fs/overlayfs/dir.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <linux/fs.h>
   8 #include <linux/namei.h>
   9 #include <linux/xattr.h>
  10 #include <linux/security.h>
  11 #include <linux/cred.h>
  12 #include <linux/module.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/posix_acl_xattr.h>
  15 #include <linux/atomic.h>
  16 #include <linux/ratelimit.h>
  17 #include "overlayfs.h"
  18
  19 static unsigned short ovl_redirect_max = 256;
  20 module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
  21 MODULE_PARM_DESC(redirect_max,
  22                  "Maximum length of absolute redirect xattr value");
  23
  24 static int ovl_set_redirect(struct dentry *dentry, bool samedir);
  25
  26 int ovl_cleanup(struct ovl_fs *ofs, struct inode *wdir, struct dentry *wdentry)
  27 {
  28         int err;
  29
  30         dget(wdentry);
  31         if (d_is_dir(wdentry))
  32                 err = ovl_do_rmdir(ofs, wdir, wdentry);
  33         else
  34                 err = ovl_do_unlink(ofs, wdir, wdentry);
  35         dput(wdentry);
  36
  37         if (err) {
  38                 pr_err("cleanup of '%pd2' failed (%i)\n",
  39                        wdentry, err);
  40         }
  41
  42         return err;
  43 }
  44
  45 struct dentry *ovl_lookup_temp(struct ovl_fs *ofs, struct dentry *workdir)
  46 {
  47         struct dentry *temp;
  48         char name[20];
  49         static atomic_t temp_id = ATOMIC_INIT(0);
  50
  51         /* counter is allowed to wrap, since temp dentries are ephemeral */
  52         snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
  53
  54         temp = ovl_lookup_upper(ofs, name, workdir, strlen(name));
  55         if (!IS_ERR(temp) && temp->d_inode) {
  56                 pr_err("workdir/%s already exists\n", name);
  57                 dput(temp);
  58                 temp = ERR_PTR(-EIO);
  59         }
  60
  61         return temp;
  62 }
  63
  64 /* caller holds i_mutex on workdir */
  65 static struct dentry *ovl_whiteout(struct ovl_fs *ofs)
  66 {
  67         int err;
  68         struct dentry *whiteout;
  69         struct dentry *workdir = ofs->workdir;
  70         struct inode *wdir = workdir->d_inode;
  71
  72         if (!ofs->whiteout) {
  73                 whiteout = ovl_lookup_temp(ofs, workdir);
  74                 if (IS_ERR(whiteout))
  75                         goto out;
  76
  77                 err = ovl_do_whiteout(ofs, wdir, whiteout);
  78                 if (err) {
  79                         dput(whiteout);
  80                         whiteout = ERR_PTR(err);
  81                         goto out;
  82                 }
  83                 ofs->whiteout = whiteout;
  84         }
  85
  86         if (ofs->share_whiteout) {
  87                 whiteout = ovl_lookup_temp(ofs, workdir);
  88                 if (IS_ERR(whiteout))
  89                         goto out;
  90
  91                 err = ovl_do_link(ofs, ofs->whiteout, wdir, whiteout);
  92                 if (!err)
  93                         goto out;
  94
  95                 if (err != -EMLINK) {
  96                         pr_warn("Failed to link whiteout - disabling whiteout inode sharing(nlink=%u, err=%i)\n",
  97                                 ofs->whiteout->d_inode->i_nlink, err);
  98                         ofs->share_whiteout = false;
  99                 }
 100                 dput(whiteout);
 101         }
 102         whiteout = ofs->whiteout;
 103         ofs->whiteout = NULL;
 104 out:
 105         return whiteout;
 106 }
 107
 108 /* Caller must hold i_mutex on both workdir and dir */
 109 int ovl_cleanup_and_whiteout(struct ovl_fs *ofs, struct inode *dir,
 110                              struct dentry *dentry)
 111 {
 112         struct inode *wdir = ofs->workdir->d_inode;
 113         struct dentry *whiteout;
 114         int err;
 115         int flags = 0;
 116
 117         whiteout = ovl_whiteout(ofs);
 118         err = PTR_ERR(whiteout);
 119         if (IS_ERR(whiteout))
 120                 return err;
 121
 122         if (d_is_dir(dentry))
 123                 flags = RENAME_EXCHANGE;
 124
 125         err = ovl_do_rename(ofs, wdir, whiteout, dir, dentry, flags);
 126         if (err)
 127                 goto kill_whiteout;
 128         if (flags)
 129                 ovl_cleanup(ofs, wdir, dentry);
 130
 131 out:
 132         dput(whiteout);
 133         return err;
 134
 135 kill_whiteout:
 136         ovl_cleanup(ofs, wdir, whiteout);
 137         goto out;
 138 }
 139
 140 int ovl_mkdir_real(struct ovl_fs *ofs, struct inode *dir,
 141                    struct dentry **newdentry, umode_t mode)
 142 {
 143         int err;
 144         struct dentry *d, *dentry = *newdentry;
 145
 146         err = ovl_do_mkdir(ofs, dir, dentry, mode);
 147         if (err)
 148                 return err;
 149
 150         if (likely(!d_unhashed(dentry)))
 151                 return 0;
 152
 153         /*
 154          * vfs_mkdir() may succeed and leave the dentry passed
 155          * to it unhashed and negative. If that happens, try to
 156          * lookup a new hashed and positive dentry.
