Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* AFS cell and server record management
3 *
4 * Copyright (C) 2002, 2017 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/slab.h>
9#include <linux/key.h>
10#include <linux/ctype.h>
11#include <linux/dns_resolver.h>
12#include <linux/sched.h>
13#include <linux/inet.h>
14#include <linux/namei.h>
15#include <keys/rxrpc-type.h>
16#include "internal.h"
17
18static unsigned __read_mostly afs_cell_gc_delay = 10;
19static unsigned __read_mostly afs_cell_min_ttl = 10 * 60;
20static unsigned __read_mostly afs_cell_max_ttl = 24 * 60 * 60;
21static atomic_t cell_debug_id;
22
23static void afs_queue_cell_manager(struct afs_net *);
24static void afs_manage_cell_work(struct work_struct *);
25
26static void afs_dec_cells_outstanding(struct afs_net *net)
27{
28 if (atomic_dec_and_test(&net->cells_outstanding))
29 wake_up_var(&net->cells_outstanding);
30}
31
32/*
33 * Set the cell timer to fire after a given delay, assuming it's not already
34 * set for an earlier time.
35 */
36static void afs_set_cell_timer(struct afs_net *net, time64_t delay)
37{
38 if (net->live) {
39 atomic_inc(&net->cells_outstanding);
40 if (timer_reduce(&net->cells_timer, jiffies + delay * HZ))
41 afs_dec_cells_outstanding(net);
42 } else {
43 afs_queue_cell_manager(net);
44 }
45}
46
47/*
48 * Look up and get an activation reference on a cell record. The caller must
49 * hold net->cells_lock at least read-locked.
50 */
51static struct afs_cell *afs_find_cell_locked(struct afs_net *net,
52 const char *name, unsigned int namesz,
53 enum afs_cell_trace reason)
54{
55 struct afs_cell *cell = NULL;
56 struct rb_node *p;
57 int n;
58
59 _enter("%*.*s", namesz, namesz, name);
60
61 if (name && namesz == 0)
62 return ERR_PTR(-EINVAL);
63 if (namesz > AFS_MAXCELLNAME)
64 return ERR_PTR(-ENAMETOOLONG);
65
66 if (!name) {
67 cell = net->ws_cell;
68 if (!cell)
69 return ERR_PTR(-EDESTADDRREQ);
70 goto found;
71 }
72
73 p = net->cells.rb_node;
74 while (p) {
75 cell = rb_entry(p, struct afs_cell, net_node);
76
77 n = strncasecmp(cell->name, name,
78 min_t(size_t, cell->name_len, namesz));
79 if (n == 0)
80 n = cell->name_len - namesz;
81 if (n < 0)
82 p = p->rb_left;
83 else if (n > 0)
84 p = p->rb_right;
85 else
86 goto found;
87 }
88
89 return ERR_PTR(-ENOENT);
90
91found:
92 return afs_use_cell(cell, reason);
93}
94
95/*
96 * Look up and get an activation reference on a cell record.
97 */
98struct afs_cell *afs_find_cell(struct afs_net *net,
99 const char *name, unsigned int namesz,
100 enum afs_cell_trace reason)
101{
102 struct afs_cell *cell;
103
104 down_read(&net->cells_lock);
105 cell = afs_find_cell_locked(net, name, namesz, reason);
106 up_read(&net->cells_lock);
107 return cell;
108}
109
110/*
111 * Set up a cell record and fill in its name, VL server address list and
112 * allocate an anonymous key
113 */
114static struct afs_cell *afs_alloc_cell(struct afs_net *net,
115 const char *name, unsigned int namelen,
116 const char *addresses)
117{
118 struct afs_vlserver_list *vllist;
119 struct afs_cell *cell;
120 int i, ret;
121
122 ASSERT(name);
123 if (namelen == 0)
124 return ERR_PTR(-EINVAL);
125 if (namelen > AFS_MAXCELLNAME) {
126 _leave(" = -ENAMETOOLONG");
127 return ERR_PTR(-ENAMETOOLONG);
128 }
129
130 /* Prohibit cell names that contain unprintable chars, '/' and '@' or
131 * that begin with a dot. This also precludes "@cell".
132 */
133 if (name[0] == '.')
134 return ERR_PTR(-EINVAL);
135 for (i = 0; i < namelen; i++) {
136 char ch = name[i];
137 if (!isprint(ch) || ch == '/' || ch == '@')
138 return ERR_PTR(-EINVAL);
139 }
140
141 _enter("%*.*s,%s", namelen, namelen, name, addresses);
142
143 cell = kzalloc(sizeof(struct afs_cell), GFP_KERNEL);
144 if (!cell) {
145 _leave(" = -ENOMEM");
146 return ERR_PTR(-ENOMEM);
147 }
148
149 cell->name = kmalloc(namelen + 1, GFP_KERNEL);
150 if (!cell->name) {
151 kfree(cell);
152 return ERR_PTR(-ENOMEM);
153 }
154
155 cell->net = net;
156 cell->name_len = namelen;
157 for (i = 0; i < namelen; i++)
158 cell->name[i] = tolower(name[i]);
159 cell->name[i] = 0;
160
161 atomic_set(&cell->ref, 1);
162 atomic_set(&cell->active, 0);
163 INIT_WORK(&cell->manager, afs_manage_cell_work);
164 cell->volumes = RB_ROOT;
165 INIT_HLIST_HEAD(&cell->proc_volumes);
166 seqlock_init(&cell->volume_lock);
167 cell->fs_servers = RB_ROOT;
168 seqlock_init(&cell->fs_lock);
169 rwlock_init(&cell->vl_servers_lock);
170 cell->flags = (1 << AFS_CELL_FL_CHECK_ALIAS);
171
172 /* Provide a VL server list, filling it in if we were given a list of
173 * addresses to use.
174 */
175 if (addresses) {
176 vllist = afs_parse_text_addrs(net,
177 addresses, strlen(addresses), ':',
178 VL_SERVICE, AFS_VL_PORT);
179 if (IS_ERR(vllist)) {
180 ret = PTR_ERR(vllist);
181 goto parse_failed;
182 }
183
184 vllist->source = DNS_RECORD_FROM_CONFIG;
185 vllist->status = DNS_LOOKUP_NOT_DONE;
186 cell->dns_expiry = TIME64_MAX;
187 } else {
188 ret = -ENOMEM;
189 vllist = afs_alloc_vlserver_list(0);
190 if (!vllist)
191 goto error;
192 vllist->source = DNS_RECORD_UNAVAILABLE;
193 vllist->status = DNS_LOOKUP_NOT_DONE;
194 cell->dns_expiry = ktime_get_real_seconds();
195 }
196
197 rcu_assign_pointer(cell->vl_servers, vllist);
198
199 cell->dns_source = vllist->source;
200 cell->dns_status = vllist->status;
201 smp_store_release(&cell->dns_lookup_count, 1); /* vs source/status */
202 atomic_inc(&net->cells_outstanding);
203 cell->debug_id = atomic_inc_return(&cell_debug_id);
204 trace_afs_cell(cell->debug_id, 1, 0, afs_cell_trace_alloc);
205
206 _leave(" = %p", cell);
207 return cell;
208
209parse_failed:
210 if (ret == -EINVAL)
211 printk(KERN_ERR "kAFS: bad VL server IP address\n");
212error:
213 kfree(cell->name);
214 kfree(cell);
215 _leave(" = %d", ret);
216 return ERR_PTR(ret);
217}
218
219/*
220 * afs_lookup_cell - Look up or create a cell record.
221 * @net: The network namespace
222 * @name: The name of the cell.
223 * @namesz: The strlen of the cell name.
224 * @vllist: A colon/comma separated list of numeric IP addresses or NULL.
225 * @excl: T if an error should be given if the cell name already exists.
226 *
227 * Look up a cell record by name and query the DNS for VL server addresses if
228 * needed. Note that that actual DNS query is punted off to the manager thread
229 * so that this function can return immediately if interrupted whilst allowing
230 * cell records to be shared even if not yet fully constructed.
