1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Handle vlserver selection and rotation.
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/sched.h>
10 #include <linux/sched/signal.h>
11 #include "internal.h"
12 #include "afs_vl.h"
13
14 /*
15 * Begin an operation on a volume location server.
16 */
afs_begin_vlserver_operation(struct afs_vl_cursor * vc,struct afs_cell * cell,struct key * key)17 bool afs_begin_vlserver_operation(struct afs_vl_cursor *vc, struct afs_cell *cell,
18 struct key *key)
19 {
20 memset(vc, 0, sizeof(*vc));
21 vc->cell = cell;
22 vc->key = key;
23 vc->error = -EDESTADDRREQ;
24 vc->ac.error = SHRT_MAX;
25
26 if (signal_pending(current)) {
27 vc->error = -EINTR;
28 vc->flags |= AFS_VL_CURSOR_STOP;
29 return false;
30 }
31
32 return true;
33 }
34
35 /*
36 * Begin iteration through a server list, starting with the last used server if
37 * possible, or the last recorded good server if not.
38 */
afs_start_vl_iteration(struct afs_vl_cursor * vc)39 static bool afs_start_vl_iteration(struct afs_vl_cursor *vc)
40 {
41 struct afs_cell *cell = vc->cell;
42 unsigned int dns_lookup_count;
43
44 if (cell->dns_source == DNS_RECORD_UNAVAILABLE ||
45 cell->dns_expiry <= ktime_get_real_seconds()) {
46 dns_lookup_count = smp_load_acquire(&cell->dns_lookup_count);
47 set_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags);
48 afs_queue_cell(cell, afs_cell_trace_get_queue_dns);
49
50 if (cell->dns_source == DNS_RECORD_UNAVAILABLE) {
51 if (wait_var_event_interruptible(
52 &cell->dns_lookup_count,
53 smp_load_acquire(&cell->dns_lookup_count)
54 != dns_lookup_count) < 0) {
55 vc->error = -ERESTARTSYS;
56 return false;
57 }
58 }
59
60 /* Status load is ordered after lookup counter load */
61 if (cell->dns_source == DNS_RECORD_UNAVAILABLE) {
62 vc->error = -EDESTADDRREQ;
63 return false;
64 }
65 }
66
67 read_lock(&cell->vl_servers_lock);
68 vc->server_list = afs_get_vlserverlist(
69 rcu_dereference_protected(cell->vl_servers,
70 lockdep_is_held(&cell->vl_servers_lock)));
71 read_unlock(&cell->vl_servers_lock);
72 if (!vc->server_list->nr_servers)
73 return false;
74
75 vc->untried = (1UL << vc->server_list->nr_servers) - 1;
76 vc->index = -1;
77 return true;
78 }
79
80 /*
81 * Select the vlserver to use. May be called multiple times to rotate
82 * through the vlservers.
83 */
afs_select_vlserver(struct afs_vl_cursor * vc)84 bool afs_select_vlserver(struct afs_vl_cursor *vc)
85 {
86 struct afs_addr_list *alist;
87 struct afs_vlserver *vlserver;
88 struct afs_error e;
89 u32 rtt;
90 int error = vc->ac.error, i;
91
92 _enter("%lx[%d],%lx[%d],%d,%d",
93 vc->untried, vc->index,
94 vc->ac.tried, vc->ac.index,
95 error, vc->ac.abort_code);
96
97 if (vc->flags & AFS_VL_CURSOR_STOP) {
98 _leave(" = f [stopped]");
99 return false;
100 }
101
102 vc->nr_iterations++;
103
104 /* Evaluate the result of the previous operation, if there was one. */
105 switch (error) {
106 case SHRT_MAX:
107 goto start;
108
109 default:
110 case 0:
111 /* Success or local failure. Stop. */
112 vc->error = error;
113 vc->flags |= AFS_VL_CURSOR_STOP;
114 _leave(" = f [okay/local %d]", vc->ac.error);
115 return false;
116
117 case -ECONNABORTED:
118 /* The far side rejected the operation on some grounds. This
119 * might involve the server being busy or the volume having been moved.
