fs/cachefiles/namei.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* CacheFiles path walking and related routines
   3  *
   4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells (dhowells@redhat.com)
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/sched.h>
  10 #include <linux/file.h>
  11 #include <linux/fs.h>
  12 #include <linux/fsnotify.h>
  13 #include <linux/quotaops.h>
  14 #include <linux/xattr.h>
  15 #include <linux/mount.h>
  16 #include <linux/namei.h>
  17 #include <linux/security.h>
  18 #include <linux/slab.h>
  19 #include "internal.h"
  20
  21 #define CACHEFILES_KEYBUF_SIZE 512
  22
  23 /*
  24  * dump debugging info about an object
  25  */
  26 static noinline
  27 void __cachefiles_printk_object(struct cachefiles_object *object,
  28                                 const char *prefix)
  29 {
  30         struct fscache_cookie *cookie;
  31         const u8 *k;
  32         unsigned loop;
  33
  34         pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
  35         pr_err("%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
  36                prefix, object->fscache.state->name,
  37                object->fscache.flags, work_busy(&object->fscache.work),
  38                object->fscache.events, object->fscache.event_mask);
  39         pr_err("%sops=%u inp=%u exc=%u\n",
  40                prefix, object->fscache.n_ops, object->fscache.n_in_progress,
  41                object->fscache.n_exclusive);
  42         pr_err("%sparent=%p\n",
  43                prefix, object->fscache.parent);
  44
  45         spin_lock(&object->fscache.lock);
  46         cookie = object->fscache.cookie;
  47         if (cookie) {
  48                 pr_err("%scookie=%p [pr=%p nd=%p fl=%lx]\n",
  49                        prefix,
  50                        object->fscache.cookie,
  51                        object->fscache.cookie->parent,
  52                        object->fscache.cookie->netfs_data,
  53                        object->fscache.cookie->flags);
  54                 pr_err("%skey=[%u] '", prefix, cookie->key_len);
  55                 k = (cookie->key_len <= sizeof(cookie->inline_key)) ?
  56                         cookie->inline_key : cookie->key;
  57                 for (loop = 0; loop < cookie->key_len; loop++)
  58                         pr_cont("%02x", k[loop]);
  59                 pr_cont("'\n");
  60         } else {
  61                 pr_err("%scookie=NULL\n", prefix);
  62         }
  63         spin_unlock(&object->fscache.lock);
  64 }
  65
  66 /*
  67  * dump debugging info about a pair of objects
  68  */
  69 static noinline void cachefiles_printk_object(struct cachefiles_object *object,
  70                                               struct cachefiles_object *xobject)
  71 {
  72         if (object)
  73                 __cachefiles_printk_object(object, "");
  74         if (xobject)
  75                 __cachefiles_printk_object(xobject, "x");
  76 }
  77
  78 /*
  79  * mark the owner of a dentry, if there is one, to indicate that that dentry
  80  * has been preemptively deleted
  81  * - the caller must hold the i_mutex on the dentry's parent as required to
  82  *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
  83  */
  84 static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
  85                                           struct dentry *dentry,
  86                                           enum fscache_why_object_killed why)
  87 {
  88         struct cachefiles_object *object;
  89         struct rb_node *p;
  90
  91         _enter(",'%pd'", dentry);
  92
  93         write_lock(&cache->active_lock);
  94
  95         p = cache->active_nodes.rb_node;
  96         while (p) {
  97                 object = rb_entry(p, struct cachefiles_object, active_node);
  98                 if (object->dentry > dentry)
  99                         p = p->rb_left;
 100                 else if (object->dentry < dentry)
 101                         p = p->rb_right;
 102                 else
 103                         goto found_dentry;
 104         }
 105
 106         write_unlock(&cache->active_lock);
 107         trace_cachefiles_mark_buried(NULL, dentry, why);
 108         _leave(" [no owner]");
 109         return;
 110
 111         /* found the dentry for  */
 112 found_dentry:
 113         kdebug("preemptive burial: OBJ%x [%s] %p",
 114                object->fscache.debug_id,
 115                object->fscache.state->name,
 116                dentry);
 117
 118         trace_cachefiles_mark_buried(object, dentry, why);
 119
 120         if (fscache_object_is_live(&object->fscache)) {
 121                 pr_err("\n");
 122                 pr_err("Error: Can't preemptively bury live object\n");
 123                 cachefiles_printk_object(object, NULL);
 124         } else {
 125                 if (why != FSCACHE_OBJECT_IS_STALE)
 126                         fscache_object_mark_killed(&object->fscache, why);
 127         }
 128
 129         write_unlock(&cache->active_lock);
 130         _leave(" [owner marked]");
