fs/overlayfs/dir.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/xattr.h>
  13 #include <linux/security.h>
  14 #include <linux/cred.h>
  15 #include <linux/module.h>
  16 #include <linux/posix_acl.h>
  17 #include <linux/posix_acl_xattr.h>
  18 #include <linux/atomic.h>
  19 #include <linux/ratelimit.h>
  20 #include "overlayfs.h"
  21
  22 static unsigned short ovl_redirect_max = 256;
  23 module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
  24 MODULE_PARM_DESC(ovl_redirect_max,
  25                  "Maximum length of absolute redirect xattr value");
  26
  27 static int ovl_set_redirect(struct dentry *dentry, bool samedir);
  28
  29 int ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
  30 {
  31         int err;
  32
  33         dget(wdentry);
  34         if (d_is_dir(wdentry))
  35                 err = ovl_do_rmdir(wdir, wdentry);
  36         else
  37                 err = ovl_do_unlink(wdir, wdentry);
  38         dput(wdentry);
  39
  40         if (err) {
  41                 pr_err("overlayfs: cleanup of '%pd2' failed (%i)\n",
  42                        wdentry, err);
  43         }
  44
  45         return err;
  46 }
  47
  48 static struct dentry *ovl_lookup_temp(struct dentry *workdir)
  49 {
  50         struct dentry *temp;
  51         char name[20];
  52         static atomic_t temp_id = ATOMIC_INIT(0);
  53
  54         /* counter is allowed to wrap, since temp dentries are ephemeral */
  55         snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
  56
  57         temp = lookup_one_len(name, workdir, strlen(name));
  58         if (!IS_ERR(temp) && temp->d_inode) {
  59                 pr_err("overlayfs: workdir/%s already exists\n", name);
  60                 dput(temp);
  61                 temp = ERR_PTR(-EIO);
  62         }
  63
  64         return temp;
  65 }
  66
  67 /* caller holds i_mutex on workdir */
  68 static struct dentry *ovl_whiteout(struct dentry *workdir)
  69 {
  70         int err;
  71         struct dentry *whiteout;
  72         struct inode *wdir = workdir->d_inode;
  73
  74         whiteout = ovl_lookup_temp(workdir);
  75         if (IS_ERR(whiteout))
  76                 return whiteout;
  77
  78         err = ovl_do_whiteout(wdir, whiteout);
  79         if (err) {
  80                 dput(whiteout);
  81                 whiteout = ERR_PTR(err);
  82         }
  83
  84         return whiteout;
  85 }
  86
  87 /* Caller must hold i_mutex on both workdir and dir */
  88 int ovl_cleanup_and_whiteout(struct dentry *workdir, struct inode *dir,
  89                              struct dentry *dentry)
  90 {
  91         struct inode *wdir = workdir->d_inode;
  92         struct dentry *whiteout;
  93         int err;
  94         int flags = 0;
  95
  96         whiteout = ovl_whiteout(workdir);
  97         err = PTR_ERR(whiteout);
  98         if (IS_ERR(whiteout))
  99                 return err;
 100
 101         if (d_is_dir(dentry))
 102                 flags = RENAME_EXCHANGE;
 103
 104         err = ovl_do_rename(wdir, whiteout, dir, dentry, flags);
 105         if (err)
 106                 goto kill_whiteout;
 107         if (flags)
 108                 ovl_cleanup(wdir, dentry);
 109
 110 out:
 111         dput(whiteout);
 112         return err;
 113
 114 kill_whiteout:
 115         ovl_cleanup(wdir, whiteout);
 116         goto out;
 117 }
 118
 119 int ovl_mkdir_real(struct inode *dir, struct dentry **newdentry, umode_t mode)
 120 {
 121         int err;
 122         struct dentry *d, *dentry = *newdentry;
 123
 124         err = ovl_do_mkdir(dir, dentry, mode);
 125         if (err)
 126                 return err;
 127
 128         if (likely(!d_unhashed(dentry)))
 129                 return 0;
 130
 131         /*
 132          * vfs_mkdir() may succeed and leave the dentry passed
 133          * to it unhashed and negative. If that happens, try to
 134          * lookup a new hashed and positive dentry.
