Merge tag 'libnvdimm-for-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdim...
[linux-2.6-microblaze.git] / fs / cachefiles / namei.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* CacheFiles path walking and related routines
3  *
4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7
8 #include <linux/module.h>
9 #include <linux/sched.h>
10 #include <linux/file.h>
11 #include <linux/fs.h>
12 #include <linux/fsnotify.h>
13 #include <linux/quotaops.h>
14 #include <linux/xattr.h>
15 #include <linux/mount.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/slab.h>
19 #include "internal.h"
20
21 #define CACHEFILES_KEYBUF_SIZE 512
22
23 /*
24  * dump debugging info about an object
25  */
26 static noinline
27 void __cachefiles_printk_object(struct cachefiles_object *object,
28                                 const char *prefix)
29 {
30         struct fscache_cookie *cookie;
31         const u8 *k;
32         unsigned loop;
33
34         pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
35         pr_err("%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
36                prefix, object->fscache.state->name,
37                object->fscache.flags, work_busy(&object->fscache.work),
38                object->fscache.events, object->fscache.event_mask);
39         pr_err("%sops=%u inp=%u exc=%u\n",
40                prefix, object->fscache.n_ops, object->fscache.n_in_progress,
41                object->fscache.n_exclusive);
42         pr_err("%sparent=%x\n",
43                prefix, object->fscache.parent ? object->fscache.parent->debug_id : 0);
44
45         spin_lock(&object->fscache.lock);
46         cookie = object->fscache.cookie;
47         if (cookie) {
48                 pr_err("%scookie=%x [pr=%x nd=%p fl=%lx]\n",
49                        prefix,
50                        cookie->debug_id,
51                        cookie->parent ? cookie->parent->debug_id : 0,
52                        cookie->netfs_data,
53                        cookie->flags);
54                 pr_err("%skey=[%u] '", prefix, cookie->key_len);
55                 k = (cookie->key_len <= sizeof(cookie->inline_key)) ?
56                         cookie->inline_key : cookie->key;
57                 for (loop = 0; loop < cookie->key_len; loop++)
58                         pr_cont("%02x", k[loop]);
59                 pr_cont("'\n");
60         } else {
61                 pr_err("%scookie=NULL\n", prefix);
62         }
63         spin_unlock(&object->fscache.lock);
64 }
65
66 /*
67  * dump debugging info about a pair of objects
68  */
69 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
70                                               struct cachefiles_object *xobject)
71 {
72         if (object)
73                 __cachefiles_printk_object(object, "");
74         if (xobject)
75                 __cachefiles_printk_object(xobject, "x");
76 }
77
78 /*
79  * mark the owner of a dentry, if there is one, to indicate that that dentry
80  * has been preemptively deleted
81  * - the caller must hold the i_mutex on the dentry's parent as required to
82  *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
83  */
84 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
85                                           struct dentry *dentry,
86                                           enum fscache_why_object_killed why)
87 {
88         struct cachefiles_object *object;
89         struct rb_node *p;
90
91         _enter(",'%pd'", dentry);
92
93         write_lock(&cache->active_lock);
94
95         p = cache->active_nodes.rb_node;
96         while (p) {
97                 object = rb_entry(p, struct cachefiles_object, active_node);
98                 if (object->dentry > dentry)
99                         p = p->rb_left;
100                 else if (object->dentry < dentry)
101                         p = p->rb_right;
102                 else
103                         goto found_dentry;
104         }
105
106         write_unlock(&cache->active_lock);
107         trace_cachefiles_mark_buried(NULL, dentry, why);
108         _leave(" [no owner]");
109         return;
110
111         /* found the dentry for  */
112 found_dentry:
113         kdebug("preemptive burial: OBJ%x [%s] %pd",
114                object->fscache.debug_id,
115                object->fscache.state->name,
116                dentry);
117
118         trace_cachefiles_mark_buried(object, dentry, why);
119
120         if (fscache_object_is_live(&object->fscache)) {
121                 pr_err("\n");
122                 pr_err("Error: Can't preemptively bury live object\n");
123                 cachefiles_printk_object(object, NULL);
124         } else {
125                 if (why != FSCACHE_OBJECT_IS_STALE)
126                         fscache_object_mark_killed(&object->fscache, why);
127         }
128
129         write_unlock(&cache->active_lock);
130         _leave(" [owner marked]");
131 }
132
133 /*
