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,
25 const struct sockaddr_rxrpc *srx)
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 if (srx->transport.family == AF_INET6) {
42 const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
43 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
44 alist = rcu_dereference(server->addresses);
45 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
46 b = &alist->addrs[i].transport.sin6;
47 diff = ((u16 __force)a->sin6_port -
48 (u16 __force)b->sin6_port);
50 diff = memcmp(&a->sin6_addr,
52 sizeof(struct in6_addr));
58 const struct sockaddr_in *a = &srx->transport.sin, *b;
59 hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
60 alist = rcu_dereference(server->addresses);
61 for (i = 0; i < alist->nr_ipv4; i++) {
62 b = &alist->addrs[i].transport.sin;
63 diff = ((u16 __force)a->sin_port -
64 (u16 __force)b->sin_port);
66 diff = ((u32 __force)a->sin_addr.s_addr -
67 (u32 __force)b->sin_addr.s_addr);
77 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
79 } while (need_seqretry(&net->fs_addr_lock, seq));
81 done_seqretry(&net->fs_addr_lock, seq);
88 * Look up a server by its UUID and mark it active.
90 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
92 struct afs_server *server = NULL;
99 /* Unfortunately, rbtree walking doesn't give reliable results
100 * under just the RCU read lock, so we have to check for
104 afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
106 seq++; /* 2 on the 1st/lockless path, otherwise odd */
107 read_seqbegin_or_lock(&net->fs_lock, &seq);
109 p = net->fs_servers.rb_node;
111 server = rb_entry(p, struct afs_server, uuid_rb);
113 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
116 } else if (diff > 0) {
119 afs_use_server(server, afs_server_trace_get_by_uuid);
125 } while (need_seqretry(&net->fs_lock, seq));
127 done_seqretry(&net->fs_lock, seq);
129 _leave(" = %p", server);
134 * Install a server record in the namespace tree. If there's a clash, we stick
135 * it into a list anchored on whichever afs_server struct is actually in the
138 static struct afs_server *afs_install_server(struct afs_cell *cell,
139 struct afs_server *candidate)
141 const struct afs_addr_list *alist;
142 struct afs_server *server, *next;
143 struct afs_net *net = cell->net;
144 struct rb_node **pp, *p;
147 _enter("%p", candidate);
149 write_seqlock(&net->fs_lock);
151 /* Firstly install the server in the UUID lookup tree */
152 pp = &net->fs_servers.rb_node;
156 _debug("- consider %p", p);
157 server = rb_entry(p, struct afs_server, uuid_rb);
158 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
160 pp = &(*pp)->rb_left;
161 } else if (diff > 0) {
162 pp = &(*pp)->rb_right;
164 if (server->cell == cell)
167 /* We have the same UUID representing servers in
168 * different cells. Append the new server to the list.
171 next = rcu_dereference_protected(
173 lockdep_is_held(&net->fs_lock.lock));
178 rcu_assign_pointer(server->uuid_next, candidate);
179 candidate->uuid_prev = server;
186 rb_link_node(&server->uuid_rb, p, pp);
187 rb_insert_color(&server->uuid_rb, &net->fs_servers);
188 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
191 write_seqlock(&net->fs_addr_lock);
192 alist = rcu_dereference_protected(server->addresses,
193 lockdep_is_held(&net->fs_addr_lock.lock));
195 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
196 * it in the IPv4 and/or IPv6 reverse-map lists.
198 * TODO: For speed we want to use something other than a flat list
199 * here; even sorting the list in terms of lowest address would help a
200 * bit, but anything we might want to do gets messy and memory
203 if (alist->nr_ipv4 > 0)
204 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
205 if (alist->nr_addrs > alist->nr_ipv4)
206 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
208 write_sequnlock(&net->fs_addr_lock);
211 afs_get_server(server, afs_server_trace_get_install);
212 write_sequnlock(&net->fs_lock);
217 * Allocate a new server record and mark it active.
219 static struct afs_server *afs_alloc_server(struct afs_cell *cell,
221 struct afs_addr_list *alist)
223 struct afs_server *server;
224 struct afs_net *net = cell->net;
228 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
232 refcount_set(&server->ref, 1);
233 atomic_set(&server->active, 1);
234 server->debug_id = atomic_inc_return(&afs_server_debug_id);
235 RCU_INIT_POINTER(server->addresses, alist);
236 server->addr_version = alist->version;
237 server->uuid = *uuid;
238 rwlock_init(&server->fs_lock);
239 INIT_WORK(&server->initcb_work, afs_server_init_callback_work);
240 init_waitqueue_head(&server->probe_wq);
241 INIT_LIST_HEAD(&server->probe_link);
242 spin_lock_init(&server->probe_lock);
244 server->rtt = UINT_MAX;
246 afs_inc_servers_outstanding(net);
247 trace_afs_server(server->debug_id, 1, 1, afs_server_trace_alloc);
248 _leave(" = %p", server);
252 _leave(" = NULL [nomem]");
257 * Look up an address record for a server
259 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
260 struct key *key, const uuid_t *uuid)
262 struct afs_vl_cursor vc;
263 struct afs_addr_list *alist = NULL;
267 if (afs_begin_vlserver_operation(&vc, cell, key)) {
268 while (afs_select_vlserver(&vc)) {
269 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
270 alist = afs_yfsvl_get_endpoints(&vc, uuid);
272 alist = afs_vl_get_addrs_u(&vc, uuid);
275 ret = afs_end_vlserver_operation(&vc);
278 return ret < 0 ? ERR_PTR(ret) : alist;
282 * Get or create a fileserver record.
