Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux.git] / fs / cachefiles / namei.c
1 /* CacheFiles path walking and related routines
2  *
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public Licence
8  * as published by the Free Software Foundation; either version
9  * 2 of the Licence, or (at your option) any later version.
10  */
11
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/file.h>
15 #include <linux/fs.h>
16 #include <linux/fsnotify.h>
17 #include <linux/quotaops.h>
18 #include <linux/xattr.h>
19 #include <linux/mount.h>
20 #include <linux/namei.h>
21 #include <linux/security.h>
22 #include <linux/slab.h>
23 #include "internal.h"
24
25 #define CACHEFILES_KEYBUF_SIZE 512
26
27 /*
28  * dump debugging info about an object
29  */
30 static noinline
31 void __cachefiles_printk_object(struct cachefiles_object *object,
32                                 const char *prefix,
33                                 u8 *keybuf)
34 {
35         struct fscache_cookie *cookie;
36         unsigned keylen, loop;
37
38         printk(KERN_ERR "%sobject: OBJ%x\n",
39                prefix, object->fscache.debug_id);
40         printk(KERN_ERR "%sobjstate=%s fl=%lx 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         printk(KERN_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         printk(KERN_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                 printk(KERN_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                 if (keybuf && cookie->def)
60                         keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
61                                                       CACHEFILES_KEYBUF_SIZE);
62                 else
63                         keylen = 0;
64         } else {
65                 printk(KERN_ERR "%scookie=NULL\n", prefix);
66                 keylen = 0;
67         }
68         spin_unlock(&object->fscache.lock);
69
70         if (keylen) {
71                 printk(KERN_ERR "%skey=[%u] '", prefix, keylen);
72                 for (loop = 0; loop < keylen; loop++)
73                         printk("%02x", keybuf[loop]);
74                 printk("'\n");
75         }
76 }
77
78 /*
79  * dump debugging info about a pair of objects
80  */
81 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
82                                               struct cachefiles_object *xobject)
83 {
84         u8 *keybuf;
85
86         keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
87         if (object)
88                 __cachefiles_printk_object(object, "", keybuf);
89         if (xobject)
90                 __cachefiles_printk_object(xobject, "x", keybuf);
91         kfree(keybuf);
92 }
93
94 /*
95  * mark the owner of a dentry, if there is one, to indicate that that dentry
96  * has been preemptively deleted
97  * - the caller must hold the i_mutex on the dentry's parent as required to
98  *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
99  */
100 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
101                                           struct dentry *dentry)
102 {
103         struct cachefiles_object *object;
104         struct rb_node *p;
105
106         _enter(",'%*.*s'",
107                dentry->d_name.len, dentry->d_name.len, dentry->d_name.name);
108
109         write_lock(&cache->active_lock);
110
111         p = cache->active_nodes.rb_node;
112         while (p) {
113                 object = rb_entry(p, struct cachefiles_object, active_node);
114                 if (object->dentry > dentry)
115                         p = p->rb_left;
116                 else if (object->dentry < dentry)
117                         p = p->rb_right;
118                 else
119                         goto found_dentry;
120         }
121
122         write_unlock(&cache->active_lock);
123         _leave(" [no owner]");
124         return;
125
126         /* found the dentry for  */
127 found_dentry:
128         kdebug("preemptive burial: OBJ%x [%s] %p",
129                object->fscache.debug_id,
130                object->fscache.state->name,
131                dentry);
132
133         if (fscache_object_is_live(&object->fscache)) {
134                 printk(KERN_ERR "\n");
135                 printk(KERN_ERR "CacheFiles: Error:"
136                        " Can't preemptively bury live object\n");
137                 cachefiles_printk_object(object, NULL);
138         } else if (test_and_set_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
139                 printk(KERN_ERR "CacheFiles: Error:"
140                        " Object already preemptively buried\n");
141         }
142
143         write_unlock(&cache->active_lock);
144         _leave(" [owner marked]");
145 }
146
147 /*
148  * record the fact that an object is now active
149  */
150 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
151                                          struct cachefiles_object *object)
152 {
153         struct cachefiles_object *xobject;
154         struct rb_node **_p, *_parent = NULL;
155         struct dentry *dentry;
156
157         _enter(",%p", object);
158
159 try_again:
160         write_lock(&cache->active_lock);