231 */
232struct afs_cell *afs_lookup_cell(struct afs_net *net,
233 const char *name, unsigned int namesz,
234 const char *vllist, bool excl)
235{
236 struct afs_cell *cell, *candidate, *cursor;
237 struct rb_node *parent, **pp;
238 enum afs_cell_state state;
239 int ret, n;
240
241 _enter("%s,%s", name, vllist);
242
243 if (!excl) {
244 cell = afs_find_cell(net, name, namesz, afs_cell_trace_use_lookup);
245 if (!IS_ERR(cell))
246 goto wait_for_cell;
247 }
248
249 /* Assume we're probably going to create a cell and preallocate and
250 * mostly set up a candidate record. We can then use this to stash the
251 * name, the net namespace and VL server addresses.
252 *
253 * We also want to do this before we hold any locks as it may involve
254 * upcalling to userspace to make DNS queries.
255 */
256 candidate = afs_alloc_cell(net, name, namesz, vllist);
257 if (IS_ERR(candidate)) {
258 _leave(" = %ld", PTR_ERR(candidate));
259 return candidate;
260 }
261
262 /* Find the insertion point and check to see if someone else added a
263 * cell whilst we were allocating.
264 */
265 down_write(&net->cells_lock);
266
267 pp = &net->cells.rb_node;
268 parent = NULL;
269 while (*pp) {
270 parent = *pp;
271 cursor = rb_entry(parent, struct afs_cell, net_node);
272
273 n = strncasecmp(cursor->name, name,
274 min_t(size_t, cursor->name_len, namesz));
275 if (n == 0)
276 n = cursor->name_len - namesz;
277 if (n < 0)
278 pp = &(*pp)->rb_left;
279 else if (n > 0)
280 pp = &(*pp)->rb_right;
281 else
282 goto cell_already_exists;
283 }
284
285 cell = candidate;
286 candidate = NULL;
287 atomic_set(&cell->active, 2);
288 trace_afs_cell(cell->debug_id, atomic_read(&cell->ref), 2, afs_cell_trace_insert);
289 rb_link_node_rcu(&cell->net_node, parent, pp);
290 rb_insert_color(&cell->net_node, &net->cells);
291 up_write(&net->cells_lock);
292
293 afs_queue_cell(cell, afs_cell_trace_get_queue_new);
294
295wait_for_cell:
296 trace_afs_cell(cell->debug_id, atomic_read(&cell->ref), atomic_read(&cell->active),
297 afs_cell_trace_wait);
298 _debug("wait_for_cell");
299 wait_var_event(&cell->state,
300 ({
301 state = smp_load_acquire(&cell->state); /* vs error */
302 state == AFS_CELL_ACTIVE || state == AFS_CELL_REMOVED;
303 }));
304
305 /* Check the state obtained from the wait check. */
306 if (state == AFS_CELL_REMOVED) {
307 ret = cell->error;
308 goto error;
309 }
310
311 _leave(" = %p [cell]", cell);
312 return cell;
313
314cell_already_exists:
315 _debug("cell exists");
316 cell = cursor;
317 if (excl) {
318 ret = -EEXIST;
319 } else {
320 afs_use_cell(cursor, afs_cell_trace_use_lookup);
321 ret = 0;
322 }
323 up_write(&net->cells_lock);
324 if (candidate)
325 afs_put_cell(candidate, afs_cell_trace_put_candidate);
326 if (ret == 0)
327 goto wait_for_cell;
328 goto error_noput;
329error:
330 afs_unuse_cell(net, cell, afs_cell_trace_unuse_lookup);
331error_noput:
332 _leave(" = %d [error]", ret);
333 return ERR_PTR(ret);
334}
335
336/*
337 * set the root cell information
338 * - can be called with a module parameter string
339 * - can be called from a write to /proc/fs/afs/rootcell
340 */
341int afs_cell_init(struct afs_net *net, const char *rootcell)
342{
343 struct afs_cell *old_root, *new_root;
344 const char *cp, *vllist;
345 size_t len;
346
347 _enter("");
348
349 if (!rootcell) {
350 /* module is loaded with no parameters, or built statically.
351 * - in the future we might initialize cell DB here.
352 */
353 _leave(" = 0 [no root]");
354 return 0;
355 }
356
357 cp = strchr(rootcell, ':');
358 if (!cp) {
359 _debug("kAFS: no VL server IP addresses specified");
360 vllist = NULL;
361 len = strlen(rootcell);
362 } else {
363 vllist = cp + 1;
364 len = cp - rootcell;
365 }
366
367 /* allocate a cell record for the root cell */
368 new_root = afs_lookup_cell(net, rootcell, len, vllist, false);
369 if (IS_ERR(new_root)) {
370 _leave(" = %ld", PTR_ERR(new_root));
371 return PTR_ERR(new_root);
372 }
373
374 if (!test_and_set_bit(AFS_CELL_FL_NO_GC, &new_root->flags))
375 afs_use_cell(new_root, afs_cell_trace_use_pin);
376
377 /* install the new cell */
378 down_write(&net->cells_lock);
379 afs_see_cell(new_root, afs_cell_trace_see_ws);
380 old_root = net->ws_cell;
381 net->ws_cell = new_root;
382 up_write(&net->cells_lock);
383
384 afs_unuse_cell(net, old_root, afs_cell_trace_unuse_ws);
385 _leave(" = 0");
386 return 0;
387}
388
389/*
390 * Update a cell's VL server address list from the DNS.
391 */
392static int afs_update_cell(struct afs_cell *cell)
393{
394 struct afs_vlserver_list *vllist, *old = NULL, *p;
395 unsigned int min_ttl = READ_ONCE(afs_cell_min_ttl);
396 unsigned int max_ttl = READ_ONCE(afs_cell_max_ttl);
397 time64_t now, expiry = 0;
398 int ret = 0;
399
400 _enter("%s", cell->name);
401
402 vllist = afs_dns_query(cell, &expiry);
403 if (IS_ERR(vllist)) {
404 ret = PTR_ERR(vllist);
405
406 _debug("%s: fail %d", cell->name, ret);
407 if (ret == -ENOMEM)
408 goto out_wake;
409
410 ret = -ENOMEM;
411 vllist = afs_alloc_vlserver_list(0);
412 if (!vllist)
413 goto out_wake;
414
415 switch (ret) {
416 case -ENODATA:
417 case -EDESTADDRREQ:
418 vllist->status = DNS_LOOKUP_GOT_NOT_FOUND;
419 break;
420 case -EAGAIN:
421 case -ECONNREFUSED:
422 vllist->status = DNS_LOOKUP_GOT_TEMP_FAILURE;
423 break;
424 default:
425 vllist->status = DNS_LOOKUP_GOT_LOCAL_FAILURE;
426 break;
427 }
428 }
429
430 _debug("%s: got list %d %d", cell->name, vllist->source, vllist->status);
431 cell->dns_status = vllist->status;
432
433 now = ktime_get_real_seconds();
434 if (min_ttl > max_ttl)
435 max_ttl = min_ttl;
436 if (expiry < now + min_ttl)
437 expiry = now + min_ttl;
438 else if (expiry > now + max_ttl)
439 expiry = now + max_ttl;
440
441 _debug("%s: status %d", cell->name, vllist->status);
442 if (vllist->source == DNS_RECORD_UNAVAILABLE) {
443 switch (vllist->status) {
444 case DNS_LOOKUP_GOT_NOT_FOUND:
445 /* The DNS said that the cell does not exist or there
446 * weren't any addresses to be had.
447 */
448 cell->dns_expiry = expiry;
449 break;
450
451 case DNS_LOOKUP_BAD:
452 case DNS_LOOKUP_GOT_LOCAL_FAILURE:
453 case DNS_LOOKUP_GOT_TEMP_FAILURE:
454 case DNS_LOOKUP_GOT_NS_FAILURE:
455 default:
456 cell->dns_expiry = now + 10;
457 break;
458 }
459 } else {
460 cell->dns_expiry = expiry;
461 }
462
463 /* Replace the VL server list if the new record has servers or the old
464 * record doesn't.
