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->no_shared_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->no_shared_whiteout = true;
  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_init_reval(dentry, newdentry, NULL);
 273
 274         if (!hardlink) {
 275                 /*
 276                  * ovl_obtain_alias() can be called after ovl_create_real()
 277                  * and before we get here, so we may get an inode from cache
 278                  * with the same real upperdentry that is not the inode we
 279                  * pre-allocated.  In this case we will use the cached inode
 280                  * to instantiate the new dentry.
 281                  *
 282                  * XXX: if we ever use ovl_obtain_alias() to decode directory
 283                  * file handles, need to use ovl_get_inode_locked() and
 284                  * d_instantiate_new() here to prevent from creating two
 285                  * hashed directory inode aliases.
 286                  */
 287                 inode = ovl_get_inode(dentry->d_sb, &oip);
 288                 if (IS_ERR(inode))
 289                         return PTR_ERR(inode);
 290                 if (inode == oip.newinode)
 291                         ovl_set_flag(OVL_UPPERDATA, inode);
 292         } else {
 293                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
 294                 dput(newdentry);
 295                 inc_nlink(inode);
 296         }
 297
 298         d_instantiate(dentry, inode);
 299         if (inode != oip.newinode) {
 300                 pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
 301                                     dentry);
 302         }
 303
 304         /* Force lookup of new upper hardlink to find its lower */
 305         if (hardlink)
 306                 d_drop(dentry);
 307
 308         return 0;
 309 }
 310
 311 static bool ovl_type_merge(struct dentry *dentry)
 312 {
 313         return OVL_TYPE_MERGE(ovl_path_type(dentry));
 314 }
 315
 316 static bool ovl_type_origin(struct dentry *dentry)
 317 {
 318         return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
 319 }
 320
 321 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
 322                             struct ovl_cattr *attr)
 323 {
 324         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 325         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 326         struct inode *udir = upperdir->d_inode;
 327         struct dentry *newdentry;
 328         int err;
 329
 330         if (!attr->hardlink && !IS_POSIXACL(udir))
 331                 attr->mode &= ~current_umask();
 332
 333         inode_lock_nested(udir, I_MUTEX_PARENT);
 334         newdentry = ovl_create_real(ofs, udir,
 335                                     ovl_lookup_upper(ofs, dentry->d_name.name,
 336                                                      upperdir, dentry->d_name.len),
 337                                     attr);
 338         err = PTR_ERR(newdentry);
 339         if (IS_ERR(newdentry))
 340                 goto out_unlock;
 341
 342         if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry) &&
 343             !ovl_allow_offline_changes(ofs)) {
 344                 /* Setting opaque here is just an optimization, allow to fail */
 345                 ovl_set_opaque(dentry, newdentry);
 346         }
 347
 348         err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
 349         if (err)
 350                 goto out_cleanup;
 351 out_unlock:
 352         inode_unlock(udir);
 353         return err;
 354
 355 out_cleanup:
 356         ovl_cleanup(ofs, udir, newdentry);
 357         dput(newdentry);
 358         goto out_unlock;
 359 }
 360
 361 static struct dentry *ovl_clear_empty(struct dentry *dentry,
 362                                       struct list_head *list)
 363 {
 364         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 365         struct dentry *workdir = ovl_workdir(dentry);
 366         struct inode *wdir = workdir->d_inode;
 367         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 368         struct inode *udir = upperdir->d_inode;
 369         struct path upperpath;
 370         struct dentry *upper;
 371         struct dentry *opaquedir;
 372         struct kstat stat;
 373         int err;
 374
 375         if (WARN_ON(!workdir))
 376                 return ERR_PTR(-EROFS);
 377
 378         err = ovl_lock_rename_workdir(workdir, upperdir);
 379         if (err)
 380                 goto out;
 381
 382         ovl_path_upper(dentry, &upperpath);
 383         err = vfs_getattr(&upperpath, &stat,
