fs/overlayfs/copy_up.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <linux/module.h>
   8 #include <linux/fs.h>
   9 #include <linux/slab.h>
  10 #include <linux/file.h>
  11 #include <linux/splice.h>
  12 #include <linux/xattr.h>
  13 #include <linux/security.h>
  14 #include <linux/uaccess.h>
  15 #include <linux/sched/signal.h>
  16 #include <linux/cred.h>
  17 #include <linux/namei.h>
  18 #include <linux/fdtable.h>
  19 #include <linux/ratelimit.h>
  20 #include <linux/exportfs.h>
  21 #include "overlayfs.h"
  22
  23 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  24
  25 static int ovl_ccup_set(const char *buf, const struct kernel_param *param)
  26 {
  27         pr_warn("\"check_copy_up\" module option is obsolete\n");
  28         return 0;
  29 }
  30
  31 static int ovl_ccup_get(char *buf, const struct kernel_param *param)
  32 {
  33         return sprintf(buf, "N\n");
  34 }
  35
  36 module_param_call(check_copy_up, ovl_ccup_set, ovl_ccup_get, NULL, 0644);
  37 MODULE_PARM_DESC(check_copy_up, "Obsolete; does nothing");
  38
  39 static bool ovl_must_copy_xattr(const char *name)
  40 {
  41         return !strcmp(name, XATTR_POSIX_ACL_ACCESS) ||
  42                !strcmp(name, XATTR_POSIX_ACL_DEFAULT) ||
  43                !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN);
  44 }
  45
  46 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  47 {
  48         ssize_t list_size, size, value_size = 0;
  49         char *buf, *name, *value = NULL;
  50         int error = 0;
  51         size_t slen;
  52
  53         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  54             !(new->d_inode->i_opflags & IOP_XATTR))
  55                 return 0;
  56
  57         list_size = vfs_listxattr(old, NULL, 0);
  58         if (list_size <= 0) {
  59                 if (list_size == -EOPNOTSUPP)
  60                         return 0;
  61                 return list_size;
  62         }
  63
  64         buf = kzalloc(list_size, GFP_KERNEL);
  65         if (!buf)
  66                 return -ENOMEM;
  67
  68         list_size = vfs_listxattr(old, buf, list_size);
  69         if (list_size <= 0) {
  70                 error = list_size;
  71                 goto out;
  72         }
  73
  74         for (name = buf; list_size; name += slen) {
  75                 slen = strnlen(name, list_size) + 1;
  76
  77                 /* underlying fs providing us with an broken xattr list? */
  78                 if (WARN_ON(slen > list_size)) {
  79                         error = -EIO;
  80                         break;
  81                 }
  82                 list_size -= slen;
  83
  84                 if (ovl_is_private_xattr(name))
  85                         continue;
  86 retry:
  87                 size = vfs_getxattr(old, name, value, value_size);
  88                 if (size == -ERANGE)
  89                         size = vfs_getxattr(old, name, NULL, 0);
  90
  91                 if (size < 0) {
  92                         error = size;
  93                         break;
  94                 }
  95
  96                 if (size > value_size) {
  97                         void *new;
  98
  99                         new = krealloc(value, size, GFP_KERNEL);
 100                         if (!new) {
 101                                 error = -ENOMEM;
 102                                 break;
 103                         }
 104                         value = new;
 105                         value_size = size;
 106                         goto retry;
 107                 }
 108
 109                 error = security_inode_copy_up_xattr(name);
 110                 if (error < 0 && error != -EOPNOTSUPP)
 111                         break;
 112                 if (error == 1) {
 113                         error = 0;
 114                         continue; /* Discard */
 115                 }
 116                 error = vfs_setxattr(new, name, value, size, 0);
 117                 if (error) {
 118                         if (error != -EOPNOTSUPP || ovl_must_copy_xattr(name))
 119                                 break;
 120
 121                         /* Ignore failure to copy unknown xattrs */
 122                         error = 0;
 123                 }
 124         }
 125         kfree(value);
 126 out:
 127         kfree(buf);
 128         return error;
 129 }
 130
 131 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 132 {
 133         struct file *old_file;
