fs/overlayfs/readdir.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <linux/fs.h>
   8 #include <linux/slab.h>
   9 #include <linux/namei.h>
  10 #include <linux/file.h>
  11 #include <linux/xattr.h>
  12 #include <linux/rbtree.h>
  13 #include <linux/security.h>
  14 #include <linux/cred.h>
  15 #include <linux/ratelimit.h>
  16 #include "overlayfs.h"
  17
  18 struct ovl_cache_entry {
  19         unsigned int len;
  20         unsigned int type;
  21         u64 real_ino;
  22         u64 ino;
  23         struct list_head l_node;
  24         struct rb_node node;
  25         struct ovl_cache_entry *next_maybe_whiteout;
  26         bool is_upper;
  27         bool is_whiteout;
  28         char name[];
  29 };
  30
  31 struct ovl_dir_cache {
  32         long refcount;
  33         u64 version;
  34         struct list_head entries;
  35         struct rb_root root;
  36 };
  37
  38 struct ovl_readdir_data {
  39         struct dir_context ctx;
  40         struct dentry *dentry;
  41         bool is_lowest;
  42         struct rb_root *root;
  43         struct list_head *list;
  44         struct list_head middle;
  45         struct ovl_cache_entry *first_maybe_whiteout;
  46         int count;
  47         int err;
  48         bool is_upper;
  49         bool d_type_supported;
  50 };
  51
  52 struct ovl_dir_file {
  53         bool is_real;
  54         bool is_upper;
  55         struct ovl_dir_cache *cache;
  56         struct list_head *cursor;
  57         struct file *realfile;
  58         struct file *upperfile;
  59 };
  60
  61 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
  62 {
  63         return rb_entry(n, struct ovl_cache_entry, node);
  64 }
  65
  66 static bool ovl_cache_entry_find_link(const char *name, int len,
  67                                       struct rb_node ***link,
  68                                       struct rb_node **parent)
  69 {
  70         bool found = false;
  71         struct rb_node **newp = *link;
  72
  73         while (!found && *newp) {
  74                 int cmp;
  75                 struct ovl_cache_entry *tmp;
  76
  77                 *parent = *newp;
  78                 tmp = ovl_cache_entry_from_node(*newp);
  79                 cmp = strncmp(name, tmp->name, len);
  80                 if (cmp > 0)
  81                         newp = &tmp->node.rb_right;
  82                 else if (cmp < 0 || len < tmp->len)
  83                         newp = &tmp->node.rb_left;
  84                 else
  85                         found = true;
  86         }
  87         *link = newp;
  88
  89         return found;
  90 }
  91
  92 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
  93                                                     const char *name, int len)
  94 {
  95         struct rb_node *node = root->rb_node;
  96         int cmp;
  97
  98         while (node) {
  99                 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
 100
 101                 cmp = strncmp(name, p->name, len);
 102                 if (cmp > 0)
 103                         node = p->node.rb_right;
 104                 else if (cmp < 0 || len < p->len)
 105                         node = p->node.rb_left;
 106                 else
 107                         return p;
 108         }
 109
 110         return NULL;
 111 }
 112
 113 static bool ovl_calc_d_ino(struct ovl_readdir_data *rdd,
 114                            struct ovl_cache_entry *p)
 115 {
 116         /* Don't care if not doing ovl_iter() */
 117         if (!rdd->dentry)
 118                 return false;
 119
 120         /* Always recalc d_ino when remapping lower inode numbers */
 121         if (ovl_xino_bits(rdd->dentry->d_sb))
 122                 return true;
 123
 124         /* Always recalc d_ino for parent */
 125         if (strcmp(p->name, "..") == 0)
 126                 return true;
 127
 128         /* If this is lower, then native d_ino will do */
 129         if (!rdd->is_upper)
 130                 return false;
 131
 132         /*
 133          * Recalc d_ino for '.' and for all entries if dir is impure (contains
 134          * copied up entries)
 135          */
 136         if ((p->name[0] == '.' && p->len == 1) ||
 137             ovl_test_flag(OVL_IMPURE, d_inode(rdd->dentry)))
 138                 return true;
 139
 140         return false;
 141 }
 142
 143 static struct ovl_cache_entry *ovl_cache_entry_new(struct ovl_readdir_data *rdd,
 144                                                    const char *name, int len,
 145                                                    u64 ino, unsigned int d_type)
 146 {
 147         struct ovl_cache_entry *p;
 148         size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
 149
 150         p = kmalloc(size, GFP_KERNEL);
 151         if (!p)
 152                 return NULL;
 153
 154         memcpy(p->name, name, len);
 155         p->name[len] = '\0';
 156         p->len = len;
 157         p->type = d_type;
 158         p->real_ino = ino;
 159         p->ino = ino;
 160         /* Defer setting d_ino for upper entry to ovl_iterate() */
 161         if (ovl_calc_d_ino(rdd, p))
 162                 p->ino = 0;
 163         p->is_upper = rdd->is_upper;
 164         p->is_whiteout = false;
 165
 166         if (d_type == DT_CHR) {
