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                                           enum fscache_why_object_killed why)
 102 {
 103         struct cachefiles_object *object;
 104         struct rb_node *p;
 105
 106         _enter(",'%pd'", dentry);
 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 {
 137                 if (why != FSCACHE_OBJECT_IS_STALE)
 138                         fscache_object_mark_killed(&object->fscache, why);
 139         }
 140
 141         write_unlock(&cache->active_lock);
 142         _leave(" [owner marked]");
 143 }
 144
 145 /*
 146  * record the fact that an object is now active
 147  */
 148 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
 149                                          struct cachefiles_object *object)
 150 {
 151         struct cachefiles_object *xobject;
 152         struct rb_node **_p, *_parent = NULL;
 153         struct dentry *dentry;
 154
 155         _enter(",%p", object);
 156
 157 try_again:
 158         write_lock(&cache->active_lock);
 159
 160         if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
 161                 pr_err("Error: Object already active\n");
 162                 cachefiles_printk_object(object, NULL);
 163                 BUG();
 164         }
 165
 166         dentry = object->dentry;
 167         _p = &cache->active_nodes.rb_node;
 168         while (*_p) {
 169                 _parent = *_p;
 170                 xobject = rb_entry(_parent,
 171                                    struct cachefiles_object, active_node);
 172
 173                 ASSERT(xobject != object);
 174
 175                 if (xobject->dentry > dentry)
 176                         _p = &(*_p)->rb_left;
 177                 else if (xobject->dentry < dentry)
 178                         _p = &(*_p)->rb_right;
 179                 else
 180                         goto wait_for_old_object;
 181         }
 182
 183         rb_link_node(&object->active_node, _parent, _p);
 184         rb_insert_color(&object->active_node, &cache->active_nodes);
 185
 186         write_unlock(&cache->active_lock);
 187         _leave(" = 0");
 188         return 0;
 189
 190         /* an old object from a previous incarnation is hogging the slot - we
 191          * need to wait for it to be destroyed */
 192 wait_for_old_object:
 193         if (fscache_object_is_live(&xobject->fscache)) {
 194                 pr_err("\n");
 195                 pr_err("Error: Unexpected object collision\n");
 196                 cachefiles_printk_object(object, xobject);
 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                                   enum fscache_why_object_killed why)
 271 {
 272         struct dentry *grave, *trap;
 273         struct path path, path_to_graveyard;
 274         char nbuffer[8 + 8 + 1];
 275         int ret;
 276
 277         _enter(",'%pd','%pd'", dir, rep);
 278
 279         _debug("remove %p from %p", rep, dir);
 280
 281         /* non-directories can just be unlinked */
 282         if (!d_is_dir(rep)) {
 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(d_inode(dir), rep, NULL);
 292
 293                         if (preemptive)
 294                                 cachefiles_mark_object_buried(cache, rep, why);
 295                 }
 296
 297                 mutex_unlock(&d_inode(dir)->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(&d_inode(dir)->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 (!d_can_lookup(cache->graveyard)) {
 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 (d_is_positive(grave)) {
 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(d_inode(dir), rep,
 393                                  d_inode(cache->graveyard), 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, why);
 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(d_backing_inode(object->dentry));
 421         ASSERT(object->dentry->d_parent);
 422
 423         dir = dget_parent(object->dentry);
 424
 425         mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
 426
 427         if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.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(&d_inode(dir)->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                                                      FSCACHE_OBJECT_WAS_RETIRED);
 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(&d_inode(dir)->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(d_backing_inode(parent->dentry));
 480
 481         if (!(d_is_dir(parent->dentry))) {
 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(&d_inode(dir)->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, d_backing_inode(next) ? "positive" : "negative");
 512
 513         if (!key)
 514                 object->new = !d_backing_inode(next);
 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 (d_is_negative(next))
 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 (d_is_negative(next)) {
 526                         ret = cachefiles_has_space(cache, 1, 0);
 527                         if (ret < 0)
 528                                 goto no_space_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(d_inode(dir), next, 0);
