2 * NVMe over Fabrics common host code.
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/init.h>
16 #include <linux/miscdevice.h>
17 #include <linux/module.h>
18 #include <linux/mutex.h>
19 #include <linux/parser.h>
20 #include <linux/seq_file.h>
24 static LIST_HEAD(nvmf_transports);
25 static DECLARE_RWSEM(nvmf_transports_rwsem);
27 static LIST_HEAD(nvmf_hosts);
28 static DEFINE_MUTEX(nvmf_hosts_mutex);
30 static struct nvmf_host *nvmf_default_host;
32 static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
34 struct nvmf_host *host;
36 list_for_each_entry(host, &nvmf_hosts, list) {
37 if (!strcmp(host->nqn, hostnqn))
44 static struct nvmf_host *nvmf_host_add(const char *hostnqn)
46 struct nvmf_host *host;
48 mutex_lock(&nvmf_hosts_mutex);
49 host = __nvmf_host_find(hostnqn);
55 host = kmalloc(sizeof(*host), GFP_KERNEL);
59 kref_init(&host->ref);
60 strlcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
62 list_add_tail(&host->list, &nvmf_hosts);
64 mutex_unlock(&nvmf_hosts_mutex);
68 static struct nvmf_host *nvmf_host_default(void)
70 struct nvmf_host *host;
72 host = kmalloc(sizeof(*host), GFP_KERNEL);
76 kref_init(&host->ref);
78 snprintf(host->nqn, NVMF_NQN_SIZE,
79 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id);
81 mutex_lock(&nvmf_hosts_mutex);
82 list_add_tail(&host->list, &nvmf_hosts);
83 mutex_unlock(&nvmf_hosts_mutex);
88 static void nvmf_host_destroy(struct kref *ref)
90 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
92 mutex_lock(&nvmf_hosts_mutex);
93 list_del(&host->list);
94 mutex_unlock(&nvmf_hosts_mutex);
99 static void nvmf_host_put(struct nvmf_host *host)
102 kref_put(&host->ref, nvmf_host_destroy);
106 * nvmf_get_address() - Get address/port
107 * @ctrl: Host NVMe controller instance which we got the address
108 * @buf: OUTPUT parameter that will contain the address/port
111 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
115 if (ctrl->opts->mask & NVMF_OPT_TRADDR)
116 len += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
117 if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
118 len += snprintf(buf + len, size - len, "%strsvcid=%s",
119 (len) ? "," : "", ctrl->opts->trsvcid);
120 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
121 len += snprintf(buf + len, size - len, "%shost_traddr=%s",
122 (len) ? "," : "", ctrl->opts->host_traddr);
123 len += snprintf(buf + len, size - len, "\n");
127 EXPORT_SYMBOL_GPL(nvmf_get_address);
130 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function.
131 * @ctrl: Host NVMe controller instance maintaining the admin
132 * queue used to submit the property read command to
133 * the allocated NVMe controller resource on the target system.
134 * @off: Starting offset value of the targeted property
135 * register (see the fabrics section of the NVMe standard).
136 * @val: OUTPUT parameter that will contain the value of
137 * the property after a successful read.
139 * Used by the host system to retrieve a 32-bit capsule property value
140 * from an NVMe controller on the target system.
142 * ("Capsule property" is an "PCIe register concept" applied to the
143 * NVMe fabrics space.)
147 * > 0: NVMe error status code
148 * < 0: Linux errno error code
150 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
152 struct nvme_command cmd;
153 union nvme_result res;
156 memset(&cmd, 0, sizeof(cmd));
157 cmd.prop_get.opcode = nvme_fabrics_command;
158 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
159 cmd.prop_get.offset = cpu_to_le32(off);
161 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
165 *val = le64_to_cpu(res.u64);
166 if (unlikely(ret != 0))
167 dev_err(ctrl->device,
168 "Property Get error: %d, offset %#x\n",
169 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
173 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
176 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function.
177 * @ctrl: Host NVMe controller instance maintaining the admin
178 * queue used to submit the property read command to
179 * the allocated controller resource on the target system.
180 * @off: Starting offset value of the targeted property
181 * register (see the fabrics section of the NVMe standard).
