]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/afs/cell.c
afs: Overhaul cell database management
[mirror_ubuntu-bionic-kernel.git] / fs / afs / cell.c
1 /* AFS cell and server record management
2 *
3 * Copyright (C) 2002, 2017 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/module.h>
13 #include <linux/slab.h>
14 #include <linux/key.h>
15 #include <linux/ctype.h>
16 #include <linux/dns_resolver.h>
17 #include <linux/sched.h>
18 #include <linux/inet.h>
19 #include <keys/rxrpc-type.h>
20 #include "internal.h"
21
22 unsigned __read_mostly afs_cell_gc_delay = 10;
23
24 static void afs_manage_cell(struct work_struct *);
25
26 static void afs_dec_cells_outstanding(struct afs_net *net)
27 {
28 if (atomic_dec_and_test(&net->cells_outstanding))
29 wake_up_atomic_t(&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 */
36 static 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 }
43 }
44
45 /*
46 * Look up and get an activation reference on a cell record under RCU
47 * conditions. The caller must hold the RCU read lock.
48 */
49 struct afs_cell *afs_lookup_cell_rcu(struct afs_net *net,
50 const char *name, unsigned int namesz)
51 {
52 struct afs_cell *cell = NULL;
53 struct rb_node *p;
54 int n, seq = 0, ret = 0;
55
56 _enter("%*.*s", namesz, namesz, name);
57
58 if (name && namesz == 0)
59 return ERR_PTR(-EINVAL);
60 if (namesz > AFS_MAXCELLNAME)
61 return ERR_PTR(-ENAMETOOLONG);
62
63 do {
64 /* Unfortunately, rbtree walking doesn't give reliable results
65 * under just the RCU read lock, so we have to check for
66 * changes.
67 */
68 if (cell)
69 afs_put_cell(net, cell);
70 cell = NULL;
71 ret = -ENOENT;
72
73 read_seqbegin_or_lock(&net->cells_lock, &seq);
74
75 if (!name) {
76 cell = rcu_dereference_raw(net->ws_cell);
77 if (cell) {
78 afs_get_cell(cell);
79 continue;
80 }
81 ret = -EDESTADDRREQ;
82 continue;
83 }
84
85 p = rcu_dereference_raw(net->cells.rb_node);
86 while (p) {
87 cell = rb_entry(p, struct afs_cell, net_node);
88
89 n = strncasecmp(cell->name, name,
90 min_t(size_t, cell->name_len, namesz));
91 if (n == 0)
92 n = cell->name_len - namesz;
93 if (n < 0) {
94 p = rcu_dereference_raw(p->rb_left);
95 } else if (n > 0) {
96 p = rcu_dereference_raw(p->rb_right);
97 } else {
98 if (atomic_inc_not_zero(&cell->usage)) {
99 ret = 0;
100 break;
101 }
102 /* We want to repeat the search, this time with
103 * the lock properly locked.
