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 #define CREATE_TRACE_POINTS
18 struct workqueue_struct *buffered_io_wq;
19 struct workqueue_struct *zbd_wq;
20 static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
21 static DEFINE_IDA(cntlid_ida);
23 struct workqueue_struct *nvmet_wq;
24 EXPORT_SYMBOL_GPL(nvmet_wq);
27 * This read/write semaphore is used to synchronize access to configuration
28 * information on a target system that will result in discovery log page
29 * information change for at least one host.
30 * The full list of resources to protected by this semaphore is:
33 * - per-subsystem allowed hosts list
34 * - allow_any_host subsystem attribute
36 * - the nvmet_transports array
38 * When updating any of those lists/structures write lock should be obtained,
39 * while when reading (popolating discovery log page or checking host-subsystem
40 * link) read lock is obtained to allow concurrent reads.
42 DECLARE_RWSEM(nvmet_config_sem);
44 u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
46 DECLARE_RWSEM(nvmet_ana_sem);
48 inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno)
52 return NVME_SC_SUCCESS;
54 req->error_loc = offsetof(struct nvme_rw_command, length);
55 return NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
57 req->error_loc = offsetof(struct nvme_rw_command, slba);
58 return NVME_SC_LBA_RANGE | NVME_SC_DNR;
60 req->error_loc = offsetof(struct nvme_common_command, opcode);
61 switch (req->cmd->common.opcode) {
63 case nvme_cmd_write_zeroes:
64 return NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
66 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
70 req->error_loc = offsetof(struct nvme_rw_command, nsid);
71 return NVME_SC_ACCESS_DENIED;
75 req->error_loc = offsetof(struct nvme_common_command, opcode);
76 return NVME_SC_INTERNAL | NVME_SC_DNR;
80 u16 nvmet_report_invalid_opcode(struct nvmet_req *req)
82 pr_debug("unhandled cmd %d on qid %d\n", req->cmd->common.opcode,
85 req->error_loc = offsetof(struct nvme_common_command, opcode);
86 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
89 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
90 const char *subsysnqn);
92 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
95 if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
96 req->error_loc = offsetof(struct nvme_common_command, dptr);
97 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
102 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len)
104 if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
105 req->error_loc = offsetof(struct nvme_common_command, dptr);
106 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
111 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len)
113 if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len) {
114 req->error_loc = offsetof(struct nvme_common_command, dptr);
115 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
120 static u32 nvmet_max_nsid(struct nvmet_subsys *subsys)
122 struct nvmet_ns *cur;
126 xa_for_each(&subsys->namespaces, idx, cur)
132 static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
134 return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16);
137 static void nvmet_async_events_failall(struct nvmet_ctrl *ctrl)
139 struct nvmet_req *req;
141 mutex_lock(&ctrl->lock);
142 while (ctrl->nr_async_event_cmds) {
143 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
144 mutex_unlock(&ctrl->lock);
145 nvmet_req_complete(req, NVME_SC_INTERNAL | NVME_SC_DNR);
146 mutex_lock(&ctrl->lock);
148 mutex_unlock(&ctrl->lock);
151 static void nvmet_async_events_process(struct nvmet_ctrl *ctrl)
153 struct nvmet_async_event *aen;
154 struct nvmet_req *req;
156 mutex_lock(&ctrl->lock);
157 while (ctrl->nr_async_event_cmds && !list_empty(&ctrl->async_events)) {
158 aen = list_first_entry(&ctrl->async_events,
159 struct nvmet_async_event, entry);
160 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
161 nvmet_set_result(req, nvmet_async_event_result(aen));
163 list_del(&aen->entry);
166 mutex_unlock(&ctrl->lock);
167 trace_nvmet_async_event(ctrl, req->cqe->result.u32);
168 nvmet_req_complete(req, 0);
169 mutex_lock(&ctrl->lock);
171 mutex_unlock(&ctrl->lock);
174 static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
176 struct nvmet_async_event *aen, *tmp;
178 mutex_lock(&ctrl->lock);
179 list_for_each_entry_safe(aen, tmp, &ctrl->async_events, entry) {
180 list_del(&aen->entry);
183 mutex_unlock(&ctrl->lock);
186 static void nvmet_async_event_work(struct work_struct *work)
188 struct nvmet_ctrl *ctrl =
189 container_of(work, struct nvmet_ctrl, async_event_work);
191 nvmet_async_events_process(ctrl);
194 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
195 u8 event_info, u8 log_page)
197 struct nvmet_async_event *aen;
199 aen = kmalloc(sizeof(*aen), GFP_KERNEL);
203 aen->event_type = event_type;
204 aen->event_info = event_info;
205 aen->log_page = log_page;
207 mutex_lock(&ctrl->lock);
208 list_add_tail(&aen->entry, &ctrl->async_events);
209 mutex_unlock(&ctrl->lock);
211 queue_work(nvmet_wq, &ctrl->async_event_work);
214 static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
218 mutex_lock(&ctrl->lock);
219 if (ctrl->nr_changed_ns > NVME_MAX_CHANGED_NAMESPACES)
222 for (i = 0; i < ctrl->nr_changed_ns; i++) {
223 if (ctrl->changed_ns_list[i] == nsid)
227 if (ctrl->nr_changed_ns == NVME_MAX_CHANGED_NAMESPACES) {
228 ctrl->changed_ns_list[0] = cpu_to_le32(0xffffffff);
229 ctrl->nr_changed_ns = U32_MAX;
