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,
85 struct ntfs_location *loc = lte->ntfs_loc;
86 ntfs_volume *vol = loc->ntfs_vol;
92 bool out_buf_malloced;
94 size_t stack_max = 32768;
96 ni = ntfs_pathname_to_inode(vol, NULL, loc->path);
98 ERROR_WITH_ERRNO("Can't find NTFS inode for \"%"TS"\"", loc->path);
99 ret = WIMLIB_ERR_NTFS_3G;
103 na = open_ntfs_attr(ni, loc);
105 ret = WIMLIB_ERR_NTFS_3G;
106 goto out_close_ntfs_inode;
109 out_buf_malloced = false;
111 if (in_chunk_size <= stack_max) {
112 out_buf = alloca(in_chunk_size);
114 out_buf = MALLOC(in_chunk_size);
115 if (out_buf == NULL) {
116 ret = WIMLIB_ERR_NOMEM;
117 goto out_close_ntfs_attr;
119 out_buf_malloced = true;
122 out_buf = ctx_or_buf;
124 pos = (loc->is_reparse_point) ? 8 : 0;
125 bytes_remaining = size;
126 while (bytes_remaining) {
127 s64 to_read = min(bytes_remaining, in_chunk_size);
128 if (ntfs_attr_pread(na, pos, to_read, out_buf) != to_read) {
129 ERROR_WITH_ERRNO("Error reading \"%"TS"\"", loc->path);
130 ret = WIMLIB_ERR_NTFS_3G;
131 goto out_free_memory;
134 bytes_remaining -= to_read;
136 ret = cb(out_buf, to_read, ctx_or_buf);
138 goto out_free_memory;
145 if (out_buf_malloced)
149 out_close_ntfs_inode:
150 ntfs_inode_close(ni);
156 read_reparse_tag(ntfs_inode *ni, struct ntfs_location *loc,
157 u32 *reparse_tag_ret)
163 na = open_ntfs_attr(ni, loc);
165 ret = WIMLIB_ERR_NTFS_3G;
169 if (ntfs_attr_pread(na, 0, sizeof(reparse_tag),
170 &reparse_tag) != sizeof(reparse_tag))
172 ERROR_WITH_ERRNO("Error reading reparse data");
173 ret = WIMLIB_ERR_NTFS_3G;
174 goto out_close_ntfs_attr;
176 *reparse_tag_ret = le32_to_cpu(reparse_tag);
177 DEBUG("ReparseTag = %#x", *reparse_tag_ret);
186 /* Load the streams from a file or reparse point in the NTFS volume into the WIM
189 capture_ntfs_streams(struct wim_inode *inode,
193 struct wim_lookup_table *lookup_table,
197 ntfs_attr_search_ctx *actx;
198 struct ntfs_location *ntfs_loc;
200 struct wim_lookup_table_entry *lte;
202 DEBUG2("Capturing NTFS data streams from `%s'", path);
204 /* Get context to search the streams of the NTFS file. */
205 actx = ntfs_attr_get_search_ctx(ni, NULL);
207 ERROR_WITH_ERRNO("Cannot get NTFS attribute search "
208 "context for \"%s\"", path);
209 return WIMLIB_ERR_NTFS_3G;
212 /* Capture each data stream or reparse data stream. */
213 while (!ntfs_attr_lookup(type, NULL, 0,
214 CASE_SENSITIVE, 0, NULL, 0, actx))
216 u64 data_size = ntfs_get_attribute_value_length(actx->attr);
217 u64 name_length = actx->attr->name_length;
220 if (data_size == 0) {
221 /* Empty stream. No lookup table entry is needed. */
225 ntfs_loc = CALLOC(1, sizeof(*ntfs_loc));
227 ret = WIMLIB_ERR_NOMEM;
230 ntfs_loc->ntfs_vol = vol;
231 ntfs_loc->path = memdup(path, path_len + 1);
232 if (!ntfs_loc->path) {
233 ret = WIMLIB_ERR_NOMEM;
234 goto out_free_ntfs_loc;
237 ntfs_loc->stream_name = memdup(attr_record_name(actx->attr),
239 if (!ntfs_loc->stream_name) {
240 ret = WIMLIB_ERR_NOMEM;
241 goto out_free_ntfs_loc;
243 ntfs_loc->stream_name_nchars = name_length;
246 lte = new_lookup_table_entry();
248 ret = WIMLIB_ERR_NOMEM;
249 goto out_free_ntfs_loc;
251 lte->resource_location = RESOURCE_IN_NTFS_VOLUME;
252 lte->ntfs_loc = ntfs_loc;
254 if (type == AT_REPARSE_POINT) {
256 ERROR("Invalid reparse data on \"%s\" "
257 "(only %u bytes)!", path, (unsigned)data_size);
