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 */
30static noinline
31void __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	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		if (keybuf && cookie->def)
59			keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
60						      CACHEFILES_KEYBUF_SIZE);
61		else
62			keylen = 0;
63	} else {
64		pr_err("%scookie=NULL\n", prefix);
65		keylen = 0;
66	}
67	spin_unlock(&object->fscache.lock);
68
69	if (keylen) {
70		pr_err("%skey=[%u] '", prefix, keylen);
71		for (loop = 0; loop < keylen; loop++)
72			pr_cont("%02x", keybuf[loop]);
73		pr_cont("'\n");
74	}
75}
76
77/*
78 * dump debugging info about a pair of objects
79 */
80static noinline void cachefiles_printk_object(struct cachefiles_object *object,
81					      struct cachefiles_object *xobject)
82{
83	u8 *keybuf;
84
85	keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
86	if (object)
87		__cachefiles_printk_object(object, "", keybuf);
88	if (xobject)
89		__cachefiles_printk_object(xobject, "x", keybuf);
90	kfree(keybuf);
91}
92
93/*
94 * mark the owner of a dentry, if there is one, to indicate that that dentry
95 * has been preemptively deleted
96 * - the caller must hold the i_mutex on the dentry's parent as required to
97 *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
98 */
99static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
100					  struct dentry *dentry)
101{
102	struct cachefiles_object *object;
103	struct rb_node *p;
104
105	_enter(",'%pd'", dentry);
106
107	write_lock(&cache->active_lock);
108
109	p = cache->active_nodes.rb_node;
110	while (p) {
111		object = rb_entry(p, struct cachefiles_object, active_node);
112		if (object->dentry > dentry)
113			p = p->rb_left;
114		else if (object->dentry < dentry)
115			p = p->rb_right;
116		else
117			goto found_dentry;
118	}
119
120	write_unlock(&cache->active_lock);
121	_leave(" [no owner]");
122	return;
123
124	/* found the dentry for  */
125found_dentry:
126	kdebug("preemptive burial: OBJ%x [%s] %p",
127	       object->fscache.debug_id,
128	       object->fscache.state->name,
129	       dentry);
130
131	if (fscache_object_is_live(&object->fscache)) {
132		pr_err("\n");
133		pr_err("Error: Can't preemptively bury live object\n");
134		cachefiles_printk_object(object, NULL);
135	} else if (test_and_set_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
136		pr_err("Error: Object already preemptively buried\n");
137	}
138
139	write_unlock(&cache->active_lock);
140	_leave(" [owner marked]");
141}
142
143/*
144 * record the fact that an object is now active
145 */
146static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
147					 struct cachefiles_object *object)
148{
149	struct cachefiles_object *xobject;
150	struct rb_node **_p, *_parent = NULL;
151	struct dentry *dentry;
152
153	_enter(",%p", object);
154
155try_again:
156	write_lock(&cache->active_lock);
157
158	if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
159		pr_err("Error: Object already active\n");
160		cachefiles_printk_object(object, NULL);
161		BUG();
162	}
163
164	dentry = object->dentry;
165	_p = &cache->active_nodes.rb_node;
166	while (*_p) {
167		_parent = *_p;
168		xobject = rb_entry(_parent,
169				   struct cachefiles_object, active_node);
170
171		ASSERT(xobject != object);
172
173		if (xobject->dentry > dentry)
174			_p = &(*_p)->rb_left;
175		else if (xobject->dentry < dentry)
176			_p = &(*_p)->rb_right;
177		else
178			goto wait_for_old_object;
179	}
180
181	rb_link_node(&object->active_node, _parent, _p);
182	rb_insert_color(&object->active_node, &cache->active_nodes);
183
184	write_unlock(&cache->active_lock);
