xref: /openbmc/linux/fs/cachefiles/namei.c (revision fee096de)
1 /* CacheFiles path walking and related routines
2  *
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
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.
10  */
11 
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/file.h>
15 #include <linux/fs.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 "internal.h"
23 
24 #define CACHEFILES_KEYBUF_SIZE 512
25 
26 /*
27  * dump debugging info about an object
28  */
29 static noinline
30 void __cachefiles_printk_object(struct cachefiles_object *object,
31 				const char *prefix,
32 				u8 *keybuf)
33 {
34 	struct fscache_cookie *cookie;
35 	unsigned keylen, loop;
36 
37 	printk(KERN_ERR "%sobject: OBJ%x\n",
38 	       prefix, object->fscache.debug_id);
39 	printk(KERN_ERR "%sobjstate=%s fl=%lx swfl=%lx ev=%lx[%lx]\n",
40 	       prefix, fscache_object_states[object->fscache.state],
41 	       object->fscache.flags, object->fscache.work.flags,
42 	       object->fscache.events,
43 	       object->fscache.event_mask & FSCACHE_OBJECT_EVENTS_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);
49 
50 	spin_lock(&object->fscache.lock);
51 	cookie = object->fscache.cookie;
52 	if (cookie) {
53 		printk(KERN_ERR "%scookie=%p [pr=%p nd=%p fl=%lx]\n",
54 		       prefix,
55 		       object->fscache.cookie,
56 		       object->fscache.cookie->parent,
57 		       object->fscache.cookie->netfs_data,
58 		       object->fscache.cookie->flags);
59 		if (keybuf)
60 			keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
61 						      CACHEFILES_KEYBUF_SIZE);
62 		else
63 			keylen = 0;
64 	} else {
65 		printk(KERN_ERR "%scookie=NULL\n", prefix);
66 		keylen = 0;
67 	}
68 	spin_unlock(&object->fscache.lock);
69 
70 	if (keylen) {
71 		printk(KERN_ERR "%skey=[%u] '", prefix, keylen);
72 		for (loop = 0; loop < keylen; loop++)
73 			printk("%02x", keybuf[loop]);
74 		printk("'\n");
75 	}
76 }
77 
78 /*
79  * dump debugging info about a pair of objects
80  */
81 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
82 					      struct cachefiles_object *xobject)
83 {
84 	u8 *keybuf;
85 
86 	keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
87 	if (object)
88 		__cachefiles_printk_object(object, "", keybuf);
89 	if (xobject)
90 		__cachefiles_printk_object(xobject, "x", keybuf);
91 	kfree(keybuf);
92 }
93 
94 /*
95  * record the fact that an object is now active
96  */
97 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
98 					 struct cachefiles_object *object)
99 {
100 	struct cachefiles_object *xobject;
101 	struct rb_node **_p, *_parent = NULL;
102 	struct dentry *dentry;
103 
104 	_enter(",%p", object);
105 
106 try_again:
107 	write_lock(&cache->active_lock);
108 
109 	if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
110 		printk(KERN_ERR "CacheFiles: Error: Object already active\n");
111 		cachefiles_printk_object(object, NULL);
112 		BUG();
113 	}
114 
115 	dentry = object->dentry;
116 	_p = &cache->active_nodes.rb_node;
117 	while (*_p) {
118 		_parent = *_p;
119 		xobject = rb_entry(_parent,
120 				   struct cachefiles_object, active_node);
121 
122 		ASSERT(xobject != object);
123 
124 		if (xobject->dentry > dentry)
125 			_p = &(*_p)->rb_left;
126 		else if (xobject->dentry < dentry)
127 			_p = &(*_p)->rb_right;
128 		else
129 			goto wait_for_old_object;
130 	}
131 
132 	rb_link_node(&object->active_node, _parent, _p);
133 	rb_insert_color(&object->active_node, &cache->active_nodes);
134 