134  * record the fact that an object is now active
135  */
136 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
137                                          struct cachefiles_object *object)
138 {
139         struct cachefiles_object *xobject;
140         struct rb_node **_p, *_parent = NULL;
141         struct dentry *dentry;
142
143         _enter(",%x", object->fscache.debug_id);
144
145 try_again:
146         write_lock(&cache->active_lock);
147
148         dentry = object->dentry;
149         trace_cachefiles_mark_active(object, dentry);
150
151         if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
152                 pr_err("Error: Object already active\n");
153                 cachefiles_printk_object(object, NULL);
154                 BUG();
155         }
156
157         _p = &cache->active_nodes.rb_node;
158         while (*_p) {
159                 _parent = *_p;
160                 xobject = rb_entry(_parent,
161                                    struct cachefiles_object, active_node);
162
163                 ASSERT(xobject != object);
164
165                 if (xobject->dentry > dentry)
166                         _p = &(*_p)->rb_left;
167                 else if (xobject->dentry < dentry)
168                         _p = &(*_p)->rb_right;
169                 else
170                         goto wait_for_old_object;
171         }
172
173         rb_link_node(&object->active_node, _parent, _p);
174         rb_insert_color(&object->active_node, &cache->active_nodes);
175
176         write_unlock(&cache->active_lock);
177         _leave(" = 0");
178         return 0;
179
180         /* an old object from a previous incarnation is hogging the slot - we
181          * need to wait for it to be destroyed */
182 wait_for_old_object:
183         trace_cachefiles_wait_active(object, dentry, xobject);
184         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
185
186         if (fscache_object_is_live(&xobject->fscache)) {
187                 pr_err("\n");
188                 pr_err("Error: Unexpected object collision\n");
189                 cachefiles_printk_object(object, xobject);
190         }
191         atomic_inc(&xobject->usage);
192         write_unlock(&cache->active_lock);
193
194         if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
195                 wait_queue_head_t *wq;
196
197                 signed long timeout = 60 * HZ;
198                 wait_queue_entry_t wait;
199                 bool requeue;
200
201                 /* if the object we're waiting for is queued for processing,
202                  * then just put ourselves on the queue behind it */
203                 if (work_pending(&xobject->fscache.work)) {
204                         _debug("queue OBJ%x behind OBJ%x immediately",
205                                object->fscache.debug_id,
206                                xobject->fscache.debug_id);
207                         goto requeue;
208                 }
209
210                 /* otherwise we sleep until either the object we're waiting for
211                  * is done, or the fscache_object is congested */
212                 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
213                 init_wait(&wait);
214                 requeue = false;
215                 do {
216                         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
217                         if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
218                                 break;
219
220                         requeue = fscache_object_sleep_till_congested(&timeout);
221                 } while (timeout > 0 && !requeue);
222                 finish_wait(wq, &wait);
223
224                 if (requeue &&
225                     test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
226                         _debug("queue OBJ%x behind OBJ%x after wait",
227                                object->fscache.debug_id,
228                                xobject->fscache.debug_id);
229                         goto requeue;
230                 }
231
232                 if (timeout <= 0) {
233                         pr_err("\n");
234                         pr_err("Error: Overlong wait for old active object to go away\n");
235                         cachefiles_printk_object(object, xobject);
236                         goto requeue;
237                 }
238         }
239
240         ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
241
242         cache->cache.ops->put_object(&xobject->fscache,
243                 (enum fscache_obj_ref_trace)cachefiles_obj_put_wait_retry);
244         goto try_again;
245
246 requeue:
247         cache->cache.ops->put_object(&xobject->fscache,
248                 (enum fscache_obj_ref_trace)cachefiles_obj_put_wait_timeo);
249         _leave(" = -ETIMEDOUT");
250         return -ETIMEDOUT;
251 }
252
253 /*
254  * Mark an object as being inactive.
