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 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 dentry *dentry = file->f_path.dentry;
 327         bool is_real;
 328
 329         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 330                 ovl_cache_put(od, dentry);
 331                 od->cache = NULL;
 332                 od->cursor = NULL;
 333         }
 334         is_real = ovl_dir_is_real(dentry);
 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
 398         cache = ovl_dir_cache(d_inode(dentry));
 399         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 400                 WARN_ON(!cache->refcount);
 401                 cache->refcount++;
 402                 return cache;
 403         }
 404         ovl_set_dir_cache(d_inode(dentry), NULL);
 405
 406         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 407         if (!cache)
 408                 return ERR_PTR(-ENOMEM);
 409
 410         cache->refcount = 1;
 411         INIT_LIST_HEAD(&cache->entries);
 412         cache->root = RB_ROOT;
 413
 414         res = ovl_dir_read_merged(dentry, &cache->entries, &cache->root);
 415         if (res) {
 416                 ovl_cache_free(&cache->entries);
 417                 kfree(cache);
 418                 return ERR_PTR(res);
 419         }
 420
 421         cache->version = ovl_dentry_version_get(dentry);
 422         ovl_set_dir_cache(d_inode(dentry), cache);
 423
 424         return cache;
 425 }
 426
 427 /* Map inode number to lower fs unique range */
 428 static u64 ovl_remap_lower_ino(u64 ino, int xinobits, int fsid,
 429                                const char *name, int namelen, bool warn)
 430 {
 431         unsigned int xinoshift = 64 - xinobits;
 432
 433         if (unlikely(ino >> xinoshift)) {
 434                 if (warn) {
 435                         pr_warn_ratelimited("d_ino too big (%.*s, ino=%llu, xinobits=%d)\n",
 436                                             namelen, name, ino, xinobits);
 437                 }
 438                 return ino;
 439         }
 440
 441         /*
 442          * The lowest xinobit is reserved for mapping the non-peresistent inode
 443          * numbers range, but this range is only exposed via st_ino, not here.
 444          */
 445         return ino | ((u64)fsid) << (xinoshift + 1);
 446 }
 447
 448 /*
 449  * Set d_ino for upper entries. Non-upper entries should always report
 450  * the uppermost real inode ino and should not call this function.
 451  *
 452  * When not all layer are on same fs, report real ino also for upper.
 453  *
 454  * When all layers are on the same fs, and upper has a reference to
 455  * copy up origin, call vfs_getattr() on the overlay entry to make
 456  * sure that d_ino will be consistent with st_ino from stat(2).
 457  */
 458 static int ovl_cache_update_ino(struct path *path, struct ovl_cache_entry *p)
 459
 460 {
 461         struct dentry *dir = path->dentry;
 462         struct dentry *this = NULL;
 463         enum ovl_path_type type;
 464         u64 ino = p->real_ino;
 465         int xinobits = ovl_xino_bits(dir->d_sb);
 466         int err = 0;
 467
 468         if (!ovl_same_dev(dir->d_sb))
 469                 goto out;
 470
 471         if (p->name[0] == '.') {
 472                 if (p->len == 1) {
 473                         this = dget(dir);
 474                         goto get;
 475                 }
 476                 if (p->len == 2 && p->name[1] == '.') {
 477                         /* we shall not be moved */
 478                         this = dget(dir->d_parent);
 479                         goto get;
 480                 }
 481         }
 482         this = lookup_one_len(p->name, dir, p->len);
 483         if (IS_ERR_OR_NULL(this) || !this->d_inode) {
 484                 if (IS_ERR(this)) {
 485                         err = PTR_ERR(this);
 486                         this = NULL;
 487                         goto fail;
 488                 }
 489                 goto out;
 490         }
 491
 492 get:
 493         type = ovl_path_type(this);
 494         if (OVL_TYPE_ORIGIN(type)) {
 495                 struct kstat stat;
 496                 struct path statpath = *path;
 497
 498                 statpath.dentry = this;
 499                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 500                 if (err)
 501                         goto fail;
 502
 503                 /*
 504                  * Directory inode is always on overlay st_dev.
 505                  * Non-dir with ovl_same_dev() could be on pseudo st_dev in case
 506                  * of xino bits overflow.
