fs/overlayfs/readdir.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/slab.h>
  12 #include <linux/namei.h>
  13 #include <linux/file.h>
  14 #include <linux/xattr.h>
  15 #include <linux/rbtree.h>
  16 #include <linux/security.h>
  17 #include <linux/cred.h>
  18 #include <linux/ratelimit.h>
  19 #include "overlayfs.h"
  20
  21 struct ovl_cache_entry {
  22         unsigned int len;
  23         unsigned int type;
  24         u64 real_ino;
  25         u64 ino;
  26         struct list_head l_node;
  27         struct rb_node node;
  28         struct ovl_cache_entry *next_maybe_whiteout;
  29         bool is_whiteout;
  30         char name[];
  31 };
  32
  33 struct ovl_dir_cache {
  34         long refcount;
  35         u64 version;
  36         struct list_head entries;
  37         struct rb_root root;
  38 };
  39
  40 struct ovl_readdir_data {
  41         struct dir_context ctx;
  42         struct dentry *dentry;
  43         bool is_lowest;
  44         struct rb_root *root;
  45         struct list_head *list;
  46         struct list_head middle;
  47         struct ovl_cache_entry *first_maybe_whiteout;
  48         int count;
  49         int err;
  50         bool is_upper;
  51         bool d_type_supported;
  52 };
  53
  54 struct ovl_dir_file {
  55         bool is_real;
  56         bool is_upper;
  57         struct ovl_dir_cache *cache;
  58         struct list_head *cursor;
  59         struct file *realfile;
  60         struct file *upperfile;
  61 };
  62
  63 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
  64 {
  65         return rb_entry(n, struct ovl_cache_entry, node);
  66 }
  67
  68 static bool ovl_cache_entry_find_link(const char *name, int len,
  69                                       struct rb_node ***link,
  70                                       struct rb_node **parent)
  71 {
  72         bool found = false;
  73         struct rb_node **newp = *link;
  74
  75         while (!found && *newp) {
  76                 int cmp;
  77                 struct ovl_cache_entry *tmp;
  78
  79                 *parent = *newp;
  80                 tmp = ovl_cache_entry_from_node(*newp);
  81                 cmp = strncmp(name, tmp->name, len);
  82                 if (cmp > 0)
  83                         newp = &tmp->node.rb_right;
  84                 else if (cmp < 0 || len < tmp->len)
  85                         newp = &tmp->node.rb_left;
  86                 else
  87                         found = true;
  88         }
  89         *link = newp;
  90
  91         return found;
  92 }
  93
  94 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
  95                                                     const char *name, int len)
  96 {
  97         struct rb_node *node = root->rb_node;
  98         int cmp;
  99
 100         while (node) {
 101                 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
 102
 103                 cmp = strncmp(name, p->name, len);
 104                 if (cmp > 0)
 105                         node = p->node.rb_right;
 106                 else if (cmp < 0 || len < p->len)
 107                         node = p->node.rb_left;
 108                 else
 109                         return p;
 110         }
 111
 112         return NULL;
 113 }
 114
 115 static bool ovl_calc_d_ino(struct ovl_readdir_data *rdd,
 116                            struct ovl_cache_entry *p)
 117 {
 118         /* Don't care if not doing ovl_iter() */
 119         if (!rdd->dentry)
 120                 return false;
 121
 122         /* Always recalc d_ino for parent */
 123         if (strcmp(p->name, "..") == 0)
 124                 return true;
 125
 126         /* If this is lower, then native d_ino will do */
 127         if (!rdd->is_upper)
 128                 return false;
 129
 130         /*
 131          * Recalc d_ino for '.' and for all entries if dir is impure (contains
 132          * copied up entries)
 133          */
 134         if ((p->name[0] == '.' && p->len == 1) ||
 135             ovl_test_flag(OVL_IMPURE, d_inode(rdd->dentry)))
 136                 return true;
 137
 138         return false;
 139 }
 140
 141 static struct ovl_cache_entry *ovl_cache_entry_new(struct ovl_readdir_data *rdd,
 142                                                    const char *name, int len,
 143                                                    u64 ino, unsigned int d_type)
 144 {
 145         struct ovl_cache_entry *p;
 146         size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
 147
 148         p = kmalloc(size, GFP_KERNEL);
 149         if (!p)
 150                 return NULL;
 151
 152         memcpy(p->name, name, len);
 153         p->name[len] = '\0';
