fs/overlayfs/super.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/namei.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/xattr.h>
  14 #include <linux/security.h>
  15 #include <linux/mount.h>
  16 #include <linux/slab.h>
  17 #include <linux/parser.h>
  18 #include <linux/module.h>
  19 #include <linux/sched.h>
  20 #include <linux/statfs.h>
  21 #include <linux/seq_file.h>
  22 #include "overlayfs.h"
  23
  24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  25 MODULE_DESCRIPTION("Overlay filesystem");
  26 MODULE_LICENSE("GPL");
  27
  28 struct ovl_config {
  29         char *lowerdir;
  30         char *upperdir;
  31         char *workdir;
  32         bool default_permissions;
  33 };
  34
  35 /* private information held for overlayfs's superblock */
  36 struct ovl_fs {
  37         struct vfsmount *upper_mnt;
  38         unsigned numlower;
  39         struct vfsmount **lower_mnt;
  40         struct dentry *workdir;
  41         long lower_namelen;
  42         /* pathnames of lower and upper dirs, for show_options */
  43         struct ovl_config config;
  44 };
  45
  46 struct ovl_dir_cache;
  47
  48 /* private information held for every overlayfs dentry */
  49 struct ovl_entry {
  50         struct dentry *__upperdentry;
  51         struct ovl_dir_cache *cache;
  52         union {
  53                 struct {
  54                         u64 version;
  55                         bool opaque;
  56                 };
  57                 struct rcu_head rcu;
  58         };
  59         unsigned numlower;
  60         struct path lowerstack[];
  61 };
  62
  63 #define OVL_MAX_STACK 500
  64
  65 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  66 {
  67         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  68 }
  69
  70 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  71 {
  72         struct ovl_entry *oe = dentry->d_fsdata;
  73         enum ovl_path_type type = 0;
  74
  75         if (oe->__upperdentry) {
  76                 type = __OVL_PATH_UPPER;
  77
  78                 if (oe->numlower) {
  79                         if (S_ISDIR(dentry->d_inode->i_mode))
  80                                 type |= __OVL_PATH_MERGE;
  81                 } else if (!oe->opaque) {
  82                         type |= __OVL_PATH_PURE;
  83                 }
  84         } else {
  85                 if (oe->numlower > 1)
  86                         type |= __OVL_PATH_MERGE;
  87         }
  88         return type;
  89 }
  90
  91 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  92 {
  93         return lockless_dereference(oe->__upperdentry);
  94 }
  95
  96 void ovl_path_upper(struct dentry *dentry, struct path *path)
  97 {
  98         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  99         struct ovl_entry *oe = dentry->d_fsdata;
 100
 101         path->mnt = ofs->upper_mnt;
 102         path->dentry = ovl_upperdentry_dereference(oe);
 103 }
 104
 105 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 106 {
 107         enum ovl_path_type type = ovl_path_type(dentry);
 108
 109         if (!OVL_TYPE_UPPER(type))
 110                 ovl_path_lower(dentry, path);
 111         else
 112                 ovl_path_upper(dentry, path);
 113
 114         return type;
 115 }
 116
 117 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 118 {
 119         struct ovl_entry *oe = dentry->d_fsdata;
 120
 121         return ovl_upperdentry_dereference(oe);
 122 }
 123
 124 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 125 {
 126         struct ovl_entry *oe = dentry->d_fsdata;
 127
 128         return __ovl_dentry_lower(oe);
 129 }
 130
 131 struct dentry *ovl_dentry_real(struct dentry *dentry)
 132 {
 133         struct ovl_entry *oe = dentry->d_fsdata;
 134         struct dentry *realdentry;
 135
 136         realdentry = ovl_upperdentry_dereference(oe);
 137         if (!realdentry)
 138                 realdentry = __ovl_dentry_lower(oe);
 139
 140         return realdentry;
 141 }
 142
 143 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 144 {
 145         struct dentry *realdentry;
 146
 147         realdentry = ovl_upperdentry_dereference(oe);
 148         if (realdentry) {
 149                 *is_upper = true;
 150         } else {
 151                 realdentry = __ovl_dentry_lower(oe);
 152                 *is_upper = false;
 153         }
 154         return realdentry;
 155 }
 156
 157 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 158                                     bool is_upper)
 159 {
 160         if (is_upper) {
 161                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 162
 163                 return ofs->upper_mnt;
 164         } else {
 165                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 166         }
 167 }
 168
 169 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 170 {
 171         struct ovl_entry *oe = dentry->d_fsdata;
 172
 173         return oe->cache;
 174 }
 175
 176 bool ovl_is_default_permissions(struct inode *inode)
 177 {
 178         struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 179
 180         return ofs->config.default_permissions;