104 */
105 }
106 cell = NULL;
107 }
108
109 } while (need_seqretry(&net->cells_lock, seq));
110
111 done_seqretry(&net->cells_lock, seq);
112
113 return ret == 0 ? cell : ERR_PTR(ret);
114 }
115
116 /*
117 * Set up a cell record and fill in its name, VL server address list and
118 * allocate an anonymous key
119 */
120 static struct afs_cell *afs_alloc_cell(struct afs_net *net,
121 const char *name, unsigned int namelen,
122 const char *vllist)
123 {
124 struct afs_cell *cell;
125 int i, ret;
126
127 ASSERT(name);
128 if (namelen == 0)
129 return ERR_PTR(-EINVAL);
130 if (namelen > AFS_MAXCELLNAME) {
131 _leave(" = -ENAMETOOLONG");
132 return ERR_PTR(-ENAMETOOLONG);
133 }
134
135 _enter("%*.*s,%s", namelen, namelen, name, vllist);
136
137 cell = kzalloc(sizeof(struct afs_cell), GFP_KERNEL);
138 if (!cell) {
139 _leave(" = -ENOMEM");
140 return ERR_PTR(-ENOMEM);
141 }
142
143 cell->net = net;
144 cell->name_len = namelen;
145 for (i = 0; i < namelen; i++)
146 cell->name[i] = tolower(name[i]);
147
148 atomic_set(&cell->usage, 2);
149 INIT_WORK(&cell->manager, afs_manage_cell);
150 rwlock_init(&cell->servers_lock);
151 INIT_LIST_HEAD(&cell->servers);
152 init_rwsem(&cell->vl_sem);
153 INIT_LIST_HEAD(&cell->vl_list);
154 spin_lock_init(&cell->vl_lock);
155 seqlock_init(&cell->vl_addrs_lock);
156 cell->flags = (1 << AFS_CELL_FL_NOT_READY);
157
158 for (i = 0; i < AFS_CELL_MAX_ADDRS; i++) {
159 struct sockaddr_rxrpc *srx = &cell->vl_addrs[i];
160 srx->srx_family = AF_RXRPC;
161 srx->srx_service = VL_SERVICE;
162 srx->transport_type = SOCK_DGRAM;
163 srx->transport.sin6.sin6_family = AF_INET6;
164 srx->transport.sin6.sin6_port = htons(AFS_VL_PORT);
165 }
166
167 /* Fill in the VL server list if we were given a list of addresses to
168 * use.
169 */
170 if (vllist) {
171 char delim = ':';
172
173 if (strchr(vllist, ',') || !strchr(vllist, '.'))
174 delim = ',';
175
176 do {
177 struct sockaddr_rxrpc *srx = &cell->vl_addrs[cell->vl_naddrs];
178
179 if (in4_pton(vllist, -1,
180 (u8 *)&srx->transport.sin6.sin6_addr.s6_addr32[3],
181 delim, &vllist)) {
182 srx->transport_len = sizeof(struct sockaddr_in6);
183 srx->transport.sin6.sin6_addr.s6_addr32[0] = 0;
184 srx->transport.sin6.sin6_addr.s6_addr32[1] = 0;
185 srx->transport.sin6.sin6_addr.s6_addr32[2] = htonl(0xffff);
186 } else if (in6_pton(vllist, -1,
187 srx->transport.sin6.sin6_addr.s6_addr,
188 delim, &vllist)) {
189 srx->transport_len = sizeof(struct sockaddr_in6);
190 srx->transport.sin6.sin6_family = AF_INET6;
191 } else {
192 goto bad_address;
193 }
194
195 cell->vl_naddrs++;
196 if (!*vllist)
197 break;
198 vllist++;
199
200 } while (cell->vl_naddrs < AFS_CELL_MAX_ADDRS && vllist);
201
202 /* Disable DNS refresh for manually-specified cells */
203 cell->dns_expiry = TIME64_MAX;
204 } else {
205 /* We're going to need to 'refresh' this cell's VL server list
206 * from the DNS before we can use it.
207 */
208 cell->dns_expiry = S64_MIN;
209 }
210
211 _leave(" = %p", cell);
212 return cell;
213
214 bad_address:
215 printk(KERN_ERR "kAFS: bad VL server IP address\n");
216 ret = -EINVAL;
217 kfree(cell);
218 _leave(" = %d", ret);
219 return ERR_PTR(ret);
220 }
221
222 /*
223 * afs_lookup_cell - Look up or create a cell record.
224 * @net: The network namespace
225 * @name: The name of the cell.
226 * @namesz: The strlen of the cell name.
227 * @vllist: A colon/comma separated list of numeric IP addresses or NULL.
228 * @excl: T if an error should be given if the cell name already exists.
229 *
230 * Look up a cell record by name and query the DNS for VL server addresses if
231 * needed. Note that that actual DNS query is punted off to the manager thread
232 * so that this function can return immediately if interrupted whilst allowing
233 * cell records to be shared even if not yet fully constructed.
