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