2 * win32_capture.c - Windows-specific code for capturing files into a WIM image.
4 * This now uses the native Windows NT API a lot and not just Win32.
8 * Copyright (C) 2013, 2014, 2015 Eric Biggers
10 * This file is free software; you can redistribute it and/or modify it under
11 * the terms of the GNU Lesser General Public License as published by the Free
12 * Software Foundation; either version 3 of the License, or (at your option) any
15 * This file is distributed in the hope that it will be useful, but WITHOUT
16 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
17 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this file; if not, see http://www.gnu.org/licenses/.
30 #include "wimlib/win32_common.h"
32 #include "wimlib/assert.h"
33 #include "wimlib/blob_table.h"
34 #include "wimlib/capture.h"
35 #include "wimlib/dentry.h"
36 #include "wimlib/encoding.h"
37 #include "wimlib/endianness.h"
38 #include "wimlib/error.h"
39 #include "wimlib/paths.h"
40 #include "wimlib/reparse.h"
42 struct winnt_scan_stats {
43 unsigned long num_get_sd_access_denied;
44 unsigned long num_get_sacl_priv_notheld;
47 static inline const wchar_t *
48 printable_path(const wchar_t *full_path)
50 /* Skip over \\?\ or \??\ */
55 * If cur_dir is not NULL, open an existing file relative to the already-open
58 * Otherwise, open the file specified by @path, which must be a Windows NT
62 winnt_openat(HANDLE cur_dir, const wchar_t *path, size_t path_nchars,
63 ACCESS_MASK perms, HANDLE *h_ret)
66 OBJECT_ATTRIBUTES attr;
70 name.Length = path_nchars * sizeof(wchar_t);
71 name.MaximumLength = name.Length + sizeof(wchar_t);
72 name.Buffer = (wchar_t *)path;
74 attr.Length = sizeof(attr);
75 attr.RootDirectory = cur_dir;
76 attr.ObjectName = &name;
78 attr.SecurityDescriptor = NULL;
79 attr.SecurityQualityOfService = NULL;
82 status = (*func_NtOpenFile)(h_ret, perms, &attr, &iosb,
83 FILE_SHARE_VALID_FLAGS,
84 FILE_OPEN_REPARSE_POINT |
85 FILE_OPEN_FOR_BACKUP_INTENT |
86 FILE_SYNCHRONOUS_IO_NONALERT |
87 FILE_SEQUENTIAL_ONLY);
88 if (!NT_SUCCESS(status)) {
89 /* Try requesting fewer permissions */
90 if (status == STATUS_ACCESS_DENIED ||
91 status == STATUS_PRIVILEGE_NOT_HELD) {
92 if (perms & ACCESS_SYSTEM_SECURITY) {
93 perms &= ~ACCESS_SYSTEM_SECURITY;
96 if (perms & READ_CONTROL) {
97 perms &= ~READ_CONTROL;
105 /* Read the first @size bytes from the file, or named data stream of a file,
106 * described by @blob. */
108 read_winnt_stream_prefix(const struct blob_descriptor *blob, u64 size,
109 const struct read_blob_callbacks *cbs)
118 /* This is an NT namespace path. */
119 path = blob->file_on_disk;
121 status = winnt_openat(NULL, path, wcslen(path),
122 FILE_READ_DATA | SYNCHRONIZE, &h);
123 if (!NT_SUCCESS(status)) {
124 winnt_error(status, L"\"%ls\": Can't open for reading",
125 printable_path(path));
126 return WIMLIB_ERR_OPEN;
130 bytes_remaining = size;
131 while (bytes_remaining) {
132 IO_STATUS_BLOCK iosb;
136 count = min(sizeof(buf), bytes_remaining);
138 status = (*func_NtReadFile)(h, NULL, NULL, NULL,
139 &iosb, buf, count, NULL, NULL);
140 if (!NT_SUCCESS(status)) {
141 winnt_error(status, L"\"%ls\": Error reading data",
142 printable_path(path));
143 ret = WIMLIB_ERR_READ;
147 bytes_read = iosb.Information;
149 bytes_remaining -= bytes_read;
150 ret = call_consume_chunk(buf, bytes_read, cbs);
158 struct win32_encrypted_read_ctx {
159 const struct read_blob_callbacks *cbs;
165 win32_encrypted_export_cb(unsigned char *data, void *_ctx, unsigned long len)
167 struct win32_encrypted_read_ctx *ctx = _ctx;
169 size_t bytes_to_consume = min(len, ctx->bytes_remaining);
171 if (bytes_to_consume == 0)
172 return ERROR_SUCCESS;
174 ret = call_consume_chunk(data, bytes_to_consume, ctx->cbs);
176 ctx->wimlib_err_code = ret;
177 /* It doesn't matter what error code is returned here, as long
178 * as it isn't ERROR_SUCCESS. */
179 return ERROR_READ_FAULT;
181 ctx->bytes_remaining -= bytes_to_consume;
182 return ERROR_SUCCESS;
186 read_win32_encrypted_file_prefix(const struct blob_descriptor *blob,
188 const struct read_blob_callbacks *cbs)
190 struct win32_encrypted_read_ctx export_ctx;
196 if (blob->file_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
197 flags |= CREATE_FOR_DIR;
199 export_ctx.cbs = cbs;
200 export_ctx.wimlib_err_code = 0;
201 export_ctx.bytes_remaining = size;
203 err = OpenEncryptedFileRaw(blob->file_on_disk, flags, &file_ctx);
204 if (err != ERROR_SUCCESS) {
206 L"Failed to open encrypted file \"%ls\" for raw read",
207 printable_path(blob->file_on_disk));
208 return WIMLIB_ERR_OPEN;
210 err = ReadEncryptedFileRaw(win32_encrypted_export_cb,
211 &export_ctx, file_ctx);
212 if (err != ERROR_SUCCESS) {
213 ret = export_ctx.wimlib_err_code;
216 L"Failed to read encrypted file \"%ls\"",
217 printable_path(blob->file_on_disk));
218 ret = WIMLIB_ERR_READ;
220 } else if (export_ctx.bytes_remaining != 0) {
221 ERROR("Only could read %"PRIu64" of %"PRIu64" bytes from "
222 "encrypted file \"%ls\"",
223 size - export_ctx.bytes_remaining, size,
224 printable_path(blob->file_on_disk));
225 ret = WIMLIB_ERR_READ;
229 CloseEncryptedFileRaw(file_ctx);
234 * Load the short name of a file into a WIM dentry.
