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 printk(KERN_ERR "%sobject: OBJ%x\n",
39 prefix, object->fscache.debug_id);
40 printk(KERN_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 printk(KERN_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 printk(KERN_ERR "%sparent=%p\n",
48 prefix, object->fscache.parent);
50 spin_lock(&object->fscache.lock);
51 cookie = object->fscache.cookie;
53 printk(KERN_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 printk(KERN_ERR "%scookie=NULL\n", prefix);
68 spin_unlock(&object->fscache.lock);
71 printk(KERN_ERR "%skey=[%u] '", prefix, keylen);
72 for (loop = 0; loop < keylen; loop++)
73 printk("%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)
103 struct cachefiles_object *object;
107 dentry->d_name.len, dentry->d_name.len, dentry->d_name.name);
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)) {
134 printk(KERN_ERR "\n");
135 printk(KERN_ERR "CacheFiles: Error:"
136 " Can't preemptively bury live object\n");
137 cachefiles_printk_object(object, NULL);
138 } else if (test_and_set_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
139 printk(KERN_ERR "CacheFiles: Error:"
140 " Object already preemptively buried\n");
143 write_unlock(&cache->active_lock);
144 _leave(" [owner marked]");
148 * record the fact that an object is now active
150 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
151 struct cachefiles_object *object)
153 struct cachefiles_object *xobject;
154 struct rb_node **_p, *_parent = NULL;
155 struct dentry *dentry;
157 _enter(",%p", object);
160 write_lock(&cache->active_lock);
162 if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
163 printk(KERN_ERR "CacheFiles: Error: Object already active\n");
164 cachefiles_printk_object(object, NULL);
168 dentry = object->dentry;
169 _p = &cache->active_nodes.rb_node;
172 xobject = rb_entry(_parent,
173 struct cachefiles_object, active_node);
175 ASSERT(xobject != object);
177 if (xobject->dentry > dentry)
178 _p = &(*_p)->rb_left;
179 else if (xobject->dentry < dentry)
180 _p = &(*_p)->rb_right;
182 goto wait_for_old_object;
185 rb_link_node(&object->active_node, _parent, _p);
186 rb_insert_color(&object->active_node, &cache->active_nodes);
188 write_unlock(&cache->active_lock);
192 /* an old object from a previous incarnation is hogging the slot - we
193 * need to wait for it to be destroyed */
195 if (fscache_object_is_live(&object->fscache)) {
196 printk(KERN_ERR "\n");
197 printk(KERN_ERR "CacheFiles: Error:"
198 " Unexpected object collision\n");
199 cachefiles_printk_object(object, xobject);
202 atomic_inc(&xobject->usage);
203 write_unlock(&cache->active_lock);
205 if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
206 wait_queue_head_t *wq;
208 signed long timeout = 60 * HZ;
212 /* if the object we're waiting for is queued for processing,
213 * then just put ourselves on the queue behind it */
214 if (work_pending(&xobject->fscache.work)) {
215 _debug("queue OBJ%x behind OBJ%x immediately",
216 object->fscache.debug_id,
217 xobject->fscache.debug_id);
221 /* otherwise we sleep until either the object we're waiting for
222 * is done, or the fscache_object is congested */
223 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
227 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
228 if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
231 requeue = fscache_object_sleep_till_congested(&timeout);
232 } while (timeout > 0 && !requeue);
233 finish_wait(wq, &wait);
236 test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
237 _debug("queue OBJ%x behind OBJ%x after wait",
238 object->fscache.debug_id,
239 xobject->fscache.debug_id);
244 printk(KERN_ERR "\n");
245 printk(KERN_ERR "CacheFiles: Error: Overlong"