161
162         if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
163                 printk(KERN_ERR "CacheFiles: Error: Object already active\n");
164                 cachefiles_printk_object(object, NULL);
165                 BUG();
166         }
167
168         dentry = object->dentry;
169         _p = &cache->active_nodes.rb_node;
170         while (*_p) {
171                 _parent = *_p;
172                 xobject = rb_entry(_parent,
173                                    struct cachefiles_object, active_node);
174
175                 ASSERT(xobject != object);
176
177                 if (xobject->dentry > dentry)
178                         _p = &(*_p)->rb_left;
179                 else if (xobject->dentry < dentry)
180                         _p = &(*_p)->rb_right;
181                 else
182                         goto wait_for_old_object;
183         }
184
185         rb_link_node(&object->active_node, _parent, _p);
186         rb_insert_color(&object->active_node, &cache->active_nodes);
187
188         write_unlock(&cache->active_lock);
189         _leave(" = 0");
190         return 0;
191
192         /* an old object from a previous incarnation is hogging the slot - we
193          * need to wait for it to be destroyed */
194 wait_for_old_object:
195         if (fscache_object_is_live(&object->fscache)) {
196                 printk(KERN_ERR "\n");
197                 printk(KERN_ERR "CacheFiles: Error:"
198                        " Unexpected object collision\n");
199                 cachefiles_printk_object(object, xobject);
200                 BUG();
201         }
202         atomic_inc(&xobject->usage);
203         write_unlock(&cache->active_lock);
204
205         if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
206                 wait_queue_head_t *wq;
207
208                 signed long timeout = 60 * HZ;
209                 wait_queue_t wait;
210                 bool requeue;
211
212                 /* if the object we're waiting for is queued for processing,
213                  * then just put ourselves on the queue behind it */
214                 if (work_pending(&xobject->fscache.work)) {
215                         _debug("queue OBJ%x behind OBJ%x immediately",
216                                object->fscache.debug_id,
217                                xobject->fscache.debug_id);
218                         goto requeue;
219                 }
220
221                 /* otherwise we sleep until either the object we're waiting for
222                  * is done, or the fscache_object is congested */
223                 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
224                 init_wait(&wait);
225                 requeue = false;
226                 do {
227                         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
228                         if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
229                                 break;
230
231                         requeue = fscache_object_sleep_till_congested(&timeout);
232                 } while (timeout > 0 && !requeue);
233                 finish_wait(wq, &wait);
234
235                 if (requeue &&
236                     test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
237                         _debug("queue OBJ%x behind OBJ%x after wait",
238                                object->fscache.debug_id,
239                                xobject->fscache.debug_id);
240                         goto requeue;
241                 }
242
243                 if (timeout <= 0) {
244                         printk(KERN_ERR "\n");
245                         printk(KERN_ERR "CacheFiles: Error: Overlong"
246                                " wait for old active object to go away\n");
247                         cachefiles_printk_object(object, xobject);
248                         goto requeue;
249                 }
250         }
251
252         ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
253
254         cache->cache.ops->put_object(&xobject->fscache);
255         goto try_again;
256
257 requeue:
258         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
259         cache->cache.ops->put_object(&xobject->fscache);
260         _leave(" = -ETIMEDOUT");
261         return -ETIMEDOUT;
262 }
263
264 /*
265  * delete an object representation from the cache
266  * - file backed objects are unlinked
267  * - directory backed objects are stuffed into the graveyard for userspace to
268  *   delete
269  * - unlocks the directory mutex
270  */
271 static int cachefiles_bury_object(struct cachefiles_cache *cache,
272                                   struct dentry *dir,
273                                   struct dentry *rep,
274                                   bool preemptive)
275 {
276         struct dentry *grave, *trap;
277         struct path path, path_to_graveyard;
278         char nbuffer[8 + 8 + 1];
279         int ret;
280
281         _enter(",'%*.*s','%*.*s'",
282                dir->d_name.len, dir->d_name.len, dir->d_name.name,
283                rep->d_name.len, rep->d_name.len, rep->d_name.name);
