1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AFS cell and server record management
4 * Copyright (C) 2002, 2017 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #include <linux/slab.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>
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;
22 static void afs_manage_cell(struct work_struct *);
24 static void afs_dec_cells_outstanding(struct afs_net *net)
26 if (atomic_dec_and_test(&net->cells_outstanding))
27 wake_up_var(&net->cells_outstanding);
31 * Set the cell timer to fire after a given delay, assuming it's not already
32 * set for an earlier time.
34 static void afs_set_cell_timer(struct afs_net *net, time64_t delay)
37 atomic_inc(&net->cells_outstanding);
38 if (timer_reduce(&net->cells_timer, jiffies + delay * HZ))
39 afs_dec_cells_outstanding(net);
44 * Look up and get an activation reference on a cell record under RCU
45 * conditions. The caller must hold the RCU read lock.
47 struct afs_cell *afs_lookup_cell_rcu(struct afs_net *net,
48 const char *name, unsigned int namesz)
50 struct afs_cell *cell = NULL;
52 int n, seq = 0, ret = 0;
54 _enter("%*.*s", namesz, namesz, name);
56 if (name && namesz == 0)
57 return ERR_PTR(-EINVAL);
58 if (namesz > AFS_MAXCELLNAME)
59 return ERR_PTR(-ENAMETOOLONG);
62 /* Unfortunately, rbtree walking doesn't give reliable results
63 * under just the RCU read lock, so we have to check for
67 afs_put_cell(net, cell);
71 read_seqbegin_or_lock(&net->cells_lock, &seq);
74 cell = rcu_dereference_raw(net->ws_cell);
84 p = rcu_dereference_raw(net->cells.rb_node);
86 cell = rb_entry(p, struct afs_cell, net_node);
88 n = strncasecmp(cell->name, name,
89 min_t(size_t, cell->name_len, namesz));
91 n = cell->name_len - namesz;
93 p = rcu_dereference_raw(p->rb_left);
95 p = rcu_dereference_raw(p->rb_right);
97 if (atomic_inc_not_zero(&cell->usage)) {
101 /* We want to repeat the search, this time with
102 * the lock properly locked.
108 } while (need_seqretry(&net->cells_lock, seq));
110 done_seqretry(&net->cells_lock, seq);
112 if (ret != 0 && cell)
113 afs_put_cell(net, cell);
115 return ret == 0 ? cell : ERR_PTR(ret);
119 * Set up a cell record and fill in its name, VL server address list and
120 * allocate an anonymous key
122 static struct afs_cell *afs_alloc_cell(struct afs_net *net,
123 const char *name, unsigned int namelen,
124 const char *addresses)
126 struct afs_vlserver_list *vllist;
127 struct afs_cell *cell;
132 return ERR_PTR(-EINVAL);
133 if (namelen > AFS_MAXCELLNAME) {
134 _leave(" = -ENAMETOOLONG");
135 return ERR_PTR(-ENAMETOOLONG);
138 /* Prohibit cell names that contain unprintable chars, '/' and '@' or
139 * that begin with a dot. This also precludes "@cell".
142 return ERR_PTR(-EINVAL);
143 for (i = 0; i < namelen; i++) {
145 if (!isprint(ch) || ch == '/' || ch == '@')
146 return ERR_PTR(-EINVAL);
149 _enter("%*.*s,%s", namelen, namelen, name, addresses);
151 cell = kzalloc(sizeof(struct afs_cell), GFP_KERNEL);
153 _leave(" = -ENOMEM");
154 return ERR_PTR(-ENOMEM);
158 cell->name_len = namelen;
159 for (i = 0; i < namelen; i++)
160 cell->name[i] = tolower(name[i]);
162 atomic_set(&cell->usage, 2);
163 INIT_WORK(&cell->manager, afs_manage_cell);
164 INIT_LIST_HEAD(&cell->proc_volumes);
165 rwlock_init(&cell->proc_lock);
166 rwlock_init(&cell->vl_servers_lock);
168 /* Provide a VL server list, filling it in if we were given a list of
172 vllist = afs_parse_text_addrs(net,
173 addresses, strlen(addresses), ':',
174 VL_SERVICE, AFS_VL_PORT);
175 if (IS_ERR(vllist)) {
176 ret = PTR_ERR(vllist);
180 vllist->source = DNS_RECORD_FROM_CONFIG;
181 vllist->status = DNS_LOOKUP_NOT_DONE;
182 cell->dns_expiry = TIME64_MAX;
185 vllist = afs_alloc_vlserver_list(0);
188 vllist->source = DNS_RECORD_UNAVAILABLE;
189 vllist->status = DNS_LOOKUP_NOT_DONE;
190 cell->dns_expiry = ktime_get_real_seconds();
193 rcu_assign_pointer(cell->vl_servers, vllist);
195 cell->dns_source = vllist->source;
196 cell->dns_status = vllist->status;
197 smp_store_release(&cell->dns_lookup_count, 1); /* vs source/status */
199 _leave(" = %p", cell);
204 printk(KERN_ERR "kAFS: bad VL server IP address\n");
207 _leave(" = %d", ret);
212 * afs_lookup_cell - Look up or create a cell record.
