1 /*
2 * win32_capture.c - Windows-specific code for capturing files into a WIM image.
3 *
4 * This now uses the native Windows NT API a lot and not just Win32.
5 */
6
7 /*
8 * Copyright (C) 2013-2018 Eric Biggers
9 *
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
13 * later version.
14 *
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
18 * details.
19 *
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/.
22 */
23
24 #ifdef __WIN32__
25
26 #ifdef HAVE_CONFIG_H
27 # include "config.h"
28 #endif
29
30 #include "wimlib/win32_common.h"
31
32 #include "wimlib/assert.h"
33 #include "wimlib/blob_table.h"
34 #include "wimlib/dentry.h"
35 #include "wimlib/encoding.h"
36 #include "wimlib/endianness.h"
37 #include "wimlib/error.h"
38 #include "wimlib/object_id.h"
39 #include "wimlib/paths.h"
40 #include "wimlib/reparse.h"
41 #include "wimlib/scan.h"
42 #include "wimlib/win32_vss.h"
43 #include "wimlib/wof.h"
44 #include "wimlib/xattr.h"
45
46 struct winnt_scan_ctx {
47 struct scan_params *params;
48 bool is_ntfs;
49 u32 vol_flags;
50 unsigned long num_get_sd_access_denied;
51 unsigned long num_get_sacl_priv_notheld;
52
53 /* True if WOF is definitely not attached to the volume being scanned;
54 * false if it may be */
55 bool wof_not_attached;
56
57 /* A reference to the VSS snapshot being used, or NULL if none */
58 struct vss_snapshot *snapshot;
59 };
60
61 static inline const wchar_t *
printable_path(const struct winnt_scan_ctx * ctx)62 printable_path(const struct winnt_scan_ctx *ctx)
63 {
64 /* Skip over \\?\ or \??\ */
65 return ctx->params->cur_path + 4;
66 }
67
68 /* Description of where data is located on a Windows filesystem */
69 struct windows_file {
70
71 /* Is the data the raw encrypted data of an EFS-encrypted file? */
72 u64 is_encrypted : 1;
73
74 /* Is this file "open by file ID" rather than the regular "open by
75 * path"? "Open by file ID" uses resources more efficiently. */
76 u64 is_file_id : 1;
77
78 /* The file's LCN (logical cluster number) for sorting, or 0 if unknown.
79 */
80 u64 sort_key : 62;
81
82 /* Length of the path in bytes, excluding the null terminator if
83 * present. */
84 size_t path_nbytes;
85
86 /* A reference to the VSS snapshot containing the file, or NULL if none.
87 */
88 struct vss_snapshot *snapshot;
89
90 /* The path to the file. If 'is_encrypted=0' this is an NT namespace
91 * path; if 'is_encrypted=1' this is a Win32 namespace path. If
92 * 'is_file_id=0', then the path is null-terminated. If 'is_file_id=1'
93 * (only allowed with 'is_encrypted=0') the path ends with a binary file
94 * ID and may not be null-terminated. */
95 wchar_t path[0];
96 };
97
98 /* Allocate a structure to describe the location of a data stream by path. */
99 static struct windows_file *
alloc_windows_file(const wchar_t * path,size_t path_nchars,const wchar_t * stream_name,size_t stream_name_nchars,struct vss_snapshot * snapshot,bool is_encrypted)100 alloc_windows_file(const wchar_t *path, size_t path_nchars,
101 const wchar_t *stream_name, size_t stream_name_nchars,
102 struct vss_snapshot *snapshot, bool is_encrypted)
103 {
104 size_t full_path_nbytes;
105 struct windows_file *file;
106 wchar_t *p;
107
108 full_path_nbytes = path_nchars * sizeof(wchar_t);
109 if (stream_name_nchars)
110 full_path_nbytes += (1 + stream_name_nchars) * sizeof(wchar_t);
111
112 file = MALLOC(sizeof(struct windows_file) + full_path_nbytes +
113 sizeof(wchar_t));
114 if (!file)
115 return NULL;
116
117 file->is_encrypted = is_encrypted;
118 file->is_file_id = 0;
119 file->sort_key = 0;
120 file->path_nbytes = full_path_nbytes;
121 file->snapshot = vss_get_snapshot(snapshot);
122 p = wmempcpy(file->path, path, path_nchars);
123 if (stream_name_nchars) {
124 /* Named data stream */
125 *p++ = L':';
126 p = wmempcpy(p, stream_name, stream_name_nchars);
127 }
128 *p = L'\0';
129 return file;
130 }
131
132 /* Allocate a structure to describe the location of a file by ID. */
133 static struct windows_file *
alloc_windows_file_for_file_id(u64 file_id,const wchar_t * root_path,size_t root_path_nchars,struct vss_snapshot * snapshot)134 alloc_windows_file_for_file_id(u64 file_id, const wchar_t *root_path,
135 size_t root_path_nchars,
136 struct vss_snapshot *snapshot)
137 {
138 size_t full_path_nbytes;
139 struct windows_file *file;
140 wchar_t *p;
141
142 full_path_nbytes = (root_path_nchars * sizeof(wchar_t)) +
143 sizeof(file_id);
144 file = MALLOC(sizeof(struct windows_file) + full_path_nbytes +
145 sizeof(wchar_t));
146 if (!file)
147 return NULL;
148
149 file->is_encrypted = 0;
150 file->is_file_id = 1;
151 file->sort_key = 0;
152 file->path_nbytes = full_path_nbytes;
153 file->snapshot = vss_get_snapshot(snapshot);
154 p = wmempcpy(file->path, root_path, root_path_nchars);
155 p = mempcpy(p, &file_id, sizeof(file_id));
156 *p = L'\0';
157 return file;
158 }
159
160 /* Add a stream, located on a Windows filesystem, to the specified WIM inode. */
161 static int
add_stream(struct wim_inode * inode,struct windows_file * windows_file,u64 stream_size,int stream_type,const utf16lechar * stream_name,struct list_head * unhashed_blobs)162 add_stream(struct wim_inode *inode, struct windows_file *windows_file,
163 u64 stream_size, int stream_type, const utf16lechar *stream_name,
164 struct list_head *unhashed_blobs)
165 {
166 struct blob_descriptor *blob = NULL;
167 struct wim_inode_stream *strm;
168 int ret;
169
170 if (!windows_file)
171 goto err_nomem;
172
173 /* If the stream is nonempty, create a blob descriptor for it. */
174 if (stream_size) {
175 blob = new_blob_descriptor();
176 if (!blob)
177 goto err_nomem;
178 blob->windows_file = windows_file;
179 blob->blob_location = BLOB_IN_WINDOWS_FILE;
180 blob->file_inode = inode;
181 blob->size = stream_size;
182 windows_file = NULL;
183 }
184
185 strm = inode_add_stream(inode, stream_type, stream_name, blob);
186 if (!strm)
187 goto err_nomem;
188
189 prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
190 ret = 0;
191 out:
192 if (windows_file)
193 free_windows_file(windows_file);
194 return ret;
195
196 err_nomem:
197 free_blob_descriptor(blob);
198 ret = WIMLIB_ERR_NOMEM;
199 goto out;
200 }
201
202 struct windows_file *
clone_windows_file(const struct windows_file * file)203 clone_windows_file(const struct windows_file *file)
204 {
205 struct windows_file *new;
206
207 new = memdup(file, sizeof(*file) + file->path_nbytes + sizeof(wchar_t));
208 if (new)
209 vss_get_snapshot(new->snapshot);
210 return new;
211 }
212
213 void
free_windows_file(struct windows_file * file)214 free_windows_file(struct windows_file *file)
215 {
216 vss_put_snapshot(file->snapshot);
217 FREE(file);
218 }
219
220 int
cmp_windows_files(const struct windows_file * file1,const struct windows_file * file2)221 cmp_windows_files(const struct windows_file *file1,
222 const struct windows_file *file2)
223 {
224 /* Compare by starting LCN (logical cluster number) */
225 int v = cmp_u64(file1->sort_key, file2->sort_key);
226 if (v)
227 return v;
228
229 /* Fall back to comparing files by path (arbitrary heuristic). */
230 v = memcmp(file1->path, file2->path,
231 min(file1->path_nbytes, file2->path_nbytes));
232 if (v)
233 return v;
234
235 return cmp_u32(file1->path_nbytes, file2->path_nbytes);
236 }
237
238 const wchar_t *
get_windows_file_path(const struct windows_file * file)239 get_windows_file_path(const struct windows_file *file)
240 {
241 return file->path;
242 }
243
244 /*
245 * Open the file named by the NT namespace path @path of length @path_nchars
246 * characters. If @cur_dir is not NULL then the path is given relative to
247 * @cur_dir; otherwise the path is absolute. @perms is the access mask of
248 * permissions to request on the handle. SYNCHRONIZE permision is always added.
249 */
250 static NTSTATUS
winnt_openat(HANDLE cur_dir,const wchar_t * path,size_t path_nchars,ACCESS_MASK perms,HANDLE * h_ret)251 winnt_openat(HANDLE cur_dir, const wchar_t *path, size_t path_nchars,
252 ACCESS_MASK perms, HANDLE *h_ret)
253 {
254 UNICODE_STRING name = {
255 .Length = path_nchars * sizeof(wchar_t),
256 .MaximumLength = path_nchars * sizeof(wchar_t),
257 .Buffer = (wchar_t *)path,
258 };
259 OBJECT_ATTRIBUTES attr = {
260 .Length = sizeof(attr),
261 .RootDirectory = cur_dir,
262 .ObjectName = &name,
263 };
264 IO_STATUS_BLOCK iosb;
265 NTSTATUS status;
266 ULONG options = FILE_OPEN_REPARSE_POINT | FILE_OPEN_FOR_BACKUP_INTENT;
267
268 perms |= SYNCHRONIZE;
269 if (perms & (FILE_READ_DATA | FILE_LIST_DIRECTORY)) {
270 options |= FILE_SYNCHRONOUS_IO_NONALERT;
271 options |= FILE_SEQUENTIAL_ONLY;
272 }
273 retry:
274 status = NtOpenFile(h_ret, perms, &attr, &iosb,
275 FILE_SHARE_VALID_FLAGS, options);
276 if (!NT_SUCCESS(status)) {
277 /* Try requesting fewer permissions */
278 if (status == STATUS_ACCESS_DENIED ||
279 status == STATUS_PRIVILEGE_NOT_HELD) {
280 if (perms & ACCESS_SYSTEM_SECURITY) {
281 perms &= ~ACCESS_SYSTEM_SECURITY;
282 goto retry;
283 }
284 if (perms & READ_CONTROL) {
285 perms &= ~READ_CONTROL;
286 goto retry;
287 }
288 }
289 }
290 return status;
291 }
292
293 static NTSTATUS
winnt_open(const wchar_t * path,size_t path_nchars,ACCESS_MASK perms,HANDLE * h_ret)294 winnt_open(const wchar_t *path, size_t path_nchars, ACCESS_MASK perms,
295 HANDLE *h_ret)
296 {
297 return winnt_openat(NULL, path, path_nchars, perms, h_ret);
298 }
299
300 static const wchar_t *
windows_file_to_string(const struct windows_file * file,u8 * buf,size_t bufsize)301 windows_file_to_string(const struct windows_file *file, u8 *buf, size_t bufsize)
302 {
303 if (file->is_file_id) {
304 u64 file_id;
305 memcpy(&file_id,
306 (u8 *)file->path + file->path_nbytes - sizeof(file_id),
307 sizeof(file_id));
308 swprintf((wchar_t *)buf, L"NTFS inode 0x%016"PRIx64, file_id);
309 } else if (file->path_nbytes + 3 * sizeof(wchar_t) <= bufsize) {
310 swprintf((wchar_t *)buf, L"\"%ls\"", file->path);
311 } else {
312 return L"(name too long)";
313 }
314 return (wchar_t *)buf;
315 }
316
317 static int
read_winnt_stream_prefix(const struct windows_file * file,u64 size,const struct consume_chunk_callback * cb)318 read_winnt_stream_prefix(const struct windows_file *file,
319 u64 size, const struct consume_chunk_callback *cb)
320 {
321 IO_STATUS_BLOCK iosb;
322 UNICODE_STRING name = {
323 .Buffer = (wchar_t *)file->path,
324 .Length = file->path_nbytes,
325 .MaximumLength = file->path_nbytes,
326 };
327 OBJECT_ATTRIBUTES attr = {
328 .Length = sizeof(attr),
329 .ObjectName = &name,
330 };
331 HANDLE h;
332 NTSTATUS status;
333 u8 buf[BUFFER_SIZE] _aligned_attribute(8);
334 u64 bytes_remaining;
335 int ret;
336
337 status = NtOpenFile(&h, FILE_READ_DATA | SYNCHRONIZE,
338 &attr, &iosb,
339 FILE_SHARE_VALID_FLAGS,
340 FILE_OPEN_REPARSE_POINT |
341 FILE_OPEN_FOR_BACKUP_INTENT |
342 FILE_SYNCHRONOUS_IO_NONALERT |
343 FILE_SEQUENTIAL_ONLY |
344 (file->is_file_id ? FILE_OPEN_BY_FILE_ID : 0));
345 if (unlikely(!NT_SUCCESS(status))) {
346 if (status == STATUS_SHARING_VIOLATION) {
347 ERROR("Can't open %ls for reading:\n"
348 " File is in use by another process! "
349 "Consider using snapshot (VSS) mode.",
350 windows_file_to_string(file, buf, sizeof(buf)));
351 } else {
352 winnt_error(status, L"Can't open %ls for reading",
353 windows_file_to_string(file, buf, sizeof(buf)));
354 }
355 return WIMLIB_ERR_OPEN;
356 }
357
358 ret = 0;
359 bytes_remaining = size;
360 while (bytes_remaining) {
361 IO_STATUS_BLOCK iosb;
362 ULONG count;
363 ULONG bytes_read;
364 const unsigned max_tries = 5;
365 unsigned tries_remaining = max_tries;
366
367 count = min(sizeof(buf), bytes_remaining);
368
369 retry_read:
370 status = NtReadFile(h, NULL, NULL, NULL,
371 &iosb, buf, count, NULL, NULL);
372 if (unlikely(!NT_SUCCESS(status))) {
373 if (status == STATUS_END_OF_FILE) {
374 ERROR("%ls: File was concurrently truncated",
375 windows_file_to_string(file, buf, sizeof(buf)));
376 ret = WIMLIB_ERR_CONCURRENT_MODIFICATION_DETECTED;
377 } else {
378 winnt_warning(status, L"Error reading data from %ls",
379 windows_file_to_string(file, buf, sizeof(buf)));
380
381 /* Currently these retries are purely a guess;
382 * there is no reproducible problem that they solve. */
383 if (--tries_remaining) {
384 int delay = 100;
385 if (status == STATUS_INSUFFICIENT_RESOURCES ||
386 status == STATUS_NO_MEMORY) {
387 delay *= 25;
388 }
389 WARNING("Retrying after %dms...", delay);
390 Sleep(delay);
391 goto retry_read;
392 }
393 ERROR("Too many retries; returning failure");
394 ret = WIMLIB_ERR_READ;
395 }
396 break;
397 } else if (unlikely(tries_remaining != max_tries)) {
398 WARNING("A read request had to be retried multiple times "
399 "before it succeeded!");
400 }
401
402 bytes_read = iosb.Information;
403
404 bytes_remaining -= bytes_read;
405 ret = consume_chunk(cb, buf, bytes_read);
406 if (ret)
407 break;
408 }
409 NtClose(h);
410 return ret;
411 }
412
413 struct win32_encrypted_read_ctx {
414 const struct consume_chunk_callback *cb;
415 int wimlib_err_code;
416 u64 bytes_remaining;
417 };
418
419 static DWORD WINAPI
win32_encrypted_export_cb(unsigned char * data,void * _ctx,unsigned long len)420 win32_encrypted_export_cb(unsigned char *data, void *_ctx, unsigned long len)
421 {
422 struct win32_encrypted_read_ctx *ctx = _ctx;
423 int ret;
424 size_t bytes_to_consume = min(len, ctx->bytes_remaining);
425
426 if (bytes_to_consume == 0)
427 return ERROR_SUCCESS;
428
429 ret = consume_chunk(ctx->cb, data, bytes_to_consume);
430 if (ret) {
431 ctx->wimlib_err_code = ret;
432 /* It doesn't matter what error code is returned here, as long
433 * as it isn't ERROR_SUCCESS. */
434 return ERROR_READ_FAULT;
435 }
436 ctx->bytes_remaining -= bytes_to_consume;
437 return ERROR_SUCCESS;
438 }
439
440 static int
read_win32_encrypted_file_prefix(const wchar_t * path,bool is_dir,u64 size,const struct consume_chunk_callback * cb)441 read_win32_encrypted_file_prefix(const wchar_t *path, bool is_dir, u64 size,
442 const struct consume_chunk_callback *cb)
443 {
444 struct win32_encrypted_read_ctx export_ctx;
445 DWORD err;
446 void *file_ctx;
447 int ret;
448 DWORD flags = 0;
449
450 if (is_dir)
451 flags |= CREATE_FOR_DIR;
452
453 export_ctx.cb = cb;
454 export_ctx.wimlib_err_code = 0;
455 export_ctx.bytes_remaining = size;
456
457 err = OpenEncryptedFileRaw(path, flags, &file_ctx);
458 if (err != ERROR_SUCCESS) {
459 win32_error(err,
460 L"Failed to open encrypted file \"%ls\" for raw read",
461 path);
462 return WIMLIB_ERR_OPEN;
463 }
464 err = ReadEncryptedFileRaw(win32_encrypted_export_cb,
465 &export_ctx, file_ctx);
466 if (err != ERROR_SUCCESS) {
467 ret = export_ctx.wimlib_err_code;
468 if (ret == 0) {
469 win32_error(err,
470 L"Failed to read encrypted file \"%ls\"",
471 path);
472 ret = WIMLIB_ERR_READ;
473 }
474 } else if (export_ctx.bytes_remaining != 0) {
475 ERROR("Only could read %"PRIu64" of %"PRIu64" bytes from "
476 "encrypted file \"%ls\"",
477 size - export_ctx.bytes_remaining, size,
478 path);
479 ret = WIMLIB_ERR_READ;
480 } else {
481 ret = 0;
482 }
483 CloseEncryptedFileRaw(file_ctx);
484 return ret;
485 }
486
487 /* Read the first @size bytes from the file, or named data stream of a file,
488 * described by @blob. */
489 int
read_windows_file_prefix(const struct blob_descriptor * blob,u64 size,const struct consume_chunk_callback * cb)490 read_windows_file_prefix(const struct blob_descriptor *blob, u64 size,
491 const struct consume_chunk_callback *cb)
492 {
493 const struct windows_file *file = blob->windows_file;
494
495 if (unlikely(file->is_encrypted)) {
496 bool is_dir = (blob->file_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY);
497 return read_win32_encrypted_file_prefix(file->path, is_dir, size, cb);
498 }
499
500 return read_winnt_stream_prefix(file, size, cb);
501 }
502
503 /*
504 * Load the short name of a file into a WIM dentry.
