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>
23 #include <linux/xattr.h>
26 #define CACHEFILES_KEYBUF_SIZE 512
29 * dump debugging info about an object
32 void __cachefiles_printk_object(struct cachefiles_object *object,
36 struct fscache_cookie *cookie;
37 unsigned keylen, loop;
39 pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
40 pr_err("%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
41 prefix, object->fscache.state->name,
42 object->fscache.flags, work_busy(&object->fscache.work),
43 object->fscache.events, object->fscache.event_mask);
44 pr_err("%sops=%u inp=%u exc=%u\n",
45 prefix, object->fscache.n_ops, object->fscache.n_in_progress,
46 object->fscache.n_exclusive);
47 pr_err("%sparent=%p\n",
48 prefix, object->fscache.parent);
50 spin_lock(&object->fscache.lock);
51 cookie = object->fscache.cookie;
53 pr_err("%scookie=%p [pr=%p nd=%p fl=%lx]\n",
55 object->fscache.cookie,
56 object->fscache.cookie->parent,
57 object->fscache.cookie->netfs_data,
58 object->fscache.cookie->flags);
59 if (keybuf && cookie->def)
60 keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
61 CACHEFILES_KEYBUF_SIZE);
65 pr_err("%scookie=NULL\n", prefix);
68 spin_unlock(&object->fscache.lock);
71 pr_err("%skey=[%u] '", prefix, keylen);
72 for (loop = 0; loop < keylen; loop++)
73 pr_cont("%02x", keybuf[loop]);
79 * dump debugging info about a pair of objects
81 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
82 struct cachefiles_object *xobject)
86 keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
88 __cachefiles_printk_object(object, "", keybuf);
90 __cachefiles_printk_object(xobject, "x", keybuf);
95 * mark the owner of a dentry, if there is one, to indicate that that dentry
96 * has been preemptively deleted
97 * - the caller must hold the i_mutex on the dentry's parent as required to
98 * call vfs_unlink(), vfs_rmdir() or vfs_rename()
100 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
101 struct dentry *dentry,
102 enum fscache_why_object_killed why)
104 struct cachefiles_object *object;
107 _enter(",'%pd'", dentry);
109 write_lock(&cache->active_lock);
111 p = cache->active_nodes.rb_node;
113 object = rb_entry(p, struct cachefiles_object, active_node);
114 if (object->dentry > dentry)
116 else if (object->dentry < dentry)
122 write_unlock(&cache->active_lock);
123 _leave(" [no owner]");
126 /* found the dentry for */
128 kdebug("preemptive burial: OBJ%x [%s] %p",
129 object->fscache.debug_id,
130 object->fscache.state->name,
133 if (fscache_object_is_live(&object->fscache)) {
135 pr_err("Error: Can't preemptively bury live object\n");
136 cachefiles_printk_object(object, NULL);
138 if (why != FSCACHE_OBJECT_IS_STALE)
139 fscache_object_mark_killed(&object->fscache, why);
142 write_unlock(&cache->active_lock);
143 _leave(" [owner marked]");
147 * record the fact that an object is now active
149 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
150 struct cachefiles_object *object)
152 struct cachefiles_object *xobject;
153 struct rb_node **_p, *_parent = NULL;
154 struct dentry *dentry;
156 _enter(",%p", object);
159 write_lock(&cache->active_lock);
161 if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
162 pr_err("Error: Object already active\n");
163 cachefiles_printk_object(object, NULL);
167 dentry = object->dentry;
168 _p = &cache->active_nodes.rb_node;
171 xobject = rb_entry(_parent,
172 struct cachefiles_object, active_node);
174 ASSERT(xobject != object);
176 if (xobject->dentry > dentry)
177 _p = &(*_p)->rb_left;
178 else if (xobject->dentry < dentry)
179 _p = &(*_p)->rb_right;
181 goto wait_for_old_object;
184 rb_link_node(&object->active_node, _parent, _p);
185 rb_insert_color(&object->active_node, &cache->active_nodes);
187 write_unlock(&cache->active_lock);
191 /* an old object from a previous incarnation is hogging the slot - we
192 * need to wait for it to be destroyed */
194 if (fscache_object_is_live(&xobject->fscache)) {
196 pr_err("Error: Unexpected object collision\n");
197 cachefiles_printk_object(object, xobject);
200 atomic_inc(&xobject->usage);
201 write_unlock(&cache->active_lock);
203 if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
204 wait_queue_head_t *wq;
206 signed long timeout = 60 * HZ;
210 /* if the object we're waiting for is queued for processing,
211 * then just put ourselves on the queue behind it */
212 if (work_pending(&xobject->fscache.work)) {
213 _debug("queue OBJ%x behind OBJ%x immediately",
214 object->fscache.debug_id,
215 xobject->fscache.debug_id);
219 /* otherwise we sleep until either the object we're waiting for
220 * is done, or the fscache_object is congested */
221 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
225 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
226 if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
229 requeue = fscache_object_sleep_till_congested(&timeout);
230 } while (timeout > 0 && !requeue);
231 finish_wait(wq, &wait);
234 test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
235 _debug("queue OBJ%x behind OBJ%x after wait",
236 object->fscache.debug_id,
237 xobject->fscache.debug_id);
243 pr_err("Error: Overlong wait for old active object to go away\n");
244 cachefiles_printk_object(object, xobject);
249 ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
251 cache->cache.ops->put_object(&xobject->fscache);
255 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
256 cache->cache.ops->put_object(&xobject->fscache);
257 _leave(" = -ETIMEDOUT");
262 * Mark an object as being inactive.