255  */
256 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
257                                      struct cachefiles_object *object,
258                                      blkcnt_t i_blocks)
259 {
260         struct dentry *dentry = object->dentry;
261         struct inode *inode = d_backing_inode(dentry);
262
263         trace_cachefiles_mark_inactive(object, dentry, inode);
264
265         write_lock(&cache->active_lock);
266         rb_erase(&object->active_node, &cache->active_nodes);
267         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
268         write_unlock(&cache->active_lock);
269
270         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
271
272         /* This object can now be culled, so we need to let the daemon know
273          * that there is something it can remove if it needs to.
274          */
275         atomic_long_add(i_blocks, &cache->b_released);
276         if (atomic_inc_return(&cache->f_released))
277                 cachefiles_state_changed(cache);
278 }
279
280 /*
281  * delete an object representation from the cache
282  * - file backed objects are unlinked
283  * - directory backed objects are stuffed into the graveyard for userspace to
284  *   delete
285  * - unlocks the directory mutex
286  */
287 static int cachefiles_bury_object(struct cachefiles_cache *cache,
288                                   struct cachefiles_object *object,
289                                   struct dentry *dir,
290                                   struct dentry *rep,
291                                   bool preemptive,
292                                   enum fscache_why_object_killed why)
293 {
294         struct dentry *grave, *trap;
295         struct path path, path_to_graveyard;
296         char nbuffer[8 + 8 + 1];
297         int ret;
298
299         _enter(",'%pd','%pd'", dir, rep);
300
301         /* non-directories can just be unlinked */
302         if (!d_is_dir(rep)) {
303                 _debug("unlink stale object");
304
305                 path.mnt = cache->mnt;
306                 path.dentry = dir;
307                 ret = security_path_unlink(&path, rep);
308                 if (ret < 0) {
309                         cachefiles_io_error(cache, "Unlink security error");
310                 } else {
311                         trace_cachefiles_unlink(object, rep, why);
312                         ret = vfs_unlink(&init_user_ns, d_inode(dir), rep,
313                                          NULL);
314
315                         if (preemptive)
316                                 cachefiles_mark_object_buried(cache, rep, why);
317                 }
318
319                 inode_unlock(d_inode(dir));
320
321                 if (ret == -EIO)
322                         cachefiles_io_error(cache, "Unlink failed");
323
324                 _leave(" = %d", ret);
325                 return ret;
326         }
327
328         /* directories have to be moved to the graveyard */
329         _debug("move stale object to graveyard");
330         inode_unlock(d_inode(dir));
331
332 try_again:
333         /* first step is to make up a grave dentry in the graveyard */
334         sprintf(nbuffer, "%08x%08x",
335                 (uint32_t) ktime_get_real_seconds(),
336                 (uint32_t) atomic_inc_return(&cache->gravecounter));
337
338         /* do the multiway lock magic */
339         trap = lock_rename(cache->graveyard, dir);
340
341         /* do some checks before getting the grave dentry */
342         if (rep->d_parent != dir || IS_DEADDIR(d_inode(rep))) {
343                 /* the entry was probably culled when we dropped the parent dir
344                  * lock */
345                 unlock_rename(cache->graveyard, dir);
346                 _leave(" = 0 [culled?]");
347                 return 0;
348         }
349
350         if (!d_can_lookup(cache->graveyard)) {
351                 unlock_rename(cache->graveyard, dir);
352                 cachefiles_io_error(cache, "Graveyard no longer a directory");
353                 return -EIO;
354         }
355
356         if (trap == rep) {
357                 unlock_rename(cache->graveyard, dir);
358                 cachefiles_io_error(cache, "May not make directory loop");
359                 return -EIO;
360         }
361
362         if (d_mountpoint(rep)) {
363                 unlock_rename(cache->graveyard, dir);
364                 cachefiles_io_error(cache, "Mountpoint in cache");
365                 return -EIO;
366         }
367
368         grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
369         if (IS_ERR(grave)) {
370                 unlock_rename(cache->graveyard, dir);
371
372                 if (PTR_ERR(grave) == -ENOMEM) {