182 * @val: OUTPUT parameter that will contain the value of
183 * the property after a successful read.
185 * Used by the host system to retrieve a 64-bit capsule property value
186 * from an NVMe controller on the target system.
188 * ("Capsule property" is an "PCIe register concept" applied to the
189 * NVMe fabrics space.)
193 * > 0: NVMe error status code
194 * < 0: Linux errno error code
196 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
198 struct nvme_command cmd;
199 union nvme_result res;
202 memset(&cmd, 0, sizeof(cmd));
203 cmd.prop_get.opcode = nvme_fabrics_command;
204 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
205 cmd.prop_get.attrib = 1;
206 cmd.prop_get.offset = cpu_to_le32(off);
208 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
212 *val = le64_to_cpu(res.u64);
213 if (unlikely(ret != 0))
214 dev_err(ctrl->device,
215 "Property Get error: %d, offset %#x\n",
216 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
219 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
222 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function.
223 * @ctrl: Host NVMe controller instance maintaining the admin
224 * queue used to submit the property read command to
225 * the allocated NVMe controller resource on the target system.
226 * @off: Starting offset value of the targeted property
227 * register (see the fabrics section of the NVMe standard).
228 * @val: Input parameter that contains the value to be
229 * written to the property.
231 * Used by the NVMe host system to write a 32-bit capsule property value
232 * to an NVMe controller on the target system.
234 * ("Capsule property" is an "PCIe register concept" applied to the
235 * NVMe fabrics space.)
238 * 0: successful write
239 * > 0: NVMe error status code
240 * < 0: Linux errno error code
242 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
244 struct nvme_command cmd;
247 memset(&cmd, 0, sizeof(cmd));
248 cmd.prop_set.opcode = nvme_fabrics_command;
249 cmd.prop_set.fctype = nvme_fabrics_type_property_set;
250 cmd.prop_set.attrib = 0;
251 cmd.prop_set.offset = cpu_to_le32(off);
252 cmd.prop_set.value = cpu_to_le64(val);
254 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
257 dev_err(ctrl->device,
258 "Property Set error: %d, offset %#x\n",
259 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
262 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
265 * nvmf_log_connect_error() - Error-parsing-diagnostic print
266 * out function for connect() errors.
268 * @ctrl: the specific /dev/nvmeX device that had the error.
270 * @errval: Error code to be decoded in a more human-friendly
273 * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
275 * @cmd: This is the SQE portion of a submission capsule.
277 * @data: This is the "Data" portion of a submission capsule.
279 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
280 int errval, int offset, struct nvme_command *cmd,
281 struct nvmf_connect_data *data)
283 int err_sctype = errval & (~NVME_SC_DNR);
285 switch (err_sctype) {
287 case (NVME_SC_CONNECT_INVALID_PARAM):
289 char *inv_data = "Connect Invalid Data Parameter";
291 switch (offset & 0xffff) {
292 case (offsetof(struct nvmf_connect_data, cntlid)):
293 dev_err(ctrl->device,
295 inv_data, data->cntlid);
297 case (offsetof(struct nvmf_connect_data, hostnqn)):
298 dev_err(ctrl->device,
299 "%s, hostnqn \"%s\"\n",
300 inv_data, data->hostnqn);
302 case (offsetof(struct nvmf_connect_data, subsysnqn)):
303 dev_err(ctrl->device,
304 "%s, subsysnqn \"%s\"\n",
305 inv_data, data->subsysnqn);
308 dev_err(ctrl->device,
309 "%s, starting byte offset: %d\n",
310 inv_data, offset & 0xffff);
314 char *inv_sqe = "Connect Invalid SQE Parameter";
317 case (offsetof(struct nvmf_connect_command, qid)):
318 dev_err(ctrl->device,
320 inv_sqe, cmd->connect.qid);
323 dev_err(ctrl->device,
324 "%s, starting byte offset: %d\n",
330 case NVME_SC_CONNECT_INVALID_HOST:
331 dev_err(ctrl->device,
332 "Connect for subsystem %s is not allowed, hostnqn: %s\n",
333 data->subsysnqn, data->hostnqn);
336 case NVME_SC_CONNECT_CTRL_BUSY:
337 dev_err(ctrl->device,
338 "Connect command failed: controller is busy or not available\n");
341 case NVME_SC_CONNECT_FORMAT:
342 dev_err(ctrl->device,
343 "Connect incompatible format: %d",
344 cmd->connect.recfmt);
348 dev_err(ctrl->device,
349 "Connect command failed, error wo/DNR bit: %d\n",
352 } /* switch (err_sctype) */
356 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
358 * @ctrl: Host nvme controller instance used to request
359 * a new NVMe controller allocation on the target
360 * system and establish an NVMe Admin connection to
363 * This function enables an NVMe host device to request a new allocation of
364 * an NVMe controller resource on a target system as well establish a
365 * fabrics-protocol connection of the NVMe Admin queue between the
366 * host system device and the allocated NVMe controller on the
367 * target system via a NVMe Fabrics "Connect" command.