213 * @net: The network namespace
214 * @name: The name of the cell.
215 * @namesz: The strlen of the cell name.
216 * @vllist: A colon/comma separated list of numeric IP addresses or NULL.
217 * @excl: T if an error should be given if the cell name already exists.
219 * Look up a cell record by name and query the DNS for VL server addresses if
220 * needed. Note that that actual DNS query is punted off to the manager thread
221 * so that this function can return immediately if interrupted whilst allowing
222 * cell records to be shared even if not yet fully constructed.
224 struct afs_cell *afs_lookup_cell(struct afs_net *net,
225 const char *name, unsigned int namesz,
226 const char *vllist, bool excl)
228 struct afs_cell *cell, *candidate, *cursor;
229 struct rb_node *parent, **pp;
230 enum afs_cell_state state;
233 _enter("%s,%s", name, vllist);
237 cell = afs_lookup_cell_rcu(net, name, namesz);
243 /* Assume we're probably going to create a cell and preallocate and
244 * mostly set up a candidate record. We can then use this to stash the
245 * name, the net namespace and VL server addresses.
247 * We also want to do this before we hold any locks as it may involve
248 * upcalling to userspace to make DNS queries.
250 candidate = afs_alloc_cell(net, name, namesz, vllist);
251 if (IS_ERR(candidate)) {
252 _leave(" = %ld", PTR_ERR(candidate));
256 /* Find the insertion point and check to see if someone else added a
257 * cell whilst we were allocating.
259 write_seqlock(&net->cells_lock);
261 pp = &net->cells.rb_node;
265 cursor = rb_entry(parent, struct afs_cell, net_node);
267 n = strncasecmp(cursor->name, name,
268 min_t(size_t, cursor->name_len, namesz));
270 n = cursor->name_len - namesz;
272 pp = &(*pp)->rb_left;
274 pp = &(*pp)->rb_right;
276 goto cell_already_exists;
281 rb_link_node_rcu(&cell->net_node, parent, pp);
282 rb_insert_color(&cell->net_node, &net->cells);
283 atomic_inc(&net->cells_outstanding);
284 write_sequnlock(&net->cells_lock);
286 queue_work(afs_wq, &cell->manager);
289 _debug("wait_for_cell");
290 wait_var_event(&cell->state,
292 state = smp_load_acquire(&cell->state); /* vs error */
293 state == AFS_CELL_ACTIVE || state == AFS_CELL_FAILED;
296 /* Check the state obtained from the wait check. */
297 if (state == AFS_CELL_FAILED) {
302 _leave(" = %p [cell]", cell);
306 _debug("cell exists");
311 afs_get_cell(cursor);
314 write_sequnlock(&net->cells_lock);
320 afs_put_cell(net, cell);
322 _leave(" = %d [error]", ret);
327 * set the root cell information
328 * - can be called with a module parameter string
329 * - can be called from a write to /proc/fs/afs/rootcell
331 int afs_cell_init(struct afs_net *net, const char *rootcell)
333 struct afs_cell *old_root, *new_root;
334 const char *cp, *vllist;
340 /* module is loaded with no parameters, or built statically.
341 * - in the future we might initialize cell DB here.
343 _leave(" = 0 [no root]");
347 cp = strchr(rootcell, ':');
349 _debug("kAFS: no VL server IP addresses specified");
351 len = strlen(rootcell);
357 /* allocate a cell record for the root cell */
358 new_root = afs_lookup_cell(net, rootcell, len, vllist, false);
359 if (IS_ERR(new_root)) {
360 _leave(" = %ld", PTR_ERR(new_root));
361 return PTR_ERR(new_root);
364 if (!test_and_set_bit(AFS_CELL_FL_NO_GC, &new_root->flags))
365 afs_get_cell(new_root);
367 /* install the new cell */
368 write_seqlock(&net->cells_lock);
369 old_root = rcu_access_pointer(net->ws_cell);
370 rcu_assign_pointer(net->ws_cell, new_root);
371 write_sequnlock(&net->cells_lock);
373 afs_put_cell(net, old_root);
379 * Update a cell's VL server address list from the DNS.
