Merge tag 'for-4.20/block-20181021' of git://git.kernel.dk/linux-block
[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 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
104                                            unsigned nsid);
105
106 static void nvme_set_queue_dying(struct nvme_ns *ns)
107 {
108         /*
109          * Revalidating a dead namespace sets capacity to 0. This will end
110          * buffered writers dirtying pages that can't be synced.
111          */
112         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
113                 return;
114         revalidate_disk(ns->disk);
115         blk_set_queue_dying(ns->queue);
116         /* Forcibly unquiesce queues to avoid blocking dispatch */
117         blk_mq_unquiesce_queue(ns->queue);
118 }
119
120 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
121 {
122         /*
123          * Only new queue scan work when admin and IO queues are both alive
124          */
125         if (ctrl->state == NVME_CTRL_LIVE)
126                 queue_work(nvme_wq, &ctrl->scan_work);
127 }
128
129 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
130 {
131         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
132                 return -EBUSY;
133         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
134                 return -EBUSY;
135         return 0;
136 }
137 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
138
139 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
140 {
141         int ret;
142
143         ret = nvme_reset_ctrl(ctrl);
144         if (!ret) {
145                 flush_work(&ctrl->reset_work);
146                 if (ctrl->state != NVME_CTRL_LIVE &&
147                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
148                         ret = -ENETRESET;
149         }
150
151         return ret;
152 }
153 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
154
155 static void nvme_delete_ctrl_work(struct work_struct *work)
156 {
157         struct nvme_ctrl *ctrl =
158                 container_of(work, struct nvme_ctrl, delete_work);
159
160         dev_info(ctrl->device,
161                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
162
163         flush_work(&ctrl->reset_work);
164         nvme_stop_ctrl(ctrl);
165         nvme_remove_namespaces(ctrl);
166         ctrl->ops->delete_ctrl(ctrl);
167         nvme_uninit_ctrl(ctrl);
168         nvme_put_ctrl(ctrl);
169 }
170
171 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
172 {
173         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
174                 return -EBUSY;
175         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
176                 return -EBUSY;
177         return 0;
178 }
179 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
180
181 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
182 {
183         int ret = 0;
184
185         /*
186          * Keep a reference until the work is flushed since ->delete_ctrl
187          * can free the controller.
188          */
189         nvme_get_ctrl(ctrl);
190         ret = nvme_delete_ctrl(ctrl);
191         if (!ret)
192                 flush_work(&ctrl->delete_work);
193         nvme_put_ctrl(ctrl);
194         return ret;
195 }
196 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
197
198 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
199 {
200         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
201 }
202
203 static blk_status_t nvme_error_status(struct request *req)
204 {
205         switch (nvme_req(req)->status & 0x7ff) {
206         case NVME_SC_SUCCESS:
207                 return BLK_STS_OK;
208         case NVME_SC_CAP_EXCEEDED:
209                 return BLK_STS_NOSPC;
210         case NVME_SC_LBA_RANGE:
211                 return BLK_STS_TARGET;
212         case NVME_SC_BAD_ATTRIBUTES:
213         case NVME_SC_ONCS_NOT_SUPPORTED:
214         case NVME_SC_INVALID_OPCODE:
215         case NVME_SC_INVALID_FIELD:
216         case NVME_SC_INVALID_NS:
217                 return BLK_STS_NOTSUPP;
218         case NVME_SC_WRITE_FAULT:
219         case NVME_SC_READ_ERROR:
220         case NVME_SC_UNWRITTEN_BLOCK:
221         case NVME_SC_ACCESS_DENIED:
222         case NVME_SC_READ_ONLY:
223         case NVME_SC_COMPARE_FAILED:
224                 return BLK_STS_MEDIUM;
225         case NVME_SC_GUARD_CHECK:
226         case NVME_SC_APPTAG_CHECK:
227         case NVME_SC_REFTAG_CHECK:
228         case NVME_SC_INVALID_PI:
229                 return BLK_STS_PROTECTION;
230         case NVME_SC_RESERVATION_CONFLICT:
231                 return BLK_STS_NEXUS;
232         default:
233                 return BLK_STS_IOERR;
234         }
235 }
236
237 static inline bool nvme_req_needs_retry(struct request *req)
238 {
239         if (blk_noretry_request(req))
240                 return false;
241         if (nvme_req(req)->status & NVME_SC_DNR)
242                 return false;
243         if (nvme_req(req)->retries >= nvme_max_retries)
244                 return false;
245         return true;
246 }
247
248 void nvme_complete_rq(struct request *req)
249 {
250         blk_status_t status = nvme_error_status(req);
251
252         trace_nvme_complete_rq(req);
253
254         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
255                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
256                     blk_path_error(status)) {
257                         nvme_failover_req(req);
258                         return;
259                 }
260
261                 if (!blk_queue_dying(req->q)) {
262                         nvme_req(req)->retries++;
263                         blk_mq_requeue_request(req, true);
264                         return;
265                 }
266         }
267         blk_mq_end_request(req, status);
268 }
269 EXPORT_SYMBOL_GPL(nvme_complete_rq);
270
271 void nvme_cancel_request(struct request *req, void *data, bool reserved)
272 {
273         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
274                                 "Cancelling I/O %d", req->tag);
275
276         nvme_req(req)->status = NVME_SC_ABORT_REQ;
277         blk_mq_complete_request(req);
278
279 }
280 EXPORT_SYMBOL_GPL(nvme_cancel_request);
281
282 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
283                 enum nvme_ctrl_state new_state)
284 {
285         enum nvme_ctrl_state old_state;
286         unsigned long flags;
287         bool changed = false;
288
289         spin_lock_irqsave(&ctrl->lock, flags);
290
291         old_state = ctrl->state;
292         switch (new_state) {
293         case NVME_CTRL_ADMIN_ONLY:
294                 switch (old_state) {
295                 case NVME_CTRL_CONNECTING:
296                         changed = true;
297                         /* FALLTHRU */
298                 default:
299                         break;
300                 }
301                 break;
302         case NVME_CTRL_LIVE:
303                 switch (old_state) {
304                 case NVME_CTRL_NEW:
305                 case NVME_CTRL_RESETTING:
306                 case NVME_CTRL_CONNECTING:
307                         changed = true;
308                         /* FALLTHRU */
309                 default:
310                         break;
311                 }
312                 break;
313         case NVME_CTRL_RESETTING:
314                 switch (old_state) {
315                 case NVME_CTRL_NEW:
316                 case NVME_CTRL_LIVE:
317                 case NVME_CTRL_ADMIN_ONLY:
318                         changed = true;
319                         /* FALLTHRU */
320                 default:
321                         break;
322                 }
323                 break;
324         case NVME_CTRL_CONNECTING:
325                 switch (old_state) {
326                 case NVME_CTRL_NEW:
327                 case NVME_CTRL_RESETTING:
328                         changed = true;
329                         /* FALLTHRU */
330                 default:
331                         break;
332                 }
333                 break;
334         case NVME_CTRL_DELETING:
335                 switch (old_state) {
336                 case NVME_CTRL_LIVE:
337                 case NVME_CTRL_ADMIN_ONLY:
338                 case NVME_CTRL_RESETTING:
339                 case NVME_CTRL_CONNECTING:
340                         changed = true;
341                         /* FALLTHRU */
342                 default:
343                         break;
344                 }
345                 break;
346         case NVME_CTRL_DEAD:
347                 switch (old_state) {
348                 case NVME_CTRL_DELETING:
349                         changed = true;
350                         /* FALLTHRU */
351                 default:
352                         break;
353                 }
354                 break;
355         default:
356                 break;
357         }
358
359         if (changed)
360                 ctrl->state = new_state;
361
362         spin_unlock_irqrestore(&ctrl->lock, flags);
363         if (changed && ctrl->state == NVME_CTRL_LIVE)
364                 nvme_kick_requeue_lists(ctrl);
365         return changed;
366 }
367 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
368
369 static void nvme_free_ns_head(struct kref *ref)
370 {
371         struct nvme_ns_head *head =
372                 container_of(ref, struct nvme_ns_head, ref);
373
374         nvme_mpath_remove_disk(head);
375         ida_simple_remove(&head->subsys->ns_ida, head->instance);
376         list_del_init(&head->entry);
377         cleanup_srcu_struct_quiesced(&head->srcu);
378         nvme_put_subsystem(head->subsys);
379         kfree(head);
380 }
381
382 static void nvme_put_ns_head(struct nvme_ns_head *head)
383 {
384         kref_put(&head->ref, nvme_free_ns_head);
385 }
386
387 static void nvme_free_ns(struct kref *kref)
388 {
389         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
390
391         if (ns->ndev)
392                 nvme_nvm_unregister(ns);
393
394         put_disk(ns->disk);
395         nvme_put_ns_head(ns->head);
396         nvme_put_ctrl(ns->ctrl);
397         kfree(ns);
398 }
399
400 static void nvme_put_ns(struct nvme_ns *ns)
401 {
402         kref_put(&ns->kref, nvme_free_ns);
403 }
404
405 static inline void nvme_clear_nvme_request(struct request *req)
406 {
407         if (!(req->rq_flags & RQF_DONTPREP)) {
408                 nvme_req(req)->retries = 0;
409                 nvme_req(req)->flags = 0;
410                 req->rq_flags |= RQF_DONTPREP;
411         }
412 }
413
414 struct request *nvme_alloc_request(struct request_queue *q,
415                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
416 {
417         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
418         struct request *req;
419
420         if (qid == NVME_QID_ANY) {
421                 req = blk_mq_alloc_request(q, op, flags);
422         } else {
423                 req = blk_mq_alloc_request_hctx(q, op, flags,
424                                 qid ? qid - 1 : 0);
425         }
426         if (IS_ERR(req))
427                 return req;
428
429         req->cmd_flags |= REQ_FAILFAST_DRIVER;
430         nvme_clear_nvme_request(req);
431         nvme_req(req)->cmd = cmd;
432
433         return req;
434 }
435 EXPORT_SYMBOL_GPL(nvme_alloc_request);
436
437 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
438 {
439         struct nvme_command c;
440
441         memset(&c, 0, sizeof(c));
442
443         c.directive.opcode = nvme_admin_directive_send;
444         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
445         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
446         c.directive.dtype = NVME_DIR_IDENTIFY;
447         c.directive.tdtype = NVME_DIR_STREAMS;
448         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
449
450         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
451 }
452
453 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
454 {
455         return nvme_toggle_streams(ctrl, false);
456 }
457
458 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
459 {
460         return nvme_toggle_streams(ctrl, true);
461 }
462
463 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
464                                   struct streams_directive_params *s, u32 nsid)
465 {
466         struct nvme_command c;
467
468         memset(&c, 0, sizeof(c));
469         memset(s, 0, sizeof(*s));
470
471         c.directive.opcode = nvme_admin_directive_recv;
472         c.directive.nsid = cpu_to_le32(nsid);
473         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
474         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
475         c.directive.dtype = NVME_DIR_STREAMS;
476
477         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
478 }
479
480 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
481 {
482         struct streams_directive_params s;
483         int ret;
484
485         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
486                 return 0;
487         if (!streams)
488                 return 0;
489
490         ret = nvme_enable_streams(ctrl);
491         if (ret)
492                 return ret;
493
494         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
495         if (ret)
496                 return ret;
497
498         ctrl->nssa = le16_to_cpu(s.nssa);
499         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
500                 dev_info(ctrl->device, "too few streams (%u) available\n",
501                                         ctrl->nssa);
502                 nvme_disable_streams(ctrl);
503                 return 0;
504         }
505
506         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
507         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
508         return 0;
509 }
510
511 /*
512  * Check if 'req' has a write hint associated with it. If it does, assign
513  * a valid namespace stream to the write.
