Loading...
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
14static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
15static atomic_t afs_server_debug_id;
16
17static struct afs_server *afs_maybe_use_server(struct afs_server *,
18 enum afs_server_trace);
19static void __afs_put_server(struct afs_net *, struct afs_server *);
20
21/*
22 * Find a server by one of its addresses.
23 */
24struct afs_server *afs_find_server(struct afs_net *net,
25 const struct sockaddr_rxrpc *srx)
26{
27 const struct afs_addr_list *alist;
28 struct afs_server *server = NULL;
29 unsigned int i;
30 int seq = 0, diff;
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 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
39
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);
48 if (diff == 0)
49 diff = memcmp(&a->sin6_addr,
50 &b->sin6_addr,
51 sizeof(struct in6_addr));
52 if (diff == 0)
53 goto found;
54 }
55 }
56 } else {
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);
64 if (diff == 0)
65 diff = ((u32 __force)a->sin_addr.s_addr -
66 (u32 __force)b->sin_addr.s_addr);
67 if (diff == 0)
68 goto found;
69 }
70 }
71 }
72
73 server = NULL;
74 continue;
75 found:
76 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr);
77
78 } while (need_seqretry(&net->fs_addr_lock, seq));
79
80 done_seqretry(&net->fs_addr_lock, seq);
81
82 rcu_read_unlock();
83 return server;
84}
85
86/*
87 * Look up a server by its UUID and mark it active.
88 */
89struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
90{
91 struct afs_server *server = NULL;
92 struct rb_node *p;
93 int diff, seq = 0;
94
95 _enter("%pU", uuid);
96
97 do {
98 /* Unfortunately, rbtree walking doesn't give reliable results
99 * under just the RCU read lock, so we have to check for
100 * changes.
101 */
102 if (server)
103 afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq);
104 server = NULL;
105
106 read_seqbegin_or_lock(&net->fs_lock, &seq);
107
108 p = net->fs_servers.rb_node;
109 while (p) {
110 server = rb_entry(p, struct afs_server, uuid_rb);
111
112 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
113 if (diff < 0) {
114 p = p->rb_left;
115 } else if (diff > 0) {
116 p = p->rb_right;
117 } else {
118 afs_use_server(server, afs_server_trace_get_by_uuid);
119 break;
120 }
121
122 server = NULL;
123 }
124 } while (need_seqretry(&net->fs_lock, seq));
125
126 done_seqretry(&net->fs_lock, seq);
127
128 _leave(" = %p", server);
129 return server;
130}
131
132/*
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
135 * tree.
136 */
137static struct afs_server *afs_install_server(struct afs_cell *cell,
138 struct afs_server *candidate)
139{
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;
144 int diff;
145
146 _enter("%p", candidate);
147
148 write_seqlock(&net->fs_lock);
149
150 /* Firstly install the server in the UUID lookup tree */
151 pp = &net->fs_servers.rb_node;
152 p = NULL;
153 while (*pp) {
154 p = *pp;
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));
158 if (diff < 0) {
159 pp = &(*pp)->rb_left;
160 } else if (diff > 0) {
161 pp = &(*pp)->rb_right;
162 } else {
163 if (server->cell == cell)
164 goto exists;
165
166 /* We have the same UUID representing servers in
167 * different cells. Append the new server to the list.
168 */
169 for (;;) {
170 next = rcu_dereference_protected(
171 server->uuid_next,
172 lockdep_is_held(&net->fs_lock.lock));
173 if (!next)
174 break;
175 server = next;
176 }
177 rcu_assign_pointer(server->uuid_next, candidate);
178 candidate->uuid_prev = server;
179 server = candidate;
180 goto added_dup;
181 }
182 }
183
184 server = candidate;
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);
188
189added_dup:
190 write_seqlock(&net->fs_addr_lock);
191 alist = rcu_dereference_protected(server->addresses,
192 lockdep_is_held(&net->fs_addr_lock.lock));
193
194 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
195 * it in the IPv4 and/or IPv6 reverse-map lists.
196 *
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
200 * intensive.
201 */
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);
206
207 write_sequnlock(&net->fs_addr_lock);
208
209exists:
210 afs_get_server(server, afs_server_trace_get_install);
211 write_sequnlock(&net->fs_lock);
212 return server;
213}
214
215/*
216 * Allocate a new server record and mark it active.
