These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / 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         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  */
80 static 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  */
99 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
100                                           struct dentry *dentry,
101                                           enum fscache_why_object_killed why)
102 {
103         struct cachefiles_object *object;
104         struct rb_node *p;
105
106         _enter(",'%pd'", dentry);
107
108         write_lock(&cache->active_lock);
109
110         p = cache->active_nodes.rb_node;
111         while (p) {
112                 object = rb_entry(p, struct cachefiles_object, active_node);
113                 if (object->dentry > dentry)
114                         p = p->rb_left;
115                 else if (object->dentry < dentry)
116                         p = p->rb_right;
117                 else
118                         goto found_dentry;
119         }
120
121         write_unlock(&cache->active_lock);
122         _leave(" [no owner]");
123         return;
124
125         /* found the dentry for  */
126 found_dentry:
127         kdebug("preemptive burial: OBJ%x [%s] %p",
128                object->fscache.debug_id,
129                object->fscache.state->name,
130                dentry);
131
132         if (fscache_object_is_live(&object->fscache)) {
133                 pr_err("\n");
134                 pr_err("Error: Can't preemptively bury live object\n");
135                 cachefiles_printk_object(object, NULL);
136         } else {
137                 if (why != FSCACHE_OBJECT_IS_STALE)
138                         fscache_object_mark_killed(&object->fscache, why);
139         }
140
141         write_unlock(&cache->active_lock);
142         _leave(" [owner marked]");
143 }
144
145 /*
146  * record the fact that an object is now active
147  */
148 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
149                                          struct cachefiles_object *object)
150 {
151         struct cachefiles_object *xobject;
152         struct rb_node **_p, *_parent = NULL;
153         struct dentry *dentry;
154
155         _enter(",%p", object);
156
157 try_again:
158         write_lock(&cache->active_lock);
159
160         if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
161                 pr_err("Error: Object already active\n");
162                 cachefiles_printk_object(object, NULL);
163                 BUG();
164         }
165
166         dentry = object->dentry;
167         _p = &cache->active_nodes.rb_node;
168         while (*_p) {
169                 _parent = *_p;
170                 xobject = rb_entry(_parent,
171                                    struct cachefiles_object, active_node);
172
173                 ASSERT(xobject != object);
174
175                 if (xobject->dentry > dentry)
176                         _p = &(*_p)->rb_left;
177                 else if (xobject->dentry < dentry)
178                         _p = &(*_p)->rb_right;
179                 else
180                         goto wait_for_old_object;
181         }
182
183         rb_link_node(&object->active_node, _parent, _p);
184         rb_insert_color(&object->active_node, &cache->active_nodes);
185
186         write_unlock(&cache->active_lock);
187         _leave(" = 0");
188         return 0;
189
190         /* an old object from a previous incarnation is hogging the slot - we
191          * need to wait for it to be destroyed */
192 wait_for_old_object:
193         if (fscache_object_is_live(&xobject->fscache)) {
194                 pr_err("\n");
195                 pr_err("Error: Unexpected object collision\n");
196                 cachefiles_printk_object(object, xobject);
197                 BUG();
198         }
199         atomic_inc(&xobject->usage);
200         write_unlock(&cache->active_lock);
201
202         if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
203                 wait_queue_head_t *wq;
204
205                 signed long timeout = 60 * HZ;
206                 wait_queue_t wait;
207                 bool requeue;
208
209                 /* if the object we're waiting for is queued for processing,
210                  * then just put ourselves on the queue behind it */
211                 if (work_pending(&xobject->fscache.work)) {
212                         _debug("queue OBJ%x behind OBJ%x immediately",
213                                object->fscache.debug_id,
214                                xobject->fscache.debug_id);
215                         goto requeue;
216                 }
217
218                 /* otherwise we sleep until either the object we're waiting for
