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 bool 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 true;
 183
 184         p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
 185         if (p == NULL) {
 186                 rdd->err = -ENOMEM;
 187                 return false;
 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 true;
 195 }
 196
 197 static bool 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 == 0;
 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 inode *inode)
 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(inode) == cache)
 246                         ovl_set_dir_cache(inode, NULL);
 247
 248                 ovl_cache_free(&cache->entries);
 249                 kfree(cache);
 250         }
 251 }
 252
 253 static bool 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(const struct path *path, struct ovl_readdir_data *rdd)
 268 {
 269         int err;
 270         struct ovl_cache_entry *p;
 271         struct dentry *dentry, *dir = path->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(mnt_idmap(path->mnt), 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(const 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, rdd);
 316
 317         fput(realfile);
 318
 319         return err;
 320 }
 321
 322 static void ovl_dir_reset(struct file *file)
 323 {
 324         struct ovl_dir_file *od = file->private_data;
 325         struct ovl_dir_cache *cache = od->cache;
 326         struct inode *inode = file_inode(file);
 327         bool is_real;
 328
 329         if (cache && ovl_inode_version_get(inode) != cache->version) {
 330                 ovl_cache_put(od, inode);
 331                 od->cache = NULL;
 332                 od->cursor = NULL;
 333         }
 334         is_real = ovl_dir_is_real(inode);
 335         if (od->is_real != is_real) {
 336                 /* is_real can only become false when dir is copied up */
 337                 if (WARN_ON(is_real))
 338                         return;
 339                 od->is_real = false;
 340         }
 341 }
 342
 343 static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list,
 344         struct rb_root *root)
 345 {
 346         int err;
 347         struct path realpath;
 348         struct ovl_readdir_data rdd = {
 349                 .ctx.actor = ovl_fill_merge,
 350                 .dentry = dentry,
 351                 .list = list,
 352                 .root = root,
 353                 .is_lowest = false,
 354         };
 355         int idx, next;
 356
 357         for (idx = 0; idx != -1; idx = next) {
 358                 next = ovl_path_next(idx, dentry, &realpath);
 359                 rdd.is_upper = ovl_dentry_upper(dentry) == realpath.dentry;
 360
 361                 if (next != -1) {
 362                         err = ovl_dir_read(&realpath, &rdd);
 363                         if (err)
 364                                 break;
 365                 } else {
 366                         /*
 367                          * Insert lowest layer entries before upper ones, this
 368                          * allows offsets to be reasonably constant
 369                          */
 370                         list_add(&rdd.middle, rdd.list);
 371                         rdd.is_lowest = true;
 372                         err = ovl_dir_read(&realpath, &rdd);
 373                         list_del(&rdd.middle);
 374                 }
 375         }
 376         return err;
 377 }
 378
 379 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 380 {
 381         struct list_head *p;
 382         loff_t off = 0;
 383
 384         list_for_each(p, &od->cache->entries) {
 385                 if (off >= pos)
 386                         break;
 387                 off++;
 388         }
 389         /* Cursor is safe since the cache is stable */
 390         od->cursor = p;
 391 }
 392
 393 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 394 {
 395         int res;
 396         struct ovl_dir_cache *cache;
 397         struct inode *inode = d_inode(dentry);
 398
 399         cache = ovl_dir_cache(inode);
 400         if (cache && ovl_inode_version_get(inode) == cache->version) {
 401                 WARN_ON(!cache->refcount);
 402                 cache->refcount++;
 403                 return cache;
 404         }
 405         ovl_set_dir_cache(d_inode(dentry), NULL);
 406
 407         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 408         if (!cache)
 409                 return ERR_PTR(-ENOMEM);
 410
 411         cache->refcount = 1;
 412         INIT_LIST_HEAD(&cache->entries);
 413         cache->root = RB_ROOT;
 414
 415         res = ovl_dir_read_merged(dentry, &cache->entries, &cache->root);
 416         if (res) {
 417                 ovl_cache_free(&cache->entries);
 418                 kfree(cache);
 419                 return ERR_PTR(res);
 420         }
 421
 422         cache->version = ovl_inode_version_get(inode);
 423         ovl_set_dir_cache(inode, cache);
 424
 425         return cache;
 426 }
 427
 428 /* Map inode number to lower fs unique range */
 429 static u64 ovl_remap_lower_ino(u64 ino, int xinobits, int fsid,
 430                                const char *name, int namelen, bool warn)
 431 {
 432         unsigned int xinoshift = 64 - xinobits;
 433
 434         if (unlikely(ino >> xinoshift)) {
 435                 if (warn) {
 436                         pr_warn_ratelimited("d_ino too big (%.*s, ino=%llu, xinobits=%d)\n",
 437                                             namelen, name, ino, xinobits);
 438                 }
 439                 return ino;
 440         }
 441
 442         /*
 443          * The lowest xinobit is reserved for mapping the non-peresistent inode
 444          * numbers range, but this range is only exposed via st_ino, not here.
