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 swfl=%lx ev=%lx[%lx]\n",
41 prefix, fscache_object_states[object->fscache.state],
42 object->fscache.flags, object->fscache.work.flags,
43 object->fscache.events,
44 object->fscache.event_mask & FSCACHE_OBJECT_EVENTS_MASK);
45 printk(KERN_ERR "%sops=%u inp=%u exc=%u\n",
46 prefix, object->fscache.n_ops, object->fscache.n_in_progress,
47 object->fscache.n_exclusive);
48 printk(KERN_ERR "%sparent=%p\n",
49 prefix, object->fscache.parent);
51 spin_lock(&object->fscache.lock);
52 cookie = object->fscache.cookie;
54 printk(KERN_ERR "%scookie=%p [pr=%p nd=%p fl=%lx]\n",
56 object->fscache.cookie,
57 object->fscache.cookie->parent,
58 object->fscache.cookie->netfs_data,
59 object->fscache.cookie->flags);
61 keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
62 CACHEFILES_KEYBUF_SIZE);
66 printk(KERN_ERR "%scookie=NULL\n", prefix);
69 spin_unlock(&object->fscache.lock);
72 printk(KERN_ERR "%skey=[%u] '", prefix, keylen);
73 for (loop = 0; loop < keylen; loop++)
74 printk("%02x", keybuf[loop]);
80 * dump debugging info about a pair of objects
82 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
83 struct cachefiles_object *xobject)
87 keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
89 __cachefiles_printk_object(object, "", keybuf);
91 __cachefiles_printk_object(xobject, "x", keybuf);
96 * record the fact that an object is now active
98 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
99 struct cachefiles_object *object)
101 struct cachefiles_object *xobject;
102 struct rb_node **_p, *_parent = NULL;
103 struct dentry *dentry;
105 _enter(",%p", object);
108 write_lock(&cache->active_lock);
110 if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
111 printk(KERN_ERR "CacheFiles: Error: Object already active\n");
112 cachefiles_printk_object(object, NULL);
116 dentry = object->dentry;
117 _p = &cache->active_nodes.rb_node;
120 xobject = rb_entry(_parent,
121 struct cachefiles_object, active_node);
123 ASSERT(xobject != object);
125 if (xobject->dentry > dentry)
126 _p = &(*_p)->rb_left;
127 else if (xobject->dentry < dentry)
128 _p = &(*_p)->rb_right;
130 goto wait_for_old_object;
133 rb_link_node(&object->active_node, _parent, _p);
134 rb_insert_color(&object->active_node, &cache->active_nodes);
136 write_unlock(&cache->active_lock);
140 /* an old object from a previous incarnation is hogging the slot - we
141 * need to wait for it to be destroyed */
143 if (xobject->fscache.state < FSCACHE_OBJECT_DYING) {
144 printk(KERN_ERR "\n");
145 printk(KERN_ERR "CacheFiles: Error:"
146 " Unexpected object collision\n");
147 cachefiles_printk_object(object, xobject);
150 atomic_inc(&xobject->usage);
151 write_unlock(&cache->active_lock);
153 if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
154 wait_queue_head_t *wq;
156 signed long timeout = 60 * HZ;
160 /* if the object we're waiting for is queued for processing,
161 * then just put ourselves on the queue behind it */
162 if (slow_work_is_queued(&xobject->fscache.work)) {
163 _debug("queue OBJ%x behind OBJ%x immediately",
164 object->fscache.debug_id,
165 xobject->fscache.debug_id);
169 /* otherwise we sleep until either the object we're waiting for
170 * is done, or the slow-work facility wants the thread back to
172 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
176 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
177 if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
179 requeue = slow_work_sleep_till_thread_needed(
180 &object->fscache.work, &timeout);
181 } while (timeout > 0 && !requeue);
182 finish_wait(wq, &wait);
185 test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
186 _debug("queue OBJ%x behind OBJ%x after wait",
187 object->fscache.debug_id,
188 xobject->fscache.debug_id);
193 printk(KERN_ERR "\n");
194 printk(KERN_ERR "CacheFiles: Error: Overlong"
195 " wait for old active object to go away\n");
196 cachefiles_printk_object(object, xobject);
201 ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
203 cache->cache.ops->put_object(&xobject->fscache);
207 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
208 cache->cache.ops->put_object(&xobject->fscache);
209 _leave(" = -ETIMEDOUT");
214 * delete an object representation from the cache
215 * - file backed objects are unlinked
216 * - directory backed objects are stuffed into the graveyard for userspace to