 131 }
 132
 133 /*
 134  * record the fact that an object is now active
 135  */
 136 static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
 137                                          struct cachefiles_object *object)
 138 {
 139         struct cachefiles_object *xobject;
 140         struct rb_node **_p, *_parent = NULL;
 141         struct dentry *dentry;
 142
 143         _enter(",%p", object);
 144
 145 try_again:
 146         write_lock(&cache->active_lock);
 147
 148         dentry = object->dentry;
 149         trace_cachefiles_mark_active(object, dentry);
 150
 151         if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
 152                 pr_err("Error: Object already active\n");
 153                 cachefiles_printk_object(object, NULL);
 154                 BUG();
 155         }
 156
 157         _p = &cache->active_nodes.rb_node;
 158         while (*_p) {
 159                 _parent = *_p;
 160                 xobject = rb_entry(_parent,
 161                                    struct cachefiles_object, active_node);
 162
 163                 ASSERT(xobject != object);
 164
 165                 if (xobject->dentry > dentry)
 166                         _p = &(*_p)->rb_left;
 167                 else if (xobject->dentry < dentry)
 168                         _p = &(*_p)->rb_right;
 169                 else
 170                         goto wait_for_old_object;
 171         }
 172
 173         rb_link_node(&object->active_node, _parent, _p);
 174         rb_insert_color(&object->active_node, &cache->active_nodes);
 175
 176         write_unlock(&cache->active_lock);
 177         _leave(" = 0");
 178         return 0;
 179
 180         /* an old object from a previous incarnation is hogging the slot - we
 181          * need to wait for it to be destroyed */
 182 wait_for_old_object:
 183         trace_cachefiles_wait_active(object, dentry, xobject);
 184         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 185
 186         if (fscache_object_is_live(&xobject->fscache)) {
 187                 pr_err("\n");
 188                 pr_err("Error: Unexpected object collision\n");
 189                 cachefiles_printk_object(object, xobject);
 190         }
 191         atomic_inc(&xobject->usage);
 192         write_unlock(&cache->active_lock);
 193
 194         if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 195                 wait_queue_head_t *wq;
 196
 197                 signed long timeout = 60 * HZ;
 198                 wait_queue_entry_t wait;
 199                 bool requeue;
 200
 201                 /* if the object we're waiting for is queued for processing,
 202                  * then just put ourselves on the queue behind it */
 203                 if (work_pending(&xobject->fscache.work)) {
 204                         _debug("queue OBJ%x behind OBJ%x immediately",
 205                                object->fscache.debug_id,
 206                                xobject->fscache.debug_id);
 207                         goto requeue;
 208                 }
 209
 210                 /* otherwise we sleep until either the object we're waiting for
 211                  * is done, or the fscache_object is congested */
 212                 wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
 213                 init_wait(&wait);
 214                 requeue = false;
 215                 do {
 216                         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 217                         if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
 218                                 break;
 219
 220                         requeue = fscache_object_sleep_till_congested(&timeout);
 221                 } while (timeout > 0 && !requeue);
 222                 finish_wait(wq, &wait);
 223
 224                 if (requeue &&
 225                     test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 226                         _debug("queue OBJ%x behind OBJ%x after wait",
 227                                object->fscache.debug_id,
 228                                xobject->fscache.debug_id);
 229                         goto requeue;
 230                 }
 231
 232                 if (timeout <= 0) {
 233                         pr_err("\n");
 234                         pr_err("Error: Overlong wait for old active object to go away\n");
 235                         cachefiles_printk_object(object, xobject);
 236                         goto requeue;
 237                 }
 238         }
 239
 240         ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
 241
 242         cache->cache.ops->put_object(&xobject->fscache,
 243                 (enum fscache_obj_ref_trace)cachefiles_obj_put_wait_retry);
 244         goto try_again;
 245
 246 requeue:
 247         cache->cache.ops->put_object(&xobject->fscache,
 248                 (enum fscache_obj_ref_trace)cachefiles_obj_put_wait_timeo);
 249         _leave(" = -ETIMEDOUT");
 250         return -ETIMEDOUT;
 251 }
 252
 253 /*
 254  * Mark an object as being inactive.