135 	write_unlock(&cache->active_lock);
136 	_leave(" = 0");
137 	return 0;
138 
139 	/* an old object from a previous incarnation is hogging the slot - we
140 	 * need to wait for it to be destroyed */
141 wait_for_old_object:
142 	if (xobject->fscache.state < FSCACHE_OBJECT_DYING) {
143 		printk(KERN_ERR "\n");
144 		printk(KERN_ERR "CacheFiles: Error:"
145 		       " Unexpected object collision\n");
146 		cachefiles_printk_object(object, xobject);
147 		BUG();
148 	}
149 	atomic_inc(&xobject->usage);
150 	write_unlock(&cache->active_lock);
151 
152 	if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
153 		wait_queue_head_t *wq;
154 
155 		signed long timeout = 60 * HZ;
156 		wait_queue_t wait;
157 		bool requeue;
158 
159 		/* if the object we're waiting for is queued for processing,
160 		 * then just put ourselves on the queue behind it */
161 		if (slow_work_is_queued(&xobject->fscache.work)) {
162 			_debug("queue OBJ%x behind OBJ%x immediately",
163 			       object->fscache.debug_id,
164 			       xobject->fscache.debug_id);
165 			goto requeue;
166 		}
167 
168 		/* otherwise we sleep until either the object we're waiting for
169 		 * is done, or the slow-work facility wants the thread back to
170 		 * do other work */
171 		wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
172 		init_wait(&wait);
173 		requeue = false;
174 		do {
175 			prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
176 			if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
177 				break;
178 			requeue = slow_work_sleep_till_thread_needed(
179 				&object->fscache.work, &timeout);
180 		} while (timeout > 0 && !requeue);
181 		finish_wait(wq, &wait);
182 
183 		if (requeue &&
184 		    test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
185 			_debug("queue OBJ%x behind OBJ%x after wait",
186 			       object->fscache.debug_id,
187 			       xobject->fscache.debug_id);
188 			goto requeue;
189 		}
190 
191 		if (timeout <= 0) {
192 			printk(KERN_ERR "\n");
193 			printk(KERN_ERR "CacheFiles: Error: Overlong"
194 			       " wait for old active object to go away\n");
195 			cachefiles_printk_object(object, xobject);
196 			goto requeue;
197 		}
198 	}
199 
200 	ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
201 
202 	cache->cache.ops->put_object(&xobject->fscache);
203 	goto try_again;
204 
205 requeue:
206 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
207 	cache->cache.ops->put_object(&xobject->fscache);
208 	_leave(" = -ETIMEDOUT");
209 	return -ETIMEDOUT;
210 }
211 
212 /*
213  * delete an object representation from the cache
214  * - file backed objects are unlinked
215  * - directory backed objects are stuffed into the graveyard for userspace to
216  *   delete
217  * - unlocks the directory mutex
218  */
219 static int cachefiles_bury_object(struct cachefiles_cache *cache,
220 				  struct dentry *dir,
221 				  struct dentry *rep)
222 {
223 	struct dentry *grave, *trap;
224 	char nbuffer[8 + 8 + 1];
225 	int ret;
226 
227 	_enter(",'%*.*s','%*.*s'",
228 	       dir->d_name.len, dir->d_name.len, dir->d_name.name,
229 	       rep->d_name.len, rep->d_name.len, rep->d_name.name);
230 
231 	/* non-directories can just be unlinked */
232 	if (!S_ISDIR(rep->d_inode->i_mode)) {
233 		_debug("unlink stale object");
234 		ret = vfs_unlink(dir->d_inode, rep);
235 
236 		mutex_unlock(&dir->d_inode->i_mutex);
237 
238 		if (ret == -EIO)
239 			cachefiles_io_error(cache, "Unlink failed");
240 
241 		_leave(" = %d", ret);
