fs/overlayfs/copy_up.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/fs.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/splice.h>
  15 #include <linux/xattr.h>
  16 #include <linux/security.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/cred.h>
  20 #include <linux/namei.h>
  21 #include <linux/fdtable.h>
  22 #include <linux/ratelimit.h>
  23 #include <linux/exportfs.h>
  24 #include "overlayfs.h"
  25
  26 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  27
  28 static int ovl_ccup_set(const char *buf, const struct kernel_param *param)
  29 {
  30         pr_warn("overlayfs: \"check_copy_up\" module option is obsolete\n");
  31         return 0;
  32 }
  33
  34 static int ovl_ccup_get(char *buf, const struct kernel_param *param)
  35 {
  36         return sprintf(buf, "N\n");
  37 }
  38
  39 module_param_call(check_copy_up, ovl_ccup_set, ovl_ccup_get, NULL, 0644);
  40 MODULE_PARM_DESC(ovl_check_copy_up, "Obsolete; does nothing");
  41
  42 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  43 {
  44         ssize_t list_size, size, value_size = 0;
  45         char *buf, *name, *value = NULL;
  46         int uninitialized_var(error);
  47         size_t slen;
  48
  49         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  50             !(new->d_inode->i_opflags & IOP_XATTR))
  51                 return 0;
  52
  53         list_size = vfs_listxattr(old, NULL, 0);
  54         if (list_size <= 0) {
  55                 if (list_size == -EOPNOTSUPP)
  56                         return 0;
  57                 return list_size;
  58         }
  59
  60         buf = kzalloc(list_size, GFP_KERNEL);
  61         if (!buf)
  62                 return -ENOMEM;
  63
  64         list_size = vfs_listxattr(old, buf, list_size);
  65         if (list_size <= 0) {
  66                 error = list_size;
  67                 goto out;
  68         }
  69
  70         for (name = buf; list_size; name += slen) {
  71                 slen = strnlen(name, list_size) + 1;
  72
  73                 /* underlying fs providing us with an broken xattr list? */
  74                 if (WARN_ON(slen > list_size)) {
  75                         error = -EIO;
  76                         break;
  77                 }
  78                 list_size -= slen;
  79
  80                 if (ovl_is_private_xattr(name))
  81                         continue;
  82 retry:
  83                 size = vfs_getxattr(old, name, value, value_size);
  84                 if (size == -ERANGE)
  85                         size = vfs_getxattr(old, name, NULL, 0);
  86
  87                 if (size < 0) {
  88                         error = size;
  89                         break;
  90                 }
  91
  92                 if (size > value_size) {
  93                         void *new;
  94
  95                         new = krealloc(value, size, GFP_KERNEL);
  96                         if (!new) {
  97                                 error = -ENOMEM;
  98                                 break;
  99                         }
 100                         value = new;
 101                         value_size = size;
 102                         goto retry;
 103                 }
 104
 105                 error = security_inode_copy_up_xattr(name);
 106                 if (error < 0 && error != -EOPNOTSUPP)
 107                         break;
 108                 if (error == 1) {
 109                         error = 0;
 110                         continue; /* Discard */
 111                 }
 112                 error = vfs_setxattr(new, name, value, size, 0);
 113                 if (error)
 114                         break;
 115         }
 116         kfree(value);
 117 out:
 118         kfree(buf);
 119         return error;
 120 }
 121
 122 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 123 {
 124         struct file *old_file;
 125         struct file *new_file;
 126         loff_t old_pos = 0;
 127         loff_t new_pos = 0;
 128         loff_t cloned;
 129         int error = 0;
 130
 131         if (len == 0)
 132                 return 0;
 133
 134         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 135         if (IS_ERR(old_file))
 136                 return PTR_ERR(old_file);
 137
 138         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 139         if (IS_ERR(new_file)) {
 140                 error = PTR_ERR(new_file);
 141                 goto out_fput;
 142         }
 143
 144         /* Try to use clone_file_range to clone up within the same fs */
 145         cloned = do_clone_file_range(old_file, 0, new_file, 0, len, 0);
