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  27         pr_warn("\"check_copy_up\" module option is obsolete\n");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 super_block *sb, struct dentry *old,
  47                    struct dentry *new)
  48 {
  49         ssize_t list_size, size, value_size = 0;
  50         char *buf, *name, *value = NULL;
  51         int error = 0;
  52         size_t slen;
  53
  54         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  55             !(new->d_inode->i_opflags & IOP_XATTR))
  56                 return 0;
  57
  58         list_size = vfs_listxattr(old, NULL, 0);
  59         if (list_size <= 0) {
  60                 if (list_size == -EOPNOTSUPP)
  61                         return 0;
  62                 return list_size;
  63         }
  64
  65         buf = kzalloc(list_size, GFP_KERNEL);
  66         if (!buf)
  67                 return -ENOMEM;
  68
  69         list_size = vfs_listxattr(old, buf, list_size);
  70         if (list_size <= 0) {
  71                 error = list_size;
  72                 goto out;
  73         }
  74
  75         for (name = buf; list_size; name += slen) {
  76                 slen = strnlen(name, list_size) + 1;
  77
  78                 /* underlying fs providing us with an broken xattr list? */
  79                 if (WARN_ON(slen > list_size)) {
  80                         error = -EIO;
  81                         break;
  82                 }
  83                 list_size -= slen;
  84
  85                 if (ovl_is_private_xattr(sb, name))
  86                         continue;
  87 retry:
  88                 size = vfs_getxattr(old, name, value, value_size);
  89                 if (size == -ERANGE)
  90                         size = vfs_getxattr(old, name, NULL, 0);
  91
  92                 if (size < 0) {
  93                         error = size;
  94                         break;
  95                 }
  96
  97                 if (size > value_size) {
  98                         void *new;
  99
 100                         new = krealloc(value, size, GFP_KERNEL);
 101                         if (!new) {
 102                                 error = -ENOMEM;
 103                                 break;
 104                         }
 105                         value = new;
 106                         value_size = size;
 107                         goto retry;
 108                 }
 109
 110                 error = security_inode_copy_up_xattr(name);
 111                 if (error < 0 && error != -EOPNOTSUPP)
 112                         break;
 113                 if (error == 1) {
 114                         error = 0;
 115                         continue; /* Discard */
 116                 }
 117                 error = vfs_setxattr(new, name, value, size, 0);
 118                 if (error) {
 119                         if (error != -EOPNOTSUPP || ovl_must_copy_xattr(name))
 120                                 break;
 121
 122                         /* Ignore failure to copy unknown xattrs */
 123                         error = 0;
 124                 }
 125         }
 126         kfree(value);
 127 out:
 128         kfree(buf);
 129         return error;
 130 }
 131
 132 static int ovl_copy_up_data(struct ovl_fs *ofs, struct path *old,
 133                             struct path *new, loff_t len)
 134 {
 135         struct file *old_file;
 136         struct file *new_file;
 137         loff_t old_pos = 0;
 138         loff_t new_pos = 0;
 139         loff_t cloned;
 140         loff_t data_pos = -1;
 141         loff_t hole_len;
 142         bool skip_hole = false;
 143         int error = 0;
 144
 145         if (len == 0)
 146                 return 0;
 147
 148         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 149         if (IS_ERR(old_file))
 150                 return PTR_ERR(old_file);
 151
 152         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 153         if (IS_ERR(new_file)) {
 154                 error = PTR_ERR(new_file);
 155                 goto out_fput;
 156         }
 157
 158         /* Try to use clone_file_range to clone up within the same fs */
 159         cloned = do_clone_file_range(old_file, 0, new_file, 0, len, 0);
 160         if (cloned == len)
 161                 goto out;
 162         /* Couldn't clone, so now we try to copy the data */
 163
 164         /* Check if lower fs supports seek operation */
 165         if (old_file->f_mode & FMODE_LSEEK &&
 166             old_file->f_op->llseek)
 167                 skip_hole = true;
 168
 169         while (len) {
 170                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 171                 long bytes;
 172
 173                 if (len < this_len)
 174                         this_len = len;
 175
 176                 if (signal_pending_state(TASK_KILLABLE, current)) {
 177                         error = -EINTR;
 178                         break;
 179                 }
 180
 181                 /*
 182                  * Fill zero for hole will cost unnecessary disk space
 183                  * and meanwhile slow down the copy-up speed, so we do
 184                  * an optimization for hole during copy-up, it relies
 185                  * on SEEK_DATA implementation in lower fs so if lower
 186                  * fs does not support it, copy-up will behave as before.