514  */
515 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
516                                      struct request *req, u16 *control,
517                                      u32 *dsmgmt)
518 {
519         enum rw_hint streamid = req->write_hint;
520
521         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
522                 streamid = 0;
523         else {
524                 streamid--;
525                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
526                         return;
527
528                 *control |= NVME_RW_DTYPE_STREAMS;
529                 *dsmgmt |= streamid << 16;
530         }
531
532         if (streamid < ARRAY_SIZE(req->q->write_hints))
533                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
534 }
535
536 static inline void nvme_setup_flush(struct nvme_ns *ns,
537                 struct nvme_command *cmnd)
538 {
539         memset(cmnd, 0, sizeof(*cmnd));
540         cmnd->common.opcode = nvme_cmd_flush;
541         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
542 }
543
544 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
545                 struct nvme_command *cmnd)
546 {
547         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
548         struct nvme_dsm_range *range;
549         struct bio *bio;
550
551         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
552         if (!range)
553                 return BLK_STS_RESOURCE;
554
555         __rq_for_each_bio(bio, req) {
556                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
557                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
558
559                 if (n < segments) {
560                         range[n].cattr = cpu_to_le32(0);
561                         range[n].nlb = cpu_to_le32(nlb);
562                         range[n].slba = cpu_to_le64(slba);
563                 }
564                 n++;
565         }
566
567         if (WARN_ON_ONCE(n != segments)) {
568                 kfree(range);
569                 return BLK_STS_IOERR;
570         }
571
572         memset(cmnd, 0, sizeof(*cmnd));
573         cmnd->dsm.opcode = nvme_cmd_dsm;
574         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
575         cmnd->dsm.nr = cpu_to_le32(segments - 1);
576         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
577
578         req->special_vec.bv_page = virt_to_page(range);
579         req->special_vec.bv_offset = offset_in_page(range);
580         req->special_vec.bv_len = sizeof(*range) * segments;
581         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
582
583         return BLK_STS_OK;
584 }
585
586 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
587                 struct request *req, struct nvme_command *cmnd)
588 {
589         struct nvme_ctrl *ctrl = ns->ctrl;
590         u16 control = 0;
591         u32 dsmgmt = 0;
592
593         if (req->cmd_flags & REQ_FUA)
594                 control |= NVME_RW_FUA;
595         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
596                 control |= NVME_RW_LR;
597
598         if (req->cmd_flags & REQ_RAHEAD)
599                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
600
601         memset(cmnd, 0, sizeof(*cmnd));
602         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
603         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
604         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
605         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
606
607         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
608                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
609
610         if (ns->ms) {
611                 /*
612                  * If formated with metadata, the block layer always provides a
613                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
614                  * we enable the PRACT bit for protection information or set the
615                  * namespace capacity to zero to prevent any I/O.
616                  */
617                 if (!blk_integrity_rq(req)) {
618                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
619                                 return BLK_STS_NOTSUPP;
620                         control |= NVME_RW_PRINFO_PRACT;
621                 } else if (req_op(req) == REQ_OP_WRITE) {
622                         t10_pi_prepare(req, ns->pi_type);
623                 }
624
625                 switch (ns->pi_type) {
626                 case NVME_NS_DPS_PI_TYPE3:
627                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
628                         break;
629                 case NVME_NS_DPS_PI_TYPE1:
630                 case NVME_NS_DPS_PI_TYPE2:
631                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
632                                         NVME_RW_PRINFO_PRCHK_REF;
633                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
634                         break;
635                 }
636         }
637
638         cmnd->rw.control = cpu_to_le16(control);
639         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
640         return 0;
641 }
642
643 void nvme_cleanup_cmd(struct request *req)
644 {
645         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
646             nvme_req(req)->status == 0) {
647                 struct nvme_ns *ns = req->rq_disk->private_data;
648
649                 t10_pi_complete(req, ns->pi_type,
650                                 blk_rq_bytes(req) >> ns->lba_shift);
651         }
652         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
653                 kfree(page_address(req->special_vec.bv_page) +
654                       req->special_vec.bv_offset);
655         }
656 }
657 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
658
659 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
660                 struct nvme_command *cmd)
661 {
662         blk_status_t ret = BLK_STS_OK;
663
664         nvme_clear_nvme_request(req);
665
666         switch (req_op(req)) {
667         case REQ_OP_DRV_IN:
668         case REQ_OP_DRV_OUT:
669                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
670                 break;
671         case REQ_OP_FLUSH:
672                 nvme_setup_flush(ns, cmd);
673                 break;
674         case REQ_OP_WRITE_ZEROES:
675                 /* currently only aliased to deallocate for a few ctrls: */
676         case REQ_OP_DISCARD:
677                 ret = nvme_setup_discard(ns, req, cmd);
678                 break;
679         case REQ_OP_READ:
680         case REQ_OP_WRITE:
681                 ret = nvme_setup_rw(ns, req, cmd);
682                 break;
683         default:
684                 WARN_ON_ONCE(1);
685                 return BLK_STS_IOERR;
686         }
687
688         cmd->common.command_id = req->tag;
689         trace_nvme_setup_cmd(req, cmd);
690         return ret;
691 }
692 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
693
694 /*
695  * Returns 0 on success.  If the result is negative, it's a Linux error code;
696  * if the result is positive, it's an NVM Express status code
697  */
698 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
699                 union nvme_result *result, void *buffer, unsigned bufflen,
700                 unsigned timeout, int qid, int at_head,
701                 blk_mq_req_flags_t flags)
702 {
703         struct request *req;
704         int ret;
705
706         req = nvme_alloc_request(q, cmd, flags, qid);
707         if (IS_ERR(req))
708                 return PTR_ERR(req);
709
710         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
711
712         if (buffer && bufflen) {
713                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
714                 if (ret)
715                         goto out;
716         }
717
718         blk_execute_rq(req->q, NULL, req, at_head);
719         if (result)
720                 *result = nvme_req(req)->result;
721         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
722                 ret = -EINTR;
723         else
724                 ret = nvme_req(req)->status;
725  out:
726         blk_mq_free_request(req);
727         return ret;
728 }
729 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
730
731 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
732                 void *buffer, unsigned bufflen)
733 {
734         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
735                         NVME_QID_ANY, 0, 0);
736 }
737 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
738
739 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
740                 unsigned len, u32 seed, bool write)
741 {
742         struct bio_integrity_payload *bip;
743         int ret = -ENOMEM;
744         void *buf;
745
746         buf = kmalloc(len, GFP_KERNEL);
747         if (!buf)
748                 goto out;
749
750         ret = -EFAULT;
751         if (write && copy_from_user(buf, ubuf, len))
752                 goto out_free_meta;
753
754         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
755         if (IS_ERR(bip)) {
756                 ret = PTR_ERR(bip);
757                 goto out_free_meta;
758         }
759
760         bip->bip_iter.bi_size = len;
761         bip->bip_iter.bi_sector = seed;
762         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
763                         offset_in_page(buf));
764         if (ret == len)
765                 return buf;
766         ret = -ENOMEM;
767 out_free_meta:
768         kfree(buf);
769 out:
770         return ERR_PTR(ret);
771 }
772
773 static int nvme_submit_user_cmd(struct request_queue *q,
774                 struct nvme_command *cmd, void __user *ubuffer,
775                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
776                 u32 meta_seed, u32 *result, unsigned timeout)
777 {
778         bool write = nvme_is_write(cmd);
779         struct nvme_ns *ns = q->queuedata;
780         struct gendisk *disk = ns ? ns->disk : NULL;
781         struct request *req;
782         struct bio *bio = NULL;
783         void *meta = NULL;
784         int ret;
785
786         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
787         if (IS_ERR(req))
788                 return PTR_ERR(req);
789
790         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
791         nvme_req(req)->flags |= NVME_REQ_USERCMD;
792
793         if (ubuffer && bufflen) {
794                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
795                                 GFP_KERNEL);
796                 if (ret)
797                         goto out;
798                 bio = req->bio;
799                 bio->bi_disk = disk;
800                 if (disk && meta_buffer && meta_len) {
801                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
802                                         meta_seed, write);
803                         if (IS_ERR(meta)) {
804                                 ret = PTR_ERR(meta);
805                                 goto out_unmap;
806                         }
807                         req->cmd_flags |= REQ_INTEGRITY;
808                 }
809         }
810
811         blk_execute_rq(req->q, disk, req, 0);
812         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
813                 ret = -EINTR;
814         else
815                 ret = nvme_req(req)->status;
816         if (result)
817                 *result = le32_to_cpu(nvme_req(req)->result.u32);
818         if (meta && !ret && !write) {
819                 if (copy_to_user(meta_buffer, meta, meta_len))
820                         ret = -EFAULT;
821         }
822         kfree(meta);
823  out_unmap:
824         if (bio)
825                 blk_rq_unmap_user(bio);
826  out:
827         blk_mq_free_request(req);
828         return ret;
829 }
830
831 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
832 {
833         struct nvme_ctrl *ctrl = rq->end_io_data;
834
835         blk_mq_free_request(rq);
836
837         if (status) {
838                 dev_err(ctrl->device,
839                         "failed nvme_keep_alive_end_io error=%d\n",
840                                 status);
841                 return;
842         }
843
844         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
845 }
846
847 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
848 {
849         struct request *rq;
850
851         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
852                         NVME_QID_ANY);
853         if (IS_ERR(rq))
854                 return PTR_ERR(rq);
855
856         rq->timeout = ctrl->kato * HZ;
857         rq->end_io_data = ctrl;
858
859         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
860
861         return 0;
862 }
863
864 static void nvme_keep_alive_work(struct work_struct *work)
865 {
866         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
867                         struct nvme_ctrl, ka_work);
868
869         if (nvme_keep_alive(ctrl)) {
870                 /* allocation failure, reset the controller */
871                 dev_err(ctrl->device, "keep-alive failed\n");
872                 nvme_reset_ctrl(ctrl);
873                 return;
874         }
875 }
876
877 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
878 {
879         if (unlikely(ctrl->kato == 0))
880                 return;
881
882         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
883 }
884
885 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
886 {
887         if (unlikely(ctrl->kato == 0))
888                 return;
889
890         cancel_delayed_work_sync(&ctrl->ka_work);
891 }
892 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
893
894 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
895 {
896         struct nvme_command c = { };
897         int error;
898
899         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
900         c.identify.opcode = nvme_admin_identify;
901         c.identify.cns = NVME_ID_CNS_CTRL;
902
903         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
904         if (!*id)
905                 return -ENOMEM;
906
907         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
908                         sizeof(struct nvme_id_ctrl));
909         if (error)
910                 kfree(*id);
911         return error;
912 }
913
914 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
915                 struct nvme_ns_ids *ids)
916 {
917         struct nvme_command c = { };
918         int status;
919         void *data;
920         int pos;
921         int len;
922
923         c.identify.opcode = nvme_admin_identify;
924         c.identify.nsid = cpu_to_le32(nsid);
925         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
926
927         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
928         if (!data)
929                 return -ENOMEM;
930
931         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
932                                       NVME_IDENTIFY_DATA_SIZE);
933         if (status)
934                 goto free_data;
935
936         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
937                 struct nvme_ns_id_desc *cur = data + pos;
938
939                 if (cur->nidl == 0)
940                         break;
941
942                 switch (cur->nidt) {
943                 case NVME_NIDT_EUI64:
944                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
945                                 dev_warn(ctrl->device,
946                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
947                                          cur->nidl);
948                                 goto free_data;
949                         }
950                         len = NVME_NIDT_EUI64_LEN;
951                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
952                         break;
953                 case NVME_NIDT_NGUID:
954                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
955                                 dev_warn(ctrl->device,
956                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
957                                          cur->nidl);
958                                 goto free_data;
959                         }
960                         len = NVME_NIDT_NGUID_LEN;
961                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
962                         break;
963                 case NVME_NIDT_UUID:
964                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
965                                 dev_warn(ctrl->device,
966                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
967                                          cur->nidl);
968                                 goto free_data;
969                         }
970                         len = NVME_NIDT_UUID_LEN;
971                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
972                         break;
973                 default:
974                         /* Skip unknown types */
975                         len = cur->nidl;
976                         break;
977                 }
978
979                 len += sizeof(*cur);
980         }
981 free_data:
982         kfree(data);
983         return status;
984 }
985
986 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
987 {
988         struct nvme_command c = { };
989
990         c.identify.opcode = nvme_admin_identify;
991         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
992         c.identify.nsid = cpu_to_le32(nsid);
993         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
994                                     NVME_IDENTIFY_DATA_SIZE);
995 }
996
997 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
998                 unsigned nsid)
999 {
1000         struct nvme_id_ns *id;
1001         struct nvme_command c = { };
1002         int error;
1003
1004         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1005         c.identify.opcode = nvme_admin_identify;
1006         c.identify.nsid = cpu_to_le32(nsid);
1007         c.identify.cns = NVME_ID_CNS_NS;
1008
1009         id = kmalloc(sizeof(*id), GFP_KERNEL);
1010         if (!id)
1011                 return NULL;
1012
1013         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1014         if (error) {
1015                 dev_warn(ctrl->device, "Identify namespace failed\n");
1016                 kfree(id);
1017                 return NULL;
1018         }
1019
1020         return id;
1021 }
1022
1023 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1024                       void *buffer, size_t buflen, u32 *result)
1025 {
1026         struct nvme_command c;
1027         union nvme_result res;
1028         int ret;
1029
1030         memset(&c, 0, sizeof(c));
1031         c.features.opcode = nvme_admin_set_features;
1032         c.features.fid = cpu_to_le32(fid);
1033         c.features.dword11 = cpu_to_le32(dword11);
1034
1035         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1036                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1037         if (ret >= 0 && result)
1038                 *result = le32_to_cpu(res.u32);
1039         return ret;
1040 }
1041
1042 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1043 {
1044         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1045         u32 result;
1046         int status, nr_io_queues;
1047
1048         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1049                         &result);
1050         if (status < 0)
1051                 return status;
1052
1053         /*
1054          * Degraded controllers might return an error when setting the queue
1055          * count.  We still want to be able to bring them online and offer
1056          * access to the admin queue, as that might be only way to fix them up.
