Merge tag 'dma-mapping-5.13' of git://git.infradead.org/users/hch/dma-mapping
[linux-2.6-microblaze.git] / drivers / nvme / host / fabrics.c
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
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
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
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
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
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  */
103 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
104 {
105         int len = 0;
106
107         if (ctrl->opts->mask & NVMF_OPT_TRADDR)
108                 len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
109         if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
110                 len += scnprintf(buf + len, size - len, "%strsvcid=%s",
111                                 (len) ? "," : "", ctrl->opts->trsvcid);
112         if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
113                 len += scnprintf(buf + len, size - len, "%shost_traddr=%s",
114                                 (len) ? "," : "", ctrl->opts->host_traddr);
115         len += scnprintf(buf + len, size - len, "\n");
116
117         return len;
118 }
119 EXPORT_SYMBOL_GPL(nvmf_get_address);
120
121 /**
122  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
123  * @ctrl:       Host NVMe controller instance maintaining the admin
124  *              queue used to submit the property read command to
125  *              the allocated NVMe controller resource on the target system.
126  * @off:        Starting offset value of the targeted property
127  *              register (see the fabrics section of the NVMe standard).
128  * @val:        OUTPUT parameter that will contain the value of
129  *              the property after a successful read.
130  *
131  * Used by the host system to retrieve a 32-bit capsule property value
132  * from an NVMe controller on the target system.
133  *
134  * ("Capsule property" is an "PCIe register concept" applied to the
135  * NVMe fabrics space.)
136  *
137  * Return:
138  *      0: successful read
139  *      > 0: NVMe error status code
140  *      < 0: Linux errno error code
141  */
142 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
143 {
144         struct nvme_command cmd;
145         union nvme_result res;
146         int ret;
147
148         memset(&cmd, 0, sizeof(cmd));
149         cmd.prop_get.opcode = nvme_fabrics_command;
150         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
151         cmd.prop_get.offset = cpu_to_le32(off);
152
153         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
154                         NVME_QID_ANY, 0, 0, false);
155
156         if (ret >= 0)
157                 *val = le64_to_cpu(res.u64);
158         if (unlikely(ret != 0))
159                 dev_err(ctrl->device,
160                         "Property Get error: %d, offset %#x\n",
161                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
162
163         return ret;
164 }
165 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
166
167 /**
168  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
169  * @ctrl:       Host NVMe controller instance maintaining the admin
170  *              queue used to submit the property read command to
171  *              the allocated controller resource on the target system.
172  * @off:        Starting offset value of the targeted property
173  *              register (see the fabrics section of the NVMe standard).
174  * @val:        OUTPUT parameter that will contain the value of
175  *              the property after a successful read.
176  *
177  * Used by the host system to retrieve a 64-bit capsule property value
178  * from an NVMe controller on the target system.
179  *
180  * ("Capsule property" is an "PCIe register concept" applied to the
181  * NVMe fabrics space.)
182  *
183  * Return:
184  *      0: successful read
185  *      > 0: NVMe error status code
186  *      < 0: Linux errno error code
187  */
188 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
189 {
190         struct nvme_command cmd;
191         union nvme_result res;
192         int ret;
193
194         memset(&cmd, 0, sizeof(cmd));
195         cmd.prop_get.opcode = nvme_fabrics_command;
196         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
197         cmd.prop_get.attrib = 1;
198         cmd.prop_get.offset = cpu_to_le32(off);
199
200         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
201                         NVME_QID_ANY, 0, 0, false);
202
203         if (ret >= 0)
204                 *val = le64_to_cpu(res.u64);
205         if (unlikely(ret != 0))
206                 dev_err(ctrl->device,
207                         "Property Get error: %d, offset %#x\n",
208                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
209         return ret;
210 }
211 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
212
213 /**
214  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
215  * @ctrl:       Host NVMe controller instance maintaining the admin
216  *              queue used to submit the property read command to
217  *              the allocated NVMe controller resource on the target system.
218  * @off:        Starting offset value of the targeted property
219  *              register (see the fabrics section of the NVMe standard).
220  * @val:        Input parameter that contains the value to be
221  *              written to the property.
222  *
223  * Used by the NVMe host system to write a 32-bit capsule property value
224  * to an NVMe controller on the target system.
225  *
226  * ("Capsule property" is an "PCIe register concept" applied to the
227  * NVMe fabrics space.)