505 */
506 static noinline_for_stack NTSTATUS
winnt_get_short_name(HANDLE h,struct wim_dentry * dentry)507 winnt_get_short_name(HANDLE h, struct wim_dentry *dentry)
508 {
509 /* It's not any harder to just make the NtQueryInformationFile() system
510 * call ourselves, and it saves a dumb call to FindFirstFile() which of
511 * course has to create its own handle. */
512 NTSTATUS status;
513 IO_STATUS_BLOCK iosb;
514 u8 buf[128] _aligned_attribute(8);
515 const FILE_NAME_INFORMATION *info;
516
517 status = NtQueryInformationFile(h, &iosb, buf, sizeof(buf),
518 FileAlternateNameInformation);
519 info = (const FILE_NAME_INFORMATION *)buf;
520 if (NT_SUCCESS(status) && info->FileNameLength != 0) {
521 dentry->d_short_name = utf16le_dupz(info->FileName,
522 info->FileNameLength);
523 if (!dentry->d_short_name)
524 return STATUS_NO_MEMORY;
525 dentry->d_short_name_nbytes = info->FileNameLength;
526 }
527 return status;
528 }
529
530 /*
531 * Load the security descriptor of a file into the corresponding inode and the
532 * WIM image's security descriptor set.
533 */
534 static noinline_for_stack int
winnt_load_security_descriptor(HANDLE h,struct wim_inode * inode,struct winnt_scan_ctx * ctx)535 winnt_load_security_descriptor(HANDLE h, struct wim_inode *inode,
536 struct winnt_scan_ctx *ctx)
537 {
538 SECURITY_INFORMATION requestedInformation;
539 u8 _buf[4096] _aligned_attribute(8);
540 u8 *buf;
541 ULONG bufsize;
542 ULONG len_needed;
543 NTSTATUS status;
544
545 /*
546 * LABEL_SECURITY_INFORMATION is needed on Windows Vista and 7 because
547 * Microsoft decided to add mandatory integrity labels to the SACL but
548 * not have them returned by SACL_SECURITY_INFORMATION.
549 *
550 * BACKUP_SECURITY_INFORMATION is needed on Windows 8 because Microsoft
551 * decided to add even more stuff to the SACL and still not have it
552 * returned by SACL_SECURITY_INFORMATION; but they did remember that
553 * backup applications exist and simply want to read the stupid thing
554 * once and for all, so they added a flag to read the entire security
555 * descriptor.
556 *
557 * Older versions of Windows tolerate these new flags being passed in.
558 */
559 requestedInformation = OWNER_SECURITY_INFORMATION |
560 GROUP_SECURITY_INFORMATION |
561 DACL_SECURITY_INFORMATION |
562 SACL_SECURITY_INFORMATION |
563 LABEL_SECURITY_INFORMATION |
564 BACKUP_SECURITY_INFORMATION;
565
566 buf = _buf;
567 bufsize = sizeof(_buf);
568
569 /*
570 * We need the file's security descriptor in
571 * SECURITY_DESCRIPTOR_RELATIVE format, and we currently have a handle
572 * opened with as many relevant permissions as possible. At this point,
573 * on Windows there are a number of options for reading a file's
574 * security descriptor:
575 *
576 * GetFileSecurity(): This takes in a path and returns the
577 * SECURITY_DESCRIPTOR_RELATIVE. Problem: this uses an internal handle,
578 * not ours, and the handle created internally doesn't specify
579 * FILE_FLAG_BACKUP_SEMANTICS. Therefore there can be access denied
580 * errors on some files and directories, even when running as the
581 * Administrator.
582 *
583 * GetSecurityInfo(): This takes in a handle and returns the security
584 * descriptor split into a bunch of different parts. This should work,
585 * but it's dumb because we have to put the security descriptor back
586 * together again.
587 *
588 * BackupRead(): This can read the security descriptor, but this is a
589 * difficult-to-use API, probably only works as the Administrator, and
590 * the format of the returned data is not well documented.
591 *
592 * NtQuerySecurityObject(): This is exactly what we need, as it takes
593 * in a handle and returns the security descriptor in
594 * SECURITY_DESCRIPTOR_RELATIVE format. Only problem is that it's a
595 * ntdll function and therefore not officially part of the Win32 API.
596 * Oh well.
597 */
598 while (!NT_SUCCESS(status = NtQuerySecurityObject(h,
599 requestedInformation,
600 (PSECURITY_DESCRIPTOR)buf,
601 bufsize,
602 &len_needed)))
603 {
604 switch (status) {
605 case STATUS_BUFFER_TOO_SMALL:
606 wimlib_assert(buf == _buf);
607 buf = MALLOC(len_needed);
608 if (!buf) {
609 status = STATUS_NO_MEMORY;
610 goto out;
611 }
612 bufsize = len_needed;
613 break;
614 case STATUS_PRIVILEGE_NOT_HELD:
615 case STATUS_ACCESS_DENIED:
616 if (ctx->params->add_flags & WIMLIB_ADD_FLAG_STRICT_ACLS) {
617 default:
618 /* Permission denied in STRICT_ACLS mode, or
619 * unknown error. */
620 goto out;
621 }
622 if (requestedInformation & SACL_SECURITY_INFORMATION) {
623 /* Try again without the SACL. */
624 ctx->num_get_sacl_priv_notheld++;
625 requestedInformation &= ~(SACL_SECURITY_INFORMATION |
626 LABEL_SECURITY_INFORMATION |
627 BACKUP_SECURITY_INFORMATION);
628 break;
629 }
630 /* Fake success (useful when capturing as
631 * non-Administrator). */
632 ctx->num_get_sd_access_denied++;
633 status = STATUS_SUCCESS;
634 goto out;
635 }
636 }
637
638 /* We can get a length of 0 with Samba. Assume that means "no security
639 * descriptor". */
640 if (len_needed == 0)
641 goto out;
642
643 /* Add the security descriptor to the WIM image, and save its ID in
644 * the file's inode. */
645 inode->i_security_id = sd_set_add_sd(ctx->params->sd_set, buf, len_needed);
646 if (unlikely(inode->i_security_id < 0))
647 status = STATUS_NO_MEMORY;
648 out:
649 if (unlikely(buf != _buf))
650 FREE(buf);
651 if (!NT_SUCCESS(status)) {
652 winnt_error(status, L"\"%ls\": Can't read security descriptor",
653 printable_path(ctx));
654 return WIMLIB_ERR_STAT;
655 }
656 return 0;
657 }
658
659 /* Load a file's object ID into the corresponding WIM inode. */
660 static noinline_for_stack int
winnt_load_object_id(HANDLE h,struct wim_inode * inode,struct winnt_scan_ctx * ctx)661 winnt_load_object_id(HANDLE h, struct wim_inode *inode,
662 struct winnt_scan_ctx *ctx)
663 {
664 FILE_OBJECTID_BUFFER buffer;
665 NTSTATUS status;
666 u32 len;
667
668 if (!(ctx->vol_flags & FILE_SUPPORTS_OBJECT_IDS))
669 return 0;
670
671 status = winnt_fsctl(h, FSCTL_GET_OBJECT_ID, NULL, 0,
672 &buffer, sizeof(buffer), &len);
673
674 if (status == STATUS_OBJECTID_NOT_FOUND) /* No object ID */
675 return 0;
676
677 if (status == STATUS_INVALID_DEVICE_REQUEST ||
678 status == STATUS_NOT_SUPPORTED /* Samba volume, WinXP */) {
679 /* The filesystem claimed to support object IDs, but we can't
680 * actually read them. This happens with Samba. */
681 ctx->vol_flags &= ~FILE_SUPPORTS_OBJECT_IDS;
682 return 0;
683 }
684
685 if (!NT_SUCCESS(status)) {
686 winnt_error(status, L"\"%ls\": Can't read object ID",
687 printable_path(ctx));
688 return WIMLIB_ERR_STAT;
689 }
690
691 if (len == 0) /* No object ID (for directories) */
692 return 0;
693
694 if (!inode_set_object_id(inode, &buffer, len))
695 return WIMLIB_ERR_NOMEM;
696
697 return 0;
698 }
699
700 /* Load a file's extended attributes into the corresponding WIM inode. */
701 static noinline_for_stack int
winnt_load_xattrs(HANDLE h,struct wim_inode * inode,struct winnt_scan_ctx * ctx,u32 ea_size)702 winnt_load_xattrs(HANDLE h, struct wim_inode *inode,
703 struct winnt_scan_ctx *ctx, u32 ea_size)
704 {
705 IO_STATUS_BLOCK iosb;
706 NTSTATUS status;
707 u8 _buf[1024] _aligned_attribute(4);
708 u8 *buf = _buf;
709 const FILE_FULL_EA_INFORMATION *ea;
710 struct wim_xattr_entry *entry;
711 int ret;
712
713
714 /*
715 * EaSize from FILE_EA_INFORMATION is apparently supposed to give the
716 * size of the buffer required for NtQueryEaFile(), but it doesn't
717 * actually work correctly; it can be off by about 4 bytes per xattr.
718 *
719 * So just start out by doubling the advertised size, and also handle
720 * STATUS_BUFFER_OVERFLOW just in case.
721 */
722 retry:
723 if (unlikely(ea_size * 2 < ea_size))
724 ea_size = UINT32_MAX;
725 else
726 ea_size *= 2;
727 if (unlikely(ea_size > sizeof(_buf))) {
728 buf = MALLOC(ea_size);
729 if (!buf) {
730 if (ea_size >= (1 << 20)) {
731 WARNING("\"%ls\": EaSize was extremely large (%u)",
732 printable_path(ctx), ea_size);
733 }
734 return WIMLIB_ERR_NOMEM;
735 }
736 }
737
738 status = NtQueryEaFile(h, &iosb, buf, ea_size,
739 FALSE, NULL, 0, NULL, TRUE);
740
741 if (unlikely(!NT_SUCCESS(status))) {
742 if (status == STATUS_BUFFER_OVERFLOW) {
743 if (buf != _buf) {
744 FREE(buf);
745 buf = NULL;
746 }
747 goto retry;
748 }
749 if (status == STATUS_NO_EAS_ON_FILE) {
750 /*
751 * FILE_EA_INFORMATION.EaSize was nonzero so this
752 * shouldn't happen, but just in case...
753 */
754 ret = 0;
755 goto out;
756 }
757 winnt_error(status, L"\"%ls\": Can't read extended attributes",
758 printable_path(ctx));
759 ret = WIMLIB_ERR_STAT;
760 goto out;
761 }
762
763 ea = (const FILE_FULL_EA_INFORMATION *)buf;
764 entry = (struct wim_xattr_entry *)buf;
765 for (;;) {
766 /*
767 * wim_xattr_entry is not larger than FILE_FULL_EA_INFORMATION,
768 * so we can reuse the same buffer by overwriting the
769 * FILE_FULL_EA_INFORMATION with the wim_xattr_entry in-place.