 135          */
 136         d = lookup_one_len(dentry->d_name.name, dentry->d_parent,
 137                            dentry->d_name.len);
 138         if (IS_ERR(d)) {
 139                 pr_warn("overlayfs: failed lookup after mkdir (%pd2, err=%i).\n",
 140                         dentry, err);
 141                 return PTR_ERR(d);
 142         }
 143         dput(dentry);
 144         *newdentry = d;
 145
 146         return 0;
 147 }
 148
 149 struct dentry *ovl_create_real(struct inode *dir, struct dentry *newdentry,
 150                                struct ovl_cattr *attr)
 151 {
 152         int err;
 153
 154         if (IS_ERR(newdentry))
 155                 return newdentry;
 156
 157         err = -ESTALE;
 158         if (newdentry->d_inode)
 159                 goto out;
 160
 161         if (attr->hardlink) {
 162                 err = ovl_do_link(attr->hardlink, dir, newdentry);
 163         } else {
 164                 switch (attr->mode & S_IFMT) {
 165                 case S_IFREG:
 166                         err = ovl_do_create(dir, newdentry, attr->mode);
 167                         break;
 168
 169                 case S_IFDIR:
 170                         /* mkdir is special... */
 171                         err =  ovl_mkdir_real(dir, &newdentry, attr->mode);
 172                         break;
 173
 174                 case S_IFCHR:
 175                 case S_IFBLK:
 176                 case S_IFIFO:
 177                 case S_IFSOCK:
 178                         err = ovl_do_mknod(dir, newdentry, attr->mode,
 179                                            attr->rdev);
 180                         break;
 181
 182                 case S_IFLNK:
 183                         err = ovl_do_symlink(dir, newdentry, attr->link);
 184                         break;
 185
 186                 default:
 187                         err = -EPERM;
 188                 }
 189         }
 190         if (!err && WARN_ON(!newdentry->d_inode)) {
 191                 /*
 192                  * Not quite sure if non-instantiated dentry is legal or not.
 193                  * VFS doesn't seem to care so check and warn here.
 194                  */
 195                 err = -EIO;
 196         }
 197 out:
 198         if (err) {
 199                 dput(newdentry);
 200                 return ERR_PTR(err);
 201         }
 202         return newdentry;
 203 }
 204
 205 struct dentry *ovl_create_temp(struct dentry *workdir, struct ovl_cattr *attr)
 206 {
 207         return ovl_create_real(d_inode(workdir), ovl_lookup_temp(workdir),
 208                                attr);
 209 }
 210
 211 static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
 212                                int xerr)
 213 {
 214         int err;
 215
 216         err = ovl_check_setxattr(dentry, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
 217         if (!err)
 218                 ovl_dentry_set_opaque(dentry);
 219
 220         return err;
 221 }
 222
 223 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
 224 {
 225         /*
 226          * Fail with -EIO when trying to create opaque dir and upper doesn't
 227          * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
 228          * return a specific error for noxattr case.
 229          */
 230         return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
 231 }
 232
 233 /*
 234  * Common operations required to be done after creation of file on upper.
 235  * If @hardlink is false, then @inode is a pre-allocated inode, we may or
 236  * may not use to instantiate the new dentry.
 237  */
 238 static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
 239                            struct dentry *newdentry, bool hardlink)
 240 {
 241         struct ovl_inode_params oip = {
 242                 .upperdentry = newdentry,
 243                 .newinode = inode,
 244         };
 245
 246         ovl_dir_modified(dentry->d_parent, false);
 247         ovl_dentry_set_upper_alias(dentry);
 248         if (!hardlink) {
 249                 /*
 250                  * ovl_obtain_alias() can be called after ovl_create_real()
 251                  * and before we get here, so we may get an inode from cache
 252                  * with the same real upperdentry that is not the inode we
 253                  * pre-allocated.  In this case we will use the cached inode
 254                  * to instantiate the new dentry.
 255                  *
 256                  * XXX: if we ever use ovl_obtain_alias() to decode directory
 257                  * file handles, need to use ovl_get_inode_locked() and
 258                  * d_instantiate_new() here to prevent from creating two
 259                  * hashed directory inode aliases.
 260                  */
 261                 inode = ovl_get_inode(dentry->d_sb, &oip);
 262                 if (IS_ERR(inode))
 263                         return PTR_ERR(inode);
 264         } else {
 265                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
 266                 dput(newdentry);
 267                 inc_nlink(inode);
 268         }
 269
 270         d_instantiate(dentry, inode);
 271         if (inode != oip.newinode) {
 272                 pr_warn_ratelimited("overlayfs: newly created inode found in cache (%pd2)\n",
 273                                     dentry);
 274         }
 275
 276         /* Force lookup of new upper hardlink to find its lower */
 277         if (hardlink)
 278                 d_drop(dentry);
 279
 280         return 0;
 281 }
 282
 283 static bool ovl_type_merge(struct dentry *dentry)
 284 {
 285         return OVL_TYPE_MERGE(ovl_path_type(dentry));
 286 }
 287
 288 static bool ovl_type_origin(struct dentry *dentry)
 289 {
 290         return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
 291 }
 292
 293 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
 294                             struct ovl_cattr *attr)
