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_LARGEFILE);
 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, bool warn)
 442 {
 443         unsigned int xinoshift = 64 - xinobits;
 444
 445         if (unlikely(ino >> xinoshift)) {
 446                 if (warn) {
 447                         pr_warn_ratelimited("d_ino too big (%.*s, ino=%llu, xinobits=%d)\n",
 448                                             namelen, name, ino, xinobits);
 449                 }
 450                 return ino;
 451         }
 452
 453         /*
 454          * The lowest xinobit is reserved for mapping the non-peresistent inode
 455          * numbers range, but this range is only exposed via st_ino, not here.
 456          */
 457         return ino | ((u64)fsid) << (xinoshift + 1);
 458 }
 459
 460 /*
 461  * Set d_ino for upper entries. Non-upper entries should always report
 462  * the uppermost real inode ino and should not call this function.
 463  *
 464  * When not all layer are on same fs, report real ino also for upper.
 465  *
 466  * When all layers are on the same fs, and upper has a reference to
 467  * copy up origin, call vfs_getattr() on the overlay entry to make
 468  * sure that d_ino will be consistent with st_ino from stat(2).
 469  */
 470 static int ovl_cache_update_ino(struct path *path, struct ovl_cache_entry *p)
 471
 472 {
 473         struct dentry *dir = path->dentry;
 474         struct dentry *this = NULL;
 475         enum ovl_path_type type;
 476         u64 ino = p->real_ino;
 477         int xinobits = ovl_xino_bits(dir->d_sb);
 478         int err = 0;
 479
 480         if (!ovl_same_dev(dir->d_sb))
 481                 goto out;
 482
 483         if (p->name[0] == '.') {
 484                 if (p->len == 1) {
 485                         this = dget(dir);
 486                         goto get;
 487                 }
 488                 if (p->len == 2 && p->name[1] == '.') {
 489                         /* we shall not be moved */
 490                         this = dget(dir->d_parent);
 491                         goto get;
 492                 }
 493         }
 494         this = lookup_one_len(p->name, dir, p->len);
 495         if (IS_ERR_OR_NULL(this) || !this->d_inode) {
 496                 if (IS_ERR(this)) {
 497                         err = PTR_ERR(this);
 498                         this = NULL;
 499                         goto fail;
 500                 }
 501                 goto out;
 502         }
 503
 504 get:
 505         type = ovl_path_type(this);
 506         if (OVL_TYPE_ORIGIN(type)) {
 507                 struct kstat stat;
 508                 struct path statpath = *path;
 509
 510                 statpath.dentry = this;
 511                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 512                 if (err)
 513                         goto fail;
 514
 515                 /*
 516                  * Directory inode is always on overlay st_dev.
 517                  * Non-dir with ovl_same_dev() could be on pseudo st_dev in case
 518                  * of xino bits overflow.
