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/pagemap.h>
  20 #include <linux/sched.h>
  21 #include <linux/statfs.h>
  22 #include <linux/seq_file.h>
  23 #include "overlayfs.h"
  24
  25 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  26 MODULE_DESCRIPTION("Overlay filesystem");
  27 MODULE_LICENSE("GPL");
  28
  29 struct ovl_config {
  30         char *lowerdir;
  31         char *upperdir;
  32         char *workdir;
  33         bool default_permissions;
  34 };
  35
  36 /* private information held for overlayfs's superblock */
  37 struct ovl_fs {
  38         struct vfsmount *upper_mnt;
  39         unsigned numlower;
  40         struct vfsmount **lower_mnt;
  41         struct dentry *workdir;
  42         long lower_namelen;
  43         /* pathnames of lower and upper dirs, for show_options */
  44         struct ovl_config config;
  45 };
  46
  47 struct ovl_dir_cache;
  48
  49 /* private information held for every overlayfs dentry */
  50 struct ovl_entry {
  51         struct dentry *__upperdentry;
  52         struct ovl_dir_cache *cache;
  53         union {
  54                 struct {
  55                         u64 version;
  56                         bool opaque;
  57                 };
  58                 struct rcu_head rcu;
  59         };
  60         unsigned numlower;
  61         struct path lowerstack[];
  62 };
  63
  64 #define OVL_MAX_STACK 500
  65
  66 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  67 {
  68         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  69 }
  70
  71 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  72 {
  73         struct ovl_entry *oe = dentry->d_fsdata;
  74         enum ovl_path_type type = 0;
  75
  76         if (oe->__upperdentry) {
  77                 type = __OVL_PATH_UPPER;
  78
  79                 if (oe->numlower) {
  80                         if (S_ISDIR(dentry->d_inode->i_mode))
  81                                 type |= __OVL_PATH_MERGE;
  82                 } else if (!oe->opaque) {
  83                         type |= __OVL_PATH_PURE;
  84                 }
  85         } else {
  86                 if (oe->numlower > 1)
  87                         type |= __OVL_PATH_MERGE;
  88         }
  89         return type;
  90 }
  91
  92 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  93 {
  94         return lockless_dereference(oe->__upperdentry);
  95 }
  96
  97 void ovl_path_upper(struct dentry *dentry, struct path *path)
  98 {
  99         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 100         struct ovl_entry *oe = dentry->d_fsdata;
 101
 102         path->mnt = ofs->upper_mnt;
 103         path->dentry = ovl_upperdentry_dereference(oe);
 104 }
 105
 106 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 107 {
 108         enum ovl_path_type type = ovl_path_type(dentry);
 109
 110         if (!OVL_TYPE_UPPER(type))
 111                 ovl_path_lower(dentry, path);
 112         else
 113                 ovl_path_upper(dentry, path);
 114
 115         return type;
 116 }
 117
 118 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 119 {
 120         struct ovl_entry *oe = dentry->d_fsdata;
 121
 122         return ovl_upperdentry_dereference(oe);
 123 }
 124
 125 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 126 {
 127         struct ovl_entry *oe = dentry->d_fsdata;
 128
 129         return __ovl_dentry_lower(oe);
 130 }
 131
 132 struct dentry *ovl_dentry_real(struct dentry *dentry)
 133 {
 134         struct ovl_entry *oe = dentry->d_fsdata;
 135         struct dentry *realdentry;
 136
 137         realdentry = ovl_upperdentry_dereference(oe);
 138         if (!realdentry)
 139                 realdentry = __ovl_dentry_lower(oe);
 140
 141         return realdentry;
 142 }
 143
 144 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 145 {
 146         struct dentry *realdentry;
 147
 148         realdentry = ovl_upperdentry_dereference(oe);
 149         if (realdentry) {
 150                 *is_upper = true;
 151         } else {
 152                 realdentry = __ovl_dentry_lower(oe);
 153                 *is_upper = false;
 154         }
 155         return realdentry;
 156 }
 157
 158 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 159                                     bool is_upper)
 160 {
 161         if (is_upper) {
 162                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 163
 164                 return ofs->upper_mnt;
 165         } else {
 166                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 167         }
 168 }
 169
 170 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 171 {
 172         struct ovl_entry *oe = dentry->d_fsdata;
 173
 174         return oe->cache;
 175 }
 176
 177 bool ovl_is_default_permissions(struct inode *inode)
 178 {
 179         struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 180
 181         return ofs->config.default_permissions;
 182 }
 183
 184 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 185 {
 186         struct ovl_entry *oe = dentry->d_fsdata;
 187