228  *
229  * Return:
230  *      0: successful write
231  *      > 0: NVMe error status code
232  *      < 0: Linux errno error code
233  */
234 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
235 {
236         struct nvme_command cmd;
237         int ret;
238
239         memset(&cmd, 0, sizeof(cmd));
240         cmd.prop_set.opcode = nvme_fabrics_command;
241         cmd.prop_set.fctype = nvme_fabrics_type_property_set;
242         cmd.prop_set.attrib = 0;
243         cmd.prop_set.offset = cpu_to_le32(off);
244         cmd.prop_set.value = cpu_to_le64(val);
245
246         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 0,
247                         NVME_QID_ANY, 0, 0, false);
248         if (unlikely(ret))
249                 dev_err(ctrl->device,
250                         "Property Set error: %d, offset %#x\n",
251                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
252         return ret;
253 }
254 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
255
256 /**
257  * nvmf_log_connect_error() - Error-parsing-diagnostic print
258  * out function for connect() errors.
259  *
260  * @ctrl: the specific /dev/nvmeX device that had the error.
261  *
262  * @errval: Error code to be decoded in a more human-friendly
263  *          printout.
264  *
265  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
266  *
267  * @cmd: This is the SQE portion of a submission capsule.
268  *
269  * @data: This is the "Data" portion of a submission capsule.
270  */
271 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
272                 int errval, int offset, struct nvme_command *cmd,
273                 struct nvmf_connect_data *data)
274 {
275         int err_sctype = errval & (~NVME_SC_DNR);
276
277         switch (err_sctype) {
278
279         case (NVME_SC_CONNECT_INVALID_PARAM):
280                 if (offset >> 16) {
281                         char *inv_data = "Connect Invalid Data Parameter";
282
283                         switch (offset & 0xffff) {
284                         case (offsetof(struct nvmf_connect_data, cntlid)):
285                                 dev_err(ctrl->device,
286                                         "%s, cntlid: %d\n",
287                                         inv_data, data->cntlid);
288                                 break;
289                         case (offsetof(struct nvmf_connect_data, hostnqn)):
290                                 dev_err(ctrl->device,
291                                         "%s, hostnqn \"%s\"\n",
292                                         inv_data, data->hostnqn);
293                                 break;
294                         case (offsetof(struct nvmf_connect_data, subsysnqn)):
295                                 dev_err(ctrl->device,
296                                         "%s, subsysnqn \"%s\"\n",
297                                         inv_data, data->subsysnqn);
298                                 break;
299                         default:
300                                 dev_err(ctrl->device,
301                                         "%s, starting byte offset: %d\n",
302                                        inv_data, offset & 0xffff);
303                                 break;
304                         }
305                 } else {
306                         char *inv_sqe = "Connect Invalid SQE Parameter";
307
308                         switch (offset) {
309                         case (offsetof(struct nvmf_connect_command, qid)):
310                                 dev_err(ctrl->device,
311                                        "%s, qid %d\n",
312                                         inv_sqe, cmd->connect.qid);
313                                 break;
314                         default:
315                                 dev_err(ctrl->device,
316                                         "%s, starting byte offset: %d\n",
317                                         inv_sqe, offset);
318                         }
319                 }
320                 break;
321
322         case NVME_SC_CONNECT_INVALID_HOST:
323                 dev_err(ctrl->device,
324                         "Connect for subsystem %s is not allowed, hostnqn: %s\n",
325                         data->subsysnqn, data->hostnqn);
326                 break;
327
328         case NVME_SC_CONNECT_CTRL_BUSY:
329                 dev_err(ctrl->device,
330                         "Connect command failed: controller is busy or not available\n");
331                 break;
332
333         case NVME_SC_CONNECT_FORMAT:
334                 dev_err(ctrl->device,
335                         "Connect incompatible format: %d",
336                         cmd->connect.recfmt);
337                 break;
338
339         default:
340                 dev_err(ctrl->device,
341                         "Connect command failed, error wo/DNR bit: %d\n",
342                         err_sctype);
343                 break;
344         } /* switch (err_sctype) */
345 }
346
347 /**
348  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
349  *                              API function.
350  * @ctrl:       Host nvme controller instance used to request
351  *              a new NVMe controller allocation on the target
352  *              system and  establish an NVMe Admin connection to
353  *              that controller.
354  *
355  * This function enables an NVMe host device to request a new allocation of
356  * an NVMe controller resource on a target system as well establish a
357  * fabrics-protocol connection of the NVMe Admin queue between the
358  * host system device and the allocated NVMe controller on the
359  * target system via a NVMe Fabrics "Connect" command.