 519                  */
 520                 WARN_ON_ONCE(S_ISDIR(stat.mode) &&
 521                              dir->d_sb->s_dev != stat.dev);
 522                 ino = stat.ino;
 523         } else if (xinobits && !OVL_TYPE_UPPER(type)) {
 524                 ino = ovl_remap_lower_ino(ino, xinobits,
 525                                           ovl_layer_lower(this)->fsid,
 526                                           p->name, p->len,
 527                                           ovl_xino_warn(dir->d_sb));
 528         }
 529
 530 out:
 531         p->ino = ino;
 532         dput(this);
 533         return err;
 534
 535 fail:
 536         pr_warn_ratelimited("failed to look up (%s) for ino (%i)\n",
 537                             p->name, err);
 538         goto out;
 539 }
 540
 541 static int ovl_fill_plain(struct dir_context *ctx, const char *name,
 542                           int namelen, loff_t offset, u64 ino,
 543                           unsigned int d_type)
 544 {
 545         struct ovl_cache_entry *p;
 546         struct ovl_readdir_data *rdd =
 547                 container_of(ctx, struct ovl_readdir_data, ctx);
 548
 549         rdd->count++;
 550         p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 551         if (p == NULL) {
 552                 rdd->err = -ENOMEM;
 553                 return -ENOMEM;
 554         }
 555         list_add_tail(&p->l_node, rdd->list);
 556
 557         return 0;
 558 }
 559
 560 static int ovl_dir_read_impure(struct path *path,  struct list_head *list,
 561                                struct rb_root *root)
 562 {
 563         int err;
 564         struct path realpath;
 565         struct ovl_cache_entry *p, *n;
 566         struct ovl_readdir_data rdd = {
 567                 .ctx.actor = ovl_fill_plain,
 568                 .list = list,
 569                 .root = root,
 570         };
 571
 572         INIT_LIST_HEAD(list);
 573         *root = RB_ROOT;
 574         ovl_path_upper(path->dentry, &realpath);
 575
 576         err = ovl_dir_read(&realpath, &rdd);
 577         if (err)
 578                 return err;
 579
 580         list_for_each_entry_safe(p, n, list, l_node) {
 581                 if (strcmp(p->name, ".") != 0 &&
 582                     strcmp(p->name, "..") != 0) {
 583                         err = ovl_cache_update_ino(path, p);
 584                         if (err)
 585                                 return err;
 586                 }
 587                 if (p->ino == p->real_ino) {
 588                         list_del(&p->l_node);
 589                         kfree(p);
 590                 } else {
 591                         struct rb_node **newp = &root->rb_node;
 592                         struct rb_node *parent = NULL;
 593
 594                         if (WARN_ON(ovl_cache_entry_find_link(p->name, p->len,
 595                                                               &newp, &parent)))
 596                                 return -EIO;
 597
 598                         rb_link_node(&p->node, parent, newp);
 599                         rb_insert_color(&p->node, root);
 600                 }
 601         }
 602         return 0;
 603 }
 604
 605 static struct ovl_dir_cache *ovl_cache_get_impure(struct path *path)
 606 {
 607         int res;
 608         struct dentry *dentry = path->dentry;
 609         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 610         struct ovl_dir_cache *cache;
 611
 612         cache = ovl_dir_cache(d_inode(dentry));
 613         if (cache && ovl_dentry_version_get(dentry) == cache->version)
 614                 return cache;
 615
 616         /* Impure cache is not refcounted, free it here */
 617         ovl_dir_cache_free(d_inode(dentry));
 618         ovl_set_dir_cache(d_inode(dentry), NULL);
 619
 620         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 621         if (!cache)
 622                 return ERR_PTR(-ENOMEM);
 623
 624         res = ovl_dir_read_impure(path, &cache->entries, &cache->root);
 625         if (res) {
 626                 ovl_cache_free(&cache->entries);
 627                 kfree(cache);
 628                 return ERR_PTR(res);
 629         }
 630         if (list_empty(&cache->entries)) {
 631                 /*
 632                  * A good opportunity to get rid of an unneeded "impure" flag.
 633                  * Removing the "impure" xattr is best effort.
