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>
15 #include <linux/nvme-keyring.h>
17 static LIST_HEAD(nvmf_transports);
18 static DECLARE_RWSEM(nvmf_transports_rwsem);
20 static LIST_HEAD(nvmf_hosts);
21 static DEFINE_MUTEX(nvmf_hosts_mutex);
23 static struct nvmf_host *nvmf_default_host;
25 static struct nvmf_host *nvmf_host_alloc(const char *hostnqn, uuid_t *id)
27 struct nvmf_host *host;
29 host = kmalloc(sizeof(*host), GFP_KERNEL);
33 kref_init(&host->ref);
34 uuid_copy(&host->id, id);
35 strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
40 static struct nvmf_host *nvmf_host_add(const char *hostnqn, uuid_t *id)
42 struct nvmf_host *host;
44 mutex_lock(&nvmf_hosts_mutex);
47 * We have defined a host as how it is perceived by the target.
48 * Therefore, we don't allow different Host NQNs with the same Host ID.
49 * Similarly, we do not allow the usage of the same Host NQN with
50 * different Host IDs. This'll maintain unambiguous host identification.
52 list_for_each_entry(host, &nvmf_hosts, list) {
53 bool same_hostnqn = !strcmp(host->nqn, hostnqn);
54 bool same_hostid = uuid_equal(&host->id, id);
56 if (same_hostnqn && same_hostid) {
61 pr_err("found same hostnqn %s but different hostid %pUb\n",
63 host = ERR_PTR(-EINVAL);
67 pr_err("found same hostid %pUb but different hostnqn %s\n",
69 host = ERR_PTR(-EINVAL);
74 host = nvmf_host_alloc(hostnqn, id);
76 host = ERR_PTR(-ENOMEM);
80 list_add_tail(&host->list, &nvmf_hosts);
82 mutex_unlock(&nvmf_hosts_mutex);
86 static struct nvmf_host *nvmf_host_default(void)
88 struct nvmf_host *host;
89 char nqn[NVMF_NQN_SIZE];
93 snprintf(nqn, NVMF_NQN_SIZE,
94 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &id);
96 host = nvmf_host_alloc(nqn, &id);
100 mutex_lock(&nvmf_hosts_mutex);
101 list_add_tail(&host->list, &nvmf_hosts);
102 mutex_unlock(&nvmf_hosts_mutex);
107 static void nvmf_host_destroy(struct kref *ref)
109 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
111 mutex_lock(&nvmf_hosts_mutex);
112 list_del(&host->list);
113 mutex_unlock(&nvmf_hosts_mutex);
118 static void nvmf_host_put(struct nvmf_host *host)
121 kref_put(&host->ref, nvmf_host_destroy);
125 * nvmf_get_address() - Get address/port
126 * @ctrl: Host NVMe controller instance which we got the address
127 * @buf: OUTPUT parameter that will contain the address/port
130 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
134 if (ctrl->opts->mask & NVMF_OPT_TRADDR)
135 len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
136 if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
137 len += scnprintf(buf + len, size - len, "%strsvcid=%s",
138 (len) ? "," : "", ctrl->opts->trsvcid);
139 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
140 len += scnprintf(buf + len, size - len, "%shost_traddr=%s",
141 (len) ? "," : "", ctrl->opts->host_traddr);
142 if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)
143 len += scnprintf(buf + len, size - len, "%shost_iface=%s",
144 (len) ? "," : "", ctrl->opts->host_iface);
145 len += scnprintf(buf + len, size - len, "\n");
149 EXPORT_SYMBOL_GPL(nvmf_get_address);
152 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function.
153 * @ctrl: Host NVMe controller instance maintaining the admin
154 * queue used to submit the property read command to
155 * the allocated NVMe controller resource on the target system.
156 * @off: Starting offset value of the targeted property
157 * register (see the fabrics section of the NVMe standard).
158 * @val: OUTPUT parameter that will contain the value of
159 * the property after a successful read.
161 * Used by the host system to retrieve a 32-bit capsule property value
162 * from an NVMe controller on the target system.
164 * ("Capsule property" is an "PCIe register concept" applied to the
165 * NVMe fabrics space.)
169 * > 0: NVMe error status code
170 * < 0: Linux errno error code
172 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
174 struct nvme_command cmd = { };
175 union nvme_result res;
178 cmd.prop_get.opcode = nvme_fabrics_command;
179 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
180 cmd.prop_get.offset = cpu_to_le32(off);
182 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
186 *val = le64_to_cpu(res.u64);
187 if (unlikely(ret != 0))
188 dev_err(ctrl->device,
189 "Property Get error: %d, offset %#x\n",
190 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
194 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
197 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function.
