Loading...
1/* AFS server record management
2 *
3 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#include <linux/sched.h>
13#include <linux/slab.h>
14#include "internal.h"
15
16static unsigned afs_server_timeout = 10; /* server timeout in seconds */
17
18static void afs_reap_server(struct work_struct *);
19
20/* tree of all the servers, indexed by IP address */
21static struct rb_root afs_servers = RB_ROOT;
22static DEFINE_RWLOCK(afs_servers_lock);
23
24/* LRU list of all the servers not currently in use */
25static LIST_HEAD(afs_server_graveyard);
26static DEFINE_SPINLOCK(afs_server_graveyard_lock);
27static DECLARE_DELAYED_WORK(afs_server_reaper, afs_reap_server);
28
29/*
30 * install a server record in the master tree
31 */
32static int afs_install_server(struct afs_server *server)
33{
34 struct afs_server *xserver;
35 struct rb_node **pp, *p;
36 int ret;
37
38 _enter("%p", server);
39
40 write_lock(&afs_servers_lock);
41
42 ret = -EEXIST;
43 pp = &afs_servers.rb_node;
44 p = NULL;
45 while (*pp) {
46 p = *pp;
47 _debug("- consider %p", p);
48 xserver = rb_entry(p, struct afs_server, master_rb);
49 if (server->addr.s_addr < xserver->addr.s_addr)
50 pp = &(*pp)->rb_left;
51 else if (server->addr.s_addr > xserver->addr.s_addr)
52 pp = &(*pp)->rb_right;
53 else
54 goto error;
55 }
56
57 rb_link_node(&server->master_rb, p, pp);
58 rb_insert_color(&server->master_rb, &afs_servers);
59 ret = 0;
60
61error:
62 write_unlock(&afs_servers_lock);
63 return ret;
64}
65
66/*
67 * allocate a new server record
68 */
69static struct afs_server *afs_alloc_server(struct afs_cell *cell,
70 const struct in_addr *addr)
71{
72 struct afs_server *server;
73
74 _enter("");
75
76 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
77 if (server) {
78 atomic_set(&server->usage, 1);
79 server->cell = cell;
80
81 INIT_LIST_HEAD(&server->link);
82 INIT_LIST_HEAD(&server->grave);
83 init_rwsem(&server->sem);
84 spin_lock_init(&server->fs_lock);
85 server->fs_vnodes = RB_ROOT;
86 server->cb_promises = RB_ROOT;
87 spin_lock_init(&server->cb_lock);
88 init_waitqueue_head(&server->cb_break_waitq);
89 INIT_DELAYED_WORK(&server->cb_break_work,
90 afs_dispatch_give_up_callbacks);
91
92 memcpy(&server->addr, addr, sizeof(struct in_addr));
93 server->addr.s_addr = addr->s_addr;
94 _leave(" = %p{%d}", server, atomic_read(&server->usage));
95 } else {
96 _leave(" = NULL [nomem]");
97 }
98 return server;
99}
100
101/*
102 * get an FS-server record for a cell
103 */
104struct afs_server *afs_lookup_server(struct afs_cell *cell,
105 const struct in_addr *addr)
106{
107 struct afs_server *server, *candidate;
108
109 _enter("%p,%pI4", cell, &addr->s_addr);
110
111 /* quick scan of the list to see if we already have the server */
112 read_lock(&cell->servers_lock);
113
114 list_for_each_entry(server, &cell->servers, link) {
115 if (server->addr.s_addr == addr->s_addr)
116 goto found_server_quickly;
117 }
118 read_unlock(&cell->servers_lock);
119
120 candidate = afs_alloc_server(cell, addr);
121 if (!candidate) {
122 _leave(" = -ENOMEM");
123 return ERR_PTR(-ENOMEM);
124 }
125
126 write_lock(&cell->servers_lock);
127
128 /* check the cell's server list again */
129 list_for_each_entry(server, &cell->servers, link) {
130 if (server->addr.s_addr == addr->s_addr)
131 goto found_server;
132 }
133
134 _debug("new");
135 server = candidate;
136 if (afs_install_server(server) < 0)
137 goto server_in_two_cells;
138
139 afs_get_cell(cell);
140 list_add_tail(&server->link, &cell->servers);
141
142 write_unlock(&cell->servers_lock);
143 _leave(" = %p{%d}", server, atomic_read(&server->usage));
144 return server;
145
146 /* found a matching server quickly */
147found_server_quickly:
148 _debug("found quickly");
149 afs_get_server(server);
