4 * Capture a WIM image directly from a NTFS volume using libntfs-3g. We capture
5 * everything we can, including security data and alternate data streams.
9 * Copyright (C) 2012, 2013 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,
80 consume_data_callback_t cb,
84 struct ntfs_location *loc = lte->ntfs_loc;
85 ntfs_volume *vol = loc->ntfs_vol;
93 ni = ntfs_pathname_to_inode(vol, NULL, loc->path);
95 ERROR_WITH_ERRNO("Can't find NTFS inode for \"%"TS"\"", loc->path);
96 ret = WIMLIB_ERR_NTFS_3G;
100 na = open_ntfs_attr(ni, loc);
102 ret = WIMLIB_ERR_NTFS_3G;
103 goto out_close_ntfs_inode;
107 out_buf = alloca(WIM_CHUNK_SIZE);
109 out_buf = ctx_or_buf;
110 pos = (loc->is_reparse_point) ? 8 : 0;
111 bytes_remaining = size;
112 while (bytes_remaining) {
113 s64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
114 if (ntfs_attr_pread(na, pos, to_read, out_buf) != to_read) {
115 ERROR_WITH_ERRNO("Error reading \"%"TS"\"", loc->path);
116 ret = WIMLIB_ERR_NTFS_3G;
117 goto out_close_ntfs_attr;
120 bytes_remaining -= to_read;
122 ret = cb(out_buf, to_read, ctx_or_buf);
124 goto out_close_ntfs_attr;
132 out_close_ntfs_inode:
133 ntfs_inode_close(ni);
139 read_reparse_tag(ntfs_inode *ni, struct ntfs_location *loc,
140 u32 *reparse_tag_ret)
146 na = open_ntfs_attr(ni, loc);
148 ret = WIMLIB_ERR_NTFS_3G;
152 if (ntfs_attr_pread(na, 0, sizeof(reparse_tag),
153 &reparse_tag) != sizeof(reparse_tag))
155 ERROR_WITH_ERRNO("Error reading reparse data");
156 ret = WIMLIB_ERR_NTFS_3G;
157 goto out_close_ntfs_attr;
159 *reparse_tag_ret = le32_to_cpu(reparse_tag);
160 DEBUG("ReparseTag = %#x", *reparse_tag_ret);
169 /* Load the streams from a file or reparse point in the NTFS volume into the WIM
172 capture_ntfs_streams(struct wim_inode *inode,
176 struct wim_lookup_table *lookup_table,
180 ntfs_attr_search_ctx *actx;
181 struct ntfs_location *ntfs_loc;
183 struct wim_lookup_table_entry *lte;
185 DEBUG2("Capturing NTFS data streams from `%s'", path);
187 /* Get context to search the streams of the NTFS file. */
188 actx = ntfs_attr_get_search_ctx(ni, NULL);
190 ERROR_WITH_ERRNO("Cannot get NTFS attribute search "
191 "context for \"%s\"", path);
192 return WIMLIB_ERR_NTFS_3G;
195 /* Capture each data stream or reparse data stream. */
196 while (!ntfs_attr_lookup(type, NULL, 0,
197 CASE_SENSITIVE, 0, NULL, 0, actx))
199 u64 data_size = ntfs_get_attribute_value_length(actx->attr);
200 u64 name_length = actx->attr->name_length;
203 if (data_size == 0) {
204 /* Empty stream. No lookup table entry is needed. */
208 ntfs_loc = CALLOC(1, sizeof(*ntfs_loc));
210 ret = WIMLIB_ERR_NOMEM;
213 ntfs_loc->ntfs_vol = vol;
214 ntfs_loc->path = memdup(path, path_len + 1);
215 if (!ntfs_loc->path) {
216 ret = WIMLIB_ERR_NOMEM;
217 goto out_free_ntfs_loc;
220 ntfs_loc->stream_name = memdup(attr_record_name(actx->attr),
222 if (!ntfs_loc->stream_name) {
223 ret = WIMLIB_ERR_NOMEM;
224 goto out_free_ntfs_loc;
226 ntfs_loc->stream_name_nchars = name_length;
229 lte = new_lookup_table_entry();
231 ret = WIMLIB_ERR_NOMEM;
232 goto out_free_ntfs_loc;
234 lte->resource_location = RESOURCE_IN_NTFS_VOLUME;
235 lte->ntfs_loc = ntfs_loc;
237 if (type == AT_REPARSE_POINT) {
239 ERROR("Invalid reparse data on \"%s\" "
240 "(only %u bytes)!", path, (unsigned)data_size);
241 ret = WIMLIB_ERR_NTFS_3G;
