xref: /linux/fs/cachefiles/namei.c (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
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 <linux/slab.h>
23 #include "internal.h"
24 
25 #define CACHEFILES_KEYBUF_SIZE 512
26 
27 /*
28  * dump debugging info about an object
29  */
30 static noinline
31 void __cachefiles_printk_object(struct cachefiles_object *object,
32 				const char *prefix,
33 				u8 *keybuf)
34 {
35 	struct fscache_cookie *cookie;
36 	unsigned keylen, loop;
37 
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);
50 
51 	spin_lock(&object->fscache.lock);
52 	cookie = object->fscache.cookie;
53 	if (cookie) {
54 		printk(KERN_ERR "%scookie=%p [pr=%p nd=%p fl=%lx]\n",
55 		       prefix,
56 		       object->fscache.cookie,
57 		       object->fscache.cookie->parent,
58 		       object->fscache.cookie->netfs_data,
59 		       object->fscache.cookie->flags);
60 		if (keybuf)
61 			keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
62 						      CACHEFILES_KEYBUF_SIZE);
63 		else
64 			keylen = 0;
65 	} else {
66 		printk(KERN_ERR "%scookie=NULL\n", prefix);
67 		keylen = 0;
68 	}
69 	spin_unlock(&object->fscache.lock);
70 
71 	if (keylen) {
72 		printk(KERN_ERR "%skey=[%u] '", prefix, keylen);
73 		for (loop = 0; loop < keylen; loop++)
74 			printk("%02x", keybuf[loop]);
75 		printk("'\n");
76 	}
77 }
78 
79 /*
80  * dump debugging info about a pair of objects
81  */
82 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
83 					      struct cachefiles_object *xobject)
84 {
85 	u8 *keybuf;
86 
87 	keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
88 	if (object)
89 		__cachefiles_printk_object(object, "", keybuf);
90 	if (xobject)
91 		__cachefiles_printk_object(xobject, "x", keybuf);
92 	kfree(keybuf);
93 }
94 
95 /*
96  * record the fact that an object is now active
97  */
98 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
99 					 struct cachefiles_object *object)
100 {
101 	struct cachefiles_object *xobject;
102 	struct rb_node **_p, *_parent = NULL;
103 	struct dentry *dentry;
104 
105 	_enter(",%p", object);
106 
107 try_again:
108 	write_lock(&cache->active_lock);
109 
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);
113 		BUG();
114 	}
115 
116 	dentry = object->dentry;
117 	_p = &cache->active_nodes.rb_node;
118 	while (*_p) {
119 		_parent = *_p;
120 		xobject = rb_entry(_parent,
121 				   struct cachefiles_object, active_node);
122 
123 		ASSERT(xobject != object);
124 
125 		if (xobject->dentry > dentry)
126 			_p = &(*_p)->rb_left;
127 		else if (xobject->dentry < dentry)
128 			_p = &(*_p)->rb_right;
129 		else
130 			goto wait_for_old_object;
131 	}
132 
133 	rb_link_node(&object->active_node, _parent, _p);
134 	rb_insert_color(&object->active_node, &cache->active_nodes);
135 
136 	write_unlock(&cache->active_lock);
137 	_leave(" = 0");
138 	return 0;
139 
140 	/* an old object from a previous incarnation is hogging the slot - we
141 	 * need to wait for it to be destroyed */
142 wait_for_old_object:
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);
148 		BUG();
149 	}
150 	atomic_inc(&xobject->usage);
151 	write_unlock(&cache->active_lock);
152 
153 	if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
154 		wait_queue_head_t *wq;
155 
156 		signed long timeout = 60 * HZ;
157 		wait_queue_t wait;
158 		bool requeue;
159 
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);
166 			goto requeue;
167 		}
168 
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
171 		 * do other work */
172 		wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
173 		init_wait(&wait);
174 		requeue = false;
175 		do {
176 			prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
177 			if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
178 				break;
179 			requeue = slow_work_sleep_till_thread_needed(
180 				&object->fscache.work, &timeout);
181 		} while (timeout > 0 && !requeue);
182 		finish_wait(wq, &wait);
183 
184 		if (requeue &&
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);
189 			goto requeue;
190 		}