770 */
771 FILE_FULL_EA_INFORMATION _ea;
772
773 STATIC_ASSERT(offsetof(struct wim_xattr_entry, name) <=
774 offsetof(FILE_FULL_EA_INFORMATION, EaName));
775 wimlib_assert((u8 *)entry <= (const u8 *)ea);
776
777 memcpy(&_ea, ea, sizeof(_ea));
778
779 entry->value_len = cpu_to_le16(_ea.EaValueLength);
780 entry->name_len = _ea.EaNameLength;
781 entry->flags = _ea.Flags;
782 memmove(entry->name, ea->EaName, _ea.EaNameLength);
783 entry->name[_ea.EaNameLength] = '\0';
784 memmove(&entry->name[_ea.EaNameLength + 1],
785 &ea->EaName[_ea.EaNameLength + 1], _ea.EaValueLength);
786 entry = (struct wim_xattr_entry *)
787 &entry->name[_ea.EaNameLength + 1 + _ea.EaValueLength];
788 if (_ea.NextEntryOffset == 0)
789 break;
790 ea = (const FILE_FULL_EA_INFORMATION *)
791 ((const u8 *)ea + _ea.NextEntryOffset);
792 }
793 wimlib_assert((u8 *)entry - buf <= ea_size);
794
795 ret = WIMLIB_ERR_NOMEM;
796 if (!inode_set_xattrs(inode, buf, (u8 *)entry - buf))
797 goto out;
798 ret = 0;
799 out:
800 if (unlikely(buf != _buf))
801 FREE(buf);
802 return ret;
803 }
804
805 static int
806 winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret,
807 HANDLE cur_dir,
808 const wchar_t *relative_path,
809 size_t relative_path_nchars,
810 const wchar_t *filename,
811 struct winnt_scan_ctx *ctx);
812
813 static int
winnt_recurse_directory(HANDLE h,struct wim_dentry * parent,struct winnt_scan_ctx * ctx)814 winnt_recurse_directory(HANDLE h,
815 struct wim_dentry *parent,
816 struct winnt_scan_ctx *ctx)
817 {
818 void *buf;
819 const size_t bufsize = 8192;
820 IO_STATUS_BLOCK iosb;
821 NTSTATUS status;
822 int ret;
823
824 buf = MALLOC(bufsize);
825 if (!buf)
826 return WIMLIB_ERR_NOMEM;
827
828 /* Using NtQueryDirectoryFile() we can re-use the same open handle,
829 * which we opened with FILE_FLAG_BACKUP_SEMANTICS. */
830
831 while (NT_SUCCESS(status = NtQueryDirectoryFile(h, NULL, NULL, NULL,
832 &iosb, buf, bufsize,
833 FileNamesInformation,
834 FALSE, NULL, FALSE)))
835 {
836 const FILE_NAMES_INFORMATION *info = buf;
837 for (;;) {
838 if (!should_ignore_filename(info->FileName,
839 info->FileNameLength / 2))
840 {
841 struct wim_dentry *child;
842 size_t orig_path_nchars;
843 const wchar_t *filename;
844
845 ret = WIMLIB_ERR_NOMEM;
846 filename = pathbuf_append_name(ctx->params,
847 info->FileName,
848 info->FileNameLength / 2,
849 &orig_path_nchars);
850 if (!filename)
851 goto out_free_buf;
852
853 ret = winnt_build_dentry_tree_recursive(
854 &child,
855 h,
856 filename,
857 info->FileNameLength / 2,
858 filename,
859 ctx);
860
861 pathbuf_truncate(ctx->params, orig_path_nchars);
862
863 if (ret)
864 goto out_free_buf;
865 attach_scanned_tree(parent, child,
866 ctx->params->blob_table);
867 }
868 if (info->NextEntryOffset == 0)
869 break;
870 info = (const FILE_NAMES_INFORMATION *)
871 ((const u8 *)info + info->NextEntryOffset);
872 }
873 }
874
875 if (unlikely(status != STATUS_NO_MORE_FILES)) {
876 winnt_error(status, L"\"%ls\": Can't read directory",
877 printable_path(ctx));
878 ret = WIMLIB_ERR_READ;
879 }
880 out_free_buf:
881 FREE(buf);
882 return ret;
883 }
884
885 /* Reparse point fixup status code */
886 #define RP_FIXED (-1)
887
888 static bool
file_has_ino_and_dev(HANDLE h,u64 ino,u64 dev)889 file_has_ino_and_dev(HANDLE h, u64 ino, u64 dev)
890 {
891 NTSTATUS status;
892 IO_STATUS_BLOCK iosb;
893 FILE_INTERNAL_INFORMATION int_info;
894 FILE_FS_VOLUME_INFORMATION vol_info;
895
896 status = NtQueryInformationFile(h, &iosb, &int_info, sizeof(int_info),
897 FileInternalInformation);
898 if (!NT_SUCCESS(status))
899 return false;
900
901 if (int_info.IndexNumber.QuadPart != ino)
902 return false;
903
904 status = NtQueryVolumeInformationFile(h, &iosb,
905 &vol_info, sizeof(vol_info),
906 FileFsVolumeInformation);
907 if (!(NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW))
908 return false;
909
910 if (iosb.Information <
911 offsetof(FILE_FS_VOLUME_INFORMATION, VolumeSerialNumber) +
912 sizeof(vol_info.VolumeSerialNumber))
913 return false;
914
915 return (vol_info.VolumeSerialNumber == dev);
916 }
917
918 /*
919 * This is the Windows equivalent of unix_relativize_link_target(); see there
920 * for general details. This version works with an "absolute" Windows link
921 * target, specified from the root of the Windows kernel object namespace. Note
922 * that we have to open directories with a trailing slash when present because
923 * \??\E: opens the E: device itself and not the filesystem root directory.
924 */
925 static const wchar_t *
winnt_relativize_link_target(const wchar_t * target,size_t target_nbytes,u64 ino,u64 dev)926 winnt_relativize_link_target(const wchar_t *target, size_t target_nbytes,
927 u64 ino, u64 dev)
928 {
929 UNICODE_STRING name;
930 OBJECT_ATTRIBUTES attr;
931 IO_STATUS_BLOCK iosb;
932 NTSTATUS status;
933 const wchar_t *target_end;
934 const wchar_t *p;
935
936 target_end = target + (target_nbytes / sizeof(wchar_t));
937
938 /* Empty path??? */
939 if (target_end == target)
940 return target;
941
942 /* No leading slash??? */
943 if (target[0] != L'\\')
944 return target;
945
946 /* UNC path??? */
947 if ((target_end - target) >= 2 &&
948 target[0] == L'\\' && target[1] == L'\\')
949 return target;
950
951 attr.Length = sizeof(attr);
952 attr.RootDirectory = NULL;
953 attr.ObjectName = &name;
954 attr.Attributes = 0;
955 attr.SecurityDescriptor = NULL;
956 attr.SecurityQualityOfService = NULL;
957
958 name.Buffer = (wchar_t *)target;
959 name.Length = 0;
960 p = target;
961 do {
962 HANDLE h;
963 const wchar_t *orig_p = p;
964
965 /* Skip non-backslashes */
966 while (p != target_end && *p != L'\\')
967 p++;
968
969 /* Skip backslashes */
970 while (p != target_end && *p == L'\\')
971 p++;
972
973 /* Append path component */
974 name.Length += (p - orig_p) * sizeof(wchar_t);
975 name.MaximumLength = name.Length;
976
977 /* Try opening the file */
978 status = NtOpenFile(&h,
979 FILE_READ_ATTRIBUTES | FILE_TRAVERSE,
980 &attr,
981 &iosb,
982 FILE_SHARE_VALID_FLAGS,
983 FILE_OPEN_FOR_BACKUP_INTENT);
984
985 if (NT_SUCCESS(status)) {
986 /* Reset root directory */
987 if (attr.RootDirectory)
988 NtClose(attr.RootDirectory);
989 attr.RootDirectory = h;
990 name.Buffer = (wchar_t *)p;
991 name.Length = 0;
992
993 if (file_has_ino_and_dev(h, ino, dev))
994 goto out_close_root_dir;
995 }
996 } while (p != target_end);
997
998 p = target;
999
1000 out_close_root_dir:
1001 if (attr.RootDirectory)
1002 NtClose(attr.RootDirectory);
1003 while (p > target && *(p - 1) == L'\\')
1004 p--;
1005 return p;
1006 }
1007
1008 static int
winnt_rpfix_progress(struct scan_params * params,const struct link_reparse_point * link,int scan_status)1009 winnt_rpfix_progress(struct scan_params *params,
1010 const struct link_reparse_point *link, int scan_status)
1011 {
1012 size_t print_name_nchars = link->print_name_nbytes / sizeof(wchar_t);
1013 wchar_t print_name0[print_name_nchars + 1];
1014
1015 wmemcpy(print_name0, link->print_name, print_name_nchars);
1016 print_name0[print_name_nchars] = L'\0';
1017
1018 params->progress.scan.symlink_target = print_name0;
1019 return do_scan_progress(params, scan_status, NULL);
1020 }
1021
1022 static int
winnt_try_rpfix(struct reparse_buffer_disk * rpbuf,u16 * rpbuflen_p,struct scan_params * params)1023 winnt_try_rpfix(struct reparse_buffer_disk *rpbuf, u16 *rpbuflen_p,
1024 struct scan_params *params)
1025 {
1026 struct link_reparse_point link;
1027 const wchar_t *rel_target;
1028 int ret;
1029
1030 if (parse_link_reparse_point(rpbuf, *rpbuflen_p, &link)) {
1031 /* Couldn't understand the reparse data; don't do the fixup. */
1032 return 0;
1033 }
1034
1035 /*
1036 * Don't do reparse point fixups on relative symbolic links.
1037 *
1038 * On Windows, a relative symbolic link is supposed to be identifiable
1039 * by having reparse tag WIM_IO_REPARSE_TAG_SYMLINK and flags
1040 * SYMBOLIC_LINK_RELATIVE. We will use this information, although this
1041 * may not always do what the user expects, since drive-relative
1042 * symbolic links such as "\Users\Public" have SYMBOLIC_LINK_RELATIVE
1043 * set, in addition to truly relative symbolic links such as "Users" or
1044 * "Users\Public". However, WIMGAPI (as of Windows 8.1) has this same
1045 * behavior.
1046 *
1047 * Otherwise, as far as I can tell, the targets of symbolic links that
1048 * are NOT relative, as well as junctions (note: a mountpoint is the
1049 * sames thing as a junction), must be NT namespace paths, for example:
1050 *
1051 * - \??\e:\Users\Public
1052 * - \DosDevices\e:\Users\Public
1053 * - \Device\HardDiskVolume4\Users\Public
1054 * - \??\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public
1055 * - \DosDevices\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public
1056 */
1057 if (link_is_relative_symlink(&link))
1058 return 0;
1059
1060 rel_target = winnt_relativize_link_target(link.substitute_name,
1061 link.substitute_name_nbytes,
1062 params->capture_root_ino,
1063 params->capture_root_dev);
1064
1065 if (rel_target == link.substitute_name) {
1066 /* Target points outside of the tree being captured or had an
1067 * unrecognized path format. Don't adjust it. */
1068 return winnt_rpfix_progress(params, &link,
1069 WIMLIB_SCAN_DENTRY_NOT_FIXED_SYMLINK);
1070 }
1071
1072 /* We have an absolute target pointing within the directory being
1073 * captured. @rel_target is the suffix of the link target that is the
1074 * part relative to the directory being captured.
1075 *
1076 * We will cut off the prefix before this part (which is the path to the
1077 * directory being captured) and add a dummy prefix. Since the process
1078 * will need to be reversed when applying the image, it doesn't matter
1079 * what exactly the prefix is, as long as it looks like an absolute
1080 * path. */
1081
1082 static const wchar_t prefix[6] = L"\\??\\X:";
1083 static const size_t num_unprintable_chars = 4;
1084
1085 size_t rel_target_nbytes =
1086 link.substitute_name_nbytes - ((const u8 *)rel_target -
1087 (const u8 *)link.substitute_name);
1088
1089 wchar_t tmp[(sizeof(prefix) + rel_target_nbytes) / sizeof(wchar_t)];
1090
1091 memcpy(tmp, prefix, sizeof(prefix));
1092 memcpy(tmp + ARRAY_LEN(prefix), rel_target, rel_target_nbytes);
1093
1094 link.substitute_name = tmp;
1095 link.substitute_name_nbytes = sizeof(tmp);
1096
1097 link.print_name = link.substitute_name + num_unprintable_chars;
1098 link.print_name_nbytes = link.substitute_name_nbytes -
1099 (num_unprintable_chars * sizeof(wchar_t));
1100
1101 if (make_link_reparse_point(&link, rpbuf, rpbuflen_p))
1102 return 0;
1103
1104 ret = winnt_rpfix_progress(params, &link,
1105 WIMLIB_SCAN_DENTRY_FIXED_SYMLINK);
1106 if (ret)
1107 return ret;
1108 return RP_FIXED;
1109 }
1110
1111 /* Load the reparse data of a file into the corresponding WIM inode. If the
1112 * reparse point is a symbolic link or junction with an absolute target and
1113 * RPFIX mode is enabled, then also rewrite its target to be relative to the
1114 * capture root. */
1115 static noinline_for_stack int
winnt_load_reparse_data(HANDLE h,struct wim_inode * inode,struct winnt_scan_ctx * ctx)1116 winnt_load_reparse_data(HANDLE h, struct wim_inode *inode,
1117 struct winnt_scan_ctx *ctx)
1118 {
1119 struct reparse_buffer_disk rpbuf;
1120 NTSTATUS status;
1121 u32 len;
1122 u16 rpbuflen;
1123 int ret;
1124
1125 if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
1126 /* See comment above assign_stream_types_encrypted() */
1127 WARNING("Ignoring reparse data of encrypted file \"%ls\"",
1128 printable_path(ctx));
1129 return 0;
1130 }
1131
1132 status = winnt_fsctl(h, FSCTL_GET_REPARSE_POINT,
1133 NULL, 0, &rpbuf, sizeof(rpbuf), &len);
1134 if (!NT_SUCCESS(status)) {
1135 winnt_error(status, L"\"%ls\": Can't get reparse point",
1136 printable_path(ctx));
1137 return WIMLIB_ERR_READLINK;
1138 }
1139
1140 rpbuflen = len;
1141
1142 if (unlikely(rpbuflen < REPARSE_DATA_OFFSET)) {
1143 ERROR("\"%ls\": reparse point buffer is too short",
1144 printable_path(ctx));
1145 return WIMLIB_ERR_INVALID_REPARSE_DATA;
1146 }
1147
1148 if (le32_to_cpu(rpbuf.rptag) == WIM_IO_REPARSE_TAG_DEDUP) {
1149 /*
1150 * Windows treats Data Deduplication reparse points specially.
1151 * Reads from the unnamed data stream actually return the
1152 * redirected file contents, even with FILE_OPEN_REPARSE_POINT.
1153 * Deduplicated files also cannot be properly restored without
1154 * also restoring the "System Volume Information" directory,
1155 * which wimlib excludes by default. Therefore, the logical
1156 * behavior for us seems to be to ignore the reparse point and
1157 * treat the file as a normal file.