 507                  */
 508                 WARN_ON_ONCE(S_ISDIR(stat.mode) &&
 509                              dir->d_sb->s_dev != stat.dev);
 510                 ino = stat.ino;
 511         } else if (xinobits && !OVL_TYPE_UPPER(type)) {
 512                 ino = ovl_remap_lower_ino(ino, xinobits,
 513                                           ovl_layer_lower(this)->fsid,
 514                                           p->name, p->len,
 515                                           ovl_xino_warn(dir->d_sb));
 516         }
 517
 518 out:
 519         p->ino = ino;
 520         dput(this);
 521         return err;
 522
 523 fail:
 524         pr_warn_ratelimited("failed to look up (%s) for ino (%i)\n",
 525                             p->name, err);
 526         goto out;
 527 }
 528
 529 static int ovl_fill_plain(struct dir_context *ctx, const char *name,
 530                           int namelen, loff_t offset, u64 ino,
 531                           unsigned int d_type)
 532 {
 533         struct ovl_cache_entry *p;
 534         struct ovl_readdir_data *rdd =
 535                 container_of(ctx, struct ovl_readdir_data, ctx);
 536
 537         rdd->count++;
 538         p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 539         if (p == NULL) {
 540                 rdd->err = -ENOMEM;
 541                 return -ENOMEM;
 542         }
 543         list_add_tail(&p->l_node, rdd->list);
 544
 545         return 0;
 546 }
 547
 548 static int ovl_dir_read_impure(struct path *path,  struct list_head *list,
 549                                struct rb_root *root)
 550 {
 551         int err;
 552         struct path realpath;
 553         struct ovl_cache_entry *p, *n;
 554         struct ovl_readdir_data rdd = {
 555                 .ctx.actor = ovl_fill_plain,
 556                 .list = list,
 557                 .root = root,
 558         };
 559
 560         INIT_LIST_HEAD(list);
 561         *root = RB_ROOT;
 562         ovl_path_upper(path->dentry, &realpath);
 563
 564         err = ovl_dir_read(&realpath, &rdd);
 565         if (err)
 566                 return err;
 567
 568         list_for_each_entry_safe(p, n, list, l_node) {
 569                 if (strcmp(p->name, ".") != 0 &&
 570                     strcmp(p->name, "..") != 0) {
 571                         err = ovl_cache_update_ino(path, p);
 572                         if (err)
 573                                 return err;
 574                 }
 575                 if (p->ino == p->real_ino) {
 576                         list_del(&p->l_node);
 577                         kfree(p);
 578                 } else {
 579                         struct rb_node **newp = &root->rb_node;
 580                         struct rb_node *parent = NULL;
 581
 582                         if (WARN_ON(ovl_cache_entry_find_link(p->name, p->len,
 583                                                               &newp, &parent)))
 584                                 return -EIO;
 585
 586                         rb_link_node(&p->node, parent, newp);
 587                         rb_insert_color(&p->node, root);
 588                 }
 589         }
 590         return 0;
 591 }
 592
 593 static struct ovl_dir_cache *ovl_cache_get_impure(struct path *path)
 594 {
 595         int res;
 596         struct dentry *dentry = path->dentry;
 597         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 598         struct ovl_dir_cache *cache;
 599
 600         cache = ovl_dir_cache(d_inode(dentry));
 601         if (cache && ovl_dentry_version_get(dentry) == cache->version)
 602                 return cache;
 603
 604         /* Impure cache is not refcounted, free it here */
 605         ovl_dir_cache_free(d_inode(dentry));
 606         ovl_set_dir_cache(d_inode(dentry), NULL);
 607
 608         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 609         if (!cache)
 610                 return ERR_PTR(-ENOMEM);
 611
 612         res = ovl_dir_read_impure(path, &cache->entries, &cache->root);
 613         if (res) {
 614                 ovl_cache_free(&cache->entries);
 615                 kfree(cache);
 616                 return ERR_PTR(res);
 617         }
 618         if (list_empty(&cache->entries)) {
 619                 /*
 620                  * A good opportunity to get rid of an unneeded "impure" flag.
 621                  * Removing the "impure" xattr is best effort.
