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);
347 case NVME_SC_HOST_PATH_ERROR:
348 dev_err(ctrl->device,
349 "Connect command failed: host path error\n");
353 dev_err(ctrl->device,
354 "Connect command failed, error wo/DNR bit: %d\n",
357 } /* switch (err_sctype) */
361 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
363 * @ctrl: Host nvme controller instance used to request
364 * a new NVMe controller allocation on the target
365 * system and establish an NVMe Admin connection to
368 * This function enables an NVMe host device to request a new allocation of
369 * an NVMe controller resource on a target system as well establish a
370 * fabrics-protocol connection of the NVMe Admin queue between the
371 * host system device and the allocated NVMe controller on the
372 * target system via a NVMe Fabrics "Connect" command.
376 * > 0: NVMe error status code
377 * < 0: Linux errno error code
380 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
382 struct nvme_command cmd;
383 union nvme_result res;
384 struct nvmf_connect_data *data;
387 memset(&cmd, 0, sizeof(cmd));
388 cmd.connect.opcode = nvme_fabrics_command;
389 cmd.connect.fctype = nvme_fabrics_type_connect;
391 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
394 * Set keep-alive timeout in seconds granularity (ms * 1000)
395 * and add a grace period for controller kato enforcement
397 cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
398 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
400 data = kzalloc(sizeof(*data), GFP_KERNEL);
404 uuid_copy(&data->hostid, &ctrl->opts->host->id);
405 data->cntlid = cpu_to_le16(0xffff);
406 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
407 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
409 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,
410 data, sizeof(*data), 0, NVME_QID_ANY, 1,
411 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
413 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
418 ctrl->cntlid = le16_to_cpu(res.u16);
424 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
427 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
429 * @ctrl: Host nvme controller instance used to establish an
430 * NVMe I/O queue connection to the already allocated NVMe
431 * controller on the target system.
432 * @qid: NVMe I/O queue number for the new I/O connection between
433 * host and target (note qid == 0 is illegal as this is
434 * the Admin queue, per NVMe standard).
436 * This function issues a fabrics-protocol connection
437 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
438 * between the host system device and the allocated NVMe controller
439 * on the target system.
443 * > 0: NVMe error status code
444 * < 0: Linux errno error code
446 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
448 struct nvme_command cmd;
449 struct nvmf_connect_data *data;
450 union nvme_result res;
453 memset(&cmd, 0, sizeof(cmd));
454 cmd.connect.opcode = nvme_fabrics_command;
455 cmd.connect.fctype = nvme_fabrics_type_connect;
456 cmd.connect.qid = cpu_to_le16(qid);
457 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
459 data = kzalloc(sizeof(*data), GFP_KERNEL);
463 uuid_copy(&data->hostid, &ctrl->opts->host->id);
464 data->cntlid = cpu_to_le16(ctrl->cntlid);
465 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
466 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
468 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
469 data, sizeof(*data), 0, qid, 1,
470 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
472 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
478 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
480 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
482 if (ctrl->opts->max_reconnects == -1 ||
483 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
488 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
491 * nvmf_register_transport() - NVMe Fabrics Library registration function.
492 * @ops: Transport ops instance to be registered to the
493 * common fabrics library.
495 * API function that registers the type of specific transport fabric
496 * being implemented to the common NVMe fabrics library. Part of
497 * the overall init sequence of starting up a fabrics driver.
499 int nvmf_register_transport(struct nvmf_transport_ops *ops)
501 if (!ops->create_ctrl)
504 down_write(&nvmf_transports_rwsem);
505 list_add_tail(&ops->entry, &nvmf_transports);
506 up_write(&nvmf_transports_rwsem);
510 EXPORT_SYMBOL_GPL(nvmf_register_transport);
513 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
514 * @ops: Transport ops instance to be unregistered from the
515 * common fabrics library.
517 * Fabrics API function that unregisters the type of specific transport
518 * fabric being implemented from the common NVMe fabrics library.
