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 ([email protected])
   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                 BUG();
 199         }
 200         atomic_inc(&xobject->usage);
 201         write_unlock(&cache->active_lock);
 202
 203         if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 204                 wait_queue_head_t *wq;
 205
 206                 signed long timeout = 60 * HZ;
 207                 wait_queue_entry_t wait;
 208                 bool requeue;
 209
 210                 /* if the object we're waiting for is queued for processing,
 211                  * then just put ourselves on the queue behind it */
 212                 if (work_pending(&xobject->fscache.work)) {
 213                         _debug("queue OBJ%x behind OBJ%x immediately",
 214                                object->fscache.debug_id,
 215                                xobject->fscache.debug_id);
 216                         goto requeue;
 217                 }
 218
 219                 /* otherwise we sleep until either the object we're waiting for
 220                  * is done, or the fscache_object is congested */
 221                 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
 222                 init_wait(&wait);
 223                 requeue = false;
 224                 do {
 225                         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 226                         if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
 227                                 break;
 228
 229                         requeue = fscache_object_sleep_till_congested(&timeout);
 230                 } while (timeout > 0 && !requeue);
 231                 finish_wait(wq, &wait);
 232
 233                 if (requeue &&
 234                     test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 235                         _debug("queue OBJ%x behind OBJ%x after wait",
 236                                object->fscache.debug_id,
 237                                xobject->fscache.debug_id);
 238                         goto requeue;
 239                 }
 240
 241                 if (timeout <= 0) {
 242                         pr_err("\n");
 243                         pr_err("Error: Overlong wait for old active object to go away\n");
 244                         cachefiles_printk_object(object, xobject);
 245                         goto requeue;
 246                 }
 247         }
 248
 249         ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
 250
 251         cache->cache.ops->put_object(&xobject->fscache);
 252         goto try_again;
 253
 254 requeue:
 255         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 256         cache->cache.ops->put_object(&xobject->fscache);
 257         _leave(" = -ETIMEDOUT");
 258         return -ETIMEDOUT;
 259 }
 260
 261 /*
 262  * Mark an object as being inactive.
 263  */
 264 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
 265                                      struct cachefiles_object *object,
 266                                      blkcnt_t i_blocks)
 267 {
 268         write_lock(&cache->active_lock);
 269         rb_erase(&object->active_node, &cache->active_nodes);
 270         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 271         write_unlock(&cache->active_lock);
 272
 273         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
 274
 275         /* This object can now be culled, so we need to let the daemon know
 276          * that there is something it can remove if it needs to.
 277          */
 278         atomic_long_add(i_blocks, &cache->b_released);
 279         if (atomic_inc_return(&cache->f_released))
 280                 cachefiles_state_changed(cache);
 281 }
 282
 283 /*
 284  * delete an object representation from the cache
 285  * - file backed objects are unlinked
 286  * - directory backed objects are stuffed into the graveyard for userspace to
 287  *   delete
 288  * - unlocks the directory mutex
 289  */
 290 static int cachefiles_bury_object(struct cachefiles_cache *cache,
 291                                   struct dentry *dir,
 292                                   struct dentry *rep,
 293                                   bool preemptive,
 294                                   enum fscache_why_object_killed why)
 295 {
 296         struct dentry *grave, *trap;
 297         struct path path, path_to_graveyard;
