xref: /linux/drivers/nvme/target/discovery.c (revision 4b99990cdf9560e8a071640baf19f312e6ae02f4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Discovery service for the NVMe over Fabrics target.
4  * Copyright (C) 2016 Intel Corporation. All rights reserved.
5  */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 #include <linux/slab.h>
8 #include <generated/utsrelease.h>
9 #include "nvmet.h"
10 
11 struct nvmet_subsys *nvmet_disc_subsys;
12 
13 static u64 nvmet_genctr;
14 
15 static void __nvmet_disc_changed(struct nvmet_port *port,
16 				 struct nvmet_ctrl *ctrl)
17 {
18 	if (ctrl->port != port)
19 		return;
20 
21 	if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_DISC_CHANGE))
22 		return;
23 
24 	nvmet_add_async_event(ctrl, NVME_AER_NOTICE,
25 			      NVME_AER_NOTICE_DISC_CHANGED, NVME_LOG_DISC);
26 }
27 
28 void nvmet_port_disc_changed(struct nvmet_port *port,
29 			     struct nvmet_subsys *subsys)
30 {
31 	struct nvmet_ctrl *ctrl;
32 
33 	lockdep_assert_held(&nvmet_config_sem);
34 	nvmet_genctr++;
35 
36 	mutex_lock(&nvmet_disc_subsys->lock);
37 	list_for_each_entry(ctrl, &nvmet_disc_subsys->ctrls, subsys_entry) {
38 		if (subsys && !nvmet_host_allowed(subsys, ctrl->hostnqn))
39 			continue;
40 
41 		__nvmet_disc_changed(port, ctrl);
42 	}
43 	mutex_unlock(&nvmet_disc_subsys->lock);
44 
45 	/* If transport can signal change, notify transport */
46 	if (port->tr_ops && port->tr_ops->discovery_chg)
47 		port->tr_ops->discovery_chg(port);
48 }
49 
50 static void __nvmet_subsys_disc_changed(struct nvmet_port *port,
51 					struct nvmet_subsys *subsys,
52 					struct nvmet_host *host)
53 {
54 	struct nvmet_ctrl *ctrl;
55 
56 	mutex_lock(&nvmet_disc_subsys->lock);
57 	list_for_each_entry(ctrl, &nvmet_disc_subsys->ctrls, subsys_entry) {
58 		if (host && strcmp(nvmet_host_name(host), ctrl->hostnqn))
59 			continue;
60 
61 		__nvmet_disc_changed(port, ctrl);
62 	}
63 	mutex_unlock(&nvmet_disc_subsys->lock);
64 }
65 
66 void nvmet_subsys_disc_changed(struct nvmet_subsys *subsys,
67 			       struct nvmet_host *host)
68 {
69 	struct nvmet_port *port;
70 	struct nvmet_subsys_link *s;
71 
72 	lockdep_assert_held(&nvmet_config_sem);
73 	nvmet_genctr++;
74 
75 	list_for_each_entry(port, nvmet_ports, global_entry)
76 		list_for_each_entry(s, &port->subsystems, entry) {
77 			if (s->subsys != subsys)
78 				continue;
79 			__nvmet_subsys_disc_changed(port, subsys, host);
80 		}
81 }
82 
83 void nvmet_referral_enable(struct nvmet_port *parent, struct nvmet_port *port)
84 {
85 	down_write(&nvmet_config_sem);
86 	if (list_empty(&port->entry)) {
87 		list_add_tail(&port->entry, &parent->referrals);
88 		port->enabled = true;
89 		nvmet_port_disc_changed(parent, NULL);
90 	}
91 	up_write(&nvmet_config_sem);
92 }
93 
94 void nvmet_referral_disable(struct nvmet_port *parent, struct nvmet_port *port)
95 {
96 	down_write(&nvmet_config_sem);
97 	if (!list_empty(&port->entry)) {
98 		port->enabled = false;
99 		list_del_init(&port->entry);
100 		nvmet_port_disc_changed(parent, NULL);
101 	}
102 	up_write(&nvmet_config_sem);
103 }
104 
105 static void nvmet_format_discovery_entry(struct nvmf_disc_rsp_page_hdr *hdr,
106 		struct nvmet_port *port, char *subsys_nqn, char *traddr,
107 		u8 type, u32 numrec)
