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(lock: &net->fs_addr_lock, seq: &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(p: &a->sin6_addr,
50 q: &b->sin6_addr,
51 size: 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(lock: &net->fs_addr_lock, seq));
79
80 done_seqretry(lock: &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(lock: &net->fs_lock, seq: &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(p: uuid, q: &server->uuid, size: 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(lock: &net->fs_lock, seq));
125
126 done_seqretry(lock: &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(sl: &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(p: &candidate->uuid, q: &server->uuid, size: 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(node: &server->uuid_rb, parent: p, rb_link: pp);
186 rb_insert_color(&server->uuid_rb, &net->fs_servers);
187 hlist_add_head_rcu(n: &server->proc_link, h: &net->fs_proc);
188
189added_dup:
190 write_seqlock(sl: &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(n: &server->addr4_link, h: &net->fs_addresses4);
204 if (alist->nr_addrs > alist->nr_ipv4)
205 hlist_add_head_rcu(n: &server->addr6_link, h: &net->fs_addresses6);
206
207 write_sequnlock(sl: &net->fs_addr_lock);
208
209exists:
210 afs_get_server(server, afs_server_trace_get_install);
211 write_sequnlock(sl: &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(size: sizeof(struct afs_server), GFP_KERNEL);
228 if (!server)
229 goto enomem;
230
231 refcount_set(r: &server->ref, n: 1);
232 atomic_set(v: &server->active, i: 1);
233 server->debug_id = atomic_inc_return(v: &afs_server_debug_id);
234 RCU_INIT_POINTER(server->addresses, alist);
235 server->addr_version = alist->version;
236 server->uuid = *uuid;
237 rwlock_init(&server->fs_lock);
238 INIT_WORK(&server->initcb_work, afs_server_init_callback_work);
239 init_waitqueue_head(&server->probe_wq);
240 INIT_LIST_HEAD(list: &server->probe_link);
241 spin_lock_init(&server->probe_lock);
242 server->cell = cell;
243 server->rtt = UINT_MAX;
244
245 afs_inc_servers_outstanding(net);
246 trace_afs_server(server_debug_id: server->debug_id, ref: 1, active: 1, reason: afs_server_trace_alloc);
247 _leave(" = %p", server);
248 return server;
249
250enomem:
251 _leave(" = NULL [nomem]");
252 return NULL;
253}
254
255/*
256 * Look up an address record for a server
257 */
258static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell,
259 struct key *key, const uuid_t *uuid)
260{
261 struct afs_vl_cursor vc;
262 struct afs_addr_list *alist = NULL;
263 int ret;
264
265 ret = -ERESTARTSYS;
266 if (afs_begin_vlserver_operation(&vc, cell, key)) {
267 while (afs_select_vlserver(&vc)) {
268 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags))
269 alist = afs_yfsvl_get_endpoints(&vc, uuid);
270 else
271 alist = afs_vl_get_addrs_u(&vc, uuid);
272 }
273
274 ret = afs_end_vlserver_operation(&vc);
275 }
276
277 return ret < 0 ? ERR_PTR(error: ret) : alist;
278}
279
280/*
281 * Get or create a fileserver record.
282 */
283struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key,
284 const uuid_t *uuid, u32 addr_version)
285{
286 struct afs_addr_list *alist;
287 struct afs_server *server, *candidate;
288
289 _enter("%p,%pU", cell->net, uuid);
290
291 server = afs_find_server_by_uuid(net: cell->net, uuid);
292 if (server) {
293 if (server->addr_version != addr_version)
294 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, addr: &server->flags);
295 return server;
296 }
297
298 alist = afs_vl_lookup_addrs(cell, key, uuid);
299 if (IS_ERR(ptr: alist))
300 return ERR_CAST(ptr: alist);
301
302 candidate = afs_alloc_server(cell, uuid, alist);
303 if (!candidate) {
304 afs_put_addrlist(alist);
305 return ERR_PTR(error: -ENOMEM);
306 }
307
308 server = afs_install_server(cell, candidate);
309 if (server != candidate) {
310 afs_put_addrlist(alist);
311 kfree(objp: candidate);
312 } else {
313 /* Immediately dispatch an asynchronous probe to each interface
314 * on the fileserver. This will make sure the repeat-probing
315 * service is started.