236 static noinline_for_stack NTSTATUS
237 winnt_get_short_name(HANDLE h, struct wim_dentry *dentry)
239 /* It's not any harder to just make the NtQueryInformationFile() system
240 * call ourselves, and it saves a dumb call to FindFirstFile() which of
241 * course has to create its own handle. */
243 IO_STATUS_BLOCK iosb;
244 u8 buf[128] _aligned_attribute(8);
245 const FILE_NAME_INFORMATION *info;
247 status = (*func_NtQueryInformationFile)(h, &iosb, buf, sizeof(buf),
248 FileAlternateNameInformation);
249 info = (const FILE_NAME_INFORMATION *)buf;
250 if (NT_SUCCESS(status) && info->FileNameLength != 0) {
251 dentry->d_short_name = utf16le_dupz(info->FileName,
252 info->FileNameLength);
253 if (!dentry->d_short_name)
254 return STATUS_NO_MEMORY;
255 dentry->d_short_name_nbytes = info->FileNameLength;
261 * Load the security descriptor of a file into the corresponding inode and the
262 * WIM image's security descriptor set.
264 static noinline_for_stack NTSTATUS
265 winnt_get_security_descriptor(HANDLE h, struct wim_inode *inode,
266 struct wim_sd_set *sd_set,
267 struct winnt_scan_stats *stats, int add_flags)
269 SECURITY_INFORMATION requestedInformation;
270 u8 _buf[4096] _aligned_attribute(8);
277 * LABEL_SECURITY_INFORMATION is needed on Windows Vista and 7 because
278 * Microsoft decided to add mandatory integrity labels to the SACL but
279 * not have them returned by SACL_SECURITY_INFORMATION.
281 * BACKUP_SECURITY_INFORMATION is needed on Windows 8 because Microsoft
282 * decided to add even more stuff to the SACL and still not have it
283 * returned by SACL_SECURITY_INFORMATION; but they did remember that
284 * backup applications exist and simply want to read the stupid thing
285 * once and for all, so they added a flag to read the entire security
288 * Older versions of Windows tolerate these new flags being passed in.
290 requestedInformation = OWNER_SECURITY_INFORMATION |
291 GROUP_SECURITY_INFORMATION |
292 DACL_SECURITY_INFORMATION |
293 SACL_SECURITY_INFORMATION |
294 LABEL_SECURITY_INFORMATION |
295 BACKUP_SECURITY_INFORMATION;
298 bufsize = sizeof(_buf);
301 * We need the file's security descriptor in
302 * SECURITY_DESCRIPTOR_RELATIVE format, and we currently have a handle
303 * opened with as many relevant permissions as possible. At this point,
304 * on Windows there are a number of options for reading a file's
305 * security descriptor:
307 * GetFileSecurity(): This takes in a path and returns the
308 * SECURITY_DESCRIPTOR_RELATIVE. Problem: this uses an internal handle,
309 * not ours, and the handle created internally doesn't specify
310 * FILE_FLAG_BACKUP_SEMANTICS. Therefore there can be access denied
311 * errors on some files and directories, even when running as the
314 * GetSecurityInfo(): This takes in a handle and returns the security
315 * descriptor split into a bunch of different parts. This should work,
316 * but it's dumb because we have to put the security descriptor back
319 * BackupRead(): This can read the security descriptor, but this is a
320 * difficult-to-use API, probably only works as the Administrator, and
321 * the format of the returned data is not well documented.
323 * NtQuerySecurityObject(): This is exactly what we need, as it takes
324 * in a handle and returns the security descriptor in
325 * SECURITY_DESCRIPTOR_RELATIVE format. Only problem is that it's a
326 * ntdll function and therefore not officially part of the Win32 API.