 188         oe->cache = cache;
 189 }
 190
 191 void ovl_path_lower(struct dentry *dentry, struct path *path)
 192 {
 193         struct ovl_entry *oe = dentry->d_fsdata;
 194
 195         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 196 }
 197
 198 int ovl_want_write(struct dentry *dentry)
 199 {
 200         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 201         return mnt_want_write(ofs->upper_mnt);
 202 }
 203
 204 void ovl_drop_write(struct dentry *dentry)
 205 {
 206         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 207         mnt_drop_write(ofs->upper_mnt);
 208 }
 209
 210 struct dentry *ovl_workdir(struct dentry *dentry)
 211 {
 212         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 213         return ofs->workdir;
 214 }
 215
 216 bool ovl_dentry_is_opaque(struct dentry *dentry)
 217 {
 218         struct ovl_entry *oe = dentry->d_fsdata;
 219         return oe->opaque;
 220 }
 221
 222 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 223 {
 224         struct ovl_entry *oe = dentry->d_fsdata;
 225         oe->opaque = opaque;
 226 }
 227
 228 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 229 {
 230         struct ovl_entry *oe = dentry->d_fsdata;
 231
 232         WARN_ON(!inode_is_locked(upperdentry->d_parent->d_inode));
 233         WARN_ON(oe->__upperdentry);
 234         BUG_ON(!upperdentry->d_inode);
 235         /*
 236          * Make sure upperdentry is consistent before making it visible to
 237          * ovl_upperdentry_dereference().
 238          */
 239         smp_wmb();
 240         oe->__upperdentry = upperdentry;
 241 }
 242
 243 void ovl_dentry_version_inc(struct dentry *dentry)
 244 {
 245         struct ovl_entry *oe = dentry->d_fsdata;
 246
 247         WARN_ON(!inode_is_locked(dentry->d_inode));
 248         oe->version++;
 249 }
 250
 251 u64 ovl_dentry_version_get(struct dentry *dentry)
 252 {
 253         struct ovl_entry *oe = dentry->d_fsdata;
 254
 255         WARN_ON(!inode_is_locked(dentry->d_inode));
 256         return oe->version;
 257 }
 258
 259 bool ovl_is_whiteout(struct dentry *dentry)
 260 {
 261         struct inode *inode = dentry->d_inode;
 262
 263         return inode && IS_WHITEOUT(inode);
 264 }
 265
 266 static bool ovl_is_opaquedir(struct dentry *dentry)
 267 {
 268         int res;
 269         char val;
 270         struct inode *inode = dentry->d_inode;
 271
 272         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 273                 return false;
 274
 275         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 276         if (res == 1 && val == 'y')
 277                 return true;
 278
 279         return false;
 280 }
 281
 282 static void ovl_dentry_release(struct dentry *dentry)
 283 {
 284         struct ovl_entry *oe = dentry->d_fsdata;
 285
 286         if (oe) {
 287                 unsigned int i;
 288
 289                 dput(oe->__upperdentry);
 290                 for (i = 0; i < oe->numlower; i++)
 291                         dput(oe->lowerstack[i].dentry);
 292                 kfree_rcu(oe, rcu);
 293         }
 294 }
 295
 296 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 297 {
 298         struct ovl_entry *oe = dentry->d_fsdata;
 299         unsigned int i;
 300         int ret = 1;
 301
 302         for (i = 0; i < oe->numlower; i++) {
 303                 struct dentry *d = oe->lowerstack[i].dentry;
 304
 305                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 306                         ret = d->d_op->d_revalidate(d, flags);
 307                         if (ret < 0)
 308                                 return ret;
 309                         if (!ret) {
 310                                 if (!(flags & LOOKUP_RCU))
 311                                         d_invalidate(d);
 312                                 return -ESTALE;
 313                         }
 314                 }
 315         }
 316         return 1;
 317 }
 318
 319 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 320 {
 321         struct ovl_entry *oe = dentry->d_fsdata;
 322         unsigned int i;
 323         int ret = 1;
 324
 325         for (i = 0; i < oe->numlower; i++) {
 326                 struct dentry *d = oe->lowerstack[i].dentry;
 327
 328                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 329                         ret = d->d_op->d_weak_revalidate(d, flags);
 330                         if (ret <= 0)
 331                                 break;
 332                 }
 333         }
 334         return ret;
 335 }
 336
 337 static const struct dentry_operations ovl_dentry_operations = {
 338         .d_release = ovl_dentry_release,
 339         .d_select_inode = ovl_d_select_inode,
 340 };
 341
 342 static const struct dentry_operations ovl_reval_dentry_operations = {
 343         .d_release = ovl_dentry_release,
 344         .d_select_inode = ovl_d_select_inode,
 345         .d_revalidate = ovl_dentry_revalidate,
 346         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 347 };
 348
 349 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 350 {