242 		return ret;
243 	}
244 
245 	/* directories have to be moved to the graveyard */
246 	_debug("move stale object to graveyard");
247 	mutex_unlock(&dir->d_inode->i_mutex);
248 
249 try_again:
250 	/* first step is to make up a grave dentry in the graveyard */
251 	sprintf(nbuffer, "%08x%08x",
252 		(uint32_t) get_seconds(),
253 		(uint32_t) atomic_inc_return(&cache->gravecounter));
254 
255 	/* do the multiway lock magic */
256 	trap = lock_rename(cache->graveyard, dir);
257 
258 	/* do some checks before getting the grave dentry */
259 	if (rep->d_parent != dir) {
260 		/* the entry was probably culled when we dropped the parent dir
261 		 * lock */
262 		unlock_rename(cache->graveyard, dir);
263 		_leave(" = 0 [culled?]");
264 		return 0;
265 	}
266 
267 	if (!S_ISDIR(cache->graveyard->d_inode->i_mode)) {
268 		unlock_rename(cache->graveyard, dir);
269 		cachefiles_io_error(cache, "Graveyard no longer a directory");
270 		return -EIO;
271 	}
272 
273 	if (trap == rep) {
274 		unlock_rename(cache->graveyard, dir);
275 		cachefiles_io_error(cache, "May not make directory loop");
276 		return -EIO;
277 	}
278 
279 	if (d_mountpoint(rep)) {
280 		unlock_rename(cache->graveyard, dir);
281 		cachefiles_io_error(cache, "Mountpoint in cache");
282 		return -EIO;
283 	}
284 
285 	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
286 	if (IS_ERR(grave)) {
287 		unlock_rename(cache->graveyard, dir);
288 
289 		if (PTR_ERR(grave) == -ENOMEM) {
290 			_leave(" = -ENOMEM");
291 			return -ENOMEM;
292 		}
293 
294 		cachefiles_io_error(cache, "Lookup error %ld",
295 				    PTR_ERR(grave));
296 		return -EIO;
297 	}
298 
299 	if (grave->d_inode) {
300 		unlock_rename(cache->graveyard, dir);
301 		dput(grave);
302 		grave = NULL;
303 		cond_resched();
304 		goto try_again;
305 	}
306 
307 	if (d_mountpoint(grave)) {
308 		unlock_rename(cache->graveyard, dir);
309 		dput(grave);
310 		cachefiles_io_error(cache, "Mountpoint in graveyard");
311 		return -EIO;
312 	}
313 
314 	/* target should not be an ancestor of source */
315 	if (trap == grave) {
316 		unlock_rename(cache->graveyard, dir);
317 		dput(grave);
318 		cachefiles_io_error(cache, "May not make directory loop");
319 		return -EIO;
320 	}
321 
322 	/* attempt the rename */
323 	ret = vfs_rename(dir->d_inode, rep, cache->graveyard->d_inode, grave);
324 	if (ret != 0 && ret != -ENOMEM)
325 		cachefiles_io_error(cache, "Rename failed with error %d", ret);
326 
327 	unlock_rename(cache->graveyard, dir);
328 	dput(grave);
329 	_leave(" = 0");
330 	return 0;
331 }
332 
333 /*
334  * delete an object representation from the cache
335  */
336 int cachefiles_delete_object(struct cachefiles_cache *cache,
337 			     struct cachefiles_object *object)
338 {
339 	struct dentry *dir;
340 	int ret;
341 
342 	_enter(",{%p}", object->dentry);
343 
344 	ASSERT(object->dentry);
345 	ASSERT(object->dentry->d_inode);
346 	ASSERT(object->dentry->d_parent);
347 
348 	dir = dget_parent(object->dentry);
349 
350 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
351 	ret = cachefiles_bury_object(cache, dir, object->dentry);
352 
353 	dput(dir);
354 	_leave(" = %d", ret);
355 	return ret;
356 }
357 
358 /*
359  * walk from the parent object to the child object through the backing
360  * filesystem, creating directories as we go
361  */
362 int cachefiles_walk_to_object(struct cachefiles_object *parent,
363 			      struct cachefiles_object *object,
364 			      const char *key,
