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 #include "ovl_entry.h"
  26
  27 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  28
  29 static bool __read_mostly ovl_check_copy_up;
  30 module_param_named(check_copy_up, ovl_check_copy_up, bool,
  31                    S_IWUSR | S_IRUGO);
  32 MODULE_PARM_DESC(ovl_check_copy_up,
  33                  "Warn on copy-up when causing process also has a R/O fd open");
  34
  35 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
  36 {
  37         const struct dentry *dentry = data;
  38
  39         if (file_inode(f) == d_inode(dentry))
  40                 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",
  41                                     f, fd, current->pid, current->comm);
  42         return 0;
  43 }
  44
  45 /*
  46  * Check the fds open by this process and warn if something like the following
  47  * scenario is about to occur:
  48  *
  49  *      fd1 = open("foo", O_RDONLY);
  50  *      fd2 = open("foo", O_RDWR);
  51  */
  52 static void ovl_do_check_copy_up(struct dentry *dentry)
  53 {
  54         if (ovl_check_copy_up)
  55                 iterate_fd(current->files, 0, ovl_check_fd, dentry);
  56 }
  57
  58 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  59 {
  60         ssize_t list_size, size, value_size = 0;
  61         char *buf, *name, *value = NULL;
  62         int uninitialized_var(error);
  63         size_t slen;
  64
  65         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  66             !(new->d_inode->i_opflags & IOP_XATTR))
  67                 return 0;
  68
  69         list_size = vfs_listxattr(old, NULL, 0);
  70         if (list_size <= 0) {
  71                 if (list_size == -EOPNOTSUPP)
  72                         return 0;
  73                 return list_size;
  74         }
  75
  76         buf = kzalloc(list_size, GFP_KERNEL);
  77         if (!buf)
  78                 return -ENOMEM;
  79
  80         list_size = vfs_listxattr(old, buf, list_size);
  81         if (list_size <= 0) {
  82                 error = list_size;
  83                 goto out;
  84         }
  85
  86         for (name = buf; list_size; name += slen) {
  87                 slen = strnlen(name, list_size) + 1;
  88
  89                 /* underlying fs providing us with an broken xattr list? */
  90                 if (WARN_ON(slen > list_size)) {
  91                         error = -EIO;
  92                         break;
  93                 }
  94                 list_size -= slen;
  95
  96                 if (ovl_is_private_xattr(name))
  97                         continue;
  98 retry:
  99                 size = vfs_getxattr(old, name, value, value_size);
 100                 if (size == -ERANGE)
 101                         size = vfs_getxattr(old, name, NULL, 0);
 102
 103                 if (size < 0) {
 104                         error = size;
 105                         break;
 106                 }
 107
 108                 if (size > value_size) {
 109                         void *new;
 110
 111                         new = krealloc(value, size, GFP_KERNEL);
 112                         if (!new) {
 113                                 error = -ENOMEM;
 114                                 break;
 115                         }
 116                         value = new;
 117                         value_size = size;
 118                         goto retry;
 119                 }
 120
 121                 error = security_inode_copy_up_xattr(name);
 122                 if (error < 0 && error != -EOPNOTSUPP)
 123                         break;
 124                 if (error == 1) {
 125                         error = 0;
 126                         continue; /* Discard */
 127                 }
 128                 error = vfs_setxattr(new, name, value, size, 0);
 129                 if (error)
 130                         break;
 131         }
 132         kfree(value);
 133 out:
 134         kfree(buf);
 135         return error;
 136 }
 137
 138 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 139 {
 140         struct file *old_file;
 141         struct file *new_file;
 142         loff_t old_pos = 0;
 143         loff_t new_pos = 0;
 144         int error = 0;
 145
 146         if (len == 0)
 147                 return 0;
 148
 149         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 150         if (IS_ERR(old_file))
 151                 return PTR_ERR(old_file);
 152
 153         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 154         if (IS_ERR(new_file)) {
 155                 error = PTR_ERR(new_file);
 156                 goto out_fput;
 157         }
 158
 159         /* Try to use clone_file_range to clone up within the same fs */
 160         error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
 161         if (!error)
 162                 goto out;
 163         /* Couldn't clone, so now we try to copy the data */