 157          */
 158         d = ovl_lookup_upper(ofs, dentry->d_name.name, dentry->d_parent,
 159                              dentry->d_name.len);
 160         if (IS_ERR(d)) {
 161                 pr_warn("failed lookup after mkdir (%pd2, err=%i).\n",
 162                         dentry, err);
 163                 return PTR_ERR(d);
 164         }
 165         dput(dentry);
 166         *newdentry = d;
 167
 168         return 0;
 169 }
 170
 171 struct dentry *ovl_create_real(struct ovl_fs *ofs, struct inode *dir,
 172                                struct dentry *newdentry, struct ovl_cattr *attr)
 173 {
 174         int err;
 175
 176         if (IS_ERR(newdentry))
 177                 return newdentry;
 178
 179         err = -ESTALE;
 180         if (newdentry->d_inode)
 181                 goto out;
 182
 183         if (attr->hardlink) {
 184                 err = ovl_do_link(ofs, attr->hardlink, dir, newdentry);
 185         } else {
 186                 switch (attr->mode & S_IFMT) {
 187                 case S_IFREG:
 188                         err = ovl_do_create(ofs, dir, newdentry, attr->mode);
 189                         break;
 190
 191                 case S_IFDIR:
 192                         /* mkdir is special... */
 193                         err =  ovl_mkdir_real(ofs, dir, &newdentry, attr->mode);
 194                         break;
 195
 196                 case S_IFCHR:
 197                 case S_IFBLK:
 198                 case S_IFIFO:
 199                 case S_IFSOCK:
 200                         err = ovl_do_mknod(ofs, dir, newdentry, attr->mode,
 201                                            attr->rdev);
 202                         break;
 203
 204                 case S_IFLNK:
 205                         err = ovl_do_symlink(ofs, dir, newdentry, attr->link);
 206                         break;
 207
 208                 default:
 209                         err = -EPERM;
 210                 }
 211         }
 212         if (!err && WARN_ON(!newdentry->d_inode)) {
 213                 /*
 214                  * Not quite sure if non-instantiated dentry is legal or not.
 215                  * VFS doesn't seem to care so check and warn here.
 216                  */
 217                 err = -EIO;
 218         }
 219 out:
 220         if (err) {
 221                 dput(newdentry);
 222                 return ERR_PTR(err);
 223         }
 224         return newdentry;
 225 }
 226
 227 struct dentry *ovl_create_temp(struct ovl_fs *ofs, struct dentry *workdir,
 228                                struct ovl_cattr *attr)
 229 {
 230         return ovl_create_real(ofs, d_inode(workdir),
 231                                ovl_lookup_temp(ofs, workdir), attr);
 232 }
 233
 234 static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
 235                                int xerr)
 236 {
 237         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 238         int err;
 239
 240         err = ovl_check_setxattr(ofs, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
 241         if (!err)
 242                 ovl_dentry_set_opaque(dentry);
 243
 244         return err;
 245 }
 246
 247 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
 248 {
 249         /*
 250          * Fail with -EIO when trying to create opaque dir and upper doesn't
 251          * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
 252          * return a specific error for noxattr case.
 253          */
 254         return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
 255 }
 256
 257 /*
 258  * Common operations required to be done after creation of file on upper.
 259  * If @hardlink is false, then @inode is a pre-allocated inode, we may or
 260  * may not use to instantiate the new dentry.
 261  */
 262 static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
 263                            struct dentry *newdentry, bool hardlink)
 264 {
 265         struct ovl_inode_params oip = {
 266                 .upperdentry = newdentry,
 267                 .newinode = inode,
 268         };
 269
 270         ovl_dir_modified(dentry->d_parent, false);
 271         ovl_dentry_set_upper_alias(dentry);
 272         ovl_dentry_update_reval(dentry, newdentry,
 273                         DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 274
 275         if (!hardlink) {
 276                 /*
 277                  * ovl_obtain_alias() can be called after ovl_create_real()
 278                  * and before we get here, so we may get an inode from cache
 279                  * with the same real upperdentry that is not the inode we
 280                  * pre-allocated.  In this case we will use the cached inode
 281                  * to instantiate the new dentry.
 282                  *
 283                  * XXX: if we ever use ovl_obtain_alias() to decode directory
 284                  * file handles, need to use ovl_get_inode_locked() and
 285                  * d_instantiate_new() here to prevent from creating two
 286                  * hashed directory inode aliases.