365 			      struct cachefiles_xattr *auxdata)
366 {
367 	struct cachefiles_cache *cache;
368 	struct dentry *dir, *next = NULL;
369 	unsigned long start;
370 	const char *name;
371 	int ret, nlen;
372 
373 	_enter("{%p},,%s,", parent->dentry, key);
374 
375 	cache = container_of(parent->fscache.cache,
376 			     struct cachefiles_cache, cache);
377 
378 	ASSERT(parent->dentry);
379 	ASSERT(parent->dentry->d_inode);
380 
381 	if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
382 		// TODO: convert file to dir
383 		_leave("looking up in none directory");
384 		return -ENOBUFS;
385 	}
386 
387 	dir = dget(parent->dentry);
388 
389 advance:
390 	/* attempt to transit the first directory component */
391 	name = key;
392 	nlen = strlen(key);
393 
394 	/* key ends in a double NUL */
395 	key = key + nlen + 1;
396 	if (!*key)
397 		key = NULL;
398 
399 lookup_again:
400 	/* search the current directory for the element name */
401 	_debug("lookup '%s'", name);
402 
403 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
404 
405 	start = jiffies;
406 	next = lookup_one_len(name, dir, nlen);
407 	cachefiles_hist(cachefiles_lookup_histogram, start);
408 	if (IS_ERR(next))
409 		goto lookup_error;
410 
411 	_debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
412 
413 	if (!key)
414 		object->new = !next->d_inode;
415 
416 	/* if this element of the path doesn't exist, then the lookup phase
417 	 * failed, and we can release any readers in the certain knowledge that
418 	 * there's nothing for them to actually read */
419 	if (!next->d_inode)
420 		fscache_object_lookup_negative(&object->fscache);
421 
422 	/* we need to create the object if it's negative */
423 	if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
424 		/* index objects and intervening tree levels must be subdirs */
425 		if (!next->d_inode) {
426 			ret = cachefiles_has_space(cache, 1, 0);
427 			if (ret < 0)
428 				goto create_error;
429 
430 			start = jiffies;
431 			ret = vfs_mkdir(dir->d_inode, next, 0);
432 			cachefiles_hist(cachefiles_mkdir_histogram, start);
433 			if (ret < 0)
434 				goto create_error;
435 
436 			ASSERT(next->d_inode);
437 
438 			_debug("mkdir -> %p{%p{ino=%lu}}",
439 			       next, next->d_inode, next->d_inode->i_ino);
440 
441 		} else if (!S_ISDIR(next->d_inode->i_mode)) {
442 			kerror("inode %lu is not a directory",
443 			       next->d_inode->i_ino);
444 			ret = -ENOBUFS;
445 			goto error;
446 		}
447 
448 	} else {
449 		/* non-index objects start out life as files */
450 		if (!next->d_inode) {
451 			ret = cachefiles_has_space(cache, 1, 0);
452 			if (ret < 0)
453 				goto create_error;
454 
455 			start = jiffies;
456 			ret = vfs_create(dir->d_inode, next, S_IFREG, NULL);
457 			cachefiles_hist(cachefiles_create_histogram, start);
458 			if (ret < 0)
459 				goto create_error;
460 
461 			ASSERT(next->d_inode);
462 
463 			_debug("create -> %p{%p{ino=%lu}}",
464 			       next, next->d_inode, next->d_inode->i_ino);
465 
466 		} else if (!S_ISDIR(next->d_inode->i_mode) &&
467 			   !S_ISREG(next->d_inode->i_mode)
468 			   ) {
469 			kerror("inode %lu is not a file or directory",
470 			       next->d_inode->i_ino);
471 			ret = -ENOBUFS;
472 			goto error;
473 		}
474 	}
475 
476 	/* process the next component */
477 	if (key) {
478 		_debug("advance");
479 		mutex_unlock(&dir->d_inode->i_mutex);
480 		dput(dir);
481 		dir = next;
482 		next = NULL;
483 		goto advance;
484 	}
485 
486 	/* we've found the object we were looking for */