329 while (!(NT_SUCCESS(status = (*func_NtQuerySecurityObject)(h,
330 requestedInformation,
331 (PSECURITY_DESCRIPTOR)buf,
336 case STATUS_BUFFER_TOO_SMALL:
337 wimlib_assert(buf == _buf);
338 buf = MALLOC(len_needed);
340 return STATUS_NO_MEMORY;
341 bufsize = len_needed;
343 case STATUS_PRIVILEGE_NOT_HELD:
344 case STATUS_ACCESS_DENIED:
345 if (add_flags & WIMLIB_ADD_FLAG_STRICT_ACLS) {
347 /* Permission denied in STRICT_ACLS mode, or
351 if (requestedInformation & SACL_SECURITY_INFORMATION) {
352 /* Try again without the SACL. */
353 stats->num_get_sacl_priv_notheld++;
354 requestedInformation &= ~(SACL_SECURITY_INFORMATION |
355 LABEL_SECURITY_INFORMATION |
356 BACKUP_SECURITY_INFORMATION);
359 /* Fake success (useful when capturing as
360 * non-Administrator). */
361 stats->num_get_sd_access_denied++;
362 status = STATUS_SUCCESS;
367 /* Add the security descriptor to the WIM image, and save its ID in
369 inode->i_security_id = sd_set_add_sd(sd_set, buf, len_needed);
370 if (unlikely(inode->i_security_id < 0))
371 status = STATUS_NO_MEMORY;
373 if (unlikely(buf != _buf))
379 winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret,
382 size_t full_path_nchars,
383 const wchar_t *filename,
384 size_t filename_nchars,
385 struct capture_params *params,
386 struct winnt_scan_stats *stats,
390 winnt_recurse_directory(HANDLE h,
392 size_t full_path_nchars,
393 struct wim_dentry *parent,
394 struct capture_params *params,
395 struct winnt_scan_stats *stats,
399 const size_t bufsize = 8192;
400 IO_STATUS_BLOCK iosb;
404 buf = MALLOC(bufsize);
406 return WIMLIB_ERR_NOMEM;
408 /* Using NtQueryDirectoryFile() we can re-use the same open handle,
409 * which we opened with FILE_FLAG_BACKUP_SEMANTICS. */
411 while (NT_SUCCESS(status = (*func_NtQueryDirectoryFile)(h, NULL, NULL, NULL,
413 FileNamesInformation,
414 FALSE, NULL, FALSE)))
416 const FILE_NAMES_INFORMATION *info = buf;
418 if (!should_ignore_filename(info->FileName,
419 info->FileNameLength / 2))
423 struct wim_dentry *child;
425 p = full_path + full_path_nchars;
426 /* Only add a backslash if we don't already have
427 * one. This prevents a duplicate backslash
428 * from being added when the path to the capture
429 * dir had a trailing backslash. */
430 if (*(p - 1) != L'\\')
433 p = wmempcpy(filename, info->FileName,
434 info->FileNameLength / 2);
437 ret = winnt_build_dentry_tree_recursive(
443 info->FileNameLength / 2,
448 full_path[full_path_nchars] = L'\0';
452 attach_scanned_tree(parent, child, params->blob_table);
454 if (info->NextEntryOffset == 0)
456 info = (const FILE_NAMES_INFORMATION *)
457 ((const u8 *)info + info->NextEntryOffset);
461 if (unlikely(status != STATUS_NO_MORE_FILES)) {
462 winnt_error(status, L"\"%ls\": Can't read directory",
463 printable_path(full_path));
464 ret = WIMLIB_ERR_READ;
471 /* Reparse point fixup status code */
472 #define RP_FIXED (-1)
475 file_has_ino_and_dev(HANDLE h, u64 ino, u64 dev)
478 IO_STATUS_BLOCK iosb;
479 FILE_INTERNAL_INFORMATION int_info;
480 FILE_FS_VOLUME_INFORMATION vol_info;
482 status = (*func_NtQueryInformationFile)(h, &iosb,
483 &int_info, sizeof(int_info),
484 FileInternalInformation);
485 if (!NT_SUCCESS(status))
488 if (int_info.IndexNumber.QuadPart != ino)
491 status = (*func_NtQueryVolumeInformationFile)(h, &iosb,
492 &vol_info, sizeof(vol_info),
493 FileFsVolumeInformation);
494 if (!(NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW))
497 if (iosb.Information <
498 offsetof(FILE_FS_VOLUME_INFORMATION, VolumeSerialNumber) +
499 sizeof(vol_info.VolumeSerialNumber))
502 return (vol_info.VolumeSerialNumber == dev);
506 * This is the Windows equivalent of unix_relativize_link_target(); see there
507 * for general details. This version works with an "absolute" Windows link
508 * target, specified from the root of the Windows kernel object namespace. Note
509 * that we have to open directories with a trailing slash when present because
510 * \??\E: opens the E: device itself and not the filesystem root directory.
512 static const wchar_t *
513 winnt_relativize_link_target(const wchar_t *target, size_t target_nbytes,
517 OBJECT_ATTRIBUTES attr;
518 IO_STATUS_BLOCK iosb;
520 const wchar_t *target_end;
523 target_end = target + (target_nbytes / sizeof(wchar_t));
526 if (target_end == target)
529 /* No leading slash??? */
530 if (target[0] != L'\\')
534 if ((target_end - target) >= 2 &&
535 target[0] == L'\\' && target[1] == L'\\')
538 attr.Length = sizeof(attr);
539 attr.RootDirectory = NULL;
540 attr.ObjectName = &name;
542 attr.SecurityDescriptor = NULL;
543 attr.SecurityQualityOfService = NULL;
545 name.Buffer = (wchar_t *)target;
550 const wchar_t *orig_p = p;
552 /* Skip non-backslashes */
553 while (p != target_end && *p != L'\\')
556 /* Skip backslashes */
557 while (p != target_end && *p == L'\\')
560 /* Append path component */
561 name.Length += (p - orig_p) * sizeof(wchar_t);
562 name.MaximumLength = name.Length;
564 /* Try opening the file */
565 status = (*func_NtOpenFile) (&h,
566 FILE_READ_ATTRIBUTES | FILE_TRAVERSE,
569 FILE_SHARE_VALID_FLAGS,
570 FILE_OPEN_FOR_BACKUP_INTENT);
572 if (NT_SUCCESS(status)) {
573 /* Reset root directory */
574 if (attr.RootDirectory)
575 (*func_NtClose)(attr.RootDirectory);
576 attr.RootDirectory = h;
577 name.Buffer = (wchar_t *)p;
580 if (file_has_ino_and_dev(h, ino, dev))
581 goto out_close_root_dir;
583 } while (p != target_end);
588 if (attr.RootDirectory)
589 (*func_NtClose)(attr.RootDirectory);
590 while (p > target && *(p - 1) == L'\\')
596 winnt_rpfix_progress(struct capture_params *params, const wchar_t *path,
597 const struct link_reparse_point *link, int scan_status)
599 size_t print_name_nchars = link->print_name_nbytes / sizeof(wchar_t);
600 wchar_t print_name0[print_name_nchars + 1];
602 wmemcpy(print_name0, link->print_name, print_name_nchars);
603 print_name0[print_name_nchars] = L'\0';
605 params->progress.scan.cur_path = printable_path(path);
606 params->progress.scan.symlink_target = print_name0;
607 return do_capture_progress(params, scan_status, NULL);
611 winnt_try_rpfix(struct reparse_buffer_disk *rpbuf, u16 *rpbuflen_p,
612 const wchar_t *path, struct capture_params *params)
614 struct link_reparse_point link;
615 const wchar_t *rel_target;
618 if (parse_link_reparse_point(rpbuf, *rpbuflen_p, &link)) {
619 /* Couldn't understand the reparse data; don't do the fixup. */
624 * Don't do reparse point fixups on relative symbolic links.
