?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dm writecache: add cond_resched to loop in persistent_memory_claim()
[linux/fpc-iii.git] / drivers / nvme / target / core.c
blob96deaf3484662de4528b307e0c3c5859def33be3
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Common code for the NVMe target.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5 */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 #include <linux/module.h>
8 #include <linux/random.h>
9 #include <linux/rculist.h>
10 #include <linux/pci-p2pdma.h>
11 #include <linux/scatterlist.h>
12
13 #define CREATE_TRACE_POINTS
14 #include "trace.h"
15
16 #include "nvmet.h"
17
18 struct workqueue_struct *buffered_io_wq;
19 static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
20 static DEFINE_IDA(cntlid_ida);
21
22 /*
23 * This read/write semaphore is used to synchronize access to configuration
24 * information on a target system that will result in discovery log page
25 * information change for at least one host.
26 * The full list of resources to protected by this semaphore is:
27 *
28 * - subsystems list
29 * - per-subsystem allowed hosts list
30 * - allow_any_host subsystem attribute
31 * - nvmet_genctr
32 * - the nvmet_transports array
33 *
34 * When updating any of those lists/structures write lock should be obtained,
35 * while when reading (popolating discovery log page or checking host-subsystem
36 * link) read lock is obtained to allow concurrent reads.
37 */
38 DECLARE_RWSEM(nvmet_config_sem);
39
40 u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
41 u64 nvmet_ana_chgcnt;
42 DECLARE_RWSEM(nvmet_ana_sem);
43
44 inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno)
45 {
46 u16 status;
47
48 switch (errno) {
49 case 0:
50 status = NVME_SC_SUCCESS;
51 break;
52 case -ENOSPC:
53 req->error_loc = offsetof(struct nvme_rw_command, length);
54 status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
55 break;
56 case -EREMOTEIO:
57 req->error_loc = offsetof(struct nvme_rw_command, slba);
58 status = NVME_SC_LBA_RANGE | NVME_SC_DNR;
59 break;
60 case -EOPNOTSUPP:
61 req->error_loc = offsetof(struct nvme_common_command, opcode);
62 switch (req->cmd->common.opcode) {
63 case nvme_cmd_dsm:
64 case nvme_cmd_write_zeroes:
65 status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
66 break;
67 default:
68 status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
69 }
70 break;
71 case -ENODATA:
72 req->error_loc = offsetof(struct nvme_rw_command, nsid);
73 status = NVME_SC_ACCESS_DENIED;
74 break;
75 case -EIO:
76 /* FALLTHRU */
77 default:
78 req->error_loc = offsetof(struct nvme_common_command, opcode);
79 status = NVME_SC_INTERNAL | NVME_SC_DNR;
80 }
81
82 return status;
83 }
84
85 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
86 const char *subsysnqn);
87
88 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
89 size_t len)
90 {
91 if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
92 req->error_loc = offsetof(struct nvme_common_command, dptr);
93 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
94 }
95 return 0;
96 }
97
98 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len)
99 {
100 if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
101 req->error_loc = offsetof(struct nvme_common_command, dptr);
102 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
103 }
104 return 0;
105 }
106
107 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len)
108 {
109 if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len) {
110 req->error_loc = offsetof(struct nvme_common_command, dptr);
111 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
112 }
113 return 0;
114 }
115
116 static unsigned int nvmet_max_nsid(struct nvmet_subsys *subsys)
117 {
118 struct nvmet_ns *ns;
119
120 if (list_empty(&subsys->namespaces))
121 return 0;
122
123 ns = list_last_entry(&subsys->namespaces, struct nvmet_ns, dev_link);
124 return ns->nsid;
125 }
126
127 static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
128 {
129 return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16);
130 }
131
132 static void nvmet_async_events_failall(struct nvmet_ctrl *ctrl)
133 {
134 u16 status = NVME_SC_INTERNAL | NVME_SC_DNR;
135 struct nvmet_req *req;
136
137 mutex_lock(&ctrl->lock);
138 while (ctrl->nr_async_event_cmds) {
139 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
140 mutex_unlock(&ctrl->lock);
141 nvmet_req_complete(req, status);
142 mutex_lock(&ctrl->lock);
143 }
144 mutex_unlock(&ctrl->lock);
145 }
146
147 static void nvmet_async_events_process(struct nvmet_ctrl *ctrl)
148 {
149 struct nvmet_async_event *aen;
150 struct nvmet_req *req;
151
152 mutex_lock(&ctrl->lock);
153 while (ctrl->nr_async_event_cmds && !list_empty(&ctrl->async_events)) {
154 aen = list_first_entry(&ctrl->async_events,
155 struct nvmet_async_event, entry);
156 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
157 nvmet_set_result(req, nvmet_async_event_result(aen));
158
159 list_del(&aen->entry);
160 kfree(aen);
161
162 mutex_unlock(&ctrl->lock);
163 nvmet_req_complete(req, 0);
164 mutex_lock(&ctrl->lock);
165 }
166 mutex_unlock(&ctrl->lock);
167 }
168
169 static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
170 {
171 struct nvmet_async_event *aen, *tmp;
172
173 mutex_lock(&ctrl->lock);
174 list_for_each_entry_safe(aen, tmp, &ctrl->async_events, entry) {
175 list_del(&aen->entry);
176 kfree(aen);
177 }
178 mutex_unlock(&ctrl->lock);
179 }
180
181 static void nvmet_async_event_work(struct work_struct *work)
182 {
183 struct nvmet_ctrl *ctrl =
184 container_of(work, struct nvmet_ctrl, async_event_work);
185
186 nvmet_async_events_process(ctrl);
187 }
188
