Merge branch 'for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck...
[sfrench/cifs-2.6.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/delay.h>
10 #include <linux/errno.h>
11 #include <linux/hdreg.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/list_sort.h>
15 #include <linux/slab.h>
16 #include <linux/types.h>
17 #include <linux/pr.h>
18 #include <linux/ptrace.h>
19 #include <linux/nvme_ioctl.h>
20 #include <linux/t10-pi.h>
21 #include <linux/pm_qos.h>
22 #include <asm/unaligned.h>
23
24 #define CREATE_TRACE_POINTS
25 #include "trace.h"
26
27 #include "nvme.h"
28 #include "fabrics.h"
29
30 #define NVME_MINORS             (1U << MINORBITS)
31
32 unsigned int admin_timeout = 60;
33 module_param(admin_timeout, uint, 0644);
34 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
35 EXPORT_SYMBOL_GPL(admin_timeout);
36
37 unsigned int nvme_io_timeout = 30;
38 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
39 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
40 EXPORT_SYMBOL_GPL(nvme_io_timeout);
41
42 static unsigned char shutdown_timeout = 5;
43 module_param(shutdown_timeout, byte, 0644);
44 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
45
46 static u8 nvme_max_retries = 5;
47 module_param_named(max_retries, nvme_max_retries, byte, 0644);
48 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
49
50 static unsigned long default_ps_max_latency_us = 100000;
51 module_param(default_ps_max_latency_us, ulong, 0644);
52 MODULE_PARM_DESC(default_ps_max_latency_us,
53                  "max power saving latency for new devices; use PM QOS to change per device");
54
55 static bool force_apst;
56 module_param(force_apst, bool, 0644);
57 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
58
59 static bool streams;
60 module_param(streams, bool, 0644);
61 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
62
63 /*
64  * nvme_wq - hosts nvme related works that are not reset or delete
65  * nvme_reset_wq - hosts nvme reset works
66  * nvme_delete_wq - hosts nvme delete works
67  *
68  * nvme_wq will host works such are scan, aen handling, fw activation,
69  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
70  * runs reset works which also flush works hosted on nvme_wq for
71  * serialization purposes. nvme_delete_wq host controller deletion
72  * works which flush reset works for serialization.
73  */
74 struct workqueue_struct *nvme_wq;
75 EXPORT_SYMBOL_GPL(nvme_wq);
76
77 struct workqueue_struct *nvme_reset_wq;
78 EXPORT_SYMBOL_GPL(nvme_reset_wq);
79
80 struct workqueue_struct *nvme_delete_wq;
81 EXPORT_SYMBOL_GPL(nvme_delete_wq);
82
83 static DEFINE_IDA(nvme_subsystems_ida);
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_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
91
92 static int nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
95                                            unsigned nsid);
96
97 static void nvme_set_queue_dying(struct nvme_ns *ns)
98 {
99         /*
100          * Revalidating a dead namespace sets capacity to 0. This will end
101          * buffered writers dirtying pages that can't be synced.
102          */
103         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
104                 return;
105         revalidate_disk(ns->disk);
106         blk_set_queue_dying(ns->queue);
107         /* Forcibly unquiesce queues to avoid blocking dispatch */
108         blk_mq_unquiesce_queue(ns->queue);
109 }
110
111 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
112 {
113         /*
114          * Only new queue scan work when admin and IO queues are both alive
115          */
116         if (ctrl->state == NVME_CTRL_LIVE)
117                 queue_work(nvme_wq, &ctrl->scan_work);
118 }
119
120 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
121 {
122         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
123                 return -EBUSY;
124         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
125                 return -EBUSY;
126         return 0;
127 }
128 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
129
130 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
131 {
132         int ret;
133
134         ret = nvme_reset_ctrl(ctrl);
135         if (!ret) {
136                 flush_work(&ctrl->reset_work);
137                 if (ctrl->state != NVME_CTRL_LIVE &&
138                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
139                         ret = -ENETRESET;
140         }
141
142         return ret;
143 }
144 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
145
146 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
147 {
148         dev_info(ctrl->device,
149                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
150
151         flush_work(&ctrl->reset_work);
152         nvme_stop_ctrl(ctrl);
153         nvme_remove_namespaces(ctrl);
154         ctrl->ops->delete_ctrl(ctrl);
155         nvme_uninit_ctrl(ctrl);
156         nvme_put_ctrl(ctrl);
157 }
158
159 static void nvme_delete_ctrl_work(struct work_struct *work)
160 {
161         struct nvme_ctrl *ctrl =
162                 container_of(work, struct nvme_ctrl, delete_work);
163
164         nvme_do_delete_ctrl(ctrl);
165 }
166
167 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
168 {
169         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
170                 return -EBUSY;
171         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
172                 return -EBUSY;
173         return 0;
174 }
175 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
176
177 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
178 {
179         int ret = 0;
180
181         /*
182          * Keep a reference until nvme_do_delete_ctrl() complete,
183          * since ->delete_ctrl can free the controller.
184          */
185         nvme_get_ctrl(ctrl);
186         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
187                 ret = -EBUSY;
188         if (!ret)
189                 nvme_do_delete_ctrl(ctrl);
190         nvme_put_ctrl(ctrl);
191         return ret;
192 }
193
194 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
195 {
196         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
197 }
198
199 static blk_status_t nvme_error_status(struct request *req)
200 {
201         switch (nvme_req(req)->status & 0x7ff) {
202         case NVME_SC_SUCCESS:
203                 return BLK_STS_OK;
204         case NVME_SC_CAP_EXCEEDED:
205                 return BLK_STS_NOSPC;
206         case NVME_SC_LBA_RANGE:
207                 return BLK_STS_TARGET;
208         case NVME_SC_BAD_ATTRIBUTES:
209         case NVME_SC_ONCS_NOT_SUPPORTED:
210         case NVME_SC_INVALID_OPCODE:
211         case NVME_SC_INVALID_FIELD:
212         case NVME_SC_INVALID_NS:
213                 return BLK_STS_NOTSUPP;
214         case NVME_SC_WRITE_FAULT:
215         case NVME_SC_READ_ERROR:
216         case NVME_SC_UNWRITTEN_BLOCK:
217         case NVME_SC_ACCESS_DENIED:
218         case NVME_SC_READ_ONLY:
219         case NVME_SC_COMPARE_FAILED:
220                 return BLK_STS_MEDIUM;
221         case NVME_SC_GUARD_CHECK:
222         case NVME_SC_APPTAG_CHECK:
223         case NVME_SC_REFTAG_CHECK:
224         case NVME_SC_INVALID_PI:
225                 return BLK_STS_PROTECTION;
226         case NVME_SC_RESERVATION_CONFLICT:
227                 return BLK_STS_NEXUS;
228         default:
229                 return BLK_STS_IOERR;
230         }
231 }
232
233 static inline bool nvme_req_needs_retry(struct request *req)
234 {
235         if (blk_noretry_request(req))
236                 return false;
237         if (nvme_req(req)->status & NVME_SC_DNR)
238                 return false;
239         if (nvme_req(req)->retries >= nvme_max_retries)
240                 return false;
241         return true;
242 }
243
244 static void nvme_retry_req(struct request *req)
245 {
246         struct nvme_ns *ns = req->q->queuedata;
247         unsigned long delay = 0;
248         u16 crd;
249
250         /* The mask and shift result must be <= 3 */
251         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
252         if (ns && crd)
253                 delay = ns->ctrl->crdt[crd - 1] * 100;
254
255         nvme_req(req)->retries++;
256         blk_mq_requeue_request(req, false);
257         blk_mq_delay_kick_requeue_list(req->q, delay);
258 }
259
260 void nvme_complete_rq(struct request *req)
261 {
262         blk_status_t status = nvme_error_status(req);
263
264         trace_nvme_complete_rq(req);
265
266         if (nvme_req(req)->ctrl->kas)
267                 nvme_req(req)->ctrl->comp_seen = true;
268
269         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
270                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
271                     blk_path_error(status)) {
272                         nvme_failover_req(req);
273                         return;
274                 }
275
276                 if (!blk_queue_dying(req->q)) {
277                         nvme_retry_req(req);
278                         return;
279                 }
280         }
281         blk_mq_end_request(req, status);
282 }
283 EXPORT_SYMBOL_GPL(nvme_complete_rq);
284
285 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
286 {
287         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
288                                 "Cancelling I/O %d", req->tag);
289
290         nvme_req(req)->status = NVME_SC_ABORT_REQ;
291         blk_mq_complete_request_sync(req);
292         return true;
293 }
294 EXPORT_SYMBOL_GPL(nvme_cancel_request);
295
296 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
297                 enum nvme_ctrl_state new_state)
298 {
299         enum nvme_ctrl_state old_state;
300         unsigned long flags;
301         bool changed = false;
302
303         spin_lock_irqsave(&ctrl->lock, flags);
304
305         old_state = ctrl->state;
306         switch (new_state) {
307         case NVME_CTRL_ADMIN_ONLY:
308                 switch (old_state) {
309                 case NVME_CTRL_CONNECTING:
310                         changed = true;
311                         /* FALLTHRU */
312                 default:
313                         break;
314                 }
315                 break;
316         case NVME_CTRL_LIVE:
317                 switch (old_state) {
318                 case NVME_CTRL_NEW:
319                 case NVME_CTRL_RESETTING:
320                 case NVME_CTRL_CONNECTING:
321                         changed = true;
322                         /* FALLTHRU */
323                 default:
324                         break;
325                 }
326                 break;
327         case NVME_CTRL_RESETTING:
328                 switch (old_state) {
329                 case NVME_CTRL_NEW:
330                 case NVME_CTRL_LIVE:
331                 case NVME_CTRL_ADMIN_ONLY:
332                         changed = true;
333                         /* FALLTHRU */
334                 default:
335                         break;
336                 }
337                 break;
338         case NVME_CTRL_CONNECTING:
339                 switch (old_state) {
340                 case NVME_CTRL_NEW:
341                 case NVME_CTRL_RESETTING:
342                         changed = true;
343                         /* FALLTHRU */
344                 default:
345                         break;
346                 }
347                 break;
348         case NVME_CTRL_DELETING:
349                 switch (old_state) {
350                 case NVME_CTRL_LIVE:
351                 case NVME_CTRL_ADMIN_ONLY:
352                 case NVME_CTRL_RESETTING:
353                 case NVME_CTRL_CONNECTING:
354                         changed = true;
355                         /* FALLTHRU */
356                 default:
357                         break;
358                 }
359                 break;
360         case NVME_CTRL_DEAD:
361                 switch (old_state) {
362                 case NVME_CTRL_DELETING:
363                         changed = true;
364                         /* FALLTHRU */
365                 default:
366                         break;
367                 }
368                 break;
369         default:
370                 break;
371         }
372
373         if (changed)
374                 ctrl->state = new_state;
375
376         spin_unlock_irqrestore(&ctrl->lock, flags);
377         if (changed && ctrl->state == NVME_CTRL_LIVE)
378                 nvme_kick_requeue_lists(ctrl);
379         return changed;
380 }
381 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
382
383 static void nvme_free_ns_head(struct kref *ref)
384 {
385         struct nvme_ns_head *head =
386                 container_of(ref, struct nvme_ns_head, ref);
387
388         nvme_mpath_remove_disk(head);
389         ida_simple_remove(&head->subsys->ns_ida, head->instance);
390         list_del_init(&head->entry);
391         cleanup_srcu_struct(&head->srcu);
392         nvme_put_subsystem(head->subsys);
393         kfree(head);
394 }
395
396 static void nvme_put_ns_head(struct nvme_ns_head *head)
397 {
398         kref_put(&head->ref, nvme_free_ns_head);
399 }
400
401 static void nvme_free_ns(struct kref *kref)
402 {
403         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
404
405         if (ns->ndev)
406                 nvme_nvm_unregister(ns);
407
408         put_disk(ns->disk);
409         nvme_put_ns_head(ns->head);
410         nvme_put_ctrl(ns->ctrl);
411         kfree(ns);
412 }
413
414 static void nvme_put_ns(struct nvme_ns *ns)
415 {
416         kref_put(&ns->kref, nvme_free_ns);
417 }
418
419 static inline void nvme_clear_nvme_request(struct request *req)
420 {
421         if (!(req->rq_flags & RQF_DONTPREP)) {
422                 nvme_req(req)->retries = 0;
423                 nvme_req(req)->flags = 0;
424                 req->rq_flags |= RQF_DONTPREP;
425         }
426 }
427
428 struct request *nvme_alloc_request(struct request_queue *q,
429                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
430 {
431         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
432         struct request *req;
433
434         if (qid == NVME_QID_ANY) {
435                 req = blk_mq_alloc_request(q, op, flags);
436         } else {
437                 req = blk_mq_alloc_request_hctx(q, op, flags,
438                                 qid ? qid - 1 : 0);
439         }
440         if (IS_ERR(req))
441                 return req;
442
443         req->cmd_flags |= REQ_FAILFAST_DRIVER;
444         nvme_clear_nvme_request(req);
445         nvme_req(req)->cmd = cmd;
446
447         return req;
448 }
449 EXPORT_SYMBOL_GPL(nvme_alloc_request);
450
451 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
452 {
453         struct nvme_command c;
454
455         memset(&c, 0, sizeof(c));
456
457         c.directive.opcode = nvme_admin_directive_send;
458         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
