xref: /linux/fs/afs/cell.c (revision f2bf88c4afc8c5ab92b40af24819933e57d0968c)
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 
18 static unsigned __read_mostly afs_cell_gc_delay = 10;
19 static unsigned __read_mostly afs_cell_min_ttl = 10 * 60;
20 static unsigned __read_mostly afs_cell_max_ttl = 24 * 60 * 60;
21 
22 static void afs_manage_cell(struct work_struct *);
23 
24 static void afs_dec_cells_outstanding(struct afs_net *net)
25 {
26 	if (atomic_dec_and_test(&net->cells_outstanding))
27 		wake_up_var(&net->cells_outstanding);
28 }
29 
30 /*
31  * Set the cell timer to fire after a given delay, assuming it's not already
32  * set for an earlier time.
33  */
34 static void afs_set_cell_timer(struct afs_net *net, time64_t delay)
35 {
36 	if (net->live) {
37 		atomic_inc(&net->cells_outstanding);
38 		if (timer_reduce(&net->cells_timer, jiffies + delay * HZ))
39 			afs_dec_cells_outstanding(net);
40 	}
41 }
42 
43 /*
44  * Look up and get an activation reference on a cell record under RCU
45  * conditions.  The caller must hold the RCU read lock.
46  */
47 struct afs_cell *afs_lookup_cell_rcu(struct afs_net *net,
48 				     const char *name, unsigned int namesz)
49 {
50 	struct afs_cell *cell = NULL;
51 	struct rb_node *p;
52 	int n, seq = 0, ret = 0;
53 
54 	_enter("%*.*s", namesz, namesz, name);
55 
56 	if (name && namesz == 0)
57 		return ERR_PTR(-EINVAL);
58 	if (namesz > AFS_MAXCELLNAME)
59 		return ERR_PTR(-ENAMETOOLONG);
60 
61 	do {
62 		/* Unfortunately, rbtree walking doesn't give reliable results
63 		 * under just the RCU read lock, so we have to check for
64 		 * changes.
65 		 */
66 		if (cell)
67 			afs_put_cell(net, cell);
68 		cell = NULL;
69 		ret = -ENOENT;
70 
71 		read_seqbegin_or_lock(&net->cells_lock, &seq);
72 
73 		if (!name) {
74 			cell = rcu_dereference_raw(net->ws_cell);
75 			if (cell) {
76 				afs_get_cell(cell);
77 				ret = 0;
78 				break;
79 			}
80 			ret = -EDESTADDRREQ;
81 			continue;
82 		}
83 
84 		p = rcu_dereference_raw(net->cells.rb_node);
85 		while (p) {
86 			cell = rb_entry(p, struct afs_cell, net_node);
87 
88 			n = strncasecmp(cell->name, name,
89 					min_t(size_t, cell->name_len, namesz));
90 			if (n == 0)
91 				n = cell->name_len - namesz;
92 			if (n < 0) {
93 				p = rcu_dereference_raw(p->rb_left);
94 			} else if (n > 0) {
95 				p = rcu_dereference_raw(p->rb_right);
96 			} else {
97 				if (atomic_inc_not_zero(&cell->usage)) {
98 					ret = 0;
99 					break;
100 				}
101 				/* We want to repeat the search, this time with
102 				 * the lock properly locked.
103 				 */
104 			}
105 			cell = NULL;
106 		}
107 
108 	} while (need_seqretry(&net->cells_lock, seq));
109 
110 	done_seqretry(&net->cells_lock, seq);
111 
112 	if (ret != 0 && cell)
113 		afs_put_cell(net, cell);
114 
115 	return ret == 0 ? cell : ERR_PTR(ret);
116 }
117 
118 /*
119  * Set up a cell record and fill in its name, VL server address list and
120  * allocate an anonymous key
121  */
122 static struct afs_cell *afs_alloc_cell(struct afs_net *net,
123 				       const char *name, unsigned int namelen,
124 				       const char *addresses)
125 {
126 	struct afs_vlserver_list *vllist;
127 	struct afs_cell *cell;
128 	int i, ret;
129 
130 	ASSERT(name);
131 	if (namelen == 0)
132 		return ERR_PTR(-EINVAL);
133 	if (namelen > AFS_MAXCELLNAME) {
134 		_leave(" = -ENAMETOOLONG");
135 		return ERR_PTR(-ENAMETOOLONG);
136 	}
137 
138 	/* Prohibit cell names that contain unprintable chars, '/' and '@' or
139 	 * that begin with a dot.  This also precludes "@cell".
