1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS server record management
4 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #include <linux/sched.h>
9 #include <linux/slab.h>
12 #include "protocol_yfs.h"
14 static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
15 static atomic_t afs_server_debug_id;
17 static struct afs_server *afs_maybe_use_server(struct afs_server *,
18 enum afs_server_trace);
19 static void __afs_put_server(struct afs_net *, struct afs_server *);
22 * Find a server by one of its addresses.
24 struct afs_server *afs_find_server(struct afs_net *net, const struct rxrpc_peer *peer)
26 const struct afs_endpoint_state *estate;
27 const struct afs_addr_list *alist;
28 struct afs_server *server = NULL;
36 afs_unuse_server_notime(net, server, afs_server_trace_put_find_rsq);
38 seq++; /* 2 on the 1st/lockless path, otherwise odd */
39 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
41 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
42 estate = rcu_dereference(server->endpoint_state);
43 alist = estate->addresses;
44 for (i = 0; i < alist->nr_addrs; i++)
45 if (alist->addrs[i].peer == peer)
52 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
54 } while (need_seqretry(&net->fs_addr_lock, seq));
56 done_seqretry(&net->fs_addr_lock, seq);
63 * Look up a server by its UUID and mark it active.
65 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
67 struct afs_server *server = NULL;
74 /* Unfortunately, rbtree walking doesn't give reliable results
75 * under just the RCU read lock, so we have to check for
79 afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
81 seq++; /* 2 on the 1st/lockless path, otherwise odd */
82 read_seqbegin_or_lock(&net->fs_lock, &seq);
84 p = net->fs_servers.rb_node;
86 server = rb_entry(p, struct afs_server, uuid_rb);
88 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
91 } else if (diff > 0) {
94 afs_use_server(server, afs_server_trace_get_by_uuid);
100 } while (need_seqretry(&net->fs_lock, seq));
102 done_seqretry(&net->fs_lock, seq);
104 _leave(" = %p", server);
109 * Install a server record in the namespace tree. If there's a clash, we stick
110 * it into a list anchored on whichever afs_server struct is actually in the
113 static struct afs_server *afs_install_server(struct afs_cell *cell,
114 struct afs_server *candidate)
116 const struct afs_endpoint_state *estate;
117 const struct afs_addr_list *alist;
118 struct afs_server *server, *next;
119 struct afs_net *net = cell->net;
120 struct rb_node **pp, *p;
123 _enter("%p", candidate);
125 write_seqlock(&net->fs_lock);
127 /* Firstly install the server in the UUID lookup tree */
128 pp = &net->fs_servers.rb_node;
132 _debug("- consider %p", p);
133 server = rb_entry(p, struct afs_server, uuid_rb);
134 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
136 pp = &(*pp)->rb_left;
137 } else if (diff > 0) {
138 pp = &(*pp)->rb_right;
140 if (server->cell == cell)
143 /* We have the same UUID representing servers in
144 * different cells. Append the new server to the list.
147 next = rcu_dereference_protected(
149 lockdep_is_held(&net->fs_lock.lock));
154 rcu_assign_pointer(server->uuid_next, candidate);
155 candidate->uuid_prev = server;
162 rb_link_node(&server->uuid_rb, p, pp);
163 rb_insert_color(&server->uuid_rb, &net->fs_servers);
164 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
167 write_seqlock(&net->fs_addr_lock);
168 estate = rcu_dereference_protected(server->endpoint_state,
169 lockdep_is_held(&net->fs_addr_lock.lock));
170 alist = estate->addresses;
172 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
173 * it in the IPv4 and/or IPv6 reverse-map lists.
175 * TODO: For speed we want to use something other than a flat list
176 * here; even sorting the list in terms of lowest address would help a
177 * bit, but anything we might want to do gets messy and memory
180 if (alist->nr_ipv4 > 0)
181 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
182 if (alist->nr_addrs > alist->nr_ipv4)
183 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
185 write_sequnlock(&net->fs_addr_lock);
188 afs_get_server(server, afs_server_trace_get_install);
189 write_sequnlock(&net->fs_lock);
194 * Allocate a new server record and mark it active.