 134         struct file *new_file;
 135         loff_t old_pos = 0;
 136         loff_t new_pos = 0;
 137         loff_t cloned;
 138         loff_t data_pos = -1;
 139         loff_t hole_len;
 140         bool skip_hole = false;
 141         int error = 0;
 142
 143         if (len == 0)
 144                 return 0;
 145
 146         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 147         if (IS_ERR(old_file))
 148                 return PTR_ERR(old_file);
 149
 150         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 151         if (IS_ERR(new_file)) {
 152                 error = PTR_ERR(new_file);
 153                 goto out_fput;
 154         }
 155
 156         /* Try to use clone_file_range to clone up within the same fs */
 157         cloned = do_clone_file_range(old_file, 0, new_file, 0, len, 0);
 158         if (cloned == len)
 159                 goto out;
 160         /* Couldn't clone, so now we try to copy the data */
 161
 162         /* Check if lower fs supports seek operation */
 163         if (old_file->f_mode & FMODE_LSEEK &&
 164             old_file->f_op->llseek)
 165                 skip_hole = true;
 166
 167         while (len) {
 168                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 169                 long bytes;
 170
 171                 if (len < this_len)
 172                         this_len = len;
 173
 174                 if (signal_pending_state(TASK_KILLABLE, current)) {
 175                         error = -EINTR;
 176                         break;
 177                 }
 178
 179                 /*
 180                  * Fill zero for hole will cost unnecessary disk space
 181                  * and meanwhile slow down the copy-up speed, so we do
 182                  * an optimization for hole during copy-up, it relies
 183                  * on SEEK_DATA implementation in lower fs so if lower
 184                  * fs does not support it, copy-up will behave as before.
 185                  *
 186                  * Detail logic of hole detection as below:
 187                  * When we detect next data position is larger than current
 188                  * position we will skip that hole, otherwise we copy
 189                  * data in the size of OVL_COPY_UP_CHUNK_SIZE. Actually,
 190                  * it may not recognize all kind of holes and sometimes
 191                  * only skips partial of hole area. However, it will be
 192                  * enough for most of the use cases.
 193                  */
 194
 195                 if (skip_hole && data_pos < old_pos) {
 196                         data_pos = vfs_llseek(old_file, old_pos, SEEK_DATA);
 197                         if (data_pos > old_pos) {
 198                                 hole_len = data_pos - old_pos;
 199                                 len -= hole_len;
 200                                 old_pos = new_pos = data_pos;
 201                                 continue;
 202                         } else if (data_pos == -ENXIO) {
 203                                 break;
 204                         } else if (data_pos < 0) {
 205                                 skip_hole = false;
 206                         }
 207                 }
 208
 209                 bytes = do_splice_direct(old_file, &old_pos,
 210                                          new_file, &new_pos,
 211                                          this_len, SPLICE_F_MOVE);
 212                 if (bytes <= 0) {
 213                         error = bytes;
 214                         break;
 215                 }
 216                 WARN_ON(old_pos != new_pos);
 217
 218                 len -= bytes;
 219         }
 220 out:
 221         if (!error)
 222                 error = vfs_fsync(new_file, 0);
 223         fput(new_file);
 224 out_fput:
 225         fput(old_file);
 226         return error;
 227 }
 228
 229 static int ovl_set_size(struct dentry *upperdentry, struct kstat *stat)
 230 {
 231         struct iattr attr = {
 232                 .ia_valid = ATTR_SIZE,
 233                 .ia_size = stat->size,
 234         };
 235
 236         return notify_change(upperdentry, &attr, NULL);
 237 }
 238
 239 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 240 {
 241         struct iattr attr = {
 242                 .ia_valid =
 243                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 244                 .ia_atime = stat->atime,