 187                  *
 188                  * Detail logic of hole detection as below:
 189                  * When we detect next data position is larger than current
 190                  * position we will skip that hole, otherwise we copy
 191                  * data in the size of OVL_COPY_UP_CHUNK_SIZE. Actually,
 192                  * it may not recognize all kind of holes and sometimes
 193                  * only skips partial of hole area. However, it will be
 194                  * enough for most of the use cases.
 195                  */
 196
 197                 if (skip_hole && data_pos < old_pos) {
 198                         data_pos = vfs_llseek(old_file, old_pos, SEEK_DATA);
 199                         if (data_pos > old_pos) {
 200                                 hole_len = data_pos - old_pos;
 201                                 len -= hole_len;
 202                                 old_pos = new_pos = data_pos;
 203                                 continue;
 204                         } else if (data_pos == -ENXIO) {
 205                                 break;
 206                         } else if (data_pos < 0) {
 207                                 skip_hole = false;
 208                         }
 209                 }
 210
 211                 bytes = do_splice_direct(old_file, &old_pos,
 212                                          new_file, &new_pos,
 213                                          this_len, SPLICE_F_MOVE);
 214                 if (bytes <= 0) {
 215                         error = bytes;
 216                         break;
 217                 }
 218                 WARN_ON(old_pos != new_pos);
 219
 220                 len -= bytes;
 221         }
 222 out:
 223         if (!error && ovl_should_sync(ofs))
 224                 error = vfs_fsync(new_file, 0);
 225         fput(new_file);
 226 out_fput:
 227         fput(old_file);
 228         return error;
 229 }
 230
 231 static int ovl_set_size(struct dentry *upperdentry, struct kstat *stat)
 232 {
 233         struct iattr attr = {
 234                 .ia_valid = ATTR_SIZE,
 235                 .ia_size = stat->size,
 236         };
 237
 238         return notify_change(upperdentry, &attr, NULL);
 239 }
 240
 241 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 242 {
 243         struct iattr attr = {
 244                 .ia_valid =
 245                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 246                 .ia_atime = stat->atime,
 247                 .ia_mtime = stat->mtime,
 248         };
 249
 250         return notify_change(upperdentry, &attr, NULL);
 251 }
 252
 253 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 254 {
 255         int err = 0;
 256
 257         if (!S_ISLNK(stat->mode)) {
 258                 struct iattr attr = {
 259                         .ia_valid = ATTR_MODE,
 260                         .ia_mode = stat->mode,
 261                 };
 262                 err = notify_change(upperdentry, &attr, NULL);
 263         }
 264         if (!err) {
 265                 struct iattr attr = {
 266                         .ia_valid = ATTR_UID | ATTR_GID,
 267                         .ia_uid = stat->uid,
 268                         .ia_gid = stat->gid,
 269                 };
 270                 err = notify_change(upperdentry, &attr, NULL);
 271         }
 272         if (!err)
 273                 ovl_set_timestamps(upperdentry, stat);
 274
 275         return err;
 276 }
 277
 278 struct ovl_fh *ovl_encode_real_fh(struct ovl_fs *ofs, struct dentry *real,
 279                                   bool is_upper)
 280 {
 281         struct ovl_fh *fh;
 282         int fh_type, dwords;
 283         int buflen = MAX_HANDLE_SZ;
 284         uuid_t *uuid = &real->d_sb->s_uuid;
 285         int err;
 286
 287         /* Make sure the real fid stays 32bit aligned */
 288         BUILD_BUG_ON(OVL_FH_FID_OFFSET % 4);
 289         BUILD_BUG_ON(MAX_HANDLE_SZ + OVL_FH_FID_OFFSET > 255);
 290
 291         fh = kzalloc(buflen + OVL_FH_FID_OFFSET, GFP_KERNEL);
 292         if (!fh)
 293                 return ERR_PTR(-ENOMEM);
 294
 295         /*
 296          * We encode a non-connectable file handle for non-dir, because we
 297          * only need to find the lower inode number and we don't want to pay
 298          * the price or reconnecting the dentry.
