Christoph Hellwig | a07b497 | 2016-06-21 18:04:20 +0200 | [diff] [blame] | 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/random.h> |
| 17 | #include <generated/utsrelease.h> |
| 18 | #include "nvmet.h" |
| 19 | |
| 20 | u32 nvmet_get_log_page_len(struct nvme_command *cmd) |
| 21 | { |
| 22 | u32 len = le16_to_cpu(cmd->get_log_page.numdu); |
| 23 | |
| 24 | len <<= 16; |
| 25 | len += le16_to_cpu(cmd->get_log_page.numdl); |
| 26 | /* NUMD is a 0's based value */ |
| 27 | len += 1; |
| 28 | len *= sizeof(u32); |
| 29 | |
| 30 | return len; |
| 31 | } |
| 32 | |
| 33 | static void nvmet_execute_get_log_page(struct nvmet_req *req) |
| 34 | { |
| 35 | size_t data_len = nvmet_get_log_page_len(req->cmd); |
| 36 | void *buf; |
| 37 | u16 status = 0; |
| 38 | |
| 39 | buf = kzalloc(data_len, GFP_KERNEL); |
| 40 | if (!buf) { |
| 41 | status = NVME_SC_INTERNAL; |
| 42 | goto out; |
| 43 | } |
| 44 | |
| 45 | switch (req->cmd->get_log_page.lid) { |
| 46 | case 0x01: |
| 47 | /* |
| 48 | * We currently never set the More bit in the status field, |
| 49 | * so all error log entries are invalid and can be zeroed out. |
| 50 | * This is called a minum viable implementation (TM) of this |
| 51 | * mandatory log page. |
| 52 | */ |
| 53 | break; |
| 54 | case 0x02: |
| 55 | /* |
| 56 | * XXX: fill out actual smart log |
| 57 | * |
| 58 | * We might have a hard time coming up with useful values for |
| 59 | * many of the fields, and even when we have useful data |
| 60 | * available (e.g. units or commands read/written) those aren't |
| 61 | * persistent over power loss. |
| 62 | */ |
| 63 | break; |
| 64 | case 0x03: |
| 65 | /* |
| 66 | * We only support a single firmware slot which always is |
| 67 | * active, so we can zero out the whole firmware slot log and |
| 68 | * still claim to fully implement this mandatory log page. |
| 69 | */ |
| 70 | break; |
| 71 | default: |
| 72 | BUG(); |
| 73 | } |
| 74 | |
| 75 | status = nvmet_copy_to_sgl(req, 0, buf, data_len); |
| 76 | |
| 77 | kfree(buf); |
| 78 | out: |
| 79 | nvmet_req_complete(req, status); |
| 80 | } |
| 81 | |
| 82 | static void nvmet_execute_identify_ctrl(struct nvmet_req *req) |
| 83 | { |
| 84 | struct nvmet_ctrl *ctrl = req->sq->ctrl; |
| 85 | struct nvme_id_ctrl *id; |
| 86 | u64 serial; |
| 87 | u16 status = 0; |
| 88 | |
| 89 | id = kzalloc(sizeof(*id), GFP_KERNEL); |
| 90 | if (!id) { |
| 91 | status = NVME_SC_INTERNAL; |
| 92 | goto out; |
| 93 | } |
| 94 | |
| 95 | /* XXX: figure out how to assign real vendors IDs. */ |
| 96 | id->vid = 0; |
| 97 | id->ssvid = 0; |
| 98 | |
| 99 | /* generate a random serial number as our controllers are ephemeral: */ |
| 100 | get_random_bytes(&serial, sizeof(serial)); |
| 101 | memset(id->sn, ' ', sizeof(id->sn)); |
| 102 | snprintf(id->sn, sizeof(id->sn), "%llx", serial); |
| 103 | |
| 104 | memset(id->mn, ' ', sizeof(id->mn)); |
| 105 | strncpy((char *)id->mn, "Linux", sizeof(id->mn)); |
| 106 | |
| 107 | memset(id->fr, ' ', sizeof(id->fr)); |