373                         _leave(" = -ENOMEM");
374                         return -ENOMEM;
375                 }
376
377                 cachefiles_io_error(cache, "Lookup error %ld",
378                                     PTR_ERR(grave));
379                 return -EIO;
380         }
381
382         if (d_is_positive(grave)) {
383                 unlock_rename(cache->graveyard, dir);
384                 dput(grave);
385                 grave = NULL;
386                 cond_resched();
387                 goto try_again;
388         }
389
390         if (d_mountpoint(grave)) {
391                 unlock_rename(cache->graveyard, dir);
392                 dput(grave);
393                 cachefiles_io_error(cache, "Mountpoint in graveyard");
394                 return -EIO;
395         }
396
397         /* target should not be an ancestor of source */
398         if (trap == grave) {
399                 unlock_rename(cache->graveyard, dir);
400                 dput(grave);
401                 cachefiles_io_error(cache, "May not make directory loop");
402                 return -EIO;
403         }
404
405         /* attempt the rename */
406         path.mnt = cache->mnt;
407         path.dentry = dir;
408         path_to_graveyard.mnt = cache->mnt;
409         path_to_graveyard.dentry = cache->graveyard;
410         ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
411         if (ret < 0) {
412                 cachefiles_io_error(cache, "Rename security error %d", ret);
413         } else {
414                 struct renamedata rd = {
415                         .old_mnt_userns = &init_user_ns,
416                         .old_dir        = d_inode(dir),
417                         .old_dentry     = rep,
418                         .new_mnt_userns = &init_user_ns,
419                         .new_dir        = d_inode(cache->graveyard),
420                         .new_dentry     = grave,
421                 };
422                 trace_cachefiles_rename(object, rep, grave, why);
423                 ret = vfs_rename(&rd);
424                 if (ret != 0 && ret != -ENOMEM)
425                         cachefiles_io_error(cache,
426                                             "Rename failed with error %d", ret);
427
428                 if (preemptive)
429                         cachefiles_mark_object_buried(cache, rep, why);
430         }
431
432         unlock_rename(cache->graveyard, dir);
433         dput(grave);
434         _leave(" = 0");
435         return 0;
436 }
437
438 /*
439  * delete an object representation from the cache
440  */
441 int cachefiles_delete_object(struct cachefiles_cache *cache,
442                              struct cachefiles_object *object)
443 {
444         struct dentry *dir;
445         int ret;
446
447         _enter(",OBJ%x{%pd}", object->fscache.debug_id, object->dentry);
448
449         ASSERT(object->dentry);
450         ASSERT(d_backing_inode(object->dentry));
451         ASSERT(object->dentry->d_parent);
452
453         dir = dget_parent(object->dentry);
454
455         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
456
457         if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
458                 /* object allocation for the same key preemptively deleted this
459                  * object's file so that it could create its own file */
460                 _debug("object preemptively buried");
461                 inode_unlock(d_inode(dir));
462                 ret = 0;
463         } else {
464                 /* we need to check that our parent is _still_ our parent - it
465                  * may have been renamed */
466                 if (dir == object->dentry->d_parent) {
467                         ret = cachefiles_bury_object(cache, object, dir,
468                                                      object->dentry, false,
469                                                      FSCACHE_OBJECT_WAS_RETIRED);
470                 } else {
471                         /* it got moved, presumably by cachefilesd culling it,
472                          * so it's no longer in the key path and we can ignore
473                          * it */
474                         inode_unlock(d_inode(dir));
475                         ret = 0;
476                 }
477         }
478
479         dput(dir);
480         _leave(" = %d", ret);
481         return ret;
482 }
483
484 /*
485  * walk from the parent object to the child object through the backing
486  * filesystem, creating directories as we go
487  */
488 int cachefiles_walk_to_object(struct cachefiles_object *parent,
489                               struct cachefiles_object *object,
490                               const char *key,
491                               struct cachefiles_xattr *auxdata)
