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 if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)
116 len += scnprintf(buf + len, size - len, "%shost_iface=%s",
117 (len) ? "," : "", ctrl->opts->host_iface);
118 len += scnprintf(buf + len, size - len, "\n");
119
120 return len;
121 }
122 EXPORT_SYMBOL_GPL(nvmf_get_address);
123
124 /**
125 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function.
126 * @ctrl: Host NVMe controller instance maintaining the admin
127 * queue used to submit the property read command to
128 * the allocated NVMe controller resource on the target system.
129 * @off: Starting offset value of the targeted property
130 * register (see the fabrics section of the NVMe standard).
131 * @val: OUTPUT parameter that will contain the value of
132 * the property after a successful read.
133 *
134 * Used by the host system to retrieve a 32-bit capsule property value
135 * from an NVMe controller on the target system.
136 *
137 * ("Capsule property" is an "PCIe register concept" applied to the
138 * NVMe fabrics space.)
139 *
140 * Return:
141 * 0: successful read
142 * > 0: NVMe error status code
143 * < 0: Linux errno error code
144 */
nvmf_reg_read32(struct nvme_ctrl * ctrl,u32 off,u32 * val)145 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
146 {
147 struct nvme_command cmd;
148 union nvme_result res;
149 int ret;
150
151 memset(&cmd, 0, sizeof(cmd));
152 cmd.prop_get.opcode = nvme_fabrics_command;
153 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
154 cmd.prop_get.offset = cpu_to_le32(off);
155
156 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
157 NVME_QID_ANY, 0, 0);
158
159 if (ret >= 0)
160 *val = le64_to_cpu(res.u64);
161 if (unlikely(ret != 0))
162 dev_err(ctrl->device,
163 "Property Get error: %d, offset %#x\n",
164 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
165
166 return ret;
167 }
168 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
169
170 /**
171 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function.
172 * @ctrl: Host NVMe controller instance maintaining the admin
173 * queue used to submit the property read command to
174 * the allocated controller resource on the target system.
175 * @off: Starting offset value of the targeted property
176 * register (see the fabrics section of the NVMe standard).
177 * @val: OUTPUT parameter that will contain the value of
178 * the property after a successful read.
179 *
180 * Used by the host system to retrieve a 64-bit capsule property value
181 * from an NVMe controller on the target system.
182 *
183 * ("Capsule property" is an "PCIe register concept" applied to the
184 * NVMe fabrics space.)
185 *
186 * Return:
187 * 0: successful read
188 * > 0: NVMe error status code
189 * < 0: Linux errno error code
190 */
nvmf_reg_read64(struct nvme_ctrl * ctrl,u32 off,u64 * val)191 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
192 {
193 struct nvme_command cmd = { };
194 union nvme_result res;
195 int ret;
196
197 cmd.prop_get.opcode = nvme_fabrics_command;
198 cmd.prop_get.fctype = nvme_fabrics_type_property_get;
199 cmd.prop_get.attrib = 1;
200 cmd.prop_get.offset = cpu_to_le32(off);
201
202 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
203 NVME_QID_ANY, 0, 0);
204
205 if (ret >= 0)
206 *val = le64_to_cpu(res.u64);
207 if (unlikely(ret != 0))
208 dev_err(ctrl->device,
209 "Property Get error: %d, offset %#x\n",
210 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
211 return ret;
212 }
213 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
214
215 /**
216 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function.
217 * @ctrl: Host NVMe controller instance maintaining the admin
218 * queue used to submit the property read command to
219 * the allocated NVMe controller resource on the target system.
220 * @off: Starting offset value of the targeted property
221 * register (see the fabrics section of the NVMe standard).
222 * @val: Input parameter that contains the value to be
223 * written to the property.
224 *
225 * Used by the NVMe host system to write a 32-bit capsule property value
226 * to an NVMe controller on the target system.
227 *
228 * ("Capsule property" is an "PCIe register concept" applied to the
229 * NVMe fabrics space.)