| 108 | strncpy((char *)id->fr, UTS_RELEASE, sizeof(id->fr)); |
| 109 | |
| 110 | id->rab = 6; |
| 111 | |
| 112 | /* |
| 113 | * XXX: figure out how we can assign a IEEE OUI, but until then |
| 114 | * the safest is to leave it as zeroes. |
| 115 | */ |
| 116 | |
| 117 | /* we support multiple ports and multiples hosts: */ |
| 118 | id->mic = (1 << 0) | (1 << 1); |
| 119 | |
| 120 | /* no limit on data transfer sizes for now */ |
| 121 | id->mdts = 0; |
| 122 | id->cntlid = cpu_to_le16(ctrl->cntlid); |
| 123 | id->ver = cpu_to_le32(ctrl->subsys->ver); |
| 124 | |
| 125 | /* XXX: figure out what to do about RTD3R/RTD3 */ |
| 126 | id->oaes = cpu_to_le32(1 << 8); |
| 127 | id->ctratt = cpu_to_le32(1 << 0); |
| 128 | |
| 129 | id->oacs = 0; |
| 130 | |
| 131 | /* |
| 132 | * We don't really have a practical limit on the number of abort |
| 133 | * comands. But we don't do anything useful for abort either, so |
| 134 | * no point in allowing more abort commands than the spec requires. |
| 135 | */ |
| 136 | id->acl = 3; |
| 137 | |
| 138 | id->aerl = NVMET_ASYNC_EVENTS - 1; |
| 139 | |
| 140 | /* first slot is read-only, only one slot supported */ |
| 141 | id->frmw = (1 << 0) | (1 << 1); |
| 142 | id->lpa = (1 << 0) | (1 << 2); |
| 143 | id->elpe = NVMET_ERROR_LOG_SLOTS - 1; |
| 144 | id->npss = 0; |
| 145 | |
| 146 | /* We support keep-alive timeout in granularity of seconds */ |
| 147 | id->kas = cpu_to_le16(NVMET_KAS); |
| 148 | |
| 149 | id->sqes = (0x6 << 4) | 0x6; |
| 150 | id->cqes = (0x4 << 4) | 0x4; |
| 151 | |
| 152 | /* no enforcement soft-limit for maxcmd - pick arbitrary high value */ |
| 153 | id->maxcmd = cpu_to_le16(NVMET_MAX_CMD); |
| 154 | |
| 155 | id->nn = cpu_to_le32(ctrl->subsys->max_nsid); |
| 156 | id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM); |
| 157 | |
| 158 | /* XXX: don't report vwc if the underlying device is write through */ |
| 159 | id->vwc = NVME_CTRL_VWC_PRESENT; |
| 160 | |
| 161 | /* |
| 162 | * We can't support atomic writes bigger than a LBA without support |
| 163 | * from the backend device. |
| 164 | */ |
| 165 | id->awun = 0; |
| 166 | id->awupf = 0; |
| 167 | |
| 168 | id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */ |
| 169 | if (ctrl->ops->has_keyed_sgls) |
| 170 | id->sgls |= cpu_to_le32(1 << 2); |
| 171 | if (ctrl->ops->sqe_inline_size) |
| 172 | id->sgls |= cpu_to_le32(1 << 20); |
| 173 | |
| 174 | strcpy(id->subnqn, ctrl->subsys->subsysnqn); |
| 175 | |
| 176 | /* Max command capsule size is sqe + single page of in-capsule data */ |
| 177 | id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) + |
| 178 | ctrl->ops->sqe_inline_size) / 16); |
| 179 | /* Max response capsule size is cqe */ |
| 180 | id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16); |
| 181 | |
| 182 | id->msdbd = ctrl->ops->msdbd; |
| 183 | |
| 184 | /* |
| 185 | * Meh, we don't really support any power state. Fake up the same |
| 186 | * values that qemu does. |
| 187 | */ |
| 188 | id->psd[0].max_power = cpu_to_le16(0x9c4); |
| 189 | id->psd[0].entry_lat = cpu_to_le32(0x10); |