258 ret = WIMLIB_ERR_NTFS_3G;
261 lte->ntfs_loc->is_reparse_point = true;
262 lte->resource_entry.original_size = data_size - 8;
263 ret = read_reparse_tag(ni, lte->ntfs_loc,
264 &inode->i_reparse_tag);
268 lte->ntfs_loc->is_reparse_point = false;
269 lte->resource_entry.original_size = data_size;
272 if (name_length == 0) {
273 /* Unnamed data stream. Put the reference to it in the
277 ERROR("Found two un-named data streams for \"%s\" "
278 "(sizes = %"PRIu64", %"PRIu64")",
279 path, wim_resource_size(inode->i_lte),
280 wim_resource_size(lte));
281 ret = WIMLIB_ERR_NTFS_3G;
289 /* Named data stream. Put the reference to it in the
290 * alternate data stream entries */
291 struct wim_ads_entry *new_ads_entry;
293 new_ads_entry = inode_add_ads_utf16le(inode,
294 attr_record_name(actx->attr),
296 if (!new_ads_entry) {
297 ret = WIMLIB_ERR_NOMEM;
300 wimlib_assert(new_ads_entry->stream_name_nbytes == name_length * 2);
301 stream_id = new_ads_entry->stream_id;
302 new_ads_entry->lte = lte;
305 lookup_table_insert_unhashed(lookup_table, lte,
309 if (errno == ENOENT) {
312 ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path);
313 ret = WIMLIB_ERR_NTFS_3G;
317 free_lookup_table_entry(lte);
320 FREE(ntfs_loc->path);
321 FREE(ntfs_loc->stream_name);
325 ntfs_attr_put_search_ctx(actx);
327 DEBUG2("Successfully captured NTFS streams from \"%s\"", path);
329 ERROR("Failed to capture NTFS streams from \"%s\"", path);
333 /* Red-black tree that maps NTFS inode numbers to DOS names */
334 struct dos_name_map {
335 struct rb_root rb_root;
338 struct dos_name_node {
339 struct rb_node rb_node;
345 /* Inserts a new DOS name into the map */
347 insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name,
348 size_t name_nbytes, le64 ntfs_ino)
350 struct dos_name_node *new_node;
352 struct rb_root *root;
353 struct rb_node *rb_parent;
355 DEBUG("DOS name_len = %zu", name_nbytes);
356 new_node = MALLOC(sizeof(struct dos_name_node));
360 /* DOS names are supposed to be 12 characters max (that's 24 bytes,
361 * assuming 2-byte ntfs characters) */
362 wimlib_assert(name_nbytes <= sizeof(new_node->dos_name));
364 /* Initialize the DOS name, DOS name length, and NTFS inode number of
365 * the red-black tree node */
366 memcpy(new_node->dos_name, dos_name, name_nbytes);
367 new_node->name_nbytes = name_nbytes;
368 new_node->ntfs_ino = ntfs_ino;
370 /* Insert the red-black tree node */
371 root = &map->rb_root;
375 struct dos_name_node *this;
377 this = container_of(*p, struct dos_name_node, rb_node);
379 if (new_node->ntfs_ino < this->ntfs_ino)
380 p = &((*p)->rb_left);
381 else if (new_node->ntfs_ino > this->ntfs_ino)
382 p = &((*p)->rb_right);
384 /* This should be impossible since a NTFS inode cannot
385 * have multiple DOS names, and we only should get each
386 * DOS name entry once from the ntfs_readdir() calls. */
387 ERROR("NTFS inode %"PRIu64" has multiple DOS names",
388 le64_to_cpu(ntfs_ino));
392 rb_link_node(&new_node->rb_node, rb_parent, p);
393 rb_insert_color(&new_node->rb_node, root);
394 DEBUG("Inserted DOS name for inode %"PRIu64, le64_to_cpu(ntfs_ino));
398 /* Returns a structure that contains the DOS name and its length for a NTFS
399 * inode, or NULL if the inode has no DOS name. */
400 static struct dos_name_node *
401 lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino)