 298         char nbuffer[8 + 8 + 1];
 299         int ret;
 300
 301         _enter(",'%pd','%pd'", dir, rep);
 302
 303         _debug("remove %p from %p", rep, dir);
 304
 305         /* non-directories can just be unlinked */
 306         if (!d_is_dir(rep)) {
 307                 _debug("unlink stale object");
 308
 309                 path.mnt = cache->mnt;
 310                 path.dentry = dir;
 311                 ret = security_path_unlink(&path, rep);
 312                 if (ret < 0) {
 313                         cachefiles_io_error(cache, "Unlink security error");
 314                 } else {
 315                         ret = vfs_unlink(d_inode(dir), rep, NULL);
 316
 317                         if (preemptive)
 318                                 cachefiles_mark_object_buried(cache, rep, why);
 319                 }
 320
 321                 inode_unlock(d_inode(dir));
 322
 323                 if (ret == -EIO)
 324                         cachefiles_io_error(cache, "Unlink failed");
 325
 326                 _leave(" = %d", ret);
 327                 return ret;
 328         }
 329
 330         /* directories have to be moved to the graveyard */
 331         _debug("move stale object to graveyard");
 332         inode_unlock(d_inode(dir));
 333
 334 try_again:
 335         /* first step is to make up a grave dentry in the graveyard */
 336         sprintf(nbuffer, "%08x%08x",
 337                 (uint32_t) get_seconds(),
 338                 (uint32_t) atomic_inc_return(&cache->gravecounter));
 339
 340         /* do the multiway lock magic */
 341         trap = lock_rename(cache->graveyard, dir);
 342
 343         /* do some checks before getting the grave dentry */
 344         if (rep->d_parent != dir) {
 345                 /* the entry was probably culled when we dropped the parent dir
 346                  * lock */
 347                 unlock_rename(cache->graveyard, dir);
 348                 _leave(" = 0 [culled?]");
 349                 return 0;
 350         }
 351
 352         if (!d_can_lookup(cache->graveyard)) {
 353                 unlock_rename(cache->graveyard, dir);
 354                 cachefiles_io_error(cache, "Graveyard no longer a directory");
 355                 return -EIO;
 356         }
 357
 358         if (trap == rep) {
 359                 unlock_rename(cache->graveyard, dir);
 360                 cachefiles_io_error(cache, "May not make directory loop");
 361                 return -EIO;
 362         }
 363
 364         if (d_mountpoint(rep)) {
 365                 unlock_rename(cache->graveyard, dir);
 366                 cachefiles_io_error(cache, "Mountpoint in cache");
 367                 return -EIO;
 368         }
 369
 370         grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
 371         if (IS_ERR(grave)) {
 372                 unlock_rename(cache->graveyard, dir);
 373
 374                 if (PTR_ERR(grave) == -ENOMEM) {
 375                         _leave(" = -ENOMEM");
 376                         return -ENOMEM;
 377                 }
 378
 379                 cachefiles_io_error(cache, "Lookup error %ld",
 380                                     PTR_ERR(grave));
 381                 return -EIO;
 382         }
 383
 384         if (d_is_positive(grave)) {
 385                 unlock_rename(cache->graveyard, dir);
 386                 dput(grave);
 387                 grave = NULL;
 388                 cond_resched();
 389                 goto try_again;
 390         }
 391
 392         if (d_mountpoint(grave)) {
 393                 unlock_rename(cache->graveyard, dir);
 394                 dput(grave);
 395                 cachefiles_io_error(cache, "Mountpoint in graveyard");
 396                 return -EIO;
 397         }
 398
 399         /* target should not be an ancestor of source */
 400         if (trap == grave) {
 401                 unlock_rename(cache->graveyard, dir);
 402                 dput(grave);
 403                 cachefiles_io_error(cache, "May not make directory loop");
 404                 return -EIO;
 405         }
 406
 407         /* attempt the rename */
 408         path.mnt = cache->mnt;
 409         path.dentry = dir;
 410         path_to_graveyard.mnt = cache->mnt;
 411         path_to_graveyard.dentry = cache->graveyard;
 412         ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
 413         if (ret < 0) {