 295 {
 296         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 297         struct inode *udir = upperdir->d_inode;
 298         struct dentry *newdentry;
 299         int err;
 300
 301         if (!attr->hardlink && !IS_POSIXACL(udir))
 302                 attr->mode &= ~current_umask();
 303
 304         inode_lock_nested(udir, I_MUTEX_PARENT);
 305         newdentry = ovl_create_real(udir,
 306                                     lookup_one_len(dentry->d_name.name,
 307                                                    upperdir,
 308                                                    dentry->d_name.len),
 309                                     attr);
 310         err = PTR_ERR(newdentry);
 311         if (IS_ERR(newdentry))
 312                 goto out_unlock;
 313
 314         if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry)) {
 315                 /* Setting opaque here is just an optimization, allow to fail */
 316                 ovl_set_opaque(dentry, newdentry);
 317         }
 318
 319         err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
 320         if (err)
 321                 goto out_cleanup;
 322 out_unlock:
 323         inode_unlock(udir);
 324         return err;
 325
 326 out_cleanup:
 327         ovl_cleanup(udir, newdentry);
 328         dput(newdentry);
 329         goto out_unlock;
 330 }
 331
 332 static struct dentry *ovl_clear_empty(struct dentry *dentry,
 333                                       struct list_head *list)
 334 {
 335         struct dentry *workdir = ovl_workdir(dentry);
 336         struct inode *wdir = workdir->d_inode;
 337         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 338         struct inode *udir = upperdir->d_inode;
 339         struct path upperpath;
 340         struct dentry *upper;
 341         struct dentry *opaquedir;
 342         struct kstat stat;
 343         int err;
 344
 345         if (WARN_ON(!workdir))
 346                 return ERR_PTR(-EROFS);
 347
 348         err = ovl_lock_rename_workdir(workdir, upperdir);
 349         if (err)
 350                 goto out;
 351
 352         ovl_path_upper(dentry, &upperpath);
 353         err = vfs_getattr(&upperpath, &stat,
 354                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 355         if (err)
 356                 goto out_unlock;
 357
 358         err = -ESTALE;
 359         if (!S_ISDIR(stat.mode))
 360                 goto out_unlock;
 361         upper = upperpath.dentry;
 362         if (upper->d_parent->d_inode != udir)
 363                 goto out_unlock;
 364
 365         opaquedir = ovl_create_temp(workdir, OVL_CATTR(stat.mode));
 366         err = PTR_ERR(opaquedir);
 367         if (IS_ERR(opaquedir))
 368                 goto out_unlock;
 369
 370         err = ovl_copy_xattr(upper, opaquedir);
 371         if (err)
 372                 goto out_cleanup;
 373
 374         err = ovl_set_opaque(dentry, opaquedir);
 375         if (err)
 376                 goto out_cleanup;
 377
 378         inode_lock(opaquedir->d_inode);
 379         err = ovl_set_attr(opaquedir, &stat);
 380         inode_unlock(opaquedir->d_inode);
 381         if (err)
 382                 goto out_cleanup;
 383
 384         err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
 385         if (err)
 386                 goto out_cleanup;
 387
 388         ovl_cleanup_whiteouts(upper, list);
 389         ovl_cleanup(wdir, upper);
 390         unlock_rename(workdir, upperdir);
 391
 392         /* dentry's upper doesn't match now, get rid of it */
 393         d_drop(dentry);
 394
 395         return opaquedir;
 396
 397 out_cleanup:
 398         ovl_cleanup(wdir, opaquedir);
 399         dput(opaquedir);
 400 out_unlock:
 401         unlock_rename(workdir, upperdir);
 402 out:
 403         return ERR_PTR(err);
 404 }
 405
 406 static int ovl_set_upper_acl(struct dentry *upperdentry, const char *name,
 407                              const struct posix_acl *acl)
 408 {
 409         void *buffer;
 410         size_t size;
 411         int err;
 412
 413         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
 414                 return 0;
 415
 416         size = posix_acl_to_xattr(NULL, acl, NULL, 0);
 417         buffer = kmalloc(size, GFP_KERNEL);
 418         if (!buffer)
 419                 return -ENOMEM;
 420
 421         size = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
 422         err = size;
 423         if (err < 0)
 424                 goto out_free;
 425
 426         err = vfs_setxattr(upperdentry, name, buffer, size, XATTR_CREATE);
 427 out_free:
 428         kfree(buffer);
 429         return err;
 430 }
 431
 432 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 433                                     struct ovl_cattr *cattr)
 434 {
 435         struct dentry *workdir = ovl_workdir(dentry);
 436         struct inode *wdir = workdir->d_inode;
 437         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 438         struct inode *udir = upperdir->d_inode;
 439         struct dentry *upper;
 440         struct dentry *newdentry;
 441         int err;
 442         struct posix_acl *acl, *default_acl;
 443         bool hardlink = !!cattr->hardlink;
 444
 445         if (WARN_ON(!workdir))
 446                 return -EROFS;
 447
 448         if (!hardlink) {
 449                 err = posix_acl_create(dentry->d_parent->d_inode,
 450                                        &cattr->mode, &default_acl, &acl);
 451                 if (err)
 452                         return err;
 453         }
 454
 455         err = ovl_lock_rename_workdir(workdir, upperdir);
 456         if (err)
 457                 goto out;
 458
 459         upper = lookup_one_len(dentry->d_name.name, upperdir,
 460                                dentry->d_name.len);
 461         err = PTR_ERR(upper);
 462         if (IS_ERR(upper))
