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 * delete an object representation from the cache
262 * - file backed objects are unlinked
263 * - directory backed objects are stuffed into the graveyard for userspace to
265 * - unlocks the directory mutex
267 static int cachefiles_bury_object(struct cachefiles_cache *cache,
271 enum fscache_why_object_killed why)
273 struct dentry *grave, *trap;
274 struct path path, path_to_graveyard;
275 char nbuffer[8 + 8 + 1];
278 _enter(",'%pd','%pd'", dir, rep);
280 _debug("remove %p from %p", rep, dir);
282 /* non-directories can just be unlinked */
283 if (!d_is_dir(rep)) {
284 _debug("unlink stale object");
286 path.mnt = cache->mnt;
288 ret = security_path_unlink(&path, rep);
290 cachefiles_io_error(cache, "Unlink security error");
292 ret = vfs_unlink(d_inode(dir), rep, NULL);
295 cachefiles_mark_object_buried(cache, rep, why);
298 mutex_unlock(&d_inode(dir)->i_mutex);
301 cachefiles_io_error(cache, "Unlink failed");
303 _leave(" = %d", ret);
307 /* directories have to be moved to the graveyard */
308 _debug("move stale object to graveyard");
309 mutex_unlock(&d_inode(dir)->i_mutex);
312 /* first step is to make up a grave dentry in the graveyard */
313 sprintf(nbuffer, "%08x%08x",
314 (uint32_t) get_seconds(),
315 (uint32_t) atomic_inc_return(&cache->gravecounter));
317 /* do the multiway lock magic */
318 trap = lock_rename(cache->graveyard, dir);
320 /* do some checks before getting the grave dentry */
321 if (rep->d_parent != dir) {
322 /* the entry was probably culled when we dropped the parent dir
324 unlock_rename(cache->graveyard, dir);
325 _leave(" = 0 [culled?]");
329 if (!d_can_lookup(cache->graveyard)) {
330 unlock_rename(cache->graveyard, dir);
331 cachefiles_io_error(cache, "Graveyard no longer a directory");
336 unlock_rename(cache->graveyard, dir);
337 cachefiles_io_error(cache, "May not make directory loop");
341 if (d_mountpoint(rep)) {
342 unlock_rename(cache->graveyard, dir);
343 cachefiles_io_error(cache, "Mountpoint in cache");
347 grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
349 unlock_rename(cache->graveyard, dir);
351 if (PTR_ERR(grave) == -ENOMEM) {
352 _leave(" = -ENOMEM");
356 cachefiles_io_error(cache, "Lookup error %ld",
361 if (d_is_positive(grave)) {
362 unlock_rename(cache->graveyard, dir);
369 if (d_mountpoint(grave)) {
370 unlock_rename(cache->graveyard, dir);
372 cachefiles_io_error(cache, "Mountpoint in graveyard");
376 /* target should not be an ancestor of source */
378 unlock_rename(cache->graveyard, dir);
380 cachefiles_io_error(cache, "May not make directory loop");
384 /* attempt the rename */
385 path.mnt = cache->mnt;
387 path_to_graveyard.mnt = cache->mnt;
388 path_to_graveyard.dentry = cache->graveyard;
389 ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
391 cachefiles_io_error(cache, "Rename security error %d", ret);
393 ret = vfs_rename(d_inode(dir), rep,
394 d_inode(cache->graveyard), grave, NULL, 0);
395 if (ret != 0 && ret != -ENOMEM)
396 cachefiles_io_error(cache,
397 "Rename failed with error %d", ret);
400 cachefiles_mark_object_buried(cache, rep, why);
403 unlock_rename(cache->graveyard, dir);
410 * delete an object representation from the cache
412 int cachefiles_delete_object(struct cachefiles_cache *cache,
413 struct cachefiles_object *object)
418 _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
420 ASSERT(object->dentry);
421 ASSERT(d_backing_inode(object->dentry));
422 ASSERT(object->dentry->d_parent);
424 dir = dget_parent(object->dentry);
426 mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