487 	object->dentry = next;
488 
489 	/* if we've found that the terminal object exists, then we need to
490 	 * check its attributes and delete it if it's out of date */
491 	if (!object->new) {
492 		_debug("validate '%*.*s'",
493 		       next->d_name.len, next->d_name.len, next->d_name.name);
494 
495 		ret = cachefiles_check_object_xattr(object, auxdata);
496 		if (ret == -ESTALE) {
497 			/* delete the object (the deleter drops the directory
498 			 * mutex) */
499 			object->dentry = NULL;
500 
501 			ret = cachefiles_bury_object(cache, dir, next);
502 			dput(next);
503 			next = NULL;
504 
505 			if (ret < 0)
506 				goto delete_error;
507 
508 			_debug("redo lookup");
509 			goto lookup_again;
510 		}
511 	}
512 
513 	/* note that we're now using this object */
514 	ret = cachefiles_mark_object_active(cache, object);
515 
516 	mutex_unlock(&dir->d_inode->i_mutex);
517 	dput(dir);
518 	dir = NULL;
519 
520 	if (ret == -ETIMEDOUT)
521 		goto mark_active_timed_out;
522 
523 	_debug("=== OBTAINED_OBJECT ===");
524 
525 	if (object->new) {
526 		/* attach data to a newly constructed terminal object */
527 		ret = cachefiles_set_object_xattr(object, auxdata);
528 		if (ret < 0)
529 			goto check_error;
530 	} else {
531 		/* always update the atime on an object we've just looked up
532 		 * (this is used to keep track of culling, and atimes are only
533 		 * updated by read, write and readdir but not lookup or
534 		 * open) */
535 		touch_atime(cache->mnt, next);
536 	}
537 
538 	/* open a file interface onto a data file */
539 	if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
540 		if (S_ISREG(object->dentry->d_inode->i_mode)) {
541 			const struct address_space_operations *aops;
542 
543 			ret = -EPERM;
544 			aops = object->dentry->d_inode->i_mapping->a_ops;
545 			if (!aops->bmap)
546 				goto check_error;
547 
548 			object->backer = object->dentry;
549 		} else {
550 			BUG(); // TODO: open file in data-class subdir
551 		}
552 	}
553 
554 	object->new = 0;
555 	fscache_obtained_object(&object->fscache);
556 
557 	_leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
558 	return 0;
559 
560 create_error:
561 	_debug("create error %d", ret);
562 	if (ret == -EIO)
563 		cachefiles_io_error(cache, "Create/mkdir failed");
564 	goto error;
565 
566 mark_active_timed_out:
567 	_debug("mark active timed out");
568 	goto release_dentry;
569 
570 check_error:
571 	_debug("check error %d", ret);
572 	write_lock(&cache->active_lock);
573 	rb_erase(&object->active_node, &cache->active_nodes);
574 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
575 	wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
576 	write_unlock(&cache->active_lock);
577 release_dentry:
578 	dput(object->dentry);
579 	object->dentry = NULL;
580 	goto error_out;
581 
582 delete_error:
583 	_debug("delete error %d", ret);
584 	goto error_out2;
585 
586 lookup_error:
587 	_debug("lookup error %ld", PTR_ERR(next));
588 	ret = PTR_ERR(next);
589 	if (ret == -EIO)
590 		cachefiles_io_error(cache, "Lookup failed");
591 	next = NULL;
592 error:
593 	mutex_unlock(&dir->d_inode->i_mutex);
594 	dput(next);
595 error_out2:
596 	dput(dir);
597 error_out:
598 	_leave(" = error %d", -ret);
599 	return ret;
600 }
601 
602 /*
603  * get a subdirectory
604  */
605 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
606 					struct dentry *dir,
607 					const char *dirname)
608 {
609 	struct dentry *subdir;
610 	unsigned long start;
611 	int ret;
