1 /* CacheFiles path walking and related routines
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/file.h>
16 #include <linux/fsnotify.h>
17 #include <linux/quotaops.h>
18 #include <linux/xattr.h>
19 #include <linux/mount.h>
20 #include <linux/namei.h>
21 #include <linux/security.h>
22 #include <linux/slab.h>
25 #define CACHEFILES_KEYBUF_SIZE 512
28 * dump debugging info about an object
31 void __cachefiles_printk_object(struct cachefiles_object *object,
35 struct fscache_cookie *cookie;
36 unsigned keylen, loop;
38 pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
39 pr_err("%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
40 prefix, object->fscache.state->name,
41 object->fscache.flags, work_busy(&object->fscache.work),
42 object->fscache.events, object->fscache.event_mask);
43 pr_err("%sops=%u inp=%u exc=%u\n",
44 prefix, object->fscache.n_ops, object->fscache.n_in_progress,
45 object->fscache.n_exclusive);
46 pr_err("%sparent=%p\n",
47 prefix, object->fscache.parent);
49 spin_lock(&object->fscache.lock);
50 cookie = object->fscache.cookie;
52 pr_err("%scookie=%p [pr=%p nd=%p fl=%lx]\n",
54 object->fscache.cookie,
55 object->fscache.cookie->parent,
56 object->fscache.cookie->netfs_data,
57 object->fscache.cookie->flags);
58 if (keybuf && cookie->def)
59 keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
60 CACHEFILES_KEYBUF_SIZE);
64 pr_err("%scookie=NULL\n", prefix);
67 spin_unlock(&object->fscache.lock);
70 pr_err("%skey=[%u] '", prefix, keylen);
71 for (loop = 0; loop < keylen; loop++)
72 pr_cont("%02x", keybuf[loop]);
78 * dump debugging info about a pair of objects
80 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
81 struct cachefiles_object *xobject)
85 keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
87 __cachefiles_printk_object(object, "", keybuf);
89 __cachefiles_printk_object(xobject, "x", keybuf);
94 * mark the owner of a dentry, if there is one, to indicate that that dentry
95 * has been preemptively deleted
96 * - the caller must hold the i_mutex on the dentry's parent as required to
97 * call vfs_unlink(), vfs_rmdir() or vfs_rename()
99 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
100 struct dentry *dentry,
101 enum fscache_why_object_killed why)
103 struct cachefiles_object *object;
106 _enter(",'%pd'", dentry);
108 write_lock(&cache->active_lock);
110 p = cache->active_nodes.rb_node;
112 object = rb_entry(p, struct cachefiles_object, active_node);
113 if (object->dentry > dentry)
115 else if (object->dentry < dentry)
121 write_unlock(&cache->active_lock);
122 _leave(" [no owner]");
125 /* found the dentry for */
127 kdebug("preemptive burial: OBJ%x [%s] %p",
128 object->fscache.debug_id,
129 object->fscache.state->name,
132 if (fscache_object_is_live(&object->fscache)) {
134 pr_err("Error: Can't preemptively bury live object\n");
135 cachefiles_printk_object(object, NULL);
137 if (why != FSCACHE_OBJECT_IS_STALE)
138 fscache_object_mark_killed(&object->fscache, why);
141 write_unlock(&cache->active_lock);
142 _leave(" [owner marked]");
146 * record the fact that an object is now active
148 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
149 struct cachefiles_object *object)
151 struct cachefiles_object *xobject;
152 struct rb_node **_p, *_parent = NULL;
153 struct dentry *dentry;
155 _enter(",%p", object);
158 write_lock(&cache->active_lock);
160 if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
161 pr_err("Error: Object already active\n");
162 cachefiles_printk_object(object, NULL);
166 dentry = object->dentry;
167 _p = &cache->active_nodes.rb_node;
170 xobject = rb_entry(_parent,
171 struct cachefiles_object, active_node);
173 ASSERT(xobject != object);
175 if (xobject->dentry > dentry)
176 _p = &(*_p)->rb_left;
177 else if (xobject->dentry < dentry)
178 _p = &(*_p)->rb_right;
180 goto wait_for_old_object;
183 rb_link_node(&object->active_node, _parent, _p);
184 rb_insert_color(&object->active_node, &cache->active_nodes);
186 write_unlock(&cache->active_lock);
190 /* an old object from a previous incarnation is hogging the slot - we
191 * need to wait for it to be destroyed */
