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 += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
109 if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
110 len += snprintf(buf + len, size - len, "%strsvcid=%s",
111 (len) ? "," : "", ctrl->opts->trsvcid);
112 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
113 len += snprintf(buf + len, size - len, "%shost_traddr=%s",
114 (len) ? "," : "", ctrl->opts->host_traddr);
115 len += snprintf(buf + len, size - len, "\n");
119 EXPORT_SYMBOL_GPL(nvmf_get_address);
122 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function.
123 * @ctrl: Host NVMe controller instance maintaining the admin
124 * queue used to submit the property read command to
125 * the allocated NVMe controller resource on the target system.
126 * @off: Starting offset value of the targeted property
127 * register (see the fabrics section of the NVMe standard).
128 * @val: OUTPUT parameter that will contain the value of
129 * the property after a successful read.
131 * Used by the host system to retrieve a 32-bit capsule property value
132 * from an NVMe controller on the target system.
134 * ("Capsule property" is an "PCIe register concept" applied to the
135 * NVMe fabrics space.)
139 * > 0: NVMe error status code
140 * < 0: Linux errno error code
142 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
144 struct nvme_command cmd;
145 union nvme_result res;
148 memset(&cmd, 0, sizeof(cmd));
149 cmd.prop_get.opcode = nvme_fabrics_command;
150 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
151 cmd.prop_get.offset = cpu_to_le32(off);
153 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
154 NVME_QID_ANY, 0, 0, false);
157 *val = le64_to_cpu(res.u64);
158 if (unlikely(ret != 0))
159 dev_err(ctrl->device,
160 "Property Get error: %d, offset %#x\n",
161 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
165 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
168 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function.
169 * @ctrl: Host NVMe controller instance maintaining the admin
170 * queue used to submit the property read command to
171 * the allocated controller resource on the target system.
172 * @off: Starting offset value of the targeted property
173 * register (see the fabrics section of the NVMe standard).
174 * @val: OUTPUT parameter that will contain the value of
175 * the property after a successful read.
177 * Used by the host system to retrieve a 64-bit capsule property value
178 * from an NVMe controller on the target system.
180 * ("Capsule property" is an "PCIe register concept" applied to the
181 * NVMe fabrics space.)
185 * > 0: NVMe error status code
186 * < 0: Linux errno error code
188 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
190 struct nvme_command cmd;
191 union nvme_result res;
194 memset(&cmd, 0, sizeof(cmd));
195 cmd.prop_get.opcode = nvme_fabrics_command;
196 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
197 cmd.prop_get.attrib = 1;
198 cmd.prop_get.offset = cpu_to_le32(off);
200 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
201 NVME_QID_ANY, 0, 0, false);
204 *val = le64_to_cpu(res.u64);
205 if (unlikely(ret != 0))
206 dev_err(ctrl->device,
207 "Property Get error: %d, offset %#x\n",
208 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
211 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
214 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function.
215 * @ctrl: Host NVMe controller instance maintaining the admin
216 * queue used to submit the property read command to
217 * the allocated NVMe controller resource on the target system.
218 * @off: Starting offset value of the targeted property
219 * register (see the fabrics section of the NVMe standard).
220 * @val: Input parameter that contains the value to be
221 * written to the property.
223 * Used by the NVMe host system to write a 32-bit capsule property value
224 * to an NVMe controller on the target system.
226 * ("Capsule property" is an "PCIe register concept" applied to the
227 * NVMe fabrics space.)
230 * 0: successful write
231 * > 0: NVMe error status code
232 * < 0: Linux errno error code
234 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
236 struct nvme_command cmd;
239 memset(&cmd, 0, sizeof(cmd));
240 cmd.prop_set.opcode = nvme_fabrics_command;
241 cmd.prop_set.fctype = nvme_fabrics_type_property_set;
242 cmd.prop_set.attrib = 0;
243 cmd.prop_set.offset = cpu_to_le32(off);
244 cmd.prop_set.value = cpu_to_le64(val);
246 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
247 NVME_QID_ANY, 0, 0, false);
249 dev_err(ctrl->device,
250 "Property Set error: %d, offset %#x\n",
251 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
254 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
257 * nvmf_log_connect_error() - Error-parsing-diagnostic print
258 * out function for connect() errors.
