xref: /openbmc/linux/drivers/nvme/host/core.c (revision ae108c48)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NVM Express device driver
4  * Copyright (c) 2011-2014, Intel Corporation.
5  */
6 
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24 
25 #include "nvme.h"
26 #include "fabrics.h"
27 #include <linux/nvme-auth.h>
28 
29 #define CREATE_TRACE_POINTS
30 #include "trace.h"
31 
32 #define NVME_MINORS		(1U << MINORBITS)
33 
34 struct nvme_ns_info {
35 	struct nvme_ns_ids ids;
36 	u32 nsid;
37 	__le32 anagrpid;
38 	bool is_shared;
39 	bool is_readonly;
40 	bool is_ready;
41 };
42 
43 unsigned int admin_timeout = 60;
44 module_param(admin_timeout, uint, 0644);
45 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
46 EXPORT_SYMBOL_GPL(admin_timeout);
47 
48 unsigned int nvme_io_timeout = 30;
49 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
50 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
51 EXPORT_SYMBOL_GPL(nvme_io_timeout);
52 
53 static unsigned char shutdown_timeout = 5;
54 module_param(shutdown_timeout, byte, 0644);
55 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
56 
57 static u8 nvme_max_retries = 5;
58 module_param_named(max_retries, nvme_max_retries, byte, 0644);
59 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
60 
61 static unsigned long default_ps_max_latency_us = 100000;
62 module_param(default_ps_max_latency_us, ulong, 0644);
63 MODULE_PARM_DESC(default_ps_max_latency_us,
64 		 "max power saving latency for new devices; use PM QOS to change per device");
65 
66 static bool force_apst;
67 module_param(force_apst, bool, 0644);
68 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
69 
70 static unsigned long apst_primary_timeout_ms = 100;
71 module_param(apst_primary_timeout_ms, ulong, 0644);
72 MODULE_PARM_DESC(apst_primary_timeout_ms,
73 	"primary APST timeout in ms");
74 
75 static unsigned long apst_secondary_timeout_ms = 2000;
76 module_param(apst_secondary_timeout_ms, ulong, 0644);
77 MODULE_PARM_DESC(apst_secondary_timeout_ms,
78 	"secondary APST timeout in ms");
79 
80 static unsigned long apst_primary_latency_tol_us = 15000;
81 module_param(apst_primary_latency_tol_us, ulong, 0644);
82 MODULE_PARM_DESC(apst_primary_latency_tol_us,
83 	"primary APST latency tolerance in us");
84 
85 static unsigned long apst_secondary_latency_tol_us = 100000;
86 module_param(apst_secondary_latency_tol_us, ulong, 0644);
87 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
88 	"secondary APST latency tolerance in us");
89 
90 /*
91  * nvme_wq - hosts nvme related works that are not reset or delete
92  * nvme_reset_wq - hosts nvme reset works
93  * nvme_delete_wq - hosts nvme delete works
94  *
95  * nvme_wq will host works such as scan, aen handling, fw activation,
96  * keep-alive, periodic reconnects etc. nvme_reset_wq
97  * runs reset works which also flush works hosted on nvme_wq for
98  * serialization purposes. nvme_delete_wq host controller deletion
99  * works which flush reset works for serialization.
100  */
101 struct workqueue_struct *nvme_wq;
102 EXPORT_SYMBOL_GPL(nvme_wq);
103 
104 struct workqueue_struct *nvme_reset_wq;
105 EXPORT_SYMBOL_GPL(nvme_reset_wq);
106 
107 struct workqueue_struct *nvme_delete_wq;
108 EXPORT_SYMBOL_GPL(nvme_delete_wq);
109 
110 static LIST_HEAD(nvme_subsystems);
111 static DEFINE_MUTEX(nvme_subsystems_lock);
112 
113 static DEFINE_IDA(nvme_instance_ida);
114 static dev_t nvme_ctrl_base_chr_devt;
115 static struct class *nvme_class;
116 static struct class *nvme_subsys_class;
117 
118 static DEFINE_IDA(nvme_ns_chr_minor_ida);
119 static dev_t nvme_ns_chr_devt;
120 static struct class *nvme_ns_chr_class;
121 
122 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
123 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
124 					   unsigned nsid);
125 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
126 				   struct nvme_command *cmd);
127 
128 void nvme_queue_scan(struct nvme_ctrl *ctrl)
129 {
130 	/*
131 	 * Only new queue scan work when admin and IO queues are both alive
132 	 */
133 	if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
134 		queue_work(nvme_wq, &ctrl->scan_work);
135 }
136 
137 /*
138  * Use this function to proceed with scheduling reset_work for a controller
139  * that had previously been set to the resetting state. This is intended for
140  * code paths that can't be interrupted by other reset attempts. A hot removal
141  * may prevent this from succeeding.
142  */
143 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
144 {
145 	if (ctrl->state != NVME_CTRL_RESETTING)
146 		return -EBUSY;
147 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
148 		return -EBUSY;
149 	return 0;
150 }
151 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
152 
153 static void nvme_failfast_work(struct work_struct *work)
154 {
155 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
156 			struct nvme_ctrl, failfast_work);
157 
158 	if (ctrl->state != NVME_CTRL_CONNECTING)
159 		return;
160 
161 	set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
162 	dev_info(ctrl->device, "failfast expired\n");
163 	nvme_kick_requeue_lists(ctrl);
164 }
165 
166 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
167 {
168 	if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
169 		return;
170 
171 	schedule_delayed_work(&ctrl->failfast_work,
172 			      ctrl->opts->fast_io_fail_tmo * HZ);
173 }
174 
175 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
176 {
177 	if (!ctrl->opts)
178 		return;
179 
180 	cancel_delayed_work_sync(&ctrl->failfast_work);
181 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
182 }
183 
184 
185 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
186 {
187 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
188 		return -EBUSY;
189 	if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
190 		return -EBUSY;
191 	return 0;
192 }
193 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
194 
195 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
196 {
197 	int ret;
198 
199 	ret = nvme_reset_ctrl(ctrl);
200 	if (!ret) {
201 		flush_work(&ctrl->reset_work);
202 		if (ctrl->state != NVME_CTRL_LIVE)
203 			ret = -ENETRESET;
204 	}
205 
206 	return ret;
207 }
208 
209 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
210 {
211 	dev_info(ctrl->device,
212 		 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
213 
214 	flush_work(&ctrl->reset_work);
215 	nvme_stop_ctrl(ctrl);
216 	nvme_remove_namespaces(ctrl);
217 	ctrl->ops->delete_ctrl(ctrl);
218 	nvme_uninit_ctrl(ctrl);
219 }
220 
221 static void nvme_delete_ctrl_work(struct work_struct *work)
222 {
223 	struct nvme_ctrl *ctrl =
224 		container_of(work, struct nvme_ctrl, delete_work);
225 
226 	nvme_do_delete_ctrl(ctrl);
227 }
228 
229 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
230 {
231 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
232 		return -EBUSY;
233 	if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
234 		return -EBUSY;
235 	return 0;
236 }
237 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
238 
239 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
240 {
241 	/*
242 	 * Keep a reference until nvme_do_delete_ctrl() complete,
243 	 * since ->delete_ctrl can free the controller.
244 	 */
245 	nvme_get_ctrl(ctrl);
246 	if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
247 		nvme_do_delete_ctrl(ctrl);
248 	nvme_put_ctrl(ctrl);
249 }
250 
251 static blk_status_t nvme_error_status(u16 status)
252 {
253 	switch (status & 0x7ff) {
254 	case NVME_SC_SUCCESS:
255 		return BLK_STS_OK;
256 	case NVME_SC_CAP_EXCEEDED:
257 		return BLK_STS_NOSPC;
258 	case NVME_SC_LBA_RANGE:
259 	case NVME_SC_CMD_INTERRUPTED:
260 	case NVME_SC_NS_NOT_READY:
261 		return BLK_STS_TARGET;
262 	case NVME_SC_BAD_ATTRIBUTES:
263 	case NVME_SC_ONCS_NOT_SUPPORTED:
264 	case NVME_SC_INVALID_OPCODE:
265 	case NVME_SC_INVALID_FIELD:
266 	case NVME_SC_INVALID_NS:
267 		return BLK_STS_NOTSUPP;
268 	case NVME_SC_WRITE_FAULT:
269 	case NVME_SC_READ_ERROR:
270 	case NVME_SC_UNWRITTEN_BLOCK:
271 	case NVME_SC_ACCESS_DENIED:
272 	case NVME_SC_READ_ONLY:
273 	case NVME_SC_COMPARE_FAILED:
274 		return BLK_STS_MEDIUM;
275 	case NVME_SC_GUARD_CHECK:
276 	case NVME_SC_APPTAG_CHECK:
277 	case NVME_SC_REFTAG_CHECK:
278 	case NVME_SC_INVALID_PI:
279 		return BLK_STS_PROTECTION;
280 	case NVME_SC_RESERVATION_CONFLICT:
281 		return BLK_STS_NEXUS;
282 	case NVME_SC_HOST_PATH_ERROR:
283 		return BLK_STS_TRANSPORT;
284 	case NVME_SC_ZONE_TOO_MANY_ACTIVE:
285 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
286 	case NVME_SC_ZONE_TOO_MANY_OPEN:
287 		return BLK_STS_ZONE_OPEN_RESOURCE;
288 	default:
289 		return BLK_STS_IOERR;
290 	}
291 }
292 
293 static void nvme_retry_req(struct request *req)
294 {
295 	unsigned long delay = 0;
296 	u16 crd;
297 
298 	/* The mask and shift result must be <= 3 */
299 	crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
300 	if (crd)
301 		delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
302 
303 	nvme_req(req)->retries++;
304 	blk_mq_requeue_request(req, false);
305 	blk_mq_delay_kick_requeue_list(req->q, delay);
306 }
307 
308 static void nvme_log_error(struct request *req)
309 {
310 	struct nvme_ns *ns = req->q->queuedata;
311 	struct nvme_request *nr = nvme_req(req);
312 
313 	if (ns) {
314 		pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
315 		       ns->disk ? ns->disk->disk_name : "?",
316 		       nvme_get_opcode_str(nr->cmd->common.opcode),
317 		       nr->cmd->common.opcode,
318 		       (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
319 		       (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
320 		       nvme_get_error_status_str(nr->status),
321 		       nr->status >> 8 & 7,	/* Status Code Type */
322 		       nr->status & 0xff,	/* Status Code */
323 		       nr->status & NVME_SC_MORE ? "MORE " : "",
324 		       nr->status & NVME_SC_DNR  ? "DNR "  : "");
325 		return;
326 	}
327 
328 	pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
329 			   dev_name(nr->ctrl->device),
330 			   nvme_get_admin_opcode_str(nr->cmd->common.opcode),
331 			   nr->cmd->common.opcode,
332 			   nvme_get_error_status_str(nr->status),
333 			   nr->status >> 8 & 7,	/* Status Code Type */
334 			   nr->status & 0xff,	/* Status Code */
335 			   nr->status & NVME_SC_MORE ? "MORE " : "",
336 			   nr->status & NVME_SC_DNR  ? "DNR "  : "");
337 }
338 
339 enum nvme_disposition {
340 	COMPLETE,
341 	RETRY,
342 	FAILOVER,
343 	AUTHENTICATE,
344 };
345 
346 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
347 {
348 	if (likely(nvme_req(req)->status == 0))
349 		return COMPLETE;
350 
351 	if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
352 		return AUTHENTICATE;
353 
354 	if (blk_noretry_request(req) ||
355 	    (nvme_req(req)->status & NVME_SC_DNR) ||
356 	    nvme_req(req)->retries >= nvme_max_retries)
357 		return COMPLETE;
358 
359 	if (req->cmd_flags & REQ_NVME_MPATH) {
360 		if (nvme_is_path_error(nvme_req(req)->status) ||
361 		    blk_queue_dying(req->q))
362 			return FAILOVER;
363 	} else {
364 		if (blk_queue_dying(req->q))
365 			return COMPLETE;
366 	}
367 
368 	return RETRY;
369 }
370 
371 static inline void nvme_end_req_zoned(struct request *req)
372 {
373 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
374 	    req_op(req) == REQ_OP_ZONE_APPEND)
375 		req->__sector = nvme_lba_to_sect(req->q->queuedata,
376 			le64_to_cpu(nvme_req(req)->result.u64));
377 }
378 
379 static inline void nvme_end_req(struct request *req)
380 {
381 	blk_status_t status = nvme_error_status(nvme_req(req)->status);
382 
383 	if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
384 		nvme_log_error(req);
385 	nvme_end_req_zoned(req);
386 	nvme_trace_bio_complete(req);
387 	blk_mq_end_request(req, status);
388 }
389 
390 void nvme_complete_rq(struct request *req)
391 {
392 	struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
393 
394 	trace_nvme_complete_rq(req);
395 	nvme_cleanup_cmd(req);
396 
397 	if (ctrl->kas)
398 		ctrl->comp_seen = true;
399 
400 	switch (nvme_decide_disposition(req)) {
401 	case COMPLETE:
402 		nvme_end_req(req);
403 		return;
404 	case RETRY:
405 		nvme_retry_req(req);
406 		return;
407 	case FAILOVER:
408 		nvme_failover_req(req);
409 		return;
410 	case AUTHENTICATE:
411 #ifdef CONFIG_NVME_AUTH
412 		queue_work(nvme_wq, &ctrl->dhchap_auth_work);
413 		nvme_retry_req(req);
414 #else
415 		nvme_end_req(req);
416 #endif
417 		return;
418 	}
419 }
420 EXPORT_SYMBOL_GPL(nvme_complete_rq);
421 
422 void nvme_complete_batch_req(struct request *req)
423 {
424 	trace_nvme_complete_rq(req);
425 	nvme_cleanup_cmd(req);
426 	nvme_end_req_zoned(req);
427 }
428 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
429 
430 /*
431  * Called to unwind from ->queue_rq on a failed command submission so that the
432  * multipathing code gets called to potentially failover to another path.
433  * The caller needs to unwind all transport specific resource allocations and
434  * must return propagate the return value.
435  */
436 blk_status_t nvme_host_path_error(struct request *req)
437 {
438 	nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
439 	blk_mq_set_request_complete(req);
440 	nvme_complete_rq(req);
441 	return BLK_STS_OK;
442 }
443 EXPORT_SYMBOL_GPL(nvme_host_path_error);
444 
445 bool nvme_cancel_request(struct request *req, void *data)
446 {
447 	dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
448 				"Cancelling I/O %d", req->tag);
449 
450 	/* don't abort one completed request */
451 	if (blk_mq_request_completed(req))
452 		return true;
453 
454 	nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
455 	nvme_req(req)->flags |= NVME_REQ_CANCELLED;
456 	blk_mq_complete_request(req);
457 	return true;
458 }
459 EXPORT_SYMBOL_GPL(nvme_cancel_request);
460 
461 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
462 {
463 	if (ctrl->tagset) {
464 		blk_mq_tagset_busy_iter(ctrl->tagset,
465 				nvme_cancel_request, ctrl);
466 		blk_mq_tagset_wait_completed_request(ctrl->tagset);
467 	}
468 }
469 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
470 
471 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
472 {
473 	if (ctrl->admin_tagset) {
474 		blk_mq_tagset_busy_iter(ctrl->admin_tagset,
475 				nvme_cancel_request, ctrl);
476 		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
477 	}
478 }
479 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
480 
481 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
482 		enum nvme_ctrl_state new_state)
483 {
484 	enum nvme_ctrl_state old_state;
485 	unsigned long flags;
486 	bool changed = false;
487 
488 	spin_lock_irqsave(&ctrl->lock, flags);
489 
490 	old_state = ctrl->state;
491 	switch (new_state) {
492 	case NVME_CTRL_LIVE:
493 		switch (old_state) {
494 		case NVME_CTRL_NEW:
495 		case NVME_CTRL_RESETTING:
496 		case NVME_CTRL_CONNECTING:
497 			changed = true;
498 			fallthrough;
499 		default:
500 			break;
501 		}
502 		break;
503 	case NVME_CTRL_RESETTING:
504 		switch (old_state) {
505 		case NVME_CTRL_NEW:
506 		case NVME_CTRL_LIVE:
507 			changed = true;
508 			fallthrough;
509 		default:
510 			break;
511 		}
512 		break;
513 	case NVME_CTRL_CONNECTING:
514 		switch (old_state) {
515 		case NVME_CTRL_NEW:
516 		case NVME_CTRL_RESETTING:
517 			changed = true;
518 			fallthrough;
519 		default:
520 			break;
521 		}
522 		break;
523 	case NVME_CTRL_DELETING:
524 		switch (old_state) {
525 		case NVME_CTRL_LIVE:
526 		case NVME_CTRL_RESETTING:
527 		case NVME_CTRL_CONNECTING:
528 			changed = true;
529 			fallthrough;
530 		default:
531 			break;
532 		}
533 		break;
534 	case NVME_CTRL_DELETING_NOIO:
535 		switch (old_state) {
536 		case NVME_CTRL_DELETING:
537 		case NVME_CTRL_DEAD:
538 			changed = true;
539 			fallthrough;
540 		default:
541 			break;
542 		}
543 		break;
544 	case NVME_CTRL_DEAD:
545 		switch (old_state) {
546 		case NVME_CTRL_DELETING:
547 			changed = true;
548 			fallthrough;
549 		default:
550 			break;
551 		}
552 		break;
553 	default:
554 		break;
555 	}
556 
557 	if (changed) {
558 		ctrl->state = new_state;
559 		wake_up_all(&ctrl->state_wq);
560 	}
561 
562 	spin_unlock_irqrestore(&ctrl->lock, flags);
563 	if (!changed)
564 		return false;
565 
566 	if (ctrl->state == NVME_CTRL_LIVE) {
567 		if (old_state == NVME_CTRL_CONNECTING)
568 			nvme_stop_failfast_work(ctrl);
569 		nvme_kick_requeue_lists(ctrl);
570 	} else if (ctrl->state == NVME_CTRL_CONNECTING &&
571 		old_state == NVME_CTRL_RESETTING) {
572 		nvme_start_failfast_work(ctrl);
573 	}
574 	return changed;
575 }
576 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
577 
578 /*
579  * Returns true for sink states that can't ever transition back to live.
580  */
581 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
582 {
583 	switch (ctrl->state) {
584 	case NVME_CTRL_NEW:
585 	case NVME_CTRL_LIVE:
586 	case NVME_CTRL_RESETTING:
587 	case NVME_CTRL_CONNECTING:
588 		return false;
589 	case NVME_CTRL_DELETING:
590 	case NVME_CTRL_DELETING_NOIO:
591 	case NVME_CTRL_DEAD:
592 		return true;
593 	default:
594 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
595 		return true;
596 	}
597 }
598 
599 /*
600  * Waits for the controller state to be resetting, or returns false if it is
601  * not possible to ever transition to that state.
602  */
603 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
604 {
605 	wait_event(ctrl->state_wq,
606 		   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
607 		   nvme_state_terminal(ctrl));
608 	return ctrl->state == NVME_CTRL_RESETTING;
609 }
610 EXPORT_SYMBOL_GPL(nvme_wait_reset);
611 
612 static void nvme_free_ns_head(struct kref *ref)
613 {
614 	struct nvme_ns_head *head =
615 		container_of(ref, struct nvme_ns_head, ref);
616 
617 	nvme_mpath_remove_disk(head);
618 	ida_free(&head->subsys->ns_ida, head->instance);
619 	cleanup_srcu_struct(&head->srcu);
620 	nvme_put_subsystem(head->subsys);
621 	kfree(head);
622 }
623 
624 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
625 {
626 	return kref_get_unless_zero(&head->ref);
627 }
628 
629 void nvme_put_ns_head(struct nvme_ns_head *head)
630 {
631 	kref_put(&head->ref, nvme_free_ns_head);
632 }
633 
634 static void nvme_free_ns(struct kref *kref)
635 {
636 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
637 
638 	put_disk(ns->disk);
639 	nvme_put_ns_head(ns->head);
640 	nvme_put_ctrl(ns->ctrl);
641 	kfree(ns);
642 }
643 
644 static inline bool nvme_get_ns(struct nvme_ns *ns)
645 {
646 	return kref_get_unless_zero(&ns->kref);
647 }
648 
649 void nvme_put_ns(struct nvme_ns *ns)
650 {
651 	kref_put(&ns->kref, nvme_free_ns);
652 }
653 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
654 
655 static inline void nvme_clear_nvme_request(struct request *req)
656 {
657 	nvme_req(req)->status = 0;
658 	nvme_req(req)->retries = 0;
659 	nvme_req(req)->flags = 0;
660 	req->rq_flags |= RQF_DONTPREP;
661 }
662 
663 /* initialize a passthrough request */
664 void nvme_init_request(struct request *req, struct nvme_command *cmd)
665 {
666 	if (req->q->queuedata)
667 		req->timeout = NVME_IO_TIMEOUT;
668 	else /* no queuedata implies admin queue */
669 		req->timeout = NVME_ADMIN_TIMEOUT;
670 
671 	/* passthru commands should let the driver set the SGL flags */
672 	cmd->common.flags &= ~NVME_CMD_SGL_ALL;
673 
674 	req->cmd_flags |= REQ_FAILFAST_DRIVER;
675 	if (req->mq_hctx->type == HCTX_TYPE_POLL)
676 		req->cmd_flags |= REQ_POLLED;
677 	nvme_clear_nvme_request(req);
678 	memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
679 }
680 EXPORT_SYMBOL_GPL(nvme_init_request);
681 
682 /*
683  * For something we're not in a state to send to the device the default action
684  * is to busy it and retry it after the controller state is recovered.  However,
685  * if the controller is deleting or if anything is marked for failfast or
686  * nvme multipath it is immediately failed.