 146         if (cloned == len)
 147                 goto out;
 148         /* Couldn't clone, so now we try to copy the data */
 149
 150         /* FIXME: copy up sparse files efficiently */
 151         while (len) {
 152                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 153                 long bytes;
 154
 155                 if (len < this_len)
 156                         this_len = len;
 157
 158                 if (signal_pending_state(TASK_KILLABLE, current)) {
 159                         error = -EINTR;
 160                         break;
 161                 }
 162
 163                 bytes = do_splice_direct(old_file, &old_pos,
 164                                          new_file, &new_pos,
 165                                          this_len, SPLICE_F_MOVE);
 166                 if (bytes <= 0) {
 167                         error = bytes;
 168                         break;
 169                 }
 170                 WARN_ON(old_pos != new_pos);
 171
 172                 len -= bytes;
 173         }
 174 out:
 175         if (!error)
 176                 error = vfs_fsync(new_file, 0);
 177         fput(new_file);
 178 out_fput:
 179         fput(old_file);
 180         return error;
 181 }
 182
 183 static int ovl_set_size(struct dentry *upperdentry, struct kstat *stat)
 184 {
 185         struct iattr attr = {
 186                 .ia_valid = ATTR_SIZE,
 187                 .ia_size = stat->size,
 188         };
 189
 190         return notify_change(upperdentry, &attr, NULL);
 191 }
 192
 193 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 194 {
 195         struct iattr attr = {
 196                 .ia_valid =
 197                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 198                 .ia_atime = stat->atime,
 199                 .ia_mtime = stat->mtime,
 200         };
 201
 202         return notify_change(upperdentry, &attr, NULL);
 203 }
 204
 205 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 206 {
 207         int err = 0;
 208
 209         if (!S_ISLNK(stat->mode)) {
 210                 struct iattr attr = {
 211                         .ia_valid = ATTR_MODE,
 212                         .ia_mode = stat->mode,
 213                 };
 214                 err = notify_change(upperdentry, &attr, NULL);
 215         }
 216         if (!err) {
 217                 struct iattr attr = {
 218                         .ia_valid = ATTR_UID | ATTR_GID,
 219                         .ia_uid = stat->uid,
 220                         .ia_gid = stat->gid,
 221                 };
 222                 err = notify_change(upperdentry, &attr, NULL);
 223         }
 224         if (!err)
 225                 ovl_set_timestamps(upperdentry, stat);
 226
 227         return err;
 228 }
 229
 230 struct ovl_fh *ovl_encode_real_fh(struct dentry *real, bool is_upper)
 231 {
 232         struct ovl_fh *fh;
 233         int fh_type, fh_len, dwords;
 234         void *buf;
 235         int buflen = MAX_HANDLE_SZ;
 236         uuid_t *uuid = &real->d_sb->s_uuid;
 237
 238         buf = kmalloc(buflen, GFP_KERNEL);
 239         if (!buf)
 240                 return ERR_PTR(-ENOMEM);
 241
 242         /*
 243          * We encode a non-connectable file handle for non-dir, because we
 244          * only need to find the lower inode number and we don't want to pay
 245          * the price or reconnecting the dentry.
 246          */
 247         dwords = buflen >> 2;
 248         fh_type = exportfs_encode_fh(real, buf, &dwords, 0);
 249         buflen = (dwords << 2);
 250
 251         fh = ERR_PTR(-EIO);
 252         if (WARN_ON(fh_type < 0) ||
 253             WARN_ON(buflen > MAX_HANDLE_SZ) ||
 254             WARN_ON(fh_type == FILEID_INVALID))
 255                 goto out;
 256
 257         BUILD_BUG_ON(MAX_HANDLE_SZ + offsetof(struct ovl_fh, fid) > 255);
 258         fh_len = offsetof(struct ovl_fh, fid) + buflen;
 259         fh = kmalloc(fh_len, GFP_KERNEL);
 260         if (!fh) {
 261                 fh = ERR_PTR(-ENOMEM);
 262                 goto out;
 263         }
 264
 265         fh->version = OVL_FH_VERSION;
 266         fh->magic = OVL_FH_MAGIC;
 267         fh->type = fh_type;
 268         fh->flags = OVL_FH_FLAG_CPU_ENDIAN;
 269         /*
 270          * When we will want to decode an overlay dentry from this handle
 271          * and all layers are on the same fs, if we get a disconncted real
 272          * dentry when we decode fid, the only way to tell if we should assign
 273          * it to upperdentry or to lowerstack is by checking this flag.
