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
11struct nvmet_subsys *nvmet_disc_subsys;
12
13static u64 nvmet_genctr;
14
15static 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_TYPE_NOTICE,
25			      NVME_AER_NOTICE_DISC_CHANGED, NVME_LOG_DISC);
26}
27
28void 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
50static 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
66void 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	nvmet_genctr++;
73
74	list_for_each_entry(port, nvmet_ports, global_entry)
75		list_for_each_entry(s, &port->subsystems, entry) {
76			if (s->subsys != subsys)
77				continue;
78			__nvmet_subsys_disc_changed(port, subsys, host);
79		}
80}
81
82void nvmet_referral_enable(struct nvmet_port *parent, struct nvmet_port *port)
83{
84	down_write(&nvmet_config_sem);
85	if (list_empty(&port->entry)) {
86		list_add_tail(&port->entry, &parent->referrals);
87		port->enabled = true;
88		nvmet_port_disc_changed(parent, NULL);
89	}
90	up_write(&nvmet_config_sem);
91}
92
93void nvmet_referral_disable(struct nvmet_port *parent, struct nvmet_port *port)
94{
95	down_write(&nvmet_config_sem);
96	if (!list_empty(&port->entry)) {
97		port->enabled = false;
98		list_del_init(&port->entry);
99		nvmet_port_disc_changed(parent, NULL);
100	}
101	up_write(&nvmet_config_sem);
102}
103
104static void nvmet_format_discovery_entry(struct nvmf_disc_rsp_page_hdr *hdr,
105		struct nvmet_port *port, char *subsys_nqn, char *traddr,
106		u8 type, u32 numrec)
107{
108	struct nvmf_disc_rsp_page_entry *e = &hdr->entries[numrec];
109
110	e->trtype = port->disc_addr.trtype;
111	e->adrfam = port->disc_addr.adrfam;
112	e->treq = port->disc_addr.treq;
113	e->portid = port->disc_addr.portid;
114	/* we support only dynamic controllers */
115	e->cntlid = cpu_to_le16(NVME_CNTLID_DYNAMIC);
116	e->asqsz = cpu_to_le16(NVME_AQ_DEPTH);
117	e->subtype = type;
118	memcpy(e->trsvcid, port->disc_addr.trsvcid, NVMF_TRSVCID_SIZE);
119	memcpy(e->traddr, traddr, NVMF_TRADDR_SIZE);
120	memcpy(e->tsas.common, port->disc_addr.tsas.common, NVMF_TSAS_SIZE);
121	strncpy(e->subnqn, subsys_nqn, NVMF_NQN_SIZE);
122}
123
124/*
125 * nvmet_set_disc_traddr - set a correct discovery log entry traddr
126 *
127 * IP based transports (e.g RDMA) can listen on "any" ipv4/ipv6 addresses
128 * (INADDR_ANY or IN6ADDR_ANY_INIT). The discovery log page traddr reply
129 * must not contain that "any" IP address. If the transport implements
130 * .disc_traddr, use it. this callback will set the discovery traddr
131 * from the req->port address in case the port in question listens
132 * "any" IP address.