196 static struct afs_server *afs_alloc_server(struct afs_cell *cell,
198 struct afs_addr_list *alist)
200 struct afs_endpoint_state *estate;
201 struct afs_server *server;
202 struct afs_net *net = cell->net;
206 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
210 estate = kzalloc(sizeof(struct afs_endpoint_state), GFP_KERNEL);
214 refcount_set(&server->ref, 1);
215 atomic_set(&server->active, 1);
216 server->debug_id = atomic_inc_return(&afs_server_debug_id);
217 server->addr_version = alist->version;
218 server->uuid = *uuid;
219 rwlock_init(&server->fs_lock);
220 INIT_LIST_HEAD(&server->volumes);
221 init_waitqueue_head(&server->probe_wq);
222 INIT_LIST_HEAD(&server->probe_link);
223 spin_lock_init(&server->probe_lock);
225 server->rtt = UINT_MAX;
226 server->service_id = FS_SERVICE;
228 server->probe_counter = 1;
229 server->probed_at = jiffies - LONG_MAX / 2;
230 refcount_set(&estate->ref, 1);
231 estate->addresses = alist;
232 estate->server_id = server->debug_id;
233 estate->probe_seq = 1;
234 rcu_assign_pointer(server->endpoint_state, estate);
236 afs_inc_servers_outstanding(net);
237 trace_afs_server(server->debug_id, 1, 1, afs_server_trace_alloc);
238 trace_afs_estate(estate->server_id, estate->probe_seq, refcount_read(&estate->ref),
239 afs_estate_trace_alloc_server);
240 _leave(" = %p", server);
246 _leave(" = NULL [nomem]");
251 * Look up an address record for a server
253 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
254 struct key *key, const uuid_t *uuid)
256 struct afs_vl_cursor vc;
257 struct afs_addr_list *alist = NULL;
261 if (afs_begin_vlserver_operation(&vc, cell, key)) {
262 while (afs_select_vlserver(&vc)) {
263 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
264 alist = afs_yfsvl_get_endpoints(&vc, uuid);
266 alist = afs_vl_get_addrs_u(&vc, uuid);
269 ret = afs_end_vlserver_operation(&vc);
272 return ret < 0 ? ERR_PTR(ret) : alist;
276 * Get or create a fileserver record.
278 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
279 const uuid_t *uuid, u32 addr_version)
281 struct afs_addr_list *alist;
282 struct afs_server *server, *candidate;
284 _enter("%p,%pU", cell->net, uuid);
286 server = afs_find_server_by_uuid(cell->net, uuid);
288 if (server->addr_version != addr_version)
289 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
293 alist = afs_vl_lookup_addrs(cell, key, uuid);
295 return ERR_CAST(alist);
297 candidate = afs_alloc_server(cell, uuid, alist);
299 afs_put_addrlist(alist, afs_alist_trace_put_server_oom);
300 return ERR_PTR(-ENOMEM);
303 server = afs_install_server(cell, candidate);
304 if (server != candidate) {
305 afs_put_addrlist(alist, afs_alist_trace_put_server_dup);
308 /* Immediately dispatch an asynchronous probe to each interface
309 * on the fileserver. This will make sure the repeat-probing
310 * service is started.
312 afs_fs_probe_fileserver(cell->net, server, alist, key);
319 * Set the server timer to fire after a given delay, assuming it's not already
320 * set for an earlier time.
322 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
325 afs_inc_servers_outstanding(net);
326 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
327 afs_dec_servers_outstanding(net);
332 * Server management timer. We have an increment on fs_outstanding that we
333 * need to pass along to the work item.
335 void afs_servers_timer(struct timer_list *timer)
337 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
340 if (!queue_work(afs_wq, &net->fs_manager))
341 afs_dec_servers_outstanding(net);
345 * Get a reference on a server object.
