xref: /openbmc/linux/drivers/nvme/target/admin-cmd.c (revision 1075b321)
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 
10 #include <generated/utsrelease.h>
11 #include <asm/unaligned.h>
12 #include "nvmet.h"
13 
14 u32 nvmet_get_log_page_len(struct nvme_command *cmd)
15 {
16 	u32 len = le16_to_cpu(cmd->get_log_page.numdu);
17 
18 	len <<= 16;
19 	len += le16_to_cpu(cmd->get_log_page.numdl);
20 	/* NUMD is a 0's based value */
21 	len += 1;
22 	len *= sizeof(u32);
23 
24 	return len;
25 }
26 
27 u64 nvmet_get_log_page_offset(struct nvme_command *cmd)
28 {
29 	return le64_to_cpu(cmd->get_log_page.lpo);
30 }
31 
32 static void nvmet_execute_get_log_page_noop(struct nvmet_req *req)
33 {
34 	nvmet_req_complete(req, nvmet_zero_sgl(req, 0, req->transfer_len));
35 }
36 
37 static void nvmet_execute_get_log_page_error(struct nvmet_req *req)
38 {
39 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
40 	unsigned long flags;
41 	off_t offset = 0;
42 	u64 slot;
43 	u64 i;
44 
45 	spin_lock_irqsave(&ctrl->error_lock, flags);
46 	slot = ctrl->err_counter % NVMET_ERROR_LOG_SLOTS;
47 
48 	for (i = 0; i < NVMET_ERROR_LOG_SLOTS; i++) {
49 		if (nvmet_copy_to_sgl(req, offset, &ctrl->slots[slot],
50 				sizeof(struct nvme_error_slot)))
51 			break;
52 
53 		if (slot == 0)
54 			slot = NVMET_ERROR_LOG_SLOTS - 1;
55 		else
56 			slot--;
57 		offset += sizeof(struct nvme_error_slot);
58 	}
59 	spin_unlock_irqrestore(&ctrl->error_lock, flags);
60 	nvmet_req_complete(req, 0);
61 }
62 
63 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req,
64 		struct nvme_smart_log *slog)
65 {
66 	struct nvmet_ns *ns;
67 	u64 host_reads, host_writes, data_units_read, data_units_written;
68 
69 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid);
70 	if (!ns) {
71 		pr_err("Could not find namespace id : %d\n",
72 				le32_to_cpu(req->cmd->get_log_page.nsid));
73 		req->error_loc = offsetof(struct nvme_rw_command, nsid);
74 		return NVME_SC_INVALID_NS;
75 	}
76 
77 	/* we don't have the right data for file backed ns */
78 	if (!ns->bdev)
79 		goto out;
80 
81 	host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]);
82 	data_units_read = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part,
83 		sectors[READ]), 1000);
84 	host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]);
85 	data_units_written = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part,
86 		sectors[WRITE]), 1000);
87 
88 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
89 	put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
90 	put_unaligned_le64(host_writes, &slog->host_writes[0]);
91 	put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
92 out:
93 	nvmet_put_namespace(ns);
94 
95 	return NVME_SC_SUCCESS;
96 }
97 
98 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
99 		struct nvme_smart_log *slog)
100 {
101 	u64 host_reads = 0, host_writes = 0;
102 	u64 data_units_read = 0, data_units_written = 0;
103 	struct nvmet_ns *ns;
104 	struct nvmet_ctrl *ctrl;
105 
106 	ctrl = req->sq->ctrl;
107 
108 	rcu_read_lock();
109 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
110 		/* we don't have the right data for file backed ns */
111 		if (!ns->bdev)
112 			continue;
113 		host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]);
114 		data_units_read += DIV_ROUND_UP(
115 			part_stat_read(ns->bdev->bd_part, sectors[READ]), 1000);
116 		host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]);
117 		data_units_written += DIV_ROUND_UP(
118 			part_stat_read(ns->bdev->bd_part, sectors[WRITE]), 1000);
119 
120 	}
121 	rcu_read_unlock();
122 
123 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
124 	put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
125 	put_unaligned_le64(host_writes, &slog->host_writes[0]);
126 	put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
127 
128 	return NVME_SC_SUCCESS;