 299          */
 300         dwords = buflen >> 2;
 301         fh_type = exportfs_encode_fh(real, (void *)fh->fb.fid, &dwords, 0);
 302         buflen = (dwords << 2);
 303
 304         err = -EIO;
 305         if (WARN_ON(fh_type < 0) ||
 306             WARN_ON(buflen > MAX_HANDLE_SZ) ||
 307             WARN_ON(fh_type == FILEID_INVALID))
 308                 goto out_err;
 309
 310         fh->fb.version = OVL_FH_VERSION;
 311         fh->fb.magic = OVL_FH_MAGIC;
 312         fh->fb.type = fh_type;
 313         fh->fb.flags = OVL_FH_FLAG_CPU_ENDIAN;
 314         /*
 315          * When we will want to decode an overlay dentry from this handle
 316          * and all layers are on the same fs, if we get a disconncted real
 317          * dentry when we decode fid, the only way to tell if we should assign
 318          * it to upperdentry or to lowerstack is by checking this flag.
 319          */
 320         if (is_upper)
 321                 fh->fb.flags |= OVL_FH_FLAG_PATH_UPPER;
 322         fh->fb.len = sizeof(fh->fb) + buflen;
 323         if (ofs->config.uuid)
 324                 fh->fb.uuid = *uuid;
 325
 326         return fh;
 327
 328 out_err:
 329         kfree(fh);
 330         return ERR_PTR(err);
 331 }
 332
 333 int ovl_set_origin(struct ovl_fs *ofs, struct dentry *dentry,
 334                    struct dentry *lower, struct dentry *upper)
 335 {
 336         const struct ovl_fh *fh = NULL;
 337         int err;
 338
 339         /*
 340          * When lower layer doesn't support export operations store a 'null' fh,
 341          * so we can use the overlay.origin xattr to distignuish between a copy
 342          * up and a pure upper inode.
 343          */
 344         if (ovl_can_decode_fh(lower->d_sb)) {
 345                 fh = ovl_encode_real_fh(ofs, lower, false);
 346                 if (IS_ERR(fh))
 347                         return PTR_ERR(fh);
 348         }
 349
 350         /*
 351          * Do not fail when upper doesn't support xattrs.
 352          */
 353         err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh->buf,
 354                                  fh ? fh->fb.len : 0, 0);
 355         kfree(fh);
 356
 357         /* Ignore -EPERM from setting "user.*" on symlink/special */
 358         return err == -EPERM ? 0 : err;
 359 }
 360
 361 /* Store file handle of @upper dir in @index dir entry */
 362 static int ovl_set_upper_fh(struct ovl_fs *ofs, struct dentry *upper,
 363                             struct dentry *index)
 364 {
 365         const struct ovl_fh *fh;
 366         int err;
 367
 368         fh = ovl_encode_real_fh(ofs, upper, true);
 369         if (IS_ERR(fh))
 370                 return PTR_ERR(fh);
 371
 372         err = ovl_do_setxattr(ofs, index, OVL_XATTR_UPPER, fh->buf, fh->fb.len);
 373
 374         kfree(fh);
 375         return err;
 376 }
 377
 378 /*
 379  * Create and install index entry.
 380  *
 381  * Caller must hold i_mutex on indexdir.
 382  */
 383 static int ovl_create_index(struct dentry *dentry, struct dentry *origin,
 384                             struct dentry *upper)
 385 {
 386         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 387         struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
 388         struct inode *dir = d_inode(indexdir);
 389         struct dentry *index = NULL;
 390         struct dentry *temp = NULL;
 391         struct qstr name = { };
 392         int err;
 393
 394         /*
 395          * For now this is only used for creating index entry for directories,
 396          * because non-dir are copied up directly to index and then hardlinked
 397          * to upper dir.