 463                 goto out_unlock;
 464
 465         err = -ESTALE;
 466         if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
 467                 goto out_dput;
 468
 469         newdentry = ovl_create_temp(workdir, cattr);
 470         err = PTR_ERR(newdentry);
 471         if (IS_ERR(newdentry))
 472                 goto out_dput;
 473
 474         /*
 475          * mode could have been mutilated due to umask (e.g. sgid directory)
 476          */
 477         if (!hardlink &&
 478             !S_ISLNK(cattr->mode) &&
 479             newdentry->d_inode->i_mode != cattr->mode) {
 480                 struct iattr attr = {
 481                         .ia_valid = ATTR_MODE,
 482                         .ia_mode = cattr->mode,
 483                 };
 484                 inode_lock(newdentry->d_inode);
 485                 err = notify_change(newdentry, &attr, NULL);
 486                 inode_unlock(newdentry->d_inode);
 487                 if (err)
 488                         goto out_cleanup;
 489         }
 490         if (!hardlink) {
 491                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
 492                                         acl);
 493                 if (err)
 494                         goto out_cleanup;
 495
 496                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
 497                                         default_acl);
 498                 if (err)
 499                         goto out_cleanup;
 500         }
 501
 502         if (!hardlink && S_ISDIR(cattr->mode)) {
 503                 err = ovl_set_opaque(dentry, newdentry);
 504                 if (err)
 505                         goto out_cleanup;
 506
 507                 err = ovl_do_rename(wdir, newdentry, udir, upper,
 508                                     RENAME_EXCHANGE);
 509                 if (err)
 510                         goto out_cleanup;
 511
 512                 ovl_cleanup(wdir, upper);
 513         } else {
 514                 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 515                 if (err)
 516                         goto out_cleanup;
 517         }
 518         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 519         if (err) {
 520                 ovl_cleanup(udir, newdentry);
 521                 dput(newdentry);
 522         }
 523 out_dput:
 524         dput(upper);
 525 out_unlock:
 526         unlock_rename(workdir, upperdir);
 527 out:
 528         if (!hardlink) {
 529                 posix_acl_release(acl);
 530                 posix_acl_release(default_acl);
 531         }
 532         return err;
 533
 534 out_cleanup:
 535         ovl_cleanup(wdir, newdentry);
 536         dput(newdentry);
 537         goto out_dput;
 538 }
 539
 540 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 541                               struct ovl_cattr *attr, bool origin)
 542 {
 543         int err;
 544         const struct cred *old_cred;
 545         struct cred *override_cred;
 546         struct dentry *parent = dentry->d_parent;
 547
 548         err = ovl_copy_up(parent);
 549         if (err)
 550                 return err;
 551
 552         old_cred = ovl_override_creds(dentry->d_sb);
 553
 554         /*
 555          * When linking a file with copy up origin into a new parent, mark the
 556          * new parent dir "impure".
 557          */
 558         if (origin) {
 559                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 560                 if (err)
 561                         goto out_revert_creds;
 562         }
 563
 564         err = -ENOMEM;
 565         override_cred = prepare_creds();
 566         if (override_cred) {
 567                 override_cred->fsuid = inode->i_uid;
 568                 override_cred->fsgid = inode->i_gid;
 569                 if (!attr->hardlink) {
 570                         err = security_dentry_create_files_as(dentry,
 571                                         attr->mode, &dentry->d_name, old_cred,
 572                                         override_cred);
 573                         if (err) {
 574                                 put_cred(override_cred);
 575                                 goto out_revert_creds;
 576                         }
 577                 }
 578                 put_cred(override_creds(override_cred));
 579                 put_cred(override_cred);
 580
 581                 if (!ovl_dentry_is_whiteout(dentry))
 582                         err = ovl_create_upper(dentry, inode, attr);
 583                 else
 584                         err = ovl_create_over_whiteout(dentry, inode, attr);
 585         }
 586 out_revert_creds:
 587         revert_creds(old_cred);
 588         return err;
 589 }
 590
 591 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 592                              const char *link)
 593 {
 594         int err;
 595         struct inode *inode;
 596         struct ovl_cattr attr = {
 597                 .rdev = rdev,
 598                 .link = link,
 599         };
 600
 601         err = ovl_want_write(dentry);
 602         if (err)
 603                 goto out;
 604
 605         /* Preallocate inode to be used by ovl_get_inode() */
 606         err = -ENOMEM;
 607         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 608         if (!inode)
 609                 goto out_drop_write;
 610
 611         spin_lock(&inode->i_lock);
 612         inode->i_state |= I_CREATING;
 613         spin_unlock(&inode->i_lock);
 614
 615         inode_init_owner(inode, dentry->d_parent->d_inode, mode);
 616         attr.mode = inode->i_mode;
 617
 618         err = ovl_create_or_link(dentry, inode, &attr, false);
 619         /* Did we end up using the preallocated inode? */
 620         if (inode != d_inode(dentry))
 621                 iput(inode);
 622
 623 out_drop_write:
 624         ovl_drop_write(dentry);
 625 out:
 626         return err;
 627 }
 628
