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, 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 */
65struct 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 */
113static 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 afs_get_cell(cell, afs_cell_trace_get_server);
167
168added_dup:
169 write_seqlock(&net->fs_addr_lock);
170 estate = rcu_dereference_protected(server->endpoint_state,
171 lockdep_is_held(&net->fs_addr_lock.lock));
172 alist = estate->addresses;
173
174 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install
175 * it in the IPv4 and/or IPv6 reverse-map lists.
176 *
177 * TODO: For speed we want to use something other than a flat list
178 * here; even sorting the list in terms of lowest address would help a
179 * bit, but anything we might want to do gets messy and memory
180 * intensive.
181 */
182 if (alist->nr_addrs > 0)
183 hlist_add_head_rcu(&server->addr_link, &net->fs_addresses);
184
185 write_sequnlock(&net->fs_addr_lock);
186
187exists:
188 afs_get_server(server, afs_server_trace_get_install);
189 write_sequnlock(&net->fs_lock);
190 return server;
191}
192
193/*
194 * Allocate a new server record and mark it active.
195 */
196static struct afs_server *afs_alloc_server(struct afs_cell *cell,
197 const uuid_t *uuid,
198 struct afs_addr_list *alist)
199{
200 struct afs_endpoint_state *estate;
201 struct afs_server *server;
202 struct afs_net *net = cell->net;
203
204 _enter("");
205
206 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL);
207 if (!server)
208 goto enomem;
209
210 estate = kzalloc(sizeof(struct afs_endpoint_state), GFP_KERNEL);
211 if (!estate)
212 goto enomem_server;
213
214 refcount_set(&server->ref, 1);
215 atomic_set(&server->active, 1);
216 server->debug_id = atomic_inc_return(&afs_server_debug_id);
217 server->addr_version = alist->version;
218 server->uuid = *uuid;
219 rwlock_init(&server->fs_lock);
220 INIT_LIST_HEAD(&server->volumes);
221 init_waitqueue_head(&server->probe_wq);
222 INIT_LIST_HEAD(&server->probe_link);
223 spin_lock_init(&server->probe_lock);
224 server->cell = cell;
225 server->rtt = UINT_MAX;
226 server->service_id = FS_SERVICE;
227
228 server->probe_counter = 1;
229 server->probed_at = jiffies - LONG_MAX / 2;
230 refcount_set(&estate->ref, 1);
231 estate->addresses = alist;
232 estate->server_id = server->debug_id;
233 estate->probe_seq = 1;
234 rcu_assign_pointer(server->endpoint_state, estate);
235
236 afs_inc_servers_outstanding(net);
237 trace_afs_server(server->debug_id, 1, 1, afs_server_trace_alloc);
238 trace_afs_estate(estate->server_id, estate->probe_seq, refcount_read(&estate->ref),
239 afs_estate_trace_alloc_server);
240 _leave(" = %p", server);
241 return server;
242
243enomem_server:
244 kfree(server);
245enomem:
246 _leave(" = NULL [nomem]");
247 return NULL;
248}
249
250/*
251 * Look up an address record for a server
252 */
253static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
254 struct key *key, const uuid_t *uuid)
255{
256 struct afs_vl_cursor vc;
257 struct afs_addr_list *alist = NULL;
258 int ret;
259
260 ret = -ERESTARTSYS;
261 if (afs_begin_vlserver_operation(&vc, cell, key)) {
262 while (afs_select_vlserver(&vc)) {
263 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
264 alist = afs_yfsvl_get_endpoints(&vc, uuid);
265 else
266 alist = afs_vl_get_addrs_u(&vc, uuid);
267 }
268
269 ret = afs_end_vlserver_operation(&vc);
270 }
271
272 return ret < 0 ? ERR_PTR(ret) : alist;
273}
274
275/*
276 * Get or create a fileserver record.