244 lte->ntfs_loc->is_reparse_point = true;
245 lte->resource_entry.original_size = data_size - 8;
246 ret = read_reparse_tag(ni, lte->ntfs_loc,
247 &inode->i_reparse_tag);
251 lte->ntfs_loc->is_reparse_point = false;
252 lte->resource_entry.original_size = data_size;
255 if (name_length == 0) {
256 /* Unnamed data stream. Put the reference to it in the
260 ERROR("Found two un-named data streams for \"%s\" "
261 "(sizes = %"PRIu64", %"PRIu64")",
262 path, wim_resource_size(inode->i_lte),
263 wim_resource_size(lte));
264 ret = WIMLIB_ERR_NTFS_3G;
272 /* Named data stream. Put the reference to it in the
273 * alternate data stream entries */
274 struct wim_ads_entry *new_ads_entry;
276 new_ads_entry = inode_add_ads_utf16le(inode,
277 attr_record_name(actx->attr),
279 if (!new_ads_entry) {
280 ret = WIMLIB_ERR_NOMEM;
283 wimlib_assert(new_ads_entry->stream_name_nbytes == name_length * 2);
284 stream_id = new_ads_entry->stream_id;
285 new_ads_entry->lte = lte;
288 lookup_table_insert_unhashed(lookup_table, lte,
292 if (errno == ENOENT) {
295 ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path);
296 ret = WIMLIB_ERR_NTFS_3G;
300 free_lookup_table_entry(lte);
303 FREE(ntfs_loc->path);
304 FREE(ntfs_loc->stream_name);
308 ntfs_attr_put_search_ctx(actx);
310 DEBUG2("Successfully captured NTFS streams from \"%s\"", path);
312 ERROR("Failed to capture NTFS streams from \"%s\"", path);
316 /* Red-black tree that maps NTFS inode numbers to DOS names */
317 struct dos_name_map {
318 struct rb_root rb_root;
321 struct dos_name_node {
322 struct rb_node rb_node;
328 /* Inserts a new DOS name into the map */
330 insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name,
331 size_t name_nbytes, le64 ntfs_ino)
333 struct dos_name_node *new_node;
335 struct rb_root *root;
336 struct rb_node *rb_parent;
338 DEBUG("DOS name_len = %zu", name_nbytes);
339 new_node = MALLOC(sizeof(struct dos_name_node));
343 /* DOS names are supposed to be 12 characters max (that's 24 bytes,
344 * assuming 2-byte ntfs characters) */
345 wimlib_assert(name_nbytes <= sizeof(new_node->dos_name));
347 /* Initialize the DOS name, DOS name length, and NTFS inode number of
348 * the red-black tree node */
349 memcpy(new_node->dos_name, dos_name, name_nbytes);
350 new_node->name_nbytes = name_nbytes;
351 new_node->ntfs_ino = ntfs_ino;
353 /* Insert the red-black tree node */
354 root = &map->rb_root;
358 struct dos_name_node *this;
360 this = container_of(*p, struct dos_name_node, rb_node);
362 if (new_node->ntfs_ino < this->ntfs_ino)
363 p = &((*p)->rb_left);
364 else if (new_node->ntfs_ino > this->ntfs_ino)
365 p = &((*p)->rb_right);
367 /* This should be impossible since a NTFS inode cannot
368 * have multiple DOS names, and we only should get each
369 * DOS name entry once from the ntfs_readdir() calls. */
370 ERROR("NTFS inode %"PRIu64" has multiple DOS names",
371 le64_to_cpu(ntfs_ino));
375 rb_link_node(&new_node->rb_node, rb_parent, p);
376 rb_insert_color(&new_node->rb_node, root);
377 DEBUG("Inserted DOS name for inode %"PRIu64, le64_to_cpu(ntfs_ino));
381 /* Returns a structure that contains the DOS name and its length for a NTFS
382 * inode, or NULL if the inode has no DOS name. */
383 static struct dos_name_node *
384 lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino)