129 }
130 
131 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req)
132 {
133 	struct nvme_smart_log *log;
134 	u16 status = NVME_SC_INTERNAL;
135 	unsigned long flags;
136 
137 	if (req->transfer_len != sizeof(*log))
138 		goto out;
139 
140 	log = kzalloc(sizeof(*log), GFP_KERNEL);
141 	if (!log)
142 		goto out;
143 
144 	if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL))
145 		status = nvmet_get_smart_log_all(req, log);
146 	else
147 		status = nvmet_get_smart_log_nsid(req, log);
148 	if (status)
149 		goto out_free_log;
150 
151 	spin_lock_irqsave(&req->sq->ctrl->error_lock, flags);
152 	put_unaligned_le64(req->sq->ctrl->err_counter,
153 			&log->num_err_log_entries);
154 	spin_unlock_irqrestore(&req->sq->ctrl->error_lock, flags);
155 
156 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
157 out_free_log:
158 	kfree(log);
159 out:
160 	nvmet_req_complete(req, status);
161 }
162 
163 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
164 {
165 	u16 status = NVME_SC_INTERNAL;
166 	struct nvme_effects_log *log;
167 
168 	log = kzalloc(sizeof(*log), GFP_KERNEL);
169 	if (!log)
170 		goto out;
171 
172 	log->acs[nvme_admin_get_log_page]	= cpu_to_le32(1 << 0);
173 	log->acs[nvme_admin_identify]		= cpu_to_le32(1 << 0);
174 	log->acs[nvme_admin_abort_cmd]		= cpu_to_le32(1 << 0);
175 	log->acs[nvme_admin_set_features]	= cpu_to_le32(1 << 0);
176 	log->acs[nvme_admin_get_features]	= cpu_to_le32(1 << 0);
177 	log->acs[nvme_admin_async_event]	= cpu_to_le32(1 << 0);
178 	log->acs[nvme_admin_keep_alive]		= cpu_to_le32(1 << 0);
179 
180 	log->iocs[nvme_cmd_read]		= cpu_to_le32(1 << 0);
181 	log->iocs[nvme_cmd_write]		= cpu_to_le32(1 << 0);
182 	log->iocs[nvme_cmd_flush]		= cpu_to_le32(1 << 0);
183 	log->iocs[nvme_cmd_dsm]			= cpu_to_le32(1 << 0);
184 	log->iocs[nvme_cmd_write_zeroes]	= cpu_to_le32(1 << 0);
185 
186 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
187 
188 	kfree(log);
189 out:
190 	nvmet_req_complete(req, status);
191 }
192 
193 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
194 {
195 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
196 	u16 status = NVME_SC_INTERNAL;
197 	size_t len;
198 
199 	if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
200 		goto out;
201 
202 	mutex_lock(&ctrl->lock);
203 	if (ctrl->nr_changed_ns == U32_MAX)
204 		len = sizeof(__le32);
205 	else
206 		len = ctrl->nr_changed_ns * sizeof(__le32);
207 	status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
208 	if (!status)
209 		status = nvmet_zero_sgl(req, len, req->transfer_len - len);
210 	ctrl->nr_changed_ns = 0;
211 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR);
212 	mutex_unlock(&ctrl->lock);
213 out:
214 	nvmet_req_complete(req, status);
215 }
216 
217 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid,
218 		struct nvme_ana_group_desc *desc)
219 {
220 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
221 	struct nvmet_ns *ns;
222 	u32 count = 0;
223 
224 	if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) {
225 		rcu_read_lock();
226 		list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link)
227 			if (ns->anagrpid == grpid)
228 				desc->nsids[count++] = cpu_to_le32(ns->nsid);
229 		rcu_read_unlock();
230 	}
231 
232 	desc->grpid = cpu_to_le32(grpid);
233 	desc->nnsids = cpu_to_le32(count);
234 	desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
235 	desc->state = req->port->ana_state[grpid];
236 	memset(desc->rsvd17, 0, sizeof(desc->rsvd17));
237 	return sizeof(struct nvme_ana_group_desc) + count * sizeof(__le32);
238 }
239 
240 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)
241 {
242 	struct nvme_ana_rsp_hdr hdr = { 0, };
243 	struct nvme_ana_group_desc *desc;
244 	size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */
245 	size_t len;
246 	u32 grpid;
247 	u16 ngrps = 0;
248 	u16 status;
249 