217 */
218static struct afs_server *afs_alloc_server(struct afs_cell *cell,
219 const uuid_t *uuid,
220 struct afs_addr_list *alist)
221{
222 struct afs_server *server;
223 struct afs_net *net = cell->net;
224
225 _enter("");
226
227 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
228 if (!server)
229 goto enomem;
230
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_waitqueue_head(&server->probe_wq);
239 INIT_LIST_HEAD(&server->probe_link);
240 spin_lock_init(&server->probe_lock);
241 server->cell = cell;
242 server->rtt = UINT_MAX;
243
244 afs_inc_servers_outstanding(net);
245 trace_afs_server(server, 1, 1, afs_server_trace_alloc);
246 _leave(" = %p", server);
247 return server;
248
249enomem:
250 _leave(" = NULL [nomem]");
251 return NULL;
252}
253
254/*
255 * Look up an address record for a server
256 */
257static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
258 struct key *key, const uuid_t *uuid)
259{
260 struct afs_vl_cursor vc;
261 struct afs_addr_list *alist = NULL;
262 int ret;
263
264 ret = -ERESTARTSYS;
265 if (afs_begin_vlserver_operation(&vc, cell, key)) {
266 while (afs_select_vlserver(&vc)) {
267 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
268 alist = afs_yfsvl_get_endpoints(&vc, uuid);
269 else
270 alist = afs_vl_get_addrs_u(&vc, uuid);
271 }
272
273 ret = afs_end_vlserver_operation(&vc);
274 }
275
276 return ret < 0 ? ERR_PTR(ret) : alist;
277}
278
279/*
280 * Get or create a fileserver record.
281 */
282struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
283 const uuid_t *uuid, u32 addr_version)
284{
285 struct afs_addr_list *alist;
286 struct afs_server *server, *candidate;
287
288 _enter("%p,%pU", cell->net, uuid);
289
290 server = afs_find_server_by_uuid(cell->net, uuid);
291 if (server) {
292 if (server->addr_version != addr_version)
293 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
294 return server;
295 }
296
297 alist = afs_vl_lookup_addrs(cell, key, uuid);
298 if (IS_ERR(alist))
299 return ERR_CAST(alist);
300
301 candidate = afs_alloc_server(cell, uuid, alist);
302 if (!candidate) {
303 afs_put_addrlist(alist);
304 return ERR_PTR(-ENOMEM);
305 }
306
307 server = afs_install_server(cell, candidate);
308 if (server != candidate) {
309 afs_put_addrlist(alist);
310 kfree(candidate);
311 } else {
312 /* Immediately dispatch an asynchronous probe to each interface
313 * on the fileserver. This will make sure the repeat-probing
314 * service is started.
315 */
316 afs_fs_probe_fileserver(cell->net, server, key, true);
317 }
318
319 return server;
320}
321
322/*
323 * Set the server timer to fire after a given delay, assuming it's not already
324 * set for an earlier time.
325 */
326static void afs_set_server_timer(struct afs_net *net, time64_t delay)
327{
328 if (net->live) {
329 afs_inc_servers_outstanding(net);
330 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
331 afs_dec_servers_outstanding(net);
332 }
333}
334
335/*
336 * Server management timer. We have an increment on fs_outstanding that we
337 * need to pass along to the work item.
338 */
339void afs_servers_timer(struct timer_list *timer)
340{
341 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
342
343 _enter("");
344 if (!queue_work(afs_wq, &net->fs_manager))
345 afs_dec_servers_outstanding(net);
346}
347
348/*
349 * Get a reference on a server object.
350 */
351struct afs_server *afs_get_server(struct afs_server *server,
352 enum afs_server_trace reason)
353{
354 unsigned int u = atomic_inc_return(&server->ref);
355
356 trace_afs_server(server, u, atomic_read(&server->active), reason);
357 return server;
358}
359
360/*
361 * Try to get a reference on a server object.
362 */
363static struct afs_server *afs_maybe_use_server(struct afs_server *server,
364 enum afs_server_trace reason)
365{
366 unsigned int r = atomic_fetch_add_unless(&server->ref, 1, 0);
367 unsigned int a;
368
369 if (r == 0)
370 return NULL;
371
372 a = atomic_inc_return(&server->active);
373 trace_afs_server(server, r, a, reason);
374 return server;
375}
376
377/*
378 * Get an active count on a server object.
379 */
380struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
381{
382 unsigned int r = atomic_inc_return(&server->ref);
383 unsigned int a = atomic_inc_return(&server->active);
384
385 trace_afs_server(server, r, a, reason);
386 return server;
387}
388
389/*
390 * Release a reference on a server record.