459         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
460         c.directive.dtype = NVME_DIR_IDENTIFY;
461         c.directive.tdtype = NVME_DIR_STREAMS;
462         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
463
464         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
465 }
466
467 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
468 {
469         return nvme_toggle_streams(ctrl, false);
470 }
471
472 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
473 {
474         return nvme_toggle_streams(ctrl, true);
475 }
476
477 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
478                                   struct streams_directive_params *s, u32 nsid)
479 {
480         struct nvme_command c;
481
482         memset(&c, 0, sizeof(c));
483         memset(s, 0, sizeof(*s));
484
485         c.directive.opcode = nvme_admin_directive_recv;
486         c.directive.nsid = cpu_to_le32(nsid);
487         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
488         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
489         c.directive.dtype = NVME_DIR_STREAMS;
490
491         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
492 }
493
494 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
495 {
496         struct streams_directive_params s;
497         int ret;
498
499         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
500                 return 0;
501         if (!streams)
502                 return 0;
503
504         ret = nvme_enable_streams(ctrl);
505         if (ret)
506                 return ret;
507
508         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
509         if (ret)
510                 return ret;
511
512         ctrl->nssa = le16_to_cpu(s.nssa);
513         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
514                 dev_info(ctrl->device, "too few streams (%u) available\n",
515                                         ctrl->nssa);
516                 nvme_disable_streams(ctrl);
517                 return 0;
518         }
519
520         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
521         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
522         return 0;
523 }
524
525 /*
526  * Check if 'req' has a write hint associated with it. If it does, assign
527  * a valid namespace stream to the write.
528  */
529 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
530                                      struct request *req, u16 *control,
531                                      u32 *dsmgmt)
532 {
533         enum rw_hint streamid = req->write_hint;
534
535         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
536                 streamid = 0;
537         else {
538                 streamid--;
539                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
540                         return;
541
542                 *control |= NVME_RW_DTYPE_STREAMS;
543                 *dsmgmt |= streamid << 16;
544         }
545
546         if (streamid < ARRAY_SIZE(req->q->write_hints))
547                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
548 }
549
550 static inline void nvme_setup_flush(struct nvme_ns *ns,
551                 struct nvme_command *cmnd)
552 {
553         cmnd->common.opcode = nvme_cmd_flush;
554         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
555 }
556
557 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
558                 struct nvme_command *cmnd)
559 {
560         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
561         struct nvme_dsm_range *range;
562         struct bio *bio;
563
564         range = kmalloc_array(segments, sizeof(*range),
565                                 GFP_ATOMIC | __GFP_NOWARN);
566         if (!range) {
567                 /*
568                  * If we fail allocation our range, fallback to the controller
569                  * discard page. If that's also busy, it's safe to return
570                  * busy, as we know we can make progress once that's freed.
571                  */
572                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
573                         return BLK_STS_RESOURCE;
574
575                 range = page_address(ns->ctrl->discard_page);
576         }
577
578         __rq_for_each_bio(bio, req) {
579                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
580                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
581
582                 if (n < segments) {
583                         range[n].cattr = cpu_to_le32(0);
584                         range[n].nlb = cpu_to_le32(nlb);
585                         range[n].slba = cpu_to_le64(slba);
586                 }
587                 n++;
588         }
589
590         if (WARN_ON_ONCE(n != segments)) {
591                 if (virt_to_page(range) == ns->ctrl->discard_page)
592                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
593                 else
594                         kfree(range);
595                 return BLK_STS_IOERR;
596         }
597
598         cmnd->dsm.opcode = nvme_cmd_dsm;
599         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
600         cmnd->dsm.nr = cpu_to_le32(segments - 1);
601         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
602
603         req->special_vec.bv_page = virt_to_page(range);
604         req->special_vec.bv_offset = offset_in_page(range);
605         req->special_vec.bv_len = sizeof(*range) * segments;
606         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
607
608         return BLK_STS_OK;
609 }
610
611 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
612                 struct request *req, struct nvme_command *cmnd)
613 {
614         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
615                 return nvme_setup_discard(ns, req, cmnd);
616
617         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
618         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
619         cmnd->write_zeroes.slba =
620                 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
621         cmnd->write_zeroes.length =
622                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
623         cmnd->write_zeroes.control = 0;
624         return BLK_STS_OK;
625 }
626
627 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
628                 struct request *req, struct nvme_command *cmnd)
629 {
630         struct nvme_ctrl *ctrl = ns->ctrl;
631         u16 control = 0;
632         u32 dsmgmt = 0;
633
634         if (req->cmd_flags & REQ_FUA)
635                 control |= NVME_RW_FUA;
636         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
637                 control |= NVME_RW_LR;
638
639         if (req->cmd_flags & REQ_RAHEAD)
640                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
641
642         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
643         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
644         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
645         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
646
647         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
648                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
649
650         if (ns->ms) {
651                 /*
652                  * If formated with metadata, the block layer always provides a
653                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
654                  * we enable the PRACT bit for protection information or set the
655                  * namespace capacity to zero to prevent any I/O.
656                  */
657                 if (!blk_integrity_rq(req)) {
658                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
659                                 return BLK_STS_NOTSUPP;
660                         control |= NVME_RW_PRINFO_PRACT;
661                 } else if (req_op(req) == REQ_OP_WRITE) {
662                         t10_pi_prepare(req, ns->pi_type);
663                 }
664
665                 switch (ns->pi_type) {
666                 case NVME_NS_DPS_PI_TYPE3:
667                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
668                         break;
669                 case NVME_NS_DPS_PI_TYPE1:
670                 case NVME_NS_DPS_PI_TYPE2:
671                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
672                                         NVME_RW_PRINFO_PRCHK_REF;
673                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
674                         break;
675                 }
676         }
677
678         cmnd->rw.control = cpu_to_le16(control);
679         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
680         return 0;
681 }
682
683 void nvme_cleanup_cmd(struct request *req)
684 {
685         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
686             nvme_req(req)->status == 0) {
687                 struct nvme_ns *ns = req->rq_disk->private_data;
688
689                 t10_pi_complete(req, ns->pi_type,
690                                 blk_rq_bytes(req) >> ns->lba_shift);
691         }
692         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
693                 struct nvme_ns *ns = req->rq_disk->private_data;
694                 struct page *page = req->special_vec.bv_page;
695
696                 if (page == ns->ctrl->discard_page)
697                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
698                 else
699                         kfree(page_address(page) + req->special_vec.bv_offset);
700         }
701 }
702 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
703
704 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
705                 struct nvme_command *cmd)
706 {
707         blk_status_t ret = BLK_STS_OK;
708
709         nvme_clear_nvme_request(req);
710
711         memset(cmd, 0, sizeof(*cmd));
712         switch (req_op(req)) {
713         case REQ_OP_DRV_IN:
714         case REQ_OP_DRV_OUT:
715                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
716                 break;
717         case REQ_OP_FLUSH:
718                 nvme_setup_flush(ns, cmd);
719                 break;
720         case REQ_OP_WRITE_ZEROES:
721                 ret = nvme_setup_write_zeroes(ns, req, cmd);
722                 break;
723         case REQ_OP_DISCARD:
724                 ret = nvme_setup_discard(ns, req, cmd);
725                 break;
726         case REQ_OP_READ:
727         case REQ_OP_WRITE:
728                 ret = nvme_setup_rw(ns, req, cmd);
729                 break;
730         default:
731                 WARN_ON_ONCE(1);
732                 return BLK_STS_IOERR;
733         }
734
735         cmd->common.command_id = req->tag;
736         trace_nvme_setup_cmd(req, cmd);
737         return ret;
738 }
739 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
740
741 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
742 {
743         struct completion *waiting = rq->end_io_data;
744
745         rq->end_io_data = NULL;
746         complete(waiting);
747 }
748
749 static void nvme_execute_rq_polled(struct request_queue *q,
750                 struct gendisk *bd_disk, struct request *rq, int at_head)
751 {
752         DECLARE_COMPLETION_ONSTACK(wait);
753
754         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
755
756         rq->cmd_flags |= REQ_HIPRI;
757         rq->end_io_data = &wait;
758         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
759
760         while (!completion_done(&wait)) {
761                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
762                 cond_resched();
763         }
764 }
765
766 /*
767  * Returns 0 on success.  If the result is negative, it's a Linux error code;
768  * if the result is positive, it's an NVM Express status code
769  */
770 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
771                 union nvme_result *result, void *buffer, unsigned bufflen,
772                 unsigned timeout, int qid, int at_head,
773                 blk_mq_req_flags_t flags, bool poll)
774 {
775         struct request *req;
776         int ret;
777
778         req = nvme_alloc_request(q, cmd, flags, qid);
779         if (IS_ERR(req))
780                 return PTR_ERR(req);
781
782         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
783
784         if (buffer && bufflen) {
785                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
786                 if (ret)
787                         goto out;
788         }
789
790         if (poll)
791                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
792         else
793                 blk_execute_rq(req->q, NULL, req, at_head);
794         if (result)
795                 *result = nvme_req(req)->result;
796         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
797                 ret = -EINTR;
798         else
799                 ret = nvme_req(req)->status;
800  out:
801         blk_mq_free_request(req);
802         return ret;
803 }
804 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
805
806 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
807                 void *buffer, unsigned bufflen)
808 {
809         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
810                         NVME_QID_ANY, 0, 0, false);
811 }
812 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
813
814 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
815                 unsigned len, u32 seed, bool write)
816 {
817         struct bio_integrity_payload *bip;
818         int ret = -ENOMEM;
819         void *buf;
820
821         buf = kmalloc(len, GFP_KERNEL);
822         if (!buf)
823                 goto out;
824
825         ret = -EFAULT;
826         if (write && copy_from_user(buf, ubuf, len))
827                 goto out_free_meta;
828
829         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
830         if (IS_ERR(bip)) {
831                 ret = PTR_ERR(bip);
832                 goto out_free_meta;
833         }
834
835         bip->bip_iter.bi_size = len;
836         bip->bip_iter.bi_sector = seed;
837         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
838                         offset_in_page(buf));
839         if (ret == len)
840                 return buf;
841         ret = -ENOMEM;
842 out_free_meta:
843         kfree(buf);
844 out:
845         return ERR_PTR(ret);
846 }
847
848 static int nvme_submit_user_cmd(struct request_queue *q,
849                 struct nvme_command *cmd, void __user *ubuffer,
850                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
851                 u32 meta_seed, u32 *result, unsigned timeout)
852 {
853         bool write = nvme_is_write(cmd);
854         struct nvme_ns *ns = q->queuedata;
855         struct gendisk *disk = ns ? ns->disk : NULL;
856         struct request *req;
857         struct bio *bio = NULL;