1057          */
1058         if (status > 0) {
1059                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1060                 *count = 0;
1061         } else {
1062                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1063                 *count = min(*count, nr_io_queues);
1064         }
1065
1066         return 0;
1067 }
1068 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1069
1070 #define NVME_AEN_SUPPORTED \
1071         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1072
1073 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1074 {
1075         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1076         int status;
1077
1078         if (!supported_aens)
1079                 return;
1080
1081         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1082                         NULL, 0, &result);
1083         if (status)
1084                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1085                          supported_aens);
1086 }
1087
1088 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1089 {
1090         struct nvme_user_io io;
1091         struct nvme_command c;
1092         unsigned length, meta_len;
1093         void __user *metadata;
1094
1095         if (copy_from_user(&io, uio, sizeof(io)))
1096                 return -EFAULT;
1097         if (io.flags)
1098                 return -EINVAL;
1099
1100         switch (io.opcode) {
1101         case nvme_cmd_write:
1102         case nvme_cmd_read:
1103         case nvme_cmd_compare:
1104                 break;
1105         default:
1106                 return -EINVAL;
1107         }
1108
1109         length = (io.nblocks + 1) << ns->lba_shift;
1110         meta_len = (io.nblocks + 1) * ns->ms;
1111         metadata = (void __user *)(uintptr_t)io.metadata;
1112
1113         if (ns->ext) {
1114                 length += meta_len;
1115                 meta_len = 0;
1116         } else if (meta_len) {
1117                 if ((io.metadata & 3) || !io.metadata)
1118                         return -EINVAL;
1119         }
1120
1121         memset(&c, 0, sizeof(c));
1122         c.rw.opcode = io.opcode;
1123         c.rw.flags = io.flags;
1124         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1125         c.rw.slba = cpu_to_le64(io.slba);
1126         c.rw.length = cpu_to_le16(io.nblocks);
1127         c.rw.control = cpu_to_le16(io.control);
1128         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1129         c.rw.reftag = cpu_to_le32(io.reftag);
1130         c.rw.apptag = cpu_to_le16(io.apptag);
1131         c.rw.appmask = cpu_to_le16(io.appmask);
1132
1133         return nvme_submit_user_cmd(ns->queue, &c,
1134                         (void __user *)(uintptr_t)io.addr, length,
1135                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1136 }
1137
1138 static u32 nvme_known_admin_effects(u8 opcode)
1139 {
1140         switch (opcode) {
1141         case nvme_admin_format_nvm:
1142                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1143                                         NVME_CMD_EFFECTS_CSE_MASK;
1144         case nvme_admin_sanitize_nvm:
1145                 return NVME_CMD_EFFECTS_CSE_MASK;
1146         default:
1147                 break;
1148         }
1149         return 0;
1150 }
1151
1152 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1153                                                                 u8 opcode)
1154 {
1155         u32 effects = 0;
1156
1157         if (ns) {
1158                 if (ctrl->effects)
1159                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1160                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1161                         dev_warn(ctrl->device,
1162                                  "IO command:%02x has unhandled effects:%08x\n",
1163                                  opcode, effects);
1164                 return 0;
1165         }
1166
1167         if (ctrl->effects)
1168                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1169         else
1170                 effects = nvme_known_admin_effects(opcode);
1171
1172         /*
1173          * For simplicity, IO to all namespaces is quiesced even if the command
1174          * effects say only one namespace is affected.
1175          */
1176         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1177                 nvme_start_freeze(ctrl);
1178                 nvme_wait_freeze(ctrl);
1179         }
1180         return effects;
1181 }
1182
1183 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1184 {
1185         struct nvme_ns *ns;
1186
1187         down_read(&ctrl->namespaces_rwsem);
1188         list_for_each_entry(ns, &ctrl->namespaces, list)
1189                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1190                         nvme_set_queue_dying(ns);
1191         up_read(&ctrl->namespaces_rwsem);
1192
1193         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1194 }
1195
1196 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1197 {
1198         /*
1199          * Revalidate LBA changes prior to unfreezing. This is necessary to
1200          * prevent memory corruption if a logical block size was changed by
1201          * this command.
1202          */
1203         if (effects & NVME_CMD_EFFECTS_LBCC)
1204                 nvme_update_formats(ctrl);
1205         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1206                 nvme_unfreeze(ctrl);
1207         if (effects & NVME_CMD_EFFECTS_CCC)
1208                 nvme_init_identify(ctrl);
1209         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1210                 nvme_queue_scan(ctrl);
1211 }
1212
1213 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1214                         struct nvme_passthru_cmd __user *ucmd)
1215 {
1216         struct nvme_passthru_cmd cmd;
1217         struct nvme_command c;
1218         unsigned timeout = 0;
1219         u32 effects;
1220         int status;
1221
1222         if (!capable(CAP_SYS_ADMIN))
1223                 return -EACCES;
1224         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1225                 return -EFAULT;
1226         if (cmd.flags)
1227                 return -EINVAL;
1228
1229         memset(&c, 0, sizeof(c));
1230         c.common.opcode = cmd.opcode;
1231         c.common.flags = cmd.flags;
1232         c.common.nsid = cpu_to_le32(cmd.nsid);
1233         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1234         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1235         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1236         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1237         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1238         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1239         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1240         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1241
1242         if (cmd.timeout_ms)
1243                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1244
1245         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1246         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1247                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1248                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1249                         0, &cmd.result, timeout);
1250         nvme_passthru_end(ctrl, effects);
1251
1252         if (status >= 0) {
1253                 if (put_user(cmd.result, &ucmd->result))
1254                         return -EFAULT;
1255         }
1256
1257         return status;
1258 }
1259
1260 /*
1261  * Issue ioctl requests on the first available path.  Note that unlike normal
1262  * block layer requests we will not retry failed request on another controller.
1263  */
1264 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1265                 struct nvme_ns_head **head, int *srcu_idx)
1266 {
1267 #ifdef CONFIG_NVME_MULTIPATH
1268         if (disk->fops == &nvme_ns_head_ops) {
1269                 *head = disk->private_data;
1270                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1271                 return nvme_find_path(*head);
1272         }
1273 #endif
1274         *head = NULL;
1275         *srcu_idx = -1;
1276         return disk->private_data;
1277 }
1278
1279 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1280 {
1281         if (head)
1282                 srcu_read_unlock(&head->srcu, idx);
1283 }
1284
1285 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1286 {
1287         switch (cmd) {
1288         case NVME_IOCTL_ID:
1289                 force_successful_syscall_return();
1290                 return ns->head->ns_id;
1291         case NVME_IOCTL_ADMIN_CMD:
1292                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1293         case NVME_IOCTL_IO_CMD:
1294                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1295         case NVME_IOCTL_SUBMIT_IO:
1296                 return nvme_submit_io(ns, (void __user *)arg);
1297         default:
1298 #ifdef CONFIG_NVM
1299                 if (ns->ndev)
1300                         return nvme_nvm_ioctl(ns, cmd, arg);
1301 #endif
1302                 if (is_sed_ioctl(cmd))
1303                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1304                                          (void __user *) arg);
1305                 return -ENOTTY;
1306         }
1307 }
1308
1309 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1310                 unsigned int cmd, unsigned long arg)
1311 {
1312         struct nvme_ns_head *head = NULL;
1313         struct nvme_ns *ns;
1314         int srcu_idx, ret;
1315
1316         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1317         if (unlikely(!ns))
1318                 ret = -EWOULDBLOCK;
1319         else
1320                 ret = nvme_ns_ioctl(ns, cmd, arg);
1321         nvme_put_ns_from_disk(head, srcu_idx);
1322         return ret;
1323 }
1324
1325 static int nvme_open(struct block_device *bdev, fmode_t mode)
1326 {
1327         struct nvme_ns *ns = bdev->bd_disk->private_data;
1328
1329 #ifdef CONFIG_NVME_MULTIPATH
1330         /* should never be called due to GENHD_FL_HIDDEN */
1331         if (WARN_ON_ONCE(ns->head->disk))
1332                 goto fail;
1333 #endif
1334         if (!kref_get_unless_zero(&ns->kref))
1335                 goto fail;
1336         if (!try_module_get(ns->ctrl->ops->module))
1337                 goto fail_put_ns;
1338
1339         return 0;
1340
1341 fail_put_ns:
1342         nvme_put_ns(ns);
1343 fail:
1344         return -ENXIO;
1345 }
1346
1347 static void nvme_release(struct gendisk *disk, fmode_t mode)
1348 {
1349         struct nvme_ns *ns = disk->private_data;
1350
1351         module_put(ns->ctrl->ops->module);
1352         nvme_put_ns(ns);
1353 }
1354
1355 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1356 {
1357         /* some standard values */
1358         geo->heads = 1 << 6;
1359         geo->sectors = 1 << 5;
1360         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1361         return 0;
1362 }
1363
1364 #ifdef CONFIG_BLK_DEV_INTEGRITY
1365 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1366 {
1367         struct blk_integrity integrity;
1368
1369         memset(&integrity, 0, sizeof(integrity));
1370         switch (pi_type) {
1371         case NVME_NS_DPS_PI_TYPE3:
1372                 integrity.profile = &t10_pi_type3_crc;
1373                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1374                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1375                 break;
1376         case NVME_NS_DPS_PI_TYPE1:
1377         case NVME_NS_DPS_PI_TYPE2:
1378                 integrity.profile = &t10_pi_type1_crc;
1379                 integrity.tag_size = sizeof(u16);
1380                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1381                 break;
1382         default:
1383                 integrity.profile = NULL;
1384                 break;
1385         }
1386         integrity.tuple_size = ms;
1387         blk_integrity_register(disk, &integrity);
1388         blk_queue_max_integrity_segments(disk->queue, 1);
1389 }
1390 #else
1391 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1392 {
1393 }
1394 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1395
1396 static void nvme_set_chunk_size(struct nvme_ns *ns)
1397 {
1398         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1399         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1400 }
1401
1402 static void nvme_config_discard(struct nvme_ns *ns)
1403 {
1404         struct nvme_ctrl *ctrl = ns->ctrl;
1405         struct request_queue *queue = ns->queue;
1406         u32 size = queue_logical_block_size(queue);
1407
1408         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1409                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1410                 return;
1411         }
1412
1413         if (ctrl->nr_streams && ns->sws && ns->sgs)
1414                 size *= ns->sws * ns->sgs;
1415
1416         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1417                         NVME_DSM_MAX_RANGES);
1418
1419         queue->limits.discard_alignment = 0;
1420         queue->limits.discard_granularity = size;
1421
1422         /* If discard is already enabled, don't reset queue limits */
1423         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1424                 return;
1425
1426         blk_queue_max_discard_sectors(queue, UINT_MAX);
1427         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1428
1429         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1430                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1431 }
1432
1433 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1434                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1435 {
1436         memset(ids, 0, sizeof(*ids));
1437
1438         if (ctrl->vs >= NVME_VS(1, 1, 0))
1439                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1440         if (ctrl->vs >= NVME_VS(1, 2, 0))
1441                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1442         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1443                  /* Don't treat error as fatal we potentially
1444                   * already have a NGUID or EUI-64
1445                   */
1446                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1447                         dev_warn(ctrl->device,
1448                                  "%s: Identify Descriptors failed\n", __func__);
1449         }
1450 }
1451
1452 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1453 {
1454         return !uuid_is_null(&ids->uuid) ||
1455                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1456                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1457 }
1458
1459 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1460 {
1461         return uuid_equal(&a->uuid, &b->uuid) &&
1462                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1463                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1464 }
1465
1466 static void nvme_update_disk_info(struct gendisk *disk,
1467                 struct nvme_ns *ns, struct nvme_id_ns *id)
1468 {
1469         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1470         unsigned short bs = 1 << ns->lba_shift;
1471
1472         blk_mq_freeze_queue(disk->queue);
1473         blk_integrity_unregister(disk);
1474
1475         blk_queue_logical_block_size(disk->queue, bs);
1476         blk_queue_physical_block_size(disk->queue, bs);
1477         blk_queue_io_min(disk->queue, bs);
1478
1479         if (ns->ms && !ns->ext &&
1480             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1481                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1482         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1483                 capacity = 0;
1484
1485         set_capacity(disk, capacity);
1486         nvme_config_discard(ns);
1487
1488         if (id->nsattr & (1 << 0))
1489                 set_disk_ro(disk, true);
1490         else
1491                 set_disk_ro(disk, false);
1492
1493         blk_mq_unfreeze_queue(disk->queue);
1494 }
1495
1496 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1497 {
1498         struct nvme_ns *ns = disk->private_data;
1499
1500         /*
1501          * If identify namespace failed, use default 512 byte block size so
1502          * block layer can use before failing read/write for 0 capacity.
