fs: Avoid premature clearing of capabilities
[cascardo/linux.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                 goto out_unlock;
52
53         host = kmalloc(sizeof(*host), GFP_KERNEL);
54         if (!host)
55                 goto out_unlock;
56
57         kref_init(&host->ref);
58         memcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
59         uuid_le_gen(&host->id);
60
61         list_add_tail(&host->list, &nvmf_hosts);
62 out_unlock:
63         mutex_unlock(&nvmf_hosts_mutex);
64         return host;
65 }
66
67 static struct nvmf_host *nvmf_host_default(void)
68 {
69         struct nvmf_host *host;
70
71         host = kmalloc(sizeof(*host), GFP_KERNEL);
72         if (!host)
73                 return NULL;
74
75         kref_init(&host->ref);
76         uuid_le_gen(&host->id);
77         snprintf(host->nqn, NVMF_NQN_SIZE,
78                 "nqn.2014-08.org.nvmexpress:NVMf:uuid:%pUl", &host->id);
79
80         mutex_lock(&nvmf_hosts_mutex);
81         list_add_tail(&host->list, &nvmf_hosts);
82         mutex_unlock(&nvmf_hosts_mutex);
83
84         return host;
85 }
86
87 static void nvmf_host_destroy(struct kref *ref)
88 {
89         struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
90
91         mutex_lock(&nvmf_hosts_mutex);
92         list_del(&host->list);
93         mutex_unlock(&nvmf_hosts_mutex);
94
95         kfree(host);
96 }
97
98 static void nvmf_host_put(struct nvmf_host *host)
99 {
100         if (host)
101                 kref_put(&host->ref, nvmf_host_destroy);
102 }
103
104 /**
105  * nvmf_get_address() -  Get address/port
106  * @ctrl:       Host NVMe controller instance which we got the address
107  * @buf:        OUTPUT parameter that will contain the address/port
108  * @size:       buffer size
109  */
110 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
111 {
112         return snprintf(buf, size, "traddr=%s,trsvcid=%s\n",
113                         ctrl->opts->traddr, ctrl->opts->trsvcid);
114 }
115 EXPORT_SYMBOL_GPL(nvmf_get_address);
116
117 /**
118  * nvmf_get_subsysnqn() - Get subsystem NQN
119  * @ctrl:       Host NVMe controller instance which we got the NQN
120  */
121 const char *nvmf_get_subsysnqn(struct nvme_ctrl *ctrl)
122 {
123         return ctrl->opts->subsysnqn;
124 }
125 EXPORT_SYMBOL_GPL(nvmf_get_subsysnqn);
126
127 /**
128  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
129  * @ctrl:       Host NVMe controller instance maintaining the admin
130  *              queue used to submit the property read command to
131  *              the allocated NVMe controller resource on the target system.
132  * @off:        Starting offset value of the targeted property
133  *              register (see the fabrics section of the NVMe standard).
134  * @val:        OUTPUT parameter that will contain the value of
135  *              the property after a successful read.
136  *
137  * Used by the host system to retrieve a 32-bit capsule property value
138  * from an NVMe controller on the target system.
139  *
140  * ("Capsule property" is an "PCIe register concept" applied to the
141  * NVMe fabrics space.)
142  *
143  * Return:
144  *      0: successful read
145  *      > 0: NVMe error status code
146  *      < 0: Linux errno error code
147  */
148 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
149 {
150         struct nvme_command cmd;
151         struct nvme_completion cqe;
152         int ret;
153
154         memset(&cmd, 0, sizeof(cmd));
155         cmd.prop_get.opcode = nvme_fabrics_command;
156         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
157         cmd.prop_get.offset = cpu_to_le32(off);
158
159         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
160                         NVME_QID_ANY, 0, 0);
161
162         if (ret >= 0)
163                 *val = le64_to_cpu(cqe.result64);
164         if (unlikely(ret != 0))
165                 dev_err(ctrl->device,
166                         "Property Get error: %d, offset %#x\n",
167                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
168
169         return ret;
170 }
171 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
172
173 /**
174  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
175  * @ctrl:       Host NVMe controller instance maintaining the admin
176  *              queue used to submit the property read command to
177  *              the allocated controller resource on the target system.