 274          */
 275         if (is_upper)
 276                 fh->flags |= OVL_FH_FLAG_PATH_UPPER;
 277         fh->len = fh_len;
 278         fh->uuid = *uuid;
 279         memcpy(fh->fid, buf, buflen);
 280
 281 out:
 282         kfree(buf);
 283         return fh;
 284 }
 285
 286 int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
 287                    struct dentry *upper)
 288 {
 289         const struct ovl_fh *fh = NULL;
 290         int err;
 291
 292         /*
 293          * When lower layer doesn't support export operations store a 'null' fh,
 294          * so we can use the overlay.origin xattr to distignuish between a copy
 295          * up and a pure upper inode.
 296          */
 297         if (ovl_can_decode_fh(lower->d_sb)) {
 298                 fh = ovl_encode_real_fh(lower, false);
 299                 if (IS_ERR(fh))
 300                         return PTR_ERR(fh);
 301         }
 302
 303         /*
 304          * Do not fail when upper doesn't support xattrs.
 305          */
 306         err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh,
 307                                  fh ? fh->len : 0, 0);
 308         kfree(fh);
 309
 310         return err;
 311 }
 312
 313 /* Store file handle of @upper dir in @index dir entry */
 314 static int ovl_set_upper_fh(struct dentry *upper, struct dentry *index)
 315 {
 316         const struct ovl_fh *fh;
 317         int err;
 318
 319         fh = ovl_encode_real_fh(upper, true);
 320         if (IS_ERR(fh))
 321                 return PTR_ERR(fh);
 322
 323         err = ovl_do_setxattr(index, OVL_XATTR_UPPER, fh, fh->len, 0);
 324
 325         kfree(fh);
 326         return err;
 327 }
 328
 329 /*
 330  * Create and install index entry.
 331  *
 332  * Caller must hold i_mutex on indexdir.
 333  */
 334 static int ovl_create_index(struct dentry *dentry, struct dentry *origin,
 335                             struct dentry *upper)
 336 {
 337         struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
 338         struct inode *dir = d_inode(indexdir);
 339         struct dentry *index = NULL;
 340         struct dentry *temp = NULL;
 341         struct qstr name = { };
 342         int err;
 343
 344         /*
 345          * For now this is only used for creating index entry for directories,
 346          * because non-dir are copied up directly to index and then hardlinked
 347          * to upper dir.
 348          *
 349          * TODO: implement create index for non-dir, so we can call it when
 350          * encoding file handle for non-dir in case index does not exist.
 351          */
 352         if (WARN_ON(!d_is_dir(dentry)))
 353                 return -EIO;
 354
 355         /* Directory not expected to be indexed before copy up */
 356         if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
 357                 return -EIO;
 358
 359         err = ovl_get_index_name(origin, &name);
 360         if (err)
 361                 return err;
 362
 363         temp = ovl_create_temp(indexdir, OVL_CATTR(S_IFDIR | 0));
 364         err = PTR_ERR(temp);
 365         if (IS_ERR(temp))
 366                 goto free_name;
 367
 368         err = ovl_set_upper_fh(upper, temp);
 369         if (err)
 370                 goto out;
 371
 372         index = lookup_one_len(name.name, indexdir, name.len);
 373         if (IS_ERR(index)) {
 374                 err = PTR_ERR(index);
 375         } else {
 376                 err = ovl_do_rename(dir, temp, dir, index, 0);
 377                 dput(index);
 378         }
 379 out:
 380         if (err)
 381                 ovl_cleanup(dir, temp);
 382         dput(temp);
 383 free_name:
 384         kfree(name.name);
 385         return err;
 386 }
 387
 388 struct ovl_copy_up_ctx {
 389         struct dentry *parent;
 390         struct dentry *dentry;
 391         struct path lowerpath;
 392         struct kstat stat;
 393         struct kstat pstat;
 394         const char *link;
 395         struct dentry *destdir;
 396         struct qstr destname;
 397         struct dentry *workdir;
 398         bool origin;
 399         bool indexed;
 400         bool metacopy;
 401 };
 402
 403 static int ovl_link_up(struct ovl_copy_up_ctx *c)
 404 {
 405         int err;
 406         struct dentry *upper;
 407         struct dentry *upperdir = ovl_dentry_upper(c->parent);
 408         struct inode *udir = d_inode(upperdir);
 409
 410         /* Mark parent "impure" because it may now contain non-pure upper */
 411         err = ovl_set_impure(c->parent, upperdir);
 412         if (err)
 413                 return err;
 414
 415         err = ovl_set_nlink_lower(c->dentry);
 416         if (err)
 417                 return err;
 418
 419         inode_lock_nested(udir, I_MUTEX_PARENT);
 420         upper = lookup_one_len(c->dentry->d_name.name, upperdir,
 421                                c->dentry->d_name.len);
 422         err = PTR_ERR(upper);
 423         if (!IS_ERR(upper)) {
 424                 err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper);
 425                 dput(upper);
 426
 427                 if (!err) {
 428                         /* Restore timestamps on parent (best effort) */
 429                         ovl_set_timestamps(upperdir, &c->pstat);
 430                         ovl_dentry_set_upper_alias(c->dentry);
 431                 }
 432         }
 433         inode_unlock(udir);
 434         if (err)
 435                 return err;
 436
 437         err = ovl_set_nlink_upper(c->dentry);
 438
 439         return err;
 440 }
 441
 442 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
 443 {
 444         int err;
 445
 446         /*
 447          * Copy up data first and then xattrs. Writing data after
 448          * xattrs will remove security.capability xattr automatically.