 287                  */
 288                 inode = ovl_get_inode(dentry->d_sb, &oip);
 289                 if (IS_ERR(inode))
 290                         return PTR_ERR(inode);
 291                 if (inode == oip.newinode)
 292                         ovl_set_flag(OVL_UPPERDATA, inode);
 293         } else {
 294                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
 295                 dput(newdentry);
 296                 inc_nlink(inode);
 297         }
 298
 299         d_instantiate(dentry, inode);
 300         if (inode != oip.newinode) {
 301                 pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
 302                                     dentry);
 303         }
 304
 305         /* Force lookup of new upper hardlink to find its lower */
 306         if (hardlink)
 307                 d_drop(dentry);
 308
 309         return 0;
 310 }
 311
 312 static bool ovl_type_merge(struct dentry *dentry)
 313 {
 314         return OVL_TYPE_MERGE(ovl_path_type(dentry));
 315 }
 316
 317 static bool ovl_type_origin(struct dentry *dentry)
 318 {
 319         return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
 320 }
 321
 322 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
 323                             struct ovl_cattr *attr)
 324 {
 325         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 326         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 327         struct inode *udir = upperdir->d_inode;
 328         struct dentry *newdentry;
 329         int err;
 330
 331         if (!attr->hardlink && !IS_POSIXACL(udir))
 332                 attr->mode &= ~current_umask();
 333
 334         inode_lock_nested(udir, I_MUTEX_PARENT);
 335         newdentry = ovl_create_real(ofs, udir,
 336                                     ovl_lookup_upper(ofs, dentry->d_name.name,
 337                                                      upperdir, dentry->d_name.len),
 338                                     attr);
 339         err = PTR_ERR(newdentry);
 340         if (IS_ERR(newdentry))
 341                 goto out_unlock;
 342
 343         if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry) &&
 344             !ovl_allow_offline_changes(ofs)) {
 345                 /* Setting opaque here is just an optimization, allow to fail */
 346                 ovl_set_opaque(dentry, newdentry);
 347         }
 348
 349         err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
 350         if (err)
 351                 goto out_cleanup;
 352 out_unlock:
 353         inode_unlock(udir);
 354         return err;
 355
 356 out_cleanup:
 357         ovl_cleanup(ofs, udir, newdentry);
 358         dput(newdentry);
 359         goto out_unlock;
 360 }
 361
 362 static struct dentry *ovl_clear_empty(struct dentry *dentry,
 363                                       struct list_head *list)
 364 {
 365         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 366         struct dentry *workdir = ovl_workdir(dentry);
 367         struct inode *wdir = workdir->d_inode;
 368         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 369         struct inode *udir = upperdir->d_inode;
 370         struct path upperpath;
 371         struct dentry *upper;
 372         struct dentry *opaquedir;
 373         struct kstat stat;
 374         int err;
 375
 376         if (WARN_ON(!workdir))
 377                 return ERR_PTR(-EROFS);
 378
 379         err = ovl_lock_rename_workdir(workdir, upperdir);
 380         if (err)
 381                 goto out;
 382
 383         ovl_path_upper(dentry, &upperpath);
 384         err = vfs_getattr(&upperpath, &stat,
 385                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 386         if (err)
 387                 goto out_unlock;
 388
 389         err = -ESTALE;
 390         if (!S_ISDIR(stat.mode))
 391                 goto out_unlock;
 392         upper = upperpath.dentry;
 393         if (upper->d_parent->d_inode != udir)
 394                 goto out_unlock;
 395
 396         opaquedir = ovl_create_temp(ofs, workdir, OVL_CATTR(stat.mode));
 397         err = PTR_ERR(opaquedir);
 398         if (IS_ERR(opaquedir))
 399                 goto out_unlock;
 400
 401         err = ovl_copy_xattr(dentry->d_sb, &upperpath, opaquedir);
 402         if (err)
 403                 goto out_cleanup;
 404
 405         err = ovl_set_opaque(dentry, opaquedir);
 406         if (err)
 407                 goto out_cleanup;
 408
 409         inode_lock(opaquedir->d_inode);
 410         err = ovl_set_attr(ofs, opaquedir, &stat);
 411         inode_unlock(opaquedir->d_inode);
 412         if (err)
 413                 goto out_cleanup;
 414
 415         err = ovl_do_rename(ofs, wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
 416         if (err)
 417                 goto out_cleanup;
 418
 419         ovl_cleanup_whiteouts(ofs, upper, list);
 420         ovl_cleanup(ofs, wdir, upper);
 421         unlock_rename(workdir, upperdir);
 422
 423         /* dentry's upper doesn't match now, get rid of it */
 424         d_drop(dentry);
 425
 426         return opaquedir;
 427
 428 out_cleanup:
 429         ovl_cleanup(ofs, wdir, opaquedir);
 430         dput(opaquedir);
 431 out_unlock:
 432         unlock_rename(workdir, upperdir);
 433 out:
 434         return ERR_PTR(err);
 435 }
 436
 437 static int ovl_set_upper_acl(struct ovl_fs *ofs, struct dentry *upperdentry,
 438                              const char *name, const struct posix_acl *acl)
 439 {
 440         void *buffer;
 441         size_t size;
 442         int err;
 443
 444         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
 445                 return 0;
 446
 447         size = posix_acl_xattr_size(acl->a_count);
 448         buffer = kmalloc(size, GFP_KERNEL);
 449         if (!buffer)
 450                 return -ENOMEM;
 451
 452         err = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
 453         if (err < 0)
 454                 goto out_free;
 455