 414                 cachefiles_io_error(cache, "Rename security error %d", ret);
 415         } else {
 416                 ret = vfs_rename(d_inode(dir), rep,
 417                                  d_inode(cache->graveyard), grave, NULL, 0);
 418                 if (ret != 0 && ret != -ENOMEM)
 419                         cachefiles_io_error(cache,
 420                                             "Rename failed with error %d", ret);
 421
 422                 if (preemptive)
 423                         cachefiles_mark_object_buried(cache, rep, why);
 424         }
 425
 426         unlock_rename(cache->graveyard, dir);
 427         dput(grave);
 428         _leave(" = 0");
 429         return 0;
 430 }
 431
 432 /*
 433  * delete an object representation from the cache
 434  */
 435 int cachefiles_delete_object(struct cachefiles_cache *cache,
 436                              struct cachefiles_object *object)
 437 {
 438         struct dentry *dir;
 439         int ret;
 440
 441         _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
 442
 443         ASSERT(object->dentry);
 444         ASSERT(d_backing_inode(object->dentry));
 445         ASSERT(object->dentry->d_parent);
 446
 447         dir = dget_parent(object->dentry);
 448
 449         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 450
 451         if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
 452                 /* object allocation for the same key preemptively deleted this
 453                  * object's file so that it could create its own file */
 454                 _debug("object preemptively buried");
 455                 inode_unlock(d_inode(dir));
 456                 ret = 0;
 457         } else {
 458                 /* we need to check that our parent is _still_ our parent - it
 459                  * may have been renamed */
 460                 if (dir == object->dentry->d_parent) {
 461                         ret = cachefiles_bury_object(cache, dir,
 462                                                      object->dentry, false,
 463                                                      FSCACHE_OBJECT_WAS_RETIRED);
 464                 } else {
 465                         /* it got moved, presumably by cachefilesd culling it,
 466                          * so it's no longer in the key path and we can ignore
 467                          * it */
 468                         inode_unlock(d_inode(dir));
 469                         ret = 0;
 470                 }
 471         }
 472
 473         dput(dir);
 474         _leave(" = %d", ret);
 475         return ret;
 476 }
 477
 478 /*
 479  * walk from the parent object to the child object through the backing
 480  * filesystem, creating directories as we go
 481  */
 482 int cachefiles_walk_to_object(struct cachefiles_object *parent,
 483                               struct cachefiles_object *object,
 484                               const char *key,
 485                               struct cachefiles_xattr *auxdata)
 486 {
 487         struct cachefiles_cache *cache;
 488         struct dentry *dir, *next = NULL;
 489         struct path path;
 490         unsigned long start;
 491         const char *name;
 492         int ret, nlen;
 493
 494         _enter("OBJ%x{%p},OBJ%x,%s,",
 495                parent->fscache.debug_id, parent->dentry,
 496                object->fscache.debug_id, key);
 497
 498         cache = container_of(parent->fscache.cache,
 499                              struct cachefiles_cache, cache);
 500         path.mnt = cache->mnt;
 501
 502         ASSERT(parent->dentry);
 503         ASSERT(d_backing_inode(parent->dentry));
 504
 505         if (!(d_is_dir(parent->dentry))) {
 506                 // TODO: convert file to dir
 507                 _leave("looking up in none directory");
 508                 return -ENOBUFS;
 509         }
 510
 511         dir = dget(parent->dentry);
 512
 513 advance:
 514         /* attempt to transit the first directory component */
 515         name = key;
 516         nlen = strlen(key);
 517
 518         /* key ends in a double NUL */
 519         key = key + nlen + 1;
 520         if (!*key)
 521                 key = NULL;
 522
 523 lookup_again:
 524         /* search the current directory for the element name */
 525         _debug("lookup '%s'", name);
 526
 527         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 528
 529         start = jiffies;