316 */
317 afs_fs_probe_fileserver(cell->net, server, key, true);
318 }
319
320 return server;
321}
322
323/*
324 * Set the server timer to fire after a given delay, assuming it's not already
325 * set for an earlier time.
326 */
327static void afs_set_server_timer(struct afs_net *net, time64_t delay)
328{
329 if (net->live) {
330 afs_inc_servers_outstanding(net);
331 if (timer_reduce(timer: &net->fs_timer, expires: jiffies + delay * HZ))
332 afs_dec_servers_outstanding(net);
333 }
334}
335
336/*
337 * Server management timer. We have an increment on fs_outstanding that we
338 * need to pass along to the work item.
339 */
340void afs_servers_timer(struct timer_list *timer)
341{
342 struct afs_net *net = container_of(timer, struct afs_net, fs_timer);
343
344 _enter("");
345 if (!queue_work(wq: afs_wq, work: &net->fs_manager))
346 afs_dec_servers_outstanding(net);
347}
348
349/*
350 * Get a reference on a server object.
351 */
352struct afs_server *afs_get_server(struct afs_server *server,
353 enum afs_server_trace reason)
354{
355 unsigned int a;
356 int r;
357
358 __refcount_inc(r: &server->ref, oldp: &r);
359 a = atomic_read(v: &server->active);
360 trace_afs_server(server_debug_id: server->debug_id, ref: r + 1, active: a, reason);
361 return server;
362}
363
364/*
365 * Try to get a reference on a server object.
366 */
367static struct afs_server *afs_maybe_use_server(struct afs_server *server,
368 enum afs_server_trace reason)
369{
370 unsigned int a;
371 int r;
372
373 if (!__refcount_inc_not_zero(r: &server->ref, oldp: &r))
374 return NULL;
375
376 a = atomic_inc_return(v: &server->active);
377 trace_afs_server(server_debug_id: server->debug_id, ref: r + 1, active: a, reason);
378 return server;
379}
380
381/*
382 * Get an active count on a server object.
383 */
384struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason)
385{
386 unsigned int a;
387 int r;
388
389 __refcount_inc(r: &server->ref, oldp: &r);
390 a = atomic_inc_return(v: &server->active);
391
392 trace_afs_server(server_debug_id: server->debug_id, ref: r + 1, active: a, reason);
393 return server;
394}
395
396/*
397 * Release a reference on a server record.
398 */
399void afs_put_server(struct afs_net *net, struct afs_server *server,
400 enum afs_server_trace reason)
401{
402 unsigned int a, debug_id = server->debug_id;
403 bool zero;
404 int r;
405
406 if (!server)
407 return;
408
409 a = atomic_read(v: &server->active);
410 zero = __refcount_dec_and_test(r: &server->ref, oldp: &r);
411 trace_afs_server(server_debug_id: debug_id, ref: r - 1, active: a, reason);
412 if (unlikely(zero))
413 __afs_put_server(net, server);
414}
415
416/*
417 * Drop an active count on a server object without updating the last-unused
418 * time.
419 */
420void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server,
421 enum afs_server_trace reason)
422{
423 if (server) {
424 unsigned int active = atomic_dec_return(v: &server->active);
425
426 if (active == 0)
427 afs_set_server_timer(net, delay: afs_server_gc_delay);
428 afs_put_server(net, server, reason);
429 }
430}
431
432/*
433 * Drop an active count on a server object.
434 */
435void afs_unuse_server(struct afs_net *net, struct afs_server *server,
436 enum afs_server_trace reason)
437{
438 if (server) {
439 server->unuse_time = ktime_get_real_seconds();
440 afs_unuse_server_notime(net, server, reason);
441 }
442}
443
444static void afs_server_rcu(struct rcu_head *rcu)
445{
446 struct afs_server *server = container_of(rcu, struct afs_server, rcu);
447
448 trace_afs_server(server_debug_id: server->debug_id, ref: refcount_read(r: &server->ref),
449 active: atomic_read(v: &server->active), reason: afs_server_trace_free);
450 afs_put_addrlist(rcu_access_pointer(server->addresses));
451 kfree(objp: server);
452}
453
454static void __afs_put_server(struct afs_net *net, struct afs_server *server)
455{
456 call_rcu(head: &server->rcu, func: afs_server_rcu);
457 afs_dec_servers_outstanding(net);
458}
459
460static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server)
461{
462 struct afs_addr_list *alist = rcu_access_pointer(server->addresses);
463 struct afs_addr_cursor ac = {
464 .alist = alist,
465 .index = alist->preferred,
466 .error = 0,
467 };
468
469 afs_fs_give_up_all_callbacks(net, server, &ac, NULL);
470}
471
472/*
473 * destroy a dead server
474 */
475static void afs_destroy_server(struct afs_net *net, struct afs_server *server)
476{
477 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags))
478 afs_give_up_callbacks(net, server);
479
480 flush_work(work: &server->initcb_work);
481 afs_put_server(net, server, reason: afs_server_trace_destroy);
482}
483
484/*
485 * Garbage collect any expired servers.