 351         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 352         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 353
 354         if (oe)
 355                 oe->numlower = numlower;
 356
 357         return oe;
 358 }
 359
 360 static bool ovl_dentry_remote(struct dentry *dentry)
 361 {
 362         return dentry->d_flags &
 363                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 364 }
 365
 366 static bool ovl_dentry_weird(struct dentry *dentry)
 367 {
 368         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 369                                   DCACHE_MANAGE_TRANSIT |
 370                                   DCACHE_OP_HASH |
 371                                   DCACHE_OP_COMPARE);
 372 }
 373
 374 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 375                                              struct qstr *name)
 376 {
 377         struct dentry *dentry;
 378
 379         inode_lock(dir->d_inode);
 380         dentry = lookup_one_len(name->name, dir, name->len);
 381         inode_unlock(dir->d_inode);
 382
 383         if (IS_ERR(dentry)) {
 384                 if (PTR_ERR(dentry) == -ENOENT)
 385                         dentry = NULL;
 386         } else if (!dentry->d_inode) {
 387                 dput(dentry);
 388                 dentry = NULL;
 389         } else if (ovl_dentry_weird(dentry)) {
 390                 dput(dentry);
 391                 /* Don't support traversing automounts and other weirdness */
 392                 dentry = ERR_PTR(-EREMOTE);
 393         }
 394         return dentry;
 395 }
 396
 397 /*
 398  * Returns next layer in stack starting from top.
 399  * Returns -1 if this is the last layer.
 400  */
 401 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 402 {
 403         struct ovl_entry *oe = dentry->d_fsdata;
 404
 405         BUG_ON(idx < 0);
 406         if (idx == 0) {
 407                 ovl_path_upper(dentry, path);
 408                 if (path->dentry)
 409                         return oe->numlower ? 1 : -1;
 410                 idx++;
 411         }
 412         BUG_ON(idx > oe->numlower);
 413         *path = oe->lowerstack[idx - 1];
 414
 415         return (idx < oe->numlower) ? idx + 1 : -1;
 416 }
 417
 418 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 419                           unsigned int flags)
 420 {
 421         struct ovl_entry *oe;
 422         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 423         struct path *stack = NULL;
 424         struct dentry *upperdir, *upperdentry = NULL;
 425         unsigned int ctr = 0;
 426         struct inode *inode = NULL;
 427         bool upperopaque = false;
 428         struct dentry *this, *prev = NULL;
 429         unsigned int i;
 430         int err;
 431
 432         upperdir = ovl_upperdentry_dereference(poe);
 433         if (upperdir) {
 434                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 435                 err = PTR_ERR(this);
 436                 if (IS_ERR(this))
 437                         goto out;
 438
 439                 if (this) {
 440                         if (unlikely(ovl_dentry_remote(this))) {
 441                                 dput(this);
 442                                 err = -EREMOTE;
 443                                 goto out;
 444                         }
 445                         if (ovl_is_whiteout(this)) {
 446                                 dput(this);
 447                                 this = NULL;
 448                                 upperopaque = true;
 449                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 450                                 upperopaque = true;
 451                         }
 452                 }
 453                 upperdentry = prev = this;
 454         }
 455
 456         if (!upperopaque && poe->numlower) {
 457                 err = -ENOMEM;
 458                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 459                 if (!stack)
 460                         goto out_put_upper;
 461         }
 462
 463         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 464                 bool opaque = false;
 465                 struct path lowerpath = poe->lowerstack[i];
 466
 467                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 468                 err = PTR_ERR(this);
 469                 if (IS_ERR(this)) {
 470                         /*
 471                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 472                          */
 473                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 474                                 continue;
 475                         goto out_put;
 476                 }
 477                 if (!this)
 478                         continue;
 479                 if (ovl_is_whiteout(this)) {
 480                         dput(this);
 481                         break;
 482                 }
 483                 /*
 484                  * Only makes sense to check opaque dir if this is not the
 485                  * lowermost layer.
