fs/overlayfs/copy_up.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/fs.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/splice.h>
  15 #include <linux/xattr.h>
  16 #include <linux/security.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/cred.h>
  20 #include <linux/namei.h>
  21 #include <linux/fdtable.h>
  22 #include <linux/ratelimit.h>
  23 #include <linux/exportfs.h>
  24 #include "overlayfs.h"
  25
  26 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  27
  28 static bool __read_mostly ovl_check_copy_up;
  29 module_param_named(check_copy_up, ovl_check_copy_up, bool,
  30                    S_IWUSR | S_IRUGO);
  31 MODULE_PARM_DESC(ovl_check_copy_up,
  32                  "Warn on copy-up when causing process also has a R/O fd open");
  33
  34 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
  35 {
  36         const struct dentry *dentry = data;
  37
  38         if (file_inode(f) == d_inode(dentry))
  39                 pr_warn_ratelimited("overlayfs: Warning: Copying up %pD, but open R/O on fd %u which will cease to be coherent [pid=%d %s]\n",
  40                                     f, fd, current->pid, current->comm);
  41         return 0;
  42 }
  43
  44 /*
  45  * Check the fds open by this process and warn if something like the following
  46  * scenario is about to occur:
  47  *
  48  *      fd1 = open("foo", O_RDONLY);
  49  *      fd2 = open("foo", O_RDWR);
  50  */
  51 static void ovl_do_check_copy_up(struct dentry *dentry)
  52 {
  53         if (ovl_check_copy_up)
  54                 iterate_fd(current->files, 0, ovl_check_fd, dentry);
  55 }
  56
  57 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  58 {
  59         ssize_t list_size, size, value_size = 0;
  60         char *buf, *name, *value = NULL;
  61         int uninitialized_var(error);
  62         size_t slen;
  63
  64         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  65             !(new->d_inode->i_opflags & IOP_XATTR))
  66                 return 0;
  67
  68         list_size = vfs_listxattr(old, NULL, 0);
  69         if (list_size <= 0) {
  70                 if (list_size == -EOPNOTSUPP)
  71                         return 0;
  72                 return list_size;
  73         }
  74
  75         buf = kzalloc(list_size, GFP_KERNEL);
  76         if (!buf)
  77                 return -ENOMEM;
  78
  79         list_size = vfs_listxattr(old, buf, list_size);
  80         if (list_size <= 0) {
  81                 error = list_size;
  82                 goto out;
  83         }
  84
  85         for (name = buf; list_size; name += slen) {
  86                 slen = strnlen(name, list_size) + 1;
  87
  88                 /* underlying fs providing us with an broken xattr list? */
  89                 if (WARN_ON(slen > list_size)) {
  90                         error = -EIO;
  91                         break;
  92                 }
  93                 list_size -= slen;
  94
  95                 if (ovl_is_private_xattr(name))
  96                         continue;
  97 retry:
  98                 size = vfs_getxattr(old, name, value, value_size);
  99                 if (size == -ERANGE)
 100                         size = vfs_getxattr(old, name, NULL, 0);
 101
 102                 if (size < 0) {
 103                         error = size;
 104                         break;
 105                 }
 106
 107                 if (size > value_size) {
 108                         void *new;
 109
 110                         new = krealloc(value, size, GFP_KERNEL);
 111                         if (!new) {
 112                                 error = -ENOMEM;
 113                                 break;
 114                         }
 115                         value = new;
 116                         value_size = size;
 117                         goto retry;
 118                 }
 119
 120                 error = security_inode_copy_up_xattr(name);
 121                 if (error < 0 && error != -EOPNOTSUPP)
 122                         break;
 123                 if (error == 1) {
 124                         error = 0;
 125                         continue; /* Discard */
 126                 }
 127                 error = vfs_setxattr(new, name, value, size, 0);
 128                 if (error)
 129                         break;
 130         }
 131         kfree(value);
 132 out:
 133         kfree(buf);
 134         return error;
 135 }
 136
 137 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 138 {
 139         struct file *old_file;
 140         struct file *new_file;
 141         loff_t old_pos = 0;
 142         loff_t new_pos = 0;
 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         error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
 160         if (!error)
 161                 goto out;
 162         /* Couldn't clone, so now we try to copy the data */
