xref: /openbmc/linux/drivers/nvme/target/admin-cmd.c (revision 2326f3cd)
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 
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 
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 
38 u64 nvmet_get_log_page_offset(struct nvme_command *cmd)
39 {
40 	return le64_to_cpu(cmd->get_log_page.lpo);
41 }
42 
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 
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 
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 
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 
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 
165 static void nvmet_get_cmd_effects_nvm(struct nvme_effects_log *log)
166 {
167 	log->acs[nvme_admin_get_log_page]	= cpu_to_le32(1 << 0);
168 	log->acs[nvme_admin_identify]		= cpu_to_le32(1 << 0);
169 	log->acs[nvme_admin_abort_cmd]		= cpu_to_le32(1 << 0);
170 	log->acs[nvme_admin_set_features]	= cpu_to_le32(1 << 0);
171 	log->acs[nvme_admin_get_features]	= cpu_to_le32(1 << 0);
172 	log->acs[nvme_admin_async_event]	= cpu_to_le32(1 << 0);
173 	log->acs[nvme_admin_keep_alive]		= cpu_to_le32(1 << 0);
174 
175 	log->iocs[nvme_cmd_read]		= cpu_to_le32(1 << 0);
176 	log->iocs[nvme_cmd_write]		= cpu_to_le32(1 << 0);
177 	log->iocs[nvme_cmd_flush]		= cpu_to_le32(1 << 0);
178 	log->iocs[nvme_cmd_dsm]			= cpu_to_le32(1 << 0);
179 	log->iocs[nvme_cmd_write_zeroes]	= cpu_to_le32(1 << 0);
180 }
181 
182 static void nvmet_get_cmd_effects_zns(struct nvme_effects_log *log)
183 {
184 	log->iocs[nvme_cmd_zone_append]		= cpu_to_le32(1 << 0);
185 	log->iocs[nvme_cmd_zone_mgmt_send]	= cpu_to_le32(1 << 0);
186 	log->iocs[nvme_cmd_zone_mgmt_recv]	= cpu_to_le32(1 << 0);
187 }
188 
189 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
190 {
191 	struct nvme_effects_log *log;
192 	u16 status = NVME_SC_SUCCESS;
193 
194 	log = kzalloc(sizeof(*log), GFP_KERNEL);
195 	if (!log) {
196 		status = NVME_SC_INTERNAL;
197 		goto out;
198 	}
199 
200 	switch (req->cmd->get_log_page.csi) {
201 	case NVME_CSI_NVM:
202 		nvmet_get_cmd_effects_nvm(log);
203 		break;
204 	case NVME_CSI_ZNS:
205 		if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
206 			status = NVME_SC_INVALID_IO_CMD_SET;
207 			goto free;
208 		}
209 		nvmet_get_cmd_effects_nvm(log);
210 		nvmet_get_cmd_effects_zns(log);
211 		break;
212 	default:
213 		status = NVME_SC_INVALID_LOG_PAGE;
214 		goto free;
215 	}
216 
217 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
218 free:
219 	kfree(log);
220 out:
221 	nvmet_req_complete(req, status);
222 }
223 
224 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
225 {
226 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
227 	u16 status = NVME_SC_INTERNAL;
228 	size_t len;
229 
230 	if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
231 		goto out;
232 
233 	mutex_lock(&ctrl->lock);
234 	if (ctrl->nr_changed_ns == U32_MAX)
235 		len = sizeof(__le32);
236 	else
237 		len = ctrl->nr_changed_ns * sizeof(__le32);
238 	status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
239 	if (!status)
240 		status = nvmet_zero_sgl(req, len, req->transfer_len - len);
241 	ctrl->nr_changed_ns = 0;
242 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR);
243 	mutex_unlock(&ctrl->lock);
244 out:
245 	nvmet_req_complete(req, status);
246 }
247 
248 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid,
249 		struct nvme_ana_group_desc *desc)
250 {
251 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
252 	struct nvmet_ns *ns;
253 	unsigned long idx;
254 	u32 count = 0;
255 
256 	if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) {
257 		xa_for_each(&ctrl->subsys->namespaces, idx, ns)
258 			if (ns->anagrpid == grpid)
259 				desc->nsids[count++] = cpu_to_le32(ns->nsid);
260 	}
261 
262 	desc->grpid = cpu_to_le32(grpid);
263 	desc->nnsids = cpu_to_le32(count);