230 *
231 * Return:
232 * 0: successful write
233 * > 0: NVMe error status code
234 * < 0: Linux errno error code
235 */
nvmf_reg_write32(struct nvme_ctrl * ctrl,u32 off,u32 val)236 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
237 {
238 struct nvme_command cmd = { };
239 int ret;
240
241 cmd.prop_set.opcode = nvme_fabrics_command;
242 cmd.prop_set.fctype = nvme_fabrics_type_property_set;
243 cmd.prop_set.attrib = 0;
244 cmd.prop_set.offset = cpu_to_le32(off);
245 cmd.prop_set.value = cpu_to_le64(val);
246
247 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 0,
248 NVME_QID_ANY, 0, 0);
249 if (unlikely(ret))
250 dev_err(ctrl->device,
251 "Property Set error: %d, offset %#x\n",
252 ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
253 return ret;
254 }
255 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
256
257 /**
258 * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for
259 * connect() errors.
260 * @ctrl: The specific /dev/nvmeX device that had the error.
261 * @errval: Error code to be decoded in a more human-friendly
262 * printout.
263 * @offset: For use with the NVMe error code
264 * NVME_SC_CONNECT_INVALID_PARAM.
265 * @cmd: This is the SQE portion of a submission capsule.
266 * @data: This is the "Data" portion of a submission capsule.
267 */
nvmf_log_connect_error(struct nvme_ctrl * ctrl,int errval,int offset,struct nvme_command * cmd,struct nvmf_connect_data * data)268 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
269 int errval, int offset, struct nvme_command *cmd,
270 struct nvmf_connect_data *data)
271 {
272 int err_sctype = errval & ~NVME_SC_DNR;
273
274 switch (err_sctype) {
275 case (NVME_SC_CONNECT_INVALID_PARAM):
276 if (offset >> 16) {
277 char *inv_data = "Connect Invalid Data Parameter";
278
279 switch (offset & 0xffff) {
280 case (offsetof(struct nvmf_connect_data, cntlid)):
281 dev_err(ctrl->device,
282 "%s, cntlid: %d\n",
283 inv_data, data->cntlid);
284 break;
285 case (offsetof(struct nvmf_connect_data, hostnqn)):
286 dev_err(ctrl->device,
287 "%s, hostnqn \"%s\"\n",
288 inv_data, data->hostnqn);
289 break;
290 case (offsetof(struct nvmf_connect_data, subsysnqn)):
291 dev_err(ctrl->device,
292 "%s, subsysnqn \"%s\"\n",
293 inv_data, data->subsysnqn);
294 break;
295 default:
296 dev_err(ctrl->device,
297 "%s, starting byte offset: %d\n",
298 inv_data, offset & 0xffff);
299 break;
300 }
301 } else {
302 char *inv_sqe = "Connect Invalid SQE Parameter";
303
304 switch (offset) {
305 case (offsetof(struct nvmf_connect_command, qid)):
306 dev_err(ctrl->device,
307 "%s, qid %d\n",
308 inv_sqe, cmd->connect.qid);
309 break;
310 default:
311 dev_err(ctrl->device,
312 "%s, starting byte offset: %d\n",
313 inv_sqe, offset);
314 }
315 }
316 break;
317 case NVME_SC_CONNECT_INVALID_HOST:
318 dev_err(ctrl->device,
319 "Connect for subsystem %s is not allowed, hostnqn: %s\n",
320 data->subsysnqn, data->hostnqn);
321 break;
322 case NVME_SC_CONNECT_CTRL_BUSY:
323 dev_err(ctrl->device,
324 "Connect command failed: controller is busy or not available\n");
325 break;
326 case NVME_SC_CONNECT_FORMAT:
327 dev_err(ctrl->device,
328 "Connect incompatible format: %d",
329 cmd->connect.recfmt);
330 break;
331 case NVME_SC_HOST_PATH_ERROR:
332 dev_err(ctrl->device,
333 "Connect command failed: host path error\n");
334 break;
335 default:
336 dev_err(ctrl->device,
337 "Connect command failed, error wo/DNR bit: %d\n",
338 err_sctype);
339 break;
340 }
341 }
342
343 /**
344 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
345 * API function.
346 * @ctrl: Host nvme controller instance used to request
347 * a new NVMe controller allocation on the target
348 * system and establish an NVMe Admin connection to
349 * that controller.
350 *
351 * This function enables an NVMe host device to request a new allocation of
352 * an NVMe controller resource on a target system as well establish a
353 * fabrics-protocol connection of the NVMe Admin queue between the
354 * host system device and the allocated NVMe controller on the
355 * target system via a NVMe Fabrics "Connect" command.
