xref: /openbmc/linux/drivers/nvme/target/admin-cmd.c (revision 8795a739)
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->data_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->data_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->data_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->data_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_identify_ctrl(struct nvmet_req *req)
286 {
287 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
288 	struct nvme_id_ctrl *id;
289 	u16 status = 0;
290 	const char model[] = "Linux";
291 
292 	id = kzalloc(sizeof(*id), GFP_KERNEL);
293 	if (!id) {
294 		status = NVME_SC_INTERNAL;
295 		goto out;
296 	}
297 
298 	/* XXX: figure out how to assign real vendors IDs. */
299 	id->vid = 0;
300 	id->ssvid = 0;
301 
302 	memset(id->sn, ' ', sizeof(id->sn));
303 	bin2hex(id->sn, &ctrl->subsys->serial,
304 		min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
305 	memcpy_and_pad(id->mn, sizeof(id->mn), model, sizeof(model) - 1, ' ');
306 	memcpy_and_pad(id->fr, sizeof(id->fr),
307 		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
308 
309 	id->rab = 6;
310 
311 	/*
312 	 * XXX: figure out how we can assign a IEEE OUI, but until then
313 	 * the safest is to leave it as zeroes.
314 	 */
315 
316 	/* we support multiple ports, multiples hosts and ANA: */
317 	id->cmic = (1 << 0) | (1 << 1) | (1 << 3);
318 
319 	/* no limit on data transfer sizes for now */
320 	id->mdts = 0;
321 	id->cntlid = cpu_to_le16(ctrl->cntlid);
322 	id->ver = cpu_to_le32(ctrl->subsys->ver);
323 
324 	/* XXX: figure out what to do about RTD3R/RTD3 */
325 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
326 	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
327 		NVME_CTRL_ATTR_TBKAS);
328 
329 	id->oacs = 0;
330 
331 	/*
332 	 * We don't really have a practical limit on the number of abort
333 	 * comands.  But we don't do anything useful for abort either, so
334 	 * no point in allowing more abort commands than the spec requires.
335 	 */
336 	id->acl = 3;
337 
338 	id->aerl = NVMET_ASYNC_EVENTS - 1;
339 
340 	/* first slot is read-only, only one slot supported */
341 	id->frmw = (1 << 0) | (1 << 1);
342 	id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
343 	id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
344 	id->npss = 0;
345 
346 	/* We support keep-alive timeout in granularity of seconds */
347 	id->kas = cpu_to_le16(NVMET_KAS);
348 
349 	id->sqes = (0x6 << 4) | 0x6;
350 	id->cqes = (0x4 << 4) | 0x4;
351 
352 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
353 	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
354 
355 	id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
356 	id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
357 	id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
358 			NVME_CTRL_ONCS_WRITE_ZEROES);
359 
360 	/* XXX: don't report vwc if the underlying device is write through */
361 	id->vwc = NVME_CTRL_VWC_PRESENT;
362 
363 	/*
364 	 * We can't support atomic writes bigger than a LBA without support
365 	 * from the backend device.
366 	 */
367 	id->awun = 0;
368 	id->awupf = 0;
369 
370 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
371 	if (ctrl->ops->has_keyed_sgls)
372 		id->sgls |= cpu_to_le32(1 << 2);
373 	if (req->port->inline_data_size)
374 		id->sgls |= cpu_to_le32(1 << 20);
375 
376 	strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
377 
378 	/* Max command capsule size is sqe + single page of in-capsule data */
379 	id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) +
380 				  req->port->inline_data_size) / 16);
381 	/* Max response capsule size is cqe */
382 	id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
383 
384 	id->msdbd = ctrl->ops->msdbd;
385 
386 	id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
387 	id->anatt = 10; /* random value */
388 	id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
389 	id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
390 
391 	/*
392 	 * Meh, we don't really support any power state.  Fake up the same
393 	 * values that qemu does.
394 	 */
395 	id->psd[0].max_power = cpu_to_le16(0x9c4);
396 	id->psd[0].entry_lat = cpu_to_le32(0x10);
397 	id->psd[0].exit_lat = cpu_to_le32(0x4);
398 
399 	id->nwpc = 1 << 0; /* write protect and no write protect */
400 
401 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
402 
403 	kfree(id);
404 out:
405 	nvmet_req_complete(req, status);
406 }
407 
408 static void nvmet_execute_identify_ns(struct nvmet_req *req)
409 {
410 	struct nvmet_ns *ns;
411 	struct nvme_id_ns *id;
412 	u16 status = 0;
413 
414 	if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
415 		req->error_loc = offsetof(struct nvme_identify, nsid);
416 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
417 		goto out;
418 	}
419 
420 	id = kzalloc(sizeof(*id), GFP_KERNEL);
421 	if (!id) {
422 		status = NVME_SC_INTERNAL;
423 		goto out;
424 	}
425 
426 	/* return an all zeroed buffer if we can't find an active namespace */
427 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
428 	if (!ns)
429 		goto done;
430 
431 	/*
432 	 * nuse = ncap = nsze isn't always true, but we have no way to find
433 	 * that out from the underlying device.
