1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS server record management
3 *
4 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8 #include <linux/sched.h>
9 #include <linux/slab.h>
10 #include "afs_fs.h"
11 #include "internal.h"
12 #include "protocol_yfs.h"
13
14 static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
15 static atomic_t afs_server_debug_id;
16
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 *);
20
21 /*
22 * Find a server by one of its addresses.
23 */
24 struct afs_server *afs_find_server(struct afs_net *net, const struct rxrpc_peer *peer)
25 {
26 const struct afs_endpoint_state *estate;
27 const struct afs_addr_list *alist;
28 struct afs_server *server = NULL;
29 unsigned int i;
30 int seq = 1;
31
32 rcu_read_lock();
33
34 do {
35 if (server)
36 afs_unuse_server_notime(net, server, afs_server_trace_put_find_rsq);
37 server = NULL;
38 seq++; /* 2 on the 1st/lockless path, otherwise odd */
39 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
40
41 hlist_for_each_entry_rcu(server, &net->fs_addresses, addr_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)
46 goto found;
47 }
48
49 server = NULL;
50 continue;
51 found:
52 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
53
54 } while (need_seqretry(&net->fs_addr_lock, seq));
55
56 done_seqretry(&net->fs_addr_lock, seq);
57
58 rcu_read_unlock();
59 return server;
60 }
61
62 /*
63 * Look up a server by its UUID and mark it active.
64 */
65 struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
66 {
67 struct afs_server *server = NULL;
68 struct rb_node *p;
69 int diff, seq = 1;
70
71 _enter("%pU", uuid);
72
73 do {
74 /* Unfortunately, rbtree walking doesn't give reliable results
75 * under just the RCU read lock, so we have to check for
76 * changes.
77 */
78 if (server)
79 afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
80 server = NULL;
81 seq++; /* 2 on the 1st/lockless path, otherwise odd */
82 read_seqbegin_or_lock(&net->fs_lock, &seq);
83
84 p = net->fs_servers.rb_node;
85 while (p) {
86 server = rb_entry(p, struct afs_server, uuid_rb);
87
88 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
89 if (diff < 0) {
90 p = p->rb_left;
91 } else if (diff > 0) {
92 p = p->rb_right;
93 } else {
94 afs_use_server(server, afs_server_trace_get_by_uuid);
95 break;
96 }
97
98 server = NULL;
99 }
100 } while (need_seqretry(&net->fs_lock, seq));
101
102 done_seqretry(&net->fs_lock, seq);
103
104 _leave(" = %p", server);
105 return server;
106 }
107
108 /*
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
111 * tree.
112 */
113 static struct afs_server *afs_install_server(struct afs_cell *cell,
114 struct afs_server *candidate)
115 {
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;
121 int diff;
122
123 _enter("%p", candidate);
124
125 write_seqlock(&net->fs_lock);
126
127 /* Firstly install the server in the UUID lookup tree */
128 pp = &net->fs_servers.rb_node;
129 p = NULL;
130 while (*pp) {
131 p = *pp;
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));
135 if (diff < 0) {
136 pp = &(*pp)->rb_left;
137 } else if (diff > 0) {
138 pp = &(*pp)->rb_right;
139 } else {
140 if (server->cell == cell)
141 goto exists;
142
143 /* We have the same UUID representing servers in
144 * different cells. Append the new server to the list.
145 */
146 for (;;) {
147 next = rcu_dereference_protected(
148 server->uuid_next,
149 lockdep_is_held(&net->fs_lock.lock));
150 if (!next)
151 break;
152 server = next;
153 }
154 rcu_assign_pointer(server->uuid_next, candidate);
155 candidate->uuid_prev = server;
156 server = candidate;
157 goto added_dup;
158 }
159 }
160
161 server = candidate;
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);
165
166 added_dup:
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;
171
172 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
173 * it in the IPv4 and/or IPv6 reverse-map lists.
174 *
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
178 * intensive.
