1 // SPDX-License-Identifier: GPL-2.0
3 * Common code for the NVMe target.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
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>
13 #include <generated/utsrelease.h>
15 #define CREATE_TRACE_POINTS
20 struct kmem_cache *nvmet_bvec_cache;
21 struct workqueue_struct *buffered_io_wq;
22 struct workqueue_struct *zbd_wq;
23 static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
24 static DEFINE_IDA(cntlid_ida);
26 struct workqueue_struct *nvmet_wq;
27 EXPORT_SYMBOL_GPL(nvmet_wq);
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:
36 * - per-subsystem allowed hosts list
37 * - allow_any_host subsystem attribute
39 * - the nvmet_transports array
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.
45 DECLARE_RWSEM(nvmet_config_sem);
47 u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
49 DECLARE_RWSEM(nvmet_ana_sem);
51 inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno)
55 return NVME_SC_SUCCESS;
57 req->error_loc = offsetof(struct nvme_rw_command, length);
58 return NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
60 req->error_loc = offsetof(struct nvme_rw_command, slba);
61 return NVME_SC_LBA_RANGE | NVME_SC_DNR;
63 req->error_loc = offsetof(struct nvme_common_command, opcode);
64 switch (req->cmd->common.opcode) {
66 case nvme_cmd_write_zeroes:
67 return NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
69 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
73 req->error_loc = offsetof(struct nvme_rw_command, nsid);
74 return NVME_SC_ACCESS_DENIED;
78 req->error_loc = offsetof(struct nvme_common_command, opcode);
79 return NVME_SC_INTERNAL | NVME_SC_DNR;
83 u16 nvmet_report_invalid_opcode(struct nvmet_req *req)
85 pr_debug("unhandled cmd %d on qid %d\n", req->cmd->common.opcode,
88 req->error_loc = offsetof(struct nvme_common_command, opcode);
89 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
92 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
93 const char *subsysnqn);
95 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
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;
105 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len)
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;
114 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len)
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;
123 static u32 nvmet_max_nsid(struct nvmet_subsys *subsys)
125 struct nvmet_ns *cur;
129 xa_for_each(&subsys->namespaces, idx, cur)
135 static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
137 return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16);
140 static void nvmet_async_events_failall(struct nvmet_ctrl *ctrl)
142 struct nvmet_req *req;
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);
151 mutex_unlock(&ctrl->lock);
154 static void nvmet_async_events_process(struct nvmet_ctrl *ctrl)
156 struct nvmet_async_event *aen;
157 struct nvmet_req *req;
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));
166 list_del(&aen->entry);
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);
174 mutex_unlock(&ctrl->lock);
177 static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
179 struct nvmet_async_event *aen, *tmp;
181 mutex_lock(&ctrl->lock);
182 list_for_each_entry_safe(aen, tmp, &ctrl->async_events, entry) {
183 list_del(&aen->entry);
186 mutex_unlock(&ctrl->lock);
189 static void nvmet_async_event_work(struct work_struct *work)
191 struct nvmet_ctrl *ctrl =
192 container_of(work, struct nvmet_ctrl, async_event_work);
194 nvmet_async_events_process(ctrl);
197 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
198 u8 event_info, u8 log_page)
200 struct nvmet_async_event *aen;
202 aen = kmalloc(sizeof(*aen), GFP_KERNEL);
206 aen->event_type = event_type;
207 aen->event_info = event_info;
208 aen->log_page = log_page;
210 mutex_lock(&ctrl->lock);
211 list_add_tail(&aen->entry, &ctrl->async_events);
212 mutex_unlock(&ctrl->lock);
214 queue_work(nvmet_wq, &ctrl->async_event_work);
217 static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
221 mutex_lock(&ctrl->lock);
222 if (ctrl->nr_changed_ns > NVME_MAX_CHANGED_NAMESPACES)
225 for (i = 0; i < ctrl->nr_changed_ns; i++) {
226 if (ctrl->changed_ns_list[i] == nsid)
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;
236 ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid;
238 mutex_unlock(&ctrl->lock);
241 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid)
243 struct nvmet_ctrl *ctrl;
245 lockdep_assert_held(&subsys->lock);
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))
251 nvmet_add_async_event(ctrl, NVME_AER_NOTICE,
252 NVME_AER_NOTICE_NS_CHANGED,
253 NVME_LOG_CHANGED_NS);
257 void nvmet_send_ana_event(struct nvmet_subsys *subsys,
258 struct nvmet_port *port)
260 struct nvmet_ctrl *ctrl;
262 mutex_lock(&subsys->lock);
263 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
264 if (port && ctrl->port != port)
266 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE))
268 nvmet_add_async_event(ctrl, NVME_AER_NOTICE,
269 NVME_AER_NOTICE_ANA, NVME_LOG_ANA);
271 mutex_unlock(&subsys->lock);
274 void nvmet_port_send_ana_event(struct nvmet_port *port)
276 struct nvmet_subsys_link *p;
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);
284 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops)
288 down_write(&nvmet_config_sem);
289 if (nvmet_transports[ops->type])
292 nvmet_transports[ops->type] = ops;
293 up_write(&nvmet_config_sem);
297 EXPORT_SYMBOL_GPL(nvmet_register_transport);
299 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops)
301 down_write(&nvmet_config_sem);
302 nvmet_transports[ops->type] = NULL;
303 up_write(&nvmet_config_sem);
305 EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
307 void nvmet_port_del_ctrls(struct nvmet_port *port, struct nvmet_subsys *subsys)
309 struct nvmet_ctrl *ctrl;
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);
316 mutex_unlock(&subsys->lock);
319 int nvmet_enable_port(struct nvmet_port *port)
321 const struct nvmet_fabrics_ops *ops;
324 lockdep_assert_held(&nvmet_config_sem);
326 ops = nvmet_transports[port->disc_addr.trtype];
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];
333 pr_err("transport type %d not supported\n",
334 port->disc_addr.trtype);
339 if (!try_module_get(ops->owner))
343 * If the user requested PI support and the transport isn't pi capable,
344 * don't enable the port.