189 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
190 u8 event_info, u8 log_page)
191 {
192 struct nvmet_async_event *aen;
193
194 aen = kmalloc(sizeof(*aen), GFP_KERNEL);
195 if (!aen)
196 return;
197
198 aen->event_type = event_type;
199 aen->event_info = event_info;
200 aen->log_page = log_page;
201
202 mutex_lock(&ctrl->lock);
203 list_add_tail(&aen->entry, &ctrl->async_events);
204 mutex_unlock(&ctrl->lock);
205
206 schedule_work(&ctrl->async_event_work);
207 }
208
209 static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
210 {
211 u32 i;
212
213 mutex_lock(&ctrl->lock);
214 if (ctrl->nr_changed_ns > NVME_MAX_CHANGED_NAMESPACES)
215 goto out_unlock;
216
217 for (i = 0; i < ctrl->nr_changed_ns; i++) {
218 if (ctrl->changed_ns_list[i] == nsid)
219 goto out_unlock;
220 }
221
222 if (ctrl->nr_changed_ns == NVME_MAX_CHANGED_NAMESPACES) {
223 ctrl->changed_ns_list[0] = cpu_to_le32(0xffffffff);
224 ctrl->nr_changed_ns = U32_MAX;
225 goto out_unlock;
226 }
227
228 ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid;
229 out_unlock:
230 mutex_unlock(&ctrl->lock);
231 }
232
233 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid)
234 {
235 struct nvmet_ctrl *ctrl;
236
237 lockdep_assert_held(&subsys->lock);
238
239 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
240 nvmet_add_to_changed_ns_log(ctrl, cpu_to_le32(nsid));
241 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_NS_ATTR))
242 continue;
243 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
244 NVME_AER_NOTICE_NS_CHANGED,
245 NVME_LOG_CHANGED_NS);
246 }
247 }
248
249 void nvmet_send_ana_event(struct nvmet_subsys *subsys,
250 struct nvmet_port *port)
251 {
252 struct nvmet_ctrl *ctrl;
253
254 mutex_lock(&subsys->lock);
255 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
256 if (port && ctrl->port != port)
257 continue;
258 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE))
259 continue;
260 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
261 NVME_AER_NOTICE_ANA, NVME_LOG_ANA);
262 }
263 mutex_unlock(&subsys->lock);
264 }
265
266 void nvmet_port_send_ana_event(struct nvmet_port *port)
267 {
268 struct nvmet_subsys_link *p;
269
270 down_read(&nvmet_config_sem);
271 list_for_each_entry(p, &port->subsystems, entry)
272 nvmet_send_ana_event(p->subsys, port);
273 up_read(&nvmet_config_sem);
274 }
275
276 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops)
277 {
278 int ret = 0;
279
280 down_write(&nvmet_config_sem);
281 if (nvmet_transports[ops->type])
282 ret = -EINVAL;
283 else
284 nvmet_transports[ops->type] = ops;
285 up_write(&nvmet_config_sem);
286
287 return ret;
288 }
289 EXPORT_SYMBOL_GPL(nvmet_register_transport);
290
291 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops)
292 {
293 down_write(&nvmet_config_sem);
294 nvmet_transports[ops->type] = NULL;
295 up_write(&nvmet_config_sem);
296 }
297 EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
298
299 void nvmet_port_del_ctrls(struct nvmet_port *port, struct nvmet_subsys *subsys)
300 {
301 struct nvmet_ctrl *ctrl;
302
303 mutex_lock(&subsys->lock);
304 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
305 if (ctrl->port == port)
306 ctrl->ops->delete_ctrl(ctrl);
307 }
308 mutex_unlock(&subsys->lock);
309 }
310
311 int nvmet_enable_port(struct nvmet_port *port)
312 {
313 const struct nvmet_fabrics_ops *ops;
314 int ret;
315
316 lockdep_assert_held(&nvmet_config_sem);
317
318 ops = nvmet_transports[port->disc_addr.trtype];
319 if (!ops) {
320 up_write(&nvmet_config_sem);
321 request_module("nvmet-transport-%d", port->disc_addr.trtype);
322 down_write(&nvmet_config_sem);
323 ops = nvmet_transports[port->disc_addr.trtype];
324 if (!ops) {
325 pr_err("transport type %d not supported\n",
326 port->disc_addr.trtype);
327 return -EINVAL;
328 }
329 }
330
331 if (!try_module_get(ops->owner))
332 return -EINVAL;
333
334 ret = ops->add_port(port);
335 if (ret) {
336 module_put(ops->owner);
337 return ret;
338 }
339
340 /* If the transport didn't set inline_data_size, then disable it. */
341 if (port->inline_data_size < 0)
342 port->inline_data_size = 0;
343
344 port->enabled = true;
345 port->tr_ops = ops;
346 return 0;
347 }
348
349 void nvmet_disable_port(struct nvmet_port *port)
350 {
351 const struct nvmet_fabrics_ops *ops;
352
353 lockdep_assert_held(&nvmet_config_sem);
354
355 port->enabled = false;
356 port->tr_ops = NULL;
357
358 ops = nvmet_transports[port->disc_addr.trtype];
359 ops->remove_port(port);
360 module_put(ops->owner);
361 }
362
363 static void nvmet_keep_alive_timer(struct work_struct *work)
364 {
365 struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
366 struct nvmet_ctrl, ka_work);
367 bool cmd_seen = ctrl->cmd_seen;
368
369 ctrl->cmd_seen = false;
370 if (cmd_seen) {
371 pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
372 ctrl->cntlid);
373 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
374 return;
375 }
376
377 pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
378 ctrl->cntlid, ctrl->kato);
379
380 nvmet_ctrl_fatal_error(ctrl);
381 }
382
383 static void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
384 {
385 pr_debug("ctrl %d start keep-alive timer for %d secs\n",
386 ctrl->cntlid, ctrl->kato);
387
388 INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer);
389 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
390 }
391
392 static void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
393 {
394 pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid);
395
396 cancel_delayed_work_sync(&ctrl->ka_work);
397 }
398
399 static struct nvmet_ns *__nvmet_find_namespace(struct nvmet_ctrl *ctrl,
400 __le32 nsid)
401 {
402 struct nvmet_ns *ns;
