xref: /openbmc/linux/drivers/nvme/target/admin-cmd.c (revision dbdc671d)
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_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
166 {
167 	u16 status = NVME_SC_INTERNAL;
168 	struct nvme_effects_log *log;
169 
170 	log = kzalloc(sizeof(*log), GFP_KERNEL);
171 	if (!log)
172 		goto out;
173 
174 	log->acs[nvme_admin_get_log_page]	= cpu_to_le32(1 << 0);
175 	log->acs[nvme_admin_identify]		= cpu_to_le32(1 << 0);
176 	log->acs[nvme_admin_abort_cmd]		= cpu_to_le32(1 << 0);
177 	log->acs[nvme_admin_set_features]	= cpu_to_le32(1 << 0);
178 	log->acs[nvme_admin_get_features]	= cpu_to_le32(1 << 0);
179 	log->acs[nvme_admin_async_event]	= cpu_to_le32(1 << 0);
180 	log->acs[nvme_admin_keep_alive]		= cpu_to_le32(1 << 0);
181 
182 	log->iocs[nvme_cmd_read]		= cpu_to_le32(1 << 0);
183 	log->iocs[nvme_cmd_write]		= cpu_to_le32(1 << 0);
184 	log->iocs[nvme_cmd_flush]		= cpu_to_le32(1 << 0);
185 	log->iocs[nvme_cmd_dsm]			= cpu_to_le32(1 << 0);
186 	log->iocs[nvme_cmd_write_zeroes]	= cpu_to_le32(1 << 0);
187 
188 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
189 
190 	kfree(log);
191 out:
192 	nvmet_req_complete(req, status);
193 }
194 
195 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
196 {
197 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
198 	u16 status = NVME_SC_INTERNAL;
199 	size_t len;
200 
201 	if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
202 		goto out;
203 
204 	mutex_lock(&ctrl->lock);
205 	if (ctrl->nr_changed_ns == U32_MAX)
206 		len = sizeof(__le32);
207 	else
208 		len = ctrl->nr_changed_ns * sizeof(__le32);
209 	status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
210 	if (!status)
211 		status = nvmet_zero_sgl(req, len, req->transfer_len - len);
212 	ctrl->nr_changed_ns = 0;
213 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR);
214 	mutex_unlock(&ctrl->lock);
215 out:
216 	nvmet_req_complete(req, status);
217 }
218 
219 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid,
220 		struct nvme_ana_group_desc *desc)
221 {
222 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
223 	struct nvmet_ns *ns;
224 	unsigned long idx;
225 	u32 count = 0;
226 
227 	if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) {
228 		xa_for_each(&ctrl->subsys->namespaces, idx, ns)
229 			if (ns->anagrpid == grpid)
230 				desc->nsids[count++] = cpu_to_le32(ns->nsid);
231 	}
232 
233 	desc->grpid = cpu_to_le32(grpid);
234 	desc->nnsids = cpu_to_le32(count);
235 	desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
236 	desc->state = req->port->ana_state[grpid];
237 	memset(desc->rsvd17, 0, sizeof(desc->rsvd17));
238 	return sizeof(struct nvme_ana_group_desc) + count * sizeof(__le32);
239 }
240 
241 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)
242 {
243 	struct nvme_ana_rsp_hdr hdr = { 0, };
244 	struct nvme_ana_group_desc *desc;
245 	size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */
246 	size_t len;
247 	u32 grpid;
248 	u16 ngrps = 0;
249 	u16 status;
250 
251 	status = NVME_SC_INTERNAL;
252 	desc = kmalloc(sizeof(struct nvme_ana_group_desc) +
253 			NVMET_MAX_NAMESPACES * sizeof(__le32), GFP_KERNEL);
254 	if (!desc)
255 		goto out;
256 
257 	down_read(&nvmet_ana_sem);
258 	for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
259 		if (!nvmet_ana_group_enabled[grpid])
260 			continue;
261 		len = nvmet_format_ana_group(req, grpid, desc);
262 		status = nvmet_copy_to_sgl(req, offset, desc, len);
