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, 2015 Eric Biggers
11 * This file is free software; you can redistribute it and/or modify it under
12 * the terms of the GNU Lesser General Public License as published by the Free
13 * Software Foundation; either version 3 of the License, or (at your option) any
16 * This file is distributed in the hope that it will be useful, but WITHOUT
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
18 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
21 * You should have received a copy of the GNU Lesser General Public License
22 * along with this file; if not, see http://www.gnu.org/licenses/.
33 #include <ntfs-3g/attrib.h>
34 #include <ntfs-3g/reparse.h>
35 #include <ntfs-3g/security.h>
36 #include <ntfs-3g/volume.h>
38 #include "wimlib/alloca.h"
39 #include "wimlib/assert.h"
40 #include "wimlib/blob_table.h"
41 #include "wimlib/capture.h"
42 #include "wimlib/dentry.h"
43 #include "wimlib/encoding.h"
44 #include "wimlib/endianness.h"
45 #include "wimlib/error.h"
46 #include "wimlib/ntfs_3g.h"
47 #include "wimlib/paths.h"
48 #include "wimlib/reparse.h"
49 #include "wimlib/security.h"
51 /* A reference-counted NTFS volume than is automatically unmounted when the
52 * reference count reaches 0 */
53 struct ntfs_volume_wrapper {
58 /* Description of where data is located in an NTFS volume */
59 struct ntfs_location {
60 struct ntfs_volume_wrapper *volume;
64 utf16lechar *attr_name;
68 static struct ntfs_volume_wrapper *
69 get_ntfs_volume(struct ntfs_volume_wrapper *volume)
76 put_ntfs_volume(struct ntfs_volume_wrapper *volume)
78 if (--volume->refcnt == 0) {
79 ntfs_umount(volume->vol, FALSE);
84 static inline const ntfschar *
85 attr_record_name(const ATTR_RECORD *record)
87 return (const ntfschar *)
88 ((const u8 *)record + le16_to_cpu(record->name_offset));
92 open_ntfs_attr(ntfs_inode *ni, const struct ntfs_location *loc)
96 na = ntfs_attr_open(ni, loc->attr_type, loc->attr_name,
97 loc->attr_name_nchars);
99 ERROR_WITH_ERRNO("Failed to open attribute of NTFS inode %"PRIu64,
106 read_ntfs_attribute_prefix(const struct blob_descriptor *blob, u64 size,
107 const struct read_blob_callbacks *cbs)
109 const struct ntfs_location *loc = blob->ntfs_loc;
110 ntfs_volume *vol = loc->volume->vol;
118 ni = ntfs_inode_open(vol, loc->mft_no);
120 ERROR_WITH_ERRNO("Failed to open NTFS inode %"PRIu64,
122 ret = WIMLIB_ERR_NTFS_3G;
126 na = open_ntfs_attr(ni, loc);
128 ret = WIMLIB_ERR_NTFS_3G;
129 goto out_close_ntfs_inode;
132 pos = (loc->attr_type == AT_REPARSE_POINT) ? REPARSE_DATA_OFFSET : 0;
133 bytes_remaining = size;
134 while (bytes_remaining) {
135 s64 to_read = min(bytes_remaining, sizeof(buf));
136 if (ntfs_attr_pread(na, pos, to_read, buf) != to_read) {
137 ERROR_WITH_ERRNO("Error reading data from NTFS inode "
138 "%"PRIu64, loc->mft_no);
139 ret = WIMLIB_ERR_NTFS_3G;
140 goto out_close_ntfs_attr;
143 bytes_remaining -= to_read;
144 ret = call_consume_chunk(buf, to_read, cbs);
146 goto out_close_ntfs_attr;
151 out_close_ntfs_inode:
152 ntfs_inode_close(ni);
158 free_ntfs_location(struct ntfs_location *loc)
160 put_ntfs_volume(loc->volume);
161 FREE(loc->attr_name);
165 struct ntfs_location *
166 clone_ntfs_location(const struct ntfs_location *loc)
168 struct ntfs_location *new = memdup(loc, sizeof(*loc));
171 if (loc->attr_name) {
172 new->attr_name = utf16le_dup(loc->attr_name);
176 new->volume = get_ntfs_volume(loc->volume);
186 cmp_ntfs_locations(const struct ntfs_location *loc1,
187 const struct ntfs_location *loc2)
189 return cmp_u64(loc1->sort_key, loc2->sort_key);
192 /* Read rptag and rpreserved from the NTFS inode and save them in the WIM inode.