403 struct rb_node *node = map->rb_root.rb_node;
405 struct dos_name_node *this;
406 this = container_of(node, struct dos_name_node, rb_node);
407 if (ntfs_ino < this->ntfs_ino)
408 node = node->rb_left;
409 else if (ntfs_ino > this->ntfs_ino)
410 node = node->rb_right;
418 set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
420 const struct dos_name_map *map = arg;
421 const struct dos_name_node *node;
423 if (dentry->is_win32_name) {
424 node = lookup_dos_name(map, dentry->d_inode->i_ino);
426 dentry->short_name = MALLOC(node->name_nbytes + 2);
427 if (!dentry->short_name)
428 return WIMLIB_ERR_NOMEM;
429 memcpy(dentry->short_name, node->dos_name,
431 dentry->short_name[node->name_nbytes / 2] = 0;
432 dentry->short_name_nbytes = node->name_nbytes;
433 DEBUG("Assigned DOS name to ino %"PRIu64,
434 dentry->d_inode->i_ino);
436 WARNING("NTFS inode %"PRIu64" has Win32 name with no "
437 "corresponding DOS name",
438 dentry->d_inode->i_ino);
445 free_dos_name_tree(struct rb_node *node) {
447 free_dos_name_tree(node->rb_left);
448 free_dos_name_tree(node->rb_right);
449 FREE(container_of(node, struct dos_name_node, rb_node));
454 destroy_dos_name_map(struct dos_name_map *map)
456 free_dos_name_tree(map->rb_root.rb_node);
460 struct wim_dentry *parent;
463 struct dos_name_map *dos_name_map;
465 struct add_image_params *params;
469 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
474 ntfs_volume *ntfs_vol,
475 struct add_image_params *params);
478 wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
479 const int name_nchars, const int name_type,
480 const s64 pos, const MFT_REF mref,
481 const unsigned dt_type)
483 struct readdir_ctx *ctx;
484 size_t mbs_name_nbytes;
486 struct wim_dentry *child;
489 size_t name_nbytes = name_nchars * sizeof(ntfschar);
492 if (name_type & FILE_NAME_DOS) {
493 /* If this is the entry for a DOS name, store it for later. */
494 ret = insert_dos_name(ctx->dos_name_map, name,
495 name_nbytes, mref & MFT_REF_MASK_CPU);
497 /* Return now if an error occurred or if this is just a DOS name
498 * and not a Win32+DOS name. */
499 if (ret != 0 || name_type == FILE_NAME_DOS)
502 ret = utf16le_to_tstr(name, name_nbytes,
503 &mbs_name, &mbs_name_nbytes);
507 if (mbs_name[0] == '.' &&
508 (mbs_name[1] == '\0' ||
509 (mbs_name[1] == '.' && mbs_name[2] == '\0'))) {
512 * note: name_type is POSIX for these, so DOS names will not
513 * have been inserted for them. */
515 goto out_free_mbs_name;
518 /* Open the inode for this directory entry and recursively capture the
519 * directory tree rooted at it */
520 ntfs_inode *ni = ntfs_inode_open(ctx->vol, mref);
522 /* XXX This used to be treated as an error, but NTFS-3g seemed
523 * to be unable to read some inodes on a Windows 8 image for
525 WARNING_WITH_ERRNO("Failed to open NTFS file \"%s/%s\"",
526 ctx->path, mbs_name);
528 goto out_free_mbs_name;
530 path_len = ctx->path_len;
532 ctx->path[path_len++] = '/';
533 memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1);
534 path_len += mbs_name_nbytes;
536 ret = build_dentry_tree_ntfs_recursive(&child, ni, ctx->path,
538 ctx->vol, ctx->params);
539 path_len -= mbs_name_nbytes + 1;
541 dentry_add_child(ctx->parent, child);
542 ntfs_inode_close(ni);
546 ctx->path[ctx->path_len] = '\0';
550 /* Recursively build a WIM dentry tree corresponding to a NTFS volume.