 384                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 385         if (err)
 386                 goto out_unlock;
 387
 388         err = -ESTALE;
 389         if (!S_ISDIR(stat.mode))
 390                 goto out_unlock;
 391         upper = upperpath.dentry;
 392         if (upper->d_parent->d_inode != udir)
 393                 goto out_unlock;
 394
 395         opaquedir = ovl_create_temp(ofs, workdir, OVL_CATTR(stat.mode));
 396         err = PTR_ERR(opaquedir);
 397         if (IS_ERR(opaquedir))
 398                 goto out_unlock;
 399
 400         err = ovl_copy_xattr(dentry->d_sb, &upperpath, opaquedir);
 401         if (err)
 402                 goto out_cleanup;
 403
 404         err = ovl_set_opaque(dentry, opaquedir);
 405         if (err)
 406                 goto out_cleanup;
 407
 408         inode_lock(opaquedir->d_inode);
 409         err = ovl_set_attr(ofs, opaquedir, &stat);
 410         inode_unlock(opaquedir->d_inode);
 411         if (err)
 412                 goto out_cleanup;
 413
 414         err = ovl_do_rename(ofs, wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
 415         if (err)
 416                 goto out_cleanup;
 417
 418         ovl_cleanup_whiteouts(ofs, upper, list);
 419         ovl_cleanup(ofs, wdir, upper);
 420         unlock_rename(workdir, upperdir);
 421
 422         /* dentry's upper doesn't match now, get rid of it */
 423         d_drop(dentry);
 424
 425         return opaquedir;
 426
 427 out_cleanup:
 428         ovl_cleanup(ofs, wdir, opaquedir);
 429         dput(opaquedir);
 430 out_unlock:
 431         unlock_rename(workdir, upperdir);
 432 out:
 433         return ERR_PTR(err);
 434 }
 435
 436 static int ovl_set_upper_acl(struct ovl_fs *ofs, struct dentry *upperdentry,
 437                              const char *acl_name, struct posix_acl *acl)
 438 {
 439         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
 440                 return 0;
 441
 442         return ovl_do_set_acl(ofs, upperdentry, acl_name, acl);
 443 }
 444
 445 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 446                                     struct ovl_cattr *cattr)
 447 {
 448         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 449         struct dentry *workdir = ovl_workdir(dentry);
 450         struct inode *wdir = workdir->d_inode;
 451         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 452         struct inode *udir = upperdir->d_inode;
 453         struct dentry *upper;
 454         struct dentry *newdentry;
 455         int err;
 456         struct posix_acl *acl, *default_acl;
 457         bool hardlink = !!cattr->hardlink;
 458
 459         if (WARN_ON(!workdir))
 460                 return -EROFS;
 461
 462         if (!hardlink) {
 463                 err = posix_acl_create(dentry->d_parent->d_inode,
 464                                        &cattr->mode, &default_acl, &acl);
 465                 if (err)
 466                         return err;
 467         }
 468
 469         err = ovl_lock_rename_workdir(workdir, upperdir);
 470         if (err)
 471                 goto out;
 472
 473         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 474                                  dentry->d_name.len);
 475         err = PTR_ERR(upper);
 476         if (IS_ERR(upper))
 477                 goto out_unlock;
 478
 479         err = -ESTALE;
 480         if (d_is_negative(upper) || !ovl_upper_is_whiteout(ofs, upper))
 481                 goto out_dput;
 482
 483         newdentry = ovl_create_temp(ofs, workdir, cattr);
 484         err = PTR_ERR(newdentry);
 485         if (IS_ERR(newdentry))
 486                 goto out_dput;
 487
 488         /*
 489          * mode could have been mutilated due to umask (e.g. sgid directory)
 490          */
 491         if (!hardlink &&
 492             !S_ISLNK(cattr->mode) &&
 493             newdentry->d_inode->i_mode != cattr->mode) {
 494                 struct iattr attr = {
 495                         .ia_valid = ATTR_MODE,
 496                         .ia_mode = cattr->mode,
 497                 };
 498                 inode_lock(newdentry->d_inode);
 499                 err = ovl_do_notify_change(ofs, newdentry, &attr);
 500                 inode_unlock(newdentry->d_inode);
 501                 if (err)
 502                         goto out_cleanup;
 503         }
 504         if (!hardlink) {
 505                 err = ovl_set_upper_acl(ofs, newdentry,
 506                                         XATTR_NAME_POSIX_ACL_ACCESS, acl);
 507                 if (err)
 508                         goto out_cleanup;
 509
 510                 err = ovl_set_upper_acl(ofs, newdentry,