246 " wait for old active object to go away\n");
247 cachefiles_printk_object(object, xobject);
252 ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
254 cache->cache.ops->put_object(&xobject->fscache);
258 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
259 cache->cache.ops->put_object(&xobject->fscache);
260 _leave(" = -ETIMEDOUT");
265 * delete an object representation from the cache
266 * - file backed objects are unlinked
267 * - directory backed objects are stuffed into the graveyard for userspace to
269 * - unlocks the directory mutex
271 static int cachefiles_bury_object(struct cachefiles_cache *cache,
276 struct dentry *grave, *trap;
277 struct path path, path_to_graveyard;
278 char nbuffer[8 + 8 + 1];
281 _enter(",'%*.*s','%*.*s'",
282 dir->d_name.len, dir->d_name.len, dir->d_name.name,
283 rep->d_name.len, rep->d_name.len, rep->d_name.name);
285 _debug("remove %p from %p", rep, dir);
287 /* non-directories can just be unlinked */
288 if (!S_ISDIR(rep->d_inode->i_mode)) {
289 _debug("unlink stale object");
291 path.mnt = cache->mnt;
293 ret = security_path_unlink(&path, rep);
295 cachefiles_io_error(cache, "Unlink security error");
297 ret = vfs_unlink(dir->d_inode, rep, NULL);
300 cachefiles_mark_object_buried(cache, rep);
303 mutex_unlock(&dir->d_inode->i_mutex);
306 cachefiles_io_error(cache, "Unlink failed");
308 _leave(" = %d", ret);
312 /* directories have to be moved to the graveyard */
313 _debug("move stale object to graveyard");
314 mutex_unlock(&dir->d_inode->i_mutex);
317 /* first step is to make up a grave dentry in the graveyard */
318 sprintf(nbuffer, "%08x%08x",
319 (uint32_t) get_seconds(),
320 (uint32_t) atomic_inc_return(&cache->gravecounter));
322 /* do the multiway lock magic */
323 trap = lock_rename(cache->graveyard, dir);
325 /* do some checks before getting the grave dentry */
326 if (rep->d_parent != dir) {
327 /* the entry was probably culled when we dropped the parent dir
329 unlock_rename(cache->graveyard, dir);
330 _leave(" = 0 [culled?]");
334 if (!S_ISDIR(cache->graveyard->d_inode->i_mode)) {
335 unlock_rename(cache->graveyard, dir);
336 cachefiles_io_error(cache, "Graveyard no longer a directory");
341 unlock_rename(cache->graveyard, dir);
342 cachefiles_io_error(cache, "May not make directory loop");
346 if (d_mountpoint(rep)) {
347 unlock_rename(cache->graveyard, dir);
348 cachefiles_io_error(cache, "Mountpoint in cache");
352 grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
354 unlock_rename(cache->graveyard, dir);
356 if (PTR_ERR(grave) == -ENOMEM) {
357 _leave(" = -ENOMEM");
361 cachefiles_io_error(cache, "Lookup error %ld",
366 if (grave->d_inode) {
367 unlock_rename(cache->graveyard, dir);
374 if (d_mountpoint(grave)) {
375 unlock_rename(cache->graveyard, dir);
377 cachefiles_io_error(cache, "Mountpoint in graveyard");
381 /* target should not be an ancestor of source */
383 unlock_rename(cache->graveyard, dir);
385 cachefiles_io_error(cache, "May not make directory loop");
389 /* attempt the rename */
390 path.mnt = cache->mnt;
392 path_to_graveyard.mnt = cache->mnt;
393 path_to_graveyard.dentry = cache->graveyard;
394 ret = security_path_rename(&path, rep, &path_to_graveyard, grave);
396 cachefiles_io_error(cache, "Rename security error %d", ret);
398 ret = vfs_rename(dir->d_inode, rep,
399 cache->graveyard->d_inode, grave, NULL, 0);
400 if (ret != 0 && ret != -ENOMEM)
401 cachefiles_io_error(cache,
402 "Rename failed with error %d", ret);
405 cachefiles_mark_object_buried(cache, rep);
408 unlock_rename(cache->graveyard, dir);
415 * delete an object representation from the cache
417 int cachefiles_delete_object(struct cachefiles_cache *cache,
