1 // SPDX-License-Identifier: GPL-2.0
3 * NVMe over Fabrics common host code.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 #include <linux/init.h>
8 #include <linux/miscdevice.h>
9 #include <linux/module.h>
10 #include <linux/mutex.h>
11 #include <linux/parser.h>
12 #include <linux/seq_file.h>
16 static LIST_HEAD(nvmf_transports);
17 static DECLARE_RWSEM(nvmf_transports_rwsem);
19 static LIST_HEAD(nvmf_hosts);
20 static DEFINE_MUTEX(nvmf_hosts_mutex);
22 static struct nvmf_host *nvmf_default_host;
24 static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
26 struct nvmf_host *host;
28 list_for_each_entry(host, &nvmf_hosts, list) {
29 if (!strcmp(host->nqn, hostnqn))
36 static struct nvmf_host *nvmf_host_add(const char *hostnqn)
38 struct nvmf_host *host;
40 mutex_lock(&nvmf_hosts_mutex);
41 host = __nvmf_host_find(hostnqn);
47 host = kmalloc(sizeof(*host), GFP_KERNEL);
51 kref_init(&host->ref);
52 strlcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
54 list_add_tail(&host->list, &nvmf_hosts);
56 mutex_unlock(&nvmf_hosts_mutex);
60 static struct nvmf_host *nvmf_host_default(void)
62 struct nvmf_host *host;
64 host = kmalloc(sizeof(*host), GFP_KERNEL);
68 kref_init(&host->ref);
70 snprintf(host->nqn, NVMF_NQN_SIZE,
71 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id);
73 mutex_lock(&nvmf_hosts_mutex);
74 list_add_tail(&host->list, &nvmf_hosts);
75 mutex_unlock(&nvmf_hosts_mutex);
80 static void nvmf_host_destroy(struct kref *ref)
82 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
84 mutex_lock(&nvmf_hosts_mutex);
85 list_del(&host->list);
86 mutex_unlock(&nvmf_hosts_mutex);
91 static void nvmf_host_put(struct nvmf_host *host)
94 kref_put(&host->ref, nvmf_host_destroy);
98 * nvmf_get_address() - Get address/port
99 * @ctrl: Host NVMe controller instance which we got the address
100 * @buf: OUTPUT parameter that will contain the address/port
103 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
107 if (ctrl->opts->mask & NVMF_OPT_TRADDR)
108 len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
109 if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
110 len += scnprintf(buf + len, size - len, "%strsvcid=%s",
111 (len) ? "," : "", ctrl->opts->trsvcid);
112 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
113 len += scnprintf(buf + len, size - len, "%shost_traddr=%s",
114 (len) ? "," : "", ctrl->opts->host_traddr);
115 if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)
116 len += scnprintf(buf + len, size - len, "%shost_iface=%s",
117 (len) ? "," : "", ctrl->opts->host_iface);
118 len += scnprintf(buf + len, size - len, "\n");
122 EXPORT_SYMBOL_GPL(nvmf_get_address);
125 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function.
126 * @ctrl: Host NVMe controller instance maintaining the admin
127 * queue used to submit the property read command to
128 * the allocated NVMe controller resource on the target system.
129 * @off: Starting offset value of the targeted property
130 * register (see the fabrics section of the NVMe standard).
131 * @val: OUTPUT parameter that will contain the value of
132 * the property after a successful read.
134 * Used by the host system to retrieve a 32-bit capsule property value
135 * from an NVMe controller on the target system.
137 * ("Capsule property" is an "PCIe register concept" applied to the
138 * NVMe fabrics space.)
142 * > 0: NVMe error status code
143 * < 0: Linux errno error code
145 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
147 struct nvme_command cmd;
148 union nvme_result res;
151 memset(&cmd, 0, sizeof(cmd));
152 cmd.prop_get.opcode = nvme_fabrics_command;
153 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
154 cmd.prop_get.offset = cpu_to_le32(off);
156 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
157 NVME_QID_ANY, 0, 0, false);
160 *val = le64_to_cpu(res.u64);
161 if (unlikely(ret != 0))
162 dev_err(ctrl->device,
163 "Property Get error: %d, offset %#x\n",
164 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
168 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
171 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function.
172 * @ctrl: Host NVMe controller instance maintaining the admin
173 * queue used to submit the property read command to
174 * the allocated controller resource on the target system.
