fs/overlayfs/readdir.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/slab.h>
  12 #include <linux/namei.h>
  13 #include <linux/file.h>
  14 #include <linux/xattr.h>
  15 #include <linux/rbtree.h>
  16 #include <linux/security.h>
  17 #include <linux/cred.h>
  18 #include "overlayfs.h"
  19
  20 struct ovl_cache_entry {
  21         unsigned int len;
  22         unsigned int type;
  23         u64 ino;
  24         struct list_head l_node;
  25         struct rb_node node;
  26         struct ovl_cache_entry *next_maybe_whiteout;
  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 };
  36
  37 struct ovl_readdir_data {
  38         struct dir_context ctx;
  39         struct dentry *dentry;
  40         bool is_lowest;
  41         struct rb_root root;
  42         struct list_head *list;
  43         struct list_head middle;
  44         struct ovl_cache_entry *first_maybe_whiteout;
  45         int count;
  46         int err;
  47         bool d_type_supported;
  48 };
  49
  50 struct ovl_dir_file {
  51         bool is_real;
  52         bool is_upper;
  53         struct ovl_dir_cache *cache;
  54         struct list_head *cursor;
  55         struct file *realfile;
  56         struct file *upperfile;
  57 };
  58
  59 static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
  60 {
  61         return container_of(n, struct ovl_cache_entry, node);
  62 }
  63
  64 static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
  65                                                     const char *name, int len)
  66 {
  67         struct rb_node *node = root->rb_node;
  68         int cmp;
  69
  70         while (node) {
  71                 struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
  72
  73                 cmp = strncmp(name, p->name, len);
  74                 if (cmp > 0)
  75                         node = p->node.rb_right;
  76                 else if (cmp < 0 || len < p->len)
  77                         node = p->node.rb_left;
  78                 else
  79                         return p;
  80         }
  81
  82         return NULL;
  83 }
  84
  85 static struct ovl_cache_entry *ovl_cache_entry_new(struct ovl_readdir_data *rdd,
  86                                                    const char *name, int len,
  87                                                    u64 ino, unsigned int d_type)
  88 {
  89         struct ovl_cache_entry *p;
  90         size_t size = offsetof(struct ovl_cache_entry, name[len + 1]);
  91
  92         p = kmalloc(size, GFP_KERNEL);
  93         if (!p)
  94                 return NULL;
  95
  96         memcpy(p->name, name, len);
  97         p->name[len] = '\0';
  98         p->len = len;
  99         p->type = d_type;
 100         p->ino = ino;
 101         p->is_whiteout = false;
 102
 103         if (d_type == DT_CHR) {
 104                 p->next_maybe_whiteout = rdd->first_maybe_whiteout;
 105                 rdd->first_maybe_whiteout = p;
 106         }
 107         return p;
 108 }
 109
 110 static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
 111                                   const char *name, int len, u64 ino,
 112                                   unsigned int d_type)
 113 {
 114         struct rb_node **newp = &rdd->root.rb_node;
 115         struct rb_node *parent = NULL;
 116         struct ovl_cache_entry *p;
 117
 118         while (*newp) {
 119                 int cmp;
 120                 struct ovl_cache_entry *tmp;
 121
 122                 parent = *newp;
 123                 tmp = ovl_cache_entry_from_node(*newp);
 124                 cmp = strncmp(name, tmp->name, len);
 125                 if (cmp > 0)
 126                         newp = &tmp->node.rb_right;
 127                 else if (cmp < 0 || len < tmp->len)
 128                         newp = &tmp->node.rb_left;
 129                 else
 130                         return 0;
 131         }
 132
 133         p = ovl_cache_entry_new(rdd, name, len, ino, d_type);
 134         if (p == NULL)
 135                 return -ENOMEM;
 136
 137         list_add_tail(&p->l_node, rdd->list);
 138         rb_link_node(&p->node, parent, newp);
 139         rb_insert_color(&p->node, &rdd->root);
 140
 141         return 0;
 142 }
 143
 144 static int ovl_fill_lowest(struct ovl_readdir_data *rdd,
 145                            const char *name, int namelen,
 146                            loff_t offset, u64 ino, unsigned int d_type)
 147 {
 148         struct ovl_cache_entry *p;
 149
 150         p = ovl_cache_entry_find(&rdd->root, name, namelen);
 151         if (p) {