 511                                         XATTR_NAME_POSIX_ACL_DEFAULT, default_acl);
 512                 if (err)
 513                         goto out_cleanup;
 514         }
 515
 516         if (!hardlink && S_ISDIR(cattr->mode)) {
 517                 err = ovl_set_opaque(dentry, newdentry);
 518                 if (err)
 519                         goto out_cleanup;
 520
 521                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper,
 522                                     RENAME_EXCHANGE);
 523                 if (err)
 524                         goto out_cleanup;
 525
 526                 ovl_cleanup(ofs, wdir, upper);
 527         } else {
 528                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper, 0);
 529                 if (err)
 530                         goto out_cleanup;
 531         }
 532         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 533         if (err) {
 534                 ovl_cleanup(ofs, udir, newdentry);
 535                 dput(newdentry);
 536         }
 537 out_dput:
 538         dput(upper);
 539 out_unlock:
 540         unlock_rename(workdir, upperdir);
 541 out:
 542         if (!hardlink) {
 543                 posix_acl_release(acl);
 544                 posix_acl_release(default_acl);
 545         }
 546         return err;
 547
 548 out_cleanup:
 549         ovl_cleanup(ofs, wdir, newdentry);
 550         dput(newdentry);
 551         goto out_dput;
 552 }
 553
 554 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 555                               struct ovl_cattr *attr, bool origin)
 556 {
 557         int err;
 558         const struct cred *old_cred;
 559         struct cred *override_cred;
 560         struct dentry *parent = dentry->d_parent;
 561
 562         old_cred = ovl_override_creds(dentry->d_sb);
 563
 564         /*
 565          * When linking a file with copy up origin into a new parent, mark the
 566          * new parent dir "impure".
 567          */
 568         if (origin) {
 569                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 570                 if (err)
 571                         goto out_revert_creds;
 572         }
 573
 574         if (!attr->hardlink) {
 575                 err = -ENOMEM;
 576                 override_cred = prepare_creds();
 577                 if (!override_cred)
 578                         goto out_revert_creds;
 579                 /*
 580                  * In the creation cases(create, mkdir, mknod, symlink),
 581                  * ovl should transfer current's fs{u,g}id to underlying
 582                  * fs. Because underlying fs want to initialize its new
 583                  * inode owner using current's fs{u,g}id. And in this
 584                  * case, the @inode is a new inode that is initialized
 585                  * in inode_init_owner() to current's fs{u,g}id. So use
 586                  * the inode's i_{u,g}id to override the cred's fs{u,g}id.
 587                  *
 588                  * But in the other hardlink case, ovl_link() does not
 589                  * create a new inode, so just use the ovl mounter's
 590                  * fs{u,g}id.
 591                  */
 592                 override_cred->fsuid = inode->i_uid;
 593                 override_cred->fsgid = inode->i_gid;
 594                 err = security_dentry_create_files_as(dentry,
 595                                 attr->mode, &dentry->d_name, old_cred,
 596                                 override_cred);
 597                 if (err) {
 598                         put_cred(override_cred);
 599                         goto out_revert_creds;
 600                 }
 601                 put_cred(override_creds(override_cred));
 602                 put_cred(override_cred);
 603         }
 604
 605         if (!ovl_dentry_is_whiteout(dentry))
 606                 err = ovl_create_upper(dentry, inode, attr);
 607         else
 608                 err = ovl_create_over_whiteout(dentry, inode, attr);
 609
 610 out_revert_creds:
 611         revert_creds(old_cred);
 612         return err;
 613 }
 614
 615 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 616                              const char *link)
 617 {
 618         int err;
 619         struct inode *inode;
 620         struct ovl_cattr attr = {
 621                 .rdev = rdev,
 622                 .link = link,
 623         };
 624
 625         err = ovl_copy_up(dentry->d_parent);
 626         if (err)
 627                 return err;
 628
 629         err = ovl_want_write(dentry);
 630         if (err)
 631                 goto out;
 632
