vfs: add helpers to get root and pwd
[linux-2.6.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, fscache_object_states[object->fscache.state],
42                object->fscache.flags, work_busy(&object->fscache.work),
43                object->fscache.events,
44                object->fscache.event_mask & FSCACHE_OBJECT_EVENTS_MASK);
45         printk(KERN_ERR "%sops=%u inp=%u exc=%u\n",
46                prefix, object->fscache.n_ops, object->fscache.n_in_progress,
47                object->fscache.n_exclusive);
48         printk(KERN_ERR "%sparent=%p\n",
49                prefix, object->fscache.parent);
50
51         spin_lock(&object->fscache.lock);
52         cookie = object->fscache.cookie;
53         if (cookie) {
54                 printk(KERN_ERR "%scookie=%p [pr=%p nd=%p fl=%lx]\n",
55                        prefix,
56                        object->fscache.cookie,
57                        object->fscache.cookie->parent,
58                        object->fscache.cookie->netfs_data,
59                        object->fscache.cookie->flags);
60                 if (keybuf)
61                         keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
62                                                       CACHEFILES_KEYBUF_SIZE);
63                 else
64                         keylen = 0;
65         } else {
66                 printk(KERN_ERR "%scookie=NULL\n", prefix);
67                 keylen = 0;
68         }
69         spin_unlock(&object->fscache.lock);
70
71         if (keylen) {
72                 printk(KERN_ERR "%skey=[%u] '", prefix, keylen);
73                 for (loop = 0; loop < keylen; loop++)
74                         printk("%02x", keybuf[loop]);
75                 printk("'\n");
76         }
77 }
78
79 /*
80  * dump debugging info about a pair of objects
81  */
82 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
83                                               struct cachefiles_object *xobject)
84 {
85         u8 *keybuf;
86
87         keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
88         if (object)
89                 __cachefiles_printk_object(object, "", keybuf);
90         if (xobject)
91                 __cachefiles_printk_object(xobject, "x", keybuf);
92         kfree(keybuf);
93 }
94
95 /*
96  * mark the owner of a dentry, if there is one, to indicate that that dentry
97  * has been preemptively deleted
98  * - the caller must hold the i_mutex on the dentry's parent as required to
99  *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
100  */
101 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
102                                           struct dentry *dentry)
103 {
104         struct cachefiles_object *object;
105         struct rb_node *p;
106
107         _enter(",'%*.*s'",
108                dentry->d_name.len, dentry->d_name.len, dentry->d_name.name);
109
110         write_lock(&cache->active_lock);
111
112         p = cache->active_nodes.rb_node;
113         while (p) {
114                 object = rb_entry(p, struct cachefiles_object, active_node);
115                 if (object->dentry > dentry)
116                         p = p->rb_left;
117                 else if (object->dentry < dentry)
118                         p = p->rb_right;
119                 else
120                         goto found_dentry;
121         }
122
123         write_unlock(&cache->active_lock);
124         _leave(" [no owner]");
125         return;
126
127         /* found the dentry for  */
128 found_dentry:
129         kdebug("preemptive burial: OBJ%x [%s] %p",
130                object->fscache.debug_id,
131                fscache_object_states[object->fscache.state],
132                dentry);
133
134         if (object->fscache.state < FSCACHE_OBJECT_DYING) {
135                 printk(KERN_ERR "\n");
136                 printk(KERN_ERR "CacheFiles: Error:"
137                        " Can't preemptively bury live object\n");
138                 cachefiles_printk_object(object, NULL);
139         } else if (test_and_set_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
140                 printk(KERN_ERR "CacheFiles: Error:"
141                        " Object already preemptively buried\n");
142         }
143
144         write_unlock(&cache->active_lock);
145         _leave(" [owner marked]");
146 }
147
148 /*
149  * record the fact that an object is now active
150  */
151 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
152                                          struct cachefiles_object *object)
153 {
154         struct cachefiles_object *xobject;
155         struct rb_node **_p, *_parent = NULL;
156         struct dentry *dentry;
157
158         _enter(",%p", object);
159
160 try_again:
161         write_lock(&cache->active_lock);
162
163         if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
164                 printk(KERN_ERR "CacheFiles: Error: Object already active\n");
165                 cachefiles_printk_object(object, NULL);
166                 BUG();
167         }
168
169         dentry = object->dentry;
170         _p = &cache->active_nodes.rb_node;