198 * @ctrl: Host NVMe controller instance maintaining the admin
199 * queue used to submit the property read command to
200 * the allocated controller resource on the target system.
201 * @off: Starting offset value of the targeted property
202 * register (see the fabrics section of the NVMe standard).
203 * @val: OUTPUT parameter that will contain the value of
204 * the property after a successful read.
206 * Used by the host system to retrieve a 64-bit capsule property value
207 * from an NVMe controller on the target system.
209 * ("Capsule property" is an "PCIe register concept" applied to the
210 * NVMe fabrics space.)
214 * > 0: NVMe error status code
215 * < 0: Linux errno error code
217 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
219 struct nvme_command cmd = { };
220 union nvme_result res;
223 cmd.prop_get.opcode = nvme_fabrics_command;
224 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
225 cmd.prop_get.attrib = 1;
226 cmd.prop_get.offset = cpu_to_le32(off);
228 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
232 *val = le64_to_cpu(res.u64);
233 if (unlikely(ret != 0))
234 dev_err(ctrl->device,
235 "Property Get error: %d, offset %#x\n",
236 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
239 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
242 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function.
243 * @ctrl: Host NVMe controller instance maintaining the admin
244 * queue used to submit the property read command to
245 * the allocated NVMe controller resource on the target system.
246 * @off: Starting offset value of the targeted property
247 * register (see the fabrics section of the NVMe standard).
248 * @val: Input parameter that contains the value to be
249 * written to the property.
251 * Used by the NVMe host system to write a 32-bit capsule property value
252 * to an NVMe controller on the target system.
254 * ("Capsule property" is an "PCIe register concept" applied to the
255 * NVMe fabrics space.)
258 * 0: successful write
259 * > 0: NVMe error status code
260 * < 0: Linux errno error code
262 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
264 struct nvme_command cmd = { };
267 cmd.prop_set.opcode = nvme_fabrics_command;
268 cmd.prop_set.fctype = nvme_fabrics_type_property_set;
269 cmd.prop_set.attrib = 0;
270 cmd.prop_set.offset = cpu_to_le32(off);
271 cmd.prop_set.value = cpu_to_le64(val);
273 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0,
276 dev_err(ctrl->device,
277 "Property Set error: %d, offset %#x\n",
278 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
281 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
284 * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for
286 * @ctrl: The specific /dev/nvmeX device that had the error.
287 * @errval: Error code to be decoded in a more human-friendly
289 * @offset: For use with the NVMe error code
290 * NVME_SC_CONNECT_INVALID_PARAM.
291 * @cmd: This is the SQE portion of a submission capsule.
292 * @data: This is the "Data" portion of a submission capsule.
294 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
295 int errval, int offset, struct nvme_command *cmd,
296 struct nvmf_connect_data *data)
298 int err_sctype = errval & ~NVME_SC_DNR;
301 dev_err(ctrl->device,
302 "Connect command failed, errno: %d\n", errval);
306 switch (err_sctype) {
307 case NVME_SC_CONNECT_INVALID_PARAM:
309 char *inv_data = "Connect Invalid Data Parameter";
311 switch (offset & 0xffff) {
312 case (offsetof(struct nvmf_connect_data, cntlid)):
313 dev_err(ctrl->device,
315 inv_data, data->cntlid);
317 case (offsetof(struct nvmf_connect_data, hostnqn)):
318 dev_err(ctrl->device,
319 "%s, hostnqn \"%s\"\n",
320 inv_data, data->hostnqn);
322 case (offsetof(struct nvmf_connect_data, subsysnqn)):
323 dev_err(ctrl->device,
324 "%s, subsysnqn \"%s\"\n",
325 inv_data, data->subsysnqn);
328 dev_err(ctrl->device,
329 "%s, starting byte offset: %d\n",
330 inv_data, offset & 0xffff);
334 char *inv_sqe = "Connect Invalid SQE Parameter";
337 case (offsetof(struct nvmf_connect_command, qid)):
338 dev_err(ctrl->device,
340 inv_sqe, cmd->connect.qid);
343 dev_err(ctrl->device,