133 */
134static void nvmet_set_disc_traddr(struct nvmet_req *req, struct nvmet_port *port,
135		char *traddr)
136{
137	if (req->ops->disc_traddr)
138		req->ops->disc_traddr(req, port, traddr);
139	else
140		memcpy(traddr, port->disc_addr.traddr, NVMF_TRADDR_SIZE);
141}
142
143static size_t discovery_log_entries(struct nvmet_req *req)
144{
145	struct nvmet_ctrl *ctrl = req->sq->ctrl;
146	struct nvmet_subsys_link *p;
147	struct nvmet_port *r;
148	size_t entries = 0;
149
150	list_for_each_entry(p, &req->port->subsystems, entry) {
151		if (!nvmet_host_allowed(p->subsys, ctrl->hostnqn))
152			continue;
153		entries++;
154	}
155	list_for_each_entry(r, &req->port->referrals, entry)
156		entries++;
157	return entries;
158}
159
160static void nvmet_execute_disc_get_log_page(struct nvmet_req *req)
161{
162	const int entry_size = sizeof(struct nvmf_disc_rsp_page_entry);
163	struct nvmet_ctrl *ctrl = req->sq->ctrl;
164	struct nvmf_disc_rsp_page_hdr *hdr;
165	u64 offset = nvmet_get_log_page_offset(req->cmd);
166	size_t data_len = nvmet_get_log_page_len(req->cmd);
167	size_t alloc_len;
168	struct nvmet_subsys_link *p;
169	struct nvmet_port *r;
170	u32 numrec = 0;
171	u16 status = 0;
172	void *buffer;
173
174	if (!nvmet_check_transfer_len(req, data_len))
175		return;
176
177	if (req->cmd->get_log_page.lid != NVME_LOG_DISC) {
178		req->error_loc =
179			offsetof(struct nvme_get_log_page_command, lid);
180		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
181		goto out;
182	}
183
184	/* Spec requires dword aligned offsets */
185	if (offset & 0x3) {
186		req->error_loc =
187			offsetof(struct nvme_get_log_page_command, lpo);
188		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
189		goto out;
190	}
191
192	/*
193	 * Make sure we're passing at least a buffer of response header size.
194	 * If host provided data len is less than the header size, only the
195	 * number of bytes requested by host will be sent to host.
196	 */
197	down_read(&nvmet_config_sem);
198	alloc_len = sizeof(*hdr) + entry_size * discovery_log_entries(req);
199	buffer = kzalloc(alloc_len, GFP_KERNEL);
200	if (!buffer) {
201		up_read(&nvmet_config_sem);
202		status = NVME_SC_INTERNAL;
203		goto out;
204	}
205
206	hdr = buffer;
207	list_for_each_entry(p, &req->port->subsystems, entry) {
208		char traddr[NVMF_TRADDR_SIZE];
209
210		if (!nvmet_host_allowed(p->subsys, ctrl->hostnqn))
211			continue;
212
213		nvmet_set_disc_traddr(req, req->port, traddr);
214		nvmet_format_discovery_entry(hdr, req->port,
215				p->subsys->subsysnqn, traddr,
216				NVME_NQN_NVME, numrec);
217		numrec++;
218	}
219
220	list_for_each_entry(r, &req->port->referrals, entry) {
221		nvmet_format_discovery_entry(hdr, r,
222				NVME_DISC_SUBSYS_NAME,
223				r->disc_addr.traddr,
224				NVME_NQN_DISC, numrec);
225		numrec++;
226	}
227
228	hdr->genctr = cpu_to_le64(nvmet_genctr);
229	hdr->numrec = cpu_to_le64(numrec);
230	hdr->recfmt = cpu_to_le16(0);
231
232	nvmet_clear_aen_bit(req, NVME_AEN_BIT_DISC_CHANGE);
233
234	up_read(&nvmet_config_sem);
235
236	status = nvmet_copy_to_sgl(req, 0, buffer + offset, data_len);
237	kfree(buffer);
238out:
239	nvmet_req_complete(req, status);
240}
241
242static void nvmet_execute_disc_identify(struct nvmet_req *req)
243{
244	struct nvmet_ctrl *ctrl = req->sq->ctrl;
245	struct nvme_id_ctrl *id;
246	const char model[] = "Linux";
247	u16 status = 0;
248
249	if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE))
250		return;
251
252	if (req->cmd->identify.cns != NVME_ID_CNS_CTRL) {
253		req->error_loc = offsetof(struct nvme_identify, cns);
254		status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
255		goto out;
256	}
257
258	id = kzalloc(sizeof(*id), GFP_KERNEL);
259	if (!id) {
260		status = NVME_SC_INTERNAL;
261		goto out;
262	}
263
264	memset(id->sn, ' ', sizeof(id->sn));
265	bin2hex(id->sn, &ctrl->subsys->serial,
