Merge branch 'for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck...
[muen/linux.git] / drivers / nvme / host / core.c
1 /*
2  * NVM Express device driver
3  * Copyright (c) 2011-2014, Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  */
14
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
25 #include <linux/pr.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
29 #include <linux/pm_qos.h>
30 #include <asm/unaligned.h>
31
32 #define CREATE_TRACE_POINTS
33 #include "trace.h"
34
35 #include "nvme.h"
36 #include "fabrics.h"
37
38 #define NVME_MINORS             (1U << MINORBITS)
39
40 unsigned int admin_timeout = 60;
41 module_param(admin_timeout, uint, 0644);
42 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
43 EXPORT_SYMBOL_GPL(admin_timeout);
44
45 unsigned int nvme_io_timeout = 30;
46 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
47 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
48 EXPORT_SYMBOL_GPL(nvme_io_timeout);
49
50 static unsigned char shutdown_timeout = 5;
51 module_param(shutdown_timeout, byte, 0644);
52 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
53
54 static u8 nvme_max_retries = 5;
55 module_param_named(max_retries, nvme_max_retries, byte, 0644);
56 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
57
58 static unsigned long default_ps_max_latency_us = 100000;
59 module_param(default_ps_max_latency_us, ulong, 0644);
60 MODULE_PARM_DESC(default_ps_max_latency_us,
61                  "max power saving latency for new devices; use PM QOS to change per device");
62
63 static bool force_apst;
64 module_param(force_apst, bool, 0644);
65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
66
67 static bool streams;
68 module_param(streams, bool, 0644);
69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
70
71 /*
72  * nvme_wq - hosts nvme related works that are not reset or delete
73  * nvme_reset_wq - hosts nvme reset works
74  * nvme_delete_wq - hosts nvme delete works
75  *
76  * nvme_wq will host works such are scan, aen handling, fw activation,
77  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
78  * runs reset works which also flush works hosted on nvme_wq for
79  * serialization purposes. nvme_delete_wq host controller deletion
80  * works which flush reset works for serialization.
81  */
82 struct workqueue_struct *nvme_wq;
83 EXPORT_SYMBOL_GPL(nvme_wq);
84
85 struct workqueue_struct *nvme_reset_wq;
86 EXPORT_SYMBOL_GPL(nvme_reset_wq);
87
88 struct workqueue_struct *nvme_delete_wq;
89 EXPORT_SYMBOL_GPL(nvme_delete_wq);
90
91 static DEFINE_IDA(nvme_subsystems_ida);
92 static LIST_HEAD(nvme_subsystems);
93 static DEFINE_MUTEX(nvme_subsystems_lock);
94
95 static DEFINE_IDA(nvme_instance_ida);
96 static dev_t nvme_chr_devt;
97 static struct class *nvme_class;
98 static struct class *nvme_subsys_class;
99
100 static void nvme_ns_remove(struct nvme_ns *ns);
101 static int nvme_revalidate_disk(struct gendisk *disk);
102 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
103
104 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
105 {
106         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
107                 return -EBUSY;
108         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
109                 return -EBUSY;
110         return 0;
111 }
112 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
113
114 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
115 {
116         int ret;
117
118         ret = nvme_reset_ctrl(ctrl);
119         if (!ret) {
120                 flush_work(&ctrl->reset_work);
121                 if (ctrl->state != NVME_CTRL_LIVE &&
122                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
123                         ret = -ENETRESET;
124         }
125
126         return ret;
127 }
128 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
129
130 static void nvme_delete_ctrl_work(struct work_struct *work)
131 {
132         struct nvme_ctrl *ctrl =
133                 container_of(work, struct nvme_ctrl, delete_work);
134
135         dev_info(ctrl->device,
136                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
137
138         flush_work(&ctrl->reset_work);
139         nvme_stop_ctrl(ctrl);
140         nvme_remove_namespaces(ctrl);
141         ctrl->ops->delete_ctrl(ctrl);
142         nvme_uninit_ctrl(ctrl);
143         nvme_put_ctrl(ctrl);
144 }
145
146 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
147 {
148         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
149                 return -EBUSY;
150         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
151                 return -EBUSY;
152         return 0;
153 }
154 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
155
156 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
157 {
158         int ret = 0;
159
160         /*
161          * Keep a reference until the work is flushed since ->delete_ctrl
162          * can free the controller.
163          */
164         nvme_get_ctrl(ctrl);
165         ret = nvme_delete_ctrl(ctrl);
166         if (!ret)
167                 flush_work(&ctrl->delete_work);
168         nvme_put_ctrl(ctrl);
169         return ret;
170 }
171 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
172
173 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
174 {
175         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
176 }
177
178 static blk_status_t nvme_error_status(struct request *req)
179 {
180         switch (nvme_req(req)->status & 0x7ff) {
181         case NVME_SC_SUCCESS:
182                 return BLK_STS_OK;
183         case NVME_SC_CAP_EXCEEDED:
184                 return BLK_STS_NOSPC;
185         case NVME_SC_LBA_RANGE:
186                 return BLK_STS_TARGET;
187         case NVME_SC_BAD_ATTRIBUTES:
188         case NVME_SC_ONCS_NOT_SUPPORTED:
189         case NVME_SC_INVALID_OPCODE:
190         case NVME_SC_INVALID_FIELD:
191         case NVME_SC_INVALID_NS:
192                 return BLK_STS_NOTSUPP;
193         case NVME_SC_WRITE_FAULT:
194         case NVME_SC_READ_ERROR:
195         case NVME_SC_UNWRITTEN_BLOCK:
196         case NVME_SC_ACCESS_DENIED:
197         case NVME_SC_READ_ONLY:
198         case NVME_SC_COMPARE_FAILED:
199                 return BLK_STS_MEDIUM;
200         case NVME_SC_GUARD_CHECK:
201         case NVME_SC_APPTAG_CHECK:
202         case NVME_SC_REFTAG_CHECK:
203         case NVME_SC_INVALID_PI:
204                 return BLK_STS_PROTECTION;
205         case NVME_SC_RESERVATION_CONFLICT:
206                 return BLK_STS_NEXUS;
207         default:
208                 return BLK_STS_IOERR;
209         }
210 }
211
212 static inline bool nvme_req_needs_retry(struct request *req)
213 {
214         if (blk_noretry_request(req))
215                 return false;
216         if (nvme_req(req)->status & NVME_SC_DNR)
217                 return false;
218         if (nvme_req(req)->retries >= nvme_max_retries)
219                 return false;
220         return true;
221 }
222
223 void nvme_complete_rq(struct request *req)
224 {
225         blk_status_t status = nvme_error_status(req);
226
227         trace_nvme_complete_rq(req);
228
229         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
230                 if (nvme_req_needs_failover(req, status)) {
231                         nvme_failover_req(req);
232                         return;
233                 }
234
235                 if (!blk_queue_dying(req->q)) {
236                         nvme_req(req)->retries++;
237                         blk_mq_requeue_request(req, true);
238                         return;
239                 }
240         }
241         blk_mq_end_request(req, status);
242 }
243 EXPORT_SYMBOL_GPL(nvme_complete_rq);
244
245 void nvme_cancel_request(struct request *req, void *data, bool reserved)
246 {
247         if (!blk_mq_request_started(req))
248                 return;
249
250         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
251                                 "Cancelling I/O %d", req->tag);
252
253         nvme_req(req)->status = NVME_SC_ABORT_REQ;
254         blk_mq_complete_request(req);
255
256 }
257 EXPORT_SYMBOL_GPL(nvme_cancel_request);
258
259 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
260                 enum nvme_ctrl_state new_state)
261 {
262         enum nvme_ctrl_state old_state;
263         unsigned long flags;
264         bool changed = false;
265
266         spin_lock_irqsave(&ctrl->lock, flags);
267
268         old_state = ctrl->state;
269         switch (new_state) {
270         case NVME_CTRL_ADMIN_ONLY:
271                 switch (old_state) {
272                 case NVME_CTRL_CONNECTING:
273                         changed = true;
274                         /* FALLTHRU */
275                 default:
276                         break;
277                 }
278                 break;
279         case NVME_CTRL_LIVE:
280                 switch (old_state) {
281                 case NVME_CTRL_NEW:
282                 case NVME_CTRL_RESETTING:
283                 case NVME_CTRL_CONNECTING:
284                         changed = true;
285                         /* FALLTHRU */
286                 default:
287                         break;
288                 }
289                 break;
290         case NVME_CTRL_RESETTING:
291                 switch (old_state) {
292                 case NVME_CTRL_NEW:
293                 case NVME_CTRL_LIVE:
294                 case NVME_CTRL_ADMIN_ONLY:
295                         changed = true;
296                         /* FALLTHRU */
297                 default:
298                         break;
299                 }
300                 break;
301         case NVME_CTRL_CONNECTING:
302                 switch (old_state) {
303                 case NVME_CTRL_NEW:
304                 case NVME_CTRL_RESETTING:
305                         changed = true;
306                         /* FALLTHRU */
307                 default:
308                         break;
309                 }
310                 break;
311         case NVME_CTRL_DELETING:
312                 switch (old_state) {
313                 case NVME_CTRL_LIVE:
314                 case NVME_CTRL_ADMIN_ONLY:
315                 case NVME_CTRL_RESETTING:
316                 case NVME_CTRL_CONNECTING:
317                         changed = true;
318                         /* FALLTHRU */
319                 default:
320                         break;
321                 }
322                 break;
323         case NVME_CTRL_DEAD:
324                 switch (old_state) {
325                 case NVME_CTRL_DELETING:
326                         changed = true;
327                         /* FALLTHRU */
328                 default:
329                         break;
330                 }
331                 break;
332         default:
333                 break;
334         }
335
336         if (changed)
337                 ctrl->state = new_state;
338
339         spin_unlock_irqrestore(&ctrl->lock, flags);
340         if (changed && ctrl->state == NVME_CTRL_LIVE)
341                 nvme_kick_requeue_lists(ctrl);
342         return changed;
343 }
344 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
345
346 static void nvme_free_ns_head(struct kref *ref)
347 {
348         struct nvme_ns_head *head =
349                 container_of(ref, struct nvme_ns_head, ref);
350
351         nvme_mpath_remove_disk(head);
352         ida_simple_remove(&head->subsys->ns_ida, head->instance);
353         list_del_init(&head->entry);
354         cleanup_srcu_struct_quiesced(&head->srcu);
355         nvme_put_subsystem(head->subsys);
356         kfree(head);
357 }
358
359 static void nvme_put_ns_head(struct nvme_ns_head *head)
360 {
361         kref_put(&head->ref, nvme_free_ns_head);
362 }
363
364 static void nvme_free_ns(struct kref *kref)
365 {
366         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
367
368         if (ns->ndev)
369                 nvme_nvm_unregister(ns);
370
371         put_disk(ns->disk);
372         nvme_put_ns_head(ns->head);
373         nvme_put_ctrl(ns->ctrl);
374         kfree(ns);
375 }
376
377 static void nvme_put_ns(struct nvme_ns *ns)
378 {
379         kref_put(&ns->kref, nvme_free_ns);
380 }
381
382 static inline void nvme_clear_nvme_request(struct request *req)
383 {
384         if (!(req->rq_flags & RQF_DONTPREP)) {
385                 nvme_req(req)->retries = 0;
386                 nvme_req(req)->flags = 0;
387                 req->rq_flags |= RQF_DONTPREP;
388         }
389 }
390
391 struct request *nvme_alloc_request(struct request_queue *q,
392                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
393 {
394         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
395         struct request *req;
396
397         if (qid == NVME_QID_ANY) {
398                 req = blk_mq_alloc_request(q, op, flags);
399         } else {
400                 req = blk_mq_alloc_request_hctx(q, op, flags,
401                                 qid ? qid - 1 : 0);
402         }
403         if (IS_ERR(req))
404                 return req;
405
406         req->cmd_flags |= REQ_FAILFAST_DRIVER;
407         nvme_clear_nvme_request(req);
408         nvme_req(req)->cmd = cmd;
409
410         return req;
411 }
412 EXPORT_SYMBOL_GPL(nvme_alloc_request);
413
414 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
415 {
416         struct nvme_command c;
417
418         memset(&c, 0, sizeof(c));
419
420         c.directive.opcode = nvme_admin_directive_send;
421         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
422         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
423         c.directive.dtype = NVME_DIR_IDENTIFY;
424         c.directive.tdtype = NVME_DIR_STREAMS;
425         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
426
427         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
428 }
429
430 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
431 {
432         return nvme_toggle_streams(ctrl, false);
433 }
434
435 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
436 {
437         return nvme_toggle_streams(ctrl, true);
438 }
439
440 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
441                                   struct streams_directive_params *s, u32 nsid)
442 {
443         struct nvme_command c;
444
445         memset(&c, 0, sizeof(c));
446         memset(s, 0, sizeof(*s));
447
448         c.directive.opcode = nvme_admin_directive_recv;
449         c.directive.nsid = cpu_to_le32(nsid);
450         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
451         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
452         c.directive.dtype = NVME_DIR_STREAMS;
453
454         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
455 }
456
457 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
458 {
459         struct streams_directive_params s;
460         int ret;
461
462         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
463                 return 0;
464         if (!streams)
465                 return 0;
466
467         ret = nvme_enable_streams(ctrl);
468         if (ret)
469                 return ret;
470
471         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
472         if (ret)
473                 return ret;
474
475         ctrl->nssa = le16_to_cpu(s.nssa);
476         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
477                 dev_info(ctrl->device, "too few streams (%u) available\n",
478                                         ctrl->nssa);
479                 nvme_disable_streams(ctrl);
480                 return 0;
481         }
482
483         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
484         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
485         return 0;
486 }
487
488 /*
489  * Check if 'req' has a write hint associated with it. If it does, assign
490  * a valid namespace stream to the write.