179 */
180 if (alist->nr_addrs > 0)
181 hlist_add_head_rcu(&server->addr_link, &net->fs_addresses);
182
183 write_sequnlock(&net->fs_addr_lock);
184
185 exists:
186 afs_get_server(server, afs_server_trace_get_install);
187 write_sequnlock(&net->fs_lock);
188 return server;
189 }
190
191 /*
192 * Allocate a new server record and mark it active.
193 */
194 static struct afs_server *afs_alloc_server(struct afs_cell *cell,
195 const uuid_t *uuid,
196 struct afs_addr_list *alist)
197 {
198 struct afs_endpoint_state *estate;
199 struct afs_server *server;
200 struct afs_net *net = cell->net;
201
202 _enter("");
203
204 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
205 if (!server)
206 goto enomem;
207
208 estate = kzalloc(sizeof(struct afs_endpoint_state), GFP_KERNEL);
209 if (!estate)
210 goto enomem_server;
211
212 refcount_set(&server->ref, 1);
213 atomic_set(&server->active, 1);
214 server->debug_id = atomic_inc_return(&afs_server_debug_id);
215 server->addr_version = alist->version;
216 server->uuid = *uuid;
217 rwlock_init(&server->fs_lock);
218 INIT_LIST_HEAD(&server->volumes);
219 init_waitqueue_head(&server->probe_wq);
220 INIT_LIST_HEAD(&server->probe_link);
221 spin_lock_init(&server->probe_lock);
222 server->cell = cell;
223 server->rtt = UINT_MAX;
224 server->service_id = FS_SERVICE;
225
226 server->probe_counter = 1;
227 server->probed_at = jiffies - LONG_MAX / 2;
228 refcount_set(&estate->ref, 1);
229 estate->addresses = alist;
230 estate->server_id = server->debug_id;
231 estate->probe_seq = 1;
232 rcu_assign_pointer(server->endpoint_state, estate);
233
234 afs_inc_servers_outstanding(net);
235 trace_afs_server(server->debug_id, 1, 1, afs_server_trace_alloc);
236 trace_afs_estate(estate->server_id, estate->probe_seq, refcount_read(&estate->ref),
237 afs_estate_trace_alloc_server);
238 _leave(" = %p", server);
239 return server;
240
241 enomem_server:
242 kfree(server);
243 enomem:
244 _leave(" = NULL [nomem]");
245 return NULL;
246 }
247
248 /*
249 * Look up an address record for a server
250 */
251 static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
252 struct key *key, const uuid_t *uuid)
253 {
254 struct afs_vl_cursor vc;
255 struct afs_addr_list *alist = NULL;
256 int ret;
257
258 ret = -ERESTARTSYS;
259 if (afs_begin_vlserver_operation(&vc, cell, key)) {
260 while (afs_select_vlserver(&vc)) {
261 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
262 alist = afs_yfsvl_get_endpoints(&vc, uuid);
263 else
264 alist = afs_vl_get_addrs_u(&vc, uuid);
265 }
266
267 ret = afs_end_vlserver_operation(&vc);
268 }
269
270 return ret < 0 ? ERR_PTR(ret) : alist;
271 }
272
273 /*
274 * Get or create a fileserver record.
275 */
276 struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
277 const uuid_t *uuid, u32 addr_version)
278 {
279 struct afs_addr_list *alist;
280 struct afs_server *server, *candidate;
281
282 _enter("%p,%pU", cell->net, uuid);
283
284 server = afs_find_server_by_uuid(cell->net, uuid);
285 if (server) {
286 if (server->addr_version != addr_version)
287 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
288 return server;
289 }
290
291 alist = afs_vl_lookup_addrs(cell, key, uuid);
292 if (IS_ERR(alist))
293 return ERR_CAST(alist);
294
295 candidate = afs_alloc_server(cell, uuid, alist);
296 if (!candidate) {
297 afs_put_addrlist(alist, afs_alist_trace_put_server_oom);
298 return ERR_PTR(-ENOMEM);
299 }
300
301 server = afs_install_server(cell, candidate);
302 if (server != candidate) {
303 afs_put_addrlist(alist, afs_alist_trace_put_server_dup);
304 kfree(candidate);
305 } else {
306 /* Immediately dispatch an asynchronous probe to each interface
307 * on the fileserver. This will make sure the repeat-probing
308 * service is started.