 445          */
 446         return ino | ((u64)fsid) << (xinoshift + 1);
 447 }
 448
 449 /*
 450  * Set d_ino for upper entries. Non-upper entries should always report
 451  * the uppermost real inode ino and should not call this function.
 452  *
 453  * When not all layer are on same fs, report real ino also for upper.
 454  *
 455  * When all layers are on the same fs, and upper has a reference to
 456  * copy up origin, call vfs_getattr() on the overlay entry to make
 457  * sure that d_ino will be consistent with st_ino from stat(2).
 458  */
 459 static int ovl_cache_update_ino(const struct path *path, struct ovl_cache_entry *p)
 460
 461 {
 462         struct dentry *dir = path->dentry;
 463         struct dentry *this = NULL;
 464         enum ovl_path_type type;
 465         u64 ino = p->real_ino;
 466         int xinobits = ovl_xino_bits(dir->d_sb);
 467         int err = 0;
 468
 469         if (!ovl_same_dev(dir->d_sb))
 470                 goto out;
 471
 472         if (p->name[0] == '.') {
 473                 if (p->len == 1) {
 474                         this = dget(dir);
 475                         goto get;
 476                 }
 477                 if (p->len == 2 && p->name[1] == '.') {
 478                         /* we shall not be moved */
 479                         this = dget(dir->d_parent);
 480                         goto get;
 481                 }
 482         }
 483         this = lookup_one(mnt_idmap(path->mnt), p->name, dir, p->len);
 484         if (IS_ERR_OR_NULL(this) || !this->d_inode) {
 485                 /* Mark a stale entry */
 486                 p->is_whiteout = true;
 487                 if (IS_ERR(this)) {
 488                         err = PTR_ERR(this);
 489                         this = NULL;
 490                         goto fail;
 491                 }
 492                 goto out;
 493         }
 494
 495 get:
 496         type = ovl_path_type(this);
 497         if (OVL_TYPE_ORIGIN(type)) {
 498                 struct kstat stat;
 499                 struct path statpath = *path;
 500
 501                 statpath.dentry = this;
 502                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 503                 if (err)
 504                         goto fail;
 505
 506                 /*
 507                  * Directory inode is always on overlay st_dev.
 508                  * Non-dir with ovl_same_dev() could be on pseudo st_dev in case
 509                  * of xino bits overflow.