150 read_unlock(&cell->servers_lock);
151no_longer_unused:
152 if (!list_empty(&server->grave)) {
153 spin_lock(&afs_server_graveyard_lock);
154 list_del_init(&server->grave);
155 spin_unlock(&afs_server_graveyard_lock);
156 }
157 _leave(" = %p{%d}", server, atomic_read(&server->usage));
158 return server;
159
160 /* found a matching server on the second pass */
161found_server:
162 _debug("found");
163 afs_get_server(server);
164 write_unlock(&cell->servers_lock);
165 kfree(candidate);
166 goto no_longer_unused;
167
168 /* found a server that seems to be in two cells */
169server_in_two_cells:
170 write_unlock(&cell->servers_lock);
171 kfree(candidate);
172 printk(KERN_NOTICE "kAFS: Server %pI4 appears to be in two cells\n",
173 addr);
174 _leave(" = -EEXIST");
175 return ERR_PTR(-EEXIST);
176}
177
178/*
179 * look up a server by its IP address
180 */
181struct afs_server *afs_find_server(const struct sockaddr_rxrpc *srx)
182{
183 struct afs_server *server = NULL;
184 struct rb_node *p;
185 struct in_addr addr = srx->transport.sin.sin_addr;
186
187 _enter("{%d,%pI4}", srx->transport.family, &addr.s_addr);
188
189 if (srx->transport.family != AF_INET) {
190 WARN(true, "AFS does not yes support non-IPv4 addresses\n");
191 return NULL;
192 }
193
194 read_lock(&afs_servers_lock);
195
196 p = afs_servers.rb_node;
197 while (p) {
198 server = rb_entry(p, struct afs_server, master_rb);
199
200 _debug("- consider %p", p);
201
202 if (addr.s_addr < server->addr.s_addr) {
203 p = p->rb_left;
204 } else if (addr.s_addr > server->addr.s_addr) {
205 p = p->rb_right;
206 } else {
207 afs_get_server(server);
208 goto found;
209 }
210 }
211
212 server = NULL;
213found:
214 read_unlock(&afs_servers_lock);
215 ASSERTIFCMP(server, server->addr.s_addr, ==, addr.s_addr);
216 _leave(" = %p", server);
217 return server;
218}
219
220/*
221 * destroy a server record
222 * - removes from the cell list
223 */
224void afs_put_server(struct afs_server *server)
225{
226 if (!server)
227 return;
228
229 _enter("%p{%d}", server, atomic_read(&server->usage));
230
231 _debug("PUT SERVER %d", atomic_read(&server->usage));
232
233 ASSERTCMP(atomic_read(&server->usage), >, 0);
234
235 if (likely(!atomic_dec_and_test(&server->usage))) {
236 _leave("");
237 return;
238 }
239
240 afs_flush_callback_breaks(server);
241
242 spin_lock(&afs_server_graveyard_lock);
243 if (atomic_read(&server->usage) == 0) {
244 list_move_tail(&server->grave, &afs_server_graveyard);
245 server->time_of_death = get_seconds();
246 queue_delayed_work(afs_wq, &afs_server_reaper,
247 afs_server_timeout * HZ);
248 }
249 spin_unlock(&afs_server_graveyard_lock);
250 _leave(" [dead]");
251}
252
253/*
254 * destroy a dead server
255 */
256static void afs_destroy_server(struct afs_server *server)
257{
258 _enter("%p", server);
259
260 ASSERTIF(server->cb_break_head != server->cb_break_tail,
261 delayed_work_pending(&server->cb_break_work));
262
263 ASSERTCMP(server->fs_vnodes.rb_node, ==, NULL);
264 ASSERTCMP(server->cb_promises.rb_node, ==, NULL);
265 ASSERTCMP(server->cb_break_head, ==, server->cb_break_tail);
266 ASSERTCMP(atomic_read(&server->cb_break_n), ==, 0);
267
268 afs_put_cell(server->cell);
269 kfree(server);
270}
271
272/*
273 * reap dead server records
274 */
275static void afs_reap_server(struct work_struct *work)
276{
277 LIST_HEAD(corpses);
278 struct afs_server *server;
279 unsigned long delay, expiry;
280 time_t now;
281
282 now = get_seconds();
283 spin_lock(&afs_server_graveyard_lock);
284
285 while (!list_empty(&afs_server_graveyard)) {
286 server = list_entry(afs_server_graveyard.next,
287 struct afs_server, grave);
288
289 /* the queue is ordered most dead first */
290 expiry = server->time_of_death + afs_server_timeout;
291 if (expiry > now) {
292 delay = (expiry - now) * HZ;