687  *
688  * Note: commands used to initialize the controller will be marked for failfast.
689  * Note: nvme cli/ioctl commands are marked for failfast.
690  */
691 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
692 		struct request *rq)
693 {
694 	if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
695 	    ctrl->state != NVME_CTRL_DELETING &&
696 	    ctrl->state != NVME_CTRL_DEAD &&
697 	    !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
698 	    !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
699 		return BLK_STS_RESOURCE;
700 	return nvme_host_path_error(rq);
701 }
702 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
703 
704 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
705 		bool queue_live)
706 {
707 	struct nvme_request *req = nvme_req(rq);
708 
709 	/*
710 	 * currently we have a problem sending passthru commands
711 	 * on the admin_q if the controller is not LIVE because we can't
712 	 * make sure that they are going out after the admin connect,
713 	 * controller enable and/or other commands in the initialization
714 	 * sequence. until the controller will be LIVE, fail with
715 	 * BLK_STS_RESOURCE so that they will be rescheduled.
716 	 */
717 	if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
718 		return false;
719 
720 	if (ctrl->ops->flags & NVME_F_FABRICS) {
721 		/*
722 		 * Only allow commands on a live queue, except for the connect
723 		 * command, which is require to set the queue live in the
724 		 * appropinquate states.
725 		 */
726 		switch (ctrl->state) {
727 		case NVME_CTRL_CONNECTING:
728 			if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
729 			    (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
730 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
731 			     req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
732 				return true;
733 			break;
734 		default:
735 			break;
736 		case NVME_CTRL_DEAD:
737 			return false;
738 		}
739 	}
740 
741 	return queue_live;
742 }
743 EXPORT_SYMBOL_GPL(__nvme_check_ready);
744 
745 static inline void nvme_setup_flush(struct nvme_ns *ns,
746 		struct nvme_command *cmnd)
747 {
748 	memset(cmnd, 0, sizeof(*cmnd));
749 	cmnd->common.opcode = nvme_cmd_flush;
750 	cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
751 }
752 
753 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
754 		struct nvme_command *cmnd)
755 {
756 	unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
757 	struct nvme_dsm_range *range;
758 	struct bio *bio;
759 
760 	/*
761 	 * Some devices do not consider the DSM 'Number of Ranges' field when
762 	 * determining how much data to DMA. Always allocate memory for maximum
763 	 * number of segments to prevent device reading beyond end of buffer.
764 	 */
765 	static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
766 
767 	range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
768 	if (!range) {
769 		/*
770 		 * If we fail allocation our range, fallback to the controller
771 		 * discard page. If that's also busy, it's safe to return
772 		 * busy, as we know we can make progress once that's freed.
773 		 */
774 		if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
775 			return BLK_STS_RESOURCE;
776 
777 		range = page_address(ns->ctrl->discard_page);
778 	}
779 
780 	__rq_for_each_bio(bio, req) {
781 		u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
782 		u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
783 
784 		if (n < segments) {
785 			range[n].cattr = cpu_to_le32(0);
786 			range[n].nlb = cpu_to_le32(nlb);
787 			range[n].slba = cpu_to_le64(slba);
788 		}
789 		n++;
790 	}
791 
792 	if (WARN_ON_ONCE(n != segments)) {
793 		if (virt_to_page(range) == ns->ctrl->discard_page)
794 			clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
795 		else
796 			kfree(range);
797 		return BLK_STS_IOERR;
798 	}
799 
800 	memset(cmnd, 0, sizeof(*cmnd));
801 	cmnd->dsm.opcode = nvme_cmd_dsm;
802 	cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
803 	cmnd->dsm.nr = cpu_to_le32(segments - 1);
804 	cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
805 
806 	req->special_vec.bv_page = virt_to_page(range);
807 	req->special_vec.bv_offset = offset_in_page(range);
808 	req->special_vec.bv_len = alloc_size;
809 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
810 
811 	return BLK_STS_OK;
812 }
813 
814 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
815 			      struct request *req)
816 {
817 	u32 upper, lower;
818 	u64 ref48;
819 
820 	/* both rw and write zeroes share the same reftag format */
821 	switch (ns->guard_type) {
822 	case NVME_NVM_NS_16B_GUARD:
823 		cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
824 		break;
825 	case NVME_NVM_NS_64B_GUARD:
826 		ref48 = ext_pi_ref_tag(req);
827 		lower = lower_32_bits(ref48);
828 		upper = upper_32_bits(ref48);
829 
830 		cmnd->rw.reftag = cpu_to_le32(lower);
831 		cmnd->rw.cdw3 = cpu_to_le32(upper);
832 		break;
833 	default:
834 		break;
835 	}
836 }
837 
838 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
839 		struct request *req, struct nvme_command *cmnd)
840 {
841 	memset(cmnd, 0, sizeof(*cmnd));
842 
843 	if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
844 		return nvme_setup_discard(ns, req, cmnd);
845 
846 	cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
847 	cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
848 	cmnd->write_zeroes.slba =
849 		cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
850 	cmnd->write_zeroes.length =
851 		cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
852 
853 	if (nvme_ns_has_pi(ns)) {
854 		cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
855 
856 		switch (ns->pi_type) {
857 		case NVME_NS_DPS_PI_TYPE1:
858 		case NVME_NS_DPS_PI_TYPE2:
859 			nvme_set_ref_tag(ns, cmnd, req);
860 			break;
861 		}
862 	}
863 
864 	return BLK_STS_OK;
865 }
866 
867 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
868 		struct request *req, struct nvme_command *cmnd,
869 		enum nvme_opcode op)
870 {
871 	u16 control = 0;
872 	u32 dsmgmt = 0;
873 
874 	if (req->cmd_flags & REQ_FUA)
875 		control |= NVME_RW_FUA;
876 	if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
877 		control |= NVME_RW_LR;
878 
879 	if (req->cmd_flags & REQ_RAHEAD)
880 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
881 
882 	cmnd->rw.opcode = op;
883 	cmnd->rw.flags = 0;
884 	cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
885 	cmnd->rw.cdw2 = 0;
886 	cmnd->rw.cdw3 = 0;
887 	cmnd->rw.metadata = 0;
888 	cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
889 	cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
890 	cmnd->rw.reftag = 0;
891 	cmnd->rw.apptag = 0;
892 	cmnd->rw.appmask = 0;
893 
894 	if (ns->ms) {
895 		/*
896 		 * If formated with metadata, the block layer always provides a
897 		 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
898 		 * we enable the PRACT bit for protection information or set the
899 		 * namespace capacity to zero to prevent any I/O.
900 		 */
901 		if (!blk_integrity_rq(req)) {
902 			if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
903 				return BLK_STS_NOTSUPP;
904 			control |= NVME_RW_PRINFO_PRACT;
905 		}
906 
907 		switch (ns->pi_type) {
908 		case NVME_NS_DPS_PI_TYPE3:
909 			control |= NVME_RW_PRINFO_PRCHK_GUARD;
910 			break;
911 		case NVME_NS_DPS_PI_TYPE1:
912 		case NVME_NS_DPS_PI_TYPE2:
913 			control |= NVME_RW_PRINFO_PRCHK_GUARD |
914 					NVME_RW_PRINFO_PRCHK_REF;
915 			if (op == nvme_cmd_zone_append)
916 				control |= NVME_RW_APPEND_PIREMAP;
917 			nvme_set_ref_tag(ns, cmnd, req);
918 			break;
919 		}
920 	}
921 
922 	cmnd->rw.control = cpu_to_le16(control);
923 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
924 	return 0;
925 }
926 
927 void nvme_cleanup_cmd(struct request *req)
928 {
929 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
930 		struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
931 
932 		if (req->special_vec.bv_page == ctrl->discard_page)
933 			clear_bit_unlock(0, &ctrl->discard_page_busy);
934 		else
935 			kfree(bvec_virt(&req->special_vec));
936 	}
937 }
938 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
939 
940 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
941 {
942 	struct nvme_command *cmd = nvme_req(req)->cmd;
943 	blk_status_t ret = BLK_STS_OK;
944 
945 	if (!(req->rq_flags & RQF_DONTPREP))
946 		nvme_clear_nvme_request(req);
947 
948 	switch (req_op(req)) {
949 	case REQ_OP_DRV_IN:
950 	case REQ_OP_DRV_OUT:
951 		/* these are setup prior to execution in nvme_init_request() */
952 		break;
953 	case REQ_OP_FLUSH:
954 		nvme_setup_flush(ns, cmd);
955 		break;
956 	case REQ_OP_ZONE_RESET_ALL:
957 	case REQ_OP_ZONE_RESET:
958 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
959 		break;
960 	case REQ_OP_ZONE_OPEN:
961 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
962 		break;
963 	case REQ_OP_ZONE_CLOSE:
964 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
965 		break;
966 	case REQ_OP_ZONE_FINISH:
967 		ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
968 		break;
969 	case REQ_OP_WRITE_ZEROES:
970 		ret = nvme_setup_write_zeroes(ns, req, cmd);
971 		break;
972 	case REQ_OP_DISCARD:
973 		ret = nvme_setup_discard(ns, req, cmd);
974 		break;
975 	case REQ_OP_READ:
976 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
977 		break;
978 	case REQ_OP_WRITE:
979 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
980 		break;
981 	case REQ_OP_ZONE_APPEND:
982 		ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
983 		break;
984 	default:
985 		WARN_ON_ONCE(1);
986 		return BLK_STS_IOERR;
987 	}
988 
989 	cmd->common.command_id = nvme_cid(req);
990 	trace_nvme_setup_cmd(req, cmd);
991 	return ret;
992 }
993 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
994 
995 /*
996  * Return values:
997  * 0:  success
998  * >0: nvme controller's cqe status response
999  * <0: kernel error in lieu of controller response
1000  */
1001 static int nvme_execute_rq(struct request *rq, bool at_head)
1002 {
1003 	blk_status_t status;
1004 
1005 	status = blk_execute_rq(rq, at_head);
1006 	if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1007 		return -EINTR;
1008 	if (nvme_req(rq)->status)
1009 		return nvme_req(rq)->status;
1010 	return blk_status_to_errno(status);
1011 }
1012 
1013 /*
1014  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1015  * if the result is positive, it's an NVM Express status code
1016  */
1017 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1018 		union nvme_result *result, void *buffer, unsigned bufflen,
1019 		int qid, int at_head, blk_mq_req_flags_t flags)
1020 {
1021 	struct request *req;
1022 	int ret;
1023 
1024 	if (qid == NVME_QID_ANY)
1025 		req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1026 	else
1027 		req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1028 						qid - 1);
1029 
1030 	if (IS_ERR(req))
1031 		return PTR_ERR(req);
1032 	nvme_init_request(req, cmd);
1033 
1034 	if (buffer && bufflen) {
1035 		ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1036 		if (ret)
1037 			goto out;
1038 	}
1039 
1040 	req->rq_flags |= RQF_QUIET;
1041 	ret = nvme_execute_rq(req, at_head);
1042 	if (result && ret >= 0)
1043 		*result = nvme_req(req)->result;
1044  out:
1045 	blk_mq_free_request(req);
1046 	return ret;
1047 }
1048 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1049 
1050 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1051 		void *buffer, unsigned bufflen)
1052 {
1053 	return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1054 			NVME_QID_ANY, 0, 0);
1055 }
1056 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1057 
1058 static u32 nvme_known_admin_effects(u8 opcode)
1059 {
1060 	switch (opcode) {
1061 	case nvme_admin_format_nvm:
1062 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1063 			NVME_CMD_EFFECTS_CSE_MASK;
1064 	case nvme_admin_sanitize_nvm:
1065 		return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1066 	default:
1067 		break;
1068 	}
1069 	return 0;
1070 }
1071 
1072 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1073 {
1074 	u32 effects = 0;
1075 
1076 	if (ns) {
1077 		if (ns->head->effects)
1078 			effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1079 		if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1080 			dev_warn_once(ctrl->device,
1081 				"IO command:%02x has unhandled effects:%08x\n",
1082 				opcode, effects);
1083 		return 0;
1084 	}
1085 
1086 	if (ctrl->effects)
1087 		effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1088 	effects |= nvme_known_admin_effects(opcode);
1089 
1090 	return effects;
1091 }
1092 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1093 
1094 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1095 			       u8 opcode)
1096 {
1097 	u32 effects = nvme_command_effects(ctrl, ns, opcode);
1098 
1099 	/*
1100 	 * For simplicity, IO to all namespaces is quiesced even if the command
1101 	 * effects say only one namespace is affected.
1102 	 */
1103 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1104 		mutex_lock(&ctrl->scan_lock);
1105 		mutex_lock(&ctrl->subsys->lock);
1106 		nvme_mpath_start_freeze(ctrl->subsys);
1107 		nvme_mpath_wait_freeze(ctrl->subsys);
1108 		nvme_start_freeze(ctrl);
1109 		nvme_wait_freeze(ctrl);
1110 	}
1111 	return effects;
1112 }
1113 
1114 void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1115 		       struct nvme_command *cmd, int status)
1116 {
1117 	if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1118 		nvme_unfreeze(ctrl);
1119 		nvme_mpath_unfreeze(ctrl->subsys);
1120 		mutex_unlock(&ctrl->subsys->lock);
1121 		nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1122 		mutex_unlock(&ctrl->scan_lock);
1123 	}
1124 	if (effects & NVME_CMD_EFFECTS_CCC)
1125 		nvme_init_ctrl_finish(ctrl);
1126 	if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1127 		nvme_queue_scan(ctrl);
1128 		flush_work(&ctrl->scan_work);
1129 	}
1130 
1131 	switch (cmd->common.opcode) {
1132 	case nvme_admin_set_features:
1133 		switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1134 		case NVME_FEAT_KATO:
1135 			/*
1136 			 * Keep alive commands interval on the host should be
1137 			 * updated when KATO is modified by Set Features
1138 			 * commands.
1139 			 */
1140 			if (!status)
1141 				nvme_update_keep_alive(ctrl, cmd);
1142 			break;
1143 		default:
1144 			break;
1145 		}
1146 		break;
1147 	default:
1148 		break;
1149 	}
1150 }
1151 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1152 
1153 int nvme_execute_passthru_rq(struct request *rq, u32 *effects)
1154 {
1155 	struct nvme_command *cmd = nvme_req(rq)->cmd;
1156 	struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1157 	struct nvme_ns *ns = rq->q->queuedata;
1158 
1159 	*effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1160 	return nvme_execute_rq(rq, false);
1161 }
1162 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1163 
1164 /*
1165  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1166  *
1167  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1168  *   accounting for transport roundtrip times [..].
1169  */
1170 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1171 {
1172 	queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1173 }
1174 
1175 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1176 						 blk_status_t status)
1177 {
1178 	struct nvme_ctrl *ctrl = rq->end_io_data;
1179 	unsigned long flags;
1180 	bool startka = false;
1181 
1182 	blk_mq_free_request(rq);
1183 
1184 	if (status) {
1185 		dev_err(ctrl->device,
1186 			"failed nvme_keep_alive_end_io error=%d\n",
1187 				status);
1188 		return RQ_END_IO_NONE;
1189 	}
1190 
1191 	ctrl->comp_seen = false;
1192 	spin_lock_irqsave(&ctrl->lock, flags);
1193 	if (ctrl->state == NVME_CTRL_LIVE ||
1194 	    ctrl->state == NVME_CTRL_CONNECTING)
1195 		startka = true;
1196 	spin_unlock_irqrestore(&ctrl->lock, flags);
1197 	if (startka)
1198 		nvme_queue_keep_alive_work(ctrl);
1199 	return RQ_END_IO_NONE;
1200 }
1201 
1202 static void nvme_keep_alive_work(struct work_struct *work)
1203 {
1204 	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1205 			struct nvme_ctrl, ka_work);
1206 	bool comp_seen = ctrl->comp_seen;
1207 	struct request *rq;
1208 
1209 	if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1210 		dev_dbg(ctrl->device,
1211 			"reschedule traffic based keep-alive timer\n");
1212 		ctrl->comp_seen = false;
1213 		nvme_queue_keep_alive_work(ctrl);
1214 		return;
1215 	}
1216 
1217 	rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1218 				  BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1219 	if (IS_ERR(rq)) {
1220 		/* allocation failure, reset the controller */
1221 		dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1222 		nvme_reset_ctrl(ctrl);
1223 		return;
1224 	}
1225 	nvme_init_request(rq, &ctrl->ka_cmd);
1226 
1227 	rq->timeout = ctrl->kato * HZ;
1228 	rq->end_io = nvme_keep_alive_end_io;
1229 	rq->end_io_data = ctrl;
1230 	rq->rq_flags |= RQF_QUIET;
1231 	blk_execute_rq_nowait(rq, false);
1232 }
1233 
1234 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1235 {
1236 	if (unlikely(ctrl->kato == 0))
1237 		return;
1238 
1239 	nvme_queue_keep_alive_work(ctrl);
1240 }
1241 
1242 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1243 {
1244 	if (unlikely(ctrl->kato == 0))
1245 		return;
1246 
1247 	cancel_delayed_work_sync(&ctrl->ka_work);
1248 }
1249 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1250 
1251 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1252 				   struct nvme_command *cmd)
1253 {
1254 	unsigned int new_kato =
1255 		DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1256 
1257 	dev_info(ctrl->device,
1258 		 "keep alive interval updated from %u ms to %u ms\n",
1259 		 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1260 
1261 	nvme_stop_keep_alive(ctrl);
1262 	ctrl->kato = new_kato;
1263 	nvme_start_keep_alive(ctrl);
1264 }
1265 
1266 /*
1267  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1268  * flag, thus sending any new CNS opcodes has a big chance of not working.
1269  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1270  * (but not for any later version).