 164         error = 0;
 165
 166         /* FIXME: copy up sparse files efficiently */
 167         while (len) {
 168                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 169                 long bytes;
 170
 171                 if (len < this_len)
 172                         this_len = len;
 173
 174                 if (signal_pending_state(TASK_KILLABLE, current)) {
 175                         error = -EINTR;
 176                         break;
 177                 }
 178
 179                 bytes = do_splice_direct(old_file, &old_pos,
 180                                          new_file, &new_pos,
 181                                          this_len, SPLICE_F_MOVE);
 182                 if (bytes <= 0) {
 183                         error = bytes;
 184                         break;
 185                 }
 186                 WARN_ON(old_pos != new_pos);
 187
 188                 len -= bytes;
 189         }
 190 out:
 191         if (!error)
 192                 error = vfs_fsync(new_file, 0);
 193         fput(new_file);
 194 out_fput:
 195         fput(old_file);
 196         return error;
 197 }
 198
 199 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 200 {
 201         struct iattr attr = {
 202                 .ia_valid =
 203                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 204                 .ia_atime = stat->atime,
 205                 .ia_mtime = stat->mtime,
 206         };
 207
 208         return notify_change(upperdentry, &attr, NULL);
 209 }
 210
 211 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 212 {
 213         int err = 0;
 214
 215         if (!S_ISLNK(stat->mode)) {
 216                 struct iattr attr = {
 217                         .ia_valid = ATTR_MODE,
 218                         .ia_mode = stat->mode,
 219                 };
 220                 err = notify_change(upperdentry, &attr, NULL);
 221         }
 222         if (!err) {
 223                 struct iattr attr = {
 224                         .ia_valid = ATTR_UID | ATTR_GID,
 225                         .ia_uid = stat->uid,
 226                         .ia_gid = stat->gid,
 227                 };
 228                 err = notify_change(upperdentry, &attr, NULL);
 229         }
 230         if (!err)
 231                 ovl_set_timestamps(upperdentry, stat);
 232
 233         return err;
 234 }
 235
 236 struct ovl_fh *ovl_encode_fh(struct dentry *lower, bool is_upper)
 237 {
 238         struct ovl_fh *fh;
 239         int fh_type, fh_len, dwords;
 240         void *buf;
 241         int buflen = MAX_HANDLE_SZ;
 242         uuid_t *uuid = &lower->d_sb->s_uuid;
 243
 244         buf = kmalloc(buflen, GFP_TEMPORARY);
 245         if (!buf)
 246                 return ERR_PTR(-ENOMEM);
 247
 248         /*
 249          * We encode a non-connectable file handle for non-dir, because we
 250          * only need to find the lower inode number and we don't want to pay
 251          * the price or reconnecting the dentry.
 252          */
 253         dwords = buflen >> 2;
 254         fh_type = exportfs_encode_fh(lower, buf, &dwords, 0);
 255         buflen = (dwords << 2);
 256
 257         fh = ERR_PTR(-EIO);
 258         if (WARN_ON(fh_type < 0) ||
 259             WARN_ON(buflen > MAX_HANDLE_SZ) ||
 260             WARN_ON(fh_type == FILEID_INVALID))
 261                 goto out;
 262
 263         BUILD_BUG_ON(MAX_HANDLE_SZ + offsetof(struct ovl_fh, fid) > 255);
 264         fh_len = offsetof(struct ovl_fh, fid) + buflen;
 265         fh = kmalloc(fh_len, GFP_KERNEL);
 266         if (!fh) {
 267                 fh = ERR_PTR(-ENOMEM);
 268                 goto out;
 269         }
 270
 271         fh->version = OVL_FH_VERSION;
 272         fh->magic = OVL_FH_MAGIC;
 273         fh->type = fh_type;
 274         fh->flags = OVL_FH_FLAG_CPU_ENDIAN;
 275         /*
 276          * When we will want to decode an overlay dentry from this handle
 277          * and all layers are on the same fs, if we get a disconncted real
 278          * dentry when we decode fid, the only way to tell if we should assign
 279          * it to upperdentry or to lowerstack is by checking this flag.
 280          */
 281         if (is_upper)
 282                 fh->flags |= OVL_FH_FLAG_PATH_UPPER;
 283         fh->len = fh_len;
 284         fh->uuid = *uuid;
 285         memcpy(fh->fid, buf, buflen);
 286
 287 out:
 288         kfree(buf);
 289         return fh;
 290 }
 291
 292 static int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
 293                           struct dentry *upper)
 294 {
 295         const struct ovl_fh *fh = NULL;
 296         int err;
 297
 298         /*
 299          * When lower layer doesn't support export operations store a 'null' fh,
 300          * so we can use the overlay.origin xattr to distignuish between a copy
 301          * up and a pure upper inode.