 456         err = ovl_do_setxattr(ofs, upperdentry, name, buffer, size, XATTR_CREATE);
 457 out_free:
 458         kfree(buffer);
 459         return err;
 460 }
 461
 462 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 463                                     struct ovl_cattr *cattr)
 464 {
 465         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 466         struct dentry *workdir = ovl_workdir(dentry);
 467         struct inode *wdir = workdir->d_inode;
 468         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 469         struct inode *udir = upperdir->d_inode;
 470         struct dentry *upper;
 471         struct dentry *newdentry;
 472         int err;
 473         struct posix_acl *acl, *default_acl;
 474         bool hardlink = !!cattr->hardlink;
 475
 476         if (WARN_ON(!workdir))
 477                 return -EROFS;
 478
 479         if (!hardlink) {
 480                 err = posix_acl_create(dentry->d_parent->d_inode,
 481                                        &cattr->mode, &default_acl, &acl);
 482                 if (err)
 483                         return err;
 484         }
 485
 486         err = ovl_lock_rename_workdir(workdir, upperdir);
 487         if (err)
 488                 goto out;
 489
 490         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 491                                  dentry->d_name.len);
 492         err = PTR_ERR(upper);
 493         if (IS_ERR(upper))
 494                 goto out_unlock;
 495
 496         err = -ESTALE;
 497         if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
 498                 goto out_dput;
 499
 500         newdentry = ovl_create_temp(ofs, workdir, cattr);
 501         err = PTR_ERR(newdentry);
 502         if (IS_ERR(newdentry))
 503                 goto out_dput;
 504
 505         /*
 506          * mode could have been mutilated due to umask (e.g. sgid directory)
 507          */
 508         if (!hardlink &&
 509             !S_ISLNK(cattr->mode) &&
 510             newdentry->d_inode->i_mode != cattr->mode) {
 511                 struct iattr attr = {
 512                         .ia_valid = ATTR_MODE,
 513                         .ia_mode = cattr->mode,
 514                 };
 515                 inode_lock(newdentry->d_inode);
 516                 err = ovl_do_notify_change(ofs, newdentry, &attr);
 517                 inode_unlock(newdentry->d_inode);
 518                 if (err)
 519                         goto out_cleanup;
 520         }
 521         if (!hardlink) {
 522                 err = ovl_set_upper_acl(ofs, newdentry,
 523                                         XATTR_NAME_POSIX_ACL_ACCESS, acl);
 524                 if (err)
 525                         goto out_cleanup;
 526
 527                 err = ovl_set_upper_acl(ofs, newdentry,
 528                                         XATTR_NAME_POSIX_ACL_DEFAULT, default_acl);
 529                 if (err)
 530                         goto out_cleanup;
 531         }
 532
 533         if (!hardlink && S_ISDIR(cattr->mode)) {
 534                 err = ovl_set_opaque(dentry, newdentry);
 535                 if (err)
 536                         goto out_cleanup;
 537
 538                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper,
 539                                     RENAME_EXCHANGE);
 540                 if (err)
 541                         goto out_cleanup;
 542
 543                 ovl_cleanup(ofs, wdir, upper);
 544         } else {
 545                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper, 0);
 546                 if (err)
 547                         goto out_cleanup;
 548         }
 549         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 550         if (err) {
 551                 ovl_cleanup(ofs, udir, newdentry);
 552                 dput(newdentry);
 553         }
 554 out_dput:
 555         dput(upper);
 556 out_unlock:
 557         unlock_rename(workdir, upperdir);
 558 out:
 559         if (!hardlink) {
 560                 posix_acl_release(acl);
 561                 posix_acl_release(default_acl);
 562         }
 563         return err;
 564
 565 out_cleanup:
 566         ovl_cleanup(ofs, wdir, newdentry);
 567         dput(newdentry);
 568         goto out_dput;
 569 }
 570
 571 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 572                               struct ovl_cattr *attr, bool origin)
 573 {
 574         int err;
 575         const struct cred *old_cred;
 576         struct cred *override_cred;
 577         struct dentry *parent = dentry->d_parent;
 578
 579         err = ovl_copy_up(parent);
 580         if (err)
 581                 return err;
 582
 583         old_cred = ovl_override_creds(dentry->d_sb);
 584
 585         /*
 586          * When linking a file with copy up origin into a new parent, mark the
 587          * new parent dir "impure".
 588          */
 589         if (origin) {
 590                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 591                 if (err)
 592                         goto out_revert_creds;
 593         }
 594
 595         if (!attr->hardlink) {
 596                 err = -ENOMEM;
 597                 override_cred = prepare_creds();
 598                 if (!override_cred)
 599                         goto out_revert_creds;
 600                 /*
 601                  * In the creation cases(create, mkdir, mknod, symlink),
 602                  * ovl should transfer current's fs{u,g}id to underlying
 603                  * fs. Because underlying fs want to initialize its new
 604                  * inode owner using current's fs{u,g}id. And in this
 605                  * case, the @inode is a new inode that is initialized
 606                  * in inode_init_owner() to current's fs{u,g}id. So use
 607                  * the inode's i_{u,g}id to override the cred's fs{u,g}id.
 608                  *
 609                  * But in the other hardlink case, ovl_link() does not
 610                  * create a new inode, so just use the ovl mounter's
 611                  * fs{u,g}id.
