4 * Capture a WIM image directly from an NTFS volume using libntfs-3g. We capture
5 * everything we can, including security data and alternate data streams.
9 * Copyright (C) 2012, 2013, 2014 Eric Biggers
11 * This file is part of wimlib, a library for working with WIM files.
13 * wimlib is free software; you can redistribute it and/or modify it under the
14 * terms of the GNU General Public License as published by the Free
15 * Software Foundation; either version 3 of the License, or (at your option)
18 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20 * A PARTICULAR PURPOSE. See the GNU General Public License for more
23 * You should have received a copy of the GNU General Public License
24 * along with wimlib; if not, see http://www.gnu.org/licenses/.
40 #include <ntfs-3g/attrib.h>
41 #include <ntfs-3g/reparse.h>
42 #include <ntfs-3g/security.h>
43 #include <ntfs-3g/volume.h>
45 #include "wimlib/capture.h"
46 #include "wimlib/dentry.h"
47 #include "wimlib/encoding.h"
48 #include "wimlib/endianness.h"
49 #include "wimlib/error.h"
50 #include "wimlib/lookup_table.h"
51 #include "wimlib/ntfs_3g.h"
52 #include "wimlib/paths.h"
53 #include "wimlib/security.h"
55 static inline ntfschar *
56 attr_record_name(ATTR_RECORD *ar)
58 return (ntfschar*)((u8*)ar + le16_to_cpu(ar->name_offset));
62 open_ntfs_attr(ntfs_inode *ni, struct ntfs_location *loc)
66 na = ntfs_attr_open(ni,
67 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
69 loc->stream_name_nchars);
71 ERROR_WITH_ERRNO("Failed to open attribute of \"%"TS"\" in "
72 "NTFS volume", loc->path);
78 read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte, u64 size,
79 consume_data_callback_t cb, void *cb_ctx)
81 struct ntfs_location *loc = lte->ntfs_loc;
82 ntfs_volume *vol = loc->ntfs_vol;
90 ni = ntfs_pathname_to_inode(vol, NULL, loc->path);
92 ERROR_WITH_ERRNO("Can't find NTFS inode for \"%"TS"\"", loc->path);
93 ret = WIMLIB_ERR_NTFS_3G;
97 na = open_ntfs_attr(ni, loc);
99 ret = WIMLIB_ERR_NTFS_3G;
100 goto out_close_ntfs_inode;
103 pos = (loc->is_reparse_point) ? 8 : 0;
104 bytes_remaining = size;
105 while (bytes_remaining) {
106 s64 to_read = min(bytes_remaining, sizeof(buf));
107 if (ntfs_attr_pread(na, pos, to_read, buf) != to_read) {
108 ERROR_WITH_ERRNO("Error reading \"%"TS"\"", loc->path);
109 ret = WIMLIB_ERR_NTFS_3G;
110 goto out_close_ntfs_attr;
113 bytes_remaining -= to_read;
114 ret = cb(buf, to_read, cb_ctx);
116 goto out_close_ntfs_attr;
121 out_close_ntfs_inode:
122 ntfs_inode_close(ni);
128 read_reparse_tag(ntfs_inode *ni, struct ntfs_location *loc,
129 u32 *reparse_tag_ret)
135 na = open_ntfs_attr(ni, loc);
137 ret = WIMLIB_ERR_NTFS_3G;
141 if (ntfs_attr_pread(na, 0, sizeof(reparse_tag),
142 &reparse_tag) != sizeof(reparse_tag))
144 ERROR_WITH_ERRNO("Error reading reparse data");
145 ret = WIMLIB_ERR_NTFS_3G;
146 goto out_close_ntfs_attr;
148 *reparse_tag_ret = le32_to_cpu(reparse_tag);
149 DEBUG("ReparseTag = %#x", *reparse_tag_ret);
158 /* Load the streams from a file or reparse point in the NTFS volume */
160 capture_ntfs_streams(struct wim_inode *inode,
164 struct list_head *unhashed_streams,
168 ntfs_attr_search_ctx *actx;
169 struct ntfs_location *ntfs_loc;
171 struct wim_lookup_table_entry *lte;
173 DEBUG("Capturing NTFS data streams from `%s'", path);
175 /* Get context to search the streams of the NTFS file. */
176 actx = ntfs_attr_get_search_ctx(ni, NULL);
178 ERROR_WITH_ERRNO("Cannot get NTFS attribute search "
179 "context for \"%s\"", path);
180 return WIMLIB_ERR_NTFS_3G;
183 /* Capture each data stream or reparse data stream. */
184 while (!ntfs_attr_lookup(type, NULL, 0,
185 CASE_SENSITIVE, 0, NULL, 0, actx))
