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 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
  29
  30 struct ovl_config {
  31         char *lowerdir;
  32         char *upperdir;
  33         char *workdir;
  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                 /*
  80                  * Non-dir dentry can hold lower dentry from previous
  81                  * location. Its purity depends only on opaque flag.
  82                  */
  83                 if (oe->numlower && S_ISDIR(dentry->d_inode->i_mode))
  84                         type |= __OVL_PATH_MERGE;
  85                 else if (!oe->opaque)
  86                         type |= __OVL_PATH_PURE;
  87         } else {
  88                 if (oe->numlower > 1)
  89                         type |= __OVL_PATH_MERGE;
  90         }
  91         return type;
  92 }
  93
  94 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  95 {
  96         return lockless_dereference(oe->__upperdentry);
  97 }
  98
  99 void ovl_path_upper(struct dentry *dentry, struct path *path)
 100 {
 101         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 102         struct ovl_entry *oe = dentry->d_fsdata;
 103
 104         path->mnt = ofs->upper_mnt;
 105         path->dentry = ovl_upperdentry_dereference(oe);
 106 }
 107
 108 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 109 {
 110         enum ovl_path_type type = ovl_path_type(dentry);
 111
 112         if (!OVL_TYPE_UPPER(type))
 113                 ovl_path_lower(dentry, path);
 114         else
 115                 ovl_path_upper(dentry, path);
 116
 117         return type;
 118 }
 119
 120 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 121 {
 122         struct ovl_entry *oe = dentry->d_fsdata;
 123
 124         return ovl_upperdentry_dereference(oe);
 125 }
 126
 127 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 128 {
 129         struct ovl_entry *oe = dentry->d_fsdata;
 130
 131         return __ovl_dentry_lower(oe);
 132 }
 133
 134 struct dentry *ovl_dentry_real(struct dentry *dentry)
 135 {
 136         struct ovl_entry *oe = dentry->d_fsdata;
 137         struct dentry *realdentry;
 138
 139         realdentry = ovl_upperdentry_dereference(oe);
 140         if (!realdentry)
 141                 realdentry = __ovl_dentry_lower(oe);
 142
 143         return realdentry;
 144 }
 145
 146 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
 147 {
 148         struct dentry *realdentry;
 149
 150         realdentry = ovl_upperdentry_dereference(oe);
 151         if (realdentry) {
 152                 *is_upper = true;
 153         } else {
 154                 realdentry = __ovl_dentry_lower(oe);
 155                 *is_upper = false;
 156         }
 157         return realdentry;
 158 }
 159
 160 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 161 {
 162         struct ovl_entry *oe = dentry->d_fsdata;
 163
 164         return oe->cache;
 165 }
 166
 167 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 168 {
 169         struct ovl_entry *oe = dentry->d_fsdata;
 170
 171         oe->cache = cache;
 172 }
 173
 174 void ovl_path_lower(struct dentry *dentry, struct path *path)
 175 {
 176         struct ovl_entry *oe = dentry->d_fsdata;
 177
 178         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 179 }
 180
 181 int ovl_want_write(struct dentry *dentry)
 182 {
 183         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 184         return mnt_want_write(ofs->upper_mnt);
 185 }
 186
 187 void ovl_drop_write(struct dentry *dentry)
 188 {
 189         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 190         mnt_drop_write(ofs->upper_mnt);
 191 }
 192
 193 struct dentry *ovl_workdir(struct dentry *dentry)
 194 {
 195         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 196         return ofs->workdir;
 197 }
 198
 199 bool ovl_dentry_is_opaque(struct dentry *dentry)
 200 {
 201         struct ovl_entry *oe = dentry->d_fsdata;
 202         return oe->opaque;
 203 }
 204
 205 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 206 {
 207         struct ovl_entry *oe = dentry->d_fsdata;
 208         oe->opaque = opaque;
 209 }
 210
 211 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 212 {
 213         struct ovl_entry *oe = dentry->d_fsdata;
 214
 215         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
 216         WARN_ON(oe->__upperdentry);
 217         BUG_ON(!upperdentry->d_inode);
 218         /*
 219          * Make sure upperdentry is consistent before making it visible to
 220          * ovl_upperdentry_dereference().