 152                 list_move_tail(&p->l_node, &rdd->middle);
 153         } else {
 154                 p = ovl_cache_entry_new(rdd, name, namelen, ino, d_type);
 155                 if (p == NULL)
 156                         rdd->err = -ENOMEM;
 157                 else
 158                         list_add_tail(&p->l_node, &rdd->middle);
 159         }
 160
 161         return rdd->err;
 162 }
 163
 164 void ovl_cache_free(struct list_head *list)
 165 {
 166         struct ovl_cache_entry *p;
 167         struct ovl_cache_entry *n;
 168
 169         list_for_each_entry_safe(p, n, list, l_node)
 170                 kfree(p);
 171
 172         INIT_LIST_HEAD(list);
 173 }
 174
 175 static void ovl_cache_put(struct ovl_dir_file *od, struct dentry *dentry)
 176 {
 177         struct ovl_dir_cache *cache = od->cache;
 178
 179         WARN_ON(cache->refcount <= 0);
 180         cache->refcount--;
 181         if (!cache->refcount) {
 182                 if (ovl_dir_cache(dentry) == cache)
 183                         ovl_set_dir_cache(dentry, NULL);
 184
 185                 ovl_cache_free(&cache->entries);
 186                 kfree(cache);
 187         }
 188 }
 189
 190 static int ovl_fill_merge(struct dir_context *ctx, const char *name,
 191                           int namelen, loff_t offset, u64 ino,
 192                           unsigned int d_type)
 193 {
 194         struct ovl_readdir_data *rdd =
 195                 container_of(ctx, struct ovl_readdir_data, ctx);
 196
 197         rdd->count++;
 198         if (!rdd->is_lowest)
 199                 return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
 200         else
 201                 return ovl_fill_lowest(rdd, name, namelen, offset, ino, d_type);
 202 }
 203
 204 static int ovl_check_whiteouts(struct dentry *dir, struct ovl_readdir_data *rdd)
 205 {
 206         int err;
 207         struct ovl_cache_entry *p;
 208         struct dentry *dentry;
 209         const struct cred *old_cred;
 210
 211         old_cred = ovl_override_creds(rdd->dentry->d_sb);
 212
 213         err = down_write_killable(&dir->d_inode->i_rwsem);
 214         if (!err) {
 215                 while (rdd->first_maybe_whiteout) {
 216                         p = rdd->first_maybe_whiteout;
 217                         rdd->first_maybe_whiteout = p->next_maybe_whiteout;
 218                         dentry = lookup_one_len(p->name, dir, p->len);
 219                         if (!IS_ERR(dentry)) {
 220                                 p->is_whiteout = ovl_is_whiteout(dentry);
 221                                 dput(dentry);
 222                         }
 223                 }
 224                 inode_unlock(dir->d_inode);
 225         }
 226         revert_creds(old_cred);
 227
 228         return err;
 229 }
 230
 231 static inline int ovl_dir_read(struct path *realpath,
 232                                struct ovl_readdir_data *rdd)
 233 {
 234         struct file *realfile;
 235         int err;
 236
 237         realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
 238         if (IS_ERR(realfile))
 239                 return PTR_ERR(realfile);
 240
 241         rdd->first_maybe_whiteout = NULL;
 242         rdd->ctx.pos = 0;
 243         do {
 244                 rdd->count = 0;
 245                 rdd->err = 0;
 246                 err = iterate_dir(realfile, &rdd->ctx);
 247                 if (err >= 0)
 248                         err = rdd->err;
 249         } while (!err && rdd->count);
 250
 251         if (!err && rdd->first_maybe_whiteout && rdd->dentry)
 252                 err = ovl_check_whiteouts(realpath->dentry, rdd);
 253
 254         fput(realfile);
 255
 256         return err;
 257 }
 258
 259 static void ovl_dir_reset(struct file *file)
 260 {
 261         struct ovl_dir_file *od = file->private_data;
 262         struct ovl_dir_cache *cache = od->cache;
 263         struct dentry *dentry = file->f_path.dentry;
 264         enum ovl_path_type type = ovl_path_type(dentry);
 265
 266         if (cache && ovl_dentry_version_get(dentry) != cache->version) {
 267                 ovl_cache_put(od, dentry);
 268                 od->cache = NULL;
 269                 od->cursor = NULL;
 270         }
 271         WARN_ON(!od->is_real && !OVL_TYPE_MERGE(type));
 272         if (od->is_real && OVL_TYPE_MERGE(type))
 273                 od->is_real = false;
 274 }
 275
 276 static int ovl_dir_read_merged(struct dentry *dentry, struct list_head *list)
 277 {
 278         int err;
 279         struct path realpath;
 280         struct ovl_readdir_data rdd = {