381 static int afs_update_cell(struct afs_cell *cell)
383 struct afs_vlserver_list *vllist, *old = NULL, *p;
384 unsigned int min_ttl = READ_ONCE(afs_cell_min_ttl);
385 unsigned int max_ttl = READ_ONCE(afs_cell_max_ttl);
386 time64_t now, expiry = 0;
389 _enter("%s", cell->name);
391 vllist = afs_dns_query(cell, &expiry);
392 if (IS_ERR(vllist)) {
393 ret = PTR_ERR(vllist);
395 _debug("%s: fail %d", cell->name, ret);
400 vllist = afs_alloc_vlserver_list(0);
407 vllist->status = DNS_LOOKUP_GOT_NOT_FOUND;
411 vllist->status = DNS_LOOKUP_GOT_TEMP_FAILURE;
414 vllist->status = DNS_LOOKUP_GOT_LOCAL_FAILURE;
419 _debug("%s: got list %d %d", cell->name, vllist->source, vllist->status);
420 cell->dns_status = vllist->status;
422 now = ktime_get_real_seconds();
423 if (min_ttl > max_ttl)
425 if (expiry < now + min_ttl)
426 expiry = now + min_ttl;
427 else if (expiry > now + max_ttl)
428 expiry = now + max_ttl;
430 _debug("%s: status %d", cell->name, vllist->status);
431 if (vllist->source == DNS_RECORD_UNAVAILABLE) {
432 switch (vllist->status) {
433 case DNS_LOOKUP_GOT_NOT_FOUND:
434 /* The DNS said that the cell does not exist or there
435 * weren't any addresses to be had.
437 cell->dns_expiry = expiry;
441 case DNS_LOOKUP_GOT_LOCAL_FAILURE:
442 case DNS_LOOKUP_GOT_TEMP_FAILURE:
443 case DNS_LOOKUP_GOT_NS_FAILURE:
445 cell->dns_expiry = now + 10;
449 cell->dns_expiry = expiry;
452 /* Replace the VL server list if the new record has servers or the old
455 write_lock(&cell->vl_servers_lock);
456 p = rcu_dereference_protected(cell->vl_servers, true);
457 if (vllist->nr_servers > 0 || p->nr_servers == 0) {
458 rcu_assign_pointer(cell->vl_servers, vllist);
459 cell->dns_source = vllist->source;
462 write_unlock(&cell->vl_servers_lock);
463 afs_put_vlserverlist(cell->net, old);
466 smp_store_release(&cell->dns_lookup_count,
467 cell->dns_lookup_count + 1); /* vs source/status */
468 wake_up_var(&cell->dns_lookup_count);
469 _leave(" = %d", ret);
474 * Destroy a cell record
476 static void afs_cell_destroy(struct rcu_head *rcu)
478 struct afs_cell *cell = container_of(rcu, struct afs_cell, rcu);
480 _enter("%p{%s}", cell, cell->name);
482 ASSERTCMP(atomic_read(&cell->usage), ==, 0);
484 afs_put_vlserverlist(cell->net, rcu_access_pointer(cell->vl_servers));
485 key_put(cell->anonymous_key);
488 _leave(" [destroyed]");
492 * Queue the cell manager.
494 static void afs_queue_cell_manager(struct afs_net *net)
496 int outstanding = atomic_inc_return(&net->cells_outstanding);
498 _enter("%d", outstanding);
500 if (!queue_work(afs_wq, &net->cells_manager))
501 afs_dec_cells_outstanding(net);
505 * Cell management timer. We have an increment on cells_outstanding that we
506 * need to pass along to the work item.
508 void afs_cells_timer(struct timer_list *timer)
510 struct afs_net *net = container_of(timer, struct afs_net, cells_timer);
513 if (!queue_work(afs_wq, &net->cells_manager))
514 afs_dec_cells_outstanding(net);
518 * Get a reference on a cell record.
520 struct afs_cell *afs_get_cell(struct afs_cell *cell)
522 atomic_inc(&cell->usage);
527 * Drop a reference on a cell record.