 510                  */
 511                 WARN_ON_ONCE(S_ISDIR(stat.mode) &&
 512                              dir->d_sb->s_dev != stat.dev);
 513                 ino = stat.ino;
 514         } else if (xinobits && !OVL_TYPE_UPPER(type)) {
 515                 ino = ovl_remap_lower_ino(ino, xinobits,
 516                                           ovl_layer_lower(this)->fsid,
 517                                           p->name, p->len,
 518                                           ovl_xino_warn(dir->d_sb));
 519         }
 520
 521 out:
 522         p->ino = ino;
 523         dput(this);
 524         return err;
 525
 526 fail:
 527         pr_warn_ratelimited("failed to look up (%s) for ino (%i)\n",
 528                             p->name, err);
 529         goto out;
 530 }
 531
 532 static bool ovl_fill_plain(struct dir_context *ctx, const char *name,
 533                           int namelen, loff_t offset, u64 ino,
 534                           unsigned int d_type)
 535 {
 536         struct ovl_cache_entry *p;
 537         struct ovl_readdir_data *rdd =
 538                 container_of(ctx, struct ovl_readdir_data, ctx);
 539
 540         rdd->count++;
 541         p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 542         if (p == NULL) {
 543                 rdd->err = -ENOMEM;
 544                 return false;
 545         }
 546         list_add_tail(&p->l_node, rdd->list);
 547
 548         return true;
 549 }
 550
 551 static int ovl_dir_read_impure(const struct path *path,  struct list_head *list,
 552                                struct rb_root *root)
 553 {
 554         int err;
 555         struct path realpath;
 556         struct ovl_cache_entry *p, *n;
 557         struct ovl_readdir_data rdd = {
 558                 .ctx.actor = ovl_fill_plain,
 559                 .list = list,
 560                 .root = root,
 561         };
 562
 563         INIT_LIST_HEAD(list);
 564         *root = RB_ROOT;
 565         ovl_path_upper(path->dentry, &realpath);
 566
 567         err = ovl_dir_read(&realpath, &rdd);
 568         if (err)
 569                 return err;
 570
 571         list_for_each_entry_safe(p, n, list, l_node) {
 572                 if (strcmp(p->name, ".") != 0 &&
 573                     strcmp(p->name, "..") != 0) {
 574                         err = ovl_cache_update_ino(path, p);
 575                         if (err)
 576                                 return err;
 577                 }
 578                 if (p->ino == p->real_ino) {
 579                         list_del(&p->l_node);
 580                         kfree(p);
 581                 } else {
 582                         struct rb_node **newp = &root->rb_node;
 583                         struct rb_node *parent = NULL;
 584
 585                         if (WARN_ON(ovl_cache_entry_find_link(p->name, p->len,
 586                                                               &newp, &parent)))
 587                                 return -EIO;
 588
 589                         rb_link_node(&p->node, parent, newp);
 590                         rb_insert_color(&p->node, root);
 591                 }
 592         }
 593         return 0;
 594 }
 595
 596 static struct ovl_dir_cache *ovl_cache_get_impure(const struct path *path)
 597 {
 598         int res;
 599         struct dentry *dentry = path->dentry;
 600         struct inode *inode = d_inode(dentry);
 601         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 602         struct ovl_dir_cache *cache;
 603
 604         cache = ovl_dir_cache(inode);
 605         if (cache && ovl_inode_version_get(inode) == cache->version)
 606                 return cache;
 607
 608         /* Impure cache is not refcounted, free it here */
 609         ovl_dir_cache_free(inode);
 610         ovl_set_dir_cache(inode, NULL);
 611
 612         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 613         if (!cache)
 614                 return ERR_PTR(-ENOMEM);
 615
 616         res = ovl_dir_read_impure(path, &cache->entries, &cache->root);
 617         if (res) {
 618                 ovl_cache_free(&cache->entries);
 619                 kfree(cache);
 620                 return ERR_PTR(res);
 621         }
 622         if (list_empty(&cache->entries)) {
 623                 /*
 624                  * A good opportunity to get rid of an unneeded "impure" flag.
 625                  * Removing the "impure" xattr is best effort.
