2 * Common code for the NVMe target.
3 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
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.
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
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/module.h>
16 #include <linux/random.h>
17 #include <linux/rculist.h>
21 static struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
22 static DEFINE_IDA(cntlid_ida);
25 * This read/write semaphore is used to synchronize access to configuration
26 * information on a target system that will result in discovery log page
27 * information change for at least one host.
28 * The full list of resources to protected by this semaphore is:
31 * - per-subsystem allowed hosts list
32 * - allow_any_host subsystem attribute
34 * - the nvmet_transports array
36 * When updating any of those lists/structures write lock should be obtained,
37 * while when reading (popolating discovery log page or checking host-subsystem
38 * link) read lock is obtained to allow concurrent reads.
40 DECLARE_RWSEM(nvmet_config_sem);
42 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
43 const char *subsysnqn);
45 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
48 if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len)
49 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
53 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len)
55 if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len)
56 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
60 static unsigned int nvmet_max_nsid(struct nvmet_subsys *subsys)
64 if (list_empty(&subsys->namespaces))
67 ns = list_last_entry(&subsys->namespaces, struct nvmet_ns, dev_link);
71 static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
73 return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16);
76 static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
78 struct nvmet_req *req;
81 mutex_lock(&ctrl->lock);
82 if (!ctrl->nr_async_event_cmds) {
83 mutex_unlock(&ctrl->lock);
87 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
88 mutex_unlock(&ctrl->lock);
89 nvmet_req_complete(req, NVME_SC_INTERNAL | NVME_SC_DNR);
93 static void nvmet_async_event_work(struct work_struct *work)
95 struct nvmet_ctrl *ctrl =
96 container_of(work, struct nvmet_ctrl, async_event_work);
97 struct nvmet_async_event *aen;
98 struct nvmet_req *req;
101 mutex_lock(&ctrl->lock);
102 aen = list_first_entry_or_null(&ctrl->async_events,
103 struct nvmet_async_event, entry);
104 if (!aen || !ctrl->nr_async_event_cmds) {
105 mutex_unlock(&ctrl->lock);
109 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
110 nvmet_set_result(req, nvmet_async_event_result(aen));
112 list_del(&aen->entry);
115 mutex_unlock(&ctrl->lock);
116 nvmet_req_complete(req, 0);
120 static void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
121 u8 event_info, u8 log_page)
123 struct nvmet_async_event *aen;
125 aen = kmalloc(sizeof(*aen), GFP_KERNEL);
129 aen->event_type = event_type;
130 aen->event_info = event_info;
131 aen->log_page = log_page;
133 mutex_lock(&ctrl->lock);
134 list_add_tail(&aen->entry, &ctrl->async_events);
135 mutex_unlock(&ctrl->lock);
137 schedule_work(&ctrl->async_event_work);
140 int nvmet_register_transport(struct nvmet_fabrics_ops *ops)
144 down_write(&nvmet_config_sem);
145 if (nvmet_transports[ops->type])
148 nvmet_transports[ops->type] = ops;
149 up_write(&nvmet_config_sem);
153 EXPORT_SYMBOL_GPL(nvmet_register_transport);
155 void nvmet_unregister_transport(struct nvmet_fabrics_ops *ops)
157 down_write(&nvmet_config_sem);
158 nvmet_transports[ops->type] = NULL;
159 up_write(&nvmet_config_sem);
161 EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
163 int nvmet_enable_port(struct nvmet_port *port)
165 struct nvmet_fabrics_ops *ops;
168 lockdep_assert_held(&nvmet_config_sem);
170 ops = nvmet_transports[port->disc_addr.trtype];
172 up_write(&nvmet_config_sem);