191 
192 		if (timeout <= 0) {
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);
197 			goto requeue;
198 		}
199 	}
200 
201 	ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
202 
203 	cache->cache.ops->put_object(&xobject->fscache);
204 	goto try_again;
205 
206 requeue:
207 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
208 	cache->cache.ops->put_object(&xobject->fscache);
209 	_leave(" = -ETIMEDOUT");
210 	return -ETIMEDOUT;
211 }
212 
213 /*
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
217  *   delete
218  * - unlocks the directory mutex
219  */
220 static int cachefiles_bury_object(struct cachefiles_cache *cache,
221 				  struct dentry *dir,
222 				  struct dentry *rep)
223 {
224 	struct dentry *grave, *trap;
225 	char nbuffer[8 + 8 + 1];
226 	int ret;
227 
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);
231 
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);
236 
237 		mutex_unlock(&dir->d_inode->i_mutex);
238 
239 		if (ret == -EIO)
240 			cachefiles_io_error(cache, "Unlink failed");
241 
242 		_leave(" = %d", ret);
243 		return ret;
244 	}
245 
246 	/* directories have to be moved to the graveyard */
247 	_debug("move stale object to graveyard");
248 	mutex_unlock(&dir->d_inode->i_mutex);
249 
250 try_again:
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));
255 
256 	/* do the multiway lock magic */
257 	trap = lock_rename(cache->graveyard, dir);
258 
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
262 		 * lock */
263 		unlock_rename(cache->graveyard, dir);
264 		_leave(" = 0 [culled?]");
265 		return 0;
266 	}
267 
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");
271 		return -EIO;
272 	}
273 
274 	if (trap == rep) {
275 		unlock_rename(cache->graveyard, dir);
276 		cachefiles_io_error(cache, "May not make directory loop");
277 		return -EIO;
278 	}
279 
280 	if (d_mountpoint(rep)) {
281 		unlock_rename(cache->graveyard, dir);
282 		cachefiles_io_error(cache, "Mountpoint in cache");
283 		return -EIO;
284 	}
285 
286 	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
287 	if (IS_ERR(grave)) {
288 		unlock_rename(cache->graveyard, dir);
289 
290 		if (PTR_ERR(grave) == -ENOMEM) {
291 			_leave(" = -ENOMEM");
292 			return -ENOMEM;
293 		}
294 
295 		cachefiles_io_error(cache, "Lookup error %ld",
296 				    PTR_ERR(grave));
297 		return -EIO;
298 	}
299 
300 	if (grave->d_inode) {
301 		unlock_rename(cache->graveyard, dir);
302 		dput(grave);
303 		grave = NULL;
304 		cond_resched();
305 		goto try_again;
306 	}
307 
308 	if (d_mountpoint(grave)) {
309 		unlock_rename(cache->graveyard, dir);
310 		dput(grave);
311 		cachefiles_io_error(cache, "Mountpoint in graveyard");
312 		return -EIO;
313 	}
314 
315 	/* target should not be an ancestor of source */
316 	if (trap == grave) {
317 		unlock_rename(cache->graveyard, dir);
318 		dput(grave);
319 		cachefiles_io_error(cache, "May not make directory loop");
320 		return -EIO;
321 	}
322 
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);
327 
328 	unlock_rename(cache->graveyard, dir);
329 	dput(grave);
330 	_leave(" = 0");
331 	return 0;
332 }
333 
334 /*
335  * delete an object representation from the cache
336  */
337 int cachefiles_delete_object(struct cachefiles_cache *cache,
338 			     struct cachefiles_object *object)
339 {
340 	struct dentry *dir;
341 	int ret;
342 
343 	_enter(",{%p}", object->dentry);
344 
345 	ASSERT(object->dentry);
346 	ASSERT(object->dentry->d_inode);
347 	ASSERT(object->dentry->d_parent);
348 
349 	dir = dget_parent(object->dentry);
350 
351 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
352 
353 	/* we need to check that our parent is _still_ our parent - it may have
354 	 * been renamed */
355 	if (dir == object->dentry->d_parent) {
356 		ret = cachefiles_bury_object(cache, dir, object->dentry);
357 	} else {
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);
361 		ret = 0;
362 	}
363 
364 	dput(dir);
365 	_leave(" = %d", ret);
366 	return ret;
367 }
368 
369 /*
370  * walk from the parent object to the child object through the backing