 154         p->len = len;
 155         p->type = d_type;
 156         p->real_ino = ino;
 157         p->ino = ino;
 158         /* Defer setting d_ino for upper entry to ovl_iterate() */
 159         if (ovl_calc_d_ino(rdd, p))
 160                 p->ino = 0;
 161         p->is_whiteout = false;
 162
 163         if (d_type == DT_CHR) {
 164                 p->next_maybe_whiteout = rdd->first_maybe_whiteout;
 165                 rdd->first_maybe_whiteout = p;
 166         }
 167         return p;
 168 }
 169
 170 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
 171                                   const char *name, int len, u64 ino,
 172                                   unsigned int d_type)
 173 {
 174         struct rb_node **newp = &rdd->root->rb_node;
 175         struct rb_node *parent = NULL;
 176         struct ovl_cache_entry *p;
 177
 178         if (ovl_cache_entry_find_link(name, len, &newp, &parent))
 179                 return 0;
 180
 181         p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
 182         if (p == NULL) {
 183                 rdd->err = -ENOMEM;
 184                 return -ENOMEM;
 185         }
 186
 187         list_add_tail(&p->l_node, rdd->list);
 188         rb_link_node(&p->node, parent, newp);
 189         rb_insert_color(&p->node, rdd->root);
 190
 191         return 0;
 192 }
 193
 194 static int ovl_fill_lowest(struct ovl_readdir_data *rdd,
 195                            const char *name, int namelen,
 196                            loff_t offset, u64 ino, unsigned int d_type)
 197 {
 198         struct ovl_cache_entry *p;
 199
 200         p = ovl_cache_entry_find(rdd->root, name, namelen);
 201         if (p) {
 202                 list_move_tail(&p->l_node, &rdd->middle);
 203         } else {
 204                 p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 205                 if (p == NULL)
 206                         rdd->err = -ENOMEM;
 207                 else
 208                         list_add_tail(&p->l_node, &rdd->middle);
 209         }
 210
 211         return rdd->err;
 212 }
 213
 214 void ovl_cache_free(struct list_head *list)
 215 {
 216         struct ovl_cache_entry *p;
 217         struct ovl_cache_entry *n;
 218
 219         list_for_each_entry_safe(p, n, list, l_node)
 220                 kfree(p);
 221
 222         INIT_LIST_HEAD(list);
 223 }
 224
 225 void ovl_dir_cache_free(struct inode *inode)
 226 {
 227         struct ovl_dir_cache *cache = ovl_dir_cache(inode);
 228
 229         if (cache) {
 230                 ovl_cache_free(&cache->entries);
 231                 kfree(cache);
 232         }
 233 }
 234
 235 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
 236 {
 237         struct ovl_dir_cache *cache = od->cache;
 238
 239         WARN_ON(cache->refcount <= 0);
 240         cache->refcount--;
 241         if (!cache->refcount) {
 242                 if (ovl_dir_cache(d_inode(dentry)) == cache)
 243                         ovl_set_dir_cache(d_inode(dentry), NULL);
 244
 245                 ovl_cache_free(&cache->entries);
 246                 kfree(cache);
 247         }
 248 }
 249
 250 static int ovl_fill_merge(struct dir_context *ctx, const char *name,
 251                           int namelen, loff_t offset, u64 ino,
 252                           unsigned int d_type)
 253 {
 254         struct ovl_readdir_data *rdd =
 255                 container_of(ctx, struct ovl_readdir_data, ctx);
 256
 257         rdd->count++;
 258         if (!rdd->is_lowest)
 259                 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
 260         else
 261                 return ovl_fill_lowest(rdd, name, namelen, offset, ino, d_type);
 262 }
 263
 264 static int ovl_check_whiteouts(struct dentry *dir, struct ovl_readdir_data *rdd)
 265 {
 266         int err;
 267         struct ovl_cache_entry *p;
 268         struct dentry *dentry;
 269         const struct cred *old_cred;
 270
 271         old_cred = ovl_override_creds(rdd->dentry->d_sb);
 272
 273         err = down_write_killable(&dir->d_inode->i_rwsem);
 274         if (!err) {
 275                 while (rdd->first_maybe_whiteout) {
 276                         p = rdd->first_maybe_whiteout;
 277                         rdd->first_maybe_whiteout = p->next_maybe_whiteout;
 278                         dentry = lookup_one_len(p->name, dir, p->len);
 279                         if (!IS_ERR(dentry)) {
 280                                 p->is_whiteout = ovl_is_whiteout(dentry);
 281                                 dput(dentry);
 282                         }
 283                 }
 284                 inode_unlock(dir->d_inode);
 285         }
 286         revert_creds(old_cred);
 287
