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 <linux/posix_acl_xattr.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         /* creds of process who forced instantiation of super block */
  46         const struct cred *creator_cred;
  47 };
  48
  49 struct ovl_dir_cache;
  50
  51 /* private information held for every overlayfs dentry */
  52 struct ovl_entry {
  53         struct dentry *__upperdentry;
  54         struct ovl_dir_cache *cache;
  55         union {
  56                 struct {
  57                         u64 version;
  58                         bool opaque;
  59                 };
  60                 struct rcu_head rcu;
  61         };
  62         unsigned numlower;
  63         struct path lowerstack[];
  64 };
  65
  66 #define OVL_MAX_STACK 500
  67
  68 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
  69 {
  70         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
  71 }
  72
  73 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  74 {
  75         struct ovl_entry *oe = dentry->d_fsdata;
  76         enum ovl_path_type type = 0;
  77
  78         if (oe->__upperdentry) {
  79                 type = __OVL_PATH_UPPER;
  80
  81                 /*
  82                  * Non-dir dentry can hold lower dentry from previous
  83                  * location. Its purity depends only on opaque flag.
  84                  */
  85                 if (oe->numlower && S_ISDIR(dentry->d_inode->i_mode))
  86                         type |= __OVL_PATH_MERGE;
  87                 else if (!oe->opaque)
  88                         type |= __OVL_PATH_PURE;
  89         } else {
  90                 if (oe->numlower > 1)
  91                         type |= __OVL_PATH_MERGE;
  92         }
  93         return type;
  94 }
  95
  96 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
  97 {
  98         return lockless_dereference(oe->__upperdentry);
  99 }
 100
 101 void ovl_path_upper(struct dentry *dentry, struct path *path)
 102 {
 103         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 104         struct ovl_entry *oe = dentry->d_fsdata;
 105
 106         path->mnt = ofs->upper_mnt;
 107         path->dentry = ovl_upperdentry_dereference(oe);
 108 }
 109
 110 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 111 {
 112         enum ovl_path_type type = ovl_path_type(dentry);
 113
 114         if (!OVL_TYPE_UPPER(type))
 115                 ovl_path_lower(dentry, path);
 116         else
 117                 ovl_path_upper(dentry, path);
 118
 119         return type;
 120 }
 121
 122 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 123 {
 124         struct ovl_entry *oe = dentry->d_fsdata;
 125
 126         return ovl_upperdentry_dereference(oe);
 127 }
 128
 129 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 130 {
 131         struct ovl_entry *oe = dentry->d_fsdata;
 132
 133         return __ovl_dentry_lower(oe);
 134 }
 135
 136 struct dentry *ovl_dentry_real(struct dentry *dentry)
 137 {
 138         struct ovl_entry *oe = dentry->d_fsdata;
 139         struct dentry *realdentry;
 140
 141         realdentry = ovl_upperdentry_dereference(oe);
 142         if (!realdentry)
 143                 realdentry = __ovl_dentry_lower(oe);
 144
 145         return realdentry;
 146 }
 147
 148 static void ovl_inode_init(struct inode *inode, struct inode *realinode,
 149                            bool is_upper)
 150 {
 151         WRITE_ONCE(inode->i_private, (unsigned long) realinode |
 152                    (is_upper ? OVL_ISUPPER_MASK : 0));
 153 }
 154
 155 struct vfsmount *ovl_entry_mnt_real(struct ovl_entry *oe, struct inode *inode,
 156                                     bool is_upper)
 157 {
 158         if (is_upper) {
 159                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 160
 161                 return ofs->upper_mnt;
 162         } else {
 163                 return oe->numlower ? oe->lowerstack[0].mnt : NULL;
 164         }
 165 }
 166
 167 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
 168 {
 169         struct ovl_entry *oe = dentry->d_fsdata;
 170
 171         return oe->cache;
 172 }
 173
 174 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
 175 {
 176         struct ovl_entry *oe = dentry->d_fsdata;
 177
 178         oe->cache = cache;
 179 }
 180
 181 void ovl_path_lower(struct dentry *dentry, struct path *path)
 182 {
 183         struct ovl_entry *oe = dentry->d_fsdata;
 184
 185         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
 186 }
 187
 188 int ovl_want_write(struct dentry *dentry)
 189 {
 190         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 191         return mnt_want_write(ofs->upper_mnt);
 192 }
 193
 194 void ovl_drop_write(struct dentry *dentry)
 195 {
 196         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 197         mnt_drop_write(ofs->upper_mnt);
 198 }
 199
 200 struct dentry *ovl_workdir(struct dentry *dentry)
 201 {
 202         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 203         return ofs->workdir;
 204 }
 205
 206 bool ovl_dentry_is_opaque(struct dentry *dentry)
 207 {
 208         struct ovl_entry *oe = dentry->d_fsdata;
 209         return oe->opaque;
 210 }
 211
 212 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
 213 {
 214         struct ovl_entry *oe = dentry->d_fsdata;
 215         oe->opaque = opaque;
 216 }
 217
 218 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
 219 {
 220         struct ovl_entry *oe = dentry->d_fsdata;
 221
 222         WARN_ON(!inode_is_locked(upperdentry->d_parent->d_inode));
 223         WARN_ON(oe->__upperdentry);
 224         /*
 225          * Make sure upperdentry is consistent before making it visible to
 226          * ovl_upperdentry_dereference().