344 "%s, starting byte offset: %d\n",
349 case NVME_SC_CONNECT_INVALID_HOST:
350 dev_err(ctrl->device,
351 "Connect for subsystem %s is not allowed, hostnqn: %s\n",
352 data->subsysnqn, data->hostnqn);
354 case NVME_SC_CONNECT_CTRL_BUSY:
355 dev_err(ctrl->device,
356 "Connect command failed: controller is busy or not available\n");
358 case NVME_SC_CONNECT_FORMAT:
359 dev_err(ctrl->device,
360 "Connect incompatible format: %d",
361 cmd->connect.recfmt);
363 case NVME_SC_HOST_PATH_ERROR:
364 dev_err(ctrl->device,
365 "Connect command failed: host path error\n");
367 case NVME_SC_AUTH_REQUIRED:
368 dev_err(ctrl->device,
369 "Connect command failed: authentication required\n");
372 dev_err(ctrl->device,
373 "Connect command failed, error wo/DNR bit: %d\n",
379 static struct nvmf_connect_data *nvmf_connect_data_prep(struct nvme_ctrl *ctrl,
382 struct nvmf_connect_data *data;
384 data = kzalloc(sizeof(*data), GFP_KERNEL);
388 uuid_copy(&data->hostid, &ctrl->opts->host->id);
389 data->cntlid = cpu_to_le16(cntlid);
390 strscpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
391 strscpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
396 static void nvmf_connect_cmd_prep(struct nvme_ctrl *ctrl, u16 qid,
397 struct nvme_command *cmd)
399 cmd->connect.opcode = nvme_fabrics_command;
400 cmd->connect.fctype = nvme_fabrics_type_connect;
401 cmd->connect.qid = cpu_to_le16(qid);
404 cmd->connect.sqsize = cpu_to_le16(ctrl->sqsize);
406 cmd->connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
409 * set keep-alive timeout in seconds granularity (ms * 1000)
411 cmd->connect.kato = cpu_to_le32(ctrl->kato * 1000);
414 if (ctrl->opts->disable_sqflow)
415 cmd->connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
419 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
421 * @ctrl: Host nvme controller instance used to request
422 * a new NVMe controller allocation on the target
423 * system and establish an NVMe Admin connection to
426 * This function enables an NVMe host device to request a new allocation of
427 * an NVMe controller resource on a target system as well establish a
428 * fabrics-protocol connection of the NVMe Admin queue between the
429 * host system device and the allocated NVMe controller on the
430 * target system via a NVMe Fabrics "Connect" command.
434 * > 0: NVMe error status code
435 * < 0: Linux errno error code
438 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
440 struct nvme_command cmd = { };
441 union nvme_result res;
442 struct nvmf_connect_data *data;
446 nvmf_connect_cmd_prep(ctrl, 0, &cmd);
448 data = nvmf_connect_data_prep(ctrl, 0xffff);
452 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
453 data, sizeof(*data), NVME_QID_ANY,
454 NVME_SUBMIT_AT_HEAD |
456 NVME_SUBMIT_RESERVED);
458 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
463 result = le32_to_cpu(res.u32);
464 ctrl->cntlid = result & 0xFFFF;
465 if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
466 /* Secure concatenation is not implemented */
467 if (result & NVME_CONNECT_AUTHREQ_ASCR) {
468 dev_warn(ctrl->device,
469 "qid 0: secure concatenation is not supported\n");
470 ret = NVME_SC_AUTH_REQUIRED;
473 /* Authentication required */
474 ret = nvme_auth_negotiate(ctrl, 0);
476 dev_warn(ctrl->device,
477 "qid 0: authentication setup failed\n");
478 ret = NVME_SC_AUTH_REQUIRED;
481 ret = nvme_auth_wait(ctrl, 0);
483 dev_warn(ctrl->device,
484 "qid 0: authentication failed\n");
486 dev_info(ctrl->device,
487 "qid 0: authenticated\n");
493 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
496 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
498 * @ctrl: Host nvme controller instance used to establish an
499 * NVMe I/O queue connection to the already allocated NVMe
500 * controller on the target system.
501 * @qid: NVMe I/O queue number for the new I/O connection between
502 * host and target (note qid == 0 is illegal as this is
503 * the Admin queue, per NVMe standard).
505 * This function issues a fabrics-protocol connection
506 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
507 * between the host system device and the allocated NVMe controller
508 * on the target system.