309 */
310 afs_fs_probe_fileserver(cell->net, server, alist, key);
311 }
312
313 return server;
314 }
315
316 /*
317 * Set the server timer to fire after a given delay, assuming it's not already
318 * set for an earlier time.
319 */
320 static void afs_set_server_timer(struct afs_net *net, time64_t delay)
321 {
322 if (net->live) {
323 afs_inc_servers_outstanding(net);
324 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
325 afs_dec_servers_outstanding(net);
326 }
327 }
328
329 /*
330 * Server management timer. We have an increment on fs_outstanding that we
331 * need to pass along to the work item.
332 */
333 void afs_servers_timer(struct timer_list *timer)
334 {
335 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
336
337 _enter("");
338 if (!queue_work(afs_wq, &net->fs_manager))
339 afs_dec_servers_outstanding(net);
340 }
341
342 /*
343 * Get a reference on a server object.
344 */
345 struct afs_server *afs_get_server(struct afs_server *server,
346 enum afs_server_trace reason)
347 {
348 unsigned int a;
349 int r;
350
351 __refcount_inc(&server->ref, &r);
352 a = atomic_read(&server->active);
353 trace_afs_server(server->debug_id, r + 1, a, reason);
354 return server;
355 }
356
357 /*
358 * Try to get a reference on a server object.
359 */
360 static struct afs_server *afs_maybe_use_server(struct afs_server *server,
361 enum afs_server_trace reason)
362 {
363 unsigned int a;
364 int r;
365
366 if (!__refcount_inc_not_zero(&server->ref, &r))
367 return NULL;
368
369 a = atomic_inc_return(&server->active);
370 trace_afs_server(server->debug_id, r + 1, a, reason);
371 return server;
372 }
373
374 /*
375 * Get an active count on a server object.
376 */
377 struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
378 {
379 unsigned int a;
380 int r;
381
382 __refcount_inc(&server->ref, &r);
383 a = atomic_inc_return(&server->active);
384
385 trace_afs_server(server->debug_id, r + 1, a, reason);
386 return server;
387 }
388
389 /*
390 * Release a reference on a server record.
391 */
392 void afs_put_server(struct afs_net *net, struct afs_server *server,
393 enum afs_server_trace reason)
394 {
395 unsigned int a, debug_id = server->debug_id;
396 bool zero;
397 int r;
398
399 if (!server)
400 return;
401
402 a = atomic_read(&server->active);
403 zero = __refcount_dec_and_test(&server->ref, &r);
404 trace_afs_server(debug_id, r - 1, a, reason);
405 if (unlikely(zero))
406 __afs_put_server(net, server);
407 }
408
409 /*
410 * Drop an active count on a server object without updating the last-unused
411 * time.
412 */
413 void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
414 enum afs_server_trace reason)
415 {
416 if (server) {
417 unsigned int active = atomic_dec_return(&server->active);
418
419 if (active == 0)
420 afs_set_server_timer(net, afs_server_gc_delay);
421 afs_put_server(net, server, reason);
422 }
423 }
424
425 /*
426 * Drop an active count on a server object.