 227          */
 228         smp_wmb();
 229         oe->__upperdentry = upperdentry;
 230 }
 231
 232 void ovl_inode_update(struct inode *inode, struct inode *upperinode)
 233 {
 234         WARN_ON(!upperinode);
 235         WARN_ON(!inode_unhashed(inode));
 236         WRITE_ONCE(inode->i_private,
 237                    (unsigned long) upperinode | OVL_ISUPPER_MASK);
 238         if (!S_ISDIR(upperinode->i_mode))
 239                 __insert_inode_hash(inode, (unsigned long) upperinode);
 240 }
 241
 242 void ovl_dentry_version_inc(struct dentry *dentry)
 243 {
 244         struct ovl_entry *oe = dentry->d_fsdata;
 245
 246         WARN_ON(!inode_is_locked(dentry->d_inode));
 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(!inode_is_locked(dentry->d_inode));
 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 const struct cred *ovl_override_creds(struct super_block *sb)
 266 {
 267         struct ovl_fs *ofs = sb->s_fs_info;
 268
 269         return override_creds(ofs->creator_cred);
 270 }
 271
 272 static bool ovl_is_opaquedir(struct dentry *dentry)
 273 {
 274         int res;
 275         char val;
 276         struct inode *inode = dentry->d_inode;
 277
 278         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
 279                 return false;
 280
 281         res = inode->i_op->getxattr(dentry, inode, OVL_XATTR_OPAQUE, &val, 1);
 282         if (res == 1 && val == 'y')
 283                 return true;
 284
 285         return false;
 286 }
 287
 288 static void ovl_dentry_release(struct dentry *dentry)
 289 {
 290         struct ovl_entry *oe = dentry->d_fsdata;
 291
 292         if (oe) {
 293                 unsigned int i;
 294
 295                 dput(oe->__upperdentry);
 296                 for (i = 0; i < oe->numlower; i++)
 297                         dput(oe->lowerstack[i].dentry);
 298                 kfree_rcu(oe, rcu);
 299         }
 300 }
 301
 302 static struct dentry *ovl_d_real(struct dentry *dentry,
 303                                  const struct inode *inode,
 304                                  unsigned int open_flags)
 305 {
 306         struct dentry *real;
 307
 308         if (d_is_dir(dentry)) {
 309                 if (!inode || inode == d_inode(dentry))
 310                         return dentry;
 311                 goto bug;
 312         }
 313
 314         if (d_is_negative(dentry))
 315                 return dentry;
 316
 317         if (open_flags) {
 318                 int err = ovl_open_maybe_copy_up(dentry, open_flags);
 319
 320                 if (err)
 321                         return ERR_PTR(err);
 322         }
 323
 324         real = ovl_dentry_upper(dentry);
 325         if (real && (!inode || inode == d_inode(real)))
 326                 return real;
 327
 328         real = ovl_dentry_lower(dentry);
 329         if (!real)
 330                 goto bug;
 331
 332         if (!inode || inode == d_inode(real))
 333                 return real;
 334
 335         /* Handle recursion */
 336         return d_real(real, inode, open_flags);
 337 bug:
 338         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
 339              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
 340         return dentry;
 341 }
 342
 343 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
 344 {
 345         struct ovl_entry *oe = dentry->d_fsdata;
 346         unsigned int i;
 347         int ret = 1;
 348
 349         for (i = 0; i < oe->numlower; i++) {
 350                 struct dentry *d = oe->lowerstack[i].dentry;
 351
 352                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
 353                         ret = d->d_op->d_revalidate(d, flags);
 354                         if (ret < 0)
 355                                 return ret;
 356                         if (!ret) {
 357                                 if (!(flags & LOOKUP_RCU))
 358                                         d_invalidate(d);
 359                                 return -ESTALE;
 360                         }
 361                 }
 362         }
 363         return 1;
 364 }
 365
 366 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
 367 {
 368         struct ovl_entry *oe = dentry->d_fsdata;
 369         unsigned int i;
 370         int ret = 1;
 371
 372         for (i = 0; i < oe->numlower; i++) {
 373                 struct dentry *d = oe->lowerstack[i].dentry;
 374
 375                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
 376                         ret = d->d_op->d_weak_revalidate(d, flags);
 377                         if (ret <= 0)
 378                                 break;
 379                 }
 380         }
 381         return ret;
 382 }
 383
 384 static const struct dentry_operations ovl_dentry_operations = {
 385         .d_release = ovl_dentry_release,
 386         .d_real = ovl_d_real,
 387 };
 388
 389 static const struct dentry_operations ovl_reval_dentry_operations = {
 390         .d_release = ovl_dentry_release,
 391         .d_real = ovl_d_real,
 392         .d_revalidate = ovl_dentry_revalidate,
 393         .d_weak_revalidate = ovl_dentry_weak_revalidate,
 394 };
 395
 396 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
 397 {
 398         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 399         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 400
