1 /*
2 * NVMe over Fabrics common host code.
3 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 */
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/init.h>
16 #include <linux/miscdevice.h>
17 #include <linux/module.h>
18 #include <linux/mutex.h>
19 #include <linux/parser.h>
20 #include <linux/seq_file.h>
21 #include "nvme.h"
22 #include "fabrics.h"
23
24 static LIST_HEAD(nvmf_transports);
25 static DECLARE_RWSEM(nvmf_transports_rwsem);
26
27 static LIST_HEAD(nvmf_hosts);
28 static DEFINE_MUTEX(nvmf_hosts_mutex);
29
30 static struct nvmf_host *nvmf_default_host;
31
__nvmf_host_find(const char * hostnqn)32 static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
33 {
34 struct nvmf_host *host;
35
36 list_for_each_entry(host, &nvmf_hosts, list) {
37 if (!strcmp(host->nqn, hostnqn))
38 return host;
39 }
40
41 return NULL;
42 }
43
nvmf_host_add(const char * hostnqn)44 static struct nvmf_host *nvmf_host_add(const char *hostnqn)
45 {
46 struct nvmf_host *host;
47
48 mutex_lock(&nvmf_hosts_mutex);
49 host = __nvmf_host_find(hostnqn);
50 if (host) {
51 kref_get(&host->ref);
52 goto out_unlock;
53 }
54
55 host = kmalloc(sizeof(*host), GFP_KERNEL);
56 if (!host)
57 goto out_unlock;
58
59 kref_init(&host->ref);
60 strlcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
61
62 list_add_tail(&host->list, &nvmf_hosts);
63 out_unlock:
64 mutex_unlock(&nvmf_hosts_mutex);
65 return host;
66 }
67
nvmf_host_default(void)68 static struct nvmf_host *nvmf_host_default(void)
69 {
70 struct nvmf_host *host;
71
72 host = kmalloc(sizeof(*host), GFP_KERNEL);
73 if (!host)
74 return NULL;
75
76 kref_init(&host->ref);
77 uuid_gen(&host->id);
78 snprintf(host->nqn, NVMF_NQN_SIZE,
79 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id);
80
81 mutex_lock(&nvmf_hosts_mutex);
82 list_add_tail(&host->list, &nvmf_hosts);
83 mutex_unlock(&nvmf_hosts_mutex);
84
85 return host;
86 }
87
nvmf_host_destroy(struct kref * ref)88 static void nvmf_host_destroy(struct kref *ref)
89 {
90 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
91
92 mutex_lock(&nvmf_hosts_mutex);
93 list_del(&host->list);
94 mutex_unlock(&nvmf_hosts_mutex);
95
96 kfree(host);
97 }
98
nvmf_host_put(struct nvmf_host * host)99 static void nvmf_host_put(struct nvmf_host *host)
100 {
101 if (host)
102 kref_put(&host->ref, nvmf_host_destroy);
103 }
104
105 /**
106 * nvmf_get_address() - Get address/port
107 * @ctrl: Host NVMe controller instance which we got the address
108 * @buf: OUTPUT parameter that will contain the address/port
109 * @size: buffer size
110 */
nvmf_get_address(struct nvme_ctrl * ctrl,char * buf,int size)111 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
112 {
113 int len = 0;
114
115 if (ctrl->opts->mask & NVMF_OPT_TRADDR)
116 len += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
117 if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
118 len += snprintf(buf + len, size - len, "%strsvcid=%s",
119 (len) ? "," : "", ctrl->opts->trsvcid);
120 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
121 len += snprintf(buf + len, size - len, "%shost_traddr=%s",
122 (len) ? "," : "", ctrl->opts->host_traddr);
123 len += snprintf(buf + len, size - len, "\n");
124
125 return len;
126 }
127 EXPORT_SYMBOL_GPL(nvmf_get_address);
128
129 /**
130 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function.
131 * @ctrl: Host NVMe controller instance maintaining the admin
132 * queue used to submit the property read command to
133 * the allocated NVMe controller resource on the target system.
134 * @off: Starting offset value of the targeted property
135 * register (see the fabrics section of the NVMe standard).
136 * @val: OUTPUT parameter that will contain the value of
137 * the property after a successful read.
138 *
139 * Used by the host system to retrieve a 32-bit capsule property value
140 * from an NVMe controller on the target system.