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);
353 ret = ops->add_port(port);
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;
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.
366 if (port->max_queue_size < 0)
367 port->max_queue_size = NVMET_MAX_QUEUE_SIZE;
369 port->max_queue_size = clamp_t(int, port->max_queue_size,
370 NVMET_MIN_QUEUE_SIZE,
371 NVMET_MAX_QUEUE_SIZE);
373 port->enabled = true;
378 module_put(ops->owner);
382 void nvmet_disable_port(struct nvmet_port *port)
384 const struct nvmet_fabrics_ops *ops;
386 lockdep_assert_held(&nvmet_config_sem);
388 port->enabled = false;
391 ops = nvmet_transports[port->disc_addr.trtype];
392 ops->remove_port(port);
393 module_put(ops->owner);
396 static void nvmet_keep_alive_timer(struct work_struct *work)
398 struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
399 struct nvmet_ctrl, ka_work);
400 bool reset_tbkas = ctrl->reset_tbkas;
402 ctrl->reset_tbkas = false;
404 pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
406 queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
410 pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
411 ctrl->cntlid, ctrl->kato);
413 nvmet_ctrl_fatal_error(ctrl);
416 void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
418 if (unlikely(ctrl->kato == 0))
421 pr_debug("ctrl %d start keep-alive timer for %d secs\n",
422 ctrl->cntlid, ctrl->kato);
424 queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
427 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
429 if (unlikely(ctrl->kato == 0))
432 pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid);
434 cancel_delayed_work_sync(&ctrl->ka_work);
437 u16 nvmet_req_find_ns(struct nvmet_req *req)
439 u32 nsid = le32_to_cpu(req->cmd->common.nsid);
441 req->ns = xa_load(&nvmet_req_subsys(req)->namespaces, nsid);
442 if (unlikely(!req->ns)) {
443 req->error_loc = offsetof(struct nvme_common_command, nsid);
444 return NVME_SC_INVALID_NS | NVME_SC_DNR;
447 percpu_ref_get(&req->ns->ref);
448 return NVME_SC_SUCCESS;
451 static void nvmet_destroy_namespace(struct percpu_ref *ref)
453 struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref);
455 complete(&ns->disable_done);
458 void nvmet_put_namespace(struct nvmet_ns *ns)
460 percpu_ref_put(&ns->ref);
463 static void nvmet_ns_dev_disable(struct nvmet_ns *ns)
465 nvmet_bdev_ns_disable(ns);
466 nvmet_file_ns_disable(ns);
469 static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns)
472 struct pci_dev *p2p_dev;
478 pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n");
482 if (!blk_queue_pci_p2pdma(ns->bdev->bd_disk->queue)) {
483 pr_err("peer-to-peer DMA is not supported by the driver of %s\n",
489 ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true);
494 * Right now we just check that there is p2pmem available so
495 * we can report an error to the user right away if there
496 * is not. We'll find the actual device to use once we
497 * setup the controller when the port's device is available.