234 */
235 struct afs_cell *afs_lookup_cell(struct afs_net *net,
236 const char *name, unsigned int namesz,
237 const char *vllist, bool excl)
238 {
239 struct afs_cell *cell, *candidate, *cursor;
240 struct rb_node *parent, **pp;
241 int ret, n;
242
243 _enter("%s,%s", name, vllist);
244
245 if (!excl) {
246 rcu_read_lock();
247 cell = afs_lookup_cell_rcu(net, name, namesz);
248 rcu_read_unlock();
249 if (!IS_ERR(cell)) {
250 if (excl) {
251 afs_put_cell(net, cell);
252 return ERR_PTR(-EEXIST);
253 }
254 goto wait_for_cell;
255 }
256 }
257
258 /* Assume we're probably going to create a cell and preallocate and
259 * mostly set up a candidate record. We can then use this to stash the
260 * name, the net namespace and VL server addresses.
261 *
262 * We also want to do this before we hold any locks as it may involve
263 * upcalling to userspace to make DNS queries.
264 */
265 candidate = afs_alloc_cell(net, name, namesz, vllist);
266 if (IS_ERR(candidate)) {
267 _leave(" = %ld", PTR_ERR(candidate));
268 return candidate;
269 }
270
271 /* Find the insertion point and check to see if someone else added a
272 * cell whilst we were allocating.
273 */
274 write_seqlock(&net->cells_lock);
275
276 pp = &net->cells.rb_node;
277 parent = NULL;
278 while (*pp) {
279 parent = *pp;
280 cursor = rb_entry(parent, struct afs_cell, net_node);
281
282 n = strncasecmp(cursor->name, name,
283 min_t(size_t, cursor->name_len, namesz));
284 if (n == 0)
285 n = cursor->name_len - namesz;
286 if (n < 0)
287 pp = &(*pp)->rb_left;
288 else if (n > 0)
289 pp = &(*pp)->rb_right;
290 else
291 goto cell_already_exists;
292 }
293
294 cell = candidate;
295 candidate = NULL;
296 rb_link_node_rcu(&cell->net_node, parent, pp);
297 rb_insert_color(&cell->net_node, &net->cells);
298 atomic_inc(&net->cells_outstanding);
299 write_sequnlock(&net->cells_lock);
300
301 queue_work(afs_wq, &cell->manager);
302
303 wait_for_cell:
304 _debug("wait_for_cell");
305 ret = wait_on_bit(&cell->flags, AFS_CELL_FL_NOT_READY, TASK_INTERRUPTIBLE);
306 smp_rmb();
307
308 switch (READ_ONCE(cell->state)) {
309 case AFS_CELL_FAILED:
310 ret = cell->error;
311 goto error;
312 default:
313 _debug("weird %u %d", cell->state, cell->error);
314 goto error;
315 case AFS_CELL_ACTIVE:
316 break;
317 }
318
319 _leave(" = %p [cell]", cell);
320 return cell;
321
322 cell_already_exists:
323 _debug("cell exists");
324 cell = cursor;
325 if (excl) {
326 ret = -EEXIST;
327 } else {
328 ASSERTCMP(atomic_read(&cursor->usage), >=, 1);
329 afs_get_cell(cursor);
330 ret = 0;
331 }
332 write_sequnlock(&net->cells_lock);
333 kfree(candidate);
334 if (ret == 0)
335 goto wait_for_cell;
336 error:
337 afs_put_cell(net, cell);
338 _leave(" = %d [error]", ret);
339 return ERR_PTR(ret);
340 }
341
342 /*
343 * set the root cell information
344 * - can be called with a module parameter string
345 * - can be called from a write to /proc/fs/afs/rootcell
346 */
347 int afs_cell_init(struct afs_net *net, const char *rootcell)
348 {
349 struct afs_cell *old_root, *new_root;
350 const char *cp, *vllist;
351 size_t len;
352
353 _enter("");
354
355 if (!rootcell) {
356 /* module is loaded with no parameters, or built statically.
357 * - in the future we might initialize cell DB here.