 626                  */
 627                 if (!ovl_want_write(dentry)) {
 628                         ovl_removexattr(ofs, ovl_dentry_upper(dentry),
 629                                         OVL_XATTR_IMPURE);
 630                         ovl_drop_write(dentry);
 631                 }
 632                 ovl_clear_flag(OVL_IMPURE, inode);
 633                 kfree(cache);
 634                 return NULL;
 635         }
 636
 637         cache->version = ovl_inode_version_get(inode);
 638         ovl_set_dir_cache(inode, cache);
 639
 640         return cache;
 641 }
 642
 643 struct ovl_readdir_translate {
 644         struct dir_context *orig_ctx;
 645         struct ovl_dir_cache *cache;
 646         struct dir_context ctx;
 647         u64 parent_ino;
 648         int fsid;
 649         int xinobits;
 650         bool xinowarn;
 651 };
 652
 653 static bool ovl_fill_real(struct dir_context *ctx, const char *name,
 654                            int namelen, loff_t offset, u64 ino,
 655                            unsigned int d_type)
 656 {
 657         struct ovl_readdir_translate *rdt =
 658                 container_of(ctx, struct ovl_readdir_translate, ctx);
 659         struct dir_context *orig_ctx = rdt->orig_ctx;
 660
 661         if (rdt->parent_ino && strcmp(name, "..") == 0) {
 662                 ino = rdt->parent_ino;
 663         } else if (rdt->cache) {
 664                 struct ovl_cache_entry *p;
 665
 666                 p = ovl_cache_entry_find(&rdt->cache->root, name, namelen);
 667                 if (p)
 668                         ino = p->ino;
 669         } else if (rdt->xinobits) {
 670                 ino = ovl_remap_lower_ino(ino, rdt->xinobits, rdt->fsid,
 671                                           name, namelen, rdt->xinowarn);
 672         }
 673
 674         return orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
 675 }
 676
 677 static bool ovl_is_impure_dir(struct file *file)
 678 {
 679         struct ovl_dir_file *od = file->private_data;
 680         struct inode *dir = file_inode(file);
 681
 682         /*
 683          * Only upper dir can be impure, but if we are in the middle of
 684          * iterating a lower real dir, dir could be copied up and marked
 685          * impure. We only want the impure cache if we started iterating
 686          * a real upper dir to begin with.
 687          */
 688         return od->is_upper && ovl_test_flag(OVL_IMPURE, dir);
 689
 690 }
 691
 692 static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
 693 {
 694         int err;
 695         struct ovl_dir_file *od = file->private_data;
 696         struct dentry *dir = file->f_path.dentry;
 697         const struct ovl_layer *lower_layer = ovl_layer_lower(dir);
 698         struct ovl_readdir_translate rdt = {
 699                 .ctx.actor = ovl_fill_real,
 700                 .orig_ctx = ctx,
 701                 .xinobits = ovl_xino_bits(dir->d_sb),
 702                 .xinowarn = ovl_xino_warn(dir->d_sb),
 703         };
 704
 705         if (rdt.xinobits && lower_layer)
 706                 rdt.fsid = lower_layer->fsid;
 707
 708         if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
 709                 struct kstat stat;
 710                 struct path statpath = file->f_path;
 711
 712                 statpath.dentry = dir->d_parent;
 713                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 714                 if (err)
 715                         return err;
 716
 717                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 718                 rdt.parent_ino = stat.ino;
 719         }
 720
 721         if (ovl_is_impure_dir(file)) {
 722                 rdt.cache = ovl_cache_get_impure(&file->f_path);
 723                 if (IS_ERR(rdt.cache))
 724                         return PTR_ERR(rdt.cache);
 725         }
 726
 727         err = iterate_dir(od->realfile, &rdt.ctx);
 728         ctx->pos = rdt.ctx.pos;
 729
 730         return err;
 731 }
 732
 733
 734 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 735 {
 736         struct ovl_dir_file *od = file->private_data;
 737         struct dentry *dentry = file->f_path.dentry;
 738         struct ovl_cache_entry *p;
 739         const struct cred *old_cred;
 740         int err;
 741
 742         old_cred = ovl_override_creds(dentry->d_sb);
 743         if (!ctx->pos)
 744                 ovl_dir_reset(file);
 745
 746         if (od->is_real) {
 747                 /*
 748                  * If parent is merge, then need to adjust d_ino for '..', if
 749                  * dir is impure then need to adjust d_ino for copied up
 750                  * entries.