 536                         cachefiles_hist(cachefiles_mkdir_histogram, start);
 537                         if (ret < 0)
 538                                 goto create_error;
 539
 540                         ASSERT(d_backing_inode(next));
 541
 542                         _debug("mkdir -> %p{%p{ino=%lu}}",
 543                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 544
 545                 } else if (!d_can_lookup(next)) {
 546                         pr_err("inode %lu is not a directory\n",
 547                                d_backing_inode(next)->i_ino);
 548                         ret = -ENOBUFS;
 549                         goto error;
 550                 }
 551
 552         } else {
 553                 /* non-index objects start out life as files */
 554                 if (d_is_negative(next)) {
 555                         ret = cachefiles_has_space(cache, 1, 0);
 556                         if (ret < 0)
 557                                 goto no_space_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(d_inode(dir), next, S_IFREG, true);
 565                         cachefiles_hist(cachefiles_create_histogram, start);
 566                         if (ret < 0)
 567                                 goto create_error;
 568
 569                         ASSERT(d_backing_inode(next));
 570
 571                         _debug("create -> %p{%p{ino=%lu}}",
 572                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 573
 574                 } else if (!d_can_lookup(next) &&
 575                            !d_is_reg(next)
 576                            ) {
 577                         pr_err("inode %lu is not a file or directory\n",
 578                                d_backing_inode(next)->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(&d_inode(dir)->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 '%pd'", next);
 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                                                      FSCACHE_OBJECT_IS_STALE);
 610                         dput(next);
 611                         next = NULL;
 612
 613                         if (ret < 0)
 614                                 goto delete_error;
 615
 616                         _debug("redo lookup");
 617                         fscache_object_retrying_stale(&object->fscache);
 618                         goto lookup_again;
 619                 }
 620         }
 621
 622         /* note that we're now using this object */
 623         ret = cachefiles_mark_object_active(cache, object);
 624
 625         mutex_unlock(&d_inode(dir)->i_mutex);
 626         dput(dir);
 627         dir = NULL;
 628
 629         if (ret == -ETIMEDOUT)
 630                 goto mark_active_timed_out;
 631
 632         _debug("=== OBTAINED_OBJECT ===");
 633
 634         if (object->new) {
 635                 /* attach data to a newly constructed terminal object */
 636                 ret = cachefiles_set_object_xattr(object, auxdata);
 637                 if (ret < 0)
 638                         goto check_error;
 639         } else {
 640                 /* always update the atime on an object we've just looked up
 641                  * (this is used to keep track of culling, and atimes are only
 642                  * updated by read, write and readdir but not lookup or
 643                  * open) */
 644                 path.dentry = next;
 645                 touch_atime(&path);
 646         }
 647
 648         /* open a file interface onto a data file */
 649         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
 650                 if (d_is_reg(object->dentry)) {
 651                         const struct address_space_operations *aops;
 652
 653                         ret = -EPERM;
 654                         aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
 655                         if (!aops->bmap)
 656                                 goto check_error;
 657                         if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
 658                                 goto check_error;
 659
 660                         object->backer = object->dentry;
 661                 } else {
 662                         BUG(); // TODO: open file in data-class subdir
 663                 }
 664         }
 665
 666         object->new = 0;
 667         fscache_obtained_object(&object->fscache);
 668
 669         _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
 670         return 0;
 671
 672 no_space_error:
 673         fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
 674 create_error:
 675         _debug("create error %d", ret);
 676         if (ret == -EIO)
 677                 cachefiles_io_error(cache, "Create/mkdir failed");
 678         goto error;
 679
 680 mark_active_timed_out:
 681         _debug("mark active timed out");
 682         goto release_dentry;
 683
 684 check_error:
 685         _debug("check error %d", ret);
 686         write_lock(&cache->active_lock);
 687         rb_erase(&object->active_node, &cache->active_nodes);