284
285         _debug("remove %p from %p", rep, dir);
286
287         /* non-directories can just be unlinked */
288         if (!S_ISDIR(rep->d_inode->i_mode)) {
289                 _debug("unlink stale object");
290
291                 path.mnt = cache->mnt;
292                 path.dentry = dir;
293                 ret = security_path_unlink(&path, rep);
294                 if (ret < 0) {
295                         cachefiles_io_error(cache, "Unlink security error");
296                 } else {
297                         ret = vfs_unlink(dir->d_inode, rep, NULL);
298
299                         if (preemptive)
300                                 cachefiles_mark_object_buried(cache, rep);
301                 }
302
303                 mutex_unlock(&dir->d_inode->i_mutex);
304
305                 if (ret == -EIO)
306                         cachefiles_io_error(cache, "Unlink failed");
307
308                 _leave(" = %d", ret);
309                 return ret;
310         }
311
312         /* directories have to be moved to the graveyard */
313         _debug("move stale object to graveyard");
314         mutex_unlock(&dir->d_inode->i_mutex);
315
316 try_again:
317         /* first step is to make up a grave dentry in the graveyard */
318         sprintf(nbuffer, "%08x%08x",
319                 (uint32_t) get_seconds(),
320                 (uint32_t) atomic_inc_return(&cache->gravecounter));
321
322         /* do the multiway lock magic */
323         trap = lock_rename(cache->graveyard, dir);
324
325         /* do some checks before getting the grave dentry */
326         if (rep->d_parent != dir) {
327                 /* the entry was probably culled when we dropped the parent dir
328                  * lock */
329                 unlock_rename(cache->graveyard, dir);
330                 _leave(" = 0 [culled?]");
331                 return 0;
332         }
333
334         if (!S_ISDIR(cache->graveyard->d_inode->i_mode)) {
335                 unlock_rename(cache->graveyard, dir);
336                 cachefiles_io_error(cache, "Graveyard no longer a directory");
337                 return -EIO;
338         }
339
340         if (trap == rep) {
341                 unlock_rename(cache->graveyard, dir);
342                 cachefiles_io_error(cache, "May not make directory loop");
343                 return -EIO;
344         }
345
346         if (d_mountpoint(rep)) {
347                 unlock_rename(cache->graveyard, dir);
348                 cachefiles_io_error(cache, "Mountpoint in cache");
349                 return -EIO;
350         }
351
352         grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
353         if (IS_ERR(grave)) {
354                 unlock_rename(cache->graveyard, dir);
355
356                 if (PTR_ERR(grave) == -ENOMEM) {
357                         _leave(" = -ENOMEM");
358                         return -ENOMEM;
359                 }
360
361                 cachefiles_io_error(cache, "Lookup error %ld",
362                                     PTR_ERR(grave));
363                 return -EIO;
364         }
365
366         if (grave->d_inode) {
367                 unlock_rename(cache->graveyard, dir);
368                 dput(grave);
369                 grave = NULL;
370                 cond_resched();
371                 goto try_again;
372         }
373
374         if (d_mountpoint(grave)) {
375                 unlock_rename(cache->graveyard, dir);
376                 dput(grave);
377                 cachefiles_io_error(cache, "Mountpoint in graveyard");
378                 return -EIO;
379         }
380
381         /* target should not be an ancestor of source */
382         if (trap == grave) {
383                 unlock_rename(cache->graveyard, dir);
384                 dput(grave);
385                 cachefiles_io_error(cache, "May not make directory loop");
386                 return -EIO;
387         }
388
389         /* attempt the rename */
390         path.mnt = cache->mnt;
391         path.dentry = dir;
392         path_to_graveyard.mnt = cache->mnt;
393         path_to_graveyard.dentry = cache->graveyard;
394         ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
395         if (ret < 0) {
396                 cachefiles_io_error(cache, "Rename security error %d", ret);
397         } else {
398                 ret = vfs_rename(dir->d_inode, rep,
399                                  cache->graveyard->d_inode, grave, NULL, 0);
400                 if (ret != 0 && ret != -ENOMEM)
401                         cachefiles_io_error(cache,
402                                             "Rename failed with error %d", ret);
403
404                 if (preemptive)
405                         cachefiles_mark_object_buried(cache, rep);
406         }
407
408         unlock_rename(cache->graveyard, dir);
409         dput(grave);
410         _leave(" = 0");
411         return 0;
412 }
413
414 /*
415  * delete an object representation from the cache
416  */
417 int cachefiles_delete_object(struct cachefiles_cache *cache,
418                              struct cachefiles_object *object)
419 {
420         struct dentry *dir;