 255  */
 256 void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
 257                                      struct cachefiles_object *object,
 258                                      blkcnt_t i_blocks)
 259 {
 260         struct dentry *dentry = object->dentry;
 261         struct inode *inode = d_backing_inode(dentry);
 262
 263         trace_cachefiles_mark_inactive(object, dentry, inode);
 264
 265         write_lock(&cache->active_lock);
 266         rb_erase(&object->active_node, &cache->active_nodes);
 267         clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 268         write_unlock(&cache->active_lock);
 269
 270         wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
 271
 272         /* This object can now be culled, so we need to let the daemon know
 273          * that there is something it can remove if it needs to.
 274          */
 275         atomic_long_add(i_blocks, &cache->b_released);
 276         if (atomic_inc_return(&cache->f_released))
 277                 cachefiles_state_changed(cache);
 278 }
 279
 280 /*
 281  * delete an object representation from the cache
 282  * - file backed objects are unlinked
 283  * - directory backed objects are stuffed into the graveyard for userspace to
 284  *   delete
 285  * - unlocks the directory mutex
 286  */
 287 static int cachefiles_bury_object(struct cachefiles_cache *cache,
 288                                   struct cachefiles_object *object,
 289                                   struct dentry *dir,
 290                                   struct dentry *rep,
 291                                   bool preemptive,
 292                                   enum fscache_why_object_killed why)
 293 {
 294         struct dentry *grave, *trap;
 295         struct path path, path_to_graveyard;
 296         char nbuffer[8 + 8 + 1];
 297         int ret;
 298
 299         _enter(",'%pd','%pd'", dir, rep);
 300
 301         _debug("remove %p from %p", rep, dir);
 302
 303         /* non-directories can just be unlinked */
 304         if (!d_is_dir(rep)) {
 305                 _debug("unlink stale object");
 306
 307                 path.mnt = cache->mnt;
 308                 path.dentry = dir;
 309                 ret = security_path_unlink(&path, rep);
 310                 if (ret < 0) {
 311                         cachefiles_io_error(cache, "Unlink security error");
 312                 } else {
 313                         trace_cachefiles_unlink(object, rep, why);
 314                         ret = vfs_unlink(&init_user_ns, d_inode(dir), rep,
 315                                          NULL);
 316
 317                         if (preemptive)
 318                                 cachefiles_mark_object_buried(cache, rep, why);
 319                 }
 320
 321                 inode_unlock(d_inode(dir));
 322
 323                 if (ret == -EIO)
 324                         cachefiles_io_error(cache, "Unlink failed");
 325
 326                 _leave(" = %d", ret);
 327                 return ret;
 328         }
 329
 330         /* directories have to be moved to the graveyard */
 331         _debug("move stale object to graveyard");
 332         inode_unlock(d_inode(dir));
 333
 334 try_again:
 335         /* first step is to make up a grave dentry in the graveyard */
 336         sprintf(nbuffer, "%08x%08x",
 337                 (uint32_t) ktime_get_real_seconds(),
 338                 (uint32_t) atomic_inc_return(&cache->gravecounter));
 339
 340         /* do the multiway lock magic */
 341         trap = lock_rename(cache->graveyard, dir);
 342
 343         /* do some checks before getting the grave dentry */
 344         if (rep->d_parent != dir || IS_DEADDIR(d_inode(rep))) {
 345                 /* the entry was probably culled when we dropped the parent dir
 346                  * lock */
 347                 unlock_rename(cache->graveyard, dir);
 348                 _leave(" = 0 [culled?]");
 349                 return 0;
 350         }
 351
 352         if (!d_can_lookup(cache->graveyard)) {
 353                 unlock_rename(cache->graveyard, dir);
 354                 cachefiles_io_error(cache, "Graveyard no longer a directory");
 355                 return -EIO;
 356         }
 357
 358         if (trap == rep) {
 359                 unlock_rename(cache->graveyard, dir);
 360                 cachefiles_io_error(cache, "May not make directory loop");
 361                 return -EIO;
 362         }
 363
 364         if (d_mountpoint(rep)) {
 365                 unlock_rename(cache->graveyard, dir);
 366                 cachefiles_io_error(cache, "Mountpoint in cache");
 367                 return -EIO;
 368         }
 369
 370         grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
 371         if (IS_ERR(grave)) {
 372                 unlock_rename(cache->graveyard, dir);
 373
 374                 if (PTR_ERR(grave) == -ENOMEM) {
 375                         _leave(" = -ENOMEM");
 376                         return -ENOMEM;