178  * @off:        Starting offset value of the targeted property
179  *              register (see the fabrics section of the NVMe standard).
180  * @val:        OUTPUT parameter that will contain the value of
181  *              the property after a successful read.
182  *
183  * Used by the host system to retrieve a 64-bit capsule property value
184  * from an NVMe controller on the target system.
185  *
186  * ("Capsule property" is an "PCIe register concept" applied to the
187  * NVMe fabrics space.)
188  *
189  * Return:
190  *      0: successful read
191  *      > 0: NVMe error status code
192  *      < 0: Linux errno error code
193  */
194 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
195 {
196         struct nvme_command cmd;
197         struct nvme_completion cqe;
198         int ret;
199
200         memset(&cmd, 0, sizeof(cmd));
201         cmd.prop_get.opcode = nvme_fabrics_command;
202         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
203         cmd.prop_get.attrib = 1;
204         cmd.prop_get.offset = cpu_to_le32(off);
205
206         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
207                         NVME_QID_ANY, 0, 0);
208
209         if (ret >= 0)
210                 *val = le64_to_cpu(cqe.result64);
211         if (unlikely(ret != 0))
212                 dev_err(ctrl->device,
213                         "Property Get error: %d, offset %#x\n",
214                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
215         return ret;
216 }
217 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
218
219 /**
220  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
221  * @ctrl:       Host NVMe controller instance maintaining the admin
222  *              queue used to submit the property read command to
223  *              the allocated NVMe controller resource on the target system.
224  * @off:        Starting offset value of the targeted property
225  *              register (see the fabrics section of the NVMe standard).
226  * @val:        Input parameter that contains the value to be
227  *              written to the property.
228  *
229  * Used by the NVMe host system to write a 32-bit capsule property value
230  * to an NVMe controller on the target system.
231  *
232  * ("Capsule property" is an "PCIe register concept" applied to the
233  * NVMe fabrics space.)
234  *
235  * Return:
236  *      0: successful write
237  *      > 0: NVMe error status code
238  *      < 0: Linux errno error code
239  */
240 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
241 {
242         struct nvme_command cmd;
243         int ret;
244
245         memset(&cmd, 0, sizeof(cmd));
246         cmd.prop_set.opcode = nvme_fabrics_command;
247         cmd.prop_set.fctype = nvme_fabrics_type_property_set;
248         cmd.prop_set.attrib = 0;
249         cmd.prop_set.offset = cpu_to_le32(off);
250         cmd.prop_set.value = cpu_to_le64(val);
251
252         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
253                         NVME_QID_ANY, 0, 0);
254         if (unlikely(ret))
255                 dev_err(ctrl->device,
256                         "Property Set error: %d, offset %#x\n",
257                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
258         return ret;
259 }
260 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
261
262 /**
263  * nvmf_log_connect_error() - Error-parsing-diagnostic print
264  * out function for connect() errors.
265  *
266  * @ctrl: the specific /dev/nvmeX device that had the error.
267  *
268  * @errval: Error code to be decoded in a more human-friendly
269  *          printout.
270  *
271  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
272  *
273  * @cmd: This is the SQE portion of a submission capsule.
274  *
275  * @data: This is the "Data" portion of a submission capsule.