421         int ret;
422
423         _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
424
425         ASSERT(object->dentry);
426         ASSERT(object->dentry->d_inode);
427         ASSERT(object->dentry->d_parent);
428
429         dir = dget_parent(object->dentry);
430
431         mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
432
433         if (test_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
434                 /* object allocation for the same key preemptively deleted this
435                  * object's file so that it could create its own file */
436                 _debug("object preemptively buried");
437                 mutex_unlock(&dir->d_inode->i_mutex);
438                 ret = 0;
439         } else {
440                 /* we need to check that our parent is _still_ our parent - it
441                  * may have been renamed */
442                 if (dir == object->dentry->d_parent) {
443                         ret = cachefiles_bury_object(cache, dir,
444                                                      object->dentry, false);
445                 } else {
446                         /* it got moved, presumably by cachefilesd culling it,
447                          * so it's no longer in the key path and we can ignore
448                          * it */
449                         mutex_unlock(&dir->d_inode->i_mutex);
450                         ret = 0;
451                 }
452         }
453
454         dput(dir);
455         _leave(" = %d", ret);
456         return ret;
457 }
458
459 /*
460  * walk from the parent object to the child object through the backing
461  * filesystem, creating directories as we go
462  */
463 int cachefiles_walk_to_object(struct cachefiles_object *parent,
464                               struct cachefiles_object *object,
465                               const char *key,
466                               struct cachefiles_xattr *auxdata)
467 {
468         struct cachefiles_cache *cache;
469         struct dentry *dir, *next = NULL;
470         struct path path;
471         unsigned long start;
472         const char *name;
473         int ret, nlen;
474
475         _enter("OBJ%x{%p},OBJ%x,%s,",
476                parent->fscache.debug_id, parent->dentry,
477                object->fscache.debug_id, key);
478
479         cache = container_of(parent->fscache.cache,
480                              struct cachefiles_cache, cache);
481         path.mnt = cache->mnt;
482
483         ASSERT(parent->dentry);
484         ASSERT(parent->dentry->d_inode);
485
486         if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
487                 // TODO: convert file to dir
488                 _leave("looking up in none directory");
489                 return -ENOBUFS;
490         }
491
492         dir = dget(parent->dentry);
493
494 advance:
495         /* attempt to transit the first directory component */
496         name = key;
497         nlen = strlen(key);
498
499         /* key ends in a double NUL */
500         key = key + nlen + 1;
501         if (!*key)
502                 key = NULL;
503
504 lookup_again:
505         /* search the current directory for the element name */
506         _debug("lookup '%s'", name);
507
508         mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
509
510         start = jiffies;
511         next = lookup_one_len(name, dir, nlen);
512         cachefiles_hist(cachefiles_lookup_histogram, start);
513         if (IS_ERR(next))
514                 goto lookup_error;
515
516         _debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
517
518         if (!key)
519                 object->new = !next->d_inode;
520
521         /* if this element of the path doesn't exist, then the lookup phase
522          * failed, and we can release any readers in the certain knowledge that
523          * there's nothing for them to actually read */
524         if (!next->d_inode)
525                 fscache_object_lookup_negative(&object->fscache);
526
527         /* we need to create the object if it's negative */
528         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
529                 /* index objects and intervening tree levels must be subdirs */
530                 if (!next->d_inode) {
531                         ret = cachefiles_has_space(cache, 1, 0);
532                         if (ret < 0)
533                                 goto create_error;
534
535                         path.dentry = dir;
536                         ret = security_path_mkdir(&path, next, 0);
537                         if (ret < 0)
538                                 goto create_error;
539                         start = jiffies;
540                         ret = vfs_mkdir(dir->d_inode, next, 0);
541                         cachefiles_hist(cachefiles_mkdir_histogram, start);
542                         if (ret < 0)
543                                 goto create_error;
544
545                         ASSERT(next->d_inode);
546
547                         _debug("mkdir -> %p{%p{ino=%lu}}",
548                                next, next->d_inode, next->d_inode->i_ino);
549