| 190 | id->psd[0].exit_lat = cpu_to_le32(0x4); |
| 191 | |
| 192 | status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); |
| 193 | |
| 194 | kfree(id); |
| 195 | out: |
| 196 | nvmet_req_complete(req, status); |
| 197 | } |
| 198 | |
| 199 | static void nvmet_execute_identify_ns(struct nvmet_req *req) |
| 200 | { |
| 201 | struct nvmet_ns *ns; |
| 202 | struct nvme_id_ns *id; |
| 203 | u16 status = 0; |
| 204 | |
| 205 | ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); |
| 206 | if (!ns) { |
| 207 | status = NVME_SC_INVALID_NS | NVME_SC_DNR; |
| 208 | goto out; |
| 209 | } |
| 210 | |
| 211 | id = kzalloc(sizeof(*id), GFP_KERNEL); |
| 212 | if (!id) { |
| 213 | status = NVME_SC_INTERNAL; |
| 214 | goto out_put_ns; |
| 215 | } |
| 216 | |
| 217 | /* |
| 218 | * nuse = ncap = nsze isn't aways true, but we have no way to find |
| 219 | * that out from the underlying device. |
| 220 | */ |
| 221 | id->ncap = id->nuse = id->nsze = |
| 222 | cpu_to_le64(ns->size >> ns->blksize_shift); |
| 223 | |
| 224 | /* |
| 225 | * We just provide a single LBA format that matches what the |
| 226 | * underlying device reports. |
| 227 | */ |
| 228 | id->nlbaf = 0; |
| 229 | id->flbas = 0; |
| 230 | |
| 231 | /* |
| 232 | * Our namespace might always be shared. Not just with other |
| 233 | * controllers, but also with any other user of the block device. |
| 234 | */ |
| 235 | id->nmic = (1 << 0); |
| 236 | |
| 237 | memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le)); |
| 238 | |
| 239 | id->lbaf[0].ds = ns->blksize_shift; |
| 240 | |
| 241 | status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); |
| 242 | |
| 243 | kfree(id); |
| 244 | out_put_ns: |
| 245 | nvmet_put_namespace(ns); |
| 246 | out: |
| 247 | nvmet_req_complete(req, status); |
| 248 | } |
| 249 | |
| 250 | static void nvmet_execute_identify_nslist(struct nvmet_req *req) |
| 251 | { |
| 252 | static const int buf_size = 4096; |
| 253 | struct nvmet_ctrl *ctrl = req->sq->ctrl; |
| 254 | struct nvmet_ns *ns; |
| 255 | u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid); |
| 256 | __le32 *list; |
| 257 | u16 status = 0; |
| 258 | int i = 0; |
| 259 | |
| 260 | list = kzalloc(buf_size, GFP_KERNEL); |
| 261 | if (!list) { |
| 262 | status = NVME_SC_INTERNAL; |
| 263 | goto out; |
| 264 | } |
| 265 | |
| 266 | rcu_read_lock(); |
| 267 | list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { |
| 268 | if (ns->nsid <= min_nsid) |
| 269 | continue; |
| 270 | list[i++] = cpu_to_le32(ns->nsid); |
| 271 | if (i == buf_size / sizeof(__le32)) |
| 272 | break; |
| 273 | } |
| 274 | rcu_read_unlock(); |
| 275 | |
| 276 | status = nvmet_copy_to_sgl(req, 0, list, buf_size); |
| 277 | |
| 278 | kfree(list); |
| 279 | out: |
| 280 | nvmet_req_complete(req, status); |
| 281 | } |
| 282 | |
| 283 | /* |
| 284 | * A "mimimum viable" abort implementation: the command is mandatory in the |
| 285 | * spec, but we are not required to do any useful work. We couldn't really |