233 ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid;
235 mutex_unlock(&ctrl->lock);
238 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid)
240 struct nvmet_ctrl *ctrl;
242 lockdep_assert_held(&subsys->lock);
244 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
245 nvmet_add_to_changed_ns_log(ctrl, cpu_to_le32(nsid));
246 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_NS_ATTR))
248 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
249 NVME_AER_NOTICE_NS_CHANGED,
250 NVME_LOG_CHANGED_NS);
254 void nvmet_send_ana_event(struct nvmet_subsys *subsys,
255 struct nvmet_port *port)
257 struct nvmet_ctrl *ctrl;
259 mutex_lock(&subsys->lock);
260 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
261 if (port && ctrl->port != port)
263 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE))
265 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
266 NVME_AER_NOTICE_ANA, NVME_LOG_ANA);
268 mutex_unlock(&subsys->lock);
271 void nvmet_port_send_ana_event(struct nvmet_port *port)
273 struct nvmet_subsys_link *p;
275 down_read(&nvmet_config_sem);
276 list_for_each_entry(p, &port->subsystems, entry)
277 nvmet_send_ana_event(p->subsys, port);
278 up_read(&nvmet_config_sem);
281 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops)
285 down_write(&nvmet_config_sem);
286 if (nvmet_transports[ops->type])
289 nvmet_transports[ops->type] = ops;
290 up_write(&nvmet_config_sem);
294 EXPORT_SYMBOL_GPL(nvmet_register_transport);
296 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops)
298 down_write(&nvmet_config_sem);
299 nvmet_transports[ops->type] = NULL;
300 up_write(&nvmet_config_sem);
302 EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
304 void nvmet_port_del_ctrls(struct nvmet_port *port, struct nvmet_subsys *subsys)
306 struct nvmet_ctrl *ctrl;
308 mutex_lock(&subsys->lock);
309 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
310 if (ctrl->port == port)
311 ctrl->ops->delete_ctrl(ctrl);
313 mutex_unlock(&subsys->lock);
316 int nvmet_enable_port(struct nvmet_port *port)
318 const struct nvmet_fabrics_ops *ops;
321 lockdep_assert_held(&nvmet_config_sem);
323 ops = nvmet_transports[port->disc_addr.trtype];
325 up_write(&nvmet_config_sem);
326 request_module("nvmet-transport-%d", port->disc_addr.trtype);
327 down_write(&nvmet_config_sem);
328 ops = nvmet_transports[port->disc_addr.trtype];
330 pr_err("transport type %d not supported\n",
331 port->disc_addr.trtype);
336 if (!try_module_get(ops->owner))
340 * If the user requested PI support and the transport isn't pi capable,
341 * don't enable the port.
343 if (port->pi_enable && !(ops->flags & NVMF_METADATA_SUPPORTED)) {
344 pr_err("T10-PI is not supported by transport type %d\n",
345 port->disc_addr.trtype);
350 ret = ops->add_port(port);
354 /* If the transport didn't set inline_data_size, then disable it. */
355 if (port->inline_data_size < 0)
356 port->inline_data_size = 0;
358 port->enabled = true;
363 module_put(ops->owner);
367 void nvmet_disable_port(struct nvmet_port *port)
369 const struct nvmet_fabrics_ops *ops;
371 lockdep_assert_held(&nvmet_config_sem);
373 port->enabled = false;
376 ops = nvmet_transports[port->disc_addr.trtype];
377 ops->remove_port(port);
378 module_put(ops->owner);
381 static void nvmet_keep_alive_timer(struct work_struct *work)
383 struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
384 struct nvmet_ctrl, ka_work);
385 bool reset_tbkas = ctrl->reset_tbkas;
387 ctrl->reset_tbkas = false;
389 pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
391 queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
395 pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
396 ctrl->cntlid, ctrl->kato);
398 nvmet_ctrl_fatal_error(ctrl);
401 void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
403 if (unlikely(ctrl->kato == 0))
406 pr_debug("ctrl %d start keep-alive timer for %d secs\n",
407 ctrl->cntlid, ctrl->kato);
409 queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
412 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
414 if (unlikely(ctrl->kato == 0))
417 pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid);
419 cancel_delayed_work_sync(&ctrl->ka_work);
422 u16 nvmet_req_find_ns(struct nvmet_req *req)
424 u32 nsid = le32_to_cpu(req->cmd->common.nsid);
426 req->ns = xa_load(&nvmet_req_subsys(req)->namespaces, nsid);
427 if (unlikely(!req->ns)) {
428 req->error_loc = offsetof(struct nvme_common_command, nsid);
429 return NVME_SC_INVALID_NS | NVME_SC_DNR;
432 percpu_ref_get(&req->ns->ref);
433 return NVME_SC_SUCCESS;
436 static void nvmet_destroy_namespace(struct percpu_ref *ref)
438 struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref);
440 complete(&ns->disable_done);
443 void nvmet_put_namespace(struct nvmet_ns *ns)
445 percpu_ref_put(&ns->ref);
448 static void nvmet_ns_dev_disable(struct nvmet_ns *ns)
450 nvmet_bdev_ns_disable(ns);
451 nvmet_file_ns_disable(ns);
454 static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns)
457 struct pci_dev *p2p_dev;
463 pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n");
467 if (!blk_queue_pci_p2pdma(ns->bdev->bd_disk->queue)) {
468 pr_err("peer-to-peer DMA is not supported by the driver of %s\n",
474 ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true);
479 * Right now we just check that there is p2pmem available so
480 * we can report an error to the user right away if there
481 * is not. We'll find the actual device to use once we
482 * setup the controller when the port's device is available.