187 u64 data_size = ntfs_get_attribute_value_length(actx->attr);
188 u64 name_length = actx->attr->name_length;
191 if (data_size == 0) {
192 /* Empty stream. No lookup table entry is needed. */
196 ntfs_loc = CALLOC(1, sizeof(*ntfs_loc));
198 ret = WIMLIB_ERR_NOMEM;
201 ntfs_loc->ntfs_vol = vol;
202 ntfs_loc->path = memdup(path, path_len + 1);
203 if (!ntfs_loc->path) {
204 ret = WIMLIB_ERR_NOMEM;
205 goto out_free_ntfs_loc;
208 ntfs_loc->stream_name = memdup(attr_record_name(actx->attr),
210 if (!ntfs_loc->stream_name) {
211 ret = WIMLIB_ERR_NOMEM;
212 goto out_free_ntfs_loc;
214 ntfs_loc->stream_name_nchars = name_length;
217 lte = new_lookup_table_entry();
219 ret = WIMLIB_ERR_NOMEM;
220 goto out_free_ntfs_loc;
222 lte->resource_location = RESOURCE_IN_NTFS_VOLUME;
223 lte->ntfs_loc = ntfs_loc;
225 if (type == AT_REPARSE_POINT) {
227 ERROR("Invalid reparse data on \"%s\" "
228 "(only %u bytes)!", path, (unsigned)data_size);
229 ret = WIMLIB_ERR_NTFS_3G;
232 lte->ntfs_loc->is_reparse_point = true;
233 lte->size = data_size - 8;
234 ret = read_reparse_tag(ni, lte->ntfs_loc,
235 &inode->i_reparse_tag);
239 lte->ntfs_loc->is_reparse_point = false;
240 lte->size = data_size;
243 if (name_length == 0) {
244 /* Unnamed data stream. Put the reference to it in the
248 if (!(inode->i_attributes &
249 FILE_ATTRIBUTE_REPARSE_POINT))
251 WARNING("Found two un-named "
252 "data streams for \"%s\" "
253 "(sizes = %"PRIu64", "
259 free_lookup_table_entry(lte);
267 /* Named data stream. Put the reference to it in the
268 * alternate data stream entries */
269 struct wim_ads_entry *new_ads_entry;
271 new_ads_entry = inode_add_ads_utf16le(inode,
272 attr_record_name(actx->attr),
274 if (!new_ads_entry) {
275 ret = WIMLIB_ERR_NOMEM;
278 wimlib_assert(new_ads_entry->stream_name_nbytes == name_length * 2);
279 stream_id = new_ads_entry->stream_id;
280 new_ads_entry->lte = lte;
283 add_unhashed_stream(lte, inode,
284 stream_id, unhashed_streams);
287 if (errno == ENOENT) {
290 ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path);
291 ret = WIMLIB_ERR_NTFS_3G;
295 free_lookup_table_entry(lte);
298 FREE(ntfs_loc->path);
299 FREE(ntfs_loc->stream_name);
303 ntfs_attr_put_search_ctx(actx);
305 DEBUG("Successfully captured NTFS streams from \"%s\"", path);
307 ERROR("Failed to capture NTFS streams from \"%s\"", path);
311 /* Binary tree that maps NTFS inode numbers to DOS names */
312 struct dos_name_map {
313 struct avl_tree_node *root;
316 struct dos_name_node {
317 struct avl_tree_node index_node;
323 #define DOS_NAME_NODE(avl_node) \
324 avl_tree_entry(avl_node, struct dos_name_node, index_node)
327 _avl_cmp_by_ntfs_ino(const struct avl_tree_node *n1,
328 const struct avl_tree_node *n2)
330 return cmp_u64(DOS_NAME_NODE(n1)->ntfs_ino,
331 DOS_NAME_NODE(n2)->ntfs_ino);
334 /* Inserts a new DOS name into the map */
336 insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name,
337 size_t name_nbytes, le64 ntfs_ino)
339 struct dos_name_node *new_node;
341 DEBUG("DOS name_len = %zu", name_nbytes);
342 new_node = MALLOC(sizeof(struct dos_name_node));
346 /* DOS names are supposed to be 12 characters max (that's 24 bytes,
347 * assuming 2-byte ntfs characters) */
348 wimlib_assert(name_nbytes <= sizeof(new_node->dos_name));
350 /* Initialize the DOS name, DOS name length, and NTFS inode number of
351 * the search tree node */
352 memcpy(new_node->dos_name, dos_name, name_nbytes);
353 new_node->name_nbytes = name_nbytes;
354 new_node->ntfs_ino = ntfs_ino;
356 /* Insert the search tree node */
357 if (avl_tree_insert(&map->root, &new_node->index_node,
358 _avl_cmp_by_ntfs_ino))
360 /* This should be impossible since an NTFS inode cannot