 288         return err;
 289 }
 290
 291 static inline int ovl_dir_read(struct path *realpath,
 292                                struct ovl_readdir_data *rdd)
 293 {
 294         struct file *realfile;
 295         int err;
 296
 297         realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
 298         if (IS_ERR(realfile))
 299                 return PTR_ERR(realfile);
 300
 301         rdd->first_maybe_whiteout = NULL;
 302         rdd->ctx.pos = 0;
 303         do {
 304                 rdd->count = 0;
 305                 rdd->err = 0;
 306                 err = iterate_dir(realfile, &rdd->ctx);
 307                 if (err >= 0)
 308                         err = rdd->err;
 309         } while (!err && rdd->count);
 310
 311         if (!err && rdd->first_maybe_whiteout && rdd->dentry)
 312                 err = ovl_check_whiteouts(realpath->dentry, rdd);
 313
 314         fput(realfile);
 315
 316         return err;
 317 }
 318
 319 static void ovl_dir_reset(struct file *file)
 320 {
 321         struct ovl_dir_file *od = file->private_data;
 322         struct ovl_dir_cache *cache = od->cache;
 323         struct dentry *dentry = file->f_path.dentry;
 324         enum ovl_path_type type = ovl_path_type(dentry);
 325
 326         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 327                 ovl_cache_put(od, dentry);
 328                 od->cache = NULL;
 329                 od->cursor = NULL;
 330         }
 331         WARN_ON(!od->is_real && !OVL_TYPE_MERGE(type));
 332         if (od->is_real && OVL_TYPE_MERGE(type))
 333                 od->is_real = false;
 334 }
 335
 336 static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list,
 337         struct rb_root *root)
 338 {
 339         int err;
 340         struct path realpath;
 341         struct ovl_readdir_data rdd = {
 342                 .ctx.actor = ovl_fill_merge,
 343                 .dentry = dentry,
 344                 .list = list,
 345                 .root = root,
 346                 .is_lowest = false,
 347         };
 348         int idx, next;
 349
 350         for (idx = 0; idx != -1; idx = next) {
 351                 next = ovl_path_next(idx, dentry, &realpath);
 352                 rdd.is_upper = ovl_dentry_upper(dentry) == realpath.dentry;
 353
 354                 if (next != -1) {
 355                         err = ovl_dir_read(&realpath, &rdd);
 356                         if (err)
 357                                 break;
 358                 } else {
 359                         /*
 360                          * Insert lowest layer entries before upper ones, this
 361                          * allows offsets to be reasonably constant
 362                          */
 363                         list_add(&rdd.middle, rdd.list);
 364                         rdd.is_lowest = true;
 365                         err = ovl_dir_read(&realpath, &rdd);
 366                         list_del(&rdd.middle);
 367                 }
 368         }
 369         return err;
 370 }
 371
 372 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 373 {
 374         struct list_head *p;
 375         loff_t off = 0;
 376
 377         list_for_each(p, &od->cache->entries) {
 378                 if (off >= pos)
 379                         break;
 380                 off++;
 381         }
 382         /* Cursor is safe since the cache is stable */
 383         od->cursor = p;
 384 }
 385
 386 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 387 {
 388         int res;
 389         struct ovl_dir_cache *cache;
 390
 391         cache = ovl_dir_cache(d_inode(dentry));
 392         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 393                 WARN_ON(!cache->refcount);
 394                 cache->refcount++;
 395                 return cache;
 396         }
 397         ovl_set_dir_cache(d_inode(dentry), NULL);
 398
 399         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 400         if (!cache)
 401                 return ERR_PTR(-ENOMEM);
 402
 403         cache->refcount = 1;
 404         INIT_LIST_HEAD(&cache->entries);
 405         cache->root = RB_ROOT;
 406
 407         res = ovl_dir_read_merged(dentry, &cache->entries, &cache->root);
 408         if (res) {
 409                 ovl_cache_free(&cache->entries);
 410                 kfree(cache);
 411                 return ERR_PTR(res);
 412         }
 413
 414         cache->version = ovl_dentry_version_get(dentry);
 415         ovl_set_dir_cache(d_inode(dentry), cache);
 416
 417         return cache;
 418 }
 419
 420 /*
 421  * Set d_ino for upper entries. Non-upper entries should always report
 422  * the uppermost real inode ino and should not call this function.
