Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[sfrench/cifs-2.6.git] / drivers / nvme / host / fabrics.c
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 DEFINE_MUTEX(nvmf_transports_mutex);
26
27 static LIST_HEAD(nvmf_hosts);
28 static DEFINE_MUTEX(nvmf_hosts_mutex);
29
30 static struct nvmf_host *nvmf_default_host;
31
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
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         memcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
61         uuid_be_gen(&host->id);
62
63         list_add_tail(&host->list, &nvmf_hosts);
64 out_unlock:
65         mutex_unlock(&nvmf_hosts_mutex);
66         return host;
67 }
68
69 static struct nvmf_host *nvmf_host_default(void)
70 {
71         struct nvmf_host *host;
72
73         host = kmalloc(sizeof(*host), GFP_KERNEL);
74         if (!host)
75                 return NULL;
76
77         kref_init(&host->ref);
78         uuid_be_gen(&host->id);
79         snprintf(host->nqn, NVMF_NQN_SIZE,
80                 "nqn.2014-08.org.nvmexpress:NVMf:uuid:%pUb", &host->id);
81
82         mutex_lock(&nvmf_hosts_mutex);
83         list_add_tail(&host->list, &nvmf_hosts);
84         mutex_unlock(&nvmf_hosts_mutex);
85
86         return host;
87 }
88
89 static void nvmf_host_destroy(struct kref *ref)
90 {
91         struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
92
93         mutex_lock(&nvmf_hosts_mutex);
94         list_del(&host->list);
95         mutex_unlock(&nvmf_hosts_mutex);
96
97         kfree(host);
98 }
99
100 static void nvmf_host_put(struct nvmf_host *host)
101 {
102         if (host)
103                 kref_put(&host->ref, nvmf_host_destroy);
104 }
105
106 /**
107  * nvmf_get_address() -  Get address/port
108  * @ctrl:       Host NVMe controller instance which we got the address
109  * @buf:        OUTPUT parameter that will contain the address/port
110  * @size:       buffer size
111  */
112 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
113 {
114         int len = 0;
115
116         if (ctrl->opts->mask & NVMF_OPT_TRADDR)
117                 len += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
118         if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
119                 len += snprintf(buf + len, size - len, "%strsvcid=%s",
120                                 (len) ? "," : "", ctrl->opts->trsvcid);
121         if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
122                 len += snprintf(buf + len, size - len, "%shost_traddr=%s",
123                                 (len) ? "," : "", ctrl->opts->host_traddr);
124         len += snprintf(buf + len, size - len, "\n");
125
126         return len;
127 }
128 EXPORT_SYMBOL_GPL(nvmf_get_address);
129
130 /**
131  * nvmf_get_subsysnqn() - Get subsystem NQN
132  * @ctrl:       Host NVMe controller instance which we got the NQN
133  */
134 const char *nvmf_get_subsysnqn(struct nvme_ctrl *ctrl)
135 {
136         return ctrl->opts->subsysnqn;
137 }
138 EXPORT_SYMBOL_GPL(nvmf_get_subsysnqn);
139
140 /**
141  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
142  * @ctrl:       Host NVMe controller instance maintaining the admin
143  *              queue used to submit the property read command to
144  *              the allocated NVMe controller resource on the target system.
145  * @off:        Starting offset value of the targeted property
146  *              register (see the fabrics section of the NVMe standard).
147  * @val:        OUTPUT parameter that will contain the value of
148  *              the property after a successful read.
149  *
150  * Used by the host system to retrieve a 32-bit capsule property value
151  * from an NVMe controller on the target system.
152  *
153  * ("Capsule property" is an "PCIe register concept" applied to the
154  * NVMe fabrics space.)
155  *
156  * Return:
157  *      0: successful read
158  *      > 0: NVMe error status code
159  *      < 0: Linux errno error code
160  */
161 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
162 {
163         struct nvme_command cmd;
164         union nvme_result res;
165         int ret;
166
167         memset(&cmd, 0, sizeof(cmd));
168         cmd.prop_get.opcode = nvme_fabrics_command;
169         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
170         cmd.prop_get.offset = cpu_to_le32(off);
171
172         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
173                         NVME_QID_ANY, 0, 0);
174
175         if (ret >= 0)
176                 *val = le64_to_cpu(res.u64);
177         if (unlikely(ret != 0))
178                 dev_err(ctrl->device,
179                         "Property Get error: %d, offset %#x\n",
180                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
181
182         return ret;
183 }
184 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
185
186 /**
187  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
188  * @ctrl:       Host NVMe controller instance maintaining the admin
189  *              queue used to submit the property read command to
190  *              the allocated controller resource on the target system.