263 		if (status)
264 			break;
265 		offset += len;
266 		ngrps++;
267 	}
268 	for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
269 		if (nvmet_ana_group_enabled[grpid])
270 			ngrps++;
271 	}
272 
273 	hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
274 	hdr.ngrps = cpu_to_le16(ngrps);
275 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE);
276 	up_read(&nvmet_ana_sem);
277 
278 	kfree(desc);
279 
280 	/* copy the header last once we know the number of groups */
281 	status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr));
282 out:
283 	nvmet_req_complete(req, status);
284 }
285 
286 static void nvmet_execute_get_log_page(struct nvmet_req *req)
287 {
288 	if (!nvmet_check_transfer_len(req, nvmet_get_log_page_len(req->cmd)))
289 		return;
290 
291 	switch (req->cmd->get_log_page.lid) {
292 	case NVME_LOG_ERROR:
293 		return nvmet_execute_get_log_page_error(req);
294 	case NVME_LOG_SMART:
295 		return nvmet_execute_get_log_page_smart(req);
296 	case NVME_LOG_FW_SLOT:
297 		/*
298 		 * We only support a single firmware slot which always is
299 		 * active, so we can zero out the whole firmware slot log and
300 		 * still claim to fully implement this mandatory log page.
301 		 */
302 		return nvmet_execute_get_log_page_noop(req);
303 	case NVME_LOG_CHANGED_NS:
304 		return nvmet_execute_get_log_changed_ns(req);
305 	case NVME_LOG_CMD_EFFECTS:
306 		return nvmet_execute_get_log_cmd_effects_ns(req);
307 	case NVME_LOG_ANA:
308 		return nvmet_execute_get_log_page_ana(req);
309 	}
310 	pr_err("unhandled lid %d on qid %d\n",
311 	       req->cmd->get_log_page.lid, req->sq->qid);
312 	req->error_loc = offsetof(struct nvme_get_log_page_command, lid);
313 	nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
314 }
315 
316 static u16 nvmet_set_model_number(struct nvmet_subsys *subsys)
317 {
318 	u16 status = 0;
319 
320 	mutex_lock(&subsys->lock);
321 	if (!subsys->model_number) {
322 		subsys->model_number =
323 			kstrdup(NVMET_DEFAULT_CTRL_MODEL, GFP_KERNEL);
324 		if (!subsys->model_number)
325 			status = NVME_SC_INTERNAL;
326 	}
327 	mutex_unlock(&subsys->lock);
328 
329 	return status;
330 }
331 
332 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
333 {
334 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
335 	struct nvmet_subsys *subsys = ctrl->subsys;
336 	struct nvme_id_ctrl *id;
337 	u32 cmd_capsule_size;
338 	u16 status = 0;
339 
340 	/*
341 	 * If there is no model number yet, set it now.  It will then remain
342 	 * stable for the life time of the subsystem.
343 	 */
344 	if (!subsys->model_number) {
345 		status = nvmet_set_model_number(subsys);
346 		if (status)
347 			goto out;
348 	}
349 
350 	id = kzalloc(sizeof(*id), GFP_KERNEL);
351 	if (!id) {
352 		status = NVME_SC_INTERNAL;
353 		goto out;
354 	}
355 
356 	/* XXX: figure out how to assign real vendors IDs. */
357 	id->vid = 0;
358 	id->ssvid = 0;
359 
360 	memset(id->sn, ' ', sizeof(id->sn));
361 	bin2hex(id->sn, &ctrl->subsys->serial,
362 		min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
363 	memcpy_and_pad(id->mn, sizeof(id->mn), subsys->model_number,
364 		       strlen(subsys->model_number), ' ');
365 	memcpy_and_pad(id->fr, sizeof(id->fr),
366 		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
367 
368 	id->rab = 6;
369 
370 	/*
371 	 * XXX: figure out how we can assign a IEEE OUI, but until then
372 	 * the safest is to leave it as zeroes.