260 * @ctrl: the specific /dev/nvmeX device that had the error.
262 * @errval: Error code to be decoded in a more human-friendly
265 * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
267 * @cmd: This is the SQE portion of a submission capsule.
269 * @data: This is the "Data" portion of a submission capsule.
271 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
272 int errval, int offset, struct nvme_command *cmd,
273 struct nvmf_connect_data *data)
275 int err_sctype = errval & (~NVME_SC_DNR);
277 switch (err_sctype) {
279 case (NVME_SC_CONNECT_INVALID_PARAM):
281 char *inv_data = "Connect Invalid Data Parameter";
283 switch (offset & 0xffff) {
284 case (offsetof(struct nvmf_connect_data, cntlid)):
285 dev_err(ctrl->device,
287 inv_data, data->cntlid);
289 case (offsetof(struct nvmf_connect_data, hostnqn)):
290 dev_err(ctrl->device,
291 "%s, hostnqn \"%s\"\n",
292 inv_data, data->hostnqn);
294 case (offsetof(struct nvmf_connect_data, subsysnqn)):
295 dev_err(ctrl->device,
296 "%s, subsysnqn \"%s\"\n",
297 inv_data, data->subsysnqn);
300 dev_err(ctrl->device,
301 "%s, starting byte offset: %d\n",
302 inv_data, offset & 0xffff);
306 char *inv_sqe = "Connect Invalid SQE Parameter";
309 case (offsetof(struct nvmf_connect_command, qid)):
310 dev_err(ctrl->device,
312 inv_sqe, cmd->connect.qid);
315 dev_err(ctrl->device,
316 "%s, starting byte offset: %d\n",
322 case NVME_SC_CONNECT_INVALID_HOST:
323 dev_err(ctrl->device,
324 "Connect for subsystem %s is not allowed, hostnqn: %s\n",
325 data->subsysnqn, data->hostnqn);
328 case NVME_SC_CONNECT_CTRL_BUSY:
329 dev_err(ctrl->device,
330 "Connect command failed: controller is busy or not available\n");
333 case NVME_SC_CONNECT_FORMAT:
334 dev_err(ctrl->device,
335 "Connect incompatible format: %d",
336 cmd->connect.recfmt);
340 dev_err(ctrl->device,
341 "Connect command failed, error wo/DNR bit: %d\n",
344 } /* switch (err_sctype) */
348 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
350 * @ctrl: Host nvme controller instance used to request
351 * a new NVMe controller allocation on the target
352 * system and establish an NVMe Admin connection to
355 * This function enables an NVMe host device to request a new allocation of
356 * an NVMe controller resource on a target system as well establish a
357 * fabrics-protocol connection of the NVMe Admin queue between the
358 * host system device and the allocated NVMe controller on the
359 * target system via a NVMe Fabrics "Connect" command.
363 * > 0: NVMe error status code
364 * < 0: Linux errno error code
367 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
369 struct nvme_command cmd;
370 union nvme_result res;
371 struct nvmf_connect_data *data;
374 memset(&cmd, 0, sizeof(cmd));
375 cmd.connect.opcode = nvme_fabrics_command;
376 cmd.connect.fctype = nvme_fabrics_type_connect;
378 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
381 * Set keep-alive timeout in seconds granularity (ms * 1000)
382 * and add a grace period for controller kato enforcement
384 cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
385 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
387 if (ctrl->opts->disable_sqflow)
388 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
390 data = kzalloc(sizeof(*data), GFP_KERNEL);
394 uuid_copy(&data->hostid, &ctrl->opts->host->id);
395 data->cntlid = cpu_to_le16(0xffff);
396 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
397 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
399 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,
400 data, sizeof(*data), 0, NVME_QID_ANY, 1,
401 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, false);
403 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
408 ctrl->cntlid = le16_to_cpu(res.u16);
414 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
417 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
419 * @ctrl: Host nvme controller instance used to establish an
420 * NVMe I/O queue connection to the already allocated NVMe
421 * controller on the target system.