486 */
487static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list)
488{
489 struct afs_server *server, *next, *prev;
490 int active;
491
492 while ((server = gc_list)) {
493 gc_list = server->gc_next;
494
495 write_seqlock(sl: &net->fs_lock);
496
497 active = atomic_read(v: &server->active);
498 if (active == 0) {
499 trace_afs_server(server_debug_id: server->debug_id, ref: refcount_read(r: &server->ref),
500 active, reason: afs_server_trace_gc);
501 next = rcu_dereference_protected(
502 server->uuid_next, lockdep_is_held(&net->fs_lock.lock));
503 prev = server->uuid_prev;
504 if (!prev) {
505 /* The one at the front is in the tree */
506 if (!next) {
507 rb_erase(&server->uuid_rb, &net->fs_servers);
508 } else {
509 rb_replace_node_rcu(victim: &server->uuid_rb,
510 new: &next->uuid_rb,
511 root: &net->fs_servers);
512 next->uuid_prev = NULL;
513 }
514 } else {
515 /* This server is not at the front */
516 rcu_assign_pointer(prev->uuid_next, next);
517 if (next)
518 next->uuid_prev = prev;
519 }
520
521 list_del(entry: &server->probe_link);
522 hlist_del_rcu(n: &server->proc_link);
523 if (!hlist_unhashed(h: &server->addr4_link))
524 hlist_del_rcu(n: &server->addr4_link);
525 if (!hlist_unhashed(h: &server->addr6_link))
526 hlist_del_rcu(n: &server->addr6_link);
527 }
528 write_sequnlock(sl: &net->fs_lock);
529
530 if (active == 0)
531 afs_destroy_server(net, server);
532 }
533}
534
535/*
536 * Manage the records of servers known to be within a network namespace. This
537 * includes garbage collecting unused servers.
538 *
539 * Note also that we were given an increment on net->servers_outstanding by
540 * whoever queued us that we need to deal with before returning.
541 */
542void afs_manage_servers(struct work_struct *work)
543{
544 struct afs_net *net = container_of(work, struct afs_net, fs_manager);
545 struct afs_server *gc_list = NULL;
546 struct rb_node *cursor;
547 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
548 bool purging = !net->live;
549
550 _enter("");
551
552 /* Trawl the server list looking for servers that have expired from
553 * lack of use.
554 */
555 read_seqlock_excl(sl: &net->fs_lock);
556
557 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) {
558 struct afs_server *server =
559 rb_entry(cursor, struct afs_server, uuid_rb);
560 int active = atomic_read(v: &server->active);
561
562 _debug("manage %pU %u", &server->uuid, active);
563
564 if (purging) {
565 trace_afs_server(server_debug_id: server->debug_id, ref: refcount_read(r: &server->ref),
566 active, reason: afs_server_trace_purging);
567 if (active != 0)
568 pr_notice("Can't purge s=%08x\n", server->debug_id);
569 }
570
571 if (active == 0) {
572 time64_t expire_at = server->unuse_time;
573
574 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) &&
575 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags))
576 expire_at += afs_server_gc_delay;
577 if (purging || expire_at <= now) {
578 server->gc_next = gc_list;
579 gc_list = server;
580 } else if (expire_at < next_manage) {
581 next_manage = expire_at;
582 }
583 }
584 }
585
586 read_sequnlock_excl(sl: &net->fs_lock);
587
588 /* Update the timer on the way out. We have to pass an increment on
589 * servers_outstanding in the namespace that we are in to the timer or
590 * the work scheduler.