512 * > 0: NVMe error status code
513 * < 0: Linux errno error code
515 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
517 struct nvme_command cmd = { };
518 struct nvmf_connect_data *data;
519 union nvme_result res;
523 nvmf_connect_cmd_prep(ctrl, qid, &cmd);
525 data = nvmf_connect_data_prep(ctrl, ctrl->cntlid);
529 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
530 data, sizeof(*data), qid,
531 NVME_SUBMIT_AT_HEAD |
532 NVME_SUBMIT_RESERVED |
535 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
539 result = le32_to_cpu(res.u32);
540 if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
541 /* Secure concatenation is not implemented */
542 if (result & NVME_CONNECT_AUTHREQ_ASCR) {
543 dev_warn(ctrl->device,
544 "qid 0: secure concatenation is not supported\n");
545 ret = NVME_SC_AUTH_REQUIRED;
548 /* Authentication required */
549 ret = nvme_auth_negotiate(ctrl, qid);
551 dev_warn(ctrl->device,
552 "qid %d: authentication setup failed\n", qid);
553 ret = NVME_SC_AUTH_REQUIRED;
555 ret = nvme_auth_wait(ctrl, qid);
557 dev_warn(ctrl->device,
558 "qid %u: authentication failed\n", qid);
565 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
567 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
569 if (ctrl->opts->max_reconnects == -1 ||
570 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
575 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
578 * nvmf_register_transport() - NVMe Fabrics Library registration function.
579 * @ops: Transport ops instance to be registered to the
580 * common fabrics library.
582 * API function that registers the type of specific transport fabric
583 * being implemented to the common NVMe fabrics library. Part of
584 * the overall init sequence of starting up a fabrics driver.
586 int nvmf_register_transport(struct nvmf_transport_ops *ops)
588 if (!ops->create_ctrl)
591 down_write(&nvmf_transports_rwsem);
592 list_add_tail(&ops->entry, &nvmf_transports);
593 up_write(&nvmf_transports_rwsem);
597 EXPORT_SYMBOL_GPL(nvmf_register_transport);
600 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
601 * @ops: Transport ops instance to be unregistered from the
602 * common fabrics library.
604 * Fabrics API function that unregisters the type of specific transport
605 * fabric being implemented from the common NVMe fabrics library.
606 * Part of the overall exit sequence of unloading the implemented driver.
608 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
610 down_write(&nvmf_transports_rwsem);
611 list_del(&ops->entry);
612 up_write(&nvmf_transports_rwsem);
614 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
616 static struct nvmf_transport_ops *nvmf_lookup_transport(
617 struct nvmf_ctrl_options *opts)
619 struct nvmf_transport_ops *ops;
621 lockdep_assert_held(&nvmf_transports_rwsem);
623 list_for_each_entry(ops, &nvmf_transports, entry) {
624 if (strcmp(ops->name, opts->transport) == 0)
631 static struct key *nvmf_parse_key(int key_id)
635 if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
636 pr_err("TLS is not supported\n");
637 return ERR_PTR(-EINVAL);
640 key = key_lookup(key_id);
642 pr_err("key id %08x not found\n", key_id);
644 pr_debug("Using key id %08x\n", key_id);
648 static const match_table_t opt_tokens = {
649 { NVMF_OPT_TRANSPORT, "transport=%s" },
650 { NVMF_OPT_TRADDR, "traddr=%s" },
651 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
652 { NVMF_OPT_NQN, "nqn=%s" },
653 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
654 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
655 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
656 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
657 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
658 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
659 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
660 { NVMF_OPT_HOST_IFACE, "host_iface=%s" },
661 { NVMF_OPT_HOST_ID, "hostid=%s" },
662 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
663 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" },
664 { NVMF_OPT_HDR_DIGEST, "hdr_digest" },