191  * @off:        Starting offset value of the targeted property
192  *              register (see the fabrics section of the NVMe standard).
193  * @val:        OUTPUT parameter that will contain the value of
194  *              the property after a successful read.
195  *
196  * Used by the host system to retrieve a 64-bit capsule property value
197  * from an NVMe controller on the target system.
198  *
199  * ("Capsule property" is an "PCIe register concept" applied to the
200  * NVMe fabrics space.)
201  *
202  * Return:
203  *      0: successful read
204  *      > 0: NVMe error status code
205  *      < 0: Linux errno error code
206  */
207 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
208 {
209         struct nvme_command cmd;
210         union nvme_result res;
211         int ret;
212
213         memset(&cmd, 0, sizeof(cmd));
214         cmd.prop_get.opcode = nvme_fabrics_command;
215         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
216         cmd.prop_get.attrib = 1;
217         cmd.prop_get.offset = cpu_to_le32(off);
218
219         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
220                         NVME_QID_ANY, 0, 0);
221
222         if (ret >= 0)
223                 *val = le64_to_cpu(res.u64);
224         if (unlikely(ret != 0))
225                 dev_err(ctrl->device,
226                         "Property Get error: %d, offset %#x\n",
227                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
228         return ret;
229 }
230 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
231
232 /**
233  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
234  * @ctrl:       Host NVMe controller instance maintaining the admin
235  *              queue used to submit the property read command to
236  *              the allocated NVMe controller resource on the target system.
237  * @off:        Starting offset value of the targeted property
238  *              register (see the fabrics section of the NVMe standard).
239  * @val:        Input parameter that contains the value to be
240  *              written to the property.
241  *
242  * Used by the NVMe host system to write a 32-bit capsule property value
243  * to an NVMe controller on the target system.
244  *
245  * ("Capsule property" is an "PCIe register concept" applied to the
246  * NVMe fabrics space.)
247  *
248  * Return:
249  *      0: successful write
250  *      > 0: NVMe error status code
251  *      < 0: Linux errno error code
252  */
253 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
254 {
255         struct nvme_command cmd;
256         int ret;
257
258         memset(&cmd, 0, sizeof(cmd));
259         cmd.prop_set.opcode = nvme_fabrics_command;
260         cmd.prop_set.fctype = nvme_fabrics_type_property_set;
261         cmd.prop_set.attrib = 0;
262         cmd.prop_set.offset = cpu_to_le32(off);
263         cmd.prop_set.value = cpu_to_le64(val);
264
265         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
266                         NVME_QID_ANY, 0, 0);
267         if (unlikely(ret))
268                 dev_err(ctrl->device,
269                         "Property Set error: %d, offset %#x\n",
270                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
271         return ret;
272 }
273 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
274
275 /**
276  * nvmf_log_connect_error() - Error-parsing-diagnostic print
277  * out function for connect() errors.
278  *
279  * @ctrl: the specific /dev/nvmeX device that had the error.
280  *
281  * @errval: Error code to be decoded in a more human-friendly
282  *          printout.
283  *
284  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
285  *
286  * @cmd: This is the SQE portion of a submission capsule.
287  *
288  * @data: This is the "Data" portion of a submission capsule.