195 read_reparse_header(ntfs_inode *ni, struct wim_inode *inode)
205 na = ntfs_attr_open(ni, AT_REPARSE_POINT, AT_UNNAMED, 0);
207 return WIMLIB_ERR_NTFS_3G;
209 res = ntfs_attr_pread(na, 0, sizeof(hdr), &hdr);
213 if (res != sizeof(hdr))
214 return WIMLIB_ERR_NTFS_3G;
216 inode->i_reparse_tag = le32_to_cpu(hdr.rptag);
217 inode->i_rp_reserved = le16_to_cpu(hdr.rpreserved);
222 attr_type_to_wimlib_stream_type(ATTR_TYPES type)
226 return STREAM_TYPE_DATA;
227 case AT_REPARSE_POINT:
228 return STREAM_TYPE_REPARSE_POINT;
231 return STREAM_TYPE_UNKNOWN;
235 /* When sorting blobs located in NTFS volumes for sequential reading, we sort
236 * first by starting LCN of the attribute if available, otherwise no sort order
237 * is defined. This usually results in better sequential access to the volume.
240 set_attr_sort_key(ntfs_inode *ni, struct ntfs_location *loc)
245 na = open_ntfs_attr(ni, loc);
247 return WIMLIB_ERR_NTFS_3G;
249 rl = ntfs_attr_find_vcn(na, 0);
250 if (rl && rl->lcn != LCN_HOLE)
251 loc->sort_key = rl->lcn;
259 /* Save information about an NTFS attribute (stream) to a WIM inode. */
261 scan_ntfs_attr(struct wim_inode *inode,
264 struct list_head *unhashed_blobs,
265 struct ntfs_volume_wrapper *volume,
267 const ATTR_RECORD *record)
269 u64 data_size = ntfs_get_attribute_value_length(record);
270 const u32 name_nchars = record->name_length;
271 struct blob_descriptor *blob = NULL;
272 utf16lechar *stream_name = NULL;
273 struct wim_inode_stream *strm;
276 if (unlikely(name_nchars)) {
278 stream_name = utf16le_dupz(attr_record_name(record),
279 name_nchars * sizeof(ntfschar));
281 ret = WIMLIB_ERR_NOMEM;
286 if (unlikely(type == AT_REPARSE_POINT)) {
287 if (data_size < REPARSE_DATA_OFFSET) {
288 ERROR("Reparse point attribute of \"%s\" "
289 "is too short!", path);
290 ret = WIMLIB_ERR_INVALID_REPARSE_DATA;
293 data_size -= REPARSE_DATA_OFFSET;
295 ret = read_reparse_header(ni, inode);
297 ERROR_WITH_ERRNO("Error reading reparse point header "
303 /* If the stream is non-empty, set up a blob descriptor for it. */
304 if (data_size != 0) {
305 blob = new_blob_descriptor();
306 if (unlikely(!blob)) {
307 ret = WIMLIB_ERR_NOMEM;
311 blob->ntfs_loc = CALLOC(1, sizeof(struct ntfs_location));
312 if (unlikely(!blob->ntfs_loc)) {
313 ret = WIMLIB_ERR_NOMEM;
317 blob->blob_location = BLOB_IN_NTFS_VOLUME;
318 blob->size = data_size;
319 blob->ntfs_loc->volume = get_ntfs_volume(volume);
320 blob->ntfs_loc->attr_type = type;
321 blob->ntfs_loc->mft_no = ni->mft_no;
323 if (unlikely(name_nchars)) {
324 blob->ntfs_loc->attr_name = utf16le_dup(stream_name);
325 if (!blob->ntfs_loc->attr_name) {
326 ret = WIMLIB_ERR_NOMEM;
329 blob->ntfs_loc->attr_name_nchars = name_nchars;
332 ret = set_attr_sort_key(ni, blob->ntfs_loc);
337 strm = inode_add_stream(inode,
338 attr_type_to_wimlib_stream_type(type),
339 stream_name ? stream_name : NO_STREAM_NAME,
341 if (unlikely(!strm)) {
342 ret = WIMLIB_ERR_NOMEM;
345 prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
349 free_blob_descriptor(blob);
354 /* Scan attributes of the specified type from a file in the NTFS volume */
356 scan_ntfs_attrs_with_type(struct wim_inode *inode,
359 struct list_head *unhashed_blobs,
360 struct ntfs_volume_wrapper *volume,
363 ntfs_attr_search_ctx *actx;
366 actx = ntfs_attr_get_search_ctx(ni, NULL);
368 ERROR_WITH_ERRNO("Failed to get NTFS attribute search "
369 "context for \"%s\"", path);
370 return WIMLIB_ERR_NTFS_3G;
373 while (!ntfs_attr_lookup(type, NULL, 0,
374 CASE_SENSITIVE, 0, NULL, 0, actx))
376 ret = scan_ntfs_attr(inode,
386 if (errno != ENOENT) {
387 ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path);
388 ret = WIMLIB_ERR_NTFS_3G;
393 ntfs_attr_put_search_ctx(actx);