360  *
361  * Return:
362  *      0: success
363  *      > 0: NVMe error status code
364  *      < 0: Linux errno error code
365  *
366  */
367 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
368 {
369         struct nvme_command cmd;
370         union nvme_result res;
371         struct nvmf_connect_data *data;
372         int ret;
373
374         memset(&cmd, 0, sizeof(cmd));
375         cmd.connect.opcode = nvme_fabrics_command;
376         cmd.connect.fctype = nvme_fabrics_type_connect;
377         cmd.connect.qid = 0;
378         cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
379
380         /*
381          * Set keep-alive timeout in seconds granularity (ms * 1000)
382          */
383         cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000);
384
385         if (ctrl->opts->disable_sqflow)
386                 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
387
388         data = kzalloc(sizeof(*data), GFP_KERNEL);
389         if (!data)
390                 return -ENOMEM;
391
392         uuid_copy(&data->hostid, &ctrl->opts->host->id);
393         data->cntlid = cpu_to_le16(0xffff);
394         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
395         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
396
397         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
398                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
399                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, false);
400         if (ret) {
401                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
402                                        &cmd, data);
403                 goto out_free_data;
404         }
405
406         ctrl->cntlid = le16_to_cpu(res.u16);
407
408 out_free_data:
409         kfree(data);
410         return ret;
411 }
412 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
413
414 /**
415  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
416  *                           API function.
417  * @ctrl:       Host nvme controller instance used to establish an
418  *              NVMe I/O queue connection to the already allocated NVMe
419  *              controller on the target system.
420  * @qid:        NVMe I/O queue number for the new I/O connection between
421  *              host and target (note qid == 0 is illegal as this is
422  *              the Admin queue, per NVMe standard).
423  * @poll:       Whether or not to poll for the completion of the connect cmd.
424  *
425  * This function issues a fabrics-protocol connection
426  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
427  * between the host system device and the allocated NVMe controller
428  * on the target system.
429  *
430  * Return:
431  *      0: success
432  *      > 0: NVMe error status code
433  *      < 0: Linux errno error code
434  */
435 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid, bool poll)
436 {
437         struct nvme_command cmd;
438         struct nvmf_connect_data *data;
439         union nvme_result res;
440         int ret;
441
442         memset(&cmd, 0, sizeof(cmd));
443         cmd.connect.opcode = nvme_fabrics_command;
444         cmd.connect.fctype = nvme_fabrics_type_connect;
445         cmd.connect.qid = cpu_to_le16(qid);
446         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
447
448         if (ctrl->opts->disable_sqflow)
449                 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
450
451         data = kzalloc(sizeof(*data), GFP_KERNEL);
452         if (!data)
453                 return -ENOMEM;
454
455         uuid_copy(&data->hostid, &ctrl->opts->host->id);
456         data->cntlid = cpu_to_le16(ctrl->cntlid);
457         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
458         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
459
460         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
461                         data, sizeof(*data), 0, qid, 1,
462                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, poll);
463         if (ret) {
464                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
465                                        &cmd, data);
466         }
467         kfree(data);
468         return ret;
469 }
470 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
471
472 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
473 {
474         if (ctrl->opts->max_reconnects == -1 ||
475             ctrl->nr_reconnects < ctrl->opts->max_reconnects)
476                 return true;
477
478         return false;
479 }
480 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
481
482 /**
483  * nvmf_register_transport() - NVMe Fabrics Library registration function.
484  * @ops:        Transport ops instance to be registered to the
485  *              common fabrics library.
486  *
487  * API function that registers the type of specific transport fabric
488  * being implemented to the common NVMe fabrics library. Part of
489  * the overall init sequence of starting up a fabrics driver.
490  */
491 int nvmf_register_transport(struct nvmf_transport_ops *ops)
492 {
493         if (!ops->create_ctrl)
494                 return -EINVAL;
495
496         down_write(&nvmf_transports_rwsem);
497         list_add_tail(&ops->entry, &nvmf_transports);
498         up_write(&nvmf_transports_rwsem);
499
500         return 0;
501 }
502 EXPORT_SYMBOL_GPL(nvmf_register_transport);
503
504 /**
505  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
506  * @ops:        Transport ops instance to be unregistered from the
507  *              common fabrics library.
508  *
509  * Fabrics API function that unregisters the type of specific transport
510  * fabric being implemented from the common NVMe fabrics library.
511  * Part of the overall exit sequence of unloading the implemented driver.