 634                  */
 635                 if (!ovl_want_write(dentry)) {
 636                         ovl_do_removexattr(ofs, ovl_dentry_upper(dentry),
 637                                            OVL_XATTR_IMPURE);
 638                         ovl_drop_write(dentry);
 639                 }
 640                 ovl_clear_flag(OVL_IMPURE, d_inode(dentry));
 641                 kfree(cache);
 642                 return NULL;
 643         }
 644
 645         cache->version = ovl_dentry_version_get(dentry);
 646         ovl_set_dir_cache(d_inode(dentry), cache);
 647
 648         return cache;
 649 }
 650
 651 struct ovl_readdir_translate {
 652         struct dir_context *orig_ctx;
 653         struct ovl_dir_cache *cache;
 654         struct dir_context ctx;
 655         u64 parent_ino;
 656         int fsid;
 657         int xinobits;
 658         bool xinowarn;
 659 };
 660
 661 static int ovl_fill_real(struct dir_context *ctx, const char *name,
 662                            int namelen, loff_t offset, u64 ino,
 663                            unsigned int d_type)
 664 {
 665         struct ovl_readdir_translate *rdt =
 666                 container_of(ctx, struct ovl_readdir_translate, ctx);
 667         struct dir_context *orig_ctx = rdt->orig_ctx;
 668
 669         if (rdt->parent_ino && strcmp(name, "..") == 0) {
 670                 ino = rdt->parent_ino;
 671         } else if (rdt->cache) {
 672                 struct ovl_cache_entry *p;
 673
 674                 p = ovl_cache_entry_find(&rdt->cache->root, name, namelen);
 675                 if (p)
 676                         ino = p->ino;
 677         } else if (rdt->xinobits) {
 678                 ino = ovl_remap_lower_ino(ino, rdt->xinobits, rdt->fsid,
 679                                           name, namelen, rdt->xinowarn);
 680         }
 681
 682         return orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
 683 }
 684
 685 static bool ovl_is_impure_dir(struct file *file)
 686 {
 687         struct ovl_dir_file *od = file->private_data;
 688         struct inode *dir = d_inode(file->f_path.dentry);
 689
 690         /*
 691          * Only upper dir can be impure, but if we are in the middle of
 692          * iterating a lower real dir, dir could be copied up and marked
 693          * impure. We only want the impure cache if we started iterating
 694          * a real upper dir to begin with.
 695          */
 696         return od->is_upper && ovl_test_flag(OVL_IMPURE, dir);
 697
 698 }
 699
 700 static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
 701 {
 702         int err;
 703         struct ovl_dir_file *od = file->private_data;
 704         struct dentry *dir = file->f_path.dentry;
 705         const struct ovl_layer *lower_layer = ovl_layer_lower(dir);
 706         struct ovl_readdir_translate rdt = {
 707                 .ctx.actor = ovl_fill_real,
 708                 .orig_ctx = ctx,
 709                 .xinobits = ovl_xino_bits(dir->d_sb),
 710                 .xinowarn = ovl_xino_warn(dir->d_sb),
 711         };
 712
 713         if (rdt.xinobits && lower_layer)
 714                 rdt.fsid = lower_layer->fsid;
 715
 716         if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
 717                 struct kstat stat;
 718                 struct path statpath = file->f_path;
 719
 720                 statpath.dentry = dir->d_parent;
 721                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 722                 if (err)
 723                         return err;
 724
 725                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 726                 rdt.parent_ino = stat.ino;
 727         }
 728
 729         if (ovl_is_impure_dir(file)) {
 730                 rdt.cache = ovl_cache_get_impure(&file->f_path);
 731                 if (IS_ERR(rdt.cache))
 732                         return PTR_ERR(rdt.cache);
 733         }
 734
 735         err = iterate_dir(od->realfile, &rdt.ctx);
 736         ctx->pos = rdt.ctx.pos;
 737
 738         return err;
 739 }
 740
 741
 742 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 743 {
 744         struct ovl_dir_file *od = file->private_data;
 745         struct dentry *dentry = file->f_path.dentry;
 746         struct ovl_cache_entry *p;
 747         const struct cred *old_cred;
 748         int err;
 749
 750         old_cred = ovl_override_creds(dentry->d_sb);
 751         if (!ctx->pos)
 752                 ovl_dir_reset(file);
 753
 754         if (od->is_real) {
 755                 /*
 756                  * If parent is merge, then need to adjust d_ino for '..', if
 757                  * dir is impure then need to adjust d_ino for copied up
 758                  * entries.
