		dev_err(ns->ctrl->device,
4381 			"identifiers changed for nsid %d\n", ns->head->ns_id);
4382 		goto out;
4383 	}
4384 
4385 	ret = nvme_update_ns_info(ns, info);
4386 out:
4387 	/*
4388 	 * Only remove the namespace if we got a fatal error back from the
4389 	 * device, otherwise ignore the error and just move on.
4390 	 *
4391 	 * TODO: we should probably schedule a delayed retry here.
4392 	 */
4393 	if (ret > 0 && (ret & NVME_SC_DNR))
4394 		nvme_ns_remove(ns);
4395 }
4396 
4397 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4398 {
4399 	struct nvme_ns_info info = { .nsid = nsid };
4400 	struct nvme_ns *ns;
4401 
4402 	if (nvme_identify_ns_descs(ctrl, &info))
4403 		return;
4404 
4405 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4406 		dev_warn(ctrl->device,
4407 			"command set not reported for nsid: %d\n", nsid);
4408 		return;
4409 	}
4410 
4411 	/*
4412 	 * If available try to use the Command Set Idependent Identify Namespace
4413 	 * data structure to find all the generic information that is needed to
4414 	 * set up a namespace.  If not fall back to the legacy version.
4415 	 */
4416 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4417 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS)) {
4418 		if (nvme_ns_info_from_id_cs_indep(ctrl, &info))
4419 			return;
4420 	} else {
4421 		if (nvme_ns_info_from_identify(ctrl, &info))
4422 			return;
4423 	}
4424 
4425 	/*
4426 	 * Ignore the namespace if it is not ready. We will get an AEN once it
4427 	 * becomes ready and restart the scan.
4428 	 */
4429 	if (!info.is_ready)
4430 		return;
4431 
4432 	ns = nvme_find_get_ns(ctrl, nsid);
4433 	if (ns) {
4434 		nvme_validate_ns(ns, &info);
4435 		nvme_put_ns(ns);
4436 	} else {
4437 		nvme_alloc_ns(ctrl, &info);
4438 	}
4439 }
4440 
4441 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4442 					unsigned nsid)
4443 {
4444 	struct nvme_ns *ns, *next;
4445 	LIST_HEAD(rm_list);
4446 
4447 	down_write(&ctrl->namespaces_rwsem);
4448 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4449 		if (ns->head->ns_id > nsid)
4450 			list_move_tail(&ns->list, &rm_list);
4451 	}
4452 	up_write(&ctrl->namespaces_rwsem);
4453 
4454 	list_for_each_entry_safe(ns, next, &rm_list, list)
4455 		nvme_ns_remove(ns);
4456 
4457 }
4458 
4459 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4460 {
4461 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4462 	__le32 *ns_list;
4463 	u32 prev = 0;
4464 	int ret = 0, i;
4465 
4466 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4467 	if (!ns_list)
4468 		return -ENOMEM;
4469 
4470 	for (;;) {
4471 		struct nvme_command cmd = {
4472 			.identify.opcode	= nvme_admin_identify,
4473 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4474 			.identify.nsid		= cpu_to_le32(prev),
4475 		};
4476 
4477 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4478 					    NVME_IDENTIFY_DATA_SIZE);
4479 		if (ret) {
4480 			dev_warn(ctrl->device,
4481 				"Identify NS List failed (status=0x%x)\n", ret);
4482 			goto free;
4483 		}
4484 
4485 		for (i = 0; i < nr_entries; i++) {
4486 			u32 nsid = le32_to_cpu(ns_list[i]);
4487 
4488 			if (!nsid)	/* end of the list? */
4489 				goto out;
4490 			nvme_scan_ns(ctrl, nsid);
4491 			while (++prev < nsid)
4492 				nvme_ns_remove_by_nsid(ctrl, prev);
4493 		}
4494 	}
4495  out:
4496 	nvme_remove_invalid_namespaces(ctrl, prev);
4497  free:
4498 	kfree(ns_list);
4499 	return ret;
4500 }
4501 
4502 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4503 {
4504 	struct nvme_id_ctrl *id;
4505 	u32 nn, i;
4506 
4507 	if (nvme_identify_ctrl(ctrl, &id))
4508 		return;
4509 	nn = le32_to_cpu(id->nn);
4510 	kfree(id);
4511 
4512 	for (i = 1; i <= nn; i++)
4513 		nvme_scan_ns(ctrl, i);
4514 
4515 	nvme_remove_invalid_namespaces(ctrl, nn);
4516 }
4517 
4518 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4519 {
4520 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4521 	__le32 *log;
4522 	int error;
4523 
4524 	log = kzalloc(log_size, GFP_KERNEL);
4525 	if (!log)
4526 		return;
4527 
4528 	/*
4529 	 * We need to read the log to clear the AEN, but we don't want to rely
4530 	 * on it for the changed namespace information as userspace could have
4531 	 * raced with us in reading the log page, which could cause us to miss
4532 	 * updates.