550                 } else if (!S_ISDIR(next->d_inode->i_mode)) {
551                         kerror("inode %lu is not a directory",
552                                next->d_inode->i_ino);
553                         ret = -ENOBUFS;
554                         goto error;
555                 }
556
557         } else {
558                 /* non-index objects start out life as files */
559                 if (!next->d_inode) {
560                         ret = cachefiles_has_space(cache, 1, 0);
561                         if (ret < 0)
562                                 goto create_error;
563
564                         path.dentry = dir;
565                         ret = security_path_mknod(&path, next, S_IFREG, 0);
566                         if (ret < 0)
567                                 goto create_error;
568                         start = jiffies;
569                         ret = vfs_create(dir->d_inode, next, S_IFREG, true);
570                         cachefiles_hist(cachefiles_create_histogram, start);
571                         if (ret < 0)
572                                 goto create_error;
573
574                         ASSERT(next->d_inode);
575
576                         _debug("create -> %p{%p{ino=%lu}}",
577                                next, next->d_inode, next->d_inode->i_ino);
578
579                 } else if (!S_ISDIR(next->d_inode->i_mode) &&
580                            !S_ISREG(next->d_inode->i_mode)
581                            ) {
582                         kerror("inode %lu is not a file or directory",
583                                next->d_inode->i_ino);
584                         ret = -ENOBUFS;
585                         goto error;
586                 }
587         }
588
589         /* process the next component */
590         if (key) {
591                 _debug("advance");
592                 mutex_unlock(&dir->d_inode->i_mutex);
593                 dput(dir);
594                 dir = next;
595                 next = NULL;
596                 goto advance;
597         }
598
599         /* we've found the object we were looking for */
600         object->dentry = next;
601
602         /* if we've found that the terminal object exists, then we need to
603          * check its attributes and delete it if it's out of date */
604         if (!object->new) {
605                 _debug("validate '%*.*s'",
606                        next->d_name.len, next->d_name.len, next->d_name.name);
607
608                 ret = cachefiles_check_object_xattr(object, auxdata);
609                 if (ret == -ESTALE) {
610                         /* delete the object (the deleter drops the directory
611                          * mutex) */
612                         object->dentry = NULL;
613
614                         ret = cachefiles_bury_object(cache, dir, next, true);
615                         dput(next);
616                         next = NULL;
617
618                         if (ret < 0)
619                                 goto delete_error;
620
621                         _debug("redo lookup");
622                         goto lookup_again;
623                 }
624         }
625
626         /* note that we're now using this object */
627         ret = cachefiles_mark_object_active(cache, object);
628
629         mutex_unlock(&dir->d_inode->i_mutex);
630         dput(dir);
631         dir = NULL;
632
633         if (ret == -ETIMEDOUT)
634                 goto mark_active_timed_out;
635
636         _debug("=== OBTAINED_OBJECT ===");
637
638         if (object->new) {
639                 /* attach data to a newly constructed terminal object */
640                 ret = cachefiles_set_object_xattr(object, auxdata);
641                 if (ret < 0)
642                         goto check_error;
643         } else {
644                 /* always update the atime on an object we've just looked up
645                  * (this is used to keep track of culling, and atimes are only
646                  * updated by read, write and readdir but not lookup or
647                  * open) */
648                 path.dentry = next;
649                 touch_atime(&path);
650         }
651
652         /* open a file interface onto a data file */
653         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
654                 if (S_ISREG(object->dentry->d_inode->i_mode)) {
655                         const struct address_space_operations *aops;
656
657                         ret = -EPERM;
658                         aops = object->dentry->d_inode->i_mapping->a_ops;
659                         if (!aops->bmap)
660                                 goto check_error;
661
662                         object->backer = object->dentry;
663                 } else {
664                         BUG(); // TODO: open file in data-class subdir
665                 }
666         }
667
668         object->new = 0;
669         fscache_obtained_object(&object->fscache);
670
671         _leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
672         return 0;
673
674 create_error:
675         _debug("create error %d", ret);
676         if (ret == -EIO)