 221          */
 222         smp_wmb();
 223         oe->__upperdentry = upperdentry;
 224 }
 225
 226 void ovl_dentry_version_inc(struct dentry *dentry)
 227 {
 228         struct ovl_entry *oe = dentry->d_fsdata;
 229
 230         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 231         oe->version++;
 232 }
 233
 234 u64 ovl_dentry_version_get(struct dentry *dentry)
 235 {
 236         struct ovl_entry *oe = dentry->d_fsdata;
 237
 238         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
 239         return oe->version;
 240 }
 241
 242 bool ovl_is_whiteout(struct dentry *dentry)
 243 {
 244         struct inode *inode = dentry->d_inode;
 245
 246         return inode && IS_WHITEOUT(inode);
 247 }
 248
 249 static bool ovl_is_opaquedir(struct dentry *dentry)
 250 {
 251         int res;
 252         char val;
 253         struct inode *inode = dentry->d_inode;
 254
 255         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 256                 return false;
 257
 258         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
 259         if (res == 1 && val == 'y')
 260                 return true;
 261
 262         return false;
 263 }
 264
 265 static void ovl_dentry_release(struct dentry *dentry)
 266 {
 267         struct ovl_entry *oe = dentry->d_fsdata;
 268
 269         if (oe) {
 270                 unsigned int i;
 271
 272                 dput(oe->__upperdentry);
 273                 for (i = 0; i < oe->numlower; i++)
 274                         dput(oe->lowerstack[i].dentry);
 275                 kfree_rcu(oe, rcu);
 276         }
 277 }
 278
 279 static struct dentry *ovl_d_real(struct dentry *dentry, struct inode *inode)
 280 {
 281         struct dentry *real;
 282
 283         if (d_is_dir(dentry)) {
 284                 if (!inode || inode == d_inode(dentry))
 285                         return dentry;
 286                 goto bug;
 287         }
 288
 289         real = ovl_dentry_upper(dentry);
 290         if (real && (!inode || inode == d_inode(real)))
 291                 return real;
 292
 293         real = ovl_dentry_lower(dentry);
 294         if (!real)
 295                 goto bug;
 296
 297         if (!inode || inode == d_inode(real))
 298                 return real;
 299
 300         /* Handle recursion */
 301         if (real->d_flags & DCACHE_OP_REAL)
 302                 return real->d_op->d_real(real, inode);
 303
 304 bug:
 305         WARN(1, "ovl_d_real(%pd4, %s:%lu\n): real dentry not found\n", dentry,
 306              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 307         return dentry;
 308 }
 309
 310 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 311 {
 312         struct ovl_entry *oe = dentry->d_fsdata;
 313         unsigned int i;
 314         int ret = 1;
 315
 316         for (i = 0; i < oe->numlower; i++) {
 317                 struct dentry *d = oe->lowerstack[i].dentry;
 318
 319                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 320                         ret = d->d_op->d_revalidate(d, flags);
 321                         if (ret < 0)
 322                                 return ret;
 323                         if (!ret) {
 324                                 if (!(flags & LOOKUP_RCU))
 325                                         d_invalidate(d);
 326                                 return -ESTALE;
 327                         }
 328                 }
 329         }
 330         return 1;
 331 }
 332
 333 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 334 {
 335         struct ovl_entry *oe = dentry->d_fsdata;
 336         unsigned int i;
 337         int ret = 1;
 338
 339         for (i = 0; i < oe->numlower; i++) {
 340                 struct dentry *d = oe->lowerstack[i].dentry;
 341
 342                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 343                         ret = d->d_op->d_weak_revalidate(d, flags);
 344                         if (ret <= 0)
 345                                 break;
 346                 }
 347         }
 348         return ret;
 349 }
 350
 351 static const struct dentry_operations ovl_dentry_operations = {
 352         .d_release = ovl_dentry_release,
 353         .d_select_inode = ovl_d_select_inode,
 354         .d_real = ovl_d_real,
 355 };
 356
 357 static const struct dentry_operations ovl_reval_dentry_operations = {
 358         .d_release = ovl_dentry_release,
 359         .d_select_inode = ovl_d_select_inode,
 360         .d_real = ovl_d_real,
 361         .d_revalidate = ovl_dentry_revalidate,
 362         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 363 };
 364
 365 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 366 {
 367         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 368         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 369
 370         if (oe)
 371                 oe->numlower = numlower;
 372