491  */
492 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
493                                      struct request *req, u16 *control,
494                                      u32 *dsmgmt)
495 {
496         enum rw_hint streamid = req->write_hint;
497
498         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
499                 streamid = 0;
500         else {
501                 streamid--;
502                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
503                         return;
504
505                 *control |= NVME_RW_DTYPE_STREAMS;
506                 *dsmgmt |= streamid << 16;
507         }
508
509         if (streamid < ARRAY_SIZE(req->q->write_hints))
510                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
511 }
512
513 static inline void nvme_setup_flush(struct nvme_ns *ns,
514                 struct nvme_command *cmnd)
515 {
516         memset(cmnd, 0, sizeof(*cmnd));
517         cmnd->common.opcode = nvme_cmd_flush;
518         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
519 }
520
521 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
522                 struct nvme_command *cmnd)
523 {
524         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
525         struct nvme_dsm_range *range;
526         struct bio *bio;
527
528         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
529         if (!range)
530                 return BLK_STS_RESOURCE;
531
532         __rq_for_each_bio(bio, req) {
533                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
534                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
535
536                 if (n < segments) {
537                         range[n].cattr = cpu_to_le32(0);
538                         range[n].nlb = cpu_to_le32(nlb);
539                         range[n].slba = cpu_to_le64(slba);
540                 }
541                 n++;
542         }
543
544         if (WARN_ON_ONCE(n != segments)) {
545                 kfree(range);
546                 return BLK_STS_IOERR;
547         }
548
549         memset(cmnd, 0, sizeof(*cmnd));
550         cmnd->dsm.opcode = nvme_cmd_dsm;
551         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
552         cmnd->dsm.nr = cpu_to_le32(segments - 1);
553         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
554
555         req->special_vec.bv_page = virt_to_page(range);
556         req->special_vec.bv_offset = offset_in_page(range);
557         req->special_vec.bv_len = sizeof(*range) * segments;
558         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
559
560         return BLK_STS_OK;
561 }
562
563 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
564                 struct request *req, struct nvme_command *cmnd)
565 {
566         struct nvme_ctrl *ctrl = ns->ctrl;
567         u16 control = 0;
568         u32 dsmgmt = 0;
569
570         if (req->cmd_flags & REQ_FUA)
571                 control |= NVME_RW_FUA;
572         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
573                 control |= NVME_RW_LR;
574
575         if (req->cmd_flags & REQ_RAHEAD)
576                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
577
578         memset(cmnd, 0, sizeof(*cmnd));
579         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
580         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
581         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
582         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
583
584         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
585                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
586
587         if (ns->ms) {
588                 /*
589                  * If formated with metadata, the block layer always provides a
590                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
591                  * we enable the PRACT bit for protection information or set the
592                  * namespace capacity to zero to prevent any I/O.
593                  */
594                 if (!blk_integrity_rq(req)) {
595                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
596                                 return BLK_STS_NOTSUPP;
597                         control |= NVME_RW_PRINFO_PRACT;
598                 }
599
600                 switch (ns->pi_type) {
601                 case NVME_NS_DPS_PI_TYPE3:
602                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
603                         break;
604                 case NVME_NS_DPS_PI_TYPE1:
605                 case NVME_NS_DPS_PI_TYPE2:
606                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
607                                         NVME_RW_PRINFO_PRCHK_REF;
608                         cmnd->rw.reftag = cpu_to_le32(
609                                         nvme_block_nr(ns, blk_rq_pos(req)));
610                         break;
611                 }
612         }
613
614         cmnd->rw.control = cpu_to_le16(control);
615         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
616         return 0;
617 }
618
619 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
620                 struct nvme_command *cmd)
621 {
622         blk_status_t ret = BLK_STS_OK;
623
624         nvme_clear_nvme_request(req);
625
626         switch (req_op(req)) {
627         case REQ_OP_DRV_IN:
628         case REQ_OP_DRV_OUT:
629                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
630                 break;
631         case REQ_OP_FLUSH:
632                 nvme_setup_flush(ns, cmd);
633                 break;
634         case REQ_OP_WRITE_ZEROES:
635                 /* currently only aliased to deallocate for a few ctrls: */
636         case REQ_OP_DISCARD:
637                 ret = nvme_setup_discard(ns, req, cmd);
638                 break;
639         case REQ_OP_READ:
640         case REQ_OP_WRITE:
641                 ret = nvme_setup_rw(ns, req, cmd);
642                 break;
643         default:
644                 WARN_ON_ONCE(1);
645                 return BLK_STS_IOERR;
646         }
647
648         cmd->common.command_id = req->tag;
649         if (ns)
650                 trace_nvme_setup_nvm_cmd(req->q->id, cmd);
651         else
652                 trace_nvme_setup_admin_cmd(cmd);
653         return ret;
654 }
655 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
656
657 /*
658  * Returns 0 on success.  If the result is negative, it's a Linux error code;
659  * if the result is positive, it's an NVM Express status code
660  */
661 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
662                 union nvme_result *result, void *buffer, unsigned bufflen,
663                 unsigned timeout, int qid, int at_head,
664                 blk_mq_req_flags_t flags)
665 {
666         struct request *req;
667         int ret;
668
669         req = nvme_alloc_request(q, cmd, flags, qid);
670         if (IS_ERR(req))
671                 return PTR_ERR(req);
672
673         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
674
675         if (buffer && bufflen) {
676                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
677                 if (ret)
678                         goto out;
679         }
680
681         blk_execute_rq(req->q, NULL, req, at_head);
682         if (result)
683                 *result = nvme_req(req)->result;
684         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
685                 ret = -EINTR;
686         else
687                 ret = nvme_req(req)->status;
688  out:
689         blk_mq_free_request(req);
690         return ret;
691 }
692 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
693
694 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
695                 void *buffer, unsigned bufflen)
696 {
697         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
698                         NVME_QID_ANY, 0, 0);
699 }
700 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
701
702 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
703                 unsigned len, u32 seed, bool write)
704 {
705         struct bio_integrity_payload *bip;
706         int ret = -ENOMEM;
707         void *buf;
708
709         buf = kmalloc(len, GFP_KERNEL);
710         if (!buf)
711                 goto out;
712
713         ret = -EFAULT;
714         if (write && copy_from_user(buf, ubuf, len))
715                 goto out_free_meta;
716
717         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
718         if (IS_ERR(bip)) {
719                 ret = PTR_ERR(bip);
720                 goto out_free_meta;
721         }
722
723         bip->bip_iter.bi_size = len;
724         bip->bip_iter.bi_sector = seed;
725         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
726                         offset_in_page(buf));
727         if (ret == len)
728                 return buf;
729         ret = -ENOMEM;
730 out_free_meta:
731         kfree(buf);
732 out:
733         return ERR_PTR(ret);
734 }
735
736 static int nvme_submit_user_cmd(struct request_queue *q,
737                 struct nvme_command *cmd, void __user *ubuffer,
738                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
739                 u32 meta_seed, u32 *result, unsigned timeout)
740 {
741         bool write = nvme_is_write(cmd);
742         struct nvme_ns *ns = q->queuedata;
743         struct gendisk *disk = ns ? ns->disk : NULL;
744         struct request *req;
745         struct bio *bio = NULL;
746         void *meta = NULL;
747         int ret;
748
749         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
750         if (IS_ERR(req))
751                 return PTR_ERR(req);
752
753         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
754         nvme_req(req)->flags |= NVME_REQ_USERCMD;
755
756         if (ubuffer && bufflen) {
757                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
758                                 GFP_KERNEL);
759                 if (ret)
760                         goto out;
761                 bio = req->bio;
762                 bio->bi_disk = disk;
763                 if (disk && meta_buffer && meta_len) {
764                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
765                                         meta_seed, write);
766                         if (IS_ERR(meta)) {
767                                 ret = PTR_ERR(meta);
768                                 goto out_unmap;
769                         }
770                         req->cmd_flags |= REQ_INTEGRITY;
771                 }
772         }
773
774         blk_execute_rq(req->q, disk, req, 0);
775         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
776                 ret = -EINTR;
777         else
778                 ret = nvme_req(req)->status;
779         if (result)
780                 *result = le32_to_cpu(nvme_req(req)->result.u32);
781         if (meta && !ret && !write) {
782                 if (copy_to_user(meta_buffer, meta, meta_len))
783                         ret = -EFAULT;
784         }
785         kfree(meta);
786  out_unmap:
787         if (bio)
788                 blk_rq_unmap_user(bio);
789  out:
790         blk_mq_free_request(req);
791         return ret;
792 }
793
794 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
795 {
796         struct nvme_ctrl *ctrl = rq->end_io_data;
797
798         blk_mq_free_request(rq);
799
800         if (status) {
801                 dev_err(ctrl->device,
802                         "failed nvme_keep_alive_end_io error=%d\n",
803                                 status);
804                 return;
805         }
806
807         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
808 }
809
810 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
811 {
812         struct request *rq;
813
814         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
815                         NVME_QID_ANY);
816         if (IS_ERR(rq))
817                 return PTR_ERR(rq);
818
819         rq->timeout = ctrl->kato * HZ;
820         rq->end_io_data = ctrl;
821
822         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
823
824         return 0;
825 }
826
827 static void nvme_keep_alive_work(struct work_struct *work)
828 {
829         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
830                         struct nvme_ctrl, ka_work);
831
832         if (nvme_keep_alive(ctrl)) {
833                 /* allocation failure, reset the controller */
834                 dev_err(ctrl->device, "keep-alive failed\n");
835                 nvme_reset_ctrl(ctrl);
836                 return;
837         }
838 }
839
840 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
841 {
842         if (unlikely(ctrl->kato == 0))
843                 return;
844
845         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
846         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
847         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
848         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
849 }
850
851 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
852 {
853         if (unlikely(ctrl->kato == 0))
854                 return;
855
856         cancel_delayed_work_sync(&ctrl->ka_work);
857 }
858 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
859
860 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
861 {
862         struct nvme_command c = { };
863         int error;
864
865         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
866         c.identify.opcode = nvme_admin_identify;
867         c.identify.cns = NVME_ID_CNS_CTRL;
868
869         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
870         if (!*id)
871                 return -ENOMEM;
872
873         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
874                         sizeof(struct nvme_id_ctrl));
875         if (error)
876                 kfree(*id);
877         return error;
878 }
879
880 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
881                 struct nvme_ns_ids *ids)
882 {
883         struct nvme_command c = { };
884         int status;
885         void *data;
886         int pos;
887         int len;
888
889         c.identify.opcode = nvme_admin_identify;
890         c.identify.nsid = cpu_to_le32(nsid);
891         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
892
893         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
894         if (!data)
895                 return -ENOMEM;
896
897         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
898                                       NVME_IDENTIFY_DATA_SIZE);
899         if (status)
900                 goto free_data;
901
902         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
903                 struct nvme_ns_id_desc *cur = data + pos;
904
905                 if (cur->nidl == 0)
906                         break;
907
908                 switch (cur->nidt) {
909                 case NVME_NIDT_EUI64:
910                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
911                                 dev_warn(ctrl->device,
912                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
913                                          cur->nidl);
914                                 goto free_data;
915                         }
916                         len = NVME_NIDT_EUI64_LEN;
917                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
918                         break;
919                 case NVME_NIDT_NGUID:
920                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
921                                 dev_warn(ctrl->device,
922                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
923                                          cur->nidl);
924                                 goto free_data;
925                         }
926                         len = NVME_NIDT_NGUID_LEN;
927                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
928                         break;
929                 case NVME_NIDT_UUID:
930                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
931                                 dev_warn(ctrl->device,
932                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
933                                          cur->nidl);
934                                 goto free_data;
935                         }
936                         len = NVME_NIDT_UUID_LEN;
937                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
938                         break;
939                 default:
940                         /* Skip unnkown types */
941                         len = cur->nidl;
942                         break;
943                 }
944
945                 len += sizeof(*cur);
946         }
947 free_data:
948         kfree(data);
949         return status;
950 }
951
952 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
953 {
954         struct nvme_command c = { };
955
956         c.identify.opcode = nvme_admin_identify;
957         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
958         c.identify.nsid = cpu_to_le32(nsid);
959         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
960                                     NVME_IDENTIFY_DATA_SIZE);
961 }
962
963 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
964                 unsigned nsid)
965 {
966         struct nvme_id_ns *id;
967         struct nvme_command c = { };
968         int error;
969
970         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
971         c.identify.opcode = nvme_admin_identify;
972         c.identify.nsid = cpu_to_le32(nsid);
973         c.identify.cns = NVME_ID_CNS_NS;
974
975         id = kmalloc(sizeof(*id), GFP_KERNEL);
976         if (!id)
977                 return NULL;
978
979         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
980         if (error) {
981                 dev_warn(ctrl->device, "Identify namespace failed\n");
982                 kfree(id);
983                 return NULL;
984         }
985
986         return id;
987 }
988
989 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
990                       void *buffer, size_t buflen, u32 *result)
991 {
992         struct nvme_command c;
993         union nvme_result res;
994         int ret;
995
996         memset(&c, 0, sizeof(c));
997         c.features.opcode = nvme_admin_set_features;
998         c.features.fid = cpu_to_le32(fid);
999         c.features.dword11 = cpu_to_le32(dword11);
1000
1001         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1002                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1003         if (ret >= 0 && result)
1004                 *result = le32_to_cpu(res.u32);
1005         return ret;
1006 }
1007
1008 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1009 {
1010         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1011         u32 result;
1012         int status, nr_io_queues;
1013
1014         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1015                         &result);
1016         if (status < 0)
1017                 return status;
1018
1019         /*
1020          * Degraded controllers might return an error when setting the queue
1021          * count.  We still want to be able to bring them online and offer
1022          * access to the admin queue, as that might be only way to fix them up.