391 */
392void afs_put_server(struct afs_net *net, struct afs_server *server,
393 enum afs_server_trace reason)
394{
395 unsigned int usage;
396
397 if (!server)
398 return;
399
400 usage = atomic_dec_return(&server->ref);
401 trace_afs_server(server, usage, atomic_read(&server->active), reason);
402 if (unlikely(usage == 0))
403 __afs_put_server(net, server);
404}
405
406/*
407 * Drop an active count on a server object without updating the last-unused
408 * time.
409 */
410void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
411 enum afs_server_trace reason)
412{
413 if (server) {
414 unsigned int active = atomic_dec_return(&server->active);
415
416 if (active == 0)
417 afs_set_server_timer(net, afs_server_gc_delay);
418 afs_put_server(net, server, reason);
419 }
420}
421
422/*
423 * Drop an active count on a server object.
424 */
425void afs_unuse_server(struct afs_net *net, struct afs_server *server,
426 enum afs_server_trace reason)
427{
428 if (server) {
429 server->unuse_time = ktime_get_real_seconds();
430 afs_unuse_server_notime(net, server, reason);
431 }
432}
433
434static void afs_server_rcu(struct rcu_head *rcu)
435{
436 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
437
438 trace_afs_server(server, atomic_read(&server->ref),
439 atomic_read(&server->active), afs_server_trace_free);
440 afs_put_addrlist(rcu_access_pointer(server->addresses));
441 kfree(server);
442}
443
444static void __afs_put_server(struct afs_net *net, struct afs_server *server)
445{
446 call_rcu(&server->rcu, afs_server_rcu);
447 afs_dec_servers_outstanding(net);
448}
449
450static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
451{
452 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
453 struct afs_addr_cursor ac = {
454 .alist = alist,
455 .index = alist->preferred,
456 .error = 0,
457 };
458
459 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
460}
461
462/*
463 * destroy a dead server
464 */
465static 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 */
476static 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, atomic_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->addr4_link))
513 hlist_del_rcu(&server->addr4_link);
514 if (!hlist_unhashed(&server->addr6_link))
515 hlist_del_rcu(&server->addr6_link);
516 }
517 write_sequnlock(&net->fs_lock);
518
519 if (active == 0)
520 afs_destroy_server(net, server);
521 }
522}
523
524/*
525 * Manage the records of servers known to be within a network namespace. This
526 * includes garbage collecting unused servers.
527 *
528 * Note also that we were given an increment on net->servers_outstanding by
529 * whoever queued us that we need to deal with before returning.
530 */
531void afs_manage_servers(struct work_struct *work)
532{
533 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
534 struct afs_server *gc_list = NULL;
535 struct rb_node *cursor;
536 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
537 bool purging = !net->live;
538
539 _enter("");
540
541 /* Trawl the server list looking for servers that have expired from
542 * lack of use.
543 */
544 read_seqlock_excl(&net->fs_lock);
545
546 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
547 struct afs_server *server =
548 rb_entry(cursor, struct afs_server, uuid_rb);
549 int active = atomic_read(&server->active);
550
551 _debug("manage %pU %u", &server->uuid, active);
552
553 if (purging) {
554 trace_afs_server(server, atomic_read(&server->ref),
555 active, afs_server_trace_purging);
556 if (active != 0)
557 pr_notice("Can't purge s=%08x\n", server->debug_id);
558 }
559
560 if (active == 0) {
561 time64_t expire_at = server->unuse_time;
562
563 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
564 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
565 expire_at += afs_server_gc_delay;
566 if (purging || expire_at <= now) {
567 server->gc_next = gc_list;
568 gc_list = server;
569 } else if (expire_at < next_manage) {
570 next_manage = expire_at;
571 }
572 }
573 }
574
575 read_sequnlock_excl(&net->fs_lock);
576
577 /* Update the timer on the way out. We have to pass an increment on
578 * servers_outstanding in the namespace that we are in to the timer or
579 * the work scheduler.
580 */
581 if (!purging && next_manage < TIME64_MAX) {
582 now = ktime_get_real_seconds();
583
584 if (next_manage - now <= 0) {
585 if (queue_work(afs_wq, &net->fs_manager))
586 afs_inc_servers_outstanding(net);
587 } else {
588 afs_set_server_timer(net, next_manage - now);
589 }
590 }
591
592 afs_gc_servers(net, gc_list);
593
594 afs_dec_servers_outstanding(net);
595 _leave(" [%d]", atomic_read(&net->servers_outstanding));
596}
597
598static void afs_queue_server_manager(struct afs_net *net)
599{
600 afs_inc_servers_outstanding(net);
601 if (!queue_work(afs_wq, &net->fs_manager))
602 afs_dec_servers_outstanding(net);
603}
604
605/*
606 * Purge list of servers.
