1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * NVMe over Fabrics common host code.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5 */
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>
13 #include "nvme.h"
14 #include "fabrics.h"
15
16 static LIST_HEAD(nvmf_transports);
17 static DECLARE_RWSEM(nvmf_transports_rwsem);
18
19 static LIST_HEAD(nvmf_hosts);
20 static DEFINE_MUTEX(nvmf_hosts_mutex);
21
22 static struct nvmf_host *nvmf_default_host;
23
__nvmf_host_find(const char * hostnqn)24 static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
25 {
26 struct nvmf_host *host;
27
28 list_for_each_entry(host, &nvmf_hosts, list) {
29 if (!strcmp(host->nqn, hostnqn))
30 return host;
31 }
32
33 return NULL;
34 }
35
nvmf_host_add(const char * hostnqn)36 static struct nvmf_host *nvmf_host_add(const char *hostnqn)
37 {
38 struct nvmf_host *host;
39
40 mutex_lock(&nvmf_hosts_mutex);
41 host = __nvmf_host_find(hostnqn);
42 if (host) {
43 kref_get(&host->ref);
44 goto out_unlock;
45 }
46
47 host = kmalloc(sizeof(*host), GFP_KERNEL);
48 if (!host)
49 goto out_unlock;
50
51 kref_init(&host->ref);
52 strlcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
53
54 list_add_tail(&host->list, &nvmf_hosts);
55 out_unlock:
56 mutex_unlock(&nvmf_hosts_mutex);
57 return host;
58 }
59
nvmf_host_default(void)60 static struct nvmf_host *nvmf_host_default(void)
61 {
62 struct nvmf_host *host;
63
64 host = kmalloc(sizeof(*host), GFP_KERNEL);
65 if (!host)
66 return NULL;
67
68 kref_init(&host->ref);
69 uuid_gen(&host->id);
70 snprintf(host->nqn, NVMF_NQN_SIZE,
71 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id);
72
73 mutex_lock(&nvmf_hosts_mutex);
74 list_add_tail(&host->list, &nvmf_hosts);
75 mutex_unlock(&nvmf_hosts_mutex);
76
77 return host;
78 }
79
nvmf_host_destroy(struct kref * ref)80 static void nvmf_host_destroy(struct kref *ref)
81 {
82 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
83
84 mutex_lock(&nvmf_hosts_mutex);
85 list_del(&host->list);
86 mutex_unlock(&nvmf_hosts_mutex);
87
88 kfree(host);
89 }
90
nvmf_host_put(struct nvmf_host * host)91 static void nvmf_host_put(struct nvmf_host *host)
92 {
93 if (host)
94 kref_put(&host->ref, nvmf_host_destroy);
95 }
96
97 /**
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
101 * @size: buffer size
102 */
nvmf_get_address(struct nvme_ctrl * ctrl,char * buf,int size)103 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
104 {
105 int len = 0;
106
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 len += scnprintf(buf + len, size - len, "\n");
116
117 return len;
118 }
119 EXPORT_SYMBOL_GPL(nvmf_get_address);
120
121 /**
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.
130 *
131 * Used by the host system to retrieve a 32-bit capsule property value
132 * from an NVMe controller on the target system.
133 *
134 * ("Capsule property" is an "PCIe register concept" applied to the
135 * NVMe fabrics space.)
136 *
137 * Return:
138 * 0: successful read
139 * > 0: NVMe error status code
140 * < 0: Linux errno error code
141 */
nvmf_reg_read32(struct nvme_ctrl * ctrl,u32 off,u32 * val)142 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
143 {
144 struct nvme_command cmd;
145 union nvme_result res;
146 int ret;
147
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);
152
153 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
154 NVME_QID_ANY, 0, 0, false);
155
156 if (ret >= 0)
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);
162
163 return ret;
164 }
165 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
166
167 /**
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.
176 *
177 * Used by the host system to retrieve a 64-bit capsule property value
178 * from an NVMe controller on the target system.
179 *
180 * ("Capsule property" is an "PCIe register concept" applied to the
181 * NVMe fabrics space.)
182 *
183 * Return:
184 * 0: successful read
185 * > 0: NVMe error status code
186 * < 0: Linux errno error code
187 */
nvmf_reg_read64(struct nvme_ctrl * ctrl,u32 off,u64 * val)188 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
189 {
190 struct nvme_command cmd;
191 union nvme_result res;
192 int ret;
193
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);
199
200 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
201 NVME_QID_ANY, 0, 0, false);
202
203 if (ret >= 0)
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);
209 return ret;
210 }
211 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
212
213 /**
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.