185	_leave(" = 0");
186	return 0;
187
188	/* an old object from a previous incarnation is hogging the slot - we
189	 * need to wait for it to be destroyed */
190wait_for_old_object:
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		BUG();
196	}
197	atomic_inc(&xobject->usage);
198	write_unlock(&cache->active_lock);
199
200	if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
201		wait_queue_head_t *wq;
202
203		signed long timeout = 60 * HZ;
204		wait_queue_t wait;
205		bool requeue;
206
207		/* if the object we're waiting for is queued for processing,
208		 * then just put ourselves on the queue behind it */
209		if (work_pending(&xobject->fscache.work)) {
210			_debug("queue OBJ%x behind OBJ%x immediately",
211			       object->fscache.debug_id,
212			       xobject->fscache.debug_id);
213			goto requeue;
214		}
215
216		/* otherwise we sleep until either the object we're waiting for
217		 * is done, or the fscache_object is congested */
218		wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
219		init_wait(&wait);
220		requeue = false;
221		do {
222			prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
223			if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
224				break;
225
226			requeue = fscache_object_sleep_till_congested(&timeout);
227		} while (timeout > 0 && !requeue);
228		finish_wait(wq, &wait);
229
230		if (requeue &&
231		    test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
232			_debug("queue OBJ%x behind OBJ%x after wait",
233			       object->fscache.debug_id,
234			       xobject->fscache.debug_id);
235			goto requeue;
236		}
237
238		if (timeout <= 0) {
239			pr_err("\n");
240			pr_err("Error: Overlong wait for old active object to go away\n");
241			cachefiles_printk_object(object, xobject);
242			goto requeue;
243		}
244	}
245
246	ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
247
248	cache->cache.ops->put_object(&xobject->fscache);
249	goto try_again;
250
251requeue:
252	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
253	cache->cache.ops->put_object(&xobject->fscache);
254	_leave(" = -ETIMEDOUT");
255	return -ETIMEDOUT;
256}
257
258/*
259 * delete an object representation from the cache
260 * - file backed objects are unlinked
261 * - directory backed objects are stuffed into the graveyard for userspace to
262 *   delete
263 * - unlocks the directory mutex
264 */
265static int cachefiles_bury_object(struct cachefiles_cache *cache,
266				  struct dentry *dir,
267				  struct dentry *rep,
268				  bool preemptive)
269{
270	struct dentry *grave, *trap;
271	struct path path, path_to_graveyard;
272	char nbuffer[8 + 8 + 1];
273	int ret;
274
275	_enter(",'%pd','%pd'", dir, rep);
276
277	_debug("remove %p from %p", rep, dir);
278
279	/* non-directories can just be unlinked */
280	if (!d_is_dir(rep)) {
281		_debug("unlink stale object");
282
283		path.mnt = cache->mnt;
284		path.dentry = dir;
285		ret = security_path_unlink(&path, rep);
286		if (ret < 0) {
287			cachefiles_io_error(cache, "Unlink security error");
288		} else {
289			ret = vfs_unlink(d_inode(dir), rep, NULL);
290
291			if (preemptive)
292				cachefiles_mark_object_buried(cache, rep);
293		}
294
295		mutex_unlock(&d_inode(dir)->i_mutex);
296
297		if (ret == -EIO)
298			cachefiles_io_error(cache, "Unlink failed");
299
300		_leave(" = %d", ret);
301		return ret;
302	}
303
304	/* directories have to be moved to the graveyard */
305	_debug("move stale object to graveyard");
306	mutex_unlock(&d_inode(dir)->i_mutex);
307
308try_again:
309	/* first step is to make up a grave dentry in the graveyard */
310	sprintf(nbuffer, "%08x%08x",
311		(uint32_t) get_seconds(),
