xref: /linux/fs/cachefiles/namei.c (revision 8f8d5745bb520c76b81abef4a2cb3023d0313bfd)
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,
248 		(enum fscache_obj_ref_trace)cachefiles_obj_put_wait_retry);
249 	goto try_again;
250 
251 requeue:
252 	cache->cache.ops->put_object(&xobject->fscache,
253 		(enum fscache_obj_ref_trace)cachefiles_obj_put_wait_timeo);
254 	_leave(" = -ETIMEDOUT");
255 	return -ETIMEDOUT;
256 }
257 
258 /*
259  * Mark an object as being inactive.
260  */
261 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
262 				     struct cachefiles_object *object,
263 				     blkcnt_t i_blocks)
264 {
265 	struct dentry *dentry = object->dentry;
266 	struct inode *inode = d_backing_inode(dentry);
267 
268 	trace_cachefiles_mark_inactive(object, dentry, inode);
269 
270 	write_lock(&cache->active_lock);
271 	rb_erase(&object->active_node, &cache->active_nodes);
272 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
273 	write_unlock(&cache->active_lock);
274 
275 	wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
276 
277 	/* This object can now be culled, so we need to let the daemon know
278 	 * that there is something it can remove if it needs to.
279 	 */
280 	atomic_long_add(i_blocks, &cache->b_released);
281 	if (atomic_inc_return(&cache->f_released))
282 		cachefiles_state_changed(cache);
283 }
284 
285 /*
286  * delete an object representation from the cache
287  * - file backed objects are unlinked
288  * - directory backed objects are stuffed into the graveyard for userspace to
289  *   delete
290  * - unlocks the directory mutex
291  */
292 static int cachefiles_bury_object(struct cachefiles_cache *cache,
293 				  struct cachefiles_object *object,
294 				  struct dentry *dir,
295 				  struct dentry *rep,
296 				  bool preemptive,
297 				  enum fscache_why_object_killed why)
298 {
299 	struct dentry *grave, *trap;
300 	struct path path, path_to_graveyard;
301 	char nbuffer[8 + 8 + 1];
302 	int ret;
303 
304 	_enter(",'%pd','%pd'", dir, rep);
305 
306 	_debug("remove %p from %p", rep, dir);
307 
308 	/* non-directories can just be unlinked */
309 	if (!d_is_dir(rep)) {
310 		_debug("unlink stale object");
311 
312 		path.mnt = cache->mnt;
313 		path.dentry = dir;
314 		ret = security_path_unlink(&path, rep);
315 		if (ret < 0) {
316 			cachefiles_io_error(cache, "Unlink security error");
317 		} else {
318 			trace_cachefiles_unlink(object, rep, why);
319 			ret = vfs_unlink(d_inode(dir), rep, NULL);
320 
321 			if (preemptive)
322 				cachefiles_mark_object_buried(cache, rep, why);
323 		}
324 
325 		inode_unlock(d_inode(dir));
326 
327 		if (ret == -EIO)
328 			cachefiles_io_error(cache, "Unlink failed");
329 
330 		_leave(" = %d", ret);
331 		return ret;
332 	}
333 
334 	/* directories have to be moved to the graveyard */
335 	_debug("move stale object to graveyard");
336 	inode_unlock(d_inode(dir));
337 
338 try_again:
339 	/* first step is to make up a grave dentry in the graveyard */
340 	sprintf(nbuffer, "%08x%08x",
341 		(uint32_t) ktime_get_real_seconds(),
342 		(uint32_t) atomic_inc_return(&cache->gravecounter));
343 
344 	/* do the multiway lock magic */
345 	trap = lock_rename(cache->graveyard, dir);
346 
347 	/* do some checks before getting the grave dentry */
348 	if (rep->d_parent != dir || IS_DEADDIR(d_inode(rep))) {
349 		/* the entry was probably culled when we dropped the parent dir
350 		 * lock */
351 		unlock_rename(cache->graveyard, dir);
352 		_leave(" = 0 [culled?]");
353 		return 0;
354 	}
355 
356 	if (!d_can_lookup(cache->graveyard)) {
357 		unlock_rename(cache->graveyard, dir);
358 		cachefiles_io_error(cache, "Graveyard no longer a directory");
359 		return -EIO;
360 	}
361 
362 	if (trap == rep) {
