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 <time.h> /* NTFS-3g headers are missing <time.h> include */
43 #include <ntfs-3g/attrib.h>
44 #include <ntfs-3g/endians.h>
45 #include <ntfs-3g/layout.h>
46 #include <ntfs-3g/misc.h>
47 #include <ntfs-3g/reparse.h>
48 #include <ntfs-3g/security.h> /* ntfs-3g/security.h before ntfs-3g/xattrs.h */
49 #include <ntfs-3g/types.h>
50 #include <ntfs-3g/volume.h>
51 #include <ntfs-3g/xattrs.h>
53 #include <ntfs-3g/acls.h> /* This should be included last as it requires
54 definitions from above not included by itself */
56 #include "wimlib/capture.h"
57 #include "wimlib/dentry.h"
58 #include "wimlib/encoding.h"
59 #include "wimlib/endianness.h"
60 #include "wimlib/error.h"
61 #include "wimlib/lookup_table.h"
62 #include "wimlib/ntfs_3g.h"
63 #include "wimlib/paths.h"
64 #include "wimlib/security.h"
66 static inline ntfschar *
67 attr_record_name(ATTR_RECORD *ar)
69 return (ntfschar*)((u8*)ar + le16_to_cpu(ar->name_offset));
73 open_ntfs_attr(ntfs_inode *ni, struct ntfs_location *loc)
77 na = ntfs_attr_open(ni,
78 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
80 loc->stream_name_nchars);
82 ERROR_WITH_ERRNO("Failed to open attribute of \"%"TS"\" in "
83 "NTFS volume", loc->path);
89 read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte,
91 consume_data_callback_t cb,
95 struct ntfs_location *loc = lte->ntfs_loc;
96 ntfs_volume *vol = loc->ntfs_vol;
104 ni = ntfs_pathname_to_inode(vol, NULL, loc->path);
106 ERROR_WITH_ERRNO("Can't find NTFS inode for \"%"TS"\"", loc->path);
107 ret = WIMLIB_ERR_NTFS_3G;
111 na = open_ntfs_attr(ni, loc);
113 ret = WIMLIB_ERR_NTFS_3G;
114 goto out_close_ntfs_inode;
118 out_buf = alloca(WIM_CHUNK_SIZE);
120 out_buf = ctx_or_buf;
121 pos = (loc->is_reparse_point) ? 8 : 0;
122 bytes_remaining = size;
123 while (bytes_remaining) {
124 s64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
125 if (ntfs_attr_pread(na, pos, to_read, out_buf) != to_read) {
126 ERROR_WITH_ERRNO("Error reading \"%"TS"\"", loc->path);
127 ret = WIMLIB_ERR_NTFS_3G;
128 goto out_close_ntfs_attr;
131 bytes_remaining -= to_read;
133 ret = cb(out_buf, to_read, ctx_or_buf);
135 goto out_close_ntfs_attr;
143 out_close_ntfs_inode:
144 ntfs_inode_close(ni);
150 read_reparse_tag(ntfs_inode *ni, struct ntfs_location *loc,
151 u32 *reparse_tag_ret)
157 na = open_ntfs_attr(ni, loc);
159 ret = WIMLIB_ERR_NTFS_3G;
163 if (ntfs_attr_pread(na, 0, sizeof(reparse_tag),
164 &reparse_tag) != sizeof(reparse_tag))
166 ERROR_WITH_ERRNO("Error reading reparse data");
167 ret = WIMLIB_ERR_NTFS_3G;
168 goto out_close_ntfs_attr;
170 *reparse_tag_ret = le32_to_cpu(reparse_tag);
171 DEBUG("ReparseTag = %#x", *reparse_tag_ret);
180 /* Load the streams from a file or reparse point in the NTFS volume into the WIM
183 capture_ntfs_streams(struct wim_inode *inode,
187 struct wim_lookup_table *lookup_table,
191 ntfs_attr_search_ctx *actx;
192 struct ntfs_location *ntfs_loc;
194 struct wim_lookup_table_entry *lte;
196 DEBUG2("Capturing NTFS data streams from `%s'", path);
198 /* Get context to search the streams of the NTFS file. */
199 actx = ntfs_attr_get_search_ctx(ni, NULL);
201 ERROR_WITH_ERRNO("Cannot get NTFS attribute search "
202 "context for \"%s\"", path);
203 return WIMLIB_ERR_NTFS_3G;
206 /* Capture each data stream or reparse data stream. */
207 while (!ntfs_attr_lookup(type, NULL, 0,
208 CASE_SENSITIVE, 0, NULL, 0, actx))