 245                 .ia_mtime = stat->mtime,
 246         };
 247
 248         return notify_change(upperdentry, &attr, NULL);
 249 }
 250
 251 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 252 {
 253         int err = 0;
 254
 255         if (!S_ISLNK(stat->mode)) {
 256                 struct iattr attr = {
 257                         .ia_valid = ATTR_MODE,
 258                         .ia_mode = stat->mode,
 259                 };
 260                 err = notify_change(upperdentry, &attr, NULL);
 261         }
 262         if (!err) {
 263                 struct iattr attr = {
 264                         .ia_valid = ATTR_UID | ATTR_GID,
 265                         .ia_uid = stat->uid,
 266                         .ia_gid = stat->gid,
 267                 };
 268                 err = notify_change(upperdentry, &attr, NULL);
 269         }
 270         if (!err)
 271                 ovl_set_timestamps(upperdentry, stat);
 272
 273         return err;
 274 }
 275
 276 struct ovl_fh *ovl_encode_real_fh(struct dentry *real, bool is_upper)
 277 {
 278         struct ovl_fh *fh;
 279         int fh_type, dwords;
 280         int buflen = MAX_HANDLE_SZ;
 281         uuid_t *uuid = &real->d_sb->s_uuid;
 282         int err;
 283
 284         /* Make sure the real fid stays 32bit aligned */
 285         BUILD_BUG_ON(OVL_FH_FID_OFFSET % 4);
 286         BUILD_BUG_ON(MAX_HANDLE_SZ + OVL_FH_FID_OFFSET > 255);
 287
 288         fh = kzalloc(buflen + OVL_FH_FID_OFFSET, GFP_KERNEL);
 289         if (!fh)
 290                 return ERR_PTR(-ENOMEM);
 291
 292         /*
 293          * We encode a non-connectable file handle for non-dir, because we
 294          * only need to find the lower inode number and we don't want to pay
 295          * the price or reconnecting the dentry.
 296          */
 297         dwords = buflen >> 2;
 298         fh_type = exportfs_encode_fh(real, (void *)fh->fb.fid, &dwords, 0);
 299         buflen = (dwords << 2);
 300
 301         err = -EIO;
 302         if (WARN_ON(fh_type < 0) ||
 303             WARN_ON(buflen > MAX_HANDLE_SZ) ||
 304             WARN_ON(fh_type == FILEID_INVALID))
 305                 goto out_err;
 306
 307         fh->fb.version = OVL_FH_VERSION;
 308         fh->fb.magic = OVL_FH_MAGIC;
 309         fh->fb.type = fh_type;
 310         fh->fb.flags = OVL_FH_FLAG_CPU_ENDIAN;
 311         /*
 312          * When we will want to decode an overlay dentry from this handle
 313          * and all layers are on the same fs, if we get a disconncted real
 314          * dentry when we decode fid, the only way to tell if we should assign
 315          * it to upperdentry or to lowerstack is by checking this flag.
 316          */
 317         if (is_upper)
 318                 fh->fb.flags |= OVL_FH_FLAG_PATH_UPPER;
 319         fh->fb.len = sizeof(fh->fb) + buflen;
 320         fh->fb.uuid = *uuid;
 321
 322         return fh;
 323
 324 out_err:
 325         kfree(fh);
 326         return ERR_PTR(err);
 327 }
 328
 329 int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
 330                    struct dentry *upper)
 331 {
 332         const struct ovl_fh *fh = NULL;
 333         int err;
 334
 335         /*
 336          * When lower layer doesn't support export operations store a 'null' fh,
 337          * so we can use the overlay.origin xattr to distignuish between a copy
 338          * up and a pure upper inode.
 339          */
 340         if (ovl_can_decode_fh(lower->d_sb)) {
 341                 fh = ovl_encode_real_fh(lower, false);
 342                 if (IS_ERR(fh))
 343                         return PTR_ERR(fh);
 344         }
 345
 346         /*
 347          * Do not fail when upper doesn't support xattrs.
 348          */
 349         err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh->buf,
 350                                  fh ? fh->fb.len : 0, 0);
 351         kfree(fh);
 352
 353         return err;
 354 }
 355
 356 /* Store file handle of @upper dir in @index dir entry */
 357 static int ovl_set_upper_fh(struct dentry *upper, struct dentry *index)
 358 {
 359         const struct ovl_fh *fh;
 360         int err;
 361
 362         fh = ovl_encode_real_fh(upper, true);
 363         if (IS_ERR(fh))
 364                 return PTR_ERR(fh);
 365
 366         err = ovl_do_setxattr(index, OVL_XATTR_UPPER, fh->buf, fh->fb.len, 0);
 367
 368         kfree(fh);
 369         return err;
 370 }
 371
 372 /*
 373  * Create and install index entry.