264 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
265 struct cachefiles_object *object)
267 blkcnt_t i_blocks = d_backing_inode(object->dentry)->i_blocks;
269 write_lock(&cache->active_lock);
270 rb_erase(&object->active_node, &cache->active_nodes);
271 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
272 write_unlock(&cache->active_lock);
274 wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
276 /* This object can now be culled, so we need to let the daemon know
277 * that there is something it can remove if it needs to.
279 atomic_long_add(i_blocks, &cache->b_released);
280 if (atomic_inc_return(&cache->f_released))
281 cachefiles_state_changed(cache);
285 * delete an object representation from the cache
286 * - file backed objects are unlinked
287 * - directory backed objects are stuffed into the graveyard for userspace to
289 * - unlocks the directory mutex
291 static int cachefiles_bury_object(struct cachefiles_cache *cache,
295 enum fscache_why_object_killed why)
297 struct dentry *grave, *trap;
298 struct path path, path_to_graveyard;
299 char nbuffer[8 + 8 + 1];
302 _enter(",'%pd','%pd'", dir, rep);
304 _debug("remove %p from %p", rep, dir);
306 /* non-directories can just be unlinked */
307 if (!d_is_dir(rep)) {
308 _debug("unlink stale object");
310 path.mnt = cache->mnt;
312 ret = security_path_unlink(&path, rep);
314 cachefiles_io_error(cache, "Unlink security error");
316 ret = vfs_unlink(d_inode(dir), rep, NULL);
319 cachefiles_mark_object_buried(cache, rep, why);
322 inode_unlock(d_inode(dir));
325 cachefiles_io_error(cache, "Unlink failed");
327 _leave(" = %d", ret);
331 /* directories have to be moved to the graveyard */
332 _debug("move stale object to graveyard");
333 inode_unlock(d_inode(dir));
336 /* first step is to make up a grave dentry in the graveyard */
337 sprintf(nbuffer, "%08x%08x",
338 (uint32_t) get_seconds(),
339 (uint32_t) atomic_inc_return(&cache->gravecounter));
341 /* do the multiway lock magic */
342 trap = lock_rename(cache->graveyard, dir);
344 /* do some checks before getting the grave dentry */
345 if (rep->d_parent != dir) {
346 /* the entry was probably culled when we dropped the parent dir
348 unlock_rename(cache->graveyard, dir);
349 _leave(" = 0 [culled?]");
353 if (!d_can_lookup(cache->graveyard)) {
354 unlock_rename(cache->graveyard, dir);
355 cachefiles_io_error(cache, "Graveyard no longer a directory");
360 unlock_rename(cache->graveyard, dir);
361 cachefiles_io_error(cache, "May not make directory loop");
365 if (d_mountpoint(rep)) {
366 unlock_rename(cache->graveyard, dir);
367 cachefiles_io_error(cache, "Mountpoint in cache");
371 grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
373 unlock_rename(cache->graveyard, dir);
375 if (PTR_ERR(grave) == -ENOMEM) {
376 _leave(" = -ENOMEM");
380 cachefiles_io_error(cache, "Lookup error %ld",
385 if (d_is_positive(grave)) {
386 unlock_rename(cache->graveyard, dir);
393 if (d_mountpoint(grave)) {
394 unlock_rename(cache->graveyard, dir);
396 cachefiles_io_error(cache, "Mountpoint in graveyard");
400 /* target should not be an ancestor of source */
402 unlock_rename(cache->graveyard, dir);
404 cachefiles_io_error(cache, "May not make directory loop");
408 /* attempt the rename */
409 path.mnt = cache->mnt;
411 path_to_graveyard.mnt = cache->mnt;
412 path_to_graveyard.dentry = cache->graveyard;
413 ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
415 cachefiles_io_error(cache, "Rename security error %d", ret);
417 ret = vfs_rename(d_inode(dir), rep,
418 d_inode(cache->graveyard), grave, NULL, 0);
419 if (ret != 0 && ret != -ENOMEM)
420 cachefiles_io_error(cache,
421 "Rename failed with error %d", ret);
424 cachefiles_mark_object_buried(cache, rep, why);