193 if (fscache_object_is_live(&xobject->fscache)) {
195 pr_err("Error: Unexpected object collision\n");
196 cachefiles_printk_object(object, xobject);
199 atomic_inc(&xobject->usage);
200 write_unlock(&cache->active_lock);
202 if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
203 wait_queue_head_t *wq;
205 signed long timeout = 60 * HZ;
209 /* if the object we're waiting for is queued for processing,
210 * then just put ourselves on the queue behind it */
211 if (work_pending(&xobject->fscache.work)) {
212 _debug("queue OBJ%x behind OBJ%x immediately",
213 object->fscache.debug_id,
214 xobject->fscache.debug_id);
218 /* otherwise we sleep until either the object we're waiting for
219 * is done, or the fscache_object is congested */
220 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
224 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
225 if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
228 requeue = fscache_object_sleep_till_congested(&timeout);
229 } while (timeout > 0 && !requeue);
230 finish_wait(wq, &wait);
233 test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
234 _debug("queue OBJ%x behind OBJ%x after wait",
235 object->fscache.debug_id,
236 xobject->fscache.debug_id);
242 pr_err("Error: Overlong wait for old active object to go away\n");
243 cachefiles_printk_object(object, xobject);
248 ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
250 cache->cache.ops->put_object(&xobject->fscache);
254 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
255 cache->cache.ops->put_object(&xobject->fscache);
256 _leave(" = -ETIMEDOUT");
261 * Mark an object as being inactive.
263 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
264 struct cachefiles_object *object)
266 write_lock(&cache->active_lock);
267 rb_erase(&object->active_node, &cache->active_nodes);
268 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
269 write_unlock(&cache->active_lock);
271 wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
273 /* This object can now be culled, so we need to let the daemon know
274 * that there is something it can remove if it needs to.
276 atomic_long_add(d_backing_inode(object->dentry)->i_blocks,
278 if (atomic_inc_return(&cache->f_released))
279 cachefiles_state_changed(cache);
283 * delete an object representation from the cache
284 * - file backed objects are unlinked
285 * - directory backed objects are stuffed into the graveyard for userspace to
287 * - unlocks the directory mutex
289 static int cachefiles_bury_object(struct cachefiles_cache *cache,
293 enum fscache_why_object_killed why)
295 struct dentry *grave, *trap;
296 struct path path, path_to_graveyard;
297 char nbuffer[8 + 8 + 1];
300 _enter(",'%pd','%pd'", dir, rep);
302 _debug("remove %p from %p", rep, dir);
304 /* non-directories can just be unlinked */
305 if (!d_is_dir(rep)) {
306 _debug("unlink stale object");
308 path.mnt = cache->mnt;
310 ret = security_path_unlink(&path, rep);
312 cachefiles_io_error(cache, "Unlink security error");
314 ret = vfs_unlink(d_inode(dir), rep, NULL);
317 cachefiles_mark_object_buried(cache, rep, why);
320 inode_unlock(d_inode(dir));
323 cachefiles_io_error(cache, "Unlink failed");
325 _leave(" = %d", ret);
329 /* directories have to be moved to the graveyard */
330 _debug("move stale object to graveyard");
331 inode_unlock(d_inode(dir));
334 /* first step is to make up a grave dentry in the graveyard */
335 sprintf(nbuffer, "%08x%08x",
336 (uint32_t) get_seconds(),
337 (uint32_t) atomic_inc_return(&cache->gravecounter));
339 /* do the multiway lock magic */
340 trap = lock_rename(cache->graveyard, dir);
342 /* do some checks before getting the grave dentry */
343 if (rep->d_parent != dir) {
344 /* the entry was probably culled when we dropped the parent dir
346 unlock_rename(cache->graveyard, dir);
347 _leave(" = 0 [culled?]");
351 if (!d_can_lookup(cache->graveyard)) {
352 unlock_rename(cache->graveyard, dir);
353 cachefiles_io_error(cache, "Graveyard no longer a directory");
358 unlock_rename(cache->graveyard, dir);
359 cachefiles_io_error(cache, "May not make directory loop");
363 if (d_mountpoint(rep)) {
