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	init_rwsem(&cell->vs_lock);
165	cell->volumes = RB_ROOT;
166	INIT_HLIST_HEAD(&cell->proc_volumes);
167	seqlock_init(&cell->volume_lock);
168	cell->fs_servers = RB_ROOT;
169	seqlock_init(&cell->fs_lock);
170	rwlock_init(&cell->vl_servers_lock);
171	cell->flags = (1 << AFS_CELL_FL_CHECK_ALIAS);
172
173	/* Provide a VL server list, filling it in if we were given a list of
174	 * addresses to use.
175	 */
176	if (addresses) {
177		vllist = afs_parse_text_addrs(net,
178					      addresses, strlen(addresses), ':',
179					      VL_SERVICE, AFS_VL_PORT);
180		if (IS_ERR(vllist)) {
181			ret = PTR_ERR(vllist);
182			goto parse_failed;
183		}
184
185		vllist->source = DNS_RECORD_FROM_CONFIG;
186		vllist->status = DNS_LOOKUP_NOT_DONE;
187		cell->dns_expiry = TIME64_MAX;
188	} else {
189		ret = -ENOMEM;
190		vllist = afs_alloc_vlserver_list(0);
191		if (!vllist)
192			goto error;
193		vllist->source = DNS_RECORD_UNAVAILABLE;
194		vllist->status = DNS_LOOKUP_NOT_DONE;
195		cell->dns_expiry = ktime_get_real_seconds();
196	}
197
198	rcu_assign_pointer(cell->vl_servers, vllist);
199
200	cell->dns_source = vllist->source;
201	cell->dns_status = vllist->status;
202	smp_store_release(&cell->dns_lookup_count, 1); /* vs source/status */
203	atomic_inc(&net->cells_outstanding);
204	cell->debug_id = atomic_inc_return(&cell_debug_id);
205	trace_afs_cell(cell->debug_id, 1, 0, afs_cell_trace_alloc);
206
207	_leave(" = %p", cell);
208	return cell;
209
210parse_failed:
211	if (ret == -EINVAL)
212		printk(KERN_ERR "kAFS: bad VL server IP address\n");
213error:
214	kfree(cell->name);
215	kfree(cell);
216	_leave(" = %d", ret);
217	return ERR_PTR(ret);
218}
219
220/*
221 * afs_lookup_cell - Look up or create a cell record.
222 * @net:	The network namespace
223 * @name:	The name of the cell.
224 * @namesz:	The strlen of the cell name.
225 * @vllist:	A colon/comma separated list of numeric IP addresses or NULL.
226 * @excl:	T if an error should be given if the cell name already exists.
227 *
228 * Look up a cell record by name and query the DNS for VL server addresses if
229 * needed.  Note that that actual DNS query is punted off to the manager thread
230 * so that this function can return immediately if interrupted whilst allowing
231 * cell records to be shared even if not yet fully constructed.
232 */
233struct afs_cell *afs_lookup_cell(struct afs_net *net,
234				 const char *name, unsigned int namesz,
235				 const char *vllist, bool excl)
236{
237	struct afs_cell *cell, *candidate, *cursor;
238	struct rb_node *parent, **pp;
239	enum afs_cell_state state;
240	int ret, n;
241
242	_enter("%s,%s", name, vllist);
243
244	if (!excl) {
245		cell = afs_find_cell(net, name, namesz, afs_cell_trace_use_lookup);
246		if (!IS_ERR(cell))
247			goto wait_for_cell;
248	}
249
250	/* Assume we're probably going to create a cell and preallocate and
251	 * mostly set up a candidate record.  We can then use this to stash the
252	 * name, the net namespace and VL server addresses.
253	 *
254	 * We also want to do this before we hold any locks as it may involve
255	 * upcalling to userspace to make DNS queries.
256	 */
257	candidate = afs_alloc_cell(net, name, namesz, vllist);
258	if (IS_ERR(candidate)) {
259		_leave(" = %ld", PTR_ERR(candidate));
260		return candidate;
261	}
262
263	/* Find the insertion point and check to see if someone else added a
264	 * cell whilst we were allocating.