171         while (*_p) {
172                 _parent = *_p;
173                 xobject = rb_entry(_parent,
174                                    struct cachefiles_object, active_node);
175
176                 ASSERT(xobject != object);
177
178                 if (xobject->dentry > dentry)
179                         _p = &(*_p)->rb_left;
180                 else if (xobject->dentry < dentry)
181                         _p = &(*_p)->rb_right;
182                 else
183                         goto wait_for_old_object;
184         }
185
186         rb_link_node(&object->active_node, _parent, _p);
187         rb_insert_color(&object->active_node, &cache->active_nodes);
188
189         write_unlock(&cache->active_lock);
190         _leave(" = 0");
191         return 0;
192
193         /* an old object from a previous incarnation is hogging the slot - we
194          * need to wait for it to be destroyed */
195 wait_for_old_object:
196         if (xobject->fscache.state < FSCACHE_OBJECT_DYING) {
197                 printk(KERN_ERR "\n");
198                 printk(KERN_ERR "CacheFiles: Error:"
199                        " Unexpected object collision\n");
200                 cachefiles_printk_object(object, xobject);
201                 BUG();
202         }
203         atomic_inc(&xobject->usage);
204         write_unlock(&cache->active_lock);
205
206         if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
207                 wait_queue_head_t *wq;
208
209                 signed long timeout = 60 * HZ;
210                 wait_queue_t wait;
211                 bool requeue;
212
213                 /* if the object we're waiting for is queued for processing,
214                  * then just put ourselves on the queue behind it */
215                 if (work_pending(&xobject->fscache.work)) {
216                         _debug("queue OBJ%x behind OBJ%x immediately",
217                                object->fscache.debug_id,
218                                xobject->fscache.debug_id);
219                         goto requeue;
220                 }
221
222                 /* otherwise we sleep until either the object we're waiting for
223                  * is done, or the fscache_object is congested */
224                 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
225                 init_wait(&wait);
226                 requeue = false;
227                 do {
228                         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
229                         if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
230                                 break;
231
232                         requeue = fscache_object_sleep_till_congested(&timeout);
233                 } while (timeout > 0 && !requeue);
234                 finish_wait(wq, &wait);
235
236                 if (requeue &&
237                     test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
238                         _debug("queue OBJ%x behind OBJ%x after wait",
239                                object->fscache.debug_id,
240                                xobject->fscache.debug_id);
241                         goto requeue;
242                 }
243
244                 if (timeout <= 0) {
245                         printk(KERN_ERR "\n");
246                         printk(KERN_ERR "CacheFiles: Error: Overlong"
247                                " wait for old active object to go away\n");
248                         cachefiles_printk_object(object, xobject);
249                         goto requeue;
250                 }
251         }
252
253         ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
254
255         cache->cache.ops->put_object(&xobject->fscache);
256         goto try_again;
257
258 requeue:
259         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
260         cache->cache.ops->put_object(&xobject->fscache);
261         _leave(" = -ETIMEDOUT");
262         return -ETIMEDOUT;
263 }
264
265 /*
266  * delete an object representation from the cache
267  * - file backed objects are unlinked
268  * - directory backed objects are stuffed into the graveyard for userspace to
269  *   delete
270  * - unlocks the directory mutex
271  */
272 static int cachefiles_bury_object(struct cachefiles_cache *cache,
273                                   struct dentry *dir,
274                                   struct dentry *rep,
275                                   bool preemptive)
276 {
277         struct dentry *grave, *trap;
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                 ret = vfs_unlink(dir->d_inode, rep);
291
292                 if (preemptive)
293                         cachefiles_mark_object_buried(cache, rep);
294
295                 mutex_unlock(&dir->d_inode->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(&dir->d_inode->i_mutex);
307
308 try_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 (!S_ISDIR(cache->graveyard->d_inode->i_mode)) {
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 (grave->d_inode) {