858         void *meta = NULL;
859         int ret;
860
861         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
862         if (IS_ERR(req))
863                 return PTR_ERR(req);
864
865         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
866         nvme_req(req)->flags |= NVME_REQ_USERCMD;
867
868         if (ubuffer && bufflen) {
869                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
870                                 GFP_KERNEL);
871                 if (ret)
872                         goto out;
873                 bio = req->bio;
874                 bio->bi_disk = disk;
875                 if (disk && meta_buffer && meta_len) {
876                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
877                                         meta_seed, write);
878                         if (IS_ERR(meta)) {
879                                 ret = PTR_ERR(meta);
880                                 goto out_unmap;
881                         }
882                         req->cmd_flags |= REQ_INTEGRITY;
883                 }
884         }
885
886         blk_execute_rq(req->q, disk, req, 0);
887         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
888                 ret = -EINTR;
889         else
890                 ret = nvme_req(req)->status;
891         if (result)
892                 *result = le32_to_cpu(nvme_req(req)->result.u32);
893         if (meta && !ret && !write) {
894                 if (copy_to_user(meta_buffer, meta, meta_len))
895                         ret = -EFAULT;
896         }
897         kfree(meta);
898  out_unmap:
899         if (bio)
900                 blk_rq_unmap_user(bio);
901  out:
902         blk_mq_free_request(req);
903         return ret;
904 }
905
906 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
907 {
908         struct nvme_ctrl *ctrl = rq->end_io_data;
909         unsigned long flags;
910         bool startka = false;
911
912         blk_mq_free_request(rq);
913
914         if (status) {
915                 dev_err(ctrl->device,
916                         "failed nvme_keep_alive_end_io error=%d\n",
917                                 status);
918                 return;
919         }
920
921         ctrl->comp_seen = false;
922         spin_lock_irqsave(&ctrl->lock, flags);
923         if (ctrl->state == NVME_CTRL_LIVE ||
924             ctrl->state == NVME_CTRL_CONNECTING)
925                 startka = true;
926         spin_unlock_irqrestore(&ctrl->lock, flags);
927         if (startka)
928                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
929 }
930
931 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
932 {
933         struct request *rq;
934
935         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
936                         NVME_QID_ANY);
937         if (IS_ERR(rq))
938                 return PTR_ERR(rq);
939
940         rq->timeout = ctrl->kato * HZ;
941         rq->end_io_data = ctrl;
942
943         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
944
945         return 0;
946 }
947
948 static void nvme_keep_alive_work(struct work_struct *work)
949 {
950         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
951                         struct nvme_ctrl, ka_work);
952         bool comp_seen = ctrl->comp_seen;
953
954         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
955                 dev_dbg(ctrl->device,
956                         "reschedule traffic based keep-alive timer\n");
957                 ctrl->comp_seen = false;
958                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
959                 return;
960         }
961
962         if (nvme_keep_alive(ctrl)) {
963                 /* allocation failure, reset the controller */
964                 dev_err(ctrl->device, "keep-alive failed\n");
965                 nvme_reset_ctrl(ctrl);
966                 return;
967         }
968 }
969
970 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
971 {
972         if (unlikely(ctrl->kato == 0))
973                 return;
974
975         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
976 }
977
978 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
979 {
980         if (unlikely(ctrl->kato == 0))
981                 return;
982
983         cancel_delayed_work_sync(&ctrl->ka_work);
984 }
985 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
986
987 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
988 {
989         struct nvme_command c = { };
990         int error;
991
992         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
993         c.identify.opcode = nvme_admin_identify;
994         c.identify.cns = NVME_ID_CNS_CTRL;
995
996         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
997         if (!*id)
998                 return -ENOMEM;
999
1000         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1001                         sizeof(struct nvme_id_ctrl));
1002         if (error)
1003                 kfree(*id);
1004         return error;
1005 }
1006
1007 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1008                 struct nvme_ns_ids *ids)
1009 {
1010         struct nvme_command c = { };
1011         int status;
1012         void *data;
1013         int pos;
1014         int len;
1015
1016         c.identify.opcode = nvme_admin_identify;
1017         c.identify.nsid = cpu_to_le32(nsid);
1018         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1019
1020         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1021         if (!data)
1022                 return -ENOMEM;
1023
1024         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1025                                       NVME_IDENTIFY_DATA_SIZE);
1026         if (status)
1027                 goto free_data;
1028
1029         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1030                 struct nvme_ns_id_desc *cur = data + pos;
1031
1032                 if (cur->nidl == 0)
1033                         break;
1034
1035                 switch (cur->nidt) {
1036                 case NVME_NIDT_EUI64:
1037                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1038                                 dev_warn(ctrl->device,
1039                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1040                                          cur->nidl);
1041                                 goto free_data;
1042                         }
1043                         len = NVME_NIDT_EUI64_LEN;
1044                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1045                         break;
1046                 case NVME_NIDT_NGUID:
1047                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1048                                 dev_warn(ctrl->device,
1049                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1050                                          cur->nidl);
1051                                 goto free_data;
1052                         }
1053                         len = NVME_NIDT_NGUID_LEN;
1054                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1055                         break;
1056                 case NVME_NIDT_UUID:
1057                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1058                                 dev_warn(ctrl->device,
1059                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1060                                          cur->nidl);
1061                                 goto free_data;
1062                         }
1063                         len = NVME_NIDT_UUID_LEN;
1064                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1065                         break;
1066                 default:
1067                         /* Skip unknown types */
1068                         len = cur->nidl;
1069                         break;
1070                 }
1071
1072                 len += sizeof(*cur);
1073         }
1074 free_data:
1075         kfree(data);
1076         return status;
1077 }
1078
1079 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1080 {
1081         struct nvme_command c = { };
1082
1083         c.identify.opcode = nvme_admin_identify;
1084         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1085         c.identify.nsid = cpu_to_le32(nsid);
1086         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1087                                     NVME_IDENTIFY_DATA_SIZE);
1088 }
1089
1090 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1091                 unsigned nsid)
1092 {
1093         struct nvme_id_ns *id;
1094         struct nvme_command c = { };
1095         int error;
1096
1097         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1098         c.identify.opcode = nvme_admin_identify;
1099         c.identify.nsid = cpu_to_le32(nsid);
1100         c.identify.cns = NVME_ID_CNS_NS;
1101
1102         id = kmalloc(sizeof(*id), GFP_KERNEL);
1103         if (!id)
1104                 return NULL;
1105
1106         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1107         if (error) {
1108                 dev_warn(ctrl->device, "Identify namespace failed\n");
1109                 kfree(id);
1110                 return NULL;
1111         }
1112
1113         return id;
1114 }
1115
1116 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1117                       void *buffer, size_t buflen, u32 *result)
1118 {
1119         struct nvme_command c;
1120         union nvme_result res;
1121         int ret;
1122
1123         memset(&c, 0, sizeof(c));
1124         c.features.opcode = nvme_admin_set_features;
1125         c.features.fid = cpu_to_le32(fid);
1126         c.features.dword11 = cpu_to_le32(dword11);
1127
1128         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1129                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1130         if (ret >= 0 && result)
1131                 *result = le32_to_cpu(res.u32);
1132         return ret;
1133 }
1134
1135 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1136 {
1137         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1138         u32 result;
1139         int status, nr_io_queues;
1140
1141         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1142                         &result);
1143         if (status < 0)
1144                 return status;
1145
1146         /*
1147          * Degraded controllers might return an error when setting the queue
1148          * count.  We still want to be able to bring them online and offer
1149          * access to the admin queue, as that might be only way to fix them up.
1150          */
1151         if (status > 0) {
1152                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1153                 *count = 0;
1154         } else {
1155                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1156                 *count = min(*count, nr_io_queues);
1157         }
1158
1159         return 0;
1160 }
1161 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1162
1163 #define NVME_AEN_SUPPORTED \
1164         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1165
1166 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1167 {
1168         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1169         int status;
1170
1171         if (!supported_aens)
1172                 return;
1173
1174         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1175                         NULL, 0, &result);
1176         if (status)
1177                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1178                          supported_aens);
1179 }
1180
1181 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1182 {
1183         struct nvme_user_io io;
1184         struct nvme_command c;
1185         unsigned length, meta_len;
1186         void __user *metadata;
1187
1188         if (copy_from_user(&io, uio, sizeof(io)))
1189                 return -EFAULT;
1190         if (io.flags)
1191                 return -EINVAL;
1192
1193         switch (io.opcode) {
1194         case nvme_cmd_write:
1195         case nvme_cmd_read:
1196         case nvme_cmd_compare:
1197                 break;
1198         default:
1199                 return -EINVAL;
1200         }
1201
1202         length = (io.nblocks + 1) << ns->lba_shift;
1203         meta_len = (io.nblocks + 1) * ns->ms;
1204         metadata = (void __user *)(uintptr_t)io.metadata;
1205
1206         if (ns->ext) {
1207                 length += meta_len;
1208                 meta_len = 0;
1209         } else if (meta_len) {
1210                 if ((io.metadata & 3) || !io.metadata)
1211                         return -EINVAL;
1212         }
1213
1214         memset(&c, 0, sizeof(c));
1215         c.rw.opcode = io.opcode;
1216         c.rw.flags = io.flags;
1217         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1218         c.rw.slba = cpu_to_le64(io.slba);
1219         c.rw.length = cpu_to_le16(io.nblocks);
1220         c.rw.control = cpu_to_le16(io.control);
1221         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1222         c.rw.reftag = cpu_to_le32(io.reftag);
1223         c.rw.apptag = cpu_to_le16(io.apptag);
1224         c.rw.appmask = cpu_to_le16(io.appmask);
1225
1226         return nvme_submit_user_cmd(ns->queue, &c,
1227                         (void __user *)(uintptr_t)io.addr, length,
1228                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1229 }
1230
1231 static u32 nvme_known_admin_effects(u8 opcode)
1232 {
1233         switch (opcode) {
1234         case nvme_admin_format_nvm:
1235                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1236                                         NVME_CMD_EFFECTS_CSE_MASK;
1237         case nvme_admin_sanitize_nvm:
1238                 return NVME_CMD_EFFECTS_CSE_MASK;
1239         default:
1240                 break;
1241         }
1242         return 0;
1243 }
1244
1245 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1246                                                                 u8 opcode)
1247 {
1248         u32 effects = 0;
1249
1250         if (ns) {
1251                 if (ctrl->effects)
1252                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1253                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1254                         dev_warn(ctrl->device,
1255                                  "IO command:%02x has unhandled effects:%08x\n",
1256                                  opcode, effects);
1257                 return 0;
1258         }
1259
1260         if (ctrl->effects)
1261                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1262         else
1263                 effects = nvme_known_admin_effects(opcode);
1264
1265         /*
1266          * For simplicity, IO to all namespaces is quiesced even if the command
1267          * effects say only one namespace is affected.