 302          */
 303         if (ovl_can_decode_fh(lower->d_sb)) {
 304                 fh = ovl_encode_fh(lower, false);
 305                 if (IS_ERR(fh))
 306                         return PTR_ERR(fh);
 307         }
 308
 309         /*
 310          * Do not fail when upper doesn't support xattrs.
 311          */
 312         err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh,
 313                                  fh ? fh->len : 0, 0);
 314         kfree(fh);
 315
 316         return err;
 317 }
 318
 319 struct ovl_copy_up_ctx {
 320         struct dentry *parent;
 321         struct dentry *dentry;
 322         struct path lowerpath;
 323         struct kstat stat;
 324         struct kstat pstat;
 325         const char *link;
 326         struct dentry *destdir;
 327         struct qstr destname;
 328         struct dentry *workdir;
 329         bool tmpfile;
 330         bool origin;
 331 };
 332
 333 static int ovl_link_up(struct ovl_copy_up_ctx *c)
 334 {
 335         int err;
 336         struct dentry *upper;
 337         struct dentry *upperdir = ovl_dentry_upper(c->parent);
 338         struct inode *udir = d_inode(upperdir);
 339
 340         /* Mark parent "impure" because it may now contain non-pure upper */
 341         err = ovl_set_impure(c->parent, upperdir);
 342         if (err)
 343                 return err;
 344
 345         err = ovl_set_nlink_lower(c->dentry);
 346         if (err)
 347                 return err;
 348
 349         inode_lock_nested(udir, I_MUTEX_PARENT);
 350         upper = lookup_one_len(c->dentry->d_name.name, upperdir,
 351                                c->dentry->d_name.len);
 352         err = PTR_ERR(upper);
 353         if (!IS_ERR(upper)) {
 354                 err = ovl_do_link(ovl_dentry_upper(c->dentry), udir, upper,
 355                                   true);
 356                 dput(upper);
 357
 358                 if (!err) {
 359                         /* Restore timestamps on parent (best effort) */
 360                         ovl_set_timestamps(upperdir, &c->pstat);
 361                         ovl_dentry_set_upper_alias(c->dentry);
 362                 }
 363         }
 364         inode_unlock(udir);
 365         ovl_set_nlink_upper(c->dentry);
 366
 367         return err;
 368 }
 369
 370 static int ovl_install_temp(struct ovl_copy_up_ctx *c, struct dentry *temp,
 371                             struct dentry **newdentry)
 372 {
 373         int err;
 374         struct dentry *upper;
 375         struct inode *udir = d_inode(c->destdir);
 376
 377         upper = lookup_one_len(c->destname.name, c->destdir, c->destname.len);
 378         if (IS_ERR(upper))
 379                 return PTR_ERR(upper);
 380
 381         if (c->tmpfile)
 382                 err = ovl_do_link(temp, udir, upper, true);
 383         else
 384                 err = ovl_do_rename(d_inode(c->workdir), temp, udir, upper, 0);
 385
 386         if (!err)
 387                 *newdentry = dget(c->tmpfile ? upper : temp);
 388         dput(upper);
 389
 390         return err;
 391 }
 392
 393 static int ovl_get_tmpfile(struct ovl_copy_up_ctx *c, struct dentry **tempp)
 394 {
 395         int err;
 396         struct dentry *temp;
 397         const struct cred *old_creds = NULL;
 398         struct cred *new_creds = NULL;
 399         struct cattr cattr = {
 400                 /* Can't properly set mode on creation because of the umask */
 401                 .mode = c->stat.mode & S_IFMT,
 402                 .rdev = c->stat.rdev,
 403                 .link = c->link
 404         };
 405
 406         err = security_inode_copy_up(c->dentry, &new_creds);
 407         if (err < 0)
 408                 goto out;
 409
 410         if (new_creds)
 411                 old_creds = override_creds(new_creds);
 412
 413         if (c->tmpfile) {
 414                 temp = ovl_do_tmpfile(c->workdir, c->stat.mode);
 415                 if (IS_ERR(temp))
 416                         goto temp_err;
 417         } else {
 418                 temp = ovl_lookup_temp(c->workdir);
 419                 if (IS_ERR(temp))
 420                         goto temp_err;
 421
 422                 err = ovl_create_real(d_inode(c->workdir), temp, &cattr,
 423                                       NULL, true);
 424                 if (err) {
 425                         dput(temp);
 426                         goto out;
 427                 }