529 void afs_put_cell(struct afs_net *net, struct afs_cell *cell)
531 time64_t now, expire_delay;
536 _enter("%s", cell->name);
538 now = ktime_get_real_seconds();
539 cell->last_inactive = now;
541 if (cell->vl_servers->nr_servers)
542 expire_delay = afs_cell_gc_delay;
544 if (atomic_dec_return(&cell->usage) > 1)
547 /* 'cell' may now be garbage collected. */
548 afs_set_cell_timer(net, expire_delay);
552 * Allocate a key to use as a placeholder for anonymous user security.
554 static int afs_alloc_anon_key(struct afs_cell *cell)
557 char keyname[4 + AFS_MAXCELLNAME + 1], *cp, *dp;
559 /* Create a key to represent an anonymous user. */
560 memcpy(keyname, "afs@", 4);
564 *dp++ = tolower(*cp);
567 key = rxrpc_get_null_key(keyname);
571 cell->anonymous_key = key;
573 _debug("anon key %p{%x}",
574 cell->anonymous_key, key_serial(cell->anonymous_key));
581 static int afs_activate_cell(struct afs_net *net, struct afs_cell *cell)
583 struct hlist_node **p;
584 struct afs_cell *pcell;
587 if (!cell->anonymous_key) {
588 ret = afs_alloc_anon_key(cell);
593 #ifdef CONFIG_AFS_FSCACHE
594 cell->cache = fscache_acquire_cookie(afs_cache_netfs.primary_index,
595 &afs_cell_cache_index_def,
596 cell->name, strlen(cell->name),
600 ret = afs_proc_cell_setup(cell);
604 mutex_lock(&net->proc_cells_lock);
605 for (p = &net->proc_cells.first; *p; p = &(*p)->next) {
606 pcell = hlist_entry(*p, struct afs_cell, proc_link);
607 if (strcmp(cell->name, pcell->name) < 0)
611 cell->proc_link.pprev = p;
612 cell->proc_link.next = *p;
613 rcu_assign_pointer(*p, &cell->proc_link.next);
614 if (cell->proc_link.next)
615 cell->proc_link.next->pprev = &cell->proc_link.next;
617 afs_dynroot_mkdir(net, cell);
618 mutex_unlock(&net->proc_cells_lock);
625 static void afs_deactivate_cell(struct afs_net *net, struct afs_cell *cell)
627 _enter("%s", cell->name);
629 afs_proc_cell_remove(cell);
631 mutex_lock(&net->proc_cells_lock);
632 hlist_del_rcu(&cell->proc_link);
633 afs_dynroot_rmdir(net, cell);
634 mutex_unlock(&net->proc_cells_lock);
636 #ifdef CONFIG_AFS_FSCACHE
637 fscache_relinquish_cookie(cell->cache, NULL, false);
645 * Manage a cell record, initialising and destroying it, maintaining its DNS
648 static void afs_manage_cell(struct work_struct *work)
650 struct afs_cell *cell = container_of(work, struct afs_cell, manager);
651 struct afs_net *net = cell->net;
655 _enter("%s", cell->name);
658 _debug("state %u", cell->state);
659 switch (cell->state) {
660 case AFS_CELL_INACTIVE:
661 case AFS_CELL_FAILED:
662 write_seqlock(&net->cells_lock);
664 deleted = atomic_try_cmpxchg_relaxed(&cell->usage, &usage, 0);
666 rb_erase(&cell->net_node, &net->cells);
667 write_sequnlock(&net->cells_lock);
669 goto final_destruction;
670 if (cell->state == AFS_CELL_FAILED)
672 smp_store_release(&cell->state, AFS_CELL_UNSET);
673 wake_up_var(&cell->state);
677 smp_store_release(&cell->state, AFS_CELL_ACTIVATING);
678 wake_up_var(&cell->state);
681 case AFS_CELL_ACTIVATING:
682 ret = afs_activate_cell(net, cell);
684 goto activation_failed;
686 smp_store_release(&cell->state, AFS_CELL_ACTIVE);
687 wake_up_var(&cell->state);
690 case AFS_CELL_ACTIVE:
691 if (atomic_read(&cell->usage) > 1) {
692 if (test_and_clear_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags)) {
693 ret = afs_update_cell(cell);
699 smp_store_release(&cell->state, AFS_CELL_DEACTIVATING);
700 wake_up_var(&cell->state);
703 case AFS_CELL_DEACTIVATING:
704 if (atomic_read(&cell->usage) > 1)
705 goto reverse_deactivation;
706 afs_deactivate_cell(net, cell);
707 smp_store_release(&cell->state, AFS_CELL_INACTIVE);
708 wake_up_var(&cell->state);
714 _debug("bad state %u", cell->state);
715 BUG(); /* Unhandled state */
719 afs_deactivate_cell(net, cell);
721 smp_store_release(&cell->state, AFS_CELL_FAILED); /* vs error */
722 wake_up_var(&cell->state);
725 reverse_deactivation:
726 smp_store_release(&cell->state, AFS_CELL_ACTIVE);
727 wake_up_var(&cell->state);
728 _leave(" [deact->act]");
732 _leave(" [done %u]", cell->state);
736 call_rcu(&cell->rcu, afs_cell_destroy);
737 afs_dec_cells_outstanding(net);
738 _leave(" [destruct %d]", atomic_read(&net->cells_outstanding));
742 * Manage the records of cells known to a network namespace. This includes
743 * updating the DNS records and garbage collecting unused cells that were
744 * automatically added.
