1 /* AFS server record management
3 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
12 #include <linux/sched.h>
13 #include <linux/slab.h>
17 static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
18 static unsigned afs_server_update_delay = 30; /* Time till VLDB recheck in secs */
20 static void afs_inc_servers_outstanding(struct afs_net *net)
22 atomic_inc(&net->servers_outstanding);
25 static void afs_dec_servers_outstanding(struct afs_net *net)
27 if (atomic_dec_and_test(&net->servers_outstanding))
28 wake_up_var(&net->servers_outstanding);
32 * Find a server by one of its addresses.
34 struct afs_server *afs_find_server(struct afs_net *net,
35 const struct sockaddr_rxrpc *srx)
37 const struct afs_addr_list *alist;
38 struct afs_server *server = NULL;
46 afs_put_server(net, server);
48 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
50 if (srx->transport.family == AF_INET6) {
51 const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
52 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
53 alist = rcu_dereference(server->addresses);
54 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
55 b = &alist->addrs[i].transport.sin6;
56 diff = ((u16 __force)a->sin6_port -
57 (u16 __force)b->sin6_port);
59 diff = memcmp(&a->sin6_addr,
61 sizeof(struct in6_addr));
67 const struct sockaddr_in *a = &srx->transport.sin, *b;
68 hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
69 alist = rcu_dereference(server->addresses);
70 for (i = 0; i < alist->nr_ipv4; i++) {
71 b = &alist->addrs[i].transport.sin;
72 diff = ((u16 __force)a->sin_port -
73 (u16 __force)b->sin_port);
75 diff = ((u32 __force)a->sin_addr.s_addr -
76 (u32 __force)b->sin_addr.s_addr);
85 if (server && !atomic_inc_not_zero(&server->usage))
88 } while (need_seqretry(&net->fs_addr_lock, seq));
90 done_seqretry(&net->fs_addr_lock, seq);
97 * Look up a server by its UUID
99 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
101 struct afs_server *server = NULL;
108 /* Unfortunately, rbtree walking doesn't give reliable results
109 * under just the RCU read lock, so we have to check for
113 afs_put_server(net, server);
116 read_seqbegin_or_lock(&net->fs_lock, &seq);
118 p = net->fs_servers.rb_node;
120 server = rb_entry(p, struct afs_server, uuid_rb);
122 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
125 } else if (diff > 0) {
128 afs_get_server(server);
134 } while (need_seqretry(&net->fs_lock, seq));
136 done_seqretry(&net->fs_lock, seq);
138 _leave(" = %p", server);
143 * Install a server record in the namespace tree
145 static struct afs_server *afs_install_server(struct afs_net *net,
146 struct afs_server *candidate)
148 const struct afs_addr_list *alist;
149 struct afs_server *server;
150 struct rb_node **pp, *p;
151 int ret = -EEXIST, diff;
153 _enter("%p", candidate);
155 write_seqlock(&net->fs_lock);
157 /* Firstly install the server in the UUID lookup tree */
158 pp = &net->fs_servers.rb_node;
162 _debug("- consider %p", p);
163 server = rb_entry(p, struct afs_server, uuid_rb);
164 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
166 pp = &(*pp)->rb_left;
168 pp = &(*pp)->rb_right;
174 rb_link_node(&server->uuid_rb, p, pp);
175 rb_insert_color(&server->uuid_rb, &net->fs_servers);
176 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
178 write_seqlock(&net->fs_addr_lock);
179 alist = rcu_dereference_protected(server->addresses,
180 lockdep_is_held(&net->fs_addr_lock.lock));
182 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
183 * it in the IPv4 and/or IPv6 reverse-map lists.