1271  */
1272 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1273 {
1274 	if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1275 		return ctrl->vs < NVME_VS(1, 2, 0);
1276 	return ctrl->vs < NVME_VS(1, 1, 0);
1277 }
1278 
1279 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1280 {
1281 	struct nvme_command c = { };
1282 	int error;
1283 
1284 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1285 	c.identify.opcode = nvme_admin_identify;
1286 	c.identify.cns = NVME_ID_CNS_CTRL;
1287 
1288 	*id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1289 	if (!*id)
1290 		return -ENOMEM;
1291 
1292 	error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1293 			sizeof(struct nvme_id_ctrl));
1294 	if (error)
1295 		kfree(*id);
1296 	return error;
1297 }
1298 
1299 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1300 		struct nvme_ns_id_desc *cur, bool *csi_seen)
1301 {
1302 	const char *warn_str = "ctrl returned bogus length:";
1303 	void *data = cur;
1304 
1305 	switch (cur->nidt) {
1306 	case NVME_NIDT_EUI64:
1307 		if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1308 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1309 				 warn_str, cur->nidl);
1310 			return -1;
1311 		}
1312 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1313 			return NVME_NIDT_EUI64_LEN;
1314 		memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1315 		return NVME_NIDT_EUI64_LEN;
1316 	case NVME_NIDT_NGUID:
1317 		if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1318 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1319 				 warn_str, cur->nidl);
1320 			return -1;
1321 		}
1322 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1323 			return NVME_NIDT_NGUID_LEN;
1324 		memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1325 		return NVME_NIDT_NGUID_LEN;
1326 	case NVME_NIDT_UUID:
1327 		if (cur->nidl != NVME_NIDT_UUID_LEN) {
1328 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1329 				 warn_str, cur->nidl);
1330 			return -1;
1331 		}
1332 		if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1333 			return NVME_NIDT_UUID_LEN;
1334 		uuid_copy(&ids->uuid, data + sizeof(*cur));
1335 		return NVME_NIDT_UUID_LEN;
1336 	case NVME_NIDT_CSI:
1337 		if (cur->nidl != NVME_NIDT_CSI_LEN) {
1338 			dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1339 				 warn_str, cur->nidl);
1340 			return -1;
1341 		}
1342 		memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1343 		*csi_seen = true;
1344 		return NVME_NIDT_CSI_LEN;
1345 	default:
1346 		/* Skip unknown types */
1347 		return cur->nidl;
1348 	}
1349 }
1350 
1351 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1352 		struct nvme_ns_info *info)
1353 {
1354 	struct nvme_command c = { };
1355 	bool csi_seen = false;
1356 	int status, pos, len;
1357 	void *data;
1358 
1359 	if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1360 		return 0;
1361 	if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1362 		return 0;
1363 
1364 	c.identify.opcode = nvme_admin_identify;
1365 	c.identify.nsid = cpu_to_le32(info->nsid);
1366 	c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1367 
1368 	data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1369 	if (!data)
1370 		return -ENOMEM;
1371 
1372 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1373 				      NVME_IDENTIFY_DATA_SIZE);
1374 	if (status) {
1375 		dev_warn(ctrl->device,
1376 			"Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1377 			info->nsid, status);
1378 		goto free_data;
1379 	}
1380 
1381 	for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1382 		struct nvme_ns_id_desc *cur = data + pos;
1383 
1384 		if (cur->nidl == 0)
1385 			break;
1386 
1387 		len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1388 		if (len < 0)
1389 			break;
1390 
1391 		len += sizeof(*cur);
1392 	}
1393 
1394 	if (nvme_multi_css(ctrl) && !csi_seen) {
1395 		dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1396 			 info->nsid);
1397 		status = -EINVAL;
1398 	}
1399 
1400 free_data:
1401 	kfree(data);
1402 	return status;
1403 }
1404 
1405 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1406 			struct nvme_id_ns **id)
1407 {
1408 	struct nvme_command c = { };
1409 	int error;
1410 
1411 	/* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1412 	c.identify.opcode = nvme_admin_identify;
1413 	c.identify.nsid = cpu_to_le32(nsid);
1414 	c.identify.cns = NVME_ID_CNS_NS;
1415 
1416 	*id = kmalloc(sizeof(**id), GFP_KERNEL);
1417 	if (!*id)
1418 		return -ENOMEM;
1419 
1420 	error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1421 	if (error) {
1422 		dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1423 		goto out_free_id;
1424 	}
1425 
1426 	error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1427 	if ((*id)->ncap == 0) /* namespace not allocated or attached */
1428 		goto out_free_id;
1429 	return 0;
1430 
1431 out_free_id:
1432 	kfree(*id);
1433 	return error;
1434 }
1435 
1436 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1437 		struct nvme_ns_info *info)
1438 {
1439 	struct nvme_ns_ids *ids = &info->ids;
1440 	struct nvme_id_ns *id;
1441 	int ret;
1442 
1443 	ret = nvme_identify_ns(ctrl, info->nsid, &id);
1444 	if (ret)
1445 		return ret;
1446 	info->anagrpid = id->anagrpid;
1447 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1448 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1449 	info->is_ready = true;
1450 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1451 		dev_info(ctrl->device,
1452 			 "Ignoring bogus Namespace Identifiers\n");
1453 	} else {
1454 		if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1455 		    !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1456 			memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1457 		if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1458 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1459 			memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1460 	}
1461 	kfree(id);
1462 	return 0;
1463 }
1464 
1465 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1466 		struct nvme_ns_info *info)
1467 {
1468 	struct nvme_id_ns_cs_indep *id;
1469 	struct nvme_command c = {
1470 		.identify.opcode	= nvme_admin_identify,
1471 		.identify.nsid		= cpu_to_le32(info->nsid),
1472 		.identify.cns		= NVME_ID_CNS_NS_CS_INDEP,
1473 	};
1474 	int ret;
1475 
1476 	id = kmalloc(sizeof(*id), GFP_KERNEL);
1477 	if (!id)
1478 		return -ENOMEM;
1479 
1480 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1481 	if (!ret) {
1482 		info->anagrpid = id->anagrpid;
1483 		info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1484 		info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1485 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1486 	}
1487 	kfree(id);
1488 	return ret;
1489 }
1490 
1491 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1492 		unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1493 {
1494 	union nvme_result res = { 0 };
1495 	struct nvme_command c = { };
1496 	int ret;
1497 
1498 	c.features.opcode = op;
1499 	c.features.fid = cpu_to_le32(fid);
1500 	c.features.dword11 = cpu_to_le32(dword11);
1501 
1502 	ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1503 			buffer, buflen, NVME_QID_ANY, 0, 0);
1504 	if (ret >= 0 && result)
1505 		*result = le32_to_cpu(res.u32);
1506 	return ret;
1507 }
1508 
1509 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1510 		      unsigned int dword11, void *buffer, size_t buflen,
1511 		      u32 *result)
1512 {
1513 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1514 			     buflen, result);
1515 }
1516 EXPORT_SYMBOL_GPL(nvme_set_features);
1517 
1518 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1519 		      unsigned int dword11, void *buffer, size_t buflen,
1520 		      u32 *result)
1521 {
1522 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1523 			     buflen, result);
1524 }
1525 EXPORT_SYMBOL_GPL(nvme_get_features);
1526 
1527 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1528 {
1529 	u32 q_count = (*count - 1) | ((*count - 1) << 16);
1530 	u32 result;
1531 	int status, nr_io_queues;
1532 
1533 	status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1534 			&result);
1535 	if (status < 0)
1536 		return status;
1537 
1538 	/*
1539 	 * Degraded controllers might return an error when setting the queue
1540 	 * count.  We still want to be able to bring them online and offer
1541 	 * access to the admin queue, as that might be only way to fix them up.
1542 	 */
1543 	if (status > 0) {
1544 		dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1545 		*count = 0;
1546 	} else {
1547 		nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1548 		*count = min(*count, nr_io_queues);
1549 	}
1550 
1551 	return 0;
1552 }
1553 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1554 
1555 #define NVME_AEN_SUPPORTED \
1556 	(NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1557 	 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1558 
1559 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1560 {
1561 	u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1562 	int status;
1563 
1564 	if (!supported_aens)
1565 		return;
1566 
1567 	status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1568 			NULL, 0, &result);
1569 	if (status)
1570 		dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1571 			 supported_aens);
1572 
1573 	queue_work(nvme_wq, &ctrl->async_event_work);
1574 }
1575 
1576 static int nvme_ns_open(struct nvme_ns *ns)
1577 {
1578 
1579 	/* should never be called due to GENHD_FL_HIDDEN */
1580 	if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1581 		goto fail;
1582 	if (!nvme_get_ns(ns))
1583 		goto fail;
1584 	if (!try_module_get(ns->ctrl->ops->module))
1585 		goto fail_put_ns;
1586 
1587 	return 0;
1588 
1589 fail_put_ns:
1590 	nvme_put_ns(ns);
1591 fail:
1592 	return -ENXIO;
1593 }
1594 
1595 static void nvme_ns_release(struct nvme_ns *ns)
1596 {
1597 
1598 	module_put(ns->ctrl->ops->module);
1599 	nvme_put_ns(ns);
1600 }
1601 
1602 static int nvme_open(struct block_device *bdev, fmode_t mode)
1603 {
1604 	return nvme_ns_open(bdev->bd_disk->private_data);
1605 }
1606 
1607 static void nvme_release(struct gendisk *disk, fmode_t mode)
1608 {
1609 	nvme_ns_release(disk->private_data);
1610 }
1611 
1612 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1613 {
1614 	/* some standard values */
1615 	geo->heads = 1 << 6;
1616 	geo->sectors = 1 << 5;
1617 	geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1618 	return 0;
1619 }
1620 
1621 #ifdef CONFIG_BLK_DEV_INTEGRITY
1622 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1623 				u32 max_integrity_segments)
1624 {
1625 	struct blk_integrity integrity = { };
1626 
1627 	switch (ns->pi_type) {
1628 	case NVME_NS_DPS_PI_TYPE3:
1629 		switch (ns->guard_type) {
1630 		case NVME_NVM_NS_16B_GUARD:
1631 			integrity.profile = &t10_pi_type3_crc;
1632 			integrity.tag_size = sizeof(u16) + sizeof(u32);
1633 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1634 			break;
1635 		case NVME_NVM_NS_64B_GUARD:
1636 			integrity.profile = &ext_pi_type3_crc64;
1637 			integrity.tag_size = sizeof(u16) + 6;
1638 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1639 			break;
1640 		default:
1641 			integrity.profile = NULL;
1642 			break;
1643 		}
1644 		break;
1645 	case NVME_NS_DPS_PI_TYPE1:
1646 	case NVME_NS_DPS_PI_TYPE2:
1647 		switch (ns->guard_type) {
1648 		case NVME_NVM_NS_16B_GUARD:
1649 			integrity.profile = &t10_pi_type1_crc;
1650 			integrity.tag_size = sizeof(u16);
1651 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1652 			break;
1653 		case NVME_NVM_NS_64B_GUARD:
1654 			integrity.profile = &ext_pi_type1_crc64;
1655 			integrity.tag_size = sizeof(u16);
1656 			integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1657 			break;
1658 		default:
1659 			integrity.profile = NULL;
1660 			break;
1661 		}
1662 		break;
1663 	default:
1664 		integrity.profile = NULL;
1665 		break;
1666 	}
1667 
1668 	integrity.tuple_size = ns->ms;
1669 	blk_integrity_register(disk, &integrity);
1670 	blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1671 }
1672 #else
1673 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1674 				u32 max_integrity_segments)
1675 {
1676 }
1677 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1678 
1679 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1680 {
1681 	struct nvme_ctrl *ctrl = ns->ctrl;
1682 	struct request_queue *queue = disk->queue;
1683 	u32 size = queue_logical_block_size(queue);
1684 
1685 	if (ctrl->max_discard_sectors == 0) {
1686 		blk_queue_max_discard_sectors(queue, 0);
1687 		return;
1688 	}
1689 
1690 	BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1691 			NVME_DSM_MAX_RANGES);
1692 
1693 	queue->limits.discard_granularity = size;
1694 
1695 	/* If discard is already enabled, don't reset queue limits */
1696 	if (queue->limits.max_discard_sectors)
1697 		return;
1698 
1699 	if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1700 		ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1701 
1702 	blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1703 	blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1704 
1705 	if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1706 		blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1707 }
1708 
1709 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1710 {
1711 	return uuid_equal(&a->uuid, &b->uuid) &&
1712 		memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1713 		memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1714 		a->csi == b->csi;
1715 }
1716 
1717 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1718 {
1719 	bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1720 	unsigned lbaf = nvme_lbaf_index(id->flbas);
1721 	struct nvme_ctrl *ctrl = ns->ctrl;
1722 	struct nvme_command c = { };
1723 	struct nvme_id_ns_nvm *nvm;
1724 	int ret = 0;
1725 	u32 elbaf;
1726 
1727 	ns->pi_size = 0;
1728 	ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1729 	if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1730 		ns->pi_size = sizeof(struct t10_pi_tuple);
1731 		ns->guard_type = NVME_NVM_NS_16B_GUARD;
1732 		goto set_pi;
1733 	}
1734 
1735 	nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1736 	if (!nvm)
1737 		return -ENOMEM;
1738 
1739 	c.identify.opcode = nvme_admin_identify;
1740 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1741 	c.identify.cns = NVME_ID_CNS_CS_NS;
1742 	c.identify.csi = NVME_CSI_NVM;
1743 
1744 	ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1745 	if (ret)
1746 		goto free_data;
1747 
1748 	elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1749 
1750 	/* no support for storage tag formats right now */
1751 	if (nvme_elbaf_sts(elbaf))
1752 		goto free_data;
1753 
1754 	ns->guard_type = nvme_elbaf_guard_type(elbaf);
1755 	switch (ns->guard_type) {
1756 	case NVME_NVM_NS_64B_GUARD:
1757 		ns->pi_size = sizeof(struct crc64_pi_tuple);
1758 		break;
1759 	case NVME_NVM_NS_16B_GUARD:
1760 		ns->pi_size = sizeof(struct t10_pi_tuple);
1761 		break;
1762 	default:
1763 		break;
1764 	}
1765 
1766 free_data:
1767 	kfree(nvm);
1768 set_pi:
1769 	if (ns->pi_size && (first || ns->ms == ns->pi_size))
1770 		ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1771 	else
1772 		ns->pi_type = 0;
1773 
1774 	return ret;
1775 }
1776 
1777 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1778 {
1779 	struct nvme_ctrl *ctrl = ns->ctrl;
1780 
1781 	if (nvme_init_ms(ns, id))
1782 		return;
1783 
1784 	ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1785 	if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1786 		return;
1787 
1788 	if (ctrl->ops->flags & NVME_F_FABRICS) {
1789 		/*
1790 		 * The NVMe over Fabrics specification only supports metadata as
1791 		 * part of the extended data LBA.  We rely on HCA/HBA support to
1792 		 * remap the separate metadata buffer from the block layer.
1793 		 */
1794 		if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1795 			return;
1796 
1797 		ns->features |= NVME_NS_EXT_LBAS;
1798 
1799 		/*
1800 		 * The current fabrics transport drivers support namespace
1801 		 * metadata formats only if nvme_ns_has_pi() returns true.
1802 		 * Suppress support for all other formats so the namespace will
1803 		 * have a 0 capacity and not be usable through the block stack.
1804 		 *
1805 		 * Note, this check will need to be modified if any drivers
1806 		 * gain the ability to use other metadata formats.
1807 		 */
1808 		if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1809 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1810 	} else {
1811 		/*
1812 		 * For PCIe controllers, we can't easily remap the separate
1813 		 * metadata buffer from the block layer and thus require a
1814 		 * separate metadata buffer for block layer metadata/PI support.
1815 		 * We allow extended LBAs for the passthrough interface, though.
1816 		 */
1817 		if (id->flbas & NVME_NS_FLBAS_META_EXT)
1818 			ns->features |= NVME_NS_EXT_LBAS;
1819 		else
1820 			ns->features |= NVME_NS_METADATA_SUPPORTED;
1821 	}
1822 }
1823 
1824 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1825 		struct request_queue *q)
1826 {
1827 	bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1828 
1829 	if (ctrl->max_hw_sectors) {
1830 		u32 max_segments =
1831 			(ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1832 
1833 		max_segments = min_not_zero(max_segments, ctrl->max_segments);
1834 		blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1835 		blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1836 	}
1837 	blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1838 	blk_queue_dma_alignment(q, 3);
1839 	blk_queue_write_cache(q, vwc, vwc);
1840 }
1841 
1842 static void nvme_update_disk_info(struct gendisk *disk,
1843 		struct nvme_ns *ns, struct nvme_id_ns *id)
1844 {
1845 	sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1846 	unsigned short bs = 1 << ns->lba_shift;
1847 	u32 atomic_bs, phys_bs, io_opt = 0;
1848 
1849 	/*
1850 	 * The block layer can't support LBA sizes larger than the page size
1851 	 * yet, so catch this early and don't allow block I/O.
1852 	 */
1853 	if (ns->lba_shift > PAGE_SHIFT) {
1854 		capacity = 0;
1855 		bs = (1 << 9);
1856 	}
1857 
1858 	blk_integrity_unregister(disk);
1859 
1860 	atomic_bs = phys_bs = bs;
1861 	if (id->nabo == 0) {
1862 		/*
1863 		 * Bit 1 indicates whether NAWUPF is defined for this namespace
1864 		 * and whether it should be used instead of AWUPF. If NAWUPF ==
1865 		 * 0 then AWUPF must be used instead.
1866 		 */
1867 		if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1868 			atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1869 		else
1870 			atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1871 	}
1872 
1873 	if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1874 		/* NPWG = Namespace Preferred Write Granularity */
1875 		phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1876 		/* NOWS = Namespace Optimal Write Size */
1877 		io_opt = bs * (1 + le16_to_cpu(id->nows));
1878 	}
1879 
1880 	blk_queue_logical_block_size(disk->queue, bs);
1881 	/*
1882 	 * Linux filesystems assume writing a single physical block is
1883 	 * an atomic operation. Hence limit the physical block size to the
1884 	 * value of the Atomic Write Unit Power Fail parameter.
1885 	 */
1886 	blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1887 	blk_queue_io_min(disk->queue, phys_bs);
1888 	blk_queue_io_opt(disk->queue, io_opt);
1889 
1890 	/*
1891 	 * Register a metadata profile for PI, or the plain non-integrity NVMe
1892 	 * metadata masquerading as Type 0 if supported, otherwise reject block
1893 	 * I/O to namespaces with metadata except when the namespace supports
1894 	 * PI, as it can strip/insert in that case.
1895 	 */
1896 	if (ns->ms) {
1897 		if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1898 		    (ns->features & NVME_NS_METADATA_SUPPORTED))
1899 			nvme_init_integrity(disk, ns,
1900 					    ns->ctrl->max_integrity_segments);
1901 		else if (!nvme_ns_has_pi(ns))
1902 			capacity = 0;
1903 	}
1904 
1905 	set_capacity_and_notify(disk, capacity);
1906 
1907 	nvme_config_discard(disk, ns);
1908 	blk_queue_max_write_zeroes_sectors(disk->queue,
1909 					   ns->ctrl->max_zeroes_sectors);
1910 }
1911 
1912 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1913 {
1914 	return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1915 }
1916 
1917 static inline bool nvme_first_scan(struct gendisk *disk)
1918 {
1919 	/* nvme_alloc_ns() scans the disk prior to adding it */
1920 	return !disk_live(disk);
1921 }
1922 
1923 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1924 {
1925 	struct nvme_ctrl *ctrl = ns->ctrl;
1926 	u32 iob;
1927 
1928 	if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1929 	    is_power_of_2(ctrl->max_hw_sectors))
1930 		iob = ctrl->max_hw_sectors;
1931 	else
1932 		iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1933 
1934 	if (!iob)
1935 		return;
1936 
1937 	if (!is_power_of_2(iob)) {
1938 		if (nvme_first_scan(ns->disk))
1939 			pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1940 				ns->disk->disk_name, iob);
1941 		return;
1942 	}
1943 
1944 	if (blk_queue_is_zoned(ns->disk->queue)) {
1945 		if (nvme_first_scan(ns->disk))
1946 			pr_warn("%s: ignoring zoned namespace IO boundary\n",
1947 				ns->disk->disk_name);
1948 		return;
1949 	}
1950 
1951 	blk_queue_chunk_sectors(ns->queue, iob);
1952 }
1953 
1954 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1955 		struct nvme_ns_info *info)
1956 {
1957 	blk_mq_freeze_queue(ns->disk->queue);
1958 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1959 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1960 	blk_mq_unfreeze_queue(ns->disk->queue);
1961 
1962 	if (nvme_ns_head_multipath(ns->head)) {
1963 		blk_mq_freeze_queue(ns->head->disk->queue);
1964 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1965 		nvme_mpath_revalidate_paths(ns);
1966 		blk_stack_limits(&ns->head->disk->queue->limits,
1967 				 &ns->queue->limits, 0);
1968 		ns->head->disk->flags |= GENHD_FL_HIDDEN;
1969 		blk_mq_unfreeze_queue(ns->head->disk->queue);
1970 	}
1971 
1972 	/* Hide the block-interface for these devices */
1973 	ns->disk->flags |= GENHD_FL_HIDDEN;
1974 	set_bit(NVME_NS_READY, &ns->flags);
1975 
1976 	return 0;
1977 }
1978 
1979 static int nvme_update_ns_info_block(struct nvme_ns *ns,
1980 		struct nvme_ns_info *info)
1981 {
1982 	struct nvme_id_ns *id;
1983 	unsigned lbaf;
1984 	int ret;
1985 
1986 	ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
1987 	if (ret)
1988 		return ret;
1989 
1990 	blk_mq_freeze_queue(ns->disk->queue);
1991 	lbaf = nvme_lbaf_index(id->flbas);
1992 	ns->lba_shift = id->lbaf[lbaf].ds;
1993 	nvme_set_queue_limits(ns->ctrl, ns->queue);
1994 
1995 	nvme_configure_metadata(ns, id);
1996 	nvme_set_chunk_sectors(ns, id);
1997 	nvme_update_disk_info(ns->disk, ns, id);
1998 
1999 	if (ns->head->ids.csi == NVME_CSI_ZNS) {
2000 		ret = nvme_update_zone_info(ns, lbaf);
2001 		if (ret) {
2002 			blk_mq_unfreeze_queue(ns->disk->queue);
2003 			goto out;
2004 		}
2005 	}
2006 
2007 	set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2008 	set_bit(NVME_NS_READY, &ns->flags);
2009 	blk_mq_unfreeze_queue(ns->disk->queue);
2010 
2011 	if (blk_queue_is_zoned(ns->queue)) {
2012 		ret = nvme_revalidate_zones(ns);
2013 		if (ret && !nvme_first_scan(ns->disk))
2014 			goto out;
2015 	}
2016 
2017 	if (nvme_ns_head_multipath(ns->head)) {
2018 		blk_mq_freeze_queue(ns->head->disk->queue);
2019 		nvme_update_disk_info(ns->head->disk, ns, id);
2020 		set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2021 		nvme_mpath_revalidate_paths(ns);
2022 		blk_stack_limits(&ns->head->disk->queue->limits,
2023 				 &ns->queue->limits, 0);
2024 		disk_update_readahead(ns->head->disk);
2025 		blk_mq_unfreeze_queue(ns->head->disk->queue);
2026 	}
2027 
2028 	ret = 0;
2029 out:
2030 	/*
2031 	 * If probing fails due an unsupported feature, hide the block device,
2032 	 * but still allow other access.