250 	status = NVME_SC_INTERNAL;
251 	desc = kmalloc(sizeof(struct nvme_ana_group_desc) +
252 			NVMET_MAX_NAMESPACES * sizeof(__le32), GFP_KERNEL);
253 	if (!desc)
254 		goto out;
255 
256 	down_read(&nvmet_ana_sem);
257 	for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
258 		if (!nvmet_ana_group_enabled[grpid])
259 			continue;
260 		len = nvmet_format_ana_group(req, grpid, desc);
261 		status = nvmet_copy_to_sgl(req, offset, desc, len);
262 		if (status)
263 			break;
264 		offset += len;
265 		ngrps++;
266 	}
267 	for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
268 		if (nvmet_ana_group_enabled[grpid])
269 			ngrps++;
270 	}
271 
272 	hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
273 	hdr.ngrps = cpu_to_le16(ngrps);
274 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE);
275 	up_read(&nvmet_ana_sem);
276 
277 	kfree(desc);
278 
279 	/* copy the header last once we know the number of groups */
280 	status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr));
281 out:
282 	nvmet_req_complete(req, status);
283 }
284 
285 static void nvmet_execute_get_log_page(struct nvmet_req *req)
286 {
287 	if (!nvmet_check_data_len(req, nvmet_get_log_page_len(req->cmd)))
288 		return;
289 
290 	switch (req->cmd->get_log_page.lid) {
291 	case NVME_LOG_ERROR:
292 		return nvmet_execute_get_log_page_error(req);
293 	case NVME_LOG_SMART:
294 		return nvmet_execute_get_log_page_smart(req);
295 	case NVME_LOG_FW_SLOT:
296 		/*
297 		 * We only support a single firmware slot which always is
298 		 * active, so we can zero out the whole firmware slot log and
299 		 * still claim to fully implement this mandatory log page.
300 		 */
301 		return nvmet_execute_get_log_page_noop(req);
302 	case NVME_LOG_CHANGED_NS:
303 		return nvmet_execute_get_log_changed_ns(req);
304 	case NVME_LOG_CMD_EFFECTS:
305 		return nvmet_execute_get_log_cmd_effects_ns(req);
306 	case NVME_LOG_ANA:
307 		return nvmet_execute_get_log_page_ana(req);
308 	}
309 	pr_err("unhandled lid %d on qid %d\n",
310 	       req->cmd->get_log_page.lid, req->sq->qid);
311 	req->error_loc = offsetof(struct nvme_get_log_page_command, lid);
312 	nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
313 }
314 
315 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
316 {
317 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
318 	struct nvme_id_ctrl *id;
319 	u16 status = 0;
320 	const char model[] = "Linux";
321 
322 	id = kzalloc(sizeof(*id), GFP_KERNEL);
323 	if (!id) {
324 		status = NVME_SC_INTERNAL;
325 		goto out;
326 	}
327 
328 	/* XXX: figure out how to assign real vendors IDs. */
329 	id->vid = 0;
330 	id->ssvid = 0;
331 
332 	memset(id->sn, ' ', sizeof(id->sn));
333 	bin2hex(id->sn, &ctrl->subsys->serial,
334 		min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
335 	memcpy_and_pad(id->mn, sizeof(id->mn), model, sizeof(model) - 1, ' ');
336 	memcpy_and_pad(id->fr, sizeof(id->fr),
337 		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
338 
339 	id->rab = 6;
340 
341 	/*
342 	 * XXX: figure out how we can assign a IEEE OUI, but until then
343 	 * the safest is to leave it as zeroes.
344 	 */
345 
346 	/* we support multiple ports, multiples hosts and ANA: */
347 	id->cmic = (1 << 0) | (1 << 1) | (1 << 3);
348 
349 	/* no limit on data transfer sizes for now */
350 	id->mdts = 0;
351 	id->cntlid = cpu_to_le16(ctrl->cntlid);
352 	id->ver = cpu_to_le32(ctrl->subsys->ver);
353 
354 	/* XXX: figure out what to do about RTD3R/RTD3 */
355 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
356 	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
357 		NVME_CTRL_ATTR_TBKAS);
358 
359 	id->oacs = 0;
360 
361 	/*
362 	 * We don't really have a practical limit on the number of abort
363 	 * comands.  But we don't do anything useful for abort either, so
364 	 * no point in allowing more abort commands than the spec requires.