 377                 }
 378
 379                 cachefiles_io_error(cache, "Lookup error %ld",
 380                                     PTR_ERR(grave));
 381                 return -EIO;
 382         }
 383
 384         if (d_is_positive(grave)) {
 385                 unlock_rename(cache->graveyard, dir);
 386                 dput(grave);
 387                 grave = NULL;
 388                 cond_resched();
 389                 goto try_again;
 390         }
 391
 392         if (d_mountpoint(grave)) {
 393                 unlock_rename(cache->graveyard, dir);
 394                 dput(grave);
 395                 cachefiles_io_error(cache, "Mountpoint in graveyard");
 396                 return -EIO;
 397         }
 398
 399         /* target should not be an ancestor of source */
 400         if (trap == grave) {
 401                 unlock_rename(cache->graveyard, dir);
 402                 dput(grave);
 403                 cachefiles_io_error(cache, "May not make directory loop");
 404                 return -EIO;
 405         }
 406
 407         /* attempt the rename */
 408         path.mnt = cache->mnt;
 409         path.dentry = dir;
 410         path_to_graveyard.mnt = cache->mnt;
 411         path_to_graveyard.dentry = cache->graveyard;
 412         ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
 413         if (ret < 0) {
 414                 cachefiles_io_error(cache, "Rename security error %d", ret);
 415         } else {
 416                 struct renamedata rd = {
 417                         .old_mnt_userns = &init_user_ns,
 418                         .old_dir        = d_inode(dir),
 419                         .old_dentry     = rep,
 420                         .new_mnt_userns = &init_user_ns,
 421                         .new_dir        = d_inode(cache->graveyard),
 422                         .new_dentry     = grave,
 423                 };
 424                 trace_cachefiles_rename(object, rep, grave, why);
 425                 ret = vfs_rename(&rd);
 426                 if (ret != 0 && ret != -ENOMEM)
 427                         cachefiles_io_error(cache,
 428                                             "Rename failed with error %d", ret);
 429
 430                 if (preemptive)
 431                         cachefiles_mark_object_buried(cache, rep, why);
 432         }
 433
 434         unlock_rename(cache->graveyard, dir);
 435         dput(grave);
 436         _leave(" = 0");
 437         return 0;
 438 }
 439
 440 /*
 441  * delete an object representation from the cache
 442  */
 443 int cachefiles_delete_object(struct cachefiles_cache *cache,
 444                              struct cachefiles_object *object)
 445 {
 446         struct dentry *dir;
 447         int ret;
 448
 449         _enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
 450
 451         ASSERT(object->dentry);
 452         ASSERT(d_backing_inode(object->dentry));
 453         ASSERT(object->dentry->d_parent);
 454
 455         dir = dget_parent(object->dentry);
 456
 457         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 458
 459         if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
 460                 /* object allocation for the same key preemptively deleted this
 461                  * object's file so that it could create its own file */
 462                 _debug("object preemptively buried");
 463                 inode_unlock(d_inode(dir));
 464                 ret = 0;
 465         } else {
 466                 /* we need to check that our parent is _still_ our parent - it
 467                  * may have been renamed */
 468                 if (dir == object->dentry->d_parent) {
 469                         ret = cachefiles_bury_object(cache, object, dir,
 470                                                      object->dentry, false,
 471                                                      FSCACHE_OBJECT_WAS_RETIRED);
 472                 } else {
 473                         /* it got moved, presumably by cachefilesd culling it,
 474                          * so it's no longer in the key path and we can ignore
 475                          * it */
 476                         inode_unlock(d_inode(dir));
 477                         ret = 0;
 478                 }
 479         }
 480
 481         dput(dir);
 482         _leave(" = %d", ret);
 483         return ret;
 484 }
 485
 486 /*
 487  * walk from the parent object to the child object through the backing
 488  * filesystem, creating directories as we go
 489  */
 490 int cachefiles_walk_to_object(struct cachefiles_object *parent,
 491                               struct cachefiles_object *object,
 492                               const char *key,
 493                               struct cachefiles_xattr *auxdata)
 494 {
 495         struct cachefiles_cache *cache;
 496         struct dentry *dir, *next = NULL;
 497         struct inode *inode;
 498         struct path path;
 499         unsigned long start;