 486                  */
 487                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 488                         opaque = true;
 489
 490                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 491                              !S_ISDIR(this->d_inode->i_mode))) {
 492                         /*
 493                          * FIXME: check for upper-opaqueness maybe better done
 494                          * in remove code.
 495                          */
 496                         if (prev == upperdentry)
 497                                 upperopaque = true;
 498                         dput(this);
 499                         break;
 500                 }
 501                 /*
 502                  * If this is a non-directory then stop here.
 503                  */
 504                 if (!S_ISDIR(this->d_inode->i_mode))
 505                         opaque = true;
 506
 507                 stack[ctr].dentry = this;
 508                 stack[ctr].mnt = lowerpath.mnt;
 509                 ctr++;
 510                 prev = this;
 511                 if (opaque)
 512                         break;
 513         }
 514
 515         oe = ovl_alloc_entry(ctr);
 516         err = -ENOMEM;
 517         if (!oe)
 518                 goto out_put;
 519
 520         if (upperdentry || ctr) {
 521                 struct dentry *realdentry;
 522
 523                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 524
 525                 err = -ENOMEM;
 526                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 527                                       oe);
 528                 if (!inode)
 529                         goto out_free_oe;
 530                 ovl_copyattr(realdentry->d_inode, inode);
 531         }
 532
 533         oe->opaque = upperopaque;
 534         oe->__upperdentry = upperdentry;
 535         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 536         kfree(stack);
 537         dentry->d_fsdata = oe;
 538         d_add(dentry, inode);
 539
 540         return NULL;
 541
 542 out_free_oe:
 543         kfree(oe);
 544 out_put:
 545         for (i = 0; i < ctr; i++)
 546                 dput(stack[i].dentry);
 547         kfree(stack);
 548 out_put_upper:
 549         dput(upperdentry);
 550 out:
 551         return ERR_PTR(err);
 552 }
 553
 554 struct file *ovl_path_open(struct path *path, int flags)
 555 {
 556         return dentry_open(path, flags, current_cred());
 557 }
 558
 559 static void ovl_put_super(struct super_block *sb)
 560 {
 561         struct ovl_fs *ufs = sb->s_fs_info;
 562         unsigned i;
 563
 564         dput(ufs->workdir);
 565         mntput(ufs->upper_mnt);
 566         for (i = 0; i < ufs->numlower; i++)
 567                 mntput(ufs->lower_mnt[i]);
 568         kfree(ufs->lower_mnt);
 569
 570         kfree(ufs->config.lowerdir);
 571         kfree(ufs->config.upperdir);
 572         kfree(ufs->config.workdir);
 573         kfree(ufs);
 574 }
 575
 576 /**
 577  * ovl_statfs
 578  * @sb: The overlayfs super block
 579  * @buf: The struct kstatfs to fill in with stats
 580  *
 581  * Get the filesystem statistics.  As writes always target the upper layer
 582  * filesystem pass the statfs to the upper filesystem (if it exists)
 583  */
 584 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 585 {
 586         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 587         struct dentry *root_dentry = dentry->d_sb->s_root;
 588         struct path path;
 589         int err;
 590
 591         ovl_path_real(root_dentry, &path);
 592
 593         err = vfs_statfs(&path, buf);
 594         if (!err) {
 595                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 596                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 597         }
 598
 599         return err;
 600 }
 601
 602 /**
 603  * ovl_show_options
 604  *
 605  * Prints the mount options for a given superblock.