 633         /* Preallocate inode to be used by ovl_get_inode() */
 634         err = -ENOMEM;
 635         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 636         if (!inode)
 637                 goto out_drop_write;
 638
 639         spin_lock(&inode->i_lock);
 640         inode->i_state |= I_CREATING;
 641         spin_unlock(&inode->i_lock);
 642
 643         inode_init_owner(&nop_mnt_idmap, inode, dentry->d_parent->d_inode, mode);
 644         attr.mode = inode->i_mode;
 645
 646         err = ovl_create_or_link(dentry, inode, &attr, false);
 647         /* Did we end up using the preallocated inode? */
 648         if (inode != d_inode(dentry))
 649                 iput(inode);
 650
 651 out_drop_write:
 652         ovl_drop_write(dentry);
 653 out:
 654         return err;
 655 }
 656
 657 static int ovl_create(struct mnt_idmap *idmap, struct inode *dir,
 658                       struct dentry *dentry, umode_t mode, bool excl)
 659 {
 660         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 661 }
 662
 663 static int ovl_mkdir(struct mnt_idmap *idmap, struct inode *dir,
 664                      struct dentry *dentry, umode_t mode)
 665 {
 666         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 667 }
 668
 669 static int ovl_mknod(struct mnt_idmap *idmap, struct inode *dir,
 670                      struct dentry *dentry, umode_t mode, dev_t rdev)
 671 {
 672         /* Don't allow creation of "whiteout" on overlay */
 673         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 674                 return -EPERM;
 675
 676         return ovl_create_object(dentry, mode, rdev, NULL);
 677 }
 678
 679 static int ovl_symlink(struct mnt_idmap *idmap, struct inode *dir,
 680                        struct dentry *dentry, const char *link)
 681 {
 682         return ovl_create_object(dentry, S_IFLNK, 0, link);
 683 }
 684
 685 static int ovl_set_link_redirect(struct dentry *dentry)
 686 {
 687         const struct cred *old_cred;
 688         int err;
 689
 690         old_cred = ovl_override_creds(dentry->d_sb);
 691         err = ovl_set_redirect(dentry, false);
 692         revert_creds(old_cred);
 693
 694         return err;
 695 }
 696
 697 static int ovl_link(struct dentry *old, struct inode *newdir,
 698                     struct dentry *new)
 699 {
 700         int err;
 701         struct inode *inode;
 702
 703         err = ovl_copy_up(old);
 704         if (err)
 705                 goto out;
 706
 707         err = ovl_copy_up(new->d_parent);
 708         if (err)
 709                 goto out;
 710
 711         err = ovl_nlink_start(old);
 712         if (err)
 713                 goto out;
 714
 715         if (ovl_is_metacopy_dentry(old)) {
 716                 err = ovl_set_link_redirect(old);
 717                 if (err)
 718                         goto out_nlink_end;
 719         }
 720
 721         inode = d_inode(old);
 722         ihold(inode);
 723
 724         err = ovl_create_or_link(new, inode,
 725                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 726                         ovl_type_origin(old));
 727         if (err)
 728                 iput(inode);
 729
 730 out_nlink_end:
 731         ovl_nlink_end(old);
 732 out:
 733         return err;
 734 }
 735
 736 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 737 {
 738         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 739 }
 740
 741 static int ovl_remove_and_whiteout(struct dentry *dentry,
 742                                    struct list_head *list)
 743 {
 744         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 745         struct dentry *workdir = ovl_workdir(dentry);
 746         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 747         struct dentry *upper;
 748         struct dentry *opaquedir = NULL;
 749         int err;
 750
 751         if (WARN_ON(!workdir))
 752                 return -EROFS;
 753
 754         if (!list_empty(list)) {
 755                 opaquedir = ovl_clear_empty(dentry, list);
 756                 err = PTR_ERR(opaquedir);
 757                 if (IS_ERR(opaquedir))
 758                         goto out;
 759         }
 760
 761         err = ovl_lock_rename_workdir(workdir, upperdir);
 762         if (err)
 763                 goto out_dput;
 764
 765         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 766                                  dentry->d_name.len);
 767         err = PTR_ERR(upper);
 768         if (IS_ERR(upper))
 769                 goto out_unlock;
 770
 771         err = -ESTALE;
 772         if ((opaquedir && upper != opaquedir) ||
 773             (!opaquedir && ovl_dentry_upper(dentry) &&
