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 | |
17 | vlserver = kzalloc(struct_size(vlserver, name, name_len + 1), |
18 | GFP_KERNEL); |
19 | if (vlserver) { |
20 | refcount_set(r: &vlserver->ref, n: 1); |
21 | rwlock_init(&vlserver->lock); |
22 | init_waitqueue_head(&vlserver->probe_wq); |
23 | spin_lock_init(&vlserver->probe_lock); |
24 | vlserver->rtt = UINT_MAX; |
25 | vlserver->name_len = name_len; |
26 | vlserver->port = port; |
27 | memcpy(vlserver->name, name, name_len); |
28 | } |
29 | return vlserver; |
30 | } |
31 | |
32 | static void afs_vlserver_rcu(struct rcu_head *rcu) |
33 | { |
34 | struct afs_vlserver *vlserver = container_of(rcu, struct afs_vlserver, rcu); |
35 | |
36 | afs_put_addrlist(rcu_access_pointer(vlserver->addresses)); |
37 | kfree_rcu(vlserver, rcu); |
38 | } |
39 | |
40 | void afs_put_vlserver(struct afs_net *net, struct afs_vlserver *vlserver) |
41 | { |
42 | if (vlserver && |
43 | refcount_dec_and_test(r: &vlserver->ref)) |
44 | call_rcu(head: &vlserver->rcu, func: afs_vlserver_rcu); |
45 | } |
46 | |
47 | struct afs_vlserver_list *afs_alloc_vlserver_list(unsigned int nr_servers) |
48 | { |
49 | struct afs_vlserver_list *vllist; |
50 | |
51 | vllist = kzalloc(struct_size(vllist, servers, nr_servers), GFP_KERNEL); |
52 | if (vllist) { |
53 | refcount_set(r: &vllist->ref, n: 1); |
54 | rwlock_init(&vllist->lock); |
55 | } |
56 | |
57 | return vllist; |
58 | } |
59 | |
60 | void afs_put_vlserverlist(struct afs_net *net, struct afs_vlserver_list *vllist) |
61 | { |
62 | if (vllist) { |
63 | if (refcount_dec_and_test(r: &vllist->ref)) { |
64 | int i; |
65 | |
66 | for (i = 0; i < vllist->nr_servers; i++) { |
67 | afs_put_vlserver(net, vlserver: vllist->servers[i].server); |
68 | } |
69 | kfree_rcu(vllist, rcu); |
70 | } |
71 | } |
72 | } |
73 | |
74 | static u16 (const u8 **_b) |
75 | { |
76 | u16 val; |
77 | |
78 | val = (u16)*(*_b)++ << 0; |
79 | val |= (u16)*(*_b)++ << 8; |
80 | return val; |
81 | } |
82 | |
83 | /* |
84 | * Build a VL server address list from a DNS queried server list. |
85 | */ |
86 | static struct afs_addr_list *(const u8 **_b, const u8 *end, |
87 | u8 nr_addrs, u16 port) |
88 | { |
89 | struct afs_addr_list *alist; |
90 | const u8 *b = *_b; |
91 | int ret = -EINVAL; |
92 | |
93 | alist = afs_alloc_addrlist(nr_addrs, VL_SERVICE, port); |
94 | if (!alist) |
95 | return ERR_PTR(error: -ENOMEM); |
96 | if (nr_addrs == 0) |
97 | return alist; |
98 | |
99 | for (; nr_addrs > 0 && end - b >= nr_addrs; nr_addrs--) { |
100 | struct dns_server_list_v1_address hdr; |
101 | __be32 x[4]; |
102 | |
103 | hdr.address_type = *b++; |
104 | |
105 | switch (hdr.address_type) { |
106 | case DNS_ADDRESS_IS_IPV4: |
107 | if (end - b < 4) { |
108 | _leave(" = -EINVAL [short inet]" ); |
109 | goto error; |
110 | } |
111 | memcpy(x, b, 4); |
112 | afs_merge_fs_addr4(alist, x[0], port); |
113 | b += 4; |
114 | break; |
115 | |
116 | case DNS_ADDRESS_IS_IPV6: |
117 | if (end - b < 16) { |
118 | _leave(" = -EINVAL [short inet6]" ); |
119 | goto error; |
120 | } |
121 | memcpy(x, b, 16); |
122 | afs_merge_fs_addr6(alist, x, port); |
123 | b += 16; |
124 | break; |
125 | |
126 | default: |
127 | _leave(" = -EADDRNOTAVAIL [unknown af %u]" , |
128 | hdr.address_type); |