358 */
359 _leave(" = 0 [no root]");
360 return 0;
361 }
362
363 cp = strchr(rootcell, ':');
364 if (!cp) {
365 _debug("kAFS: no VL server IP addresses specified");
366 vllist = NULL;
367 len = strlen(rootcell);
368 } else {
369 vllist = cp + 1;
370 len = cp - rootcell;
371 }
372
373 /* allocate a cell record for the root cell */
374 new_root = afs_lookup_cell(net, rootcell, len, vllist, false);
375 if (IS_ERR(new_root)) {
376 _leave(" = %ld", PTR_ERR(new_root));
377 return PTR_ERR(new_root);
378 }
379
380 set_bit(AFS_CELL_FL_NO_GC, &new_root->flags);
381 afs_get_cell(new_root);
382
383 /* install the new cell */
384 write_seqlock(&net->cells_lock);
385 old_root = net->ws_cell;
386 net->ws_cell = new_root;
387 write_sequnlock(&net->cells_lock);
388
389 afs_put_cell(net, old_root);
390 _leave(" = 0");
391 return 0;
392 }
393
394 /*
395 * Update a cell's VL server address list from the DNS.
396 */
397 static void afs_update_cell(struct afs_cell *cell)
398 {
399 time64_t now, expiry;
400 char *vllist = NULL;
401 int ret;
402
403 _enter("%s", cell->name);
404
405 ret = dns_query("afsdb", cell->name, cell->name_len,
406 "ipv4", &vllist, &expiry);
407 _debug("query %d", ret);
408 switch (ret) {
409 case 0 ... INT_MAX:
410 clear_bit(AFS_CELL_FL_DNS_FAIL, &cell->flags);
411 clear_bit(AFS_CELL_FL_NOT_FOUND, &cell->flags);
412 goto parse_dns_data;
413
414 case -ENODATA:
415 clear_bit(AFS_CELL_FL_DNS_FAIL, &cell->flags);
416 set_bit(AFS_CELL_FL_NOT_FOUND, &cell->flags);
417 cell->dns_expiry = ktime_get_real_seconds() + 61;
418 cell->error = -EDESTADDRREQ;
419 goto out;
420
421 case -EAGAIN:
422 case -ECONNREFUSED:
423 default:
424 /* Unable to query DNS. */
425 set_bit(AFS_CELL_FL_DNS_FAIL, &cell->flags);
426 cell->dns_expiry = ktime_get_real_seconds() + 10;
427 cell->error = -EDESTADDRREQ;
428 goto out;
429 }
430
431 parse_dns_data:
432 write_seqlock(&cell->vl_addrs_lock);
433
434 ret = -EINVAL;
435 do {
436 struct sockaddr_rxrpc *srx = &cell->vl_addrs[cell->vl_naddrs];
437
438 if (in4_pton(vllist, -1,
439 (u8 *)&srx->transport.sin6.sin6_addr.s6_addr32[3],
440 ',', (const char **)&vllist)) {
441 srx->transport_len = sizeof(struct sockaddr_in6);
442 srx->transport.sin6.sin6_addr.s6_addr32[0] = 0;
443 srx->transport.sin6.sin6_addr.s6_addr32[1] = 0;
444 srx->transport.sin6.sin6_addr.s6_addr32[2] = htonl(0xffff);
445 } else if (in6_pton(vllist, -1,
446 srx->transport.sin6.sin6_addr.s6_addr,
447 ',', (const char **)&vllist)) {
448 srx->transport_len = sizeof(struct sockaddr_in6);
449 srx->transport.sin6.sin6_family = AF_INET6;
450 } else {
451 goto bad_address;
452 }
453
454 cell->vl_naddrs++;
455 if (!*vllist)
456 break;
457 vllist++;
458
459 } while (cell->vl_naddrs < AFS_CELL_MAX_ADDRS);
460
461 if (cell->vl_naddrs < AFS_CELL_MAX_ADDRS)
462 memset(cell->vl_addrs + cell->vl_naddrs, 0,
463 (AFS_CELL_MAX_ADDRS - cell->vl_naddrs) * sizeof(cell->vl_addrs[0]));
464
465 now = ktime_get_real_seconds();
466 cell->dns_expiry = expiry;
467 afs_set_cell_timer(cell->net, expiry - now);
468 bad_address:
469 write_sequnlock(&cell->vl_addrs_lock);
470 out:
471 _leave("");
472 }
473
474 /*
475 * Destroy a cell record
476 */
477 static void afs_cell_destroy(struct rcu_head *rcu)
478 {
479 struct afs_cell *cell = container_of(rcu, struct afs_cell, rcu);
480
481 _enter("%p{%s}", cell, cell->name);
482
483 ASSERTCMP(atomic_read(&cell->usage), ==, 0);
484
485 key_put(cell->anonymous_key);
486 kfree(cell);
487
488 _leave(" [destroyed]");
489 }
490
491 /*
492 * Queue the cell manager.