373 	 */
374 
375 	/* we support multiple ports, multiples hosts and ANA: */
376 	id->cmic = (1 << 0) | (1 << 1) | (1 << 3);
377 
378 	/* Limit MDTS according to transport capability */
379 	if (ctrl->ops->get_mdts)
380 		id->mdts = ctrl->ops->get_mdts(ctrl);
381 	else
382 		id->mdts = 0;
383 
384 	id->cntlid = cpu_to_le16(ctrl->cntlid);
385 	id->ver = cpu_to_le32(ctrl->subsys->ver);
386 
387 	/* XXX: figure out what to do about RTD3R/RTD3 */
388 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
389 	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
390 		NVME_CTRL_ATTR_TBKAS);
391 
392 	id->oacs = 0;
393 
394 	/*
395 	 * We don't really have a practical limit on the number of abort
396 	 * comands.  But we don't do anything useful for abort either, so
397 	 * no point in allowing more abort commands than the spec requires.
398 	 */
399 	id->acl = 3;
400 
401 	id->aerl = NVMET_ASYNC_EVENTS - 1;
402 
403 	/* first slot is read-only, only one slot supported */
404 	id->frmw = (1 << 0) | (1 << 1);
405 	id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
406 	id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
407 	id->npss = 0;
408 
409 	/* We support keep-alive timeout in granularity of seconds */
410 	id->kas = cpu_to_le16(NVMET_KAS);
411 
412 	id->sqes = (0x6 << 4) | 0x6;
413 	id->cqes = (0x4 << 4) | 0x4;
414 
415 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
416 	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
417 
418 	id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
419 	id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
420 	id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
421 			NVME_CTRL_ONCS_WRITE_ZEROES);
422 
423 	/* XXX: don't report vwc if the underlying device is write through */
424 	id->vwc = NVME_CTRL_VWC_PRESENT;
425 
426 	/*
427 	 * We can't support atomic writes bigger than a LBA without support
428 	 * from the backend device.
429 	 */
430 	id->awun = 0;
431 	id->awupf = 0;
432 
433 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
434 	if (ctrl->ops->flags & NVMF_KEYED_SGLS)
435 		id->sgls |= cpu_to_le32(1 << 2);
436 	if (req->port->inline_data_size)
437 		id->sgls |= cpu_to_le32(1 << 20);
438 
439 	strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
440 
441 	/*
442 	 * Max command capsule size is sqe + in-capsule data size.
443 	 * Disable in-capsule data for Metadata capable controllers.
444 	 */
445 	cmd_capsule_size = sizeof(struct nvme_command);
446 	if (!ctrl->pi_support)
447 		cmd_capsule_size += req->port->inline_data_size;
448 	id->ioccsz = cpu_to_le32(cmd_capsule_size / 16);
449 
450 	/* Max response capsule size is cqe */
451 	id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
452 
453 	id->msdbd = ctrl->ops->msdbd;
454 
455 	id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
456 	id->anatt = 10; /* random value */
457 	id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
458 	id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
459 
460 	/*
461 	 * Meh, we don't really support any power state.  Fake up the same
462 	 * values that qemu does.