2033 	 */
2034 	if (ret == -ENODEV) {
2035 		ns->disk->flags |= GENHD_FL_HIDDEN;
2036 		set_bit(NVME_NS_READY, &ns->flags);
2037 		ret = 0;
2038 	}
2039 	kfree(id);
2040 	return ret;
2041 }
2042 
2043 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2044 {
2045 	switch (info->ids.csi) {
2046 	case NVME_CSI_ZNS:
2047 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2048 			dev_info(ns->ctrl->device,
2049 	"block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2050 				info->nsid);
2051 			return nvme_update_ns_info_generic(ns, info);
2052 		}
2053 		return nvme_update_ns_info_block(ns, info);
2054 	case NVME_CSI_NVM:
2055 		return nvme_update_ns_info_block(ns, info);
2056 	default:
2057 		dev_info(ns->ctrl->device,
2058 			"block device for nsid %u not supported (csi %u)\n",
2059 			info->nsid, info->ids.csi);
2060 		return nvme_update_ns_info_generic(ns, info);
2061 	}
2062 }
2063 
2064 static char nvme_pr_type(enum pr_type type)
2065 {
2066 	switch (type) {
2067 	case PR_WRITE_EXCLUSIVE:
2068 		return 1;
2069 	case PR_EXCLUSIVE_ACCESS:
2070 		return 2;
2071 	case PR_WRITE_EXCLUSIVE_REG_ONLY:
2072 		return 3;
2073 	case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2074 		return 4;
2075 	case PR_WRITE_EXCLUSIVE_ALL_REGS:
2076 		return 5;
2077 	case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2078 		return 6;
2079 	default:
2080 		return 0;
2081 	}
2082 }
2083 
2084 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
2085 		struct nvme_command *c, u8 data[16])
2086 {
2087 	struct nvme_ns_head *head = bdev->bd_disk->private_data;
2088 	int srcu_idx = srcu_read_lock(&head->srcu);
2089 	struct nvme_ns *ns = nvme_find_path(head);
2090 	int ret = -EWOULDBLOCK;
2091 
2092 	if (ns) {
2093 		c->common.nsid = cpu_to_le32(ns->head->ns_id);
2094 		ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2095 	}
2096 	srcu_read_unlock(&head->srcu, srcu_idx);
2097 	return ret;
2098 }
2099 
2100 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2101 		u8 data[16])
2102 {
2103 	c->common.nsid = cpu_to_le32(ns->head->ns_id);
2104 	return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2105 }
2106 
2107 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2108 				u64 key, u64 sa_key, u8 op)
2109 {
2110 	struct nvme_command c = { };
2111 	u8 data[16] = { 0, };
2112 
2113 	put_unaligned_le64(key, &data[0]);
2114 	put_unaligned_le64(sa_key, &data[8]);
2115 
2116 	c.common.opcode = op;
2117 	c.common.cdw10 = cpu_to_le32(cdw10);
2118 
2119 	if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2120 	    bdev->bd_disk->fops == &nvme_ns_head_ops)
2121 		return nvme_send_ns_head_pr_command(bdev, &c, data);
2122 	return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2123 }
2124 
2125 static int nvme_pr_register(struct block_device *bdev, u64 old,
2126 		u64 new, unsigned flags)
2127 {
2128 	u32 cdw10;
2129 
2130 	if (flags & ~PR_FL_IGNORE_KEY)
2131 		return -EOPNOTSUPP;
2132 
2133 	cdw10 = old ? 2 : 0;
2134 	cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2135 	cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2136 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2137 }
2138 
2139 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2140 		enum pr_type type, unsigned flags)
2141 {
2142 	u32 cdw10;
2143 
2144 	if (flags & ~PR_FL_IGNORE_KEY)
2145 		return -EOPNOTSUPP;
2146 
2147 	cdw10 = nvme_pr_type(type) << 8;
2148 	cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2149 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2150 }
2151 
2152 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2153 		enum pr_type type, bool abort)
2154 {
2155 	u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2156 
2157 	return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2158 }
2159 
2160 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2161 {
2162 	u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2163 
2164 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2165 }
2166 
2167 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2168 {
2169 	u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2170 
2171 	return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2172 }
2173 
2174 const struct pr_ops nvme_pr_ops = {
2175 	.pr_register	= nvme_pr_register,
2176 	.pr_reserve	= nvme_pr_reserve,
2177 	.pr_release	= nvme_pr_release,
2178 	.pr_preempt	= nvme_pr_preempt,
2179 	.pr_clear	= nvme_pr_clear,
2180 };
2181 
2182 #ifdef CONFIG_BLK_SED_OPAL
2183 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2184 		bool send)
2185 {
2186 	struct nvme_ctrl *ctrl = data;
2187 	struct nvme_command cmd = { };
2188 
2189 	if (send)
2190 		cmd.common.opcode = nvme_admin_security_send;
2191 	else
2192 		cmd.common.opcode = nvme_admin_security_recv;
2193 	cmd.common.nsid = 0;
2194 	cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2195 	cmd.common.cdw11 = cpu_to_le32(len);
2196 
2197 	return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2198 			NVME_QID_ANY, 1, 0);
2199 }
2200 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2201 #endif /* CONFIG_BLK_SED_OPAL */
2202 
2203 #ifdef CONFIG_BLK_DEV_ZONED
2204 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2205 		unsigned int nr_zones, report_zones_cb cb, void *data)
2206 {
2207 	return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2208 			data);
2209 }
2210 #else
2211 #define nvme_report_zones	NULL
2212 #endif /* CONFIG_BLK_DEV_ZONED */
2213 
2214 static const struct block_device_operations nvme_bdev_ops = {
2215 	.owner		= THIS_MODULE,
2216 	.ioctl		= nvme_ioctl,
2217 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
2218 	.open		= nvme_open,
2219 	.release	= nvme_release,
2220 	.getgeo		= nvme_getgeo,
2221 	.report_zones	= nvme_report_zones,
2222 	.pr_ops		= &nvme_pr_ops,
2223 };
2224 
2225 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 timeout, bool enabled)
2226 {
2227 	unsigned long timeout_jiffies = ((timeout + 1) * HZ / 2) + jiffies;
2228 	u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2229 	int ret;
2230 
2231 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2232 		if (csts == ~0)
2233 			return -ENODEV;
2234 		if ((csts & NVME_CSTS_RDY) == bit)
2235 			break;
2236 
2237 		usleep_range(1000, 2000);
2238 		if (fatal_signal_pending(current))
2239 			return -EINTR;
2240 		if (time_after(jiffies, timeout_jiffies)) {
2241 			dev_err(ctrl->device,
2242 				"Device not ready; aborting %s, CSTS=0x%x\n",
2243 				enabled ? "initialisation" : "reset", csts);
2244 			return -ENODEV;
2245 		}
2246 	}
2247 
2248 	return ret;
2249 }
2250 
2251 /*
2252  * If the device has been passed off to us in an enabled state, just clear
2253  * the enabled bit.  The spec says we should set the 'shutdown notification
2254  * bits', but doing so may cause the device to complete commands to the
2255  * admin queue ... and we don't know what memory that might be pointing at!
2256  */
2257 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2258 {
2259 	int ret;
2260 
2261 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2262 	ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2263 
2264 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2265 	if (ret)
2266 		return ret;
2267 
2268 	if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2269 		msleep(NVME_QUIRK_DELAY_AMOUNT);
2270 
2271 	return nvme_wait_ready(ctrl, NVME_CAP_TIMEOUT(ctrl->cap), false);
2272 }
2273 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2274 
2275 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2276 {
2277 	unsigned dev_page_min;
2278 	u32 timeout;
2279 	int ret;
2280 
2281 	ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2282 	if (ret) {
2283 		dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2284 		return ret;
2285 	}
2286 	dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2287 
2288 	if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2289 		dev_err(ctrl->device,
2290 			"Minimum device page size %u too large for host (%u)\n",
2291 			1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2292 		return -ENODEV;
2293 	}
2294 
2295 	if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2296 		ctrl->ctrl_config = NVME_CC_CSS_CSI;
2297 	else
2298 		ctrl->ctrl_config = NVME_CC_CSS_NVM;
2299 
2300 	if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2301 		u32 crto;
2302 
2303 		ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2304 		if (ret) {
2305 			dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2306 				ret);
2307 			return ret;
2308 		}
2309 
2310 		if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2311 			ctrl->ctrl_config |= NVME_CC_CRIME;
2312 			timeout = NVME_CRTO_CRIMT(crto);
2313 		} else {
2314 			timeout = NVME_CRTO_CRWMT(crto);
2315 		}
2316 	} else {
2317 		timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2318 	}
2319 
2320 	ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2321 	ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2322 	ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2323 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2324 	if (ret)
2325 		return ret;
2326 
2327 	/* Flush write to device (required if transport is PCI) */
2328 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2329 	if (ret)
2330 		return ret;
2331 
2332 	ctrl->ctrl_config |= NVME_CC_ENABLE;
2333 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2334 	if (ret)
2335 		return ret;
2336 	return nvme_wait_ready(ctrl, timeout, true);
2337 }
2338 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2339 
2340 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2341 {
2342 	unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2343 	u32 csts;
2344 	int ret;
2345 
2346 	ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2347 	ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2348 
2349 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2350 	if (ret)
2351 		return ret;
2352 
2353 	while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2354 		if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2355 			break;
2356 
2357 		msleep(100);
2358 		if (fatal_signal_pending(current))
2359 			return -EINTR;
2360 		if (time_after(jiffies, timeout)) {
2361 			dev_err(ctrl->device,
2362 				"Device shutdown incomplete; abort shutdown\n");
2363 			return -ENODEV;
2364 		}
2365 	}
2366 
2367 	return ret;
2368 }
2369 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2370 
2371 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2372 {
2373 	__le64 ts;
2374 	int ret;
2375 
2376 	if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2377 		return 0;
2378 
2379 	ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2380 	ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2381 			NULL);
2382 	if (ret)
2383 		dev_warn_once(ctrl->device,
2384 			"could not set timestamp (%d)\n", ret);
2385 	return ret;
2386 }
2387 
2388 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2389 {
2390 	struct nvme_feat_host_behavior *host;
2391 	u8 acre = 0, lbafee = 0;
2392 	int ret;
2393 
2394 	/* Don't bother enabling the feature if retry delay is not reported */
2395 	if (ctrl->crdt[0])
2396 		acre = NVME_ENABLE_ACRE;
2397 	if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2398 		lbafee = NVME_ENABLE_LBAFEE;
2399 
2400 	if (!acre && !lbafee)
2401 		return 0;
2402 
2403 	host = kzalloc(sizeof(*host), GFP_KERNEL);
2404 	if (!host)
2405 		return 0;
2406 
2407 	host->acre = acre;
2408 	host->lbafee = lbafee;
2409 	ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2410 				host, sizeof(*host), NULL);
2411 	kfree(host);
2412 	return ret;
2413 }
2414 
2415 /*
2416  * The function checks whether the given total (exlat + enlat) latency of
2417  * a power state allows the latter to be used as an APST transition target.
2418  * It does so by comparing the latency to the primary and secondary latency
2419  * tolerances defined by module params. If there's a match, the corresponding
2420  * timeout value is returned and the matching tolerance index (1 or 2) is
2421  * reported.
2422  */
2423 static bool nvme_apst_get_transition_time(u64 total_latency,
2424 		u64 *transition_time, unsigned *last_index)
2425 {
2426 	if (total_latency <= apst_primary_latency_tol_us) {
2427 		if (*last_index == 1)
2428 			return false;
2429 		*last_index = 1;
2430 		*transition_time = apst_primary_timeout_ms;
2431 		return true;
2432 	}
2433 	if (apst_secondary_timeout_ms &&
2434 		total_latency <= apst_secondary_latency_tol_us) {
2435 		if (*last_index <= 2)
2436 			return false;
2437 		*last_index = 2;
2438 		*transition_time = apst_secondary_timeout_ms;
2439 		return true;
2440 	}
2441 	return false;
2442 }
2443 
2444 /*
2445  * APST (Autonomous Power State Transition) lets us program a table of power
2446  * state transitions that the controller will perform automatically.
2447  *
2448  * Depending on module params, one of the two supported techniques will be used:
2449  *
2450  * - If the parameters provide explicit timeouts and tolerances, they will be
2451  *   used to build a table with up to 2 non-operational states to transition to.
2452  *   The default parameter values were selected based on the values used by
2453  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2454  *   regeneration of the APST table in the event of switching between external
2455  *   and battery power, the timeouts and tolerances reflect a compromise
2456  *   between values used by Microsoft for AC and battery scenarios.
2457  * - If not, we'll configure the table with a simple heuristic: we are willing
2458  *   to spend at most 2% of the time transitioning between power states.
2459  *   Therefore, when running in any given state, we will enter the next
2460  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2461  *   microseconds, as long as that state's exit latency is under the requested
2462  *   maximum latency.
2463  *
2464  * We will not autonomously enter any non-operational state for which the total
2465  * latency exceeds ps_max_latency_us.
2466  *
2467  * Users can set ps_max_latency_us to zero to turn off APST.
2468  */
2469 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2470 {
2471 	struct nvme_feat_auto_pst *table;
2472 	unsigned apste = 0;
2473 	u64 max_lat_us = 0;
2474 	__le64 target = 0;
2475 	int max_ps = -1;
2476 	int state;
2477 	int ret;
2478 	unsigned last_lt_index = UINT_MAX;
2479 
2480 	/*
2481 	 * If APST isn't supported or if we haven't been initialized yet,
2482 	 * then don't do anything.
2483 	 */
2484 	if (!ctrl->apsta)
2485 		return 0;
2486 
2487 	if (ctrl->npss > 31) {
2488 		dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2489 		return 0;
2490 	}
2491 
2492 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2493 	if (!table)
2494 		return 0;
2495 
2496 	if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2497 		/* Turn off APST. */
2498 		dev_dbg(ctrl->device, "APST disabled\n");
2499 		goto done;
2500 	}
2501 
2502 	/*
2503 	 * Walk through all states from lowest- to highest-power.
2504 	 * According to the spec, lower-numbered states use more power.  NPSS,
2505 	 * despite the name, is the index of the lowest-power state, not the
2506 	 * number of states.
2507 	 */
2508 	for (state = (int)ctrl->npss; state >= 0; state--) {
2509 		u64 total_latency_us, exit_latency_us, transition_ms;
2510 
2511 		if (target)
2512 			table->entries[state] = target;
2513 
2514 		/*
2515 		 * Don't allow transitions to the deepest state if it's quirked
2516 		 * off.
2517 		 */
2518 		if (state == ctrl->npss &&
2519 		    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2520 			continue;
2521 
2522 		/*
2523 		 * Is this state a useful non-operational state for higher-power
2524 		 * states to autonomously transition to?
2525 		 */
2526 		if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2527 			continue;
2528 
2529 		exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2530 		if (exit_latency_us > ctrl->ps_max_latency_us)
2531 			continue;
2532 
2533 		total_latency_us = exit_latency_us +
2534 			le32_to_cpu(ctrl->psd[state].entry_lat);
2535 
2536 		/*
2537 		 * This state is good. It can be used as the APST idle target
2538 		 * for higher power states.
2539 		 */
2540 		if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2541 			if (!nvme_apst_get_transition_time(total_latency_us,
2542 					&transition_ms, &last_lt_index))
2543 				continue;
2544 		} else {
2545 			transition_ms = total_latency_us + 19;
2546 			do_div(transition_ms, 20);
2547 			if (transition_ms > (1 << 24) - 1)
2548 				transition_ms = (1 << 24) - 1;
2549 		}
2550 
2551 		target = cpu_to_le64((state << 3) | (transition_ms << 8));
2552 		if (max_ps == -1)
2553 			max_ps = state;
2554 		if (total_latency_us > max_lat_us)
2555 			max_lat_us = total_latency_us;
2556 	}
2557 
2558 	if (max_ps == -1)
2559 		dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2560 	else
2561 		dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2562 			max_ps, max_lat_us, (int)sizeof(*table), table);
2563 	apste = 1;
2564 
2565 done:
2566 	ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2567 				table, sizeof(*table), NULL);
2568 	if (ret)
2569 		dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2570 	kfree(table);
2571 	return ret;
2572 }
2573 
2574 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2575 {
2576 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2577 	u64 latency;
2578 
2579 	switch (val) {
2580 	case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2581 	case PM_QOS_LATENCY_ANY:
2582 		latency = U64_MAX;
2583 		break;
2584 
2585 	default:
2586 		latency = val;
2587 	}
2588 
2589 	if (ctrl->ps_max_latency_us != latency) {
2590 		ctrl->ps_max_latency_us = latency;
2591 		if (ctrl->state == NVME_CTRL_LIVE)
2592 			nvme_configure_apst(ctrl);
2593 	}
2594 }
2595 
2596 struct nvme_core_quirk_entry {
2597 	/*
2598 	 * NVMe model and firmware strings are padded with spaces.  For
2599 	 * simplicity, strings in the quirk table are padded with NULLs
2600 	 * instead.
2601 	 */
2602 	u16 vid;
2603 	const char *mn;
2604 	const char *fr;
2605 	unsigned long quirks;
2606 };
2607 
2608 static const struct nvme_core_quirk_entry core_quirks[] = {
2609 	{
2610 		/*
2611 		 * This Toshiba device seems to die using any APST states.  See:
2612 		 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2613 		 */
2614 		.vid = 0x1179,
2615 		.mn = "THNSF5256GPUK TOSHIBA",
2616 		.quirks = NVME_QUIRK_NO_APST,
2617 	},
2618 	{
2619 		/*
2620 		 * This LiteON CL1-3D*-Q11 firmware version has a race
2621 		 * condition associated with actions related to suspend to idle
2622 		 * LiteON has resolved the problem in future firmware
2623 		 */
2624 		.vid = 0x14a4,
2625 		.fr = "22301111",
2626 		.quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2627 	},
2628 	{
2629 		/*
2630 		 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2631 		 * aborts I/O during any load, but more easily reproducible
2632 		 * with discards (fstrim).
2633 		 *
2634 		 * The device is left in a state where it is also not possible
2635 		 * to use "nvme set-feature" to disable APST, but booting with
2636 		 * nvme_core.default_ps_max_latency=0 works.
2637 		 */
2638 		.vid = 0x1e0f,
2639 		.mn = "KCD6XVUL6T40",
2640 		.quirks = NVME_QUIRK_NO_APST,
2641 	},
2642 	{
2643 		/*
2644 		 * The external Samsung X5 SSD fails initialization without a
2645 		 * delay before checking if it is ready and has a whole set of
2646 		 * other problems.  To make this even more interesting, it
2647 		 * shares the PCI ID with internal Samsung 970 Evo Plus that
2648 		 * does not need or want these quirks.
2649 		 */
2650 		.vid = 0x144d,
2651 		.mn = "Samsung Portable SSD X5",
2652 		.quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2653 			  NVME_QUIRK_NO_DEEPEST_PS |
2654 			  NVME_QUIRK_IGNORE_DEV_SUBNQN,
2655 	}
2656 };
2657 
2658 /* match is null-terminated but idstr is space-padded. */
2659 static bool string_matches(const char *idstr, const char *match, size_t len)
2660 {
2661 	size_t matchlen;
2662 
2663 	if (!match)
2664 		return true;
2665 
2666 	matchlen = strlen(match);
2667 	WARN_ON_ONCE(matchlen > len);
2668 
2669 	if (memcmp(idstr, match, matchlen))
2670 		return false;
2671 
2672 	for (; matchlen < len; matchlen++)
2673 		if (idstr[matchlen] != ' ')
2674 			return false;
2675 
2676 	return true;
2677 }
2678 
2679 static bool quirk_matches(const struct nvme_id_ctrl *id,
2680 			  const struct nvme_core_quirk_entry *q)
2681 {
2682 	return q->vid == le16_to_cpu(id->vid) &&
2683 		string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2684 		string_matches(id->fr, q->fr, sizeof(id->fr));
2685 }
2686 
2687 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2688 		struct nvme_id_ctrl *id)
2689 {
2690 	size_t nqnlen;
2691 	int off;
2692 
2693 	if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2694 		nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2695 		if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2696 			strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2697 			return;
2698 		}
2699 
2700 		if (ctrl->vs >= NVME_VS(1, 2, 1))
2701 			dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2702 	}
2703 
2704 	/*
2705 	 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2706 	 * Base Specification 2.0.  It is slightly different from the format
2707 	 * specified there due to historic reasons, and we can't change it now.
2708 	 */
2709 	off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2710 			"nqn.2014.08.org.nvmexpress:%04x%04x",
2711 			le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2712 	memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2713 	off += sizeof(id->sn);
2714 	memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2715 	off += sizeof(id->mn);
2716 	memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2717 }
2718 
2719 static void nvme_release_subsystem(struct device *dev)
2720 {
2721 	struct nvme_subsystem *subsys =
2722 		container_of(dev, struct nvme_subsystem, dev);
2723 
2724 	if (subsys->instance >= 0)
2725 		ida_free(&nvme_instance_ida, subsys->instance);
2726 	kfree(subsys);
2727 }
2728 
2729 static void nvme_destroy_subsystem(struct kref *ref)
2730 {
2731 	struct nvme_subsystem *subsys =
2732 			container_of(ref, struct nvme_subsystem, ref);
2733 
2734 	mutex_lock(&nvme_subsystems_lock);
2735 	list_del(&subsys->entry);
2736 	mutex_unlock(&nvme_subsystems_lock);
2737 
2738 	ida_destroy(&subsys->ns_ida);
2739 	device_del(&subsys->dev);
2740 	put_device(&subsys->dev);
2741 }
2742 
2743 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2744 {
2745 	kref_put(&subsys->ref, nvme_destroy_subsystem);
2746 }
2747 
2748 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2749 {
2750 	struct nvme_subsystem *subsys;
2751 
2752 	lockdep_assert_held(&nvme_subsystems_lock);
2753 
2754 	/*
2755 	 * Fail matches for discovery subsystems. This results
2756 	 * in each discovery controller bound to a unique subsystem.