 423  *
 424  * When not all layer are on same fs, report real ino also for upper.
 425  *
 426  * When all layers are on the same fs, and upper has a reference to
 427  * copy up origin, call vfs_getattr() on the overlay entry to make
 428  * sure that d_ino will be consistent with st_ino from stat(2).
 429  */
 430 static int ovl_cache_update_ino(struct path *path, struct ovl_cache_entry *p)
 431
 432 {
 433         struct dentry *dir = path->dentry;
 434         struct dentry *this = NULL;
 435         enum ovl_path_type type;
 436         u64 ino = p->real_ino;
 437         int err = 0;
 438
 439         if (!ovl_same_sb(dir->d_sb))
 440                 goto out;
 441
 442         if (p->name[0] == '.') {
 443                 if (p->len == 1) {
 444                         this = dget(dir);
 445                         goto get;
 446                 }
 447                 if (p->len == 2 && p->name[1] == '.') {
 448                         /* we shall not be moved */
 449                         this = dget(dir->d_parent);
 450                         goto get;
 451                 }
 452         }
 453         this = lookup_one_len(p->name, dir, p->len);
 454         if (IS_ERR_OR_NULL(this) || !this->d_inode) {
 455                 if (IS_ERR(this)) {
 456                         err = PTR_ERR(this);
 457                         this = NULL;
 458                         goto fail;
 459                 }
 460                 goto out;
 461         }
 462
 463 get:
 464         type = ovl_path_type(this);
 465         if (OVL_TYPE_ORIGIN(type)) {
 466                 struct kstat stat;
 467                 struct path statpath = *path;
 468
 469                 statpath.dentry = this;
 470                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 471                 if (err)
 472                         goto fail;
 473
 474                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 475                 ino = stat.ino;
 476         }
 477
 478 out:
 479         p->ino = ino;
 480         dput(this);
 481         return err;
 482
 483 fail:
 484         pr_warn_ratelimited("overlay: failed to look up (%s) for ino (%i)\n",
 485                             p->name, err);
 486         goto out;
 487 }
 488
 489 static int ovl_fill_plain(struct dir_context *ctx, const char *name,
 490                           int namelen, loff_t offset, u64 ino,
 491                           unsigned int d_type)
 492 {
 493         struct ovl_cache_entry *p;
 494         struct ovl_readdir_data *rdd =
 495                 container_of(ctx, struct ovl_readdir_data, ctx);
 496
 497         rdd->count++;
 498         p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 499         if (p == NULL) {
 500                 rdd->err = -ENOMEM;
 501                 return -ENOMEM;
 502         }
 503         list_add_tail(&p->l_node, rdd->list);
 504
 505         return 0;
 506 }
 507
 508 static int ovl_dir_read_impure(struct path *path,  struct list_head *list,
 509                                struct rb_root *root)
 510 {
 511         int err;
 512         struct path realpath;
 513         struct ovl_cache_entry *p, *n;
 514         struct ovl_readdir_data rdd = {
 515                 .ctx.actor = ovl_fill_plain,
 516                 .list = list,
 517                 .root = root,
 518         };
 519
 520         INIT_LIST_HEAD(list);
 521         *root = RB_ROOT;
 522         ovl_path_upper(path->dentry, &realpath);
 523
 524         err = ovl_dir_read(&realpath, &rdd);
 525         if (err)
 526                 return err;
 527
 528         list_for_each_entry_safe(p, n, list, l_node) {
 529                 if (strcmp(p->name, ".") != 0 &&
 530                     strcmp(p->name, "..") != 0) {
 531                         err = ovl_cache_update_ino(path, p);
 532                         if (err)
 533                                 return err;
 534                 }
 535                 if (p->ino == p->real_ino) {