1503          */
1504         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1505         if (ns->lba_shift == 0)
1506                 ns->lba_shift = 9;
1507         ns->noiob = le16_to_cpu(id->noiob);
1508         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1509         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1510         /* the PI implementation requires metadata equal t10 pi tuple size */
1511         if (ns->ms == sizeof(struct t10_pi_tuple))
1512                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1513         else
1514                 ns->pi_type = 0;
1515
1516         if (ns->noiob)
1517                 nvme_set_chunk_size(ns);
1518         nvme_update_disk_info(disk, ns, id);
1519         if (ns->ndev)
1520                 nvme_nvm_update_nvm_info(ns);
1521 #ifdef CONFIG_NVME_MULTIPATH
1522         if (ns->head->disk)
1523                 nvme_update_disk_info(ns->head->disk, ns, id);
1524 #endif
1525 }
1526
1527 static int nvme_revalidate_disk(struct gendisk *disk)
1528 {
1529         struct nvme_ns *ns = disk->private_data;
1530         struct nvme_ctrl *ctrl = ns->ctrl;
1531         struct nvme_id_ns *id;
1532         struct nvme_ns_ids ids;
1533         int ret = 0;
1534
1535         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1536                 set_capacity(disk, 0);
1537                 return -ENODEV;
1538         }
1539
1540         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1541         if (!id)
1542                 return -ENODEV;
1543
1544         if (id->ncap == 0) {
1545                 ret = -ENODEV;
1546                 goto out;
1547         }
1548
1549         __nvme_revalidate_disk(disk, id);
1550         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1551         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1552                 dev_err(ctrl->device,
1553                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1554                 ret = -ENODEV;
1555         }
1556
1557 out:
1558         kfree(id);
1559         return ret;
1560 }
1561
1562 static char nvme_pr_type(enum pr_type type)
1563 {
1564         switch (type) {
1565         case PR_WRITE_EXCLUSIVE:
1566                 return 1;
1567         case PR_EXCLUSIVE_ACCESS:
1568                 return 2;
1569         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1570                 return 3;
1571         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1572                 return 4;
1573         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1574                 return 5;
1575         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1576                 return 6;
1577         default:
1578                 return 0;
1579         }
1580 };
1581
1582 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1583                                 u64 key, u64 sa_key, u8 op)
1584 {
1585         struct nvme_ns_head *head = NULL;
1586         struct nvme_ns *ns;
1587         struct nvme_command c;
1588         int srcu_idx, ret;
1589         u8 data[16] = { 0, };
1590
1591         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1592         if (unlikely(!ns))
1593                 return -EWOULDBLOCK;
1594
1595         put_unaligned_le64(key, &data[0]);
1596         put_unaligned_le64(sa_key, &data[8]);
1597
1598         memset(&c, 0, sizeof(c));
1599         c.common.opcode = op;
1600         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1601         c.common.cdw10[0] = cpu_to_le32(cdw10);
1602
1603         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1604         nvme_put_ns_from_disk(head, srcu_idx);
1605         return ret;
1606 }
1607
1608 static int nvme_pr_register(struct block_device *bdev, u64 old,
1609                 u64 new, unsigned flags)
1610 {
1611         u32 cdw10;
1612
1613         if (flags & ~PR_FL_IGNORE_KEY)
1614                 return -EOPNOTSUPP;
1615
1616         cdw10 = old ? 2 : 0;
1617         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1618         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1619         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1620 }
1621
1622 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1623                 enum pr_type type, unsigned flags)
1624 {
1625         u32 cdw10;
1626
1627         if (flags & ~PR_FL_IGNORE_KEY)
1628                 return -EOPNOTSUPP;
1629
1630         cdw10 = nvme_pr_type(type) << 8;
1631         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1632         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1633 }
1634
1635 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1636                 enum pr_type type, bool abort)
1637 {
1638         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1639         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1640 }
1641
1642 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1643 {
1644         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1645         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1646 }
1647
1648 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1649 {
1650         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1651         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1652 }
1653
1654 static const struct pr_ops nvme_pr_ops = {
1655         .pr_register    = nvme_pr_register,
1656         .pr_reserve     = nvme_pr_reserve,
1657         .pr_release     = nvme_pr_release,
1658         .pr_preempt     = nvme_pr_preempt,
1659         .pr_clear       = nvme_pr_clear,
1660 };
1661
1662 #ifdef CONFIG_BLK_SED_OPAL
1663 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1664                 bool send)
1665 {
1666         struct nvme_ctrl *ctrl = data;
1667         struct nvme_command cmd;
1668
1669         memset(&cmd, 0, sizeof(cmd));
1670         if (send)
1671                 cmd.common.opcode = nvme_admin_security_send;
1672         else
1673                 cmd.common.opcode = nvme_admin_security_recv;
1674         cmd.common.nsid = 0;
1675         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1676         cmd.common.cdw10[1] = cpu_to_le32(len);
1677
1678         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1679                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1680 }
1681 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1682 #endif /* CONFIG_BLK_SED_OPAL */
1683
1684 static const struct block_device_operations nvme_fops = {
1685         .owner          = THIS_MODULE,
1686         .ioctl          = nvme_ioctl,
1687         .compat_ioctl   = nvme_ioctl,
1688         .open           = nvme_open,
1689         .release        = nvme_release,
1690         .getgeo         = nvme_getgeo,
1691         .revalidate_disk= nvme_revalidate_disk,
1692         .pr_ops         = &nvme_pr_ops,
1693 };
1694
1695 #ifdef CONFIG_NVME_MULTIPATH
1696 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1697 {
1698         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1699
1700         if (!kref_get_unless_zero(&head->ref))
1701                 return -ENXIO;
1702         return 0;
1703 }
1704
1705 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1706 {
1707         nvme_put_ns_head(disk->private_data);
1708 }
1709
1710 const struct block_device_operations nvme_ns_head_ops = {
1711         .owner          = THIS_MODULE,
1712         .open           = nvme_ns_head_open,
1713         .release        = nvme_ns_head_release,
1714         .ioctl          = nvme_ioctl,
1715         .compat_ioctl   = nvme_ioctl,
1716         .getgeo         = nvme_getgeo,
1717         .pr_ops         = &nvme_pr_ops,
1718 };
1719 #endif /* CONFIG_NVME_MULTIPATH */
1720
1721 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1722 {
1723         unsigned long timeout =
1724                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1725         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1726         int ret;
1727
1728         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1729                 if (csts == ~0)
1730                         return -ENODEV;
1731                 if ((csts & NVME_CSTS_RDY) == bit)
1732                         break;
1733
1734                 msleep(100);
1735                 if (fatal_signal_pending(current))
1736                         return -EINTR;
1737                 if (time_after(jiffies, timeout)) {
1738                         dev_err(ctrl->device,
1739                                 "Device not ready; aborting %s\n", enabled ?
1740                                                 "initialisation" : "reset");
1741                         return -ENODEV;
1742                 }
1743         }
1744
1745         return ret;
1746 }
1747
1748 /*
1749  * If the device has been passed off to us in an enabled state, just clear
1750  * the enabled bit.  The spec says we should set the 'shutdown notification
1751  * bits', but doing so may cause the device to complete commands to the
1752  * admin queue ... and we don't know what memory that might be pointing at!
1753  */
1754 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1755 {
1756         int ret;
1757
1758         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1759         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1760
1761         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1762         if (ret)
1763                 return ret;
1764
1765         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1766                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1767
1768         return nvme_wait_ready(ctrl, cap, false);
1769 }
1770 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1771
1772 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1773 {
1774         /*
1775          * Default to a 4K page size, with the intention to update this
1776          * path in the future to accomodate architectures with differing
1777          * kernel and IO page sizes.
1778          */
1779         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1780         int ret;
1781
1782         if (page_shift < dev_page_min) {
1783                 dev_err(ctrl->device,
1784                         "Minimum device page size %u too large for host (%u)\n",
1785                         1 << dev_page_min, 1 << page_shift);
1786                 return -ENODEV;
1787         }
1788
1789         ctrl->page_size = 1 << page_shift;
1790
1791         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1792         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1793         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1794         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1795         ctrl->ctrl_config |= NVME_CC_ENABLE;
1796
1797         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1798         if (ret)
1799                 return ret;
1800         return nvme_wait_ready(ctrl, cap, true);
1801 }
1802 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1803
1804 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1805 {
1806         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1807         u32 csts;
1808         int ret;
1809
1810         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1811         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1812
1813         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1814         if (ret)
1815                 return ret;
1816
1817         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1818                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1819                         break;
1820
1821                 msleep(100);
1822                 if (fatal_signal_pending(current))
1823                         return -EINTR;
1824                 if (time_after(jiffies, timeout)) {
1825                         dev_err(ctrl->device,
1826                                 "Device shutdown incomplete; abort shutdown\n");
1827                         return -ENODEV;
1828                 }
1829         }
1830
1831         return ret;
1832 }
1833 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1834
1835 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1836                 struct request_queue *q)
1837 {
1838         bool vwc = false;
1839
1840         if (ctrl->max_hw_sectors) {
1841                 u32 max_segments =
1842                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1843
1844                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1845                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1846                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1847         }
1848         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1849             is_power_of_2(ctrl->max_hw_sectors))
1850                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1851         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1852         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1853                 vwc = true;
1854         blk_queue_write_cache(q, vwc, vwc);
1855 }
1856
1857 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1858 {
1859         __le64 ts;
1860         int ret;
1861
1862         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1863                 return 0;
1864
1865         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1866         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1867                         NULL);
1868         if (ret)
1869                 dev_warn_once(ctrl->device,
1870                         "could not set timestamp (%d)\n", ret);
1871         return ret;
1872 }
1873
1874 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1875 {
1876         /*
1877          * APST (Autonomous Power State Transition) lets us program a
1878          * table of power state transitions that the controller will
1879          * perform automatically.  We configure it with a simple
1880          * heuristic: we are willing to spend at most 2% of the time
1881          * transitioning between power states.  Therefore, when running
1882          * in any given state, we will enter the next lower-power
1883          * non-operational state after waiting 50 * (enlat + exlat)
1884          * microseconds, as long as that state's exit latency is under
1885          * the requested maximum latency.
