]> git.neil.brown.name Git - linux.git/blob - drivers/nvme/target/admin-cmd.c
nvmet: add AEN configuration support
[linux.git] / drivers / nvme / target / admin-cmd.c
1 /*
2  * NVMe admin command implementation.
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/module.h>
16 #include <linux/rculist.h>
17
18 #include <generated/utsrelease.h>
19 #include <asm/unaligned.h>
20 #include "nvmet.h"
21
22 u32 nvmet_get_log_page_len(struct nvme_command *cmd)
23 {
24         u32 len = le16_to_cpu(cmd->get_log_page.numdu);
25
26         len <<= 16;
27         len += le16_to_cpu(cmd->get_log_page.numdl);
28         /* NUMD is a 0's based value */
29         len += 1;
30         len *= sizeof(u32);
31
32         return len;
33 }
34
35 static void nvmet_execute_get_log_page_noop(struct nvmet_req *req)
36 {
37         nvmet_req_complete(req, nvmet_zero_sgl(req, 0, req->data_len));
38 }
39
40 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req,
41                 struct nvme_smart_log *slog)
42 {
43         struct nvmet_ns *ns;
44         u64 host_reads, host_writes, data_units_read, data_units_written;
45
46         ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid);
47         if (!ns) {
48                 pr_err("nvmet : Could not find namespace id : %d\n",
49                                 le32_to_cpu(req->cmd->get_log_page.nsid));
50                 return NVME_SC_INVALID_NS;
51         }
52
53         /* we don't have the right data for file backed ns */
54         if (!ns->bdev)
55                 goto out;
56
57         host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]);
58         data_units_read = part_stat_read(ns->bdev->bd_part, sectors[READ]);
59         host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]);
60         data_units_written = part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
61
62         put_unaligned_le64(host_reads, &slog->host_reads[0]);
63         put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
64         put_unaligned_le64(host_writes, &slog->host_writes[0]);
65         put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
66 out:
67         nvmet_put_namespace(ns);
68
69         return NVME_SC_SUCCESS;
70 }
71
72 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
73                 struct nvme_smart_log *slog)
74 {
75         u64 host_reads = 0, host_writes = 0;
76         u64 data_units_read = 0, data_units_written = 0;
77         struct nvmet_ns *ns;
78         struct nvmet_ctrl *ctrl;
79
80         ctrl = req->sq->ctrl;
81
82         rcu_read_lock();
83         list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
84                 /* we don't have the right data for file backed ns */
85                 if (!ns->bdev)
86                         continue;
87                 host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]);
88                 data_units_read +=
89                         part_stat_read(ns->bdev->bd_part, sectors[READ]);
90                 host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]);
91                 data_units_written +=
92                         part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
93
94         }
95         rcu_read_unlock();
96
97         put_unaligned_le64(host_reads, &slog->host_reads[0]);
98         put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
99         put_unaligned_le64(host_writes, &slog->host_writes[0]);
100         put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
101
102         return NVME_SC_SUCCESS;
103 }
104
105 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req)
106 {
107         struct nvme_smart_log *log;
108         u16 status = NVME_SC_INTERNAL;
109
110         if (req->data_len != sizeof(*log))
111                 goto out;
112
113         log = kzalloc(sizeof(*log), GFP_KERNEL);
114         if (!log)
115                 goto out;
116
117         if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL))
118                 status = nvmet_get_smart_log_all(req, log);
119         else
120                 status = nvmet_get_smart_log_nsid(req, log);
121         if (status)
122                 goto out;
123
124         status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
125 out:
126         nvmet_req_complete(req, status);
127 }
128
129 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
130 {
131         struct nvmet_ctrl *ctrl = req->sq->ctrl;
132         u16 status = NVME_SC_INTERNAL;
133         size_t len;
134
135         if (req->data_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
136                 goto out;
137
138         mutex_lock(&ctrl->lock);
139         if (ctrl->nr_changed_ns == U32_MAX)
140                 len = sizeof(__le32);
141         else
142                 len = ctrl->nr_changed_ns * sizeof(__le32);
143         status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
144         if (!status)
145                 status = nvmet_zero_sgl(req, len, req->data_len - len);
146         ctrl->nr_changed_ns = 0;
147         mutex_unlock(&ctrl->lock);
148 out:
149         nvmet_req_complete(req, status);
150 }
151
152 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
153 {
154         struct nvmet_ctrl *ctrl = req->sq->ctrl;
155         struct nvme_id_ctrl *id;
156         u16 status = 0;
157         const char model[] = "Linux";
158
159         id = kzalloc(sizeof(*id), GFP_KERNEL);
160         if (!id) {
161                 status = NVME_SC_INTERNAL;
162                 goto out;
163         }
164
165         /* XXX: figure out how to assign real vendors IDs. */
166         id->vid = 0;
167         id->ssvid = 0;
168
169         memset(id->sn, ' ', sizeof(id->sn));
170         bin2hex(id->sn, &ctrl->subsys->serial,
171                 min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
172         memcpy_and_pad(id->mn, sizeof(id->mn), model, sizeof(model) - 1, ' ');
173         memcpy_and_pad(id->fr, sizeof(id->fr),
174                        UTS_RELEASE, strlen(UTS_RELEASE), ' ');
175
176         id->rab = 6;
177
178         /*
179          * XXX: figure out how we can assign a IEEE OUI, but until then
180          * the safest is to leave it as zeroes.