 167                 p->next_maybe_whiteout = rdd->first_maybe_whiteout;
 168                 rdd->first_maybe_whiteout = p;
 169         }
 170         return p;
 171 }
 172
 173 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
 174                                   const char *name, int len, u64 ino,
 175                                   unsigned int d_type)
 176 {
 177         struct rb_node **newp = &rdd->root->rb_node;
 178         struct rb_node *parent = NULL;
 179         struct ovl_cache_entry *p;
 180
 181         if (ovl_cache_entry_find_link(name, len, &newp, &parent))
 182                 return 0;
 183
 184         p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
 185         if (p == NULL) {
 186                 rdd->err = -ENOMEM;
 187                 return -ENOMEM;
 188         }
 189
 190         list_add_tail(&p->l_node, rdd->list);
 191         rb_link_node(&p->node, parent, newp);
 192         rb_insert_color(&p->node, rdd->root);
 193
 194         return 0;
 195 }
 196
 197 static int ovl_fill_lowest(struct ovl_readdir_data *rdd,
 198                            const char *name, int namelen,
 199                            loff_t offset, u64 ino, unsigned int d_type)
 200 {
 201         struct ovl_cache_entry *p;
 202
 203         p = ovl_cache_entry_find(rdd->root, name, namelen);
 204         if (p) {
 205                 list_move_tail(&p->l_node, &rdd->middle);
 206         } else {
 207                 p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 208                 if (p == NULL)
 209                         rdd->err = -ENOMEM;
 210                 else
 211                         list_add_tail(&p->l_node, &rdd->middle);
 212         }
 213
 214         return rdd->err;
 215 }
 216
 217 void ovl_cache_free(struct list_head *list)
 218 {
 219         struct ovl_cache_entry *p;
 220         struct ovl_cache_entry *n;
 221
 222         list_for_each_entry_safe(p, n, list, l_node)
 223                 kfree(p);
 224
 225         INIT_LIST_HEAD(list);
 226 }
 227
 228 void ovl_dir_cache_free(struct inode *inode)
 229 {
 230         struct ovl_dir_cache *cache = ovl_dir_cache(inode);
 231
 232         if (cache) {
 233                 ovl_cache_free(&cache->entries);
 234                 kfree(cache);
 235         }
 236 }
 237
 238 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
 239 {
 240         struct ovl_dir_cache *cache = od->cache;
 241
 242         WARN_ON(cache->refcount <= 0);
 243         cache->refcount--;
 244         if (!cache->refcount) {
 245                 if (ovl_dir_cache(d_inode(dentry)) == cache)
 246                         ovl_set_dir_cache(d_inode(dentry), NULL);
 247
 248                 ovl_cache_free(&cache->entries);
 249                 kfree(cache);
 250         }
 251 }
 252
 253 static int ovl_fill_merge(struct dir_context *ctx, const char *name,
 254                           int namelen, loff_t offset, u64 ino,
 255                           unsigned int d_type)
 256 {
 257         struct ovl_readdir_data *rdd =
 258                 container_of(ctx, struct ovl_readdir_data, ctx);
 259
 260         rdd->count++;
 261         if (!rdd->is_lowest)
 262                 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
 263         else
 264                 return ovl_fill_lowest(rdd, name, namelen, offset, ino, d_type);
 265 }
 266
 267 static int ovl_check_whiteouts(struct dentry *dir, struct ovl_readdir_data *rdd)
 268 {
 269         int err;
 270         struct ovl_cache_entry *p;
 271         struct dentry *dentry;
 272         const struct cred *old_cred;
 273
 274         old_cred = ovl_override_creds(rdd->dentry->d_sb);
 275
 276         err = down_write_killable(&dir->d_inode->i_rwsem);
 277         if (!err) {
 278                 while (rdd->first_maybe_whiteout) {
 279                         p = rdd->first_maybe_whiteout;
 280                         rdd->first_maybe_whiteout = p->next_maybe_whiteout;
 281                         dentry = lookup_one_len(p->name, dir, p->len);
 282                         if (!IS_ERR(dentry)) {
 283                                 p->is_whiteout = ovl_is_whiteout(dentry);
 284                                 dput(dentry);
 285                         }
 286                 }
 287                 inode_unlock(dir->d_inode);
 288         }
 289         revert_creds(old_cred);
 290
 291         return err;
 292 }
 293
 294 static inline int ovl_dir_read(struct path *realpath,
 295                                struct ovl_readdir_data *rdd)
 296 {
 297         struct file *realfile;
 298         int err;
 299
 300         realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
 301         if (IS_ERR(realfile))
 302                 return PTR_ERR(realfile);
 303
 304         rdd->first_maybe_whiteout = NULL;
 305         rdd->ctx.pos = 0;
 306         do {
 307                 rdd->count = 0;
 308                 rdd->err = 0;
 309                 err = iterate_dir(realfile, &rdd->ctx);
 310                 if (err >= 0)
 311                         err = rdd->err;
 312         } while (!err && rdd->count);
 313
 314         if (!err && rdd->first_maybe_whiteout && rdd->dentry)
 315                 err = ovl_check_whiteouts(realpath->dentry, rdd);
 316
 317         fput(realfile);
 318
 319         return err;
 320 }
 321
 322 /*
 323  * Can we iterate real dir directly?
