| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Discovery service for the NVMe over Fabrics target. |
| 4 | * Copyright (C) 2016 Intel Corporation. All rights reserved. |
| 5 | */ |
| 6 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 7 | #include <linux/slab.h> |
| 8 | #include <generated/utsrelease.h> |
| 9 | #include "nvmet.h" |
| 10 | |
| 11 | struct nvmet_subsys *nvmet_disc_subsys; |
| 12 | |
| 13 | static u64 nvmet_genctr; |
| 14 | |
| 15 | static void __nvmet_disc_changed(struct nvmet_port *port, |
| 16 | struct nvmet_ctrl *ctrl) |
| 17 | { |
| 18 | if (ctrl->port != port) |
| 19 | return; |
| 20 | |
| 21 | if (nvmet_aen_bit_disabled(ctrl, bn: NVME_AEN_BIT_DISC_CHANGE)) |
| 22 | return; |
| 23 | |
| 24 | nvmet_add_async_event(ctrl, event_type: NVME_AER_NOTICE, |
| 25 | event_info: NVME_AER_NOTICE_DISC_CHANGED, log_page: NVME_LOG_DISC); |
| 26 | } |
| 27 | |
| 28 | void nvmet_port_disc_changed(struct nvmet_port *port, |
| 29 | struct nvmet_subsys *subsys) |
| 30 | { |
| 31 | struct nvmet_ctrl *ctrl; |
| 32 | |
| 33 | lockdep_assert_held(&nvmet_config_sem); |
| 34 | nvmet_genctr++; |
| 35 | |
| 36 | mutex_lock(&nvmet_disc_subsys->lock); |
| 37 | list_for_each_entry(ctrl, &nvmet_disc_subsys->ctrls, subsys_entry) { |
| 38 | if (subsys && !nvmet_host_allowed(subsys, hostnqn: ctrl->hostnqn)) |
| 39 | continue; |
| 40 | |
| 41 | __nvmet_disc_changed(port, ctrl); |
| 42 | } |
| 43 | mutex_unlock(lock: &nvmet_disc_subsys->lock); |
| 44 | |
| 45 | /* If transport can signal change, notify transport */ |
| 46 | if (port->tr_ops && port->tr_ops->discovery_chg) |
| 47 | port->tr_ops->discovery_chg(port); |
| 48 | } |
| 49 | |
| 50 | static void __nvmet_subsys_disc_changed(struct nvmet_port *port, |
| 51 | struct nvmet_subsys *subsys, |
| 52 | struct nvmet_host *host) |
| 53 | { |
| 54 | struct nvmet_ctrl *ctrl; |
| 55 | |
| 56 | mutex_lock(&nvmet_disc_subsys->lock); |
| 57 | list_for_each_entry(ctrl, &nvmet_disc_subsys->ctrls, subsys_entry) { |
| 58 | if (host && strcmp(nvmet_host_name(host), ctrl->hostnqn)) |
| 59 | continue; |
| 60 | |
| 61 | __nvmet_disc_changed(port, ctrl); |
| 62 | } |
| 63 | mutex_unlock(lock: &nvmet_disc_subsys->lock); |
| 64 | } |
| 65 | |
| 66 | void nvmet_subsys_disc_changed(struct nvmet_subsys *subsys, |
| 67 | struct nvmet_host *host) |
| 68 | { |
| 69 | struct nvmet_port *port; |
| 70 | struct nvmet_subsys_link *s; |
| 71 | |
| 72 | lockdep_assert_held(&nvmet_config_sem); |
| 73 | nvmet_genctr++; |
| 74 | |
| 75 | list_for_each_entry(port, nvmet_ports, global_entry) |
| 76 | list_for_each_entry(s, &port->subsystems, entry) { |
| 77 | if (s->subsys != subsys) |
| 78 | continue; |
| 79 | __nvmet_subsys_disc_changed(port, subsys, host); |
| 80 | } |
| 81 | } |
| 82 | |
| 83 | void nvmet_referral_enable(struct nvmet_port *parent, struct nvmet_port *port) |
| 84 | { |
| 85 | down_write(sem: &nvmet_config_sem); |
| 86 | if (list_empty(head: &port->entry)) { |
| 87 | list_add_tail(new: &port->entry, head: &parent->referrals); |
| 88 | port->enabled = true; |