 281                 .ctx.actor = ovl_fill_merge,
 282                 .dentry = dentry,
 283                 .list = list,
 284                 .root = RB_ROOT,
 285                 .is_lowest = false,
 286         };
 287         int idx, next;
 288
 289         for (idx = 0; idx != -1; idx = next) {
 290                 next = ovl_path_next(idx, dentry, &realpath);
 291
 292                 if (next != -1) {
 293                         err = ovl_dir_read(&realpath, &rdd);
 294                         if (err)
 295                                 break;
 296                 } else {
 297                         /*
 298                          * Insert lowest layer entries before upper ones, this
 299                          * allows offsets to be reasonably constant
 300                          */
 301                         list_add(&rdd.middle, rdd.list);
 302                         rdd.is_lowest = true;
 303                         err = ovl_dir_read(&realpath, &rdd);
 304                         list_del(&rdd.middle);
 305                 }
 306         }
 307         return err;
 308 }
 309
 310 static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
 311 {
 312         struct list_head *p;
 313         loff_t off = 0;
 314
 315         list_for_each(p, &od->cache->entries) {
 316                 if (off >= pos)
 317                         break;
 318                 off++;
 319         }
 320         /* Cursor is safe since the cache is stable */
 321         od->cursor = p;
 322 }
 323
 324 static struct ovl_dir_cache *ovl_cache_get(struct dentry *dentry)
 325 {
 326         int res;
 327         struct ovl_dir_cache *cache;
 328
 329         cache = ovl_dir_cache(dentry);
 330         if (cache && ovl_dentry_version_get(dentry) == cache->version) {
 331                 cache->refcount++;
 332                 return cache;
 333         }
 334         ovl_set_dir_cache(dentry, NULL);
 335
 336         cache = kzalloc(sizeof(struct ovl_dir_cache), GFP_KERNEL);
 337         if (!cache)
 338                 return ERR_PTR(-ENOMEM);
 339
 340         cache->refcount = 1;
 341         INIT_LIST_HEAD(&cache->entries);
 342
 343         res = ovl_dir_read_merged(dentry, &cache->entries);
 344         if (res) {
 345                 ovl_cache_free(&cache->entries);
 346                 kfree(cache);
 347                 return ERR_PTR(res);
 348         }
 349
 350         cache->version = ovl_dentry_version_get(dentry);
 351         ovl_set_dir_cache(dentry, cache);
 352
 353         return cache;
 354 }
 355
 356 static int ovl_iterate(struct file *file, struct dir_context *ctx)
 357 {
 358         struct ovl_dir_file *od = file->private_data;
 359         struct dentry *dentry = file->f_path.dentry;
 360         struct ovl_cache_entry *p;
 361
 362         if (!ctx->pos)
 363                 ovl_dir_reset(file);
 364
 365         if (od->is_real)
 366                 return iterate_dir(od->realfile, ctx);
 367
 368         if (!od->cache) {
 369                 struct ovl_dir_cache *cache;
 370
 371                 cache = ovl_cache_get(dentry);
 372                 if (IS_ERR(cache))
 373                         return PTR_ERR(cache);
 374
 375                 od->cache = cache;
 376                 ovl_seek_cursor(od, ctx->pos);
 377         }
 378
 379         while (od->cursor != &od->cache->entries) {
 380                 p = list_entry(od->cursor, struct ovl_cache_entry, l_node);
 381                 if (!p->is_whiteout)
 382                         if (!dir_emit(ctx, p->name, p->len, p->ino, p->type))
 383                                 break;
 384                 od->cursor = p->l_node.next;
 385                 ctx->pos++;
 386         }
 387         return 0;
 388 }
 389
 390 static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
 391 {
 392         loff_t res;
 393         struct ovl_dir_file *od = file->private_data;
 394
 395         inode_lock(file_inode(file));
 396         if (!file->f_pos)
 397                 ovl_dir_reset(file);
 398
 399         if (od->is_real) {
 400                 res = vfs_llseek(od->realfile, offset, origin);
 401                 file->f_pos = od->realfile->f_pos;
 402         } else {
 403                 res = -EINVAL;
 404
 405                 switch (origin) {
 406                 case SEEK_CUR:
 407                         offset += file->f_pos;
 408                         break;
 409                 case SEEK_SET:
 410                         break;
 411                 default:
 412                         goto out_unlock;
 413                 }