293 mod_delayed_work(afs_wq, &afs_server_reaper, delay);
294 break;
295 }
296
297 write_lock(&server->cell->servers_lock);
298 write_lock(&afs_servers_lock);
299 if (atomic_read(&server->usage) > 0) {
300 list_del_init(&server->grave);
301 } else {
302 list_move_tail(&server->grave, &corpses);
303 list_del_init(&server->link);
304 rb_erase(&server->master_rb, &afs_servers);
305 }
306 write_unlock(&afs_servers_lock);
307 write_unlock(&server->cell->servers_lock);
308 }
309
310 spin_unlock(&afs_server_graveyard_lock);
311
312 /* now reap the corpses we've extracted */
313 while (!list_empty(&corpses)) {
314 server = list_entry(corpses.next, struct afs_server, grave);
315 list_del(&server->grave);
316 afs_destroy_server(server);
317 }
318}
319
320/*
321 * discard all the server records for rmmod
322 */
323void __exit afs_purge_servers(void)
324{
325 afs_server_timeout = 0;
326 mod_delayed_work(afs_wq, &afs_server_reaper, 0);
327}
1/* AFS server record management
2 *
3 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#include <linux/sched.h>
13#include <linux/slab.h>
14#include "afs_fs.h"
15#include "internal.h"
16
17static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
18static unsigned afs_server_update_delay = 30; /* Time till VLDB recheck in secs */
19
20static void afs_inc_servers_outstanding(struct afs_net *net)
21{
22 atomic_inc(&net->servers_outstanding);
23}
24
25static void afs_dec_servers_outstanding(struct afs_net *net)
26{
27 if (atomic_dec_and_test(&net->servers_outstanding))
28 wake_up_var(&net->servers_outstanding);
29}
30
31/*
32 * Find a server by one of its addresses.
33 */
34struct afs_server *afs_find_server(struct afs_net *net,
35 const struct sockaddr_rxrpc *srx)
36{
37 const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
38 const struct afs_addr_list *alist;
39 struct afs_server *server = NULL;
40 unsigned int i;
41 bool ipv6 = true;
42 int seq = 0, diff;
43
44 if (srx->transport.sin6.sin6_addr.s6_addr32[0] == 0 ||
45 srx->transport.sin6.sin6_addr.s6_addr32[1] == 0 ||
46 srx->transport.sin6.sin6_addr.s6_addr32[2] == htonl(0xffff))
47 ipv6 = false;
48
49 rcu_read_lock();
50
51 do {
52 if (server)
53 afs_put_server(net, server);
54 server = NULL;
55 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
56
57 if (ipv6) {
58 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
59 alist = rcu_dereference(server->addresses);
60 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
61 b = &alist->addrs[i].transport.sin6;
62 diff = ((u16 __force)a->sin6_port -
63 (u16 __force)b->sin6_port);
64 if (diff == 0)
65 diff = memcmp(&a->sin6_addr,
66 &b->sin6_addr,
67 sizeof(struct in6_addr));
68 if (diff == 0)
69 goto found;
70 }
71 }
72 } else {
73 hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
74 alist = rcu_dereference(server->addresses);
75 for (i = 0; i < alist->nr_ipv4; i++) {
76 b = &alist->addrs[i].transport.sin6;
77 diff = ((u16 __force)a->sin6_port -
78 (u16 __force)b->sin6_port);
79 if (diff == 0)
80 diff = ((u32 __force)a->sin6_addr.s6_addr32[3] -
81 (u32 __force)b->sin6_addr.s6_addr32[3]);
82 if (diff == 0)
83 goto found;
84 }
85 }
86 }
87
88 server = NULL;
89 found:
90 if (server && !atomic_inc_not_zero(&server->usage))
91 server = NULL;
92
93 } while (need_seqretry(&net->fs_addr_lock, seq));
94
95 done_seqretry(&net->fs_addr_lock, seq);
96
97 rcu_read_unlock();
98 return server;
99}
100
101/*
102 * Look up a server by its UUID
103 */
104struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
105{
106 struct afs_server *server = NULL;
107 struct rb_node *p;
108 int diff, seq = 0;
109
110 _enter("%pU", uuid);
111
112 do {
113 /* Unfortunately, rbtree walking doesn't give reliable results
114 * under just the RCU read lock, so we have to check for
115 * changes.
