]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - drivers/nvme/host/core.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
UBUNTU: Ubuntu-5.11.0-22.23
[mirror_ubuntu-hirsute-kernel.git] / drivers / nvme / host / core.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
5 */
6
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/compat.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/hdreg.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/list_sort.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/pr.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
24
25 #include "nvme.h"
26 #include "fabrics.h"
27
28 #define CREATE_TRACE_POINTS
29 #include "trace.h"
30
31 #define NVME_MINORS (1U << MINORBITS)
32
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
37
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
42
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
46
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
50
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
55
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
59
60 static bool streams;
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
63
64 /*
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
68 *
69 * nvme_wq will host works such as scan, aen handling, fw activation,
70 * keep-alive, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
74 */
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
77
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
80
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
83
84 static LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
86
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_ctrl_base_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
91
92 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
93 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
94 unsigned nsid);
95
96 /*
97 * Prepare a queue for teardown.
98 *
99 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
100 * the capacity to 0 after that to avoid blocking dispatchers that may be
101 * holding bd_butex. This will end buffered writers dirtying pages that can't
102 * be synced.
103 */
104 static void nvme_set_queue_dying(struct nvme_ns *ns)
105 {
106 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
107 return;
108
109 blk_set_queue_dying(ns->queue);
110 blk_mq_unquiesce_queue(ns->queue);
111
112 set_capacity_and_notify(ns->disk, 0);
113 }
114
115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
116 {
117 /*
118 * Only new queue scan work when admin and IO queues are both alive
119 */
120 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
121 queue_work(nvme_wq, &ctrl->scan_work);
122 }
123
124 /*
125 * Use this function to proceed with scheduling reset_work for a controller
126 * that had previously been set to the resetting state. This is intended for
127 * code paths that can't be interrupted by other reset attempts. A hot removal
128 * may prevent this from succeeding.
129 */
130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
131 {
132 if (ctrl->state != NVME_CTRL_RESETTING)
133 return -EBUSY;
134 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
135 return -EBUSY;
136 return 0;
137 }
138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
139
140 static void nvme_failfast_work(struct work_struct *work)
141 {
142 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
143 struct nvme_ctrl, failfast_work);
144
145 if (ctrl->state != NVME_CTRL_CONNECTING)
146 return;
147
148 set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
149 dev_info(ctrl->device, "failfast expired\n");
150 nvme_kick_requeue_lists(ctrl);
151 }
152
153 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
154 {
155 if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
156 return;
157
158 schedule_delayed_work(&ctrl->failfast_work,
159 ctrl->opts->fast_io_fail_tmo * HZ);
160 }
161
162 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
163 {
164 if (!ctrl->opts)
165 return;
166
167 cancel_delayed_work_sync(&ctrl->failfast_work);
168 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
169 }
170
171
172 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
173 {
174 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
175 return -EBUSY;
176 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
177 return -EBUSY;
178 return 0;
179 }
180 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
181
182 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
183 {
184 int ret;
185
186 ret = nvme_reset_ctrl(ctrl);
187 if (!ret) {
188 flush_work(&ctrl->reset_work);
189 if (ctrl->state != NVME_CTRL_LIVE)
190 ret = -ENETRESET;
191 }
192
193 return ret;
194 }
195
196 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
197 {
198 dev_info(ctrl->device,
199 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
200
201 flush_work(&ctrl->reset_work);
202 nvme_stop_ctrl(ctrl);
203 nvme_remove_namespaces(ctrl);
204 ctrl->ops->delete_ctrl(ctrl);
205 nvme_uninit_ctrl(ctrl);
206 }
207
208 static void nvme_delete_ctrl_work(struct work_struct *work)
209 {
210 struct nvme_ctrl *ctrl =
211 container_of(work, struct nvme_ctrl, delete_work);
212
213 nvme_do_delete_ctrl(ctrl);
214 }
215
216 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
217 {
218 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
219 return -EBUSY;
220 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
221 return -EBUSY;
222 return 0;
223 }
224 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
225
226 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
227 {
228 /*
229 * Keep a reference until nvme_do_delete_ctrl() complete,
230 * since ->delete_ctrl can free the controller.
231 */
232 nvme_get_ctrl(ctrl);
233 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
234 nvme_do_delete_ctrl(ctrl);
235 nvme_put_ctrl(ctrl);
236 }
237
238 static blk_status_t nvme_error_status(u16 status)
239 {
240 switch (status & 0x7ff) {
241 case NVME_SC_SUCCESS:
242 return BLK_STS_OK;
243 case NVME_SC_CAP_EXCEEDED:
244 return BLK_STS_NOSPC;
245 case NVME_SC_LBA_RANGE:
246 case NVME_SC_CMD_INTERRUPTED:
247 case NVME_SC_NS_NOT_READY:
248 return BLK_STS_TARGET;
249 case NVME_SC_BAD_ATTRIBUTES:
250 case NVME_SC_ONCS_NOT_SUPPORTED:
251 case NVME_SC_INVALID_OPCODE:
252 case NVME_SC_INVALID_FIELD:
253 case NVME_SC_INVALID_NS:
254 return BLK_STS_NOTSUPP;
255 case NVME_SC_WRITE_FAULT:
256 case NVME_SC_READ_ERROR:
257 case NVME_SC_UNWRITTEN_BLOCK:
258 case NVME_SC_ACCESS_DENIED:
259 case NVME_SC_READ_ONLY:
260 case NVME_SC_COMPARE_FAILED:
261 return BLK_STS_MEDIUM;
262 case NVME_SC_GUARD_CHECK:
263 case NVME_SC_APPTAG_CHECK:
264 case NVME_SC_REFTAG_CHECK:
265 case NVME_SC_INVALID_PI:
266 return BLK_STS_PROTECTION;
267 case NVME_SC_RESERVATION_CONFLICT:
268 return BLK_STS_NEXUS;
269 case NVME_SC_HOST_PATH_ERROR:
270 return BLK_STS_TRANSPORT;
271 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
272 return BLK_STS_ZONE_ACTIVE_RESOURCE;
273 case NVME_SC_ZONE_TOO_MANY_OPEN:
274 return BLK_STS_ZONE_OPEN_RESOURCE;
275 default:
276 return BLK_STS_IOERR;
277 }
278 }
279
280 static void nvme_retry_req(struct request *req)
281 {
282 struct nvme_ns *ns = req->q->queuedata;
283 unsigned long delay = 0;
284 u16 crd;
285
286 /* The mask and shift result must be <= 3 */
287 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
288 if (ns && crd)
289 delay = ns->ctrl->crdt[crd - 1] * 100;
290
291 nvme_req(req)->retries++;
292 blk_mq_requeue_request(req, false);
293 blk_mq_delay_kick_requeue_list(req->q, delay);
294 }
295
296 enum nvme_disposition {
297 COMPLETE,
298 RETRY,
299 FAILOVER,
300 };
301
302 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
303 {
304 if (likely(nvme_req(req)->status == 0))
305 return COMPLETE;
306
307 if (blk_noretry_request(req) ||
308 (nvme_req(req)->status & NVME_SC_DNR) ||
309 nvme_req(req)->retries >= nvme_max_retries)
310 return COMPLETE;
311
312 if (req->cmd_flags & REQ_NVME_MPATH) {
313 if (nvme_is_path_error(nvme_req(req)->status) ||
314 blk_queue_dying(req->q))
315 return FAILOVER;
316 } else {
317 if (blk_queue_dying(req->q))
318 return COMPLETE;
319 }
320
321 return RETRY;
322 }
323
324 static inline void nvme_end_req(struct request *req)
325 {
326 blk_status_t status = nvme_error_status(nvme_req(req)->status);
327
328 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
329 req_op(req) == REQ_OP_ZONE_APPEND)
330 req->__sector = nvme_lba_to_sect(req->q->queuedata,
331 le64_to_cpu(nvme_req(req)->result.u64));
332
333 nvme_trace_bio_complete(req);
334 blk_mq_end_request(req, status);
335 }
336
337 void nvme_complete_rq(struct request *req)
338 {
339 trace_nvme_complete_rq(req);
340 nvme_cleanup_cmd(req);
341
342 if (nvme_req(req)->ctrl->kas)
343 nvme_req(req)->ctrl->comp_seen = true;
344
345 switch (nvme_decide_disposition(req)) {
346 case COMPLETE:
347 nvme_end_req(req);
348 return;
349 case RETRY:
350 nvme_retry_req(req);
351 return;
352 case FAILOVER:
353 nvme_failover_req(req);
354 return;
355 }
356 }
357 EXPORT_SYMBOL_GPL(nvme_complete_rq);
358
359 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
360 {
361 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
362 "Cancelling I/O %d", req->tag);
363
364 /* don't abort one completed request */
365 if (blk_mq_request_completed(req))
366 return true;
367
368 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
369 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
370 blk_mq_complete_request(req);
371 return true;
372 }
373 EXPORT_SYMBOL_GPL(nvme_cancel_request);
374
375 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
376 {
377 if (ctrl->tagset) {
378 blk_mq_tagset_busy_iter(ctrl->tagset,
379 nvme_cancel_request, ctrl);
380 blk_mq_tagset_wait_completed_request(ctrl->tagset);
381 }
382 }
383 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
384
385 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
386 {
387 if (ctrl->admin_tagset) {
388 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
389 nvme_cancel_request, ctrl);
390 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
391 }
392 }
393 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
394
395 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
396 enum nvme_ctrl_state new_state)
397 {
398 enum nvme_ctrl_state old_state;
399 unsigned long flags;
400 bool changed = false;
401
402 spin_lock_irqsave(&ctrl->lock, flags);
403
404 old_state = ctrl->state;
405 switch (new_state) {
406 case NVME_CTRL_LIVE:
407 switch (old_state) {
408 case NVME_CTRL_NEW:
409 case NVME_CTRL_RESETTING:
410 case NVME_CTRL_CONNECTING:
411 changed = true;
412 fallthrough;
413 default:
414 break;
415 }
416 break;
417 case NVME_CTRL_RESETTING:
418 switch (old_state) {
419 case NVME_CTRL_NEW:
420 case NVME_CTRL_LIVE:
421 changed = true;
422 fallthrough;
423 default:
424 break;
425 }
426 break;
427 case NVME_CTRL_CONNECTING:
428 switch (old_state) {
429 case NVME_CTRL_NEW:
430 case NVME_CTRL_RESETTING:
431 changed = true;
432 fallthrough;
433 default:
434 break;
435 }
436 break;
437 case NVME_CTRL_DELETING:
438 switch (old_state) {
439 case NVME_CTRL_LIVE:
440 case NVME_CTRL_RESETTING:
441 case NVME_CTRL_CONNECTING:
442 changed = true;
443 fallthrough;
444 default:
445 break;
446 }
447 break;
448 case NVME_CTRL_DELETING_NOIO:
449 switch (old_state) {
450 case NVME_CTRL_DELETING:
451 case NVME_CTRL_DEAD:
452 changed = true;
453 fallthrough;
454 default:
455 break;
456 }
457 break;
458 case NVME_CTRL_DEAD:
459 switch (old_state) {
460 case NVME_CTRL_DELETING:
461 changed = true;
462 fallthrough;
463 default:
464 break;
465 }
466 break;
467 default:
468 break;
469 }
470
471 if (changed) {
472 ctrl->state = new_state;
473 wake_up_all(&ctrl->state_wq);
474 }
475
476 spin_unlock_irqrestore(&ctrl->lock, flags);
477 if (!changed)
478 return false;
479
480 if (ctrl->state == NVME_CTRL_LIVE) {
481 if (old_state == NVME_CTRL_CONNECTING)
482 nvme_stop_failfast_work(ctrl);
483 nvme_kick_requeue_lists(ctrl);
484 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
485 old_state == NVME_CTRL_RESETTING) {
486 nvme_start_failfast_work(ctrl);
487 }
488 return changed;
489 }
490 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
491
492 /*
493 * Returns true for sink states that can't ever transition back to live.
494 */
495 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
496 {
497 switch (ctrl->state) {
498 case NVME_CTRL_NEW:
499 case NVME_CTRL_LIVE:
500 case NVME_CTRL_RESETTING:
501 case NVME_CTRL_CONNECTING:
502 return false;
503 case NVME_CTRL_DELETING:
504 case NVME_CTRL_DELETING_NOIO:
505 case NVME_CTRL_DEAD:
506 return true;
507 default:
508 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
509 return true;
510 }
511 }
512
513 /*
514 * Waits for the controller state to be resetting, or returns false if it is
515 * not possible to ever transition to that state.
516 */
517 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
518 {
519 wait_event(ctrl->state_wq,
520 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
521 nvme_state_terminal(ctrl));
522 return ctrl->state == NVME_CTRL_RESETTING;
523 }
524 EXPORT_SYMBOL_GPL(nvme_wait_reset);
525
526 static void nvme_free_ns_head(struct kref *ref)
527 {
528 struct nvme_ns_head *head =
529 container_of(ref, struct nvme_ns_head, ref);
530
531 nvme_mpath_remove_disk(head);
532 ida_simple_remove(&head->subsys->ns_ida, head->instance);
533 cleanup_srcu_struct(&head->srcu);
534 nvme_put_subsystem(head->subsys);
535 kfree(head);
536 }
537
538 static void nvme_put_ns_head(struct nvme_ns_head *head)
539 {
540 kref_put(&head->ref, nvme_free_ns_head);
541 }
542
543 static void nvme_free_ns(struct kref *kref)
544 {
545 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
546
547 if (ns->ndev)
548 nvme_nvm_unregister(ns);
549
550 put_disk(ns->disk);
551 nvme_put_ns_head(ns->head);
552 nvme_put_ctrl(ns->ctrl);
553 kfree(ns);
554 }
555
556 void nvme_put_ns(struct nvme_ns *ns)
557 {
558 kref_put(&ns->kref, nvme_free_ns);
559 }
560 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
561
562 static inline void nvme_clear_nvme_request(struct request *req)
563 {
564 if (!(req->rq_flags & RQF_DONTPREP)) {
565 nvme_req(req)->retries = 0;
566 nvme_req(req)->flags = 0;
567 req->rq_flags |= RQF_DONTPREP;
568 }
569 }
570
571 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
572 {
573 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
574 }
575
576 static inline void nvme_init_request(struct request *req,
577 struct nvme_command *cmd)
578 {
579 if (req->q->queuedata)
580 req->timeout = NVME_IO_TIMEOUT;
581 else /* no queuedata implies admin queue */
582 req->timeout = NVME_ADMIN_TIMEOUT;
583
584 req->cmd_flags |= REQ_FAILFAST_DRIVER;
585 nvme_clear_nvme_request(req);
586 nvme_req(req)->cmd = cmd;
587 }
588
589 struct request *nvme_alloc_request(struct request_queue *q,
590 struct nvme_command *cmd, blk_mq_req_flags_t flags)
591 {
592 struct request *req;
593
594 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
595 if (!IS_ERR(req))
596 nvme_init_request(req, cmd);
597 return req;
598 }
599 EXPORT_SYMBOL_GPL(nvme_alloc_request);
600
601 static struct request *nvme_alloc_request_qid(struct request_queue *q,
602 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
603 {
604 struct request *req;
605
606 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
607 qid ? qid - 1 : 0);
608 if (!IS_ERR(req))
609 nvme_init_request(req, cmd);
610 return req;
611 }
612
613 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
614 {
615 struct nvme_command c;
616
617 memset(&c, 0, sizeof(c));
618
619 c.directive.opcode = nvme_admin_directive_send;
620 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
621 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
622 c.directive.dtype = NVME_DIR_IDENTIFY;
623 c.directive.tdtype = NVME_DIR_STREAMS;
624 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
625
626 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
627 }
628
629 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
630 {
631 return nvme_toggle_streams(ctrl, false);
632 }
633
634 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
635 {
636 return nvme_toggle_streams(ctrl, true);
637 }
638
639 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
640 struct streams_directive_params *s, u32 nsid)
641 {
642 struct nvme_command c;
643
644 memset(&c, 0, sizeof(c));
645 memset(s, 0, sizeof(*s));
646
647 c.directive.opcode = nvme_admin_directive_recv;
648 c.directive.nsid = cpu_to_le32(nsid);
649 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
650 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
651 c.directive.dtype = NVME_DIR_STREAMS;
652
653 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
654 }
655
656 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
657 {
658 struct streams_directive_params s;
659 int ret;
660
661 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
662 return 0;
663 if (!streams)
664 return 0;
665
666 ret = nvme_enable_streams(ctrl);
667 if (ret)
668 return ret;
669
670 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
671 if (ret)
672 goto out_disable_stream;
673
674 ctrl->nssa = le16_to_cpu(s.nssa);
675 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
676 dev_info(ctrl->device, "too few streams (%u) available\n",
677 ctrl->nssa);
678 goto out_disable_stream;
679 }
680
681 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
682 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
683 return 0;
684
685 out_disable_stream:
686 nvme_disable_streams(ctrl);
687 return ret;
688 }
689
690 /*
691 * Check if 'req' has a write hint associated with it. If it does, assign
692 * a valid namespace stream to the write.
