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 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
40 if (srx->transport.family == AF_INET6) {
41 const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
42 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
43 alist = rcu_dereference(server->addresses);
44 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
45 b = &alist->addrs[i].transport.sin6;
46 diff = ((u16 __force)a->sin6_port -
47 (u16 __force)b->sin6_port);
49 diff = memcmp(&a->sin6_addr,
51 sizeof(struct in6_addr));
57 const struct sockaddr_in *a = &srx->transport.sin, *b;
58 hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
59 alist = rcu_dereference(server->addresses);
60 for (i = 0; i < alist->nr_ipv4; i++) {
61 b = &alist->addrs[i].transport.sin;
62 diff = ((u16 __force)a->sin_port -
63 (u16 __force)b->sin_port);
65 diff = ((u32 __force)a->sin_addr.s_addr -
66 (u32 __force)b->sin_addr.s_addr);
76 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
78 } while (need_seqretry(&net->fs_addr_lock, seq));
80 done_seqretry(&net->fs_addr_lock, seq);
87 * Look up a server by its UUID and mark it active.
89 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
91 struct afs_server *server = NULL;
98 /* Unfortunately, rbtree walking doesn't give reliable results
99 * under just the RCU read lock, so we have to check for
103 afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
106 read_seqbegin_or_lock(&net->fs_lock, &seq);
108 p = net->fs_servers.rb_node;
110 server = rb_entry(p, struct afs_server, uuid_rb);
112 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
115 } else if (diff > 0) {
118 afs_use_server(server, afs_server_trace_get_by_uuid);
124 } while (need_seqretry(&net->fs_lock, seq));
126 done_seqretry(&net->fs_lock, seq);
128 _leave(" = %p", server);
133 * Install a server record in the namespace tree. If there's a clash, we stick
134 * it into a list anchored on whichever afs_server struct is actually in the
137 static struct afs_server *afs_install_server(struct afs_cell *cell,
138 struct afs_server *candidate)
140 const struct afs_addr_list *alist;
141 struct afs_server *server, *next;
142 struct afs_net *net = cell->net;
143 struct rb_node **pp, *p;
146 _enter("%p", candidate);
148 write_seqlock(&net->fs_lock);
150 /* Firstly install the server in the UUID lookup tree */
151 pp = &net->fs_servers.rb_node;
155 _debug("- consider %p", p);
156 server = rb_entry(p, struct afs_server, uuid_rb);
157 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
159 pp = &(*pp)->rb_left;
160 } else if (diff > 0) {
161 pp = &(*pp)->rb_right;
163 if (server->cell == cell)
166 /* We have the same UUID representing servers in
167 * different cells. Append the new server to the list.
170 next = rcu_dereference_protected(
172 lockdep_is_held(&net->fs_lock.lock));
177 rcu_assign_pointer(server->uuid_next, candidate);
178 candidate->uuid_prev = server;
185 rb_link_node(&server->uuid_rb, p, pp);
186 rb_insert_color(&server->uuid_rb, &net->fs_servers);
187 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
190 write_seqlock(&net->fs_addr_lock);
191 alist = rcu_dereference_protected(server->addresses,
192 lockdep_is_held(&net->fs_addr_lock.lock));
194 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
195 * it in the IPv4 and/or IPv6 reverse-map lists.
197 * TODO: For speed we want to use something other than a flat list
198 * here; even sorting the list in terms of lowest address would help a
199 * bit, but anything we might want to do gets messy and memory
202 if (alist->nr_ipv4 > 0)
203 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
204 if (alist->nr_addrs > alist->nr_ipv4)
205 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
207 write_sequnlock(&net->fs_addr_lock);
210 afs_get_server(server, afs_server_trace_get_install);
211 write_sequnlock(&net->fs_lock);
216 * Allocate a new server record and mark it active.
218 static struct afs_server *afs_alloc_server(struct afs_cell *cell,
220 struct afs_addr_list *alist)
222 struct afs_server *server;
223 struct afs_net *net = cell->net;
227 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
231 atomic_set(&server->ref, 1);
232 atomic_set(&server->active, 1);
233 server->debug_id = atomic_inc_return(&afs_server_debug_id);
234 RCU_INIT_POINTER(server->addresses, alist);
235 server->addr_version = alist->version;
236 server->uuid = *uuid;
237 rwlock_init(&server->fs_lock);
238 INIT_WORK(&server->initcb_work, afs_server_init_callback_work);
239 init_waitqueue_head(&server->probe_wq);
240 INIT_LIST_HEAD(&server->probe_link);
241 spin_lock_init(&server->probe_lock);
243 server->rtt = UINT_MAX;
245 afs_inc_servers_outstanding(net);
246 trace_afs_server(server, 1, 1, afs_server_trace_alloc);
247 _leave(" = %p", server);
251 _leave(" = NULL [nomem]");
256 * Look up an address record for a server
258 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
259 struct key *key, const uuid_t *uuid)
261 struct afs_vl_cursor vc;
262 struct afs_addr_list *alist = NULL;
266 if (afs_begin_vlserver_operation(&vc, cell, key)) {
267 while (afs_select_vlserver(&vc)) {
268 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
269 alist = afs_yfsvl_get_endpoints(&vc, uuid);
271 alist = afs_vl_get_addrs_u(&vc, uuid);
274 ret = afs_end_vlserver_operation(&vc);
277 return ret < 0 ? ERR_PTR(ret) : alist;
281 * Get or create a fileserver record.