434 	 */
435 	id->ncap = id->nsze = cpu_to_le64(ns->size >> ns->blksize_shift);
436 	switch (req->port->ana_state[ns->anagrpid]) {
437 	case NVME_ANA_INACCESSIBLE:
438 	case NVME_ANA_PERSISTENT_LOSS:
439 		break;
440 	default:
441 		id->nuse = id->nsze;
442 		break;
443         }
444 
445 	if (ns->bdev)
446 		nvmet_bdev_set_limits(ns->bdev, id);
447 
448 	/*
449 	 * We just provide a single LBA format that matches what the
450 	 * underlying device reports.
451 	 */
452 	id->nlbaf = 0;
453 	id->flbas = 0;
454 
455 	/*
456 	 * Our namespace might always be shared.  Not just with other
457 	 * controllers, but also with any other user of the block device.
458 	 */
459 	id->nmic = (1 << 0);
460 	id->anagrpid = cpu_to_le32(ns->anagrpid);
461 
462 	memcpy(&id->nguid, &ns->nguid, sizeof(id->nguid));
463 
464 	id->lbaf[0].ds = ns->blksize_shift;
465 
466 	if (ns->readonly)
467 		id->nsattr |= (1 << 0);
468 	nvmet_put_namespace(ns);
469 done:
470 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
471 	kfree(id);
472 out:
473 	nvmet_req_complete(req, status);
474 }
475 
476 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
477 {
478 	static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
479 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
480 	struct nvmet_ns *ns;
481 	u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
482 	__le32 *list;
483 	u16 status = 0;
484 	int i = 0;
485 
486 	list = kzalloc(buf_size, GFP_KERNEL);
487 	if (!list) {
488 		status = NVME_SC_INTERNAL;
489 		goto out;
490 	}
491 
492 	rcu_read_lock();
493 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
494 		if (ns->nsid <= min_nsid)
495 			continue;
496 		list[i++] = cpu_to_le32(ns->nsid);
497 		if (i == buf_size / sizeof(__le32))
498 			break;
499 	}
500 	rcu_read_unlock();
501 
502 	status = nvmet_copy_to_sgl(req, 0, list, buf_size);
503 
504 	kfree(list);
505 out:
506 	nvmet_req_complete(req, status);
507 }
508 
509 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
510 				    void *id, off_t *off)
511 {
512 	struct nvme_ns_id_desc desc = {
513 		.nidt = type,
514 		.nidl = len,
515 	};
516 	u16 status;
517 
518 	status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
519 	if (status)
520 		return status;
521 	*off += sizeof(desc);
522 
523 	status = nvmet_copy_to_sgl(req, *off, id, len);
524 	if (status)
525 		return status;
526 	*off += len;
527 
528 	return 0;
529 }
530 
531 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
532 {
533 	struct nvmet_ns *ns;
534 	u16 status = 0;
535 	off_t off = 0;
536 
537 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
538 	if (!ns) {
539 		req->error_loc = offsetof(struct nvme_identify, nsid);
540 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
541 		goto out;
542 	}
543 
544 	if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) {
545 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
546 						  NVME_NIDT_UUID_LEN,
547 						  &ns->uuid, &off);
548 		if (status)
549 			goto out_put_ns;
550 	}
551 	if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) {
552 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
553 						  NVME_NIDT_NGUID_LEN,
554 						  &ns->nguid, &off);
555 		if (status)
556 			goto out_put_ns;
557 	}
558 
559 	if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
560 			off) != NVME_IDENTIFY_DATA_SIZE - off)
561 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
562 out_put_ns:
563 	nvmet_put_namespace(ns);
564 out:
565 	nvmet_req_complete(req, status);
566 }
567 
568 /*
569  * A "minimum viable" abort implementation: the command is mandatory in the
570  * spec, but we are not required to do any useful work.  We couldn't really
571  * do a useful abort, so don't bother even with waiting for the command
572  * to be exectuted and return immediately telling the command to abort
573  * wasn't found.