4533 	 */
4534 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4535 			NVME_CSI_NVM, log, log_size, 0);
4536 	if (error)
4537 		dev_warn(ctrl->device,
4538 			"reading changed ns log failed: %d\n", error);
4539 
4540 	kfree(log);
4541 }
4542 
4543 static void nvme_scan_work(struct work_struct *work)
4544 {
4545 	struct nvme_ctrl *ctrl =
4546 		container_of(work, struct nvme_ctrl, scan_work);
4547 	int ret;
4548 
4549 	/* No tagset on a live ctrl means IO queues could not created */
4550 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4551 		return;
4552 
4553 	/*
4554 	 * Identify controller limits can change at controller reset due to
4555 	 * new firmware download, even though it is not common we cannot ignore
4556 	 * such scenario. Controller's non-mdts limits are reported in the unit
4557 	 * of logical blocks that is dependent on the format of attached
4558 	 * namespace. Hence re-read the limits at the time of ns allocation.
4559 	 */
4560 	ret = nvme_init_non_mdts_limits(ctrl);
4561 	if (ret < 0) {
4562 		dev_warn(ctrl->device,
4563 			"reading non-mdts-limits failed: %d\n", ret);
4564 		return;
4565 	}
4566 
4567 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4568 		dev_info(ctrl->device, "rescanning namespaces.\n");
4569 		nvme_clear_changed_ns_log(ctrl);
4570 	}
4571 
4572 	mutex_lock(&ctrl->scan_lock);
4573 	if (nvme_ctrl_limited_cns(ctrl)) {
4574 		nvme_scan_ns_sequential(ctrl);
4575 	} else {
4576 		/*
4577 		 * Fall back to sequential scan if DNR is set to handle broken
4578 		 * devices which should support Identify NS List (as per the VS
4579 		 * they report) but don't actually support it.
4580 		 */
4581 		ret = nvme_scan_ns_list(ctrl);
4582 		if (ret > 0 && ret & NVME_SC_DNR)
4583 			nvme_scan_ns_sequential(ctrl);
4584 	}
4585 	mutex_unlock(&ctrl->scan_lock);
4586 }
4587 
4588 /*
4589  * This function iterates the namespace list unlocked to allow recovery from
4590  * controller failure. It is up to the caller to ensure the namespace list is
4591  * not modified by scan work while this function is executing.
4592  */
4593 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4594 {
4595 	struct nvme_ns *ns, *next;
4596 	LIST_HEAD(ns_list);
4597 
4598 	/*
4599 	 * make sure to requeue I/O to all namespaces as these
4600 	 * might result from the scan itself and must complete
4601 	 * for the scan_work to make progress
4602 	 */
4603 	nvme_mpath_clear_ctrl_paths(ctrl);
4604 
4605 	/* prevent racing with ns scanning */
4606 	flush_work(&ctrl->scan_work);
4607 
4608 	/*
4609 	 * The dead states indicates the controller was not gracefully
4610 	 * disconnected. In that case, we won't be able to flush any data while
4611 	 * removing the namespaces' disks; fail all the queues now to avoid
4612 	 * potentially having to clean up the failed sync later.