219                  * is done, or the fscache_object is congested */
220                 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
221                 init_wait(&wait);
222                 requeue = false;
223                 do {
224                         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
225                         if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
226                                 break;
227
228                         requeue = fscache_object_sleep_till_congested(&timeout);
229                 } while (timeout > 0 && !requeue);
230                 finish_wait(wq, &wait);
231
232                 if (requeue &&
233                     test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
234                         _debug("queue OBJ%x behind OBJ%x after wait",
235                                object->fscache.debug_id,
236                                xobject->fscache.debug_id);
237                         goto requeue;
238                 }
239
240                 if (timeout <= 0) {
241                         pr_err("\n");
242                         pr_err("Error: Overlong wait for old active object to go away\n");
243                         cachefiles_printk_object(object, xobject);
244                         goto requeue;
245                 }
246         }
247
248         ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
249
250         cache->cache.ops->put_object(&xobject->fscache);
251         goto try_again;
252
253 requeue:
254         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
255         cache->cache.ops->put_object(&xobject->fscache);
256         _leave(" = -ETIMEDOUT");
257         return -ETIMEDOUT;
258 }
259
260 /*
261  * delete an object representation from the cache
262  * - file backed objects are unlinked
263  * - directory backed objects are stuffed into the graveyard for userspace to
264  *   delete
265  * - unlocks the directory mutex
266  */
267 static int cachefiles_bury_object(struct cachefiles_cache *cache,
268                                   struct dentry *dir,
269                                   struct dentry *rep,
270                                   bool preemptive,
271                                   enum fscache_why_object_killed why)
272 {
273         struct dentry *grave, *trap;
274         struct path path, path_to_graveyard;
275         char nbuffer[8 + 8 + 1];
276         int ret;
277
278         _enter(",'%pd','%pd'", dir, rep);
279
280         _debug("remove %p from %p", rep, dir);
281
282         /* non-directories can just be unlinked */
283         if (!d_is_dir(rep)) {
284                 _debug("unlink stale object");
285
286                 path.mnt = cache->mnt;
287                 path.dentry = dir;
288                 ret = security_path_unlink(&path, rep);
289                 if (ret < 0) {
290                         cachefiles_io_error(cache, "Unlink security error");
291                 } else {
292                         ret = vfs_unlink(d_inode(dir), rep, NULL);
293
294                         if (preemptive)
295                                 cachefiles_mark_object_buried(cache, rep, why);
296                 }
297
298                 mutex_unlock(&d_inode(dir)->i_mutex);
299
300                 if (ret == -EIO)
301                         cachefiles_io_error(cache, "Unlink failed");
302
303                 _leave(" = %d", ret);
304                 return ret;
305         }
306
307         /* directories have to be moved to the graveyard */
308         _debug("move stale object to graveyard");
309         mutex_unlock(&d_inode(dir)->i_mutex);
310
311 try_again:
312         /* first step is to make up a grave dentry in the graveyard */
313         sprintf(nbuffer, "%08x%08x",
314                 (uint32_t) get_seconds(),
315                 (uint32_t) atomic_inc_return(&cache->gravecounter));
316
317         /* do the multiway lock magic */
318         trap = lock_rename(cache->graveyard, dir);
319
320         /* do some checks before getting the grave dentry */
321         if (rep->d_parent != dir) {
322                 /* the entry was probably culled when we dropped the parent dir
323                  * lock */
324                 unlock_rename(cache->graveyard, dir);
325                 _leave(" = 0 [culled?]");
326                 return 0;
327         }
328
329         if (!d_can_lookup(cache->graveyard)) {
330                 unlock_rename(cache->graveyard, dir);
331                 cachefiles_io_error(cache, "Graveyard no longer a directory");
332                 return -EIO;
333         }
334
335         if (trap == rep) {
336                 unlock_rename(cache->graveyard, dir);
337                 cachefiles_io_error(cache, "May not make directory loop");