493 */
494 static void afs_queue_cell_manager(struct afs_net *net)
495 {
496 int outstanding = atomic_inc_return(&net->cells_outstanding);
497
498 _enter("%d", outstanding);
499
500 if (!queue_work(afs_wq, &net->cells_manager))
501 afs_dec_cells_outstanding(net);
502 }
503
504 /*
505 * Cell management timer. We have an increment on cells_outstanding that we
506 * need to pass along to the work item.
507 */
508 void afs_cells_timer(struct timer_list *timer)
509 {
510 struct afs_net *net = container_of(timer, struct afs_net, cells_timer);
511
512 _enter("");
513 if (!queue_work(afs_wq, &net->cells_manager))
514 afs_dec_cells_outstanding(net);
515 }
516
517 /*
518 * Drop a reference on a cell record.
519 */
520 void afs_put_cell(struct afs_net *net, struct afs_cell *cell)
521 {
522 time64_t now, expire_delay;
523
524 if (!cell)
525 return;
526
527 _enter("%s", cell->name);
528
529 now = ktime_get_real_seconds();
530 cell->last_inactive = now;
531 expire_delay = 0;
532 if (!test_bit(AFS_CELL_FL_DNS_FAIL, &cell->flags) &&
533 !test_bit(AFS_CELL_FL_NOT_FOUND, &cell->flags))
534 expire_delay = afs_cell_gc_delay;
535
536 if (atomic_dec_return(&cell->usage) > 1)
537 return;
538
539 /* 'cell' may now be garbage collected. */
540 afs_set_cell_timer(net, expire_delay);
541 }
542
543 /*
544 * Allocate a key to use as a placeholder for anonymous user security.
545 */
546 static int afs_alloc_anon_key(struct afs_cell *cell)
547 {
548 struct key *key;
549 char keyname[4 + AFS_MAXCELLNAME + 1], *cp, *dp;
550
551 /* Create a key to represent an anonymous user. */
552 memcpy(keyname, "afs@", 4);
553 dp = keyname + 4;
554 cp = cell->name;
555 do {
556 *dp++ = tolower(*cp);
557 } while (*cp++);
558
559 key = rxrpc_get_null_key(keyname);
560 if (IS_ERR(key))
561 return PTR_ERR(key);
562
563 cell->anonymous_key = key;
564
565 _debug("anon key %p{%x}",
566 cell->anonymous_key, key_serial(cell->anonymous_key));
567 return 0;
568 }
569
570 /*
571 * Activate a cell.
572 */
573 static int afs_activate_cell(struct afs_net *net, struct afs_cell *cell)
574 {
575 int ret;
576
577 if (!cell->anonymous_key) {
578 ret = afs_alloc_anon_key(cell);
579 if (ret < 0)
580 return ret;
581 }
582
583 #ifdef CONFIG_AFS_FSCACHE
584 cell->cache = fscache_acquire_cookie(afs_cache_netfs.primary_index,
585 &afs_cell_cache_index_def,
586 cell, true);
587 #endif
588 ret = afs_proc_cell_setup(net, cell);
589 if (ret < 0)
590 return ret;
591 spin_lock(&net->proc_cells_lock);
592 list_add_tail(&cell->proc_link, &net->proc_cells);
593 spin_unlock(&net->proc_cells_lock);
594 return 0;
595 }
596
597 /*
598 * Deactivate a cell.