312		(uint32_t) atomic_inc_return(&cache->gravecounter));
313
314	/* do the multiway lock magic */
315	trap = lock_rename(cache->graveyard, dir);
316
317	/* do some checks before getting the grave dentry */
318	if (rep->d_parent != dir) {
319		/* the entry was probably culled when we dropped the parent dir
320		 * lock */
321		unlock_rename(cache->graveyard, dir);
322		_leave(" = 0 [culled?]");
323		return 0;
324	}
325
326	if (!d_can_lookup(cache->graveyard)) {
327		unlock_rename(cache->graveyard, dir);
328		cachefiles_io_error(cache, "Graveyard no longer a directory");
329		return -EIO;
330	}
331
332	if (trap == rep) {
333		unlock_rename(cache->graveyard, dir);
334		cachefiles_io_error(cache, "May not make directory loop");
335		return -EIO;
336	}
337
338	if (d_mountpoint(rep)) {
339		unlock_rename(cache->graveyard, dir);
340		cachefiles_io_error(cache, "Mountpoint in cache");
341		return -EIO;
342	}
343
344	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
345	if (IS_ERR(grave)) {
346		unlock_rename(cache->graveyard, dir);
347
348		if (PTR_ERR(grave) == -ENOMEM) {
349			_leave(" = -ENOMEM");
350			return -ENOMEM;
351		}
352
353		cachefiles_io_error(cache, "Lookup error %ld",
354				    PTR_ERR(grave));
355		return -EIO;
356	}
357
358	if (d_is_positive(grave)) {
359		unlock_rename(cache->graveyard, dir);
360		dput(grave);
361		grave = NULL;
362		cond_resched();
363		goto try_again;
364	}
365
366	if (d_mountpoint(grave)) {
367		unlock_rename(cache->graveyard, dir);
368		dput(grave);
369		cachefiles_io_error(cache, "Mountpoint in graveyard");
370		return -EIO;
371	}
372
373	/* target should not be an ancestor of source */
374	if (trap == grave) {
375		unlock_rename(cache->graveyard, dir);
376		dput(grave);
377		cachefiles_io_error(cache, "May not make directory loop");
378		return -EIO;
379	}
380
381	/* attempt the rename */
382	path.mnt = cache->mnt;
383	path.dentry = dir;
384	path_to_graveyard.mnt = cache->mnt;
385	path_to_graveyard.dentry = cache->graveyard;
386	ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
387	if (ret < 0) {
388		cachefiles_io_error(cache, "Rename security error %d", ret);
389	} else {
390		ret = vfs_rename(d_inode(dir), rep,
391				 d_inode(cache->graveyard), grave, NULL, 0);
392		if (ret != 0 && ret != -ENOMEM)
393			cachefiles_io_error(cache,
394					    "Rename failed with error %d", ret);
395
396		if (preemptive)
397			cachefiles_mark_object_buried(cache, rep);
398	}
399
400	unlock_rename(cache->graveyard, dir);
401	dput(grave);
402	_leave(" = 0");
403	return 0;
404}
405
406/*
407 * delete an object representation from the cache
408 */
409int cachefiles_delete_object(struct cachefiles_cache *cache,
410			     struct cachefiles_object *object)
411{
412	struct dentry *dir;
413	int ret;
414
415	_enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
416
417	ASSERT(object->dentry);
418	ASSERT(d_backing_inode(object->dentry));
419	ASSERT(object->dentry->d_parent);
420
421	dir = dget_parent(object->dentry);
422
423	mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
424
425	if (test_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
426		/* object allocation for the same key preemptively deleted this
427		 * object's file so that it could create its own file */
428		_debug("object preemptively buried");
429		mutex_unlock(&d_inode(dir)->i_mutex);
430		ret = 0;
431	} else {
432		/* we need to check that our parent is _still_ our parent - it
433		 * may have been renamed */
434		if (dir == object->dentry->d_parent) {
435			ret = cachefiles_bury_object(cache, dir,
436						     object->dentry, false);
437		} else {
