xref: /openbmc/linux/drivers/nvme/target/admin-cmd.c (revision 58f9d806)
1 /*
2  * NVMe admin command implementation.
3  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  */
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/module.h>
16 #include <linux/rculist.h>
17 
18 #include <generated/utsrelease.h>
19 #include <asm/unaligned.h>
20 #include "nvmet.h"
21 
22 /*
23  * This helper allows us to clear the AEN based on the RAE bit,
24  * Please use this helper when processing the log pages which are
25  * associated with the AEN.
26  */
27 static inline void nvmet_clear_aen(struct nvmet_req *req, u32 aen_bit)
28 {
29 	int rae = le32_to_cpu(req->cmd->common.cdw10[0]) & 1 << 15;
30 
31 	if (!rae)
32 		clear_bit(aen_bit, &req->sq->ctrl->aen_masked);
33 }
34 
35 u32 nvmet_get_log_page_len(struct nvme_command *cmd)
36 {
37 	u32 len = le16_to_cpu(cmd->get_log_page.numdu);
38 
39 	len <<= 16;
40 	len += le16_to_cpu(cmd->get_log_page.numdl);
41 	/* NUMD is a 0's based value */
42 	len += 1;
43 	len *= sizeof(u32);
44 
45 	return len;
46 }
47 
48 static void nvmet_execute_get_log_page_noop(struct nvmet_req *req)
49 {
50 	nvmet_req_complete(req, nvmet_zero_sgl(req, 0, req->data_len));
51 }
52 
53 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req,
54 		struct nvme_smart_log *slog)
55 {
56 	struct nvmet_ns *ns;
57 	u64 host_reads, host_writes, data_units_read, data_units_written;
58 
59 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid);
60 	if (!ns) {
61 		pr_err("Could not find namespace id : %d\n",
62 				le32_to_cpu(req->cmd->get_log_page.nsid));
63 		return NVME_SC_INVALID_NS;
64 	}
65 
66 	/* we don't have the right data for file backed ns */
67 	if (!ns->bdev)
68 		goto out;
69 
70 	host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]);
71 	data_units_read = part_stat_read(ns->bdev->bd_part, sectors[READ]);
72 	host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]);
73 	data_units_written = part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
74 
75 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
76 	put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
77 	put_unaligned_le64(host_writes, &slog->host_writes[0]);
78 	put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
79 out:
80 	nvmet_put_namespace(ns);
81 
82 	return NVME_SC_SUCCESS;
83 }
84 
85 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
86 		struct nvme_smart_log *slog)
87 {
88 	u64 host_reads = 0, host_writes = 0;
89 	u64 data_units_read = 0, data_units_written = 0;
90 	struct nvmet_ns *ns;
91 	struct nvmet_ctrl *ctrl;
92 
93 	ctrl = req->sq->ctrl;
94 
95 	rcu_read_lock();
96 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
97 		/* we don't have the right data for file backed ns */
98 		if (!ns->bdev)
99 			continue;
100 		host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]);
101 		data_units_read +=
102 			part_stat_read(ns->bdev->bd_part, sectors[READ]);
103 		host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]);
104 		data_units_written +=
105 			part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
106 
107 	}
108 	rcu_read_unlock();
109 
110 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
111 	put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
112 	put_unaligned_le64(host_writes, &slog->host_writes[0]);
113 	put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
114 
115 	return NVME_SC_SUCCESS;
116 }
117 
118 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req)
119 {
120 	struct nvme_smart_log *log;
121 	u16 status = NVME_SC_INTERNAL;
122 
123 	if (req->data_len != sizeof(*log))
124 		goto out;
125 
126 	log = kzalloc(sizeof(*log), GFP_KERNEL);
127 	if (!log)
128 		goto out;
129 
130 	if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL))
131 		status = nvmet_get_smart_log_all(req, log);
132 	else
133 		status = nvmet_get_smart_log_nsid(req, log);
134 	if (status)
135 		goto out_free_log;
136 
137 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
138 out_free_log:
139 	kfree(log);
140 out:
141 	nvmet_req_complete(req, status);
142 }
143 
144 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
145 {
146 	u16 status = NVME_SC_INTERNAL;
147 	struct nvme_effects_log *log;
148 
149 	log = kzalloc(sizeof(*log), GFP_KERNEL);
150 	if (!log)
151 		goto out;
152 
153 	log->acs[nvme_admin_get_log_page]	= cpu_to_le32(1 << 0);
154 	log->acs[nvme_admin_identify]		= cpu_to_le32(1 << 0);
155 	log->acs[nvme_admin_abort_cmd]		= cpu_to_le32(1 << 0);