693 */
694 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
695 struct request *req, u16 *control,
696 u32 *dsmgmt)
697 {
698 enum rw_hint streamid = req->write_hint;
699
700 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
701 streamid = 0;
702 else {
703 streamid--;
704 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
705 return;
706
707 *control |= NVME_RW_DTYPE_STREAMS;
708 *dsmgmt |= streamid << 16;
709 }
710
711 if (streamid < ARRAY_SIZE(req->q->write_hints))
712 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
713 }
714
715 static void nvme_setup_passthrough(struct request *req,
716 struct nvme_command *cmd)
717 {
718 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
719 /* passthru commands should let the driver set the SGL flags */
720 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
721 }
722
723 static inline void nvme_setup_flush(struct nvme_ns *ns,
724 struct nvme_command *cmnd)
725 {
726 cmnd->common.opcode = nvme_cmd_flush;
727 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
728 }
729
730 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
731 struct nvme_command *cmnd)
732 {
733 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
734 struct nvme_dsm_range *range;
735 struct bio *bio;
736
737 /*
738 * Some devices do not consider the DSM 'Number of Ranges' field when
739 * determining how much data to DMA. Always allocate memory for maximum
740 * number of segments to prevent device reading beyond end of buffer.
741 */
742 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
743
744 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
745 if (!range) {
746 /*
747 * If we fail allocation our range, fallback to the controller
748 * discard page. If that's also busy, it's safe to return
749 * busy, as we know we can make progress once that's freed.
750 */
751 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
752 return BLK_STS_RESOURCE;
753
754 range = page_address(ns->ctrl->discard_page);
755 }
756
757 __rq_for_each_bio(bio, req) {
758 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
759 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
760
761 if (n < segments) {
762 range[n].cattr = cpu_to_le32(0);
763 range[n].nlb = cpu_to_le32(nlb);
764 range[n].slba = cpu_to_le64(slba);
765 }
766 n++;
767 }
768
769 if (WARN_ON_ONCE(n != segments)) {
770 if (virt_to_page(range) == ns->ctrl->discard_page)
771 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
772 else
773 kfree(range);
774 return BLK_STS_IOERR;
775 }
776
777 cmnd->dsm.opcode = nvme_cmd_dsm;
778 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
779 cmnd->dsm.nr = cpu_to_le32(segments - 1);
780 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
781
782 req->special_vec.bv_page = virt_to_page(range);
783 req->special_vec.bv_offset = offset_in_page(range);
784 req->special_vec.bv_len = alloc_size;
785 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
786
787 return BLK_STS_OK;
788 }
789
790 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
791 struct request *req, struct nvme_command *cmnd)
792 {
793 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
794 return nvme_setup_discard(ns, req, cmnd);
795
796 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
797 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
798 cmnd->write_zeroes.slba =
799 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
800 cmnd->write_zeroes.length =
801 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
802 cmnd->write_zeroes.control = 0;
803 return BLK_STS_OK;
804 }
805
806 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
807 struct request *req, struct nvme_command *cmnd,
808 enum nvme_opcode op)
809 {
810 struct nvme_ctrl *ctrl = ns->ctrl;
811 u16 control = 0;
812 u32 dsmgmt = 0;
813
814 if (req->cmd_flags & REQ_FUA)
815 control |= NVME_RW_FUA;
816 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
817 control |= NVME_RW_LR;
818
819 if (req->cmd_flags & REQ_RAHEAD)
820 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
821
822 cmnd->rw.opcode = op;
823 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
824 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
825 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
826
827 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
828 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
829
830 if (ns->ms) {
831 /*
832 * If formated with metadata, the block layer always provides a
833 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
834 * we enable the PRACT bit for protection information or set the
835 * namespace capacity to zero to prevent any I/O.
836 */
837 if (!blk_integrity_rq(req)) {
838 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
839 return BLK_STS_NOTSUPP;
840 control |= NVME_RW_PRINFO_PRACT;
841 }
842
843 switch (ns->pi_type) {
844 case NVME_NS_DPS_PI_TYPE3:
845 control |= NVME_RW_PRINFO_PRCHK_GUARD;
846 break;
847 case NVME_NS_DPS_PI_TYPE1:
848 case NVME_NS_DPS_PI_TYPE2:
849 control |= NVME_RW_PRINFO_PRCHK_GUARD |
850 NVME_RW_PRINFO_PRCHK_REF;
851 if (op == nvme_cmd_zone_append)
852 control |= NVME_RW_APPEND_PIREMAP;
853 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
854 break;
855 }
856 }
857
858 cmnd->rw.control = cpu_to_le16(control);
859 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
860 return 0;
861 }
862
863 void nvme_cleanup_cmd(struct request *req)
864 {
865 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
866 struct nvme_ns *ns = req->rq_disk->private_data;
867 struct page *page = req->special_vec.bv_page;
868
869 if (page == ns->ctrl->discard_page)
870 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
871 else
872 kfree(page_address(page) + req->special_vec.bv_offset);
873 }
874 }
875 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
876
877 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
878 struct nvme_command *cmd)
879 {
880 blk_status_t ret = BLK_STS_OK;
881
882 nvme_clear_nvme_request(req);
883
884 memset(cmd, 0, sizeof(*cmd));
885 switch (req_op(req)) {
886 case REQ_OP_DRV_IN:
887 case REQ_OP_DRV_OUT:
888 nvme_setup_passthrough(req, cmd);
889 break;
890 case REQ_OP_FLUSH:
891 nvme_setup_flush(ns, cmd);
892 break;
893 case REQ_OP_ZONE_RESET_ALL:
894 case REQ_OP_ZONE_RESET:
895 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
896 break;
897 case REQ_OP_ZONE_OPEN:
898 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
899 break;
900 case REQ_OP_ZONE_CLOSE:
901 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
902 break;
903 case REQ_OP_ZONE_FINISH:
904 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
905 break;
906 case REQ_OP_WRITE_ZEROES:
907 ret = nvme_setup_write_zeroes(ns, req, cmd);
908 break;
909 case REQ_OP_DISCARD:
910 ret = nvme_setup_discard(ns, req, cmd);
911 break;
912 case REQ_OP_READ:
913 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
914 break;
915 case REQ_OP_WRITE:
916 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
917 break;
918 case REQ_OP_ZONE_APPEND:
919 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
920 break;
921 default:
922 WARN_ON_ONCE(1);
923 return BLK_STS_IOERR;
924 }
925
926 cmd->common.command_id = req->tag;
927 trace_nvme_setup_cmd(req, cmd);
928 return ret;
929 }
930 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
931
932 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
933 {
934 struct completion *waiting = rq->end_io_data;
935
936 rq->end_io_data = NULL;
937 complete(waiting);
938 }
939
940 static void nvme_execute_rq_polled(struct request_queue *q,
941 struct gendisk *bd_disk, struct request *rq, int at_head)
942 {
943 DECLARE_COMPLETION_ONSTACK(wait);
944
945 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
946
947 rq->cmd_flags |= REQ_HIPRI;
948 rq->end_io_data = &wait;
949 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
950
951 while (!completion_done(&wait)) {
952 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
953 cond_resched();
954 }
955 }
956
957 /*
958 * Returns 0 on success. If the result is negative, it's a Linux error code;
959 * if the result is positive, it's an NVM Express status code
960 */
961 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
962 union nvme_result *result, void *buffer, unsigned bufflen,
963 unsigned timeout, int qid, int at_head,
964 blk_mq_req_flags_t flags, bool poll)
965 {
966 struct request *req;
967 int ret;
968
969 if (qid == NVME_QID_ANY)
970 req = nvme_alloc_request(q, cmd, flags);
971 else
972 req = nvme_alloc_request_qid(q, cmd, flags, qid);
973 if (IS_ERR(req))
974 return PTR_ERR(req);
975
976 if (timeout)
977 req->timeout = timeout;
978
979 if (buffer && bufflen) {
980 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
981 if (ret)
982 goto out;
983 }
984
985 if (poll)
986 nvme_execute_rq_polled(req->q, NULL, req, at_head);
987 else
988 blk_execute_rq(req->q, NULL, req, at_head);
989 if (result)
990 *result = nvme_req(req)->result;
991 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
992 ret = -EINTR;
993 else
994 ret = nvme_req(req)->status;
995 out:
996 blk_mq_free_request(req);
997 return ret;
998 }
999 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1000
1001 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1002 void *buffer, unsigned bufflen)
1003 {
1004 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1005 NVME_QID_ANY, 0, 0, false);
1006 }
1007 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1008
1009 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
1010 unsigned len, u32 seed, bool write)
1011 {
1012 struct bio_integrity_payload *bip;
1013 int ret = -ENOMEM;
1014 void *buf;
1015
1016 buf = kmalloc(len, GFP_KERNEL);
1017 if (!buf)
1018 goto out;
1019
1020 ret = -EFAULT;
1021 if (write && copy_from_user(buf, ubuf, len))
1022 goto out_free_meta;
1023
1024 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1025 if (IS_ERR(bip)) {
1026 ret = PTR_ERR(bip);
1027 goto out_free_meta;
1028 }
1029
1030 bip->bip_iter.bi_size = len;
1031 bip->bip_iter.bi_sector = seed;
1032 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1033 offset_in_page(buf));
1034 if (ret == len)
1035 return buf;
1036 ret = -ENOMEM;
1037 out_free_meta:
1038 kfree(buf);
1039 out:
1040 return ERR_PTR(ret);
1041 }
1042
1043 static u32 nvme_known_admin_effects(u8 opcode)
1044 {
1045 switch (opcode) {
1046 case nvme_admin_format_nvm:
1047 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1048 NVME_CMD_EFFECTS_CSE_MASK;
1049 case nvme_admin_sanitize_nvm:
1050 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1051 default:
1052 break;
1053 }
1054 return 0;
1055 }
1056
1057 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1058 {
1059 u32 effects = 0;
1060
1061 if (ns) {
1062 if (ns->head->effects)
1063 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1064 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1065 dev_warn(ctrl->device,
1066 "IO command:%02x has unhandled effects:%08x\n",
1067 opcode, effects);
1068 return 0;
1069 }
1070
1071 if (ctrl->effects)
1072 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1073 effects |= nvme_known_admin_effects(opcode);
1074
1075 return effects;
1076 }
1077 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1078
1079 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1080 u8 opcode)
1081 {
1082 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1083
1084 /*
1085 * For simplicity, IO to all namespaces is quiesced even if the command
1086 * effects say only one namespace is affected.
1087 */
1088 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1089 mutex_lock(&ctrl->scan_lock);
1090 mutex_lock(&ctrl->subsys->lock);
1091 nvme_mpath_start_freeze(ctrl->subsys);
1092 nvme_mpath_wait_freeze(ctrl->subsys);
1093 nvme_start_freeze(ctrl);
1094 nvme_wait_freeze(ctrl);
1095 }
1096 return effects;
1097 }
1098
1099 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1100 {
1101 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1102 nvme_unfreeze(ctrl);
1103 nvme_mpath_unfreeze(ctrl->subsys);
1104 mutex_unlock(&ctrl->subsys->lock);
1105 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1106 mutex_unlock(&ctrl->scan_lock);
1107 }
1108 if (effects & NVME_CMD_EFFECTS_CCC)
1109 nvme_init_identify(ctrl);
1110 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1111 nvme_queue_scan(ctrl);
1112 flush_work(&ctrl->scan_work);
1113 }
1114 }
1115
1116 void nvme_execute_passthru_rq(struct request *rq)
1117 {
1118 struct nvme_command *cmd = nvme_req(rq)->cmd;
1119 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1120 struct nvme_ns *ns = rq->q->queuedata;
1121 struct gendisk *disk = ns ? ns->disk : NULL;
1122 u32 effects;
1123
1124 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1125 blk_execute_rq(rq->q, disk, rq, 0);
1126 nvme_passthru_end(ctrl, effects);
1127 }
1128 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1129
1130 static int nvme_submit_user_cmd(struct request_queue *q,
1131 struct nvme_command *cmd, void __user *ubuffer,
1132 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1133 u32 meta_seed, u64 *result, unsigned timeout)
1134 {
1135 bool write = nvme_is_write(cmd);
1136 struct nvme_ns *ns = q->queuedata;
1137 struct gendisk *disk = ns ? ns->disk : NULL;
1138 struct request *req;
1139 struct bio *bio = NULL;
1140 void *meta = NULL;
1141 int ret;
1142
1143 req = nvme_alloc_request(q, cmd, 0);
1144 if (IS_ERR(req))
1145 return PTR_ERR(req);
1146
1147 if (timeout)
1148 req->timeout = timeout;
1149 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1150
1151 if (ubuffer && bufflen) {
1152 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1153 GFP_KERNEL);
1154 if (ret)
1155 goto out;
1156 bio = req->bio;
1157 bio->bi_disk = disk;
1158 if (disk && meta_buffer && meta_len) {
1159 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1160 meta_seed, write);
1161 if (IS_ERR(meta)) {
1162 ret = PTR_ERR(meta);
1163 goto out_unmap;
1164 }
1165 req->cmd_flags |= REQ_INTEGRITY;
1166 }
1167 }
1168
1169 nvme_execute_passthru_rq(req);
1170 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1171 ret = -EINTR;
1172 else
1173 ret = nvme_req(req)->status;
1174 if (result)
1175 *result = le64_to_cpu(nvme_req(req)->result.u64);
1176 if (meta && !ret && !write) {
1177 if (copy_to_user(meta_buffer, meta, meta_len))
1178 ret = -EFAULT;
1179 }
1180 kfree(meta);
1181 out_unmap:
1182 if (bio)
1183 blk_rq_unmap_user(bio);
1184 out:
1185 blk_mq_free_request(req);
1186 return ret;
1187 }
1188
1189 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1190 {
1191 struct nvme_ctrl *ctrl = rq->end_io_data;
1192 unsigned long flags;
1193 bool startka = false;
1194
1195 blk_mq_free_request(rq);
1196
1197 if (status) {
1198 dev_err(ctrl->device,
1199 "failed nvme_keep_alive_end_io error=%d\n",
1200 status);
1201 return;
1202 }
1203
1204 ctrl->comp_seen = false;
1205 spin_lock_irqsave(&ctrl->lock, flags);
1206 if (ctrl->state == NVME_CTRL_LIVE ||
1207 ctrl->state == NVME_CTRL_CONNECTING)
1208 startka = true;
1209 spin_unlock_irqrestore(&ctrl->lock, flags);
1210 if (startka)
1211 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1212 }
1213
1214 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1215 {
1216 struct request *rq;
1217
1218 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1219 BLK_MQ_REQ_RESERVED);
1220 if (IS_ERR(rq))
1221 return PTR_ERR(rq);
1222
1223 rq->timeout = ctrl->kato * HZ;
1224 rq->end_io_data = ctrl;
1225
1226 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1227
1228 return 0;
1229 }
1230
1231 static void nvme_keep_alive_work(struct work_struct *work)
1232 {
1233 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1234 struct nvme_ctrl, ka_work);
1235 bool comp_seen = ctrl->comp_seen;
1236
1237 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1238 dev_dbg(ctrl->device,
1239 "reschedule traffic based keep-alive timer\n");
1240 ctrl->comp_seen = false;
1241 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1242 return;
1243 }
1244
1245 if (nvme_keep_alive(ctrl)) {
1246 /* allocation failure, reset the controller */
1247 dev_err(ctrl->device, "keep-alive failed\n");
1248 nvme_reset_ctrl(ctrl);
1249 return;
1250 }
1251 }
1252
1253 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1254 {
1255 if (unlikely(ctrl->kato == 0))
1256 return;
1257
1258 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1259 }
1260
1261 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1262 {
1263 if (unlikely(ctrl->kato == 0))
1264 return;
1265
1266 cancel_delayed_work_sync(&ctrl->ka_work);
1267 }
1268 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1269
1270 /*
1271 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1272 * flag, thus sending any new CNS opcodes has a big chance of not working.