1886          *
1887          * We will not autonomously enter any non-operational state for
1888          * which the total latency exceeds ps_max_latency_us.  Users
1889          * can set ps_max_latency_us to zero to turn off APST.
1890          */
1891
1892         unsigned apste;
1893         struct nvme_feat_auto_pst *table;
1894         u64 max_lat_us = 0;
1895         int max_ps = -1;
1896         int ret;
1897
1898         /*
1899          * If APST isn't supported or if we haven't been initialized yet,
1900          * then don't do anything.
1901          */
1902         if (!ctrl->apsta)
1903                 return 0;
1904
1905         if (ctrl->npss > 31) {
1906                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1907                 return 0;
1908         }
1909
1910         table = kzalloc(sizeof(*table), GFP_KERNEL);
1911         if (!table)
1912                 return 0;
1913
1914         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1915                 /* Turn off APST. */
1916                 apste = 0;
1917                 dev_dbg(ctrl->device, "APST disabled\n");
1918         } else {
1919                 __le64 target = cpu_to_le64(0);
1920                 int state;
1921
1922                 /*
1923                  * Walk through all states from lowest- to highest-power.
1924                  * According to the spec, lower-numbered states use more
1925                  * power.  NPSS, despite the name, is the index of the
1926                  * lowest-power state, not the number of states.
1927                  */
1928                 for (state = (int)ctrl->npss; state >= 0; state--) {
1929                         u64 total_latency_us, exit_latency_us, transition_ms;
1930
1931                         if (target)
1932                                 table->entries[state] = target;
1933
1934                         /*
1935                          * Don't allow transitions to the deepest state
1936                          * if it's quirked off.
1937                          */
1938                         if (state == ctrl->npss &&
1939                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1940                                 continue;
1941
1942                         /*
1943                          * Is this state a useful non-operational state for
1944                          * higher-power states to autonomously transition to?
1945                          */
1946                         if (!(ctrl->psd[state].flags &
1947                               NVME_PS_FLAGS_NON_OP_STATE))
1948                                 continue;
1949
1950                         exit_latency_us =
1951                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1952                         if (exit_latency_us > ctrl->ps_max_latency_us)
1953                                 continue;
1954
1955                         total_latency_us =
1956                                 exit_latency_us +
1957                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1958
1959                         /*
1960                          * This state is good.  Use it as the APST idle
1961                          * target for higher power states.
1962                          */
1963                         transition_ms = total_latency_us + 19;
1964                         do_div(transition_ms, 20);
1965                         if (transition_ms > (1 << 24) - 1)
1966                                 transition_ms = (1 << 24) - 1;
1967
1968                         target = cpu_to_le64((state << 3) |
1969                                              (transition_ms << 8));
1970
1971                         if (max_ps == -1)
1972                                 max_ps = state;
1973
1974                         if (total_latency_us > max_lat_us)
1975                                 max_lat_us = total_latency_us;
1976                 }
1977
1978                 apste = 1;
1979
1980                 if (max_ps == -1) {
1981                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1982                 } else {
1983                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1984                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1985                 }
1986         }
1987
1988         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1989                                 table, sizeof(*table), NULL);
1990         if (ret)
1991                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1992
1993         kfree(table);
1994         return ret;
1995 }
1996
1997 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1998 {
1999         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2000         u64 latency;
2001
2002         switch (val) {
2003         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2004         case PM_QOS_LATENCY_ANY:
2005                 latency = U64_MAX;
2006                 break;
2007
2008         default:
2009                 latency = val;
2010         }
2011
2012         if (ctrl->ps_max_latency_us != latency) {
2013                 ctrl->ps_max_latency_us = latency;
2014                 nvme_configure_apst(ctrl);
2015         }
2016 }
2017
2018 struct nvme_core_quirk_entry {
2019         /*
2020          * NVMe model and firmware strings are padded with spaces.  For
2021          * simplicity, strings in the quirk table are padded with NULLs
2022          * instead.
2023          */
2024         u16 vid;
2025         const char *mn;
2026         const char *fr;
2027         unsigned long quirks;
2028 };
2029
2030 static const struct nvme_core_quirk_entry core_quirks[] = {
2031         {
2032                 /*
2033                  * This Toshiba device seems to die using any APST states.  See:
2034                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2035                  */
2036                 .vid = 0x1179,
2037                 .mn = "THNSF5256GPUK TOSHIBA",
2038                 .quirks = NVME_QUIRK_NO_APST,
2039         }
2040 };
2041
2042 /* match is null-terminated but idstr is space-padded. */
2043 static bool string_matches(const char *idstr, const char *match, size_t len)
2044 {
2045         size_t matchlen;
2046
2047         if (!match)
2048                 return true;
2049
2050         matchlen = strlen(match);
2051         WARN_ON_ONCE(matchlen > len);
2052
2053         if (memcmp(idstr, match, matchlen))
2054                 return false;
2055
2056         for (; matchlen < len; matchlen++)
2057                 if (idstr[matchlen] != ' ')
2058                         return false;
2059
2060         return true;
2061 }
2062
2063 static bool quirk_matches(const struct nvme_id_ctrl *id,
2064                           const struct nvme_core_quirk_entry *q)
2065 {
2066         return q->vid == le16_to_cpu(id->vid) &&
2067                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2068                 string_matches(id->fr, q->fr, sizeof(id->fr));
2069 }
2070
2071 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2072                 struct nvme_id_ctrl *id)
2073 {
2074         size_t nqnlen;
2075         int off;
2076
2077         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2078         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2079                 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2080                 return;
2081         }
2082
2083         if (ctrl->vs >= NVME_VS(1, 2, 1))
2084                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2085
2086         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2087         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2088                         "nqn.2014.08.org.nvmexpress:%4x%4x",
2089                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2090         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2091         off += sizeof(id->sn);
2092         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2093         off += sizeof(id->mn);
2094         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2095 }
2096
2097 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2098 {
2099         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2100         kfree(subsys);
2101 }
2102
2103 static void nvme_release_subsystem(struct device *dev)
2104 {
2105         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2106 }
2107
2108 static void nvme_destroy_subsystem(struct kref *ref)
2109 {
2110         struct nvme_subsystem *subsys =
2111                         container_of(ref, struct nvme_subsystem, ref);
2112
2113         mutex_lock(&nvme_subsystems_lock);
2114         list_del(&subsys->entry);
2115         mutex_unlock(&nvme_subsystems_lock);
2116
2117         ida_destroy(&subsys->ns_ida);
2118         device_del(&subsys->dev);
2119         put_device(&subsys->dev);
2120 }
2121
2122 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2123 {
2124         kref_put(&subsys->ref, nvme_destroy_subsystem);
2125 }
2126
2127 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2128 {
2129         struct nvme_subsystem *subsys;
2130
2131         lockdep_assert_held(&nvme_subsystems_lock);
2132
2133         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2134                 if (strcmp(subsys->subnqn, subsysnqn))
2135                         continue;
2136                 if (!kref_get_unless_zero(&subsys->ref))
2137                         continue;
2138                 return subsys;
2139         }
2140
2141         return NULL;
2142 }
2143
2144 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2145         struct device_attribute subsys_attr_##_name = \
2146                 __ATTR(_name, _mode, _show, NULL)
2147
2148 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2149                                     struct device_attribute *attr,
2150                                     char *buf)
2151 {
2152         struct nvme_subsystem *subsys =
2153                 container_of(dev, struct nvme_subsystem, dev);
2154
2155         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2156 }
2157 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2158
2159 #define nvme_subsys_show_str_function(field)                            \
2160 static ssize_t subsys_##field##_show(struct device *dev,                \
2161                             struct device_attribute *attr, char *buf)   \
2162 {                                                                       \
2163         struct nvme_subsystem *subsys =                                 \
2164                 container_of(dev, struct nvme_subsystem, dev);          \
2165         return sprintf(buf, "%.*s\n",                                   \
2166                        (int)sizeof(subsys->field), subsys->field);      \
2167 }                                                                       \
2168 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2169
2170 nvme_subsys_show_str_function(model);
2171 nvme_subsys_show_str_function(serial);
2172 nvme_subsys_show_str_function(firmware_rev);
2173
2174 static struct attribute *nvme_subsys_attrs[] = {
2175         &subsys_attr_model.attr,
2176         &subsys_attr_serial.attr,
2177         &subsys_attr_firmware_rev.attr,
2178         &subsys_attr_subsysnqn.attr,
2179         NULL,
2180 };
2181
2182 static struct attribute_group nvme_subsys_attrs_group = {
2183         .attrs = nvme_subsys_attrs,
2184 };
2185
2186 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2187         &nvme_subsys_attrs_group,
2188         NULL,
2189 };
2190
2191 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2192 {
2193         int count = 0;
2194         struct nvme_ctrl *ctrl;
2195
2196         mutex_lock(&subsys->lock);
2197         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2198                 if (ctrl->state != NVME_CTRL_DELETING &&
2199                     ctrl->state != NVME_CTRL_DEAD)
2200                         count++;
2201         }
2202         mutex_unlock(&subsys->lock);
2203
2204         return count;
2205 }
2206
2207 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2208 {
2209         struct nvme_subsystem *subsys, *found;
2210         int ret;
2211
2212         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2213         if (!subsys)
2214                 return -ENOMEM;
2215         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2216         if (ret < 0) {
2217                 kfree(subsys);
2218                 return ret;
2219         }
2220         subsys->instance = ret;
2221         mutex_init(&subsys->lock);
2222         kref_init(&subsys->ref);
2223         INIT_LIST_HEAD(&subsys->ctrls);
2224         INIT_LIST_HEAD(&subsys->nsheads);
2225         nvme_init_subnqn(subsys, ctrl, id);
2226         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2227         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2228         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2229         subsys->vendor_id = le16_to_cpu(id->vid);
2230         subsys->cmic = id->cmic;
2231
2232         subsys->dev.class = nvme_subsys_class;
2233         subsys->dev.release = nvme_release_subsystem;
2234         subsys->dev.groups = nvme_subsys_attrs_groups;
2235         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2236         device_initialize(&subsys->dev);
2237
2238         mutex_lock(&nvme_subsystems_lock);
2239         found = __nvme_find_get_subsystem(subsys->subnqn);
2240         if (found) {
2241                 /*
2242                  * Verify that the subsystem actually supports multiple
2243                  * controllers, else bail out.