 629 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 630                       bool excl)
 631 {
 632         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 633 }
 634
 635 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 636 {
 637         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 638 }
 639
 640 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
 641                      dev_t rdev)
 642 {
 643         /* Don't allow creation of "whiteout" on overlay */
 644         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 645                 return -EPERM;
 646
 647         return ovl_create_object(dentry, mode, rdev, NULL);
 648 }
 649
 650 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
 651                        const char *link)
 652 {
 653         return ovl_create_object(dentry, S_IFLNK, 0, link);
 654 }
 655
 656 static int ovl_set_link_redirect(struct dentry *dentry)
 657 {
 658         const struct cred *old_cred;
 659         int err;
 660
 661         old_cred = ovl_override_creds(dentry->d_sb);
 662         err = ovl_set_redirect(dentry, false);
 663         revert_creds(old_cred);
 664
 665         return err;
 666 }
 667
 668 static int ovl_link(struct dentry *old, struct inode *newdir,
 669                     struct dentry *new)
 670 {
 671         int err;
 672         bool locked = false;
 673         struct inode *inode;
 674
 675         err = ovl_want_write(old);
 676         if (err)
 677                 goto out;
 678
 679         err = ovl_copy_up(old);
 680         if (err)
 681                 goto out_drop_write;
 682
 683         err = ovl_copy_up(new->d_parent);
 684         if (err)
 685                 goto out_drop_write;
 686
 687         if (ovl_is_metacopy_dentry(old)) {
 688                 err = ovl_set_link_redirect(old);
 689                 if (err)
 690                         goto out_drop_write;
 691         }
 692
 693         err = ovl_nlink_start(old, &locked);
 694         if (err)
 695                 goto out_drop_write;
 696
 697         inode = d_inode(old);
 698         ihold(inode);
 699
 700         err = ovl_create_or_link(new, inode,
 701                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 702                         ovl_type_origin(old));
 703         if (err)
 704                 iput(inode);
 705
 706         ovl_nlink_end(old, locked);
 707 out_drop_write:
 708         ovl_drop_write(old);
 709 out:
 710         return err;
 711 }
 712
 713 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 714 {
 715         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 716 }
 717
 718 static int ovl_remove_and_whiteout(struct dentry *dentry,
 719                                    struct list_head *list)
 720 {
 721         struct dentry *workdir = ovl_workdir(dentry);
 722         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 723         struct dentry *upper;
 724         struct dentry *opaquedir = NULL;
 725         int err;
 726
 727         if (WARN_ON(!workdir))
 728                 return -EROFS;
 729
 730         if (!list_empty(list)) {
 731                 opaquedir = ovl_clear_empty(dentry, list);
 732                 err = PTR_ERR(opaquedir);
 733                 if (IS_ERR(opaquedir))
 734                         goto out;
 735         }
 736
 737         err = ovl_lock_rename_workdir(workdir, upperdir);
 738         if (err)
 739                 goto out_dput;
 740
 741         upper = lookup_one_len(dentry->d_name.name, upperdir,
 742                                dentry->d_name.len);
 743         err = PTR_ERR(upper);
 744         if (IS_ERR(upper))
 745                 goto out_unlock;
 746
 747         err = -ESTALE;
 748         if ((opaquedir && upper != opaquedir) ||
 749             (!opaquedir && ovl_dentry_upper(dentry) &&
 750              !ovl_matches_upper(dentry, upper))) {
 751                 goto out_dput_upper;
 752         }
 753
 754         err = ovl_cleanup_and_whiteout(workdir, d_inode(upperdir), upper);
 755         if (err)
 756                 goto out_d_drop;
 757
 758         ovl_dir_modified(dentry->d_parent, true);
 759 out_d_drop:
 760         d_drop(dentry);
 761 out_dput_upper:
 762         dput(upper);
 763 out_unlock:
 764         unlock_rename(workdir, upperdir);
 765 out_dput:
 766         dput(opaquedir);
 767 out:
 768         return err;
 769 }
 770
 771 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 772                             struct list_head *list)
 773 {
 774         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 775         struct inode *dir = upperdir->d_inode;
 776         struct dentry *upper;
 777         struct dentry *opaquedir = NULL;
 778         int err;
 779
 780         if (!list_empty(list)) {
 781                 opaquedir = ovl_clear_empty(dentry, list);
 782                 err = PTR_ERR(opaquedir);
 783                 if (IS_ERR(opaquedir))
 784                         goto out;
 785         }
 786
 787         inode_lock_nested(dir, I_MUTEX_PARENT);
 788         upper = lookup_one_len(dentry->d_name.name, upperdir,
 789                                dentry->d_name.len);
 790         err = PTR_ERR(upper);
 791         if (IS_ERR(upper))
 792                 goto out_unlock;
 793
 794         err = -ESTALE;
 795         if ((opaquedir && upper != opaquedir) ||
 796             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 797                 goto out_dput_upper;
 798
 799         if (is_dir)
 800                 err = vfs_rmdir(dir, upper);
 801         else
 802                 err = vfs_unlink(dir, upper, NULL);
 803         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 804
 805         /*
 806          * Keeping this dentry hashed would mean having to release
 807          * upperpath/lowerpath, which could only be done if we are the
 808          * sole user of this dentry.  Too tricky...  Just unhash for
 809          * now.
