2 * NVMe admin command implementation.
3 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
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.
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
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/module.h>
16 #include <linux/rculist.h>
18 #include <generated/utsrelease.h>
19 #include <asm/unaligned.h>
22 u32
nvmet_get_log_page_len(struct nvme_command
*cmd
)
24 u32 len
= le16_to_cpu(cmd
->get_log_page
.numdu
);
27 len
+= le16_to_cpu(cmd
->get_log_page
.numdl
);
28 /* NUMD is a 0's based value */
35 static u16
nvmet_get_smart_log_nsid(struct nvmet_req
*req
,
36 struct nvme_smart_log
*slog
)
40 u64 host_reads
, host_writes
, data_units_read
, data_units_written
;
42 status
= NVME_SC_SUCCESS
;
43 ns
= nvmet_find_namespace(req
->sq
->ctrl
, req
->cmd
->get_log_page
.nsid
);
45 status
= NVME_SC_INVALID_NS
;
46 pr_err("nvmet : Could not find namespace id : %d\n",
47 le32_to_cpu(req
->cmd
->get_log_page
.nsid
));
51 host_reads
= part_stat_read(ns
->bdev
->bd_part
, ios
[READ
]);
52 data_units_read
= part_stat_read(ns
->bdev
->bd_part
, sectors
[READ
]);
53 host_writes
= part_stat_read(ns
->bdev
->bd_part
, ios
[WRITE
]);
54 data_units_written
= part_stat_read(ns
->bdev
->bd_part
, sectors
[WRITE
]);
56 put_unaligned_le64(host_reads
, &slog
->host_reads
[0]);
57 put_unaligned_le64(data_units_read
, &slog
->data_units_read
[0]);
58 put_unaligned_le64(host_writes
, &slog
->host_writes
[0]);
59 put_unaligned_le64(data_units_written
, &slog
->data_units_written
[0]);
60 nvmet_put_namespace(ns
);
65 static u16
nvmet_get_smart_log_all(struct nvmet_req
*req
,
66 struct nvme_smart_log
*slog
)
69 u64 host_reads
= 0, host_writes
= 0;
70 u64 data_units_read
= 0, data_units_written
= 0;
72 struct nvmet_ctrl
*ctrl
;
74 status
= NVME_SC_SUCCESS
;
78 list_for_each_entry_rcu(ns
, &ctrl
->subsys
->namespaces
, dev_link
) {
79 host_reads
+= part_stat_read(ns
->bdev
->bd_part
, ios
[READ
]);
81 part_stat_read(ns
->bdev
->bd_part
, sectors
[READ
]);
82 host_writes
+= part_stat_read(ns
->bdev
->bd_part
, ios
[WRITE
]);
84 part_stat_read(ns
->bdev
->bd_part
, sectors
[WRITE
]);
89 put_unaligned_le64(host_reads
, &slog
->host_reads
[0]);
90 put_unaligned_le64(data_units_read
, &slog
->data_units_read
[0]);
91 put_unaligned_le64(host_writes
, &slog
->host_writes
[0]);
92 put_unaligned_le64(data_units_written
, &slog
->data_units_written
[0]);
97 static u16
nvmet_get_smart_log(struct nvmet_req
*req
,
98 struct nvme_smart_log
*slog
)
102 WARN_ON(req
== NULL
|| slog
== NULL
);
103 if (req
->cmd
->get_log_page
.nsid
== cpu_to_le32(0xFFFFFFFF))
104 status
= nvmet_get_smart_log_all(req
, slog
);
106 status
= nvmet_get_smart_log_nsid(req
, slog
);
110 static void nvmet_execute_get_log_page(struct nvmet_req
*req
)
112 struct nvme_smart_log
*smart_log
;
113 size_t data_len
= nvmet_get_log_page_len(req
->cmd
);
117 buf
= kzalloc(data_len
, GFP_KERNEL
);
119 status
= NVME_SC_INTERNAL
;
123 switch (req
->cmd
->get_log_page
.lid
) {
126 * We currently never set the More bit in the status field,
127 * so all error log entries are invalid and can be zeroed out.
128 * This is called a minum viable implementation (TM) of this
129 * mandatory log page.
134 * XXX: fill out actual smart log
136 * We might have a hard time coming up with useful values for
137 * many of the fields, and even when we have useful data
138 * available (e.g. units or commands read/written) those aren't
139 * persistent over power loss.