356 *
357 * Return:
358 * 0: success
359 * > 0: NVMe error status code
360 * < 0: Linux errno error code
361 *
362 */
nvmf_connect_admin_queue(struct nvme_ctrl * ctrl)363 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
364 {
365 struct nvme_command cmd = { };
366 union nvme_result res;
367 struct nvmf_connect_data *data;
368 int ret;
369
370 cmd.connect.opcode = nvme_fabrics_command;
371 cmd.connect.fctype = nvme_fabrics_type_connect;
372 cmd.connect.qid = 0;
373 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
374
375 /*
376 * Set keep-alive timeout in seconds granularity (ms * 1000)
377 */
378 cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000);
379
380 if (ctrl->opts->disable_sqflow)
381 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
382
383 data = kzalloc(sizeof(*data), GFP_KERNEL);
384 if (!data)
385 return -ENOMEM;
386
387 uuid_copy(&data->hostid, &ctrl->opts->host->id);
388 data->cntlid = cpu_to_le16(0xffff);
389 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
390 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
391
392 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
393 data, sizeof(*data), 0, NVME_QID_ANY, 1,
394 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
395 if (ret) {
396 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
397 &cmd, data);
398 goto out_free_data;
399 }
400
401 ctrl->cntlid = le16_to_cpu(res.u16);
402
403 out_free_data:
404 kfree(data);
405 return ret;
406 }
407 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
408
409 /**
410 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
411 * API function.
412 * @ctrl: Host nvme controller instance used to establish an
413 * NVMe I/O queue connection to the already allocated NVMe
414 * controller on the target system.
415 * @qid: NVMe I/O queue number for the new I/O connection between
416 * host and target (note qid == 0 is illegal as this is
417 * the Admin queue, per NVMe standard).
418 *
419 * This function issues a fabrics-protocol connection
420 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
421 * between the host system device and the allocated NVMe controller
422 * on the target system.
423 *
424 * Return:
425 * 0: success
426 * > 0: NVMe error status code
427 * < 0: Linux errno error code
428 */
nvmf_connect_io_queue(struct nvme_ctrl * ctrl,u16 qid)429 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
430 {
431 struct nvme_command cmd = { };
432 struct nvmf_connect_data *data;
433 union nvme_result res;
434 int ret;
435
436 cmd.connect.opcode = nvme_fabrics_command;
437 cmd.connect.fctype = nvme_fabrics_type_connect;
438 cmd.connect.qid = cpu_to_le16(qid);
439 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
440
441 if (ctrl->opts->disable_sqflow)
442 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
443
444 data = kzalloc(sizeof(*data), GFP_KERNEL);
445 if (!data)
446 return -ENOMEM;
447
448 uuid_copy(&data->hostid, &ctrl->opts->host->id);
449 data->cntlid = cpu_to_le16(ctrl->cntlid);
450 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
451 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
452
453 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
454 data, sizeof(*data), 0, qid, 1,
455 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
456 if (ret) {
457 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
458 &cmd, data);
459 }
460 kfree(data);
461 return ret;
462 }
463 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
464
nvmf_should_reconnect(struct nvme_ctrl * ctrl)465 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
466 {
467 if (ctrl->opts->max_reconnects == -1 ||
468 ctrl->nr_reconnects < ctrl->opts->max_reconnects)
469 return true;
470
471 return false;
472 }
473 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
474
475 /**
476 * nvmf_register_transport() - NVMe Fabrics Library registration function.
477 * @ops: Transport ops instance to be registered to the
478 * common fabrics library.
479 *
480 * API function that registers the type of specific transport fabric
481 * being implemented to the common NVMe fabrics library. Part of
482 * the overall init sequence of starting up a fabrics driver.
483 */
nvmf_register_transport(struct nvmf_transport_ops * ops)484 int nvmf_register_transport(struct nvmf_transport_ops *ops)
485 {
486 if (!ops->create_ctrl)
487 return -EINVAL;
488
489 down_write(&nvmf_transports_rwsem);
490 list_add_tail(&ops->entry, &nvmf_transports);
491 up_write(&nvmf_transports_rwsem);
492
493 return 0;
494 }
495 EXPORT_SYMBOL_GPL(nvmf_register_transport);
496
497 /**
498 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
499 * @ops: Transport ops instance to be unregistered from the
500 * common fabrics library.
