1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * NVMe admin command implementation.
4 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
5 */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 #include <linux/module.h>
8 #include <linux/rculist.h>
9 #include <linux/part_stat.h>
10
11 #include <generated/utsrelease.h>
12 #include <asm/unaligned.h>
13 #include "nvmet.h"
14
nvmet_get_log_page_len(struct nvme_command * cmd)15 u32 nvmet_get_log_page_len(struct nvme_command *cmd)
16 {
17 u32 len = le16_to_cpu(cmd->get_log_page.numdu);
18
19 len <<= 16;
20 len += le16_to_cpu(cmd->get_log_page.numdl);
21 /* NUMD is a 0's based value */
22 len += 1;
23 len *= sizeof(u32);
24
25 return len;
26 }
27
nvmet_feat_data_len(struct nvmet_req * req,u32 cdw10)28 static u32 nvmet_feat_data_len(struct nvmet_req *req, u32 cdw10)
29 {
30 switch (cdw10 & 0xff) {
31 case NVME_FEAT_HOST_ID:
32 return sizeof(req->sq->ctrl->hostid);
33 default:
34 return 0;
35 }
36 }
37
nvmet_get_log_page_offset(struct nvme_command * cmd)38 u64 nvmet_get_log_page_offset(struct nvme_command *cmd)
39 {
40 return le64_to_cpu(cmd->get_log_page.lpo);
41 }
42
nvmet_execute_get_log_page_noop(struct nvmet_req * req)43 static void nvmet_execute_get_log_page_noop(struct nvmet_req *req)
44 {
45 nvmet_req_complete(req, nvmet_zero_sgl(req, 0, req->transfer_len));
46 }
47
nvmet_execute_get_log_page_error(struct nvmet_req * req)48 static void nvmet_execute_get_log_page_error(struct nvmet_req *req)
49 {
50 struct nvmet_ctrl *ctrl = req->sq->ctrl;
51 unsigned long flags;
52 off_t offset = 0;
53 u64 slot;
54 u64 i;
55
56 spin_lock_irqsave(&ctrl->error_lock, flags);
57 slot = ctrl->err_counter % NVMET_ERROR_LOG_SLOTS;
58
59 for (i = 0; i < NVMET_ERROR_LOG_SLOTS; i++) {
60 if (nvmet_copy_to_sgl(req, offset, &ctrl->slots[slot],
61 sizeof(struct nvme_error_slot)))
62 break;
63
64 if (slot == 0)
65 slot = NVMET_ERROR_LOG_SLOTS - 1;
66 else
67 slot--;
68 offset += sizeof(struct nvme_error_slot);
69 }
70 spin_unlock_irqrestore(&ctrl->error_lock, flags);
71 nvmet_req_complete(req, 0);
72 }
73
nvmet_get_smart_log_nsid(struct nvmet_req * req,struct nvme_smart_log * slog)74 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req,
75 struct nvme_smart_log *slog)
76 {
77 u64 host_reads, host_writes, data_units_read, data_units_written;
78 u16 status;
79
80 status = nvmet_req_find_ns(req);
81 if (status)
82 return status;
83
84 /* we don't have the right data for file backed ns */
85 if (!req->ns->bdev)
86 return NVME_SC_SUCCESS;
87
88 host_reads = part_stat_read(req->ns->bdev, ios[READ]);
89 data_units_read =
90 DIV_ROUND_UP(part_stat_read(req->ns->bdev, sectors[READ]), 1000);
91 host_writes = part_stat_read(req->ns->bdev, ios[WRITE]);
92 data_units_written =
93 DIV_ROUND_UP(part_stat_read(req->ns->bdev, sectors[WRITE]), 1000);
94
95 put_unaligned_le64(host_reads, &slog->host_reads[0]);
96 put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
97 put_unaligned_le64(host_writes, &slog->host_writes[0]);
98 put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
99
100 return NVME_SC_SUCCESS;
101 }
102
nvmet_get_smart_log_all(struct nvmet_req * req,struct nvme_smart_log * slog)103 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
104 struct nvme_smart_log *slog)
105 {
106 u64 host_reads = 0, host_writes = 0;
107 u64 data_units_read = 0, data_units_written = 0;
108 struct nvmet_ns *ns;
109 struct nvmet_ctrl *ctrl;
110 unsigned long idx;
111
112 ctrl = req->sq->ctrl;
113 xa_for_each(&ctrl->subsys->namespaces, idx, ns) {
114 /* we don't have the right data for file backed ns */
115 if (!ns->bdev)
116 continue;
117 host_reads += part_stat_read(ns->bdev, ios[READ]);
118 data_units_read += DIV_ROUND_UP(
119 part_stat_read(ns->bdev, sectors[READ]), 1000);
120 host_writes += part_stat_read(ns->bdev, ios[WRITE]);
121 data_units_written += DIV_ROUND_UP(
122 part_stat_read(ns->bdev, sectors[WRITE]), 1000);
123 }
124
125 put_unaligned_le64(host_reads, &slog->host_reads[0]);
126 put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
127 put_unaligned_le64(host_writes, &slog->host_writes[0]);
