854428e4dfd20266780858558bed72b608c9f465
[wimlib] / src / win32_capture.c
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, 2014, 2015 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/capture.h"
35 #include "wimlib/dentry.h"
36 #include "wimlib/encoding.h"
37 #include "wimlib/endianness.h"
38 #include "wimlib/error.h"
39 #include "wimlib/paths.h"
40 #include "wimlib/reparse.h"
41
42 struct winnt_scan_stats {
43         unsigned long num_get_sd_access_denied;
44         unsigned long num_get_sacl_priv_notheld;
45 };
46
47 static inline const wchar_t *
48 printable_path(const wchar_t *full_path)
49 {
50         /* Skip over \\?\ or \??\  */
51         return full_path + 4;
52 }
53
54 /*
55  * If cur_dir is not NULL, open an existing file relative to the already-open
56  * directory cur_dir.
57  *
58  * Otherwise, open the file specified by @path, which must be a Windows NT
59  * namespace path.
60  */
61 static NTSTATUS
62 winnt_openat(HANDLE cur_dir, const wchar_t *path, size_t path_nchars,
63              ACCESS_MASK perms, HANDLE *h_ret)
64 {
65         UNICODE_STRING name;
66         OBJECT_ATTRIBUTES attr;
67         IO_STATUS_BLOCK iosb;
68         NTSTATUS status;
69
70         name.Length = path_nchars * sizeof(wchar_t);
71         name.MaximumLength = name.Length + sizeof(wchar_t);
72         name.Buffer = (wchar_t *)path;
73
74         attr.Length = sizeof(attr);
75         attr.RootDirectory = cur_dir;
76         attr.ObjectName = &name;
77         attr.Attributes = 0;
78         attr.SecurityDescriptor = NULL;
79         attr.SecurityQualityOfService = NULL;
80
81 retry:
82         status = (*func_NtOpenFile)(h_ret, perms, &attr, &iosb,
83                                     FILE_SHARE_VALID_FLAGS,
84                                     FILE_OPEN_REPARSE_POINT |
85                                             FILE_OPEN_FOR_BACKUP_INTENT |
86                                             FILE_SYNCHRONOUS_IO_NONALERT |
87                                             FILE_SEQUENTIAL_ONLY);
88         if (!NT_SUCCESS(status)) {
89                 /* Try requesting fewer permissions  */
90                 if (status == STATUS_ACCESS_DENIED ||
91                     status == STATUS_PRIVILEGE_NOT_HELD) {
92                         if (perms & ACCESS_SYSTEM_SECURITY) {
93                                 perms &= ~ACCESS_SYSTEM_SECURITY;
94                                 goto retry;
95                         }
96                         if (perms & READ_CONTROL) {
97                                 perms &= ~READ_CONTROL;
98                                 goto retry;
99                         }
100                 }
101         }
102         return status;
103 }
104
105 /* Read the first @size bytes from the file, or named data stream of a file,
106  * described by @blob.  */
107 int
108 read_winnt_stream_prefix(const struct blob_descriptor *blob, u64 size,
109                          const struct read_blob_callbacks *cbs)
110 {
111         const wchar_t *path;
112         HANDLE h;
113         NTSTATUS status;
114         u8 buf[BUFFER_SIZE];
115         u64 bytes_remaining;
116         int ret;
117
118         /* This is an NT namespace path.  */
119         path = blob->file_on_disk;
120
121         status = winnt_openat(NULL, path, wcslen(path),
122                               FILE_READ_DATA | SYNCHRONIZE, &h);
123         if (!NT_SUCCESS(status)) {
124                 winnt_error(status, L"\"%ls\": Can't open for reading",
125                             printable_path(path));
126                 return WIMLIB_ERR_OPEN;
127         }
128
129         ret = 0;
130         bytes_remaining = size;
131         while (bytes_remaining) {
132                 IO_STATUS_BLOCK iosb;
133                 ULONG count;
134                 ULONG bytes_read;
135
136                 count = min(sizeof(buf), bytes_remaining);
137
138                 status = (*func_NtReadFile)(h, NULL, NULL, NULL,
139                                             &iosb, buf, count, NULL, NULL);
140                 if (!NT_SUCCESS(status)) {
141                         winnt_error(status, L"\"%ls\": Error reading data",
142                                     printable_path(path));
143                         ret = WIMLIB_ERR_READ;
144                         break;
145                 }
146
147                 bytes_read = iosb.Information;
148
149                 bytes_remaining -= bytes_read;
150                 ret = call_consume_chunk(buf, bytes_read, cbs);
151                 if (ret)
152                         break;
153         }
154         (*func_NtClose)(h);
155         return ret;
156 }
157
158 struct win32_encrypted_read_ctx {
159         const struct read_blob_callbacks *cbs;
160         int wimlib_err_code;
161         u64 bytes_remaining;
162 };
163
164 static DWORD WINAPI
165 win32_encrypted_export_cb(unsigned char *data, void *_ctx, unsigned long len)
166 {
167         struct win32_encrypted_read_ctx *ctx = _ctx;
168         int ret;
169         size_t bytes_to_consume = min(len, ctx->bytes_remaining);
170
171         if (bytes_to_consume == 0)