 774              !ovl_matches_upper(dentry, upper))) {
 775                 goto out_dput_upper;
 776         }
 777
 778         err = ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
 779         if (err)
 780                 goto out_d_drop;
 781
 782         ovl_dir_modified(dentry->d_parent, true);
 783 out_d_drop:
 784         d_drop(dentry);
 785 out_dput_upper:
 786         dput(upper);
 787 out_unlock:
 788         unlock_rename(workdir, upperdir);
 789 out_dput:
 790         dput(opaquedir);
 791 out:
 792         return err;
 793 }
 794
 795 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 796                             struct list_head *list)
 797 {
 798         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 799         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 800         struct inode *dir = upperdir->d_inode;
 801         struct dentry *upper;
 802         struct dentry *opaquedir = NULL;
 803         int err;
 804
 805         if (!list_empty(list)) {
 806                 opaquedir = ovl_clear_empty(dentry, list);
 807                 err = PTR_ERR(opaquedir);
 808                 if (IS_ERR(opaquedir))
 809                         goto out;
 810         }
 811
 812         inode_lock_nested(dir, I_MUTEX_PARENT);
 813         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 814                                  dentry->d_name.len);
 815         err = PTR_ERR(upper);
 816         if (IS_ERR(upper))
 817                 goto out_unlock;
 818
 819         err = -ESTALE;
 820         if ((opaquedir && upper != opaquedir) ||
 821             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 822                 goto out_dput_upper;
 823
 824         if (is_dir)
 825                 err = ovl_do_rmdir(ofs, dir, upper);
 826         else
 827                 err = ovl_do_unlink(ofs, dir, upper);
 828         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 829
 830         /*
 831          * Keeping this dentry hashed would mean having to release
 832          * upperpath/lowerpath, which could only be done if we are the
 833          * sole user of this dentry.  Too tricky...  Just unhash for
 834          * now.
 835          */
 836         if (!err)
 837                 d_drop(dentry);
 838 out_dput_upper:
 839         dput(upper);
 840 out_unlock:
 841         inode_unlock(dir);
 842         dput(opaquedir);
 843 out:
 844         return err;
 845 }
 846
 847 static bool ovl_pure_upper(struct dentry *dentry)
 848 {
 849         return !ovl_dentry_lower(dentry) &&
 850                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 851 }
 852
 853 static void ovl_drop_nlink(struct dentry *dentry)
 854 {
 855         struct inode *inode = d_inode(dentry);
 856         struct dentry *alias;
 857
 858         /* Try to find another, hashed alias */
 859         spin_lock(&inode->i_lock);
 860         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 861                 if (alias != dentry && !d_unhashed(alias))
 862                         break;
 863         }
 864         spin_unlock(&inode->i_lock);
 865
 866         /*
 867          * Changes to underlying layers may cause i_nlink to lose sync with
 868          * reality.  In this case prevent the link count from going to zero
 869          * prematurely.
 870          */
 871         if (inode->i_nlink > !!alias)
 872                 drop_nlink(inode);
 873 }
 874
 875 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 876 {
 877         int err;
 878         const struct cred *old_cred;
 879         bool lower_positive = ovl_lower_positive(dentry);
 880         LIST_HEAD(list);
 881
 882         /* No need to clean pure upper removed by vfs_rmdir() */
 883         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 884                 err = ovl_check_empty_dir(dentry, &list);
 885                 if (err)
 886                         goto out;
 887         }
 888
 889         err = ovl_copy_up(dentry->d_parent);
 890         if (err)
 891                 goto out;
 892
 893         err = ovl_nlink_start(dentry);
 894         if (err)
 895                 goto out;
 896
 897         old_cred = ovl_override_creds(dentry->d_sb);
 898         if (!lower_positive)
 899                 err = ovl_remove_upper(dentry, is_dir, &list);
 900         else
 901                 err = ovl_remove_and_whiteout(dentry, &list);
 902         revert_creds(old_cred);
 903         if (!err) {
 904                 if (is_dir)
 905                         clear_nlink(dentry->d_inode);
 906                 else
 907                         ovl_drop_nlink(dentry);
 908         }
 909         ovl_nlink_end(dentry);