222 *
223 * Used by the NVMe host system to write a 32-bit capsule property value
224 * to an NVMe controller on the target system.
225 *
226 * ("Capsule property" is an "PCIe register concept" applied to the
227 * NVMe fabrics space.)
228 *
229 * Return:
230 * 0: successful write
231 * > 0: NVMe error status code
232 * < 0: Linux errno error code
233 */
nvmf_reg_write32(struct nvme_ctrl * ctrl,u32 off,u32 val)234 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
235 {
236 struct nvme_command cmd;
237 int ret;
238
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);
245
246 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 0,
247 NVME_QID_ANY, 0, 0, false);
248 if (unlikely(ret))
249 dev_err(ctrl->device,
250 "Property Set error: %d, offset %#x\n",
251 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
252 return ret;
253 }
254 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
255
256 /**
257 * nvmf_log_connect_error() - Error-parsing-diagnostic print
258 * out function for connect() errors.
259 *
260 * @ctrl: the specific /dev/nvmeX device that had the error.
261 *
262 * @errval: Error code to be decoded in a more human-friendly
263 * printout.
264 *
265 * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
266 *
267 * @cmd: This is the SQE portion of a submission capsule.
268 *
269 * @data: This is the "Data" portion of a submission capsule.
270 */
nvmf_log_connect_error(struct nvme_ctrl * ctrl,int errval,int offset,struct nvme_command * cmd,struct nvmf_connect_data * data)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)
274 {
275 int err_sctype = errval & (~NVME_SC_DNR);
276
277 switch (err_sctype) {
278
279 case (NVME_SC_CONNECT_INVALID_PARAM):
280 if (offset >> 16) {
281 char *inv_data = "Connect Invalid Data Parameter";
282
283 switch (offset & 0xffff) {
284 case (offsetof(struct nvmf_connect_data, cntlid)):
285 dev_err(ctrl->device,
286 "%s, cntlid: %d\n",
287 inv_data, data->cntlid);
288 break;
289 case (offsetof(struct nvmf_connect_data, hostnqn)):
290 dev_err(ctrl->device,
291 "%s, hostnqn \"%s\"\n",
292 inv_data, data->hostnqn);
293 break;
294 case (offsetof(struct nvmf_connect_data, subsysnqn)):
295 dev_err(ctrl->device,
296 "%s, subsysnqn \"%s\"\n",
297 inv_data, data->subsysnqn);
298 break;
299 default:
300 dev_err(ctrl->device,
301 "%s, starting byte offset: %d\n",
302 inv_data, offset & 0xffff);
303 break;
304 }
305 } else {
306 char *inv_sqe = "Connect Invalid SQE Parameter";
307
308 switch (offset) {
309 case (offsetof(struct nvmf_connect_command, qid)):
310 dev_err(ctrl->device,
311 "%s, qid %d\n",
312 inv_sqe, cmd->connect.qid);
313 break;
314 default:
315 dev_err(ctrl->device,
316 "%s, starting byte offset: %d\n",
317 inv_sqe, offset);
318 }
319 }
320 break;
321
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);
326 break;
327
328 case NVME_SC_CONNECT_CTRL_BUSY:
329 dev_err(ctrl->device,
330 "Connect command failed: controller is busy or not available\n");
331 break;
332
333 case NVME_SC_CONNECT_FORMAT:
334 dev_err(ctrl->device,
335 "Connect incompatible format: %d",
336 cmd->connect.recfmt);
337 break;
338
339 case NVME_SC_HOST_PATH_ERROR:
340 dev_err(ctrl->device,
341 "Connect command failed: host path error\n");
342 break;
343
344 default:
345 dev_err(ctrl->device,
346 "Connect command failed, error wo/DNR bit: %d\n",
347 err_sctype);
348 break;
349 } /* switch (err_sctype) */
350 }
351
352 /**
353 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
354 * API function.
355 * @ctrl: Host nvme controller instance used to request
356 * a new NVMe controller allocation on the target
357 * system and establish an NVMe Admin connection to
358 * that controller.
359 *
360 * This function enables an NVMe host device to request a new allocation of
361 * an NVMe controller resource on a target system as well establish a
362 * fabrics-protocol connection of the NVMe Admin queue between the
363 * host system device and the allocated NVMe controller on the
364 * target system via a NVMe Fabrics "Connect" command.
