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                 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_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 {
 267         blkcnt_t i_blocks = d_backing_inode(object->dentry)->i_blocks;
 268
 269         write_lock(&cache->active_lock);
 270         rb_erase(&object->active_node, &cache->active_nodes);
 271         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 272         write_unlock(&cache->active_lock);
 273
 274         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
 275
 276         /* This object can now be culled, so we need to let the daemon know
 277          * that there is something it can remove if it needs to.
 278          */
 279         atomic_long_add(i_blocks, &cache->b_released);
 280         if (atomic_inc_return(&cache->f_released))
 281                 cachefiles_state_changed(cache);
 282 }
 283
 284 /*
 285  * delete an object representation from the cache
 286  * - file backed objects are unlinked
 287  * - directory backed objects are stuffed into the graveyard for userspace to
 288  *   delete
 289  * - unlocks the directory mutex
 290  */
 291 static int cachefiles_bury_object(struct cachefiles_cache *cache,
 292                                   struct dentry *dir,
 293                                   struct dentry *rep,
 294                                   bool preemptive,
 295                                   enum fscache_why_object_killed why)
 296 {
 297         struct dentry *grave, *trap;
 298         struct path path, path_to_graveyard;
 299         char nbuffer[8 + 8 + 1];
 300         int ret;
 301
 302         _enter(",'%pd','%pd'", dir, rep);
 303
 304         _debug("remove %p from %p", rep, dir);
 305
 306         /* non-directories can just be unlinked */
 307         if (!d_is_dir(rep)) {
 308                 _debug("unlink stale object");
 309
 310                 path.mnt = cache->mnt;
 311                 path.dentry = dir;
 312                 ret = security_path_unlink(&path, rep);
 313                 if (ret < 0) {
 314                         cachefiles_io_error(cache, "Unlink security error");
 315                 } else {
 316                         ret = vfs_unlink(d_inode(dir), rep, NULL);
 317
 318                         if (preemptive)
 319                                 cachefiles_mark_object_buried(cache, rep, why);
 320                 }
 321
 322                 inode_unlock(d_inode(dir));
 323
 324                 if (ret == -EIO)
 325                         cachefiles_io_error(cache, "Unlink failed");
 326
 327                 _leave(" = %d", ret);
 328                 return ret;
 329         }
 330
 331         /* directories have to be moved to the graveyard */
 332         _debug("move stale object to graveyard");
 333         inode_unlock(d_inode(dir));
 334
 335 try_again:
 336         /* first step is to make up a grave dentry in the graveyard */
 337         sprintf(nbuffer, "%08x%08x",
 338                 (uint32_t) get_seconds(),
 339                 (uint32_t) atomic_inc_return(&cache->gravecounter));
 340
 341         /* do the multiway lock magic */
 342         trap = lock_rename(cache->graveyard, dir);
 343
 344         /* do some checks before getting the grave dentry */
 345         if (rep->d_parent != dir) {
 346                 /* the entry was probably culled when we dropped the parent dir
 347                  * lock */
 348                 unlock_rename(cache->graveyard, dir);
 349                 _leave(" = 0 [culled?]");
 350                 return 0;
 351         }
 352
 353         if (!d_can_lookup(cache->graveyard)) {
 354                 unlock_rename(cache->graveyard, dir);
 355                 cachefiles_io_error(cache, "Graveyard no longer a directory");
 356                 return -EIO;
 357         }
 358
 359         if (trap == rep) {
 360                 unlock_rename(cache->graveyard, dir);
 361                 cachefiles_io_error(cache, "May not make directory loop");
 362                 return -EIO;
 363         }
 364
 365         if (d_mountpoint(rep)) {
 366                 unlock_rename(cache->graveyard, dir);
 367                 cachefiles_io_error(cache, "Mountpoint in cache");
 368                 return -EIO;
 369         }
 370
 371         grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
 372         if (IS_ERR(grave)) {
 373                 unlock_rename(cache->graveyard, dir);
 374
 375                 if (PTR_ERR(grave) == -ENOMEM) {
 376                         _leave(" = -ENOMEM");
 377                         return -ENOMEM;
 378                 }
 379
 380                 cachefiles_io_error(cache, "Lookup error %ld",
 381                                     PTR_ERR(grave));
 382                 return -EIO;
 383         }
 384
 385         if (d_is_positive(grave)) {
 386                 unlock_rename(cache->graveyard, dir);
 387                 dput(grave);