128 put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
129
130 return NVME_SC_SUCCESS;
131 }
132
nvmet_execute_get_log_page_smart(struct nvmet_req * req)133 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req)
134 {
135 struct nvme_smart_log *log;
136 u16 status = NVME_SC_INTERNAL;
137 unsigned long flags;
138
139 if (req->transfer_len != sizeof(*log))
140 goto out;
141
142 log = kzalloc(sizeof(*log), GFP_KERNEL);
143 if (!log)
144 goto out;
145
146 if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL))
147 status = nvmet_get_smart_log_all(req, log);
148 else
149 status = nvmet_get_smart_log_nsid(req, log);
150 if (status)
151 goto out_free_log;
152
153 spin_lock_irqsave(&req->sq->ctrl->error_lock, flags);
154 put_unaligned_le64(req->sq->ctrl->err_counter,
155 &log->num_err_log_entries);
156 spin_unlock_irqrestore(&req->sq->ctrl->error_lock, flags);
157
158 status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
159 out_free_log:
160 kfree(log);
161 out:
162 nvmet_req_complete(req, status);
163 }
164
nvmet_get_cmd_effects_nvm(struct nvme_effects_log * log)165 static void nvmet_get_cmd_effects_nvm(struct nvme_effects_log *log)
166 {
167 log->acs[nvme_admin_get_log_page] =
168 log->acs[nvme_admin_identify] =
169 log->acs[nvme_admin_abort_cmd] =
170 log->acs[nvme_admin_set_features] =
171 log->acs[nvme_admin_get_features] =
172 log->acs[nvme_admin_async_event] =
173 log->acs[nvme_admin_keep_alive] =
174 cpu_to_le32(NVME_CMD_EFFECTS_CSUPP);
175
176 log->iocs[nvme_cmd_read] =
177 log->iocs[nvme_cmd_write] =
178 log->iocs[nvme_cmd_flush] =
179 log->iocs[nvme_cmd_dsm] =
180 log->iocs[nvme_cmd_write_zeroes] =
181 cpu_to_le32(NVME_CMD_EFFECTS_CSUPP);
182 }
183
nvmet_get_cmd_effects_zns(struct nvme_effects_log * log)184 static void nvmet_get_cmd_effects_zns(struct nvme_effects_log *log)
185 {
186 log->iocs[nvme_cmd_zone_append] =
187 log->iocs[nvme_cmd_zone_mgmt_send] =
188 log->iocs[nvme_cmd_zone_mgmt_recv] =
189 cpu_to_le32(NVME_CMD_EFFECTS_CSUPP);
190 }
191
nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req * req)192 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
193 {
194 struct nvme_effects_log *log;
195 u16 status = NVME_SC_SUCCESS;
196
197 log = kzalloc(sizeof(*log), GFP_KERNEL);
198 if (!log) {
199 status = NVME_SC_INTERNAL;
200 goto out;
201 }
202
203 switch (req->cmd->get_log_page.csi) {
204 case NVME_CSI_NVM:
205 nvmet_get_cmd_effects_nvm(log);
206 break;
207 case NVME_CSI_ZNS:
208 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
209 status = NVME_SC_INVALID_IO_CMD_SET;
210 goto free;
211 }
212 nvmet_get_cmd_effects_nvm(log);
213 nvmet_get_cmd_effects_zns(log);
214 break;
215 default:
216 status = NVME_SC_INVALID_LOG_PAGE;
217 goto free;
218 }
219
220 status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
221 free:
222 kfree(log);
223 out:
224 nvmet_req_complete(req, status);
225 }
226
nvmet_execute_get_log_changed_ns(struct nvmet_req * req)227 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
228 {
229 struct nvmet_ctrl *ctrl = req->sq->ctrl;
230 u16 status = NVME_SC_INTERNAL;
231 size_t len;
232
233 if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
234 goto out;
235
236 mutex_lock(&ctrl->lock);
237 if (ctrl->nr_changed_ns == U32_MAX)
238 len = sizeof(__le32);
239 else
240 len = ctrl->nr_changed_ns * sizeof(__le32);
241 status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
242 if (!status)
243 status = nvmet_zero_sgl(req, len, req->transfer_len - len);
244 ctrl->nr_changed_ns = 0;
245 nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR);
246 mutex_unlock(&ctrl->lock);
247 out:
248 nvmet_req_complete(req, status);
249 }
250
nvmet_format_ana_group(struct nvmet_req * req,u32 grpid,struct nvme_ana_group_desc * desc)251 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid,
252 struct nvme_ana_group_desc *desc)
253 {
254 struct nvmet_ctrl *ctrl = req->sq->ctrl;
255 struct nvmet_ns *ns;
256 unsigned long idx;
257 u32 count = 0;
258
259 if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) {