500 p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns));
502 pr_err("no peer-to-peer memory is available for %s\n",
507 pci_dev_put(p2p_dev);
514 * Note: ctrl->subsys->lock should be held when calling this function
516 static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl,
519 struct device *clients[2];
520 struct pci_dev *p2p_dev;
523 if (!ctrl->p2p_client || !ns->use_p2pmem)
527 ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true);
531 p2p_dev = pci_dev_get(ns->p2p_dev);
533 clients[0] = ctrl->p2p_client;
534 clients[1] = nvmet_ns_dev(ns);
536 p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients));
538 pr_err("no peer-to-peer memory is available that's supported by %s and %s\n",
539 dev_name(ctrl->p2p_client), ns->device_path);
544 ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev);
546 pci_dev_put(p2p_dev);
548 pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev),
552 bool nvmet_ns_revalidate(struct nvmet_ns *ns)
554 loff_t oldsize = ns->size;
557 nvmet_bdev_ns_revalidate(ns);
559 nvmet_file_ns_revalidate(ns);
561 return oldsize != ns->size;
564 int nvmet_ns_enable(struct nvmet_ns *ns)
566 struct nvmet_subsys *subsys = ns->subsys;
567 struct nvmet_ctrl *ctrl;
570 mutex_lock(&subsys->lock);
573 if (nvmet_is_passthru_subsys(subsys)) {
574 pr_info("cannot enable both passthru and regular namespaces for a single subsystem");
582 if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES)
585 ret = nvmet_bdev_ns_enable(ns);
587 ret = nvmet_file_ns_enable(ns);
591 ret = nvmet_p2pmem_ns_enable(ns);
593 goto out_dev_disable;
595 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
596 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
598 ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
603 if (ns->nsid > subsys->max_nsid)
604 subsys->max_nsid = ns->nsid;
606 ret = xa_insert(&subsys->namespaces, ns->nsid, ns, GFP_KERNEL);
608 goto out_restore_subsys_maxnsid;
610 subsys->nr_namespaces++;
612 nvmet_ns_changed(subsys, ns->nsid);
616 mutex_unlock(&subsys->lock);
619 out_restore_subsys_maxnsid:
620 subsys->max_nsid = nvmet_max_nsid(subsys);
621 percpu_ref_exit(&ns->ref);
623 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
624 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
626 nvmet_ns_dev_disable(ns);
630 void nvmet_ns_disable(struct nvmet_ns *ns)
632 struct nvmet_subsys *subsys = ns->subsys;
633 struct nvmet_ctrl *ctrl;
635 mutex_lock(&subsys->lock);
640 xa_erase(&ns->subsys->namespaces, ns->nsid);
641 if (ns->nsid == subsys->max_nsid)
642 subsys->max_nsid = nvmet_max_nsid(subsys);
644 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
645 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
647 mutex_unlock(&subsys->lock);
650 * Now that we removed the namespaces from the lookup list, we
651 * can kill the per_cpu ref and wait for any remaining references
652 * to be dropped, as well as a RCU grace period for anyone only
653 * using the namepace under rcu_read_lock(). Note that we can't
654 * use call_rcu here as we need to ensure the namespaces have
655 * been fully destroyed before unloading the module.
657 percpu_ref_kill(&ns->ref);
659 wait_for_completion(&ns->disable_done);
660 percpu_ref_exit(&ns->ref);
662 mutex_lock(&subsys->lock);
664 subsys->nr_namespaces--;
665 nvmet_ns_changed(subsys, ns->nsid);
666 nvmet_ns_dev_disable(ns);
668 mutex_unlock(&subsys->lock);
671 void nvmet_ns_free(struct nvmet_ns *ns)
673 nvmet_ns_disable(ns);
675 down_write(&nvmet_ana_sem);
676 nvmet_ana_group_enabled[ns->anagrpid]--;
677 up_write(&nvmet_ana_sem);
679 kfree(ns->device_path);
683 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
687 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
691 init_completion(&ns->disable_done);
696 down_write(&nvmet_ana_sem);
697 ns->anagrpid = NVMET_DEFAULT_ANA_GRPID;
698 nvmet_ana_group_enabled[ns->anagrpid]++;
699 up_write(&nvmet_ana_sem);
702 ns->buffered_io = false;
703 ns->csi = NVME_CSI_NVM;
708 static void nvmet_update_sq_head(struct nvmet_req *req)
711 u32 old_sqhd, new_sqhd;
713 old_sqhd = READ_ONCE(req->sq->sqhd);
715 new_sqhd = (old_sqhd + 1) % req->sq->size;
716 } while (!try_cmpxchg(&req->sq->sqhd, &old_sqhd, new_sqhd));
718 req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
721 static void nvmet_set_error(struct nvmet_req *req, u16 status)
723 struct nvmet_ctrl *ctrl = req->sq->ctrl;
724 struct nvme_error_slot *new_error_slot;
727 req->cqe->status = cpu_to_le16(status << 1);
729 if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC)
732 spin_lock_irqsave(&ctrl->error_lock, flags);
735 &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS];
737 new_error_slot->error_count = cpu_to_le64(ctrl->err_counter);
738 new_error_slot->sqid = cpu_to_le16(req->sq->qid);
739 new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id);
740 new_error_slot->status_field = cpu_to_le16(status << 1);
741 new_error_slot->param_error_location = cpu_to_le16(req->error_loc);
742 new_error_slot->lba = cpu_to_le64(req->error_slba);
743 new_error_slot->nsid = req->cmd->common.nsid;
744 spin_unlock_irqrestore(&ctrl->error_lock, flags);
746 /* set the more bit for this request */
747 req->cqe->status |= cpu_to_le16(1 << 14);
750 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
752 struct nvmet_ns *ns = req->ns;
754 if (!req->sq->sqhd_disabled)
755 nvmet_update_sq_head(req);
756 req->cqe->sq_id = cpu_to_le16(req->sq->qid);
757 req->cqe->command_id = req->cmd->common.command_id;
759 if (unlikely(status))
760 nvmet_set_error(req, status);
762 trace_nvmet_req_complete(req);
764 req->ops->queue_response(req);
766 nvmet_put_namespace(ns);
769 void nvmet_req_complete(struct nvmet_req *req, u16 status)
771 struct nvmet_sq *sq = req->sq;
773 __nvmet_req_complete(req, status);
774 percpu_ref_put(&sq->ref);
776 EXPORT_SYMBOL_GPL(nvmet_req_complete);
778 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq,
785 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq,
795 static void nvmet_confirm_sq(struct percpu_ref *ref)
797 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
799 complete(&sq->confirm_done);
802 void nvmet_sq_destroy(struct nvmet_sq *sq)
804 struct nvmet_ctrl *ctrl = sq->ctrl;
807 * If this is the admin queue, complete all AERs so that our
808 * queue doesn't have outstanding requests on it.