463 	 */
464 	id->psd[0].max_power = cpu_to_le16(0x9c4);
465 	id->psd[0].entry_lat = cpu_to_le32(0x10);
466 	id->psd[0].exit_lat = cpu_to_le32(0x4);
467 
468 	id->nwpc = 1 << 0; /* write protect and no write protect */
469 
470 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
471 
472 	kfree(id);
473 out:
474 	nvmet_req_complete(req, status);
475 }
476 
477 static void nvmet_execute_identify_ns(struct nvmet_req *req)
478 {
479 	struct nvme_id_ns *id;
480 	u16 status;
481 
482 	if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
483 		req->error_loc = offsetof(struct nvme_identify, nsid);
484 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
485 		goto out;
486 	}
487 
488 	id = kzalloc(sizeof(*id), GFP_KERNEL);
489 	if (!id) {
490 		status = NVME_SC_INTERNAL;
491 		goto out;
492 	}
493 
494 	/* return an all zeroed buffer if we can't find an active namespace */
495 	status = nvmet_req_find_ns(req);
496 	if (status) {
497 		status = 0;
498 		goto done;
499 	}
500 
501 	nvmet_ns_revalidate(req->ns);
502 
503 	/*
504 	 * nuse = ncap = nsze isn't always true, but we have no way to find
505 	 * that out from the underlying device.
506 	 */
507 	id->ncap = id->nsze =
508 		cpu_to_le64(req->ns->size >> req->ns->blksize_shift);
509 	switch (req->port->ana_state[req->ns->anagrpid]) {
510 	case NVME_ANA_INACCESSIBLE:
511 	case NVME_ANA_PERSISTENT_LOSS:
512 		break;
513 	default:
514 		id->nuse = id->nsze;
515 		break;
516         }
517 
518 	if (req->ns->bdev)
519 		nvmet_bdev_set_limits(req->ns->bdev, id);
520 
521 	/*
522 	 * We just provide a single LBA format that matches what the
523 	 * underlying device reports.
524 	 */
525 	id->nlbaf = 0;
526 	id->flbas = 0;
527 
528 	/*
529 	 * Our namespace might always be shared.  Not just with other
530 	 * controllers, but also with any other user of the block device.
531 	 */
532 	id->nmic = (1 << 0);
533 	id->anagrpid = cpu_to_le32(req->ns->anagrpid);
534 
535 	memcpy(&id->nguid, &req->ns->nguid, sizeof(id->nguid));
536 
537 	id->lbaf[0].ds = req->ns->blksize_shift;
538 
539 	if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns)) {
540 		id->dpc = NVME_NS_DPC_PI_FIRST | NVME_NS_DPC_PI_LAST |
541 			  NVME_NS_DPC_PI_TYPE1 | NVME_NS_DPC_PI_TYPE2 |
542 			  NVME_NS_DPC_PI_TYPE3;
543 		id->mc = NVME_MC_EXTENDED_LBA;
544 		id->dps = req->ns->pi_type;
545 		id->flbas = NVME_NS_FLBAS_META_EXT;
546 		id->lbaf[0].ms = cpu_to_le16(req->ns->metadata_size);
547 	}
548 
549 	if (req->ns->readonly)
550 		id->nsattr |= (1 << 0);
551 done:
552 	if (!status)
553 		status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
554 
555 	kfree(id);
556 out:
557 	nvmet_req_complete(req, status);
558 }
559 
560 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
561 {
562 	static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
563 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
564 	struct nvmet_ns *ns;
565 	unsigned long idx;
566 	u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
567 	__le32 *list;
568 	u16 status = 0;
569 	int i = 0;
570 
571 	list = kzalloc(buf_size, GFP_KERNEL);
572 	if (!list) {
573 		status = NVME_SC_INTERNAL;
574 		goto out;
575 	}
576 
577 	xa_for_each(&ctrl->subsys->namespaces, idx, ns) {
578 		if (ns->nsid <= min_nsid)
579 			continue;
580 		list[i++] = cpu_to_le32(ns->nsid);
581 		if (i == buf_size / sizeof(__le32))
582 			break;
583 	}
584 
585 	status = nvmet_copy_to_sgl(req, 0, list, buf_size);
586 
587 	kfree(list);
588 out:
589 	nvmet_req_complete(req, status);
590 }
591 
592 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
