xref: /openbmc/linux/drivers/nvme/target/admin-cmd.c (revision 82df5b73)
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 	struct nvmet_ns *ns;
78 	u64 host_reads, host_writes, data_units_read, data_units_written;
79 
80 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid);
81 	if (!ns) {
82 		pr_err("Could not find namespace id : %d\n",
83 				le32_to_cpu(req->cmd->get_log_page.nsid));
84 		req->error_loc = offsetof(struct nvme_rw_command, nsid);
85 		return NVME_SC_INVALID_NS;
86 	}
87 
88 	/* we don't have the right data for file backed ns */
89 	if (!ns->bdev)
90 		goto out;
91 
92 	host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]);
93 	data_units_read = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part,
94 		sectors[READ]), 1000);
95 	host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]);
96 	data_units_written = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part,
97 		sectors[WRITE]), 1000);
98 
99 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
100 	put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
101 	put_unaligned_le64(host_writes, &slog->host_writes[0]);
102 	put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
103 out:
104 	nvmet_put_namespace(ns);
105 
106 	return NVME_SC_SUCCESS;
107 }
108 
109 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
110 		struct nvme_smart_log *slog)
111 {
112 	u64 host_reads = 0, host_writes = 0;
113 	u64 data_units_read = 0, data_units_written = 0;
114 	struct nvmet_ns *ns;
115 	struct nvmet_ctrl *ctrl;
116 
117 	ctrl = req->sq->ctrl;
118 
119 	rcu_read_lock();
120 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
121 		/* we don't have the right data for file backed ns */
122 		if (!ns->bdev)
123 			continue;
124 		host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]);
125 		data_units_read += DIV_ROUND_UP(
126 			part_stat_read(ns->bdev->bd_part, sectors[READ]), 1000);
127 		host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]);
128 		data_units_written += DIV_ROUND_UP(
129 			part_stat_read(ns->bdev->bd_part, sectors[WRITE]), 1000);
130 
131 	}
132 	rcu_read_unlock();
133 
134 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
135 	put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
136 	put_unaligned_le64(host_writes, &slog->host_writes[0]);
137 	put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
138 
139 	return NVME_SC_SUCCESS;
140 }
141 
142 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req)
143 {
144 	struct nvme_smart_log *log;
145 	u16 status = NVME_SC_INTERNAL;
146 	unsigned long flags;
147 
148 	if (req->transfer_len != sizeof(*log))
149 		goto out;
150 
151 	log = kzalloc(sizeof(*log), GFP_KERNEL);
152 	if (!log)
153 		goto out;
154 
155 	if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL))
156 		status = nvmet_get_smart_log_all(req, log);
157 	else
158 		status = nvmet_get_smart_log_nsid(req, log);
159 	if (status)
160 		goto out_free_log;
161 
162 	spin_lock_irqsave(&req->sq->ctrl->error_lock, flags);
163 	put_unaligned_le64(req->sq->ctrl->err_counter,
164 			&log->num_err_log_entries);
165 	spin_unlock_irqrestore(&req->sq->ctrl->error_lock, flags);
166 
167 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
168 out_free_log:
169 	kfree(log);
170 out:
171 	nvmet_req_complete(req, status);
172 }
173 
174 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
175 {
176 	u16 status = NVME_SC_INTERNAL;
177 	struct nvme_effects_log *log;
178 
179 	log = kzalloc(sizeof(*log), GFP_KERNEL);
180 	if (!log)
181 		goto out;
182 
183 	log->acs[nvme_admin_get_log_page]	= cpu_to_le32(1 << 0);
184 	log->acs[nvme_admin_identify]		= cpu_to_le32(1 << 0);
185 	log->acs[nvme_admin_abort_cmd]		= cpu_to_le32(1 << 0);
186 	log->acs[nvme_admin_set_features]	= cpu_to_le32(1 << 0);
187 	log->acs[nvme_admin_get_features]	= cpu_to_le32(1 << 0);
188 	log->acs[nvme_admin_async_event]	= cpu_to_le32(1 << 0);
189 	log->acs[nvme_admin_keep_alive]		= cpu_to_le32(1 << 0);