492 {
493         struct cachefiles_cache *cache;
494         struct dentry *dir, *next = NULL;
495         struct inode *inode;
496         struct path path;
497         const char *name;
498         int ret, nlen;
499
500         _enter("OBJ%x{%pd},OBJ%x,%s,",
501                parent->fscache.debug_id, parent->dentry,
502                object->fscache.debug_id, key);
503
504         cache = container_of(parent->fscache.cache,
505                              struct cachefiles_cache, cache);
506         path.mnt = cache->mnt;
507
508         ASSERT(parent->dentry);
509         ASSERT(d_backing_inode(parent->dentry));
510
511         if (!(d_is_dir(parent->dentry))) {
512                 // TODO: convert file to dir
513                 _leave("looking up in none directory");
514                 return -ENOBUFS;
515         }
516
517         dir = dget(parent->dentry);
518
519 advance:
520         /* attempt to transit the first directory component */
521         name = key;
522         nlen = strlen(key);
523
524         /* key ends in a double NUL */
525         key = key + nlen + 1;
526         if (!*key)
527                 key = NULL;
528
529 lookup_again:
530         /* search the current directory for the element name */
531         _debug("lookup '%s'", name);
532
533         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
534
535         next = lookup_one_len(name, dir, nlen);
536         if (IS_ERR(next)) {
537                 trace_cachefiles_lookup(object, next, NULL);
538                 goto lookup_error;
539         }
540
541         inode = d_backing_inode(next);
542         trace_cachefiles_lookup(object, next, inode);
543         _debug("next -> %pd %s", next, inode ? "positive" : "negative");
544
545         if (!key)
546                 object->new = !inode;
547
548         /* if this element of the path doesn't exist, then the lookup phase
549          * failed, and we can release any readers in the certain knowledge that
550          * there's nothing for them to actually read */
551         if (d_is_negative(next))
552                 fscache_object_lookup_negative(&object->fscache);
553
554         /* we need to create the object if it's negative */
555         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
556                 /* index objects and intervening tree levels must be subdirs */
557                 if (d_is_negative(next)) {
558                         ret = cachefiles_has_space(cache, 1, 0);
559                         if (ret < 0)
560                                 goto no_space_error;
561
562                         path.dentry = dir;
563                         ret = security_path_mkdir(&path, next, 0);
564                         if (ret < 0)
565                                 goto create_error;
566                         ret = vfs_mkdir(&init_user_ns, d_inode(dir), next, 0);
567                         if (!key)
568                                 trace_cachefiles_mkdir(object, next, ret);
569                         if (ret < 0)
570                                 goto create_error;
571
572                         if (unlikely(d_unhashed(next))) {
573                                 dput(next);
574                                 inode_unlock(d_inode(dir));
575                                 goto lookup_again;
576                         }
577                         ASSERT(d_backing_inode(next));
578
579                         _debug("mkdir -> %pd{ino=%lu}",
580                                next, d_backing_inode(next)->i_ino);
581
582                 } else if (!d_can_lookup(next)) {
583                         pr_err("inode %lu is not a directory\n",
584                                d_backing_inode(next)->i_ino);
585                         ret = -ENOBUFS;
586                         goto error;
587                 }
588
589         } else {
590                 /* non-index objects start out life as files */
591                 if (d_is_negative(next)) {
592                         ret = cachefiles_has_space(cache, 1, 0);
593                         if (ret < 0)
594                                 goto no_space_error;
595
596                         path.dentry = dir;
597                         ret = security_path_mknod(&path, next, S_IFREG, 0);
598                         if (ret < 0)
599                                 goto create_error;
600                         ret = vfs_create(&init_user_ns, d_inode(dir), next,
601                                          S_IFREG, true);
602                         trace_cachefiles_create(object, next, ret);
603                         if (ret < 0)
604                                 goto create_error;
605
606                         ASSERT(d_backing_inode(next));