 373         return oe;
 374 }
 375
 376 static bool ovl_dentry_remote(struct dentry *dentry)
 377 {
 378         return dentry->d_flags &
 379                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 380 }
 381
 382 static bool ovl_dentry_weird(struct dentry *dentry)
 383 {
 384         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 385                                   DCACHE_MANAGE_TRANSIT |
 386                                   DCACHE_OP_HASH |
 387                                   DCACHE_OP_COMPARE);
 388 }
 389
 390 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
 391                                              struct qstr *name)
 392 {
 393         struct dentry *dentry;
 394
 395         mutex_lock(&dir->d_inode->i_mutex);
 396         dentry = lookup_one_len(name->name, dir, name->len);
 397         mutex_unlock(&dir->d_inode->i_mutex);
 398
 399         if (IS_ERR(dentry)) {
 400                 if (PTR_ERR(dentry) == -ENOENT)
 401                         dentry = NULL;
 402         } else if (!dentry->d_inode) {
 403                 dput(dentry);
 404                 dentry = NULL;
 405         } else if (ovl_dentry_weird(dentry)) {
 406                 dput(dentry);
 407                 /* Don't support traversing automounts and other weirdness */
 408                 dentry = ERR_PTR(-EREMOTE);
 409         }
 410         return dentry;
 411 }
 412
 413 /*
 414  * Returns next layer in stack starting from top.
 415  * Returns -1 if this is the last layer.
 416  */
 417 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 418 {
 419         struct ovl_entry *oe = dentry->d_fsdata;
 420
 421         BUG_ON(idx < 0);
 422         if (idx == 0) {
 423                 ovl_path_upper(dentry, path);
 424                 if (path->dentry)
 425                         return oe->numlower ? 1 : -1;
 426                 idx++;
 427         }
 428         BUG_ON(idx > oe->numlower);
 429         *path = oe->lowerstack[idx - 1];
 430
 431         return (idx < oe->numlower) ? idx + 1 : -1;
 432 }
 433
 434 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 435                           unsigned int flags)
 436 {
 437         struct ovl_entry *oe;
 438         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 439         struct path *stack = NULL;
 440         struct dentry *upperdir, *upperdentry = NULL;
 441         unsigned int ctr = 0;
 442         struct inode *inode = NULL;
 443         bool upperopaque = false;
 444         struct dentry *this, *prev = NULL;
 445         unsigned int i;
 446         int err;
 447
 448         upperdir = ovl_upperdentry_dereference(poe);
 449         if (upperdir) {
 450                 this = ovl_lookup_real(upperdir, &dentry->d_name);
 451                 err = PTR_ERR(this);
 452                 if (IS_ERR(this))
 453                         goto out;
 454
 455                 if (this) {
 456                         if (unlikely(ovl_dentry_remote(this))) {
 457                                 dput(this);
 458                                 err = -EREMOTE;
 459                                 goto out;
 460                         }
 461                         if (ovl_is_whiteout(this)) {
 462                                 dput(this);
 463                                 this = NULL;
 464                                 upperopaque = true;
 465                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 466                                 upperopaque = true;
 467                         }
 468                 }
 469                 upperdentry = prev = this;
 470         }
 471
 472         if (!upperopaque && poe->numlower) {
 473                 err = -ENOMEM;
 474                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 475                 if (!stack)
 476                         goto out_put_upper;
 477         }
 478
 479         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 480                 bool opaque = false;
 481                 struct path lowerpath = poe->lowerstack[i];
 482
 483                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
 484                 err = PTR_ERR(this);
 485                 if (IS_ERR(this)) {
 486                         /*
 487                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 488                          */
 489                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 490                                 continue;
 491                         goto out_put;
 492                 }
 493                 if (!this)
 494                         continue;
 495                 if (ovl_is_whiteout(this)) {
 496                         dput(this);
 497                         break;
 498                 }
 499                 /*
 500                  * Only makes sense to check opaque dir if this is not the
 501                  * lowermost layer.