265	 */
266	down_write(&net->cells_lock);
267
268	pp = &net->cells.rb_node;
269	parent = NULL;
270	while (*pp) {
271		parent = *pp;
272		cursor = rb_entry(parent, struct afs_cell, net_node);
273
274		n = strncasecmp(cursor->name, name,
275				min_t(size_t, cursor->name_len, namesz));
276		if (n == 0)
277			n = cursor->name_len - namesz;
278		if (n < 0)
279			pp = &(*pp)->rb_left;
280		else if (n > 0)
281			pp = &(*pp)->rb_right;
282		else
283			goto cell_already_exists;
284	}
285
286	cell = candidate;
287	candidate = NULL;
288	atomic_set(&cell->active, 2);
289	trace_afs_cell(cell->debug_id, refcount_read(&cell->ref), 2, afs_cell_trace_insert);
290	rb_link_node_rcu(&cell->net_node, parent, pp);
291	rb_insert_color(&cell->net_node, &net->cells);
292	up_write(&net->cells_lock);
293
294	afs_queue_cell(cell, afs_cell_trace_get_queue_new);
295
296wait_for_cell:
297	trace_afs_cell(cell->debug_id, refcount_read(&cell->ref), atomic_read(&cell->active),
298		       afs_cell_trace_wait);
299	_debug("wait_for_cell");
300	wait_var_event(&cell->state,
301		       ({
302			       state = smp_load_acquire(&cell->state); /* vs error */
303			       state == AFS_CELL_ACTIVE || state == AFS_CELL_REMOVED;
304		       }));
305
306	/* Check the state obtained from the wait check. */
307	if (state == AFS_CELL_REMOVED) {
308		ret = cell->error;
309		goto error;
310	}
311
312	_leave(" = %p [cell]", cell);
313	return cell;
314
315cell_already_exists:
316	_debug("cell exists");
317	cell = cursor;
318	if (excl) {
319		ret = -EEXIST;
320	} else {
321		afs_use_cell(cursor, afs_cell_trace_use_lookup);
322		ret = 0;
323	}
324	up_write(&net->cells_lock);
325	if (candidate)
326		afs_put_cell(candidate, afs_cell_trace_put_candidate);
327	if (ret == 0)
328		goto wait_for_cell;
329	goto error_noput;
330error:
331	afs_unuse_cell(net, cell, afs_cell_trace_unuse_lookup);
332error_noput:
333	_leave(" = %d [error]", ret);
334	return ERR_PTR(ret);
335}
336
337/*
338 * set the root cell information
339 * - can be called with a module parameter string
340 * - can be called from a write to /proc/fs/afs/rootcell
341 */
342int afs_cell_init(struct afs_net *net, const char *rootcell)
343{
344	struct afs_cell *old_root, *new_root;
345	const char *cp, *vllist;
346	size_t len;
347
348	_enter("");
349
350	if (!rootcell) {
351		/* module is loaded with no parameters, or built statically.
352		 * - in the future we might initialize cell DB here.
353		 */
354		_leave(" = 0 [no root]");
355		return 0;
356	}
357
358	cp = strchr(rootcell, ':');
359	if (!cp) {
360		_debug("kAFS: no VL server IP addresses specified");
361		vllist = NULL;
362		len = strlen(rootcell);
363	} else {
364		vllist = cp + 1;
365		len = cp - rootcell;
366	}
367
368	/* allocate a cell record for the root cell */
369	new_root = afs_lookup_cell(net, rootcell, len, vllist, false);
370	if (IS_ERR(new_root)) {
371		_leave(" = %ld", PTR_ERR(new_root));
372		return PTR_ERR(new_root);
373	}
374
375	if (!test_and_set_bit(AFS_CELL_FL_NO_GC, &new_root->flags))
376		afs_use_cell(new_root, afs_cell_trace_use_pin);
377
378	/* install the new cell */
379	down_write(&net->cells_lock);
380	afs_see_cell(new_root, afs_cell_trace_see_ws);
381	old_root = net->ws_cell;
382	net->ws_cell = new_root;
383	up_write(&net->cells_lock);
384
385	afs_unuse_cell(net, old_root, afs_cell_trace_unuse_ws);
386	_leave(" = 0");
387	return 0;
388}
389
390/*
391 * Update a cell's VL server address list from the DNS.