371  * filesystem, creating directories as we go
372  */
373 int cachefiles_walk_to_object(struct cachefiles_object *parent,
374 			      struct cachefiles_object *object,
375 			      const char *key,
376 			      struct cachefiles_xattr *auxdata)
377 {
378 	struct cachefiles_cache *cache;
379 	struct dentry *dir, *next = NULL;
380 	unsigned long start;
381 	const char *name;
382 	int ret, nlen;
383 
384 	_enter("{%p},,%s,", parent->dentry, key);
385 
386 	cache = container_of(parent->fscache.cache,
387 			     struct cachefiles_cache, cache);
388 
389 	ASSERT(parent->dentry);
390 	ASSERT(parent->dentry->d_inode);
391 
392 	if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
393 		// TODO: convert file to dir
394 		_leave("looking up in none directory");
395 		return -ENOBUFS;
396 	}
397 
398 	dir = dget(parent->dentry);
399 
400 advance:
401 	/* attempt to transit the first directory component */
402 	name = key;
403 	nlen = strlen(key);
404 
405 	/* key ends in a double NUL */
406 	key = key + nlen + 1;
407 	if (!*key)
408 		key = NULL;
409 
410 lookup_again:
411 	/* search the current directory for the element name */
412 	_debug("lookup '%s'", name);
413 
414 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
415 
416 	start = jiffies;
417 	next = lookup_one_len(name, dir, nlen);
418 	cachefiles_hist(cachefiles_lookup_histogram, start);
419 	if (IS_ERR(next))
420 		goto lookup_error;
421 
422 	_debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
423 
424 	if (!key)
425 		object->new = !next->d_inode;
426 
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 */
430 	if (!next->d_inode)
431 		fscache_object_lookup_negative(&object->fscache);
432 
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);
438 			if (ret < 0)
439 				goto create_error;
440 
441 			start = jiffies;
442 			ret = vfs_mkdir(dir->d_inode, next, 0);
443 			cachefiles_hist(cachefiles_mkdir_histogram, start);
444 			if (ret < 0)
445 				goto create_error;
446 
447 			ASSERT(next->d_inode);
448 
449 			_debug("mkdir -> %p{%p{ino=%lu}}",
450 			       next, next->d_inode, next->d_inode->i_ino);
451 
452 		} else if (!S_ISDIR(next->d_inode->i_mode)) {
453 			kerror("inode %lu is not a directory",
454 			       next->d_inode->i_ino);
455 			ret = -ENOBUFS;
456 			goto error;
457 		}
458 
459 	} else {
460 		/* non-index objects start out life as files */
461 		if (!next->d_inode) {
462 			ret = cachefiles_has_space(cache, 1, 0);
463 			if (ret < 0)
464 				goto create_error;
465 
466 			start = jiffies;
467 			ret = vfs_create(dir->d_inode, next, S_IFREG, NULL);
468 			cachefiles_hist(cachefiles_create_histogram, start);
469 			if (ret < 0)
470 				goto create_error;
471 
472 			ASSERT(next->d_inode);
473 
474 			_debug("create -> %p{%p{ino=%lu}}",
475 			       next, next->d_inode, next->d_inode->i_ino);
476 
477 		} else if (!S_ISDIR(next->d_inode->i_mode) &&
478 			   !S_ISREG(next->d_inode->i_mode)
479 			   ) {
480 			kerror("inode %lu is not a file or directory",
481 			       next->d_inode->i_ino);
482 			ret = -ENOBUFS;
483 			goto error;
484 		}
485 	}
486 
487 	/* process the next component */
488 	if (key) {
489 		_debug("advance");
490 		mutex_unlock(&dir->d_inode->i_mutex);
491 		dput(dir);
492 		dir = next;
493 		next = NULL;
494 		goto advance;
495 	}
496 
497 	/* we've found the object we were looking for */
498 	object->dentry = next;
499 
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 */
502 	if (!object->new) {
503 		_debug("validate '%*.*s'",
504 		       next->d_name.len, next->d_name.len, next->d_name.name);
505 
506 		ret = cachefiles_check_object_xattr(object, auxdata);
507 		if (ret == -ESTALE) {
508 			/* delete the object (the deleter drops the directory
509 			 * mutex) */
510 			object->dentry = NULL;
511 
512 			ret = cachefiles_bury_object(cache, dir, next);
513 			dput(next);
514 			next = NULL;
515 
516 			if (ret < 0)
517 				goto delete_error;
518 
519 			_debug("redo lookup");
520 			goto lookup_again;
521 		}
522 	}
523 