397 /* Load the security descriptor of an NTFS inode into the corresponding WIM
398 * inode and the WIM image's security descriptor set. */
399 static noinline_for_stack int
400 get_security_descriptor(ntfs_inode *ni, struct wim_inode *inode,
401 ntfs_volume *vol, struct wim_sd_set *sd_set)
403 struct SECURITY_CONTEXT scx = {.vol = vol};
406 size_t avail_size = sizeof(_buf);
410 ret = ntfs_get_ntfs_acl(&scx, ni, buf, avail_size);
411 if (unlikely(ret < 0)) {
412 ret = WIMLIB_ERR_NTFS_3G;
416 if (unlikely(ret > avail_size)) {
417 if (unlikely(buf != _buf))
421 ret = WIMLIB_ERR_NOMEM;
428 if (likely(ret > 0)) {
429 inode->i_security_id = sd_set_add_sd(sd_set, buf, ret);
430 if (unlikely(inode->i_security_id < 0)) {
431 ret = WIMLIB_ERR_NOMEM;
438 if (unlikely(buf != _buf))
443 /* Binary tree that maps NTFS inode numbers to DOS names */
444 struct dos_name_map {
445 struct avl_tree_node *root;
448 struct dos_name_node {
449 struct avl_tree_node index_node;
455 #define DOS_NAME_NODE(avl_node) \
456 avl_tree_entry(avl_node, struct dos_name_node, index_node)
459 _avl_cmp_by_ntfs_ino(const struct avl_tree_node *n1,
460 const struct avl_tree_node *n2)
462 return cmp_u64(DOS_NAME_NODE(n1)->ntfs_ino,
463 DOS_NAME_NODE(n2)->ntfs_ino);
466 /* Inserts a new DOS name into the map */
468 insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name,
469 size_t name_nbytes, u64 ntfs_ino)
471 struct dos_name_node *new_node;
473 new_node = MALLOC(sizeof(struct dos_name_node));
475 return WIMLIB_ERR_NOMEM;
477 /* DOS names are supposed to be 12 characters max (that's 24 bytes,
478 * assuming 2-byte ntfs characters) */
479 wimlib_assert(name_nbytes <= sizeof(new_node->dos_name));
481 /* Initialize the DOS name, DOS name length, and NTFS inode number of
482 * the search tree node */
483 memcpy(new_node->dos_name, dos_name, name_nbytes);
484 new_node->name_nbytes = name_nbytes;
485 new_node->ntfs_ino = ntfs_ino;
487 /* Insert the search tree node */
488 if (avl_tree_insert(&map->root, &new_node->index_node,
489 _avl_cmp_by_ntfs_ino))
491 /* This should be impossible since an NTFS inode cannot have
492 * multiple DOS names, and we only should get each DOS name
493 * entry once from the ntfs_readdir() calls. */
494 WARNING("NTFS inode %"PRIu64" has multiple DOS names", ntfs_ino);
500 /* Returns a structure that contains the DOS name and its length for an NTFS
501 * inode, or NULL if the inode has no DOS name. */
502 static struct dos_name_node *
503 lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino)
505 struct dos_name_node dummy;
506 struct avl_tree_node *res;
508 dummy.ntfs_ino = ntfs_ino;
510 res = avl_tree_lookup_node(map->root, &dummy.index_node,
511 _avl_cmp_by_ntfs_ino);
514 return DOS_NAME_NODE(res);
518 set_dentry_dos_name(struct wim_dentry *dentry, const struct dos_name_map *map)
520 const struct dos_name_node *node;
522 if (dentry->d_is_win32_name) {
523 node = lookup_dos_name(map, dentry->d_inode->i_ino);
525 dentry->d_short_name = utf16le_dupz(node->dos_name,
527 if (!dentry->d_short_name)
528 return WIMLIB_ERR_NOMEM;
529 dentry->d_short_name_nbytes = node->name_nbytes;
531 WARNING("NTFS inode %"PRIu64" has Win32 name with no "
532 "corresponding DOS name",
533 dentry->d_inode->i_ino);
540 destroy_dos_name_map(struct dos_name_map *map)
542 struct dos_name_node *node;
543 avl_tree_for_each_in_postorder(node, map->root,
544 struct dos_name_node, index_node)
549 struct wim_dentry *parent;
552 struct dos_name_map dos_name_map;
553 struct ntfs_volume_wrapper *volume;
554 struct capture_params *params;
559 ntfs_3g_build_dentry_tree_recursive(struct wim_dentry **root_p,
564 struct ntfs_volume_wrapper *volume,
565 struct capture_params *params);
568 filldir(void *_ctx, const ntfschar *name, const int name_nchars,