 374  *
 375  * Caller must hold i_mutex on indexdir.
 376  */
 377 static int ovl_create_index(struct dentry *dentry, struct dentry *origin,
 378                             struct dentry *upper)
 379 {
 380         struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
 381         struct inode *dir = d_inode(indexdir);
 382         struct dentry *index = NULL;
 383         struct dentry *temp = NULL;
 384         struct qstr name = { };
 385         int err;
 386
 387         /*
 388          * For now this is only used for creating index entry for directories,
 389          * because non-dir are copied up directly to index and then hardlinked
 390          * to upper dir.
 391          *
 392          * TODO: implement create index for non-dir, so we can call it when
 393          * encoding file handle for non-dir in case index does not exist.
 394          */
 395         if (WARN_ON(!d_is_dir(dentry)))
 396                 return -EIO;
 397
 398         /* Directory not expected to be indexed before copy up */
 399         if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
 400                 return -EIO;
 401
 402         err = ovl_get_index_name(origin, &name);
 403         if (err)
 404                 return err;
 405
 406         temp = ovl_create_temp(indexdir, OVL_CATTR(S_IFDIR | 0));
 407         err = PTR_ERR(temp);
 408         if (IS_ERR(temp))
 409                 goto free_name;
 410
 411         err = ovl_set_upper_fh(upper, temp);
 412         if (err)
 413                 goto out;
 414
 415         index = lookup_one_len(name.name, indexdir, name.len);
 416         if (IS_ERR(index)) {
 417                 err = PTR_ERR(index);
 418         } else {
 419                 err = ovl_do_rename(dir, temp, dir, index, 0);
 420                 dput(index);
 421         }
 422 out:
 423         if (err)
 424                 ovl_cleanup(dir, temp);
 425         dput(temp);
 426 free_name:
 427         kfree(name.name);
 428         return err;
 429 }
 430
 431 struct ovl_copy_up_ctx {
 432         struct dentry *parent;
 433         struct dentry *dentry;
 434         struct path lowerpath;
 435         struct kstat stat;
 436         struct kstat pstat;
 437         const char *link;
 438         struct dentry *destdir;
 439         struct qstr destname;
 440         struct dentry *workdir;
 441         bool origin;
 442         bool indexed;
 443         bool metacopy;
 444 };
 445
 446 static int ovl_link_up(struct ovl_copy_up_ctx *c)
 447 {
 448         int err;
 449         struct dentry *upper;
 450         struct dentry *upperdir = ovl_dentry_upper(c->parent);
 451         struct inode *udir = d_inode(upperdir);
 452
 453         /* Mark parent "impure" because it may now contain non-pure upper */
 454         err = ovl_set_impure(c->parent, upperdir);
 455         if (err)
 456                 return err;
 457
 458         err = ovl_set_nlink_lower(c->dentry);
 459         if (err)
 460                 return err;
 461
 462         inode_lock_nested(udir, I_MUTEX_PARENT);
 463         upper = lookup_one_len(c->dentry->d_name.name, upperdir,
 464                                c->dentry->d_name.len);
 465         err = PTR_ERR(upper);
 466         if (!IS_ERR(upper)) {
 467                 err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper);
 468                 dput(upper);
 469
 470                 if (!err) {
 471                         /* Restore timestamps on parent (best effort) */
 472                         ovl_set_timestamps(upperdir, &c->pstat);
 473                         ovl_dentry_set_upper_alias(c->dentry);
 474                 }
 475         }
 476         inode_unlock(udir);
 477         if (err)
 478                 return err;
 479
 480         err = ovl_set_nlink_upper(c->dentry);
 481
 482         return err;
 483 }
 484
 485 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
 486 {
 487         int err;
 488
 489         /*
 490          * Copy up data first and then xattrs. Writing data after
 491          * xattrs will remove security.capability xattr automatically.
