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_upper;
  30         bool is_whiteout;
  31         char name[];
  32 };
  33
  34 struct ovl_dir_cache {
  35         long refcount;
  36         u64 version;
  37         struct list_head entries;
  38         struct rb_root root;
  39 };
  40
  41 struct ovl_readdir_data {
  42         struct dir_context ctx;
  43         struct dentry *dentry;
  44         bool is_lowest;
  45         struct rb_root *root;
  46         struct list_head *list;
  47         struct list_head middle;
  48         struct ovl_cache_entry *first_maybe_whiteout;
  49         int count;
  50         int err;
  51         bool is_upper;
  52         bool d_type_supported;
  53 };
  54
  55 struct ovl_dir_file {
  56         bool is_real;
  57         bool is_upper;
  58         struct ovl_dir_cache *cache;
  59         struct list_head *cursor;
  60         struct file *realfile;
  61         struct file *upperfile;
  62 };
  63
  64 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
  65 {
  66         return rb_entry(n, struct ovl_cache_entry, node);
  67 }
  68
  69 static bool ovl_cache_entry_find_link(const char *name, int len,
  70                                       struct rb_node ***link,
  71                                       struct rb_node **parent)
  72 {
  73         bool found = false;
  74         struct rb_node **newp = *link;
  75
  76         while (!found && *newp) {
  77                 int cmp;
  78                 struct ovl_cache_entry *tmp;
  79
  80                 *parent = *newp;
  81                 tmp = ovl_cache_entry_from_node(*newp);
  82                 cmp = strncmp(name, tmp->name, len);
  83                 if (cmp > 0)
  84                         newp = &tmp->node.rb_right;
  85                 else if (cmp < 0 || len < tmp->len)
  86                         newp = &tmp->node.rb_left;
  87                 else
  88                         found = true;
  89         }
  90         *link = newp;
  91
  92         return found;
  93 }
  94
  95 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
  96                                                     const char *name, int len)
  97 {
  98         struct rb_node *node = root->rb_node;
  99         int cmp;
 100
 101         while (node) {
 102                 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
 103
 104                 cmp = strncmp(name, p->name, len);
 105                 if (cmp > 0)
 106                         node = p->node.rb_right;
 107                 else if (cmp < 0 || len < p->len)
 108                         node = p->node.rb_left;
 109                 else
 110                         return p;
 111         }
 112
 113         return NULL;
 114 }
 115
 116 static bool ovl_calc_d_ino(struct ovl_readdir_data *rdd,
 117                            struct ovl_cache_entry *p)
 118 {
 119         /* Don't care if not doing ovl_iter() */
 120         if (!rdd->dentry)
 121                 return false;
 122
 123         /* Always recalc d_ino for parent */
 124         if (strcmp(p->name, "..") == 0)
 125                 return true;
 126
 127         /* If this is lower, then native d_ino will do */
 128         if (!rdd->is_upper)
 129                 return false;
 130
 131         /*
 132          * Recalc d_ino for '.' and for all entries if dir is impure (contains
 133          * copied up entries)
 134          */
 135         if ((p->name[0] == '.' && p->len == 1) ||
 136             ovl_test_flag(OVL_IMPURE, d_inode(rdd->dentry)))
 137                 return true;
 138
 139         return false;
 140 }
 141
 142 static struct ovl_cache_entry *ovl_cache_entry_new(struct ovl_readdir_data *rdd,
 143                                                    const char *name, int len,
 144                                                    u64 ino, unsigned int d_type)
 145 {
 146         struct ovl_cache_entry *p;
 147         size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
 148
 149         p = kmalloc(size, GFP_KERNEL);
 150         if (!p)
 151                 return NULL;
 152
 153         memcpy(p->name, name, len);
 154         p->name[len] = '\0';
 155         p->len = len;
 156         p->type = d_type;
 157         p->real_ino = ino;
 158         p->ino = ino;
 159         /* Defer setting d_ino for upper entry to ovl_iterate() */
 160         if (ovl_calc_d_ino(rdd, p))
 161                 p->ino = 0;
 162         p->is_upper = rdd->is_upper;
 163         p->is_whiteout = false;
 164
 165         if (d_type == DT_CHR) {
 166                 p->next_maybe_whiteout = rdd->first_maybe_whiteout;
 167                 rdd->first_maybe_whiteout = p;
 168         }
 169         return p;
 170 }
 171
 172 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
 173                                   const char *name, int len, u64 ino,
 174                                   unsigned int d_type)
 175 {
 176         struct rb_node **newp = &rdd->root->rb_node;