1158 */
1159 inode->i_attributes &= ~FILE_ATTRIBUTE_REPARSE_POINT;
1160 return 0;
1161 }
1162
1163 if (ctx->params->add_flags & WIMLIB_ADD_FLAG_RPFIX) {
1164 ret = winnt_try_rpfix(&rpbuf, &rpbuflen, ctx->params);
1165 if (ret == RP_FIXED)
1166 inode->i_rp_flags &= ~WIM_RP_FLAG_NOT_FIXED;
1167 else if (ret)
1168 return ret;
1169 }
1170
1171 inode->i_reparse_tag = le32_to_cpu(rpbuf.rptag);
1172 inode->i_rp_reserved = le16_to_cpu(rpbuf.rpreserved);
1173
1174 if (!inode_add_stream_with_data(inode,
1175 STREAM_TYPE_REPARSE_POINT,
1176 NO_STREAM_NAME,
1177 rpbuf.rpdata,
1178 rpbuflen - REPARSE_DATA_OFFSET,
1179 ctx->params->blob_table))
1180 return WIMLIB_ERR_NOMEM;
1181
1182 return 0;
1183 }
1184
1185 static DWORD WINAPI
win32_tally_encrypted_size_cb(unsigned char * _data,void * _size_ret,unsigned long len)1186 win32_tally_encrypted_size_cb(unsigned char *_data, void *_size_ret,
1187 unsigned long len)
1188 {
1189 *(u64*)_size_ret += len;
1190 return ERROR_SUCCESS;
1191 }
1192
1193 static int
win32_get_encrypted_file_size(const wchar_t * path,bool is_dir,u64 * size_ret)1194 win32_get_encrypted_file_size(const wchar_t *path, bool is_dir, u64 *size_ret)
1195 {
1196 DWORD err;
1197 void *file_ctx;
1198 int ret;
1199 DWORD flags = 0;
1200
1201 if (is_dir)
1202 flags |= CREATE_FOR_DIR;
1203
1204 err = OpenEncryptedFileRaw(path, flags, &file_ctx);
1205 if (err != ERROR_SUCCESS) {
1206 win32_error(err,
1207 L"Failed to open encrypted file \"%ls\" for raw read",
1208 path);
1209 return WIMLIB_ERR_OPEN;
1210 }
1211 *size_ret = 0;
1212 err = ReadEncryptedFileRaw(win32_tally_encrypted_size_cb,
1213 size_ret, file_ctx);
1214 if (err != ERROR_SUCCESS) {
1215 win32_error(err,
1216 L"Failed to read raw encrypted data from \"%ls\"",
1217 path);
1218 ret = WIMLIB_ERR_READ;
1219 } else {
1220 ret = 0;
1221 }
1222 CloseEncryptedFileRaw(file_ctx);
1223 return ret;
1224 }
1225
1226 static int
winnt_scan_efsrpc_raw_data(struct wim_inode * inode,struct winnt_scan_ctx * ctx)1227 winnt_scan_efsrpc_raw_data(struct wim_inode *inode,
1228 struct winnt_scan_ctx *ctx)
1229 {
1230 wchar_t *path = ctx->params->cur_path;
1231 size_t path_nchars = ctx->params->cur_path_nchars;
1232 const bool is_dir = (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY);
1233 struct windows_file *windows_file;
1234 u64 size;
1235 int ret;
1236
1237 /* OpenEncryptedFileRaw() expects a Win32 name. */
1238 wimlib_assert(!wmemcmp(path, L"\\??\\", 4));
1239 path[1] = L'\\';
1240
1241 ret = win32_get_encrypted_file_size(path, is_dir, &size);
1242 if (ret)
1243 goto out;
1244
1245 /* Empty EFSRPC data does not make sense */
1246 wimlib_assert(size != 0);
1247
1248 windows_file = alloc_windows_file(path, path_nchars, NULL, 0,
1249 ctx->snapshot, true);
1250 ret = add_stream(inode, windows_file, size, STREAM_TYPE_EFSRPC_RAW_DATA,
1251 NO_STREAM_NAME, ctx->params->unhashed_blobs);
1252 out:
1253 path[1] = L'?';
1254 return ret;
1255 }
1256
1257 static bool
get_data_stream_name(const wchar_t * raw_stream_name,size_t raw_stream_name_nchars,const wchar_t ** stream_name_ret,size_t * stream_name_nchars_ret)1258 get_data_stream_name(const wchar_t *raw_stream_name, size_t raw_stream_name_nchars,
1259 const wchar_t **stream_name_ret, size_t *stream_name_nchars_ret)
1260 {
1261 const wchar_t *sep, *type, *end;
1262
1263 /* The stream name should be returned as :NAME:TYPE */
1264 if (raw_stream_name_nchars < 1)
1265 return false;
1266 if (raw_stream_name[0] != L':')
1267 return false;
1268
1269 raw_stream_name++;
1270 raw_stream_name_nchars--;
1271
1272 end = raw_stream_name + raw_stream_name_nchars;
1273
1274 sep = wmemchr(raw_stream_name, L':', raw_stream_name_nchars);
1275 if (!sep)
1276 return false;
1277
1278 type = sep + 1;
1279 if (end - type != 5)
1280 return false;
1281
1282 if (wmemcmp(type, L"$DATA", 5))
1283 return false;
1284
1285 *stream_name_ret = raw_stream_name;
1286 *stream_name_nchars_ret = sep - raw_stream_name;
1287 return true;
1288 }
1289
1290 static int
winnt_scan_data_stream(wchar_t * raw_stream_name,size_t raw_stream_name_nchars,u64 stream_size,struct wim_inode * inode,struct winnt_scan_ctx * ctx)1291 winnt_scan_data_stream(wchar_t *raw_stream_name, size_t raw_stream_name_nchars,
1292 u64 stream_size, struct wim_inode *inode,
1293 struct winnt_scan_ctx *ctx)
1294 {
1295 wchar_t *stream_name;
1296 size_t stream_name_nchars;
1297 struct windows_file *windows_file;
1298
1299 /* Given the raw stream name (which is something like
1300 * :streamname:$DATA), extract just the stream name part (streamname).
1301 * Ignore any non-$DATA streams. */
1302 if (!get_data_stream_name(raw_stream_name, raw_stream_name_nchars,
1303 (const wchar_t **)&stream_name,
1304 &stream_name_nchars))
1305 return 0;
1306
1307 stream_name[stream_name_nchars] = L'\0';
1308
1309 windows_file = alloc_windows_file(ctx->params->cur_path,
1310 ctx->params->cur_path_nchars,
1311 stream_name, stream_name_nchars,
1312 ctx->snapshot, false);
1313 return add_stream(inode, windows_file, stream_size, STREAM_TYPE_DATA,
1314 stream_name, ctx->params->unhashed_blobs);
1315 }
1316
1317 /*
1318 * Load information about the data streams of an open file into a WIM inode.
1319 *
1320 * We use the NtQueryInformationFile() system call instead of FindFirstStream()
1321 * and FindNextStream(). This is done for two reasons:
1322 *
1323 * - FindFirstStream() opens its own handle to the file or directory and
1324 * apparently does so without specifying FILE_FLAG_BACKUP_SEMANTICS, thereby
1325 * causing access denied errors on certain files (even when running as the
1326 * Administrator).
1327 * - FindFirstStream() and FindNextStream() is only available on Windows Vista
1328 * and later, whereas the stream support in NtQueryInformationFile() was
1329 * already present in Windows XP.
1330 */
1331 static noinline_for_stack int
winnt_scan_data_streams(HANDLE h,struct wim_inode * inode,u64 file_size,struct winnt_scan_ctx * ctx)1332 winnt_scan_data_streams(HANDLE h, struct wim_inode *inode, u64 file_size,
1333 struct winnt_scan_ctx *ctx)
1334 {
1335 int ret;
1336 u8 _buf[4096] _aligned_attribute(8);
1337 u8 *buf;
1338 size_t bufsize;
1339 IO_STATUS_BLOCK iosb;
1340 NTSTATUS status;
1341 FILE_STREAM_INFORMATION *info;
1342
1343 buf = _buf;
1344 bufsize = sizeof(_buf);
1345
1346 if (!(ctx->vol_flags & FILE_NAMED_STREAMS))
1347 goto unnamed_only;
1348
1349 /* Get a buffer containing the stream information. */
1350 while (!NT_SUCCESS(status = NtQueryInformationFile(h,
1351 &iosb,
1352 buf,
1353 bufsize,
1354 FileStreamInformation)))
1355 {
1356
1357 switch (status) {
1358 case STATUS_BUFFER_OVERFLOW:
1359 {
1360 u8 *newbuf;
1361
1362 bufsize *= 2;
1363 if (buf == _buf)
1364 newbuf = MALLOC(bufsize);
1365 else
1366 newbuf = REALLOC(buf, bufsize);
1367 if (!newbuf) {
1368 ret = WIMLIB_ERR_NOMEM;
1369 goto out_free_buf;
1370 }
1371 buf = newbuf;
1372 }
1373 break;
1374 case STATUS_NOT_IMPLEMENTED:
1375 case STATUS_NOT_SUPPORTED:
1376 case STATUS_INVALID_INFO_CLASS:
1377 goto unnamed_only;
1378 default:
1379 winnt_error(status,
1380 L"\"%ls\": Failed to query stream information",
1381 printable_path(ctx));
1382 ret = WIMLIB_ERR_READ;
1383 goto out_free_buf;
1384 }
1385 }
1386
1387 if (iosb.Information == 0) {
1388 /* No stream information. */
1389 ret = 0;
1390 goto out_free_buf;
1391 }
1392
1393 /* Parse one or more stream information structures. */
1394 info = (FILE_STREAM_INFORMATION *)buf;
1395 for (;;) {
1396 /* Load the stream information. */
1397 ret = winnt_scan_data_stream(info->StreamName,
1398 info->StreamNameLength / 2,
1399 info->StreamSize.QuadPart,
1400 inode, ctx);
1401 if (ret)
1402 goto out_free_buf;
1403
1404 if (info->NextEntryOffset == 0) {
1405 /* No more stream information. */
1406 break;
1407 }
1408 /* Advance to next stream information. */
1409 info = (FILE_STREAM_INFORMATION *)
1410 ((u8 *)info + info->NextEntryOffset);
1411 }
1412 ret = 0;
1413 goto out_free_buf;
1414
1415 unnamed_only:
1416 /* The volume does not support named streams. Only capture the unnamed
1417 * data stream. */
1418 if (inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY |
1419 FILE_ATTRIBUTE_REPARSE_POINT))
1420 {
1421 ret = 0;
1422 goto out_free_buf;
1423 }
1424
1425 {
1426 wchar_t stream_name[] = L"::$DATA";
1427 ret = winnt_scan_data_stream(stream_name, 7, file_size,
1428 inode, ctx);
1429 }
1430 out_free_buf:
1431 /* Free buffer if allocated on heap. */
1432 if (unlikely(buf != _buf))
1433 FREE(buf);
1434 return ret;
1435 }
1436
1437 static u64
extract_starting_lcn(const RETRIEVAL_POINTERS_BUFFER * extents)1438 extract_starting_lcn(const RETRIEVAL_POINTERS_BUFFER *extents)
1439 {
1440 if (extents->ExtentCount < 1)
1441 return 0;
1442
1443 return extents->Extents[0].Lcn.QuadPart;
1444 }
1445
1446 static noinline_for_stack u64
get_sort_key(HANDLE h)1447 get_sort_key(HANDLE h)
1448 {
1449 STARTING_VCN_INPUT_BUFFER in = { .StartingVcn.QuadPart = 0 };
1450 RETRIEVAL_POINTERS_BUFFER out;
1451
1452 if (!NT_SUCCESS(winnt_fsctl(h, FSCTL_GET_RETRIEVAL_POINTERS,
1453 &in, sizeof(in), &out, sizeof(out), NULL)))
1454 return 0;
1455
1456 return extract_starting_lcn(&out);
1457 }
1458
1459 static void
set_sort_key(struct wim_inode * inode,u64 sort_key)1460 set_sort_key(struct wim_inode *inode, u64 sort_key)
1461 {
1462 for (unsigned i = 0; i < inode->i_num_streams; i++) {
1463 struct wim_inode_stream *strm = &inode->i_streams[i];
1464 struct blob_descriptor *blob = stream_blob_resolved(strm);
1465 if (blob && blob->blob_location == BLOB_IN_WINDOWS_FILE)
1466 blob->windows_file->sort_key = sort_key;
1467 }
1468 }
1469
1470 static inline bool
should_try_to_use_wimboot_hash(const struct wim_inode * inode,const struct winnt_scan_ctx * ctx)1471 should_try_to_use_wimboot_hash(const struct wim_inode *inode,
1472 const struct winnt_scan_ctx *ctx)
1473 {
1474 /* Directories and encrypted files aren't valid for external backing. */
1475 if (inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY |
1476 FILE_ATTRIBUTE_ENCRYPTED))
1477 return false;
1478
1479 /* If the file is a reparse point, then try the hash fixup if it's a WOF
1480 * reparse point and we're in WIMBOOT mode. Otherwise, try the hash
1481 * fixup if WOF may be attached. */
1482 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)
1483 return (inode->i_reparse_tag == WIM_IO_REPARSE_TAG_WOF) &&
1484 (ctx->params->add_flags & WIMLIB_ADD_FLAG_WIMBOOT);
1485 return !ctx->wof_not_attached;
1486 }
1487
1488 /*
1489 * This function implements an optimization for capturing files from a
1490 * filesystem with a backing WIM(s). If a file is WIM-backed, then we can
1491 * retrieve the SHA-1 message digest of its original contents from its reparse
1492 * point. This may eliminate the need to read the file's data and/or allow the
1493 * file's data to be immediately deduplicated with existing data in the WIM.
1494 *
1495 * If WOF is attached, then this function is merely an optimization, but
1496 * potentially a very effective one. If WOF is detached, then this function
1497 * really causes WIM-backed files to be, effectively, automatically
1498 * "dereferenced" when possible; the unnamed data stream is updated to reference
1499 * the original contents and the reparse point is removed.
1500 *
1501 * This function returns 0 if the fixup succeeded or was intentionally not
1502 * executed. Otherwise it returns an error code.
1503 */
1504 static noinline_for_stack int
try_to_use_wimboot_hash(HANDLE h,struct wim_inode * inode,struct winnt_scan_ctx * ctx)1505 try_to_use_wimboot_hash(HANDLE h, struct wim_inode *inode,
1506 struct winnt_scan_ctx *ctx)
1507 {
1508 struct blob_table *blob_table = ctx->params->blob_table;
1509 struct wim_inode_stream *reparse_strm = NULL;
1510 struct wim_inode_stream *strm;
1511 struct blob_descriptor *blob;
1512 u8 hash[SHA1_HASH_SIZE];
1513 int ret;
1514
1515 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1516 struct reparse_buffer_disk rpbuf;
1517 struct {
1518 struct wof_external_info wof_info;
1519 struct wim_provider_rpdata wim_info;
1520 } *rpdata = (void *)rpbuf.rpdata;
1521 struct blob_descriptor *reparse_blob;
1522
1523 /* The file has a WOF reparse point, so WOF must be detached.
1524 * We can read the reparse point directly. */
1525 ctx->wof_not_attached = true;
1526 reparse_strm = inode_get_unnamed_stream(inode, STREAM_TYPE_REPARSE_POINT);
1527 reparse_blob = stream_blob_resolved(reparse_strm);
1528
1529 if (!reparse_blob || reparse_blob->size < sizeof(*rpdata))
1530 return 0; /* Not a WIM-backed file */
1531
1532 ret = read_blob_into_buf(reparse_blob, rpdata);
1533 if (ret)
1534 return ret;
1535
1536 if (rpdata->wof_info.version != WOF_CURRENT_VERSION ||
1537 rpdata->wof_info.provider != WOF_PROVIDER_WIM ||
1538 rpdata->wim_info.version != 2)
1539 return 0; /* Not a WIM-backed file */
1540
1541 /* Okay, this is a WIM backed file. Get its SHA-1 hash. */
1542 copy_hash(hash, rpdata->wim_info.unnamed_data_stream_hash);
1543 } else {
1544 struct {
1545 struct wof_external_info wof_info;
1546 struct wim_provider_external_info wim_info;
1547 } out;
1548 NTSTATUS status;
1549
1550 /* WOF may be attached. Try reading this file's external
1551 * backing info. */
1552 status = winnt_fsctl(h, FSCTL_GET_EXTERNAL_BACKING,
1553 NULL, 0, &out, sizeof(out), NULL);
1554
1555 /* Is WOF not attached? */
1556 if (status == STATUS_INVALID_DEVICE_REQUEST ||
1557 status == STATUS_NOT_SUPPORTED) {
1558 ctx->wof_not_attached = true;
1559 return 0;
1560 }
1561
1562 /* Is this file not externally backed? */
1563 if (status == STATUS_OBJECT_NOT_EXTERNALLY_BACKED)
1564 return 0;
1565
1566 /* Does this file have an unknown type of external backing that
1567 * needed a larger information buffer? */
1568 if (status == STATUS_BUFFER_TOO_SMALL)
1569 return 0;
1570
1571 /* Was there some other failure? */
1572 if (status != STATUS_SUCCESS) {
1573 winnt_error(status,
1574 L"\"%ls\": FSCTL_GET_EXTERNAL_BACKING failed",
1575 printable_path(ctx));
1576 return WIMLIB_ERR_STAT;
1577 }
1578
1579 /* Is this file backed by a WIM? */
1580 if (out.wof_info.version != WOF_CURRENT_VERSION ||
1581 out.wof_info.provider != WOF_PROVIDER_WIM ||
1582 out.wim_info.version != WIM_PROVIDER_CURRENT_VERSION)
1583 return 0;
1584
1585 /* Okay, this is a WIM backed file. Get its SHA-1 hash. */
1586 copy_hash(hash, out.wim_info.unnamed_data_stream_hash);
1587 }
1588
1589 /* If the file's unnamed data stream is nonempty, then fill in its hash
1590 * and deduplicate it if possible.