 398          *
 399          * TODO: implement create index for non-dir, so we can call it when
 400          * encoding file handle for non-dir in case index does not exist.
 401          */
 402         if (WARN_ON(!d_is_dir(dentry)))
 403                 return -EIO;
 404
 405         /* Directory not expected to be indexed before copy up */
 406         if (WARN_ON(ovl_test_flag(OVL_INDEX, d_inode(dentry))))
 407                 return -EIO;
 408
 409         err = ovl_get_index_name(ofs, origin, &name);
 410         if (err)
 411                 return err;
 412
 413         temp = ovl_create_temp(indexdir, OVL_CATTR(S_IFDIR | 0));
 414         err = PTR_ERR(temp);
 415         if (IS_ERR(temp))
 416                 goto free_name;
 417
 418         err = ovl_set_upper_fh(ofs, upper, temp);
 419         if (err)
 420                 goto out;
 421
 422         index = lookup_one_len(name.name, indexdir, name.len);
 423         if (IS_ERR(index)) {
 424                 err = PTR_ERR(index);
 425         } else {
 426                 err = ovl_do_rename(dir, temp, dir, index, 0);
 427                 dput(index);
 428         }
 429 out:
 430         if (err)
 431                 ovl_cleanup(dir, temp);
 432         dput(temp);
 433 free_name:
 434         kfree(name.name);
 435         return err;
 436 }
 437
 438 struct ovl_copy_up_ctx {
 439         struct dentry *parent;
 440         struct dentry *dentry;
 441         struct path lowerpath;
 442         struct kstat stat;
 443         struct kstat pstat;
 444         const char *link;
 445         struct dentry *destdir;
 446         struct qstr destname;
 447         struct dentry *workdir;
 448         bool origin;
 449         bool indexed;
 450         bool metacopy;
 451 };
 452
 453 static int ovl_link_up(struct ovl_copy_up_ctx *c)
 454 {
 455         int err;
 456         struct dentry *upper;
 457         struct dentry *upperdir = ovl_dentry_upper(c->parent);
 458         struct inode *udir = d_inode(upperdir);
 459
 460         /* Mark parent "impure" because it may now contain non-pure upper */
 461         err = ovl_set_impure(c->parent, upperdir);
 462         if (err)
 463                 return err;
 464
 465         err = ovl_set_nlink_lower(c->dentry);
 466         if (err)
 467                 return err;
 468
 469         inode_lock_nested(udir, I_MUTEX_PARENT);
 470         upper = lookup_one_len(c->dentry->d_name.name, upperdir,
 471                                c->dentry->d_name.len);
 472         err = PTR_ERR(upper);
 473         if (!IS_ERR(upper)) {
 474                 err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper);
 475                 dput(upper);
 476
 477                 if (!err) {
 478                         /* Restore timestamps on parent (best effort) */
 479                         ovl_set_timestamps(upperdir, &c->pstat);
 480                         ovl_dentry_set_upper_alias(c->dentry);
 481                 }
 482         }
 483         inode_unlock(udir);
 484         if (err)
 485                 return err;
 486
 487         err = ovl_set_nlink_upper(c->dentry);
 488
 489         return err;
 490 }
 491
 492 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
 493 {
 494         struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
 495         int err;
 496
 497         /*
 498          * Copy up data first and then xattrs. Writing data after
 499          * xattrs will remove security.capability xattr automatically.
 500          */
 501         if (S_ISREG(c->stat.mode) && !c->metacopy) {
 502                 struct path upperpath, datapath;
 503
 504                 ovl_path_upper(c->dentry, &upperpath);
 505                 if (WARN_ON(upperpath.dentry != NULL))
 506                         return -EIO;
 507                 upperpath.dentry = temp;
 508
 509                 ovl_path_lowerdata(c->dentry, &datapath);
 510                 err = ovl_copy_up_data(ofs, &datapath, &upperpath,
 511                                        c->stat.size);
 512                 if (err)
 513                         return err;
 514         }
 515
 516         err = ovl_copy_xattr(c->dentry->d_sb, c->lowerpath.dentry, temp);
 517         if (err)
 518                 return err;
 519
 520         /*
 521          * Store identifier of lower inode in upper inode xattr to
 522          * allow lookup of the copy up origin inode.
