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