519 * Part of the overall exit sequence of unloading the implemented driver.
521 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
523 down_write(&nvmf_transports_rwsem);
524 list_del(&ops->entry);
525 up_write(&nvmf_transports_rwsem);
527 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
529 static struct nvmf_transport_ops *nvmf_lookup_transport(
530 struct nvmf_ctrl_options *opts)
532 struct nvmf_transport_ops *ops;
534 lockdep_assert_held(&nvmf_transports_rwsem);
536 list_for_each_entry(ops, &nvmf_transports, entry) {
537 if (strcmp(ops->name, opts->transport) == 0)
545 * For something we're not in a state to send to the device the default action
546 * is to busy it and retry it after the controller state is recovered. However,
547 * if the controller is deleting or if anything is marked for failfast or
548 * nvme multipath it is immediately failed.
550 * Note: commands used to initialize the controller will be marked for failfast.
551 * Note: nvme cli/ioctl commands are marked for failfast.
553 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
556 if (ctrl->state != NVME_CTRL_DELETING &&
557 ctrl->state != NVME_CTRL_DEAD &&
558 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
559 return BLK_STS_RESOURCE;
561 nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR;
562 blk_mq_start_request(rq);
563 nvme_complete_rq(rq);
566 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
568 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
571 struct nvme_request *req = nvme_req(rq);
574 * If we are in some state of setup or teardown only allow
575 * internally generated commands.
577 if (!blk_rq_is_passthrough(rq) || (req->flags & NVME_REQ_USERCMD))
581 * Only allow commands on a live queue, except for the connect command,
582 * which is require to set the queue live in the appropinquate states.
584 switch (ctrl->state) {
585 case NVME_CTRL_CONNECTING:
586 if (req->cmd->common.opcode == nvme_fabrics_command &&
587 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
598 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
600 static const match_table_t opt_tokens = {
601 { NVMF_OPT_TRANSPORT, "transport=%s" },
602 { NVMF_OPT_TRADDR, "traddr=%s" },
603 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
604 { NVMF_OPT_NQN, "nqn=%s" },
605 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
606 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
607 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
608 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
609 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
610 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
611 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
612 { NVMF_OPT_HOST_ID, "hostid=%s" },
613 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
614 { NVMF_OPT_ERR, NULL }
617 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
620 substring_t args[MAX_OPT_ARGS];
621 char *options, *o, *p;
624 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
628 opts->queue_size = NVMF_DEF_QUEUE_SIZE;
629 opts->nr_io_queues = num_online_cpus();
630 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
631 opts->kato = NVME_DEFAULT_KATO;
632 opts->duplicate_connect = false;
634 options = o = kstrdup(buf, GFP_KERNEL);
640 while ((p = strsep(&o, ",\n")) != NULL) {
644 token = match_token(p, opt_tokens, args);
647 case NVMF_OPT_TRANSPORT:
648 p = match_strdup(args);
653 kfree(opts->transport);
657 p = match_strdup(args);
662 kfree(opts->subsysnqn);
664 nqnlen = strlen(opts->subsysnqn);
665 if (nqnlen >= NVMF_NQN_SIZE) {
666 pr_err("%s needs to be < %d bytes\n",
667 opts->subsysnqn, NVMF_NQN_SIZE);
671 opts->discovery_nqn =
672 !(strcmp(opts->subsysnqn,
673 NVME_DISC_SUBSYS_NAME));
675 case NVMF_OPT_TRADDR:
676 p = match_strdup(args);
684 case NVMF_OPT_TRSVCID:
685 p = match_strdup(args);
690 kfree(opts->trsvcid);
