xref: /linux/fs/afs/vl_list.c (revision 40d269c000bda9fcd276a0412a9cebd3f6e344c5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS vlserver list management.
3  *
4  * Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/slab.h>
10 #include "internal.h"
11 
12 struct afs_vlserver *afs_alloc_vlserver(const char *name, size_t name_len,
13 					unsigned short port)
14 {
15 	struct afs_vlserver *vlserver;
16 	static atomic_t debug_ids;
17 
18 	vlserver = kzalloc(struct_size(vlserver, name, name_len + 1),
19 			   GFP_KERNEL);
20 	if (vlserver) {
21 		refcount_set(&vlserver->ref, 1);
22 		rwlock_init(&vlserver->lock);
23 		init_waitqueue_head(&vlserver->probe_wq);
24 		spin_lock_init(&vlserver->probe_lock);
25 		vlserver->debug_id = atomic_inc_return(&debug_ids);
26 		vlserver->rtt = UINT_MAX;
27 		vlserver->name_len = name_len;
28 		vlserver->service_id = VL_SERVICE;
29 		vlserver->port = port;
30 		memcpy(vlserver->name, name, name_len);
31 	}
32 	return vlserver;
33 }
34 
35 static void afs_vlserver_rcu(struct rcu_head *rcu)
36 {
37 	struct afs_vlserver *vlserver = container_of(rcu, struct afs_vlserver, rcu);
38 
39 	afs_put_addrlist(rcu_access_pointer(vlserver->addresses),
40 			 afs_alist_trace_put_vlserver);
41 	kfree_rcu(vlserver, rcu);
42 }
43 
44 void afs_put_vlserver(struct afs_net *net, struct afs_vlserver *vlserver)
45 {
46 	if (vlserver &&
47 	    refcount_dec_and_test(&vlserver->ref))
48 		call_rcu(&vlserver->rcu, afs_vlserver_rcu);
49 }
50 
51 struct afs_vlserver_list *afs_alloc_vlserver_list(unsigned int nr_servers)
52 {
53 	struct afs_vlserver_list *vllist;
54 
55 	vllist = kzalloc(struct_size(vllist, servers, nr_servers), GFP_KERNEL);
56 	if (vllist) {
57 		refcount_set(&vllist->ref, 1);
58 		rwlock_init(&vllist->lock);
59 	}
60 
61 	return vllist;
62 }
63 
64 void afs_put_vlserverlist(struct afs_net *net, struct afs_vlserver_list *vllist)
65 {
66 	if (vllist) {
67 		if (refcount_dec_and_test(&vllist->ref)) {
68 			int i;
69 
70 			for (i = 0; i < vllist->nr_servers; i++) {
71 				afs_put_vlserver(net, vllist->servers[i].server);
72 			}
73 			kfree_rcu(vllist, rcu);
74 		}
75 	}
76 }
77 
78 static u16 afs_extract_le16(const u8 **_b)
79 {
80 	u16 val;
81 
82 	val  = (u16)*(*_b)++ << 0;
83 	val |= (u16)*(*_b)++ << 8;
84 	return val;
85 }
86 
87 /*
88  * Build a VL server address list from a DNS queried server list.