210 u64 data_size = ntfs_get_attribute_value_length(actx->attr);
211 u64 name_length = actx->attr->name_length;
214 if (data_size == 0) {
215 /* Empty stream. No lookup table entry is needed. */
219 ntfs_loc = CALLOC(1, sizeof(*ntfs_loc));
221 ret = WIMLIB_ERR_NOMEM;
224 ntfs_loc->ntfs_vol = vol;
225 ntfs_loc->path = memdup(path, path_len + 1);
226 if (!ntfs_loc->path) {
227 ret = WIMLIB_ERR_NOMEM;
228 goto out_free_ntfs_loc;
231 ntfs_loc->stream_name = memdup(attr_record_name(actx->attr),
233 if (!ntfs_loc->stream_name) {
234 ret = WIMLIB_ERR_NOMEM;
235 goto out_free_ntfs_loc;
237 ntfs_loc->stream_name_nchars = name_length;
240 lte = new_lookup_table_entry();
242 ret = WIMLIB_ERR_NOMEM;
243 goto out_free_ntfs_loc;
245 lte->resource_location = RESOURCE_IN_NTFS_VOLUME;
246 lte->ntfs_loc = ntfs_loc;
248 if (type == AT_REPARSE_POINT) {
250 ERROR("Invalid reparse data on \"%s\" "
251 "(only %u bytes)!", path, (unsigned)data_size);
252 ret = WIMLIB_ERR_NTFS_3G;
255 lte->ntfs_loc->is_reparse_point = true;
256 lte->resource_entry.original_size = data_size - 8;
257 ret = read_reparse_tag(ni, lte->ntfs_loc,
258 &inode->i_reparse_tag);
262 lte->ntfs_loc->is_reparse_point = false;
263 lte->resource_entry.original_size = data_size;
266 if (name_length == 0) {
267 /* Unnamed data stream. Put the reference to it in the
271 ERROR("Found two un-named data streams for \"%s\" "
272 "(sizes = %"PRIu64", %"PRIu64")",
273 path, wim_resource_size(inode->i_lte),
274 wim_resource_size(lte));
275 ret = WIMLIB_ERR_NTFS_3G;
283 /* Named data stream. Put the reference to it in the
284 * alternate data stream entries */
285 struct wim_ads_entry *new_ads_entry;
287 new_ads_entry = inode_add_ads_utf16le(inode,
288 attr_record_name(actx->attr),
290 if (!new_ads_entry) {
291 ret = WIMLIB_ERR_NOMEM;
294 wimlib_assert(new_ads_entry->stream_name_nbytes == name_length * 2);
295 stream_id = new_ads_entry->stream_id;
296 new_ads_entry->lte = lte;
299 lookup_table_insert_unhashed(lookup_table, lte,
303 if (errno == ENOENT) {
306 ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path);
307 ret = WIMLIB_ERR_NTFS_3G;
311 free_lookup_table_entry(lte);
314 FREE(ntfs_loc->path);
315 FREE(ntfs_loc->stream_name);
319 ntfs_attr_put_search_ctx(actx);
321 DEBUG2("Successfully captured NTFS streams from \"%s\"", path);
323 ERROR("Failed to capture NTFS streams from \"%s\"", path);
327 /* Red-black tree that maps NTFS inode numbers to DOS names */
328 struct dos_name_map {
329 struct rb_root rb_root;
332 struct dos_name_node {
333 struct rb_node rb_node;
339 /* Inserts a new DOS name into the map */
341 insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name,
342 size_t name_nbytes, u64 ntfs_ino)
344 struct dos_name_node *new_node;
346 struct rb_root *root;
347 struct rb_node *rb_parent;
349 DEBUG("DOS name_len = %zu", name_nbytes);
350 new_node = MALLOC(sizeof(struct dos_name_node));
354 /* DOS names are supposed to be 12 characters max (that's 24 bytes,
355 * assuming 2-byte ntfs characters) */
356 wimlib_assert(name_nbytes <= sizeof(new_node->dos_name));
358 /* Initialize the DOS name, DOS name length, and NTFS inode number of
359 * the red-black tree node */
360 memcpy(new_node->dos_name, dos_name, name_nbytes);
361 new_node->name_nbytes = name_nbytes;
362 new_node->ntfs_ino = ntfs_ino;
364 /* Insert the red-black tree node */
365 root = &map->rb_root;
369 struct dos_name_node *this;
