xref: /openbmc/linux/fs/cachefiles/namei.c (revision 081d7d35)
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 {
34 	struct fscache_cookie *cookie;
35 	const u8 *k;
36 	unsigned loop;
37 
38 	pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
39 	pr_err("%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
40 	       prefix, object->fscache.state->name,
41 	       object->fscache.flags, work_busy(&object->fscache.work),
42 	       object->fscache.events, object->fscache.event_mask);
43 	pr_err("%sops=%u inp=%u exc=%u\n",
44 	       prefix, object->fscache.n_ops, object->fscache.n_in_progress,
45 	       object->fscache.n_exclusive);
46 	pr_err("%sparent=%p\n",
47 	       prefix, object->fscache.parent);
48 
49 	spin_lock(&object->fscache.lock);
50 	cookie = object->fscache.cookie;
51 	if (cookie) {
52 		pr_err("%scookie=%p [pr=%p nd=%p fl=%lx]\n",
53 		       prefix,
54 		       object->fscache.cookie,
55 		       object->fscache.cookie->parent,
56 		       object->fscache.cookie->netfs_data,
57 		       object->fscache.cookie->flags);
58 		pr_err("%skey=[%u] '", prefix, cookie->key_len);
59 		k = (cookie->key_len <= sizeof(cookie->inline_key)) ?
60 			cookie->inline_key : cookie->key;
61 		for (loop = 0; loop < cookie->key_len; loop++)
62 			pr_cont("%02x", k[loop]);
63 		pr_cont("'\n");
64 	} else {
65 		pr_err("%scookie=NULL\n", prefix);
66 	}
67 	spin_unlock(&object->fscache.lock);
68 }
69 
70 /*
71  * dump debugging info about a pair of objects
72  */
73 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
74 					      struct cachefiles_object *xobject)
75 {
76 	if (object)
77 		__cachefiles_printk_object(object, "");
78 	if (xobject)
79 		__cachefiles_printk_object(xobject, "x");
80 }
81 
82 /*
83  * mark the owner of a dentry, if there is one, to indicate that that dentry
84  * has been preemptively deleted
85  * - the caller must hold the i_mutex on the dentry's parent as required to
86  *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
87  */
88 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
89 					  struct dentry *dentry,
90 					  enum fscache_why_object_killed why)
91 {
92 	struct cachefiles_object *object;
93 	struct rb_node *p;
94 
95 	_enter(",'%pd'", dentry);
96 
97 	write_lock(&cache->active_lock);
98 
99 	p = cache->active_nodes.rb_node;
100 	while (p) {
101 		object = rb_entry(p, struct cachefiles_object, active_node);
102 		if (object->dentry > dentry)
103 			p = p->rb_left;
104 		else if (object->dentry < dentry)
105 			p = p->rb_right;
106 		else
107 			goto found_dentry;
108 	}
109 
110 	write_unlock(&cache->active_lock);
111 	trace_cachefiles_mark_buried(NULL, dentry, why);
112 	_leave(" [no owner]");
113 	return;
114 
115 	/* found the dentry for  */
116 found_dentry:
117 	kdebug("preemptive burial: OBJ%x [%s] %p",
118 	       object->fscache.debug_id,
119 	       object->fscache.state->name,
120 	       dentry);
121 
122 	trace_cachefiles_mark_buried(object, dentry, why);
123 
124 	if (fscache_object_is_live(&object->fscache)) {
125 		pr_err("\n");
126 		pr_err("Error: Can't preemptively bury live object\n");
127 		cachefiles_printk_object(object, NULL);
128 	} else {
129 		if (why != FSCACHE_OBJECT_IS_STALE)
130 			fscache_object_mark_killed(&object->fscache, why);
131 	}
132 
133 	write_unlock(&cache->active_lock);
134 	_leave(" [owner marked]");
135 }
136 
137 /*
138  * record the fact that an object is now active
139  */
140 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
141 					 struct cachefiles_object *object)
142 {
143 	struct cachefiles_object *xobject;
144 	struct rb_node **_p, *_parent = NULL;
145 	struct dentry *dentry;
146 
147 	_enter(",%p", object);
148 
149 try_again:
150 	write_lock(&cache->active_lock);
151 
152 	dentry = object->dentry;
153 	trace_cachefiles_mark_active(object, dentry);
154 
155 	if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
156 		pr_err("Error: Object already active\n");