 606  * Returns zero; does not fail.
 607  */
 608 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 609 {
 610         struct super_block *sb = dentry->d_sb;
 611         struct ovl_fs *ufs = sb->s_fs_info;
 612
 613         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 614         if (ufs->config.upperdir) {
 615                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 616                 seq_show_option(m, "workdir", ufs->config.workdir);
 617         }
 618         if (ufs->config.default_permissions)
 619                 seq_puts(m, ",default_permissions");
 620         return 0;
 621 }
 622
 623 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 624 {
 625         struct ovl_fs *ufs = sb->s_fs_info;
 626
 627         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 628                 return -EROFS;
 629
 630         return 0;
 631 }
 632
 633 static const struct super_operations ovl_super_operations = {
 634         .put_super      = ovl_put_super,
 635         .statfs         = ovl_statfs,
 636         .show_options   = ovl_show_options,
 637         .remount_fs     = ovl_remount,
 638 };
 639
 640 enum {
 641         OPT_LOWERDIR,
 642         OPT_UPPERDIR,
 643         OPT_WORKDIR,
 644         OPT_DEFAULT_PERMISSIONS,
 645         OPT_ERR,
 646 };
 647
 648 static const match_table_t ovl_tokens = {
 649         {OPT_LOWERDIR,                  "lowerdir=%s"},
 650         {OPT_UPPERDIR,                  "upperdir=%s"},
 651         {OPT_WORKDIR,                   "workdir=%s"},
 652         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 653         {OPT_ERR,                       NULL}
 654 };
 655
 656 static char *ovl_next_opt(char **s)
 657 {
 658         char *sbegin = *s;
 659         char *p;
 660
 661         if (sbegin == NULL)
 662                 return NULL;
 663
 664         for (p = sbegin; *p; p++) {
 665                 if (*p == '\\') {
 666                         p++;
 667                         if (!*p)
 668                                 break;
 669                 } else if (*p == ',') {
 670                         *p = '\0';
 671                         *s = p + 1;
 672                         return sbegin;
 673                 }
 674         }
 675         *s = NULL;
 676         return sbegin;
 677 }
 678
 679 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 680 {
 681         char *p;
 682
 683         while ((p = ovl_next_opt(&opt)) != NULL) {
 684                 int token;
 685                 substring_t args[MAX_OPT_ARGS];
 686
 687                 if (!*p)
 688                         continue;
 689
 690                 token = match_token(p, ovl_tokens, args);
 691                 switch (token) {
 692                 case OPT_UPPERDIR:
 693                         kfree(config->upperdir);
 694                         config->upperdir = match_strdup(&args[0]);
 695                         if (!config->upperdir)
 696                                 return -ENOMEM;
 697                         break;
 698
 699                 case OPT_LOWERDIR:
 700                         kfree(config->lowerdir);
 701                         config->lowerdir = match_strdup(&args[0]);
 702                         if (!config->lowerdir)
 703                                 return -ENOMEM;
 704                         break;
 705
 706                 case OPT_WORKDIR:
 707                         kfree(config->workdir);
 708                         config->workdir = match_strdup(&args[0]);
 709                         if (!config->workdir)
 710                                 return -ENOMEM;
 711                         break;
 712
 713                 case OPT_DEFAULT_PERMISSIONS:
 714                         config->default_permissions = true;
 715                         break;
 716
 717                 default:
 718                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 719                         return -EINVAL;
 720                 }
 721         }
 722
 723         /* Workdir is useless in non-upper mount */
 724         if (!config->upperdir && config->workdir) {
 725                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 726                         config->workdir);
 727                 kfree(config->workdir);
 728                 config->workdir = NULL;
 729         }
 730
 731         return 0;
 732 }
 733
 734 #define OVL_WORKDIR_NAME "work"
 735
 736 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 737                                          struct dentry *dentry)
 738 {
 739         struct inode *dir = dentry->d_inode;
 740         struct dentry *work;
 741         int err;
 742         bool retried = false;
 743
 744         err = mnt_want_write(mnt);
 745         if (err)
 746                 return ERR_PTR(err);
 747
 748         inode_lock_nested(dir, I_MUTEX_PARENT);
 749 retry:
 750         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 751                               strlen(OVL_WORKDIR_NAME));
 752
 753         if (!IS_ERR(work)) {
 754                 struct kstat stat = {