190 
191 	log->iocs[nvme_cmd_read]		= cpu_to_le32(1 << 0);
192 	log->iocs[nvme_cmd_write]		= cpu_to_le32(1 << 0);
193 	log->iocs[nvme_cmd_flush]		= cpu_to_le32(1 << 0);
194 	log->iocs[nvme_cmd_dsm]			= cpu_to_le32(1 << 0);
195 	log->iocs[nvme_cmd_write_zeroes]	= cpu_to_le32(1 << 0);
196 
197 	status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
198 
199 	kfree(log);
200 out:
201 	nvmet_req_complete(req, status);
202 }
203 
204 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
205 {
206 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
207 	u16 status = NVME_SC_INTERNAL;
208 	size_t len;
209 
210 	if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
211 		goto out;
212 
213 	mutex_lock(&ctrl->lock);
214 	if (ctrl->nr_changed_ns == U32_MAX)
215 		len = sizeof(__le32);
216 	else
217 		len = ctrl->nr_changed_ns * sizeof(__le32);
218 	status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
219 	if (!status)
220 		status = nvmet_zero_sgl(req, len, req->transfer_len - len);
221 	ctrl->nr_changed_ns = 0;
222 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR);
223 	mutex_unlock(&ctrl->lock);
224 out:
225 	nvmet_req_complete(req, status);
226 }
227 
228 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid,
229 		struct nvme_ana_group_desc *desc)
230 {
231 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
232 	struct nvmet_ns *ns;
233 	u32 count = 0;
234 
235 	if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) {
236 		rcu_read_lock();
237 		list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link)
238 			if (ns->anagrpid == grpid)
239 				desc->nsids[count++] = cpu_to_le32(ns->nsid);
240 		rcu_read_unlock();
241 	}
242 
243 	desc->grpid = cpu_to_le32(grpid);
244 	desc->nnsids = cpu_to_le32(count);
245 	desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
246 	desc->state = req->port->ana_state[grpid];
247 	memset(desc->rsvd17, 0, sizeof(desc->rsvd17));
248 	return sizeof(struct nvme_ana_group_desc) + count * sizeof(__le32);
249 }
250 
251 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)
252 {
253 	struct nvme_ana_rsp_hdr hdr = { 0, };
254 	struct nvme_ana_group_desc *desc;
255 	size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */
256 	size_t len;
257 	u32 grpid;
258 	u16 ngrps = 0;
259 	u16 status;
260 
261 	status = NVME_SC_INTERNAL;
262 	desc = kmalloc(sizeof(struct nvme_ana_group_desc) +
263 			NVMET_MAX_NAMESPACES * sizeof(__le32), GFP_KERNEL);
264 	if (!desc)
265 		goto out;
266 
267 	down_read(&nvmet_ana_sem);
268 	for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
269 		if (!nvmet_ana_group_enabled[grpid])
270 			continue;
271 		len = nvmet_format_ana_group(req, grpid, desc);
272 		status = nvmet_copy_to_sgl(req, offset, desc, len);
273 		if (status)
274 			break;
275 		offset += len;
276 		ngrps++;
277 	}
278 	for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
279 		if (nvmet_ana_group_enabled[grpid])
280 			ngrps++;
281 	}
282 
283 	hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
284 	hdr.ngrps = cpu_to_le16(ngrps);
285 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE);
286 	up_read(&nvmet_ana_sem);
287 
288 	kfree(desc);
289 
290 	/* copy the header last once we know the number of groups */
291 	status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr));
292 out:
293 	nvmet_req_complete(req, status);
294 }
295 
296 static void nvmet_execute_get_log_page(struct nvmet_req *req)
297 {
298 	if (!nvmet_check_transfer_len(req, nvmet_get_log_page_len(req->cmd)))
299 		return;
300 
301 	switch (req->cmd->get_log_page.lid) {
302 	case NVME_LOG_ERROR:
303 		return nvmet_execute_get_log_page_error(req);
304 	case NVME_LOG_SMART:
305 		return nvmet_execute_get_log_page_smart(req);
306 	case NVME_LOG_FW_SLOT:
307 		/*
308 		 * We only support a single firmware slot which always is
309 		 * active, so we can zero out the whole firmware slot log and
310 		 * still claim to fully implement this mandatory log page.