1023          */
1024         if (status > 0) {
1025                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1026                 *count = 0;
1027         } else {
1028                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1029                 *count = min(*count, nr_io_queues);
1030         }
1031
1032         return 0;
1033 }
1034 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1035
1036 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1037 {
1038         struct nvme_user_io io;
1039         struct nvme_command c;
1040         unsigned length, meta_len;
1041         void __user *metadata;
1042
1043         if (copy_from_user(&io, uio, sizeof(io)))
1044                 return -EFAULT;
1045         if (io.flags)
1046                 return -EINVAL;
1047
1048         switch (io.opcode) {
1049         case nvme_cmd_write:
1050         case nvme_cmd_read:
1051         case nvme_cmd_compare:
1052                 break;
1053         default:
1054                 return -EINVAL;
1055         }
1056
1057         length = (io.nblocks + 1) << ns->lba_shift;
1058         meta_len = (io.nblocks + 1) * ns->ms;
1059         metadata = (void __user *)(uintptr_t)io.metadata;
1060
1061         if (ns->ext) {
1062                 length += meta_len;
1063                 meta_len = 0;
1064         } else if (meta_len) {
1065                 if ((io.metadata & 3) || !io.metadata)
1066                         return -EINVAL;
1067         }
1068
1069         memset(&c, 0, sizeof(c));
1070         c.rw.opcode = io.opcode;
1071         c.rw.flags = io.flags;
1072         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1073         c.rw.slba = cpu_to_le64(io.slba);
1074         c.rw.length = cpu_to_le16(io.nblocks);
1075         c.rw.control = cpu_to_le16(io.control);
1076         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1077         c.rw.reftag = cpu_to_le32(io.reftag);
1078         c.rw.apptag = cpu_to_le16(io.apptag);
1079         c.rw.appmask = cpu_to_le16(io.appmask);
1080
1081         return nvme_submit_user_cmd(ns->queue, &c,
1082                         (void __user *)(uintptr_t)io.addr, length,
1083                         metadata, meta_len, io.slba, NULL, 0);
1084 }
1085
1086 static u32 nvme_known_admin_effects(u8 opcode)
1087 {
1088         switch (opcode) {
1089         case nvme_admin_format_nvm:
1090                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1091                                         NVME_CMD_EFFECTS_CSE_MASK;
1092         case nvme_admin_sanitize_nvm:
1093                 return NVME_CMD_EFFECTS_CSE_MASK;
1094         default:
1095                 break;
1096         }
1097         return 0;
1098 }
1099
1100 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1101                                                                 u8 opcode)
1102 {
1103         u32 effects = 0;
1104
1105         if (ns) {
1106                 if (ctrl->effects)
1107                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1108                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1109                         dev_warn(ctrl->device,
1110                                  "IO command:%02x has unhandled effects:%08x\n",
1111                                  opcode, effects);
1112                 return 0;
1113         }
1114
1115         if (ctrl->effects)
1116                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1117         else
1118                 effects = nvme_known_admin_effects(opcode);
1119
1120         /*
1121          * For simplicity, IO to all namespaces is quiesced even if the command
1122          * effects say only one namespace is affected.
1123          */
1124         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1125                 nvme_start_freeze(ctrl);
1126                 nvme_wait_freeze(ctrl);
1127         }
1128         return effects;
1129 }
1130
1131 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1132 {
1133         struct nvme_ns *ns, *next;
1134         LIST_HEAD(rm_list);
1135
1136         down_write(&ctrl->namespaces_rwsem);
1137         list_for_each_entry(ns, &ctrl->namespaces, list) {
1138                 if (ns->disk && nvme_revalidate_disk(ns->disk)) {
1139                         list_move_tail(&ns->list, &rm_list);
1140                 }
1141         }
1142         up_write(&ctrl->namespaces_rwsem);
1143
1144         list_for_each_entry_safe(ns, next, &rm_list, list)
1145                 nvme_ns_remove(ns);
1146 }
1147
1148 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1149 {
1150         /*
1151          * Revalidate LBA changes prior to unfreezing. This is necessary to
1152          * prevent memory corruption if a logical block size was changed by
1153          * this command.
1154          */
1155         if (effects & NVME_CMD_EFFECTS_LBCC)
1156                 nvme_update_formats(ctrl);
1157         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1158                 nvme_unfreeze(ctrl);
1159         if (effects & NVME_CMD_EFFECTS_CCC)
1160                 nvme_init_identify(ctrl);
1161         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1162                 nvme_queue_scan(ctrl);
1163 }
1164
1165 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1166                         struct nvme_passthru_cmd __user *ucmd)
1167 {
1168         struct nvme_passthru_cmd cmd;
1169         struct nvme_command c;
1170         unsigned timeout = 0;
1171         u32 effects;
1172         int status;
1173
1174         if (!capable(CAP_SYS_ADMIN))
1175                 return -EACCES;
1176         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1177                 return -EFAULT;
1178         if (cmd.flags)
1179                 return -EINVAL;
1180
1181         memset(&c, 0, sizeof(c));
1182         c.common.opcode = cmd.opcode;
1183         c.common.flags = cmd.flags;
1184         c.common.nsid = cpu_to_le32(cmd.nsid);
1185         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1186         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1187         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1188         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1189         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1190         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1191         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1192         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1193
1194         if (cmd.timeout_ms)
1195                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1196
1197         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1198         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1199                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1200                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1201                         0, &cmd.result, timeout);
1202         nvme_passthru_end(ctrl, effects);
1203
1204         if (status >= 0) {
1205                 if (put_user(cmd.result, &ucmd->result))
1206                         return -EFAULT;
1207         }
1208
1209         return status;
1210 }
1211
1212 /*
1213  * Issue ioctl requests on the first available path.  Note that unlike normal
1214  * block layer requests we will not retry failed request on another controller.
1215  */
1216 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1217                 struct nvme_ns_head **head, int *srcu_idx)
1218 {
1219 #ifdef CONFIG_NVME_MULTIPATH
1220         if (disk->fops == &nvme_ns_head_ops) {
1221                 *head = disk->private_data;
1222                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1223                 return nvme_find_path(*head);
1224         }
1225 #endif
1226         *head = NULL;
1227         *srcu_idx = -1;
1228         return disk->private_data;
1229 }
1230
1231 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1232 {
1233         if (head)
1234                 srcu_read_unlock(&head->srcu, idx);
1235 }
1236
1237 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1238 {
1239         switch (cmd) {
1240         case NVME_IOCTL_ID:
1241                 force_successful_syscall_return();
1242                 return ns->head->ns_id;
1243         case NVME_IOCTL_ADMIN_CMD:
1244                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1245         case NVME_IOCTL_IO_CMD:
1246                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1247         case NVME_IOCTL_SUBMIT_IO:
1248                 return nvme_submit_io(ns, (void __user *)arg);
1249         default:
1250 #ifdef CONFIG_NVM
1251                 if (ns->ndev)
1252                         return nvme_nvm_ioctl(ns, cmd, arg);
1253 #endif
1254                 if (is_sed_ioctl(cmd))
1255                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1256                                          (void __user *) arg);
1257                 return -ENOTTY;
1258         }
1259 }
1260
1261 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1262                 unsigned int cmd, unsigned long arg)
1263 {
1264         struct nvme_ns_head *head = NULL;
1265         struct nvme_ns *ns;
1266         int srcu_idx, ret;
1267
1268         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1269         if (unlikely(!ns))
1270                 ret = -EWOULDBLOCK;
1271         else
1272                 ret = nvme_ns_ioctl(ns, cmd, arg);
1273         nvme_put_ns_from_disk(head, srcu_idx);
1274         return ret;
1275 }
1276
1277 static int nvme_open(struct block_device *bdev, fmode_t mode)
1278 {
1279         struct nvme_ns *ns = bdev->bd_disk->private_data;
1280
1281 #ifdef CONFIG_NVME_MULTIPATH
1282         /* should never be called due to GENHD_FL_HIDDEN */
1283         if (WARN_ON_ONCE(ns->head->disk))
1284                 goto fail;
1285 #endif
1286         if (!kref_get_unless_zero(&ns->kref))
1287                 goto fail;
1288         if (!try_module_get(ns->ctrl->ops->module))
1289                 goto fail_put_ns;
1290
1291         return 0;
1292
1293 fail_put_ns:
1294         nvme_put_ns(ns);
1295 fail:
1296         return -ENXIO;
1297 }
1298
1299 static void nvme_release(struct gendisk *disk, fmode_t mode)
1300 {
1301         struct nvme_ns *ns = disk->private_data;
1302
1303         module_put(ns->ctrl->ops->module);
1304         nvme_put_ns(ns);
1305 }
1306
1307 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1308 {
1309         /* some standard values */
1310         geo->heads = 1 << 6;
1311         geo->sectors = 1 << 5;
1312         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1313         return 0;
1314 }
1315
1316 #ifdef CONFIG_BLK_DEV_INTEGRITY
1317 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1318 {
1319         struct blk_integrity integrity;
1320
1321         memset(&integrity, 0, sizeof(integrity));
1322         switch (pi_type) {
1323         case NVME_NS_DPS_PI_TYPE3:
1324                 integrity.profile = &t10_pi_type3_crc;
1325                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1326                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1327                 break;
1328         case NVME_NS_DPS_PI_TYPE1:
1329         case NVME_NS_DPS_PI_TYPE2:
1330                 integrity.profile = &t10_pi_type1_crc;
1331                 integrity.tag_size = sizeof(u16);
1332                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1333                 break;
1334         default:
1335                 integrity.profile = NULL;
1336                 break;
1337         }
1338         integrity.tuple_size = ms;
1339         blk_integrity_register(disk, &integrity);
1340         blk_queue_max_integrity_segments(disk->queue, 1);
1341 }
1342 #else
1343 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1344 {
1345 }
1346 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1347
1348 static void nvme_set_chunk_size(struct nvme_ns *ns)
1349 {
1350         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1351         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1352 }
1353
1354 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1355                 unsigned stream_alignment, struct request_queue *queue)
1356 {
1357         u32 size = queue_logical_block_size(queue);
1358
1359         if (stream_alignment)
1360                 size *= stream_alignment;
1361
1362         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1363                         NVME_DSM_MAX_RANGES);
1364
1365         queue->limits.discard_alignment = 0;
1366         queue->limits.discard_granularity = size;
1367
1368         blk_queue_max_discard_sectors(queue, UINT_MAX);
1369         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1370         blk_queue_flag_set(QUEUE_FLAG_DISCARD, queue);
1371
1372         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1373                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1374 }
1375
1376 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1377                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1378 {
1379         memset(ids, 0, sizeof(*ids));
1380
1381         if (ctrl->vs >= NVME_VS(1, 1, 0))
1382                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1383         if (ctrl->vs >= NVME_VS(1, 2, 0))
1384                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1385         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1386                  /* Don't treat error as fatal we potentially
1387                   * already have a NGUID or EUI-64
1388                   */
1389                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1390                         dev_warn(ctrl->device,
1391                                  "%s: Identify Descriptors failed\n", __func__);
1392         }
1393 }
1394
1395 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1396 {
1397         return !uuid_is_null(&ids->uuid) ||
1398                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1399                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1400 }
1401
1402 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1403 {
1404         return uuid_equal(&a->uuid, &b->uuid) &&
1405                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1406                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1407 }
1408
1409 static void nvme_update_disk_info(struct gendisk *disk,
1410                 struct nvme_ns *ns, struct nvme_id_ns *id)
1411 {
1412         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1413         unsigned short bs = 1 << ns->lba_shift;
1414         unsigned stream_alignment = 0;
1415
1416         if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1417                 stream_alignment = ns->sws * ns->sgs;
1418
1419         blk_mq_freeze_queue(disk->queue);
1420         blk_integrity_unregister(disk);
1421
1422         blk_queue_logical_block_size(disk->queue, bs);
1423         blk_queue_physical_block_size(disk->queue, bs);
1424         blk_queue_io_min(disk->queue, bs);
1425
1426         if (ns->ms && !ns->ext &&
1427             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1428                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1429         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1430                 capacity = 0;
1431         set_capacity(disk, capacity);
1432
1433         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1434                 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1435         blk_mq_unfreeze_queue(disk->queue);
1436 }
1437
1438 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1439 {
1440         struct nvme_ns *ns = disk->private_data;
1441
1442         /*
1443          * If identify namespace failed, use default 512 byte block size so
1444          * block layer can use before failing read/write for 0 capacity.