156 	log->acs[nvme_admin_set_features]	= cpu_to_le32(1 << 0);
157 	log->acs[nvme_admin_get_features]	= cpu_to_le32(1 << 0);
158 	log->acs[nvme_admin_async_event]	= cpu_to_le32(1 << 0);
159 	log->acs[nvme_admin_keep_alive]		= cpu_to_le32(1 << 0);
160 
161 	log->iocs[nvme_cmd_read]		= cpu_to_le32(1 << 0);
162 	log->iocs[nvme_cmd_write]		= cpu_to_le32(1 << 0);
163 	log->iocs[nvme_cmd_flush]		= cpu_to_le32(1 << 0);
164 	log->iocs[nvme_cmd_dsm]			= cpu_to_le32(1 << 0);
165 	log->iocs[nvme_cmd_write_zeroes]	= cpu_to_le32(1 << 0);
166 
167 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
168 
169 	kfree(log);
170 out:
171 	nvmet_req_complete(req, status);
172 }
173 
174 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
175 {
176 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
177 	u16 status = NVME_SC_INTERNAL;
178 	size_t len;
179 
180 	if (req->data_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
181 		goto out;
182 
183 	mutex_lock(&ctrl->lock);
184 	if (ctrl->nr_changed_ns == U32_MAX)
185 		len = sizeof(__le32);
186 	else
187 		len = ctrl->nr_changed_ns * sizeof(__le32);
188 	status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
189 	if (!status)
190 		status = nvmet_zero_sgl(req, len, req->data_len - len);
191 	ctrl->nr_changed_ns = 0;
192 	nvmet_clear_aen(req, NVME_AEN_CFG_NS_ATTR);
193 	mutex_unlock(&ctrl->lock);
194 out:
195 	nvmet_req_complete(req, status);
196 }
197 
198 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid,
199 		struct nvme_ana_group_desc *desc)
200 {
201 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
202 	struct nvmet_ns *ns;
203 	u32 count = 0;
204 
205 	if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) {
206 		rcu_read_lock();
207 		list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link)
208 			if (ns->anagrpid == grpid)
209 				desc->nsids[count++] = cpu_to_le32(ns->nsid);
210 		rcu_read_unlock();
211 	}
212 
213 	desc->grpid = cpu_to_le32(grpid);
214 	desc->nnsids = cpu_to_le32(count);
215 	desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
216 	desc->state = req->port->ana_state[grpid];
217 	memset(desc->rsvd17, 0, sizeof(desc->rsvd17));
218 	return sizeof(struct nvme_ana_group_desc) + count * sizeof(__le32);
219 }
220 
221 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)
222 {
223 	struct nvme_ana_rsp_hdr hdr = { 0, };
224 	struct nvme_ana_group_desc *desc;
225 	size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */
226 	size_t len;
227 	u32 grpid;
228 	u16 ngrps = 0;
229 	u16 status;
230 
231 	status = NVME_SC_INTERNAL;
232 	desc = kmalloc(sizeof(struct nvme_ana_group_desc) +
233 			NVMET_MAX_NAMESPACES * sizeof(__le32), GFP_KERNEL);
234 	if (!desc)
235 		goto out;
236 
237 	down_read(&nvmet_ana_sem);
238 	for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
239 		if (!nvmet_ana_group_enabled[grpid])
240 			continue;
241 		len = nvmet_format_ana_group(req, grpid, desc);
242 		status = nvmet_copy_to_sgl(req, offset, desc, len);
243 		if (status)
244 			break;
245 		offset += len;
246 		ngrps++;
247 	}
248 	for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
249 		if (nvmet_ana_group_enabled[grpid])
250 			ngrps++;
251 	}
252 
253 	hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
254 	hdr.ngrps = cpu_to_le16(ngrps);
255 	nvmet_clear_aen(req, NVME_AEN_CFG_ANA_CHANGE);
256 	up_read(&nvmet_ana_sem);
257 
258 	kfree(desc);
259 
260 	/* copy the header last once we know the number of groups */
261 	status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr));
262 out:
263 	nvmet_req_complete(req, status);
264 }
265 
266 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
267 {
268 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
269 	struct nvme_id_ctrl *id;
270 	u16 status = 0;
271 	const char model[] = "Linux";
272 
273 	id = kzalloc(sizeof(*id), GFP_KERNEL);
274 	if (!id) {
275 		status = NVME_SC_INTERNAL;
276 		goto out;
277 	}
278 
279 	/* XXX: figure out how to assign real vendors IDs. */
280 	id->vid = 0;
281 	id->ssvid = 0;
282 
283 	memset(id->sn, ' ', sizeof(id->sn));
284 	bin2hex(id->sn, &ctrl->subsys->serial,
285 		min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
286 	memcpy_and_pad(id->mn, sizeof(id->mn), model, sizeof(model) - 1, ' ');
287 	memcpy_and_pad(id->fr, sizeof(id->fr),
288 		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
289 
290 	id->rab = 6;
291 
292 	/*
293 	 * XXX: figure out how we can assign a IEEE OUI, but until then
294 	 * the safest is to leave it as zeroes.