593 				    void *id, off_t *off)
594 {
595 	struct nvme_ns_id_desc desc = {
596 		.nidt = type,
597 		.nidl = len,
598 	};
599 	u16 status;
600 
601 	status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
602 	if (status)
603 		return status;
604 	*off += sizeof(desc);
605 
606 	status = nvmet_copy_to_sgl(req, *off, id, len);
607 	if (status)
608 		return status;
609 	*off += len;
610 
611 	return 0;
612 }
613 
614 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
615 {
616 	off_t off = 0;
617 	u16 status;
618 
619 	status = nvmet_req_find_ns(req);
620 	if (status)
621 		goto out;
622 
623 	if (memchr_inv(&req->ns->uuid, 0, sizeof(req->ns->uuid))) {
624 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
625 						  NVME_NIDT_UUID_LEN,
626 						  &req->ns->uuid, &off);
627 		if (status)
628 			goto out;
629 	}
630 	if (memchr_inv(req->ns->nguid, 0, sizeof(req->ns->nguid))) {
631 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
632 						  NVME_NIDT_NGUID_LEN,
633 						  &req->ns->nguid, &off);
634 		if (status)
635 			goto out;
636 	}
637 
638 	if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
639 			off) != NVME_IDENTIFY_DATA_SIZE - off)
640 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
641 
642 out:
643 	nvmet_req_complete(req, status);
644 }
645 
646 static void nvmet_execute_identify(struct nvmet_req *req)
647 {
648 	if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE))
649 		return;
650 
651 	switch (req->cmd->identify.cns) {
652 	case NVME_ID_CNS_NS:
653 		return nvmet_execute_identify_ns(req);
654 	case NVME_ID_CNS_CTRL:
655 		return nvmet_execute_identify_ctrl(req);
656 	case NVME_ID_CNS_NS_ACTIVE_LIST:
657 		return nvmet_execute_identify_nslist(req);
658 	case NVME_ID_CNS_NS_DESC_LIST:
659 		return nvmet_execute_identify_desclist(req);
660 	}
661 
662 	pr_err("unhandled identify cns %d on qid %d\n",
663 	       req->cmd->identify.cns, req->sq->qid);
664 	req->error_loc = offsetof(struct nvme_identify, cns);
665 	nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
666 }
667 
668 /*
669  * A "minimum viable" abort implementation: the command is mandatory in the
670  * spec, but we are not required to do any useful work.  We couldn't really
671  * do a useful abort, so don't bother even with waiting for the command
672  * to be exectuted and return immediately telling the command to abort
673  * wasn't found.
674  */
675 static void nvmet_execute_abort(struct nvmet_req *req)
676 {
677 	if (!nvmet_check_transfer_len(req, 0))
678 		return;
679 	nvmet_set_result(req, 1);
680 	nvmet_req_complete(req, 0);
681 }
682 
683 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
684 {
685 	u16 status;
686 
687 	if (req->ns->file)
688 		status = nvmet_file_flush(req);
689 	else
690 		status = nvmet_bdev_flush(req);
691 
692 	if (status)
693 		pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
694 	return status;
695 }
696 
697 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req)
698 {
699 	u32 write_protect = le32_to_cpu(req->cmd->common.cdw11);
700 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
701 	u16 status;
702 
703 	status = nvmet_req_find_ns(req);
704 	if (status)
705 		return status;
706 
707 	mutex_lock(&subsys->lock);
708 	switch (write_protect) {
709 	case NVME_NS_WRITE_PROTECT:
710 		req->ns->readonly = true;
711 		status = nvmet_write_protect_flush_sync(req);
712 		if (status)
713 			req->ns->readonly = false;
714 		break;
715 	case NVME_NS_NO_WRITE_PROTECT:
716 		req->ns->readonly = false;
717 		status = 0;