501 *
502 * Fabrics API function that unregisters the type of specific transport
503 * fabric being implemented from the common NVMe fabrics library.
504 * Part of the overall exit sequence of unloading the implemented driver.
505 */
nvmf_unregister_transport(struct nvmf_transport_ops * ops)506 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
507 {
508 down_write(&nvmf_transports_rwsem);
509 list_del(&ops->entry);
510 up_write(&nvmf_transports_rwsem);
511 }
512 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
513
nvmf_lookup_transport(struct nvmf_ctrl_options * opts)514 static struct nvmf_transport_ops *nvmf_lookup_transport(
515 struct nvmf_ctrl_options *opts)
516 {
517 struct nvmf_transport_ops *ops;
518
519 lockdep_assert_held(&nvmf_transports_rwsem);
520
521 list_for_each_entry(ops, &nvmf_transports, entry) {
522 if (strcmp(ops->name, opts->transport) == 0)
523 return ops;
524 }
525
526 return NULL;
527 }
528
529 static const match_table_t opt_tokens = {
530 { NVMF_OPT_TRANSPORT, "transport=%s" },
531 { NVMF_OPT_TRADDR, "traddr=%s" },
532 { NVMF_OPT_TRSVCID, "trsvcid=%s" },
533 { NVMF_OPT_NQN, "nqn=%s" },
534 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" },
535 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" },
536 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" },
537 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" },
538 { NVMF_OPT_KATO, "keep_alive_tmo=%d" },
539 { NVMF_OPT_HOSTNQN, "hostnqn=%s" },
540 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" },
541 { NVMF_OPT_HOST_IFACE, "host_iface=%s" },
542 { NVMF_OPT_HOST_ID, "hostid=%s" },
543 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" },
544 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" },
545 { NVMF_OPT_HDR_DIGEST, "hdr_digest" },
546 { NVMF_OPT_DATA_DIGEST, "data_digest" },
547 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" },
548 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" },
549 { NVMF_OPT_TOS, "tos=%d" },
550 { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" },
551 { NVMF_OPT_ERR, NULL }
552 };
553
nvmf_parse_options(struct nvmf_ctrl_options * opts,const char * buf)554 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
555 const char *buf)
556 {
557 substring_t args[MAX_OPT_ARGS];
558 char *options, *o, *p;
559 int token, ret = 0;
560 size_t nqnlen = 0;
561 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
562 uuid_t hostid;
563
564 /* Set defaults */
565 opts->queue_size = NVMF_DEF_QUEUE_SIZE;
566 opts->nr_io_queues = num_online_cpus();
567 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
568 opts->kato = 0;
569 opts->duplicate_connect = false;
570 opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
571 opts->hdr_digest = false;
572 opts->data_digest = false;
573 opts->tos = -1; /* < 0 == use transport default */
574
575 options = o = kstrdup(buf, GFP_KERNEL);
576 if (!options)
577 return -ENOMEM;
578
579 uuid_gen(&hostid);
580
581 while ((p = strsep(&o, ",\n")) != NULL) {
582 if (!*p)
583 continue;
584
585 token = match_token(p, opt_tokens, args);
586 opts->mask |= token;
587 switch (token) {
588 case NVMF_OPT_TRANSPORT:
589 p = match_strdup(args);
590 if (!p) {
591 ret = -ENOMEM;
592 goto out;
593 }
594 kfree(opts->transport);
595 opts->transport = p;
596 break;
597 case NVMF_OPT_NQN:
598 p = match_strdup(args);
599 if (!p) {
600 ret = -ENOMEM;
601 goto out;
602 }
603 kfree(opts->subsysnqn);
604 opts->subsysnqn = p;
605 nqnlen = strlen(opts->subsysnqn);