218 * - unlocks the directory mutex
220 static int cachefiles_bury_object(struct cachefiles_cache *cache,
224 struct dentry *grave, *trap;
225 char nbuffer[8 + 8 + 1];
228 _enter(",'%*.*s','%*.*s'",
229 dir->d_name.len, dir->d_name.len, dir->d_name.name,
230 rep->d_name.len, rep->d_name.len, rep->d_name.name);
232 /* non-directories can just be unlinked */
233 if (!S_ISDIR(rep->d_inode->i_mode)) {
234 _debug("unlink stale object");
235 ret = vfs_unlink(dir->d_inode, rep);
237 mutex_unlock(&dir->d_inode->i_mutex);
240 cachefiles_io_error(cache, "Unlink failed");
242 _leave(" = %d", ret);
246 /* directories have to be moved to the graveyard */
247 _debug("move stale object to graveyard");
248 mutex_unlock(&dir->d_inode->i_mutex);
251 /* first step is to make up a grave dentry in the graveyard */
252 sprintf(nbuffer, "%08x%08x",
253 (uint32_t) get_seconds(),
254 (uint32_t) atomic_inc_return(&cache->gravecounter));
256 /* do the multiway lock magic */
257 trap = lock_rename(cache->graveyard, dir);
259 /* do some checks before getting the grave dentry */
260 if (rep->d_parent != dir) {
261 /* the entry was probably culled when we dropped the parent dir
263 unlock_rename(cache->graveyard, dir);
264 _leave(" = 0 [culled?]");
268 if (!S_ISDIR(cache->graveyard->d_inode->i_mode)) {
269 unlock_rename(cache->graveyard, dir);
270 cachefiles_io_error(cache, "Graveyard no longer a directory");
275 unlock_rename(cache->graveyard, dir);
276 cachefiles_io_error(cache, "May not make directory loop");
280 if (d_mountpoint(rep)) {
281 unlock_rename(cache->graveyard, dir);
282 cachefiles_io_error(cache, "Mountpoint in cache");
286 grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
288 unlock_rename(cache->graveyard, dir);
290 if (PTR_ERR(grave) == -ENOMEM) {
291 _leave(" = -ENOMEM");
295 cachefiles_io_error(cache, "Lookup error %ld",
300 if (grave->d_inode) {
301 unlock_rename(cache->graveyard, dir);
308 if (d_mountpoint(grave)) {
309 unlock_rename(cache->graveyard, dir);
311 cachefiles_io_error(cache, "Mountpoint in graveyard");
315 /* target should not be an ancestor of source */
317 unlock_rename(cache->graveyard, dir);
319 cachefiles_io_error(cache, "May not make directory loop");
323 /* attempt the rename */
324 ret = vfs_rename(dir->d_inode, rep, cache->graveyard->d_inode, grave);
325 if (ret != 0 && ret != -ENOMEM)
326 cachefiles_io_error(cache, "Rename failed with error %d", ret);
328 unlock_rename(cache->graveyard, dir);
335 * delete an object representation from the cache
337 int cachefiles_delete_object(struct cachefiles_cache *cache,
338 struct cachefiles_object *object)
343 _enter(",{%p}", object->dentry);
345 ASSERT(object->dentry);
346 ASSERT(object->dentry->d_inode);
347 ASSERT(object->dentry->d_parent);
349 dir = dget_parent(object->dentry);
351 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
353 /* we need to check that our parent is _still_ our parent - it may have
355 if (dir == object->dentry->d_parent) {
356 ret = cachefiles_bury_object(cache, dir, object->dentry);
358 /* it got moved, presumably by cachefilesd culling it, so it's
359 * no longer in the key path and we can ignore it */
360 mutex_unlock(&dir->d_inode->i_mutex);
365 _leave(" = %d", ret);
370 * walk from the parent object to the child object through the backing
371 * filesystem, creating directories as we go
373 int cachefiles_walk_to_object(struct cachefiles_object *parent,
374 struct cachefiles_object *object,
376 struct cachefiles_xattr *auxdata)
378 struct cachefiles_cache *cache;
379 struct dentry *dir, *next = NULL;
384 _enter("{%p},,%s,", parent->dentry, key);
386 cache = container_of(parent->fscache.cache,
387 struct cachefiles_cache, cache);
389 ASSERT(parent->dentry);
390 ASSERT(parent->dentry->d_inode);
392 if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
393 // TODO: convert file to dir
394 _leave("looking up in none directory");
398 dir = dget(parent->dentry);
401 /* attempt to transit the first directory component */
405 /* key ends in a double NUL */
406 key = key + nlen + 1;
411 /* search the current directory for the element name */
412 _debug("lookup '%s'", name);
414 mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
417 next = lookup_one_len(name, dir, nlen);
418 cachefiles_hist(cachefiles_lookup_histogram, start);
422 _debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