175 * @off: Starting offset value of the targeted property
176 * register (see the fabrics section of the NVMe standard).
177 * @val: OUTPUT parameter that will contain the value of
178 * the property after a successful read.
180 * Used by the host system to retrieve a 64-bit capsule property value
181 * from an NVMe controller on the target system.
183 * ("Capsule property" is an "PCIe register concept" applied to the
184 * NVMe fabrics space.)
188 * > 0: NVMe error status code
189 * < 0: Linux errno error code
191 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
193 struct nvme_command cmd = { };
194 union nvme_result res;
197 cmd.prop_get.opcode = nvme_fabrics_command;
198 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
199 cmd.prop_get.attrib = 1;
200 cmd.prop_get.offset = cpu_to_le32(off);
202 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
203 NVME_QID_ANY, 0, 0, false);
206 *val = le64_to_cpu(res.u64);
207 if (unlikely(ret != 0))
208 dev_err(ctrl->device,
209 "Property Get error: %d, offset %#x\n",
210 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
213 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
216 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function.
217 * @ctrl: Host NVMe controller instance maintaining the admin
218 * queue used to submit the property read command to
219 * the allocated NVMe controller resource on the target system.
220 * @off: Starting offset value of the targeted property
221 * register (see the fabrics section of the NVMe standard).
222 * @val: Input parameter that contains the value to be
223 * written to the property.
225 * Used by the NVMe host system to write a 32-bit capsule property value
226 * to an NVMe controller on the target system.
228 * ("Capsule property" is an "PCIe register concept" applied to the
229 * NVMe fabrics space.)
232 * 0: successful write
233 * > 0: NVMe error status code
234 * < 0: Linux errno error code
236 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
238 struct nvme_command cmd = { };
241 cmd.prop_set.opcode = nvme_fabrics_command;
242 cmd.prop_set.fctype = nvme_fabrics_type_property_set;
243 cmd.prop_set.attrib = 0;
244 cmd.prop_set.offset = cpu_to_le32(off);
245 cmd.prop_set.value = cpu_to_le64(val);
247 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 0,
248 NVME_QID_ANY, 0, 0, false);
250 dev_err(ctrl->device,
251 "Property Set error: %d, offset %#x\n",
252 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
255 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
258 * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for
260 * @ctrl: The specific /dev/nvmeX device that had the error.
261 * @errval: Error code to be decoded in a more human-friendly
263 * @offset: For use with the NVMe error code
264 * NVME_SC_CONNECT_INVALID_PARAM.
265 * @cmd: This is the SQE portion of a submission capsule.
266 * @data: This is the "Data" portion of a submission capsule.
268 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
269 int errval, int offset, struct nvme_command *cmd,
270 struct nvmf_connect_data *data)
272 int err_sctype = errval & ~NVME_SC_DNR;
274 switch (err_sctype) {
275 case (NVME_SC_CONNECT_INVALID_PARAM):
277 char *inv_data = "Connect Invalid Data Parameter";
279 switch (offset & 0xffff) {
280 case (offsetof(struct nvmf_connect_data, cntlid)):
281 dev_err(ctrl->device,
283 inv_data, data->cntlid);
285 case (offsetof(struct nvmf_connect_data, hostnqn)):
286 dev_err(ctrl->device,
287 "%s, hostnqn \"%s\"\n",
288 inv_data, data->hostnqn);
290 case (offsetof(struct nvmf_connect_data, subsysnqn)):
291 dev_err(ctrl->device,
292 "%s, subsysnqn \"%s\"\n",
293 inv_data, data->subsysnqn);
296 dev_err(ctrl->device,
297 "%s, starting byte offset: %d\n",
298 inv_data, offset & 0xffff);
302 char *inv_sqe = "Connect Invalid SQE Parameter";
305 case (offsetof(struct nvmf_connect_command, qid)):
306 dev_err(ctrl->device,
308 inv_sqe, cmd->connect.qid);
311 dev_err(ctrl->device,
312 "%s, starting byte offset: %d\n",
317 case NVME_SC_CONNECT_INVALID_HOST:
318 dev_err(ctrl->device,
319 "Connect for subsystem %s is not allowed, hostnqn: %s\n",
320 data->subsysnqn, data->hostnqn);
322 case NVME_SC_CONNECT_CTRL_BUSY:
323 dev_err(ctrl->device,
324 "Connect command failed: controller is busy or not available\n");
326 case NVME_SC_CONNECT_FORMAT:
327 dev_err(ctrl->device,
328 "Connect incompatible format: %d",
329 cmd->connect.recfmt);
331 case NVME_SC_HOST_PATH_ERROR:
332 dev_err(ctrl->device,
333 "Connect command failed: host path error\n");
336 dev_err(ctrl->device,
337 "Connect command failed, error wo/DNR bit: %d\n",
344 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
346 * @ctrl: Host nvme controller instance used to request
347 * a new NVMe controller allocation on the target
348 * system and establish an NVMe Admin connection to
351 * This function enables an NVMe host device to request a new allocation of
352 * an NVMe controller resource on a target system as well establish a
353 * fabrics-protocol connection of the NVMe Admin queue between the
354 * host system device and the allocated NVMe controller on the
355 * target system via a NVMe Fabrics "Connect" command.