260 xa_for_each(&ctrl->subsys->namespaces, idx, ns)
261 if (ns->anagrpid == grpid)
262 desc->nsids[count++] = cpu_to_le32(ns->nsid);
263 }
264
265 desc->grpid = cpu_to_le32(grpid);
266 desc->nnsids = cpu_to_le32(count);
267 desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
268 desc->state = req->port->ana_state[grpid];
269 memset(desc->rsvd17, 0, sizeof(desc->rsvd17));
270 return struct_size(desc, nsids, count);
271 }
272
nvmet_execute_get_log_page_ana(struct nvmet_req * req)273 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)
274 {
275 struct nvme_ana_rsp_hdr hdr = { 0, };
276 struct nvme_ana_group_desc *desc;
277 size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */
278 size_t len;
279 u32 grpid;
280 u16 ngrps = 0;
281 u16 status;
282
283 status = NVME_SC_INTERNAL;
284 desc = kmalloc(struct_size(desc, nsids, NVMET_MAX_NAMESPACES),
285 GFP_KERNEL);
286 if (!desc)
287 goto out;
288
289 down_read(&nvmet_ana_sem);
290 for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
291 if (!nvmet_ana_group_enabled[grpid])
292 continue;
293 len = nvmet_format_ana_group(req, grpid, desc);
294 status = nvmet_copy_to_sgl(req, offset, desc, len);
295 if (status)
296 break;
297 offset += len;
298 ngrps++;
299 }
300 for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
301 if (nvmet_ana_group_enabled[grpid])
302 ngrps++;
303 }
304
305 hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
306 hdr.ngrps = cpu_to_le16(ngrps);
307 nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE);
308 up_read(&nvmet_ana_sem);
309
310 kfree(desc);
311
312 /* copy the header last once we know the number of groups */
313 status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr));
314 out:
315 nvmet_req_complete(req, status);
316 }
317
nvmet_execute_get_log_page(struct nvmet_req * req)318 static void nvmet_execute_get_log_page(struct nvmet_req *req)
319 {
320 if (!nvmet_check_transfer_len(req, nvmet_get_log_page_len(req->cmd)))
321 return;
322
323 switch (req->cmd->get_log_page.lid) {
324 case NVME_LOG_ERROR:
325 return nvmet_execute_get_log_page_error(req);
326 case NVME_LOG_SMART:
327 return nvmet_execute_get_log_page_smart(req);
328 case NVME_LOG_FW_SLOT:
329 /*
330 * We only support a single firmware slot which always is
331 * active, so we can zero out the whole firmware slot log and
332 * still claim to fully implement this mandatory log page.
333 */
334 return nvmet_execute_get_log_page_noop(req);
335 case NVME_LOG_CHANGED_NS:
336 return nvmet_execute_get_log_changed_ns(req);
337 case NVME_LOG_CMD_EFFECTS:
338 return nvmet_execute_get_log_cmd_effects_ns(req);
339 case NVME_LOG_ANA:
340 return nvmet_execute_get_log_page_ana(req);
341 }
342 pr_debug("unhandled lid %d on qid %d\n",
343 req->cmd->get_log_page.lid, req->sq->qid);
344 req->error_loc = offsetof(struct nvme_get_log_page_command, lid);
345 nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
346 }
347
nvmet_execute_identify_ctrl(struct nvmet_req * req)348 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
349 {
350 struct nvmet_ctrl *ctrl = req->sq->ctrl;
351 struct nvmet_subsys *subsys = ctrl->subsys;
352 struct nvme_id_ctrl *id;
353 u32 cmd_capsule_size;
354 u16 status = 0;
355
356 if (!subsys->subsys_discovered) {
357 mutex_lock(&subsys->lock);
358 subsys->subsys_discovered = true;
359 mutex_unlock(&subsys->lock);
360 }
361
362 id = kzalloc(sizeof(*id), GFP_KERNEL);
363 if (!id) {
364 status = NVME_SC_INTERNAL;
365 goto out;
366 }
367
368 /* XXX: figure out how to assign real vendors IDs. */
369 id->vid = 0;
370 id->ssvid = 0;
371
372 memcpy(id->sn, ctrl->subsys->serial, NVMET_SN_MAX_SIZE);
373 memcpy_and_pad(id->mn, sizeof(id->mn), subsys->model_number,
374 strlen(subsys->model_number), ' ');
375 memcpy_and_pad(id->fr, sizeof(id->fr),
376 UTS_RELEASE, strlen(UTS_RELEASE), ' ');
377
378 id->rab = 6;
379
380 if (nvmet_is_disc_subsys(ctrl->subsys))
381 id->cntrltype = NVME_CTRL_DISC;
382 else
383 id->cntrltype = NVME_CTRL_IO;
384
385 /*
386 * XXX: figure out how we can assign a IEEE OUI, but until then
387 * the safest is to leave it as zeroes.