361 * have multiple DOS names, and we only should get each
362 * DOS name entry once from the ntfs_readdir() calls. */
363 ERROR("NTFS inode %"PRIu64" has multiple DOS names",
364 le64_to_cpu(ntfs_ino));
368 DEBUG("Inserted DOS name for inode %"PRIu64, le64_to_cpu(ntfs_ino));
372 /* Returns a structure that contains the DOS name and its length for an NTFS
373 * inode, or NULL if the inode has no DOS name. */
374 static struct dos_name_node *
375 lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino)
377 struct dos_name_node dummy;
378 struct avl_tree_node *res;
380 dummy.ntfs_ino = cpu_to_le64(ntfs_ino);
382 res = avl_tree_lookup_node(map->root, &dummy.index_node,
383 _avl_cmp_by_ntfs_ino);
386 return DOS_NAME_NODE(res);
390 set_dentry_dos_name(struct wim_dentry *dentry, const struct dos_name_map *map)
392 const struct dos_name_node *node;
394 if (dentry->is_win32_name) {
395 node = lookup_dos_name(map, dentry->d_inode->i_ino);
397 dentry->short_name = utf16le_dupz((const utf16lechar *)node->dos_name,
399 if (!dentry->short_name)
400 return WIMLIB_ERR_NOMEM;
401 dentry->short_name_nbytes = node->name_nbytes;
402 DEBUG("Assigned DOS name to ino %"PRIu64,
403 dentry->d_inode->i_ino);
405 WARNING("NTFS inode %"PRIu64" has Win32 name with no "
406 "corresponding DOS name",
407 dentry->d_inode->i_ino);
414 free_dos_name_tree(struct avl_tree_node *node) {
416 free_dos_name_tree(node->left);
417 free_dos_name_tree(node->right);
418 FREE(DOS_NAME_NODE(node));
423 destroy_dos_name_map(struct dos_name_map *map)
425 free_dos_name_tree(map->root);
429 struct wim_dentry *parent;
432 struct dos_name_map *dos_name_map;
434 struct add_image_params *params;
438 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
443 ntfs_volume *ntfs_vol,
444 struct add_image_params *params);
447 wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
448 const int name_nchars, const int name_type,
449 const s64 pos, const MFT_REF mref,
450 const unsigned dt_type)
452 struct readdir_ctx *ctx;
453 size_t mbs_name_nbytes;
455 struct wim_dentry *child;
458 size_t name_nbytes = name_nchars * sizeof(ntfschar);
461 if (name_type & FILE_NAME_DOS) {
462 /* If this is the entry for a DOS name, store it for later. */
463 ret = insert_dos_name(ctx->dos_name_map, name,
464 name_nbytes, mref & MFT_REF_MASK_CPU);
466 /* Return now if an error occurred or if this is just a DOS name
467 * and not a Win32+DOS name. */
468 if (ret != 0 || name_type == FILE_NAME_DOS)
471 ret = utf16le_to_tstr(name, name_nbytes,
472 &mbs_name, &mbs_name_nbytes);
476 if (mbs_name[0] == '.' &&
477 (mbs_name[1] == '\0' ||
478 (mbs_name[1] == '.' && mbs_name[2] == '\0'))) {
481 * note: name_type is POSIX for these, so DOS names will not
482 * have been inserted for them. */
484 goto out_free_mbs_name;
487 /* Open the inode for this directory entry and recursively capture the
488 * directory tree rooted at it */
489 ntfs_inode *ni = ntfs_inode_open(ctx->vol, mref);
491 /* XXX This used to be treated as an error, but NTFS-3g seemed
492 * to be unable to read some inodes on a Windows 8 image for
494 WARNING_WITH_ERRNO("Failed to open NTFS file \"%s/%s\"",
495 ctx->path, mbs_name);
497 goto out_free_mbs_name;
499 path_len = ctx->path_len;
501 ctx->path[path_len++] = '/';
502 memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1);
503 path_len += mbs_name_nbytes;
505 ret = build_dentry_tree_ntfs_recursive(&child, ni, ctx->path,
507 ctx->vol, ctx->params);
508 path_len -= mbs_name_nbytes + 1;
510 dentry_add_child(ctx->parent, child);
511 ntfs_inode_close(ni);
515 ctx->path[ctx->path_len] = '\0';
519 /* Recursively build a WIM dentry tree corresponding to an NTFS volume.