359 * > 0: NVMe error status code
360 * < 0: Linux errno error code
363 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
365 struct nvme_command cmd = { };
366 union nvme_result res;
367 struct nvmf_connect_data *data;
370 cmd.connect.opcode = nvme_fabrics_command;
371 cmd.connect.fctype = nvme_fabrics_type_connect;
373 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
376 * Set keep-alive timeout in seconds granularity (ms * 1000)
378 cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000);
380 if (ctrl->opts->disable_sqflow)
381 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
383 data = kzalloc(sizeof(*data), GFP_KERNEL);
387 uuid_copy(&data->hostid, &ctrl->opts->host->id);
388 data->cntlid = cpu_to_le16(0xffff);
389 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
390 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
392 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
393 data, sizeof(*data), 0, NVME_QID_ANY, 1,
394 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, false);
396 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
401 ctrl->cntlid = le16_to_cpu(res.u16);
407 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
410 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
412 * @ctrl: Host nvme controller instance used to establish an
413 * NVMe I/O queue connection to the already allocated NVMe
414 * controller on the target system.
415 * @qid: NVMe I/O queue number for the new I/O connection between
416 * host and target (note qid == 0 is illegal as this is
417 * the Admin queue, per NVMe standard).
418 * @poll: Whether or not to poll for the completion of the connect cmd.
420 * This function issues a fabrics-protocol connection
421 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
422 * between the host system device and the allocated NVMe controller
423 * on the target system.
427 * > 0: NVMe error status code
428 * < 0: Linux errno error code
430 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid, bool poll)
432 struct nvme_command cmd = { };
433 struct nvmf_connect_data *data;
434 union nvme_result res;
437 cmd.connect.opcode = nvme_fabrics_command;
438 cmd.connect.fctype = nvme_fabrics_type_connect;
439 cmd.connect.qid = cpu_to_le16(qid);
440 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
442 if (ctrl->opts->disable_sqflow)
443 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
445 data = kzalloc(sizeof(*data), GFP_KERNEL);
449 uuid_copy(&data->hostid, &ctrl->opts->host->id);
450 data->cntlid = cpu_to_le16(ctrl->cntlid);
451 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
452 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
454 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
455 data, sizeof(*data), 0, qid, 1,
456 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, poll);
458 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
464 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
466 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
468 if (ctrl->opts->max_reconnects == -1 ||
469 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
474 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
477 * nvmf_register_transport() - NVMe Fabrics Library registration function.
478 * @ops: Transport ops instance to be registered to the
479 * common fabrics library.
481 * API function that registers the type of specific transport fabric
482 * being implemented to the common NVMe fabrics library. Part of
483 * the overall init sequence of starting up a fabrics driver.
485 int nvmf_register_transport(struct nvmf_transport_ops *ops)
487 if (!ops->create_ctrl)
490 down_write(&nvmf_transports_rwsem);
491 list_add_tail(&ops->entry, &nvmf_transports);
492 up_write(&nvmf_transports_rwsem);
496 EXPORT_SYMBOL_GPL(nvmf_register_transport);
499 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
500 * @ops: Transport ops instance to be unregistered from the
501 * common fabrics library.
503 * Fabrics API function that unregisters the type of specific transport
504 * fabric being implemented from the common NVMe fabrics library.
505 * Part of the overall exit sequence of unloading the implemented driver.