612 
613 	_enter(",,%s", dirname);
614 
615 	/* search the current directory for the element name */
616 	mutex_lock(&dir->d_inode->i_mutex);
617 
618 	start = jiffies;
619 	subdir = lookup_one_len(dirname, dir, strlen(dirname));
620 	cachefiles_hist(cachefiles_lookup_histogram, start);
621 	if (IS_ERR(subdir)) {
622 		if (PTR_ERR(subdir) == -ENOMEM)
623 			goto nomem_d_alloc;
624 		goto lookup_error;
625 	}
626 
627 	_debug("subdir -> %p %s",
628 	       subdir, subdir->d_inode ? "positive" : "negative");
629 
630 	/* we need to create the subdir if it doesn't exist yet */
631 	if (!subdir->d_inode) {
632 		ret = cachefiles_has_space(cache, 1, 0);
633 		if (ret < 0)
634 			goto mkdir_error;
635 
636 		_debug("attempt mkdir");
637 
638 		ret = vfs_mkdir(dir->d_inode, subdir, 0700);
639 		if (ret < 0)
640 			goto mkdir_error;
641 
642 		ASSERT(subdir->d_inode);
643 
644 		_debug("mkdir -> %p{%p{ino=%lu}}",
645 		       subdir,
646 		       subdir->d_inode,
647 		       subdir->d_inode->i_ino);
648 	}
649 
650 	mutex_unlock(&dir->d_inode->i_mutex);
651 
652 	/* we need to make sure the subdir is a directory */
653 	ASSERT(subdir->d_inode);
654 
655 	if (!S_ISDIR(subdir->d_inode->i_mode)) {
656 		kerror("%s is not a directory", dirname);
657 		ret = -EIO;
658 		goto check_error;
659 	}
660 
661 	ret = -EPERM;
662 	if (!subdir->d_inode->i_op ||
663 	    !subdir->d_inode->i_op->setxattr ||
664 	    !subdir->d_inode->i_op->getxattr ||
665 	    !subdir->d_inode->i_op->lookup ||
666 	    !subdir->d_inode->i_op->mkdir ||
667 	    !subdir->d_inode->i_op->create ||
668 	    !subdir->d_inode->i_op->rename ||
669 	    !subdir->d_inode->i_op->rmdir ||
670 	    !subdir->d_inode->i_op->unlink)
671 		goto check_error;
672 
673 	_leave(" = [%lu]", subdir->d_inode->i_ino);
674 	return subdir;
675 
676 check_error:
677 	dput(subdir);
678 	_leave(" = %d [check]", ret);
679 	return ERR_PTR(ret);
680 
681 mkdir_error:
682 	mutex_unlock(&dir->d_inode->i_mutex);
683 	dput(subdir);
684 	kerror("mkdir %s failed with error %d", dirname, ret);
685 	return ERR_PTR(ret);
686 
687 lookup_error:
688 	mutex_unlock(&dir->d_inode->i_mutex);
689 	ret = PTR_ERR(subdir);
690 	kerror("Lookup %s failed with error %d", dirname, ret);
691 	return ERR_PTR(ret);
692 
693 nomem_d_alloc:
694 	mutex_unlock(&dir->d_inode->i_mutex);
695 	_leave(" = -ENOMEM");
696 	return ERR_PTR(-ENOMEM);
697 }
698 
699 /*
700  * find out if an object is in use or not
701  * - if finds object and it's not in use:
702  *   - returns a pointer to the object and a reference on it
703  *   - returns with the directory locked
704  */
705 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
706 					      struct dentry *dir,
707 					      char *filename)
708 {
709 	struct cachefiles_object *object;
710 	struct rb_node *_n;
711 	struct dentry *victim;
712 	unsigned long start;
713 	int ret;
714 
715 	//_enter(",%*.*s/,%s",
716 	//       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
717 
718 	/* look up the victim */
719 	mutex_lock_nested(&dir->d_inode->i_mutex, 1);
720 
721 	start = jiffies;
722 	victim = lookup_one_len(filename, dir, strlen(filename));
723 	cachefiles_hist(cachefiles_lookup_histogram, start);
724 	if (IS_ERR(victim))
725 		goto lookup_error;
726 
727 	//_debug("victim -> %p %s",
728 	//       victim, victim->d_inode ? "positive" : "negative");
729 
730 	/* if the object is no longer there then we probably retired the object
731 	 * at the netfs's request whilst the cull was in progress