 910
 911         /*
 912          * Copy ctime
 913          *
 914          * Note: we fail to update ctime if there was no copy-up, only a
 915          * whiteout
 916          */
 917         if (ovl_dentry_upper(dentry))
 918                 ovl_copyattr(d_inode(dentry));
 919
 920 out:
 921         ovl_cache_free(&list);
 922         return err;
 923 }
 924
 925 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 926 {
 927         return ovl_do_remove(dentry, false);
 928 }
 929
 930 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 931 {
 932         return ovl_do_remove(dentry, true);
 933 }
 934
 935 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 936 {
 937         enum ovl_path_type type = ovl_path_type(dentry);
 938
 939         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 940 }
 941
 942 static bool ovl_can_move(struct dentry *dentry)
 943 {
 944         return ovl_redirect_dir(OVL_FS(dentry->d_sb)) ||
 945                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 946 }
 947
 948 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 949 {
 950         char *buf, *ret;
 951         struct dentry *d, *tmp;
 952         int buflen = ovl_redirect_max + 1;
 953
 954         if (!abs_redirect) {
 955                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 956                                GFP_KERNEL);
 957                 goto out;
 958         }
 959
 960         buf = ret = kmalloc(buflen, GFP_KERNEL);
 961         if (!buf)
 962                 goto out;
 963
 964         buflen--;
 965         buf[buflen] = '\0';
 966         for (d = dget(dentry); !IS_ROOT(d);) {
 967                 const char *name;
 968                 int thislen;
 969
 970                 spin_lock(&d->d_lock);
 971                 name = ovl_dentry_get_redirect(d);
 972                 if (name) {
 973                         thislen = strlen(name);
 974                 } else {
 975                         name = d->d_name.name;
 976                         thislen = d->d_name.len;
 977                 }
 978
 979                 /* If path is too long, fall back to userspace move */
 980                 if (thislen + (name[0] != '/') > buflen) {
 981                         ret = ERR_PTR(-EXDEV);
 982                         spin_unlock(&d->d_lock);
 983                         goto out_put;
 984                 }
 985
 986                 buflen -= thislen;
 987                 memcpy(&buf[buflen], name, thislen);
 988                 spin_unlock(&d->d_lock);
 989                 tmp = dget_parent(d);
 990
 991                 dput(d);
 992                 d = tmp;
 993
 994                 /* Absolute redirect: finished */
 995                 if (buf[buflen] == '/')
 996                         break;
 997                 buflen--;
 998                 buf[buflen] = '/';
 999         }
1000         ret = kstrdup(&buf[buflen], GFP_KERNEL);
1001 out_put:
1002         dput(d);
1003         kfree(buf);
1004 out:
1005         return ret ? ret : ERR_PTR(-ENOMEM);
1006 }
1007
1008 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1009 {
1010         struct dentry *lowerdentry;
1011
1012         if (!samedir)
1013                 return true;
1014
1015         if (d_is_dir(dentry))
1016                 return false;
1017
1018         /*
1019          * For non-dir hardlinked files, we need absolute redirects
1020          * in general as two upper hardlinks could be in different
1021          * dirs. We could put a relative redirect now and convert
1022          * it to absolute redirect later. But when nlink > 1 and
1023          * indexing is on, that means relative redirect needs to be
1024          * converted to absolute during copy up of another lower
1025          * hardllink as well.
1026          *
1027          * So without optimizing too much, just check if lower is
1028          * a hard link or not. If lower is hard link, put absolute
1029          * redirect.
1030          */
1031         lowerdentry = ovl_dentry_lower(dentry);
1032         return (d_inode(lowerdentry)->i_nlink > 1);
1033 }
1034
1035 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1036 {
1037         int err;
1038         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1039         const char *redirect = ovl_dentry_get_redirect(dentry);
1040         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1041
1042         if (redirect && (!absolute_redirect || redirect[0] == '/'))
1043                 return 0;
1044
1045         redirect = ovl_get_redirect(dentry, absolute_redirect);
1046         if (IS_ERR(redirect))
1047                 return PTR_ERR(redirect);
1048