 502                  */
 503                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 504                         opaque = true;
 505
 506                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 507                              !S_ISDIR(this->d_inode->i_mode))) {
 508                         /*
 509                          * FIXME: check for upper-opaqueness maybe better done
 510                          * in remove code.
 511                          */
 512                         if (prev == upperdentry)
 513                                 upperopaque = true;
 514                         dput(this);
 515                         break;
 516                 }
 517                 /*
 518                  * If this is a non-directory then stop here.
 519                  */
 520                 if (!S_ISDIR(this->d_inode->i_mode))
 521                         opaque = true;
 522
 523                 stack[ctr].dentry = this;
 524                 stack[ctr].mnt = lowerpath.mnt;
 525                 ctr++;
 526                 prev = this;
 527                 if (opaque)
 528                         break;
 529         }
 530
 531         oe = ovl_alloc_entry(ctr);
 532         err = -ENOMEM;
 533         if (!oe)
 534                 goto out_put;
 535
 536         if (upperdentry || ctr) {
 537                 struct dentry *realdentry;
 538
 539                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 540
 541                 err = -ENOMEM;
 542                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
 543                                       oe);
 544                 if (!inode)
 545                         goto out_free_oe;
 546                 ovl_copyattr(realdentry->d_inode, inode);
 547         }
 548
 549         oe->opaque = upperopaque;
 550         oe->__upperdentry = upperdentry;
 551         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 552         kfree(stack);
 553         dentry->d_fsdata = oe;
 554         d_add(dentry, inode);
 555
 556         return NULL;
 557
 558 out_free_oe:
 559         kfree(oe);
 560 out_put:
 561         for (i = 0; i < ctr; i++)
 562                 dput(stack[i].dentry);
 563         kfree(stack);
 564 out_put_upper:
 565         dput(upperdentry);
 566 out:
 567         return ERR_PTR(err);
 568 }
 569
 570 struct file *ovl_path_open(struct path *path, int flags)
 571 {
 572         return dentry_open(path, flags, current_cred());
 573 }
 574
 575 static void ovl_put_super(struct super_block *sb)
 576 {
 577         struct ovl_fs *ufs = sb->s_fs_info;
 578         unsigned i;
 579
 580         dput(ufs->workdir);
 581         mntput(ufs->upper_mnt);
 582         for (i = 0; i < ufs->numlower; i++)
 583                 mntput(ufs->lower_mnt[i]);
 584         kfree(ufs->lower_mnt);
 585
 586         kfree(ufs->config.lowerdir);
 587         kfree(ufs->config.upperdir);
 588         kfree(ufs->config.workdir);
 589         kfree(ufs);
 590 }
 591
 592 /**
 593  * ovl_statfs
 594  * @sb: The overlayfs super block
 595  * @buf: The struct kstatfs to fill in with stats
 596  *
 597  * Get the filesystem statistics.  As writes always target the upper layer
 598  * filesystem pass the statfs to the upper filesystem (if it exists)
 599  */
 600 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 601 {
 602         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 603         struct dentry *root_dentry = dentry->d_sb->s_root;
 604         struct path path;
 605         int err;
 606
 607         ovl_path_real(root_dentry, &path);
 608
 609         err = vfs_statfs(&path, buf);
 610         if (!err) {
 611                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 612                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 613         }
 614
 615         return err;
 616 }
 617
 618 /**
 619  * ovl_show_options
 620  *
 621  * Prints the mount options for a given superblock.
