Loading...
Note: File does not exist in v3.15.
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#include <generated/utsrelease.h>
14
15#define CREATE_TRACE_POINTS
16#include "trace.h"
17
18#include "nvmet.h"
19
20struct kmem_cache *nvmet_bvec_cache;
21struct workqueue_struct *buffered_io_wq;
22struct workqueue_struct *zbd_wq;
23static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
24static DEFINE_IDA(cntlid_ida);
25
26struct workqueue_struct *nvmet_wq;
27EXPORT_SYMBOL_GPL(nvmet_wq);
28
29/*
30 * This read/write semaphore is used to synchronize access to configuration
31 * information on a target system that will result in discovery log page
32 * information change for at least one host.
33 * The full list of resources to protected by this semaphore is:
34 *
35 * - subsystems list
36 * - per-subsystem allowed hosts list
37 * - allow_any_host subsystem attribute
38 * - nvmet_genctr
39 * - the nvmet_transports array
40 *
41 * When updating any of those lists/structures write lock should be obtained,
42 * while when reading (popolating discovery log page or checking host-subsystem
43 * link) read lock is obtained to allow concurrent reads.
44 */
45DECLARE_RWSEM(nvmet_config_sem);
46
47u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
48u64 nvmet_ana_chgcnt;
49DECLARE_RWSEM(nvmet_ana_sem);
50
51inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno)
52{
53 switch (errno) {
54 case 0:
55 return NVME_SC_SUCCESS;
56 case -ENOSPC:
57 req->error_loc = offsetof(struct nvme_rw_command, length);
58 return NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
59 case -EREMOTEIO:
60 req->error_loc = offsetof(struct nvme_rw_command, slba);
61 return NVME_SC_LBA_RANGE | NVME_SC_DNR;
62 case -EOPNOTSUPP:
63 req->error_loc = offsetof(struct nvme_common_command, opcode);
64 switch (req->cmd->common.opcode) {
65 case nvme_cmd_dsm:
66 case nvme_cmd_write_zeroes:
67 return NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
68 default:
69 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
70 }
71 break;
72 case -ENODATA:
73 req->error_loc = offsetof(struct nvme_rw_command, nsid);
74 return NVME_SC_ACCESS_DENIED;
75 case -EIO:
76 fallthrough;
77 default:
78 req->error_loc = offsetof(struct nvme_common_command, opcode);
79 return NVME_SC_INTERNAL | NVME_SC_DNR;
80 }
81}
82
83u16 nvmet_report_invalid_opcode(struct nvmet_req *req)
84{
85 pr_debug("unhandled cmd %d on qid %d\n", req->cmd->common.opcode,
86 req->sq->qid);
87
88 req->error_loc = offsetof(struct nvme_common_command, opcode);
89 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
90}
91
92static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
93 const char *subsysnqn);
94
95u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
96 size_t len)
97{
98 if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
99 req->error_loc = offsetof(struct nvme_common_command, dptr);
100 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
101 }
102 return 0;
103}
104
105u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len)
106{
107 if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
108 req->error_loc = offsetof(struct nvme_common_command, dptr);
109 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
110 }
111 return 0;
112}
113
114u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len)
115{
116 if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len) {
117 req->error_loc = offsetof(struct nvme_common_command, dptr);
118 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
119 }
120 return 0;
121}
122
123static u32 nvmet_max_nsid(struct nvmet_subsys *subsys)
124{
125 struct nvmet_ns *cur;
126 unsigned long idx;
127 u32 nsid = 0;
128
129 xa_for_each(&subsys->namespaces, idx, cur)
130 nsid = cur->nsid;
131
132 return nsid;
133}
134
135static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
136{
137 return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16);
138}
139
140static void nvmet_async_events_failall(struct nvmet_ctrl *ctrl)
141{
142 struct nvmet_req *req;
143
144 mutex_lock(&ctrl->lock);
145 while (ctrl->nr_async_event_cmds) {
146 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
147 mutex_unlock(&ctrl->lock);
148 nvmet_req_complete(req, NVME_SC_INTERNAL | NVME_SC_DNR);
149 mutex_lock(&ctrl->lock);
150 }
151 mutex_unlock(&ctrl->lock);
152}
153
154static void nvmet_async_events_process(struct nvmet_ctrl *ctrl)
155{
156 struct nvmet_async_event *aen;
157 struct nvmet_req *req;
158
159 mutex_lock(&ctrl->lock);
160 while (ctrl->nr_async_event_cmds && !list_empty(&ctrl->async_events)) {
161 aen = list_first_entry(&ctrl->async_events,
162 struct nvmet_async_event, entry);
163 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
164 nvmet_set_result(req, nvmet_async_event_result(aen));
165
166 list_del(&aen->entry);
167 kfree(aen);
168
169 mutex_unlock(&ctrl->lock);
170 trace_nvmet_async_event(ctrl, req->cqe->result.u32);
171 nvmet_req_complete(req, 0);
172 mutex_lock(&ctrl->lock);
173 }
174 mutex_unlock(&ctrl->lock);
175}
176
177static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
178{
179 struct nvmet_async_event *aen, *tmp;
180
181 mutex_lock(&ctrl->lock);
182 list_for_each_entry_safe(aen, tmp, &ctrl->async_events, entry) {
183 list_del(&aen->entry);
184 kfree(aen);
185 }
186 mutex_unlock(&ctrl->lock);
187}
188
189static void nvmet_async_event_work(struct work_struct *work)
190{
191 struct nvmet_ctrl *ctrl =
192 container_of(work, struct nvmet_ctrl, async_event_work);
193
194 nvmet_async_events_process(ctrl);
195}
196
197void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
198 u8 event_info, u8 log_page)
199{
200 struct nvmet_async_event *aen;
201
202 aen = kmalloc(sizeof(*aen), GFP_KERNEL);
203 if (!aen)
204 return;
205
206 aen->event_type = event_type;
207 aen->event_info = event_info;
208 aen->log_page = log_page;
209
210 mutex_lock(&ctrl->lock);
211 list_add_tail(&aen->entry, &ctrl->async_events);
212 mutex_unlock(&ctrl->lock);
213
214 queue_work(nvmet_wq, &ctrl->async_event_work);
215}
216
217static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
218{
219 u32 i;
220
221 mutex_lock(&ctrl->lock);
222 if (ctrl->nr_changed_ns > NVME_MAX_CHANGED_NAMESPACES)
223 goto out_unlock;
224
225 for (i = 0; i < ctrl->nr_changed_ns; i++) {
226 if (ctrl->changed_ns_list[i] == nsid)
227 goto out_unlock;
228 }
229
230 if (ctrl->nr_changed_ns == NVME_MAX_CHANGED_NAMESPACES) {
231 ctrl->changed_ns_list[0] = cpu_to_le32(0xffffffff);
232 ctrl->nr_changed_ns = U32_MAX;
233 goto out_unlock;
234 }
235
236 ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid;
237out_unlock:
238 mutex_unlock(&ctrl->lock);
239}
240
241void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid)
242{
243 struct nvmet_ctrl *ctrl;
244
245 lockdep_assert_held(&subsys->lock);
246
247 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