2244                  */
2245                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2246                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2247                         dev_err(ctrl->device,
2248                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2249                                 found->subnqn);
2250                         nvme_put_subsystem(found);
2251                         ret = -EINVAL;
2252                         goto out_unlock;
2253                 }
2254
2255                 __nvme_release_subsystem(subsys);
2256                 subsys = found;
2257         } else {
2258                 ret = device_add(&subsys->dev);
2259                 if (ret) {
2260                         dev_err(ctrl->device,
2261                                 "failed to register subsystem device.\n");
2262                         goto out_unlock;
2263                 }
2264                 ida_init(&subsys->ns_ida);
2265                 list_add_tail(&subsys->entry, &nvme_subsystems);
2266         }
2267
2268         ctrl->subsys = subsys;
2269         mutex_unlock(&nvme_subsystems_lock);
2270
2271         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2272                         dev_name(ctrl->device))) {
2273                 dev_err(ctrl->device,
2274                         "failed to create sysfs link from subsystem.\n");
2275                 /* the transport driver will eventually put the subsystem */
2276                 return -EINVAL;
2277         }
2278
2279         mutex_lock(&subsys->lock);
2280         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2281         mutex_unlock(&subsys->lock);
2282
2283         return 0;
2284
2285 out_unlock:
2286         mutex_unlock(&nvme_subsystems_lock);
2287         put_device(&subsys->dev);
2288         return ret;
2289 }
2290
2291 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2292                 void *log, size_t size, u64 offset)
2293 {
2294         struct nvme_command c = { };
2295         unsigned long dwlen = size / 4 - 1;
2296
2297         c.get_log_page.opcode = nvme_admin_get_log_page;
2298         c.get_log_page.nsid = cpu_to_le32(nsid);
2299         c.get_log_page.lid = log_page;
2300         c.get_log_page.lsp = lsp;
2301         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2302         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2303         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2304         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2305
2306         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2307 }
2308
2309 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2310 {
2311         int ret;
2312
2313         if (!ctrl->effects)
2314                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2315
2316         if (!ctrl->effects)
2317                 return 0;
2318
2319         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2320                         ctrl->effects, sizeof(*ctrl->effects), 0);
2321         if (ret) {
2322                 kfree(ctrl->effects);
2323                 ctrl->effects = NULL;
2324         }
2325         return ret;
2326 }
2327
2328 /*
2329  * Initialize the cached copies of the Identify data and various controller
2330  * register in our nvme_ctrl structure.  This should be called as soon as
2331  * the admin queue is fully up and running.
2332  */
2333 int nvme_init_identify(struct nvme_ctrl *ctrl)
2334 {
2335         struct nvme_id_ctrl *id;
2336         u64 cap;
2337         int ret, page_shift;
2338         u32 max_hw_sectors;
2339         bool prev_apst_enabled;
2340
2341         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2342         if (ret) {
2343                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2344                 return ret;
2345         }
2346
2347         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2348         if (ret) {
2349                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2350                 return ret;
2351         }
2352         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2353
2354         if (ctrl->vs >= NVME_VS(1, 1, 0))
2355                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2356
2357         ret = nvme_identify_ctrl(ctrl, &id);
2358         if (ret) {
2359                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2360                 return -EIO;
2361         }
2362
2363         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2364                 ret = nvme_get_effects_log(ctrl);
2365                 if (ret < 0)
2366                         goto out_free;
2367         }
2368
2369         if (!ctrl->identified) {
2370                 int i;
2371
2372                 ret = nvme_init_subsystem(ctrl, id);
2373                 if (ret)
2374                         goto out_free;
2375
2376                 /*
2377                  * Check for quirks.  Quirk can depend on firmware version,
2378                  * so, in principle, the set of quirks present can change
2379                  * across a reset.  As a possible future enhancement, we
2380                  * could re-scan for quirks every time we reinitialize
2381                  * the device, but we'd have to make sure that the driver
2382                  * behaves intelligently if the quirks change.
2383                  */
2384                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2385                         if (quirk_matches(id, &core_quirks[i]))
2386                                 ctrl->quirks |= core_quirks[i].quirks;
2387                 }
2388         }
2389
2390         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2391                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2392                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2393         }
2394
2395         ctrl->oacs = le16_to_cpu(id->oacs);
2396         ctrl->oncs = le16_to_cpup(&id->oncs);
2397         ctrl->oaes = le32_to_cpu(id->oaes);
2398         atomic_set(&ctrl->abort_limit, id->acl + 1);
2399         ctrl->vwc = id->vwc;
2400         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2401         if (id->mdts)
2402                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2403         else
2404                 max_hw_sectors = UINT_MAX;
2405         ctrl->max_hw_sectors =
2406                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2407
2408         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2409         ctrl->sgls = le32_to_cpu(id->sgls);
2410         ctrl->kas = le16_to_cpu(id->kas);
2411         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2412
2413         if (id->rtd3e) {
2414                 /* us -> s */
2415                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2416
2417                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2418                                                  shutdown_timeout, 60);
2419
2420                 if (ctrl->shutdown_timeout != shutdown_timeout)
2421                         dev_info(ctrl->device,
2422                                  "Shutdown timeout set to %u seconds\n",
2423                                  ctrl->shutdown_timeout);
2424         } else
2425                 ctrl->shutdown_timeout = shutdown_timeout;
2426
2427         ctrl->npss = id->npss;
2428         ctrl->apsta = id->apsta;
2429         prev_apst_enabled = ctrl->apst_enabled;
2430         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2431                 if (force_apst && id->apsta) {
2432                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2433                         ctrl->apst_enabled = true;
2434                 } else {
2435                         ctrl->apst_enabled = false;
2436                 }
2437         } else {
2438                 ctrl->apst_enabled = id->apsta;
2439         }
2440         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2441
2442         if (ctrl->ops->flags & NVME_F_FABRICS) {
2443                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2444                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2445                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2446                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2447
2448                 /*
2449                  * In fabrics we need to verify the cntlid matches the
2450                  * admin connect
2451                  */
2452                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2453                         ret = -EINVAL;
2454                         goto out_free;
2455                 }
2456
2457                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2458                         dev_err(ctrl->device,
2459                                 "keep-alive support is mandatory for fabrics\n");
2460                         ret = -EINVAL;
2461                         goto out_free;
2462                 }
2463         } else {
2464                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2465                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2466                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2467                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2468                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2469         }
2470
2471         ret = nvme_mpath_init(ctrl, id);
2472         kfree(id);
2473
2474         if (ret < 0)
2475                 return ret;
2476
2477         if (ctrl->apst_enabled && !prev_apst_enabled)
2478                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2479         else if (!ctrl->apst_enabled && prev_apst_enabled)
2480                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2481
2482         ret = nvme_configure_apst(ctrl);
2483         if (ret < 0)
2484                 return ret;
2485         
2486         ret = nvme_configure_timestamp(ctrl);
2487         if (ret < 0)
2488                 return ret;
2489
2490         ret = nvme_configure_directives(ctrl);
2491         if (ret < 0)
2492                 return ret;
2493
2494         ctrl->identified = true;
2495
2496         return 0;
2497
2498 out_free:
2499         kfree(id);
2500         return ret;
2501 }
2502 EXPORT_SYMBOL_GPL(nvme_init_identify);
2503
2504 static int nvme_dev_open(struct inode *inode, struct file *file)
2505 {
2506         struct nvme_ctrl *ctrl =
2507                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2508
2509         switch (ctrl->state) {
2510         case NVME_CTRL_LIVE:
2511         case NVME_CTRL_ADMIN_ONLY:
2512                 break;
2513         default:
2514                 return -EWOULDBLOCK;
2515         }
2516
2517         file->private_data = ctrl;
2518         return 0;
2519 }
2520
2521 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2522 {
2523         struct nvme_ns *ns;
2524         int ret;
2525
2526         down_read(&ctrl->namespaces_rwsem);
2527         if (list_empty(&ctrl->namespaces)) {
2528                 ret = -ENOTTY;
2529                 goto out_unlock;
2530         }
2531
2532         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2533         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2534                 dev_warn(ctrl->device,
2535                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2536                 ret = -EINVAL;
2537                 goto out_unlock;
2538         }
2539
2540         dev_warn(ctrl->device,
2541                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2542         kref_get(&ns->kref);
2543         up_read(&ctrl->namespaces_rwsem);
2544
2545         ret = nvme_user_cmd(ctrl, ns, argp);
2546         nvme_put_ns(ns);
2547         return ret;
2548
2549 out_unlock:
2550         up_read(&ctrl->namespaces_rwsem);
2551         return ret;
2552 }
2553
2554 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2555                 unsigned long arg)
2556 {
2557         struct nvme_ctrl *ctrl = file->private_data;
2558         void __user *argp = (void __user *)arg;
2559
2560         switch (cmd) {
2561         case NVME_IOCTL_ADMIN_CMD:
2562                 return nvme_user_cmd(ctrl, NULL, argp);
2563         case NVME_IOCTL_IO_CMD:
2564                 return nvme_dev_user_cmd(ctrl, argp);
2565         case NVME_IOCTL_RESET:
2566                 dev_warn(ctrl->device, "resetting controller\n");
2567                 return nvme_reset_ctrl_sync(ctrl);
2568         case NVME_IOCTL_SUBSYS_RESET:
2569                 return nvme_reset_subsystem(ctrl);
2570         case NVME_IOCTL_RESCAN:
2571                 nvme_queue_scan(ctrl);
2572                 return 0;
2573         default:
2574                 return -ENOTTY;
2575         }
2576 }
2577
2578 static const struct file_operations nvme_dev_fops = {
2579         .owner          = THIS_MODULE,
2580         .open           = nvme_dev_open,
2581         .unlocked_ioctl = nvme_dev_ioctl,
2582         .compat_ioctl   = nvme_dev_ioctl,
2583 };
2584
2585 static ssize_t nvme_sysfs_reset(struct device *dev,
2586                                 struct device_attribute *attr, const char *buf,
2587                                 size_t count)
2588 {
2589         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2590         int ret;
2591
2592         ret = nvme_reset_ctrl_sync(ctrl);
2593         if (ret < 0)
2594                 return ret;
2595         return count;
2596 }
2597 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2598
2599 static ssize_t nvme_sysfs_rescan(struct device *dev,
2600                                 struct device_attribute *attr, const char *buf,
2601                                 size_t count)
2602 {
2603         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2604
2605         nvme_queue_scan(ctrl);
2606         return count;
2607 }
2608 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2609
2610 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2611 {
2612         struct gendisk *disk = dev_to_disk(dev);
2613
2614         if (disk->fops == &nvme_fops)
2615                 return nvme_get_ns_from_dev(dev)->head;
2616         else
2617                 return disk->private_data;
2618 }
2619
2620 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2621                 char *buf)
2622 {
2623         struct nvme_ns_head *head = dev_to_ns_head(dev);
2624         struct nvme_ns_ids *ids = &head->ids;
2625         struct nvme_subsystem *subsys = head->subsys;
2626         int serial_len = sizeof(subsys->serial);
2627         int model_len = sizeof(subsys->model);
2628
2629         if (!uuid_is_null(&ids->uuid))
2630                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2631
2632         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2633                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2634
2635         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2636                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2637