364 unlock_rename(cache->graveyard, dir);
365 cachefiles_io_error(cache, "Mountpoint in cache");
369 grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
371 unlock_rename(cache->graveyard, dir);
373 if (PTR_ERR(grave) == -ENOMEM) {
374 _leave(" = -ENOMEM");
378 cachefiles_io_error(cache, "Lookup error %ld",
383 if (d_is_positive(grave)) {
384 unlock_rename(cache->graveyard, dir);
391 if (d_mountpoint(grave)) {
392 unlock_rename(cache->graveyard, dir);
394 cachefiles_io_error(cache, "Mountpoint in graveyard");
398 /* target should not be an ancestor of source */
400 unlock_rename(cache->graveyard, dir);
402 cachefiles_io_error(cache, "May not make directory loop");
406 /* attempt the rename */
407 path.mnt = cache->mnt;
409 path_to_graveyard.mnt = cache->mnt;
410 path_to_graveyard.dentry = cache->graveyard;
411 ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
413 cachefiles_io_error(cache, "Rename security error %d", ret);
415 ret = vfs_rename(d_inode(dir), rep,
416 d_inode(cache->graveyard), grave, NULL, 0);
417 if (ret != 0 && ret != -ENOMEM)
418 cachefiles_io_error(cache,
419 "Rename failed with error %d", ret);
422 cachefiles_mark_object_buried(cache, rep, why);
425 unlock_rename(cache->graveyard, dir);
432 * delete an object representation from the cache
434 int cachefiles_delete_object(struct cachefiles_cache *cache,
435 struct cachefiles_object *object)
440 _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
442 ASSERT(object->dentry);
443 ASSERT(d_backing_inode(object->dentry));
444 ASSERT(object->dentry->d_parent);
446 dir = dget_parent(object->dentry);
448 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
450 if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
451 /* object allocation for the same key preemptively deleted this
452 * object's file so that it could create its own file */
453 _debug("object preemptively buried");
454 inode_unlock(d_inode(dir));
457 /* we need to check that our parent is _still_ our parent - it
458 * may have been renamed */
459 if (dir == object->dentry->d_parent) {
460 ret = cachefiles_bury_object(cache, dir,
461 object->dentry, false,
462 FSCACHE_OBJECT_WAS_RETIRED);
464 /* it got moved, presumably by cachefilesd culling it,
465 * so it's no longer in the key path and we can ignore
467 inode_unlock(d_inode(dir));
473 _leave(" = %d", ret);
478 * walk from the parent object to the child object through the backing
479 * filesystem, creating directories as we go
481 int cachefiles_walk_to_object(struct cachefiles_object *parent,
482 struct cachefiles_object *object,
484 struct cachefiles_xattr *auxdata)
486 struct cachefiles_cache *cache;
487 struct dentry *dir, *next = NULL;
493 _enter("OBJ%x{%p},OBJ%x,%s,",
494 parent->fscache.debug_id, parent->dentry,
495 object->fscache.debug_id, key);
497 cache = container_of(parent->fscache.cache,
498 struct cachefiles_cache, cache);
499 path.mnt = cache->mnt;
501 ASSERT(parent->dentry);
502 ASSERT(d_backing_inode(parent->dentry));
504 if (!(d_is_dir(parent->dentry))) {
505 // TODO: convert file to dir
506 _leave("looking up in none directory");
510 dir = dget(parent->dentry);
513 /* attempt to transit the first directory component */
517 /* key ends in a double NUL */
518 key = key + nlen + 1;
523 /* search the current directory for the element name */
524 _debug("lookup '%s'", name);
526 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
529 next = lookup_one_len(name, dir, nlen);
530 cachefiles_hist(cachefiles_lookup_histogram, start);
534 _debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
537 object->new = !d_backing_inode(next);
539 /* if this element of the path doesn't exist, then the lookup phase
540 * failed, and we can release any readers in the certain knowledge that
541 * there's nothing for them to actually read */
542 if (d_is_negative(next))
543 fscache_object_lookup_negative(&object->fscache);
545 /* we need to create the object if it's negative */
546 if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
547 /* index objects and intervening tree levels must be subdirs */
548 if (d_is_negative(next)) {