599 */
600 static void afs_deactivate_cell(struct afs_net *net, struct afs_cell *cell)
601 {
602 _enter("%s", cell->name);
603
604 afs_proc_cell_remove(net, cell);
605
606 spin_lock(&net->proc_cells_lock);
607 list_del_init(&cell->proc_link);
608 spin_unlock(&net->proc_cells_lock);
609
610 #ifdef CONFIG_AFS_FSCACHE
611 fscache_relinquish_cookie(cell->cache, 0);
612 cell->cache = NULL;
613 #endif
614
615 _leave("");
616 }
617
618 /*
619 * Manage a cell record, initialising and destroying it, maintaining its DNS
620 * records.
621 */
622 static void afs_manage_cell(struct work_struct *work)
623 {
624 struct afs_cell *cell = container_of(work, struct afs_cell, manager);
625 struct afs_net *net = cell->net;
626 bool deleted;
627 int ret, usage;
628
629 _enter("%s", cell->name);
630
631 again:
632 _debug("state %u", cell->state);
633 switch (cell->state) {
634 case AFS_CELL_INACTIVE:
635 case AFS_CELL_FAILED:
636 write_seqlock(&net->cells_lock);
637 usage = 1;
638 deleted = atomic_try_cmpxchg_relaxed(&cell->usage, &usage, 0);
639 if (deleted)
640 rb_erase(&cell->net_node, &net->cells);
641 write_sequnlock(&net->cells_lock);
642 if (deleted)
643 goto final_destruction;
644 if (cell->state == AFS_CELL_FAILED)
645 goto done;
646 cell->state = AFS_CELL_UNSET;
647 goto again;
648
649 case AFS_CELL_UNSET:
650 cell->state = AFS_CELL_ACTIVATING;
651 goto again;
652
653 case AFS_CELL_ACTIVATING:
654 ret = afs_activate_cell(net, cell);
655 if (ret < 0)
656 goto activation_failed;
657
658 cell->state = AFS_CELL_ACTIVE;
659 smp_wmb();
660 clear_bit(AFS_CELL_FL_NOT_READY, &cell->flags);
661 wake_up_bit(&cell->flags, AFS_CELL_FL_NOT_READY);
662 goto again;
663
664 case AFS_CELL_ACTIVE:
665 if (atomic_read(&cell->usage) > 1) {
666 time64_t now = ktime_get_real_seconds();
667 if (cell->dns_expiry <= now && net->live)
668 afs_update_cell(cell);
669 goto done;
670 }
671 cell->state = AFS_CELL_DEACTIVATING;
672 goto again;
673
674 case AFS_CELL_DEACTIVATING:
675 set_bit(AFS_CELL_FL_NOT_READY, &cell->flags);
676 if (atomic_read(&cell->usage) > 1)
677 goto reverse_deactivation;
678 afs_deactivate_cell(net, cell);
679 cell->state = AFS_CELL_INACTIVE;
680 goto again;
681
682 default:
683 break;
684 }
685 _debug("bad state %u", cell->state);
686 BUG(); /* Unhandled state */
687
688 activation_failed:
689 cell->error = ret;
690 afs_deactivate_cell(net, cell);
691
692 cell->state = AFS_CELL_FAILED;
693 smp_wmb();
694 if (test_and_clear_bit(AFS_CELL_FL_NOT_READY, &cell->flags))
695 wake_up_bit(&cell->flags, AFS_CELL_FL_NOT_READY);
696 goto again;
697
698 reverse_deactivation:
699 cell->state = AFS_CELL_ACTIVE;
700 smp_wmb();
701 clear_bit(AFS_CELL_FL_NOT_READY, &cell->flags);
702 wake_up_bit(&cell->flags, AFS_CELL_FL_NOT_READY);
703 _leave(" [deact->act]");
704 return;
705
706 done:
707 _leave(" [done %u]", cell->state);
708 return;
709
710 final_destruction:
711 call_rcu(&cell->rcu, afs_cell_destroy);
712 afs_dec_cells_outstanding(net);
713 _leave(" [destruct %d]", atomic_read(&net->cells_outstanding));
714 }
715
716 /*
717 * Manage the records of cells known to a network namespace. This includes
718 * updating the DNS records and garbage collecting unused cells that were
719 * automatically added.