 536                         list_del(&p->l_node);
 537                         kfree(p);
 538                 } else {
 539                         struct rb_node **newp = &root->rb_node;
 540                         struct rb_node *parent = NULL;
 541
 542                         if (WARN_ON(ovl_cache_entry_find_link(p->name, p->len,
 543                                                               &newp, &parent)))
 544                                 return -EIO;
 545
 546                         rb_link_node(&p->node, parent, newp);
 547                         rb_insert_color(&p->node, root);
 548                 }
 549         }
 550         return 0;
 551 }
 552
 553 static struct ovl_dir_cache *ovl_cache_get_impure(struct path *path)
 554 {
 555         int res;
 556         struct dentry *dentry = path->dentry;
 557         struct ovl_dir_cache *cache;
 558
 559         cache = ovl_dir_cache(d_inode(dentry));
 560         if (cache && ovl_dentry_version_get(dentry) == cache->version)
 561                 return cache;
 562
 563         /* Impure cache is not refcounted, free it here */
 564         ovl_dir_cache_free(d_inode(dentry));
 565         ovl_set_dir_cache(d_inode(dentry), NULL);
 566
 567         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 568         if (!cache)
 569                 return ERR_PTR(-ENOMEM);
 570
 571         res = ovl_dir_read_impure(path, &cache->entries, &cache->root);
 572         if (res) {
 573                 ovl_cache_free(&cache->entries);
 574                 kfree(cache);
 575                 return ERR_PTR(res);
 576         }
 577         if (list_empty(&cache->entries)) {
 578                 /*
 579                  * A good opportunity to get rid of an unneeded "impure" flag.
 580                  * Removing the "impure" xattr is best effort.
 581                  */
 582                 if (!ovl_want_write(dentry)) {
 583                         ovl_do_removexattr(ovl_dentry_upper(dentry),
 584                                            OVL_XATTR_IMPURE);
 585                         ovl_drop_write(dentry);
 586                 }
 587                 ovl_clear_flag(OVL_IMPURE, d_inode(dentry));
 588                 kfree(cache);
 589                 return NULL;
 590         }
 591
 592         cache->version = ovl_dentry_version_get(dentry);
 593         ovl_set_dir_cache(d_inode(dentry), cache);
 594
 595         return cache;
 596 }
 597
 598 struct ovl_readdir_translate {
 599         struct dir_context *orig_ctx;
 600         struct ovl_dir_cache *cache;
 601         struct dir_context ctx;
 602         u64 parent_ino;
 603 };
 604
 605 static int ovl_fill_real(struct dir_context *ctx, const char *name,
 606                            int namelen, loff_t offset, u64 ino,
 607                            unsigned int d_type)
 608 {
 609         struct ovl_readdir_translate *rdt =
 610                 container_of(ctx, struct ovl_readdir_translate, ctx);
 611         struct dir_context *orig_ctx = rdt->orig_ctx;
 612
 613         if (rdt->parent_ino && strcmp(name, "..") == 0)
 614                 ino = rdt->parent_ino;
 615         else if (rdt->cache) {
 616                 struct ovl_cache_entry *p;
 617
 618                 p = ovl_cache_entry_find(&rdt->cache->root, name, namelen);
 619                 if (p)
 620                         ino = p->ino;
 621         }
 622
 623         return orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
 624 }
 625
 626 static bool ovl_is_impure_dir(struct file *file)
 627 {
 628         struct ovl_dir_file *od = file->private_data;
 629         struct inode *dir = d_inode(file->f_path.dentry);
 630
 631         /*
 632          * Only upper dir can be impure, but if we are in the middle of
 633          * iterating a lower real dir, dir could be copied up and marked
 634          * impure. We only want the impure cache if we started iterating
 635          * a real upper dir to begin with.