 530         next = lookup_one_len(name, dir, nlen);
 531         cachefiles_hist(cachefiles_lookup_histogram, start);
 532         if (IS_ERR(next))
 533                 goto lookup_error;
 534
 535         _debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
 536
 537         if (!key)
 538                 object->new = !d_backing_inode(next);
 539
 540         /* if this element of the path doesn't exist, then the lookup phase
 541          * failed, and we can release any readers in the certain knowledge that
 542          * there's nothing for them to actually read */
 543         if (d_is_negative(next))
 544                 fscache_object_lookup_negative(&object->fscache);
 545
 546         /* we need to create the object if it's negative */
 547         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
 548                 /* index objects and intervening tree levels must be subdirs */
 549                 if (d_is_negative(next)) {
 550                         ret = cachefiles_has_space(cache, 1, 0);
 551                         if (ret < 0)
 552                                 goto no_space_error;
 553
 554                         path.dentry = dir;
 555                         ret = security_path_mkdir(&path, next, 0);
 556                         if (ret < 0)
 557                                 goto create_error;
 558                         start = jiffies;
 559                         ret = vfs_mkdir(d_inode(dir), next, 0);
 560                         cachefiles_hist(cachefiles_mkdir_histogram, start);
 561                         if (ret < 0)
 562                                 goto create_error;
 563
 564                         ASSERT(d_backing_inode(next));
 565
 566                         _debug("mkdir -> %p{%p{ino=%lu}}",
 567                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 568
 569                 } else if (!d_can_lookup(next)) {
 570                         pr_err("inode %lu is not a directory\n",
 571                                d_backing_inode(next)->i_ino);
 572                         ret = -ENOBUFS;
 573                         goto error;
 574                 }
 575
 576         } else {
 577                 /* non-index objects start out life as files */
 578                 if (d_is_negative(next)) {
 579                         ret = cachefiles_has_space(cache, 1, 0);
 580                         if (ret < 0)
 581                                 goto no_space_error;
 582
 583                         path.dentry = dir;
 584                         ret = security_path_mknod(&path, next, S_IFREG, 0);
 585                         if (ret < 0)
 586                                 goto create_error;
 587                         start = jiffies;
 588                         ret = vfs_create(d_inode(dir), next, S_IFREG, true);
 589                         cachefiles_hist(cachefiles_create_histogram, start);
 590                         if (ret < 0)
 591                                 goto create_error;
 592
 593                         ASSERT(d_backing_inode(next));
 594
 595                         _debug("create -> %p{%p{ino=%lu}}",
 596                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 597
 598                 } else if (!d_can_lookup(next) &&
 599                            !d_is_reg(next)
 600                            ) {
 601                         pr_err("inode %lu is not a file or directory\n",
 602                                d_backing_inode(next)->i_ino);
 603                         ret = -ENOBUFS;
 604                         goto error;
 605                 }
 606         }
 607
 608         /* process the next component */
 609         if (key) {
 610                 _debug("advance");
 611                 inode_unlock(d_inode(dir));
 612                 dput(dir);
 613                 dir = next;
 614                 next = NULL;
 615                 goto advance;
 616         }
 617
 618         /* we've found the object we were looking for */
 619         object->dentry = next;
 620
 621         /* if we've found that the terminal object exists, then we need to
 622          * check its attributes and delete it if it's out of date */
 623         if (!object->new) {
 624                 _debug("validate '%pd'", next);
 625
 626                 ret = cachefiles_check_object_xattr(object, auxdata);
 627                 if (ret == -ESTALE) {
 628                         /* delete the object (the deleter drops the directory