2757 	 * This avoids issues with validating controller values
2758 	 * that can only be true when there is a single unique subsystem.
2759 	 * There may be multiple and completely independent entities
2760 	 * that provide discovery controllers.
2761 	 */
2762 	if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2763 		return NULL;
2764 
2765 	list_for_each_entry(subsys, &nvme_subsystems, entry) {
2766 		if (strcmp(subsys->subnqn, subsysnqn))
2767 			continue;
2768 		if (!kref_get_unless_zero(&subsys->ref))
2769 			continue;
2770 		return subsys;
2771 	}
2772 
2773 	return NULL;
2774 }
2775 
2776 #define SUBSYS_ATTR_RO(_name, _mode, _show)			\
2777 	struct device_attribute subsys_attr_##_name = \
2778 		__ATTR(_name, _mode, _show, NULL)
2779 
2780 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2781 				    struct device_attribute *attr,
2782 				    char *buf)
2783 {
2784 	struct nvme_subsystem *subsys =
2785 		container_of(dev, struct nvme_subsystem, dev);
2786 
2787 	return sysfs_emit(buf, "%s\n", subsys->subnqn);
2788 }
2789 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2790 
2791 static ssize_t nvme_subsys_show_type(struct device *dev,
2792 				    struct device_attribute *attr,
2793 				    char *buf)
2794 {
2795 	struct nvme_subsystem *subsys =
2796 		container_of(dev, struct nvme_subsystem, dev);
2797 
2798 	switch (subsys->subtype) {
2799 	case NVME_NQN_DISC:
2800 		return sysfs_emit(buf, "discovery\n");
2801 	case NVME_NQN_NVME:
2802 		return sysfs_emit(buf, "nvm\n");
2803 	default:
2804 		return sysfs_emit(buf, "reserved\n");
2805 	}
2806 }
2807 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2808 
2809 #define nvme_subsys_show_str_function(field)				\
2810 static ssize_t subsys_##field##_show(struct device *dev,		\
2811 			    struct device_attribute *attr, char *buf)	\
2812 {									\
2813 	struct nvme_subsystem *subsys =					\
2814 		container_of(dev, struct nvme_subsystem, dev);		\
2815 	return sysfs_emit(buf, "%.*s\n",				\
2816 			   (int)sizeof(subsys->field), subsys->field);	\
2817 }									\
2818 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2819 
2820 nvme_subsys_show_str_function(model);
2821 nvme_subsys_show_str_function(serial);
2822 nvme_subsys_show_str_function(firmware_rev);
2823 
2824 static struct attribute *nvme_subsys_attrs[] = {
2825 	&subsys_attr_model.attr,
2826 	&subsys_attr_serial.attr,
2827 	&subsys_attr_firmware_rev.attr,
2828 	&subsys_attr_subsysnqn.attr,
2829 	&subsys_attr_subsystype.attr,
2830 #ifdef CONFIG_NVME_MULTIPATH
2831 	&subsys_attr_iopolicy.attr,
2832 #endif
2833 	NULL,
2834 };
2835 
2836 static const struct attribute_group nvme_subsys_attrs_group = {
2837 	.attrs = nvme_subsys_attrs,
2838 };
2839 
2840 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2841 	&nvme_subsys_attrs_group,
2842 	NULL,
2843 };
2844 
2845 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2846 {
2847 	return ctrl->opts && ctrl->opts->discovery_nqn;
2848 }
2849 
2850 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2851 		struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2852 {
2853 	struct nvme_ctrl *tmp;
2854 
2855 	lockdep_assert_held(&nvme_subsystems_lock);
2856 
2857 	list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2858 		if (nvme_state_terminal(tmp))
2859 			continue;
2860 
2861 		if (tmp->cntlid == ctrl->cntlid) {
2862 			dev_err(ctrl->device,
2863 				"Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2864 				ctrl->cntlid, dev_name(tmp->device),
2865 				subsys->subnqn);
2866 			return false;
2867 		}
2868 
2869 		if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2870 		    nvme_discovery_ctrl(ctrl))
2871 			continue;
2872 
2873 		dev_err(ctrl->device,
2874 			"Subsystem does not support multiple controllers\n");
2875 		return false;
2876 	}
2877 
2878 	return true;
2879 }
2880 
2881 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2882 {
2883 	struct nvme_subsystem *subsys, *found;
2884 	int ret;
2885 
2886 	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2887 	if (!subsys)
2888 		return -ENOMEM;
2889 
2890 	subsys->instance = -1;
2891 	mutex_init(&subsys->lock);
2892 	kref_init(&subsys->ref);
2893 	INIT_LIST_HEAD(&subsys->ctrls);
2894 	INIT_LIST_HEAD(&subsys->nsheads);
2895 	nvme_init_subnqn(subsys, ctrl, id);
2896 	memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2897 	memcpy(subsys->model, id->mn, sizeof(subsys->model));
2898 	subsys->vendor_id = le16_to_cpu(id->vid);
2899 	subsys->cmic = id->cmic;
2900 
2901 	/* Versions prior to 1.4 don't necessarily report a valid type */
2902 	if (id->cntrltype == NVME_CTRL_DISC ||
2903 	    !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2904 		subsys->subtype = NVME_NQN_DISC;
2905 	else
2906 		subsys->subtype = NVME_NQN_NVME;
2907 
2908 	if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2909 		dev_err(ctrl->device,
2910 			"Subsystem %s is not a discovery controller",
2911 			subsys->subnqn);
2912 		kfree(subsys);
2913 		return -EINVAL;
2914 	}
2915 	subsys->awupf = le16_to_cpu(id->awupf);
2916 	nvme_mpath_default_iopolicy(subsys);
2917 
2918 	subsys->dev.class = nvme_subsys_class;
2919 	subsys->dev.release = nvme_release_subsystem;
2920 	subsys->dev.groups = nvme_subsys_attrs_groups;
2921 	dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2922 	device_initialize(&subsys->dev);
2923 
2924 	mutex_lock(&nvme_subsystems_lock);
2925 	found = __nvme_find_get_subsystem(subsys->subnqn);
2926 	if (found) {
2927 		put_device(&subsys->dev);
2928 		subsys = found;
2929 
2930 		if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2931 			ret = -EINVAL;
2932 			goto out_put_subsystem;
2933 		}
2934 	} else {
2935 		ret = device_add(&subsys->dev);
2936 		if (ret) {
2937 			dev_err(ctrl->device,
2938 				"failed to register subsystem device.\n");
2939 			put_device(&subsys->dev);
2940 			goto out_unlock;
2941 		}
2942 		ida_init(&subsys->ns_ida);
2943 		list_add_tail(&subsys->entry, &nvme_subsystems);
2944 	}
2945 
2946 	ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2947 				dev_name(ctrl->device));
2948 	if (ret) {
2949 		dev_err(ctrl->device,
2950 			"failed to create sysfs link from subsystem.\n");
2951 		goto out_put_subsystem;
2952 	}
2953 
2954 	if (!found)
2955 		subsys->instance = ctrl->instance;
2956 	ctrl->subsys = subsys;
2957 	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2958 	mutex_unlock(&nvme_subsystems_lock);
2959 	return 0;
2960 
2961 out_put_subsystem:
2962 	nvme_put_subsystem(subsys);
2963 out_unlock:
2964 	mutex_unlock(&nvme_subsystems_lock);
2965 	return ret;
2966 }
2967 
2968 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2969 		void *log, size_t size, u64 offset)
2970 {
2971 	struct nvme_command c = { };
2972 	u32 dwlen = nvme_bytes_to_numd(size);
2973 
2974 	c.get_log_page.opcode = nvme_admin_get_log_page;
2975 	c.get_log_page.nsid = cpu_to_le32(nsid);
2976 	c.get_log_page.lid = log_page;
2977 	c.get_log_page.lsp = lsp;
2978 	c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2979 	c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2980 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2981 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2982 	c.get_log_page.csi = csi;
2983 
2984 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2985 }
2986 
2987 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2988 				struct nvme_effects_log **log)
2989 {
2990 	struct nvme_effects_log	*cel = xa_load(&ctrl->cels, csi);
2991 	int ret;
2992 
2993 	if (cel)
2994 		goto out;
2995 
2996 	cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2997 	if (!cel)
2998 		return -ENOMEM;
2999 
3000 	ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3001 			cel, sizeof(*cel), 0);
3002 	if (ret) {
3003 		kfree(cel);
3004 		return ret;
3005 	}
3006 
3007 	xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3008 out:
3009 	*log = cel;
3010 	return 0;
3011 }
3012 
3013 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3014 {
3015 	u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3016 
3017 	if (check_shl_overflow(1U, units + page_shift - 9, &val))
3018 		return UINT_MAX;
3019 	return val;
3020 }
3021 
3022 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3023 {
3024 	struct nvme_command c = { };
3025 	struct nvme_id_ctrl_nvm *id;
3026 	int ret;
3027 
3028 	if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3029 		ctrl->max_discard_sectors = UINT_MAX;
3030 		ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3031 	} else {
3032 		ctrl->max_discard_sectors = 0;
3033 		ctrl->max_discard_segments = 0;
3034 	}
3035 
3036 	/*
3037 	 * Even though NVMe spec explicitly states that MDTS is not applicable
3038 	 * to the write-zeroes, we are cautious and limit the size to the
3039 	 * controllers max_hw_sectors value, which is based on the MDTS field
3040 	 * and possibly other limiting factors.
3041 	 */
3042 	if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3043 	    !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3044 		ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3045 	else
3046 		ctrl->max_zeroes_sectors = 0;
3047 
3048 	if (nvme_ctrl_limited_cns(ctrl))
3049 		return 0;
3050 
3051 	id = kzalloc(sizeof(*id), GFP_KERNEL);
3052 	if (!id)
3053 		return 0;
3054 
3055 	c.identify.opcode = nvme_admin_identify;
3056 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
3057 	c.identify.csi = NVME_CSI_NVM;
3058 
3059 	ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3060 	if (ret)
3061 		goto free_data;
3062 
3063 	if (id->dmrl)
3064 		ctrl->max_discard_segments = id->dmrl;
3065 	ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3066 	if (id->wzsl)
3067 		ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3068 
3069 free_data:
3070 	kfree(id);
3071 	return ret;
3072 }
3073 
3074 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3075 {
3076 	struct nvme_id_ctrl *id;
3077 	u32 max_hw_sectors;
3078 	bool prev_apst_enabled;
3079 	int ret;
3080 
3081 	ret = nvme_identify_ctrl(ctrl, &id);
3082 	if (ret) {
3083 		dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3084 		return -EIO;
3085 	}
3086 
3087 	if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3088 		ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3089 		if (ret < 0)
3090 			goto out_free;
3091 	}
3092 
3093 	if (!(ctrl->ops->flags & NVME_F_FABRICS))
3094 		ctrl->cntlid = le16_to_cpu(id->cntlid);
3095 
3096 	if (!ctrl->identified) {
3097 		unsigned int i;
3098 
3099 		ret = nvme_init_subsystem(ctrl, id);
3100 		if (ret)
3101 			goto out_free;
3102 
3103 		/*
3104 		 * Check for quirks.  Quirk can depend on firmware version,
3105 		 * so, in principle, the set of quirks present can change
3106 		 * across a reset.  As a possible future enhancement, we
3107 		 * could re-scan for quirks every time we reinitialize
3108 		 * the device, but we'd have to make sure that the driver
3109 		 * behaves intelligently if the quirks change.
3110 		 */
3111 		for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3112 			if (quirk_matches(id, &core_quirks[i]))
3113 				ctrl->quirks |= core_quirks[i].quirks;
3114 		}
3115 	}
3116 	memcpy(ctrl->subsys->firmware_rev, id->fr,
3117 	       sizeof(ctrl->subsys->firmware_rev));
3118 
3119 	if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3120 		dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3121 		ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3122 	}
3123 
3124 	ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3125 	ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3126 	ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3127 
3128 	ctrl->oacs = le16_to_cpu(id->oacs);
3129 	ctrl->oncs = le16_to_cpu(id->oncs);
3130 	ctrl->mtfa = le16_to_cpu(id->mtfa);
3131 	ctrl->oaes = le32_to_cpu(id->oaes);
3132 	ctrl->wctemp = le16_to_cpu(id->wctemp);
3133 	ctrl->cctemp = le16_to_cpu(id->cctemp);
3134 
3135 	atomic_set(&ctrl->abort_limit, id->acl + 1);
3136 	ctrl->vwc = id->vwc;
3137 	if (id->mdts)
3138 		max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3139 	else
3140 		max_hw_sectors = UINT_MAX;
3141 	ctrl->max_hw_sectors =
3142 		min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3143 
3144 	nvme_set_queue_limits(ctrl, ctrl->admin_q);
3145 	ctrl->sgls = le32_to_cpu(id->sgls);
3146 	ctrl->kas = le16_to_cpu(id->kas);
3147 	ctrl->max_namespaces = le32_to_cpu(id->mnan);
3148 	ctrl->ctratt = le32_to_cpu(id->ctratt);
3149 
3150 	ctrl->cntrltype = id->cntrltype;
3151 	ctrl->dctype = id->dctype;
3152 
3153 	if (id->rtd3e) {
3154 		/* us -> s */
3155 		u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3156 
3157 		ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3158 						 shutdown_timeout, 60);
3159 
3160 		if (ctrl->shutdown_timeout != shutdown_timeout)
3161 			dev_info(ctrl->device,
3162 				 "Shutdown timeout set to %u seconds\n",
3163 				 ctrl->shutdown_timeout);
3164 	} else
3165 		ctrl->shutdown_timeout = shutdown_timeout;
3166 
3167 	ctrl->npss = id->npss;
3168 	ctrl->apsta = id->apsta;
3169 	prev_apst_enabled = ctrl->apst_enabled;
3170 	if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3171 		if (force_apst && id->apsta) {
3172 			dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3173 			ctrl->apst_enabled = true;
3174 		} else {
3175 			ctrl->apst_enabled = false;
3176 		}
3177 	} else {
3178 		ctrl->apst_enabled = id->apsta;
3179 	}
3180 	memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3181 
3182 	if (ctrl->ops->flags & NVME_F_FABRICS) {
3183 		ctrl->icdoff = le16_to_cpu(id->icdoff);
3184 		ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3185 		ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3186 		ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3187 
3188 		/*
3189 		 * In fabrics we need to verify the cntlid matches the
3190 		 * admin connect
3191 		 */
3192 		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3193 			dev_err(ctrl->device,
3194 				"Mismatching cntlid: Connect %u vs Identify "
3195 				"%u, rejecting\n",
3196 				ctrl->cntlid, le16_to_cpu(id->cntlid));
3197 			ret = -EINVAL;
3198 			goto out_free;
3199 		}
3200 
3201 		if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3202 			dev_err(ctrl->device,
3203 				"keep-alive support is mandatory for fabrics\n");
3204 			ret = -EINVAL;
3205 			goto out_free;
3206 		}
3207 	} else {
3208 		ctrl->hmpre = le32_to_cpu(id->hmpre);
3209 		ctrl->hmmin = le32_to_cpu(id->hmmin);
3210 		ctrl->hmminds = le32_to_cpu(id->hmminds);
3211 		ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3212 	}
3213 
3214 	ret = nvme_mpath_init_identify(ctrl, id);
3215 	if (ret < 0)
3216 		goto out_free;
3217 
3218 	if (ctrl->apst_enabled && !prev_apst_enabled)
3219 		dev_pm_qos_expose_latency_tolerance(ctrl->device);
3220 	else if (!ctrl->apst_enabled && prev_apst_enabled)
3221 		dev_pm_qos_hide_latency_tolerance(ctrl->device);
3222 
3223 out_free:
3224 	kfree(id);
3225 	return ret;
3226 }
3227 
3228 /*
3229  * Initialize the cached copies of the Identify data and various controller
3230  * register in our nvme_ctrl structure.  This should be called as soon as
3231  * the admin queue is fully up and running.