365 	 */
366 	id->acl = 3;
367 
368 	id->aerl = NVMET_ASYNC_EVENTS - 1;
369 
370 	/* first slot is read-only, only one slot supported */
371 	id->frmw = (1 << 0) | (1 << 1);
372 	id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
373 	id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
374 	id->npss = 0;
375 
376 	/* We support keep-alive timeout in granularity of seconds */
377 	id->kas = cpu_to_le16(NVMET_KAS);
378 
379 	id->sqes = (0x6 << 4) | 0x6;
380 	id->cqes = (0x4 << 4) | 0x4;
381 
382 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
383 	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
384 
385 	id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
386 	id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
387 	id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
388 			NVME_CTRL_ONCS_WRITE_ZEROES);
389 
390 	/* XXX: don't report vwc if the underlying device is write through */
391 	id->vwc = NVME_CTRL_VWC_PRESENT;
392 
393 	/*
394 	 * We can't support atomic writes bigger than a LBA without support
395 	 * from the backend device.
396 	 */
397 	id->awun = 0;
398 	id->awupf = 0;
399 
400 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
401 	if (ctrl->ops->has_keyed_sgls)
402 		id->sgls |= cpu_to_le32(1 << 2);
403 	if (req->port->inline_data_size)
404 		id->sgls |= cpu_to_le32(1 << 20);
405 
406 	strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
407 
408 	/* Max command capsule size is sqe + single page of in-capsule data */
409 	id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) +
410 				  req->port->inline_data_size) / 16);
411 	/* Max response capsule size is cqe */
412 	id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
413 
414 	id->msdbd = ctrl->ops->msdbd;
415 
416 	id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
417 	id->anatt = 10; /* random value */
418 	id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
419 	id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
420 
421 	/*
422 	 * Meh, we don't really support any power state.  Fake up the same
423 	 * values that qemu does.
424 	 */
425 	id->psd[0].max_power = cpu_to_le16(0x9c4);
426 	id->psd[0].entry_lat = cpu_to_le32(0x10);
427 	id->psd[0].exit_lat = cpu_to_le32(0x4);
428 
429 	id->nwpc = 1 << 0; /* write protect and no write protect */
430 
431 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
432 
433 	kfree(id);
434 out:
435 	nvmet_req_complete(req, status);
436 }
437 
438 static void nvmet_execute_identify_ns(struct nvmet_req *req)
439 {
440 	struct nvmet_ns *ns;
441 	struct nvme_id_ns *id;
442 	u16 status = 0;
443 
444 	if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
445 		req->error_loc = offsetof(struct nvme_identify, nsid);
446 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
447 		goto out;
448 	}
449 
450 	id = kzalloc(sizeof(*id), GFP_KERNEL);
451 	if (!id) {
452 		status = NVME_SC_INTERNAL;
453 		goto out;
454 	}
455 
456 	/* return an all zeroed buffer if we can't find an active namespace */
457 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
458 	if (!ns)
459 		goto done;
460 
461 	/*
462 	 * nuse = ncap = nsze isn't always true, but we have no way to find
463 	 * that out from the underlying device.
464 	 */
465 	id->ncap = id->nsze = cpu_to_le64(ns->size >> ns->blksize_shift);
466 	switch (req->port->ana_state[ns->anagrpid]) {
467 	case NVME_ANA_INACCESSIBLE:
468 	case NVME_ANA_PERSISTENT_LOSS:
469 		break;
470 	default:
471 		id->nuse = id->nsze;
472 		break;
473         }
474 
475 	if (ns->bdev)
476 		nvmet_bdev_set_limits(ns->bdev, id);
477 
478 	/*
479 	 * We just provide a single LBA format that matches what the
480 	 * underlying device reports.
481 	 */
482 	id->nlbaf = 0;
483 	id->flbas = 0;
484 
485 	/*
486 	 * Our namespace might always be shared.  Not just with other
487 	 * controllers, but also with any other user of the block device.