 523          *
 524          * Don't set origin when we are breaking the association with a lower
 525          * hard link.
 526          */
 527         if (c->origin) {
 528                 err = ovl_set_origin(ofs, c->dentry, c->lowerpath.dentry, temp);
 529                 if (err)
 530                         return err;
 531         }
 532
 533         if (c->metacopy) {
 534                 err = ovl_check_setxattr(c->dentry, temp, OVL_XATTR_METACOPY,
 535                                          NULL, 0, -EOPNOTSUPP);
 536                 if (err)
 537                         return err;
 538         }
 539
 540         inode_lock(temp->d_inode);
 541         if (S_ISREG(c->stat.mode))
 542                 err = ovl_set_size(temp, &c->stat);
 543         if (!err)
 544                 err = ovl_set_attr(temp, &c->stat);
 545         inode_unlock(temp->d_inode);
 546
 547         return err;
 548 }
 549
 550 struct ovl_cu_creds {
 551         const struct cred *old;
 552         struct cred *new;
 553 };
 554
 555 static int ovl_prep_cu_creds(struct dentry *dentry, struct ovl_cu_creds *cc)
 556 {
 557         int err;
 558
 559         cc->old = cc->new = NULL;
 560         err = security_inode_copy_up(dentry, &cc->new);
 561         if (err < 0)
 562                 return err;
 563
 564         if (cc->new)
 565                 cc->old = override_creds(cc->new);
 566
 567         return 0;
 568 }
 569
 570 static void ovl_revert_cu_creds(struct ovl_cu_creds *cc)
 571 {
 572         if (cc->new) {
 573                 revert_creds(cc->old);
 574                 put_cred(cc->new);
 575         }
 576 }
 577
 578 /*
 579  * Copyup using workdir to prepare temp file.  Used when copying up directories,
 580  * special files or when upper fs doesn't support O_TMPFILE.
 581  */
 582 static int ovl_copy_up_workdir(struct ovl_copy_up_ctx *c)
 583 {
 584         struct inode *inode;
 585         struct inode *udir = d_inode(c->destdir), *wdir = d_inode(c->workdir);
 586         struct dentry *temp, *upper;
 587         struct ovl_cu_creds cc;
 588         int err;
 589         struct ovl_cattr cattr = {
 590                 /* Can't properly set mode on creation because of the umask */
 591                 .mode = c->stat.mode & S_IFMT,
 592                 .rdev = c->stat.rdev,
 593                 .link = c->link
 594         };
 595
 596         /* workdir and destdir could be the same when copying up to indexdir */
 597         err = -EIO;
 598         if (lock_rename(c->workdir, c->destdir) != NULL)
 599                 goto unlock;
 600
 601         err = ovl_prep_cu_creds(c->dentry, &cc);
 602         if (err)
 603                 goto unlock;
 604
 605         temp = ovl_create_temp(c->workdir, &cattr);
 606         ovl_revert_cu_creds(&cc);
 607
 608         err = PTR_ERR(temp);
 609         if (IS_ERR(temp))
 610                 goto unlock;
 611
 612         err = ovl_copy_up_inode(c, temp);
 613         if (err)
 614                 goto cleanup;
 615
 616         if (S_ISDIR(c->stat.mode) && c->indexed) {
 617                 err = ovl_create_index(c->dentry, c->lowerpath.dentry, temp);
 618                 if (err)
 619                         goto cleanup;
 620         }
 621
 622         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 623         err = PTR_ERR(upper);
 624         if (IS_ERR(upper))
 625                 goto cleanup;
 626
 627         err = ovl_do_rename(wdir, temp, udir, upper, 0);
 628         dput(upper);
 629         if (err)
 630                 goto cleanup;
 631
 632         if (!c->metacopy)
 633                 ovl_set_upperdata(d_inode(c->dentry));
 634         inode = d_inode(c->dentry);
 635         ovl_inode_update(inode, temp);
 636         if (S_ISDIR(inode->i_mode))
 637                 ovl_set_flag(OVL_WHITEOUTS, inode);
 638 unlock:
 639         unlock_rename(c->workdir, c->destdir);
 640
 641         return err;
 642
 643 cleanup:
 644         ovl_cleanup(wdir, temp);
 645         dput(temp);
 646         goto unlock;
 647 }
 648
 649 /* Copyup using O_TMPFILE which does not require cross dir locking */
 650 static int ovl_copy_up_tmpfile(struct ovl_copy_up_ctx *c)
 651 {