427 */
428 void afs_unuse_server(struct afs_net *net, struct afs_server *server,
429 enum afs_server_trace reason)
430 {
431 if (server) {
432 server->unuse_time = ktime_get_real_seconds();
433 afs_unuse_server_notime(net, server, reason);
434 }
435 }
436
437 static void afs_server_rcu(struct rcu_head *rcu)
438 {
439 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
440
441 trace_afs_server(server->debug_id, refcount_read(&server->ref),
442 atomic_read(&server->active), afs_server_trace_free);
443 afs_put_endpoint_state(rcu_access_pointer(server->endpoint_state),
444 afs_estate_trace_put_server);
445 kfree(server);
446 }
447
448 static void __afs_put_server(struct afs_net *net, struct afs_server *server)
449 {
450 call_rcu(&server->rcu, afs_server_rcu);
451 afs_dec_servers_outstanding(net);
452 }
453
454 static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
455 {
456 struct afs_endpoint_state *estate = rcu_access_pointer(server->endpoint_state);
457 struct afs_addr_list *alist = estate->addresses;
458
459 afs_fs_give_up_all_callbacks(net, server, &alist->addrs[alist->preferred], NULL);
460 }
461
462 /*
463 * destroy a dead server
464 */
465 static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
466 {
467 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
468 afs_give_up_callbacks(net, server);
469
470 afs_put_server(net, server, afs_server_trace_destroy);
471 }
472
473 /*
474 * Garbage collect any expired servers.
475 */
476 static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
477 {
478 struct afs_server *server, *next, *prev;
479 int active;
480
481 while ((server = gc_list)) {
482 gc_list = server->gc_next;
483
484 write_seqlock(&net->fs_lock);
485
486 active = atomic_read(&server->active);
487 if (active == 0) {
488 trace_afs_server(server->debug_id, refcount_read(&server->ref),
489 active, afs_server_trace_gc);
490 next = rcu_dereference_protected(
491 server->uuid_next, lockdep_is_held(&net->fs_lock.lock));
492 prev = server->uuid_prev;
493 if (!prev) {
494 /* The one at the front is in the tree */
495 if (!next) {
496 rb_erase(&server->uuid_rb, &net->fs_servers);
497 } else {
498 rb_replace_node_rcu(&server->uuid_rb,
499 &next->uuid_rb,
500 &net->fs_servers);
501 next->uuid_prev = NULL;
502 }
503 } else {
504 /* This server is not at the front */
505 rcu_assign_pointer(prev->uuid_next, next);
506 if (next)
507 next->uuid_prev = prev;
508 }
509
510 list_del(&server->probe_link);
511 hlist_del_rcu(&server->proc_link);
512 if (!hlist_unhashed(&server->addr_link))
513 hlist_del_rcu(&server->addr_link);
514 }
515 write_sequnlock(&net->fs_lock);
516
517 if (active == 0)
518 afs_destroy_server(net, server);
519 }
520 }
521
522 /*
523 * Manage the records of servers known to be within a network namespace. This
524 * includes garbage collecting unused servers.
525 *
526 * Note also that we were given an increment on net->servers_outstanding by
527 * whoever queued us that we need to deal with before returning.
528 */
529 void afs_manage_servers(struct work_struct *work)
530 {
531 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
532 struct afs_server *gc_list = NULL;
533 struct rb_node *cursor;
534 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
535 bool purging = !net->live;
536
537 _enter("");
538
539 /* Trawl the server list looking for servers that have expired from
540 * lack of use.
541 */
542 read_seqlock_excl(&net->fs_lock);
543
544 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
545 struct afs_server *server =
546 rb_entry(cursor, struct afs_server, uuid_rb);
547 int active = atomic_read(&server->active);
548
549 _debug("manage %pU %u", &server->uuid, active);
550
551 if (purging) {
552 trace_afs_server(server->debug_id, refcount_read(&server->ref),
553 active, afs_server_trace_purging);
554 if (active != 0)
555 pr_notice("Can't purge s=%08x\n", server->debug_id);
556 }
557
558 if (active == 0) {
559 time64_t expire_at = server->unuse_time;
560
561 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
562 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
563 expire_at += afs_server_gc_delay;
564 if (purging || expire_at <= now) {
565 server->gc_next = gc_list;
566 gc_list = server;
567 } else if (expire_at < next_manage) {
568 next_manage = expire_at;
569 }
570 }
571 }
572
573 read_sequnlock_excl(&net->fs_lock);
574
575 /* Update the timer on the way out. We have to pass an increment on
576 * servers_outstanding in the namespace that we are in to the timer or
577 * the work scheduler.