3232  */
3233 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3234 {
3235 	int ret;
3236 
3237 	ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3238 	if (ret) {
3239 		dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3240 		return ret;
3241 	}
3242 
3243 	ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3244 
3245 	if (ctrl->vs >= NVME_VS(1, 1, 0))
3246 		ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3247 
3248 	ret = nvme_init_identify(ctrl);
3249 	if (ret)
3250 		return ret;
3251 
3252 	ret = nvme_configure_apst(ctrl);
3253 	if (ret < 0)
3254 		return ret;
3255 
3256 	ret = nvme_configure_timestamp(ctrl);
3257 	if (ret < 0)
3258 		return ret;
3259 
3260 	ret = nvme_configure_host_options(ctrl);
3261 	if (ret < 0)
3262 		return ret;
3263 
3264 	if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3265 		ret = nvme_hwmon_init(ctrl);
3266 		if (ret < 0)
3267 			return ret;
3268 	}
3269 
3270 	ctrl->identified = true;
3271 
3272 	return 0;
3273 }
3274 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3275 
3276 static int nvme_dev_open(struct inode *inode, struct file *file)
3277 {
3278 	struct nvme_ctrl *ctrl =
3279 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3280 
3281 	switch (ctrl->state) {
3282 	case NVME_CTRL_LIVE:
3283 		break;
3284 	default:
3285 		return -EWOULDBLOCK;
3286 	}
3287 
3288 	nvme_get_ctrl(ctrl);
3289 	if (!try_module_get(ctrl->ops->module)) {
3290 		nvme_put_ctrl(ctrl);
3291 		return -EINVAL;
3292 	}
3293 
3294 	file->private_data = ctrl;
3295 	return 0;
3296 }
3297 
3298 static int nvme_dev_release(struct inode *inode, struct file *file)
3299 {
3300 	struct nvme_ctrl *ctrl =
3301 		container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3302 
3303 	module_put(ctrl->ops->module);
3304 	nvme_put_ctrl(ctrl);
3305 	return 0;
3306 }
3307 
3308 static const struct file_operations nvme_dev_fops = {
3309 	.owner		= THIS_MODULE,
3310 	.open		= nvme_dev_open,
3311 	.release	= nvme_dev_release,
3312 	.unlocked_ioctl	= nvme_dev_ioctl,
3313 	.compat_ioctl	= compat_ptr_ioctl,
3314 	.uring_cmd	= nvme_dev_uring_cmd,
3315 };
3316 
3317 static ssize_t nvme_sysfs_reset(struct device *dev,
3318 				struct device_attribute *attr, const char *buf,
3319 				size_t count)
3320 {
3321 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3322 	int ret;
3323 
3324 	ret = nvme_reset_ctrl_sync(ctrl);
3325 	if (ret < 0)
3326 		return ret;
3327 	return count;
3328 }
3329 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3330 
3331 static ssize_t nvme_sysfs_rescan(struct device *dev,
3332 				struct device_attribute *attr, const char *buf,
3333 				size_t count)
3334 {
3335 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3336 
3337 	nvme_queue_scan(ctrl);
3338 	return count;
3339 }
3340 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3341 
3342 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3343 {
3344 	struct gendisk *disk = dev_to_disk(dev);
3345 
3346 	if (disk->fops == &nvme_bdev_ops)
3347 		return nvme_get_ns_from_dev(dev)->head;
3348 	else
3349 		return disk->private_data;
3350 }
3351 
3352 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3353 		char *buf)
3354 {
3355 	struct nvme_ns_head *head = dev_to_ns_head(dev);
3356 	struct nvme_ns_ids *ids = &head->ids;
3357 	struct nvme_subsystem *subsys = head->subsys;
3358 	int serial_len = sizeof(subsys->serial);
3359 	int model_len = sizeof(subsys->model);
3360 
3361 	if (!uuid_is_null(&ids->uuid))
3362 		return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3363 
3364 	if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3365 		return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3366 
3367 	if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3368 		return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3369 
3370 	while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3371 				  subsys->serial[serial_len - 1] == '\0'))
3372 		serial_len--;
3373 	while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3374 				 subsys->model[model_len - 1] == '\0'))
3375 		model_len--;
3376 
3377 	return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3378 		serial_len, subsys->serial, model_len, subsys->model,
3379 		head->ns_id);
3380 }
3381 static DEVICE_ATTR_RO(wwid);
3382 
3383 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3384 		char *buf)
3385 {
3386 	return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3387 }
3388 static DEVICE_ATTR_RO(nguid);
3389 
3390 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3391 		char *buf)
3392 {
3393 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3394 
3395 	/* For backward compatibility expose the NGUID to userspace if
3396 	 * we have no UUID set
3397 	 */
3398 	if (uuid_is_null(&ids->uuid)) {
3399 		dev_warn_ratelimited(dev,
3400 			"No UUID available providing old NGUID\n");
3401 		return sysfs_emit(buf, "%pU\n", ids->nguid);
3402 	}
3403 	return sysfs_emit(buf, "%pU\n", &ids->uuid);
3404 }
3405 static DEVICE_ATTR_RO(uuid);
3406 
3407 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3408 		char *buf)
3409 {
3410 	return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3411 }
3412 static DEVICE_ATTR_RO(eui);
3413 
3414 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3415 		char *buf)
3416 {
3417 	return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3418 }
3419 static DEVICE_ATTR_RO(nsid);
3420 
3421 static struct attribute *nvme_ns_id_attrs[] = {
3422 	&dev_attr_wwid.attr,
3423 	&dev_attr_uuid.attr,
3424 	&dev_attr_nguid.attr,
3425 	&dev_attr_eui.attr,
3426 	&dev_attr_nsid.attr,
3427 #ifdef CONFIG_NVME_MULTIPATH
3428 	&dev_attr_ana_grpid.attr,
3429 	&dev_attr_ana_state.attr,
3430 #endif
3431 	NULL,
3432 };
3433 
3434 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3435 		struct attribute *a, int n)
3436 {
3437 	struct device *dev = container_of(kobj, struct device, kobj);
3438 	struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3439 
3440 	if (a == &dev_attr_uuid.attr) {
3441 		if (uuid_is_null(&ids->uuid) &&
3442 		    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3443 			return 0;
3444 	}
3445 	if (a == &dev_attr_nguid.attr) {
3446 		if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3447 			return 0;
3448 	}
3449 	if (a == &dev_attr_eui.attr) {
3450 		if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3451 			return 0;
3452 	}
3453 #ifdef CONFIG_NVME_MULTIPATH
3454 	if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3455 		if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3456 			return 0;
3457 		if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3458 			return 0;
3459 	}
3460 #endif
3461 	return a->mode;
3462 }
3463 
3464 static const struct attribute_group nvme_ns_id_attr_group = {
3465 	.attrs		= nvme_ns_id_attrs,
3466 	.is_visible	= nvme_ns_id_attrs_are_visible,
3467 };
3468 
3469 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3470 	&nvme_ns_id_attr_group,
3471 	NULL,
3472 };
3473 
3474 #define nvme_show_str_function(field)						\
3475 static ssize_t  field##_show(struct device *dev,				\
3476 			    struct device_attribute *attr, char *buf)		\
3477 {										\
3478         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3479         return sysfs_emit(buf, "%.*s\n",					\
3480 		(int)sizeof(ctrl->subsys->field), ctrl->subsys->field);		\
3481 }										\
3482 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3483 
3484 nvme_show_str_function(model);
3485 nvme_show_str_function(serial);
3486 nvme_show_str_function(firmware_rev);
3487 
3488 #define nvme_show_int_function(field)						\
3489 static ssize_t  field##_show(struct device *dev,				\
3490 			    struct device_attribute *attr, char *buf)		\
3491 {										\
3492         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);				\
3493         return sysfs_emit(buf, "%d\n", ctrl->field);				\
3494 }										\
3495 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3496 
3497 nvme_show_int_function(cntlid);
3498 nvme_show_int_function(numa_node);
3499 nvme_show_int_function(queue_count);
3500 nvme_show_int_function(sqsize);
3501 nvme_show_int_function(kato);
3502 
3503 static ssize_t nvme_sysfs_delete(struct device *dev,
3504 				struct device_attribute *attr, const char *buf,
3505 				size_t count)
3506 {
3507 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3508 
3509 	if (device_remove_file_self(dev, attr))
3510 		nvme_delete_ctrl_sync(ctrl);
3511 	return count;
3512 }
3513 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3514 
3515 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3516 					 struct device_attribute *attr,
3517 					 char *buf)
3518 {
3519 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3520 
3521 	return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3522 }
3523 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3524 
3525 static ssize_t nvme_sysfs_show_state(struct device *dev,
3526 				     struct device_attribute *attr,
3527 				     char *buf)
3528 {
3529 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3530 	static const char *const state_name[] = {
3531 		[NVME_CTRL_NEW]		= "new",
3532 		[NVME_CTRL_LIVE]	= "live",
3533 		[NVME_CTRL_RESETTING]	= "resetting",
3534 		[NVME_CTRL_CONNECTING]	= "connecting",
3535 		[NVME_CTRL_DELETING]	= "deleting",
3536 		[NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3537 		[NVME_CTRL_DEAD]	= "dead",
3538 	};
3539 
3540 	if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3541 	    state_name[ctrl->state])
3542 		return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3543 
3544 	return sysfs_emit(buf, "unknown state\n");
3545 }
3546 
3547 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3548 
3549 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3550 					 struct device_attribute *attr,
3551 					 char *buf)
3552 {
3553 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3554 
3555 	return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3556 }
3557 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3558 
3559 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3560 					struct device_attribute *attr,
3561 					char *buf)
3562 {
3563 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3564 
3565 	return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3566 }
3567 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3568 
3569 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3570 					struct device_attribute *attr,
3571 					char *buf)
3572 {
3573 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3574 
3575 	return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3576 }
3577 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3578 
3579 static ssize_t nvme_sysfs_show_address(struct device *dev,
3580 					 struct device_attribute *attr,
3581 					 char *buf)
3582 {
3583 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3584 
3585 	return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3586 }
3587 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3588 
3589 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3590 		struct device_attribute *attr, char *buf)
3591 {
3592 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3593 	struct nvmf_ctrl_options *opts = ctrl->opts;
3594 
3595 	if (ctrl->opts->max_reconnects == -1)
3596 		return sysfs_emit(buf, "off\n");
3597 	return sysfs_emit(buf, "%d\n",
3598 			  opts->max_reconnects * opts->reconnect_delay);
3599 }
3600 
3601 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3602 		struct device_attribute *attr, const char *buf, size_t count)
3603 {
3604 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3605 	struct nvmf_ctrl_options *opts = ctrl->opts;
3606 	int ctrl_loss_tmo, err;
3607 
3608 	err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3609 	if (err)
3610 		return -EINVAL;
3611 
3612 	if (ctrl_loss_tmo < 0)
3613 		opts->max_reconnects = -1;
3614 	else
3615 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3616 						opts->reconnect_delay);
3617 	return count;
3618 }
3619 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3620 	nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3621 
3622 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3623 		struct device_attribute *attr, char *buf)
3624 {
3625 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3626 
3627 	if (ctrl->opts->reconnect_delay == -1)
3628 		return sysfs_emit(buf, "off\n");
3629 	return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3630 }
3631 
3632 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3633 		struct device_attribute *attr, const char *buf, size_t count)
3634 {
3635 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3636 	unsigned int v;
3637 	int err;
3638 
3639 	err = kstrtou32(buf, 10, &v);
3640 	if (err)
3641 		return err;
3642 
3643 	ctrl->opts->reconnect_delay = v;
3644 	return count;
3645 }
3646 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3647 	nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3648 
3649 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3650 		struct device_attribute *attr, char *buf)
3651 {
3652 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3653 
3654 	if (ctrl->opts->fast_io_fail_tmo == -1)
3655 		return sysfs_emit(buf, "off\n");
3656 	return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3657 }
3658 
3659 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3660 		struct device_attribute *attr, const char *buf, size_t count)
3661 {
3662 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3663 	struct nvmf_ctrl_options *opts = ctrl->opts;
3664 	int fast_io_fail_tmo, err;
3665 
3666 	err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3667 	if (err)
3668 		return -EINVAL;
3669 
3670 	if (fast_io_fail_tmo < 0)
3671 		opts->fast_io_fail_tmo = -1;
3672 	else
3673 		opts->fast_io_fail_tmo = fast_io_fail_tmo;
3674 	return count;
3675 }
3676 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3677 	nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3678 
3679 static ssize_t cntrltype_show(struct device *dev,
3680 			      struct device_attribute *attr, char *buf)
3681 {
3682 	static const char * const type[] = {
3683 		[NVME_CTRL_IO] = "io\n",
3684 		[NVME_CTRL_DISC] = "discovery\n",
3685 		[NVME_CTRL_ADMIN] = "admin\n",
3686 	};
3687 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3688 
3689 	if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3690 		return sysfs_emit(buf, "reserved\n");
3691 
3692 	return sysfs_emit(buf, type[ctrl->cntrltype]);
3693 }
3694 static DEVICE_ATTR_RO(cntrltype);
3695 
3696 static ssize_t dctype_show(struct device *dev,
3697 			   struct device_attribute *attr, char *buf)
3698 {
3699 	static const char * const type[] = {
3700 		[NVME_DCTYPE_NOT_REPORTED] = "none\n",
3701 		[NVME_DCTYPE_DDC] = "ddc\n",
3702 		[NVME_DCTYPE_CDC] = "cdc\n",
3703 	};
3704 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3705 
3706 	if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3707 		return sysfs_emit(buf, "reserved\n");
3708 
3709 	return sysfs_emit(buf, type[ctrl->dctype]);
3710 }
3711 static DEVICE_ATTR_RO(dctype);
3712 
3713 #ifdef CONFIG_NVME_AUTH
3714 static ssize_t nvme_ctrl_dhchap_secret_show(struct device *dev,
3715 		struct device_attribute *attr, char *buf)
3716 {
3717 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3718 	struct nvmf_ctrl_options *opts = ctrl->opts;
3719 
3720 	if (!opts->dhchap_secret)
3721 		return sysfs_emit(buf, "none\n");
3722 	return sysfs_emit(buf, "%s\n", opts->dhchap_secret);
3723 }
3724 
3725 static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
3726 		struct device_attribute *attr, const char *buf, size_t count)
3727 {
3728 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3729 	struct nvmf_ctrl_options *opts = ctrl->opts;
3730 	char *dhchap_secret;
3731 
3732 	if (!ctrl->opts->dhchap_secret)
3733 		return -EINVAL;
3734 	if (count < 7)
3735 		return -EINVAL;
3736 	if (memcmp(buf, "DHHC-1:", 7))
3737 		return -EINVAL;
3738 
3739 	dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3740 	if (!dhchap_secret)
3741 		return -ENOMEM;
3742 	memcpy(dhchap_secret, buf, count);
3743 	nvme_auth_stop(ctrl);
3744 	if (strcmp(dhchap_secret, opts->dhchap_secret)) {
3745 		int ret;
3746 
3747 		ret = nvme_auth_generate_key(dhchap_secret, &ctrl->host_key);
3748 		if (ret)
3749 			return ret;
3750 		kfree(opts->dhchap_secret);
3751 		opts->dhchap_secret = dhchap_secret;
3752 		/* Key has changed; re-authentication with new key */
3753 		nvme_auth_reset(ctrl);
3754 	}
3755 	/* Start re-authentication */
3756 	dev_info(ctrl->device, "re-authenticating controller\n");
3757 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3758 
3759 	return count;
3760 }
3761 static DEVICE_ATTR(dhchap_secret, S_IRUGO | S_IWUSR,
3762 	nvme_ctrl_dhchap_secret_show, nvme_ctrl_dhchap_secret_store);
3763 
3764 static ssize_t nvme_ctrl_dhchap_ctrl_secret_show(struct device *dev,
3765 		struct device_attribute *attr, char *buf)
3766 {
3767 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3768 	struct nvmf_ctrl_options *opts = ctrl->opts;
3769 
3770 	if (!opts->dhchap_ctrl_secret)
3771 		return sysfs_emit(buf, "none\n");
3772 	return sysfs_emit(buf, "%s\n", opts->dhchap_ctrl_secret);
3773 }
3774 
3775 static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
3776 		struct device_attribute *attr, const char *buf, size_t count)
3777 {
3778 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3779 	struct nvmf_ctrl_options *opts = ctrl->opts;
3780 	char *dhchap_secret;
3781 
3782 	if (!ctrl->opts->dhchap_ctrl_secret)
3783 		return -EINVAL;
3784 	if (count < 7)
3785 		return -EINVAL;
3786 	if (memcmp(buf, "DHHC-1:", 7))
3787 		return -EINVAL;
3788 
3789 	dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3790 	if (!dhchap_secret)
3791 		return -ENOMEM;
3792 	memcpy(dhchap_secret, buf, count);
3793 	nvme_auth_stop(ctrl);
3794 	if (strcmp(dhchap_secret, opts->dhchap_ctrl_secret)) {
3795 		int ret;
3796 
3797 		ret = nvme_auth_generate_key(dhchap_secret, &ctrl->ctrl_key);
3798 		if (ret)
3799 			return ret;
3800 		kfree(opts->dhchap_ctrl_secret);
3801 		opts->dhchap_ctrl_secret = dhchap_secret;
3802 		/* Key has changed; re-authentication with new key */
3803 		nvme_auth_reset(ctrl);
3804 	}
3805 	/* Start re-authentication */
3806 	dev_info(ctrl->device, "re-authenticating controller\n");
3807 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3808 
3809 	return count;
3810 }
3811 static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR,
3812 	nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);
3813 #endif
3814 
3815 static struct attribute *nvme_dev_attrs[] = {
3816 	&dev_attr_reset_controller.attr,
3817 	&dev_attr_rescan_controller.attr,
3818 	&dev_attr_model.attr,
3819 	&dev_attr_serial.attr,
3820 	&dev_attr_firmware_rev.attr,
3821 	&dev_attr_cntlid.attr,
3822 	&dev_attr_delete_controller.attr,
3823 	&dev_attr_transport.attr,
3824 	&dev_attr_subsysnqn.attr,
3825 	&dev_attr_address.attr,
3826 	&dev_attr_state.attr,
3827 	&dev_attr_numa_node.attr,
3828 	&dev_attr_queue_count.attr,
3829 	&dev_attr_sqsize.attr,
3830 	&dev_attr_hostnqn.attr,
3831 	&dev_attr_hostid.attr,
3832 	&dev_attr_ctrl_loss_tmo.attr,
3833 	&dev_attr_reconnect_delay.attr,
3834 	&dev_attr_fast_io_fail_tmo.attr,
3835 	&dev_attr_kato.attr,
3836 	&dev_attr_cntrltype.attr,
3837 	&dev_attr_dctype.attr,
3838 #ifdef CONFIG_NVME_AUTH
3839 	&dev_attr_dhchap_secret.attr,
3840 	&dev_attr_dhchap_ctrl_secret.attr,
3841 #endif
3842 	NULL
3843 };
3844 
3845 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3846 		struct attribute *a, int n)
3847 {
3848 	struct device *dev = container_of(kobj, struct device, kobj);
3849 	struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3850 
3851 	if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3852 		return 0;
3853 	if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3854 		return 0;
3855 	if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3856 		return 0;
3857 	if (a == &dev_attr_hostid.attr && !ctrl->opts)
3858 		return 0;
3859 	if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3860 		return 0;
3861 	if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3862 		return 0;
3863 	if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3864 		return 0;
3865 #ifdef CONFIG_NVME_AUTH
3866 	if (a == &dev_attr_dhchap_secret.attr && !ctrl->opts)
3867 		return 0;
3868 	if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)
3869 		return 0;
3870 #endif
3871 
3872 	return a->mode;
3873 }
3874 
3875 static const struct attribute_group nvme_dev_attrs_group = {
3876 	.attrs		= nvme_dev_attrs,
3877 	.is_visible	= nvme_dev_attrs_are_visible,
3878 };
3879 
3880 static const struct attribute_group *nvme_dev_attr_groups[] = {
3881 	&nvme_dev_attrs_group,
3882 	NULL,
3883 };
3884 
3885 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3886 		unsigned nsid)
3887 {
3888 	struct nvme_ns_head *h;
3889 
3890 	lockdep_assert_held(&ctrl->subsys->lock);
3891 
3892 	list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3893 		/*
3894 		 * Private namespaces can share NSIDs under some conditions.
3895 		 * In that case we can't use the same ns_head for namespaces
3896 		 * with the same NSID.
3897 		 */
3898 		if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3899 			continue;
3900 		if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3901 			return h;
3902 	}
3903 
3904 	return NULL;
3905 }
3906 
3907 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3908 		struct nvme_ns_ids *ids)
3909 {
3910 	bool has_uuid = !uuid_is_null(&ids->uuid);
3911 	bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3912 	bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3913 	struct nvme_ns_head *h;
3914 
3915 	lockdep_assert_held(&subsys->lock);
3916 
3917 	list_for_each_entry(h, &subsys->nsheads, entry) {
3918 		if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3919 			return -EINVAL;
3920 		if (has_nguid &&
3921 		    memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3922 			return -EINVAL;
3923 		if (has_eui64 &&
3924 		    memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3925 			return -EINVAL;
3926 	}
3927 
3928 	return 0;
3929 }
3930 
3931 static void nvme_cdev_rel(struct device *dev)
3932 {
3933 	ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3934 }
3935 
3936 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3937 {
3938 	cdev_device_del(cdev, cdev_device);
3939 	put_device(cdev_device);
3940 }
3941 
3942 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3943 		const struct file_operations *fops, struct module *owner)
3944 {
3945 	int minor, ret;
3946 
3947 	minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3948 	if (minor < 0)
3949 		return minor;
3950 	cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3951 	cdev_device->class = nvme_ns_chr_class;
3952 	cdev_device->release = nvme_cdev_rel;
3953 	device_initialize(cdev_device);
3954 	cdev_init(cdev, fops);
3955 	cdev->owner = owner;
3956 	ret = cdev_device_add(cdev, cdev_device);
3957 	if (ret)
3958 		put_device(cdev_device);
3959 
3960 	return ret;
3961 }
3962 
3963 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3964 {
3965 	return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3966 }
3967 
3968 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3969 {
3970 	nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3971 	return 0;
3972 }
3973 
3974 static const struct file_operations nvme_ns_chr_fops = {
3975 	.owner		= THIS_MODULE,
3976 	.open		= nvme_ns_chr_open,
3977 	.release	= nvme_ns_chr_release,
3978 	.unlocked_ioctl	= nvme_ns_chr_ioctl,
3979 	.compat_ioctl	= compat_ptr_ioctl,
3980 	.uring_cmd	= nvme_ns_chr_uring_cmd,
3981 	.uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3982 };
3983 
3984 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3985 {
3986 	int ret;
3987 
3988 	ns->cdev_device.parent = ns->ctrl->device;
3989 	ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3990 			   ns->ctrl->instance, ns->head->instance);
3991 	if (ret)
3992 		return ret;
3993 
3994 	return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3995 			     ns->ctrl->ops->module);
3996 }
3997 
3998 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3999 		struct nvme_ns_info *info)
4000 {
4001 	struct nvme_ns_head *head;
4002 	size_t size = sizeof(*head);
4003 	int ret = -ENOMEM;
4004 
4005 #ifdef CONFIG_NVME_MULTIPATH
4006 	size += num_possible_nodes() * sizeof(struct nvme_ns *);
4007 #endif
4008 
4009 	head = kzalloc(size, GFP_KERNEL);
4010 	if (!head)
4011 		goto out;
4012 	ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
4013 	if (ret < 0)
4014 		goto out_free_head;
4015 	head->instance = ret;
4016 	INIT_LIST_HEAD(&head->list);
4017 	ret = init_srcu_struct(&head->srcu);
4018 	if (ret)
4019 		goto out_ida_remove;
4020 	head->subsys = ctrl->subsys;
4021 	head->ns_id = info->nsid;
4022 	head->ids = info->ids;
4023 	head->shared = info->is_shared;
4024 	kref_init(&head->ref);
4025 
4026 	if (head->ids.csi) {
4027 		ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
4028 		if (ret)
4029 			goto out_cleanup_srcu;
4030 	} else
4031 		head->effects = ctrl->effects;
4032 
4033 	ret = nvme_mpath_alloc_disk(ctrl, head);
4034 	if (ret)
4035 		goto out_cleanup_srcu;
4036 
4037 	list_add_tail(&head->entry, &ctrl->subsys->nsheads);
4038 
4039 	kref_get(&ctrl->subsys->ref);
4040 
4041 	return head;
4042 out_cleanup_srcu:
4043 	cleanup_srcu_struct(&head->srcu);
4044 out_ida_remove:
4045 	ida_free(&ctrl->subsys->ns_ida, head->instance);
4046 out_free_head:
4047 	kfree(head);
4048 out:
4049 	if (ret > 0)
4050 		ret = blk_status_to_errno(nvme_error_status(ret));
4051 	return ERR_PTR(ret);
4052 }
4053 
4054 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
4055 		struct nvme_ns_ids *ids)
4056 {
4057 	struct nvme_subsystem *s;
4058 	int ret = 0;
4059 
4060 	/*
4061 	 * Note that this check is racy as we try to avoid holding the global
4062 	 * lock over the whole ns_head creation.  But it is only intended as
4063 	 * a sanity check anyway.