 181 }
 182
 183 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 184 {
 185         struct ovl_entry *oe = dentry->d_fsdata;
 186
 187         oe->cache = cache;
 188 }
 189
 190 void ovl_path_lower(struct dentry *dentry, struct path *path)
 191 {
 192         struct ovl_entry *oe = dentry->d_fsdata;
 193
 194         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 195 }
 196
 197 int ovl_want_write(struct dentry *dentry)
 198 {
 199         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 200         return mnt_want_write(ofs->upper_mnt);
 201 }
 202
 203 void ovl_drop_write(struct dentry *dentry)
 204 {
 205         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 206         mnt_drop_write(ofs->upper_mnt);
 207 }
 208
 209 struct dentry *ovl_workdir(struct dentry *dentry)
 210 {
 211         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 212         return ofs->workdir;
 213 }
 214
 215 bool ovl_dentry_is_opaque(struct dentry *dentry)
 216 {
 217         struct ovl_entry *oe = dentry->d_fsdata;
 218         return oe->opaque;
 219 }
 220
 221 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 222 {
 223         struct ovl_entry *oe = dentry->d_fsdata;
 224         oe->opaque = opaque;
 225 }
 226
 227 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 228 {
 229         struct ovl_entry *oe = dentry->d_fsdata;
 230
 231         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
 232         WARN_ON(oe->__upperdentry);
 233         BUG_ON(!upperdentry->d_inode);
 234         /*
 235          * Make sure upperdentry is consistent before making it visible to
 236          * ovl_upperdentry_dereference().
 237          */
 238         smp_wmb();
 239         oe->__upperdentry = upperdentry;
 240 }
 241
 242 void ovl_dentry_version_inc(struct dentry *dentry)
 243 {
 244         struct ovl_entry *oe = dentry->d_fsdata;
 245
 246         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 247         oe->version++;
 248 }
 249
 250 u64 ovl_dentry_version_get(struct dentry *dentry)
 251 {
 252         struct ovl_entry *oe = dentry->d_fsdata;
 253
 254         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 255         return oe->version;
 256 }
 257
 258 bool ovl_is_whiteout(struct dentry *dentry)
 259 {
 260         struct inode *inode = dentry->d_inode;
 261
 262         return inode && IS_WHITEOUT(inode);
 263 }
 264
 265 static bool ovl_is_opaquedir(struct dentry *dentry)
 266 {
 267         int res;
 268         char val;
 269         struct inode *inode = dentry->d_inode;
 270
 271         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 272                 return false;
 273
 274         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 275         if (res == 1 && val == 'y')
 276                 return true;
 277
 278         return false;
 279 }
 280
 281 static void ovl_dentry_release(struct dentry *dentry)
 282 {
 283         struct ovl_entry *oe = dentry->d_fsdata;
 284
 285         if (oe) {
 286                 unsigned int i;
 287
 288                 dput(oe->__upperdentry);
 289                 for (i = 0; i < oe->numlower; i++)
 290                         dput(oe->lowerstack[i].dentry);
 291                 kfree_rcu(oe, rcu);
 292         }
 293 }
 294
 295 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 296 {
 297         struct ovl_entry *oe = dentry->d_fsdata;
 298         unsigned int i;
 299         int ret = 1;
 300
 301         for (i = 0; i < oe->numlower; i++) {
 302                 struct dentry *d = oe->lowerstack[i].dentry;
 303
 304                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 305                         ret = d->d_op->d_revalidate(d, flags);
 306                         if (ret < 0)
 307                                 return ret;
 308                         if (!ret) {
 309                                 if (!(flags & LOOKUP_RCU))
 310                                         d_invalidate(d);
 311                                 return -ESTALE;
 312                         }
 313                 }
 314         }
 315         return 1;
 316 }
 317
 318 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 319 {
 320         struct ovl_entry *oe = dentry->d_fsdata;
 321         unsigned int i;
 322         int ret = 1;
 323
 324         for (i = 0; i < oe->numlower; i++) {
 325                 struct dentry *d = oe->lowerstack[i].dentry;
 326
 327                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 328                         ret = d->d_op->d_weak_revalidate(d, flags);
 329                         if (ret <= 0)
 330                                 break;
 331                 }
 332         }
 333         return ret;
 334 }
 335
 336 static const struct dentry_operations ovl_dentry_operations = {
 337         .d_release = ovl_dentry_release,
 338         .d_select_inode = ovl_d_select_inode,
 339 };
 340
 341 static const struct dentry_operations ovl_reval_dentry_operations = {
 342         .d_release = ovl_dentry_release,
 343         .d_revalidate = ovl_dentry_revalidate,
 344         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 345 };
 346