157 		cachefiles_printk_object(object, NULL);
158 		BUG();
159 	}
160 
161 	_p = &cache->active_nodes.rb_node;
162 	while (*_p) {
163 		_parent = *_p;
164 		xobject = rb_entry(_parent,
165 				   struct cachefiles_object, active_node);
166 
167 		ASSERT(xobject != object);
168 
169 		if (xobject->dentry > dentry)
170 			_p = &(*_p)->rb_left;
171 		else if (xobject->dentry < dentry)
172 			_p = &(*_p)->rb_right;
173 		else
174 			goto wait_for_old_object;
175 	}
176 
177 	rb_link_node(&object->active_node, _parent, _p);
178 	rb_insert_color(&object->active_node, &cache->active_nodes);
179 
180 	write_unlock(&cache->active_lock);
181 	_leave(" = 0");
182 	return 0;
183 
184 	/* an old object from a previous incarnation is hogging the slot - we
185 	 * need to wait for it to be destroyed */
186 wait_for_old_object:
187 	trace_cachefiles_wait_active(object, dentry, xobject);
188 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
189 
190 	if (fscache_object_is_live(&xobject->fscache)) {
191 		pr_err("\n");
192 		pr_err("Error: Unexpected object collision\n");
193 		cachefiles_printk_object(object, xobject);
194 	}
195 	atomic_inc(&xobject->usage);
196 	write_unlock(&cache->active_lock);
197 
198 	if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
199 		wait_queue_head_t *wq;
200 
201 		signed long timeout = 60 * HZ;
202 		wait_queue_entry_t wait;
203 		bool requeue;
204 
205 		/* if the object we're waiting for is queued for processing,
206 		 * then just put ourselves on the queue behind it */
207 		if (work_pending(&xobject->fscache.work)) {
208 			_debug("queue OBJ%x behind OBJ%x immediately",
209 			       object->fscache.debug_id,
210 			       xobject->fscache.debug_id);
211 			goto requeue;
212 		}
213 
214 		/* otherwise we sleep until either the object we're waiting for
215 		 * is done, or the fscache_object is congested */
216 		wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
217 		init_wait(&wait);
218 		requeue = false;
219 		do {
220 			prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
221 			if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
222 				break;
223 
224 			requeue = fscache_object_sleep_till_congested(&timeout);
225 		} while (timeout > 0 && !requeue);
226 		finish_wait(wq, &wait);
227 
228 		if (requeue &&
229 		    test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
230 			_debug("queue OBJ%x behind OBJ%x after wait",
231 			       object->fscache.debug_id,
232 			       xobject->fscache.debug_id);
233 			goto requeue;
234 		}
235 
236 		if (timeout <= 0) {
237 			pr_err("\n");
238 			pr_err("Error: Overlong wait for old active object to go away\n");
239 			cachefiles_printk_object(object, xobject);
240 			goto requeue;
241 		}
242 	}
243 
244 	ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
245 
246 	cache->cache.ops->put_object(&xobject->fscache,
247 		(enum fscache_obj_ref_trace)cachefiles_obj_put_wait_retry);
248 	goto try_again;
249 
250 requeue:
251 	cache->cache.ops->put_object(&xobject->fscache,
252 		(enum fscache_obj_ref_trace)cachefiles_obj_put_wait_timeo);
253 	_leave(" = -ETIMEDOUT");
254 	return -ETIMEDOUT;
255 }
256 
257 /*
258  * Mark an object as being inactive.
259  */
260 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
261 				     struct cachefiles_object *object,
262 				     blkcnt_t i_blocks)
263 {
264 	struct dentry *dentry = object->dentry;
265 	struct inode *inode = d_backing_inode(dentry);
266 
267 	trace_cachefiles_mark_inactive(object, dentry, inode);
268 
269 	write_lock(&cache->active_lock);
270 	rb_erase(&object->active_node, &cache->active_nodes);
271 	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
272 	write_unlock(&cache->active_lock);
273 
274 	wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
275 
276 	/* This object can now be culled, so we need to let the daemon know
277 	 * that there is something it can remove if it needs to.