465 */
466 write_lock(&cell->vl_servers_lock);
467 p = rcu_dereference_protected(cell->vl_servers, true);
468 if (vllist->nr_servers > 0 || p->nr_servers == 0) {
469 rcu_assign_pointer(cell->vl_servers, vllist);
470 cell->dns_source = vllist->source;
471 old = p;
472 }
473 write_unlock(&cell->vl_servers_lock);
474 afs_put_vlserverlist(cell->net, old);
475
476out_wake:
477 smp_store_release(&cell->dns_lookup_count,
478 cell->dns_lookup_count + 1); /* vs source/status */
479 wake_up_var(&cell->dns_lookup_count);
480 _leave(" = %d", ret);
481 return ret;
482}
483
484/*
485 * Destroy a cell record
486 */
487static void afs_cell_destroy(struct rcu_head *rcu)
488{
489 struct afs_cell *cell = container_of(rcu, struct afs_cell, rcu);
490 struct afs_net *net = cell->net;
491 int u;
492
493 _enter("%p{%s}", cell, cell->name);
494
495 u = atomic_read(&cell->ref);
496 ASSERTCMP(u, ==, 0);
497 trace_afs_cell(cell->debug_id, u, atomic_read(&cell->active), afs_cell_trace_free);
498
499 afs_put_vlserverlist(net, rcu_access_pointer(cell->vl_servers));
500 afs_unuse_cell(net, cell->alias_of, afs_cell_trace_unuse_alias);
501 key_put(cell->anonymous_key);
502 kfree(cell->name);
503 kfree(cell);
504
505 afs_dec_cells_outstanding(net);
506 _leave(" [destroyed]");
507}
508
509/*
510 * Queue the cell manager.
511 */
512static void afs_queue_cell_manager(struct afs_net *net)
513{
514 int outstanding = atomic_inc_return(&net->cells_outstanding);
515
516 _enter("%d", outstanding);
517
518 if (!queue_work(afs_wq, &net->cells_manager))
519 afs_dec_cells_outstanding(net);
520}
521
522/*
523 * Cell management timer. We have an increment on cells_outstanding that we
524 * need to pass along to the work item.
525 */
526void afs_cells_timer(struct timer_list *timer)
527{
528 struct afs_net *net = container_of(timer, struct afs_net, cells_timer);
529
530 _enter("");
531 if (!queue_work(afs_wq, &net->cells_manager))
532 afs_dec_cells_outstanding(net);
533}
534
535/*
536 * Get a reference on a cell record.
537 */
538struct afs_cell *afs_get_cell(struct afs_cell *cell, enum afs_cell_trace reason)
539{
540 int u;
541
542 if (atomic_read(&cell->ref) <= 0)
543 BUG();
544
545 u = atomic_inc_return(&cell->ref);
546 trace_afs_cell(cell->debug_id, u, atomic_read(&cell->active), reason);
547 return cell;
548}
549
550/*
551 * Drop a reference on a cell record.
552 */
553void afs_put_cell(struct afs_cell *cell, enum afs_cell_trace reason)
554{
555 if (cell) {
556 unsigned int debug_id = cell->debug_id;
557 unsigned int u, a;
558
559 a = atomic_read(&cell->active);
560 u = atomic_dec_return(&cell->ref);
561 trace_afs_cell(debug_id, u, a, reason);
562 if (u == 0) {
563 a = atomic_read(&cell->active);
564 WARN(a != 0, "Cell active count %u > 0\n", a);
565 call_rcu(&cell->rcu, afs_cell_destroy);
566 }
567 }
568}
569
570/*
571 * Note a cell becoming more active.
572 */
573struct afs_cell *afs_use_cell(struct afs_cell *cell, enum afs_cell_trace reason)
574{
575 int u, a;
576
577 if (atomic_read(&cell->ref) <= 0)
578 BUG();
579
580 u = atomic_read(&cell->ref);
581 a = atomic_inc_return(&cell->active);
582 trace_afs_cell(cell->debug_id, u, a, reason);
583 return cell;
584}
585
586/*
587 * Record a cell becoming less active. When the active counter reaches 1, it
588 * is scheduled for destruction, but may get reactivated.
589 */
590void afs_unuse_cell(struct afs_net *net, struct afs_cell *cell, enum afs_cell_trace reason)
591{
592 unsigned int debug_id;
593 time64_t now, expire_delay;
594 int u, a;
595
596 if (!cell)
597 return;
598
599 _enter("%s", cell->name);
600
601 now = ktime_get_real_seconds();
602 cell->last_inactive = now;
603 expire_delay = 0;
604 if (cell->vl_servers->nr_servers)
605 expire_delay = afs_cell_gc_delay;
606
607 debug_id = cell->debug_id;
608 u = atomic_read(&cell->ref);
609 a = atomic_dec_return(&cell->active);
610 trace_afs_cell(debug_id, u, a, reason);
611 WARN_ON(a == 0);
612 if (a == 1)
613 /* 'cell' may now be garbage collected. */
614 afs_set_cell_timer(net, expire_delay);
615}
616
617/*
618 * Note that a cell has been seen.
619 */
620void afs_see_cell(struct afs_cell *cell, enum afs_cell_trace reason)
621{
622 int u, a;
623
624 u = atomic_read(&cell->ref);
625 a = atomic_read(&cell->active);
626 trace_afs_cell(cell->debug_id, u, a, reason);
627}
628
629/*
630 * Queue a cell for management, giving the workqueue a ref to hold.
631 */
632void afs_queue_cell(struct afs_cell *cell, enum afs_cell_trace reason)
633{
634 afs_get_cell(cell, reason);
635 if (!queue_work(afs_wq, &cell->manager))
636 afs_put_cell(cell, afs_cell_trace_put_queue_fail);
637}
638
639/*
640 * Allocate a key to use as a placeholder for anonymous user security.
641 */
642static int afs_alloc_anon_key(struct afs_cell *cell)
643{
644 struct key *key;
645 char keyname[4 + AFS_MAXCELLNAME + 1], *cp, *dp;
646
647 /* Create a key to represent an anonymous user. */
648 memcpy(keyname, "afs@", 4);
649 dp = keyname + 4;
650 cp = cell->name;
651 do {
652 *dp++ = tolower(*cp);
653 } while (*cp++);
654
655 key = rxrpc_get_null_key(keyname);
656 if (IS_ERR(key))
657 return PTR_ERR(key);
658
659 cell->anonymous_key = key;
660
661 _debug("anon key %p{%x}",
662 cell->anonymous_key, key_serial(cell->anonymous_key));
663 return 0;
664}
665
666/*
667 * Activate a cell.
668 */
669static int afs_activate_cell(struct afs_net *net, struct afs_cell *cell)
670{
671 struct hlist_node **p;
672 struct afs_cell *pcell;
673 int ret;
674
675 if (!cell->anonymous_key) {
676 ret = afs_alloc_anon_key(cell);
677 if (ret < 0)
678 return ret;
679 }
680
681#ifdef CONFIG_AFS_FSCACHE
682 cell->cache = fscache_acquire_cookie(afs_cache_netfs.primary_index,
683 &afs_cell_cache_index_def,
684 cell->name, strlen(cell->name),
685 NULL, 0,
686 cell, 0, true);
687#endif
688 ret = afs_proc_cell_setup(cell);
689 if (ret < 0)
690 return ret;
691
692 mutex_lock(&net->proc_cells_lock);
693 for (p = &net->proc_cells.first; *p; p = &(*p)->next) {
694 pcell = hlist_entry(*p, struct afs_cell, proc_link);
695 if (strcmp(cell->name, pcell->name) < 0)
696 break;
697 }
698
699 cell->proc_link.pprev = p;
700 cell->proc_link.next = *p;
701 rcu_assign_pointer(*p, &cell->proc_link.next);
702 if (cell->proc_link.next)
703 cell->proc_link.next->pprev = &cell->proc_link.next;
704
705 afs_dynroot_mkdir(net, cell);
706 mutex_unlock(&net->proc_cells_lock);
707 return 0;
708}
709
710/*
711 * Deactivate a cell.