425 object->new = !next->d_inode;
427 /* if this element of the path doesn't exist, then the lookup phase
428 * failed, and we can release any readers in the certain knowledge that
429 * there's nothing for them to actually read */
431 fscache_object_lookup_negative(&object->fscache);
433 /* we need to create the object if it's negative */
434 if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
435 /* index objects and intervening tree levels must be subdirs */
436 if (!next->d_inode) {
437 ret = cachefiles_has_space(cache, 1, 0);
442 ret = vfs_mkdir(dir->d_inode, next, 0);
443 cachefiles_hist(cachefiles_mkdir_histogram, start);
447 ASSERT(next->d_inode);
449 _debug("mkdir -> %p{%p{ino=%lu}}",
450 next, next->d_inode, next->d_inode->i_ino);
452 } else if (!S_ISDIR(next->d_inode->i_mode)) {
453 kerror("inode %lu is not a directory",
454 next->d_inode->i_ino);
460 /* non-index objects start out life as files */
461 if (!next->d_inode) {
462 ret = cachefiles_has_space(cache, 1, 0);
467 ret = vfs_create(dir->d_inode, next, S_IFREG, NULL);
468 cachefiles_hist(cachefiles_create_histogram, start);
472 ASSERT(next->d_inode);
474 _debug("create -> %p{%p{ino=%lu}}",
475 next, next->d_inode, next->d_inode->i_ino);
477 } else if (!S_ISDIR(next->d_inode->i_mode) &&
478 !S_ISREG(next->d_inode->i_mode)
480 kerror("inode %lu is not a file or directory",
481 next->d_inode->i_ino);
487 /* process the next component */
490 mutex_unlock(&dir->d_inode->i_mutex);
497 /* we've found the object we were looking for */
498 object->dentry = next;
500 /* if we've found that the terminal object exists, then we need to
501 * check its attributes and delete it if it's out of date */
503 _debug("validate '%*.*s'",
504 next->d_name.len, next->d_name.len, next->d_name.name);
506 ret = cachefiles_check_object_xattr(object, auxdata);
507 if (ret == -ESTALE) {
508 /* delete the object (the deleter drops the directory
510 object->dentry = NULL;
512 ret = cachefiles_bury_object(cache, dir, next);
519 _debug("redo lookup");
524 /* note that we're now using this object */
525 ret = cachefiles_mark_object_active(cache, object);
527 mutex_unlock(&dir->d_inode->i_mutex);
531 if (ret == -ETIMEDOUT)
532 goto mark_active_timed_out;
534 _debug("=== OBTAINED_OBJECT ===");
537 /* attach data to a newly constructed terminal object */
538 ret = cachefiles_set_object_xattr(object, auxdata);
542 /* always update the atime on an object we've just looked up
543 * (this is used to keep track of culling, and atimes are only
544 * updated by read, write and readdir but not lookup or
546 touch_atime(cache->mnt, next);
549 /* open a file interface onto a data file */
550 if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
551 if (S_ISREG(object->dentry->d_inode->i_mode)) {
552 const struct address_space_operations *aops;
555 aops = object->dentry->d_inode->i_mapping->a_ops;
559 object->backer = object->dentry;
561 BUG(); // TODO: open file in data-class subdir
566 fscache_obtained_object(&object->fscache);
568 _leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
572 _debug("create error %d", ret);
574 cachefiles_io_error(cache, "Create/mkdir failed");
577 mark_active_timed_out:
578 _debug("mark active timed out");
582 _debug("check error %d", ret);
583 write_lock(&cache->active_lock);
584 rb_erase(&object->active_node, &cache->active_nodes);
585 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
586 wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
587 write_unlock(&cache->active_lock);
589 dput(object->dentry);
590 object->dentry = NULL;
594 _debug("delete error %d", ret);
598 _debug("lookup error %ld", PTR_ERR(next));
601 cachefiles_io_error(cache, "Lookup failed");
604 mutex_unlock(&dir->d_inode->i_mutex);
609 _leave(" = error %d", -ret);
616 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
620 struct dentry *subdir;
624 _enter(",,%s", dirname);
626 /* search the current directory for the element name */
627 mutex_lock(&dir->d_inode->i_mutex);
630 subdir = lookup_one_len(dirname, dir, strlen(dirname));
631 cachefiles_hist(cachefiles_lookup_histogram, start);
632 if (IS_ERR(subdir)) {
633 if (PTR_ERR(subdir) == -ENOMEM)
638 _debug("subdir -> %p %s",
639 subdir, subdir->d_inode ? "positive" : "negative");
641 /* we need to create the subdir if it doesn't exist yet */
642 if (!subdir->d_inode) {
643 ret = cachefiles_has_space(cache, 1, 0);