295 	 */
296 
297 	/* we support multiple ports, multiples hosts and ANA: */
298 	id->cmic = (1 << 0) | (1 << 1) | (1 << 3);
299 
300 	/* no limit on data transfer sizes for now */
301 	id->mdts = 0;
302 	id->cntlid = cpu_to_le16(ctrl->cntlid);
303 	id->ver = cpu_to_le32(ctrl->subsys->ver);
304 
305 	/* XXX: figure out what to do about RTD3R/RTD3 */
306 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
307 	id->ctratt = cpu_to_le32(1 << 0);
308 
309 	id->oacs = 0;
310 
311 	/*
312 	 * We don't really have a practical limit on the number of abort
313 	 * comands.  But we don't do anything useful for abort either, so
314 	 * no point in allowing more abort commands than the spec requires.
315 	 */
316 	id->acl = 3;
317 
318 	id->aerl = NVMET_ASYNC_EVENTS - 1;
319 
320 	/* first slot is read-only, only one slot supported */
321 	id->frmw = (1 << 0) | (1 << 1);
322 	id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
323 	id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
324 	id->npss = 0;
325 
326 	/* We support keep-alive timeout in granularity of seconds */
327 	id->kas = cpu_to_le16(NVMET_KAS);
328 
329 	id->sqes = (0x6 << 4) | 0x6;
330 	id->cqes = (0x4 << 4) | 0x4;
331 
332 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
333 	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
334 
335 	id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
336 	id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
337 	id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
338 			NVME_CTRL_ONCS_WRITE_ZEROES);
339 
340 	/* XXX: don't report vwc if the underlying device is write through */
341 	id->vwc = NVME_CTRL_VWC_PRESENT;
342 
343 	/*
344 	 * We can't support atomic writes bigger than a LBA without support
345 	 * from the backend device.
346 	 */
347 	id->awun = 0;
348 	id->awupf = 0;
349 
350 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
351 	if (ctrl->ops->has_keyed_sgls)
352 		id->sgls |= cpu_to_le32(1 << 2);
353 	if (req->port->inline_data_size)
354 		id->sgls |= cpu_to_le32(1 << 20);
355 
356 	strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
357 
358 	/* Max command capsule size is sqe + single page of in-capsule data */
359 	id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) +
360 				  req->port->inline_data_size) / 16);
361 	/* Max response capsule size is cqe */
362 	id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
363 
364 	id->msdbd = ctrl->ops->msdbd;
365 
366 	id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
367 	id->anatt = 10; /* random value */
368 	id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
369 	id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
370 
371 	/*
372 	 * Meh, we don't really support any power state.  Fake up the same
373 	 * values that qemu does.
374 	 */
375 	id->psd[0].max_power = cpu_to_le16(0x9c4);
376 	id->psd[0].entry_lat = cpu_to_le32(0x10);
377 	id->psd[0].exit_lat = cpu_to_le32(0x4);
378 
379 	id->nwpc = 1 << 0; /* write protect and no write protect */
380 
381 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
382 
383 	kfree(id);
384 out:
385 	nvmet_req_complete(req, status);
386 }
387 
388 static void nvmet_execute_identify_ns(struct nvmet_req *req)
389 {
390 	struct nvmet_ns *ns;
391 	struct nvme_id_ns *id;
392 	u16 status = 0;
393 
394 	if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
395 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
396 		goto out;
397 	}
398 
399 	id = kzalloc(sizeof(*id), GFP_KERNEL);
400 	if (!id) {
401 		status = NVME_SC_INTERNAL;
402 		goto out;
403 	}
404 
405 	/* return an all zeroed buffer if we can't find an active namespace */
406 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
407 	if (!ns)
408 		goto done;
409 
410 	/*
411 	 * nuse = ncap = nsze isn't always true, but we have no way to find
412 	 * that out from the underlying device.