665 { NVMF_OPT_DATA_DIGEST, "data_digest" },
666 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" },
667 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" },
668 { NVMF_OPT_TOS, "tos=%d" },
669 #ifdef CONFIG_NVME_TCP_TLS
670 { NVMF_OPT_KEYRING, "keyring=%d" },
671 { NVMF_OPT_TLS_KEY, "tls_key=%d" },
673 { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" },
674 { NVMF_OPT_DISCOVERY, "discovery" },
675 #ifdef CONFIG_NVME_HOST_AUTH
676 { NVMF_OPT_DHCHAP_SECRET, "dhchap_secret=%s" },
677 { NVMF_OPT_DHCHAP_CTRL_SECRET, "dhchap_ctrl_secret=%s" },
679 #ifdef CONFIG_NVME_TCP_TLS
680 { NVMF_OPT_TLS, "tls" },
682 { NVMF_OPT_ERR, NULL }
685 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
688 substring_t args[MAX_OPT_ARGS];
689 char *options, *o, *p;
692 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO, key_id;
694 char hostnqn[NVMF_NQN_SIZE];
698 opts->queue_size = NVMF_DEF_QUEUE_SIZE;
699 opts->nr_io_queues = num_online_cpus();
700 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
702 opts->duplicate_connect = false;
703 opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
704 opts->hdr_digest = false;
705 opts->data_digest = false;
706 opts->tos = -1; /* < 0 == use transport default */
708 opts->tls_key = NULL;
709 opts->keyring = NULL;
711 options = o = kstrdup(buf, GFP_KERNEL);
715 /* use default host if not given by user space */
716 uuid_copy(&hostid, &nvmf_default_host->id);
717 strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
719 while ((p = strsep(&o, ",\n")) != NULL) {
723 token = match_token(p, opt_tokens, args);
726 case NVMF_OPT_TRANSPORT:
727 p = match_strdup(args);
732 kfree(opts->transport);
736 p = match_strdup(args);
741 kfree(opts->subsysnqn);
743 nqnlen = strlen(opts->subsysnqn);
744 if (nqnlen >= NVMF_NQN_SIZE) {
745 pr_err("%s needs to be < %d bytes\n",
746 opts->subsysnqn, NVMF_NQN_SIZE);
750 opts->discovery_nqn =
751 !(strcmp(opts->subsysnqn,
752 NVME_DISC_SUBSYS_NAME));
754 case NVMF_OPT_TRADDR:
755 p = match_strdup(args);
763 case NVMF_OPT_TRSVCID:
764 p = match_strdup(args);
769 kfree(opts->trsvcid);
772 case NVMF_OPT_QUEUE_SIZE:
773 if (match_int(args, &token)) {
777 if (token < NVMF_MIN_QUEUE_SIZE ||
778 token > NVMF_MAX_QUEUE_SIZE) {
779 pr_err("Invalid queue_size %d\n", token);
783 opts->queue_size = token;
785 case NVMF_OPT_NR_IO_QUEUES:
786 if (match_int(args, &token)) {
791 pr_err("Invalid number of IOQs %d\n", token);
795 if (opts->discovery_nqn) {
796 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
800 opts->nr_io_queues = min_t(unsigned int,
801 num_online_cpus(), token);
804 if (match_int(args, &token)) {
810 pr_err("Invalid keep_alive_tmo %d\n", token);
813 } else if (token == 0 && !opts->discovery_nqn) {
814 /* Allowed for debug */
815 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
819 case NVMF_OPT_CTRL_LOSS_TMO:
820 if (match_int(args, &token)) {
826 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
827 ctrl_loss_tmo = token;
829 case NVMF_OPT_FAIL_FAST_TMO:
830 if (match_int(args, &token)) {
836 pr_warn("I/O fail on reconnect controller after %d sec\n",
841 opts->fast_io_fail_tmo = token;
843 case NVMF_OPT_HOSTNQN:
845 pr_err("hostnqn already user-assigned: %s\n",
850 p = match_strdup(args);
856 if (nqnlen >= NVMF_NQN_SIZE) {
857 pr_err("%s needs to be < %d bytes\n",
863 strscpy(hostnqn, p, NVMF_NQN_SIZE);
866 case NVMF_OPT_RECONNECT_DELAY:
867 if (match_int(args, &token)) {
872 pr_err("Invalid reconnect_delay %d\n", token);
876 opts->reconnect_delay = token;
878 case NVMF_OPT_HOST_TRADDR:
879 p = match_strdup(args);
884 kfree(opts->host_traddr);
885 opts->host_traddr = p;
887 case NVMF_OPT_HOST_IFACE:
888 p = match_strdup(args);
893 kfree(opts->host_iface);
894 opts->host_iface = p;
896 case NVMF_OPT_HOST_ID:
897 p = match_strdup(args);
902 ret = uuid_parse(p, &hostid);
904 pr_err("Invalid hostid %s\n", p);
911 case NVMF_OPT_DUP_CONNECT:
912 opts->duplicate_connect = true;
914 case NVMF_OPT_DISABLE_SQFLOW:
915 opts->disable_sqflow = true;
917 case NVMF_OPT_HDR_DIGEST:
918 opts->hdr_digest = true;
920 case NVMF_OPT_DATA_DIGEST:
921 opts->data_digest = true;
923 case NVMF_OPT_NR_WRITE_QUEUES:
924 if (match_int(args, &token)) {
929 pr_err("Invalid nr_write_queues %d\n", token);
933 opts->nr_write_queues = token;
935 case NVMF_OPT_NR_POLL_QUEUES:
936 if (match_int(args, &token)) {
941 pr_err("Invalid nr_poll_queues %d\n", token);
945 opts->nr_poll_queues = token;
948 if (match_int(args, &token)) {
953 pr_err("Invalid type of service %d\n", token);
958 pr_warn("Clamping type of service to 255\n");
963 case NVMF_OPT_KEYRING:
964 if (match_int(args, &key_id) || key_id <= 0) {
968 key = nvmf_parse_key(key_id);
973 key_put(opts->keyring);
976 case NVMF_OPT_TLS_KEY:
977 if (match_int(args, &key_id) || key_id <= 0) {
981 key = nvmf_parse_key(key_id);
986 key_put(opts->tls_key);
989 case NVMF_OPT_DISCOVERY:
990 opts->discovery_nqn = true;
992 case NVMF_OPT_DHCHAP_SECRET:
993 p = match_strdup(args);
998 if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
999 pr_err("Invalid DH-CHAP secret %s\n", p);
1003 kfree(opts->dhchap_secret);
1004 opts->dhchap_secret = p;
1006 case NVMF_OPT_DHCHAP_CTRL_SECRET:
1007 p = match_strdup(args);
1012 if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1013 pr_err("Invalid DH-CHAP secret %s\n", p);
1017 kfree(opts->dhchap_ctrl_secret);
1018 opts->dhchap_ctrl_secret = p;
1021 if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1022 pr_err("TLS is not supported\n");
1029 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
1036 if (opts->discovery_nqn) {
1037 opts->nr_io_queues = 0;
1038 opts->nr_write_queues = 0;
1039 opts->nr_poll_queues = 0;
1040 opts->duplicate_connect = true;
1043 opts->kato = NVME_DEFAULT_KATO;
1045 if (ctrl_loss_tmo < 0) {
1046 opts->max_reconnects = -1;
1048 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
1049 opts->reconnect_delay);
1050 if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
1051 pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
1052 opts->fast_io_fail_tmo, ctrl_loss_tmo);
1055 opts->host = nvmf_host_add(hostnqn, &hostid);
1056 if (IS_ERR(opts->host)) {
1057 ret = PTR_ERR(opts->host);
1067 void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
1068 u32 io_queues[HCTX_MAX_TYPES])
1070 if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
1072 * separate read/write queues
1073 * hand out dedicated default queues only after we have
1074 * sufficient read queues.
1076 io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
1077 nr_io_queues -= io_queues[HCTX_TYPE_READ];
1078 io_queues[HCTX_TYPE_DEFAULT] =
1079 min(opts->nr_write_queues, nr_io_queues);
1080 nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1083 * shared read/write queues
1084 * either no write queues were requested, or we don't have
1085 * sufficient queue count to have dedicated default queues.
1087 io_queues[HCTX_TYPE_DEFAULT] =
1088 min(opts->nr_io_queues, nr_io_queues);
1089 nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1092 if (opts->nr_poll_queues && nr_io_queues) {
1093 /* map dedicated poll queues only if we have queues left */
1094 io_queues[HCTX_TYPE_POLL] =
1095 min(opts->nr_poll_queues, nr_io_queues);
1098 EXPORT_SYMBOL_GPL(nvmf_set_io_queues);
1100 void nvmf_map_queues(struct blk_mq_tag_set *set, struct nvme_ctrl *ctrl,
1101 u32 io_queues[HCTX_MAX_TYPES])
1103 struct nvmf_ctrl_options *opts = ctrl->opts;
1105 if (opts->nr_write_queues && io_queues[HCTX_TYPE_READ]) {
1106 /* separate read/write queues */
1107 set->map[HCTX_TYPE_DEFAULT].nr_queues =
1108 io_queues[HCTX_TYPE_DEFAULT];
1109 set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1110 set->map[HCTX_TYPE_READ].nr_queues =
1111 io_queues[HCTX_TYPE_READ];
1112 set->map[HCTX_TYPE_READ].queue_offset =
1113 io_queues[HCTX_TYPE_DEFAULT];
1115 /* shared read/write queues */
1116 set->map[HCTX_TYPE_DEFAULT].nr_queues =
1117 io_queues[HCTX_TYPE_DEFAULT];
1118 set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1119 set->map[HCTX_TYPE_READ].nr_queues =
1120 io_queues[HCTX_TYPE_DEFAULT];
1121 set->map[HCTX_TYPE_READ].queue_offset = 0;
1124 blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1125 blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