264 	desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
265 	desc->state = req->port->ana_state[grpid];
266 	memset(desc->rsvd17, 0, sizeof(desc->rsvd17));
267 	return sizeof(struct nvme_ana_group_desc) + count * sizeof(__le32);
268 }
269 
270 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)
271 {
272 	struct nvme_ana_rsp_hdr hdr = { 0, };
273 	struct nvme_ana_group_desc *desc;
274 	size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */
275 	size_t len;
276 	u32 grpid;
277 	u16 ngrps = 0;
278 	u16 status;
279 
280 	status = NVME_SC_INTERNAL;
281 	desc = kmalloc(sizeof(struct nvme_ana_group_desc) +
282 			NVMET_MAX_NAMESPACES * sizeof(__le32), GFP_KERNEL);
283 	if (!desc)
284 		goto out;
285 
286 	down_read(&nvmet_ana_sem);
287 	for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
288 		if (!nvmet_ana_group_enabled[grpid])
289 			continue;
290 		len = nvmet_format_ana_group(req, grpid, desc);
291 		status = nvmet_copy_to_sgl(req, offset, desc, len);
292 		if (status)
293 			break;
294 		offset += len;
295 		ngrps++;
296 	}
297 	for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
298 		if (nvmet_ana_group_enabled[grpid])
299 			ngrps++;
300 	}
301 
302 	hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
303 	hdr.ngrps = cpu_to_le16(ngrps);
304 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE);
305 	up_read(&nvmet_ana_sem);
306 
307 	kfree(desc);
308 
309 	/* copy the header last once we know the number of groups */
310 	status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr));
311 out:
312 	nvmet_req_complete(req, status);
313 }
314 
315 static void nvmet_execute_get_log_page(struct nvmet_req *req)
316 {
317 	if (!nvmet_check_transfer_len(req, nvmet_get_log_page_len(req->cmd)))
318 		return;
319 
320 	switch (req->cmd->get_log_page.lid) {
321 	case NVME_LOG_ERROR:
322 		return nvmet_execute_get_log_page_error(req);
323 	case NVME_LOG_SMART:
324 		return nvmet_execute_get_log_page_smart(req);
325 	case NVME_LOG_FW_SLOT:
326 		/*
327 		 * We only support a single firmware slot which always is
328 		 * active, so we can zero out the whole firmware slot log and
329 		 * still claim to fully implement this mandatory log page.
330 		 */
331 		return nvmet_execute_get_log_page_noop(req);
332 	case NVME_LOG_CHANGED_NS:
333 		return nvmet_execute_get_log_changed_ns(req);
334 	case NVME_LOG_CMD_EFFECTS:
335 		return nvmet_execute_get_log_cmd_effects_ns(req);
336 	case NVME_LOG_ANA:
337 		return nvmet_execute_get_log_page_ana(req);
338 	}
339 	pr_debug("unhandled lid %d on qid %d\n",
340 	       req->cmd->get_log_page.lid, req->sq->qid);
341 	req->error_loc = offsetof(struct nvme_get_log_page_command, lid);
342 	nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
343 }
344 
345 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
346 {
347 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
348 	struct nvmet_subsys *subsys = ctrl->subsys;
349 	struct nvme_id_ctrl *id;
350 	u32 cmd_capsule_size;
351 	u16 status = 0;
352 
353 	if (!subsys->subsys_discovered) {
354 		mutex_lock(&subsys->lock);
355 		subsys->subsys_discovered = true;
356 		mutex_unlock(&subsys->lock);
357 	}
358 
359 	id = kzalloc(sizeof(*id), GFP_KERNEL);
360 	if (!id) {
361 		status = NVME_SC_INTERNAL;
362 		goto out;
363 	}
364 
365 	/* XXX: figure out how to assign real vendors IDs. */
366 	id->vid = 0;
367 	id->ssvid = 0;
368 
369 	memcpy(id->sn, ctrl->subsys->serial, NVMET_SN_MAX_SIZE);
370 	memcpy_and_pad(id->mn, sizeof(id->mn), subsys->model_number,
371 		       strlen(subsys->model_number), ' ');
372 	memcpy_and_pad(id->fr, sizeof(id->fr),
373 		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
374 
375 	id->rab = 6;
376 
377 	/*
378 	 * XXX: figure out how we can assign a IEEE OUI, but until then
379 	 * the safest is to leave it as zeroes.