278 	 */
279 	atomic_long_add(i_blocks, &cache->b_released);
280 	if (atomic_inc_return(&cache->f_released))
281 		cachefiles_state_changed(cache);
282 }
283 
284 /*
285  * delete an object representation from the cache
286  * - file backed objects are unlinked
287  * - directory backed objects are stuffed into the graveyard for userspace to
288  *   delete
289  * - unlocks the directory mutex
290  */
291 static int cachefiles_bury_object(struct cachefiles_cache *cache,
292 				  struct cachefiles_object *object,
293 				  struct dentry *dir,
294 				  struct dentry *rep,
295 				  bool preemptive,
296 				  enum fscache_why_object_killed why)
297 {
298 	struct dentry *grave, *trap;
299 	struct path path, path_to_graveyard;
300 	char nbuffer[8 + 8 + 1];
301 	int ret;
302 
303 	_enter(",'%pd','%pd'", dir, rep);
304 
305 	_debug("remove %p from %p", rep, dir);
306 
307 	/* non-directories can just be unlinked */
308 	if (!d_is_dir(rep)) {
309 		_debug("unlink stale object");
310 
311 		path.mnt = cache->mnt;
312 		path.dentry = dir;
313 		ret = security_path_unlink(&path, rep);
314 		if (ret < 0) {
315 			cachefiles_io_error(cache, "Unlink security error");
316 		} else {
317 			trace_cachefiles_unlink(object, rep, why);
318 			ret = vfs_unlink(d_inode(dir), rep, NULL);
319 
320 			if (preemptive)
321 				cachefiles_mark_object_buried(cache, rep, why);
322 		}
323 
324 		inode_unlock(d_inode(dir));
325 
326 		if (ret == -EIO)
327 			cachefiles_io_error(cache, "Unlink failed");
328 
329 		_leave(" = %d", ret);
330 		return ret;
331 	}
332 
333 	/* directories have to be moved to the graveyard */
334 	_debug("move stale object to graveyard");
335 	inode_unlock(d_inode(dir));
336 
337 try_again:
338 	/* first step is to make up a grave dentry in the graveyard */
339 	sprintf(nbuffer, "%08x%08x",
340 		(uint32_t) ktime_get_real_seconds(),
341 		(uint32_t) atomic_inc_return(&cache->gravecounter));
342 
343 	/* do the multiway lock magic */
344 	trap = lock_rename(cache->graveyard, dir);
345 
346 	/* do some checks before getting the grave dentry */
347 	if (rep->d_parent != dir || IS_DEADDIR(d_inode(rep))) {
348 		/* the entry was probably culled when we dropped the parent dir
349 		 * lock */
350 		unlock_rename(cache->graveyard, dir);
351 		_leave(" = 0 [culled?]");
352 		return 0;
353 	}
354 
355 	if (!d_can_lookup(cache->graveyard)) {
356 		unlock_rename(cache->graveyard, dir);
357 		cachefiles_io_error(cache, "Graveyard no longer a directory");
358 		return -EIO;
359 	}
360 
361 	if (trap == rep) {
362 		unlock_rename(cache->graveyard, dir);
363 		cachefiles_io_error(cache, "May not make directory loop");
364 		return -EIO;
365 	}
366 
367 	if (d_mountpoint(rep)) {
368 		unlock_rename(cache->graveyard, dir);
369 		cachefiles_io_error(cache, "Mountpoint in cache");
370 		return -EIO;
371 	}
372 
373 	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
374 	if (IS_ERR(grave)) {
375 		unlock_rename(cache->graveyard, dir);
376 
377 		if (PTR_ERR(grave) == -ENOMEM) {
378 			_leave(" = -ENOMEM");
379 			return -ENOMEM;
380 		}
381 
382 		cachefiles_io_error(cache, "Lookup error %ld",
383 				    PTR_ERR(grave));
384 		return -EIO;
385 	}
386 
387 	if (d_is_positive(grave)) {
388 		unlock_rename(cache->graveyard, dir);
389 		dput(grave);
390 		grave = NULL;
391 		cond_resched();
392 		goto try_again;
393 	}
394 
395 	if (d_mountpoint(grave)) {
396 		unlock_rename(cache->graveyard, dir);
397 		dput(grave);
398 		cachefiles_io_error(cache, "Mountpoint in graveyard");
399 		return -EIO;
400 	}
401 