488 	 */
489 	id->nmic = (1 << 0);
490 	id->anagrpid = cpu_to_le32(ns->anagrpid);
491 
492 	memcpy(&id->nguid, &ns->nguid, sizeof(id->nguid));
493 
494 	id->lbaf[0].ds = ns->blksize_shift;
495 
496 	if (ns->readonly)
497 		id->nsattr |= (1 << 0);
498 	nvmet_put_namespace(ns);
499 done:
500 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
501 	kfree(id);
502 out:
503 	nvmet_req_complete(req, status);
504 }
505 
506 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
507 {
508 	static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
509 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
510 	struct nvmet_ns *ns;
511 	u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
512 	__le32 *list;
513 	u16 status = 0;
514 	int i = 0;
515 
516 	list = kzalloc(buf_size, GFP_KERNEL);
517 	if (!list) {
518 		status = NVME_SC_INTERNAL;
519 		goto out;
520 	}
521 
522 	rcu_read_lock();
523 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
524 		if (ns->nsid <= min_nsid)
525 			continue;
526 		list[i++] = cpu_to_le32(ns->nsid);
527 		if (i == buf_size / sizeof(__le32))
528 			break;
529 	}
530 	rcu_read_unlock();
531 
532 	status = nvmet_copy_to_sgl(req, 0, list, buf_size);
533 
534 	kfree(list);
535 out:
536 	nvmet_req_complete(req, status);
537 }
538 
539 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
540 				    void *id, off_t *off)
541 {
542 	struct nvme_ns_id_desc desc = {
543 		.nidt = type,
544 		.nidl = len,
545 	};
546 	u16 status;
547 
548 	status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
549 	if (status)
550 		return status;
551 	*off += sizeof(desc);
552 
553 	status = nvmet_copy_to_sgl(req, *off, id, len);
554 	if (status)
555 		return status;
556 	*off += len;
557 
558 	return 0;
559 }
560 
561 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
562 {
563 	struct nvmet_ns *ns;
564 	u16 status = 0;
565 	off_t off = 0;
566 
567 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
568 	if (!ns) {
569 		req->error_loc = offsetof(struct nvme_identify, nsid);
570 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
571 		goto out;
572 	}
573 
574 	if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) {
575 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
576 						  NVME_NIDT_UUID_LEN,
577 						  &ns->uuid, &off);
578 		if (status)
579 			goto out_put_ns;
580 	}
581 	if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) {
582 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
583 						  NVME_NIDT_NGUID_LEN,
584 						  &ns->nguid, &off);
585 		if (status)
586 			goto out_put_ns;
587 	}
588 
589 	if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
590 			off) != NVME_IDENTIFY_DATA_SIZE - off)
591 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
592 out_put_ns:
593 	nvmet_put_namespace(ns);
594 out:
595 	nvmet_req_complete(req, status);
596 }
597 
598 static void nvmet_execute_identify(struct nvmet_req *req)
599 {
600 	if (!nvmet_check_data_len(req, NVME_IDENTIFY_DATA_SIZE))
601 		return;
602 
603 	switch (req->cmd->identify.cns) {
604 	case NVME_ID_CNS_NS:
605 		return nvmet_execute_identify_ns(req);
606 	case NVME_ID_CNS_CTRL:
607 		return nvmet_execute_identify_ctrl(req);
608 	case NVME_ID_CNS_NS_ACTIVE_LIST:
609 		return nvmet_execute_identify_nslist(req);
610 	case NVME_ID_CNS_NS_DESC_LIST:
611 		return nvmet_execute_identify_desclist(req);
612 	}
613 
614 	pr_err("unhandled identify cns %d on qid %d\n",
615 	       req->cmd->identify.cns, req->sq->qid);
616 	req->error_loc = offsetof(struct nvme_identify, cns);
617 	nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
618 }
619 
620 /*
621  * A "minimum viable" abort implementation: the command is mandatory in the
622  * spec, but we are not required to do any useful work.  We couldn't really
623  * do a useful abort, so don't bother even with waiting for the command
624  * to be exectuted and return immediately telling the command to abort
625  * wasn't found.