141 if (data_len
!= sizeof(*smart_log
)) {
142 status
= NVME_SC_INTERNAL
;
146 status
= nvmet_get_smart_log(req
, smart_log
);
148 memset(buf
, '\0', data_len
);
154 * We only support a single firmware slot which always is
155 * active, so we can zero out the whole firmware slot log and
156 * still claim to fully implement this mandatory log page.
163 status
= nvmet_copy_to_sgl(req
, 0, buf
, data_len
);
168 nvmet_req_complete(req
, status
);
171 static void nvmet_execute_identify_ctrl(struct nvmet_req
*req
)
173 struct nvmet_ctrl
*ctrl
= req
->sq
->ctrl
;
174 struct nvme_id_ctrl
*id
;
177 id
= kzalloc(sizeof(*id
), GFP_KERNEL
);
179 status
= NVME_SC_INTERNAL
;
183 /* XXX: figure out how to assign real vendors IDs. */
187 memset(id
->sn
, ' ', sizeof(id
->sn
));
188 snprintf(id
->sn
, sizeof(id
->sn
), "%llx", ctrl
->serial
);
190 memset(id
->mn
, ' ', sizeof(id
->mn
));
191 strncpy((char *)id
->mn
, "Linux", sizeof(id
->mn
));
193 memset(id
->fr
, ' ', sizeof(id
->fr
));
194 strncpy((char *)id
->fr
, UTS_RELEASE
, sizeof(id
->fr
));
199 * XXX: figure out how we can assign a IEEE OUI, but until then
200 * the safest is to leave it as zeroes.
203 /* we support multiple ports and multiples hosts: */
204 id
->cmic
= (1 << 0) | (1 << 1);
206 /* no limit on data transfer sizes for now */
208 id
->cntlid
= cpu_to_le16(ctrl
->cntlid
);
209 id
->ver
= cpu_to_le32(ctrl
->subsys
->ver
);
211 /* XXX: figure out what to do about RTD3R/RTD3 */
212 id
->oaes
= cpu_to_le32(1 << 8);
213 id
->ctratt
= cpu_to_le32(1 << 0);
218 * We don't really have a practical limit on the number of abort
219 * comands. But we don't do anything useful for abort either, so
220 * no point in allowing more abort commands than the spec requires.
224 id
->aerl
= NVMET_ASYNC_EVENTS
- 1;
226 /* first slot is read-only, only one slot supported */
227 id
->frmw
= (1 << 0) | (1 << 1);
228 id
->lpa
= (1 << 0) | (1 << 2);
229 id
->elpe
= NVMET_ERROR_LOG_SLOTS
- 1;
232 /* We support keep-alive timeout in granularity of seconds */
233 id
->kas
= cpu_to_le16(NVMET_KAS
);
235 id
->sqes
= (0x6 << 4) | 0x6;
236 id
->cqes
= (0x4 << 4) | 0x4;
238 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */
239 id
->maxcmd
= cpu_to_le16(NVMET_MAX_CMD
);
241 id
->nn
= cpu_to_le32(ctrl
->subsys
->max_nsid
);
242 id
->oncs
= cpu_to_le16(NVME_CTRL_ONCS_DSM
|
243 NVME_CTRL_ONCS_WRITE_ZEROES
);
245 /* XXX: don't report vwc if the underlying device is write through */
246 id
->vwc
= NVME_CTRL_VWC_PRESENT
;
249 * We can't support atomic writes bigger than a LBA without support
250 * from the backend device.
255 id
->sgls
= cpu_to_le32(1 << 0); /* we always support SGLs */
256 if (ctrl
->ops
->has_keyed_sgls
)
257 id
->sgls
|= cpu_to_le32(1 << 2);
258 if (ctrl
->ops
->sqe_inline_size
)
259 id
->sgls
|= cpu_to_le32(1 << 20);
261 strcpy(id
->subnqn
, ctrl
->subsys
->subsysnqn
);
263 /* Max command capsule size is sqe + single page of in-capsule data */
264 id
->ioccsz
= cpu_to_le32((sizeof(struct nvme_command
) +
265 ctrl
->ops
->sqe_inline_size
) / 16);
266 /* Max response capsule size is cqe */
267 id
->iorcsz
= cpu_to_le32(sizeof(struct nvme_completion
) / 16);
269 id
->msdbd
= ctrl
->ops
->msdbd
;
272 * Meh, we don't really support any power state. Fake up the same
273 * values that qemu does.