422 * @qid: NVMe I/O queue number for the new I/O connection between
423 * host and target (note qid == 0 is illegal as this is
424 * the Admin queue, per NVMe standard).
425 * @poll: Whether or not to poll for the completion of the connect cmd.
427 * This function issues a fabrics-protocol connection
428 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
429 * between the host system device and the allocated NVMe controller
430 * on the target system.
434 * > 0: NVMe error status code
435 * < 0: Linux errno error code
437 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid, bool poll)
439 struct nvme_command cmd;
440 struct nvmf_connect_data *data;
441 union nvme_result res;
444 memset(&cmd, 0, sizeof(cmd));
445 cmd.connect.opcode = nvme_fabrics_command;
446 cmd.connect.fctype = nvme_fabrics_type_connect;
447 cmd.connect.qid = cpu_to_le16(qid);
448 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
450 if (ctrl->opts->disable_sqflow)
451 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
453 data = kzalloc(sizeof(*data), GFP_KERNEL);
457 uuid_copy(&data->hostid, &ctrl->opts->host->id);
458 data->cntlid = cpu_to_le16(ctrl->cntlid);
459 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
460 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
462 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
463 data, sizeof(*data), 0, qid, 1,
464 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, poll);
466 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
472 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
474 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
476 if (ctrl->opts->max_reconnects == -1 ||
477 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
482 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
485 * nvmf_register_transport() - NVMe Fabrics Library registration function.
486 * @ops: Transport ops instance to be registered to the
487 * common fabrics library.
489 * API function that registers the type of specific transport fabric
490 * being implemented to the common NVMe fabrics library. Part of
491 * the overall init sequence of starting up a fabrics driver.
493 int nvmf_register_transport(struct nvmf_transport_ops *ops)
495 if (!ops->create_ctrl)
498 down_write(&nvmf_transports_rwsem);
499 list_add_tail(&ops->entry, &nvmf_transports);
500 up_write(&nvmf_transports_rwsem);
504 EXPORT_SYMBOL_GPL(nvmf_register_transport);
507 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
508 * @ops: Transport ops instance to be unregistered from the
509 * common fabrics library.
511 * Fabrics API function that unregisters the type of specific transport
512 * fabric being implemented from the common NVMe fabrics library.
513 * Part of the overall exit sequence of unloading the implemented driver.
515 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
517 down_write(&nvmf_transports_rwsem);
518 list_del(&ops->entry);
519 up_write(&nvmf_transports_rwsem);
521 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
523 static struct nvmf_transport_ops *nvmf_lookup_transport(
524 struct nvmf_ctrl_options *opts)
526 struct nvmf_transport_ops *ops;
528 lockdep_assert_held(&nvmf_transports_rwsem);
530 list_for_each_entry(ops, &nvmf_transports, entry) {
531 if (strcmp(ops->name, opts->transport) == 0)
539 * For something we're not in a state to send to the device the default action
540 * is to busy it and retry it after the controller state is recovered. However,
541 * if the controller is deleting or if anything is marked for failfast or
542 * nvme multipath it is immediately failed.
544 * Note: commands used to initialize the controller will be marked for failfast.
545 * Note: nvme cli/ioctl commands are marked for failfast.
547 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
550 if (ctrl->state != NVME_CTRL_DELETING &&
551 ctrl->state != NVME_CTRL_DEAD &&
552 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
553 return BLK_STS_RESOURCE;
555 nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR;
556 blk_mq_start_request(rq);
557 nvme_complete_rq(rq);
560 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
562 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
565 struct nvme_request *req = nvme_req(rq);
568 * If we are in some state of setup or teardown only allow
569 * internally generated commands.