380 	 */
381 
382 	/* we support multiple ports, multiples hosts and ANA: */
383 	id->cmic = (1 << 0) | (1 << 1) | (1 << 3);
384 
385 	/* Limit MDTS according to transport capability */
386 	if (ctrl->ops->get_mdts)
387 		id->mdts = ctrl->ops->get_mdts(ctrl);
388 	else
389 		id->mdts = 0;
390 
391 	id->cntlid = cpu_to_le16(ctrl->cntlid);
392 	id->ver = cpu_to_le32(ctrl->subsys->ver);
393 
394 	/* XXX: figure out what to do about RTD3R/RTD3 */
395 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
396 	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
397 		NVME_CTRL_ATTR_TBKAS);
398 
399 	id->oacs = 0;
400 
401 	/*
402 	 * We don't really have a practical limit on the number of abort
403 	 * comands.  But we don't do anything useful for abort either, so
404 	 * no point in allowing more abort commands than the spec requires.
405 	 */
406 	id->acl = 3;
407 
408 	id->aerl = NVMET_ASYNC_EVENTS - 1;
409 
410 	/* first slot is read-only, only one slot supported */
411 	id->frmw = (1 << 0) | (1 << 1);
412 	id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
413 	id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
414 	id->npss = 0;
415 
416 	/* We support keep-alive timeout in granularity of seconds */
417 	id->kas = cpu_to_le16(NVMET_KAS);
418 
419 	id->sqes = (0x6 << 4) | 0x6;
420 	id->cqes = (0x4 << 4) | 0x4;
421 
422 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
423 	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
424 
425 	id->nn = cpu_to_le32(NVMET_MAX_NAMESPACES);
426 	id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
427 	id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
428 			NVME_CTRL_ONCS_WRITE_ZEROES);
429 
430 	/* XXX: don't report vwc if the underlying device is write through */
431 	id->vwc = NVME_CTRL_VWC_PRESENT;
432 
433 	/*
434 	 * We can't support atomic writes bigger than a LBA without support
435 	 * from the backend device.
436 	 */
437 	id->awun = 0;
438 	id->awupf = 0;
439 
440 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
441 	if (ctrl->ops->flags & NVMF_KEYED_SGLS)
442 		id->sgls |= cpu_to_le32(1 << 2);
443 	if (req->port->inline_data_size)
444 		id->sgls |= cpu_to_le32(1 << 20);
445 
446 	strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
447 
448 	/*
449 	 * Max command capsule size is sqe + in-capsule data size.
450 	 * Disable in-capsule data for Metadata capable controllers.
451 	 */
452 	cmd_capsule_size = sizeof(struct nvme_command);
453 	if (!ctrl->pi_support)
454 		cmd_capsule_size += req->port->inline_data_size;
455 	id->ioccsz = cpu_to_le32(cmd_capsule_size / 16);
456 
457 	/* Max response capsule size is cqe */
458 	id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
459 
460 	id->msdbd = ctrl->ops->msdbd;
461 
462 	id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
463 	id->anatt = 10; /* random value */
464 	id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
465 	id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
466 
467 	/*
468 	 * Meh, we don't really support any power state.  Fake up the same
469 	 * values that qemu does.
470 	 */
471 	id->psd[0].max_power = cpu_to_le16(0x9c4);
472 	id->psd[0].entry_lat = cpu_to_le32(0x10);
473 	id->psd[0].exit_lat = cpu_to_le32(0x4);
474 
475 	id->nwpc = 1 << 0; /* write protect and no write protect */
476 
477 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
478 
479 	kfree(id);
480 out:
481 	nvmet_req_complete(req, status);
482 }
483 
484 static void nvmet_execute_identify_ns(struct nvmet_req *req)
485 {
486 	struct nvme_id_ns *id;
487 	u16 status;
488 
489 	if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
490 		req->error_loc = offsetof(struct nvme_identify, nsid);
491 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
492 		goto out;
493 	}
494 
495 	id = kzalloc(sizeof(*id), GFP_KERNEL);
496 	if (!id) {
497 		status = NVME_SC_INTERNAL;
498 		goto out;
499 	}
500 
501 	/* return an all zeroed buffer if we can't find an active namespace */
502 	status = nvmet_req_find_ns(req);
503 	if (status) {
504 		status = 0;
505 		goto done;
506 	}
507 
508 	nvmet_ns_revalidate(req->ns);
509 
510 	/*
511 	 * nuse = ncap = nsze isn't always true, but we have no way to find
512 	 * that out from the underlying device.