4613 	 */
4614 	if (ctrl->state == NVME_CTRL_DEAD) {
4615 		nvme_mark_namespaces_dead(ctrl);
4616 		nvme_unquiesce_io_queues(ctrl);
4617 	}
4618 
4619 	/* this is a no-op when called from the controller reset handler */
4620 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4621 
4622 	down_write(&ctrl->namespaces_rwsem);
4623 	list_splice_init(&ctrl->namespaces, &ns_list);
4624 	up_write(&ctrl->namespaces_rwsem);
4625 
4626 	list_for_each_entry_safe(ns, next, &ns_list, list)
4627 		nvme_ns_remove(ns);
4628 }
4629 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4630 
4631 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4632 {
4633 	const struct nvme_ctrl *ctrl =
4634 		container_of(dev, struct nvme_ctrl, ctrl_device);
4635 	struct nvmf_ctrl_options *opts = ctrl->opts;
4636 	int ret;
4637 
4638 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4639 	if (ret)
4640 		return ret;
4641 
4642 	if (opts) {
4643 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4644 		if (ret)
4645 			return ret;
4646 
4647 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4648 				opts->trsvcid ?: "none");
4649 		if (ret)
4650 			return ret;
4651 
4652 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4653 				opts->host_traddr ?: "none");
4654 		if (ret)
4655 			return ret;
4656 
4657 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4658 				opts->host_iface ?: "none");
4659 	}
4660 	return ret;
4661 }
4662 
4663 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4664 {
4665 	char *envp[2] = { envdata, NULL };
4666 
4667 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4668 }
4669 
4670 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4671 {
4672 	char *envp[2] = { NULL, NULL };
4673 	u32 aen_result = ctrl->aen_result;
4674 
4675 	ctrl->aen_result = 0;
4676 	if (!aen_result)
4677 		return;
4678 
4679 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4680 	if (!envp[0])
4681 		return;
4682 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4683 	kfree(envp[0]);
4684 }
4685 
4686 static void nvme_async_event_work(struct work_struct *work)
4687 {
4688 	struct nvme_ctrl *ctrl =
4689 		container_of(work, struct nvme_ctrl, async_event_work);
4690 
4691 	nvme_aen_uevent(ctrl);
4692 
4693 	/*
4694 	 * The transport drivers must guarantee AER submission here is safe by
4695 	 * flushing ctrl async_event_work after changing the controller state
4696 	 * from LIVE and before freeing the admin queue.
4697 	*/
4698 	if (ctrl->state == NVME_CTRL_LIVE)
4699 		ctrl->ops->submit_async_event(ctrl);
4700 }
4701 
4702 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4703 {
4704 
4705 	u32 csts;
4706 
4707 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4708 		return false;
4709 
4710 	if (csts == ~0)
4711 		return false;
4712 
4713 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4714 }
4715 
4716 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4717 {
4718 	struct nvme_fw_slot_info_log *log;
4719 
4720 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4721 	if (!log)
4722 		return;
4723 
4724 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4725 			log, sizeof(*log), 0))
4726 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4727 	kfree(log);
4728 }
4729 
4730 static void nvme_fw_act_work(struct work_struct *work)
4731 {
4732 	struct nvme_ctrl *ctrl = container_of(work,
4733 				struct nvme_ctrl, fw_act_work);
4734 	unsigned long fw_act_timeout;
4735 
4736 	if (ctrl->mtfa)
4737 		fw_act_timeout = jiffies +
4738 				msecs_to_jiffies(ctrl->mtfa * 100);
4739 	else
4740 		fw_act_timeout = jiffies +
4741 				msecs_to_jiffies(admin_timeout * 1000);
4742 
4743 	nvme_quiesce_io_queues(ctrl);
4744 	while (nvme_ctrl_pp_status(ctrl)) {
4745 		if (time_after(jiffies, fw_act_timeout)) {
4746 			dev_warn(ctrl->device,
4747 				"Fw activation timeout, reset controller\n");
4748 			nvme_try_sched_reset(ctrl);
4749 			return;
4750 		}
4751 		msleep(100);
4752 	}
4753 
4754 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4755 		return;
4756 
4757 	nvme_unquiesce_io_queues(ctrl);
4758 	/* read FW slot information to clear the AER */
4759 	nvme_get_fw_slot_info(ctrl);
4760 
4761 	queue_work(nvme_wq, &ctrl->async_event_work);
4762 }
4763 
4764 static u32 nvme_aer_type(u32 result)
4765 {
4766 	return result & 0x7;
4767 }
4768 
4769 static u32 nvme_aer_subtype(u32 result)
4770 {
4771 	return (result & 0xff00) >> 8;
4772 }
4773 
4774 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4775 {
4776 	u32 aer_notice_type = nvme_aer_subtype(result);
4777 	bool requeue = true;
4778 
4779 	trace_nvme_async_event(ctrl, aer_notice_type);
4780 
4781 	switch (aer_notice_type) {
4782 	case NVME_AER_NOTICE_NS_CHANGED:
4783 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4784 		nvme_queue_scan(ctrl);
4785 		break;
4786 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4787 		/*
4788 		 * We are (ab)using the RESETTING state to prevent subsequent
4789 		 * recovery actions from interfering with the controller's
4790 		 * firmware activation.