| 89 | nvmet_port_disc_changed(port: parent, NULL); |
| 90 | } |
| 91 | up_write(sem: &nvmet_config_sem); |
| 92 | } |
| 93 | |
| 94 | void nvmet_referral_disable(struct nvmet_port *parent, struct nvmet_port *port) |
| 95 | { |
| 96 | down_write(sem: &nvmet_config_sem); |
| 97 | if (!list_empty(head: &port->entry)) { |
| 98 | port->enabled = false; |
| 99 | list_del_init(entry: &port->entry); |
| 100 | nvmet_port_disc_changed(port: parent, NULL); |
| 101 | } |
| 102 | up_write(sem: &nvmet_config_sem); |
| 103 | } |
| 104 | |
| 105 | static void nvmet_format_discovery_entry(struct nvmf_disc_rsp_page_hdr *hdr, |
| 106 | struct nvmet_port *port, char *subsys_nqn, char *traddr, |
| 107 | u8 type, u32 numrec) |
| 108 | { |
| 109 | struct nvmf_disc_rsp_page_entry *e = &hdr->entries[numrec]; |
| 110 | |
| 111 | e->trtype = port->disc_addr.trtype; |
| 112 | e->adrfam = port->disc_addr.adrfam; |
| 113 | e->treq = port->disc_addr.treq; |
| 114 | e->portid = port->disc_addr.portid; |
| 115 | /* we support only dynamic controllers */ |
| 116 | e->cntlid = cpu_to_le16(NVME_CNTLID_DYNAMIC); |
| 117 | e->asqsz = cpu_to_le16(NVME_AQ_DEPTH); |
| 118 | e->subtype = type; |
| 119 | memcpy(e->trsvcid, port->disc_addr.trsvcid, NVMF_TRSVCID_SIZE); |
| 120 | memcpy(e->traddr, traddr, NVMF_TRADDR_SIZE); |
| 121 | memcpy(e->tsas.common, port->disc_addr.tsas.common, NVMF_TSAS_SIZE); |
| 122 | strscpy(e->subnqn, subsys_nqn, NVMF_NQN_SIZE); |
| 123 | } |
| 124 | |
| 125 | /* |
| 126 | * nvmet_set_disc_traddr - set a correct discovery log entry traddr |
| 127 | * |
| 128 | * IP based transports (e.g RDMA) can listen on "any" ipv4/ipv6 addresses |
| 129 | * (INADDR_ANY or IN6ADDR_ANY_INIT). The discovery log page traddr reply |
| 130 | * must not contain that "any" IP address. If the transport implements |
| 131 | * .disc_traddr, use it. this callback will set the discovery traddr |
| 132 | * from the req->port address in case the port in question listens |
| 133 | * "any" IP address. |
| 134 | */ |
| 135 | static void nvmet_set_disc_traddr(struct nvmet_req *req, struct nvmet_port *port, |
| 136 | char *traddr) |
| 137 | { |
| 138 | if (req->ops->disc_traddr) |
| 139 | req->ops->disc_traddr(req, port, traddr); |
| 140 | else |
| 141 | memcpy(traddr, port->disc_addr.traddr, NVMF_TRADDR_SIZE); |
| 142 | } |
| 143 | |
| 144 | static size_t discovery_log_entries(struct nvmet_req *req) |
| 145 | { |
| 146 | struct nvmet_ctrl *ctrl = req->sq->ctrl; |
| 147 | struct nvmet_subsys_link *p; |
| 148 | struct nvmet_port *r; |
| 149 | size_t entries = 1; |
| 150 | |
| 151 | list_for_each_entry(p, &req->port->subsystems, entry) { |
| 152 | if (!nvmet_host_allowed(subsys: p->subsys, hostnqn: ctrl->hostnqn)) |
| 153 | continue; |
| 154 | entries++; |
| 155 | } |
| 156 | list_for_each_entry(r, &req->port->referrals, entry) |
| 157 | entries++; |
| 158 | return entries; |
| 159 | } |
| 160 | |
| 161 | static void nvmet_execute_disc_get_log_page(struct nvmet_req *req) |
| 162 | { |
| 163 | const int entry_size = sizeof(struct nvmf_disc_rsp_page_entry); |