513 	 */
514 	id->ncap = id->nsze =
515 		cpu_to_le64(req->ns->size >> req->ns->blksize_shift);
516 	switch (req->port->ana_state[req->ns->anagrpid]) {
517 	case NVME_ANA_INACCESSIBLE:
518 	case NVME_ANA_PERSISTENT_LOSS:
519 		break;
520 	default:
521 		id->nuse = id->nsze;
522 		break;
523 	}
524 
525 	if (req->ns->bdev)
526 		nvmet_bdev_set_limits(req->ns->bdev, id);
527 
528 	/*
529 	 * We just provide a single LBA format that matches what the
530 	 * underlying device reports.
531 	 */
532 	id->nlbaf = 0;
533 	id->flbas = 0;
534 
535 	/*
536 	 * Our namespace might always be shared.  Not just with other
537 	 * controllers, but also with any other user of the block device.
538 	 */
539 	id->nmic = (1 << 0);
540 	id->anagrpid = cpu_to_le32(req->ns->anagrpid);
541 
542 	memcpy(&id->nguid, &req->ns->nguid, sizeof(id->nguid));
543 
544 	id->lbaf[0].ds = req->ns->blksize_shift;
545 
546 	if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns)) {
547 		id->dpc = NVME_NS_DPC_PI_FIRST | NVME_NS_DPC_PI_LAST |
548 			  NVME_NS_DPC_PI_TYPE1 | NVME_NS_DPC_PI_TYPE2 |
549 			  NVME_NS_DPC_PI_TYPE3;
550 		id->mc = NVME_MC_EXTENDED_LBA;
551 		id->dps = req->ns->pi_type;
552 		id->flbas = NVME_NS_FLBAS_META_EXT;
553 		id->lbaf[0].ms = cpu_to_le16(req->ns->metadata_size);
554 	}
555 
556 	if (req->ns->readonly)
557 		id->nsattr |= (1 << 0);
558 done:
559 	if (!status)
560 		status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
561 
562 	kfree(id);
563 out:
564 	nvmet_req_complete(req, status);
565 }
566 
567 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
568 {
569 	static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
570 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
571 	struct nvmet_ns *ns;
572 	unsigned long idx;
573 	u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
574 	__le32 *list;
575 	u16 status = 0;
576 	int i = 0;
577 
578 	list = kzalloc(buf_size, GFP_KERNEL);
579 	if (!list) {
580 		status = NVME_SC_INTERNAL;
581 		goto out;
582 	}
583 
584 	xa_for_each(&ctrl->subsys->namespaces, idx, ns) {
585 		if (ns->nsid <= min_nsid)
586 			continue;
587 		list[i++] = cpu_to_le32(ns->nsid);
588 		if (i == buf_size / sizeof(__le32))
589 			break;
590 	}
591 
592 	status = nvmet_copy_to_sgl(req, 0, list, buf_size);
593 
594 	kfree(list);
595 out:
596 	nvmet_req_complete(req, status);
597 }
598 
599 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
600 				    void *id, off_t *off)
601 {
602 	struct nvme_ns_id_desc desc = {
603 		.nidt = type,
604 		.nidl = len,
605 	};
606 	u16 status;
607 
608 	status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
609 	if (status)
610 		return status;
611 	*off += sizeof(desc);
612 
613 	status = nvmet_copy_to_sgl(req, *off, id, len);
614 	if (status)
615 		return status;
616 	*off += len;
617 
618 	return 0;
619 }
620 
621 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
622 {
623 	off_t off = 0;
624 	u16 status;
625 
626 	status = nvmet_req_find_ns(req);
627 	if (status)
628 		goto out;
629 
630 	if (memchr_inv(&req->ns->uuid, 0, sizeof(req->ns->uuid))) {