311 		 */
312 		return nvmet_execute_get_log_page_noop(req);
313 	case NVME_LOG_CHANGED_NS:
314 		return nvmet_execute_get_log_changed_ns(req);
315 	case NVME_LOG_CMD_EFFECTS:
316 		return nvmet_execute_get_log_cmd_effects_ns(req);
317 	case NVME_LOG_ANA:
318 		return nvmet_execute_get_log_page_ana(req);
319 	}
320 	pr_err("unhandled lid %d on qid %d\n",
321 	       req->cmd->get_log_page.lid, req->sq->qid);
322 	req->error_loc = offsetof(struct nvme_get_log_page_command, lid);
323 	nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
324 }
325 
326 static void nvmet_id_set_model_number(struct nvme_id_ctrl *id,
327 				      struct nvmet_subsys *subsys)
328 {
329 	const char *model = NVMET_DEFAULT_CTRL_MODEL;
330 	struct nvmet_subsys_model *subsys_model;
331 
332 	rcu_read_lock();
333 	subsys_model = rcu_dereference(subsys->model);
334 	if (subsys_model)
335 		model = subsys_model->number;
336 	memcpy_and_pad(id->mn, sizeof(id->mn), model, strlen(model), ' ');
337 	rcu_read_unlock();
338 }
339 
340 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
341 {
342 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
343 	struct nvme_id_ctrl *id;
344 	u32 cmd_capsule_size;
345 	u16 status = 0;
346 
347 	id = kzalloc(sizeof(*id), GFP_KERNEL);
348 	if (!id) {
349 		status = NVME_SC_INTERNAL;
350 		goto out;
351 	}
352 
353 	/* XXX: figure out how to assign real vendors IDs. */
354 	id->vid = 0;
355 	id->ssvid = 0;
356 
357 	memset(id->sn, ' ', sizeof(id->sn));
358 	bin2hex(id->sn, &ctrl->subsys->serial,
359 		min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
360 	nvmet_id_set_model_number(id, ctrl->subsys);
361 	memcpy_and_pad(id->fr, sizeof(id->fr),
362 		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
363 
364 	id->rab = 6;
365 
366 	/*
367 	 * XXX: figure out how we can assign a IEEE OUI, but until then
368 	 * the safest is to leave it as zeroes.
369 	 */
370 
371 	/* we support multiple ports, multiples hosts and ANA: */
372 	id->cmic = (1 << 0) | (1 << 1) | (1 << 3);
373 
374 	/* Limit MDTS according to transport capability */
375 	if (ctrl->ops->get_mdts)
376 		id->mdts = ctrl->ops->get_mdts(ctrl);
377 	else
378 		id->mdts = 0;
379 
380 	id->cntlid = cpu_to_le16(ctrl->cntlid);
381 	id->ver = cpu_to_le32(ctrl->subsys->ver);
382 
383 	/* XXX: figure out what to do about RTD3R/RTD3 */
384 	id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
385 	id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
386 		NVME_CTRL_ATTR_TBKAS);
387 
388 	id->oacs = 0;
389 
390 	/*
391 	 * We don't really have a practical limit on the number of abort
392 	 * comands.  But we don't do anything useful for abort either, so
393 	 * no point in allowing more abort commands than the spec requires.
394 	 */
395 	id->acl = 3;
396 
397 	id->aerl = NVMET_ASYNC_EVENTS - 1;
398 
399 	/* first slot is read-only, only one slot supported */
400 	id->frmw = (1 << 0) | (1 << 1);
401 	id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
402 	id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
403 	id->npss = 0;
404 
405 	/* We support keep-alive timeout in granularity of seconds */
406 	id->kas = cpu_to_le16(NVMET_KAS);
407 
408 	id->sqes = (0x6 << 4) | 0x6;
409 	id->cqes = (0x4 << 4) | 0x4;
410 
411 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
412 	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
413 
414 	id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
415 	id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
416 	id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
417 			NVME_CTRL_ONCS_WRITE_ZEROES);
418 
419 	/* XXX: don't report vwc if the underlying device is write through */
420 	id->vwc = NVME_CTRL_VWC_PRESENT;
421 
422 	/*
423 	 * We can't support atomic writes bigger than a LBA without support
424 	 * from the backend device.
425 	 */
426 	id->awun = 0;
427 	id->awupf = 0;
428 
429 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
430 	if (ctrl->ops->has_keyed_sgls)
431 		id->sgls |= cpu_to_le32(1 << 2);
432 	if (req->port->inline_data_size)
433 		id->sgls |= cpu_to_le32(1 << 20);
434 
435 	strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
436 
437 	/*
438 	 * Max command capsule size is sqe + in-capsule data size.
439 	 * Disable in-capsule data for Metadata capable controllers.