89  */
90 static struct afs_addr_list *afs_extract_vl_addrs(struct afs_net *net,
91 						  const u8 **_b, const u8 *end,
92 						  u8 nr_addrs, u16 port)
93 {
94 	struct afs_addr_list *alist;
95 	const u8 *b = *_b;
96 	int ret = -EINVAL;
97 
98 	alist = afs_alloc_addrlist(nr_addrs);
99 	if (!alist)
100 		return ERR_PTR(-ENOMEM);
101 	if (nr_addrs == 0)
102 		return alist;
103 
104 	for (; nr_addrs > 0 && end - b >= nr_addrs; nr_addrs--) {
105 		struct dns_server_list_v1_address hdr;
106 		__be32 x[4];
107 
108 		hdr.address_type = *b++;
109 
110 		switch (hdr.address_type) {
111 		case DNS_ADDRESS_IS_IPV4:
112 			if (end - b < 4) {
113 				_leave(" = -EINVAL [short inet]");
114 				goto error;
115 			}
116 			memcpy(x, b, 4);
117 			ret = afs_merge_fs_addr4(net, alist, x[0], port);
118 			if (ret < 0)
119 				goto error;
120 			b += 4;
121 			break;
122 
123 		case DNS_ADDRESS_IS_IPV6:
124 			if (end - b < 16) {
125 				_leave(" = -EINVAL [short inet6]");
126 				goto error;
127 			}
128 			memcpy(x, b, 16);
129 			ret = afs_merge_fs_addr6(net, alist, x, port);
130 			if (ret < 0)
131 				goto error;
132 			b += 16;
133 			break;
134 
135 		default:
136 			_leave(" = -EADDRNOTAVAIL [unknown af %u]",
137 			       hdr.address_type);
138 			ret = -EADDRNOTAVAIL;
139 			goto error;
140 		}
141 	}
142 
143 	/* Start with IPv6 if available. */
144 	if (alist->nr_ipv4 < alist->nr_addrs)
145 		alist->preferred = alist->nr_ipv4;
146 
147 	*_b = b;
148 	return alist;
149 
150 error:
151 	*_b = b;
152 	afs_put_addrlist(alist, afs_alist_trace_put_parse_error);
153 	return ERR_PTR(ret);
154 }
155 
156 /*
157  * Build a VL server list from a DNS queried server list.
158  */
159 struct afs_vlserver_list *afs_extract_vlserver_list(struct afs_cell *cell,
160 						    const void *buffer,
161 						    size_t buffer_size)
162 {
163 	const struct dns_server_list_v1_header *hdr = buffer;
164 	struct dns_server_list_v1_server bs;
165 	struct afs_vlserver_list *vllist, *previous;
166 	struct afs_addr_list *addrs;
167 	struct afs_vlserver *server;
168 	const u8 *b = buffer, *end = buffer + buffer_size;
169 	int ret = -ENOMEM, nr_servers, i, j;
170 
171 	_enter("");
172 
173 	/* Check that it's a server list, v1 */
174 	if (end - b < sizeof(*hdr) ||
175 	    hdr->hdr.content != DNS_PAYLOAD_IS_SERVER_LIST ||
176 	    hdr->hdr.version != 1) {
177 		pr_notice("kAFS: Got DNS record [%u,%u] len %zu\n",
178 			  hdr->hdr.content, hdr->hdr.version, end - b);
179 		ret = -EDESTADDRREQ;
180 		goto dump;
181 	}
182 
183 	nr_servers = hdr->nr_servers;
184 
185 	vllist = afs_alloc_vlserver_list(nr_servers);
186 	if (!vllist)
187 		return ERR_PTR(-ENOMEM);
188 
189 	vllist->source = (hdr->source < NR__dns_record_source) ?
190 		hdr->source : NR__dns_record_source;
191 	vllist->status = (hdr->status < NR__dns_lookup_status) ?
192 		hdr->status : NR__dns_lookup_status;
193 
194 	read_lock(&cell->vl_servers_lock);
195 	previous = afs_get_vlserverlist(
196 		rcu_dereference_protected(cell->vl_servers,
197 					  lockdep_is_held(&cell->vl_servers_lock)));
198 	read_unlock(&cell->vl_servers_lock);
199 
200 	b += sizeof(*hdr);
201 	while (end - b >= sizeof(bs)) {
202 		bs.name_len	= afs_extract_le16(&b);
203 		bs.priority	= afs_extract_le16(&b);
204 		bs.weight	= afs_extract_le16(&b);
205 		bs.port		= afs_extract_le16(&b);
206 		bs.source	= *b++;
207 		bs.status	= *b++;
208 		bs.protocol	= *b++;
209 		bs.nr_addrs	= *b++;
210 
211 		_debug("extract %u %u %u %u %u %u %*.*s",
212 		       bs.name_len, bs.priority, bs.weight,
213 		       bs.port, bs.protocol, bs.nr_addrs,
214 		       bs.name_len, bs.name_len, b);
215 
216 		if (end - b < bs.name_len)
217 			break;
218 
219 		ret = -EPROTONOSUPPORT;
220 		if (bs.protocol == DNS_SERVER_PROTOCOL_UNSPECIFIED) {
221 			bs.protocol = DNS_SERVER_PROTOCOL_UDP;
222 		} else if (bs.protocol != DNS_SERVER_PROTOCOL_UDP) {
223 			_leave(" = [proto %u]", bs.protocol);
224 			goto error;
225 		}
226 
227 		if (bs.port == 0)
228 			bs.port = AFS_VL_PORT;
229 		if (bs.source > NR__dns_record_source)
230 			bs.source = NR__dns_record_source;
231 		if (bs.status > NR__dns_lookup_status)
232 			bs.status = NR__dns_lookup_status;
233 
234 		/* See if we can update an old server record */
235 		server = NULL;
236 		for (i = 0; i < previous->nr_servers; i++) {
237 			struct afs_vlserver *p = previous->servers[i].server;
238 
239 			if (p->name_len == bs.name_len &&
240 			    p->port == bs.port &&
241 			    strncasecmp(b, p->name, bs.name_len) == 0) {
242 				server = afs_get_vlserver(p);
243 				break;
244 			}
245 		}
246 
247 		if (!server) {
248 			ret = -ENOMEM;
249 			server = afs_alloc_vlserver(b, bs.name_len, bs.port);
250 			if (!server)
251 				goto error;
252 		}
253 
254 		b += bs.name_len;
255 
256 		/* Extract the addresses - note that we can't skip this as we
257 		 * have to advance the payload pointer.