365 *
366 * Return:
367 * 0: success
368 * > 0: NVMe error status code
369 * < 0: Linux errno error code
370 *
371 */
nvmf_connect_admin_queue(struct nvme_ctrl * ctrl)372 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
373 {
374 struct nvme_command cmd;
375 union nvme_result res;
376 struct nvmf_connect_data *data;
377 int ret;
378
379 memset(&cmd, 0, sizeof(cmd));
380 cmd.connect.opcode = nvme_fabrics_command;
381 cmd.connect.fctype = nvme_fabrics_type_connect;
382 cmd.connect.qid = 0;
383 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
384
385 /*
386 * Set keep-alive timeout in seconds granularity (ms * 1000)
387 * and add a grace period for controller kato enforcement
388 */
389 cmd.connect.kato = ctrl->kato ?
390 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000) : 0;
391
392 if (ctrl->opts->disable_sqflow)
393 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
394
395 data = kzalloc(sizeof(*data), GFP_KERNEL);
396 if (!data)
397 return -ENOMEM;
398
399 uuid_copy(&data->hostid, &ctrl->opts->host->id);
400 data->cntlid = cpu_to_le16(0xffff);
401 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
402 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
403
404 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
405 data, sizeof(*data), 0, NVME_QID_ANY, 1,
406 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, false);
407 if (ret) {
408 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
409 &cmd, data);
410 goto out_free_data;
411 }
412
413 ctrl->cntlid = le16_to_cpu(res.u16);
414
415 out_free_data:
416 kfree(data);
417 return ret;
418 }
419 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
420
421 /**
422 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
423 * API function.
424 * @ctrl: Host nvme controller instance used to establish an
425 * NVMe I/O queue connection to the already allocated NVMe
426 * controller on the target system.
427 * @qid: NVMe I/O queue number for the new I/O connection between
428 * host and target (note qid == 0 is illegal as this is
429 * the Admin queue, per NVMe standard).
430 * @poll: Whether or not to poll for the completion of the connect cmd.
431 *
432 * This function issues a fabrics-protocol connection
433 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
434 * between the host system device and the allocated NVMe controller
435 * on the target system.
436 *
437 * Return:
438 * 0: success
439 * > 0: NVMe error status code
440 * < 0: Linux errno error code
441 */
nvmf_connect_io_queue(struct nvme_ctrl * ctrl,u16 qid,bool poll)442 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid, bool poll)
443 {
444 struct nvme_command cmd;
445 struct nvmf_connect_data *data;
446 union nvme_result res;
447 int ret;
448
449 memset(&cmd, 0, sizeof(cmd));
450 cmd.connect.opcode = nvme_fabrics_command;
451 cmd.connect.fctype = nvme_fabrics_type_connect;
452 cmd.connect.qid = cpu_to_le16(qid);
453 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
454
455 if (ctrl->opts->disable_sqflow)
456 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
457
458 data = kzalloc(sizeof(*data), GFP_KERNEL);
459 if (!data)
460 return -ENOMEM;
461
462 uuid_copy(&data->hostid, &ctrl->opts->host->id);
463 data->cntlid = cpu_to_le16(ctrl->cntlid);
464 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
465 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
466
467 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
468 data, sizeof(*data), 0, qid, 1,
469 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, poll);
470 if (ret) {
471 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
472 &cmd, data);
473 }
474 kfree(data);
475 return ret;
476 }
477 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
478
nvmf_should_reconnect(struct nvme_ctrl * ctrl)479 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
480 {
481 if (ctrl->opts->max_reconnects == -1 ||
482 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
483 return true;
484
485 return false;
486 }
487 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
488
489 /**
490 * nvmf_register_transport() - NVMe Fabrics Library registration function.
491 * @ops: Transport ops instance to be registered to the
492 * common fabrics library.
493 *
494 * API function that registers the type of specific transport fabric
495 * being implemented to the common NVMe fabrics library. Part of
496 * the overall init sequence of starting up a fabrics driver.
497 */
nvmf_register_transport(struct nvmf_transport_ops * ops)498 int nvmf_register_transport(struct nvmf_transport_ops *ops)
499 {
500 if (!ops->create_ctrl)
501 return -EINVAL;
502
503 down_write(&nvmf_transports_rwsem);
504 list_add_tail(&ops->entry, &nvmf_transports);
505 up_write(&nvmf_transports_rwsem);
506
507 return 0;
508 }
509 EXPORT_SYMBOL_GPL(nvmf_register_transport);
510
511 /**
512 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
513 * @ops: Transport ops instance to be unregistered from the
514 * common fabrics library.