712 */
713static void afs_deactivate_cell(struct afs_net *net, struct afs_cell *cell)
714{
715 _enter("%s", cell->name);
716
717 afs_proc_cell_remove(cell);
718
719 mutex_lock(&net->proc_cells_lock);
720 hlist_del_rcu(&cell->proc_link);
721 afs_dynroot_rmdir(net, cell);
722 mutex_unlock(&net->proc_cells_lock);
723
724#ifdef CONFIG_AFS_FSCACHE
725 fscache_relinquish_cookie(cell->cache, NULL, false);
726 cell->cache = NULL;
727#endif
728
729 _leave("");
730}
731
732/*
733 * Manage a cell record, initialising and destroying it, maintaining its DNS
734 * records.
735 */
736static void afs_manage_cell(struct afs_cell *cell)
737{
738 struct afs_net *net = cell->net;
739 int ret, active;
740
741 _enter("%s", cell->name);
742
743again:
744 _debug("state %u", cell->state);
745 switch (cell->state) {
746 case AFS_CELL_INACTIVE:
747 case AFS_CELL_FAILED:
748 down_write(&net->cells_lock);
749 active = 1;
750 if (atomic_try_cmpxchg_relaxed(&cell->active, &active, 0)) {
751 rb_erase(&cell->net_node, &net->cells);
752 trace_afs_cell(cell->debug_id, atomic_read(&cell->ref), 0,
753 afs_cell_trace_unuse_delete);
754 smp_store_release(&cell->state, AFS_CELL_REMOVED);
755 }
756 up_write(&net->cells_lock);
757 if (cell->state == AFS_CELL_REMOVED) {
758 wake_up_var(&cell->state);
759 goto final_destruction;
760 }
761 if (cell->state == AFS_CELL_FAILED)
762 goto done;
763 smp_store_release(&cell->state, AFS_CELL_UNSET);
764 wake_up_var(&cell->state);
765 goto again;
766
767 case AFS_CELL_UNSET:
768 smp_store_release(&cell->state, AFS_CELL_ACTIVATING);
769 wake_up_var(&cell->state);
770 goto again;
771
772 case AFS_CELL_ACTIVATING:
773 ret = afs_activate_cell(net, cell);
774 if (ret < 0)
775 goto activation_failed;
776
777 smp_store_release(&cell->state, AFS_CELL_ACTIVE);
778 wake_up_var(&cell->state);
779 goto again;
780
781 case AFS_CELL_ACTIVE:
782 if (atomic_read(&cell->active) > 1) {
783 if (test_and_clear_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags)) {
784 ret = afs_update_cell(cell);
785 if (ret < 0)
786 cell->error = ret;
787 }
788 goto done;
789 }
790 smp_store_release(&cell->state, AFS_CELL_DEACTIVATING);
791 wake_up_var(&cell->state);
792 goto again;
793
794 case AFS_CELL_DEACTIVATING:
795 if (atomic_read(&cell->active) > 1)
796 goto reverse_deactivation;
797 afs_deactivate_cell(net, cell);
798 smp_store_release(&cell->state, AFS_CELL_INACTIVE);
799 wake_up_var(&cell->state);
800 goto again;
801
802 case AFS_CELL_REMOVED:
803 goto done;
804
805 default:
806 break;
807 }
808 _debug("bad state %u", cell->state);
809 BUG(); /* Unhandled state */
810
811activation_failed:
812 cell->error = ret;
813 afs_deactivate_cell(net, cell);
814
815 smp_store_release(&cell->state, AFS_CELL_FAILED); /* vs error */
816 wake_up_var(&cell->state);
817 goto again;
818
819reverse_deactivation:
820 smp_store_release(&cell->state, AFS_CELL_ACTIVE);
821 wake_up_var(&cell->state);
822 _leave(" [deact->act]");
823 return;
824
825done:
826 _leave(" [done %u]", cell->state);
827 return;
828
829final_destruction:
830 /* The root volume is pinning the cell */
831 afs_put_volume(cell->net, cell->root_volume, afs_volume_trace_put_cell_root);
832 cell->root_volume = NULL;
833 afs_put_cell(cell, afs_cell_trace_put_destroy);
834}
835
836static void afs_manage_cell_work(struct work_struct *work)
837{
838 struct afs_cell *cell = container_of(work, struct afs_cell, manager);
839
840 afs_manage_cell(cell);
841 afs_put_cell(cell, afs_cell_trace_put_queue_work);
842}
843
844/*
845 * Manage the records of cells known to a network namespace. This includes
846 * updating the DNS records and garbage collecting unused cells that were
847 * automatically added.
848 *
849 * Note that constructed cell records may only be removed from net->cells by
850 * this work item, so it is safe for this work item to stash a cursor pointing
851 * into the tree and then return to caller (provided it skips cells that are
852 * still under construction).
853 *
854 * Note also that we were given an increment on net->cells_outstanding by
855 * whoever queued us that we need to deal with before returning.
856 */
857void afs_manage_cells(struct work_struct *work)
858{
859 struct afs_net *net = container_of(work, struct afs_net, cells_manager);
860 struct rb_node *cursor;
861 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
862 bool purging = !net->live;
863
864 _enter("");
865
866 /* Trawl the cell database looking for cells that have expired from
867 * lack of use and cells whose DNS results have expired and dispatch
868 * their managers.
869 */
870 down_read(&net->cells_lock);
871
872 for (cursor = rb_first(&net->cells); cursor; cursor = rb_next(cursor)) {
873 struct afs_cell *cell =
874 rb_entry(cursor, struct afs_cell, net_node);
875 unsigned active;
876 bool sched_cell = false;
877
878 active = atomic_read(&cell->active);
879 trace_afs_cell(cell->debug_id, atomic_read(&cell->ref),
880 active, afs_cell_trace_manage);
881
882 ASSERTCMP(active, >=, 1);
883
884 if (purging) {
885 if (test_and_clear_bit(AFS_CELL_FL_NO_GC, &cell->flags)) {
886 active = atomic_dec_return(&cell->active);
887 trace_afs_cell(cell->debug_id, atomic_read(&cell->ref),
888 active, afs_cell_trace_unuse_pin);
889 }
890 }
891
892 if (active == 1) {
893 struct afs_vlserver_list *vllist;
894 time64_t expire_at = cell->last_inactive;
895
896 read_lock(&cell->vl_servers_lock);
897 vllist = rcu_dereference_protected(
898 cell->vl_servers,
899 lockdep_is_held(&cell->vl_servers_lock));
900 if (vllist->nr_servers > 0)
901 expire_at += afs_cell_gc_delay;
902 read_unlock(&cell->vl_servers_lock);
903 if (purging || expire_at <= now)
904 sched_cell = true;
905 else if (expire_at < next_manage)
906 next_manage = expire_at;
907 }
908
909 if (!purging) {
910 if (test_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags))
911 sched_cell = true;
912 }
913
914 if (sched_cell)
915 afs_queue_cell(cell, afs_cell_trace_get_queue_manage);
916 }
917
918 up_read(&net->cells_lock);
919
920 /* Update the timer on the way out. We have to pass an increment on
921 * cells_outstanding in the namespace that we are in to the timer or
922 * the work scheduler.
923 */
924 if (!purging && next_manage < TIME64_MAX) {
925 now = ktime_get_real_seconds();
926
927 if (next_manage - now <= 0) {
928 if (queue_work(afs_wq, &net->cells_manager))
929 atomic_inc(&net->cells_outstanding);
930 } else {
931 afs_set_cell_timer(net, next_manage - now);
932 }
933 }
934
935 afs_dec_cells_outstanding(net);
936 _leave(" [%d]", atomic_read(&net->cells_outstanding));
937}
938
939/*
940 * Purge in-memory cell database.