485 p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns));
487 pr_err("no peer-to-peer memory is available for %s\n",
492 pci_dev_put(p2p_dev);
499 * Note: ctrl->subsys->lock should be held when calling this function
501 static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl,
504 struct device *clients[2];
505 struct pci_dev *p2p_dev;
508 if (!ctrl->p2p_client || !ns->use_p2pmem)
512 ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true);
516 p2p_dev = pci_dev_get(ns->p2p_dev);
518 clients[0] = ctrl->p2p_client;
519 clients[1] = nvmet_ns_dev(ns);
521 p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients));
523 pr_err("no peer-to-peer memory is available that's supported by %s and %s\n",
524 dev_name(ctrl->p2p_client), ns->device_path);
529 ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev);
531 pci_dev_put(p2p_dev);
533 pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev),
537 bool nvmet_ns_revalidate(struct nvmet_ns *ns)
539 loff_t oldsize = ns->size;
542 nvmet_bdev_ns_revalidate(ns);
544 nvmet_file_ns_revalidate(ns);
546 return oldsize != ns->size;
549 int nvmet_ns_enable(struct nvmet_ns *ns)
551 struct nvmet_subsys *subsys = ns->subsys;
552 struct nvmet_ctrl *ctrl;
555 mutex_lock(&subsys->lock);
558 if (nvmet_is_passthru_subsys(subsys)) {
559 pr_info("cannot enable both passthru and regular namespaces for a single subsystem");
567 if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES)
570 ret = nvmet_bdev_ns_enable(ns);
572 ret = nvmet_file_ns_enable(ns);
576 ret = nvmet_p2pmem_ns_enable(ns);
578 goto out_dev_disable;
580 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
581 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
583 ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
588 if (ns->nsid > subsys->max_nsid)
589 subsys->max_nsid = ns->nsid;
591 ret = xa_insert(&subsys->namespaces, ns->nsid, ns, GFP_KERNEL);
593 goto out_restore_subsys_maxnsid;
595 subsys->nr_namespaces++;
597 nvmet_ns_changed(subsys, ns->nsid);
601 mutex_unlock(&subsys->lock);
604 out_restore_subsys_maxnsid:
605 subsys->max_nsid = nvmet_max_nsid(subsys);
606 percpu_ref_exit(&ns->ref);
608 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
609 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
611 nvmet_ns_dev_disable(ns);
615 void nvmet_ns_disable(struct nvmet_ns *ns)
617 struct nvmet_subsys *subsys = ns->subsys;
618 struct nvmet_ctrl *ctrl;
620 mutex_lock(&subsys->lock);
625 xa_erase(&ns->subsys->namespaces, ns->nsid);
626 if (ns->nsid == subsys->max_nsid)
627 subsys->max_nsid = nvmet_max_nsid(subsys);
629 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
630 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
632 mutex_unlock(&subsys->lock);
635 * Now that we removed the namespaces from the lookup list, we
636 * can kill the per_cpu ref and wait for any remaining references
637 * to be dropped, as well as a RCU grace period for anyone only
638 * using the namepace under rcu_read_lock(). Note that we can't
639 * use call_rcu here as we need to ensure the namespaces have
640 * been fully destroyed before unloading the module.