388 */
389
390 /* we support multiple ports, multiples hosts and ANA: */
391 id->cmic = NVME_CTRL_CMIC_MULTI_PORT | NVME_CTRL_CMIC_MULTI_CTRL |
392 NVME_CTRL_CMIC_ANA;
393
394 /* Limit MDTS according to transport capability */
395 if (ctrl->ops->get_mdts)
396 id->mdts = ctrl->ops->get_mdts(ctrl);
397 else
398 id->mdts = 0;
399
400 id->cntlid = cpu_to_le16(ctrl->cntlid);
401 id->ver = cpu_to_le32(ctrl->subsys->ver);
402
403 /* XXX: figure out what to do about RTD3R/RTD3 */
404 id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
405 id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
406 NVME_CTRL_ATTR_TBKAS);
407
408 id->oacs = 0;
409
410 /*
411 * We don't really have a practical limit on the number of abort
412 * comands. But we don't do anything useful for abort either, so
413 * no point in allowing more abort commands than the spec requires.
414 */
415 id->acl = 3;
416
417 id->aerl = NVMET_ASYNC_EVENTS - 1;
418
419 /* first slot is read-only, only one slot supported */
420 id->frmw = (1 << 0) | (1 << 1);
421 id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
422 id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
423 id->npss = 0;
424
425 /* We support keep-alive timeout in granularity of seconds */
426 id->kas = cpu_to_le16(NVMET_KAS);
427
428 id->sqes = (0x6 << 4) | 0x6;
429 id->cqes = (0x4 << 4) | 0x4;
430
431 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */
432 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
433
434 id->nn = cpu_to_le32(NVMET_MAX_NAMESPACES);
435 id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
436 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
437 NVME_CTRL_ONCS_WRITE_ZEROES);
438
439 /* XXX: don't report vwc if the underlying device is write through */
440 id->vwc = NVME_CTRL_VWC_PRESENT;
441
442 /*
443 * We can't support atomic writes bigger than a LBA without support
444 * from the backend device.
445 */
446 id->awun = 0;
447 id->awupf = 0;
448
449 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */
450 if (ctrl->ops->flags & NVMF_KEYED_SGLS)
451 id->sgls |= cpu_to_le32(1 << 2);
452 if (req->port->inline_data_size)
453 id->sgls |= cpu_to_le32(1 << 20);
454
455 strscpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
456
457 /*
458 * Max command capsule size is sqe + in-capsule data size.
459 * Disable in-capsule data for Metadata capable controllers.
460 */
461 cmd_capsule_size = sizeof(struct nvme_command);
462 if (!ctrl->pi_support)
463 cmd_capsule_size += req->port->inline_data_size;
464 id->ioccsz = cpu_to_le32(cmd_capsule_size / 16);
465
466 /* Max response capsule size is cqe */
467 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
468
469 id->msdbd = ctrl->ops->msdbd;
470
471 id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
472 id->anatt = 10; /* random value */
473 id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
474 id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
475
476 /*
477 * Meh, we don't really support any power state. Fake up the same
478 * values that qemu does.