 751                  */
 752                 if (ovl_xino_bits(dentry->d_sb) ||
 753                     (ovl_same_fs(dentry->d_sb) &&
 754                      (ovl_is_impure_dir(file) ||
 755                       OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent))))) {
 756                         err = ovl_iterate_real(file, ctx);
 757                 } else {
 758                         err = iterate_dir(od->realfile, ctx);
 759                 }
 760                 goto out;
 761         }
 762
 763         if (!od->cache) {
 764                 struct ovl_dir_cache *cache;
 765
 766                 cache = ovl_cache_get(dentry);
 767                 err = PTR_ERR(cache);
 768                 if (IS_ERR(cache))
 769                         goto out;
 770
 771                 od->cache = cache;
 772                 ovl_seek_cursor(od, ctx->pos);
 773         }
 774
 775         while (od->cursor != &od->cache->entries) {
 776                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 777                 if (!p->is_whiteout) {
 778                         if (!p->ino) {
 779                                 err = ovl_cache_update_ino(&file->f_path, p);
 780                                 if (err)
 781                                         goto out;
 782                         }
 783                 }
 784                 /* ovl_cache_update_ino() sets is_whiteout on stale entry */
 785                 if (!p->is_whiteout) {
 786                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 787                                 break;
 788                 }
 789                 od->cursor = p->l_node.next;
 790                 ctx->pos++;
 791         }
 792         err = 0;
 793 out:
 794         revert_creds(old_cred);
 795         return err;
 796 }
 797
 798 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 799 {
 800         loff_t res;
 801         struct ovl_dir_file *od = file->private_data;
 802
 803         inode_lock(file_inode(file));
 804         if (!file->f_pos)
 805                 ovl_dir_reset(file);
 806
 807         if (od->is_real) {
 808                 res = vfs_llseek(od->realfile, offset, origin);
 809                 file->f_pos = od->realfile->f_pos;
 810         } else {
 811                 res = -EINVAL;
 812
 813                 switch (origin) {
 814                 case SEEK_CUR:
 815                         offset += file->f_pos;
 816                         break;
 817                 case SEEK_SET:
 818                         break;
 819                 default:
 820                         goto out_unlock;
 821                 }
 822                 if (offset < 0)
 823                         goto out_unlock;
 824
 825                 if (offset != file->f_pos) {
 826                         file->f_pos = offset;
 827                         if (od->cache)
 828                                 ovl_seek_cursor(od, offset);
 829                 }
 830                 res = offset;
 831         }
 832 out_unlock:
 833         inode_unlock(file_inode(file));
 834
 835         return res;
 836 }
 837
 838 static struct file *ovl_dir_open_realfile(const struct file *file,
 839                                           const struct path *realpath)
 840 {
 841         struct file *res;
 842         const struct cred *old_cred;
 843
 844         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 845         res = ovl_path_open(realpath, O_RDONLY | (file->f_flags & O_LARGEFILE));
 846         revert_creds(old_cred);
 847
 848         return res;
 849 }
 850
 851 /*
 852  * Like ovl_real_fdget(), returns upperfile if dir was copied up since open.
 853  * Unlike ovl_real_fdget(), this caches upperfile in file->private_data.
 854  *
 855  * TODO: use same abstract type for file->private_data of dir and file so
 856  * upperfile could also be cached for files as well.