403
404 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
405 if (ns->nsid == le32_to_cpu(nsid))
406 return ns;
407 }
408
409 return NULL;
410 }
411
412 struct nvmet_ns *nvmet_find_namespace(struct nvmet_ctrl *ctrl, __le32 nsid)
413 {
414 struct nvmet_ns *ns;
415
416 rcu_read_lock();
417 ns = __nvmet_find_namespace(ctrl, nsid);
418 if (ns)
419 percpu_ref_get(&ns->ref);
420 rcu_read_unlock();
421
422 return ns;
423 }
424
425 static void nvmet_destroy_namespace(struct percpu_ref *ref)
426 {
427 struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref);
428
429 complete(&ns->disable_done);
430 }
431
432 void nvmet_put_namespace(struct nvmet_ns *ns)
433 {
434 percpu_ref_put(&ns->ref);
435 }
436
437 static void nvmet_ns_dev_disable(struct nvmet_ns *ns)
438 {
439 nvmet_bdev_ns_disable(ns);
440 nvmet_file_ns_disable(ns);
441 }
442
443 static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns)
444 {
445 int ret;
446 struct pci_dev *p2p_dev;
447
448 if (!ns->use_p2pmem)
449 return 0;
450
451 if (!ns->bdev) {
452 pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n");
453 return -EINVAL;
454 }
455
456 if (!blk_queue_pci_p2pdma(ns->bdev->bd_queue)) {
457 pr_err("peer-to-peer DMA is not supported by the driver of %s\n",
458 ns->device_path);
459 return -EINVAL;
460 }
461
462 if (ns->p2p_dev) {
463 ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true);
464 if (ret < 0)
465 return -EINVAL;
466 } else {
467 /*
468 * Right now we just check that there is p2pmem available so
469 * we can report an error to the user right away if there
470 * is not. We'll find the actual device to use once we
471 * setup the controller when the port's device is available.
472 */
473
474 p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns));
475 if (!p2p_dev) {
476 pr_err("no peer-to-peer memory is available for %s\n",
477 ns->device_path);
478 return -EINVAL;
479 }
480
481 pci_dev_put(p2p_dev);
482 }
483
484 return 0;
485 }
486
487 /*
488 * Note: ctrl->subsys->lock should be held when calling this function
489 */
490 static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl,
491 struct nvmet_ns *ns)
492 {
493 struct device *clients[2];
494 struct pci_dev *p2p_dev;
495 int ret;
496
497 if (!ctrl->p2p_client || !ns->use_p2pmem)
498 return;
499
500 if (ns->p2p_dev) {
501 ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true);
502 if (ret < 0)
503 return;
504
505 p2p_dev = pci_dev_get(ns->p2p_dev);
506 } else {
507 clients[0] = ctrl->p2p_client;
508 clients[1] = nvmet_ns_dev(ns);
509
510 p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients));
511 if (!p2p_dev) {
512 pr_err("no peer-to-peer memory is available that's supported by %s and %s\n",
513 dev_name(ctrl->p2p_client), ns->device_path);
514 return;
515 }
516 }
517
518 ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev);
519 if (ret < 0)
520 pci_dev_put(p2p_dev);
521
522 pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev),
523 ns->nsid);
524 }
525
526 int nvmet_ns_enable(struct nvmet_ns *ns)
527 {
528 struct nvmet_subsys *subsys = ns->subsys;
529 struct nvmet_ctrl *ctrl;
530 int ret;
531
532 mutex_lock(&subsys->lock);
533 ret = 0;
534 if (ns->enabled)
535 goto out_unlock;
536
537 ret = -EMFILE;
538 if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES)
539 goto out_unlock;
540
541 ret = nvmet_bdev_ns_enable(ns);
542 if (ret == -ENOTBLK)
543 ret = nvmet_file_ns_enable(ns);
544 if (ret)
545 goto out_unlock;
546
547 ret = nvmet_p2pmem_ns_enable(ns);
548 if (ret)
549 goto out_dev_disable;
550
551 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
552 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
553
554 ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
555 0, GFP_KERNEL);
556 if (ret)
557 goto out_dev_put;
558
559 if (ns->nsid > subsys->max_nsid)
560 subsys->max_nsid = ns->nsid;
561
562 /*
563 * The namespaces list needs to be sorted to simplify the implementation
564 * of the Identify Namepace List subcommand.
565 */
566 if (list_empty(&subsys->namespaces)) {
567 list_add_tail_rcu(&ns->dev_link, &subsys->namespaces);
568 } else {
569 struct nvmet_ns *old;
570
571 list_for_each_entry_rcu(old, &subsys->namespaces, dev_link,
572 lockdep_is_held(&subsys->lock)) {
573 BUG_ON(ns->nsid == old->nsid);
574 if (ns->nsid < old->nsid)
575 break;
576 }
577
578 list_add_tail_rcu(&ns->dev_link, &old->dev_link);
579 }
580 subsys->nr_namespaces++;
581
582 nvmet_ns_changed(subsys, ns->nsid);
583 ns->enabled = true;
584 ret = 0;
585 out_unlock:
586 mutex_unlock(&subsys->lock);
587 return ret;
588 out_dev_put:
589 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
590 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
591 out_dev_disable:
592 nvmet_ns_dev_disable(ns);
593 goto out_unlock;
594 }
595
596 void nvmet_ns_disable(struct nvmet_ns *ns)
597 {
598 struct nvmet_subsys *subsys = ns->subsys;
599 struct nvmet_ctrl *ctrl;
600
601 mutex_lock(&subsys->lock);
602 if (!ns->enabled)
603 goto out_unlock;
604
605 ns->enabled = false;
606 list_del_rcu(&ns->dev_link);
607 if (ns->nsid == subsys->max_nsid)
608 subsys->max_nsid = nvmet_max_nsid(subsys);
609
610 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
611 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
612
613 mutex_unlock(&subsys->lock);
614
615 /*
616 * Now that we removed the namespaces from the lookup list, we
617 * can kill the per_cpu ref and wait for any remaining references
618 * to be dropped, as well as a RCU grace period for anyone only
619 * using the namepace under rcu_read_lock(). Note that we can't
620 * use call_rcu here as we need to ensure the namespaces have
621 * been fully destroyed before unloading the module.