718 		break;
719 	default:
720 		break;
721 	}
722 
723 	if (!status)
724 		nvmet_ns_changed(subsys, req->ns->nsid);
725 	mutex_unlock(&subsys->lock);
726 	return status;
727 }
728 
729 u16 nvmet_set_feat_kato(struct nvmet_req *req)
730 {
731 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
732 
733 	nvmet_stop_keep_alive_timer(req->sq->ctrl);
734 	req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
735 	nvmet_start_keep_alive_timer(req->sq->ctrl);
736 
737 	nvmet_set_result(req, req->sq->ctrl->kato);
738 
739 	return 0;
740 }
741 
742 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask)
743 {
744 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
745 
746 	if (val32 & ~mask) {
747 		req->error_loc = offsetof(struct nvme_common_command, cdw11);
748 		return NVME_SC_INVALID_FIELD | NVME_SC_DNR;
749 	}
750 
751 	WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
752 	nvmet_set_result(req, val32);
753 
754 	return 0;
755 }
756 
757 void nvmet_execute_set_features(struct nvmet_req *req)
758 {
759 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
760 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
761 	u32 cdw11 = le32_to_cpu(req->cmd->common.cdw11);
762 	u16 status = 0;
763 	u16 nsqr;
764 	u16 ncqr;
765 
766 	if (!nvmet_check_transfer_len(req, 0))
767 		return;
768 
769 	switch (cdw10 & 0xff) {
770 	case NVME_FEAT_NUM_QUEUES:
771 		ncqr = (cdw11 >> 16) & 0xffff;
772 		nsqr = cdw11 & 0xffff;
773 		if (ncqr == 0xffff || nsqr == 0xffff) {
774 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
775 			break;
776 		}
777 		nvmet_set_result(req,
778 			(subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
779 		break;
780 	case NVME_FEAT_KATO:
781 		status = nvmet_set_feat_kato(req);
782 		break;
783 	case NVME_FEAT_ASYNC_EVENT:
784 		status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL);
785 		break;
786 	case NVME_FEAT_HOST_ID:
787 		status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
788 		break;
789 	case NVME_FEAT_WRITE_PROTECT:
790 		status = nvmet_set_feat_write_protect(req);
791 		break;
792 	default:
793 		req->error_loc = offsetof(struct nvme_common_command, cdw10);
794 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
795 		break;
796 	}
797 
798 	nvmet_req_complete(req, status);
799 }
800 
801 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req)
802 {
803 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
804 	u32 result;
805 
806 	result = nvmet_req_find_ns(req);
807 	if (result)
808 		return result;
809 
810 	mutex_lock(&subsys->lock);
811 	if (req->ns->readonly == true)
812 		result = NVME_NS_WRITE_PROTECT;
813 	else
814 		result = NVME_NS_NO_WRITE_PROTECT;
815 	nvmet_set_result(req, result);
816 	mutex_unlock(&subsys->lock);
817 
818 	return 0;
819 }
820 
821 void nvmet_get_feat_kato(struct nvmet_req *req)
822 {
823 	nvmet_set_result(req, req->sq->ctrl->kato * 1000);
824 }
825 
826 void nvmet_get_feat_async_event(struct nvmet_req *req)
827 {
828 	nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
829 }
830 
831 void nvmet_execute_get_features(struct nvmet_req *req)
832 {
833 	struct nvmet_subsys *subsys = nvmet_req_subsys(req);
834 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
835 	u16 status = 0;
836 
837 	if (!nvmet_check_transfer_len(req, nvmet_feat_data_len(req, cdw10)))
838 		return;
839 
840 	switch (cdw10 & 0xff) {
841 	/*
842 	 * These features are mandatory in the spec, but we don't
843 	 * have a useful way to implement them.  We'll eventually
844 	 * need to come up with some fake values for these.
