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_dir_cache *cache;
 610
 611         cache = ovl_dir_cache(d_inode(dentry));
 612         if (cache && ovl_dentry_version_get(dentry) == cache->version)
 613                 return cache;
 614
 615         /* Impure cache is not refcounted, free it here */
 616         ovl_dir_cache_free(d_inode(dentry));
 617         ovl_set_dir_cache(d_inode(dentry), NULL);
 618
 619         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 620         if (!cache)
 621                 return ERR_PTR(-ENOMEM);
 622
 623         res = ovl_dir_read_impure(path, &cache->entries, &cache->root);
 624         if (res) {
 625                 ovl_cache_free(&cache->entries);
 626                 kfree(cache);
 627                 return ERR_PTR(res);
 628         }
 629         if (list_empty(&cache->entries)) {
 630                 /*
 631                  * A good opportunity to get rid of an unneeded "impure" flag.
 632                  * Removing the "impure" xattr is best effort.
 633                  */
 634                 if (!ovl_want_write(dentry)) {
 635                         ovl_do_removexattr(ovl_dentry_upper(dentry),
 636                                            OVL_XATTR_IMPURE);
 637                         ovl_drop_write(dentry);
 638                 }
 639                 ovl_clear_flag(OVL_IMPURE, d_inode(dentry));
 640                 kfree(cache);
 641                 return NULL;
 642         }
 643
 644         cache->version = ovl_dentry_version_get(dentry);
 645         ovl_set_dir_cache(d_inode(dentry), cache);
 646
 647         return cache;
 648 }
 649
 650 struct ovl_readdir_translate {
 651         struct dir_context *orig_ctx;
 652         struct ovl_dir_cache *cache;
 653         struct dir_context ctx;
 654         u64 parent_ino;
 655         int fsid;
 656         int xinobits;
 657         bool xinowarn;
 658 };
 659
 660 static int ovl_fill_real(struct dir_context *ctx, const char *name,
 661                            int namelen, loff_t offset, u64 ino,
 662                            unsigned int d_type)
 663 {
 664         struct ovl_readdir_translate *rdt =
 665                 container_of(ctx, struct ovl_readdir_translate, ctx);
 666         struct dir_context *orig_ctx = rdt->orig_ctx;
 667
 668         if (rdt->parent_ino && strcmp(name, "..") == 0) {
 669                 ino = rdt->parent_ino;
 670         } else if (rdt->cache) {
 671                 struct ovl_cache_entry *p;
 672
 673                 p = ovl_cache_entry_find(&rdt->cache->root, name, namelen);
 674                 if (p)
 675                         ino = p->ino;
 676         } else if (rdt->xinobits) {
 677                 ino = ovl_remap_lower_ino(ino, rdt->xinobits, rdt->fsid,
 678                                           name, namelen, rdt->xinowarn);
 679         }
 680
 681         return orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
 682 }
 683
 684 static bool ovl_is_impure_dir(struct file *file)
 685 {
 686         struct ovl_dir_file *od = file->private_data;
 687         struct inode *dir = d_inode(file->f_path.dentry);
 688
 689         /*
 690          * Only upper dir can be impure, but if we are in the middle of
 691          * iterating a lower real dir, dir could be copied up and marked
 692          * impure. We only want the impure cache if we started iterating
 693          * a real upper dir to begin with.