338                 return -EIO;
339         }
340
341         if (d_mountpoint(rep)) {
342                 unlock_rename(cache->graveyard, dir);
343                 cachefiles_io_error(cache, "Mountpoint in cache");
344                 return -EIO;
345         }
346
347         grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
348         if (IS_ERR(grave)) {
349                 unlock_rename(cache->graveyard, dir);
350
351                 if (PTR_ERR(grave) == -ENOMEM) {
352                         _leave(" = -ENOMEM");
353                         return -ENOMEM;
354                 }
355
356                 cachefiles_io_error(cache, "Lookup error %ld",
357                                     PTR_ERR(grave));
358                 return -EIO;
359         }
360
361         if (d_is_positive(grave)) {
362                 unlock_rename(cache->graveyard, dir);
363                 dput(grave);
364                 grave = NULL;
365                 cond_resched();
366                 goto try_again;
367         }
368
369         if (d_mountpoint(grave)) {
370                 unlock_rename(cache->graveyard, dir);
371                 dput(grave);
372                 cachefiles_io_error(cache, "Mountpoint in graveyard");
373                 return -EIO;
374         }
375
376         /* target should not be an ancestor of source */
377         if (trap == grave) {
378                 unlock_rename(cache->graveyard, dir);
379                 dput(grave);
380                 cachefiles_io_error(cache, "May not make directory loop");
381                 return -EIO;
382         }
383
384         /* attempt the rename */
385         path.mnt = cache->mnt;
386         path.dentry = dir;
387         path_to_graveyard.mnt = cache->mnt;
388         path_to_graveyard.dentry = cache->graveyard;
389         ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
390         if (ret < 0) {
391                 cachefiles_io_error(cache, "Rename security error %d", ret);
392         } else {
393                 ret = vfs_rename(d_inode(dir), rep,
394                                  d_inode(cache->graveyard), grave, NULL, 0);
395                 if (ret != 0 && ret != -ENOMEM)
396                         cachefiles_io_error(cache,
397                                             "Rename failed with error %d", ret);
398
399                 if (preemptive)
400                         cachefiles_mark_object_buried(cache, rep, why);
401         }
402
403         unlock_rename(cache->graveyard, dir);
404         dput(grave);
405         _leave(" = 0");
406         return 0;
407 }
408
409 /*
410  * delete an object representation from the cache
411  */
412 int cachefiles_delete_object(struct cachefiles_cache *cache,
413                              struct cachefiles_object *object)
414 {
415         struct dentry *dir;
416         int ret;
417
418         _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
419
420         ASSERT(object->dentry);
421         ASSERT(d_backing_inode(object->dentry));
422         ASSERT(object->dentry->d_parent);
423
424         dir = dget_parent(object->dentry);
425
426         mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
427
428         if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
429                 /* object allocation for the same key preemptively deleted this
430                  * object's file so that it could create its own file */
431                 _debug("object preemptively buried");
432                 mutex_unlock(&d_inode(dir)->i_mutex);
433                 ret = 0;
434         } else {
435                 /* we need to check that our parent is _still_ our parent - it
436                  * may have been renamed */
437                 if (dir == object->dentry->d_parent) {
438                         ret = cachefiles_bury_object(cache, dir,
439                                                      object->dentry, false,
440                                                      FSCACHE_OBJECT_WAS_RETIRED);
441                 } else {
442                         /* it got moved, presumably by cachefilesd culling it,
443                          * so it's no longer in the key path and we can ignore
444                          * it */
445                         mutex_unlock(&d_inode(dir)->i_mutex);
446                         ret = 0;
447                 }
448         }
449
450         dput(dir);
451         _leave(" = %d", ret);
452         return ret;
453 }
454
455 /*
456  * walk from the parent object to the child object through the backing
457  * filesystem, creating directories as we go
458  */
459 int cachefiles_walk_to_object(struct cachefiles_object *parent,
460                               struct cachefiles_object *object,
461                               const char *key,