 622  * Returns zero; does not fail.
 623  */
 624 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 625 {
 626         struct super_block *sb = dentry->d_sb;
 627         struct ovl_fs *ufs = sb->s_fs_info;
 628
 629         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 630         if (ufs->config.upperdir) {
 631                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 632                 seq_show_option(m, "workdir", ufs->config.workdir);
 633         }
 634         return 0;
 635 }
 636
 637 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 638 {
 639         struct ovl_fs *ufs = sb->s_fs_info;
 640
 641         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 642                 return -EROFS;
 643
 644         return 0;
 645 }
 646
 647 static const struct super_operations ovl_super_operations = {
 648         .put_super      = ovl_put_super,
 649         .statfs         = ovl_statfs,
 650         .show_options   = ovl_show_options,
 651         .remount_fs     = ovl_remount,
 652 };
 653
 654 enum {
 655         OPT_LOWERDIR,
 656         OPT_UPPERDIR,
 657         OPT_WORKDIR,
 658         OPT_ERR,
 659 };
 660
 661 static const match_table_t ovl_tokens = {
 662         {OPT_LOWERDIR,                  "lowerdir=%s"},
 663         {OPT_UPPERDIR,                  "upperdir=%s"},
 664         {OPT_WORKDIR,                   "workdir=%s"},
 665         {OPT_ERR,                       NULL}
 666 };
 667
 668 static char *ovl_next_opt(char **s)
 669 {
 670         char *sbegin = *s;
 671         char *p;
 672
 673         if (sbegin == NULL)
 674                 return NULL;
 675
 676         for (p = sbegin; *p; p++) {
 677                 if (*p == '\\') {
 678                         p++;
 679                         if (!*p)
 680                                 break;
 681                 } else if (*p == ',') {
 682                         *p = '\0';
 683                         *s = p + 1;
 684                         return sbegin;
 685                 }
 686         }
 687         *s = NULL;
 688         return sbegin;
 689 }
 690
 691 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 692 {
 693         char *p;
 694
 695         while ((p = ovl_next_opt(&opt)) != NULL) {
 696                 int token;
 697                 substring_t args[MAX_OPT_ARGS];
 698
 699                 if (!*p)
 700                         continue;
 701
 702                 token = match_token(p, ovl_tokens, args);
 703                 switch (token) {
 704                 case OPT_UPPERDIR:
 705                         kfree(config->upperdir);
 706                         config->upperdir = match_strdup(&args[0]);
 707                         if (!config->upperdir)
 708                                 return -ENOMEM;
 709                         break;
 710
 711                 case OPT_LOWERDIR:
 712                         kfree(config->lowerdir);
 713                         config->lowerdir = match_strdup(&args[0]);
 714                         if (!config->lowerdir)
 715                                 return -ENOMEM;
 716                         break;
 717
 718                 case OPT_WORKDIR:
 719                         kfree(config->workdir);
 720                         config->workdir = match_strdup(&args[0]);
 721                         if (!config->workdir)
 722                                 return -ENOMEM;
 723                         break;
 724
 725                 default:
 726                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 727                         return -EINVAL;
 728                 }
 729         }
 730
 731         /* Workdir is useless in non-upper mount */
 732         if (!config->upperdir && config->workdir) {
 733                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 734                         config->workdir);
 735                 kfree(config->workdir);
 736                 config->workdir = NULL;
 737         }
 738
 739         return 0;
 740 }
 741
 742 #define OVL_WORKDIR_NAME "work"
 743
 744 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 745                                          struct dentry *dentry)
 746 {
 747         struct inode *dir = dentry->d_inode;
 748         struct dentry *work;
 749         int err;
 750         bool retried = false;
 751
 752         err = mnt_want_write(mnt);
 753         if (err)
 754                 return ERR_PTR(err);
 755
 756         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 757 retry:
 758         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 759                               strlen(OVL_WORKDIR_NAME));
 760
 761         if (!IS_ERR(work)) {
 762                 struct kstat stat = {
 763                         .mode = S_IFDIR | 0,
 764                 };
 765
 766                 if (work->d_inode) {
 767                         err = -EEXIST;
 768                         if (retried)
 769                                 goto out_dput;
 770