284 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
285 const uuid_t *uuid, u32 addr_version)
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);
294 if (server->addr_version != addr_version)
295 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
299 alist = afs_vl_lookup_addrs(cell, key, uuid);
301 return ERR_CAST(alist);
303 candidate = afs_alloc_server(cell, uuid, alist);
305 afs_put_addrlist(alist);
306 return ERR_PTR(-ENOMEM);
309 server = afs_install_server(cell, candidate);
310 if (server != candidate) {
311 afs_put_addrlist(alist);
314 /* Immediately dispatch an asynchronous probe to each interface
315 * on the fileserver. This will make sure the repeat-probing
316 * service is started.
318 afs_fs_probe_fileserver(cell->net, server, key, true);
325 * Set the server timer to fire after a given delay, assuming it's not already
326 * set for an earlier time.
328 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
331 afs_inc_servers_outstanding(net);
332 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
333 afs_dec_servers_outstanding(net);
338 * Server management timer. We have an increment on fs_outstanding that we
339 * need to pass along to the work item.
341 void afs_servers_timer(struct timer_list *timer)
343 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
346 if (!queue_work(afs_wq, &net->fs_manager))
347 afs_dec_servers_outstanding(net);
351 * Get a reference on a server object.
353 struct afs_server *afs_get_server(struct afs_server *server,
354 enum afs_server_trace reason)
359 __refcount_inc(&server->ref, &r);
360 a = atomic_read(&server->active);
361 trace_afs_server(server->debug_id, r + 1, a, reason);
366 * Try to get a reference on a server object.
368 static struct afs_server *afs_maybe_use_server(struct afs_server *server,
369 enum afs_server_trace reason)
374 if (!__refcount_inc_not_zero(&server->ref, &r))
377 a = atomic_inc_return(&server->active);
378 trace_afs_server(server->debug_id, r + 1, a, reason);
383 * Get an active count on a server object.
385 struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
390 __refcount_inc(&server->ref, &r);
391 a = atomic_inc_return(&server->active);
393 trace_afs_server(server->debug_id, r + 1, a, reason);
398 * Release a reference on a server record.
400 void afs_put_server(struct afs_net *net, struct afs_server *server,
401 enum afs_server_trace reason)
403 unsigned int a, debug_id = server->debug_id;
410 a = atomic_read(&server->active);
411 zero = __refcount_dec_and_test(&server->ref, &r);
412 trace_afs_server(debug_id, r - 1, a, reason);
414 __afs_put_server(net, server);
418 * Drop an active count on a server object without updating the last-unused
421 void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
422 enum afs_server_trace reason)
425 unsigned int active = atomic_dec_return(&server->active);
428 afs_set_server_timer(net, afs_server_gc_delay);
429 afs_put_server(net, server, reason);
434 * Drop an active count on a server object.
436 void afs_unuse_server(struct afs_net *net, struct afs_server *server,
437 enum afs_server_trace reason)
440 server->unuse_time = ktime_get_real_seconds();
441 afs_unuse_server_notime(net, server, reason);
445 static void afs_server_rcu(struct rcu_head *rcu)
447 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
449 trace_afs_server(server->debug_id, refcount_read(&server->ref),
450 atomic_read(&server->active), afs_server_trace_free);
451 afs_put_addrlist(rcu_access_pointer(server->addresses));
455 static void __afs_put_server(struct afs_net *net, struct afs_server *server)
457 call_rcu(&server->rcu, afs_server_rcu);
458 afs_dec_servers_outstanding(net);
461 static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
463 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
464 struct afs_addr_cursor ac = {
466 .index = alist->preferred,
470 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
474 * destroy a dead server
476 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
478 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
479 afs_give_up_callbacks(net, server);
481 flush_work(&server->initcb_work);
482 afs_put_server(net, server, afs_server_trace_destroy);
486 * Garbage collect any expired servers.