462                               struct cachefiles_xattr *auxdata)
463 {
464         struct cachefiles_cache *cache;
465         struct dentry *dir, *next = NULL;
466         struct path path;
467         unsigned long start;
468         const char *name;
469         int ret, nlen;
470
471         _enter("OBJ%x{%p},OBJ%x,%s,",
472                parent->fscache.debug_id, parent->dentry,
473                object->fscache.debug_id, key);
474
475         cache = container_of(parent->fscache.cache,
476                              struct cachefiles_cache, cache);
477         path.mnt = cache->mnt;
478
479         ASSERT(parent->dentry);
480         ASSERT(d_backing_inode(parent->dentry));
481
482         if (!(d_is_dir(parent->dentry))) {
483                 // TODO: convert file to dir
484                 _leave("looking up in none directory");
485                 return -ENOBUFS;
486         }
487
488         dir = dget(parent->dentry);
489
490 advance:
491         /* attempt to transit the first directory component */
492         name = key;
493         nlen = strlen(key);
494
495         /* key ends in a double NUL */
496         key = key + nlen + 1;
497         if (!*key)
498                 key = NULL;
499
500 lookup_again:
501         /* search the current directory for the element name */
502         _debug("lookup '%s'", name);
503
504         mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
505
506         start = jiffies;
507         next = lookup_one_len(name, dir, nlen);
508         cachefiles_hist(cachefiles_lookup_histogram, start);
509         if (IS_ERR(next))
510                 goto lookup_error;
511
512         _debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
513
514         if (!key)
515                 object->new = !d_backing_inode(next);
516
517         /* if this element of the path doesn't exist, then the lookup phase
518          * failed, and we can release any readers in the certain knowledge that
519          * there's nothing for them to actually read */
520         if (d_is_negative(next))
521                 fscache_object_lookup_negative(&object->fscache);
522
523         /* we need to create the object if it's negative */
524         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
525                 /* index objects and intervening tree levels must be subdirs */
526                 if (d_is_negative(next)) {
527                         ret = cachefiles_has_space(cache, 1, 0);
528                         if (ret < 0)
529                                 goto no_space_error;
530
531                         path.dentry = dir;
532                         ret = security_path_mkdir(&path, next, 0);
533                         if (ret < 0)
534                                 goto create_error;
535                         start = jiffies;
536                         ret = vfs_mkdir(d_inode(dir), next, 0);
537                         cachefiles_hist(cachefiles_mkdir_histogram, start);
538                         if (ret < 0)
539                                 goto create_error;
540
541                         ASSERT(d_backing_inode(next));
542
543                         _debug("mkdir -> %p{%p{ino=%lu}}",
544                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
545
546                 } else if (!d_can_lookup(next)) {
547                         pr_err("inode %lu is not a directory\n",
548                                d_backing_inode(next)->i_ino);
549                         ret = -ENOBUFS;
550                         goto error;
551                 }
552
553         } else {
554                 /* non-index objects start out life as files */
555                 if (d_is_negative(next)) {
556                         ret = cachefiles_has_space(cache, 1, 0);
557                         if (ret < 0)
558                                 goto no_space_error;
559
560                         path.dentry = dir;
561                         ret = security_path_mknod(&path, next, S_IFREG, 0);
562                         if (ret < 0)
563                                 goto create_error;
564                         start = jiffies;
565                         ret = vfs_create(d_inode(dir), next, S_IFREG, true);
566                         cachefiles_hist(cachefiles_create_histogram, start);
567                         if (ret < 0)
568                                 goto create_error;
569
570                         ASSERT(d_backing_inode(next));
571
572                         _debug("create -> %p{%p{ino=%lu}}",
573                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
574
575                 } else if (!d_can_lookup(next) &&
576                            !d_is_reg(next)
577                            ) {
578                         pr_err("inode %lu is not a file or directory\n",