 857  */
 858 struct file *ovl_dir_real_file(const struct file *file, bool want_upper)
 859 {
 860
 861         struct ovl_dir_file *od = file->private_data;
 862         struct dentry *dentry = file->f_path.dentry;
 863         struct file *old, *realfile = od->realfile;
 864
 865         if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
 866                 return want_upper ? NULL : realfile;
 867
 868         /*
 869          * Need to check if we started out being a lower dir, but got copied up
 870          */
 871         if (!od->is_upper) {
 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                         if (IS_ERR(realfile))
 879                                 return realfile;
 880
 881                         old = cmpxchg_release(&od->upperfile, NULL, realfile);
 882                         if (old) {
 883                                 fput(realfile);
 884                                 realfile = old;
 885                         }
 886                 }
 887         }
 888
 889         return realfile;
 890 }
 891
 892 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 893                          int datasync)
 894 {
 895         struct file *realfile;
 896         int err;
 897
 898         err = ovl_sync_status(OVL_FS(file_inode(file)->i_sb));
 899         if (err <= 0)
 900                 return err;
 901
 902         realfile = ovl_dir_real_file(file, true);
 903         err = PTR_ERR_OR_ZERO(realfile);
 904
 905         /* Nothing to sync for lower */
 906         if (!realfile || err)
 907                 return err;
 908
 909         return vfs_fsync_range(realfile, start, end, datasync);
 910 }
 911
 912 static int ovl_dir_release(struct inode *inode, struct file *file)
 913 {
 914         struct ovl_dir_file *od = file->private_data;
 915
 916         if (od->cache) {
 917                 inode_lock(inode);
 918                 ovl_cache_put(od, inode);
 919                 inode_unlock(inode);
 920         }
 921         fput(od->realfile);
 922         if (od->upperfile)
 923                 fput(od->upperfile);
 924         kfree(od);
 925
 926         return 0;
 927 }
 928
 929 static int ovl_dir_open(struct inode *inode, struct file *file)
 930 {
 931         struct path realpath;
 932         struct file *realfile;
 933         struct ovl_dir_file *od;
 934         enum ovl_path_type type;
 935
 936         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 937         if (!od)
 938                 return -ENOMEM;
 939
 940         type = ovl_path_real(file->f_path.dentry, &realpath);
 941         realfile = ovl_dir_open_realfile(file, &realpath);
 942         if (IS_ERR(realfile)) {
 943                 kfree(od);
 944                 return PTR_ERR(realfile);
 945         }
 946         od->realfile = realfile;
 947         od->is_real = ovl_dir_is_real(inode);
 948         od->is_upper = OVL_TYPE_UPPER(type);
 949         file->private_data = od;
 950
 951         return 0;
 952 }
 953
 954 const struct file_operations ovl_dir_operations = {
 955         .read           = generic_read_dir,
 956         .open           = ovl_dir_open,
 957         .iterate        = ovl_iterate,
 958         .llseek         = ovl_dir_llseek,
 959         .fsync          = ovl_dir_fsync,
 960         .release        = ovl_dir_release,
 961 };
 962
 963 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 964 {
 965         int err;
 966         struct ovl_cache_entry *p, *n;
 967         struct rb_root root = RB_ROOT;
 968         const struct cred *old_cred;
 969
 970         old_cred = ovl_override_creds(dentry->d_sb);
 971         err = ovl_dir_read_merged(dentry, list, &root);
 972         revert_creds(old_cred);
 973         if (err)
 974                 return err;
 975
 976         err = 0;
 977
 978         list_for_each_entry_safe(p, n, list, l_node) {
 979                 /*
 980                  * Select whiteouts in upperdir, they should
 981                  * be cleared when deleting this directory.
 982                  */
 983                 if (p->is_whiteout) {
 984                         if (p->is_upper)
 985                                 continue;
 986                         goto del_entry;
 987                 }
 988
 989                 if (p->name[0] == '.') {
 990                         if (p->len == 1)
 991                                 goto del_entry;
 992                         if (p->len == 2 && p->name[1] == '.')