 694          */
 695         return od->is_upper && ovl_test_flag(OVL_IMPURE, dir);
 696
 697 }
 698
 699 static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
 700 {
 701         int err;
 702         struct ovl_dir_file *od = file->private_data;
 703         struct dentry *dir = file->f_path.dentry;
 704         const struct ovl_layer *lower_layer = ovl_layer_lower(dir);
 705         struct ovl_readdir_translate rdt = {
 706                 .ctx.actor = ovl_fill_real,
 707                 .orig_ctx = ctx,
 708                 .xinobits = ovl_xino_bits(dir->d_sb),
 709                 .xinowarn = ovl_xino_warn(dir->d_sb),
 710         };
 711
 712         if (rdt.xinobits && lower_layer)
 713                 rdt.fsid = lower_layer->fsid;
 714
 715         if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
 716                 struct kstat stat;
 717                 struct path statpath = file->f_path;
 718
 719                 statpath.dentry = dir->d_parent;
 720                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 721                 if (err)
 722                         return err;
 723
 724                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 725                 rdt.parent_ino = stat.ino;
 726         }
 727
 728         if (ovl_is_impure_dir(file)) {
 729                 rdt.cache = ovl_cache_get_impure(&file->f_path);
 730                 if (IS_ERR(rdt.cache))
 731                         return PTR_ERR(rdt.cache);
 732         }
 733
 734         err = iterate_dir(od->realfile, &rdt.ctx);
 735         ctx->pos = rdt.ctx.pos;
 736
 737         return err;
 738 }
 739
 740
 741 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 742 {
 743         struct ovl_dir_file *od = file->private_data;
 744         struct dentry *dentry = file->f_path.dentry;
 745         struct ovl_cache_entry *p;
 746         const struct cred *old_cred;
 747         int err;
 748
 749         old_cred = ovl_override_creds(dentry->d_sb);
 750         if (!ctx->pos)
 751                 ovl_dir_reset(file);
 752
 753         if (od->is_real) {
 754                 /*
 755                  * If parent is merge, then need to adjust d_ino for '..', if
 756                  * dir is impure then need to adjust d_ino for copied up
 757                  * entries.
 758                  */
 759                 if (ovl_xino_bits(dentry->d_sb) ||
 760                     (ovl_same_fs(dentry->d_sb) &&
 761                      (ovl_is_impure_dir(file) ||
 762                       OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent))))) {
 763                         err = ovl_iterate_real(file, ctx);
 764                 } else {
 765                         err = iterate_dir(od->realfile, ctx);
 766                 }
 767                 goto out;
 768         }
 769
 770         if (!od->cache) {
 771                 struct ovl_dir_cache *cache;
 772
 773                 cache = ovl_cache_get(dentry);
 774                 err = PTR_ERR(cache);
 775                 if (IS_ERR(cache))
 776                         goto out;
 777
 778                 od->cache = cache;
 779                 ovl_seek_cursor(od, ctx->pos);
 780         }
 781
 782         while (od->cursor != &od->cache->entries) {
 783                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 784                 if (!p->is_whiteout) {
 785                         if (!p->ino) {
 786                                 err = ovl_cache_update_ino(&file->f_path, p);
 787                                 if (err)
 788                                         goto out;
 789                         }
 790                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 791                                 break;
 792                 }
 793                 od->cursor = p->l_node.next;
 794                 ctx->pos++;
 795         }
 796         err = 0;
 797 out:
 798         revert_creds(old_cred);
 799         return err;
 800 }
 801
 802 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 803 {
 804         loff_t res;
 805         struct ovl_dir_file *od = file->private_data;
 806
 807         inode_lock(file_inode(file));
 808         if (!file->f_pos)
 809                 ovl_dir_reset(file);
 810
 811         if (od->is_real) {
 812                 res = vfs_llseek(od->realfile, offset, origin);
 813                 file->f_pos = od->realfile->f_pos;
 814         } else {
 815                 res = -EINVAL;
 816
 817                 switch (origin) {
 818                 case SEEK_CUR:
 819                         offset += file->f_pos;
 820                         break;
 821                 case SEEK_SET:
 822                         break;
 823                 default:
 824                         goto out_unlock;
 825                 }
 826                 if (offset < 0)
 827                         goto out_unlock;
 828
 829                 if (offset != file->f_pos) {
 830                         file->f_pos = offset;