172                 return ERROR_SUCCESS;
173
174         ret = call_consume_chunk(data, bytes_to_consume, ctx->cbs);
175         if (ret) {
176                 ctx->wimlib_err_code = ret;
177                 /* It doesn't matter what error code is returned here, as long
178                  * as it isn't ERROR_SUCCESS.  */
179                 return ERROR_READ_FAULT;
180         }
181         ctx->bytes_remaining -= bytes_to_consume;
182         return ERROR_SUCCESS;
183 }
184
185 int
186 read_win32_encrypted_file_prefix(const struct blob_descriptor *blob,
187                                  u64 size,
188                                  const struct read_blob_callbacks *cbs)
189 {
190         struct win32_encrypted_read_ctx export_ctx;
191         DWORD err;
192         void *file_ctx;
193         int ret;
194         DWORD flags = 0;
195
196         if (blob->file_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
197                 flags |= CREATE_FOR_DIR;
198
199         export_ctx.cbs = cbs;
200         export_ctx.wimlib_err_code = 0;
201         export_ctx.bytes_remaining = size;
202
203         err = OpenEncryptedFileRaw(blob->file_on_disk, flags, &file_ctx);
204         if (err != ERROR_SUCCESS) {
205                 win32_error(err,
206                             L"Failed to open encrypted file \"%ls\" for raw read",
207                             printable_path(blob->file_on_disk));
208                 return WIMLIB_ERR_OPEN;
209         }
210         err = ReadEncryptedFileRaw(win32_encrypted_export_cb,
211                                    &export_ctx, file_ctx);
212         if (err != ERROR_SUCCESS) {
213                 ret = export_ctx.wimlib_err_code;
214                 if (ret == 0) {
215                         win32_error(err,
216                                     L"Failed to read encrypted file \"%ls\"",
217                                     printable_path(blob->file_on_disk));
218                         ret = WIMLIB_ERR_READ;
219                 }
220         } else if (export_ctx.bytes_remaining != 0) {
221                 ERROR("Only could read %"PRIu64" of %"PRIu64" bytes from "
222                       "encrypted file \"%ls\"",
223                       size - export_ctx.bytes_remaining, size,
224                       printable_path(blob->file_on_disk));
225                 ret = WIMLIB_ERR_READ;
226         } else {
227                 ret = 0;
228         }
229         CloseEncryptedFileRaw(file_ctx);
230         return ret;
231 }
232
233 /*
234  * Load the short name of a file into a WIM dentry.
235  */
236 static noinline_for_stack NTSTATUS
237 winnt_get_short_name(HANDLE h, struct wim_dentry *dentry)
238 {
239         /* It's not any harder to just make the NtQueryInformationFile() system
240          * call ourselves, and it saves a dumb call to FindFirstFile() which of
241          * course has to create its own handle.  */
242         NTSTATUS status;
243         IO_STATUS_BLOCK iosb;
244         u8 buf[128] _aligned_attribute(8);
245         const FILE_NAME_INFORMATION *info;
246
247         status = (*func_NtQueryInformationFile)(h, &iosb, buf, sizeof(buf),
248                                                 FileAlternateNameInformation);
249         info = (const FILE_NAME_INFORMATION *)buf;
250         if (NT_SUCCESS(status) && info->FileNameLength != 0) {
251                 dentry->d_short_name = utf16le_dupz(info->FileName,
252                                                     info->FileNameLength);
253                 if (!dentry->d_short_name)
254                         return STATUS_NO_MEMORY;
255                 dentry->d_short_name_nbytes = info->FileNameLength;
256         }
257         return status;
258 }
259
260 /*
261  * Load the security descriptor of a file into the corresponding inode and the
262  * WIM image's security descriptor set.
263  */
264 static noinline_for_stack NTSTATUS
265 winnt_get_security_descriptor(HANDLE h, struct wim_inode *inode,
266                               struct wim_sd_set *sd_set,
267                               struct winnt_scan_stats *stats, int add_flags)
268 {
269         SECURITY_INFORMATION requestedInformation;
270         u8 _buf[4096] _aligned_attribute(8);
271         u8 *buf;
272         ULONG bufsize;
273         ULONG len_needed;
274         NTSTATUS status;
275
276         /*
277          * LABEL_SECURITY_INFORMATION is needed on Windows Vista and 7 because
278          * Microsoft decided to add mandatory integrity labels to the SACL but
279          * not have them returned by SACL_SECURITY_INFORMATION.
280          *
281          * BACKUP_SECURITY_INFORMATION is needed on Windows 8 because Microsoft
282          * decided to add even more stuff to the SACL and still not have it
283          * returned by SACL_SECURITY_INFORMATION; but they did remember that
284          * backup applications exist and simply want to read the stupid thing
285          * once and for all, so they added a flag to read the entire security
286          * descriptor.
287          *
288          * Older versions of Windows tolerate these new flags being passed in.