248 nvmet_add_to_changed_ns_log(ctrl, cpu_to_le32(nsid));
249 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_NS_ATTR))
250 continue;
251 nvmet_add_async_event(ctrl, NVME_AER_NOTICE,
252 NVME_AER_NOTICE_NS_CHANGED,
253 NVME_LOG_CHANGED_NS);
254 }
255}
256
257void nvmet_send_ana_event(struct nvmet_subsys *subsys,
258 struct nvmet_port *port)
259{
260 struct nvmet_ctrl *ctrl;
261
262 mutex_lock(&subsys->lock);
263 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
264 if (port && ctrl->port != port)
265 continue;
266 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE))
267 continue;
268 nvmet_add_async_event(ctrl, NVME_AER_NOTICE,
269 NVME_AER_NOTICE_ANA, NVME_LOG_ANA);
270 }
271 mutex_unlock(&subsys->lock);
272}
273
274void nvmet_port_send_ana_event(struct nvmet_port *port)
275{
276 struct nvmet_subsys_link *p;
277
278 down_read(&nvmet_config_sem);
279 list_for_each_entry(p, &port->subsystems, entry)
280 nvmet_send_ana_event(p->subsys, port);
281 up_read(&nvmet_config_sem);
282}
283
284int nvmet_register_transport(const struct nvmet_fabrics_ops *ops)
285{
286 int ret = 0;
287
288 down_write(&nvmet_config_sem);
289 if (nvmet_transports[ops->type])
290 ret = -EINVAL;
291 else
292 nvmet_transports[ops->type] = ops;
293 up_write(&nvmet_config_sem);
294
295 return ret;
296}
297EXPORT_SYMBOL_GPL(nvmet_register_transport);
298
299void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops)
300{
301 down_write(&nvmet_config_sem);
302 nvmet_transports[ops->type] = NULL;
303 up_write(&nvmet_config_sem);
304}
305EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
306
307void nvmet_port_del_ctrls(struct nvmet_port *port, struct nvmet_subsys *subsys)
308{
309 struct nvmet_ctrl *ctrl;
310
311 mutex_lock(&subsys->lock);
312 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
313 if (ctrl->port == port)
314 ctrl->ops->delete_ctrl(ctrl);
315 }
316 mutex_unlock(&subsys->lock);
317}
318
319int nvmet_enable_port(struct nvmet_port *port)
320{
321 const struct nvmet_fabrics_ops *ops;
322 int ret;
323
324 lockdep_assert_held(&nvmet_config_sem);
325
326 ops = nvmet_transports[port->disc_addr.trtype];
327 if (!ops) {
328 up_write(&nvmet_config_sem);
329 request_module("nvmet-transport-%d", port->disc_addr.trtype);
330 down_write(&nvmet_config_sem);
331 ops = nvmet_transports[port->disc_addr.trtype];
332 if (!ops) {
333 pr_err("transport type %d not supported\n",
334 port->disc_addr.trtype);
335 return -EINVAL;
336 }
337 }
338
339 if (!try_module_get(ops->owner))
340 return -EINVAL;
341
342 /*
343 * If the user requested PI support and the transport isn't pi capable,
344 * don't enable the port.
345 */
346 if (port->pi_enable && !(ops->flags & NVMF_METADATA_SUPPORTED)) {
347 pr_err("T10-PI is not supported by transport type %d\n",
348 port->disc_addr.trtype);
349 ret = -EINVAL;
350 goto out_put;
351 }
352
353 ret = ops->add_port(port);
354 if (ret)
355 goto out_put;
356
357 /* If the transport didn't set inline_data_size, then disable it. */
358 if (port->inline_data_size < 0)
359 port->inline_data_size = 0;
360
361 /*
362 * If the transport didn't set the max_queue_size properly, then clamp
363 * it to the target limits. Also set default values in case the
364 * transport didn't set it at all.
365 */
366 if (port->max_queue_size < 0)
367 port->max_queue_size = NVMET_MAX_QUEUE_SIZE;
368 else
369 port->max_queue_size = clamp_t(int, port->max_queue_size,
370 NVMET_MIN_QUEUE_SIZE,
371 NVMET_MAX_QUEUE_SIZE);
372
373 port->enabled = true;
374 port->tr_ops = ops;
375 return 0;
376
377out_put:
378 module_put(ops->owner);
379 return ret;
380}
381
382void nvmet_disable_port(struct nvmet_port *port)
383{
384 const struct nvmet_fabrics_ops *ops;
385
386 lockdep_assert_held(&nvmet_config_sem);
387
388 port->enabled = false;
389 port->tr_ops = NULL;
390
391 ops = nvmet_transports[port->disc_addr.trtype];
392 ops->remove_port(port);
393 module_put(ops->owner);
394}
395
396static void nvmet_keep_alive_timer(struct work_struct *work)
397{
398 struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
399 struct nvmet_ctrl, ka_work);
400 bool reset_tbkas = ctrl->reset_tbkas;
401
402 ctrl->reset_tbkas = false;
403 if (reset_tbkas) {
404 pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
405 ctrl->cntlid);
406 queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
407 return;
408 }
409
410 pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
411 ctrl->cntlid, ctrl->kato);
412
413 nvmet_ctrl_fatal_error(ctrl);
414}
415
416void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
417{
418 if (unlikely(ctrl->kato == 0))
419 return;
420
421 pr_debug("ctrl %d start keep-alive timer for %d secs\n",
422 ctrl->cntlid, ctrl->kato);
423
424 queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
425}
426
427void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
428{
429 if (unlikely(ctrl->kato == 0))
430 return;
431
432 pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid);
433
434 cancel_delayed_work_sync(&ctrl->ka_work);
435}
436
437u16 nvmet_req_find_ns(struct nvmet_req *req)
438{
439 u32 nsid = le32_to_cpu(req->cmd->common.nsid);
440 struct nvmet_subsys *subsys = nvmet_req_subsys(req);
441
442 req->ns = xa_load(&subsys->namespaces, nsid);
443 if (unlikely(!req->ns)) {
444 req->error_loc = offsetof(struct nvme_common_command, nsid);
445 if (nvmet_subsys_nsid_exists(subsys, nsid))
446 return NVME_SC_INTERNAL_PATH_ERROR;
447 return NVME_SC_INVALID_NS | NVME_SC_DNR;
448 }
449
450 percpu_ref_get(&req->ns->ref);
451 return NVME_SC_SUCCESS;
452}
453
454static void nvmet_destroy_namespace(struct percpu_ref *ref)
455{
456 struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref);
457
458 complete(&ns->disable_done);
459}
460
461void nvmet_put_namespace(struct nvmet_ns *ns)
462{
463 percpu_ref_put(&ns->ref);
464}
465
466static void nvmet_ns_dev_disable(struct nvmet_ns *ns)
467{
468 nvmet_bdev_ns_disable(ns);
469 nvmet_file_ns_disable(ns);
470}
471
472static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns)
473{
474 int ret;
475 struct pci_dev *p2p_dev;
476
477 if (!ns->use_p2pmem)
478 return 0;
479
480 if (!ns->bdev) {
481 pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n");
482 return -EINVAL;
483 }
484
485 if (!blk_queue_pci_p2pdma(ns->bdev->bd_disk->queue)) {
486 pr_err("peer-to-peer DMA is not supported by the driver of %s\n",
487 ns->device_path);
488 return -EINVAL;
489 }
490
491 if (ns->p2p_dev) {
492 ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true);
493 if (ret < 0)
494 return -EINVAL;
495 } else {
496 /*
497 * Right now we just check that there is p2pmem available so
498 * we can report an error to the user right away if there
499 * is not. We'll find the actual device to use once we
500 * setup the controller when the port's device is available.