440 	 */
441 	cmd_capsule_size = sizeof(struct nvme_command);
442 	if (!ctrl->pi_support)
443 		cmd_capsule_size += req->port->inline_data_size;
444 	id->ioccsz = cpu_to_le32(cmd_capsule_size / 16);
445 
446 	/* Max response capsule size is cqe */
447 	id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
448 
449 	id->msdbd = ctrl->ops->msdbd;
450 
451 	id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
452 	id->anatt = 10; /* random value */
453 	id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
454 	id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
455 
456 	/*
457 	 * Meh, we don't really support any power state.  Fake up the same
458 	 * values that qemu does.
459 	 */
460 	id->psd[0].max_power = cpu_to_le16(0x9c4);
461 	id->psd[0].entry_lat = cpu_to_le32(0x10);
462 	id->psd[0].exit_lat = cpu_to_le32(0x4);
463 
464 	id->nwpc = 1 << 0; /* write protect and no write protect */
465 
466 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
467 
468 	kfree(id);
469 out:
470 	nvmet_req_complete(req, status);
471 }
472 
473 static void nvmet_execute_identify_ns(struct nvmet_req *req)
474 {
475 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
476 	struct nvmet_ns *ns;
477 	struct nvme_id_ns *id;
478 	u16 status = 0;
479 
480 	if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
481 		req->error_loc = offsetof(struct nvme_identify, nsid);
482 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
483 		goto out;
484 	}
485 
486 	id = kzalloc(sizeof(*id), GFP_KERNEL);
487 	if (!id) {
488 		status = NVME_SC_INTERNAL;
489 		goto out;
490 	}
491 
492 	/* return an all zeroed buffer if we can't find an active namespace */
493 	ns = nvmet_find_namespace(ctrl, req->cmd->identify.nsid);
494 	if (!ns)
495 		goto done;
496 
497 	nvmet_ns_revalidate(ns);
498 
499 	/*
500 	 * nuse = ncap = nsze isn't always true, but we have no way to find
501 	 * that out from the underlying device.
502 	 */
503 	id->ncap = id->nsze = cpu_to_le64(ns->size >> ns->blksize_shift);
504 	switch (req->port->ana_state[ns->anagrpid]) {
505 	case NVME_ANA_INACCESSIBLE:
506 	case NVME_ANA_PERSISTENT_LOSS:
507 		break;
508 	default:
509 		id->nuse = id->nsze;
510 		break;
511         }
512 
513 	if (ns->bdev)
514 		nvmet_bdev_set_limits(ns->bdev, id);
515 
516 	/*
517 	 * We just provide a single LBA format that matches what the
518 	 * underlying device reports.
519 	 */
520 	id->nlbaf = 0;
521 	id->flbas = 0;
522 
523 	/*
524 	 * Our namespace might always be shared.  Not just with other
525 	 * controllers, but also with any other user of the block device.
526 	 */
527 	id->nmic = (1 << 0);
528 	id->anagrpid = cpu_to_le32(ns->anagrpid);
529 
530 	memcpy(&id->nguid, &ns->nguid, sizeof(id->nguid));
531 
532 	id->lbaf[0].ds = ns->blksize_shift;
533 
534 	if (ctrl->pi_support && nvmet_ns_has_pi(ns)) {
535 		id->dpc = NVME_NS_DPC_PI_FIRST | NVME_NS_DPC_PI_LAST |
536 			  NVME_NS_DPC_PI_TYPE1 | NVME_NS_DPC_PI_TYPE2 |
537 			  NVME_NS_DPC_PI_TYPE3;
538 		id->mc = NVME_MC_EXTENDED_LBA;
539 		id->dps = ns->pi_type;
540 		id->flbas = NVME_NS_FLBAS_META_EXT;
541 		id->lbaf[0].ms = cpu_to_le16(ns->metadata_size);
542 	}
543 
544 	if (ns->readonly)
545 		id->nsattr |= (1 << 0);
546 	nvmet_put_namespace(ns);
547 done:
548 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
549 	kfree(id);
550 out:
551 	nvmet_req_complete(req, status);
552 }
553 
554 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
555 {
556 	static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
557 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
558 	struct nvmet_ns *ns;
559 	u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
560 	__le32 *list;
561 	u16 status = 0;
562 	int i = 0;
563 
564 	list = kzalloc(buf_size, GFP_KERNEL);
565 	if (!list) {
566 		status = NVME_SC_INTERNAL;
567 		goto out;
568 	}
569 
570 	rcu_read_lock();
571 	list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
572 		if (ns->nsid <= min_nsid)
573 			continue;
574 		list[i++] = cpu_to_le32(ns->nsid);
575 		if (i == buf_size / sizeof(__le32))
576 			break;
577 	}
578 	rcu_read_unlock();
579 
580 	status = nvmet_copy_to_sgl(req, 0, list, buf_size);
581 