428 if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
429 /* object allocation for the same key preemptively deleted this
430 * object's file so that it could create its own file */
431 _debug("object preemptively buried");
432 mutex_unlock(&d_inode(dir)->i_mutex);
435 /* we need to check that our parent is _still_ our parent - it
436 * may have been renamed */
437 if (dir == object->dentry->d_parent) {
438 ret = cachefiles_bury_object(cache, dir,
439 object->dentry, false,
440 FSCACHE_OBJECT_WAS_RETIRED);
442 /* it got moved, presumably by cachefilesd culling it,
443 * so it's no longer in the key path and we can ignore
445 mutex_unlock(&d_inode(dir)->i_mutex);
451 _leave(" = %d", ret);
456 * walk from the parent object to the child object through the backing
457 * filesystem, creating directories as we go
459 int cachefiles_walk_to_object(struct cachefiles_object *parent,
460 struct cachefiles_object *object,
462 struct cachefiles_xattr *auxdata)
464 struct cachefiles_cache *cache;
465 struct dentry *dir, *next = NULL;
471 _enter("OBJ%x{%p},OBJ%x,%s,",
472 parent->fscache.debug_id, parent->dentry,
473 object->fscache.debug_id, key);
475 cache = container_of(parent->fscache.cache,
476 struct cachefiles_cache, cache);
477 path.mnt = cache->mnt;
479 ASSERT(parent->dentry);
480 ASSERT(d_backing_inode(parent->dentry));
482 if (!(d_is_dir(parent->dentry))) {
483 // TODO: convert file to dir
484 _leave("looking up in none directory");
488 dir = dget(parent->dentry);
491 /* attempt to transit the first directory component */
495 /* key ends in a double NUL */
496 key = key + nlen + 1;
501 /* search the current directory for the element name */
502 _debug("lookup '%s'", name);
504 mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
507 next = lookup_one_len(name, dir, nlen);
508 cachefiles_hist(cachefiles_lookup_histogram, start);
512 _debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
515 object->new = !d_backing_inode(next);
517 /* if this element of the path doesn't exist, then the lookup phase
518 * failed, and we can release any readers in the certain knowledge that
519 * there's nothing for them to actually read */
520 if (d_is_negative(next))
521 fscache_object_lookup_negative(&object->fscache);
523 /* we need to create the object if it's negative */
524 if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
525 /* index objects and intervening tree levels must be subdirs */
526 if (d_is_negative(next)) {
527 ret = cachefiles_has_space(cache, 1, 0);
532 ret = security_path_mkdir(&path, next, 0);
536 ret = vfs_mkdir(d_inode(dir), next, 0);
537 cachefiles_hist(cachefiles_mkdir_histogram, start);
541 ASSERT(d_backing_inode(next));
543 _debug("mkdir -> %p{%p{ino=%lu}}",
544 next, d_backing_inode(next), d_backing_inode(next)->i_ino);
546 } else if (!d_can_lookup(next)) {
547 pr_err("inode %lu is not a directory\n",
548 d_backing_inode(next)->i_ino);
554 /* non-index objects start out life as files */
555 if (d_is_negative(next)) {
556 ret = cachefiles_has_space(cache, 1, 0);
561 ret = security_path_mknod(&path, next, S_IFREG, 0);
565 ret = vfs_create(d_inode(dir), next, S_IFREG, true);
566 cachefiles_hist(cachefiles_create_histogram, start);
570 ASSERT(d_backing_inode(next));
572 _debug("create -> %p{%p{ino=%lu}}",
573 next, d_backing_inode(next), d_backing_inode(next)->i_ino);
575 } else if (!d_can_lookup(next) &&
578 pr_err("inode %lu is not a file or directory\n",
579 d_backing_inode(next)->i_ino);
585 /* process the next component */
588 mutex_unlock(&d_inode(dir)->i_mutex);
595 /* we've found the object we were looking for */