579                                d_backing_inode(next)->i_ino);
580                         ret = -ENOBUFS;
581                         goto error;
582                 }
583         }
584
585         /* process the next component */
586         if (key) {
587                 _debug("advance");
588                 mutex_unlock(&d_inode(dir)->i_mutex);
589                 dput(dir);
590                 dir = next;
591                 next = NULL;
592                 goto advance;
593         }
594
595         /* we've found the object we were looking for */
596         object->dentry = next;
597
598         /* if we've found that the terminal object exists, then we need to
599          * check its attributes and delete it if it's out of date */
600         if (!object->new) {
601                 _debug("validate '%pd'", next);
602
603                 ret = cachefiles_check_object_xattr(object, auxdata);
604                 if (ret == -ESTALE) {
605                         /* delete the object (the deleter drops the directory
606                          * mutex) */
607                         object->dentry = NULL;
608
609                         ret = cachefiles_bury_object(cache, dir, next, true,
610                                                      FSCACHE_OBJECT_IS_STALE);
611                         dput(next);
612                         next = NULL;
613
614                         if (ret < 0)
615                                 goto delete_error;
616
617                         _debug("redo lookup");
618                         fscache_object_retrying_stale(&object->fscache);
619                         goto lookup_again;
620                 }
621         }
622
623         /* note that we're now using this object */
624         ret = cachefiles_mark_object_active(cache, object);
625
626         mutex_unlock(&d_inode(dir)->i_mutex);
627         dput(dir);
628         dir = NULL;
629
630         if (ret == -ETIMEDOUT)
631                 goto mark_active_timed_out;
632
633         _debug("=== OBTAINED_OBJECT ===");
634
635         if (object->new) {
636                 /* attach data to a newly constructed terminal object */
637                 ret = cachefiles_set_object_xattr(object, auxdata);
638                 if (ret < 0)
639                         goto check_error;
640         } else {
641                 /* always update the atime on an object we've just looked up
642                  * (this is used to keep track of culling, and atimes are only
643                  * updated by read, write and readdir but not lookup or
644                  * open) */
645                 path.dentry = next;
646                 touch_atime(&path);
647         }
648
649         /* open a file interface onto a data file */
650         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
651                 if (d_is_reg(object->dentry)) {
652                         const struct address_space_operations *aops;
653
654                         ret = -EPERM;
655                         aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
656                         if (!aops->bmap)
657                                 goto check_error;
658                         if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
659                                 goto check_error;
660
661                         object->backer = object->dentry;
662                 } else {
663                         BUG(); // TODO: open file in data-class subdir
664                 }
665         }
666
667         object->new = 0;
668         fscache_obtained_object(&object->fscache);
669
670         _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
671         return 0;
672
673 no_space_error:
674         fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
675 create_error:
676         _debug("create error %d", ret);
677         if (ret == -EIO)
678                 cachefiles_io_error(cache, "Create/mkdir failed");
679         goto error;
680
681 mark_active_timed_out:
682         _debug("mark active timed out");
683         goto release_dentry;
684
685 check_error:
686         _debug("check error %d", ret);
687         write_lock(&cache->active_lock);
688         rb_erase(&object->active_node, &cache->active_nodes);
689         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
690         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
691         write_unlock(&cache->active_lock);
692 release_dentry:
693         dput(object->dentry);
694         object->dentry = NULL;
695         goto error_out;
696
697 delete_error:
698         _debug("delete error %d", ret);
699         goto error_out2;
700
701 lookup_error:
702         _debug("lookup error %ld", PTR_ERR(next));
703         ret = PTR_ERR(next);