1268          */
1269         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1270                 mutex_lock(&ctrl->scan_lock);
1271                 nvme_start_freeze(ctrl);
1272                 nvme_wait_freeze(ctrl);
1273         }
1274         return effects;
1275 }
1276
1277 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1278 {
1279         struct nvme_ns *ns;
1280
1281         down_read(&ctrl->namespaces_rwsem);
1282         list_for_each_entry(ns, &ctrl->namespaces, list)
1283                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1284                         nvme_set_queue_dying(ns);
1285         up_read(&ctrl->namespaces_rwsem);
1286
1287         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1288 }
1289
1290 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1291 {
1292         /*
1293          * Revalidate LBA changes prior to unfreezing. This is necessary to
1294          * prevent memory corruption if a logical block size was changed by
1295          * this command.
1296          */
1297         if (effects & NVME_CMD_EFFECTS_LBCC)
1298                 nvme_update_formats(ctrl);
1299         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1300                 nvme_unfreeze(ctrl);
1301                 mutex_unlock(&ctrl->scan_lock);
1302         }
1303         if (effects & NVME_CMD_EFFECTS_CCC)
1304                 nvme_init_identify(ctrl);
1305         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1306                 nvme_queue_scan(ctrl);
1307 }
1308
1309 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1310                         struct nvme_passthru_cmd __user *ucmd)
1311 {
1312         struct nvme_passthru_cmd cmd;
1313         struct nvme_command c;
1314         unsigned timeout = 0;
1315         u32 effects;
1316         int status;
1317
1318         if (!capable(CAP_SYS_ADMIN))
1319                 return -EACCES;
1320         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1321                 return -EFAULT;
1322         if (cmd.flags)
1323                 return -EINVAL;
1324
1325         memset(&c, 0, sizeof(c));
1326         c.common.opcode = cmd.opcode;
1327         c.common.flags = cmd.flags;
1328         c.common.nsid = cpu_to_le32(cmd.nsid);
1329         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1330         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1331         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1332         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1333         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1334         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1335         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1336         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1337
1338         if (cmd.timeout_ms)
1339                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1340
1341         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1342         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1343                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1344                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1345                         0, &cmd.result, timeout);
1346         nvme_passthru_end(ctrl, effects);
1347
1348         if (status >= 0) {
1349                 if (put_user(cmd.result, &ucmd->result))
1350                         return -EFAULT;
1351         }
1352
1353         return status;
1354 }
1355
1356 /*
1357  * Issue ioctl requests on the first available path.  Note that unlike normal
1358  * block layer requests we will not retry failed request on another controller.
1359  */
1360 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1361                 struct nvme_ns_head **head, int *srcu_idx)
1362 {
1363 #ifdef CONFIG_NVME_MULTIPATH
1364         if (disk->fops == &nvme_ns_head_ops) {
1365                 *head = disk->private_data;
1366                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1367                 return nvme_find_path(*head);
1368         }
1369 #endif
1370         *head = NULL;
1371         *srcu_idx = -1;
1372         return disk->private_data;
1373 }
1374
1375 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1376 {
1377         if (head)
1378                 srcu_read_unlock(&head->srcu, idx);
1379 }
1380
1381 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1382 {
1383         switch (cmd) {
1384         case NVME_IOCTL_ID:
1385                 force_successful_syscall_return();
1386                 return ns->head->ns_id;
1387         case NVME_IOCTL_ADMIN_CMD:
1388                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1389         case NVME_IOCTL_IO_CMD:
1390                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1391         case NVME_IOCTL_SUBMIT_IO:
1392                 return nvme_submit_io(ns, (void __user *)arg);
1393         default:
1394 #ifdef CONFIG_NVM
1395                 if (ns->ndev)
1396                         return nvme_nvm_ioctl(ns, cmd, arg);
1397 #endif
1398                 if (is_sed_ioctl(cmd))
1399                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1400                                          (void __user *) arg);
1401                 return -ENOTTY;
1402         }
1403 }
1404
1405 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1406                 unsigned int cmd, unsigned long arg)
1407 {
1408         struct nvme_ns_head *head = NULL;
1409         struct nvme_ns *ns;
1410         int srcu_idx, ret;
1411
1412         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1413         if (unlikely(!ns))
1414                 ret = -EWOULDBLOCK;
1415         else
1416                 ret = nvme_ns_ioctl(ns, cmd, arg);
1417         nvme_put_ns_from_disk(head, srcu_idx);
1418         return ret;
1419 }
1420
1421 static int nvme_open(struct block_device *bdev, fmode_t mode)
1422 {
1423         struct nvme_ns *ns = bdev->bd_disk->private_data;
1424
1425 #ifdef CONFIG_NVME_MULTIPATH
1426         /* should never be called due to GENHD_FL_HIDDEN */
1427         if (WARN_ON_ONCE(ns->head->disk))
1428                 goto fail;
1429 #endif
1430         if (!kref_get_unless_zero(&ns->kref))
1431                 goto fail;
1432         if (!try_module_get(ns->ctrl->ops->module))
1433                 goto fail_put_ns;
1434
1435         return 0;
1436
1437 fail_put_ns:
1438         nvme_put_ns(ns);
1439 fail:
1440         return -ENXIO;
1441 }
1442
1443 static void nvme_release(struct gendisk *disk, fmode_t mode)
1444 {
1445         struct nvme_ns *ns = disk->private_data;
1446
1447         module_put(ns->ctrl->ops->module);
1448         nvme_put_ns(ns);
1449 }
1450
1451 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1452 {
1453         /* some standard values */
1454         geo->heads = 1 << 6;
1455         geo->sectors = 1 << 5;
1456         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1457         return 0;
1458 }
1459
1460 #ifdef CONFIG_BLK_DEV_INTEGRITY
1461 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1462 {
1463         struct blk_integrity integrity;
1464
1465         memset(&integrity, 0, sizeof(integrity));
1466         switch (pi_type) {
1467         case NVME_NS_DPS_PI_TYPE3:
1468                 integrity.profile = &t10_pi_type3_crc;
1469                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1470                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1471                 break;
1472         case NVME_NS_DPS_PI_TYPE1:
1473         case NVME_NS_DPS_PI_TYPE2:
1474                 integrity.profile = &t10_pi_type1_crc;
1475                 integrity.tag_size = sizeof(u16);
1476                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1477                 break;
1478         default:
1479                 integrity.profile = NULL;
1480                 break;
1481         }
1482         integrity.tuple_size = ms;
1483         blk_integrity_register(disk, &integrity);
1484         blk_queue_max_integrity_segments(disk->queue, 1);
1485 }
1486 #else
1487 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1488 {
1489 }
1490 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1491
1492 static void nvme_set_chunk_size(struct nvme_ns *ns)
1493 {
1494         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1495         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1496 }
1497
1498 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1499 {
1500         struct nvme_ctrl *ctrl = ns->ctrl;
1501         struct request_queue *queue = disk->queue;
1502         u32 size = queue_logical_block_size(queue);
1503
1504         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1505                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1506                 return;
1507         }
1508
1509         if (ctrl->nr_streams && ns->sws && ns->sgs)
1510                 size *= ns->sws * ns->sgs;
1511
1512         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1513                         NVME_DSM_MAX_RANGES);
1514
1515         queue->limits.discard_alignment = 0;
1516         queue->limits.discard_granularity = size;
1517
1518         /* If discard is already enabled, don't reset queue limits */
1519         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1520                 return;
1521
1522         blk_queue_max_discard_sectors(queue, UINT_MAX);
1523         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1524
1525         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1526                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1527 }
1528
1529 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1530 {
1531         u32 max_sectors;
1532         unsigned short bs = 1 << ns->lba_shift;
1533
1534         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1535             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1536                 return;
1537         /*
1538          * Even though NVMe spec explicitly states that MDTS is not
1539          * applicable to the write-zeroes:- "The restriction does not apply to
1540          * commands that do not transfer data between the host and the
1541          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1542          * In order to be more cautious use controller's max_hw_sectors value
1543          * to configure the maximum sectors for the write-zeroes which is
1544          * configured based on the controller's MDTS field in the
1545          * nvme_init_identify() if available.
1546          */
1547         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1548                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1549         else
1550                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1551
1552         blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1553 }
1554
1555 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1556                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1557 {
1558         memset(ids, 0, sizeof(*ids));
1559
1560         if (ctrl->vs >= NVME_VS(1, 1, 0))
1561                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1562         if (ctrl->vs >= NVME_VS(1, 2, 0))
1563                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1564         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1565                  /* Don't treat error as fatal we potentially
1566                   * already have a NGUID or EUI-64
1567                   */
1568                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1569                         dev_warn(ctrl->device,
1570                                  "%s: Identify Descriptors failed\n", __func__);
1571         }
1572 }
1573
1574 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1575 {
1576         return !uuid_is_null(&ids->uuid) ||
1577                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1578                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1579 }
1580
1581 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1582 {
1583         return uuid_equal(&a->uuid, &b->uuid) &&
1584                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1585                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1586 }
1587
1588 static void nvme_update_disk_info(struct gendisk *disk,
1589                 struct nvme_ns *ns, struct nvme_id_ns *id)
1590 {
1591         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1592         unsigned short bs = 1 << ns->lba_shift;
1593
1594         blk_mq_freeze_queue(disk->queue);
1595         blk_integrity_unregister(disk);
1596
1597         blk_queue_logical_block_size(disk->queue, bs);
1598         blk_queue_physical_block_size(disk->queue, bs);
1599         blk_queue_io_min(disk->queue, bs);
1600
1601         if (ns->ms && !ns->ext &&
1602             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1603                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1604         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1605                 capacity = 0;
1606
1607         set_capacity(disk, capacity);
1608
1609         nvme_config_discard(disk, ns);
1610         nvme_config_write_zeroes(disk, ns);
1611
1612         if (id->nsattr & (1 << 0))
1613                 set_disk_ro(disk, true);
1614         else
1615                 set_disk_ro(disk, false);
1616
1617         blk_mq_unfreeze_queue(disk->queue);
1618 }
1619
1620 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1621 {
1622         struct nvme_ns *ns = disk->private_data;
1623
1624         /*
1625          * If identify namespace failed, use default 512 byte block size so
1626          * block layer can use before failing read/write for 0 capacity.
