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(&xobject->fscache)) {
 193                 pr_err("\n");
 194                 pr_err("Error: Unexpected object collision\n");
 195                 cachefiles_printk_object(object, xobject);
 196         }
 197         atomic_inc(&xobject->usage);
 198         write_unlock(&cache->active_lock);
 199
 200         if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 201                 wait_queue_head_t *wq;
 202
 203                 signed long timeout = 60 * HZ;
 204                 wait_queue_t wait;
 205                 bool requeue;
 206
 207                 /* if the object we're waiting for is queued for processing,
 208                  * then just put ourselves on the queue behind it */
 209                 if (work_pending(&xobject->fscache.work)) {
 210                         _debug("queue OBJ%x behind OBJ%x immediately",
 211                                object->fscache.debug_id,
 212                                xobject->fscache.debug_id);
 213                         goto requeue;
 214                 }
 215
 216                 /* otherwise we sleep until either the object we're waiting for
 217                  * is done, or the fscache_object is congested */
 218                 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
 219                 init_wait(&wait);
 220                 requeue = false;
 221                 do {
 222                         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 223                         if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
 224                                 break;
 225
 226                         requeue = fscache_object_sleep_till_congested(&timeout);
 227                 } while (timeout > 0 && !requeue);
 228                 finish_wait(wq, &wait);
 229
 230                 if (requeue &&
 231                     test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 232                         _debug("queue OBJ%x behind OBJ%x after wait",
 233                                object->fscache.debug_id,
 234                                xobject->fscache.debug_id);
 235                         goto requeue;
 236                 }
 237
 238                 if (timeout <= 0) {
 239                         pr_err("\n");
 240                         pr_err("Error: Overlong wait for old active object to go away\n");
 241                         cachefiles_printk_object(object, xobject);
 242                         goto requeue;
 243                 }
 244         }
 245
 246         ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
 247
 248         cache->cache.ops->put_object(&xobject->fscache);
 249         goto try_again;
 250
 251 requeue:
 252         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 253         cache->cache.ops->put_object(&xobject->fscache);
 254         _leave(" = -ETIMEDOUT");
 255         return -ETIMEDOUT;
 256 }
 257
 258 /*
 259  * delete an object representation from the cache
 260  * - file backed objects are unlinked
 261  * - directory backed objects are stuffed into the graveyard for userspace to
 262  *   delete
 263  * - unlocks the directory mutex
 264  */
 265 static int cachefiles_bury_object(struct cachefiles_cache *cache,
 266                                   struct dentry *dir,
 267                                   struct dentry *rep,
 268                                   bool preemptive)
 269 {
 270         struct dentry *grave, *trap;
 271         struct path path, path_to_graveyard;
 272         char nbuffer[8 + 8 + 1];
 273         int ret;
 274
 275         _enter(",'%*.*s','%*.*s'",
 276                dir->d_name.len, dir->d_name.len, dir->d_name.name,
 277                rep->d_name.len, rep->d_name.len, rep->d_name.name);
 278
 279         _debug("remove %p from %p", rep, dir);
 280
 281         /* non-directories can just be unlinked */
 282         if (!S_ISDIR(rep->d_inode->i_mode)) {
 283                 _debug("unlink stale object");
 284
 285                 path.mnt = cache->mnt;
 286                 path.dentry = dir;
 287                 ret = security_path_unlink(&path, rep);
 288                 if (ret < 0) {
 289                         cachefiles_io_error(cache, "Unlink security error");
 290                 } else {
 291                         ret = vfs_unlink(dir->d_inode, rep, NULL);
 292
 293                         if (preemptive)
 294                                 cachefiles_mark_object_buried(cache, rep);
 295                 }
 296
 297                 mutex_unlock(&dir->d_inode->i_mutex);
 298
 299                 if (ret == -EIO)
 300                         cachefiles_io_error(cache, "Unlink failed");
 301
 302                 _leave(" = %d", ret);
 303                 return ret;