479 */
480 id->psd[0].max_power = cpu_to_le16(0x9c4);
481 id->psd[0].entry_lat = cpu_to_le32(0x10);
482 id->psd[0].exit_lat = cpu_to_le32(0x4);
483
484 id->nwpc = 1 << 0; /* write protect and no write protect */
485
486 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
487
488 kfree(id);
489 out:
490 nvmet_req_complete(req, status);
491 }
492
nvmet_execute_identify_ns(struct nvmet_req * req)493 static void nvmet_execute_identify_ns(struct nvmet_req *req)
494 {
495 struct nvme_id_ns *id;
496 u16 status;
497
498 if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
499 req->error_loc = offsetof(struct nvme_identify, nsid);
500 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
501 goto out;
502 }
503
504 id = kzalloc(sizeof(*id), GFP_KERNEL);
505 if (!id) {
506 status = NVME_SC_INTERNAL;
507 goto out;
508 }
509
510 /* return an all zeroed buffer if we can't find an active namespace */
511 status = nvmet_req_find_ns(req);
512 if (status) {
513 status = 0;
514 goto done;
515 }
516
517 if (nvmet_ns_revalidate(req->ns)) {
518 mutex_lock(&req->ns->subsys->lock);
519 nvmet_ns_changed(req->ns->subsys, req->ns->nsid);
520 mutex_unlock(&req->ns->subsys->lock);
521 }
522
523 /*
524 * nuse = ncap = nsze isn't always true, but we have no way to find
525 * that out from the underlying device.
526 */
527 id->ncap = id->nsze =
528 cpu_to_le64(req->ns->size >> req->ns->blksize_shift);
529 switch (req->port->ana_state[req->ns->anagrpid]) {
530 case NVME_ANA_INACCESSIBLE:
531 case NVME_ANA_PERSISTENT_LOSS:
532 break;
533 default:
534 id->nuse = id->nsze;
535 break;
536 }
537
538 if (req->ns->bdev)
539 nvmet_bdev_set_limits(req->ns->bdev, id);
540
541 /*
542 * We just provide a single LBA format that matches what the
543 * underlying device reports.
544 */
545 id->nlbaf = 0;
546 id->flbas = 0;
547
548 /*
549 * Our namespace might always be shared. Not just with other
550 * controllers, but also with any other user of the block device.
551 */
552 id->nmic = NVME_NS_NMIC_SHARED;
553 id->anagrpid = cpu_to_le32(req->ns->anagrpid);
554
555 memcpy(&id->nguid, &req->ns->nguid, sizeof(id->nguid));
556
557 id->lbaf[0].ds = req->ns->blksize_shift;
558
559 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns)) {
560 id->dpc = NVME_NS_DPC_PI_FIRST | NVME_NS_DPC_PI_LAST |
561 NVME_NS_DPC_PI_TYPE1 | NVME_NS_DPC_PI_TYPE2 |
562 NVME_NS_DPC_PI_TYPE3;
563 id->mc = NVME_MC_EXTENDED_LBA;
564 id->dps = req->ns->pi_type;
565 id->flbas = NVME_NS_FLBAS_META_EXT;
566 id->lbaf[0].ms = cpu_to_le16(req->ns->metadata_size);
567 }
568
569 if (req->ns->readonly)
570 id->nsattr |= (1 << 0);
571 done:
572 if (!status)
573 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
574
575 kfree(id);
576 out:
577 nvmet_req_complete(req, status);
578 }
579
nvmet_execute_identify_nslist(struct nvmet_req * req)580 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
581 {
582 static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
583 struct nvmet_ctrl *ctrl = req->sq->ctrl;
584 struct nvmet_ns *ns;
585 unsigned long idx;
586 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
587 __le32 *list;
588 u16 status = 0;
589 int i = 0;
590
591 list = kzalloc(buf_size, GFP_KERNEL);
592 if (!list) {
593 status = NVME_SC_INTERNAL;
594 goto out;
595 }
596
597 xa_for_each(&ctrl->subsys->namespaces, idx, ns) {
598 if (ns->nsid <= min_nsid)
599 continue;
600 list[i++] = cpu_to_le32(ns->nsid);
601 if (i == buf_size / sizeof(__le32))
602 break;
603 }
604
605 status = nvmet_copy_to_sgl(req, 0, list, buf_size);
606
607 kfree(list);
608 out:
609 nvmet_req_complete(req, status);
610 }
611
nvmet_copy_ns_identifier(struct nvmet_req * req,u8 type,u8 len,void * id,off_t * off)612 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
