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, 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 ERROR("Found two un-named data streams for \"%s\" "
249 "(sizes = %"PRIu64", %"PRIu64")",
250 path, inode->i_lte->size,
252 ret = WIMLIB_ERR_NTFS_3G;
260 /* Named data stream. Put the reference to it in the
261 * alternate data stream entries */
262 struct wim_ads_entry *new_ads_entry;
264 new_ads_entry = inode_add_ads_utf16le(inode,
265 attr_record_name(actx->attr),
267 if (!new_ads_entry) {
268 ret = WIMLIB_ERR_NOMEM;
271 wimlib_assert(new_ads_entry->stream_name_nbytes == name_length * 2);
272 stream_id = new_ads_entry->stream_id;
273 new_ads_entry->lte = lte;
276 add_unhashed_stream(lte, inode,
277 stream_id, unhashed_streams);
280 if (errno == ENOENT) {
283 ERROR_WITH_ERRNO("Error listing NTFS attributes of \"%s\"", path);
284 ret = WIMLIB_ERR_NTFS_3G;
288 free_lookup_table_entry(lte);
291 FREE(ntfs_loc->path);
292 FREE(ntfs_loc->stream_name);
296 ntfs_attr_put_search_ctx(actx);
298 DEBUG("Successfully captured NTFS streams from \"%s\"", path);
300 ERROR("Failed to capture NTFS streams from \"%s\"", path);
304 /* Red-black tree that maps NTFS inode numbers to DOS names */
305 struct dos_name_map {
306 struct rb_root rb_root;
309 struct dos_name_node {
310 struct rb_node rb_node;
316 /* Inserts a new DOS name into the map */
318 insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name,
319 size_t name_nbytes, le64 ntfs_ino)
321 struct dos_name_node *new_node;
323 struct rb_root *root;
324 struct rb_node *rb_parent;
326 DEBUG("DOS name_len = %zu", name_nbytes);
327 new_node = MALLOC(sizeof(struct dos_name_node));
331 /* DOS names are supposed to be 12 characters max (that's 24 bytes,
332 * assuming 2-byte ntfs characters) */
333 wimlib_assert(name_nbytes <= sizeof(new_node->dos_name));
335 /* Initialize the DOS name, DOS name length, and NTFS inode number of
336 * the red-black tree node */
337 memcpy(new_node->dos_name, dos_name, name_nbytes);
338 new_node->name_nbytes = name_nbytes;
339 new_node->ntfs_ino = ntfs_ino;
341 /* Insert the red-black tree node */
342 root = &map->rb_root;
346 struct dos_name_node *this;
348 this = container_of(*p, struct dos_name_node, rb_node);
350 if (new_node->ntfs_ino < this->ntfs_ino)
351 p = &((*p)->rb_left);
352 else if (new_node->ntfs_ino > this->ntfs_ino)
353 p = &((*p)->rb_right);
355 /* This should be impossible since a NTFS inode cannot
356 * have multiple DOS names, and we only should get each
357 * DOS name entry once from the ntfs_readdir() calls. */
358 ERROR("NTFS inode %"PRIu64" has multiple DOS names",
359 le64_to_cpu(ntfs_ino));
363 rb_link_node(&new_node->rb_node, rb_parent, p);
364 rb_insert_color(&new_node->rb_node, root);
365 DEBUG("Inserted DOS name for inode %"PRIu64, le64_to_cpu(ntfs_ino));
369 /* Returns a structure that contains the DOS name and its length for a NTFS
370 * inode, or NULL if the inode has no DOS name. */
371 static struct dos_name_node *
372 lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino)
374 struct rb_node *node = map->rb_root.rb_node;
376 struct dos_name_node *this;
377 this = container_of(node, struct dos_name_node, rb_node);
378 if (ntfs_ino < this->ntfs_ino)
379 node = node->rb_left;
380 else if (ntfs_ino > this->ntfs_ino)
381 node = node->rb_right;
389 set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
391 const struct dos_name_map *map = arg;
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 = MALLOC(node->name_nbytes + 2);
398 if (!dentry->short_name)
399 return WIMLIB_ERR_NOMEM;
400 memcpy(dentry->short_name, node->dos_name,
402 dentry->short_name[node->name_nbytes / 2] = 0;
403 dentry->short_name_nbytes = node->name_nbytes;
404 DEBUG("Assigned DOS name to ino %"PRIu64,
405 dentry->d_inode->i_ino);
407 WARNING("NTFS inode %"PRIu64" has Win32 name with no "
408 "corresponding DOS name",
409 dentry->d_inode->i_ino);
416 free_dos_name_tree(struct rb_node *node) {
418 free_dos_name_tree(node->rb_left);
419 free_dos_name_tree(node->rb_right);
420 FREE(container_of(node, struct dos_name_node, rb_node));
425 destroy_dos_name_map(struct dos_name_map *map)
427 free_dos_name_tree(map->rb_root.rb_node);