371 * > 0: NVMe error status code
372 * < 0: Linux errno error code
375 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
377 struct nvme_command cmd;
378 union nvme_result res;
379 struct nvmf_connect_data *data;
382 memset(&cmd, 0, sizeof(cmd));
383 cmd.connect.opcode = nvme_fabrics_command;
384 cmd.connect.fctype = nvme_fabrics_type_connect;
386 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
389 * Set keep-alive timeout in seconds granularity (ms * 1000)
390 * and add a grace period for controller kato enforcement
392 cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
393 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
395 data = kzalloc(sizeof(*data), GFP_KERNEL);
399 uuid_copy(&data->hostid, &ctrl->opts->host->id);
400 data->cntlid = cpu_to_le16(0xffff);
401 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
402 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
404 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,
405 data, sizeof(*data), 0, NVME_QID_ANY, 1,
406 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
408 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
413 ctrl->cntlid = le16_to_cpu(res.u16);
419 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
422 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
424 * @ctrl: Host nvme controller instance used to establish an
425 * NVMe I/O queue connection to the already allocated NVMe
426 * controller on the target system.
427 * @qid: NVMe I/O queue number for the new I/O connection between
428 * host and target (note qid == 0 is illegal as this is
429 * the Admin queue, per NVMe standard).
431 * This function issues a fabrics-protocol connection
432 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
433 * between the host system device and the allocated NVMe controller
434 * on the target system.
438 * > 0: NVMe error status code
439 * < 0: Linux errno error code
441 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
443 struct nvme_command cmd;
444 struct nvmf_connect_data *data;
445 union nvme_result res;
448 memset(&cmd, 0, sizeof(cmd));
449 cmd.connect.opcode = nvme_fabrics_command;
450 cmd.connect.fctype = nvme_fabrics_type_connect;
451 cmd.connect.qid = cpu_to_le16(qid);
452 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
454 data = kzalloc(sizeof(*data), GFP_KERNEL);
458 uuid_copy(&data->hostid, &ctrl->opts->host->id);
459 data->cntlid = cpu_to_le16(ctrl->cntlid);
460 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
461 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
463 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
464 data, sizeof(*data), 0, qid, 1,
465 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
467 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
473 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
475 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
477 if (ctrl->opts->max_reconnects == -1 ||
478 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
483 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
486 * nvmf_register_transport() - NVMe Fabrics Library registration function.
487 * @ops: Transport ops instance to be registered to the
488 * common fabrics library.
490 * API function that registers the type of specific transport fabric
491 * being implemented to the common NVMe fabrics library. Part of
492 * the overall init sequence of starting up a fabrics driver.
494 int nvmf_register_transport(struct nvmf_transport_ops *ops)
496 if (!ops->create_ctrl)
499 down_write(&nvmf_transports_rwsem);
500 list_add_tail(&ops->entry, &nvmf_transports);
501 up_write(&nvmf_transports_rwsem);
505 EXPORT_SYMBOL_GPL(nvmf_register_transport);
508 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
509 * @ops: Transport ops instance to be unregistered from the
510 * common fabrics library.
512 * Fabrics API function that unregisters the type of specific transport
513 * fabric being implemented from the common NVMe fabrics library.