277 */
278struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
279 const uuid_t *uuid, u32 addr_version)
280{
281 struct afs_addr_list *alist;
282 struct afs_server *server, *candidate;
283
284 _enter("%p,%pU", cell->net, uuid);
285
286 server = afs_find_server_by_uuid(cell->net, uuid);
287 if (server) {
288 if (server->addr_version != addr_version)
289 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
290 return server;
291 }
292
293 alist = afs_vl_lookup_addrs(cell, key, uuid);
294 if (IS_ERR(alist))
295 return ERR_CAST(alist);
296
297 candidate = afs_alloc_server(cell, uuid, alist);
298 if (!candidate) {
299 afs_put_addrlist(alist, afs_alist_trace_put_server_oom);
300 return ERR_PTR(-ENOMEM);
301 }
302
303 server = afs_install_server(cell, candidate);
304 if (server != candidate) {
305 afs_put_addrlist(alist, afs_alist_trace_put_server_dup);
306 kfree(candidate);
307 } else {
308 /* Immediately dispatch an asynchronous probe to each interface
309 * on the fileserver. This will make sure the repeat-probing
310 * service is started.
311 */
312 afs_fs_probe_fileserver(cell->net, server, alist, key);
313 }
314
315 return server;
316}
317
318/*
319 * Set the server timer to fire after a given delay, assuming it's not already
320 * set for an earlier time.
321 */
322static void afs_set_server_timer(struct afs_net *net, time64_t delay)
323{
324 if (net->live) {
325 afs_inc_servers_outstanding(net);
326 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ))
327 afs_dec_servers_outstanding(net);
328 }
329}
330
331/*
332 * Server management timer. We have an increment on fs_outstanding that we
333 * need to pass along to the work item.
334 */
335void afs_servers_timer(struct timer_list *timer)
336{
337 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
338
339 _enter("");
340 if (!queue_work(afs_wq, &net->fs_manager))
341 afs_dec_servers_outstanding(net);
342}
343
344/*
345 * Get a reference on a server object.
346 */
347struct afs_server *afs_get_server(struct afs_server *server,
348 enum afs_server_trace reason)
349{
350 unsigned int a;
351 int r;
352
353 __refcount_inc(&server->ref, &r);
354 a = atomic_read(&server->active);
355 trace_afs_server(server->debug_id, r + 1, a, reason);
356 return server;
357}
358
359/*
360 * Try to get a reference on a server object.
361 */
362static struct afs_server *afs_maybe_use_server(struct afs_server *server,
363 enum afs_server_trace reason)
364{
365 unsigned int a;
366 int r;
367
368 if (!__refcount_inc_not_zero(&server->ref, &r))
369 return NULL;
370
371 a = atomic_inc_return(&server->active);
372 trace_afs_server(server->debug_id, r + 1, a, reason);
373 return server;
374}
375
376/*
377 * Get an active count on a server object.
378 */
379struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
380{
381 unsigned int a;
382 int r;
383
384 __refcount_inc(&server->ref, &r);
385 a = atomic_inc_return(&server->active);
386
387 trace_afs_server(server->debug_id, r + 1, a, reason);
388 return server;
389}
390
391/*
392 * Release a reference on a server record.
393 */
394void afs_put_server(struct afs_net *net, struct afs_server *server,
395 enum afs_server_trace reason)
396{
397 unsigned int a, debug_id = server->debug_id;
398 bool zero;
399 int r;
400
401 if (!server)
402 return;
403
404 a = atomic_read(&server->active);
405 zero = __refcount_dec_and_test(&server->ref, &r);
406 trace_afs_server(debug_id, r - 1, a, reason);
407 if (unlikely(zero))
408 __afs_put_server(net, server);
409}
410
411/*
412 * Drop an active count on a server object without updating the last-unused
413 * time.
414 */
415void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
416 enum afs_server_trace reason)
417{
418 if (server) {
419 unsigned int active = atomic_dec_return(&server->active);
420
421 if (active == 0)
422 afs_set_server_timer(net, afs_server_gc_delay);
423 afs_put_server(net, server, reason);
424 }
425}
426
427/*
428 * Drop an active count on a server object.