140 	 */
141 	if (name[0] == '.')
142 		return ERR_PTR(-EINVAL);
143 	for (i = 0; i < namelen; i++) {
144 		char ch = name[i];
145 		if (!isprint(ch) || ch == '/' || ch == '@')
146 			return ERR_PTR(-EINVAL);
147 	}
148 
149 	_enter("%*.*s,%s", namelen, namelen, name, addresses);
150 
151 	cell = kzalloc(sizeof(struct afs_cell), GFP_KERNEL);
152 	if (!cell) {
153 		_leave(" = -ENOMEM");
154 		return ERR_PTR(-ENOMEM);
155 	}
156 
157 	cell->name = kmalloc(namelen + 1, GFP_KERNEL);
158 	if (!cell->name) {
159 		kfree(cell);
160 		return ERR_PTR(-ENOMEM);
161 	}
162 
163 	cell->net = net;
164 	cell->name_len = namelen;
165 	for (i = 0; i < namelen; i++)
166 		cell->name[i] = tolower(name[i]);
167 	cell->name[i] = 0;
168 
169 	atomic_set(&cell->usage, 2);
170 	INIT_WORK(&cell->manager, afs_manage_cell);
171 	cell->volumes = RB_ROOT;
172 	INIT_HLIST_HEAD(&cell->proc_volumes);
173 	seqlock_init(&cell->volume_lock);
174 	cell->fs_servers = RB_ROOT;
175 	seqlock_init(&cell->fs_lock);
176 	rwlock_init(&cell->vl_servers_lock);
177 	cell->flags = (1 << AFS_CELL_FL_CHECK_ALIAS);
178 
179 	/* Provide a VL server list, filling it in if we were given a list of
180 	 * addresses to use.
181 	 */
182 	if (addresses) {
183 		vllist = afs_parse_text_addrs(net,
184 					      addresses, strlen(addresses), ':',
185 					      VL_SERVICE, AFS_VL_PORT);
186 		if (IS_ERR(vllist)) {
187 			ret = PTR_ERR(vllist);
188 			goto parse_failed;
189 		}
190 
191 		vllist->source = DNS_RECORD_FROM_CONFIG;
192 		vllist->status = DNS_LOOKUP_NOT_DONE;
193 		cell->dns_expiry = TIME64_MAX;
194 	} else {
195 		ret = -ENOMEM;
196 		vllist = afs_alloc_vlserver_list(0);
197 		if (!vllist)
198 			goto error;
199 		vllist->source = DNS_RECORD_UNAVAILABLE;
200 		vllist->status = DNS_LOOKUP_NOT_DONE;
201 		cell->dns_expiry = ktime_get_real_seconds();
202 	}
203 
204 	rcu_assign_pointer(cell->vl_servers, vllist);
205 
206 	cell->dns_source = vllist->source;
207 	cell->dns_status = vllist->status;
208 	smp_store_release(&cell->dns_lookup_count, 1); /* vs source/status */
209 
210 	_leave(" = %p", cell);
211 	return cell;
212 
213 parse_failed:
214 	if (ret == -EINVAL)
215 		printk(KERN_ERR "kAFS: bad VL server IP address\n");
216 error:
217 	kfree(cell->name);
218 	kfree(cell);
219 	_leave(" = %d", ret);
220 	return ERR_PTR(ret);
221 }
222 
223 /*
224  * afs_lookup_cell - Look up or create a cell record.
225  * @net:	The network namespace
226  * @name:	The name of the cell.
227  * @namesz:	The strlen of the cell name.
228  * @vllist:	A colon/comma separated list of numeric IP addresses or NULL.
229  * @excl:	T if an error should be given if the cell name already exists.
230  *
231  * Look up a cell record by name and query the DNS for VL server addresses if
232  * needed.  Note that that actual DNS query is punted off to the manager thread
233  * so that this function can return immediately if interrupted whilst allowing
234  * cell records to be shared even if not yet fully constructed.