129 | ret = -EADDRNOTAVAIL; |
130 | goto error; |
131 | } |
132 | } |
133 | |
134 | /* Start with IPv6 if available. */ |
135 | if (alist->nr_ipv4 < alist->nr_addrs) |
136 | alist->preferred = alist->nr_ipv4; |
137 | |
138 | *_b = b; |
139 | return alist; |
140 | |
141 | error: |
142 | *_b = b; |
143 | afs_put_addrlist(alist); |
144 | return ERR_PTR(error: ret); |
145 | } |
146 | |
147 | /* |
148 | * Build a VL server list from a DNS queried server list. |
149 | */ |
150 | struct afs_vlserver_list *(struct afs_cell *cell, |
151 | const void *buffer, |
152 | size_t buffer_size) |
153 | { |
154 | const struct dns_server_list_v1_header *hdr = buffer; |
155 | struct dns_server_list_v1_server bs; |
156 | struct afs_vlserver_list *vllist, *previous; |
157 | struct afs_addr_list *addrs; |
158 | struct afs_vlserver *server; |
159 | const u8 *b = buffer, *end = buffer + buffer_size; |
160 | int ret = -ENOMEM, nr_servers, i, j; |
161 | |
162 | _enter("" ); |
163 | |
164 | /* Check that it's a server list, v1 */ |
165 | if (end - b < sizeof(*hdr) || |
166 | hdr->hdr.content != DNS_PAYLOAD_IS_SERVER_LIST || |
167 | hdr->hdr.version != 1) { |
168 | pr_notice("kAFS: Got DNS record [%u,%u] len %zu\n" , |
169 | hdr->hdr.content, hdr->hdr.version, end - b); |
170 | ret = -EDESTADDRREQ; |
171 | goto dump; |
172 | } |
173 | |
174 | nr_servers = hdr->nr_servers; |
175 | |
176 | vllist = afs_alloc_vlserver_list(nr_servers); |
177 | if (!vllist) |
178 | return ERR_PTR(error: -ENOMEM); |
179 | |
180 | vllist->source = (hdr->source < NR__dns_record_source) ? |
181 | hdr->source : NR__dns_record_source; |
182 | vllist->status = (hdr->status < NR__dns_lookup_status) ? |
183 | hdr->status : NR__dns_lookup_status; |
184 | |
185 | read_lock(&cell->vl_servers_lock); |
186 | previous = afs_get_vlserverlist( |
187 | rcu_dereference_protected(cell->vl_servers, |
188 | lockdep_is_held(&cell->vl_servers_lock))); |
189 | read_unlock(&cell->vl_servers_lock); |
190 | |
191 | b += sizeof(*hdr); |
192 | while (end - b >= sizeof(bs)) { |
193 | bs.name_len = afs_extract_le16(b: &b); |
194 | bs.priority = afs_extract_le16(b: &b); |
195 | bs.weight = afs_extract_le16(b: &b); |
196 | bs.port = afs_extract_le16(b: &b); |
197 | bs.source = *b++; |
198 | bs.status = *b++; |
199 | bs.protocol = *b++; |
200 | bs.nr_addrs = *b++; |
201 | |
202 | _debug("extract %u %u %u %u %u %u %*.*s" , |
203 | bs.name_len, bs.priority, bs.weight, |
204 | bs.port, bs.protocol, bs.nr_addrs, |
205 | bs.name_len, bs.name_len, b); |
206 | |
207 | if (end - b < bs.name_len) |
208 | break; |
209 | |
210 | ret = -EPROTONOSUPPORT; |
211 | if (bs.protocol == DNS_SERVER_PROTOCOL_UNSPECIFIED) { |
212 | bs.protocol = DNS_SERVER_PROTOCOL_UDP; |
213 | } else if (bs.protocol != DNS_SERVER_PROTOCOL_UDP) { |
214 | _leave(" = [proto %u]" , bs.protocol); |
215 | goto error; |
216 | } |
217 | |
218 | if (bs.port == 0) |
219 | bs.port = AFS_VL_PORT; |
220 | if (bs.source > NR__dns_record_source) |
221 | bs.source = NR__dns_record_source; |
222 | if (bs.status > NR__dns_lookup_status) |
223 | bs.status = NR__dns_lookup_status; |
224 | |
225 | /* See if we can update an old server record */ |
226 | server = NULL; |
227 | for (i = 0; i < previous->nr_servers; i++) { |