141 *
142 * ("Capsule property" is an "PCIe register concept" applied to the
143 * NVMe fabrics space.)
144 *
145 * Return:
146 * 0: successful read
147 * > 0: NVMe error status code
148 * < 0: Linux errno error code
149 */
nvmf_reg_read32(struct nvme_ctrl * ctrl,u32 off,u32 * val)150 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
151 {
152 struct nvme_command cmd;
153 union nvme_result res;
154 int ret;
155
156 memset(&cmd, 0, sizeof(cmd));
157 cmd.prop_get.opcode = nvme_fabrics_command;
158 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
159 cmd.prop_get.offset = cpu_to_le32(off);
160
161 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
162 NVME_QID_ANY, 0, 0);
163
164 if (ret >= 0)
165 *val = le64_to_cpu(res.u64);
166 if (unlikely(ret != 0))
167 dev_err(ctrl->device,
168 "Property Get error: %d, offset %#x\n",
169 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
170
171 return ret;
172 }
173 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
174
175 /**
176 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function.
177 * @ctrl: Host NVMe controller instance maintaining the admin
178 * queue used to submit the property read command to
179 * the allocated controller resource on the target system.
180 * @off: Starting offset value of the targeted property
181 * register (see the fabrics section of the NVMe standard).
182 * @val: OUTPUT parameter that will contain the value of
183 * the property after a successful read.
184 *
185 * Used by the host system to retrieve a 64-bit capsule property value
186 * from an NVMe controller on the target system.
187 *
188 * ("Capsule property" is an "PCIe register concept" applied to the
189 * NVMe fabrics space.)
190 *
191 * Return:
192 * 0: successful read
193 * > 0: NVMe error status code
194 * < 0: Linux errno error code
195 */
nvmf_reg_read64(struct nvme_ctrl * ctrl,u32 off,u64 * val)196 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
197 {
198 struct nvme_command cmd;
199 union nvme_result res;
200 int ret;
201
202 memset(&cmd, 0, sizeof(cmd));
203 cmd.prop_get.opcode = nvme_fabrics_command;
204 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
205 cmd.prop_get.attrib = 1;
206 cmd.prop_get.offset = cpu_to_le32(off);
207
208 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
209 NVME_QID_ANY, 0, 0);
210
211 if (ret >= 0)
212 *val = le64_to_cpu(res.u64);
213 if (unlikely(ret != 0))
214 dev_err(ctrl->device,
215 "Property Get error: %d, offset %#x\n",
216 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
217 return ret;
218 }
219 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
220
221 /**
222 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function.
223 * @ctrl: Host NVMe controller instance maintaining the admin
224 * queue used to submit the property read command to
225 * the allocated NVMe controller resource on the target system.
226 * @off: Starting offset value of the targeted property
227 * register (see the fabrics section of the NVMe standard).
228 * @val: Input parameter that contains the value to be
229 * written to the property.
230 *
231 * Used by the NVMe host system to write a 32-bit capsule property value
232 * to an NVMe controller on the target system.
233 *
234 * ("Capsule property" is an "PCIe register concept" applied to the
235 * NVMe fabrics space.)
236 *
237 * Return:
238 * 0: successful write
239 * > 0: NVMe error status code
240 * < 0: Linux errno error code
241 */
nvmf_reg_write32(struct nvme_ctrl * ctrl,u32 off,u32 val)242 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
243 {
244 struct nvme_command cmd;
245 int ret;
246
247 memset(&cmd, 0, sizeof(cmd));
248 cmd.prop_set.opcode = nvme_fabrics_command;
249 cmd.prop_set.fctype = nvme_fabrics_type_property_set;
250 cmd.prop_set.attrib = 0;
251 cmd.prop_set.offset = cpu_to_le32(off);
252 cmd.prop_set.value = cpu_to_le64(val);
253
254 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
255 NVME_QID_ANY, 0, 0);
256 if (unlikely(ret))
257 dev_err(ctrl->device,
258 "Property Set error: %d, offset %#x\n",
259 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
260 return ret;
261 }
262 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
263
264 /**
265 * nvmf_log_connect_error() - Error-parsing-diagnostic print
266 * out function for connect() errors.
267 *
268 * @ctrl: the specific /dev/nvmeX device that had the error.