 492          */
 493         if (S_ISREG(c->stat.mode) && !c->metacopy) {
 494                 struct path upperpath, datapath;
 495
 496                 ovl_path_upper(c->dentry, &upperpath);
 497                 if (WARN_ON(upperpath.dentry != NULL))
 498                         return -EIO;
 499                 upperpath.dentry = temp;
 500
 501                 ovl_path_lowerdata(c->dentry, &datapath);
 502                 err = ovl_copy_up_data(&datapath, &upperpath, c->stat.size);
 503                 if (err)
 504                         return err;
 505         }
 506
 507         err = ovl_copy_xattr(c->lowerpath.dentry, temp);
 508         if (err)
 509                 return err;
 510
 511         /*
 512          * Store identifier of lower inode in upper inode xattr to
 513          * allow lookup of the copy up origin inode.
 514          *
 515          * Don't set origin when we are breaking the association with a lower
 516          * hard link.
 517          */
 518         if (c->origin) {
 519                 err = ovl_set_origin(c->dentry, c->lowerpath.dentry, temp);
 520                 if (err)
 521                         return err;
 522         }
 523
 524         if (c->metacopy) {
 525                 err = ovl_check_setxattr(c->dentry, temp, OVL_XATTR_METACOPY,
 526                                          NULL, 0, -EOPNOTSUPP);
 527                 if (err)
 528                         return err;
 529         }
 530
 531         inode_lock(temp->d_inode);
 532         if (S_ISREG(c->stat.mode))
 533                 err = ovl_set_size(temp, &c->stat);
 534         if (!err)
 535                 err = ovl_set_attr(temp, &c->stat);
 536         inode_unlock(temp->d_inode);
 537
 538         return err;
 539 }
 540
 541 struct ovl_cu_creds {
 542         const struct cred *old;
 543         struct cred *new;
 544 };
 545
 546 static int ovl_prep_cu_creds(struct dentry *dentry, struct ovl_cu_creds *cc)
 547 {
 548         int err;
 549
 550         cc->old = cc->new = NULL;
 551         err = security_inode_copy_up(dentry, &cc->new);
 552         if (err < 0)
 553                 return err;
 554
 555         if (cc->new)
 556                 cc->old = override_creds(cc->new);
 557
 558         return 0;
 559 }
 560
 561 static void ovl_revert_cu_creds(struct ovl_cu_creds *cc)
 562 {
 563         if (cc->new) {
 564                 revert_creds(cc->old);
 565                 put_cred(cc->new);
 566         }
 567 }
 568
 569 /*
 570  * Copyup using workdir to prepare temp file.  Used when copying up directories,
 571  * special files or when upper fs doesn't support O_TMPFILE.
 572  */
 573 static int ovl_copy_up_workdir(struct ovl_copy_up_ctx *c)
 574 {
 575         struct inode *inode;
 576         struct inode *udir = d_inode(c->destdir), *wdir = d_inode(c->workdir);
 577         struct dentry *temp, *upper;
 578         struct ovl_cu_creds cc;
 579         int err;
 580         struct ovl_cattr cattr = {
 581                 /* Can't properly set mode on creation because of the umask */
 582                 .mode = c->stat.mode & S_IFMT,
 583                 .rdev = c->stat.rdev,
 584                 .link = c->link
 585         };
 586
 587         /* workdir and destdir could be the same when copying up to indexdir */
 588         err = -EIO;
 589         if (lock_rename(c->workdir, c->destdir) != NULL)
 590                 goto unlock;
 591
 592         err = ovl_prep_cu_creds(c->dentry, &cc);
 593         if (err)
 594                 goto unlock;
 595
 596         temp = ovl_create_temp(c->workdir, &cattr);