606 if (nqnlen >= NVMF_NQN_SIZE) {
607 pr_err("%s needs to be < %d bytes\n",
608 opts->subsysnqn, NVMF_NQN_SIZE);
609 ret = -EINVAL;
610 goto out;
611 }
612 opts->discovery_nqn =
613 !(strcmp(opts->subsysnqn,
614 NVME_DISC_SUBSYS_NAME));
615 break;
616 case NVMF_OPT_TRADDR:
617 p = match_strdup(args);
618 if (!p) {
619 ret = -ENOMEM;
620 goto out;
621 }
622 kfree(opts->traddr);
623 opts->traddr = p;
624 break;
625 case NVMF_OPT_TRSVCID:
626 p = match_strdup(args);
627 if (!p) {
628 ret = -ENOMEM;
629 goto out;
630 }
631 kfree(opts->trsvcid);
632 opts->trsvcid = p;
633 break;
634 case NVMF_OPT_QUEUE_SIZE:
635 if (match_int(args, &token)) {
636 ret = -EINVAL;
637 goto out;
638 }
639 if (token < NVMF_MIN_QUEUE_SIZE ||
640 token > NVMF_MAX_QUEUE_SIZE) {
641 pr_err("Invalid queue_size %d\n", token);
642 ret = -EINVAL;
643 goto out;
644 }
645 opts->queue_size = token;
646 break;
647 case NVMF_OPT_NR_IO_QUEUES:
648 if (match_int(args, &token)) {
649 ret = -EINVAL;
650 goto out;
651 }
652 if (token <= 0) {
653 pr_err("Invalid number of IOQs %d\n", token);
654 ret = -EINVAL;
655 goto out;
656 }
657 if (opts->discovery_nqn) {
658 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
659 break;
660 }
661
662 opts->nr_io_queues = min_t(unsigned int,
663 num_online_cpus(), token);
664 break;
665 case NVMF_OPT_KATO:
666 if (match_int(args, &token)) {
667 ret = -EINVAL;
668 goto out;
669 }
670
671 if (token < 0) {
672 pr_err("Invalid keep_alive_tmo %d\n", token);
673 ret = -EINVAL;
674 goto out;
675 } else if (token == 0 && !opts->discovery_nqn) {
676 /* Allowed for debug */
677 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
678 }
679 opts->kato = token;
680 break;
681 case NVMF_OPT_CTRL_LOSS_TMO:
682 if (match_int(args, &token)) {
683 ret = -EINVAL;
684 goto out;
685 }
686
687 if (token < 0)
688 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
689 ctrl_loss_tmo = token;
690 break;
691 case NVMF_OPT_FAIL_FAST_TMO:
692 if (match_int(args, &token)) {
693 ret = -EINVAL;
694 goto out;
695 }
696
697 if (token >= 0)
698 pr_warn("I/O fail on reconnect controller after %d sec\n",
699 token);
700 opts->fast_io_fail_tmo = token;
701 break;
702 case NVMF_OPT_HOSTNQN:
703 if (opts->host) {
704 pr_err("hostnqn already user-assigned: %s\n",
705 opts->host->nqn);
706 ret = -EADDRINUSE;
707 goto out;
708 }
709 p = match_strdup(args);
710 if (!p) {
711 ret = -ENOMEM;
712 goto out;
713 }
714 nqnlen = strlen(p);
715 if (nqnlen >= NVMF_NQN_SIZE) {
716 pr_err("%s needs to be < %d bytes\n",
717 p, NVMF_NQN_SIZE);
718 kfree(p);
719 ret = -EINVAL;
720 goto out;
721 }
722 opts->host = nvmf_host_add(p);
723 kfree(p);
724 if (!opts->host) {
725 ret = -ENOMEM;
726 goto out;
727 }
728 break;
729 case NVMF_OPT_RECONNECT_DELAY:
730 if (match_int(args, &token)) {
731 ret = -EINVAL;
732 goto out;
733 }
734 if (token <= 0) {
735 pr_err("Invalid reconnect_delay %d\n", token);
736 ret = -EINVAL;
737 goto out;
738 }
739 opts->reconnect_delay = token;
740 break;
741 case NVMF_OPT_HOST_TRADDR:
742 p = match_strdup(args);
743 if (!p) {
744 ret = -ENOMEM;
745 goto out;
746 }
747 kfree(opts->host_traddr);
748 opts->host_traddr = p;
749 break;
750 case NVMF_OPT_HOST_IFACE:
751 p = match_strdup(args);
752 if (!p) {
753 ret = -ENOMEM;
754 goto out;
755 }
756 kfree(opts->host_iface);
757 opts->host_iface = p;
758 break;
759 case NVMF_OPT_HOST_ID:
760 p = match_strdup(args);
761 if (!p) {
762 ret = -ENOMEM;
763 goto out;
764 }