845 	 */
846 #if 0
847 	case NVME_FEAT_ARBITRATION:
848 		break;
849 	case NVME_FEAT_POWER_MGMT:
850 		break;
851 	case NVME_FEAT_TEMP_THRESH:
852 		break;
853 	case NVME_FEAT_ERR_RECOVERY:
854 		break;
855 	case NVME_FEAT_IRQ_COALESCE:
856 		break;
857 	case NVME_FEAT_IRQ_CONFIG:
858 		break;
859 	case NVME_FEAT_WRITE_ATOMIC:
860 		break;
861 #endif
862 	case NVME_FEAT_ASYNC_EVENT:
863 		nvmet_get_feat_async_event(req);
864 		break;
865 	case NVME_FEAT_VOLATILE_WC:
866 		nvmet_set_result(req, 1);
867 		break;
868 	case NVME_FEAT_NUM_QUEUES:
869 		nvmet_set_result(req,
870 			(subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
871 		break;
872 	case NVME_FEAT_KATO:
873 		nvmet_get_feat_kato(req);
874 		break;
875 	case NVME_FEAT_HOST_ID:
876 		/* need 128-bit host identifier flag */
877 		if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) {
878 			req->error_loc =
879 				offsetof(struct nvme_common_command, cdw11);
880 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
881 			break;
882 		}
883 
884 		status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
885 				sizeof(req->sq->ctrl->hostid));
886 		break;
887 	case NVME_FEAT_WRITE_PROTECT:
888 		status = nvmet_get_feat_write_protect(req);
889 		break;
890 	default:
891 		req->error_loc =
892 			offsetof(struct nvme_common_command, cdw10);
893 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
894 		break;
895 	}
896 
897 	nvmet_req_complete(req, status);
898 }
899 
900 void nvmet_execute_async_event(struct nvmet_req *req)
901 {
902 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
903 
904 	if (!nvmet_check_transfer_len(req, 0))
905 		return;
906 
907 	mutex_lock(&ctrl->lock);
908 	if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
909 		mutex_unlock(&ctrl->lock);
910 		nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
911 		return;
912 	}
913 	ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
914 	mutex_unlock(&ctrl->lock);
915 
916 	schedule_work(&ctrl->async_event_work);
917 }
918 
919 void nvmet_execute_keep_alive(struct nvmet_req *req)
920 {
921 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
922 
923 	if (!nvmet_check_transfer_len(req, 0))
924 		return;
925 
926 	pr_debug("ctrl %d update keep-alive timer for %d secs\n",
927 		ctrl->cntlid, ctrl->kato);
928 
929 	mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
930 	nvmet_req_complete(req, 0);
931 }
932 
933 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
934 {
935 	struct nvme_command *cmd = req->cmd;
936 	u16 ret;
937 
938 	if (nvme_is_fabrics(cmd))
939 		return nvmet_parse_fabrics_cmd(req);
940 	if (nvmet_req_subsys(req)->type == NVME_NQN_DISC)
941 		return nvmet_parse_discovery_cmd(req);
942 
943 	ret = nvmet_check_ctrl_status(req, cmd);
944 	if (unlikely(ret))
945 		return ret;
946 
947 	if (nvmet_req_passthru_ctrl(req))
948 		return nvmet_parse_passthru_admin_cmd(req);
949 
950 	switch (cmd->common.opcode) {
951 	case nvme_admin_get_log_page:
952 		req->execute = nvmet_execute_get_log_page;
953 		return 0;
954 	case nvme_admin_identify:
955 		req->execute = nvmet_execute_identify;
956 		return 0;
957 	case nvme_admin_abort_cmd:
958 		req->execute = nvmet_execute_abort;
959 		return 0;
960 	case nvme_admin_set_features:
961 		req->execute = nvmet_execute_set_features;
962 		return 0;
963 	case nvme_admin_get_features:
964 		req->execute = nvmet_execute_get_features;
965 		return 0;
966 	case nvme_admin_async_event:
967 		req->execute = nvmet_execute_async_event;
968 		return 0;
969 	case nvme_admin_keep_alive:
970 		req->execute = nvmet_execute_keep_alive;
971 		return 0;
972 	}
973 
974 	pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
975 	       req->sq->qid);
976 	req->error_loc = offsetof(struct nvme_common_command, opcode);
977 	return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
978 }
979