1273 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1274 * (but not for any later version).
1275 */
1276 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1277 {
1278 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1279 return ctrl->vs < NVME_VS(1, 2, 0);
1280 return ctrl->vs < NVME_VS(1, 1, 0);
1281 }
1282
1283 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1284 {
1285 struct nvme_command c = { };
1286 int error;
1287
1288 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1289 c.identify.opcode = nvme_admin_identify;
1290 c.identify.cns = NVME_ID_CNS_CTRL;
1291
1292 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1293 if (!*id)
1294 return -ENOMEM;
1295
1296 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1297 sizeof(struct nvme_id_ctrl));
1298 if (error)
1299 kfree(*id);
1300 return error;
1301 }
1302
1303 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1304 {
1305 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1306 }
1307
1308 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1309 struct nvme_ns_id_desc *cur, bool *csi_seen)
1310 {
1311 const char *warn_str = "ctrl returned bogus length:";
1312 void *data = cur;
1313
1314 switch (cur->nidt) {
1315 case NVME_NIDT_EUI64:
1316 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1317 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1318 warn_str, cur->nidl);
1319 return -1;
1320 }
1321 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1322 return NVME_NIDT_EUI64_LEN;
1323 case NVME_NIDT_NGUID:
1324 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1325 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1326 warn_str, cur->nidl);
1327 return -1;
1328 }
1329 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1330 return NVME_NIDT_NGUID_LEN;
1331 case NVME_NIDT_UUID:
1332 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1333 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1334 warn_str, cur->nidl);
1335 return -1;
1336 }
1337 uuid_copy(&ids->uuid, data + sizeof(*cur));
1338 return NVME_NIDT_UUID_LEN;
1339 case NVME_NIDT_CSI:
1340 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1341 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1342 warn_str, cur->nidl);
1343 return -1;
1344 }
1345 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1346 *csi_seen = true;
1347 return NVME_NIDT_CSI_LEN;
1348 default:
1349 /* Skip unknown types */
1350 return cur->nidl;
1351 }
1352 }
1353
1354 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1355 struct nvme_ns_ids *ids)
1356 {
1357 struct nvme_command c = { };
1358 bool csi_seen = false;
1359 int status, pos, len;
1360 void *data;
1361
1362 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1363 return 0;
1364 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1365 return 0;
1366
1367 c.identify.opcode = nvme_admin_identify;
1368 c.identify.nsid = cpu_to_le32(nsid);
1369 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1370
1371 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1372 if (!data)
1373 return -ENOMEM;
1374
1375 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1376 NVME_IDENTIFY_DATA_SIZE);
1377 if (status) {
1378 dev_warn(ctrl->device,
1379 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1380 nsid, status);
1381 goto free_data;
1382 }
1383
1384 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1385 struct nvme_ns_id_desc *cur = data + pos;
1386
1387 if (cur->nidl == 0)
1388 break;
1389
1390 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1391 if (len < 0)
1392 break;
1393
1394 len += sizeof(*cur);
1395 }
1396
1397 if (nvme_multi_css(ctrl) && !csi_seen) {
1398 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1399 nsid);
1400 status = -EINVAL;
1401 }
1402
1403 free_data:
1404 kfree(data);
1405 return status;
1406 }
1407
1408 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1409 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1410 {
1411 struct nvme_command c = { };
1412 int error;
1413
1414 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1415 c.identify.opcode = nvme_admin_identify;
1416 c.identify.nsid = cpu_to_le32(nsid);
1417 c.identify.cns = NVME_ID_CNS_NS;
1418
1419 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1420 if (!*id)
1421 return -ENOMEM;
1422
1423 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1424 if (error) {
1425 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1426 goto out_free_id;
1427 }
1428
1429 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1430 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1431 goto out_free_id;
1432
1433 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1434 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1435 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1436 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1437 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1438 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1439
1440 return 0;
1441
1442 out_free_id:
1443 kfree(*id);
1444 return error;
1445 }
1446
1447 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1448 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1449 {
1450 union nvme_result res = { 0 };
1451 struct nvme_command c;
1452 int ret;
1453
1454 memset(&c, 0, sizeof(c));
1455 c.features.opcode = op;
1456 c.features.fid = cpu_to_le32(fid);
1457 c.features.dword11 = cpu_to_le32(dword11);
1458
1459 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1460 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1461 if (ret >= 0 && result)
1462 *result = le32_to_cpu(res.u32);
1463 return ret;
1464 }
1465
1466 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1467 unsigned int dword11, void *buffer, size_t buflen,
1468 u32 *result)
1469 {
1470 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1471 buflen, result);
1472 }
1473 EXPORT_SYMBOL_GPL(nvme_set_features);
1474
1475 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1476 unsigned int dword11, void *buffer, size_t buflen,
1477 u32 *result)
1478 {
1479 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1480 buflen, result);
1481 }
1482 EXPORT_SYMBOL_GPL(nvme_get_features);
1483
1484 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1485 {
1486 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1487 u32 result;
1488 int status, nr_io_queues;
1489
1490 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1491 &result);
1492 if (status < 0)
1493 return status;
1494
1495 /*
1496 * Degraded controllers might return an error when setting the queue
1497 * count. We still want to be able to bring them online and offer
1498 * access to the admin queue, as that might be only way to fix them up.
1499 */
1500 if (status > 0) {
1501 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1502 *count = 0;
1503 } else {
1504 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1505 *count = min(*count, nr_io_queues);
1506 }
1507
1508 return 0;
1509 }
1510 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1511
1512 #define NVME_AEN_SUPPORTED \
1513 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1514 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1515
1516 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1517 {
1518 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1519 int status;
1520
1521 if (!supported_aens)
1522 return;
1523
1524 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1525 NULL, 0, &result);
1526 if (status)
1527 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1528 supported_aens);
1529
1530 queue_work(nvme_wq, &ctrl->async_event_work);
1531 }
1532
1533 /*
1534 * Convert integer values from ioctl structures to user pointers, silently
1535 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1536 * kernels.
1537 */
1538 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1539 {
1540 if (in_compat_syscall())
1541 ptrval = (compat_uptr_t)ptrval;
1542 return (void __user *)ptrval;
1543 }
1544
1545 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1546 {
1547 struct nvme_user_io io;
1548 struct nvme_command c;
1549 unsigned length, meta_len;
1550 void __user *metadata;
1551
1552 if (copy_from_user(&io, uio, sizeof(io)))
1553 return -EFAULT;
1554 if (io.flags)
1555 return -EINVAL;
1556
1557 switch (io.opcode) {
1558 case nvme_cmd_write:
1559 case nvme_cmd_read:
1560 case nvme_cmd_compare:
1561 break;
1562 default:
1563 return -EINVAL;
1564 }
1565
1566 length = (io.nblocks + 1) << ns->lba_shift;
1567
1568 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1569 ns->ms == sizeof(struct t10_pi_tuple)) {
1570 /*
1571 * Protection information is stripped/inserted by the
1572 * controller.
1573 */
1574 if (nvme_to_user_ptr(io.metadata))
1575 return -EINVAL;
1576 meta_len = 0;
1577 metadata = NULL;
1578 } else {
1579 meta_len = (io.nblocks + 1) * ns->ms;
1580 metadata = nvme_to_user_ptr(io.metadata);
1581 }
1582
1583 if (ns->features & NVME_NS_EXT_LBAS) {
1584 length += meta_len;
1585 meta_len = 0;
1586 } else if (meta_len) {
1587 if ((io.metadata & 3) || !io.metadata)
1588 return -EINVAL;
1589 }
1590
1591 memset(&c, 0, sizeof(c));
1592 c.rw.opcode = io.opcode;
1593 c.rw.flags = io.flags;
1594 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1595 c.rw.slba = cpu_to_le64(io.slba);
1596 c.rw.length = cpu_to_le16(io.nblocks);
1597 c.rw.control = cpu_to_le16(io.control);
1598 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1599 c.rw.reftag = cpu_to_le32(io.reftag);
1600 c.rw.apptag = cpu_to_le16(io.apptag);
1601 c.rw.appmask = cpu_to_le16(io.appmask);
1602
1603 return nvme_submit_user_cmd(ns->queue, &c,
1604 nvme_to_user_ptr(io.addr), length,
1605 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1606 }
1607
1608 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1609 struct nvme_passthru_cmd __user *ucmd)
1610 {
1611 struct nvme_passthru_cmd cmd;
1612 struct nvme_command c;
1613 unsigned timeout = 0;
1614 u64 result;
1615 int status;
1616
1617 if (!capable(CAP_SYS_ADMIN))
1618 return -EACCES;
1619 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1620 return -EFAULT;
1621 if (cmd.flags)
1622 return -EINVAL;
1623
1624 memset(&c, 0, sizeof(c));
1625 c.common.opcode = cmd.opcode;
1626 c.common.flags = cmd.flags;
1627 c.common.nsid = cpu_to_le32(cmd.nsid);
1628 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1629 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1630 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1631 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1632 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1633 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1634 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1635 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1636
1637 if (cmd.timeout_ms)
1638 timeout = msecs_to_jiffies(cmd.timeout_ms);
1639
1640 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1641 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1642 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1643 0, &result, timeout);
1644
1645 if (status >= 0) {
1646 if (put_user(result, &ucmd->result))
1647 return -EFAULT;
1648 }
1649
1650 return status;
1651 }
1652
1653 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1654 struct nvme_passthru_cmd64 __user *ucmd)
1655 {
1656 struct nvme_passthru_cmd64 cmd;
1657 struct nvme_command c;
1658 unsigned timeout = 0;
1659 int status;
1660
1661 if (!capable(CAP_SYS_ADMIN))
1662 return -EACCES;
1663 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1664 return -EFAULT;
1665 if (cmd.flags)
1666 return -EINVAL;
1667
1668 memset(&c, 0, sizeof(c));
1669 c.common.opcode = cmd.opcode;
1670 c.common.flags = cmd.flags;
1671 c.common.nsid = cpu_to_le32(cmd.nsid);
1672 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1673 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1674 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1675 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1676 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1677 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1678 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1679 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1680
1681 if (cmd.timeout_ms)
1682 timeout = msecs_to_jiffies(cmd.timeout_ms);
1683
1684 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1685 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1686 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1687 0, &cmd.result, timeout);
1688
1689 if (status >= 0) {
1690 if (put_user(cmd.result, &ucmd->result))
1691 return -EFAULT;
1692 }
1693
1694 return status;
1695 }
1696
1697 /*
1698 * Issue ioctl requests on the first available path. Note that unlike normal
1699 * block layer requests we will not retry failed request on another controller.
1700 */
1701 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1702 struct nvme_ns_head **head, int *srcu_idx)
1703 {
1704 #ifdef CONFIG_NVME_MULTIPATH
1705 if (disk->fops == &nvme_ns_head_ops) {
1706 struct nvme_ns *ns;
1707
1708 *head = disk->private_data;
1709 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1710 ns = nvme_find_path(*head);
1711 if (!ns)
1712 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1713 return ns;
1714 }
1715 #endif
1716 *head = NULL;
1717 *srcu_idx = -1;
1718 return disk->private_data;
1719 }
1720
1721 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1722 {
1723 if (head)
1724 srcu_read_unlock(&head->srcu, idx);
1725 }
1726
1727 static bool is_ctrl_ioctl(unsigned int cmd)
1728 {
1729 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1730 return true;
1731 if (is_sed_ioctl(cmd))
1732 return true;
1733 return false;
1734 }
1735
1736 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1737 void __user *argp,
1738 struct nvme_ns_head *head,
1739 int srcu_idx)
1740 {
1741 struct nvme_ctrl *ctrl = ns->ctrl;
1742 int ret;
1743
1744 nvme_get_ctrl(ns->ctrl);
1745 nvme_put_ns_from_disk(head, srcu_idx);
1746
1747 switch (cmd) {
1748 case NVME_IOCTL_ADMIN_CMD:
1749 ret = nvme_user_cmd(ctrl, NULL, argp);
1750 break;
1751 case NVME_IOCTL_ADMIN64_CMD:
1752 ret = nvme_user_cmd64(ctrl, NULL, argp);
1753 break;
1754 default:
1755 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1756 break;
1757 }
1758 nvme_put_ctrl(ctrl);
1759 return ret;
1760 }
1761
1762 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1763 unsigned int cmd, unsigned long arg)
1764 {
1765 struct nvme_ns_head *head = NULL;
1766 void __user *argp = (void __user *)arg;
1767 struct nvme_ns *ns;
1768 int srcu_idx, ret;
1769
1770 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1771 if (unlikely(!ns))
1772 return -EWOULDBLOCK;
1773
1774 /*
1775 * Handle ioctls that apply to the controller instead of the namespace
1776 * seperately and drop the ns SRCU reference early. This avoids a
1777 * deadlock when deleting namespaces using the passthrough interface.
1778 */
1779 if (is_ctrl_ioctl(cmd))
1780 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1781
1782 switch (cmd) {
1783 case NVME_IOCTL_ID:
1784 force_successful_syscall_return();
1785 ret = ns->head->ns_id;
1786 break;
1787 case NVME_IOCTL_IO_CMD:
1788 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1789 break;
1790 case NVME_IOCTL_SUBMIT_IO:
1791 ret = nvme_submit_io(ns, argp);
1792 break;
1793 case NVME_IOCTL_IO64_CMD:
1794 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1795 break;
1796 default:
1797 if (ns->ndev)
1798 ret = nvme_nvm_ioctl(ns, cmd, arg);
1799 else
1800 ret = -ENOTTY;
1801 }
1802
1803 nvme_put_ns_from_disk(head, srcu_idx);
1804 return ret;
1805 }
1806
1807 #ifdef CONFIG_COMPAT
1808 struct nvme_user_io32 {
1809 __u8 opcode;
1810 __u8 flags;
1811 __u16 control;
1812 __u16 nblocks;
1813 __u16 rsvd;
1814 __u64 metadata;
1815 __u64 addr;
1816 __u64 slba;
1817 __u32 dsmgmt;
1818 __u32 reftag;
1819 __u16 apptag;
1820 __u16 appmask;
1821 } __attribute__((__packed__));
1822
1823 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1824
1825 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1826 unsigned int cmd, unsigned long arg)
1827 {
1828 /*
1829 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1830 * between 32 bit programs and 64 bit kernel.
1831 * The cause is that the results of sizeof(struct nvme_user_io),
1832 * which is used to define NVME_IOCTL_SUBMIT_IO,
1833 * are not same between 32 bit compiler and 64 bit compiler.
1834 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1835 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1836 * Other IOCTL numbers are same between 32 bit and 64 bit.
1837 * So there is nothing to do regarding to other IOCTL numbers.
1838 */
1839 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1840 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1841
1842 return nvme_ioctl(bdev, mode, cmd, arg);
1843 }
1844 #else
1845 #define nvme_compat_ioctl NULL
1846 #endif /* CONFIG_COMPAT */
1847
1848 static int nvme_open(struct block_device *bdev, fmode_t mode)
1849 {
1850 struct nvme_ns *ns = bdev->bd_disk->private_data;
1851
1852 #ifdef CONFIG_NVME_MULTIPATH
1853 /* should never be called due to GENHD_FL_HIDDEN */
1854 if (WARN_ON_ONCE(ns->head->disk))
1855 goto fail;
1856 #endif
1857 if (!kref_get_unless_zero(&ns->kref))
1858 goto fail;
1859 if (!try_module_get(ns->ctrl->ops->module))
1860 goto fail_put_ns;
1861
1862 return 0;
1863
1864 fail_put_ns:
1865 nvme_put_ns(ns);
1866 fail:
1867 return -ENXIO;
1868 }
1869
1870 static void nvme_release(struct gendisk *disk, fmode_t mode)
1871 {
1872 struct nvme_ns *ns = disk->private_data;
1873
1874 module_put(ns->ctrl->ops->module);
1875 nvme_put_ns(ns);
1876 }
1877
1878 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1879 {
1880 /* some standard values */
1881 geo->heads = 1 << 6;
1882 geo->sectors = 1 << 5;
1883 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1884 return 0;
1885 }
1886
1887 #ifdef CONFIG_BLK_DEV_INTEGRITY
1888 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1889 u32 max_integrity_segments)
1890 {
1891 struct blk_integrity integrity;
1892
1893 memset(&integrity, 0, sizeof(integrity));
1894 switch (pi_type) {
1895 case NVME_NS_DPS_PI_TYPE3:
1896 integrity.profile = &t10_pi_type3_crc;
1897 integrity.tag_size = sizeof(u16) + sizeof(u32);
1898 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1899 break;
1900 case NVME_NS_DPS_PI_TYPE1:
1901 case NVME_NS_DPS_PI_TYPE2:
1902 integrity.profile = &t10_pi_type1_crc;
1903 integrity.tag_size = sizeof(u16);
1904 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1905 break;
1906 default:
1907 integrity.profile = NULL;
1908 break;
1909 }
1910 integrity.tuple_size = ms;
1911 blk_integrity_register(disk, &integrity);
1912 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1913 }
1914 #else
1915 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1916 u32 max_integrity_segments)
1917 {
1918 }
1919 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1920
1921 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1922 {
1923 struct nvme_ctrl *ctrl = ns->ctrl;
1924 struct request_queue *queue = disk->queue;
1925 u32 size = queue_logical_block_size(queue);
1926
1927 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1928 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1929 return;
1930 }
1931
1932 if (ctrl->nr_streams && ns->sws && ns->sgs)
1933 size *= ns->sws * ns->sgs;
1934
1935 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1936 NVME_DSM_MAX_RANGES);
1937
1938 queue->limits.discard_alignment = 0;
1939 queue->limits.discard_granularity = size;
1940
1941 /* If discard is already enabled, don't reset queue limits */
1942 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1943 return;
1944
1945 blk_queue_max_discard_sectors(queue, UINT_MAX);
1946 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1947
1948 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1949 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1950 }
1951
1952 /*
1953 * Even though NVMe spec explicitly states that MDTS is not applicable to the
1954 * write-zeroes, we are cautious and limit the size to the controllers
1955 * max_hw_sectors value, which is based on the MDTS field and possibly other
1956 * limiting factors.