765 ret = uuid_parse(p, &hostid);
766 if (ret) {
767 pr_err("Invalid hostid %s\n", p);
768 ret = -EINVAL;
769 kfree(p);
770 goto out;
771 }
772 kfree(p);
773 break;
774 case NVMF_OPT_DUP_CONNECT:
775 opts->duplicate_connect = true;
776 break;
777 case NVMF_OPT_DISABLE_SQFLOW:
778 opts->disable_sqflow = true;
779 break;
780 case NVMF_OPT_HDR_DIGEST:
781 opts->hdr_digest = true;
782 break;
783 case NVMF_OPT_DATA_DIGEST:
784 opts->data_digest = true;
785 break;
786 case NVMF_OPT_NR_WRITE_QUEUES:
787 if (match_int(args, &token)) {
788 ret = -EINVAL;
789 goto out;
790 }
791 if (token <= 0) {
792 pr_err("Invalid nr_write_queues %d\n", token);
793 ret = -EINVAL;
794 goto out;
795 }
796 opts->nr_write_queues = token;
797 break;
798 case NVMF_OPT_NR_POLL_QUEUES:
799 if (match_int(args, &token)) {
800 ret = -EINVAL;
801 goto out;
802 }
803 if (token <= 0) {
804 pr_err("Invalid nr_poll_queues %d\n", token);
805 ret = -EINVAL;
806 goto out;
807 }
808 opts->nr_poll_queues = token;
809 break;
810 case NVMF_OPT_TOS:
811 if (match_int(args, &token)) {
812 ret = -EINVAL;
813 goto out;
814 }
815 if (token < 0) {
816 pr_err("Invalid type of service %d\n", token);
817 ret = -EINVAL;
818 goto out;
819 }
820 if (token > 255) {
821 pr_warn("Clamping type of service to 255\n");
822 token = 255;
823 }
824 opts->tos = token;
825 break;
826 default:
827 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
828 p);
829 ret = -EINVAL;
830 goto out;
831 }
832 }
833
834 if (opts->discovery_nqn) {
835 opts->nr_io_queues = 0;
836 opts->nr_write_queues = 0;
837 opts->nr_poll_queues = 0;
838 opts->duplicate_connect = true;
839 } else {
840 if (!opts->kato)
841 opts->kato = NVME_DEFAULT_KATO;
842 }
843 if (ctrl_loss_tmo < 0) {
844 opts->max_reconnects = -1;
845 } else {
846 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
847 opts->reconnect_delay);
848 if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
849 pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
850 opts->fast_io_fail_tmo, ctrl_loss_tmo);
851 }
852
853 if (!opts->host) {
854 kref_get(&nvmf_default_host->ref);
855 opts->host = nvmf_default_host;
856 }
857
858 uuid_copy(&opts->host->id, &hostid);
859
860 out:
861 kfree(options);
862 return ret;
863 }
864
nvmf_check_required_opts(struct nvmf_ctrl_options * opts,unsigned int required_opts)865 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
866 unsigned int required_opts)
867 {
868 if ((opts->mask & required_opts) != required_opts) {
869 int i;
870
871 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
872 if ((opt_tokens[i].token & required_opts) &&
873 !(opt_tokens[i].token & opts->mask)) {
874 pr_warn("missing parameter '%s'\n",
875 opt_tokens[i].pattern);
876 }
877 }
878
879 return -EINVAL;
880 }
881
882 return 0;
883 }
884
nvmf_ip_options_match(struct nvme_ctrl * ctrl,struct nvmf_ctrl_options * opts)885 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
886 struct nvmf_ctrl_options *opts)
887 {
888 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
889 strcmp(opts->traddr, ctrl->opts->traddr) ||
890 strcmp(opts->trsvcid, ctrl->opts->trsvcid))
891 return false;
892
893 /*
894 * Checking the local address is rough. In most cases, none is specified
895 * and the host port is selected by the stack.
896 *
897 * Assume no match if:
898 * - local address is specified and address is not the same
899 * - local address is not specified but remote is, or vice versa
900 * (admin using specific host_traddr when it matters).