269 *
270 * @errval: Error code to be decoded in a more human-friendly
271 * printout.
272 *
273 * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
274 *
275 * @cmd: This is the SQE portion of a submission capsule.
276 *
277 * @data: This is the "Data" portion of a submission capsule.
278 */
nvmf_log_connect_error(struct nvme_ctrl * ctrl,int errval,int offset,struct nvme_command * cmd,struct nvmf_connect_data * data)279 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
280 int errval, int offset, struct nvme_command *cmd,
281 struct nvmf_connect_data *data)
282 {
283 int err_sctype = errval & (~NVME_SC_DNR);
284
285 switch (err_sctype) {
286
287 case (NVME_SC_CONNECT_INVALID_PARAM):
288 if (offset >> 16) {
289 char *inv_data = "Connect Invalid Data Parameter";
290
291 switch (offset & 0xffff) {
292 case (offsetof(struct nvmf_connect_data, cntlid)):
293 dev_err(ctrl->device,
294 "%s, cntlid: %d\n",
295 inv_data, data->cntlid);
296 break;
297 case (offsetof(struct nvmf_connect_data, hostnqn)):
298 dev_err(ctrl->device,
299 "%s, hostnqn \"%s\"\n",
300 inv_data, data->hostnqn);
301 break;
302 case (offsetof(struct nvmf_connect_data, subsysnqn)):
303 dev_err(ctrl->device,
304 "%s, subsysnqn \"%s\"\n",
305 inv_data, data->subsysnqn);
306 break;
307 default:
308 dev_err(ctrl->device,
309 "%s, starting byte offset: %d\n",
310 inv_data, offset & 0xffff);
311 break;
312 }
313 } else {
314 char *inv_sqe = "Connect Invalid SQE Parameter";
315
316 switch (offset) {
317 case (offsetof(struct nvmf_connect_command, qid)):
318 dev_err(ctrl->device,
319 "%s, qid %d\n",
320 inv_sqe, cmd->connect.qid);
321 break;
322 default:
323 dev_err(ctrl->device,
324 "%s, starting byte offset: %d\n",
325 inv_sqe, offset);
326 }
327 }
328 break;
329
330 case NVME_SC_CONNECT_INVALID_HOST:
331 dev_err(ctrl->device,
332 "Connect for subsystem %s is not allowed, hostnqn: %s\n",
333 data->subsysnqn, data->hostnqn);
334 break;
335
336 case NVME_SC_CONNECT_CTRL_BUSY:
337 dev_err(ctrl->device,
338 "Connect command failed: controller is busy or not available\n");
339 break;
340
341 case NVME_SC_CONNECT_FORMAT:
342 dev_err(ctrl->device,
343 "Connect incompatible format: %d",
344 cmd->connect.recfmt);
345 break;
346
347 default:
348 dev_err(ctrl->device,
349 "Connect command failed, error wo/DNR bit: %d\n",
350 err_sctype);
351 break;
352 } /* switch (err_sctype) */
353 }
354
355 /**
356 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
357 * API function.
358 * @ctrl: Host nvme controller instance used to request
359 * a new NVMe controller allocation on the target
360 * system and establish an NVMe Admin connection to
361 * that controller.
362 *
363 * This function enables an NVMe host device to request a new allocation of
364 * an NVMe controller resource on a target system as well establish a
365 * fabrics-protocol connection of the NVMe Admin queue between the
366 * host system device and the allocated NVMe controller on the
367 * target system via a NVMe Fabrics "Connect" command.
368 *
369 * Return:
370 * 0: success
371 * > 0: NVMe error status code
372 * < 0: Linux errno error code
373 *
374 */
nvmf_connect_admin_queue(struct nvme_ctrl * ctrl)375 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
376 {
377 struct nvme_command cmd;
378 union nvme_result res;
379 struct nvmf_connect_data *data;
380 int ret;
381
382 memset(&cmd, 0, sizeof(cmd));
383 cmd.connect.opcode = nvme_fabrics_command;
384 cmd.connect.fctype = nvme_fabrics_type_connect;
385 cmd.connect.qid = 0;
386 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
387
388 /*
389 * Set keep-alive timeout in seconds granularity (ms * 1000)
390 * and add a grace period for controller kato enforcement
391 */
392 cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
393 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
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->admin_q, &cmd, &res,
405 data, sizeof(*data), 0, NVME_QID_ANY, 1,
406 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
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 *
431 * This function issues a fabrics-protocol connection
432 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
433 * between the host system device and the allocated NVMe controller
434 * on the target system.