363 		unlock_rename(cache->graveyard, dir);
364 		cachefiles_io_error(cache, "May not make directory loop");
365 		return -EIO;
366 	}
367 
368 	if (d_mountpoint(rep)) {
369 		unlock_rename(cache->graveyard, dir);
370 		cachefiles_io_error(cache, "Mountpoint in cache");
371 		return -EIO;
372 	}
373 
374 	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
375 	if (IS_ERR(grave)) {
376 		unlock_rename(cache->graveyard, dir);
377 
378 		if (PTR_ERR(grave) == -ENOMEM) {
379 			_leave(" = -ENOMEM");
380 			return -ENOMEM;
381 		}
382 
383 		cachefiles_io_error(cache, "Lookup error %ld",
384 				    PTR_ERR(grave));
385 		return -EIO;
386 	}
387 
388 	if (d_is_positive(grave)) {
389 		unlock_rename(cache->graveyard, dir);
390 		dput(grave);
391 		grave = NULL;
392 		cond_resched();
393 		goto try_again;
394 	}
395 
396 	if (d_mountpoint(grave)) {
397 		unlock_rename(cache->graveyard, dir);
398 		dput(grave);
399 		cachefiles_io_error(cache, "Mountpoint in graveyard");
400 		return -EIO;
401 	}
402 
403 	/* target should not be an ancestor of source */
404 	if (trap == grave) {
405 		unlock_rename(cache->graveyard, dir);
406 		dput(grave);
407 		cachefiles_io_error(cache, "May not make directory loop");
408 		return -EIO;
409 	}
410 
411 	/* attempt the rename */
412 	path.mnt = cache->mnt;
413 	path.dentry = dir;
414 	path_to_graveyard.mnt = cache->mnt;
415 	path_to_graveyard.dentry = cache->graveyard;
416 	ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
417 	if (ret < 0) {
418 		cachefiles_io_error(cache, "Rename security error %d", ret);
419 	} else {
420 		trace_cachefiles_rename(object, rep, grave, why);
421 		ret = vfs_rename(d_inode(dir), rep,
422 				 d_inode(cache->graveyard), grave, NULL, 0);
423 		if (ret != 0 && ret != -ENOMEM)
424 			cachefiles_io_error(cache,
425 					    "Rename failed with error %d", ret);
426 
427 		if (preemptive)
428 			cachefiles_mark_object_buried(cache, rep, why);
429 	}
430 
431 	unlock_rename(cache->graveyard, dir);
432 	dput(grave);
433 	_leave(" = 0");
434 	return 0;
435 }
436 
437 /*
438  * delete an object representation from the cache
439  */
440 int cachefiles_delete_object(struct cachefiles_cache *cache,
441 			     struct cachefiles_object *object)
442 {
443 	struct dentry *dir;
444 	int ret;
445 
446 	_enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
447 
448 	ASSERT(object->dentry);
449 	ASSERT(d_backing_inode(object->dentry));
450 	ASSERT(object->dentry->d_parent);
451 
452 	dir = dget_parent(object->dentry);
453 
454 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
455 
456 	if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
457 		/* object allocation for the same key preemptively deleted this
458 		 * object's file so that it could create its own file */
459 		_debug("object preemptively buried");
460 		inode_unlock(d_inode(dir));
461 		ret = 0;
462 	} else {
463 		/* we need to check that our parent is _still_ our parent - it
464 		 * may have been renamed */
465 		if (dir == object->dentry->d_parent) {
466 			ret = cachefiles_bury_object(cache, object, dir,
467 						     object->dentry, false,
468 						     FSCACHE_OBJECT_WAS_RETIRED);
469 		} else {
470 			/* it got moved, presumably by cachefilesd culling it,
471 			 * so it's no longer in the key path and we can ignore
472 			 * it */
473 			inode_unlock(d_inode(dir));
474 			ret = 0;
475 		}
476 	}
477 
478 	dput(dir);
479 	_leave(" = %d", ret);
480 	return ret;
481 }
482 
483 /*
484  * walk from the parent object to the child object through the backing
485  * filesystem, creating directories as we go
486  */
487 int cachefiles_walk_to_object(struct cachefiles_object *parent,
488 			      struct cachefiles_object *object,
489 			      const char *key,
490 			      struct cachefiles_xattr *auxdata)