 401         if (oe)
 402                 oe->numlower = numlower;
 403
 404         return oe;
 405 }
 406
 407 static bool ovl_dentry_remote(struct dentry *dentry)
 408 {
 409         return dentry->d_flags &
 410                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE |
 411                  DCACHE_OP_REAL);
 412 }
 413
 414 static bool ovl_dentry_weird(struct dentry *dentry)
 415 {
 416         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 417                                   DCACHE_MANAGE_TRANSIT |
 418                                   DCACHE_OP_HASH |
 419                                   DCACHE_OP_COMPARE);
 420 }
 421
 422 static inline struct dentry *ovl_lookup_real(struct super_block *ovl_sb,
 423                                              struct dentry *dir,
 424                                              const struct qstr *name)
 425 {
 426         const struct cred *old_cred;
 427         struct dentry *dentry;
 428
 429         old_cred = ovl_override_creds(ovl_sb);
 430         dentry = lookup_one_len_unlocked(name->name, dir, name->len);
 431         revert_creds(old_cred);
 432
 433         if (IS_ERR(dentry)) {
 434                 if (PTR_ERR(dentry) == -ENOENT)
 435                         dentry = NULL;
 436         } else if (!dentry->d_inode) {
 437                 dput(dentry);
 438                 dentry = NULL;
 439         } else if (ovl_dentry_weird(dentry)) {
 440                 dput(dentry);
 441                 /* Don't support traversing automounts and other weirdness */
 442                 dentry = ERR_PTR(-EREMOTE);
 443         }
 444         return dentry;
 445 }
 446
 447 /*
 448  * Returns next layer in stack starting from top.
 449  * Returns -1 if this is the last layer.
 450  */
 451 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 452 {
 453         struct ovl_entry *oe = dentry->d_fsdata;
 454
 455         BUG_ON(idx < 0);
 456         if (idx == 0) {
 457                 ovl_path_upper(dentry, path);
 458                 if (path->dentry)
 459                         return oe->numlower ? 1 : -1;
 460                 idx++;
 461         }
 462         BUG_ON(idx > oe->numlower);
 463         *path = oe->lowerstack[idx - 1];
 464
 465         return (idx < oe->numlower) ? idx + 1 : -1;
 466 }
 467
 468 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 469                           unsigned int flags)
 470 {
 471         struct ovl_entry *oe;
 472         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 473         struct path *stack = NULL;
 474         struct dentry *upperdir, *upperdentry = NULL;
 475         unsigned int ctr = 0;
 476         struct inode *inode = NULL;
 477         bool upperopaque = false;
 478         struct dentry *this, *prev = NULL;
 479         unsigned int i;
 480         int err;
 481
 482         upperdir = ovl_upperdentry_dereference(poe);
 483         if (upperdir) {
 484                 this = ovl_lookup_real(dentry->d_sb, upperdir, &dentry->d_name);
 485                 err = PTR_ERR(this);
 486                 if (IS_ERR(this))
 487                         goto out;
 488
 489                 if (this) {
 490                         if (unlikely(ovl_dentry_remote(this))) {
 491                                 dput(this);
 492                                 err = -EREMOTE;
 493                                 goto out;
 494                         }
 495                         if (ovl_is_whiteout(this)) {
 496                                 dput(this);
 497                                 this = NULL;
 498                                 upperopaque = true;
 499                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
 500                                 upperopaque = true;
 501                         }
 502                 }
 503                 upperdentry = prev = this;
 504         }
 505
 506         if (!upperopaque && poe->numlower) {
 507                 err = -ENOMEM;
 508                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
 509                 if (!stack)
 510                         goto out_put_upper;
 511         }
 512
 513         for (i = 0; !upperopaque && i < poe->numlower; i++) {
 514                 bool opaque = false;
 515                 struct path lowerpath = poe->lowerstack[i];
 516
 517                 this = ovl_lookup_real(dentry->d_sb,
 518                                        lowerpath.dentry, &dentry->d_name);
 519                 err = PTR_ERR(this);
 520                 if (IS_ERR(this)) {
 521                         /*
 522                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
 523                          */
 524                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
 525                                 continue;
 526                         goto out_put;
 527                 }
 528                 if (!this)
 529                         continue;
 530                 if (ovl_is_whiteout(this)) {
 531                         dput(this);
 532                         break;
 533                 }
 534                 /*
 535                  * Only makes sense to check opaque dir if this is not the
 536                  * lowermost layer.