607 */
608void afs_purge_servers(struct afs_net *net)
609{
610 _enter("");
611
612 if (del_timer_sync(&net->fs_timer))
613 afs_dec_servers_outstanding(net);
614
615 afs_queue_server_manager(net);
616
617 _debug("wait");
618 atomic_dec(&net->servers_outstanding);
619 wait_var_event(&net->servers_outstanding,
620 !atomic_read(&net->servers_outstanding));
621 _leave("");
622}
623
624/*
625 * Get an update for a server's address list.
626 */
627static noinline bool afs_update_server_record(struct afs_operation *op,
628 struct afs_server *server)
629{
630 struct afs_addr_list *alist, *discard;
631
632 _enter("");
633
634 trace_afs_server(server, atomic_read(&server->ref), atomic_read(&server->active),
635 afs_server_trace_update);
636
637 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
638 if (IS_ERR(alist)) {
639 if ((PTR_ERR(alist) == -ERESTARTSYS ||
640 PTR_ERR(alist) == -EINTR) &&
641 (op->flags & AFS_OPERATION_UNINTR) &&
642 server->addresses) {
643 _leave(" = t [intr]");
644 return true;
645 }
646 op->error = PTR_ERR(alist);
647 _leave(" = f [%d]", op->error);
648 return false;
649 }
650
651 discard = alist;
652 if (server->addr_version != alist->version) {
653 write_lock(&server->fs_lock);
654 discard = rcu_dereference_protected(server->addresses,
655 lockdep_is_held(&server->fs_lock));
656 rcu_assign_pointer(server->addresses, alist);
657 server->addr_version = alist->version;
658 write_unlock(&server->fs_lock);
659 }
660
661 afs_put_addrlist(discard);
662 _leave(" = t");
663 return true;
664}
665
666/*
667 * See if a server's address list needs updating.
668 */
669bool afs_check_server_record(struct afs_operation *op, struct afs_server *server)
670{
671 bool success;
672 int ret, retries = 0;
673
674 _enter("");
675
676 ASSERT(server);
677
678retry:
679 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
680 goto wait;
681 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
682 goto update;
683 _leave(" = t [good]");
684 return true;
685
686update:
687 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
688 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
689 success = afs_update_server_record(op, server);
690 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
691 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
692 _leave(" = %d", success);
693 return success;
694 }
695
696wait:
697 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
698 (op->flags & AFS_OPERATION_UNINTR) ?
699 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
700 if (ret == -ERESTARTSYS) {
701 op->error = ret;
702 _leave(" = f [intr]");
703 return false;
704 }
705
706 retries++;
707 if (retries == 4) {
708 _leave(" = f [stale]");
709 ret = -ESTALE;
710 return false;
711 }
712 goto retry;
713}
1/* AFS server record management
2 *
3 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#include <linux/sched.h>
13#include <linux/slab.h>
14#include "afs_fs.h"
15#include "internal.h"
16
17static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */
18static unsigned afs_server_update_delay = 30; /* Time till VLDB recheck in secs */
19
20static void afs_inc_servers_outstanding(struct afs_net *net)
21{
22 atomic_inc(&net->servers_outstanding);
23}
24
25static void afs_dec_servers_outstanding(struct afs_net *net)
26{
27 if (atomic_dec_and_test(&net->servers_outstanding))
28 wake_up_var(&net->servers_outstanding);
29}
30
31/*
32 * Find a server by one of its addresses.