275 id
->psd
[0].max_power
= cpu_to_le16(0x9c4);
276 id
->psd
[0].entry_lat
= cpu_to_le32(0x10);
277 id
->psd
[0].exit_lat
= cpu_to_le32(0x4);
279 status
= nvmet_copy_to_sgl(req
, 0, id
, sizeof(*id
));
283 nvmet_req_complete(req
, status
);
286 static void nvmet_execute_identify_ns(struct nvmet_req
*req
)
289 struct nvme_id_ns
*id
;
292 ns
= nvmet_find_namespace(req
->sq
->ctrl
, req
->cmd
->identify
.nsid
);
294 status
= NVME_SC_INVALID_NS
| NVME_SC_DNR
;
298 id
= kzalloc(sizeof(*id
), GFP_KERNEL
);
300 status
= NVME_SC_INTERNAL
;
305 * nuse = ncap = nsze isn't aways true, but we have no way to find
306 * that out from the underlying device.
308 id
->ncap
= id
->nuse
= id
->nsze
=
309 cpu_to_le64(ns
->size
>> ns
->blksize_shift
);
312 * We just provide a single LBA format that matches what the
313 * underlying device reports.
319 * Our namespace might always be shared. Not just with other
320 * controllers, but also with any other user of the block device.
324 memcpy(&id
->nguid
, &ns
->nguid
, sizeof(uuid_le
));
326 id
->lbaf
[0].ds
= ns
->blksize_shift
;
328 status
= nvmet_copy_to_sgl(req
, 0, id
, sizeof(*id
));
332 nvmet_put_namespace(ns
);
334 nvmet_req_complete(req
, status
);
337 static void nvmet_execute_identify_nslist(struct nvmet_req
*req
)
339 static const int buf_size
= 4096;
340 struct nvmet_ctrl
*ctrl
= req
->sq
->ctrl
;
342 u32 min_nsid
= le32_to_cpu(req
->cmd
->identify
.nsid
);
347 list
= kzalloc(buf_size
, GFP_KERNEL
);
349 status
= NVME_SC_INTERNAL
;
354 list_for_each_entry_rcu(ns
, &ctrl
->subsys
->namespaces
, dev_link
) {
355 if (ns
->nsid
<= min_nsid
)
357 list
[i
++] = cpu_to_le32(ns
->nsid
);
358 if (i
== buf_size
/ sizeof(__le32
))
363 status
= nvmet_copy_to_sgl(req
, 0, list
, buf_size
);
367 nvmet_req_complete(req
, status
);
371 * A "mimimum viable" abort implementation: the command is mandatory in the
372 * spec, but we are not required to do any useful work. We couldn't really
373 * do a useful abort, so don't bother even with waiting for the command
374 * to be exectuted and return immediately telling the command to abort
377 static void nvmet_execute_abort(struct nvmet_req
*req
)
379 nvmet_set_result(req
, 1);
380 nvmet_req_complete(req
, 0);
383 static void nvmet_execute_set_features(struct nvmet_req
*req
)
385 struct nvmet_subsys
*subsys
= req
->sq
->ctrl
->subsys
;
386 u32 cdw10
= le32_to_cpu(req
->cmd
->common
.cdw10
[0]);
390 switch (cdw10
& 0xf) {
391 case NVME_FEAT_NUM_QUEUES
:
392 nvmet_set_result(req
,
393 (subsys
->max_qid
- 1) | ((subsys
->max_qid
- 1) << 16));
396 val32
= le32_to_cpu(req
->cmd
->common
.cdw10
[1]);
397 req
->sq
->ctrl
->kato
= DIV_ROUND_UP(val32
, 1000);
398 nvmet_set_result(req
, req
->sq
->ctrl
->kato
);
401 status
= NVME_SC_INVALID_FIELD
| NVME_SC_DNR
;
405 nvmet_req_complete(req
, status
);
408 static void nvmet_execute_get_features(struct nvmet_req
*req
)
410 struct nvmet_subsys
*subsys
= req
->sq
->ctrl
->subsys
;
411 u32 cdw10
= le32_to_cpu(req
->cmd
->common
.cdw10
[0]);
414 switch (cdw10
& 0xf) {
416 * These features are mandatory in the spec, but we don't
417 * have a useful way to implement them. We'll eventually
418 * need to come up with some fake values for these.