1445          */
1446         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1447         if (ns->lba_shift == 0)
1448                 ns->lba_shift = 9;
1449         ns->noiob = le16_to_cpu(id->noiob);
1450         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1451         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1452         /* the PI implementation requires metadata equal t10 pi tuple size */
1453         if (ns->ms == sizeof(struct t10_pi_tuple))
1454                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1455         else
1456                 ns->pi_type = 0;
1457
1458         if (ns->noiob)
1459                 nvme_set_chunk_size(ns);
1460         nvme_update_disk_info(disk, ns, id);
1461         if (ns->ndev)
1462                 nvme_nvm_update_nvm_info(ns);
1463 #ifdef CONFIG_NVME_MULTIPATH
1464         if (ns->head->disk)
1465                 nvme_update_disk_info(ns->head->disk, ns, id);
1466 #endif
1467 }
1468
1469 static int nvme_revalidate_disk(struct gendisk *disk)
1470 {
1471         struct nvme_ns *ns = disk->private_data;
1472         struct nvme_ctrl *ctrl = ns->ctrl;
1473         struct nvme_id_ns *id;
1474         struct nvme_ns_ids ids;
1475         int ret = 0;
1476
1477         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1478                 set_capacity(disk, 0);
1479                 return -ENODEV;
1480         }
1481
1482         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1483         if (!id)
1484                 return -ENODEV;
1485
1486         if (id->ncap == 0) {
1487                 ret = -ENODEV;
1488                 goto out;
1489         }
1490
1491         __nvme_revalidate_disk(disk, id);
1492         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1493         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1494                 dev_err(ctrl->device,
1495                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1496                 ret = -ENODEV;
1497         }
1498
1499 out:
1500         kfree(id);
1501         return ret;
1502 }
1503
1504 static char nvme_pr_type(enum pr_type type)
1505 {
1506         switch (type) {
1507         case PR_WRITE_EXCLUSIVE:
1508                 return 1;
1509         case PR_EXCLUSIVE_ACCESS:
1510                 return 2;
1511         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1512                 return 3;
1513         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1514                 return 4;
1515         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1516                 return 5;
1517         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1518                 return 6;
1519         default:
1520                 return 0;
1521         }
1522 };
1523
1524 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1525                                 u64 key, u64 sa_key, u8 op)
1526 {
1527         struct nvme_ns_head *head = NULL;
1528         struct nvme_ns *ns;
1529         struct nvme_command c;
1530         int srcu_idx, ret;
1531         u8 data[16] = { 0, };
1532
1533         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1534         if (unlikely(!ns))
1535                 return -EWOULDBLOCK;
1536
1537         put_unaligned_le64(key, &data[0]);
1538         put_unaligned_le64(sa_key, &data[8]);
1539
1540         memset(&c, 0, sizeof(c));
1541         c.common.opcode = op;
1542         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1543         c.common.cdw10[0] = cpu_to_le32(cdw10);
1544
1545         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1546         nvme_put_ns_from_disk(head, srcu_idx);
1547         return ret;
1548 }
1549
1550 static int nvme_pr_register(struct block_device *bdev, u64 old,
1551                 u64 new, unsigned flags)
1552 {
1553         u32 cdw10;
1554
1555         if (flags & ~PR_FL_IGNORE_KEY)
1556                 return -EOPNOTSUPP;
1557
1558         cdw10 = old ? 2 : 0;
1559         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1560         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1561         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1562 }
1563
1564 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1565                 enum pr_type type, unsigned flags)
1566 {
1567         u32 cdw10;
1568
1569         if (flags & ~PR_FL_IGNORE_KEY)
1570                 return -EOPNOTSUPP;
1571
1572         cdw10 = nvme_pr_type(type) << 8;
1573         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1574         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1575 }
1576
1577 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1578                 enum pr_type type, bool abort)
1579 {
1580         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1581         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1582 }
1583
1584 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1585 {
1586         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1587         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1588 }
1589
1590 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1591 {
1592         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1593         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1594 }
1595
1596 static const struct pr_ops nvme_pr_ops = {
1597         .pr_register    = nvme_pr_register,
1598         .pr_reserve     = nvme_pr_reserve,
1599         .pr_release     = nvme_pr_release,
1600         .pr_preempt     = nvme_pr_preempt,
1601         .pr_clear       = nvme_pr_clear,
1602 };
1603
1604 #ifdef CONFIG_BLK_SED_OPAL
1605 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1606                 bool send)
1607 {
1608         struct nvme_ctrl *ctrl = data;
1609         struct nvme_command cmd;
1610
1611         memset(&cmd, 0, sizeof(cmd));
1612         if (send)
1613                 cmd.common.opcode = nvme_admin_security_send;
1614         else
1615                 cmd.common.opcode = nvme_admin_security_recv;
1616         cmd.common.nsid = 0;
1617         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1618         cmd.common.cdw10[1] = cpu_to_le32(len);
1619
1620         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1621                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1622 }
1623 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1624 #endif /* CONFIG_BLK_SED_OPAL */
1625
1626 static const struct block_device_operations nvme_fops = {
1627         .owner          = THIS_MODULE,
1628         .ioctl          = nvme_ioctl,
1629         .compat_ioctl   = nvme_ioctl,
1630         .open           = nvme_open,
1631         .release        = nvme_release,
1632         .getgeo         = nvme_getgeo,
1633         .revalidate_disk= nvme_revalidate_disk,
1634         .pr_ops         = &nvme_pr_ops,
1635 };
1636
1637 #ifdef CONFIG_NVME_MULTIPATH
1638 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1639 {
1640         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1641
1642         if (!kref_get_unless_zero(&head->ref))
1643                 return -ENXIO;
1644         return 0;
1645 }
1646
1647 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1648 {
1649         nvme_put_ns_head(disk->private_data);
1650 }
1651
1652 const struct block_device_operations nvme_ns_head_ops = {
1653         .owner          = THIS_MODULE,
1654         .open           = nvme_ns_head_open,
1655         .release        = nvme_ns_head_release,
1656         .ioctl          = nvme_ioctl,
1657         .compat_ioctl   = nvme_ioctl,
1658         .getgeo         = nvme_getgeo,
1659         .pr_ops         = &nvme_pr_ops,
1660 };
1661 #endif /* CONFIG_NVME_MULTIPATH */
1662
1663 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1664 {
1665         unsigned long timeout =
1666                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1667         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1668         int ret;
1669
1670         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1671                 if (csts == ~0)
1672                         return -ENODEV;
1673                 if ((csts & NVME_CSTS_RDY) == bit)
1674                         break;
1675
1676                 msleep(100);
1677                 if (fatal_signal_pending(current))
1678                         return -EINTR;
1679                 if (time_after(jiffies, timeout)) {
1680                         dev_err(ctrl->device,
1681                                 "Device not ready; aborting %s\n", enabled ?
1682                                                 "initialisation" : "reset");
1683                         return -ENODEV;
1684                 }
1685         }
1686
1687         return ret;
1688 }
1689
1690 /*
1691  * If the device has been passed off to us in an enabled state, just clear
1692  * the enabled bit.  The spec says we should set the 'shutdown notification
1693  * bits', but doing so may cause the device to complete commands to the
1694  * admin queue ... and we don't know what memory that might be pointing at!
1695  */
1696 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1697 {
1698         int ret;
1699
1700         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1701         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1702
1703         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1704         if (ret)
1705                 return ret;
1706
1707         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1708                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1709
1710         return nvme_wait_ready(ctrl, cap, false);
1711 }
1712 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1713
1714 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1715 {
1716         /*
1717          * Default to a 4K page size, with the intention to update this
1718          * path in the future to accomodate architectures with differing
1719          * kernel and IO page sizes.
1720          */
1721         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1722         int ret;
1723
1724         if (page_shift < dev_page_min) {
1725                 dev_err(ctrl->device,
1726                         "Minimum device page size %u too large for host (%u)\n",
1727                         1 << dev_page_min, 1 << page_shift);
1728                 return -ENODEV;
1729         }
1730
1731         ctrl->page_size = 1 << page_shift;
1732
1733         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1734         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1735         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1736         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1737         ctrl->ctrl_config |= NVME_CC_ENABLE;
1738
1739         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1740         if (ret)
1741                 return ret;
1742         return nvme_wait_ready(ctrl, cap, true);
1743 }
1744 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1745
1746 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1747 {
1748         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1749         u32 csts;
1750         int ret;
1751
1752         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1753         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1754
1755         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1756         if (ret)
1757                 return ret;
1758
1759         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1760                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1761                         break;
1762
1763                 msleep(100);
1764                 if (fatal_signal_pending(current))
1765                         return -EINTR;
1766                 if (time_after(jiffies, timeout)) {
1767                         dev_err(ctrl->device,
1768                                 "Device shutdown incomplete; abort shutdown\n");
1769                         return -ENODEV;
1770                 }
1771         }
1772
1773         return ret;
1774 }
1775 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1776
1777 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1778                 struct request_queue *q)
1779 {
1780         bool vwc = false;
1781
1782         if (ctrl->max_hw_sectors) {
1783                 u32 max_segments =
1784                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1785
1786                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1787                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1788         }
1789         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1790             is_power_of_2(ctrl->max_hw_sectors))
1791                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1792         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1793         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1794                 vwc = true;
1795         blk_queue_write_cache(q, vwc, vwc);
1796 }
1797
1798 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1799 {
1800         __le64 ts;
1801         int ret;
1802
1803         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1804                 return 0;
1805
1806         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1807         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1808                         NULL);
1809         if (ret)
1810                 dev_warn_once(ctrl->device,
1811                         "could not set timestamp (%d)\n", ret);
1812         return ret;
1813 }
1814
1815 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1816 {
1817         /*
1818          * APST (Autonomous Power State Transition) lets us program a
1819          * table of power state transitions that the controller will
1820          * perform automatically.  We configure it with a simple
1821          * heuristic: we are willing to spend at most 2% of the time
1822          * transitioning between power states.  Therefore, when running
1823          * in any given state, we will enter the next lower-power
1824          * non-operational state after waiting 50 * (enlat + exlat)
1825          * microseconds, as long as that state's exit latency is under
1826          * the requested maximum latency.