520 * At the same time, update the WIM lookup table with lookup table entries for
521 * the NTFS streams, and build an array of security descriptors.
524 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
530 struct add_image_params *params)
534 struct wim_dentry *root = NULL;
535 struct wim_inode *inode = NULL;
537 if (should_exclude_path(path, path_len, params->config)) {
538 /* Exclude a file or directory tree based on the capture
539 * configuration file. */
543 /* Get file attributes */
544 ret = ntfs_get_ntfs_attrib(ni, (char*)&attributes, sizeof(attributes));
545 if (ret != sizeof(attributes)) {
546 ERROR_WITH_ERRNO("Failed to get NTFS attributes from \"%s\"", path);
547 ret = WIMLIB_ERR_NTFS_3G;
551 if ((attributes & (FILE_ATTRIBUTE_DIRECTORY |
552 FILE_ATTRIBUTE_ENCRYPTED)) == FILE_ATTRIBUTE_ENCRYPTED)
554 if (params->add_flags & WIMLIB_ADD_FLAG_NO_UNSUPPORTED_EXCLUDE)
556 ERROR("Can't archive unsupported encrypted file \"%s\"", path);
557 ret = WIMLIB_ERR_UNSUPPORTED_FILE;
560 params->progress.scan.cur_path = path;
561 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_UNSUPPORTED, NULL);
565 /* Create a WIM dentry with an associated inode, which may be shared */
566 ret = inode_table_new_dentry(params->inode_table,
567 path_basename_with_len(path, path_len),
568 ni->mft_no, 0, false, &root);
572 if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */
573 root->is_win32_name = 1;
575 inode = root->d_inode;
577 if (inode->i_nlink > 1) {
578 /* Shared inode; nothing more to do */
582 inode->i_creation_time = le64_to_cpu(ni->creation_time);
583 inode->i_last_write_time = le64_to_cpu(ni->last_data_change_time);
584 inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
585 inode->i_attributes = le32_to_cpu(attributes);
586 inode->i_resolved = 1;
588 /* Capture streams. */
590 if (attributes & FILE_ATTR_REPARSE_POINT) {
591 /* Capture reparse data stream. */
592 ret = capture_ntfs_streams(inode, ni, path, path_len,
593 params->unhashed_streams,
594 vol, AT_REPARSE_POINT);
599 /* Capture data streams.
601 * Directories should not have an unnamed data stream, but they may have
602 * named data streams.