181          */
182
183         /* we support multiple ports and multiples hosts: */
184         id->cmic = (1 << 0) | (1 << 1);
185
186         /* no limit on data transfer sizes for now */
187         id->mdts = 0;
188         id->cntlid = cpu_to_le16(ctrl->cntlid);
189         id->ver = cpu_to_le32(ctrl->subsys->ver);
190
191         /* XXX: figure out what to do about RTD3R/RTD3 */
192         id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
193         id->ctratt = cpu_to_le32(1 << 0);
194
195         id->oacs = 0;
196
197         /*
198          * We don't really have a practical limit on the number of abort
199          * comands.  But we don't do anything useful for abort either, so
200          * no point in allowing more abort commands than the spec requires.
201          */
202         id->acl = 3;
203
204         id->aerl = NVMET_ASYNC_EVENTS - 1;
205
206         /* first slot is read-only, only one slot supported */
207         id->frmw = (1 << 0) | (1 << 1);
208         id->lpa = (1 << 0) | (1 << 2);
209         id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
210         id->npss = 0;
211
212         /* We support keep-alive timeout in granularity of seconds */
213         id->kas = cpu_to_le16(NVMET_KAS);
214
215         id->sqes = (0x6 << 4) | 0x6;
216         id->cqes = (0x4 << 4) | 0x4;
217
218         /* no enforcement soft-limit for maxcmd - pick arbitrary high value */
219         id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
220
221         id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
222         id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
223                         NVME_CTRL_ONCS_WRITE_ZEROES);
224
225         /* XXX: don't report vwc if the underlying device is write through */
226         id->vwc = NVME_CTRL_VWC_PRESENT;
227
228         /*
229          * We can't support atomic writes bigger than a LBA without support
230          * from the backend device.
231          */
232         id->awun = 0;
233         id->awupf = 0;
234
235         id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */
236         if (ctrl->ops->has_keyed_sgls)
237                 id->sgls |= cpu_to_le32(1 << 2);
238         if (ctrl->ops->sqe_inline_size)
239                 id->sgls |= cpu_to_le32(1 << 20);
240
241         strcpy(id->subnqn, ctrl->subsys->subsysnqn);
242
243         /* Max command capsule size is sqe + single page of in-capsule data */
244         id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) +
245                                   ctrl->ops->sqe_inline_size) / 16);
246         /* Max response capsule size is cqe */
247         id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
248
249         id->msdbd = ctrl->ops->msdbd;
250
251         /*
252          * Meh, we don't really support any power state.  Fake up the same
253          * values that qemu does.
254          */
255         id->psd[0].max_power = cpu_to_le16(0x9c4);
256         id->psd[0].entry_lat = cpu_to_le32(0x10);
257         id->psd[0].exit_lat = cpu_to_le32(0x4);
258
259         status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
260
261         kfree(id);
262 out:
263         nvmet_req_complete(req, status);
264 }
265
266 static void nvmet_execute_identify_ns(struct nvmet_req *req)
267 {
268         struct nvmet_ns *ns;
269         struct nvme_id_ns *id;
270         u16 status = 0;
271
272         ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
273         if (!ns) {
274                 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
275                 goto out;
276         }
277
278         id = kzalloc(sizeof(*id), GFP_KERNEL);
279         if (!id) {
280                 status = NVME_SC_INTERNAL;
281                 goto out_put_ns;
282         }
283
284         /*
285          * nuse = ncap = nsze isn't always true, but we have no way to find
286          * that out from the underlying device.
287          */
288         id->ncap = id->nuse = id->nsze =
289                 cpu_to_le64(ns->size >> ns->blksize_shift);
290
291         /*
292          * We just provide a single LBA format that matches what the
293          * underlying device reports.
294          */
295         id->nlbaf = 0;
296         id->flbas = 0;
297
298         /*
299          * Our namespace might always be shared.  Not just with other
300          * controllers, but also with any other user of the block device.