642 percpu_ref_kill(&ns->ref);
644 wait_for_completion(&ns->disable_done);
645 percpu_ref_exit(&ns->ref);
647 mutex_lock(&subsys->lock);
649 subsys->nr_namespaces--;
650 nvmet_ns_changed(subsys, ns->nsid);
651 nvmet_ns_dev_disable(ns);
653 mutex_unlock(&subsys->lock);
656 void nvmet_ns_free(struct nvmet_ns *ns)
658 nvmet_ns_disable(ns);
660 down_write(&nvmet_ana_sem);
661 nvmet_ana_group_enabled[ns->anagrpid]--;
662 up_write(&nvmet_ana_sem);
664 kfree(ns->device_path);
668 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
672 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
676 init_completion(&ns->disable_done);
681 down_write(&nvmet_ana_sem);
682 ns->anagrpid = NVMET_DEFAULT_ANA_GRPID;
683 nvmet_ana_group_enabled[ns->anagrpid]++;
684 up_write(&nvmet_ana_sem);
687 ns->buffered_io = false;
688 ns->csi = NVME_CSI_NVM;
693 static void nvmet_update_sq_head(struct nvmet_req *req)
696 u32 old_sqhd, new_sqhd;
699 old_sqhd = req->sq->sqhd;
700 new_sqhd = (old_sqhd + 1) % req->sq->size;
701 } while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
704 req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
707 static void nvmet_set_error(struct nvmet_req *req, u16 status)
709 struct nvmet_ctrl *ctrl = req->sq->ctrl;
710 struct nvme_error_slot *new_error_slot;
713 req->cqe->status = cpu_to_le16(status << 1);
715 if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC)
718 spin_lock_irqsave(&ctrl->error_lock, flags);
721 &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS];
723 new_error_slot->error_count = cpu_to_le64(ctrl->err_counter);
724 new_error_slot->sqid = cpu_to_le16(req->sq->qid);
725 new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id);
726 new_error_slot->status_field = cpu_to_le16(status << 1);
727 new_error_slot->param_error_location = cpu_to_le16(req->error_loc);
728 new_error_slot->lba = cpu_to_le64(req->error_slba);
729 new_error_slot->nsid = req->cmd->common.nsid;
730 spin_unlock_irqrestore(&ctrl->error_lock, flags);
732 /* set the more bit for this request */
733 req->cqe->status |= cpu_to_le16(1 << 14);
736 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
738 if (!req->sq->sqhd_disabled)
739 nvmet_update_sq_head(req);
740 req->cqe->sq_id = cpu_to_le16(req->sq->qid);
741 req->cqe->command_id = req->cmd->common.command_id;
743 if (unlikely(status))
744 nvmet_set_error(req, status);
746 trace_nvmet_req_complete(req);
749 nvmet_put_namespace(req->ns);
750 req->ops->queue_response(req);
753 void nvmet_req_complete(struct nvmet_req *req, u16 status)
755 __nvmet_req_complete(req, status);
756 percpu_ref_put(&req->sq->ref);
758 EXPORT_SYMBOL_GPL(nvmet_req_complete);
760 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq,
767 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq,
777 static void nvmet_confirm_sq(struct percpu_ref *ref)
779 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
781 complete(&sq->confirm_done);
784 void nvmet_sq_destroy(struct nvmet_sq *sq)
786 struct nvmet_ctrl *ctrl = sq->ctrl;
789 * If this is the admin queue, complete all AERs so that our
790 * queue doesn't have outstanding requests on it.
792 if (ctrl && ctrl->sqs && ctrl->sqs[0] == sq)
793 nvmet_async_events_failall(ctrl);
794 percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq);
795 wait_for_completion(&sq->confirm_done);
796 wait_for_completion(&sq->free_done);
797 percpu_ref_exit(&sq->ref);
801 * The teardown flow may take some time, and the host may not
802 * send us keep-alive during this period, hence reset the
803 * traffic based keep-alive timer so we don't trigger a
804 * controller teardown as a result of a keep-alive expiration.
806 ctrl->reset_tbkas = true;
807 sq->ctrl->sqs[sq->qid] = NULL;
808 nvmet_ctrl_put(ctrl);
809 sq->ctrl = NULL; /* allows reusing the queue later */
812 EXPORT_SYMBOL_GPL(nvmet_sq_destroy);
814 static void nvmet_sq_free(struct percpu_ref *ref)
816 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
818 complete(&sq->free_done);
821 int nvmet_sq_init(struct nvmet_sq *sq)
825 ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL);
827 pr_err("percpu_ref init failed!\n");
830 init_completion(&sq->free_done);
831 init_completion(&sq->confirm_done);
835 EXPORT_SYMBOL_GPL(nvmet_sq_init);
837 static inline u16 nvmet_check_ana_state(struct nvmet_port *port,
840 enum nvme_ana_state state = port->ana_state[ns->anagrpid];
842 if (unlikely(state == NVME_ANA_INACCESSIBLE))
843 return NVME_SC_ANA_INACCESSIBLE;
844 if (unlikely(state == NVME_ANA_PERSISTENT_LOSS))
845 return NVME_SC_ANA_PERSISTENT_LOSS;
846 if (unlikely(state == NVME_ANA_CHANGE))
847 return NVME_SC_ANA_TRANSITION;
851 static inline u16 nvmet_io_cmd_check_access(struct nvmet_req *req)
853 if (unlikely(req->ns->readonly)) {
854 switch (req->cmd->common.opcode) {
859 return NVME_SC_NS_WRITE_PROTECTED;
866 static u16 nvmet_parse_io_cmd(struct nvmet_req *req)
870 ret = nvmet_check_ctrl_status(req);
874 if (nvmet_is_passthru_req(req))
875 return nvmet_parse_passthru_io_cmd(req);
877 ret = nvmet_req_find_ns(req);
881 ret = nvmet_check_ana_state(req->port, req->ns);
883 req->error_loc = offsetof(struct nvme_common_command, nsid);
886 ret = nvmet_io_cmd_check_access(req);