1627          */
1628         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1629         if (ns->lba_shift == 0)
1630                 ns->lba_shift = 9;
1631         ns->noiob = le16_to_cpu(id->noiob);
1632         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1633         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1634         /* the PI implementation requires metadata equal t10 pi tuple size */
1635         if (ns->ms == sizeof(struct t10_pi_tuple))
1636                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1637         else
1638                 ns->pi_type = 0;
1639
1640         if (ns->noiob)
1641                 nvme_set_chunk_size(ns);
1642         nvme_update_disk_info(disk, ns, id);
1643 #ifdef CONFIG_NVME_MULTIPATH
1644         if (ns->head->disk) {
1645                 nvme_update_disk_info(ns->head->disk, ns, id);
1646                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1647         }
1648 #endif
1649 }
1650
1651 static int nvme_revalidate_disk(struct gendisk *disk)
1652 {
1653         struct nvme_ns *ns = disk->private_data;
1654         struct nvme_ctrl *ctrl = ns->ctrl;
1655         struct nvme_id_ns *id;
1656         struct nvme_ns_ids ids;
1657         int ret = 0;
1658
1659         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1660                 set_capacity(disk, 0);
1661                 return -ENODEV;
1662         }
1663
1664         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1665         if (!id)
1666                 return -ENODEV;
1667
1668         if (id->ncap == 0) {
1669                 ret = -ENODEV;
1670                 goto out;
1671         }
1672
1673         __nvme_revalidate_disk(disk, id);
1674         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1675         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1676                 dev_err(ctrl->device,
1677                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1678                 ret = -ENODEV;
1679         }
1680
1681 out:
1682         kfree(id);
1683         return ret;
1684 }
1685
1686 static char nvme_pr_type(enum pr_type type)
1687 {
1688         switch (type) {
1689         case PR_WRITE_EXCLUSIVE:
1690                 return 1;
1691         case PR_EXCLUSIVE_ACCESS:
1692                 return 2;
1693         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1694                 return 3;
1695         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1696                 return 4;
1697         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1698                 return 5;
1699         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1700                 return 6;
1701         default:
1702                 return 0;
1703         }
1704 };
1705
1706 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1707                                 u64 key, u64 sa_key, u8 op)
1708 {
1709         struct nvme_ns_head *head = NULL;
1710         struct nvme_ns *ns;
1711         struct nvme_command c;
1712         int srcu_idx, ret;
1713         u8 data[16] = { 0, };
1714
1715         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1716         if (unlikely(!ns))
1717                 return -EWOULDBLOCK;
1718
1719         put_unaligned_le64(key, &data[0]);
1720         put_unaligned_le64(sa_key, &data[8]);
1721
1722         memset(&c, 0, sizeof(c));
1723         c.common.opcode = op;
1724         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1725         c.common.cdw10 = cpu_to_le32(cdw10);
1726
1727         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1728         nvme_put_ns_from_disk(head, srcu_idx);
1729         return ret;
1730 }
1731
1732 static int nvme_pr_register(struct block_device *bdev, u64 old,
1733                 u64 new, unsigned flags)
1734 {
1735         u32 cdw10;
1736
1737         if (flags & ~PR_FL_IGNORE_KEY)
1738                 return -EOPNOTSUPP;
1739
1740         cdw10 = old ? 2 : 0;
1741         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1742         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1743         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1744 }
1745
1746 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1747                 enum pr_type type, unsigned flags)
1748 {
1749         u32 cdw10;
1750
1751         if (flags & ~PR_FL_IGNORE_KEY)
1752                 return -EOPNOTSUPP;
1753
1754         cdw10 = nvme_pr_type(type) << 8;
1755         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1756         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1757 }
1758
1759 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1760                 enum pr_type type, bool abort)
1761 {
1762         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1763         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1764 }
1765
1766 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1767 {
1768         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1769         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1770 }
1771
1772 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1773 {
1774         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1775         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1776 }
1777
1778 static const struct pr_ops nvme_pr_ops = {
1779         .pr_register    = nvme_pr_register,
1780         .pr_reserve     = nvme_pr_reserve,
1781         .pr_release     = nvme_pr_release,
1782         .pr_preempt     = nvme_pr_preempt,
1783         .pr_clear       = nvme_pr_clear,
1784 };
1785
1786 #ifdef CONFIG_BLK_SED_OPAL
1787 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1788                 bool send)
1789 {
1790         struct nvme_ctrl *ctrl = data;
1791         struct nvme_command cmd;
1792
1793         memset(&cmd, 0, sizeof(cmd));
1794         if (send)
1795                 cmd.common.opcode = nvme_admin_security_send;
1796         else
1797                 cmd.common.opcode = nvme_admin_security_recv;
1798         cmd.common.nsid = 0;
1799         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1800         cmd.common.cdw11 = cpu_to_le32(len);
1801
1802         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1803                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1804 }
1805 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1806 #endif /* CONFIG_BLK_SED_OPAL */
1807
1808 static const struct block_device_operations nvme_fops = {
1809         .owner          = THIS_MODULE,
1810         .ioctl          = nvme_ioctl,
1811         .compat_ioctl   = nvme_ioctl,
1812         .open           = nvme_open,
1813         .release        = nvme_release,
1814         .getgeo         = nvme_getgeo,
1815         .revalidate_disk= nvme_revalidate_disk,
1816         .pr_ops         = &nvme_pr_ops,
1817 };
1818
1819 #ifdef CONFIG_NVME_MULTIPATH
1820 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1821 {
1822         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1823
1824         if (!kref_get_unless_zero(&head->ref))
1825                 return -ENXIO;
1826         return 0;
1827 }
1828
1829 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1830 {
1831         nvme_put_ns_head(disk->private_data);
1832 }
1833
1834 const struct block_device_operations nvme_ns_head_ops = {
1835         .owner          = THIS_MODULE,
1836         .open           = nvme_ns_head_open,
1837         .release        = nvme_ns_head_release,
1838         .ioctl          = nvme_ioctl,
1839         .compat_ioctl   = nvme_ioctl,
1840         .getgeo         = nvme_getgeo,
1841         .pr_ops         = &nvme_pr_ops,
1842 };
1843 #endif /* CONFIG_NVME_MULTIPATH */
1844
1845 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1846 {
1847         unsigned long timeout =
1848                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1849         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1850         int ret;
1851
1852         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1853                 if (csts == ~0)
1854                         return -ENODEV;
1855                 if ((csts & NVME_CSTS_RDY) == bit)
1856                         break;
1857
1858                 msleep(100);
1859                 if (fatal_signal_pending(current))
1860                         return -EINTR;
1861                 if (time_after(jiffies, timeout)) {
1862                         dev_err(ctrl->device,
1863                                 "Device not ready; aborting %s\n", enabled ?
1864                                                 "initialisation" : "reset");
1865                         return -ENODEV;
1866                 }
1867         }
1868
1869         return ret;
1870 }
1871
1872 /*
1873  * If the device has been passed off to us in an enabled state, just clear
1874  * the enabled bit.  The spec says we should set the 'shutdown notification
1875  * bits', but doing so may cause the device to complete commands to the
1876  * admin queue ... and we don't know what memory that might be pointing at!
1877  */
1878 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1879 {
1880         int ret;
1881
1882         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1883         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1884
1885         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1886         if (ret)
1887                 return ret;
1888
1889         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1890                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1891
1892         return nvme_wait_ready(ctrl, cap, false);
1893 }
1894 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1895
1896 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1897 {
1898         /*
1899          * Default to a 4K page size, with the intention to update this
1900          * path in the future to accomodate architectures with differing
1901          * kernel and IO page sizes.
1902          */
1903         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1904         int ret;
1905
1906         if (page_shift < dev_page_min) {
1907                 dev_err(ctrl->device,
1908                         "Minimum device page size %u too large for host (%u)\n",
1909                         1 << dev_page_min, 1 << page_shift);
1910                 return -ENODEV;
1911         }
1912
1913         ctrl->page_size = 1 << page_shift;
1914
1915         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1916         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1917         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1918         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1919         ctrl->ctrl_config |= NVME_CC_ENABLE;
1920
1921         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1922         if (ret)
1923                 return ret;
1924         return nvme_wait_ready(ctrl, cap, true);
1925 }
1926 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1927
1928 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1929 {
1930         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1931         u32 csts;
1932         int ret;
1933
1934         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1935         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1936
1937         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1938         if (ret)
1939                 return ret;
1940
1941         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1942                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1943                         break;
1944
1945                 msleep(100);
1946                 if (fatal_signal_pending(current))
1947                         return -EINTR;
1948                 if (time_after(jiffies, timeout)) {
1949                         dev_err(ctrl->device,
1950                                 "Device shutdown incomplete; abort shutdown\n");
1951                         return -ENODEV;
1952                 }
1953         }
1954
1955         return ret;
1956 }
1957 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1958
1959 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1960                 struct request_queue *q)
1961 {
1962         bool vwc = false;
1963
1964         if (ctrl->max_hw_sectors) {
1965                 u32 max_segments =
1966                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1967
1968                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1969                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1970                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1971         }
1972         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1973             is_power_of_2(ctrl->max_hw_sectors))
1974                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1975         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1976         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1977                 vwc = true;
1978         blk_queue_write_cache(q, vwc, vwc);
1979 }
1980
1981 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1982 {
1983         __le64 ts;
1984         int ret;
1985
1986         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1987                 return 0;
1988
1989         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1990         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1991                         NULL);
1992         if (ret)
1993                 dev_warn_once(ctrl->device,
1994                         "could not set timestamp (%d)\n", ret);
1995         return ret;
1996 }
1997
1998 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
1999 {
2000         struct nvme_feat_host_behavior *host;
2001         int ret;
2002
2003         /* Don't bother enabling the feature if retry delay is not reported */
2004         if (!ctrl->crdt[0])
2005                 return 0;
2006
2007         host = kzalloc(sizeof(*host), GFP_KERNEL);
2008         if (!host)
2009                 return 0;
2010
2011         host->acre = NVME_ENABLE_ACRE;
2012         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2013                                 host, sizeof(*host), NULL);
2014         kfree(host);
2015         return ret;
2016 }
2017
2018 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2019 {
2020         /*
2021          * APST (Autonomous Power State Transition) lets us program a
2022          * table of power state transitions that the controller will
2023          * perform automatically.  We configure it with a simple
2024          * heuristic: we are willing to spend at most 2% of the time
2025          * transitioning between power states.  Therefore, when running
2026          * in any given state, we will enter the next lower-power
2027          * non-operational state after waiting 50 * (enlat + exlat)
2028          * microseconds, as long as that state's exit latency is under
2029          * the requested maximum latency.
2030          *
2031          * We will not autonomously enter any non-operational state for
2032          * which the total latency exceeds ps_max_latency_us.  Users
2033          * can set ps_max_latency_us to zero to turn off APST.
2034          */
2035
2036         unsigned apste;
2037         struct nvme_feat_auto_pst *table;
2038         u64 max_lat_us = 0;
2039         int max_ps = -1;
2040         int ret;
2041
2042         /*
2043          * If APST isn't supported or if we haven't been initialized yet,
2044          * then don't do anything.
2045          */
2046         if (!ctrl->apsta)
2047                 return 0;
2048
2049         if (ctrl->npss > 31) {
2050                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2051                 return 0;
2052         }
2053
2054         table = kzalloc(sizeof(*table), GFP_KERNEL);
2055         if (!table)
2056                 return 0;
2057
2058         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2059                 /* Turn off APST. */
2060                 apste = 0;
2061                 dev_dbg(ctrl->device, "APST disabled\n");
2062         } else {
2063                 __le64 target = cpu_to_le64(0);
2064                 int state;
2065
2066                 /*
2067                  * Walk through all states from lowest- to highest-power.
2068                  * According to the spec, lower-numbered states use more
2069                  * power.  NPSS, despite the name, is the index of the
2070                  * lowest-power state, not the number of states.