418 struct cachefiles_object *object)
423 _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
425 ASSERT(object->dentry);
426 ASSERT(object->dentry->d_inode);
427 ASSERT(object->dentry->d_parent);
429 dir = dget_parent(object->dentry);
431 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
433 if (test_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
434 /* object allocation for the same key preemptively deleted this
435 * object's file so that it could create its own file */
436 _debug("object preemptively buried");
437 mutex_unlock(&dir->d_inode->i_mutex);
440 /* we need to check that our parent is _still_ our parent - it
441 * may have been renamed */
442 if (dir == object->dentry->d_parent) {
443 ret = cachefiles_bury_object(cache, dir,
444 object->dentry, false);
446 /* it got moved, presumably by cachefilesd culling it,
447 * so it's no longer in the key path and we can ignore
449 mutex_unlock(&dir->d_inode->i_mutex);
455 _leave(" = %d", ret);
460 * walk from the parent object to the child object through the backing
461 * filesystem, creating directories as we go
463 int cachefiles_walk_to_object(struct cachefiles_object *parent,
464 struct cachefiles_object *object,
466 struct cachefiles_xattr *auxdata)
468 struct cachefiles_cache *cache;
469 struct dentry *dir, *next = NULL;
475 _enter("OBJ%x{%p},OBJ%x,%s,",
476 parent->fscache.debug_id, parent->dentry,
477 object->fscache.debug_id, key);
479 cache = container_of(parent->fscache.cache,
480 struct cachefiles_cache, cache);
481 path.mnt = cache->mnt;
483 ASSERT(parent->dentry);
484 ASSERT(parent->dentry->d_inode);
486 if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
487 // TODO: convert file to dir
488 _leave("looking up in none directory");
492 dir = dget(parent->dentry);
495 /* attempt to transit the first directory component */
499 /* key ends in a double NUL */
500 key = key + nlen + 1;
505 /* search the current directory for the element name */
506 _debug("lookup '%s'", name);
508 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
511 next = lookup_one_len(name, dir, nlen);
512 cachefiles_hist(cachefiles_lookup_histogram, start);
516 _debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
519 object->new = !next->d_inode;
521 /* if this element of the path doesn't exist, then the lookup phase
522 * failed, and we can release any readers in the certain knowledge that
523 * there's nothing for them to actually read */
525 fscache_object_lookup_negative(&object->fscache);
527 /* we need to create the object if it's negative */
528 if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
529 /* index objects and intervening tree levels must be subdirs */
530 if (!next->d_inode) {
531 ret = cachefiles_has_space(cache, 1, 0);
536 ret = security_path_mkdir(&path, next, 0);
540 ret = vfs_mkdir(dir->d_inode, next, 0);
541 cachefiles_hist(cachefiles_mkdir_histogram, start);
545 ASSERT(next->d_inode);
547 _debug("mkdir -> %p{%p{ino=%lu}}",
548 next, next->d_inode, next->d_inode->i_ino);
550 } else if (!S_ISDIR(next->d_inode->i_mode)) {
551 kerror("inode %lu is not a directory",
552 next->d_inode->i_ino);
558 /* non-index objects start out life as files */
559 if (!next->d_inode) {
560 ret = cachefiles_has_space(cache, 1, 0);
565 ret = security_path_mknod(&path, next, S_IFREG, 0);
569 ret = vfs_create(dir->d_inode, next, S_IFREG, true);
570 cachefiles_hist(cachefiles_create_histogram, start);
574 ASSERT(next->d_inode);
576 _debug("create -> %p{%p{ino=%lu}}",
577 next, next->d_inode, next->d_inode->i_ino);
579 } else if (!S_ISDIR(next->d_inode->i_mode) &&
580 !S_ISREG(next->d_inode->i_mode)
582 kerror("inode %lu is not a file or directory",
583 next->d_inode->i_ino);