704         if (ret == -EIO)
705                 cachefiles_io_error(cache, "Lookup failed");
706         next = NULL;
707 error:
708         mutex_unlock(&d_inode(dir)->i_mutex);
709         dput(next);
710 error_out2:
711         dput(dir);
712 error_out:
713         _leave(" = error %d", -ret);
714         return ret;
715 }
716
717 /*
718  * get a subdirectory
719  */
720 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
721                                         struct dentry *dir,
722                                         const char *dirname)
723 {
724         struct dentry *subdir;
725         unsigned long start;
726         struct path path;
727         int ret;
728
729         _enter(",,%s", dirname);
730
731         /* search the current directory for the element name */
732         mutex_lock(&d_inode(dir)->i_mutex);
733
734         start = jiffies;
735         subdir = lookup_one_len(dirname, dir, strlen(dirname));
736         cachefiles_hist(cachefiles_lookup_histogram, start);
737         if (IS_ERR(subdir)) {
738                 if (PTR_ERR(subdir) == -ENOMEM)
739                         goto nomem_d_alloc;
740                 goto lookup_error;
741         }
742
743         _debug("subdir -> %p %s",
744                subdir, d_backing_inode(subdir) ? "positive" : "negative");
745
746         /* we need to create the subdir if it doesn't exist yet */
747         if (d_is_negative(subdir)) {
748                 ret = cachefiles_has_space(cache, 1, 0);
749                 if (ret < 0)
750                         goto mkdir_error;
751
752                 _debug("attempt mkdir");
753
754                 path.mnt = cache->mnt;
755                 path.dentry = dir;
756                 ret = security_path_mkdir(&path, subdir, 0700);
757                 if (ret < 0)
758                         goto mkdir_error;
759                 ret = vfs_mkdir(d_inode(dir), subdir, 0700);
760                 if (ret < 0)
761                         goto mkdir_error;
762
763                 ASSERT(d_backing_inode(subdir));
764
765                 _debug("mkdir -> %p{%p{ino=%lu}}",
766                        subdir,
767                        d_backing_inode(subdir),
768                        d_backing_inode(subdir)->i_ino);
769         }
770
771         mutex_unlock(&d_inode(dir)->i_mutex);
772
773         /* we need to make sure the subdir is a directory */
774         ASSERT(d_backing_inode(subdir));
775
776         if (!d_can_lookup(subdir)) {
777                 pr_err("%s is not a directory\n", dirname);
778                 ret = -EIO;
779                 goto check_error;
780         }
781
782         ret = -EPERM;
783         if (!d_backing_inode(subdir)->i_op->setxattr ||
784             !d_backing_inode(subdir)->i_op->getxattr ||
785             !d_backing_inode(subdir)->i_op->lookup ||
786             !d_backing_inode(subdir)->i_op->mkdir ||
787             !d_backing_inode(subdir)->i_op->create ||
788             (!d_backing_inode(subdir)->i_op->rename &&
789              !d_backing_inode(subdir)->i_op->rename2) ||
790             !d_backing_inode(subdir)->i_op->rmdir ||
791             !d_backing_inode(subdir)->i_op->unlink)
792                 goto check_error;
793
794         _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
795         return subdir;
796
797 check_error:
798         dput(subdir);
799         _leave(" = %d [check]", ret);
800         return ERR_PTR(ret);
801
802 mkdir_error:
803         mutex_unlock(&d_inode(dir)->i_mutex);
804         dput(subdir);
805         pr_err("mkdir %s failed with error %d\n", dirname, ret);
806         return ERR_PTR(ret);
807
808 lookup_error:
809         mutex_unlock(&d_inode(dir)->i_mutex);
810         ret = PTR_ERR(subdir);
811         pr_err("Lookup %s failed with error %d\n", dirname, ret);
812         return ERR_PTR(ret);
813
814 nomem_d_alloc:
815         mutex_unlock(&d_inode(dir)->i_mutex);
816         _leave(" = -ENOMEM");
817         return ERR_PTR(-ENOMEM);
818 }
819
820 /*
821  * find out if an object is in use or not
822  * - if finds object and it's not in use:
823  *   - returns a pointer to the object and a reference on it
824  *   - returns with the directory locked
825  */
826 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
827                                               struct dentry *dir,
828                                               char *filename)
829 {
830         struct cachefiles_object *object;
831         struct rb_node *_n;
832         struct dentry *victim;
833         unsigned long start;
834         int ret;
835
836         //_enter(",%pd/,%s",
837         //       dir, filename);
838
839         /* look up the victim */
840         mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
841
842         start = jiffies;