435 *
436 * Return:
437 * 0: success
438 * > 0: NVMe error status code
439 * < 0: Linux errno error code
440 */
nvmf_connect_io_queue(struct nvme_ctrl * ctrl,u16 qid)441 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
442 {
443 struct nvme_command cmd;
444 struct nvmf_connect_data *data;
445 union nvme_result res;
446 int ret;
447
448 memset(&cmd, 0, sizeof(cmd));
449 cmd.connect.opcode = nvme_fabrics_command;
450 cmd.connect.fctype = nvme_fabrics_type_connect;
451 cmd.connect.qid = cpu_to_le16(qid);
452 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
453
454 data = kzalloc(sizeof(*data), GFP_KERNEL);
455 if (!data)
456 return -ENOMEM;
457
458 uuid_copy(&data->hostid, &ctrl->opts->host->id);
459 data->cntlid = cpu_to_le16(ctrl->cntlid);
460 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
461 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
462
463 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
464 data, sizeof(*data), 0, qid, 1,
465 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
466 if (ret) {
467 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
468 &cmd, data);
469 }
470 kfree(data);
471 return ret;
472 }
473 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
474
nvmf_should_reconnect(struct nvme_ctrl * ctrl)475 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
476 {
477 if (ctrl->opts->max_reconnects == -1 ||
478 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
479 return true;
480
481 return false;
482 }
483 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
484
485 /**
486 * nvmf_register_transport() - NVMe Fabrics Library registration function.
487 * @ops: Transport ops instance to be registered to the
488 * common fabrics library.
489 *
490 * API function that registers the type of specific transport fabric
491 * being implemented to the common NVMe fabrics library. Part of
492 * the overall init sequence of starting up a fabrics driver.
493 */
nvmf_register_transport(struct nvmf_transport_ops * ops)494 int nvmf_register_transport(struct nvmf_transport_ops *ops)
495 {
496 if (!ops->create_ctrl)
497 return -EINVAL;
498
499 down_write(&nvmf_transports_rwsem);
500 list_add_tail(&ops->entry, &nvmf_transports);
501 up_write(&nvmf_transports_rwsem);
502
503 return 0;
504 }
505 EXPORT_SYMBOL_GPL(nvmf_register_transport);
506
507 /**
508 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
509 * @ops: Transport ops instance to be unregistered from the
510 * common fabrics library.
511 *
512 * Fabrics API function that unregisters the type of specific transport
513 * fabric being implemented from the common NVMe fabrics library.
514 * Part of the overall exit sequence of unloading the implemented driver.
515 */
nvmf_unregister_transport(struct nvmf_transport_ops * ops)516 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
517 {
518 down_write(&nvmf_transports_rwsem);
519 list_del(&ops->entry);
520 up_write(&nvmf_transports_rwsem);
521 }
522 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
523
nvmf_lookup_transport(struct nvmf_ctrl_options * opts)524 static struct nvmf_transport_ops *nvmf_lookup_transport(
525 struct nvmf_ctrl_options *opts)
526 {
527 struct nvmf_transport_ops *ops;
528
529 lockdep_assert_held(&nvmf_transports_rwsem);
530
531 list_for_each_entry(ops, &nvmf_transports, entry) {
532 if (strcmp(ops->name, opts->transport) == 0)
533 return ops;
534 }
535
536 return NULL;
537 }
538
539 /*
540 * For something we're not in a state to send to the device the default action
541 * is to busy it and retry it after the controller state is recovered. However,
542 * if the controller is deleting or if anything is marked for failfast or
543 * nvme multipath it is immediately failed.
544 *
545 * Note: commands used to initialize the controller will be marked for failfast.
546 * Note: nvme cli/ioctl commands are marked for failfast.
547 */
nvmf_fail_nonready_command(struct nvme_ctrl * ctrl,struct request * rq)548 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
549 struct request *rq)
550 {
551 if (ctrl->state != NVME_CTRL_DELETING &&
552 ctrl->state != NVME_CTRL_DEAD &&
553 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
554 return BLK_STS_RESOURCE;
555 nvme_req(rq)->status = NVME_SC_ABORT_REQ;
556 return BLK_STS_IOERR;
557 }
558 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
559
__nvmf_check_ready(struct nvme_ctrl * ctrl,struct request * rq,bool queue_live)560 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
561 bool queue_live)
562 {
563 struct nvme_request *req = nvme_req(rq);
564
565 /*
566 * If we are in some state of setup or teardown only allow
567 * internally generated commands.