512  */
513 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
514 {
515         down_write(&nvmf_transports_rwsem);
516         list_del(&ops->entry);
517         up_write(&nvmf_transports_rwsem);
518 }
519 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
520
521 static struct nvmf_transport_ops *nvmf_lookup_transport(
522                 struct nvmf_ctrl_options *opts)
523 {
524         struct nvmf_transport_ops *ops;
525
526         lockdep_assert_held(&nvmf_transports_rwsem);
527
528         list_for_each_entry(ops, &nvmf_transports, entry) {
529                 if (strcmp(ops->name, opts->transport) == 0)
530                         return ops;
531         }
532
533         return NULL;
534 }
535
536 /*
537  * For something we're not in a state to send to the device the default action
538  * is to busy it and retry it after the controller state is recovered.  However,
539  * if the controller is deleting or if anything is marked for failfast or
540  * nvme multipath it is immediately failed.
541  *
542  * Note: commands used to initialize the controller will be marked for failfast.
543  * Note: nvme cli/ioctl commands are marked for failfast.
544  */
545 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
546                 struct request *rq)
547 {
548         if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
549             ctrl->state != NVME_CTRL_DEAD &&
550             !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
551             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
552                 return BLK_STS_RESOURCE;
553         return nvme_host_path_error(rq);
554 }
555 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
556
557 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
558                 bool queue_live)
559 {
560         struct nvme_request *req = nvme_req(rq);
561
562         /*
563          * currently we have a problem sending passthru commands
564          * on the admin_q if the controller is not LIVE because we can't
565          * make sure that they are going out after the admin connect,
566          * controller enable and/or other commands in the initialization
567          * sequence. until the controller will be LIVE, fail with
568          * BLK_STS_RESOURCE so that they will be rescheduled.
569          */
570         if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
571                 return false;
572
573         /*
574          * Only allow commands on a live queue, except for the connect command,
575          * which is require to set the queue live in the appropinquate states.
576          */
577         switch (ctrl->state) {
578         case NVME_CTRL_CONNECTING:
579                 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
580                     req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
581                         return true;
582                 break;
583         default:
584                 break;
585         case NVME_CTRL_DEAD:
586                 return false;
587         }
588
589         return queue_live;
590 }
591 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
592
593 static const match_table_t opt_tokens = {
594         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
595         { NVMF_OPT_TRADDR,              "traddr=%s"             },
596         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
597         { NVMF_OPT_NQN,                 "nqn=%s"                },
598         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
599         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
600         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
601         { NVMF_OPT_CTRL_LOSS_TMO,       "ctrl_loss_tmo=%d"      },
602         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
603         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
604         { NVMF_OPT_HOST_TRADDR,         "host_traddr=%s"        },
605         { NVMF_OPT_HOST_ID,             "hostid=%s"             },
606         { NVMF_OPT_DUP_CONNECT,         "duplicate_connect"     },
607         { NVMF_OPT_DISABLE_SQFLOW,      "disable_sqflow"        },
608         { NVMF_OPT_HDR_DIGEST,          "hdr_digest"            },
609         { NVMF_OPT_DATA_DIGEST,         "data_digest"           },
610         { NVMF_OPT_NR_WRITE_QUEUES,     "nr_write_queues=%d"    },
611         { NVMF_OPT_NR_POLL_QUEUES,      "nr_poll_queues=%d"     },
612         { NVMF_OPT_TOS,                 "tos=%d"                },
613         { NVMF_OPT_FAIL_FAST_TMO,       "fast_io_fail_tmo=%d"   },
614         { NVMF_OPT_ERR,                 NULL                    }
615 };
616
617 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
618                 const char *buf)
619 {
620         substring_t args[MAX_OPT_ARGS];
621         char *options, *o, *p;
622         int token, ret = 0;
623         size_t nqnlen  = 0;
624         int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
625         uuid_t hostid;
626
627         /* Set defaults */
628         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
629         opts->nr_io_queues = num_online_cpus();
630         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
631         opts->kato = 0;
632         opts->duplicate_connect = false;
633         opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
634         opts->hdr_digest = false;
635         opts->data_digest = false;
636         opts->tos = -1; /* < 0 == use transport default */
637
638         options = o = kstrdup(buf, GFP_KERNEL);
639         if (!options)
640                 return -ENOMEM;
641
642         uuid_gen(&hostid);
643
644         while ((p = strsep(&o, ",\n")) != NULL) {
645                 if (!*p)
646                         continue;
647
648                 token = match_token(p, opt_tokens, args);
649                 opts->mask |= token;
650                 switch (token) {
651                 case NVMF_OPT_TRANSPORT:
652                         p = match_strdup(args);
653                         if (!p) {
654                                 ret = -ENOMEM;
655                                 goto out;
656                         }
657                         kfree(opts->transport);
658                         opts->transport = p;
659                         break;
660                 case NVMF_OPT_NQN:
661                         p = match_strdup(args);
662                         if (!p) {
663                                 ret = -ENOMEM;