607
608                         _debug("create -> %pd{ino=%lu}",
609                                next, d_backing_inode(next)->i_ino);
610
611                 } else if (!d_can_lookup(next) &&
612                            !d_is_reg(next)
613                            ) {
614                         pr_err("inode %lu is not a file or directory\n",
615                                d_backing_inode(next)->i_ino);
616                         ret = -ENOBUFS;
617                         goto error;
618                 }
619         }
620
621         /* process the next component */
622         if (key) {
623                 _debug("advance");
624                 inode_unlock(d_inode(dir));
625                 dput(dir);
626                 dir = next;
627                 next = NULL;
628                 goto advance;
629         }
630
631         /* we've found the object we were looking for */
632         object->dentry = next;
633
634         /* if we've found that the terminal object exists, then we need to
635          * check its attributes and delete it if it's out of date */
636         if (!object->new) {
637                 _debug("validate '%pd'", next);
638
639                 ret = cachefiles_check_object_xattr(object, auxdata);
640                 if (ret == -ESTALE) {
641                         /* delete the object (the deleter drops the directory
642                          * mutex) */
643                         object->dentry = NULL;
644
645                         ret = cachefiles_bury_object(cache, object, dir, next,
646                                                      true,
647                                                      FSCACHE_OBJECT_IS_STALE);
648                         dput(next);
649                         next = NULL;
650
651                         if (ret < 0)
652                                 goto delete_error;
653
654                         _debug("redo lookup");
655                         fscache_object_retrying_stale(&object->fscache);
656                         goto lookup_again;
657                 }
658         }
659
660         /* note that we're now using this object */
661         ret = cachefiles_mark_object_active(cache, object);
662
663         inode_unlock(d_inode(dir));
664         dput(dir);
665         dir = NULL;
666
667         if (ret == -ETIMEDOUT)
668                 goto mark_active_timed_out;
669
670         _debug("=== OBTAINED_OBJECT ===");
671
672         if (object->new) {
673                 /* attach data to a newly constructed terminal object */
674                 ret = cachefiles_set_object_xattr(object, auxdata);
675                 if (ret < 0)
676                         goto check_error;
677         } else {
678                 /* always update the atime on an object we've just looked up
679                  * (this is used to keep track of culling, and atimes are only
680                  * updated by read, write and readdir but not lookup or
681                  * open) */
682                 path.dentry = next;
683                 touch_atime(&path);
684         }
685
686         /* open a file interface onto a data file */
687         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
688                 if (d_is_reg(object->dentry)) {
689                         const struct address_space_operations *aops;
690
691                         ret = -EPERM;
692                         aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
693                         if (!aops->bmap)
694                                 goto check_error;
695                         if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
696                                 goto check_error;
697
698                         object->backer = object->dentry;
699                 } else {
700                         BUG(); // TODO: open file in data-class subdir
701                 }
702         }
703
704         object->new = 0;
705         fscache_obtained_object(&object->fscache);
706
707         _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
708         return 0;
709
710 no_space_error:
711         fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
712 create_error:
713         _debug("create error %d", ret);
714         if (ret == -EIO)
715                 cachefiles_io_error(cache, "Create/mkdir failed");
716         goto error;
717
718 mark_active_timed_out:
719         _debug("mark active timed out");
720         goto release_dentry;
721
722 check_error:
723         _debug("check error %d", ret);
724         cachefiles_mark_object_inactive(
725                 cache, object, d_backing_inode(object->dentry)->i_blocks);
726 release_dentry:
727         dput(object->dentry);