677                 cachefiles_io_error(cache, "Create/mkdir failed");
678         goto error;
679
680 mark_active_timed_out:
681         _debug("mark active timed out");
682         goto release_dentry;
683
684 check_error:
685         _debug("check error %d", ret);
686         write_lock(&cache->active_lock);
687         rb_erase(&object->active_node, &cache->active_nodes);
688         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
689         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
690         write_unlock(&cache->active_lock);
691 release_dentry:
692         dput(object->dentry);
693         object->dentry = NULL;
694         goto error_out;
695
696 delete_error:
697         _debug("delete error %d", ret);
698         goto error_out2;
699
700 lookup_error:
701         _debug("lookup error %ld", PTR_ERR(next));
702         ret = PTR_ERR(next);
703         if (ret == -EIO)
704                 cachefiles_io_error(cache, "Lookup failed");
705         next = NULL;
706 error:
707         mutex_unlock(&dir->d_inode->i_mutex);
708         dput(next);
709 error_out2:
710         dput(dir);
711 error_out:
712         _leave(" = error %d", -ret);
713         return ret;
714 }
715
716 /*
717  * get a subdirectory
718  */
719 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
720                                         struct dentry *dir,
721                                         const char *dirname)
722 {
723         struct dentry *subdir;
724         unsigned long start;
725         struct path path;
726         int ret;
727
728         _enter(",,%s", dirname);
729
730         /* search the current directory for the element name */
731         mutex_lock(&dir->d_inode->i_mutex);
732
733         start = jiffies;
734         subdir = lookup_one_len(dirname, dir, strlen(dirname));
735         cachefiles_hist(cachefiles_lookup_histogram, start);
736         if (IS_ERR(subdir)) {
737                 if (PTR_ERR(subdir) == -ENOMEM)
738                         goto nomem_d_alloc;
739                 goto lookup_error;
740         }
741
742         _debug("subdir -> %p %s",
743                subdir, subdir->d_inode ? "positive" : "negative");
744
745         /* we need to create the subdir if it doesn't exist yet */
746         if (!subdir->d_inode) {
747                 ret = cachefiles_has_space(cache, 1, 0);
748                 if (ret < 0)
749                         goto mkdir_error;
750
751                 _debug("attempt mkdir");
752
753                 path.mnt = cache->mnt;
754                 path.dentry = dir;
755                 ret = security_path_mkdir(&path, subdir, 0700);
756                 if (ret < 0)
757                         goto mkdir_error;
758                 ret = vfs_mkdir(dir->d_inode, subdir, 0700);
759                 if (ret < 0)
760                         goto mkdir_error;
761
762                 ASSERT(subdir->d_inode);
763
764                 _debug("mkdir -> %p{%p{ino=%lu}}",
765                        subdir,
766                        subdir->d_inode,
767                        subdir->d_inode->i_ino);
768         }
769
770         mutex_unlock(&dir->d_inode->i_mutex);
771
772         /* we need to make sure the subdir is a directory */
773         ASSERT(subdir->d_inode);
774
775         if (!S_ISDIR(subdir->d_inode->i_mode)) {
776                 kerror("%s is not a directory", dirname);
777                 ret = -EIO;
778                 goto check_error;
779         }
780
781         ret = -EPERM;
782         if (!subdir->d_inode->i_op->setxattr ||
783             !subdir->d_inode->i_op->getxattr ||
784             !subdir->d_inode->i_op->lookup ||
785             !subdir->d_inode->i_op->mkdir ||
786             !subdir->d_inode->i_op->create ||
787             !subdir->d_inode->i_op->rename ||
788             !subdir->d_inode->i_op->rmdir ||
789             !subdir->d_inode->i_op->unlink)
790                 goto check_error;
791
792         _leave(" = [%lu]", subdir->d_inode->i_ino);
793         return subdir;
794
795 check_error:
796         dput(subdir);
797         _leave(" = %d [check]", ret);
798         return ERR_PTR(ret);
799
800 mkdir_error:
801         mutex_unlock(&dir->d_inode->i_mutex);
802         dput(subdir);
803         kerror("mkdir %s failed with error %d", dirname, ret);
804         return ERR_PTR(ret);
805
806 lookup_error:
807         mutex_unlock(&dir->d_inode->i_mutex);
808         ret = PTR_ERR(subdir);
809         kerror("Lookup %s failed with error %d", dirname, ret);
810         return ERR_PTR(ret);
811
812 nomem_d_alloc:
813         mutex_unlock(&dir->d_inode->i_mutex);
814         _leave(" = -ENOMEM");
815         return ERR_PTR(-ENOMEM);
816 }
817
818 /*
819  * find out if an object is in use or not
820  * - if finds object and it's not in use:
821  *   - returns a pointer to the object and a reference on it
822  *   - returns with the directory locked
823  */
824 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
825                                               struct dentry *dir,
826                                               char *filename)
827 {
828         struct cachefiles_object *object;