120 */
121 switch (vc->ac.abort_code) {
122 case AFSVL_IO:
123 case AFSVL_BADVOLOPER:
124 case AFSVL_NOMEM:
125 /* The server went weird. */
126 vc->error = -EREMOTEIO;
127 //write_lock(&vc->cell->vl_servers_lock);
128 //vc->server_list->weird_mask |= 1 << vc->index;
129 //write_unlock(&vc->cell->vl_servers_lock);
130 goto next_server;
131
132 default:
133 vc->error = afs_abort_to_error(vc->ac.abort_code);
134 goto failed;
135 }
136
137 case -ERFKILL:
138 case -EADDRNOTAVAIL:
139 case -ENETUNREACH:
140 case -EHOSTUNREACH:
141 case -EHOSTDOWN:
142 case -ECONNREFUSED:
143 case -ETIMEDOUT:
144 case -ETIME:
145 _debug("no conn %d", error);
146 vc->error = error;
147 goto iterate_address;
148
149 case -ECONNRESET:
150 _debug("call reset");
151 vc->error = error;
152 vc->flags |= AFS_VL_CURSOR_RETRY;
153 goto next_server;
154
155 case -EOPNOTSUPP:
156 _debug("notsupp");
157 goto next_server;
158 }
159
160 restart_from_beginning:
161 _debug("restart");
162 afs_end_cursor(&vc->ac);
163 afs_put_vlserverlist(vc->cell->net, vc->server_list);
164 vc->server_list = NULL;
165 if (vc->flags & AFS_VL_CURSOR_RETRIED)
166 goto failed;
167 vc->flags |= AFS_VL_CURSOR_RETRIED;
168 start:
169 _debug("start");
170
171 if (!afs_start_vl_iteration(vc))
172 goto failed;
173
174 error = afs_send_vl_probes(vc->cell->net, vc->key, vc->server_list);
175 if (error < 0)
176 goto failed_set_error;
177
178 pick_server:
179 _debug("pick [%lx]", vc->untried);
180
181 error = afs_wait_for_vl_probes(vc->server_list, vc->untried);
182 if (error < 0)
183 goto failed_set_error;
184
185 /* Pick the untried server with the lowest RTT. */
186 vc->index = vc->server_list->preferred;
187 if (test_bit(vc->index, &vc->untried))
188 goto selected_server;
189
190 vc->index = -1;
191 rtt = U32_MAX;
192 for (i = 0; i < vc->server_list->nr_servers; i++) {
193 struct afs_vlserver *s = vc->server_list->servers[i].server;
194
195 if (!test_bit(i, &vc->untried) ||
196 !test_bit(AFS_VLSERVER_FL_RESPONDING, &s->flags))
197 continue;
198 if (s->probe.rtt < rtt) {
199 vc->index = i;
200 rtt = s->probe.rtt;
201 }
202 }
203
204 if (vc->index == -1)
205 goto no_more_servers;
206
207 selected_server:
208 _debug("use %d", vc->index);
209 __clear_bit(vc->index, &vc->untried);
210
211 /* We're starting on a different vlserver from the list. We need to
212 * check it, find its address list and probe its capabilities before we
213 * use it.
214 */
215 ASSERTCMP(vc->ac.alist, ==, NULL);
216 vlserver = vc->server_list->servers[vc->index].server;
217 vc->server = vlserver;
218
219 _debug("USING VLSERVER: %s", vlserver->name);
220
221 read_lock(&vlserver->lock);
222 alist = rcu_dereference_protected(vlserver->addresses,
223 lockdep_is_held(&vlserver->lock));
224 afs_get_addrlist(alist);
225 read_unlock(&vlserver->lock);
226
227 memset(&vc->ac, 0, sizeof(vc->ac));
228
229 if (!vc->ac.alist)
230 vc->ac.alist = alist;
231 else
232 afs_put_addrlist(alist);
233
234 vc->ac.index = -1;
235
236 iterate_address:
237 ASSERT(vc->ac.alist);
238 /* Iterate over the current server's address list to try and find an
239 * address on which it will respond to us.