427 unlock_rename(cache->graveyard, dir);
434 * delete an object representation from the cache
436 int cachefiles_delete_object(struct cachefiles_cache *cache,
437 struct cachefiles_object *object)
442 _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
444 ASSERT(object->dentry);
445 ASSERT(d_backing_inode(object->dentry));
446 ASSERT(object->dentry->d_parent);
448 dir = dget_parent(object->dentry);
450 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
452 if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
453 /* object allocation for the same key preemptively deleted this
454 * object's file so that it could create its own file */
455 _debug("object preemptively buried");
456 inode_unlock(d_inode(dir));
459 /* we need to check that our parent is _still_ our parent - it
460 * may have been renamed */
461 if (dir == object->dentry->d_parent) {
462 ret = cachefiles_bury_object(cache, dir,
463 object->dentry, false,
464 FSCACHE_OBJECT_WAS_RETIRED);
466 /* it got moved, presumably by cachefilesd culling it,
467 * so it's no longer in the key path and we can ignore
469 inode_unlock(d_inode(dir));
475 _leave(" = %d", ret);
480 * walk from the parent object to the child object through the backing
481 * filesystem, creating directories as we go
483 int cachefiles_walk_to_object(struct cachefiles_object *parent,
484 struct cachefiles_object *object,
486 struct cachefiles_xattr *auxdata)
488 struct cachefiles_cache *cache;
489 struct dentry *dir, *next = NULL;
495 _enter("OBJ%x{%p},OBJ%x,%s,",
496 parent->fscache.debug_id, parent->dentry,
497 object->fscache.debug_id, key);
499 cache = container_of(parent->fscache.cache,
500 struct cachefiles_cache, cache);
501 path.mnt = cache->mnt;
503 ASSERT(parent->dentry);
504 ASSERT(d_backing_inode(parent->dentry));
506 if (!(d_is_dir(parent->dentry))) {
507 // TODO: convert file to dir
508 _leave("looking up in none directory");
512 dir = dget(parent->dentry);
515 /* attempt to transit the first directory component */
519 /* key ends in a double NUL */
520 key = key + nlen + 1;
525 /* search the current directory for the element name */
526 _debug("lookup '%s'", name);
528 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
531 next = lookup_one_len(name, dir, nlen);
532 cachefiles_hist(cachefiles_lookup_histogram, start);
536 _debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
539 object->new = !d_backing_inode(next);
541 /* if this element of the path doesn't exist, then the lookup phase
542 * failed, and we can release any readers in the certain knowledge that
543 * there's nothing for them to actually read */
544 if (d_is_negative(next))
545 fscache_object_lookup_negative(&object->fscache);
547 /* we need to create the object if it's negative */
548 if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
549 /* index objects and intervening tree levels must be subdirs */
550 if (d_is_negative(next)) {
551 ret = cachefiles_has_space(cache, 1, 0);
556 ret = security_path_mkdir(&path, next, 0);
560 ret = vfs_mkdir(d_inode(dir), next, 0);
561 cachefiles_hist(cachefiles_mkdir_histogram, start);
565 ASSERT(d_backing_inode(next));
567 _debug("mkdir -> %p{%p{ino=%lu}}",
568 next, d_backing_inode(next), d_backing_inode(next)->i_ino);
570 } else if (!d_can_lookup(next)) {
571 pr_err("inode %lu is not a directory\n",
572 d_backing_inode(next)->i_ino);
578 /* non-index objects start out life as files */
579 if (d_is_negative(next)) {
580 ret = cachefiles_has_space(cache, 1, 0);
585 ret = security_path_mknod(&path, next, S_IFREG, 0);
589 ret = vfs_create(d_inode(dir), next, S_IFREG, true);
590 cachefiles_hist(cachefiles_create_histogram, start);
594 ASSERT(d_backing_inode(next));