 652         struct inode *udir = d_inode(c->destdir);
 653         struct dentry *temp, *upper;
 654         struct ovl_cu_creds cc;
 655         int err;
 656
 657         err = ovl_prep_cu_creds(c->dentry, &cc);
 658         if (err)
 659                 return err;
 660
 661         temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
 662         ovl_revert_cu_creds(&cc);
 663
 664         if (IS_ERR(temp))
 665                 return PTR_ERR(temp);
 666
 667         err = ovl_copy_up_inode(c, temp);
 668         if (err)
 669                 goto out_dput;
 670
 671         inode_lock_nested(udir, I_MUTEX_PARENT);
 672
 673         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 674         err = PTR_ERR(upper);
 675         if (!IS_ERR(upper)) {
 676                 err = ovl_do_link(temp, udir, upper);
 677                 dput(upper);
 678         }
 679         inode_unlock(udir);
 680
 681         if (err)
 682                 goto out_dput;
 683
 684         if (!c->metacopy)
 685                 ovl_set_upperdata(d_inode(c->dentry));
 686         ovl_inode_update(d_inode(c->dentry), temp);
 687
 688         return 0;
 689
 690 out_dput:
 691         dput(temp);
 692         return err;
 693 }
 694
 695 /*
 696  * Copy up a single dentry
 697  *
 698  * All renames start with copy up of source if necessary.  The actual
 699  * rename will only proceed once the copy up was successful.  Copy up uses
 700  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 701  * is possible that the copy up will lock the old parent.  At that point
 702  * the file will have already been copied up anyway.
 703  */
 704 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
 705 {
 706         int err;
 707         struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
 708         bool to_index = false;
 709
 710         /*
 711          * Indexed non-dir is copied up directly to the index entry and then
 712          * hardlinked to upper dir. Indexed dir is copied up to indexdir,
 713          * then index entry is created and then copied up dir installed.
 714          * Copying dir up to indexdir instead of workdir simplifies locking.
 715          */
 716         if (ovl_need_index(c->dentry)) {
 717                 c->indexed = true;
 718                 if (S_ISDIR(c->stat.mode))
 719                         c->workdir = ovl_indexdir(c->dentry->d_sb);
 720                 else
 721                         to_index = true;
 722         }
 723
 724         if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || to_index)
 725                 c->origin = true;
 726
 727         if (to_index) {
 728                 c->destdir = ovl_indexdir(c->dentry->d_sb);
 729                 err = ovl_get_index_name(ofs, c->lowerpath.dentry, &c->destname);
 730                 if (err)
 731                         return err;
 732         } else if (WARN_ON(!c->parent)) {
 733                 /* Disconnected dentry must be copied up to index dir */
 734                 return -EIO;
 735         } else {
 736                 /*
 737                  * Mark parent "impure" because it may now contain non-pure
 738                  * upper
 739                  */
 740                 err = ovl_set_impure(c->parent, c->destdir);
 741                 if (err)
 742                         return err;
 743         }
 744
 745         /* Should we copyup with O_TMPFILE or with workdir? */
 746         if (S_ISREG(c->stat.mode) && ofs->tmpfile)
 747                 err = ovl_copy_up_tmpfile(c);
 748         else
 749                 err = ovl_copy_up_workdir(c);
 750         if (err)
 751                 goto out;
 752
 753         if (c->indexed)
 754                 ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
 755
 756         if (to_index) {
 757                 /* Initialize nlink for copy up of disconnected dentry */
 758                 err = ovl_set_nlink_upper(c->dentry);
 759         } else {
 760                 struct inode *udir = d_inode(c->destdir);
 761
 762                 /* Restore timestamps on parent (best effort) */
 763                 inode_lock(udir);
 764                 ovl_set_timestamps(c->destdir, &c->pstat);
