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  27         pr_warn("overlayfs: \"check_copy_up\" module option is obsolete\n");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