596 _debug("create -> %p{%p{ino=%lu}}",
597 next, d_backing_inode(next), d_backing_inode(next)->i_ino);
599 } else if (!d_can_lookup(next) &&
602 pr_err("inode %lu is not a file or directory\n",
603 d_backing_inode(next)->i_ino);
609 /* process the next component */
612 inode_unlock(d_inode(dir));
619 /* we've found the object we were looking for */
620 object->dentry = next;
622 /* if we've found that the terminal object exists, then we need to
623 * check its attributes and delete it if it's out of date */
625 _debug("validate '%pd'", next);
627 ret = cachefiles_check_object_xattr(object, auxdata);
628 if (ret == -ESTALE) {
629 /* delete the object (the deleter drops the directory
631 object->dentry = NULL;
633 ret = cachefiles_bury_object(cache, dir, next, true,
634 FSCACHE_OBJECT_IS_STALE);
641 _debug("redo lookup");
642 fscache_object_retrying_stale(&object->fscache);
647 /* note that we're now using this object */
648 ret = cachefiles_mark_object_active(cache, object);
650 inode_unlock(d_inode(dir));
654 if (ret == -ETIMEDOUT)
655 goto mark_active_timed_out;
657 _debug("=== OBTAINED_OBJECT ===");
660 /* attach data to a newly constructed terminal object */
661 ret = cachefiles_set_object_xattr(object, auxdata);
665 /* always update the atime on an object we've just looked up
666 * (this is used to keep track of culling, and atimes are only
667 * updated by read, write and readdir but not lookup or
673 /* open a file interface onto a data file */
674 if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
675 if (d_is_reg(object->dentry)) {
676 const struct address_space_operations *aops;
679 aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
682 if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
685 object->backer = object->dentry;
687 BUG(); // TODO: open file in data-class subdir
692 fscache_obtained_object(&object->fscache);
694 _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
698 fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
700 _debug("create error %d", ret);
702 cachefiles_io_error(cache, "Create/mkdir failed");
705 mark_active_timed_out:
706 _debug("mark active timed out");
710 _debug("check error %d", ret);
711 cachefiles_mark_object_inactive(cache, object);
713 dput(object->dentry);
714 object->dentry = NULL;
718 _debug("delete error %d", ret);
722 _debug("lookup error %ld", PTR_ERR(next));
725 cachefiles_io_error(cache, "Lookup failed");
728 inode_unlock(d_inode(dir));
733 _leave(" = error %d", -ret);
740 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
744 struct dentry *subdir;
749 _enter(",,%s", dirname);
751 /* search the current directory for the element name */
752 inode_lock(d_inode(dir));
755 subdir = lookup_one_len(dirname, dir, strlen(dirname));
756 cachefiles_hist(cachefiles_lookup_histogram, start);
757 if (IS_ERR(subdir)) {
758 if (PTR_ERR(subdir) == -ENOMEM)
763 _debug("subdir -> %p %s",
764 subdir, d_backing_inode(subdir) ? "positive" : "negative");
766 /* we need to create the subdir if it doesn't exist yet */
767 if (d_is_negative(subdir)) {
768 ret = cachefiles_has_space(cache, 1, 0);
772 _debug("attempt mkdir");
774 path.mnt = cache->mnt;
776 ret = security_path_mkdir(&path, subdir, 0700);
779 ret = vfs_mkdir(d_inode(dir), subdir, 0700);
783 ASSERT(d_backing_inode(subdir));
785 _debug("mkdir -> %p{%p{ino=%lu}}",
787 d_backing_inode(subdir),
788 d_backing_inode(subdir)->i_ino);
791 inode_unlock(d_inode(dir));
793 /* we need to make sure the subdir is a directory */
794 ASSERT(d_backing_inode(subdir));
796 if (!d_can_lookup(subdir)) {
797 pr_err("%s is not a directory\n", dirname);
803 if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
804 !d_backing_inode(subdir)->i_op->lookup ||