 414                 if (offset < 0)
 415                         goto out_unlock;
 416
 417                 if (offset != file->f_pos) {
 418                         file->f_pos = offset;
 419                         if (od->cache)
 420                                 ovl_seek_cursor(od, offset);
 421                 }
 422                 res = offset;
 423         }
 424 out_unlock:
 425         inode_unlock(file_inode(file));
 426
 427         return res;
 428 }
 429
 430 static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
 431                          int datasync)
 432 {
 433         struct ovl_dir_file *od = file->private_data;
 434         struct dentry *dentry = file->f_path.dentry;
 435         struct file *realfile = od->realfile;
 436
 437         /*
 438          * Need to check if we started out being a lower dir, but got copied up
 439          */
 440         if (!od->is_upper && OVL_TYPE_UPPER(ovl_path_type(dentry))) {
 441                 struct inode *inode = file_inode(file);
 442
 443                 realfile = lockless_dereference(od->upperfile);
 444                 if (!realfile) {
 445                         struct path upperpath;
 446
 447                         ovl_path_upper(dentry, &upperpath);
 448                         realfile = ovl_path_open(&upperpath, O_RDONLY);
 449                         smp_mb__before_spinlock();
 450                         inode_lock(inode);
 451                         if (!od->upperfile) {
 452                                 if (IS_ERR(realfile)) {
 453                                         inode_unlock(inode);
 454                                         return PTR_ERR(realfile);
 455                                 }
 456                                 od->upperfile = realfile;
 457                         } else {
 458                                 /* somebody has beaten us to it */
 459                                 if (!IS_ERR(realfile))
 460                                         fput(realfile);
 461                                 realfile = od->upperfile;
 462                         }
 463                         inode_unlock(inode);
 464                 }
 465         }
 466
 467         return vfs_fsync_range(realfile, start, end, datasync);
 468 }
 469
 470 static int ovl_dir_release(struct inode *inode, struct file *file)
 471 {
 472         struct ovl_dir_file *od = file->private_data;
 473
 474         if (od->cache) {
 475                 inode_lock(inode);
 476                 ovl_cache_put(od, file->f_path.dentry);
 477                 inode_unlock(inode);
 478         }
 479         fput(od->realfile);
 480         if (od->upperfile)
 481                 fput(od->upperfile);
 482         kfree(od);
 483
 484         return 0;
 485 }
 486
 487 static int ovl_dir_open(struct inode *inode, struct file *file)
 488 {
 489         struct path realpath;
 490         struct file *realfile;
 491         struct ovl_dir_file *od;
 492         enum ovl_path_type type;
 493
 494         od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
 495         if (!od)
 496                 return -ENOMEM;
 497
 498         type = ovl_path_real(file->f_path.dentry, &realpath);
 499         realfile = ovl_path_open(&realpath, file->f_flags);
 500         if (IS_ERR(realfile)) {
 501                 kfree(od);
 502                 return PTR_ERR(realfile);
 503         }
 504         od->realfile = realfile;
 505         od->is_real = !OVL_TYPE_MERGE(type);
 506         od->is_upper = OVL_TYPE_UPPER(type);
 507         file->private_data = od;
 508
 509         return 0;
 510 }
 511
 512 const struct file_operations ovl_dir_operations = {
 513         .read           = generic_read_dir,
 514         .open           = ovl_dir_open,
 515         .iterate        = ovl_iterate,
 516         .llseek         = ovl_dir_llseek,
 517         .fsync          = ovl_dir_fsync,
 518         .release        = ovl_dir_release,
 519 };
 520
 521 int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
 522 {
 523         int err;
 524         struct ovl_cache_entry *p;
 525
 526         err = ovl_dir_read_merged(dentry, list);
 527         if (err)
 528                 return err;
 529
 530         err = 0;
 531
 532         list_for_each_entry(p, list, l_node) {
 533                 if (p->is_whiteout)
 534                         continue;
 535
 536                 if (p->name[0] == '.') {
 537                         if (p->len == 1)
 538                                 continue;
 539                         if (p->len == 2 && p->name[1] == '.')