 629                          * mutex) */
 630                         object->dentry = NULL;
 631
 632                         ret = cachefiles_bury_object(cache, dir, next, true,
 633                                                      FSCACHE_OBJECT_IS_STALE);
 634                         dput(next);
 635                         next = NULL;
 636
 637                         if (ret < 0)
 638                                 goto delete_error;
 639
 640                         _debug("redo lookup");
 641                         fscache_object_retrying_stale(&object->fscache);
 642                         goto lookup_again;
 643                 }
 644         }
 645
 646         /* note that we're now using this object */
 647         ret = cachefiles_mark_object_active(cache, object);
 648
 649         inode_unlock(d_inode(dir));
 650         dput(dir);
 651         dir = NULL;
 652
 653         if (ret == -ETIMEDOUT)
 654                 goto mark_active_timed_out;
 655
 656         _debug("=== OBTAINED_OBJECT ===");
 657
 658         if (object->new) {
 659                 /* attach data to a newly constructed terminal object */
 660                 ret = cachefiles_set_object_xattr(object, auxdata);
 661                 if (ret < 0)
 662                         goto check_error;
 663         } else {
 664                 /* always update the atime on an object we've just looked up
 665                  * (this is used to keep track of culling, and atimes are only
 666                  * updated by read, write and readdir but not lookup or
 667                  * open) */
 668                 path.dentry = next;
 669                 touch_atime(&path);
 670         }
 671
 672         /* open a file interface onto a data file */
 673         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
 674                 if (d_is_reg(object->dentry)) {
 675                         const struct address_space_operations *aops;
 676
 677                         ret = -EPERM;
 678                         aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
 679                         if (!aops->bmap)
 680                                 goto check_error;
 681                         if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
 682                                 goto check_error;
 683
 684                         object->backer = object->dentry;
 685                 } else {
 686                         BUG(); // TODO: open file in data-class subdir
 687                 }
 688         }
 689
 690         object->new = 0;
 691         fscache_obtained_object(&object->fscache);
 692
 693         _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
 694         return 0;
 695
 696 no_space_error:
 697         fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
 698 create_error:
 699         _debug("create error %d", ret);
 700         if (ret == -EIO)
 701                 cachefiles_io_error(cache, "Create/mkdir failed");
 702         goto error;
 703
 704 mark_active_timed_out:
 705         _debug("mark active timed out");
 706         goto release_dentry;
 707
 708 check_error:
 709         _debug("check error %d", ret);
 710         cachefiles_mark_object_inactive(
 711                 cache, object, d_backing_inode(object->dentry)->i_blocks);
 712 release_dentry:
 713         dput(object->dentry);
 714         object->dentry = NULL;
 715         goto error_out;
 716
 717 delete_error:
 718         _debug("delete error %d", ret);
 719         goto error_out2;
 720
 721 lookup_error:
 722         _debug("lookup error %ld", PTR_ERR(next));
 723         ret = PTR_ERR(next);
 724         if (ret == -EIO)
 725                 cachefiles_io_error(cache, "Lookup failed");
 726         next = NULL;
 727 error:
 728         inode_unlock(d_inode(dir));
 729         dput(next);
 730 error_out2:
 731         dput(dir);
 732 error_out:
 733         _leave(" = error %d", -ret);
 734         return ret;
 735 }
 736
 737 /*
 738  * get a subdirectory
 739  */
 740 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 741                                         struct dentry *dir,
 742                                         const char *dirname)
 743 {
 744         struct dentry *subdir;
 745         unsigned long start;
 746         struct path path;
 747         int ret;
 748
 749         _enter(",,%s", dirname);
 750