392 */
393static 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		vllist = afs_alloc_vlserver_list(0);
412		if (!vllist) {
413			if (ret >= 0)
414				ret = -ENOMEM;
415			goto out_wake;
416		}
417
418		switch (ret) {
419		case -ENODATA:
420		case -EDESTADDRREQ:
421			vllist->status = DNS_LOOKUP_GOT_NOT_FOUND;
422			break;
423		case -EAGAIN:
424		case -ECONNREFUSED:
425			vllist->status = DNS_LOOKUP_GOT_TEMP_FAILURE;
426			break;
427		default:
428			vllist->status = DNS_LOOKUP_GOT_LOCAL_FAILURE;
429			break;
430		}
431	}
432
433	_debug("%s: got list %d %d", cell->name, vllist->source, vllist->status);
434	cell->dns_status = vllist->status;
435
436	now = ktime_get_real_seconds();
437	if (min_ttl > max_ttl)
438		max_ttl = min_ttl;
439	if (expiry < now + min_ttl)
440		expiry = now + min_ttl;
441	else if (expiry > now + max_ttl)
442		expiry = now + max_ttl;
443
444	_debug("%s: status %d", cell->name, vllist->status);
445	if (vllist->source == DNS_RECORD_UNAVAILABLE) {
446		switch (vllist->status) {
447		case DNS_LOOKUP_GOT_NOT_FOUND:
448			/* The DNS said that the cell does not exist or there
449			 * weren't any addresses to be had.
450			 */
451			cell->dns_expiry = expiry;
452			break;
453
454		case DNS_LOOKUP_BAD:
455		case DNS_LOOKUP_GOT_LOCAL_FAILURE:
456		case DNS_LOOKUP_GOT_TEMP_FAILURE:
457		case DNS_LOOKUP_GOT_NS_FAILURE:
458		default:
459			cell->dns_expiry = now + 10;
460			break;
461		}
462	} else {
463		cell->dns_expiry = expiry;
464	}
465
466	/* Replace the VL server list if the new record has servers or the old
467	 * record doesn't.
468	 */
469	write_lock(&cell->vl_servers_lock);
470	p = rcu_dereference_protected(cell->vl_servers, true);
471	if (vllist->nr_servers > 0 || p->nr_servers == 0) {
472		rcu_assign_pointer(cell->vl_servers, vllist);
473		cell->dns_source = vllist->source;
474		old = p;
475	}
476	write_unlock(&cell->vl_servers_lock);
477	afs_put_vlserverlist(cell->net, old);
478
479out_wake:
480	smp_store_release(&cell->dns_lookup_count,
481			  cell->dns_lookup_count + 1); /* vs source/status */
482	wake_up_var(&cell->dns_lookup_count);
483	_leave(" = %d", ret);
484	return ret;
485}
486
487/*
488 * Destroy a cell record
489 */
490static void afs_cell_destroy(struct rcu_head *rcu)
491{
492	struct afs_cell *cell = container_of(rcu, struct afs_cell, rcu);
493	struct afs_net *net = cell->net;
494	int r;
495
496	_enter("%p{%s}", cell, cell->name);
497
498	r = refcount_read(&cell->ref);
499	ASSERTCMP(r, ==, 0);
500	trace_afs_cell(cell->debug_id, r, atomic_read(&cell->active), afs_cell_trace_free);
501
502	afs_put_vlserverlist(net, rcu_access_pointer(cell->vl_servers));
503	afs_unuse_cell(net, cell->alias_of, afs_cell_trace_unuse_alias);
504	key_put(cell->anonymous_key);
505	kfree(cell->name);
506	kfree(cell);
507
508	afs_dec_cells_outstanding(net);
509	_leave(" [destroyed]");
510}
511
512/*
513 * Queue the cell manager.