941 */
942void afs_cell_purge(struct afs_net *net)
943{
944 struct afs_cell *ws;
945
946 _enter("");
947
948 down_write(&net->cells_lock);
949 ws = net->ws_cell;
950 net->ws_cell = NULL;
951 up_write(&net->cells_lock);
952 afs_unuse_cell(net, ws, afs_cell_trace_unuse_ws);
953
954 _debug("del timer");
955 if (del_timer_sync(&net->cells_timer))
956 atomic_dec(&net->cells_outstanding);
957
958 _debug("kick mgr");
959 afs_queue_cell_manager(net);
960
961 _debug("wait");
962 wait_var_event(&net->cells_outstanding,
963 !atomic_read(&net->cells_outstanding));
964 _leave("");
965}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* AFS cell and server record management
3 *
4 * Copyright (C) 2002, 2017 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/slab.h>
9#include <linux/key.h>
10#include <linux/ctype.h>
11#include <linux/dns_resolver.h>
12#include <linux/sched.h>
13#include <linux/inet.h>
14#include <linux/namei.h>
15#include <keys/rxrpc-type.h>
16#include "internal.h"
17
18static unsigned __read_mostly afs_cell_gc_delay = 10;
19static unsigned __read_mostly afs_cell_min_ttl = 10 * 60;
20static unsigned __read_mostly afs_cell_max_ttl = 24 * 60 * 60;
21static atomic_t cell_debug_id;
22
23static void afs_queue_cell_manager(struct afs_net *);
24static void afs_manage_cell_work(struct work_struct *);
25
26static void afs_dec_cells_outstanding(struct afs_net *net)
27{
28 if (atomic_dec_and_test(&net->cells_outstanding))
29 wake_up_var(&net->cells_outstanding);
30}
31
32/*
33 * Set the cell timer to fire after a given delay, assuming it's not already
34 * set for an earlier time.
35 */
36static void afs_set_cell_timer(struct afs_net *net, time64_t delay)
37{
38 if (net->live) {
39 atomic_inc(&net->cells_outstanding);
40 if (timer_reduce(&net->cells_timer, jiffies + delay * HZ))
41 afs_dec_cells_outstanding(net);
42 } else {
43 afs_queue_cell_manager(net);
44 }
45}
46
47/*
48 * Look up and get an activation reference on a cell record. The caller must
49 * hold net->cells_lock at least read-locked.
50 */
51static struct afs_cell *afs_find_cell_locked(struct afs_net *net,
52 const char *name, unsigned int namesz,
53 enum afs_cell_trace reason)
54{
55 struct afs_cell *cell = NULL;
56 struct rb_node *p;
57 int n;
58
59 _enter("%*.*s", namesz, namesz, name);
60
61 if (name && namesz == 0)
62 return ERR_PTR(-EINVAL);
63 if (namesz > AFS_MAXCELLNAME)
64 return ERR_PTR(-ENAMETOOLONG);
65
66 if (!name) {
67 cell = net->ws_cell;
68 if (!cell)
69 return ERR_PTR(-EDESTADDRREQ);
70 goto found;
71 }
72
73 p = net->cells.rb_node;
74 while (p) {
75 cell = rb_entry(p, struct afs_cell, net_node);
76
77 n = strncasecmp(cell->name, name,
78 min_t(size_t, cell->name_len, namesz));
79 if (n == 0)
80 n = cell->name_len - namesz;
81 if (n < 0)
82 p = p->rb_left;
83 else if (n > 0)
84 p = p->rb_right;
85 else
86 goto found;
87 }
88
89 return ERR_PTR(-ENOENT);
90
91found:
92 return afs_use_cell(cell, reason);
93}
94
95/*
96 * Look up and get an activation reference on a cell record.
97 */
98struct afs_cell *afs_find_cell(struct afs_net *net,
99 const char *name, unsigned int namesz,
100 enum afs_cell_trace reason)
101{
102 struct afs_cell *cell;
103
104 down_read(&net->cells_lock);
105 cell = afs_find_cell_locked(net, name, namesz, reason);
106 up_read(&net->cells_lock);
107 return cell;
108}
109
110/*
111 * Set up a cell record and fill in its name, VL server address list and
112 * allocate an anonymous key
113 */
114static struct afs_cell *afs_alloc_cell(struct afs_net *net,
115 const char *name, unsigned int namelen,
116 const char *addresses)
117{
118 struct afs_vlserver_list *vllist;
119 struct afs_cell *cell;
120 int i, ret;
121
122 ASSERT(name);
123 if (namelen == 0)
124 return ERR_PTR(-EINVAL);
125 if (namelen > AFS_MAXCELLNAME) {
126 _leave(" = -ENAMETOOLONG");
127 return ERR_PTR(-ENAMETOOLONG);
128 }
129
130 /* Prohibit cell names that contain unprintable chars, '/' and '@' or
131 * that begin with a dot. This also precludes "@cell".
132 */
133 if (name[0] == '.')
134 return ERR_PTR(-EINVAL);
135 for (i = 0; i < namelen; i++) {
136 char ch = name[i];
137 if (!isprint(ch) || ch == '/' || ch == '@')
138 return ERR_PTR(-EINVAL);
139 }
140
141 _enter("%*.*s,%s", namelen, namelen, name, addresses);
142
143 cell = kzalloc(sizeof(struct afs_cell), GFP_KERNEL);
144 if (!cell) {
145 _leave(" = -ENOMEM");
146 return ERR_PTR(-ENOMEM);
147 }
148
149 cell->name = kmalloc(namelen + 1, GFP_KERNEL);
150 if (!cell->name) {
151 kfree(cell);
152 return ERR_PTR(-ENOMEM);
153 }
154
155 cell->net = net;
156 cell->name_len = namelen;
157 for (i = 0; i < namelen; i++)
158 cell->name[i] = tolower(name[i]);
159 cell->name[i] = 0;
160
161 refcount_set(&cell->ref, 1);
162 atomic_set(&cell->active, 0);
163 INIT_WORK(&cell->manager, afs_manage_cell_work);
164 cell->volumes = RB_ROOT;
165 INIT_HLIST_HEAD(&cell->proc_volumes);
166 seqlock_init(&cell->volume_lock);
167 cell->fs_servers = RB_ROOT;
168 seqlock_init(&cell->fs_lock);
169 INIT_LIST_HEAD(&cell->fs_open_mmaps);
170 init_rwsem(&cell->fs_open_mmaps_lock);
171 rwlock_init(&cell->vl_servers_lock);
172 cell->flags = (1 << AFS_CELL_FL_CHECK_ALIAS);
173
174 /* Provide a VL server list, filling it in if we were given a list of
175 * addresses to use.
176 */
177 if (addresses) {
178 vllist = afs_parse_text_addrs(net,
179 addresses, strlen(addresses), ':',
180 VL_SERVICE, AFS_VL_PORT);
181 if (IS_ERR(vllist)) {
182 ret = PTR_ERR(vllist);
183 goto parse_failed;
184 }
185
186 vllist->source = DNS_RECORD_FROM_CONFIG;
187 vllist->status = DNS_LOOKUP_NOT_DONE;
188 cell->dns_expiry = TIME64_MAX;
189 } else {
190 ret = -ENOMEM;
191 vllist = afs_alloc_vlserver_list(0);
192 if (!vllist)
193 goto error;
194 vllist->source = DNS_RECORD_UNAVAILABLE;
195 vllist->status = DNS_LOOKUP_NOT_DONE;
196 cell->dns_expiry = ktime_get_real_seconds();
197 }
198
199 rcu_assign_pointer(cell->vl_servers, vllist);
200
201 cell->dns_source = vllist->source;
202 cell->dns_status = vllist->status;
203 smp_store_release(&cell->dns_lookup_count, 1); /* vs source/status */
204 atomic_inc(&net->cells_outstanding);
205 cell->debug_id = atomic_inc_return(&cell_debug_id);
206 trace_afs_cell(cell->debug_id, 1, 0, afs_cell_trace_alloc);
207
208 _leave(" = %p", cell);
209 return cell;
210
211parse_failed:
212 if (ret == -EINVAL)
213 printk(KERN_ERR "kAFS: bad VL server IP address\n");
214error:
215 kfree(cell->name);
216 kfree(cell);
217 _leave(" = %d", ret);
218 return ERR_PTR(ret);
219}
220
221/*
222 * afs_lookup_cell - Look up or create a cell record.
223 * @net: The network namespace
224 * @name: The name of the cell.
225 * @namesz: The strlen of the cell name.
226 * @vllist: A colon/comma separated list of numeric IP addresses or NULL.
227 * @excl: T if an error should be given if the cell name already exists.
