xref: /openbmc/linux/fs/cachefiles/namei.c (revision 0b003749)
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 <linux/xattr.h>
24 #include "internal.h"
25 
26 #define CACHEFILES_KEYBUF_SIZE 512
27 
28 /*
29  * dump debugging info about an object
30  */
31 static noinline
32 void __cachefiles_printk_object(struct cachefiles_object *object,
33 				const char *prefix)
34 {
35 	struct fscache_cookie *cookie;
36 	const u8 *k;
37 	unsigned loop;
38 
39 	pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
40 	pr_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 	pr_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 	pr_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 		pr_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 		pr_err("%skey=[%u] '", prefix, cookie->key_len);
60 		k = (cookie->key_len <= sizeof(cookie->inline_key)) ?
61 			cookie->inline_key : cookie->key;
62 		for (loop = 0; loop < cookie->key_len; loop++)
63 			pr_cont("%02x", k[loop]);
64 		pr_cont("'\n");
65 	} else {
66 		pr_err("%scookie=NULL\n", prefix);
67 	}
68 	spin_unlock(&object->fscache.lock);
69 }
70 
71 /*
72  * dump debugging info about a pair of objects
73  */
74 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
75 					      struct cachefiles_object *xobject)
76 {
77 	if (object)
78 		__cachefiles_printk_object(object, "");
79 	if (xobject)
80 		__cachefiles_printk_object(xobject, "x");
81 }
82 
83 /*
84  * mark the owner of a dentry, if there is one, to indicate that that dentry
85  * has been preemptively deleted
86  * - the caller must hold the i_mutex on the dentry's parent as required to
87  *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
88  */
89 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
90 					  struct dentry *dentry,
91 					  enum fscache_why_object_killed why)
92 {
93 	struct cachefiles_object *object;
94 	struct rb_node *p;
95 
96 	_enter(",'%pd'", dentry);
97 
98 	write_lock(&cache->active_lock);
99 
100 	p = cache->active_nodes.rb_node;
101 	while (p) {
102 		object = rb_entry(p, struct cachefiles_object, active_node);
103 		if (object->dentry > dentry)
104 			p = p->rb_left;
105 		else if (object->dentry < dentry)
106 			p = p->rb_right;
107 		else
108 			goto found_dentry;
109 	}
110 
111 	write_unlock(&cache->active_lock);
112 	trace_cachefiles_mark_buried(NULL, dentry, why);
113 	_leave(" [no owner]");
114 	return;
115 
116 	/* found the dentry for  */
117 found_dentry:
118 	kdebug("preemptive burial: OBJ%x [%s] %p",
119 	       object->fscache.debug_id,
120 	       object->fscache.state->name,
121 	       dentry);
122 
123 	trace_cachefiles_mark_buried(object, dentry, why);
124 
125 	if (fscache_object_is_live(&object->fscache)) {
126 		pr_err("\n");
127 		pr_err("Error: Can't preemptively bury live object\n");
128 		cachefiles_printk_object(object, NULL);
129 	} else {
130 		if (why != FSCACHE_OBJECT_IS_STALE)
131 			fscache_object_mark_killed(&object->fscache, why);
132 	}
133 
134 	write_unlock(&cache->active_lock);
135 	_leave(" [owner marked]");
136 }
137 
138 /*
139  * record the fact that an object is now active
140  */
141 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
142 					 struct cachefiles_object *object)
143 {
144 	struct cachefiles_object *xobject;
145 	struct rb_node **_p, *_parent = NULL;
146 	struct dentry *dentry;
147 
148 	_enter(",%p", object);
149 
150 try_again:
151 	write_lock(&cache->active_lock);
152 
153 	dentry = object->dentry;
154 	trace_cachefiles_mark_active(object, dentry);
155 
156 	if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
157 		pr_err("Error: Object already active\n");
158 		cachefiles_printk_object(object, NULL);
159 		BUG();
160 	}
161 
162 	_p = &cache->active_nodes.rb_node;