514 * Part of the overall exit sequence of unloading the implemented driver.
516 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
518 down_write(&nvmf_transports_rwsem);
519 list_del(&ops->entry);
520 up_write(&nvmf_transports_rwsem);
522 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
524 static struct nvmf_transport_ops *nvmf_lookup_transport(
525 struct nvmf_ctrl_options *opts)
527 struct nvmf_transport_ops *ops;
529 lockdep_assert_held(&nvmf_transports_rwsem);
531 list_for_each_entry(ops, &nvmf_transports, entry) {
532 if (strcmp(ops->name, opts->transport) == 0)
540 * For something we're not in a state to send to the device the default action
541 * is to busy it and retry it after the controller state is recovered. However,
542 * if the controller is deleting or if anything is marked for failfast or
543 * nvme multipath it is immediately failed.
545 * Note: commands used to initialize the controller will be marked for failfast.
546 * Note: nvme cli/ioctl commands are marked for failfast.
548 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
551 if (ctrl->state != NVME_CTRL_DELETING &&
552 ctrl->state != NVME_CTRL_DEAD &&
553 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
554 return BLK_STS_RESOURCE;
556 nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR;
557 blk_mq_start_request(rq);
558 nvme_complete_rq(rq);
561 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
563 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
566 struct nvme_request *req = nvme_req(rq);
569 * If we are in some state of setup or teardown only allow
570 * internally generated commands.
572 if (!blk_rq_is_passthrough(rq) || (req->flags & NVME_REQ_USERCMD))
576 * Only allow commands on a live queue, except for the connect command,
577 * which is require to set the queue live in the appropinquate states.
579 switch (ctrl->state) {
581 case NVME_CTRL_CONNECTING:
582 if (req->cmd->common.opcode == nvme_fabrics_command &&
583 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
594 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
596 static const match_table_t opt_tokens = {
597 { NVMF_OPT_TRANSPORT, "transport=%s" },
598 { NVMF_OPT_TRADDR, "traddr=%s" },
599 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
600 { NVMF_OPT_NQN, "nqn=%s" },
601 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
602 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
603 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
604 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
605 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
606 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
607 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
608 { NVMF_OPT_HOST_ID, "hostid=%s" },
609 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
610 { NVMF_OPT_ERR, NULL }
613 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
616 substring_t args[MAX_OPT_ARGS];
617 char *options, *o, *p;
620 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
624 opts->queue_size = NVMF_DEF_QUEUE_SIZE;
625 opts->nr_io_queues = num_online_cpus();
626 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
627 opts->kato = NVME_DEFAULT_KATO;
628 opts->duplicate_connect = false;
630 options = o = kstrdup(buf, GFP_KERNEL);
636 while ((p = strsep(&o, ",\n")) != NULL) {
640 token = match_token(p, opt_tokens, args);
643 case NVMF_OPT_TRANSPORT:
644 p = match_strdup(args);
649 kfree(opts->transport);
653 p = match_strdup(args);
658 kfree(opts->subsysnqn);
660 nqnlen = strlen(opts->subsysnqn);
661 if (nqnlen >= NVMF_NQN_SIZE) {
662 pr_err("%s needs to be < %d bytes\n",
663 opts->subsysnqn, NVMF_NQN_SIZE);
667 opts->discovery_nqn =
668 !(strcmp(opts->subsysnqn,
669 NVME_DISC_SUBSYS_NAME));
671 case NVMF_OPT_TRADDR:
672 p = match_strdup(args);
680 case NVMF_OPT_TRSVCID:
681 p = match_strdup(args);
686 kfree(opts->trsvcid);
689 case NVMF_OPT_QUEUE_SIZE:
690 if (match_int(args, &token)) {
694 if (token < NVMF_MIN_QUEUE_SIZE ||
695 token > NVMF_MAX_QUEUE_SIZE) {
696 pr_err("Invalid queue_size %d\n", token);
700 opts->queue_size = token;
702 case NVMF_OPT_NR_IO_QUEUES:
703 if (match_int(args, &token)) {
708 pr_err("Invalid number of IOQs %d\n", token);
712 if (opts->discovery_nqn) {
713 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
717 opts->nr_io_queues = min_t(unsigned int,