347 struct afs_server *afs_get_server(struct afs_server *server,
348 enum afs_server_trace reason)
353 __refcount_inc(&server->ref, &r);
354 a = atomic_read(&server->active);
355 trace_afs_server(server->debug_id, r + 1, a, reason);
360 * Try to get a reference on a server object.
362 static struct afs_server *afs_maybe_use_server(struct afs_server *server,
363 enum afs_server_trace reason)
368 if (!__refcount_inc_not_zero(&server->ref, &r))
371 a = atomic_inc_return(&server->active);
372 trace_afs_server(server->debug_id, r + 1, a, reason);
377 * Get an active count on a server object.
379 struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
384 __refcount_inc(&server->ref, &r);
385 a = atomic_inc_return(&server->active);
387 trace_afs_server(server->debug_id, r + 1, a, reason);
392 * Release a reference on a server record.
394 void afs_put_server(struct afs_net *net, struct afs_server *server,
395 enum afs_server_trace reason)
397 unsigned int a, debug_id = server->debug_id;
404 a = atomic_read(&server->active);
405 zero = __refcount_dec_and_test(&server->ref, &r);
406 trace_afs_server(debug_id, r - 1, a, reason);
408 __afs_put_server(net, server);
412 * Drop an active count on a server object without updating the last-unused
415 void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
416 enum afs_server_trace reason)
419 unsigned int active = atomic_dec_return(&server->active);
422 afs_set_server_timer(net, afs_server_gc_delay);
423 afs_put_server(net, server, reason);
428 * Drop an active count on a server object.
430 void afs_unuse_server(struct afs_net *net, struct afs_server *server,
431 enum afs_server_trace reason)
434 server->unuse_time = ktime_get_real_seconds();
435 afs_unuse_server_notime(net, server, reason);
439 static void afs_server_rcu(struct rcu_head *rcu)
441 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
443 trace_afs_server(server->debug_id, refcount_read(&server->ref),
444 atomic_read(&server->active), afs_server_trace_free);
445 afs_put_endpoint_state(rcu_access_pointer(server->endpoint_state),
446 afs_estate_trace_put_server);
450 static void __afs_put_server(struct afs_net *net, struct afs_server *server)
452 call_rcu(&server->rcu, afs_server_rcu);
453 afs_dec_servers_outstanding(net);
456 static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
458 struct afs_endpoint_state *estate = rcu_access_pointer(server->endpoint_state);
459 struct afs_addr_list *alist = estate->addresses;
461 afs_fs_give_up_all_callbacks(net, server, &alist->addrs[alist->preferred], NULL);
465 * destroy a dead server
467 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
469 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
470 afs_give_up_callbacks(net, server);
472 afs_put_server(net, server, afs_server_trace_destroy);
476 * Garbage collect any expired servers.
478 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
480 struct afs_server *server, *next, *prev;
483 while ((server = gc_list)) {
484 gc_list = server->gc_next;
486 write_seqlock(&net->fs_lock);
488 active = atomic_read(&server->active);
490 trace_afs_server(server->debug_id, refcount_read(&server->ref),
491 active, afs_server_trace_gc);
492 next = rcu_dereference_protected(
493 server->uuid_next, lockdep_is_held(&net->fs_lock.lock));
494 prev = server->uuid_prev;
496 /* The one at the front is in the tree */
498 rb_erase(&server->uuid_rb, &net->fs_servers);
500 rb_replace_node_rcu(&server->uuid_rb,
503 next->uuid_prev = NULL;
506 /* This server is not at the front */
507 rcu_assign_pointer(prev->uuid_next, next);
509 next->uuid_prev = prev;
512 list_del(&server->probe_link);
513 hlist_del_rcu(&server->proc_link);
514 if (!hlist_unhashed(&server->addr4_link))
515 hlist_del_rcu(&server->addr4_link);
516 if (!hlist_unhashed(&server->addr6_link))
517 hlist_del_rcu(&server->addr6_link);
519 write_sequnlock(&net->fs_lock);
522 afs_destroy_server(net, server);
527 * Manage the records of servers known to be within a network namespace. This
528 * includes garbage collecting unused servers.