515 *
516 * Fabrics API function that unregisters the type of specific transport
517 * fabric being implemented from the common NVMe fabrics library.
518 * Part of the overall exit sequence of unloading the implemented driver.
519 */
nvmf_unregister_transport(struct nvmf_transport_ops * ops)520 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
521 {
522 down_write(&nvmf_transports_rwsem);
523 list_del(&ops->entry);
524 up_write(&nvmf_transports_rwsem);
525 }
526 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
527
nvmf_lookup_transport(struct nvmf_ctrl_options * opts)528 static struct nvmf_transport_ops *nvmf_lookup_transport(
529 struct nvmf_ctrl_options *opts)
530 {
531 struct nvmf_transport_ops *ops;
532
533 lockdep_assert_held(&nvmf_transports_rwsem);
534
535 list_for_each_entry(ops, &nvmf_transports, entry) {
536 if (strcmp(ops->name, opts->transport) == 0)
537 return ops;
538 }
539
540 return NULL;
541 }
542
543 /*
544 * For something we're not in a state to send to the device the default action
545 * is to busy it and retry it after the controller state is recovered. However,
546 * if the controller is deleting or if anything is marked for failfast or
547 * nvme multipath it is immediately failed.
548 *
549 * Note: commands used to initialize the controller will be marked for failfast.
550 * Note: nvme cli/ioctl commands are marked for failfast.
551 */
nvmf_fail_nonready_command(struct nvme_ctrl * ctrl,struct request * rq)552 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
553 struct request *rq)
554 {
555 if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
556 ctrl->state != NVME_CTRL_DEAD &&
557 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
558 return BLK_STS_RESOURCE;
559
560 nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR;
561 blk_mq_start_request(rq);
562 nvme_complete_rq(rq);
563 return BLK_STS_OK;
564 }
565 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
566
__nvmf_check_ready(struct nvme_ctrl * ctrl,struct request * rq,bool queue_live)567 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
568 bool queue_live)
569 {
570 struct nvme_request *req = nvme_req(rq);
571
572 /*
573 * currently we have a problem sending passthru commands
574 * on the admin_q if the controller is not LIVE because we can't
575 * make sure that they are going out after the admin connect,
576 * controller enable and/or other commands in the initialization
577 * sequence. until the controller will be LIVE, fail with
578 * BLK_STS_RESOURCE so that they will be rescheduled.
579 */
580 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
581 return false;
582
583 /*
584 * Only allow commands on a live queue, except for the connect command,
585 * which is require to set the queue live in the appropinquate states.
586 */
587 switch (ctrl->state) {
588 case NVME_CTRL_CONNECTING:
589 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
590 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
591 return true;
592 break;
593 default:
594 break;
595 case NVME_CTRL_DEAD:
596 return false;
597 }
598
599 return queue_live;
600 }
601 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
602
603 static const match_table_t opt_tokens = {
604 { NVMF_OPT_TRANSPORT, "transport=%s" },
605 { NVMF_OPT_TRADDR, "traddr=%s" },
606 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
607 { NVMF_OPT_NQN, "nqn=%s" },
608 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
609 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
610 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
611 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
612 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
613 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
614 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
615 { NVMF_OPT_HOST_ID, "hostid=%s" },
616 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