 993                                 goto del_entry;
 994                 }
 995                 err = -ENOTEMPTY;
 996                 break;
 997
 998 del_entry:
 999                 list_del(&p->l_node);
1000                 kfree(p);
1001         }
1002
1003         return err;
1004 }
1005
1006 void ovl_cleanup_whiteouts(struct ovl_fs *ofs, struct dentry *upper,
1007                            struct list_head *list)
1008 {
1009         struct ovl_cache_entry *p;
1010
1011         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
1012         list_for_each_entry(p, list, l_node) {
1013                 struct dentry *dentry;
1014
1015                 if (WARN_ON(!p->is_whiteout || !p->is_upper))
1016                         continue;
1017
1018                 dentry = ovl_lookup_upper(ofs, p->name, upper, p->len);
1019                 if (IS_ERR(dentry)) {
1020                         pr_err("lookup '%s/%.*s' failed (%i)\n",
1021                                upper->d_name.name, p->len, p->name,
1022                                (int) PTR_ERR(dentry));
1023                         continue;
1024                 }
1025                 if (dentry->d_inode)
1026                         ovl_cleanup(ofs, upper->d_inode, dentry);
1027                 dput(dentry);
1028         }
1029         inode_unlock(upper->d_inode);
1030 }
1031
1032 static bool ovl_check_d_type(struct dir_context *ctx, const char *name,
1033                           int namelen, loff_t offset, u64 ino,
1034                           unsigned int d_type)
1035 {
1036         struct ovl_readdir_data *rdd =
1037                 container_of(ctx, struct ovl_readdir_data, ctx);
1038
1039         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
1040         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
1041                 return true;
1042
1043         if (d_type != DT_UNKNOWN)
1044                 rdd->d_type_supported = true;
1045
1046         return true;
1047 }
1048
1049 /*
1050  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
1051  * if error is encountered.
1052  */
1053 int ovl_check_d_type_supported(const struct path *realpath)
1054 {
1055         int err;
1056         struct ovl_readdir_data rdd = {
1057                 .ctx.actor = ovl_check_d_type,
1058                 .d_type_supported = false,
1059         };
1060
1061         err = ovl_dir_read(realpath, &rdd);
1062         if (err)
1063                 return err;
1064
1065         return rdd.d_type_supported;
1066 }
1067
1068 #define OVL_INCOMPATDIR_NAME "incompat"
1069
1070 static int ovl_workdir_cleanup_recurse(struct ovl_fs *ofs, const struct path *path,
1071                                        int level)
1072 {
1073         int err;
1074         struct inode *dir = path->dentry->d_inode;
1075         LIST_HEAD(list);
1076         struct ovl_cache_entry *p;
1077         struct ovl_readdir_data rdd = {
1078                 .ctx.actor = ovl_fill_plain,
1079                 .list = &list,
1080         };
1081         bool incompat = false;
1082
1083         /*
1084          * The "work/incompat" directory is treated specially - if it is not
1085          * empty, instead of printing a generic error and mounting read-only,
1086          * we will error about incompat features and fail the mount.
1087          *
1088          * When called from ovl_indexdir_cleanup(), path->dentry->d_name.name
1089          * starts with '#'.
1090          */
1091         if (level == 2 &&
1092             !strcmp(path->dentry->d_name.name, OVL_INCOMPATDIR_NAME))
1093                 incompat = true;
1094
1095         err = ovl_dir_read(path, &rdd);
1096         if (err)
1097                 goto out;
1098
1099         inode_lock_nested(dir, I_MUTEX_PARENT);
1100         list_for_each_entry(p, &list, l_node) {
1101                 struct dentry *dentry;
1102
1103                 if (p->name[0] == '.') {
1104                         if (p->len == 1)
1105                                 continue;
1106                         if (p->len == 2 && p->name[1] == '.')