901 */
902 if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
903 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
904 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
905 return false;
906 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
907 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
908 return false;
909 }
910
911 return true;
912 }
913 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
914
nvmf_check_allowed_opts(struct nvmf_ctrl_options * opts,unsigned int allowed_opts)915 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
916 unsigned int allowed_opts)
917 {
918 if (opts->mask & ~allowed_opts) {
919 int i;
920
921 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
922 if ((opt_tokens[i].token & opts->mask) &&
923 (opt_tokens[i].token & ~allowed_opts)) {
924 pr_warn("invalid parameter '%s'\n",
925 opt_tokens[i].pattern);
926 }
927 }
928
929 return -EINVAL;
930 }
931
932 return 0;
933 }
934
nvmf_free_options(struct nvmf_ctrl_options * opts)935 void nvmf_free_options(struct nvmf_ctrl_options *opts)
936 {
937 nvmf_host_put(opts->host);
938 kfree(opts->transport);
939 kfree(opts->traddr);
940 kfree(opts->trsvcid);
941 kfree(opts->subsysnqn);
942 kfree(opts->host_traddr);
943 kfree(opts->host_iface);
944 kfree(opts);
945 }
946 EXPORT_SYMBOL_GPL(nvmf_free_options);
947
948 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
949 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
950 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
951 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
952 NVMF_OPT_DISABLE_SQFLOW |\
953 NVMF_OPT_FAIL_FAST_TMO)
954
955 static struct nvme_ctrl *
nvmf_create_ctrl(struct device * dev,const char * buf)956 nvmf_create_ctrl(struct device *dev, const char *buf)
957 {
958 struct nvmf_ctrl_options *opts;
959 struct nvmf_transport_ops *ops;
960 struct nvme_ctrl *ctrl;
961 int ret;
962
963 opts = kzalloc(sizeof(*opts), GFP_KERNEL);
964 if (!opts)
965 return ERR_PTR(-ENOMEM);
966
967 ret = nvmf_parse_options(opts, buf);
968 if (ret)
969 goto out_free_opts;
970
971
972 request_module("nvme-%s", opts->transport);
973
974 /*
975 * Check the generic options first as we need a valid transport for
976 * the lookup below. Then clear the generic flags so that transport
977 * drivers don't have to care about them.
978 */
979 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
980 if (ret)
981 goto out_free_opts;
982 opts->mask &= ~NVMF_REQUIRED_OPTS;
983
984 down_read(&nvmf_transports_rwsem);
985 ops = nvmf_lookup_transport(opts);
986 if (!ops) {
987 pr_info("no handler found for transport %s.\n",
988 opts->transport);
989 ret = -EINVAL;
990 goto out_unlock;
991 }
992
993 if (!try_module_get(ops->module)) {
994 ret = -EBUSY;
995 goto out_unlock;
996 }
997 up_read(&nvmf_transports_rwsem);
998
999 ret = nvmf_check_required_opts(opts, ops->required_opts);
1000 if (ret)
1001 goto out_module_put;
1002 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1003 ops->allowed_opts | ops->required_opts);
1004 if (ret)
1005 goto out_module_put;
1006
1007 ctrl = ops->create_ctrl(dev, opts);
1008 if (IS_ERR(ctrl)) {
1009 ret = PTR_ERR(ctrl);
1010 goto out_module_put;
1011 }
1012
1013 module_put(ops->module);
1014 return ctrl;
1015
1016 out_module_put:
1017 module_put(ops->module);
1018 goto out_free_opts;
1019 out_unlock:
1020 up_read(&nvmf_transports_rwsem);
1021 out_free_opts:
1022 nvmf_free_options(opts);
1023 return ERR_PTR(ret);
1024 }
1025
1026 static struct class *nvmf_class;
1027 static struct device *nvmf_device;
1028 static DEFINE_MUTEX(nvmf_dev_mutex);
1029
nvmf_dev_write(struct file * file,const char __user * ubuf,size_t count,loff_t * pos)1030 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1031 size_t count, loff_t *pos)
1032 {
1033 struct seq_file *seq_file = file->private_data;