116 */
117 if (server)
118 afs_put_server(net, server);
119 server = NULL;
120
121 read_seqbegin_or_lock(&net->fs_lock, &seq);
122
123 p = net->fs_servers.rb_node;
124 while (p) {
125 server = rb_entry(p, struct afs_server, uuid_rb);
126
127 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
128 if (diff < 0) {
129 p = p->rb_left;
130 } else if (diff > 0) {
131 p = p->rb_right;
132 } else {
133 afs_get_server(server);
134 break;
135 }
136
137 server = NULL;
138 }
139 } while (need_seqretry(&net->fs_lock, seq));
140
141 done_seqretry(&net->fs_lock, seq);
142
143 _leave(" = %p", server);
144 return server;
145}
146
147/*
148 * Install a server record in the namespace tree
149 */
150static struct afs_server *afs_install_server(struct afs_net *net,
151 struct afs_server *candidate)
152{
153 const struct afs_addr_list *alist;
154 struct afs_server *server;
155 struct rb_node **pp, *p;
156 int ret = -EEXIST, diff;
157
158 _enter("%p", candidate);
159
160 write_seqlock(&net->fs_lock);
161
162 /* Firstly install the server in the UUID lookup tree */
163 pp = &net->fs_servers.rb_node;
164 p = NULL;
165 while (*pp) {
166 p = *pp;
167 _debug("- consider %p", p);
168 server = rb_entry(p, struct afs_server, uuid_rb);
169 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
170 if (diff < 0)
171 pp = &(*pp)->rb_left;
172 else if (diff > 0)
173 pp = &(*pp)->rb_right;
174 else
175 goto exists;
176 }
177
178 server = candidate;
179 rb_link_node(&server->uuid_rb, p, pp);
180 rb_insert_color(&server->uuid_rb, &net->fs_servers);
181 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
182
183 write_seqlock(&net->fs_addr_lock);
184 alist = rcu_dereference_protected(server->addresses,
185 lockdep_is_held(&net->fs_addr_lock.lock));
186
187 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
188 * it in the IPv4 and/or IPv6 reverse-map lists.
189 *
190 * TODO: For speed we want to use something other than a flat list
191 * here; even sorting the list in terms of lowest address would help a
192 * bit, but anything we might want to do gets messy and memory
193 * intensive.