1126 if (opts->nr_poll_queues && io_queues[HCTX_TYPE_POLL]) {
1127 /* map dedicated poll queues only if we have queues left */
1128 set->map[HCTX_TYPE_POLL].nr_queues = io_queues[HCTX_TYPE_POLL];
1129 set->map[HCTX_TYPE_POLL].queue_offset =
1130 io_queues[HCTX_TYPE_DEFAULT] +
1131 io_queues[HCTX_TYPE_READ];
1132 blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
1135 dev_info(ctrl->device,
1136 "mapped %d/%d/%d default/read/poll queues.\n",
1137 io_queues[HCTX_TYPE_DEFAULT],
1138 io_queues[HCTX_TYPE_READ],
1139 io_queues[HCTX_TYPE_POLL]);
1141 EXPORT_SYMBOL_GPL(nvmf_map_queues);
1143 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
1144 unsigned int required_opts)
1146 if ((opts->mask & required_opts) != required_opts) {
1149 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1150 if ((opt_tokens[i].token & required_opts) &&
1151 !(opt_tokens[i].token & opts->mask)) {
1152 pr_warn("missing parameter '%s'\n",
1153 opt_tokens[i].pattern);
1163 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
1164 struct nvmf_ctrl_options *opts)
1166 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
1167 strcmp(opts->traddr, ctrl->opts->traddr) ||
1168 strcmp(opts->trsvcid, ctrl->opts->trsvcid))
1172 * Checking the local address or host interfaces is rough.
1174 * In most cases, none is specified and the host port or
1175 * host interface is selected by the stack.
1177 * Assume no match if:
1178 * - local address or host interface is specified and address
1179 * or host interface is not the same
1180 * - local address or host interface is not specified but
1181 * remote is, or vice versa (admin using specific
1182 * host_traddr/host_iface when it matters).
1184 if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
1185 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1186 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
1188 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
1189 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1193 if ((opts->mask & NVMF_OPT_HOST_IFACE) &&
1194 (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1195 if (strcmp(opts->host_iface, ctrl->opts->host_iface))
1197 } else if ((opts->mask & NVMF_OPT_HOST_IFACE) ||
1198 (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1204 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
1206 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
1207 unsigned int allowed_opts)
1209 if (opts->mask & ~allowed_opts) {
1212 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1213 if ((opt_tokens[i].token & opts->mask) &&
1214 (opt_tokens[i].token & ~allowed_opts)) {
1215 pr_warn("invalid parameter '%s'\n",
1216 opt_tokens[i].pattern);
1226 void nvmf_free_options(struct nvmf_ctrl_options *opts)
1228 nvmf_host_put(opts->host);
1229 key_put(opts->keyring);
1230 key_put(opts->tls_key);
1231 kfree(opts->transport);
1232 kfree(opts->traddr);
1233 kfree(opts->trsvcid);
1234 kfree(opts->subsysnqn);
1235 kfree(opts->host_traddr);
1236 kfree(opts->host_iface);
1237 kfree(opts->dhchap_secret);
1238 kfree(opts->dhchap_ctrl_secret);
1241 EXPORT_SYMBOL_GPL(nvmf_free_options);
1243 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1244 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1245 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1246 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1247 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\
1248 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\
1249 NVMF_OPT_DHCHAP_CTRL_SECRET)
1251 static struct nvme_ctrl *
1252 nvmf_create_ctrl(struct device *dev, const char *buf)
1254 struct nvmf_ctrl_options *opts;
1255 struct nvmf_transport_ops *ops;
1256 struct nvme_ctrl *ctrl;
1259 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1261 return ERR_PTR(-ENOMEM);
1263 ret = nvmf_parse_options(opts, buf);
1268 request_module("nvme-%s", opts->transport);
1271 * Check the generic options first as we need a valid transport for
1272 * the lookup below. Then clear the generic flags so that transport
1273 * drivers don't have to care about them.