 324  *
 325  * Non-merge dir may contain whiteouts from a time it was a merge upper, before
 326  * lower dir was removed under it and possibly before it was rotated from upper
 327  * to lower layer.
 328  */
 329 static bool ovl_dir_is_real(struct dentry *dir)
 330 {
 331         return !ovl_test_flag(OVL_WHITEOUTS, d_inode(dir));
 332 }
 333
 334 static void ovl_dir_reset(struct file *file)
 335 {
 336         struct ovl_dir_file *od = file->private_data;
 337         struct ovl_dir_cache *cache = od->cache;
 338         struct dentry *dentry = file->f_path.dentry;
 339         bool is_real;
 340
 341         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 342                 ovl_cache_put(od, dentry);
 343                 od->cache = NULL;
 344                 od->cursor = NULL;
 345         }
 346         is_real = ovl_dir_is_real(dentry);
 347         if (od->is_real != is_real) {
 348                 /* is_real can only become false when dir is copied up */
 349                 if (WARN_ON(is_real))
 350                         return;
 351                 od->is_real = false;
 352         }
 353 }
 354
 355 static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list,
 356         struct rb_root *root)
 357 {
 358         int err;
 359         struct path realpath;
 360         struct ovl_readdir_data rdd = {
 361                 .ctx.actor = ovl_fill_merge,
 362                 .dentry = dentry,
 363                 .list = list,
 364                 .root = root,
 365                 .is_lowest = false,
 366         };
 367         int idx, next;
 368
 369         for (idx = 0; idx != -1; idx = next) {
 370                 next = ovl_path_next(idx, dentry, &realpath);
 371                 rdd.is_upper = ovl_dentry_upper(dentry) == realpath.dentry;
 372
 373                 if (next != -1) {
 374                         err = ovl_dir_read(&realpath, &rdd);
 375                         if (err)
 376                                 break;
 377                 } else {
 378                         /*
 379                          * Insert lowest layer entries before upper ones, this
 380                          * allows offsets to be reasonably constant
 381                          */
 382                         list_add(&rdd.middle, rdd.list);
 383                         rdd.is_lowest = true;
 384                         err = ovl_dir_read(&realpath, &rdd);
 385                         list_del(&rdd.middle);
 386                 }
 387         }
 388         return err;
 389 }
 390
 391 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 392 {
 393         struct list_head *p;
 394         loff_t off = 0;
 395
 396         list_for_each(p, &od->cache->entries) {
 397                 if (off >= pos)
 398                         break;
 399                 off++;
 400         }
 401         /* Cursor is safe since the cache is stable */
 402         od->cursor = p;
 403 }
 404
 405 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 406 {
 407         int res;
 408         struct ovl_dir_cache *cache;
 409
 410         cache = ovl_dir_cache(d_inode(dentry));
 411         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 412                 WARN_ON(!cache->refcount);
 413                 cache->refcount++;
 414                 return cache;
 415         }
 416         ovl_set_dir_cache(d_inode(dentry), NULL);
 417
 418         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 419         if (!cache)
 420                 return ERR_PTR(-ENOMEM);
 421
 422         cache->refcount = 1;
 423         INIT_LIST_HEAD(&cache->entries);
 424         cache->root = RB_ROOT;
 425
 426         res = ovl_dir_read_merged(dentry, &cache->entries, &cache->root);
 427         if (res) {
 428                 ovl_cache_free(&cache->entries);
 429                 kfree(cache);
 430                 return ERR_PTR(res);
 431         }
 432
 433         cache->version = ovl_dentry_version_get(dentry);
 434         ovl_set_dir_cache(d_inode(dentry), cache);
 435
 436         return cache;
 437 }
 438
 439 /* Map inode number to lower fs unique range */
 440 static u64 ovl_remap_lower_ino(u64 ino, int xinobits, int fsid,
 441                                const char *name, int namelen)
 442 {
 443         if (ino >> (64 - xinobits)) {
 444                 pr_warn_ratelimited("overlayfs: d_ino too big (%.*s, ino=%llu, xinobits=%d)\n",
 445                                     namelen, name, ino, xinobits);
 446                 return ino;
 447         }
 448
 449         return ino | ((u64)fsid) << (64 - xinobits);
 450 }
 451
 452 /*
 453  * Set d_ino for upper entries. Non-upper entries should always report
 454  * the uppermost real inode ino and should not call this function.