549 ret = cachefiles_has_space(cache, 1, 0);
554 ret = security_path_mkdir(&path, next, 0);
558 ret = vfs_mkdir(d_inode(dir), next, 0);
559 cachefiles_hist(cachefiles_mkdir_histogram, start);
563 ASSERT(d_backing_inode(next));
565 _debug("mkdir -> %p{%p{ino=%lu}}",
566 next, d_backing_inode(next), d_backing_inode(next)->i_ino);
568 } else if (!d_can_lookup(next)) {
569 pr_err("inode %lu is not a directory\n",
570 d_backing_inode(next)->i_ino);
576 /* non-index objects start out life as files */
577 if (d_is_negative(next)) {
578 ret = cachefiles_has_space(cache, 1, 0);
583 ret = security_path_mknod(&path, next, S_IFREG, 0);
587 ret = vfs_create(d_inode(dir), next, S_IFREG, true);
588 cachefiles_hist(cachefiles_create_histogram, start);
592 ASSERT(d_backing_inode(next));
594 _debug("create -> %p{%p{ino=%lu}}",
595 next, d_backing_inode(next), d_backing_inode(next)->i_ino);
597 } else if (!d_can_lookup(next) &&
600 pr_err("inode %lu is not a file or directory\n",
601 d_backing_inode(next)->i_ino);
607 /* process the next component */
610 inode_unlock(d_inode(dir));
617 /* we've found the object we were looking for */
618 object->dentry = next;
620 /* if we've found that the terminal object exists, then we need to
621 * check its attributes and delete it if it's out of date */
623 _debug("validate '%pd'", next);
625 ret = cachefiles_check_object_xattr(object, auxdata);
626 if (ret == -ESTALE) {
627 /* delete the object (the deleter drops the directory
629 object->dentry = NULL;
631 ret = cachefiles_bury_object(cache, dir, next, true,
632 FSCACHE_OBJECT_IS_STALE);
639 _debug("redo lookup");
640 fscache_object_retrying_stale(&object->fscache);
645 /* note that we're now using this object */
646 ret = cachefiles_mark_object_active(cache, object);
648 inode_unlock(d_inode(dir));
652 if (ret == -ETIMEDOUT)
653 goto mark_active_timed_out;
655 _debug("=== OBTAINED_OBJECT ===");
658 /* attach data to a newly constructed terminal object */
659 ret = cachefiles_set_object_xattr(object, auxdata);
663 /* always update the atime on an object we've just looked up
664 * (this is used to keep track of culling, and atimes are only
665 * updated by read, write and readdir but not lookup or
671 /* open a file interface onto a data file */
672 if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
673 if (d_is_reg(object->dentry)) {
674 const struct address_space_operations *aops;
677 aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
680 if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
683 object->backer = object->dentry;
685 BUG(); // TODO: open file in data-class subdir
690 fscache_obtained_object(&object->fscache);
692 _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
696 fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
698 _debug("create error %d", ret);
700 cachefiles_io_error(cache, "Create/mkdir failed");
703 mark_active_timed_out:
704 _debug("mark active timed out");
708 _debug("check error %d", ret);
709 cachefiles_mark_object_inactive(cache, object);
711 dput(object->dentry);
712 object->dentry = NULL;
716 _debug("delete error %d", ret);
720 _debug("lookup error %ld", PTR_ERR(next));
723 cachefiles_io_error(cache, "Lookup failed");
726 inode_unlock(d_inode(dir));
731 _leave(" = error %d", -ret);
738 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
742 struct dentry *subdir;
747 _enter(",,%s", dirname);
749 /* search the current directory for the element name */
750 inode_lock(d_inode(dir));
753 subdir = lookup_one_len(dirname, dir, strlen(dirname));
754 cachefiles_hist(cachefiles_lookup_histogram, start);
755 if (IS_ERR(subdir)) {
756 if (PTR_ERR(subdir) == -ENOMEM)
761 _debug("subdir -> %p %s",
762 subdir, d_backing_inode(subdir) ? "positive" : "negative");
764 /* we need to create the subdir if it doesn't exist yet */
765 if (d_is_negative(subdir)) {
766 ret = cachefiles_has_space(cache, 1, 0);
770 _debug("attempt mkdir");
772 path.mnt = cache->mnt;
774 ret = security_path_mkdir(&path, subdir, 0700);
777 ret = vfs_mkdir(d_inode(dir), subdir, 0700);
781 ASSERT(d_backing_inode(subdir));
783 _debug("mkdir -> %p{%p{ino=%lu}}",
785 d_backing_inode(subdir),