108 {
109 	struct nvmf_disc_rsp_page_entry *e = &hdr->entries[numrec];
110 
111 	e->trtype = port->disc_addr.trtype;
112 	e->adrfam = port->disc_addr.adrfam;
113 	e->treq = port->disc_addr.treq;
114 	e->portid = port->disc_addr.portid;
115 	/* we support only dynamic controllers */
116 	e->cntlid = cpu_to_le16(NVME_CNTLID_DYNAMIC);
117 	e->asqsz = cpu_to_le16(NVME_AQ_DEPTH);
118 	e->subtype = type;
119 	memcpy(e->trsvcid, port->disc_addr.trsvcid, NVMF_TRSVCID_SIZE);
120 	memcpy(e->traddr, traddr, NVMF_TRADDR_SIZE);
121 	memcpy(e->tsas.common, port->disc_addr.tsas.common, NVMF_TSAS_SIZE);
122 	strscpy(e->subnqn, subsys_nqn, NVMF_NQN_SIZE);
123 }
124 
125 /*
126  * nvmet_set_disc_traddr - set a correct discovery log entry traddr
127  *
128  * IP based transports (e.g RDMA) can listen on "any" ipv4/ipv6 addresses
129  * (INADDR_ANY or IN6ADDR_ANY_INIT). The discovery log page traddr reply
130  * must not contain that "any" IP address. If the transport implements
131  * .disc_traddr, use it. this callback will set the discovery traddr
132  * from the req->port address in case the port in question listens
133  * "any" IP address.
134  */
135 static void nvmet_set_disc_traddr(struct nvmet_req *req, struct nvmet_port *port,
136 		char *traddr)
137 {
138 	if (req->ops->disc_traddr)
139 		req->ops->disc_traddr(req, port, traddr);
140 	else
141 		memcpy(traddr, port->disc_addr.traddr, NVMF_TRADDR_SIZE);
142 }
143 
144 static size_t discovery_log_entries(struct nvmet_req *req)
145 {
146 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
147 	struct nvmet_subsys_link *p;
148 	struct nvmet_port *r;
149 	size_t entries = 1;
150 
151 	list_for_each_entry(p, &req->port->subsystems, entry) {
152 		if (!nvmet_host_allowed(p->subsys, ctrl->hostnqn))
153 			continue;
154 		entries++;
155 	}
156 	list_for_each_entry(r, &req->port->referrals, entry)
157 		entries++;
158 	return entries;
159 }
160 
161 static void nvmet_execute_disc_get_log_page(struct nvmet_req *req)
162 {
163 	const int entry_size = sizeof(struct nvmf_disc_rsp_page_entry);
164 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
165 	struct nvmf_disc_rsp_page_hdr *hdr;
166 	u64 offset = nvmet_get_log_page_offset(req->cmd);
167 	size_t data_len = nvmet_get_log_page_len(req->cmd);
168 	size_t alloc_len;
169 	size_t copy_len;
170 	struct nvmet_subsys_link *p;
171 	struct nvmet_port *r;
172 	u32 numrec = 0;
173 	u16 status = 0;
174 	void *buffer;
175 	char traddr[NVMF_TRADDR_SIZE];
176 
177 	if (!nvmet_check_transfer_len(req, data_len))
178 		return;
179 
180 	if (req->cmd->get_log_page.lid != NVME_LOG_DISC) {
181 		req->error_loc =
182 			offsetof(struct nvme_get_log_page_command, lid);
183 		status = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR;
184 		goto out;
185 	}
186 
187 	/* Spec requires dword aligned offsets */
188 	if (offset & 0x3) {
189 		req->error_loc =
190 			offsetof(struct nvme_get_log_page_command, lpo);
191 		status = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR;
192 		goto out;
193 	}
194 
195 	/*
196 	 * Make sure we're passing at least a buffer of response header size.
197 	 * If host provided data len is less than the header size, only the
198 	 * number of bytes requested by host will be sent to host.