1957 */
1958 static void nvme_config_write_zeroes(struct request_queue *q,
1959 struct nvme_ctrl *ctrl)
1960 {
1961 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
1962 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1963 blk_queue_max_write_zeroes_sectors(q, ctrl->max_hw_sectors);
1964 }
1965
1966 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1967 {
1968 return !uuid_is_null(&ids->uuid) ||
1969 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1970 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1971 }
1972
1973 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1974 {
1975 return uuid_equal(&a->uuid, &b->uuid) &&
1976 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1977 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1978 a->csi == b->csi;
1979 }
1980
1981 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1982 u32 *phys_bs, u32 *io_opt)
1983 {
1984 struct streams_directive_params s;
1985 int ret;
1986
1987 if (!ctrl->nr_streams)
1988 return 0;
1989
1990 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1991 if (ret)
1992 return ret;
1993
1994 ns->sws = le32_to_cpu(s.sws);
1995 ns->sgs = le16_to_cpu(s.sgs);
1996
1997 if (ns->sws) {
1998 *phys_bs = ns->sws * (1 << ns->lba_shift);
1999 if (ns->sgs)
2000 *io_opt = *phys_bs * ns->sgs;
2001 }
2002
2003 return 0;
2004 }
2005
2006 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2007 {
2008 struct nvme_ctrl *ctrl = ns->ctrl;
2009
2010 /*
2011 * The PI implementation requires the metadata size to be equal to the
2012 * t10 pi tuple size.
2013 */
2014 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2015 if (ns->ms == sizeof(struct t10_pi_tuple))
2016 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2017 else
2018 ns->pi_type = 0;
2019
2020 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2021 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2022 return 0;
2023 if (ctrl->ops->flags & NVME_F_FABRICS) {
2024 /*
2025 * The NVMe over Fabrics specification only supports metadata as
2026 * part of the extended data LBA. We rely on HCA/HBA support to
2027 * remap the separate metadata buffer from the block layer.
2028 */
2029 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2030 return -EINVAL;
2031 if (ctrl->max_integrity_segments)
2032 ns->features |=
2033 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2034 } else {
2035 /*
2036 * For PCIe controllers, we can't easily remap the separate
2037 * metadata buffer from the block layer and thus require a
2038 * separate metadata buffer for block layer metadata/PI support.
2039 * We allow extended LBAs for the passthrough interface, though.
2040 */
2041 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2042 ns->features |= NVME_NS_EXT_LBAS;
2043 else
2044 ns->features |= NVME_NS_METADATA_SUPPORTED;
2045 }
2046
2047 return 0;
2048 }
2049
2050 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2051 struct request_queue *q)
2052 {
2053 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2054
2055 if (ctrl->max_hw_sectors) {
2056 u32 max_segments =
2057 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2058
2059 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2060 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2061 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2062 }
2063 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2064 blk_queue_dma_alignment(q, 7);
2065 blk_queue_write_cache(q, vwc, vwc);
2066 }
2067
2068 static void nvme_update_disk_info(struct gendisk *disk,
2069 struct nvme_ns *ns, struct nvme_id_ns *id)
2070 {
2071 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2072 unsigned short bs = 1 << ns->lba_shift;
2073 u32 atomic_bs, phys_bs, io_opt = 0;
2074
2075 /*
2076 * The block layer can't support LBA sizes larger than the page size
2077 * yet, so catch this early and don't allow block I/O.
2078 */
2079 if (ns->lba_shift > PAGE_SHIFT) {
2080 capacity = 0;
2081 bs = (1 << 9);
2082 }
2083
2084 blk_integrity_unregister(disk);
2085
2086 atomic_bs = phys_bs = bs;
2087 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2088 if (id->nabo == 0) {
2089 /*
2090 * Bit 1 indicates whether NAWUPF is defined for this namespace
2091 * and whether it should be used instead of AWUPF. If NAWUPF ==
2092 * 0 then AWUPF must be used instead.
2093 */
2094 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2095 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2096 else
2097 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2098 }
2099
2100 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2101 /* NPWG = Namespace Preferred Write Granularity */
2102 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2103 /* NOWS = Namespace Optimal Write Size */
2104 io_opt = bs * (1 + le16_to_cpu(id->nows));
2105 }
2106
2107 blk_queue_logical_block_size(disk->queue, bs);
2108 /*
2109 * Linux filesystems assume writing a single physical block is
2110 * an atomic operation. Hence limit the physical block size to the
2111 * value of the Atomic Write Unit Power Fail parameter.
2112 */
2113 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2114 blk_queue_io_min(disk->queue, phys_bs);
2115 blk_queue_io_opt(disk->queue, io_opt);
2116
2117 /*
2118 * Register a metadata profile for PI, or the plain non-integrity NVMe
2119 * metadata masquerading as Type 0 if supported, otherwise reject block
2120 * I/O to namespaces with metadata except when the namespace supports
2121 * PI, as it can strip/insert in that case.
2122 */
2123 if (ns->ms) {
2124 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2125 (ns->features & NVME_NS_METADATA_SUPPORTED))
2126 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2127 ns->ctrl->max_integrity_segments);
2128 else if (!nvme_ns_has_pi(ns))
2129 capacity = 0;
2130 }
2131
2132 set_capacity_and_notify(disk, capacity);
2133
2134 nvme_config_discard(disk, ns);
2135 nvme_config_write_zeroes(disk->queue, ns->ctrl);
2136
2137 if ((id->nsattr & NVME_NS_ATTR_RO) ||
2138 test_bit(NVME_NS_FORCE_RO, &ns->flags))
2139 set_disk_ro(disk, true);
2140 }
2141
2142 static inline bool nvme_first_scan(struct gendisk *disk)
2143 {
2144 /* nvme_alloc_ns() scans the disk prior to adding it */
2145 return !(disk->flags & GENHD_FL_UP);
2146 }
2147
2148 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2149 {
2150 struct nvme_ctrl *ctrl = ns->ctrl;
2151 u32 iob;
2152
2153 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2154 is_power_of_2(ctrl->max_hw_sectors))
2155 iob = ctrl->max_hw_sectors;
2156 else
2157 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2158
2159 if (!iob)
2160 return;
2161
2162 if (!is_power_of_2(iob)) {
2163 if (nvme_first_scan(ns->disk))
2164 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2165 ns->disk->disk_name, iob);
2166 return;
2167 }
2168
2169 if (blk_queue_is_zoned(ns->disk->queue)) {
2170 if (nvme_first_scan(ns->disk))
2171 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2172 ns->disk->disk_name);
2173 return;
2174 }
2175
2176 blk_queue_chunk_sectors(ns->queue, iob);
2177 }
2178
2179 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2180 {
2181 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2182 int ret;
2183
2184 blk_mq_freeze_queue(ns->disk->queue);
2185 ns->lba_shift = id->lbaf[lbaf].ds;
2186 nvme_set_queue_limits(ns->ctrl, ns->queue);
2187
2188 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2189 ret = nvme_update_zone_info(ns, lbaf);
2190 if (ret)
2191 goto out_unfreeze;
2192 }
2193
2194 ret = nvme_configure_metadata(ns, id);
2195 if (ret)
2196 goto out_unfreeze;
2197 nvme_set_chunk_sectors(ns, id);
2198 nvme_update_disk_info(ns->disk, ns, id);
2199 blk_mq_unfreeze_queue(ns->disk->queue);
2200
2201 if (blk_queue_is_zoned(ns->queue)) {
2202 ret = nvme_revalidate_zones(ns);
2203 if (ret && !nvme_first_scan(ns->disk))
2204 return ret;
2205 }
2206
2207 #ifdef CONFIG_NVME_MULTIPATH
2208 if (ns->head->disk) {
2209 blk_mq_freeze_queue(ns->head->disk->queue);
2210 nvme_update_disk_info(ns->head->disk, ns, id);
2211 blk_stack_limits(&ns->head->disk->queue->limits,
2212 &ns->queue->limits, 0);
2213 blk_queue_update_readahead(ns->head->disk->queue);
2214 blk_mq_unfreeze_queue(ns->head->disk->queue);
2215 }
2216 #endif
2217 return 0;
2218
2219 out_unfreeze:
2220 blk_mq_unfreeze_queue(ns->disk->queue);
2221 return ret;
2222 }
2223
2224 static char nvme_pr_type(enum pr_type type)
2225 {
2226 switch (type) {
2227 case PR_WRITE_EXCLUSIVE:
2228 return 1;
2229 case PR_EXCLUSIVE_ACCESS:
2230 return 2;
2231 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2232 return 3;
2233 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2234 return 4;
2235 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2236 return 5;
2237 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2238 return 6;
2239 default:
2240 return 0;
2241 }
2242 };
2243
2244 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2245 u64 key, u64 sa_key, u8 op)
2246 {
2247 struct nvme_ns_head *head = NULL;
2248 struct nvme_ns *ns;
2249 struct nvme_command c;
2250 int srcu_idx, ret;
2251 u8 data[16] = { 0, };
2252
2253 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2254 if (unlikely(!ns))
2255 return -EWOULDBLOCK;
2256
2257 put_unaligned_le64(key, &data[0]);
2258 put_unaligned_le64(sa_key, &data[8]);
2259
2260 memset(&c, 0, sizeof(c));
2261 c.common.opcode = op;
2262 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2263 c.common.cdw10 = cpu_to_le32(cdw10);
2264
2265 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2266 nvme_put_ns_from_disk(head, srcu_idx);
2267 return ret;
2268 }
2269
2270 static int nvme_pr_register(struct block_device *bdev, u64 old,
2271 u64 new, unsigned flags)
2272 {
2273 u32 cdw10;
2274
2275 if (flags & ~PR_FL_IGNORE_KEY)
2276 return -EOPNOTSUPP;
2277
2278 cdw10 = old ? 2 : 0;
2279 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2280 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2281 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2282 }
2283
2284 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2285 enum pr_type type, unsigned flags)
2286 {
2287 u32 cdw10;
2288
2289 if (flags & ~PR_FL_IGNORE_KEY)
2290 return -EOPNOTSUPP;
2291
2292 cdw10 = nvme_pr_type(type) << 8;
2293 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2294 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2295 }
2296
2297 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2298 enum pr_type type, bool abort)
2299 {
2300 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2301 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2302 }
2303
2304 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2305 {
2306 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2307 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2308 }
2309
2310 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2311 {
2312 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2313 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2314 }
2315
2316 static const struct pr_ops nvme_pr_ops = {
2317 .pr_register = nvme_pr_register,
2318 .pr_reserve = nvme_pr_reserve,
2319 .pr_release = nvme_pr_release,
2320 .pr_preempt = nvme_pr_preempt,
2321 .pr_clear = nvme_pr_clear,
2322 };
2323
2324 #ifdef CONFIG_BLK_SED_OPAL
2325 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2326 bool send)
2327 {
2328 struct nvme_ctrl *ctrl = data;
2329 struct nvme_command cmd;
2330
2331 memset(&cmd, 0, sizeof(cmd));
2332 if (send)
2333 cmd.common.opcode = nvme_admin_security_send;
2334 else
2335 cmd.common.opcode = nvme_admin_security_recv;
2336 cmd.common.nsid = 0;
2337 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2338 cmd.common.cdw11 = cpu_to_le32(len);
2339
2340 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2341 NVME_QID_ANY, 1, 0, false);
2342 }
2343 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2344 #endif /* CONFIG_BLK_SED_OPAL */
2345
2346 static const struct block_device_operations nvme_bdev_ops = {
2347 .owner = THIS_MODULE,
2348 .ioctl = nvme_ioctl,
2349 .compat_ioctl = nvme_compat_ioctl,
2350 .open = nvme_open,
2351 .release = nvme_release,
2352 .getgeo = nvme_getgeo,
2353 .report_zones = nvme_report_zones,
2354 .pr_ops = &nvme_pr_ops,
2355 };
2356
2357 #ifdef CONFIG_NVME_MULTIPATH
2358 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2359 {
2360 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2361
2362 if (!kref_get_unless_zero(&head->ref))
2363 return -ENXIO;
2364 return 0;
2365 }
2366
2367 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2368 {
2369 nvme_put_ns_head(disk->private_data);
2370 }
2371
2372 const struct block_device_operations nvme_ns_head_ops = {
2373 .owner = THIS_MODULE,
2374 .submit_bio = nvme_ns_head_submit_bio,
2375 .open = nvme_ns_head_open,
2376 .release = nvme_ns_head_release,
2377 .ioctl = nvme_ioctl,
2378 .compat_ioctl = nvme_compat_ioctl,
2379 .getgeo = nvme_getgeo,
2380 .report_zones = nvme_report_zones,
2381 .pr_ops = &nvme_pr_ops,
2382 };
2383 #endif /* CONFIG_NVME_MULTIPATH */
2384
2385 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2386 {
2387 unsigned long timeout =
2388 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2389 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2390 int ret;
2391
2392 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2393 if (csts == ~0)
2394 return -ENODEV;
2395 if ((csts & NVME_CSTS_RDY) == bit)
2396 break;
2397
2398 usleep_range(1000, 2000);
2399 if (fatal_signal_pending(current))
2400 return -EINTR;
2401 if (time_after(jiffies, timeout)) {
2402 dev_err(ctrl->device,
2403 "Device not ready; aborting %s, CSTS=0x%x\n",
2404 enabled ? "initialisation" : "reset", csts);
2405 return -ENODEV;
2406 }
2407 }
2408
2409 return ret;
2410 }
2411
2412 /*
2413 * If the device has been passed off to us in an enabled state, just clear
2414 * the enabled bit. The spec says we should set the 'shutdown notification
2415 * bits', but doing so may cause the device to complete commands to the
2416 * admin queue ... and we don't know what memory that might be pointing at!