 622                  */
 623                 if (!ovl_want_write(dentry)) {
 624                         ovl_do_removexattr(ofs, ovl_dentry_upper(dentry),
 625                                            OVL_XATTR_IMPURE);
 626                         ovl_drop_write(dentry);
 627                 }
 628                 ovl_clear_flag(OVL_IMPURE, d_inode(dentry));
 629                 kfree(cache);
 630                 return NULL;
 631         }
 632
 633         cache->version = ovl_dentry_version_get(dentry);
 634         ovl_set_dir_cache(d_inode(dentry), cache);
 635
 636         return cache;
 637 }
 638
 639 struct ovl_readdir_translate {
 640         struct dir_context *orig_ctx;
 641         struct ovl_dir_cache *cache;
 642         struct dir_context ctx;
 643         u64 parent_ino;
 644         int fsid;
 645         int xinobits;
 646         bool xinowarn;
 647 };
 648
 649 static int ovl_fill_real(struct dir_context *ctx, const char *name,
 650                            int namelen, loff_t offset, u64 ino,
 651                            unsigned int d_type)
 652 {
 653         struct ovl_readdir_translate *rdt =
 654                 container_of(ctx, struct ovl_readdir_translate, ctx);
 655         struct dir_context *orig_ctx = rdt->orig_ctx;
 656
 657         if (rdt->parent_ino && strcmp(name, "..") == 0) {
 658                 ino = rdt->parent_ino;
 659         } else if (rdt->cache) {
 660                 struct ovl_cache_entry *p;
 661
 662                 p = ovl_cache_entry_find(&rdt->cache->root, name, namelen);
 663                 if (p)
 664                         ino = p->ino;
 665         } else if (rdt->xinobits) {
 666                 ino = ovl_remap_lower_ino(ino, rdt->xinobits, rdt->fsid,
 667                                           name, namelen, rdt->xinowarn);
 668         }
 669
 670         return orig_ctx->actor(orig_ctx, name, namelen, offset, ino, d_type);
 671 }
 672
 673 static bool ovl_is_impure_dir(struct file *file)
 674 {
 675         struct ovl_dir_file *od = file->private_data;
 676         struct inode *dir = d_inode(file->f_path.dentry);
 677
 678         /*
 679          * Only upper dir can be impure, but if we are in the middle of
 680          * iterating a lower real dir, dir could be copied up and marked
 681          * impure. We only want the impure cache if we started iterating
 682          * a real upper dir to begin with.
 683          */
 684         return od->is_upper && ovl_test_flag(OVL_IMPURE, dir);
 685
 686 }
 687
 688 static int ovl_iterate_real(struct file *file, struct dir_context *ctx)
 689 {
 690         int err;
 691         struct ovl_dir_file *od = file->private_data;
 692         struct dentry *dir = file->f_path.dentry;
 693         const struct ovl_layer *lower_layer = ovl_layer_lower(dir);
 694         struct ovl_readdir_translate rdt = {
 695                 .ctx.actor = ovl_fill_real,
 696                 .orig_ctx = ctx,
 697                 .xinobits = ovl_xino_bits(dir->d_sb),
 698                 .xinowarn = ovl_xino_warn(dir->d_sb),
 699         };
 700
 701         if (rdt.xinobits && lower_layer)
 702                 rdt.fsid = lower_layer->fsid;
 703
 704         if (OVL_TYPE_MERGE(ovl_path_type(dir->d_parent))) {
 705                 struct kstat stat;
 706                 struct path statpath = file->f_path;
 707
 708                 statpath.dentry = dir->d_parent;
 709                 err = vfs_getattr(&statpath, &stat, STATX_INO, 0);
 710                 if (err)
 711                         return err;
 712
 713                 WARN_ON_ONCE(dir->d_sb->s_dev != stat.dev);
 714                 rdt.parent_ino = stat.ino;
 715         }
 716
 717         if (ovl_is_impure_dir(file)) {
 718                 rdt.cache = ovl_cache_get_impure(&file->f_path);
 719                 if (IS_ERR(rdt.cache))
 720                         return PTR_ERR(rdt.cache);
 721         }
 722
 723         err = iterate_dir(od->realfile, &rdt.ctx);
 724         ctx->pos = rdt.ctx.pos;
 725
 726         return err;
 727 }
 728
 729
 730 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 731 {
 732         struct ovl_dir_file *od = file->private_data;
 733         struct dentry *dentry = file->f_path.dentry;
 734         struct ovl_cache_entry *p;
 735         const struct cred *old_cred;
 736         int err;
 737
 738         old_cred = ovl_override_creds(dentry->d_sb);
 739         if (!ctx->pos)
 740                 ovl_dir_reset(file);
 741
 742         if (od->is_real) {
 743                 /*
 744                  * If parent is merge, then need to adjust d_ino for '..', if
 745                  * dir is impure then need to adjust d_ino for copied up
 746                  * entries.