173 request_module("nvmet-transport-%d", port->disc_addr.trtype);
174 down_write(&nvmet_config_sem);
175 ops = nvmet_transports[port->disc_addr.trtype];
177 pr_err("transport type %d not supported\n",
178 port->disc_addr.trtype);
183 if (!try_module_get(ops->owner))
186 ret = ops->add_port(port);
188 module_put(ops->owner);
192 port->enabled = true;
196 void nvmet_disable_port(struct nvmet_port *port)
198 struct nvmet_fabrics_ops *ops;
200 lockdep_assert_held(&nvmet_config_sem);
202 port->enabled = false;
204 ops = nvmet_transports[port->disc_addr.trtype];
205 ops->remove_port(port);
206 module_put(ops->owner);
209 static void nvmet_keep_alive_timer(struct work_struct *work)
211 struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
212 struct nvmet_ctrl, ka_work);
214 pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
215 ctrl->cntlid, ctrl->kato);
217 nvmet_ctrl_fatal_error(ctrl);
220 static void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
222 pr_debug("ctrl %d start keep-alive timer for %d secs\n",
223 ctrl->cntlid, ctrl->kato);
225 INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer);
226 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
229 static void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
231 pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid);
233 cancel_delayed_work_sync(&ctrl->ka_work);
236 static struct nvmet_ns *__nvmet_find_namespace(struct nvmet_ctrl *ctrl,
241 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
242 if (ns->nsid == le32_to_cpu(nsid))
249 struct nvmet_ns *nvmet_find_namespace(struct nvmet_ctrl *ctrl, __le32 nsid)
254 ns = __nvmet_find_namespace(ctrl, nsid);
256 percpu_ref_get(&ns->ref);
262 static void nvmet_destroy_namespace(struct percpu_ref *ref)
264 struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref);
266 complete(&ns->disable_done);
269 void nvmet_put_namespace(struct nvmet_ns *ns)
271 percpu_ref_put(&ns->ref);
274 int nvmet_ns_enable(struct nvmet_ns *ns)
276 struct nvmet_subsys *subsys = ns->subsys;
277 struct nvmet_ctrl *ctrl;
280 mutex_lock(&subsys->lock);
284 ns->bdev = blkdev_get_by_path(ns->device_path, FMODE_READ | FMODE_WRITE,
286 if (IS_ERR(ns->bdev)) {
287 pr_err("failed to open block device %s: (%ld)\n",
288 ns->device_path, PTR_ERR(ns->bdev));
289 ret = PTR_ERR(ns->bdev);
294 ns->size = i_size_read(ns->bdev->bd_inode);
295 ns->blksize_shift = blksize_bits(bdev_logical_block_size(ns->bdev));
297 ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
302 if (ns->nsid > subsys->max_nsid)
303 subsys->max_nsid = ns->nsid;
306 * The namespaces list needs to be sorted to simplify the implementation
307 * of the Identify Namepace List subcommand.
309 if (list_empty(&subsys->namespaces)) {
310 list_add_tail_rcu(&ns->dev_link, &subsys->namespaces);
312 struct nvmet_ns *old;
314 list_for_each_entry_rcu(old, &subsys->namespaces, dev_link) {
315 BUG_ON(ns->nsid == old->nsid);
316 if (ns->nsid < old->nsid)
320 list_add_tail_rcu(&ns->dev_link, &old->dev_link);
323 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
324 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, 0, 0);
329 mutex_unlock(&subsys->lock);
332 blkdev_put(ns->bdev, FMODE_WRITE|FMODE_READ);
337 void nvmet_ns_disable(struct nvmet_ns *ns)
339 struct nvmet_subsys *subsys = ns->subsys;
340 struct nvmet_ctrl *ctrl;
342 mutex_lock(&subsys->lock);
347 list_del_rcu(&ns->dev_link);
348 if (ns->nsid == subsys->max_nsid)
349 subsys->max_nsid = nvmet_max_nsid(subsys);
350 mutex_unlock(&subsys->lock);
353 * Now that we removed the namespaces from the lookup list, we
354 * can kill the per_cpu ref and wait for any remaining references
355 * to be dropped, as well as a RCU grace period for anyone only
356 * using the namepace under rcu_read_lock(). Note that we can't
357 * use call_rcu here as we need to ensure the namespaces have
358 * been fully destroyed before unloading the module.