386 struct rb_node *node = map->rb_root.rb_node;
388 struct dos_name_node *this;
389 this = container_of(node, struct dos_name_node, rb_node);
390 if (ntfs_ino < this->ntfs_ino)
391 node = node->rb_left;
392 else if (ntfs_ino > this->ntfs_ino)
393 node = node->rb_right;
401 set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
403 const struct dos_name_map *map = arg;
404 const struct dos_name_node *node;
406 if (dentry->is_win32_name) {
407 node = lookup_dos_name(map, dentry->d_inode->i_ino);
409 dentry->short_name = MALLOC(node->name_nbytes + 2);
410 if (!dentry->short_name)
411 return WIMLIB_ERR_NOMEM;
412 memcpy(dentry->short_name, node->dos_name,
414 dentry->short_name[node->name_nbytes / 2] = 0;
415 dentry->short_name_nbytes = node->name_nbytes;
416 DEBUG("Assigned DOS name to ino %"PRIu64,
417 dentry->d_inode->i_ino);
419 WARNING("NTFS inode %"PRIu64" has Win32 name with no "
420 "corresponding DOS name",
421 dentry->d_inode->i_ino);
428 free_dos_name_tree(struct rb_node *node) {
430 free_dos_name_tree(node->rb_left);
431 free_dos_name_tree(node->rb_right);
432 FREE(container_of(node, struct dos_name_node, rb_node));
437 destroy_dos_name_map(struct dos_name_map *map)
439 free_dos_name_tree(map->rb_root.rb_node);
443 struct wim_dentry *parent;
446 struct dos_name_map *dos_name_map;
448 struct add_image_params *params;
452 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
457 ntfs_volume *ntfs_vol,
458 struct add_image_params *params);
461 wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
462 const int name_nchars, const int name_type,
463 const s64 pos, const MFT_REF mref,
464 const unsigned dt_type)
466 struct readdir_ctx *ctx;
467 size_t mbs_name_nbytes;
469 struct wim_dentry *child;
472 size_t name_nbytes = name_nchars * sizeof(ntfschar);
475 if (name_type & FILE_NAME_DOS) {
476 /* If this is the entry for a DOS name, store it for later. */
477 ret = insert_dos_name(ctx->dos_name_map, name,
478 name_nbytes, mref & MFT_REF_MASK_CPU);
480 /* Return now if an error occurred or if this is just a DOS name
481 * and not a Win32+DOS name. */
482 if (ret != 0 || name_type == FILE_NAME_DOS)
485 ret = utf16le_to_tstr(name, name_nbytes,
486 &mbs_name, &mbs_name_nbytes);
490 if (mbs_name[0] == '.' &&
491 (mbs_name[1] == '\0' ||
492 (mbs_name[1] == '.' && mbs_name[2] == '\0'))) {
495 * note: name_type is POSIX for these, so DOS names will not
496 * have been inserted for them. */
498 goto out_free_mbs_name;
501 /* Open the inode for this directory entry and recursively capture the
502 * directory tree rooted at it */
503 ntfs_inode *ni = ntfs_inode_open(ctx->vol, mref);
505 /* XXX This used to be treated as an error, but NTFS-3g seemed
506 * to be unable to read some inodes on a Windows 8 image for
508 WARNING_WITH_ERRNO("Failed to open NTFS file \"%s/%s\"",
509 ctx->path, mbs_name);
511 goto out_free_mbs_name;
513 path_len = ctx->path_len;
515 ctx->path[path_len++] = '/';
516 memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1);
517 path_len += mbs_name_nbytes;
519 ret = build_dentry_tree_ntfs_recursive(&child, ni, ctx->path,
521 ctx->vol, ctx->params);
522 path_len -= mbs_name_nbytes + 1;
524 dentry_add_child(ctx->parent, child);
525 ntfs_inode_close(ni);
529 ctx->path[ctx->path_len] = '\0';
533 /* Recursively build a WIM dentry tree corresponding to a NTFS volume.