 688         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 689         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
 690         write_unlock(&cache->active_lock);
 691 release_dentry:
 692         dput(object->dentry);
 693         object->dentry = NULL;
 694         goto error_out;
 695
 696 delete_error:
 697         _debug("delete error %d", ret);
 698         goto error_out2;
 699
 700 lookup_error:
 701         _debug("lookup error %ld", PTR_ERR(next));
 702         ret = PTR_ERR(next);
 703         if (ret == -EIO)
 704                 cachefiles_io_error(cache, "Lookup failed");
 705         next = NULL;
 706 error:
 707         mutex_unlock(&d_inode(dir)->i_mutex);
 708         dput(next);
 709 error_out2:
 710         dput(dir);
 711 error_out:
 712         _leave(" = error %d", -ret);
 713         return ret;
 714 }
 715
 716 /*
 717  * get a subdirectory
 718  */
 719 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 720                                         struct dentry *dir,
 721                                         const char *dirname)
 722 {
 723         struct dentry *subdir;
 724         unsigned long start;
 725         struct path path;
 726         int ret;
 727
 728         _enter(",,%s", dirname);
 729
 730         /* search the current directory for the element name */
 731         mutex_lock(&d_inode(dir)->i_mutex);
 732
 733         start = jiffies;
 734         subdir = lookup_one_len(dirname, dir, strlen(dirname));
 735         cachefiles_hist(cachefiles_lookup_histogram, start);
 736         if (IS_ERR(subdir)) {
 737                 if (PTR_ERR(subdir) == -ENOMEM)
 738                         goto nomem_d_alloc;
 739                 goto lookup_error;
 740         }
 741
 742         _debug("subdir -> %p %s",
 743                subdir, d_backing_inode(subdir) ? "positive" : "negative");
 744
 745         /* we need to create the subdir if it doesn't exist yet */
 746         if (d_is_negative(subdir)) {
 747                 ret = cachefiles_has_space(cache, 1, 0);
 748                 if (ret < 0)
 749                         goto mkdir_error;
 750
 751                 _debug("attempt mkdir");
 752
 753                 path.mnt = cache->mnt;
 754                 path.dentry = dir;
 755                 ret = security_path_mkdir(&path, subdir, 0700);
 756                 if (ret < 0)
 757                         goto mkdir_error;
 758                 ret = vfs_mkdir(d_inode(dir), subdir, 0700);
 759                 if (ret < 0)
 760                         goto mkdir_error;
 761
 762                 ASSERT(d_backing_inode(subdir));
 763
 764                 _debug("mkdir -> %p{%p{ino=%lu}}",
 765                        subdir,
 766                        d_backing_inode(subdir),
 767                        d_backing_inode(subdir)->i_ino);
 768         }
 769
 770         mutex_unlock(&d_inode(dir)->i_mutex);
 771
 772         /* we need to make sure the subdir is a directory */
 773         ASSERT(d_backing_inode(subdir));
 774
 775         if (!d_can_lookup(subdir)) {
 776                 pr_err("%s is not a directory\n", dirname);
 777                 ret = -EIO;
 778                 goto check_error;
 779         }
 780
 781         ret = -EPERM;
 782         if (!d_backing_inode(subdir)->i_op->setxattr ||
 783             !d_backing_inode(subdir)->i_op->getxattr ||
 784             !d_backing_inode(subdir)->i_op->lookup ||
 785             !d_backing_inode(subdir)->i_op->mkdir ||
 786             !d_backing_inode(subdir)->i_op->create ||
 787             (!d_backing_inode(subdir)->i_op->rename &&
 788              !d_backing_inode(subdir)->i_op->rename2) ||
 789             !d_backing_inode(subdir)->i_op->rmdir ||
 790             !d_backing_inode(subdir)->i_op->unlink)
 791                 goto check_error;
 792
 793         _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
 794         return subdir;
 795
 796 check_error:
 797         dput(subdir);
 798         _leave(" = %d [check]", ret);
 799         return ERR_PTR(ret);
 800
 801 mkdir_error:
 802         mutex_unlock(&d_inode(dir)->i_mutex);
 803         dput(subdir);
 804         pr_err("mkdir %s failed with error %d\n", dirname, ret);
 805         return ERR_PTR(ret);
 806
 807 lookup_error:
 808         mutex_unlock(&d_inode(dir)->i_mutex);
 809         ret = PTR_ERR(subdir);
 810         pr_err("Lookup %s failed with error %d\n", dirname, ret);
 811         return ERR_PTR(ret);
 812
 813 nomem_d_alloc:
 814         mutex_unlock(&d_inode(dir)->i_mutex);
 815         _leave(" = -ENOMEM");
 816         return ERR_PTR(-ENOMEM);
 817 }
 818
 819 /*
 820  * find out if an object is in use or not
 821  * - if finds object and it's not in use:
 822  *   - returns a pointer to the object and a reference on it
 823  *   - returns with the directory locked
 824  */
 825 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
 826                                               struct dentry *dir,
 827                                               char *filename)
 828 {
 829         struct cachefiles_object *object;
 830         struct rb_node *_n;
 831         struct dentry *victim;
 832         unsigned long start;
 833         int ret;