568 */
569 if (!blk_rq_is_passthrough(rq) || (req->flags & NVME_REQ_USERCMD))
570 return false;
571
572 /*
573 * Only allow commands on a live queue, except for the connect command,
574 * which is require to set the queue live in the appropinquate states.
575 */
576 switch (ctrl->state) {
577 case NVME_CTRL_NEW:
578 case NVME_CTRL_CONNECTING:
579 if (req->cmd->common.opcode == nvme_fabrics_command &&
580 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
581 return true;
582 break;
583 default:
584 break;
585 case NVME_CTRL_DEAD:
586 return false;
587 }
588
589 return queue_live;
590 }
591 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
592
593 static const match_table_t opt_tokens = {
594 { NVMF_OPT_TRANSPORT, "transport=%s" },
595 { NVMF_OPT_TRADDR, "traddr=%s" },
596 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
597 { NVMF_OPT_NQN, "nqn=%s" },
598 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
599 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
600 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
601 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
602 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
603 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
604 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
605 { NVMF_OPT_HOST_ID, "hostid=%s" },
606 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
607 { NVMF_OPT_ERR, NULL }
608 };
609
nvmf_parse_options(struct nvmf_ctrl_options * opts,const char * buf)610 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
611 const char *buf)
612 {
613 substring_t args[MAX_OPT_ARGS];
614 char *options, *o, *p;
615 int token, ret = 0;
616 size_t nqnlen = 0;
617 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
618 uuid_t hostid;
619
620 /* Set defaults */
621 opts->queue_size = NVMF_DEF_QUEUE_SIZE;
622 opts->nr_io_queues = num_online_cpus();
623 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
624 opts->kato = NVME_DEFAULT_KATO;
625 opts->duplicate_connect = false;
626
627 options = o = kstrdup(buf, GFP_KERNEL);
628 if (!options)
629 return -ENOMEM;
630
631 uuid_gen(&hostid);
632
633 while ((p = strsep(&o, ",\n")) != NULL) {
634 if (!*p)
635 continue;
636
637 token = match_token(p, opt_tokens, args);
638 opts->mask |= token;
639 switch (token) {
640 case NVMF_OPT_TRANSPORT:
641 p = match_strdup(args);
642 if (!p) {
643 ret = -ENOMEM;
644 goto out;
645 }
646 kfree(opts->transport);
647 opts->transport = p;
648 break;
649 case NVMF_OPT_NQN:
650 p = match_strdup(args);
651 if (!p) {
652 ret = -ENOMEM;
653 goto out;
654 }
655 kfree(opts->subsysnqn);
656 opts->subsysnqn = p;
657 nqnlen = strlen(opts->subsysnqn);
658 if (nqnlen >= NVMF_NQN_SIZE) {
659 pr_err("%s needs to be < %d bytes\n",
660 opts->subsysnqn, NVMF_NQN_SIZE);
661 ret = -EINVAL;
662 goto out;
663 }
664 opts->discovery_nqn =
665 !(strcmp(opts->subsysnqn,
666 NVME_DISC_SUBSYS_NAME));
667 break;
668 case NVMF_OPT_TRADDR:
669 p = match_strdup(args);
670 if (!p) {
671 ret = -ENOMEM;
672 goto out;
673 }
674 kfree(opts->traddr);
675 opts->traddr = p;
676 break;
677 case NVMF_OPT_TRSVCID:
678 p = match_strdup(args);
679 if (!p) {
680 ret = -ENOMEM;
681 goto out;
682 }
683 kfree(opts->trsvcid);
684 opts->trsvcid = p;
685 break;
686 case NVMF_OPT_QUEUE_SIZE:
687 if (match_int(args, &token)) {
688 ret = -EINVAL;
689 goto out;
690 }
691 if (token < NVMF_MIN_QUEUE_SIZE ||
692 token > NVMF_MAX_QUEUE_SIZE) {
693 pr_err("Invalid queue_size %d\n", token);
694 ret = -EINVAL;
695 goto out;
696 }
697 opts->queue_size = token;
698 break;
699 case NVMF_OPT_NR_IO_QUEUES:
700 if (match_int(args, &token)) {
701 ret = -EINVAL;
702 goto out;
703 }
704 if (token <= 0) {
705 pr_err("Invalid number of IOQs %d\n", token);
706 ret = -EINVAL;
707 goto out;
708 }
709 if (opts->discovery_nqn) {
710 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
711 break;
712 }
713
714 opts->nr_io_queues = min_t(unsigned int,
715 num_online_cpus(), token);
716 break;
717 case NVMF_OPT_KATO:
718 if (match_int(args, &token)) {
719 ret = -EINVAL;
720 goto out;
721 }
722
723 if (token < 0) {
724 pr_err("Invalid keep_alive_tmo %d\n", token);
725 ret = -EINVAL;
726 goto out;
727 } else if (token == 0 && !opts->discovery_nqn) {
728 /* Allowed for debug */
729 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
730 }
731 opts->kato = token;
732
733 if (opts->discovery_nqn && opts->kato) {
734 pr_err("Discovery controllers cannot accept KATO != 0\n");
735 ret = -EINVAL;
736 goto out;
737 }
738
739 break;
740 case NVMF_OPT_CTRL_LOSS_TMO:
741 if (match_int(args, &token)) {
742 ret = -EINVAL;
743 goto out;
744 }
745
746 if (token < 0)
747 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
748 ctrl_loss_tmo = token;
749 break;
750 case NVMF_OPT_HOSTNQN:
751 if (opts->host) {
752 pr_err("hostnqn already user-assigned: %s\n",
753 opts->host->nqn);
754 ret = -EADDRINUSE;
755 goto out;
756 }
757 p = match_strdup(args);
758 if (!p) {
759 ret = -ENOMEM;
760 goto out;
761 }
762 nqnlen = strlen(p);
763 if (nqnlen >= NVMF_NQN_SIZE) {
764 pr_err("%s needs to be < %d bytes\n",
765 p, NVMF_NQN_SIZE);
766 kfree(p);
767 ret = -EINVAL;
768 goto out;
769 }
770 nvmf_host_put(opts->host);
771 opts->host = nvmf_host_add(p);
772 kfree(p);
773 if (!opts->host) {
774 ret = -ENOMEM;
775 goto out;
776 }
777 break;
778 case NVMF_OPT_RECONNECT_DELAY:
779 if (match_int(args, &token)) {
780 ret = -EINVAL;
781 goto out;
782 }
783 if (token <= 0) {
784 pr_err("Invalid reconnect_delay %d\n", token);
785 ret = -EINVAL;
786 goto out;
787 }
788 opts->reconnect_delay = token;
789 break;
790 case NVMF_OPT_HOST_TRADDR:
791 p = match_strdup(args);
792 if (!p) {
793 ret = -ENOMEM;
794 goto out;
795 }
796 kfree(opts->host_traddr);
797 opts->host_traddr = p;
798 break;
799 case NVMF_OPT_HOST_ID:
800 p = match_strdup(args);
801 if (!p) {
802 ret = -ENOMEM;
803 goto out;
804 }
805 ret = uuid_parse(p, &hostid);
806 if (ret) {
807 pr_err("Invalid hostid %s\n", p);
808 ret = -EINVAL;
809 kfree(p);
810 goto out;
811 }
812 kfree(p);
813 break;
814 case NVMF_OPT_DUP_CONNECT:
815 opts->duplicate_connect = true;
816 break;
817 default:
818 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
819 p);
820 ret = -EINVAL;
821 goto out;
822 }
823 }
824
825 if (opts->discovery_nqn) {
826 opts->kato = 0;
827 opts->nr_io_queues = 0;
828 opts->duplicate_connect = true;
829 }
830 if (ctrl_loss_tmo < 0)
831 opts->max_reconnects = -1;
832 else
833 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
834 opts->reconnect_delay);
835
836 if (!opts->host) {
837 kref_get(&nvmf_default_host->ref);
838 opts->host = nvmf_default_host;
839 }
840
841 uuid_copy(&opts->host->id, &hostid);
842
843 out:
844 kfree(options);
845 return ret;
846 }
847
nvmf_check_required_opts(struct nvmf_ctrl_options * opts,unsigned int required_opts)848 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
849 unsigned int required_opts)
850 {
851 if ((opts->mask & required_opts) != required_opts) {
852 int i;
853
854 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
855 if ((opt_tokens[i].token & required_opts) &&
856 !(opt_tokens[i].token & opts->mask)) {
857 pr_warn("missing parameter '%s'\n",