626 * On Windows, a relative symbolic link is supposed to be identifiable
627 * by having reparse tag WIM_IO_REPARSE_TAG_SYMLINK and flags
628 * SYMBOLIC_LINK_RELATIVE. We will use this information, although this
629 * may not always do what the user expects, since drive-relative
630 * symbolic links such as "\Users\Public" have SYMBOLIC_LINK_RELATIVE
631 * set, in addition to truely relative symbolic links such as "Users" or
632 * "Users\Public". However, WIMGAPI (as of Windows 8.1) has this same
635 * Otherwise, as far as I can tell, the targets of symbolic links that
636 * are NOT relative, as well as junctions (note: a mountpoint is the
637 * sames thing as a junction), must be NT namespace paths, for example:
639 * - \??\e:\Users\Public
640 * - \DosDevices\e:\Users\Public
641 * - \Device\HardDiskVolume4\Users\Public
642 * - \??\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public
643 * - \DosDevices\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public
645 if (link_is_relative_symlink(&link))
648 rel_target = winnt_relativize_link_target(link.substitute_name,
649 link.substitute_name_nbytes,
650 params->capture_root_ino,
651 params->capture_root_dev);
653 if (rel_target == link.substitute_name) {
654 /* Target points outside of the tree being captured or had an
655 * unrecognized path format. Don't adjust it. */
656 return winnt_rpfix_progress(params, path, &link,
657 WIMLIB_SCAN_DENTRY_NOT_FIXED_SYMLINK);
660 /* We have an absolute target pointing within the directory being
661 * captured. @rel_target is the suffix of the link target that is the
662 * part relative to the directory being captured.
664 * We will cut off the prefix before this part (which is the path to the
665 * directory being captured) and add a dummy prefix. Since the process
666 * will need to be reversed when applying the image, it doesn't matter
667 * what exactly the prefix is, as long as it looks like an absolute
670 static const wchar_t prefix[6] = L"\\??\\X:";
671 static const size_t num_unprintable_chars = 4;
673 size_t rel_target_nbytes =
674 link.substitute_name_nbytes - ((const u8 *)rel_target -
675 (const u8 *)link.substitute_name);
677 wchar_t tmp[(sizeof(prefix) + rel_target_nbytes) / sizeof(wchar_t)];
679 memcpy(tmp, prefix, sizeof(prefix));
680 memcpy(tmp + ARRAY_LEN(prefix), rel_target, rel_target_nbytes);
682 link.substitute_name = tmp;
683 link.substitute_name_nbytes = sizeof(tmp);
685 link.print_name = link.substitute_name + num_unprintable_chars;
686 link.print_name_nbytes = link.substitute_name_nbytes -
687 (num_unprintable_chars * sizeof(wchar_t));
689 if (make_link_reparse_point(&link, rpbuf, rpbuflen_p))
692 ret = winnt_rpfix_progress(params, path, &link,
693 WIMLIB_SCAN_DENTRY_FIXED_SYMLINK);
699 /* Load the reparse data of a file into the corresponding WIM inode. If the
700 * reparse point is a symbolic link or junction with an absolute target and
701 * RPFIX mode is enabled, then also rewrite its target to be relative to the
703 static noinline_for_stack int
704 winnt_load_reparse_data(HANDLE h, struct wim_inode *inode,
705 const wchar_t *full_path, struct capture_params *params)
707 struct reparse_buffer_disk rpbuf;
708 DWORD bytes_returned;
712 if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
713 /* See comment above assign_stream_types_encrypted() */
714 WARNING("Ignoring reparse data of encrypted file \"%ls\"",
715 printable_path(full_path));
719 if (!DeviceIoControl(h, FSCTL_GET_REPARSE_POINT,
720 NULL, 0, &rpbuf, REPARSE_POINT_MAX_SIZE,
721 &bytes_returned, NULL))
723 win32_error(GetLastError(), L"\"%ls\": Can't get reparse point",
724 printable_path(full_path));
725 return WIMLIB_ERR_READLINK;
728 rpbuflen = bytes_returned;
730 if (unlikely(rpbuflen < REPARSE_DATA_OFFSET)) {
731 ERROR("\"%ls\": reparse point buffer is too short",
732 printable_path(full_path));
733 return WIMLIB_ERR_INVALID_REPARSE_DATA;
736 if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX) {
737 ret = winnt_try_rpfix(&rpbuf, &rpbuflen, full_path, params);
739 inode->i_rp_flags &= ~WIM_RP_FLAG_NOT_FIXED;
744 inode->i_reparse_tag = le32_to_cpu(rpbuf.rptag);
745 inode->i_rp_reserved = le16_to_cpu(rpbuf.rpreserved);
747 if (!inode_add_stream_with_data(inode,
748 STREAM_TYPE_REPARSE_POINT,