33 */
34struct afs_server *afs_find_server(struct afs_net *net,
35 const struct sockaddr_rxrpc *srx)
36{
37 const struct sockaddr_in6 *a = &srx->transport.sin6, *b;
38 const struct afs_addr_list *alist;
39 struct afs_server *server = NULL;
40 unsigned int i;
41 bool ipv6 = true;
42 int seq = 0, diff;
43
44 if (srx->transport.sin6.sin6_addr.s6_addr32[0] == 0 ||
45 srx->transport.sin6.sin6_addr.s6_addr32[1] == 0 ||
46 srx->transport.sin6.sin6_addr.s6_addr32[2] == htonl(0xffff))
47 ipv6 = false;
48
49 rcu_read_lock();
50
51 do {
52 if (server)
53 afs_put_server(net, server);
54 server = NULL;
55 read_seqbegin_or_lock(&net->fs_addr_lock, &seq);
56
57 if (ipv6) {
58 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) {
59 alist = rcu_dereference(server->addresses);
60 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) {
61 b = &alist->addrs[i].transport.sin6;
62 diff = ((u16 __force)a->sin6_port -
63 (u16 __force)b->sin6_port);
64 if (diff == 0)
65 diff = memcmp(&a->sin6_addr,
66 &b->sin6_addr,
67 sizeof(struct in6_addr));
68 if (diff == 0)
69 goto found;
70 }
71 }
72 } else {
73 hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) {
74 alist = rcu_dereference(server->addresses);
75 for (i = 0; i < alist->nr_ipv4; i++) {
76 b = &alist->addrs[i].transport.sin6;
77 diff = ((u16 __force)a->sin6_port -
78 (u16 __force)b->sin6_port);
79 if (diff == 0)
80 diff = ((u32 __force)a->sin6_addr.s6_addr32[3] -
81 (u32 __force)b->sin6_addr.s6_addr32[3]);
82 if (diff == 0)
83 goto found;
84 }
85 }
86 }
87
88 server = NULL;
89 found:
90 if (server && !atomic_inc_not_zero(&server->usage))
91 server = NULL;
92
93 } while (need_seqretry(&net->fs_addr_lock, seq));
94
95 done_seqretry(&net->fs_addr_lock, seq);
96
97 rcu_read_unlock();
98 return server;
99}
100
101/*
102 * Look up a server by its UUID
103 */
104struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid)
105{
106 struct afs_server *server = NULL;
107 struct rb_node *p;
108 int diff, seq = 0;
109
110 _enter("%pU", uuid);
111
112 do {
113 /* Unfortunately, rbtree walking doesn't give reliable results
114 * under just the RCU read lock, so we have to check for
115 * changes.
116 */
117 if (server)
118 afs_put_server(net, server);
119 server = NULL;
120
121 read_seqbegin_or_lock(&net->fs_lock, &seq);
122
123 p = net->fs_servers.rb_node;
124 while (p) {
125 server = rb_entry(p, struct afs_server, uuid_rb);
126
127 diff = memcmp(uuid, &server->uuid, sizeof(*uuid));
128 if (diff < 0) {
129 p = p->rb_left;
130 } else if (diff > 0) {
131 p = p->rb_right;
132 } else {
133 afs_get_server(server);
134 break;
135 }
136
137 server = NULL;
138 }
139 } while (need_seqretry(&net->fs_lock, seq));
140
141 done_seqretry(&net->fs_lock, seq);
142
143 _leave(" = %p", server);
144 return server;
145}
146
147/*
148 * Install a server record in the namespace tree
149 */
150static struct afs_server *afs_install_server(struct afs_net *net,
151 struct afs_server *candidate)
152{
153 const struct afs_addr_list *alist;
154 struct afs_server *server;
155 struct rb_node **pp, *p;
156 int ret = -EEXIST, diff;
157
158 _enter("%p", candidate);
159
160 write_seqlock(&net->fs_lock);
161
162 /* Firstly install the server in the UUID lookup tree */
163 pp = &net->fs_servers.rb_node;
164 p = NULL;
165 while (*pp) {
166 p = *pp;
167 _debug("- consider %p", p);
168 server = rb_entry(p, struct afs_server, uuid_rb);
169 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t));
170 if (diff < 0)
171 pp = &(*pp)->rb_left;
172 else if (diff > 0)
173 pp = &(*pp)->rb_right;
174 else
175 goto exists;
176 }
177
178 server = candidate;
179 rb_link_node(&server->uuid_rb, p, pp);
180 rb_insert_color(&server->uuid_rb, &net->fs_servers);
181 hlist_add_head_rcu(&server->proc_link, &net->fs_proc);
182
183 write_seqlock(&net->fs_addr_lock);
184 alist = rcu_dereference_protected(server->addresses,
185 lockdep_is_held(&net->fs_addr_lock.lock));
186
187 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
188 * it in the IPv4 and/or IPv6 reverse-map lists.
189 *
190 * TODO: For speed we want to use something other than a flat list
191 * here; even sorting the list in terms of lowest address would help a
192 * bit, but anything we might want to do gets messy and memory
193 * intensive.