491 {
492 	struct cachefiles_cache *cache;
493 	struct dentry *dir, *next = NULL;
494 	struct inode *inode;
495 	struct path path;
496 	unsigned long start;
497 	const char *name;
498 	int ret, nlen;
499 
500 	_enter("OBJ%x{%p},OBJ%x,%s,",
501 	       parent->fscache.debug_id, parent->dentry,
502 	       object->fscache.debug_id, key);
503 
504 	cache = container_of(parent->fscache.cache,
505 			     struct cachefiles_cache, cache);
506 	path.mnt = cache->mnt;
507 
508 	ASSERT(parent->dentry);
509 	ASSERT(d_backing_inode(parent->dentry));
510 
511 	if (!(d_is_dir(parent->dentry))) {
512 		// TODO: convert file to dir
513 		_leave("looking up in none directory");
514 		return -ENOBUFS;
515 	}
516 
517 	dir = dget(parent->dentry);
518 
519 advance:
520 	/* attempt to transit the first directory component */
521 	name = key;
522 	nlen = strlen(key);
523 
524 	/* key ends in a double NUL */
525 	key = key + nlen + 1;
526 	if (!*key)
527 		key = NULL;
528 
529 lookup_again:
530 	/* search the current directory for the element name */
531 	_debug("lookup '%s'", name);
532 
533 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
534 
535 	start = jiffies;
536 	next = lookup_one_len(name, dir, nlen);
537 	cachefiles_hist(cachefiles_lookup_histogram, start);
538 	if (IS_ERR(next)) {
539 		trace_cachefiles_lookup(object, next, NULL);
540 		goto lookup_error;
541 	}
542 
543 	inode = d_backing_inode(next);
544 	trace_cachefiles_lookup(object, next, inode);
545 	_debug("next -> %p %s", next, inode ? "positive" : "negative");
546 
547 	if (!key)
548 		object->new = !inode;
549 
550 	/* if this element of the path doesn't exist, then the lookup phase
551 	 * failed, and we can release any readers in the certain knowledge that
552 	 * there's nothing for them to actually read */
553 	if (d_is_negative(next))
554 		fscache_object_lookup_negative(&object->fscache);
555 
556 	/* we need to create the object if it's negative */
557 	if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
558 		/* index objects and intervening tree levels must be subdirs */
559 		if (d_is_negative(next)) {
560 			ret = cachefiles_has_space(cache, 1, 0);
561 			if (ret < 0)
562 				goto no_space_error;
563 
564 			path.dentry = dir;
565 			ret = security_path_mkdir(&path, next, 0);
566 			if (ret < 0)
567 				goto create_error;
568 			start = jiffies;
569 			ret = vfs_mkdir(d_inode(dir), next, 0);
570 			cachefiles_hist(cachefiles_mkdir_histogram, start);
571 			if (!key)
572 				trace_cachefiles_mkdir(object, next, ret);
573 			if (ret < 0)
574 				goto create_error;
575 
576 			if (unlikely(d_unhashed(next))) {
577 				dput(next);
578 				inode_unlock(d_inode(dir));
579 				goto lookup_again;
580 			}
581 			ASSERT(d_backing_inode(next));
582 
583 			_debug("mkdir -> %p{%p{ino=%lu}}",
584 			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
585 
586 		} else if (!d_can_lookup(next)) {
587 			pr_err("inode %lu is not a directory\n",
588 			       d_backing_inode(next)->i_ino);
589 			ret = -ENOBUFS;
590 			goto error;
591 		}
592 
593 	} else {
594 		/* non-index objects start out life as files */
595 		if (d_is_negative(next)) {
596 			ret = cachefiles_has_space(cache, 1, 0);
597 			if (ret < 0)
598 				goto no_space_error;
599 
600 			path.dentry = dir;
601 			ret = security_path_mknod(&path, next, S_IFREG, 0);
602 			if (ret < 0)
603 				goto create_error;
604 			start = jiffies;
605 			ret = vfs_create(d_inode(dir), next, S_IFREG, true);
606 			cachefiles_hist(cachefiles_create_histogram, start);
607 			trace_cachefiles_create(object, next, ret);
608 			if (ret < 0)
609 				goto create_error;
610 
611 			ASSERT(d_backing_inode(next));
612 
613 			_debug("create -> %p{%p{ino=%lu}}",