289  */
290 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
291                 int errval, int offset, struct nvme_command *cmd,
292                 struct nvmf_connect_data *data)
293 {
294         int err_sctype = errval & (~NVME_SC_DNR);
295
296         switch (err_sctype) {
297
298         case (NVME_SC_CONNECT_INVALID_PARAM):
299                 if (offset >> 16) {
300                         char *inv_data = "Connect Invalid Data Parameter";
301
302                         switch (offset & 0xffff) {
303                         case (offsetof(struct nvmf_connect_data, cntlid)):
304                                 dev_err(ctrl->device,
305                                         "%s, cntlid: %d\n",
306                                         inv_data, data->cntlid);
307                                 break;
308                         case (offsetof(struct nvmf_connect_data, hostnqn)):
309                                 dev_err(ctrl->device,
310                                         "%s, hostnqn \"%s\"\n",
311                                         inv_data, data->hostnqn);
312                                 break;
313                         case (offsetof(struct nvmf_connect_data, subsysnqn)):
314                                 dev_err(ctrl->device,
315                                         "%s, subsysnqn \"%s\"\n",
316                                         inv_data, data->subsysnqn);
317                                 break;
318                         default:
319                                 dev_err(ctrl->device,
320                                         "%s, starting byte offset: %d\n",
321                                        inv_data, offset & 0xffff);
322                                 break;
323                         }
324                 } else {
325                         char *inv_sqe = "Connect Invalid SQE Parameter";
326
327                         switch (offset) {
328                         case (offsetof(struct nvmf_connect_command, qid)):
329                                 dev_err(ctrl->device,
330                                        "%s, qid %d\n",
331                                         inv_sqe, cmd->connect.qid);
332                                 break;
333                         default:
334                                 dev_err(ctrl->device,
335                                         "%s, starting byte offset: %d\n",
336                                         inv_sqe, offset);
337                         }
338                 }
339                 break;
340         default:
341                 dev_err(ctrl->device,
342                         "Connect command failed, error wo/DNR bit: %d\n",
343                         err_sctype);
344                 break;
345         } /* switch (err_sctype) */
346 }
347
348 /**
349  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
350  *                              API function.
351  * @ctrl:       Host nvme controller instance used to request
352  *              a new NVMe controller allocation on the target
353  *              system and  establish an NVMe Admin connection to
354  *              that controller.
355  *
356  * This function enables an NVMe host device to request a new allocation of
357  * an NVMe controller resource on a target system as well establish a
358  * fabrics-protocol connection of the NVMe Admin queue between the
359  * host system device and the allocated NVMe controller on the
360  * target system via a NVMe Fabrics "Connect" command.
361  *
362  * Return:
363  *      0: success
364  *      > 0: NVMe error status code
365  *      < 0: Linux errno error code
366  *
367  */
368 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
369 {
370         struct nvme_command cmd;
371         union nvme_result res;
372         struct nvmf_connect_data *data;
373         int ret;
374
375         memset(&cmd, 0, sizeof(cmd));
376         cmd.connect.opcode = nvme_fabrics_command;
377         cmd.connect.fctype = nvme_fabrics_type_connect;
378         cmd.connect.qid = 0;
379
380         /*
381          * fabrics spec sets a minimum of depth 32 for admin queue,
382          * so set the queue with this depth always until
383          * justification otherwise.
384          */
385         cmd.connect.sqsize = cpu_to_le16(NVMF_AQ_DEPTH - 1);
386
387         /*
388          * Set keep-alive timeout in seconds granularity (ms * 1000)
389          * and add a grace period for controller kato enforcement
390          */
391         cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
392                 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
393
394         data = kzalloc(sizeof(*data), GFP_KERNEL);
395         if (!data)
396                 return -ENOMEM;
397
398         memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_be));
399         data->cntlid = cpu_to_le16(0xffff);
400         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
401         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
402
403         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,
404                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
405                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
406         if (ret) {
407                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
408                                        &cmd, data);
409                 goto out_free_data;
410         }
411
412         ctrl->cntlid = le16_to_cpu(res.u16);
413
414 out_free_data:
415         kfree(data);
416         return ret;
417 }
418 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
419
420 /**
421  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
422  *                           API function.
423  * @ctrl:       Host nvme controller instance used to establish an
424  *              NVMe I/O queue connection to the already allocated NVMe
425  *              controller on the target system.
426  * @qid:        NVMe I/O queue number for the new I/O connection between
427  *              host and target (note qid == 0 is illegal as this is
428  *              the Admin queue, per NVMe standard).
429  *
430  * This function issues a fabrics-protocol connection
431  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
432  * between the host system device and the allocated NVMe controller
433  * on the target system.