 500         const char *name;
 501         int ret, nlen;
 502
 503         _enter("OBJ%x{%p},OBJ%x,%s,",
 504                parent->fscache.debug_id, parent->dentry,
 505                object->fscache.debug_id, key);
 506
 507         cache = container_of(parent->fscache.cache,
 508                              struct cachefiles_cache, cache);
 509         path.mnt = cache->mnt;
 510
 511         ASSERT(parent->dentry);
 512         ASSERT(d_backing_inode(parent->dentry));
 513
 514         if (!(d_is_dir(parent->dentry))) {
 515                 // TODO: convert file to dir
 516                 _leave("looking up in none directory");
 517                 return -ENOBUFS;
 518         }
 519
 520         dir = dget(parent->dentry);
 521
 522 advance:
 523         /* attempt to transit the first directory component */
 524         name = key;
 525         nlen = strlen(key);
 526
 527         /* key ends in a double NUL */
 528         key = key + nlen + 1;
 529         if (!*key)
 530                 key = NULL;
 531
 532 lookup_again:
 533         /* search the current directory for the element name */
 534         _debug("lookup '%s'", name);
 535
 536         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 537
 538         start = jiffies;
 539         next = lookup_one_len(name, dir, nlen);
 540         cachefiles_hist(cachefiles_lookup_histogram, start);
 541         if (IS_ERR(next)) {
 542                 trace_cachefiles_lookup(object, next, NULL);
 543                 goto lookup_error;
 544         }
 545
 546         inode = d_backing_inode(next);
 547         trace_cachefiles_lookup(object, next, inode);
 548         _debug("next -> %p %s", next, inode ? "positive" : "negative");
 549
 550         if (!key)
 551                 object->new = !inode;
 552
 553         /* if this element of the path doesn't exist, then the lookup phase
 554          * failed, and we can release any readers in the certain knowledge that
 555          * there's nothing for them to actually read */
 556         if (d_is_negative(next))
 557                 fscache_object_lookup_negative(&object->fscache);
 558
 559         /* we need to create the object if it's negative */
 560         if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
 561                 /* index objects and intervening tree levels must be subdirs */
 562                 if (d_is_negative(next)) {
 563                         ret = cachefiles_has_space(cache, 1, 0);
 564                         if (ret < 0)
 565                                 goto no_space_error;
 566
 567                         path.dentry = dir;
 568                         ret = security_path_mkdir(&path, next, 0);
 569                         if (ret < 0)
 570                                 goto create_error;
 571                         start = jiffies;
 572                         ret = vfs_mkdir(&init_user_ns, d_inode(dir), next, 0);
 573                         cachefiles_hist(cachefiles_mkdir_histogram, start);
 574                         if (!key)
 575                                 trace_cachefiles_mkdir(object, next, ret);
 576                         if (ret < 0)
 577                                 goto create_error;
 578
 579                         if (unlikely(d_unhashed(next))) {
 580                                 dput(next);
 581                                 inode_unlock(d_inode(dir));
 582                                 goto lookup_again;
 583                         }
 584                         ASSERT(d_backing_inode(next));
 585
 586                         _debug("mkdir -> %p{%p{ino=%lu}}",
 587                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 588
 589                 } else if (!d_can_lookup(next)) {
 590                         pr_err("inode %lu is not a directory\n",
 591                                d_backing_inode(next)->i_ino);
 592                         ret = -ENOBUFS;
 593                         goto error;
 594                 }
 595
 596         } else {
 597                 /* non-index objects start out life as files */
 598                 if (d_is_negative(next)) {
 599                         ret = cachefiles_has_space(cache, 1, 0);
 600                         if (ret < 0)
 601                                 goto no_space_error;
 602
 603                         path.dentry = dir;
 604                         ret = security_path_mknod(&path, next, S_IFREG, 0);
 605                         if (ret < 0)
 606                                 goto create_error;
 607                         start = jiffies;
 608                         ret = vfs_create(&init_user_ns, d_inode(dir), next,
 609                                          S_IFREG, true);
 610                         cachefiles_hist(cachefiles_create_histogram, start);
 611                         trace_cachefiles_create(object, next, ret);
 612                         if (ret < 0)
 613                                 goto create_error;
 614
 615                         ASSERT(d_backing_inode(next));