1591 *
1592 * With WOF detached, we require that the blob *must* de-duplicable for
1593 * any action can be taken, since without WOF we can't fall back to
1594 * getting the "dereferenced" data by reading the stream (the real
1595 * stream is sparse and contains all zeroes). */
1596 strm = inode_get_unnamed_data_stream(inode);
1597 if (strm && (blob = stream_blob_resolved(strm))) {
1598 struct blob_descriptor **back_ptr;
1599
1600 if (reparse_strm && !lookup_blob(blob_table, hash))
1601 return 0;
1602 back_ptr = retrieve_pointer_to_unhashed_blob(blob);
1603 copy_hash(blob->hash, hash);
1604 if (after_blob_hashed(blob, back_ptr, blob_table) != blob)
1605 free_blob_descriptor(blob);
1606 }
1607
1608 /* Remove the reparse point, if present. */
1609 if (reparse_strm) {
1610 inode_remove_stream(inode, reparse_strm, blob_table);
1611 inode->i_attributes &= ~(FILE_ATTRIBUTE_REPARSE_POINT |
1612 FILE_ATTRIBUTE_SPARSE_FILE);
1613 if (inode->i_attributes == 0)
1614 inode->i_attributes = FILE_ATTRIBUTE_NORMAL;
1615 }
1616
1617 return 0;
1618 }
1619
1620 struct file_info {
1621 u32 attributes;
1622 u32 num_links;
1623 u64 creation_time;
1624 u64 last_write_time;
1625 u64 last_access_time;
1626 u64 ino;
1627 u64 end_of_file;
1628 u32 ea_size;
1629 };
1630
1631 static noinline_for_stack NTSTATUS
get_file_info(HANDLE h,struct file_info * info)1632 get_file_info(HANDLE h, struct file_info *info)
1633 {
1634 IO_STATUS_BLOCK iosb;
1635 NTSTATUS status;
1636 FILE_ALL_INFORMATION all_info;
1637
1638 status = NtQueryInformationFile(h, &iosb, &all_info, sizeof(all_info),
1639 FileAllInformation);
1640
1641 if (unlikely(!NT_SUCCESS(status) && status != STATUS_BUFFER_OVERFLOW))
1642 return status;
1643
1644 info->attributes = all_info.BasicInformation.FileAttributes;
1645 info->num_links = all_info.StandardInformation.NumberOfLinks;
1646 info->creation_time = all_info.BasicInformation.CreationTime.QuadPart;
1647 info->last_write_time = all_info.BasicInformation.LastWriteTime.QuadPart;
1648 info->last_access_time = all_info.BasicInformation.LastAccessTime.QuadPart;
1649 info->ino = all_info.InternalInformation.IndexNumber.QuadPart;
1650 info->end_of_file = all_info.StandardInformation.EndOfFile.QuadPart;
1651 info->ea_size = all_info.EaInformation.EaSize;
1652 return STATUS_SUCCESS;
1653 }
1654
1655 static void
get_volume_information(HANDLE h,struct winnt_scan_ctx * ctx)1656 get_volume_information(HANDLE h, struct winnt_scan_ctx *ctx)
1657 {
1658 u8 _attr_info[sizeof(FILE_FS_ATTRIBUTE_INFORMATION) + 128] _aligned_attribute(8);
1659 FILE_FS_ATTRIBUTE_INFORMATION *attr_info = (void *)_attr_info;
1660 FILE_FS_VOLUME_INFORMATION vol_info;
1661 struct file_info file_info;
1662 IO_STATUS_BLOCK iosb;
1663 NTSTATUS status;
1664
1665 /* Get volume flags */
1666 status = NtQueryVolumeInformationFile(h, &iosb, attr_info,
1667 sizeof(_attr_info),
1668 FileFsAttributeInformation);
1669 if (NT_SUCCESS(status)) {
1670 ctx->vol_flags = attr_info->FileSystemAttributes;
1671 ctx->is_ntfs = (attr_info->FileSystemNameLength == 4 * sizeof(wchar_t)) &&
1672 !wmemcmp(attr_info->FileSystemName, L"NTFS", 4);
1673 } else {
1674 winnt_warning(status, L"\"%ls\": Can't get volume attributes",
1675 printable_path(ctx));
1676 }
1677
1678 /* Get volume ID. */
1679 status = NtQueryVolumeInformationFile(h, &iosb, &vol_info,
1680 sizeof(vol_info),
1681 FileFsVolumeInformation);
1682 if ((NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW) &&
1683 (iosb.Information >= offsetof(FILE_FS_VOLUME_INFORMATION,
1684 VolumeSerialNumber) +
1685 sizeof(vol_info.VolumeSerialNumber)))
1686 {
1687 ctx->params->capture_root_dev = vol_info.VolumeSerialNumber;
1688 } else {
1689 winnt_warning(status, L"\"%ls\": Can't get volume ID",
1690 printable_path(ctx));
1691 }
1692
1693 /* Get inode number. */
1694 status = get_file_info(h, &file_info);
1695 if (NT_SUCCESS(status)) {
1696 ctx->params->capture_root_ino = file_info.ino;
1697 } else {
1698 winnt_warning(status, L"\"%ls\": Can't get file information",
1699 printable_path(ctx));
1700 }
1701 }
1702
1703 static int
winnt_build_dentry_tree_recursive(struct wim_dentry ** root_ret,HANDLE cur_dir,const wchar_t * relative_path,size_t relative_path_nchars,const wchar_t * filename,struct winnt_scan_ctx * ctx)1704 winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret,
1705 HANDLE cur_dir,
1706 const wchar_t *relative_path,
1707 size_t relative_path_nchars,
1708 const wchar_t *filename,
1709 struct winnt_scan_ctx *ctx)
1710 {
1711 struct wim_dentry *root = NULL;
1712 struct wim_inode *inode = NULL;
1713 HANDLE h = NULL;
1714 int ret;
1715 NTSTATUS status;
1716 struct file_info file_info;
1717 u64 sort_key;
1718
1719 ret = try_exclude(ctx->params);
1720 if (unlikely(ret < 0)) /* Excluded? */
1721 goto out_progress;
1722 if (unlikely(ret > 0)) /* Error? */
1723 goto out;
1724
1725 /* Open the file with permission to read metadata. Although we will
1726 * later need a handle with FILE_LIST_DIRECTORY permission (or,
1727 * equivalently, FILE_READ_DATA; they're the same numeric value) if the
1728 * file is a directory, it can significantly slow things down to request
1729 * this permission on all nondirectories. Perhaps it causes Windows to
1730 * start prefetching the file contents... */
1731 status = winnt_openat(cur_dir, relative_path, relative_path_nchars,
1732 FILE_READ_ATTRIBUTES | FILE_READ_EA |
1733 READ_CONTROL | ACCESS_SYSTEM_SECURITY,
1734 &h);
1735 if (unlikely(!NT_SUCCESS(status))) {
1736 if (status == STATUS_DELETE_PENDING) {
1737 WARNING("\"%ls\": Deletion pending; skipping file",
1738 printable_path(ctx));
1739 ret = 0;
1740 goto out;
1741 }
1742 if (status == STATUS_SHARING_VIOLATION) {
1743 ERROR("Can't open \"%ls\":\n"
1744 " File is in use by another process! "
1745 "Consider using snapshot (VSS) mode.",
1746 printable_path(ctx));
1747 ret = WIMLIB_ERR_OPEN;
1748 goto out;
1749 }
1750 winnt_error(status, L"\"%ls\": Can't open file",
1751 printable_path(ctx));
1752 if (status == STATUS_FVE_LOCKED_VOLUME)
1753 ret = WIMLIB_ERR_FVE_LOCKED_VOLUME;
1754 else
1755 ret = WIMLIB_ERR_OPEN;
1756 goto out;
1757 }
1758
1759 /* Get information about the file. */
1760 status = get_file_info(h, &file_info);
1761 if (!NT_SUCCESS(status)) {
1762 winnt_error(status, L"\"%ls\": Can't get file information",
1763 printable_path(ctx));
1764 ret = WIMLIB_ERR_STAT;
1765 goto out;
1766 }
1767
1768 /* Create a WIM dentry with an associated inode, which may be shared.
1769 *
1770 * However, we need to explicitly check for directories and files with
1771 * only 1 link and refuse to hard link them. This is because Windows
1772 * has a bug where it can return duplicate File IDs for files and
1773 * directories on the FAT filesystem.
1774 *
1775 * Since we don't follow mount points on Windows, we don't need to query
1776 * the volume ID per-file. Just once, for the root, is enough. But we
1777 * can't simply pass 0, because then there could be inode collisions
1778 * among multiple calls to win32_build_dentry_tree() that are scanning
1779 * files on different volumes. */
1780 ret = inode_table_new_dentry(ctx->params->inode_table,
1781 filename,
1782 file_info.ino,
1783 ctx->params->capture_root_dev,
1784 (file_info.num_links <= 1),
1785 &root);
1786 if (ret)
1787 goto out;
1788
1789 /* Get the short (DOS) name of the file. */
1790 status = winnt_get_short_name(h, root);
1791
1792 /* If we can't read the short filename for any reason other than
1793 * out-of-memory, just ignore the error and assume the file has no short
1794 * name. This shouldn't be an issue, since the short names are
1795 * essentially obsolete anyway. */
1796 if (unlikely(status == STATUS_NO_MEMORY)) {
1797 ret = WIMLIB_ERR_NOMEM;
1798 goto out;
1799 }
1800
1801 inode = root->d_inode;
1802
1803 if (inode->i_nlink > 1) {
1804 /* Shared inode (hard link); skip reading per-inode information.
1805 */
1806 goto out_progress;
1807 }
1808
1809 inode->i_attributes = file_info.attributes;
1810 inode->i_creation_time = file_info.creation_time;
1811 inode->i_last_write_time = file_info.last_write_time;
1812 inode->i_last_access_time = file_info.last_access_time;
1813
1814 /* Get the file's security descriptor, unless we are capturing in
1815 * NO_ACLS mode or the volume does not support security descriptors. */
1816 if (!(ctx->params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)
1817 && (ctx->vol_flags & FILE_PERSISTENT_ACLS))
1818 {
1819 ret = winnt_load_security_descriptor(h, inode, ctx);
1820 if (ret)
1821 goto out;
1822 }
1823
1824 /* Get the file's object ID. */
1825 ret = winnt_load_object_id(h, inode, ctx);
1826 if (ret)
1827 goto out;
1828
1829 /* Get the file's extended attributes. */
1830 if (unlikely(file_info.ea_size != 0)) {
1831 ret = winnt_load_xattrs(h, inode, ctx, file_info.ea_size);
1832 if (ret)
1833 goto out;
1834 }
1835
1836 /* If this is a reparse point, load the reparse data. */
1837 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)) {
1838 ret = winnt_load_reparse_data(h, inode, ctx);
1839 if (ret)
1840 goto out;
1841 }
1842
1843 sort_key = get_sort_key(h);
1844
1845 if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
1846 /* Load information about the raw encrypted data. This is
1847 * needed for any directory or non-directory that has
1848 * FILE_ATTRIBUTE_ENCRYPTED set.
1849 *
1850 * Note: since OpenEncryptedFileRaw() fails with
1851 * ERROR_SHARING_VIOLATION if there are any open handles to the
1852 * file, we have to close the file and re-open it later if
1853 * needed. */
1854 NtClose(h);
1855 h = NULL;
1856 ret = winnt_scan_efsrpc_raw_data(inode, ctx);
1857 if (ret)
1858 goto out;
1859 } else {
1860 /*
1861 * Load information about data streams (unnamed and named).
1862 *
1863 * Skip this step for encrypted files, since the data from
1864 * ReadEncryptedFileRaw() already contains all data streams (and
1865 * they do in fact all get restored by WriteEncryptedFileRaw().)
1866 *
1867 * Note: WIMGAPI (as of Windows 8.1) gets wrong and stores both
1868 * the EFSRPC data and the named data stream(s)...!