289          */
290         requestedInformation = OWNER_SECURITY_INFORMATION |
291                                GROUP_SECURITY_INFORMATION |
292                                DACL_SECURITY_INFORMATION |
293                                SACL_SECURITY_INFORMATION |
294                                LABEL_SECURITY_INFORMATION |
295                                BACKUP_SECURITY_INFORMATION;
296
297         buf = _buf;
298         bufsize = sizeof(_buf);
299
300         /*
301          * We need the file's security descriptor in
302          * SECURITY_DESCRIPTOR_RELATIVE format, and we currently have a handle
303          * opened with as many relevant permissions as possible.  At this point,
304          * on Windows there are a number of options for reading a file's
305          * security descriptor:
306          *
307          * GetFileSecurity():  This takes in a path and returns the
308          * SECURITY_DESCRIPTOR_RELATIVE.  Problem: this uses an internal handle,
309          * not ours, and the handle created internally doesn't specify
310          * FILE_FLAG_BACKUP_SEMANTICS.  Therefore there can be access denied
311          * errors on some files and directories, even when running as the
312          * Administrator.
313          *
314          * GetSecurityInfo():  This takes in a handle and returns the security
315          * descriptor split into a bunch of different parts.  This should work,
316          * but it's dumb because we have to put the security descriptor back
317          * together again.
318          *
319          * BackupRead():  This can read the security descriptor, but this is a
320          * difficult-to-use API, probably only works as the Administrator, and
321          * the format of the returned data is not well documented.
322          *
323          * NtQuerySecurityObject():  This is exactly what we need, as it takes
324          * in a handle and returns the security descriptor in
325          * SECURITY_DESCRIPTOR_RELATIVE format.  Only problem is that it's a
326          * ntdll function and therefore not officially part of the Win32 API.
327          * Oh well.
328          */
329         while (!(NT_SUCCESS(status = (*func_NtQuerySecurityObject)(h,
330                                                                    requestedInformation,
331                                                                    (PSECURITY_DESCRIPTOR)buf,
332                                                                    bufsize,
333                                                                    &len_needed))))
334         {
335                 switch (status) {
336                 case STATUS_BUFFER_TOO_SMALL:
337                         wimlib_assert(buf == _buf);
338                         buf = MALLOC(len_needed);
339                         if (!buf)
340                                 return STATUS_NO_MEMORY;
341                         bufsize = len_needed;
342                         break;
343                 case STATUS_PRIVILEGE_NOT_HELD:
344                 case STATUS_ACCESS_DENIED:
345                         if (add_flags & WIMLIB_ADD_FLAG_STRICT_ACLS) {
346                 default:
347                                 /* Permission denied in STRICT_ACLS mode, or
348                                  * unknown error.  */
349                                 goto out_free_buf;
350                         }
351                         if (requestedInformation & SACL_SECURITY_INFORMATION) {
352                                 /* Try again without the SACL.  */
353                                 stats->num_get_sacl_priv_notheld++;
354                                 requestedInformation &= ~(SACL_SECURITY_INFORMATION |
355                                                           LABEL_SECURITY_INFORMATION |
356                                                           BACKUP_SECURITY_INFORMATION);
357                                 break;
358                         }
359                         /* Fake success (useful when capturing as
360                          * non-Administrator).  */
361                         stats->num_get_sd_access_denied++;
362                         status = STATUS_SUCCESS;
363                         goto out_free_buf;
364                 }
365         }
366
367         /* Add the security descriptor to the WIM image, and save its ID in
368          * file's inode.  */
369         inode->i_security_id = sd_set_add_sd(sd_set, buf, len_needed);
370         if (unlikely(inode->i_security_id < 0))
371                 status = STATUS_NO_MEMORY;
372 out_free_buf:
373         if (unlikely(buf != _buf))
374                 FREE(buf);
375         return status;
376 }
377
378 static int
379 winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret,
380                                   HANDLE cur_dir,
381                                   wchar_t *full_path,
382                                   size_t full_path_nchars,
383                                   const wchar_t *filename,
384                                   size_t filename_nchars,
385                                   struct capture_params *params,
386                                   struct winnt_scan_stats *stats,
387                                   u32 vol_flags);
388
389 static int
390 winnt_recurse_directory(HANDLE h,
391                         wchar_t *full_path,
392                         size_t full_path_nchars,
393                         struct wim_dentry *parent,
394                         struct capture_params *params,
395                         struct winnt_scan_stats *stats,
396                         u32 vol_flags)
397 {
398         void *buf;
399         const size_t bufsize = 8192;
400         IO_STATUS_BLOCK iosb;
401         NTSTATUS status;
402         int ret;
403
404         buf = MALLOC(bufsize);