1049         err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1050                                  OVL_XATTR_REDIRECT,
1051                                  redirect, strlen(redirect), -EXDEV);
1052         if (!err) {
1053                 spin_lock(&dentry->d_lock);
1054                 ovl_dentry_set_redirect(dentry, redirect);
1055                 spin_unlock(&dentry->d_lock);
1056         } else {
1057                 kfree(redirect);
1058                 pr_warn_ratelimited("failed to set redirect (%i)\n",
1059                                     err);
1060                 /* Fall back to userspace copy-up */
1061                 err = -EXDEV;
1062         }
1063         return err;
1064 }
1065
1066 static int ovl_rename(struct mnt_idmap *idmap, struct inode *olddir,
1067                       struct dentry *old, struct inode *newdir,
1068                       struct dentry *new, unsigned int flags)
1069 {
1070         int err;
1071         struct dentry *old_upperdir;
1072         struct dentry *new_upperdir;
1073         struct dentry *olddentry;
1074         struct dentry *newdentry;
1075         struct dentry *trap;
1076         bool old_opaque;
1077         bool new_opaque;
1078         bool cleanup_whiteout = false;
1079         bool update_nlink = false;
1080         bool overwrite = !(flags & RENAME_EXCHANGE);
1081         bool is_dir = d_is_dir(old);
1082         bool new_is_dir = d_is_dir(new);
1083         bool samedir = olddir == newdir;
1084         struct dentry *opaquedir = NULL;
1085         const struct cred *old_cred = NULL;
1086         struct ovl_fs *ofs = OVL_FS(old->d_sb);
1087         LIST_HEAD(list);
1088
1089         err = -EINVAL;
1090         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1091                 goto out;
1092
1093         flags &= ~RENAME_NOREPLACE;
1094
1095         /* Don't copy up directory trees */
1096         err = -EXDEV;
1097         if (!ovl_can_move(old))
1098                 goto out;
1099         if (!overwrite && !ovl_can_move(new))
1100                 goto out;
1101
1102         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1103                 err = ovl_check_empty_dir(new, &list);
1104                 if (err)
1105                         goto out;
1106         }
1107
1108         if (overwrite) {
1109                 if (ovl_lower_positive(old)) {
1110                         if (!ovl_dentry_is_whiteout(new)) {
1111                                 /* Whiteout source */
1112                                 flags |= RENAME_WHITEOUT;
1113                         } else {
1114                                 /* Switch whiteouts */
1115                                 flags |= RENAME_EXCHANGE;
1116                         }
1117                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1118                         flags |= RENAME_EXCHANGE;
1119                         cleanup_whiteout = true;
1120                 }
1121         }
1122
1123         err = ovl_copy_up(old);
1124         if (err)
1125                 goto out;
1126
1127         err = ovl_copy_up(new->d_parent);
1128         if (err)
1129                 goto out;
1130         if (!overwrite) {
1131                 err = ovl_copy_up(new);
1132                 if (err)
1133                         goto out;
1134         } else if (d_inode(new)) {
1135                 err = ovl_nlink_start(new);
1136                 if (err)
1137                         goto out;
1138
1139                 update_nlink = true;
1140         }
1141
1142         if (!update_nlink) {
1143                 /* ovl_nlink_start() took ovl_want_write() */
1144                 err = ovl_want_write(old);
1145                 if (err)
1146                         goto out;
1147         }
1148
1149         old_cred = ovl_override_creds(old->d_sb);
1150
1151         if (!list_empty(&list)) {
1152                 opaquedir = ovl_clear_empty(new, &list);
1153                 err = PTR_ERR(opaquedir);
1154                 if (IS_ERR(opaquedir)) {
1155                         opaquedir = NULL;
1156                         goto out_revert_creds;
1157                 }
1158         }
1159
1160         old_upperdir = ovl_dentry_upper(old->d_parent);
1161         new_upperdir = ovl_dentry_upper(new->d_parent);
1162
1163         if (!samedir) {
1164                 /*
1165                  * When moving a merge dir or non-dir with copy up origin into
1166                  * a new parent, we are marking the new parent dir "impure".
1167                  * When ovl_iterate() iterates an "impure" upper dir, it will
1168                  * lookup the origin inodes of the entries to fill d_ino.