163 	while (*_p) {
164 		_parent = *_p;
165 		xobject = rb_entry(_parent,
166 				   struct cachefiles_object, active_node);
167 
168 		ASSERT(xobject != object);
169 
170 		if (xobject->dentry > dentry)
171 			_p = &(*_p)->rb_left;
172 		else if (xobject->dentry < dentry)
173 			_p = &(*_p)->rb_right;
174 		else
175 			goto wait_for_old_object;
176 	}
177 
178 	rb_link_node(&object->active_node, _parent, _p);
179 	rb_insert_color(&object->active_node, &cache->active_nodes);
180 
181 	write_unlock(&cache->active_lock);
182 	_leave(" = 0");
183 	return 0;
184 
185 	/* an old object from a previous incarnation is hogging the slot - we
186 	 * need to wait for it to be destroyed */
187 wait_for_old_object:
188 	trace_cachefiles_wait_active(object, dentry, xobject);
189 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
190 
191 	if (fscache_object_is_live(&xobject->fscache)) {
192 		pr_err("\n");
193 		pr_err("Error: Unexpected object collision\n");
194 		cachefiles_printk_object(object, xobject);
195 	}
196 	atomic_inc(&xobject->usage);
197 	write_unlock(&cache->active_lock);
198 
199 	if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
200 		wait_queue_head_t *wq;
201 
202 		signed long timeout = 60 * HZ;
203 		wait_queue_entry_t wait;
204 		bool requeue;
205 
206 		/* if the object we're waiting for is queued for processing,
207 		 * then just put ourselves on the queue behind it */
208 		if (work_pending(&xobject->fscache.work)) {
209 			_debug("queue OBJ%x behind OBJ%x immediately",
210 			       object->fscache.debug_id,
211 			       xobject->fscache.debug_id);
212 			goto requeue;
213 		}
214 
215 		/* otherwise we sleep until either the object we're waiting for
216 		 * is done, or the fscache_object is congested */
217 		wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
218 		init_wait(&wait);
219 		requeue = false;
220 		do {
221 			prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
222 			if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
223 				break;
224 
225 			requeue = fscache_object_sleep_till_congested(&timeout);
226 		} while (timeout > 0 && !requeue);
227 		finish_wait(wq, &wait);
228 
229 		if (requeue &&
230 		    test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
231 			_debug("queue OBJ%x behind OBJ%x after wait",
232 			       object->fscache.debug_id,
233 			       xobject->fscache.debug_id);
234 			goto requeue;
235 		}
236 
237 		if (timeout <= 0) {
238 			pr_err("\n");
239 			pr_err("Error: Overlong wait for old active object to go away\n");
240 			cachefiles_printk_object(object, xobject);
241 			goto requeue;
242 		}
243 	}
244 
245 	ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
246 
247 	cache->cache.ops->put_object(&xobject->fscache, cachefiles_obj_put_wait_retry);
248 	goto try_again;
249 
250 requeue:
251 	cache->cache.ops->put_object(&xobject->fscache, cachefiles_obj_put_wait_timeo);
252 	_leave(" = -ETIMEDOUT");
253 	return -ETIMEDOUT;
254 }
255 
256 /*
257  * Mark an object as being inactive.
258  */
259 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
260 				     struct cachefiles_object *object,
261 				     blkcnt_t i_blocks)
262 {
263 	struct dentry *dentry = object->dentry;
264 	struct inode *inode = d_backing_inode(dentry);
265 
266 	trace_cachefiles_mark_inactive(object, dentry, inode);
267 
268 	write_lock(&cache->active_lock);
269 	rb_erase(&object->active_node, &cache->active_nodes);
270 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
271 	write_unlock(&cache->active_lock);
272 
273 	wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
274 
275 	/* This object can now be culled, so we need to let the daemon know
276 	 * that there is something it can remove if it needs to.