 455  *
 456  * When not all layer are on same fs, report real ino also for upper.
 457  *
 458  * When all layers are on the same fs, and upper has a reference to
 459  * copy up origin, call vfs_getattr() on the overlay entry to make
 460  * sure that d_ino will be consistent with st_ino from stat(2).
 461  */
 462 static int ovl_cache_update_ino(struct path *path, struct ovl_cache_entry *p)
 463
 464 {
 465         struct dentry *dir = path->dentry;
 466         struct dentry *this = NULL;
 467         enum ovl_path_type type;
 468         u64 ino = p->real_ino;
 469         int xinobits = ovl_xino_bits(dir->d_sb);
 470         int err = 0;
 471
 472         if (!ovl_same_sb(dir->d_sb) && !xinobits)
 473                 goto out;
 474
 475         if (p->name[0] == '.') {
 476                 if (p->len == 1) {
 477                         this = dget(dir);
 478                         goto get;
 479                 }
 480                 if (p->len == 2 && p->name[1] == '.') {
 481                         /* we shall not be moved */
 482                         this = dget(dir->d_parent);
 483                         goto get;
 484                 }
 485         }
 486         this = lookup_one_len(p->name, dir, p->len);
 487         if (IS_ERR_OR_NULL(this) || !this->d_inode) {
 488                 if (IS_ERR(this)) {
 489                         err = PTR_ERR(this);
 490                         this = NULL;
 491                         goto fail;
 492                 }
 493                 goto out;
 494         }
 495
 496 get:
 497         type = ovl_path_type(this);
 498         if (OVL_TYPE_ORIGIN(type)) {
 499                 struct kstat stat;
 500                 struct path statpath = *path;
 501
 502                 statpath.dentry = this;
 503                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 504                 if (err)
 505                         goto fail;
 506
 507                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 508                 ino = stat.ino;
 509         } else if (xinobits && !OVL_TYPE_UPPER(type)) {
 510                 ino = ovl_remap_lower_ino(ino, xinobits,
 511                                           ovl_layer_lower(this)->fsid,
 512                                           p->name, p->len);
 513         }
 514
 515 out:
 516         p->ino = ino;
 517         dput(this);
 518         return err;
 519
 520 fail:
 521         pr_warn_ratelimited("overlayfs: failed to look up (%s) for ino (%i)\n",
 522                             p->name, err);
 523         goto out;
 524 }
 525
 526 static int ovl_fill_plain(struct dir_context *ctx, const char *name,
 527                           int namelen, loff_t offset, u64 ino,
 528                           unsigned int d_type)
 529 {
 530         struct ovl_cache_entry *p;
 531         struct ovl_readdir_data *rdd =
 532                 container_of(ctx, struct ovl_readdir_data, ctx);
 533
 534         rdd->count++;
 535         p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 536         if (p == NULL) {
 537                 rdd->err = -ENOMEM;
 538                 return -ENOMEM;
 539         }
 540         list_add_tail(&p->l_node, rdd->list);
 541
 542         return 0;
 543 }
 544
 545 static int ovl_dir_read_impure(struct path *path,  struct list_head *list,
 546                                struct rb_root *root)
 547 {
 548         int err;
 549         struct path realpath;
 550         struct ovl_cache_entry *p, *n;
 551         struct ovl_readdir_data rdd = {
 552                 .ctx.actor = ovl_fill_plain,
 553                 .list = list,
 554                 .root = root,
 555         };
 556
 557         INIT_LIST_HEAD(list);
 558         *root = RB_ROOT;
 559         ovl_path_upper(path->dentry, &realpath);
 560
 561         err = ovl_dir_read(&realpath, &rdd);
 562         if (err)
 563                 return err;
 564
 565         list_for_each_entry_safe(p, n, list, l_node) {
 566                 if (strcmp(p->name, ".") != 0 &&
 567                     strcmp(p->name, "..") != 0) {
 568                         err = ovl_cache_update_ino(path, p);
 569                         if (err)
 570                                 return err;
 571                 }
 572                 if (p->ino == p->real_ino) {
 573                         list_del(&p->l_node);
 574                         kfree(p);
 575                 } else {
 576                         struct rb_node **newp = &root->rb_node;
 577                         struct rb_node *parent = NULL;
 578
 579                         if (WARN_ON(ovl_cache_entry_find_link(p->name, p->len,
 580                                                               &newp, &parent)))
 581                                 return -EIO;
 582
 583                         rb_link_node(&p->node, parent, newp);
 584                         rb_insert_color(&p->node, root);
 585                 }
 586         }
 587         return 0;
 588 }
 589
 590 static struct ovl_dir_cache *ovl_cache_get_impure(struct path *path)
 591 {
 592         int res;
 593         struct dentry *dentry = path->dentry;
 594         struct ovl_dir_cache *cache;
 595
 596         cache = ovl_dir_cache(d_inode(dentry));
 597         if (cache && ovl_dentry_version_get(dentry) == cache->version)
 598                 return cache;
 599
 600         /* Impure cache is not refcounted, free it here */
 601         ovl_dir_cache_free(d_inode(dentry));
 602         ovl_set_dir_cache(d_inode(dentry), NULL);
 603
 604         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 605         if (!cache)
 606                 return ERR_PTR(-ENOMEM);
 607
 608         res = ovl_dir_read_impure(path, &cache->entries, &cache->root);
 609         if (res) {
 610                 ovl_cache_free(&cache->entries);
 611                 kfree(cache);
 612                 return ERR_PTR(res);
 613         }
 614         if (list_empty(&cache->entries)) {
 615                 /*
 616                  * A good opportunity to get rid of an unneeded "impure" flag.