2417 */
2418 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2419 {
2420 int ret;
2421
2422 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2423 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2424
2425 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2426 if (ret)
2427 return ret;
2428
2429 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2430 msleep(NVME_QUIRK_DELAY_AMOUNT);
2431
2432 return nvme_wait_ready(ctrl, ctrl->cap, false);
2433 }
2434 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2435
2436 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2437 {
2438 unsigned dev_page_min;
2439 int ret;
2440
2441 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2442 if (ret) {
2443 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2444 return ret;
2445 }
2446 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2447
2448 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2449 dev_err(ctrl->device,
2450 "Minimum device page size %u too large for host (%u)\n",
2451 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2452 return -ENODEV;
2453 }
2454
2455 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2456 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2457 else
2458 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2459 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2460 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2461 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2462 ctrl->ctrl_config |= NVME_CC_ENABLE;
2463
2464 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2465 if (ret)
2466 return ret;
2467 return nvme_wait_ready(ctrl, ctrl->cap, true);
2468 }
2469 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2470
2471 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2472 {
2473 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2474 u32 csts;
2475 int ret;
2476
2477 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2478 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2479
2480 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2481 if (ret)
2482 return ret;
2483
2484 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2485 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2486 break;
2487
2488 msleep(100);
2489 if (fatal_signal_pending(current))
2490 return -EINTR;
2491 if (time_after(jiffies, timeout)) {
2492 dev_err(ctrl->device,
2493 "Device shutdown incomplete; abort shutdown\n");
2494 return -ENODEV;
2495 }
2496 }
2497
2498 return ret;
2499 }
2500 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2501
2502 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2503 {
2504 __le64 ts;
2505 int ret;
2506
2507 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2508 return 0;
2509
2510 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2511 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2512 NULL);
2513 if (ret)
2514 dev_warn_once(ctrl->device,
2515 "could not set timestamp (%d)\n", ret);
2516 return ret;
2517 }
2518
2519 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2520 {
2521 struct nvme_feat_host_behavior *host;
2522 int ret;
2523
2524 /* Don't bother enabling the feature if retry delay is not reported */
2525 if (!ctrl->crdt[0])
2526 return 0;
2527
2528 host = kzalloc(sizeof(*host), GFP_KERNEL);
2529 if (!host)
2530 return 0;
2531
2532 host->acre = NVME_ENABLE_ACRE;
2533 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2534 host, sizeof(*host), NULL);
2535 kfree(host);
2536 return ret;
2537 }
2538
2539 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2540 {
2541 /*
2542 * APST (Autonomous Power State Transition) lets us program a
2543 * table of power state transitions that the controller will
2544 * perform automatically. We configure it with a simple
2545 * heuristic: we are willing to spend at most 2% of the time
2546 * transitioning between power states. Therefore, when running
2547 * in any given state, we will enter the next lower-power
2548 * non-operational state after waiting 50 * (enlat + exlat)
2549 * microseconds, as long as that state's exit latency is under
2550 * the requested maximum latency.
2551 *
2552 * We will not autonomously enter any non-operational state for
2553 * which the total latency exceeds ps_max_latency_us. Users
2554 * can set ps_max_latency_us to zero to turn off APST.
2555 */
2556
2557 unsigned apste;
2558 struct nvme_feat_auto_pst *table;
2559 u64 max_lat_us = 0;
2560 int max_ps = -1;
2561 int ret;
2562
2563 /*
2564 * If APST isn't supported or if we haven't been initialized yet,
2565 * then don't do anything.
2566 */
2567 if (!ctrl->apsta)
2568 return 0;
2569
2570 if (ctrl->npss > 31) {
2571 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2572 return 0;
2573 }
2574
2575 table = kzalloc(sizeof(*table), GFP_KERNEL);
2576 if (!table)
2577 return 0;
2578
2579 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2580 /* Turn off APST. */
2581 apste = 0;
2582 dev_dbg(ctrl->device, "APST disabled\n");
2583 } else {
2584 __le64 target = cpu_to_le64(0);
2585 int state;
2586
2587 /*
2588 * Walk through all states from lowest- to highest-power.
2589 * According to the spec, lower-numbered states use more
2590 * power. NPSS, despite the name, is the index of the
2591 * lowest-power state, not the number of states.
2592 */
2593 for (state = (int)ctrl->npss; state >= 0; state--) {
2594 u64 total_latency_us, exit_latency_us, transition_ms;
2595
2596 if (target)
2597 table->entries[state] = target;
2598
2599 /*
2600 * Don't allow transitions to the deepest state
2601 * if it's quirked off.
2602 */
2603 if (state == ctrl->npss &&
2604 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2605 continue;
2606
2607 /*
2608 * Is this state a useful non-operational state for
2609 * higher-power states to autonomously transition to?
2610 */
2611 if (!(ctrl->psd[state].flags &
2612 NVME_PS_FLAGS_NON_OP_STATE))
2613 continue;
2614
2615 exit_latency_us =
2616 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2617 if (exit_latency_us > ctrl->ps_max_latency_us)
2618 continue;
2619
2620 total_latency_us =
2621 exit_latency_us +
2622 le32_to_cpu(ctrl->psd[state].entry_lat);
2623
2624 /*
2625 * This state is good. Use it as the APST idle
2626 * target for higher power states.
2627 */
2628 transition_ms = total_latency_us + 19;
2629 do_div(transition_ms, 20);
2630 if (transition_ms > (1 << 24) - 1)
2631 transition_ms = (1 << 24) - 1;
2632
2633 target = cpu_to_le64((state << 3) |
2634 (transition_ms << 8));
2635
2636 if (max_ps == -1)
2637 max_ps = state;
2638
2639 if (total_latency_us > max_lat_us)
2640 max_lat_us = total_latency_us;
2641 }
2642
2643 apste = 1;
2644
2645 if (max_ps == -1) {
2646 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2647 } else {
2648 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2649 max_ps, max_lat_us, (int)sizeof(*table), table);
2650 }
2651 }
2652
2653 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2654 table, sizeof(*table), NULL);
2655 if (ret)
2656 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2657
2658 kfree(table);
2659 return ret;
2660 }
2661
2662 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2663 {
2664 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2665 u64 latency;
2666
2667 switch (val) {
2668 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2669 case PM_QOS_LATENCY_ANY:
2670 latency = U64_MAX;
2671 break;
2672
2673 default:
2674 latency = val;
2675 }
2676
2677 if (ctrl->ps_max_latency_us != latency) {
2678 ctrl->ps_max_latency_us = latency;
2679 nvme_configure_apst(ctrl);
2680 }
2681 }
2682
2683 struct nvme_core_quirk_entry {
2684 /*
2685 * NVMe model and firmware strings are padded with spaces. For
2686 * simplicity, strings in the quirk table are padded with NULLs
2687 * instead.
2688 */
2689 u16 vid;
2690 const char *mn;
2691 const char *fr;
2692 unsigned long quirks;
2693 };
2694
2695 static const struct nvme_core_quirk_entry core_quirks[] = {
2696 {
2697 /*
2698 * This Toshiba device seems to die using any APST states. See:
2699 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2700 */
2701 .vid = 0x1179,
2702 .mn = "THNSF5256GPUK TOSHIBA",
2703 .quirks = NVME_QUIRK_NO_APST,
2704 },
2705 {
2706 /*
2707 * This LiteON CL1-3D*-Q11 firmware version has a race
2708 * condition associated with actions related to suspend to idle
2709 * LiteON has resolved the problem in future firmware
2710 */
2711 .vid = 0x14a4,
2712 .fr = "22301111",
2713 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2714 }
2715 };
2716
2717 /* match is null-terminated but idstr is space-padded. */
2718 static bool string_matches(const char *idstr, const char *match, size_t len)
2719 {
2720 size_t matchlen;
2721
2722 if (!match)
2723 return true;
2724
2725 matchlen = strlen(match);
2726 WARN_ON_ONCE(matchlen > len);
2727
2728 if (memcmp(idstr, match, matchlen))
2729 return false;
2730
2731 for (; matchlen < len; matchlen++)
2732 if (idstr[matchlen] != ' ')
2733 return false;
2734
2735 return true;
2736 }
2737
2738 static bool quirk_matches(const struct nvme_id_ctrl *id,
2739 const struct nvme_core_quirk_entry *q)
2740 {
2741 return q->vid == le16_to_cpu(id->vid) &&
2742 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2743 string_matches(id->fr, q->fr, sizeof(id->fr));
2744 }
2745
2746 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2747 struct nvme_id_ctrl *id)
2748 {
2749 size_t nqnlen;
2750 int off;
2751
2752 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2753 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2754 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2755 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2756 return;
2757 }
2758
2759 if (ctrl->vs >= NVME_VS(1, 2, 1))
2760 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2761 }
2762
2763 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2764 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2765 "nqn.2014.08.org.nvmexpress:%04x%04x",
2766 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2767 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2768 off += sizeof(id->sn);
2769 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2770 off += sizeof(id->mn);
2771 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2772 }
2773
2774 static void nvme_release_subsystem(struct device *dev)
2775 {
2776 struct nvme_subsystem *subsys =
2777 container_of(dev, struct nvme_subsystem, dev);
2778
2779 if (subsys->instance >= 0)
2780 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2781 kfree(subsys);
2782 }
2783
2784 static void nvme_destroy_subsystem(struct kref *ref)
2785 {
2786 struct nvme_subsystem *subsys =
2787 container_of(ref, struct nvme_subsystem, ref);
2788
2789 mutex_lock(&nvme_subsystems_lock);
2790 list_del(&subsys->entry);
2791 mutex_unlock(&nvme_subsystems_lock);
2792
2793 ida_destroy(&subsys->ns_ida);
2794 device_del(&subsys->dev);
2795 put_device(&subsys->dev);
2796 }
2797
2798 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2799 {
2800 kref_put(&subsys->ref, nvme_destroy_subsystem);
2801 }
2802
2803 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2804 {
2805 struct nvme_subsystem *subsys;
2806
2807 lockdep_assert_held(&nvme_subsystems_lock);
2808
2809 /*
2810 * Fail matches for discovery subsystems. This results
2811 * in each discovery controller bound to a unique subsystem.
2812 * This avoids issues with validating controller values
2813 * that can only be true when there is a single unique subsystem.
2814 * There may be multiple and completely independent entities
2815 * that provide discovery controllers.
2816 */
2817 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2818 return NULL;
2819
2820 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2821 if (strcmp(subsys->subnqn, subsysnqn))
2822 continue;
2823 if (!kref_get_unless_zero(&subsys->ref))
2824 continue;
2825 return subsys;
2826 }
2827
2828 return NULL;
2829 }
2830
2831 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2832 struct device_attribute subsys_attr_##_name = \
2833 __ATTR(_name, _mode, _show, NULL)
2834
2835 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2836 struct device_attribute *attr,
2837 char *buf)
2838 {
2839 struct nvme_subsystem *subsys =
2840 container_of(dev, struct nvme_subsystem, dev);
2841
2842 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2843 }
2844 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2845
2846 #define nvme_subsys_show_str_function(field) \
2847 static ssize_t subsys_##field##_show(struct device *dev, \
2848 struct device_attribute *attr, char *buf) \
2849 { \
2850 struct nvme_subsystem *subsys = \
2851 container_of(dev, struct nvme_subsystem, dev); \
2852 return sprintf(buf, "%.*s\n", \
2853 (int)sizeof(subsys->field), subsys->field); \
2854 } \
2855 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2856
2857 nvme_subsys_show_str_function(model);
2858 nvme_subsys_show_str_function(serial);
2859 nvme_subsys_show_str_function(firmware_rev);
2860
2861 static struct attribute *nvme_subsys_attrs[] = {
2862 &subsys_attr_model.attr,
2863 &subsys_attr_serial.attr,
2864 &subsys_attr_firmware_rev.attr,
2865 &subsys_attr_subsysnqn.attr,
2866 #ifdef CONFIG_NVME_MULTIPATH
2867 &subsys_attr_iopolicy.attr,
2868 #endif
2869 NULL,
2870 };
2871
2872 static struct attribute_group nvme_subsys_attrs_group = {
2873 .attrs = nvme_subsys_attrs,
2874 };
2875
2876 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2877 &nvme_subsys_attrs_group,
2878 NULL,
2879 };
2880
2881 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2882 {
2883 return ctrl->opts && ctrl->opts->discovery_nqn;
2884 }
2885
2886 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2887 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2888 {
2889 struct nvme_ctrl *tmp;
2890
2891 lockdep_assert_held(&nvme_subsystems_lock);
2892
2893 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2894 if (nvme_state_terminal(tmp))
2895 continue;
2896
2897 if (tmp->cntlid == ctrl->cntlid) {
2898 dev_err(ctrl->device,
2899 "Duplicate cntlid %u with %s, rejecting\n",
2900 ctrl->cntlid, dev_name(tmp->device));
2901 return false;
2902 }
2903
2904 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2905 nvme_discovery_ctrl(ctrl))
2906 continue;
2907
2908 dev_err(ctrl->device,
2909 "Subsystem does not support multiple controllers\n");
2910 return false;
2911 }
2912
2913 return true;
2914 }
2915
2916 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2917 {
2918 struct nvme_subsystem *subsys, *found;
2919 int ret;
2920
2921 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2922 if (!subsys)
2923 return -ENOMEM;
2924
2925 subsys->instance = -1;
2926 mutex_init(&subsys->lock);
2927 kref_init(&subsys->ref);
2928 INIT_LIST_HEAD(&subsys->ctrls);
2929 INIT_LIST_HEAD(&subsys->nsheads);
2930 nvme_init_subnqn(subsys, ctrl, id);
2931 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2932 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2933 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2934 subsys->vendor_id = le16_to_cpu(id->vid);
2935 subsys->cmic = id->cmic;
2936 subsys->awupf = le16_to_cpu(id->awupf);
2937 #ifdef CONFIG_NVME_MULTIPATH
2938 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2939 #endif
2940
2941 subsys->dev.class = nvme_subsys_class;
2942 subsys->dev.release = nvme_release_subsystem;
2943 subsys->dev.groups = nvme_subsys_attrs_groups;
2944 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2945 device_initialize(&subsys->dev);
2946
2947 mutex_lock(&nvme_subsystems_lock);
2948 found = __nvme_find_get_subsystem(subsys->subnqn);
2949 if (found) {
2950 put_device(&subsys->dev);
2951 subsys = found;
2952
2953 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2954 ret = -EINVAL;
2955 goto out_put_subsystem;
2956 }
2957 } else {
2958 ret = device_add(&subsys->dev);
2959 if (ret) {
2960 dev_err(ctrl->device,
2961 "failed to register subsystem device.\n");
2962 put_device(&subsys->dev);
2963 goto out_unlock;
2964 }
2965 ida_init(&subsys->ns_ida);
2966 list_add_tail(&subsys->entry, &nvme_subsystems);
2967 }
2968
2969 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2970 dev_name(ctrl->device));
2971 if (ret) {
2972 dev_err(ctrl->device,
2973 "failed to create sysfs link from subsystem.\n");
2974 goto out_put_subsystem;
2975 }
2976
2977 if (!found)
2978 subsys->instance = ctrl->instance;
2979 ctrl->subsys = subsys;
2980 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2981 mutex_unlock(&nvme_subsystems_lock);
2982 return 0;
2983
2984 out_put_subsystem:
2985 nvme_put_subsystem(subsys);
2986 out_unlock:
2987 mutex_unlock(&nvme_subsystems_lock);
2988 return ret;
2989 }
2990
2991 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2992 void *log, size_t size, u64 offset)
2993 {
2994 struct nvme_command c = { };
2995 u32 dwlen = nvme_bytes_to_numd(size);
2996
2997 c.get_log_page.opcode = nvme_admin_get_log_page;
2998 c.get_log_page.nsid = cpu_to_le32(nsid);
2999 c.get_log_page.lid = log_page;
3000 c.get_log_page.lsp = lsp;
3001 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3002 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3003 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3004 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3005 c.get_log_page.csi = csi;
3006
3007 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3008 }
3009
3010 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3011 struct nvme_effects_log **log)
3012 {
3013 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3014 int ret;
3015
3016 if (cel)
3017 goto out;
3018
3019 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3020 if (!cel)
3021 return -ENOMEM;
3022
3023 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3024 cel, sizeof(*cel), 0);
3025 if (ret) {
3026 kfree(cel);
3027 return ret;
3028 }
3029
3030 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3031 out:
3032 *log = cel;
3033 return 0;
3034 }
3035
3036 /*
3037 * Initialize the cached copies of the Identify data and various controller
3038 * register in our nvme_ctrl structure. This should be called as soon as
3039 * the admin queue is fully up and running.
3040 */
3041 int nvme_init_identify(struct nvme_ctrl *ctrl)
3042 {
3043 struct nvme_id_ctrl *id;
3044 int ret, page_shift;
3045 u32 max_hw_sectors;
3046 bool prev_apst_enabled;
3047
3048 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3049 if (ret) {
3050 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3051 return ret;
3052 }
3053 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3054 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3055
3056 if (ctrl->vs >= NVME_VS(1, 1, 0))
3057 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3058
3059 ret = nvme_identify_ctrl(ctrl, &id);
3060 if (ret) {
3061 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3062 return -EIO;
3063 }
3064
3065 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3066 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3067 if (ret < 0)
3068 goto out_free;
3069 }
3070
3071 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3072 ctrl->cntlid = le16_to_cpu(id->cntlid);
3073
3074 if (!ctrl->identified) {
3075 int i;
3076
3077 ret = nvme_init_subsystem(ctrl, id);
3078 if (ret)
3079 goto out_free;
3080
3081 /*
3082 * Check for quirks. Quirk can depend on firmware version,
3083 * so, in principle, the set of quirks present can change
3084 * across a reset. As a possible future enhancement, we
3085 * could re-scan for quirks every time we reinitialize
3086 * the device, but we'd have to make sure that the driver
3087 * behaves intelligently if the quirks change.