888 req->error_loc = offsetof(struct nvme_common_command, nsid);
892 switch (req->ns->csi) {
895 return nvmet_file_parse_io_cmd(req);
896 return nvmet_bdev_parse_io_cmd(req);
898 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED))
899 return nvmet_bdev_zns_parse_io_cmd(req);
900 return NVME_SC_INVALID_IO_CMD_SET;
902 return NVME_SC_INVALID_IO_CMD_SET;
906 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
907 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops)
909 u8 flags = req->cmd->common.flags;
916 req->metadata_sg = NULL;
918 req->metadata_sg_cnt = 0;
919 req->transfer_len = 0;
920 req->metadata_len = 0;
921 req->cqe->status = 0;
922 req->cqe->sq_head = 0;
924 req->error_loc = NVMET_NO_ERROR_LOC;
927 /* no support for fused commands yet */
928 if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) {
929 req->error_loc = offsetof(struct nvme_common_command, flags);
930 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
935 * For fabrics, PSDT field shall describe metadata pointer (MPTR) that
936 * contains an address of a single contiguous physical buffer that is
939 if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) {
940 req->error_loc = offsetof(struct nvme_common_command, flags);
941 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
945 if (unlikely(!req->sq->ctrl))
946 /* will return an error for any non-connect command: */
947 status = nvmet_parse_connect_cmd(req);
948 else if (likely(req->sq->qid != 0))
949 status = nvmet_parse_io_cmd(req);
951 status = nvmet_parse_admin_cmd(req);
956 trace_nvmet_req_init(req, req->cmd);
958 if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
959 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
964 sq->ctrl->reset_tbkas = true;
969 __nvmet_req_complete(req, status);
972 EXPORT_SYMBOL_GPL(nvmet_req_init);
974 void nvmet_req_uninit(struct nvmet_req *req)
976 percpu_ref_put(&req->sq->ref);
978 nvmet_put_namespace(req->ns);
980 EXPORT_SYMBOL_GPL(nvmet_req_uninit);
982 bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len)
984 if (unlikely(len != req->transfer_len)) {
985 req->error_loc = offsetof(struct nvme_common_command, dptr);
986 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
992 EXPORT_SYMBOL_GPL(nvmet_check_transfer_len);
994 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len)
996 if (unlikely(data_len > req->transfer_len)) {
997 req->error_loc = offsetof(struct nvme_common_command, dptr);
998 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
1005 static unsigned int nvmet_data_transfer_len(struct nvmet_req *req)
1007 return req->transfer_len - req->metadata_len;
1010 static int nvmet_req_alloc_p2pmem_sgls(struct pci_dev *p2p_dev,
1011 struct nvmet_req *req)
1013 req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt,
1014 nvmet_data_transfer_len(req));
1018 if (req->metadata_len) {
1019 req->metadata_sg = pci_p2pmem_alloc_sgl(p2p_dev,
1020 &req->metadata_sg_cnt, req->metadata_len);
1021 if (!req->metadata_sg)
1025 req->p2p_dev = p2p_dev;
1029 pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1034 static struct pci_dev *nvmet_req_find_p2p_dev(struct nvmet_req *req)
1036 if (!IS_ENABLED(CONFIG_PCI_P2PDMA) ||
1037 !req->sq->ctrl || !req->sq->qid || !req->ns)
1039 return radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
1042 int nvmet_req_alloc_sgls(struct nvmet_req *req)
1044 struct pci_dev *p2p_dev = nvmet_req_find_p2p_dev(req);
1046 if (p2p_dev && !nvmet_req_alloc_p2pmem_sgls(p2p_dev, req))
1049 req->sg = sgl_alloc(nvmet_data_transfer_len(req), GFP_KERNEL,
1051 if (unlikely(!req->sg))
1054 if (req->metadata_len) {
1055 req->metadata_sg = sgl_alloc(req->metadata_len, GFP_KERNEL,
1056 &req->metadata_sg_cnt);
1057 if (unlikely(!req->metadata_sg))
1067 EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgls);
1069 void nvmet_req_free_sgls(struct nvmet_req *req)
1072 pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1073 if (req->metadata_sg)
1074 pci_p2pmem_free_sgl(req->p2p_dev, req->metadata_sg);
1075 req->p2p_dev = NULL;
1078 if (req->metadata_sg)
1079 sgl_free(req->metadata_sg);
1083 req->metadata_sg = NULL;
1085 req->metadata_sg_cnt = 0;
1087 EXPORT_SYMBOL_GPL(nvmet_req_free_sgls);
1089 static inline bool nvmet_cc_en(u32 cc)
1091 return (cc >> NVME_CC_EN_SHIFT) & 0x1;
1094 static inline u8 nvmet_cc_css(u32 cc)
1096 return (cc >> NVME_CC_CSS_SHIFT) & 0x7;
1099 static inline u8 nvmet_cc_mps(u32 cc)
1101 return (cc >> NVME_CC_MPS_SHIFT) & 0xf;
1104 static inline u8 nvmet_cc_ams(u32 cc)
1106 return (cc >> NVME_CC_AMS_SHIFT) & 0x7;
1109 static inline u8 nvmet_cc_shn(u32 cc)
1111 return (cc >> NVME_CC_SHN_SHIFT) & 0x3;
1114 static inline u8 nvmet_cc_iosqes(u32 cc)
1116 return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf;
1119 static inline u8 nvmet_cc_iocqes(u32 cc)
1121 return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf;
1124 static inline bool nvmet_css_supported(u8 cc_css)
1126 switch (cc_css << NVME_CC_CSS_SHIFT) {
1127 case NVME_CC_CSS_NVM:
1128 case NVME_CC_CSS_CSI:
1135 static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl)
1137 lockdep_assert_held(&ctrl->lock);
1140 * Only I/O controllers should verify iosqes,iocqes.