 617                  * Removing the "impure" xattr is best effort.
 618                  */
 619                 if (!ovl_want_write(dentry)) {
 620                         ovl_do_removexattr(ovl_dentry_upper(dentry),
 621                                            OVL_XATTR_IMPURE);
 622                         ovl_drop_write(dentry);
 623                 }
 624                 ovl_clear_flag(OVL_IMPURE, d_inode(dentry));
 625                 kfree(cache);
 626                 return NULL;
 627         }
 628
 629         cache->version = ovl_dentry_version_get(dentry);
 630         ovl_set_dir_cache(d_inode(dentry), cache);
 631
 632         return cache;
 633 }
 634
 635 struct ovl_readdir_translate {
 636         struct dir_context *orig_ctx;
 637         struct ovl_dir_cache *cache;
 638         struct dir_context ctx;
 639         u64 parent_ino;
 640         int fsid;
 641         int xinobits;
 642 };
 643
 644 static int ovl_fill_real(struct dir_context *ctx, const char *name,
 645                            int namelen, loff_t offset, u64 ino,
 646                            unsigned int d_type)
 647 {
 648         struct ovl_readdir_translate *rdt =
 649                 container_of(ctx, struct ovl_readdir_translate, ctx);
 650         struct dir_context *orig_ctx = rdt->orig_ctx;
 651
 652         if (rdt->parent_ino && strcmp(name, "..") == 0) {
 653                 ino = rdt->parent_ino;
 654         } else if (rdt->cache) {
 655                 struct ovl_cache_entry *p;
 656
 657                 p = ovl_cache_entry_find(&rdt->cache->root, name, namelen);
 658                 if (p)
 659                         ino = p->ino;
 660         } else if (rdt->xinobits) {
 661                 ino = ovl_remap_lower_ino(ino, rdt->xinobits, rdt->fsid,
 662                                           name, namelen);
 663         }
 664
 665         return orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
 666 }
 667
 668 static bool ovl_is_impure_dir(struct file *file)
 669 {
 670         struct ovl_dir_file *od = file->private_data;
 671         struct inode *dir = d_inode(file->f_path.dentry);
 672
 673         /*
 674          * Only upper dir can be impure, but if we are in the middle of
 675          * iterating a lower real dir, dir could be copied up and marked
 676          * impure. We only want the impure cache if we started iterating
 677          * a real upper dir to begin with.
 678          */
 679         return od->is_upper && ovl_test_flag(OVL_IMPURE, dir);
 680
 681 }
 682
 683 static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
 684 {
 685         int err;
 686         struct ovl_dir_file *od = file->private_data;
 687         struct dentry *dir = file->f_path.dentry;
 688         struct ovl_layer *lower_layer = ovl_layer_lower(dir);
 689         struct ovl_readdir_translate rdt = {
 690                 .ctx.actor = ovl_fill_real,
 691                 .orig_ctx = ctx,
 692                 .xinobits = ovl_xino_bits(dir->d_sb),
 693         };
 694
 695         if (rdt.xinobits && lower_layer)
 696                 rdt.fsid = lower_layer->fsid;
 697
 698         if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
 699                 struct kstat stat;
 700                 struct path statpath = file->f_path;
 701
 702                 statpath.dentry = dir->d_parent;
 703                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 704                 if (err)
 705                         return err;
 706
 707                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 708                 rdt.parent_ino = stat.ino;
 709         }
 710
 711         if (ovl_is_impure_dir(file)) {
 712                 rdt.cache = ovl_cache_get_impure(&file->f_path);
 713                 if (IS_ERR(rdt.cache))
 714                         return PTR_ERR(rdt.cache);
 715         }
 716
 717         err = iterate_dir(od->realfile, &rdt.ctx);
 718         ctx->pos = rdt.ctx.pos;
 719
 720         return err;
 721 }
 722
 723
 724 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 725 {
 726         struct ovl_dir_file *od = file->private_data;
 727         struct dentry *dentry = file->f_path.dentry;
 728         struct ovl_cache_entry *p;
 729         int err;
 730
 731         if (!ctx->pos)
 732                 ovl_dir_reset(file);
 733
 734         if (od->is_real) {
 735                 /*
 736                  * If parent is merge, then need to adjust d_ino for '..', if
 737                  * dir is impure then need to adjust d_ino for copied up
 738                  * entries.