434  *
435  * Return:
436  *      0: success
437  *      > 0: NVMe error status code
438  *      < 0: Linux errno error code
439  */
440 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
441 {
442         struct nvme_command cmd;
443         struct nvmf_connect_data *data;
444         union nvme_result res;
445         int ret;
446
447         memset(&cmd, 0, sizeof(cmd));
448         cmd.connect.opcode = nvme_fabrics_command;
449         cmd.connect.fctype = nvme_fabrics_type_connect;
450         cmd.connect.qid = cpu_to_le16(qid);
451         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
452
453         data = kzalloc(sizeof(*data), GFP_KERNEL);
454         if (!data)
455                 return -ENOMEM;
456
457         memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_be));
458         data->cntlid = cpu_to_le16(ctrl->cntlid);
459         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
460         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
461
462         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
463                         data, sizeof(*data), 0, qid, 1,
464                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
465         if (ret) {
466                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
467                                        &cmd, data);
468         }
469         kfree(data);
470         return ret;
471 }
472 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
473
474 /**
475  * nvmf_register_transport() - NVMe Fabrics Library registration function.
476  * @ops:        Transport ops instance to be registered to the
477  *              common fabrics library.
478  *
479  * API function that registers the type of specific transport fabric
480  * being implemented to the common NVMe fabrics library. Part of
481  * the overall init sequence of starting up a fabrics driver.
482  */
483 int nvmf_register_transport(struct nvmf_transport_ops *ops)
484 {
485         if (!ops->create_ctrl)
486                 return -EINVAL;
487
488         mutex_lock(&nvmf_transports_mutex);
489         list_add_tail(&ops->entry, &nvmf_transports);
490         mutex_unlock(&nvmf_transports_mutex);
491
492         return 0;
493 }
494 EXPORT_SYMBOL_GPL(nvmf_register_transport);
495
496 /**
497  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
498  * @ops:        Transport ops instance to be unregistered from the
499  *              common fabrics library.
500  *
501  * Fabrics API function that unregisters the type of specific transport
502  * fabric being implemented from the common NVMe fabrics library.
503  * Part of the overall exit sequence of unloading the implemented driver.
504  */
505 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
506 {
507         mutex_lock(&nvmf_transports_mutex);
508         list_del(&ops->entry);
509         mutex_unlock(&nvmf_transports_mutex);
510 }
511 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
512
513 static struct nvmf_transport_ops *nvmf_lookup_transport(
514                 struct nvmf_ctrl_options *opts)
515 {
516         struct nvmf_transport_ops *ops;
517
518         lockdep_assert_held(&nvmf_transports_mutex);
519
520         list_for_each_entry(ops, &nvmf_transports, entry) {
521                 if (strcmp(ops->name, opts->transport) == 0)
522                         return ops;
523         }
524
525         return NULL;
526 }
527
528 static const match_table_t opt_tokens = {
529         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
530         { NVMF_OPT_TRADDR,              "traddr=%s"             },
531         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
532         { NVMF_OPT_NQN,                 "nqn=%s"                },
533         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
534         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
535         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
536         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
537         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
538         { NVMF_OPT_HOST_TRADDR,         "host_traddr=%s"        },
539         { NVMF_OPT_ERR,                 NULL                    }
540 };
541
542 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
543                 const char *buf)
544 {
545         substring_t args[MAX_OPT_ARGS];
546         char *options, *o, *p;
547         int token, ret = 0;
548         size_t nqnlen  = 0;
549
550         /* Set defaults */
551         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
552         opts->nr_io_queues = num_online_cpus();
553         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
554
555         options = o = kstrdup(buf, GFP_KERNEL);
556         if (!options)
557                 return -ENOMEM;
558
559         while ((p = strsep(&o, ",\n")) != NULL) {
560                 if (!*p)
561                         continue;
562
563                 token = match_token(p, opt_tokens, args);
564                 opts->mask |= token;
565                 switch (token) {
566                 case NVMF_OPT_TRANSPORT:
567                         p = match_strdup(args);
568                         if (!p) {
