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
12struct 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
35static 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
44void 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
51struct 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
64void 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
78static 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 */
90static 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
150error:
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 */
159struct 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
326error_2:
327	afs_put_vlserver(cell->net, server);
328error:
329	afs_put_vlserverlist(cell->net, vllist);
330	afs_put_vlserverlist(cell->net, previous);
331dump:
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