574  */
575 static void nvmet_execute_abort(struct nvmet_req *req)
576 {
577 	nvmet_set_result(req, 1);
578 	nvmet_req_complete(req, 0);
579 }
580 
581 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
582 {
583 	u16 status;
584 
585 	if (req->ns->file)
586 		status = nvmet_file_flush(req);
587 	else
588 		status = nvmet_bdev_flush(req);
589 
590 	if (status)
591 		pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
592 	return status;
593 }
594 
595 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req)
596 {
597 	u32 write_protect = le32_to_cpu(req->cmd->common.cdw11);
598 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
599 	u16 status = NVME_SC_FEATURE_NOT_CHANGEABLE;
600 
601 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->rw.nsid);
602 	if (unlikely(!req->ns)) {
603 		req->error_loc = offsetof(struct nvme_common_command, nsid);
604 		return status;
605 	}
606 
607 	mutex_lock(&subsys->lock);
608 	switch (write_protect) {
609 	case NVME_NS_WRITE_PROTECT:
610 		req->ns->readonly = true;
611 		status = nvmet_write_protect_flush_sync(req);
612 		if (status)
613 			req->ns->readonly = false;
614 		break;
615 	case NVME_NS_NO_WRITE_PROTECT:
616 		req->ns->readonly = false;
617 		status = 0;
618 		break;
619 	default:
620 		break;
621 	}
622 
623 	if (!status)
624 		nvmet_ns_changed(subsys, req->ns->nsid);
625 	mutex_unlock(&subsys->lock);
626 	return status;
627 }
628 
629 u16 nvmet_set_feat_kato(struct nvmet_req *req)
630 {
631 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
632 
633 	req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
634 
635 	nvmet_set_result(req, req->sq->ctrl->kato);
636 
637 	return 0;
638 }
639 
640 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask)
641 {
642 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
643 
644 	if (val32 & ~mask) {
645 		req->error_loc = offsetof(struct nvme_common_command, cdw11);
646 		return NVME_SC_INVALID_FIELD | NVME_SC_DNR;
647 	}
648 
649 	WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
650 	nvmet_set_result(req, val32);
651 
652 	return 0;
653 }
654 
655 static void nvmet_execute_set_features(struct nvmet_req *req)
656 {
657 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
658 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
659 	u16 status = 0;
660 
661 	switch (cdw10 & 0xff) {
662 	case NVME_FEAT_NUM_QUEUES:
663 		nvmet_set_result(req,
664 			(subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
665 		break;
666 	case NVME_FEAT_KATO:
667 		status = nvmet_set_feat_kato(req);
668 		break;
669 	case NVME_FEAT_ASYNC_EVENT:
670 		status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL);
671 		break;
672 	case NVME_FEAT_HOST_ID:
673 		status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
674 		break;
675 	case NVME_FEAT_WRITE_PROTECT:
676 		status = nvmet_set_feat_write_protect(req);
677 		break;
678 	default:
679 		req->error_loc = offsetof(struct nvme_common_command, cdw10);
680 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
681 		break;
682 	}
683 
684 	nvmet_req_complete(req, status);
685 }
686 
687 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req)
688 {
689 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
690 	u32 result;
691 
692 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->common.nsid);
693 	if (!req->ns)  {
694 		req->error_loc = offsetof(struct nvme_common_command, nsid);
695 		return NVME_SC_INVALID_NS | NVME_SC_DNR;
696 	}
697 	mutex_lock(&subsys->lock);
698 	if (req->ns->readonly == true)
699 		result = NVME_NS_WRITE_PROTECT;
700 	else
701 		result = NVME_NS_NO_WRITE_PROTECT;
702 	nvmet_set_result(req, result);
703 	mutex_unlock(&subsys->lock);
704 
705 	return 0;
706 }
707 
708 void nvmet_get_feat_kato(struct nvmet_req *req)
709 {
710 	nvmet_set_result(req, req->sq->ctrl->kato * 1000);
711 }
712 
713 void nvmet_get_feat_async_event(struct nvmet_req *req)
714 {
715 	nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
716 }
717 
718 static void nvmet_execute_get_features(struct nvmet_req *req)
719 {
720 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
721 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
722 	u16 status = 0;
723 
724 	switch (cdw10 & 0xff) {
725 	/*
726 	 * These features are mandatory in the spec, but we don't
727 	 * have a useful way to implement them.  We'll eventually
728 	 * need to come up with some fake values for these.