732 	 */
733 	if (!victim->d_inode) {
734 		mutex_unlock(&dir->d_inode->i_mutex);
735 		dput(victim);
736 		_leave(" = -ENOENT [absent]");
737 		return ERR_PTR(-ENOENT);
738 	}
739 
740 	/* check to see if we're using this object */
741 	read_lock(&cache->active_lock);
742 
743 	_n = cache->active_nodes.rb_node;
744 
745 	while (_n) {
746 		object = rb_entry(_n, struct cachefiles_object, active_node);
747 
748 		if (object->dentry > victim)
749 			_n = _n->rb_left;
750 		else if (object->dentry < victim)
751 			_n = _n->rb_right;
752 		else
753 			goto object_in_use;
754 	}
755 
756 	read_unlock(&cache->active_lock);
757 
758 	//_leave(" = %p", victim);
759 	return victim;
760 
761 object_in_use:
762 	read_unlock(&cache->active_lock);
763 	mutex_unlock(&dir->d_inode->i_mutex);
764 	dput(victim);
765 	//_leave(" = -EBUSY [in use]");
766 	return ERR_PTR(-EBUSY);
767 
768 lookup_error:
769 	mutex_unlock(&dir->d_inode->i_mutex);
770 	ret = PTR_ERR(victim);
771 	if (ret == -ENOENT) {
772 		/* file or dir now absent - probably retired by netfs */
773 		_leave(" = -ESTALE [absent]");
774 		return ERR_PTR(-ESTALE);
775 	}
776 
777 	if (ret == -EIO) {
778 		cachefiles_io_error(cache, "Lookup failed");
779 	} else if (ret != -ENOMEM) {
780 		kerror("Internal error: %d", ret);
781 		ret = -EIO;
782 	}
783 
784 	_leave(" = %d", ret);
785 	return ERR_PTR(ret);
786 }
787 
788 /*
789  * cull an object if it's not in use
790  * - called only by cache manager daemon
791  */
792 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
793 		    char *filename)
794 {
795 	struct dentry *victim;
796 	int ret;
797 
798 	_enter(",%*.*s/,%s",
799 	       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
800 
801 	victim = cachefiles_check_active(cache, dir, filename);
802 	if (IS_ERR(victim))
803 		return PTR_ERR(victim);
804 
805 	_debug("victim -> %p %s",
806 	       victim, victim->d_inode ? "positive" : "negative");
807 
808 	/* okay... the victim is not being used so we can cull it
809 	 * - start by marking it as stale
810 	 */
811 	_debug("victim is cullable");
812 
813 	ret = cachefiles_remove_object_xattr(cache, victim);
814 	if (ret < 0)
815 		goto error_unlock;
816 
817 	/*  actually remove the victim (drops the dir mutex) */
818 	_debug("bury");
819 
820 	ret = cachefiles_bury_object(cache, dir, victim);
821 	if (ret < 0)
822 		goto error;
823 
824 	dput(victim);
825 	_leave(" = 0");
826 	return 0;
827 
828 error_unlock:
829 	mutex_unlock(&dir->d_inode->i_mutex);
830 error:
831 	dput(victim);
832 	if (ret == -ENOENT) {
833 		/* file or dir now absent - probably retired by netfs */
834 		_leave(" = -ESTALE [absent]");
835 		return -ESTALE;
836 	}
837 
838 	if (ret != -ENOMEM) {
839 		kerror("Internal error: %d", ret);
840 		ret = -EIO;
841 	}
842 
843 	_leave(" = %d", ret);
844 	return ret;
845 }
846 
847 /*
848  * find out if an object is in use or not
849  * - called only by cache manager daemon
850  * - returns -EBUSY or 0 to indicate whether an object is in use or not
851  */
852 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
853 			    char *filename)
854 {
855 	struct dentry *victim;
856 
857 	//_enter(",%*.*s/,%s",
858 	//       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
859 
860 	victim = cachefiles_check_active(cache, dir, filename);
861 	if (IS_ERR(victim))
862 		return PTR_ERR(victim);
863 
864 	mutex_unlock(&dir->d_inode->i_mutex);
865 	dput(victim);
866 	//_leave(" = 0");
867 	return 0;
868 }
869