 755                         .mode = S_IFDIR | 0,
 756                 };
 757
 758                 if (work->d_inode) {
 759                         err = -EEXIST;
 760                         if (retried)
 761                                 goto out_dput;
 762
 763                         retried = true;
 764                         ovl_cleanup(dir, work);
 765                         dput(work);
 766                         goto retry;
 767                 }
 768
 769                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 770                 if (err)
 771                         goto out_dput;
 772         }
 773 out_unlock:
 774         inode_unlock(dir);
 775         mnt_drop_write(mnt);
 776
 777         return work;
 778
 779 out_dput:
 780         dput(work);
 781         work = ERR_PTR(err);
 782         goto out_unlock;
 783 }
 784
 785 static void ovl_unescape(char *s)
 786 {
 787         char *d = s;
 788
 789         for (;; s++, d++) {
 790                 if (*s == '\\')
 791                         s++;
 792                 *d = *s;
 793                 if (!*s)
 794                         break;
 795         }
 796 }
 797
 798 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 799 {
 800         int err = -EINVAL;
 801
 802         if (!*name) {
 803                 pr_err("overlayfs: empty lowerdir\n");
 804                 goto out;
 805         }
 806         err = kern_path(name, LOOKUP_FOLLOW, path);
 807         if (err) {
 808                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 809                 goto out;
 810         }
 811         err = -EINVAL;
 812         if (ovl_dentry_weird(path->dentry)) {
 813                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 814                 goto out_put;
 815         }
 816         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 817                 pr_err("overlayfs: '%s' not a directory\n", name);
 818                 goto out_put;
 819         }
 820         return 0;
 821
 822 out_put:
 823         path_put(path);
 824 out:
 825         return err;
 826 }
 827
 828 static int ovl_mount_dir(const char *name, struct path *path)
 829 {
 830         int err = -ENOMEM;
 831         char *tmp = kstrdup(name, GFP_KERNEL);
 832
 833         if (tmp) {
 834                 ovl_unescape(tmp);
 835                 err = ovl_mount_dir_noesc(tmp, path);
 836
 837                 if (!err)
 838                         if (ovl_dentry_remote(path->dentry)) {
 839                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 840                                        tmp);
 841                                 path_put(path);
 842                                 err = -EINVAL;
 843                         }
 844                 kfree(tmp);
 845         }
 846         return err;
 847 }
 848
 849 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 850                          int *stack_depth, bool *remote)
 851 {
 852         int err;
 853         struct kstatfs statfs;
 854
 855         err = ovl_mount_dir_noesc(name, path);
 856         if (err)
 857                 goto out;
 858
 859         err = vfs_statfs(path, &statfs);
 860         if (err) {
 861                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 862                 goto out_put;
 863         }
 864         *namelen = max(*namelen, statfs.f_namelen);
 865         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 866
 867         if (ovl_dentry_remote(path->dentry))
 868                 *remote = true;
 869
 870         return 0;
 871
 872 out_put:
 873         path_put(path);
 874 out:
 875         return err;
 876 }
 877
 878 /* Workdir should not be subdir of upperdir and vice versa */
 879 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 880 {
 881         bool ok = false;
 882
 883         if (workdir != upperdir) {
 884                 ok = (lock_rename(workdir, upperdir) == NULL);
 885                 unlock_rename(workdir, upperdir);
 886         }
 887         return ok;
 888 }
 889
 890 static unsigned int ovl_split_lowerdirs(char *str)
 891 {
 892         unsigned int ctr = 1;
 893         char *s, *d;
 894
 895         for (s = d = str;; s++, d++) {
 896                 if (*s == '\\') {
 897                         s++;
 898                 } else if (*s == ':') {
 899                         *d = '\0';
 900                         ctr++;
 901                         continue;
 902                 }
 903                 *d = *s;
 904                 if (!*s)
 905                         break;
 906         }
 907         return ctr;
 908 }
 909
 910 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 911 {
 912         struct path upperpath = { NULL, NULL };
 913         struct path workpath = { NULL, NULL };
 914         struct dentry *root_dentry;
 915         struct ovl_entry *oe;
 916         struct ovl_fs *ufs;
 917         struct path *stack = NULL;
 918         char *lowertmp;
 919         char *lower;
 920         unsigned int numlower;
 921         unsigned int stacklen = 0;
 922         unsigned int i;
 923         bool remote = false;