569                                 ret = -ENOMEM;
570                                 goto out;
571                         }
572                         opts->transport = p;
573                         break;
574                 case NVMF_OPT_NQN:
575                         p = match_strdup(args);
576                         if (!p) {
577                                 ret = -ENOMEM;
578                                 goto out;
579                         }
580                         opts->subsysnqn = p;
581                         nqnlen = strlen(opts->subsysnqn);
582                         if (nqnlen >= NVMF_NQN_SIZE) {
583                                 pr_err("%s needs to be < %d bytes\n",
584                                         opts->subsysnqn, NVMF_NQN_SIZE);
585                                 ret = -EINVAL;
586                                 goto out;
587                         }
588                         opts->discovery_nqn =
589                                 !(strcmp(opts->subsysnqn,
590                                          NVME_DISC_SUBSYS_NAME));
591                         if (opts->discovery_nqn)
592                                 opts->nr_io_queues = 0;
593                         break;
594                 case NVMF_OPT_TRADDR:
595                         p = match_strdup(args);
596                         if (!p) {
597                                 ret = -ENOMEM;
598                                 goto out;
599                         }
600                         opts->traddr = p;
601                         break;
602                 case NVMF_OPT_TRSVCID:
603                         p = match_strdup(args);
604                         if (!p) {
605                                 ret = -ENOMEM;
606                                 goto out;
607                         }
608                         opts->trsvcid = p;
609                         break;
610                 case NVMF_OPT_QUEUE_SIZE:
611                         if (match_int(args, &token)) {
612                                 ret = -EINVAL;
613                                 goto out;
614                         }
615                         if (token < NVMF_MIN_QUEUE_SIZE ||
616                             token > NVMF_MAX_QUEUE_SIZE) {
617                                 pr_err("Invalid queue_size %d\n", token);
618                                 ret = -EINVAL;
619                                 goto out;
620                         }
621                         opts->queue_size = token;
622                         break;
623                 case NVMF_OPT_NR_IO_QUEUES:
624                         if (match_int(args, &token)) {
625                                 ret = -EINVAL;
626                                 goto out;
627                         }
628                         if (token <= 0) {
629                                 pr_err("Invalid number of IOQs %d\n", token);
630                                 ret = -EINVAL;
631                                 goto out;
632                         }
633                         opts->nr_io_queues = min_t(unsigned int,
634                                         num_online_cpus(), token);
635                         break;
636                 case NVMF_OPT_KATO:
637                         if (match_int(args, &token)) {
638                                 ret = -EINVAL;
639                                 goto out;
640                         }
641
642                         if (opts->discovery_nqn) {
643                                 pr_err("Discovery controllers cannot accept keep_alive_tmo != 0\n");
644                                 ret = -EINVAL;
645                                 goto out;
646                         }
647
648                         if (token < 0) {
649                                 pr_err("Invalid keep_alive_tmo %d\n", token);
650                                 ret = -EINVAL;
651                                 goto out;
652                         } else if (token == 0) {
653                                 /* Allowed for debug */
654                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
655                         }
656                         opts->kato = token;
657                         break;
658                 case NVMF_OPT_HOSTNQN:
659                         if (opts->host) {
660                                 pr_err("hostnqn already user-assigned: %s\n",
661                                        opts->host->nqn);
662                                 ret = -EADDRINUSE;
663                                 goto out;
664                         }
665                         p = match_strdup(args);
666                         if (!p) {
667                                 ret = -ENOMEM;
668                                 goto out;
669                         }
670                         nqnlen = strlen(p);
671                         if (nqnlen >= NVMF_NQN_SIZE) {
672                                 pr_err("%s needs to be < %d bytes\n",
673                                         p, NVMF_NQN_SIZE);
674                                 kfree(p);
675                                 ret = -EINVAL;