228 *
229 * Look up a cell record by name and query the DNS for VL server addresses if
230 * needed. Note that that actual DNS query is punted off to the manager thread
231 * so that this function can return immediately if interrupted whilst allowing
232 * cell records to be shared even if not yet fully constructed.
233 */
234struct afs_cell *afs_lookup_cell(struct afs_net *net,
235 const char *name, unsigned int namesz,
236 const char *vllist, bool excl)
237{
238 struct afs_cell *cell, *candidate, *cursor;
239 struct rb_node *parent, **pp;
240 enum afs_cell_state state;
241 int ret, n;
242
243 _enter("%s,%s", name, vllist);
244
245 if (!excl) {
246 cell = afs_find_cell(net, name, namesz, afs_cell_trace_use_lookup);
247 if (!IS_ERR(cell))
248 goto wait_for_cell;
249 }
250
251 /* Assume we're probably going to create a cell and preallocate and
252 * mostly set up a candidate record. We can then use this to stash the
253 * name, the net namespace and VL server addresses.
254 *
255 * We also want to do this before we hold any locks as it may involve
256 * upcalling to userspace to make DNS queries.
257 */
258 candidate = afs_alloc_cell(net, name, namesz, vllist);
259 if (IS_ERR(candidate)) {
260 _leave(" = %ld", PTR_ERR(candidate));
261 return candidate;
262 }
263
264 /* Find the insertion point and check to see if someone else added a
265 * cell whilst we were allocating.
266 */
267 down_write(&net->cells_lock);
268
269 pp = &net->cells.rb_node;
270 parent = NULL;
271 while (*pp) {
272 parent = *pp;
273 cursor = rb_entry(parent, struct afs_cell, net_node);
274
275 n = strncasecmp(cursor->name, name,
276 min_t(size_t, cursor->name_len, namesz));
277 if (n == 0)
278 n = cursor->name_len - namesz;
279 if (n < 0)
280 pp = &(*pp)->rb_left;
281 else if (n > 0)
282 pp = &(*pp)->rb_right;
283 else
284 goto cell_already_exists;
285 }
286
287 cell = candidate;
288 candidate = NULL;
289 atomic_set(&cell->active, 2);
290 trace_afs_cell(cell->debug_id, refcount_read(&cell->ref), 2, afs_cell_trace_insert);
291 rb_link_node_rcu(&cell->net_node, parent, pp);
292 rb_insert_color(&cell->net_node, &net->cells);
293 up_write(&net->cells_lock);
294
295 afs_queue_cell(cell, afs_cell_trace_get_queue_new);
296
297wait_for_cell:
298 trace_afs_cell(cell->debug_id, refcount_read(&cell->ref), atomic_read(&cell->active),
299 afs_cell_trace_wait);
300 _debug("wait_for_cell");
301 wait_var_event(&cell->state,
302 ({
303 state = smp_load_acquire(&cell->state); /* vs error */
304 state == AFS_CELL_ACTIVE || state == AFS_CELL_REMOVED;
305 }));
306
307 /* Check the state obtained from the wait check. */
308 if (state == AFS_CELL_REMOVED) {
309 ret = cell->error;
310 goto error;
311 }
312
313 _leave(" = %p [cell]", cell);
314 return cell;
315
316cell_already_exists:
317 _debug("cell exists");
318 cell = cursor;
319 if (excl) {
320 ret = -EEXIST;
321 } else {
322 afs_use_cell(cursor, afs_cell_trace_use_lookup);
323 ret = 0;
324 }
325 up_write(&net->cells_lock);
326 if (candidate)
327 afs_put_cell(candidate, afs_cell_trace_put_candidate);
328 if (ret == 0)
329 goto wait_for_cell;
330 goto error_noput;
331error:
332 afs_unuse_cell(net, cell, afs_cell_trace_unuse_lookup);
333error_noput:
334 _leave(" = %d [error]", ret);
335 return ERR_PTR(ret);
336}
337
338/*
339 * set the root cell information
340 * - can be called with a module parameter string
341 * - can be called from a write to /proc/fs/afs/rootcell
342 */
343int afs_cell_init(struct afs_net *net, const char *rootcell)
344{
345 struct afs_cell *old_root, *new_root;
346 const char *cp, *vllist;
347 size_t len;
348
349 _enter("");
350
351 if (!rootcell) {
352 /* module is loaded with no parameters, or built statically.
353 * - in the future we might initialize cell DB here.
354 */
355 _leave(" = 0 [no root]");
356 return 0;
357 }
358
359 cp = strchr(rootcell, ':');
360 if (!cp) {
361 _debug("kAFS: no VL server IP addresses specified");
362 vllist = NULL;
363 len = strlen(rootcell);
364 } else {
365 vllist = cp + 1;
366 len = cp - rootcell;
367 }
368
369 /* allocate a cell record for the root cell */
370 new_root = afs_lookup_cell(net, rootcell, len, vllist, false);
371 if (IS_ERR(new_root)) {
372 _leave(" = %ld", PTR_ERR(new_root));
373 return PTR_ERR(new_root);
374 }
375
376 if (!test_and_set_bit(AFS_CELL_FL_NO_GC, &new_root->flags))
377 afs_use_cell(new_root, afs_cell_trace_use_pin);
378
379 /* install the new cell */
380 down_write(&net->cells_lock);
381 afs_see_cell(new_root, afs_cell_trace_see_ws);
382 old_root = net->ws_cell;
383 net->ws_cell = new_root;
384 up_write(&net->cells_lock);
385
386 afs_unuse_cell(net, old_root, afs_cell_trace_unuse_ws);
387 _leave(" = 0");
388 return 0;
389}
390
391/*
392 * Update a cell's VL server address list from the DNS.
393 */
394static int afs_update_cell(struct afs_cell *cell)
395{
396 struct afs_vlserver_list *vllist, *old = NULL, *p;
397 unsigned int min_ttl = READ_ONCE(afs_cell_min_ttl);
398 unsigned int max_ttl = READ_ONCE(afs_cell_max_ttl);
399 time64_t now, expiry = 0;
400 int ret = 0;
401
402 _enter("%s", cell->name);
403
404 vllist = afs_dns_query(cell, &expiry);
405 if (IS_ERR(vllist)) {
406 ret = PTR_ERR(vllist);
407
408 _debug("%s: fail %d", cell->name, ret);
409 if (ret == -ENOMEM)
410 goto out_wake;
411
412 ret = -ENOMEM;
413 vllist = afs_alloc_vlserver_list(0);
414 if (!vllist)
415 goto out_wake;
416
417 switch (ret) {
418 case -ENODATA:
419 case -EDESTADDRREQ:
420 vllist->status = DNS_LOOKUP_GOT_NOT_FOUND;
421 break;
422 case -EAGAIN:
423 case -ECONNREFUSED:
424 vllist->status = DNS_LOOKUP_GOT_TEMP_FAILURE;
425 break;
426 default:
427 vllist->status = DNS_LOOKUP_GOT_LOCAL_FAILURE;
428 break;
429 }
430 }
431
432 _debug("%s: got list %d %d", cell->name, vllist->source, vllist->status);
433 cell->dns_status = vllist->status;
434
435 now = ktime_get_real_seconds();
436 if (min_ttl > max_ttl)
437 max_ttl = min_ttl;
438 if (expiry < now + min_ttl)
439 expiry = now + min_ttl;
440 else if (expiry > now + max_ttl)
441 expiry = now + max_ttl;
442
443 _debug("%s: status %d", cell->name, vllist->status);
444 if (vllist->source == DNS_RECORD_UNAVAILABLE) {
445 switch (vllist->status) {
446 case DNS_LOOKUP_GOT_NOT_FOUND:
447 /* The DNS said that the cell does not exist or there
448 * weren't any addresses to be had.
449 */
450 cell->dns_expiry = expiry;
451 break;
452
453 case DNS_LOOKUP_BAD:
454 case DNS_LOOKUP_GOT_LOCAL_FAILURE:
455 case DNS_LOOKUP_GOT_TEMP_FAILURE:
456 case DNS_LOOKUP_GOT_NS_FAILURE:
457 default:
458 cell->dns_expiry = now + 10;
459 break;
460 }
461 } else {
462 cell->dns_expiry = expiry;
463 }
464
465 /* Replace the VL server list if the new record has servers or the old
466 * record doesn't.