 177         struct rb_node *parent = NULL;
 178         struct ovl_cache_entry *p;
 179
 180         if (ovl_cache_entry_find_link(name, len, &newp, &parent))
 181                 return 0;
 182
 183         p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
 184         if (p == NULL) {
 185                 rdd->err = -ENOMEM;
 186                 return -ENOMEM;
 187         }
 188
 189         list_add_tail(&p->l_node, rdd->list);
 190         rb_link_node(&p->node, parent, newp);
 191         rb_insert_color(&p->node, rdd->root);
 192
 193         return 0;
 194 }
 195
 196 static int ovl_fill_lowest(struct ovl_readdir_data *rdd,
 197                            const char *name, int namelen,
 198                            loff_t offset, u64 ino, unsigned int d_type)
 199 {
 200         struct ovl_cache_entry *p;
 201
 202         p = ovl_cache_entry_find(rdd->root, name, namelen);
 203         if (p) {
 204                 list_move_tail(&p->l_node, &rdd->middle);
 205         } else {
 206                 p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 207                 if (p == NULL)
 208                         rdd->err = -ENOMEM;
 209                 else
 210                         list_add_tail(&p->l_node, &rdd->middle);
 211         }
 212
 213         return rdd->err;
 214 }
 215
 216 void ovl_cache_free(struct list_head *list)
 217 {
 218         struct ovl_cache_entry *p;
 219         struct ovl_cache_entry *n;
 220
 221         list_for_each_entry_safe(p, n, list, l_node)
 222                 kfree(p);
 223
 224         INIT_LIST_HEAD(list);
 225 }
 226
 227 void ovl_dir_cache_free(struct inode *inode)
 228 {
 229         struct ovl_dir_cache *cache = ovl_dir_cache(inode);
 230
 231         if (cache) {
 232                 ovl_cache_free(&cache->entries);
 233                 kfree(cache);
 234         }
 235 }
 236
 237 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
 238 {
 239         struct ovl_dir_cache *cache = od->cache;
 240
 241         WARN_ON(cache->refcount <= 0);
 242         cache->refcount--;
 243         if (!cache->refcount) {
 244                 if (ovl_dir_cache(d_inode(dentry)) == cache)
 245                         ovl_set_dir_cache(d_inode(dentry), NULL);
 246
 247                 ovl_cache_free(&cache->entries);
 248                 kfree(cache);
 249         }
 250 }
 251
 252 static int ovl_fill_merge(struct dir_context *ctx, const char *name,
 253                           int namelen, loff_t offset, u64 ino,
 254                           unsigned int d_type)
 255 {
 256         struct ovl_readdir_data *rdd =
 257                 container_of(ctx, struct ovl_readdir_data, ctx);
 258
 259         rdd->count++;
 260         if (!rdd->is_lowest)
 261                 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
 262         else
 263                 return ovl_fill_lowest(rdd, name, namelen, offset, ino, d_type);
 264 }
 265
 266 static int ovl_check_whiteouts(struct dentry *dir, struct ovl_readdir_data *rdd)
 267 {
 268         int err;
 269         struct ovl_cache_entry *p;
 270         struct dentry *dentry;
 271         const struct cred *old_cred;
 272
 273         old_cred = ovl_override_creds(rdd->dentry->d_sb);
 274
 275         err = down_write_killable(&dir->d_inode->i_rwsem);
 276         if (!err) {
 277                 while (rdd->first_maybe_whiteout) {
 278                         p = rdd->first_maybe_whiteout;
 279                         rdd->first_maybe_whiteout = p->next_maybe_whiteout;
 280                         dentry = lookup_one_len(p->name, dir, p->len);
 281                         if (!IS_ERR(dentry)) {
 282                                 p->is_whiteout = ovl_is_whiteout(dentry);
 283                                 dput(dentry);
 284                         }
 285                 }
 286                 inode_unlock(dir->d_inode);
 287         }
 288         revert_creds(old_cred);
 289
 290         return err;
 291 }
 292
 293 static inline int ovl_dir_read(struct path *realpath,
 294                                struct ovl_readdir_data *rdd)
 295 {
 296         struct file *realfile;
 297         int err;
 298
 299         realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
 300         if (IS_ERR(realfile))
 301                 return PTR_ERR(realfile);
 302
 303         rdd->first_maybe_whiteout = NULL;
 304         rdd->ctx.pos = 0;
 305         do {
 306                 rdd->count = 0;
 307                 rdd->err = 0;
 308                 err = iterate_dir(realfile, &rdd->ctx);
 309                 if (err >= 0)
 310                         err = rdd->err;
 311         } while (!err && rdd->count);
 312
 313         if (!err && rdd->first_maybe_whiteout && rdd->dentry)
 314                 err = ovl_check_whiteouts(realpath->dentry, rdd);
 315
 316         fput(realfile);
 317
 318         return err;
 319 }
 320
 321 /*
 322  * Can we iterate real dir directly?