1275 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1278 opts->mask &= ~NVMF_REQUIRED_OPTS;
1280 down_read(&nvmf_transports_rwsem);
1281 ops = nvmf_lookup_transport(opts);
1283 pr_info("no handler found for transport %s.\n",
1289 if (!try_module_get(ops->module)) {
1293 up_read(&nvmf_transports_rwsem);
1295 ret = nvmf_check_required_opts(opts, ops->required_opts);
1297 goto out_module_put;
1298 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1299 ops->allowed_opts | ops->required_opts);
1301 goto out_module_put;
1303 ctrl = ops->create_ctrl(dev, opts);
1305 ret = PTR_ERR(ctrl);
1306 goto out_module_put;
1309 module_put(ops->module);
1313 module_put(ops->module);
1316 up_read(&nvmf_transports_rwsem);
1318 nvmf_free_options(opts);
1319 return ERR_PTR(ret);
1322 static const struct class nvmf_class = {
1323 .name = "nvme-fabrics",
1326 static struct device *nvmf_device;
1327 static DEFINE_MUTEX(nvmf_dev_mutex);
1329 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1330 size_t count, loff_t *pos)
1332 struct seq_file *seq_file = file->private_data;
1333 struct nvme_ctrl *ctrl;
1337 if (count > PAGE_SIZE)
1340 buf = memdup_user_nul(ubuf, count);
1342 return PTR_ERR(buf);
1344 mutex_lock(&nvmf_dev_mutex);
1345 if (seq_file->private) {
1350 ctrl = nvmf_create_ctrl(nvmf_device, buf);
1352 ret = PTR_ERR(ctrl);
1356 seq_file->private = ctrl;
1359 mutex_unlock(&nvmf_dev_mutex);
1361 return ret ? ret : count;
1364 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file)
1366 const struct match_token *tok;
1370 * Add dummy entries for instance and cntlid to
1371 * signal an invalid/non-existing controller
1373 seq_puts(seq_file, "instance=-1,cntlid=-1");
1374 for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) {
1375 tok = &opt_tokens[idx];
1376 if (tok->token == NVMF_OPT_ERR)
1378 seq_puts(seq_file, ",");
1379 seq_puts(seq_file, tok->pattern);
1381 seq_puts(seq_file, "\n");
1384 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1386 struct nvme_ctrl *ctrl;
1388 mutex_lock(&nvmf_dev_mutex);
1389 ctrl = seq_file->private;
1391 __nvmf_concat_opt_tokens(seq_file);
1395 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1396 ctrl->instance, ctrl->cntlid);
1399 mutex_unlock(&nvmf_dev_mutex);
1403 static int nvmf_dev_open(struct inode *inode, struct file *file)
1406 * The miscdevice code initializes file->private_data, but doesn't
1407 * make use of it later.
1409 file->private_data = NULL;
1410 return single_open(file, nvmf_dev_show, NULL);
1413 static int nvmf_dev_release(struct inode *inode, struct file *file)
1415 struct seq_file *seq_file = file->private_data;
1416 struct nvme_ctrl *ctrl = seq_file->private;
1419 nvme_put_ctrl(ctrl);
1420 return single_release(inode, file);
1423 static const struct file_operations nvmf_dev_fops = {
1424 .owner = THIS_MODULE,
1425 .write = nvmf_dev_write,
1427 .open = nvmf_dev_open,
1428 .release = nvmf_dev_release,
1431 static struct miscdevice nvmf_misc = {
1432 .minor = MISC_DYNAMIC_MINOR,
1433 .name = "nvme-fabrics",
1434 .fops = &nvmf_dev_fops,
1437 static int __init nvmf_init(void)
1441 nvmf_default_host = nvmf_host_default();
1442 if (!nvmf_default_host)
1445 ret = class_register(&nvmf_class);
1447 pr_err("couldn't register class nvme-fabrics\n");
1452 device_create(&nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1453 if (IS_ERR(nvmf_device)) {
1454 pr_err("couldn't create nvme-fabrics device!\n");
1455 ret = PTR_ERR(nvmf_device);
1456 goto out_destroy_class;
1459 ret = misc_register(&nvmf_misc);
1461 pr_err("couldn't register misc device: %d\n", ret);
1462 goto out_destroy_device;
1468 device_destroy(&nvmf_class, MKDEV(0, 0));
1470 class_unregister(&nvmf_class);
1472 nvmf_host_put(nvmf_default_host);
1476 static void __exit nvmf_exit(void)
1478 misc_deregister(&nvmf_misc);
1479 device_destroy(&nvmf_class, MKDEV(0, 0));
1480 class_unregister(&nvmf_class);
1481 nvmf_host_put(nvmf_default_host);
1483 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1484 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1485 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1486 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1487 BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64);
1488 BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64);
1489 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1490 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8);
1491 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16);
1492 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16);
1493 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16);
1494 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16);
1497 MODULE_LICENSE("GPL v2");
1498 MODULE_DESCRIPTION("NVMe host fabrics library");
1500 module_init(nvmf_init);
1501 module_exit(nvmf_exit);