571 if (!blk_rq_is_passthrough(rq) || (req->flags & NVME_REQ_USERCMD))
575 * Only allow commands on a live queue, except for the connect command,
576 * which is require to set the queue live in the appropinquate states.
578 switch (ctrl->state) {
580 case NVME_CTRL_CONNECTING:
581 if (nvme_is_fabrics(req->cmd) &&
582 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
593 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
595 static const match_table_t opt_tokens = {
596 { NVMF_OPT_TRANSPORT, "transport=%s" },
597 { NVMF_OPT_TRADDR, "traddr=%s" },
598 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
599 { NVMF_OPT_NQN, "nqn=%s" },
600 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
601 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
602 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
603 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
604 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
605 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
606 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
607 { NVMF_OPT_HOST_ID, "hostid=%s" },
608 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
609 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" },
610 { NVMF_OPT_HDR_DIGEST, "hdr_digest" },
611 { NVMF_OPT_DATA_DIGEST, "data_digest" },
612 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" },
613 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" },
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;
633 opts->hdr_digest = false;
634 opts->data_digest = false;
636 options = o = kstrdup(buf, GFP_KERNEL);
642 while ((p = strsep(&o, ",\n")) != NULL) {
646 token = match_token(p, opt_tokens, args);
649 case NVMF_OPT_TRANSPORT:
650 p = match_strdup(args);
655 kfree(opts->transport);
659 p = match_strdup(args);
664 kfree(opts->subsysnqn);
666 nqnlen = strlen(opts->subsysnqn);
667 if (nqnlen >= NVMF_NQN_SIZE) {
668 pr_err("%s needs to be < %d bytes\n",
669 opts->subsysnqn, NVMF_NQN_SIZE);
673 opts->discovery_nqn =
674 !(strcmp(opts->subsysnqn,
675 NVME_DISC_SUBSYS_NAME));
677 case NVMF_OPT_TRADDR:
678 p = match_strdup(args);
686 case NVMF_OPT_TRSVCID:
687 p = match_strdup(args);
692 kfree(opts->trsvcid);
695 case NVMF_OPT_QUEUE_SIZE:
696 if (match_int(args, &token)) {
700 if (token < NVMF_MIN_QUEUE_SIZE ||
701 token > NVMF_MAX_QUEUE_SIZE) {
702 pr_err("Invalid queue_size %d\n", token);
706 opts->queue_size = token;
708 case NVMF_OPT_NR_IO_QUEUES:
709 if (match_int(args, &token)) {
714 pr_err("Invalid number of IOQs %d\n", token);
718 if (opts->discovery_nqn) {
719 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
723 opts->nr_io_queues = min_t(unsigned int,
724 num_online_cpus(), token);
727 if (match_int(args, &token)) {
733 pr_err("Invalid keep_alive_tmo %d\n", token);
736 } else if (token == 0 && !opts->discovery_nqn) {
737 /* Allowed for debug */
738 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
742 if (opts->discovery_nqn && opts->kato) {
743 pr_err("Discovery controllers cannot accept KATO != 0\n");
749 case NVMF_OPT_CTRL_LOSS_TMO:
750 if (match_int(args, &token)) {
756 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
757 ctrl_loss_tmo = token;
759 case NVMF_OPT_HOSTNQN:
761 pr_err("hostnqn already user-assigned: %s\n",
766 p = match_strdup(args);
772 if (nqnlen >= NVMF_NQN_SIZE) {
773 pr_err("%s needs to be < %d bytes\n",
779 nvmf_host_put(opts->host);
780 opts->host = nvmf_host_add(p);
787 case NVMF_OPT_RECONNECT_DELAY:
788 if (match_int(args, &token)) {
793 pr_err("Invalid reconnect_delay %d\n", token);
797 opts->reconnect_delay = token;
799 case NVMF_OPT_HOST_TRADDR:
800 p = match_strdup(args);
805 kfree(opts->host_traddr);
806 opts->host_traddr = p;
808 case NVMF_OPT_HOST_ID:
809 p = match_strdup(args);
814 ret = uuid_parse(p, &hostid);