 347 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 348 {
 349         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 350         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 351
 352         if (oe)
 353                 oe->numlower = numlower;
 354
 355         return oe;
 356 }
 357
 358 static bool ovl_dentry_remote(struct dentry *dentry)
 359 {
 360         return dentry->d_flags &
 361                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 362 }
 363
 364 static bool ovl_dentry_weird(struct dentry *dentry)
 365 {
 366         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 367                                   DCACHE_MANAGE_TRANSIT |
 368                                   DCACHE_OP_HASH |
 369                                   DCACHE_OP_COMPARE);
 370 }
 371
 372 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 373                                              struct qstr *name)
 374 {
 375         struct dentry *dentry;
 376
 377         mutex_lock(&dir->d_inode->i_mutex);
 378         dentry = lookup_one_len(name->name, dir, name->len);
 379         mutex_unlock(&dir->d_inode->i_mutex);
 380
 381         if (IS_ERR(dentry)) {
 382                 if (PTR_ERR(dentry) == -ENOENT)
 383                         dentry = NULL;
 384         } else if (!dentry->d_inode) {
 385                 dput(dentry);
 386                 dentry = NULL;
 387         } else if (ovl_dentry_weird(dentry)) {
 388                 dput(dentry);
 389                 /* Don't support traversing automounts and other weirdness */
 390                 dentry = ERR_PTR(-EREMOTE);
 391         }
 392         return dentry;
 393 }
 394
 395 /*
 396  * Returns next layer in stack starting from top.
 397  * Returns -1 if this is the last layer.
 398  */
 399 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 400 {
 401         struct ovl_entry *oe = dentry->d_fsdata;
 402
 403         BUG_ON(idx < 0);
 404         if (idx == 0) {
 405                 ovl_path_upper(dentry, path);
 406                 if (path->dentry)
 407                         return oe->numlower ? 1 : -1;
 408                 idx++;
 409         }
 410         BUG_ON(idx > oe->numlower);
 411         *path = oe->lowerstack[idx - 1];
 412
 413         return (idx < oe->numlower) ? idx + 1 : -1;
 414 }
 415
 416 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 417                           unsigned int flags)
 418 {
 419         struct ovl_entry *oe;
 420         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 421         struct path *stack = NULL;
 422         struct dentry *upperdir, *upperdentry = NULL;
 423         unsigned int ctr = 0;
 424         struct inode *inode = NULL;
 425         bool upperopaque = false;
 426         struct dentry *this, *prev = NULL;
 427         unsigned int i;
 428         int err;
 429
 430         upperdir = ovl_upperdentry_dereference(poe);
 431         if (upperdir) {
 432                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 433                 err = PTR_ERR(this);
 434                 if (IS_ERR(this))
 435                         goto out;
 436
 437                 if (this) {
 438                         if (unlikely(ovl_dentry_remote(this))) {
 439                                 dput(this);
 440                                 err = -EREMOTE;
 441                                 goto out;
 442                         }
 443                         if (ovl_is_whiteout(this)) {
 444                                 dput(this);
 445                                 this = NULL;
 446                                 upperopaque = true;
 447                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 448                                 upperopaque = true;
 449                         }
 450                 }
 451                 upperdentry = prev = this;
 452         }
 453
 454         if (!upperopaque && poe->numlower) {
 455                 err = -ENOMEM;
 456                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 457                 if (!stack)
 458                         goto out_put_upper;
 459         }
 460
 461         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 462                 bool opaque = false;
 463                 struct path lowerpath = poe->lowerstack[i];
 464
 465                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 466                 err = PTR_ERR(this);
 467                 if (IS_ERR(this)) {
 468                         /*
 469                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 470                          */
 471                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 472                                 continue;
 473                         goto out_put;
 474                 }
 475                 if (!this)
 476                         continue;
 477                 if (ovl_is_whiteout(this)) {
 478                         dput(this);
 479                         break;
 480                 }
 481                 /*
 482                  * Only makes sense to check opaque dir if this is not the
 483                  * lowermost layer.