194 */
195 if (alist->nr_ipv4 > 0)
196 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
197 if (alist->nr_addrs > alist->nr_ipv4)
198 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
199
200 write_sequnlock(&net->fs_addr_lock);
201 ret = 0;
202
203exists:
204 afs_get_server(server);
205 write_sequnlock(&net->fs_lock);
206 return server;
207}
208
209/*
210 * allocate a new server record
211 */
212static struct afs_server *afs_alloc_server(struct afs_net *net,
213 const uuid_t *uuid,
214 struct afs_addr_list *alist)
215{
216 struct afs_server *server;
217
218 _enter("");
219
220 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
221 if (!server)
222 goto enomem;
223
224 atomic_set(&server->usage, 1);
225 RCU_INIT_POINTER(server->addresses, alist);
226 server->addr_version = alist->version;
227 server->uuid = *uuid;
228 server->flags = (1UL << AFS_SERVER_FL_NEW);
229 server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
230 rwlock_init(&server->fs_lock);
231 INIT_LIST_HEAD(&server->cb_interests);
232 rwlock_init(&server->cb_break_lock);
233
234 afs_inc_servers_outstanding(net);
235 _leave(" = %p", server);
236 return server;
237
238enomem:
239 _leave(" = NULL [nomem]");
240 return NULL;
241}
242
243/*
244 * Look up an address record for a server
245 */
246static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
247 struct key *key, const uuid_t *uuid)
248{
249 struct afs_addr_cursor ac;
250 struct afs_addr_list *alist;
251 int ret;
252
253 ret = afs_set_vl_cursor(&ac, cell);
254 if (ret < 0)
255 return ERR_PTR(ret);
256
257 while (afs_iterate_addresses(&ac)) {
258 if (test_bit(ac.index, &ac.alist->yfs))
259 alist = afs_yfsvl_get_endpoints(cell->net, &ac, key, uuid);
260 else
261 alist = afs_vl_get_addrs_u(cell->net, &ac, key, uuid);
262 switch (ac.error) {
263 case 0:
264 afs_end_cursor(&ac);
265 return alist;
266 case -ECONNABORTED:
267 ac.error = afs_abort_to_error(ac.abort_code);
268 goto error;
269 case -ENOMEM:
270 case -ENONET:
271 goto error;
272 case -ENETUNREACH:
273 case -EHOSTUNREACH:
274 case -ECONNREFUSED:
275 break;
276 default:
277 ac.error = -EIO;
278 goto error;
279 }
280 }
281
282error:
283 return ERR_PTR(afs_end_cursor(&ac));
284}
285
286/*
287 * Get or create a fileserver record.
288 */
289struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
290 const uuid_t *uuid)
291{
292 struct afs_addr_list *alist;
293 struct afs_server *server, *candidate;
294
295 _enter("%p,%pU", cell->net, uuid);
296
297 server = afs_find_server_by_uuid(cell->net, uuid);
298 if (server)
299 return server;
300
301 alist = afs_vl_lookup_addrs(cell, key, uuid);
302 if (IS_ERR(alist))
303 return ERR_CAST(alist);
304
305 candidate = afs_alloc_server(cell->net, uuid, alist);
306 if (!candidate) {
307 afs_put_addrlist(alist);
308 return ERR_PTR(-ENOMEM);
309 }
310
311 server = afs_install_server(cell->net, candidate);
312 if (server != candidate) {
313 afs_put_addrlist(alist);
314 kfree(candidate);
315 }
316
317 _leave(" = %p{%d}", server, atomic_read(&server->usage));
318 return server;
319}
320
321/*
322 * Set the server timer to fire after a given delay, assuming it's not already
323 * set for an earlier time.
324 */
325static void afs_set_server_timer(struct afs_net *net, time64_t delay)
326{
327 if (net->live) {
328 afs_inc_servers_outstanding(net);
329 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
330 afs_dec_servers_outstanding(net);
331 }
332}
333
334/*
335 * Server management timer. We have an increment on fs_outstanding that we
336 * need to pass along to the work item.
337 */
338void afs_servers_timer(struct timer_list *timer)
339{
340 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
341
342 _enter("");
343 if (!queue_work(afs_wq, &net->fs_manager))
344 afs_dec_servers_outstanding(net);
345}
346
347/*
348 * Release a reference on a server record.
349 */
350void afs_put_server(struct afs_net *net, struct afs_server *server)
351{
352 unsigned int usage;
353
354 if (!server)
355 return;
356
357 server->put_time = ktime_get_real_seconds();
358
359 usage = atomic_dec_return(&server->usage);
360
361 _enter("{%u}", usage);
362
363 if (likely(usage > 0))
364 return;
365
366 afs_set_server_timer(net, afs_server_gc_delay);
367}
368
369static void afs_server_rcu(struct rcu_head *rcu)
370{
371 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
372
373 afs_put_addrlist(rcu_access_pointer(server->addresses));
374 kfree(server);
375}
376
377/*
378 * destroy a dead server
379 */
380static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
381{
382 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
383 struct afs_addr_cursor ac = {
384 .alist = alist,
385 .start = alist->index,
386 .index = 0,
387 .addr = &alist->addrs[alist->index],
388 .error = 0,
389 };
390 _enter("%p", server);
391
392 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
393 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
394
395 call_rcu(&server->rcu, afs_server_rcu);
396 afs_dec_servers_outstanding(net);
397}
398
399/*
400 * Garbage collect any expired servers.