429 */
430void afs_unuse_server(struct afs_net *net, struct afs_server *server,
431 enum afs_server_trace reason)
432{
433 if (server) {
434 server->unuse_time = ktime_get_real_seconds();
435 afs_unuse_server_notime(net, server, reason);
436 }
437}
438
439static void afs_server_rcu(struct rcu_head *rcu)
440{
441 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
442
443 trace_afs_server(server->debug_id, refcount_read(&server->ref),
444 atomic_read(&server->active), afs_server_trace_free);
445 afs_put_endpoint_state(rcu_access_pointer(server->endpoint_state),
446 afs_estate_trace_put_server);
447 afs_put_cell(server->cell, afs_cell_trace_put_server);
448 kfree(server);
449}
450
451static void __afs_put_server(struct afs_net *net, struct afs_server *server)
452{
453 call_rcu(&server->rcu, afs_server_rcu);
454 afs_dec_servers_outstanding(net);
455}
456
457static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
458{
459 struct afs_endpoint_state *estate = rcu_access_pointer(server->endpoint_state);
460 struct afs_addr_list *alist = estate->addresses;
461
462 afs_fs_give_up_all_callbacks(net, server, &alist->addrs[alist->preferred], NULL);
463}
464
465/*
466 * destroy a dead server
467 */
468static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
469{
470 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
471 afs_give_up_callbacks(net, server);
472
473 afs_put_server(net, server, afs_server_trace_destroy);
474}
475
476/*
477 * Garbage collect any expired servers.
478 */
479static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
480{
481 struct afs_server *server, *next, *prev;
482 int active;
483
484 while ((server = gc_list)) {
485 gc_list = server->gc_next;
486
487 write_seqlock(&net->fs_lock);
488
489 active = atomic_read(&server->active);
490 if (active == 0) {
491 trace_afs_server(server->debug_id, refcount_read(&server->ref),
492 active, afs_server_trace_gc);
493 next = rcu_dereference_protected(
494 server->uuid_next, lockdep_is_held(&net->fs_lock.lock));
495 prev = server->uuid_prev;
496 if (!prev) {
497 /* The one at the front is in the tree */
498 if (!next) {
499 rb_erase(&server->uuid_rb, &net->fs_servers);
500 } else {
501 rb_replace_node_rcu(&server->uuid_rb,
502 &next->uuid_rb,
503 &net->fs_servers);
504 next->uuid_prev = NULL;
505 }
506 } else {
507 /* This server is not at the front */
508 rcu_assign_pointer(prev->uuid_next, next);
509 if (next)
510 next->uuid_prev = prev;
511 }
512
513 list_del(&server->probe_link);
514 hlist_del_rcu(&server->proc_link);
515 if (!hlist_unhashed(&server->addr_link))
516 hlist_del_rcu(&server->addr_link);
517 }
518 write_sequnlock(&net->fs_lock);
519
520 if (active == 0)
521 afs_destroy_server(net, server);
522 }
523}
524
525/*
526 * Manage the records of servers known to be within a network namespace. This
527 * includes garbage collecting unused servers.
528 *
529 * Note also that we were given an increment on net->servers_outstanding by
530 * whoever queued us that we need to deal with before returning.
531 */
532void afs_manage_servers(struct work_struct *work)
533{
534 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
535 struct afs_server *gc_list = NULL;
536 struct rb_node *cursor;
537 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
538 bool purging = !net->live;
539
540 _enter("");
541
542 /* Trawl the server list looking for servers that have expired from
543 * lack of use.
544 */
545 read_seqlock_excl(&net->fs_lock);
546
547 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
548 struct afs_server *server =
549 rb_entry(cursor, struct afs_server, uuid_rb);
550 int active = atomic_read(&server->active);
551
552 _debug("manage %pU %u", &server->uuid, active);
553
554 if (purging) {
555 trace_afs_server(server->debug_id, refcount_read(&server->ref),
556 active, afs_server_trace_purging);
557 if (active != 0)
558 pr_notice("Can't purge s=%08x\n", server->debug_id);
559 }
560
561 if (active == 0) {
562 time64_t expire_at = server->unuse_time;
563
564 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
565 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
566 expire_at += afs_server_gc_delay;
567 if (purging || expire_at <= now) {
568 server->gc_next = gc_list;
569 gc_list = server;
570 } else if (expire_at < next_manage) {
571 next_manage = expire_at;
572 }
573 }
574 }
575
576 read_sequnlock_excl(&net->fs_lock);
577
578 /* Update the timer on the way out. We have to pass an increment on
579 * servers_outstanding in the namespace that we are in to the timer or
580 * the work scheduler.