438			/* it got moved, presumably by cachefilesd culling it,
439			 * so it's no longer in the key path and we can ignore
440			 * it */
441			mutex_unlock(&d_inode(dir)->i_mutex);
442			ret = 0;
443		}
444	}
445
446	dput(dir);
447	_leave(" = %d", ret);
448	return ret;
449}
450
451/*
452 * walk from the parent object to the child object through the backing
453 * filesystem, creating directories as we go
454 */
455int cachefiles_walk_to_object(struct cachefiles_object *parent,
456			      struct cachefiles_object *object,
457			      const char *key,
458			      struct cachefiles_xattr *auxdata)
459{
460	struct cachefiles_cache *cache;
461	struct dentry *dir, *next = NULL;
462	struct path path;
463	unsigned long start;
464	const char *name;
465	int ret, nlen;
466
467	_enter("OBJ%x{%p},OBJ%x,%s,",
468	       parent->fscache.debug_id, parent->dentry,
469	       object->fscache.debug_id, key);
470
471	cache = container_of(parent->fscache.cache,
472			     struct cachefiles_cache, cache);
473	path.mnt = cache->mnt;
474
475	ASSERT(parent->dentry);
476	ASSERT(d_backing_inode(parent->dentry));
477
478	if (!(d_is_dir(parent->dentry))) {
479		// TODO: convert file to dir
480		_leave("looking up in none directory");
481		return -ENOBUFS;
482	}
483
484	dir = dget(parent->dentry);
485
486advance:
487	/* attempt to transit the first directory component */
488	name = key;
489	nlen = strlen(key);
490
491	/* key ends in a double NUL */
492	key = key + nlen + 1;
493	if (!*key)
494		key = NULL;
495
496lookup_again:
497	/* search the current directory for the element name */
498	_debug("lookup '%s'", name);
499
500	mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
501
502	start = jiffies;
503	next = lookup_one_len(name, dir, nlen);
504	cachefiles_hist(cachefiles_lookup_histogram, start);
505	if (IS_ERR(next))
506		goto lookup_error;
507
508	_debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
509
510	if (!key)
511		object->new = !d_backing_inode(next);
512
513	/* if this element of the path doesn't exist, then the lookup phase
514	 * failed, and we can release any readers in the certain knowledge that
515	 * there's nothing for them to actually read */
516	if (d_is_negative(next))
517		fscache_object_lookup_negative(&object->fscache);
518
519	/* we need to create the object if it's negative */
520	if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
521		/* index objects and intervening tree levels must be subdirs */
522		if (d_is_negative(next)) {
523			ret = cachefiles_has_space(cache, 1, 0);
524			if (ret < 0)
525				goto create_error;
526
527			path.dentry = dir;
528			ret = security_path_mkdir(&path, next, 0);
529			if (ret < 0)
530				goto create_error;
531			start = jiffies;
532			ret = vfs_mkdir(d_inode(dir), next, 0);
533			cachefiles_hist(cachefiles_mkdir_histogram, start);
534			if (ret < 0)
535				goto create_error;
536
537			ASSERT(d_backing_inode(next));
538
539			_debug("mkdir -> %p{%p{ino=%lu}}",
540			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
541
542		} else if (!d_can_lookup(next)) {
543			pr_err("inode %lu is not a directory\n",
544			       d_backing_inode(next)->i_ino);
545			ret = -ENOBUFS;
546			goto error;
547		}
548
549	} else {
550		/* non-index objects start out life as files */
551		if (d_is_negative(next)) {
552			ret = cachefiles_has_space(cache, 1, 0);
553			if (ret < 0)
554				goto create_error;
555
556			path.dentry = dir;
557			ret = security_path_mknod(&path, next, S_IFREG, 0);
558			if (ret < 0)
559				goto create_error;
560			start = jiffies;
561			ret = vfs_create(d_inode(dir), next, S_IFREG, true);
562			cachefiles_hist(cachefiles_create_histogram, start);