283 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
284 const uuid_t *uuid, u32 addr_version)
286 struct afs_addr_list *alist;
287 struct afs_server *server, *candidate;
289 _enter("%p,%pU", cell->net, uuid);
291 server = afs_find_server_by_uuid(cell->net, uuid);
293 if (server->addr_version != addr_version)
294 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
298 alist = afs_vl_lookup_addrs(cell, key, uuid);
300 return ERR_CAST(alist);
302 candidate = afs_alloc_server(cell, uuid, alist);
304 afs_put_addrlist(alist);
305 return ERR_PTR(-ENOMEM);
308 server = afs_install_server(cell, candidate);
309 if (server != candidate) {
310 afs_put_addrlist(alist);
313 /* Immediately dispatch an asynchronous probe to each interface
314 * on the fileserver. This will make sure the repeat-probing
315 * service is started.
317 afs_fs_probe_fileserver(cell->net, server, key, true);
324 * Set the server timer to fire after a given delay, assuming it's not already
325 * set for an earlier time.
327 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
330 afs_inc_servers_outstanding(net);
331 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
332 afs_dec_servers_outstanding(net);
337 * Server management timer. We have an increment on fs_outstanding that we
338 * need to pass along to the work item.
340 void afs_servers_timer(struct timer_list *timer)
342 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
345 if (!queue_work(afs_wq, &net->fs_manager))
346 afs_dec_servers_outstanding(net);
350 * Get a reference on a server object.
352 struct afs_server *afs_get_server(struct afs_server *server,
353 enum afs_server_trace reason)
355 unsigned int u = atomic_inc_return(&server->ref);
357 trace_afs_server(server, u, atomic_read(&server->active), reason);
362 * Try to get a reference on a server object.
364 static struct afs_server *afs_maybe_use_server(struct afs_server *server,
365 enum afs_server_trace reason)
367 unsigned int r = atomic_fetch_add_unless(&server->ref, 1, 0);
373 a = atomic_inc_return(&server->active);
374 trace_afs_server(server, r, a, reason);
379 * Get an active count on a server object.
381 struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
383 unsigned int r = atomic_inc_return(&server->ref);
384 unsigned int a = atomic_inc_return(&server->active);
386 trace_afs_server(server, r, a, reason);
391 * Release a reference on a server record.
393 void afs_put_server(struct afs_net *net, struct afs_server *server,
394 enum afs_server_trace reason)
401 usage = atomic_dec_return(&server->ref);
402 trace_afs_server(server, usage, atomic_read(&server->active), reason);
403 if (unlikely(usage == 0))
404 __afs_put_server(net, server);
408 * Drop an active count on a server object without updating the last-unused
411 void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
412 enum afs_server_trace reason)
415 unsigned int active = atomic_dec_return(&server->active);
418 afs_set_server_timer(net, afs_server_gc_delay);
419 afs_put_server(net, server, reason);
424 * Drop an active count on a server object.
426 void afs_unuse_server(struct afs_net *net, struct afs_server *server,
427 enum afs_server_trace reason)
430 server->unuse_time = ktime_get_real_seconds();
431 afs_unuse_server_notime(net, server, reason);
435 static void afs_server_rcu(struct rcu_head *rcu)
437 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
439 trace_afs_server(server, atomic_read(&server->ref),
440 atomic_read(&server->active), afs_server_trace_free);
441 afs_put_addrlist(rcu_access_pointer(server->addresses));
445 static void __afs_put_server(struct afs_net *net, struct afs_server *server)
447 call_rcu(&server->rcu, afs_server_rcu);
448 afs_dec_servers_outstanding(net);
451 static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
453 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
454 struct afs_addr_cursor ac = {
456 .index = alist->preferred,
460 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
464 * destroy a dead server
466 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
468 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
469 afs_give_up_callbacks(net, server);
471 flush_work(&server->initcb_work);
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, atomic_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, atomic_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)
632 struct afs_addr_list *alist, *discard;
636 trace_afs_server(server, atomic_read(&server->ref), atomic_read(&server->active),
637 afs_server_trace_update);
639 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
641 if ((PTR_ERR(alist) == -ERESTARTSYS ||
642 PTR_ERR(alist) == -EINTR) &&
643 (op->flags & AFS_OPERATION_UNINTR) &&
645 _leave(" = t [intr]");
648 op->error = PTR_ERR(alist);
649 _leave(" = f [%d]", op->error);
654 if (server->addr_version != alist->version) {
655 write_lock(&server->fs_lock);
656 discard = rcu_dereference_protected(server->addresses,
657 lockdep_is_held(&server->fs_lock));
658 rcu_assign_pointer(server->addresses, alist);
659 server->addr_version = alist->version;
660 write_unlock(&server->fs_lock);
663 afs_put_addrlist(discard);
669 * See if a server's address list needs updating.
671 bool afs_check_server_record(struct afs_operation *op, struct afs_server *server)
674 int ret, retries = 0;
681 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
683 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
685 _leave(" = t [good]");
689 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
690 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
691 success = afs_update_server_record(op, server);
692 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
693 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
694 _leave(" = %d", success);
699 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
700 (op->flags & AFS_OPERATION_UNINTR) ?
701 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
702 if (ret == -ERESTARTSYS) {
704 _leave(" = f [intr]");
710 _leave(" = f [stale]");