810 if (ctrl && ctrl->sqs && ctrl->sqs[0] == sq)
811 nvmet_async_events_failall(ctrl);
812 percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq);
813 wait_for_completion(&sq->confirm_done);
814 wait_for_completion(&sq->free_done);
815 percpu_ref_exit(&sq->ref);
816 nvmet_auth_sq_free(sq);
820 * The teardown flow may take some time, and the host may not
821 * send us keep-alive during this period, hence reset the
822 * traffic based keep-alive timer so we don't trigger a
823 * controller teardown as a result of a keep-alive expiration.
825 ctrl->reset_tbkas = true;
826 sq->ctrl->sqs[sq->qid] = NULL;
827 nvmet_ctrl_put(ctrl);
828 sq->ctrl = NULL; /* allows reusing the queue later */
831 EXPORT_SYMBOL_GPL(nvmet_sq_destroy);
833 static void nvmet_sq_free(struct percpu_ref *ref)
835 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
837 complete(&sq->free_done);
840 int nvmet_sq_init(struct nvmet_sq *sq)
844 ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL);
846 pr_err("percpu_ref init failed!\n");
849 init_completion(&sq->free_done);
850 init_completion(&sq->confirm_done);
851 nvmet_auth_sq_init(sq);
855 EXPORT_SYMBOL_GPL(nvmet_sq_init);
857 static inline u16 nvmet_check_ana_state(struct nvmet_port *port,
860 enum nvme_ana_state state = port->ana_state[ns->anagrpid];
862 if (unlikely(state == NVME_ANA_INACCESSIBLE))
863 return NVME_SC_ANA_INACCESSIBLE;
864 if (unlikely(state == NVME_ANA_PERSISTENT_LOSS))
865 return NVME_SC_ANA_PERSISTENT_LOSS;
866 if (unlikely(state == NVME_ANA_CHANGE))
867 return NVME_SC_ANA_TRANSITION;
871 static inline u16 nvmet_io_cmd_check_access(struct nvmet_req *req)
873 if (unlikely(req->ns->readonly)) {
874 switch (req->cmd->common.opcode) {
879 return NVME_SC_NS_WRITE_PROTECTED;
886 static u16 nvmet_parse_io_cmd(struct nvmet_req *req)
888 struct nvme_command *cmd = req->cmd;
891 if (nvme_is_fabrics(cmd))
892 return nvmet_parse_fabrics_io_cmd(req);
894 if (unlikely(!nvmet_check_auth_status(req)))
895 return NVME_SC_AUTH_REQUIRED | NVME_SC_DNR;
897 ret = nvmet_check_ctrl_status(req);
901 if (nvmet_is_passthru_req(req))
902 return nvmet_parse_passthru_io_cmd(req);
904 ret = nvmet_req_find_ns(req);
908 ret = nvmet_check_ana_state(req->port, req->ns);
910 req->error_loc = offsetof(struct nvme_common_command, nsid);
913 ret = nvmet_io_cmd_check_access(req);
915 req->error_loc = offsetof(struct nvme_common_command, nsid);
919 switch (req->ns->csi) {
922 return nvmet_file_parse_io_cmd(req);
923 return nvmet_bdev_parse_io_cmd(req);
925 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED))
926 return nvmet_bdev_zns_parse_io_cmd(req);
927 return NVME_SC_INVALID_IO_CMD_SET;
929 return NVME_SC_INVALID_IO_CMD_SET;
933 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
934 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops)
936 u8 flags = req->cmd->common.flags;
943 req->metadata_sg = NULL;
945 req->metadata_sg_cnt = 0;
946 req->transfer_len = 0;
947 req->metadata_len = 0;
948 req->cqe->status = 0;
949 req->cqe->sq_head = 0;
951 req->error_loc = NVMET_NO_ERROR_LOC;
954 /* no support for fused commands yet */
955 if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) {
956 req->error_loc = offsetof(struct nvme_common_command, flags);
957 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
962 * For fabrics, PSDT field shall describe metadata pointer (MPTR) that
963 * contains an address of a single contiguous physical buffer that is
966 if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) {
967 req->error_loc = offsetof(struct nvme_common_command, flags);
968 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
972 if (unlikely(!req->sq->ctrl))
973 /* will return an error for any non-connect command: */
974 status = nvmet_parse_connect_cmd(req);
975 else if (likely(req->sq->qid != 0))
976 status = nvmet_parse_io_cmd(req);