 612                  */
 613                 override_cred->fsuid = inode->i_uid;
 614                 override_cred->fsgid = inode->i_gid;
 615                 err = security_dentry_create_files_as(dentry,
 616                                 attr->mode, &dentry->d_name, old_cred,
 617                                 override_cred);
 618                 if (err) {
 619                         put_cred(override_cred);
 620                         goto out_revert_creds;
 621                 }
 622                 put_cred(override_creds(override_cred));
 623                 put_cred(override_cred);
 624         }
 625
 626         if (!ovl_dentry_is_whiteout(dentry))
 627                 err = ovl_create_upper(dentry, inode, attr);
 628         else
 629                 err = ovl_create_over_whiteout(dentry, inode, attr);
 630
 631 out_revert_creds:
 632         revert_creds(old_cred);
 633         return err;
 634 }
 635
 636 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 637                              const char *link)
 638 {
 639         int err;
 640         struct inode *inode;
 641         struct ovl_cattr attr = {
 642                 .rdev = rdev,
 643                 .link = link,
 644         };
 645
 646         err = ovl_want_write(dentry);
 647         if (err)
 648                 goto out;
 649
 650         /* Preallocate inode to be used by ovl_get_inode() */
 651         err = -ENOMEM;
 652         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 653         if (!inode)
 654                 goto out_drop_write;
 655
 656         spin_lock(&inode->i_lock);
 657         inode->i_state |= I_CREATING;
 658         spin_unlock(&inode->i_lock);
 659
 660         inode_init_owner(&init_user_ns, inode, dentry->d_parent->d_inode, mode);
 661         attr.mode = inode->i_mode;
 662
 663         err = ovl_create_or_link(dentry, inode, &attr, false);
 664         /* Did we end up using the preallocated inode? */
 665         if (inode != d_inode(dentry))
 666                 iput(inode);
 667
 668 out_drop_write:
 669         ovl_drop_write(dentry);
 670 out:
 671         return err;
 672 }
 673
 674 static int ovl_create(struct user_namespace *mnt_userns, struct inode *dir,
 675                       struct dentry *dentry, umode_t mode, bool excl)
 676 {
 677         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 678 }
 679
 680 static int ovl_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 681                      struct dentry *dentry, umode_t mode)
 682 {
 683         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 684 }
 685
 686 static int ovl_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 687                      struct dentry *dentry, umode_t mode, dev_t rdev)
 688 {
 689         /* Don't allow creation of "whiteout" on overlay */
 690         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 691                 return -EPERM;
 692
 693         return ovl_create_object(dentry, mode, rdev, NULL);
 694 }
 695
 696 static int ovl_symlink(struct user_namespace *mnt_userns, struct inode *dir,
 697                        struct dentry *dentry, const char *link)
 698 {
 699         return ovl_create_object(dentry, S_IFLNK, 0, link);
 700 }
 701
 702 static int ovl_set_link_redirect(struct dentry *dentry)
 703 {
 704         const struct cred *old_cred;
 705         int err;
 706
 707         old_cred = ovl_override_creds(dentry->d_sb);
 708         err = ovl_set_redirect(dentry, false);
 709         revert_creds(old_cred);
 710
 711         return err;
 712 }
 713
 714 static int ovl_link(struct dentry *old, struct inode *newdir,
 715                     struct dentry *new)
 716 {
 717         int err;
 718         struct inode *inode;
 719
 720         err = ovl_want_write(old);
 721         if (err)
 722                 goto out;
 723
 724         err = ovl_copy_up(old);
 725         if (err)
 726                 goto out_drop_write;
 727
 728         err = ovl_copy_up(new->d_parent);
 729         if (err)
 730                 goto out_drop_write;
 731
 732         if (ovl_is_metacopy_dentry(old)) {
 733                 err = ovl_set_link_redirect(old);
 734                 if (err)
 735                         goto out_drop_write;
 736         }
 737
 738         err = ovl_nlink_start(old);
 739         if (err)
 740                 goto out_drop_write;
 741
 742         inode = d_inode(old);
 743         ihold(inode);
 744
 745         err = ovl_create_or_link(new, inode,
 746                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 747                         ovl_type_origin(old));
 748         if (err)
 749                 iput(inode);
 750
 751         ovl_nlink_end(old);
 752 out_drop_write:
 753         ovl_drop_write(old);
 754 out:
 755         return err;
 756 }
 757
 758 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 759 {
 760         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 761 }
 762
 763 static int ovl_remove_and_whiteout(struct dentry *dentry,
 764                                    struct list_head *list)
 765 {
 766         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 767         struct dentry *workdir = ovl_workdir(dentry);
 768         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 769         struct dentry *upper;
 770         struct dentry *opaquedir = NULL;
 771         int err;
 772
 773         if (WARN_ON(!workdir))
 774                 return -EROFS;
 775
 776         if (!list_empty(list)) {
 777                 opaquedir = ovl_clear_empty(dentry, list);
 778                 err = PTR_ERR(opaquedir);
 779                 if (IS_ERR(opaquedir))
 780                         goto out;
 781         }
 782
 783         err = ovl_lock_rename_workdir(workdir, upperdir);
 784         if (err)
 785                 goto out_dput;
 786
 787         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 788                                  dentry->d_name.len);
 789         err = PTR_ERR(upper);
 790         if (IS_ERR(upper))
 791                 goto out_unlock;
 792
 793         err = -ESTALE;
 794         if ((opaquedir && upper != opaquedir) ||
 795             (!opaquedir && ovl_dentry_upper(dentry) &&
 796              !ovl_matches_upper(dentry, upper))) {
 797                 goto out_dput_upper;
 798         }
 799
 800         err = ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
 801         if (err)
 802                 goto out_d_drop;
 803
 804         ovl_dir_modified(dentry->d_parent, true);
 805 out_d_drop:
 806         d_drop(dentry);
 807 out_dput_upper:
 808         dput(upper);
 809 out_unlock:
 810         unlock_rename(workdir, upperdir);
 811 out_dput:
 812         dput(opaquedir);
 813 out:
 814         return err;
 815 }
 816
 817 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 818                             struct list_head *list)
 819 {
 820         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 821         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 822         struct inode *dir = upperdir->d_inode;
 823         struct dentry *upper;
 824         struct dentry *opaquedir = NULL;
 825         int err;
 826
 827         if (!list_empty(list)) {
 828                 opaquedir = ovl_clear_empty(dentry, list);
 829                 err = PTR_ERR(opaquedir);
 830                 if (IS_ERR(opaquedir))
 831                         goto out;
 832         }
 833
 834         inode_lock_nested(dir, I_MUTEX_PARENT);
 835         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 836                                  dentry->d_name.len);
 837         err = PTR_ERR(upper);
 838         if (IS_ERR(upper))
 839                 goto out_unlock;
 840
 841         err = -ESTALE;
 842         if ((opaquedir && upper != opaquedir) ||
 843             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 844                 goto out_dput_upper;
 845
 846         if (is_dir)
 847                 err = ovl_do_rmdir(ofs, dir, upper);
 848         else
 849                 err = ovl_do_unlink(ofs, dir, upper);
 850         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 851
 852         /*
 853          * Keeping this dentry hashed would mean having to release
 854          * upperpath/lowerpath, which could only be done if we are the
 855          * sole user of this dentry.  Too tricky...  Just unhash for
 856          * now.