501 */
502
503 p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns));
504 if (!p2p_dev) {
505 pr_err("no peer-to-peer memory is available for %s\n",
506 ns->device_path);
507 return -EINVAL;
508 }
509
510 pci_dev_put(p2p_dev);
511 }
512
513 return 0;
514}
515
516/*
517 * Note: ctrl->subsys->lock should be held when calling this function
518 */
519static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl,
520 struct nvmet_ns *ns)
521{
522 struct device *clients[2];
523 struct pci_dev *p2p_dev;
524 int ret;
525
526 if (!ctrl->p2p_client || !ns->use_p2pmem)
527 return;
528
529 if (ns->p2p_dev) {
530 ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true);
531 if (ret < 0)
532 return;
533
534 p2p_dev = pci_dev_get(ns->p2p_dev);
535 } else {
536 clients[0] = ctrl->p2p_client;
537 clients[1] = nvmet_ns_dev(ns);
538
539 p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients));
540 if (!p2p_dev) {
541 pr_err("no peer-to-peer memory is available that's supported by %s and %s\n",
542 dev_name(ctrl->p2p_client), ns->device_path);
543 return;
544 }
545 }
546
547 ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev);
548 if (ret < 0)
549 pci_dev_put(p2p_dev);
550
551 pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev),
552 ns->nsid);
553}
554
555bool nvmet_ns_revalidate(struct nvmet_ns *ns)
556{
557 loff_t oldsize = ns->size;
558
559 if (ns->bdev)
560 nvmet_bdev_ns_revalidate(ns);
561 else
562 nvmet_file_ns_revalidate(ns);
563
564 return oldsize != ns->size;
565}
566
567int nvmet_ns_enable(struct nvmet_ns *ns)
568{
569 struct nvmet_subsys *subsys = ns->subsys;
570 struct nvmet_ctrl *ctrl;
571 int ret;
572
573 mutex_lock(&subsys->lock);
574 ret = 0;
575
576 if (nvmet_is_passthru_subsys(subsys)) {
577 pr_info("cannot enable both passthru and regular namespaces for a single subsystem");
578 goto out_unlock;
579 }
580
581 if (ns->enabled)
582 goto out_unlock;
583
584 ret = -EMFILE;
585 if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES)
586 goto out_unlock;
587
588 ret = nvmet_bdev_ns_enable(ns);
589 if (ret == -ENOTBLK)
590 ret = nvmet_file_ns_enable(ns);
591 if (ret)
592 goto out_unlock;
593
594 ret = nvmet_p2pmem_ns_enable(ns);
595 if (ret)
596 goto out_dev_disable;
597
598 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
599 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
600
601 ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
602 0, GFP_KERNEL);
603 if (ret)
604 goto out_dev_put;
605
606 if (ns->nsid > subsys->max_nsid)
607 subsys->max_nsid = ns->nsid;
608
609 ret = xa_insert(&subsys->namespaces, ns->nsid, ns, GFP_KERNEL);
610 if (ret)
611 goto out_restore_subsys_maxnsid;
612
613 subsys->nr_namespaces++;
614
615 nvmet_ns_changed(subsys, ns->nsid);
616 ns->enabled = true;
617 ret = 0;
618out_unlock:
619 mutex_unlock(&subsys->lock);
620 return ret;
621
622out_restore_subsys_maxnsid:
623 subsys->max_nsid = nvmet_max_nsid(subsys);
624 percpu_ref_exit(&ns->ref);
625out_dev_put:
626 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
627 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
628out_dev_disable:
629 nvmet_ns_dev_disable(ns);
630 goto out_unlock;
631}
632
633void nvmet_ns_disable(struct nvmet_ns *ns)
634{
635 struct nvmet_subsys *subsys = ns->subsys;
636 struct nvmet_ctrl *ctrl;
637
638 mutex_lock(&subsys->lock);
639 if (!ns->enabled)
640 goto out_unlock;
641
642 ns->enabled = false;
643 xa_erase(&ns->subsys->namespaces, ns->nsid);
644 if (ns->nsid == subsys->max_nsid)
645 subsys->max_nsid = nvmet_max_nsid(subsys);
646
647 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
648 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
649
650 mutex_unlock(&subsys->lock);
651
652 /*
653 * Now that we removed the namespaces from the lookup list, we
654 * can kill the per_cpu ref and wait for any remaining references
655 * to be dropped, as well as a RCU grace period for anyone only
656 * using the namepace under rcu_read_lock(). Note that we can't
657 * use call_rcu here as we need to ensure the namespaces have
658 * been fully destroyed before unloading the module.