786 d_backing_inode(subdir)->i_ino);
789 inode_unlock(d_inode(dir));
791 /* we need to make sure the subdir is a directory */
792 ASSERT(d_backing_inode(subdir));
794 if (!d_can_lookup(subdir)) {
795 pr_err("%s is not a directory\n", dirname);
801 if (!d_backing_inode(subdir)->i_op->setxattr ||
802 !d_backing_inode(subdir)->i_op->getxattr ||
803 !d_backing_inode(subdir)->i_op->lookup ||
804 !d_backing_inode(subdir)->i_op->mkdir ||
805 !d_backing_inode(subdir)->i_op->create ||
806 (!d_backing_inode(subdir)->i_op->rename &&
807 !d_backing_inode(subdir)->i_op->rename2) ||
808 !d_backing_inode(subdir)->i_op->rmdir ||
809 !d_backing_inode(subdir)->i_op->unlink)
812 _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
817 _leave(" = %d [check]", ret);
821 inode_unlock(d_inode(dir));
823 pr_err("mkdir %s failed with error %d\n", dirname, ret);
827 inode_unlock(d_inode(dir));
828 ret = PTR_ERR(subdir);
829 pr_err("Lookup %s failed with error %d\n", dirname, ret);
833 inode_unlock(d_inode(dir));
834 _leave(" = -ENOMEM");
835 return ERR_PTR(-ENOMEM);
839 * find out if an object is in use or not
840 * - if finds object and it's not in use:
841 * - returns a pointer to the object and a reference on it
842 * - returns with the directory locked
844 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
848 struct cachefiles_object *object;
850 struct dentry *victim;
857 /* look up the victim */
858 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
861 victim = lookup_one_len(filename, dir, strlen(filename));
862 cachefiles_hist(cachefiles_lookup_histogram, start);
866 //_debug("victim -> %p %s",
867 // victim, d_backing_inode(victim) ? "positive" : "negative");
869 /* if the object is no longer there then we probably retired the object
870 * at the netfs's request whilst the cull was in progress
872 if (d_is_negative(victim)) {
873 inode_unlock(d_inode(dir));
875 _leave(" = -ENOENT [absent]");
876 return ERR_PTR(-ENOENT);
879 /* check to see if we're using this object */
880 read_lock(&cache->active_lock);
882 _n = cache->active_nodes.rb_node;
885 object = rb_entry(_n, struct cachefiles_object, active_node);
887 if (object->dentry > victim)
889 else if (object->dentry < victim)
895 read_unlock(&cache->active_lock);
897 //_leave(" = %p", victim);
901 read_unlock(&cache->active_lock);
902 inode_unlock(d_inode(dir));
904 //_leave(" = -EBUSY [in use]");
905 return ERR_PTR(-EBUSY);
908 inode_unlock(d_inode(dir));
909 ret = PTR_ERR(victim);
910 if (ret == -ENOENT) {
911 /* file or dir now absent - probably retired by netfs */
912 _leave(" = -ESTALE [absent]");
913 return ERR_PTR(-ESTALE);
917 cachefiles_io_error(cache, "Lookup failed");
918 } else if (ret != -ENOMEM) {
919 pr_err("Internal error: %d\n", ret);
923 _leave(" = %d", ret);
928 * cull an object if it's not in use
929 * - called only by cache manager daemon
931 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
934 struct dentry *victim;
937 _enter(",%pd/,%s", dir, filename);
939 victim = cachefiles_check_active(cache, dir, filename);
941 return PTR_ERR(victim);
943 _debug("victim -> %p %s",
944 victim, d_backing_inode(victim) ? "positive" : "negative");
946 /* okay... the victim is not being used so we can cull it
947 * - start by marking it as stale
949 _debug("victim is cullable");
951 ret = cachefiles_remove_object_xattr(cache, victim);
955 /* actually remove the victim (drops the dir mutex) */
958 ret = cachefiles_bury_object(cache, dir, victim, false,
959 FSCACHE_OBJECT_WAS_CULLED);
968 inode_unlock(d_inode(dir));
971 if (ret == -ENOENT) {
972 /* file or dir now absent - probably retired by netfs */
973 _leave(" = -ESTALE [absent]");
977 if (ret != -ENOMEM) {
978 pr_err("Internal error: %d\n", ret);
982 _leave(" = %d", ret);
987 * find out if an object is in use or not
988 * - called only by cache manager daemon
989 * - returns -EBUSY or 0 to indicate whether an object is in use or not
991 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
994 struct dentry *victim;
999 victim = cachefiles_check_active(cache, dir, filename);
1001 return PTR_ERR(victim);
1003 inode_unlock(d_inode(dir));