235  */
236 struct afs_cell *afs_lookup_cell(struct afs_net *net,
237 				 const char *name, unsigned int namesz,
238 				 const char *vllist, bool excl)
239 {
240 	struct afs_cell *cell, *candidate, *cursor;
241 	struct rb_node *parent, **pp;
242 	enum afs_cell_state state;
243 	int ret, n;
244 
245 	_enter("%s,%s", name, vllist);
246 
247 	if (!excl) {
248 		rcu_read_lock();
249 		cell = afs_lookup_cell_rcu(net, name, namesz);
250 		rcu_read_unlock();
251 		if (!IS_ERR(cell))
252 			goto wait_for_cell;
253 	}
254 
255 	/* Assume we're probably going to create a cell and preallocate and
256 	 * mostly set up a candidate record.  We can then use this to stash the
257 	 * name, the net namespace and VL server addresses.
258 	 *
259 	 * We also want to do this before we hold any locks as it may involve
260 	 * upcalling to userspace to make DNS queries.
261 	 */
262 	candidate = afs_alloc_cell(net, name, namesz, vllist);
263 	if (IS_ERR(candidate)) {
264 		_leave(" = %ld", PTR_ERR(candidate));
265 		return candidate;
266 	}
267 
268 	/* Find the insertion point and check to see if someone else added a
269 	 * cell whilst we were allocating.
270 	 */
271 	write_seqlock(&net->cells_lock);
272 
273 	pp = &net->cells.rb_node;
274 	parent = NULL;
275 	while (*pp) {
276 		parent = *pp;
277 		cursor = rb_entry(parent, struct afs_cell, net_node);
278 
279 		n = strncasecmp(cursor->name, name,
280 				min_t(size_t, cursor->name_len, namesz));
281 		if (n == 0)
282 			n = cursor->name_len - namesz;
283 		if (n < 0)
284 			pp = &(*pp)->rb_left;
285 		else if (n > 0)
286 			pp = &(*pp)->rb_right;
287 		else
288 			goto cell_already_exists;
289 	}
290 
291 	cell = candidate;
292 	candidate = NULL;
293 	rb_link_node_rcu(&cell->net_node, parent, pp);
294 	rb_insert_color(&cell->net_node, &net->cells);
295 	atomic_inc(&net->cells_outstanding);
296 	write_sequnlock(&net->cells_lock);
297 
298 	queue_work(afs_wq, &cell->manager);
299 
300 wait_for_cell:
301 	_debug("wait_for_cell");
302 	wait_var_event(&cell->state,
303 		       ({
304 			       state = smp_load_acquire(&cell->state); /* vs error */
305 			       state == AFS_CELL_ACTIVE || state == AFS_CELL_FAILED;
306 		       }));
307 
308 	/* Check the state obtained from the wait check. */
309 	if (state == AFS_CELL_FAILED) {
310 		ret = cell->error;
311 		goto error;
312 	}
313 
314 	_leave(" = %p [cell]", cell);
315 	return cell;
316 
317 cell_already_exists:
318 	_debug("cell exists");
319 	cell = cursor;
320 	if (excl) {
321 		ret = -EEXIST;
322 	} else {
323 		afs_get_cell(cursor);
324 		ret = 0;
325 	}
326 	write_sequnlock(&net->cells_lock);
327 	kfree(candidate);
328 	if (ret == 0)
329 		goto wait_for_cell;
330 	goto error_noput;
331 error:
332 	afs_put_cell(net, cell);
333 error_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  */
343 int 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_get_cell(new_root);
378 
379 	/* install the new cell */
380 	write_seqlock(&net->cells_lock);
381 	old_root = rcu_access_pointer(net->ws_cell);
382 	rcu_assign_pointer(net->ws_cell, new_root);
383 	write_sequnlock(&net->cells_lock);
384 
385 	afs_put_cell(net, old_root);
386 	_leave(" = 0");
387 	return 0;
388 }
389 
390 /*
391  * Update a cell's VL server address list from the DNS.