277 	 */
278 	atomic_long_add(i_blocks, &cache->b_released);
279 	if (atomic_inc_return(&cache->f_released))
280 		cachefiles_state_changed(cache);
281 }
282 
283 /*
284  * delete an object representation from the cache
285  * - file backed objects are unlinked
286  * - directory backed objects are stuffed into the graveyard for userspace to
287  *   delete
288  * - unlocks the directory mutex
289  */
290 static int cachefiles_bury_object(struct cachefiles_cache *cache,
291 				  struct cachefiles_object *object,
292 				  struct dentry *dir,
293 				  struct dentry *rep,
294 				  bool preemptive,
295 				  enum fscache_why_object_killed why)
296 {
297 	struct dentry *grave, *trap;
298 	struct path path, path_to_graveyard;
299 	char nbuffer[8 + 8 + 1];
300 	int ret;
301 
302 	_enter(",'%pd','%pd'", dir, rep);
303 
304 	_debug("remove %p from %p", rep, dir);
305 
306 	/* non-directories can just be unlinked */
307 	if (!d_is_dir(rep)) {
308 		_debug("unlink stale object");
309 
310 		path.mnt = cache->mnt;
311 		path.dentry = dir;
312 		ret = security_path_unlink(&path, rep);
313 		if (ret < 0) {
314 			cachefiles_io_error(cache, "Unlink security error");
315 		} else {
316 			trace_cachefiles_unlink(object, rep, why);
317 			ret = vfs_unlink(d_inode(dir), rep, NULL);
318 
319 			if (preemptive)
320 				cachefiles_mark_object_buried(cache, rep, why);
321 		}
322 
323 		inode_unlock(d_inode(dir));
324 
325 		if (ret == -EIO)
326 			cachefiles_io_error(cache, "Unlink failed");
327 
328 		_leave(" = %d", ret);
329 		return ret;
330 	}
331 
332 	/* directories have to be moved to the graveyard */
333 	_debug("move stale object to graveyard");
334 	inode_unlock(d_inode(dir));
335 
336 try_again:
337 	/* first step is to make up a grave dentry in the graveyard */
338 	sprintf(nbuffer, "%08x%08x",
339 		(uint32_t) get_seconds(),
340 		(uint32_t) atomic_inc_return(&cache->gravecounter));
341 
342 	/* do the multiway lock magic */
343 	trap = lock_rename(cache->graveyard, dir);
344 
345 	/* do some checks before getting the grave dentry */
346 	if (rep->d_parent != dir || IS_DEADDIR(d_inode(rep))) {
347 		/* the entry was probably culled when we dropped the parent dir
348 		 * lock */
349 		unlock_rename(cache->graveyard, dir);
350 		_leave(" = 0 [culled?]");
351 		return 0;
352 	}
353 
354 	if (!d_can_lookup(cache->graveyard)) {
355 		unlock_rename(cache->graveyard, dir);
356 		cachefiles_io_error(cache, "Graveyard no longer a directory");
357 		return -EIO;
358 	}
359 
360 	if (trap == rep) {
361 		unlock_rename(cache->graveyard, dir);
362 		cachefiles_io_error(cache, "May not make directory loop");
363 		return -EIO;
364 	}
365 
366 	if (d_mountpoint(rep)) {
367 		unlock_rename(cache->graveyard, dir);
368 		cachefiles_io_error(cache, "Mountpoint in cache");
369 		return -EIO;
370 	}
371 
372 	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
373 	if (IS_ERR(grave)) {
374 		unlock_rename(cache->graveyard, dir);
375 
376 		if (PTR_ERR(grave) == -ENOMEM) {
377 			_leave(" = -ENOMEM");
378 			return -ENOMEM;
379 		}
380 
381 		cachefiles_io_error(cache, "Lookup error %ld",
382 				    PTR_ERR(grave));
383 		return -EIO;
384 	}
385 
386 	if (d_is_positive(grave)) {
387 		unlock_rename(cache->graveyard, dir);
388 		dput(grave);
389 		grave = NULL;
390 		cond_resched();
391 		goto try_again;
392 	}
393 
394 	if (d_mountpoint(grave)) {
395 		unlock_rename(cache->graveyard, dir);
396 		dput(grave);
397 		cachefiles_io_error(cache, "Mountpoint in graveyard");
398 		return -EIO;
399 	}
400 