816 pr_err("Invalid hostid %s\n", p);
823 case NVMF_OPT_DUP_CONNECT:
824 opts->duplicate_connect = true;
826 case NVMF_OPT_DISABLE_SQFLOW:
827 opts->disable_sqflow = true;
829 case NVMF_OPT_HDR_DIGEST:
830 opts->hdr_digest = true;
832 case NVMF_OPT_DATA_DIGEST:
833 opts->data_digest = true;
835 case NVMF_OPT_NR_WRITE_QUEUES:
836 if (match_int(args, &token)) {
841 pr_err("Invalid nr_write_queues %d\n", token);
845 opts->nr_write_queues = token;
847 case NVMF_OPT_NR_POLL_QUEUES:
848 if (match_int(args, &token)) {
853 pr_err("Invalid nr_poll_queues %d\n", token);
857 opts->nr_poll_queues = token;
860 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
867 if (opts->discovery_nqn) {
869 opts->nr_io_queues = 0;
870 opts->nr_write_queues = 0;
871 opts->nr_poll_queues = 0;
872 opts->duplicate_connect = true;
874 if (ctrl_loss_tmo < 0)
875 opts->max_reconnects = -1;
877 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
878 opts->reconnect_delay);
881 kref_get(&nvmf_default_host->ref);
882 opts->host = nvmf_default_host;
885 uuid_copy(&opts->host->id, &hostid);
892 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
893 unsigned int required_opts)
895 if ((opts->mask & required_opts) != required_opts) {
898 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
899 if ((opt_tokens[i].token & required_opts) &&
900 !(opt_tokens[i].token & opts->mask)) {
901 pr_warn("missing parameter '%s'\n",
902 opt_tokens[i].pattern);
912 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
913 struct nvmf_ctrl_options *opts)
915 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
916 strcmp(opts->traddr, ctrl->opts->traddr) ||
917 strcmp(opts->trsvcid, ctrl->opts->trsvcid))
921 * Checking the local address is rough. In most cases, none is specified
922 * and the host port is selected by the stack.
924 * Assume no match if:
925 * - local address is specified and address is not the same
926 * - local address is not specified but remote is, or vice versa
927 * (admin using specific host_traddr when it matters).
929 if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
930 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
931 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
933 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
934 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
940 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
942 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
943 unsigned int allowed_opts)
945 if (opts->mask & ~allowed_opts) {
948 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
949 if ((opt_tokens[i].token & opts->mask) &&
950 (opt_tokens[i].token & ~allowed_opts)) {
951 pr_warn("invalid parameter '%s'\n",
952 opt_tokens[i].pattern);
962 void nvmf_free_options(struct nvmf_ctrl_options *opts)
964 nvmf_host_put(opts->host);
965 kfree(opts->transport);
967 kfree(opts->trsvcid);
968 kfree(opts->subsysnqn);
969 kfree(opts->host_traddr);
972 EXPORT_SYMBOL_GPL(nvmf_free_options);
974 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
975 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
976 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
977 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
978 NVMF_OPT_DISABLE_SQFLOW)
980 static struct nvme_ctrl *
981 nvmf_create_ctrl(struct device *dev, const char *buf)
983 struct nvmf_ctrl_options *opts;
984 struct nvmf_transport_ops *ops;
985 struct nvme_ctrl *ctrl;
988 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
990 return ERR_PTR(-ENOMEM);
992 ret = nvmf_parse_options(opts, buf);
997 request_module("nvme-%s", opts->transport);
1000 * Check the generic options first as we need a valid transport for
1001 * the lookup below. Then clear the generic flags so that transport
1002 * drivers don't have to care about them.