 304         }
 305
 306         /* directories have to be moved to the graveyard */
 307         _debug("move stale object to graveyard");
 308         mutex_unlock(&dir->d_inode->i_mutex);
 309
 310 try_again:
 311         /* first step is to make up a grave dentry in the graveyard */
 312         sprintf(nbuffer, "%08x%08x",
 313                 (uint32_t) get_seconds(),
 314                 (uint32_t) atomic_inc_return(&cache->gravecounter));
 315
 316         /* do the multiway lock magic */
 317         trap = lock_rename(cache->graveyard, dir);
 318
 319         /* do some checks before getting the grave dentry */
 320         if (rep->d_parent != dir || IS_DEADDIR(d_inode(rep))) {
 321                 /* the entry was probably culled when we dropped the parent dir
 322                  * lock */
 323                 unlock_rename(cache->graveyard, dir);
 324                 _leave(" = 0 [culled?]");
 325                 return 0;
 326         }
 327
 328         if (!S_ISDIR(cache->graveyard->d_inode->i_mode)) {
 329                 unlock_rename(cache->graveyard, dir);
 330                 cachefiles_io_error(cache, "Graveyard no longer a directory");
 331                 return -EIO;
 332         }
 333
 334         if (trap == rep) {
 335                 unlock_rename(cache->graveyard, dir);
 336                 cachefiles_io_error(cache, "May not make directory loop");
 337                 return -EIO;
 338         }
 339
 340         if (d_mountpoint(rep)) {
 341                 unlock_rename(cache->graveyard, dir);
 342                 cachefiles_io_error(cache, "Mountpoint in cache");
 343                 return -EIO;
 344         }
 345
 346         grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
 347         if (IS_ERR(grave)) {
 348                 unlock_rename(cache->graveyard, dir);
 349
 350                 if (PTR_ERR(grave) == -ENOMEM) {
 351                         _leave(" = -ENOMEM");
 352                         return -ENOMEM;
 353                 }
 354
 355                 cachefiles_io_error(cache, "Lookup error %ld",
 356                                     PTR_ERR(grave));
 357                 return -EIO;
 358         }
 359
 360         if (grave->d_inode) {
 361                 unlock_rename(cache->graveyard, dir);
 362                 dput(grave);
 363                 grave = NULL;
 364                 cond_resched();
 365                 goto try_again;
 366         }
 367
 368         if (d_mountpoint(grave)) {
 369                 unlock_rename(cache->graveyard, dir);
 370                 dput(grave);
 371                 cachefiles_io_error(cache, "Mountpoint in graveyard");
 372                 return -EIO;
 373         }
 374
 375         /* target should not be an ancestor of source */
 376         if (trap == grave) {
 377                 unlock_rename(cache->graveyard, dir);
 378                 dput(grave);
 379                 cachefiles_io_error(cache, "May not make directory loop");
 380                 return -EIO;
 381         }
 382
 383         /* attempt the rename */
 384         path.mnt = cache->mnt;
 385         path.dentry = dir;
 386         path_to_graveyard.mnt = cache->mnt;
 387         path_to_graveyard.dentry = cache->graveyard;
 388         ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
 389         if (ret < 0) {
 390                 cachefiles_io_error(cache, "Rename security error %d", ret);
 391         } else {
 392                 ret = vfs_rename(dir->d_inode, rep,
 393                                  cache->graveyard->d_inode, grave, NULL, 0);
 394                 if (ret != 0 && ret != -ENOMEM)
 395                         cachefiles_io_error(cache,
 396                                             "Rename failed with error %d", ret);
 397
 398                 if (preemptive)
 399                         cachefiles_mark_object_buried(cache, rep);
 400         }
 401
 402         unlock_rename(cache->graveyard, dir);
 403         dput(grave);
 404         _leave(" = 0");
 405         return 0;
 406 }
 407
 408 /*
 409  * delete an object representation from the cache
 410  */
 411 int cachefiles_delete_object(struct cachefiles_cache *cache,
 412                              struct cachefiles_object *object)
 413 {
 414         struct dentry *dir;
 415         int ret;
 416
 417         _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
 418
 419         ASSERT(object->dentry);
 420         ASSERT(object->dentry->d_inode);
 421         ASSERT(object->dentry->d_parent);
 422
 423         dir = dget_parent(object->dentry);
 424
 425         mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
 426
 427         if (test_bit(CACHEFILES_OBJECT_BURIED, &object->flags)) {