659 */
660 percpu_ref_kill(&ns->ref);
661 synchronize_rcu();
662 wait_for_completion(&ns->disable_done);
663 percpu_ref_exit(&ns->ref);
664
665 mutex_lock(&subsys->lock);
666
667 subsys->nr_namespaces--;
668 nvmet_ns_changed(subsys, ns->nsid);
669 nvmet_ns_dev_disable(ns);
670out_unlock:
671 mutex_unlock(&subsys->lock);
672}
673
674void nvmet_ns_free(struct nvmet_ns *ns)
675{
676 nvmet_ns_disable(ns);
677
678 down_write(&nvmet_ana_sem);
679 nvmet_ana_group_enabled[ns->anagrpid]--;
680 up_write(&nvmet_ana_sem);
681
682 kfree(ns->device_path);
683 kfree(ns);
684}
685
686struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
687{
688 struct nvmet_ns *ns;
689
690 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
691 if (!ns)
692 return NULL;
693
694 init_completion(&ns->disable_done);
695
696 ns->nsid = nsid;
697 ns->subsys = subsys;
698
699 down_write(&nvmet_ana_sem);
700 ns->anagrpid = NVMET_DEFAULT_ANA_GRPID;
701 nvmet_ana_group_enabled[ns->anagrpid]++;
702 up_write(&nvmet_ana_sem);
703
704 uuid_gen(&ns->uuid);
705 ns->buffered_io = false;
706 ns->csi = NVME_CSI_NVM;
707
708 return ns;
709}
710
711static void nvmet_update_sq_head(struct nvmet_req *req)
712{
713 if (req->sq->size) {
714 u32 old_sqhd, new_sqhd;
715
716 old_sqhd = READ_ONCE(req->sq->sqhd);
717 do {
718 new_sqhd = (old_sqhd + 1) % req->sq->size;
719 } while (!try_cmpxchg(&req->sq->sqhd, &old_sqhd, new_sqhd));
720 }
721 req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
722}
723
724static void nvmet_set_error(struct nvmet_req *req, u16 status)
725{
726 struct nvmet_ctrl *ctrl = req->sq->ctrl;
727 struct nvme_error_slot *new_error_slot;
728 unsigned long flags;
729
730 req->cqe->status = cpu_to_le16(status << 1);
731
732 if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC)
733 return;
734
735 spin_lock_irqsave(&ctrl->error_lock, flags);
736 ctrl->err_counter++;
737 new_error_slot =
738 &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS];
739
740 new_error_slot->error_count = cpu_to_le64(ctrl->err_counter);
741 new_error_slot->sqid = cpu_to_le16(req->sq->qid);
742 new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id);
743 new_error_slot->status_field = cpu_to_le16(status << 1);
744 new_error_slot->param_error_location = cpu_to_le16(req->error_loc);
745 new_error_slot->lba = cpu_to_le64(req->error_slba);
746 new_error_slot->nsid = req->cmd->common.nsid;
747 spin_unlock_irqrestore(&ctrl->error_lock, flags);
748
749 /* set the more bit for this request */
750 req->cqe->status |= cpu_to_le16(1 << 14);
751}
752
753static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
754{
755 struct nvmet_ns *ns = req->ns;
756
757 if (!req->sq->sqhd_disabled)
758 nvmet_update_sq_head(req);
759 req->cqe->sq_id = cpu_to_le16(req->sq->qid);
760 req->cqe->command_id = req->cmd->common.command_id;
761
762 if (unlikely(status))
763 nvmet_set_error(req, status);
764
765 trace_nvmet_req_complete(req);
766
767 req->ops->queue_response(req);
768 if (ns)
769 nvmet_put_namespace(ns);
770}
771
772void nvmet_req_complete(struct nvmet_req *req, u16 status)
773{
774 struct nvmet_sq *sq = req->sq;
775
776 __nvmet_req_complete(req, status);
777 percpu_ref_put(&sq->ref);
778}
779EXPORT_SYMBOL_GPL(nvmet_req_complete);
780
781void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq,
782 u16 qid, u16 size)
783{
784 cq->qid = qid;
785 cq->size = size;
786}
787
788void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq,
789 u16 qid, u16 size)
790{
791 sq->sqhd = 0;
792 sq->qid = qid;
793 sq->size = size;
794
795 ctrl->sqs[qid] = sq;
796}
797
798static void nvmet_confirm_sq(struct percpu_ref *ref)
799{
800 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
801
802 complete(&sq->confirm_done);
803}
804
805void nvmet_sq_destroy(struct nvmet_sq *sq)
806{
807 struct nvmet_ctrl *ctrl = sq->ctrl;
808
809 /*
810 * If this is the admin queue, complete all AERs so that our
811 * queue doesn't have outstanding requests on it.
812 */
813 if (ctrl && ctrl->sqs && ctrl->sqs[0] == sq)
814 nvmet_async_events_failall(ctrl);
815 percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq);
816 wait_for_completion(&sq->confirm_done);
817 wait_for_completion(&sq->free_done);
818 percpu_ref_exit(&sq->ref);
819 nvmet_auth_sq_free(sq);
820
821 if (ctrl) {
822 /*
823 * The teardown flow may take some time, and the host may not
824 * send us keep-alive during this period, hence reset the
825 * traffic based keep-alive timer so we don't trigger a
826 * controller teardown as a result of a keep-alive expiration.
827 */
828 ctrl->reset_tbkas = true;
829 sq->ctrl->sqs[sq->qid] = NULL;
830 nvmet_ctrl_put(ctrl);
831 sq->ctrl = NULL; /* allows reusing the queue later */
832 }
833}
834EXPORT_SYMBOL_GPL(nvmet_sq_destroy);
835
836static void nvmet_sq_free(struct percpu_ref *ref)
837{
838 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
839
840 complete(&sq->free_done);
841}
842
843int nvmet_sq_init(struct nvmet_sq *sq)
844{
845 int ret;
846
847 ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL);
848 if (ret) {
849 pr_err("percpu_ref init failed!\n");
850 return ret;
851 }
852 init_completion(&sq->free_done);
853 init_completion(&sq->confirm_done);
854 nvmet_auth_sq_init(sq);
855
856 return 0;
857}
858EXPORT_SYMBOL_GPL(nvmet_sq_init);
859
860static inline u16 nvmet_check_ana_state(struct nvmet_port *port,
861 struct nvmet_ns *ns)
862{
863 enum nvme_ana_state state = port->ana_state[ns->anagrpid];
864
865 if (unlikely(state == NVME_ANA_INACCESSIBLE))
866 return NVME_SC_ANA_INACCESSIBLE;
867 if (unlikely(state == NVME_ANA_PERSISTENT_LOSS))
868 return NVME_SC_ANA_PERSISTENT_LOSS;
869 if (unlikely(state == NVME_ANA_CHANGE))
870 return NVME_SC_ANA_TRANSITION;
871 return 0;
872}
873
874static inline u16 nvmet_io_cmd_check_access(struct nvmet_req *req)
875{
876 if (unlikely(req->ns->readonly)) {
877 switch (req->cmd->common.opcode) {
878 case nvme_cmd_read:
879 case nvme_cmd_flush:
880 break;
881 default:
882 return NVME_SC_NS_WRITE_PROTECTED;
883 }
884 }
885
886 return 0;
887}
888
889static u16 nvmet_parse_io_cmd(struct nvmet_req *req)
890{
891 struct nvme_command *cmd = req->cmd;
892 u16 ret;
893
894 if (nvme_is_fabrics(cmd))
895 return nvmet_parse_fabrics_io_cmd(req);
896
897 if (unlikely(!nvmet_check_auth_status(req)))
898 return NVME_SC_AUTH_REQUIRED | NVME_SC_DNR;
899
900 ret = nvmet_check_ctrl_status(req);
901 if (unlikely(ret))
902 return ret;
903
904 if (nvmet_is_passthru_req(req))
905 return nvmet_parse_passthru_io_cmd(req);
906
907 ret = nvmet_req_find_ns(req);
908 if (unlikely(ret))
909 return ret;
910
911 ret = nvmet_check_ana_state(req->port, req->ns);
912 if (unlikely(ret)) {
913 req->error_loc = offsetof(struct nvme_common_command, nsid);
914 return ret;
915 }
916 ret = nvmet_io_cmd_check_access(req);
917 if (unlikely(ret)) {
918 req->error_loc = offsetof(struct nvme_common_command, nsid);
919 return ret;
920 }
921
922 switch (req->ns->csi) {
923 case NVME_CSI_NVM:
924 if (req->ns->file)
925 return nvmet_file_parse_io_cmd(req);
926 return nvmet_bdev_parse_io_cmd(req);
927 case NVME_CSI_ZNS:
928 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED))
929 return nvmet_bdev_zns_parse_io_cmd(req);
930 return NVME_SC_INVALID_IO_CMD_SET;
931 default:
932 return NVME_SC_INVALID_IO_CMD_SET;
933 }
934}
935
936bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
937 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops)
938{
939 u8 flags = req->cmd->common.flags;
940 u16 status;
941
942 req->cq = cq;
943 req->sq = sq;
944 req->ops = ops;
945 req->sg = NULL;
946 req->metadata_sg = NULL;
947 req->sg_cnt = 0;
948 req->metadata_sg_cnt = 0;
949 req->transfer_len = 0;
950 req->metadata_len = 0;
951 req->cqe->status = 0;
952 req->cqe->sq_head = 0;
953 req->ns = NULL;
954 req->error_loc = NVMET_NO_ERROR_LOC;
955 req->error_slba = 0;
956
957 /* no support for fused commands yet */
958 if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) {
959 req->error_loc = offsetof(struct nvme_common_command, flags);
960 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
961 goto fail;
962 }
963
964 /*
965 * For fabrics, PSDT field shall describe metadata pointer (MPTR) that
966 * contains an address of a single contiguous physical buffer that is
967 * byte aligned.
