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 <generated/utsrelease.h>
17 #include <asm/unaligned.h>
20 u32 nvmet_get_log_page_len(struct nvme_command *cmd)
22 u32 len = le16_to_cpu(cmd->get_log_page.numdu);
25 len += le16_to_cpu(cmd->get_log_page.numdl);
26 /* NUMD is a 0's based value */
33 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req,
34 struct nvme_smart_log *slog)
38 u64 host_reads, host_writes, data_units_read, data_units_written;
40 status = NVME_SC_SUCCESS;
41 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid);
43 status = NVME_SC_INVALID_NS;
44 pr_err("nvmet : Could not find namespace id : %d\n",
45 le32_to_cpu(req->cmd->get_log_page.nsid));
49 host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]);
50 data_units_read = part_stat_read(ns->bdev->bd_part, sectors[READ]);
51 host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]);
52 data_units_written = part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
54 put_unaligned_le64(host_reads, &slog->host_reads[0]);
55 put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
56 put_unaligned_le64(host_writes, &slog->host_writes[0]);
57 put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
58 nvmet_put_namespace(ns);
63 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
64 struct nvme_smart_log *slog)
67 u64 host_reads = 0, host_writes = 0;
68 u64 data_units_read = 0, data_units_written = 0;
70 struct nvmet_ctrl *ctrl;
72 status = NVME_SC_SUCCESS;
76 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
77 host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]);
79 part_stat_read(ns->bdev->bd_part, sectors[READ]);
80 host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]);
82 part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
87 put_unaligned_le64(host_reads, &slog->host_reads[0]);
88 put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
89 put_unaligned_le64(host_writes, &slog->host_writes[0]);
90 put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
95 static u16 nvmet_get_smart_log(struct nvmet_req *req,
96 struct nvme_smart_log *slog)
100 WARN_ON(req == NULL || slog == NULL);
101 if (req->cmd->get_log_page.nsid == 0xFFFFFFFF)
102 status = nvmet_get_smart_log_all(req, slog);
104 status = nvmet_get_smart_log_nsid(req, slog);
108 static void nvmet_execute_get_log_page(struct nvmet_req *req)
110 struct nvme_smart_log *smart_log;
111 size_t data_len = nvmet_get_log_page_len(req->cmd);
115 buf = kzalloc(data_len, GFP_KERNEL);
117 status = NVME_SC_INTERNAL;
121 switch (req->cmd->get_log_page.lid) {
124 * We currently never set the More bit in the status field,
125 * so all error log entries are invalid and can be zeroed out.
126 * This is called a minum viable implementation (TM) of this
127 * mandatory log page.
132 * XXX: fill out actual smart log
134 * We might have a hard time coming up with useful values for
135 * many of the fields, and even when we have useful data
136 * available (e.g. units or commands read/written) those aren't
137 * persistent over power loss.
139 if (data_len != sizeof(*smart_log)) {
140 status = NVME_SC_INTERNAL;
144 status = nvmet_get_smart_log(req, smart_log);
146 memset(buf, '\0', data_len);
152 * We only support a single firmware slot which always is
153 * active, so we can zero out the whole firmware slot log and
154 * still claim to fully implement this mandatory log page.
161 status = nvmet_copy_to_sgl(req, 0, buf, data_len);
166 nvmet_req_complete(req, status);
169 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
171 struct nvmet_ctrl *ctrl = req->sq->ctrl;
172 struct nvme_id_ctrl *id;
175 id = kzalloc(sizeof(*id), GFP_KERNEL);
177 status = NVME_SC_INTERNAL;
181 /* XXX: figure out how to assign real vendors IDs. */
185 memset(id->sn, ' ', sizeof(id->sn));
186 snprintf(id->sn, sizeof(id->sn), "%llx", ctrl->serial);
188 memset(id->mn, ' ', sizeof(id->mn));
189 strncpy((char *)id->mn, "Linux", sizeof(id->mn));
191 memset(id->fr, ' ', sizeof(id->fr));
192 strncpy((char *)id->fr, UTS_RELEASE, sizeof(id->fr));
197 * XXX: figure out how we can assign a IEEE OUI, but until then
198 * the safest is to leave it as zeroes.