843         victim = lookup_one_len(filename, dir, strlen(filename));
844         cachefiles_hist(cachefiles_lookup_histogram, start);
845         if (IS_ERR(victim))
846                 goto lookup_error;
847
848         //_debug("victim -> %p %s",
849         //       victim, d_backing_inode(victim) ? "positive" : "negative");
850
851         /* if the object is no longer there then we probably retired the object
852          * at the netfs's request whilst the cull was in progress
853          */
854         if (d_is_negative(victim)) {
855                 mutex_unlock(&d_inode(dir)->i_mutex);
856                 dput(victim);
857                 _leave(" = -ENOENT [absent]");
858                 return ERR_PTR(-ENOENT);
859         }
860
861         /* check to see if we're using this object */
862         read_lock(&cache->active_lock);
863
864         _n = cache->active_nodes.rb_node;
865
866         while (_n) {
867                 object = rb_entry(_n, struct cachefiles_object, active_node);
868
869                 if (object->dentry > victim)
870                         _n = _n->rb_left;
871                 else if (object->dentry < victim)
872                         _n = _n->rb_right;
873                 else
874                         goto object_in_use;
875         }
876
877         read_unlock(&cache->active_lock);
878
879         //_leave(" = %p", victim);
880         return victim;
881
882 object_in_use:
883         read_unlock(&cache->active_lock);
884         mutex_unlock(&d_inode(dir)->i_mutex);
885         dput(victim);
886         //_leave(" = -EBUSY [in use]");
887         return ERR_PTR(-EBUSY);
888
889 lookup_error:
890         mutex_unlock(&d_inode(dir)->i_mutex);
891         ret = PTR_ERR(victim);
892         if (ret == -ENOENT) {
893                 /* file or dir now absent - probably retired by netfs */
894                 _leave(" = -ESTALE [absent]");
895                 return ERR_PTR(-ESTALE);
896         }
897
898         if (ret == -EIO) {
899                 cachefiles_io_error(cache, "Lookup failed");
900         } else if (ret != -ENOMEM) {
901                 pr_err("Internal error: %d\n", ret);
902                 ret = -EIO;
903         }
904
905         _leave(" = %d", ret);
906         return ERR_PTR(ret);
907 }
908
909 /*
910  * cull an object if it's not in use
911  * - called only by cache manager daemon
912  */
913 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
914                     char *filename)
915 {
916         struct dentry *victim;
917         int ret;
918
919         _enter(",%pd/,%s", dir, filename);
920
921         victim = cachefiles_check_active(cache, dir, filename);
922         if (IS_ERR(victim))
923                 return PTR_ERR(victim);
924
925         _debug("victim -> %p %s",
926                victim, d_backing_inode(victim) ? "positive" : "negative");
927
928         /* okay... the victim is not being used so we can cull it
929          * - start by marking it as stale
930          */
931         _debug("victim is cullable");
932
933         ret = cachefiles_remove_object_xattr(cache, victim);
934         if (ret < 0)
935                 goto error_unlock;
936
937         /*  actually remove the victim (drops the dir mutex) */
938         _debug("bury");
939
940         ret = cachefiles_bury_object(cache, dir, victim, false,
941                                      FSCACHE_OBJECT_WAS_CULLED);
942         if (ret < 0)
943                 goto error;
944
945         dput(victim);
946         _leave(" = 0");
947         return 0;
948
949 error_unlock:
950         mutex_unlock(&d_inode(dir)->i_mutex);
951 error:
952         dput(victim);
953         if (ret == -ENOENT) {
954                 /* file or dir now absent - probably retired by netfs */
955                 _leave(" = -ESTALE [absent]");
956                 return -ESTALE;
957         }
958
959         if (ret != -ENOMEM) {
960                 pr_err("Internal error: %d\n", ret);
961                 ret = -EIO;
962         }
963
964         _leave(" = %d", ret);
965         return ret;
966 }
967
968 /*
969  * find out if an object is in use or not
970  * - called only by cache manager daemon
971  * - returns -EBUSY or 0 to indicate whether an object is in use or not
972  */
973 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
974                             char *filename)
975 {
976         struct dentry *victim;
977
978         //_enter(",%pd/,%s",
979         //       dir, filename);
980
981         victim = cachefiles_check_active(cache, dir, filename);
982         if (IS_ERR(victim))
983                 return PTR_ERR(victim);
984
985         mutex_unlock(&d_inode(dir)->i_mutex);
986         dput(victim);
987         //_leave(" = 0");
988         return 0;
989 }