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