 163         error = 0;
 164
 165         /* FIXME: copy up sparse files efficiently */
 166         while (len) {
 167                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 168                 long bytes;
 169
 170                 if (len < this_len)
 171                         this_len = len;
 172
 173                 if (signal_pending_state(TASK_KILLABLE, current)) {
 174                         error = -EINTR;
 175                         break;
 176                 }
 177
 178                 bytes = do_splice_direct(old_file, &old_pos,
 179                                          new_file, &new_pos,
 180                                          this_len, SPLICE_F_MOVE);
 181                 if (bytes <= 0) {
 182                         error = bytes;
 183                         break;
 184                 }
 185                 WARN_ON(old_pos != new_pos);
 186
 187                 len -= bytes;
 188         }
 189 out:
 190         if (!error)
 191                 error = vfs_fsync(new_file, 0);
 192         fput(new_file);
 193 out_fput:
 194         fput(old_file);
 195         return error;
 196 }
 197
 198 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 199 {
 200         struct iattr attr = {
 201                 .ia_valid =
 202                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 203                 .ia_atime = stat->atime,
 204                 .ia_mtime = stat->mtime,
 205         };
 206
 207         return notify_change(upperdentry, &attr, NULL);
 208 }
 209
 210 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 211 {
 212         int err = 0;
 213
 214         if (!S_ISLNK(stat->mode)) {
 215                 struct iattr attr = {
 216                         .ia_valid = ATTR_MODE,
 217                         .ia_mode = stat->mode,
 218                 };
 219                 err = notify_change(upperdentry, &attr, NULL);
 220         }
 221         if (!err) {
 222                 struct iattr attr = {
 223                         .ia_valid = ATTR_UID | ATTR_GID,
 224                         .ia_uid = stat->uid,
 225                         .ia_gid = stat->gid,
 226                 };
 227                 err = notify_change(upperdentry, &attr, NULL);
 228         }
 229         if (!err)
 230                 ovl_set_timestamps(upperdentry, stat);
 231
 232         return err;
 233 }
 234
 235 struct ovl_fh *ovl_encode_fh(struct dentry *lower, bool is_upper)
 236 {
 237         struct ovl_fh *fh;
 238         int fh_type, fh_len, dwords;
 239         void *buf;
 240         int buflen = MAX_HANDLE_SZ;
 241         uuid_t *uuid = &lower->d_sb->s_uuid;
 242
 243         buf = kmalloc(buflen, GFP_KERNEL);
 244         if (!buf)
 245                 return ERR_PTR(-ENOMEM);
 246
 247         /*
 248          * We encode a non-connectable file handle for non-dir, because we
 249          * only need to find the lower inode number and we don't want to pay
 250          * the price or reconnecting the dentry.
 251          */
 252         dwords = buflen >> 2;
 253         fh_type = exportfs_encode_fh(lower, buf, &dwords, 0);
 254         buflen = (dwords << 2);
 255
 256         fh = ERR_PTR(-EIO);
 257         if (WARN_ON(fh_type < 0) ||
 258             WARN_ON(buflen > MAX_HANDLE_SZ) ||
 259             WARN_ON(fh_type == FILEID_INVALID))
 260                 goto out;
 261
 262         BUILD_BUG_ON(MAX_HANDLE_SZ + offsetof(struct ovl_fh, fid) > 255);
 263         fh_len = offsetof(struct ovl_fh, fid) + buflen;
 264         fh = kmalloc(fh_len, GFP_KERNEL);
 265         if (!fh) {
 266                 fh = ERR_PTR(-ENOMEM);
 267                 goto out;
 268         }
 269
 270         fh->version = OVL_FH_VERSION;
 271         fh->magic = OVL_FH_MAGIC;
 272         fh->type = fh_type;
 273         fh->flags = OVL_FH_FLAG_CPU_ENDIAN;
 274         /*
 275          * When we will want to decode an overlay dentry from this handle
 276          * and all layers are on the same fs, if we get a disconncted real
 277          * dentry when we decode fid, the only way to tell if we should assign
 278          * it to upperdentry or to lowerstack is by checking this flag.
 279          */
 280         if (is_upper)
 281                 fh->flags |= OVL_FH_FLAG_PATH_UPPER;
 282         fh->len = fh_len;
 283         fh->uuid = *uuid;
 284         memcpy(fh->fid, buf, buflen);
 285
 286 out:
 287         kfree(buf);
 288         return fh;
 289 }
 290
 291 static int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
 292                           struct dentry *upper)
 293 {
 294         const struct ovl_fh *fh = NULL;
 295         int err;
 296
 297         /*
 298          * When lower layer doesn't support export operations store a 'null' fh,
 299          * so we can use the overlay.origin xattr to distignuish between a copy
 300          * up and a pure upper inode.
