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>
19 u32 nvmet_get_log_page_len(struct nvme_command *cmd)
21 u32 len = le16_to_cpu(cmd->get_log_page.numdu);
24 len += le16_to_cpu(cmd->get_log_page.numdl);
25 /* NUMD is a 0's based value */
32 static void nvmet_execute_get_log_page(struct nvmet_req *req)
34 size_t data_len = nvmet_get_log_page_len(req->cmd);
38 buf = kzalloc(data_len, GFP_KERNEL);
40 status = NVME_SC_INTERNAL;
44 switch (req->cmd->get_log_page.lid) {
47 * We currently never set the More bit in the status field,
48 * so all error log entries are invalid and can be zeroed out.
49 * This is called a minum viable implementation (TM) of this
55 * XXX: fill out actual smart log
57 * We might have a hard time coming up with useful values for
58 * many of the fields, and even when we have useful data
59 * available (e.g. units or commands read/written) those aren't
60 * persistent over power loss.
65 * We only support a single firmware slot which always is
66 * active, so we can zero out the whole firmware slot log and
67 * still claim to fully implement this mandatory log page.
74 status = nvmet_copy_to_sgl(req, 0, buf, data_len);
78 nvmet_req_complete(req, status);
81 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
83 struct nvmet_ctrl *ctrl = req->sq->ctrl;
84 struct nvme_id_ctrl *id;
87 id = kzalloc(sizeof(*id), GFP_KERNEL);
89 status = NVME_SC_INTERNAL;
93 /* XXX: figure out how to assign real vendors IDs. */
97 memset(id->sn, ' ', sizeof(id->sn));
98 snprintf(id->sn, sizeof(id->sn), "%llx", ctrl->serial);
100 memset(id->mn, ' ', sizeof(id->mn));
101 strncpy((char *)id->mn, "Linux", sizeof(id->mn));
103 memset(id->fr, ' ', sizeof(id->fr));
104 strncpy((char *)id->fr, UTS_RELEASE, sizeof(id->fr));
109 * XXX: figure out how we can assign a IEEE OUI, but until then
110 * the safest is to leave it as zeroes.
113 /* we support multiple ports and multiples hosts: */
114 id->mic = (1 << 0) | (1 << 1);
116 /* no limit on data transfer sizes for now */
118 id->cntlid = cpu_to_le16(ctrl->cntlid);
119 id->ver = cpu_to_le32(ctrl->subsys->ver);
121 /* XXX: figure out what to do about RTD3R/RTD3 */
122 id->oaes = cpu_to_le32(1 << 8);
123 id->ctratt = cpu_to_le32(1 << 0);
128 * We don't really have a practical limit on the number of abort
129 * comands. But we don't do anything useful for abort either, so
130 * no point in allowing more abort commands than the spec requires.
134 id->aerl = NVMET_ASYNC_EVENTS - 1;
136 /* first slot is read-only, only one slot supported */
137 id->frmw = (1 << 0) | (1 << 1);
138 id->lpa = (1 << 0) | (1 << 2);
139 id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
142 /* We support keep-alive timeout in granularity of seconds */
143 id->kas = cpu_to_le16(NVMET_KAS);
145 id->sqes = (0x6 << 4) | 0x6;
146 id->cqes = (0x4 << 4) | 0x4;
148 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */
149 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
151 id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
152 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM);
154 /* XXX: don't report vwc if the underlying device is write through */
155 id->vwc = NVME_CTRL_VWC_PRESENT;
158 * We can't support atomic writes bigger than a LBA without support
159 * from the backend device.
164 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */
165 if (ctrl->ops->has_keyed_sgls)
166 id->sgls |= cpu_to_le32(1 << 2);
167 if (ctrl->ops->sqe_inline_size)
168 id->sgls |= cpu_to_le32(1 << 20);
170 strcpy(id->subnqn, ctrl->subsys->subsysnqn);
172 /* Max command capsule size is sqe + single page of in-capsule data */
173 id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) +
174 ctrl->ops->sqe_inline_size) / 16);
175 /* Max response capsule size is cqe */
176 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
178 id->msdbd = ctrl->ops->msdbd;
181 * Meh, we don't really support any power state. Fake up the same
182 * values that qemu does.
184 id->psd[0].max_power = cpu_to_le16(0x9c4);
185 id->psd[0].entry_lat = cpu_to_le32(0x10);
186 id->psd[0].exit_lat = cpu_to_le32(0x4);
188 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
192 nvmet_req_complete(req, status);
195 static void nvmet_execute_identify_ns(struct nvmet_req *req)
198 struct nvme_id_ns *id;
201 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
203 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
207 id = kzalloc(sizeof(*id), GFP_KERNEL);
209 status = NVME_SC_INTERNAL;
214 * nuse = ncap = nsze isn't aways true, but we have no way to find
215 * that out from the underlying device.
217 id->ncap = id->nuse = id->nsze =
218 cpu_to_le64(ns->size >> ns->blksize_shift);
221 * We just provide a single LBA format that matches what the
222 * underlying device reports.