 449          */
 450         if (S_ISREG(c->stat.mode) && !c->metacopy) {
 451                 struct path upperpath, datapath;
 452
 453                 ovl_path_upper(c->dentry, &upperpath);
 454                 if (WARN_ON(upperpath.dentry != NULL))
 455                         return -EIO;
 456                 upperpath.dentry = temp;
 457
 458                 ovl_path_lowerdata(c->dentry, &datapath);
 459                 err = ovl_copy_up_data(&datapath, &upperpath, c->stat.size);
 460                 if (err)
 461                         return err;
 462         }
 463
 464         err = ovl_copy_xattr(c->lowerpath.dentry, temp);
 465         if (err)
 466                 return err;
 467
 468         /*
 469          * Store identifier of lower inode in upper inode xattr to
 470          * allow lookup of the copy up origin inode.
 471          *
 472          * Don't set origin when we are breaking the association with a lower
 473          * hard link.
 474          */
 475         if (c->origin) {
 476                 err = ovl_set_origin(c->dentry, c->lowerpath.dentry, temp);
 477                 if (err)
 478                         return err;
 479         }
 480
 481         if (c->metacopy) {
 482                 err = ovl_check_setxattr(c->dentry, temp, OVL_XATTR_METACOPY,
 483                                          NULL, 0, -EOPNOTSUPP);
 484                 if (err)
 485                         return err;
 486         }
 487
 488         inode_lock(temp->d_inode);
 489         if (c->metacopy)
 490                 err = ovl_set_size(temp, &c->stat);
 491         if (!err)
 492                 err = ovl_set_attr(temp, &c->stat);
 493         inode_unlock(temp->d_inode);
 494
 495         return err;
 496 }
 497
 498 struct ovl_cu_creds {
 499         const struct cred *old;
 500         struct cred *new;
 501 };
 502
 503 static int ovl_prep_cu_creds(struct dentry *dentry, struct ovl_cu_creds *cc)
 504 {
 505         int err;
 506
 507         cc->old = cc->new = NULL;
 508         err = security_inode_copy_up(dentry, &cc->new);
 509         if (err < 0)
 510                 return err;
 511
 512         if (cc->new)
 513                 cc->old = override_creds(cc->new);
 514
 515         return 0;
 516 }
 517
 518 static void ovl_revert_cu_creds(struct ovl_cu_creds *cc)
 519 {
 520         if (cc->new) {
 521                 revert_creds(cc->old);
 522                 put_cred(cc->new);
 523         }
 524 }
 525
 526 /*
 527  * Copyup using workdir to prepare temp file.  Used when copying up directories,
 528  * special files or when upper fs doesn't support O_TMPFILE.