617 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" },
618 { NVMF_OPT_HDR_DIGEST, "hdr_digest" },
619 { NVMF_OPT_DATA_DIGEST, "data_digest" },
620 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" },
621 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" },
622 { NVMF_OPT_TOS, "tos=%d" },
623 { NVMF_OPT_ERR, NULL }
624 };
625
nvmf_parse_options(struct nvmf_ctrl_options * opts,const char * buf)626 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
627 const char *buf)
628 {
629 substring_t args[MAX_OPT_ARGS];
630 char *options, *o, *p;
631 int token, ret = 0;
632 size_t nqnlen = 0;
633 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
634 uuid_t hostid;
635
636 /* Set defaults */
637 opts->queue_size = NVMF_DEF_QUEUE_SIZE;
638 opts->nr_io_queues = num_online_cpus();
639 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
640 opts->kato = NVME_DEFAULT_KATO;
641 opts->duplicate_connect = false;
642 opts->hdr_digest = false;
643 opts->data_digest = false;
644 opts->tos = -1; /* < 0 == use transport default */
645
646 options = o = kstrdup(buf, GFP_KERNEL);
647 if (!options)
648 return -ENOMEM;
649
650 uuid_gen(&hostid);
651
652 while ((p = strsep(&o, ",\n")) != NULL) {
653 if (!*p)
654 continue;
655
656 token = match_token(p, opt_tokens, args);
657 opts->mask |= token;
658 switch (token) {
659 case NVMF_OPT_TRANSPORT:
660 p = match_strdup(args);
661 if (!p) {
662 ret = -ENOMEM;
663 goto out;
664 }
665 kfree(opts->transport);
666 opts->transport = p;
667 break;
668 case NVMF_OPT_NQN:
669 p = match_strdup(args);
670 if (!p) {
671 ret = -ENOMEM;
672 goto out;
673 }
674 kfree(opts->subsysnqn);
675 opts->subsysnqn = p;
676 nqnlen = strlen(opts->subsysnqn);
677 if (nqnlen >= NVMF_NQN_SIZE) {
678 pr_err("%s needs to be < %d bytes\n",
679 opts->subsysnqn, NVMF_NQN_SIZE);
680 ret = -EINVAL;
681 goto out;
682 }
683 opts->discovery_nqn =
684 !(strcmp(opts->subsysnqn,
685 NVME_DISC_SUBSYS_NAME));
686 break;
687 case NVMF_OPT_TRADDR:
688 p = match_strdup(args);
689 if (!p) {
690 ret = -ENOMEM;
691 goto out;
692 }
693 kfree(opts->traddr);
694 opts->traddr = p;
695 break;
696 case NVMF_OPT_TRSVCID:
697 p = match_strdup(args);
698 if (!p) {
699 ret = -ENOMEM;
700 goto out;
701 }
702 kfree(opts->trsvcid);
703 opts->trsvcid = p;
704 break;
705 case NVMF_OPT_QUEUE_SIZE:
706 if (match_int(args, &token)) {
707 ret = -EINVAL;
708 goto out;
709 }
710 if (token < NVMF_MIN_QUEUE_SIZE ||
711 token > NVMF_MAX_QUEUE_SIZE) {
712 pr_err("Invalid queue_size %d\n", token);
713 ret = -EINVAL;
714 goto out;
715 }
716 opts->queue_size = token;
717 break;
718 case NVMF_OPT_NR_IO_QUEUES:
719 if (match_int(args, &token)) {
720 ret = -EINVAL;
721 goto out;
722 }
723 if (token <= 0) {
724 pr_err("Invalid number of IOQs %d\n", token);
725 ret = -EINVAL;
726 goto out;
727 }
728 if (opts->discovery_nqn) {
729 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
730 break;
731 }
732
733 opts->nr_io_queues = min_t(unsigned int,
734 num_online_cpus(), token);
735 break;
736 case NVMF_OPT_KATO:
737 if (match_int(args, &token)) {
738 ret = -EINVAL;
739 goto out;
740 }
741
742 if (token < 0) {
743 pr_err("Invalid keep_alive_tmo %d\n", token);
744 ret = -EINVAL;
745 goto out;
746 } else if (token == 0 && !opts->discovery_nqn) {
747 /* Allowed for debug */
748 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
749 }
750 opts->kato = token;
751 break;
752 case NVMF_OPT_CTRL_LOSS_TMO:
753 if (match_int(args, &token)) {
754 ret = -EINVAL;
755 goto out;
756 }
757
758 if (token < 0)
759 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
760 ctrl_loss_tmo = token;
761 break;
762 case NVMF_OPT_HOSTNQN:
763 if (opts->host) {
764 pr_err("hostnqn already user-assigned: %s\n",
765 opts->host->nqn);
766 ret = -EADDRINUSE;
767 goto out;
768 }
769 p = match_strdup(args);
770 if (!p) {
771 ret = -ENOMEM;
772 goto out;
773 }
774 nqnlen = strlen(p);