1869 */
1870 ret = winnt_scan_data_streams(h,
1871 inode,
1872 file_info.end_of_file,
1873 ctx);
1874 if (ret)
1875 goto out;
1876 }
1877
1878 if (unlikely(should_try_to_use_wimboot_hash(inode, ctx))) {
1879 ret = try_to_use_wimboot_hash(h, inode, ctx);
1880 if (ret)
1881 goto out;
1882 }
1883
1884 set_sort_key(inode, sort_key);
1885
1886 if (inode_is_directory(inode)) {
1887
1888 /* Directory: recurse to children. */
1889
1890 /* Re-open the directory with FILE_LIST_DIRECTORY access. */
1891 if (h) {
1892 NtClose(h);
1893 h = NULL;
1894 }
1895 status = winnt_openat(cur_dir, relative_path,
1896 relative_path_nchars, FILE_LIST_DIRECTORY,
1897 &h);
1898 if (!NT_SUCCESS(status)) {
1899 winnt_error(status, L"\"%ls\": Can't open directory",
1900 printable_path(ctx));
1901 ret = WIMLIB_ERR_OPEN;
1902 goto out;
1903 }
1904 ret = winnt_recurse_directory(h, root, ctx);
1905 if (ret)
1906 goto out;
1907 }
1908
1909 out_progress:
1910 if (likely(root))
1911 ret = do_scan_progress(ctx->params, WIMLIB_SCAN_DENTRY_OK, inode);
1912 else
1913 ret = do_scan_progress(ctx->params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL);
1914 out:
1915 if (likely(h))
1916 NtClose(h);
1917 if (unlikely(ret)) {
1918 free_dentry_tree(root, ctx->params->blob_table);
1919 root = NULL;
1920 ret = report_scan_error(ctx->params, ret);
1921 }
1922 *root_ret = root;
1923 return ret;
1924 }
1925
1926 static void
winnt_do_scan_warnings(const wchar_t * path,const struct winnt_scan_ctx * ctx)1927 winnt_do_scan_warnings(const wchar_t *path, const struct winnt_scan_ctx *ctx)
1928 {
1929 if (likely(ctx->num_get_sacl_priv_notheld == 0 &&
1930 ctx->num_get_sd_access_denied == 0))
1931 return;
1932
1933 WARNING("Scan of \"%ls\" complete, but with one or more warnings:", path);
1934 if (ctx->num_get_sacl_priv_notheld != 0) {
1935 WARNING("- Could not capture SACL (System Access Control List)\n"
1936 " on %lu files or directories.",
1937 ctx->num_get_sacl_priv_notheld);
1938 }
1939 if (ctx->num_get_sd_access_denied != 0) {
1940 WARNING("- Could not capture security descriptor at all\n"
1941 " on %lu files or directories.",
1942 ctx->num_get_sd_access_denied);
1943 }
1944 WARNING("To fully capture all security descriptors, run the program\n"
1945 " with Administrator rights.");
1946 }
1947
1948 /*----------------------------------------------------------------------------*
1949 * Fast MFT scan implementation *
1950 *----------------------------------------------------------------------------*/
1951
1952 #define ENABLE_FAST_MFT_SCAN 1
1953
1954 #ifdef ENABLE_FAST_MFT_SCAN
1955
1956 typedef struct {
1957 u64 StartingCluster;
1958 u64 ClusterCount;
1959 } CLUSTER_RANGE;
1960
1961 typedef struct {
1962 u64 StartingFileReferenceNumber;
1963 u64 EndingFileReferenceNumber;
1964 } FILE_REFERENCE_RANGE;
1965
1966 /* The FSCTL_QUERY_FILE_LAYOUT ioctl. This ioctl can be used on Windows 8 and
1967 * later to scan the MFT of an NTFS volume. */
1968 #define FSCTL_QUERY_FILE_LAYOUT CTL_CODE(FILE_DEVICE_FILE_SYSTEM, 157, METHOD_NEITHER, FILE_ANY_ACCESS)
1969
1970 /* The input to FSCTL_QUERY_FILE_LAYOUT */
1971 typedef struct {
1972 u32 NumberOfPairs;
1973 #define QUERY_FILE_LAYOUT_RESTART 0x00000001
1974 #define QUERY_FILE_LAYOUT_INCLUDE_NAMES 0x00000002
1975 #define QUERY_FILE_LAYOUT_INCLUDE_STREAMS 0x00000004
1976 #define QUERY_FILE_LAYOUT_INCLUDE_EXTENTS 0x00000008
1977 #define QUERY_FILE_LAYOUT_INCLUDE_EXTRA_INFO 0x00000010
1978 #define QUERY_FILE_LAYOUT_INCLUDE_STREAMS_WITH_NO_CLUSTERS_ALLOCATED 0x00000020
1979 u32 Flags;
1980 #define QUERY_FILE_LAYOUT_FILTER_TYPE_NONE 0
1981 #define QUERY_FILE_LAYOUT_FILTER_TYPE_CLUSTERS 1
1982 #define QUERY_FILE_LAYOUT_FILTER_TYPE_FILEID 2
1983 #define QUERY_FILE_LAYOUT_NUM_FILTER_TYPES 3
1984 u32 FilterType;
1985 u32 Reserved;
1986 union {
1987 CLUSTER_RANGE ClusterRanges[1];
1988 FILE_REFERENCE_RANGE FileReferenceRanges[1];
1989 } Filter;
1990 } QUERY_FILE_LAYOUT_INPUT;
1991
1992 /* The header of the buffer returned by FSCTL_QUERY_FILE_LAYOUT */
1993 typedef struct {
1994 u32 FileEntryCount;
1995 u32 FirstFileOffset;
1996 #define QUERY_FILE_LAYOUT_SINGLE_INSTANCED 0x00000001
1997 u32 Flags;
1998 u32 Reserved;
1999 } QUERY_FILE_LAYOUT_OUTPUT;
2000
2001 /* Inode information returned by FSCTL_QUERY_FILE_LAYOUT */
2002 typedef struct {
2003 u32 Version;
2004 u32 NextFileOffset;
2005 u32 Flags;
2006 u32 FileAttributes;
2007 u64 FileReferenceNumber;
2008 u32 FirstNameOffset;
2009 u32 FirstStreamOffset;
2010 u32 ExtraInfoOffset;
2011 u32 Reserved;
2012 } FILE_LAYOUT_ENTRY;
2013
2014 /* Extra inode information returned by FSCTL_QUERY_FILE_LAYOUT */
2015 typedef struct {
2016 struct {
2017 u64 CreationTime;
2018 u64 LastAccessTime;
2019 u64 LastWriteTime;
2020 u64 ChangeTime;
2021 u32 FileAttributes;
2022 } BasicInformation;
2023 u32 OwnerId;
2024 u32 SecurityId;
2025 s64 Usn;
2026 } FILE_LAYOUT_INFO_ENTRY;
2027
2028 /* Filename (or dentry) information returned by FSCTL_QUERY_FILE_LAYOUT */
2029 typedef struct {
2030 u32 NextNameOffset;
2031 #define FILE_LAYOUT_NAME_ENTRY_PRIMARY 0x00000001
2032 #define FILE_LAYOUT_NAME_ENTRY_DOS 0x00000002
2033 u32 Flags;
2034 u64 ParentFileReferenceNumber;
2035 u32 FileNameLength;
2036 u32 Reserved;
2037 wchar_t FileName[1];
2038 } FILE_LAYOUT_NAME_ENTRY;
2039
2040 /* Stream information returned by FSCTL_QUERY_FILE_LAYOUT */
2041 typedef struct {
2042 u32 Version;
2043 u32 NextStreamOffset;
2044 #define STREAM_LAYOUT_ENTRY_IMMOVABLE 0x00000001
2045 #define STREAM_LAYOUT_ENTRY_PINNED 0x00000002
2046 #define STREAM_LAYOUT_ENTRY_RESIDENT 0x00000004
2047 #define STREAM_LAYOUT_ENTRY_NO_CLUSTERS_ALLOCATED 0x00000008
2048 u32 Flags;
2049 u32 ExtentInformationOffset;
2050 u64 AllocationSize;
2051 u64 EndOfFile;
2052 u64 Reserved;
2053 u32 AttributeFlags;
2054 u32 StreamIdentifierLength;
2055 wchar_t StreamIdentifier[1];
2056 } STREAM_LAYOUT_ENTRY;
2057
2058
2059 typedef struct {
2060 #define STREAM_EXTENT_ENTRY_AS_RETRIEVAL_POINTERS 0x00000001
2061 #define STREAM_EXTENT_ENTRY_ALL_EXTENTS 0x00000002
2062 u32 Flags;
2063 union {
2064 RETRIEVAL_POINTERS_BUFFER RetrievalPointers;
2065 } ExtentInformation;
2066 } STREAM_EXTENT_ENTRY;
2067
2068 /* Extract the MFT number part of the full inode number */
2069 #define NTFS_MFT_NO(ref) ((ref) & (((u64)1 << 48) - 1))
2070
2071 /* Is the file the root directory of the NTFS volume? The root directory always
2072 * occupies MFT record 5. */
2073 #define NTFS_IS_ROOT_FILE(ino) (NTFS_MFT_NO(ino) == 5)
2074
2075 /* Is the file a special NTFS file, other than the root directory? The special
2076 * files are the first 16 records in the MFT. */
2077 #define NTFS_IS_SPECIAL_FILE(ino) \
2078 (NTFS_MFT_NO(ino) <= 15 && !NTFS_IS_ROOT_FILE(ino))
2079
2080 #define NTFS_SPECIAL_STREAM_OBJECT_ID 0x00000001
2081 #define NTFS_SPECIAL_STREAM_EA 0x00000002
2082 #define NTFS_SPECIAL_STREAM_EA_INFORMATION 0x00000004
2083
2084 /* Intermediate inode structure. This is used to temporarily save information
2085 * from FSCTL_QUERY_FILE_LAYOUT before creating the full 'struct wim_inode'. */
2086 struct ntfs_inode {
2087 struct avl_tree_node index_node;
2088 u64 ino;
2089 u64 creation_time;
2090 u64 last_access_time;
2091 u64 last_write_time;
2092 u64 starting_lcn;
2093 u32 attributes;
2094 u32 security_id;
2095 u32 num_aliases;
2096 u32 num_streams;
2097 u32 special_streams;
2098 u32 first_stream_offset;
2099 struct ntfs_dentry *first_child;
2100 wchar_t short_name[13];
2101 };
2102
2103 /* Intermediate dentry structure. This is used to temporarily save information
2104 * from FSCTL_QUERY_FILE_LAYOUT before creating the full 'struct wim_dentry'. */
2105 struct ntfs_dentry {
2106 u32 offset_from_inode : 31;
2107 u32 is_primary : 1;
2108 union {
2109 /* Note: build_children_lists() replaces 'parent_ino' with
2110 * 'next_child'. */
2111 u64 parent_ino;
2112 struct ntfs_dentry *next_child;
2113 };
2114 wchar_t name[0];
2115 };
2116
2117 /* Intermediate stream structure. This is used to temporarily save information
2118 * from FSCTL_QUERY_FILE_LAYOUT before creating the full 'struct
2119 * wim_inode_stream'. */
2120 struct ntfs_stream {
2121 u64 size;
2122 wchar_t name[0];
2123 };
2124
2125 /* Map of all known NTFS inodes, keyed by inode number */
2126 struct ntfs_inode_map {
2127 struct avl_tree_node *root;
2128 };
2129
2130 #define NTFS_INODE(node) \
2131 avl_tree_entry((node), struct ntfs_inode, index_node)
2132
2133 #define SKIP_ALIGNED(p, size) ((void *)(p) + ALIGN((size), 8))
2134
2135 /* Get a pointer to the first dentry of the inode. */
2136 #define FIRST_DENTRY(ni) SKIP_ALIGNED((ni), sizeof(struct ntfs_inode))
2137
2138 /* Get a pointer to the first stream of the inode. */
2139 #define FIRST_STREAM(ni) ((const void *)ni + ni->first_stream_offset)
2140
2141 /* Advance to the next dentry of the inode. */
2142 #define NEXT_DENTRY(nd) SKIP_ALIGNED((nd), sizeof(struct ntfs_dentry) + \
2143 (wcslen((nd)->name) + 1) * sizeof(wchar_t))
2144
2145 /* Advance to the next stream of the inode. */
2146 #define NEXT_STREAM(ns) SKIP_ALIGNED((ns), sizeof(struct ntfs_stream) + \
2147 (wcslen((ns)->name) + 1) * sizeof(wchar_t))
2148
2149 static int
_avl_cmp_ntfs_inodes(const struct avl_tree_node * node1,const struct avl_tree_node * node2)2150 _avl_cmp_ntfs_inodes(const struct avl_tree_node *node1,
2151 const struct avl_tree_node *node2)
2152 {
2153 return cmp_u64(NTFS_INODE(node1)->ino, NTFS_INODE(node2)->ino);
2154 }
2155
2156 /* Adds an NTFS inode to the map. */
2157 static void
ntfs_inode_map_add_inode(struct ntfs_inode_map * map,struct ntfs_inode * ni)2158 ntfs_inode_map_add_inode(struct ntfs_inode_map *map, struct ntfs_inode *ni)
2159 {
2160 if (avl_tree_insert(&map->root, &ni->index_node, _avl_cmp_ntfs_inodes)) {
2161 WARNING("Inode 0x%016"PRIx64" is a duplicate!", ni->ino);
2162 FREE(ni);
2163 }
2164 }
2165
2166 /* Find an ntfs_inode in the map by inode number. Returns NULL if not found. */
2167 static struct ntfs_inode *
ntfs_inode_map_lookup(struct ntfs_inode_map * map,u64 ino)2168 ntfs_inode_map_lookup(struct ntfs_inode_map *map, u64 ino)
2169 {
2170 struct ntfs_inode tmp;
2171 struct avl_tree_node *res;
2172
2173 tmp.ino = ino;
2174 res = avl_tree_lookup_node(map->root, &tmp.index_node, _avl_cmp_ntfs_inodes);
2175 if (!res)
2176 return NULL;
2177 return NTFS_INODE(res);
2178 }
2179
2180 /* Remove an ntfs_inode from the map and free it. */
2181 static void
ntfs_inode_map_remove(struct ntfs_inode_map * map,struct ntfs_inode * ni)2182 ntfs_inode_map_remove(struct ntfs_inode_map *map, struct ntfs_inode *ni)
2183 {
2184 avl_tree_remove(&map->root, &ni->index_node);
2185 FREE(ni);
2186 }
2187
2188 /* Free all ntfs_inodes in the map. */
2189 static void
ntfs_inode_map_destroy(struct ntfs_inode_map * map)2190 ntfs_inode_map_destroy(struct ntfs_inode_map *map)
2191 {
2192 struct ntfs_inode *ni;
2193
2194 avl_tree_for_each_in_postorder(ni, map->root, struct ntfs_inode, index_node)
2195 FREE(ni);
2196 }
2197
2198 static bool
file_has_streams(const FILE_LAYOUT_ENTRY * file)2199 file_has_streams(const FILE_LAYOUT_ENTRY *file)
2200 {
2201 return (file->FirstStreamOffset != 0) &&
2202 !(file->FileAttributes & FILE_ATTRIBUTE_ENCRYPTED);
2203 }
2204
2205 static bool
is_valid_name_entry(const FILE_LAYOUT_NAME_ENTRY * name)2206 is_valid_name_entry(const FILE_LAYOUT_NAME_ENTRY *name)
2207 {
2208 return name->FileNameLength > 0 &&
2209 name->FileNameLength % 2 == 0 &&
2210 !wmemchr(name->FileName, L'\0', name->FileNameLength / 2) &&
2211 (!(name->Flags & FILE_LAYOUT_NAME_ENTRY_DOS) ||
2212 name->FileNameLength <= 24);
2213 }
2214
2215 /* Validate the FILE_LAYOUT_NAME_ENTRYs of the specified file and compute the
2216 * total length in bytes of the ntfs_dentry structures needed to hold the name
2217 * information. */
2218 static int
validate_names_and_compute_total_length(const FILE_LAYOUT_ENTRY * file,size_t * total_length_ret)2219 validate_names_and_compute_total_length(const FILE_LAYOUT_ENTRY *file,
2220 size_t *total_length_ret)
2221 {
2222 const FILE_LAYOUT_NAME_ENTRY *name =
2223 (const void *)file + file->FirstNameOffset;
2224 size_t total = 0;
2225 size_t num_long_names = 0;
2226
2227 for (;;) {
2228 if (unlikely(!is_valid_name_entry(name))) {
2229 ERROR("Invalid FILE_LAYOUT_NAME_ENTRY! "
2230 "FileReferenceNumber=0x%016"PRIx64", "
2231 "FileNameLength=%"PRIu32", "
2232 "FileName=%.*ls, Flags=0x%08"PRIx32,
2233 file->FileReferenceNumber,
2234 name->FileNameLength,
2235 (int)(name->FileNameLength / 2),
2236 name->FileName, name->Flags);
2237 return WIMLIB_ERR_UNSUPPORTED;
2238 }
2239 if (name->Flags != FILE_LAYOUT_NAME_ENTRY_DOS) {
2240 num_long_names++;
2241 total += ALIGN(sizeof(struct ntfs_dentry) +
2242 name->FileNameLength + sizeof(wchar_t),
2243 8);
2244 }
2245 if (name->NextNameOffset == 0)
2246 break;
2247 name = (const void *)name + name->NextNameOffset;
2248 }
2249
2250 if (unlikely(num_long_names == 0)) {
2251 ERROR("Inode 0x%016"PRIx64" has no long names!",
2252 file->FileReferenceNumber);
2253 return WIMLIB_ERR_UNSUPPORTED;
2254 }
2255
2256 *total_length_ret = total;
2257 return 0;
2258 }
2259
2260 static bool
is_valid_stream_entry(const STREAM_LAYOUT_ENTRY * stream)2261 is_valid_stream_entry(const STREAM_LAYOUT_ENTRY *stream)
2262 {
2263 return stream->StreamIdentifierLength % 2 == 0 &&
2264 !wmemchr(stream->StreamIdentifier , L'\0',
2265 stream->StreamIdentifierLength / 2);
2266 }
2267
2268 /* assumes that 'id' is a wide string literal */
2269 #define stream_has_identifier(stream, id) \
2270 ((stream)->StreamIdentifierLength == sizeof(id) - 2 && \
2271 !memcmp((stream)->StreamIdentifier, id, sizeof(id) - 2))
2272 /*
2273 * If the specified STREAM_LAYOUT_ENTRY represents a DATA stream as opposed to
2274 * some other type of NTFS stream such as a STANDARD_INFORMATION stream, return
2275 * true and set *stream_name_ret and *stream_name_nchars_ret to specify just the
2276 * stream name. For example, ":foo:$DATA" would become "foo" with length 3
2277 * characters. Otherwise return false.