405         if (!buf)
406                 return WIMLIB_ERR_NOMEM;
407
408         /* Using NtQueryDirectoryFile() we can re-use the same open handle,
409          * which we opened with FILE_FLAG_BACKUP_SEMANTICS.  */
410
411         while (NT_SUCCESS(status = (*func_NtQueryDirectoryFile)(h, NULL, NULL, NULL,
412                                                                 &iosb, buf, bufsize,
413                                                                 FileNamesInformation,
414                                                                 FALSE, NULL, FALSE)))
415         {
416                 const FILE_NAMES_INFORMATION *info = buf;
417                 for (;;) {
418                         if (!should_ignore_filename(info->FileName,
419                                                     info->FileNameLength / 2))
420                         {
421                                 wchar_t *p;
422                                 wchar_t *filename;
423                                 struct wim_dentry *child;
424
425                                 p = full_path + full_path_nchars;
426                                 /* Only add a backslash if we don't already have
427                                  * one.  This prevents a duplicate backslash
428                                  * from being added when the path to the capture
429                                  * dir had a trailing backslash.  */
430                                 if (*(p - 1) != L'\\')
431                                         *p++ = L'\\';
432                                 filename = p;
433                                 p = wmempcpy(filename, info->FileName,
434                                              info->FileNameLength / 2);
435                                 *p = '\0';
436
437                                 ret = winnt_build_dentry_tree_recursive(
438                                                         &child,
439                                                         h,
440                                                         full_path,
441                                                         p - full_path,
442                                                         filename,
443                                                         info->FileNameLength / 2,
444                                                         params,
445                                                         stats,
446                                                         vol_flags);
447
448                                 full_path[full_path_nchars] = L'\0';
449
450                                 if (ret)
451                                         goto out_free_buf;
452                                 if (child)
453                                         dentry_add_child(parent, child);
454                         }
455                         if (info->NextEntryOffset == 0)
456                                 break;
457                         info = (const FILE_NAMES_INFORMATION *)
458                                         ((const u8 *)info + info->NextEntryOffset);
459                 }
460         }
461
462         if (unlikely(status != STATUS_NO_MORE_FILES)) {
463                 winnt_error(status, L"\"%ls\": Can't read directory",
464                             printable_path(full_path));
465                 ret = WIMLIB_ERR_READ;
466         }
467 out_free_buf:
468         FREE(buf);
469         return ret;
470 }
471
472 /* Reparse point fixup status code  */
473 #define RP_FIXED        (-1)
474
475 static bool
476 file_has_ino_and_dev(HANDLE h, u64 ino, u64 dev)
477 {
478         NTSTATUS status;
479         IO_STATUS_BLOCK iosb;
480         FILE_INTERNAL_INFORMATION int_info;
481         FILE_FS_VOLUME_INFORMATION vol_info;
482
483         status = (*func_NtQueryInformationFile)(h, &iosb,
484                                                 &int_info, sizeof(int_info),
485                                                 FileInternalInformation);
486         if (!NT_SUCCESS(status))
487                 return false;
488
489         if (int_info.IndexNumber.QuadPart != ino)
490                 return false;
491
492         status = (*func_NtQueryVolumeInformationFile)(h, &iosb,
493                                                       &vol_info, sizeof(vol_info),
494                                                       FileFsVolumeInformation);
495         if (!(NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW))
496                 return false;
497
498         if (iosb.Information <
499              offsetof(FILE_FS_VOLUME_INFORMATION, VolumeSerialNumber) +
500              sizeof(vol_info.VolumeSerialNumber))
501                 return false;
502
503         return (vol_info.VolumeSerialNumber == dev);
504 }
505
506 /*
507  * This is the Windows equivalent of unix_relativize_link_target(); see there
508  * for general details.  This version works with an "absolute" Windows link
509  * target, specified from the root of the Windows kernel object namespace.  Note
510  * that we have to open directories with a trailing slash when present because
511  * \??\E: opens the E: device itself and not the filesystem root directory.
512  */
513 static const wchar_t *
514 winnt_relativize_link_target(const wchar_t *target, size_t target_nbytes,
515                              u64 ino, u64 dev)
516 {
517         UNICODE_STRING name;
518         OBJECT_ATTRIBUTES attr;
519         IO_STATUS_BLOCK iosb;
520         NTSTATUS status;
521         const wchar_t *target_end;
522         const wchar_t *p;
523
524         target_end = target + (target_nbytes / sizeof(wchar_t));
525
526         /* Empty path??? */
527         if (target_end == target)
528                 return target;
529
530         /* No leading slash???  */
531         if (target[0] != L'\\')