| 164 | struct nvmet_ctrl *ctrl = req->sq->ctrl; |
| 165 | struct nvmf_disc_rsp_page_hdr *hdr; |
| 166 | u64 offset = nvmet_get_log_page_offset(cmd: req->cmd); |
| 167 | size_t data_len = nvmet_get_log_page_len(cmd: req->cmd); |
| 168 | size_t alloc_len; |
| 169 | struct nvmet_subsys_link *p; |
| 170 | struct nvmet_port *r; |
| 171 | u32 numrec = 0; |
| 172 | u16 status = 0; |
| 173 | void *buffer; |
| 174 | char traddr[NVMF_TRADDR_SIZE]; |
| 175 | |
| 176 | if (!nvmet_check_transfer_len(req, len: data_len)) |
| 177 | return; |
| 178 | |
| 179 | if (req->cmd->get_log_page.lid != NVME_LOG_DISC) { |
| 180 | req->error_loc = |
| 181 | offsetof(struct nvme_get_log_page_command, lid); |
| 182 | status = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR; |
| 183 | goto out; |
| 184 | } |
| 185 | |
| 186 | /* Spec requires dword aligned offsets */ |
| 187 | if (offset & 0x3) { |
| 188 | req->error_loc = |
| 189 | offsetof(struct nvme_get_log_page_command, lpo); |
| 190 | status = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR; |
| 191 | goto out; |
| 192 | } |
| 193 | |
| 194 | /* |
| 195 | * Make sure we're passing at least a buffer of response header size. |
| 196 | * If host provided data len is less than the header size, only the |
| 197 | * number of bytes requested by host will be sent to host. |
| 198 | */ |
| 199 | down_read(sem: &nvmet_config_sem); |
| 200 | alloc_len = sizeof(*hdr) + entry_size * discovery_log_entries(req); |
| 201 | buffer = kzalloc(alloc_len, GFP_KERNEL); |
| 202 | if (!buffer) { |
| 203 | up_read(sem: &nvmet_config_sem); |
| 204 | status = NVME_SC_INTERNAL; |
| 205 | goto out; |
| 206 | } |
| 207 | hdr = buffer; |
| 208 | |
| 209 | nvmet_set_disc_traddr(req, port: req->port, traddr); |
| 210 | |
| 211 | nvmet_format_discovery_entry(hdr, port: req->port, |
| 212 | subsys_nqn: nvmet_disc_subsys->subsysnqn, |
| 213 | traddr, type: NVME_NQN_CURR, numrec); |
| 214 | numrec++; |
| 215 | |
| 216 | list_for_each_entry(p, &req->port->subsystems, entry) { |
| 217 | if (!nvmet_host_allowed(subsys: p->subsys, hostnqn: ctrl->hostnqn)) |
| 218 | continue; |
| 219 | |
| 220 | nvmet_format_discovery_entry(hdr, port: req->port, |
| 221 | subsys_nqn: p->subsys->subsysnqn, traddr, |
| 222 | type: NVME_NQN_NVME, numrec); |
| 223 | numrec++; |
| 224 | } |
| 225 | |
| 226 | list_for_each_entry(r, &req->port->referrals, entry) { |
| 227 | if (r->disc_addr.trtype == NVMF_TRTYPE_PCI) |
| 228 | continue; |
| 229 | |
| 230 | nvmet_format_discovery_entry(hdr, port: r, |
| 231 | NVME_DISC_SUBSYS_NAME, |
| 232 | traddr: r->disc_addr.traddr, |
| 233 | type: NVME_NQN_DISC, numrec); |
| 234 | numrec++; |
| 235 | } |
| 236 | |
| 237 | hdr->genctr = cpu_to_le64(nvmet_genctr); |
| 238 | hdr->numrec = cpu_to_le64(numrec); |
| 239 | hdr->recfmt = cpu_to_le16(0); |
| 240 | |
| 241 | nvmet_clear_aen_bit(req, bn: NVME_AEN_BIT_DISC_CHANGE); |
| 242 | |
| 243 | up_read(sem: &nvmet_config_sem); |
| 244 | |
| 245 | status = nvmet_copy_to_sgl(req, off: 0, buf: buffer + offset, len: data_len); |