266		min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
267	memset(id->fr, ' ', sizeof(id->fr));
268	memcpy_and_pad(id->mn, sizeof(id->mn), model, sizeof(model) - 1, ' ');
269	memcpy_and_pad(id->fr, sizeof(id->fr),
270		       UTS_RELEASE, strlen(UTS_RELEASE), ' ');
271
272	/* no limit on data transfer sizes for now */
273	id->mdts = 0;
274	id->cntlid = cpu_to_le16(ctrl->cntlid);
275	id->ver = cpu_to_le32(ctrl->subsys->ver);
276	id->lpa = (1 << 2);
277
278	/* no enforcement soft-limit for maxcmd - pick arbitrary high value */
279	id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
280
281	id->sgls = cpu_to_le32(1 << 0);	/* we always support SGLs */
282	if (ctrl->ops->flags & NVMF_KEYED_SGLS)
283		id->sgls |= cpu_to_le32(1 << 2);
284	if (req->port->inline_data_size)
285		id->sgls |= cpu_to_le32(1 << 20);
286
287	id->oaes = cpu_to_le32(NVMET_DISC_AEN_CFG_OPTIONAL);
288
289	strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
290
291	status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
292
293	kfree(id);
294out:
295	nvmet_req_complete(req, status);
296}
297
298static void nvmet_execute_disc_set_features(struct nvmet_req *req)
299{
300	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
301	u16 stat;
302
303	if (!nvmet_check_transfer_len(req, 0))
304		return;
305
306	switch (cdw10 & 0xff) {
307	case NVME_FEAT_KATO:
308		stat = nvmet_set_feat_kato(req);
309		break;
310	case NVME_FEAT_ASYNC_EVENT:
311		stat = nvmet_set_feat_async_event(req,
312						  NVMET_DISC_AEN_CFG_OPTIONAL);
313		break;
314	default:
315		req->error_loc =
316			offsetof(struct nvme_common_command, cdw10);
317		stat = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
318		break;
319	}
320
321	nvmet_req_complete(req, stat);
322}
323
324static void nvmet_execute_disc_get_features(struct nvmet_req *req)
325{
326	u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
327	u16 stat = 0;
328
329	if (!nvmet_check_transfer_len(req, 0))
330		return;
331
332	switch (cdw10 & 0xff) {
333	case NVME_FEAT_KATO:
334		nvmet_get_feat_kato(req);
335		break;
336	case NVME_FEAT_ASYNC_EVENT:
337		nvmet_get_feat_async_event(req);
338		break;
339	default:
340		req->error_loc =
341			offsetof(struct nvme_common_command, cdw10);
342		stat = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
343		break;
344	}
345
346	nvmet_req_complete(req, stat);
347}
348
349u16 nvmet_parse_discovery_cmd(struct nvmet_req *req)
350{
351	struct nvme_command *cmd = req->cmd;
352
353	if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
354		pr_err("got cmd %d while not ready\n",
355		       cmd->common.opcode);
356		req->error_loc =
357			offsetof(struct nvme_common_command, opcode);
358		return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
359	}
360
361	switch (cmd->common.opcode) {
362	case nvme_admin_set_features:
363		req->execute = nvmet_execute_disc_set_features;
364		return 0;
365	case nvme_admin_get_features:
366		req->execute = nvmet_execute_disc_get_features;
367		return 0;
368	case nvme_admin_async_event:
369		req->execute = nvmet_execute_async_event;
370		return 0;
371	case nvme_admin_keep_alive:
372		req->execute = nvmet_execute_keep_alive;
373		return 0;
374	case nvme_admin_get_log_page:
375		req->execute = nvmet_execute_disc_get_log_page;
376		return 0;
377	case nvme_admin_identify:
378		req->execute = nvmet_execute_disc_identify;
379		return 0;
380	default:
381		pr_err("unhandled cmd %d\n", cmd->common.opcode);
382		req->error_loc = offsetof(struct nvme_common_command, opcode);
383		return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
384	}
385
386}
387
388int __init nvmet_init_discovery(void)
389{
390	nvmet_disc_subsys =
391		nvmet_subsys_alloc(NVME_DISC_SUBSYS_NAME, NVME_NQN_DISC);
392	return PTR_ERR_OR_ZERO(nvmet_disc_subsys);
393}
394
395void nvmet_exit_discovery(void)
396{
397	nvmet_subsys_put(nvmet_disc_subsys);
398}
399