 739                  */
 740                 if (ovl_xino_bits(dentry->d_sb) ||
 741                     (ovl_same_sb(dentry->d_sb) &&
 742                      (ovl_is_impure_dir(file) ||
 743                       OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent))))) {
 744                         return ovl_iterate_real(file, ctx);
 745                 }
 746                 return iterate_dir(od->realfile, ctx);
 747         }
 748
 749         if (!od->cache) {
 750                 struct ovl_dir_cache *cache;
 751
 752                 cache = ovl_cache_get(dentry);
 753                 if (IS_ERR(cache))
 754                         return PTR_ERR(cache);
 755
 756                 od->cache = cache;
 757                 ovl_seek_cursor(od, ctx->pos);
 758         }
 759
 760         while (od->cursor != &od->cache->entries) {
 761                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 762                 if (!p->is_whiteout) {
 763                         if (!p->ino) {
 764                                 err = ovl_cache_update_ino(&file->f_path, p);
 765                                 if (err)
 766                                         return err;
 767                         }
 768                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 769                                 break;
 770                 }
 771                 od->cursor = p->l_node.next;
 772                 ctx->pos++;
 773         }
 774         return 0;
 775 }
 776
 777 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 778 {
 779         loff_t res;
 780         struct ovl_dir_file *od = file->private_data;
 781
 782         inode_lock(file_inode(file));
 783         if (!file->f_pos)
 784                 ovl_dir_reset(file);
 785
 786         if (od->is_real) {
 787                 res = vfs_llseek(od->realfile, offset, origin);
 788                 file->f_pos = od->realfile->f_pos;
 789         } else {
 790                 res = -EINVAL;
 791
 792                 switch (origin) {
 793                 case SEEK_CUR:
 794                         offset += file->f_pos;
 795                         break;
 796                 case SEEK_SET:
 797                         break;
 798                 default:
 799                         goto out_unlock;
 800                 }
 801                 if (offset < 0)
 802                         goto out_unlock;
 803
 804                 if (offset != file->f_pos) {
 805                         file->f_pos = offset;
 806                         if (od->cache)
 807                                 ovl_seek_cursor(od, offset);
 808                 }
 809                 res = offset;
 810         }
 811 out_unlock:
 812         inode_unlock(file_inode(file));
 813
 814         return res;
 815 }
 816
 817 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 818                          int datasync)
 819 {
 820         struct ovl_dir_file *od = file->private_data;
 821         struct dentry *dentry = file->f_path.dentry;
 822         struct file *realfile = od->realfile;
 823
 824         /* Nothing to sync for lower */
 825         if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
 826                 return 0;
 827
 828         /*
 829          * Need to check if we started out being a lower dir, but got copied up
 830          */
 831         if (!od->is_upper) {
 832                 struct inode *inode = file_inode(file);
 833
 834                 realfile = READ_ONCE(od->upperfile);
 835                 if (!realfile) {
 836                         struct path upperpath;
 837
 838                         ovl_path_upper(dentry, &upperpath);
 839                         realfile = ovl_path_open(&upperpath, O_RDONLY);
 840
 841                         inode_lock(inode);
 842                         if (!od->upperfile) {
 843                                 if (IS_ERR(realfile)) {
 844                                         inode_unlock(inode);
 845                                         return PTR_ERR(realfile);
 846                                 }
 847                                 smp_store_release(&od->upperfile, realfile);
 848                         } else {
 849                                 /* somebody has beaten us to it */
 850                                 if (!IS_ERR(realfile))
 851                                         fput(realfile);
 852                                 realfile = od->upperfile;
 853                         }
 854                         inode_unlock(inode);
 855                 }
 856         }
 857
 858         return vfs_fsync_range(realfile, start, end, datasync);
 859 }
 860
 861 static int ovl_dir_release(struct inode *inode, struct file *file)