413 	 */
414 	id->ncap = id->nsze = cpu_to_le64(ns->size >> ns->blksize_shift);
415 	switch (req->port->ana_state[ns->anagrpid]) {
416 	case NVME_ANA_INACCESSIBLE:
417 	case NVME_ANA_PERSISTENT_LOSS:
418 		break;
419 	default:
420 		id->nuse = id->nsze;
421 		break;
422         }
423 
424 	/*
425 	 * We just provide a single LBA format that matches what the
426 	 * underlying device reports.
427 	 */
428 	id->nlbaf = 0;
429 	id->flbas = 0;
430 
431 	/*
432 	 * Our namespace might always be shared.  Not just with other
433 	 * controllers, but also with any other user of the block device.
434 	 */
435 	id->nmic = (1 << 0);
436 	id->anagrpid = cpu_to_le32(ns->anagrpid);
437 
438 	memcpy(&id->nguid, &ns->nguid, sizeof(id->nguid));
439 
440 	id->lbaf[0].ds = ns->blksize_shift;
441 
442 	if (ns->readonly)
443 		id->nsattr |= (1 << 0);
444 	nvmet_put_namespace(ns);
445 done:
446 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
447 	kfree(id);
448 out:
449 	nvmet_req_complete(req, status);
450 }
451 
452 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
453 {
454 	static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
455 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
456 	struct nvmet_ns *ns;
457 	u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
458 	__le32 *list;
459 	u16 status = 0;
460 	int i = 0;
461 
462 	list = kzalloc(buf_size, GFP_KERNEL);
463 	if (!list) {
464 		status = NVME_SC_INTERNAL;
465 		goto out;
466 	}
467 
468 	rcu_read_lock();
469 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
470 		if (ns->nsid <= min_nsid)
471 			continue;
472 		list[i++] = cpu_to_le32(ns->nsid);
473 		if (i == buf_size / sizeof(__le32))
474 			break;
475 	}
476 	rcu_read_unlock();
477 
478 	status = nvmet_copy_to_sgl(req, 0, list, buf_size);
479 
480 	kfree(list);
481 out:
482 	nvmet_req_complete(req, status);
483 }
484 
485 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
486 				    void *id, off_t *off)
487 {
488 	struct nvme_ns_id_desc desc = {
489 		.nidt = type,
490 		.nidl = len,
491 	};
492 	u16 status;
493 
494 	status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
495 	if (status)
496 		return status;
497 	*off += sizeof(desc);
498 
499 	status = nvmet_copy_to_sgl(req, *off, id, len);
500 	if (status)
501 		return status;
502 	*off += len;
503 
504 	return 0;
505 }
506 
507 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
508 {
509 	struct nvmet_ns *ns;
510 	u16 status = 0;
511 	off_t off = 0;
512 
513 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
514 	if (!ns) {
515 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
516 		goto out;
517 	}
518 
519 	if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) {
520 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
521 						  NVME_NIDT_UUID_LEN,
522 						  &ns->uuid, &off);
523 		if (status)
524 			goto out_put_ns;
525 	}
526 	if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) {
527 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
528 						  NVME_NIDT_NGUID_LEN,
529 						  &ns->nguid, &off);
530 		if (status)
531 			goto out_put_ns;
532 	}
533 
534 	if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
535 			off) != NVME_IDENTIFY_DATA_SIZE - off)
536 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
537 out_put_ns:
538 	nvmet_put_namespace(ns);
539 out:
540 	nvmet_req_complete(req, status);
541 }
542 
543 /*
544  * A "minimum viable" abort implementation: the command is mandatory in the
545  * spec, but we are not required to do any useful work.  We couldn't really
546  * do a useful abort, so don't bother even with waiting for the command
547  * to be exectuted and return immediately telling the command to abort
548  * wasn't found.