 537                  */
 538                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
 539                         opaque = true;
 540
 541                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
 542                              !S_ISDIR(this->d_inode->i_mode))) {
 543                         /*
 544                          * FIXME: check for upper-opaqueness maybe better done
 545                          * in remove code.
 546                          */
 547                         if (prev == upperdentry)
 548                                 upperopaque = true;
 549                         dput(this);
 550                         break;
 551                 }
 552                 /*
 553                  * If this is a non-directory then stop here.
 554                  */
 555                 if (!S_ISDIR(this->d_inode->i_mode))
 556                         opaque = true;
 557
 558                 stack[ctr].dentry = this;
 559                 stack[ctr].mnt = lowerpath.mnt;
 560                 ctr++;
 561                 prev = this;
 562                 if (opaque)
 563                         break;
 564         }
 565
 566         oe = ovl_alloc_entry(ctr);
 567         err = -ENOMEM;
 568         if (!oe)
 569                 goto out_put;
 570
 571         if (upperdentry || ctr) {
 572                 struct dentry *realdentry;
 573                 struct inode *realinode;
 574
 575                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
 576                 realinode = d_inode(realdentry);
 577
 578                 err = -ENOMEM;
 579                 if (upperdentry && !d_is_dir(upperdentry)) {
 580                         inode = ovl_get_inode(dentry->d_sb, realinode);
 581                 } else {
 582                         inode = ovl_new_inode(dentry->d_sb, realinode->i_mode);
 583                         if (inode)
 584                                 ovl_inode_init(inode, realinode, !!upperdentry);
 585                 }
 586                 if (!inode)
 587                         goto out_free_oe;
 588                 ovl_copyattr(realdentry->d_inode, inode);
 589         }
 590
 591         oe->opaque = upperopaque;
 592         oe->__upperdentry = upperdentry;
 593         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
 594         kfree(stack);
 595         dentry->d_fsdata = oe;
 596         d_add(dentry, inode);
 597
 598         return NULL;
 599
 600 out_free_oe:
 601         kfree(oe);
 602 out_put:
 603         for (i = 0; i < ctr; i++)
 604                 dput(stack[i].dentry);
 605         kfree(stack);
 606 out_put_upper:
 607         dput(upperdentry);
 608 out:
 609         return ERR_PTR(err);
 610 }
 611
 612 struct file *ovl_path_open(struct path *path, int flags)
 613 {
 614         return dentry_open(path, flags | O_NOATIME, current_cred());
 615 }
 616
 617 static void ovl_put_super(struct super_block *sb)
 618 {
 619         struct ovl_fs *ufs = sb->s_fs_info;
 620         unsigned i;
 621
 622         dput(ufs->workdir);
 623         mntput(ufs->upper_mnt);
 624         for (i = 0; i < ufs->numlower; i++)
 625                 mntput(ufs->lower_mnt[i]);
 626         kfree(ufs->lower_mnt);
 627
 628         kfree(ufs->config.lowerdir);
 629         kfree(ufs->config.upperdir);
 630         kfree(ufs->config.workdir);
 631         put_cred(ufs->creator_cred);
 632         kfree(ufs);
 633 }
 634
 635 /**
 636  * ovl_statfs
 637  * @sb: The overlayfs super block
 638  * @buf: The struct kstatfs to fill in with stats
 639  *
 640  * Get the filesystem statistics.  As writes always target the upper layer
 641  * filesystem pass the statfs to the upper filesystem (if it exists)
 642  */
 643 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
 644 {
 645         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 646         struct dentry *root_dentry = dentry->d_sb->s_root;
 647         struct path path;
 648         int err;
 649
 650         ovl_path_real(root_dentry, &path);
 651
 652         err = vfs_statfs(&path, buf);
 653         if (!err) {
 654                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
 655                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
 656         }
 657
 658         return err;
 659 }
 660
 661 /**
 662  * ovl_show_options
 663  *
 664  * Prints the mount options for a given superblock.
 665  * Returns zero; does not fail.