1004 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1007 opts->mask &= ~NVMF_REQUIRED_OPTS;
1009 down_read(&nvmf_transports_rwsem);
1010 ops = nvmf_lookup_transport(opts);
1012 pr_info("no handler found for transport %s.\n",
1018 if (!try_module_get(ops->module)) {
1022 up_read(&nvmf_transports_rwsem);
1024 ret = nvmf_check_required_opts(opts, ops->required_opts);
1026 goto out_module_put;
1027 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1028 ops->allowed_opts | ops->required_opts);
1030 goto out_module_put;
1032 ctrl = ops->create_ctrl(dev, opts);
1034 ret = PTR_ERR(ctrl);
1035 goto out_module_put;
1038 module_put(ops->module);
1042 module_put(ops->module);
1045 up_read(&nvmf_transports_rwsem);
1047 nvmf_free_options(opts);
1048 return ERR_PTR(ret);
1051 static struct class *nvmf_class;
1052 static struct device *nvmf_device;
1053 static DEFINE_MUTEX(nvmf_dev_mutex);
1055 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1056 size_t count, loff_t *pos)
1058 struct seq_file *seq_file = file->private_data;
1059 struct nvme_ctrl *ctrl;
1063 if (count > PAGE_SIZE)
1066 buf = memdup_user_nul(ubuf, count);
1068 return PTR_ERR(buf);
1070 mutex_lock(&nvmf_dev_mutex);
1071 if (seq_file->private) {
1076 ctrl = nvmf_create_ctrl(nvmf_device, buf);
1078 ret = PTR_ERR(ctrl);
1082 seq_file->private = ctrl;
1085 mutex_unlock(&nvmf_dev_mutex);
1087 return ret ? ret : count;
1090 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1092 struct nvme_ctrl *ctrl;
1095 mutex_lock(&nvmf_dev_mutex);
1096 ctrl = seq_file->private;
1102 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1103 ctrl->instance, ctrl->cntlid);
1106 mutex_unlock(&nvmf_dev_mutex);
1110 static int nvmf_dev_open(struct inode *inode, struct file *file)
1113 * The miscdevice code initializes file->private_data, but doesn't
1114 * make use of it later.
1116 file->private_data = NULL;
1117 return single_open(file, nvmf_dev_show, NULL);
1120 static int nvmf_dev_release(struct inode *inode, struct file *file)
1122 struct seq_file *seq_file = file->private_data;
1123 struct nvme_ctrl *ctrl = seq_file->private;
1126 nvme_put_ctrl(ctrl);
1127 return single_release(inode, file);
1130 static const struct file_operations nvmf_dev_fops = {
1131 .owner = THIS_MODULE,
1132 .write = nvmf_dev_write,
1134 .open = nvmf_dev_open,
1135 .release = nvmf_dev_release,
1138 static struct miscdevice nvmf_misc = {
1139 .minor = MISC_DYNAMIC_MINOR,
1140 .name = "nvme-fabrics",
1141 .fops = &nvmf_dev_fops,
1144 static int __init nvmf_init(void)
1148 nvmf_default_host = nvmf_host_default();
1149 if (!nvmf_default_host)
1152 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1153 if (IS_ERR(nvmf_class)) {
1154 pr_err("couldn't register class nvme-fabrics\n");
1155 ret = PTR_ERR(nvmf_class);
1160 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1161 if (IS_ERR(nvmf_device)) {
1162 pr_err("couldn't create nvme-fabris device!\n");
1163 ret = PTR_ERR(nvmf_device);
1164 goto out_destroy_class;
1167 ret = misc_register(&nvmf_misc);
1169 pr_err("couldn't register misc device: %d\n", ret);
1170 goto out_destroy_device;
1176 device_destroy(nvmf_class, MKDEV(0, 0));
1178 class_destroy(nvmf_class);
1180 nvmf_host_put(nvmf_default_host);
1184 static void __exit nvmf_exit(void)
1186 misc_deregister(&nvmf_misc);
1187 device_destroy(nvmf_class, MKDEV(0, 0));
1188 class_destroy(nvmf_class);
1189 nvmf_host_put(nvmf_default_host);
1191 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1192 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1193 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1194 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1195 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1198 MODULE_LICENSE("GPL v2");
1200 module_init(nvmf_init);
1201 module_exit(nvmf_exit);