360 percpu_ref_kill(&ns->ref);
362 wait_for_completion(&ns->disable_done);
363 percpu_ref_exit(&ns->ref);
365 mutex_lock(&subsys->lock);
366 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
367 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, 0, 0);
370 blkdev_put(ns->bdev, FMODE_WRITE|FMODE_READ);
372 mutex_unlock(&subsys->lock);
375 void nvmet_ns_free(struct nvmet_ns *ns)
377 nvmet_ns_disable(ns);
379 kfree(ns->device_path);
383 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
387 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
391 INIT_LIST_HEAD(&ns->dev_link);
392 init_completion(&ns->disable_done);
401 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
403 u32 old_sqhd, new_sqhd;
407 nvmet_set_status(req, status);
411 old_sqhd = req->sq->sqhd;
412 new_sqhd = (old_sqhd + 1) % req->sq->size;
413 } while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
416 sqhd = req->sq->sqhd & 0x0000FFFF;
417 req->rsp->sq_head = cpu_to_le16(sqhd);
418 req->rsp->sq_id = cpu_to_le16(req->sq->qid);
419 req->rsp->command_id = req->cmd->common.command_id;
422 nvmet_put_namespace(req->ns);
423 req->ops->queue_response(req);
426 void nvmet_req_complete(struct nvmet_req *req, u16 status)
428 __nvmet_req_complete(req, status);
429 percpu_ref_put(&req->sq->ref);
431 EXPORT_SYMBOL_GPL(nvmet_req_complete);
433 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq,
442 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq,
452 static void nvmet_confirm_sq(struct percpu_ref *ref)
454 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
456 complete(&sq->confirm_done);
459 void nvmet_sq_destroy(struct nvmet_sq *sq)
462 * If this is the admin queue, complete all AERs so that our
463 * queue doesn't have outstanding requests on it.
465 if (sq->ctrl && sq->ctrl->sqs && sq->ctrl->sqs[0] == sq)
466 nvmet_async_events_free(sq->ctrl);
467 percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq);
468 wait_for_completion(&sq->confirm_done);
469 wait_for_completion(&sq->free_done);
470 percpu_ref_exit(&sq->ref);
473 nvmet_ctrl_put(sq->ctrl);
474 sq->ctrl = NULL; /* allows reusing the queue later */
477 EXPORT_SYMBOL_GPL(nvmet_sq_destroy);
479 static void nvmet_sq_free(struct percpu_ref *ref)
481 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
483 complete(&sq->free_done);
486 int nvmet_sq_init(struct nvmet_sq *sq)
490 ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL);
492 pr_err("percpu_ref init failed!\n");
495 init_completion(&sq->free_done);
496 init_completion(&sq->confirm_done);
500 EXPORT_SYMBOL_GPL(nvmet_sq_init);
502 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
503 struct nvmet_sq *sq, struct nvmet_fabrics_ops *ops)
505 u8 flags = req->cmd->common.flags;
513 req->transfer_len = 0;
514 req->rsp->status = 0;
517 /* no support for fused commands yet */
518 if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) {
519 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
523 /* either variant of SGLs is fine, as we don't support metadata */
524 if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF &&
525 (flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METASEG)) {
526 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
530 if (unlikely(!req->sq->ctrl))
531 /* will return an error for any Non-connect command: */
532 status = nvmet_parse_connect_cmd(req);
533 else if (likely(req->sq->qid != 0))
534 status = nvmet_parse_io_cmd(req);
535 else if (req->cmd->common.opcode == nvme_fabrics_command)
536 status = nvmet_parse_fabrics_cmd(req);
537 else if (req->sq->ctrl->subsys->type == NVME_NQN_DISC)