 616
 617                         _debug("create -> %p{%p{ino=%lu}}",
 618                                next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 619
 620                 } else if (!d_can_lookup(next) &&
 621                            !d_is_reg(next)
 622                            ) {
 623                         pr_err("inode %lu is not a file or directory\n",
 624                                d_backing_inode(next)->i_ino);
 625                         ret = -ENOBUFS;
 626                         goto error;
 627                 }
 628         }
 629
 630         /* process the next component */
 631         if (key) {
 632                 _debug("advance");
 633                 inode_unlock(d_inode(dir));
 634                 dput(dir);
 635                 dir = next;
 636                 next = NULL;
 637                 goto advance;
 638         }
 639
 640         /* we've found the object we were looking for */
 641         object->dentry = next;
 642
 643         /* if we've found that the terminal object exists, then we need to
 644          * check its attributes and delete it if it's out of date */
 645         if (!object->new) {
 646                 _debug("validate '%pd'", next);
 647
 648                 ret = cachefiles_check_object_xattr(object, auxdata);
 649                 if (ret == -ESTALE) {
 650                         /* delete the object (the deleter drops the directory
 651                          * mutex) */
 652                         object->dentry = NULL;
 653
 654                         ret = cachefiles_bury_object(cache, object, dir, next,
 655                                                      true,
 656                                                      FSCACHE_OBJECT_IS_STALE);
 657                         dput(next);
 658                         next = NULL;
 659
 660                         if (ret < 0)
 661                                 goto delete_error;
 662
 663                         _debug("redo lookup");
 664                         fscache_object_retrying_stale(&object->fscache);
 665                         goto lookup_again;
 666                 }
 667         }
 668
 669         /* note that we're now using this object */
 670         ret = cachefiles_mark_object_active(cache, object);
 671
 672         inode_unlock(d_inode(dir));
 673         dput(dir);
 674         dir = NULL;
 675
 676         if (ret == -ETIMEDOUT)
 677                 goto mark_active_timed_out;
 678
 679         _debug("=== OBTAINED_OBJECT ===");
 680
 681         if (object->new) {
 682                 /* attach data to a newly constructed terminal object */
 683                 ret = cachefiles_set_object_xattr(object, auxdata);
 684                 if (ret < 0)
 685                         goto check_error;
 686         } else {
 687                 /* always update the atime on an object we've just looked up
 688                  * (this is used to keep track of culling, and atimes are only
 689                  * updated by read, write and readdir but not lookup or
 690                  * open) */
 691                 path.dentry = next;
 692                 touch_atime(&path);
 693         }
 694
 695         /* open a file interface onto a data file */
 696         if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
 697                 if (d_is_reg(object->dentry)) {
 698                         const struct address_space_operations *aops;
 699
 700                         ret = -EPERM;
 701                         aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
 702                         if (!aops->bmap)
 703                                 goto check_error;
 704                         if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
 705                                 goto check_error;
 706
 707                         object->backer = object->dentry;
 708                 } else {
 709                         BUG(); // TODO: open file in data-class subdir
 710                 }
 711         }
 712
 713         object->new = 0;
 714         fscache_obtained_object(&object->fscache);
 715
 716         _leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
 717         return 0;
 718
 719 no_space_error:
 720         fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
 721 create_error:
 722         _debug("create error %d", ret);
 723         if (ret == -EIO)
 724                 cachefiles_io_error(cache, "Create/mkdir failed");
 725         goto error;
 726
 727 mark_active_timed_out:
 728         _debug("mark active timed out");
 729         goto release_dentry;
 730
 731 check_error:
 732         _debug("check error %d", ret);
 733         cachefiles_mark_object_inactive(
 734                 cache, object, d_backing_inode(object->dentry)->i_blocks);
 735 release_dentry:
 736         dput(object->dentry);
 737         object->dentry = NULL;
 738         goto error_out;
 739
 740 delete_error:
 741         _debug("delete error %d", ret);
 742         goto error_out2;
 743
 744 lookup_error:
 745         _debug("lookup error %ld", PTR_ERR(next));
 746         ret = PTR_ERR(next);
 747         if (ret == -EIO)
 748                 cachefiles_io_error(cache, "Lookup failed");
 749         next = NULL;
 750 error:
 751         inode_unlock(d_inode(dir));
 752         dput(next);
 753 error_out2:
 754         dput(dir);
 755 error_out:
 756         _leave(" = error %d", -ret);
 757         return ret;
 758 }
 759
 760 /*
 761  * get a subdirectory
 762  */
 763 struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 764                                         struct dentry *dir,
 765                                         const char *dirname)
 766 {
 767         struct dentry *subdir;
 768         unsigned long start;
 769         struct path path;
 770         int ret;
 771
 772         _enter(",,%s", dirname);
 773
 774         /* search the current directory for the element name */
 775         inode_lock(d_inode(dir));
 776
 777 retry:
 778         start = jiffies;
 779         subdir = lookup_one_len(dirname, dir, strlen(dirname));
 780         cachefiles_hist(cachefiles_lookup_histogram, start);
 781         if (IS_ERR(subdir)) {
 782                 if (PTR_ERR(subdir) == -ENOMEM)
 783                         goto nomem_d_alloc;
 784                 goto lookup_error;
 785         }
 786
 787         _debug("subdir -> %p %s",
 788                subdir, d_backing_inode(subdir) ? "positive" : "negative");
 789
 790         /* we need to create the subdir if it doesn't exist yet */
 791         if (d_is_negative(subdir)) {
 792                 ret = cachefiles_has_space(cache, 1, 0);
 793                 if (ret < 0)
 794                         goto mkdir_error;
 795
 796                 _debug("attempt mkdir");
 797
 798                 path.mnt = cache->mnt;
 799                 path.dentry = dir;
 800                 ret = security_path_mkdir(&path, subdir, 0700);
 801                 if (ret < 0)
 802                         goto mkdir_error;
 803                 ret = vfs_mkdir(&init_user_ns, d_inode(dir), subdir, 0700);
 804                 if (ret < 0)
 805                         goto mkdir_error;
 806
 807                 if (unlikely(d_unhashed(subdir))) {
 808                         dput(subdir);
 809                         goto retry;
 810                 }
 811                 ASSERT(d_backing_inode(subdir));
 812
 813                 _debug("mkdir -> %p{%p{ino=%lu}}",
 814                        subdir,
 815                        d_backing_inode(subdir),
 816                        d_backing_inode(subdir)->i_ino);
 817         }
 818
 819         inode_unlock(d_inode(dir));
 820
 821         /* we need to make sure the subdir is a directory */
 822         ASSERT(d_backing_inode(subdir));
 823
 824         if (!d_can_lookup(subdir)) {
 825                 pr_err("%s is not a directory\n", dirname);
 826                 ret = -EIO;
 827                 goto check_error;
 828         }
 829
 830         ret = -EPERM;
 831         if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
 832             !d_backing_inode(subdir)->i_op->lookup ||
 833             !d_backing_inode(subdir)->i_op->mkdir ||
 834             !d_backing_inode(subdir)->i_op->create ||
 835             !d_backing_inode(subdir)->i_op->rename ||
 836             !d_backing_inode(subdir)->i_op->rmdir ||
 837             !d_backing_inode(subdir)->i_op->unlink)
 838                 goto check_error;
 839
 840         _leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
 841         return subdir;
 842
 843 check_error:
 844         dput(subdir);
 845         _leave(" = %d [check]", ret);
 846         return ERR_PTR(ret);
 847
 848 mkdir_error:
 849         inode_unlock(d_inode(dir));
 850         dput(subdir);
 851         pr_err("mkdir %s failed with error %d\n", dirname, ret);
 852         return ERR_PTR(ret);
 853
 854 lookup_error:
 855         inode_unlock(d_inode(dir));
 856         ret = PTR_ERR(subdir);
 857         pr_err("Lookup %s failed with error %d\n", dirname, ret);
 858         return ERR_PTR(ret);
 859
 860 nomem_d_alloc:
 861         inode_unlock(d_inode(dir));
 862         _leave(" = -ENOMEM");
 863         return ERR_PTR(-ENOMEM);
 864 }
 865
 866 /*
 867  * find out if an object is in use or not
 868  * - if finds object and it's not in use:
 869  *   - returns a pointer to the object and a reference on it
 870  *   - returns with the directory locked
 871  */
 872 static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
 873                                               struct dentry *dir,
 874                                               char *filename)
 875 {
 876         struct cachefiles_object *object;
 877         struct rb_node *_n;
 878         struct dentry *victim;
 879         unsigned long start;
 880         int ret;
 881
 882         //_enter(",%pd/,%s",
 883         //       dir, filename);
 884