4791 		 */
4792 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4793 			nvme_auth_stop(ctrl);
4794 			requeue = false;
4795 			queue_work(nvme_wq, &ctrl->fw_act_work);
4796 		}
4797 		break;
4798 #ifdef CONFIG_NVME_MULTIPATH
4799 	case NVME_AER_NOTICE_ANA:
4800 		if (!ctrl->ana_log_buf)
4801 			break;
4802 		queue_work(nvme_wq, &ctrl->ana_work);
4803 		break;
4804 #endif
4805 	case NVME_AER_NOTICE_DISC_CHANGED:
4806 		ctrl->aen_result = result;
4807 		break;
4808 	default:
4809 		dev_warn(ctrl->device, "async event result %08x\n", result);
4810 	}
4811 	return requeue;
4812 }
4813 
4814 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4815 {
4816 	trace_nvme_async_event(ctrl, NVME_AER_ERROR);
4817 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4818 	nvme_reset_ctrl(ctrl);
4819 }
4820 
4821 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4822 		volatile union nvme_result *res)
4823 {
4824 	u32 result = le32_to_cpu(res->u32);
4825 	u32 aer_type = nvme_aer_type(result);
4826 	u32 aer_subtype = nvme_aer_subtype(result);
4827 	bool requeue = true;
4828 
4829 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4830 		return;
4831 
4832 	switch (aer_type) {
4833 	case NVME_AER_NOTICE:
4834 		requeue = nvme_handle_aen_notice(ctrl, result);
4835 		break;
4836 	case NVME_AER_ERROR:
4837 		/*
4838 		 * For a persistent internal error, don't run async_event_work
4839 		 * to submit a new AER. The controller reset will do it.
4840 		 */
4841 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4842 			nvme_handle_aer_persistent_error(ctrl);
4843 			return;
4844 		}
4845 		fallthrough;
4846 	case NVME_AER_SMART:
4847 	case NVME_AER_CSS:
4848 	case NVME_AER_VS:
4849 		trace_nvme_async_event(ctrl, aer_type);
4850 		ctrl->aen_result = result;
4851 		break;
4852 	default:
4853 		break;
4854 	}
4855 
4856 	if (requeue)
4857 		queue_work(nvme_wq, &ctrl->async_event_work);
4858 }
4859 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4860 
4861 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4862 		const struct blk_mq_ops *ops, unsigned int cmd_size)
4863 {
4864 	int ret;
4865 
4866 	memset(set, 0, sizeof(*set));
4867 	set->ops = ops;
4868 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4869 	if (ctrl->ops->flags & NVME_F_FABRICS)
4870 		set->reserved_tags = NVMF_RESERVED_TAGS;
4871 	set->numa_node = ctrl->numa_node;
4872 	set->flags = BLK_MQ_F_NO_SCHED;
4873 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4874 		set->flags |= BLK_MQ_F_BLOCKING;
4875 	set->cmd_size = cmd_size;
4876 	set->driver_data = ctrl;
4877 	set->nr_hw_queues = 1;
4878 	set->timeout = NVME_ADMIN_TIMEOUT;
4879 	ret = blk_mq_alloc_tag_set(set);
4880 	if (ret)
4881 		return ret;
4882 
4883 	ctrl->admin_q = blk_mq_init_queue(set);
4884 	if (IS_ERR(ctrl->admin_q)) {
4885 		ret = PTR_ERR(ctrl->admin_q);
4886 		goto out_free_tagset;
4887 	}
4888 
4889 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4890 		ctrl->fabrics_q = blk_mq_init_queue(set);
4891 		if (IS_ERR(ctrl->fabrics_q)) {
4892 			ret = PTR_ERR(ctrl->fabrics_q);
4893 			goto out_cleanup_admin_q;
4894 		}
4895 	}
4896 
4897 	ctrl->admin_tagset = set;
4898 	return 0;
4899 
4900 out_cleanup_admin_q:
4901 	blk_mq_destroy_queue(ctrl->admin_q);
4902 	blk_put_queue(ctrl->admin_q);
4903 out_free_tagset:
4904 	blk_mq_free_tag_set(ctrl->admin_tagset);
4905 	return ret;
4906 }
4907 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4908 
4909 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4910 {
4911 	blk_mq_destroy_queue(ctrl->admin_q);
4912 	blk_put_queue(ctrl->admin_q);
4913 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4914 		blk_mq_destroy_queue(ctrl->fabrics_q);
4915 		blk_put_queue(ctrl->fabrics_q);
4916 	}
4917 	blk_mq_free_tag_set(ctrl->admin_tagset);
4918 }
4919 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4920 
4921 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4922 		const struct blk_mq_ops *ops, unsigned int nr_maps,
4923 		unsigned int cmd_size)
4924 {
4925 	int ret;
4926 
4927 	memset(set, 0, sizeof(*set));
4928 	set->ops = ops;
4929 	set->queue_depth = ctrl->sqsize + 1;
4930 	/*
4931 	 * Some Apple controllers requires tags to be unique across admin and
4932 	 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4933 	 */
4934 	if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4935 		set->reserved_tags = NVME_AQ_DEPTH;
4936 	else if (ctrl->ops->flags & NVME_F_FABRICS)
4937 		set->reserved_tags = NVMF_RESERVED_TAGS;
4938 	set->numa_node = ctrl->numa_node;
4939 	set->flags = BLK_MQ_F_SHOULD_MERGE;
4940 	if (ctrl->ops->flags & NVME_F_BLOCKING)
4941 		set->flags |= BLK_MQ_F_BLOCKING;
4942 	set->cmd_size = cmd_size,
4943 	set->driver_data = ctrl;
4944 	set->nr_hw_queues = ctrl->queue_count - 1;
4945 	set->timeout = NVME_IO_TIMEOUT;
4946 	set->nr_maps = nr_maps;
4947 	ret = blk_mq_alloc_tag_set(set);
4948 	if (ret)
4949 		return ret;
4950 
4951 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4952 		ctrl->connect_q = blk_mq_init_queue(set);
4953         	if (IS_ERR(ctrl->connect_q)) {
4954 			ret = PTR_ERR(ctrl->connect_q);
4955 			goto out_free_tag_set;
4956 		}
4957 		blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4958 				   ctrl->connect_q);
4959 	}
4960 
4961 	ctrl->tagset = set;
4962 	return 0;
4963 
4964 out_free_tag_set:
4965 	blk_mq_free_tag_set(set);
4966 	return ret;
4967 }
4968 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4969 
4970 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4971 {
4972 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4973 		blk_mq_destroy_queue(ctrl->connect_q);
4974 		blk_put_queue(ctrl->connect_q);
4975 	}
4976 	blk_mq_free_tag_set(ctrl->tagset);
4977 }
4978 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4979 
4980 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4981 {
4982 	nvme_mpath_stop(ctrl);
4983 	nvme_auth_stop(ctrl);
4984 	nvme_stop_keep_alive(ctrl);
4985 	nvme_stop_failfast_work(ctrl);
4986 	flush_work(&ctrl->async_event_work);
4987 	cancel_work_sync(&ctrl->fw_act_work);
4988 	if (ctrl->ops->stop_ctrl)
4989 		ctrl->ops->stop_ctrl(ctrl);
4990 }
4991 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4992 
4993 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4994 {
4995 	nvme_start_keep_alive(ctrl);
4996 
4997 	nvme_enable_aen(ctrl);
4998 
4999 	/*
5000 	 * persistent discovery controllers need to send indication to userspace
5001 	 * to re-read the discovery log page to learn about possible changes
5002 	 * that were missed. We identify persistent discovery controllers by
5003 	 * checking that they started once before, hence are reconnecting back.