 388                 grave = NULL;
 389                 cond_resched();
 390                 goto try_again;
 391         }
 392
 393         if (d_mountpoint(grave)) {
 394                 unlock_rename(cache->graveyard, dir);
 395                 dput(grave);
 396                 cachefiles_io_error(cache, "Mountpoint in graveyard");
 397                 return -EIO;
 398         }
 399
 400         /* target should not be an ancestor of source */
 401         if (trap == grave) {
 402                 unlock_rename(cache->graveyard, dir);
 403                 dput(grave);
 404                 cachefiles_io_error(cache, "May not make directory loop");
 405                 return -EIO;
 406         }
 407
 408         /* attempt the rename */
 409         path.mnt = cache->mnt;
 410         path.dentry = dir;
 411         path_to_graveyard.mnt = cache->mnt;
 412         path_to_graveyard.dentry = cache->graveyard;
 413         ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
 414         if (ret < 0) {
 415                 cachefiles_io_error(cache, "Rename security error %d", ret);
 416         } else {
 417                 ret = vfs_rename(d_inode(dir), rep,
 418                                  d_inode(cache->graveyard), grave, NULL, 0);
 419                 if (ret != 0 && ret != -ENOMEM)
 420                         cachefiles_io_error(cache,
 421                                             "Rename failed with error %d", ret);
 422
 423                 if (preemptive)
 424                         cachefiles_mark_object_buried(cache, rep, why);
 425         }
 426
 427         unlock_rename(cache->graveyard, dir);
 428         dput(grave);
 429         _leave(" = 0");
 430         return 0;
 431 }
 432
 433 /*
 434  * delete an object representation from the cache
 435  */
 436 int cachefiles_delete_object(struct cachefiles_cache *cache,
 437                              struct cachefiles_object *object)
 438 {
 439         struct dentry *dir;
 440         int ret;
 441
 442         _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
 443
 444         ASSERT(object->dentry);
 445         ASSERT(d_backing_inode(object->dentry));
 446         ASSERT(object->dentry->d_parent);
 447
 448         dir = dget_parent(object->dentry);
 449
 450         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 451
 452         if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
 453                 /* object allocation for the same key preemptively deleted this
 454                  * object's file so that it could create its own file */
 455                 _debug("object preemptively buried");
 456                 inode_unlock(d_inode(dir));
 457                 ret = 0;
 458         } else {
 459                 /* we need to check that our parent is _still_ our parent - it
 460                  * may have been renamed */
 461                 if (dir == object->dentry->d_parent) {
 462                         ret = cachefiles_bury_object(cache, dir,
 463                                                      object->dentry, false,
 464                                                      FSCACHE_OBJECT_WAS_RETIRED);
 465                 } else {
 466                         /* it got moved, presumably by cachefilesd culling it,
 467                          * so it's no longer in the key path and we can ignore
 468                          * it */
 469                         inode_unlock(d_inode(dir));
 470                         ret = 0;
 471                 }
 472         }
 473
 474         dput(dir);
 475         _leave(" = %d", ret);
 476         return ret;
 477 }
 478
 479 /*
 480  * walk from the parent object to the child object through the backing
 481  * filesystem, creating directories as we go
 482  */
 483 int cachefiles_walk_to_object(struct cachefiles_object *parent,
 484                               struct cachefiles_object *object,
 485                               const char *key,
 486                               struct cachefiles_xattr *auxdata)
 487 {
 488         struct cachefiles_cache *cache;
 489         struct dentry *dir, *next = NULL;
 490         struct path path;
 491         unsigned long start;
 492         const char *name;
 493         int ret, nlen;
 494
 495         _enter("OBJ%x{%p},OBJ%x,%s,",
 496                parent->fscache.debug_id, parent->dentry,
 497                object->fscache.debug_id, key);
 498
 499         cache = container_of(parent->fscache.cache,
 500                              struct cachefiles_cache, cache);
 501         path.mnt = cache->mnt;
 502
 503         ASSERT(parent->dentry);
 504         ASSERT(d_backing_inode(parent->dentry));
 505
 506         if (!(d_is_dir(parent->dentry))) {
 507                 // TODO: convert file to dir
 508                 _leave("looking up in none directory");
 509                 return -ENOBUFS;
 510         }
 511
 512         dir = dget(parent->dentry);
 513
 514 advance:
 515         /* attempt to transit the first directory component */
 516         name = key;
 517         nlen = strlen(key);
 518
 519         /* key ends in a double NUL */
 520         key = key + nlen + 1;
 521         if (!*key)
 522                 key = NULL;
 523
 524 lookup_again:
 525         /* search the current directory for the element name */
 526         _debug("lookup '%s'", name);
 527
 528         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 529
 530         start = jiffies;
 531         next = lookup_one_len(name, dir, nlen);
 532         cachefiles_hist(cachefiles_lookup_histogram, start);
 533         if (IS_ERR(next))
 534                 goto lookup_error;
 535
 536         _debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
 537
 538         if (!key)
 539                 object->new = !d_backing_inode(next);
 540
 541         /* if this element of the path doesn't exist, then the lookup phase
 542          * failed, and we can release any readers in the certain knowledge that
 543          * there's nothing for them to actually read */
 544         if (d_is_negative(next))
 545                 fscache_object_lookup_negative(&object->fscache);
 546
 547         /* we need to create the object if it's negative */
 548         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
 549                 /* index objects and intervening tree levels must be subdirs */
 550                 if (d_is_negative(next)) {
 551                         ret = cachefiles_has_space(cache, 1, 0);
 552                         if (ret < 0)
 553                                 goto no_space_error;
 554
 555                         path.dentry = dir;
 556                         ret = security_path_mkdir(&path, next, 0);
 557                         if (ret < 0)
 558                                 goto create_error;
 559                         start = jiffies;
 560                         ret = vfs_mkdir(d_inode(dir), next, 0);
 561                         cachefiles_hist(cachefiles_mkdir_histogram, start);
 562                         if (ret < 0)
 563                                 goto create_error;
 564
 565                         ASSERT(d_backing_inode(next));
 566
 567                         _debug("mkdir -> %p{%p{ino=%lu}}",
 568                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 569
 570                 } else if (!d_can_lookup(next)) {
 571                         pr_err("inode %lu is not a directory\n",
 572                                d_backing_inode(next)->i_ino);
 573                         ret = -ENOBUFS;
 574                         goto error;
 575                 }
 576
 577         } else {
 578                 /* non-index objects start out life as files */
 579                 if (d_is_negative(next)) {
 580                         ret = cachefiles_has_space(cache, 1, 0);
 581                         if (ret < 0)
 582                                 goto no_space_error;
 583
 584                         path.dentry = dir;
 585                         ret = security_path_mknod(&path, next, S_IFREG, 0);
 586                         if (ret < 0)
 587                                 goto create_error;
 588                         start = jiffies;
 589                         ret = vfs_create(d_inode(dir), next, S_IFREG, true);
 590                         cachefiles_hist(cachefiles_create_histogram, start);
 591                         if (ret < 0)
 592                                 goto create_error;
 593
 594                         ASSERT(d_backing_inode(next));
 595
 596                         _debug("create -> %p{%p{ino=%lu}}",
 597                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 598
 599                 } else if (!d_can_lookup(next) &&
 600                            !d_is_reg(next)
 601                            ) {
 602                         pr_err("inode %lu is not a file or directory\n",
 603                                d_backing_inode(next)->i_ino);
 604                         ret = -ENOBUFS;
 605                         goto error;
 606                 }
 607         }
 608
 609         /* process the next component */
 610         if (key) {
 611                 _debug("advance");
 612                 inode_unlock(d_inode(dir));
 613                 dput(dir);
 614                 dir = next;
 615                 next = NULL;
 616                 goto advance;
 617         }
 618
 619         /* we've found the object we were looking for */
 620         object->dentry = next;
 621
 622         /* if we've found that the terminal object exists, then we need to
 623          * check its attributes and delete it if it's out of date */
 624         if (!object->new) {
 625                 _debug("validate '%pd'", next);
 626
 627                 ret = cachefiles_check_object_xattr(object, auxdata);
 628                 if (ret == -ESTALE) {
 629                         /* delete the object (the deleter drops the directory