 323  *
 324  * Non-merge dir may contain whiteouts from a time it was a merge upper, before
 325  * lower dir was removed under it and possibly before it was rotated from upper
 326  * to lower layer.
 327  */
 328 static bool ovl_dir_is_real(struct dentry *dir)
 329 {
 330         return !ovl_test_flag(OVL_WHITEOUTS, d_inode(dir));
 331 }
 332
 333 static void ovl_dir_reset(struct file *file)
 334 {
 335         struct ovl_dir_file *od = file->private_data;
 336         struct ovl_dir_cache *cache = od->cache;
 337         struct dentry *dentry = file->f_path.dentry;
 338         bool is_real;
 339
 340         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 341                 ovl_cache_put(od, dentry);
 342                 od->cache = NULL;
 343                 od->cursor = NULL;
 344         }
 345         is_real = ovl_dir_is_real(dentry);
 346         if (od->is_real != is_real) {
 347                 /* is_real can only become false when dir is copied up */
 348                 if (WARN_ON(is_real))
 349                         return;
 350                 od->is_real = false;
 351         }
 352 }
 353
 354 static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list,
 355         struct rb_root *root)
 356 {
 357         int err;
 358         struct path realpath;
 359         struct ovl_readdir_data rdd = {
 360                 .ctx.actor = ovl_fill_merge,
 361                 .dentry = dentry,
 362                 .list = list,
 363                 .root = root,
 364                 .is_lowest = false,
 365         };
 366         int idx, next;
 367
 368         for (idx = 0; idx != -1; idx = next) {
 369                 next = ovl_path_next(idx, dentry, &realpath);
 370                 rdd.is_upper = ovl_dentry_upper(dentry) == realpath.dentry;
 371
 372                 if (next != -1) {
 373                         err = ovl_dir_read(&realpath, &rdd);
 374                         if (err)
 375                                 break;
 376                 } else {
 377                         /*
 378                          * Insert lowest layer entries before upper ones, this
 379                          * allows offsets to be reasonably constant
 380                          */
 381                         list_add(&rdd.middle, rdd.list);
 382                         rdd.is_lowest = true;
 383                         err = ovl_dir_read(&realpath, &rdd);
 384                         list_del(&rdd.middle);
 385                 }
 386         }
 387         return err;
 388 }
 389
 390 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 391 {
 392         struct list_head *p;
 393         loff_t off = 0;
 394
 395         list_for_each(p, &od->cache->entries) {
 396                 if (off >= pos)
 397                         break;
 398                 off++;
 399         }
 400         /* Cursor is safe since the cache is stable */
 401         od->cursor = p;
 402 }
 403
 404 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 405 {
 406         int res;
 407         struct ovl_dir_cache *cache;
 408
 409         cache = ovl_dir_cache(d_inode(dentry));
 410         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 411                 WARN_ON(!cache->refcount);
 412                 cache->refcount++;
 413                 return cache;
 414         }
 415         ovl_set_dir_cache(d_inode(dentry), NULL);
 416
 417         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 418         if (!cache)
 419                 return ERR_PTR(-ENOMEM);
 420
 421         cache->refcount = 1;
 422         INIT_LIST_HEAD(&cache->entries);
 423         cache->root = RB_ROOT;
 424
 425         res = ovl_dir_read_merged(dentry, &cache->entries, &cache->root);
 426         if (res) {
 427                 ovl_cache_free(&cache->entries);
 428                 kfree(cache);
 429                 return ERR_PTR(res);
 430         }
 431
 432         cache->version = ovl_dentry_version_get(dentry);
 433         ovl_set_dir_cache(d_inode(dentry), cache);
 434
 435         return cache;
 436 }
 437
 438 /*
 439  * Set d_ino for upper entries. Non-upper entries should always report
 440  * the uppermost real inode ino and should not call this function.