 301          */
 302         if (ovl_can_decode_fh(lower->d_sb)) {
 303                 fh = ovl_encode_fh(lower, false);
 304                 if (IS_ERR(fh))
 305                         return PTR_ERR(fh);
 306         }
 307
 308         /*
 309          * Do not fail when upper doesn't support xattrs.
 310          */
 311         err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh,
 312                                  fh ? fh->len : 0, 0);
 313         kfree(fh);
 314
 315         return err;
 316 }
 317
 318 struct ovl_copy_up_ctx {
 319         struct dentry *parent;
 320         struct dentry *dentry;
 321         struct path lowerpath;
 322         struct kstat stat;
 323         struct kstat pstat;
 324         const char *link;
 325         struct dentry *destdir;
 326         struct qstr destname;
 327         struct dentry *workdir;
 328         bool tmpfile;
 329         bool origin;
 330 };
 331
 332 static int ovl_link_up(struct ovl_copy_up_ctx *c)
 333 {
 334         int err;
 335         struct dentry *upper;
 336         struct dentry *upperdir = ovl_dentry_upper(c->parent);
 337         struct inode *udir = d_inode(upperdir);
 338
 339         /* Mark parent "impure" because it may now contain non-pure upper */
 340         err = ovl_set_impure(c->parent, upperdir);
 341         if (err)
 342                 return err;
 343
 344         err = ovl_set_nlink_lower(c->dentry);
 345         if (err)
 346                 return err;
 347
 348         inode_lock_nested(udir, I_MUTEX_PARENT);
 349         upper = lookup_one_len(c->dentry->d_name.name, upperdir,
 350                                c->dentry->d_name.len);
 351         err = PTR_ERR(upper);
 352         if (!IS_ERR(upper)) {
 353                 err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper,
 354                                   true);
 355                 dput(upper);
 356
 357                 if (!err) {
 358                         /* Restore timestamps on parent (best effort) */
 359                         ovl_set_timestamps(upperdir, &c->pstat);
 360                         ovl_dentry_set_upper_alias(c->dentry);
 361                 }
 362         }
 363         inode_unlock(udir);
 364         ovl_set_nlink_upper(c->dentry);
 365
 366         return err;
 367 }
 368
 369 static int ovl_install_temp(struct ovl_copy_up_ctx *c, struct dentry *temp,
 370                             struct dentry **newdentry)
 371 {
 372         int err;
 373         struct dentry *upper;
 374         struct inode *udir = d_inode(c->destdir);
 375
 376         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 377         if (IS_ERR(upper))
 378                 return PTR_ERR(upper);
 379
 380         if (c->tmpfile)
 381                 err = ovl_do_link(temp, udir, upper, true);
 382         else
 383                 err = ovl_do_rename(d_inode(c->workdir), temp, udir, upper, 0);
 384
 385         if (!err)
 386                 *newdentry = dget(c->tmpfile ? upper : temp);
 387         dput(upper);
 388
 389         return err;
 390 }
 391
 392 static int ovl_get_tmpfile(struct ovl_copy_up_ctx *c, struct dentry **tempp)
 393 {
 394         int err;
 395         struct dentry *temp;
 396         const struct cred *old_creds = NULL;
 397         struct cred *new_creds = NULL;
 398         struct cattr cattr = {
 399                 /* Can't properly set mode on creation because of the umask */
 400                 .mode = c->stat.mode & S_IFMT,
 401                 .rdev = c->stat.rdev,
 402                 .link = c->link
 403         };
 404
 405         err = security_inode_copy_up(c->dentry, &new_creds);
 406         if (err < 0)
 407                 goto out;
 408
 409         if (new_creds)
 410                 old_creds = override_creds(new_creds);
 411
 412         if (c->tmpfile) {
 413                 temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
 414                 if (IS_ERR(temp))
 415                         goto temp_err;
 416         } else {
 417                 temp = ovl_lookup_temp(c->workdir);
 418                 if (IS_ERR(temp))
 419                         goto temp_err;
 420
 421                 err = ovl_create_real(d_inode(c->workdir), temp, &cattr,
 422                                       NULL, true);
 423                 if (err) {
 424                         dput(temp);
 425                         goto out;
 426                 }
 427         }
 428         err = 0;
 429         *tempp = temp;
 430 out:
 431         if (new_creds) {
 432                 revert_creds(old_creds);
 433                 put_cred(new_creds);
 434         }
 435
 436         return err;
 437
 438 temp_err:
 439         err = PTR_ERR(temp);
 440         goto out;
 441 }
 442
 443 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
 444 {
 445         int err;
 446
 447         if (S_ISREG(c->stat.mode)) {
 448                 struct path upperpath;
 449
 450                 ovl_path_upper(c->dentry, &upperpath);
 451                 BUG_ON(upperpath.dentry != NULL);
 452                 upperpath.dentry = temp;
 453
 454                 err = ovl_copy_up_data(&c->lowerpath, &upperpath, c->stat.size);
 455                 if (err)
 456                         return err;
 457         }
 458
 459         err = ovl_copy_xattr(c->lowerpath.dentry, temp);
 460         if (err)
 461                 return err;
 462
 463         inode_lock(temp->d_inode);
 464         err = ovl_set_attr(temp, &c->stat);
 465         inode_unlock(temp->d_inode);
 466         if (err)
 467                 return err;
 468
 469         /*
 470          * Store identifier of lower inode in upper inode xattr to
 471          * allow lookup of the copy up origin inode.