 831                         if (od->cache)
 832                                 ovl_seek_cursor(od, offset);
 833                 }
 834                 res = offset;
 835         }
 836 out_unlock:
 837         inode_unlock(file_inode(file));
 838
 839         return res;
 840 }
 841
 842 static struct file *ovl_dir_open_realfile(struct file *file,
 843                                           struct path *realpath)
 844 {
 845         struct file *res;
 846         const struct cred *old_cred;
 847
 848         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 849         res = ovl_path_open(realpath, O_RDONLY | (file->f_flags & O_LARGEFILE));
 850         revert_creds(old_cred);
 851
 852         return res;
 853 }
 854
 855 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 856                          int datasync)
 857 {
 858         struct ovl_dir_file *od = file->private_data;
 859         struct dentry *dentry = file->f_path.dentry;
 860         struct file *realfile = od->realfile;
 861
 862         /* Nothing to sync for lower */
 863         if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
 864                 return 0;
 865
 866         /*
 867          * Need to check if we started out being a lower dir, but got copied up
 868          */
 869         if (!od->is_upper) {
 870                 struct inode *inode = file_inode(file);
 871
 872                 realfile = READ_ONCE(od->upperfile);
 873                 if (!realfile) {
 874                         struct path upperpath;
 875
 876                         ovl_path_upper(dentry, &upperpath);
 877                         realfile = ovl_dir_open_realfile(file, &upperpath);
 878
 879                         inode_lock(inode);
 880                         if (!od->upperfile) {
 881                                 if (IS_ERR(realfile)) {
 882                                         inode_unlock(inode);
 883                                         return PTR_ERR(realfile);
 884                                 }
 885                                 smp_store_release(&od->upperfile, realfile);
 886                         } else {
 887                                 /* somebody has beaten us to it */
 888                                 if (!IS_ERR(realfile))
 889                                         fput(realfile);
 890                                 realfile = od->upperfile;
 891                         }
 892                         inode_unlock(inode);
 893                 }
 894         }
 895
 896         return vfs_fsync_range(realfile, start, end, datasync);
 897 }
 898
 899 static int ovl_dir_release(struct inode *inode, struct file *file)
 900 {
 901         struct ovl_dir_file *od = file->private_data;
 902
 903         if (od->cache) {
 904                 inode_lock(inode);
 905                 ovl_cache_put(od, file->f_path.dentry);
 906                 inode_unlock(inode);
 907         }
 908         fput(od->realfile);
 909         if (od->upperfile)
 910                 fput(od->upperfile);
 911         kfree(od);
 912
 913         return 0;
 914 }
 915
 916 static int ovl_dir_open(struct inode *inode, struct file *file)
 917 {
 918         struct path realpath;
 919         struct file *realfile;
 920         struct ovl_dir_file *od;
 921         enum ovl_path_type type;
 922
 923         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 924         if (!od)
 925                 return -ENOMEM;
 926
 927         type = ovl_path_real(file->f_path.dentry, &realpath);
 928         realfile = ovl_dir_open_realfile(file, &realpath);
 929         if (IS_ERR(realfile)) {
 930                 kfree(od);
 931                 return PTR_ERR(realfile);
 932         }
 933         od->realfile = realfile;
 934         od->is_real = ovl_dir_is_real(file->f_path.dentry);
 935         od->is_upper = OVL_TYPE_UPPER(type);
 936         file->private_data = od;
 937
 938         return 0;
 939 }
 940
 941 const struct file_operations ovl_dir_operations = {
 942         .read           = generic_read_dir,
 943         .open           = ovl_dir_open,
 944         .iterate        = ovl_iterate,
 945         .llseek         = ovl_dir_llseek,
 946         .fsync          = ovl_dir_fsync,
 947         .release        = ovl_dir_release,
 948 };
 949
 950 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 951 {
 952         int err;
 953         struct ovl_cache_entry *p, *n;
 954         struct rb_root root = RB_ROOT;
 955         const struct cred *old_cred;
 956
 957         old_cred = ovl_override_creds(dentry->d_sb);
 958         err = ovl_dir_read_merged(dentry, list, &root);
 959         revert_creds(old_cred);
 960         if (err)
 961                 return err;
 962
 963         err = 0;
 964
 965         list_for_each_entry_safe(p, n, list, l_node) {
 966                 /*
 967                  * Select whiteouts in upperdir, they should
 968                  * be cleared when deleting this directory.