751 rpbuflen - REPARSE_DATA_OFFSET,
753 return WIMLIB_ERR_NOMEM;
759 win32_tally_encrypted_size_cb(unsigned char *_data, void *_size_ret,
762 *(u64*)_size_ret += len;
763 return ERROR_SUCCESS;
767 win32_get_encrypted_file_size(const wchar_t *path, bool is_dir, u64 *size_ret)
775 flags |= CREATE_FOR_DIR;
777 err = OpenEncryptedFileRaw(path, flags, &file_ctx);
778 if (err != ERROR_SUCCESS) {
780 L"Failed to open encrypted file \"%ls\" for raw read",
781 printable_path(path));
782 return WIMLIB_ERR_OPEN;
785 err = ReadEncryptedFileRaw(win32_tally_encrypted_size_cb,
787 if (err != ERROR_SUCCESS) {
789 L"Failed to read raw encrypted data from \"%ls\"",
790 printable_path(path));
791 ret = WIMLIB_ERR_READ;
795 CloseEncryptedFileRaw(file_ctx);
800 winnt_scan_efsrpc_raw_data(struct wim_inode *inode, const wchar_t *nt_path,
801 struct list_head *unhashed_blobs)
803 struct blob_descriptor *blob;
804 struct wim_inode_stream *strm;
807 blob = new_blob_descriptor();
811 blob->file_on_disk = WCSDUP(nt_path);
812 if (!blob->file_on_disk)
814 blob->blob_location = BLOB_WIN32_ENCRYPTED;
816 /* OpenEncryptedFileRaw() expects a Win32 name. */
817 wimlib_assert(!wmemcmp(blob->file_on_disk, L"\\??\\", 4));
818 blob->file_on_disk[1] = L'\\';
820 blob->file_inode = inode;
822 ret = win32_get_encrypted_file_size(blob->file_on_disk,
823 (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY),
828 /* Empty EFSRPC data does not make sense */
829 wimlib_assert(blob->size != 0);
831 strm = inode_add_stream(inode, STREAM_TYPE_EFSRPC_RAW_DATA,
832 NO_STREAM_NAME, blob);
836 prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
840 ret = WIMLIB_ERR_NOMEM;
842 free_blob_descriptor(blob);
847 get_data_stream_name(wchar_t *raw_stream_name, size_t raw_stream_name_nchars,
848 wchar_t **stream_name_ret, size_t *stream_name_nchars_ret)
850 const wchar_t *sep, *type, *end;
852 /* The stream name should be returned as :NAME:TYPE */
853 if (raw_stream_name_nchars < 1)
855 if (raw_stream_name[0] != L':')
859 raw_stream_name_nchars--;
861 end = raw_stream_name + raw_stream_name_nchars;
863 sep = wmemchr(raw_stream_name, L':', raw_stream_name_nchars);
871 if (wmemcmp(type, L"$DATA", 5))
874 *stream_name_ret = raw_stream_name;
875 *stream_name_nchars_ret = sep - raw_stream_name;
879 /* Build the path to the data stream. For unnamed streams, this is simply the
880 * path to the file. For named streams, this is the path to the file, followed
881 * by a colon, followed by the stream name. */
883 build_data_stream_path(const wchar_t *path, size_t path_nchars,
884 const wchar_t *stream_name, size_t stream_name_nchars)
886 size_t stream_path_nchars;
887 wchar_t *stream_path;
890 stream_path_nchars = path_nchars;
891 if (stream_name_nchars)
892 stream_path_nchars += 1 + stream_name_nchars;
894 stream_path = MALLOC((stream_path_nchars + 1) * sizeof(wchar_t));
896 p = wmempcpy(stream_path, path, path_nchars);
897 if (stream_name_nchars) {
899 p = wmempcpy(p, stream_name, stream_name_nchars);
907 winnt_scan_data_stream(const wchar_t *path, size_t path_nchars,
908 wchar_t *raw_stream_name, size_t raw_stream_name_nchars,
910 struct wim_inode *inode, struct list_head *unhashed_blobs)
912 wchar_t *stream_name;
913 size_t stream_name_nchars;
914 struct blob_descriptor *blob;
915 struct wim_inode_stream *strm;
917 /* Given the raw stream name (which is something like
918 * :streamname:$DATA), extract just the stream name part (streamname).
919 * Ignore any non-$DATA streams. */
920 if (!get_data_stream_name(raw_stream_name, raw_stream_name_nchars,
921 &stream_name, &stream_name_nchars))
924 stream_name[stream_name_nchars] = L'\0';
926 /* If the stream is non-empty, set up a blob descriptor for it. */
927 if (stream_size != 0) {
928 blob = new_blob_descriptor();
931 blob->file_on_disk = build_data_stream_path(path,
935 if (!blob->file_on_disk)
937 blob->blob_location = BLOB_IN_WINNT_FILE_ON_DISK;
938 blob->size = stream_size;
939 blob->file_inode = inode;
944 strm = inode_add_stream(inode, STREAM_TYPE_DATA, stream_name, blob);
948 prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
952 free_blob_descriptor(blob);
953 return WIMLIB_ERR_NOMEM;
957 * Load information about the data streams of an open file into a WIM inode.