596 object->dentry = next;
598 /* if we've found that the terminal object exists, then we need to
599 * check its attributes and delete it if it's out of date */
601 _debug("validate '%pd'", next);
603 ret = cachefiles_check_object_xattr(object, auxdata);
604 if (ret == -ESTALE) {
605 /* delete the object (the deleter drops the directory
607 object->dentry = NULL;
609 ret = cachefiles_bury_object(cache, dir, next, true,
610 FSCACHE_OBJECT_IS_STALE);
617 _debug("redo lookup");
618 fscache_object_retrying_stale(&object->fscache);
623 /* note that we're now using this object */
624 ret = cachefiles_mark_object_active(cache, object);
626 mutex_unlock(&d_inode(dir)->i_mutex);
630 if (ret == -ETIMEDOUT)
631 goto mark_active_timed_out;
633 _debug("=== OBTAINED_OBJECT ===");
636 /* attach data to a newly constructed terminal object */
637 ret = cachefiles_set_object_xattr(object, auxdata);
641 /* always update the atime on an object we've just looked up
642 * (this is used to keep track of culling, and atimes are only
643 * updated by read, write and readdir but not lookup or
649 /* open a file interface onto a data file */
650 if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
651 if (d_is_reg(object->dentry)) {
652 const struct address_space_operations *aops;
655 aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
658 if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
661 object->backer = object->dentry;
663 BUG(); // TODO: open file in data-class subdir
668 fscache_obtained_object(&object->fscache);
670 _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
674 fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
676 _debug("create error %d", ret);
678 cachefiles_io_error(cache, "Create/mkdir failed");
681 mark_active_timed_out:
682 _debug("mark active timed out");
686 _debug("check error %d", ret);
687 write_lock(&cache->active_lock);
688 rb_erase(&object->active_node, &cache->active_nodes);
689 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
690 wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
691 write_unlock(&cache->active_lock);
693 dput(object->dentry);
694 object->dentry = NULL;
698 _debug("delete error %d", ret);
702 _debug("lookup error %ld", PTR_ERR(next));
705 cachefiles_io_error(cache, "Lookup failed");
708 mutex_unlock(&d_inode(dir)->i_mutex);
713 _leave(" = error %d", -ret);
720 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
724 struct dentry *subdir;
729 _enter(",,%s", dirname);
731 /* search the current directory for the element name */
732 mutex_lock(&d_inode(dir)->i_mutex);
735 subdir = lookup_one_len(dirname, dir, strlen(dirname));
736 cachefiles_hist(cachefiles_lookup_histogram, start);
737 if (IS_ERR(subdir)) {
738 if (PTR_ERR(subdir) == -ENOMEM)
743 _debug("subdir -> %p %s",
744 subdir, d_backing_inode(subdir) ? "positive" : "negative");
746 /* we need to create the subdir if it doesn't exist yet */
747 if (d_is_negative(subdir)) {
748 ret = cachefiles_has_space(cache, 1, 0);
752 _debug("attempt mkdir");
754 path.mnt = cache->mnt;
756 ret = security_path_mkdir(&path, subdir, 0700);
759 ret = vfs_mkdir(d_inode(dir), subdir, 0700);
763 ASSERT(d_backing_inode(subdir));
765 _debug("mkdir -> %p{%p{ino=%lu}}",
767 d_backing_inode(subdir),
768 d_backing_inode(subdir)->i_ino);
771 mutex_unlock(&d_inode(dir)->i_mutex);
773 /* we need to make sure the subdir is a directory */
774 ASSERT(d_backing_inode(subdir));
776 if (!d_can_lookup(subdir)) {
777 pr_err("%s is not a directory\n", dirname);
783 if (!d_backing_inode(subdir)->i_op->setxattr ||
784 !d_backing_inode(subdir)->i_op->getxattr ||