3088 */
3089 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3090 if (quirk_matches(id, &core_quirks[i]))
3091 ctrl->quirks |= core_quirks[i].quirks;
3092 }
3093 }
3094
3095 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3096 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3097 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3098 }
3099
3100 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3101 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3102 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3103
3104 ctrl->oacs = le16_to_cpu(id->oacs);
3105 ctrl->oncs = le16_to_cpu(id->oncs);
3106 ctrl->mtfa = le16_to_cpu(id->mtfa);
3107 ctrl->oaes = le32_to_cpu(id->oaes);
3108 ctrl->wctemp = le16_to_cpu(id->wctemp);
3109 ctrl->cctemp = le16_to_cpu(id->cctemp);
3110
3111 atomic_set(&ctrl->abort_limit, id->acl + 1);
3112 ctrl->vwc = id->vwc;
3113 if (id->mdts)
3114 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3115 else
3116 max_hw_sectors = UINT_MAX;
3117 ctrl->max_hw_sectors =
3118 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3119
3120 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3121 ctrl->sgls = le32_to_cpu(id->sgls);
3122 ctrl->kas = le16_to_cpu(id->kas);
3123 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3124 ctrl->ctratt = le32_to_cpu(id->ctratt);
3125
3126 if (id->rtd3e) {
3127 /* us -> s */
3128 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3129
3130 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3131 shutdown_timeout, 60);
3132
3133 if (ctrl->shutdown_timeout != shutdown_timeout)
3134 dev_info(ctrl->device,
3135 "Shutdown timeout set to %u seconds\n",
3136 ctrl->shutdown_timeout);
3137 } else
3138 ctrl->shutdown_timeout = shutdown_timeout;
3139
3140 ctrl->npss = id->npss;
3141 ctrl->apsta = id->apsta;
3142 prev_apst_enabled = ctrl->apst_enabled;
3143 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3144 if (force_apst && id->apsta) {
3145 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3146 ctrl->apst_enabled = true;
3147 } else {
3148 ctrl->apst_enabled = false;
3149 }
3150 } else {
3151 ctrl->apst_enabled = id->apsta;
3152 }
3153 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3154
3155 if (ctrl->ops->flags & NVME_F_FABRICS) {
3156 ctrl->icdoff = le16_to_cpu(id->icdoff);
3157 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3158 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3159 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3160
3161 /*
3162 * In fabrics we need to verify the cntlid matches the
3163 * admin connect
3164 */
3165 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3166 dev_err(ctrl->device,
3167 "Mismatching cntlid: Connect %u vs Identify "
3168 "%u, rejecting\n",
3169 ctrl->cntlid, le16_to_cpu(id->cntlid));
3170 ret = -EINVAL;
3171 goto out_free;
3172 }
3173
3174 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3175 dev_err(ctrl->device,
3176 "keep-alive support is mandatory for fabrics\n");
3177 ret = -EINVAL;
3178 goto out_free;
3179 }
3180 } else {
3181 ctrl->hmpre = le32_to_cpu(id->hmpre);
3182 ctrl->hmmin = le32_to_cpu(id->hmmin);
3183 ctrl->hmminds = le32_to_cpu(id->hmminds);
3184 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3185 }
3186
3187 ret = nvme_mpath_init(ctrl, id);
3188 kfree(id);
3189
3190 if (ret < 0)
3191 return ret;
3192
3193 if (ctrl->apst_enabled && !prev_apst_enabled)
3194 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3195 else if (!ctrl->apst_enabled && prev_apst_enabled)
3196 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3197
3198 ret = nvme_configure_apst(ctrl);
3199 if (ret < 0)
3200 return ret;
3201
3202 ret = nvme_configure_timestamp(ctrl);
3203 if (ret < 0)
3204 return ret;
3205
3206 ret = nvme_configure_directives(ctrl);
3207 if (ret < 0)
3208 return ret;
3209
3210 ret = nvme_configure_acre(ctrl);
3211 if (ret < 0)
3212 return ret;
3213
3214 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3215 ret = nvme_hwmon_init(ctrl);
3216 if (ret < 0)
3217 return ret;
3218 }
3219
3220 ctrl->identified = true;
3221
3222 return 0;
3223
3224 out_free:
3225 kfree(id);
3226 return ret;
3227 }
3228 EXPORT_SYMBOL_GPL(nvme_init_identify);
3229
3230 static int nvme_dev_open(struct inode *inode, struct file *file)
3231 {
3232 struct nvme_ctrl *ctrl =
3233 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3234
3235 switch (ctrl->state) {
3236 case NVME_CTRL_LIVE:
3237 break;
3238 default:
3239 return -EWOULDBLOCK;
3240 }
3241
3242 nvme_get_ctrl(ctrl);
3243 if (!try_module_get(ctrl->ops->module)) {
3244 nvme_put_ctrl(ctrl);
3245 return -EINVAL;
3246 }
3247
3248 file->private_data = ctrl;
3249 return 0;
3250 }
3251
3252 static int nvme_dev_release(struct inode *inode, struct file *file)
3253 {
3254 struct nvme_ctrl *ctrl =
3255 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3256
3257 module_put(ctrl->ops->module);
3258 nvme_put_ctrl(ctrl);
3259 return 0;
3260 }
3261
3262 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3263 {
3264 struct nvme_ns *ns;
3265 int ret;
3266
3267 down_read(&ctrl->namespaces_rwsem);
3268 if (list_empty(&ctrl->namespaces)) {
3269 ret = -ENOTTY;
3270 goto out_unlock;
3271 }
3272
3273 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3274 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3275 dev_warn(ctrl->device,
3276 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3277 ret = -EINVAL;
3278 goto out_unlock;
3279 }
3280
3281 dev_warn(ctrl->device,
3282 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3283 kref_get(&ns->kref);
3284 up_read(&ctrl->namespaces_rwsem);
3285
3286 ret = nvme_user_cmd(ctrl, ns, argp);
3287 nvme_put_ns(ns);
3288 return ret;
3289
3290 out_unlock:
3291 up_read(&ctrl->namespaces_rwsem);
3292 return ret;
3293 }
3294
3295 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3296 unsigned long arg)
3297 {
3298 struct nvme_ctrl *ctrl = file->private_data;
3299 void __user *argp = (void __user *)arg;
3300
3301 switch (cmd) {
3302 case NVME_IOCTL_ADMIN_CMD:
3303 return nvme_user_cmd(ctrl, NULL, argp);
3304 case NVME_IOCTL_ADMIN64_CMD:
3305 return nvme_user_cmd64(ctrl, NULL, argp);
3306 case NVME_IOCTL_IO_CMD:
3307 return nvme_dev_user_cmd(ctrl, argp);
3308 case NVME_IOCTL_RESET:
3309 dev_warn(ctrl->device, "resetting controller\n");
3310 return nvme_reset_ctrl_sync(ctrl);
3311 case NVME_IOCTL_SUBSYS_RESET:
3312 return nvme_reset_subsystem(ctrl);
3313 case NVME_IOCTL_RESCAN:
3314 nvme_queue_scan(ctrl);
3315 return 0;
3316 default:
3317 return -ENOTTY;
3318 }
3319 }
3320
3321 static const struct file_operations nvme_dev_fops = {
3322 .owner = THIS_MODULE,
3323 .open = nvme_dev_open,
3324 .release = nvme_dev_release,
3325 .unlocked_ioctl = nvme_dev_ioctl,
3326 .compat_ioctl = compat_ptr_ioctl,
3327 };
3328
3329 static ssize_t nvme_sysfs_reset(struct device *dev,
3330 struct device_attribute *attr, const char *buf,
3331 size_t count)
3332 {
3333 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3334 int ret;
3335
3336 ret = nvme_reset_ctrl_sync(ctrl);
3337 if (ret < 0)
3338 return ret;
3339 return count;
3340 }
3341 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3342
3343 static ssize_t nvme_sysfs_rescan(struct device *dev,
3344 struct device_attribute *attr, const char *buf,
3345 size_t count)
3346 {
3347 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3348
3349 nvme_queue_scan(ctrl);
3350 return count;
3351 }
3352 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3353
3354 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3355 {
3356 struct gendisk *disk = dev_to_disk(dev);
3357
3358 if (disk->fops == &nvme_bdev_ops)
3359 return nvme_get_ns_from_dev(dev)->head;
3360 else
3361 return disk->private_data;
3362 }
3363
3364 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3365 char *buf)
3366 {
3367 struct nvme_ns_head *head = dev_to_ns_head(dev);
3368 struct nvme_ns_ids *ids = &head->ids;
3369 struct nvme_subsystem *subsys = head->subsys;
3370 int serial_len = sizeof(subsys->serial);
3371 int model_len = sizeof(subsys->model);
3372
3373 if (!uuid_is_null(&ids->uuid))
3374 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3375
3376 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3377 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3378
3379 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3380 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3381
3382 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3383 subsys->serial[serial_len - 1] == '\0'))
3384 serial_len--;
3385 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3386 subsys->model[model_len - 1] == '\0'))
3387 model_len--;
3388
3389 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3390 serial_len, subsys->serial, model_len, subsys->model,
3391 head->ns_id);
3392 }
3393 static DEVICE_ATTR_RO(wwid);
3394
3395 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3396 char *buf)
3397 {
3398 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3399 }
3400 static DEVICE_ATTR_RO(nguid);
3401
3402 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3403 char *buf)
3404 {
3405 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3406
3407 /* For backward compatibility expose the NGUID to userspace if
3408 * we have no UUID set
3409 */
3410 if (uuid_is_null(&ids->uuid)) {
3411 printk_ratelimited(KERN_WARNING
3412 "No UUID available providing old NGUID\n");
3413 return sprintf(buf, "%pU\n", ids->nguid);
3414 }
3415 return sprintf(buf, "%pU\n", &ids->uuid);
3416 }
3417 static DEVICE_ATTR_RO(uuid);
3418
3419 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3420 char *buf)
3421 {
3422 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3423 }
3424 static DEVICE_ATTR_RO(eui);
3425
3426 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3427 char *buf)
3428 {
3429 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3430 }
3431 static DEVICE_ATTR_RO(nsid);
3432
3433 static struct attribute *nvme_ns_id_attrs[] = {
3434 &dev_attr_wwid.attr,
3435 &dev_attr_uuid.attr,
3436 &dev_attr_nguid.attr,
3437 &dev_attr_eui.attr,
3438 &dev_attr_nsid.attr,
3439 #ifdef CONFIG_NVME_MULTIPATH
3440 &dev_attr_ana_grpid.attr,
3441 &dev_attr_ana_state.attr,
3442 #endif
3443 NULL,
3444 };
3445
3446 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3447 struct attribute *a, int n)
3448 {
3449 struct device *dev = container_of(kobj, struct device, kobj);
3450 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3451
3452 if (a == &dev_attr_uuid.attr) {
3453 if (uuid_is_null(&ids->uuid) &&
3454 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3455 return 0;
3456 }
3457 if (a == &dev_attr_nguid.attr) {
3458 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3459 return 0;
3460 }
3461 if (a == &dev_attr_eui.attr) {
3462 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3463 return 0;
3464 }
3465 #ifdef CONFIG_NVME_MULTIPATH
3466 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3467 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3468 return 0;
3469 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3470 return 0;
3471 }
3472 #endif
3473 return a->mode;
3474 }
3475
3476 static const struct attribute_group nvme_ns_id_attr_group = {
3477 .attrs = nvme_ns_id_attrs,
3478 .is_visible = nvme_ns_id_attrs_are_visible,
3479 };
3480
3481 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3482 &nvme_ns_id_attr_group,
3483 #ifdef CONFIG_NVM
3484 &nvme_nvm_attr_group,
3485 #endif
3486 NULL,
3487 };
3488
3489 #define nvme_show_str_function(field) \
3490 static ssize_t field##_show(struct device *dev, \
3491 struct device_attribute *attr, char *buf) \
3492 { \
3493 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3494 return sprintf(buf, "%.*s\n", \
3495 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3496 } \
3497 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3498
3499 nvme_show_str_function(model);
3500 nvme_show_str_function(serial);
3501 nvme_show_str_function(firmware_rev);
3502
3503 #define nvme_show_int_function(field) \
3504 static ssize_t field##_show(struct device *dev, \
3505 struct device_attribute *attr, char *buf) \
3506 { \
3507 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3508 return sprintf(buf, "%d\n", ctrl->field); \
3509 } \
3510 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3511
3512 nvme_show_int_function(cntlid);
3513 nvme_show_int_function(numa_node);
3514 nvme_show_int_function(queue_count);
3515 nvme_show_int_function(sqsize);
3516
3517 static ssize_t nvme_sysfs_delete(struct device *dev,
3518 struct device_attribute *attr, const char *buf,
3519 size_t count)
3520 {
3521 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3522
3523 if (device_remove_file_self(dev, attr))
3524 nvme_delete_ctrl_sync(ctrl);
3525 return count;
3526 }
3527 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3528
3529 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3530 struct device_attribute *attr,
3531 char *buf)
3532 {
3533 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3534
3535 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3536 }
3537 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3538
3539 static ssize_t nvme_sysfs_show_state(struct device *dev,
3540 struct device_attribute *attr,
3541 char *buf)
3542 {
3543 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3544 static const char *const state_name[] = {
3545 [NVME_CTRL_NEW] = "new",
3546 [NVME_CTRL_LIVE] = "live",
3547 [NVME_CTRL_RESETTING] = "resetting",
3548 [NVME_CTRL_CONNECTING] = "connecting",
3549 [NVME_CTRL_DELETING] = "deleting",
3550 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3551 [NVME_CTRL_DEAD] = "dead",
3552 };
3553
3554 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3555 state_name[ctrl->state])
3556 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3557
3558 return sprintf(buf, "unknown state\n");
3559 }
3560
3561 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3562
3563 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3564 struct device_attribute *attr,
3565 char *buf)
3566 {
3567 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3568
3569 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3570 }
3571 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3572
3573 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3574 struct device_attribute *attr,
3575 char *buf)
3576 {
3577 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3578
3579 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3580 }
3581 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3582
3583 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3584 struct device_attribute *attr,
3585 char *buf)
3586 {
3587 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3588
3589 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3590 }
3591 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3592
3593 static ssize_t nvme_sysfs_show_address(struct device *dev,
3594 struct device_attribute *attr,
3595 char *buf)
3596 {
3597 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3598
3599 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3600 }
3601 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3602
3603 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3604 struct device_attribute *attr, char *buf)
3605 {
3606 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3607 struct nvmf_ctrl_options *opts = ctrl->opts;
3608
3609 if (ctrl->opts->max_reconnects == -1)
3610 return sprintf(buf, "off\n");
3611 return sprintf(buf, "%d\n",
3612 opts->max_reconnects * opts->reconnect_delay);
3613 }
3614
3615 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3616 struct device_attribute *attr, const char *buf, size_t count)
3617 {
3618 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3619 struct nvmf_ctrl_options *opts = ctrl->opts;
3620 int ctrl_loss_tmo, err;
3621
3622 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3623 if (err)
3624 return -EINVAL;
3625
3626 else if (ctrl_loss_tmo < 0)
3627 opts->max_reconnects = -1;
3628 else
3629 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3630 opts->reconnect_delay);
3631 return count;
3632 }
3633 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3634 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3635
3636 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3637 struct device_attribute *attr, char *buf)
3638 {
3639 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3640
3641 if (ctrl->opts->reconnect_delay == -1)
3642 return sprintf(buf, "off\n");