301          */
302         id->nmic = (1 << 0);
303
304         memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le));
305
306         id->lbaf[0].ds = ns->blksize_shift;
307
308         status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
309
310         kfree(id);
311 out_put_ns:
312         nvmet_put_namespace(ns);
313 out:
314         nvmet_req_complete(req, status);
315 }
316
317 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
318 {
319         static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
320         struct nvmet_ctrl *ctrl = req->sq->ctrl;
321         struct nvmet_ns *ns;
322         u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
323         __le32 *list;
324         u16 status = 0;
325         int i = 0;
326
327         list = kzalloc(buf_size, GFP_KERNEL);
328         if (!list) {
329                 status = NVME_SC_INTERNAL;
330                 goto out;
331         }
332
333         rcu_read_lock();
334         list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
335                 if (ns->nsid <= min_nsid)
336                         continue;
337                 list[i++] = cpu_to_le32(ns->nsid);
338                 if (i == buf_size / sizeof(__le32))
339                         break;
340         }
341         rcu_read_unlock();
342
343         status = nvmet_copy_to_sgl(req, 0, list, buf_size);
344
345         kfree(list);
346 out:
347         nvmet_req_complete(req, status);
348 }
349
350 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
351                                     void *id, off_t *off)
352 {
353         struct nvme_ns_id_desc desc = {
354                 .nidt = type,
355                 .nidl = len,
356         };
357         u16 status;
358
359         status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
360         if (status)
361                 return status;
362         *off += sizeof(desc);
363
364         status = nvmet_copy_to_sgl(req, *off, id, len);
365         if (status)
366                 return status;
367         *off += len;
368
369         return 0;
370 }
371
372 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
373 {
374         struct nvmet_ns *ns;
375         u16 status = 0;
376         off_t off = 0;
377
378         ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
379         if (!ns) {
380                 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
381                 goto out;
382         }
383
384         if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) {
385                 status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
386                                                   NVME_NIDT_UUID_LEN,
387                                                   &ns->uuid, &off);
388                 if (status)
389                         goto out_put_ns;
390         }
391         if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) {
392                 status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
393                                                   NVME_NIDT_NGUID_LEN,
394                                                   &ns->nguid, &off);
395                 if (status)
396                         goto out_put_ns;
397         }
398
399         if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
400                         off) != NVME_IDENTIFY_DATA_SIZE - off)
401                 status = NVME_SC_INTERNAL | NVME_SC_DNR;
402 out_put_ns:
403         nvmet_put_namespace(ns);
404 out:
405         nvmet_req_complete(req, status);
406 }
407
408 /*
409  * A "minimum viable" abort implementation: the command is mandatory in the
410  * spec, but we are not required to do any useful work.  We couldn't really
411  * do a useful abort, so don't bother even with waiting for the command
412  * to be exectuted and return immediately telling the command to abort
413  * wasn't found.
414  */
415 static void nvmet_execute_abort(struct nvmet_req *req)
416 {
417         nvmet_set_result(req, 1);
418         nvmet_req_complete(req, 0);
419 }
420
421 static void nvmet_execute_set_features(struct nvmet_req *req)
422 {
423         struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
424         u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
425         u32 val32;
426         u16 status = 0;
427
428         switch (cdw10 & 0xff) {
429         case NVME_FEAT_NUM_QUEUES:
430                 nvmet_set_result(req,
431                         (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
432                 break;
433         case NVME_FEAT_KATO:
434                 val32 = le32_to_cpu(req->cmd->common.cdw10[1]);
435                 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
436                 nvmet_set_result(req, req->sq->ctrl->kato);
437                 break;
438         case NVME_FEAT_ASYNC_EVENT:
439                 val32 = le32_to_cpu(req->cmd->common.cdw10[1]);
440                 if (val32 & ~NVMET_AEN_CFG_ALL) {
441                         status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
442                         break;
443                 }
444
445                 WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
446                 nvmet_set_result(req, val32);
447                 break;
448         case NVME_FEAT_HOST_ID:
449                 status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
450                 break;
451         default:
452                 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
453                 break;
454         }
455
456         nvmet_req_complete(req, status);
457 }
458
459 static void nvmet_execute_get_features(struct nvmet_req *req)
460 {
461         struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
462         u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
463         u16 status = 0;
464
465         switch (cdw10 & 0xff) {
466         /*
467          * These features are mandatory in the spec, but we don't
468          * have a useful way to implement them.  We'll eventually
469          * need to come up with some fake values for these.