569 const int name_type, const s64 pos, const MFT_REF mref,
570 const unsigned dt_type)
572 struct readdir_ctx *ctx = _ctx;
573 const size_t name_nbytes = name_nchars * sizeof(ntfschar);
575 size_t mbs_name_nbytes;
577 struct wim_dentry *child;
580 if (name_type & FILE_NAME_DOS) {
581 /* If this is the entry for a DOS name, store it for later. */
582 ret = insert_dos_name(&ctx->dos_name_map, name,
583 name_nbytes, MREF(mref));
585 /* Return now if an error occurred or if this is just a DOS name
586 * and not a Win32+DOS name. */
587 if (ret != 0 || name_type == FILE_NAME_DOS)
591 ret = utf16le_to_tstr(name, name_nbytes, &mbs_name, &mbs_name_nbytes);
595 if (should_ignore_filename(mbs_name, mbs_name_nbytes))
596 goto out_free_mbs_name;
598 path_len = ctx->path_len;
600 ctx->path[path_len++] = '/';
601 memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1);
602 path_len += mbs_name_nbytes;
604 ret = ntfs_3g_build_dentry_tree_recursive(&child, mref, ctx->path,
606 ctx->volume, ctx->params);
607 attach_scanned_tree(ctx->parent, child, ctx->params->blob_table);
616 ntfs_3g_recurse_directory(ntfs_inode *ni, char *path, size_t path_len,
617 struct wim_dentry *parent,
618 struct ntfs_volume_wrapper *volume,
619 struct capture_params *params)
623 struct readdir_ctx ctx = {
626 .path_len = path_len,
627 .dos_name_map = { .root = NULL },
632 ret = ntfs_readdir(ni, &pos, &ctx, filldir);
633 path[path_len] = '\0';
639 /* error from ntfs_readdir() itself */
640 ERROR_WITH_ERRNO("Error reading directory \"%s\"", path);
641 ret = WIMLIB_ERR_NTFS_3G;
644 struct wim_dentry *child;
647 for_dentry_child(child, parent) {
648 ret = set_dentry_dos_name(child, &ctx.dos_name_map);
653 destroy_dos_name_map(&ctx.dos_name_map);
658 ntfs_3g_build_dentry_tree_recursive(struct wim_dentry **root_ret,
663 struct ntfs_volume_wrapper *volume,
664 struct capture_params *params)
668 struct wim_dentry *root = NULL;
669 struct wim_inode *inode = NULL;
670 ntfs_inode *ni = NULL;
672 ret = try_exclude(path, params);
673 if (unlikely(ret < 0)) /* Excluded? */
675 if (unlikely(ret > 0)) /* Error? */
678 ni = ntfs_inode_open(volume->vol, mref);
680 ERROR_WITH_ERRNO("Failed to open NTFS file \"%s\"", path);
681 ret = WIMLIB_ERR_NTFS_3G;
685 /* Get file attributes */
686 ret = ntfs_get_ntfs_attrib(ni, (char*)&attributes, sizeof(attributes));
687 if (ret != sizeof(attributes)) {
688 ERROR_WITH_ERRNO("Failed to get NTFS attributes from \"%s\"", path);
689 ret = WIMLIB_ERR_NTFS_3G;
693 if (unlikely(attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
694 if (params->add_flags & WIMLIB_ADD_FLAG_NO_UNSUPPORTED_EXCLUDE)
696 ERROR("Can't archive \"%s\" because NTFS-3g capture mode "
697 "does not support encrypted files and directories", path);
698 ret = WIMLIB_ERR_UNSUPPORTED_FILE;
701 params->progress.scan.cur_path = path;
702 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_UNSUPPORTED, NULL);
706 /* Create a WIM dentry with an associated inode, which may be shared */
707 ret = inode_table_new_dentry(params->inode_table,
708 path_basename_with_len(path, path_len),
709 ni->mft_no, 0, false, &root);
713 if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */
714 root->d_is_win32_name = 1;
716 inode = root->d_inode;
718 if (inode->i_nlink > 1) {
719 /* Shared inode; nothing more to do */
723 inode->i_creation_time = le64_to_cpu(ni->creation_time);
724 inode->i_last_write_time = le64_to_cpu(ni->last_data_change_time);
725 inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
726 inode->i_attributes = attributes;
728 if (attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
729 /* Scan the reparse point stream. */
730 ret = scan_ntfs_attrs_with_type(inode, ni, path,
731 params->unhashed_blobs,
732 volume, AT_REPARSE_POINT);
737 /* Scan the data streams.