 765                 inode_unlock(udir);
 766
 767                 ovl_dentry_set_upper_alias(c->dentry);
 768         }
 769
 770 out:
 771         if (to_index)
 772                 kfree(c->destname.name);
 773         return err;
 774 }
 775
 776 static bool ovl_need_meta_copy_up(struct dentry *dentry, umode_t mode,
 777                                   int flags)
 778 {
 779         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 780
 781         if (!ofs->config.metacopy)
 782                 return false;
 783
 784         if (!S_ISREG(mode))
 785                 return false;
 786
 787         if (flags && ((OPEN_FMODE(flags) & FMODE_WRITE) || (flags & O_TRUNC)))
 788                 return false;
 789
 790         return true;
 791 }
 792
 793 static ssize_t ovl_getxattr(struct dentry *dentry, char *name, char **value)
 794 {
 795         ssize_t res;
 796         char *buf;
 797
 798         res = vfs_getxattr(dentry, name, NULL, 0);
 799         if (res == -ENODATA || res == -EOPNOTSUPP)
 800                 res = 0;
 801
 802         if (res > 0) {
 803                 buf = kzalloc(res, GFP_KERNEL);
 804                 if (!buf)
 805                         return -ENOMEM;
 806
 807                 res = vfs_getxattr(dentry, name, buf, res);
 808                 if (res < 0)
 809                         kfree(buf);
 810                 else
 811                         *value = buf;
 812         }
 813         return res;
 814 }
 815
 816 /* Copy up data of an inode which was copied up metadata only in the past. */
 817 static int ovl_copy_up_meta_inode_data(struct ovl_copy_up_ctx *c)
 818 {
 819         struct ovl_fs *ofs = OVL_FS(c->dentry->d_sb);
 820         struct path upperpath, datapath;
 821         int err;
 822         char *capability = NULL;
 823         ssize_t cap_size;
 824
 825         ovl_path_upper(c->dentry, &upperpath);
 826         if (WARN_ON(upperpath.dentry == NULL))
 827                 return -EIO;
 828
 829         ovl_path_lowerdata(c->dentry, &datapath);
 830         if (WARN_ON(datapath.dentry == NULL))
 831                 return -EIO;
 832
 833         if (c->stat.size) {
 834                 err = cap_size = ovl_getxattr(upperpath.dentry, XATTR_NAME_CAPS,
 835                                               &capability);
 836                 if (cap_size < 0)
 837                         goto out;
 838         }
 839
 840         err = ovl_copy_up_data(ofs, &datapath, &upperpath, c->stat.size);
 841         if (err)
 842                 goto out_free;
 843
 844         /*
 845          * Writing to upper file will clear security.capability xattr. We
 846          * don't want that to happen for normal copy-up operation.
 847          */
 848         if (capability) {
 849                 err = vfs_setxattr(upperpath.dentry, XATTR_NAME_CAPS,
 850                                    capability, cap_size, 0);
 851                 if (err)
 852                         goto out_free;
 853         }
 854
 855
 856         err = ovl_do_removexattr(ofs, upperpath.dentry, OVL_XATTR_METACOPY);
 857         if (err)
 858                 goto out_free;
 859
 860         ovl_set_upperdata(d_inode(c->dentry));
 861 out_free:
 862         kfree(capability);
 863 out:
 864         return err;
 865 }
 866
 867 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 868                            int flags)
 869 {
 870         int err;
 871         DEFINE_DELAYED_CALL(done);
 872         struct path parentpath;
 873         struct ovl_copy_up_ctx ctx = {
 874                 .parent = parent,
 875                 .dentry = dentry,
 876                 .workdir = ovl_workdir(dentry),
 877         };
 878
 879         if (WARN_ON(!ctx.workdir))
 880                 return -EROFS;
 881
 882         ovl_path_lower(dentry, &ctx.lowerpath);
 883         err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
 884                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 885         if (err)
 886                 return err;
 887
 888         ctx.metacopy = ovl_need_meta_copy_up(dentry, ctx.stat.mode, flags);
 889
 890         if (parent) {
 891                 ovl_path_upper(parent, &parentpath);