 597         ovl_revert_cu_creds(&cc);
 598
 599         err = PTR_ERR(temp);
 600         if (IS_ERR(temp))
 601                 goto unlock;
 602
 603         err = ovl_copy_up_inode(c, temp);
 604         if (err)
 605                 goto cleanup;
 606
 607         if (S_ISDIR(c->stat.mode) && c->indexed) {
 608                 err = ovl_create_index(c->dentry, c->lowerpath.dentry, temp);
 609                 if (err)
 610                         goto cleanup;
 611         }
 612
 613         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 614         err = PTR_ERR(upper);
 615         if (IS_ERR(upper))
 616                 goto cleanup;
 617
 618         err = ovl_do_rename(wdir, temp, udir, upper, 0);
 619         dput(upper);
 620         if (err)
 621                 goto cleanup;
 622
 623         if (!c->metacopy)
 624                 ovl_set_upperdata(d_inode(c->dentry));
 625         inode = d_inode(c->dentry);
 626         ovl_inode_update(inode, temp);
 627         if (S_ISDIR(inode->i_mode))
 628                 ovl_set_flag(OVL_WHITEOUTS, inode);
 629 unlock:
 630         unlock_rename(c->workdir, c->destdir);
 631
 632         return err;
 633
 634 cleanup:
 635         ovl_cleanup(wdir, temp);
 636         dput(temp);
 637         goto unlock;
 638 }
 639
 640 /* Copyup using O_TMPFILE which does not require cross dir locking */
 641 static int ovl_copy_up_tmpfile(struct ovl_copy_up_ctx *c)
 642 {
 643         struct inode *udir = d_inode(c->destdir);
 644         struct dentry *temp, *upper;
 645         struct ovl_cu_creds cc;
 646         int err;
 647
 648         err = ovl_prep_cu_creds(c->dentry, &cc);
 649         if (err)
 650                 return err;
 651
 652         temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
 653         ovl_revert_cu_creds(&cc);
 654
 655         if (IS_ERR(temp))
 656                 return PTR_ERR(temp);
 657
 658         err = ovl_copy_up_inode(c, temp);
 659         if (err)
 660                 goto out_dput;
 661
 662         inode_lock_nested(udir, I_MUTEX_PARENT);
 663
 664         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 665         err = PTR_ERR(upper);
 666         if (!IS_ERR(upper)) {
 667                 err = ovl_do_link(temp, udir, upper);
 668                 dput(upper);
 669         }
 670         inode_unlock(udir);
 671
 672         if (err)
 673                 goto out_dput;
 674
 675         if (!c->metacopy)
 676                 ovl_set_upperdata(d_inode(c->dentry));
 677         ovl_inode_update(d_inode(c->dentry), temp);
 678
 679         return 0;
 680
 681 out_dput:
 682         dput(temp);
 683         return err;
 684 }
 685
 686 /*
 687  * Copy up a single dentry
 688  *
 689  * All renames start with copy up of source if necessary.  The actual
 690  * rename will only proceed once the copy up was successful.  Copy up uses
 691  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 692  * is possible that the copy up will lock the old parent.  At that point
 693  * the file will have already been copied up anyway.
 694  */
 695 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
 696 {
 697         int err;
 698         struct ovl_fs *ofs = c->dentry->d_sb->s_fs_info;
 699         bool to_index = false;
 700
 701         /*
 702          * Indexed non-dir is copied up directly to the index entry and then
 703          * hardlinked to upper dir. Indexed dir is copied up to indexdir,
 704          * then index entry is created and then copied up dir installed.
 705          * Copying dir up to indexdir instead of workdir simplifies locking.