 441  *
 442  * When not all layer are on same fs, report real ino also for upper.
 443  *
 444  * When all layers are on the same fs, and upper has a reference to
 445  * copy up origin, call vfs_getattr() on the overlay entry to make
 446  * sure that d_ino will be consistent with st_ino from stat(2).
 447  */
 448 static int ovl_cache_update_ino(struct path *path, struct ovl_cache_entry *p)
 449
 450 {
 451         struct dentry *dir = path->dentry;
 452         struct dentry *this = NULL;
 453         enum ovl_path_type type;
 454         u64 ino = p->real_ino;
 455         int err = 0;
 456
 457         if (!ovl_same_sb(dir->d_sb))
 458                 goto out;
 459
 460         if (p->name[0] == '.') {
 461                 if (p->len == 1) {
 462                         this = dget(dir);
 463                         goto get;
 464                 }
 465                 if (p->len == 2 && p->name[1] == '.') {
 466                         /* we shall not be moved */
 467                         this = dget(dir->d_parent);
 468                         goto get;
 469                 }
 470         }
 471         this = lookup_one_len(p->name, dir, p->len);
 472         if (IS_ERR_OR_NULL(this) || !this->d_inode) {
 473                 if (IS_ERR(this)) {
 474                         err = PTR_ERR(this);
 475                         this = NULL;
 476                         goto fail;
 477                 }
 478                 goto out;
 479         }
 480
 481 get:
 482         type = ovl_path_type(this);
 483         if (OVL_TYPE_ORIGIN(type)) {
 484                 struct kstat stat;
 485                 struct path statpath = *path;
 486
 487                 statpath.dentry = this;
 488                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 489                 if (err)
 490                         goto fail;
 491
 492                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 493                 ino = stat.ino;
 494         }
 495
 496 out:
 497         p->ino = ino;
 498         dput(this);
 499         return err;
 500
 501 fail:
 502         pr_warn_ratelimited("overlayfs: failed to look up (%s) for ino (%i)\n",
 503                             p->name, err);
 504         goto out;
 505 }
 506
 507 static int ovl_fill_plain(struct dir_context *ctx, const char *name,
 508                           int namelen, loff_t offset, u64 ino,
 509                           unsigned int d_type)
 510 {
 511         struct ovl_cache_entry *p;
 512         struct ovl_readdir_data *rdd =
 513                 container_of(ctx, struct ovl_readdir_data, ctx);
 514
 515         rdd->count++;
 516         p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 517         if (p == NULL) {
 518                 rdd->err = -ENOMEM;
 519                 return -ENOMEM;
 520         }
 521         list_add_tail(&p->l_node, rdd->list);
 522
 523         return 0;
 524 }
 525
 526 static int ovl_dir_read_impure(struct path *path,  struct list_head *list,
 527                                struct rb_root *root)
 528 {
 529         int err;
 530         struct path realpath;
 531         struct ovl_cache_entry *p, *n;
 532         struct ovl_readdir_data rdd = {
 533                 .ctx.actor = ovl_fill_plain,
 534                 .list = list,
 535                 .root = root,
 536         };
 537
 538         INIT_LIST_HEAD(list);
 539         *root = RB_ROOT;
 540         ovl_path_upper(path->dentry, &realpath);
 541
 542         err = ovl_dir_read(&realpath, &rdd);
 543         if (err)
 544                 return err;
 545
 546         list_for_each_entry_safe(p, n, list, l_node) {
 547                 if (strcmp(p->name, ".") != 0 &&
 548                     strcmp(p->name, "..") != 0) {
 549                         err = ovl_cache_update_ino(path, p);
 550                         if (err)
 551                                 return err;
 552                 }
 553                 if (p->ino == p->real_ino) {
 554                         list_del(&p->l_node);
 555                         kfree(p);
 556                 } else {
 557                         struct rb_node **newp = &root->rb_node;
 558                         struct rb_node *parent = NULL;
 559
 560                         if (WARN_ON(ovl_cache_entry_find_link(p->name, p->len,
 561                                                               &newp, &parent)))
 562                                 return -EIO;
 563
 564                         rb_link_node(&p->node, parent, newp);
 565                         rb_insert_color(&p->node, root);
 566                 }
 567         }
 568         return 0;
 569 }
 570
 571 static struct ovl_dir_cache *ovl_cache_get_impure(struct path *path)
 572 {
 573         int res;
 574         struct dentry *dentry = path->dentry;
 575         struct ovl_dir_cache *cache;
 576
 577         cache = ovl_dir_cache(d_inode(dentry));
 578         if (cache && ovl_dentry_version_get(dentry) == cache->version)
 579                 return cache;
 580
 581         /* Impure cache is not refcounted, free it here */
 582         ovl_dir_cache_free(d_inode(dentry));
 583         ovl_set_dir_cache(d_inode(dentry), NULL);
 584
 585         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 586         if (!cache)
 587                 return ERR_PTR(-ENOMEM);
 588
 589         res = ovl_dir_read_impure(path, &cache->entries, &cache->root);
 590         if (res) {
 591                 ovl_cache_free(&cache->entries);
 592                 kfree(cache);
 593                 return ERR_PTR(res);
 594         }
 595         if (list_empty(&cache->entries)) {
 596                 /*
 597                  * A good opportunity to get rid of an unneeded "impure" flag.