 472          *
 473          * Don't set origin when we are breaking the association with a lower
 474          * hard link.
 475          */
 476         if (c->origin) {
 477                 err = ovl_set_origin(c->dentry, c->lowerpath.dentry, temp);
 478                 if (err)
 479                         return err;
 480         }
 481
 482         return 0;
 483 }
 484
 485 static int ovl_copy_up_locked(struct ovl_copy_up_ctx *c)
 486 {
 487         struct inode *udir = c->destdir->d_inode;
 488         struct inode *inode;
 489         struct dentry *newdentry = NULL;
 490         struct dentry *temp = NULL;
 491         int err;
 492
 493         err = ovl_get_tmpfile(c, &temp);
 494         if (err)
 495                 goto out;
 496
 497         err = ovl_copy_up_inode(c, temp);
 498         if (err)
 499                 goto out_cleanup;
 500
 501         if (c->tmpfile) {
 502                 inode_lock_nested(udir, I_MUTEX_PARENT);
 503                 err = ovl_install_temp(c, temp, &newdentry);
 504                 inode_unlock(udir);
 505         } else {
 506                 err = ovl_install_temp(c, temp, &newdentry);
 507         }
 508         if (err)
 509                 goto out_cleanup;
 510
 511         inode = d_inode(c->dentry);
 512         ovl_inode_update(inode, newdentry);
 513         if (S_ISDIR(inode->i_mode))
 514                 ovl_set_flag(OVL_WHITEOUTS, inode);
 515
 516 out:
 517         dput(temp);
 518         return err;
 519
 520 out_cleanup:
 521         if (!c->tmpfile)
 522                 ovl_cleanup(d_inode(c->workdir), temp);
 523         goto out;
 524 }
 525
 526 /*
 527  * Copy up a single dentry
 528  *
 529  * All renames start with copy up of source if necessary.  The actual
 530  * rename will only proceed once the copy up was successful.  Copy up uses
 531  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 532  * is possible that the copy up will lock the old parent.  At that point
 533  * the file will have already been copied up anyway.