 747                  */
 748                 if (ovl_xino_bits(dentry->d_sb) ||
 749                     (ovl_same_fs(dentry->d_sb) &&
 750                      (ovl_is_impure_dir(file) ||
 751                       OVL_TYPE_MERGE(ovl_path_type(dentry->d_parent))))) {
 752                         err = ovl_iterate_real(file, ctx);
 753                 } else {
 754                         err = iterate_dir(od->realfile, ctx);
 755                 }
 756                 goto out;
 757         }
 758
 759         if (!od->cache) {
 760                 struct ovl_dir_cache *cache;
 761
 762                 cache = ovl_cache_get(dentry);
 763                 err = PTR_ERR(cache);
 764                 if (IS_ERR(cache))
 765                         goto out;
 766
 767                 od->cache = cache;
 768                 ovl_seek_cursor(od, ctx->pos);
 769         }
 770
 771         while (od->cursor != &od->cache->entries) {
 772                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 773                 if (!p->is_whiteout) {
 774                         if (!p->ino) {
 775                                 err = ovl_cache_update_ino(&file->f_path, p);
 776                                 if (err)
 777                                         goto out;
 778                         }
 779                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 780                                 break;
 781                 }
 782                 od->cursor = p->l_node.next;
 783                 ctx->pos++;
 784         }
 785         err = 0;
 786 out:
 787         revert_creds(old_cred);
 788         return err;
 789 }
 790
 791 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 792 {
 793         loff_t res;
 794         struct ovl_dir_file *od = file->private_data;
 795
 796         inode_lock(file_inode(file));
 797         if (!file->f_pos)
 798                 ovl_dir_reset(file);
 799
 800         if (od->is_real) {
 801                 res = vfs_llseek(od->realfile, offset, origin);
 802                 file->f_pos = od->realfile->f_pos;
 803         } else {
 804                 res = -EINVAL;
 805
 806                 switch (origin) {
 807                 case SEEK_CUR:
 808                         offset += file->f_pos;
 809                         break;
 810                 case SEEK_SET:
 811                         break;
 812                 default:
 813                         goto out_unlock;
 814                 }
 815                 if (offset < 0)
 816                         goto out_unlock;
 817
 818                 if (offset != file->f_pos) {
 819                         file->f_pos = offset;
 820                         if (od->cache)
 821                                 ovl_seek_cursor(od, offset);
 822                 }
 823                 res = offset;
 824         }
 825 out_unlock:
 826         inode_unlock(file_inode(file));
 827
 828         return res;
 829 }
 830
 831 static struct file *ovl_dir_open_realfile(const struct file *file,
 832                                           struct path *realpath)
 833 {
 834         struct file *res;
 835         const struct cred *old_cred;
 836
 837         old_cred = ovl_override_creds(file_inode(file)->i_sb);
 838         res = ovl_path_open(realpath, O_RDONLY | (file->f_flags & O_LARGEFILE));
 839         revert_creds(old_cred);
 840
 841         return res;
 842 }
 843
 844 /*
 845  * Like ovl_real_fdget(), returns upperfile if dir was copied up since open.
 846  * Unlike ovl_real_fdget(), this caches upperfile in file->private_data.
 847  *
 848  * TODO: use same abstract type for file->private_data of dir and file so
 849  * upperfile could also be cached for files as well.