2638         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2639                                   subsys->serial[serial_len - 1] == '\0'))
2640                 serial_len--;
2641         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2642                                  subsys->model[model_len - 1] == '\0'))
2643                 model_len--;
2644
2645         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2646                 serial_len, subsys->serial, model_len, subsys->model,
2647                 head->ns_id);
2648 }
2649 static DEVICE_ATTR_RO(wwid);
2650
2651 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2652                 char *buf)
2653 {
2654         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2655 }
2656 static DEVICE_ATTR_RO(nguid);
2657
2658 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2659                 char *buf)
2660 {
2661         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2662
2663         /* For backward compatibility expose the NGUID to userspace if
2664          * we have no UUID set
2665          */
2666         if (uuid_is_null(&ids->uuid)) {
2667                 printk_ratelimited(KERN_WARNING
2668                                    "No UUID available providing old NGUID\n");
2669                 return sprintf(buf, "%pU\n", ids->nguid);
2670         }
2671         return sprintf(buf, "%pU\n", &ids->uuid);
2672 }
2673 static DEVICE_ATTR_RO(uuid);
2674
2675 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2676                 char *buf)
2677 {
2678         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2679 }
2680 static DEVICE_ATTR_RO(eui);
2681
2682 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2683                 char *buf)
2684 {
2685         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2686 }
2687 static DEVICE_ATTR_RO(nsid);
2688
2689 static struct attribute *nvme_ns_id_attrs[] = {
2690         &dev_attr_wwid.attr,
2691         &dev_attr_uuid.attr,
2692         &dev_attr_nguid.attr,
2693         &dev_attr_eui.attr,
2694         &dev_attr_nsid.attr,
2695 #ifdef CONFIG_NVME_MULTIPATH
2696         &dev_attr_ana_grpid.attr,
2697         &dev_attr_ana_state.attr,
2698 #endif
2699         NULL,
2700 };
2701
2702 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2703                 struct attribute *a, int n)
2704 {
2705         struct device *dev = container_of(kobj, struct device, kobj);
2706         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2707
2708         if (a == &dev_attr_uuid.attr) {
2709                 if (uuid_is_null(&ids->uuid) &&
2710                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2711                         return 0;
2712         }
2713         if (a == &dev_attr_nguid.attr) {
2714                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2715                         return 0;
2716         }
2717         if (a == &dev_attr_eui.attr) {
2718                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2719                         return 0;
2720         }
2721 #ifdef CONFIG_NVME_MULTIPATH
2722         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2723                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2724                         return 0;
2725                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2726                         return 0;
2727         }
2728 #endif
2729         return a->mode;
2730 }
2731
2732 static const struct attribute_group nvme_ns_id_attr_group = {
2733         .attrs          = nvme_ns_id_attrs,
2734         .is_visible     = nvme_ns_id_attrs_are_visible,
2735 };
2736
2737 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2738         &nvme_ns_id_attr_group,
2739 #ifdef CONFIG_NVM
2740         &nvme_nvm_attr_group,
2741 #endif
2742         NULL,
2743 };
2744
2745 #define nvme_show_str_function(field)                                           \
2746 static ssize_t  field##_show(struct device *dev,                                \
2747                             struct device_attribute *attr, char *buf)           \
2748 {                                                                               \
2749         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2750         return sprintf(buf, "%.*s\n",                                           \
2751                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2752 }                                                                               \
2753 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2754
2755 nvme_show_str_function(model);
2756 nvme_show_str_function(serial);
2757 nvme_show_str_function(firmware_rev);
2758
2759 #define nvme_show_int_function(field)                                           \
2760 static ssize_t  field##_show(struct device *dev,                                \
2761                             struct device_attribute *attr, char *buf)           \
2762 {                                                                               \
2763         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2764         return sprintf(buf, "%d\n", ctrl->field);       \
2765 }                                                                               \
2766 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2767
2768 nvme_show_int_function(cntlid);
2769
2770 static ssize_t nvme_sysfs_delete(struct device *dev,
2771                                 struct device_attribute *attr, const char *buf,
2772                                 size_t count)
2773 {
2774         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2775
2776         if (device_remove_file_self(dev, attr))
2777                 nvme_delete_ctrl_sync(ctrl);
2778         return count;
2779 }
2780 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2781
2782 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2783                                          struct device_attribute *attr,
2784                                          char *buf)
2785 {
2786         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2787
2788         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2789 }
2790 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2791
2792 static ssize_t nvme_sysfs_show_state(struct device *dev,
2793                                      struct device_attribute *attr,
2794                                      char *buf)
2795 {
2796         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2797         static const char *const state_name[] = {
2798                 [NVME_CTRL_NEW]         = "new",
2799                 [NVME_CTRL_LIVE]        = "live",
2800                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2801                 [NVME_CTRL_RESETTING]   = "resetting",
2802                 [NVME_CTRL_CONNECTING]  = "connecting",
2803                 [NVME_CTRL_DELETING]    = "deleting",
2804                 [NVME_CTRL_DEAD]        = "dead",
2805         };
2806
2807         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2808             state_name[ctrl->state])
2809                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2810
2811         return sprintf(buf, "unknown state\n");
2812 }
2813
2814 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2815
2816 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2817                                          struct device_attribute *attr,
2818                                          char *buf)
2819 {
2820         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2821
2822         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2823 }
2824 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2825
2826 static ssize_t nvme_sysfs_show_address(struct device *dev,
2827                                          struct device_attribute *attr,
2828                                          char *buf)
2829 {
2830         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2831
2832         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2833 }
2834 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2835
2836 static struct attribute *nvme_dev_attrs[] = {
2837         &dev_attr_reset_controller.attr,
2838         &dev_attr_rescan_controller.attr,
2839         &dev_attr_model.attr,
2840         &dev_attr_serial.attr,
2841         &dev_attr_firmware_rev.attr,
2842         &dev_attr_cntlid.attr,
2843         &dev_attr_delete_controller.attr,
2844         &dev_attr_transport.attr,
2845         &dev_attr_subsysnqn.attr,
2846         &dev_attr_address.attr,
2847         &dev_attr_state.attr,
2848         NULL
2849 };
2850
2851 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2852                 struct attribute *a, int n)
2853 {
2854         struct device *dev = container_of(kobj, struct device, kobj);
2855         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2856
2857         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2858                 return 0;
2859         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2860                 return 0;
2861
2862         return a->mode;
2863 }
2864
2865 static struct attribute_group nvme_dev_attrs_group = {
2866         .attrs          = nvme_dev_attrs,
2867         .is_visible     = nvme_dev_attrs_are_visible,
2868 };
2869
2870 static const struct attribute_group *nvme_dev_attr_groups[] = {
2871         &nvme_dev_attrs_group,
2872         NULL,
2873 };
2874
2875 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2876                 unsigned nsid)
2877 {
2878         struct nvme_ns_head *h;
2879
2880         lockdep_assert_held(&subsys->lock);
2881
2882         list_for_each_entry(h, &subsys->nsheads, entry) {
2883                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2884                         return h;
2885         }
2886
2887         return NULL;
2888 }
2889
2890 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2891                 struct nvme_ns_head *new)
2892 {
2893         struct nvme_ns_head *h;
2894
2895         lockdep_assert_held(&subsys->lock);
2896
2897         list_for_each_entry(h, &subsys->nsheads, entry) {
2898                 if (nvme_ns_ids_valid(&new->ids) &&
2899                     !list_empty(&h->list) &&
2900                     nvme_ns_ids_equal(&new->ids, &h->ids))
2901                         return -EINVAL;
2902         }
2903
2904         return 0;
2905 }
2906
2907 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2908                 unsigned nsid, struct nvme_id_ns *id)
2909 {
2910         struct nvme_ns_head *head;
2911         size_t size = sizeof(*head);
2912         int ret = -ENOMEM;
2913
2914 #ifdef CONFIG_NVME_MULTIPATH
2915         size += num_possible_nodes() * sizeof(struct nvme_ns *);
2916 #endif
2917
2918         head = kzalloc(size, GFP_KERNEL);
2919         if (!head)
2920                 goto out;
2921         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2922         if (ret < 0)
2923                 goto out_free_head;
2924         head->instance = ret;
2925         INIT_LIST_HEAD(&head->list);
2926         ret = init_srcu_struct(&head->srcu);
2927         if (ret)
2928                 goto out_ida_remove;
2929         head->subsys = ctrl->subsys;
2930         head->ns_id = nsid;
2931         kref_init(&head->ref);
2932
2933         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2934
2935         ret = __nvme_check_ids(ctrl->subsys, head);
2936         if (ret) {
2937                 dev_err(ctrl->device,
2938                         "duplicate IDs for nsid %d\n", nsid);
2939                 goto out_cleanup_srcu;
2940         }
2941
2942         ret = nvme_mpath_alloc_disk(ctrl, head);
2943         if (ret)
2944                 goto out_cleanup_srcu;
2945
2946         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2947
2948         kref_get(&ctrl->subsys->ref);
2949
2950         return head;
2951 out_cleanup_srcu:
2952         cleanup_srcu_struct(&head->srcu);
2953 out_ida_remove:
2954         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2955 out_free_head:
2956         kfree(head);
2957 out:
2958         return ERR_PTR(ret);
2959 }
2960
2961 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2962                 struct nvme_id_ns *id)
2963 {
2964         struct nvme_ctrl *ctrl = ns->ctrl;
2965         bool is_shared = id->nmic & (1 << 0);
2966         struct nvme_ns_head *head = NULL;
2967         int ret = 0;
2968
2969         mutex_lock(&ctrl->subsys->lock);
2970         if (is_shared)
2971                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2972         if (!head) {
2973                 head = nvme_alloc_ns_head(ctrl, nsid, id);
2974                 if (IS_ERR(head)) {
2975                         ret = PTR_ERR(head);
2976                         goto out_unlock;
2977                 }
2978         } else {
2979                 struct nvme_ns_ids ids;
2980
2981                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2982                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2983                         dev_err(ctrl->device,
2984                                 "IDs don't match for shared namespace %d\n",
2985                                         nsid);
2986                         ret = -EINVAL;
2987                         goto out_unlock;
2988                 }
2989         }
2990
2991         list_add_tail(&ns->siblings, &head->list);
2992         ns->head = head;
2993
2994 out_unlock:
2995         mutex_unlock(&ctrl->subsys->lock);
2996         return ret;
2997 }
2998
2999 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3000 {
3001         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3002         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3003
3004         return nsa->head->ns_id - nsb->head->ns_id;
3005 }
3006
3007 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3008 {
3009         struct nvme_ns *ns, *ret = NULL;
3010
3011         down_read(&ctrl->namespaces_rwsem);
3012         list_for_each_entry(ns, &ctrl->namespaces, list) {
3013                 if (ns->head->ns_id == nsid) {
3014                         if (!kref_get_unless_zero(&ns->kref))
3015                                 continue;
3016                         ret = ns;
3017                         break;
3018                 }
3019                 if (ns->head->ns_id > nsid)
3020                         break;
3021         }
3022         up_read(&ctrl->namespaces_rwsem);
3023         return ret;
3024 }
3025
3026 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3027 {
3028         struct streams_directive_params s;
3029         int ret;
3030
3031         if (!ctrl->nr_streams)
3032                 return 0;
3033
3034         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3035         if (ret)
3036                 return ret;
3037
3038         ns->sws = le32_to_cpu(s.sws);
3039         ns->sgs = le16_to_cpu(s.sgs);
3040
3041         if (ns->sws) {
3042                 unsigned int bs = 1 << ns->lba_shift;
3043