693 case NVMF_OPT_QUEUE_SIZE:
694 if (match_int(args, &token)) {
698 if (token < NVMF_MIN_QUEUE_SIZE ||
699 token > NVMF_MAX_QUEUE_SIZE) {
700 pr_err("Invalid queue_size %d\n", token);
704 opts->queue_size = token;
706 case NVMF_OPT_NR_IO_QUEUES:
707 if (match_int(args, &token)) {
712 pr_err("Invalid number of IOQs %d\n", token);
716 if (opts->discovery_nqn) {
717 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
721 opts->nr_io_queues = min_t(unsigned int,
722 num_online_cpus(), token);
725 if (match_int(args, &token)) {
731 pr_err("Invalid keep_alive_tmo %d\n", token);
734 } else if (token == 0 && !opts->discovery_nqn) {
735 /* Allowed for debug */
736 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
740 if (opts->discovery_nqn && opts->kato) {
741 pr_err("Discovery controllers cannot accept KATO != 0\n");
747 case NVMF_OPT_CTRL_LOSS_TMO:
748 if (match_int(args, &token)) {
754 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
755 ctrl_loss_tmo = token;
757 case NVMF_OPT_HOSTNQN:
759 pr_err("hostnqn already user-assigned: %s\n",
764 p = match_strdup(args);
770 if (nqnlen >= NVMF_NQN_SIZE) {
771 pr_err("%s needs to be < %d bytes\n",
777 nvmf_host_put(opts->host);
778 opts->host = nvmf_host_add(p);
785 case NVMF_OPT_RECONNECT_DELAY:
786 if (match_int(args, &token)) {
791 pr_err("Invalid reconnect_delay %d\n", token);
795 opts->reconnect_delay = token;
797 case NVMF_OPT_HOST_TRADDR:
798 p = match_strdup(args);
803 kfree(opts->host_traddr);
804 opts->host_traddr = p;
806 case NVMF_OPT_HOST_ID:
807 p = match_strdup(args);
812 ret = uuid_parse(p, &hostid);
814 pr_err("Invalid hostid %s\n", p);
821 case NVMF_OPT_DUP_CONNECT:
822 opts->duplicate_connect = true;
825 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
832 if (opts->discovery_nqn) {
834 opts->nr_io_queues = 0;
835 opts->duplicate_connect = true;
837 if (ctrl_loss_tmo < 0)
838 opts->max_reconnects = -1;
840 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
841 opts->reconnect_delay);
844 kref_get(&nvmf_default_host->ref);
845 opts->host = nvmf_default_host;
848 uuid_copy(&opts->host->id, &hostid);
855 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
856 unsigned int required_opts)
858 if ((opts->mask & required_opts) != required_opts) {
861 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
862 if ((opt_tokens[i].token & required_opts) &&
863 !(opt_tokens[i].token & opts->mask)) {
864 pr_warn("missing parameter '%s'\n",
865 opt_tokens[i].pattern);
875 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
876 unsigned int allowed_opts)
878 if (opts->mask & ~allowed_opts) {
881 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
882 if ((opt_tokens[i].token & opts->mask) &&
883 (opt_tokens[i].token & ~allowed_opts)) {
884 pr_warn("invalid parameter '%s'\n",
885 opt_tokens[i].pattern);
895 void nvmf_free_options(struct nvmf_ctrl_options *opts)
897 nvmf_host_put(opts->host);
898 kfree(opts->transport);
900 kfree(opts->trsvcid);
901 kfree(opts->subsysnqn);
902 kfree(opts->host_traddr);
905 EXPORT_SYMBOL_GPL(nvmf_free_options);
907 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
908 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
909 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
910 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT)
912 static struct nvme_ctrl *
913 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
915 struct nvmf_ctrl_options *opts;
916 struct nvmf_transport_ops *ops;
917 struct nvme_ctrl *ctrl;
920 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
922 return ERR_PTR(-ENOMEM);
924 ret = nvmf_parse_options(opts, buf);
929 request_module("nvme-%s", opts->transport);
932 * Check the generic options first as we need a valid transport for
933 * the lookup below. Then clear the generic flags so that transport
934 * drivers don't have to care about them.