402 	/* target should not be an ancestor of source */
403 	if (trap == grave) {
404 		unlock_rename(cache->graveyard, dir);
405 		dput(grave);
406 		cachefiles_io_error(cache, "May not make directory loop");
407 		return -EIO;
408 	}
409 
410 	/* attempt the rename */
411 	path.mnt = cache->mnt;
412 	path.dentry = dir;
413 	path_to_graveyard.mnt = cache->mnt;
414 	path_to_graveyard.dentry = cache->graveyard;
415 	ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
416 	if (ret < 0) {
417 		cachefiles_io_error(cache, "Rename security error %d", ret);
418 	} else {
419 		trace_cachefiles_rename(object, rep, grave, why);
420 		ret = vfs_rename(d_inode(dir), rep,
421 				 d_inode(cache->graveyard), grave, NULL, 0);
422 		if (ret != 0 && ret != -ENOMEM)
423 			cachefiles_io_error(cache,
424 					    "Rename failed with error %d", ret);
425 
426 		if (preemptive)
427 			cachefiles_mark_object_buried(cache, rep, why);
428 	}
429 
430 	unlock_rename(cache->graveyard, dir);
431 	dput(grave);
432 	_leave(" = 0");
433 	return 0;
434 }
435 
436 /*
437  * delete an object representation from the cache
438  */
439 int cachefiles_delete_object(struct cachefiles_cache *cache,
440 			     struct cachefiles_object *object)
441 {
442 	struct dentry *dir;
443 	int ret;
444 
445 	_enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
446 
447 	ASSERT(object->dentry);
448 	ASSERT(d_backing_inode(object->dentry));
449 	ASSERT(object->dentry->d_parent);
450 
451 	dir = dget_parent(object->dentry);
452 
453 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
454 
455 	if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
456 		/* object allocation for the same key preemptively deleted this
457 		 * object's file so that it could create its own file */
458 		_debug("object preemptively buried");
459 		inode_unlock(d_inode(dir));
460 		ret = 0;
461 	} else {
462 		/* we need to check that our parent is _still_ our parent - it
463 		 * may have been renamed */
464 		if (dir == object->dentry->d_parent) {
465 			ret = cachefiles_bury_object(cache, object, dir,
466 						     object->dentry, false,
467 						     FSCACHE_OBJECT_WAS_RETIRED);
468 		} else {
469 			/* it got moved, presumably by cachefilesd culling it,
470 			 * so it's no longer in the key path and we can ignore
471 			 * it */
472 			inode_unlock(d_inode(dir));
473 			ret = 0;
474 		}
475 	}
476 
477 	dput(dir);
478 	_leave(" = %d", ret);
479 	return ret;
480 }
481 
482 /*
483  * walk from the parent object to the child object through the backing
484  * filesystem, creating directories as we go
485  */
486 int cachefiles_walk_to_object(struct cachefiles_object *parent,
487 			      struct cachefiles_object *object,
488 			      const char *key,
489 			      struct cachefiles_xattr *auxdata)
490 {
491 	struct cachefiles_cache *cache;
492 	struct dentry *dir, *next = NULL;
493 	struct inode *inode;
494 	struct path path;
495 	unsigned long start;
496 	const char *name;
497 	int ret, nlen;
498 
499 	_enter("OBJ%x{%p},OBJ%x,%s,",
500 	       parent->fscache.debug_id, parent->dentry,
501 	       object->fscache.debug_id, key);
502 
503 	cache = container_of(parent->fscache.cache,
504 			     struct cachefiles_cache, cache);
505 	path.mnt = cache->mnt;
506 
507 	ASSERT(parent->dentry);
508 	ASSERT(d_backing_inode(parent->dentry));
509 
510 	if (!(d_is_dir(parent->dentry))) {
511 		// TODO: convert file to dir
512 		_leave("looking up in none directory");
513 		return -ENOBUFS;
514 	}
515 
516 	dir = dget(parent->dentry);
517 
518 advance:
519 	/* attempt to transit the first directory component */
520 	name = key;
521 	nlen = strlen(key);
522 
523 	/* key ends in a double NUL */
524 	key = key + nlen + 1;
525 	if (!*key)
526 		key = NULL;
527 
528 lookup_again:
529 	/* search the current directory for the element name */
530 	_debug("lookup '%s'", name);
531 
532 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
533 
534 	start = jiffies;
535 	next = lookup_one_len(name, dir, nlen);
536 	cachefiles_hist(cachefiles_lookup_histogram, start);
537 	if (IS_ERR(next)) {
538 		trace_cachefiles_lookup(object, next, NULL);
539 		goto lookup_error;
540 	}
541 
542 	inode = d_backing_inode(next);
543 	trace_cachefiles_lookup(object, next, inode);
544 	_debug("next -> %p %s", next, inode ? "positive" : "negative");
545 
546 	if (!key)
547 		object->new = !inode;
548 
549 	/* if this element of the path doesn't exist, then the lookup phase
550 	 * failed, and we can release any readers in the certain knowledge that
551 	 * there's nothing for them to actually read */
552 	if (d_is_negative(next))
553 		fscache_object_lookup_negative(&object->fscache);
554 
555 	/* we need to create the object if it's negative */
556 	if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
557 		/* index objects and intervening tree levels must be subdirs */
558 		if (d_is_negative(next)) {
559 			ret = cachefiles_has_space(cache, 1, 0);
560 			if (ret < 0)
561 				goto no_space_error;
562 
563 			path.dentry = dir;
564 			ret = security_path_mkdir(&path, next, 0);
565 			if (ret < 0)
566 				goto create_error;
567 			start = jiffies;
568 			ret = vfs_mkdir(d_inode(dir), next, 0);
569 			cachefiles_hist(cachefiles_mkdir_histogram, start);
570 			if (!key)
571 				trace_cachefiles_mkdir(object, next, ret);
572 			if (ret < 0)
573 				goto create_error;
574 
575 			if (unlikely(d_unhashed(next))) {
576 				dput(next);
577 				inode_unlock(d_inode(dir));
578 				goto lookup_again;
579 			}
580 			ASSERT(d_backing_inode(next));
581 
582 			_debug("mkdir -> %p{%p{ino=%lu}}",
583 			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
584 
585 		} else if (!d_can_lookup(next)) {
586 			pr_err("inode %lu is not a directory\n",
587 			       d_backing_inode(next)->i_ino);
588 			ret = -ENOBUFS;
589 			goto error;
590 		}
591 
592 	} else {
593 		/* non-index objects start out life as files */
594 		if (d_is_negative(next)) {
595 			ret = cachefiles_has_space(cache, 1, 0);
596 			if (ret < 0)
597 				goto no_space_error;
598 
599 			path.dentry = dir;
600 			ret = security_path_mknod(&path, next, S_IFREG, 0);
601 			if (ret < 0)
602 				goto create_error;
603 			start = jiffies;
604 			ret = vfs_create(d_inode(dir), next, S_IFREG, true);
605 			cachefiles_hist(cachefiles_create_histogram, start);
606 			trace_cachefiles_create(object, next, ret);
607 			if (ret < 0)
608 				goto create_error;
609 
610 			ASSERT(d_backing_inode(next));
611 
612 			_debug("create -> %p{%p{ino=%lu}}",
613 			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
614 
615 		} else if (!d_can_lookup(next) &&
616 			   !d_is_reg(next)
617 			   ) {
618 			pr_err("inode %lu is not a file or directory\n",
619 			       d_backing_inode(next)->i_ino);
620 			ret = -ENOBUFS;
621 			goto error;
622 		}
623 	}
624 
625 	/* process the next component */
626 	if (key) {
627 		_debug("advance");
628 		inode_unlock(d_inode(dir));
629 		dput(dir);
630 		dir = next;
631 		next = NULL;
632 		goto advance;
633 	}
634 
635 	/* we've found the object we were looking for */
636 	object->dentry = next;
637 