4064 	 */
4065 	mutex_lock(&nvme_subsystems_lock);
4066 	list_for_each_entry(s, &nvme_subsystems, entry) {
4067 		if (s == this)
4068 			continue;
4069 		mutex_lock(&s->lock);
4070 		ret = nvme_subsys_check_duplicate_ids(s, ids);
4071 		mutex_unlock(&s->lock);
4072 		if (ret)
4073 			break;
4074 	}
4075 	mutex_unlock(&nvme_subsystems_lock);
4076 
4077 	return ret;
4078 }
4079 
4080 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
4081 {
4082 	struct nvme_ctrl *ctrl = ns->ctrl;
4083 	struct nvme_ns_head *head = NULL;
4084 	int ret;
4085 
4086 	ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
4087 	if (ret) {
4088 		dev_err(ctrl->device,
4089 			"globally duplicate IDs for nsid %d\n", info->nsid);
4090 		nvme_print_device_info(ctrl);
4091 		return ret;
4092 	}
4093 
4094 	mutex_lock(&ctrl->subsys->lock);
4095 	head = nvme_find_ns_head(ctrl, info->nsid);
4096 	if (!head) {
4097 		ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4098 		if (ret) {
4099 			dev_err(ctrl->device,
4100 				"duplicate IDs in subsystem for nsid %d\n",
4101 				info->nsid);
4102 			goto out_unlock;
4103 		}
4104 		head = nvme_alloc_ns_head(ctrl, info);
4105 		if (IS_ERR(head)) {
4106 			ret = PTR_ERR(head);
4107 			goto out_unlock;
4108 		}
4109 	} else {
4110 		ret = -EINVAL;
4111 		if (!info->is_shared || !head->shared) {
4112 			dev_err(ctrl->device,
4113 				"Duplicate unshared namespace %d\n",
4114 				info->nsid);
4115 			goto out_put_ns_head;
4116 		}
4117 		if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4118 			dev_err(ctrl->device,
4119 				"IDs don't match for shared namespace %d\n",
4120 					info->nsid);
4121 			goto out_put_ns_head;
4122 		}
4123 
4124 		if (!multipath && !list_empty(&head->list)) {
4125 			dev_warn(ctrl->device,
4126 				"Found shared namespace %d, but multipathing not supported.\n",
4127 				info->nsid);
4128 			dev_warn_once(ctrl->device,
4129 				"Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
4130 		}
4131 	}
4132 
4133 	list_add_tail_rcu(&ns->siblings, &head->list);
4134 	ns->head = head;
4135 	mutex_unlock(&ctrl->subsys->lock);
4136 	return 0;
4137 
4138 out_put_ns_head:
4139 	nvme_put_ns_head(head);
4140 out_unlock:
4141 	mutex_unlock(&ctrl->subsys->lock);
4142 	return ret;
4143 }
4144 
4145 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4146 {
4147 	struct nvme_ns *ns, *ret = NULL;
4148 
4149 	down_read(&ctrl->namespaces_rwsem);
4150 	list_for_each_entry(ns, &ctrl->namespaces, list) {
4151 		if (ns->head->ns_id == nsid) {
4152 			if (!nvme_get_ns(ns))
4153 				continue;
4154 			ret = ns;
4155 			break;
4156 		}
4157 		if (ns->head->ns_id > nsid)
4158 			break;
4159 	}
4160 	up_read(&ctrl->namespaces_rwsem);
4161 	return ret;
4162 }
4163 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
4164 
4165 /*
4166  * Add the namespace to the controller list while keeping the list ordered.
4167  */
4168 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4169 {
4170 	struct nvme_ns *tmp;
4171 
4172 	list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4173 		if (tmp->head->ns_id < ns->head->ns_id) {
4174 			list_add(&ns->list, &tmp->list);
4175 			return;
4176 		}
4177 	}
4178 	list_add(&ns->list, &ns->ctrl->namespaces);
4179 }
4180 
4181 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4182 {
4183 	struct nvme_ns *ns;
4184 	struct gendisk *disk;
4185 	int node = ctrl->numa_node;
4186 
4187 	ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4188 	if (!ns)
4189 		return;
4190 
4191 	disk = blk_mq_alloc_disk(ctrl->tagset, ns);
4192 	if (IS_ERR(disk))
4193 		goto out_free_ns;
4194 	disk->fops = &nvme_bdev_ops;
4195 	disk->private_data = ns;
4196 
4197 	ns->disk = disk;
4198 	ns->queue = disk->queue;
4199 
4200 	if (ctrl->opts && ctrl->opts->data_digest)
4201 		blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4202 
4203 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4204 	if (ctrl->ops->supports_pci_p2pdma &&
4205 	    ctrl->ops->supports_pci_p2pdma(ctrl))
4206 		blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4207 
4208 	ns->ctrl = ctrl;
4209 	kref_init(&ns->kref);
4210 
4211 	if (nvme_init_ns_head(ns, info))
4212 		goto out_cleanup_disk;
4213 
4214 	/*
4215 	 * If multipathing is enabled, the device name for all disks and not
4216 	 * just those that represent shared namespaces needs to be based on the
4217 	 * subsystem instance.  Using the controller instance for private
4218 	 * namespaces could lead to naming collisions between shared and private
4219 	 * namespaces if they don't use a common numbering scheme.
4220 	 *
4221 	 * If multipathing is not enabled, disk names must use the controller
4222 	 * instance as shared namespaces will show up as multiple block
4223 	 * devices.
4224 	 */
4225 	if (ns->head->disk) {
4226 		sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4227 			ctrl->instance, ns->head->instance);
4228 		disk->flags |= GENHD_FL_HIDDEN;
4229 	} else if (multipath) {
4230 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4231 			ns->head->instance);
4232 	} else {
4233 		sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4234 			ns->head->instance);
4235 	}
4236 
4237 	if (nvme_update_ns_info(ns, info))
4238 		goto out_unlink_ns;
4239 
4240 	down_write(&ctrl->namespaces_rwsem);
4241 	nvme_ns_add_to_ctrl_list(ns);
4242 	up_write(&ctrl->namespaces_rwsem);
4243 	nvme_get_ctrl(ctrl);
4244 
4245 	if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4246 		goto out_cleanup_ns_from_list;
4247 
4248 	if (!nvme_ns_head_multipath(ns->head))
4249 		nvme_add_ns_cdev(ns);
4250 
4251 	nvme_mpath_add_disk(ns, info->anagrpid);
4252 	nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4253 
4254 	return;
4255 
4256  out_cleanup_ns_from_list:
4257 	nvme_put_ctrl(ctrl);
4258 	down_write(&ctrl->namespaces_rwsem);
4259 	list_del_init(&ns->list);
4260 	up_write(&ctrl->namespaces_rwsem);
4261  out_unlink_ns:
4262 	mutex_lock(&ctrl->subsys->lock);
4263 	list_del_rcu(&ns->siblings);
4264 	if (list_empty(&ns->head->list))
4265 		list_del_init(&ns->head->entry);
4266 	mutex_unlock(&ctrl->subsys->lock);
4267 	nvme_put_ns_head(ns->head);
4268  out_cleanup_disk:
4269 	put_disk(disk);
4270  out_free_ns:
4271 	kfree(ns);
4272 }
4273 
4274 static void nvme_ns_remove(struct nvme_ns *ns)
4275 {
4276 	bool last_path = false;
4277 
4278 	if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4279 		return;
4280 
4281 	clear_bit(NVME_NS_READY, &ns->flags);
4282 	set_capacity(ns->disk, 0);
4283 	nvme_fault_inject_fini(&ns->fault_inject);
4284 
4285 	/*
4286 	 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4287 	 * this ns going back into current_path.
4288 	 */
4289 	synchronize_srcu(&ns->head->srcu);
4290 
4291 	/* wait for concurrent submissions */
4292 	if (nvme_mpath_clear_current_path(ns))
4293 		synchronize_srcu(&ns->head->srcu);
4294 
4295 	mutex_lock(&ns->ctrl->subsys->lock);
4296 	list_del_rcu(&ns->siblings);
4297 	if (list_empty(&ns->head->list)) {
4298 		list_del_init(&ns->head->entry);
4299 		last_path = true;
4300 	}
4301 	mutex_unlock(&ns->ctrl->subsys->lock);
4302 
4303 	/* guarantee not available in head->list */
4304 	synchronize_rcu();
4305 
4306 	if (!nvme_ns_head_multipath(ns->head))
4307 		nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4308 	del_gendisk(ns->disk);
4309 
4310 	down_write(&ns->ctrl->namespaces_rwsem);
4311 	list_del_init(&ns->list);
4312 	up_write(&ns->ctrl->namespaces_rwsem);
4313 
4314 	if (last_path)
4315 		nvme_mpath_shutdown_disk(ns->head);
4316 	nvme_put_ns(ns);
4317 }
4318 
4319 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4320 {
4321 	struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4322 
4323 	if (ns) {
4324 		nvme_ns_remove(ns);
4325 		nvme_put_ns(ns);
4326 	}
4327 }
4328 
4329 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4330 {
4331 	int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4332 
4333 	if (test_bit(NVME_NS_DEAD, &ns->flags))
4334 		goto out;
4335 
4336 	ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4337 	if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4338 		dev_err(ns->ctrl->device,
4339 			"identifiers changed for nsid %d\n", ns->head->ns_id);
4340 		goto out;
4341 	}
4342 
4343 	ret = nvme_update_ns_info(ns, info);
4344 out:
4345 	/*
4346 	 * Only remove the namespace if we got a fatal error back from the
4347 	 * device, otherwise ignore the error and just move on.
4348 	 *
4349 	 * TODO: we should probably schedule a delayed retry here.
4350 	 */
4351 	if (ret > 0 && (ret & NVME_SC_DNR))
4352 		nvme_ns_remove(ns);
4353 }
4354 
4355 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4356 {
4357 	struct nvme_ns_info info = { .nsid = nsid };
4358 	struct nvme_ns *ns;
4359 
4360 	if (nvme_identify_ns_descs(ctrl, &info))
4361 		return;
4362 
4363 	if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4364 		dev_warn(ctrl->device,
4365 			"command set not reported for nsid: %d\n", nsid);
4366 		return;
4367 	}
4368 
4369 	/*
4370 	 * If available try to use the Command Set Idependent Identify Namespace
4371 	 * data structure to find all the generic information that is needed to
4372 	 * set up a namespace.  If not fall back to the legacy version.
4373 	 */
4374 	if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4375 	    (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS)) {
4376 		if (nvme_ns_info_from_id_cs_indep(ctrl, &info))
4377 			return;
4378 	} else {
4379 		if (nvme_ns_info_from_identify(ctrl, &info))
4380 			return;
4381 	}
4382 
4383 	/*
4384 	 * Ignore the namespace if it is not ready. We will get an AEN once it
4385 	 * becomes ready and restart the scan.
4386 	 */
4387 	if (!info.is_ready)
4388 		return;
4389 
4390 	ns = nvme_find_get_ns(ctrl, nsid);
4391 	if (ns) {
4392 		nvme_validate_ns(ns, &info);
4393 		nvme_put_ns(ns);
4394 	} else {
4395 		nvme_alloc_ns(ctrl, &info);
4396 	}
4397 }
4398 
4399 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4400 					unsigned nsid)
4401 {
4402 	struct nvme_ns *ns, *next;
4403 	LIST_HEAD(rm_list);
4404 
4405 	down_write(&ctrl->namespaces_rwsem);
4406 	list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4407 		if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4408 			list_move_tail(&ns->list, &rm_list);
4409 	}
4410 	up_write(&ctrl->namespaces_rwsem);
4411 
4412 	list_for_each_entry_safe(ns, next, &rm_list, list)
4413 		nvme_ns_remove(ns);
4414 
4415 }
4416 
4417 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4418 {
4419 	const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4420 	__le32 *ns_list;
4421 	u32 prev = 0;
4422 	int ret = 0, i;
4423 
4424 	if (nvme_ctrl_limited_cns(ctrl))
4425 		return -EOPNOTSUPP;
4426 
4427 	ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4428 	if (!ns_list)
4429 		return -ENOMEM;
4430 
4431 	for (;;) {
4432 		struct nvme_command cmd = {
4433 			.identify.opcode	= nvme_admin_identify,
4434 			.identify.cns		= NVME_ID_CNS_NS_ACTIVE_LIST,
4435 			.identify.nsid		= cpu_to_le32(prev),
4436 		};
4437 
4438 		ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4439 					    NVME_IDENTIFY_DATA_SIZE);
4440 		if (ret) {
4441 			dev_warn(ctrl->device,
4442 				"Identify NS List failed (status=0x%x)\n", ret);
4443 			goto free;
4444 		}
4445 
4446 		for (i = 0; i < nr_entries; i++) {
4447 			u32 nsid = le32_to_cpu(ns_list[i]);
4448 
4449 			if (!nsid)	/* end of the list? */
4450 				goto out;
4451 			nvme_scan_ns(ctrl, nsid);
4452 			while (++prev < nsid)
4453 				nvme_ns_remove_by_nsid(ctrl, prev);
4454 		}
4455 	}
4456  out:
4457 	nvme_remove_invalid_namespaces(ctrl, prev);
4458  free:
4459 	kfree(ns_list);
4460 	return ret;
4461 }
4462 
4463 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4464 {
4465 	struct nvme_id_ctrl *id;
4466 	u32 nn, i;
4467 
4468 	if (nvme_identify_ctrl(ctrl, &id))
4469 		return;
4470 	nn = le32_to_cpu(id->nn);
4471 	kfree(id);
4472 
4473 	for (i = 1; i <= nn; i++)
4474 		nvme_scan_ns(ctrl, i);
4475 
4476 	nvme_remove_invalid_namespaces(ctrl, nn);
4477 }
4478 
4479 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4480 {
4481 	size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4482 	__le32 *log;
4483 	int error;
4484 
4485 	log = kzalloc(log_size, GFP_KERNEL);
4486 	if (!log)
4487 		return;
4488 
4489 	/*
4490 	 * We need to read the log to clear the AEN, but we don't want to rely
4491 	 * on it for the changed namespace information as userspace could have
4492 	 * raced with us in reading the log page, which could cause us to miss
4493 	 * updates.
4494 	 */
4495 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4496 			NVME_CSI_NVM, log, log_size, 0);
4497 	if (error)
4498 		dev_warn(ctrl->device,
4499 			"reading changed ns log failed: %d\n", error);
4500 
4501 	kfree(log);
4502 }
4503 
4504 static void nvme_scan_work(struct work_struct *work)
4505 {
4506 	struct nvme_ctrl *ctrl =
4507 		container_of(work, struct nvme_ctrl, scan_work);
4508 	int ret;
4509 
4510 	/* No tagset on a live ctrl means IO queues could not created */
4511 	if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4512 		return;
4513 
4514 	/*
4515 	 * Identify controller limits can change at controller reset due to
4516 	 * new firmware download, even though it is not common we cannot ignore
4517 	 * such scenario. Controller's non-mdts limits are reported in the unit
4518 	 * of logical blocks that is dependent on the format of attached
4519 	 * namespace. Hence re-read the limits at the time of ns allocation.
4520 	 */
4521 	ret = nvme_init_non_mdts_limits(ctrl);
4522 	if (ret < 0) {
4523 		dev_warn(ctrl->device,
4524 			"reading non-mdts-limits failed: %d\n", ret);
4525 		return;
4526 	}
4527 
4528 	if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4529 		dev_info(ctrl->device, "rescanning namespaces.\n");
4530 		nvme_clear_changed_ns_log(ctrl);
4531 	}
4532 
4533 	mutex_lock(&ctrl->scan_lock);
4534 	if (nvme_scan_ns_list(ctrl) != 0)
4535 		nvme_scan_ns_sequential(ctrl);
4536 	mutex_unlock(&ctrl->scan_lock);
4537 }
4538 
4539 /*
4540  * This function iterates the namespace list unlocked to allow recovery from
4541  * controller failure. It is up to the caller to ensure the namespace list is
4542  * not modified by scan work while this function is executing.
4543  */
4544 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4545 {
4546 	struct nvme_ns *ns, *next;
4547 	LIST_HEAD(ns_list);
4548 
4549 	/*
4550 	 * make sure to requeue I/O to all namespaces as these
4551 	 * might result from the scan itself and must complete
4552 	 * for the scan_work to make progress
4553 	 */
4554 	nvme_mpath_clear_ctrl_paths(ctrl);
4555 
4556 	/* prevent racing with ns scanning */
4557 	flush_work(&ctrl->scan_work);
4558 
4559 	/*
4560 	 * The dead states indicates the controller was not gracefully
4561 	 * disconnected. In that case, we won't be able to flush any data while
4562 	 * removing the namespaces' disks; fail all the queues now to avoid
4563 	 * potentially having to clean up the failed sync later.
4564 	 */
4565 	if (ctrl->state == NVME_CTRL_DEAD)
4566 		nvme_kill_queues(ctrl);
4567 
4568 	/* this is a no-op when called from the controller reset handler */
4569 	nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4570 
4571 	down_write(&ctrl->namespaces_rwsem);
4572 	list_splice_init(&ctrl->namespaces, &ns_list);
4573 	up_write(&ctrl->namespaces_rwsem);
4574 
4575 	list_for_each_entry_safe(ns, next, &ns_list, list)
4576 		nvme_ns_remove(ns);
4577 }
4578 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4579 
4580 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4581 {
4582 	struct nvme_ctrl *ctrl =
4583 		container_of(dev, struct nvme_ctrl, ctrl_device);
4584 	struct nvmf_ctrl_options *opts = ctrl->opts;
4585 	int ret;
4586 
4587 	ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4588 	if (ret)
4589 		return ret;
4590 
4591 	if (opts) {
4592 		ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4593 		if (ret)
4594 			return ret;
4595 
4596 		ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4597 				opts->trsvcid ?: "none");
4598 		if (ret)
4599 			return ret;
4600 
4601 		ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4602 				opts->host_traddr ?: "none");
4603 		if (ret)
4604 			return ret;
4605 
4606 		ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4607 				opts->host_iface ?: "none");
4608 	}
4609 	return ret;
4610 }
4611 
4612 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4613 {
4614 	char *envp[2] = { envdata, NULL };
4615 
4616 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4617 }
4618 
4619 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4620 {
4621 	char *envp[2] = { NULL, NULL };
4622 	u32 aen_result = ctrl->aen_result;
4623 
4624 	ctrl->aen_result = 0;
4625 	if (!aen_result)
4626 		return;
4627 
4628 	envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4629 	if (!envp[0])
4630 		return;
4631 	kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4632 	kfree(envp[0]);
4633 }
4634 
4635 static void nvme_async_event_work(struct work_struct *work)
4636 {
4637 	struct nvme_ctrl *ctrl =
4638 		container_of(work, struct nvme_ctrl, async_event_work);
4639 
4640 	nvme_aen_uevent(ctrl);
4641 
4642 	/*
4643 	 * The transport drivers must guarantee AER submission here is safe by
4644 	 * flushing ctrl async_event_work after changing the controller state
4645 	 * from LIVE and before freeing the admin queue.
4646 	*/
4647 	if (ctrl->state == NVME_CTRL_LIVE)
4648 		ctrl->ops->submit_async_event(ctrl);
4649 }
4650 
4651 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4652 {
4653 
4654 	u32 csts;
4655 
4656 	if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4657 		return false;
4658 
4659 	if (csts == ~0)
4660 		return false;
4661 
4662 	return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4663 }
4664 
4665 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4666 {
4667 	struct nvme_fw_slot_info_log *log;
4668 
4669 	log = kmalloc(sizeof(*log), GFP_KERNEL);
4670 	if (!log)
4671 		return;
4672 
4673 	if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4674 			log, sizeof(*log), 0))
4675 		dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4676 	kfree(log);
4677 }
4678 
4679 static void nvme_fw_act_work(struct work_struct *work)
4680 {
4681 	struct nvme_ctrl *ctrl = container_of(work,
4682 				struct nvme_ctrl, fw_act_work);
4683 	unsigned long fw_act_timeout;
4684 
4685 	if (ctrl->mtfa)
4686 		fw_act_timeout = jiffies +
4687 				msecs_to_jiffies(ctrl->mtfa * 100);
4688 	else
4689 		fw_act_timeout = jiffies +
4690 				msecs_to_jiffies(admin_timeout * 1000);
4691 
4692 	nvme_stop_queues(ctrl);
4693 	while (nvme_ctrl_pp_status(ctrl)) {
4694 		if (time_after(jiffies, fw_act_timeout)) {
4695 			dev_warn(ctrl->device,
4696 				"Fw activation timeout, reset controller\n");
4697 			nvme_try_sched_reset(ctrl);
4698 			return;
4699 		}
4700 		msleep(100);
4701 	}
4702 
4703 	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4704 		return;
4705 
4706 	nvme_start_queues(ctrl);
4707 	/* read FW slot information to clear the AER */
4708 	nvme_get_fw_slot_info(ctrl);
4709 
4710 	queue_work(nvme_wq, &ctrl->async_event_work);
4711 }
4712 
4713 static u32 nvme_aer_type(u32 result)
4714 {
4715 	return result & 0x7;
4716 }
4717 
4718 static u32 nvme_aer_subtype(u32 result)
4719 {
4720 	return (result & 0xff00) >> 8;
4721 }
4722 
4723 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4724 {
4725 	u32 aer_notice_type = nvme_aer_subtype(result);
4726 	bool requeue = true;
4727 
4728 	trace_nvme_async_event(ctrl, aer_notice_type);
4729 
4730 	switch (aer_notice_type) {
4731 	case NVME_AER_NOTICE_NS_CHANGED:
4732 		set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4733 		nvme_queue_scan(ctrl);
4734 		break;
4735 	case NVME_AER_NOTICE_FW_ACT_STARTING:
4736 		/*
4737 		 * We are (ab)using the RESETTING state to prevent subsequent
4738 		 * recovery actions from interfering with the controller's
4739 		 * firmware activation.