538 status = nvmet_parse_discovery_cmd(req);
540 status = nvmet_parse_admin_cmd(req);
545 if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
546 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
553 __nvmet_req_complete(req, status);
556 EXPORT_SYMBOL_GPL(nvmet_req_init);
558 void nvmet_req_uninit(struct nvmet_req *req)
560 percpu_ref_put(&req->sq->ref);
562 nvmet_put_namespace(req->ns);
564 EXPORT_SYMBOL_GPL(nvmet_req_uninit);
566 void nvmet_req_execute(struct nvmet_req *req)
568 if (unlikely(req->data_len != req->transfer_len))
569 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
573 EXPORT_SYMBOL_GPL(nvmet_req_execute);
575 static inline bool nvmet_cc_en(u32 cc)
577 return (cc >> NVME_CC_EN_SHIFT) & 0x1;
580 static inline u8 nvmet_cc_css(u32 cc)
582 return (cc >> NVME_CC_CSS_SHIFT) & 0x7;
585 static inline u8 nvmet_cc_mps(u32 cc)
587 return (cc >> NVME_CC_MPS_SHIFT) & 0xf;
590 static inline u8 nvmet_cc_ams(u32 cc)
592 return (cc >> NVME_CC_AMS_SHIFT) & 0x7;
595 static inline u8 nvmet_cc_shn(u32 cc)
597 return (cc >> NVME_CC_SHN_SHIFT) & 0x3;
600 static inline u8 nvmet_cc_iosqes(u32 cc)
602 return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf;
605 static inline u8 nvmet_cc_iocqes(u32 cc)
607 return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf;
610 static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl)
612 lockdep_assert_held(&ctrl->lock);
614 if (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES ||
615 nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES ||
616 nvmet_cc_mps(ctrl->cc) != 0 ||
617 nvmet_cc_ams(ctrl->cc) != 0 ||
618 nvmet_cc_css(ctrl->cc) != 0) {
619 ctrl->csts = NVME_CSTS_CFS;
623 ctrl->csts = NVME_CSTS_RDY;
626 static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl)
628 lockdep_assert_held(&ctrl->lock);
630 /* XXX: tear down queues? */
631 ctrl->csts &= ~NVME_CSTS_RDY;
635 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new)
639 mutex_lock(&ctrl->lock);
643 if (nvmet_cc_en(new) && !nvmet_cc_en(old))
644 nvmet_start_ctrl(ctrl);
645 if (!nvmet_cc_en(new) && nvmet_cc_en(old))
646 nvmet_clear_ctrl(ctrl);
647 if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) {
648 nvmet_clear_ctrl(ctrl);
649 ctrl->csts |= NVME_CSTS_SHST_CMPLT;
651 if (!nvmet_cc_shn(new) && nvmet_cc_shn(old))
652 ctrl->csts &= ~NVME_CSTS_SHST_CMPLT;
653 mutex_unlock(&ctrl->lock);
656 static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
658 /* command sets supported: NVMe command set: */
659 ctrl->cap = (1ULL << 37);
660 /* CC.EN timeout in 500msec units: */
661 ctrl->cap |= (15ULL << 24);
662 /* maximum queue entries supported: */
663 ctrl->cap |= NVMET_QUEUE_SIZE - 1;
666 u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid,
667 struct nvmet_req *req, struct nvmet_ctrl **ret)
669 struct nvmet_subsys *subsys;
670 struct nvmet_ctrl *ctrl;
673 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
675 pr_warn("connect request for invalid subsystem %s!\n",
677 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
678 return NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
681 mutex_lock(&subsys->lock);
682 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
683 if (ctrl->cntlid == cntlid) {
684 if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) {
685 pr_warn("hostnqn mismatch.\n");
688 if (!kref_get_unless_zero(&ctrl->ref))
696 pr_warn("could not find controller %d for subsys %s / host %s\n",
697 cntlid, subsysnqn, hostnqn);
698 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
699 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