959 * We use the NtQueryInformationFile() system call instead of FindFirstStream()
960 * and FindNextStream(). This is done for two reasons:
962 * - FindFirstStream() opens its own handle to the file or directory and
963 * apparently does so without specifying FILE_FLAG_BACKUP_SEMANTICS, thereby
964 * causing access denied errors on certain files (even when running as the
966 * - FindFirstStream() and FindNextStream() is only available on Windows Vista
967 * and later, whereas the stream support in NtQueryInformationFile() was
968 * already present in Windows XP.
970 static noinline_for_stack int
971 winnt_scan_data_streams(HANDLE h, const wchar_t *path, size_t path_nchars,
972 struct wim_inode *inode, struct list_head *unhashed_blobs,
973 u64 file_size, u32 vol_flags)
976 u8 _buf[4096] _aligned_attribute(8);
979 IO_STATUS_BLOCK iosb;
981 FILE_STREAM_INFORMATION *info;
984 bufsize = sizeof(_buf);
986 if (!(vol_flags & FILE_NAMED_STREAMS))
989 /* Get a buffer containing the stream information. */
990 while (!NT_SUCCESS(status = (*func_NtQueryInformationFile)(h,
994 FileStreamInformation)))
998 case STATUS_BUFFER_OVERFLOW:
1004 newbuf = MALLOC(bufsize);
1006 newbuf = REALLOC(buf, bufsize);
1008 ret = WIMLIB_ERR_NOMEM;
1014 case STATUS_NOT_IMPLEMENTED:
1015 case STATUS_NOT_SUPPORTED:
1016 case STATUS_INVALID_INFO_CLASS:
1020 L"\"%ls\": Failed to query stream information",
1021 printable_path(path));
1022 ret = WIMLIB_ERR_READ;
1027 if (iosb.Information == 0) {
1028 /* No stream information. */
1033 /* Parse one or more stream information structures. */
1034 info = (FILE_STREAM_INFORMATION *)buf;
1036 /* Load the stream information. */
1037 ret = winnt_scan_data_stream(path, path_nchars,
1039 info->StreamNameLength / 2,
1040 info->StreamSize.QuadPart,
1041 inode, unhashed_blobs);
1045 if (info->NextEntryOffset == 0) {
1046 /* No more stream information. */
1049 /* Advance to next stream information. */
1050 info = (FILE_STREAM_INFORMATION *)
1051 ((u8 *)info + info->NextEntryOffset);
1057 /* The volume does not support named streams. Only capture the unnamed
1059 if (inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY |
1060 FILE_ATTRIBUTE_REPARSE_POINT))
1067 wchar_t stream_name[] = L"::$DATA";
1068 ret = winnt_scan_data_stream(path, path_nchars, stream_name, 7,
1069 file_size, inode, unhashed_blobs);
1072 /* Free buffer if allocated on heap. */
1073 if (unlikely(buf != _buf))
1078 static noinline_for_stack u64
1079 get_sort_key(HANDLE h)
1081 STARTING_VCN_INPUT_BUFFER in = { .StartingVcn.QuadPart = 0 };
1082 RETRIEVAL_POINTERS_BUFFER out;
1083 DWORD bytesReturned;
1085 if (!DeviceIoControl(h, FSCTL_GET_RETRIEVAL_POINTERS,
1088 &bytesReturned, NULL))
1091 if (out.ExtentCount < 1)
1094 return out.Extents[0].Lcn.QuadPart;
1098 set_sort_key(struct wim_inode *inode, u64 sort_key)
1100 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1101 struct wim_inode_stream *strm = &inode->i_streams[i];
1102 struct blob_descriptor *blob = stream_blob_resolved(strm);
1103 if (blob && (blob->blob_location == BLOB_IN_WINNT_FILE_ON_DISK ||
1104 blob->blob_location == BLOB_WIN32_ENCRYPTED))
1105 blob->sort_key = sort_key;
1109 static noinline_for_stack u32
1110 get_volume_information(HANDLE h, const wchar_t *full_path,
1111 struct capture_params *params)
1113 FILE_FS_ATTRIBUTE_INFORMATION attr_info;
1114 FILE_FS_VOLUME_INFORMATION vol_info;
1115 IO_STATUS_BLOCK iosb;
1119 /* Get volume flags */
1120 status = (*func_NtQueryVolumeInformationFile)(h, &iosb,
1123 FileFsAttributeInformation);
1124 if (likely((NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW) &&
1125 (iosb.Information >=
1126 offsetof(FILE_FS_ATTRIBUTE_INFORMATION,
1127 FileSystemAttributes) +
1128 sizeof(attr_info.FileSystemAttributes))))
1130 vol_flags = attr_info.FileSystemAttributes;
1132 winnt_warning(status, L"\"%ls\": Can't get volume attributes",
1133 printable_path(full_path));
1137 /* Get volume ID. */
1138 status = (*func_NtQueryVolumeInformationFile)(h, &iosb,
1141 FileFsVolumeInformation);
1142 if (likely((NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW) &&
1143 (iosb.Information >=
1144 offsetof(FILE_FS_VOLUME_INFORMATION,
1145 VolumeSerialNumber) +
1146 sizeof(vol_info.VolumeSerialNumber))))
1148 params->capture_root_dev = vol_info.VolumeSerialNumber;