563			if (ret < 0)
564				goto create_error;
565
566			ASSERT(d_backing_inode(next));
567
568			_debug("create -> %p{%p{ino=%lu}}",
569			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
570
571		} else if (!d_can_lookup(next) &&
572			   !d_is_reg(next)
573			   ) {
574			pr_err("inode %lu is not a file or directory\n",
575			       d_backing_inode(next)->i_ino);
576			ret = -ENOBUFS;
577			goto error;
578		}
579	}
580
581	/* process the next component */
582	if (key) {
583		_debug("advance");
584		mutex_unlock(&d_inode(dir)->i_mutex);
585		dput(dir);
586		dir = next;
587		next = NULL;
588		goto advance;
589	}
590
591	/* we've found the object we were looking for */
592	object->dentry = next;
593
594	/* if we've found that the terminal object exists, then we need to
595	 * check its attributes and delete it if it's out of date */
596	if (!object->new) {
597		_debug("validate '%pd'", next);
598
599		ret = cachefiles_check_object_xattr(object, auxdata);
600		if (ret == -ESTALE) {
601			/* delete the object (the deleter drops the directory
602			 * mutex) */
603			object->dentry = NULL;
604
605			ret = cachefiles_bury_object(cache, dir, next, true);
606			dput(next);
607			next = NULL;
608
609			if (ret < 0)
610				goto delete_error;
611
612			_debug("redo lookup");
613			goto lookup_again;
614		}
615	}
616
617	/* note that we're now using this object */
618	ret = cachefiles_mark_object_active(cache, object);
619
620	mutex_unlock(&d_inode(dir)->i_mutex);
621	dput(dir);
622	dir = NULL;
623
624	if (ret == -ETIMEDOUT)
625		goto mark_active_timed_out;
626
627	_debug("=== OBTAINED_OBJECT ===");
628
629	if (object->new) {
630		/* attach data to a newly constructed terminal object */
631		ret = cachefiles_set_object_xattr(object, auxdata);
632		if (ret < 0)
633			goto check_error;
634	} else {
635		/* always update the atime on an object we've just looked up
636		 * (this is used to keep track of culling, and atimes are only
637		 * updated by read, write and readdir but not lookup or
638		 * open) */
639		path.dentry = next;
640		touch_atime(&path);
641	}
642
643	/* open a file interface onto a data file */
644	if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
645		if (d_is_reg(object->dentry)) {
646			const struct address_space_operations *aops;
647
648			ret = -EPERM;
649			aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
650			if (!aops->bmap)
651				goto check_error;
652
653			object->backer = object->dentry;
654		} else {
655			BUG(); // TODO: open file in data-class subdir
656		}
657	}
658
659	object->new = 0;
660	fscache_obtained_object(&object->fscache);
661
662	_leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
663	return 0;
664
665create_error:
666	_debug("create error %d", ret);
667	if (ret == -EIO)
668		cachefiles_io_error(cache, "Create/mkdir failed");
669	goto error;
670
671mark_active_timed_out:
672	_debug("mark active timed out");
673	goto release_dentry;
674
675check_error:
676	_debug("check error %d", ret);
677	write_lock(&cache->active_lock);
678	rb_erase(&object->active_node, &cache->active_nodes);
679	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
680	wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
681	write_unlock(&cache->active_lock);
682release_dentry:
683	dput(object->dentry);
684	object->dentry = NULL;
685	goto error_out;
686
687delete_error:
688	_debug("delete error %d", ret);
689	goto error_out2;
690
691lookup_error:
692	_debug("lookup error %ld", PTR_ERR(next));
693	ret = PTR_ERR(next);
694	if (ret == -EIO)
695		cachefiles_io_error(cache, "Lookup failed");
696	next = NULL;
697error:
698	mutex_unlock(&d_inode(dir)->i_mutex);