785 !d_backing_inode(subdir)->i_op->lookup ||
786 !d_backing_inode(subdir)->i_op->mkdir ||
787 !d_backing_inode(subdir)->i_op->create ||
788 (!d_backing_inode(subdir)->i_op->rename &&
789 !d_backing_inode(subdir)->i_op->rename2) ||
790 !d_backing_inode(subdir)->i_op->rmdir ||
791 !d_backing_inode(subdir)->i_op->unlink)
794 _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
799 _leave(" = %d [check]", ret);
803 mutex_unlock(&d_inode(dir)->i_mutex);
805 pr_err("mkdir %s failed with error %d\n", dirname, ret);
809 mutex_unlock(&d_inode(dir)->i_mutex);
810 ret = PTR_ERR(subdir);
811 pr_err("Lookup %s failed with error %d\n", dirname, ret);
815 mutex_unlock(&d_inode(dir)->i_mutex);
816 _leave(" = -ENOMEM");
817 return ERR_PTR(-ENOMEM);
821 * find out if an object is in use or not
822 * - if finds object and it's not in use:
823 * - returns a pointer to the object and a reference on it
824 * - returns with the directory locked
826 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
830 struct cachefiles_object *object;
832 struct dentry *victim;
839 /* look up the victim */
840 mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
843 victim = lookup_one_len(filename, dir, strlen(filename));
844 cachefiles_hist(cachefiles_lookup_histogram, start);
848 //_debug("victim -> %p %s",
849 // victim, d_backing_inode(victim) ? "positive" : "negative");
851 /* if the object is no longer there then we probably retired the object
852 * at the netfs's request whilst the cull was in progress
854 if (d_is_negative(victim)) {
855 mutex_unlock(&d_inode(dir)->i_mutex);
857 _leave(" = -ENOENT [absent]");
858 return ERR_PTR(-ENOENT);
861 /* check to see if we're using this object */
862 read_lock(&cache->active_lock);
864 _n = cache->active_nodes.rb_node;
867 object = rb_entry(_n, struct cachefiles_object, active_node);
869 if (object->dentry > victim)
871 else if (object->dentry < victim)
877 read_unlock(&cache->active_lock);
879 //_leave(" = %p", victim);
883 read_unlock(&cache->active_lock);
884 mutex_unlock(&d_inode(dir)->i_mutex);
886 //_leave(" = -EBUSY [in use]");
887 return ERR_PTR(-EBUSY);
890 mutex_unlock(&d_inode(dir)->i_mutex);
891 ret = PTR_ERR(victim);
892 if (ret == -ENOENT) {
893 /* file or dir now absent - probably retired by netfs */
894 _leave(" = -ESTALE [absent]");
895 return ERR_PTR(-ESTALE);
899 cachefiles_io_error(cache, "Lookup failed");
900 } else if (ret != -ENOMEM) {
901 pr_err("Internal error: %d\n", ret);
905 _leave(" = %d", ret);
910 * cull an object if it's not in use
911 * - called only by cache manager daemon
913 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
916 struct dentry *victim;
919 _enter(",%pd/,%s", dir, filename);
921 victim = cachefiles_check_active(cache, dir, filename);
923 return PTR_ERR(victim);
925 _debug("victim -> %p %s",
926 victim, d_backing_inode(victim) ? "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,
941 FSCACHE_OBJECT_WAS_CULLED);
950 mutex_unlock(&d_inode(dir)->i_mutex);
953 if (ret == -ENOENT) {
954 /* file or dir now absent - probably retired by netfs */
955 _leave(" = -ESTALE [absent]");
959 if (ret != -ENOMEM) {
960 pr_err("Internal error: %d\n", ret);
964 _leave(" = %d", ret);
969 * find out if an object is in use or not
970 * - called only by cache manager daemon
971 * - returns -EBUSY or 0 to indicate whether an object is in use or not
973 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
976 struct dentry *victim;
981 victim = cachefiles_check_active(cache, dir, filename);
983 return PTR_ERR(victim);
985 mutex_unlock(&d_inode(dir)->i_mutex);