 484                  */
 485                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 486                         opaque = true;
 487
 488                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 489                              !S_ISDIR(this->d_inode->i_mode))) {
 490                         /*
 491                          * FIXME: check for upper-opaqueness maybe better done
 492                          * in remove code.
 493                          */
 494                         if (prev == upperdentry)
 495                                 upperopaque = true;
 496                         dput(this);
 497                         break;
 498                 }
 499                 /*
 500                  * If this is a non-directory then stop here.
 501                  */
 502                 if (!S_ISDIR(this->d_inode->i_mode))
 503                         opaque = true;
 504
 505                 stack[ctr].dentry = this;
 506                 stack[ctr].mnt = lowerpath.mnt;
 507                 ctr++;
 508                 prev = this;
 509                 if (opaque)
 510                         break;
 511         }
 512
 513         oe = ovl_alloc_entry(ctr);
 514         err = -ENOMEM;
 515         if (!oe)
 516                 goto out_put;
 517
 518         if (upperdentry || ctr) {
 519                 struct dentry *realdentry;
 520
 521                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 522
 523                 err = -ENOMEM;
 524                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 525                                       oe);
 526                 if (!inode)
 527                         goto out_free_oe;
 528                 ovl_copyattr(realdentry->d_inode, inode);
 529         }
 530
 531         oe->opaque = upperopaque;
 532         oe->__upperdentry = upperdentry;
 533         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 534         kfree(stack);
 535         dentry->d_fsdata = oe;
 536         d_add(dentry, inode);
 537
 538         return NULL;
 539
 540 out_free_oe:
 541         kfree(oe);
 542 out_put:
 543         for (i = 0; i < ctr; i++)
 544                 dput(stack[i].dentry);
 545         kfree(stack);
 546 out_put_upper:
 547         dput(upperdentry);
 548 out:
 549         return ERR_PTR(err);
 550 }
 551
 552 struct file *ovl_path_open(struct path *path, int flags)
 553 {
 554         return dentry_open(path, flags, current_cred());
 555 }
 556
 557 static void ovl_put_super(struct super_block *sb)
 558 {
 559         struct ovl_fs *ufs = sb->s_fs_info;
 560         unsigned i;
 561
 562         dput(ufs->workdir);
 563         mntput(ufs->upper_mnt);
 564         for (i = 0; i < ufs->numlower; i++)
 565                 mntput(ufs->lower_mnt[i]);
 566         kfree(ufs->lower_mnt);
 567
 568         kfree(ufs->config.lowerdir);
 569         kfree(ufs->config.upperdir);
 570         kfree(ufs->config.workdir);
 571         kfree(ufs);
 572 }
 573
 574 /**
 575  * ovl_statfs
 576  * @sb: The overlayfs super block
 577  * @buf: The struct kstatfs to fill in with stats
 578  *
 579  * Get the filesystem statistics.  As writes always target the upper layer
 580  * filesystem pass the statfs to the upper filesystem (if it exists)
 581  */
 582 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 583 {
 584         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 585         struct dentry *root_dentry = dentry->d_sb->s_root;
 586         struct path path;
 587         int err;
 588
 589         ovl_path_real(root_dentry, &path);
 590
 591         err = vfs_statfs(&path, buf);
 592         if (!err) {
 593                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 594                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 595         }
 596
 597         return err;
 598 }
 599
 600 /**
 601  * ovl_show_options
 602  *
 603  * Prints the mount options for a given superblock.