 850  */
 851 struct file *ovl_dir_real_file(const struct file *file, bool want_upper)
 852 {
 853
 854         struct ovl_dir_file *od = file->private_data;
 855         struct dentry *dentry = file->f_path.dentry;
 856         struct file *old, *realfile = od->realfile;
 857
 858         if (!OVL_TYPE_UPPER(ovl_path_type(dentry)))
 859                 return want_upper ? NULL : realfile;
 860
 861         /*
 862          * Need to check if we started out being a lower dir, but got copied up
 863          */
 864         if (!od->is_upper) {
 865                 realfile = READ_ONCE(od->upperfile);
 866                 if (!realfile) {
 867                         struct path upperpath;
 868
 869                         ovl_path_upper(dentry, &upperpath);
 870                         realfile = ovl_dir_open_realfile(file, &upperpath);
 871                         if (IS_ERR(realfile))
 872                                 return realfile;
 873
 874                         old = cmpxchg_release(&od->upperfile, NULL, realfile);
 875                         if (old) {
 876                                 fput(realfile);
 877                                 realfile = old;
 878                         }
 879                 }
 880         }
 881
 882         return realfile;
 883 }
 884
 885 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 886                          int datasync)
 887 {
 888         struct file *realfile;
 889         int err;
 890
 891         err = ovl_sync_status(OVL_FS(file->f_path.dentry->d_sb));
 892         if (err <= 0)
 893                 return err;
 894
 895         realfile = ovl_dir_real_file(file, true);
 896         err = PTR_ERR_OR_ZERO(realfile);
 897
 898         /* Nothing to sync for lower */
 899         if (!realfile || err)
 900                 return err;
 901
 902         return vfs_fsync_range(realfile, start, end, datasync);
 903 }
 904
 905 static int ovl_dir_release(struct inode *inode, struct file *file)
 906 {
 907         struct ovl_dir_file *od = file->private_data;
 908
 909         if (od->cache) {
 910                 inode_lock(inode);
 911                 ovl_cache_put(od, file->f_path.dentry);
 912                 inode_unlock(inode);
 913         }
 914         fput(od->realfile);
 915         if (od->upperfile)
 916                 fput(od->upperfile);
 917         kfree(od);
 918
 919         return 0;
 920 }
 921
 922 static int ovl_dir_open(struct inode *inode, struct file *file)
 923 {
 924         struct path realpath;
 925         struct file *realfile;
 926         struct ovl_dir_file *od;
 927         enum ovl_path_type type;
 928
 929         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 930         if (!od)
 931                 return -ENOMEM;
 932
 933         type = ovl_path_real(file->f_path.dentry, &realpath);
 934         realfile = ovl_dir_open_realfile(file, &realpath);
 935         if (IS_ERR(realfile)) {
 936                 kfree(od);
 937                 return PTR_ERR(realfile);
 938         }
 939         od->realfile = realfile;
 940         od->is_real = ovl_dir_is_real(file->f_path.dentry);
 941         od->is_upper = OVL_TYPE_UPPER(type);
 942         file->private_data = od;
 943
 944         return 0;
 945 }
 946
 947 const struct file_operations ovl_dir_operations = {
 948         .read           = generic_read_dir,
 949         .open           = ovl_dir_open,
 950         .iterate        = ovl_iterate,
 951         .llseek         = ovl_dir_llseek,
 952         .fsync          = ovl_dir_fsync,
 953         .release        = ovl_dir_release,
 954 };
 955
 956 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 957 {
 958         int err;
 959         struct ovl_cache_entry *p, *n;
 960         struct rb_root root = RB_ROOT;
 961         const struct cred *old_cred;
 962
 963         old_cred = ovl_override_creds(dentry->d_sb);
 964         err = ovl_dir_read_merged(dentry, list, &root);
 965         revert_creds(old_cred);
 966         if (err)
 967                 return err;
 968
 969         err = 0;
 970
 971         list_for_each_entry_safe(p, n, list, l_node) {
 972                 /*
 973                  * Select whiteouts in upperdir, they should
 974                  * be cleared when deleting this directory.
 975                  */
 976                 if (p->is_whiteout) {
 977                         if (p->is_upper)
 978                                 continue;
 979                         goto del_entry;
 980                 }
 981
 982                 if (p->name[0] == '.') {
 983                         if (p->len == 1)
 984                                 goto del_entry;
 985                         if (p->len == 2 && p->name[1] == '.')