 636          */
 637         return od->is_upper && ovl_test_flag(OVL_IMPURE, dir);
 638
 639 }
 640
 641 static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
 642 {
 643         int err;
 644         struct ovl_dir_file *od = file->private_data;
 645         struct dentry *dir = file->f_path.dentry;
 646         struct ovl_readdir_translate rdt = {
 647                 .ctx.actor = ovl_fill_real,
 648                 .orig_ctx = ctx,
 649         };
 650
 651         if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
 652                 struct kstat stat;
 653                 struct path statpath = file->f_path;
 654
 655                 statpath.dentry = dir->d_parent;
 656                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 657                 if (err)
 658                         return err;
 659
 660                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 661                 rdt.parent_ino = stat.ino;
 662         }
 663
 664         if (ovl_is_impure_dir(file)) {
 665                 rdt.cache = ovl_cache_get_impure(&file->f_path);
 666                 if (IS_ERR(rdt.cache))
 667                         return PTR_ERR(rdt.cache);
 668         }
 669
 670         err = iterate_dir(od->realfile, &rdt.ctx);
 671         ctx->pos = rdt.ctx.pos;
 672
 673         return err;
 674 }
 675
 676
 677 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 678 {
 679         struct ovl_dir_file *od = file->private_data;
 680         struct dentry *dentry = file->f_path.dentry;
 681         struct ovl_cache_entry *p;
 682         int err;
 683
 684         if (!ctx->pos)
 685                 ovl_dir_reset(file);
 686
 687         if (od->is_real) {
 688                 /*
 689                  * If parent is merge, then need to adjust d_ino for '..', if
 690                  * dir is impure then need to adjust d_ino for copied up
 691                  * entries.
 692                  */
 693                 if (ovl_same_sb(dentry->d_sb) &&
 694                     (ovl_is_impure_dir(file) ||
 695                      OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent)))) {
 696                         return ovl_iterate_real(file, ctx);
 697                 }
 698                 return iterate_dir(od->realfile, ctx);
 699         }
 700
 701         if (!od->cache) {
 702                 struct ovl_dir_cache *cache;
 703
 704                 cache = ovl_cache_get(dentry);
 705                 if (IS_ERR(cache))
 706                         return PTR_ERR(cache);
 707
 708                 od->cache = cache;
 709                 ovl_seek_cursor(od, ctx->pos);
 710         }
 711
 712         while (od->cursor != &od->cache->entries) {
 713                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 714                 if (!p->is_whiteout) {
 715                         if (!p->ino) {
 716                                 err = ovl_cache_update_ino(&file->f_path, p);
 717                                 if (err)
 718                                         return err;
 719                         }
 720                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 721                                 break;
 722                 }
 723                 od->cursor = p->l_node.next;
 724                 ctx->pos++;
 725         }
 726         return 0;
 727 }
 728
 729 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 730 {
 731         loff_t res;
 732         struct ovl_dir_file *od = file->private_data;
 733
 734         inode_lock(file_inode(file));
 735         if (!file->f_pos)
 736                 ovl_dir_reset(file);
 737
 738         if (od->is_real) {
 739                 res = vfs_llseek(od->realfile, offset, origin);
 740                 file->f_pos = od->realfile->f_pos;
 741         } else {
 742                 res = -EINVAL;
 743
 744                 switch (origin) {
 745                 case SEEK_CUR:
 746                         offset += file->f_pos;
 747                         break;
 748                 case SEEK_SET:
 749                         break;
 750                 default:
 751                         goto out_unlock;
 752                 }
 753                 if (offset < 0)
 754                         goto out_unlock;
 755
 756                 if (offset != file->f_pos) {
 757                         file->f_pos = offset;
 758                         if (od->cache)
 759                                 ovl_seek_cursor(od, offset);
 760                 }
 761                 res = offset;
 762         }
 763 out_unlock:
 764         inode_unlock(file_inode(file));
 765
 766         return res;
 767 }
 768
 769 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 770                          int datasync)
 771 {
 772         struct ovl_dir_file *od = file->private_data;
 773         struct dentry *dentry = file->f_path.dentry;
 774         struct file *realfile = od->realfile;
 775
 776         /* Nothing to sync for lower */
 777         if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
 778                 return 0;
 779
 780         /*
 781          * Need to check if we started out being a lower dir, but got copied up
 782          */
 783         if (!od->is_upper) {
 784                 struct inode *inode = file_inode(file);
 785
 786                 realfile = READ_ONCE(od->upperfile);