467 */
468 write_lock(&cell->vl_servers_lock);
469 p = rcu_dereference_protected(cell->vl_servers, true);
470 if (vllist->nr_servers > 0 || p->nr_servers == 0) {
471 rcu_assign_pointer(cell->vl_servers, vllist);
472 cell->dns_source = vllist->source;
473 old = p;
474 }
475 write_unlock(&cell->vl_servers_lock);
476 afs_put_vlserverlist(cell->net, old);
477
478out_wake:
479 smp_store_release(&cell->dns_lookup_count,
480 cell->dns_lookup_count + 1); /* vs source/status */
481 wake_up_var(&cell->dns_lookup_count);
482 _leave(" = %d", ret);
483 return ret;
484}
485
486/*
487 * Destroy a cell record
488 */
489static void afs_cell_destroy(struct rcu_head *rcu)
490{
491 struct afs_cell *cell = container_of(rcu, struct afs_cell, rcu);
492 struct afs_net *net = cell->net;
493 int r;
494
495 _enter("%p{%s}", cell, cell->name);
496
497 r = refcount_read(&cell->ref);
498 ASSERTCMP(r, ==, 0);
499 trace_afs_cell(cell->debug_id, r, atomic_read(&cell->active), afs_cell_trace_free);
500
501 afs_put_vlserverlist(net, rcu_access_pointer(cell->vl_servers));
502 afs_unuse_cell(net, cell->alias_of, afs_cell_trace_unuse_alias);
503 key_put(cell->anonymous_key);
504 kfree(cell->name);
505 kfree(cell);
506
507 afs_dec_cells_outstanding(net);
508 _leave(" [destroyed]");
509}
510
511/*
512 * Queue the cell manager.
513 */
514static void afs_queue_cell_manager(struct afs_net *net)
515{
516 int outstanding = atomic_inc_return(&net->cells_outstanding);
517
518 _enter("%d", outstanding);
519
520 if (!queue_work(afs_wq, &net->cells_manager))
521 afs_dec_cells_outstanding(net);
522}
523
524/*
525 * Cell management timer. We have an increment on cells_outstanding that we
526 * need to pass along to the work item.
527 */
528void afs_cells_timer(struct timer_list *timer)
529{
530 struct afs_net *net = container_of(timer, struct afs_net, cells_timer);
531
532 _enter("");
533 if (!queue_work(afs_wq, &net->cells_manager))
534 afs_dec_cells_outstanding(net);
535}
536
537/*
538 * Get a reference on a cell record.
539 */
540struct afs_cell *afs_get_cell(struct afs_cell *cell, enum afs_cell_trace reason)
541{
542 int r;
543
544 __refcount_inc(&cell->ref, &r);
545 trace_afs_cell(cell->debug_id, r + 1, atomic_read(&cell->active), reason);
546 return cell;
547}
548
549/*
550 * Drop a reference on a cell record.
551 */
552void afs_put_cell(struct afs_cell *cell, enum afs_cell_trace reason)
553{
554 if (cell) {
555 unsigned int debug_id = cell->debug_id;
556 unsigned int a;
557 bool zero;
558 int r;
559
560 a = atomic_read(&cell->active);
561 zero = __refcount_dec_and_test(&cell->ref, &r);
562 trace_afs_cell(debug_id, r - 1, a, reason);
563 if (zero) {
564 a = atomic_read(&cell->active);
565 WARN(a != 0, "Cell active count %u > 0\n", a);
566 call_rcu(&cell->rcu, afs_cell_destroy);
567 }
568 }
569}
570
571/*
572 * Note a cell becoming more active.
573 */
574struct afs_cell *afs_use_cell(struct afs_cell *cell, enum afs_cell_trace reason)
575{
576 int r, a;
577
578 r = refcount_read(&cell->ref);
579 WARN_ON(r == 0);
580 a = atomic_inc_return(&cell->active);
581 trace_afs_cell(cell->debug_id, r, a, reason);
582 return cell;
583}
584
585/*
586 * Record a cell becoming less active. When the active counter reaches 1, it
587 * is scheduled for destruction, but may get reactivated.
588 */
589void afs_unuse_cell(struct afs_net *net, struct afs_cell *cell, enum afs_cell_trace reason)
590{
591 unsigned int debug_id;
592 time64_t now, expire_delay;
593 int r, a;
594
595 if (!cell)
596 return;
597
598 _enter("%s", cell->name);
599
600 now = ktime_get_real_seconds();
601 cell->last_inactive = now;
602 expire_delay = 0;
603 if (cell->vl_servers->nr_servers)
604 expire_delay = afs_cell_gc_delay;
605
606 debug_id = cell->debug_id;
607 r = refcount_read(&cell->ref);
608 a = atomic_dec_return(&cell->active);
609 trace_afs_cell(debug_id, r, a, reason);
610 WARN_ON(a == 0);
611 if (a == 1)
612 /* 'cell' may now be garbage collected. */
613 afs_set_cell_timer(net, expire_delay);
614}
615
616/*
617 * Note that a cell has been seen.
618 */
619void afs_see_cell(struct afs_cell *cell, enum afs_cell_trace reason)
620{
621 int r, a;
622
623 r = refcount_read(&cell->ref);
624 a = atomic_read(&cell->active);
625 trace_afs_cell(cell->debug_id, r, a, reason);
626}
627
628/*
629 * Queue a cell for management, giving the workqueue a ref to hold.
630 */
631void afs_queue_cell(struct afs_cell *cell, enum afs_cell_trace reason)
632{
633 afs_get_cell(cell, reason);
634 if (!queue_work(afs_wq, &cell->manager))
635 afs_put_cell(cell, afs_cell_trace_put_queue_fail);
636}
637
638/*
639 * Allocate a key to use as a placeholder for anonymous user security.
640 */
641static int afs_alloc_anon_key(struct afs_cell *cell)
642{
643 struct key *key;
644 char keyname[4 + AFS_MAXCELLNAME + 1], *cp, *dp;
645
646 /* Create a key to represent an anonymous user. */
647 memcpy(keyname, "afs@", 4);
648 dp = keyname + 4;
649 cp = cell->name;
650 do {
651 *dp++ = tolower(*cp);
652 } while (*cp++);
653
654 key = rxrpc_get_null_key(keyname);
655 if (IS_ERR(key))
656 return PTR_ERR(key);
657
658 cell->anonymous_key = key;
659
660 _debug("anon key %p{%x}",
661 cell->anonymous_key, key_serial(cell->anonymous_key));
662 return 0;
663}
664
665/*
666 * Activate a cell.
667 */
668static int afs_activate_cell(struct afs_net *net, struct afs_cell *cell)
669{
670 struct hlist_node **p;
671 struct afs_cell *pcell;
672 int ret;
673
674 if (!cell->anonymous_key) {
675 ret = afs_alloc_anon_key(cell);
676 if (ret < 0)
677 return ret;
678 }
679
680 ret = afs_proc_cell_setup(cell);
681 if (ret < 0)
682 return ret;
683
684 mutex_lock(&net->proc_cells_lock);
685 for (p = &net->proc_cells.first; *p; p = &(*p)->next) {
686 pcell = hlist_entry(*p, struct afs_cell, proc_link);
687 if (strcmp(cell->name, pcell->name) < 0)
688 break;
689 }
690
691 cell->proc_link.pprev = p;
692 cell->proc_link.next = *p;
693 rcu_assign_pointer(*p, &cell->proc_link.next);
694 if (cell->proc_link.next)
695 cell->proc_link.next->pprev = &cell->proc_link.next;
696
697 afs_dynroot_mkdir(net, cell);
698 mutex_unlock(&net->proc_cells_lock);
699 return 0;
700}
701
702/*
703 * Deactivate a cell.