622 */
623 percpu_ref_kill(&ns->ref);
624 synchronize_rcu();
625 wait_for_completion(&ns->disable_done);
626 percpu_ref_exit(&ns->ref);
627
628 mutex_lock(&subsys->lock);
629
630 subsys->nr_namespaces--;
631 nvmet_ns_changed(subsys, ns->nsid);
632 nvmet_ns_dev_disable(ns);
633 out_unlock:
634 mutex_unlock(&subsys->lock);
635 }
636
637 void nvmet_ns_free(struct nvmet_ns *ns)
638 {
639 nvmet_ns_disable(ns);
640
641 down_write(&nvmet_ana_sem);
642 nvmet_ana_group_enabled[ns->anagrpid]--;
643 up_write(&nvmet_ana_sem);
644
645 kfree(ns->device_path);
646 kfree(ns);
647 }
648
649 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
650 {
651 struct nvmet_ns *ns;
652
653 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
654 if (!ns)
655 return NULL;
656
657 INIT_LIST_HEAD(&ns->dev_link);
658 init_completion(&ns->disable_done);
659
660 ns->nsid = nsid;
661 ns->subsys = subsys;
662
663 down_write(&nvmet_ana_sem);
664 ns->anagrpid = NVMET_DEFAULT_ANA_GRPID;
665 nvmet_ana_group_enabled[ns->anagrpid]++;
666 up_write(&nvmet_ana_sem);
667
668 uuid_gen(&ns->uuid);
669 ns->buffered_io = false;
670
671 return ns;
672 }
673
674 static void nvmet_update_sq_head(struct nvmet_req *req)
675 {
676 if (req->sq->size) {
677 u32 old_sqhd, new_sqhd;
678
679 do {
680 old_sqhd = req->sq->sqhd;
681 new_sqhd = (old_sqhd + 1) % req->sq->size;
682 } while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
683 old_sqhd);
684 }
685 req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
686 }
687
688 static void nvmet_set_error(struct nvmet_req *req, u16 status)
689 {
690 struct nvmet_ctrl *ctrl = req->sq->ctrl;
691 struct nvme_error_slot *new_error_slot;
692 unsigned long flags;
693
694 req->cqe->status = cpu_to_le16(status << 1);
695
696 if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC)
697 return;
698
699 spin_lock_irqsave(&ctrl->error_lock, flags);
700 ctrl->err_counter++;
701 new_error_slot =
702 &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS];
703
704 new_error_slot->error_count = cpu_to_le64(ctrl->err_counter);
705 new_error_slot->sqid = cpu_to_le16(req->sq->qid);
706 new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id);
707 new_error_slot->status_field = cpu_to_le16(status << 1);
708 new_error_slot->param_error_location = cpu_to_le16(req->error_loc);
709 new_error_slot->lba = cpu_to_le64(req->error_slba);
710 new_error_slot->nsid = req->cmd->common.nsid;
711 spin_unlock_irqrestore(&ctrl->error_lock, flags);
712
713 /* set the more bit for this request */
714 req->cqe->status |= cpu_to_le16(1 << 14);
715 }
716
717 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
718 {
719 if (!req->sq->sqhd_disabled)
720 nvmet_update_sq_head(req);
721 req->cqe->sq_id = cpu_to_le16(req->sq->qid);
722 req->cqe->command_id = req->cmd->common.command_id;
723
724 if (unlikely(status))
725 nvmet_set_error(req, status);
726
727 trace_nvmet_req_complete(req);
728
729 if (req->ns)
730 nvmet_put_namespace(req->ns);
731 req->ops->queue_response(req);
732 }
733
734 void nvmet_req_complete(struct nvmet_req *req, u16 status)
735 {
736 __nvmet_req_complete(req, status);
737 percpu_ref_put(&req->sq->ref);
738 }
739 EXPORT_SYMBOL_GPL(nvmet_req_complete);
740
741 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq,
742 u16 qid, u16 size)
743 {
744 cq->qid = qid;
745 cq->size = size;
746
747 ctrl->cqs[qid] = cq;
748 }
749
750 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq,
751 u16 qid, u16 size)
752 {
753 sq->sqhd = 0;
754 sq->qid = qid;
755 sq->size = size;
756
757 ctrl->sqs[qid] = sq;
758 }
759
760 static void nvmet_confirm_sq(struct percpu_ref *ref)
761 {
762 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
763
764 complete(&sq->confirm_done);
765 }
766
767 void nvmet_sq_destroy(struct nvmet_sq *sq)
768 {
769 struct nvmet_ctrl *ctrl = sq->ctrl;
770
771 /*
772 * If this is the admin queue, complete all AERs so that our
773 * queue doesn't have outstanding requests on it.