 604  * Returns zero; does not fail.
 605  */
 606 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 607 {
 608         struct super_block *sb = dentry->d_sb;
 609         struct ovl_fs *ufs = sb->s_fs_info;
 610
 611         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 612         if (ufs->config.upperdir) {
 613                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 614                 seq_show_option(m, "workdir", ufs->config.workdir);
 615         }
 616         if (ufs->config.default_permissions)
 617                 seq_puts(m, ",default_permissions");
 618         return 0;
 619 }
 620
 621 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 622 {
 623         struct ovl_fs *ufs = sb->s_fs_info;
 624
 625         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 626                 return -EROFS;
 627
 628         return 0;
 629 }
 630
 631 static const struct super_operations ovl_super_operations = {
 632         .put_super      = ovl_put_super,
 633         .statfs         = ovl_statfs,
 634         .show_options   = ovl_show_options,
 635         .remount_fs     = ovl_remount,
 636 };
 637
 638 enum {
 639         OPT_LOWERDIR,
 640         OPT_UPPERDIR,
 641         OPT_WORKDIR,
 642         OPT_DEFAULT_PERMISSIONS,
 643         OPT_ERR,
 644 };
 645
 646 static const match_table_t ovl_tokens = {
 647         {OPT_LOWERDIR,                  "lowerdir=%s"},
 648         {OPT_UPPERDIR,                  "upperdir=%s"},
 649         {OPT_WORKDIR,                   "workdir=%s"},
 650         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 651         {OPT_ERR,                       NULL}
 652 };
 653
 654 static char *ovl_next_opt(char **s)
 655 {
 656         char *sbegin = *s;
 657         char *p;
 658
 659         if (sbegin == NULL)
 660                 return NULL;
 661
 662         for (p = sbegin; *p; p++) {
 663                 if (*p == '\\') {
 664                         p++;
 665                         if (!*p)
 666                                 break;
 667                 } else if (*p == ',') {
 668                         *p = '\0';
 669                         *s = p + 1;
 670                         return sbegin;
 671                 }
 672         }
 673         *s = NULL;
 674         return sbegin;
 675 }
 676
 677 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 678 {
 679         char *p;
 680
 681         while ((p = ovl_next_opt(&opt)) != NULL) {
 682                 int token;
 683                 substring_t args[MAX_OPT_ARGS];
 684
 685                 if (!*p)
 686                         continue;
 687
 688                 token = match_token(p, ovl_tokens, args);
 689                 switch (token) {
 690                 case OPT_UPPERDIR:
 691                         kfree(config->upperdir);
 692                         config->upperdir = match_strdup(&args[0]);
 693                         if (!config->upperdir)
 694                                 return -ENOMEM;
 695                         break;
 696
 697                 case OPT_LOWERDIR:
 698                         kfree(config->lowerdir);
 699                         config->lowerdir = match_strdup(&args[0]);
 700                         if (!config->lowerdir)
 701                                 return -ENOMEM;
 702                         break;
 703
 704                 case OPT_WORKDIR:
 705                         kfree(config->workdir);
 706                         config->workdir = match_strdup(&args[0]);
 707                         if (!config->workdir)
 708                                 return -ENOMEM;
 709                         break;
 710
 711                 case OPT_DEFAULT_PERMISSIONS:
 712                         config->default_permissions = true;
 713                         break;
 714
 715                 default:
 716                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 717                         return -EINVAL;
 718                 }
 719         }
 720
 721         /* Workdir is useless in non-upper mount */
 722         if (!config->upperdir && config->workdir) {
 723                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 724                         config->workdir);
 725                 kfree(config->workdir);
 726                 config->workdir = NULL;
 727         }
 728
 729         return 0;
 730 }
 731
 732 #define OVL_WORKDIR_NAME "work"
 733
 734 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 735                                          struct dentry *dentry)
 736 {
 737         struct inode *dir = dentry->d_inode;
 738         struct dentry *work;
 739         int err;
 740         bool retried = false;
 741
 742         err = mnt_want_write(mnt);
 743         if (err)
 744                 return ERR_PTR(err);
 745
 746         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 747 retry:
 748         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 749                               strlen(OVL_WORKDIR_NAME));
 750
 751         if (!IS_ERR(work)) {
 752                 struct kstat stat = {