| 286 | * do a useful abort, so don't bother even with waiting for the command |
| 287 | * to be exectuted and return immediately telling the command to abort |
| 288 | * wasn't found. |
| 289 | */ |
| 290 | static void nvmet_execute_abort(struct nvmet_req *req) |
| 291 | { |
| 292 | nvmet_set_result(req, 1); |
| 293 | nvmet_req_complete(req, 0); |
| 294 | } |
| 295 | |
| 296 | static void nvmet_execute_set_features(struct nvmet_req *req) |
| 297 | { |
| 298 | struct nvmet_subsys *subsys = req->sq->ctrl->subsys; |
| 299 | u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); |
| 300 | u64 val; |
| 301 | u32 val32; |
| 302 | u16 status = 0; |
| 303 | |
| 304 | switch (cdw10 & 0xf) { |
| 305 | case NVME_FEAT_NUM_QUEUES: |
| 306 | nvmet_set_result(req, |
| 307 | (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16)); |
| 308 | break; |
| 309 | case NVME_FEAT_KATO: |
| 310 | val = le64_to_cpu(req->cmd->prop_set.value); |
| 311 | val32 = val & 0xffff; |
| 312 | req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000); |
| 313 | nvmet_set_result(req, req->sq->ctrl->kato); |
| 314 | break; |
| 315 | default: |
| 316 | status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; |
| 317 | break; |
| 318 | } |
| 319 | |
| 320 | nvmet_req_complete(req, status); |
| 321 | } |
| 322 | |
| 323 | static void nvmet_execute_get_features(struct nvmet_req *req) |
| 324 | { |
| 325 | struct nvmet_subsys *subsys = req->sq->ctrl->subsys; |
| 326 | u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); |
| 327 | u16 status = 0; |
| 328 | |
| 329 | switch (cdw10 & 0xf) { |
| 330 | /* |
| 331 | * These features are mandatory in the spec, but we don't |
| 332 | * have a useful way to implement them. We'll eventually |
| 333 | * need to come up with some fake values for these. |
| 334 | */ |
| 335 | #if 0 |
| 336 | case NVME_FEAT_ARBITRATION: |
| 337 | break; |
| 338 | case NVME_FEAT_POWER_MGMT: |
| 339 | break; |
| 340 | case NVME_FEAT_TEMP_THRESH: |
| 341 | break; |
| 342 | case NVME_FEAT_ERR_RECOVERY: |
| 343 | break; |
| 344 | case NVME_FEAT_IRQ_COALESCE: |
| 345 | break; |
| 346 | case NVME_FEAT_IRQ_CONFIG: |
| 347 | break; |
| 348 | case NVME_FEAT_WRITE_ATOMIC: |
| 349 | break; |
| 350 | case NVME_FEAT_ASYNC_EVENT: |
| 351 | break; |
| 352 | #endif |
| 353 | case NVME_FEAT_VOLATILE_WC: |
| 354 | nvmet_set_result(req, 1); |
| 355 | break; |
| 356 | case NVME_FEAT_NUM_QUEUES: |
| 357 | nvmet_set_result(req, |
| 358 | (subsys->max_qid-1) | ((subsys->max_qid-1) << 16)); |
| 359 | break; |
| 360 | case NVME_FEAT_KATO: |
| 361 | nvmet_set_result(req, req->sq->ctrl->kato * 1000); |
| 362 | break; |
| 363 | default: |
| 364 | status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; |
| 365 | break; |
| 366 | } |
| 367 | |
| 368 | nvmet_req_complete(req, status); |
| 369 | } |
| 370 | |
| 371 | static void nvmet_execute_async_event(struct nvmet_req *req) |
| 372 | { |
| 373 | struct nvmet_ctrl *ctrl = req->sq->ctrl; |
| 374 | |
| 375 | mutex_lock(&ctrl->lock); |
| 376 | if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) { |