488 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
490 struct afs_server *server, *next, *prev;
493 while ((server = gc_list)) {
494 gc_list = server->gc_next;
496 write_seqlock(&net->fs_lock);
498 active = atomic_read(&server->active);
500 trace_afs_server(server->debug_id, refcount_read(&server->ref),
501 active, afs_server_trace_gc);
502 next = rcu_dereference_protected(
503 server->uuid_next, lockdep_is_held(&net->fs_lock.lock));
504 prev = server->uuid_prev;
506 /* The one at the front is in the tree */
508 rb_erase(&server->uuid_rb, &net->fs_servers);
510 rb_replace_node_rcu(&server->uuid_rb,
513 next->uuid_prev = NULL;
516 /* This server is not at the front */
517 rcu_assign_pointer(prev->uuid_next, next);
519 next->uuid_prev = prev;
522 list_del(&server->probe_link);
523 hlist_del_rcu(&server->proc_link);
524 if (!hlist_unhashed(&server->addr4_link))
525 hlist_del_rcu(&server->addr4_link);
526 if (!hlist_unhashed(&server->addr6_link))
527 hlist_del_rcu(&server->addr6_link);
529 write_sequnlock(&net->fs_lock);
532 afs_destroy_server(net, server);
537 * Manage the records of servers known to be within a network namespace. This
538 * includes garbage collecting unused servers.
540 * Note also that we were given an increment on net->servers_outstanding by
541 * whoever queued us that we need to deal with before returning.
543 void afs_manage_servers(struct work_struct *work)
545 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
546 struct afs_server *gc_list = NULL;
547 struct rb_node *cursor;
548 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
549 bool purging = !net->live;
553 /* Trawl the server list looking for servers that have expired from
556 read_seqlock_excl(&net->fs_lock);
558 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
559 struct afs_server *server =
560 rb_entry(cursor, struct afs_server, uuid_rb);
561 int active = atomic_read(&server->active);
563 _debug("manage %pU %u", &server->uuid, active);
566 trace_afs_server(server->debug_id, refcount_read(&server->ref),
567 active, afs_server_trace_purging);
569 pr_notice("Can't purge s=%08x\n", server->debug_id);
573 time64_t expire_at = server->unuse_time;
575 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
576 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
577 expire_at += afs_server_gc_delay;
578 if (purging || expire_at <= now) {
579 server->gc_next = gc_list;
581 } else if (expire_at < next_manage) {
582 next_manage = expire_at;
587 read_sequnlock_excl(&net->fs_lock);
589 /* Update the timer on the way out. We have to pass an increment on
590 * servers_outstanding in the namespace that we are in to the timer or
591 * the work scheduler.
593 if (!purging && next_manage < TIME64_MAX) {
594 now = ktime_get_real_seconds();
596 if (next_manage - now <= 0) {
597 if (queue_work(afs_wq, &net->fs_manager))
598 afs_inc_servers_outstanding(net);
600 afs_set_server_timer(net, next_manage - now);
604 afs_gc_servers(net, gc_list);
606 afs_dec_servers_outstanding(net);
607 _leave(" [%d]", atomic_read(&net->servers_outstanding));
610 static void afs_queue_server_manager(struct afs_net *net)
612 afs_inc_servers_outstanding(net);
613 if (!queue_work(afs_wq, &net->fs_manager))
614 afs_dec_servers_outstanding(net);
618 * Purge list of servers.
620 void afs_purge_servers(struct afs_net *net)
624 if (del_timer_sync(&net->fs_timer))
625 afs_dec_servers_outstanding(net);
627 afs_queue_server_manager(net);
630 atomic_dec(&net->servers_outstanding);
631 wait_var_event(&net->servers_outstanding,
632 !atomic_read(&net->servers_outstanding));
637 * Get an update for a server's address list.
639 static noinline bool afs_update_server_record(struct afs_operation *op,
640 struct afs_server *server)
642 struct afs_addr_list *alist, *discard;
646 trace_afs_server(server->debug_id, refcount_read(&server->ref),
647 atomic_read(&server->active),
648 afs_server_trace_update);
650 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
652 if ((PTR_ERR(alist) == -ERESTARTSYS ||
653 PTR_ERR(alist) == -EINTR) &&
654 (op->flags & AFS_OPERATION_UNINTR) &&
656 _leave(" = t [intr]");
659 op->error = PTR_ERR(alist);
660 _leave(" = f [%d]", op->error);
665 if (server->addr_version != alist->version) {
666 write_lock(&server->fs_lock);
667 discard = rcu_dereference_protected(server->addresses,
668 lockdep_is_held(&server->fs_lock));
669 rcu_assign_pointer(server->addresses, alist);
670 server->addr_version = alist->version;
671 write_unlock(&server->fs_lock);
674 afs_put_addrlist(discard);
680 * See if a server's address list needs updating.
682 bool afs_check_server_record(struct afs_operation *op, struct afs_server *server)
685 int ret, retries = 0;
692 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
694 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
696 _leave(" = t [good]");
700 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
701 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
702 success = afs_update_server_record(op, server);
703 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
704 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
705 _leave(" = %d", success);
710 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
711 (op->flags & AFS_OPERATION_UNINTR) ?
712 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
713 if (ret == -ERESTARTSYS) {
715 _leave(" = f [intr]");
721 _leave(" = f [stale]");