968 */
969 if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) {
970 req->error_loc = offsetof(struct nvme_common_command, flags);
971 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
972 goto fail;
973 }
974
975 if (unlikely(!req->sq->ctrl))
976 /* will return an error for any non-connect command: */
977 status = nvmet_parse_connect_cmd(req);
978 else if (likely(req->sq->qid != 0))
979 status = nvmet_parse_io_cmd(req);
980 else
981 status = nvmet_parse_admin_cmd(req);
982
983 if (status)
984 goto fail;
985
986 trace_nvmet_req_init(req, req->cmd);
987
988 if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
989 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
990 goto fail;
991 }
992
993 if (sq->ctrl)
994 sq->ctrl->reset_tbkas = true;
995
996 return true;
997
998fail:
999 __nvmet_req_complete(req, status);
1000 return false;
1001}
1002EXPORT_SYMBOL_GPL(nvmet_req_init);
1003
1004void nvmet_req_uninit(struct nvmet_req *req)
1005{
1006 percpu_ref_put(&req->sq->ref);
1007 if (req->ns)
1008 nvmet_put_namespace(req->ns);
1009}
1010EXPORT_SYMBOL_GPL(nvmet_req_uninit);
1011
1012bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len)
1013{
1014 if (unlikely(len != req->transfer_len)) {
1015 req->error_loc = offsetof(struct nvme_common_command, dptr);
1016 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
1017 return false;
1018 }
1019
1020 return true;
1021}
1022EXPORT_SYMBOL_GPL(nvmet_check_transfer_len);
1023
1024bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len)
1025{
1026 if (unlikely(data_len > req->transfer_len)) {
1027 req->error_loc = offsetof(struct nvme_common_command, dptr);
1028 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
1029 return false;
1030 }
1031
1032 return true;
1033}
1034
1035static unsigned int nvmet_data_transfer_len(struct nvmet_req *req)
1036{
1037 return req->transfer_len - req->metadata_len;
1038}
1039
1040static int nvmet_req_alloc_p2pmem_sgls(struct pci_dev *p2p_dev,
1041 struct nvmet_req *req)
1042{
1043 req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt,
1044 nvmet_data_transfer_len(req));
1045 if (!req->sg)
1046 goto out_err;
1047
1048 if (req->metadata_len) {
1049 req->metadata_sg = pci_p2pmem_alloc_sgl(p2p_dev,
1050 &req->metadata_sg_cnt, req->metadata_len);
1051 if (!req->metadata_sg)
1052 goto out_free_sg;
1053 }
1054
1055 req->p2p_dev = p2p_dev;
1056
1057 return 0;
1058out_free_sg:
1059 pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1060out_err:
1061 return -ENOMEM;
1062}
1063
1064static struct pci_dev *nvmet_req_find_p2p_dev(struct nvmet_req *req)
1065{
1066 if (!IS_ENABLED(CONFIG_PCI_P2PDMA) ||
1067 !req->sq->ctrl || !req->sq->qid || !req->ns)
1068 return NULL;
1069 return radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
1070}
1071
1072int nvmet_req_alloc_sgls(struct nvmet_req *req)
1073{
1074 struct pci_dev *p2p_dev = nvmet_req_find_p2p_dev(req);
1075
1076 if (p2p_dev && !nvmet_req_alloc_p2pmem_sgls(p2p_dev, req))
1077 return 0;
1078
1079 req->sg = sgl_alloc(nvmet_data_transfer_len(req), GFP_KERNEL,
1080 &req->sg_cnt);
1081 if (unlikely(!req->sg))
1082 goto out;
1083
1084 if (req->metadata_len) {
1085 req->metadata_sg = sgl_alloc(req->metadata_len, GFP_KERNEL,
1086 &req->metadata_sg_cnt);
1087 if (unlikely(!req->metadata_sg))
1088 goto out_free;
1089 }
1090
1091 return 0;
1092out_free:
1093 sgl_free(req->sg);
1094out:
1095 return -ENOMEM;
1096}
1097EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgls);
1098
1099void nvmet_req_free_sgls(struct nvmet_req *req)
1100{
1101 if (req->p2p_dev) {
1102 pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1103 if (req->metadata_sg)
1104 pci_p2pmem_free_sgl(req->p2p_dev, req->metadata_sg);
1105 req->p2p_dev = NULL;
1106 } else {
1107 sgl_free(req->sg);
1108 if (req->metadata_sg)
1109 sgl_free(req->metadata_sg);
1110 }
1111
1112 req->sg = NULL;
1113 req->metadata_sg = NULL;
1114 req->sg_cnt = 0;
1115 req->metadata_sg_cnt = 0;
1116}
1117EXPORT_SYMBOL_GPL(nvmet_req_free_sgls);
1118
1119static inline bool nvmet_cc_en(u32 cc)
1120{
1121 return (cc >> NVME_CC_EN_SHIFT) & 0x1;
1122}
1123
1124static inline u8 nvmet_cc_css(u32 cc)
1125{
1126 return (cc >> NVME_CC_CSS_SHIFT) & 0x7;
1127}
1128
1129static inline u8 nvmet_cc_mps(u32 cc)
1130{
1131 return (cc >> NVME_CC_MPS_SHIFT) & 0xf;
1132}
1133
1134static inline u8 nvmet_cc_ams(u32 cc)
1135{
1136 return (cc >> NVME_CC_AMS_SHIFT) & 0x7;
1137}
1138
1139static inline u8 nvmet_cc_shn(u32 cc)
1140{
1141 return (cc >> NVME_CC_SHN_SHIFT) & 0x3;
1142}
1143
1144static inline u8 nvmet_cc_iosqes(u32 cc)
1145{
1146 return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf;
1147}
1148
1149static inline u8 nvmet_cc_iocqes(u32 cc)
1150{
1151 return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf;
1152}
1153
1154static inline bool nvmet_css_supported(u8 cc_css)
1155{
1156 switch (cc_css << NVME_CC_CSS_SHIFT) {
1157 case NVME_CC_CSS_NVM:
1158 case NVME_CC_CSS_CSI:
1159 return true;
1160 default:
1161 return false;
1162 }
1163}
1164
1165static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl)
1166{
1167 lockdep_assert_held(&ctrl->lock);
1168
1169 /*
1170 * Only I/O controllers should verify iosqes,iocqes.