589 /* process the next component */
592 mutex_unlock(&dir->d_inode->i_mutex);
599 /* we've found the object we were looking for */
600 object->dentry = next;
602 /* if we've found that the terminal object exists, then we need to
603 * check its attributes and delete it if it's out of date */
605 _debug("validate '%*.*s'",
606 next->d_name.len, next->d_name.len, next->d_name.name);
608 ret = cachefiles_check_object_xattr(object, auxdata);
609 if (ret == -ESTALE) {
610 /* delete the object (the deleter drops the directory
612 object->dentry = NULL;
614 ret = cachefiles_bury_object(cache, dir, next, true);
621 _debug("redo lookup");
626 /* note that we're now using this object */
627 ret = cachefiles_mark_object_active(cache, object);
629 mutex_unlock(&dir->d_inode->i_mutex);
633 if (ret == -ETIMEDOUT)
634 goto mark_active_timed_out;
636 _debug("=== OBTAINED_OBJECT ===");
639 /* attach data to a newly constructed terminal object */
640 ret = cachefiles_set_object_xattr(object, auxdata);
644 /* always update the atime on an object we've just looked up
645 * (this is used to keep track of culling, and atimes are only
646 * updated by read, write and readdir but not lookup or
652 /* open a file interface onto a data file */
653 if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
654 if (S_ISREG(object->dentry->d_inode->i_mode)) {
655 const struct address_space_operations *aops;
658 aops = object->dentry->d_inode->i_mapping->a_ops;
662 object->backer = object->dentry;
664 BUG(); // TODO: open file in data-class subdir
669 fscache_obtained_object(&object->fscache);
671 _leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
675 _debug("create error %d", ret);
677 cachefiles_io_error(cache, "Create/mkdir failed");
680 mark_active_timed_out:
681 _debug("mark active timed out");
685 _debug("check error %d", ret);
686 write_lock(&cache->active_lock);
687 rb_erase(&object->active_node, &cache->active_nodes);
688 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
689 wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
690 write_unlock(&cache->active_lock);
692 dput(object->dentry);
693 object->dentry = NULL;
697 _debug("delete error %d", ret);
701 _debug("lookup error %ld", PTR_ERR(next));
704 cachefiles_io_error(cache, "Lookup failed");
707 mutex_unlock(&dir->d_inode->i_mutex);
712 _leave(" = error %d", -ret);
719 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
723 struct dentry *subdir;
728 _enter(",,%s", dirname);
730 /* search the current directory for the element name */
731 mutex_lock(&dir->d_inode->i_mutex);
734 subdir = lookup_one_len(dirname, dir, strlen(dirname));
735 cachefiles_hist(cachefiles_lookup_histogram, start);
736 if (IS_ERR(subdir)) {
737 if (PTR_ERR(subdir) == -ENOMEM)
742 _debug("subdir -> %p %s",
743 subdir, subdir->d_inode ? "positive" : "negative");
745 /* we need to create the subdir if it doesn't exist yet */
746 if (!subdir->d_inode) {
747 ret = cachefiles_has_space(cache, 1, 0);
751 _debug("attempt mkdir");
753 path.mnt = cache->mnt;
755 ret = security_path_mkdir(&path, subdir, 0700);
758 ret = vfs_mkdir(dir->d_inode, subdir, 0700);
762 ASSERT(subdir->d_inode);
764 _debug("mkdir -> %p{%p{ino=%lu}}",
767 subdir->d_inode->i_ino);
770 mutex_unlock(&dir->d_inode->i_mutex);
772 /* we need to make sure the subdir is a directory */
773 ASSERT(subdir->d_inode);
775 if (!S_ISDIR(subdir->d_inode->i_mode)) {
776 kerror("%s is not a directory", dirname);
782 if (!subdir->d_inode->i_op ||
783 !subdir->d_inode->i_op->setxattr ||
784 !subdir->d_inode->i_op->getxattr ||
785 !subdir->d_inode->i_op->lookup ||
786 !subdir->d_inode->i_op->mkdir ||
787 !subdir->d_inode->i_op->create ||
788 !subdir->d_inode->i_op->rename ||
789 !subdir->d_inode->i_op->rmdir ||