729 	 */
730 #if 0
731 	case NVME_FEAT_ARBITRATION:
732 		break;
733 	case NVME_FEAT_POWER_MGMT:
734 		break;
735 	case NVME_FEAT_TEMP_THRESH:
736 		break;
737 	case NVME_FEAT_ERR_RECOVERY:
738 		break;
739 	case NVME_FEAT_IRQ_COALESCE:
740 		break;
741 	case NVME_FEAT_IRQ_CONFIG:
742 		break;
743 	case NVME_FEAT_WRITE_ATOMIC:
744 		break;
745 #endif
746 	case NVME_FEAT_ASYNC_EVENT:
747 		nvmet_get_feat_async_event(req);
748 		break;
749 	case NVME_FEAT_VOLATILE_WC:
750 		nvmet_set_result(req, 1);
751 		break;
752 	case NVME_FEAT_NUM_QUEUES:
753 		nvmet_set_result(req,
754 			(subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
755 		break;
756 	case NVME_FEAT_KATO:
757 		nvmet_get_feat_kato(req);
758 		break;
759 	case NVME_FEAT_HOST_ID:
760 		/* need 128-bit host identifier flag */
761 		if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) {
762 			req->error_loc =
763 				offsetof(struct nvme_common_command, cdw11);
764 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
765 			break;
766 		}
767 
768 		status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
769 				sizeof(req->sq->ctrl->hostid));
770 		break;
771 	case NVME_FEAT_WRITE_PROTECT:
772 		status = nvmet_get_feat_write_protect(req);
773 		break;
774 	default:
775 		req->error_loc =
776 			offsetof(struct nvme_common_command, cdw10);
777 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
778 		break;
779 	}
780 
781 	nvmet_req_complete(req, status);
782 }
783 
784 void nvmet_execute_async_event(struct nvmet_req *req)
785 {
786 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
787 
788 	mutex_lock(&ctrl->lock);
789 	if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
790 		mutex_unlock(&ctrl->lock);
791 		nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
792 		return;
793 	}
794 	ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
795 	mutex_unlock(&ctrl->lock);
796 
797 	schedule_work(&ctrl->async_event_work);
798 }
799 
800 void nvmet_execute_keep_alive(struct nvmet_req *req)
801 {
802 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
803 
804 	pr_debug("ctrl %d update keep-alive timer for %d secs\n",
805 		ctrl->cntlid, ctrl->kato);
806 
807 	mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
808 	nvmet_req_complete(req, 0);
809 }
810 
811 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
812 {
813 	struct nvme_command *cmd = req->cmd;
814 	u16 ret;
815 
816 	ret = nvmet_check_ctrl_status(req, cmd);
817 	if (unlikely(ret))
818 		return ret;
819 
820 	switch (cmd->common.opcode) {
821 	case nvme_admin_get_log_page:
822 		req->data_len = nvmet_get_log_page_len(cmd);
823 
824 		switch (cmd->get_log_page.lid) {
825 		case NVME_LOG_ERROR:
826 			req->execute = nvmet_execute_get_log_page_error;
827 			return 0;
828 		case NVME_LOG_SMART:
829 			req->execute = nvmet_execute_get_log_page_smart;
830 			return 0;
831 		case NVME_LOG_FW_SLOT:
832 			/*
833 			 * We only support a single firmware slot which always
834 			 * is active, so we can zero out the whole firmware slot
835 			 * log and still claim to fully implement this mandatory
836 			 * log page.
837 			 */
838 			req->execute = nvmet_execute_get_log_page_noop;
839 			return 0;
840 		case NVME_LOG_CHANGED_NS:
841 			req->execute = nvmet_execute_get_log_changed_ns;
842 			return 0;
843 		case NVME_LOG_CMD_EFFECTS:
844 			req->execute = nvmet_execute_get_log_cmd_effects_ns;
845 			return 0;
846 		case NVME_LOG_ANA:
847 			req->execute = nvmet_execute_get_log_page_ana;
848 			return 0;
849 		}
850 		break;
851 	case nvme_admin_identify:
852 		req->data_len = NVME_IDENTIFY_DATA_SIZE;
853 		switch (cmd->identify.cns) {
854 		case NVME_ID_CNS_NS:
855 			req->execute = nvmet_execute_identify_ns;
856 			return 0;
857 		case NVME_ID_CNS_CTRL:
858 			req->execute = nvmet_execute_identify_ctrl;
859 			return 0;
860 		case NVME_ID_CNS_NS_ACTIVE_LIST:
861 			req->execute = nvmet_execute_identify_nslist;
862 			return 0;
863 		case NVME_ID_CNS_NS_DESC_LIST:
864 			req->execute = nvmet_execute_identify_desclist;
865 			return 0;
866 		}
867 		break;
868 	case nvme_admin_abort_cmd:
869 		req->execute = nvmet_execute_abort;
870 		req->data_len = 0;
871 		return 0;
872 	case nvme_admin_set_features:
873 		req->execute = nvmet_execute_set_features;
874 		req->data_len = 0;
875 		return 0;
876 	case nvme_admin_get_features:
877 		req->execute = nvmet_execute_get_features;
878 		req->data_len = 0;
879 		return 0;
880 	case nvme_admin_async_event:
881 		req->execute = nvmet_execute_async_event;
882 		req->data_len = 0;
883 		return 0;
884 	case nvme_admin_keep_alive:
885 		req->execute = nvmet_execute_keep_alive;
886 		req->data_len = 0;
887 		return 0;
888 	}
889 
890 	pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
891 	       req->sq->qid);
892 	req->error_loc = offsetof(struct nvme_common_command, opcode);
893 	return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
894 }
895