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         ret = vfs_rename(dir->d_inode, rep, cache->graveyard->d_inode, grave);
383         if (ret != 0 && ret != -ENOMEM)
384                 cachefiles_io_error(cache, "Rename failed with error %d", ret);
385
386         if (preemptive)
387                 cachefiles_mark_object_buried(cache, rep);
388
389         unlock_rename(cache->graveyard, dir);
390         dput(grave);
391         _leave(" = 0");
392         return 0;
393 }
394
395 /*
396  * delete an object representation from the cache
397  */
398 int cachefiles_delete_object(struct cachefiles_cache *cache,
399                              struct cachefiles_object *object)
400 {
401         struct dentry *dir;
402         int ret;
403
404         _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
405
406         ASSERT(object->dentry);
407         ASSERT(object->dentry->d_inode);
408         ASSERT(object->dentry->d_parent);
409
410         dir = dget_parent(object->dentry);
411
412         mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
413
414         if (test_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
415                 /* object allocation for the same key preemptively deleted this
416                  * object's file so that it could create its own file */
417                 _debug("object preemptively buried");
418                 mutex_unlock(&dir->d_inode->i_mutex);
419                 ret = 0;
420         } else {
421                 /* we need to check that our parent is _still_ our parent - it
422                  * may have been renamed */
423                 if (dir == object->dentry->d_parent) {
424                         ret = cachefiles_bury_object(cache, dir,
425                                                      object->dentry, false);
426                 } else {
427                         /* it got moved, presumably by cachefilesd culling it,
428                          * so it's no longer in the key path and we can ignore
429                          * it */
430                         mutex_unlock(&dir->d_inode->i_mutex);
431                         ret = 0;
432                 }
433         }
434
435         dput(dir);
436         _leave(" = %d", ret);
437         return ret;
438 }
439
440 /*
441  * walk from the parent object to the child object through the backing
442  * filesystem, creating directories as we go
443  */
444 int cachefiles_walk_to_object(struct cachefiles_object *parent,
445                               struct cachefiles_object *object,
446                               const char *key,
447                               struct cachefiles_xattr *auxdata)
448 {
449         struct cachefiles_cache *cache;
450         struct dentry *dir, *next = NULL;
451         unsigned long start;
452         const char *name;
453         int ret, nlen;
454
455         _enter("OBJ%x{%p},OBJ%x,%s,",
456                parent->fscache.debug_id, parent->dentry,
457                object->fscache.debug_id, key);
458
459         cache = container_of(parent->fscache.cache,
460                              struct cachefiles_cache, cache);
461
462         ASSERT(parent->dentry);
463         ASSERT(parent->dentry->d_inode);
464
465         if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
466                 // TODO: convert file to dir
467                 _leave("looking up in none directory");
468                 return -ENOBUFS;
469         }
470
471         dir = dget(parent->dentry);
472
473 advance:
474         /* attempt to transit the first directory component */
475         name = key;
476         nlen = strlen(key);
477
478         /* key ends in a double NUL */
479         key = key + nlen + 1;
480         if (!*key)
481                 key = NULL;
482
483 lookup_again:
484         /* search the current directory for the element name */
485         _debug("lookup '%s'", name);
486
487         mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
488
489         start = jiffies;
490         next = lookup_one_len(name, dir, nlen);
491         cachefiles_hist(cachefiles_lookup_histogram, start);
492         if (IS_ERR(next))
493                 goto lookup_error;
494
495         _debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
496
497         if (!key)
498                 object->new = !next->d_inode;
499
500         /* if this element of the path doesn't exist, then the lookup phase
501          * failed, and we can release any readers in the certain knowledge that
502          * there's nothing for them to actually read */
503         if (!next->d_inode)
504                 fscache_object_lookup_negative(&object->fscache);
505
506         /* we need to create the object if it's negative */
507         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
508                 /* index objects and intervening tree levels must be subdirs */
509                 if (!next->d_inode) {
510                         ret = cachefiles_has_space(cache, 1, 0);
511                         if (ret < 0)
512                                 goto create_error;
513
514                         start = jiffies;