194 */
195 if (alist->nr_ipv4 > 0)
196 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4);
197 if (alist->nr_addrs > alist->nr_ipv4)
198 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6);
199
200 write_sequnlock(&net->fs_addr_lock);
201 ret = 0;
202
203exists:
204 afs_get_server(server);
205 write_sequnlock(&net->fs_lock);
206 return server;
207}
208
209/*
210 * allocate a new server record
211 */
212static struct afs_server *afs_alloc_server(struct afs_net *net,
213 const uuid_t *uuid,
214 struct afs_addr_list *alist)
215{
216 struct afs_server *server;
217
218 _enter("");
219
220 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
221 if (!server)
222 goto enomem;
223
224 atomic_set(&server->usage, 1);
225 RCU_INIT_POINTER(server->addresses, alist);
226 server->addr_version = alist->version;
227 server->uuid = *uuid;
228 server->flags = (1UL << AFS_SERVER_FL_NEW);
229 server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
230 rwlock_init(&server->fs_lock);
231 INIT_LIST_HEAD(&server->cb_interests);
232 rwlock_init(&server->cb_break_lock);
233
234 afs_inc_servers_outstanding(net);
235 _leave(" = %p", server);
236 return server;
237
238enomem:
239 _leave(" = NULL [nomem]");
240 return NULL;
241}
242
243/*
244 * Look up an address record for a server
245 */
246static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
247 struct key *key, const uuid_t *uuid)
248{
249 struct afs_addr_cursor ac;
250 struct afs_addr_list *alist;
251 int ret;
252
253 ret = afs_set_vl_cursor(&ac, cell);
254 if (ret < 0)
255 return ERR_PTR(ret);
256
257 while (afs_iterate_addresses(&ac)) {
258 if (test_bit(ac.index, &ac.alist->yfs))
259 alist = afs_yfsvl_get_endpoints(cell->net, &ac, key, uuid);
260 else
261 alist = afs_vl_get_addrs_u(cell->net, &ac, key, uuid);
262 switch (ac.error) {
263 case 0:
264 afs_end_cursor(&ac);
265 return alist;
266 case -ECONNABORTED:
267 ac.error = afs_abort_to_error(ac.abort_code);
268 goto error;
269 case -ENOMEM:
270 case -ENONET:
271 goto error;
272 case -ENETUNREACH:
273 case -EHOSTUNREACH:
274 case -ECONNREFUSED:
275 break;
276 default:
277 ac.error = -EIO;
278 goto error;
279 }
280 }
281
282error:
283 return ERR_PTR(afs_end_cursor(&ac));
284}
285
286/*
287 * Get or create a fileserver record.
288 */
289struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
290 const uuid_t *uuid)
291{
292 struct afs_addr_list *alist;
293 struct afs_server *server, *candidate;
294
295 _enter("%p,%pU", cell->net, uuid);
296
297 server = afs_find_server_by_uuid(cell->net, uuid);
298 if (server)
299 return server;
300
301 alist = afs_vl_lookup_addrs(cell, key, uuid);
302 if (IS_ERR(alist))
303 return ERR_CAST(alist);
304
305 candidate = afs_alloc_server(cell->net, uuid, alist);
306 if (!candidate) {
307 afs_put_addrlist(alist);
308 return ERR_PTR(-ENOMEM);
309 }
310
311 server = afs_install_server(cell->net, candidate);
312 if (server != candidate) {
313 afs_put_addrlist(alist);
314 kfree(candidate);
315 }
316
317 _leave(" = %p{%d}", server, atomic_read(&server->usage));
318 return server;
319}
320
321/*
322 * Set the server timer to fire after a given delay, assuming it's not already
323 * set for an earlier time.
324 */
325static void afs_set_server_timer(struct afs_net *net, time64_t delay)
326{
327 if (net->live) {
328 afs_inc_servers_outstanding(net);
329 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
330 afs_dec_servers_outstanding(net);
331 }
332}
333
334/*
335 * Server management timer. We have an increment on fs_outstanding that we
336 * need to pass along to the work item.
337 */
338void afs_servers_timer(struct timer_list *timer)
339{
340 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
341
342 _enter("");
343 if (!queue_work(afs_wq, &net->fs_manager))
344 afs_dec_servers_outstanding(net);
345}
346
347/*
348 * Release a reference on a server record.
349 */
350void afs_put_server(struct afs_net *net, struct afs_server *server)
351{
352 unsigned int usage;
353
354 if (!server)
355 return;
356
357 server->put_time = ktime_get_real_seconds();
358
359 usage = atomic_dec_return(&server->usage);
360
361 _enter("{%u}", usage);
362
363 if (likely(usage > 0))
364 return;
365
366 afs_set_server_timer(net, afs_server_gc_delay);
367}
368
369static void afs_server_rcu(struct rcu_head *rcu)
370{
371 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
372
373 afs_put_addrlist(rcu_access_pointer(server->addresses));
374 kfree(server);
375}
376
377/*
378 * destroy a dead server
379 */
380static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
381{
382 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
383 struct afs_addr_cursor ac = {
384 .alist = alist,
385 .start = alist->index,
386 .index = 0,
387 .addr = &alist->addrs[alist->index],
388 .error = 0,
389 };
390 _enter("%p", server);
391
392 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
393 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
394
395 call_rcu(&server->rcu, afs_server_rcu);
396 afs_dec_servers_outstanding(net);
397}
398
399/*
400 * Garbage collect any expired servers.