 529  */
 530 static int ovl_copy_up_workdir(struct ovl_copy_up_ctx *c)
 531 {
 532         struct inode *inode;
 533         struct inode *udir = d_inode(c->destdir), *wdir = d_inode(c->workdir);
 534         struct dentry *temp, *upper;
 535         struct ovl_cu_creds cc;
 536         int err;
 537         struct ovl_cattr cattr = {
 538                 /* Can't properly set mode on creation because of the umask */
 539                 .mode = c->stat.mode & S_IFMT,
 540                 .rdev = c->stat.rdev,
 541                 .link = c->link
 542         };
 543
 544         err = ovl_lock_rename_workdir(c->workdir, c->destdir);
 545         if (err)
 546                 return err;
 547
 548         err = ovl_prep_cu_creds(c->dentry, &cc);
 549         if (err)
 550                 goto unlock;
 551
 552         temp = ovl_create_temp(c->workdir, &cattr);
 553         ovl_revert_cu_creds(&cc);
 554
 555         err = PTR_ERR(temp);
 556         if (IS_ERR(temp))
 557                 goto unlock;
 558
 559         err = ovl_copy_up_inode(c, temp);
 560         if (err)
 561                 goto cleanup;
 562
 563         if (S_ISDIR(c->stat.mode) && c->indexed) {
 564                 err = ovl_create_index(c->dentry, c->lowerpath.dentry, temp);
 565                 if (err)
 566                         goto cleanup;
 567         }
 568
 569         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 570         err = PTR_ERR(upper);
 571         if (IS_ERR(upper))
 572                 goto cleanup;
 573
 574         err = ovl_do_rename(wdir, temp, udir, upper, 0);
 575         dput(upper);
 576         if (err)
 577                 goto cleanup;
 578
 579         if (!c->metacopy)
 580                 ovl_set_upperdata(d_inode(c->dentry));
 581         inode = d_inode(c->dentry);
 582         ovl_inode_update(inode, temp);
 583         if (S_ISDIR(inode->i_mode))
 584                 ovl_set_flag(OVL_WHITEOUTS, inode);
 585 unlock:
 586         unlock_rename(c->workdir, c->destdir);
 587
 588         return err;
 589
 590 cleanup:
 591         ovl_cleanup(wdir, temp);
 592         dput(temp);
 593         goto unlock;
 594 }
 595
 596 /* Copyup using O_TMPFILE which does not require cross dir locking */
 597 static int ovl_copy_up_tmpfile(struct ovl_copy_up_ctx *c)
 598 {
 599         struct inode *udir = d_inode(c->destdir);
 600         struct dentry *temp, *upper;
 601         struct ovl_cu_creds cc;
 602         int err;
 603
 604         err = ovl_prep_cu_creds(c->dentry, &cc);
 605         if (err)
 606                 return err;
 607
 608         temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
 609         ovl_revert_cu_creds(&cc);
 610
 611         if (IS_ERR(temp))
 612                 return PTR_ERR(temp);
 613
 614         err = ovl_copy_up_inode(c, temp);
 615         if (err)
 616                 goto out_dput;
 617
 618         inode_lock_nested(udir, I_MUTEX_PARENT);
 619
 620         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 621         err = PTR_ERR(upper);
 622         if (!IS_ERR(upper)) {
 623                 err = ovl_do_link(temp, udir, upper);
 624                 dput(upper);
 625         }
 626         inode_unlock(udir);
 627
 628         if (err)
 629                 goto out_dput;
 630
 631         if (!c->metacopy)
 632                 ovl_set_upperdata(d_inode(c->dentry));
 633         ovl_inode_update(d_inode(c->dentry), temp);
 634
 635         return 0;
 636
 637 out_dput:
 638         dput(temp);
 639         return err;
 640 }
 641
 642 /*
 643  * Copy up a single dentry
 644  *
 645  * All renames start with copy up of source if necessary.  The actual
 646  * rename will only proceed once the copy up was successful.  Copy up uses
 647  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 648  * is possible that the copy up will lock the old parent.  At that point
 649  * the file will have already been copied up anyway.
 650  */
 651 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
 652 {
 653         int err;
 654         struct ovl_fs *ofs = c->dentry->d_sb->s_fs_info;
 655         bool to_index = false;
 656
 657         /*
 658          * Indexed non-dir is copied up directly to the index entry and then
 659          * hardlinked to upper dir. Indexed dir is copied up to indexdir,
 660          * then index entry is created and then copied up dir installed.
 661          * Copying dir up to indexdir instead of workdir simplifies locking.