514 */
515static void afs_queue_cell_manager(struct afs_net *net)
516{
517	int outstanding = atomic_inc_return(&net->cells_outstanding);
518
519	_enter("%d", outstanding);
520
521	if (!queue_work(afs_wq, &net->cells_manager))
522		afs_dec_cells_outstanding(net);
523}
524
525/*
526 * Cell management timer.  We have an increment on cells_outstanding that we
527 * need to pass along to the work item.
528 */
529void afs_cells_timer(struct timer_list *timer)
530{
531	struct afs_net *net = container_of(timer, struct afs_net, cells_timer);
532
533	_enter("");
534	if (!queue_work(afs_wq, &net->cells_manager))
535		afs_dec_cells_outstanding(net);
536}
537
538/*
539 * Get a reference on a cell record.
540 */
541struct afs_cell *afs_get_cell(struct afs_cell *cell, enum afs_cell_trace reason)
542{
543	int r;
544
545	__refcount_inc(&cell->ref, &r);
546	trace_afs_cell(cell->debug_id, r + 1, 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 a;
558		bool zero;
559		int r;
560
561		a = atomic_read(&cell->active);
562		zero = __refcount_dec_and_test(&cell->ref, &r);
563		trace_afs_cell(debug_id, r - 1, a, reason);
564		if (zero) {
565			a = atomic_read(&cell->active);
566			WARN(a != 0, "Cell active count %u > 0\n", a);
567			call_rcu(&cell->rcu, afs_cell_destroy);
568		}
569	}
570}
571
572/*
573 * Note a cell becoming more active.
574 */
575struct afs_cell *afs_use_cell(struct afs_cell *cell, enum afs_cell_trace reason)
576{
577	int r, a;
578
579	r = refcount_read(&cell->ref);
580	WARN_ON(r == 0);
581	a = atomic_inc_return(&cell->active);
582	trace_afs_cell(cell->debug_id, r, 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 r, 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	r = refcount_read(&cell->ref);
609	a = atomic_dec_return(&cell->active);
610	trace_afs_cell(debug_id, r, 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 r, a;
623
624	r = refcount_read(&cell->ref);
625	a = atomic_read(&cell->active);
626	trace_afs_cell(cell->debug_id, r, 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	ret = afs_proc_cell_setup(cell);
682	if (ret < 0)
683		return ret;
684
685	mutex_lock(&net->proc_cells_lock);
686	for (p = &net->proc_cells.first; *p; p = &(*p)->next) {
687		pcell = hlist_entry(*p, struct afs_cell, proc_link);
688		if (strcmp(cell->name, pcell->name) < 0)
689			break;
690	}
691
692	cell->proc_link.pprev = p;
693	cell->proc_link.next = *p;
694	rcu_assign_pointer(*p, &cell->proc_link.next);
695	if (cell->proc_link.next)
696		cell->proc_link.next->pprev = &cell->proc_link.next;
697
698	afs_dynroot_mkdir(net, cell);
699	mutex_unlock(&net->proc_cells_lock);
700	return 0;
701}
702
703/*
704 * Deactivate a cell.
705 */
706static void afs_deactivate_cell(struct afs_net *net, struct afs_cell *cell)
707{
708	_enter("%s", cell->name);
709
710	afs_proc_cell_remove(cell);
711
712	mutex_lock(&net->proc_cells_lock);
713	hlist_del_rcu(&cell->proc_link);
714	afs_dynroot_rmdir(net, cell);
715	mutex_unlock(&net->proc_cells_lock);
716
717	_leave("");
718}
719
720/*
721 * Manage a cell record, initialising and destroying it, maintaining its DNS
722 * records.
