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