2278 */
2279 static bool
use_stream(const FILE_LAYOUT_ENTRY * file,const STREAM_LAYOUT_ENTRY * stream,const wchar_t ** stream_name_ret,size_t * stream_name_nchars_ret)2280 use_stream(const FILE_LAYOUT_ENTRY *file, const STREAM_LAYOUT_ENTRY *stream,
2281 const wchar_t **stream_name_ret, size_t *stream_name_nchars_ret)
2282 {
2283 const wchar_t *stream_name;
2284 size_t stream_name_nchars;
2285
2286 if (stream->StreamIdentifierLength == 0) {
2287 /* The unnamed data stream may be given as an empty string
2288 * rather than as "::$DATA". Handle it both ways. */
2289 stream_name = L"";
2290 stream_name_nchars = 0;
2291 } else if (!get_data_stream_name(stream->StreamIdentifier,
2292 stream->StreamIdentifierLength / 2,
2293 &stream_name, &stream_name_nchars))
2294 return false;
2295
2296 /* Skip the unnamed data stream for directories. */
2297 if (stream_name_nchars == 0 &&
2298 (file->FileAttributes & FILE_ATTRIBUTE_DIRECTORY))
2299 return false;
2300
2301 *stream_name_ret = stream_name;
2302 *stream_name_nchars_ret = stream_name_nchars;
2303 return true;
2304 }
2305
2306 /* Validate the STREAM_LAYOUT_ENTRYs of the specified file and compute the total
2307 * length in bytes of the ntfs_stream structures needed to hold the stream
2308 * information. In addition, set *special_streams_ret to a bitmask of special
2309 * stream types that were found. */
2310 static int
validate_streams_and_compute_total_length(const FILE_LAYOUT_ENTRY * file,size_t * total_length_ret,u32 * special_streams_ret)2311 validate_streams_and_compute_total_length(const FILE_LAYOUT_ENTRY *file,
2312 size_t *total_length_ret,
2313 u32 *special_streams_ret)
2314 {
2315 const STREAM_LAYOUT_ENTRY *stream =
2316 (const void *)file + file->FirstStreamOffset;
2317 size_t total = 0;
2318 u32 special_streams = 0;
2319
2320 for (;;) {
2321 const wchar_t *name;
2322 size_t name_nchars;
2323
2324 if (unlikely(!is_valid_stream_entry(stream))) {
2325 WARNING("Invalid STREAM_LAYOUT_ENTRY! "
2326 "FileReferenceNumber=0x%016"PRIx64", "
2327 "StreamIdentifierLength=%"PRIu32", "
2328 "StreamIdentifier=%.*ls",
2329 file->FileReferenceNumber,
2330 stream->StreamIdentifierLength,
2331 (int)(stream->StreamIdentifierLength / 2),
2332 stream->StreamIdentifier);
2333 return WIMLIB_ERR_UNSUPPORTED;
2334 }
2335
2336 if (use_stream(file, stream, &name, &name_nchars)) {
2337 total += ALIGN(sizeof(struct ntfs_stream) +
2338 (name_nchars + 1) * sizeof(wchar_t), 8);
2339 } else if (stream_has_identifier(stream, L"::$OBJECT_ID")) {
2340 special_streams |= NTFS_SPECIAL_STREAM_OBJECT_ID;
2341 } else if (stream_has_identifier(stream, L"::$EA")) {
2342 special_streams |= NTFS_SPECIAL_STREAM_EA;
2343 } else if (stream_has_identifier(stream, L"::$EA_INFORMATION")) {
2344 special_streams |= NTFS_SPECIAL_STREAM_EA_INFORMATION;
2345 }
2346 if (stream->NextStreamOffset == 0)
2347 break;
2348 stream = (const void *)stream + stream->NextStreamOffset;
2349 }
2350
2351 *total_length_ret = total;
2352 *special_streams_ret = special_streams;
2353 return 0;
2354 }
2355
2356 static void *
load_name_information(const FILE_LAYOUT_ENTRY * file,struct ntfs_inode * ni,void * p)2357 load_name_information(const FILE_LAYOUT_ENTRY *file, struct ntfs_inode *ni,
2358 void *p)
2359 {
2360 const FILE_LAYOUT_NAME_ENTRY *name =
2361 (const void *)file + file->FirstNameOffset;
2362 for (;;) {
2363 struct ntfs_dentry *nd = p;
2364 /* Note that a name may be just a short (DOS) name, just a long
2365 * name, or both a short name and a long name. If there is a
2366 * short name, one name should also be marked as "primary" to
2367 * indicate which long name the short name is associated with.
2368 * Also, there should be at most one short name per inode. */
2369 if (name->Flags & FILE_LAYOUT_NAME_ENTRY_DOS) {
2370 memcpy(ni->short_name,
2371 name->FileName, name->FileNameLength);
2372 ni->short_name[name->FileNameLength / 2] = L'\0';
2373 }
2374 if (name->Flags != FILE_LAYOUT_NAME_ENTRY_DOS) {
2375 ni->num_aliases++;
2376 nd->offset_from_inode = (u8 *)nd - (u8 *)ni;
2377 nd->is_primary = ((name->Flags &
2378 FILE_LAYOUT_NAME_ENTRY_PRIMARY) != 0);
2379 nd->parent_ino = name->ParentFileReferenceNumber;
2380 memcpy(nd->name, name->FileName, name->FileNameLength);
2381 nd->name[name->FileNameLength / 2] = L'\0';
2382 p += ALIGN(sizeof(struct ntfs_dentry) +
2383 name->FileNameLength + sizeof(wchar_t), 8);
2384 }
2385 if (name->NextNameOffset == 0)
2386 break;
2387 name = (const void *)name + name->NextNameOffset;
2388 }
2389 return p;
2390 }
2391
2392 static u64
load_starting_lcn(const STREAM_LAYOUT_ENTRY * stream)2393 load_starting_lcn(const STREAM_LAYOUT_ENTRY *stream)
2394 {
2395 const STREAM_EXTENT_ENTRY *entry;
2396
2397 if (stream->ExtentInformationOffset == 0)
2398 return 0;
2399
2400 entry = (const void *)stream + stream->ExtentInformationOffset;
2401
2402 if (!(entry->Flags & STREAM_EXTENT_ENTRY_AS_RETRIEVAL_POINTERS))
2403 return 0;
2404
2405 return extract_starting_lcn(&entry->ExtentInformation.RetrievalPointers);
2406 }
2407
2408 static void *
load_stream_information(const FILE_LAYOUT_ENTRY * file,struct ntfs_inode * ni,void * p)2409 load_stream_information(const FILE_LAYOUT_ENTRY *file, struct ntfs_inode *ni,
2410 void *p)
2411 {
2412 const STREAM_LAYOUT_ENTRY *stream =
2413 (const void *)file + file->FirstStreamOffset;
2414 const u32 first_stream_offset = (const u8 *)p - (const u8 *)ni;
2415 for (;;) {
2416 struct ntfs_stream *ns = p;
2417 const wchar_t *name;
2418 size_t name_nchars;
2419
2420 if (use_stream(file, stream, &name, &name_nchars)) {
2421 ni->first_stream_offset = first_stream_offset;
2422 ni->num_streams++;
2423 if (name_nchars == 0)
2424 ni->starting_lcn = load_starting_lcn(stream);
2425 ns->size = stream->EndOfFile;
2426 wmemcpy(ns->name, name, name_nchars);
2427 ns->name[name_nchars] = L'\0';
2428 p += ALIGN(sizeof(struct ntfs_stream) +
2429 (name_nchars + 1) * sizeof(wchar_t), 8);
2430 }
2431 if (stream->NextStreamOffset == 0)
2432 break;
2433 stream = (const void *)stream + stream->NextStreamOffset;
2434 }
2435 return p;
2436 }
2437
2438 /* Process the information for a file given by FSCTL_QUERY_FILE_LAYOUT. */
2439 static int
load_one_file(const FILE_LAYOUT_ENTRY * file,struct ntfs_inode_map * inode_map)2440 load_one_file(const FILE_LAYOUT_ENTRY *file, struct ntfs_inode_map *inode_map)
2441 {
2442 const FILE_LAYOUT_INFO_ENTRY *info =
2443 (const void *)file + file->ExtraInfoOffset;
2444 size_t inode_size;
2445 struct ntfs_inode *ni;
2446 size_t n;
2447 int ret;
2448 void *p;
2449 u32 special_streams = 0;
2450
2451 inode_size = ALIGN(sizeof(struct ntfs_inode), 8);
2452
2453 /* The root file should have no names, and all other files should have
2454 * at least one name. But just in case, we ignore the names of the root
2455 * file, and we ignore any non-root file with no names. */
2456 if (!NTFS_IS_ROOT_FILE(file->FileReferenceNumber)) {
2457 if (file->FirstNameOffset == 0)
2458 return 0;
2459 ret = validate_names_and_compute_total_length(file, &n);
2460 if (ret)
2461 return ret;
2462 inode_size += n;
2463 }
2464
2465 if (file_has_streams(file)) {
2466 ret = validate_streams_and_compute_total_length(file, &n,
2467 &special_streams);
2468 if (ret)
2469 return ret;
2470 inode_size += n;
2471 }
2472
2473 /* To save memory, we allocate the ntfs_dentry's and ntfs_stream's in
2474 * the same memory block as their ntfs_inode. */
2475 ni = CALLOC(1, inode_size);
2476 if (!ni)
2477 return WIMLIB_ERR_NOMEM;
2478
2479 ni->ino = file->FileReferenceNumber;
2480 ni->attributes = info->BasicInformation.FileAttributes;
2481 ni->creation_time = info->BasicInformation.CreationTime;
2482 ni->last_write_time = info->BasicInformation.LastWriteTime;
2483 ni->last_access_time = info->BasicInformation.LastAccessTime;
2484 ni->security_id = info->SecurityId;
2485 ni->special_streams = special_streams;
2486
2487 p = FIRST_DENTRY(ni);
2488
2489 if (!NTFS_IS_ROOT_FILE(file->FileReferenceNumber))
2490 p = load_name_information(file, ni, p);
2491
2492 if (file_has_streams(file))
2493 p = load_stream_information(file, ni, p);
2494
2495 wimlib_assert((u8 *)p - (u8 *)ni == inode_size);
2496
2497 ntfs_inode_map_add_inode(inode_map, ni);
2498 return 0;
2499 }
2500
2501 /*
2502 * Quickly find all files on an NTFS volume by using FSCTL_QUERY_FILE_LAYOUT to
2503 * scan the MFT. The NTFS volume is specified by the NT namespace path @path.
2504 * For each file, allocate an 'ntfs_inode' structure for each file and add it to
2505 * 'inode_map' keyed by inode number. Include NTFS special files such as
2506 * $Bitmap (they will be removed later).
2507 */
2508 static int
load_files_from_mft(const wchar_t * path,struct ntfs_inode_map * inode_map)2509 load_files_from_mft(const wchar_t *path, struct ntfs_inode_map *inode_map)
2510 {
2511 HANDLE h = NULL;
2512 QUERY_FILE_LAYOUT_INPUT in = (QUERY_FILE_LAYOUT_INPUT) {
2513 .NumberOfPairs = 0,
2514 .Flags = QUERY_FILE_LAYOUT_RESTART |
2515 QUERY_FILE_LAYOUT_INCLUDE_NAMES |
2516 QUERY_FILE_LAYOUT_INCLUDE_STREAMS |
2517 QUERY_FILE_LAYOUT_INCLUDE_EXTENTS |
2518 QUERY_FILE_LAYOUT_INCLUDE_EXTRA_INFO |
2519 QUERY_FILE_LAYOUT_INCLUDE_STREAMS_WITH_NO_CLUSTERS_ALLOCATED,
2520 .FilterType = QUERY_FILE_LAYOUT_FILTER_TYPE_NONE,
2521 };
2522 size_t outsize = 32768;
2523 QUERY_FILE_LAYOUT_OUTPUT *out = NULL;
2524 int ret;
2525 NTSTATUS status;
2526
2527 status = winnt_open(path, wcslen(path),
2528 FILE_READ_DATA | FILE_READ_ATTRIBUTES, &h);
2529 if (!NT_SUCCESS(status)) {
2530 ret = -1; /* Silently try standard recursive scan instead */
2531 goto out;
2532 }
2533
2534 for (;;) {
2535 /* Allocate a buffer for the output of the ioctl. */
2536 out = MALLOC(outsize);
2537 if (!out) {
2538 ret = WIMLIB_ERR_NOMEM;
2539 goto out;
2540 }
2541
2542 /* Execute FSCTL_QUERY_FILE_LAYOUT until it fails. */
2543 while (NT_SUCCESS(status = winnt_fsctl(h,
2544 FSCTL_QUERY_FILE_LAYOUT,
2545 &in, sizeof(in),
2546 out, outsize, NULL)))
2547 {
2548 const FILE_LAYOUT_ENTRY *file =
2549 (const void *)out + out->FirstFileOffset;
2550 for (;;) {
2551 ret = load_one_file(file, inode_map);
2552 if (ret)
2553 goto out;
2554 if (file->NextFileOffset == 0)
2555 break;
2556 file = (const void *)file + file->NextFileOffset;
2557 }
2558 in.Flags &= ~QUERY_FILE_LAYOUT_RESTART;
2559 }
2560
2561 /* Enlarge the buffer if needed. */
2562 if (status != STATUS_BUFFER_TOO_SMALL)
2563 break;
2564 FREE(out);
2565 outsize *= 2;
2566 }
2567
2568 /* Normally, FSCTL_QUERY_FILE_LAYOUT fails with STATUS_END_OF_FILE after
2569 * all files have been enumerated. */
2570 if (status != STATUS_END_OF_FILE) {
2571 if (status == STATUS_INVALID_DEVICE_REQUEST /* old OS */ ||
2572 status == STATUS_NOT_SUPPORTED /* Samba volume, WinXP */ ||
2573 status == STATUS_INVALID_PARAMETER /* not root directory */ )
2574 {
2575 /* Silently try standard recursive scan instead */
2576 ret = -1;
2577 } else {
2578 winnt_error(status,
2579 L"Error enumerating files on volume \"%ls\"",
2580 path);
2581 /* Try standard recursive scan instead */
2582 ret = WIMLIB_ERR_UNSUPPORTED;
2583 }
2584 goto out;
2585 }
2586 ret = 0;
2587 out:
2588 FREE(out);
2589 NtClose(h);
2590 return ret;
2591 }
2592
2593 /* Build the list of child dentries for each inode in @map. This is done by
2594 * iterating through each name of each inode and adding it to its parent's
2595 * children list. Note that every name should have a parent, i.e. should belong
2596 * to some directory. The root directory does not have any names. */
2597 static int
build_children_lists(struct ntfs_inode_map * map,struct ntfs_inode ** root_ret)2598 build_children_lists(struct ntfs_inode_map *map, struct ntfs_inode **root_ret)
2599 {
2600 struct ntfs_inode *ni;
2601
2602 avl_tree_for_each_in_order(ni, map->root, struct ntfs_inode, index_node)
2603 {
2604 struct ntfs_dentry *nd;
2605 u32 n;
2606
2607 if (NTFS_IS_ROOT_FILE(ni->ino)) {
2608 *root_ret = ni;
2609 continue;
2610 }
2611
2612 n = ni->num_aliases;
2613 nd = FIRST_DENTRY(ni);
2614 for (;;) {
2615 struct ntfs_inode *parent;
2616
2617 parent = ntfs_inode_map_lookup(map, nd->parent_ino);
2618 if (unlikely(!parent)) {
2619 ERROR("Parent inode 0x%016"PRIx64" of"
2620 "directory entry \"%ls\" (inode "
2621 "0x%016"PRIx64") was missing from the "
2622 "MFT listing!",
2623 nd->parent_ino, nd->name, ni->ino);
2624 return WIMLIB_ERR_UNSUPPORTED;
2625 }
2626 nd->next_child = parent->first_child;
2627 parent->first_child = nd;
2628 if (!--n)
2629 break;
2630 nd = NEXT_DENTRY(nd);
2631 }
2632 }
2633 return 0;
2634 }
2635
2636 struct security_map_node {
2637 struct avl_tree_node index_node;
2638 u32 disk_security_id;
2639 u32 wim_security_id;
2640 };
2641
2642 /* Map from disk security IDs to WIM security IDs */
2643 struct security_map {
2644 struct avl_tree_node *root;
2645 };
2646
2647 #define SECURITY_MAP_NODE(node) \
2648 avl_tree_entry((node), struct security_map_node, index_node)
2649
2650 static int
_avl_cmp_security_map_nodes(const struct avl_tree_node * node1,const struct avl_tree_node * node2)2651 _avl_cmp_security_map_nodes(const struct avl_tree_node *node1,
2652 const struct avl_tree_node *node2)
2653 {
2654 return cmp_u32(SECURITY_MAP_NODE(node1)->disk_security_id,
2655 SECURITY_MAP_NODE(node2)->disk_security_id);
2656 }
2657
2658 static s32
security_map_lookup(struct security_map * map,u32 disk_security_id)2659 security_map_lookup(struct security_map *map, u32 disk_security_id)
2660 {
2661 struct security_map_node tmp;
2662 const struct avl_tree_node *res;
2663
2664 if (disk_security_id == 0) /* No on-disk security ID; uncacheable */
2665 return -1;
2666
2667 tmp.disk_security_id = disk_security_id;
2668 res = avl_tree_lookup_node(map->root, &tmp.index_node,
2669 _avl_cmp_security_map_nodes);
2670 if (!res)
2671 return -1;
2672 return SECURITY_MAP_NODE(res)->wim_security_id;
2673 }
2674
2675 static int
security_map_insert(struct security_map * map,u32 disk_security_id,u32 wim_security_id)2676 security_map_insert(struct security_map *map, u32 disk_security_id,
2677 u32 wim_security_id)
2678 {
2679 struct security_map_node *node;
2680
2681 if (disk_security_id == 0) /* No on-disk security ID; uncacheable */
2682 return 0;
2683
2684 node = MALLOC(sizeof(*node));
2685 if (!node)
2686 return WIMLIB_ERR_NOMEM;
2687
2688 node->disk_security_id = disk_security_id;
2689 node->wim_security_id = wim_security_id;
2690 avl_tree_insert(&map->root, &node->index_node,
2691 _avl_cmp_security_map_nodes);
2692 return 0;
2693 }
2694
2695 static void
security_map_destroy(struct security_map * map)2696 security_map_destroy(struct security_map *map)
2697 {
2698 struct security_map_node *node;
2699
2700 avl_tree_for_each_in_postorder(node, map->root,
2701 struct security_map_node, index_node)
2702 FREE(node);
2703 }
2704
2705 /*
2706 * Turn our temporary NTFS structures into the final WIM structures:
2707 *
2708 * ntfs_inode => wim_inode
2709 * ntfs_dentry => wim_dentry
2710 * ntfs_stream => wim_inode_stream
2711 *
2712 * This also handles things such as exclusions and issuing progress messages.