805 !d_backing_inode(subdir)->i_op->mkdir ||
806 !d_backing_inode(subdir)->i_op->create ||
807 (!d_backing_inode(subdir)->i_op->rename &&
808 !d_backing_inode(subdir)->i_op->rename2) ||
809 !d_backing_inode(subdir)->i_op->rmdir ||
810 !d_backing_inode(subdir)->i_op->unlink)
813 _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
818 _leave(" = %d [check]", ret);
822 inode_unlock(d_inode(dir));
824 pr_err("mkdir %s failed with error %d\n", dirname, ret);
828 inode_unlock(d_inode(dir));
829 ret = PTR_ERR(subdir);
830 pr_err("Lookup %s failed with error %d\n", dirname, ret);
834 inode_unlock(d_inode(dir));
835 _leave(" = -ENOMEM");
836 return ERR_PTR(-ENOMEM);
840 * find out if an object is in use or not
841 * - if finds object and it's not in use:
842 * - returns a pointer to the object and a reference on it
843 * - returns with the directory locked
845 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
849 struct cachefiles_object *object;
851 struct dentry *victim;
858 /* look up the victim */
859 inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
862 victim = lookup_one_len(filename, dir, strlen(filename));
863 cachefiles_hist(cachefiles_lookup_histogram, start);
867 //_debug("victim -> %p %s",
868 // victim, d_backing_inode(victim) ? "positive" : "negative");
870 /* if the object is no longer there then we probably retired the object
871 * at the netfs's request whilst the cull was in progress
873 if (d_is_negative(victim)) {
874 inode_unlock(d_inode(dir));
876 _leave(" = -ENOENT [absent]");
877 return ERR_PTR(-ENOENT);
880 /* check to see if we're using this object */
881 read_lock(&cache->active_lock);
883 _n = cache->active_nodes.rb_node;
886 object = rb_entry(_n, struct cachefiles_object, active_node);
888 if (object->dentry > victim)
890 else if (object->dentry < victim)
896 read_unlock(&cache->active_lock);
898 //_leave(" = %p", victim);
902 read_unlock(&cache->active_lock);
903 inode_unlock(d_inode(dir));
905 //_leave(" = -EBUSY [in use]");
906 return ERR_PTR(-EBUSY);
909 inode_unlock(d_inode(dir));
910 ret = PTR_ERR(victim);
911 if (ret == -ENOENT) {
912 /* file or dir now absent - probably retired by netfs */
913 _leave(" = -ESTALE [absent]");
914 return ERR_PTR(-ESTALE);
918 cachefiles_io_error(cache, "Lookup failed");
919 } else if (ret != -ENOMEM) {
920 pr_err("Internal error: %d\n", ret);
924 _leave(" = %d", ret);
929 * cull an object if it's not in use
930 * - called only by cache manager daemon
932 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
935 struct dentry *victim;
938 _enter(",%pd/,%s", dir, filename);
940 victim = cachefiles_check_active(cache, dir, filename);
942 return PTR_ERR(victim);
944 _debug("victim -> %p %s",
945 victim, d_backing_inode(victim) ? "positive" : "negative");
947 /* okay... the victim is not being used so we can cull it
948 * - start by marking it as stale
950 _debug("victim is cullable");
952 ret = cachefiles_remove_object_xattr(cache, victim);
956 /* actually remove the victim (drops the dir mutex) */
959 ret = cachefiles_bury_object(cache, dir, victim, false,
960 FSCACHE_OBJECT_WAS_CULLED);
969 inode_unlock(d_inode(dir));
972 if (ret == -ENOENT) {
973 /* file or dir now absent - probably retired by netfs */
974 _leave(" = -ESTALE [absent]");
978 if (ret != -ENOMEM) {
979 pr_err("Internal error: %d\n", ret);
983 _leave(" = %d", ret);
988 * find out if an object is in use or not
989 * - called only by cache manager daemon
990 * - returns -EBUSY or 0 to indicate whether an object is in use or not
992 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
995 struct dentry *victim;
1000 victim = cachefiles_check_active(cache, dir, filename);
1002 return PTR_ERR(victim);
1004 inode_unlock(d_inode(dir));