2071                  */
2072                 for (state = (int)ctrl->npss; state >= 0; state--) {
2073                         u64 total_latency_us, exit_latency_us, transition_ms;
2074
2075                         if (target)
2076                                 table->entries[state] = target;
2077
2078                         /*
2079                          * Don't allow transitions to the deepest state
2080                          * if it's quirked off.
2081                          */
2082                         if (state == ctrl->npss &&
2083                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2084                                 continue;
2085
2086                         /*
2087                          * Is this state a useful non-operational state for
2088                          * higher-power states to autonomously transition to?
2089                          */
2090                         if (!(ctrl->psd[state].flags &
2091                               NVME_PS_FLAGS_NON_OP_STATE))
2092                                 continue;
2093
2094                         exit_latency_us =
2095                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2096                         if (exit_latency_us > ctrl->ps_max_latency_us)
2097                                 continue;
2098
2099                         total_latency_us =
2100                                 exit_latency_us +
2101                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2102
2103                         /*
2104                          * This state is good.  Use it as the APST idle
2105                          * target for higher power states.
2106                          */
2107                         transition_ms = total_latency_us + 19;
2108                         do_div(transition_ms, 20);
2109                         if (transition_ms > (1 << 24) - 1)
2110                                 transition_ms = (1 << 24) - 1;
2111
2112                         target = cpu_to_le64((state << 3) |
2113                                              (transition_ms << 8));
2114
2115                         if (max_ps == -1)
2116                                 max_ps = state;
2117
2118                         if (total_latency_us > max_lat_us)
2119                                 max_lat_us = total_latency_us;
2120                 }
2121
2122                 apste = 1;
2123
2124                 if (max_ps == -1) {
2125                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2126                 } else {
2127                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2128                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2129                 }
2130         }
2131
2132         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2133                                 table, sizeof(*table), NULL);
2134         if (ret)
2135                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2136
2137         kfree(table);
2138         return ret;
2139 }
2140
2141 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2142 {
2143         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2144         u64 latency;
2145
2146         switch (val) {
2147         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2148         case PM_QOS_LATENCY_ANY:
2149                 latency = U64_MAX;
2150                 break;
2151
2152         default:
2153                 latency = val;
2154         }
2155
2156         if (ctrl->ps_max_latency_us != latency) {
2157                 ctrl->ps_max_latency_us = latency;
2158                 nvme_configure_apst(ctrl);
2159         }
2160 }
2161
2162 struct nvme_core_quirk_entry {
2163         /*
2164          * NVMe model and firmware strings are padded with spaces.  For
2165          * simplicity, strings in the quirk table are padded with NULLs
2166          * instead.
2167          */
2168         u16 vid;
2169         const char *mn;
2170         const char *fr;
2171         unsigned long quirks;
2172 };
2173
2174 static const struct nvme_core_quirk_entry core_quirks[] = {
2175         {
2176                 /*
2177                  * This Toshiba device seems to die using any APST states.  See:
2178                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2179                  */
2180                 .vid = 0x1179,
2181                 .mn = "THNSF5256GPUK TOSHIBA",
2182                 .quirks = NVME_QUIRK_NO_APST,
2183         }
2184 };
2185
2186 /* match is null-terminated but idstr is space-padded. */
2187 static bool string_matches(const char *idstr, const char *match, size_t len)
2188 {
2189         size_t matchlen;
2190
2191         if (!match)
2192                 return true;
2193
2194         matchlen = strlen(match);
2195         WARN_ON_ONCE(matchlen > len);
2196
2197         if (memcmp(idstr, match, matchlen))
2198                 return false;
2199
2200         for (; matchlen < len; matchlen++)
2201                 if (idstr[matchlen] != ' ')
2202                         return false;
2203
2204         return true;
2205 }
2206
2207 static bool quirk_matches(const struct nvme_id_ctrl *id,
2208                           const struct nvme_core_quirk_entry *q)
2209 {
2210         return q->vid == le16_to_cpu(id->vid) &&
2211                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2212                 string_matches(id->fr, q->fr, sizeof(id->fr));
2213 }
2214
2215 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2216                 struct nvme_id_ctrl *id)
2217 {
2218         size_t nqnlen;
2219         int off;
2220
2221         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2222                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2223                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2224                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2225                         return;
2226                 }
2227
2228                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2229                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2230         }
2231
2232         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2233         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2234                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2235                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2236         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2237         off += sizeof(id->sn);
2238         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2239         off += sizeof(id->mn);
2240         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2241 }
2242
2243 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2244 {
2245         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2246         kfree(subsys);
2247 }
2248
2249 static void nvme_release_subsystem(struct device *dev)
2250 {
2251         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2252 }
2253
2254 static void nvme_destroy_subsystem(struct kref *ref)
2255 {
2256         struct nvme_subsystem *subsys =
2257                         container_of(ref, struct nvme_subsystem, ref);
2258
2259         mutex_lock(&nvme_subsystems_lock);
2260         list_del(&subsys->entry);
2261         mutex_unlock(&nvme_subsystems_lock);
2262
2263         ida_destroy(&subsys->ns_ida);
2264         device_del(&subsys->dev);
2265         put_device(&subsys->dev);
2266 }
2267
2268 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2269 {
2270         kref_put(&subsys->ref, nvme_destroy_subsystem);
2271 }
2272
2273 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2274 {
2275         struct nvme_subsystem *subsys;
2276
2277         lockdep_assert_held(&nvme_subsystems_lock);
2278
2279         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2280                 if (strcmp(subsys->subnqn, subsysnqn))
2281                         continue;
2282                 if (!kref_get_unless_zero(&subsys->ref))
2283                         continue;
2284                 return subsys;
2285         }
2286
2287         return NULL;
2288 }
2289
2290 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2291         struct device_attribute subsys_attr_##_name = \
2292                 __ATTR(_name, _mode, _show, NULL)
2293
2294 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2295                                     struct device_attribute *attr,
2296                                     char *buf)
2297 {
2298         struct nvme_subsystem *subsys =
2299                 container_of(dev, struct nvme_subsystem, dev);
2300
2301         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2302 }
2303 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2304
2305 #define nvme_subsys_show_str_function(field)                            \
2306 static ssize_t subsys_##field##_show(struct device *dev,                \
2307                             struct device_attribute *attr, char *buf)   \
2308 {                                                                       \
2309         struct nvme_subsystem *subsys =                                 \
2310                 container_of(dev, struct nvme_subsystem, dev);          \
2311         return sprintf(buf, "%.*s\n",                                   \
2312                        (int)sizeof(subsys->field), subsys->field);      \
2313 }                                                                       \
2314 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2315
2316 nvme_subsys_show_str_function(model);
2317 nvme_subsys_show_str_function(serial);
2318 nvme_subsys_show_str_function(firmware_rev);
2319
2320 static struct attribute *nvme_subsys_attrs[] = {
2321         &subsys_attr_model.attr,
2322         &subsys_attr_serial.attr,
2323         &subsys_attr_firmware_rev.attr,
2324         &subsys_attr_subsysnqn.attr,
2325 #ifdef CONFIG_NVME_MULTIPATH
2326         &subsys_attr_iopolicy.attr,
2327 #endif
2328         NULL,
2329 };
2330
2331 static struct attribute_group nvme_subsys_attrs_group = {
2332         .attrs = nvme_subsys_attrs,
2333 };
2334
2335 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2336         &nvme_subsys_attrs_group,
2337         NULL,
2338 };
2339
2340 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2341 {
2342         int count = 0;
2343         struct nvme_ctrl *ctrl;
2344
2345         mutex_lock(&subsys->lock);
2346         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2347                 if (ctrl->state != NVME_CTRL_DELETING &&
2348                     ctrl->state != NVME_CTRL_DEAD)
2349                         count++;
2350         }
2351         mutex_unlock(&subsys->lock);
2352
2353         return count;
2354 }
2355
2356 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2357 {
2358         struct nvme_subsystem *subsys, *found;
2359         int ret;
2360
2361         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2362         if (!subsys)
2363                 return -ENOMEM;
2364         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2365         if (ret < 0) {
2366                 kfree(subsys);
2367                 return ret;
2368         }
2369         subsys->instance = ret;
2370         mutex_init(&subsys->lock);
2371         kref_init(&subsys->ref);
2372         INIT_LIST_HEAD(&subsys->ctrls);
2373         INIT_LIST_HEAD(&subsys->nsheads);
2374         nvme_init_subnqn(subsys, ctrl, id);
2375         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2376         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2377         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2378         subsys->vendor_id = le16_to_cpu(id->vid);
2379         subsys->cmic = id->cmic;
2380 #ifdef CONFIG_NVME_MULTIPATH
2381         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2382 #endif
2383
2384         subsys->dev.class = nvme_subsys_class;
2385         subsys->dev.release = nvme_release_subsystem;
2386         subsys->dev.groups = nvme_subsys_attrs_groups;
2387         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2388         device_initialize(&subsys->dev);
2389
2390         mutex_lock(&nvme_subsystems_lock);
2391         found = __nvme_find_get_subsystem(subsys->subnqn);
2392         if (found) {
2393                 /*
2394                  * Verify that the subsystem actually supports multiple
2395                  * controllers, else bail out.
2396                  */
2397                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2398                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2399                         dev_err(ctrl->device,
2400                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2401                                 found->subnqn);
2402                         nvme_put_subsystem(found);
2403                         ret = -EINVAL;
2404                         goto out_unlock;
2405                 }
2406
2407                 __nvme_release_subsystem(subsys);
2408                 subsys = found;
2409         } else {
2410                 ret = device_add(&subsys->dev);
2411                 if (ret) {
2412                         dev_err(ctrl->device,
2413                                 "failed to register subsystem device.\n");
2414                         goto out_unlock;
2415                 }
2416                 ida_init(&subsys->ns_ida);
2417                 list_add_tail(&subsys->entry, &nvme_subsystems);
2418         }
2419
2420         ctrl->subsys = subsys;
2421         mutex_unlock(&nvme_subsystems_lock);
2422
2423         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2424                         dev_name(ctrl->device))) {
2425                 dev_err(ctrl->device,
2426                         "failed to create sysfs link from subsystem.\n");
2427                 /* the transport driver will eventually put the subsystem */
2428                 return -EINVAL;
2429         }
2430
2431         mutex_lock(&subsys->lock);
2432         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2433         mutex_unlock(&subsys->lock);
2434
2435         return 0;
2436
2437 out_unlock:
2438         mutex_unlock(&nvme_subsystems_lock);
2439         put_device(&subsys->dev);
2440         return ret;
2441 }
2442
2443 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2444                 void *log, size_t size, u64 offset)
2445 {
2446         struct nvme_command c = { };
2447         unsigned long dwlen = size / 4 - 1;
2448
2449         c.get_log_page.opcode = nvme_admin_get_log_page;
2450         c.get_log_page.nsid = cpu_to_le32(nsid);
2451         c.get_log_page.lid = log_page;
2452         c.get_log_page.lsp = lsp;
2453         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2454         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2455         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2456         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2457
2458         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2459 }
2460
2461 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2462 {
2463         int ret;
2464
2465         if (!ctrl->effects)
2466                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2467
2468         if (!ctrl->effects)
2469                 return 0;
2470
2471         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2472                         ctrl->effects, sizeof(*ctrl->effects), 0);
2473         if (ret) {
2474                 kfree(ctrl->effects);
2475                 ctrl->effects = NULL;
2476         }
2477         return ret;
2478 }
2479
2480 /*
2481  * Initialize the cached copies of the Identify data and various controller
2482  * register in our nvme_ctrl structure.  This should be called as soon as
2483  * the admin queue is fully up and running.