1150 winnt_warning(status, L"\"%ls\": Can't get volume ID",
1151 printable_path(full_path));
1152 params->capture_root_dev = 0;
1161 u64 last_write_time;
1162 u64 last_access_time;
1167 static noinline_for_stack NTSTATUS
1168 get_file_info(HANDLE h, struct file_info *info)
1170 IO_STATUS_BLOCK iosb;
1172 FILE_ALL_INFORMATION all_info;
1174 status = (*func_NtQueryInformationFile)(h, &iosb, &all_info,
1176 FileAllInformation);
1178 if (unlikely(!NT_SUCCESS(status) && status != STATUS_BUFFER_OVERFLOW))
1181 info->attributes = all_info.BasicInformation.FileAttributes;
1182 info->num_links = all_info.StandardInformation.NumberOfLinks;
1183 info->creation_time = all_info.BasicInformation.CreationTime.QuadPart;
1184 info->last_write_time = all_info.BasicInformation.LastWriteTime.QuadPart;
1185 info->last_access_time = all_info.BasicInformation.LastAccessTime.QuadPart;
1186 info->ino = all_info.InternalInformation.IndexNumber.QuadPart;
1187 info->end_of_file = all_info.StandardInformation.EndOfFile.QuadPart;
1188 return STATUS_SUCCESS;
1192 winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret,
1195 size_t full_path_nchars,
1196 const wchar_t *filename,
1197 size_t filename_nchars,
1198 struct capture_params *params,
1199 struct winnt_scan_stats *stats,
1202 struct wim_dentry *root = NULL;
1203 struct wim_inode *inode = NULL;
1207 struct file_info file_info;
1208 ACCESS_MASK requestedPerms;
1211 ret = try_exclude(full_path, params);
1212 if (unlikely(ret < 0)) /* Excluded? */
1214 if (unlikely(ret > 0)) /* Error? */
1217 /* Open the file. */
1218 requestedPerms = FILE_READ_DATA |
1219 FILE_READ_ATTRIBUTES |
1221 ACCESS_SYSTEM_SECURITY |
1224 status = winnt_openat(cur_dir,
1225 (cur_dir ? filename : full_path),
1226 (cur_dir ? filename_nchars : full_path_nchars),
1229 if (unlikely(!NT_SUCCESS(status))) {
1230 if (status == STATUS_DELETE_PENDING) {
1231 WARNING("\"%ls\": Deletion pending; skipping file",
1232 printable_path(full_path));
1236 if (status == STATUS_ACCESS_DENIED &&
1237 (requestedPerms & FILE_READ_DATA)) {
1238 /* This happens on encrypted files. */
1239 requestedPerms &= ~FILE_READ_DATA;
1243 winnt_error(status, L"\"%ls\": Can't open file",
1244 printable_path(full_path));
1245 if (status == STATUS_FVE_LOCKED_VOLUME)
1246 ret = WIMLIB_ERR_FVE_LOCKED_VOLUME;
1248 ret = WIMLIB_ERR_OPEN;
1252 /* Get information about the file. */
1253 status = get_file_info(h, &file_info);
1254 if (!NT_SUCCESS(status)) {
1255 winnt_error(status, L"\"%ls\": Can't get file information",
1256 printable_path(full_path));
1257 ret = WIMLIB_ERR_STAT;
1261 if (unlikely(!(requestedPerms & FILE_READ_DATA)) &&
1262 !(file_info.attributes & FILE_ATTRIBUTE_ENCRYPTED))
1264 ERROR("\"%ls\": Permission to read data was denied",
1265 printable_path(full_path));
1266 ret = WIMLIB_ERR_OPEN;
1270 if (unlikely(!cur_dir)) {
1271 /* Root of tree being captured; get volume information. */
1272 vol_flags = get_volume_information(h, full_path, params);
1273 params->capture_root_ino = file_info.ino;
1277 /* Create a WIM dentry with an associated inode, which may be shared.
1279 * However, we need to explicitly check for directories and files with
1280 * only 1 link and refuse to hard link them. This is because Windows
1281 * has a bug where it can return duplicate File IDs for files and
1282 * directories on the FAT filesystem.
1284 * Since we don't follow mount points on Windows, we don't need to query
1285 * the volume ID per-file. Just once, for the root, is enough. But we
1286 * can't simply pass 0, because then there could be inode collisions
1287 * among multiple calls to win32_build_dentry_tree() that are scanning
1288 * files on different volumes. */
1289 ret = inode_table_new_dentry(params->inode_table,
1292 params->capture_root_dev,
1293 (file_info.num_links <= 1),
1298 /* Get the short (DOS) name of the file. */
1299 status = winnt_get_short_name(h, root);
1301 /* If we can't read the short filename for any reason other than
1302 * out-of-memory, just ignore the error and assume the file has no short
1303 * name. This shouldn't be an issue, since the short names are
1304 * essentially obsolete anyway. */
1305 if (unlikely(status == STATUS_NO_MEMORY)) {
1306 ret = WIMLIB_ERR_NOMEM;
1310 inode = root->d_inode;
1312 if (inode->i_nlink > 1) {
1313 /* Shared inode (hard link); skip reading per-inode information.