613 void *id, off_t *off)
614 {
615 struct nvme_ns_id_desc desc = {
616 .nidt = type,
617 .nidl = len,
618 };
619 u16 status;
620
621 status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
622 if (status)
623 return status;
624 *off += sizeof(desc);
625
626 status = nvmet_copy_to_sgl(req, *off, id, len);
627 if (status)
628 return status;
629 *off += len;
630
631 return 0;
632 }
633
nvmet_execute_identify_desclist(struct nvmet_req * req)634 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
635 {
636 off_t off = 0;
637 u16 status;
638
639 status = nvmet_req_find_ns(req);
640 if (status)
641 goto out;
642
643 if (memchr_inv(&req->ns->uuid, 0, sizeof(req->ns->uuid))) {
644 status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
645 NVME_NIDT_UUID_LEN,
646 &req->ns->uuid, &off);
647 if (status)
648 goto out;
649 }
650 if (memchr_inv(req->ns->nguid, 0, sizeof(req->ns->nguid))) {
651 status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
652 NVME_NIDT_NGUID_LEN,
653 &req->ns->nguid, &off);
654 if (status)
655 goto out;
656 }
657
658 status = nvmet_copy_ns_identifier(req, NVME_NIDT_CSI,
659 NVME_NIDT_CSI_LEN,
660 &req->ns->csi, &off);
661 if (status)
662 goto out;
663
664 if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
665 off) != NVME_IDENTIFY_DATA_SIZE - off)
666 status = NVME_SC_INTERNAL | NVME_SC_DNR;
667
668 out:
669 nvmet_req_complete(req, status);
670 }
671
nvmet_handle_identify_desclist(struct nvmet_req * req)672 static bool nvmet_handle_identify_desclist(struct nvmet_req *req)
673 {
674 switch (req->cmd->identify.csi) {
675 case NVME_CSI_NVM:
676 nvmet_execute_identify_desclist(req);
677 return true;
678 case NVME_CSI_ZNS:
679 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
680 nvmet_execute_identify_desclist(req);
681 return true;
682 }
683 return false;
684 default:
685 return false;
686 }
687 }
688
nvmet_execute_identify(struct nvmet_req * req)689 static void nvmet_execute_identify(struct nvmet_req *req)
690 {
691 if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE))
692 return;
693
694 switch (req->cmd->identify.cns) {
695 case NVME_ID_CNS_NS:
696 switch (req->cmd->identify.csi) {
697 case NVME_CSI_NVM:
698 return nvmet_execute_identify_ns(req);
699 default:
700 break;
701 }
702 break;
703 case NVME_ID_CNS_CS_NS:
704 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
705 switch (req->cmd->identify.csi) {
706 case NVME_CSI_ZNS:
707 return nvmet_execute_identify_cns_cs_ns(req);
708 default:
709 break;
710 }
711 }
712 break;
713 case NVME_ID_CNS_CTRL:
714 switch (req->cmd->identify.csi) {
715 case NVME_CSI_NVM:
716 return nvmet_execute_identify_ctrl(req);
717 }
718 break;
719 case NVME_ID_CNS_CS_CTRL:
720 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
721 switch (req->cmd->identify.csi) {
722 case NVME_CSI_ZNS:
723 return nvmet_execute_identify_cns_cs_ctrl(req);
724 default:
725 break;
726 }
727 }
728 break;
729 case NVME_ID_CNS_NS_ACTIVE_LIST:
730 switch (req->cmd->identify.csi) {
731 case NVME_CSI_NVM:
732 return nvmet_execute_identify_nslist(req);
733 default:
734 break;
735 }
736 break;
737 case NVME_ID_CNS_NS_DESC_LIST:
738 if (nvmet_handle_identify_desclist(req) == true)
739 return;
740 break;
741 }
742
743 nvmet_req_cns_error_complete(req);
744 }
745
746 /*
747 * A "minimum viable" abort implementation: the command is mandatory in the
748 * spec, but we are not required to do any useful work. We couldn't really
749 * do a useful abort, so don't bother even with waiting for the command
750 * to be exectuted and return immediately telling the command to abort
751 * wasn't found.