1827          *
1828          * We will not autonomously enter any non-operational state for
1829          * which the total latency exceeds ps_max_latency_us.  Users
1830          * can set ps_max_latency_us to zero to turn off APST.
1831          */
1832
1833         unsigned apste;
1834         struct nvme_feat_auto_pst *table;
1835         u64 max_lat_us = 0;
1836         int max_ps = -1;
1837         int ret;
1838
1839         /*
1840          * If APST isn't supported or if we haven't been initialized yet,
1841          * then don't do anything.
1842          */
1843         if (!ctrl->apsta)
1844                 return 0;
1845
1846         if (ctrl->npss > 31) {
1847                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1848                 return 0;
1849         }
1850
1851         table = kzalloc(sizeof(*table), GFP_KERNEL);
1852         if (!table)
1853                 return 0;
1854
1855         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1856                 /* Turn off APST. */
1857                 apste = 0;
1858                 dev_dbg(ctrl->device, "APST disabled\n");
1859         } else {
1860                 __le64 target = cpu_to_le64(0);
1861                 int state;
1862
1863                 /*
1864                  * Walk through all states from lowest- to highest-power.
1865                  * According to the spec, lower-numbered states use more
1866                  * power.  NPSS, despite the name, is the index of the
1867                  * lowest-power state, not the number of states.
1868                  */
1869                 for (state = (int)ctrl->npss; state >= 0; state--) {
1870                         u64 total_latency_us, exit_latency_us, transition_ms;
1871
1872                         if (target)
1873                                 table->entries[state] = target;
1874
1875                         /*
1876                          * Don't allow transitions to the deepest state
1877                          * if it's quirked off.
1878                          */
1879                         if (state == ctrl->npss &&
1880                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1881                                 continue;
1882
1883                         /*
1884                          * Is this state a useful non-operational state for
1885                          * higher-power states to autonomously transition to?
1886                          */
1887                         if (!(ctrl->psd[state].flags &
1888                               NVME_PS_FLAGS_NON_OP_STATE))
1889                                 continue;
1890
1891                         exit_latency_us =
1892                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1893                         if (exit_latency_us > ctrl->ps_max_latency_us)
1894                                 continue;
1895
1896                         total_latency_us =
1897                                 exit_latency_us +
1898                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1899
1900                         /*
1901                          * This state is good.  Use it as the APST idle
1902                          * target for higher power states.
1903                          */
1904                         transition_ms = total_latency_us + 19;
1905                         do_div(transition_ms, 20);
1906                         if (transition_ms > (1 << 24) - 1)
1907                                 transition_ms = (1 << 24) - 1;
1908
1909                         target = cpu_to_le64((state << 3) |
1910                                              (transition_ms << 8));
1911
1912                         if (max_ps == -1)
1913                                 max_ps = state;
1914
1915                         if (total_latency_us > max_lat_us)
1916                                 max_lat_us = total_latency_us;
1917                 }
1918
1919                 apste = 1;
1920
1921                 if (max_ps == -1) {
1922                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1923                 } else {
1924                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1925                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1926                 }
1927         }
1928
1929         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1930                                 table, sizeof(*table), NULL);
1931         if (ret)
1932                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1933
1934         kfree(table);
1935         return ret;
1936 }
1937
1938 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1939 {
1940         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1941         u64 latency;
1942
1943         switch (val) {
1944         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1945         case PM_QOS_LATENCY_ANY:
1946                 latency = U64_MAX;
1947                 break;
1948
1949         default:
1950                 latency = val;
1951         }
1952
1953         if (ctrl->ps_max_latency_us != latency) {
1954                 ctrl->ps_max_latency_us = latency;
1955                 nvme_configure_apst(ctrl);
1956         }
1957 }
1958
1959 struct nvme_core_quirk_entry {
1960         /*
1961          * NVMe model and firmware strings are padded with spaces.  For
1962          * simplicity, strings in the quirk table are padded with NULLs
1963          * instead.
1964          */
1965         u16 vid;
1966         const char *mn;
1967         const char *fr;
1968         unsigned long quirks;
1969 };
1970
1971 static const struct nvme_core_quirk_entry core_quirks[] = {
1972         {
1973                 /*
1974                  * This Toshiba device seems to die using any APST states.  See:
1975                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1976                  */
1977                 .vid = 0x1179,
1978                 .mn = "THNSF5256GPUK TOSHIBA",
1979                 .quirks = NVME_QUIRK_NO_APST,
1980         }
1981 };
1982
1983 /* match is null-terminated but idstr is space-padded. */
1984 static bool string_matches(const char *idstr, const char *match, size_t len)
1985 {
1986         size_t matchlen;
1987
1988         if (!match)
1989                 return true;
1990
1991         matchlen = strlen(match);
1992         WARN_ON_ONCE(matchlen > len);
1993
1994         if (memcmp(idstr, match, matchlen))
1995                 return false;
1996
1997         for (; matchlen < len; matchlen++)
1998                 if (idstr[matchlen] != ' ')
1999                         return false;
2000
2001         return true;
2002 }
2003
2004 static bool quirk_matches(const struct nvme_id_ctrl *id,
2005                           const struct nvme_core_quirk_entry *q)
2006 {
2007         return q->vid == le16_to_cpu(id->vid) &&
2008                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2009                 string_matches(id->fr, q->fr, sizeof(id->fr));
2010 }
2011
2012 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2013                 struct nvme_id_ctrl *id)
2014 {
2015         size_t nqnlen;
2016         int off;
2017
2018         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2019         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2020                 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2021                 return;
2022         }
2023
2024         if (ctrl->vs >= NVME_VS(1, 2, 1))
2025                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2026
2027         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2028         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2029                         "nqn.2014.08.org.nvmexpress:%4x%4x",
2030                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2031         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2032         off += sizeof(id->sn);
2033         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2034         off += sizeof(id->mn);
2035         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2036 }
2037
2038 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2039 {
2040         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2041         kfree(subsys);
2042 }
2043
2044 static void nvme_release_subsystem(struct device *dev)
2045 {
2046         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2047 }
2048
2049 static void nvme_destroy_subsystem(struct kref *ref)
2050 {
2051         struct nvme_subsystem *subsys =
2052                         container_of(ref, struct nvme_subsystem, ref);
2053
2054         mutex_lock(&nvme_subsystems_lock);
2055         list_del(&subsys->entry);
2056         mutex_unlock(&nvme_subsystems_lock);
2057
2058         ida_destroy(&subsys->ns_ida);
2059         device_del(&subsys->dev);
2060         put_device(&subsys->dev);
2061 }
2062
2063 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2064 {
2065         kref_put(&subsys->ref, nvme_destroy_subsystem);
2066 }
2067
2068 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2069 {
2070         struct nvme_subsystem *subsys;
2071
2072         lockdep_assert_held(&nvme_subsystems_lock);
2073
2074         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2075                 if (strcmp(subsys->subnqn, subsysnqn))
2076                         continue;
2077                 if (!kref_get_unless_zero(&subsys->ref))
2078                         continue;
2079                 return subsys;
2080         }
2081
2082         return NULL;
2083 }
2084
2085 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2086         struct device_attribute subsys_attr_##_name = \
2087                 __ATTR(_name, _mode, _show, NULL)
2088
2089 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2090                                     struct device_attribute *attr,
2091                                     char *buf)
2092 {
2093         struct nvme_subsystem *subsys =
2094                 container_of(dev, struct nvme_subsystem, dev);
2095
2096         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2097 }
2098 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2099
2100 #define nvme_subsys_show_str_function(field)                            \
2101 static ssize_t subsys_##field##_show(struct device *dev,                \
2102                             struct device_attribute *attr, char *buf)   \
2103 {                                                                       \
2104         struct nvme_subsystem *subsys =                                 \
2105                 container_of(dev, struct nvme_subsystem, dev);          \
2106         return sprintf(buf, "%.*s\n",                                   \
2107                        (int)sizeof(subsys->field), subsys->field);      \
2108 }                                                                       \
2109 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2110
2111 nvme_subsys_show_str_function(model);
2112 nvme_subsys_show_str_function(serial);
2113 nvme_subsys_show_str_function(firmware_rev);
2114
2115 static struct attribute *nvme_subsys_attrs[] = {
2116         &subsys_attr_model.attr,
2117         &subsys_attr_serial.attr,
2118         &subsys_attr_firmware_rev.attr,
2119         &subsys_attr_subsysnqn.attr,
2120         NULL,
2121 };
2122
2123 static struct attribute_group nvme_subsys_attrs_group = {
2124         .attrs = nvme_subsys_attrs,
2125 };
2126
2127 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2128         &nvme_subsys_attrs_group,
2129         NULL,
2130 };
2131
2132 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2133 {
2134         int count = 0;
2135         struct nvme_ctrl *ctrl;
2136
2137         mutex_lock(&subsys->lock);
2138         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2139                 if (ctrl->state != NVME_CTRL_DELETING &&
2140                     ctrl->state != NVME_CTRL_DEAD)
2141                         count++;
2142         }
2143         mutex_unlock(&subsys->lock);
2144
2145         return count;
2146 }
2147
2148 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2149 {
2150         struct nvme_subsystem *subsys, *found;
2151         int ret;
2152
2153         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2154         if (!subsys)
2155                 return -ENOMEM;
2156         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2157         if (ret < 0) {
2158                 kfree(subsys);
2159                 return ret;
2160         }
2161         subsys->instance = ret;
2162         mutex_init(&subsys->lock);
2163         kref_init(&subsys->ref);
2164         INIT_LIST_HEAD(&subsys->ctrls);
2165         INIT_LIST_HEAD(&subsys->nsheads);
2166         nvme_init_subnqn(subsys, ctrl, id);
2167         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2168         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2169         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2170         subsys->vendor_id = le16_to_cpu(id->vid);
2171         subsys->cmic = id->cmic;
2172
2173         subsys->dev.class = nvme_subsys_class;
2174         subsys->dev.release = nvme_release_subsystem;
2175         subsys->dev.groups = nvme_subsys_attrs_groups;
2176         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2177         device_initialize(&subsys->dev);
2178
2179         mutex_lock(&nvme_subsystems_lock);
2180         found = __nvme_find_get_subsystem(subsys->subnqn);
2181         if (found) {
2182                 /*
2183                  * Verify that the subsystem actually supports multiple
2184                  * controllers, else bail out.