532                 return target;
533
534         /* UNC path???  */
535         if ((target_end - target) >= 2 &&
536             target[0] == L'\\' && target[1] == L'\\')
537                 return target;
538
539         attr.Length = sizeof(attr);
540         attr.RootDirectory = NULL;
541         attr.ObjectName = &name;
542         attr.Attributes = 0;
543         attr.SecurityDescriptor = NULL;
544         attr.SecurityQualityOfService = NULL;
545
546         name.Buffer = (wchar_t *)target;
547         name.Length = 0;
548         p = target;
549         do {
550                 HANDLE h;
551                 const wchar_t *orig_p = p;
552
553                 /* Skip non-backslashes  */
554                 while (p != target_end && *p != L'\\')
555                         p++;
556
557                 /* Skip backslashes  */
558                 while (p != target_end && *p == L'\\')
559                         p++;
560
561                 /* Append path component  */
562                 name.Length += (p - orig_p) * sizeof(wchar_t);
563                 name.MaximumLength = name.Length;
564
565                 /* Try opening the file  */
566                 status = (*func_NtOpenFile) (&h,
567                                              FILE_READ_ATTRIBUTES | FILE_TRAVERSE,
568                                              &attr,
569                                              &iosb,
570                                              FILE_SHARE_VALID_FLAGS,
571                                              FILE_OPEN_FOR_BACKUP_INTENT);
572
573                 if (NT_SUCCESS(status)) {
574                         /* Reset root directory  */
575                         if (attr.RootDirectory)
576                                 (*func_NtClose)(attr.RootDirectory);
577                         attr.RootDirectory = h;
578                         name.Buffer = (wchar_t *)p;
579                         name.Length = 0;
580
581                         if (file_has_ino_and_dev(h, ino, dev))
582                                 goto out_close_root_dir;
583                 }
584         } while (p != target_end);
585
586         p = target;
587
588 out_close_root_dir:
589         if (attr.RootDirectory)
590                 (*func_NtClose)(attr.RootDirectory);
591         while (p > target && *(p - 1) == L'\\')
592                 p--;
593         return p;
594 }
595
596 static int
597 winnt_rpfix_progress(struct capture_params *params, const wchar_t *path,
598                      const struct link_reparse_point *link, int scan_status)
599 {
600         size_t print_name_nchars = link->print_name_nbytes / sizeof(wchar_t);
601         wchar_t print_name0[print_name_nchars + 1];
602
603         wmemcpy(print_name0, link->print_name, print_name_nchars);
604         print_name0[print_name_nchars] = L'\0';
605
606         params->progress.scan.cur_path = printable_path(path);
607         params->progress.scan.symlink_target = print_name0;
608         return do_capture_progress(params, scan_status, NULL);
609 }
610
611 static int
612 winnt_try_rpfix(struct reparse_buffer_disk *rpbuf, u16 *rpbuflen_p,
613                 const wchar_t *path, struct capture_params *params)
614 {
615         struct link_reparse_point link;
616         const wchar_t *rel_target;
617         int ret;
618
619         if (parse_link_reparse_point(rpbuf, *rpbuflen_p, &link)) {
620                 /* Couldn't understand the reparse data; don't do the fixup.  */
621                 return 0;
622         }
623
624         /*
625          * Don't do reparse point fixups on relative symbolic links.
626          *
627          * On Windows, a relative symbolic link is supposed to be identifiable
628          * by having reparse tag WIM_IO_REPARSE_TAG_SYMLINK and flags
629          * SYMBOLIC_LINK_RELATIVE.  We will use this information, although this
630          * may not always do what the user expects, since drive-relative
631          * symbolic links such as "\Users\Public" have SYMBOLIC_LINK_RELATIVE
632          * set, in addition to truely relative symbolic links such as "Users" or
633          * "Users\Public".  However, WIMGAPI (as of Windows 8.1) has this same
634          * behavior.
635          *
636          * Otherwise, as far as I can tell, the targets of symbolic links that
637          * are NOT relative, as well as junctions (note: a mountpoint is the
638          * sames thing as a junction), must be NT namespace paths, for example:
639          *
640          *     - \??\e:\Users\Public
641          *     - \DosDevices\e:\Users\Public
642          *     - \Device\HardDiskVolume4\Users\Public
643          *     - \??\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public
644          *     - \DosDevices\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public
645          */
646         if (link_is_relative_symlink(&link))
647                 return 0;
648
649         rel_target = winnt_relativize_link_target(link.substitute_name,
650                                                   link.substitute_name_nbytes,
651                                                   params->capture_root_ino,
652                                                   params->capture_root_dev);
653
654         if (rel_target == link.substitute_name) {
655                 /* Target points outside of the tree being captured or had an
656                  * unrecognized path format.  Don't adjust it.  */
657                 return winnt_rpfix_progress(params, path, &link,
658                                             WIMLIB_SCAN_DENTRY_NOT_FIXED_SYMLINK);
659         }
660
661         /* We have an absolute target pointing within the directory being
662          * captured. @rel_target is the suffix of the link target that is the
663          * part relative to the directory being captured.