524 	/* note that we're now using this object */
525 	ret = cachefiles_mark_object_active(cache, object);
526 
527 	mutex_unlock(&dir->d_inode->i_mutex);
528 	dput(dir);
529 	dir = NULL;
530 
531 	if (ret == -ETIMEDOUT)
532 		goto mark_active_timed_out;
533 
534 	_debug("=== OBTAINED_OBJECT ===");
535 
536 	if (object->new) {
537 		/* attach data to a newly constructed terminal object */
538 		ret = cachefiles_set_object_xattr(object, auxdata);
539 		if (ret < 0)
540 			goto check_error;
541 	} else {
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
545 		 * open) */
546 		touch_atime(cache->mnt, next);
547 	}
548 
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;
553 
554 			ret = -EPERM;
555 			aops = object->dentry->d_inode->i_mapping->a_ops;
556 			if (!aops->bmap)
557 				goto check_error;
558 
559 			object->backer = object->dentry;
560 		} else {
561 			BUG(); // TODO: open file in data-class subdir
562 		}
563 	}
564 
565 	object->new = 0;
566 	fscache_obtained_object(&object->fscache);
567 
568 	_leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
569 	return 0;
570 
571 create_error:
572 	_debug("create error %d", ret);
573 	if (ret == -EIO)
574 		cachefiles_io_error(cache, "Create/mkdir failed");
575 	goto error;
576 
577 mark_active_timed_out:
578 	_debug("mark active timed out");
579 	goto release_dentry;
580 
581 check_error:
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);
588 release_dentry:
589 	dput(object->dentry);
590 	object->dentry = NULL;
591 	goto error_out;
592 
593 delete_error:
594 	_debug("delete error %d", ret);
595 	goto error_out2;
596 
597 lookup_error:
598 	_debug("lookup error %ld", PTR_ERR(next));
599 	ret = PTR_ERR(next);
600 	if (ret == -EIO)
601 		cachefiles_io_error(cache, "Lookup failed");
602 	next = NULL;
603 error:
604 	mutex_unlock(&dir->d_inode->i_mutex);
605 	dput(next);
606 error_out2:
607 	dput(dir);
608 error_out:
609 	_leave(" = error %d", -ret);
610 	return ret;
611 }
612 
613 /*
614  * get a subdirectory
615  */
616 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
617 					struct dentry *dir,
618 					const char *dirname)
619 {
620 	struct dentry *subdir;
621 	unsigned long start;
622 	int ret;
623 
624 	_enter(",,%s", dirname);
625 
626 	/* search the current directory for the element name */
627 	mutex_lock(&dir->d_inode->i_mutex);
628 
629 	start = jiffies;
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)
634 			goto nomem_d_alloc;
635 		goto lookup_error;
636 	}
637 
638 	_debug("subdir -> %p %s",
639 	       subdir, subdir->d_inode ? "positive" : "negative");
640 
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);
644 		if (ret < 0)
645 			goto mkdir_error;
646 
647 		_debug("attempt mkdir");
648 
649 		ret = vfs_mkdir(dir->d_inode, subdir, 0700);
650 		if (ret < 0)
651 			goto mkdir_error;
652 
653 		ASSERT(subdir->d_inode);
654 
655 		_debug("mkdir -> %p{%p{ino=%lu}}",
656 		       subdir,
657 		       subdir->d_inode,
658 		       subdir->d_inode->i_ino);
659 	}
660 
661 	mutex_unlock(&dir->d_inode->i_mutex);
662 
663 	/* we need to make sure the subdir is a directory */
664 	ASSERT(subdir->d_inode);
665 
666 	if (!S_ISDIR(subdir->d_inode->i_mode)) {
667 		kerror("%s is not a directory", dirname);
668 		ret = -EIO;
669 		goto check_error;
670 	}
671 
672 	ret = -EPERM;
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)
682 		goto check_error;
683 
684 	_leave(" = [%lu]", subdir->d_inode->i_ino);
685 	return subdir;
686 
687 check_error:
688 	dput(subdir);
689 	_leave(" = %d [check]", ret);
690 	return ERR_PTR(ret);
691 
692 mkdir_error:
693 	mutex_unlock(&dir->d_inode->i_mutex);
694 	dput(subdir);
695 	kerror("mkdir %s failed with error %d", dirname, ret);
696 	return ERR_PTR(ret);
697 
698 lookup_error:
699 	mutex_unlock(&dir->d_inode->i_mutex);
700 	ret = PTR_ERR(subdir);
701 	kerror("Lookup %s failed with error %d", dirname, ret);
702 	return ERR_PTR(ret);
703 
704 nomem_d_alloc:
705 	mutex_unlock(&dir->d_inode->i_mutex);
706 	_leave(" = -ENOMEM");
707 	return ERR_PTR(-ENOMEM);
708 }
709 
710 /*