470          */
471 #if 0
472         case NVME_FEAT_ARBITRATION:
473                 break;
474         case NVME_FEAT_POWER_MGMT:
475                 break;
476         case NVME_FEAT_TEMP_THRESH:
477                 break;
478         case NVME_FEAT_ERR_RECOVERY:
479                 break;
480         case NVME_FEAT_IRQ_COALESCE:
481                 break;
482         case NVME_FEAT_IRQ_CONFIG:
483                 break;
484         case NVME_FEAT_WRITE_ATOMIC:
485                 break;
486 #endif
487         case NVME_FEAT_ASYNC_EVENT:
488                 nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
489                 break;
490         case NVME_FEAT_VOLATILE_WC:
491                 nvmet_set_result(req, 1);
492                 break;
493         case NVME_FEAT_NUM_QUEUES:
494                 nvmet_set_result(req,
495                         (subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
496                 break;
497         case NVME_FEAT_KATO:
498                 nvmet_set_result(req, req->sq->ctrl->kato * 1000);
499                 break;
500         case NVME_FEAT_HOST_ID:
501                 /* need 128-bit host identifier flag */
502                 if (!(req->cmd->common.cdw10[1] & cpu_to_le32(1 << 0))) {
503                         status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
504                         break;
505                 }
506
507                 status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
508                                 sizeof(req->sq->ctrl->hostid));
509                 break;
510         default:
511                 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
512                 break;
513         }
514
515         nvmet_req_complete(req, status);
516 }
517
518 static void nvmet_execute_async_event(struct nvmet_req *req)
519 {
520         struct nvmet_ctrl *ctrl = req->sq->ctrl;
521
522         mutex_lock(&ctrl->lock);
523         if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
524                 mutex_unlock(&ctrl->lock);
525                 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
526                 return;
527         }
528         ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
529         mutex_unlock(&ctrl->lock);
530
531         schedule_work(&ctrl->async_event_work);
532 }
533
534 static void nvmet_execute_keep_alive(struct nvmet_req *req)
535 {
536         struct nvmet_ctrl *ctrl = req->sq->ctrl;
537
538         pr_debug("ctrl %d update keep-alive timer for %d secs\n",
539                 ctrl->cntlid, ctrl->kato);
540
541         mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
542         nvmet_req_complete(req, 0);
543 }
544
545 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
546 {
547         struct nvme_command *cmd = req->cmd;
548         u16 ret;
549
550         ret = nvmet_check_ctrl_status(req, cmd);
551         if (unlikely(ret))
552                 return ret;
553
554         switch (cmd->common.opcode) {
555         case nvme_admin_get_log_page:
556                 req->data_len = nvmet_get_log_page_len(cmd);
557
558                 switch (cmd->get_log_page.lid) {
559                 case NVME_LOG_ERROR:
560                         /*
561                          * We currently never set the More bit in the status
562                          * field, so all error log entries are invalid and can
563                          * be zeroed out.  This is called a minum viable
564                          * implementation (TM) of this mandatory log page.
565                          */
566                         req->execute = nvmet_execute_get_log_page_noop;
567                         return 0;
568                 case NVME_LOG_SMART:
569                         req->execute = nvmet_execute_get_log_page_smart;
570                         return 0;
571                 case NVME_LOG_FW_SLOT:
572                         /*
573                          * We only support a single firmware slot which always
574                          * is active, so we can zero out the whole firmware slot
575                          * log and still claim to fully implement this mandatory
576                          * log page.
577                          */
578                         req->execute = nvmet_execute_get_log_page_noop;
579                         return 0;
580                 case NVME_LOG_CHANGED_NS:
581                         req->execute = nvmet_execute_get_log_changed_ns;
582                         return 0;
583                 }
584                 break;
585         case nvme_admin_identify:
586                 req->data_len = NVME_IDENTIFY_DATA_SIZE;
587                 switch (cmd->identify.cns) {
588                 case NVME_ID_CNS_NS:
589                         req->execute = nvmet_execute_identify_ns;
590                         return 0;
591                 case NVME_ID_CNS_CTRL:
592                         req->execute = nvmet_execute_identify_ctrl;
593                         return 0;
594                 case NVME_ID_CNS_NS_ACTIVE_LIST:
595                         req->execute = nvmet_execute_identify_nslist;
596                         return 0;
597                 case NVME_ID_CNS_NS_DESC_LIST:
598                         req->execute = nvmet_execute_identify_desclist;
599                         return 0;
600                 }
601                 break;
602         case nvme_admin_abort_cmd:
603                 req->execute = nvmet_execute_abort;
604                 req->data_len = 0;
605                 return 0;
606         case nvme_admin_set_features:
607                 req->execute = nvmet_execute_set_features;
608                 req->data_len = 0;
609                 return 0;
610         case nvme_admin_get_features:
611                 req->execute = nvmet_execute_get_features;
612                 req->data_len = 0;
613                 return 0;
614         case nvme_admin_async_event:
615                 req->execute = nvmet_execute_async_event;
616                 req->data_len = 0;
617                 return 0;
618         case nvme_admin_keep_alive:
619                 req->execute = nvmet_execute_keep_alive;
620                 req->data_len = 0;
621                 return 0;
622         }
623
624         pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
625                req->sq->qid);
626         return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
627 }