704 */
705static void afs_deactivate_cell(struct afs_net *net, struct afs_cell *cell)
706{
707 _enter("%s", cell->name);
708
709 afs_proc_cell_remove(cell);
710
711 mutex_lock(&net->proc_cells_lock);
712 hlist_del_rcu(&cell->proc_link);
713 afs_dynroot_rmdir(net, cell);
714 mutex_unlock(&net->proc_cells_lock);
715
716 _leave("");
717}
718
719/*
720 * Manage a cell record, initialising and destroying it, maintaining its DNS
721 * records.
722 */
723static void afs_manage_cell(struct afs_cell *cell)
724{
725 struct afs_net *net = cell->net;
726 int ret, active;
727
728 _enter("%s", cell->name);
729
730again:
731 _debug("state %u", cell->state);
732 switch (cell->state) {
733 case AFS_CELL_INACTIVE:
734 case AFS_CELL_FAILED:
735 down_write(&net->cells_lock);
736 active = 1;
737 if (atomic_try_cmpxchg_relaxed(&cell->active, &active, 0)) {
738 rb_erase(&cell->net_node, &net->cells);
739 trace_afs_cell(cell->debug_id, refcount_read(&cell->ref), 0,
740 afs_cell_trace_unuse_delete);
741 smp_store_release(&cell->state, AFS_CELL_REMOVED);
742 }
743 up_write(&net->cells_lock);
744 if (cell->state == AFS_CELL_REMOVED) {
745 wake_up_var(&cell->state);
746 goto final_destruction;
747 }
748 if (cell->state == AFS_CELL_FAILED)
749 goto done;
750 smp_store_release(&cell->state, AFS_CELL_UNSET);
751 wake_up_var(&cell->state);
752 goto again;
753
754 case AFS_CELL_UNSET:
755 smp_store_release(&cell->state, AFS_CELL_ACTIVATING);
756 wake_up_var(&cell->state);
757 goto again;
758
759 case AFS_CELL_ACTIVATING:
760 ret = afs_activate_cell(net, cell);
761 if (ret < 0)
762 goto activation_failed;
763
764 smp_store_release(&cell->state, AFS_CELL_ACTIVE);
765 wake_up_var(&cell->state);
766 goto again;
767
768 case AFS_CELL_ACTIVE:
769 if (atomic_read(&cell->active) > 1) {
770 if (test_and_clear_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags)) {
771 ret = afs_update_cell(cell);
772 if (ret < 0)
773 cell->error = ret;
774 }
775 goto done;
776 }
777 smp_store_release(&cell->state, AFS_CELL_DEACTIVATING);
778 wake_up_var(&cell->state);
779 goto again;
780
781 case AFS_CELL_DEACTIVATING:
782 if (atomic_read(&cell->active) > 1)
783 goto reverse_deactivation;
784 afs_deactivate_cell(net, cell);
785 smp_store_release(&cell->state, AFS_CELL_INACTIVE);
786 wake_up_var(&cell->state);
787 goto again;
788
789 case AFS_CELL_REMOVED:
790 goto done;
791
792 default:
793 break;
794 }
795 _debug("bad state %u", cell->state);
796 BUG(); /* Unhandled state */
797
798activation_failed:
799 cell->error = ret;
800 afs_deactivate_cell(net, cell);
801
802 smp_store_release(&cell->state, AFS_CELL_FAILED); /* vs error */
803 wake_up_var(&cell->state);
804 goto again;
805
806reverse_deactivation:
807 smp_store_release(&cell->state, AFS_CELL_ACTIVE);
808 wake_up_var(&cell->state);
809 _leave(" [deact->act]");
810 return;
811
812done:
813 _leave(" [done %u]", cell->state);
814 return;
815
816final_destruction:
817 /* The root volume is pinning the cell */
818 afs_put_volume(cell->net, cell->root_volume, afs_volume_trace_put_cell_root);
819 cell->root_volume = NULL;
820 afs_put_cell(cell, afs_cell_trace_put_destroy);
821}
822
823static void afs_manage_cell_work(struct work_struct *work)
824{
825 struct afs_cell *cell = container_of(work, struct afs_cell, manager);
826
827 afs_manage_cell(cell);
828 afs_put_cell(cell, afs_cell_trace_put_queue_work);
829}
830
831/*
832 * Manage the records of cells known to a network namespace. This includes
833 * updating the DNS records and garbage collecting unused cells that were
834 * automatically added.
835 *
836 * Note that constructed cell records may only be removed from net->cells by
837 * this work item, so it is safe for this work item to stash a cursor pointing
838 * into the tree and then return to caller (provided it skips cells that are
839 * still under construction).
840 *
841 * Note also that we were given an increment on net->cells_outstanding by
842 * whoever queued us that we need to deal with before returning.
843 */
844void afs_manage_cells(struct work_struct *work)
845{
846 struct afs_net *net = container_of(work, struct afs_net, cells_manager);
847 struct rb_node *cursor;
848 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
849 bool purging = !net->live;
850
851 _enter("");
852
853 /* Trawl the cell database looking for cells that have expired from
854 * lack of use and cells whose DNS results have expired and dispatch
855 * their managers.
856 */
857 down_read(&net->cells_lock);
858
859 for (cursor = rb_first(&net->cells); cursor; cursor = rb_next(cursor)) {
860 struct afs_cell *cell =
861 rb_entry(cursor, struct afs_cell, net_node);
862 unsigned active;
863 bool sched_cell = false;
864
865 active = atomic_read(&cell->active);
866 trace_afs_cell(cell->debug_id, refcount_read(&cell->ref),
867 active, afs_cell_trace_manage);
868
869 ASSERTCMP(active, >=, 1);
870
871 if (purging) {
872 if (test_and_clear_bit(AFS_CELL_FL_NO_GC, &cell->flags)) {
873 active = atomic_dec_return(&cell->active);
874 trace_afs_cell(cell->debug_id, refcount_read(&cell->ref),
875 active, afs_cell_trace_unuse_pin);
876 }
877 }
878
879 if (active == 1) {
880 struct afs_vlserver_list *vllist;
881 time64_t expire_at = cell->last_inactive;
882
883 read_lock(&cell->vl_servers_lock);
884 vllist = rcu_dereference_protected(
885 cell->vl_servers,
886 lockdep_is_held(&cell->vl_servers_lock));
887 if (vllist->nr_servers > 0)
888 expire_at += afs_cell_gc_delay;
889 read_unlock(&cell->vl_servers_lock);
890 if (purging || expire_at <= now)
891 sched_cell = true;
892 else if (expire_at < next_manage)
893 next_manage = expire_at;
894 }
895
896 if (!purging) {
897 if (test_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags))
898 sched_cell = true;
899 }
900
901 if (sched_cell)
902 afs_queue_cell(cell, afs_cell_trace_get_queue_manage);
903 }
904
905 up_read(&net->cells_lock);
906
907 /* Update the timer on the way out. We have to pass an increment on
908 * cells_outstanding in the namespace that we are in to the timer or
909 * the work scheduler.
910 */
911 if (!purging && next_manage < TIME64_MAX) {
912 now = ktime_get_real_seconds();
913
914 if (next_manage - now <= 0) {
915 if (queue_work(afs_wq, &net->cells_manager))
916 atomic_inc(&net->cells_outstanding);
917 } else {
918 afs_set_cell_timer(net, next_manage - now);
919 }
920 }
921
922 afs_dec_cells_outstanding(net);
923 _leave(" [%d]", atomic_read(&net->cells_outstanding));
924}
925
926/*
927 * Purge in-memory cell database.
928 */
929void afs_cell_purge(struct afs_net *net)
930{
931 struct afs_cell *ws;
932
933 _enter("");
934
935 down_write(&net->cells_lock);
936 ws = net->ws_cell;
937 net->ws_cell = NULL;
938 up_write(&net->cells_lock);
939 afs_unuse_cell(net, ws, afs_cell_trace_unuse_ws);
940
941 _debug("del timer");
942 if (del_timer_sync(&net->cells_timer))
943 atomic_dec(&net->cells_outstanding);
944
945 _debug("kick mgr");
946 afs_queue_cell_manager(net);
947
948 _debug("wait");
949 wait_var_event(&net->cells_outstanding,
950 !atomic_read(&net->cells_outstanding));
951 _leave("");
952}