631 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
632 						  NVME_NIDT_UUID_LEN,
633 						  &req->ns->uuid, &off);
634 		if (status)
635 			goto out;
636 	}
637 	if (memchr_inv(req->ns->nguid, 0, sizeof(req->ns->nguid))) {
638 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
639 						  NVME_NIDT_NGUID_LEN,
640 						  &req->ns->nguid, &off);
641 		if (status)
642 			goto out;
643 	}
644 
645 	status = nvmet_copy_ns_identifier(req, NVME_NIDT_CSI,
646 					  NVME_NIDT_CSI_LEN,
647 					  &req->ns->csi, &off);
648 	if (status)
649 		goto out;
650 
651 	if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
652 			off) != NVME_IDENTIFY_DATA_SIZE - off)
653 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
654 
655 out:
656 	nvmet_req_complete(req, status);
657 }
658 
659 static bool nvmet_handle_identify_desclist(struct nvmet_req *req)
660 {
661 	switch (req->cmd->identify.csi) {
662 	case NVME_CSI_NVM:
663 		nvmet_execute_identify_desclist(req);
664 		return true;
665 	case NVME_CSI_ZNS:
666 		if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
667 			nvmet_execute_identify_desclist(req);
668 			return true;
669 		}
670 		return false;
671 	default:
672 		return false;
673 	}
674 }
675 
676 static void nvmet_execute_identify(struct nvmet_req *req)
677 {
678 	if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE))
679 		return;
680 
681 	switch (req->cmd->identify.cns) {
682 	case NVME_ID_CNS_NS:
683 		switch (req->cmd->identify.csi) {
684 		case NVME_CSI_NVM:
685 			return nvmet_execute_identify_ns(req);
686 		default:
687 			break;
688 		}
689 		break;
690 	case NVME_ID_CNS_CS_NS:
691 		if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
692 			switch (req->cmd->identify.csi) {
693 			case NVME_CSI_ZNS:
694 				return nvmet_execute_identify_cns_cs_ns(req);
695 			default:
696 				break;
697 			}
698 		}
699 		break;
700 	case NVME_ID_CNS_CTRL:
701 		switch (req->cmd->identify.csi) {
702 		case NVME_CSI_NVM:
703 			return nvmet_execute_identify_ctrl(req);
704 		}
705 		break;
706 	case NVME_ID_CNS_CS_CTRL:
707 		if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
708 			switch (req->cmd->identify.csi) {
709 			case NVME_CSI_ZNS:
710 				return nvmet_execute_identify_cns_cs_ctrl(req);
711 			default:
712 				break;
713 			}
714 		}
715 		break;
716 	case NVME_ID_CNS_NS_ACTIVE_LIST:
717 		switch (req->cmd->identify.csi) {
718 		case NVME_CSI_NVM:
719 			return nvmet_execute_identify_nslist(req);
720 		default:
721 			break;
722 		}
723 		break;
724 	case NVME_ID_CNS_NS_DESC_LIST:
725 		if (nvmet_handle_identify_desclist(req) == true)
726 			return;
727 		break;
728 	}
729 
730 	nvmet_req_cns_error_complete(req);
731 }
732 
733 /*
734  * A "minimum viable" abort implementation: the command is mandatory in the
735  * spec, but we are not required to do any useful work.  We couldn't really
736  * do a useful abort, so don't bother even with waiting for the command
737  * to be exectuted and return immediately telling the command to abort
738  * wasn't found.