276  */
277 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
278                 int errval, int offset, struct nvme_command *cmd,
279                 struct nvmf_connect_data *data)
280 {
281         int err_sctype = errval & (~NVME_SC_DNR);
282
283         switch (err_sctype) {
284
285         case (NVME_SC_CONNECT_INVALID_PARAM):
286                 if (offset >> 16) {
287                         char *inv_data = "Connect Invalid Data Parameter";
288
289                         switch (offset & 0xffff) {
290                         case (offsetof(struct nvmf_connect_data, cntlid)):
291                                 dev_err(ctrl->device,
292                                         "%s, cntlid: %d\n",
293                                         inv_data, data->cntlid);
294                                 break;
295                         case (offsetof(struct nvmf_connect_data, hostnqn)):
296                                 dev_err(ctrl->device,
297                                         "%s, hostnqn \"%s\"\n",
298                                         inv_data, data->hostnqn);
299                                 break;
300                         case (offsetof(struct nvmf_connect_data, subsysnqn)):
301                                 dev_err(ctrl->device,
302                                         "%s, subsysnqn \"%s\"\n",
303                                         inv_data, data->subsysnqn);
304                                 break;
305                         default:
306                                 dev_err(ctrl->device,
307                                         "%s, starting byte offset: %d\n",
308                                        inv_data, offset & 0xffff);
309                                 break;
310                         }
311                 } else {
312                         char *inv_sqe = "Connect Invalid SQE Parameter";
313
314                         switch (offset) {
315                         case (offsetof(struct nvmf_connect_command, qid)):
316                                 dev_err(ctrl->device,
317                                        "%s, qid %d\n",
318                                         inv_sqe, cmd->connect.qid);
319                                 break;
320                         default:
321                                 dev_err(ctrl->device,
322                                         "%s, starting byte offset: %d\n",
323                                         inv_sqe, offset);
324                         }
325                 }
326                 break;
327         default:
328                 dev_err(ctrl->device,
329                         "Connect command failed, error wo/DNR bit: %d\n",
330                         err_sctype);
331                 break;
332         } /* switch (err_sctype) */
333 }
334
335 /**
336  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
337  *                              API function.
338  * @ctrl:       Host nvme controller instance used to request
339  *              a new NVMe controller allocation on the target
340  *              system and  establish an NVMe Admin connection to
341  *              that controller.
342  *
343  * This function enables an NVMe host device to request a new allocation of
344  * an NVMe controller resource on a target system as well establish a
345  * fabrics-protocol connection of the NVMe Admin queue between the
346  * host system device and the allocated NVMe controller on the
347  * target system via a NVMe Fabrics "Connect" command.
348  *
349  * Return:
350  *      0: success
351  *      > 0: NVMe error status code
352  *      < 0: Linux errno error code
353  *
354  */
355 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
356 {
357         struct nvme_command cmd;
358         struct nvme_completion cqe;
359         struct nvmf_connect_data *data;
360         int ret;
361
362         memset(&cmd, 0, sizeof(cmd));
363         cmd.connect.opcode = nvme_fabrics_command;
364         cmd.connect.fctype = nvme_fabrics_type_connect;
365         cmd.connect.qid = 0;
366         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
367         /*
368          * Set keep-alive timeout in seconds granularity (ms * 1000)
369          * and add a grace period for controller kato enforcement
370          */
371         cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
372                 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
373
374         data = kzalloc(sizeof(*data), GFP_KERNEL);
375         if (!data)
376                 return -ENOMEM;
377
378         memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_le));
379         data->cntlid = cpu_to_le16(0xffff);
380         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
381         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
382
383         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe,
384                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
385                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
386         if (ret) {
387                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(cqe.result),
388                                        &cmd, data);
389                 goto out_free_data;
390         }
391
392         ctrl->cntlid = le16_to_cpu(cqe.result16);
393
394 out_free_data:
395         kfree(data);
396         return ret;
397 }
398 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
399
400 /**
401  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
402  *                           API function.
403  * @ctrl:       Host nvme controller instance used to establish an
404  *              NVMe I/O queue connection to the already allocated NVMe
405  *              controller on the target system.
406  * @qid:        NVMe I/O queue number for the new I/O connection between
407  *              host and target (note qid == 0 is illegal as this is
408  *              the Admin queue, per NVMe standard).
409  *
410  * This function issues a fabrics-protocol connection
411  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
412  * between the host system device and the allocated NVMe controller
413  * on the target system.