1171 * Strictly speaking, the spec says a discovery controller
1172 * should verify iosqes,iocqes are zeroed, however that
1173 * would break backwards compatibility, so don't enforce it.
1174 */
1175 if (!nvmet_is_disc_subsys(ctrl->subsys) &&
1176 (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES ||
1177 nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES)) {
1178 ctrl->csts = NVME_CSTS_CFS;
1179 return;
1180 }
1181
1182 if (nvmet_cc_mps(ctrl->cc) != 0 ||
1183 nvmet_cc_ams(ctrl->cc) != 0 ||
1184 !nvmet_css_supported(nvmet_cc_css(ctrl->cc))) {
1185 ctrl->csts = NVME_CSTS_CFS;
1186 return;
1187 }
1188
1189 ctrl->csts = NVME_CSTS_RDY;
1190
1191 /*
1192 * Controllers that are not yet enabled should not really enforce the
1193 * keep alive timeout, but we still want to track a timeout and cleanup
1194 * in case a host died before it enabled the controller. Hence, simply
1195 * reset the keep alive timer when the controller is enabled.
1196 */
1197 if (ctrl->kato)
1198 mod_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
1199}
1200
1201static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl)
1202{
1203 lockdep_assert_held(&ctrl->lock);
1204
1205 /* XXX: tear down queues? */
1206 ctrl->csts &= ~NVME_CSTS_RDY;
1207 ctrl->cc = 0;
1208}
1209
1210void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new)
1211{
1212 u32 old;
1213
1214 mutex_lock(&ctrl->lock);
1215 old = ctrl->cc;
1216 ctrl->cc = new;
1217
1218 if (nvmet_cc_en(new) && !nvmet_cc_en(old))
1219 nvmet_start_ctrl(ctrl);
1220 if (!nvmet_cc_en(new) && nvmet_cc_en(old))
1221 nvmet_clear_ctrl(ctrl);
1222 if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) {
1223 nvmet_clear_ctrl(ctrl);
1224 ctrl->csts |= NVME_CSTS_SHST_CMPLT;
1225 }
1226 if (!nvmet_cc_shn(new) && nvmet_cc_shn(old))
1227 ctrl->csts &= ~NVME_CSTS_SHST_CMPLT;
1228 mutex_unlock(&ctrl->lock);
1229}
1230
1231static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
1232{
1233 /* command sets supported: NVMe command set: */
1234 ctrl->cap = (1ULL << 37);
1235 /* Controller supports one or more I/O Command Sets */
1236 ctrl->cap |= (1ULL << 43);
1237 /* CC.EN timeout in 500msec units: */
1238 ctrl->cap |= (15ULL << 24);
1239 /* maximum queue entries supported: */
1240 if (ctrl->ops->get_max_queue_size)
1241 ctrl->cap |= min_t(u16, ctrl->ops->get_max_queue_size(ctrl),
1242 ctrl->port->max_queue_size) - 1;
1243 else
1244 ctrl->cap |= ctrl->port->max_queue_size - 1;
1245
1246 if (nvmet_is_passthru_subsys(ctrl->subsys))
1247 nvmet_passthrough_override_cap(ctrl);
1248}
1249
1250struct nvmet_ctrl *nvmet_ctrl_find_get(const char *subsysnqn,
1251 const char *hostnqn, u16 cntlid,
1252 struct nvmet_req *req)
1253{
1254 struct nvmet_ctrl *ctrl = NULL;
1255 struct nvmet_subsys *subsys;
1256
1257 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1258 if (!subsys) {
1259 pr_warn("connect request for invalid subsystem %s!\n",
1260 subsysnqn);
1261 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1262 goto out;
1263 }
1264
1265 mutex_lock(&subsys->lock);
1266 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
1267 if (ctrl->cntlid == cntlid) {
1268 if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) {
1269 pr_warn("hostnqn mismatch.\n");
1270 continue;
1271 }
1272 if (!kref_get_unless_zero(&ctrl->ref))
1273 continue;
1274
1275 /* ctrl found */
1276 goto found;
1277 }
1278 }
1279
1280 ctrl = NULL; /* ctrl not found */
1281 pr_warn("could not find controller %d for subsys %s / host %s\n",
1282 cntlid, subsysnqn, hostnqn);
1283 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
1284
1285found:
1286 mutex_unlock(&subsys->lock);
1287 nvmet_subsys_put(subsys);
1288out:
1289 return ctrl;
1290}
1291
1292u16 nvmet_check_ctrl_status(struct nvmet_req *req)
1293{
1294 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
1295 pr_err("got cmd %d while CC.EN == 0 on qid = %d\n",
1296 req->cmd->common.opcode, req->sq->qid);
1297 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1298 }
1299
1300 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
1301 pr_err("got cmd %d while CSTS.RDY == 0 on qid = %d\n",
1302 req->cmd->common.opcode, req->sq->qid);
1303 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1304 }
1305
1306 if (unlikely(!nvmet_check_auth_status(req))) {
1307 pr_warn("qid %d not authenticated\n", req->sq->qid);
1308 return NVME_SC_AUTH_REQUIRED | NVME_SC_DNR;
1309 }
1310 return 0;
1311}
1312
1313bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn)
1314{
1315 struct nvmet_host_link *p;
1316
1317 lockdep_assert_held(&nvmet_config_sem);
1318
1319 if (subsys->allow_any_host)
1320 return true;
1321
1322 if (nvmet_is_disc_subsys(subsys)) /* allow all access to disc subsys */
1323 return true;
1324
1325 list_for_each_entry(p, &subsys->hosts, entry) {
1326 if (!strcmp(nvmet_host_name(p->host), hostnqn))
1327 return true;
1328 }
1329
1330 return false;
1331}
1332
1333/*
1334 * Note: ctrl->subsys->lock should be held when calling this function
1335 */
1336static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl,
1337 struct nvmet_req *req)
1338{
1339 struct nvmet_ns *ns;
1340 unsigned long idx;
1341
1342 if (!req->p2p_client)
1343 return;
1344
1345 ctrl->p2p_client = get_device(req->p2p_client);
1346
1347 xa_for_each(&ctrl->subsys->namespaces, idx, ns)
1348 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
1349}
1350
1351/*
1352 * Note: ctrl->subsys->lock should be held when calling this function