 924         int err;
 925
 926         err = -ENOMEM;
 927         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 928         if (!ufs)
 929                 goto out;
 930
 931         err = ovl_parse_opt((char *) data, &ufs->config);
 932         if (err)
 933                 goto out_free_config;
 934
 935         err = -EINVAL;
 936         if (!ufs->config.lowerdir) {
 937                 pr_err("overlayfs: missing 'lowerdir'\n");
 938                 goto out_free_config;
 939         }
 940
 941         sb->s_stack_depth = 0;
 942         sb->s_maxbytes = MAX_LFS_FILESIZE;
 943         if (ufs->config.upperdir) {
 944                 if (!ufs->config.workdir) {
 945                         pr_err("overlayfs: missing 'workdir'\n");
 946                         goto out_free_config;
 947                 }
 948
 949                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 950                 if (err)
 951                         goto out_free_config;
 952
 953                 /* Upper fs should not be r/o */
 954                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 955                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 956                         err = -EINVAL;
 957                         goto out_put_upperpath;
 958                 }
 959
 960                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 961                 if (err)
 962                         goto out_put_upperpath;
 963
 964                 err = -EINVAL;
 965                 if (upperpath.mnt != workpath.mnt) {
 966                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 967                         goto out_put_workpath;
 968                 }
 969                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 970                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 971                         goto out_put_workpath;
 972                 }
 973                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 974         }
 975         err = -ENOMEM;
 976         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 977         if (!lowertmp)
 978                 goto out_put_workpath;
 979
 980         err = -EINVAL;
 981         stacklen = ovl_split_lowerdirs(lowertmp);
 982         if (stacklen > OVL_MAX_STACK) {
 983                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 984                        OVL_MAX_STACK);
 985                 goto out_free_lowertmp;
 986         } else if (!ufs->config.upperdir && stacklen == 1) {
 987                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 988                 goto out_free_lowertmp;
 989         }
 990
 991         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
 992         if (!stack)
 993                 goto out_free_lowertmp;
 994
 995         lower = lowertmp;
 996         for (numlower = 0; numlower < stacklen; numlower++) {
 997                 err = ovl_lower_dir(lower, &stack[numlower],
 998                                     &ufs->lower_namelen, &sb->s_stack_depth,
 999                                     &remote);
1000                 if (err)
1001                         goto out_put_lowerpath;
1002
1003                 lower = strchr(lower, '\0') + 1;
1004         }
1005
1006         err = -EINVAL;
1007         sb->s_stack_depth++;
1008         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1009                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1010                 goto out_put_lowerpath;
1011         }
1012
1013         if (ufs->config.upperdir) {
1014                 ufs->upper_mnt = clone_private_mount(&upperpath);
1015                 err = PTR_ERR(ufs->upper_mnt);
1016                 if (IS_ERR(ufs->upper_mnt)) {
1017                         pr_err("overlayfs: failed to clone upperpath\n");
1018                         goto out_put_lowerpath;
1019                 }
1020
1021                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1022                 err = PTR_ERR(ufs->workdir);
1023                 if (IS_ERR(ufs->workdir)) {
1024                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1025                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1026                         sb->s_flags |= MS_RDONLY;
1027                         ufs->workdir = NULL;
1028                 }
1029         }
1030
1031         err = -ENOMEM;
1032         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1033         if (ufs->lower_mnt == NULL)
1034                 goto out_put_workdir;
1035         for (i = 0; i < numlower; i++) {
1036                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1037
1038                 err = PTR_ERR(mnt);
1039                 if (IS_ERR(mnt)) {
1040                         pr_err("overlayfs: failed to clone lowerpath\n");
1041                         goto out_put_lower_mnt;
1042                 }
1043                 /*
1044                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1045                  * will fail instead of modifying lower fs.