978 status = nvmet_parse_admin_cmd(req);
983 trace_nvmet_req_init(req, req->cmd);
985 if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
986 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
991 sq->ctrl->reset_tbkas = true;
996 __nvmet_req_complete(req, status);
999 EXPORT_SYMBOL_GPL(nvmet_req_init);
1001 void nvmet_req_uninit(struct nvmet_req *req)
1003 percpu_ref_put(&req->sq->ref);
1005 nvmet_put_namespace(req->ns);
1007 EXPORT_SYMBOL_GPL(nvmet_req_uninit);
1009 bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len)
1011 if (unlikely(len != req->transfer_len)) {
1012 req->error_loc = offsetof(struct nvme_common_command, dptr);
1013 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
1019 EXPORT_SYMBOL_GPL(nvmet_check_transfer_len);
1021 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len)
1023 if (unlikely(data_len > req->transfer_len)) {
1024 req->error_loc = offsetof(struct nvme_common_command, dptr);
1025 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
1032 static unsigned int nvmet_data_transfer_len(struct nvmet_req *req)
1034 return req->transfer_len - req->metadata_len;
1037 static int nvmet_req_alloc_p2pmem_sgls(struct pci_dev *p2p_dev,
1038 struct nvmet_req *req)
1040 req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt,
1041 nvmet_data_transfer_len(req));
1045 if (req->metadata_len) {
1046 req->metadata_sg = pci_p2pmem_alloc_sgl(p2p_dev,
1047 &req->metadata_sg_cnt, req->metadata_len);
1048 if (!req->metadata_sg)
1052 req->p2p_dev = p2p_dev;
1056 pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1061 static struct pci_dev *nvmet_req_find_p2p_dev(struct nvmet_req *req)
1063 if (!IS_ENABLED(CONFIG_PCI_P2PDMA) ||
1064 !req->sq->ctrl || !req->sq->qid || !req->ns)
1066 return radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
1069 int nvmet_req_alloc_sgls(struct nvmet_req *req)
1071 struct pci_dev *p2p_dev = nvmet_req_find_p2p_dev(req);
1073 if (p2p_dev && !nvmet_req_alloc_p2pmem_sgls(p2p_dev, req))
1076 req->sg = sgl_alloc(nvmet_data_transfer_len(req), GFP_KERNEL,
1078 if (unlikely(!req->sg))
1081 if (req->metadata_len) {
1082 req->metadata_sg = sgl_alloc(req->metadata_len, GFP_KERNEL,
1083 &req->metadata_sg_cnt);
1084 if (unlikely(!req->metadata_sg))
1094 EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgls);
1096 void nvmet_req_free_sgls(struct nvmet_req *req)
1099 pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1100 if (req->metadata_sg)
1101 pci_p2pmem_free_sgl(req->p2p_dev, req->metadata_sg);
1102 req->p2p_dev = NULL;
1105 if (req->metadata_sg)
1106 sgl_free(req->metadata_sg);
1110 req->metadata_sg = NULL;
1112 req->metadata_sg_cnt = 0;
1114 EXPORT_SYMBOL_GPL(nvmet_req_free_sgls);
1116 static inline bool nvmet_cc_en(u32 cc)
1118 return (cc >> NVME_CC_EN_SHIFT) & 0x1;
1121 static inline u8 nvmet_cc_css(u32 cc)
1123 return (cc >> NVME_CC_CSS_SHIFT) & 0x7;
1126 static inline u8 nvmet_cc_mps(u32 cc)
1128 return (cc >> NVME_CC_MPS_SHIFT) & 0xf;
1131 static inline u8 nvmet_cc_ams(u32 cc)
1133 return (cc >> NVME_CC_AMS_SHIFT) & 0x7;
1136 static inline u8 nvmet_cc_shn(u32 cc)
1138 return (cc >> NVME_CC_SHN_SHIFT) & 0x3;
1141 static inline u8 nvmet_cc_iosqes(u32 cc)
1143 return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf;
1146 static inline u8 nvmet_cc_iocqes(u32 cc)
1148 return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf;
1151 static inline bool nvmet_css_supported(u8 cc_css)
1153 switch (cc_css << NVME_CC_CSS_SHIFT) {
1154 case NVME_CC_CSS_NVM:
1155 case NVME_CC_CSS_CSI:
1162 static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl)
1164 lockdep_assert_held(&ctrl->lock);
1167 * Only I/O controllers should verify iosqes,iocqes.
1168 * Strictly speaking, the spec says a discovery controller
1169 * should verify iosqes,iocqes are zeroed, however that
1170 * would break backwards compatibility, so don't enforce it.