414  *
415  * Return:
416  *      0: success
417  *      > 0: NVMe error status code
418  *      < 0: Linux errno error code
419  */
420 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
421 {
422         struct nvme_command cmd;
423         struct nvmf_connect_data *data;
424         struct nvme_completion cqe;
425         int ret;
426
427         memset(&cmd, 0, sizeof(cmd));
428         cmd.connect.opcode = nvme_fabrics_command;
429         cmd.connect.fctype = nvme_fabrics_type_connect;
430         cmd.connect.qid = cpu_to_le16(qid);
431         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
432
433         data = kzalloc(sizeof(*data), GFP_KERNEL);
434         if (!data)
435                 return -ENOMEM;
436
437         memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_le));
438         data->cntlid = cpu_to_le16(ctrl->cntlid);
439         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
440         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
441
442         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &cqe,
443                         data, sizeof(*data), 0, qid, 1,
444                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
445         if (ret) {
446                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(cqe.result),
447                                        &cmd, data);
448         }
449         kfree(data);
450         return ret;
451 }
452 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
453
454 /**
455  * nvmf_register_transport() - NVMe Fabrics Library registration function.
456  * @ops:        Transport ops instance to be registered to the
457  *              common fabrics library.
458  *
459  * API function that registers the type of specific transport fabric
460  * being implemented to the common NVMe fabrics library. Part of
461  * the overall init sequence of starting up a fabrics driver.
462  */
463 void nvmf_register_transport(struct nvmf_transport_ops *ops)
464 {
465         mutex_lock(&nvmf_transports_mutex);
466         list_add_tail(&ops->entry, &nvmf_transports);
467         mutex_unlock(&nvmf_transports_mutex);
468 }
469 EXPORT_SYMBOL_GPL(nvmf_register_transport);
470
471 /**
472  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
473  * @ops:        Transport ops instance to be unregistered from the
474  *              common fabrics library.
475  *
476  * Fabrics API function that unregisters the type of specific transport
477  * fabric being implemented from the common NVMe fabrics library.
478  * Part of the overall exit sequence of unloading the implemented driver.
479  */
480 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
481 {
482         mutex_lock(&nvmf_transports_mutex);
483         list_del(&ops->entry);
484         mutex_unlock(&nvmf_transports_mutex);
485 }
486 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
487
488 static struct nvmf_transport_ops *nvmf_lookup_transport(
489                 struct nvmf_ctrl_options *opts)
490 {
491         struct nvmf_transport_ops *ops;
492
493         lockdep_assert_held(&nvmf_transports_mutex);
494
495         list_for_each_entry(ops, &nvmf_transports, entry) {
496                 if (strcmp(ops->name, opts->transport) == 0)
497                         return ops;
498         }
499
500         return NULL;
501 }
502
503 static const match_table_t opt_tokens = {
504         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
505         { NVMF_OPT_TRADDR,              "traddr=%s"             },
506         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
507         { NVMF_OPT_NQN,                 "nqn=%s"                },
508         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
509         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
510         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
511         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
512         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
513         { NVMF_OPT_ERR,                 NULL                    }
514 };
515
516 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
517                 const char *buf)
518 {
519         substring_t args[MAX_OPT_ARGS];
520         char *options, *o, *p;
521         int token, ret = 0;
522         size_t nqnlen  = 0;
523
524         /* Set defaults */
525         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
526         opts->nr_io_queues = num_online_cpus();
527         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
528
529         options = o = kstrdup(buf, GFP_KERNEL);
530         if (!options)
531                 return -ENOMEM;
532
533         while ((p = strsep(&o, ",\n")) != NULL) {
534                 if (!*p)
535                         continue;
536
537                 token = match_token(p, opt_tokens, args);
538                 opts->mask |= token;
539                 switch (token) {
540                 case NVMF_OPT_TRANSPORT:
541                         p = match_strdup(args);
542                         if (!p) {