4740 		 */
4741 		if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4742 			nvme_auth_stop(ctrl);
4743 			requeue = false;
4744 			queue_work(nvme_wq, &ctrl->fw_act_work);
4745 		}
4746 		break;
4747 #ifdef CONFIG_NVME_MULTIPATH
4748 	case NVME_AER_NOTICE_ANA:
4749 		if (!ctrl->ana_log_buf)
4750 			break;
4751 		queue_work(nvme_wq, &ctrl->ana_work);
4752 		break;
4753 #endif
4754 	case NVME_AER_NOTICE_DISC_CHANGED:
4755 		ctrl->aen_result = result;
4756 		break;
4757 	default:
4758 		dev_warn(ctrl->device, "async event result %08x\n", result);
4759 	}
4760 	return requeue;
4761 }
4762 
4763 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4764 {
4765 	trace_nvme_async_event(ctrl, NVME_AER_ERROR);
4766 	dev_warn(ctrl->device, "resetting controller due to AER\n");
4767 	nvme_reset_ctrl(ctrl);
4768 }
4769 
4770 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4771 		volatile union nvme_result *res)
4772 {
4773 	u32 result = le32_to_cpu(res->u32);
4774 	u32 aer_type = nvme_aer_type(result);
4775 	u32 aer_subtype = nvme_aer_subtype(result);
4776 	bool requeue = true;
4777 
4778 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4779 		return;
4780 
4781 	switch (aer_type) {
4782 	case NVME_AER_NOTICE:
4783 		requeue = nvme_handle_aen_notice(ctrl, result);
4784 		break;
4785 	case NVME_AER_ERROR:
4786 		/*
4787 		 * For a persistent internal error, don't run async_event_work
4788 		 * to submit a new AER. The controller reset will do it.
4789 		 */
4790 		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4791 			nvme_handle_aer_persistent_error(ctrl);
4792 			return;
4793 		}
4794 		fallthrough;
4795 	case NVME_AER_SMART:
4796 	case NVME_AER_CSS:
4797 	case NVME_AER_VS:
4798 		trace_nvme_async_event(ctrl, aer_type);
4799 		ctrl->aen_result = result;
4800 		break;
4801 	default:
4802 		break;
4803 	}
4804 
4805 	if (requeue)
4806 		queue_work(nvme_wq, &ctrl->async_event_work);
4807 }
4808 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4809 
4810 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4811 		const struct blk_mq_ops *ops, unsigned int flags,
4812 		unsigned int cmd_size)
4813 {
4814 	int ret;
4815 
4816 	memset(set, 0, sizeof(*set));
4817 	set->ops = ops;
4818 	set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4819 	if (ctrl->ops->flags & NVME_F_FABRICS)
4820 		set->reserved_tags = NVMF_RESERVED_TAGS;
4821 	set->numa_node = ctrl->numa_node;
4822 	set->flags = flags;
4823 	set->cmd_size = cmd_size;
4824 	set->driver_data = ctrl;
4825 	set->nr_hw_queues = 1;
4826 	set->timeout = NVME_ADMIN_TIMEOUT;
4827 	ret = blk_mq_alloc_tag_set(set);
4828 	if (ret)
4829 		return ret;
4830 
4831 	ctrl->admin_q = blk_mq_init_queue(set);
4832 	if (IS_ERR(ctrl->admin_q)) {
4833 		ret = PTR_ERR(ctrl->admin_q);
4834 		goto out_free_tagset;
4835 	}
4836 
4837 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4838 		ctrl->fabrics_q = blk_mq_init_queue(set);
4839 		if (IS_ERR(ctrl->fabrics_q)) {
4840 			ret = PTR_ERR(ctrl->fabrics_q);
4841 			goto out_cleanup_admin_q;
4842 		}
4843 	}
4844 
4845 	ctrl->admin_tagset = set;
4846 	return 0;
4847 
4848 out_cleanup_admin_q:
4849 	blk_mq_destroy_queue(ctrl->fabrics_q);
4850 out_free_tagset:
4851 	blk_mq_free_tag_set(ctrl->admin_tagset);
4852 	return ret;
4853 }
4854 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4855 
4856 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4857 {
4858 	blk_mq_destroy_queue(ctrl->admin_q);
4859 	if (ctrl->ops->flags & NVME_F_FABRICS)
4860 		blk_mq_destroy_queue(ctrl->fabrics_q);
4861 	blk_mq_free_tag_set(ctrl->admin_tagset);
4862 }
4863 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4864 
4865 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4866 		const struct blk_mq_ops *ops, unsigned int flags,
4867 		unsigned int cmd_size)
4868 {
4869 	int ret;
4870 
4871 	memset(set, 0, sizeof(*set));
4872 	set->ops = ops;
4873 	set->queue_depth = ctrl->sqsize + 1;
4874 	set->reserved_tags = NVMF_RESERVED_TAGS;
4875 	set->numa_node = ctrl->numa_node;
4876 	set->flags = flags;
4877 	set->cmd_size = cmd_size,
4878 	set->driver_data = ctrl;
4879 	set->nr_hw_queues = ctrl->queue_count - 1;
4880 	set->timeout = NVME_IO_TIMEOUT;
4881 	if (ops->map_queues)
4882 		set->nr_maps = ctrl->opts->nr_poll_queues ? HCTX_MAX_TYPES : 2;
4883 	ret = blk_mq_alloc_tag_set(set);
4884 	if (ret)
4885 		return ret;
4886 
4887 	if (ctrl->ops->flags & NVME_F_FABRICS) {
4888 		ctrl->connect_q = blk_mq_init_queue(set);
4889         	if (IS_ERR(ctrl->connect_q)) {
4890 			ret = PTR_ERR(ctrl->connect_q);
4891 			goto out_free_tag_set;
4892 		}
4893 	}
4894 
4895 	ctrl->tagset = set;
4896 	return 0;
4897 
4898 out_free_tag_set:
4899 	blk_mq_free_tag_set(set);
4900 	return ret;
4901 }
4902 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4903 
4904 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4905 {
4906 	if (ctrl->ops->flags & NVME_F_FABRICS)
4907 		blk_mq_destroy_queue(ctrl->connect_q);
4908 	blk_mq_free_tag_set(ctrl->tagset);
4909 }
4910 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4911 
4912 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4913 {
4914 	nvme_mpath_stop(ctrl);
4915 	nvme_auth_stop(ctrl);
4916 	nvme_stop_keep_alive(ctrl);
4917 	nvme_stop_failfast_work(ctrl);
4918 	flush_work(&ctrl->async_event_work);
4919 	cancel_work_sync(&ctrl->fw_act_work);
4920 	if (ctrl->ops->stop_ctrl)
4921 		ctrl->ops->stop_ctrl(ctrl);
4922 }
4923 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4924 
4925 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4926 {
4927 	nvme_start_keep_alive(ctrl);
4928 
4929 	nvme_enable_aen(ctrl);
4930 
4931 	/*
4932 	 * persistent discovery controllers need to send indication to userspace
4933 	 * to re-read the discovery log page to learn about possible changes
4934 	 * that were missed. We identify persistent discovery controllers by
4935 	 * checking that they started once before, hence are reconnecting back.
4936 	 */
4937 	if (test_and_set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4938 	    nvme_discovery_ctrl(ctrl))
4939 		nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4940 
4941 	if (ctrl->queue_count > 1) {
4942 		nvme_queue_scan(ctrl);
4943 		nvme_start_queues(ctrl);
4944 		nvme_mpath_update(ctrl);
4945 	}
4946 
4947 	nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4948 }
4949 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4950 
4951 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4952 {
4953 	nvme_hwmon_exit(ctrl);
4954 	nvme_fault_inject_fini(&ctrl->fault_inject);
4955 	dev_pm_qos_hide_latency_tolerance(ctrl->device);
4956 	cdev_device_del(&ctrl->cdev, ctrl->device);
4957 	nvme_put_ctrl(ctrl);
4958 }
4959 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4960 
4961 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4962 {
4963 	struct nvme_effects_log	*cel;
4964 	unsigned long i;
4965 
4966 	xa_for_each(&ctrl->cels, i, cel) {
4967 		xa_erase(&ctrl->cels, i);
4968 		kfree(cel);
4969 	}
4970 
4971 	xa_destroy(&ctrl->cels);
4972 }
4973 
4974 static void nvme_free_ctrl(struct device *dev)
4975 {
4976 	struct nvme_ctrl *ctrl =
4977 		container_of(dev, struct nvme_ctrl, ctrl_device);
4978 	struct nvme_subsystem *subsys = ctrl->subsys;
4979 
4980 	if (!subsys || ctrl->instance != subsys->instance)
4981 		ida_free(&nvme_instance_ida, ctrl->instance);
4982 
4983 	nvme_free_cels(ctrl);
4984 	nvme_mpath_uninit(ctrl);
4985 	nvme_auth_stop(ctrl);
4986 	nvme_auth_free(ctrl);
4987 	__free_page(ctrl->discard_page);
4988 
4989 	if (subsys) {
4990 		mutex_lock(&nvme_subsystems_lock);
4991 		list_del(&ctrl->subsys_entry);
4992 		sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4993 		mutex_unlock(&nvme_subsystems_lock);
4994 	}
4995 
4996 	ctrl->ops->free_ctrl(ctrl);
4997 
4998 	if (subsys)
4999 		nvme_put_subsystem(subsys);
5000 }
5001 
5002 /*
5003  * Initialize a NVMe controller structures.  This needs to be called during
5004  * earliest initialization so that we have the initialized structured around
5005  * during probing.
5006  */
5007 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
5008 		const struct nvme_ctrl_ops *ops, unsigned long quirks)
5009 {
5010 	int ret;
5011 
5012 	ctrl->state = NVME_CTRL_NEW;
5013 	clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
5014 	spin_lock_init(&ctrl->lock);
5015 	mutex_init(&ctrl->scan_lock);
5016 	INIT_LIST_HEAD(&ctrl->namespaces);
5017 	xa_init(&ctrl->cels);
5018 	init_rwsem(&ctrl->namespaces_rwsem);
5019 	ctrl->dev = dev;
5020 	ctrl->ops = ops;
5021 	ctrl->quirks = quirks;
5022 	ctrl->numa_node = NUMA_NO_NODE;
5023 	INIT_WORK(&ctrl->scan_work, nvme_scan_work);
5024 	INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
5025 	INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
5026 	INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
5027 	init_waitqueue_head(&ctrl->state_wq);
5028 
5029 	INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
5030 	INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
5031 	memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
5032 	ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
5033 
5034 	BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
5035 			PAGE_SIZE);
5036 	ctrl->discard_page = alloc_page(GFP_KERNEL);
5037 	if (!ctrl->discard_page) {
5038 		ret = -ENOMEM;
5039 		goto out;
5040 	}
5041 
5042 	ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
5043 	if (ret < 0)
5044 		goto out;
5045 	ctrl->instance = ret;
5046 
5047 	device_initialize(&ctrl->ctrl_device);
5048 	ctrl->device = &ctrl->ctrl_device;
5049 	ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
5050 			ctrl->instance);
5051 	ctrl->device->class = nvme_class;
5052 	ctrl->device->parent = ctrl->dev;
5053 	ctrl->device->groups = nvme_dev_attr_groups;
5054 	ctrl->device->release = nvme_free_ctrl;
5055 	dev_set_drvdata(ctrl->device, ctrl);
5056 	ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
5057 	if (ret)
5058 		goto out_release_instance;
5059 
5060 	nvme_get_ctrl(ctrl);
5061 	cdev_init(&ctrl->cdev, &nvme_dev_fops);
5062 	ctrl->cdev.owner = ops->module;
5063 	ret = cdev_device_add(&ctrl->cdev, ctrl->device);
5064 	if (ret)
5065 		goto out_free_name;
5066 
5067 	/*
5068 	 * Initialize latency tolerance controls.  The sysfs files won't
5069 	 * be visible to userspace unless the device actually supports APST.
5070 	 */
5071 	ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
5072 	dev_pm_qos_update_user_latency_tolerance(ctrl->device,
5073 		min(default_ps_max_latency_us, (unsigned long)S32_MAX));
5074 
5075 	nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
5076 	nvme_mpath_init_ctrl(ctrl);
5077 	nvme_auth_init_ctrl(ctrl);
5078 
5079 	return 0;
5080 out_free_name:
5081 	nvme_put_ctrl(ctrl);
5082 	kfree_const(ctrl->device->kobj.name);
5083 out_release_instance:
5084 	ida_free(&nvme_instance_ida, ctrl->instance);
5085 out:
5086 	if (ctrl->discard_page)
5087 		__free_page(ctrl->discard_page);
5088 	return ret;
5089 }
5090 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
5091 
5092 static void nvme_start_ns_queue(struct nvme_ns *ns)
5093 {
5094 	if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
5095 		blk_mq_unquiesce_queue(ns->queue);
5096 }
5097 
5098 static void nvme_stop_ns_queue(struct nvme_ns *ns)
5099 {
5100 	if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
5101 		blk_mq_quiesce_queue(ns->queue);
5102 	else
5103 		blk_mq_wait_quiesce_done(ns->queue);
5104 }
5105 
5106 /*
5107  * Prepare a queue for teardown.
5108  *
5109  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
5110  * the capacity to 0 after that to avoid blocking dispatchers that may be
5111  * holding bd_butex.  This will end buffered writers dirtying pages that can't
5112  * be synced.
5113  */
5114 static void nvme_set_queue_dying(struct nvme_ns *ns)
5115 {
5116 	if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
5117 		return;
5118 
5119 	blk_mark_disk_dead(ns->disk);
5120 	nvme_start_ns_queue(ns);
5121 
5122 	set_capacity_and_notify(ns->disk, 0);
5123 }
5124 
5125 /**
5126  * nvme_kill_queues(): Ends all namespace queues
5127  * @ctrl: the dead controller that needs to end
5128  *
5129  * Call this function when the driver determines it is unable to get the
5130  * controller in a state capable of servicing IO.
5131  */
5132 void nvme_kill_queues(struct nvme_ctrl *ctrl)
5133 {
5134 	struct nvme_ns *ns;
5135 
5136 	down_read(&ctrl->namespaces_rwsem);
5137 
5138 	/* Forcibly unquiesce queues to avoid blocking dispatch */
5139 	if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
5140 		nvme_start_admin_queue(ctrl);
5141 
5142 	list_for_each_entry(ns, &ctrl->namespaces, list)
5143 		nvme_set_queue_dying(ns);
5144 
5145 	up_read(&ctrl->namespaces_rwsem);
5146 }
5147 EXPORT_SYMBOL_GPL(nvme_kill_queues);
5148 
5149 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5150 {
5151 	struct nvme_ns *ns;
5152 
5153 	down_read(&ctrl->namespaces_rwsem);
5154 	list_for_each_entry(ns, &ctrl->namespaces, list)
5155 		blk_mq_unfreeze_queue(ns->queue);
5156 	up_read(&ctrl->namespaces_rwsem);
5157 }
5158 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5159 
5160 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5161 {
5162 	struct nvme_ns *ns;
5163 
5164 	down_read(&ctrl->namespaces_rwsem);
5165 	list_for_each_entry(ns, &ctrl->namespaces, list) {
5166 		timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5167 		if (timeout <= 0)
5168 			break;
5169 	}
5170 	up_read(&ctrl->namespaces_rwsem);
5171 	return timeout;
5172 }
5173 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5174 
5175 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5176 {
5177 	struct nvme_ns *ns;
5178 
5179 	down_read(&ctrl->namespaces_rwsem);
5180 	list_for_each_entry(ns, &ctrl->namespaces, list)
5181 		blk_mq_freeze_queue_wait(ns->queue);
5182 	up_read(&ctrl->namespaces_rwsem);
5183 }
5184 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5185 
5186 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5187 {
5188 	struct nvme_ns *ns;
5189 
5190 	down_read(&ctrl->namespaces_rwsem);
5191 	list_for_each_entry(ns, &ctrl->namespaces, list)
5192 		blk_freeze_queue_start(ns->queue);
5193 	up_read(&ctrl->namespaces_rwsem);
5194 }
5195 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5196 
5197 void nvme_stop_queues(struct nvme_ctrl *ctrl)
5198 {
5199 	struct nvme_ns *ns;
5200 
5201 	down_read(&ctrl->namespaces_rwsem);
5202 	list_for_each_entry(ns, &ctrl->namespaces, list)
5203 		nvme_stop_ns_queue(ns);
5204 	up_read(&ctrl->namespaces_rwsem);
5205 }
5206 EXPORT_SYMBOL_GPL(nvme_stop_queues);
5207 
5208 void nvme_start_queues(struct nvme_ctrl *ctrl)
5209 {
5210 	struct nvme_ns *ns;
5211 
5212 	down_read(&ctrl->namespaces_rwsem);
5213 	list_for_each_entry(ns, &ctrl->namespaces, list)
5214 		nvme_start_ns_queue(ns);
5215 	up_read(&ctrl->namespaces_rwsem);
5216 }
5217 EXPORT_SYMBOL_GPL(nvme_start_queues);
5218 
5219 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
5220 {
5221 	if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5222 		blk_mq_quiesce_queue(ctrl->admin_q);
5223 	else
5224 		blk_mq_wait_quiesce_done(ctrl->admin_q);
5225 }
5226 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
5227 
5228 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
5229 {
5230 	if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5231 		blk_mq_unquiesce_queue(ctrl->admin_q);
5232 }
5233 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
5234 
5235 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5236 {
5237 	struct nvme_ns *ns;
5238 
5239 	down_read(&ctrl->namespaces_rwsem);
5240 	list_for_each_entry(ns, &ctrl->namespaces, list)
5241 		blk_sync_queue(ns->queue);
5242 	up_read(&ctrl->namespaces_rwsem);
5243 }
5244 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5245 
5246 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5247 {
5248 	nvme_sync_io_queues(ctrl);
5249 	if (ctrl->admin_q)
5250 		blk_sync_queue(ctrl->admin_q);
5251 }
5252 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5253 
5254 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5255 {
5256 	if (file->f_op != &nvme_dev_fops)
5257 		return NULL;
5258 	return file->private_data;
5259 }
5260 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
5261 
5262 /*
5263  * Check we didn't inadvertently grow the command structure sizes:
5264  */
5265 static inline void _nvme_check_size(void)
5266 {
5267 	BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5268 	BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5269 	BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5270 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5271 	BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5272 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5273 	BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5274 	BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5275 	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5276 	BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5277 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5278 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5279 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5280 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5281 			NVME_IDENTIFY_DATA_SIZE);
5282 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5283 	BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5284 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5285 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5286 	BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5287 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5288 	BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5289 	BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5290 	BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5291 }
5292 
5293 
5294 static int __init nvme_core_init(void)
5295 {
5296 	int result = -ENOMEM;
5297 
5298 	_nvme_check_size();
5299 
5300 	nvme_wq = alloc_workqueue("nvme-wq",
5301 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5302 	if (!nvme_wq)
5303 		goto out;
5304 
5305 	nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
5306 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5307 	if (!nvme_reset_wq)
5308 		goto destroy_wq;
5309 
5310 	nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
5311 			WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5312 	if (!nvme_delete_wq)
5313 		goto destroy_reset_wq;
5314 
5315 	result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5316 			NVME_MINORS, "nvme");
5317 	if (result < 0)
5318 		goto destroy_delete_wq;
5319 
5320 	nvme_class = class_create(THIS_MODULE, "nvme");
5321 	if (IS_ERR(nvme_class)) {
5322 		result = PTR_ERR(nvme_class);
5323 		goto unregister_chrdev;
5324 	}
5325 	nvme_class->dev_uevent = nvme_class_uevent;
5326 
5327 	nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
5328 	if (IS_ERR(nvme_subsys_class)) {
5329 		result = PTR_ERR(nvme_subsys_class);
5330 		goto destroy_class;
5331 	}
5332 
5333 	result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5334 				     "nvme-generic");
5335 	if (result < 0)
5336 		goto destroy_subsys_class;
5337 
5338 	nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
5339 	if (IS_ERR(nvme_ns_chr_class)) {
5340 		result = PTR_ERR(nvme_ns_chr_class);
5341 		goto unregister_generic_ns;
5342 	}
5343 
5344 	return 0;
5345 
5346 unregister_generic_ns:
5347 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5348 destroy_subsys_class:
5349 	class_destroy(nvme_subsys_class);
5350 destroy_class:
5351 	class_destroy(nvme_class);
5352 unregister_chrdev:
5353 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5354 destroy_delete_wq:
5355 	destroy_workqueue(nvme_delete_wq);
5356 destroy_reset_wq:
5357 	destroy_workqueue(nvme_reset_wq);
5358 destroy_wq:
5359 	destroy_workqueue(nvme_wq);
5360 out:
5361 	return result;
5362 }
5363 
5364 static void __exit nvme_core_exit(void)
5365 {
5366 	class_destroy(nvme_ns_chr_class);
5367 	class_destroy(nvme_subsys_class);
5368 	class_destroy(nvme_class);
5369 	unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5370 	unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5371 	destroy_workqueue(nvme_delete_wq);
5372 	destroy_workqueue(nvme_reset_wq);
5373 	destroy_workqueue(nvme_wq);
5374 	ida_destroy(&nvme_ns_chr_minor_ida);
5375 	ida_destroy(&nvme_instance_ida);
5376 }
5377 
5378 MODULE_LICENSE("GPL");
5379 MODULE_VERSION("1.0");
5380 module_init(nvme_core_init);
5381 module_exit(nvme_core_exit);
5382