 810          */
 811         if (!err)
 812                 d_drop(dentry);
 813 out_dput_upper:
 814         dput(upper);
 815 out_unlock:
 816         inode_unlock(dir);
 817         dput(opaquedir);
 818 out:
 819         return err;
 820 }
 821
 822 static bool ovl_pure_upper(struct dentry *dentry)
 823 {
 824         return !ovl_dentry_lower(dentry) &&
 825                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 826 }
 827
 828 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 829 {
 830         int err;
 831         bool locked = false;
 832         const struct cred *old_cred;
 833         struct dentry *upperdentry;
 834         bool lower_positive = ovl_lower_positive(dentry);
 835         LIST_HEAD(list);
 836
 837         /* No need to clean pure upper removed by vfs_rmdir() */
 838         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 839                 err = ovl_check_empty_dir(dentry, &list);
 840                 if (err)
 841                         goto out;
 842         }
 843
 844         err = ovl_want_write(dentry);
 845         if (err)
 846                 goto out;
 847
 848         err = ovl_copy_up(dentry->d_parent);
 849         if (err)
 850                 goto out_drop_write;
 851
 852         err = ovl_nlink_start(dentry, &locked);
 853         if (err)
 854                 goto out_drop_write;
 855
 856         old_cred = ovl_override_creds(dentry->d_sb);
 857         if (!lower_positive)
 858                 err = ovl_remove_upper(dentry, is_dir, &list);
 859         else
 860                 err = ovl_remove_and_whiteout(dentry, &list);
 861         revert_creds(old_cred);
 862         if (!err) {
 863                 if (is_dir)
 864                         clear_nlink(dentry->d_inode);
 865                 else
 866                         drop_nlink(dentry->d_inode);
 867         }
 868         ovl_nlink_end(dentry, locked);
 869
 870         /*
 871          * Copy ctime
 872          *
 873          * Note: we fail to update ctime if there was no copy-up, only a
 874          * whiteout
 875          */
 876         upperdentry = ovl_dentry_upper(dentry);
 877         if (upperdentry)
 878                 ovl_copyattr(d_inode(upperdentry), d_inode(dentry));
 879
 880 out_drop_write:
 881         ovl_drop_write(dentry);
 882 out:
 883         ovl_cache_free(&list);
 884         return err;
 885 }
 886
 887 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 888 {
 889         return ovl_do_remove(dentry, false);
 890 }
 891
 892 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 893 {
 894         return ovl_do_remove(dentry, true);
 895 }
 896
 897 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 898 {
 899         enum ovl_path_type type = ovl_path_type(dentry);
 900
 901         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 902 }
 903
 904 static bool ovl_can_move(struct dentry *dentry)
 905 {
 906         return ovl_redirect_dir(dentry->d_sb) ||
 907                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 908 }
 909
 910 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 911 {
 912         char *buf, *ret;
 913         struct dentry *d, *tmp;
 914         int buflen = ovl_redirect_max + 1;
 915
 916         if (!abs_redirect) {
 917                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 918                                GFP_KERNEL);
 919                 goto out;
 920         }
 921
 922         buf = ret = kmalloc(buflen, GFP_KERNEL);
 923         if (!buf)
 924                 goto out;
 925
 926         buflen--;
 927         buf[buflen] = '\0';
 928         for (d = dget(dentry); !IS_ROOT(d);) {
 929                 const char *name;
 930                 int thislen;
 931
 932                 spin_lock(&d->d_lock);
 933                 name = ovl_dentry_get_redirect(d);
 934                 if (name) {
 935                         thislen = strlen(name);
 936                 } else {
 937                         name = d->d_name.name;
 938                         thislen = d->d_name.len;
 939                 }
 940
 941                 /* If path is too long, fall back to userspace move */
 942                 if (thislen + (name[0] != '/') > buflen) {
 943                         ret = ERR_PTR(-EXDEV);
 944                         spin_unlock(&d->d_lock);
 945                         goto out_put;
 946                 }
 947
 948                 buflen -= thislen;
 949                 memcpy(&buf[buflen], name, thislen);
 950                 spin_unlock(&d->d_lock);
 951                 tmp = dget_parent(d);
 952
 953                 dput(d);
 954                 d = tmp;
 955
 956                 /* Absolute redirect: finished */
 957                 if (buf[buflen] == '/')
 958                         break;
 959                 buflen--;
 960                 buf[buflen] = '/';
 961         }
 962         ret = kstrdup(&buf[buflen], GFP_KERNEL);
 963 out_put:
 964         dput(d);
 965         kfree(buf);
 966 out:
 967         return ret ? ret : ERR_PTR(-ENOMEM);
 968 }
 969
 970 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
 971 {
 972         struct dentry *lowerdentry;
 973
 974         if (!samedir)
 975                 return true;
 976
 977         if (d_is_dir(dentry))
 978                 return false;
 979
 980         /*
 981          * For non-dir hardlinked files, we need absolute redirects
 982          * in general as two upper hardlinks could be in different
 983          * dirs. We could put a relative redirect now and convert
 984          * it to absolute redirect later. But when nlink > 1 and
 985          * indexing is on, that means relative redirect needs to be
 986          * converted to absolute during copy up of another lower
 987          * hardllink as well.