392  */
393 static int afs_update_cell(struct afs_cell *cell)
394 {
395 	struct afs_vlserver_list *vllist, *old = NULL, *p;
396 	unsigned int min_ttl = READ_ONCE(afs_cell_min_ttl);
397 	unsigned int max_ttl = READ_ONCE(afs_cell_max_ttl);
398 	time64_t now, expiry = 0;
399 	int ret = 0;
400 
401 	_enter("%s", cell->name);
402 
403 	vllist = afs_dns_query(cell, &expiry);
404 	if (IS_ERR(vllist)) {
405 		ret = PTR_ERR(vllist);
406 
407 		_debug("%s: fail %d", cell->name, ret);
408 		if (ret == -ENOMEM)
409 			goto out_wake;
410 
411 		ret = -ENOMEM;
412 		vllist = afs_alloc_vlserver_list(0);
413 		if (!vllist)
414 			goto out_wake;
415 
416 		switch (ret) {
417 		case -ENODATA:
418 		case -EDESTADDRREQ:
419 			vllist->status = DNS_LOOKUP_GOT_NOT_FOUND;
420 			break;
421 		case -EAGAIN:
422 		case -ECONNREFUSED:
423 			vllist->status = DNS_LOOKUP_GOT_TEMP_FAILURE;
424 			break;
425 		default:
426 			vllist->status = DNS_LOOKUP_GOT_LOCAL_FAILURE;
427 			break;
428 		}
429 	}
430 
431 	_debug("%s: got list %d %d", cell->name, vllist->source, vllist->status);
432 	cell->dns_status = vllist->status;
433 
434 	now = ktime_get_real_seconds();
435 	if (min_ttl > max_ttl)
436 		max_ttl = min_ttl;
437 	if (expiry < now + min_ttl)
438 		expiry = now + min_ttl;
439 	else if (expiry > now + max_ttl)
440 		expiry = now + max_ttl;
441 
442 	_debug("%s: status %d", cell->name, vllist->status);
443 	if (vllist->source == DNS_RECORD_UNAVAILABLE) {
444 		switch (vllist->status) {
445 		case DNS_LOOKUP_GOT_NOT_FOUND:
446 			/* The DNS said that the cell does not exist or there
447 			 * weren't any addresses to be had.
448 			 */
449 			cell->dns_expiry = expiry;
450 			break;
451 
452 		case DNS_LOOKUP_BAD:
453 		case DNS_LOOKUP_GOT_LOCAL_FAILURE:
454 		case DNS_LOOKUP_GOT_TEMP_FAILURE:
455 		case DNS_LOOKUP_GOT_NS_FAILURE:
456 		default:
457 			cell->dns_expiry = now + 10;
458 			break;
459 		}
460 	} else {
461 		cell->dns_expiry = expiry;
462 	}
463 
464 	/* Replace the VL server list if the new record has servers or the old
465 	 * record doesn't.
466 	 */
467 	write_lock(&cell->vl_servers_lock);
468 	p = rcu_dereference_protected(cell->vl_servers, true);
469 	if (vllist->nr_servers > 0 || p->nr_servers == 0) {
470 		rcu_assign_pointer(cell->vl_servers, vllist);
471 		cell->dns_source = vllist->source;
472 		old = p;
473 	}
474 	write_unlock(&cell->vl_servers_lock);
475 	afs_put_vlserverlist(cell->net, old);
476 
477 out_wake:
478 	smp_store_release(&cell->dns_lookup_count,
479 			  cell->dns_lookup_count + 1); /* vs source/status */
480 	wake_up_var(&cell->dns_lookup_count);
481 	_leave(" = %d", ret);
482 	return ret;
483 }
484 
485 /*
486  * Destroy a cell record
487  */
488 static void afs_cell_destroy(struct rcu_head *rcu)
489 {
490 	struct afs_cell *cell = container_of(rcu, struct afs_cell, rcu);
491 
492 	_enter("%p{%s}", cell, cell->name);
493 
494 	ASSERTCMP(atomic_read(&cell->usage), ==, 0);
495 
496 	afs_put_volume(cell->net, cell->root_volume, afs_volume_trace_put_cell_root);
497 	afs_put_vlserverlist(cell->net, rcu_access_pointer(cell->vl_servers));
498 	afs_put_cell(cell->net, cell->alias_of);
499 	key_put(cell->anonymous_key);
500 	kfree(cell->name);
501 	kfree(cell);
502 
503 	_leave(" [destroyed]");
504 }
505 
506 /*
507  * Queue the cell manager.