752 */
nvmet_execute_abort(struct nvmet_req * req)753 static void nvmet_execute_abort(struct nvmet_req *req)
754 {
755 if (!nvmet_check_transfer_len(req, 0))
756 return;
757 nvmet_set_result(req, 1);
758 nvmet_req_complete(req, 0);
759 }
760
nvmet_write_protect_flush_sync(struct nvmet_req * req)761 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
762 {
763 u16 status;
764
765 if (req->ns->file)
766 status = nvmet_file_flush(req);
767 else
768 status = nvmet_bdev_flush(req);
769
770 if (status)
771 pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
772 return status;
773 }
774
nvmet_set_feat_write_protect(struct nvmet_req * req)775 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req)
776 {
777 u32 write_protect = le32_to_cpu(req->cmd->common.cdw11);
778 struct nvmet_subsys *subsys = nvmet_req_subsys(req);
779 u16 status;
780
781 status = nvmet_req_find_ns(req);
782 if (status)
783 return status;
784
785 mutex_lock(&subsys->lock);
786 switch (write_protect) {
787 case NVME_NS_WRITE_PROTECT:
788 req->ns->readonly = true;
789 status = nvmet_write_protect_flush_sync(req);
790 if (status)
791 req->ns->readonly = false;
792 break;
793 case NVME_NS_NO_WRITE_PROTECT:
794 req->ns->readonly = false;
795 status = 0;
796 break;
797 default:
798 break;
799 }
800
801 if (!status)
802 nvmet_ns_changed(subsys, req->ns->nsid);
803 mutex_unlock(&subsys->lock);
804 return status;
805 }
806
nvmet_set_feat_kato(struct nvmet_req * req)807 u16 nvmet_set_feat_kato(struct nvmet_req *req)
808 {
809 u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
810
811 nvmet_stop_keep_alive_timer(req->sq->ctrl);
812 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
813 nvmet_start_keep_alive_timer(req->sq->ctrl);
814
815 nvmet_set_result(req, req->sq->ctrl->kato);
816
817 return 0;
818 }
819
nvmet_set_feat_async_event(struct nvmet_req * req,u32 mask)820 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask)
821 {
822 u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
823
824 if (val32 & ~mask) {
825 req->error_loc = offsetof(struct nvme_common_command, cdw11);
826 return NVME_SC_INVALID_FIELD | NVME_SC_DNR;
827 }
828
829 WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
830 nvmet_set_result(req, val32);
831
832 return 0;
833 }
834
nvmet_execute_set_features(struct nvmet_req * req)835 void nvmet_execute_set_features(struct nvmet_req *req)
836 {
837 struct nvmet_subsys *subsys = nvmet_req_subsys(req);
838 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
839 u32 cdw11 = le32_to_cpu(req->cmd->common.cdw11);
840 u16 status = 0;
841 u16 nsqr;
842 u16 ncqr;
843
844 if (!nvmet_check_transfer_len(req, 0))
845 return;
846
847 switch (cdw10 & 0xff) {
848 case NVME_FEAT_NUM_QUEUES:
849 ncqr = (cdw11 >> 16) & 0xffff;
850 nsqr = cdw11 & 0xffff;
851 if (ncqr == 0xffff || nsqr == 0xffff) {
852 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
853 break;
854 }
855 nvmet_set_result(req,
856 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
857 break;
858 case NVME_FEAT_KATO:
859 status = nvmet_set_feat_kato(req);
860 break;
861 case NVME_FEAT_ASYNC_EVENT:
862 status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL);
863 break;
864 case NVME_FEAT_HOST_ID:
865 status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
866 break;
867 case NVME_FEAT_WRITE_PROTECT:
868 status = nvmet_set_feat_write_protect(req);
869 break;
870 default:
871 req->error_loc = offsetof(struct nvme_common_command, cdw10);
872 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
873 break;
874 }
875
876 nvmet_req_complete(req, status);
877 }
878
nvmet_get_feat_write_protect(struct nvmet_req * req)879 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req)
880 {
881 struct nvmet_subsys *subsys = nvmet_req_subsys(req);
882 u32 result;
883
884 result = nvmet_req_find_ns(req);
885 if (result)
886 return result;
887
888 mutex_lock(&subsys->lock);
889 if (req->ns->readonly == true)
890 result = NVME_NS_WRITE_PROTECT;
891 else
892 result = NVME_NS_NO_WRITE_PROTECT;
893 nvmet_set_result(req, result);
894 mutex_unlock(&subsys->lock);
895
896 return 0;
897 }
898
nvmet_get_feat_kato(struct nvmet_req * req)899 void nvmet_get_feat_kato(struct nvmet_req *req)
900 {
901 nvmet_set_result(req, req->sq->ctrl->kato * 1000);
902 }
903
nvmet_get_feat_async_event(struct nvmet_req * req)904 void nvmet_get_feat_async_event(struct nvmet_req *req)
905 {
906 nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
907 }
908
nvmet_execute_get_features(struct nvmet_req * req)909 void nvmet_execute_get_features(struct nvmet_req *req)
910 {
911 struct nvmet_subsys *subsys = nvmet_req_subsys(req);
912 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
913 u16 status = 0;
914
915 if (!nvmet_check_transfer_len(req, nvmet_feat_data_len(req, cdw10)))
916 return;
917
918 switch (cdw10 & 0xff) {
919 /*
920 * These features are mandatory in the spec, but we don't
921 * have a useful way to implement them. We'll eventually
922 * need to come up with some fake values for these.