543                                 ret = -ENOMEM;
544                                 goto out;
545                         }
546                         opts->transport = p;
547                         break;
548                 case NVMF_OPT_NQN:
549                         p = match_strdup(args);
550                         if (!p) {
551                                 ret = -ENOMEM;
552                                 goto out;
553                         }
554                         opts->subsysnqn = p;
555                         nqnlen = strlen(opts->subsysnqn);
556                         if (nqnlen >= NVMF_NQN_SIZE) {
557                                 pr_err("%s needs to be < %d bytes\n",
558                                 opts->subsysnqn, NVMF_NQN_SIZE);
559                                 ret = -EINVAL;
560                                 goto out;
561                         }
562                         opts->discovery_nqn =
563                                 !(strcmp(opts->subsysnqn,
564                                          NVME_DISC_SUBSYS_NAME));
565                         if (opts->discovery_nqn)
566                                 opts->nr_io_queues = 0;
567                         break;
568                 case NVMF_OPT_TRADDR:
569                         p = match_strdup(args);
570                         if (!p) {
571                                 ret = -ENOMEM;
572                                 goto out;
573                         }
574                         opts->traddr = p;
575                         break;
576                 case NVMF_OPT_TRSVCID:
577                         p = match_strdup(args);
578                         if (!p) {
579                                 ret = -ENOMEM;
580                                 goto out;
581                         }
582                         opts->trsvcid = p;
583                         break;
584                 case NVMF_OPT_QUEUE_SIZE:
585                         if (match_int(args, &token)) {
586                                 ret = -EINVAL;
587                                 goto out;
588                         }
589                         if (token < NVMF_MIN_QUEUE_SIZE ||
590                             token > NVMF_MAX_QUEUE_SIZE) {
591                                 pr_err("Invalid queue_size %d\n", token);
592                                 ret = -EINVAL;
593                                 goto out;
594                         }
595                         opts->queue_size = token;
596                         break;
597                 case NVMF_OPT_NR_IO_QUEUES:
598                         if (match_int(args, &token)) {
599                                 ret = -EINVAL;
600                                 goto out;
601                         }
602                         if (token <= 0) {
603                                 pr_err("Invalid number of IOQs %d\n", token);
604                                 ret = -EINVAL;
605                                 goto out;
606                         }
607                         opts->nr_io_queues = min_t(unsigned int,
608                                         num_online_cpus(), token);
609                         break;
610                 case NVMF_OPT_KATO:
611                         if (match_int(args, &token)) {
612                                 ret = -EINVAL;
613                                 goto out;
614                         }
615
616                         if (opts->discovery_nqn) {
617                                 pr_err("Discovery controllers cannot accept keep_alive_tmo != 0\n");
618                                 ret = -EINVAL;
619                                 goto out;
620                         }
621
622                         if (token < 0) {
623                                 pr_err("Invalid keep_alive_tmo %d\n", token);
624                                 ret = -EINVAL;
625                                 goto out;
626                         } else if (token == 0) {
627                                 /* Allowed for debug */
628                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
629                         }
630                         opts->kato = token;
631                         break;
632                 case NVMF_OPT_HOSTNQN:
633                         if (opts->host) {
634                                 pr_err("hostnqn already user-assigned: %s\n",
635                                        opts->host->nqn);
636                                 ret = -EADDRINUSE;
637                                 goto out;
638                         }
639                         p = match_strdup(args);
640                         if (!p) {
641                                 ret = -ENOMEM;
642                                 goto out;
643                         }
644                         nqnlen = strlen(p);
645                         if (nqnlen >= NVMF_NQN_SIZE) {
646                                 pr_err("%s needs to be < %d bytes\n",