507 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
509 down_write(&nvmf_transports_rwsem);
510 list_del(&ops->entry);
511 up_write(&nvmf_transports_rwsem);
513 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
515 static struct nvmf_transport_ops *nvmf_lookup_transport(
516 struct nvmf_ctrl_options *opts)
518 struct nvmf_transport_ops *ops;
520 lockdep_assert_held(&nvmf_transports_rwsem);
522 list_for_each_entry(ops, &nvmf_transports, entry) {
523 if (strcmp(ops->name, opts->transport) == 0)
530 static const match_table_t opt_tokens = {
531 { NVMF_OPT_TRANSPORT, "transport=%s" },
532 { NVMF_OPT_TRADDR, "traddr=%s" },
533 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
534 { NVMF_OPT_NQN, "nqn=%s" },
535 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
536 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
537 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
538 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
539 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
540 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
541 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
542 { NVMF_OPT_HOST_IFACE, "host_iface=%s" },
543 { NVMF_OPT_HOST_ID, "hostid=%s" },
544 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
545 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" },
546 { NVMF_OPT_HDR_DIGEST, "hdr_digest" },
547 { NVMF_OPT_DATA_DIGEST, "data_digest" },
548 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" },
549 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" },
550 { NVMF_OPT_TOS, "tos=%d" },
551 { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" },
552 { NVMF_OPT_ERR, NULL }
555 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
558 substring_t args[MAX_OPT_ARGS];
559 char *options, *o, *p;
562 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
566 opts->queue_size = NVMF_DEF_QUEUE_SIZE;
567 opts->nr_io_queues = num_online_cpus();
568 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
570 opts->duplicate_connect = false;
571 opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
572 opts->hdr_digest = false;
573 opts->data_digest = false;
574 opts->tos = -1; /* < 0 == use transport default */
576 options = o = kstrdup(buf, GFP_KERNEL);
582 while ((p = strsep(&o, ",\n")) != NULL) {
586 token = match_token(p, opt_tokens, args);
589 case NVMF_OPT_TRANSPORT:
590 p = match_strdup(args);
595 kfree(opts->transport);
599 p = match_strdup(args);
604 kfree(opts->subsysnqn);
606 nqnlen = strlen(opts->subsysnqn);
607 if (nqnlen >= NVMF_NQN_SIZE) {
608 pr_err("%s needs to be < %d bytes\n",
609 opts->subsysnqn, NVMF_NQN_SIZE);
613 opts->discovery_nqn =
614 !(strcmp(opts->subsysnqn,
615 NVME_DISC_SUBSYS_NAME));
617 case NVMF_OPT_TRADDR:
618 p = match_strdup(args);
626 case NVMF_OPT_TRSVCID:
627 p = match_strdup(args);
632 kfree(opts->trsvcid);
635 case NVMF_OPT_QUEUE_SIZE:
636 if (match_int(args, &token)) {
640 if (token < NVMF_MIN_QUEUE_SIZE ||
641 token > NVMF_MAX_QUEUE_SIZE) {
642 pr_err("Invalid queue_size %d\n", token);
646 opts->queue_size = token;
648 case NVMF_OPT_NR_IO_QUEUES:
649 if (match_int(args, &token)) {
654 pr_err("Invalid number of IOQs %d\n", token);
658 if (opts->discovery_nqn) {
659 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
663 opts->nr_io_queues = min_t(unsigned int,
664 num_online_cpus(), token);
667 if (match_int(args, &token)) {
673 pr_err("Invalid keep_alive_tmo %d\n", token);
676 } else if (token == 0 && !opts->discovery_nqn) {
677 /* Allowed for debug */
678 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
682 case NVMF_OPT_CTRL_LOSS_TMO:
683 if (match_int(args, &token)) {
689 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
690 ctrl_loss_tmo = token;
692 case NVMF_OPT_FAIL_FAST_TMO:
693 if (match_int(args, &token)) {
699 pr_warn("I/O fail on reconnect controller after %d sec\n",
701 opts->fast_io_fail_tmo = token;
703 case NVMF_OPT_HOSTNQN:
705 pr_err("hostnqn already user-assigned: %s\n",
710 p = match_strdup(args);
716 if (nqnlen >= NVMF_NQN_SIZE) {
717 pr_err("%s needs to be < %d bytes\n",
723 nvmf_host_put(opts->host);
724 opts->host = nvmf_host_add(p);
731 case NVMF_OPT_RECONNECT_DELAY:
732 if (match_int(args, &token)) {
737 pr_err("Invalid reconnect_delay %d\n", token);
741 opts->reconnect_delay = token;
743 case NVMF_OPT_HOST_TRADDR:
744 p = match_strdup(args);
749 kfree(opts->host_traddr);