936 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
939 opts->mask &= ~NVMF_REQUIRED_OPTS;
941 down_read(&nvmf_transports_rwsem);
942 ops = nvmf_lookup_transport(opts);
944 pr_info("no handler found for transport %s.\n",
950 if (!try_module_get(ops->module)) {
954 up_read(&nvmf_transports_rwsem);
956 ret = nvmf_check_required_opts(opts, ops->required_opts);
959 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
960 ops->allowed_opts | ops->required_opts);
964 ctrl = ops->create_ctrl(dev, opts);
970 module_put(ops->module);
974 module_put(ops->module);
977 up_read(&nvmf_transports_rwsem);
979 nvmf_free_options(opts);
983 static struct class *nvmf_class;
984 static struct device *nvmf_device;
985 static DEFINE_MUTEX(nvmf_dev_mutex);
987 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
988 size_t count, loff_t *pos)
990 struct seq_file *seq_file = file->private_data;
991 struct nvme_ctrl *ctrl;
995 if (count > PAGE_SIZE)
998 buf = memdup_user_nul(ubuf, count);
1000 return PTR_ERR(buf);
1002 mutex_lock(&nvmf_dev_mutex);
1003 if (seq_file->private) {
1008 ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
1010 ret = PTR_ERR(ctrl);
1014 seq_file->private = ctrl;
1017 mutex_unlock(&nvmf_dev_mutex);
1019 return ret ? ret : count;
1022 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1024 struct nvme_ctrl *ctrl;
1027 mutex_lock(&nvmf_dev_mutex);
1028 ctrl = seq_file->private;
1034 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1035 ctrl->instance, ctrl->cntlid);
1038 mutex_unlock(&nvmf_dev_mutex);
1042 static int nvmf_dev_open(struct inode *inode, struct file *file)
1045 * The miscdevice code initializes file->private_data, but doesn't
1046 * make use of it later.
1048 file->private_data = NULL;
1049 return single_open(file, nvmf_dev_show, NULL);
1052 static int nvmf_dev_release(struct inode *inode, struct file *file)
1054 struct seq_file *seq_file = file->private_data;
1055 struct nvme_ctrl *ctrl = seq_file->private;
1058 nvme_put_ctrl(ctrl);
1059 return single_release(inode, file);
1062 static const struct file_operations nvmf_dev_fops = {
1063 .owner = THIS_MODULE,
1064 .write = nvmf_dev_write,
1066 .open = nvmf_dev_open,
1067 .release = nvmf_dev_release,
1070 static struct miscdevice nvmf_misc = {
1071 .minor = MISC_DYNAMIC_MINOR,
1072 .name = "nvme-fabrics",
1073 .fops = &nvmf_dev_fops,
1076 static int __init nvmf_init(void)
1080 nvmf_default_host = nvmf_host_default();
1081 if (!nvmf_default_host)
1084 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1085 if (IS_ERR(nvmf_class)) {
1086 pr_err("couldn't register class nvme-fabrics\n");
1087 ret = PTR_ERR(nvmf_class);
1092 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1093 if (IS_ERR(nvmf_device)) {
1094 pr_err("couldn't create nvme-fabris device!\n");
1095 ret = PTR_ERR(nvmf_device);
1096 goto out_destroy_class;
1099 ret = misc_register(&nvmf_misc);
1101 pr_err("couldn't register misc device: %d\n", ret);
1102 goto out_destroy_device;
1108 device_destroy(nvmf_class, MKDEV(0, 0));
1110 class_destroy(nvmf_class);
1112 nvmf_host_put(nvmf_default_host);
1116 static void __exit nvmf_exit(void)
1118 misc_deregister(&nvmf_misc);
1119 device_destroy(nvmf_class, MKDEV(0, 0));
1120 class_destroy(nvmf_class);
1121 nvmf_host_put(nvmf_default_host);
1123 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1124 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1125 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1126 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1129 MODULE_LICENSE("GPL v2");
1131 module_init(nvmf_init);
1132 module_exit(nvmf_exit);