 666  */
 667 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
 668 {
 669         struct super_block *sb = dentry->d_sb;
 670         struct ovl_fs *ufs = sb->s_fs_info;
 671
 672         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
 673         if (ufs->config.upperdir) {
 674                 seq_show_option(m, "upperdir", ufs->config.upperdir);
 675                 seq_show_option(m, "workdir", ufs->config.workdir);
 676         }
 677         if (ufs->config.default_permissions)
 678                 seq_puts(m, ",default_permissions");
 679         return 0;
 680 }
 681
 682 static int ovl_remount(struct super_block *sb, int *flags, char *data)
 683 {
 684         struct ovl_fs *ufs = sb->s_fs_info;
 685
 686         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
 687                 return -EROFS;
 688
 689         return 0;
 690 }
 691
 692 static const struct super_operations ovl_super_operations = {
 693         .put_super      = ovl_put_super,
 694         .statfs         = ovl_statfs,
 695         .show_options   = ovl_show_options,
 696         .remount_fs     = ovl_remount,
 697         .drop_inode     = generic_delete_inode,
 698 };
 699
 700 enum {
 701         OPT_LOWERDIR,
 702         OPT_UPPERDIR,
 703         OPT_WORKDIR,
 704         OPT_DEFAULT_PERMISSIONS,
 705         OPT_ERR,
 706 };
 707
 708 static const match_table_t ovl_tokens = {
 709         {OPT_LOWERDIR,                  "lowerdir=%s"},
 710         {OPT_UPPERDIR,                  "upperdir=%s"},
 711         {OPT_WORKDIR,                   "workdir=%s"},
 712         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 713         {OPT_ERR,                       NULL}
 714 };
 715
 716 static char *ovl_next_opt(char **s)
 717 {
 718         char *sbegin = *s;
 719         char *p;
 720
 721         if (sbegin == NULL)
 722                 return NULL;
 723
 724         for (p = sbegin; *p; p++) {
 725                 if (*p == '\\') {
 726                         p++;
 727                         if (!*p)
 728                                 break;
 729                 } else if (*p == ',') {
 730                         *p = '\0';
 731                         *s = p + 1;
 732                         return sbegin;
 733                 }
 734         }
 735         *s = NULL;
 736         return sbegin;
 737 }
 738
 739 static int ovl_parse_opt(char *opt, struct ovl_config *config)
 740 {
 741         char *p;
 742
 743         while ((p = ovl_next_opt(&opt)) != NULL) {
 744                 int token;
 745                 substring_t args[MAX_OPT_ARGS];
 746
 747                 if (!*p)
 748                         continue;
 749
 750                 token = match_token(p, ovl_tokens, args);
 751                 switch (token) {
 752                 case OPT_UPPERDIR:
 753                         kfree(config->upperdir);
 754                         config->upperdir = match_strdup(&args[0]);
 755                         if (!config->upperdir)
 756                                 return -ENOMEM;
 757                         break;
 758
 759                 case OPT_LOWERDIR:
 760                         kfree(config->lowerdir);
 761                         config->lowerdir = match_strdup(&args[0]);
 762                         if (!config->lowerdir)
 763                                 return -ENOMEM;
 764                         break;
 765
 766                 case OPT_WORKDIR:
 767                         kfree(config->workdir);
 768                         config->workdir = match_strdup(&args[0]);
 769                         if (!config->workdir)
 770                                 return -ENOMEM;
 771                         break;
 772
 773                 case OPT_DEFAULT_PERMISSIONS:
 774                         config->default_permissions = true;
 775                         break;
 776
 777                 default:
 778                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
 779                         return -EINVAL;
 780                 }
 781         }
 782
 783         /* Workdir is useless in non-upper mount */
 784         if (!config->upperdir && config->workdir) {
 785                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
 786                         config->workdir);
 787                 kfree(config->workdir);
 788                 config->workdir = NULL;
 789         }
 790
 791         return 0;
 792 }
 793
 794 #define OVL_WORKDIR_NAME "work"
 795
 796 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
 797                                          struct dentry *dentry)
 798 {
 799         struct inode *dir = dentry->d_inode;
 800         struct dentry *work;
 801         int err;
 802         bool retried = false;
 803
 804         err = mnt_want_write(mnt);
 805         if (err)
 806                 return ERR_PTR(err);
 807
 808         inode_lock_nested(dir, I_MUTEX_PARENT);
 809 retry:
 810         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
 811                               strlen(OVL_WORKDIR_NAME));
 812
 813         if (!IS_ERR(work)) {
 814                 struct kstat stat = {
 815                         .mode = S_IFDIR | 0,
 816                 };
 817
 818                 if (work->d_inode) {
 819                         err = -EEXIST;
 820                         if (retried)
 821                                 goto out_dput;
 822
 823                         retried = true;
 824                         ovl_cleanup(dir, work);
 825                         dput(work);
 826                         goto retry;
 827                 }
 828
 829                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
 830                 if (err)
 831                         goto out_dput;
 832         }
 833 out_unlock:
 834         inode_unlock(dir);
 835         mnt_drop_write(mnt);
 836
 837         return work;
 838
 839 out_dput:
 840         dput(work);
 841         work = ERR_PTR(err);
 842         goto out_unlock;
 843 }
 844
 845 static void ovl_unescape(char *s)
 846 {
 847         char *d = s;
 848
 849         for (;; s++, d++) {
 850                 if (*s == '\\')