750 opts->host_traddr = p;
752 case NVMF_OPT_HOST_IFACE:
753 p = match_strdup(args);
758 kfree(opts->host_iface);
759 opts->host_iface = p;
761 case NVMF_OPT_HOST_ID:
762 p = match_strdup(args);
767 ret = uuid_parse(p, &hostid);
769 pr_err("Invalid hostid %s\n", p);
776 case NVMF_OPT_DUP_CONNECT:
777 opts->duplicate_connect = true;
779 case NVMF_OPT_DISABLE_SQFLOW:
780 opts->disable_sqflow = true;
782 case NVMF_OPT_HDR_DIGEST:
783 opts->hdr_digest = true;
785 case NVMF_OPT_DATA_DIGEST:
786 opts->data_digest = true;
788 case NVMF_OPT_NR_WRITE_QUEUES:
789 if (match_int(args, &token)) {
794 pr_err("Invalid nr_write_queues %d\n", token);
798 opts->nr_write_queues = token;
800 case NVMF_OPT_NR_POLL_QUEUES:
801 if (match_int(args, &token)) {
806 pr_err("Invalid nr_poll_queues %d\n", token);
810 opts->nr_poll_queues = token;
813 if (match_int(args, &token)) {
818 pr_err("Invalid type of service %d\n", token);
823 pr_warn("Clamping type of service to 255\n");
829 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
836 if (opts->discovery_nqn) {
837 opts->nr_io_queues = 0;
838 opts->nr_write_queues = 0;
839 opts->nr_poll_queues = 0;
840 opts->duplicate_connect = true;
843 opts->kato = NVME_DEFAULT_KATO;
845 if (ctrl_loss_tmo < 0) {
846 opts->max_reconnects = -1;
848 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
849 opts->reconnect_delay);
850 if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
851 pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
852 opts->fast_io_fail_tmo, ctrl_loss_tmo);
856 kref_get(&nvmf_default_host->ref);
857 opts->host = nvmf_default_host;
860 uuid_copy(&opts->host->id, &hostid);
867 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
868 unsigned int required_opts)
870 if ((opts->mask & required_opts) != required_opts) {
873 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
874 if ((opt_tokens[i].token & required_opts) &&
875 !(opt_tokens[i].token & opts->mask)) {
876 pr_warn("missing parameter '%s'\n",
877 opt_tokens[i].pattern);
887 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
888 struct nvmf_ctrl_options *opts)
890 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
891 strcmp(opts->traddr, ctrl->opts->traddr) ||
892 strcmp(opts->trsvcid, ctrl->opts->trsvcid))
896 * Checking the local address is rough. In most cases, none is specified
897 * and the host port is selected by the stack.
899 * Assume no match if:
900 * - local address is specified and address is not the same
901 * - local address is not specified but remote is, or vice versa
902 * (admin using specific host_traddr when it matters).
904 if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
905 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
906 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
908 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
909 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
915 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
917 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
918 unsigned int allowed_opts)
920 if (opts->mask & ~allowed_opts) {
923 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
924 if ((opt_tokens[i].token & opts->mask) &&
925 (opt_tokens[i].token & ~allowed_opts)) {
926 pr_warn("invalid parameter '%s'\n",
927 opt_tokens[i].pattern);
937 void nvmf_free_options(struct nvmf_ctrl_options *opts)
939 nvmf_host_put(opts->host);
940 kfree(opts->transport);
942 kfree(opts->trsvcid);
943 kfree(opts->subsysnqn);
944 kfree(opts->host_traddr);
945 kfree(opts->host_iface);
948 EXPORT_SYMBOL_GPL(nvmf_free_options);
950 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
951 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
952 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
953 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
954 NVMF_OPT_DISABLE_SQFLOW |\
955 NVMF_OPT_FAIL_FAST_TMO)
957 static struct nvme_ctrl *
958 nvmf_create_ctrl(struct device *dev, const char *buf)
960 struct nvmf_ctrl_options *opts;
961 struct nvmf_transport_ops *ops;
962 struct nvme_ctrl *ctrl;
965 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
967 return ERR_PTR(-ENOMEM);
969 ret = nvmf_parse_options(opts, buf);
974 request_module("nvme-%s", opts->transport);
977 * Check the generic options first as we need a valid transport for
978 * the lookup below. Then clear the generic flags so that transport
979 * drivers don't have to care about them.