647 _debug("attempt mkdir");
649 ret = vfs_mkdir(dir->d_inode, subdir, 0700);
653 ASSERT(subdir->d_inode);
655 _debug("mkdir -> %p{%p{ino=%lu}}",
658 subdir->d_inode->i_ino);
661 mutex_unlock(&dir->d_inode->i_mutex);
663 /* we need to make sure the subdir is a directory */
664 ASSERT(subdir->d_inode);
666 if (!S_ISDIR(subdir->d_inode->i_mode)) {
667 kerror("%s is not a directory", dirname);
673 if (!subdir->d_inode->i_op ||
674 !subdir->d_inode->i_op->setxattr ||
675 !subdir->d_inode->i_op->getxattr ||
676 !subdir->d_inode->i_op->lookup ||
677 !subdir->d_inode->i_op->mkdir ||
678 !subdir->d_inode->i_op->create ||
679 !subdir->d_inode->i_op->rename ||
680 !subdir->d_inode->i_op->rmdir ||
681 !subdir->d_inode->i_op->unlink)
684 _leave(" = [%lu]", subdir->d_inode->i_ino);
689 _leave(" = %d [check]", ret);
693 mutex_unlock(&dir->d_inode->i_mutex);
695 kerror("mkdir %s failed with error %d", dirname, ret);
699 mutex_unlock(&dir->d_inode->i_mutex);
700 ret = PTR_ERR(subdir);
701 kerror("Lookup %s failed with error %d", dirname, ret);
705 mutex_unlock(&dir->d_inode->i_mutex);
706 _leave(" = -ENOMEM");
707 return ERR_PTR(-ENOMEM);
711 * find out if an object is in use or not
712 * - if finds object and it's not in use:
713 * - returns a pointer to the object and a reference on it
714 * - returns with the directory locked
716 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
720 struct cachefiles_object *object;
722 struct dentry *victim;
726 //_enter(",%*.*s/,%s",
727 // dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
729 /* look up the victim */
730 mutex_lock_nested(&dir->d_inode->i_mutex, 1);
733 victim = lookup_one_len(filename, dir, strlen(filename));
734 cachefiles_hist(cachefiles_lookup_histogram, start);
738 //_debug("victim -> %p %s",
739 // victim, victim->d_inode ? "positive" : "negative");
741 /* if the object is no longer there then we probably retired the object
742 * at the netfs's request whilst the cull was in progress
744 if (!victim->d_inode) {
745 mutex_unlock(&dir->d_inode->i_mutex);
747 _leave(" = -ENOENT [absent]");
748 return ERR_PTR(-ENOENT);
751 /* check to see if we're using this object */
752 read_lock(&cache->active_lock);
754 _n = cache->active_nodes.rb_node;
757 object = rb_entry(_n, struct cachefiles_object, active_node);
759 if (object->dentry > victim)
761 else if (object->dentry < victim)
767 read_unlock(&cache->active_lock);
769 //_leave(" = %p", victim);
773 read_unlock(&cache->active_lock);
774 mutex_unlock(&dir->d_inode->i_mutex);
776 //_leave(" = -EBUSY [in use]");
777 return ERR_PTR(-EBUSY);
780 mutex_unlock(&dir->d_inode->i_mutex);
781 ret = PTR_ERR(victim);
782 if (ret == -ENOENT) {
783 /* file or dir now absent - probably retired by netfs */
784 _leave(" = -ESTALE [absent]");
785 return ERR_PTR(-ESTALE);
789 cachefiles_io_error(cache, "Lookup failed");
790 } else if (ret != -ENOMEM) {
791 kerror("Internal error: %d", ret);
795 _leave(" = %d", ret);
800 * cull an object if it's not in use
801 * - called only by cache manager daemon
803 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
806 struct dentry *victim;
810 dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
812 victim = cachefiles_check_active(cache, dir, filename);
814 return PTR_ERR(victim);
816 _debug("victim -> %p %s",
817 victim, victim->d_inode ? "positive" : "negative");
819 /* okay... the victim is not being used so we can cull it
820 * - start by marking it as stale
822 _debug("victim is cullable");
824 ret = cachefiles_remove_object_xattr(cache, victim);
828 /* actually remove the victim (drops the dir mutex) */
831 ret = cachefiles_bury_object(cache, dir, victim);
840 mutex_unlock(&dir->d_inode->i_mutex);
843 if (ret == -ENOENT) {
844 /* file or dir now absent - probably retired by netfs */
845 _leave(" = -ESTALE [absent]");
849 if (ret != -ENOMEM) {
850 kerror("Internal error: %d", ret);
854 _leave(" = %d", ret);
859 * find out if an object is in use or not
860 * - called only by cache manager daemon
861 * - returns -EBUSY or 0 to indicate whether an object is in use or not
863 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
866 struct dentry *victim;
868 //_enter(",%*.*s/,%s",
869 // dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
871 victim = cachefiles_check_active(cache, dir, filename);
873 return PTR_ERR(victim);
875 mutex_unlock(&dir->d_inode->i_mutex);