401 */
402static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
403{
404 struct afs_server *server;
405 bool deleted;
406 int usage;
407
408 while ((server = gc_list)) {
409 gc_list = server->gc_next;
410
411 write_seqlock(&net->fs_lock);
412 usage = 1;
413 deleted = atomic_try_cmpxchg(&server->usage, &usage, 0);
414 if (deleted) {
415 rb_erase(&server->uuid_rb, &net->fs_servers);
416 hlist_del_rcu(&server->proc_link);
417 }
418 write_sequnlock(&net->fs_lock);
419
420 if (deleted) {
421 write_seqlock(&net->fs_addr_lock);
422 if (!hlist_unhashed(&server->addr4_link))
423 hlist_del_rcu(&server->addr4_link);
424 if (!hlist_unhashed(&server->addr6_link))
425 hlist_del_rcu(&server->addr6_link);
426 write_sequnlock(&net->fs_addr_lock);
427 afs_destroy_server(net, server);
428 }
429 }
430}
431
432/*
433 * Manage the records of servers known to be within a network namespace. This
434 * includes garbage collecting unused servers.
435 *
436 * Note also that we were given an increment on net->servers_outstanding by
437 * whoever queued us that we need to deal with before returning.
438 */
439void afs_manage_servers(struct work_struct *work)
440{
441 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
442 struct afs_server *gc_list = NULL;
443 struct rb_node *cursor;
444 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
445 bool purging = !net->live;
446
447 _enter("");
448
449 /* Trawl the server list looking for servers that have expired from
450 * lack of use.
451 */
452 read_seqlock_excl(&net->fs_lock);
453
454 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
455 struct afs_server *server =
456 rb_entry(cursor, struct afs_server, uuid_rb);
457 int usage = atomic_read(&server->usage);
458
459 _debug("manage %pU %u", &server->uuid, usage);
460
461 ASSERTCMP(usage, >=, 1);
462 ASSERTIFCMP(purging, usage, ==, 1);
463
464 if (usage == 1) {
465 time64_t expire_at = server->put_time;
466
467 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
468 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
469 expire_at += afs_server_gc_delay;
470 if (purging || expire_at <= now) {
471 server->gc_next = gc_list;
472 gc_list = server;
473 } else if (expire_at < next_manage) {
474 next_manage = expire_at;
475 }
476 }
477 }
478
479 read_sequnlock_excl(&net->fs_lock);
480
481 /* Update the timer on the way out. We have to pass an increment on
482 * servers_outstanding in the namespace that we are in to the timer or
483 * the work scheduler.
484 */
485 if (!purging && next_manage < TIME64_MAX) {
486 now = ktime_get_real_seconds();
487
488 if (next_manage - now <= 0) {
489 if (queue_work(afs_wq, &net->fs_manager))
490 afs_inc_servers_outstanding(net);
491 } else {
492 afs_set_server_timer(net, next_manage - now);
493 }
494 }
495
496 afs_gc_servers(net, gc_list);
497
498 afs_dec_servers_outstanding(net);
499 _leave(" [%d]", atomic_read(&net->servers_outstanding));
500}
501
502static void afs_queue_server_manager(struct afs_net *net)
503{
504 afs_inc_servers_outstanding(net);
505 if (!queue_work(afs_wq, &net->fs_manager))
506 afs_dec_servers_outstanding(net);
507}
508
509/*
510 * Purge list of servers.
511 */
512void afs_purge_servers(struct afs_net *net)
513{
514 _enter("");
515
516 if (del_timer_sync(&net->fs_timer))
517 atomic_dec(&net->servers_outstanding);
518
519 afs_queue_server_manager(net);
520
521 _debug("wait");
522 wait_var_event(&net->servers_outstanding,
523 !atomic_read(&net->servers_outstanding));
524 _leave("");
525}
526
527/*
528 * Probe a fileserver to find its capabilities.
529 *
530 * TODO: Try service upgrade.