199 	 */
200 	down_read(&nvmet_config_sem);
201 	alloc_len = sizeof(*hdr) + entry_size * discovery_log_entries(req);
202 	buffer = kzalloc(alloc_len, GFP_KERNEL);
203 	if (!buffer) {
204 		up_read(&nvmet_config_sem);
205 		status = NVME_SC_INTERNAL;
206 		goto out;
207 	}
208 	hdr = buffer;
209 
210 	nvmet_set_disc_traddr(req, req->port, traddr);
211 
212 	nvmet_format_discovery_entry(hdr, req->port,
213 				     nvmet_disc_subsys->subsysnqn,
214 				     traddr, NVME_NQN_CURR, numrec);
215 	numrec++;
216 
217 	list_for_each_entry(p, &req->port->subsystems, entry) {
218 		if (!nvmet_host_allowed(p->subsys, ctrl->hostnqn))
219 			continue;
220 
221 		nvmet_format_discovery_entry(hdr, req->port,
222 				p->subsys->subsysnqn, traddr,
223 				NVME_NQN_NVME, numrec);
224 		numrec++;
225 	}
226 
227 	list_for_each_entry(r, &req->port->referrals, entry) {
228 		if (r->disc_addr.trtype == NVMF_TRTYPE_PCI)
229 			continue;
230 
231 		nvmet_format_discovery_entry(hdr, r,
232 				NVME_DISC_SUBSYS_NAME,
233 				r->disc_addr.traddr,
234 				NVME_NQN_DISC, numrec);
235 		numrec++;
236 	}
237 
238 	hdr->genctr = cpu_to_le64(nvmet_genctr);
239 	hdr->numrec = cpu_to_le64(numrec);
240 	hdr->recfmt = cpu_to_le16(0);
241 
242 	nvmet_clear_aen_bit(req, NVME_AEN_BIT_DISC_CHANGE);
243 
244 	up_read(&nvmet_config_sem);
245 
246 	/*
247 	 * Validate the host-supplied log page offset before copying out.
248 	 * Without this check, the host controls a 64-bit byte offset into
249 	 * a small kzalloc'd buffer: a value past the log page lets the
250 	 * subsequent memcpy read adjacent kernel heap, and a value aimed
251 	 * at unmapped kernel memory faults the in-kernel copy and crashes
252 	 * the target host. The Discovery controller is unauthenticated,
253 	 * so the bug is reachable from any reachable fabric peer.
254 	 */
255 	if (offset > alloc_len) {
256 		req->error_loc =
257 			offsetof(struct nvme_get_log_page_command, lpo);
258 		status = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR;
259 		goto out_free_buffer;
260 	}
261 
262 	copy_len = min_t(size_t, data_len, alloc_len - offset);
263 	status = nvmet_copy_to_sgl(req, 0, buffer + offset, copy_len);
264 	if (!status && copy_len < data_len)
265 		status = nvmet_zero_sgl(req, copy_len, data_len - copy_len);
266 out_free_buffer:
267 	kfree(buffer);
268 out:
269 	nvmet_req_complete(req, status);
270 }
271 
272 static void nvmet_execute_disc_identify(struct nvmet_req *req)
273 {
274 	struct nvmet_ctrl *ctrl = req->sq->ctrl;
275 	struct nvme_id_ctrl *id;
276 	u16 status = 0;
277 
278 	if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE))
279 		return;
280 
281 	if (req->cmd->identify.cns != NVME_ID_CNS_CTRL) {
282 		req->error_loc = offsetof(struct nvme_identify, cns);
283 		status = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR;
284 		goto out;
285 	}
286 
287 	id = kzalloc_obj(*id);
288 	if (!id) {
289 		status = NVME_SC_INTERNAL;
290 		goto out;
291 	}
292 
293 	memcpy(id->sn, ctrl->subsys->serial, NVMET_SN_MAX_SIZE);
294 	memset(id->fr, ' ', sizeof(id->fr));
295 	memcpy_and_pad(id->mn, sizeof(id->mn), ctrl->subsys->model_number,
296 		       strlen(ctrl->subsys->model_number), ' ');
297 	memcpy_and_pad(id->fr, sizeof(id->fr),
298 		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
299 
300 	id->cntrltype = NVME_CTRL_DISC;
301 
302 	/* no limit on data transfer sizes for now */
303 	id->mdts = 0;
304 	id->cntlid = cpu_to_le16(ctrl->cntlid);
305 	id->ver = cpu_to_le32(ctrl->subsys->ver);
306 	id->lpa = (1 << 2);
307 
308 	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
309 	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD(ctrl));
310 