702 mutex_unlock(&subsys->lock);
703 nvmet_subsys_put(subsys);
707 u16 nvmet_check_ctrl_status(struct nvmet_req *req, struct nvme_command *cmd)
709 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
710 pr_err("got io cmd %d while CC.EN == 0 on qid = %d\n",
711 cmd->common.opcode, req->sq->qid);
712 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
715 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
716 pr_err("got io cmd %d while CSTS.RDY == 0 on qid = %d\n",
717 cmd->common.opcode, req->sq->qid);
719 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
724 static bool __nvmet_host_allowed(struct nvmet_subsys *subsys,
727 struct nvmet_host_link *p;
729 if (subsys->allow_any_host)
732 list_for_each_entry(p, &subsys->hosts, entry) {
733 if (!strcmp(nvmet_host_name(p->host), hostnqn))
740 static bool nvmet_host_discovery_allowed(struct nvmet_req *req,
743 struct nvmet_subsys_link *s;
745 list_for_each_entry(s, &req->port->subsystems, entry) {
746 if (__nvmet_host_allowed(s->subsys, hostnqn))
753 bool nvmet_host_allowed(struct nvmet_req *req, struct nvmet_subsys *subsys,
756 lockdep_assert_held(&nvmet_config_sem);
758 if (subsys->type == NVME_NQN_DISC)
759 return nvmet_host_discovery_allowed(req, hostnqn);
761 return __nvmet_host_allowed(subsys, hostnqn);
764 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
765 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp)
767 struct nvmet_subsys *subsys;
768 struct nvmet_ctrl *ctrl;
772 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
773 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
775 pr_warn("connect request for invalid subsystem %s!\n",
777 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
781 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
782 down_read(&nvmet_config_sem);
783 if (!nvmet_host_allowed(req, subsys, hostnqn)) {
784 pr_info("connect by host %s for subsystem %s not allowed\n",
786 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
787 up_read(&nvmet_config_sem);
788 status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
789 goto out_put_subsystem;
791 up_read(&nvmet_config_sem);
793 status = NVME_SC_INTERNAL;
794 ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
796 goto out_put_subsystem;
797 mutex_init(&ctrl->lock);
799 nvmet_init_cap(ctrl);
801 INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
802 INIT_LIST_HEAD(&ctrl->async_events);
804 memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
805 memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE);
807 kref_init(&ctrl->ref);
808 ctrl->subsys = subsys;
810 ctrl->cqs = kcalloc(subsys->max_qid + 1,
811 sizeof(struct nvmet_cq *),
816 ctrl->sqs = kcalloc(subsys->max_qid + 1,
817 sizeof(struct nvmet_sq *),
822 ret = ida_simple_get(&cntlid_ida,
823 NVME_CNTLID_MIN, NVME_CNTLID_MAX,
826 status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR;
831 ctrl->ops = req->ops;
832 if (ctrl->subsys->type == NVME_NQN_DISC) {
833 /* Don't accept keep-alive timeout for discovery controllers */
835 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
840 * Discovery controllers use some arbitrary high value in order
841 * to cleanup stale discovery sessions
843 * From the latest base diff RC:
844 * "The Keep Alive command is not supported by
845 * Discovery controllers. A transport may specify a
846 * fixed Discovery controller activity timeout value
847 * (e.g., 2 minutes). If no commands are received
848 * by a Discovery controller within that time
849 * period, the controller may perform the
850 * actions for Keep Alive Timer expiration".