530 * Note also that we were given an increment on net->servers_outstanding by
531 * whoever queued us that we need to deal with before returning.
533 void afs_manage_servers(struct work_struct *work)
535 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
536 struct afs_server *gc_list = NULL;
537 struct rb_node *cursor;
538 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
539 bool purging = !net->live;
543 /* Trawl the server list looking for servers that have expired from
546 read_seqlock_excl(&net->fs_lock);
548 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
549 struct afs_server *server =
550 rb_entry(cursor, struct afs_server, uuid_rb);
551 int active = atomic_read(&server->active);
553 _debug("manage %pU %u", &server->uuid, active);
556 trace_afs_server(server->debug_id, refcount_read(&server->ref),
557 active, afs_server_trace_purging);
559 pr_notice("Can't purge s=%08x\n", server->debug_id);
563 time64_t expire_at = server->unuse_time;
565 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
566 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
567 expire_at += afs_server_gc_delay;
568 if (purging || expire_at <= now) {
569 server->gc_next = gc_list;
571 } else if (expire_at < next_manage) {
572 next_manage = expire_at;
577 read_sequnlock_excl(&net->fs_lock);
579 /* Update the timer on the way out. We have to pass an increment on
580 * servers_outstanding in the namespace that we are in to the timer or
581 * the work scheduler.
583 if (!purging && next_manage < TIME64_MAX) {
584 now = ktime_get_real_seconds();
586 if (next_manage - now <= 0) {
587 if (queue_work(afs_wq, &net->fs_manager))
588 afs_inc_servers_outstanding(net);
590 afs_set_server_timer(net, next_manage - now);
594 afs_gc_servers(net, gc_list);
596 afs_dec_servers_outstanding(net);
597 _leave(" [%d]", atomic_read(&net->servers_outstanding));
600 static void afs_queue_server_manager(struct afs_net *net)
602 afs_inc_servers_outstanding(net);
603 if (!queue_work(afs_wq, &net->fs_manager))
604 afs_dec_servers_outstanding(net);
608 * Purge list of servers.
610 void afs_purge_servers(struct afs_net *net)
614 if (del_timer_sync(&net->fs_timer))
615 afs_dec_servers_outstanding(net);
617 afs_queue_server_manager(net);
620 atomic_dec(&net->servers_outstanding);
621 wait_var_event(&net->servers_outstanding,
622 !atomic_read(&net->servers_outstanding));
627 * Get an update for a server's address list.
629 static noinline bool afs_update_server_record(struct afs_operation *op,
630 struct afs_server *server,
633 struct afs_endpoint_state *estate;
634 struct afs_addr_list *alist;
639 trace_afs_server(server->debug_id, refcount_read(&server->ref),
640 atomic_read(&server->active),
641 afs_server_trace_update);
643 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
646 estate = rcu_dereference(server->endpoint_state);
647 has_addrs = estate->addresses;
650 if ((PTR_ERR(alist) == -ERESTARTSYS ||
651 PTR_ERR(alist) == -EINTR) &&
652 (op->flags & AFS_OPERATION_UNINTR) &&
654 _leave(" = t [intr]");
657 afs_op_set_error(op, PTR_ERR(alist));
658 _leave(" = f [%d]", afs_op_error(op));
662 if (server->addr_version != alist->version)
663 afs_fs_probe_fileserver(op->net, server, alist, key);
665 afs_put_addrlist(alist, afs_alist_trace_put_server_update);
671 * See if a server's address list needs updating.
673 bool afs_check_server_record(struct afs_operation *op, struct afs_server *server,
677 int ret, retries = 0;
684 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
686 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
688 _leave(" = t [good]");
692 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
693 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
694 success = afs_update_server_record(op, server, key);
695 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
696 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
697 _leave(" = %d", success);
702 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
703 (op->flags & AFS_OPERATION_UNINTR) ?
704 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
705 if (ret == -ERESTARTSYS) {
706 afs_op_set_error(op, ret);
707 _leave(" = f [intr]");
713 _leave(" = f [stale]");