3643 return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3644 }
3645
3646 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3647 struct device_attribute *attr, const char *buf, size_t count)
3648 {
3649 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3650 unsigned int v;
3651 int err;
3652
3653 err = kstrtou32(buf, 10, &v);
3654 if (err)
3655 return err;
3656
3657 ctrl->opts->reconnect_delay = v;
3658 return count;
3659 }
3660 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3661 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3662
3663 static struct attribute *nvme_dev_attrs[] = {
3664 &dev_attr_reset_controller.attr,
3665 &dev_attr_rescan_controller.attr,
3666 &dev_attr_model.attr,
3667 &dev_attr_serial.attr,
3668 &dev_attr_firmware_rev.attr,
3669 &dev_attr_cntlid.attr,
3670 &dev_attr_delete_controller.attr,
3671 &dev_attr_transport.attr,
3672 &dev_attr_subsysnqn.attr,
3673 &dev_attr_address.attr,
3674 &dev_attr_state.attr,
3675 &dev_attr_numa_node.attr,
3676 &dev_attr_queue_count.attr,
3677 &dev_attr_sqsize.attr,
3678 &dev_attr_hostnqn.attr,
3679 &dev_attr_hostid.attr,
3680 &dev_attr_ctrl_loss_tmo.attr,
3681 &dev_attr_reconnect_delay.attr,
3682 NULL
3683 };
3684
3685 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3686 struct attribute *a, int n)
3687 {
3688 struct device *dev = container_of(kobj, struct device, kobj);
3689 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3690
3691 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3692 return 0;
3693 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3694 return 0;
3695 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3696 return 0;
3697 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3698 return 0;
3699 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3700 return 0;
3701 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3702 return 0;
3703
3704 return a->mode;
3705 }
3706
3707 static struct attribute_group nvme_dev_attrs_group = {
3708 .attrs = nvme_dev_attrs,
3709 .is_visible = nvme_dev_attrs_are_visible,
3710 };
3711
3712 static const struct attribute_group *nvme_dev_attr_groups[] = {
3713 &nvme_dev_attrs_group,
3714 NULL,
3715 };
3716
3717 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3718 unsigned nsid)
3719 {
3720 struct nvme_ns_head *h;
3721
3722 lockdep_assert_held(&subsys->lock);
3723
3724 list_for_each_entry(h, &subsys->nsheads, entry) {
3725 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3726 return h;
3727 }
3728
3729 return NULL;
3730 }
3731
3732 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3733 struct nvme_ns_head *new)
3734 {
3735 struct nvme_ns_head *h;
3736
3737 lockdep_assert_held(&subsys->lock);
3738
3739 list_for_each_entry(h, &subsys->nsheads, entry) {
3740 if (nvme_ns_ids_valid(&new->ids) &&
3741 nvme_ns_ids_equal(&new->ids, &h->ids))
3742 return -EINVAL;
3743 }
3744
3745 return 0;
3746 }
3747
3748 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3749 unsigned nsid, struct nvme_ns_ids *ids)
3750 {
3751 struct nvme_ns_head *head;
3752 size_t size = sizeof(*head);
3753 int ret = -ENOMEM;
3754
3755 #ifdef CONFIG_NVME_MULTIPATH
3756 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3757 #endif
3758
3759 head = kzalloc(size, GFP_KERNEL);
3760 if (!head)
3761 goto out;
3762 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3763 if (ret < 0)
3764 goto out_free_head;
3765 head->instance = ret;
3766 INIT_LIST_HEAD(&head->list);
3767 ret = init_srcu_struct(&head->srcu);
3768 if (ret)
3769 goto out_ida_remove;
3770 head->subsys = ctrl->subsys;
3771 head->ns_id = nsid;
3772 head->ids = *ids;
3773 kref_init(&head->ref);
3774
3775 ret = __nvme_check_ids(ctrl->subsys, head);
3776 if (ret) {
3777 dev_err(ctrl->device,
3778 "duplicate IDs for nsid %d\n", nsid);
3779 goto out_cleanup_srcu;
3780 }
3781
3782 if (head->ids.csi) {
3783 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3784 if (ret)
3785 goto out_cleanup_srcu;
3786 } else
3787 head->effects = ctrl->effects;
3788
3789 ret = nvme_mpath_alloc_disk(ctrl, head);
3790 if (ret)
3791 goto out_cleanup_srcu;
3792
3793 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3794
3795 kref_get(&ctrl->subsys->ref);
3796
3797 return head;
3798 out_cleanup_srcu:
3799 cleanup_srcu_struct(&head->srcu);
3800 out_ida_remove:
3801 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3802 out_free_head:
3803 kfree(head);
3804 out:
3805 if (ret > 0)
3806 ret = blk_status_to_errno(nvme_error_status(ret));
3807 return ERR_PTR(ret);
3808 }
3809
3810 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3811 struct nvme_ns_ids *ids, bool is_shared)
3812 {
3813 struct nvme_ctrl *ctrl = ns->ctrl;
3814 struct nvme_ns_head *head = NULL;
3815 int ret = 0;
3816
3817 mutex_lock(&ctrl->subsys->lock);
3818 head = nvme_find_ns_head(ctrl->subsys, nsid);
3819 if (!head) {
3820 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3821 if (IS_ERR(head)) {
3822 ret = PTR_ERR(head);
3823 goto out_unlock;
3824 }
3825 head->shared = is_shared;
3826 } else {
3827 ret = -EINVAL;
3828 if (!is_shared || !head->shared) {
3829 dev_err(ctrl->device,
3830 "Duplicate unshared namespace %d\n", nsid);
3831 goto out_put_ns_head;
3832 }
3833 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3834 dev_err(ctrl->device,
3835 "IDs don't match for shared namespace %d\n",
3836 nsid);
3837 goto out_put_ns_head;
3838 }
3839 }
3840
3841 list_add_tail_rcu(&ns->siblings, &head->list);
3842 ns->head = head;
3843 mutex_unlock(&ctrl->subsys->lock);
3844 return 0;
3845
3846 out_put_ns_head:
3847 nvme_put_ns_head(head);
3848 out_unlock:
3849 mutex_unlock(&ctrl->subsys->lock);
3850 return ret;
3851 }
3852
3853 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3854 {
3855 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3856 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3857
3858 return nsa->head->ns_id - nsb->head->ns_id;
3859 }
3860
3861 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3862 {
3863 struct nvme_ns *ns, *ret = NULL;
3864
3865 down_read(&ctrl->namespaces_rwsem);
3866 list_for_each_entry(ns, &ctrl->namespaces, list) {
3867 if (ns->head->ns_id == nsid) {
3868 if (!kref_get_unless_zero(&ns->kref))
3869 continue;
3870 ret = ns;
3871 break;
3872 }
3873 if (ns->head->ns_id > nsid)
3874 break;
3875 }
3876 up_read(&ctrl->namespaces_rwsem);
3877 return ret;
3878 }
3879 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3880
3881 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3882 struct nvme_ns_ids *ids)
3883 {
3884 struct nvme_ns *ns;
3885 struct gendisk *disk;
3886 struct nvme_id_ns *id;
3887 char disk_name[DISK_NAME_LEN];
3888 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3889
3890 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3891 return;
3892
3893 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3894 if (!ns)
3895 goto out_free_id;
3896
3897 ns->queue = blk_mq_init_queue(ctrl->tagset);
3898 if (IS_ERR(ns->queue))
3899 goto out_free_ns;
3900
3901 if (ctrl->opts && ctrl->opts->data_digest)
3902 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3903
3904 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3905 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3906 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3907
3908 ns->queue->queuedata = ns;
3909 ns->ctrl = ctrl;
3910 kref_init(&ns->kref);
3911
3912 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3913 goto out_free_queue;
3914 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3915
3916 disk = alloc_disk_node(0, node);
3917 if (!disk)
3918 goto out_unlink_ns;
3919
3920 disk->fops = &nvme_bdev_ops;
3921 disk->private_data = ns;
3922 disk->queue = ns->queue;
3923 disk->flags = flags;
3924 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3925 ns->disk = disk;
3926
3927 if (nvme_update_ns_info(ns, id))
3928 goto out_put_disk;
3929
3930 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3931 if (nvme_nvm_register(ns, disk_name, node)) {
3932 dev_warn(ctrl->device, "LightNVM init failure\n");
3933 goto out_put_disk;
3934 }
3935 }
3936
3937 down_write(&ctrl->namespaces_rwsem);
3938 list_add_tail(&ns->list, &ctrl->namespaces);
3939 up_write(&ctrl->namespaces_rwsem);
3940
3941 nvme_get_ctrl(ctrl);
3942
3943 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3944
3945 nvme_mpath_add_disk(ns, id);
3946 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3947 kfree(id);
3948
3949 return;
3950 out_put_disk:
3951 /* prevent double queue cleanup */
3952 ns->disk->queue = NULL;
3953 put_disk(ns->disk);
3954 out_unlink_ns:
3955 mutex_lock(&ctrl->subsys->lock);
3956 list_del_rcu(&ns->siblings);
3957 if (list_empty(&ns->head->list))
3958 list_del_init(&ns->head->entry);
3959 mutex_unlock(&ctrl->subsys->lock);
3960 nvme_put_ns_head(ns->head);
3961 out_free_queue:
3962 blk_cleanup_queue(ns->queue);
3963 out_free_ns:
3964 kfree(ns);
3965 out_free_id:
3966 kfree(id);
3967 }
3968
3969 static void nvme_ns_remove(struct nvme_ns *ns)
3970 {
3971 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3972 return;
3973
3974 set_capacity(ns->disk, 0);
3975 nvme_fault_inject_fini(&ns->fault_inject);
3976
3977 mutex_lock(&ns->ctrl->subsys->lock);
3978 list_del_rcu(&ns->siblings);
3979 if (list_empty(&ns->head->list))
3980 list_del_init(&ns->head->entry);
3981 mutex_unlock(&ns->ctrl->subsys->lock);
3982
3983 synchronize_rcu(); /* guarantee not available in head->list */
3984 nvme_mpath_clear_current_path(ns);
3985 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3986
3987 if (ns->disk->flags & GENHD_FL_UP) {
3988 del_gendisk(ns->disk);
3989 blk_cleanup_queue(ns->queue);
3990 if (blk_get_integrity(ns->disk))
3991 blk_integrity_unregister(ns->disk);
3992 }
3993
3994 down_write(&ns->ctrl->namespaces_rwsem);
3995 list_del_init(&ns->list);
3996 up_write(&ns->ctrl->namespaces_rwsem);
3997
3998 nvme_mpath_check_last_path(ns);
3999 nvme_put_ns(ns);
4000 }
4001
4002 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4003 {
4004 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4005
4006 if (ns) {
4007 nvme_ns_remove(ns);
4008 nvme_put_ns(ns);
4009 }
4010 }
4011
4012 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4013 {
4014 struct nvme_id_ns *id;
4015 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4016
4017 if (test_bit(NVME_NS_DEAD, &ns->flags))
4018 goto out;
4019
4020 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4021 if (ret)
4022 goto out;
4023
4024 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4025 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4026 dev_err(ns->ctrl->device,
4027 "identifiers changed for nsid %d\n", ns->head->ns_id);
4028 goto out_free_id;
4029 }
4030
4031 ret = nvme_update_ns_info(ns, id);
4032
4033 out_free_id:
4034 kfree(id);
4035 out:
4036 /*
4037 * Only remove the namespace if we got a fatal error back from the
4038 * device, otherwise ignore the error and just move on.
4039 *
4040 * TODO: we should probably schedule a delayed retry here.
4041 */
4042 if (ret > 0 && (ret & NVME_SC_DNR))
4043 nvme_ns_remove(ns);
4044 }
4045
4046 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4047 {
4048 struct nvme_ns_ids ids = { };
4049 struct nvme_ns *ns;
4050
4051 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4052 return;
4053
4054 ns = nvme_find_get_ns(ctrl, nsid);
4055 if (ns) {
4056 nvme_validate_ns(ns, &ids);
4057 nvme_put_ns(ns);
4058 return;
4059 }
4060
4061 switch (ids.csi) {
4062 case NVME_CSI_NVM:
4063 nvme_alloc_ns(ctrl, nsid, &ids);
4064 break;
4065 case NVME_CSI_ZNS:
4066 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4067 dev_warn(ctrl->device,
4068 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4069 nsid);
4070 break;
4071 }
4072 if (!nvme_multi_css(ctrl)) {
4073 dev_warn(ctrl->device,
4074 "command set not reported for nsid: %d\n",
4075 nsid);
4076 break;
4077 }
4078 nvme_alloc_ns(ctrl, nsid, &ids);
4079 break;
4080 default:
4081 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4082 ids.csi, nsid);
4083 break;
4084 }
4085 }
4086
4087 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4088 unsigned nsid)
4089 {
4090 struct nvme_ns *ns, *next;
4091 LIST_HEAD(rm_list);
4092
4093 down_write(&ctrl->namespaces_rwsem);
4094 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4095 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4096 list_move_tail(&ns->list, &rm_list);
4097 }
4098 up_write(&ctrl->namespaces_rwsem);
4099
4100 list_for_each_entry_safe(ns, next, &rm_list, list)
4101 nvme_ns_remove(ns);
4102
4103 }
4104
4105 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4106 {
4107 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4108 __le32 *ns_list;
4109 u32 prev = 0;
4110 int ret = 0, i;
4111
4112 if (nvme_ctrl_limited_cns(ctrl))
4113 return -EOPNOTSUPP;
4114
4115 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4116 if (!ns_list)
4117 return -ENOMEM;
4118
4119 for (;;) {
4120 struct nvme_command cmd = {
4121 .identify.opcode = nvme_admin_identify,
4122 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4123 .identify.nsid = cpu_to_le32(prev),
4124 };
4125
4126 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4127 NVME_IDENTIFY_DATA_SIZE);
4128 if (ret) {
4129 dev_warn(ctrl->device,
4130 "Identify NS List failed (status=0x%x)\n", ret);
4131 goto free;
4132 }
4133
4134 for (i = 0; i < nr_entries; i++) {
4135 u32 nsid = le32_to_cpu(ns_list[i]);
4136
4137 if (!nsid) /* end of the list? */
4138 goto out;
4139 nvme_validate_or_alloc_ns(ctrl, nsid);
4140 while (++prev < nsid)
4141 nvme_ns_remove_by_nsid(ctrl, prev);
4142 }
4143 }
4144 out:
4145 nvme_remove_invalid_namespaces(ctrl, prev);
4146 free:
4147 kfree(ns_list);
4148 return ret;
4149 }
4150
4151 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4152 {
4153 struct nvme_id_ctrl *id;
4154 u32 nn, i;
4155
4156 if (nvme_identify_ctrl(ctrl, &id))
4157 return;
4158 nn = le32_to_cpu(id->nn);
4159 kfree(id);
4160
4161 for (i = 1; i <= nn; i++)
4162 nvme_validate_or_alloc_ns(ctrl, i);
4163
4164 nvme_remove_invalid_namespaces(ctrl, nn);
4165 }
4166
4167 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4168 {
4169 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4170 __le32 *log;
4171 int error;
4172
4173 log = kzalloc(log_size, GFP_KERNEL);
4174 if (!log)
4175 return;
4176
4177 /*
4178 * We need to read the log to clear the AEN, but we don't want to rely
4179 * on it for the changed namespace information as userspace could have
4180 * raced with us in reading the log page, which could cause us to miss
4181 * updates.
4182 */
4183 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4184 NVME_CSI_NVM, log, log_size, 0);
4185 if (error)
4186 dev_warn(ctrl->device,
4187 "reading changed ns log failed: %d\n", error);
4188
4189 kfree(log);
4190 }
4191
4192 static void nvme_scan_work(struct work_struct *work)
4193 {
4194 struct nvme_ctrl *ctrl =
4195 container_of(work, struct nvme_ctrl, scan_work);
4196
4197 /* No tagset on a live ctrl means IO queues could not created */
4198 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4199 return;
4200
4201 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4202 dev_info(ctrl->device, "rescanning namespaces.\n");
4203 nvme_clear_changed_ns_log(ctrl);
4204 }
4205
4206 mutex_lock(&ctrl->scan_lock);
4207 if (nvme_scan_ns_list(ctrl) != 0)
4208 nvme_scan_ns_sequential(ctrl);
4209 mutex_unlock(&ctrl->scan_lock);
4210
4211 down_write(&ctrl->namespaces_rwsem);
4212 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4213 up_write(&ctrl->namespaces_rwsem);
4214 }
4215
4216 /*
4217 * This function iterates the namespace list unlocked to allow recovery from
4218 * controller failure. It is up to the caller to ensure the namespace list is
4219 * not modified by scan work while this function is executing.
4220 */
4221 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4222 {
4223 struct nvme_ns *ns, *next;
4224 LIST_HEAD(ns_list);
4225
4226 /*
4227 * make sure to requeue I/O to all namespaces as these
4228 * might result from the scan itself and must complete
4229 * for the scan_work to make progress
4230 */
4231 nvme_mpath_clear_ctrl_paths(ctrl);
4232
4233 /* prevent racing with ns scanning */
4234 flush_work(&ctrl->scan_work);
4235
4236 /*
4237 * The dead states indicates the controller was not gracefully
4238 * disconnected. In that case, we won't be able to flush any data while
4239 * removing the namespaces' disks; fail all the queues now to avoid
4240 * potentially having to clean up the failed sync later.