371 this = container_of(*p, struct dos_name_node, rb_node);
373 if (new_node->ntfs_ino < this->ntfs_ino)
374 p = &((*p)->rb_left);
375 else if (new_node->ntfs_ino > this->ntfs_ino)
376 p = &((*p)->rb_right);
378 /* This should be impossible since a NTFS inode cannot
379 * have multiple DOS names, and we only should get each
380 * DOS name entry once from the ntfs_readdir() calls. */
381 ERROR("NTFS inode %"PRIu64" has multiple DOS names",
386 rb_link_node(&new_node->rb_node, rb_parent, p);
387 rb_insert_color(&new_node->rb_node, root);
388 DEBUG("Inserted DOS name for inode %"PRIu64, ntfs_ino);
392 /* Returns a structure that contains the DOS name and its length for a NTFS
393 * inode, or NULL if the inode has no DOS name. */
394 static struct dos_name_node *
395 lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino)
397 struct rb_node *node = map->rb_root.rb_node;
399 struct dos_name_node *this;
400 this = container_of(node, struct dos_name_node, rb_node);
401 if (ntfs_ino < this->ntfs_ino)
402 node = node->rb_left;
403 else if (ntfs_ino > this->ntfs_ino)
404 node = node->rb_right;
412 set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
414 const struct dos_name_map *map = arg;
415 const struct dos_name_node *node;
417 if (dentry->is_win32_name) {
418 node = lookup_dos_name(map, dentry->d_inode->i_ino);
420 dentry->short_name = MALLOC(node->name_nbytes + 2);
421 if (!dentry->short_name)
422 return WIMLIB_ERR_NOMEM;
423 memcpy(dentry->short_name, node->dos_name,
425 dentry->short_name[node->name_nbytes / 2] = 0;
426 dentry->short_name_nbytes = node->name_nbytes;
427 DEBUG("Assigned DOS name to ino %"PRIu64,
428 dentry->d_inode->i_ino);
430 WARNING("NTFS inode %"PRIu64" has Win32 name with no "
431 "corresponding DOS name",
432 dentry->d_inode->i_ino);
439 free_dos_name_tree(struct rb_node *node) {
441 free_dos_name_tree(node->rb_left);
442 free_dos_name_tree(node->rb_right);
443 FREE(container_of(node, struct dos_name_node, rb_node));
448 destroy_dos_name_map(struct dos_name_map *map)
450 free_dos_name_tree(map->rb_root.rb_node);
454 struct wim_dentry *parent;
458 struct dos_name_map *dos_name_map;
460 struct add_image_params *params;
464 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
470 ntfs_volume *ntfs_vol,
471 struct add_image_params *params);
474 wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
475 const int name_nchars, const int name_type,
476 const s64 pos, const MFT_REF mref,
477 const unsigned dt_type)
479 struct readdir_ctx *ctx;
480 size_t mbs_name_nbytes;
482 struct wim_dentry *child;
485 size_t name_nbytes = name_nchars * sizeof(ntfschar);
488 if (name_type & FILE_NAME_DOS) {
489 /* If this is the entry for a DOS name, store it for later. */
490 ret = insert_dos_name(ctx->dos_name_map, name,
491 name_nbytes, mref & MFT_REF_MASK_CPU);
493 /* Return now if an error occurred or if this is just a DOS name
494 * and not a Win32+DOS name. */
495 if (ret != 0 || name_type == FILE_NAME_DOS)
498 ret = utf16le_to_tstr(name, name_nbytes,
499 &mbs_name, &mbs_name_nbytes);
503 if (mbs_name[0] == '.' &&
504 (mbs_name[1] == '\0' ||
505 (mbs_name[1] == '.' && mbs_name[2] == '\0'))) {
508 * note: name_type is POSIX for these, so DOS names will not
509 * have been inserted for them. */
511 goto out_free_mbs_name;
514 /* Open the inode for this directory entry and recursively capture the
515 * directory tree rooted at it */
516 ntfs_inode *ni = ntfs_inode_open(ctx->dir_ni->vol, mref);