739  */
740 static void nvmet_execute_abort(struct nvmet_req *req)
741 {
742 	if (!nvmet_check_transfer_len(req, 0))
743 		return;
744 	nvmet_set_result(req, 1);
745 	nvmet_req_complete(req, 0);
746 }
747 
748 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
749 {
750 	u16 status;
751 
752 	if (req->ns->file)
753 		status = nvmet_file_flush(req);
754 	else
755 		status = nvmet_bdev_flush(req);
756 
757 	if (status)
758 		pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
759 	return status;
760 }
761 
762 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req)
763 {
764 	u32 write_protect = le32_to_cpu(req->cmd->common.cdw11);
765 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
766 	u16 status;
767 
768 	status = nvmet_req_find_ns(req);
769 	if (status)
770 		return status;
771 
772 	mutex_lock(&subsys->lock);
773 	switch (write_protect) {
774 	case NVME_NS_WRITE_PROTECT:
775 		req->ns->readonly = true;
776 		status = nvmet_write_protect_flush_sync(req);
777 		if (status)
778 			req->ns->readonly = false;
779 		break;
780 	case NVME_NS_NO_WRITE_PROTECT:
781 		req->ns->readonly = false;
782 		status = 0;
783 		break;
784 	default:
785 		break;
786 	}
787 
788 	if (!status)
789 		nvmet_ns_changed(subsys, req->ns->nsid);
790 	mutex_unlock(&subsys->lock);
791 	return status;
792 }
793 
794 u16 nvmet_set_feat_kato(struct nvmet_req *req)
795 {
796 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
797 
798 	nvmet_stop_keep_alive_timer(req->sq->ctrl);
799 	req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
800 	nvmet_start_keep_alive_timer(req->sq->ctrl);
801 
802 	nvmet_set_result(req, req->sq->ctrl->kato);
803 
804 	return 0;
805 }
806 
807 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask)
808 {
809 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
810 
811 	if (val32 & ~mask) {
812 		req->error_loc = offsetof(struct nvme_common_command, cdw11);
813 		return NVME_SC_INVALID_FIELD | NVME_SC_DNR;
814 	}
815 
816 	WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
817 	nvmet_set_result(req, val32);
818 
819 	return 0;
820 }
821 
822 void nvmet_execute_set_features(struct nvmet_req *req)
823 {
824 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
825 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
826 	u32 cdw11 = le32_to_cpu(req->cmd->common.cdw11);
827 	u16 status = 0;
828 	u16 nsqr;
829 	u16 ncqr;
830 
831 	if (!nvmet_check_transfer_len(req, 0))
832 		return;
833 
834 	switch (cdw10 & 0xff) {
835 	case NVME_FEAT_NUM_QUEUES:
836 		ncqr = (cdw11 >> 16) & 0xffff;
837 		nsqr = cdw11 & 0xffff;
838 		if (ncqr == 0xffff || nsqr == 0xffff) {
839 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
840 			break;
841 		}
842 		nvmet_set_result(req,
843 			(subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
844 		break;
845 	case NVME_FEAT_KATO:
846 		status = nvmet_set_feat_kato(req);
847 		break;
848 	case NVME_FEAT_ASYNC_EVENT:
849 		status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL);
850 		break;
851 	case NVME_FEAT_HOST_ID:
852 		status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
853 		break;
854 	case NVME_FEAT_WRITE_PROTECT:
855 		status = nvmet_set_feat_write_protect(req);
856 		break;
857 	default:
858 		req->error_loc = offsetof(struct nvme_common_command, cdw10);
859 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
860 		break;
861 	}
862 
863 	nvmet_req_complete(req, status);
864 }
865 
866 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req)
867 {
868 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
869 	u32 result;
870 
871 	result = nvmet_req_find_ns(req);
872 	if (result)
873 		return result;
874 
875 	mutex_lock(&subsys->lock);
876 	if (req->ns->readonly == true)
877 		result = NVME_NS_WRITE_PROTECT;
878 	else
879 		result = NVME_NS_NO_WRITE_PROTECT;
880 	nvmet_set_result(req, result);
881 	mutex_unlock(&subsys->lock);
882 
883 	return 0;
884 }
885 
886 void nvmet_get_feat_kato(struct nvmet_req *req)
887 {
888 	nvmet_set_result(req, req->sq->ctrl->kato * 1000);
889 }
890 
891 void nvmet_get_feat_async_event(struct nvmet_req *req)
892 {
893 	nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
894 }
895 
896 void nvmet_execute_get_features(struct nvmet_req *req)
897 {
898 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
899 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
900 	u16 status = 0;
901 
902 	if (!nvmet_check_transfer_len(req, nvmet_feat_data_len(req, cdw10)))
903 		return;
904 
905 	switch (cdw10 & 0xff) {
906 	/*
907 	 * These features are mandatory in the spec, but we don't
908 	 * have a useful way to implement them.  We'll eventually
909 	 * need to come up with some fake values for these.