604 * Reparse points may have an unnamed data stream (which will be ignored
605 * in favor of the reparse data stream), and they also may have named
608 * Regular files can have an unnamed data stream as well as named data
610 ret = capture_ntfs_streams(inode, ni, path, path_len,
611 params->unhashed_streams, vol, AT_DATA);
615 if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) {
617 /* Recurse to directory children */
619 struct dos_name_map dos_name_map = { .root = NULL };
620 struct readdir_ctx ctx = {
623 .path_len = path_len,
624 .dos_name_map = &dos_name_map,
628 ret = ntfs_readdir(ni, &pos, &ctx, wim_ntfs_capture_filldir);
630 ERROR_WITH_ERRNO("Error reading directory \"%s\"", path);
631 ret = WIMLIB_ERR_NTFS_3G;
633 struct wim_dentry *child;
636 for_dentry_child(child, root) {
637 ret = set_dentry_dos_name(child, &dos_name_map);
642 destroy_dos_name_map(&dos_name_map);
646 path[path_len] = '\0';
648 /* Reparse-point fixups are a no-op because in NTFS-3g capture mode we
649 * only allow capturing an entire volume. */
650 if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX &&
651 inode_is_symlink(inode))
652 inode->i_not_rpfixed = 0;
654 if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
655 struct SECURITY_CONTEXT sec_ctx;
659 /* Get security descriptor */
660 memset(&sec_ctx, 0, sizeof(sec_ctx));
665 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, sizeof(_sd));
666 if (ret > sizeof(_sd)) {
668 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, ret);
671 inode->i_security_id = sd_set_add_sd(params->sd_set,
673 if (inode->i_security_id == -1) {
674 ERROR("Out of memory");
675 ret = WIMLIB_ERR_NOMEM;
678 DEBUG("Added security ID = %u for `%s'",
679 inode->i_security_id, path);
681 } else if (ret < 0) {
682 ERROR_WITH_ERRNO("Failed to get security information from "
684 ret = WIMLIB_ERR_NTFS_3G;
686 inode->i_security_id = -1;
687 DEBUG("No security ID for `%s'", path);
694 params->progress.scan.cur_path = path;
696 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL);
698 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK, inode);
703 free_dentry_tree(root, params->lookup_table);
709 do_ntfs_umount(struct _ntfs_volume *vol)
711 DEBUG("Unmounting NTFS volume");
712 if (ntfs_umount(vol, FALSE))
713 return WIMLIB_ERR_NTFS_3G;
719 build_dentry_tree_ntfs(struct wim_dentry **root_p,
721 struct add_image_params *params)
727 DEBUG("Mounting NTFS volume `%s' read-only", device);
729 /* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
732 * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
733 * NTFS_MNT_RDONLY is an enumerated constant. Also, the NTFS-3g headers don't
734 * seem to contain any explicit version information. So we have to rely on a
735 * test done at configure time to detect whether NTFS_MNT_RDONLY should be used.
737 #ifdef HAVE_NTFS_MNT_RDONLY
739 vol = ntfs_mount(device, NTFS_MNT_RDONLY);
740 #elif defined(MS_RDONLY)
741 /* NTFS-3g 2011, 2012 */
742 vol = ntfs_mount(device, MS_RDONLY);
744 #error "Can't find NTFS_MNT_RDONLY or MS_RDONLY flags"
747 ERROR_WITH_ERRNO("Failed to mount NTFS volume `%s' read-only",
749 return WIMLIB_ERR_NTFS_3G;
751 ntfs_open_secure(vol);
753 /* We don't want to capture the special NTFS files such as $Bitmap. Not
754 * to be confused with "hidden" or "system" files which are real files
755 * that we do need to capture. */
756 NVolClearShowSysFiles(vol);
758 DEBUG("Opening root NTFS dentry");
759 root_ni = ntfs_inode_open(vol, FILE_root);
761 ERROR_WITH_ERRNO("Failed to open root inode of NTFS volume "
763 ret = WIMLIB_ERR_NTFS_3G;
767 /* Currently we assume that all the paths fit into this length and there
768 * is no check for overflow. */
769 char *path = MALLOC(32768);
771 ERROR("Could not allocate memory for NTFS pathname");
772 ret = WIMLIB_ERR_NOMEM;
778 ret = build_dentry_tree_ntfs_recursive(root_p, root_ni, path, 1,
779 FILE_NAME_POSIX, vol, params);
782 ntfs_inode_close(root_ni);
784 ntfs_index_ctx_put(vol->secure_xsii);
785 ntfs_index_ctx_put(vol->secure_xsdh);
786 ntfs_inode_close(vol->secure_ni);
789 if (do_ntfs_umount(vol)) {
790 ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'",
794 /* We need to leave the NTFS volume mounted so that we can read
795 * the NTFS files again when we are actually writing the WIM */
796 *(ntfs_volume**)params->extra_arg = vol;
800 #endif /* WITH_NTFS_3G */