829         struct rb_node *_n;
830         struct dentry *victim;
831         unsigned long start;
832         int ret;
833
834         //_enter(",%*.*s/,%s",
835         //       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
836
837         /* look up the victim */
838         mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
839
840         start = jiffies;
841         victim = lookup_one_len(filename, dir, strlen(filename));
842         cachefiles_hist(cachefiles_lookup_histogram, start);
843         if (IS_ERR(victim))
844                 goto lookup_error;
845
846         //_debug("victim -> %p %s",
847         //       victim, victim->d_inode ? "positive" : "negative");
848
849         /* if the object is no longer there then we probably retired the object
850          * at the netfs's request whilst the cull was in progress
851          */
852         if (!victim->d_inode) {
853                 mutex_unlock(&dir->d_inode->i_mutex);
854                 dput(victim);
855                 _leave(" = -ENOENT [absent]");
856                 return ERR_PTR(-ENOENT);
857         }
858
859         /* check to see if we're using this object */
860         read_lock(&cache->active_lock);
861
862         _n = cache->active_nodes.rb_node;
863
864         while (_n) {
865                 object = rb_entry(_n, struct cachefiles_object, active_node);
866
867                 if (object->dentry > victim)
868                         _n = _n->rb_left;
869                 else if (object->dentry < victim)
870                         _n = _n->rb_right;
871                 else
872                         goto object_in_use;
873         }
874
875         read_unlock(&cache->active_lock);
876
877         //_leave(" = %p", victim);
878         return victim;
879
880 object_in_use:
881         read_unlock(&cache->active_lock);
882         mutex_unlock(&dir->d_inode->i_mutex);
883         dput(victim);
884         //_leave(" = -EBUSY [in use]");
885         return ERR_PTR(-EBUSY);
886
887 lookup_error:
888         mutex_unlock(&dir->d_inode->i_mutex);
889         ret = PTR_ERR(victim);
890         if (ret == -ENOENT) {
891                 /* file or dir now absent - probably retired by netfs */
892                 _leave(" = -ESTALE [absent]");
893                 return ERR_PTR(-ESTALE);
894         }
895
896         if (ret == -EIO) {
897                 cachefiles_io_error(cache, "Lookup failed");
898         } else if (ret != -ENOMEM) {
899                 kerror("Internal error: %d", ret);
900                 ret = -EIO;
901         }
902
903         _leave(" = %d", ret);
904         return ERR_PTR(ret);
905 }
906
907 /*
908  * cull an object if it's not in use
909  * - called only by cache manager daemon
910  */
911 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
912                     char *filename)
913 {
914         struct dentry *victim;
915         int ret;
916
917         _enter(",%*.*s/,%s",
918                dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
919
920         victim = cachefiles_check_active(cache, dir, filename);
921         if (IS_ERR(victim))
922                 return PTR_ERR(victim);
923
924         _debug("victim -> %p %s",
925                victim, victim->d_inode ? "positive" : "negative");
926
927         /* okay... the victim is not being used so we can cull it
928          * - start by marking it as stale
929          */
930         _debug("victim is cullable");
931
932         ret = cachefiles_remove_object_xattr(cache, victim);
933         if (ret < 0)
934                 goto error_unlock;
935
936         /*  actually remove the victim (drops the dir mutex) */
937         _debug("bury");
938
939         ret = cachefiles_bury_object(cache, dir, victim, false);
940         if (ret < 0)
941                 goto error;
942
943         dput(victim);
944         _leave(" = 0");
945         return 0;
946
947 error_unlock:
948         mutex_unlock(&dir->d_inode->i_mutex);
949 error:
950         dput(victim);
951         if (ret == -ENOENT) {
952                 /* file or dir now absent - probably retired by netfs */
953                 _leave(" = -ESTALE [absent]");
954                 return -ESTALE;
955         }
956
957         if (ret != -ENOMEM) {
958                 kerror("Internal error: %d", ret);
959                 ret = -EIO;
960         }
961
962         _leave(" = %d", ret);
963         return ret;
964 }
965
966 /*
967  * find out if an object is in use or not
968  * - called only by cache manager daemon
969  * - returns -EBUSY or 0 to indicate whether an object is in use or not
970  */
971 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
972                             char *filename)
973 {
974         struct dentry *victim;
975
976         //_enter(",%*.*s/,%s",
977         //       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
978
979         victim = cachefiles_check_active(cache, dir, filename);
980         if (IS_ERR(victim))
981                 return PTR_ERR(victim);
982
983         mutex_unlock(&dir->d_inode->i_mutex);
984         dput(victim);
985         //_leave(" = 0");
986         return 0;
987 }