578 */
579 if (!purging && next_manage < TIME64_MAX) {
580 now = ktime_get_real_seconds();
581
582 if (next_manage - now <= 0) {
583 if (queue_work(afs_wq, &net->fs_manager))
584 afs_inc_servers_outstanding(net);
585 } else {
586 afs_set_server_timer(net, next_manage - now);
587 }
588 }
589
590 afs_gc_servers(net, gc_list);
591
592 afs_dec_servers_outstanding(net);
593 _leave(" [%d]", atomic_read(&net->servers_outstanding));
594 }
595
596 static void afs_queue_server_manager(struct afs_net *net)
597 {
598 afs_inc_servers_outstanding(net);
599 if (!queue_work(afs_wq, &net->fs_manager))
600 afs_dec_servers_outstanding(net);
601 }
602
603 /*
604 * Purge list of servers.
605 */
606 void afs_purge_servers(struct afs_net *net)
607 {
608 _enter("");
609
610 if (del_timer_sync(&net->fs_timer))
611 afs_dec_servers_outstanding(net);
612
613 afs_queue_server_manager(net);
614
615 _debug("wait");
616 atomic_dec(&net->servers_outstanding);
617 wait_var_event(&net->servers_outstanding,
618 !atomic_read(&net->servers_outstanding));
619 _leave("");
620 }
621
622 /*
623 * Get an update for a server's address list.
624 */
625 static noinline bool afs_update_server_record(struct afs_operation *op,
626 struct afs_server *server,
627 struct key *key)
628 {
629 struct afs_endpoint_state *estate;
630 struct afs_addr_list *alist;
631 bool has_addrs;
632
633 _enter("");
634
635 trace_afs_server(server->debug_id, refcount_read(&server->ref),
636 atomic_read(&server->active),
637 afs_server_trace_update);
638
639 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
640 if (IS_ERR(alist)) {
641 rcu_read_lock();
642 estate = rcu_dereference(server->endpoint_state);
643 has_addrs = estate->addresses;
644 rcu_read_unlock();
645
646 if ((PTR_ERR(alist) == -ERESTARTSYS ||
647 PTR_ERR(alist) == -EINTR) &&
648 (op->flags & AFS_OPERATION_UNINTR) &&
649 has_addrs) {
650 _leave(" = t [intr]");
651 return true;
652 }
653 afs_op_set_error(op, PTR_ERR(alist));
654 _leave(" = f [%d]", afs_op_error(op));
655 return false;
656 }
657
658 if (server->addr_version != alist->version)
659 afs_fs_probe_fileserver(op->net, server, alist, key);
660
661 afs_put_addrlist(alist, afs_alist_trace_put_server_update);
662 _leave(" = t");
663 return true;
664 }
665
666 /*
667 * See if a server's address list needs updating.
668 */
669 bool afs_check_server_record(struct afs_operation *op, struct afs_server *server,
670 struct key *key)
671 {
672 bool success;
673 int ret, retries = 0;
674
675 _enter("");
676
677 ASSERT(server);
678
679 retry:
680 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
681 goto wait;
682 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
683 goto update;
684 _leave(" = t [good]");
685 return true;
686
687 update:
688 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
689 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
690 success = afs_update_server_record(op, server, key);
691 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
692 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
693 _leave(" = %d", success);
694 return success;
695 }
696
697 wait:
698 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
699 (op->flags & AFS_OPERATION_UNINTR) ?
700 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
701 if (ret == -ERESTARTSYS) {
702 afs_op_set_error(op, ret);
703 _leave(" = f [intr]");
704 return false;
705 }
706
707 retries++;
708 if (retries == 4) {
709 _leave(" = f [stale]");
710 ret = -ESTALE;
711 return false;
712 }
713 goto retry;
714 }
715
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.