699	dput(next);
700error_out2:
701	dput(dir);
702error_out:
703	_leave(" = error %d", -ret);
704	return ret;
705}
706
707/*
708 * get a subdirectory
709 */
710struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
711					struct dentry *dir,
712					const char *dirname)
713{
714	struct dentry *subdir;
715	unsigned long start;
716	struct path path;
717	int ret;
718
719	_enter(",,%s", dirname);
720
721	/* search the current directory for the element name */
722	mutex_lock(&d_inode(dir)->i_mutex);
723
724	start = jiffies;
725	subdir = lookup_one_len(dirname, dir, strlen(dirname));
726	cachefiles_hist(cachefiles_lookup_histogram, start);
727	if (IS_ERR(subdir)) {
728		if (PTR_ERR(subdir) == -ENOMEM)
729			goto nomem_d_alloc;
730		goto lookup_error;
731	}
732
733	_debug("subdir -> %p %s",
734	       subdir, d_backing_inode(subdir) ? "positive" : "negative");
735
736	/* we need to create the subdir if it doesn't exist yet */
737	if (d_is_negative(subdir)) {
738		ret = cachefiles_has_space(cache, 1, 0);
739		if (ret < 0)
740			goto mkdir_error;
741
742		_debug("attempt mkdir");
743
744		path.mnt = cache->mnt;
745		path.dentry = dir;
746		ret = security_path_mkdir(&path, subdir, 0700);
747		if (ret < 0)
748			goto mkdir_error;
749		ret = vfs_mkdir(d_inode(dir), subdir, 0700);
750		if (ret < 0)
751			goto mkdir_error;
752
753		ASSERT(d_backing_inode(subdir));
754
755		_debug("mkdir -> %p{%p{ino=%lu}}",
756		       subdir,
757		       d_backing_inode(subdir),
758		       d_backing_inode(subdir)->i_ino);
759	}
760
761	mutex_unlock(&d_inode(dir)->i_mutex);
762
763	/* we need to make sure the subdir is a directory */
764	ASSERT(d_backing_inode(subdir));
765
766	if (!d_can_lookup(subdir)) {
767		pr_err("%s is not a directory\n", dirname);
768		ret = -EIO;
769		goto check_error;
770	}
771
772	ret = -EPERM;
773	if (!d_backing_inode(subdir)->i_op->setxattr ||
774	    !d_backing_inode(subdir)->i_op->getxattr ||
775	    !d_backing_inode(subdir)->i_op->lookup ||
776	    !d_backing_inode(subdir)->i_op->mkdir ||
777	    !d_backing_inode(subdir)->i_op->create ||
778	    (!d_backing_inode(subdir)->i_op->rename &&
779	     !d_backing_inode(subdir)->i_op->rename2) ||
780	    !d_backing_inode(subdir)->i_op->rmdir ||
781	    !d_backing_inode(subdir)->i_op->unlink)
782		goto check_error;
783
784	_leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
785	return subdir;
786
787check_error:
788	dput(subdir);
789	_leave(" = %d [check]", ret);
790	return ERR_PTR(ret);
791
792mkdir_error:
793	mutex_unlock(&d_inode(dir)->i_mutex);
794	dput(subdir);
795	pr_err("mkdir %s failed with error %d\n", dirname, ret);
796	return ERR_PTR(ret);
797
798lookup_error:
799	mutex_unlock(&d_inode(dir)->i_mutex);
800	ret = PTR_ERR(subdir);
801	pr_err("Lookup %s failed with error %d\n", dirname, ret);
802	return ERR_PTR(ret);
803
804nomem_d_alloc:
805	mutex_unlock(&d_inode(dir)->i_mutex);
806	_leave(" = -ENOMEM");
807	return ERR_PTR(-ENOMEM);
808}
809
810/*
811 * find out if an object is in use or not
812 * - if finds object and it's not in use:
813 *   - returns a pointer to the object and a reference on it
814 *   - returns with the directory locked
815 */
816static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
817					      struct dentry *dir,
818					      char *filename)
819{
820	struct cachefiles_object *object;
821	struct rb_node *_n;
822	struct dentry *victim;
823	unsigned long start;
824	int ret;
825
826	//_enter(",%pd/,%s",
827	//       dir, filename);
828
829	/* look up the victim */
830	mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
831
832	start = jiffies;
833	victim = lookup_one_len(filename, dir, strlen(filename));