518 /* XXX This used to be treated as an error, but NTFS-3g seemed
519 * to be unable to read some inodes on a Windows 8 image for
521 WARNING_WITH_ERRNO("Failed to open NTFS file \"%s/%s\"",
522 ctx->path, mbs_name);
524 goto out_free_mbs_name;
526 path_len = ctx->path_len;
528 ctx->path[path_len++] = '/';
529 memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1);
530 path_len += mbs_name_nbytes;
532 ret = build_dentry_tree_ntfs_recursive(&child, ctx->dir_ni,
533 ni, ctx->path, path_len, name_type,
534 ctx->vol, ctx->params);
535 path_len -= mbs_name_nbytes + 1;
537 dentry_add_child(ctx->parent, child);
538 ntfs_inode_close(ni);
542 ctx->path[ctx->path_len] = '\0';
546 /* Recursively build a WIM dentry tree corresponding to a NTFS volume.
547 * At the same time, update the WIM lookup table with lookup table entries for
548 * the NTFS streams, and build an array of security descriptors.
551 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
558 struct add_image_params *params)
562 struct wim_dentry *root;
563 struct wim_inode *inode;
564 ATTR_TYPES stream_type;
566 if (exclude_path(path, path_len, params->config, false)) {
567 /* Exclude a file or directory tree based on the capture
568 * configuration file */
569 if ((params->add_flags & WIMLIB_ADD_FLAG_EXCLUDE_VERBOSE)
570 && params->progress_func)
572 union wimlib_progress_info info;
573 info.scan.cur_path = path;
574 info.scan.excluded = true;
575 params->progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info);
582 /* Get file attributes */
583 struct SECURITY_CONTEXT ctx;
584 memset(&ctx, 0, sizeof(ctx));
586 ret = ntfs_xattr_system_getxattr(&ctx, XATTR_NTFS_ATTRIB,
587 ni, dir_ni, (char *)&attributes,
589 if (ret != sizeof(attributes)) {
590 ERROR_WITH_ERRNO("Failed to get NTFS attributes from \"%s\"", path);
591 return WIMLIB_ERR_NTFS_3G;
594 if ((params->add_flags & WIMLIB_ADD_FLAG_VERBOSE)
595 && params->progress_func)
597 union wimlib_progress_info info;
598 info.scan.cur_path = path;
599 info.scan.excluded = false;
600 params->progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info);
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 inode = root->d_inode;
612 if (inode->i_nlink > 1) /* Shared inode; nothing more to do */
615 if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */
616 root->is_win32_name = 1;
617 inode->i_creation_time = le64_to_cpu(ni->creation_time);
618 inode->i_last_write_time = le64_to_cpu(ni->last_data_change_time);
619 inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
620 inode->i_attributes = le32_to_cpu(attributes);
621 inode->i_resolved = 1;
623 if (attributes & FILE_ATTR_REPARSE_POINT)
624 stream_type = AT_REPARSE_POINT;
626 stream_type = AT_DATA;
628 /* Capture the file's streams; more specifically, this is supposed to:
630 * - Regular files: capture unnamed data stream and any named data
632 * - Directories: capture any named data streams
633 * - Reparse points: capture reparse data only
635 ret = capture_ntfs_streams(inode, ni, path, path_len,
636 params->lookup_table, vol, stream_type);
640 if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) {
642 /* Recurse to directory children */
644 struct dos_name_map dos_name_map = { .rb_root = {.rb_node = NULL} };
645 struct readdir_ctx ctx = {
649 .path_len = path_len,
650 .dos_name_map = &dos_name_map,
654 ret = ntfs_readdir(ni, &pos, &ctx, wim_ntfs_capture_filldir);