531 */
532static bool afs_do_probe_fileserver(struct afs_fs_cursor *fc)
533{
534 _enter("");
535
536 fc->ac.addr = NULL;
537 fc->ac.start = READ_ONCE(fc->ac.alist->index);
538 fc->ac.index = fc->ac.start;
539 fc->ac.error = 0;
540 fc->ac.begun = false;
541
542 while (afs_iterate_addresses(&fc->ac)) {
543 afs_fs_get_capabilities(afs_v2net(fc->vnode), fc->cbi->server,
544 &fc->ac, fc->key);
545 switch (fc->ac.error) {
546 case 0:
547 afs_end_cursor(&fc->ac);
548 set_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags);
549 return true;
550 case -ECONNABORTED:
551 fc->ac.error = afs_abort_to_error(fc->ac.abort_code);
552 goto error;
553 case -ENOMEM:
554 case -ENONET:
555 goto error;
556 case -ENETUNREACH:
557 case -EHOSTUNREACH:
558 case -ECONNREFUSED:
559 case -ETIMEDOUT:
560 case -ETIME:
561 break;
562 default:
563 fc->ac.error = -EIO;
564 goto error;
565 }
566 }
567
568error:
569 afs_end_cursor(&fc->ac);
570 return false;
571}
572
573/*
574 * If we haven't already, try probing the fileserver to get its capabilities.
575 * We try not to instigate parallel probes, but it's possible that the parallel
576 * probes will fail due to authentication failure when ours would succeed.
577 *
578 * TODO: Try sending an anonymous probe if an authenticated probe fails.
579 */
580bool afs_probe_fileserver(struct afs_fs_cursor *fc)
581{
582 bool success;
583 int ret, retries = 0;
584
585 _enter("");
586
587retry:
588 if (test_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags)) {
589 _leave(" = t");
590 return true;
591 }
592
593 if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags)) {
594 success = afs_do_probe_fileserver(fc);
595 clear_bit_unlock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags);
596 wake_up_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING);
597 _leave(" = t");
598 return success;
599 }
600
601 _debug("wait");
602 ret = wait_on_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING,
603 TASK_INTERRUPTIBLE);
604 if (ret == -ERESTARTSYS) {
605 fc->ac.error = ret;
606 _leave(" = f [%d]", ret);
607 return false;
608 }
609
610 retries++;
611 if (retries == 4) {
612 fc->ac.error = -ESTALE;
613 _leave(" = f [stale]");
614 return false;
615 }
616 _debug("retry");
617 goto retry;
618}
619
620/*
621 * Get an update for a server's address list.
622 */
623static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
624{
625 struct afs_addr_list *alist, *discard;
626
627 _enter("");
628
629 alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
630 &server->uuid);
631 if (IS_ERR(alist)) {
632 fc->ac.error = PTR_ERR(alist);
633 _leave(" = f [%d]", fc->ac.error);
634 return false;
635 }
636
637 discard = alist;
638 if (server->addr_version != alist->version) {
639 write_lock(&server->fs_lock);
640 discard = rcu_dereference_protected(server->addresses,
641 lockdep_is_held(&server->fs_lock));
642 rcu_assign_pointer(server->addresses, alist);
643 server->addr_version = alist->version;
644 write_unlock(&server->fs_lock);
645 }
646
647 server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
648 afs_put_addrlist(discard);
649 _leave(" = t");
650 return true;
651}
652
653/*
654 * See if a server's address list needs updating.
655 */
656bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
657{
658 time64_t now = ktime_get_real_seconds();
659 long diff;
660 bool success;
661 int ret, retries = 0;
662
663 _enter("");
664
665 ASSERT(server);
666
667retry:
668 diff = READ_ONCE(server->update_at) - now;
669 if (diff > 0) {
670 _leave(" = t [not now %ld]", diff);
671 return true;
672 }
673
674 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
675 success = afs_update_server_record(fc, server);
676 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
677 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
678 _leave(" = %d", success);
679 return success;
680 }
681
682 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
683 TASK_INTERRUPTIBLE);
684 if (ret == -ERESTARTSYS) {
685 fc->ac.error = ret;
686 _leave(" = f [intr]");
687 return false;
688 }
689
690 retries++;
691 if (retries == 4) {
692 _leave(" = f [stale]");
693 ret = -ESTALE;
694 return false;
695 }
696 goto retry;
697}