2185                  */
2186                 if (nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2187                         dev_err(ctrl->device,
2188                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2189                                 found->subnqn);
2190                         nvme_put_subsystem(found);
2191                         ret = -EINVAL;
2192                         goto out_unlock;
2193                 }
2194
2195                 __nvme_release_subsystem(subsys);
2196                 subsys = found;
2197         } else {
2198                 ret = device_add(&subsys->dev);
2199                 if (ret) {
2200                         dev_err(ctrl->device,
2201                                 "failed to register subsystem device.\n");
2202                         goto out_unlock;
2203                 }
2204                 ida_init(&subsys->ns_ida);
2205                 list_add_tail(&subsys->entry, &nvme_subsystems);
2206         }
2207
2208         ctrl->subsys = subsys;
2209         mutex_unlock(&nvme_subsystems_lock);
2210
2211         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2212                         dev_name(ctrl->device))) {
2213                 dev_err(ctrl->device,
2214                         "failed to create sysfs link from subsystem.\n");
2215                 /* the transport driver will eventually put the subsystem */
2216                 return -EINVAL;
2217         }
2218
2219         mutex_lock(&subsys->lock);
2220         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2221         mutex_unlock(&subsys->lock);
2222
2223         return 0;
2224
2225 out_unlock:
2226         mutex_unlock(&nvme_subsystems_lock);
2227         put_device(&subsys->dev);
2228         return ret;
2229 }
2230
2231 int nvme_get_log_ext(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
2232                      u8 log_page, void *log,
2233                      size_t size, u64 offset)
2234 {
2235         struct nvme_command c = { };
2236         unsigned long dwlen = size / 4 - 1;
2237
2238         c.get_log_page.opcode = nvme_admin_get_log_page;
2239
2240         if (ns)
2241                 c.get_log_page.nsid = cpu_to_le32(ns->head->ns_id);
2242         else
2243                 c.get_log_page.nsid = cpu_to_le32(NVME_NSID_ALL);
2244
2245         c.get_log_page.lid = log_page;
2246         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2247         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2248         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2249         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2250
2251         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2252 }
2253
2254 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2255                         size_t size)
2256 {
2257         return nvme_get_log_ext(ctrl, NULL, log_page, log, size, 0);
2258 }
2259
2260 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2261 {
2262         int ret;
2263
2264         if (!ctrl->effects)
2265                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2266
2267         if (!ctrl->effects)
2268                 return 0;
2269
2270         ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2271                                         sizeof(*ctrl->effects));
2272         if (ret) {
2273                 kfree(ctrl->effects);
2274                 ctrl->effects = NULL;
2275         }
2276         return ret;
2277 }
2278
2279 /*
2280  * Initialize the cached copies of the Identify data and various controller
2281  * register in our nvme_ctrl structure.  This should be called as soon as
2282  * the admin queue is fully up and running.
2283  */
2284 int nvme_init_identify(struct nvme_ctrl *ctrl)
2285 {
2286         struct nvme_id_ctrl *id;
2287         u64 cap;
2288         int ret, page_shift;
2289         u32 max_hw_sectors;
2290         bool prev_apst_enabled;
2291
2292         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2293         if (ret) {
2294                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2295                 return ret;
2296         }
2297
2298         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2299         if (ret) {
2300                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2301                 return ret;
2302         }
2303         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2304
2305         if (ctrl->vs >= NVME_VS(1, 1, 0))
2306                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2307
2308         ret = nvme_identify_ctrl(ctrl, &id);
2309         if (ret) {
2310                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2311                 return -EIO;
2312         }
2313
2314         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2315                 ret = nvme_get_effects_log(ctrl);
2316                 if (ret < 0)
2317                         return ret;
2318         }
2319
2320         if (!ctrl->identified) {
2321                 int i;
2322
2323                 ret = nvme_init_subsystem(ctrl, id);
2324                 if (ret)
2325                         goto out_free;
2326
2327                 /*
2328                  * Check for quirks.  Quirk can depend on firmware version,
2329                  * so, in principle, the set of quirks present can change
2330                  * across a reset.  As a possible future enhancement, we
2331                  * could re-scan for quirks every time we reinitialize
2332                  * the device, but we'd have to make sure that the driver
2333                  * behaves intelligently if the quirks change.
2334                  */
2335                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2336                         if (quirk_matches(id, &core_quirks[i]))
2337                                 ctrl->quirks |= core_quirks[i].quirks;
2338                 }
2339         }
2340
2341         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2342                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2343                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2344         }
2345
2346         ctrl->oacs = le16_to_cpu(id->oacs);
2347         ctrl->oncs = le16_to_cpup(&id->oncs);
2348         atomic_set(&ctrl->abort_limit, id->acl + 1);
2349         ctrl->vwc = id->vwc;
2350         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2351         if (id->mdts)
2352                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2353         else
2354                 max_hw_sectors = UINT_MAX;
2355         ctrl->max_hw_sectors =
2356                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2357
2358         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2359         ctrl->sgls = le32_to_cpu(id->sgls);
2360         ctrl->kas = le16_to_cpu(id->kas);
2361
2362         if (id->rtd3e) {
2363                 /* us -> s */
2364                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2365
2366                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2367                                                  shutdown_timeout, 60);
2368
2369                 if (ctrl->shutdown_timeout != shutdown_timeout)
2370                         dev_info(ctrl->device,
2371                                  "Shutdown timeout set to %u seconds\n",
2372                                  ctrl->shutdown_timeout);
2373         } else
2374                 ctrl->shutdown_timeout = shutdown_timeout;
2375
2376         ctrl->npss = id->npss;
2377         ctrl->apsta = id->apsta;
2378         prev_apst_enabled = ctrl->apst_enabled;
2379         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2380                 if (force_apst && id->apsta) {
2381                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2382                         ctrl->apst_enabled = true;
2383                 } else {
2384                         ctrl->apst_enabled = false;
2385                 }
2386         } else {
2387                 ctrl->apst_enabled = id->apsta;
2388         }
2389         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2390
2391         if (ctrl->ops->flags & NVME_F_FABRICS) {
2392                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2393                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2394                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2395                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2396
2397                 /*
2398                  * In fabrics we need to verify the cntlid matches the
2399                  * admin connect
2400                  */
2401                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2402                         ret = -EINVAL;
2403                         goto out_free;
2404                 }
2405
2406                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2407                         dev_err(ctrl->device,
2408                                 "keep-alive support is mandatory for fabrics\n");
2409                         ret = -EINVAL;
2410                         goto out_free;
2411                 }
2412         } else {
2413                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2414                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2415                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2416                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2417                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2418         }
2419
2420         kfree(id);
2421
2422         if (ctrl->apst_enabled && !prev_apst_enabled)
2423                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2424         else if (!ctrl->apst_enabled && prev_apst_enabled)
2425                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2426
2427         ret = nvme_configure_apst(ctrl);
2428         if (ret < 0)
2429                 return ret;
2430         
2431         ret = nvme_configure_timestamp(ctrl);
2432         if (ret < 0)
2433                 return ret;
2434
2435         ret = nvme_configure_directives(ctrl);
2436         if (ret < 0)
2437                 return ret;
2438
2439         ctrl->identified = true;
2440
2441         return 0;
2442
2443 out_free:
2444         kfree(id);
2445         return ret;
2446 }
2447 EXPORT_SYMBOL_GPL(nvme_init_identify);
2448
2449 static int nvme_dev_open(struct inode *inode, struct file *file)
2450 {
2451         struct nvme_ctrl *ctrl =
2452                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2453
2454         switch (ctrl->state) {
2455         case NVME_CTRL_LIVE:
2456         case NVME_CTRL_ADMIN_ONLY:
2457                 break;
2458         default:
2459                 return -EWOULDBLOCK;
2460         }
2461
2462         file->private_data = ctrl;
2463         return 0;
2464 }
2465
2466 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2467 {
2468         struct nvme_ns *ns;
2469         int ret;
2470
2471         down_read(&ctrl->namespaces_rwsem);
2472         if (list_empty(&ctrl->namespaces)) {
2473                 ret = -ENOTTY;
2474                 goto out_unlock;
2475         }
2476
2477         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2478         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2479                 dev_warn(ctrl->device,
2480                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2481                 ret = -EINVAL;
2482                 goto out_unlock;
2483         }
2484
2485         dev_warn(ctrl->device,
2486                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2487         kref_get(&ns->kref);
2488         up_read(&ctrl->namespaces_rwsem);
2489
2490         ret = nvme_user_cmd(ctrl, ns, argp);
2491         nvme_put_ns(ns);
2492         return ret;
2493
2494 out_unlock:
2495         up_read(&ctrl->namespaces_rwsem);
2496         return ret;
2497 }
2498
2499 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2500                 unsigned long arg)
2501 {
2502         struct nvme_ctrl *ctrl = file->private_data;
2503         void __user *argp = (void __user *)arg;
2504
2505         switch (cmd) {
2506         case NVME_IOCTL_ADMIN_CMD:
2507                 return nvme_user_cmd(ctrl, NULL, argp);
2508         case NVME_IOCTL_IO_CMD:
2509                 return nvme_dev_user_cmd(ctrl, argp);
2510         case NVME_IOCTL_RESET:
2511                 dev_warn(ctrl->device, "resetting controller\n");
2512                 return nvme_reset_ctrl_sync(ctrl);
2513         case NVME_IOCTL_SUBSYS_RESET:
2514                 return nvme_reset_subsystem(ctrl);
2515         case NVME_IOCTL_RESCAN:
2516                 nvme_queue_scan(ctrl);
2517                 return 0;
2518         default:
2519                 return -ENOTTY;
2520         }
2521 }
2522
2523 static const struct file_operations nvme_dev_fops = {
2524         .owner          = THIS_MODULE,
2525         .open           = nvme_dev_open,
2526         .unlocked_ioctl = nvme_dev_ioctl,
2527         .compat_ioctl   = nvme_dev_ioctl,
2528 };
2529
2530 static ssize_t nvme_sysfs_reset(struct device *dev,
2531                                 struct device_attribute *attr, const char *buf,
2532                                 size_t count)
2533 {
2534         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2535         int ret;
2536
2537         ret = nvme_reset_ctrl_sync(ctrl);
2538         if (ret < 0)
2539                 return ret;
2540         return count;
2541 }
2542 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2543
2544 static ssize_t nvme_sysfs_rescan(struct device *dev,
2545                                 struct device_attribute *attr, const char *buf,
2546                                 size_t count)
2547 {
2548         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2549
2550         nvme_queue_scan(ctrl);
2551         return count;
2552 }
2553 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2554
2555 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2556 {
2557         struct gendisk *disk = dev_to_disk(dev);
2558
2559         if (disk->fops == &nvme_fops)
2560                 return nvme_get_ns_from_dev(dev)->head;
2561         else
2562                 return disk->private_data;
2563 }
2564
2565 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2566                 char *buf)
2567 {
2568         struct nvme_ns_head *head = dev_to_ns_head(dev);
2569         struct nvme_ns_ids *ids = &head->ids;
2570         struct nvme_subsystem *subsys = head->subsys;
2571         int serial_len = sizeof(subsys->serial);
2572         int model_len = sizeof(subsys->model);
2573
2574         if (!uuid_is_null(&ids->uuid))
2575                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2576
2577         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2578                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2579
2580         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2581                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2582
2583         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2584                                   subsys->serial[serial_len - 1] == '\0'))
2585                 serial_len--;
2586         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2587                                  subsys->model[model_len - 1] == '\0'))
2588                 model_len--;
2589
2590         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2591                 serial_len, subsys->serial, model_len, subsys->model,
2592                 head->ns_id);
2593 }
2594 static DEVICE_ATTR_RO(wwid);
2595
2596 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2597                 char *buf)
2598 {
2599         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2600 }