723 */
724static void afs_manage_cell(struct afs_cell *cell)
725{
726	struct afs_net *net = cell->net;
727	int ret, active;
728
729	_enter("%s", cell->name);
730
731again:
732	_debug("state %u", cell->state);
733	switch (cell->state) {
734	case AFS_CELL_INACTIVE:
735	case AFS_CELL_FAILED:
736		down_write(&net->cells_lock);
737		active = 1;
738		if (atomic_try_cmpxchg_relaxed(&cell->active, &active, 0)) {
739			rb_erase(&cell->net_node, &net->cells);
740			trace_afs_cell(cell->debug_id, refcount_read(&cell->ref), 0,
741				       afs_cell_trace_unuse_delete);
742			smp_store_release(&cell->state, AFS_CELL_REMOVED);
743		}
744		up_write(&net->cells_lock);
745		if (cell->state == AFS_CELL_REMOVED) {
746			wake_up_var(&cell->state);
747			goto final_destruction;
748		}
749		if (cell->state == AFS_CELL_FAILED)
750			goto done;
751		smp_store_release(&cell->state, AFS_CELL_UNSET);
752		wake_up_var(&cell->state);
753		goto again;
754
755	case AFS_CELL_UNSET:
756		smp_store_release(&cell->state, AFS_CELL_ACTIVATING);
757		wake_up_var(&cell->state);
758		goto again;
759
760	case AFS_CELL_ACTIVATING:
761		ret = afs_activate_cell(net, cell);
762		if (ret < 0)
763			goto activation_failed;
764
765		smp_store_release(&cell->state, AFS_CELL_ACTIVE);
766		wake_up_var(&cell->state);
767		goto again;
768
769	case AFS_CELL_ACTIVE:
770		if (atomic_read(&cell->active) > 1) {
771			if (test_and_clear_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags)) {
772				ret = afs_update_cell(cell);
773				if (ret < 0)
774					cell->error = ret;
775			}
776			goto done;
777		}
778		smp_store_release(&cell->state, AFS_CELL_DEACTIVATING);
779		wake_up_var(&cell->state);
780		goto again;
781
782	case AFS_CELL_DEACTIVATING:
783		if (atomic_read(&cell->active) > 1)
784			goto reverse_deactivation;
785		afs_deactivate_cell(net, cell);
786		smp_store_release(&cell->state, AFS_CELL_INACTIVE);
787		wake_up_var(&cell->state);
788		goto again;
789
790	case AFS_CELL_REMOVED:
791		goto done;
792
793	default:
794		break;
795	}
796	_debug("bad state %u", cell->state);
797	BUG(); /* Unhandled state */
798
799activation_failed:
800	cell->error = ret;
801	afs_deactivate_cell(net, cell);
802
803	smp_store_release(&cell->state, AFS_CELL_FAILED); /* vs error */
804	wake_up_var(&cell->state);
805	goto again;
806
807reverse_deactivation:
808	smp_store_release(&cell->state, AFS_CELL_ACTIVE);
809	wake_up_var(&cell->state);
810	_leave(" [deact->act]");
811	return;
812
813done:
814	_leave(" [done %u]", cell->state);
815	return;
816
817final_destruction:
818	/* The root volume is pinning the cell */
819	afs_put_volume(cell->root_volume, afs_volume_trace_put_cell_root);
820	cell->root_volume = NULL;
821	afs_put_cell(cell, afs_cell_trace_put_destroy);
822}
823
824static void afs_manage_cell_work(struct work_struct *work)
825{
826	struct afs_cell *cell = container_of(work, struct afs_cell, manager);
827
828	afs_manage_cell(cell);
829	afs_put_cell(cell, afs_cell_trace_put_queue_work);
830}
831
832/*
833 * Manage the records of cells known to a network namespace.  This includes
834 * updating the DNS records and garbage collecting unused cells that were
835 * automatically added.