551 * At the same time, update the WIM lookup table with lookup table entries for
552 * the NTFS streams, and build an array of security descriptors.
555 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
561 struct add_image_params *params)
565 struct wim_dentry *root = NULL;
566 struct wim_inode *inode;
567 ATTR_TYPES stream_type;
569 params->progress.scan.cur_path = path;
571 if (exclude_path(path, path_len, params->config, false)) {
572 /* Exclude a file or directory tree based on the capture
573 * configuration file */
574 do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED);
579 /* Get file attributes */
580 ret = ntfs_get_ntfs_attrib(ni, (char*)&attributes, sizeof(attributes));
581 if (ret != sizeof(attributes)) {
582 ERROR_WITH_ERRNO("Failed to get NTFS attributes from \"%s\"", path);
583 ret = WIMLIB_ERR_NTFS_3G;
587 if ((attributes & (FILE_ATTRIBUTE_DIRECTORY |
588 FILE_ATTRIBUTE_ENCRYPTED)) == FILE_ATTRIBUTE_ENCRYPTED)
590 if (params->add_flags & WIMLIB_ADD_FLAG_NO_UNSUPPORTED_EXCLUDE)
592 ERROR("Can't archive unsupported encrypted file \"%s\"", path);
593 ret = WIMLIB_ERR_UNSUPPORTED_FILE;
596 do_capture_progress(params, WIMLIB_SCAN_DENTRY_UNSUPPORTED);
601 do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK);
603 /* Create a WIM dentry with an associated inode, which may be shared */
604 ret = inode_table_new_dentry(¶ms->inode_table,
605 path_basename_with_len(path, path_len),
606 ni->mft_no, 0, false, &root);
610 if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */
611 root->is_win32_name = 1;
613 inode = root->d_inode;
615 if (inode->i_nlink > 1) {
616 /* Shared inode; nothing more to do */
621 inode->i_creation_time = le64_to_cpu(ni->creation_time);
622 inode->i_last_write_time = le64_to_cpu(ni->last_data_change_time);
623 inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
624 inode->i_attributes = le32_to_cpu(attributes);
625 inode->i_resolved = 1;
627 if (attributes & FILE_ATTR_REPARSE_POINT)
628 stream_type = AT_REPARSE_POINT;
630 stream_type = AT_DATA;
632 /* Capture the file's streams; more specifically, this is supposed to:
634 * - Regular files: capture unnamed data stream and any named data
636 * - Directories: capture any named data streams
637 * - Reparse points: capture reparse data only
639 ret = capture_ntfs_streams(inode, ni, path, path_len,
640 params->lookup_table, vol, stream_type);
644 if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) {
646 /* Recurse to directory children */
648 struct dos_name_map dos_name_map = { .rb_root = {.rb_node = NULL} };
649 struct readdir_ctx ctx = {
652 .path_len = path_len,
653 .dos_name_map = &dos_name_map,
657 ret = ntfs_readdir(ni, &pos, &ctx, wim_ntfs_capture_filldir);
659 ERROR_WITH_ERRNO("Error reading directory \"%s\"", path);
660 ret = WIMLIB_ERR_NTFS_3G;
662 ret = for_dentry_child(root, set_dentry_dos_name,
665 destroy_dos_name_map(&dos_name_map);
670 /* Reparse-point fixups are a no-op because in NTFS-3g capture mode we
671 * only allow capturing an entire volume. */
672 if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX &&