775 if (nqnlen >= NVMF_NQN_SIZE) {
776 pr_err("%s needs to be < %d bytes\n",
777 p, NVMF_NQN_SIZE);
778 kfree(p);
779 ret = -EINVAL;
780 goto out;
781 }
782 nvmf_host_put(opts->host);
783 opts->host = nvmf_host_add(p);
784 kfree(p);
785 if (!opts->host) {
786 ret = -ENOMEM;
787 goto out;
788 }
789 break;
790 case NVMF_OPT_RECONNECT_DELAY:
791 if (match_int(args, &token)) {
792 ret = -EINVAL;
793 goto out;
794 }
795 if (token <= 0) {
796 pr_err("Invalid reconnect_delay %d\n", token);
797 ret = -EINVAL;
798 goto out;
799 }
800 opts->reconnect_delay = token;
801 break;
802 case NVMF_OPT_HOST_TRADDR:
803 p = match_strdup(args);
804 if (!p) {
805 ret = -ENOMEM;
806 goto out;
807 }
808 kfree(opts->host_traddr);
809 opts->host_traddr = p;
810 break;
811 case NVMF_OPT_HOST_ID:
812 p = match_strdup(args);
813 if (!p) {
814 ret = -ENOMEM;
815 goto out;
816 }
817 ret = uuid_parse(p, &hostid);
818 if (ret) {
819 pr_err("Invalid hostid %s\n", p);
820 ret = -EINVAL;
821 kfree(p);
822 goto out;
823 }
824 kfree(p);
825 break;
826 case NVMF_OPT_DUP_CONNECT:
827 opts->duplicate_connect = true;
828 break;
829 case NVMF_OPT_DISABLE_SQFLOW:
830 opts->disable_sqflow = true;
831 break;
832 case NVMF_OPT_HDR_DIGEST:
833 opts->hdr_digest = true;
834 break;
835 case NVMF_OPT_DATA_DIGEST:
836 opts->data_digest = true;
837 break;
838 case NVMF_OPT_NR_WRITE_QUEUES:
839 if (match_int(args, &token)) {
840 ret = -EINVAL;
841 goto out;
842 }
843 if (token <= 0) {
844 pr_err("Invalid nr_write_queues %d\n", token);
845 ret = -EINVAL;
846 goto out;
847 }
848 opts->nr_write_queues = token;
849 break;
850 case NVMF_OPT_NR_POLL_QUEUES:
851 if (match_int(args, &token)) {
852 ret = -EINVAL;
853 goto out;
854 }
855 if (token <= 0) {
856 pr_err("Invalid nr_poll_queues %d\n", token);
857 ret = -EINVAL;
858 goto out;
859 }
860 opts->nr_poll_queues = token;
861 break;
862 case NVMF_OPT_TOS:
863 if (match_int(args, &token)) {
864 ret = -EINVAL;
865 goto out;
866 }
867 if (token < 0) {
868 pr_err("Invalid type of service %d\n", token);
869 ret = -EINVAL;
870 goto out;
871 }
872 if (token > 255) {
873 pr_warn("Clamping type of service to 255\n");
874 token = 255;
875 }
876 opts->tos = token;
877 break;
878 default:
879 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
880 p);
881 ret = -EINVAL;
882 goto out;
883 }
884 }
885
886 if (opts->discovery_nqn) {
887 opts->nr_io_queues = 0;
888 opts->nr_write_queues = 0;
889 opts->nr_poll_queues = 0;
890 opts->duplicate_connect = true;
891 }
892 if (ctrl_loss_tmo < 0)
893 opts->max_reconnects = -1;
894 else
895 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
896 opts->reconnect_delay);
897
898 if (!opts->host) {
899 kref_get(&nvmf_default_host->ref);
900 opts->host = nvmf_default_host;
901 }
902
903 uuid_copy(&opts->host->id, &hostid);
904
905 out:
906 kfree(options);
907 return ret;
908 }
909
nvmf_check_required_opts(struct nvmf_ctrl_options * opts,unsigned int required_opts)910 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
911 unsigned int required_opts)
912 {
913 if ((opts->mask & required_opts) != required_opts) {
914 int i;
915
916 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
917 if ((opt_tokens[i].token & required_opts) &&
918 !(opt_tokens[i].token & opts->mask)) {
919 pr_warn("missing parameter '%s'\n",
920 opt_tokens[i].pattern);
921 }
922 }
923
924 return -EINVAL;
925 }
926
927 return 0;
928 }
929
nvmf_ip_options_match(struct nvme_ctrl * ctrl,struct nvmf_ctrl_options * opts)930 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
931 struct nvmf_ctrl_options *opts)
932 {
933 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
934 strcmp(opts->traddr, ctrl->opts->traddr) ||
935 strcmp(opts->trsvcid, ctrl->opts->trsvcid))
936 return false;
937
938 /*
939 * Checking the local address is rough. In most cases, none is specified
940 * and the host port is selected by the stack.