 598                  * Removing the "impure" xattr is best effort.
 599                  */
 600                 if (!ovl_want_write(dentry)) {
 601                         ovl_do_removexattr(ovl_dentry_upper(dentry),
 602                                            OVL_XATTR_IMPURE);
 603                         ovl_drop_write(dentry);
 604                 }
 605                 ovl_clear_flag(OVL_IMPURE, d_inode(dentry));
 606                 kfree(cache);
 607                 return NULL;
 608         }
 609
 610         cache->version = ovl_dentry_version_get(dentry);
 611         ovl_set_dir_cache(d_inode(dentry), cache);
 612
 613         return cache;
 614 }
 615
 616 struct ovl_readdir_translate {
 617         struct dir_context *orig_ctx;
 618         struct ovl_dir_cache *cache;
 619         struct dir_context ctx;
 620         u64 parent_ino;
 621 };
 622
 623 static int ovl_fill_real(struct dir_context *ctx, const char *name,
 624                            int namelen, loff_t offset, u64 ino,
 625                            unsigned int d_type)
 626 {
 627         struct ovl_readdir_translate *rdt =
 628                 container_of(ctx, struct ovl_readdir_translate, ctx);
 629         struct dir_context *orig_ctx = rdt->orig_ctx;
 630
 631         if (rdt->parent_ino && strcmp(name, "..") == 0)
 632                 ino = rdt->parent_ino;
 633         else if (rdt->cache) {
 634                 struct ovl_cache_entry *p;
 635
 636                 p = ovl_cache_entry_find(&rdt->cache->root, name, namelen);
 637                 if (p)
 638                         ino = p->ino;
 639         }
 640
 641         return orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
 642 }
 643
 644 static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
 645 {
 646         int err;
 647         struct ovl_dir_file *od = file->private_data;
 648         struct dentry *dir = file->f_path.dentry;
 649         struct ovl_readdir_translate rdt = {
 650                 .ctx.actor = ovl_fill_real,
 651                 .orig_ctx = ctx,
 652         };
 653
 654         if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
 655                 struct kstat stat;
 656                 struct path statpath = file->f_path;
 657
 658                 statpath.dentry = dir->d_parent;
 659                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 660                 if (err)
 661                         return err;
 662
 663                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 664                 rdt.parent_ino = stat.ino;
 665         }
 666
 667         if (ovl_test_flag(OVL_IMPURE, d_inode(dir))) {
 668                 rdt.cache = ovl_cache_get_impure(&file->f_path);
 669                 if (IS_ERR(rdt.cache))
 670                         return PTR_ERR(rdt.cache);
 671         }
 672
 673         err = iterate_dir(od->realfile, &rdt.ctx);
 674         ctx->pos = rdt.ctx.pos;
 675
 676         return err;
 677 }
 678
 679
 680 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 681 {
 682         struct ovl_dir_file *od = file->private_data;
 683         struct dentry *dentry = file->f_path.dentry;
 684         struct ovl_cache_entry *p;
 685         int err;
 686
 687         if (!ctx->pos)
 688                 ovl_dir_reset(file);
 689
 690         if (od->is_real) {
 691                 /*
 692                  * If parent is merge, then need to adjust d_ino for '..', if
 693                  * dir is impure then need to adjust d_ino for copied up
 694                  * entries.