 986                                 goto del_entry;
 987                 }
 988                 err = -ENOTEMPTY;
 989                 break;
 990
 991 del_entry:
 992                 list_del(&p->l_node);
 993                 kfree(p);
 994         }
 995
 996         return err;
 997 }
 998
 999 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
1000 {
1001         struct ovl_cache_entry *p;
1002
1003         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
1004         list_for_each_entry(p, list, l_node) {
1005                 struct dentry *dentry;
1006
1007                 if (WARN_ON(!p->is_whiteout || !p->is_upper))
1008                         continue;
1009
1010                 dentry = lookup_one_len(p->name, upper, p->len);
1011                 if (IS_ERR(dentry)) {
1012                         pr_err("lookup '%s/%.*s' failed (%i)\n",
1013                                upper->d_name.name, p->len, p->name,
1014                                (int) PTR_ERR(dentry));
1015                         continue;
1016                 }
1017                 if (dentry->d_inode)
1018                         ovl_cleanup(upper->d_inode, dentry);
1019                 dput(dentry);
1020         }
1021         inode_unlock(upper->d_inode);
1022 }
1023
1024 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
1025                           int namelen, loff_t offset, u64 ino,
1026                           unsigned int d_type)
1027 {
1028         struct ovl_readdir_data *rdd =
1029                 container_of(ctx, struct ovl_readdir_data, ctx);
1030
1031         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
1032         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
1033                 return 0;
1034
1035         if (d_type != DT_UNKNOWN)
1036                 rdd->d_type_supported = true;
1037
1038         return 0;
1039 }
1040
1041 /*
1042  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
1043  * if error is encountered.
1044  */
1045 int ovl_check_d_type_supported(struct path *realpath)
1046 {
1047         int err;
1048         struct ovl_readdir_data rdd = {
1049                 .ctx.actor = ovl_check_d_type,
1050                 .d_type_supported = false,
1051         };
1052
1053         err = ovl_dir_read(realpath, &rdd);
1054         if (err)
1055                 return err;
1056
1057         return rdd.d_type_supported;
1058 }
1059
1060 #define OVL_INCOMPATDIR_NAME "incompat"
1061
1062 static int ovl_workdir_cleanup_recurse(struct path *path, int level)
1063 {
1064         int err;
1065         struct inode *dir = path->dentry->d_inode;
1066         LIST_HEAD(list);
1067         struct rb_root root = RB_ROOT;
1068         struct ovl_cache_entry *p;
1069         struct ovl_readdir_data rdd = {
1070                 .ctx.actor = ovl_fill_merge,
1071                 .dentry = NULL,
1072                 .list = &list,
1073                 .root = &root,
1074                 .is_lowest = false,
1075         };
1076         bool incompat = false;
1077
1078         /*
1079          * The "work/incompat" directory is treated specially - if it is not
1080          * empty, instead of printing a generic error and mounting read-only,
1081          * we will error about incompat features and fail the mount.
1082          *
1083          * When called from ovl_indexdir_cleanup(), path->dentry->d_name.name
1084          * starts with '#'.
1085          */
1086         if (level == 2 &&
1087             !strcmp(path->dentry->d_name.name, OVL_INCOMPATDIR_NAME))
1088                 incompat = true;
1089
1090         err = ovl_dir_read(path, &rdd);
1091         if (err)
1092                 goto out;
1093
1094         inode_lock_nested(dir, I_MUTEX_PARENT);
1095         list_for_each_entry(p, &list, l_node) {
1096                 struct dentry *dentry;
1097
1098                 if (p->name[0] == '.') {
1099                         if (p->len == 1)
1100                                 continue;
1101                         if (p->len == 2 && p->name[1] == '.')