 753                         .mode = S_IFDIR | 0,
 754                 };
 755
 756                 if (work->d_inode) {
 757                         err = -EEXIST;
 758                         if (retried)
 759                                 goto out_dput;
 760
 761                         retried = true;
 762                         ovl_cleanup(dir, work);
 763                         dput(work);
 764                         goto retry;
 765                 }
 766
 767                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 768                 if (err)
 769                         goto out_dput;
 770         }
 771 out_unlock:
 772         mutex_unlock(&dir->i_mutex);
 773         mnt_drop_write(mnt);
 774
 775         return work;
 776
 777 out_dput:
 778         dput(work);
 779         work = ERR_PTR(err);
 780         goto out_unlock;
 781 }
 782
 783 static void ovl_unescape(char *s)
 784 {
 785         char *d = s;
 786
 787         for (;; s++, d++) {
 788                 if (*s == '\\')
 789                         s++;
 790                 *d = *s;
 791                 if (!*s)
 792                         break;
 793         }
 794 }
 795
 796 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 797 {
 798         int err = -EINVAL;
 799
 800         if (!*name) {
 801                 pr_err("overlayfs: empty lowerdir\n");
 802                 goto out;
 803         }
 804         err = kern_path(name, LOOKUP_FOLLOW, path);
 805         if (err) {
 806                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 807                 goto out;
 808         }
 809         err = -EINVAL;
 810         if (ovl_dentry_weird(path->dentry)) {
 811                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 812                 goto out_put;
 813         }
 814         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 815                 pr_err("overlayfs: '%s' not a directory\n", name);
 816                 goto out_put;
 817         }
 818         return 0;
 819
 820 out_put:
 821         path_put(path);
 822 out:
 823         return err;
 824 }
 825
 826 static int ovl_mount_dir(const char *name, struct path *path)
 827 {
 828         int err = -ENOMEM;
 829         char *tmp = kstrdup(name, GFP_KERNEL);
 830
 831         if (tmp) {
 832                 ovl_unescape(tmp);
 833                 err = ovl_mount_dir_noesc(tmp, path);
 834
 835                 if (!err)
 836                         if (ovl_dentry_remote(path->dentry)) {
 837                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 838                                        tmp);
 839                                 path_put(path);
 840                                 err = -EINVAL;
 841                         }
 842                 kfree(tmp);
 843         }
 844         return err;
 845 }
 846
 847 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 848                          int *stack_depth, bool *remote)
 849 {
 850         int err;
 851         struct kstatfs statfs;
 852
 853         err = ovl_mount_dir_noesc(name, path);
 854         if (err)
 855                 goto out;
 856
 857         err = vfs_statfs(path, &statfs);
 858         if (err) {
 859                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 860                 goto out_put;
 861         }
 862         *namelen = max(*namelen, statfs.f_namelen);
 863         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 864
 865         if (ovl_dentry_remote(path->dentry))
 866                 *remote = true;
 867
 868         return 0;
 869
 870 out_put:
 871         path_put(path);
 872 out:
 873         return err;
 874 }
 875
 876 /* Workdir should not be subdir of upperdir and vice versa */
 877 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 878 {
 879         bool ok = false;
 880
 881         if (workdir != upperdir) {
 882                 ok = (lock_rename(workdir, upperdir) == NULL);
 883                 unlock_rename(workdir, upperdir);
 884         }
 885         return ok;
 886 }
 887
 888 static unsigned int ovl_split_lowerdirs(char *str)
 889 {
 890         unsigned int ctr = 1;
 891         char *s, *d;
 892
 893         for (s = d = str;; s++, d++) {
 894                 if (*s == '\\') {
 895                         s++;
 896                 } else if (*s == ':') {
 897                         *d = '\0';
 898                         ctr++;
 899                         continue;
 900                 }
 901                 *d = *s;
 902                 if (!*s)
 903                         break;
 904         }
 905         return ctr;
 906 }
 907
 908 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 909 {
 910         struct path upperpath = { NULL, NULL };
 911         struct path workpath = { NULL, NULL };
 912         struct dentry *root_dentry;
 913         struct ovl_entry *oe;
 914         struct ovl_fs *ufs;
 915         struct path *stack = NULL;
 916         char *lowertmp;
 917         char *lower;
 918         unsigned int numlower;
 919         unsigned int stacklen = 0;
 920         unsigned int i;