 695                  */
 696                 if (ovl_same_sb(dentry->d_sb) &&
 697                     (ovl_test_flag(OVL_IMPURE, d_inode(dentry)) ||
 698                      OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent)))) {
 699                         return ovl_iterate_real(file, ctx);
 700                 }
 701                 return iterate_dir(od->realfile, ctx);
 702         }
 703
 704         if (!od->cache) {
 705                 struct ovl_dir_cache *cache;
 706
 707                 cache = ovl_cache_get(dentry);
 708                 if (IS_ERR(cache))
 709                         return PTR_ERR(cache);
 710
 711                 od->cache = cache;
 712                 ovl_seek_cursor(od, ctx->pos);
 713         }
 714
 715         while (od->cursor != &od->cache->entries) {
 716                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 717                 if (!p->is_whiteout) {
 718                         if (!p->ino) {
 719                                 err = ovl_cache_update_ino(&file->f_path, p);
 720                                 if (err)
 721                                         return err;
 722                         }
 723                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 724                                 break;
 725                 }
 726                 od->cursor = p->l_node.next;
 727                 ctx->pos++;
 728         }
 729         return 0;
 730 }
 731
 732 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 733 {
 734         loff_t res;
 735         struct ovl_dir_file *od = file->private_data;
 736
 737         inode_lock(file_inode(file));
 738         if (!file->f_pos)
 739                 ovl_dir_reset(file);
 740
 741         if (od->is_real) {
 742                 res = vfs_llseek(od->realfile, offset, origin);
 743                 file->f_pos = od->realfile->f_pos;
 744         } else {
 745                 res = -EINVAL;
 746
 747                 switch (origin) {
 748                 case SEEK_CUR:
 749                         offset += file->f_pos;
 750                         break;
 751                 case SEEK_SET:
 752                         break;
 753                 default:
 754                         goto out_unlock;
 755                 }
 756                 if (offset < 0)
 757                         goto out_unlock;
 758
 759                 if (offset != file->f_pos) {
 760                         file->f_pos = offset;
 761                         if (od->cache)
 762                                 ovl_seek_cursor(od, offset);
 763                 }
 764                 res = offset;
 765         }
 766 out_unlock:
 767         inode_unlock(file_inode(file));
 768
 769         return res;
 770 }
 771
 772 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 773                          int datasync)
 774 {
 775         struct ovl_dir_file *od = file->private_data;
 776         struct dentry *dentry = file->f_path.dentry;
 777         struct file *realfile = od->realfile;
 778
 779         /* Nothing to sync for lower */
 780         if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
 781                 return 0;
 782
 783         /*
 784          * Need to check if we started out being a lower dir, but got copied up
 785          */
 786         if (!od->is_upper) {
 787                 struct inode *inode = file_inode(file);
 788
 789                 realfile = READ_ONCE(od->upperfile);
 790                 if (!realfile) {
 791                         struct path upperpath;
 792
 793                         ovl_path_upper(dentry, &upperpath);
 794                         realfile = ovl_path_open(&upperpath, O_RDONLY);
 795
 796                         inode_lock(inode);
 797                         if (!od->upperfile) {
 798                                 if (IS_ERR(realfile)) {
 799                                         inode_unlock(inode);
 800                                         return PTR_ERR(realfile);
 801                                 }
 802                                 smp_store_release(&od->upperfile, realfile);
 803                         } else {
 804                                 /* somebody has beaten us to it */
 805                                 if (!IS_ERR(realfile))
 806                                         fput(realfile);
 807                                 realfile = od->upperfile;
 808                         }
 809                         inode_unlock(inode);
 810                 }
 811         }
 812
 813         return vfs_fsync_range(realfile, start, end, datasync);
 814 }
 815
 816 static int ovl_dir_release(struct inode *inode, struct file *file)
 817 {
 818         struct ovl_dir_file *od = file->private_data;
 819
 820         if (od->cache) {
 821                 inode_lock(inode);
 822                 ovl_cache_put(od, file->f_path.dentry);
 823                 inode_unlock(inode);
 824         }
 825         fput(od->realfile);
 826         if (od->upperfile)
 827                 fput(od->upperfile);
 828         kfree(od);
 829
 830         return 0;
 831 }
 832
 833 static int ovl_dir_open(struct inode *inode, struct file *file)
 834 {
 835         struct path realpath;
 836         struct file *realfile;
 837         struct ovl_dir_file *od;
 838         enum ovl_path_type type;
 839
 840         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 841         if (!od)