581 */
582 if (!purging && next_manage < TIME64_MAX) {
583 now = ktime_get_real_seconds();
584
585 if (next_manage - now <= 0) {
586 if (queue_work(afs_wq, &net->fs_manager))
587 afs_inc_servers_outstanding(net);
588 } else {
589 afs_set_server_timer(net, next_manage - now);
590 }
591 }
592
593 afs_gc_servers(net, gc_list);
594
595 afs_dec_servers_outstanding(net);
596 _leave(" [%d]", atomic_read(&net->servers_outstanding));
597}
598
599static void afs_queue_server_manager(struct afs_net *net)
600{
601 afs_inc_servers_outstanding(net);
602 if (!queue_work(afs_wq, &net->fs_manager))
603 afs_dec_servers_outstanding(net);
604}
605
606/*
607 * Purge list of servers.
608 */
609void afs_purge_servers(struct afs_net *net)
610{
611 _enter("");
612
613 if (del_timer_sync(&net->fs_timer))
614 afs_dec_servers_outstanding(net);
615
616 afs_queue_server_manager(net);
617
618 _debug("wait");
619 atomic_dec(&net->servers_outstanding);
620 wait_var_event(&net->servers_outstanding,
621 !atomic_read(&net->servers_outstanding));
622 _leave("");
623}
624
625/*
626 * Get an update for a server's address list.
627 */
628static noinline bool afs_update_server_record(struct afs_operation *op,
629 struct afs_server *server,
630 struct key *key)
631{
632 struct afs_endpoint_state *estate;
633 struct afs_addr_list *alist;
634 bool has_addrs;
635
636 _enter("");
637
638 trace_afs_server(server->debug_id, refcount_read(&server->ref),
639 atomic_read(&server->active),
640 afs_server_trace_update);
641
642 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid);
643 if (IS_ERR(alist)) {
644 rcu_read_lock();
645 estate = rcu_dereference(server->endpoint_state);
646 has_addrs = estate->addresses;
647 rcu_read_unlock();
648
649 if ((PTR_ERR(alist) == -ERESTARTSYS ||
650 PTR_ERR(alist) == -EINTR) &&
651 (op->flags & AFS_OPERATION_UNINTR) &&
652 has_addrs) {
653 _leave(" = t [intr]");
654 return true;
655 }
656 afs_op_set_error(op, PTR_ERR(alist));
657 _leave(" = f [%d]", afs_op_error(op));
658 return false;
659 }
660
661 if (server->addr_version != alist->version)
662 afs_fs_probe_fileserver(op->net, server, alist, key);
663
664 afs_put_addrlist(alist, afs_alist_trace_put_server_update);
665 _leave(" = t");
666 return true;
667}
668
669/*
670 * See if a server's address list needs updating.
671 */
672bool afs_check_server_record(struct afs_operation *op, struct afs_server *server,
673 struct key *key)
674{
675 bool success;
676 int ret, retries = 0;
677
678 _enter("");
679
680 ASSERT(server);
681
682retry:
683 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
684 goto wait;
685 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
686 goto update;
687 _leave(" = t [good]");
688 return true;
689
690update:
691 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) {
692 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags);
693 success = afs_update_server_record(op, server, key);
694 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags);
695 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING);
696 _leave(" = %d", success);
697 return success;
698 }
699
700wait:
701 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING,
702 (op->flags & AFS_OPERATION_UNINTR) ?
703 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
704 if (ret == -ERESTARTSYS) {
705 afs_op_set_error(op, ret);
706 _leave(" = f [intr]");
707 return false;
708 }
709
710 retries++;
711 if (retries == 4) {
712 _leave(" = f [stale]");
713 ret = -ESTALE;
714 return false;
715 }
716 goto retry;
717}
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}