718 num_online_cpus(), token);
721 if (match_int(args, &token)) {
727 pr_err("Invalid keep_alive_tmo %d\n", token);
730 } else if (token == 0 && !opts->discovery_nqn) {
731 /* Allowed for debug */
732 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
736 if (opts->discovery_nqn && opts->kato) {
737 pr_err("Discovery controllers cannot accept KATO != 0\n");
743 case NVMF_OPT_CTRL_LOSS_TMO:
744 if (match_int(args, &token)) {
750 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
751 ctrl_loss_tmo = token;
753 case NVMF_OPT_HOSTNQN:
755 pr_err("hostnqn already user-assigned: %s\n",
760 p = match_strdup(args);
766 if (nqnlen >= NVMF_NQN_SIZE) {
767 pr_err("%s needs to be < %d bytes\n",
773 nvmf_host_put(opts->host);
774 opts->host = nvmf_host_add(p);
781 case NVMF_OPT_RECONNECT_DELAY:
782 if (match_int(args, &token)) {
787 pr_err("Invalid reconnect_delay %d\n", token);
791 opts->reconnect_delay = token;
793 case NVMF_OPT_HOST_TRADDR:
794 p = match_strdup(args);
799 kfree(opts->host_traddr);
800 opts->host_traddr = p;
802 case NVMF_OPT_HOST_ID:
803 p = match_strdup(args);
808 ret = uuid_parse(p, &hostid);
810 pr_err("Invalid hostid %s\n", p);
817 case NVMF_OPT_DUP_CONNECT:
818 opts->duplicate_connect = true;
821 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
828 if (opts->discovery_nqn) {
830 opts->nr_io_queues = 0;
831 opts->duplicate_connect = true;
833 if (ctrl_loss_tmo < 0)
834 opts->max_reconnects = -1;
836 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
837 opts->reconnect_delay);
840 kref_get(&nvmf_default_host->ref);
841 opts->host = nvmf_default_host;
844 uuid_copy(&opts->host->id, &hostid);
851 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
852 unsigned int required_opts)
854 if ((opts->mask & required_opts) != required_opts) {
857 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
858 if ((opt_tokens[i].token & required_opts) &&
859 !(opt_tokens[i].token & opts->mask)) {
860 pr_warn("missing parameter '%s'\n",
861 opt_tokens[i].pattern);
871 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
872 struct nvmf_ctrl_options *opts)
874 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
875 strcmp(opts->traddr, ctrl->opts->traddr) ||
876 strcmp(opts->trsvcid, ctrl->opts->trsvcid))
880 * Checking the local address is rough. In most cases, none is specified
881 * and the host port is selected by the stack.
883 * Assume no match if:
884 * - local address is specified and address is not the same
885 * - local address is not specified but remote is, or vice versa
886 * (admin using specific host_traddr when it matters).
888 if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
889 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
890 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
892 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
893 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
899 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
901 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
902 unsigned int allowed_opts)
904 if (opts->mask & ~allowed_opts) {
907 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
908 if ((opt_tokens[i].token & opts->mask) &&
909 (opt_tokens[i].token & ~allowed_opts)) {
910 pr_warn("invalid parameter '%s'\n",
911 opt_tokens[i].pattern);
921 void nvmf_free_options(struct nvmf_ctrl_options *opts)
923 nvmf_host_put(opts->host);
924 kfree(opts->transport);
926 kfree(opts->trsvcid);
927 kfree(opts->subsysnqn);
928 kfree(opts->host_traddr);
931 EXPORT_SYMBOL_GPL(nvmf_free_options);
933 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
934 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
935 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
936 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT)
938 static struct nvme_ctrl *
939 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
941 struct nvmf_ctrl_options *opts;
942 struct nvmf_transport_ops *ops;
943 struct nvme_ctrl *ctrl;
946 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
948 return ERR_PTR(-ENOMEM);
950 ret = nvmf_parse_options(opts, buf);
955 request_module("nvme-%s", opts->transport);
958 * Check the generic options first as we need a valid transport for
959 * the lookup below. Then clear the generic flags so that transport
960 * drivers don't have to care about them.