 662          */
 663         if (ovl_need_index(c->dentry)) {
 664                 c->indexed = true;
 665                 if (S_ISDIR(c->stat.mode))
 666                         c->workdir = ovl_indexdir(c->dentry->d_sb);
 667                 else
 668                         to_index = true;
 669         }
 670
 671         if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index)
 672                 c->origin = true;
 673
 674         if (to_index) {
 675                 c->destdir = ovl_indexdir(c->dentry->d_sb);
 676                 err = ovl_get_index_name(c->lowerpath.dentry, &c->destname);
 677                 if (err)
 678                         return err;
 679         } else if (WARN_ON(!c->parent)) {
 680                 /* Disconnected dentry must be copied up to index dir */
 681                 return -EIO;
 682         } else {
 683                 /*
 684                  * Mark parent "impure" because it may now contain non-pure
 685                  * upper
 686                  */
 687                 err = ovl_set_impure(c->parent, c->destdir);
 688                 if (err)
 689                         return err;
 690         }
 691
 692         /* Should we copyup with O_TMPFILE or with workdir? */
 693         if (S_ISREG(c->stat.mode) && ofs->tmpfile)
 694                 err = ovl_copy_up_tmpfile(c);
 695         else
 696                 err = ovl_copy_up_workdir(c);
 697         if (err)
 698                 goto out;
 699
 700         if (c->indexed)
 701                 ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
 702
 703         if (to_index) {
 704                 /* Initialize nlink for copy up of disconnected dentry */
 705                 err = ovl_set_nlink_upper(c->dentry);
 706         } else {
 707                 struct inode *udir = d_inode(c->destdir);
 708
 709                 /* Restore timestamps on parent (best effort) */
 710                 inode_lock(udir);
 711                 ovl_set_timestamps(c->destdir, &c->pstat);
 712                 inode_unlock(udir);
 713
 714                 ovl_dentry_set_upper_alias(c->dentry);
 715         }
 716
 717 out:
 718         if (to_index)
 719                 kfree(c->destname.name);
 720         return err;
 721 }
 722
 723 static bool ovl_need_meta_copy_up(struct dentry *dentry, umode_t mode,
 724                                   int flags)
 725 {
 726         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 727
 728         if (!ofs->config.metacopy)
 729                 return false;
 730
 731         if (!S_ISREG(mode))
 732                 return false;
 733
 734         if (flags && ((OPEN_FMODE(flags) & FMODE_WRITE) || (flags & O_TRUNC)))
 735                 return false;
 736
 737         return true;
 738 }
 739
 740 /* Copy up data of an inode which was copied up metadata only in the past. */
 741 static int ovl_copy_up_meta_inode_data(struct ovl_copy_up_ctx *c)
 742 {
 743         struct path upperpath, datapath;
 744         int err;
 745         char *capability = NULL;
 746         ssize_t uninitialized_var(cap_size);
 747
 748         ovl_path_upper(c->dentry, &upperpath);
 749         if (WARN_ON(upperpath.dentry == NULL))
 750                 return -EIO;
 751
 752         ovl_path_lowerdata(c->dentry, &datapath);
 753         if (WARN_ON(datapath.dentry == NULL))
 754                 return -EIO;
 755
 756         if (c->stat.size) {
 757                 err = cap_size = ovl_getxattr(upperpath.dentry, XATTR_NAME_CAPS,
 758                                               &capability, 0);
 759                 if (err < 0 && err != -ENODATA)
 760                         goto out;
 761         }
 762
 763         err = ovl_copy_up_data(&datapath, &upperpath, c->stat.size);
 764         if (err)
 765                 goto out_free;
 766
 767         /*
 768          * Writing to upper file will clear security.capability xattr. We
 769          * don't want that to happen for normal copy-up operation.