228 * Our namespace might always be shared. Not just with other
229 * controllers, but also with any other user of the block device.
233 memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le));
235 id->lbaf[0].ds = ns->blksize_shift;
237 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
241 nvmet_put_namespace(ns);
243 nvmet_req_complete(req, status);
246 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
248 static const int buf_size = 4096;
249 struct nvmet_ctrl *ctrl = req->sq->ctrl;
251 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
256 list = kzalloc(buf_size, GFP_KERNEL);
258 status = NVME_SC_INTERNAL;
263 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
264 if (ns->nsid <= min_nsid)
266 list[i++] = cpu_to_le32(ns->nsid);
267 if (i == buf_size / sizeof(__le32))
272 status = nvmet_copy_to_sgl(req, 0, list, buf_size);
276 nvmet_req_complete(req, status);
280 * A "mimimum viable" abort implementation: the command is mandatory in the
281 * spec, but we are not required to do any useful work. We couldn't really
282 * do a useful abort, so don't bother even with waiting for the command
283 * to be exectuted and return immediately telling the command to abort
286 static void nvmet_execute_abort(struct nvmet_req *req)
288 nvmet_set_result(req, 1);
289 nvmet_req_complete(req, 0);
292 static void nvmet_execute_set_features(struct nvmet_req *req)
294 struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
295 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
300 switch (cdw10 & 0xf) {
301 case NVME_FEAT_NUM_QUEUES:
302 nvmet_set_result(req,
303 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
306 val = le64_to_cpu(req->cmd->prop_set.value);
307 val32 = val & 0xffff;
308 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
309 nvmet_set_result(req, req->sq->ctrl->kato);
312 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
316 nvmet_req_complete(req, status);
319 static void nvmet_execute_get_features(struct nvmet_req *req)
321 struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
322 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
325 switch (cdw10 & 0xf) {
327 * These features are mandatory in the spec, but we don't
328 * have a useful way to implement them. We'll eventually
329 * need to come up with some fake values for these.
332 case NVME_FEAT_ARBITRATION:
334 case NVME_FEAT_POWER_MGMT:
336 case NVME_FEAT_TEMP_THRESH:
338 case NVME_FEAT_ERR_RECOVERY:
340 case NVME_FEAT_IRQ_COALESCE:
342 case NVME_FEAT_IRQ_CONFIG:
344 case NVME_FEAT_WRITE_ATOMIC:
346 case NVME_FEAT_ASYNC_EVENT:
349 case NVME_FEAT_VOLATILE_WC:
350 nvmet_set_result(req, 1);
352 case NVME_FEAT_NUM_QUEUES:
353 nvmet_set_result(req,
354 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
357 nvmet_set_result(req, req->sq->ctrl->kato * 1000);
360 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
364 nvmet_req_complete(req, status);
367 static void nvmet_execute_async_event(struct nvmet_req *req)
369 struct nvmet_ctrl *ctrl = req->sq->ctrl;
371 mutex_lock(&ctrl->lock);
372 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
373 mutex_unlock(&ctrl->lock);
374 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
377 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
378 mutex_unlock(&ctrl->lock);
380 schedule_work(&ctrl->async_event_work);
383 static void nvmet_execute_keep_alive(struct nvmet_req *req)
385 struct nvmet_ctrl *ctrl = req->sq->ctrl;
387 pr_debug("ctrl %d update keep-alive timer for %d secs\n",
388 ctrl->cntlid, ctrl->kato);
390 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
391 nvmet_req_complete(req, 0);
394 int nvmet_parse_admin_cmd(struct nvmet_req *req)
396 struct nvme_command *cmd = req->cmd;
400 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
401 pr_err("nvmet: got admin cmd %d while CC.EN == 0\n",
403 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
405 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
406 pr_err("nvmet: got admin cmd %d while CSTS.RDY == 0\n",
408 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
411 switch (cmd->common.opcode) {
412 case nvme_admin_get_log_page:
413 req->data_len = nvmet_get_log_page_len(cmd);
415 switch (cmd->get_log_page.lid) {
419 req->execute = nvmet_execute_get_log_page;
423 case nvme_admin_identify:
424 req->data_len = 4096;
425 switch (le32_to_cpu(cmd->identify.cns)) {
427 req->execute = nvmet_execute_identify_ns;
430 req->execute = nvmet_execute_identify_ctrl;
433 req->execute = nvmet_execute_identify_nslist;
437 case nvme_admin_abort_cmd:
438 req->execute = nvmet_execute_abort;
441 case nvme_admin_set_features:
442 req->execute = nvmet_execute_set_features;
445 case nvme_admin_get_features:
446 req->execute = nvmet_execute_get_features;
449 case nvme_admin_async_event:
450 req->execute = nvmet_execute_async_event;
453 case nvme_admin_keep_alive:
454 req->execute = nvmet_execute_keep_alive;
459 pr_err("nvmet: unhandled cmd %d\n", cmd->common.opcode);
460 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;