549  */
550 static void nvmet_execute_abort(struct nvmet_req *req)
551 {
552 	nvmet_set_result(req, 1);
553 	nvmet_req_complete(req, 0);
554 }
555 
556 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
557 {
558 	u16 status;
559 
560 	if (req->ns->file)
561 		status = nvmet_file_flush(req);
562 	else
563 		status = nvmet_bdev_flush(req);
564 
565 	if (status)
566 		pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
567 	return status;
568 }
569 
570 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req)
571 {
572 	u32 write_protect = le32_to_cpu(req->cmd->common.cdw10[1]);
573 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
574 	u16 status = NVME_SC_FEATURE_NOT_CHANGEABLE;
575 
576 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->rw.nsid);
577 	if (unlikely(!req->ns))
578 		return status;
579 
580 	mutex_lock(&subsys->lock);
581 	switch (write_protect) {
582 	case NVME_NS_WRITE_PROTECT:
583 		req->ns->readonly = true;
584 		status = nvmet_write_protect_flush_sync(req);
585 		if (status)
586 			req->ns->readonly = false;
587 		break;
588 	case NVME_NS_NO_WRITE_PROTECT:
589 		req->ns->readonly = false;
590 		status = 0;
591 		break;
592 	default:
593 		break;
594 	}
595 
596 	if (!status)
597 		nvmet_ns_changed(subsys, req->ns->nsid);
598 	mutex_unlock(&subsys->lock);
599 	return status;
600 }
601 
602 static void nvmet_execute_set_features(struct nvmet_req *req)
603 {
604 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
605 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
606 	u32 val32;
607 	u16 status = 0;
608 
609 	switch (cdw10 & 0xff) {
610 	case NVME_FEAT_NUM_QUEUES:
611 		nvmet_set_result(req,
612 			(subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
613 		break;
614 	case NVME_FEAT_KATO:
615 		val32 = le32_to_cpu(req->cmd->common.cdw10[1]);
616 		req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
617 		nvmet_set_result(req, req->sq->ctrl->kato);
618 		break;
619 	case NVME_FEAT_ASYNC_EVENT:
620 		val32 = le32_to_cpu(req->cmd->common.cdw10[1]);
621 		if (val32 & ~NVMET_AEN_CFG_ALL) {
622 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
623 			break;
624 		}
625 
626 		WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
627 		nvmet_set_result(req, val32);
628 		break;
629 	case NVME_FEAT_HOST_ID:
630 		status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
631 		break;
632 	case NVME_FEAT_WRITE_PROTECT:
633 		status = nvmet_set_feat_write_protect(req);
634 		break;
635 	default:
636 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
637 		break;
638 	}
639 
640 	nvmet_req_complete(req, status);
641 }
642 
643 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req)
644 {
645 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
646 	u32 result;
647 
648 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->common.nsid);
649 	if (!req->ns)
650 		return NVME_SC_INVALID_NS | NVME_SC_DNR;
651 
652 	mutex_lock(&subsys->lock);
653 	if (req->ns->readonly == true)
654 		result = NVME_NS_WRITE_PROTECT;
655 	else
656 		result = NVME_NS_NO_WRITE_PROTECT;
657 	nvmet_set_result(req, result);
658 	mutex_unlock(&subsys->lock);
659 
660 	return 0;
661 }
662 
663 static void nvmet_execute_get_features(struct nvmet_req *req)
664 {
665 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
666 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
667 	u16 status = 0;
668 
669 	switch (cdw10 & 0xff) {
670 	/*
671 	 * These features are mandatory in the spec, but we don't
672 	 * have a useful way to implement them.  We'll eventually
673 	 * need to come up with some fake values for these.