| 246 | kfree(objp: buffer); |
| 247 | out: |
| 248 | nvmet_req_complete(req, status); |
| 249 | } |
| 250 | |
| 251 | static void nvmet_execute_disc_identify(struct nvmet_req *req) |
| 252 | { |
| 253 | struct nvmet_ctrl *ctrl = req->sq->ctrl; |
| 254 | struct nvme_id_ctrl *id; |
| 255 | u16 status = 0; |
| 256 | |
| 257 | if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE)) |
| 258 | return; |
| 259 | |
| 260 | if (req->cmd->identify.cns != NVME_ID_CNS_CTRL) { |
| 261 | req->error_loc = offsetof(struct nvme_identify, cns); |
| 262 | status = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR; |
| 263 | goto out; |
| 264 | } |
| 265 | |
| 266 | id = kzalloc(sizeof(*id), GFP_KERNEL); |
| 267 | if (!id) { |
| 268 | status = NVME_SC_INTERNAL; |
| 269 | goto out; |
| 270 | } |
| 271 | |
| 272 | memcpy(id->sn, ctrl->subsys->serial, NVMET_SN_MAX_SIZE); |
| 273 | memset(id->fr, ' ', sizeof(id->fr)); |
| 274 | memcpy_and_pad(dest: id->mn, dest_len: sizeof(id->mn), src: ctrl->subsys->model_number, |
| 275 | strlen(ctrl->subsys->model_number), pad: ' '); |
| 276 | memcpy_and_pad(dest: id->fr, dest_len: sizeof(id->fr), |
| 277 | UTS_RELEASE, strlen(UTS_RELEASE), pad: ' '); |
| 278 | |
| 279 | id->cntrltype = NVME_CTRL_DISC; |
| 280 | |
| 281 | /* no limit on data transfer sizes for now */ |
| 282 | id->mdts = 0; |
| 283 | id->cntlid = cpu_to_le16(ctrl->cntlid); |
| 284 | id->ver = cpu_to_le32(ctrl->subsys->ver); |
| 285 | id->lpa = (1 << 2); |
| 286 | |
| 287 | /* no enforcement soft-limit for maxcmd - pick arbitrary high value */ |
| 288 | id->maxcmd = cpu_to_le16(NVMET_MAX_CMD(ctrl)); |
| 289 | |
| 290 | id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */ |
| 291 | if (ctrl->ops->flags & NVMF_KEYED_SGLS) |
| 292 | id->sgls |= cpu_to_le32(1 << 2); |
| 293 | if (req->port->inline_data_size) |
| 294 | id->sgls |= cpu_to_le32(1 << 20); |
| 295 | |
| 296 | id->oaes = cpu_to_le32(NVMET_DISC_AEN_CFG_OPTIONAL); |
| 297 | |
| 298 | strscpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn)); |
| 299 | |
| 300 | status = nvmet_copy_to_sgl(req, off: 0, buf: id, len: sizeof(*id)); |
| 301 | |
| 302 | kfree(objp: id); |
| 303 | out: |
| 304 | nvmet_req_complete(req, status); |
| 305 | } |
| 306 | |
| 307 | static void nvmet_execute_disc_set_features(struct nvmet_req *req) |
| 308 | { |
| 309 | u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10); |
| 310 | u16 stat; |
| 311 | |
| 312 | if (!nvmet_check_transfer_len(req, len: 0)) |
| 313 | return; |
| 314 | |
| 315 | switch (cdw10 & 0xff) { |
| 316 | case NVME_FEAT_KATO: |
| 317 | stat = nvmet_set_feat_kato(req); |
| 318 | break; |
| 319 | case NVME_FEAT_ASYNC_EVENT: |
| 320 | stat = nvmet_set_feat_async_event(req, |
| 321 | NVMET_DISC_AEN_CFG_OPTIONAL); |
| 322 | break; |
| 323 | default: |
| 324 | req->error_loc = |
| 325 | offsetof(struct nvme_common_command, cdw10); |
| 326 | stat = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR; |
| 327 | break; |
| 328 | } |
| 329 | |
| 330 | nvmet_req_complete(req, status: stat); |