 751         /* search the current directory for the element name */
 752         inode_lock(d_inode(dir));
 753
 754         start = jiffies;
 755         subdir = lookup_one_len(dirname, dir, strlen(dirname));
 756         cachefiles_hist(cachefiles_lookup_histogram, start);
 757         if (IS_ERR(subdir)) {
 758                 if (PTR_ERR(subdir) == -ENOMEM)
 759                         goto nomem_d_alloc;
 760                 goto lookup_error;
 761         }
 762
 763         _debug("subdir -> %p %s",
 764                subdir, d_backing_inode(subdir) ? "positive" : "negative");
 765
 766         /* we need to create the subdir if it doesn't exist yet */
 767         if (d_is_negative(subdir)) {
 768                 ret = cachefiles_has_space(cache, 1, 0);
 769                 if (ret < 0)
 770                         goto mkdir_error;
 771
 772                 _debug("attempt mkdir");
 773
 774                 path.mnt = cache->mnt;
 775                 path.dentry = dir;
 776                 ret = security_path_mkdir(&path, subdir, 0700);
 777                 if (ret < 0)
 778                         goto mkdir_error;
 779                 ret = vfs_mkdir(d_inode(dir), subdir, 0700);
 780                 if (ret < 0)
 781                         goto mkdir_error;
 782
 783                 ASSERT(d_backing_inode(subdir));
 784
 785                 _debug("mkdir -> %p{%p{ino=%lu}}",
 786                        subdir,
 787                        d_backing_inode(subdir),
 788                        d_backing_inode(subdir)->i_ino);
 789         }
 790
 791         inode_unlock(d_inode(dir));
 792
 793         /* we need to make sure the subdir is a directory */
 794         ASSERT(d_backing_inode(subdir));
 795
 796         if (!d_can_lookup(subdir)) {
 797                 pr_err("%s is not a directory\n", dirname);
 798                 ret = -EIO;
 799                 goto check_error;
 800         }
 801
 802         ret = -EPERM;
 803         if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
 804             !d_backing_inode(subdir)->i_op->lookup ||
 805             !d_backing_inode(subdir)->i_op->mkdir ||
 806             !d_backing_inode(subdir)->i_op->create ||
 807             !d_backing_inode(subdir)->i_op->rename ||
 808             !d_backing_inode(subdir)->i_op->rmdir ||
 809             !d_backing_inode(subdir)->i_op->unlink)
 810                 goto check_error;
 811
 812         _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
 813         return subdir;
 814
 815 check_error:
 816         dput(subdir);
 817         _leave(" = %d [check]", ret);
 818         return ERR_PTR(ret);
 819
 820 mkdir_error:
 821         inode_unlock(d_inode(dir));
 822         dput(subdir);
 823         pr_err("mkdir %s failed with error %d\n", dirname, ret);
 824         return ERR_PTR(ret);
 825
 826 lookup_error:
 827         inode_unlock(d_inode(dir));
 828         ret = PTR_ERR(subdir);
 829         pr_err("Lookup %s failed with error %d\n", dirname, ret);
 830         return ERR_PTR(ret);
 831
 832 nomem_d_alloc:
 833         inode_unlock(d_inode(dir));
 834         _leave(" = -ENOMEM");
 835         return ERR_PTR(-ENOMEM);
 836 }
 837
 838 /*
 839  * find out if an object is in use or not
 840  * - if finds object and it's not in use:
 841  *   - returns a pointer to the object and a reference on it
 842  *   - returns with the directory locked
 843  */
 844 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
 845                                               struct dentry *dir,
 846                                               char *filename)
 847 {
 848         struct cachefiles_object *object;
 849         struct rb_node *_n;
 850         struct dentry *victim;
 851         unsigned long start;
 852         int ret;
 853
 854         //_enter(",%pd/,%s",
 855         //       dir, filename);
 856
 857         /* look up the victim */
 858         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 859
 860         start = jiffies;
 861         victim = lookup_one_len(filename, dir, strlen(filename));
 862         cachefiles_hist(cachefiles_lookup_histogram, start);
 863         if (IS_ERR(victim))
 864                 goto lookup_error;
 865
 866         //_debug("victim -> %p %s",
 867         //       victim, d_backing_inode(victim) ? "positive" : "negative");
 868
 869         /* if the object is no longer there then we probably retired the object