676                                 goto out;
677                         }
678                         opts->host = nvmf_host_add(p);
679                         kfree(p);
680                         if (!opts->host) {
681                                 ret = -ENOMEM;
682                                 goto out;
683                         }
684                         break;
685                 case NVMF_OPT_RECONNECT_DELAY:
686                         if (match_int(args, &token)) {
687                                 ret = -EINVAL;
688                                 goto out;
689                         }
690                         if (token <= 0) {
691                                 pr_err("Invalid reconnect_delay %d\n", token);
692                                 ret = -EINVAL;
693                                 goto out;
694                         }
695                         opts->reconnect_delay = token;
696                         break;
697                 case NVMF_OPT_HOST_TRADDR:
698                         p = match_strdup(args);
699                         if (!p) {
700                                 ret = -ENOMEM;
701                                 goto out;
702                         }
703                         opts->host_traddr = p;
704                         break;
705                 default:
706                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
707                                 p);
708                         ret = -EINVAL;
709                         goto out;
710                 }
711         }
712
713         if (!opts->host) {
714                 kref_get(&nvmf_default_host->ref);
715                 opts->host = nvmf_default_host;
716         }
717
718 out:
719         if (!opts->discovery_nqn && !opts->kato)
720                 opts->kato = NVME_DEFAULT_KATO;
721         kfree(options);
722         return ret;
723 }
724
725 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
726                 unsigned int required_opts)
727 {
728         if ((opts->mask & required_opts) != required_opts) {
729                 int i;
730
731                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
732                         if ((opt_tokens[i].token & required_opts) &&
733                             !(opt_tokens[i].token & opts->mask)) {
734                                 pr_warn("missing parameter '%s'\n",
735                                         opt_tokens[i].pattern);
736                         }
737                 }
738
739                 return -EINVAL;
740         }
741
742         return 0;
743 }
744
745 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
746                 unsigned int allowed_opts)
747 {
748         if (opts->mask & ~allowed_opts) {
749                 int i;
750
751                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
752                         if (opt_tokens[i].token & ~allowed_opts) {
753                                 pr_warn("invalid parameter '%s'\n",
754                                         opt_tokens[i].pattern);
755                         }
756                 }
757
758                 return -EINVAL;
759         }
760
761         return 0;
762 }
763
764 void nvmf_free_options(struct nvmf_ctrl_options *opts)
765 {
766         nvmf_host_put(opts->host);
767         kfree(opts->transport);
768         kfree(opts->traddr);
769         kfree(opts->trsvcid);
770         kfree(opts->subsysnqn);
771         kfree(opts->host_traddr);
772         kfree(opts);
773 }
774 EXPORT_SYMBOL_GPL(nvmf_free_options);
775
776 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
777 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
778                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN)
779
780 static struct nvme_ctrl *
781 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
782 {
783         struct nvmf_ctrl_options *opts;
784         struct nvmf_transport_ops *ops;
785         struct nvme_ctrl *ctrl;
786         int ret;
787
788         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
789         if (!opts)
790                 return ERR_PTR(-ENOMEM);
791
792         ret = nvmf_parse_options(opts, buf);
793         if (ret)
794                 goto out_free_opts;
795
796         /*
797          * Check the generic options first as we need a valid transport for
798          * the lookup below.  Then clear the generic flags so that transport
799          * drivers don't have to care about them.
800          */
801         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
802         if (ret)
803                 goto out_free_opts;
804         opts->mask &= ~NVMF_REQUIRED_OPTS;
805
806         mutex_lock(&nvmf_transports_mutex);
807         ops = nvmf_lookup_transport(opts);
808         if (!ops) {
809                 pr_info("no handler found for transport %s.\n",
810                         opts->transport);
811                 ret = -EINVAL;
812                 goto out_unlock;
813         }
814
815         ret = nvmf_check_required_opts(opts, ops->required_opts);
816         if (ret)
817                 goto out_unlock;
818         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