431 struct wim_dentry *parent;
434 struct dos_name_map *dos_name_map;
436 struct add_image_params *params;
440 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
445 ntfs_volume *ntfs_vol,
446 struct add_image_params *params);
449 wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
450 const int name_nchars, const int name_type,
451 const s64 pos, const MFT_REF mref,
452 const unsigned dt_type)
454 struct readdir_ctx *ctx;
455 size_t mbs_name_nbytes;
457 struct wim_dentry *child;
460 size_t name_nbytes = name_nchars * sizeof(ntfschar);
463 if (name_type & FILE_NAME_DOS) {
464 /* If this is the entry for a DOS name, store it for later. */
465 ret = insert_dos_name(ctx->dos_name_map, name,
466 name_nbytes, mref & MFT_REF_MASK_CPU);
468 /* Return now if an error occurred or if this is just a DOS name
469 * and not a Win32+DOS name. */
470 if (ret != 0 || name_type == FILE_NAME_DOS)
473 ret = utf16le_to_tstr(name, name_nbytes,
474 &mbs_name, &mbs_name_nbytes);
478 if (mbs_name[0] == '.' &&
479 (mbs_name[1] == '\0' ||
480 (mbs_name[1] == '.' && mbs_name[2] == '\0'))) {
483 * note: name_type is POSIX for these, so DOS names will not
484 * have been inserted for them. */
486 goto out_free_mbs_name;
489 /* Open the inode for this directory entry and recursively capture the
490 * directory tree rooted at it */
491 ntfs_inode *ni = ntfs_inode_open(ctx->vol, mref);
493 /* XXX This used to be treated as an error, but NTFS-3g seemed
494 * to be unable to read some inodes on a Windows 8 image for
496 WARNING_WITH_ERRNO("Failed to open NTFS file \"%s/%s\"",
497 ctx->path, mbs_name);
499 goto out_free_mbs_name;
501 path_len = ctx->path_len;
503 ctx->path[path_len++] = '/';
504 memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1);
505 path_len += mbs_name_nbytes;
507 ret = build_dentry_tree_ntfs_recursive(&child, ni, ctx->path,
509 ctx->vol, ctx->params);
510 path_len -= mbs_name_nbytes + 1;
512 dentry_add_child(ctx->parent, child);
513 ntfs_inode_close(ni);
517 ctx->path[ctx->path_len] = '\0';
521 /* Recursively build a WIM dentry tree corresponding to a NTFS volume.
522 * At the same time, update the WIM lookup table with lookup table entries for
523 * the NTFS streams, and build an array of security descriptors.
526 build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
532 struct add_image_params *params)
536 struct wim_dentry *root = NULL;
537 struct wim_inode *inode = NULL;
538 ATTR_TYPES stream_type;
540 if (exclude_path(path, path_len, params->config, false)) {
541 /* Exclude a file or directory tree based on the capture
542 * configuration file. */
547 /* Get file attributes */
548 ret = ntfs_get_ntfs_attrib(ni, (char*)&attributes, sizeof(attributes));
549 if (ret != sizeof(attributes)) {
550 ERROR_WITH_ERRNO("Failed to get NTFS attributes from \"%s\"", path);
551 ret = WIMLIB_ERR_NTFS_3G;
555 if ((attributes & (FILE_ATTRIBUTE_DIRECTORY |
556 FILE_ATTRIBUTE_ENCRYPTED)) == FILE_ATTRIBUTE_ENCRYPTED)
558 if (params->add_flags & WIMLIB_ADD_FLAG_NO_UNSUPPORTED_EXCLUDE)
560 ERROR("Can't archive unsupported encrypted file \"%s\"", path);
561 ret = WIMLIB_ERR_UNSUPPORTED_FILE;
564 params->progress.scan.cur_path = path;
565 do_capture_progress(params, WIMLIB_SCAN_DENTRY_UNSUPPORTED, NULL);
570 /* Create a WIM dentry with an associated inode, which may be shared */
571 ret = inode_table_new_dentry(¶ms->inode_table,
572 path_basename_with_len(path, path_len),
573 ni->mft_no, 0, false, &root);
577 if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */
578 root->is_win32_name = 1;
580 inode = root->d_inode;
582 if (inode->i_nlink > 1) {
583 /* Shared inode; nothing more to do */
588 inode->i_creation_time = le64_to_cpu(ni->creation_time);
589 inode->i_last_write_time = le64_to_cpu(ni->last_data_change_time);
590 inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
591 inode->i_attributes = le32_to_cpu(attributes);
592 inode->i_resolved = 1;
594 if (attributes & FILE_ATTR_REPARSE_POINT)
595 stream_type = AT_REPARSE_POINT;
597 stream_type = AT_DATA;
599 /* Capture the file's streams; more specifically, this is supposed to:
601 * - Regular files: capture unnamed data stream and any named data
603 * - Directories: capture any named data streams