774 */
775 if (ctrl && ctrl->sqs && ctrl->sqs[0] == sq)
776 nvmet_async_events_failall(ctrl);
777 percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq);
778 wait_for_completion(&sq->confirm_done);
779 wait_for_completion(&sq->free_done);
780 percpu_ref_exit(&sq->ref);
781
782 if (ctrl) {
783 nvmet_ctrl_put(ctrl);
784 sq->ctrl = NULL; /* allows reusing the queue later */
785 }
786 }
787 EXPORT_SYMBOL_GPL(nvmet_sq_destroy);
788
789 static void nvmet_sq_free(struct percpu_ref *ref)
790 {
791 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
792
793 complete(&sq->free_done);
794 }
795
796 int nvmet_sq_init(struct nvmet_sq *sq)
797 {
798 int ret;
799
800 ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL);
801 if (ret) {
802 pr_err("percpu_ref init failed!\n");
803 return ret;
804 }
805 init_completion(&sq->free_done);
806 init_completion(&sq->confirm_done);
807
808 return 0;
809 }
810 EXPORT_SYMBOL_GPL(nvmet_sq_init);
811
812 static inline u16 nvmet_check_ana_state(struct nvmet_port *port,
813 struct nvmet_ns *ns)
814 {
815 enum nvme_ana_state state = port->ana_state[ns->anagrpid];
816
817 if (unlikely(state == NVME_ANA_INACCESSIBLE))
818 return NVME_SC_ANA_INACCESSIBLE;
819 if (unlikely(state == NVME_ANA_PERSISTENT_LOSS))
820 return NVME_SC_ANA_PERSISTENT_LOSS;
821 if (unlikely(state == NVME_ANA_CHANGE))
822 return NVME_SC_ANA_TRANSITION;
823 return 0;
824 }
825
826 static inline u16 nvmet_io_cmd_check_access(struct nvmet_req *req)
827 {
828 if (unlikely(req->ns->readonly)) {
829 switch (req->cmd->common.opcode) {
830 case nvme_cmd_read:
831 case nvme_cmd_flush:
832 break;
833 default:
834 return NVME_SC_NS_WRITE_PROTECTED;
835 }
836 }
837
838 return 0;
839 }
840
841 static u16 nvmet_parse_io_cmd(struct nvmet_req *req)
842 {
843 struct nvme_command *cmd = req->cmd;
844 u16 ret;
845
846 ret = nvmet_check_ctrl_status(req, cmd);
847 if (unlikely(ret))
848 return ret;
849
850 req->ns = nvmet_find_namespace(req->sq->ctrl, cmd->rw.nsid);
851 if (unlikely(!req->ns)) {
852 req->error_loc = offsetof(struct nvme_common_command, nsid);
853 return NVME_SC_INVALID_NS | NVME_SC_DNR;
854 }
855 ret = nvmet_check_ana_state(req->port, req->ns);
856 if (unlikely(ret)) {
857 req->error_loc = offsetof(struct nvme_common_command, nsid);
858 return ret;
859 }
860 ret = nvmet_io_cmd_check_access(req);
861 if (unlikely(ret)) {
862 req->error_loc = offsetof(struct nvme_common_command, nsid);
863 return ret;
864 }
865
866 if (req->ns->file)
867 return nvmet_file_parse_io_cmd(req);
868 else
869 return nvmet_bdev_parse_io_cmd(req);
870 }
871
872 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
873 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops)
874 {
875 u8 flags = req->cmd->common.flags;
876 u16 status;
877
878 req->cq = cq;
879 req->sq = sq;
880 req->ops = ops;
881 req->sg = NULL;
882 req->sg_cnt = 0;
883 req->transfer_len = 0;
884 req->cqe->status = 0;
885 req->cqe->sq_head = 0;
886 req->ns = NULL;
887 req->error_loc = NVMET_NO_ERROR_LOC;
888 req->error_slba = 0;
889
890 trace_nvmet_req_init(req, req->cmd);
891
892 /* no support for fused commands yet */
893 if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) {
894 req->error_loc = offsetof(struct nvme_common_command, flags);
895 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
896 goto fail;
897 }
898
899 /*
900 * For fabrics, PSDT field shall describe metadata pointer (MPTR) that
901 * contains an address of a single contiguous physical buffer that is
902 * byte aligned.
903 */
904 if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) {
905 req->error_loc = offsetof(struct nvme_common_command, flags);
906 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
907 goto fail;
908 }
909
910 if (unlikely(!req->sq->ctrl))
911 /* will return an error for any non-connect command: */
912 status = nvmet_parse_connect_cmd(req);
913 else if (likely(req->sq->qid != 0))
914 status = nvmet_parse_io_cmd(req);
915 else
916 status = nvmet_parse_admin_cmd(req);
917
918 if (status)
919 goto fail;
920
921 if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
922 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
923 goto fail;
924 }
925
926 if (sq->ctrl)
927 sq->ctrl->cmd_seen = true;
928
929 return true;
930
931 fail:
932 __nvmet_req_complete(req, status);
933 return false;
934 }
935 EXPORT_SYMBOL_GPL(nvmet_req_init);
936
937 void nvmet_req_uninit(struct nvmet_req *req)
938 {
939 percpu_ref_put(&req->sq->ref);
940 if (req->ns)
941 nvmet_put_namespace(req->ns);
942 }
943 EXPORT_SYMBOL_GPL(nvmet_req_uninit);
944
945 bool nvmet_check_data_len(struct nvmet_req *req, size_t data_len)
946 {
947 if (unlikely(data_len != req->transfer_len)) {
948 req->error_loc = offsetof(struct nvme_common_command, dptr);
949 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
950 return false;
951 }
952
953 return true;
954 }
955 EXPORT_SYMBOL_GPL(nvmet_check_data_len);
956
957 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len)
958 {
959 if (unlikely(data_len > req->transfer_len)) {
960 req->error_loc = offsetof(struct nvme_common_command, dptr);
961 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
962 return false;
963 }
964
965 return true;
966 }
967
968 int nvmet_req_alloc_sgl(struct nvmet_req *req)