401 	/* target should not be an ancestor of source */
402 	if (trap == grave) {
403 		unlock_rename(cache->graveyard, dir);
404 		dput(grave);
405 		cachefiles_io_error(cache, "May not make directory loop");
406 		return -EIO;
407 	}
408 
409 	/* attempt the rename */
410 	path.mnt = cache->mnt;
411 	path.dentry = dir;
412 	path_to_graveyard.mnt = cache->mnt;
413 	path_to_graveyard.dentry = cache->graveyard;
414 	ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
415 	if (ret < 0) {
416 		cachefiles_io_error(cache, "Rename security error %d", ret);
417 	} else {
418 		trace_cachefiles_rename(object, rep, grave, why);
419 		ret = vfs_rename(d_inode(dir), rep,
420 				 d_inode(cache->graveyard), grave, NULL, 0);
421 		if (ret != 0 && ret != -ENOMEM)
422 			cachefiles_io_error(cache,
423 					    "Rename failed with error %d", ret);
424 
425 		if (preemptive)
426 			cachefiles_mark_object_buried(cache, rep, why);
427 	}
428 
429 	unlock_rename(cache->graveyard, dir);
430 	dput(grave);
431 	_leave(" = 0");
432 	return 0;
433 }
434 
435 /*
436  * delete an object representation from the cache
437  */
438 int cachefiles_delete_object(struct cachefiles_cache *cache,
439 			     struct cachefiles_object *object)
440 {
441 	struct dentry *dir;
442 	int ret;
443 
444 	_enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
445 
446 	ASSERT(object->dentry);
447 	ASSERT(d_backing_inode(object->dentry));
448 	ASSERT(object->dentry->d_parent);
449 
450 	dir = dget_parent(object->dentry);
451 
452 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
453 
454 	if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
455 		/* object allocation for the same key preemptively deleted this
456 		 * object's file so that it could create its own file */
457 		_debug("object preemptively buried");
458 		inode_unlock(d_inode(dir));
459 		ret = 0;
460 	} else {
461 		/* we need to check that our parent is _still_ our parent - it
462 		 * may have been renamed */
463 		if (dir == object->dentry->d_parent) {
464 			ret = cachefiles_bury_object(cache, object, dir,
465 						     object->dentry, false,
466 						     FSCACHE_OBJECT_WAS_RETIRED);
467 		} else {
468 			/* it got moved, presumably by cachefilesd culling it,
469 			 * so it's no longer in the key path and we can ignore
470 			 * it */
471 			inode_unlock(d_inode(dir));
472 			ret = 0;
473 		}
474 	}
475 
476 	dput(dir);
477 	_leave(" = %d", ret);
478 	return ret;
479 }
480 
481 /*
482  * walk from the parent object to the child object through the backing
483  * filesystem, creating directories as we go
484  */
485 int cachefiles_walk_to_object(struct cachefiles_object *parent,
486 			      struct cachefiles_object *object,
487 			      const char *key,
488 			      struct cachefiles_xattr *auxdata)
489 {
490 	struct cachefiles_cache *cache;
491 	struct dentry *dir, *next = NULL;
492 	struct inode *inode;
493 	struct path path;
494 	unsigned long start;
495 	const char *name;
496 	int ret, nlen;
497 
498 	_enter("OBJ%x{%p},OBJ%x,%s,",
499 	       parent->fscache.debug_id, parent->dentry,
500 	       object->fscache.debug_id, key);
501 
502 	cache = container_of(parent->fscache.cache,
503 			     struct cachefiles_cache, cache);
504 	path.mnt = cache->mnt;
505 
506 	ASSERT(parent->dentry);
507 	ASSERT(d_backing_inode(parent->dentry));
508 
509 	if (!(d_is_dir(parent->dentry))) {
510 		// TODO: convert file to dir
511 		_leave("looking up in none directory");
512 		return -ENOBUFS;
513 	}
514 
515 	dir = dget(parent->dentry);
516 
517 advance:
518 	/* attempt to transit the first directory component */
519 	name = key;
520 	nlen = strlen(key);
521 
522 	/* key ends in a double NUL */
523 	key = key + nlen + 1;
524 	if (!*key)
525 		key = NULL;
526 
527 lookup_again:
528 	/* search the current directory for the element name */
529 	_debug("lookup '%s'", name);
530 
531 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
532 
533 	start = jiffies;
534 	next = lookup_one_len(name, dir, nlen);
535 	cachefiles_hist(cachefiles_lookup_histogram, start);
536 	if (IS_ERR(next)) {
537 		trace_cachefiles_lookup(object, next, NULL);
538 		goto lookup_error;
539 	}
540 
541 	inode = d_backing_inode(next);
542 	trace_cachefiles_lookup(object, next, inode);
543 	_debug("next -> %p %s", next, inode ? "positive" : "negative");
544 
545 	if (!key)
546 		object->new = !inode;
547 
548 	/* if this element of the path doesn't exist, then the lookup phase
549 	 * failed, and we can release any readers in the certain knowledge that
550 	 * there's nothing for them to actually read */
551 	if (d_is_negative(next))
552 		fscache_object_lookup_negative(&object->fscache);
553 
554 	/* we need to create the object if it's negative */
555 	if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
556 		/* index objects and intervening tree levels must be subdirs */
557 		if (d_is_negative(next)) {
558 			ret = cachefiles_has_space(cache, 1, 0);
559 			if (ret < 0)
560 				goto no_space_error;
561 
562 			path.dentry = dir;
563 			ret = security_path_mkdir(&path, next, 0);
564 			if (ret < 0)
565 				goto create_error;
566 			start = jiffies;
567 			ret = vfs_mkdir(d_inode(dir), next, 0);
568 			cachefiles_hist(cachefiles_mkdir_histogram, start);
569 			if (!key)
570 				trace_cachefiles_mkdir(object, next, ret);
571 			if (ret < 0)
572 				goto create_error;
573 
574 			if (unlikely(d_unhashed(next))) {
575 				dput(next);
576 				inode_unlock(d_inode(dir));
577 				goto lookup_again;
578 			}
579 			ASSERT(d_backing_inode(next));
580 
581 			_debug("mkdir -> %p{%p{ino=%lu}}",
582 			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
583 
584 		} else if (!d_can_lookup(next)) {
585 			pr_err("inode %lu is not a directory\n",
586 			       d_backing_inode(next)->i_ino);
587 			ret = -ENOBUFS;
588 			goto error;
589 		}
590 
591 	} else {
592 		/* non-index objects start out life as files */
593 		if (d_is_negative(next)) {
594 			ret = cachefiles_has_space(cache, 1, 0);
595 			if (ret < 0)
596 				goto no_space_error;
597 
598 			path.dentry = dir;
599 			ret = security_path_mknod(&path, next, S_IFREG, 0);
600 			if (ret < 0)
601 				goto create_error;
602 			start = jiffies;
603 			ret = vfs_create(d_inode(dir), next, S_IFREG, true);
604 			cachefiles_hist(cachefiles_create_histogram, start);
605 			trace_cachefiles_create(object, next, ret);
606 			if (ret < 0)
607 				goto create_error;
608 
609 			ASSERT(d_backing_inode(next));
610 
611 			_debug("create -> %p{%p{ino=%lu}}",
612 			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
613 
614 		} else if (!d_can_lookup(next) &&
615 			   !d_is_reg(next)
616 			   ) {
617 			pr_err("inode %lu is not a file or directory\n",
618 			       d_backing_inode(next)->i_ino);
619 			ret = -ENOBUFS;
620 			goto error;
621 		}
622 	}
623 
624 	/* process the next component */
625 	if (key) {
626 		_debug("advance");
627 		inode_unlock(d_inode(dir));
628 		dput(dir);
629 		dir = next;
630 		next = NULL;
631 		goto advance;
632 	}
633 
634 	/* we've found the object we were looking for */
635 	object->dentry = next;
636 