2601 static DEVICE_ATTR_RO(nguid);
2602
2603 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2604                 char *buf)
2605 {
2606         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2607
2608         /* For backward compatibility expose the NGUID to userspace if
2609          * we have no UUID set
2610          */
2611         if (uuid_is_null(&ids->uuid)) {
2612                 printk_ratelimited(KERN_WARNING
2613                                    "No UUID available providing old NGUID\n");
2614                 return sprintf(buf, "%pU\n", ids->nguid);
2615         }
2616         return sprintf(buf, "%pU\n", &ids->uuid);
2617 }
2618 static DEVICE_ATTR_RO(uuid);
2619
2620 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2621                 char *buf)
2622 {
2623         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2624 }
2625 static DEVICE_ATTR_RO(eui);
2626
2627 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2628                 char *buf)
2629 {
2630         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2631 }
2632 static DEVICE_ATTR_RO(nsid);
2633
2634 static struct attribute *nvme_ns_id_attrs[] = {
2635         &dev_attr_wwid.attr,
2636         &dev_attr_uuid.attr,
2637         &dev_attr_nguid.attr,
2638         &dev_attr_eui.attr,
2639         &dev_attr_nsid.attr,
2640         NULL,
2641 };
2642
2643 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2644                 struct attribute *a, int n)
2645 {
2646         struct device *dev = container_of(kobj, struct device, kobj);
2647         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2648
2649         if (a == &dev_attr_uuid.attr) {
2650                 if (uuid_is_null(&ids->uuid) &&
2651                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2652                         return 0;
2653         }
2654         if (a == &dev_attr_nguid.attr) {
2655                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2656                         return 0;
2657         }
2658         if (a == &dev_attr_eui.attr) {
2659                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2660                         return 0;
2661         }
2662         return a->mode;
2663 }
2664
2665 const struct attribute_group nvme_ns_id_attr_group = {
2666         .attrs          = nvme_ns_id_attrs,
2667         .is_visible     = nvme_ns_id_attrs_are_visible,
2668 };
2669
2670 #define nvme_show_str_function(field)                                           \
2671 static ssize_t  field##_show(struct device *dev,                                \
2672                             struct device_attribute *attr, char *buf)           \
2673 {                                                                               \
2674         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2675         return sprintf(buf, "%.*s\n",                                           \
2676                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2677 }                                                                               \
2678 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2679
2680 nvme_show_str_function(model);
2681 nvme_show_str_function(serial);
2682 nvme_show_str_function(firmware_rev);
2683
2684 #define nvme_show_int_function(field)                                           \
2685 static ssize_t  field##_show(struct device *dev,                                \
2686                             struct device_attribute *attr, char *buf)           \
2687 {                                                                               \
2688         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2689         return sprintf(buf, "%d\n", ctrl->field);       \
2690 }                                                                               \
2691 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2692
2693 nvme_show_int_function(cntlid);
2694
2695 static ssize_t nvme_sysfs_delete(struct device *dev,
2696                                 struct device_attribute *attr, const char *buf,
2697                                 size_t count)
2698 {
2699         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2700
2701         if (device_remove_file_self(dev, attr))
2702                 nvme_delete_ctrl_sync(ctrl);
2703         return count;
2704 }
2705 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2706
2707 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2708                                          struct device_attribute *attr,
2709                                          char *buf)
2710 {
2711         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2712
2713         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2714 }
2715 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2716
2717 static ssize_t nvme_sysfs_show_state(struct device *dev,
2718                                      struct device_attribute *attr,
2719                                      char *buf)
2720 {
2721         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2722         static const char *const state_name[] = {
2723                 [NVME_CTRL_NEW]         = "new",
2724                 [NVME_CTRL_LIVE]        = "live",
2725                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2726                 [NVME_CTRL_RESETTING]   = "resetting",
2727                 [NVME_CTRL_CONNECTING]  = "connecting",
2728                 [NVME_CTRL_DELETING]    = "deleting",
2729                 [NVME_CTRL_DEAD]        = "dead",
2730         };
2731
2732         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2733             state_name[ctrl->state])
2734                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2735
2736         return sprintf(buf, "unknown state\n");
2737 }
2738
2739 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2740
2741 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2742                                          struct device_attribute *attr,
2743                                          char *buf)
2744 {
2745         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2746
2747         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2748 }
2749 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2750
2751 static ssize_t nvme_sysfs_show_address(struct device *dev,
2752                                          struct device_attribute *attr,
2753                                          char *buf)
2754 {
2755         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2756
2757         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2758 }
2759 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2760
2761 static struct attribute *nvme_dev_attrs[] = {
2762         &dev_attr_reset_controller.attr,
2763         &dev_attr_rescan_controller.attr,
2764         &dev_attr_model.attr,
2765         &dev_attr_serial.attr,
2766         &dev_attr_firmware_rev.attr,
2767         &dev_attr_cntlid.attr,
2768         &dev_attr_delete_controller.attr,
2769         &dev_attr_transport.attr,
2770         &dev_attr_subsysnqn.attr,
2771         &dev_attr_address.attr,
2772         &dev_attr_state.attr,
2773         NULL
2774 };
2775
2776 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2777                 struct attribute *a, int n)
2778 {
2779         struct device *dev = container_of(kobj, struct device, kobj);
2780         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2781
2782         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2783                 return 0;
2784         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2785                 return 0;
2786
2787         return a->mode;
2788 }
2789
2790 static struct attribute_group nvme_dev_attrs_group = {
2791         .attrs          = nvme_dev_attrs,
2792         .is_visible     = nvme_dev_attrs_are_visible,
2793 };
2794
2795 static const struct attribute_group *nvme_dev_attr_groups[] = {
2796         &nvme_dev_attrs_group,
2797         NULL,
2798 };
2799
2800 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2801                 unsigned nsid)
2802 {
2803         struct nvme_ns_head *h;
2804
2805         lockdep_assert_held(&subsys->lock);
2806
2807         list_for_each_entry(h, &subsys->nsheads, entry) {
2808                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2809                         return h;
2810         }
2811
2812         return NULL;
2813 }
2814
2815 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2816                 struct nvme_ns_head *new)
2817 {
2818         struct nvme_ns_head *h;
2819
2820         lockdep_assert_held(&subsys->lock);
2821
2822         list_for_each_entry(h, &subsys->nsheads, entry) {
2823                 if (nvme_ns_ids_valid(&new->ids) &&
2824                     !list_empty(&h->list) &&
2825                     nvme_ns_ids_equal(&new->ids, &h->ids))
2826                         return -EINVAL;
2827         }
2828
2829         return 0;
2830 }
2831
2832 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2833                 unsigned nsid, struct nvme_id_ns *id)
2834 {
2835         struct nvme_ns_head *head;
2836         int ret = -ENOMEM;
2837
2838         head = kzalloc(sizeof(*head), GFP_KERNEL);
2839         if (!head)
2840                 goto out;
2841         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2842         if (ret < 0)
2843                 goto out_free_head;
2844         head->instance = ret;
2845         INIT_LIST_HEAD(&head->list);
2846         ret = init_srcu_struct(&head->srcu);
2847         if (ret)
2848                 goto out_ida_remove;
2849         head->subsys = ctrl->subsys;
2850         head->ns_id = nsid;
2851         kref_init(&head->ref);
2852
2853         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2854
2855         ret = __nvme_check_ids(ctrl->subsys, head);
2856         if (ret) {
2857                 dev_err(ctrl->device,
2858                         "duplicate IDs for nsid %d\n", nsid);
2859                 goto out_cleanup_srcu;
2860         }
2861
2862         ret = nvme_mpath_alloc_disk(ctrl, head);
2863         if (ret)
2864                 goto out_cleanup_srcu;
2865
2866         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2867
2868         kref_get(&ctrl->subsys->ref);
2869
2870         return head;
2871 out_cleanup_srcu:
2872         cleanup_srcu_struct(&head->srcu);
2873 out_ida_remove:
2874         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2875 out_free_head:
2876         kfree(head);
2877 out:
2878         return ERR_PTR(ret);
2879 }
2880
2881 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2882                 struct nvme_id_ns *id)
2883 {
2884         struct nvme_ctrl *ctrl = ns->ctrl;
2885         bool is_shared = id->nmic & (1 << 0);
2886         struct nvme_ns_head *head = NULL;
2887         int ret = 0;
2888
2889         mutex_lock(&ctrl->subsys->lock);
2890         if (is_shared)
2891                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2892         if (!head) {
2893                 head = nvme_alloc_ns_head(ctrl, nsid, id);
2894                 if (IS_ERR(head)) {
2895                         ret = PTR_ERR(head);
2896                         goto out_unlock;
2897                 }
2898         } else {
2899                 struct nvme_ns_ids ids;
2900
2901                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2902                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2903                         dev_err(ctrl->device,
2904                                 "IDs don't match for shared namespace %d\n",
2905                                         nsid);
2906                         ret = -EINVAL;
2907                         goto out_unlock;
2908                 }
2909         }
2910
2911         list_add_tail(&ns->siblings, &head->list);
2912         ns->head = head;
2913
2914 out_unlock:
2915         mutex_unlock(&ctrl->subsys->lock);
2916         return ret;
2917 }
2918
2919 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2920 {
2921         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2922         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2923
2924         return nsa->head->ns_id - nsb->head->ns_id;
2925 }
2926
2927 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2928 {
2929         struct nvme_ns *ns, *ret = NULL;
2930
2931         down_read(&ctrl->namespaces_rwsem);
2932         list_for_each_entry(ns, &ctrl->namespaces, list) {
2933                 if (ns->head->ns_id == nsid) {
2934                         if (!kref_get_unless_zero(&ns->kref))
2935                                 continue;
2936                         ret = ns;
2937                         break;
2938                 }
2939                 if (ns->head->ns_id > nsid)
2940                         break;
2941         }
2942         up_read(&ctrl->namespaces_rwsem);
2943         return ret;
2944 }
2945
2946 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2947 {
2948         struct streams_directive_params s;
2949         int ret;
2950
2951         if (!ctrl->nr_streams)
2952                 return 0;
2953
2954         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2955         if (ret)
2956                 return ret;
2957
2958         ns->sws = le32_to_cpu(s.sws);
2959         ns->sgs = le16_to_cpu(s.sgs);
2960
2961         if (ns->sws) {
2962                 unsigned int bs = 1 << ns->lba_shift;
2963
2964                 blk_queue_io_min(ns->queue, bs * ns->sws);
2965                 if (ns->sgs)
2966                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2967         }
2968
2969         return 0;
2970 }
2971
2972 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2973 {
2974         struct nvme_ns *ns;
2975         struct gendisk *disk;
2976         struct nvme_id_ns *id;
2977         char disk_name[DISK_NAME_LEN];
2978         int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2979
2980         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2981         if (!ns)
2982                 return;
2983
2984         ns->queue = blk_mq_init_queue(ctrl->tagset);
2985         if (IS_ERR(ns->queue))
2986                 goto out_free_ns;
2987         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
2988         ns->queue->queuedata = ns;
2989         ns->ctrl = ctrl;
2990
2991         kref_init(&ns->kref);
2992         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2993
2994         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2995         nvme_set_queue_limits(ctrl, ns->queue);
2996
2997         id = nvme_identify_ns(ctrl, nsid);
2998         if (!id)
2999                 goto out_free_queue;
3000
3001         if (id->ncap == 0)
3002                 goto out_free_id;
3003
3004         if (nvme_init_ns_head(ns, nsid, id))
3005                 goto out_free_id;
3006         nvme_setup_streams_ns(ctrl, ns);
3007         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3008
3009         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3010                 if (nvme_nvm_register(ns, disk_name, node)) {
3011                         dev_warn(ctrl->device, "LightNVM init failure\n");
3012                         goto out_unlink_ns;
3013                 }
3014         }
3015
3016         disk = alloc_disk_node(0, node);
3017         if (!disk)
3018                 goto out_unlink_ns;
3019
3020         disk->fops = &nvme_fops;
3021         disk->private_data = ns;
3022         disk->queue = ns->queue;
3023         disk->flags = flags;
3024         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3025         ns->disk = disk;
3026
3027         __nvme_revalidate_disk(disk, id);
3028
3029         down_write(&ctrl->namespaces_rwsem);
3030         list_add_tail(&ns->list, &ctrl->namespaces);
3031         up_write(&ctrl->namespaces_rwsem);