969 {
970 struct pci_dev *p2p_dev = NULL;
971
972 if (IS_ENABLED(CONFIG_PCI_P2PDMA)) {
973 if (req->sq->ctrl && req->ns)
974 p2p_dev = radix_tree_lookup(&req->sq->ctrl->p2p_ns_map,
975 req->ns->nsid);
976
977 req->p2p_dev = NULL;
978 if (req->sq->qid && p2p_dev) {
979 req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt,
980 req->transfer_len);
981 if (req->sg) {
982 req->p2p_dev = p2p_dev;
983 return 0;
984 }
985 }
986
987 /*
988 * If no P2P memory was available we fallback to using
989 * regular memory
990 */
991 }
992
993 req->sg = sgl_alloc(req->transfer_len, GFP_KERNEL, &req->sg_cnt);
994 if (unlikely(!req->sg))
995 return -ENOMEM;
996
997 return 0;
998 }
999 EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgl);
1000
1001 void nvmet_req_free_sgl(struct nvmet_req *req)
1002 {
1003 if (req->p2p_dev)
1004 pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1005 else
1006 sgl_free(req->sg);
1007
1008 req->sg = NULL;
1009 req->sg_cnt = 0;
1010 }
1011 EXPORT_SYMBOL_GPL(nvmet_req_free_sgl);
1012
1013 static inline bool nvmet_cc_en(u32 cc)
1014 {
1015 return (cc >> NVME_CC_EN_SHIFT) & 0x1;
1016 }
1017
1018 static inline u8 nvmet_cc_css(u32 cc)
1019 {
1020 return (cc >> NVME_CC_CSS_SHIFT) & 0x7;
1021 }
1022
1023 static inline u8 nvmet_cc_mps(u32 cc)
1024 {
1025 return (cc >> NVME_CC_MPS_SHIFT) & 0xf;
1026 }
1027
1028 static inline u8 nvmet_cc_ams(u32 cc)
1029 {
1030 return (cc >> NVME_CC_AMS_SHIFT) & 0x7;
1031 }
1032
1033 static inline u8 nvmet_cc_shn(u32 cc)
1034 {
1035 return (cc >> NVME_CC_SHN_SHIFT) & 0x3;
1036 }
1037
1038 static inline u8 nvmet_cc_iosqes(u32 cc)
1039 {
1040 return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf;
1041 }
1042
1043 static inline u8 nvmet_cc_iocqes(u32 cc)
1044 {
1045 return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf;
1046 }
1047
1048 static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl)
1049 {
1050 lockdep_assert_held(&ctrl->lock);
1051
1052 if (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES ||
1053 nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES ||
1054 nvmet_cc_mps(ctrl->cc) != 0 ||
1055 nvmet_cc_ams(ctrl->cc) != 0 ||
1056 nvmet_cc_css(ctrl->cc) != 0) {
1057 ctrl->csts = NVME_CSTS_CFS;
1058 return;
1059 }
1060
1061 ctrl->csts = NVME_CSTS_RDY;
1062
1063 /*
1064 * Controllers that are not yet enabled should not really enforce the
1065 * keep alive timeout, but we still want to track a timeout and cleanup
1066 * in case a host died before it enabled the controller. Hence, simply
1067 * reset the keep alive timer when the controller is enabled.
1068 */
1069 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
1070 }
1071
1072 static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl)
1073 {
1074 lockdep_assert_held(&ctrl->lock);
1075
1076 /* XXX: tear down queues? */
1077 ctrl->csts &= ~NVME_CSTS_RDY;
1078 ctrl->cc = 0;
1079 }
1080
1081 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new)
1082 {
1083 u32 old;
1084
1085 mutex_lock(&ctrl->lock);
1086 old = ctrl->cc;
1087 ctrl->cc = new;
1088
1089 if (nvmet_cc_en(new) && !nvmet_cc_en(old))
1090 nvmet_start_ctrl(ctrl);
1091 if (!nvmet_cc_en(new) && nvmet_cc_en(old))
1092 nvmet_clear_ctrl(ctrl);
1093 if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) {
1094 nvmet_clear_ctrl(ctrl);
1095 ctrl->csts |= NVME_CSTS_SHST_CMPLT;
1096 }
1097 if (!nvmet_cc_shn(new) && nvmet_cc_shn(old))
1098 ctrl->csts &= ~NVME_CSTS_SHST_CMPLT;
1099 mutex_unlock(&ctrl->lock);
1100 }
1101
1102 static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
1103 {
1104 /* command sets supported: NVMe command set: */
1105 ctrl->cap = (1ULL << 37);
1106 /* CC.EN timeout in 500msec units: */
1107 ctrl->cap |= (15ULL << 24);
1108 /* maximum queue entries supported: */
1109 ctrl->cap |= NVMET_QUEUE_SIZE - 1;
1110 }
1111
1112 u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid,
1113 struct nvmet_req *req, struct nvmet_ctrl **ret)
1114 {
1115 struct nvmet_subsys *subsys;
1116 struct nvmet_ctrl *ctrl;
1117 u16 status = 0;
1118
1119 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1120 if (!subsys) {
1121 pr_warn("connect request for invalid subsystem %s!\n",
1122 subsysnqn);
1123 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1124 return NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1125 }
1126
1127 mutex_lock(&subsys->lock);
1128 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
1129 if (ctrl->cntlid == cntlid) {
1130 if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) {
1131 pr_warn("hostnqn mismatch.\n");
1132 continue;
1133 }
1134 if (!kref_get_unless_zero(&ctrl->ref))
1135 continue;
1136
1137 *ret = ctrl;
1138 goto out;
1139 }
1140 }
1141
1142 pr_warn("could not find controller %d for subsys %s / host %s\n",
1143 cntlid, subsysnqn, hostnqn);
1144 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
1145 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1146
1147 out:
1148 mutex_unlock(&subsys->lock);
1149 nvmet_subsys_put(subsys);
1150 return status;
1151 }
1152
1153 u16 nvmet_check_ctrl_status(struct nvmet_req *req, struct nvme_command *cmd)
1154 {
1155 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
1156 pr_err("got cmd %d while CC.EN == 0 on qid = %d\n",