 428                 /* object allocation for the same key preemptively deleted this
 429                  * object's file so that it could create its own file */
 430                 _debug("object preemptively buried");
 431                 mutex_unlock(&dir->d_inode->i_mutex);
 432                 ret = 0;
 433         } else {
 434                 /* we need to check that our parent is _still_ our parent - it
 435                  * may have been renamed */
 436                 if (dir == object->dentry->d_parent) {
 437                         ret = cachefiles_bury_object(cache, dir,
 438                                                      object->dentry, false);
 439                 } else {
 440                         /* it got moved, presumably by cachefilesd culling it,
 441                          * so it's no longer in the key path and we can ignore
 442                          * it */
 443                         mutex_unlock(&dir->d_inode->i_mutex);
 444                         ret = 0;
 445                 }
 446         }
 447
 448         dput(dir);
 449         _leave(" = %d", ret);
 450         return ret;
 451 }
 452
 453 /*
 454  * walk from the parent object to the child object through the backing
 455  * filesystem, creating directories as we go
 456  */
 457 int cachefiles_walk_to_object(struct cachefiles_object *parent,
 458                               struct cachefiles_object *object,
 459                               const char *key,
 460                               struct cachefiles_xattr *auxdata)
 461 {
 462         struct cachefiles_cache *cache;
 463         struct dentry *dir, *next = NULL;
 464         struct path path;
 465         unsigned long start;
 466         const char *name;
 467         int ret, nlen;
 468
 469         _enter("OBJ%x{%p},OBJ%x,%s,",
 470                parent->fscache.debug_id, parent->dentry,
 471                object->fscache.debug_id, key);
 472
 473         cache = container_of(parent->fscache.cache,
 474                              struct cachefiles_cache, cache);
 475         path.mnt = cache->mnt;
 476
 477         ASSERT(parent->dentry);
 478         ASSERT(parent->dentry->d_inode);
 479
 480         if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
 481                 // TODO: convert file to dir
 482                 _leave("looking up in none directory");
 483                 return -ENOBUFS;
 484         }
 485
 486         dir = dget(parent->dentry);
 487
 488 advance:
 489         /* attempt to transit the first directory component */
 490         name = key;
 491         nlen = strlen(key);
 492
 493         /* key ends in a double NUL */
 494         key = key + nlen + 1;
 495         if (!*key)
 496                 key = NULL;
 497
 498 lookup_again:
 499         /* search the current directory for the element name */
 500         _debug("lookup '%s'", name);
 501
 502         mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_PARENT);
 503
 504         start = jiffies;
 505         next = lookup_one_len(name, dir, nlen);
 506         cachefiles_hist(cachefiles_lookup_histogram, start);
 507         if (IS_ERR(next))
 508                 goto lookup_error;
 509
 510         _debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
 511
 512         if (!key)
 513                 object->new = !next->d_inode;
 514
 515         /* if this element of the path doesn't exist, then the lookup phase
 516          * failed, and we can release any readers in the certain knowledge that
 517          * there's nothing for them to actually read */
 518         if (!next->d_inode)
 519                 fscache_object_lookup_negative(&object->fscache);
 520
 521         /* we need to create the object if it's negative */
 522         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
 523                 /* index objects and intervening tree levels must be subdirs */
 524                 if (!next->d_inode) {
 525                         ret = cachefiles_has_space(cache, 1, 0);
 526                         if (ret < 0)
 527                                 goto create_error;
 528
 529                         path.dentry = dir;
 530                         ret = security_path_mkdir(&path, next, 0);
 531                         if (ret < 0)
 532                                 goto create_error;
 533                         start = jiffies;
 534                         ret = vfs_mkdir(dir->d_inode, next, 0);
 535                         cachefiles_hist(cachefiles_mkdir_histogram, start);
 536                         if (ret < 0)
 537                                 goto create_error;
 538
 539                         ASSERT(next->d_inode);
 540
 541                         _debug("mkdir -> %p{%p{ino=%lu}}",
 542                                next, next->d_inode, next->d_inode->i_ino);