819                                 ops->allowed_opts | ops->required_opts);
820         if (ret)
821                 goto out_unlock;
822
823         ctrl = ops->create_ctrl(dev, opts);
824         if (IS_ERR(ctrl)) {
825                 ret = PTR_ERR(ctrl);
826                 goto out_unlock;
827         }
828
829         mutex_unlock(&nvmf_transports_mutex);
830         return ctrl;
831
832 out_unlock:
833         mutex_unlock(&nvmf_transports_mutex);
834 out_free_opts:
835         nvmf_free_options(opts);
836         return ERR_PTR(ret);
837 }
838
839 static struct class *nvmf_class;
840 static struct device *nvmf_device;
841 static DEFINE_MUTEX(nvmf_dev_mutex);
842
843 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
844                 size_t count, loff_t *pos)
845 {
846         struct seq_file *seq_file = file->private_data;
847         struct nvme_ctrl *ctrl;
848         const char *buf;
849         int ret = 0;
850
851         if (count > PAGE_SIZE)
852                 return -ENOMEM;
853
854         buf = memdup_user_nul(ubuf, count);
855         if (IS_ERR(buf))
856                 return PTR_ERR(buf);
857
858         mutex_lock(&nvmf_dev_mutex);
859         if (seq_file->private) {
860                 ret = -EINVAL;
861                 goto out_unlock;
862         }
863
864         ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
865         if (IS_ERR(ctrl)) {
866                 ret = PTR_ERR(ctrl);
867                 goto out_unlock;
868         }
869
870         seq_file->private = ctrl;
871
872 out_unlock:
873         mutex_unlock(&nvmf_dev_mutex);
874         kfree(buf);
875         return ret ? ret : count;
876 }
877
878 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
879 {
880         struct nvme_ctrl *ctrl;
881         int ret = 0;
882
883         mutex_lock(&nvmf_dev_mutex);
884         ctrl = seq_file->private;
885         if (!ctrl) {
886                 ret = -EINVAL;
887                 goto out_unlock;
888         }
889
890         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
891                         ctrl->instance, ctrl->cntlid);
892
893 out_unlock:
894         mutex_unlock(&nvmf_dev_mutex);
895         return ret;
896 }
897
898 static int nvmf_dev_open(struct inode *inode, struct file *file)
899 {
900         /*
901          * The miscdevice code initializes file->private_data, but doesn't
902          * make use of it later.
903          */
904         file->private_data = NULL;
905         return single_open(file, nvmf_dev_show, NULL);
906 }
907
908 static int nvmf_dev_release(struct inode *inode, struct file *file)
909 {
910         struct seq_file *seq_file = file->private_data;
911         struct nvme_ctrl *ctrl = seq_file->private;
912
913         if (ctrl)
914                 nvme_put_ctrl(ctrl);
915         return single_release(inode, file);
916 }
917
918 static const struct file_operations nvmf_dev_fops = {
919         .owner          = THIS_MODULE,
920         .write          = nvmf_dev_write,
921         .read           = seq_read,
922         .open           = nvmf_dev_open,
923         .release        = nvmf_dev_release,
924 };
925
926 static struct miscdevice nvmf_misc = {
927         .minor          = MISC_DYNAMIC_MINOR,
928         .name           = "nvme-fabrics",
929         .fops           = &nvmf_dev_fops,
930 };
931
932 static int __init nvmf_init(void)
933 {
934         int ret;
935
936         nvmf_default_host = nvmf_host_default();
937         if (!nvmf_default_host)
938                 return -ENOMEM;
939
940         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
941         if (IS_ERR(nvmf_class)) {
942                 pr_err("couldn't register class nvme-fabrics\n");
943                 ret = PTR_ERR(nvmf_class);
944                 goto out_free_host;
945         }
946
947         nvmf_device =
948                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
949         if (IS_ERR(nvmf_device)) {
950                 pr_err("couldn't create nvme-fabris device!\n");
951                 ret = PTR_ERR(nvmf_device);
952                 goto out_destroy_class;
953         }
954
955         ret = misc_register(&nvmf_misc);
956         if (ret) {
957                 pr_err("couldn't register misc device: %d\n", ret);
958                 goto out_destroy_device;
959         }
960
961         return 0;
962
963 out_destroy_device:
964         device_destroy(nvmf_class, MKDEV(0, 0));
965 out_destroy_class:
966         class_destroy(nvmf_class);
967 out_free_host:
968         nvmf_host_put(nvmf_default_host);
969         return ret;
970 }
971
972 static void __exit nvmf_exit(void)
973 {
974         misc_deregister(&nvmf_misc);
975         device_destroy(nvmf_class, MKDEV(0, 0));
976         class_destroy(nvmf_class);
977         nvmf_host_put(nvmf_default_host);
978
979         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
980         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
981         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
982         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
983 }
984
985 MODULE_LICENSE("GPL v2");
986
987 module_init(nvmf_init);
988 module_exit(nvmf_exit);