664          *
665          * We will cut off the prefix before this part (which is the path to the
666          * directory being captured) and add a dummy prefix.  Since the process
667          * will need to be reversed when applying the image, it doesn't matter
668          * what exactly the prefix is, as long as it looks like an absolute
669          * path.  */
670
671         static const wchar_t prefix[6] = L"\\??\\X:";
672         static const size_t num_unprintable_chars = 4;
673
674         size_t rel_target_nbytes =
675                 link.substitute_name_nbytes - ((const u8 *)rel_target -
676                                                (const u8 *)link.substitute_name);
677
678         wchar_t tmp[(sizeof(prefix) + rel_target_nbytes) / sizeof(wchar_t)];
679
680         memcpy(tmp, prefix, sizeof(prefix));
681         memcpy(tmp + ARRAY_LEN(prefix), rel_target, rel_target_nbytes);
682
683         link.substitute_name = tmp;
684         link.substitute_name_nbytes = sizeof(tmp);
685
686         link.print_name = link.substitute_name + num_unprintable_chars;
687         link.print_name_nbytes = link.substitute_name_nbytes -
688                                  (num_unprintable_chars * sizeof(wchar_t));
689
690         if (make_link_reparse_point(&link, rpbuf, rpbuflen_p))
691                 return 0;
692
693         ret = winnt_rpfix_progress(params, path, &link,
694                                    WIMLIB_SCAN_DENTRY_FIXED_SYMLINK);
695         if (ret)
696                 return ret;
697         return RP_FIXED;
698 }
699
700 /* Load the reparse data of a file into the corresponding WIM inode.  If the
701  * reparse point is a symbolic link or junction with an absolute target and
702  * RPFIX mode is enabled, then also rewrite its target to be relative to the
703  * capture root.  */
704 static noinline_for_stack int
705 winnt_load_reparse_data(HANDLE h, struct wim_inode *inode,
706                         const wchar_t *full_path, struct capture_params *params)
707 {
708         struct reparse_buffer_disk rpbuf;
709         DWORD bytes_returned;
710         u16 rpbuflen;
711         int ret;
712
713         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
714                 /* See comment above assign_stream_types_encrypted()  */
715                 WARNING("Ignoring reparse data of encrypted file \"%ls\"",
716                         printable_path(full_path));
717                 return 0;
718         }
719
720         if (!DeviceIoControl(h, FSCTL_GET_REPARSE_POINT,
721                              NULL, 0, &rpbuf, REPARSE_POINT_MAX_SIZE,
722                              &bytes_returned, NULL))
723         {
724                 win32_error(GetLastError(), L"\"%ls\": Can't get reparse point",
725                             printable_path(full_path));
726                 return WIMLIB_ERR_READLINK;
727         }
728
729         rpbuflen = bytes_returned;
730
731         if (unlikely(rpbuflen < REPARSE_DATA_OFFSET)) {
732                 ERROR("\"%ls\": reparse point buffer is too short",
733                       printable_path(full_path));
734                 return WIMLIB_ERR_INVALID_REPARSE_DATA;
735         }
736
737         if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX) {
738                 ret = winnt_try_rpfix(&rpbuf, &rpbuflen, full_path, params);
739                 if (ret == RP_FIXED)
740                         inode->i_rp_flags &= ~WIM_RP_FLAG_NOT_FIXED;
741                 else if (ret)
742                         return ret;
743         }
744
745         inode->i_reparse_tag = le32_to_cpu(rpbuf.rptag);
746         inode->i_rp_reserved = le16_to_cpu(rpbuf.rpreserved);
747
748         if (!inode_add_stream_with_data(inode,
749                                         STREAM_TYPE_REPARSE_POINT,
750                                         NO_STREAM_NAME,
751                                         rpbuf.rpdata,
752                                         rpbuflen - REPARSE_DATA_OFFSET,
753                                         params->blob_table))
754                 return WIMLIB_ERR_NOMEM;
755
756         return 0;
757 }
758
759 static DWORD WINAPI
760 win32_tally_encrypted_size_cb(unsigned char *_data, void *_size_ret,
761                               unsigned long len)
762 {
763         *(u64*)_size_ret += len;
764         return ERROR_SUCCESS;
765 }
766
767 static int
768 win32_get_encrypted_file_size(const wchar_t *path, bool is_dir, u64 *size_ret)
769 {
770         DWORD err;
771         void *file_ctx;
772         int ret;
773         DWORD flags = 0;
774
775         if (is_dir)
776                 flags |= CREATE_FOR_DIR;
777
778         err = OpenEncryptedFileRaw(path, flags, &file_ctx);
779         if (err != ERROR_SUCCESS) {
780                 win32_error(err,
781                             L"Failed to open encrypted file \"%ls\" for raw read",
782                             printable_path(path));
783                 return WIMLIB_ERR_OPEN;
784         }
785         *size_ret = 0;
786         err = ReadEncryptedFileRaw(win32_tally_encrypted_size_cb,
787                                    size_ret, file_ctx);
788         if (err != ERROR_SUCCESS) {
789                 win32_error(err,
790                             L"Failed to read raw encrypted data from \"%ls\"",
791                             printable_path(path));
792                 ret = WIMLIB_ERR_READ;
793         } else {
794                 ret = 0;
795         }
796         CloseEncryptedFileRaw(file_ctx);
797         return ret;
798 }
799
800 static int
801 winnt_scan_efsrpc_raw_data(struct wim_inode *inode, const wchar_t *nt_path,
802                            struct list_head *unhashed_blobs)
803 {
804         struct blob_descriptor *blob;
805         struct wim_inode_stream *strm;
806         int ret;
807
808         blob = new_blob_descriptor();
809         if (!blob)
810                 goto err_nomem;
811
812         blob->file_on_disk = WCSDUP(nt_path);
813         if (!blob->file_on_disk)
814                 goto err_nomem;