 770          */
 771         if (capability) {
 772                 err = ovl_do_setxattr(upperpath.dentry, XATTR_NAME_CAPS,
 773                                       capability, cap_size, 0);
 774                 if (err)
 775                         goto out_free;
 776         }
 777
 778
 779         err = vfs_removexattr(upperpath.dentry, OVL_XATTR_METACOPY);
 780         if (err)
 781                 goto out_free;
 782
 783         ovl_set_upperdata(d_inode(c->dentry));
 784 out_free:
 785         kfree(capability);
 786 out:
 787         return err;
 788 }
 789
 790 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 791                            int flags)
 792 {
 793         int err;
 794         DEFINE_DELAYED_CALL(done);
 795         struct path parentpath;
 796         struct ovl_copy_up_ctx ctx = {
 797                 .parent = parent,
 798                 .dentry = dentry,
 799                 .workdir = ovl_workdir(dentry),
 800         };
 801
 802         if (WARN_ON(!ctx.workdir))
 803                 return -EROFS;
 804
 805         ovl_path_lower(dentry, &ctx.lowerpath);
 806         err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
 807                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 808         if (err)
 809                 return err;
 810
 811         ctx.metacopy = ovl_need_meta_copy_up(dentry, ctx.stat.mode, flags);
 812
 813         if (parent) {
 814                 ovl_path_upper(parent, &parentpath);
 815                 ctx.destdir = parentpath.dentry;
 816                 ctx.destname = dentry->d_name;
 817
 818                 err = vfs_getattr(&parentpath, &ctx.pstat,
 819                                   STATX_ATIME | STATX_MTIME,
 820                                   AT_STATX_SYNC_AS_STAT);
 821                 if (err)
 822                         return err;
 823         }
 824
 825         /* maybe truncate regular file. this has no effect on dirs */
 826         if (flags & O_TRUNC)
 827                 ctx.stat.size = 0;
 828
 829         if (S_ISLNK(ctx.stat.mode)) {
 830                 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
 831                 if (IS_ERR(ctx.link))
 832                         return PTR_ERR(ctx.link);
 833         }
 834
 835         err = ovl_copy_up_start(dentry, flags);
 836         /* err < 0: interrupted, err > 0: raced with another copy-up */
 837         if (unlikely(err)) {
 838                 if (err > 0)
 839                         err = 0;
 840         } else {
 841                 if (!ovl_dentry_upper(dentry))
 842                         err = ovl_do_copy_up(&ctx);
 843                 if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
 844                         err = ovl_link_up(&ctx);
 845                 if (!err && ovl_dentry_needs_data_copy_up_locked(dentry, flags))
 846                         err = ovl_copy_up_meta_inode_data(&ctx);
 847                 ovl_copy_up_end(dentry);
 848         }
 849         do_delayed_call(&done);
 850
 851         return err;
 852 }
 853
 854 int ovl_copy_up_flags(struct dentry *dentry, int flags)
 855 {
 856         int err = 0;
 857         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 858         bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
 859
 860         /*
 861          * With NFS export, copy up can get called for a disconnected non-dir.
 862          * In this case, we will copy up lower inode to index dir without
 863          * linking it to upper dir.
 864          */
 865         if (WARN_ON(disconnected && d_is_dir(dentry)))
 866                 return -EIO;
 867
 868         while (!err) {
 869                 struct dentry *next;
 870                 struct dentry *parent = NULL;
 871
 872                 if (ovl_already_copied_up(dentry, flags))
 873                         break;
 874
 875                 next = dget(dentry);
 876                 /* find the topmost dentry not yet copied up */
 877                 for (; !disconnected;) {
 878                         parent = dget_parent(next);
 879
 880                         if (ovl_dentry_upper(parent))
 881                                 break;
 882
 883                         dput(next);
 884                         next = parent;
 885                 }
 886
 887                 err = ovl_copy_up_one(parent, next, flags);
 888
 889                 dput(parent);
 890                 dput(next);
 891         }
 892         revert_creds(old_cred);
 893
 894         return err;
 895 }
 896
 897 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
 898 {
 899         /* Copy up of disconnected dentry does not set upper alias */
 900         if (ovl_already_copied_up(dentry, flags))
 901                 return false;
 902
 903         if (special_file(d_inode(dentry)->i_mode))
 904                 return false;
 905
 906         if (!ovl_open_flags_need_copy_up(flags))
 907                 return false;
 908
 909         return true;
 910 }
 911
 912 int ovl_maybe_copy_up(struct dentry *dentry, int flags)
 913 {
 914         int err = 0;
 915
 916         if (ovl_open_need_copy_up(dentry, flags)) {
 917                 err = ovl_want_write(dentry);
 918                 if (!err) {
 919                         err = ovl_copy_up_flags(dentry, flags);
 920                         ovl_drop_write(dentry);
 921                 }
 922         }
 923
 924         return err;
 925 }
 926
 927 int ovl_copy_up_with_data(struct dentry *dentry)
 928 {
 929         return ovl_copy_up_flags(dentry, O_WRONLY);
 930 }
 931
 932 int ovl_copy_up(struct dentry *dentry)
 933 {
 934         return ovl_copy_up_flags(dentry, 0);
 935 }