638 	/* if we've found that the terminal object exists, then we need to
639 	 * check its attributes and delete it if it's out of date */
640 	if (!object->new) {
641 		_debug("validate '%pd'", next);
642 
643 		ret = cachefiles_check_object_xattr(object, auxdata);
644 		if (ret == -ESTALE) {
645 			/* delete the object (the deleter drops the directory
646 			 * mutex) */
647 			object->dentry = NULL;
648 
649 			ret = cachefiles_bury_object(cache, object, dir, next,
650 						     true,
651 						     FSCACHE_OBJECT_IS_STALE);
652 			dput(next);
653 			next = NULL;
654 
655 			if (ret < 0)
656 				goto delete_error;
657 
658 			_debug("redo lookup");
659 			fscache_object_retrying_stale(&object->fscache);
660 			goto lookup_again;
661 		}
662 	}
663 
664 	/* note that we're now using this object */
665 	ret = cachefiles_mark_object_active(cache, object);
666 
667 	inode_unlock(d_inode(dir));
668 	dput(dir);
669 	dir = NULL;
670 
671 	if (ret == -ETIMEDOUT)
672 		goto mark_active_timed_out;
673 
674 	_debug("=== OBTAINED_OBJECT ===");
675 
676 	if (object->new) {
677 		/* attach data to a newly constructed terminal object */
678 		ret = cachefiles_set_object_xattr(object, auxdata);
679 		if (ret < 0)
680 			goto check_error;
681 	} else {
682 		/* always update the atime on an object we've just looked up
683 		 * (this is used to keep track of culling, and atimes are only
684 		 * updated by read, write and readdir but not lookup or
685 		 * open) */
686 		path.dentry = next;
687 		touch_atime(&path);
688 	}
689 
690 	/* open a file interface onto a data file */
691 	if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
692 		if (d_is_reg(object->dentry)) {
693 			const struct address_space_operations *aops;
694 
695 			ret = -EPERM;
696 			aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
697 			if (!aops->bmap)
698 				goto check_error;
699 			if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
700 				goto check_error;
701 
702 			object->backer = object->dentry;
703 		} else {
704 			BUG(); // TODO: open file in data-class subdir
705 		}
706 	}
707 
708 	object->new = 0;
709 	fscache_obtained_object(&object->fscache);
710 
711 	_leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
712 	return 0;
713 
714 no_space_error:
715 	fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
716 create_error:
717 	_debug("create error %d", ret);
718 	if (ret == -EIO)
719 		cachefiles_io_error(cache, "Create/mkdir failed");
720 	goto error;
721 
722 mark_active_timed_out:
723 	_debug("mark active timed out");
724 	goto release_dentry;
725 
726 check_error:
727 	_debug("check error %d", ret);
728 	cachefiles_mark_object_inactive(
729 		cache, object, d_backing_inode(object->dentry)->i_blocks);
730 release_dentry:
731 	dput(object->dentry);
732 	object->dentry = NULL;
733 	goto error_out;
734 
735 delete_error:
736 	_debug("delete error %d", ret);
737 	goto error_out2;
738 
739 lookup_error:
740 	_debug("lookup error %ld", PTR_ERR(next));
741 	ret = PTR_ERR(next);
742 	if (ret == -EIO)
743 		cachefiles_io_error(cache, "Lookup failed");
744 	next = NULL;
745 error:
746 	inode_unlock(d_inode(dir));
747 	dput(next);
748 error_out2:
749 	dput(dir);
750 error_out:
751 	_leave(" = error %d", -ret);
752 	return ret;
753 }
754 
755 /*
756  * get a subdirectory
757  */
758 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
759 					struct dentry *dir,
760 					const char *dirname)
761 {
762 	struct dentry *subdir;
763 	unsigned long start;
764 	struct path path;
765 	int ret;
766 
767 	_enter(",,%s", dirname);
768 