3032
3033         nvme_get_ctrl(ctrl);
3034
3035         kfree(id);
3036
3037         device_add_disk(ctrl->device, ns->disk);
3038         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3039                                         &nvme_ns_id_attr_group))
3040                 pr_warn("%s: failed to create sysfs group for identification\n",
3041                         ns->disk->disk_name);
3042         if (ns->ndev && nvme_nvm_register_sysfs(ns))
3043                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3044                         ns->disk->disk_name);
3045
3046         nvme_mpath_add_disk(ns->head);
3047         nvme_fault_inject_init(ns);
3048         return;
3049  out_unlink_ns:
3050         mutex_lock(&ctrl->subsys->lock);
3051         list_del_rcu(&ns->siblings);
3052         mutex_unlock(&ctrl->subsys->lock);
3053  out_free_id:
3054         kfree(id);
3055  out_free_queue:
3056         blk_cleanup_queue(ns->queue);
3057  out_free_ns:
3058         kfree(ns);
3059 }
3060
3061 static void nvme_ns_remove(struct nvme_ns *ns)
3062 {
3063         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3064                 return;
3065
3066         nvme_fault_inject_fini(ns);
3067         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3068                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3069                                         &nvme_ns_id_attr_group);
3070                 if (ns->ndev)
3071                         nvme_nvm_unregister_sysfs(ns);
3072                 del_gendisk(ns->disk);
3073                 blk_cleanup_queue(ns->queue);
3074                 if (blk_get_integrity(ns->disk))
3075                         blk_integrity_unregister(ns->disk);
3076         }
3077
3078         mutex_lock(&ns->ctrl->subsys->lock);
3079         nvme_mpath_clear_current_path(ns);
3080         list_del_rcu(&ns->siblings);
3081         mutex_unlock(&ns->ctrl->subsys->lock);
3082
3083         down_write(&ns->ctrl->namespaces_rwsem);
3084         list_del_init(&ns->list);
3085         up_write(&ns->ctrl->namespaces_rwsem);
3086
3087         synchronize_srcu(&ns->head->srcu);
3088         nvme_mpath_check_last_path(ns);
3089         nvme_put_ns(ns);
3090 }
3091
3092 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3093 {
3094         struct nvme_ns *ns;
3095
3096         ns = nvme_find_get_ns(ctrl, nsid);
3097         if (ns) {
3098                 if (ns->disk && revalidate_disk(ns->disk))
3099                         nvme_ns_remove(ns);
3100                 nvme_put_ns(ns);
3101         } else
3102                 nvme_alloc_ns(ctrl, nsid);
3103 }
3104
3105 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3106                                         unsigned nsid)
3107 {
3108         struct nvme_ns *ns, *next;
3109         LIST_HEAD(rm_list);
3110
3111         down_write(&ctrl->namespaces_rwsem);
3112         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3113                 if (ns->head->ns_id > nsid)
3114                         list_move_tail(&ns->list, &rm_list);
3115         }
3116         up_write(&ctrl->namespaces_rwsem);
3117
3118         list_for_each_entry_safe(ns, next, &rm_list, list)
3119                 nvme_ns_remove(ns);
3120
3121 }
3122
3123 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3124 {
3125         struct nvme_ns *ns;
3126         __le32 *ns_list;
3127         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3128         int ret = 0;
3129
3130         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3131         if (!ns_list)
3132                 return -ENOMEM;
3133
3134         for (i = 0; i < num_lists; i++) {
3135                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3136                 if (ret)
3137                         goto free;
3138
3139                 for (j = 0; j < min(nn, 1024U); j++) {
3140                         nsid = le32_to_cpu(ns_list[j]);
3141                         if (!nsid)
3142                                 goto out;
3143
3144                         nvme_validate_ns(ctrl, nsid);
3145
3146                         while (++prev < nsid) {
3147                                 ns = nvme_find_get_ns(ctrl, prev);
3148                                 if (ns) {
3149                                         nvme_ns_remove(ns);
3150                                         nvme_put_ns(ns);
3151                                 }
3152                         }
3153                 }
3154                 nn -= j;
3155         }
3156  out:
3157         nvme_remove_invalid_namespaces(ctrl, prev);
3158  free:
3159         kfree(ns_list);
3160         return ret;
3161 }
3162
3163 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3164 {
3165         unsigned i;
3166
3167         for (i = 1; i <= nn; i++)
3168                 nvme_validate_ns(ctrl, i);
3169
3170         nvme_remove_invalid_namespaces(ctrl, nn);
3171 }
3172
3173 static void nvme_scan_work(struct work_struct *work)
3174 {
3175         struct nvme_ctrl *ctrl =
3176                 container_of(work, struct nvme_ctrl, scan_work);
3177         struct nvme_id_ctrl *id;
3178         unsigned nn;
3179
3180         if (ctrl->state != NVME_CTRL_LIVE)
3181                 return;
3182
3183         WARN_ON_ONCE(!ctrl->tagset);
3184
3185         if (nvme_identify_ctrl(ctrl, &id))
3186                 return;
3187
3188         nn = le32_to_cpu(id->nn);
3189         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3190             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3191                 if (!nvme_scan_ns_list(ctrl, nn))
3192                         goto done;
3193         }
3194         nvme_scan_ns_sequential(ctrl, nn);
3195  done:
3196         down_write(&ctrl->namespaces_rwsem);
3197         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3198         up_write(&ctrl->namespaces_rwsem);
3199         kfree(id);
3200 }
3201
3202 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3203 {
3204         /*
3205          * Only new queue scan work when admin and IO queues are both alive
3206          */
3207         if (ctrl->state == NVME_CTRL_LIVE)
3208                 queue_work(nvme_wq, &ctrl->scan_work);
3209 }
3210 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3211
3212 /*
3213  * This function iterates the namespace list unlocked to allow recovery from
3214  * controller failure. It is up to the caller to ensure the namespace list is
3215  * not modified by scan work while this function is executing.
3216  */
3217 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3218 {
3219         struct nvme_ns *ns, *next;
3220         LIST_HEAD(ns_list);
3221
3222         /*
3223          * The dead states indicates the controller was not gracefully
3224          * disconnected. In that case, we won't be able to flush any data while
3225          * removing the namespaces' disks; fail all the queues now to avoid
3226          * potentially having to clean up the failed sync later.
3227          */
3228         if (ctrl->state == NVME_CTRL_DEAD)
3229                 nvme_kill_queues(ctrl);
3230
3231         down_write(&ctrl->namespaces_rwsem);
3232         list_splice_init(&ctrl->namespaces, &ns_list);
3233         up_write(&ctrl->namespaces_rwsem);
3234
3235         list_for_each_entry_safe(ns, next, &ns_list, list)
3236                 nvme_ns_remove(ns);
3237 }
3238 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3239
3240 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3241 {
3242         char *envp[2] = { NULL, NULL };
3243         u32 aen_result = ctrl->aen_result;
3244
3245         ctrl->aen_result = 0;
3246         if (!aen_result)
3247                 return;
3248
3249         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3250         if (!envp[0])
3251                 return;
3252         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3253         kfree(envp[0]);
3254 }
3255
3256 static void nvme_async_event_work(struct work_struct *work)
3257 {
3258         struct nvme_ctrl *ctrl =
3259                 container_of(work, struct nvme_ctrl, async_event_work);
3260
3261         nvme_aen_uevent(ctrl);
3262         ctrl->ops->submit_async_event(ctrl);
3263 }
3264
3265 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3266 {
3267
3268         u32 csts;
3269
3270         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3271                 return false;
3272
3273         if (csts == ~0)
3274                 return false;
3275
3276         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3277 }
3278
3279 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3280 {
3281         struct nvme_fw_slot_info_log *log;
3282
3283         log = kmalloc(sizeof(*log), GFP_KERNEL);
3284         if (!log)
3285                 return;
3286
3287         if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3288                 dev_warn(ctrl->device,
3289                                 "Get FW SLOT INFO log error\n");
3290         kfree(log);
3291 }
3292
3293 static void nvme_fw_act_work(struct work_struct *work)
3294 {
3295         struct nvme_ctrl *ctrl = container_of(work,
3296                                 struct nvme_ctrl, fw_act_work);
3297         unsigned long fw_act_timeout;
3298
3299         if (ctrl->mtfa)
3300                 fw_act_timeout = jiffies +
3301                                 msecs_to_jiffies(ctrl->mtfa * 100);
3302         else
3303                 fw_act_timeout = jiffies +
3304                                 msecs_to_jiffies(admin_timeout * 1000);
3305
3306         nvme_stop_queues(ctrl);
3307         while (nvme_ctrl_pp_status(ctrl)) {
3308                 if (time_after(jiffies, fw_act_timeout)) {
3309                         dev_warn(ctrl->device,
3310                                 "Fw activation timeout, reset controller\n");
3311                         nvme_reset_ctrl(ctrl);
3312                         break;
3313                 }
3314                 msleep(100);
3315         }
3316
3317         if (ctrl->state != NVME_CTRL_LIVE)
3318                 return;
3319
3320         nvme_start_queues(ctrl);
3321         /* read FW slot information to clear the AER */
3322         nvme_get_fw_slot_info(ctrl);
3323 }
3324
3325 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3326                 union nvme_result *res)
3327 {
3328         u32 result = le32_to_cpu(res->u32);
3329
3330         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3331                 return;
3332
3333         switch (result & 0x7) {
3334         case NVME_AER_ERROR:
3335         case NVME_AER_SMART:
3336         case NVME_AER_CSS:
3337         case NVME_AER_VS:
3338                 ctrl->aen_result = result;
3339                 break;
3340         default:
3341                 break;
3342         }
3343
3344         switch (result & 0xff07) {
3345         case NVME_AER_NOTICE_NS_CHANGED:
3346                 dev_info(ctrl->device, "rescanning\n");
3347                 nvme_queue_scan(ctrl);
3348                 break;
3349         case NVME_AER_NOTICE_FW_ACT_STARTING:
3350                 queue_work(nvme_wq, &ctrl->fw_act_work);
3351                 break;
3352         default:
3353                 dev_warn(ctrl->device, "async event result %08x\n", result);
3354         }
3355         queue_work(nvme_wq, &ctrl->async_event_work);
3356 }
3357 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3358
3359 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3360 {
3361         nvme_stop_keep_alive(ctrl);
3362         flush_work(&ctrl->async_event_work);
3363         flush_work(&ctrl->scan_work);
3364         cancel_work_sync(&ctrl->fw_act_work);
3365         if (ctrl->ops->stop_ctrl)
3366                 ctrl->ops->stop_ctrl(ctrl);
3367 }
3368 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3369
3370 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3371 {
3372         if (ctrl->kato)
3373                 nvme_start_keep_alive(ctrl);
3374
3375         if (ctrl->queue_count > 1) {
3376                 nvme_queue_scan(ctrl);
3377                 queue_work(nvme_wq, &ctrl->async_event_work);
3378                 nvme_start_queues(ctrl);
3379         }
3380 }
3381 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3382
3383 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3384 {
3385         cdev_device_del(&ctrl->cdev, ctrl->device);
3386 }
3387 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3388
3389 static void nvme_free_ctrl(struct device *dev)
3390 {
3391         struct nvme_ctrl *ctrl =
3392                 container_of(dev, struct nvme_ctrl, ctrl_device);
3393         struct nvme_subsystem *subsys = ctrl->subsys;
3394
3395         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3396         kfree(ctrl->effects);
3397
3398         if (subsys) {
3399                 mutex_lock(&subsys->lock);
3400                 list_del(&ctrl->subsys_entry);
3401                 mutex_unlock(&subsys->lock);
3402                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3403         }
3404
3405         ctrl->ops->free_ctrl(ctrl);
3406
3407         if (subsys)
3408                 nvme_put_subsystem(subsys);
3409 }
3410
3411 /*
3412  * Initialize a NVMe controller structures.  This needs to be called during
3413  * earliest initialization so that we have the initialized structured around
3414  * during probing.
3415  */
3416 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3417                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3418 {
3419         int ret;
3420
3421         ctrl->state = NVME_CTRL_NEW;
3422         spin_lock_init(&ctrl->lock);
3423         INIT_LIST_HEAD(&ctrl->namespaces);
3424         init_rwsem(&ctrl->namespaces_rwsem);
3425         ctrl->dev = dev;
3426         ctrl->ops = ops;
3427         ctrl->quirks = quirks;
3428         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3429         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3430         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3431         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3432
3433         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3434         if (ret < 0)
3435                 goto out;
3436         ctrl->instance = ret;
3437
3438         device_initialize(&ctrl->ctrl_device);
3439         ctrl->device = &ctrl->ctrl_device;
3440         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3441         ctrl->device->class = nvme_class;
3442         ctrl->device->parent = ctrl->dev;
3443         ctrl->device->groups = nvme_dev_attr_groups;
3444         ctrl->device->release = nvme_free_ctrl;
3445         dev_set_drvdata(ctrl->device, ctrl);
3446         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3447         if (ret)
3448                 goto out_release_instance;
3449
3450         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3451         ctrl->cdev.owner = ops->module;
3452         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3453         if (ret)
3454                 goto out_free_name;
3455
3456         /*
3457          * Initialize latency tolerance controls.  The sysfs files won't
3458          * be visible to userspace unless the device actually supports APST.
3459          */
3460         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3461         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3462                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3463
3464         return 0;
3465 out_free_name:
3466         kfree_const(dev->kobj.name);
3467 out_release_instance:
3468         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3469 out:
3470         return ret;
3471 }
3472 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3473
3474 /**
3475  * nvme_kill_queues(): Ends all namespace queues
3476  * @ctrl: the dead controller that needs to end