 857          */
 858         if (!err)
 859                 d_drop(dentry);
 860 out_dput_upper:
 861         dput(upper);
 862 out_unlock:
 863         inode_unlock(dir);
 864         dput(opaquedir);
 865 out:
 866         return err;
 867 }
 868
 869 static bool ovl_pure_upper(struct dentry *dentry)
 870 {
 871         return !ovl_dentry_lower(dentry) &&
 872                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 873 }
 874
 875 static void ovl_drop_nlink(struct dentry *dentry)
 876 {
 877         struct inode *inode = d_inode(dentry);
 878         struct dentry *alias;
 879
 880         /* Try to find another, hashed alias */
 881         spin_lock(&inode->i_lock);
 882         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 883                 if (alias != dentry && !d_unhashed(alias))
 884                         break;
 885         }
 886         spin_unlock(&inode->i_lock);
 887
 888         /*
 889          * Changes to underlying layers may cause i_nlink to lose sync with
 890          * reality.  In this case prevent the link count from going to zero
 891          * prematurely.
 892          */
 893         if (inode->i_nlink > !!alias)
 894                 drop_nlink(inode);
 895 }
 896
 897 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 898 {
 899         int err;
 900         const struct cred *old_cred;
 901         bool lower_positive = ovl_lower_positive(dentry);
 902         LIST_HEAD(list);
 903
 904         /* No need to clean pure upper removed by vfs_rmdir() */
 905         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 906                 err = ovl_check_empty_dir(dentry, &list);
 907                 if (err)
 908                         goto out;
 909         }
 910
 911         err = ovl_want_write(dentry);
 912         if (err)
 913                 goto out;
 914
 915         err = ovl_copy_up(dentry->d_parent);
 916         if (err)
 917                 goto out_drop_write;
 918
 919         err = ovl_nlink_start(dentry);
 920         if (err)
 921                 goto out_drop_write;
 922
 923         old_cred = ovl_override_creds(dentry->d_sb);
 924         if (!lower_positive)
 925                 err = ovl_remove_upper(dentry, is_dir, &list);
 926         else
 927                 err = ovl_remove_and_whiteout(dentry, &list);
 928         revert_creds(old_cred);
 929         if (!err) {
 930                 if (is_dir)
 931                         clear_nlink(dentry->d_inode);
 932                 else
 933                         ovl_drop_nlink(dentry);
 934         }
 935         ovl_nlink_end(dentry);
 936
 937         /*
 938          * Copy ctime
 939          *
 940          * Note: we fail to update ctime if there was no copy-up, only a
 941          * whiteout
 942          */
 943         if (ovl_dentry_upper(dentry))
 944                 ovl_copyattr(d_inode(dentry));
 945
 946 out_drop_write:
 947         ovl_drop_write(dentry);
 948 out:
 949         ovl_cache_free(&list);
 950         return err;
 951 }
 952
 953 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 954 {
 955         return ovl_do_remove(dentry, false);
 956 }
 957
 958 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 959 {
 960         return ovl_do_remove(dentry, true);
 961 }
 962
 963 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 964 {
 965         enum ovl_path_type type = ovl_path_type(dentry);
 966
 967         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 968 }
 969
 970 static bool ovl_can_move(struct dentry *dentry)
 971 {
 972         return ovl_redirect_dir(dentry->d_sb) ||
 973                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 974 }
 975
 976 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 977 {
 978         char *buf, *ret;
 979         struct dentry *d, *tmp;
 980         int buflen = ovl_redirect_max + 1;
 981
 982         if (!abs_redirect) {
 983                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 984                                GFP_KERNEL);
 985                 goto out;
 986         }
 987
 988         buf = ret = kmalloc(buflen, GFP_KERNEL);
 989         if (!buf)
 990                 goto out;
 991
 992         buflen--;
 993         buf[buflen] = '\0';
 994         for (d = dget(dentry); !IS_ROOT(d);) {
 995                 const char *name;
 996                 int thislen;
 997
 998                 spin_lock(&d->d_lock);
 999                 name = ovl_dentry_get_redirect(d);
1000                 if (name) {
1001                         thislen = strlen(name);
1002                 } else {
1003                         name = d->d_name.name;
1004                         thislen = d->d_name.len;
1005                 }
1006
1007                 /* If path is too long, fall back to userspace move */
1008                 if (thislen + (name[0] != '/') > buflen) {
1009                         ret = ERR_PTR(-EXDEV);
1010                         spin_unlock(&d->d_lock);
1011                         goto out_put;
1012                 }
1013
1014                 buflen -= thislen;
1015                 memcpy(&buf[buflen], name, thislen);
1016                 spin_unlock(&d->d_lock);
1017                 tmp = dget_parent(d);
1018
1019                 dput(d);
1020                 d = tmp;
1021
1022                 /* Absolute redirect: finished */
1023                 if (buf[buflen] == '/')
1024                         break;
1025                 buflen--;
1026                 buf[buflen] = '/';
1027         }
1028         ret = kstrdup(&buf[buflen], GFP_KERNEL);
1029 out_put:
1030         dput(d);
1031         kfree(buf);
1032 out:
1033         return ret ? ret : ERR_PTR(-ENOMEM);
1034 }
1035
1036 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1037 {
1038         struct dentry *lowerdentry;
1039
1040         if (!samedir)
1041                 return true;
1042
1043         if (d_is_dir(dentry))
1044                 return false;
1045
1046         /*
1047          * For non-dir hardlinked files, we need absolute redirects
1048          * in general as two upper hardlinks could be in different
1049          * dirs. We could put a relative redirect now and convert
1050          * it to absolute redirect later. But when nlink > 1 and
1051          * indexing is on, that means relative redirect needs to be
1052          * converted to absolute during copy up of another lower
1053          * hardllink as well.