1353 */
1354static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl)
1355{
1356 struct radix_tree_iter iter;
1357 void __rcu **slot;
1358
1359 radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0)
1360 pci_dev_put(radix_tree_deref_slot(slot));
1361
1362 put_device(ctrl->p2p_client);
1363}
1364
1365static void nvmet_fatal_error_handler(struct work_struct *work)
1366{
1367 struct nvmet_ctrl *ctrl =
1368 container_of(work, struct nvmet_ctrl, fatal_err_work);
1369
1370 pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid);
1371 ctrl->ops->delete_ctrl(ctrl);
1372}
1373
1374u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
1375 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp)
1376{
1377 struct nvmet_subsys *subsys;
1378 struct nvmet_ctrl *ctrl;
1379 int ret;
1380 u16 status;
1381
1382 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1383 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1384 if (!subsys) {
1385 pr_warn("connect request for invalid subsystem %s!\n",
1386 subsysnqn);
1387 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1388 req->error_loc = offsetof(struct nvme_common_command, dptr);
1389 goto out;
1390 }
1391
1392 down_read(&nvmet_config_sem);
1393 if (!nvmet_host_allowed(subsys, hostnqn)) {
1394 pr_info("connect by host %s for subsystem %s not allowed\n",
1395 hostnqn, subsysnqn);
1396 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
1397 up_read(&nvmet_config_sem);
1398 status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
1399 req->error_loc = offsetof(struct nvme_common_command, dptr);
1400 goto out_put_subsystem;
1401 }
1402 up_read(&nvmet_config_sem);
1403
1404 status = NVME_SC_INTERNAL;
1405 ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
1406 if (!ctrl)
1407 goto out_put_subsystem;
1408 mutex_init(&ctrl->lock);
1409
1410 ctrl->port = req->port;
1411 ctrl->ops = req->ops;
1412
1413#ifdef CONFIG_NVME_TARGET_PASSTHRU
1414 /* By default, set loop targets to clear IDS by default */
1415 if (ctrl->port->disc_addr.trtype == NVMF_TRTYPE_LOOP)
1416 subsys->clear_ids = 1;
1417#endif
1418
1419 INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
1420 INIT_LIST_HEAD(&ctrl->async_events);
1421 INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL);
1422 INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
1423 INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer);
1424
1425 memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
1426 memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE);
1427
1428 kref_init(&ctrl->ref);
1429 ctrl->subsys = subsys;
1430 ctrl->pi_support = ctrl->port->pi_enable && ctrl->subsys->pi_support;
1431 nvmet_init_cap(ctrl);
1432 WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL);
1433
1434 ctrl->changed_ns_list = kmalloc_array(NVME_MAX_CHANGED_NAMESPACES,
1435 sizeof(__le32), GFP_KERNEL);
1436 if (!ctrl->changed_ns_list)
1437 goto out_free_ctrl;
1438
1439 ctrl->sqs = kcalloc(subsys->max_qid + 1,
1440 sizeof(struct nvmet_sq *),
1441 GFP_KERNEL);
1442 if (!ctrl->sqs)
1443 goto out_free_changed_ns_list;
1444
1445 ret = ida_alloc_range(&cntlid_ida,
1446 subsys->cntlid_min, subsys->cntlid_max,
1447 GFP_KERNEL);
1448 if (ret < 0) {
1449 status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR;
1450 goto out_free_sqs;
1451 }
1452 ctrl->cntlid = ret;
1453
1454 /*
1455 * Discovery controllers may use some arbitrary high value
1456 * in order to cleanup stale discovery sessions
1457 */
1458 if (nvmet_is_disc_subsys(ctrl->subsys) && !kato)
1459 kato = NVMET_DISC_KATO_MS;
1460
1461 /* keep-alive timeout in seconds */
1462 ctrl->kato = DIV_ROUND_UP(kato, 1000);
1463
1464 ctrl->err_counter = 0;
1465 spin_lock_init(&ctrl->error_lock);
1466
1467 nvmet_start_keep_alive_timer(ctrl);
1468
1469 mutex_lock(&subsys->lock);
1470 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
1471 nvmet_setup_p2p_ns_map(ctrl, req);
1472 mutex_unlock(&subsys->lock);
1473
1474 *ctrlp = ctrl;
1475 return 0;
1476
1477out_free_sqs:
1478 kfree(ctrl->sqs);
1479out_free_changed_ns_list:
1480 kfree(ctrl->changed_ns_list);
1481out_free_ctrl:
1482 kfree(ctrl);
1483out_put_subsystem:
1484 nvmet_subsys_put(subsys);
1485out:
1486 return status;
1487}
1488
1489static void nvmet_ctrl_free(struct kref *ref)
1490{
1491 struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref);
1492 struct nvmet_subsys *subsys = ctrl->subsys;
1493
1494 mutex_lock(&subsys->lock);
1495 nvmet_release_p2p_ns_map(ctrl);
1496 list_del(&ctrl->subsys_entry);
1497 mutex_unlock(&subsys->lock);
1498
1499 nvmet_stop_keep_alive_timer(ctrl);
1500
1501 flush_work(&ctrl->async_event_work);
1502 cancel_work_sync(&ctrl->fatal_err_work);
1503
1504 nvmet_destroy_auth(ctrl);
1505
1506 ida_free(&cntlid_ida, ctrl->cntlid);
1507
1508 nvmet_async_events_free(ctrl);
1509 kfree(ctrl->sqs);
1510 kfree(ctrl->changed_ns_list);
1511 kfree(ctrl);
1512
1513 nvmet_subsys_put(subsys);
1514}
1515
1516void nvmet_ctrl_put(struct nvmet_ctrl *ctrl)
1517{
1518 kref_put(&ctrl->ref, nvmet_ctrl_free);
1519}
1520
1521void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl)
1522{
1523 mutex_lock(&ctrl->lock);
1524 if (!(ctrl->csts & NVME_CSTS_CFS)) {
1525 ctrl->csts |= NVME_CSTS_CFS;
1526 queue_work(nvmet_wq, &ctrl->fatal_err_work);
1527 }
1528 mutex_unlock(&ctrl->lock);
1529}
1530EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error);
1531
1532static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