941 *
942 * Assume no match if:
943 * - local address is specified and address is not the same
944 * - local address is not specified but remote is, or vice versa
945 * (admin using specific host_traddr when it matters).
946 */
947 if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
948 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
949 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
950 return false;
951 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
952 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
953 return false;
954 }
955
956 return true;
957 }
958 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
959
nvmf_check_allowed_opts(struct nvmf_ctrl_options * opts,unsigned int allowed_opts)960 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
961 unsigned int allowed_opts)
962 {
963 if (opts->mask & ~allowed_opts) {
964 int i;
965
966 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
967 if ((opt_tokens[i].token & opts->mask) &&
968 (opt_tokens[i].token & ~allowed_opts)) {
969 pr_warn("invalid parameter '%s'\n",
970 opt_tokens[i].pattern);
971 }
972 }
973
974 return -EINVAL;
975 }
976
977 return 0;
978 }
979
nvmf_free_options(struct nvmf_ctrl_options * opts)980 void nvmf_free_options(struct nvmf_ctrl_options *opts)
981 {
982 nvmf_host_put(opts->host);
983 kfree(opts->transport);
984 kfree(opts->traddr);
985 kfree(opts->trsvcid);
986 kfree(opts->subsysnqn);
987 kfree(opts->host_traddr);
988 kfree(opts);
989 }
990 EXPORT_SYMBOL_GPL(nvmf_free_options);
991
992 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
993 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
994 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
995 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
996 NVMF_OPT_DISABLE_SQFLOW)
997
998 static struct nvme_ctrl *
nvmf_create_ctrl(struct device * dev,const char * buf)999 nvmf_create_ctrl(struct device *dev, const char *buf)
1000 {
1001 struct nvmf_ctrl_options *opts;
1002 struct nvmf_transport_ops *ops;
1003 struct nvme_ctrl *ctrl;
1004 int ret;
1005
1006 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1007 if (!opts)
1008 return ERR_PTR(-ENOMEM);
1009
1010 ret = nvmf_parse_options(opts, buf);
1011 if (ret)
1012 goto out_free_opts;
1013
1014
1015 request_module("nvme-%s", opts->transport);
1016
1017 /*
1018 * Check the generic options first as we need a valid transport for
1019 * the lookup below. Then clear the generic flags so that transport
1020 * drivers don't have to care about them.
1021 */
1022 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1023 if (ret)
1024 goto out_free_opts;
1025 opts->mask &= ~NVMF_REQUIRED_OPTS;
1026
1027 down_read(&nvmf_transports_rwsem);
1028 ops = nvmf_lookup_transport(opts);
1029 if (!ops) {
1030 pr_info("no handler found for transport %s.\n",
1031 opts->transport);
1032 ret = -EINVAL;
1033 goto out_unlock;
1034 }
1035
1036 if (!try_module_get(ops->module)) {
1037 ret = -EBUSY;
1038 goto out_unlock;
1039 }
1040 up_read(&nvmf_transports_rwsem);
1041
1042 ret = nvmf_check_required_opts(opts, ops->required_opts);
1043 if (ret)
1044 goto out_module_put;
1045 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1046 ops->allowed_opts | ops->required_opts);
1047 if (ret)
1048 goto out_module_put;
1049
1050 ctrl = ops->create_ctrl(dev, opts);
1051 if (IS_ERR(ctrl)) {
1052 ret = PTR_ERR(ctrl);
1053 goto out_module_put;
1054 }
1055
1056 module_put(ops->module);
1057 return ctrl;
1058
1059 out_module_put:
1060 module_put(ops->module);
1061 goto out_free_opts;
1062 out_unlock:
1063 up_read(&nvmf_transports_rwsem);
1064 out_free_opts:
1065 nvmf_free_options(opts);
1066 return ERR_PTR(ret);
1067 }
1068
1069 static struct class *nvmf_class;
1070 static struct device *nvmf_device;
1071 static DEFINE_MUTEX(nvmf_dev_mutex);
1072
nvmf_dev_write(struct file * file,const char __user * ubuf,size_t count,loff_t * pos)1073 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1074 size_t count, loff_t *pos)