508  */
509 static void afs_queue_cell_manager(struct afs_net *net)
510 {
511 	int outstanding = atomic_inc_return(&net->cells_outstanding);
512 
513 	_enter("%d", outstanding);
514 
515 	if (!queue_work(afs_wq, &net->cells_manager))
516 		afs_dec_cells_outstanding(net);
517 }
518 
519 /*
520  * Cell management timer.  We have an increment on cells_outstanding that we
521  * need to pass along to the work item.
522  */
523 void afs_cells_timer(struct timer_list *timer)
524 {
525 	struct afs_net *net = container_of(timer, struct afs_net, cells_timer);
526 
527 	_enter("");
528 	if (!queue_work(afs_wq, &net->cells_manager))
529 		afs_dec_cells_outstanding(net);
530 }
531 
532 /*
533  * Get a reference on a cell record.
534  */
535 struct afs_cell *afs_get_cell(struct afs_cell *cell)
536 {
537 	atomic_inc(&cell->usage);
538 	return cell;
539 }
540 
541 /*
542  * Drop a reference on a cell record.
543  */
544 void afs_put_cell(struct afs_net *net, struct afs_cell *cell)
545 {
546 	time64_t now, expire_delay;
547 
548 	if (!cell)
549 		return;
550 
551 	_enter("%s", cell->name);
552 
553 	now = ktime_get_real_seconds();
554 	cell->last_inactive = now;
555 	expire_delay = 0;
556 	if (cell->vl_servers->nr_servers)
557 		expire_delay = afs_cell_gc_delay;
558 
559 	if (atomic_dec_return(&cell->usage) > 1)
560 		return;
561 
562 	/* 'cell' may now be garbage collected. */
563 	afs_set_cell_timer(net, expire_delay);
564 }
565 
566 /*
567  * Allocate a key to use as a placeholder for anonymous user security.
568  */
569 static int afs_alloc_anon_key(struct afs_cell *cell)
570 {
571 	struct key *key;
572 	char keyname[4 + AFS_MAXCELLNAME + 1], *cp, *dp;
573 
574 	/* Create a key to represent an anonymous user. */
575 	memcpy(keyname, "afs@", 4);
576 	dp = keyname + 4;
577 	cp = cell->name;
578 	do {
579 		*dp++ = tolower(*cp);
580 	} while (*cp++);
581 
582 	key = rxrpc_get_null_key(keyname);
583 	if (IS_ERR(key))
584 		return PTR_ERR(key);
585 
586 	cell->anonymous_key = key;
587 
588 	_debug("anon key %p{%x}",
589 	       cell->anonymous_key, key_serial(cell->anonymous_key));
590 	return 0;
591 }
592 
593 /*
594  * Activate a cell.
595  */
596 static int afs_activate_cell(struct afs_net *net, struct afs_cell *cell)
597 {
598 	struct hlist_node **p;
599 	struct afs_cell *pcell;
600 	int ret;
601 
602 	if (!cell->anonymous_key) {
603 		ret = afs_alloc_anon_key(cell);
604 		if (ret < 0)
605 			return ret;
606 	}
607 
608 #ifdef CONFIG_AFS_FSCACHE
609 	cell->cache = fscache_acquire_cookie(afs_cache_netfs.primary_index,
610 					     &afs_cell_cache_index_def,
611 					     cell->name, strlen(cell->name),
612 					     NULL, 0,
613 					     cell, 0, true);
614 #endif
615 	ret = afs_proc_cell_setup(cell);
616 	if (ret < 0)
617 		return ret;
618 
619 	mutex_lock(&net->proc_cells_lock);
620 	for (p = &net->proc_cells.first; *p; p = &(*p)->next) {
621 		pcell = hlist_entry(*p, struct afs_cell, proc_link);
622 		if (strcmp(cell->name, pcell->name) < 0)
623 			break;
624 	}
625 
626 	cell->proc_link.pprev = p;
627 	cell->proc_link.next = *p;
628 	rcu_assign_pointer(*p, &cell->proc_link.next);
629 	if (cell->proc_link.next)
630 		cell->proc_link.next->pprev = &cell->proc_link.next;
631 
632 	afs_dynroot_mkdir(net, cell);
633 	mutex_unlock(&net->proc_cells_lock);
634 	return 0;
635 }
636 
637 /*
638  * Deactivate a cell.