258 		 */
259 		addrs = afs_extract_vl_addrs(cell->net, &b, end, bs.nr_addrs, bs.port);
260 		if (IS_ERR(addrs)) {
261 			ret = PTR_ERR(addrs);
262 			goto error_2;
263 		}
264 
265 		if (vllist->nr_servers >= nr_servers) {
266 			_debug("skip %u >= %u", vllist->nr_servers, nr_servers);
267 			afs_put_addrlist(addrs, afs_alist_trace_put_parse_empty);
268 			afs_put_vlserver(cell->net, server);
269 			continue;
270 		}
271 
272 		addrs->source = bs.source;
273 		addrs->status = bs.status;
274 
275 		if (addrs->nr_addrs == 0) {
276 			afs_put_addrlist(addrs, afs_alist_trace_put_parse_empty);
277 			if (!rcu_access_pointer(server->addresses)) {
278 				afs_put_vlserver(cell->net, server);
279 				continue;
280 			}
281 		} else {
282 			struct afs_addr_list *old = addrs;
283 
284 			write_lock(&server->lock);
285 			old = rcu_replace_pointer(server->addresses, old,
286 						  lockdep_is_held(&server->lock));
287 			write_unlock(&server->lock);
288 			afs_put_addrlist(old, afs_alist_trace_put_vlserver_old);
289 		}
290 
291 
292 		/* TODO: Might want to check for duplicates */
293 
294 		/* Insertion-sort by priority and weight */
295 		for (j = 0; j < vllist->nr_servers; j++) {
296 			if (bs.priority < vllist->servers[j].priority)
297 				break; /* Lower preferable */
298 			if (bs.priority == vllist->servers[j].priority &&
299 			    bs.weight > vllist->servers[j].weight)
300 				break; /* Higher preferable */
301 		}
302 
303 		if (j < vllist->nr_servers) {
304 			memmove(vllist->servers + j + 1,
305 				vllist->servers + j,
306 				(vllist->nr_servers - j) * sizeof(struct afs_vlserver_entry));
307 		}
308 
309 		clear_bit(AFS_VLSERVER_FL_PROBED, &server->flags);
310 
311 		vllist->servers[j].priority = bs.priority;
312 		vllist->servers[j].weight = bs.weight;
313 		vllist->servers[j].server = server;
314 		vllist->nr_servers++;
315 	}
316 
317 	if (b != end) {
318 		_debug("parse error %zd", b - end);
319 		goto error;
320 	}
321 
322 	afs_put_vlserverlist(cell->net, previous);
323 	_leave(" = ok [%u]", vllist->nr_servers);
324 	return vllist;
325 
326 error_2:
327 	afs_put_vlserver(cell->net, server);
328 error:
329 	afs_put_vlserverlist(cell->net, vllist);
330 	afs_put_vlserverlist(cell->net, previous);
331 dump:
332 	if (ret != -ENOMEM) {
333 		printk(KERN_DEBUG "DNS: at %zu\n", (const void *)b - buffer);
334 		print_hex_dump_bytes("DNS: ", DUMP_PREFIX_NONE, buffer, buffer_size);
335 	}
336 	return ERR_PTR(ret);
337 }
338