5004 	 */
5005 	if (test_and_set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
5006 	    nvme_discovery_ctrl(ctrl))
5007 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
5008 
5009 	if (ctrl->queue_count > 1) {
5010 		nvme_queue_scan(ctrl);
5011 		nvme_unquiesce_io_queues(ctrl);
5012 		nvme_mpath_update(ctrl);
5013 	}
5014 
5015 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
5016 }
5017 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
5018 
5019 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
5020 {
5021 	nvme_hwmon_exit(ctrl);
5022 	nvme_fault_inject_fini(&ctrl->fault_inject);
5023 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
5024 	cdev_device_del(&ctrl->cdev, ctrl->device);
5025 	nvme_put_ctrl(ctrl);
5026 }
5027 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
5028 
5029 static void nvme_free_cels(struct nvme_ctrl *ctrl)
5030 {
5031 	struct nvme_effects_log	*cel;
5032 	unsigned long i;
5033 
5034 	xa_for_each(&ctrl->cels, i, cel) {
5035 		xa_erase(&ctrl->cels, i);
5036 		kfree(cel);
5037 	}
5038 
5039 	xa_destroy(&ctrl->cels);
5040 }
5041 
5042 static void nvme_free_ctrl(struct device *dev)
5043 {
5044 	struct nvme_ctrl *ctrl =
5045 		container_of(dev, struct nvme_ctrl, ctrl_device);
5046 	struct nvme_subsystem *subsys = ctrl->subsys;
5047 
5048 	if (!subsys || ctrl->instance != subsys->instance)
5049 		ida_free(&nvme_instance_ida, ctrl->instance);
5050 
5051 	nvme_free_cels(ctrl);
5052 	nvme_mpath_uninit(ctrl);
5053 	nvme_auth_stop(ctrl);
5054 	nvme_auth_free(ctrl);
5055 	__free_page(ctrl->discard_page);
5056 	free_opal_dev(ctrl->opal_dev);
5057 
5058 	if (subsys) {
5059 		mutex_lock(&nvme_subsystems_lock);
5060 		list_del(&ctrl->subsys_entry);
5061 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
5062 		mutex_unlock(&nvme_subsystems_lock);
5063 	}
5064 
5065 	ctrl->ops->free_ctrl(ctrl);
5066 
5067 	if (subsys)
5068 		nvme_put_subsystem(subsys);
5069 }
5070 
5071 /*
5072  * Initialize a NVMe controller structures.  This needs to be called during
5073  * earliest initialization so that we have the initialized structured around
5074  * during probing.
5075  */
5076 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
5077 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
5078 {
5079 	int ret;
5080 
5081 	ctrl->state = NVME_CTRL_NEW;
5082 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
5083 	spin_lock_init(&ctrl->lock);
5084 	mutex_init(&ctrl->scan_lock);
5085 	INIT_LIST_HEAD(&ctrl->namespaces);
5086 	xa_init(&ctrl->cels);
5087 	init_rwsem(&ctrl->namespaces_rwsem);
5088 	ctrl->dev = dev;
5089 	ctrl->ops = ops;
5090 	ctrl->quirks = quirks;
5091 	ctrl->numa_node = NUMA_NO_NODE;
5092 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
5093 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
5094 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
5095 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
5096 	init_waitqueue_head(&ctrl->state_wq);
5097 
5098 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
5099 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
5100 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
5101 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
5102 
5103 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
5104 			PAGE_SIZE);
5105 	ctrl->discard_page = alloc_page(GFP_KERNEL);
5106 	if (!ctrl->discard_page) {
5107 		ret = -ENOMEM;
5108 		goto out;
5109 	}
5110 
5111 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
5112 	if (ret < 0)
5113 		goto out;
5114 	ctrl->instance = ret;
5115 
5116 	device_initialize(&ctrl->ctrl_device);
5117 	ctrl->device = &ctrl->ctrl_device;
5118 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
5119 			ctrl->instance);
5120 	ctrl->device->class = nvme_class;
5121 	ctrl->device->parent = ctrl->dev;
5122 	if (ops->dev_attr_groups)
5123 		ctrl->device->groups = ops->dev_attr_groups;
5124 	else
5125 		ctrl->device->groups = nvme_dev_attr_groups;
5126 	ctrl->device->release = nvme_free_ctrl;
5127 	dev_set_drvdata(ctrl->device, ctrl);
5128 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
5129 	if (ret)
5130 		goto out_release_instance;
5131 
5132 	nvme_get_ctrl(ctrl);
5133 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
5134 	ctrl->cdev.owner = ops->module;
5135 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
5136 	if (ret)
5137 		goto out_free_name;
5138 
5139 	/*
5140 	 * Initialize latency tolerance controls.  The sysfs files won't
5141 	 * be visible to userspace unless the device actually supports APST.