 870          * at the netfs's request whilst the cull was in progress
 871          */
 872         if (d_is_negative(victim)) {
 873                 inode_unlock(d_inode(dir));
 874                 dput(victim);
 875                 _leave(" = -ENOENT [absent]");
 876                 return ERR_PTR(-ENOENT);
 877         }
 878
 879         /* check to see if we're using this object */
 880         read_lock(&cache->active_lock);
 881
 882         _n = cache->active_nodes.rb_node;
 883
 884         while (_n) {
 885                 object = rb_entry(_n, struct cachefiles_object, active_node);
 886
 887                 if (object->dentry > victim)
 888                         _n = _n->rb_left;
 889                 else if (object->dentry < victim)
 890                         _n = _n->rb_right;
 891                 else
 892                         goto object_in_use;
 893         }
 894
 895         read_unlock(&cache->active_lock);
 896
 897         //_leave(" = %p", victim);
 898         return victim;
 899
 900 object_in_use:
 901         read_unlock(&cache->active_lock);
 902         inode_unlock(d_inode(dir));
 903         dput(victim);
 904         //_leave(" = -EBUSY [in use]");
 905         return ERR_PTR(-EBUSY);
 906
 907 lookup_error:
 908         inode_unlock(d_inode(dir));
 909         ret = PTR_ERR(victim);
 910         if (ret == -ENOENT) {
 911                 /* file or dir now absent - probably retired by netfs */
 912                 _leave(" = -ESTALE [absent]");
 913                 return ERR_PTR(-ESTALE);
 914         }
 915
 916         if (ret == -EIO) {
 917                 cachefiles_io_error(cache, "Lookup failed");
 918         } else if (ret != -ENOMEM) {
 919                 pr_err("Internal error: %d\n", ret);
 920                 ret = -EIO;
 921         }
 922
 923         _leave(" = %d", ret);
 924         return ERR_PTR(ret);
 925 }
 926
 927 /*
 928  * cull an object if it's not in use
 929  * - called only by cache manager daemon
 930  */
 931 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
 932                     char *filename)
 933 {
 934         struct dentry *victim;
 935         int ret;
 936
 937         _enter(",%pd/,%s", dir, filename);
 938
 939         victim = cachefiles_check_active(cache, dir, filename);
 940         if (IS_ERR(victim))
 941                 return PTR_ERR(victim);
 942
 943         _debug("victim -> %p %s",
 944                victim, d_backing_inode(victim) ? "positive" : "negative");
 945
 946         /* okay... the victim is not being used so we can cull it
 947          * - start by marking it as stale
 948          */
 949         _debug("victim is cullable");
 950
 951         ret = cachefiles_remove_object_xattr(cache, victim);
 952         if (ret < 0)
 953                 goto error_unlock;
 954
 955         /*  actually remove the victim (drops the dir mutex) */
 956         _debug("bury");
 957
 958         ret = cachefiles_bury_object(cache, dir, victim, false,
 959                                      FSCACHE_OBJECT_WAS_CULLED);
 960         if (ret < 0)
 961                 goto error;
 962
 963         dput(victim);
 964         _leave(" = 0");
 965         return 0;
 966
 967 error_unlock:
 968         inode_unlock(d_inode(dir));
 969 error:
 970         dput(victim);
 971         if (ret == -ENOENT) {
 972                 /* file or dir now absent - probably retired by netfs */
 973                 _leave(" = -ESTALE [absent]");
 974                 return -ESTALE;
 975         }
 976
 977         if (ret != -ENOMEM) {
 978                 pr_err("Internal error: %d\n", ret);
 979                 ret = -EIO;
 980         }
 981
 982         _leave(" = %d", ret);
 983         return ret;
 984 }
 985
 986 /*
 987  * find out if an object is in use or not
 988  * - called only by cache manager daemon
 989  * - returns -EBUSY or 0 to indicate whether an object is in use or not
 990  */
 991 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
 992                             char *filename)
 993 {
 994         struct dentry *victim;
 995
 996         //_enter(",%pd/,%s",
 997         //       dir, filename);
 998
 999         victim = cachefiles_check_active(cache, dir, filename);
1000         if (IS_ERR(victim))
1001                 return PTR_ERR(victim);
1002
1003         inode_unlock(d_inode(dir));
1004         dput(victim);
1005         //_leave(" = 0");
1006         return 0;
1007 }