 851                         s++;
 852                 *d = *s;
 853                 if (!*s)
 854                         break;
 855         }
 856 }
 857
 858 static int ovl_mount_dir_noesc(const char *name, struct path *path)
 859 {
 860         int err = -EINVAL;
 861
 862         if (!*name) {
 863                 pr_err("overlayfs: empty lowerdir\n");
 864                 goto out;
 865         }
 866         err = kern_path(name, LOOKUP_FOLLOW, path);
 867         if (err) {
 868                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
 869                 goto out;
 870         }
 871         err = -EINVAL;
 872         if (ovl_dentry_weird(path->dentry)) {
 873                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
 874                 goto out_put;
 875         }
 876         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
 877                 pr_err("overlayfs: '%s' not a directory\n", name);
 878                 goto out_put;
 879         }
 880         return 0;
 881
 882 out_put:
 883         path_put(path);
 884 out:
 885         return err;
 886 }
 887
 888 static int ovl_mount_dir(const char *name, struct path *path)
 889 {
 890         int err = -ENOMEM;
 891         char *tmp = kstrdup(name, GFP_KERNEL);
 892
 893         if (tmp) {
 894                 ovl_unescape(tmp);
 895                 err = ovl_mount_dir_noesc(tmp, path);
 896
 897                 if (!err)
 898                         if (ovl_dentry_remote(path->dentry)) {
 899                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
 900                                        tmp);
 901                                 path_put(path);
 902                                 err = -EINVAL;
 903                         }
 904                 kfree(tmp);
 905         }
 906         return err;
 907 }
 908
 909 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
 910                          int *stack_depth, bool *remote)
 911 {
 912         int err;
 913         struct kstatfs statfs;
 914
 915         err = ovl_mount_dir_noesc(name, path);
 916         if (err)
 917                 goto out;
 918
 919         err = vfs_statfs(path, &statfs);
 920         if (err) {
 921                 pr_err("overlayfs: statfs failed on '%s'\n", name);
 922                 goto out_put;
 923         }
 924         *namelen = max(*namelen, statfs.f_namelen);
 925         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
 926
 927         if (ovl_dentry_remote(path->dentry))
 928                 *remote = true;
 929
 930         return 0;
 931
 932 out_put:
 933         path_put(path);
 934 out:
 935         return err;
 936 }
 937
 938 /* Workdir should not be subdir of upperdir and vice versa */
 939 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
 940 {
 941         bool ok = false;
 942
 943         if (workdir != upperdir) {
 944                 ok = (lock_rename(workdir, upperdir) == NULL);
 945                 unlock_rename(workdir, upperdir);
 946         }
 947         return ok;
 948 }
 949
 950 static unsigned int ovl_split_lowerdirs(char *str)
 951 {
 952         unsigned int ctr = 1;
 953         char *s, *d;
 954
 955         for (s = d = str;; s++, d++) {
 956                 if (*s == '\\') {
 957                         s++;
 958                 } else if (*s == ':') {
 959                         *d = '\0';
 960                         ctr++;
 961                         continue;
 962                 }
 963                 *d = *s;
 964                 if (!*s)
 965                         break;
 966         }
 967         return ctr;
 968 }
 969
 970 static int ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
 971                                    struct dentry *dentry, struct inode *inode,
 972                                    const char *name, const void *value,
 973                                    size_t size, int flags)
 974 {
 975         struct dentry *workdir = ovl_workdir(dentry);
 976         struct inode *realinode = ovl_inode_real(inode, NULL);
 977         struct posix_acl *acl = NULL;
 978         int err;
 979
 980         /* Check that everything is OK before copy-up */
 981         if (value) {
 982                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
 983                 if (IS_ERR(acl))
 984                         return PTR_ERR(acl);
 985         }
 986         err = -EOPNOTSUPP;
 987         if (!IS_POSIXACL(d_inode(workdir)))
 988                 goto out_acl_release;
 989         if (!realinode->i_op->set_acl)
 990                 goto out_acl_release;
 991         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
 992                 err = acl ? -EACCES : 0;
 993                 goto out_acl_release;
 994         }
 995         err = -EPERM;
 996         if (!inode_owner_or_capable(inode))
 997                 goto out_acl_release;
 998
 999         posix_acl_release(acl);
1000
1001         return ovl_setxattr(dentry, inode, handler->name, value, size, flags);
1002
1003 out_acl_release:
1004         posix_acl_release(acl);
1005         return err;
1006 }
1007
1008 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1009                                struct dentry *dentry, struct inode *inode,
1010                                const char *name, const void *value,
1011                                size_t size, int flags)
1012 {
1013         return ovl_setxattr(dentry, inode, name, value, size, flags);
1014 }
1015
1016 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1017                              struct dentry *dentry, struct inode *inode,
1018                              const char *name, const void *value,
1019                              size_t size, int flags)
1020 {
1021         return -EPERM;
1022 }
1023
1024 static const struct xattr_handler ovl_posix_acl_access_xattr_handler = {
1025         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1026         .flags = ACL_TYPE_ACCESS,
1027         .set = ovl_posix_acl_xattr_set,
1028 };
1029
1030 static const struct xattr_handler ovl_posix_acl_default_xattr_handler = {
1031         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1032         .flags = ACL_TYPE_DEFAULT,