673 inode_is_symlink(inode))
674 inode->i_not_rpfixed = 0;
676 if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
677 struct SECURITY_CONTEXT sec_ctx;
681 /* Get security descriptor */
682 memset(&sec_ctx, 0, sizeof(sec_ctx));
687 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, sizeof(_sd));
688 if (ret > sizeof(_sd)) {
690 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, ret);
693 inode->i_security_id = sd_set_add_sd(¶ms->sd_set,
695 if (inode->i_security_id == -1) {
696 ERROR("Out of memory");
697 ret = WIMLIB_ERR_NOMEM;
700 DEBUG("Added security ID = %u for `%s'",
701 inode->i_security_id, path);
703 } else if (ret < 0) {
704 ERROR_WITH_ERRNO("Failed to get security information from "
706 ret = WIMLIB_ERR_NTFS_3G;
708 inode->i_security_id = -1;
709 DEBUG("No security ID for `%s'", path);
716 free_dentry_tree(root, params->lookup_table);
722 do_ntfs_umount(struct _ntfs_volume *vol)
724 DEBUG("Unmounting NTFS volume");
725 if (ntfs_umount(vol, FALSE))
726 return WIMLIB_ERR_NTFS_3G;
732 build_dentry_tree_ntfs(struct wim_dentry **root_p,
734 struct add_image_params *params)
740 DEBUG("Mounting NTFS volume `%s' read-only", device);
742 /* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
745 * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
746 * NTFS_MNT_RDONLY is an enumerated constant. Also, the NTFS-3g headers don't
747 * seem to contain any explicit version information. So we have to rely on a
748 * test done at configure time to detect whether NTFS_MNT_RDONLY should be used.
750 #ifdef HAVE_NTFS_MNT_RDONLY
752 vol = ntfs_mount(device, NTFS_MNT_RDONLY);
753 #elif defined(MS_RDONLY)
754 /* NTFS-3g 2011, 2012 */
755 vol = ntfs_mount(device, MS_RDONLY);
757 #error "Can't find NTFS_MNT_RDONLY or MS_RDONLY flags"
760 ERROR_WITH_ERRNO("Failed to mount NTFS volume `%s' read-only",
762 return WIMLIB_ERR_NTFS_3G;
764 ntfs_open_secure(vol);
766 /* We don't want to capture the special NTFS files such as $Bitmap. Not
767 * to be confused with "hidden" or "system" files which are real files
768 * that we do need to capture. */
769 NVolClearShowSysFiles(vol);
771 DEBUG("Opening root NTFS dentry");
772 root_ni = ntfs_inode_open(vol, FILE_root);
774 ERROR_WITH_ERRNO("Failed to open root inode of NTFS volume "
776 ret = WIMLIB_ERR_NTFS_3G;
780 /* Currently we assume that all the paths fit into this length and there
781 * is no check for overflow. */
782 char *path = MALLOC(32768);
784 ERROR("Could not allocate memory for NTFS pathname");
785 ret = WIMLIB_ERR_NOMEM;
791 ret = build_dentry_tree_ntfs_recursive(root_p, root_ni, path, 1,
792 FILE_NAME_POSIX, vol, params);
795 ntfs_inode_close(root_ni);
797 ntfs_index_ctx_put(vol->secure_xsii);
798 ntfs_index_ctx_put(vol->secure_xsdh);
799 ntfs_inode_close(vol->secure_ni);
802 if (do_ntfs_umount(vol)) {
803 ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'",
806 ret = WIMLIB_ERR_NTFS_3G;
809 /* We need to leave the NTFS volume mounted so that we can read
810 * the NTFS files again when we are actually writing the WIM */
811 *(ntfs_volume**)params->extra_arg = vol;
815 #endif /* WITH_NTFS_3G */