 787                 if (!realfile) {
 788                         struct path upperpath;
 789
 790                         ovl_path_upper(dentry, &upperpath);
 791                         realfile = ovl_path_open(&upperpath, O_RDONLY);
 792
 793                         inode_lock(inode);
 794                         if (!od->upperfile) {
 795                                 if (IS_ERR(realfile)) {
 796                                         inode_unlock(inode);
 797                                         return PTR_ERR(realfile);
 798                                 }
 799                                 smp_store_release(&od->upperfile, realfile);
 800                         } else {
 801                                 /* somebody has beaten us to it */
 802                                 if (!IS_ERR(realfile))
 803                                         fput(realfile);
 804                                 realfile = od->upperfile;
 805                         }
 806                         inode_unlock(inode);
 807                 }
 808         }
 809
 810         return vfs_fsync_range(realfile, start, end, datasync);
 811 }
 812
 813 static int ovl_dir_release(struct inode *inode, struct file *file)
 814 {
 815         struct ovl_dir_file *od = file->private_data;
 816
 817         if (od->cache) {
 818                 inode_lock(inode);
 819                 ovl_cache_put(od, file->f_path.dentry);
 820                 inode_unlock(inode);
 821         }
 822         fput(od->realfile);
 823         if (od->upperfile)
 824                 fput(od->upperfile);
 825         kfree(od);
 826
 827         return 0;
 828 }
 829
 830 static int ovl_dir_open(struct inode *inode, struct file *file)
 831 {
 832         struct path realpath;
 833         struct file *realfile;
 834         struct ovl_dir_file *od;
 835         enum ovl_path_type type;
 836
 837         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 838         if (!od)
 839                 return -ENOMEM;
 840
 841         type = ovl_path_real(file->f_path.dentry, &realpath);
 842         realfile = ovl_path_open(&realpath, file->f_flags);
 843         if (IS_ERR(realfile)) {
 844                 kfree(od);
 845                 return PTR_ERR(realfile);
 846         }
 847         od->realfile = realfile;
 848         od->is_real = !OVL_TYPE_MERGE(type);
 849         od->is_upper = OVL_TYPE_UPPER(type);
 850         file->private_data = od;
 851
 852         return 0;
 853 }
 854
 855 const struct file_operations ovl_dir_operations = {
 856         .read           = generic_read_dir,
 857         .open           = ovl_dir_open,
 858         .iterate        = ovl_iterate,
 859         .llseek         = ovl_dir_llseek,
 860         .fsync          = ovl_dir_fsync,
 861         .release        = ovl_dir_release,
 862 };
 863
 864 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 865 {
 866         int err;
 867         struct ovl_cache_entry *p;
 868         struct rb_root root = RB_ROOT;
 869
 870         err = ovl_dir_read_merged(dentry, list, &root);
 871         if (err)
 872                 return err;
 873
 874         err = 0;
 875
 876         list_for_each_entry(p, list, l_node) {
 877                 if (p->is_whiteout)
 878                         continue;
 879
 880                 if (p->name[0] == '.') {
 881                         if (p->len == 1)
 882                                 continue;
 883                         if (p->len == 2 && p->name[1] == '.')
 884                                 continue;
 885                 }
 886                 err = -ENOTEMPTY;
 887                 break;
 888         }
 889
 890         return err;
 891 }
 892
 893 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
 894 {
 895         struct ovl_cache_entry *p;
 896
 897         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
 898         list_for_each_entry(p, list, l_node) {
 899                 struct dentry *dentry;
 900
 901                 if (!p->is_whiteout)
 902                         continue;
 903
 904                 dentry = lookup_one_len(p->name, upper, p->len);
 905                 if (IS_ERR(dentry)) {
 906                         pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
 907                                upper->d_name.name, p->len, p->name,
 908                                (int) PTR_ERR(dentry));
 909                         continue;
 910                 }
 911                 if (dentry->d_inode)
 912                         ovl_cleanup(upper->d_inode, dentry);
 913                 dput(dentry);
 914         }
 915         inode_unlock(upper->d_inode);
 916 }
 917
 918 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
 919                           int namelen, loff_t offset, u64 ino,
 920                           unsigned int d_type)
 921 {
 922         struct ovl_readdir_data *rdd =
 923                 container_of(ctx, struct ovl_readdir_data, ctx);
 924
 925         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
 926         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
 927                 return 0;
 928
 929         if (d_type != DT_UNKNOWN)
 930                 rdd->d_type_supported = true;
 931
 932         return 0;
 933 }
 934
 935 /*
 936  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
 937  * if error is encountered.