 771                         retried = true;
 772                         ovl_cleanup(dir, work);
 773                         dput(work);
 774                         goto retry;
 775                 }
 776
 777                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 778                 if (err)
 779                         goto out_dput;
 780         }
 781 out_unlock:
 782         mutex_unlock(&dir->i_mutex);
 783         mnt_drop_write(mnt);
 784
 785         return work;
 786
 787 out_dput:
 788         dput(work);
 789         work = ERR_PTR(err);
 790         goto out_unlock;
 791 }
 792
 793 static void ovl_unescape(char *s)
 794 {
 795         char *d = s;
 796
 797         for (;; s++, d++) {
 798                 if (*s == '\\')
 799                         s++;
 800                 *d = *s;
 801                 if (!*s)
 802                         break;
 803         }
 804 }
 805
 806 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 807 {
 808         int err = -EINVAL;
 809
 810         if (!*name) {
 811                 pr_err("overlayfs: empty lowerdir\n");
 812                 goto out;
 813         }
 814         err = kern_path(name, LOOKUP_FOLLOW, path);
 815         if (err) {
 816                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 817                 goto out;
 818         }
 819         err = -EINVAL;
 820         if (ovl_dentry_weird(path->dentry)) {
 821                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 822                 goto out_put;
 823         }
 824         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 825                 pr_err("overlayfs: '%s' not a directory\n", name);
 826                 goto out_put;
 827         }
 828         return 0;
 829
 830 out_put:
 831         path_put(path);
 832 out:
 833         return err;
 834 }
 835
 836 static int ovl_mount_dir(const char *name, struct path *path)
 837 {
 838         int err = -ENOMEM;
 839         char *tmp = kstrdup(name, GFP_KERNEL);
 840
 841         if (tmp) {
 842                 ovl_unescape(tmp);
 843                 err = ovl_mount_dir_noesc(tmp, path);
 844
 845                 if (!err)
 846                         if (ovl_dentry_remote(path->dentry)) {
 847                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 848                                        tmp);
 849                                 path_put(path);
 850                                 err = -EINVAL;
 851                         }
 852                 kfree(tmp);
 853         }
 854         return err;
 855 }
 856
 857 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 858                          int *stack_depth, bool *remote)
 859 {
 860         int err;
 861         struct kstatfs statfs;
 862
 863         err = ovl_mount_dir_noesc(name, path);
 864         if (err)
 865                 goto out;
 866
 867         err = vfs_statfs(path, &statfs);
 868         if (err) {
 869                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 870                 goto out_put;
 871         }
 872         *namelen = max(*namelen, statfs.f_namelen);
 873         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 874
 875         if (ovl_dentry_remote(path->dentry))
 876                 *remote = true;
 877
 878         return 0;
 879
 880 out_put:
 881         path_put(path);
 882 out:
 883         return err;
 884 }
 885
 886 /* Workdir should not be subdir of upperdir and vice versa */
 887 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 888 {
 889         bool ok = false;
 890
 891         if (workdir != upperdir) {
 892                 ok = (lock_rename(workdir, upperdir) == NULL);
 893                 unlock_rename(workdir, upperdir);
 894         }
 895         return ok;
 896 }
 897
 898 static unsigned int ovl_split_lowerdirs(char *str)
 899 {
 900         unsigned int ctr = 1;
 901         char *s, *d;
 902
 903         for (s = d = str;; s++, d++) {
 904                 if (*s == '\\') {
 905                         s++;
 906                 } else if (*s == ':') {
 907                         *d = '\0';
 908                         ctr++;
 909                         continue;
 910                 }
 911                 *d = *s;
 912                 if (!*s)
 913                         break;
 914         }
 915         return ctr;
 916 }
 917
 918 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
 919 {
 920         struct path upperpath = { NULL, NULL };
 921         struct path workpath = { NULL, NULL };
 922         struct dentry *root_dentry;
 923         struct ovl_entry *oe;
 924         struct ovl_fs *ufs;
 925         struct path *stack = NULL;
 926         char *lowertmp;
 927         char *lower;
 928         unsigned int numlower;
 929         unsigned int stacklen = 0;
 930         unsigned int i;
 931         bool remote = false;
 932         int err;
 933
 934         err = -ENOMEM;
 935         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