 759                  */
 760                 if (ovl_xino_bits(dentry->d_sb) ||
 761                     (ovl_same_fs(dentry->d_sb) &&
 762                      (ovl_is_impure_dir(file) ||
 763                       OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent))))) {
 764                         err = ovl_iterate_real(file, ctx);
 765                 } else {
 766                         err = iterate_dir(od->realfile, ctx);
 767                 }
 768                 goto out;
 769         }
 770
 771         if (!od->cache) {
 772                 struct ovl_dir_cache *cache;
 773
 774                 cache = ovl_cache_get(dentry);
 775                 err = PTR_ERR(cache);
 776                 if (IS_ERR(cache))
 777                         goto out;
 778
 779                 od->cache = cache;
 780                 ovl_seek_cursor(od, ctx->pos);
 781         }
 782
 783         while (od->cursor != &od->cache->entries) {
 784                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 785                 if (!p->is_whiteout) {
 786                         if (!p->ino) {
 787                                 err = ovl_cache_update_ino(&file->f_path, p);
 788                                 if (err)
 789                                         goto out;
 790                         }
 791                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 792                                 break;
 793                 }
 794                 od->cursor = p->l_node.next;
 795                 ctx->pos++;
 796         }
 797         err = 0;
 798 out:
 799         revert_creds(old_cred);
 800         return err;
 801 }
 802
 803 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 804 {
 805         loff_t res;
 806         struct ovl_dir_file *od = file->private_data;
 807
 808         inode_lock(file_inode(file));
 809         if (!file->f_pos)
 810                 ovl_dir_reset(file);
 811
 812         if (od->is_real) {
 813                 res = vfs_llseek(od->realfile, offset, origin);
 814                 file->f_pos = od->realfile->f_pos;
 815         } else {
 816                 res = -EINVAL;
 817
 818                 switch (origin) {
 819                 case SEEK_CUR:
 820                         offset += file->f_pos;
 821                         break;
 822                 case SEEK_SET:
 823                         break;
 824                 default:
 825                         goto out_unlock;
 826                 }
 827                 if (offset < 0)
 828                         goto out_unlock;
 829
 830                 if (offset != file->f_pos) {
 831                         file->f_pos = offset;
 832                         if (od->cache)
 833                                 ovl_seek_cursor(od, offset);
 834                 }
 835                 res = offset;
 836         }
 837 out_unlock:
 838         inode_unlock(file_inode(file));
 839
 840         return res;
 841 }
 842
 843 static struct file *ovl_dir_open_realfile(const struct file *file,
 844                                           struct path *realpath)
 845 {
 846         struct file *res;
 847         const struct cred *old_cred;
 848
 849         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 850         res = ovl_path_open(realpath, O_RDONLY | (file->f_flags & O_LARGEFILE));
 851         revert_creds(old_cred);
 852
 853         return res;
 854 }
 855
 856 /*
 857  * Like ovl_real_fdget(), returns upperfile if dir was copied up since open.
 858  * Unlike ovl_real_fdget(), this caches upperfile in file->private_data.
 859  *
 860  * TODO: use same abstract type for file->private_data of dir and file so
 861  * upperfile could also be cached for files as well.