311 	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
312 	if (ctrl->ops->flags & NVMF_KEYED_SGLS)
313 		id->sgls |= cpu_to_le32(1 << 2);
314 	if (req->port->inline_data_size)
315 		id->sgls |= cpu_to_le32(1 << 20);
316 
317 	id->oaes = cpu_to_le32(NVMET_DISC_AEN_CFG_OPTIONAL);
318 
319 	strscpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
320 
321 	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
322 
323 	kfree(id);
324 out:
325 	nvmet_req_complete(req, status);
326 }
327 
328 static void nvmet_execute_disc_set_features(struct nvmet_req *req)
329 {
330 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
331 	u16 stat;
332 
333 	if (!nvmet_check_transfer_len(req, 0))
334 		return;
335 
336 	switch (cdw10 & 0xff) {
337 	case NVME_FEAT_KATO:
338 		stat = nvmet_set_feat_kato(req);
339 		break;
340 	case NVME_FEAT_ASYNC_EVENT:
341 		stat = nvmet_set_feat_async_event(req,
342 						  NVMET_DISC_AEN_CFG_OPTIONAL);
343 		break;
344 	default:
345 		req->error_loc =
346 			offsetof(struct nvme_common_command, cdw10);
347 		stat = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR;
348 		break;
349 	}
350 
351 	nvmet_req_complete(req, stat);
352 }
353 
354 static void nvmet_execute_disc_get_features(struct nvmet_req *req)
355 {
356 	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
357 	u16 stat = 0;
358 
359 	if (!nvmet_check_transfer_len(req, 0))
360 		return;
361 
362 	switch (cdw10 & 0xff) {
363 	case NVME_FEAT_KATO:
364 		nvmet_get_feat_kato(req);
365 		break;
366 	case NVME_FEAT_ASYNC_EVENT:
367 		nvmet_get_feat_async_event(req);
368 		break;
369 	default:
370 		req->error_loc =
371 			offsetof(struct nvme_common_command, cdw10);
372 		stat = NVME_SC_INVALID_FIELD | NVME_STATUS_DNR;
373 		break;
374 	}
375 
376 	nvmet_req_complete(req, stat);
377 }
378 
379 u32 nvmet_discovery_cmd_data_len(struct nvmet_req *req)
380 {
381 	struct nvme_command *cmd = req->cmd;
382 
383 	switch (cmd->common.opcode) {
384 	case nvme_admin_get_log_page:
385 		return nvmet_get_log_page_len(req->cmd);
386 	case nvme_admin_identify:
387 		return NVME_IDENTIFY_DATA_SIZE;
388 	default:
389 		return 0;
390 	}
391 }
392 
393 u16 nvmet_parse_discovery_cmd(struct nvmet_req *req)
394 {
395 	struct nvme_command *cmd = req->cmd;
396 
397 	if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
398 		pr_err("got cmd %d while not ready\n",
399 		       cmd->common.opcode);
400 		req->error_loc =
401 			offsetof(struct nvme_common_command, opcode);
402 		return NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR;
403 	}
404 
405 	switch (cmd->common.opcode) {
406 	case nvme_admin_set_features:
407 		req->execute = nvmet_execute_disc_set_features;
408 		return 0;
409 	case nvme_admin_get_features:
410 		req->execute = nvmet_execute_disc_get_features;
411 		return 0;
412 	case nvme_admin_async_event:
413 		req->execute = nvmet_execute_async_event;
414 		return 0;
415 	case nvme_admin_keep_alive:
416 		req->execute = nvmet_execute_keep_alive;
417 		return 0;
418 	case nvme_admin_get_log_page:
419 		req->execute = nvmet_execute_disc_get_log_page;
420 		return 0;
421 	case nvme_admin_identify:
422 		req->execute = nvmet_execute_disc_identify;
423 		return 0;
424 	default:
425 		pr_debug("unhandled cmd %d\n", cmd->common.opcode);
426 		req->error_loc = offsetof(struct nvme_common_command, opcode);
427 		return NVME_SC_INVALID_OPCODE | NVME_STATUS_DNR;
428 	}
429 
430 }
431 
432 int __init nvmet_init_discovery(void)
433 {
434 	nvmet_disc_subsys =
435 		nvmet_subsys_alloc(NVME_DISC_SUBSYS_NAME, NVME_NQN_CURR);
436 	return PTR_ERR_OR_ZERO(nvmet_disc_subsys);
437 }
438 
439 void nvmet_exit_discovery(void)
440 {
441 	nvmet_subsys_put(nvmet_disc_subsys);
442 }
443