1157 cmd->common.opcode, req->sq->qid);
1158 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1159 }
1160
1161 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
1162 pr_err("got cmd %d while CSTS.RDY == 0 on qid = %d\n",
1163 cmd->common.opcode, req->sq->qid);
1164 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1165 }
1166 return 0;
1167 }
1168
1169 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn)
1170 {
1171 struct nvmet_host_link *p;
1172
1173 lockdep_assert_held(&nvmet_config_sem);
1174
1175 if (subsys->allow_any_host)
1176 return true;
1177
1178 if (subsys->type == NVME_NQN_DISC) /* allow all access to disc subsys */
1179 return true;
1180
1181 list_for_each_entry(p, &subsys->hosts, entry) {
1182 if (!strcmp(nvmet_host_name(p->host), hostnqn))
1183 return true;
1184 }
1185
1186 return false;
1187 }
1188
1189 /*
1190 * Note: ctrl->subsys->lock should be held when calling this function
1191 */
1192 static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl,
1193 struct nvmet_req *req)
1194 {
1195 struct nvmet_ns *ns;
1196
1197 if (!req->p2p_client)
1198 return;
1199
1200 ctrl->p2p_client = get_device(req->p2p_client);
1201
1202 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link,
1203 lockdep_is_held(&ctrl->subsys->lock))
1204 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
1205 }
1206
1207 /*
1208 * Note: ctrl->subsys->lock should be held when calling this function
1209 */
1210 static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl)
1211 {
1212 struct radix_tree_iter iter;
1213 void __rcu **slot;
1214
1215 radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0)
1216 pci_dev_put(radix_tree_deref_slot(slot));
1217
1218 put_device(ctrl->p2p_client);
1219 }
1220
1221 static void nvmet_fatal_error_handler(struct work_struct *work)
1222 {
1223 struct nvmet_ctrl *ctrl =
1224 container_of(work, struct nvmet_ctrl, fatal_err_work);
1225
1226 pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid);
1227 ctrl->ops->delete_ctrl(ctrl);
1228 }
1229
1230 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
1231 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp)
1232 {
1233 struct nvmet_subsys *subsys;
1234 struct nvmet_ctrl *ctrl;
1235 int ret;
1236 u16 status;
1237
1238 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1239 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1240 if (!subsys) {
1241 pr_warn("connect request for invalid subsystem %s!\n",
1242 subsysnqn);
1243 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1244 goto out;
1245 }
1246
1247 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1248 down_read(&nvmet_config_sem);
1249 if (!nvmet_host_allowed(subsys, hostnqn)) {
1250 pr_info("connect by host %s for subsystem %s not allowed\n",
1251 hostnqn, subsysnqn);
1252 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
1253 up_read(&nvmet_config_sem);
1254 status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
1255 goto out_put_subsystem;
1256 }
1257 up_read(&nvmet_config_sem);
1258
1259 status = NVME_SC_INTERNAL;
1260 ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
1261 if (!ctrl)
1262 goto out_put_subsystem;
1263 mutex_init(&ctrl->lock);
1264
1265 nvmet_init_cap(ctrl);
1266
1267 ctrl->port = req->port;
1268
1269 INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
1270 INIT_LIST_HEAD(&ctrl->async_events);
1271 INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL);
1272 INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
1273
1274 memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
1275 memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE);
1276
1277 kref_init(&ctrl->ref);
1278 ctrl->subsys = subsys;
1279 WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL);
1280
1281 ctrl->changed_ns_list = kmalloc_array(NVME_MAX_CHANGED_NAMESPACES,
1282 sizeof(__le32), GFP_KERNEL);
1283 if (!ctrl->changed_ns_list)
1284 goto out_free_ctrl;
1285
1286 ctrl->cqs = kcalloc(subsys->max_qid + 1,
1287 sizeof(struct nvmet_cq *),
1288 GFP_KERNEL);
1289 if (!ctrl->cqs)
1290 goto out_free_changed_ns_list;
1291
1292 ctrl->sqs = kcalloc(subsys->max_qid + 1,
1293 sizeof(struct nvmet_sq *),
1294 GFP_KERNEL);
1295 if (!ctrl->sqs)
1296 goto out_free_cqs;
1297
1298 if (subsys->cntlid_min > subsys->cntlid_max)
1299 goto out_free_cqs;
1300
1301 ret = ida_simple_get(&cntlid_ida,
1302 subsys->cntlid_min, subsys->cntlid_max,
1303 GFP_KERNEL);
1304 if (ret < 0) {
1305 status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR;
1306 goto out_free_sqs;
1307 }
1308 ctrl->cntlid = ret;
1309
1310 ctrl->ops = req->ops;
1311
1312 /*
1313 * Discovery controllers may use some arbitrary high value
1314 * in order to cleanup stale discovery sessions
1315 */
1316 if ((ctrl->subsys->type == NVME_NQN_DISC) && !kato)
1317 kato = NVMET_DISC_KATO_MS;
1318
1319 /* keep-alive timeout in seconds */
1320 ctrl->kato = DIV_ROUND_UP(kato, 1000);
1321
1322 ctrl->err_counter = 0;
1323 spin_lock_init(&ctrl->error_lock);
1324
1325 nvmet_start_keep_alive_timer(ctrl);
1326
1327 mutex_lock(&subsys->lock);
1328 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
1329 nvmet_setup_p2p_ns_map(ctrl, req);
1330 mutex_unlock(&subsys->lock);
1331
1332 *ctrlp = ctrl;
1333 return 0;