 938  */
 939 int ovl_check_d_type_supported(struct path *realpath)
 940 {
 941         int err;
 942         struct ovl_readdir_data rdd = {
 943                 .ctx.actor = ovl_check_d_type,
 944                 .d_type_supported = false,
 945         };
 946
 947         err = ovl_dir_read(realpath, &rdd);
 948         if (err)
 949                 return err;
 950
 951         return rdd.d_type_supported;
 952 }
 953
 954 static void ovl_workdir_cleanup_recurse(struct path *path, int level)
 955 {
 956         int err;
 957         struct inode *dir = path->dentry->d_inode;
 958         LIST_HEAD(list);
 959         struct rb_root root = RB_ROOT;
 960         struct ovl_cache_entry *p;
 961         struct ovl_readdir_data rdd = {
 962                 .ctx.actor = ovl_fill_merge,
 963                 .dentry = NULL,
 964                 .list = &list,
 965                 .root = &root,
 966                 .is_lowest = false,
 967         };
 968
 969         err = ovl_dir_read(path, &rdd);
 970         if (err)
 971                 goto out;
 972
 973         inode_lock_nested(dir, I_MUTEX_PARENT);
 974         list_for_each_entry(p, &list, l_node) {
 975                 struct dentry *dentry;
 976
 977                 if (p->name[0] == '.') {
 978                         if (p->len == 1)
 979                                 continue;
 980                         if (p->len == 2 && p->name[1] == '.')
 981                                 continue;
 982                 }
 983                 dentry = lookup_one_len(p->name, path->dentry, p->len);
 984                 if (IS_ERR(dentry))
 985                         continue;
 986                 if (dentry->d_inode)
 987                         ovl_workdir_cleanup(dir, path->mnt, dentry, level);
 988                 dput(dentry);
 989         }
 990         inode_unlock(dir);
 991 out:
 992         ovl_cache_free(&list);
 993 }
 994
 995 void ovl_workdir_cleanup(struct inode *dir, struct vfsmount *mnt,
 996                          struct dentry *dentry, int level)
 997 {
 998         int err;
 999
1000         if (!d_is_dir(dentry) || level > 1) {
1001                 ovl_cleanup(dir, dentry);
1002                 return;
1003         }
1004
1005         err = ovl_do_rmdir(dir, dentry);
1006         if (err) {
1007                 struct path path = { .mnt = mnt, .dentry = dentry };
1008
1009                 inode_unlock(dir);
1010                 ovl_workdir_cleanup_recurse(&path, level + 1);
1011                 inode_lock_nested(dir, I_MUTEX_PARENT);
1012                 ovl_cleanup(dir, dentry);
1013         }
1014 }
1015
1016 int ovl_indexdir_cleanup(struct dentry *dentry, struct vfsmount *mnt,
1017                          struct path *lowerstack, unsigned int numlower)
1018 {
1019         int err;
1020         struct dentry *index = NULL;
1021         struct inode *dir = dentry->d_inode;
1022         struct path path = { .mnt = mnt, .dentry = dentry };
1023         LIST_HEAD(list);
1024         struct rb_root root = RB_ROOT;
1025         struct ovl_cache_entry *p;
1026         struct ovl_readdir_data rdd = {
1027                 .ctx.actor = ovl_fill_merge,
1028                 .dentry = NULL,
1029                 .list = &list,
1030                 .root = &root,
1031                 .is_lowest = false,
1032         };
1033
1034         err = ovl_dir_read(&path, &rdd);
1035         if (err)
1036                 goto out;
1037
1038         inode_lock_nested(dir, I_MUTEX_PARENT);
1039         list_for_each_entry(p, &list, l_node) {
1040                 if (p->name[0] == '.') {
1041                         if (p->len == 1)
1042                                 continue;
1043                         if (p->len == 2 && p->name[1] == '.')
1044                                 continue;
1045                 }
1046                 index = lookup_one_len(p->name, dentry, p->len);
1047                 if (IS_ERR(index)) {
1048                         err = PTR_ERR(index);
1049                         index = NULL;
1050                         break;
1051                 }
1052                 err = ovl_verify_index(index, lowerstack, numlower);
1053                 /* Cleanup stale and orphan index entries */
1054                 if (err && (err == -ESTALE || err == -ENOENT))
1055                         err = ovl_cleanup(dir, index);
1056                 if (err)
1057                         break;
1058
1059                 dput(index);
1060                 index = NULL;
1061         }
1062         dput(index);
1063         inode_unlock(dir);
1064 out:
1065         ovl_cache_free(&list);
1066         if (err)
1067                 pr_err("overlayfs: failed index dir cleanup (%i)\n", err);
1068         return err;
1069 }