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
715  */
716 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
717 					      struct dentry *dir,
718 					      char *filename)
719 {
720 	struct cachefiles_object *object;
721 	struct rb_node *_n;
722 	struct dentry *victim;
723 	unsigned long start;
724 	int ret;
725 
726 	//_enter(",%*.*s/,%s",
727 	//       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
728 
729 	/* look up the victim */
730 	mutex_lock_nested(&dir->d_inode->i_mutex, 1);
731 
732 	start = jiffies;
733 	victim = lookup_one_len(filename, dir, strlen(filename));
734 	cachefiles_hist(cachefiles_lookup_histogram, start);
735 	if (IS_ERR(victim))
736 		goto lookup_error;
737 
738 	//_debug("victim -> %p %s",
739 	//       victim, victim->d_inode ? "positive" : "negative");
740 
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
743 	 */
744 	if (!victim->d_inode) {
745 		mutex_unlock(&dir->d_inode->i_mutex);
746 		dput(victim);
747 		_leave(" = -ENOENT [absent]");
748 		return ERR_PTR(-ENOENT);
749 	}
750 
751 	/* check to see if we're using this object */
752 	read_lock(&cache->active_lock);
753 
754 	_n = cache->active_nodes.rb_node;
755 
756 	while (_n) {
757 		object = rb_entry(_n, struct cachefiles_object, active_node);
758 
759 		if (object->dentry > victim)
760 			_n = _n->rb_left;
761 		else if (object->dentry < victim)
762 			_n = _n->rb_right;
763 		else
764 			goto object_in_use;
765 	}
766 
767 	read_unlock(&cache->active_lock);
768 
769 	//_leave(" = %p", victim);
770 	return victim;
771 
772 object_in_use:
773 	read_unlock(&cache->active_lock);
774 	mutex_unlock(&dir->d_inode->i_mutex);
775 	dput(victim);
776 	//_leave(" = -EBUSY [in use]");
777 	return ERR_PTR(-EBUSY);
778 
779 lookup_error:
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);
786 	}
787 
788 	if (ret == -EIO) {
789 		cachefiles_io_error(cache, "Lookup failed");
790 	} else if (ret != -ENOMEM) {
791 		kerror("Internal error: %d", ret);
792 		ret = -EIO;
793 	}
794 
795 	_leave(" = %d", ret);
796 	return ERR_PTR(ret);
797 }
798 
799 /*
800  * cull an object if it's not in use
801  * - called only by cache manager daemon
802  */
803 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
804 		    char *filename)
805 {
806 	struct dentry *victim;
807 	int ret;
808 
809 	_enter(",%*.*s/,%s",
810 	       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
811 
812 	victim = cachefiles_check_active(cache, dir, filename);
813 	if (IS_ERR(victim))
814 		return PTR_ERR(victim);
815 
816 	_debug("victim -> %p %s",
817 	       victim, victim->d_inode ? "positive" : "negative");
818 
819 	/* okay... the victim is not being used so we can cull it
820 	 * - start by marking it as stale
821 	 */
822 	_debug("victim is cullable");
823 
824 	ret = cachefiles_remove_object_xattr(cache, victim);
825 	if (ret < 0)
826 		goto error_unlock;
827 
828 	/*  actually remove the victim (drops the dir mutex) */
829 	_debug("bury");
830 
831 	ret = cachefiles_bury_object(cache, dir, victim);
832 	if (ret < 0)
833 		goto error;
834 
835 	dput(victim);
836 	_leave(" = 0");
837 	return 0;
838 
839 error_unlock:
840 	mutex_unlock(&dir->d_inode->i_mutex);
841 error:
842 	dput(victim);
843 	if (ret == -ENOENT) {
844 		/* file or dir now absent - probably retired by netfs */
845 		_leave(" = -ESTALE [absent]");
846 		return -ESTALE;
847 	}
848 
849 	if (ret != -ENOMEM) {
850 		kerror("Internal error: %d", ret);
851 		ret = -EIO;
852 	}
853 
854 	_leave(" = %d", ret);
855 	return ret;
856 }
857 
858 /*
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
862  */
863 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
864 			    char *filename)
865 {
866 	struct dentry *victim;
867 
868 	//_enter(",%*.*s/,%s",
869 	//       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
870 
871 	victim = cachefiles_check_active(cache, dir, filename);
872 	if (IS_ERR(victim))
873 		return PTR_ERR(victim);
874 
875 	mutex_unlock(&dir->d_inode->i_mutex);
876 	dput(victim);
877 	//_leave(" = 0");
878 	return 0;
879 }
880