228 | struct afs_vlserver *p = previous->servers[i].server; |
229 | |
230 | if (p->name_len == bs.name_len && |
231 | p->port == bs.port && |
232 | strncasecmp(s1: b, s2: p->name, n: bs.name_len) == 0) { |
233 | server = afs_get_vlserver(vlserver: p); |
234 | break; |
235 | } |
236 | } |
237 | |
238 | if (!server) { |
239 | ret = -ENOMEM; |
240 | server = afs_alloc_vlserver(name: b, name_len: bs.name_len, port: bs.port); |
241 | if (!server) |
242 | goto error; |
243 | } |
244 | |
245 | b += bs.name_len; |
246 | |
247 | /* Extract the addresses - note that we can't skip this as we |
248 | * have to advance the payload pointer. |
249 | */ |
250 | addrs = afs_extract_vl_addrs(b: &b, end, nr_addrs: bs.nr_addrs, port: bs.port); |
251 | if (IS_ERR(ptr: addrs)) { |
252 | ret = PTR_ERR(ptr: addrs); |
253 | goto error_2; |
254 | } |
255 | |
256 | if (vllist->nr_servers >= nr_servers) { |
257 | _debug("skip %u >= %u" , vllist->nr_servers, nr_servers); |
258 | afs_put_addrlist(addrs); |
259 | afs_put_vlserver(net: cell->net, vlserver: server); |
260 | continue; |
261 | } |
262 | |
263 | addrs->source = bs.source; |
264 | addrs->status = bs.status; |
265 | |
266 | if (addrs->nr_addrs == 0) { |
267 | afs_put_addrlist(addrs); |
268 | if (!rcu_access_pointer(server->addresses)) { |
269 | afs_put_vlserver(net: cell->net, vlserver: server); |
270 | continue; |
271 | } |
272 | } else { |
273 | struct afs_addr_list *old = addrs; |
274 | |
275 | write_lock(&server->lock); |
276 | old = rcu_replace_pointer(server->addresses, old, |
277 | lockdep_is_held(&server->lock)); |
278 | write_unlock(&server->lock); |
279 | afs_put_addrlist(old); |
280 | } |
281 | |
282 | |
283 | /* TODO: Might want to check for duplicates */ |
284 | |
285 | /* Insertion-sort by priority and weight */ |
286 | for (j = 0; j < vllist->nr_servers; j++) { |
287 | if (bs.priority < vllist->servers[j].priority) |
288 | break; /* Lower preferable */ |
289 | if (bs.priority == vllist->servers[j].priority && |
290 | bs.weight > vllist->servers[j].weight) |
291 | break; /* Higher preferable */ |
292 | } |
293 | |
294 | if (j < vllist->nr_servers) { |
295 | memmove(vllist->servers + j + 1, |
296 | vllist->servers + j, |
297 | (vllist->nr_servers - j) * sizeof(struct afs_vlserver_entry)); |
298 | } |
299 | |
300 | clear_bit(AFS_VLSERVER_FL_PROBED, addr: &server->flags); |
301 | |
302 | vllist->servers[j].priority = bs.priority; |
303 | vllist->servers[j].weight = bs.weight; |
304 | vllist->servers[j].server = server; |
305 | vllist->nr_servers++; |
306 | } |
307 | |
308 | if (b != end) { |
309 | _debug("parse error %zd" , b - end); |
310 | goto error; |
311 | } |
312 | |
313 | afs_put_vlserverlist(net: cell->net, vllist: previous); |
314 | _leave(" = ok [%u]" , vllist->nr_servers); |
315 | return vllist; |
316 | |
317 | error_2: |
318 | afs_put_vlserver(net: cell->net, vlserver: server); |
319 | error: |
320 | afs_put_vlserverlist(net: cell->net, vllist); |
321 | afs_put_vlserverlist(net: cell->net, vllist: previous); |
322 | dump: |
323 | if (ret != -ENOMEM) { |
324 | printk(KERN_DEBUG "DNS: at %zu\n" , (const void *)b - buffer); |
325 | print_hex_dump_bytes("DNS: " , DUMP_PREFIX_NONE, buffer, buffer_size); |
326 | } |
327 | return ERR_PTR(error: ret); |
328 | } |
329 | |