664                                 goto out;
665                         }
666                         kfree(opts->subsysnqn);
667                         opts->subsysnqn = p;
668                         nqnlen = strlen(opts->subsysnqn);
669                         if (nqnlen >= NVMF_NQN_SIZE) {
670                                 pr_err("%s needs to be < %d bytes\n",
671                                         opts->subsysnqn, NVMF_NQN_SIZE);
672                                 ret = -EINVAL;
673                                 goto out;
674                         }
675                         opts->discovery_nqn =
676                                 !(strcmp(opts->subsysnqn,
677                                          NVME_DISC_SUBSYS_NAME));
678                         break;
679                 case NVMF_OPT_TRADDR:
680                         p = match_strdup(args);
681                         if (!p) {
682                                 ret = -ENOMEM;
683                                 goto out;
684                         }
685                         kfree(opts->traddr);
686                         opts->traddr = p;
687                         break;
688                 case NVMF_OPT_TRSVCID:
689                         p = match_strdup(args);
690                         if (!p) {
691                                 ret = -ENOMEM;
692                                 goto out;
693                         }
694                         kfree(opts->trsvcid);
695                         opts->trsvcid = p;
696                         break;
697                 case NVMF_OPT_QUEUE_SIZE:
698                         if (match_int(args, &token)) {
699                                 ret = -EINVAL;
700                                 goto out;
701                         }
702                         if (token < NVMF_MIN_QUEUE_SIZE ||
703                             token > NVMF_MAX_QUEUE_SIZE) {
704                                 pr_err("Invalid queue_size %d\n", token);
705                                 ret = -EINVAL;
706                                 goto out;
707                         }
708                         opts->queue_size = token;
709                         break;
710                 case NVMF_OPT_NR_IO_QUEUES:
711                         if (match_int(args, &token)) {
712                                 ret = -EINVAL;
713                                 goto out;
714                         }
715                         if (token <= 0) {
716                                 pr_err("Invalid number of IOQs %d\n", token);
717                                 ret = -EINVAL;
718                                 goto out;
719                         }
720                         if (opts->discovery_nqn) {
721                                 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
722                                 break;
723                         }
724
725                         opts->nr_io_queues = min_t(unsigned int,
726                                         num_online_cpus(), token);
727                         break;
728                 case NVMF_OPT_KATO:
729                         if (match_int(args, &token)) {
730                                 ret = -EINVAL;
731                                 goto out;
732                         }
733
734                         if (token < 0) {
735                                 pr_err("Invalid keep_alive_tmo %d\n", token);
736                                 ret = -EINVAL;
737                                 goto out;
738                         } else if (token == 0 && !opts->discovery_nqn) {
739                                 /* Allowed for debug */
740                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
741                         }
742                         opts->kato = token;
743                         break;
744                 case NVMF_OPT_CTRL_LOSS_TMO:
745                         if (match_int(args, &token)) {
746                                 ret = -EINVAL;
747                                 goto out;
748                         }
749
750                         if (token < 0)
751                                 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
752                         ctrl_loss_tmo = token;
753                         break;
754                 case NVMF_OPT_FAIL_FAST_TMO:
755                         if (match_int(args, &token)) {
756                                 ret = -EINVAL;
757                                 goto out;
758                         }
759
760                         if (token >= 0)
761                                 pr_warn("I/O fail on reconnect controller after %d sec\n",
762                                         token);
763                         opts->fast_io_fail_tmo = token;
764                         break;
765                 case NVMF_OPT_HOSTNQN:
766                         if (opts->host) {
767                                 pr_err("hostnqn already user-assigned: %s\n",
768                                        opts->host->nqn);
769                                 ret = -EADDRINUSE;
770                                 goto out;
771                         }
772                         p = match_strdup(args);
773                         if (!p) {
774                                 ret = -ENOMEM;
775                                 goto out;
776                         }
777                         nqnlen = strlen(p);
778                         if (nqnlen >= NVMF_NQN_SIZE) {
779                                 pr_err("%s needs to be < %d bytes\n",
780                                         p, NVMF_NQN_SIZE);
781                                 kfree(p);
782                                 ret = -EINVAL;
783                                 goto out;
784                         }
785                         nvmf_host_put(opts->host);
786                         opts->host = nvmf_host_add(p);
787                         kfree(p);
788                         if (!opts->host) {
789                                 ret = -ENOMEM;
790                                 goto out;
791                         }
792                         break;
793                 case NVMF_OPT_RECONNECT_DELAY:
794                         if (match_int(args, &token)) {
795                                 ret = -EINVAL;
796                                 goto out;
797                         }
798                         if (token <= 0) {