1141 * Strictly speaking, the spec says a discovery controller
1142 * should verify iosqes,iocqes are zeroed, however that
1143 * would break backwards compatibility, so don't enforce it.
1145 if (!nvmet_is_disc_subsys(ctrl->subsys) &&
1146 (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES ||
1147 nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES)) {
1148 ctrl->csts = NVME_CSTS_CFS;
1152 if (nvmet_cc_mps(ctrl->cc) != 0 ||
1153 nvmet_cc_ams(ctrl->cc) != 0 ||
1154 !nvmet_css_supported(nvmet_cc_css(ctrl->cc))) {
1155 ctrl->csts = NVME_CSTS_CFS;
1159 ctrl->csts = NVME_CSTS_RDY;
1162 * Controllers that are not yet enabled should not really enforce the
1163 * keep alive timeout, but we still want to track a timeout and cleanup
1164 * in case a host died before it enabled the controller. Hence, simply
1165 * reset the keep alive timer when the controller is enabled.
1168 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
1171 static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl)
1173 lockdep_assert_held(&ctrl->lock);
1175 /* XXX: tear down queues? */
1176 ctrl->csts &= ~NVME_CSTS_RDY;
1180 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new)
1184 mutex_lock(&ctrl->lock);
1188 if (nvmet_cc_en(new) && !nvmet_cc_en(old))
1189 nvmet_start_ctrl(ctrl);
1190 if (!nvmet_cc_en(new) && nvmet_cc_en(old))
1191 nvmet_clear_ctrl(ctrl);
1192 if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) {
1193 nvmet_clear_ctrl(ctrl);
1194 ctrl->csts |= NVME_CSTS_SHST_CMPLT;
1196 if (!nvmet_cc_shn(new) && nvmet_cc_shn(old))
1197 ctrl->csts &= ~NVME_CSTS_SHST_CMPLT;
1198 mutex_unlock(&ctrl->lock);
1201 static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
1203 /* command sets supported: NVMe command set: */
1204 ctrl->cap = (1ULL << 37);
1205 /* Controller supports one or more I/O Command Sets */
1206 ctrl->cap |= (1ULL << 43);
1207 /* CC.EN timeout in 500msec units: */
1208 ctrl->cap |= (15ULL << 24);
1209 /* maximum queue entries supported: */
1210 if (ctrl->ops->get_max_queue_size)
1211 ctrl->cap |= ctrl->ops->get_max_queue_size(ctrl) - 1;
1213 ctrl->cap |= NVMET_QUEUE_SIZE - 1;
1215 if (nvmet_is_passthru_subsys(ctrl->subsys))
1216 nvmet_passthrough_override_cap(ctrl);
1219 struct nvmet_ctrl *nvmet_ctrl_find_get(const char *subsysnqn,
1220 const char *hostnqn, u16 cntlid,
1221 struct nvmet_req *req)
1223 struct nvmet_ctrl *ctrl = NULL;
1224 struct nvmet_subsys *subsys;
1226 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1228 pr_warn("connect request for invalid subsystem %s!\n",
1230 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1234 mutex_lock(&subsys->lock);
1235 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
1236 if (ctrl->cntlid == cntlid) {
1237 if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) {
1238 pr_warn("hostnqn mismatch.\n");
1241 if (!kref_get_unless_zero(&ctrl->ref))
1249 ctrl = NULL; /* ctrl not found */
1250 pr_warn("could not find controller %d for subsys %s / host %s\n",
1251 cntlid, subsysnqn, hostnqn);
1252 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
1255 mutex_unlock(&subsys->lock);
1256 nvmet_subsys_put(subsys);
1261 u16 nvmet_check_ctrl_status(struct nvmet_req *req)
1263 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
1264 pr_err("got cmd %d while CC.EN == 0 on qid = %d\n",
1265 req->cmd->common.opcode, req->sq->qid);
1266 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1269 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
1270 pr_err("got cmd %d while CSTS.RDY == 0 on qid = %d\n",
1271 req->cmd->common.opcode, req->sq->qid);
1272 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1277 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn)
1279 struct nvmet_host_link *p;
1281 lockdep_assert_held(&nvmet_config_sem);
1283 if (subsys->allow_any_host)
1286 if (nvmet_is_disc_subsys(subsys)) /* allow all access to disc subsys */
1289 list_for_each_entry(p, &subsys->hosts, entry) {
1290 if (!strcmp(nvmet_host_name(p->host), hostnqn))
1298 * Note: ctrl->subsys->lock should be held when calling this function
1300 static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl,
1301 struct nvmet_req *req)
1303 struct nvmet_ns *ns;