 428         }
 429         err = 0;
 430         *tempp = temp;
 431 out:
 432         if (new_creds) {
 433                 revert_creds(old_creds);
 434                 put_cred(new_creds);
 435         }
 436
 437         return err;
 438
 439 temp_err:
 440         err = PTR_ERR(temp);
 441         goto out;
 442 }
 443
 444 static int ovl_copy_up_inode(struct ovl_copy_up_ctx *c, struct dentry *temp)
 445 {
 446         int err;
 447
 448         if (S_ISREG(c->stat.mode)) {
 449                 struct path upperpath;
 450
 451                 ovl_path_upper(c->dentry, &upperpath);
 452                 BUG_ON(upperpath.dentry != NULL);
 453                 upperpath.dentry = temp;
 454
 455                 err = ovl_copy_up_data(&c->lowerpath, &upperpath, c->stat.size);
 456                 if (err)
 457                         return err;
 458         }
 459
 460         err = ovl_copy_xattr(c->lowerpath.dentry, temp);
 461         if (err)
 462                 return err;
 463
 464         inode_lock(temp->d_inode);
 465         err = ovl_set_attr(temp, &c->stat);
 466         inode_unlock(temp->d_inode);
 467         if (err)
 468                 return err;
 469
 470         /*
 471          * Store identifier of lower inode in upper inode xattr to
 472          * allow lookup of the copy up origin inode.
 473          *
 474          * Don't set origin when we are breaking the association with a lower
 475          * hard link.
 476          */
 477         if (c->origin) {
 478                 err = ovl_set_origin(c->dentry, c->lowerpath.dentry, temp);
 479                 if (err)
 480                         return err;
 481         }
 482
 483         return 0;
 484 }
 485
 486 static int ovl_copy_up_locked(struct ovl_copy_up_ctx *c)
 487 {
 488         struct inode *udir = c->destdir->d_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         ovl_inode_update(d_inode(c->dentry), newdentry);
 512 out:
 513         dput(temp);
 514         return err;
 515
 516 out_cleanup:
 517         if (!c->tmpfile)
 518                 ovl_cleanup(d_inode(c->workdir), temp);
 519         goto out;
 520 }
 521
 522 /*
 523  * Copy up a single dentry
 524  *
 525  * All renames start with copy up of source if necessary.  The actual
 526  * rename will only proceed once the copy up was successful.  Copy up uses
 527  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 528  * is possible that the copy up will lock the old parent.  At that point
 529  * the file will have already been copied up anyway.
 530  */
 531 static int ovl_do_copy_up(struct ovl_copy_up_ctx *c)
 532 {
 533         int err;
 534         struct ovl_fs *ofs = c->dentry->d_sb->s_fs_info;
 535         bool indexed = false;
 536
 537         if (ovl_indexdir(c->dentry->d_sb) && !S_ISDIR(c->stat.mode) &&
 538             c->stat.nlink > 1)
 539                 indexed = true;
 540
 541         if (S_ISDIR(c->stat.mode) || c->stat.nlink == 1 || indexed)
 542                 c->origin = true;
 543
 544         if (indexed) {
 545                 c->destdir = ovl_indexdir(c->dentry->d_sb);
 546                 err = ovl_get_index_name(c->lowerpath.dentry, &c->destname);
 547                 if (err)
 548                         return err;
 549         } else {
 550                 /*
 551                  * Mark parent "impure" because it may now contain non-pure
 552                  * upper
 553                  */
 554                 err = ovl_set_impure(c->parent, c->destdir);
 555                 if (err)
 556                         return err;
 557         }
 558
 559         /* Should we copyup with O_TMPFILE or with workdir? */
 560         if (S_ISREG(c->stat.mode) && ofs->tmpfile) {
 561                 c->tmpfile = true;
 562                 err = ovl_copy_up_locked(c);
 563         } else {
 564                 err = -EIO;
 565                 if (lock_rename(c->workdir, c->destdir) != NULL) {
 566                         pr_err("overlayfs: failed to lock workdir+upperdir\n");
 567                 } else {
 568                         err = ovl_copy_up_locked(c);
 569                         unlock_rename(c->workdir, c->destdir);
 570                 }
 571         }
 572
 573         if (indexed) {
 574                 if (!err)
 575                         ovl_set_flag(OVL_INDEX, d_inode(c->dentry));
 576                 kfree(c->destname.name);
 577         } else if (!err) {