1318 inode->i_attributes = file_info.attributes;
1319 inode->i_creation_time = file_info.creation_time;
1320 inode->i_last_write_time = file_info.last_write_time;
1321 inode->i_last_access_time = file_info.last_access_time;
1323 /* Get the file's security descriptor, unless we are capturing in
1324 * NO_ACLS mode or the volume does not support security descriptors. */
1325 if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)
1326 && (vol_flags & FILE_PERSISTENT_ACLS))
1328 status = winnt_get_security_descriptor(h, inode,
1329 params->sd_set, stats,
1331 if (!NT_SUCCESS(status)) {
1333 L"\"%ls\": Can't read security descriptor",
1334 printable_path(full_path));
1335 ret = WIMLIB_ERR_STAT;
1340 /* If this is a reparse point, load the reparse data. */
1341 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)) {
1342 ret = winnt_load_reparse_data(h, inode, full_path, params);
1347 sort_key = get_sort_key(h);
1349 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
1350 /* Load information about the raw encrypted data. This is
1351 * needed for any directory or non-directory that has
1352 * FILE_ATTRIBUTE_ENCRYPTED set.
1354 * Note: since OpenEncryptedFileRaw() fails with
1355 * ERROR_SHARING_VIOLATION if there are any open handles to the
1356 * file, we have to close the file and re-open it later if
1360 ret = winnt_scan_efsrpc_raw_data(inode, full_path,
1361 params->unhashed_blobs);
1366 * Load information about data streams (unnamed and named).
1368 * Skip this step for encrypted files, since the data from
1369 * ReadEncryptedFileRaw() already contains all data streams (and
1370 * they do in fact all get restored by WriteEncryptedFileRaw().)
1372 * Note: WIMGAPI (as of Windows 8.1) gets wrong and stores both
1373 * the EFSRPC data and the named data stream(s)...!
1375 ret = winnt_scan_data_streams(h,
1379 params->unhashed_blobs,
1380 file_info.end_of_file,
1386 set_sort_key(inode, sort_key);
1388 if (inode_is_directory(inode)) {
1390 /* Directory: recurse to children. */
1393 /* Re-open handle that was closed to read raw encrypted
1395 status = winnt_openat(cur_dir,
1397 filename : full_path),
1399 filename_nchars : full_path_nchars),
1400 FILE_LIST_DIRECTORY | SYNCHRONIZE,
1402 if (!NT_SUCCESS(status)) {
1404 L"\"%ls\": Can't re-open file",
1405 printable_path(full_path));
1406 ret = WIMLIB_ERR_OPEN;
1410 ret = winnt_recurse_directory(h,
1422 params->progress.scan.cur_path = printable_path(full_path);
1424 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK, inode);
1426 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL);
1430 if (unlikely(ret)) {
1431 free_dentry_tree(root, params->blob_table);
1433 ret = report_capture_error(params, ret, full_path);
1440 winnt_do_scan_warnings(const wchar_t *path, const struct winnt_scan_stats *stats)
1442 if (likely(stats->num_get_sacl_priv_notheld == 0 &&
1443 stats->num_get_sd_access_denied == 0))
1446 WARNING("Scan of \"%ls\" complete, but with one or more warnings:", path);
1447 if (stats->num_get_sacl_priv_notheld != 0) {
1448 WARNING("- Could not capture SACL (System Access Control List)\n"
1449 " on %lu files or directories.",
1450 stats->num_get_sacl_priv_notheld);
1452 if (stats->num_get_sd_access_denied != 0) {
1453 WARNING("- Could not capture security descriptor at all\n"
1454 " on %lu files or directories.",
1455 stats->num_get_sd_access_denied);
1457 WARNING("To fully capture all security descriptors, run the program\n"
1458 " with Administrator rights.");
1461 #define WINDOWS_NT_MAX_PATH 32768
1463 /* Win32 version of capturing a directory tree. */
1465 win32_build_dentry_tree(struct wim_dentry **root_ret,
1466 const wchar_t *root_disk_path,
1467 struct capture_params *params)
1471 UNICODE_STRING ntpath;
1472 struct winnt_scan_stats stats;
1473 size_t ntpath_nchars;
1475 /* WARNING: There is no check for overflow later when this buffer is
1476 * being used! But it's as long as the maximum path length understood
1477 * by Windows NT (which is NOT the same as MAX_PATH). */
1478 path = MALLOC((WINDOWS_NT_MAX_PATH + 1) * sizeof(wchar_t));
1480 return WIMLIB_ERR_NOMEM;
1482 ret = win32_path_to_nt_path(root_disk_path, &ntpath);
1486 if (ntpath.Length < 4 * sizeof(wchar_t) ||
1487 ntpath.Length > WINDOWS_NT_MAX_PATH * sizeof(wchar_t) ||
1488 wmemcmp(ntpath.Buffer, L"\\??\\", 4))
1490 ERROR("\"%ls\": unrecognized path format", root_disk_path);
1491 ret = WIMLIB_ERR_INVALID_PARAM;
1493 ntpath_nchars = ntpath.Length / sizeof(wchar_t);
1494 wmemcpy(path, ntpath.Buffer, ntpath_nchars);
1495 path[ntpath_nchars] = L'\0';
1497 params->capture_root_nchars = ntpath_nchars;
1498 if (path[ntpath_nchars - 1] == L'\\')
1499 params->capture_root_nchars--;
1502 HeapFree(GetProcessHeap(), 0, ntpath.Buffer);
1506 memset(&stats, 0, sizeof(stats));
1508 ret = winnt_build_dentry_tree_recursive(root_ret, NULL,
1509 path, ntpath_nchars,
1510 L"", 0, params, &stats, 0);
1514 winnt_do_scan_warnings(root_disk_path, &stats);
1518 #endif /* __WIN32__ */