1533 const char *subsysnqn)
1534{
1535 struct nvmet_subsys_link *p;
1536
1537 if (!port)
1538 return NULL;
1539
1540 if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) {
1541 if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
1542 return NULL;
1543 return nvmet_disc_subsys;
1544 }
1545
1546 down_read(&nvmet_config_sem);
1547 if (!strncmp(nvmet_disc_subsys->subsysnqn, subsysnqn,
1548 NVMF_NQN_SIZE)) {
1549 if (kref_get_unless_zero(&nvmet_disc_subsys->ref)) {
1550 up_read(&nvmet_config_sem);
1551 return nvmet_disc_subsys;
1552 }
1553 }
1554 list_for_each_entry(p, &port->subsystems, entry) {
1555 if (!strncmp(p->subsys->subsysnqn, subsysnqn,
1556 NVMF_NQN_SIZE)) {
1557 if (!kref_get_unless_zero(&p->subsys->ref))
1558 break;
1559 up_read(&nvmet_config_sem);
1560 return p->subsys;
1561 }
1562 }
1563 up_read(&nvmet_config_sem);
1564 return NULL;
1565}
1566
1567struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
1568 enum nvme_subsys_type type)
1569{
1570 struct nvmet_subsys *subsys;
1571 char serial[NVMET_SN_MAX_SIZE / 2];
1572 int ret;
1573
1574 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
1575 if (!subsys)
1576 return ERR_PTR(-ENOMEM);
1577
1578 subsys->ver = NVMET_DEFAULT_VS;
1579 /* generate a random serial number as our controllers are ephemeral: */
1580 get_random_bytes(&serial, sizeof(serial));
1581 bin2hex(subsys->serial, &serial, sizeof(serial));
1582
1583 subsys->model_number = kstrdup(NVMET_DEFAULT_CTRL_MODEL, GFP_KERNEL);
1584 if (!subsys->model_number) {
1585 ret = -ENOMEM;
1586 goto free_subsys;
1587 }
1588
1589 subsys->ieee_oui = 0;
1590
1591 subsys->firmware_rev = kstrndup(UTS_RELEASE, NVMET_FR_MAX_SIZE, GFP_KERNEL);
1592 if (!subsys->firmware_rev) {
1593 ret = -ENOMEM;
1594 goto free_mn;
1595 }
1596
1597 switch (type) {
1598 case NVME_NQN_NVME:
1599 subsys->max_qid = NVMET_NR_QUEUES;
1600 break;
1601 case NVME_NQN_DISC:
1602 case NVME_NQN_CURR:
1603 subsys->max_qid = 0;
1604 break;
1605 default:
1606 pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
1607 ret = -EINVAL;
1608 goto free_fr;
1609 }
1610 subsys->type = type;
1611 subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
1612 GFP_KERNEL);
1613 if (!subsys->subsysnqn) {
1614 ret = -ENOMEM;
1615 goto free_fr;
1616 }
1617 subsys->cntlid_min = NVME_CNTLID_MIN;
1618 subsys->cntlid_max = NVME_CNTLID_MAX;
1619 kref_init(&subsys->ref);
1620
1621 mutex_init(&subsys->lock);
1622 xa_init(&subsys->namespaces);
1623 INIT_LIST_HEAD(&subsys->ctrls);
1624 INIT_LIST_HEAD(&subsys->hosts);
1625
1626 return subsys;
1627
1628free_fr:
1629 kfree(subsys->firmware_rev);
1630free_mn:
1631 kfree(subsys->model_number);
1632free_subsys:
1633 kfree(subsys);
1634 return ERR_PTR(ret);
1635}
1636
1637static void nvmet_subsys_free(struct kref *ref)
1638{
1639 struct nvmet_subsys *subsys =
1640 container_of(ref, struct nvmet_subsys, ref);
1641
1642 WARN_ON_ONCE(!xa_empty(&subsys->namespaces));
1643
1644 xa_destroy(&subsys->namespaces);
1645 nvmet_passthru_subsys_free(subsys);
1646
1647 kfree(subsys->subsysnqn);
1648 kfree(subsys->model_number);
1649 kfree(subsys->firmware_rev);
1650 kfree(subsys);
1651}
1652
1653void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys)
1654{
1655 struct nvmet_ctrl *ctrl;
1656
1657 mutex_lock(&subsys->lock);
1658 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
1659 ctrl->ops->delete_ctrl(ctrl);
1660 mutex_unlock(&subsys->lock);
1661}
1662
1663void nvmet_subsys_put(struct nvmet_subsys *subsys)
1664{
1665 kref_put(&subsys->ref, nvmet_subsys_free);
1666}
1667
1668static int __init nvmet_init(void)
1669{
1670 int error = -ENOMEM;
1671
1672 nvmet_ana_group_enabled[NVMET_DEFAULT_ANA_GRPID] = 1;
1673
1674 nvmet_bvec_cache = kmem_cache_create("nvmet-bvec",
1675 NVMET_MAX_MPOOL_BVEC * sizeof(struct bio_vec), 0,
1676 SLAB_HWCACHE_ALIGN, NULL);
1677 if (!nvmet_bvec_cache)
1678 return -ENOMEM;
1679
1680 zbd_wq = alloc_workqueue("nvmet-zbd-wq", WQ_MEM_RECLAIM, 0);
1681 if (!zbd_wq)
1682 goto out_destroy_bvec_cache;
1683
1684 buffered_io_wq = alloc_workqueue("nvmet-buffered-io-wq",
1685 WQ_MEM_RECLAIM, 0);
1686 if (!buffered_io_wq)
1687 goto out_free_zbd_work_queue;
1688
1689 nvmet_wq = alloc_workqueue("nvmet-wq",
1690 WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1691 if (!nvmet_wq)
1692 goto out_free_buffered_work_queue;
1693
1694 error = nvmet_init_discovery();
1695 if (error)
1696 goto out_free_nvmet_work_queue;
1697
1698 error = nvmet_init_configfs();
1699 if (error)
1700 goto out_exit_discovery;
1701 return 0;
1702
1703out_exit_discovery:
1704 nvmet_exit_discovery();
1705out_free_nvmet_work_queue:
1706 destroy_workqueue(nvmet_wq);
1707out_free_buffered_work_queue:
1708 destroy_workqueue(buffered_io_wq);
1709out_free_zbd_work_queue:
1710 destroy_workqueue(zbd_wq);
1711out_destroy_bvec_cache:
1712 kmem_cache_destroy(nvmet_bvec_cache);
1713 return error;
1714}
1715
1716static void __exit nvmet_exit(void)
1717{
1718 nvmet_exit_configfs();
1719 nvmet_exit_discovery();
1720 ida_destroy(&cntlid_ida);
1721 destroy_workqueue(nvmet_wq);
1722 destroy_workqueue(buffered_io_wq);
1723 destroy_workqueue(zbd_wq);
1724 kmem_cache_destroy(nvmet_bvec_cache);
1725
1726 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024);
1727 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024);
1728}
1729
1730module_init(nvmet_init);
1731module_exit(nvmet_exit);
1732
1733MODULE_DESCRIPTION("NVMe target core framework");
1734MODULE_LICENSE("GPL v2");