534 * At the same time, update the WIM lookup table with lookup table entries for
535 * the NTFS streams, and build an array of security descriptors.
538 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
544 struct add_image_params *params)
548 struct wim_dentry *root;
549 struct wim_inode *inode;
550 ATTR_TYPES stream_type;
552 if (exclude_path(path, path_len, params->config, false)) {
553 /* Exclude a file or directory tree based on the capture
554 * configuration file */
555 if ((params->add_flags & WIMLIB_ADD_FLAG_EXCLUDE_VERBOSE)
556 && params->progress_func)
558 union wimlib_progress_info info;
559 info.scan.cur_path = path;
560 info.scan.status = WIMLIB_SCAN_DENTRY_EXCLUDED;
561 params->progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info);
568 /* Get file attributes */
569 ret = ntfs_get_ntfs_attrib(ni, (char*)&attributes, sizeof(attributes));
570 if (ret != sizeof(attributes)) {
571 ERROR_WITH_ERRNO("Failed to get NTFS attributes from \"%s\"", path);
572 return WIMLIB_ERR_NTFS_3G;
575 if ((attributes & (FILE_ATTRIBUTE_DIRECTORY |
576 FILE_ATTRIBUTE_ENCRYPTED)) == FILE_ATTRIBUTE_ENCRYPTED)
578 if (params->add_flags & WIMLIB_ADD_FLAG_NO_UNSUPPORTED_EXCLUDE)
580 ERROR("Can't archive unsupported encrypted file \"%s\"", path);
581 return WIMLIB_ERR_UNSUPPORTED_FILE;
583 if ((params->add_flags & WIMLIB_ADD_FLAG_EXCLUDE_VERBOSE)
584 && params->progress_func)
586 union wimlib_progress_info info;
587 info.scan.cur_path = path;
588 info.scan.status = WIMLIB_SCAN_DENTRY_UNSUPPORTED;
589 params->progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info);
596 if ((params->add_flags & WIMLIB_ADD_FLAG_VERBOSE)
597 && params->progress_func)
599 union wimlib_progress_info info;
600 info.scan.cur_path = path;
601 info.scan.status = WIMLIB_SCAN_DENTRY_OK;
602 params->progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info);
605 /* Create a WIM dentry with an associated inode, which may be shared */
606 ret = inode_table_new_dentry(¶ms->inode_table,
607 path_basename_with_len(path, path_len),
608 ni->mft_no, 0, false, &root);
612 if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */
613 root->is_win32_name = 1;
615 inode = root->d_inode;
617 if (inode->i_nlink > 1) /* Shared inode; nothing more to do */
620 inode->i_creation_time = le64_to_cpu(ni->creation_time);
621 inode->i_last_write_time = le64_to_cpu(ni->last_data_change_time);
622 inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
623 inode->i_attributes = le32_to_cpu(attributes);
624 inode->i_resolved = 1;
626 if (attributes & FILE_ATTR_REPARSE_POINT)
627 stream_type = AT_REPARSE_POINT;
629 stream_type = AT_DATA;
631 /* Capture the file's streams; more specifically, this is supposed to:
633 * - Regular files: capture unnamed data stream and any named data
635 * - Directories: capture any named data streams
636 * - Reparse points: capture reparse data only
638 ret = capture_ntfs_streams(inode, ni, path, path_len,
639 params->lookup_table, vol, stream_type);
643 if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) {
645 /* Recurse to directory children */
647 struct dos_name_map dos_name_map = { .rb_root = {.rb_node = NULL} };
648 struct readdir_ctx ctx = {
651 .path_len = path_len,
652 .dos_name_map = &dos_name_map,
656 ret = ntfs_readdir(ni, &pos, &ctx, wim_ntfs_capture_filldir);
658 ERROR_WITH_ERRNO("Error reading directory \"%s\"", path);
659 ret = WIMLIB_ERR_NTFS_3G;