647                                         p, NVMF_NQN_SIZE);
648                                 ret = -EINVAL;
649                                 goto out;
650                         }
651                         opts->host = nvmf_host_add(p);
652                         if (!opts->host) {
653                                 ret = -ENOMEM;
654                                 goto out;
655                         }
656                         break;
657                 case NVMF_OPT_RECONNECT_DELAY:
658                         if (match_int(args, &token)) {
659                                 ret = -EINVAL;
660                                 goto out;
661                         }
662                         if (token <= 0) {
663                                 pr_err("Invalid reconnect_delay %d\n", token);
664                                 ret = -EINVAL;
665                                 goto out;
666                         }
667                         opts->reconnect_delay = token;
668                         break;
669                 default:
670                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
671                                 p);
672                         ret = -EINVAL;
673                         goto out;
674                 }
675         }
676
677         if (!opts->host) {
678                 kref_get(&nvmf_default_host->ref);
679                 opts->host = nvmf_default_host;
680         }
681
682 out:
683         if (!opts->discovery_nqn && !opts->kato)
684                 opts->kato = NVME_DEFAULT_KATO;
685         kfree(options);
686         return ret;
687 }
688
689 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
690                 unsigned int required_opts)
691 {
692         if ((opts->mask & required_opts) != required_opts) {
693                 int i;
694
695                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
696                         if ((opt_tokens[i].token & required_opts) &&
697                             !(opt_tokens[i].token & opts->mask)) {
698                                 pr_warn("missing parameter '%s'\n",
699                                         opt_tokens[i].pattern);
700                         }
701                 }
702
703                 return -EINVAL;
704         }
705
706         return 0;
707 }
708
709 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
710                 unsigned int allowed_opts)
711 {
712         if (opts->mask & ~allowed_opts) {
713                 int i;
714
715                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
716                         if (opt_tokens[i].token & ~allowed_opts) {
717                                 pr_warn("invalid parameter '%s'\n",
718                                         opt_tokens[i].pattern);
719                         }
720                 }
721
722                 return -EINVAL;
723         }
724
725         return 0;
726 }
727
728 void nvmf_free_options(struct nvmf_ctrl_options *opts)
729 {
730         nvmf_host_put(opts->host);
731         kfree(opts->transport);
732         kfree(opts->traddr);
733         kfree(opts->trsvcid);
734         kfree(opts->subsysnqn);
735         kfree(opts);
736 }
737 EXPORT_SYMBOL_GPL(nvmf_free_options);
738
739 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
740 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
741                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN)
742
743 static struct nvme_ctrl *
744 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
745 {
746         struct nvmf_ctrl_options *opts;
747         struct nvmf_transport_ops *ops;
748         struct nvme_ctrl *ctrl;
749         int ret;
750
751         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
752         if (!opts)
753                 return ERR_PTR(-ENOMEM);
754
755         ret = nvmf_parse_options(opts, buf);
756         if (ret)
757                 goto out_free_opts;
758
759         /*
760          * Check the generic options first as we need a valid transport for
761          * the lookup below.  Then clear the generic flags so that transport
762          * drivers don't have to care about them.
763          */
764         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
765         if (ret)
766                 goto out_free_opts;
767         opts->mask &= ~NVMF_REQUIRED_OPTS;
768
769         mutex_lock(&nvmf_transports_mutex);
770         ops = nvmf_lookup_transport(opts);
771         if (!ops) {
772                 pr_info("no handler found for transport %s.\n",
773                         opts->transport);
774                 ret = -EINVAL;
775                 goto out_unlock;
776         }
777
778         ret = nvmf_check_required_opts(opts, ops->required_opts);
779         if (ret)
780                 goto out_unlock;
781         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
782                                 ops->allowed_opts | ops->required_opts);
783         if (ret)
784                 goto out_unlock;
785
786         ctrl = ops->create_ctrl(dev, opts);