858 opt_tokens[i].pattern);
859 }
860 }
861
862 return -EINVAL;
863 }
864
865 return 0;
866 }
867
nvmf_check_allowed_opts(struct nvmf_ctrl_options * opts,unsigned int allowed_opts)868 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
869 unsigned int allowed_opts)
870 {
871 if (opts->mask & ~allowed_opts) {
872 int i;
873
874 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
875 if ((opt_tokens[i].token & opts->mask) &&
876 (opt_tokens[i].token & ~allowed_opts)) {
877 pr_warn("invalid parameter '%s'\n",
878 opt_tokens[i].pattern);
879 }
880 }
881
882 return -EINVAL;
883 }
884
885 return 0;
886 }
887
nvmf_free_options(struct nvmf_ctrl_options * opts)888 void nvmf_free_options(struct nvmf_ctrl_options *opts)
889 {
890 nvmf_host_put(opts->host);
891 kfree(opts->transport);
892 kfree(opts->traddr);
893 kfree(opts->trsvcid);
894 kfree(opts->subsysnqn);
895 kfree(opts->host_traddr);
896 kfree(opts);
897 }
898 EXPORT_SYMBOL_GPL(nvmf_free_options);
899
900 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
901 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
902 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
903 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT)
904
905 static struct nvme_ctrl *
nvmf_create_ctrl(struct device * dev,const char * buf,size_t count)906 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
907 {
908 struct nvmf_ctrl_options *opts;
909 struct nvmf_transport_ops *ops;
910 struct nvme_ctrl *ctrl;
911 int ret;
912
913 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
914 if (!opts)
915 return ERR_PTR(-ENOMEM);
916
917 ret = nvmf_parse_options(opts, buf);
918 if (ret)
919 goto out_free_opts;
920
921
922 request_module("nvme-%s", opts->transport);
923
924 /*
925 * Check the generic options first as we need a valid transport for
926 * the lookup below. Then clear the generic flags so that transport
927 * drivers don't have to care about them.
928 */
929 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
930 if (ret)
931 goto out_free_opts;
932 opts->mask &= ~NVMF_REQUIRED_OPTS;
933
934 down_read(&nvmf_transports_rwsem);
935 ops = nvmf_lookup_transport(opts);
936 if (!ops) {
937 pr_info("no handler found for transport %s.\n",
938 opts->transport);
939 ret = -EINVAL;
940 goto out_unlock;
941 }
942
943 if (!try_module_get(ops->module)) {
944 ret = -EBUSY;
945 goto out_unlock;
946 }
947 up_read(&nvmf_transports_rwsem);
948
949 ret = nvmf_check_required_opts(opts, ops->required_opts);
950 if (ret)
951 goto out_module_put;
952 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
953 ops->allowed_opts | ops->required_opts);
954 if (ret)
955 goto out_module_put;
956
957 ctrl = ops->create_ctrl(dev, opts);
958 if (IS_ERR(ctrl)) {
959 ret = PTR_ERR(ctrl);
960 goto out_module_put;
961 }
962
963 module_put(ops->module);
964 return ctrl;
965
966 out_module_put:
967 module_put(ops->module);
968 goto out_free_opts;
969 out_unlock:
970 up_read(&nvmf_transports_rwsem);
971 out_free_opts:
972 nvmf_free_options(opts);
973 return ERR_PTR(ret);
974 }
975
976 static struct class *nvmf_class;
977 static struct device *nvmf_device;
978 static DEFINE_MUTEX(nvmf_dev_mutex);
979
nvmf_dev_write(struct file * file,const char __user * ubuf,size_t count,loff_t * pos)980 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
981 size_t count, loff_t *pos)
982 {
983 struct seq_file *seq_file = file->private_data;
984 struct nvme_ctrl *ctrl;
985 const char *buf;
986 int ret = 0;
987
988 if (count > PAGE_SIZE)
989 return -ENOMEM;
990
991 buf = memdup_user_nul(ubuf, count);