582 	kfree(list);
583 out:
584 	nvmet_req_complete(req, status);
585 }
586 
587 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
588 				    void *id, off_t *off)
589 {
590 	struct nvme_ns_id_desc desc = {
591 		.nidt = type,
592 		.nidl = len,
593 	};
594 	u16 status;
595 
596 	status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
597 	if (status)
598 		return status;
599 	*off += sizeof(desc);
600 
601 	status = nvmet_copy_to_sgl(req, *off, id, len);
602 	if (status)
603 		return status;
604 	*off += len;
605 
606 	return 0;
607 }
608 
609 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
610 {
611 	struct nvmet_ns *ns;
612 	u16 status = 0;
613 	off_t off = 0;
614 
615 	ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
616 	if (!ns) {
617 		req->error_loc = offsetof(struct nvme_identify, nsid);
618 		status = NVME_SC_INVALID_NS | NVME_SC_DNR;
619 		goto out;
620 	}
621 
622 	if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) {
623 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
624 						  NVME_NIDT_UUID_LEN,
625 						  &ns->uuid, &off);
626 		if (status)
627 			goto out_put_ns;
628 	}
629 	if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) {
630 		status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
631 						  NVME_NIDT_NGUID_LEN,
632 						  &ns->nguid, &off);
633 		if (status)
634 			goto out_put_ns;
635 	}
636 
637 	if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
638 			off) != NVME_IDENTIFY_DATA_SIZE - off)
639 		status = NVME_SC_INTERNAL | NVME_SC_DNR;
640 out_put_ns:
641 	nvmet_put_namespace(ns);
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 = req->sq->ctrl->subsys;
701 	u16 status = NVME_SC_FEATURE_NOT_CHANGEABLE;
702 
703 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->rw.nsid);
704 	if (unlikely(!req->ns)) {
705 		req->error_loc = offsetof(struct nvme_common_command, nsid);
706 		return status;
707 	}
708 
709 	mutex_lock(&subsys->lock);
710 	switch (write_protect) {
711 	case NVME_NS_WRITE_PROTECT:
712 		req->ns->readonly = true;
713 		status = nvmet_write_protect_flush_sync(req);
714 		if (status)
715 			req->ns->readonly = false;
716 		break;
717 	case NVME_NS_NO_WRITE_PROTECT:
718 		req->ns->readonly = false;
719 		status = 0;
720 		break;
721 	default:
722 		break;
723 	}
724 
725 	if (!status)
726 		nvmet_ns_changed(subsys, req->ns->nsid);
727 	mutex_unlock(&subsys->lock);
728 	return status;
729 }
730 
731 u16 nvmet_set_feat_kato(struct nvmet_req *req)
732 {
733 	u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
734 
735 	req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
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 static void nvmet_execute_set_features(struct nvmet_req *req)
758 {
759 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
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 = req->sq->ctrl->subsys;
804 	u32 result;
805 
806 	req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->common.nsid);
807 	if (!req->ns)  {
808 		req->error_loc = offsetof(struct nvme_common_command, nsid);
809 		return NVME_SC_INVALID_NS | NVME_SC_DNR;
810 	}
811 	mutex_lock(&subsys->lock);
812 	if (req->ns->readonly == true)
813 		result = NVME_NS_WRITE_PROTECT;
814 	else
815 		result = NVME_NS_NO_WRITE_PROTECT;
816 	nvmet_set_result(req, result);
817 	mutex_unlock(&subsys->lock);
818 
819 	return 0;
820 }
821 
822 void nvmet_get_feat_kato(struct nvmet_req *req)
823 {
824 	nvmet_set_result(req, req->sq->ctrl->kato * 1000);
825 }
826 
827 void nvmet_get_feat_async_event(struct nvmet_req *req)
828 {
829 	nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
830 }
831 
832 static void nvmet_execute_get_features(struct nvmet_req *req)
833 {
834 	struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
835 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
836 	u16 status = 0;
837 
838 	if (!nvmet_check_transfer_len(req, nvmet_feat_data_len(req, cdw10)))
839 		return;
840 
841 	switch (cdw10 & 0xff) {
842 	/*
843 	 * These features are mandatory in the spec, but we don't
844 	 * have a useful way to implement them.  We'll eventually
845 	 * need to come up with some fake values for these.