 921         bool remote = false;
 922         int err;
 923
 924         err = -ENOMEM;
 925         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 926         if (!ufs)
 927                 goto out;
 928
 929         err = ovl_parse_opt((char *) data, &ufs->config);
 930         if (err)
 931                 goto out_free_config;
 932
 933         err = -EINVAL;
 934         if (!ufs->config.lowerdir) {
 935                 pr_err("overlayfs: missing 'lowerdir'\n");
 936                 goto out_free_config;
 937         }
 938
 939         sb->s_stack_depth = 0;
 940         sb->s_maxbytes = MAX_LFS_FILESIZE;
 941         if (ufs->config.upperdir) {
 942                 if (!ufs->config.workdir) {
 943                         pr_err("overlayfs: missing 'workdir'\n");
 944                         goto out_free_config;
 945                 }
 946
 947                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 948                 if (err)
 949                         goto out_free_config;
 950
 951                 /* Upper fs should not be r/o */
 952                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 953                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 954                         err = -EINVAL;
 955                         goto out_put_upperpath;
 956                 }
 957
 958                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 959                 if (err)
 960                         goto out_put_upperpath;
 961
 962                 err = -EINVAL;
 963                 if (upperpath.mnt != workpath.mnt) {
 964                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 965                         goto out_put_workpath;
 966                 }
 967                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 968                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 969                         goto out_put_workpath;
 970                 }
 971                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 972         }
 973         err = -ENOMEM;
 974         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 975         if (!lowertmp)
 976                 goto out_put_workpath;
 977
 978         err = -EINVAL;
 979         stacklen = ovl_split_lowerdirs(lowertmp);
 980         if (stacklen > OVL_MAX_STACK) {
 981                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 982                        OVL_MAX_STACK);
 983                 goto out_free_lowertmp;
 984         } else if (!ufs->config.upperdir && stacklen == 1) {
 985                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 986                 goto out_free_lowertmp;
 987         }
 988
 989         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 990         if (!stack)
 991                 goto out_free_lowertmp;
 992
 993         lower = lowertmp;
 994         for (numlower = 0; numlower < stacklen; numlower++) {
 995                 err = ovl_lower_dir(lower, &stack[numlower],
 996                                     &ufs->lower_namelen, &sb->s_stack_depth,
 997                                     &remote);
 998                 if (err)
 999                         goto out_put_lowerpath;
1000
1001                 lower = strchr(lower, '\0') + 1;
1002         }
1003
1004         err = -EINVAL;
1005         sb->s_stack_depth++;
1006         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1007                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1008                 goto out_put_lowerpath;
1009         }
1010
1011         if (ufs->config.upperdir) {
1012                 ufs->upper_mnt = clone_private_mount(&upperpath);
1013                 err = PTR_ERR(ufs->upper_mnt);
1014                 if (IS_ERR(ufs->upper_mnt)) {
1015                         pr_err("overlayfs: failed to clone upperpath\n");
1016                         goto out_put_lowerpath;
1017                 }
1018
1019                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1020                 err = PTR_ERR(ufs->workdir);
1021                 if (IS_ERR(ufs->workdir)) {
1022                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1023                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1024                         sb->s_flags |= MS_RDONLY;
1025                         ufs->workdir = NULL;
1026                 }
1027         }
1028
1029         err = -ENOMEM;
1030         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1031         if (ufs->lower_mnt == NULL)
1032                 goto out_put_workdir;
1033         for (i = 0; i < numlower; i++) {
1034                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1035
1036                 err = PTR_ERR(mnt);
1037                 if (IS_ERR(mnt)) {
1038                         pr_err("overlayfs: failed to clone lowerpath\n");
1039                         goto out_put_lower_mnt;
1040                 }
1041                 /*
1042                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1043                  * will fail instead of modifying lower fs.