 862  */
 863 struct file *ovl_dir_real_file(const struct file *file, bool want_upper)
 864 {
 865
 866         struct ovl_dir_file *od = file->private_data;
 867         struct dentry *dentry = file->f_path.dentry;
 868         struct file *old, *realfile = od->realfile;
 869
 870         if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
 871                 return want_upper ? NULL : realfile;
 872
 873         /*
 874          * Need to check if we started out being a lower dir, but got copied up
 875          */
 876         if (!od->is_upper) {
 877                 realfile = READ_ONCE(od->upperfile);
 878                 if (!realfile) {
 879                         struct path upperpath;
 880
 881                         ovl_path_upper(dentry, &upperpath);
 882                         realfile = ovl_dir_open_realfile(file, &upperpath);
 883                         if (IS_ERR(realfile))
 884                                 return realfile;
 885
 886                         old = cmpxchg_release(&od->upperfile, NULL, realfile);
 887                         if (old) {
 888                                 fput(realfile);
 889                                 realfile = old;
 890                         }
 891                 }
 892         }
 893
 894         return realfile;
 895 }
 896
 897 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 898                          int datasync)
 899 {
 900         struct file *realfile;
 901         int err;
 902
 903         err = ovl_sync_status(OVL_FS(file->f_path.dentry->d_sb));
 904         if (err <= 0)
 905                 return err;
 906
 907         realfile = ovl_dir_real_file(file, true);
 908         err = PTR_ERR_OR_ZERO(realfile);
 909
 910         /* Nothing to sync for lower */
 911         if (!realfile || err)
 912                 return err;
 913
 914         return vfs_fsync_range(realfile, start, end, datasync);
 915 }
 916
 917 static int ovl_dir_release(struct inode *inode, struct file *file)
 918 {
 919         struct ovl_dir_file *od = file->private_data;
 920
 921         if (od->cache) {
 922                 inode_lock(inode);
 923                 ovl_cache_put(od, file->f_path.dentry);
 924                 inode_unlock(inode);
 925         }
 926         fput(od->realfile);
 927         if (od->upperfile)
 928                 fput(od->upperfile);
 929         kfree(od);
 930
 931         return 0;
 932 }
 933
 934 static int ovl_dir_open(struct inode *inode, struct file *file)
 935 {
 936         struct path realpath;
 937         struct file *realfile;
 938         struct ovl_dir_file *od;
 939         enum ovl_path_type type;
 940
 941         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 942         if (!od)
 943                 return -ENOMEM;
 944
 945         type = ovl_path_real(file->f_path.dentry, &realpath);
 946         realfile = ovl_dir_open_realfile(file, &realpath);
 947         if (IS_ERR(realfile)) {
 948                 kfree(od);
 949                 return PTR_ERR(realfile);
 950         }
 951         od->realfile = realfile;
 952         od->is_real = ovl_dir_is_real(file->f_path.dentry);
 953         od->is_upper = OVL_TYPE_UPPER(type);
 954         file->private_data = od;
 955
 956         return 0;
 957 }
 958
 959 const struct file_operations ovl_dir_operations = {
 960         .read           = generic_read_dir,
 961         .open           = ovl_dir_open,
 962         .iterate        = ovl_iterate,
 963         .llseek         = ovl_dir_llseek,
 964         .fsync          = ovl_dir_fsync,
 965         .release        = ovl_dir_release,
 966 };
 967
 968 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 969 {
 970         int err;
 971         struct ovl_cache_entry *p, *n;
 972         struct rb_root root = RB_ROOT;
 973         const struct cred *old_cred;
 974
 975         old_cred = ovl_override_creds(dentry->d_sb);
 976         err = ovl_dir_read_merged(dentry, list, &root);
 977         revert_creds(old_cred);
 978         if (err)
 979                 return err;
 980
 981         err = 0;
 982
 983         list_for_each_entry_safe(p, n, list, l_node) {
 984                 /*
 985                  * Select whiteouts in upperdir, they should
 986                  * be cleared when deleting this directory.
 987                  */
 988                 if (p->is_whiteout) {
 989                         if (p->is_upper)
 990                                 continue;
 991                         goto del_entry;
 992                 }
 993
 994                 if (p->name[0] == '.') {
 995                         if (p->len == 1)
 996                                 goto del_entry;
 997                         if (p->len == 2 && p->name[1] == '.')