 706          */
 707         if (ovl_need_index(c->dentry)) {
 708                 c->indexed = true;
 709                 if (S_ISDIR(c->stat.mode))
 710                         c->workdir = ovl_indexdir(c->dentry->d_sb);
 711                 else
 712                         to_index = true;
 713         }
 714
 715         if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index)
 716                 c->origin = true;
 717
 718         if (to_index) {
 719                 c->destdir = ovl_indexdir(c->dentry->d_sb);
 720                 err = ovl_get_index_name(c->lowerpath.dentry, &c->destname);
 721                 if (err)
 722                         return err;
 723         } else if (WARN_ON(!c->parent)) {
 724                 /* Disconnected dentry must be copied up to index dir */
 725                 return -EIO;
 726         } else {
 727                 /*
 728                  * Mark parent "impure" because it may now contain non-pure
 729                  * upper
 730                  */
 731                 err = ovl_set_impure(c->parent, c->destdir);
 732                 if (err)
 733                         return err;
 734         }
 735
 736         /* Should we copyup with O_TMPFILE or with workdir? */
 737         if (S_ISREG(c->stat.mode) && ofs->tmpfile)
 738                 err = ovl_copy_up_tmpfile(c);
 739         else
 740                 err = ovl_copy_up_workdir(c);
 741         if (err)
 742                 goto out;
 743
 744         if (c->indexed)
 745                 ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
 746
 747         if (to_index) {
 748                 /* Initialize nlink for copy up of disconnected dentry */
 749                 err = ovl_set_nlink_upper(c->dentry);
 750         } else {
 751                 struct inode *udir = d_inode(c->destdir);
 752
 753                 /* Restore timestamps on parent (best effort) */
 754                 inode_lock(udir);
 755                 ovl_set_timestamps(c->destdir, &c->pstat);
 756                 inode_unlock(udir);
 757
 758                 ovl_dentry_set_upper_alias(c->dentry);
 759         }
 760
 761 out:
 762         if (to_index)
 763                 kfree(c->destname.name);
 764         return err;
 765 }
 766
 767 static bool ovl_need_meta_copy_up(struct dentry *dentry, umode_t mode,
 768                                   int flags)
 769 {
 770         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 771
 772         if (!ofs->config.metacopy)
 773                 return false;
 774
 775         if (!S_ISREG(mode))
 776                 return false;
 777
 778         if (flags && ((OPEN_FMODE(flags) & FMODE_WRITE) || (flags & O_TRUNC)))
 779                 return false;
 780
 781         return true;
 782 }
 783
 784 /* Copy up data of an inode which was copied up metadata only in the past. */
 785 static int ovl_copy_up_meta_inode_data(struct ovl_copy_up_ctx *c)
 786 {
 787         struct path upperpath, datapath;
 788         int err;
 789         char *capability = NULL;
 790         ssize_t uninitialized_var(cap_size);
 791
 792         ovl_path_upper(c->dentry, &upperpath);
 793         if (WARN_ON(upperpath.dentry == NULL))
 794                 return -EIO;
 795
 796         ovl_path_lowerdata(c->dentry, &datapath);
 797         if (WARN_ON(datapath.dentry == NULL))
 798                 return -EIO;
 799
 800         if (c->stat.size) {
 801                 err = cap_size = ovl_getxattr(upperpath.dentry, XATTR_NAME_CAPS,
 802                                               &capability, 0);
 803                 if (err < 0 && err != -ENODATA)
 804                         goto out;
 805         }
 806
 807         err = ovl_copy_up_data(&datapath, &upperpath, c->stat.size);
 808         if (err)
 809                 goto out_free;
 810
 811         /*
 812          * Writing to upper file will clear security.capability xattr. We
 813          * don't want that to happen for normal copy-up operation.