637 	/* if we've found that the terminal object exists, then we need to
638 	 * check its attributes and delete it if it's out of date */
639 	if (!object->new) {
640 		_debug("validate '%pd'", next);
641 
642 		ret = cachefiles_check_object_xattr(object, auxdata);
643 		if (ret == -ESTALE) {
644 			/* delete the object (the deleter drops the directory
645 			 * mutex) */
646 			object->dentry = NULL;
647 
648 			ret = cachefiles_bury_object(cache, object, dir, next,
649 						     true,
650 						     FSCACHE_OBJECT_IS_STALE);
651 			dput(next);
652 			next = NULL;
653 
654 			if (ret < 0)
655 				goto delete_error;
656 
657 			_debug("redo lookup");
658 			fscache_object_retrying_stale(&object->fscache);
659 			goto lookup_again;
660 		}
661 	}
662 
663 	/* note that we're now using this object */
664 	ret = cachefiles_mark_object_active(cache, object);
665 
666 	inode_unlock(d_inode(dir));
667 	dput(dir);
668 	dir = NULL;
669 
670 	if (ret == -ETIMEDOUT)
671 		goto mark_active_timed_out;
672 
673 	_debug("=== OBTAINED_OBJECT ===");
674 
675 	if (object->new) {
676 		/* attach data to a newly constructed terminal object */
677 		ret = cachefiles_set_object_xattr(object, auxdata);
678 		if (ret < 0)
679 			goto check_error;
680 	} else {
681 		/* always update the atime on an object we've just looked up
682 		 * (this is used to keep track of culling, and atimes are only
683 		 * updated by read, write and readdir but not lookup or
684 		 * open) */
685 		path.dentry = next;
686 		touch_atime(&path);
687 	}
688 
689 	/* open a file interface onto a data file */
690 	if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
691 		if (d_is_reg(object->dentry)) {
692 			const struct address_space_operations *aops;
693 
694 			ret = -EPERM;
695 			aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
696 			if (!aops->bmap)
697 				goto check_error;
698 			if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
699 				goto check_error;
700 
701 			object->backer = object->dentry;
702 		} else {
703 			BUG(); // TODO: open file in data-class subdir
704 		}
705 	}
706 
707 	object->new = 0;
708 	fscache_obtained_object(&object->fscache);
709 
710 	_leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
711 	return 0;
712 
713 no_space_error:
714 	fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
715 create_error:
716 	_debug("create error %d", ret);
717 	if (ret == -EIO)
718 		cachefiles_io_error(cache, "Create/mkdir failed");
719 	goto error;
720 
721 mark_active_timed_out:
722 	_debug("mark active timed out");
723 	goto release_dentry;
724 
725 check_error:
726 	_debug("check error %d", ret);
727 	cachefiles_mark_object_inactive(
728 		cache, object, d_backing_inode(object->dentry)->i_blocks);
729 release_dentry:
730 	dput(object->dentry);
731 	object->dentry = NULL;
732 	goto error_out;
733 
734 delete_error:
735 	_debug("delete error %d", ret);
736 	goto error_out2;
737 
738 lookup_error:
739 	_debug("lookup error %ld", PTR_ERR(next));
740 	ret = PTR_ERR(next);
741 	if (ret == -EIO)
742 		cachefiles_io_error(cache, "Lookup failed");
743 	next = NULL;
744 error:
745 	inode_unlock(d_inode(dir));
746 	dput(next);
747 error_out2:
748 	dput(dir);
749 error_out:
750 	_leave(" = error %d", -ret);
751 	return ret;
752 }
753 
754 /*
755  * get a subdirectory
756  */
757 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
758 					struct dentry *dir,
759 					const char *dirname)
760 {
761 	struct dentry *subdir;
762 	unsigned long start;
763 	struct path path;
764 	int ret;
765 
766 	_enter(",,%s", dirname);
767 