728         object->dentry = NULL;
729         goto error_out;
730
731 delete_error:
732         _debug("delete error %d", ret);
733         goto error_out2;
734
735 lookup_error:
736         _debug("lookup error %ld", PTR_ERR(next));
737         ret = PTR_ERR(next);
738         if (ret == -EIO)
739                 cachefiles_io_error(cache, "Lookup failed");
740         next = NULL;
741 error:
742         inode_unlock(d_inode(dir));
743         dput(next);
744 error_out2:
745         dput(dir);
746 error_out:
747         _leave(" = error %d", -ret);
748         return ret;
749 }
750
751 /*
752  * get a subdirectory
753  */
754 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
755                                         struct dentry *dir,
756                                         const char *dirname)
757 {
758         struct dentry *subdir;
759         struct path path;
760         int ret;
761
762         _enter(",,%s", dirname);
763
764         /* search the current directory for the element name */
765         inode_lock(d_inode(dir));
766
767 retry:
768         subdir = lookup_one_len(dirname, dir, strlen(dirname));
769         if (IS_ERR(subdir)) {
770                 if (PTR_ERR(subdir) == -ENOMEM)
771                         goto nomem_d_alloc;
772                 goto lookup_error;
773         }
774
775         _debug("subdir -> %pd %s",
776                subdir, d_backing_inode(subdir) ? "positive" : "negative");
777
778         /* we need to create the subdir if it doesn't exist yet */
779         if (d_is_negative(subdir)) {
780                 ret = cachefiles_has_space(cache, 1, 0);
781                 if (ret < 0)
782                         goto mkdir_error;
783
784                 _debug("attempt mkdir");
785
786                 path.mnt = cache->mnt;
787                 path.dentry = dir;
788                 ret = security_path_mkdir(&path, subdir, 0700);
789                 if (ret < 0)
790                         goto mkdir_error;
791                 ret = vfs_mkdir(&init_user_ns, d_inode(dir), subdir, 0700);
792                 if (ret < 0)
793                         goto mkdir_error;
794
795                 if (unlikely(d_unhashed(subdir))) {
796                         dput(subdir);
797                         goto retry;
798                 }
799                 ASSERT(d_backing_inode(subdir));
800
801                 _debug("mkdir -> %pd{ino=%lu}",
802                        subdir, d_backing_inode(subdir)->i_ino);
803         }
804
805         inode_unlock(d_inode(dir));
806
807         /* we need to make sure the subdir is a directory */
808         ASSERT(d_backing_inode(subdir));
809
810         if (!d_can_lookup(subdir)) {
811                 pr_err("%s is not a directory\n", dirname);
812                 ret = -EIO;
813                 goto check_error;
814         }
815
816         ret = -EPERM;
817         if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
818             !d_backing_inode(subdir)->i_op->lookup ||
819             !d_backing_inode(subdir)->i_op->mkdir ||
820             !d_backing_inode(subdir)->i_op->create ||
821             !d_backing_inode(subdir)->i_op->rename ||
822             !d_backing_inode(subdir)->i_op->rmdir ||
823             !d_backing_inode(subdir)->i_op->unlink)
824                 goto check_error;
825
826         _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
827         return subdir;
828
829 check_error:
830         dput(subdir);
831         _leave(" = %d [check]", ret);
832         return ERR_PTR(ret);
833
834 mkdir_error:
835         inode_unlock(d_inode(dir));
836         dput(subdir);
837         pr_err("mkdir %s failed with error %d\n", dirname, ret);
838         return ERR_PTR(ret);
839
840 lookup_error:
841         inode_unlock(d_inode(dir));
842         ret = PTR_ERR(subdir);
843         pr_err("Lookup %s failed with error %d\n", dirname, ret);
844         return ERR_PTR(ret);
845
846 nomem_d_alloc:
847         inode_unlock(d_inode(dir));
848         _leave(" = -ENOMEM");
849         return ERR_PTR(-ENOMEM);
850 }
851
852 /*
853  * find out if an object is in use or not
854  * - if finds object and it's not in use:
855  *   - returns a pointer to the object and a reference on it
856  *   - returns with the directory locked
857  */
858 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
859                                               struct dentry *dir,
860                                               char *filename)
861 {
862         struct cachefiles_object *object;
863         struct rb_node *_n;
864         struct dentry *victim;
865         int ret;
866
867         //_enter(",%pd/,%s",
868         //       dir, filename);
869
870         /* look up the victim */