| 331 | } |
| 332 | |
| 333 | static void nvmet_execute_disc_get_features(struct nvmet_req *req) |
| 334 | { |
| 335 | u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10); |
| 336 | u16 stat = 0; |
| 337 | |
| 338 | if (!nvmet_check_transfer_len(req, len: 0)) |
| 339 | return; |
| 340 | |
| 341 | switch (cdw10 & 0xff) { |
| 342 | case NVME_FEAT_KATO: |
| 343 | nvmet_get_feat_kato(req); |
| 344 | break; |
| 345 | case NVME_FEAT_ASYNC_EVENT: |
| 346 | nvmet_get_feat_async_event(req); |
| 347 | break; |
| 348 | default: |
| 349 | req->error_loc = |
| 350 | offsetof(struct nvme_common_command, cdw10); |
| 351 | stat = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR; |
| 352 | break; |
| 353 | } |
| 354 | |
| 355 | nvmet_req_complete(req, status: stat); |
| 356 | } |
| 357 | |
| 358 | u32 nvmet_discovery_cmd_data_len(struct nvmet_req *req) |
| 359 | { |
| 360 | struct nvme_command *cmd = req->cmd; |
| 361 | |
| 362 | switch (cmd->common.opcode) { |
| 363 | case nvme_admin_get_log_page: |
| 364 | return nvmet_get_log_page_len(cmd: req->cmd); |
| 365 | case nvme_admin_identify: |
| 366 | return NVME_IDENTIFY_DATA_SIZE; |
| 367 | default: |
| 368 | return 0; |
| 369 | } |
| 370 | } |
| 371 | |
| 372 | u16 nvmet_parse_discovery_cmd(struct nvmet_req *req) |
| 373 | { |
| 374 | struct nvme_command *cmd = req->cmd; |
| 375 | |
| 376 | if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) { |
| 377 | pr_err("got cmd %d while not ready\n" , |
| 378 | cmd->common.opcode); |
| 379 | req->error_loc = |
| 380 | offsetof(struct nvme_common_command, opcode); |
| 381 | return NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR; |
| 382 | } |
| 383 | |
| 384 | switch (cmd->common.opcode) { |
| 385 | case nvme_admin_set_features: |
| 386 | req->execute = nvmet_execute_disc_set_features; |
| 387 | return 0; |
| 388 | case nvme_admin_get_features: |
| 389 | req->execute = nvmet_execute_disc_get_features; |
| 390 | return 0; |
| 391 | case nvme_admin_async_event: |
| 392 | req->execute = nvmet_execute_async_event; |
| 393 | return 0; |
| 394 | case nvme_admin_keep_alive: |
| 395 | req->execute = nvmet_execute_keep_alive; |
| 396 | return 0; |
| 397 | case nvme_admin_get_log_page: |
| 398 | req->execute = nvmet_execute_disc_get_log_page; |
| 399 | return 0; |
| 400 | case nvme_admin_identify: |
| 401 | req->execute = nvmet_execute_disc_identify; |
| 402 | return 0; |
| 403 | default: |
| 404 | pr_debug("unhandled cmd %d\n" , cmd->common.opcode); |
| 405 | req->error_loc = offsetof(struct nvme_common_command, opcode); |
| 406 | return NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR; |
| 407 | } |
| 408 | |
| 409 | } |
| 410 | |
| 411 | int __init nvmet_init_discovery(void) |
| 412 | { |
| 413 | nvmet_disc_subsys = |
| 414 | nvmet_subsys_alloc(NVME_DISC_SUBSYS_NAME, type: NVME_NQN_CURR); |
| 415 | return PTR_ERR_OR_ZERO(ptr: nvmet_disc_subsys); |
| 416 | } |
| 417 | |
| 418 | void nvmet_exit_discovery(void) |
| 419 | { |
| 420 | nvmet_subsys_put(subsys: nvmet_disc_subsys); |
| 421 | } |
| 422 | |