656 ERROR_WITH_ERRNO("Error reading directory \"%s\"", path);
657 ret = WIMLIB_ERR_NTFS_3G;
659 ret = for_dentry_child(root, set_dentry_dos_name,
662 destroy_dos_name_map(&dos_name_map);
667 /* Reparse-point fixups are a no-op because in NTFS-3g capture mode we
668 * only allow capturing an entire volume. */
669 if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX &&
670 inode_is_symlink(inode))
671 inode->i_not_rpfixed = 0;
673 if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
674 /* Get security descriptor */
678 ret = ntfs_xattr_system_getxattr(&ctx, XATTR_NTFS_ACL,
681 if (ret > sizeof(sd)) {
683 ret = ntfs_xattr_system_getxattr(&ctx, XATTR_NTFS_ACL,
684 ni, dir_ni, sd, ret);
687 inode->i_security_id = sd_set_add_sd(¶ms->sd_set,
689 if (inode->i_security_id == -1) {
690 ERROR("Out of memory");
691 ret = WIMLIB_ERR_NOMEM;
694 DEBUG("Added security ID = %u for `%s'",
695 inode->i_security_id, path);
697 } else if (ret < 0) {
698 ERROR_WITH_ERRNO("Failed to get security information from "
700 ret = WIMLIB_ERR_NTFS_3G;
702 inode->i_security_id = -1;
703 DEBUG("No security ID for `%s'", path);
710 free_dentry_tree(root, params->lookup_table);
716 do_ntfs_umount(struct _ntfs_volume *vol)
718 DEBUG("Unmounting NTFS volume");
719 if (ntfs_umount(vol, FALSE))
720 return WIMLIB_ERR_NTFS_3G;
726 build_dentry_tree_ntfs(struct wim_dentry **root_p,
728 struct add_image_params *params)
734 DEBUG("Mounting NTFS volume `%s' read-only", device);
736 /* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
739 * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
740 * NTFS_MNT_RDONLY is an enumerated constant. Also, the NTFS-3g headers don't
741 * seem to contain any explicit version information. So we have to rely on a
742 * test done at configure time to detect whether NTFS_MNT_RDONLY should be used.
744 #ifdef HAVE_NTFS_MNT_RDONLY
746 vol = ntfs_mount(device, NTFS_MNT_RDONLY);
747 #elif defined(MS_RDONLY)
748 /* NTFS-3g 2011, 2012 */
749 vol = ntfs_mount(device, MS_RDONLY);
751 #error "Can't find NTFS_MNT_RDONLY or MS_RDONLY flags"
754 ERROR_WITH_ERRNO("Failed to mount NTFS volume `%s' read-only",
756 return WIMLIB_ERR_NTFS_3G;
758 ntfs_open_secure(vol);
760 /* We don't want to capture the special NTFS files such as $Bitmap. Not
761 * to be confused with "hidden" or "system" files which are real files
762 * that we do need to capture. */
763 NVolClearShowSysFiles(vol);
765 DEBUG("Opening root NTFS dentry");
766 root_ni = ntfs_inode_open(vol, FILE_root);
768 ERROR_WITH_ERRNO("Failed to open root inode of NTFS volume "
770 ret = WIMLIB_ERR_NTFS_3G;
774 /* Currently we assume that all the paths fit into this length and there
775 * is no check for overflow. */
776 char *path = MALLOC(32768);
778 ERROR("Could not allocate memory for NTFS pathname");
779 ret = WIMLIB_ERR_NOMEM;
785 ret = build_dentry_tree_ntfs_recursive(root_p, NULL, root_ni, path, 1,
786 FILE_NAME_POSIX, vol, params);
789 ntfs_inode_close(root_ni);
791 ntfs_index_ctx_put(vol->secure_xsii);
792 ntfs_index_ctx_put(vol->secure_xsdh);
793 ntfs_inode_close(vol->secure_ni);
796 if (do_ntfs_umount(vol)) {
797 ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'",
800 ret = WIMLIB_ERR_NTFS_3G;
803 /* We need to leave the NTFS volume mounted so that we can read
804 * the NTFS files again when we are actually writing the WIM */
805 *(ntfs_volume**)params->extra_arg = vol;
809 #endif /* WITH_NTFS_3G */