604 * - Reparse points: capture reparse data only
606 ret = capture_ntfs_streams(inode, ni, path, path_len,
607 params->unhashed_streams, vol, stream_type);
611 if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) {
613 /* Recurse to directory children */
615 struct dos_name_map dos_name_map = { .rb_root = {.rb_node = NULL} };
616 struct readdir_ctx ctx = {
619 .path_len = path_len,
620 .dos_name_map = &dos_name_map,
624 ret = ntfs_readdir(ni, &pos, &ctx, wim_ntfs_capture_filldir);
626 ERROR_WITH_ERRNO("Error reading directory \"%s\"", path);
627 ret = WIMLIB_ERR_NTFS_3G;
629 ret = for_dentry_child(root, set_dentry_dos_name,
632 destroy_dos_name_map(&dos_name_map);
636 path[path_len] = '\0';
638 /* Reparse-point fixups are a no-op because in NTFS-3g capture mode we
639 * only allow capturing an entire volume. */
640 if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX &&
641 inode_is_symlink(inode))
642 inode->i_not_rpfixed = 0;
644 if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
645 struct SECURITY_CONTEXT sec_ctx;
649 /* Get security descriptor */
650 memset(&sec_ctx, 0, sizeof(sec_ctx));
655 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, sizeof(_sd));
656 if (ret > sizeof(_sd)) {
658 ret = ntfs_get_ntfs_acl(&sec_ctx, ni, sd, ret);
661 inode->i_security_id = sd_set_add_sd(¶ms->sd_set,
663 if (inode->i_security_id == -1) {
664 ERROR("Out of memory");
665 ret = WIMLIB_ERR_NOMEM;
668 DEBUG("Added security ID = %u for `%s'",
669 inode->i_security_id, path);
671 } else if (ret < 0) {
672 ERROR_WITH_ERRNO("Failed to get security information from "
674 ret = WIMLIB_ERR_NTFS_3G;
676 inode->i_security_id = -1;
677 DEBUG("No security ID for `%s'", path);
684 params->progress.scan.cur_path = path;
686 do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL);
688 do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK, inode);
693 free_dentry_tree(root, params->lookup_table);
699 do_ntfs_umount(struct _ntfs_volume *vol)
701 DEBUG("Unmounting NTFS volume");
702 if (ntfs_umount(vol, FALSE))
703 return WIMLIB_ERR_NTFS_3G;
709 build_dentry_tree_ntfs(struct wim_dentry **root_p,
711 struct add_image_params *params)
717 DEBUG("Mounting NTFS volume `%s' read-only", device);
719 /* NTFS-3g 2013 renamed the "read-only" mount flag from MS_RDONLY to
722 * Unfortunately we can't check for defined(NTFS_MNT_RDONLY) because
723 * NTFS_MNT_RDONLY is an enumerated constant. Also, the NTFS-3g headers don't
724 * seem to contain any explicit version information. So we have to rely on a
725 * test done at configure time to detect whether NTFS_MNT_RDONLY should be used.
727 #ifdef HAVE_NTFS_MNT_RDONLY
729 vol = ntfs_mount(device, NTFS_MNT_RDONLY);
730 #elif defined(MS_RDONLY)
731 /* NTFS-3g 2011, 2012 */
732 vol = ntfs_mount(device, MS_RDONLY);
734 #error "Can't find NTFS_MNT_RDONLY or MS_RDONLY flags"
737 ERROR_WITH_ERRNO("Failed to mount NTFS volume `%s' read-only",
739 return WIMLIB_ERR_NTFS_3G;
741 ntfs_open_secure(vol);
743 /* We don't want to capture the special NTFS files such as $Bitmap. Not
744 * to be confused with "hidden" or "system" files which are real files
745 * that we do need to capture. */
746 NVolClearShowSysFiles(vol);
748 DEBUG("Opening root NTFS dentry");
749 root_ni = ntfs_inode_open(vol, FILE_root);
751 ERROR_WITH_ERRNO("Failed to open root inode of NTFS volume "
753 ret = WIMLIB_ERR_NTFS_3G;
757 /* Currently we assume that all the paths fit into this length and there
758 * is no check for overflow. */
759 char *path = MALLOC(32768);
761 ERROR("Could not allocate memory for NTFS pathname");
762 ret = WIMLIB_ERR_NOMEM;
768 ret = build_dentry_tree_ntfs_recursive(root_p, root_ni, path, 1,
769 FILE_NAME_POSIX, vol, params);
772 ntfs_inode_close(root_ni);
774 ntfs_index_ctx_put(vol->secure_xsii);
775 ntfs_index_ctx_put(vol->secure_xsdh);
776 ntfs_inode_close(vol->secure_ni);
779 if (do_ntfs_umount(vol)) {
780 ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'",
783 ret = WIMLIB_ERR_NTFS_3G;
786 /* We need to leave the NTFS volume mounted so that we can read
787 * the NTFS files again when we are actually writing the WIM */
788 *(ntfs_volume**)params->extra_arg = vol;
792 #endif /* WITH_NTFS_3G */