2484  */
2485 int nvme_init_identify(struct nvme_ctrl *ctrl)
2486 {
2487         struct nvme_id_ctrl *id;
2488         u64 cap;
2489         int ret, page_shift;
2490         u32 max_hw_sectors;
2491         bool prev_apst_enabled;
2492
2493         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2494         if (ret) {
2495                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2496                 return ret;
2497         }
2498
2499         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2500         if (ret) {
2501                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2502                 return ret;
2503         }
2504         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2505
2506         if (ctrl->vs >= NVME_VS(1, 1, 0))
2507                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2508
2509         ret = nvme_identify_ctrl(ctrl, &id);
2510         if (ret) {
2511                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2512                 return -EIO;
2513         }
2514
2515         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2516                 ret = nvme_get_effects_log(ctrl);
2517                 if (ret < 0)
2518                         goto out_free;
2519         }
2520
2521         if (!ctrl->identified) {
2522                 int i;
2523
2524                 ret = nvme_init_subsystem(ctrl, id);
2525                 if (ret)
2526                         goto out_free;
2527
2528                 /*
2529                  * Check for quirks.  Quirk can depend on firmware version,
2530                  * so, in principle, the set of quirks present can change
2531                  * across a reset.  As a possible future enhancement, we
2532                  * could re-scan for quirks every time we reinitialize
2533                  * the device, but we'd have to make sure that the driver
2534                  * behaves intelligently if the quirks change.
2535                  */
2536                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2537                         if (quirk_matches(id, &core_quirks[i]))
2538                                 ctrl->quirks |= core_quirks[i].quirks;
2539                 }
2540         }
2541
2542         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2543                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2544                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2545         }
2546
2547         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2548         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2549         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2550
2551         ctrl->oacs = le16_to_cpu(id->oacs);
2552         ctrl->oncs = le16_to_cpup(&id->oncs);
2553         ctrl->oaes = le32_to_cpu(id->oaes);
2554         atomic_set(&ctrl->abort_limit, id->acl + 1);
2555         ctrl->vwc = id->vwc;
2556         if (id->mdts)
2557                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2558         else
2559                 max_hw_sectors = UINT_MAX;
2560         ctrl->max_hw_sectors =
2561                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2562
2563         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2564         ctrl->sgls = le32_to_cpu(id->sgls);
2565         ctrl->kas = le16_to_cpu(id->kas);
2566         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2567         ctrl->ctratt = le32_to_cpu(id->ctratt);
2568
2569         if (id->rtd3e) {
2570                 /* us -> s */
2571                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2572
2573                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2574                                                  shutdown_timeout, 60);
2575
2576                 if (ctrl->shutdown_timeout != shutdown_timeout)
2577                         dev_info(ctrl->device,
2578                                  "Shutdown timeout set to %u seconds\n",
2579                                  ctrl->shutdown_timeout);
2580         } else
2581                 ctrl->shutdown_timeout = shutdown_timeout;
2582
2583         ctrl->npss = id->npss;
2584         ctrl->apsta = id->apsta;
2585         prev_apst_enabled = ctrl->apst_enabled;
2586         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2587                 if (force_apst && id->apsta) {
2588                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2589                         ctrl->apst_enabled = true;
2590                 } else {
2591                         ctrl->apst_enabled = false;
2592                 }
2593         } else {
2594                 ctrl->apst_enabled = id->apsta;
2595         }
2596         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2597
2598         if (ctrl->ops->flags & NVME_F_FABRICS) {
2599                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2600                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2601                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2602                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2603
2604                 /*
2605                  * In fabrics we need to verify the cntlid matches the
2606                  * admin connect
2607                  */
2608                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2609                         ret = -EINVAL;
2610                         goto out_free;
2611                 }
2612
2613                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2614                         dev_err(ctrl->device,
2615                                 "keep-alive support is mandatory for fabrics\n");
2616                         ret = -EINVAL;
2617                         goto out_free;
2618                 }
2619         } else {
2620                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2621                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2622                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2623                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2624                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2625         }
2626
2627         ret = nvme_mpath_init(ctrl, id);
2628         kfree(id);
2629
2630         if (ret < 0)
2631                 return ret;
2632
2633         if (ctrl->apst_enabled && !prev_apst_enabled)
2634                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2635         else if (!ctrl->apst_enabled && prev_apst_enabled)
2636                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2637
2638         ret = nvme_configure_apst(ctrl);
2639         if (ret < 0)
2640                 return ret;
2641         
2642         ret = nvme_configure_timestamp(ctrl);
2643         if (ret < 0)
2644                 return ret;
2645
2646         ret = nvme_configure_directives(ctrl);
2647         if (ret < 0)
2648                 return ret;
2649
2650         ret = nvme_configure_acre(ctrl);
2651         if (ret < 0)
2652                 return ret;
2653
2654         ctrl->identified = true;
2655
2656         return 0;
2657
2658 out_free:
2659         kfree(id);
2660         return ret;
2661 }
2662 EXPORT_SYMBOL_GPL(nvme_init_identify);
2663
2664 static int nvme_dev_open(struct inode *inode, struct file *file)
2665 {
2666         struct nvme_ctrl *ctrl =
2667                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2668
2669         switch (ctrl->state) {
2670         case NVME_CTRL_LIVE:
2671         case NVME_CTRL_ADMIN_ONLY:
2672                 break;
2673         default:
2674                 return -EWOULDBLOCK;
2675         }
2676
2677         file->private_data = ctrl;
2678         return 0;
2679 }
2680
2681 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2682 {
2683         struct nvme_ns *ns;
2684         int ret;
2685
2686         down_read(&ctrl->namespaces_rwsem);
2687         if (list_empty(&ctrl->namespaces)) {
2688                 ret = -ENOTTY;
2689                 goto out_unlock;
2690         }
2691
2692         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2693         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2694                 dev_warn(ctrl->device,
2695                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2696                 ret = -EINVAL;
2697                 goto out_unlock;
2698         }
2699
2700         dev_warn(ctrl->device,
2701                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2702         kref_get(&ns->kref);
2703         up_read(&ctrl->namespaces_rwsem);
2704
2705         ret = nvme_user_cmd(ctrl, ns, argp);
2706         nvme_put_ns(ns);
2707         return ret;
2708
2709 out_unlock:
2710         up_read(&ctrl->namespaces_rwsem);
2711         return ret;
2712 }
2713
2714 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2715                 unsigned long arg)
2716 {
2717         struct nvme_ctrl *ctrl = file->private_data;
2718         void __user *argp = (void __user *)arg;
2719
2720         switch (cmd) {
2721         case NVME_IOCTL_ADMIN_CMD:
2722                 return nvme_user_cmd(ctrl, NULL, argp);
2723         case NVME_IOCTL_IO_CMD:
2724                 return nvme_dev_user_cmd(ctrl, argp);
2725         case NVME_IOCTL_RESET:
2726                 dev_warn(ctrl->device, "resetting controller\n");
2727                 return nvme_reset_ctrl_sync(ctrl);
2728         case NVME_IOCTL_SUBSYS_RESET:
2729                 return nvme_reset_subsystem(ctrl);
2730         case NVME_IOCTL_RESCAN:
2731                 nvme_queue_scan(ctrl);
2732                 return 0;
2733         default:
2734                 return -ENOTTY;
2735         }
2736 }
2737
2738 static const struct file_operations nvme_dev_fops = {
2739         .owner          = THIS_MODULE,
2740         .open           = nvme_dev_open,
2741         .unlocked_ioctl = nvme_dev_ioctl,
2742         .compat_ioctl   = nvme_dev_ioctl,
2743 };
2744
2745 static ssize_t nvme_sysfs_reset(struct device *dev,
2746                                 struct device_attribute *attr, const char *buf,
2747                                 size_t count)
2748 {
2749         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2750         int ret;
2751
2752         ret = nvme_reset_ctrl_sync(ctrl);
2753         if (ret < 0)
2754                 return ret;
2755         return count;
2756 }
2757 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2758
2759 static ssize_t nvme_sysfs_rescan(struct device *dev,
2760                                 struct device_attribute *attr, const char *buf,
2761                                 size_t count)
2762 {
2763         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2764
2765         nvme_queue_scan(ctrl);
2766         return count;
2767 }
2768 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2769
2770 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2771 {
2772         struct gendisk *disk = dev_to_disk(dev);
2773
2774         if (disk->fops == &nvme_fops)
2775                 return nvme_get_ns_from_dev(dev)->head;
2776         else
2777                 return disk->private_data;
2778 }
2779
2780 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2781                 char *buf)
2782 {
2783         struct nvme_ns_head *head = dev_to_ns_head(dev);
2784         struct nvme_ns_ids *ids = &head->ids;
2785         struct nvme_subsystem *subsys = head->subsys;
2786         int serial_len = sizeof(subsys->serial);
2787         int model_len = sizeof(subsys->model);
2788
2789         if (!uuid_is_null(&ids->uuid))
2790                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2791
2792         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2793                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2794
2795         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2796                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2797
2798         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2799                                   subsys->serial[serial_len - 1] == '\0'))
2800                 serial_len--;
2801         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2802                                  subsys->model[model_len - 1] == '\0'))
2803                 model_len--;
2804
2805         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2806                 serial_len, subsys->serial, model_len, subsys->model,
2807                 head->ns_id);
2808 }
2809 static DEVICE_ATTR_RO(wwid);
2810
2811 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2812                 char *buf)
2813 {
2814         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2815 }
2816 static DEVICE_ATTR_RO(nguid);
2817
2818 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2819                 char *buf)
2820 {
2821         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2822
2823         /* For backward compatibility expose the NGUID to userspace if
2824          * we have no UUID set
2825          */
2826         if (uuid_is_null(&ids->uuid)) {
2827                 printk_ratelimited(KERN_WARNING
2828                                    "No UUID available providing old NGUID\n");
2829                 return sprintf(buf, "%pU\n", ids->nguid);
2830         }
2831         return sprintf(buf, "%pU\n", &ids->uuid);
2832 }
2833 static DEVICE_ATTR_RO(uuid);
2834
2835 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2836                 char *buf)
2837 {
2838         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2839 }
2840 static DEVICE_ATTR_RO(eui);
2841
2842 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2843                 char *buf)
2844 {
2845         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2846 }
2847 static DEVICE_ATTR_RO(nsid);
2848
2849 static struct attribute *nvme_ns_id_attrs[] = {
2850         &dev_attr_wwid.attr,
2851         &dev_attr_uuid.attr,
2852         &dev_attr_nguid.attr,
2853         &dev_attr_eui.attr,
2854         &dev_attr_nsid.attr,
2855 #ifdef CONFIG_NVME_MULTIPATH
2856         &dev_attr_ana_grpid.attr,
2857         &dev_attr_ana_state.attr,
2858 #endif
2859         NULL,
2860 };
2861
2862 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2863                 struct attribute *a, int n)
2864 {
2865         struct device *dev = container_of(kobj, struct device, kobj);
2866         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2867
2868         if (a == &dev_attr_uuid.attr) {
2869                 if (uuid_is_null(&ids->uuid) &&
2870                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2871                         return 0;
2872         }
2873         if (a == &dev_attr_nguid.attr) {
2874                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2875                         return 0;
2876         }
2877         if (a == &dev_attr_eui.attr) {