834	cachefiles_hist(cachefiles_lookup_histogram, start);
835	if (IS_ERR(victim))
836		goto lookup_error;
837
838	//_debug("victim -> %p %s",
839	//       victim, d_backing_inode(victim) ? "positive" : "negative");
840
841	/* if the object is no longer there then we probably retired the object
842	 * at the netfs's request whilst the cull was in progress
843	 */
844	if (d_is_negative(victim)) {
845		mutex_unlock(&d_inode(dir)->i_mutex);
846		dput(victim);
847		_leave(" = -ENOENT [absent]");
848		return ERR_PTR(-ENOENT);
849	}
850
851	/* check to see if we're using this object */
852	read_lock(&cache->active_lock);
853
854	_n = cache->active_nodes.rb_node;
855
856	while (_n) {
857		object = rb_entry(_n, struct cachefiles_object, active_node);
858
859		if (object->dentry > victim)
860			_n = _n->rb_left;
861		else if (object->dentry < victim)
862			_n = _n->rb_right;
863		else
864			goto object_in_use;
865	}
866
867	read_unlock(&cache->active_lock);
868
869	//_leave(" = %p", victim);
870	return victim;
871
872object_in_use:
873	read_unlock(&cache->active_lock);
874	mutex_unlock(&d_inode(dir)->i_mutex);
875	dput(victim);
876	//_leave(" = -EBUSY [in use]");
877	return ERR_PTR(-EBUSY);
878
879lookup_error:
880	mutex_unlock(&d_inode(dir)->i_mutex);
881	ret = PTR_ERR(victim);
882	if (ret == -ENOENT) {
883		/* file or dir now absent - probably retired by netfs */
884		_leave(" = -ESTALE [absent]");
885		return ERR_PTR(-ESTALE);
886	}
887
888	if (ret == -EIO) {
889		cachefiles_io_error(cache, "Lookup failed");
890	} else if (ret != -ENOMEM) {
891		pr_err("Internal error: %d\n", ret);
892		ret = -EIO;
893	}
894
895	_leave(" = %d", ret);
896	return ERR_PTR(ret);
897}
898
899/*
900 * cull an object if it's not in use
901 * - called only by cache manager daemon
902 */
903int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
904		    char *filename)
905{
906	struct dentry *victim;
907	int ret;
908
909	_enter(",%pd/,%s", dir, filename);
910
911	victim = cachefiles_check_active(cache, dir, filename);
912	if (IS_ERR(victim))
913		return PTR_ERR(victim);
914
915	_debug("victim -> %p %s",
916	       victim, d_backing_inode(victim) ? "positive" : "negative");
917
918	/* okay... the victim is not being used so we can cull it
919	 * - start by marking it as stale
920	 */
921	_debug("victim is cullable");
922
923	ret = cachefiles_remove_object_xattr(cache, victim);
924	if (ret < 0)
925		goto error_unlock;
926
927	/*  actually remove the victim (drops the dir mutex) */
928	_debug("bury");
929
930	ret = cachefiles_bury_object(cache, dir, victim, false);
931	if (ret < 0)
932		goto error;
933
934	dput(victim);
935	_leave(" = 0");
936	return 0;
937
938error_unlock:
939	mutex_unlock(&d_inode(dir)->i_mutex);
940error:
941	dput(victim);
942	if (ret == -ENOENT) {
943		/* file or dir now absent - probably retired by netfs */
944		_leave(" = -ESTALE [absent]");
945		return -ESTALE;
946	}
947
948	if (ret != -ENOMEM) {
949		pr_err("Internal error: %d\n", ret);
950		ret = -EIO;
951	}
952
953	_leave(" = %d", ret);
954	return ret;
955}
956
957/*
958 * find out if an object is in use or not
959 * - called only by cache manager daemon
960 * - returns -EBUSY or 0 to indicate whether an object is in use or not
961 */
962int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
963			    char *filename)
964{
965	struct dentry *victim;
966
967	//_enter(",%pd/,%s",
968	//       dir, filename);
969
970	victim = cachefiles_check_active(cache, dir, filename);
971	if (IS_ERR(victim))
972		return PTR_ERR(victim);
973
974	mutex_unlock(&d_inode(dir)->i_mutex);
975	dput(victim);
976	//_leave(" = 0");
977	return 0;
978}
979