 540                                 continue;
 541                 }
 542                 err = -ENOTEMPTY;
 543                 break;
 544         }
 545
 546         return err;
 547 }
 548
 549 void ovl_cleanup_whiteouts(struct dentry *upper, struct list_head *list)
 550 {
 551         struct ovl_cache_entry *p;
 552
 553         inode_lock_nested(upper->d_inode, I_MUTEX_CHILD);
 554         list_for_each_entry(p, list, l_node) {
 555                 struct dentry *dentry;
 556
 557                 if (!p->is_whiteout)
 558                         continue;
 559
 560                 dentry = lookup_one_len(p->name, upper, p->len);
 561                 if (IS_ERR(dentry)) {
 562                         pr_err("overlayfs: lookup '%s/%.*s' failed (%i)\n",
 563                                upper->d_name.name, p->len, p->name,
 564                                (int) PTR_ERR(dentry));
 565                         continue;
 566                 }
 567                 if (dentry->d_inode)
 568                         ovl_cleanup(upper->d_inode, dentry);
 569                 dput(dentry);
 570         }
 571         inode_unlock(upper->d_inode);
 572 }
 573
 574 static int ovl_check_d_type(struct dir_context *ctx, const char *name,
 575                           int namelen, loff_t offset, u64 ino,
 576                           unsigned int d_type)
 577 {
 578         struct ovl_readdir_data *rdd =
 579                 container_of(ctx, struct ovl_readdir_data, ctx);
 580
 581         /* Even if d_type is not supported, DT_DIR is returned for . and .. */
 582         if (!strncmp(name, ".", namelen) || !strncmp(name, "..", namelen))
 583                 return 0;
 584
 585         if (d_type != DT_UNKNOWN)
 586                 rdd->d_type_supported = true;
 587
 588         return 0;
 589 }
 590
 591 /*
 592  * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
 593  * if error is encountered.
 594  */
 595 int ovl_check_d_type_supported(struct path *realpath)
 596 {
 597         int err;
 598         struct ovl_readdir_data rdd = {
 599                 .ctx.actor = ovl_check_d_type,
 600                 .d_type_supported = false,
 601         };
 602
 603         err = ovl_dir_read(realpath, &rdd);
 604         if (err)
 605                 return err;
 606
 607         return rdd.d_type_supported;
 608 }
 609
 610 static void ovl_workdir_cleanup_recurse(struct path *path, int level)
 611 {
 612         int err;
 613         struct inode *dir = path->dentry->d_inode;
 614         LIST_HEAD(list);
 615         struct ovl_cache_entry *p;
 616         struct ovl_readdir_data rdd = {
 617                 .ctx.actor = ovl_fill_merge,
 618                 .dentry = NULL,
 619                 .list = &list,
 620                 .root = RB_ROOT,
 621                 .is_lowest = false,
 622         };
 623
 624         err = ovl_dir_read(path, &rdd);
 625         if (err)
 626                 goto out;
 627
 628         inode_lock_nested(dir, I_MUTEX_PARENT);
 629         list_for_each_entry(p, &list, l_node) {
 630                 struct dentry *dentry;
 631
 632                 if (p->name[0] == '.') {
 633                         if (p->len == 1)
 634                                 continue;
 635                         if (p->len == 2 && p->name[1] == '.')
 636                                 continue;
 637                 }
 638                 dentry = lookup_one_len(p->name, path->dentry, p->len);
 639                 if (IS_ERR(dentry))
 640                         continue;
 641                 if (dentry->d_inode)
 642                         ovl_workdir_cleanup(dir, path->mnt, dentry, level);
 643                 dput(dentry);
 644         }
 645         inode_unlock(dir);
 646 out:
 647         ovl_cache_free(&list);
 648 }
 649
 650 void ovl_workdir_cleanup(struct inode *dir, struct vfsmount *mnt,
 651                          struct dentry *dentry, int level)
 652 {
 653         int err;
 654
 655         if (!d_is_dir(dentry) || level > 1) {
 656                 ovl_cleanup(dir, dentry);
 657                 return;
 658         }
 659
 660         err = ovl_do_rmdir(dir, dentry);
 661         if (err) {
 662                 struct path path = { .mnt = mnt, .dentry = dentry };
 663
 664                 inode_unlock(dir);
 665                 ovl_workdir_cleanup_recurse(&path, level + 1);
 666                 inode_lock_nested(dir, I_MUTEX_PARENT);
 667                 ovl_cleanup(dir, dentry);
 668         }
 669 }