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_addr_list *alist;
27 struct afs_server *server = NULL;
35 afs_unuse_server_notime(net, server, afs_server_trace_put_find_rsq);
37 seq++; /* 2 on the 1st/lockless path, otherwise odd */
38 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
40 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
41 alist = rcu_dereference(server->addresses);
42 for (i = 0; i < alist->nr_addrs; i++)
43 if (alist->addrs[i].peer == peer)
50 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
52 } while (need_seqretry(&net->fs_addr_lock, seq));
54 done_seqretry(&net->fs_addr_lock, seq);
61 * Look up a server by its UUID and mark it active.
63 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
65 struct afs_server *server = NULL;
72 /* Unfortunately, rbtree walking doesn't give reliable results
73 * under just the RCU read lock, so we have to check for
77 afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
79 seq++; /* 2 on the 1st/lockless path, otherwise odd */
80 read_seqbegin_or_lock(&net->fs_lock, &seq);
82 p = net->fs_servers.rb_node;
84 server = rb_entry(p, struct afs_server, uuid_rb);
86 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
89 } else if (diff > 0) {
92 afs_use_server(server, afs_server_trace_get_by_uuid);
98 } while (need_seqretry(&net->fs_lock, seq));
100 done_seqretry(&net->fs_lock, seq);
102 _leave(" = %p", server);
107 * Install a server record in the namespace tree. If there's a clash, we stick
108 * it into a list anchored on whichever afs_server struct is actually in the
111 static struct afs_server *afs_install_server(struct afs_cell *cell,
112 struct afs_server *candidate)
114 const struct afs_addr_list *alist;
115 struct afs_server *server, *next;
116 struct afs_net *net = cell->net;
117 struct rb_node **pp, *p;
120 _enter("%p", candidate);
122 write_seqlock(&net->fs_lock);
124 /* Firstly install the server in the UUID lookup tree */
125 pp = &net->fs_servers.rb_node;
129 _debug("- consider %p", p);
130 server = rb_entry(p, struct afs_server, uuid_rb);
131 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
133 pp = &(*pp)->rb_left;
134 } else if (diff > 0) {
135 pp = &(*pp)->rb_right;
137 if (server->cell == cell)
140 /* We have the same UUID representing servers in
141 * different cells. Append the new server to the list.
144 next = rcu_dereference_protected(
146 lockdep_is_held(&net->fs_lock.lock));
151 rcu_assign_pointer(server->uuid_next, candidate);
152 candidate->uuid_prev = server;
159 rb_link_node(&server->uuid_rb, p, pp);
160 rb_insert_color(&server->uuid_rb, &net->fs_servers);
161 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
164 write_seqlock(&net->fs_addr_lock);
165 alist = rcu_dereference_protected(server->addresses,
166 lockdep_is_held(&net->fs_addr_lock.lock));
168 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
169 * it in the IPv4 and/or IPv6 reverse-map lists.
171 * TODO: For speed we want to use something other than a flat list
172 * here; even sorting the list in terms of lowest address would help a
173 * bit, but anything we might want to do gets messy and memory
176 if (alist->nr_ipv4 > 0)
177 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
178 if (alist->nr_addrs > alist->nr_ipv4)
179 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
181 write_sequnlock(&net->fs_addr_lock);
184 afs_get_server(server, afs_server_trace_get_install);
185 write_sequnlock(&net->fs_lock);
190 * Allocate a new server record and mark it active.
192 static struct afs_server *afs_alloc_server(struct afs_cell *cell,
194 struct afs_addr_list *alist)
196 struct afs_server *server;
197 struct afs_net *net = cell->net;
201 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
205 refcount_set(&server->ref, 1);
206 atomic_set(&server->active, 1);
207 server->debug_id = atomic_inc_return(&afs_server_debug_id);
208 RCU_INIT_POINTER(server->addresses, alist);
209 server->addr_version = alist->version;
210 server->uuid = *uuid;
211 rwlock_init(&server->fs_lock);
212 INIT_WORK(&server->initcb_work, afs_server_init_callback_work);
213 init_waitqueue_head(&server->probe_wq);
214 INIT_LIST_HEAD(&server->probe_link);
215 spin_lock_init(&server->probe_lock);
217 server->rtt = UINT_MAX;
219 afs_inc_servers_outstanding(net);
220 trace_afs_server(server->debug_id, 1, 1, afs_server_trace_alloc);
221 _leave(" = %p", server);
225 _leave(" = NULL [nomem]");
230 * Look up an address record for a server
232 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
233 struct key *key, const uuid_t *uuid)
235 struct afs_vl_cursor vc;
236 struct afs_addr_list *alist = NULL;
240 if (afs_begin_vlserver_operation(&vc, cell, key)) {
241 while (afs_select_vlserver(&vc)) {
242 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
243 alist = afs_yfsvl_get_endpoints(&vc, uuid);
245 alist = afs_vl_get_addrs_u(&vc, uuid);
248 ret = afs_end_vlserver_operation(&vc);
251 return ret < 0 ? ERR_PTR(ret) : alist;
255 * Get or create a fileserver record.