| 377 | mutex_unlock(&ctrl->lock); |
| 378 | nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR); |
| 379 | return; |
| 380 | } |
| 381 | ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req; |
| 382 | mutex_unlock(&ctrl->lock); |
| 383 | |
| 384 | schedule_work(&ctrl->async_event_work); |
| 385 | } |
| 386 | |
| 387 | static void nvmet_execute_keep_alive(struct nvmet_req *req) |
| 388 | { |
| 389 | struct nvmet_ctrl *ctrl = req->sq->ctrl; |
| 390 | |
| 391 | pr_debug("ctrl %d update keep-alive timer for %d secs\n", |
| 392 | ctrl->cntlid, ctrl->kato); |
| 393 | |
| 394 | mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); |
| 395 | nvmet_req_complete(req, 0); |
| 396 | } |
| 397 | |
| 398 | int nvmet_parse_admin_cmd(struct nvmet_req *req) |
| 399 | { |
| 400 | struct nvme_command *cmd = req->cmd; |
| 401 | |
| 402 | req->ns = NULL; |
| 403 | |
| 404 | if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) { |
| 405 | pr_err("nvmet: got admin cmd %d while CC.EN == 0\n", |
| 406 | cmd->common.opcode); |
| 407 | return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; |
| 408 | } |
| 409 | if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) { |
| 410 | pr_err("nvmet: got admin cmd %d while CSTS.RDY == 0\n", |
| 411 | cmd->common.opcode); |
| 412 | return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; |
| 413 | } |
| 414 | |
| 415 | switch (cmd->common.opcode) { |
| 416 | case nvme_admin_get_log_page: |
| 417 | req->data_len = nvmet_get_log_page_len(cmd); |
| 418 | |
| 419 | switch (cmd->get_log_page.lid) { |
| 420 | case 0x01: |
| 421 | case 0x02: |
| 422 | case 0x03: |
| 423 | req->execute = nvmet_execute_get_log_page; |
| 424 | return 0; |
| 425 | } |
| 426 | break; |
| 427 | case nvme_admin_identify: |
| 428 | req->data_len = 4096; |
| 429 | switch (le32_to_cpu(cmd->identify.cns)) { |
| 430 | case 0x00: |
| 431 | req->execute = nvmet_execute_identify_ns; |
| 432 | return 0; |
| 433 | case 0x01: |
| 434 | req->execute = nvmet_execute_identify_ctrl; |
| 435 | return 0; |
| 436 | case 0x02: |
| 437 | req->execute = nvmet_execute_identify_nslist; |
| 438 | return 0; |
| 439 | } |
| 440 | break; |
| 441 | case nvme_admin_abort_cmd: |
| 442 | req->execute = nvmet_execute_abort; |
| 443 | req->data_len = 0; |
| 444 | return 0; |
| 445 | case nvme_admin_set_features: |
| 446 | req->execute = nvmet_execute_set_features; |
| 447 | req->data_len = 0; |
| 448 | return 0; |
| 449 | case nvme_admin_get_features: |
| 450 | req->execute = nvmet_execute_get_features; |
| 451 | req->data_len = 0; |
| 452 | return 0; |
| 453 | case nvme_admin_async_event: |
| 454 | req->execute = nvmet_execute_async_event; |
| 455 | req->data_len = 0; |
| 456 | return 0; |
| 457 | case nvme_admin_keep_alive: |
| 458 | req->execute = nvmet_execute_keep_alive; |
| 459 | req->data_len = 0; |
| 460 | return 0; |
| 461 | } |
| 462 | |
| 463 | pr_err("nvmet: unhandled cmd %d\n", cmd->common.opcode); |
| 464 | return NVME_SC_INVALID_OPCODE | NVME_SC_DNR; |
| 465 | } |