5142 	 */
5143 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
5144 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
5145 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
5146 
5147 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
5148 	nvme_mpath_init_ctrl(ctrl);
5149 	ret = nvme_auth_init_ctrl(ctrl);
5150 	if (ret)
5151 		goto out_free_cdev;
5152 
5153 	return 0;
5154 out_free_cdev:
5155 	cdev_device_del(&ctrl->cdev, ctrl->device);
5156 out_free_name:
5157 	nvme_put_ctrl(ctrl);
5158 	kfree_const(ctrl->device->kobj.name);
5159 out_release_instance:
5160 	ida_free(&nvme_instance_ida, ctrl->instance);
5161 out:
5162 	if (ctrl->discard_page)
5163 		__free_page(ctrl->discard_page);
5164 	return ret;
5165 }
5166 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
5167 
5168 /* let I/O to all namespaces fail in preparation for surprise removal */
5169 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
5170 {
5171 	struct nvme_ns *ns;
5172 
5173 	down_read(&ctrl->namespaces_rwsem);
5174 	list_for_each_entry(ns, &ctrl->namespaces, list)
5175 		blk_mark_disk_dead(ns->disk);
5176 	up_read(&ctrl->namespaces_rwsem);
5177 }
5178 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
5179 
5180 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5181 {
5182 	struct nvme_ns *ns;
5183 
5184 	down_read(&ctrl->namespaces_rwsem);
5185 	list_for_each_entry(ns, &ctrl->namespaces, list)
5186 		blk_mq_unfreeze_queue(ns->queue);
5187 	up_read(&ctrl->namespaces_rwsem);
5188 }
5189 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5190 
5191 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5192 {
5193 	struct nvme_ns *ns;
5194 
5195 	down_read(&ctrl->namespaces_rwsem);
5196 	list_for_each_entry(ns, &ctrl->namespaces, list) {
5197 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5198 		if (timeout <= 0)
5199 			break;
5200 	}
5201 	up_read(&ctrl->namespaces_rwsem);
5202 	return timeout;
5203 }
5204 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5205 
5206 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5207 {
5208 	struct nvme_ns *ns;
5209 
5210 	down_read(&ctrl->namespaces_rwsem);
5211 	list_for_each_entry(ns, &ctrl->namespaces, list)
5212 		blk_mq_freeze_queue_wait(ns->queue);
5213 	up_read(&ctrl->namespaces_rwsem);
5214 }
5215 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5216 
5217 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5218 {
5219 	struct nvme_ns *ns;
5220 
5221 	down_read(&ctrl->namespaces_rwsem);
5222 	list_for_each_entry(ns, &ctrl->namespaces, list)
5223 		blk_freeze_queue_start(ns->queue);
5224 	up_read(&ctrl->namespaces_rwsem);
5225 }
5226 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5227 
5228 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
5229 {
5230 	if (!ctrl->tagset)
5231 		return;
5232 	if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5233 		blk_mq_quiesce_tagset(ctrl->tagset);
5234 	else
5235 		blk_mq_wait_quiesce_done(ctrl->tagset);
5236 }
5237 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
5238 
5239 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
5240 {
5241 	if (!ctrl->tagset)
5242 		return;
5243 	if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5244 		blk_mq_unquiesce_tagset(ctrl->tagset);
5245 }
5246 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
5247 
5248 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
5249 {
5250 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5251 		blk_mq_quiesce_queue(ctrl->admin_q);
5252 	else
5253 		blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
5254 }
5255 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
5256 
5257 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
5258 {
5259 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5260 		blk_mq_unquiesce_queue(ctrl->admin_q);
5261 }
5262 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
5263 
5264 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5265 {
5266 	struct nvme_ns *ns;
5267 
5268 	down_read(&ctrl->namespaces_rwsem);
5269 	list_for_each_entry(ns, &ctrl->namespaces, list)
5270 		blk_sync_queue(ns->queue);
5271 	up_read(&ctrl->namespaces_rwsem);
5272 }
5273 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5274 
5275 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5276 {
5277 	nvme_sync_io_queues(ctrl);
5278 	if (ctrl->admin_q)
5279 		blk_sync_queue(ctrl->admin_q);
5280 }
5281 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5282 
5283 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5284 {