769 	/* search the current directory for the element name */
770 	inode_lock(d_inode(dir));
771 
772 retry:
773 	start = jiffies;
774 	subdir = lookup_one_len(dirname, dir, strlen(dirname));
775 	cachefiles_hist(cachefiles_lookup_histogram, start);
776 	if (IS_ERR(subdir)) {
777 		if (PTR_ERR(subdir) == -ENOMEM)
778 			goto nomem_d_alloc;
779 		goto lookup_error;
780 	}
781 
782 	_debug("subdir -> %p %s",
783 	       subdir, d_backing_inode(subdir) ? "positive" : "negative");
784 
785 	/* we need to create the subdir if it doesn't exist yet */
786 	if (d_is_negative(subdir)) {
787 		ret = cachefiles_has_space(cache, 1, 0);
788 		if (ret < 0)
789 			goto mkdir_error;
790 
791 		_debug("attempt mkdir");
792 
793 		path.mnt = cache->mnt;
794 		path.dentry = dir;
795 		ret = security_path_mkdir(&path, subdir, 0700);
796 		if (ret < 0)
797 			goto mkdir_error;
798 		ret = vfs_mkdir(d_inode(dir), subdir, 0700);
799 		if (ret < 0)
800 			goto mkdir_error;
801 
802 		if (unlikely(d_unhashed(subdir))) {
803 			dput(subdir);
804 			goto retry;
805 		}
806 		ASSERT(d_backing_inode(subdir));
807 
808 		_debug("mkdir -> %p{%p{ino=%lu}}",
809 		       subdir,
810 		       d_backing_inode(subdir),
811 		       d_backing_inode(subdir)->i_ino);
812 	}
813 
814 	inode_unlock(d_inode(dir));
815 
816 	/* we need to make sure the subdir is a directory */
817 	ASSERT(d_backing_inode(subdir));
818 
819 	if (!d_can_lookup(subdir)) {
820 		pr_err("%s is not a directory\n", dirname);
821 		ret = -EIO;
822 		goto check_error;
823 	}
824 
825 	ret = -EPERM;
826 	if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
827 	    !d_backing_inode(subdir)->i_op->lookup ||
828 	    !d_backing_inode(subdir)->i_op->mkdir ||
829 	    !d_backing_inode(subdir)->i_op->create ||
830 	    !d_backing_inode(subdir)->i_op->rename ||
831 	    !d_backing_inode(subdir)->i_op->rmdir ||
832 	    !d_backing_inode(subdir)->i_op->unlink)
833 		goto check_error;
834 
835 	_leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
836 	return subdir;
837 
838 check_error:
839 	dput(subdir);
840 	_leave(" = %d [check]", ret);
841 	return ERR_PTR(ret);
842 
843 mkdir_error:
844 	inode_unlock(d_inode(dir));
845 	dput(subdir);
846 	pr_err("mkdir %s failed with error %d\n", dirname, ret);
847 	return ERR_PTR(ret);
848 
849 lookup_error:
850 	inode_unlock(d_inode(dir));
851 	ret = PTR_ERR(subdir);
852 	pr_err("Lookup %s failed with error %d\n", dirname, ret);
853 	return ERR_PTR(ret);
854 
855 nomem_d_alloc:
856 	inode_unlock(d_inode(dir));
857 	_leave(" = -ENOMEM");
858 	return ERR_PTR(-ENOMEM);
859 }
860 
861 /*
862  * find out if an object is in use or not
863  * - if finds object and it's not in use:
864  *   - returns a pointer to the object and a reference on it
865  *   - returns with the directory locked
866  */
867 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
868 					      struct dentry *dir,
869 					      char *filename)
870 {
871 	struct cachefiles_object *object;
872 	struct rb_node *_n;
873 	struct dentry *victim;
874 	unsigned long start;
875 	int ret;
876 
877 	//_enter(",%pd/,%s",
878 	//       dir, filename);
879 
880 	/* look up the victim */
881 	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
882 
883 	start = jiffies;
884 	victim = lookup_one_len(filename, dir, strlen(filename));
885 	cachefiles_hist(cachefiles_lookup_histogram, start);
886 	if (IS_ERR(victim))
887 		goto lookup_error;
888 
889 	//_debug("victim -> %p %s",
890 	//       victim, d_backing_inode(victim) ? "positive" : "negative");
891 