799                                 pr_err("Invalid reconnect_delay %d\n", token);
800                                 ret = -EINVAL;
801                                 goto out;
802                         }
803                         opts->reconnect_delay = token;
804                         break;
805                 case NVMF_OPT_HOST_TRADDR:
806                         p = match_strdup(args);
807                         if (!p) {
808                                 ret = -ENOMEM;
809                                 goto out;
810                         }
811                         kfree(opts->host_traddr);
812                         opts->host_traddr = p;
813                         break;
814                 case NVMF_OPT_HOST_ID:
815                         p = match_strdup(args);
816                         if (!p) {
817                                 ret = -ENOMEM;
818                                 goto out;
819                         }
820                         ret = uuid_parse(p, &hostid);
821                         if (ret) {
822                                 pr_err("Invalid hostid %s\n", p);
823                                 ret = -EINVAL;
824                                 kfree(p);
825                                 goto out;
826                         }
827                         kfree(p);
828                         break;
829                 case NVMF_OPT_DUP_CONNECT:
830                         opts->duplicate_connect = true;
831                         break;
832                 case NVMF_OPT_DISABLE_SQFLOW:
833                         opts->disable_sqflow = true;
834                         break;
835                 case NVMF_OPT_HDR_DIGEST:
836                         opts->hdr_digest = true;
837                         break;
838                 case NVMF_OPT_DATA_DIGEST:
839                         opts->data_digest = true;
840                         break;
841                 case NVMF_OPT_NR_WRITE_QUEUES:
842                         if (match_int(args, &token)) {
843                                 ret = -EINVAL;
844                                 goto out;
845                         }
846                         if (token <= 0) {
847                                 pr_err("Invalid nr_write_queues %d\n", token);
848                                 ret = -EINVAL;
849                                 goto out;
850                         }
851                         opts->nr_write_queues = token;
852                         break;
853                 case NVMF_OPT_NR_POLL_QUEUES:
854                         if (match_int(args, &token)) {
855                                 ret = -EINVAL;
856                                 goto out;
857                         }
858                         if (token <= 0) {
859                                 pr_err("Invalid nr_poll_queues %d\n", token);
860                                 ret = -EINVAL;
861                                 goto out;
862                         }
863                         opts->nr_poll_queues = token;
864                         break;
865                 case NVMF_OPT_TOS:
866                         if (match_int(args, &token)) {
867                                 ret = -EINVAL;
868                                 goto out;
869                         }
870                         if (token < 0) {
871                                 pr_err("Invalid type of service %d\n", token);
872                                 ret = -EINVAL;
873                                 goto out;
874                         }
875                         if (token > 255) {
876                                 pr_warn("Clamping type of service to 255\n");
877                                 token = 255;
878                         }
879                         opts->tos = token;
880                         break;
881                 default:
882                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
883                                 p);
884                         ret = -EINVAL;
885                         goto out;
886                 }
887         }
888
889         if (opts->discovery_nqn) {
890                 opts->nr_io_queues = 0;
891                 opts->nr_write_queues = 0;
892                 opts->nr_poll_queues = 0;
893                 opts->duplicate_connect = true;
894         } else {
895                 if (!opts->kato)
896                         opts->kato = NVME_DEFAULT_KATO;
897         }
898         if (ctrl_loss_tmo < 0) {
899                 opts->max_reconnects = -1;
900         } else {
901                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
902                                                 opts->reconnect_delay);
903                 if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
904                         pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
905                                 opts->fast_io_fail_tmo, ctrl_loss_tmo);
906         }
907
908         if (!opts->host) {
909                 kref_get(&nvmf_default_host->ref);
910                 opts->host = nvmf_default_host;
911         }
912
913         uuid_copy(&opts->host->id, &hostid);
914
915 out:
916         kfree(options);
917         return ret;
918 }
919
920 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
921                 unsigned int required_opts)
922 {
923         if ((opts->mask & required_opts) != required_opts) {
924                 int i;
925
926                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
927                         if ((opt_tokens[i].token & required_opts) &&
928                             !(opt_tokens[i].token & opts->mask)) {
929                                 pr_warn("missing parameter '%s'\n",
930                                         opt_tokens[i].pattern);
931                         }
932                 }
933
934                 return -EINVAL;
935         }
936
937         return 0;
938 }
939
940 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
941                 struct nvmf_ctrl_options *opts)
942 {
943         if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
944             strcmp(opts->traddr, ctrl->opts->traddr) ||
945             strcmp(opts->trsvcid, ctrl->opts->trsvcid))
946                 return false;
947
948         /*
949          * Checking the local address is rough. In most cases, none is specified
950          * and the host port is selected by the stack.