 862 {
 863         struct ovl_dir_file *od = file->private_data;
 864
 865         if (od->cache) {
 866                 inode_lock(inode);
 867                 ovl_cache_put(od, file->f_path.dentry);
 868                 inode_unlock(inode);
 869         }
 870         fput(od->realfile);
 871         if (od->upperfile)
 872                 fput(od->upperfile);
 873         kfree(od);
 874
 875         return 0;
 876 }
 877
 878 static int ovl_dir_open(struct inode *inode, struct file *file)
 879 {
 880         struct path realpath;
 881         struct file *realfile;
 882         struct ovl_dir_file *od;
 883         enum ovl_path_type type;
 884
 885         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 886         if (!od)
 887                 return -ENOMEM;
 888
 889         type = ovl_path_real(file->f_path.dentry, &realpath);
 890         realfile = ovl_path_open(&realpath, file->f_flags);
 891         if (IS_ERR(realfile)) {
 892                 kfree(od);
 893                 return PTR_ERR(realfile);
 894         }
 895         od->realfile = realfile;
 896         od->is_real = ovl_dir_is_real(file->f_path.dentry);
 897         od->is_upper = OVL_TYPE_UPPER(type);
 898         file->private_data = od;
 899
 900         return 0;
 901 }
 902
 903 const struct file_operations ovl_dir_operations = {
 904         .read           = generic_read_dir,
 905         .open           = ovl_dir_open,
 906         .iterate        = ovl_iterate,
 907         .llseek         = ovl_dir_llseek,
 908         .fsync          = ovl_dir_fsync,
 909         .release        = ovl_dir_release,
 910 };
 911
 912 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 913 {
 914         int err;
 915         struct ovl_cache_entry *p, *n;
 916         struct rb_root root = RB_ROOT;
 917         const struct cred *old_cred;
 918
 919         old_cred = ovl_override_creds(dentry->d_sb);
 920         err = ovl_dir_read_merged(dentry, list, &root);
 921         revert_creds(old_cred);
 922         if (err)
 923                 return err;
 924
 925         err = 0;
 926
 927         list_for_each_entry_safe(p, n, list, l_node) {
 928                 /*
 929                  * Select whiteouts in upperdir, they should
 930                  * be cleared when deleting this directory.
 931                  */
 932                 if (p->is_whiteout) {
 933                         if (p->is_upper)
 934                                 continue;
 935                         goto del_entry;
 936                 }
 937
 938                 if (p->name[0] == '.') {
 939                         if (p->len == 1)
 940                                 goto del_entry;
 941                         if (p->len == 2 && p->name[1] == '.')
 942                                 goto del_entry;
 943                 }
 944                 err = -ENOTEMPTY;
 945                 break;
 946
 947 del_entry:
 948                 list_del(&p->l_node);
 949                 kfree(p);
 950         }
 951
 952         return err;
 953 }
 954
 955 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
 956 {
 957         struct ovl_cache_entry *p;
 958
 959         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
 960         list_for_each_entry(p, list, l_node) {
 961                 struct dentry *dentry;
 962
 963                 if (WARN_ON(!p->is_whiteout || !p->is_upper))
 964                         continue;
 965
 966                 dentry = lookup_one_len(p->name, upper, p->len);
 967                 if (IS_ERR(dentry)) {
 968                         pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
 969                                upper->d_name.name, p->len, p->name,
 970                                (int) PTR_ERR(dentry));
 971                         continue;
 972                 }
 973                 if (dentry->d_inode)
 974                         ovl_cleanup(upper->d_inode, dentry);
 975                 dput(dentry);
 976         }
 977         inode_unlock(upper->d_inode);
 978 }
 979
 980 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
 981                           int namelen, loff_t offset, u64 ino,
 982                           unsigned int d_type)
 983 {
 984         struct ovl_readdir_data *rdd =
 985                 container_of(ctx, struct ovl_readdir_data, ctx);
 986
 987         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
 988         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
 989                 return 0;
 990
 991         if (d_type != DT_UNKNOWN)
 992                 rdd->d_type_supported = true;
 993
 994         return 0;
 995 }
 996
 997 /*
 998  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
 999  * if error is encountered.