910 	 */
911 #if 0
912 	case NVME_FEAT_ARBITRATION:
913 		break;
914 	case NVME_FEAT_POWER_MGMT:
915 		break;
916 	case NVME_FEAT_TEMP_THRESH:
917 		break;
918 	case NVME_FEAT_ERR_RECOVERY:
919 		break;
920 	case NVME_FEAT_IRQ_COALESCE:
921 		break;
922 	case NVME_FEAT_IRQ_CONFIG:
923 		break;
924 	case NVME_FEAT_WRITE_ATOMIC:
925 		break;
926 #endif
927 	case NVME_FEAT_ASYNC_EVENT:
928 		nvmet_get_feat_async_event(req);
929 		break;
930 	case NVME_FEAT_VOLATILE_WC:
931 		nvmet_set_result(req, 1);
932 		break;
933 	case NVME_FEAT_NUM_QUEUES:
934 		nvmet_set_result(req,
935 			(subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
936 		break;
937 	case NVME_FEAT_KATO:
938 		nvmet_get_feat_kato(req);
939 		break;
940 	case NVME_FEAT_HOST_ID:
941 		/* need 128-bit host identifier flag */
942 		if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) {
943 			req->error_loc =
944 				offsetof(struct nvme_common_command, cdw11);
945 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
946 			break;
947 		}
948 
949 		status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
950 				sizeof(req->sq->ctrl->hostid));
951 		break;
952 	case NVME_FEAT_WRITE_PROTECT:
953 		status = nvmet_get_feat_write_protect(req);
954 		break;
955 	default:
956 		req->error_loc =
957 			offsetof(struct nvme_common_command, cdw10);
958 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
959 		break;
960 	}
961 
962 	nvmet_req_complete(req, status);
963 }
964 
965 void nvmet_execute_async_event(struct nvmet_req *req)
966 {
967 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
968 
969 	if (!nvmet_check_transfer_len(req, 0))
970 		return;
971 
972 	mutex_lock(&ctrl->lock);
973 	if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
974 		mutex_unlock(&ctrl->lock);
975 		nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
976 		return;
977 	}
978 	ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
979 	mutex_unlock(&ctrl->lock);
980 
981 	schedule_work(&ctrl->async_event_work);
982 }
983 
984 void nvmet_execute_keep_alive(struct nvmet_req *req)
985 {
986 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
987 	u16 status = 0;
988 
989 	if (!nvmet_check_transfer_len(req, 0))
990 		return;
991 
992 	if (!ctrl->kato) {
993 		status = NVME_SC_KA_TIMEOUT_INVALID;
994 		goto out;
995 	}
996 
997 	pr_debug("ctrl %d update keep-alive timer for %d secs\n",
998 		ctrl->cntlid, ctrl->kato);
999 	mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
1000 out:
1001 	nvmet_req_complete(req, status);
1002 }
1003 
1004 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
1005 {
1006 	struct nvme_command *cmd = req->cmd;
1007 	u16 ret;
1008 
1009 	if (nvme_is_fabrics(cmd))
1010 		return nvmet_parse_fabrics_cmd(req);
1011 	if (nvmet_req_subsys(req)->type == NVME_NQN_DISC)
1012 		return nvmet_parse_discovery_cmd(req);
1013 
1014 	ret = nvmet_check_ctrl_status(req);
1015 	if (unlikely(ret))
1016 		return ret;
1017 
1018 	if (nvmet_is_passthru_req(req))
1019 		return nvmet_parse_passthru_admin_cmd(req);
1020 
1021 	switch (cmd->common.opcode) {
1022 	case nvme_admin_get_log_page:
1023 		req->execute = nvmet_execute_get_log_page;
1024 		return 0;
1025 	case nvme_admin_identify:
1026 		req->execute = nvmet_execute_identify;
1027 		return 0;
1028 	case nvme_admin_abort_cmd:
1029 		req->execute = nvmet_execute_abort;
1030 		return 0;
1031 	case nvme_admin_set_features:
1032 		req->execute = nvmet_execute_set_features;
1033 		return 0;
1034 	case nvme_admin_get_features:
1035 		req->execute = nvmet_execute_get_features;
1036 		return 0;
1037 	case nvme_admin_async_event:
1038 		req->execute = nvmet_execute_async_event;
1039 		return 0;
1040 	case nvme_admin_keep_alive:
1041 		req->execute = nvmet_execute_keep_alive;
1042 		return 0;
1043 	default:
1044 		return nvmet_report_invalid_opcode(req);
1045 	}
1046 }
1047