 885         /* look up the victim */
 886         inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 887
 888         start = jiffies;
 889         victim = lookup_one_len(filename, dir, strlen(filename));
 890         cachefiles_hist(cachefiles_lookup_histogram, start);
 891         if (IS_ERR(victim))
 892                 goto lookup_error;
 893
 894         //_debug("victim -> %p %s",
 895         //       victim, d_backing_inode(victim) ? "positive" : "negative");
 896
 897         /* if the object is no longer there then we probably retired the object
 898          * at the netfs's request whilst the cull was in progress
 899          */
 900         if (d_is_negative(victim)) {
 901                 inode_unlock(d_inode(dir));
 902                 dput(victim);
 903                 _leave(" = -ENOENT [absent]");
 904                 return ERR_PTR(-ENOENT);
 905         }
 906
 907         /* check to see if we're using this object */
 908         read_lock(&cache->active_lock);
 909
 910         _n = cache->active_nodes.rb_node;
 911
 912         while (_n) {
 913                 object = rb_entry(_n, struct cachefiles_object, active_node);
 914
 915                 if (object->dentry > victim)
 916                         _n = _n->rb_left;
 917                 else if (object->dentry < victim)
 918                         _n = _n->rb_right;
 919                 else
 920                         goto object_in_use;
 921         }
 922
 923         read_unlock(&cache->active_lock);
 924
 925         //_leave(" = %p", victim);
 926         return victim;
 927
 928 object_in_use:
 929         read_unlock(&cache->active_lock);
 930         inode_unlock(d_inode(dir));
 931         dput(victim);
 932         //_leave(" = -EBUSY [in use]");
 933         return ERR_PTR(-EBUSY);
 934
 935 lookup_error:
 936         inode_unlock(d_inode(dir));
 937         ret = PTR_ERR(victim);
 938         if (ret == -ENOENT) {
 939                 /* file or dir now absent - probably retired by netfs */
 940                 _leave(" = -ESTALE [absent]");
 941                 return ERR_PTR(-ESTALE);
 942         }
 943
 944         if (ret == -EIO) {
 945                 cachefiles_io_error(cache, "Lookup failed");
 946         } else if (ret != -ENOMEM) {
 947                 pr_err("Internal error: %d\n", ret);
 948                 ret = -EIO;
 949         }
 950
 951         _leave(" = %d", ret);
 952         return ERR_PTR(ret);
 953 }
 954
 955 /*
 956  * cull an object if it's not in use
 957  * - called only by cache manager daemon
 958  */
 959 int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
 960                     char *filename)
 961 {
 962         struct dentry *victim;
 963         int ret;
 964
 965         _enter(",%pd/,%s", dir, filename);
 966
 967         victim = cachefiles_check_active(cache, dir, filename);
 968         if (IS_ERR(victim))
 969                 return PTR_ERR(victim);
 970
 971         _debug("victim -> %p %s",
 972                victim, d_backing_inode(victim) ? "positive" : "negative");
 973
 974         /* okay... the victim is not being used so we can cull it
 975          * - start by marking it as stale
 976          */
 977         _debug("victim is cullable");
 978
 979         ret = cachefiles_remove_object_xattr(cache, victim);
 980         if (ret < 0)
 981                 goto error_unlock;
 982
 983         /*  actually remove the victim (drops the dir mutex) */
 984         _debug("bury");
 985
 986         ret = cachefiles_bury_object(cache, NULL, dir, victim, false,
 987                                      FSCACHE_OBJECT_WAS_CULLED);
 988         if (ret < 0)
 989                 goto error;
 990
 991         dput(victim);
 992         _leave(" = 0");
 993         return 0;
 994
 995 error_unlock:
 996         inode_unlock(d_inode(dir));
 997 error:
 998         dput(victim);
 999         if (ret == -ENOENT) {
1000                 /* file or dir now absent - probably retired by netfs */
1001                 _leave(" = -ESTALE [absent]");
1002                 return -ESTALE;
1003         }
1004
1005         if (ret != -ENOMEM) {
1006                 pr_err("Internal error: %d\n", ret);
1007                 ret = -EIO;
1008         }
1009
1010         _leave(" = %d", ret);
1011         return ret;
1012 }
1013
1014 /*
1015  * find out if an object is in use or not
1016  * - called only by cache manager daemon
1017  * - returns -EBUSY or 0 to indicate whether an object is in use or not
1018  */
1019 int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
1020                             char *filename)
1021 {
1022         struct dentry *victim;
1023
1024         //_enter(",%pd/,%s",
1025         //       dir, filename);
1026
1027         victim = cachefiles_check_active(cache, dir, filename);
1028         if (IS_ERR(victim))
1029                 return PTR_ERR(victim);
1030
1031         inode_unlock(d_inode(dir));
1032         dput(victim);
1033         //_leave(" = 0");
1034         return 0;
1035 }