614 			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
615 
616 		} else if (!d_can_lookup(next) &&
617 			   !d_is_reg(next)
618 			   ) {
619 			pr_err("inode %lu is not a file or directory\n",
620 			       d_backing_inode(next)->i_ino);
621 			ret = -ENOBUFS;
622 			goto error;
623 		}
624 	}
625 
626 	/* process the next component */
627 	if (key) {
628 		_debug("advance");
629 		inode_unlock(d_inode(dir));
630 		dput(dir);
631 		dir = next;
632 		next = NULL;
633 		goto advance;
634 	}
635 
636 	/* we've found the object we were looking for */
637 	object->dentry = next;
638 
639 	/* if we've found that the terminal object exists, then we need to
640 	 * check its attributes and delete it if it's out of date */
641 	if (!object->new) {
642 		_debug("validate '%pd'", next);
643 
644 		ret = cachefiles_check_object_xattr(object, auxdata);
645 		if (ret == -ESTALE) {
646 			/* delete the object (the deleter drops the directory
647 			 * mutex) */
648 			object->dentry = NULL;
649 
650 			ret = cachefiles_bury_object(cache, object, dir, next,
651 						     true,
652 						     FSCACHE_OBJECT_IS_STALE);
653 			dput(next);
654 			next = NULL;
655 
656 			if (ret < 0)
657 				goto delete_error;
658 
659 			_debug("redo lookup");
660 			fscache_object_retrying_stale(&object->fscache);
661 			goto lookup_again;
662 		}
663 	}
664 
665 	/* note that we're now using this object */
666 	ret = cachefiles_mark_object_active(cache, object);
667 
668 	inode_unlock(d_inode(dir));
669 	dput(dir);
670 	dir = NULL;
671 
672 	if (ret == -ETIMEDOUT)
673 		goto mark_active_timed_out;
674 
675 	_debug("=== OBTAINED_OBJECT ===");
676 
677 	if (object->new) {
678 		/* attach data to a newly constructed terminal object */
679 		ret = cachefiles_set_object_xattr(object, auxdata);
680 		if (ret < 0)
681 			goto check_error;
682 	} else {
683 		/* always update the atime on an object we've just looked up
684 		 * (this is used to keep track of culling, and atimes are only
685 		 * updated by read, write and readdir but not lookup or
686 		 * open) */
687 		path.dentry = next;
688 		touch_atime(&path);
689 	}
690 
691 	/* open a file interface onto a data file */
692 	if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
693 		if (d_is_reg(object->dentry)) {
694 			const struct address_space_operations *aops;
695 
696 			ret = -EPERM;
697 			aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
698 			if (!aops->bmap)
699 				goto check_error;
700 			if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
701 				goto check_error;
702 
703 			object->backer = object->dentry;
704 		} else {
705 			BUG(); // TODO: open file in data-class subdir
706 		}
707 	}
708 
709 	object->new = 0;
710 	fscache_obtained_object(&object->fscache);
711 
712 	_leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
713 	return 0;
714 
715 no_space_error:
716 	fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
717 create_error:
718 	_debug("create error %d", ret);
719 	if (ret == -EIO)
720 		cachefiles_io_error(cache, "Create/mkdir failed");
721 	goto error;
722 
723 mark_active_timed_out:
724 	_debug("mark active timed out");
725 	goto release_dentry;
726 
727 check_error:
728 	_debug("check error %d", ret);
729 	cachefiles_mark_object_inactive(
730 		cache, object, d_backing_inode(object->dentry)->i_blocks);
731 release_dentry:
732 	dput(object->dentry);
733 	object->dentry = NULL;
734 	goto error_out;
735 
736 delete_error:
737 	_debug("delete error %d", ret);
738 	goto error_out2;
739 
740 lookup_error:
741 	_debug("lookup error %ld", PTR_ERR(next));
742 	ret = PTR_ERR(next);
743 	if (ret == -EIO)
744 		cachefiles_io_error(cache, "Lookup failed");
745 	next = NULL;
746 error:
747 	inode_unlock(d_inode(dir));
748 	dput(next);