2878                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2879                         return 0;
2880         }
2881 #ifdef CONFIG_NVME_MULTIPATH
2882         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2883                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2884                         return 0;
2885                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2886                         return 0;
2887         }
2888 #endif
2889         return a->mode;
2890 }
2891
2892 static const struct attribute_group nvme_ns_id_attr_group = {
2893         .attrs          = nvme_ns_id_attrs,
2894         .is_visible     = nvme_ns_id_attrs_are_visible,
2895 };
2896
2897 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2898         &nvme_ns_id_attr_group,
2899 #ifdef CONFIG_NVM
2900         &nvme_nvm_attr_group,
2901 #endif
2902         NULL,
2903 };
2904
2905 #define nvme_show_str_function(field)                                           \
2906 static ssize_t  field##_show(struct device *dev,                                \
2907                             struct device_attribute *attr, char *buf)           \
2908 {                                                                               \
2909         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2910         return sprintf(buf, "%.*s\n",                                           \
2911                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2912 }                                                                               \
2913 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2914
2915 nvme_show_str_function(model);
2916 nvme_show_str_function(serial);
2917 nvme_show_str_function(firmware_rev);
2918
2919 #define nvme_show_int_function(field)                                           \
2920 static ssize_t  field##_show(struct device *dev,                                \
2921                             struct device_attribute *attr, char *buf)           \
2922 {                                                                               \
2923         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2924         return sprintf(buf, "%d\n", ctrl->field);       \
2925 }                                                                               \
2926 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2927
2928 nvme_show_int_function(cntlid);
2929 nvme_show_int_function(numa_node);
2930
2931 static ssize_t nvme_sysfs_delete(struct device *dev,
2932                                 struct device_attribute *attr, const char *buf,
2933                                 size_t count)
2934 {
2935         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2936
2937         if (device_remove_file_self(dev, attr))
2938                 nvme_delete_ctrl_sync(ctrl);
2939         return count;
2940 }
2941 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2942
2943 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2944                                          struct device_attribute *attr,
2945                                          char *buf)
2946 {
2947         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2948
2949         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2950 }
2951 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2952
2953 static ssize_t nvme_sysfs_show_state(struct device *dev,
2954                                      struct device_attribute *attr,
2955                                      char *buf)
2956 {
2957         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2958         static const char *const state_name[] = {
2959                 [NVME_CTRL_NEW]         = "new",
2960                 [NVME_CTRL_LIVE]        = "live",
2961                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2962                 [NVME_CTRL_RESETTING]   = "resetting",
2963                 [NVME_CTRL_CONNECTING]  = "connecting",
2964                 [NVME_CTRL_DELETING]    = "deleting",
2965                 [NVME_CTRL_DEAD]        = "dead",
2966         };
2967
2968         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2969             state_name[ctrl->state])
2970                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2971
2972         return sprintf(buf, "unknown state\n");
2973 }
2974
2975 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2976
2977 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2978                                          struct device_attribute *attr,
2979                                          char *buf)
2980 {
2981         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2982
2983         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2984 }
2985 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2986
2987 static ssize_t nvme_sysfs_show_address(struct device *dev,
2988                                          struct device_attribute *attr,
2989                                          char *buf)
2990 {
2991         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2992
2993         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2994 }
2995 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2996
2997 static struct attribute *nvme_dev_attrs[] = {
2998         &dev_attr_reset_controller.attr,
2999         &dev_attr_rescan_controller.attr,
3000         &dev_attr_model.attr,
3001         &dev_attr_serial.attr,
3002         &dev_attr_firmware_rev.attr,
3003         &dev_attr_cntlid.attr,
3004         &dev_attr_delete_controller.attr,
3005         &dev_attr_transport.attr,
3006         &dev_attr_subsysnqn.attr,
3007         &dev_attr_address.attr,
3008         &dev_attr_state.attr,
3009         &dev_attr_numa_node.attr,
3010         NULL
3011 };
3012
3013 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3014                 struct attribute *a, int n)
3015 {
3016         struct device *dev = container_of(kobj, struct device, kobj);
3017         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3018
3019         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3020                 return 0;
3021         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3022                 return 0;
3023
3024         return a->mode;
3025 }
3026
3027 static struct attribute_group nvme_dev_attrs_group = {
3028         .attrs          = nvme_dev_attrs,
3029         .is_visible     = nvme_dev_attrs_are_visible,
3030 };
3031
3032 static const struct attribute_group *nvme_dev_attr_groups[] = {
3033         &nvme_dev_attrs_group,
3034         NULL,
3035 };
3036
3037 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3038                 unsigned nsid)
3039 {
3040         struct nvme_ns_head *h;
3041
3042         lockdep_assert_held(&subsys->lock);
3043
3044         list_for_each_entry(h, &subsys->nsheads, entry) {
3045                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3046                         return h;
3047         }
3048
3049         return NULL;
3050 }
3051
3052 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3053                 struct nvme_ns_head *new)
3054 {
3055         struct nvme_ns_head *h;
3056
3057         lockdep_assert_held(&subsys->lock);
3058
3059         list_for_each_entry(h, &subsys->nsheads, entry) {
3060                 if (nvme_ns_ids_valid(&new->ids) &&
3061                     !list_empty(&h->list) &&
3062                     nvme_ns_ids_equal(&new->ids, &h->ids))
3063                         return -EINVAL;
3064         }
3065
3066         return 0;
3067 }
3068
3069 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3070                 unsigned nsid, struct nvme_id_ns *id)
3071 {
3072         struct nvme_ns_head *head;
3073         size_t size = sizeof(*head);
3074         int ret = -ENOMEM;
3075
3076 #ifdef CONFIG_NVME_MULTIPATH
3077         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3078 #endif
3079
3080         head = kzalloc(size, GFP_KERNEL);
3081         if (!head)
3082                 goto out;
3083         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3084         if (ret < 0)
3085                 goto out_free_head;
3086         head->instance = ret;
3087         INIT_LIST_HEAD(&head->list);
3088         ret = init_srcu_struct(&head->srcu);
3089         if (ret)
3090                 goto out_ida_remove;
3091         head->subsys = ctrl->subsys;
3092         head->ns_id = nsid;
3093         kref_init(&head->ref);
3094
3095         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3096
3097         ret = __nvme_check_ids(ctrl->subsys, head);
3098         if (ret) {
3099                 dev_err(ctrl->device,
3100                         "duplicate IDs for nsid %d\n", nsid);
3101                 goto out_cleanup_srcu;
3102         }
3103
3104         ret = nvme_mpath_alloc_disk(ctrl, head);
3105         if (ret)
3106                 goto out_cleanup_srcu;
3107
3108         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3109
3110         kref_get(&ctrl->subsys->ref);
3111
3112         return head;
3113 out_cleanup_srcu:
3114         cleanup_srcu_struct(&head->srcu);
3115 out_ida_remove:
3116         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3117 out_free_head:
3118         kfree(head);
3119 out:
3120         return ERR_PTR(ret);
3121 }
3122
3123 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3124                 struct nvme_id_ns *id)
3125 {
3126         struct nvme_ctrl *ctrl = ns->ctrl;
3127         bool is_shared = id->nmic & (1 << 0);
3128         struct nvme_ns_head *head = NULL;
3129         int ret = 0;
3130
3131         mutex_lock(&ctrl->subsys->lock);
3132         if (is_shared)
3133                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3134         if (!head) {
3135                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3136                 if (IS_ERR(head)) {
3137                         ret = PTR_ERR(head);
3138                         goto out_unlock;
3139                 }
3140         } else {
3141                 struct nvme_ns_ids ids;
3142
3143                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3144                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3145                         dev_err(ctrl->device,
3146                                 "IDs don't match for shared namespace %d\n",
3147                                         nsid);
3148                         ret = -EINVAL;
3149                         goto out_unlock;
3150                 }
3151         }
3152
3153         list_add_tail(&ns->siblings, &head->list);
3154         ns->head = head;
3155
3156 out_unlock:
3157         mutex_unlock(&ctrl->subsys->lock);
3158         return ret;
3159 }
3160
3161 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3162 {
3163         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3164         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3165
3166         return nsa->head->ns_id - nsb->head->ns_id;
3167 }
3168
3169 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3170 {
3171         struct nvme_ns *ns, *ret = NULL;
3172
3173         down_read(&ctrl->namespaces_rwsem);
3174         list_for_each_entry(ns, &ctrl->namespaces, list) {
3175                 if (ns->head->ns_id == nsid) {
3176                         if (!kref_get_unless_zero(&ns->kref))
3177                                 continue;
3178                         ret = ns;
3179                         break;
3180                 }
3181                 if (ns->head->ns_id > nsid)
3182                         break;
3183         }
3184         up_read(&ctrl->namespaces_rwsem);
3185         return ret;
3186 }
3187
3188 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3189 {
3190         struct streams_directive_params s;
3191         int ret;
3192
3193         if (!ctrl->nr_streams)
3194                 return 0;
3195
3196         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3197         if (ret)
3198                 return ret;
3199
3200         ns->sws = le32_to_cpu(s.sws);
3201         ns->sgs = le16_to_cpu(s.sgs);
3202
3203         if (ns->sws) {
3204                 unsigned int bs = 1 << ns->lba_shift;
3205
3206                 blk_queue_io_min(ns->queue, bs * ns->sws);
3207                 if (ns->sgs)
3208                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3209         }
3210
3211         return 0;
3212 }
3213
3214 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3215 {
3216         struct nvme_ns *ns;
3217         struct gendisk *disk;
3218         struct nvme_id_ns *id;
3219         char disk_name[DISK_NAME_LEN];
3220         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3221
3222         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3223         if (!ns)
3224                 return -ENOMEM;
3225
3226         ns->queue = blk_mq_init_queue(ctrl->tagset);
3227         if (IS_ERR(ns->queue)) {
3228                 ret = PTR_ERR(ns->queue);
3229                 goto out_free_ns;
3230         }
3231
3232         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3233         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3234                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3235
3236         ns->queue->queuedata = ns;
3237         ns->ctrl = ctrl;
3238
3239         kref_init(&ns->kref);
3240         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3241
3242         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3243         nvme_set_queue_limits(ctrl, ns->queue);
3244
3245         id = nvme_identify_ns(ctrl, nsid);
3246         if (!id) {
3247                 ret = -EIO;
3248                 goto out_free_queue;
3249         }
3250
3251         if (id->ncap == 0) {
3252                 ret = -EINVAL;
3253                 goto out_free_id;
3254         }
3255
3256         ret = nvme_init_ns_head(ns, nsid, id);
3257         if (ret)
3258                 goto out_free_id;
3259         nvme_setup_streams_ns(ctrl, ns);
3260         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3261
3262         disk = alloc_disk_node(0, node);
3263         if (!disk) {
3264                 ret = -ENOMEM;
3265                 goto out_unlink_ns;
3266         }
3267
3268         disk->fops = &nvme_fops;
3269         disk->private_data = ns;
3270         disk->queue = ns->queue;
3271         disk->flags = flags;
3272         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3273         ns->disk = disk;
3274
3275         __nvme_revalidate_disk(disk, id);
3276
3277         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3278                 ret = nvme_nvm_register(ns, disk_name, node);
3279                 if (ret) {
3280                         dev_warn(ctrl->device, "LightNVM init failure\n");
3281                         goto out_put_disk;
3282                 }
3283         }
3284
3285         down_write(&ctrl->namespaces_rwsem);
3286         list_add_tail(&ns->list, &ctrl->namespaces);
3287         up_write(&ctrl->namespaces_rwsem);