923 */
924 #if 0
925 case NVME_FEAT_ARBITRATION:
926 break;
927 case NVME_FEAT_POWER_MGMT:
928 break;
929 case NVME_FEAT_TEMP_THRESH:
930 break;
931 case NVME_FEAT_ERR_RECOVERY:
932 break;
933 case NVME_FEAT_IRQ_COALESCE:
934 break;
935 case NVME_FEAT_IRQ_CONFIG:
936 break;
937 case NVME_FEAT_WRITE_ATOMIC:
938 break;
939 #endif
940 case NVME_FEAT_ASYNC_EVENT:
941 nvmet_get_feat_async_event(req);
942 break;
943 case NVME_FEAT_VOLATILE_WC:
944 nvmet_set_result(req, 1);
945 break;
946 case NVME_FEAT_NUM_QUEUES:
947 nvmet_set_result(req,
948 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
949 break;
950 case NVME_FEAT_KATO:
951 nvmet_get_feat_kato(req);
952 break;
953 case NVME_FEAT_HOST_ID:
954 /* need 128-bit host identifier flag */
955 if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) {
956 req->error_loc =
957 offsetof(struct nvme_common_command, cdw11);
958 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
959 break;
960 }
961
962 status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
963 sizeof(req->sq->ctrl->hostid));
964 break;
965 case NVME_FEAT_WRITE_PROTECT:
966 status = nvmet_get_feat_write_protect(req);
967 break;
968 default:
969 req->error_loc =
970 offsetof(struct nvme_common_command, cdw10);
971 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
972 break;
973 }
974
975 nvmet_req_complete(req, status);
976 }
977
nvmet_execute_async_event(struct nvmet_req * req)978 void nvmet_execute_async_event(struct nvmet_req *req)
979 {
980 struct nvmet_ctrl *ctrl = req->sq->ctrl;
981
982 if (!nvmet_check_transfer_len(req, 0))
983 return;
984
985 mutex_lock(&ctrl->lock);
986 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
987 mutex_unlock(&ctrl->lock);
988 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
989 return;
990 }
991 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
992 mutex_unlock(&ctrl->lock);
993
994 queue_work(nvmet_wq, &ctrl->async_event_work);
995 }
996
nvmet_execute_keep_alive(struct nvmet_req * req)997 void nvmet_execute_keep_alive(struct nvmet_req *req)
998 {
999 struct nvmet_ctrl *ctrl = req->sq->ctrl;
1000 u16 status = 0;
1001
1002 if (!nvmet_check_transfer_len(req, 0))
1003 return;
1004
1005 if (!ctrl->kato) {
1006 status = NVME_SC_KA_TIMEOUT_INVALID;
1007 goto out;
1008 }
1009
1010 pr_debug("ctrl %d update keep-alive timer for %d secs\n",
1011 ctrl->cntlid, ctrl->kato);
1012 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
1013 out:
1014 nvmet_req_complete(req, status);
1015 }
1016
nvmet_parse_admin_cmd(struct nvmet_req * req)1017 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
1018 {
1019 struct nvme_command *cmd = req->cmd;
1020 u16 ret;
1021
1022 if (nvme_is_fabrics(cmd))
1023 return nvmet_parse_fabrics_admin_cmd(req);
1024 if (unlikely(!nvmet_check_auth_status(req)))
1025 return NVME_SC_AUTH_REQUIRED | NVME_SC_DNR;
1026 if (nvmet_is_disc_subsys(nvmet_req_subsys(req)))
1027 return nvmet_parse_discovery_cmd(req);
1028
1029 ret = nvmet_check_ctrl_status(req);
1030 if (unlikely(ret))
1031 return ret;
1032
1033 if (nvmet_is_passthru_req(req))
1034 return nvmet_parse_passthru_admin_cmd(req);
1035
1036 switch (cmd->common.opcode) {
1037 case nvme_admin_get_log_page:
1038 req->execute = nvmet_execute_get_log_page;
1039 return 0;
1040 case nvme_admin_identify:
1041 req->execute = nvmet_execute_identify;
1042 return 0;
1043 case nvme_admin_abort_cmd:
1044 req->execute = nvmet_execute_abort;
1045 return 0;
1046 case nvme_admin_set_features:
1047 req->execute = nvmet_execute_set_features;
1048 return 0;
1049 case nvme_admin_get_features:
1050 req->execute = nvmet_execute_get_features;
1051 return 0;
1052 case nvme_admin_async_event:
1053 req->execute = nvmet_execute_async_event;
1054 return 0;
1055 case nvme_admin_keep_alive:
1056 req->execute = nvmet_execute_keep_alive;
1057 return 0;
1058 default:
1059 return nvmet_report_invalid_opcode(req);
1060 }
1061 }
1062