185 * TODO: For speed we want to use something other than a flat list
186 * here; even sorting the list in terms of lowest address would help a
187 * bit, but anything we might want to do gets messy and memory
190 if (alist->nr_ipv4 > 0)
191 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
192 if (alist->nr_addrs > alist->nr_ipv4)
193 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
195 write_sequnlock(&net->fs_addr_lock);
199 afs_get_server(server);
200 write_sequnlock(&net->fs_lock);
205 * allocate a new server record
207 static struct afs_server *afs_alloc_server(struct afs_net *net,
209 struct afs_addr_list *alist)
211 struct afs_server *server;
215 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
219 atomic_set(&server->usage, 1);
220 RCU_INIT_POINTER(server->addresses, alist);
221 server->addr_version = alist->version;
222 server->uuid = *uuid;
223 server->flags = (1UL << AFS_SERVER_FL_NEW);
224 server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
225 rwlock_init(&server->fs_lock);
226 INIT_HLIST_HEAD(&server->cb_volumes);
227 rwlock_init(&server->cb_break_lock);
229 afs_inc_servers_outstanding(net);
230 _leave(" = %p", server);
234 _leave(" = NULL [nomem]");
239 * Look up an address record for a server
241 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
242 struct key *key, const uuid_t *uuid)
244 struct afs_addr_cursor ac;
245 struct afs_addr_list *alist;
248 ret = afs_set_vl_cursor(&ac, cell);
252 while (afs_iterate_addresses(&ac)) {
253 if (test_bit(ac.index, &ac.alist->yfs))
254 alist = afs_yfsvl_get_endpoints(cell->net, &ac, key, uuid);
256 alist = afs_vl_get_addrs_u(cell->net, &ac, key, uuid);
262 ac.error = afs_abort_to_error(ac.abort_code);
278 return ERR_PTR(afs_end_cursor(&ac));
282 * Get or create a fileserver record.
284 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
287 struct afs_addr_list *alist;
288 struct afs_server *server, *candidate;
290 _enter("%p,%pU", cell->net, uuid);
292 server = afs_find_server_by_uuid(cell->net, uuid);
296 alist = afs_vl_lookup_addrs(cell, key, uuid);
298 return ERR_CAST(alist);
300 candidate = afs_alloc_server(cell->net, uuid, alist);
302 afs_put_addrlist(alist);
303 return ERR_PTR(-ENOMEM);
306 server = afs_install_server(cell->net, candidate);
307 if (server != candidate) {
308 afs_put_addrlist(alist);
312 _leave(" = %p{%d}", server, atomic_read(&server->usage));
317 * Set the server timer to fire after a given delay, assuming it's not already
318 * set for an earlier time.
320 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
323 afs_inc_servers_outstanding(net);
324 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
325 afs_dec_servers_outstanding(net);
330 * Server management timer. We have an increment on fs_outstanding that we
331 * need to pass along to the work item.
333 void afs_servers_timer(struct timer_list *timer)
335 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
338 if (!queue_work(afs_wq, &net->fs_manager))
339 afs_dec_servers_outstanding(net);
343 * Release a reference on a server record.
345 void afs_put_server(struct afs_net *net, struct afs_server *server)
352 server->put_time = ktime_get_real_seconds();
354 usage = atomic_dec_return(&server->usage);
356 _enter("{%u}", usage);
358 if (likely(usage > 0))
361 afs_set_server_timer(net, afs_server_gc_delay);
364 static void afs_server_rcu(struct rcu_head *rcu)
366 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
368 afs_put_addrlist(rcu_access_pointer(server->addresses));
373 * destroy a dead server
375 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
377 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
378 struct afs_addr_cursor ac = {
380 .start = alist->index,
382 .addr = &alist->addrs[alist->index],
385 _enter("%p", server);
387 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
388 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
390 call_rcu(&server->rcu, afs_server_rcu);
391 afs_dec_servers_outstanding(net);
395 * Garbage collect any expired servers.
397 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
399 struct afs_server *server;
403 while ((server = gc_list)) {
404 gc_list = server->gc_next;
406 write_seqlock(&net->fs_lock);
408 deleted = atomic_try_cmpxchg(&server->usage, &usage, 0);
410 rb_erase(&server->uuid_rb, &net->fs_servers);
411 hlist_del_rcu(&server->proc_link);
413 write_sequnlock(&net->fs_lock);
416 write_seqlock(&net->fs_addr_lock);
417 if (!hlist_unhashed(&server->addr4_link))
418 hlist_del_rcu(&server->addr4_link);
419 if (!hlist_unhashed(&server->addr6_link))
420 hlist_del_rcu(&server->addr6_link);
421 write_sequnlock(&net->fs_addr_lock);
422 afs_destroy_server(net, server);
428 * Manage the records of servers known to be within a network namespace. This
429 * includes garbage collecting unused servers.
431 * Note also that we were given an increment on net->servers_outstanding by
432 * whoever queued us that we need to deal with before returning.