 834
 835         //_enter(",%pd/,%s",
 836         //       dir, filename);
 837
 838         /* look up the victim */
 839         mutex_lock_nested(&d_inode(dir)->i_mutex, I_MUTEX_PARENT);
 840
 841         start = jiffies;
 842         victim = lookup_one_len(filename, dir, strlen(filename));
 843         cachefiles_hist(cachefiles_lookup_histogram, start);
 844         if (IS_ERR(victim))
 845                 goto lookup_error;
 846
 847         //_debug("victim -> %p %s",
 848         //       victim, d_backing_inode(victim) ? "positive" : "negative");
 849
 850         /* if the object is no longer there then we probably retired the object
 851          * at the netfs's request whilst the cull was in progress
 852          */
 853         if (d_is_negative(victim)) {
 854                 mutex_unlock(&d_inode(dir)->i_mutex);
 855                 dput(victim);
 856                 _leave(" = -ENOENT [absent]");
 857                 return ERR_PTR(-ENOENT);
 858         }
 859
 860         /* check to see if we're using this object */
 861         read_lock(&cache->active_lock);
 862
 863         _n = cache->active_nodes.rb_node;
 864
 865         while (_n) {
 866                 object = rb_entry(_n, struct cachefiles_object, active_node);
 867
 868                 if (object->dentry > victim)
 869                         _n = _n->rb_left;
 870                 else if (object->dentry < victim)
 871                         _n = _n->rb_right;
 872                 else
 873                         goto object_in_use;
 874         }
 875
 876         read_unlock(&cache->active_lock);
 877
 878         //_leave(" = %p", victim);
 879         return victim;
 880
 881 object_in_use:
 882         read_unlock(&cache->active_lock);
 883         mutex_unlock(&d_inode(dir)->i_mutex);
 884         dput(victim);
 885         //_leave(" = -EBUSY [in use]");
 886         return ERR_PTR(-EBUSY);
 887
 888 lookup_error:
 889         mutex_unlock(&d_inode(dir)->i_mutex);
 890         ret = PTR_ERR(victim);
 891         if (ret == -ENOENT) {
 892                 /* file or dir now absent - probably retired by netfs */
 893                 _leave(" = -ESTALE [absent]");
 894                 return ERR_PTR(-ESTALE);
 895         }
 896
 897         if (ret == -EIO) {
 898                 cachefiles_io_error(cache, "Lookup failed");
 899         } else if (ret != -ENOMEM) {
 900                 pr_err("Internal error: %d\n", ret);
 901                 ret = -EIO;
 902         }
 903
 904         _leave(" = %d", ret);
 905         return ERR_PTR(ret);
 906 }
 907
 908 /*
 909  * cull an object if it's not in use
 910  * - called only by cache manager daemon
 911  */
 912 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
 913                     char *filename)
 914 {
 915         struct dentry *victim;
 916         int ret;
 917
 918         _enter(",%pd/,%s", dir, filename);
 919
 920         victim = cachefiles_check_active(cache, dir, filename);
 921         if (IS_ERR(victim))
 922                 return PTR_ERR(victim);
 923
 924         _debug("victim -> %p %s",
 925                victim, d_backing_inode(victim) ? "positive" : "negative");
 926
 927         /* okay... the victim is not being used so we can cull it
 928          * - start by marking it as stale
 929          */
 930         _debug("victim is cullable");
 931
 932         ret = cachefiles_remove_object_xattr(cache, victim);
 933         if (ret < 0)
 934                 goto error_unlock;
 935
 936         /*  actually remove the victim (drops the dir mutex) */
 937         _debug("bury");
 938
 939         ret = cachefiles_bury_object(cache, dir, victim, false,
 940                                      FSCACHE_OBJECT_WAS_CULLED);
 941         if (ret < 0)
 942                 goto error;
 943
 944         dput(victim);
 945         _leave(" = 0");
 946         return 0;
 947
 948 error_unlock:
 949         mutex_unlock(&d_inode(dir)->i_mutex);
 950 error:
 951         dput(victim);
 952         if (ret == -ENOENT) {
 953                 /* file or dir now absent - probably retired by netfs */
 954                 _leave(" = -ESTALE [absent]");
 955                 return -ESTALE;
 956         }
 957
 958         if (ret != -ENOMEM) {
 959                 pr_err("Internal error: %d\n", ret);
 960                 ret = -EIO;
 961         }
 962
 963         _leave(" = %d", ret);
 964         return ret;
 965 }
 966
 967 /*
 968  * find out if an object is in use or not
 969  * - called only by cache manager daemon
 970  * - returns -EBUSY or 0 to indicate whether an object is in use or not
 971  */
 972 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
 973                             char *filename)
 974 {
 975         struct dentry *victim;
 976
 977         //_enter(",%pd/,%s",
 978         //       dir, filename);
 979
 980         victim = cachefiles_check_active(cache, dir, filename);
 981         if (IS_ERR(victim))
 982                 return PTR_ERR(victim);
 983
 984         mutex_unlock(&d_inode(dir)->i_mutex);
 985         dput(victim);
 986         //_leave(" = 0");
 987         return 0;
 988 }