962 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
965 opts->mask &= ~NVMF_REQUIRED_OPTS;
967 down_read(&nvmf_transports_rwsem);
968 ops = nvmf_lookup_transport(opts);
970 pr_info("no handler found for transport %s.\n",
976 if (!try_module_get(ops->module)) {
980 up_read(&nvmf_transports_rwsem);
982 ret = nvmf_check_required_opts(opts, ops->required_opts);
985 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
986 ops->allowed_opts | ops->required_opts);
990 ctrl = ops->create_ctrl(dev, opts);
996 module_put(ops->module);
1000 module_put(ops->module);
1003 up_read(&nvmf_transports_rwsem);
1005 nvmf_free_options(opts);
1006 return ERR_PTR(ret);
1009 static struct class *nvmf_class;
1010 static struct device *nvmf_device;
1011 static DEFINE_MUTEX(nvmf_dev_mutex);
1013 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1014 size_t count, loff_t *pos)
1016 struct seq_file *seq_file = file->private_data;
1017 struct nvme_ctrl *ctrl;
1021 if (count > PAGE_SIZE)
1024 buf = memdup_user_nul(ubuf, count);
1026 return PTR_ERR(buf);
1028 mutex_lock(&nvmf_dev_mutex);
1029 if (seq_file->private) {
1034 ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
1036 ret = PTR_ERR(ctrl);
1040 seq_file->private = ctrl;
1043 mutex_unlock(&nvmf_dev_mutex);
1045 return ret ? ret : count;
1048 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1050 struct nvme_ctrl *ctrl;
1053 mutex_lock(&nvmf_dev_mutex);
1054 ctrl = seq_file->private;
1060 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1061 ctrl->instance, ctrl->cntlid);
1064 mutex_unlock(&nvmf_dev_mutex);
1068 static int nvmf_dev_open(struct inode *inode, struct file *file)
1071 * The miscdevice code initializes file->private_data, but doesn't
1072 * make use of it later.
1074 file->private_data = NULL;
1075 return single_open(file, nvmf_dev_show, NULL);
1078 static int nvmf_dev_release(struct inode *inode, struct file *file)
1080 struct seq_file *seq_file = file->private_data;
1081 struct nvme_ctrl *ctrl = seq_file->private;
1084 nvme_put_ctrl(ctrl);
1085 return single_release(inode, file);
1088 static const struct file_operations nvmf_dev_fops = {
1089 .owner = THIS_MODULE,
1090 .write = nvmf_dev_write,
1092 .open = nvmf_dev_open,
1093 .release = nvmf_dev_release,
1096 static struct miscdevice nvmf_misc = {
1097 .minor = MISC_DYNAMIC_MINOR,
1098 .name = "nvme-fabrics",
1099 .fops = &nvmf_dev_fops,
1102 static int __init nvmf_init(void)
1106 nvmf_default_host = nvmf_host_default();
1107 if (!nvmf_default_host)
1110 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1111 if (IS_ERR(nvmf_class)) {
1112 pr_err("couldn't register class nvme-fabrics\n");
1113 ret = PTR_ERR(nvmf_class);
1118 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1119 if (IS_ERR(nvmf_device)) {
1120 pr_err("couldn't create nvme-fabris device!\n");
1121 ret = PTR_ERR(nvmf_device);
1122 goto out_destroy_class;
1125 ret = misc_register(&nvmf_misc);
1127 pr_err("couldn't register misc device: %d\n", ret);
1128 goto out_destroy_device;
1134 device_destroy(nvmf_class, MKDEV(0, 0));
1136 class_destroy(nvmf_class);
1138 nvmf_host_put(nvmf_default_host);
1142 static void __exit nvmf_exit(void)
1144 misc_deregister(&nvmf_misc);
1145 device_destroy(nvmf_class, MKDEV(0, 0));
1146 class_destroy(nvmf_class);
1147 nvmf_host_put(nvmf_default_host);
1149 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1150 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1151 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1152 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1155 MODULE_LICENSE("GPL v2");
1157 module_init(nvmf_init);
1158 module_exit(nvmf_exit);