1334
1335 out_free_sqs:
1336 kfree(ctrl->sqs);
1337 out_free_cqs:
1338 kfree(ctrl->cqs);
1339 out_free_changed_ns_list:
1340 kfree(ctrl->changed_ns_list);
1341 out_free_ctrl:
1342 kfree(ctrl);
1343 out_put_subsystem:
1344 nvmet_subsys_put(subsys);
1345 out:
1346 return status;
1347 }
1348
1349 static void nvmet_ctrl_free(struct kref *ref)
1350 {
1351 struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref);
1352 struct nvmet_subsys *subsys = ctrl->subsys;
1353
1354 mutex_lock(&subsys->lock);
1355 nvmet_release_p2p_ns_map(ctrl);
1356 list_del(&ctrl->subsys_entry);
1357 mutex_unlock(&subsys->lock);
1358
1359 nvmet_stop_keep_alive_timer(ctrl);
1360
1361 flush_work(&ctrl->async_event_work);
1362 cancel_work_sync(&ctrl->fatal_err_work);
1363
1364 ida_simple_remove(&cntlid_ida, ctrl->cntlid);
1365
1366 nvmet_async_events_free(ctrl);
1367 kfree(ctrl->sqs);
1368 kfree(ctrl->cqs);
1369 kfree(ctrl->changed_ns_list);
1370 kfree(ctrl);
1371
1372 nvmet_subsys_put(subsys);
1373 }
1374
1375 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl)
1376 {
1377 kref_put(&ctrl->ref, nvmet_ctrl_free);
1378 }
1379
1380 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl)
1381 {
1382 mutex_lock(&ctrl->lock);
1383 if (!(ctrl->csts & NVME_CSTS_CFS)) {
1384 ctrl->csts |= NVME_CSTS_CFS;
1385 schedule_work(&ctrl->fatal_err_work);
1386 }
1387 mutex_unlock(&ctrl->lock);
1388 }
1389 EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error);
1390
1391 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
1392 const char *subsysnqn)
1393 {
1394 struct nvmet_subsys_link *p;
1395
1396 if (!port)
1397 return NULL;
1398
1399 if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) {
1400 if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
1401 return NULL;
1402 return nvmet_disc_subsys;
1403 }
1404
1405 down_read(&nvmet_config_sem);
1406 list_for_each_entry(p, &port->subsystems, entry) {
1407 if (!strncmp(p->subsys->subsysnqn, subsysnqn,
1408 NVMF_NQN_SIZE)) {
1409 if (!kref_get_unless_zero(&p->subsys->ref))
1410 break;
1411 up_read(&nvmet_config_sem);
1412 return p->subsys;
1413 }
1414 }
1415 up_read(&nvmet_config_sem);
1416 return NULL;
1417 }
1418
1419 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
1420 enum nvme_subsys_type type)
1421 {
1422 struct nvmet_subsys *subsys;
1423
1424 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
1425 if (!subsys)
1426 return ERR_PTR(-ENOMEM);
1427
1428 subsys->ver = NVME_VS(1, 3, 0); /* NVMe 1.3.0 */
1429 /* generate a random serial number as our controllers are ephemeral: */
1430 get_random_bytes(&subsys->serial, sizeof(subsys->serial));
1431
1432 switch (type) {
1433 case NVME_NQN_NVME:
1434 subsys->max_qid = NVMET_NR_QUEUES;
1435 break;
1436 case NVME_NQN_DISC:
1437 subsys->max_qid = 0;
1438 break;
1439 default:
1440 pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
1441 kfree(subsys);
1442 return ERR_PTR(-EINVAL);
1443 }
1444 subsys->type = type;
1445 subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
1446 GFP_KERNEL);
1447 if (!subsys->subsysnqn) {
1448 kfree(subsys);
1449 return ERR_PTR(-ENOMEM);
1450 }
1451 subsys->cntlid_min = NVME_CNTLID_MIN;
1452 subsys->cntlid_max = NVME_CNTLID_MAX;
1453 kref_init(&subsys->ref);
1454
1455 mutex_init(&subsys->lock);
1456 INIT_LIST_HEAD(&subsys->namespaces);
1457 INIT_LIST_HEAD(&subsys->ctrls);
1458 INIT_LIST_HEAD(&subsys->hosts);
1459
1460 return subsys;
1461 }
1462
1463 static void nvmet_subsys_free(struct kref *ref)
1464 {
1465 struct nvmet_subsys *subsys =
1466 container_of(ref, struct nvmet_subsys, ref);
1467
1468 WARN_ON_ONCE(!list_empty(&subsys->namespaces));
1469
1470 kfree(subsys->subsysnqn);
1471 kfree_rcu(subsys->model, rcuhead);
1472 kfree(subsys);
1473 }
1474
1475 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys)
1476 {
1477 struct nvmet_ctrl *ctrl;
1478
1479 mutex_lock(&subsys->lock);
1480 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
1481 ctrl->ops->delete_ctrl(ctrl);
1482 mutex_unlock(&subsys->lock);
1483 }
1484
1485 void nvmet_subsys_put(struct nvmet_subsys *subsys)
1486 {
1487 kref_put(&subsys->ref, nvmet_subsys_free);
1488 }
1489
1490 static int __init nvmet_init(void)
1491 {
1492 int error;
1493
1494 nvmet_ana_group_enabled[NVMET_DEFAULT_ANA_GRPID] = 1;
1495
1496 buffered_io_wq = alloc_workqueue("nvmet-buffered-io-wq",
1497 WQ_MEM_RECLAIM, 0);
1498 if (!buffered_io_wq) {
1499 error = -ENOMEM;
1500 goto out;
1501 }
1502
1503 error = nvmet_init_discovery();
1504 if (error)
1505 goto out_free_work_queue;
1506
1507 error = nvmet_init_configfs();
1508 if (error)
1509 goto out_exit_discovery;
1510 return 0;
1511
1512 out_exit_discovery:
1513 nvmet_exit_discovery();
1514 out_free_work_queue:
1515 destroy_workqueue(buffered_io_wq);
1516 out:
1517 return error;
1518 }
1519
1520 static void __exit nvmet_exit(void)
1521 {
1522 nvmet_exit_configfs();
1523 nvmet_exit_discovery();
1524 ida_destroy(&cntlid_ida);
1525 destroy_workqueue(buffered_io_wq);
1526
1527 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024);
1528 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024);
1529 }
1530
1531 module_init(nvmet_init);
1532 module_exit(nvmet_exit);
1533
1534 MODULE_LICENSE("GPL v2");
Linux with added FPC-III support