515                         ret = vfs_mkdir(dir->d_inode, next, 0);
516                         cachefiles_hist(cachefiles_mkdir_histogram, start);
517                         if (ret < 0)
518                                 goto create_error;
519
520                         ASSERT(next->d_inode);
521
522                         _debug("mkdir -> %p{%p{ino=%lu}}",
523                                next, next->d_inode, next->d_inode->i_ino);
524
525                 } else if (!S_ISDIR(next->d_inode->i_mode)) {
526                         kerror("inode %lu is not a directory",
527                                next->d_inode->i_ino);
528                         ret = -ENOBUFS;
529                         goto error;
530                 }
531
532         } else {
533                 /* non-index objects start out life as files */
534                 if (!next->d_inode) {
535                         ret = cachefiles_has_space(cache, 1, 0);
536                         if (ret < 0)
537                                 goto create_error;
538
539                         start = jiffies;
540                         ret = vfs_create(dir->d_inode, next, S_IFREG, NULL);
541                         cachefiles_hist(cachefiles_create_histogram, start);
542                         if (ret < 0)
543                                 goto create_error;
544
545                         ASSERT(next->d_inode);
546
547                         _debug("create -> %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                            !S_ISREG(next->d_inode->i_mode)
552                            ) {
553                         kerror("inode %lu is not a file or directory",
554                                next->d_inode->i_ino);
555                         ret = -ENOBUFS;
556                         goto error;
557                 }
558         }
559
560         /* process the next component */
561         if (key) {
562                 _debug("advance");
563                 mutex_unlock(&dir->d_inode->i_mutex);
564                 dput(dir);
565                 dir = next;
566                 next = NULL;
567                 goto advance;
568         }
569
570         /* we've found the object we were looking for */
571         object->dentry = next;
572
573         /* if we've found that the terminal object exists, then we need to
574          * check its attributes and delete it if it's out of date */
575         if (!object->new) {
576                 _debug("validate '%*.*s'",
577                        next->d_name.len, next->d_name.len, next->d_name.name);
578
579                 ret = cachefiles_check_object_xattr(object, auxdata);
580                 if (ret == -ESTALE) {
581                         /* delete the object (the deleter drops the directory
582                          * mutex) */
583                         object->dentry = NULL;
584
585                         ret = cachefiles_bury_object(cache, dir, next, true);
586                         dput(next);
587                         next = NULL;
588
589                         if (ret < 0)
590                                 goto delete_error;
591
592                         _debug("redo lookup");
593                         goto lookup_again;
594                 }
595         }
596
597         /* note that we're now using this object */
598         ret = cachefiles_mark_object_active(cache, object);
599
600         mutex_unlock(&dir->d_inode->i_mutex);
601         dput(dir);
602         dir = NULL;
603
604         if (ret == -ETIMEDOUT)
605                 goto mark_active_timed_out;
606
607         _debug("=== OBTAINED_OBJECT ===");
608
609         if (object->new) {
610                 /* attach data to a newly constructed terminal object */
611                 ret = cachefiles_set_object_xattr(object, auxdata);
612                 if (ret < 0)
613                         goto check_error;
614         } else {
615                 /* always update the atime on an object we've just looked up
616                  * (this is used to keep track of culling, and atimes are only
617                  * updated by read, write and readdir but not lookup or
618                  * open) */
619                 touch_atime(cache->mnt, next);
620         }
621
622         /* open a file interface onto a data file */
623         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
624                 if (S_ISREG(object->dentry->d_inode->i_mode)) {
625                         const struct address_space_operations *aops;
626
627                         ret = -EPERM;
628                         aops = object->dentry->d_inode->i_mapping->a_ops;
629                         if (!aops->bmap)
630                                 goto check_error;
631
632                         object->backer = object->dentry;
633                 } else {
634                         BUG(); // TODO: open file in data-class subdir
635                 }
636         }
637
638         object->new = 0;
639         fscache_obtained_object(&object->fscache);
640
641         _leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