746 * Note that constructed cell records may only be removed from net->cells by
747 * this work item, so it is safe for this work item to stash a cursor pointing
748 * into the tree and then return to caller (provided it skips cells that are
749 * still under construction).
751 * Note also that we were given an increment on net->cells_outstanding by
752 * whoever queued us that we need to deal with before returning.
754 void afs_manage_cells(struct work_struct *work)
756 struct afs_net *net = container_of(work, struct afs_net, cells_manager);
757 struct rb_node *cursor;
758 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX;
759 bool purging = !net->live;
763 /* Trawl the cell database looking for cells that have expired from
764 * lack of use and cells whose DNS results have expired and dispatch
767 read_seqlock_excl(&net->cells_lock);
769 for (cursor = rb_first(&net->cells); cursor; cursor = rb_next(cursor)) {
770 struct afs_cell *cell =
771 rb_entry(cursor, struct afs_cell, net_node);
773 bool sched_cell = false;
775 usage = atomic_read(&cell->usage);
776 _debug("manage %s %u", cell->name, usage);
778 ASSERTCMP(usage, >=, 1);
781 if (test_and_clear_bit(AFS_CELL_FL_NO_GC, &cell->flags))
782 usage = atomic_dec_return(&cell->usage);
783 ASSERTCMP(usage, ==, 1);
787 struct afs_vlserver_list *vllist;
788 time64_t expire_at = cell->last_inactive;
790 read_lock(&cell->vl_servers_lock);
791 vllist = rcu_dereference_protected(
793 lockdep_is_held(&cell->vl_servers_lock));
794 if (vllist->nr_servers > 0)
795 expire_at += afs_cell_gc_delay;
796 read_unlock(&cell->vl_servers_lock);
797 if (purging || expire_at <= now)
799 else if (expire_at < next_manage)
800 next_manage = expire_at;
804 if (test_bit(AFS_CELL_FL_DO_LOOKUP, &cell->flags))
809 queue_work(afs_wq, &cell->manager);
812 read_sequnlock_excl(&net->cells_lock);
814 /* Update the timer on the way out. We have to pass an increment on
815 * cells_outstanding in the namespace that we are in to the timer or
816 * the work scheduler.
818 if (!purging && next_manage < TIME64_MAX) {
819 now = ktime_get_real_seconds();
821 if (next_manage - now <= 0) {
822 if (queue_work(afs_wq, &net->cells_manager))
823 atomic_inc(&net->cells_outstanding);
825 afs_set_cell_timer(net, next_manage - now);
829 afs_dec_cells_outstanding(net);
830 _leave(" [%d]", atomic_read(&net->cells_outstanding));
834 * Purge in-memory cell database.
836 void afs_cell_purge(struct afs_net *net)
842 write_seqlock(&net->cells_lock);
843 ws = rcu_access_pointer(net->ws_cell);
844 RCU_INIT_POINTER(net->ws_cell, NULL);
845 write_sequnlock(&net->cells_lock);
846 afs_put_cell(net, ws);
849 if (del_timer_sync(&net->cells_timer))
850 atomic_dec(&net->cells_outstanding);
853 afs_queue_cell_manager(net);
856 wait_var_event(&net->cells_outstanding,
857 !atomic_read(&net->cells_outstanding));