1034 struct nvme_ctrl *ctrl;
1035 const char *buf;
1036 int ret = 0;
1037
1038 if (count > PAGE_SIZE)
1039 return -ENOMEM;
1040
1041 buf = memdup_user_nul(ubuf, count);
1042 if (IS_ERR(buf))
1043 return PTR_ERR(buf);
1044
1045 mutex_lock(&nvmf_dev_mutex);
1046 if (seq_file->private) {
1047 ret = -EINVAL;
1048 goto out_unlock;
1049 }
1050
1051 ctrl = nvmf_create_ctrl(nvmf_device, buf);
1052 if (IS_ERR(ctrl)) {
1053 ret = PTR_ERR(ctrl);
1054 goto out_unlock;
1055 }
1056
1057 seq_file->private = ctrl;
1058
1059 out_unlock:
1060 mutex_unlock(&nvmf_dev_mutex);
1061 kfree(buf);
1062 return ret ? ret : count;
1063 }
1064
nvmf_dev_show(struct seq_file * seq_file,void * private)1065 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1066 {
1067 struct nvme_ctrl *ctrl;
1068 int ret = 0;
1069
1070 mutex_lock(&nvmf_dev_mutex);
1071 ctrl = seq_file->private;
1072 if (!ctrl) {
1073 ret = -EINVAL;
1074 goto out_unlock;
1075 }
1076
1077 seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1078 ctrl->instance, ctrl->cntlid);
1079
1080 out_unlock:
1081 mutex_unlock(&nvmf_dev_mutex);
1082 return ret;
1083 }
1084
nvmf_dev_open(struct inode * inode,struct file * file)1085 static int nvmf_dev_open(struct inode *inode, struct file *file)
1086 {
1087 /*
1088 * The miscdevice code initializes file->private_data, but doesn't
1089 * make use of it later.
1090 */
1091 file->private_data = NULL;
1092 return single_open(file, nvmf_dev_show, NULL);
1093 }
1094
nvmf_dev_release(struct inode * inode,struct file * file)1095 static int nvmf_dev_release(struct inode *inode, struct file *file)
1096 {
1097 struct seq_file *seq_file = file->private_data;
1098 struct nvme_ctrl *ctrl = seq_file->private;
1099
1100 if (ctrl)
1101 nvme_put_ctrl(ctrl);
1102 return single_release(inode, file);
1103 }
1104
1105 static const struct file_operations nvmf_dev_fops = {
1106 .owner = THIS_MODULE,
1107 .write = nvmf_dev_write,
1108 .read = seq_read,
1109 .open = nvmf_dev_open,
1110 .release = nvmf_dev_release,
1111 };
1112
1113 static struct miscdevice nvmf_misc = {
1114 .minor = MISC_DYNAMIC_MINOR,
1115 .name = "nvme-fabrics",
1116 .fops = &nvmf_dev_fops,
1117 };
1118
nvmf_init(void)1119 static int __init nvmf_init(void)
1120 {
1121 int ret;
1122
1123 nvmf_default_host = nvmf_host_default();
1124 if (!nvmf_default_host)
1125 return -ENOMEM;
1126
1127 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1128 if (IS_ERR(nvmf_class)) {
1129 pr_err("couldn't register class nvme-fabrics\n");
1130 ret = PTR_ERR(nvmf_class);
1131 goto out_free_host;
1132 }
1133
1134 nvmf_device =
1135 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1136 if (IS_ERR(nvmf_device)) {
1137 pr_err("couldn't create nvme-fabris device!\n");
1138 ret = PTR_ERR(nvmf_device);
1139 goto out_destroy_class;
1140 }
1141
1142 ret = misc_register(&nvmf_misc);
1143 if (ret) {
1144 pr_err("couldn't register misc device: %d\n", ret);
1145 goto out_destroy_device;
1146 }
1147
1148 return 0;
1149
1150 out_destroy_device:
1151 device_destroy(nvmf_class, MKDEV(0, 0));
1152 out_destroy_class:
1153 class_destroy(nvmf_class);
1154 out_free_host:
1155 nvmf_host_put(nvmf_default_host);
1156 return ret;
1157 }
1158
nvmf_exit(void)1159 static void __exit nvmf_exit(void)
1160 {
1161 misc_deregister(&nvmf_misc);
1162 device_destroy(nvmf_class, MKDEV(0, 0));
1163 class_destroy(nvmf_class);
1164 nvmf_host_put(nvmf_default_host);
1165
1166 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1167 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1168 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1169 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1170 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1171 }
1172
1173 MODULE_LICENSE("GPL v2");
1174
1175 module_init(nvmf_init);
1176 module_exit(nvmf_exit);
1177