642         return 0;
643
644 create_error:
645         _debug("create error %d", ret);
646         if (ret == -EIO)
647                 cachefiles_io_error(cache, "Create/mkdir failed");
648         goto error;
649
650 mark_active_timed_out:
651         _debug("mark active timed out");
652         goto release_dentry;
653
654 check_error:
655         _debug("check error %d", ret);
656         write_lock(&cache->active_lock);
657         rb_erase(&object->active_node, &cache->active_nodes);
658         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
659         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
660         write_unlock(&cache->active_lock);
661 release_dentry:
662         dput(object->dentry);
663         object->dentry = NULL;
664         goto error_out;
665
666 delete_error:
667         _debug("delete error %d", ret);
668         goto error_out2;
669
670 lookup_error:
671         _debug("lookup error %ld", PTR_ERR(next));
672         ret = PTR_ERR(next);
673         if (ret == -EIO)
674                 cachefiles_io_error(cache, "Lookup failed");
675         next = NULL;
676 error:
677         mutex_unlock(&dir->d_inode->i_mutex);
678         dput(next);
679 error_out2:
680         dput(dir);
681 error_out:
682         _leave(" = error %d", -ret);
683         return ret;
684 }
685
686 /*
687  * get a subdirectory
688  */
689 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
690                                         struct dentry *dir,
691                                         const char *dirname)
692 {
693         struct dentry *subdir;
694         unsigned long start;
695         int ret;
696
697         _enter(",,%s", dirname);
698
699         /* search the current directory for the element name */
700         mutex_lock(&dir->d_inode->i_mutex);
701
702         start = jiffies;
703         subdir = lookup_one_len(dirname, dir, strlen(dirname));
704         cachefiles_hist(cachefiles_lookup_histogram, start);
705         if (IS_ERR(subdir)) {
706                 if (PTR_ERR(subdir) == -ENOMEM)
707                         goto nomem_d_alloc;
708                 goto lookup_error;
709         }
710
711         _debug("subdir -> %p %s",
712                subdir, subdir->d_inode ? "positive" : "negative");
713
714         /* we need to create the subdir if it doesn't exist yet */
715         if (!subdir->d_inode) {
716                 ret = cachefiles_has_space(cache, 1, 0);
717                 if (ret < 0)
718                         goto mkdir_error;
719
720                 _debug("attempt mkdir");
721
722                 ret = vfs_mkdir(dir->d_inode, subdir, 0700);
723                 if (ret < 0)
724                         goto mkdir_error;
725
726                 ASSERT(subdir->d_inode);
727
728                 _debug("mkdir -> %p{%p{ino=%lu}}",
729                        subdir,
730                        subdir->d_inode,
731                        subdir->d_inode->i_ino);
732         }
733
734         mutex_unlock(&dir->d_inode->i_mutex);
735
736         /* we need to make sure the subdir is a directory */
737         ASSERT(subdir->d_inode);
738
739         if (!S_ISDIR(subdir->d_inode->i_mode)) {
740                 kerror("%s is not a directory", dirname);
741                 ret = -EIO;
742                 goto check_error;
743         }
744
745         ret = -EPERM;
746         if (!subdir->d_inode->i_op ||
747             !subdir->d_inode->i_op->setxattr ||
748             !subdir->d_inode->i_op->getxattr ||
749             !subdir->d_inode->i_op->lookup ||
750             !subdir->d_inode->i_op->mkdir ||
751             !subdir->d_inode->i_op->create ||
752             !subdir->d_inode->i_op->rename ||
753             !subdir->d_inode->i_op->rmdir ||
754             !subdir->d_inode->i_op->unlink)
755                 goto check_error;
756
757         _leave(" = [%lu]", subdir->d_inode->i_ino);
758         return subdir;
759
760 check_error:
761         dput(subdir);
762         _leave(" = %d [check]", ret);
763         return ERR_PTR(ret);
764
765 mkdir_error:
766         mutex_unlock(&dir->d_inode->i_mutex);
767         dput(subdir);
768         kerror("mkdir %s failed with error %d", dirname, ret);
769         return ERR_PTR(ret);
770
771 lookup_error:
772         mutex_unlock(&dir->d_inode->i_mutex);
773         ret = PTR_ERR(subdir);
774         kerror("Lookup %s failed with error %d", dirname, ret);
775         return ERR_PTR(ret);
776
777 nomem_d_alloc:
778         mutex_unlock(&dir->d_inode->i_mutex);
779         _leave(" = -ENOMEM");
780         return ERR_PTR(-ENOMEM);
781 }
782
783 /*
784  * find out if an object is in use or not
785  * - if finds object and it's not in use:
786  *   - returns a pointer to the object and a reference on it
787  *   - returns with the directory locked
788  */
789 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
790                                               struct dentry *dir,
791                                               char *filename)
792 {
793         struct cachefiles_object *object;
794         struct rb_node *_n;