 969                  */
 970                 if (p->is_whiteout) {
 971                         if (p->is_upper)
 972                                 continue;
 973                         goto del_entry;
 974                 }
 975
 976                 if (p->name[0] == '.') {
 977                         if (p->len == 1)
 978                                 goto del_entry;
 979                         if (p->len == 2 && p->name[1] == '.')
 980                                 goto del_entry;
 981                 }
 982                 err = -ENOTEMPTY;
 983                 break;
 984
 985 del_entry:
 986                 list_del(&p->l_node);
 987                 kfree(p);
 988         }
 989
 990         return err;
 991 }
 992
 993 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
 994 {
 995         struct ovl_cache_entry *p;
 996
 997         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
 998         list_for_each_entry(p, list, l_node) {
 999                 struct dentry *dentry;
1000
1001                 if (WARN_ON(!p->is_whiteout || !p->is_upper))
1002                         continue;
1003
1004                 dentry = lookup_one_len(p->name, upper, p->len);
1005                 if (IS_ERR(dentry)) {
1006                         pr_err("lookup '%s/%.*s' failed (%i)\n",
1007                                upper->d_name.name, p->len, p->name,
1008                                (int) PTR_ERR(dentry));
1009                         continue;
1010                 }
1011                 if (dentry->d_inode)
1012                         ovl_cleanup(upper->d_inode, dentry);
1013                 dput(dentry);
1014         }
1015         inode_unlock(upper->d_inode);
1016 }
1017
1018 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
1019                           int namelen, loff_t offset, u64 ino,
1020                           unsigned int d_type)
1021 {
1022         struct ovl_readdir_data *rdd =
1023                 container_of(ctx, struct ovl_readdir_data, ctx);
1024
1025         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
1026         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
1027                 return 0;
1028
1029         if (d_type != DT_UNKNOWN)
1030                 rdd->d_type_supported = true;
1031
1032         return 0;
1033 }
1034
1035 /*
1036  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
1037  * if error is encountered.
1038  */
1039 int ovl_check_d_type_supported(struct path *realpath)
1040 {
1041         int err;
1042         struct ovl_readdir_data rdd = {
1043                 .ctx.actor = ovl_check_d_type,
1044                 .d_type_supported = false,
1045         };
1046
1047         err = ovl_dir_read(realpath, &rdd);
1048         if (err)
1049                 return err;
1050
1051         return rdd.d_type_supported;
1052 }
1053
1054 static void ovl_workdir_cleanup_recurse(struct path *path, int level)
1055 {
1056         int err;
1057         struct inode *dir = path->dentry->d_inode;
1058         LIST_HEAD(list);
1059         struct rb_root root = RB_ROOT;
1060         struct ovl_cache_entry *p;
1061         struct ovl_readdir_data rdd = {
1062                 .ctx.actor = ovl_fill_merge,
1063                 .dentry = NULL,
1064                 .list = &list,
1065                 .root = &root,
1066                 .is_lowest = false,
1067         };
1068
1069         err = ovl_dir_read(path, &rdd);
1070         if (err)
1071                 goto out;
1072
1073         inode_lock_nested(dir, I_MUTEX_PARENT);
1074         list_for_each_entry(p, &list, l_node) {
1075                 struct dentry *dentry;
1076
1077                 if (p->name[0] == '.') {
1078                         if (p->len == 1)
1079                                 continue;
1080                         if (p->len == 2 && p->name[1] == '.')