5285 	if (file->f_op != &nvme_dev_fops)
5286 		return NULL;
5287 	return file->private_data;
5288 }
5289 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
5290 
5291 /*
5292  * Check we didn't inadvertently grow the command structure sizes:
5293  */
5294 static inline void _nvme_check_size(void)
5295 {
5296 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5297 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5298 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5299 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5300 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5301 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5302 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5303 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5304 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5305 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5306 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5307 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5308 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5309 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5310 			NVME_IDENTIFY_DATA_SIZE);
5311 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5312 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5313 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5314 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5315 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5316 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5317 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5318 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5319 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5320 }
5321 
5322 
5323 static int __init nvme_core_init(void)
5324 {
5325 	int result = -ENOMEM;
5326 
5327 	_nvme_check_size();
5328 
5329 	nvme_wq = alloc_workqueue("nvme-wq",
5330 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5331 	if (!nvme_wq)
5332 		goto out;
5333 
5334 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
5335 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5336 	if (!nvme_reset_wq)
5337 		goto destroy_wq;
5338 
5339 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
5340 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5341 	if (!nvme_delete_wq)
5342 		goto destroy_reset_wq;
5343 
5344 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5345 			NVME_MINORS, "nvme");
5346 	if (result < 0)
5347 		goto destroy_delete_wq;
5348 
5349 	nvme_class = class_create(THIS_MODULE, "nvme");
5350 	if (IS_ERR(nvme_class)) {
5351 		result = PTR_ERR(nvme_class);
5352 		goto unregister_chrdev;
5353 	}
5354 	nvme_class->dev_uevent = nvme_class_uevent;
5355 
5356 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
5357 	if (IS_ERR(nvme_subsys_class)) {
5358 		result = PTR_ERR(nvme_subsys_class);
5359 		goto destroy_class;
5360 	}
5361 
5362 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5363 				     "nvme-generic");
5364 	if (result < 0)
5365 		goto destroy_subsys_class;
5366 
5367 	nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
5368 	if (IS_ERR(nvme_ns_chr_class)) {
5369 		result = PTR_ERR(nvme_ns_chr_class);
5370 		goto unregister_generic_ns;
5371 	}
5372 
5373 	result = nvme_init_auth();
5374 	if (result)
5375 		goto destroy_ns_chr;
5376 	return 0;
5377 
5378 destroy_ns_chr:
5379 	class_destroy(nvme_ns_chr_class);
5380 unregister_generic_ns:
5381 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5382 destroy_subsys_class:
5383 	class_destroy(nvme_subsys_class);
5384 destroy_class:
5385 	class_destroy(nvme_class);
5386 unregister_chrdev:
5387 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5388 destroy_delete_wq:
5389 	destroy_workqueue(nvme_delete_wq);
5390 destroy_reset_wq:
5391 	destroy_workqueue(nvme_reset_wq);
5392 destroy_wq:
5393 	destroy_workqueue(nvme_wq);
5394 out:
5395 	return result;
5396 }
5397 
5398 static void __exit nvme_core_exit(void)
5399 {
5400 	nvme_exit_auth();
5401 	class_destroy(nvme_ns_chr_class);
5402 	class_destroy(nvme_subsys_class);
5403 	class_destroy(nvme_class);
5404 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5405 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5406 	destroy_workqueue(nvme_delete_wq);
5407 	destroy_workqueue(nvme_reset_wq);
5408 	destroy_workqueue(nvme_wq);
5409 	ida_destroy(&nvme_ns_chr_minor_ida);
5410 	ida_destroy(&nvme_instance_ida);
5411 }
5412 
5413 MODULE_LICENSE("GPL");
5414 MODULE_VERSION("1.0");
5415 module_init(nvme_core_init);
5416 module_exit(nvme_core_exit);
5417