871         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
872
873         victim = lookup_one_len(filename, dir, strlen(filename));
874         if (IS_ERR(victim))
875                 goto lookup_error;
876
877         //_debug("victim -> %pd %s",
878         //       victim, d_backing_inode(victim) ? "positive" : "negative");
879
880         /* if the object is no longer there then we probably retired the object
881          * at the netfs's request whilst the cull was in progress
882          */
883         if (d_is_negative(victim)) {
884                 inode_unlock(d_inode(dir));
885                 dput(victim);
886                 _leave(" = -ENOENT [absent]");
887                 return ERR_PTR(-ENOENT);
888         }
889
890         /* check to see if we're using this object */
891         read_lock(&cache->active_lock);
892
893         _n = cache->active_nodes.rb_node;
894
895         while (_n) {
896                 object = rb_entry(_n, struct cachefiles_object, active_node);
897
898                 if (object->dentry > victim)
899                         _n = _n->rb_left;
900                 else if (object->dentry < victim)
901                         _n = _n->rb_right;
902                 else
903                         goto object_in_use;
904         }
905
906         read_unlock(&cache->active_lock);
907
908         //_leave(" = %pd", victim);
909         return victim;
910
911 object_in_use:
912         read_unlock(&cache->active_lock);
913         inode_unlock(d_inode(dir));
914         dput(victim);
915         //_leave(" = -EBUSY [in use]");
916         return ERR_PTR(-EBUSY);
917
918 lookup_error:
919         inode_unlock(d_inode(dir));
920         ret = PTR_ERR(victim);
921         if (ret == -ENOENT) {
922                 /* file or dir now absent - probably retired by netfs */
923                 _leave(" = -ESTALE [absent]");
924                 return ERR_PTR(-ESTALE);
925         }
926
927         if (ret == -EIO) {
928                 cachefiles_io_error(cache, "Lookup failed");
929         } else if (ret != -ENOMEM) {
930                 pr_err("Internal error: %d\n", ret);
931                 ret = -EIO;
932         }
933
934         _leave(" = %d", ret);
935         return ERR_PTR(ret);
936 }
937
938 /*
939  * cull an object if it's not in use
940  * - called only by cache manager daemon
941  */
942 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
943                     char *filename)
944 {
945         struct dentry *victim;
946         int ret;
947
948         _enter(",%pd/,%s", dir, filename);
949
950         victim = cachefiles_check_active(cache, dir, filename);
951         if (IS_ERR(victim))
952                 return PTR_ERR(victim);
953
954         _debug("victim -> %pd %s",
955                victim, d_backing_inode(victim) ? "positive" : "negative");
956
957         /* okay... the victim is not being used so we can cull it
958          * - start by marking it as stale
959          */
960         _debug("victim is cullable");
961
962         ret = cachefiles_remove_object_xattr(cache, victim);
963         if (ret < 0)
964                 goto error_unlock;
965
966         /*  actually remove the victim (drops the dir mutex) */
967         _debug("bury");
968
969         ret = cachefiles_bury_object(cache, NULL, dir, victim, false,
970                                      FSCACHE_OBJECT_WAS_CULLED);
971         if (ret < 0)
972                 goto error;
973
974         dput(victim);
975         _leave(" = 0");
976         return 0;
977
978 error_unlock:
979         inode_unlock(d_inode(dir));
980 error:
981         dput(victim);
982         if (ret == -ENOENT) {
983                 /* file or dir now absent - probably retired by netfs */
984                 _leave(" = -ESTALE [absent]");
985                 return -ESTALE;
986         }
987
988         if (ret != -ENOMEM) {
989                 pr_err("Internal error: %d\n", ret);
990                 ret = -EIO;
991         }
992
993         _leave(" = %d", ret);
994         return ret;
995 }
996
997 /*
998  * find out if an object is in use or not
999  * - called only by cache manager daemon
1000  * - returns -EBUSY or 0 to indicate whether an object is in use or not
1001  */
1002 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
1003                             char *filename)
1004 {
1005         struct dentry *victim;
1006
1007         //_enter(",%pd/,%s",
1008         //       dir, filename);
1009
1010         victim = cachefiles_check_active(cache, dir, filename);
1011         if (IS_ERR(victim))
1012                 return PTR_ERR(victim);
1013
1014         inode_unlock(d_inode(dir));
1015         dput(victim);
1016         //_leave(" = 0");
1017         return 0;
1018 }