787         if (IS_ERR(ctrl)) {
788                 ret = PTR_ERR(ctrl);
789                 goto out_unlock;
790         }
791
792         mutex_unlock(&nvmf_transports_mutex);
793         return ctrl;
794
795 out_unlock:
796         mutex_unlock(&nvmf_transports_mutex);
797 out_free_opts:
798         nvmf_host_put(opts->host);
799         kfree(opts);
800         return ERR_PTR(ret);
801 }
802
803 static struct class *nvmf_class;
804 static struct device *nvmf_device;
805 static DEFINE_MUTEX(nvmf_dev_mutex);
806
807 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
808                 size_t count, loff_t *pos)
809 {
810         struct seq_file *seq_file = file->private_data;
811         struct nvme_ctrl *ctrl;
812         const char *buf;
813         int ret = 0;
814
815         if (count > PAGE_SIZE)
816                 return -ENOMEM;
817
818         buf = memdup_user_nul(ubuf, count);
819         if (IS_ERR(buf))
820                 return PTR_ERR(buf);
821
822         mutex_lock(&nvmf_dev_mutex);
823         if (seq_file->private) {
824                 ret = -EINVAL;
825                 goto out_unlock;
826         }
827
828         ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
829         if (IS_ERR(ctrl)) {
830                 ret = PTR_ERR(ctrl);
831                 goto out_unlock;
832         }
833
834         seq_file->private = ctrl;
835
836 out_unlock:
837         mutex_unlock(&nvmf_dev_mutex);
838         kfree(buf);
839         return ret ? ret : count;
840 }
841
842 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
843 {
844         struct nvme_ctrl *ctrl;
845         int ret = 0;
846
847         mutex_lock(&nvmf_dev_mutex);
848         ctrl = seq_file->private;
849         if (!ctrl) {
850                 ret = -EINVAL;
851                 goto out_unlock;
852         }
853
854         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
855                         ctrl->instance, ctrl->cntlid);
856
857 out_unlock:
858         mutex_unlock(&nvmf_dev_mutex);
859         return ret;
860 }
861
862 static int nvmf_dev_open(struct inode *inode, struct file *file)
863 {
864         /*
865          * The miscdevice code initializes file->private_data, but doesn't
866          * make use of it later.
867          */
868         file->private_data = NULL;
869         return single_open(file, nvmf_dev_show, NULL);
870 }
871
872 static int nvmf_dev_release(struct inode *inode, struct file *file)
873 {
874         struct seq_file *seq_file = file->private_data;
875         struct nvme_ctrl *ctrl = seq_file->private;
876
877         if (ctrl)
878                 nvme_put_ctrl(ctrl);
879         return single_release(inode, file);
880 }
881
882 static const struct file_operations nvmf_dev_fops = {
883         .owner          = THIS_MODULE,
884         .write          = nvmf_dev_write,
885         .read           = seq_read,
886         .open           = nvmf_dev_open,
887         .release        = nvmf_dev_release,
888 };
889
890 static struct miscdevice nvmf_misc = {
891         .minor          = MISC_DYNAMIC_MINOR,
892         .name           = "nvme-fabrics",
893         .fops           = &nvmf_dev_fops,
894 };
895
896 static int __init nvmf_init(void)
897 {
898         int ret;
899
900         nvmf_default_host = nvmf_host_default();
901         if (!nvmf_default_host)
902                 return -ENOMEM;
903
904         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
905         if (IS_ERR(nvmf_class)) {
906                 pr_err("couldn't register class nvme-fabrics\n");
907                 ret = PTR_ERR(nvmf_class);
908                 goto out_free_host;
909         }
910
911         nvmf_device =
912                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
913         if (IS_ERR(nvmf_device)) {
914                 pr_err("couldn't create nvme-fabris device!\n");
915                 ret = PTR_ERR(nvmf_device);
916                 goto out_destroy_class;
917         }
918
919         ret = misc_register(&nvmf_misc);
920         if (ret) {
921                 pr_err("couldn't register misc device: %d\n", ret);
922                 goto out_destroy_device;
923         }
924
925         return 0;
926
927 out_destroy_device:
928         device_destroy(nvmf_class, MKDEV(0, 0));
929 out_destroy_class:
930         class_destroy(nvmf_class);
931 out_free_host:
932         nvmf_host_put(nvmf_default_host);
933         return ret;
934 }
935
936 static void __exit nvmf_exit(void)
937 {
938         misc_deregister(&nvmf_misc);
939         device_destroy(nvmf_class, MKDEV(0, 0));
940         class_destroy(nvmf_class);
941         nvmf_host_put(nvmf_default_host);
942
943         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
944         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
945         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
946         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
947 }
948
949 MODULE_LICENSE("GPL v2");
950
951 module_init(nvmf_init);
952 module_exit(nvmf_exit);