4241 */
4242 if (ctrl->state == NVME_CTRL_DEAD)
4243 nvme_kill_queues(ctrl);
4244
4245 /* this is a no-op when called from the controller reset handler */
4246 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4247
4248 down_write(&ctrl->namespaces_rwsem);
4249 list_splice_init(&ctrl->namespaces, &ns_list);
4250 up_write(&ctrl->namespaces_rwsem);
4251
4252 list_for_each_entry_safe(ns, next, &ns_list, list)
4253 nvme_ns_remove(ns);
4254 }
4255 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4256
4257 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4258 {
4259 struct nvme_ctrl *ctrl =
4260 container_of(dev, struct nvme_ctrl, ctrl_device);
4261 struct nvmf_ctrl_options *opts = ctrl->opts;
4262 int ret;
4263
4264 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4265 if (ret)
4266 return ret;
4267
4268 if (opts) {
4269 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4270 if (ret)
4271 return ret;
4272
4273 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4274 opts->trsvcid ?: "none");
4275 if (ret)
4276 return ret;
4277
4278 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4279 opts->host_traddr ?: "none");
4280 }
4281 return ret;
4282 }
4283
4284 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4285 {
4286 char *envp[2] = { NULL, NULL };
4287 u32 aen_result = ctrl->aen_result;
4288
4289 ctrl->aen_result = 0;
4290 if (!aen_result)
4291 return;
4292
4293 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4294 if (!envp[0])
4295 return;
4296 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4297 kfree(envp[0]);
4298 }
4299
4300 static void nvme_async_event_work(struct work_struct *work)
4301 {
4302 struct nvme_ctrl *ctrl =
4303 container_of(work, struct nvme_ctrl, async_event_work);
4304
4305 nvme_aen_uevent(ctrl);
4306 ctrl->ops->submit_async_event(ctrl);
4307 }
4308
4309 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4310 {
4311
4312 u32 csts;
4313
4314 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4315 return false;
4316
4317 if (csts == ~0)
4318 return false;
4319
4320 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4321 }
4322
4323 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4324 {
4325 struct nvme_fw_slot_info_log *log;
4326
4327 log = kmalloc(sizeof(*log), GFP_KERNEL);
4328 if (!log)
4329 return;
4330
4331 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4332 log, sizeof(*log), 0))
4333 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4334 kfree(log);
4335 }
4336
4337 static void nvme_fw_act_work(struct work_struct *work)
4338 {
4339 struct nvme_ctrl *ctrl = container_of(work,
4340 struct nvme_ctrl, fw_act_work);
4341 unsigned long fw_act_timeout;
4342
4343 if (ctrl->mtfa)
4344 fw_act_timeout = jiffies +
4345 msecs_to_jiffies(ctrl->mtfa * 100);
4346 else
4347 fw_act_timeout = jiffies +
4348 msecs_to_jiffies(admin_timeout * 1000);
4349
4350 nvme_stop_queues(ctrl);
4351 while (nvme_ctrl_pp_status(ctrl)) {
4352 if (time_after(jiffies, fw_act_timeout)) {
4353 dev_warn(ctrl->device,
4354 "Fw activation timeout, reset controller\n");
4355 nvme_try_sched_reset(ctrl);
4356 return;
4357 }
4358 msleep(100);
4359 }
4360
4361 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4362 return;
4363
4364 nvme_start_queues(ctrl);
4365 /* read FW slot information to clear the AER */
4366 nvme_get_fw_slot_info(ctrl);
4367 }
4368
4369 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4370 {
4371 u32 aer_notice_type = (result & 0xff00) >> 8;
4372
4373 trace_nvme_async_event(ctrl, aer_notice_type);
4374
4375 switch (aer_notice_type) {
4376 case NVME_AER_NOTICE_NS_CHANGED:
4377 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4378 nvme_queue_scan(ctrl);
4379 break;
4380 case NVME_AER_NOTICE_FW_ACT_STARTING:
4381 /*
4382 * We are (ab)using the RESETTING state to prevent subsequent
4383 * recovery actions from interfering with the controller's
4384 * firmware activation.
4385 */
4386 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4387 queue_work(nvme_wq, &ctrl->fw_act_work);
4388 break;
4389 #ifdef CONFIG_NVME_MULTIPATH
4390 case NVME_AER_NOTICE_ANA:
4391 if (!ctrl->ana_log_buf)
4392 break;
4393 queue_work(nvme_wq, &ctrl->ana_work);
4394 break;
4395 #endif
4396 case NVME_AER_NOTICE_DISC_CHANGED:
4397 ctrl->aen_result = result;
4398 break;
4399 default:
4400 dev_warn(ctrl->device, "async event result %08x\n", result);
4401 }
4402 }
4403
4404 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4405 volatile union nvme_result *res)
4406 {
4407 u32 result = le32_to_cpu(res->u32);
4408 u32 aer_type = result & 0x07;
4409
4410 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4411 return;
4412
4413 switch (aer_type) {
4414 case NVME_AER_NOTICE:
4415 nvme_handle_aen_notice(ctrl, result);
4416 break;
4417 case NVME_AER_ERROR:
4418 case NVME_AER_SMART:
4419 case NVME_AER_CSS:
4420 case NVME_AER_VS:
4421 trace_nvme_async_event(ctrl, aer_type);
4422 ctrl->aen_result = result;
4423 break;
4424 default:
4425 break;
4426 }
4427 queue_work(nvme_wq, &ctrl->async_event_work);
4428 }
4429 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4430
4431 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4432 {
4433 nvme_mpath_stop(ctrl);
4434 nvme_stop_keep_alive(ctrl);
4435 nvme_stop_failfast_work(ctrl);
4436 flush_work(&ctrl->async_event_work);
4437 cancel_work_sync(&ctrl->fw_act_work);
4438 }
4439 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4440
4441 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4442 {
4443 nvme_start_keep_alive(ctrl);
4444
4445 nvme_enable_aen(ctrl);
4446
4447 if (ctrl->queue_count > 1) {
4448 nvme_queue_scan(ctrl);
4449 nvme_start_queues(ctrl);
4450 }
4451 }
4452 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4453
4454 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4455 {
4456 nvme_fault_inject_fini(&ctrl->fault_inject);
4457 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4458 cdev_device_del(&ctrl->cdev, ctrl->device);
4459 nvme_put_ctrl(ctrl);
4460 }
4461 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4462
4463 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4464 {
4465 struct nvme_effects_log *cel;
4466 unsigned long i;
4467
4468 xa_for_each (&ctrl->cels, i, cel) {
4469 xa_erase(&ctrl->cels, i);
4470 kfree(cel);
4471 }
4472
4473 xa_destroy(&ctrl->cels);
4474 }
4475
4476 static void nvme_free_ctrl(struct device *dev)
4477 {
4478 struct nvme_ctrl *ctrl =
4479 container_of(dev, struct nvme_ctrl, ctrl_device);
4480 struct nvme_subsystem *subsys = ctrl->subsys;
4481
4482 if (!subsys || ctrl->instance != subsys->instance)
4483 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4484
4485 nvme_free_cels(ctrl);
4486 nvme_mpath_uninit(ctrl);
4487 __free_page(ctrl->discard_page);
4488
4489 if (subsys) {
4490 mutex_lock(&nvme_subsystems_lock);
4491 list_del(&ctrl->subsys_entry);
4492 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4493 mutex_unlock(&nvme_subsystems_lock);
4494 }
4495
4496 ctrl->ops->free_ctrl(ctrl);
4497
4498 if (subsys)
4499 nvme_put_subsystem(subsys);
4500 }
4501
4502 /*
4503 * Initialize a NVMe controller structures. This needs to be called during
4504 * earliest initialization so that we have the initialized structured around
4505 * during probing.
4506 */
4507 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4508 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4509 {
4510 int ret;
4511
4512 ctrl->state = NVME_CTRL_NEW;
4513 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4514 spin_lock_init(&ctrl->lock);
4515 mutex_init(&ctrl->scan_lock);
4516 INIT_LIST_HEAD(&ctrl->namespaces);
4517 xa_init(&ctrl->cels);
4518 init_rwsem(&ctrl->namespaces_rwsem);
4519 ctrl->dev = dev;
4520 ctrl->ops = ops;
4521 ctrl->quirks = quirks;
4522 ctrl->numa_node = NUMA_NO_NODE;
4523 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4524 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4525 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4526 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4527 init_waitqueue_head(&ctrl->state_wq);
4528
4529 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4530 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4531 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4532 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4533
4534 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4535 PAGE_SIZE);
4536 ctrl->discard_page = alloc_page(GFP_KERNEL);
4537 if (!ctrl->discard_page) {
4538 ret = -ENOMEM;
4539 goto out;
4540 }
4541
4542 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4543 if (ret < 0)
4544 goto out;
4545 ctrl->instance = ret;
4546
4547 device_initialize(&ctrl->ctrl_device);
4548 ctrl->device = &ctrl->ctrl_device;
4549 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4550 ctrl->instance);
4551 ctrl->device->class = nvme_class;
4552 ctrl->device->parent = ctrl->dev;
4553 ctrl->device->groups = nvme_dev_attr_groups;
4554 ctrl->device->release = nvme_free_ctrl;
4555 dev_set_drvdata(ctrl->device, ctrl);
4556 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4557 if (ret)
4558 goto out_release_instance;
4559
4560 nvme_get_ctrl(ctrl);
4561 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4562 ctrl->cdev.owner = ops->module;
4563 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4564 if (ret)
4565 goto out_free_name;
4566
4567 /*
4568 * Initialize latency tolerance controls. The sysfs files won't
4569 * be visible to userspace unless the device actually supports APST.
4570 */
4571 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4572 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4573 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4574
4575 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4576
4577 return 0;
4578 out_free_name:
4579 nvme_put_ctrl(ctrl);
4580 kfree_const(ctrl->device->kobj.name);
4581 out_release_instance:
4582 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4583 out:
4584 if (ctrl->discard_page)
4585 __free_page(ctrl->discard_page);
4586 return ret;
4587 }
4588 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4589
4590 /**
4591 * nvme_kill_queues(): Ends all namespace queues
4592 * @ctrl: the dead controller that needs to end
4593 *
4594 * Call this function when the driver determines it is unable to get the
4595 * controller in a state capable of servicing IO.
4596 */
4597 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4598 {
4599 struct nvme_ns *ns;
4600
4601 down_read(&ctrl->namespaces_rwsem);
4602
4603 /* Forcibly unquiesce queues to avoid blocking dispatch */
4604 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4605 blk_mq_unquiesce_queue(ctrl->admin_q);
4606
4607 list_for_each_entry(ns, &ctrl->namespaces, list)
4608 nvme_set_queue_dying(ns);
4609
4610 up_read(&ctrl->namespaces_rwsem);
4611 }
4612 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4613
4614 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4615 {
4616 struct nvme_ns *ns;
4617
4618 down_read(&ctrl->namespaces_rwsem);
4619 list_for_each_entry(ns, &ctrl->namespaces, list)
4620 blk_mq_unfreeze_queue(ns->queue);
4621 up_read(&ctrl->namespaces_rwsem);
4622 }
4623 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4624
4625 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4626 {
4627 struct nvme_ns *ns;
4628
4629 down_read(&ctrl->namespaces_rwsem);
4630 list_for_each_entry(ns, &ctrl->namespaces, list) {
4631 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4632 if (timeout <= 0)
4633 break;
4634 }
4635 up_read(&ctrl->namespaces_rwsem);
4636 return timeout;
4637 }
4638 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4639
4640 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4641 {
4642 struct nvme_ns *ns;
4643
4644 down_read(&ctrl->namespaces_rwsem);
4645 list_for_each_entry(ns, &ctrl->namespaces, list)
4646 blk_mq_freeze_queue_wait(ns->queue);
4647 up_read(&ctrl->namespaces_rwsem);
4648 }
4649 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4650
4651 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4652 {
4653 struct nvme_ns *ns;
4654
4655 down_read(&ctrl->namespaces_rwsem);
4656 list_for_each_entry(ns, &ctrl->namespaces, list)
4657 blk_freeze_queue_start(ns->queue);
4658 up_read(&ctrl->namespaces_rwsem);
4659 }
4660 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4661
4662 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4663 {
4664 struct nvme_ns *ns;
4665
4666 down_read(&ctrl->namespaces_rwsem);
4667 list_for_each_entry(ns, &ctrl->namespaces, list)
4668 blk_mq_quiesce_queue(ns->queue);
4669 up_read(&ctrl->namespaces_rwsem);
4670 }
4671 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4672
4673 void nvme_start_queues(struct nvme_ctrl *ctrl)
4674 {
4675 struct nvme_ns *ns;
4676
4677 down_read(&ctrl->namespaces_rwsem);
4678 list_for_each_entry(ns, &ctrl->namespaces, list)
4679 blk_mq_unquiesce_queue(ns->queue);
4680 up_read(&ctrl->namespaces_rwsem);
4681 }
4682 EXPORT_SYMBOL_GPL(nvme_start_queues);
4683
4684 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4685 {
4686 struct nvme_ns *ns;
4687
4688 down_read(&ctrl->namespaces_rwsem);
4689 list_for_each_entry(ns, &ctrl->namespaces, list)
4690 blk_sync_queue(ns->queue);
4691 up_read(&ctrl->namespaces_rwsem);
4692 }
4693 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4694
4695 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4696 {
4697 nvme_sync_io_queues(ctrl);
4698 if (ctrl->admin_q)
4699 blk_sync_queue(ctrl->admin_q);
4700 }
4701 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4702
4703 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4704 {
4705 if (file->f_op != &nvme_dev_fops)
4706 return NULL;
4707 return file->private_data;
4708 }
4709 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4710
4711 /*
4712 * Check we didn't inadvertently grow the command structure sizes:
4713 */
4714 static inline void _nvme_check_size(void)
4715 {
4716 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4717 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4718 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4719 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4720 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4721 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4722 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4723 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4724 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4725 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4726 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4727 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4728 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4729 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4730 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4731 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4732 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4733 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4734 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4735 }
4736
4737
4738 static int __init nvme_core_init(void)
4739 {
4740 int result = -ENOMEM;
4741
4742 _nvme_check_size();
4743
4744 nvme_wq = alloc_workqueue("nvme-wq",
4745 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4746 if (!nvme_wq)
4747 goto out;
4748
4749 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4750 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4751 if (!nvme_reset_wq)
4752 goto destroy_wq;
4753
4754 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4755 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4756 if (!nvme_delete_wq)
4757 goto destroy_reset_wq;
4758
4759 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4760 NVME_MINORS, "nvme");
4761 if (result < 0)
4762 goto destroy_delete_wq;
4763
4764 nvme_class = class_create(THIS_MODULE, "nvme");
4765 if (IS_ERR(nvme_class)) {
4766 result = PTR_ERR(nvme_class);
4767 goto unregister_chrdev;
4768 }
4769 nvme_class->dev_uevent = nvme_class_uevent;
4770
4771 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4772 if (IS_ERR(nvme_subsys_class)) {
4773 result = PTR_ERR(nvme_subsys_class);
4774 goto destroy_class;
4775 }
4776 return 0;
4777
4778 destroy_class:
4779 class_destroy(nvme_class);
4780 unregister_chrdev:
4781 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4782 destroy_delete_wq:
4783 destroy_workqueue(nvme_delete_wq);
4784 destroy_reset_wq:
4785 destroy_workqueue(nvme_reset_wq);
4786 destroy_wq:
4787 destroy_workqueue(nvme_wq);
4788 out:
4789 return result;
4790 }
4791
4792 static void __exit nvme_core_exit(void)
4793 {
4794 class_destroy(nvme_subsys_class);
4795 class_destroy(nvme_class);
4796 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4797 destroy_workqueue(nvme_delete_wq);
4798 destroy_workqueue(nvme_reset_wq);
4799 destroy_workqueue(nvme_wq);
4800 ida_destroy(&nvme_instance_ida);
4801 }
4802
4803 MODULE_LICENSE("GPL");
4804 MODULE_VERSION("1.0");
4805 module_init(nvme_core_init);
4806 module_exit(nvme_core_exit);
Ubuntu Hirsute Linux kernel mirror