795         struct dentry *victim;
796         unsigned long start;
797         int ret;
798
799         //_enter(",%*.*s/,%s",
800         //       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
801
802         /* look up the victim */
803         mutex_lock_nested(&dir->d_inode->i_mutex, 1);
804
805         start = jiffies;
806         victim = lookup_one_len(filename, dir, strlen(filename));
807         cachefiles_hist(cachefiles_lookup_histogram, start);
808         if (IS_ERR(victim))
809                 goto lookup_error;
810
811         //_debug("victim -> %p %s",
812         //       victim, victim->d_inode ? "positive" : "negative");
813
814         /* if the object is no longer there then we probably retired the object
815          * at the netfs's request whilst the cull was in progress
816          */
817         if (!victim->d_inode) {
818                 mutex_unlock(&dir->d_inode->i_mutex);
819                 dput(victim);
820                 _leave(" = -ENOENT [absent]");
821                 return ERR_PTR(-ENOENT);
822         }
823
824         /* check to see if we're using this object */
825         read_lock(&cache->active_lock);
826
827         _n = cache->active_nodes.rb_node;
828
829         while (_n) {
830                 object = rb_entry(_n, struct cachefiles_object, active_node);
831
832                 if (object->dentry > victim)
833                         _n = _n->rb_left;
834                 else if (object->dentry < victim)
835                         _n = _n->rb_right;
836                 else
837                         goto object_in_use;
838         }
839
840         read_unlock(&cache->active_lock);
841
842         //_leave(" = %p", victim);
843         return victim;
844
845 object_in_use:
846         read_unlock(&cache->active_lock);
847         mutex_unlock(&dir->d_inode->i_mutex);
848         dput(victim);
849         //_leave(" = -EBUSY [in use]");
850         return ERR_PTR(-EBUSY);
851
852 lookup_error:
853         mutex_unlock(&dir->d_inode->i_mutex);
854         ret = PTR_ERR(victim);
855         if (ret == -ENOENT) {
856                 /* file or dir now absent - probably retired by netfs */
857                 _leave(" = -ESTALE [absent]");
858                 return ERR_PTR(-ESTALE);
859         }
860
861         if (ret == -EIO) {
862                 cachefiles_io_error(cache, "Lookup failed");
863         } else if (ret != -ENOMEM) {
864                 kerror("Internal error: %d", ret);
865                 ret = -EIO;
866         }
867
868         _leave(" = %d", ret);
869         return ERR_PTR(ret);
870 }
871
872 /*
873  * cull an object if it's not in use
874  * - called only by cache manager daemon
875  */
876 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
877                     char *filename)
878 {
879         struct dentry *victim;
880         int ret;
881
882         _enter(",%*.*s/,%s",
883                dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
884
885         victim = cachefiles_check_active(cache, dir, filename);
886         if (IS_ERR(victim))
887                 return PTR_ERR(victim);
888
889         _debug("victim -> %p %s",
890                victim, victim->d_inode ? "positive" : "negative");
891
892         /* okay... the victim is not being used so we can cull it
893          * - start by marking it as stale
894          */
895         _debug("victim is cullable");
896
897         ret = cachefiles_remove_object_xattr(cache, victim);
898         if (ret < 0)
899                 goto error_unlock;
900
901         /*  actually remove the victim (drops the dir mutex) */
902         _debug("bury");
903
904         ret = cachefiles_bury_object(cache, dir, victim, false);
905         if (ret < 0)
906                 goto error;
907
908         dput(victim);
909         _leave(" = 0");
910         return 0;
911
912 error_unlock:
913         mutex_unlock(&dir->d_inode->i_mutex);
914 error:
915         dput(victim);
916         if (ret == -ENOENT) {
917                 /* file or dir now absent - probably retired by netfs */
918                 _leave(" = -ESTALE [absent]");
919                 return -ESTALE;
920         }
921
922         if (ret != -ENOMEM) {
923                 kerror("Internal error: %d", ret);
924                 ret = -EIO;
925         }
926
927         _leave(" = %d", ret);
928         return ret;
929 }
930
931 /*
932  * find out if an object is in use or not
933  * - called only by cache manager daemon
934  * - returns -EBUSY or 0 to indicate whether an object is in use or not
935  */
936 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
937                             char *filename)
938 {
939         struct dentry *victim;
940
941         //_enter(",%*.*s/,%s",
942         //       dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
943
944         victim = cachefiles_check_active(cache, dir, filename);
945         if (IS_ERR(victim))
946                 return PTR_ERR(victim);
947
948         mutex_unlock(&dir->d_inode->i_mutex);
949         dput(victim);
950         //_leave(" = 0");
951         return 0;
952 }