739 * Note: directories should not have an unnamed data stream, but they
740 * may have named data streams. Nondirectories (including reparse
741 * points) can have an unnamed data stream as well as named data
743 ret = scan_ntfs_attrs_with_type(inode, ni, path, params->unhashed_blobs,
748 /* Reparse-point fixups are a no-op because in NTFS-3g capture mode we
749 * only allow capturing an entire volume. */
750 if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX &&
751 inode_is_symlink(inode))
752 inode->i_rp_flags &= ~WIM_RP_FLAG_NOT_FIXED;
754 if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
755 ret = get_security_descriptor(ni, inode, volume->vol,
758 ERROR_WITH_ERRNO("Error reading security descriptor "
764 if (inode_is_directory(inode)) {
765 ret = ntfs_3g_recurse_directory(ni, path, path_len, root,
772 params->progress.scan.cur_path = path;
774 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL);
776 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK, inode);
779 ntfs_inode_close(ni);
781 free_dentry_tree(root, params->blob_table);
783 ret = report_capture_error(params, ret, path);
790 ntfs_3g_build_dentry_tree(struct wim_dentry **root_ret,
792 struct capture_params *params)
794 struct ntfs_volume_wrapper *volume;
799 volume = MALLOC(sizeof(struct ntfs_volume_wrapper));
801 return WIMLIB_ERR_NOMEM;
803 /* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
806 * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
807 * NTFS_MNT_RDONLY is an enumerated constant. Also, the NTFS-3g headers
808 * don't seem to contain any explicit version information. So we have
809 * to rely on a test done at configure time to detect whether
810 * NTFS_MNT_RDONLY should be used. */
811 #ifdef HAVE_NTFS_MNT_RDONLY
813 vol = ntfs_mount(device, NTFS_MNT_RDONLY);
814 #elif defined(MS_RDONLY)
815 /* NTFS-3g 2011, 2012 */
816 vol = ntfs_mount(device, MS_RDONLY);
818 #error "Can't find NTFS_MNT_RDONLY or MS_RDONLY flags"
821 ERROR_WITH_ERRNO("Failed to mount NTFS volume \"%s\" read-only",
824 return WIMLIB_ERR_NTFS_3G;
830 ntfs_open_secure(vol);
832 /* We don't want to capture the special NTFS files such as $Bitmap. Not
833 * to be confused with "hidden" or "system" files which are real files
834 * that we do need to capture. */
835 NVolClearShowSysFiles(vol);
837 /* Currently we assume that all the paths fit into this length and there
838 * is no check for overflow. */
839 path = MALLOC(32768);
841 ret = WIMLIB_ERR_NOMEM;
842 goto out_put_ntfs_volume;
847 ret = ntfs_3g_build_dentry_tree_recursive(root_ret, FILE_root, path, 1,
848 FILE_NAME_POSIX, volume,
852 ntfs_index_ctx_put(vol->secure_xsii);
853 ntfs_index_ctx_put(vol->secure_xsdh);
854 ntfs_inode_close(vol->secure_ni);
855 put_ntfs_volume(volume);
858 #endif /* WITH_NTFS_3G */