815         blob->blob_location = BLOB_WIN32_ENCRYPTED;
816
817         /* OpenEncryptedFileRaw() expects a Win32 name.  */
818         wimlib_assert(!wmemcmp(blob->file_on_disk, L"\\??\\", 4));
819         blob->file_on_disk[1] = L'\\';
820
821         blob->file_inode = inode;
822
823         ret = win32_get_encrypted_file_size(blob->file_on_disk,
824                                             (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY),
825                                             &blob->size);
826         if (ret)
827                 goto err;
828
829         /* Empty EFSRPC data does not make sense  */
830         wimlib_assert(blob->size != 0);
831
832         strm = inode_add_stream(inode, STREAM_TYPE_EFSRPC_RAW_DATA,
833                                 NO_STREAM_NAME, blob);
834         if (!strm)
835                 goto err_nomem;
836
837         prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
838         return 0;
839
840 err_nomem:
841         ret = WIMLIB_ERR_NOMEM;
842 err:
843         free_blob_descriptor(blob);
844         return ret;
845 }
846
847 static bool
848 get_data_stream_name(wchar_t *raw_stream_name, size_t raw_stream_name_nchars,
849                      wchar_t **stream_name_ret, size_t *stream_name_nchars_ret)
850 {
851         const wchar_t *sep, *type, *end;
852
853         /* The stream name should be returned as :NAME:TYPE  */
854         if (raw_stream_name_nchars < 1)
855                 return false;
856         if (raw_stream_name[0] != L':')
857                 return false;
858
859         raw_stream_name++;
860         raw_stream_name_nchars--;
861
862         end = raw_stream_name + raw_stream_name_nchars;
863
864         sep = wmemchr(raw_stream_name, L':', raw_stream_name_nchars);
865         if (!sep)
866                 return false;
867
868         type = sep + 1;
869         if (end - type != 5)
870                 return false;
871
872         if (wmemcmp(type, L"$DATA", 5))
873                 return false;
874
875         *stream_name_ret = raw_stream_name;
876         *stream_name_nchars_ret = sep - raw_stream_name;
877         return true;
878 }
879
880 /* Build the path to the data stream.  For unnamed streams, this is simply the
881  * path to the file.  For named streams, this is the path to the file, followed
882  * by a colon, followed by the stream name.  */
883 static wchar_t *
884 build_data_stream_path(const wchar_t *path, size_t path_nchars,
885                        const wchar_t *stream_name, size_t stream_name_nchars)
886 {
887         size_t stream_path_nchars;
888         wchar_t *stream_path;
889         wchar_t *p;
890
891         stream_path_nchars = path_nchars;
892         if (stream_name_nchars)
893                 stream_path_nchars += 1 + stream_name_nchars;
894
895         stream_path = MALLOC((stream_path_nchars + 1) * sizeof(wchar_t));
896         if (stream_path) {
897                 p = wmempcpy(stream_path, path, path_nchars);
898                 if (stream_name_nchars) {
899                         *p++ = L':';
900                         p = wmempcpy(p, stream_name, stream_name_nchars);
901                 }
902                 *p++ = L'\0';
903         }
904         return stream_path;
905 }
906
907 static int
908 winnt_scan_data_stream(const wchar_t *path, size_t path_nchars,
909                        wchar_t *raw_stream_name, size_t raw_stream_name_nchars,
910                        u64 stream_size,
911                        struct wim_inode *inode, struct list_head *unhashed_blobs)
912 {
913         wchar_t *stream_name;
914         size_t stream_name_nchars;
915         struct blob_descriptor *blob;
916         struct wim_inode_stream *strm;
917
918         /* Given the raw stream name (which is something like
919          * :streamname:$DATA), extract just the stream name part (streamname).
920          * Ignore any non-$DATA streams.  */
921         if (!get_data_stream_name(raw_stream_name, raw_stream_name_nchars,
922                                   &stream_name, &stream_name_nchars))
923                 return 0;
924
925         stream_name[stream_name_nchars] = L'\0';
926
927         /* If the stream is non-empty, set up a blob descriptor for it.  */
928         if (stream_size != 0) {
929                 blob = new_blob_descriptor();
930                 if (!blob)
931                         goto err_nomem;
932                 blob->file_on_disk = build_data_stream_path(path,
933                                                             path_nchars,
934                                                             stream_name,
935                                                             stream_name_nchars);
936                 if (!blob->file_on_disk)
937                         goto err_nomem;
938                 blob->blob_location = BLOB_IN_WINNT_FILE_ON_DISK;
939                 blob->size = stream_size;
940                 blob->file_inode = inode;
941         } else {
942                 blob = NULL;
943         }
944
945         strm = inode_add_stream(inode, STREAM_TYPE_DATA, stream_name, blob);
946         if (!strm)
947                 goto err_nomem;
948
949         prepare_unhashed_blob(blob, inode, strm->stream_id, unhashed_blobs);
950         return 0;
951
952 err_nomem:
953         free_blob_descriptor(blob);
954         return WIMLIB_ERR_NOMEM;
955 }
956
957 /*
958  * Load information about the data streams of an open file into a WIM inode.
959  *
960  * We use the NtQueryInformationFile() system call instead of FindFirstStream()
961  * and FindNextStream().  This is done for two reasons:
962  *
963  * - FindFirstStream() opens its own handle to the file or directory and
964  *   apparently does so without specifying FILE_FLAG_BACKUP_SEMANTICS, thereby
965  *   causing access denied errors on certain files (even when running as the
966  *   Administrator).
967  * - FindFirstStream() and FindNextStream() is only available on Windows Vista
968  *   and later, whereas the stream support in NtQueryInformationFile() was
969  *   already present in Windows XP.