401 */
402static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
403{
404 struct afs_server *server;
405 bool deleted;
406 int usage;
407
408 while ((server = gc_list)) {
409 gc_list = server->gc_next;
410
411 write_seqlock(&net->fs_lock);
412 usage = 1;
413 deleted = atomic_try_cmpxchg(&server->usage, &usage, 0);
414 if (deleted) {
415 rb_erase(&server->uuid_rb, &net->fs_servers);
416 hlist_del_rcu(&server->proc_link);
417 }
418 write_sequnlock(&net->fs_lock);
419
420 if (deleted) {
421 write_seqlock(&net->fs_addr_lock);
422 if (!hlist_unhashed(&server->addr4_link))
423 hlist_del_rcu(&server->addr4_link);
424 if (!hlist_unhashed(&server->addr6_link))
425 hlist_del_rcu(&server->addr6_link);
426 write_sequnlock(&net->fs_addr_lock);
427 afs_destroy_server(net, server);
428 }
429 }
430}
431
432/*
433 * Manage the records of servers known to be within a network namespace. This
434 * includes garbage collecting unused servers.
435 *
436 * Note also that we were given an increment on net->servers_outstanding by
437 * whoever queued us that we need to deal with before returning.
438 */
439void afs_manage_servers(struct work_struct *work)
440{
441 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
442 struct afs_server *gc_list = NULL;
443 struct rb_node *cursor;
444 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
445 bool purging = !net->live;
446
447 _enter("");
448
449 /* Trawl the server list looking for servers that have expired from
450 * lack of use.
451 */
452 read_seqlock_excl(&net->fs_lock);
453
454 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
455 struct afs_server *server =
456 rb_entry(cursor, struct afs_server, uuid_rb);
457 int usage = atomic_read(&server->usage);
458
459 _debug("manage %pU %u", &server->uuid, usage);
460
461 ASSERTCMP(usage, >=, 1);
462 ASSERTIFCMP(purging, usage, ==, 1);
463
464 if (usage == 1) {
465 time64_t expire_at = server->put_time;
466
467 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
468 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
469 expire_at += afs_server_gc_delay;
470 if (purging || expire_at <= now) {
471 server->gc_next = gc_list;
472 gc_list = server;
473 } else if (expire_at < next_manage) {
474 next_manage = expire_at;
475 }
476 }
477 }
478
479 read_sequnlock_excl(&net->fs_lock);
480
481 /* Update the timer on the way out. We have to pass an increment on
482 * servers_outstanding in the namespace that we are in to the timer or
483 * the work scheduler.
484 */
485 if (!purging && next_manage < TIME64_MAX) {
486 now = ktime_get_real_seconds();
487
488 if (next_manage - now <= 0) {
489 if (queue_work(afs_wq, &net->fs_manager))
490 afs_inc_servers_outstanding(net);
491 } else {
492 afs_set_server_timer(net, next_manage - now);
493 }
494 }
495
496 afs_gc_servers(net, gc_list);
497
498 afs_dec_servers_outstanding(net);
499 _leave(" [%d]", atomic_read(&net->servers_outstanding));
500}
501
502static void afs_queue_server_manager(struct afs_net *net)
503{
504 afs_inc_servers_outstanding(net);
505 if (!queue_work(afs_wq, &net->fs_manager))
506 afs_dec_servers_outstanding(net);
507}
508
509/*
510 * Purge list of servers.
511 */
512void afs_purge_servers(struct afs_net *net)
513{
514 _enter("");
515
516 if (del_timer_sync(&net->fs_timer))
517 atomic_dec(&net->servers_outstanding);
518
519 afs_queue_server_manager(net);
520
521 _debug("wait");
522 wait_var_event(&net->servers_outstanding,
523 !atomic_read(&net->servers_outstanding));
524 _leave("");
525}
526
527/*
528 * Probe a fileserver to find its capabilities.
529 *
530 * TODO: Try service upgrade.