1169                  */
1170                 if (ovl_type_origin(old)) {
1171                         err = ovl_set_impure(new->d_parent, new_upperdir);
1172                         if (err)
1173                                 goto out_revert_creds;
1174                 }
1175                 if (!overwrite && ovl_type_origin(new)) {
1176                         err = ovl_set_impure(old->d_parent, old_upperdir);
1177                         if (err)
1178                                 goto out_revert_creds;
1179                 }
1180         }
1181
1182         trap = lock_rename(new_upperdir, old_upperdir);
1183
1184         olddentry = ovl_lookup_upper(ofs, old->d_name.name, old_upperdir,
1185                                      old->d_name.len);
1186         err = PTR_ERR(olddentry);
1187         if (IS_ERR(olddentry))
1188                 goto out_unlock;
1189
1190         err = -ESTALE;
1191         if (!ovl_matches_upper(old, olddentry))
1192                 goto out_dput_old;
1193
1194         newdentry = ovl_lookup_upper(ofs, new->d_name.name, new_upperdir,
1195                                      new->d_name.len);
1196         err = PTR_ERR(newdentry);
1197         if (IS_ERR(newdentry))
1198                 goto out_dput_old;
1199
1200         old_opaque = ovl_dentry_is_opaque(old);
1201         new_opaque = ovl_dentry_is_opaque(new);
1202
1203         err = -ESTALE;
1204         if (d_inode(new) && ovl_dentry_upper(new)) {
1205                 if (opaquedir) {
1206                         if (newdentry != opaquedir)
1207                                 goto out_dput;
1208                 } else {
1209                         if (!ovl_matches_upper(new, newdentry))
1210                                 goto out_dput;
1211                 }
1212         } else {
1213                 if (!d_is_negative(newdentry)) {
1214                         if (!new_opaque || !ovl_upper_is_whiteout(ofs, newdentry))
1215                                 goto out_dput;
1216                 } else {
1217                         if (flags & RENAME_EXCHANGE)
1218                                 goto out_dput;
1219                 }
1220         }
1221
1222         if (olddentry == trap)
1223                 goto out_dput;
1224         if (newdentry == trap)
1225                 goto out_dput;
1226
1227         if (olddentry->d_inode == newdentry->d_inode)
1228                 goto out_dput;
1229
1230         err = 0;
1231         if (ovl_type_merge_or_lower(old))
1232                 err = ovl_set_redirect(old, samedir);
1233         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1234                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1235         if (err)
1236                 goto out_dput;
1237
1238         if (!overwrite && ovl_type_merge_or_lower(new))
1239                 err = ovl_set_redirect(new, samedir);
1240         else if (!overwrite && new_is_dir && !new_opaque &&
1241                  ovl_type_merge(old->d_parent))
1242                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1243         if (err)
1244                 goto out_dput;
1245
1246         err = ovl_do_rename(ofs, old_upperdir->d_inode, olddentry,
1247                             new_upperdir->d_inode, newdentry, flags);
1248         if (err)
1249                 goto out_dput;
1250
1251         if (cleanup_whiteout)
1252                 ovl_cleanup(ofs, old_upperdir->d_inode, newdentry);
1253
1254         if (overwrite && d_inode(new)) {
1255                 if (new_is_dir)
1256                         clear_nlink(d_inode(new));
1257                 else
1258                         ovl_drop_nlink(new);
1259         }
1260
1261         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1262                          (!overwrite && ovl_type_origin(new)));
1263         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1264                          (d_inode(new) && ovl_type_origin(new)));
1265
1266         /* copy ctime: */
1267         ovl_copyattr(d_inode(old));
1268         if (d_inode(new) && ovl_dentry_upper(new))
1269                 ovl_copyattr(d_inode(new));
1270
1271 out_dput:
1272         dput(newdentry);
1273 out_dput_old:
1274         dput(olddentry);
1275 out_unlock:
1276         unlock_rename(new_upperdir, old_upperdir);
1277 out_revert_creds:
1278         revert_creds(old_cred);
1279         if (update_nlink)
1280                 ovl_nlink_end(new);
1281         else
1282                 ovl_drop_write(old);
1283 out:
1284         dput(opaquedir);
1285         ovl_cache_free(&list);
1286         return err;
1287 }
1288
1289 const struct inode_operations ovl_dir_inode_operations = {
1290         .lookup         = ovl_lookup,
1291         .mkdir          = ovl_mkdir,
1292         .symlink        = ovl_symlink,
1293         .unlink         = ovl_unlink,
1294         .rmdir          = ovl_rmdir,
1295         .rename         = ovl_rename,
1296         .link           = ovl_link,
1297         .setattr        = ovl_setattr,
1298         .create         = ovl_create,
1299         .mknod          = ovl_mknod,
1300         .permission     = ovl_permission,
1301         .getattr        = ovl_getattr,
1302         .listxattr      = ovl_listxattr,
1303         .get_inode_acl  = ovl_get_inode_acl,
1304         .get_acl        = ovl_get_acl,
1305         .set_acl        = ovl_set_acl,
1306         .update_time    = ovl_update_time,
1307         .fileattr_get   = ovl_fileattr_get,
1308         .fileattr_set   = ovl_fileattr_set,
1309 };