 998                                 goto del_entry;
 999                 }
1000                 err = -ENOTEMPTY;
1001                 break;
1002
1003 del_entry:
1004                 list_del(&p->l_node);
1005                 kfree(p);
1006         }
1007
1008         return err;
1009 }
1010
1011 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
1012 {
1013         struct ovl_cache_entry *p;
1014
1015         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
1016         list_for_each_entry(p, list, l_node) {
1017                 struct dentry *dentry;
1018
1019                 if (WARN_ON(!p->is_whiteout || !p->is_upper))
1020                         continue;
1021
1022                 dentry = lookup_one_len(p->name, upper, p->len);
1023                 if (IS_ERR(dentry)) {
1024                         pr_err("lookup '%s/%.*s' failed (%i)\n",
1025                                upper->d_name.name, p->len, p->name,
1026                                (int) PTR_ERR(dentry));
1027                         continue;
1028                 }
1029                 if (dentry->d_inode)
1030                         ovl_cleanup(upper->d_inode, dentry);
1031                 dput(dentry);
1032         }
1033         inode_unlock(upper->d_inode);
1034 }
1035
1036 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
1037                           int namelen, loff_t offset, u64 ino,
1038                           unsigned int d_type)
1039 {
1040         struct ovl_readdir_data *rdd =
1041                 container_of(ctx, struct ovl_readdir_data, ctx);
1042
1043         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
1044         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
1045                 return 0;
1046
1047         if (d_type != DT_UNKNOWN)
1048                 rdd->d_type_supported = true;
1049
1050         return 0;
1051 }
1052
1053 /*
1054  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
1055  * if error is encountered.
1056  */
1057 int ovl_check_d_type_supported(struct path *realpath)
1058 {
1059         int err;
1060         struct ovl_readdir_data rdd = {
1061                 .ctx.actor = ovl_check_d_type,
1062                 .d_type_supported = false,
1063         };
1064
1065         err = ovl_dir_read(realpath, &rdd);
1066         if (err)
1067                 return err;
1068
1069         return rdd.d_type_supported;
1070 }
1071
1072 #define OVL_INCOMPATDIR_NAME "incompat"
1073
1074 static int ovl_workdir_cleanup_recurse(struct path *path, int level)
1075 {
1076         int err;
1077         struct inode *dir = path->dentry->d_inode;
1078         LIST_HEAD(list);
1079         struct rb_root root = RB_ROOT;
1080         struct ovl_cache_entry *p;
1081         struct ovl_readdir_data rdd = {
1082                 .ctx.actor = ovl_fill_merge,
1083                 .dentry = NULL,
1084                 .list = &list,
1085                 .root = &root,
1086                 .is_lowest = false,
1087         };
1088         bool incompat = false;
1089
1090         /*
1091          * The "work/incompat" directory is treated specially - if it is not
1092          * empty, instead of printing a generic error and mounting read-only,
1093          * we will error about incompat features and fail the mount.
1094          *
1095          * When called from ovl_indexdir_cleanup(), path->dentry->d_name.name
1096          * starts with '#'.
1097          */
1098         if (level == 2 &&
1099             !strcmp(path->dentry->d_name.name, OVL_INCOMPATDIR_NAME))
1100                 incompat = true;
1101
1102         err = ovl_dir_read(path, &rdd);
1103         if (err)
1104                 goto out;
1105
1106         inode_lock_nested(dir, I_MUTEX_PARENT);
1107         list_for_each_entry(p, &list, l_node) {
1108                 struct dentry *dentry;
1109
1110                 if (p->name[0] == '.') {
1111                         if (p->len == 1)
1112                                 continue;
1113                         if (p->len == 2 && p->name[1] == '.')