674 	 */
675 #if 0
676 	case NVME_FEAT_ARBITRATION:
677 		break;
678 	case NVME_FEAT_POWER_MGMT:
679 		break;
680 	case NVME_FEAT_TEMP_THRESH:
681 		break;
682 	case NVME_FEAT_ERR_RECOVERY:
683 		break;
684 	case NVME_FEAT_IRQ_COALESCE:
685 		break;
686 	case NVME_FEAT_IRQ_CONFIG:
687 		break;
688 	case NVME_FEAT_WRITE_ATOMIC:
689 		break;
690 #endif
691 	case NVME_FEAT_ASYNC_EVENT:
692 		nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
693 		break;
694 	case NVME_FEAT_VOLATILE_WC:
695 		nvmet_set_result(req, 1);
696 		break;
697 	case NVME_FEAT_NUM_QUEUES:
698 		nvmet_set_result(req,
699 			(subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
700 		break;
701 	case NVME_FEAT_KATO:
702 		nvmet_set_result(req, req->sq->ctrl->kato * 1000);
703 		break;
704 	case NVME_FEAT_HOST_ID:
705 		/* need 128-bit host identifier flag */
706 		if (!(req->cmd->common.cdw10[1] & cpu_to_le32(1 << 0))) {
707 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
708 			break;
709 		}
710 
711 		status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
712 				sizeof(req->sq->ctrl->hostid));
713 		break;
714 	case NVME_FEAT_WRITE_PROTECT:
715 		status = nvmet_get_feat_write_protect(req);
716 		break;
717 	default:
718 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
719 		break;
720 	}
721 
722 	nvmet_req_complete(req, status);
723 }
724 
725 static void nvmet_execute_async_event(struct nvmet_req *req)
726 {
727 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
728 
729 	mutex_lock(&ctrl->lock);
730 	if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
731 		mutex_unlock(&ctrl->lock);
732 		nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
733 		return;
734 	}
735 	ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
736 	mutex_unlock(&ctrl->lock);
737 
738 	schedule_work(&ctrl->async_event_work);
739 }
740 
741 static void nvmet_execute_keep_alive(struct nvmet_req *req)
742 {
743 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
744 
745 	pr_debug("ctrl %d update keep-alive timer for %d secs\n",
746 		ctrl->cntlid, ctrl->kato);
747 
748 	mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
749 	nvmet_req_complete(req, 0);
750 }
751 
752 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
753 {
754 	struct nvme_command *cmd = req->cmd;
755 	u16 ret;
756 
757 	ret = nvmet_check_ctrl_status(req, cmd);
758 	if (unlikely(ret))
759 		return ret;
760 
761 	switch (cmd->common.opcode) {
762 	case nvme_admin_get_log_page:
763 		req->data_len = nvmet_get_log_page_len(cmd);
764 
765 		switch (cmd->get_log_page.lid) {
766 		case NVME_LOG_ERROR:
767 			/*
768 			 * We currently never set the More bit in the status
769 			 * field, so all error log entries are invalid and can
770 			 * be zeroed out.  This is called a minum viable
771 			 * implementation (TM) of this mandatory log page.
772 			 */
773 			req->execute = nvmet_execute_get_log_page_noop;
774 			return 0;
775 		case NVME_LOG_SMART:
776 			req->execute = nvmet_execute_get_log_page_smart;
777 			return 0;
778 		case NVME_LOG_FW_SLOT:
779 			/*
780 			 * We only support a single firmware slot which always
781 			 * is active, so we can zero out the whole firmware slot
782 			 * log and still claim to fully implement this mandatory
783 			 * log page.
784 			 */
785 			req->execute = nvmet_execute_get_log_page_noop;
786 			return 0;
787 		case NVME_LOG_CHANGED_NS:
788 			req->execute = nvmet_execute_get_log_changed_ns;
789 			return 0;
790 		case NVME_LOG_CMD_EFFECTS:
791 			req->execute = nvmet_execute_get_log_cmd_effects_ns;
792 			return 0;
793 		case NVME_LOG_ANA:
794 			req->execute = nvmet_execute_get_log_page_ana;
795 			return 0;
796 		}
797 		break;
798 	case nvme_admin_identify:
799 		req->data_len = NVME_IDENTIFY_DATA_SIZE;
800 		switch (cmd->identify.cns) {
801 		case NVME_ID_CNS_NS:
802 			req->execute = nvmet_execute_identify_ns;
803 			return 0;
804 		case NVME_ID_CNS_CTRL:
805 			req->execute = nvmet_execute_identify_ctrl;
806 			return 0;
807 		case NVME_ID_CNS_NS_ACTIVE_LIST:
808 			req->execute = nvmet_execute_identify_nslist;
809 			return 0;
810 		case NVME_ID_CNS_NS_DESC_LIST:
811 			req->execute = nvmet_execute_identify_desclist;
812 			return 0;
813 		}
814 		break;
815 	case nvme_admin_abort_cmd:
816 		req->execute = nvmet_execute_abort;
817 		req->data_len = 0;
818 		return 0;
819 	case nvme_admin_set_features:
820 		req->execute = nvmet_execute_set_features;
821 		req->data_len = 0;
822 		return 0;
823 	case nvme_admin_get_features:
824 		req->execute = nvmet_execute_get_features;
825 		req->data_len = 0;
826 		return 0;
827 	case nvme_admin_async_event:
828 		req->execute = nvmet_execute_async_event;
829 		req->data_len = 0;
830 		return 0;
831 	case nvme_admin_keep_alive:
832 		req->execute = nvmet_execute_keep_alive;
833 		req->data_len = 0;
834 		return 0;
835 	}
836 
837 	pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
838 	       req->sq->qid);
839 	return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
840 }
841