1102                                 continue;
1103                 } else if (incompat) {
1104                         pr_err("overlay with incompat feature '%s' cannot be mounted\n",
1105                                 p->name);
1106                         err = -EINVAL;
1107                         break;
1108                 }
1109                 dentry = lookup_one_len(p->name, path->dentry, p->len);
1110                 if (IS_ERR(dentry))
1111                         continue;
1112                 if (dentry->d_inode)
1113                         err = ovl_workdir_cleanup(dir, path->mnt, dentry, level);
1114                 dput(dentry);
1115                 if (err)
1116                         break;
1117         }
1118         inode_unlock(dir);
1119 out:
1120         ovl_cache_free(&list);
1121         return err;
1122 }
1123
1124 int ovl_workdir_cleanup(struct inode *dir, struct vfsmount *mnt,
1125                          struct dentry *dentry, int level)
1126 {
1127         int err;
1128
1129         if (!d_is_dir(dentry) || level > 1) {
1130                 return ovl_cleanup(dir, dentry);
1131         }
1132
1133         err = ovl_do_rmdir(dir, dentry);
1134         if (err) {
1135                 struct path path = { .mnt = mnt, .dentry = dentry };
1136
1137                 inode_unlock(dir);
1138                 err = ovl_workdir_cleanup_recurse(&path, level + 1);
1139                 inode_lock_nested(dir, I_MUTEX_PARENT);
1140                 if (!err)
1141                         err = ovl_cleanup(dir, dentry);
1142         }
1143
1144         return err;
1145 }
1146
1147 int ovl_indexdir_cleanup(struct ovl_fs *ofs)
1148 {
1149         int err;
1150         struct dentry *indexdir = ofs->indexdir;
1151         struct dentry *index = NULL;
1152         struct inode *dir = indexdir->d_inode;
1153         struct path path = { .mnt = ovl_upper_mnt(ofs), .dentry = indexdir };
1154         LIST_HEAD(list);
1155         struct rb_root root = RB_ROOT;
1156         struct ovl_cache_entry *p;
1157         struct ovl_readdir_data rdd = {
1158                 .ctx.actor = ovl_fill_merge,
1159                 .dentry = NULL,
1160                 .list = &list,
1161                 .root = &root,
1162                 .is_lowest = false,
1163         };
1164
1165         err = ovl_dir_read(&path, &rdd);
1166         if (err)
1167                 goto out;
1168
1169         inode_lock_nested(dir, I_MUTEX_PARENT);
1170         list_for_each_entry(p, &list, l_node) {
1171                 if (p->name[0] == '.') {
1172                         if (p->len == 1)
1173                                 continue;
1174                         if (p->len == 2 && p->name[1] == '.')
1175                                 continue;
1176                 }
1177                 index = lookup_one_len(p->name, indexdir, p->len);
1178                 if (IS_ERR(index)) {
1179                         err = PTR_ERR(index);
1180                         index = NULL;
1181                         break;
1182                 }
1183                 /* Cleanup leftover from index create/cleanup attempt */
1184                 if (index->d_name.name[0] == '#') {
1185                         err = ovl_workdir_cleanup(dir, path.mnt, index, 1);
1186                         if (err)
1187                                 break;
1188                         goto next;
1189                 }
1190                 err = ovl_verify_index(ofs, index);
1191                 if (!err) {
1192                         goto next;
1193                 } else if (err == -ESTALE) {
1194                         /* Cleanup stale index entries */
1195                         err = ovl_cleanup(dir, index);
1196                 } else if (err != -ENOENT) {
1197                         /*
1198                          * Abort mount to avoid corrupting the index if
1199                          * an incompatible index entry was found or on out
1200                          * of memory.
1201                          */
1202                         break;
1203                 } else if (ofs->config.nfs_export) {
1204                         /*
1205                          * Whiteout orphan index to block future open by
1206                          * handle after overlay nlink dropped to zero.
1207                          */
1208                         err = ovl_cleanup_and_whiteout(ofs, dir, index);
1209                 } else {
1210                         /* Cleanup orphan index entries */
1211                         err = ovl_cleanup(dir, index);
1212                 }
1213
1214                 if (err)
1215                         break;
1216
1217 next:
1218                 dput(index);
1219                 index = NULL;
1220         }
1221         dput(index);
1222         inode_unlock(dir);
1223 out:
1224         ovl_cache_free(&list);
1225         if (err)
1226                 pr_err("failed index dir cleanup (%i)\n", err);
1227         return err;
1228 }