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  27         pr_warn("\"check_copy_up\" module option is obsolete\n");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