201 /* we support multiple ports and multiples hosts: */
202 id->cmic = (1 << 0) | (1 << 1);
204 /* no limit on data transfer sizes for now */
206 id->cntlid = cpu_to_le16(ctrl->cntlid);
207 id->ver = cpu_to_le32(ctrl->subsys->ver);
209 /* XXX: figure out what to do about RTD3R/RTD3 */
210 id->oaes = cpu_to_le32(1 << 8);
211 id->ctratt = cpu_to_le32(1 << 0);
216 * We don't really have a practical limit on the number of abort
217 * comands. But we don't do anything useful for abort either, so
218 * no point in allowing more abort commands than the spec requires.
222 id->aerl = NVMET_ASYNC_EVENTS - 1;
224 /* first slot is read-only, only one slot supported */
225 id->frmw = (1 << 0) | (1 << 1);
226 id->lpa = (1 << 0) | (1 << 2);
227 id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
230 /* We support keep-alive timeout in granularity of seconds */
231 id->kas = cpu_to_le16(NVMET_KAS);
233 id->sqes = (0x6 << 4) | 0x6;
234 id->cqes = (0x4 << 4) | 0x4;
236 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */
237 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
239 id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
240 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
241 NVME_CTRL_ONCS_WRITE_ZEROES);
243 /* XXX: don't report vwc if the underlying device is write through */
244 id->vwc = NVME_CTRL_VWC_PRESENT;
247 * We can't support atomic writes bigger than a LBA without support
248 * from the backend device.
253 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */
254 if (ctrl->ops->has_keyed_sgls)
255 id->sgls |= cpu_to_le32(1 << 2);
256 if (ctrl->ops->sqe_inline_size)
257 id->sgls |= cpu_to_le32(1 << 20);
259 strcpy(id->subnqn, ctrl->subsys->subsysnqn);
261 /* Max command capsule size is sqe + single page of in-capsule data */
262 id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) +
263 ctrl->ops->sqe_inline_size) / 16);
264 /* Max response capsule size is cqe */
265 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
267 id->msdbd = ctrl->ops->msdbd;
270 * Meh, we don't really support any power state. Fake up the same
271 * values that qemu does.
273 id->psd[0].max_power = cpu_to_le16(0x9c4);
274 id->psd[0].entry_lat = cpu_to_le32(0x10);
275 id->psd[0].exit_lat = cpu_to_le32(0x4);
277 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
281 nvmet_req_complete(req, status);
284 static void nvmet_execute_identify_ns(struct nvmet_req *req)
287 struct nvme_id_ns *id;
290 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
292 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
296 id = kzalloc(sizeof(*id), GFP_KERNEL);
298 status = NVME_SC_INTERNAL;
303 * nuse = ncap = nsze isn't aways true, but we have no way to find
304 * that out from the underlying device.
306 id->ncap = id->nuse = id->nsze =
307 cpu_to_le64(ns->size >> ns->blksize_shift);
310 * We just provide a single LBA format that matches what the
311 * underlying device reports.
317 * Our namespace might always be shared. Not just with other
318 * controllers, but also with any other user of the block device.