1306 if (!req->p2p_client)
1309 ctrl->p2p_client = get_device(req->p2p_client);
1311 xa_for_each(&ctrl->subsys->namespaces, idx, ns)
1312 nvmet_p2pmem_ns_add_p2p(ctrl, ns);
1316 * Note: ctrl->subsys->lock should be held when calling this function
1318 static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl)
1320 struct radix_tree_iter iter;
1323 radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0)
1324 pci_dev_put(radix_tree_deref_slot(slot));
1326 put_device(ctrl->p2p_client);
1329 static void nvmet_fatal_error_handler(struct work_struct *work)
1331 struct nvmet_ctrl *ctrl =
1332 container_of(work, struct nvmet_ctrl, fatal_err_work);
1334 pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid);
1335 ctrl->ops->delete_ctrl(ctrl);
1338 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
1339 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp)
1341 struct nvmet_subsys *subsys;
1342 struct nvmet_ctrl *ctrl;
1346 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1347 subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1349 pr_warn("connect request for invalid subsystem %s!\n",
1351 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1352 req->error_loc = offsetof(struct nvme_common_command, dptr);
1356 down_read(&nvmet_config_sem);
1357 if (!nvmet_host_allowed(subsys, hostnqn)) {
1358 pr_info("connect by host %s for subsystem %s not allowed\n",
1359 hostnqn, subsysnqn);
1360 req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
1361 up_read(&nvmet_config_sem);
1362 status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
1363 req->error_loc = offsetof(struct nvme_common_command, dptr);
1364 goto out_put_subsystem;
1366 up_read(&nvmet_config_sem);
1368 status = NVME_SC_INTERNAL;
1369 ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
1371 goto out_put_subsystem;
1372 mutex_init(&ctrl->lock);
1374 ctrl->port = req->port;
1375 ctrl->ops = req->ops;
1377 #ifdef CONFIG_NVME_TARGET_PASSTHRU
1378 /* By default, set loop targets to clear IDS by default */
1379 if (ctrl->port->disc_addr.trtype == NVMF_TRTYPE_LOOP)
1380 subsys->clear_ids = 1;
1383 INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
1384 INIT_LIST_HEAD(&ctrl->async_events);
1385 INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL);
1386 INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
1387 INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer);
1389 memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
1390 memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE);
1392 kref_init(&ctrl->ref);
1393 ctrl->subsys = subsys;
1394 nvmet_init_cap(ctrl);
1395 WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL);
1397 ctrl->changed_ns_list = kmalloc_array(NVME_MAX_CHANGED_NAMESPACES,
1398 sizeof(__le32), GFP_KERNEL);
1399 if (!ctrl->changed_ns_list)
1402 ctrl->sqs = kcalloc(subsys->max_qid + 1,
1403 sizeof(struct nvmet_sq *),
1406 goto out_free_changed_ns_list;
1408 if (subsys->cntlid_min > subsys->cntlid_max)
1411 ret = ida_alloc_range(&cntlid_ida,
1412 subsys->cntlid_min, subsys->cntlid_max,
1415 status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR;
1421 * Discovery controllers may use some arbitrary high value
1422 * in order to cleanup stale discovery sessions
1424 if (nvmet_is_disc_subsys(ctrl->subsys) && !kato)
1425 kato = NVMET_DISC_KATO_MS;
1427 /* keep-alive timeout in seconds */
1428 ctrl->kato = DIV_ROUND_UP(kato, 1000);
1430 ctrl->err_counter = 0;
1431 spin_lock_init(&ctrl->error_lock);
1433 nvmet_start_keep_alive_timer(ctrl);
1435 mutex_lock(&subsys->lock);
1436 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
1437 nvmet_setup_p2p_ns_map(ctrl, req);
1438 mutex_unlock(&subsys->lock);
1445 out_free_changed_ns_list:
1446 kfree(ctrl->changed_ns_list);
1450 nvmet_subsys_put(subsys);
1455 static void nvmet_ctrl_free(struct kref *ref)
1457 struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref);
1458 struct nvmet_subsys *subsys = ctrl->subsys;
1460 mutex_lock(&subsys->lock);
1461 nvmet_release_p2p_ns_map(ctrl);
1462 list_del(&ctrl->subsys_entry);
1463 mutex_unlock(&subsys->lock);
1465 nvmet_stop_keep_alive_timer(ctrl);
1467 flush_work(&ctrl->async_event_work);