1033         .set = ovl_posix_acl_xattr_set,
1034 };
1035
1036 static const struct xattr_handler ovl_own_xattr_handler = {
1037         .prefix = OVL_XATTR_PREFIX,
1038         .set = ovl_own_xattr_set,
1039 };
1040
1041 static const struct xattr_handler ovl_other_xattr_handler = {
1042         .prefix = "", /* catch all */
1043         .set = ovl_other_xattr_set,
1044 };
1045
1046 static const struct xattr_handler *ovl_xattr_handlers[] = {
1047         &ovl_posix_acl_access_xattr_handler,
1048         &ovl_posix_acl_default_xattr_handler,
1049         &ovl_own_xattr_handler,
1050         &ovl_other_xattr_handler,
1051         NULL
1052 };
1053
1054 static const struct xattr_handler *ovl_xattr_noacl_handlers[] = {
1055         &ovl_own_xattr_handler,
1056         &ovl_other_xattr_handler,
1057         NULL,
1058 };
1059
1060 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1061 {
1062         struct path upperpath = { NULL, NULL };
1063         struct path workpath = { NULL, NULL };
1064         struct dentry *root_dentry;
1065         struct inode *realinode;
1066         struct ovl_entry *oe;
1067         struct ovl_fs *ufs;
1068         struct path *stack = NULL;
1069         char *lowertmp;
1070         char *lower;
1071         unsigned int numlower;
1072         unsigned int stacklen = 0;
1073         unsigned int i;
1074         bool remote = false;
1075         int err;
1076
1077         err = -ENOMEM;
1078         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1079         if (!ufs)
1080                 goto out;
1081
1082         err = ovl_parse_opt((char *) data, &ufs->config);
1083         if (err)
1084                 goto out_free_config;
1085
1086         err = -EINVAL;
1087         if (!ufs->config.lowerdir) {
1088                 if (!silent)
1089                         pr_err("overlayfs: missing 'lowerdir'\n");
1090                 goto out_free_config;
1091         }
1092
1093         sb->s_stack_depth = 0;
1094         sb->s_maxbytes = MAX_LFS_FILESIZE;
1095         if (ufs->config.upperdir) {
1096                 if (!ufs->config.workdir) {
1097                         pr_err("overlayfs: missing 'workdir'\n");
1098                         goto out_free_config;
1099                 }
1100
1101                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
1102                 if (err)
1103                         goto out_free_config;
1104
1105                 /* Upper fs should not be r/o */
1106                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
1107                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1108                         err = -EINVAL;
1109                         goto out_put_upperpath;
1110                 }
1111
1112                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
1113                 if (err)
1114                         goto out_put_upperpath;
1115
1116                 err = -EINVAL;
1117                 if (upperpath.mnt != workpath.mnt) {
1118                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1119                         goto out_put_workpath;
1120                 }
1121                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
1122                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1123                         goto out_put_workpath;
1124                 }
1125                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1126         }
1127         err = -ENOMEM;
1128         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1129         if (!lowertmp)
1130                 goto out_put_workpath;
1131
1132         err = -EINVAL;
1133         stacklen = ovl_split_lowerdirs(lowertmp);
1134         if (stacklen > OVL_MAX_STACK) {
1135                 pr_err("overlayfs: too many lower directries, limit is %d\n",
1136                        OVL_MAX_STACK);
1137                 goto out_free_lowertmp;
1138         } else if (!ufs->config.upperdir && stacklen == 1) {
1139                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1140                 goto out_free_lowertmp;
1141         }
1142
1143         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1144         if (!stack)
1145                 goto out_free_lowertmp;
1146
1147         lower = lowertmp;
1148         for (numlower = 0; numlower < stacklen; numlower++) {
1149                 err = ovl_lower_dir(lower, &stack[numlower],
1150                                     &ufs->lower_namelen, &sb->s_stack_depth,
1151                                     &remote);
1152                 if (err)
1153                         goto out_put_lowerpath;
1154
1155                 lower = strchr(lower, '\0') + 1;
1156         }
1157
1158         err = -EINVAL;
1159         sb->s_stack_depth++;
1160         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1161                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1162                 goto out_put_lowerpath;
1163         }
1164
1165         if (ufs->config.upperdir) {
1166                 ufs->upper_mnt = clone_private_mount(&upperpath);
1167                 err = PTR_ERR(ufs->upper_mnt);
1168                 if (IS_ERR(ufs->upper_mnt)) {
1169                         pr_err("overlayfs: failed to clone upperpath\n");
1170                         goto out_put_lowerpath;
1171                 }
1172                 /* Don't inherit atime flags */
1173                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1174
1175                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
1176
1177                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1178                 err = PTR_ERR(ufs->workdir);
1179                 if (IS_ERR(ufs->workdir)) {
1180                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1181                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1182                         sb->s_flags |= MS_RDONLY;
1183                         ufs->workdir = NULL;
1184                 }
1185
1186                 /*
1187                  * Upper should support d_type, else whiteouts are visible.