257 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
258 const uuid_t *uuid, u32 addr_version)
260 struct afs_addr_list *alist;
261 struct afs_server *server, *candidate;
263 _enter("%p,%pU", cell->net, uuid);
265 server = afs_find_server_by_uuid(cell->net, uuid);
267 if (server->addr_version != addr_version)
268 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
272 alist = afs_vl_lookup_addrs(cell, key, uuid);
274 return ERR_CAST(alist);
276 candidate = afs_alloc_server(cell, uuid, alist);
278 afs_put_addrlist(alist);
279 return ERR_PTR(-ENOMEM);
282 server = afs_install_server(cell, candidate);
283 if (server != candidate) {
284 afs_put_addrlist(alist);
287 /* Immediately dispatch an asynchronous probe to each interface
288 * on the fileserver. This will make sure the repeat-probing
289 * service is started.
291 afs_fs_probe_fileserver(cell->net, server, key, true);
298 * Set the server timer to fire after a given delay, assuming it's not already
299 * set for an earlier time.
301 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
304 afs_inc_servers_outstanding(net);
305 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
306 afs_dec_servers_outstanding(net);
311 * Server management timer. We have an increment on fs_outstanding that we
312 * need to pass along to the work item.
314 void afs_servers_timer(struct timer_list *timer)
316 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
319 if (!queue_work(afs_wq, &net->fs_manager))
320 afs_dec_servers_outstanding(net);
324 * Get a reference on a server object.
326 struct afs_server *afs_get_server(struct afs_server *server,
327 enum afs_server_trace reason)
332 __refcount_inc(&server->ref, &r);
333 a = atomic_read(&server->active);
334 trace_afs_server(server->debug_id, r + 1, a, reason);
339 * Try to get a reference on a server object.
341 static struct afs_server *afs_maybe_use_server(struct afs_server *server,
342 enum afs_server_trace reason)
347 if (!__refcount_inc_not_zero(&server->ref, &r))
350 a = atomic_inc_return(&server->active);
351 trace_afs_server(server->debug_id, r + 1, a, reason);
356 * Get an active count on a server object.
358 struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
363 __refcount_inc(&server->ref, &r);
364 a = atomic_inc_return(&server->active);
366 trace_afs_server(server->debug_id, r + 1, a, reason);
371 * Release a reference on a server record.
373 void afs_put_server(struct afs_net *net, struct afs_server *server,
374 enum afs_server_trace reason)
376 unsigned int a, debug_id = server->debug_id;
383 a = atomic_read(&server->active);
384 zero = __refcount_dec_and_test(&server->ref, &r);
385 trace_afs_server(debug_id, r - 1, a, reason);
387 __afs_put_server(net, server);
391 * Drop an active count on a server object without updating the last-unused
394 void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
395 enum afs_server_trace reason)
398 unsigned int active = atomic_dec_return(&server->active);
401 afs_set_server_timer(net, afs_server_gc_delay);
402 afs_put_server(net, server, reason);
407 * Drop an active count on a server object.
409 void afs_unuse_server(struct afs_net *net, struct afs_server *server,
410 enum afs_server_trace reason)
413 server->unuse_time = ktime_get_real_seconds();
414 afs_unuse_server_notime(net, server, reason);
418 static void afs_server_rcu(struct rcu_head *rcu)
420 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
422 trace_afs_server(server->debug_id, refcount_read(&server->ref),
423 atomic_read(&server->active), afs_server_trace_free);
424 afs_put_addrlist(rcu_access_pointer(server->addresses));
428 static void __afs_put_server(struct afs_net *net, struct afs_server *server)
430 call_rcu(&server->rcu, afs_server_rcu);
431 afs_dec_servers_outstanding(net);
434 static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
436 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
437 struct afs_addr_cursor ac = {
439 .index = alist->preferred,
442 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
446 * destroy a dead server
448 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
450 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
451 afs_give_up_callbacks(net, server);
453 flush_work(&server->initcb_work);
454 afs_put_server(net, server, afs_server_trace_destroy);
458 * Garbage collect any expired servers.