 988          *
 989          * So without optimizing too much, just check if lower is
 990          * a hard link or not. If lower is hard link, put absolute
 991          * redirect.
 992          */
 993         lowerdentry = ovl_dentry_lower(dentry);
 994         return (d_inode(lowerdentry)->i_nlink > 1);
 995 }
 996
 997 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
 998 {
 999         int err;
1000         const char *redirect = ovl_dentry_get_redirect(dentry);
1001         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1002
1003         if (redirect && (!absolute_redirect || redirect[0] == '/'))
1004                 return 0;
1005
1006         redirect = ovl_get_redirect(dentry, absolute_redirect);
1007         if (IS_ERR(redirect))
1008                 return PTR_ERR(redirect);
1009
1010         err = ovl_check_setxattr(dentry, ovl_dentry_upper(dentry),
1011                                  OVL_XATTR_REDIRECT,
1012                                  redirect, strlen(redirect), -EXDEV);
1013         if (!err) {
1014                 spin_lock(&dentry->d_lock);
1015                 ovl_dentry_set_redirect(dentry, redirect);
1016                 spin_unlock(&dentry->d_lock);
1017         } else {
1018                 kfree(redirect);
1019                 pr_warn_ratelimited("overlayfs: failed to set redirect (%i)\n",
1020                                     err);
1021                 /* Fall back to userspace copy-up */
1022                 err = -EXDEV;
1023         }
1024         return err;
1025 }
1026
1027 static int ovl_rename(struct inode *olddir, struct dentry *old,
1028                       struct inode *newdir, struct dentry *new,
1029                       unsigned int flags)
1030 {
1031         int err;
1032         bool locked = false;
1033         struct dentry *old_upperdir;
1034         struct dentry *new_upperdir;
1035         struct dentry *olddentry;
1036         struct dentry *newdentry;
1037         struct dentry *trap;
1038         bool old_opaque;
1039         bool new_opaque;
1040         bool cleanup_whiteout = false;
1041         bool overwrite = !(flags & RENAME_EXCHANGE);
1042         bool is_dir = d_is_dir(old);
1043         bool new_is_dir = d_is_dir(new);
1044         bool samedir = olddir == newdir;
1045         struct dentry *opaquedir = NULL;
1046         const struct cred *old_cred = NULL;
1047         LIST_HEAD(list);
1048
1049         err = -EINVAL;
1050         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1051                 goto out;
1052
1053         flags &= ~RENAME_NOREPLACE;
1054
1055         /* Don't copy up directory trees */
1056         err = -EXDEV;
1057         if (!ovl_can_move(old))
1058                 goto out;
1059         if (!overwrite && !ovl_can_move(new))
1060                 goto out;
1061
1062         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1063                 err = ovl_check_empty_dir(new, &list);
1064                 if (err)
1065                         goto out;
1066         }
1067
1068         if (overwrite) {
1069                 if (ovl_lower_positive(old)) {
1070                         if (!ovl_dentry_is_whiteout(new)) {
1071                                 /* Whiteout source */
1072                                 flags |= RENAME_WHITEOUT;
1073                         } else {
1074                                 /* Switch whiteouts */
1075                                 flags |= RENAME_EXCHANGE;
1076                         }
1077                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1078                         flags |= RENAME_EXCHANGE;
1079                         cleanup_whiteout = true;
1080                 }
1081         }
1082
1083         err = ovl_want_write(old);
1084         if (err)
1085                 goto out;
1086
1087         err = ovl_copy_up(old);
1088         if (err)
1089                 goto out_drop_write;
1090
1091         err = ovl_copy_up(new->d_parent);
1092         if (err)
1093                 goto out_drop_write;
1094         if (!overwrite) {
1095                 err = ovl_copy_up(new);
1096                 if (err)
1097                         goto out_drop_write;
1098         } else {
1099                 err = ovl_nlink_start(new, &locked);
1100                 if (err)
1101                         goto out_drop_write;
1102         }
1103
1104         old_cred = ovl_override_creds(old->d_sb);
1105
1106         if (!list_empty(&list)) {
1107                 opaquedir = ovl_clear_empty(new, &list);
1108                 err = PTR_ERR(opaquedir);
1109                 if (IS_ERR(opaquedir)) {
1110                         opaquedir = NULL;
1111                         goto out_revert_creds;
1112                 }
1113         }
1114
1115         old_upperdir = ovl_dentry_upper(old->d_parent);
1116         new_upperdir = ovl_dentry_upper(new->d_parent);
1117
1118         if (!samedir) {
1119                 /*
1120                  * When moving a merge dir or non-dir with copy up origin into
1121                  * a new parent, we are marking the new parent dir "impure".