 630                          * mutex) */
 631                         object->dentry = NULL;
 632
 633                         ret = cachefiles_bury_object(cache, dir, next, true,
 634                                                      FSCACHE_OBJECT_IS_STALE);
 635                         dput(next);
 636                         next = NULL;
 637
 638                         if (ret < 0)
 639                                 goto delete_error;
 640
 641                         _debug("redo lookup");
 642                         fscache_object_retrying_stale(&object->fscache);
 643                         goto lookup_again;
 644                 }
 645         }
 646
 647         /* note that we're now using this object */
 648         ret = cachefiles_mark_object_active(cache, object);
 649
 650         inode_unlock(d_inode(dir));
 651         dput(dir);
 652         dir = NULL;
 653
 654         if (ret == -ETIMEDOUT)
 655                 goto mark_active_timed_out;
 656
 657         _debug("=== OBTAINED_OBJECT ===");
 658
 659         if (object->new) {
 660                 /* attach data to a newly constructed terminal object */
 661                 ret = cachefiles_set_object_xattr(object, auxdata);
 662                 if (ret < 0)
 663                         goto check_error;
 664         } else {
 665                 /* always update the atime on an object we've just looked up
 666                  * (this is used to keep track of culling, and atimes are only
 667                  * updated by read, write and readdir but not lookup or
 668                  * open) */
 669                 path.dentry = next;
 670                 touch_atime(&path);
 671         }
 672
 673         /* open a file interface onto a data file */
 674         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
 675                 if (d_is_reg(object->dentry)) {
 676                         const struct address_space_operations *aops;
 677
 678                         ret = -EPERM;
 679                         aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
 680                         if (!aops->bmap)
 681                                 goto check_error;
 682                         if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
 683                                 goto check_error;
 684
 685                         object->backer = object->dentry;
 686                 } else {
 687                         BUG(); // TODO: open file in data-class subdir
 688                 }
 689         }
 690
 691         object->new = 0;
 692         fscache_obtained_object(&object->fscache);
 693
 694         _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
 695         return 0;
 696
 697 no_space_error:
 698         fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
 699 create_error:
 700         _debug("create error %d", ret);
 701         if (ret == -EIO)
 702                 cachefiles_io_error(cache, "Create/mkdir failed");
 703         goto error;
 704
 705 mark_active_timed_out:
 706         _debug("mark active timed out");
 707         goto release_dentry;
 708
 709 check_error:
 710         _debug("check error %d", ret);
 711         cachefiles_mark_object_inactive(cache, object);
 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->rename2) ||
 809             !d_backing_inode(subdir)->i_op->rmdir ||
 810             !d_backing_inode(subdir)->i_op->unlink)
 811                 goto check_error;
 812
 813         _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
 814         return subdir;
 815
 816 check_error:
 817         dput(subdir);
 818         _leave(" = %d [check]", ret);
 819         return ERR_PTR(ret);
 820
 821 mkdir_error:
 822         inode_unlock(d_inode(dir));
 823         dput(subdir);
 824         pr_err("mkdir %s failed with error %d\n", dirname, ret);
 825         return ERR_PTR(ret);
 826
 827 lookup_error:
 828         inode_unlock(d_inode(dir));
 829         ret = PTR_ERR(subdir);
 830         pr_err("Lookup %s failed with error %d\n", dirname, ret);
 831         return ERR_PTR(ret);
 832
 833 nomem_d_alloc:
 834         inode_unlock(d_inode(dir));
 835         _leave(" = -ENOMEM");
 836         return ERR_PTR(-ENOMEM);
 837 }
 838
 839 /*
 840  * find out if an object is in use or not
 841  * - if finds object and it's not in use:
 842  *   - returns a pointer to the object and a reference on it
 843  *   - returns with the directory locked
 844  */
 845 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
 846                                               struct dentry *dir,
 847                                               char *filename)
 848 {
 849         struct cachefiles_object *object;
 850         struct rb_node *_n;
 851         struct dentry *victim;
 852         unsigned long start;
 853         int ret;
 854
 855         //_enter(",%pd/,%s",
 856         //       dir, filename);
 857
 858         /* look up the victim */