 814          */
 815         if (capability) {
 816                 err = ovl_do_setxattr(upperpath.dentry, XATTR_NAME_CAPS,
 817                                       capability, cap_size, 0);
 818                 if (err)
 819                         goto out_free;
 820         }
 821
 822
 823         err = vfs_removexattr(upperpath.dentry, OVL_XATTR_METACOPY);
 824         if (err)
 825                 goto out_free;
 826
 827         ovl_set_upperdata(d_inode(c->dentry));
 828 out_free:
 829         kfree(capability);
 830 out:
 831         return err;
 832 }
 833
 834 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 835                            int flags)
 836 {
 837         int err;
 838         DEFINE_DELAYED_CALL(done);
 839         struct path parentpath;
 840         struct ovl_copy_up_ctx ctx = {
 841                 .parent = parent,
 842                 .dentry = dentry,
 843                 .workdir = ovl_workdir(dentry),
 844         };
 845
 846         if (WARN_ON(!ctx.workdir))
 847                 return -EROFS;
 848
 849         ovl_path_lower(dentry, &ctx.lowerpath);
 850         err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
 851                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 852         if (err)
 853                 return err;
 854
 855         ctx.metacopy = ovl_need_meta_copy_up(dentry, ctx.stat.mode, flags);
 856
 857         if (parent) {
 858                 ovl_path_upper(parent, &parentpath);
 859                 ctx.destdir = parentpath.dentry;
 860                 ctx.destname = dentry->d_name;
 861
 862                 err = vfs_getattr(&parentpath, &ctx.pstat,
 863                                   STATX_ATIME | STATX_MTIME,
 864                                   AT_STATX_SYNC_AS_STAT);
 865                 if (err)
 866                         return err;
 867         }
 868
 869         /* maybe truncate regular file. this has no effect on dirs */
 870         if (flags & O_TRUNC)
 871                 ctx.stat.size = 0;
 872
 873         if (S_ISLNK(ctx.stat.mode)) {
 874                 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
 875                 if (IS_ERR(ctx.link))
 876                         return PTR_ERR(ctx.link);
 877         }
 878
 879         err = ovl_copy_up_start(dentry, flags);
 880         /* err < 0: interrupted, err > 0: raced with another copy-up */
 881         if (unlikely(err)) {
 882                 if (err > 0)
 883                         err = 0;
 884         } else {
 885                 if (!ovl_dentry_upper(dentry))
 886                         err = ovl_do_copy_up(&ctx);
 887                 if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
 888                         err = ovl_link_up(&ctx);
 889                 if (!err && ovl_dentry_needs_data_copy_up_locked(dentry, flags))
 890                         err = ovl_copy_up_meta_inode_data(&ctx);
 891                 ovl_copy_up_end(dentry);
 892         }
 893         do_delayed_call(&done);
 894
 895         return err;
 896 }
 897
 898 static int ovl_copy_up_flags(struct dentry *dentry, int flags)
 899 {
 900         int err = 0;
 901         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 902         bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
 903
 904         /*
 905          * With NFS export, copy up can get called for a disconnected non-dir.
 906          * In this case, we will copy up lower inode to index dir without
 907          * linking it to upper dir.
 908          */
 909         if (WARN_ON(disconnected && d_is_dir(dentry)))
 910                 return -EIO;
 911
 912         while (!err) {
 913                 struct dentry *next;
 914                 struct dentry *parent = NULL;
 915
 916                 if (ovl_already_copied_up(dentry, flags))
 917                         break;
 918
 919                 next = dget(dentry);
 920                 /* find the topmost dentry not yet copied up */
 921                 for (; !disconnected;) {
 922                         parent = dget_parent(next);
 923
 924                         if (ovl_dentry_upper(parent))
 925                                 break;
 926
 927                         dput(next);
 928                         next = parent;
 929                 }
 930
 931                 err = ovl_copy_up_one(parent, next, flags);
 932
 933                 dput(parent);
 934                 dput(next);
 935         }
 936         revert_creds(old_cred);
 937
 938         return err;
 939 }
 940
 941 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
 942 {
 943         /* Copy up of disconnected dentry does not set upper alias */
 944         if (ovl_already_copied_up(dentry, flags))
 945                 return false;
 946
 947         if (special_file(d_inode(dentry)->i_mode))
 948                 return false;
 949
 950         if (!ovl_open_flags_need_copy_up(flags))
 951                 return false;
 952
 953         return true;
 954 }
 955
 956 int ovl_maybe_copy_up(struct dentry *dentry, int flags)
 957 {
 958         int err = 0;
 959
 960         if (ovl_open_need_copy_up(dentry, flags)) {
 961                 err = ovl_want_write(dentry);
 962                 if (!err) {
 963                         err = ovl_copy_up_flags(dentry, flags);
 964                         ovl_drop_write(dentry);
 965                 }
 966         }
 967
 968         return err;
 969 }
 970
 971 int ovl_copy_up_with_data(struct dentry *dentry)
 972 {
 973         return ovl_copy_up_flags(dentry, O_WRONLY);
 974 }
 975
 976 int ovl_copy_up(struct dentry *dentry)
 977 {
 978         return ovl_copy_up_flags(dentry, 0);
 979 }