1046                  */
1047                 mnt->mnt_flags |= MNT_READONLY;
1048
1049                 ufs->lower_mnt[ufs->numlower] = mnt;
1050                 ufs->numlower++;
1051         }
1052
1053         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1054         if (!ufs->upper_mnt)
1055                 sb->s_flags |= MS_RDONLY;
1056
1057         if (remote)
1058                 sb->s_d_op = &ovl_reval_dentry_operations;
1059         else
1060                 sb->s_d_op = &ovl_dentry_operations;
1061
1062         err = -ENOMEM;
1063         oe = ovl_alloc_entry(numlower);
1064         if (!oe)
1065                 goto out_put_lower_mnt;
1066
1067         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1068         if (!root_dentry)
1069                 goto out_free_oe;
1070
1071         mntput(upperpath.mnt);
1072         for (i = 0; i < numlower; i++)
1073                 mntput(stack[i].mnt);
1074         path_put(&workpath);
1075         kfree(lowertmp);
1076
1077         oe->__upperdentry = upperpath.dentry;
1078         for (i = 0; i < numlower; i++) {
1079                 oe->lowerstack[i].dentry = stack[i].dentry;
1080                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1081         }
1082         kfree(stack);
1083
1084         root_dentry->d_fsdata = oe;
1085
1086         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1087                      root_dentry->d_inode);
1088
1089         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1090         sb->s_op = &ovl_super_operations;
1091         sb->s_root = root_dentry;
1092         sb->s_fs_info = ufs;
1093
1094         return 0;
1095
1096 out_free_oe:
1097         kfree(oe);
1098 out_put_lower_mnt:
1099         for (i = 0; i < ufs->numlower; i++)
1100                 mntput(ufs->lower_mnt[i]);
1101         kfree(ufs->lower_mnt);
1102 out_put_workdir:
1103         dput(ufs->workdir);
1104         mntput(ufs->upper_mnt);
1105 out_put_lowerpath:
1106         for (i = 0; i < numlower; i++)
1107                 path_put(&stack[i]);
1108         kfree(stack);
1109 out_free_lowertmp:
1110         kfree(lowertmp);
1111 out_put_workpath:
1112         path_put(&workpath);
1113 out_put_upperpath:
1114         path_put(&upperpath);
1115 out_free_config:
1116         kfree(ufs->config.lowerdir);
1117         kfree(ufs->config.upperdir);
1118         kfree(ufs->config.workdir);
1119         kfree(ufs);
1120 out:
1121         return err;
1122 }
1123
1124 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1125                                 const char *dev_name, void *raw_data)
1126 {
1127         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1128 }
1129
1130 static struct file_system_type ovl_fs_type = {
1131         .owner          = THIS_MODULE,
1132         .name           = "overlay",
1133         .mount          = ovl_mount,
1134         .kill_sb        = kill_anon_super,
1135 };
1136 MODULE_ALIAS_FS("overlay");
1137
1138 static int __init ovl_init(void)
1139 {
1140         return register_filesystem(&ovl_fs_type);
1141 }
1142
1143 static void __exit ovl_exit(void)
1144 {
1145         unregister_filesystem(&ovl_fs_type);
1146 }
1147
1148 module_init(ovl_init);
1149 module_exit(ovl_exit);