fs/cachefiles/namei.c

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