1122                  * When ovl_iterate() iterates an "impure" upper dir, it will
1123                  * lookup the origin inodes of the entries to fill d_ino.
1124                  */
1125                 if (ovl_type_origin(old)) {
1126                         err = ovl_set_impure(new->d_parent, new_upperdir);
1127                         if (err)
1128                                 goto out_revert_creds;
1129                 }
1130                 if (!overwrite && ovl_type_origin(new)) {
1131                         err = ovl_set_impure(old->d_parent, old_upperdir);
1132                         if (err)
1133                                 goto out_revert_creds;
1134                 }
1135         }
1136
1137         trap = lock_rename(new_upperdir, old_upperdir);
1138
1139         olddentry = lookup_one_len(old->d_name.name, old_upperdir,
1140                                    old->d_name.len);
1141         err = PTR_ERR(olddentry);
1142         if (IS_ERR(olddentry))
1143                 goto out_unlock;
1144
1145         err = -ESTALE;
1146         if (!ovl_matches_upper(old, olddentry))
1147                 goto out_dput_old;
1148
1149         newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1150                                    new->d_name.len);
1151         err = PTR_ERR(newdentry);
1152         if (IS_ERR(newdentry))
1153                 goto out_dput_old;
1154
1155         old_opaque = ovl_dentry_is_opaque(old);
1156         new_opaque = ovl_dentry_is_opaque(new);
1157
1158         err = -ESTALE;
1159         if (d_inode(new) && ovl_dentry_upper(new)) {
1160                 if (opaquedir) {
1161                         if (newdentry != opaquedir)
1162                                 goto out_dput;
1163                 } else {
1164                         if (!ovl_matches_upper(new, newdentry))
1165                                 goto out_dput;
1166                 }
1167         } else {
1168                 if (!d_is_negative(newdentry)) {
1169                         if (!new_opaque || !ovl_is_whiteout(newdentry))
1170                                 goto out_dput;
1171                 } else {
1172                         if (flags & RENAME_EXCHANGE)
1173                                 goto out_dput;
1174                 }
1175         }
1176
1177         if (olddentry == trap)
1178                 goto out_dput;
1179         if (newdentry == trap)
1180                 goto out_dput;
1181
1182         if (olddentry->d_inode == newdentry->d_inode)
1183                 goto out_dput;
1184
1185         err = 0;
1186         if (ovl_type_merge_or_lower(old))
1187                 err = ovl_set_redirect(old, samedir);
1188         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1189                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1190         if (err)
1191                 goto out_dput;
1192
1193         if (!overwrite && ovl_type_merge_or_lower(new))
1194                 err = ovl_set_redirect(new, samedir);
1195         else if (!overwrite && new_is_dir && !new_opaque &&
1196                  ovl_type_merge(old->d_parent))
1197                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1198         if (err)
1199                 goto out_dput;
1200
1201         err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1202                             new_upperdir->d_inode, newdentry, flags);
1203         if (err)
1204                 goto out_dput;
1205
1206         if (cleanup_whiteout)
1207                 ovl_cleanup(old_upperdir->d_inode, newdentry);
1208
1209         if (overwrite && d_inode(new)) {
1210                 if (new_is_dir)
1211                         clear_nlink(d_inode(new));
1212                 else
1213                         drop_nlink(d_inode(new));
1214         }
1215
1216         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1217                          (!overwrite && ovl_type_origin(new)));
1218         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1219                          (d_inode(new) && ovl_type_origin(new)));
1220
1221         /* copy ctime: */
1222         ovl_copyattr(d_inode(olddentry), d_inode(old));
1223         if (d_inode(new) && ovl_dentry_upper(new))
1224                 ovl_copyattr(d_inode(newdentry), d_inode(new));
1225
1226 out_dput:
1227         dput(newdentry);
1228 out_dput_old:
1229         dput(olddentry);
1230 out_unlock:
1231         unlock_rename(new_upperdir, old_upperdir);
1232 out_revert_creds:
1233         revert_creds(old_cred);
1234         ovl_nlink_end(new, locked);
1235 out_drop_write:
1236         ovl_drop_write(old);
1237 out:
1238         dput(opaquedir);
1239         ovl_cache_free(&list);
1240         return err;
1241 }
1242
1243 const struct inode_operations ovl_dir_inode_operations = {
1244         .lookup         = ovl_lookup,
1245         .mkdir          = ovl_mkdir,
1246         .symlink        = ovl_symlink,
1247         .unlink         = ovl_unlink,
1248         .rmdir          = ovl_rmdir,
1249         .rename         = ovl_rename,
1250         .link           = ovl_link,
1251         .setattr        = ovl_setattr,
1252         .create         = ovl_create,
1253         .mknod          = ovl_mknod,
1254         .permission     = ovl_permission,
1255         .getattr        = ovl_getattr,
1256         .listxattr      = ovl_listxattr,
1257         .get_acl        = ovl_get_acl,
1258         .update_time    = ovl_update_time,
1259 };