846 	 */
847 #if 0
848 	case NVME_FEAT_ARBITRATION:
849 		break;
850 	case NVME_FEAT_POWER_MGMT:
851 		break;
852 	case NVME_FEAT_TEMP_THRESH:
853 		break;
854 	case NVME_FEAT_ERR_RECOVERY:
855 		break;
856 	case NVME_FEAT_IRQ_COALESCE:
857 		break;
858 	case NVME_FEAT_IRQ_CONFIG:
859 		break;
860 	case NVME_FEAT_WRITE_ATOMIC:
861 		break;
862 #endif
863 	case NVME_FEAT_ASYNC_EVENT:
864 		nvmet_get_feat_async_event(req);
865 		break;
866 	case NVME_FEAT_VOLATILE_WC:
867 		nvmet_set_result(req, 1);
868 		break;
869 	case NVME_FEAT_NUM_QUEUES:
870 		nvmet_set_result(req,
871 			(subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
872 		break;
873 	case NVME_FEAT_KATO:
874 		nvmet_get_feat_kato(req);
875 		break;
876 	case NVME_FEAT_HOST_ID:
877 		/* need 128-bit host identifier flag */
878 		if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) {
879 			req->error_loc =
880 				offsetof(struct nvme_common_command, cdw11);
881 			status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
882 			break;
883 		}
884 
885 		status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
886 				sizeof(req->sq->ctrl->hostid));
887 		break;
888 	case NVME_FEAT_WRITE_PROTECT:
889 		status = nvmet_get_feat_write_protect(req);
890 		break;
891 	default:
892 		req->error_loc =
893 			offsetof(struct nvme_common_command, cdw10);
894 		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
895 		break;
896 	}
897 
898 	nvmet_req_complete(req, status);
899 }
900 
901 void nvmet_execute_async_event(struct nvmet_req *req)
902 {
903 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
904 
905 	if (!nvmet_check_transfer_len(req, 0))
906 		return;
907 
908 	mutex_lock(&ctrl->lock);
909 	if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
910 		mutex_unlock(&ctrl->lock);
911 		nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
912 		return;
913 	}
914 	ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
915 	mutex_unlock(&ctrl->lock);
916 
917 	schedule_work(&ctrl->async_event_work);
918 }
919 
920 void nvmet_execute_keep_alive(struct nvmet_req *req)
921 {
922 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
923 
924 	if (!nvmet_check_transfer_len(req, 0))
925 		return;
926 
927 	pr_debug("ctrl %d update keep-alive timer for %d secs\n",
928 		ctrl->cntlid, ctrl->kato);
929 
930 	mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
931 	nvmet_req_complete(req, 0);
932 }
933 
934 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
935 {
936 	struct nvme_command *cmd = req->cmd;
937 	u16 ret;
938 
939 	if (nvme_is_fabrics(cmd))
940 		return nvmet_parse_fabrics_cmd(req);
941 	if (req->sq->ctrl->subsys->type == NVME_NQN_DISC)
942 		return nvmet_parse_discovery_cmd(req);
943 
944 	ret = nvmet_check_ctrl_status(req, cmd);
945 	if (unlikely(ret))
946 		return ret;
947 
948 	switch (cmd->common.opcode) {
949 	case nvme_admin_get_log_page:
950 		req->execute = nvmet_execute_get_log_page;
951 		return 0;
952 	case nvme_admin_identify:
953 		req->execute = nvmet_execute_identify;
954 		return 0;
955 	case nvme_admin_abort_cmd:
956 		req->execute = nvmet_execute_abort;
957 		return 0;
958 	case nvme_admin_set_features:
959 		req->execute = nvmet_execute_set_features;
960 		return 0;
961 	case nvme_admin_get_features:
962 		req->execute = nvmet_execute_get_features;
963 		return 0;
964 	case nvme_admin_async_event:
965 		req->execute = nvmet_execute_async_event;
966 		return 0;
967 	case nvme_admin_keep_alive:
968 		req->execute = nvmet_execute_keep_alive;
969 		return 0;
970 	}
971 
972 	pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
973 	       req->sq->qid);
974 	req->error_loc = offsetof(struct nvme_common_command, opcode);
975 	return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
976 }
977