639  */
640 static void afs_deactivate_cell(struct afs_net *net, struct afs_cell *cell)
641 {
642 	_enter("%s", cell->name);
643 
644 	afs_proc_cell_remove(cell);
645 
646 	mutex_lock(&net->proc_cells_lock);
647 	hlist_del_rcu(&cell->proc_link);
648 	afs_dynroot_rmdir(net, cell);
649 	mutex_unlock(&net->proc_cells_lock);
650 
651 #ifdef CONFIG_AFS_FSCACHE
652 	fscache_relinquish_cookie(cell->cache, NULL, false);
653 	cell->cache = NULL;
654 #endif
655 
656 	_leave("");
657 }
658 
659 /*
660  * Manage a cell record, initialising and destroying it, maintaining its DNS
661  * records.
662  */
663 static void afs_manage_cell(struct work_struct *work)
664 {
665 	struct afs_cell *cell = container_of(work, struct afs_cell, manager);
666 	struct afs_net *net = cell->net;
667 	bool deleted;
668 	int ret, usage;
669 
670 	_enter("%s", cell->name);
671 
672 again:
673 	_debug("state %u", cell->state);
674 	switch (cell->state) {
675 	case AFS_CELL_INACTIVE:
676 	case AFS_CELL_FAILED:
677 		write_seqlock(&net->cells_lock);
678 		usage = 1;
679 		deleted = atomic_try_cmpxchg_relaxed(&cell->usage, &usage, 0);
680 		if (deleted)
681 			rb_erase(&cell->net_node, &net->cells);
682 		write_sequnlock(&net->cells_lock);
683 		if (deleted)
684 			goto final_destruction;
685 		if (cell->state == AFS_CELL_FAILED)
686 			goto done;
687 		smp_store_release(&cell->state, AFS_CELL_UNSET);
688 		wake_up_var(&cell->state);
689 		goto again;
690 
691 	case AFS_CELL_UNSET:
692 		smp_store_release(&cell->state, AFS_CELL_ACTIVATING);
693 		wake_up_var(&cell->state);
694 		goto again;
695 
696 	case AFS_CELL_ACTIVATING:
697 		ret = afs_activate_cell(net, cell);
698 		if (ret < 0)
699 			goto activation_failed;
700 
701 		smp_store_release(&cell->state, AFS_CELL_ACTIVE);
702 		wake_up_var(&cell->state);
703 		goto again;
704 
705 	case AFS_CELL_ACTIVE:
706 		if (atomic_read(&cell->usage) > 1) {
707 			if (test_and_clear_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags)) {
708 				ret = afs_update_cell(cell);
709 				if (ret < 0)
710 					cell->error = ret;
711 			}
712 			goto done;
713 		}
714 		smp_store_release(&cell->state, AFS_CELL_DEACTIVATING);
715 		wake_up_var(&cell->state);
716 		goto again;
717 
718 	case AFS_CELL_DEACTIVATING:
719 		if (atomic_read(&cell->usage) > 1)
720 			goto reverse_deactivation;
721 		afs_deactivate_cell(net, cell);
722 		smp_store_release(&cell->state, AFS_CELL_INACTIVE);
723 		wake_up_var(&cell->state);
724 		goto again;
725 
726 	default:
727 		break;
728 	}
729 	_debug("bad state %u", cell->state);
730 	BUG(); /* Unhandled state */
731 
732 activation_failed:
733 	cell->error = ret;
734 	afs_deactivate_cell(net, cell);
735 
736 	smp_store_release(&cell->state, AFS_CELL_FAILED); /* vs error */
737 	wake_up_var(&cell->state);
738 	goto again;
739 
740 reverse_deactivation:
741 	smp_store_release(&cell->state, AFS_CELL_ACTIVE);
742 	wake_up_var(&cell->state);
743 	_leave(" [deact->act]");
744 	return;
745 
746 done:
747 	_leave(" [done %u]", cell->state);
748 	return;
749 
750 final_destruction:
751 	call_rcu(&cell->rcu, afs_cell_destroy);
752 	afs_dec_cells_outstanding(net);
753 	_leave(" [destruct %d]", atomic_read(&net->cells_outstanding));
754 }
755 
756 /*
757  * Manage the records of cells known to a network namespace.  This includes
758  * updating the DNS records and garbage collecting unused cells that were
759  * automatically added.