460 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
462 struct afs_server *server, *next, *prev;
465 while ((server = gc_list)) {
466 gc_list = server->gc_next;
468 write_seqlock(&net->fs_lock);
470 active = atomic_read(&server->active);
472 trace_afs_server(server->debug_id, refcount_read(&server->ref),
473 active, afs_server_trace_gc);
474 next = rcu_dereference_protected(
475 server->uuid_next, lockdep_is_held(&net->fs_lock.lock));
476 prev = server->uuid_prev;
478 /* The one at the front is in the tree */
480 rb_erase(&server->uuid_rb, &net->fs_servers);
482 rb_replace_node_rcu(&server->uuid_rb,
485 next->uuid_prev = NULL;
488 /* This server is not at the front */
489 rcu_assign_pointer(prev->uuid_next, next);
491 next->uuid_prev = prev;
494 list_del(&server->probe_link);
495 hlist_del_rcu(&server->proc_link);
496 if (!hlist_unhashed(&server->addr4_link))
497 hlist_del_rcu(&server->addr4_link);
498 if (!hlist_unhashed(&server->addr6_link))
499 hlist_del_rcu(&server->addr6_link);
501 write_sequnlock(&net->fs_lock);
504 afs_destroy_server(net, server);
509 * Manage the records of servers known to be within a network namespace. This
510 * includes garbage collecting unused servers.
512 * Note also that we were given an increment on net->servers_outstanding by
513 * whoever queued us that we need to deal with before returning.
515 void afs_manage_servers(struct work_struct *work)
517 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
518 struct afs_server *gc_list = NULL;
519 struct rb_node *cursor;
520 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
521 bool purging = !net->live;
525 /* Trawl the server list looking for servers that have expired from
528 read_seqlock_excl(&net->fs_lock);
530 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
531 struct afs_server *server =
532 rb_entry(cursor, struct afs_server, uuid_rb);
533 int active = atomic_read(&server->active);
535 _debug("manage %pU %u", &server->uuid, active);
538 trace_afs_server(server->debug_id, refcount_read(&server->ref),
539 active, afs_server_trace_purging);
541 pr_notice("Can't purge s=%08x\n", server->debug_id);
545 time64_t expire_at = server->unuse_time;
547 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
548 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
549 expire_at += afs_server_gc_delay;
550 if (purging || expire_at <= now) {
551 server->gc_next = gc_list;
553 } else if (expire_at < next_manage) {
554 next_manage = expire_at;
559 read_sequnlock_excl(&net->fs_lock);
561 /* Update the timer on the way out. We have to pass an increment on
562 * servers_outstanding in the namespace that we are in to the timer or
563 * the work scheduler.
565 if (!purging && next_manage < TIME64_MAX) {
566 now = ktime_get_real_seconds();
568 if (next_manage - now <= 0) {
569 if (queue_work(afs_wq, &net->fs_manager))
570 afs_inc_servers_outstanding(net);
572 afs_set_server_timer(net, next_manage - now);
576 afs_gc_servers(net, gc_list);
578 afs_dec_servers_outstanding(net);
579 _leave(" [%d]", atomic_read(&net->servers_outstanding));
582 static void afs_queue_server_manager(struct afs_net *net)
584 afs_inc_servers_outstanding(net);
585 if (!queue_work(afs_wq, &net->fs_manager))
586 afs_dec_servers_outstanding(net);
590 * Purge list of servers.
592 void afs_purge_servers(struct afs_net *net)
596 if (del_timer_sync(&net->fs_timer))
597 afs_dec_servers_outstanding(net);
599 afs_queue_server_manager(net);
602 atomic_dec(&net->servers_outstanding);
603 wait_var_event(&net->servers_outstanding,
604 !atomic_read(&net->servers_outstanding));
609 * Get an update for a server's address list.
611 static noinline bool afs_update_server_record(struct afs_operation *op,
612 struct afs_server *server)
614 struct afs_addr_list *alist, *discard;
618 trace_afs_server(server->debug_id, refcount_read(&server->ref),
619 atomic_read(&server->active),
620 afs_server_trace_update);
622 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
624 if ((PTR_ERR(alist) == -ERESTARTSYS ||
625 PTR_ERR(alist) == -EINTR) &&
626 (op->flags & AFS_OPERATION_UNINTR) &&
628 _leave(" = t [intr]");
631 afs_op_set_error(op, PTR_ERR(alist));
632 _leave(" = f [%d]", afs_op_error(op));
637 if (server->addr_version != alist->version) {
638 write_lock(&server->fs_lock);
639 discard = rcu_dereference_protected(server->addresses,
640 lockdep_is_held(&server->fs_lock));
641 rcu_assign_pointer(server->addresses, alist);
642 server->addr_version = alist->version;
643 write_unlock(&server->fs_lock);
646 afs_put_addrlist(discard);
652 * See if a server's address list needs updating.
654 bool afs_check_server_record(struct afs_operation *op, struct afs_server *server)
657 int ret, retries = 0;
664 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
666 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
668 _leave(" = t [good]");
672 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
673 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
674 success = afs_update_server_record(op, server);
675 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
676 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
677 _leave(" = %d", success);
682 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
683 (op->flags & AFS_OPERATION_UNINTR) ?
684 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
685 if (ret == -ERESTARTSYS) {
686 afs_op_set_error(op, ret);
687 _leave(" = f [intr]");
693 _leave(" = f [stale]");