1172 if (!nvmet_is_disc_subsys(ctrl->subsys) &&
1173 (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES ||
1174 nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES)) {
1175 ctrl->csts = NVME_CSTS_CFS;
1179 if (nvmet_cc_mps(ctrl->cc) != 0 ||
1180 nvmet_cc_ams(ctrl->cc) != 0 ||
1181 !nvmet_css_supported(nvmet_cc_css(ctrl->cc))) {
1182 ctrl->csts = NVME_CSTS_CFS;
1186 ctrl->csts = NVME_CSTS_RDY;
1189 * Controllers that are not yet enabled should not really enforce the
1190 * keep alive timeout, but we still want to track a timeout and cleanup
1191 * in case a host died before it enabled the controller. Hence, simply
1192 * reset the keep alive timer when the controller is enabled.
1195 mod_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
1198 static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl)
1200 lockdep_assert_held(&ctrl->lock);
1202 /* XXX: tear down queues? */
1203 ctrl->csts &= ~NVME_CSTS_RDY;
1207 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new)
1211 mutex_lock(&ctrl->lock);
1215 if (nvmet_cc_en(new) && !nvmet_cc_en(old))
1216 nvmet_start_ctrl(ctrl);
1217 if (!nvmet_cc_en(new) && nvmet_cc_en(old))
1218 nvmet_clear_ctrl(ctrl);
1219 if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) {
1220 nvmet_clear_ctrl(ctrl);
1221 ctrl->csts |= NVME_CSTS_SHST_CMPLT;
1223 if (!nvmet_cc_shn(new) && nvmet_cc_shn(old))
1224 ctrl->csts &= ~NVME_CSTS_SHST_CMPLT;
1225 mutex_unlock(&ctrl->lock);
1228 static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
1230 /* command sets supported: NVMe command set: */
1231 ctrl->cap = (1ULL << 37);
1232 /* Controller supports one or more I/O Command Sets */
1233 ctrl->cap |= (1ULL << 43);
1234 /* CC.EN timeout in 500msec units: */
1235 ctrl->cap |= (15ULL << 24);
1236 /* maximum queue entries supported: */
1237 if (ctrl->ops->get_max_queue_size)
1238 ctrl->cap |= min_t(u16, ctrl->ops->get_max_queue_size(ctrl),
1239 ctrl->port->max_queue_size) - 1;
1241 ctrl->cap |= ctrl->port->max_queue_size - 1;
1243 if (nvmet_is_passthru_subsys(ctrl->subsys))
1244 nvmet_passthrough_override_cap(ctrl);
1247 struct nvmet_ctrl *nvmet_ctrl_find_get(const char *subsysnqn,
1248 const char *hostnqn, u16 cntlid,
1249 struct nvmet_req *req)
1251 struct nvmet_ctrl *ctrl = NULL;
1252 struct nvmet_subsys *subsys;
1254 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1256 pr_warn("connect request for invalid subsystem %s!\n",
1258 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1262 mutex_lock(&subsys->lock);
1263 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
1264 if (ctrl->cntlid == cntlid) {
1265 if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) {
1266 pr_warn("hostnqn mismatch.\n");
1269 if (!kref_get_unless_zero(&ctrl->ref))
1277 ctrl = NULL; /* ctrl not found */
1278 pr_warn("could not find controller %d for subsys %s / host %s\n",
1279 cntlid, subsysnqn, hostnqn);
1280 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
1283 mutex_unlock(&subsys->lock);
1284 nvmet_subsys_put(subsys);
1289 u16 nvmet_check_ctrl_status(struct nvmet_req *req)
1291 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
1292 pr_err("got cmd %d while CC.EN == 0 on qid = %d\n",
1293 req->cmd->common.opcode, req->sq->qid);
1294 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1297 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
1298 pr_err("got cmd %d while CSTS.RDY == 0 on qid = %d\n",
1299 req->cmd->common.opcode, req->sq->qid);
1300 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1303 if (unlikely(!nvmet_check_auth_status(req))) {
1304 pr_warn("qid %d not authenticated\n", req->sq->qid);
1305 return NVME_SC_AUTH_REQUIRED | NVME_SC_DNR;
1310 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn)
1312 struct nvmet_host_link *p;
1314 lockdep_assert_held(&nvmet_config_sem);
1316 if (subsys->allow_any_host)
1319 if (nvmet_is_disc_subsys(subsys)) /* allow all access to disc subsys */
1322 list_for_each_entry(p, &subsys->hosts, entry) {
1323 if (!strcmp(nvmet_host_name(p->host), hostnqn))
1331 * Note: ctrl->subsys->lock should be held when calling this function
1333 static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl,
1334 struct nvmet_req *req)
1336 struct nvmet_ns *ns;
1339 if (!req->p2p_client)
1342 ctrl->p2p_client = get_device(req->p2p_client);
1344 xa_for_each(&ctrl->subsys->namespaces, idx, ns)
1345 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