1081                                 continue;
1082                 }
1083                 dentry = lookup_one_len(p->name, path->dentry, p->len);
1084                 if (IS_ERR(dentry))
1085                         continue;
1086                 if (dentry->d_inode)
1087                         ovl_workdir_cleanup(dir, path->mnt, dentry, level);
1088                 dput(dentry);
1089         }
1090         inode_unlock(dir);
1091 out:
1092         ovl_cache_free(&list);
1093 }
1094
1095 int ovl_workdir_cleanup(struct inode *dir, struct vfsmount *mnt,
1096                          struct dentry *dentry, int level)
1097 {
1098         int err;
1099
1100         if (!d_is_dir(dentry) || level > 1) {
1101                 return ovl_cleanup(dir, dentry);
1102         }
1103
1104         err = ovl_do_rmdir(dir, dentry);
1105         if (err) {
1106                 struct path path = { .mnt = mnt, .dentry = dentry };
1107
1108                 inode_unlock(dir);
1109                 ovl_workdir_cleanup_recurse(&path, level + 1);
1110                 inode_lock_nested(dir, I_MUTEX_PARENT);
1111                 err = ovl_cleanup(dir, dentry);
1112         }
1113
1114         return err;
1115 }
1116
1117 int ovl_indexdir_cleanup(struct ovl_fs *ofs)
1118 {
1119         int err;
1120         struct dentry *indexdir = ofs->indexdir;
1121         struct dentry *index = NULL;
1122         struct inode *dir = indexdir->d_inode;
1123         struct path path = { .mnt = ovl_upper_mnt(ofs), .dentry = indexdir };
1124         LIST_HEAD(list);
1125         struct rb_root root = RB_ROOT;
1126         struct ovl_cache_entry *p;
1127         struct ovl_readdir_data rdd = {
1128                 .ctx.actor = ovl_fill_merge,
1129                 .dentry = NULL,
1130                 .list = &list,
1131                 .root = &root,
1132                 .is_lowest = false,
1133         };
1134
1135         err = ovl_dir_read(&path, &rdd);
1136         if (err)
1137                 goto out;
1138
1139         inode_lock_nested(dir, I_MUTEX_PARENT);
1140         list_for_each_entry(p, &list, l_node) {
1141                 if (p->name[0] == '.') {
1142                         if (p->len == 1)
1143                                 continue;
1144                         if (p->len == 2 && p->name[1] == '.')
1145                                 continue;
1146                 }
1147                 index = lookup_one_len(p->name, indexdir, p->len);
1148                 if (IS_ERR(index)) {
1149                         err = PTR_ERR(index);
1150                         index = NULL;
1151                         break;
1152                 }
1153                 /* Cleanup leftover from index create/cleanup attempt */
1154                 if (index->d_name.name[0] == '#') {
1155                         err = ovl_workdir_cleanup(dir, path.mnt, index, 1);
1156                         if (err)
1157                                 break;
1158                         goto next;
1159                 }
1160                 err = ovl_verify_index(ofs, index);
1161                 if (!err) {
1162                         goto next;
1163                 } else if (err == -ESTALE) {
1164                         /* Cleanup stale index entries */
1165                         err = ovl_cleanup(dir, index);
1166                 } else if (err != -ENOENT) {
1167                         /*
1168                          * Abort mount to avoid corrupting the index if
1169                          * an incompatible index entry was found or on out
1170                          * of memory.
1171                          */
1172                         break;
1173                 } else if (ofs->config.nfs_export) {
1174                         /*
1175                          * Whiteout orphan index to block future open by
1176                          * handle after overlay nlink dropped to zero.
1177                          */
1178                         err = ovl_cleanup_and_whiteout(ofs, dir, index);
1179                 } else {
1180                         /* Cleanup orphan index entries */
1181                         err = ovl_cleanup(dir, index);
1182                 }
1183
1184                 if (err)
1185                         break;
1186
1187 next:
1188                 dput(index);
1189                 index = NULL;
1190         }
1191         dput(index);
1192         inode_unlock(dir);
1193 out:
1194         ovl_cache_free(&list);
1195         if (err)
1196                 pr_err("failed index dir cleanup (%i)\n", err);
1197         return err;
1198 }