626  */
627 static void nvmet_execute_abort(struct nvmet_req *req)
628 {
629 	if (!nvmet_check_data_len(req, 0))
630 		return;
631 	nvmet_set_result(req, 1);
632 	nvmet_req_complete(req, 0);
633 }
634 
635 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
636 {
637 	u16 status;
638 
639 	if (req->ns->file)
640 		status = nvmet_file_flush(req);
641 	else
642 		status = nvmet_bdev_flush(req);
643 
644 	if (status)
645 		pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
646 	return status;
647 }
648 
649 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req)
650 {
651 	u32 write_protect = le32_to_cpu(req->cmd->common.cdw11);
652 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
653 	u16 status = NVME_SC_FEATURE_NOT_CHANGEABLE;
654 
655 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->rw.nsid);
656 	if (unlikely(!req->ns)) {
657 		req->error_loc = offsetof(struct nvme_common_command, nsid);
658 		return status;
659 	}
660 
661 	mutex_lock(&subsys->lock);
662 	switch (write_protect) {
663 	case NVME_NS_WRITE_PROTECT:
664 		req->ns->readonly = true;
665 		status = nvmet_write_protect_flush_sync(req);
666 		if (status)
667 			req->ns->readonly = false;
668 		break;
669 	case NVME_NS_NO_WRITE_PROTECT:
670 		req->ns->readonly = false;
671 		status = 0;
672 		break;
673 	default:
674 		break;
675 	}
676 
677 	if (!status)
678 		nvmet_ns_changed(subsys, req->ns->nsid);
679 	mutex_unlock(&subsys->lock);
680 	return status;
681 }
682 
683 u16 nvmet_set_feat_kato(struct nvmet_req *req)
684 {
685 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
686 
687 	req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
688 
689 	nvmet_set_result(req, req->sq->ctrl->kato);
690 
691 	return 0;
692 }
693 
694 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask)
695 {
696 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
697 
698 	if (val32 & ~mask) {
699 		req->error_loc = offsetof(struct nvme_common_command, cdw11);
700 		return NVME_SC_INVALID_FIELD | NVME_SC_DNR;
701 	}
702 
703 	WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
704 	nvmet_set_result(req, val32);
705 
706 	return 0;
707 }
708 
709 static void nvmet_execute_set_features(struct nvmet_req *req)
710 {
711 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
712 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
713 	u16 status = 0;
714 
715 	if (!nvmet_check_data_len(req, 0))
716 		return;
717 
718 	switch (cdw10 & 0xff) {
719 	case NVME_FEAT_NUM_QUEUES:
720 		nvmet_set_result(req,
721 			(subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
722 		break;
723 	case NVME_FEAT_KATO:
724 		status = nvmet_set_feat_kato(req);
725 		break;
726 	case NVME_FEAT_ASYNC_EVENT:
727 		status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL);
728 		break;
729 	case NVME_FEAT_HOST_ID:
730 		status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
731 		break;
732 	case NVME_FEAT_WRITE_PROTECT:
733 		status = nvmet_set_feat_write_protect(req);
734 		break;
735 	default:
736 		req->error_loc = offsetof(struct nvme_common_command, cdw10);
737 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
738 		break;
739 	}
740 
741 	nvmet_req_complete(req, status);
742 }
743 
744 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req)
745 {
746 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
747 	u32 result;
748 
749 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->common.nsid);
750 	if (!req->ns)  {
751 		req->error_loc = offsetof(struct nvme_common_command, nsid);
752 		return NVME_SC_INVALID_NS | NVME_SC_DNR;
753 	}
754 	mutex_lock(&subsys->lock);
755 	if (req->ns->readonly == true)
756 		result = NVME_NS_WRITE_PROTECT;
757 	else
758 		result = NVME_NS_NO_WRITE_PROTECT;
759 	nvmet_set_result(req, result);
760 	mutex_unlock(&subsys->lock);
761 
762 	return 0;
763 }
764 
765 void nvmet_get_feat_kato(struct nvmet_req *req)
766 {
767 	nvmet_set_result(req, req->sq->ctrl->kato * 1000);
768 }
769 
770 void nvmet_get_feat_async_event(struct nvmet_req *req)
771 {
772 	nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
773 }
774 
775 static void nvmet_execute_get_features(struct nvmet_req *req)
776 {
777 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
778 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
779 	u16 status = 0;
780 
781 	if (!nvmet_check_data_len(req, 0))
782 		return;
783 
784 	switch (cdw10 & 0xff) {
785 	/*
786 	 * These features are mandatory in the spec, but we don't
787 	 * have a useful way to implement them.  We'll eventually
788 	 * need to come up with some fake values for these.