 543
 544                 } else if (!S_ISDIR(next->d_inode->i_mode)) {
 545                         pr_err("inode %lu is not a directory\n",
 546                                next->d_inode->i_ino);
 547                         ret = -ENOBUFS;
 548                         goto error;
 549                 }
 550
 551         } else {
 552                 /* non-index objects start out life as files */
 553                 if (!next->d_inode) {
 554                         ret = cachefiles_has_space(cache, 1, 0);
 555                         if (ret < 0)
 556                                 goto create_error;
 557
 558                         path.dentry = dir;
 559                         ret = security_path_mknod(&path, next, S_IFREG, 0);
 560                         if (ret < 0)
 561                                 goto create_error;
 562                         start = jiffies;
 563                         ret = vfs_create(dir->d_inode, next, S_IFREG, true);
 564                         cachefiles_hist(cachefiles_create_histogram, start);
 565                         if (ret < 0)
 566                                 goto create_error;
 567
 568                         ASSERT(next->d_inode);
 569
 570                         _debug("create -> %p{%p{ino=%lu}}",
 571                                next, next->d_inode, next->d_inode->i_ino);
 572
 573                 } else if (!S_ISDIR(next->d_inode->i_mode) &&
 574                            !S_ISREG(next->d_inode->i_mode)
 575                            ) {
 576                         pr_err("inode %lu is not a file or directory\n",
 577                                next->d_inode->i_ino);
 578                         ret = -ENOBUFS;
 579                         goto error;
 580                 }
 581         }
 582
 583         /* process the next component */
 584         if (key) {
 585                 _debug("advance");
 586                 mutex_unlock(&dir->d_inode->i_mutex);
 587                 dput(dir);
 588                 dir = next;
 589                 next = NULL;
 590                 goto advance;
 591         }
 592
 593         /* we've found the object we were looking for */
 594         object->dentry = next;
 595
 596         /* if we've found that the terminal object exists, then we need to
 597          * check its attributes and delete it if it's out of date */
 598         if (!object->new) {
 599                 _debug("validate '%*.*s'",
 600                        next->d_name.len, next->d_name.len, next->d_name.name);
 601
 602                 ret = cachefiles_check_object_xattr(object, auxdata);
 603                 if (ret == -ESTALE) {
 604                         /* delete the object (the deleter drops the directory
 605                          * mutex) */
 606                         object->dentry = NULL;
 607
 608                         ret = cachefiles_bury_object(cache, dir, next, true);
 609                         dput(next);
 610                         next = NULL;
 611
 612                         if (ret < 0)
 613                                 goto delete_error;
 614
 615                         _debug("redo lookup");
 616                         goto lookup_again;
 617                 }
 618         }
 619
 620         /* note that we're now using this object */
 621         ret = cachefiles_mark_object_active(cache, object);
 622
 623         mutex_unlock(&dir->d_inode->i_mutex);
 624         dput(dir);
 625         dir = NULL;
 626
 627         if (ret == -ETIMEDOUT)
 628                 goto mark_active_timed_out;
 629
 630         _debug("=== OBTAINED_OBJECT ===");
 631
 632         if (object->new) {
 633                 /* attach data to a newly constructed terminal object */
 634                 ret = cachefiles_set_object_xattr(object, auxdata);
 635                 if (ret < 0)
 636                         goto check_error;
 637         } else {
 638                 /* always update the atime on an object we've just looked up
 639                  * (this is used to keep track of culling, and atimes are only
 640                  * updated by read, write and readdir but not lookup or
 641                  * open) */
 642                 path.dentry = next;
 643                 touch_atime(&path);
 644         }
 645
 646         /* open a file interface onto a data file */
 647         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
 648                 if (S_ISREG(object->dentry->d_inode->i_mode)) {
 649                         const struct address_space_operations *aops;
 650
 651                         ret = -EPERM;
 652                         aops = object->dentry->d_inode->i_mapping->a_ops;
 653                         if (!aops->bmap)
 654                                 goto check_error;