790 !subdir->d_inode->i_op->unlink)
793 _leave(" = [%lu]", subdir->d_inode->i_ino);
798 _leave(" = %d [check]", ret);
802 mutex_unlock(&dir->d_inode->i_mutex);
804 kerror("mkdir %s failed with error %d", dirname, ret);
808 mutex_unlock(&dir->d_inode->i_mutex);
809 ret = PTR_ERR(subdir);
810 kerror("Lookup %s failed with error %d", dirname, ret);
814 mutex_unlock(&dir->d_inode->i_mutex);
815 _leave(" = -ENOMEM");
816 return ERR_PTR(-ENOMEM);
820 * find out if an object is in use or not
821 * - if finds object and it's not in use:
822 * - returns a pointer to the object and a reference on it
823 * - returns with the directory locked
825 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
829 struct cachefiles_object *object;
831 struct dentry *victim;
835 //_enter(",%*.*s/,%s",
836 // dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
838 /* look up the victim */
839 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
842 victim = lookup_one_len(filename, dir, strlen(filename));
843 cachefiles_hist(cachefiles_lookup_histogram, start);
847 //_debug("victim -> %p %s",
848 // victim, victim->d_inode ? "positive" : "negative");
850 /* if the object is no longer there then we probably retired the object
851 * at the netfs's request whilst the cull was in progress
853 if (!victim->d_inode) {
854 mutex_unlock(&dir->d_inode->i_mutex);
856 _leave(" = -ENOENT [absent]");
857 return ERR_PTR(-ENOENT);
860 /* check to see if we're using this object */
861 read_lock(&cache->active_lock);
863 _n = cache->active_nodes.rb_node;
866 object = rb_entry(_n, struct cachefiles_object, active_node);
868 if (object->dentry > victim)
870 else if (object->dentry < victim)
876 read_unlock(&cache->active_lock);
878 //_leave(" = %p", victim);
882 read_unlock(&cache->active_lock);
883 mutex_unlock(&dir->d_inode->i_mutex);
885 //_leave(" = -EBUSY [in use]");
886 return ERR_PTR(-EBUSY);
889 mutex_unlock(&dir->d_inode->i_mutex);
890 ret = PTR_ERR(victim);
891 if (ret == -ENOENT) {
892 /* file or dir now absent - probably retired by netfs */
893 _leave(" = -ESTALE [absent]");
894 return ERR_PTR(-ESTALE);
898 cachefiles_io_error(cache, "Lookup failed");
899 } else if (ret != -ENOMEM) {
900 kerror("Internal error: %d", ret);
904 _leave(" = %d", ret);
909 * cull an object if it's not in use
910 * - called only by cache manager daemon
912 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
915 struct dentry *victim;
919 dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
921 victim = cachefiles_check_active(cache, dir, filename);
923 return PTR_ERR(victim);
925 _debug("victim -> %p %s",
926 victim, victim->d_inode ? "positive" : "negative");
928 /* okay... the victim is not being used so we can cull it
929 * - start by marking it as stale
931 _debug("victim is cullable");
933 ret = cachefiles_remove_object_xattr(cache, victim);
937 /* actually remove the victim (drops the dir mutex) */
940 ret = cachefiles_bury_object(cache, dir, victim, false);
949 mutex_unlock(&dir->d_inode->i_mutex);
952 if (ret == -ENOENT) {
953 /* file or dir now absent - probably retired by netfs */
954 _leave(" = -ESTALE [absent]");
958 if (ret != -ENOMEM) {
959 kerror("Internal error: %d", ret);
963 _leave(" = %d", ret);
968 * find out if an object is in use or not
969 * - called only by cache manager daemon
970 * - returns -EBUSY or 0 to indicate whether an object is in use or not
972 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
975 struct dentry *victim;
977 //_enter(",%*.*s/,%s",
978 // dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
980 victim = cachefiles_check_active(cache, dir, filename);
982 return PTR_ERR(victim);
984 mutex_unlock(&dir->d_inode->i_mutex);