1349 * Note: ctrl->subsys->lock should be held when calling this function
1351 static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl)
1353 struct radix_tree_iter iter;
1356 radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0)
1357 pci_dev_put(radix_tree_deref_slot(slot));
1359 put_device(ctrl->p2p_client);
1362 static void nvmet_fatal_error_handler(struct work_struct *work)
1364 struct nvmet_ctrl *ctrl =
1365 container_of(work, struct nvmet_ctrl, fatal_err_work);
1367 pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid);
1368 ctrl->ops->delete_ctrl(ctrl);
1371 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
1372 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp)
1374 struct nvmet_subsys *subsys;
1375 struct nvmet_ctrl *ctrl;
1379 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1380 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1382 pr_warn("connect request for invalid subsystem %s!\n",
1384 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1385 req->error_loc = offsetof(struct nvme_common_command, dptr);
1389 down_read(&nvmet_config_sem);
1390 if (!nvmet_host_allowed(subsys, hostnqn)) {
1391 pr_info("connect by host %s for subsystem %s not allowed\n",
1392 hostnqn, subsysnqn);
1393 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
1394 up_read(&nvmet_config_sem);
1395 status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
1396 req->error_loc = offsetof(struct nvme_common_command, dptr);
1397 goto out_put_subsystem;
1399 up_read(&nvmet_config_sem);
1401 status = NVME_SC_INTERNAL;
1402 ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
1404 goto out_put_subsystem;
1405 mutex_init(&ctrl->lock);
1407 ctrl->port = req->port;
1408 ctrl->ops = req->ops;
1410 #ifdef CONFIG_NVME_TARGET_PASSTHRU
1411 /* By default, set loop targets to clear IDS by default */
1412 if (ctrl->port->disc_addr.trtype == NVMF_TRTYPE_LOOP)
1413 subsys->clear_ids = 1;
1416 INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
1417 INIT_LIST_HEAD(&ctrl->async_events);
1418 INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL);
1419 INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
1420 INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer);
1422 memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
1423 memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE);
1425 kref_init(&ctrl->ref);
1426 ctrl->subsys = subsys;
1427 ctrl->pi_support = ctrl->port->pi_enable && ctrl->subsys->pi_support;
1428 nvmet_init_cap(ctrl);
1429 WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL);
1431 ctrl->changed_ns_list = kmalloc_array(NVME_MAX_CHANGED_NAMESPACES,
1432 sizeof(__le32), GFP_KERNEL);
1433 if (!ctrl->changed_ns_list)
1436 ctrl->sqs = kcalloc(subsys->max_qid + 1,
1437 sizeof(struct nvmet_sq *),
1440 goto out_free_changed_ns_list;
1442 ret = ida_alloc_range(&cntlid_ida,
1443 subsys->cntlid_min, subsys->cntlid_max,
1446 status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR;
1452 * Discovery controllers may use some arbitrary high value
1453 * in order to cleanup stale discovery sessions
1455 if (nvmet_is_disc_subsys(ctrl->subsys) && !kato)
1456 kato = NVMET_DISC_KATO_MS;
1458 /* keep-alive timeout in seconds */
1459 ctrl->kato = DIV_ROUND_UP(kato, 1000);
1461 ctrl->err_counter = 0;
1462 spin_lock_init(&ctrl->error_lock);
1464 nvmet_start_keep_alive_timer(ctrl);
1466 mutex_lock(&subsys->lock);
1467 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
1468 nvmet_setup_p2p_ns_map(ctrl, req);
1469 mutex_unlock(&subsys->lock);
1476 out_free_changed_ns_list:
1477 kfree(ctrl->changed_ns_list);
1481 nvmet_subsys_put(subsys);
1486 static void nvmet_ctrl_free(struct kref *ref)
1488 struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref);
1489 struct nvmet_subsys *subsys = ctrl->subsys;
1491 mutex_lock(&subsys->lock);
1492 nvmet_release_p2p_ns_map(ctrl);
1493 list_del(&ctrl->subsys_entry);
1494 mutex_unlock(&subsys->lock);
1496 nvmet_stop_keep_alive_timer(ctrl);
1498 flush_work(&ctrl->async_event_work);
1499 cancel_work_sync(&ctrl->fatal_err_work);
1501 nvmet_destroy_auth(ctrl);
1503 ida_free(&cntlid_ida, ctrl->cntlid);
1505 nvmet_async_events_free(ctrl);
1507 kfree(ctrl->changed_ns_list);
1510 nvmet_subsys_put(subsys);
1513 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl)
1515 kref_put(&ctrl->ref, nvmet_ctrl_free);
1518 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl)
1520 mutex_lock(&ctrl->lock);