 936         if (!ufs)
 937                 goto out;
 938
 939         err = ovl_parse_opt((char *) data, &ufs->config);
 940         if (err)
 941                 goto out_free_config;
 942
 943         err = -EINVAL;
 944         if (!ufs->config.lowerdir) {
 945                 pr_err("overlayfs: missing 'lowerdir'\n");
 946                 goto out_free_config;
 947         }
 948
 949         sb->s_stack_depth = 0;
 950         sb->s_maxbytes = MAX_LFS_FILESIZE;
 951         if (ufs->config.upperdir) {
 952                 if (!ufs->config.workdir) {
 953                         pr_err("overlayfs: missing 'workdir'\n");
 954                         goto out_free_config;
 955                 }
 956
 957                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
 958                 if (err)
 959                         goto out_free_config;
 960
 961                 /* Upper fs should not be r/o */
 962                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
 963                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
 964                         err = -EINVAL;
 965                         goto out_put_upperpath;
 966                 }
 967
 968                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
 969                 if (err)
 970                         goto out_put_upperpath;
 971
 972                 err = -EINVAL;
 973                 if (upperpath.mnt != workpath.mnt) {
 974                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
 975                         goto out_put_workpath;
 976                 }
 977                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
 978                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
 979                         goto out_put_workpath;
 980                 }
 981                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
 982         }
 983         err = -ENOMEM;
 984         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
 985         if (!lowertmp)
 986                 goto out_put_workpath;
 987
 988         err = -EINVAL;
 989         stacklen = ovl_split_lowerdirs(lowertmp);
 990         if (stacklen > OVL_MAX_STACK) {
 991                 pr_err("overlayfs: too many lower directries, limit is %d\n",
 992                        OVL_MAX_STACK);
 993                 goto out_free_lowertmp;
 994         } else if (!ufs->config.upperdir && stacklen == 1) {
 995                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
 996                 goto out_free_lowertmp;
 997         }
 998
 999         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1000         if (!stack)
1001                 goto out_free_lowertmp;
1002
1003         lower = lowertmp;
1004         for (numlower = 0; numlower < stacklen; numlower++) {
1005                 err = ovl_lower_dir(lower, &stack[numlower],
1006                                     &ufs->lower_namelen, &sb->s_stack_depth,
1007                                     &remote);
1008                 if (err)
1009                         goto out_put_lowerpath;
1010
1011                 lower = strchr(lower, '\0') + 1;
1012         }
1013
1014         err = -EINVAL;
1015         sb->s_stack_depth++;
1016         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1017                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1018                 goto out_put_lowerpath;
1019         }
1020
1021         if (ufs->config.upperdir) {
1022                 ufs->upper_mnt = clone_private_mount(&upperpath);
1023                 err = PTR_ERR(ufs->upper_mnt);
1024                 if (IS_ERR(ufs->upper_mnt)) {
1025                         pr_err("overlayfs: failed to clone upperpath\n");
1026                         goto out_put_lowerpath;
1027                 }
1028
1029                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1030                 err = PTR_ERR(ufs->workdir);
1031                 if (IS_ERR(ufs->workdir)) {
1032                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1033                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1034                         sb->s_flags |= MS_RDONLY;
1035                         ufs->workdir = NULL;
1036                 }
1037         }
1038
1039         err = -ENOMEM;
1040         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1041         if (ufs->lower_mnt == NULL)
1042                 goto out_put_workdir;
1043         for (i = 0; i < numlower; i++) {
1044                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1045
1046                 err = PTR_ERR(mnt);
1047                 if (IS_ERR(mnt)) {
1048                         pr_err("overlayfs: failed to clone lowerpath\n");
1049                         goto out_put_lower_mnt;
1050                 }
1051                 /*
1052                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1053                  * will fail instead of modifying lower fs.