951          *
952          * Assume no match if:
953          * -  local address is specified and address is not the same
954          * -  local address is not specified but remote is, or vice versa
955          *    (admin using specific host_traddr when it matters).
956          */
957         if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
958             (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
959                 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
960                         return false;
961         } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
962                    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
963                 return false;
964         }
965
966         return true;
967 }
968 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
969
970 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
971                 unsigned int allowed_opts)
972 {
973         if (opts->mask & ~allowed_opts) {
974                 int i;
975
976                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
977                         if ((opt_tokens[i].token & opts->mask) &&
978                             (opt_tokens[i].token & ~allowed_opts)) {
979                                 pr_warn("invalid parameter '%s'\n",
980                                         opt_tokens[i].pattern);
981                         }
982                 }
983
984                 return -EINVAL;
985         }
986
987         return 0;
988 }
989
990 void nvmf_free_options(struct nvmf_ctrl_options *opts)
991 {
992         nvmf_host_put(opts->host);
993         kfree(opts->transport);
994         kfree(opts->traddr);
995         kfree(opts->trsvcid);
996         kfree(opts->subsysnqn);
997         kfree(opts->host_traddr);
998         kfree(opts);
999 }
1000 EXPORT_SYMBOL_GPL(nvmf_free_options);
1001
1002 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1003 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1004                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1005                                  NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1006                                  NVMF_OPT_DISABLE_SQFLOW |\
1007                                  NVMF_OPT_FAIL_FAST_TMO)
1008
1009 static struct nvme_ctrl *
1010 nvmf_create_ctrl(struct device *dev, const char *buf)
1011 {
1012         struct nvmf_ctrl_options *opts;
1013         struct nvmf_transport_ops *ops;
1014         struct nvme_ctrl *ctrl;
1015         int ret;
1016
1017         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1018         if (!opts)
1019                 return ERR_PTR(-ENOMEM);
1020
1021         ret = nvmf_parse_options(opts, buf);
1022         if (ret)
1023                 goto out_free_opts;
1024
1025
1026         request_module("nvme-%s", opts->transport);
1027
1028         /*
1029          * Check the generic options first as we need a valid transport for
1030          * the lookup below.  Then clear the generic flags so that transport
1031          * drivers don't have to care about them.
1032          */
1033         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1034         if (ret)
1035                 goto out_free_opts;
1036         opts->mask &= ~NVMF_REQUIRED_OPTS;
1037
1038         down_read(&nvmf_transports_rwsem);
1039         ops = nvmf_lookup_transport(opts);
1040         if (!ops) {
1041                 pr_info("no handler found for transport %s.\n",
1042                         opts->transport);
1043                 ret = -EINVAL;
1044                 goto out_unlock;
1045         }
1046
1047         if (!try_module_get(ops->module)) {
1048                 ret = -EBUSY;
1049                 goto out_unlock;
1050         }
1051         up_read(&nvmf_transports_rwsem);
1052
1053         ret = nvmf_check_required_opts(opts, ops->required_opts);
1054         if (ret)
1055                 goto out_module_put;
1056         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1057                                 ops->allowed_opts | ops->required_opts);
1058         if (ret)
1059                 goto out_module_put;
1060
1061         ctrl = ops->create_ctrl(dev, opts);
1062         if (IS_ERR(ctrl)) {
1063                 ret = PTR_ERR(ctrl);
1064                 goto out_module_put;
1065         }
1066
1067         module_put(ops->module);
1068         return ctrl;
1069
1070 out_module_put:
1071         module_put(ops->module);
1072         goto out_free_opts;
1073 out_unlock:
1074         up_read(&nvmf_transports_rwsem);
1075 out_free_opts:
1076         nvmf_free_options(opts);
1077         return ERR_PTR(ret);
1078 }
1079
1080 static struct class *nvmf_class;
1081 static struct device *nvmf_device;