789 	 */
790 #if 0
791 	case NVME_FEAT_ARBITRATION:
792 		break;
793 	case NVME_FEAT_POWER_MGMT:
794 		break;
795 	case NVME_FEAT_TEMP_THRESH:
796 		break;
797 	case NVME_FEAT_ERR_RECOVERY:
798 		break;
799 	case NVME_FEAT_IRQ_COALESCE:
800 		break;
801 	case NVME_FEAT_IRQ_CONFIG:
802 		break;
803 	case NVME_FEAT_WRITE_ATOMIC:
804 		break;
805 #endif
806 	case NVME_FEAT_ASYNC_EVENT:
807 		nvmet_get_feat_async_event(req);
808 		break;
809 	case NVME_FEAT_VOLATILE_WC:
810 		nvmet_set_result(req, 1);
811 		break;
812 	case NVME_FEAT_NUM_QUEUES:
813 		nvmet_set_result(req,
814 			(subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
815 		break;
816 	case NVME_FEAT_KATO:
817 		nvmet_get_feat_kato(req);
818 		break;
819 	case NVME_FEAT_HOST_ID:
820 		/* need 128-bit host identifier flag */
821 		if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) {
822 			req->error_loc =
823 				offsetof(struct nvme_common_command, cdw11);
824 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
825 			break;
826 		}
827 
828 		status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
829 				sizeof(req->sq->ctrl->hostid));
830 		break;
831 	case NVME_FEAT_WRITE_PROTECT:
832 		status = nvmet_get_feat_write_protect(req);
833 		break;
834 	default:
835 		req->error_loc =
836 			offsetof(struct nvme_common_command, cdw10);
837 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
838 		break;
839 	}
840 
841 	nvmet_req_complete(req, status);
842 }
843 
844 void nvmet_execute_async_event(struct nvmet_req *req)
845 {
846 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
847 
848 	if (!nvmet_check_data_len(req, 0))
849 		return;
850 
851 	mutex_lock(&ctrl->lock);
852 	if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
853 		mutex_unlock(&ctrl->lock);
854 		nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
855 		return;
856 	}
857 	ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
858 	mutex_unlock(&ctrl->lock);
859 
860 	schedule_work(&ctrl->async_event_work);
861 }
862 
863 void nvmet_execute_keep_alive(struct nvmet_req *req)
864 {
865 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
866 
867 	if (!nvmet_check_data_len(req, 0))
868 		return;
869 
870 	pr_debug("ctrl %d update keep-alive timer for %d secs\n",
871 		ctrl->cntlid, ctrl->kato);
872 
873 	mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
874 	nvmet_req_complete(req, 0);
875 }
876 
877 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
878 {
879 	struct nvme_command *cmd = req->cmd;
880 	u16 ret;
881 
882 	if (nvme_is_fabrics(cmd))
883 		return nvmet_parse_fabrics_cmd(req);
884 	if (req->sq->ctrl->subsys->type == NVME_NQN_DISC)
885 		return nvmet_parse_discovery_cmd(req);
886 
887 	ret = nvmet_check_ctrl_status(req, cmd);
888 	if (unlikely(ret))
889 		return ret;
890 
891 	switch (cmd->common.opcode) {
892 	case nvme_admin_get_log_page:
893 		req->execute = nvmet_execute_get_log_page;
894 		return 0;
895 	case nvme_admin_identify:
896 		req->execute = nvmet_execute_identify;
897 		return 0;
898 	case nvme_admin_abort_cmd:
899 		req->execute = nvmet_execute_abort;
900 		return 0;
901 	case nvme_admin_set_features:
902 		req->execute = nvmet_execute_set_features;
903 		return 0;
904 	case nvme_admin_get_features:
905 		req->execute = nvmet_execute_get_features;
906 		return 0;
907 	case nvme_admin_async_event:
908 		req->execute = nvmet_execute_async_event;
909 		return 0;
910 	case nvme_admin_keep_alive:
911 		req->execute = nvmet_execute_keep_alive;
912 		return 0;
913 	}
914 
915 	pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
916 	       req->sq->qid);
917 	req->error_loc = offsetof(struct nvme_common_command, opcode);
918 	return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
919 }
920