760  *
761  * Note that constructed cell records may only be removed from net->cells by
762  * this work item, so it is safe for this work item to stash a cursor pointing
763  * into the tree and then return to caller (provided it skips cells that are
764  * still under construction).
765  *
766  * Note also that we were given an increment on net->cells_outstanding by
767  * whoever queued us that we need to deal with before returning.
768  */
769 void afs_manage_cells(struct work_struct *work)
770 {
771 	struct afs_net *net = container_of(work, struct afs_net, cells_manager);
772 	struct rb_node *cursor;
773 	time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
774 	bool purging = !net->live;
775 
776 	_enter("");
777 
778 	/* Trawl the cell database looking for cells that have expired from
779 	 * lack of use and cells whose DNS results have expired and dispatch
780 	 * their managers.
781 	 */
782 	read_seqlock_excl(&net->cells_lock);
783 
784 	for (cursor = rb_first(&net->cells); cursor; cursor = rb_next(cursor)) {
785 		struct afs_cell *cell =
786 			rb_entry(cursor, struct afs_cell, net_node);
787 		unsigned usage;
788 		bool sched_cell = false;
789 
790 		usage = atomic_read(&cell->usage);
791 		_debug("manage %s %u", cell->name, usage);
792 
793 		ASSERTCMP(usage, >=, 1);
794 
795 		if (purging) {
796 			if (test_and_clear_bit(AFS_CELL_FL_NO_GC, &cell->flags))
797 				usage = atomic_dec_return(&cell->usage);
798 			ASSERTCMP(usage, ==, 1);
799 		}
800 
801 		if (usage == 1) {
802 			struct afs_vlserver_list *vllist;
803 			time64_t expire_at = cell->last_inactive;
804 
805 			read_lock(&cell->vl_servers_lock);
806 			vllist = rcu_dereference_protected(
807 				cell->vl_servers,
808 				lockdep_is_held(&cell->vl_servers_lock));
809 			if (vllist->nr_servers > 0)
810 				expire_at += afs_cell_gc_delay;
811 			read_unlock(&cell->vl_servers_lock);
812 			if (purging || expire_at <= now)
813 				sched_cell = true;
814 			else if (expire_at < next_manage)
815 				next_manage = expire_at;
816 		}
817 
818 		if (!purging) {
819 			if (test_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags))
820 				sched_cell = true;
821 		}
822 
823 		if (sched_cell)
824 			queue_work(afs_wq, &cell->manager);
825 	}
826 
827 	read_sequnlock_excl(&net->cells_lock);
828 
829 	/* Update the timer on the way out.  We have to pass an increment on
830 	 * cells_outstanding in the namespace that we are in to the timer or
831 	 * the work scheduler.
832 	 */
833 	if (!purging && next_manage < TIME64_MAX) {
834 		now = ktime_get_real_seconds();
835 
836 		if (next_manage - now <= 0) {
837 			if (queue_work(afs_wq, &net->cells_manager))
838 				atomic_inc(&net->cells_outstanding);
839 		} else {
840 			afs_set_cell_timer(net, next_manage - now);
841 		}
842 	}
843 
844 	afs_dec_cells_outstanding(net);
845 	_leave(" [%d]", atomic_read(&net->cells_outstanding));
846 }
847 
848 /*
849  * Purge in-memory cell database.
850  */
851 void afs_cell_purge(struct afs_net *net)
852 {
853 	struct afs_cell *ws;
854 
855 	_enter("");
856 
857 	write_seqlock(&net->cells_lock);
858 	ws = rcu_access_pointer(net->ws_cell);
859 	RCU_INIT_POINTER(net->ws_cell, NULL);
860 	write_sequnlock(&net->cells_lock);
861 	afs_put_cell(net, ws);
862 
863 	_debug("del timer");
864 	if (del_timer_sync(&net->cells_timer))
865 		atomic_dec(&net->cells_outstanding);
866 
867 	_debug("kick mgr");
868 	afs_queue_cell_manager(net);
869 
870 	_debug("wait");
871 	wait_var_event(&net->cells_outstanding,
872 		       !atomic_read(&net->cells_outstanding));
873 	_leave("");
874 }
875