981 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
984 opts->mask &= ~NVMF_REQUIRED_OPTS;
986 down_read(&nvmf_transports_rwsem);
987 ops = nvmf_lookup_transport(opts);
989 pr_info("no handler found for transport %s.\n",
995 if (!try_module_get(ops->module)) {
999 up_read(&nvmf_transports_rwsem);
1001 ret = nvmf_check_required_opts(opts, ops->required_opts);
1003 goto out_module_put;
1004 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1005 ops->allowed_opts | ops->required_opts);
1007 goto out_module_put;
1009 ctrl = ops->create_ctrl(dev, opts);
1011 ret = PTR_ERR(ctrl);
1012 goto out_module_put;
1015 module_put(ops->module);
1019 module_put(ops->module);
1022 up_read(&nvmf_transports_rwsem);
1024 nvmf_free_options(opts);
1025 return ERR_PTR(ret);
1028 static struct class *nvmf_class;
1029 static struct device *nvmf_device;
1030 static DEFINE_MUTEX(nvmf_dev_mutex);
1032 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1033 size_t count, loff_t *pos)
1035 struct seq_file *seq_file = file->private_data;
1036 struct nvme_ctrl *ctrl;
1040 if (count > PAGE_SIZE)
1043 buf = memdup_user_nul(ubuf, count);
1045 return PTR_ERR(buf);
1047 mutex_lock(&nvmf_dev_mutex);
1048 if (seq_file->private) {
1053 ctrl = nvmf_create_ctrl(nvmf_device, buf);
1055 ret = PTR_ERR(ctrl);
1059 seq_file->private = ctrl;
1062 mutex_unlock(&nvmf_dev_mutex);
1064 return ret ? ret : count;
1067 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1069 struct nvme_ctrl *ctrl;
1072 mutex_lock(&nvmf_dev_mutex);
1073 ctrl = seq_file->private;
1079 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1080 ctrl->instance, ctrl->cntlid);
1083 mutex_unlock(&nvmf_dev_mutex);
1087 static int nvmf_dev_open(struct inode *inode, struct file *file)
1090 * The miscdevice code initializes file->private_data, but doesn't
1091 * make use of it later.
1093 file->private_data = NULL;
1094 return single_open(file, nvmf_dev_show, NULL);
1097 static int nvmf_dev_release(struct inode *inode, struct file *file)
1099 struct seq_file *seq_file = file->private_data;
1100 struct nvme_ctrl *ctrl = seq_file->private;
1103 nvme_put_ctrl(ctrl);
1104 return single_release(inode, file);
1107 static const struct file_operations nvmf_dev_fops = {
1108 .owner = THIS_MODULE,
1109 .write = nvmf_dev_write,
1111 .open = nvmf_dev_open,
1112 .release = nvmf_dev_release,
1115 static struct miscdevice nvmf_misc = {
1116 .minor = MISC_DYNAMIC_MINOR,
1117 .name = "nvme-fabrics",
1118 .fops = &nvmf_dev_fops,
1121 static int __init nvmf_init(void)
1125 nvmf_default_host = nvmf_host_default();
1126 if (!nvmf_default_host)
1129 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1130 if (IS_ERR(nvmf_class)) {
1131 pr_err("couldn't register class nvme-fabrics\n");
1132 ret = PTR_ERR(nvmf_class);
1137 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1138 if (IS_ERR(nvmf_device)) {
1139 pr_err("couldn't create nvme-fabris device!\n");
1140 ret = PTR_ERR(nvmf_device);
1141 goto out_destroy_class;
1144 ret = misc_register(&nvmf_misc);
1146 pr_err("couldn't register misc device: %d\n", ret);
1147 goto out_destroy_device;
1153 device_destroy(nvmf_class, MKDEV(0, 0));
1155 class_destroy(nvmf_class);
1157 nvmf_host_put(nvmf_default_host);
1161 static void __exit nvmf_exit(void)
1163 misc_deregister(&nvmf_misc);
1164 device_destroy(nvmf_class, MKDEV(0, 0));
1165 class_destroy(nvmf_class);
1166 nvmf_host_put(nvmf_default_host);
1168 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1169 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1170 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1171 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1172 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1175 MODULE_LICENSE("GPL v2");
1177 module_init(nvmf_init);
1178 module_exit(nvmf_exit);