720 *
721 * Note that constructed cell records may only be removed from net->cells by
722 * this work item, so it is safe for this work item to stash a cursor pointing
723 * into the tree and then return to caller (provided it skips cells that are
724 * still under construction).
725 *
726 * Note also that we were given an increment on net->cells_outstanding by
727 * whoever queued us that we need to deal with before returning.
728 */
729 void afs_manage_cells(struct work_struct *work)
730 {
731 struct afs_net *net = container_of(work, struct afs_net, cells_manager);
732 struct rb_node *cursor;
733 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
734 bool purging = !net->live;
735
736 _enter("");
737
738 /* Trawl the cell database looking for cells that have expired from
739 * lack of use and cells whose DNS results have expired and dispatch
740 * their managers.
741 */
742 read_seqlock_excl(&net->cells_lock);
743
744 for (cursor = rb_first(&net->cells); cursor; cursor = rb_next(cursor)) {
745 struct afs_cell *cell =
746 rb_entry(cursor, struct afs_cell, net_node);
747 unsigned usage;
748 bool sched_cell = false;
749
750 usage = atomic_read(&cell->usage);
751 _debug("manage %s %u", cell->name, usage);
752
753 ASSERTCMP(usage, >=, 1);
754
755 if (purging) {
756 if (test_and_clear_bit(AFS_CELL_FL_NO_GC, &cell->flags))
757 usage = atomic_dec_return(&cell->usage);
758 ASSERTCMP(usage, ==, 1);
759 }
760
761 if (usage == 1) {
762 time64_t expire_at = cell->last_inactive;
763
764 if (!test_bit(AFS_CELL_FL_DNS_FAIL, &cell->flags) &&
765 !test_bit(AFS_CELL_FL_NOT_FOUND, &cell->flags))
766 expire_at += afs_cell_gc_delay;
767 if (purging || expire_at <= now)
768 sched_cell = true;
769 else if (expire_at < next_manage)
770 next_manage = expire_at;
771 }
772
773 if (!purging) {
774 if (cell->dns_expiry <= now)
775 sched_cell = true;
776 else if (cell->dns_expiry <= next_manage)
777 next_manage = cell->dns_expiry;
778 }
779
780 if (sched_cell)
781 queue_work(afs_wq, &cell->manager);
782 }
783
784 read_sequnlock_excl(&net->cells_lock);
785
786 /* Update the timer on the way out. We have to pass an increment on
787 * cells_outstanding in the namespace that we are in to the timer or
788 * the work scheduler.
789 */
790 if (!purging && next_manage < TIME64_MAX) {
791 now = ktime_get_real_seconds();
792
793 if (next_manage - now <= 0) {
794 if (queue_work(afs_wq, &net->cells_manager))
795 atomic_inc(&net->cells_outstanding);
796 } else {
797 afs_set_cell_timer(net, next_manage - now);
798 }
799 }
800
801 afs_dec_cells_outstanding(net);
802 _leave(" [%d]", atomic_read(&net->cells_outstanding));
803 }
804
805 /*
806 * Purge in-memory cell database.
807 */
808 void afs_cell_purge(struct afs_net *net)
809 {
810 struct afs_cell *ws;
811
812 _enter("");
813
814 write_seqlock(&net->cells_lock);
815 ws = net->ws_cell;
816 net->ws_cell = NULL;
817 write_sequnlock(&net->cells_lock);
818 afs_put_cell(net, ws);
819
820 _debug("del timer");
821 if (del_timer_sync(&net->cells_timer))
822 atomic_dec(&net->cells_outstanding);
823
824 _debug("kick mgr");
825 afs_queue_cell_manager(net);
826
827 _debug("wait");
828 wait_on_atomic_t(&net->cells_outstanding, atomic_t_wait,
829 TASK_UNINTERRUPTIBLE);
830 _leave("");
831 }