1114                                 continue;
1115                 } else if (incompat) {
1116                         pr_err("overlay with incompat feature '%s' cannot be mounted\n",
1117                                 p->name);
1118                         err = -EINVAL;
1119                         break;
1120                 }
1121                 dentry = lookup_one_len(p->name, path->dentry, p->len);
1122                 if (IS_ERR(dentry))
1123                         continue;
1124                 if (dentry->d_inode)
1125                         err = ovl_workdir_cleanup(dir, path->mnt, dentry, level);
1126                 dput(dentry);
1127                 if (err)
1128                         break;
1129         }
1130         inode_unlock(dir);
1131 out:
1132         ovl_cache_free(&list);
1133         return err;
1134 }
1135
1136 int ovl_workdir_cleanup(struct inode *dir, struct vfsmount *mnt,
1137                          struct dentry *dentry, int level)
1138 {
1139         int err;
1140
1141         if (!d_is_dir(dentry) || level > 1) {
1142                 return ovl_cleanup(dir, dentry);
1143         }
1144
1145         err = ovl_do_rmdir(dir, dentry);
1146         if (err) {
1147                 struct path path = { .mnt = mnt, .dentry = dentry };
1148
1149                 inode_unlock(dir);
1150                 err = ovl_workdir_cleanup_recurse(&path, level + 1);
1151                 inode_lock_nested(dir, I_MUTEX_PARENT);
1152                 if (!err)
1153                         err = ovl_cleanup(dir, dentry);
1154         }
1155
1156         return err;
1157 }
1158
1159 int ovl_indexdir_cleanup(struct ovl_fs *ofs)
1160 {
1161         int err;
1162         struct dentry *indexdir = ofs->indexdir;
1163         struct dentry *index = NULL;
1164         struct inode *dir = indexdir->d_inode;
1165         struct path path = { .mnt = ovl_upper_mnt(ofs), .dentry = indexdir };
1166         LIST_HEAD(list);
1167         struct rb_root root = RB_ROOT;
1168         struct ovl_cache_entry *p;
1169         struct ovl_readdir_data rdd = {
1170                 .ctx.actor = ovl_fill_merge,
1171                 .dentry = NULL,
1172                 .list = &list,
1173                 .root = &root,
1174                 .is_lowest = false,
1175         };
1176
1177         err = ovl_dir_read(&path, &rdd);
1178         if (err)
1179                 goto out;
1180
1181         inode_lock_nested(dir, I_MUTEX_PARENT);
1182         list_for_each_entry(p, &list, l_node) {
1183                 if (p->name[0] == '.') {
1184                         if (p->len == 1)
1185                                 continue;
1186                         if (p->len == 2 && p->name[1] == '.')
1187                                 continue;
1188                 }
1189                 index = lookup_one_len(p->name, indexdir, p->len);
1190                 if (IS_ERR(index)) {
1191                         err = PTR_ERR(index);
1192                         index = NULL;
1193                         break;
1194                 }
1195                 /* Cleanup leftover from index create/cleanup attempt */
1196                 if (index->d_name.name[0] == '#') {
1197                         err = ovl_workdir_cleanup(dir, path.mnt, index, 1);
1198                         if (err)
1199                                 break;
1200                         goto next;
1201                 }
1202                 err = ovl_verify_index(ofs, index);
1203                 if (!err) {
1204                         goto next;
1205                 } else if (err == -ESTALE) {
1206                         /* Cleanup stale index entries */
1207                         err = ovl_cleanup(dir, index);
1208                 } else if (err != -ENOENT) {
1209                         /*
1210                          * Abort mount to avoid corrupting the index if
1211                          * an incompatible index entry was found or on out
1212                          * of memory.
1213                          */
1214                         break;
1215                 } else if (ofs->config.nfs_export) {
1216                         /*
1217                          * Whiteout orphan index to block future open by
1218                          * handle after overlay nlink dropped to zero.
1219                          */
1220                         err = ovl_cleanup_and_whiteout(ofs, dir, index);
1221                 } else {
1222                         /* Cleanup orphan index entries */
1223                         err = ovl_cleanup(dir, index);
1224                 }
1225
1226                 if (err)
1227                         break;
1228
1229 next:
1230                 dput(index);
1231                 index = NULL;
1232         }
1233         dput(index);
1234         inode_unlock(dir);
1235 out:
1236         ovl_cache_free(&list);
1237         if (err)
1238                 pr_err("failed index dir cleanup (%i)\n", err);
1239         return err;
1240 }