591 */
592 if (!purging && next_manage < TIME64_MAX) {
593 now = ktime_get_real_seconds();
594
595 if (next_manage - now <= 0) {
596 if (queue_work(wq: afs_wq, work: &net->fs_manager))
597 afs_inc_servers_outstanding(net);
598 } else {
599 afs_set_server_timer(net, delay: next_manage - now);
600 }
601 }
602
603 afs_gc_servers(net, gc_list);
604
605 afs_dec_servers_outstanding(net);
606 _leave(" [%d]", atomic_read(&net->servers_outstanding));
607}
608
609static void afs_queue_server_manager(struct afs_net *net)
610{
611 afs_inc_servers_outstanding(net);
612 if (!queue_work(wq: afs_wq, work: &net->fs_manager))
613 afs_dec_servers_outstanding(net);
614}
615
616/*
617 * Purge list of servers.
618 */
619void afs_purge_servers(struct afs_net *net)
620{
621 _enter("");
622
623 if (del_timer_sync(timer: &net->fs_timer))
624 afs_dec_servers_outstanding(net);
625
626 afs_queue_server_manager(net);
627
628 _debug("wait");
629 atomic_dec(v: &net->servers_outstanding);
630 wait_var_event(&net->servers_outstanding,
631 !atomic_read(&net->servers_outstanding));
632 _leave("");
633}
634
635/*
636 * Get an update for a server's address list.
637 */
638static noinline bool afs_update_server_record(struct afs_operation *op,
639 struct afs_server *server)
640{
641 struct afs_addr_list *alist, *discard;
642
643 _enter("");
644
645 trace_afs_server(server_debug_id: server->debug_id, ref: refcount_read(r: &server->ref),
646 active: atomic_read(v: &server->active),
647 reason: afs_server_trace_update);
648
649 alist = afs_vl_lookup_addrs(cell: op->volume->cell, key: op->key, uuid: &server->uuid);
650 if (IS_ERR(ptr: alist)) {
651 if ((PTR_ERR(ptr: alist) == -ERESTARTSYS ||
652 PTR_ERR(ptr: alist) == -EINTR) &&
653 (op->flags & AFS_OPERATION_UNINTR) &&
654 server->addresses) {
655 _leave(" = t [intr]");
656 return true;
657 }
658 op->error = PTR_ERR(ptr: alist);
659 _leave(" = f [%d]", op->error);
660 return false;
661 }
662
663 discard = alist;
664 if (server->addr_version != alist->version) {
665 write_lock(&server->fs_lock);
666 discard = rcu_dereference_protected(server->addresses,
667 lockdep_is_held(&server->fs_lock));
668 rcu_assign_pointer(server->addresses, alist);
669 server->addr_version = alist->version;
670 write_unlock(&server->fs_lock);
671 }
672
673 afs_put_addrlist(discard);
674 _leave(" = t");
675 return true;
676}
677
678/*
679 * See if a server's address list needs updating.
680 */
681bool afs_check_server_record(struct afs_operation *op, struct afs_server *server)
682{
683 bool success;
684 int ret, retries = 0;
685
686 _enter("");
687
688 ASSERT(server);
689
690retry:
691 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags))
692 goto wait;
693 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags))
694 goto update;
695 _leave(" = t [good]");
696 return true;
697
698update:
699 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, addr: &server->flags)) {
700 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, addr: &server->flags);
701 success = afs_update_server_record(op, server);
702 clear_bit_unlock(AFS_SERVER_FL_UPDATING, addr: &server->flags);
703 wake_up_bit(word: &server->flags, AFS_SERVER_FL_UPDATING);
704 _leave(" = %d", success);
705 return success;
706 }
707
708wait:
709 ret = wait_on_bit(word: &server->flags, AFS_SERVER_FL_UPDATING,
710 mode: (op->flags & AFS_OPERATION_UNINTR) ?
711 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
712 if (ret == -ERESTARTSYS) {
713 op->error = ret;
714 _leave(" = f [intr]");
715 return false;
716 }
717
718 retries++;
719 if (retries == 4) {
720 _leave(" = f [stale]");
721 ret = -ESTALE;
722 return false;
723 }
724 goto retry;
725}
726

source code of linux/fs/afs/server.c