2713 * It's similar to winnt_build_dentry_tree_recursive(), but this is much faster
2714 * because almost all information we need is already loaded in memory in the
2715 * ntfs_* structures. However, in some cases we still fall back to
2716 * winnt_build_dentry_tree_recursive() and/or opening the file.
2717 */
2718 static int
generate_wim_structures_recursive(struct wim_dentry ** root_ret,const wchar_t * filename,bool is_primary_name,struct ntfs_inode * ni,struct winnt_scan_ctx * ctx,struct ntfs_inode_map * inode_map,struct security_map * security_map)2719 generate_wim_structures_recursive(struct wim_dentry **root_ret,
2720 const wchar_t *filename, bool is_primary_name,
2721 struct ntfs_inode *ni,
2722 struct winnt_scan_ctx *ctx,
2723 struct ntfs_inode_map *inode_map,
2724 struct security_map *security_map)
2725 {
2726 int ret = 0;
2727 struct wim_dentry *root = NULL;
2728 struct wim_inode *inode = NULL;
2729 const struct ntfs_stream *ns;
2730
2731 /* Completely ignore NTFS special files. */
2732 if (NTFS_IS_SPECIAL_FILE(ni->ino))
2733 goto out;
2734
2735 /* Fall back to a recursive scan for unhandled cases. Reparse points,
2736 * in particular, can't be properly handled here because a commonly used
2737 * filter driver (WOF) hides reparse points from regular filesystem APIs
2738 * but not from FSCTL_QUERY_FILE_LAYOUT. */
2739 if (ni->attributes & (FILE_ATTRIBUTE_REPARSE_POINT |
2740 FILE_ATTRIBUTE_ENCRYPTED) ||
2741 ni->special_streams != 0)
2742 {
2743 ret = winnt_build_dentry_tree_recursive(&root,
2744 NULL,
2745 ctx->params->cur_path,
2746 ctx->params->cur_path_nchars,
2747 filename,
2748 ctx);
2749 goto out;
2750 }
2751
2752 /* Test for exclusion based on path. */
2753 ret = try_exclude(ctx->params);
2754 if (unlikely(ret < 0)) /* Excluded? */
2755 goto out_progress;
2756 if (unlikely(ret > 0)) /* Error? */
2757 goto out;
2758
2759 /* Create the WIM dentry and possibly a new WIM inode */
2760 ret = inode_table_new_dentry(ctx->params->inode_table, filename,
2761 ni->ino, ctx->params->capture_root_dev,
2762 false, &root);
2763 if (ret)
2764 goto out;
2765
2766 inode = root->d_inode;
2767
2768 /* Set the short name if needed. */
2769 if (is_primary_name && *ni->short_name) {
2770 size_t nbytes = wcslen(ni->short_name) * sizeof(wchar_t);
2771 root->d_short_name = memdup(ni->short_name,
2772 nbytes + sizeof(wchar_t));
2773 if (!root->d_short_name) {
2774 ret = WIMLIB_ERR_NOMEM;
2775 goto out;
2776 }
2777 root->d_short_name_nbytes = nbytes;
2778 }
2779
2780 if (inode->i_nlink > 1) { /* Already seen this inode? */
2781 ret = 0;
2782 goto out_progress;
2783 }
2784
2785 /* The file attributes and timestamps were cached from the MFT. */
2786 inode->i_attributes = ni->attributes;
2787 inode->i_creation_time = ni->creation_time;
2788 inode->i_last_write_time = ni->last_write_time;
2789 inode->i_last_access_time = ni->last_access_time;
2790
2791 /* Set the security descriptor if needed. */
2792 if (!(ctx->params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)) {
2793 /* Look up the WIM security ID that corresponds to the on-disk
2794 * security ID. */
2795 s32 wim_security_id =
2796 security_map_lookup(security_map, ni->security_id);
2797 if (likely(wim_security_id >= 0)) {
2798 /* The mapping for this security ID is already cached.*/
2799 inode->i_security_id = wim_security_id;
2800 } else {
2801 HANDLE h;
2802 NTSTATUS status;
2803
2804 /* Create a mapping for this security ID and insert it
2805 * into the security map. */
2806
2807 status = winnt_open(ctx->params->cur_path,
2808 ctx->params->cur_path_nchars,
2809 READ_CONTROL |
2810 ACCESS_SYSTEM_SECURITY, &h);
2811 if (!NT_SUCCESS(status)) {
2812 winnt_error(status, L"Can't open \"%ls\" to "
2813 "read security descriptor",
2814 printable_path(ctx));
2815 ret = WIMLIB_ERR_OPEN;
2816 goto out;
2817 }
2818 ret = winnt_load_security_descriptor(h, inode, ctx);
2819 NtClose(h);
2820 if (ret)
2821 goto out;
2822
2823 ret = security_map_insert(security_map, ni->security_id,
2824 inode->i_security_id);
2825 if (ret)
2826 goto out;
2827 }
2828 }
2829
2830 /* Add data streams based on the cached information from the MFT. */
2831 ns = FIRST_STREAM(ni);
2832 for (u32 i = 0; i < ni->num_streams; i++) {
2833 struct windows_file *windows_file;
2834
2835 /* Reference the stream by path if it's a named data stream, or
2836 * if the volume doesn't support "open by file ID", or if the
2837 * application hasn't explicitly opted in to "open by file ID".
2838 * Otherwise, only save the inode number (file ID). */
2839 if (*ns->name ||
2840 !(ctx->vol_flags & FILE_SUPPORTS_OPEN_BY_FILE_ID) ||
2841 !(ctx->params->add_flags & WIMLIB_ADD_FLAG_FILE_PATHS_UNNEEDED))
2842 {
2843 windows_file = alloc_windows_file(ctx->params->cur_path,
2844 ctx->params->cur_path_nchars,
2845 ns->name,
2846 wcslen(ns->name),
2847 ctx->snapshot,
2848 false);
2849 } else {
2850 windows_file = alloc_windows_file_for_file_id(ni->ino,
2851 ctx->params->cur_path,
2852 ctx->params->root_path_nchars,
2853 ctx->snapshot);
2854 }
2855
2856 ret = add_stream(inode, windows_file, ns->size,
2857 STREAM_TYPE_DATA, ns->name,
2858 ctx->params->unhashed_blobs);
2859 if (ret)
2860 goto out;
2861 ns = NEXT_STREAM(ns);
2862 }
2863
2864 set_sort_key(inode, ni->starting_lcn);
2865
2866 /* If processing a directory, then recurse to its children. In this
2867 * version there is no need to go to disk, as we already have the list
2868 * of children cached from the MFT. */
2869 if (inode_is_directory(inode)) {
2870 const struct ntfs_dentry *nd = ni->first_child;
2871
2872 while (nd != NULL) {
2873 size_t orig_path_nchars;
2874 struct wim_dentry *child;
2875 const struct ntfs_dentry *next = nd->next_child;
2876
2877 ret = WIMLIB_ERR_NOMEM;
2878 if (!pathbuf_append_name(ctx->params, nd->name,
2879 wcslen(nd->name),
2880 &orig_path_nchars))
2881 goto out;
2882
2883 ret = generate_wim_structures_recursive(
2884 &child,
2885 nd->name,
2886 nd->is_primary,
2887 (void *)nd - nd->offset_from_inode,
2888 ctx,
2889 inode_map,
2890 security_map);
2891
2892 pathbuf_truncate(ctx->params, orig_path_nchars);
2893
2894 if (ret)
2895 goto out;
2896
2897 attach_scanned_tree(root, child, ctx->params->blob_table);
2898 nd = next;
2899 }
2900 }
2901
2902 out_progress:
2903 if (likely(root))
2904 ret = do_scan_progress(ctx->params, WIMLIB_SCAN_DENTRY_OK, inode);
2905 else
2906 ret = do_scan_progress(ctx->params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL);
2907 out:
2908 if (--ni->num_aliases == 0) {
2909 /* Memory usage optimization: when we don't need the ntfs_inode
2910 * (and its names and streams) anymore, free it. */
2911 ntfs_inode_map_remove(inode_map, ni);
2912 }
2913 if (unlikely(ret)) {
2914 free_dentry_tree(root, ctx->params->blob_table);
2915 root = NULL;
2916 }
2917 *root_ret = root;
2918 return ret;
2919 }
2920
2921 static int
winnt_build_dentry_tree_fast(struct wim_dentry ** root_ret,struct winnt_scan_ctx * ctx)2922 winnt_build_dentry_tree_fast(struct wim_dentry **root_ret,
2923 struct winnt_scan_ctx *ctx)
2924 {
2925 struct ntfs_inode_map inode_map = { .root = NULL };
2926 struct security_map security_map = { .root = NULL };
2927 struct ntfs_inode *root = NULL;
2928 wchar_t *path = ctx->params->cur_path;
2929 size_t path_nchars = ctx->params->cur_path_nchars;
2930 bool adjust_path;
2931 int ret;
2932
2933 adjust_path = (path[path_nchars - 1] == L'\\');
2934 if (adjust_path)
2935 path[path_nchars - 1] = L'\0';
2936
2937 ret = load_files_from_mft(path, &inode_map);
2938
2939 if (adjust_path)
2940 path[path_nchars - 1] = L'\\';
2941
2942 if (ret)
2943 goto out;
2944
2945 ret = build_children_lists(&inode_map, &root);
2946 if (ret)
2947 goto out;
2948
2949 if (!root) {
2950 ERROR("The MFT listing for volume \"%ls\" did not include a "
2951 "root directory!", path);
2952 ret = WIMLIB_ERR_UNSUPPORTED;
2953 goto out;
2954 }
2955
2956 root->num_aliases = 1;
2957
2958 ret = generate_wim_structures_recursive(root_ret, L"", false, root, ctx,
2959 &inode_map, &security_map);
2960 out:
2961 ntfs_inode_map_destroy(&inode_map);
2962 security_map_destroy(&security_map);
2963 return ret;
2964 }
2965
2966 #endif /* ENABLE_FAST_MFT_SCAN */
2967
2968 /*----------------------------------------------------------------------------*
2969 * Entry point for directory tree scans on Windows *
2970 *----------------------------------------------------------------------------*/
2971
2972 int
win32_build_dentry_tree(struct wim_dentry ** root_ret,const wchar_t * root_disk_path,struct scan_params * params)2973 win32_build_dentry_tree(struct wim_dentry **root_ret,
2974 const wchar_t *root_disk_path,
2975 struct scan_params *params)
2976 {
2977 struct winnt_scan_ctx ctx = { .params = params };
2978 UNICODE_STRING ntpath;
2979 HANDLE h = NULL;
2980 NTSTATUS status;
2981 int ret;
2982
2983 if (params->add_flags & WIMLIB_ADD_FLAG_SNAPSHOT)
2984 ret = vss_create_snapshot(root_disk_path, &ntpath, &ctx.snapshot);
2985 else
2986 ret = win32_path_to_nt_path(root_disk_path, &ntpath);
2987
2988 if (ret)
2989 goto out;
2990
2991 if (ntpath.Length < 4 * sizeof(wchar_t) ||
2992 wmemcmp(ntpath.Buffer, L"\\??\\", 4))
2993 {
2994 ERROR("\"%ls\": unrecognized path format", root_disk_path);
2995 ret = WIMLIB_ERR_INVALID_PARAM;
2996 } else {
2997 ret = pathbuf_init(params, ntpath.Buffer);
2998 }
2999 HeapFree(GetProcessHeap(), 0, ntpath.Buffer);
3000 if (ret)
3001 goto out;
3002
3003 status = winnt_open(params->cur_path, params->cur_path_nchars,
3004 FILE_READ_ATTRIBUTES, &h);
3005 if (!NT_SUCCESS(status)) {
3006 winnt_error(status, L"Can't open \"%ls\"", root_disk_path);
3007 if (status == STATUS_FVE_LOCKED_VOLUME)
3008 ret = WIMLIB_ERR_FVE_LOCKED_VOLUME;
3009 else
3010 ret = WIMLIB_ERR_OPEN;
3011 goto out;
3012 }
3013
3014 get_volume_information(h, &ctx);
3015
3016 NtClose(h);
3017
3018 #ifdef ENABLE_FAST_MFT_SCAN
3019 if (ctx.is_ntfs && !_wgetenv(L"WIMLIB_DISABLE_QUERY_FILE_LAYOUT")) {
3020 ret = winnt_build_dentry_tree_fast(root_ret, &ctx);
3021 if (ret >= 0 && ret != WIMLIB_ERR_UNSUPPORTED)
3022 goto out;
3023 if (ret >= 0) {
3024 WARNING("A problem occurred during the fast MFT scan.\n"
3025 " Falling back to the standard "
3026 "recursive directory tree scan.");
3027 }
3028 }
3029 #endif
3030 ret = winnt_build_dentry_tree_recursive(root_ret, NULL,
3031 params->cur_path,
3032 params->cur_path_nchars,
3033 L"", &ctx);
3034 out:
3035 vss_put_snapshot(ctx.snapshot);
3036 if (ret == 0)
3037 winnt_do_scan_warnings(root_disk_path, &ctx);
3038 return ret;
3039 }
3040
3041 #endif /* __WIN32__ */
3042