1082 static DEFINE_MUTEX(nvmf_dev_mutex);
1083
1084 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1085                 size_t count, loff_t *pos)
1086 {
1087         struct seq_file *seq_file = file->private_data;
1088         struct nvme_ctrl *ctrl;
1089         const char *buf;
1090         int ret = 0;
1091
1092         if (count > PAGE_SIZE)
1093                 return -ENOMEM;
1094
1095         buf = memdup_user_nul(ubuf, count);
1096         if (IS_ERR(buf))
1097                 return PTR_ERR(buf);
1098
1099         mutex_lock(&nvmf_dev_mutex);
1100         if (seq_file->private) {
1101                 ret = -EINVAL;
1102                 goto out_unlock;
1103         }
1104
1105         ctrl = nvmf_create_ctrl(nvmf_device, buf);
1106         if (IS_ERR(ctrl)) {
1107                 ret = PTR_ERR(ctrl);
1108                 goto out_unlock;
1109         }
1110
1111         seq_file->private = ctrl;
1112
1113 out_unlock:
1114         mutex_unlock(&nvmf_dev_mutex);
1115         kfree(buf);
1116         return ret ? ret : count;
1117 }
1118
1119 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1120 {
1121         struct nvme_ctrl *ctrl;
1122         int ret = 0;
1123
1124         mutex_lock(&nvmf_dev_mutex);
1125         ctrl = seq_file->private;
1126         if (!ctrl) {
1127                 ret = -EINVAL;
1128                 goto out_unlock;
1129         }
1130
1131         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1132                         ctrl->instance, ctrl->cntlid);
1133
1134 out_unlock:
1135         mutex_unlock(&nvmf_dev_mutex);
1136         return ret;
1137 }
1138
1139 static int nvmf_dev_open(struct inode *inode, struct file *file)
1140 {
1141         /*
1142          * The miscdevice code initializes file->private_data, but doesn't
1143          * make use of it later.
1144          */
1145         file->private_data = NULL;
1146         return single_open(file, nvmf_dev_show, NULL);
1147 }
1148
1149 static int nvmf_dev_release(struct inode *inode, struct file *file)
1150 {
1151         struct seq_file *seq_file = file->private_data;
1152         struct nvme_ctrl *ctrl = seq_file->private;
1153
1154         if (ctrl)
1155                 nvme_put_ctrl(ctrl);
1156         return single_release(inode, file);
1157 }
1158
1159 static const struct file_operations nvmf_dev_fops = {
1160         .owner          = THIS_MODULE,
1161         .write          = nvmf_dev_write,
1162         .read           = seq_read,
1163         .open           = nvmf_dev_open,
1164         .release        = nvmf_dev_release,
1165 };
1166
1167 static struct miscdevice nvmf_misc = {
1168         .minor          = MISC_DYNAMIC_MINOR,
1169         .name           = "nvme-fabrics",
1170         .fops           = &nvmf_dev_fops,
1171 };
1172
1173 static int __init nvmf_init(void)
1174 {
1175         int ret;
1176
1177         nvmf_default_host = nvmf_host_default();
1178         if (!nvmf_default_host)
1179                 return -ENOMEM;
1180
1181         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1182         if (IS_ERR(nvmf_class)) {
1183                 pr_err("couldn't register class nvme-fabrics\n");
1184                 ret = PTR_ERR(nvmf_class);
1185                 goto out_free_host;
1186         }
1187
1188         nvmf_device =
1189                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1190         if (IS_ERR(nvmf_device)) {
1191                 pr_err("couldn't create nvme-fabris device!\n");
1192                 ret = PTR_ERR(nvmf_device);
1193                 goto out_destroy_class;
1194         }
1195
1196         ret = misc_register(&nvmf_misc);
1197         if (ret) {
1198                 pr_err("couldn't register misc device: %d\n", ret);
1199                 goto out_destroy_device;
1200         }
1201
1202         return 0;
1203
1204 out_destroy_device:
1205         device_destroy(nvmf_class, MKDEV(0, 0));
1206 out_destroy_class:
1207         class_destroy(nvmf_class);
1208 out_free_host:
1209         nvmf_host_put(nvmf_default_host);
1210         return ret;
1211 }
1212
1213 static void __exit nvmf_exit(void)
1214 {
1215         misc_deregister(&nvmf_misc);
1216         device_destroy(nvmf_class, MKDEV(0, 0));
1217         class_destroy(nvmf_class);
1218         nvmf_host_put(nvmf_default_host);
1219
1220         BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1221         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1222         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1223         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1224         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1225 }
1226
1227 MODULE_LICENSE("GPL v2");
1228
1229 module_init(nvmf_init);
1230 module_exit(nvmf_exit);