 892                 ctx.destdir = parentpath.dentry;
 893                 ctx.destname = dentry->d_name;
 894
 895                 err = vfs_getattr(&parentpath, &ctx.pstat,
 896                                   STATX_ATIME | STATX_MTIME,
 897                                   AT_STATX_SYNC_AS_STAT);
 898                 if (err)
 899                         return err;
 900         }
 901
 902         /* maybe truncate regular file. this has no effect on dirs */
 903         if (flags & O_TRUNC)
 904                 ctx.stat.size = 0;
 905
 906         if (S_ISLNK(ctx.stat.mode)) {
 907                 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
 908                 if (IS_ERR(ctx.link))
 909                         return PTR_ERR(ctx.link);
 910         }
 911
 912         err = ovl_copy_up_start(dentry, flags);
 913         /* err < 0: interrupted, err > 0: raced with another copy-up */
 914         if (unlikely(err)) {
 915                 if (err > 0)
 916                         err = 0;
 917         } else {
 918                 if (!ovl_dentry_upper(dentry))
 919                         err = ovl_do_copy_up(&ctx);
 920                 if (!err && parent && !ovl_dentry_has_upper_alias(dentry))
 921                         err = ovl_link_up(&ctx);
 922                 if (!err && ovl_dentry_needs_data_copy_up_locked(dentry, flags))
 923                         err = ovl_copy_up_meta_inode_data(&ctx);
 924                 ovl_copy_up_end(dentry);
 925         }
 926         do_delayed_call(&done);
 927
 928         return err;
 929 }
 930
 931 static int ovl_copy_up_flags(struct dentry *dentry, int flags)
 932 {
 933         int err = 0;
 934         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 935         bool disconnected = (dentry->d_flags & DCACHE_DISCONNECTED);
 936
 937         /*
 938          * With NFS export, copy up can get called for a disconnected non-dir.
 939          * In this case, we will copy up lower inode to index dir without
 940          * linking it to upper dir.
 941          */
 942         if (WARN_ON(disconnected && d_is_dir(dentry)))
 943                 return -EIO;
 944
 945         while (!err) {
 946                 struct dentry *next;
 947                 struct dentry *parent = NULL;
 948
 949                 if (ovl_already_copied_up(dentry, flags))
 950                         break;
 951
 952                 next = dget(dentry);
 953                 /* find the topmost dentry not yet copied up */
 954                 for (; !disconnected;) {
 955                         parent = dget_parent(next);
 956
 957                         if (ovl_dentry_upper(parent))
 958                                 break;
 959
 960                         dput(next);
 961                         next = parent;
 962                 }
 963
 964                 err = ovl_copy_up_one(parent, next, flags);
 965
 966                 dput(parent);
 967                 dput(next);
 968         }
 969         revert_creds(old_cred);
 970
 971         return err;
 972 }
 973
 974 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
 975 {
 976         /* Copy up of disconnected dentry does not set upper alias */
 977         if (ovl_already_copied_up(dentry, flags))
 978                 return false;
 979
 980         if (special_file(d_inode(dentry)->i_mode))
 981                 return false;
 982
 983         if (!ovl_open_flags_need_copy_up(flags))
 984                 return false;
 985
 986         return true;
 987 }
 988
 989 int ovl_maybe_copy_up(struct dentry *dentry, int flags)
 990 {
 991         int err = 0;
 992
 993         if (ovl_open_need_copy_up(dentry, flags)) {
 994                 err = ovl_want_write(dentry);
 995                 if (!err) {
 996                         err = ovl_copy_up_flags(dentry, flags);
 997                         ovl_drop_write(dentry);
 998                 }
 999         }
1000
1001         return err;
1002 }
1003
1004 int ovl_copy_up_with_data(struct dentry *dentry)
1005 {
1006         return ovl_copy_up_flags(dentry, O_WRONLY);
1007 }
1008
1009 int ovl_copy_up(struct dentry *dentry)
1010 {
1011         return ovl_copy_up_flags(dentry, 0);
1012 }