 859         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 860
 861         start = jiffies;
 862         victim = lookup_one_len(filename, dir, strlen(filename));
 863         cachefiles_hist(cachefiles_lookup_histogram, start);
 864         if (IS_ERR(victim))
 865                 goto lookup_error;
 866
 867         //_debug("victim -> %p %s",
 868         //       victim, d_backing_inode(victim) ? "positive" : "negative");
 869
 870         /* if the object is no longer there then we probably retired the object
 871          * at the netfs's request whilst the cull was in progress
 872          */
 873         if (d_is_negative(victim)) {
 874                 inode_unlock(d_inode(dir));
 875                 dput(victim);
 876                 _leave(" = -ENOENT [absent]");
 877                 return ERR_PTR(-ENOENT);
 878         }
 879
 880         /* check to see if we're using this object */
 881         read_lock(&cache->active_lock);
 882
 883         _n = cache->active_nodes.rb_node;
 884
 885         while (_n) {
 886                 object = rb_entry(_n, struct cachefiles_object, active_node);
 887
 888                 if (object->dentry > victim)
 889                         _n = _n->rb_left;
 890                 else if (object->dentry < victim)
 891                         _n = _n->rb_right;
 892                 else
 893                         goto object_in_use;
 894         }
 895
 896         read_unlock(&cache->active_lock);
 897
 898         //_leave(" = %p", victim);
 899         return victim;
 900
 901 object_in_use:
 902         read_unlock(&cache->active_lock);
 903         inode_unlock(d_inode(dir));
 904         dput(victim);
 905         //_leave(" = -EBUSY [in use]");
 906         return ERR_PTR(-EBUSY);
 907
 908 lookup_error:
 909         inode_unlock(d_inode(dir));
 910         ret = PTR_ERR(victim);
 911         if (ret == -ENOENT) {
 912                 /* file or dir now absent - probably retired by netfs */
 913                 _leave(" = -ESTALE [absent]");
 914                 return ERR_PTR(-ESTALE);
 915         }
 916
 917         if (ret == -EIO) {
 918                 cachefiles_io_error(cache, "Lookup failed");
 919         } else if (ret != -ENOMEM) {
 920                 pr_err("Internal error: %d\n", ret);
 921                 ret = -EIO;
 922         }
 923
 924         _leave(" = %d", ret);
 925         return ERR_PTR(ret);
 926 }
 927
 928 /*
 929  * cull an object if it's not in use
 930  * - called only by cache manager daemon
 931  */
 932 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
 933                     char *filename)
 934 {
 935         struct dentry *victim;
 936         int ret;
 937
 938         _enter(",%pd/,%s", dir, filename);
 939
 940         victim = cachefiles_check_active(cache, dir, filename);
 941         if (IS_ERR(victim))
 942                 return PTR_ERR(victim);
 943
 944         _debug("victim -> %p %s",
 945                victim, d_backing_inode(victim) ? "positive" : "negative");
 946
 947         /* okay... the victim is not being used so we can cull it
 948          * - start by marking it as stale
 949          */
 950         _debug("victim is cullable");
 951
 952         ret = cachefiles_remove_object_xattr(cache, victim);
 953         if (ret < 0)
 954                 goto error_unlock;
 955
 956         /*  actually remove the victim (drops the dir mutex) */
 957         _debug("bury");
 958
 959         ret = cachefiles_bury_object(cache, dir, victim, false,
 960                                      FSCACHE_OBJECT_WAS_CULLED);
 961         if (ret < 0)
 962                 goto error;
 963
 964         dput(victim);
 965         _leave(" = 0");
 966         return 0;
 967
 968 error_unlock:
 969         inode_unlock(d_inode(dir));
 970 error:
 971         dput(victim);
 972         if (ret == -ENOENT) {
 973                 /* file or dir now absent - probably retired by netfs */
 974                 _leave(" = -ESTALE [absent]");
 975                 return -ESTALE;
 976         }
 977
 978         if (ret != -ENOMEM) {
 979                 pr_err("Internal error: %d\n", ret);
 980                 ret = -EIO;
 981         }
 982
 983         _leave(" = %d", ret);
 984         return ret;
 985 }
 986
 987 /*
 988  * find out if an object is in use or not
 989  * - called only by cache manager daemon
 990  * - returns -EBUSY or 0 to indicate whether an object is in use or not
 991  */
 992 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
 993                             char *filename)
 994 {
 995         struct dentry *victim;
 996
 997         //_enter(",%pd/,%s",
 998         //       dir, filename);
 999
1000         victim = cachefiles_check_active(cache, dir, filename);
1001         if (IS_ERR(victim))
1002                 return PTR_ERR(victim);
1003
1004         inode_unlock(d_inode(dir));
1005         dput(victim);
1006         //_leave(" = 0");
1007         return 0;
1008 }