 842                 return -ENOMEM;
 843
 844         type = ovl_path_real(file->f_path.dentry, &realpath);
 845         realfile = ovl_path_open(&realpath, file->f_flags);
 846         if (IS_ERR(realfile)) {
 847                 kfree(od);
 848                 return PTR_ERR(realfile);
 849         }
 850         od->realfile = realfile;
 851         od->is_real = ovl_dir_is_real(file->f_path.dentry);
 852         od->is_upper = OVL_TYPE_UPPER(type);
 853         file->private_data = od;
 854
 855         return 0;
 856 }
 857
 858 const struct file_operations ovl_dir_operations = {
 859         .read           = generic_read_dir,
 860         .open           = ovl_dir_open,
 861         .iterate        = ovl_iterate,
 862         .llseek         = ovl_dir_llseek,
 863         .fsync          = ovl_dir_fsync,
 864         .release        = ovl_dir_release,
 865 };
 866
 867 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 868 {
 869         int err;
 870         struct ovl_cache_entry *p, *n;
 871         struct rb_root root = RB_ROOT;
 872         const struct cred *old_cred;
 873
 874         old_cred = ovl_override_creds(dentry->d_sb);
 875         err = ovl_dir_read_merged(dentry, list, &root);
 876         revert_creds(old_cred);
 877         if (err)
 878                 return err;
 879
 880         err = 0;
 881
 882         list_for_each_entry_safe(p, n, list, l_node) {
 883                 /*
 884                  * Select whiteouts in upperdir, they should
 885                  * be cleared when deleting this directory.
 886                  */
 887                 if (p->is_whiteout) {
 888                         if (p->is_upper)
 889                                 continue;
 890                         goto del_entry;
 891                 }
 892
 893                 if (p->name[0] == '.') {
 894                         if (p->len == 1)
 895                                 goto del_entry;
 896                         if (p->len == 2 && p->name[1] == '.')
 897                                 goto del_entry;
 898                 }
 899                 err = -ENOTEMPTY;
 900                 break;
 901
 902 del_entry:
 903                 list_del(&p->l_node);
 904                 kfree(p);
 905         }
 906
 907         return err;
 908 }
 909
 910 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
 911 {
 912         struct ovl_cache_entry *p;
 913
 914         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
 915         list_for_each_entry(p, list, l_node) {
 916                 struct dentry *dentry;
 917
 918                 if (WARN_ON(!p->is_whiteout || !p->is_upper))
 919                         continue;
 920
 921                 dentry = lookup_one_len(p->name, upper, p->len);
 922                 if (IS_ERR(dentry)) {
 923                         pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
 924                                upper->d_name.name, p->len, p->name,
 925                                (int) PTR_ERR(dentry));
 926                         continue;
 927                 }
 928                 if (dentry->d_inode)
 929                         ovl_cleanup(upper->d_inode, dentry);
 930                 dput(dentry);
 931         }
 932         inode_unlock(upper->d_inode);
 933 }
 934
 935 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
 936                           int namelen, loff_t offset, u64 ino,
 937                           unsigned int d_type)
 938 {
 939         struct ovl_readdir_data *rdd =
 940                 container_of(ctx, struct ovl_readdir_data, ctx);
 941
 942         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
 943         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
 944                 return 0;
 945
 946         if (d_type != DT_UNKNOWN)
 947                 rdd->d_type_supported = true;
 948
 949         return 0;
 950 }
 951
 952 /*
 953  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
 954  * if error is encountered.
 955  */
 956 int ovl_check_d_type_supported(struct path *realpath)
 957 {
 958         int err;
 959         struct ovl_readdir_data rdd = {
 960                 .ctx.actor = ovl_check_d_type,
 961                 .d_type_supported = false,
 962         };
 963
 964         err = ovl_dir_read(realpath, &rdd);
 965         if (err)
 966                 return err;
 967
 968         return rdd.d_type_supported;
 969 }
 970
 971 static void ovl_workdir_cleanup_recurse(struct path *path, int level)
 972 {
 973         int err;
 974         struct inode *dir = path->dentry->d_inode;
 975         LIST_HEAD(list);
 976         struct rb_root root = RB_ROOT;
 977         struct ovl_cache_entry *p;
 978         struct ovl_readdir_data rdd = {
 979                 .ctx.actor = ovl_fill_merge,
 980                 .dentry = NULL,
 981                 .list = &list,
 982                 .root = &root,
 983                 .is_lowest = false,
 984         };
 985
 986         err = ovl_dir_read(path, &rdd);
 987         if (err)
 988                 goto out;
 989
 990         inode_lock_nested(dir, I_MUTEX_PARENT);
 991         list_for_each_entry(p, &list, l_node) {
 992                 struct dentry *dentry;
 993
 994                 if (p->name[0] == '.') {
 995                         if (p->len == 1)
 996                                 continue;
 997                         if (p->len == 2 && p->name[1] == '.')