 655
 656                         object->backer = object->dentry;
 657                 } else {
 658                         BUG(); // TODO: open file in data-class subdir
 659                 }
 660         }
 661
 662         object->new = 0;
 663         fscache_obtained_object(&object->fscache);
 664
 665         _leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
 666         return 0;
 667
 668 create_error:
 669         _debug("create error %d", ret);
 670         if (ret == -EIO)
 671                 cachefiles_io_error(cache, "Create/mkdir failed");
 672         goto error;
 673
 674 mark_active_timed_out:
 675         _debug("mark active timed out");
 676         goto release_dentry;
 677
 678 check_error:
 679         _debug("check error %d", ret);
 680         write_lock(&cache->active_lock);
 681         rb_erase(&object->active_node, &cache->active_nodes);
 682         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 683         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
 684         write_unlock(&cache->active_lock);
 685 release_dentry:
 686         dput(object->dentry);
 687         object->dentry = NULL;
 688         goto error_out;
 689
 690 delete_error:
 691         _debug("delete error %d", ret);
 692         goto error_out2;
 693
 694 lookup_error:
 695         _debug("lookup error %ld", PTR_ERR(next));
 696         ret = PTR_ERR(next);
 697         if (ret == -EIO)
 698                 cachefiles_io_error(cache, "Lookup failed");
 699         next = NULL;
 700 error:
 701         mutex_unlock(&dir->d_inode->i_mutex);
 702         dput(next);
 703 error_out2:
 704         dput(dir);
 705 error_out:
 706         _leave(" = error %d", -ret);
 707         return ret;
 708 }
 709
 710 /*
 711  * get a subdirectory
 712  */
 713 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 714                                         struct dentry *dir,
 715                                         const char *dirname)
 716 {
 717         struct dentry *subdir;
 718         unsigned long start;
 719         struct path path;
 720         int ret;
 721
 722         _enter(",,%s", dirname);
 723
 724         /* search the current directory for the element name */
 725         mutex_lock(&dir->d_inode->i_mutex);
 726
 727         start = jiffies;
 728         subdir = lookup_one_len(dirname, dir, strlen(dirname));
 729         cachefiles_hist(cachefiles_lookup_histogram, start);
 730         if (IS_ERR(subdir)) {
 731                 if (PTR_ERR(subdir) == -ENOMEM)
 732                         goto nomem_d_alloc;
 733                 goto lookup_error;
 734         }
 735
 736         _debug("subdir -> %p %s",
 737                subdir, subdir->d_inode ? "positive" : "negative");
 738
 739         /* we need to create the subdir if it doesn't exist yet */
 740         if (!subdir->d_inode) {
 741                 ret = cachefiles_has_space(cache, 1, 0);
 742                 if (ret < 0)
 743                         goto mkdir_error;
 744
 745                 _debug("attempt mkdir");
 746
 747                 path.mnt = cache->mnt;
 748                 path.dentry = dir;
 749                 ret = security_path_mkdir(&path, subdir, 0700);
 750                 if (ret < 0)
 751                         goto mkdir_error;
 752                 ret = vfs_mkdir(dir->d_inode, subdir, 0700);
 753                 if (ret < 0)
 754                         goto mkdir_error;
 755
 756                 ASSERT(subdir->d_inode);
 757
 758                 _debug("mkdir -> %p{%p{ino=%lu}}",
 759                        subdir,
 760                        subdir->d_inode,
 761                        subdir->d_inode->i_ino);
 762         }
 763
 764         mutex_unlock(&dir->d_inode->i_mutex);
 765
 766         /* we need to make sure the subdir is a directory */
 767         ASSERT(subdir->d_inode);
 768
 769         if (!S_ISDIR(subdir->d_inode->i_mode)) {
 770                 pr_err("%s is not a directory\n", dirname);
 771                 ret = -EIO;
 772                 goto check_error;
 773         }
 774
 775         ret = -EPERM;
 776         if (!subdir->d_inode->i_op->setxattr ||
 777             !subdir->d_inode->i_op->getxattr ||
 778             !subdir->d_inode->i_op->lookup ||
 779             !subdir->d_inode->i_op->mkdir ||
 780             !subdir->d_inode->i_op->create ||
 781             (!subdir->d_inode->i_op->rename &&
 782              !subdir->d_inode->i_op->rename2) ||
 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\n", 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\n", 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\n", 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\n", 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 }