892 	/* if the object is no longer there then we probably retired the object
893 	 * at the netfs's request whilst the cull was in progress
894 	 */
895 	if (d_is_negative(victim)) {
896 		inode_unlock(d_inode(dir));
897 		dput(victim);
898 		_leave(" = -ENOENT [absent]");
899 		return ERR_PTR(-ENOENT);
900 	}
901 
902 	/* check to see if we're using this object */
903 	read_lock(&cache->active_lock);
904 
905 	_n = cache->active_nodes.rb_node;
906 
907 	while (_n) {
908 		object = rb_entry(_n, struct cachefiles_object, active_node);
909 
910 		if (object->dentry > victim)
911 			_n = _n->rb_left;
912 		else if (object->dentry < victim)
913 			_n = _n->rb_right;
914 		else
915 			goto object_in_use;
916 	}
917 
918 	read_unlock(&cache->active_lock);
919 
920 	//_leave(" = %p", victim);
921 	return victim;
922 
923 object_in_use:
924 	read_unlock(&cache->active_lock);
925 	inode_unlock(d_inode(dir));
926 	dput(victim);
927 	//_leave(" = -EBUSY [in use]");
928 	return ERR_PTR(-EBUSY);
929 
930 lookup_error:
931 	inode_unlock(d_inode(dir));
932 	ret = PTR_ERR(victim);
933 	if (ret == -ENOENT) {
934 		/* file or dir now absent - probably retired by netfs */
935 		_leave(" = -ESTALE [absent]");
936 		return ERR_PTR(-ESTALE);
937 	}
938 
939 	if (ret == -EIO) {
940 		cachefiles_io_error(cache, "Lookup failed");
941 	} else if (ret != -ENOMEM) {
942 		pr_err("Internal error: %d\n", ret);
943 		ret = -EIO;
944 	}
945 
946 	_leave(" = %d", ret);
947 	return ERR_PTR(ret);
948 }
949 
950 /*
951  * cull an object if it's not in use
952  * - called only by cache manager daemon
953  */
954 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
955 		    char *filename)
956 {
957 	struct dentry *victim;
958 	int ret;
959 
960 	_enter(",%pd/,%s", dir, filename);
961 
962 	victim = cachefiles_check_active(cache, dir, filename);
963 	if (IS_ERR(victim))
964 		return PTR_ERR(victim);
965 
966 	_debug("victim -> %p %s",
967 	       victim, d_backing_inode(victim) ? "positive" : "negative");
968 
969 	/* okay... the victim is not being used so we can cull it
970 	 * - start by marking it as stale
971 	 */
972 	_debug("victim is cullable");
973 
974 	ret = cachefiles_remove_object_xattr(cache, victim);
975 	if (ret < 0)
976 		goto error_unlock;
977 
978 	/*  actually remove the victim (drops the dir mutex) */
979 	_debug("bury");
980 
981 	ret = cachefiles_bury_object(cache, NULL, dir, victim, false,
982 				     FSCACHE_OBJECT_WAS_CULLED);
983 	if (ret < 0)
984 		goto error;
985 
986 	dput(victim);
987 	_leave(" = 0");
988 	return 0;
989 
990 error_unlock:
991 	inode_unlock(d_inode(dir));
992 error:
993 	dput(victim);
994 	if (ret == -ENOENT) {
995 		/* file or dir now absent - probably retired by netfs */
996 		_leave(" = -ESTALE [absent]");
997 		return -ESTALE;
998 	}
999 
1000 	if (ret != -ENOMEM) {
1001 		pr_err("Internal error: %d\n", ret);
1002 		ret = -EIO;
1003 	}
1004 
1005 	_leave(" = %d", ret);
1006 	return ret;
1007 }
1008 
1009 /*
1010  * find out if an object is in use or not
1011  * - called only by cache manager daemon
1012  * - returns -EBUSY or 0 to indicate whether an object is in use or not
1013  */
1014 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
1015 			    char *filename)
1016 {
1017 	struct dentry *victim;
1018 
1019 	//_enter(",%pd/,%s",
1020 	//       dir, filename);
1021 
1022 	victim = cachefiles_check_active(cache, dir, filename);
1023 	if (IS_ERR(victim))
1024 		return PTR_ERR(victim);
1025 
1026 	inode_unlock(d_inode(dir));
1027 	dput(victim);
1028 	//_leave(" = 0");
1029 	return 0;
1030 }
1031