421 case NVME_FEAT_ARBITRATION
:
423 case NVME_FEAT_POWER_MGMT
:
425 case NVME_FEAT_TEMP_THRESH
:
427 case NVME_FEAT_ERR_RECOVERY
:
429 case NVME_FEAT_IRQ_COALESCE
:
431 case NVME_FEAT_IRQ_CONFIG
:
433 case NVME_FEAT_WRITE_ATOMIC
:
435 case NVME_FEAT_ASYNC_EVENT
:
438 case NVME_FEAT_VOLATILE_WC
:
439 nvmet_set_result(req
, 1);
441 case NVME_FEAT_NUM_QUEUES
:
442 nvmet_set_result(req
,
443 (subsys
->max_qid
-1) | ((subsys
->max_qid
-1) << 16));
446 nvmet_set_result(req
, req
->sq
->ctrl
->kato
* 1000);
449 status
= NVME_SC_INVALID_FIELD
| NVME_SC_DNR
;
453 nvmet_req_complete(req
, status
);
456 static void nvmet_execute_async_event(struct nvmet_req
*req
)
458 struct nvmet_ctrl
*ctrl
= req
->sq
->ctrl
;
460 mutex_lock(&ctrl
->lock
);
461 if (ctrl
->nr_async_event_cmds
>= NVMET_ASYNC_EVENTS
) {
462 mutex_unlock(&ctrl
->lock
);
463 nvmet_req_complete(req
, NVME_SC_ASYNC_LIMIT
| NVME_SC_DNR
);
466 ctrl
->async_event_cmds
[ctrl
->nr_async_event_cmds
++] = req
;
467 mutex_unlock(&ctrl
->lock
);
469 schedule_work(&ctrl
->async_event_work
);
472 static void nvmet_execute_keep_alive(struct nvmet_req
*req
)
474 struct nvmet_ctrl
*ctrl
= req
->sq
->ctrl
;
476 pr_debug("ctrl %d update keep-alive timer for %d secs\n",
477 ctrl
->cntlid
, ctrl
->kato
);
479 mod_delayed_work(system_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
);
480 nvmet_req_complete(req
, 0);
483 int nvmet_parse_admin_cmd(struct nvmet_req
*req
)
485 struct nvme_command
*cmd
= req
->cmd
;
489 if (unlikely(!(req
->sq
->ctrl
->cc
& NVME_CC_ENABLE
))) {
490 pr_err("nvmet: got admin cmd %d while CC.EN == 0\n",
492 return NVME_SC_CMD_SEQ_ERROR
| NVME_SC_DNR
;
494 if (unlikely(!(req
->sq
->ctrl
->csts
& NVME_CSTS_RDY
))) {
495 pr_err("nvmet: got admin cmd %d while CSTS.RDY == 0\n",
497 return NVME_SC_CMD_SEQ_ERROR
| NVME_SC_DNR
;
500 switch (cmd
->common
.opcode
) {
501 case nvme_admin_get_log_page
:
502 req
->data_len
= nvmet_get_log_page_len(cmd
);
504 switch (cmd
->get_log_page
.lid
) {
508 req
->execute
= nvmet_execute_get_log_page
;
512 case nvme_admin_identify
:
513 req
->data_len
= 4096;
514 switch (cmd
->identify
.cns
) {
516 req
->execute
= nvmet_execute_identify_ns
;
518 case NVME_ID_CNS_CTRL
:
519 req
->execute
= nvmet_execute_identify_ctrl
;
521 case NVME_ID_CNS_NS_ACTIVE_LIST
:
522 req
->execute
= nvmet_execute_identify_nslist
;
526 case nvme_admin_abort_cmd
:
527 req
->execute
= nvmet_execute_abort
;
530 case nvme_admin_set_features
:
531 req
->execute
= nvmet_execute_set_features
;
534 case nvme_admin_get_features
:
535 req
->execute
= nvmet_execute_get_features
;
538 case nvme_admin_async_event
:
539 req
->execute
= nvmet_execute_async_event
;
542 case nvme_admin_keep_alive
:
543 req
->execute
= nvmet_execute_keep_alive
;
548 pr_err("nvmet: unhandled cmd %d\n", cmd
->common
.opcode
);
549 return NVME_SC_INVALID_OPCODE
| NVME_SC_DNR
;