1107                                 continue;
1108                 } else if (incompat) {
1109                         pr_err("overlay with incompat feature '%s' cannot be mounted\n",
1110                                 p->name);
1111                         err = -EINVAL;
1112                         break;
1113                 }
1114                 dentry = ovl_lookup_upper(ofs, p->name, path->dentry, p->len);
1115                 if (IS_ERR(dentry))
1116                         continue;
1117                 if (dentry->d_inode)
1118                         err = ovl_workdir_cleanup(ofs, dir, path->mnt, dentry, level);
1119                 dput(dentry);
1120                 if (err)
1121                         break;
1122         }
1123         inode_unlock(dir);
1124 out:
1125         ovl_cache_free(&list);
1126         return err;
1127 }
1128
1129 int ovl_workdir_cleanup(struct ovl_fs *ofs, struct inode *dir,
1130                         struct vfsmount *mnt, struct dentry *dentry, int level)
1131 {
1132         int err;
1133
1134         if (!d_is_dir(dentry) || level > 1) {
1135                 return ovl_cleanup(ofs, dir, dentry);
1136         }
1137
1138         err = ovl_do_rmdir(ofs, dir, dentry);
1139         if (err) {
1140                 struct path path = { .mnt = mnt, .dentry = dentry };
1141
1142                 inode_unlock(dir);
1143                 err = ovl_workdir_cleanup_recurse(ofs, &path, level + 1);
1144                 inode_lock_nested(dir, I_MUTEX_PARENT);
1145                 if (!err)
1146                         err = ovl_cleanup(ofs, dir, dentry);
1147         }
1148
1149         return err;
1150 }
1151
1152 int ovl_indexdir_cleanup(struct ovl_fs *ofs)
1153 {
1154         int err;
1155         struct dentry *indexdir = ofs->indexdir;
1156         struct dentry *index = NULL;
1157         struct inode *dir = indexdir->d_inode;
1158         struct path path = { .mnt = ovl_upper_mnt(ofs), .dentry = indexdir };
1159         LIST_HEAD(list);
1160         struct ovl_cache_entry *p;
1161         struct ovl_readdir_data rdd = {
1162                 .ctx.actor = ovl_fill_plain,
1163                 .list = &list,
1164         };
1165
1166         err = ovl_dir_read(&path, &rdd);
1167         if (err)
1168                 goto out;
1169
1170         inode_lock_nested(dir, I_MUTEX_PARENT);
1171         list_for_each_entry(p, &list, l_node) {
1172                 if (p->name[0] == '.') {
1173                         if (p->len == 1)
1174                                 continue;
1175                         if (p->len == 2 && p->name[1] == '.')
1176                                 continue;
1177                 }
1178                 index = ovl_lookup_upper(ofs, p->name, indexdir, p->len);
1179                 if (IS_ERR(index)) {
1180                         err = PTR_ERR(index);
1181                         index = NULL;
1182                         break;
1183                 }
1184                 /* Cleanup leftover from index create/cleanup attempt */
1185                 if (index->d_name.name[0] == '#') {
1186                         err = ovl_workdir_cleanup(ofs, dir, path.mnt, index, 1);
1187                         if (err)
1188                                 break;
1189                         goto next;
1190                 }
1191                 err = ovl_verify_index(ofs, index);
1192                 if (!err) {
1193                         goto next;
1194                 } else if (err == -ESTALE) {
1195                         /* Cleanup stale index entries */
1196                         err = ovl_cleanup(ofs, dir, index);
1197                 } else if (err != -ENOENT) {
1198                         /*
1199                          * Abort mount to avoid corrupting the index if
1200                          * an incompatible index entry was found or on out
1201                          * of memory.
1202                          */
1203                         break;
1204                 } else if (ofs->config.nfs_export) {
1205                         /*
1206                          * Whiteout orphan index to block future open by
1207                          * handle after overlay nlink dropped to zero.
1208                          */
1209                         err = ovl_cleanup_and_whiteout(ofs, dir, index);
1210                 } else {
1211                         /* Cleanup orphan index entries */
1212                         err = ovl_cleanup(ofs, dir, index);
1213                 }
1214
1215                 if (err)
1216                         break;
1217
1218 next:
1219                 dput(index);
1220                 index = NULL;
1221         }
1222         dput(index);
1223         inode_unlock(dir);
1224 out:
1225         ovl_cache_free(&list);
1226         if (err)
1227                 pr_err("failed index dir cleanup (%i)\n", err);
1228         return err;
1229 }