1468 cancel_work_sync(&ctrl->fatal_err_work);
1470 ida_free(&cntlid_ida, ctrl->cntlid);
1472 nvmet_async_events_free(ctrl);
1474 kfree(ctrl->changed_ns_list);
1477 nvmet_subsys_put(subsys);
1480 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl)
1482 kref_put(&ctrl->ref, nvmet_ctrl_free);
1485 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl)
1487 mutex_lock(&ctrl->lock);
1488 if (!(ctrl->csts & NVME_CSTS_CFS)) {
1489 ctrl->csts |= NVME_CSTS_CFS;
1490 queue_work(nvmet_wq, &ctrl->fatal_err_work);
1492 mutex_unlock(&ctrl->lock);
1494 EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error);
1496 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
1497 const char *subsysnqn)
1499 struct nvmet_subsys_link *p;
1504 if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) {
1505 if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
1507 return nvmet_disc_subsys;
1510 down_read(&nvmet_config_sem);
1511 list_for_each_entry(p, &port->subsystems, entry) {
1512 if (!strncmp(p->subsys->subsysnqn, subsysnqn,
1514 if (!kref_get_unless_zero(&p->subsys->ref))
1516 up_read(&nvmet_config_sem);
1520 up_read(&nvmet_config_sem);
1524 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
1525 enum nvme_subsys_type type)
1527 struct nvmet_subsys *subsys;
1528 char serial[NVMET_SN_MAX_SIZE / 2];
1531 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
1533 return ERR_PTR(-ENOMEM);
1535 subsys->ver = NVMET_DEFAULT_VS;
1536 /* generate a random serial number as our controllers are ephemeral: */
1537 get_random_bytes(&serial, sizeof(serial));
1538 bin2hex(subsys->serial, &serial, sizeof(serial));
1540 subsys->model_number = kstrdup(NVMET_DEFAULT_CTRL_MODEL, GFP_KERNEL);
1541 if (!subsys->model_number) {
1548 subsys->max_qid = NVMET_NR_QUEUES;
1552 subsys->max_qid = 0;
1555 pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
1559 subsys->type = type;
1560 subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
1562 if (!subsys->subsysnqn) {
1566 subsys->cntlid_min = NVME_CNTLID_MIN;
1567 subsys->cntlid_max = NVME_CNTLID_MAX;
1568 kref_init(&subsys->ref);
1570 mutex_init(&subsys->lock);
1571 xa_init(&subsys->namespaces);
1572 INIT_LIST_HEAD(&subsys->ctrls);
1573 INIT_LIST_HEAD(&subsys->hosts);
1578 kfree(subsys->model_number);
1581 return ERR_PTR(ret);
1584 static void nvmet_subsys_free(struct kref *ref)
1586 struct nvmet_subsys *subsys =
1587 container_of(ref, struct nvmet_subsys, ref);
1589 WARN_ON_ONCE(!xa_empty(&subsys->namespaces));
1591 xa_destroy(&subsys->namespaces);
1592 nvmet_passthru_subsys_free(subsys);
1594 kfree(subsys->subsysnqn);
1595 kfree(subsys->model_number);
1599 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys)
1601 struct nvmet_ctrl *ctrl;
1603 mutex_lock(&subsys->lock);
1604 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
1605 ctrl->ops->delete_ctrl(ctrl);
1606 mutex_unlock(&subsys->lock);
1609 void nvmet_subsys_put(struct nvmet_subsys *subsys)
1611 kref_put(&subsys->ref, nvmet_subsys_free);
1614 static int __init nvmet_init(void)
1618 nvmet_ana_group_enabled[NVMET_DEFAULT_ANA_GRPID] = 1;
1620 zbd_wq = alloc_workqueue("nvmet-zbd-wq", WQ_MEM_RECLAIM, 0);
1624 buffered_io_wq = alloc_workqueue("nvmet-buffered-io-wq",
1626 if (!buffered_io_wq) {
1628 goto out_free_zbd_work_queue;
1631 nvmet_wq = alloc_workqueue("nvmet-wq", WQ_MEM_RECLAIM, 0);
1634 goto out_free_buffered_work_queue;
1637 error = nvmet_init_discovery();
1639 goto out_free_nvmet_work_queue;
1641 error = nvmet_init_configfs();
1643 goto out_exit_discovery;
1647 nvmet_exit_discovery();
1648 out_free_nvmet_work_queue:
1649 destroy_workqueue(nvmet_wq);
1650 out_free_buffered_work_queue:
1651 destroy_workqueue(buffered_io_wq);
1652 out_free_zbd_work_queue:
1653 destroy_workqueue(zbd_wq);
1657 static void __exit nvmet_exit(void)
1659 nvmet_exit_configfs();
1660 nvmet_exit_discovery();
1661 ida_destroy(&cntlid_ida);
1662 destroy_workqueue(nvmet_wq);
1663 destroy_workqueue(buffered_io_wq);
1664 destroy_workqueue(zbd_wq);
1666 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024);
1667 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024);
1670 module_init(nvmet_init);
1671 module_exit(nvmet_exit);
1673 MODULE_LICENSE("GPL v2");