1521 if (!(ctrl->csts & NVME_CSTS_CFS)) {
1522 ctrl->csts |= NVME_CSTS_CFS;
1523 queue_work(nvmet_wq, &ctrl->fatal_err_work);
1525 mutex_unlock(&ctrl->lock);
1527 EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error);
1529 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
1530 const char *subsysnqn)
1532 struct nvmet_subsys_link *p;
1537 if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) {
1538 if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
1540 return nvmet_disc_subsys;
1543 down_read(&nvmet_config_sem);
1544 list_for_each_entry(p, &port->subsystems, entry) {
1545 if (!strncmp(p->subsys->subsysnqn, subsysnqn,
1547 if (!kref_get_unless_zero(&p->subsys->ref))
1549 up_read(&nvmet_config_sem);
1553 up_read(&nvmet_config_sem);
1557 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
1558 enum nvme_subsys_type type)
1560 struct nvmet_subsys *subsys;
1561 char serial[NVMET_SN_MAX_SIZE / 2];
1564 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
1566 return ERR_PTR(-ENOMEM);
1568 subsys->ver = NVMET_DEFAULT_VS;
1569 /* generate a random serial number as our controllers are ephemeral: */
1570 get_random_bytes(&serial, sizeof(serial));
1571 bin2hex(subsys->serial, &serial, sizeof(serial));
1573 subsys->model_number = kstrdup(NVMET_DEFAULT_CTRL_MODEL, GFP_KERNEL);
1574 if (!subsys->model_number) {
1579 subsys->ieee_oui = 0;
1581 subsys->firmware_rev = kstrndup(UTS_RELEASE, NVMET_FR_MAX_SIZE, GFP_KERNEL);
1582 if (!subsys->firmware_rev) {
1589 subsys->max_qid = NVMET_NR_QUEUES;
1593 subsys->max_qid = 0;
1596 pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
1600 subsys->type = type;
1601 subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
1603 if (!subsys->subsysnqn) {
1607 subsys->cntlid_min = NVME_CNTLID_MIN;
1608 subsys->cntlid_max = NVME_CNTLID_MAX;
1609 kref_init(&subsys->ref);
1611 mutex_init(&subsys->lock);
1612 xa_init(&subsys->namespaces);
1613 INIT_LIST_HEAD(&subsys->ctrls);
1614 INIT_LIST_HEAD(&subsys->hosts);
1619 kfree(subsys->firmware_rev);
1621 kfree(subsys->model_number);
1624 return ERR_PTR(ret);
1627 static void nvmet_subsys_free(struct kref *ref)
1629 struct nvmet_subsys *subsys =
1630 container_of(ref, struct nvmet_subsys, ref);
1632 WARN_ON_ONCE(!xa_empty(&subsys->namespaces));
1634 xa_destroy(&subsys->namespaces);
1635 nvmet_passthru_subsys_free(subsys);
1637 kfree(subsys->subsysnqn);
1638 kfree(subsys->model_number);
1639 kfree(subsys->firmware_rev);
1643 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys)
1645 struct nvmet_ctrl *ctrl;
1647 mutex_lock(&subsys->lock);
1648 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
1649 ctrl->ops->delete_ctrl(ctrl);
1650 mutex_unlock(&subsys->lock);
1653 void nvmet_subsys_put(struct nvmet_subsys *subsys)
1655 kref_put(&subsys->ref, nvmet_subsys_free);
1658 static int __init nvmet_init(void)
1660 int error = -ENOMEM;
1662 nvmet_ana_group_enabled[NVMET_DEFAULT_ANA_GRPID] = 1;
1664 nvmet_bvec_cache = kmem_cache_create("nvmet-bvec",
1665 NVMET_MAX_MPOOL_BVEC * sizeof(struct bio_vec), 0,
1666 SLAB_HWCACHE_ALIGN, NULL);
1667 if (!nvmet_bvec_cache)
1670 zbd_wq = alloc_workqueue("nvmet-zbd-wq", WQ_MEM_RECLAIM, 0);
1672 goto out_destroy_bvec_cache;
1674 buffered_io_wq = alloc_workqueue("nvmet-buffered-io-wq",
1676 if (!buffered_io_wq)
1677 goto out_free_zbd_work_queue;
1679 nvmet_wq = alloc_workqueue("nvmet-wq", WQ_MEM_RECLAIM, 0);
1681 goto out_free_buffered_work_queue;
1683 error = nvmet_init_discovery();
1685 goto out_free_nvmet_work_queue;
1687 error = nvmet_init_configfs();
1689 goto out_exit_discovery;
1693 nvmet_exit_discovery();
1694 out_free_nvmet_work_queue:
1695 destroy_workqueue(nvmet_wq);
1696 out_free_buffered_work_queue:
1697 destroy_workqueue(buffered_io_wq);
1698 out_free_zbd_work_queue:
1699 destroy_workqueue(zbd_wq);
1700 out_destroy_bvec_cache:
1701 kmem_cache_destroy(nvmet_bvec_cache);
1705 static void __exit nvmet_exit(void)
1707 nvmet_exit_configfs();
1708 nvmet_exit_discovery();
1709 ida_destroy(&cntlid_ida);
1710 destroy_workqueue(nvmet_wq);
1711 destroy_workqueue(buffered_io_wq);
1712 destroy_workqueue(zbd_wq);
1713 kmem_cache_destroy(nvmet_bvec_cache);
1715 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024);
1716 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024);
1719 module_init(nvmet_init);
1720 module_exit(nvmet_exit);
1722 MODULE_DESCRIPTION("NVMe target core framework");
1723 MODULE_LICENSE("GPL v2");