992 if (IS_ERR(buf))
993 return PTR_ERR(buf);
994
995 mutex_lock(&nvmf_dev_mutex);
996 if (seq_file->private) {
997 ret = -EINVAL;
998 goto out_unlock;
999 }
1000
1001 ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
1002 if (IS_ERR(ctrl)) {
1003 ret = PTR_ERR(ctrl);
1004 goto out_unlock;
1005 }
1006
1007 seq_file->private = ctrl;
1008
1009 out_unlock:
1010 mutex_unlock(&nvmf_dev_mutex);
1011 kfree(buf);
1012 return ret ? ret : count;
1013 }
1014
nvmf_dev_show(struct seq_file * seq_file,void * private)1015 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1016 {
1017 struct nvme_ctrl *ctrl;
1018 int ret = 0;
1019
1020 mutex_lock(&nvmf_dev_mutex);
1021 ctrl = seq_file->private;
1022 if (!ctrl) {
1023 ret = -EINVAL;
1024 goto out_unlock;
1025 }
1026
1027 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1028 ctrl->instance, ctrl->cntlid);
1029
1030 out_unlock:
1031 mutex_unlock(&nvmf_dev_mutex);
1032 return ret;
1033 }
1034
nvmf_dev_open(struct inode * inode,struct file * file)1035 static int nvmf_dev_open(struct inode *inode, struct file *file)
1036 {
1037 /*
1038 * The miscdevice code initializes file->private_data, but doesn't
1039 * make use of it later.
1040 */
1041 file->private_data = NULL;
1042 return single_open(file, nvmf_dev_show, NULL);
1043 }
1044
nvmf_dev_release(struct inode * inode,struct file * file)1045 static int nvmf_dev_release(struct inode *inode, struct file *file)
1046 {
1047 struct seq_file *seq_file = file->private_data;
1048 struct nvme_ctrl *ctrl = seq_file->private;
1049
1050 if (ctrl)
1051 nvme_put_ctrl(ctrl);
1052 return single_release(inode, file);
1053 }
1054
1055 static const struct file_operations nvmf_dev_fops = {
1056 .owner = THIS_MODULE,
1057 .write = nvmf_dev_write,
1058 .read = seq_read,
1059 .open = nvmf_dev_open,
1060 .release = nvmf_dev_release,
1061 };
1062
1063 static struct miscdevice nvmf_misc = {
1064 .minor = MISC_DYNAMIC_MINOR,
1065 .name = "nvme-fabrics",
1066 .fops = &nvmf_dev_fops,
1067 };
1068
nvmf_init(void)1069 static int __init nvmf_init(void)
1070 {
1071 int ret;
1072
1073 nvmf_default_host = nvmf_host_default();
1074 if (!nvmf_default_host)
1075 return -ENOMEM;
1076
1077 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1078 if (IS_ERR(nvmf_class)) {
1079 pr_err("couldn't register class nvme-fabrics\n");
1080 ret = PTR_ERR(nvmf_class);
1081 goto out_free_host;
1082 }
1083
1084 nvmf_device =
1085 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1086 if (IS_ERR(nvmf_device)) {
1087 pr_err("couldn't create nvme-fabris device!\n");
1088 ret = PTR_ERR(nvmf_device);
1089 goto out_destroy_class;
1090 }
1091
1092 ret = misc_register(&nvmf_misc);
1093 if (ret) {
1094 pr_err("couldn't register misc device: %d\n", ret);
1095 goto out_destroy_device;
1096 }
1097
1098 return 0;
1099
1100 out_destroy_device:
1101 device_destroy(nvmf_class, MKDEV(0, 0));
1102 out_destroy_class:
1103 class_destroy(nvmf_class);
1104 out_free_host:
1105 nvmf_host_put(nvmf_default_host);
1106 return ret;
1107 }
1108
nvmf_exit(void)1109 static void __exit nvmf_exit(void)
1110 {
1111 misc_deregister(&nvmf_misc);
1112 device_destroy(nvmf_class, MKDEV(0, 0));
1113 class_destroy(nvmf_class);
1114 nvmf_host_put(nvmf_default_host);
1115
1116 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1117 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1118 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1119 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1120 }
1121
1122 MODULE_LICENSE("GPL v2");
1123
1124 module_init(nvmf_init);
1125 module_exit(nvmf_exit);
1126