434 void afs_manage_servers(struct work_struct *work)
436 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
437 struct afs_server *gc_list = NULL;
438 struct rb_node *cursor;
439 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
440 bool purging = !net->live;
444 /* Trawl the server list looking for servers that have expired from
447 read_seqlock_excl(&net->fs_lock);
449 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
450 struct afs_server *server =
451 rb_entry(cursor, struct afs_server, uuid_rb);
452 int usage = atomic_read(&server->usage);
454 _debug("manage %pU %u", &server->uuid, usage);
456 ASSERTCMP(usage, >=, 1);
457 ASSERTIFCMP(purging, usage, ==, 1);
460 time64_t expire_at = server->put_time;
462 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
463 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
464 expire_at += afs_server_gc_delay;
465 if (purging || expire_at <= now) {
466 server->gc_next = gc_list;
468 } else if (expire_at < next_manage) {
469 next_manage = expire_at;
474 read_sequnlock_excl(&net->fs_lock);
476 /* Update the timer on the way out. We have to pass an increment on
477 * servers_outstanding in the namespace that we are in to the timer or
478 * the work scheduler.
480 if (!purging && next_manage < TIME64_MAX) {
481 now = ktime_get_real_seconds();
483 if (next_manage - now <= 0) {
484 if (queue_work(afs_wq, &net->fs_manager))
485 afs_inc_servers_outstanding(net);
487 afs_set_server_timer(net, next_manage - now);
491 afs_gc_servers(net, gc_list);
493 afs_dec_servers_outstanding(net);
494 _leave(" [%d]", atomic_read(&net->servers_outstanding));
497 static void afs_queue_server_manager(struct afs_net *net)
499 afs_inc_servers_outstanding(net);
500 if (!queue_work(afs_wq, &net->fs_manager))
501 afs_dec_servers_outstanding(net);
505 * Purge list of servers.
507 void afs_purge_servers(struct afs_net *net)
511 if (del_timer_sync(&net->fs_timer))
512 atomic_dec(&net->servers_outstanding);
514 afs_queue_server_manager(net);
517 wait_var_event(&net->servers_outstanding,
518 !atomic_read(&net->servers_outstanding));
523 * Probe a fileserver to find its capabilities.
525 * TODO: Try service upgrade.
527 static bool afs_do_probe_fileserver(struct afs_fs_cursor *fc)
532 fc->ac.start = READ_ONCE(fc->ac.alist->index);
533 fc->ac.index = fc->ac.start;
535 fc->ac.begun = false;
537 while (afs_iterate_addresses(&fc->ac)) {
538 afs_fs_get_capabilities(afs_v2net(fc->vnode), fc->cbi->server,
540 switch (fc->ac.error) {
542 afs_end_cursor(&fc->ac);
543 set_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags);
546 fc->ac.error = afs_abort_to_error(fc->ac.abort_code);
564 afs_end_cursor(&fc->ac);
569 * If we haven't already, try probing the fileserver to get its capabilities.
570 * We try not to instigate parallel probes, but it's possible that the parallel
571 * probes will fail due to authentication failure when ours would succeed.
573 * TODO: Try sending an anonymous probe if an authenticated probe fails.
575 bool afs_probe_fileserver(struct afs_fs_cursor *fc)
578 int ret, retries = 0;
583 if (test_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags)) {
588 if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags)) {
589 success = afs_do_probe_fileserver(fc);
590 clear_bit_unlock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags);
591 wake_up_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING);
597 ret = wait_on_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING,
599 if (ret == -ERESTARTSYS) {
601 _leave(" = f [%d]", ret);
607 fc->ac.error = -ESTALE;
608 _leave(" = f [stale]");
616 * Get an update for a server's address list.
618 static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
620 struct afs_addr_list *alist, *discard;
624 alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
627 fc->ac.error = PTR_ERR(alist);
628 _leave(" = f [%d]", fc->ac.error);
633 if (server->addr_version != alist->version) {
634 write_lock(&server->fs_lock);
635 discard = rcu_dereference_protected(server->addresses,
636 lockdep_is_held(&server->fs_lock));
637 rcu_assign_pointer(server->addresses, alist);
638 server->addr_version = alist->version;
639 write_unlock(&server->fs_lock);
642 server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
643 afs_put_addrlist(discard);
649 * See if a server's address list needs updating.
651 bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
653 time64_t now = ktime_get_real_seconds();
656 int ret, retries = 0;
663 diff = READ_ONCE(server->update_at) - now;
665 _leave(" = t [not now %ld]", diff);
669 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
670 success = afs_update_server_record(fc, server);
671 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
672 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
673 _leave(" = %d", success);
677 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
679 if (ret == -ERESTARTSYS) {
681 _leave(" = f [intr]");
687 _leave(" = f [stale]");