 534  */
 535 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
 536 {
 537         int err;
 538         struct ovl_fs *ofs = c->dentry->d_sb->s_fs_info;
 539         bool indexed = false;
 540
 541         if (ovl_indexdir(c->dentry->d_sb) && !S_ISDIR(c->stat.mode) &&
 542             c->stat.nlink > 1)
 543                 indexed = true;
 544
 545         if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || indexed)
 546                 c->origin = true;
 547
 548         if (indexed) {
 549                 c->destdir = ovl_indexdir(c->dentry->d_sb);
 550                 err = ovl_get_index_name(c->lowerpath.dentry, &c->destname);
 551                 if (err)
 552                         return err;
 553         } else {
 554                 /*
 555                  * Mark parent "impure" because it may now contain non-pure
 556                  * upper
 557                  */
 558                 err = ovl_set_impure(c->parent, c->destdir);
 559                 if (err)
 560                         return err;
 561         }
 562
 563         /* Should we copyup with O_TMPFILE or with workdir? */
 564         if (S_ISREG(c->stat.mode) && ofs->tmpfile) {
 565                 c->tmpfile = true;
 566                 err = ovl_copy_up_locked(c);
 567         } else {
 568                 err = ovl_lock_rename_workdir(c->workdir, c->destdir);
 569                 if (!err) {
 570                         err = ovl_copy_up_locked(c);
 571                         unlock_rename(c->workdir, c->destdir);
 572                 }
 573         }
 574
 575         if (indexed) {
 576                 if (!err)
 577                         ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
 578                 kfree(c->destname.name);
 579         } else if (!err) {
 580                 struct inode *udir = d_inode(c->destdir);
 581
 582                 /* Restore timestamps on parent (best effort) */
 583                 inode_lock(udir);
 584                 ovl_set_timestamps(c->destdir, &c->pstat);
 585                 inode_unlock(udir);
 586
 587                 ovl_dentry_set_upper_alias(c->dentry);
 588         }
 589
 590         return err;
 591 }
 592
 593 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 594                            int flags)
 595 {
 596         int err;
 597         DEFINE_DELAYED_CALL(done);
 598         struct path parentpath;
 599         struct ovl_copy_up_ctx ctx = {
 600                 .parent = parent,
 601                 .dentry = dentry,
 602                 .workdir = ovl_workdir(dentry),
 603         };
 604
 605         if (WARN_ON(!ctx.workdir))
 606                 return -EROFS;
 607
 608         ovl_path_lower(dentry, &ctx.lowerpath);
 609         err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
 610                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 611         if (err)
 612                 return err;
 613
 614         ovl_path_upper(parent, &parentpath);
 615         ctx.destdir = parentpath.dentry;
 616         ctx.destname = dentry->d_name;
 617
 618         err = vfs_getattr(&parentpath, &ctx.pstat,
 619                           STATX_ATIME | STATX_MTIME, AT_STATX_SYNC_AS_STAT);
 620         if (err)
 621                 return err;
 622
 623         /* maybe truncate regular file. this has no effect on dirs */
 624         if (flags & O_TRUNC)
 625                 ctx.stat.size = 0;
 626
 627         if (S_ISLNK(ctx.stat.mode)) {
 628                 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
 629                 if (IS_ERR(ctx.link))
 630                         return PTR_ERR(ctx.link);
 631         }
 632         ovl_do_check_copy_up(ctx.lowerpath.dentry);
 633
 634         err = ovl_copy_up_start(dentry);
 635         /* err < 0: interrupted, err > 0: raced with another copy-up */
 636         if (unlikely(err)) {
 637                 if (err > 0)
 638                         err = 0;
 639         } else {
 640                 if (!ovl_dentry_upper(dentry))
 641                         err = ovl_do_copy_up(&ctx);
 642                 if (!err && !ovl_dentry_has_upper_alias(dentry))
 643                         err = ovl_link_up(&ctx);
 644                 ovl_copy_up_end(dentry);
 645         }
 646         do_delayed_call(&done);
 647
 648         return err;
 649 }
 650
 651 int ovl_copy_up_flags(struct dentry *dentry, int flags)
 652 {
 653         int err = 0;
 654         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 655
 656         while (!err) {
 657                 struct dentry *next;
 658                 struct dentry *parent;
 659
 660                 /*
 661                  * Check if copy-up has happened as well as for upper alias (in
 662                  * case of hard links) is there.
 663                  *
 664                  * Both checks are lockless:
 665                  *  - false negatives: will recheck under oi->lock
 666                  *  - false positives:
 667                  *    + ovl_dentry_upper() uses memory barriers to ensure the
 668                  *      upper dentry is up-to-date
 669                  *    + ovl_dentry_has_upper_alias() relies on locking of
 670                  *      upper parent i_rwsem to prevent reordering copy-up
 671                  *      with rename.
 672                  */
 673                 if (ovl_dentry_upper(dentry) &&
 674                     ovl_dentry_has_upper_alias(dentry))
 675                         break;
 676
 677                 next = dget(dentry);
 678                 /* find the topmost dentry not yet copied up */
 679                 for (;;) {
 680                         parent = dget_parent(next);
 681
 682                         if (ovl_dentry_upper(parent))
 683                                 break;
 684
 685                         dput(next);
 686                         next = parent;
 687                 }
 688
 689                 err = ovl_copy_up_one(parent, next, flags);
 690
 691                 dput(parent);
 692                 dput(next);
 693         }
 694         revert_creds(old_cred);
 695
 696         return err;
 697 }
 698
 699 int ovl_copy_up(struct dentry *dentry)
 700 {
 701         return ovl_copy_up_flags(dentry, 0);
 702 }