531 */
532static bool afs_do_probe_fileserver(struct afs_fs_cursor *fc)
533{
534 _enter("");
535
536 fc->ac.addr = NULL;
537 fc->ac.start = READ_ONCE(fc->ac.alist->index);
538 fc->ac.index = fc->ac.start;
539 fc->ac.error = 0;
540 fc->ac.begun = false;
541
542 while (afs_iterate_addresses(&fc->ac)) {
543 afs_fs_get_capabilities(afs_v2net(fc->vnode), fc->cbi->server,
544 &fc->ac, fc->key);
545 switch (fc->ac.error) {
546 case 0:
547 afs_end_cursor(&fc->ac);
548 set_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags);
549 return true;
550 case -ECONNABORTED:
551 fc->ac.error = afs_abort_to_error(fc->ac.abort_code);
552 goto error;
553 case -ENOMEM:
554 case -ENONET:
555 goto error;
556 case -ENETUNREACH:
557 case -EHOSTUNREACH:
558 case -ECONNREFUSED:
559 case -ETIMEDOUT:
560 case -ETIME:
561 break;
562 default:
563 fc->ac.error = -EIO;
564 goto error;
565 }
566 }
567
568error:
569 afs_end_cursor(&fc->ac);
570 return false;
571}
572
573/*
574 * If we haven't already, try probing the fileserver to get its capabilities.
575 * We try not to instigate parallel probes, but it's possible that the parallel
576 * probes will fail due to authentication failure when ours would succeed.
577 *
578 * TODO: Try sending an anonymous probe if an authenticated probe fails.
579 */
580bool afs_probe_fileserver(struct afs_fs_cursor *fc)
581{
582 bool success;
583 int ret, retries = 0;
584
585 _enter("");
586
587retry:
588 if (test_bit(AFS_SERVER_FL_PROBED, &fc->cbi->server->flags)) {
589 _leave(" = t");
590 return true;
591 }
592
593 if (!test_and_set_bit_lock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags)) {
594 success = afs_do_probe_fileserver(fc);
595 clear_bit_unlock(AFS_SERVER_FL_PROBING, &fc->cbi->server->flags);
596 wake_up_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING);
597 _leave(" = t");
598 return success;
599 }
600
601 _debug("wait");
602 ret = wait_on_bit(&fc->cbi->server->flags, AFS_SERVER_FL_PROBING,
603 TASK_INTERRUPTIBLE);
604 if (ret == -ERESTARTSYS) {
605 fc->ac.error = ret;
606 _leave(" = f [%d]", ret);
607 return false;
608 }
609
610 retries++;
611 if (retries == 4) {
612 fc->ac.error = -ESTALE;
613 _leave(" = f [stale]");
614 return false;
615 }
616 _debug("retry");
617 goto retry;
618}
619
620/*
621 * Get an update for a server's address list.
622 */
623static noinline bool afs_update_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
624{
625 struct afs_addr_list *alist, *discard;
626
627 _enter("");
628
629 alist = afs_vl_lookup_addrs(fc->vnode->volume->cell, fc->key,
630 &server->uuid);
631 if (IS_ERR(alist)) {
632 fc->ac.error = PTR_ERR(alist);
633 _leave(" = f [%d]", fc->ac.error);
634 return false;
635 }
636
637 discard = alist;
638 if (server->addr_version != alist->version) {
639 write_lock(&server->fs_lock);
640 discard = rcu_dereference_protected(server->addresses,
641 lockdep_is_held(&server->fs_lock));
642 rcu_assign_pointer(server->addresses, alist);
643 server->addr_version = alist->version;
644 write_unlock(&server->fs_lock);
645 }
646
647 server->update_at = ktime_get_real_seconds() + afs_server_update_delay;
648 afs_put_addrlist(discard);
649 _leave(" = t");
650 return true;
651}
652
653/*
654 * See if a server's address list needs updating.
655 */
656bool afs_check_server_record(struct afs_fs_cursor *fc, struct afs_server *server)
657{
658 time64_t now = ktime_get_real_seconds();
659 long diff;
660 bool success;
661 int ret, retries = 0;
662
663 _enter("");
664
665 ASSERT(server);
666
667retry:
668 diff = READ_ONCE(server->update_at) - now;
669 if (diff > 0) {
670 _leave(" = t [not now %ld]", diff);
671 return true;
672 }
673
674 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
675 success = afs_update_server_record(fc, server);
676 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
677 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
678 _leave(" = %d", success);
679 return success;
680 }
681
682 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
683 TASK_INTERRUPTIBLE);
684 if (ret == -ERESTARTSYS) {
685 fc->ac.error = ret;
686 _leave(" = f [intr]");
687 return false;
688 }
689
690 retries++;
691 if (retries == 4) {
692 _leave(" = f [stale]");
693 ret = -ESTALE;
694 return false;
695 }
696 goto retry;
697}