1044                  */
1045                 mnt->mnt_flags |= MNT_READONLY;
1046
1047                 ufs->lower_mnt[ufs->numlower] = mnt;
1048                 ufs->numlower++;
1049         }
1050
1051         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1052         if (!ufs->upper_mnt)
1053                 sb->s_flags |= MS_RDONLY;
1054
1055         if (remote)
1056                 sb->s_d_op = &ovl_reval_dentry_operations;
1057         else
1058                 sb->s_d_op = &ovl_dentry_operations;
1059
1060         err = -ENOMEM;
1061         oe = ovl_alloc_entry(numlower);
1062         if (!oe)
1063                 goto out_put_lower_mnt;
1064
1065         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1066         if (!root_dentry)
1067                 goto out_free_oe;
1068
1069         mntput(upperpath.mnt);
1070         for (i = 0; i < numlower; i++)
1071                 mntput(stack[i].mnt);
1072         path_put(&workpath);
1073         kfree(lowertmp);
1074
1075         oe->__upperdentry = upperpath.dentry;
1076         for (i = 0; i < numlower; i++) {
1077                 oe->lowerstack[i].dentry = stack[i].dentry;
1078                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1079         }
1080         kfree(stack);
1081
1082         root_dentry->d_fsdata = oe;
1083
1084         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1085                      root_dentry->d_inode);
1086
1087         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1088         sb->s_op = &ovl_super_operations;
1089         sb->s_root = root_dentry;
1090         sb->s_fs_info = ufs;
1091
1092         return 0;
1093
1094 out_free_oe:
1095         kfree(oe);
1096 out_put_lower_mnt:
1097         for (i = 0; i < ufs->numlower; i++)
1098                 mntput(ufs->lower_mnt[i]);
1099         kfree(ufs->lower_mnt);
1100 out_put_workdir:
1101         dput(ufs->workdir);
1102         mntput(ufs->upper_mnt);
1103 out_put_lowerpath:
1104         for (i = 0; i < numlower; i++)
1105                 path_put(&stack[i]);
1106         kfree(stack);
1107 out_free_lowertmp:
1108         kfree(lowertmp);
1109 out_put_workpath:
1110         path_put(&workpath);
1111 out_put_upperpath:
1112         path_put(&upperpath);
1113 out_free_config:
1114         kfree(ufs->config.lowerdir);
1115         kfree(ufs->config.upperdir);
1116         kfree(ufs->config.workdir);
1117         kfree(ufs);
1118 out:
1119         return err;
1120 }
1121
1122 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1123                                 const char *dev_name, void *raw_data)
1124 {
1125         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1126 }
1127
1128 static struct file_system_type ovl_fs_type = {
1129         .owner          = THIS_MODULE,
1130         .name           = "overlay",
1131         .mount          = ovl_mount,
1132         .kill_sb        = kill_anon_super,
1133 };
1134 MODULE_ALIAS_FS("overlay");
1135
1136 static int __init ovl_init(void)
1137 {
1138         return register_filesystem(&ovl_fs_type);
1139 }
1140
1141 static void __exit ovl_exit(void)
1142 {
1143         unregister_filesystem(&ovl_fs_type);
1144 }
1145
1146 module_init(ovl_init);
1147 module_exit(ovl_exit);