322 memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le));
324 id->lbaf[0].ds = ns->blksize_shift;
326 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
330 nvmet_put_namespace(ns);
332 nvmet_req_complete(req, status);
335 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
337 static const int buf_size = 4096;
338 struct nvmet_ctrl *ctrl = req->sq->ctrl;
340 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
345 list = kzalloc(buf_size, GFP_KERNEL);
347 status = NVME_SC_INTERNAL;
352 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
353 if (ns->nsid <= min_nsid)
355 list[i++] = cpu_to_le32(ns->nsid);
356 if (i == buf_size / sizeof(__le32))
361 status = nvmet_copy_to_sgl(req, 0, list, buf_size);
365 nvmet_req_complete(req, status);
369 * A "mimimum viable" abort implementation: the command is mandatory in the
370 * spec, but we are not required to do any useful work. We couldn't really
371 * do a useful abort, so don't bother even with waiting for the command
372 * to be exectuted and return immediately telling the command to abort
375 static void nvmet_execute_abort(struct nvmet_req *req)
377 nvmet_set_result(req, 1);
378 nvmet_req_complete(req, 0);
381 static void nvmet_execute_set_features(struct nvmet_req *req)
383 struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
384 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
388 switch (cdw10 & 0xf) {
389 case NVME_FEAT_NUM_QUEUES:
390 nvmet_set_result(req,
391 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
394 val32 = le32_to_cpu(req->cmd->common.cdw10[1]);
395 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
396 nvmet_set_result(req, req->sq->ctrl->kato);
399 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
403 nvmet_req_complete(req, status);
406 static void nvmet_execute_get_features(struct nvmet_req *req)
408 struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
409 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
412 switch (cdw10 & 0xf) {
414 * These features are mandatory in the spec, but we don't
415 * have a useful way to implement them. We'll eventually
416 * need to come up with some fake values for these.
419 case NVME_FEAT_ARBITRATION:
421 case NVME_FEAT_POWER_MGMT:
423 case NVME_FEAT_TEMP_THRESH:
425 case NVME_FEAT_ERR_RECOVERY:
427 case NVME_FEAT_IRQ_COALESCE:
429 case NVME_FEAT_IRQ_CONFIG:
431 case NVME_FEAT_WRITE_ATOMIC:
433 case NVME_FEAT_ASYNC_EVENT:
436 case NVME_FEAT_VOLATILE_WC:
437 nvmet_set_result(req, 1);
439 case NVME_FEAT_NUM_QUEUES:
440 nvmet_set_result(req,
441 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
444 nvmet_set_result(req, req->sq->ctrl->kato * 1000);
447 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
451 nvmet_req_complete(req, status);
454 static void nvmet_execute_async_event(struct nvmet_req *req)
456 struct nvmet_ctrl *ctrl = req->sq->ctrl;
458 mutex_lock(&ctrl->lock);
459 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
460 mutex_unlock(&ctrl->lock);
461 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
464 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
465 mutex_unlock(&ctrl->lock);
467 schedule_work(&ctrl->async_event_work);
470 static void nvmet_execute_keep_alive(struct nvmet_req *req)
472 struct nvmet_ctrl *ctrl = req->sq->ctrl;
474 pr_debug("ctrl %d update keep-alive timer for %d secs\n",
475 ctrl->cntlid, ctrl->kato);
477 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
478 nvmet_req_complete(req, 0);
481 int nvmet_parse_admin_cmd(struct nvmet_req *req)
483 struct nvme_command *cmd = req->cmd;
487 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
488 pr_err("nvmet: got admin cmd %d while CC.EN == 0\n",
490 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
492 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
493 pr_err("nvmet: got admin cmd %d while CSTS.RDY == 0\n",
495 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
498 switch (cmd->common.opcode) {
499 case nvme_admin_get_log_page:
500 req->data_len = nvmet_get_log_page_len(cmd);
502 switch (cmd->get_log_page.lid) {
506 req->execute = nvmet_execute_get_log_page;
510 case nvme_admin_identify:
511 req->data_len = 4096;
512 switch (cmd->identify.cns) {
514 req->execute = nvmet_execute_identify_ns;
516 case NVME_ID_CNS_CTRL:
517 req->execute = nvmet_execute_identify_ctrl;
519 case NVME_ID_CNS_NS_ACTIVE_LIST:
520 req->execute = nvmet_execute_identify_nslist;
524 case nvme_admin_abort_cmd:
525 req->execute = nvmet_execute_abort;
528 case nvme_admin_set_features:
529 req->execute = nvmet_execute_set_features;
532 case nvme_admin_get_features:
533 req->execute = nvmet_execute_get_features;
536 case nvme_admin_async_event:
537 req->execute = nvmet_execute_async_event;
540 case nvme_admin_keep_alive:
541 req->execute = nvmet_execute_keep_alive;
546 pr_err("nvmet: unhandled cmd %d\n", cmd->common.opcode);
547 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;