836 *
837 * Note that constructed cell records may only be removed from net->cells by
838 * this work item, so it is safe for this work item to stash a cursor pointing
839 * into the tree and then return to caller (provided it skips cells that are
840 * still under construction).
841 *
842 * Note also that we were given an increment on net->cells_outstanding by
843 * whoever queued us that we need to deal with before returning.
844 */
845void afs_manage_cells(struct work_struct *work)
846{
847	struct afs_net *net = container_of(work, struct afs_net, cells_manager);
848	struct rb_node *cursor;
849	time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
850	bool purging = !net->live;
851
852	_enter("");
853
854	/* Trawl the cell database looking for cells that have expired from
855	 * lack of use and cells whose DNS results have expired and dispatch
856	 * their managers.
857	 */
858	down_read(&net->cells_lock);
859
860	for (cursor = rb_first(&net->cells); cursor; cursor = rb_next(cursor)) {
861		struct afs_cell *cell =
862			rb_entry(cursor, struct afs_cell, net_node);
863		unsigned active;
864		bool sched_cell = false;
865
866		active = atomic_read(&cell->active);
867		trace_afs_cell(cell->debug_id, refcount_read(&cell->ref),
868			       active, afs_cell_trace_manage);
869
870		ASSERTCMP(active, >=, 1);
871
872		if (purging) {
873			if (test_and_clear_bit(AFS_CELL_FL_NO_GC, &cell->flags)) {
874				active = atomic_dec_return(&cell->active);
875				trace_afs_cell(cell->debug_id, refcount_read(&cell->ref),
876					       active, afs_cell_trace_unuse_pin);
877			}
878		}
879
880		if (active == 1) {
881			struct afs_vlserver_list *vllist;
882			time64_t expire_at = cell->last_inactive;
883
884			read_lock(&cell->vl_servers_lock);
885			vllist = rcu_dereference_protected(
886				cell->vl_servers,
887				lockdep_is_held(&cell->vl_servers_lock));
888			if (vllist->nr_servers > 0)
889				expire_at += afs_cell_gc_delay;
890			read_unlock(&cell->vl_servers_lock);
891			if (purging || expire_at <= now)
892				sched_cell = true;
893			else if (expire_at < next_manage)
894				next_manage = expire_at;
895		}
896
897		if (!purging) {
898			if (test_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags))
899				sched_cell = true;
900		}
901
902		if (sched_cell)
903			afs_queue_cell(cell, afs_cell_trace_get_queue_manage);
904	}
905
906	up_read(&net->cells_lock);
907
908	/* Update the timer on the way out.  We have to pass an increment on
909	 * cells_outstanding in the namespace that we are in to the timer or
910	 * the work scheduler.
911	 */
912	if (!purging && next_manage < TIME64_MAX) {
913		now = ktime_get_real_seconds();
914
915		if (next_manage - now <= 0) {
916			if (queue_work(afs_wq, &net->cells_manager))
917				atomic_inc(&net->cells_outstanding);
918		} else {
919			afs_set_cell_timer(net, next_manage - now);
920		}
921	}
922
923	afs_dec_cells_outstanding(net);
924	_leave(" [%d]", atomic_read(&net->cells_outstanding));
925}
926
927/*
928 * Purge in-memory cell database.
929 */
930void afs_cell_purge(struct afs_net *net)
931{
932	struct afs_cell *ws;
933
934	_enter("");
935
936	down_write(&net->cells_lock);
937	ws = net->ws_cell;
938	net->ws_cell = NULL;
939	up_write(&net->cells_lock);
940	afs_unuse_cell(net, ws, afs_cell_trace_unuse_ws);
941
942	_debug("del timer");
943	if (del_timer_sync(&net->cells_timer))
944		atomic_dec(&net->cells_outstanding);
945
946	_debug("kick mgr");
947	afs_queue_cell_manager(net);
948
949	_debug("wait");
950	wait_var_event(&net->cells_outstanding,
951		       !atomic_read(&net->cells_outstanding));
952	_leave("");
953}
954