1075 {
1076 struct seq_file *seq_file = file->private_data;
1077 struct nvme_ctrl *ctrl;
1078 const char *buf;
1079 int ret = 0;
1080
1081 if (count > PAGE_SIZE)
1082 return -ENOMEM;
1083
1084 buf = memdup_user_nul(ubuf, count);
1085 if (IS_ERR(buf))
1086 return PTR_ERR(buf);
1087
1088 mutex_lock(&nvmf_dev_mutex);
1089 if (seq_file->private) {
1090 ret = -EINVAL;
1091 goto out_unlock;
1092 }
1093
1094 ctrl = nvmf_create_ctrl(nvmf_device, buf);
1095 if (IS_ERR(ctrl)) {
1096 ret = PTR_ERR(ctrl);
1097 goto out_unlock;
1098 }
1099
1100 seq_file->private = ctrl;
1101
1102 out_unlock:
1103 mutex_unlock(&nvmf_dev_mutex);
1104 kfree(buf);
1105 return ret ? ret : count;
1106 }
1107
nvmf_dev_show(struct seq_file * seq_file,void * private)1108 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1109 {
1110 struct nvme_ctrl *ctrl;
1111 int ret = 0;
1112
1113 mutex_lock(&nvmf_dev_mutex);
1114 ctrl = seq_file->private;
1115 if (!ctrl) {
1116 ret = -EINVAL;
1117 goto out_unlock;
1118 }
1119
1120 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1121 ctrl->instance, ctrl->cntlid);
1122
1123 out_unlock:
1124 mutex_unlock(&nvmf_dev_mutex);
1125 return ret;
1126 }
1127
nvmf_dev_open(struct inode * inode,struct file * file)1128 static int nvmf_dev_open(struct inode *inode, struct file *file)
1129 {
1130 /*
1131 * The miscdevice code initializes file->private_data, but doesn't
1132 * make use of it later.
1133 */
1134 file->private_data = NULL;
1135 return single_open(file, nvmf_dev_show, NULL);
1136 }
1137
nvmf_dev_release(struct inode * inode,struct file * file)1138 static int nvmf_dev_release(struct inode *inode, struct file *file)
1139 {
1140 struct seq_file *seq_file = file->private_data;
1141 struct nvme_ctrl *ctrl = seq_file->private;
1142
1143 if (ctrl)
1144 nvme_put_ctrl(ctrl);
1145 return single_release(inode, file);
1146 }
1147
1148 static const struct file_operations nvmf_dev_fops = {
1149 .owner = THIS_MODULE,
1150 .write = nvmf_dev_write,
1151 .read = seq_read,
1152 .open = nvmf_dev_open,
1153 .release = nvmf_dev_release,
1154 };
1155
1156 static struct miscdevice nvmf_misc = {
1157 .minor = MISC_DYNAMIC_MINOR,
1158 .name = "nvme-fabrics",
1159 .fops = &nvmf_dev_fops,
1160 };
1161
nvmf_init(void)1162 static int __init nvmf_init(void)
1163 {
1164 int ret;
1165
1166 nvmf_default_host = nvmf_host_default();
1167 if (!nvmf_default_host)
1168 return -ENOMEM;
1169
1170 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1171 if (IS_ERR(nvmf_class)) {
1172 pr_err("couldn't register class nvme-fabrics\n");
1173 ret = PTR_ERR(nvmf_class);
1174 goto out_free_host;
1175 }
1176
1177 nvmf_device =
1178 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1179 if (IS_ERR(nvmf_device)) {
1180 pr_err("couldn't create nvme-fabris device!\n");
1181 ret = PTR_ERR(nvmf_device);
1182 goto out_destroy_class;
1183 }
1184
1185 ret = misc_register(&nvmf_misc);
1186 if (ret) {
1187 pr_err("couldn't register misc device: %d\n", ret);
1188 goto out_destroy_device;
1189 }
1190
1191 return 0;
1192
1193 out_destroy_device:
1194 device_destroy(nvmf_class, MKDEV(0, 0));
1195 out_destroy_class:
1196 class_destroy(nvmf_class);
1197 out_free_host:
1198 nvmf_host_put(nvmf_default_host);
1199 return ret;
1200 }
1201
nvmf_exit(void)1202 static void __exit nvmf_exit(void)
1203 {
1204 misc_deregister(&nvmf_misc);
1205 device_destroy(nvmf_class, MKDEV(0, 0));
1206 class_destroy(nvmf_class);
1207 nvmf_host_put(nvmf_default_host);
1208
1209 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1210 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1211 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1212 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1213 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1214 }
1215
1216 MODULE_LICENSE("GPL v2");
1217
1218 module_init(nvmf_init);
1219 module_exit(nvmf_exit);
1220