661 ret = for_dentry_child(root, set_dentry_dos_name,
664 destroy_dos_name_map(&dos_name_map);
669 /* Reparse-point fixups are a no-op because in NTFS-3g capture mode we
670 * only allow capturing an entire volume. */
671 if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX &&
672 inode_is_symlink(inode))
673 inode->i_not_rpfixed = 0;
675 if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
676 struct SECURITY_CONTEXT sec_ctx;
680 /* Get security descriptor */
681 memset(&sec_ctx, 0, sizeof(sec_ctx));
685 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, _sd, sizeof(_sd));
686 if (ret > sizeof(_sd)) {
688 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, ret);
691 inode->i_security_id = sd_set_add_sd(¶ms->sd_set,
693 if (inode->i_security_id == -1) {
694 ERROR("Out of memory");
695 ret = WIMLIB_ERR_NOMEM;
698 DEBUG("Added security ID = %u for `%s'",
699 inode->i_security_id, path);
701 } else if (ret < 0) {
702 ERROR_WITH_ERRNO("Failed to get security information from "
704 ret = WIMLIB_ERR_NTFS_3G;
706 inode->i_security_id = -1;
707 DEBUG("No security ID for `%s'", path);
714 free_dentry_tree(root, params->lookup_table);
720 do_ntfs_umount(struct _ntfs_volume *vol)
722 DEBUG("Unmounting NTFS volume");
723 if (ntfs_umount(vol, FALSE))
724 return WIMLIB_ERR_NTFS_3G;
730 build_dentry_tree_ntfs(struct wim_dentry **root_p,
732 struct add_image_params *params)
738 DEBUG("Mounting NTFS volume `%s' read-only", device);
740 /* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
743 * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
744 * NTFS_MNT_RDONLY is an enumerated constant. Also, the NTFS-3g headers don't
745 * seem to contain any explicit version information. So we have to rely on a
746 * test done at configure time to detect whether NTFS_MNT_RDONLY should be used.
748 #ifdef HAVE_NTFS_MNT_RDONLY
750 vol = ntfs_mount(device, NTFS_MNT_RDONLY);
751 #elif defined(MS_RDONLY)
752 /* NTFS-3g 2011, 2012 */
753 vol = ntfs_mount(device, MS_RDONLY);
755 #error "Can't find NTFS_MNT_RDONLY or MS_RDONLY flags"
758 ERROR_WITH_ERRNO("Failed to mount NTFS volume `%s' read-only",
760 return WIMLIB_ERR_NTFS_3G;
762 ntfs_open_secure(vol);
764 /* We don't want to capture the special NTFS files such as $Bitmap. Not
765 * to be confused with "hidden" or "system" files which are real files
766 * that we do need to capture. */
767 NVolClearShowSysFiles(vol);
769 DEBUG("Opening root NTFS dentry");
770 root_ni = ntfs_inode_open(vol, FILE_root);
772 ERROR_WITH_ERRNO("Failed to open root inode of NTFS volume "
774 ret = WIMLIB_ERR_NTFS_3G;
778 /* Currently we assume that all the paths fit into this length and there
779 * is no check for overflow. */
780 char *path = MALLOC(32768);
782 ERROR("Could not allocate memory for NTFS pathname");
783 ret = WIMLIB_ERR_NOMEM;
789 ret = build_dentry_tree_ntfs_recursive(root_p, root_ni, path, 1,
790 FILE_NAME_POSIX, vol, params);
793 ntfs_inode_close(root_ni);
795 ntfs_index_ctx_put(vol->secure_xsii);
796 ntfs_index_ctx_put(vol->secure_xsdh);
797 ntfs_inode_close(vol->secure_ni);
800 if (do_ntfs_umount(vol)) {
801 ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'",
804 ret = WIMLIB_ERR_NTFS_3G;
807 /* We need to leave the NTFS volume mounted so that we can read
808 * the NTFS files again when we are actually writing the WIM */
809 *(ntfs_volume**)params->extra_arg = vol;
813 #endif /* WITH_NTFS_3G */