749 error_out2:
750 	dput(dir);
751 error_out:
752 	_leave(" = error %d", -ret);
753 	return ret;
754 }
755 
756 /*
757  * get a subdirectory
758  */
759 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
760 					struct dentry *dir,
761 					const char *dirname)
762 {
763 	struct dentry *subdir;
764 	unsigned long start;
765 	struct path path;
766 	int ret;
767 
768 	_enter(",,%s", dirname);
769 
770 	/* search the current directory for the element name */
771 	inode_lock(d_inode(dir));
772 
773 retry:
774 	start = jiffies;
775 	subdir = lookup_one_len(dirname, dir, strlen(dirname));
776 	cachefiles_hist(cachefiles_lookup_histogram, start);
777 	if (IS_ERR(subdir)) {
778 		if (PTR_ERR(subdir) == -ENOMEM)
779 			goto nomem_d_alloc;
780 		goto lookup_error;
781 	}
782 
783 	_debug("subdir -> %p %s",
784 	       subdir, d_backing_inode(subdir) ? "positive" : "negative");
785 
786 	/* we need to create the subdir if it doesn't exist yet */
787 	if (d_is_negative(subdir)) {
788 		ret = cachefiles_has_space(cache, 1, 0);
789 		if (ret < 0)
790 			goto mkdir_error;
791 
792 		_debug("attempt mkdir");
793 
794 		path.mnt = cache->mnt;
795 		path.dentry = dir;
796 		ret = security_path_mkdir(&path, subdir, 0700);
797 		if (ret < 0)
798 			goto mkdir_error;
799 		ret = vfs_mkdir(d_inode(dir), subdir, 0700);
800 		if (ret < 0)
801 			goto mkdir_error;
802 
803 		if (unlikely(d_unhashed(subdir))) {
804 			dput(subdir);
805 			goto retry;
806 		}
807 		ASSERT(d_backing_inode(subdir));
808 
809 		_debug("mkdir -> %p{%p{ino=%lu}}",
810 		       subdir,
811 		       d_backing_inode(subdir),
812 		       d_backing_inode(subdir)->i_ino);
813 	}
814 
815 	inode_unlock(d_inode(dir));
816 
817 	/* we need to make sure the subdir is a directory */
818 	ASSERT(d_backing_inode(subdir));
819 
820 	if (!d_can_lookup(subdir)) {
821 		pr_err("%s is not a directory\n", dirname);
822 		ret = -EIO;
823 		goto check_error;
824 	}
825 
826 	ret = -EPERM;
827 	if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
828 	    !d_backing_inode(subdir)->i_op->lookup ||
829 	    !d_backing_inode(subdir)->i_op->mkdir ||
830 	    !d_backing_inode(subdir)->i_op->create ||
831 	    !d_backing_inode(subdir)->i_op->rename ||
832 	    !d_backing_inode(subdir)->i_op->rmdir ||
833 	    !d_backing_inode(subdir)->i_op->unlink)
834 		goto check_error;
835 
836 	_leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
837 	return subdir;
838 
839 check_error:
840 	dput(subdir);
841 	_leave(" = %d [check]", ret);
842 	return ERR_PTR(ret);
843 
844 mkdir_error:
845 	inode_unlock(d_inode(dir));
846 	dput(subdir);
847 	pr_err("mkdir %s failed with error %d\n", dirname, ret);
848 	return ERR_PTR(ret);
849 
850 lookup_error:
851 	inode_unlock(d_inode(dir));
852 	ret = PTR_ERR(subdir);
853 	pr_err("Lookup %s failed with error %d\n", dirname, ret);
854 	return ERR_PTR(ret);
855 
856 nomem_d_alloc:
857 	inode_unlock(d_inode(dir));
858 	_leave(" = -ENOMEM");
859 	return ERR_PTR(-ENOMEM);
860 }
861 
862 /*
863  * find out if an object is in use or not
864  * - if finds object and it's not in use:
865  *   - returns a pointer to the object and a reference on it
866  *   - returns with the directory locked
867  */
868 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
869 					      struct dentry *dir,
870 					      char *filename)
871 {
872 	struct cachefiles_object *object;
873 	struct rb_node *_n;
874 	struct dentry *victim;
875 	unsigned long start;
876 	int ret;
877 
878 	//_enter(",%pd/,%s",
879 	//       dir, filename);
880 
881 	/* look up the victim */
882 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
883 
884 	start = jiffies;
885 	victim = lookup_one_len(filename, dir, strlen(filename));
886 	cachefiles_hist(cachefiles_lookup_histogram, start);