1188                  * Given workdir and upper are on same fs, we can do
1189                  * iterate_dir() on workdir. This check requires successful
1190                  * creation of workdir in previous step.
1191                  */
1192                 if (ufs->workdir) {
1193                         err = ovl_check_d_type_supported(&workpath);
1194                         if (err < 0)
1195                                 goto out_put_workdir;
1196
1197                         /*
1198                          * We allowed this configuration and don't want to
1199                          * break users over kernel upgrade. So warn instead
1200                          * of erroring out.
1201                          */
1202                         if (!err)
1203                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1204                 }
1205         }
1206
1207         err = -ENOMEM;
1208         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1209         if (ufs->lower_mnt == NULL)
1210                 goto out_put_workdir;
1211         for (i = 0; i < numlower; i++) {
1212                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1213
1214                 err = PTR_ERR(mnt);
1215                 if (IS_ERR(mnt)) {
1216                         pr_err("overlayfs: failed to clone lowerpath\n");
1217                         goto out_put_lower_mnt;
1218                 }
1219                 /*
1220                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1221                  * will fail instead of modifying lower fs.
1222                  */
1223                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1224
1225                 ufs->lower_mnt[ufs->numlower] = mnt;
1226                 ufs->numlower++;
1227         }
1228
1229         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1230         if (!ufs->upper_mnt)
1231                 sb->s_flags |= MS_RDONLY;
1232
1233         if (remote)
1234                 sb->s_d_op = &ovl_reval_dentry_operations;
1235         else
1236                 sb->s_d_op = &ovl_dentry_operations;
1237
1238         ufs->creator_cred = prepare_creds();
1239         if (!ufs->creator_cred)
1240                 goto out_put_lower_mnt;
1241
1242         err = -ENOMEM;
1243         oe = ovl_alloc_entry(numlower);
1244         if (!oe)
1245                 goto out_put_cred;
1246
1247         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR));
1248         if (!root_dentry)
1249                 goto out_free_oe;
1250
1251         mntput(upperpath.mnt);
1252         for (i = 0; i < numlower; i++)
1253                 mntput(stack[i].mnt);
1254         path_put(&workpath);
1255         kfree(lowertmp);
1256
1257         oe->__upperdentry = upperpath.dentry;
1258         for (i = 0; i < numlower; i++) {
1259                 oe->lowerstack[i].dentry = stack[i].dentry;
1260                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1261         }
1262         kfree(stack);
1263
1264         root_dentry->d_fsdata = oe;
1265
1266         realinode = d_inode(ovl_dentry_real(root_dentry));
1267         ovl_inode_init(d_inode(root_dentry), realinode, !!upperpath.dentry);
1268         ovl_copyattr(realinode, d_inode(root_dentry));
1269
1270         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1271         sb->s_op = &ovl_super_operations;
1272         if (IS_ENABLED(CONFIG_FS_POSIX_ACL))
1273                 sb->s_xattr = ovl_xattr_handlers;
1274         else
1275                 sb->s_xattr = ovl_xattr_noacl_handlers;
1276         sb->s_root = root_dentry;
1277         sb->s_fs_info = ufs;
1278         sb->s_flags |= MS_POSIXACL;
1279
1280         return 0;
1281
1282 out_free_oe:
1283         kfree(oe);
1284 out_put_cred:
1285         put_cred(ufs->creator_cred);
1286 out_put_lower_mnt:
1287         for (i = 0; i < ufs->numlower; i++)
1288                 mntput(ufs->lower_mnt[i]);
1289         kfree(ufs->lower_mnt);
1290 out_put_workdir:
1291         dput(ufs->workdir);
1292         mntput(ufs->upper_mnt);
1293 out_put_lowerpath:
1294         for (i = 0; i < numlower; i++)
1295                 path_put(&stack[i]);
1296         kfree(stack);
1297 out_free_lowertmp:
1298         kfree(lowertmp);
1299 out_put_workpath:
1300         path_put(&workpath);
1301 out_put_upperpath:
1302         path_put(&upperpath);
1303 out_free_config:
1304         kfree(ufs->config.lowerdir);
1305         kfree(ufs->config.upperdir);
1306         kfree(ufs->config.workdir);
1307         kfree(ufs);
1308 out:
1309         return err;
1310 }
1311
1312 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1313                                 const char *dev_name, void *raw_data)
1314 {
1315         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1316 }
1317
1318 static struct file_system_type ovl_fs_type = {
1319         .owner          = THIS_MODULE,
1320         .name           = "overlay",
1321         .mount          = ovl_mount,
1322         .kill_sb        = kill_anon_super,
1323 };
1324 MODULE_ALIAS_FS("overlay");
1325
1326 static int __init ovl_init(void)
1327 {
1328         return register_filesystem(&ovl_fs_type);
1329 }
1330
1331 static void __exit ovl_exit(void)
1332 {
1333         unregister_filesystem(&ovl_fs_type);
1334 }
1335
1336 module_init(ovl_init);
1337 module_exit(ovl_exit);