3044                 blk_queue_io_min(ns->queue, bs * ns->sws);
3045                 if (ns->sgs)
3046                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3047         }
3048
3049         return 0;
3050 }
3051
3052 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3053 {
3054         struct nvme_ns *ns;
3055         struct gendisk *disk;
3056         struct nvme_id_ns *id;
3057         char disk_name[DISK_NAME_LEN];
3058         int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
3059
3060         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3061         if (!ns)
3062                 return;
3063
3064         ns->queue = blk_mq_init_queue(ctrl->tagset);
3065         if (IS_ERR(ns->queue))
3066                 goto out_free_ns;
3067         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3068         ns->queue->queuedata = ns;
3069         ns->ctrl = ctrl;
3070
3071         kref_init(&ns->kref);
3072         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3073
3074         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3075         nvme_set_queue_limits(ctrl, ns->queue);
3076
3077         id = nvme_identify_ns(ctrl, nsid);
3078         if (!id)
3079                 goto out_free_queue;
3080
3081         if (id->ncap == 0)
3082                 goto out_free_id;
3083
3084         if (nvme_init_ns_head(ns, nsid, id))
3085                 goto out_free_id;
3086         nvme_setup_streams_ns(ctrl, ns);
3087         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3088
3089         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3090                 if (nvme_nvm_register(ns, disk_name, node)) {
3091                         dev_warn(ctrl->device, "LightNVM init failure\n");
3092                         goto out_unlink_ns;
3093                 }
3094         }
3095
3096         disk = alloc_disk_node(0, node);
3097         if (!disk)
3098                 goto out_unlink_ns;
3099
3100         disk->fops = &nvme_fops;
3101         disk->private_data = ns;
3102         disk->queue = ns->queue;
3103         disk->flags = flags;
3104         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3105         ns->disk = disk;
3106
3107         __nvme_revalidate_disk(disk, id);
3108
3109         down_write(&ctrl->namespaces_rwsem);
3110         list_add_tail(&ns->list, &ctrl->namespaces);
3111         up_write(&ctrl->namespaces_rwsem);
3112
3113         nvme_get_ctrl(ctrl);
3114
3115         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3116
3117         nvme_mpath_add_disk(ns, id);
3118         nvme_fault_inject_init(ns);
3119         kfree(id);
3120
3121         return;
3122  out_unlink_ns:
3123         mutex_lock(&ctrl->subsys->lock);
3124         list_del_rcu(&ns->siblings);
3125         mutex_unlock(&ctrl->subsys->lock);
3126  out_free_id:
3127         kfree(id);
3128  out_free_queue:
3129         blk_cleanup_queue(ns->queue);
3130  out_free_ns:
3131         kfree(ns);
3132 }
3133
3134 static void nvme_ns_remove(struct nvme_ns *ns)
3135 {
3136         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3137                 return;
3138
3139         nvme_fault_inject_fini(ns);
3140         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3141                 del_gendisk(ns->disk);
3142                 blk_cleanup_queue(ns->queue);
3143                 if (blk_get_integrity(ns->disk))
3144                         blk_integrity_unregister(ns->disk);
3145         }
3146
3147         mutex_lock(&ns->ctrl->subsys->lock);
3148         list_del_rcu(&ns->siblings);
3149         nvme_mpath_clear_current_path(ns);
3150         mutex_unlock(&ns->ctrl->subsys->lock);
3151
3152         down_write(&ns->ctrl->namespaces_rwsem);
3153         list_del_init(&ns->list);
3154         up_write(&ns->ctrl->namespaces_rwsem);
3155
3156         synchronize_srcu(&ns->head->srcu);
3157         nvme_mpath_check_last_path(ns);
3158         nvme_put_ns(ns);
3159 }
3160
3161 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3162 {
3163         struct nvme_ns *ns;
3164
3165         ns = nvme_find_get_ns(ctrl, nsid);
3166         if (ns) {
3167                 if (ns->disk && revalidate_disk(ns->disk))
3168                         nvme_ns_remove(ns);
3169                 nvme_put_ns(ns);
3170         } else
3171                 nvme_alloc_ns(ctrl, nsid);
3172 }
3173
3174 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3175                                         unsigned nsid)
3176 {
3177         struct nvme_ns *ns, *next;
3178         LIST_HEAD(rm_list);
3179
3180         down_write(&ctrl->namespaces_rwsem);
3181         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3182                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3183                         list_move_tail(&ns->list, &rm_list);
3184         }
3185         up_write(&ctrl->namespaces_rwsem);
3186
3187         list_for_each_entry_safe(ns, next, &rm_list, list)
3188                 nvme_ns_remove(ns);
3189
3190 }
3191
3192 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3193 {
3194         struct nvme_ns *ns;
3195         __le32 *ns_list;
3196         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3197         int ret = 0;
3198
3199         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3200         if (!ns_list)
3201                 return -ENOMEM;
3202
3203         for (i = 0; i < num_lists; i++) {
3204                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3205                 if (ret)
3206                         goto free;
3207
3208                 for (j = 0; j < min(nn, 1024U); j++) {
3209                         nsid = le32_to_cpu(ns_list[j]);
3210                         if (!nsid)
3211                                 goto out;
3212
3213                         nvme_validate_ns(ctrl, nsid);
3214
3215                         while (++prev < nsid) {
3216                                 ns = nvme_find_get_ns(ctrl, prev);
3217                                 if (ns) {
3218                                         nvme_ns_remove(ns);
3219                                         nvme_put_ns(ns);
3220                                 }
3221                         }
3222                 }
3223                 nn -= j;
3224         }
3225  out:
3226         nvme_remove_invalid_namespaces(ctrl, prev);
3227  free:
3228         kfree(ns_list);
3229         return ret;
3230 }
3231
3232 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3233 {
3234         unsigned i;
3235
3236         for (i = 1; i <= nn; i++)
3237                 nvme_validate_ns(ctrl, i);
3238
3239         nvme_remove_invalid_namespaces(ctrl, nn);
3240 }
3241
3242 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3243 {
3244         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3245         __le32 *log;
3246         int error;
3247
3248         log = kzalloc(log_size, GFP_KERNEL);
3249         if (!log)
3250                 return;
3251
3252         /*
3253          * We need to read the log to clear the AEN, but we don't want to rely
3254          * on it for the changed namespace information as userspace could have
3255          * raced with us in reading the log page, which could cause us to miss
3256          * updates.
3257          */
3258         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3259                         log_size, 0);
3260         if (error)
3261                 dev_warn(ctrl->device,
3262                         "reading changed ns log failed: %d\n", error);
3263
3264         kfree(log);
3265 }
3266
3267 static void nvme_scan_work(struct work_struct *work)
3268 {
3269         struct nvme_ctrl *ctrl =
3270                 container_of(work, struct nvme_ctrl, scan_work);
3271         struct nvme_id_ctrl *id;
3272         unsigned nn;
3273
3274         if (ctrl->state != NVME_CTRL_LIVE)
3275                 return;
3276
3277         WARN_ON_ONCE(!ctrl->tagset);
3278
3279         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3280                 dev_info(ctrl->device, "rescanning namespaces.\n");
3281                 nvme_clear_changed_ns_log(ctrl);
3282         }
3283
3284         if (nvme_identify_ctrl(ctrl, &id))
3285                 return;
3286
3287         nn = le32_to_cpu(id->nn);
3288         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3289             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3290                 if (!nvme_scan_ns_list(ctrl, nn))
3291                         goto out_free_id;
3292         }
3293         nvme_scan_ns_sequential(ctrl, nn);
3294 out_free_id:
3295         kfree(id);
3296         down_write(&ctrl->namespaces_rwsem);
3297         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3298         up_write(&ctrl->namespaces_rwsem);
3299 }
3300
3301 /*
3302  * This function iterates the namespace list unlocked to allow recovery from
3303  * controller failure. It is up to the caller to ensure the namespace list is
3304  * not modified by scan work while this function is executing.
3305  */
3306 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3307 {
3308         struct nvme_ns *ns, *next;
3309         LIST_HEAD(ns_list);
3310
3311         /*
3312          * The dead states indicates the controller was not gracefully
3313          * disconnected. In that case, we won't be able to flush any data while
3314          * removing the namespaces' disks; fail all the queues now to avoid
3315          * potentially having to clean up the failed sync later.
3316          */
3317         if (ctrl->state == NVME_CTRL_DEAD)
3318                 nvme_kill_queues(ctrl);
3319
3320         down_write(&ctrl->namespaces_rwsem);
3321         list_splice_init(&ctrl->namespaces, &ns_list);
3322         up_write(&ctrl->namespaces_rwsem);
3323
3324         list_for_each_entry_safe(ns, next, &ns_list, list)
3325                 nvme_ns_remove(ns);
3326 }
3327 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3328
3329 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3330 {
3331         char *envp[2] = { NULL, NULL };
3332         u32 aen_result = ctrl->aen_result;
3333
3334         ctrl->aen_result = 0;
3335         if (!aen_result)
3336                 return;
3337
3338         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3339         if (!envp[0])
3340                 return;
3341         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3342         kfree(envp[0]);
3343 }
3344
3345 static void nvme_async_event_work(struct work_struct *work)
3346 {
3347         struct nvme_ctrl *ctrl =
3348                 container_of(work, struct nvme_ctrl, async_event_work);
3349
3350         nvme_aen_uevent(ctrl);
3351         ctrl->ops->submit_async_event(ctrl);
3352 }
3353
3354 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3355 {
3356
3357         u32 csts;
3358
3359         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3360                 return false;
3361
3362         if (csts == ~0)
3363                 return false;
3364
3365         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3366 }
3367
3368 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3369 {
3370         struct nvme_fw_slot_info_log *log;
3371
3372         log = kmalloc(sizeof(*log), GFP_KERNEL);
3373         if (!log)
3374                 return;
3375
3376         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3377                         sizeof(*log), 0))
3378                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3379         kfree(log);
3380 }
3381
3382 static void nvme_fw_act_work(struct work_struct *work)
3383 {
3384         struct nvme_ctrl *ctrl = container_of(work,
3385                                 struct nvme_ctrl, fw_act_work);
3386         unsigned long fw_act_timeout;
3387
3388         if (ctrl->mtfa)
3389                 fw_act_timeout = jiffies +
3390                                 msecs_to_jiffies(ctrl->mtfa * 100);
3391         else
3392                 fw_act_timeout = jiffies +
3393                                 msecs_to_jiffies(admin_timeout * 1000);
3394
3395         nvme_stop_queues(ctrl);
3396         while (nvme_ctrl_pp_status(ctrl)) {
3397                 if (time_after(jiffies, fw_act_timeout)) {
3398                         dev_warn(ctrl->device,
3399                                 "Fw activation timeout, reset controller\n");
3400                         nvme_reset_ctrl(ctrl);
3401                         break;
3402                 }
3403                 msleep(100);
3404         }
3405
3406         if (ctrl->state != NVME_CTRL_LIVE)
3407                 return;
3408
3409         nvme_start_queues(ctrl);
3410         /* read FW slot information to clear the AER */
3411         nvme_get_fw_slot_info(ctrl);
3412 }
3413
3414 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3415 {
3416         u32 aer_notice_type = (result & 0xff00) >> 8;
3417
3418         switch (aer_notice_type) {
3419         case NVME_AER_NOTICE_NS_CHANGED:
3420                 trace_nvme_async_event(ctrl, aer_notice_type);
3421                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3422                 nvme_queue_scan(ctrl);
3423                 break;
3424         case NVME_AER_NOTICE_FW_ACT_STARTING:
3425                 trace_nvme_async_event(ctrl, aer_notice_type);
3426                 queue_work(nvme_wq, &ctrl->fw_act_work);
3427                 break;
3428 #ifdef CONFIG_NVME_MULTIPATH
3429         case NVME_AER_NOTICE_ANA:
3430                 trace_nvme_async_event(ctrl, aer_notice_type);
3431                 if (!ctrl->ana_log_buf)
3432                         break;
3433                 queue_work(nvme_wq, &ctrl->ana_work);
3434                 break;
3435 #endif
3436         default:
3437                 dev_warn(ctrl->device, "async event result %08x\n", result);
3438         }
3439 }
3440
3441 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3442                 volatile union nvme_result *res)
3443 {
3444         u32 result = le32_to_cpu(res->u32);
3445         u32 aer_type = result & 0x07;
3446
3447         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3448                 return;
3449
3450         switch (aer_type) {
3451         case NVME_AER_NOTICE:
3452                 nvme_handle_aen_notice(ctrl, result);
3453                 break;
3454         case NVME_AER_ERROR:
3455         case NVME_AER_SMART:
3456         case NVME_AER_CSS:
3457         case NVME_AER_VS:
3458                 trace_nvme_async_event(ctrl, aer_type);
3459                 ctrl->aen_result = result;
3460                 break;
3461         default:
3462                 break;
3463         }
3464         queue_work(nvme_wq, &ctrl->async_event_work);
3465 }
3466 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3467
3468 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3469 {
3470         nvme_mpath_stop(ctrl);
3471         nvme_stop_keep_alive(ctrl);
3472         flush_work(&ctrl->async_event_work);
3473         flush_work(&ctrl->scan_work);
3474         cancel_work_sync(&ctrl->fw_act_work);
3475       &nb