1054          *
1055          * So without optimizing too much, just check if lower is
1056          * a hard link or not. If lower is hard link, put absolute
1057          * redirect.
1058          */
1059         lowerdentry = ovl_dentry_lower(dentry);
1060         return (d_inode(lowerdentry)->i_nlink > 1);
1061 }
1062
1063 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1064 {
1065         int err;
1066         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1067         const char *redirect = ovl_dentry_get_redirect(dentry);
1068         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1069
1070         if (redirect && (!absolute_redirect || redirect[0] == '/'))
1071                 return 0;
1072
1073         redirect = ovl_get_redirect(dentry, absolute_redirect);
1074         if (IS_ERR(redirect))
1075                 return PTR_ERR(redirect);
1076
1077         err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1078                                  OVL_XATTR_REDIRECT,
1079                                  redirect, strlen(redirect), -EXDEV);
1080         if (!err) {
1081                 spin_lock(&dentry->d_lock);
1082                 ovl_dentry_set_redirect(dentry, redirect);
1083                 spin_unlock(&dentry->d_lock);
1084         } else {
1085                 kfree(redirect);
1086                 pr_warn_ratelimited("failed to set redirect (%i)\n",
1087                                     err);
1088                 /* Fall back to userspace copy-up */
1089                 err = -EXDEV;
1090         }
1091         return err;
1092 }
1093
1094 static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
1095                       struct dentry *old, struct inode *newdir,
1096                       struct dentry *new, unsigned int flags)
1097 {
1098         int err;
1099         struct dentry *old_upperdir;
1100         struct dentry *new_upperdir;
1101         struct dentry *olddentry;
1102         struct dentry *newdentry;
1103         struct dentry *trap;
1104         bool old_opaque;
1105         bool new_opaque;
1106         bool cleanup_whiteout = false;
1107         bool update_nlink = false;
1108         bool overwrite = !(flags & RENAME_EXCHANGE);
1109         bool is_dir = d_is_dir(old);
1110         bool new_is_dir = d_is_dir(new);
1111         bool samedir = olddir == newdir;
1112         struct dentry *opaquedir = NULL;
1113         const struct cred *old_cred = NULL;
1114         struct ovl_fs *ofs = OVL_FS(old->d_sb);
1115         LIST_HEAD(list);
1116
1117         err = -EINVAL;
1118         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1119                 goto out;
1120
1121         flags &= ~RENAME_NOREPLACE;
1122
1123         /* Don't copy up directory trees */
1124         err = -EXDEV;
1125         if (!ovl_can_move(old))
1126                 goto out;
1127         if (!overwrite && !ovl_can_move(new))
1128                 goto out;
1129
1130         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1131                 err = ovl_check_empty_dir(new, &list);
1132                 if (err)
1133                         goto out;
1134         }
1135
1136         if (overwrite) {
1137                 if (ovl_lower_positive(old)) {
1138                         if (!ovl_dentry_is_whiteout(new)) {
1139                                 /* Whiteout source */
1140                                 flags |= RENAME_WHITEOUT;
1141                         } else {
1142                                 /* Switch whiteouts */
1143                                 flags |= RENAME_EXCHANGE;
1144                         }
1145                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1146                         flags |= RENAME_EXCHANGE;
1147                         cleanup_whiteout = true;
1148                 }
1149         }
1150
1151         err = ovl_want_write(old);
1152         if (err)
1153                 goto out;
1154
1155         err = ovl_copy_up(old);
1156         if (err)
1157                 goto out_drop_write;
1158
1159         err = ovl_copy_up(new->d_parent);
1160         if (err)
1161                 goto out_drop_write;
1162         if (!overwrite) {
1163                 err = ovl_copy_up(new);
1164                 if (err)
1165                         goto out_drop_write;
1166         } else if (d_inode(new)) {
1167                 err = ovl_nlink_start(new);
1168                 if (err)
1169                         goto out_drop_write;
1170
1171                 update_nlink = true;
1172         }
1173
1174         old_cred = ovl_override_creds(old->d_sb);
1175
1176         if (!list_empty(&list)) {
1177                 opaquedir = ovl_clear_empty(new, &list);
1178                 err = PTR_ERR(opaquedir);
1179                 if (IS_ERR(opaquedir)) {
1180                         opaquedir = NULL;
1181                         goto out_revert_creds;
1182                 }
1183         }
1184
1185         old_upperdir = ovl_dentry_upper(old->d_parent);
1186         new_upperdir = ovl_dentry_upper(new->d_parent);
1187
1188         if (!samedir) {
1189                 /*
1190                  * When moving a merge dir or non-dir with copy up origin into
1191                  * a new parent, we are marking the new parent dir "impure".