1000  */
1001 int ovl_check_d_type_supported(struct path *realpath)
1002 {
1003         int err;
1004         struct ovl_readdir_data rdd = {
1005                 .ctx.actor = ovl_check_d_type,
1006                 .d_type_supported = false,
1007         };
1008
1009         err = ovl_dir_read(realpath, &rdd);
1010         if (err)
1011                 return err;
1012
1013         return rdd.d_type_supported;
1014 }
1015
1016 static void ovl_workdir_cleanup_recurse(struct path *path, int level)
1017 {
1018         int err;
1019         struct inode *dir = path->dentry->d_inode;
1020         LIST_HEAD(list);
1021         struct rb_root root = RB_ROOT;
1022         struct ovl_cache_entry *p;
1023         struct ovl_readdir_data rdd = {
1024                 .ctx.actor = ovl_fill_merge,
1025                 .dentry = NULL,
1026                 .list = &list,
1027                 .root = &root,
1028                 .is_lowest = false,
1029         };
1030
1031         err = ovl_dir_read(path, &rdd);
1032         if (err)
1033                 goto out;
1034
1035         inode_lock_nested(dir, I_MUTEX_PARENT);
1036         list_for_each_entry(p, &list, l_node) {
1037                 struct dentry *dentry;
1038
1039                 if (p->name[0] == '.') {
1040                         if (p->len == 1)
1041                                 continue;
1042                         if (p->len == 2 && p->name[1] == '.')
1043                                 continue;
1044                 }
1045                 dentry = lookup_one_len(p->name, path->dentry, p->len);
1046                 if (IS_ERR(dentry))
1047                         continue;
1048                 if (dentry->d_inode)
1049                         ovl_workdir_cleanup(dir, path->mnt, dentry, level);
1050                 dput(dentry);
1051         }
1052         inode_unlock(dir);
1053 out:
1054         ovl_cache_free(&list);
1055 }
1056
1057 void ovl_workdir_cleanup(struct inode *dir, struct vfsmount *mnt,
1058                          struct dentry *dentry, int level)
1059 {
1060         int err;
1061
1062         if (!d_is_dir(dentry) || level > 1) {
1063                 ovl_cleanup(dir, dentry);
1064                 return;
1065         }
1066
1067         err = ovl_do_rmdir(dir, dentry);
1068         if (err) {
1069                 struct path path = { .mnt = mnt, .dentry = dentry };
1070
1071                 inode_unlock(dir);
1072                 ovl_workdir_cleanup_recurse(&path, level + 1);
1073                 inode_lock_nested(dir, I_MUTEX_PARENT);
1074                 ovl_cleanup(dir, dentry);
1075         }
1076 }
1077
1078 int ovl_indexdir_cleanup(struct ovl_fs *ofs)
1079 {
1080         int err;
1081         struct dentry *indexdir = ofs->indexdir;
1082         struct dentry *index = NULL;
1083         struct inode *dir = indexdir->d_inode;
1084         struct path path = { .mnt = ofs->upper_mnt, .dentry = indexdir };
1085         LIST_HEAD(list);
1086         struct rb_root root = RB_ROOT;
1087         struct ovl_cache_entry *p;
1088         struct ovl_readdir_data rdd = {
1089                 .ctx.actor = ovl_fill_merge,
1090                 .dentry = NULL,
1091                 .list = &list,
1092                 .root = &root,
1093                 .is_lowest = false,
1094         };
1095
1096         err = ovl_dir_read(&path, &rdd);
1097         if (err)
1098                 goto out;
1099
1100         inode_lock_nested(dir, I_MUTEX_PARENT);
1101         list_for_each_entry(p, &list, l_node) {
1102                 if (p->name[0] == '.') {
1103                         if (p->len == 1)
1104                                 continue;
1105                         if (p->len == 2 && p->name[1] == '.')
1106                                 continue;
1107                 }
1108                 index = lookup_one_len(p->name, indexdir, p->len);
1109                 if (IS_ERR(index)) {
1110                         err = PTR_ERR(index);
1111                         index = NULL;
1112                         break;
1113                 }
1114                 err = ovl_verify_index(ofs, index);
1115                 if (!err) {
1116                         goto next;
1117                 } else if (err == -ESTALE) {
1118                         /* Cleanup stale index entries */
1119                         err = ovl_cleanup(dir, index);
1120                 } else if (err != -ENOENT) {
1121                         /*
1122                          * Abort mount to avoid corrupting the index if
1123                          * an incompatible index entry was found or on out
1124                          * of memory.
1125                          */
1126                         break;
1127                 } else if (ofs->config.nfs_export) {
1128                         /*
1129                          * Whiteout orphan index to block future open by
1130                          * handle after overlay nlink dropped to zero.
1131                          */
1132                         err = ovl_cleanup_and_whiteout(indexdir, dir, index);
1133                 } else {
1134                         /* Cleanup orphan index entries */
1135                         err = ovl_cleanup(dir, index);
1136                 }
1137
1138                 if (err)
1139                         break;
1140
1141 next:
1142                 dput(index);
1143                 index = NULL;
1144         }
1145         dput(index);
1146         inode_unlock(dir);
1147 out:
1148         ovl_cache_free(&list);
1149         if (err)
1150                 pr_err("overlayfs: failed index dir cleanup (%i)\n", err);
1151         return err;
1152 }