1054                  */
1055                 mnt->mnt_flags |= MNT_READONLY;
1056
1057                 ufs->lower_mnt[ufs->numlower] = mnt;
1058                 ufs->numlower++;
1059         }
1060
1061         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1062         if (!ufs->upper_mnt)
1063                 sb->s_flags |= MS_RDONLY;
1064
1065         if (remote)
1066                 sb->s_d_op = &ovl_reval_dentry_operations;
1067         else
1068                 sb->s_d_op = &ovl_dentry_operations;
1069
1070         err = -ENOMEM;
1071         oe = ovl_alloc_entry(numlower);
1072         if (!oe)
1073                 goto out_put_lower_mnt;
1074
1075         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1076         if (!root_dentry)
1077                 goto out_free_oe;
1078
1079         mntput(upperpath.mnt);
1080         for (i = 0; i < numlower; i++)
1081                 mntput(stack[i].mnt);
1082         path_put(&workpath);
1083         kfree(lowertmp);
1084
1085         oe->__upperdentry = upperpath.dentry;
1086         for (i = 0; i < numlower; i++) {
1087                 oe->lowerstack[i].dentry = stack[i].dentry;
1088                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1089         }
1090         kfree(stack);
1091
1092         root_dentry->d_fsdata = oe;
1093
1094         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1095                      root_dentry->d_inode);
1096
1097         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1098         sb->s_op = &ovl_super_operations;
1099         sb->s_root = root_dentry;
1100         sb->s_fs_info = ufs;
1101
1102         return 0;
1103
1104 out_free_oe:
1105         kfree(oe);
1106 out_put_lower_mnt:
1107         for (i = 0; i < ufs->numlower; i++)
1108                 mntput(ufs->lower_mnt[i]);
1109         kfree(ufs->lower_mnt);
1110 out_put_workdir:
1111         dput(ufs->workdir);
1112         mntput(ufs->upper_mnt);
1113 out_put_lowerpath:
1114         for (i = 0; i < numlower; i++)
1115                 path_put(&stack[i]);
1116         kfree(stack);
1117 out_free_lowertmp:
1118         kfree(lowertmp);
1119 out_put_workpath:
1120         path_put(&workpath);
1121 out_put_upperpath:
1122         path_put(&upperpath);
1123 out_free_config:
1124         kfree(ufs->config.lowerdir);
1125         kfree(ufs->config.upperdir);
1126         kfree(ufs->config.workdir);
1127         kfree(ufs);
1128 out:
1129         return err;
1130 }
1131
1132 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1133                                 const char *dev_name, void *raw_data)
1134 {
1135         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1136 }
1137
1138 static struct file_system_type ovl_fs_type = {
1139         .owner          = THIS_MODULE,
1140         .name           = "overlay",
1141         .mount          = ovl_mount,
1142         .kill_sb        = kill_anon_super,
1143 };
1144 MODULE_ALIAS_FS("overlay");
1145
1146 static int __init ovl_init(void)
1147 {
1148         return register_filesystem(&ovl_fs_type);
1149 }
1150
1151 static void __exit ovl_exit(void)
1152 {
1153         unregister_filesystem(&ovl_fs_type);
1154 }
1155
1156 module_init(ovl_init);
1157 module_exit(ovl_exit);