768 	/* search the current directory for the element name */
769 	inode_lock(d_inode(dir));
770 
771 retry:
772 	start = jiffies;
773 	subdir = lookup_one_len(dirname, dir, strlen(dirname));
774 	cachefiles_hist(cachefiles_lookup_histogram, start);
775 	if (IS_ERR(subdir)) {
776 		if (PTR_ERR(subdir) == -ENOMEM)
777 			goto nomem_d_alloc;
778 		goto lookup_error;
779 	}
780 
781 	_debug("subdir -> %p %s",
782 	       subdir, d_backing_inode(subdir) ? "positive" : "negative");
783 
784 	/* we need to create the subdir if it doesn't exist yet */
785 	if (d_is_negative(subdir)) {
786 		ret = cachefiles_has_space(cache, 1, 0);
787 		if (ret < 0)
788 			goto mkdir_error;
789 
790 		_debug("attempt mkdir");
791 
792 		path.mnt = cache->mnt;
793 		path.dentry = dir;
794 		ret = security_path_mkdir(&path, subdir, 0700);
795 		if (ret < 0)
796 			goto mkdir_error;
797 		ret = vfs_mkdir(d_inode(dir), subdir, 0700);
798 		if (ret < 0)
799 			goto mkdir_error;
800 
801 		if (unlikely(d_unhashed(subdir))) {
802 			dput(subdir);
803 			goto retry;
804 		}
805 		ASSERT(d_backing_inode(subdir));
806 
807 		_debug("mkdir -> %p{%p{ino=%lu}}",
808 		       subdir,
809 		       d_backing_inode(subdir),
810 		       d_backing_inode(subdir)->i_ino);
811 	}
812 
813 	inode_unlock(d_inode(dir));
814 
815 	/* we need to make sure the subdir is a directory */
816 	ASSERT(d_backing_inode(subdir));
817 
818 	if (!d_can_lookup(subdir)) {
819 		pr_err("%s is not a directory\n", dirname);
820 		ret = -EIO;
821 		goto check_error;
822 	}
823 
824 	ret = -EPERM;
825 	if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
826 	    !d_backing_inode(subdir)->i_op->lookup ||
827 	    !d_backing_inode(subdir)->i_op->mkdir ||
828 	    !d_backing_inode(subdir)->i_op->create ||
829 	    !d_backing_inode(subdir)->i_op->rename ||
830 	    !d_backing_inode(subdir)->i_op->rmdir ||
831 	    !d_backing_inode(subdir)->i_op->unlink)
832 		goto check_error;
833 
834 	_leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
835 	return subdir;
836 
837 check_error:
838 	dput(subdir);
839 	_leave(" = %d [check]", ret);
840 	return ERR_PTR(ret);
841 
842 mkdir_error:
843 	inode_unlock(d_inode(dir));
844 	dput(subdir);
845 	pr_err("mkdir %s failed with error %d\n", dirname, ret);
846 	return ERR_PTR(ret);
847 
848 lookup_error:
849 	inode_unlock(d_inode(dir));
850 	ret = PTR_ERR(subdir);
851 	pr_err("Lookup %s failed with error %d\n", dirname, ret);
852 	return ERR_PTR(ret);
853 
854 nomem_d_alloc:
855 	inode_unlock(d_inode(dir));
856 	_leave(" = -ENOMEM");
857 	return ERR_PTR(-ENOMEM);
858 }
859 
860 /*
861  * find out if an object is in use or not
862  * - if finds object and it's not in use:
863  *   - returns a pointer to the object and a reference on it
864  *   - returns with the directory locked
865  */
866 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
867 					      struct dentry *dir,
868 					      char *filename)
869 {
870 	struct cachefiles_object *object;
871 	struct rb_node *_n;
872 	struct dentry *victim;
873 	unsigned long start;
874 	int ret;
875 
876 	//_enter(",%pd/,%s",
877 	//       dir, filename);
878 
879 	/* look up the victim */
880 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
881 
882 	start = jiffies;
883 	victim = lookup_one_len(filename, dir, strlen(filename));
884 	cachefiles_hist(cachefiles_lookup_histogram, start);
885 	if (IS_ERR(victim))
886 		goto lookup_error;
887 
888 	//_debug("victim -> %p %s",
889 	//       victim, d_backing_inode(victim) ? "positive" : "negative");
890 