1192                  * When ovl_iterate() iterates an "impure" upper dir, it will
1193                  * lookup the origin inodes of the entries to fill d_ino.
1194                  */
1195                 if (ovl_type_origin(old)) {
1196                         err = ovl_set_impure(new->d_parent, new_upperdir);
1197                         if (err)
1198                                 goto out_revert_creds;
1199                 }
1200                 if (!overwrite && ovl_type_origin(new)) {
1201                         err = ovl_set_impure(old->d_parent, old_upperdir);
1202                         if (err)
1203                                 goto out_revert_creds;
1204                 }
1205         }
1206
1207         trap = lock_rename(new_upperdir, old_upperdir);
1208
1209         olddentry = ovl_lookup_upper(ofs, old->d_name.name, old_upperdir,
1210                                      old->d_name.len);
1211         err = PTR_ERR(olddentry);
1212         if (IS_ERR(olddentry))
1213                 goto out_unlock;
1214
1215         err = -ESTALE;
1216         if (!ovl_matches_upper(old, olddentry))
1217                 goto out_dput_old;
1218
1219         newdentry = ovl_lookup_upper(ofs, new->d_name.name, new_upperdir,
1220                                      new->d_name.len);
1221         err = PTR_ERR(newdentry);
1222         if (IS_ERR(newdentry))
1223                 goto out_dput_old;
1224
1225         old_opaque = ovl_dentry_is_opaque(old);
1226         new_opaque = ovl_dentry_is_opaque(new);
1227
1228         err = -ESTALE;
1229         if (d_inode(new) && ovl_dentry_upper(new)) {
1230                 if (opaquedir) {
1231                         if (newdentry != opaquedir)
1232                                 goto out_dput;
1233                 } else {
1234                         if (!ovl_matches_upper(new, newdentry))
1235                                 goto out_dput;
1236                 }
1237         } else {
1238                 if (!d_is_negative(newdentry)) {
1239                         if (!new_opaque || !ovl_is_whiteout(newdentry))
1240                                 goto out_dput;
1241                 } else {
1242                         if (flags & RENAME_EXCHANGE)
1243                                 goto out_dput;
1244                 }
1245         }
1246
1247         if (olddentry == trap)
1248                 goto out_dput;
1249         if (newdentry == trap)
1250                 goto out_dput;
1251
1252         if (olddentry->d_inode == newdentry->d_inode)
1253                 goto out_dput;
1254
1255         err = 0;
1256         if (ovl_type_merge_or_lower(old))
1257                 err = ovl_set_redirect(old, samedir);
1258         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1259                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1260         if (err)
1261                 goto out_dput;
1262
1263         if (!overwrite && ovl_type_merge_or_lower(new))
1264                 err = ovl_set_redirect(new, samedir);
1265         else if (!overwrite && new_is_dir && !new_opaque &&
1266                  ovl_type_merge(old->d_parent))
1267                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1268         if (err)
1269                 goto out_dput;
1270
1271         err = ovl_do_rename(ofs, old_upperdir->d_inode, olddentry,
1272                             new_upperdir->d_inode, newdentry, flags);
1273         if (err)
1274                 goto out_dput;
1275
1276         if (cleanup_whiteout)
1277                 ovl_cleanup(ofs, old_upperdir->d_inode, newdentry);
1278
1279         if (overwrite && d_inode(new)) {
1280                 if (new_is_dir)
1281                         clear_nlink(d_inode(new));
1282                 else
1283                         ovl_drop_nlink(new);
1284         }
1285
1286         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1287                          (!overwrite && ovl_type_origin(new)));
1288         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1289                          (d_inode(new) && ovl_type_origin(new)));
1290
1291         /* copy ctime: */
1292         ovl_copyattr(d_inode(old));
1293         if (d_inode(new) && ovl_dentry_upper(new))
1294                 ovl_copyattr(d_inode(new));
1295
1296 out_dput:
1297         dput(newdentry);
1298 out_dput_old:
1299         dput(olddentry);
1300 out_unlock:
1301         unlock_rename(new_upperdir, old_upperdir);
1302 out_revert_creds:
1303         revert_creds(old_cred);
1304         if (update_nlink)
1305                 ovl_nlink_end(new);
1306 out_drop_write:
1307         ovl_drop_write(old);
1308 out:
1309         dput(opaquedir);
1310         ovl_cache_free(&list);
1311         return err;
1312 }
1313
1314 const struct inode_operations ovl_dir_inode_operations = {
1315         .lookup         = ovl_lookup,
1316         .mkdir          = ovl_mkdir,
1317         .symlink        = ovl_symlink,
1318         .unlink         = ovl_unlink,
1319         .rmdir          = ovl_rmdir,
1320         .rename         = ovl_rename,
1321         .link           = ovl_link,
1322         .setattr        = ovl_setattr,
1323         .create         = ovl_create,
1324         .mknod          = ovl_mknod,
1325         .permission     = ovl_permission,
1326         .getattr        = ovl_getattr,
1327         .listxattr      = ovl_listxattr,
1328         .get_acl        = ovl_get_acl,
1329         .update_time    = ovl_update_time,
1330         .fileattr_get   = ovl_fileattr_get,
1331         .fileattr_set   = ovl_fileattr_set,
1332 };