852 ctrl->kato = NVMET_DISC_KATO;
854 /* keep-alive timeout in seconds */
855 ctrl->kato = DIV_ROUND_UP(kato, 1000);
857 nvmet_start_keep_alive_timer(ctrl);
859 mutex_lock(&subsys->lock);
860 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
861 mutex_unlock(&subsys->lock);
867 ida_simple_remove(&cntlid_ida, ctrl->cntlid);
875 nvmet_subsys_put(subsys);
880 static void nvmet_ctrl_free(struct kref *ref)
882 struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref);
883 struct nvmet_subsys *subsys = ctrl->subsys;
885 mutex_lock(&subsys->lock);
886 list_del(&ctrl->subsys_entry);
887 mutex_unlock(&subsys->lock);
889 nvmet_stop_keep_alive_timer(ctrl);
891 flush_work(&ctrl->async_event_work);
892 cancel_work_sync(&ctrl->fatal_err_work);
894 ida_simple_remove(&cntlid_ida, ctrl->cntlid);
900 nvmet_subsys_put(subsys);
903 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl)
905 kref_put(&ctrl->ref, nvmet_ctrl_free);
908 static void nvmet_fatal_error_handler(struct work_struct *work)
910 struct nvmet_ctrl *ctrl =
911 container_of(work, struct nvmet_ctrl, fatal_err_work);
913 pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid);
914 ctrl->ops->delete_ctrl(ctrl);
917 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl)
919 mutex_lock(&ctrl->lock);
920 if (!(ctrl->csts & NVME_CSTS_CFS)) {
921 ctrl->csts |= NVME_CSTS_CFS;
922 INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
923 schedule_work(&ctrl->fatal_err_work);
925 mutex_unlock(&ctrl->lock);
927 EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error);
929 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
930 const char *subsysnqn)
932 struct nvmet_subsys_link *p;
937 if (!strncmp(NVME_DISC_SUBSYS_NAME, subsysnqn,
939 if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
941 return nvmet_disc_subsys;
944 down_read(&nvmet_config_sem);
945 list_for_each_entry(p, &port->subsystems, entry) {
946 if (!strncmp(p->subsys->subsysnqn, subsysnqn,
948 if (!kref_get_unless_zero(&p->subsys->ref))
950 up_read(&nvmet_config_sem);
954 up_read(&nvmet_config_sem);
958 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
959 enum nvme_subsys_type type)
961 struct nvmet_subsys *subsys;
963 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
967 subsys->ver = NVME_VS(1, 3, 0); /* NVMe 1.3.0 */
968 /* generate a random serial number as our controllers are ephemeral: */
969 get_random_bytes(&subsys->serial, sizeof(subsys->serial));
973 subsys->max_qid = NVMET_NR_QUEUES;
979 pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
984 subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
986 if (!subsys->subsysnqn) {
991 kref_init(&subsys->ref);
993 mutex_init(&subsys->lock);
994 INIT_LIST_HEAD(&subsys->namespaces);
995 INIT_LIST_HEAD(&subsys->ctrls);
996 INIT_LIST_HEAD(&subsys->hosts);
1001 static void nvmet_subsys_free(struct kref *ref)
1003 struct nvmet_subsys *subsys =
1004 container_of(ref, struct nvmet_subsys, ref);
1006 WARN_ON_ONCE(!list_empty(&subsys->namespaces));
1008 kfree(subsys->subsysnqn);
1012 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys)
1014 struct nvmet_ctrl *ctrl;
1016 mutex_lock(&subsys->lock);
1017 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
1018 ctrl->ops->delete_ctrl(ctrl);
1019 mutex_unlock(&subsys->lock);
1022 void nvmet_subsys_put(struct nvmet_subsys *subsys)
1024 kref_put(&subsys->ref, nvmet_subsys_free);
1027 static int __init nvmet_init(void)
1031 error = nvmet_init_discovery();
1035 error = nvmet_init_configfs();
1037 goto out_exit_discovery;
1041 nvmet_exit_discovery();
1046 static void __exit nvmet_exit(void)
1048 nvmet_exit_configfs();
1049 nvmet_exit_discovery();
1050 ida_destroy(&cntlid_ida);
1052 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024);
1053 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024);
1056 module_init(nvmet_init);
1057 module_exit(nvmet_exit);
1059 MODULE_LICENSE("GPL v2");