 578                 struct inode *udir = d_inode(c->destdir);
 579
 580                 /* Restore timestamps on parent (best effort) */
 581                 inode_lock(udir);
 582                 ovl_set_timestamps(c->destdir, &c->pstat);
 583                 inode_unlock(udir);
 584
 585                 ovl_dentry_set_upper_alias(c->dentry);
 586         }
 587
 588         return err;
 589 }
 590
 591 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 592                            int flags)
 593 {
 594         int err;
 595         DEFINE_DELAYED_CALL(done);
 596         struct path parentpath;
 597         struct ovl_copy_up_ctx ctx = {
 598                 .parent = parent,
 599                 .dentry = dentry,
 600                 .workdir = ovl_workdir(dentry),
 601         };
 602
 603         if (WARN_ON(!ctx.workdir))
 604                 return -EROFS;
 605
 606         ovl_path_lower(dentry, &ctx.lowerpath);
 607         err = vfs_getattr(&ctx.lowerpath, &ctx.stat,
 608                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 609         if (err)
 610                 return err;
 611
 612         ovl_path_upper(parent, &parentpath);
 613         ctx.destdir = parentpath.dentry;
 614         ctx.destname = dentry->d_name;
 615
 616         err = vfs_getattr(&parentpath, &ctx.pstat,
 617                           STATX_ATIME | STATX_MTIME, AT_STATX_SYNC_AS_STAT);
 618         if (err)
 619                 return err;
 620
 621         /* maybe truncate regular file. this has no effect on dirs */
 622         if (flags & O_TRUNC)
 623                 ctx.stat.size = 0;
 624
 625         if (S_ISLNK(ctx.stat.mode)) {
 626                 ctx.link = vfs_get_link(ctx.lowerpath.dentry, &done);
 627                 if (IS_ERR(ctx.link))
 628                         return PTR_ERR(ctx.link);
 629         }
 630         ovl_do_check_copy_up(ctx.lowerpath.dentry);
 631
 632         err = ovl_copy_up_start(dentry);
 633         /* err < 0: interrupted, err > 0: raced with another copy-up */
 634         if (unlikely(err)) {
 635                 if (err > 0)
 636                         err = 0;
 637         } else {
 638                 if (!ovl_dentry_upper(dentry))
 639                         err = ovl_do_copy_up(&ctx);
 640                 if (!err && !ovl_dentry_has_upper_alias(dentry))
 641                         err = ovl_link_up(&ctx);
 642                 ovl_copy_up_end(dentry);
 643         }
 644         do_delayed_call(&done);
 645
 646         return err;
 647 }
 648
 649 int ovl_copy_up_flags(struct dentry *dentry, int flags)
 650 {
 651         int err = 0;
 652         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 653
 654         while (!err) {
 655                 struct dentry *next;
 656                 struct dentry *parent;
 657
 658                 /*
 659                  * Check if copy-up has happened as well as for upper alias (in
 660                  * case of hard links) is there.
 661                  *
 662                  * Both checks are lockless:
 663                  *  - false negatives: will recheck under oi->lock
 664                  *  - false positives:
 665                  *    + ovl_dentry_upper() uses memory barriers to ensure the
 666                  *      upper dentry is up-to-date
 667                  *    + ovl_dentry_has_upper_alias() relies on locking of
 668                  *      upper parent i_rwsem to prevent reordering copy-up
 669                  *      with rename.
 670                  */
 671                 if (ovl_dentry_upper(dentry) &&
 672                     ovl_dentry_has_upper_alias(dentry))
 673                         break;
 674
 675                 next = dget(dentry);
 676                 /* find the topmost dentry not yet copied up */
 677                 for (;;) {
 678                         parent = dget_parent(next);
 679
 680                         if (ovl_dentry_upper(parent))
 681                                 break;
 682
 683                         dput(next);
 684                         next = parent;
 685                 }
 686
 687                 err = ovl_copy_up_one(parent, next, flags);
 688
 689                 dput(parent);
 690                 dput(next);
 691         }
 692         revert_creds(old_cred);
 693
 694         return err;
 695 }
 696
 697 int ovl_copy_up(struct dentry *dentry)
 698 {
 699         return ovl_copy_up_flags(dentry, 0);
 700 }