 998                                 continue;
 999                 }
1000                 dentry = lookup_one_len(p->name, path->dentry, p->len);
1001                 if (IS_ERR(dentry))
1002                         continue;
1003                 if (dentry->d_inode)
1004                         ovl_workdir_cleanup(dir, path->mnt, dentry, level);
1005                 dput(dentry);
1006         }
1007         inode_unlock(dir);
1008 out:
1009         ovl_cache_free(&list);
1010 }
1011
1012 void ovl_workdir_cleanup(struct inode *dir, struct vfsmount *mnt,
1013                          struct dentry *dentry, int level)
1014 {
1015         int err;
1016
1017         if (!d_is_dir(dentry) || level > 1) {
1018                 ovl_cleanup(dir, dentry);
1019                 return;
1020         }
1021
1022         err = ovl_do_rmdir(dir, dentry);
1023         if (err) {
1024                 struct path path = { .mnt = mnt, .dentry = dentry };
1025
1026                 inode_unlock(dir);
1027                 ovl_workdir_cleanup_recurse(&path, level + 1);
1028                 inode_lock_nested(dir, I_MUTEX_PARENT);
1029                 ovl_cleanup(dir, dentry);
1030         }
1031 }
1032
1033 int ovl_indexdir_cleanup(struct ovl_fs *ofs)
1034 {
1035         int err;
1036         struct dentry *indexdir = ofs->indexdir;
1037         struct dentry *index = NULL;
1038         struct inode *dir = indexdir->d_inode;
1039         struct path path = { .mnt = ofs->upper_mnt, .dentry = indexdir };
1040         LIST_HEAD(list);
1041         struct rb_root root = RB_ROOT;
1042         struct ovl_cache_entry *p;
1043         struct ovl_readdir_data rdd = {
1044                 .ctx.actor = ovl_fill_merge,
1045                 .dentry = NULL,
1046                 .list = &list,
1047                 .root = &root,
1048                 .is_lowest = false,
1049         };
1050
1051         err = ovl_dir_read(&path, &rdd);
1052         if (err)
1053                 goto out;
1054
1055         inode_lock_nested(dir, I_MUTEX_PARENT);
1056         list_for_each_entry(p, &list, l_node) {
1057                 if (p->name[0] == '.') {
1058                         if (p->len == 1)
1059                                 continue;
1060                         if (p->len == 2 && p->name[1] == '.')
1061                                 continue;
1062                 }
1063                 index = lookup_one_len(p->name, indexdir, p->len);
1064                 if (IS_ERR(index)) {
1065                         err = PTR_ERR(index);
1066                         index = NULL;
1067                         break;
1068                 }
1069                 err = ovl_verify_index(ofs, index);
1070                 if (!err) {
1071                         goto next;
1072                 } else if (err == -ESTALE) {
1073                         /* Cleanup stale index entries */
1074                         err = ovl_cleanup(dir, index);
1075                 } else if (err != -ENOENT) {
1076                         /*
1077                          * Abort mount to avoid corrupting the index if
1078                          * an incompatible index entry was found or on out
1079                          * of memory.
1080                          */
1081                         break;
1082                 } else if (ofs->config.nfs_export) {
1083                         /*
1084                          * Whiteout orphan index to block future open by
1085                          * handle after overlay nlink dropped to zero.
1086                          */
1087                         err = ovl_cleanup_and_whiteout(indexdir, dir, index);
1088                 } else {
1089                         /* Cleanup orphan index entries */
1090                         err = ovl_cleanup(dir, index);
1091                 }
1092
1093                 if (err)
1094                         break;
1095
1096 next:
1097                 dput(index);
1098                 index = NULL;
1099         }
1100         dput(index);
1101         inode_unlock(dir);
1102 out:
1103         ovl_cache_free(&list);
1104         if (err)
1105                 pr_err("overlayfs: failed index dir cleanup (%i)\n", err);
1106         return err;
1107 }