240 */
241 if (!afs_iterate_addresses(&vc->ac))
242 goto next_server;
243
244 _debug("VL address %d/%d", vc->ac.index, vc->ac.alist->nr_addrs);
245
246 _leave(" = t %pISpc", &vc->ac.alist->addrs[vc->ac.index].transport);
247 return true;
248
249 next_server:
250 _debug("next");
251 afs_end_cursor(&vc->ac);
252 goto pick_server;
253
254 no_more_servers:
255 /* That's all the servers poked to no good effect. Try again if some
256 * of them were busy.
257 */
258 if (vc->flags & AFS_VL_CURSOR_RETRY)
259 goto restart_from_beginning;
260
261 e.error = -EDESTADDRREQ;
262 e.responded = false;
263 for (i = 0; i < vc->server_list->nr_servers; i++) {
264 struct afs_vlserver *s = vc->server_list->servers[i].server;
265
266 if (test_bit(AFS_VLSERVER_FL_RESPONDING, &s->flags))
267 e.responded = true;
268 afs_prioritise_error(&e, READ_ONCE(s->probe.error),
269 s->probe.abort_code);
270 }
271
272 error = e.error;
273
274 failed_set_error:
275 vc->error = error;
276 failed:
277 vc->flags |= AFS_VL_CURSOR_STOP;
278 afs_end_cursor(&vc->ac);
279 _leave(" = f [failed %d]", vc->error);
280 return false;
281 }
282
283 /*
284 * Dump cursor state in the case of the error being EDESTADDRREQ.
285 */
afs_vl_dump_edestaddrreq(const struct afs_vl_cursor * vc)286 static void afs_vl_dump_edestaddrreq(const struct afs_vl_cursor *vc)
287 {
288 static int count;
289 int i;
290
291 if (!IS_ENABLED(CONFIG_AFS_DEBUG_CURSOR) || count > 3)
292 return;
293 count++;
294
295 rcu_read_lock();
296 pr_notice("EDESTADDR occurred\n");
297 pr_notice("VC: ut=%lx ix=%u ni=%hu fl=%hx err=%hd\n",
298 vc->untried, vc->index, vc->nr_iterations, vc->flags, vc->error);
299
300 if (vc->server_list) {
301 const struct afs_vlserver_list *sl = vc->server_list;
302 pr_notice("VC: SL nr=%u ix=%u\n",
303 sl->nr_servers, sl->index);
304 for (i = 0; i < sl->nr_servers; i++) {
305 const struct afs_vlserver *s = sl->servers[i].server;
306 pr_notice("VC: server %s+%hu fl=%lx E=%hd\n",
307 s->name, s->port, s->flags, s->probe.error);
308 if (s->addresses) {
309 const struct afs_addr_list *a =
310 rcu_dereference(s->addresses);
311 pr_notice("VC: - nr=%u/%u/%u pf=%u\n",
312 a->nr_ipv4, a->nr_addrs, a->max_addrs,
313 a->preferred);
314 pr_notice("VC: - R=%lx F=%lx\n",
315 a->responded, a->failed);
316 if (a == vc->ac.alist)
317 pr_notice("VC: - current\n");
318 }
319 }
320 }
321
322 pr_notice("AC: t=%lx ax=%u ac=%d er=%d r=%u ni=%u\n",
323 vc->ac.tried, vc->ac.index, vc->ac.abort_code, vc->ac.error,
324 vc->ac.responded, vc->ac.nr_iterations);
325 rcu_read_unlock();
326 }
327
328 /*
329 * Tidy up a volume location server cursor and unlock the vnode.
330 */
afs_end_vlserver_operation(struct afs_vl_cursor * vc)331 int afs_end_vlserver_operation(struct afs_vl_cursor *vc)
332 {
333 struct afs_net *net = vc->cell->net;
334
335 if (vc->error == -EDESTADDRREQ ||
336 vc->error == -EADDRNOTAVAIL ||
337 vc->error == -ENETUNREACH ||
338 vc->error == -EHOSTUNREACH)
339 afs_vl_dump_edestaddrreq(vc);
340
341 afs_end_cursor(&vc->ac);
342 afs_put_vlserverlist(net, vc->server_list);
343
344 if (vc->error == -ECONNABORTED)
345 vc->error = afs_abort_to_error(vc->ac.abort_code);
346
347 return vc->error;
348 }
349