891 	/* if the object is no longer there then we probably retired the object
892 	 * at the netfs's request whilst the cull was in progress
893 	 */
894 	if (d_is_negative(victim)) {
895 		inode_unlock(d_inode(dir));
896 		dput(victim);
897 		_leave(" = -ENOENT [absent]");
898 		return ERR_PTR(-ENOENT);
899 	}
900 
901 	/* check to see if we're using this object */
902 	read_lock(&cache->active_lock);
903 
904 	_n = cache->active_nodes.rb_node;
905 
906 	while (_n) {
907 		object = rb_entry(_n, struct cachefiles_object, active_node);
908 
909 		if (object->dentry > victim)
910 			_n = _n->rb_left;
911 		else if (object->dentry < victim)
912 			_n = _n->rb_right;
913 		else
914 			goto object_in_use;
915 	}
916 
917 	read_unlock(&cache->active_lock);
918 
919 	//_leave(" = %p", victim);
920 	return victim;
921 
922 object_in_use:
923 	read_unlock(&cache->active_lock);
924 	inode_unlock(d_inode(dir));
925 	dput(victim);
926 	//_leave(" = -EBUSY [in use]");
927 	return ERR_PTR(-EBUSY);
928 
929 lookup_error:
930 	inode_unlock(d_inode(dir));
931 	ret = PTR_ERR(victim);
932 	if (ret == -ENOENT) {
933 		/* file or dir now absent - probably retired by netfs */
934 		_leave(" = -ESTALE [absent]");
935 		return ERR_PTR(-ESTALE);
936 	}
937 
938 	if (ret == -EIO) {
939 		cachefiles_io_error(cache, "Lookup failed");
940 	} else if (ret != -ENOMEM) {
941 		pr_err("Internal error: %d\n", ret);
942 		ret = -EIO;
943 	}
944 
945 	_leave(" = %d", ret);
946 	return ERR_PTR(ret);
947 }
948 
949 /*
950  * cull an object if it's not in use
951  * - called only by cache manager daemon
952  */
953 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
954 		    char *filename)
955 {
956 	struct dentry *victim;
957 	int ret;
958 
959 	_enter(",%pd/,%s", dir, filename);
960 
961 	victim = cachefiles_check_active(cache, dir, filename);
962 	if (IS_ERR(victim))
963 		return PTR_ERR(victim);
964 
965 	_debug("victim -> %p %s",
966 	       victim, d_backing_inode(victim) ? "positive" : "negative");
967 
968 	/* okay... the victim is not being used so we can cull it
969 	 * - start by marking it as stale
970 	 */
971 	_debug("victim is cullable");
972 
973 	ret = cachefiles_remove_object_xattr(cache, victim);
974 	if (ret < 0)
975 		goto error_unlock;
976 
977 	/*  actually remove the victim (drops the dir mutex) */
978 	_debug("bury");
979 
980 	ret = cachefiles_bury_object(cache, NULL, dir, victim, false,
981 				     FSCACHE_OBJECT_WAS_CULLED);
982 	if (ret < 0)
983 		goto error;
984 
985 	dput(victim);
986 	_leave(" = 0");
987 	return 0;
988 
989 error_unlock:
990 	inode_unlock(d_inode(dir));
991 error:
992 	dput(victim);
993 	if (ret == -ENOENT) {
994 		/* file or dir now absent - probably retired by netfs */
995 		_leave(" = -ESTALE [absent]");
996 		return -ESTALE;
997 	}
998 
999 	if (ret != -ENOMEM) {
1000 		pr_err("Internal error: %d\n", ret);
1001 		ret = -EIO;
1002 	}
1003 
1004 	_leave(" = %d", ret);
1005 	return ret;
1006 }
1007 
1008 /*
1009  * find out if an object is in use or not
1010  * - called only by cache manager daemon
1011  * - returns -EBUSY or 0 to indicate whether an object is in use or not
1012  */
1013 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
1014 			    char *filename)
1015 {
1016 	struct dentry *victim;
1017 
1018 	//_enter(",%pd/,%s",
1019 	//       dir, filename);
1020 
1021 	victim = cachefiles_check_active(cache, dir, filename);
1022 	if (IS_ERR(victim))
1023 		return PTR_ERR(victim);
1024 
1025 	inode_unlock(d_inode(dir));
1026 	dput(victim);
1027 	//_leave(" = 0");
1028 	return 0;
1029 }
1030