887 	if (IS_ERR(victim))
888 		goto lookup_error;
889 
890 	//_debug("victim -> %p %s",
891 	//       victim, d_backing_inode(victim) ? "positive" : "negative");
892 
893 	/* if the object is no longer there then we probably retired the object
894 	 * at the netfs's request whilst the cull was in progress
895 	 */
896 	if (d_is_negative(victim)) {
897 		inode_unlock(d_inode(dir));
898 		dput(victim);
899 		_leave(" = -ENOENT [absent]");
900 		return ERR_PTR(-ENOENT);
901 	}
902 
903 	/* check to see if we're using this object */
904 	read_lock(&cache->active_lock);
905 
906 	_n = cache->active_nodes.rb_node;
907 
908 	while (_n) {
909 		object = rb_entry(_n, struct cachefiles_object, active_node);
910 
911 		if (object->dentry > victim)
912 			_n = _n->rb_left;
913 		else if (object->dentry < victim)
914 			_n = _n->rb_right;
915 		else
916 			goto object_in_use;
917 	}
918 
919 	read_unlock(&cache->active_lock);
920 
921 	//_leave(" = %p", victim);
922 	return victim;
923 
924 object_in_use:
925 	read_unlock(&cache->active_lock);
926 	inode_unlock(d_inode(dir));
927 	dput(victim);
928 	//_leave(" = -EBUSY [in use]");
929 	return ERR_PTR(-EBUSY);
930 
931 lookup_error:
932 	inode_unlock(d_inode(dir));
933 	ret = PTR_ERR(victim);
934 	if (ret == -ENOENT) {
935 		/* file or dir now absent - probably retired by netfs */
936 		_leave(" = -ESTALE [absent]");
937 		return ERR_PTR(-ESTALE);
938 	}
939 
940 	if (ret == -EIO) {
941 		cachefiles_io_error(cache, "Lookup failed");
942 	} else if (ret != -ENOMEM) {
943 		pr_err("Internal error: %d\n", ret);
944 		ret = -EIO;
945 	}
946 
947 	_leave(" = %d", ret);
948 	return ERR_PTR(ret);
949 }
950 
951 /*
952  * cull an object if it's not in use
953  * - called only by cache manager daemon
954  */
955 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
956 		    char *filename)
957 {
958 	struct dentry *victim;
959 	int ret;
960 
961 	_enter(",%pd/,%s", dir, filename);
962 
963 	victim = cachefiles_check_active(cache, dir, filename);
964 	if (IS_ERR(victim))
965 		return PTR_ERR(victim);
966 
967 	_debug("victim -> %p %s",
968 	       victim, d_backing_inode(victim) ? "positive" : "negative");
969 
970 	/* okay... the victim is not being used so we can cull it
971 	 * - start by marking it as stale
972 	 */
973 	_debug("victim is cullable");
974 
975 	ret = cachefiles_remove_object_xattr(cache, victim);
976 	if (ret < 0)
977 		goto error_unlock;
978 
979 	/*  actually remove the victim (drops the dir mutex) */
980 	_debug("bury");
981 
982 	ret = cachefiles_bury_object(cache, NULL, dir, victim, false,
983 				     FSCACHE_OBJECT_WAS_CULLED);
984 	if (ret < 0)
985 		goto error;
986 
987 	dput(victim);
988 	_leave(" = 0");
989 	return 0;
990 
991 error_unlock:
992 	inode_unlock(d_inode(dir));
993 error:
994 	dput(victim);
995 	if (ret == -ENOENT) {
996 		/* file or dir now absent - probably retired by netfs */
997 		_leave(" = -ESTALE [absent]");
998 		return -ESTALE;
999 	}
1000 
1001 	if (ret != -ENOMEM) {
1002 		pr_err("Internal error: %d\n", ret);
1003 		ret = -EIO;
1004 	}
1005 
1006 	_leave(" = %d", ret);
1007 	return ret;
1008 }
1009 
1010 /*
1011  * find out if an object is in use or not
1012  * - called only by cache manager daemon
1013  * - returns -EBUSY or 0 to indicate whether an object is in use or not
1014  */
1015 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
1016 			    char *filename)
1017 {
1018 	struct dentry *victim;
1019 
1020 	//_enter(",%pd/,%s",
1021 	//       dir, filename);
1022 
1023 	victim = cachefiles_check_active(cache, dir, filename);
1024 	if (IS_ERR(victim))
1025 		return PTR_ERR(victim);
1026 
1027 	inode_unlock(d_inode(dir));
1028 	dput(victim);
1029 	//_leave(" = 0");
1030 	return 0;
1031 }
1032