970  */
971 static noinline_for_stack int
972 winnt_scan_data_streams(HANDLE h, const wchar_t *path, size_t path_nchars,
973                         struct wim_inode *inode, struct list_head *unhashed_blobs,
974                         u64 file_size, u32 vol_flags)
975 {
976         int ret;
977         u8 _buf[4096] _aligned_attribute(8);
978         u8 *buf;
979         size_t bufsize;
980         IO_STATUS_BLOCK iosb;
981         NTSTATUS status;
982         FILE_STREAM_INFORMATION *info;
983
984         buf = _buf;
985         bufsize = sizeof(_buf);
986
987         if (!(vol_flags & FILE_NAMED_STREAMS))
988                 goto unnamed_only;
989
990         /* Get a buffer containing the stream information.  */
991         while (!NT_SUCCESS(status = (*func_NtQueryInformationFile)(h,
992                                                                    &iosb,
993                                                                    buf,
994                                                                    bufsize,
995                                                                    FileStreamInformation)))
996         {
997
998                 switch (status) {
999                 case STATUS_BUFFER_OVERFLOW:
1000                         {
1001                                 u8 *newbuf;
1002
1003                                 bufsize *= 2;
1004                                 if (buf == _buf)
1005                                         newbuf = MALLOC(bufsize);
1006                                 else
1007                                         newbuf = REALLOC(buf, bufsize);
1008                                 if (!newbuf) {
1009                                         ret = WIMLIB_ERR_NOMEM;
1010                                         goto out_free_buf;
1011                                 }
1012                                 buf = newbuf;
1013                         }
1014                         break;
1015                 case STATUS_NOT_IMPLEMENTED:
1016                 case STATUS_NOT_SUPPORTED:
1017                 case STATUS_INVALID_INFO_CLASS:
1018                         goto unnamed_only;
1019                 default:
1020                         winnt_error(status,
1021                                     L"\"%ls\": Failed to query stream information",
1022                                     printable_path(path));
1023                         ret = WIMLIB_ERR_READ;
1024                         goto out_free_buf;
1025                 }
1026         }
1027
1028         if (iosb.Information == 0) {
1029                 /* No stream information.  */
1030                 ret = 0;
1031                 goto out_free_buf;
1032         }
1033
1034         /* Parse one or more stream information structures.  */
1035         info = (FILE_STREAM_INFORMATION *)buf;
1036         for (;;) {
1037                 /* Load the stream information.  */
1038                 ret = winnt_scan_data_stream(path, path_nchars,
1039                                              info->StreamName,
1040                                              info->StreamNameLength / 2,
1041                                              info->StreamSize.QuadPart,
1042                                              inode, unhashed_blobs);
1043                 if (ret)
1044                         goto out_free_buf;
1045
1046                 if (info->NextEntryOffset == 0) {
1047                         /* No more stream information.  */
1048                         break;
1049                 }
1050                 /* Advance to next stream information.  */
1051                 info = (FILE_STREAM_INFORMATION *)
1052                                 ((u8 *)info + info->NextEntryOffset);
1053         }
1054         ret = 0;
1055         goto out_free_buf;
1056
1057 unnamed_only:
1058         /* The volume does not support named streams.  Only capture the unnamed
1059          * data stream.  */
1060         if (inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY |
1061                                    FILE_ATTRIBUTE_REPARSE_POINT))
1062         {
1063                 ret = 0;
1064                 goto out_free_buf;
1065         }
1066
1067         {
1068                 wchar_t stream_name[] = L"::$DATA";
1069                 ret = winnt_scan_data_stream(path, path_nchars, stream_name, 7,
1070                                              file_size, inode, unhashed_blobs);
1071         }
1072 out_free_buf:
1073         /* Free buffer if allocated on heap.  */
1074         if (unlikely(buf != _buf))
1075                 FREE(buf);
1076         return ret;
1077 }
1078
1079 static noinline_for_stack u64
1080 get_sort_key(HANDLE h)
1081 {
1082         STARTING_VCN_INPUT_BUFFER in = { .StartingVcn.QuadPart = 0 };
1083         RETRIEVAL_POINTERS_BUFFER out;
1084         DWORD bytesReturned;
1085
1086         if (!DeviceIoControl(h, FSCTL_GET_RETRIEVAL_POINTERS,
1087                              &in, sizeof(in),
1088                              &out, sizeof(out),
1089                              &bytesReturned, NULL))
1090                 return 0;
1091
1092         if (out.ExtentCount < 1)
1093                 return 0;
1094
1095         return out.Extents[0].Lcn.QuadPart;
1096 }
1097
1098 static void
1099 set_sort_key(struct wim_inode *inode, u64 sort_key)
1100 {
1101         for (unsigned i = 0; i < inode->i_num_streams; i++) {
1102                 struct wim_inode_stream *strm = &inode->i_streams[i];
1103                 struct blob_descriptor *blob = stream_blob_resolved(strm);
1104                 if (blob && (blob->blob_location == BLOB_IN_WINNT_FILE_ON_DISK ||
1105                              blob->blob_location == BLOB_WIN32_ENCRYPTED))
1106                         blob->sort_key = sort_key;
1107         }
1108 }
1109
1110 static noinline_for_stack u32
1111 get_volume_information(HANDLE h, const wchar_t *full_path,
1112                        struct capture_params *params)
1113 {
1114         FILE_FS_ATTRIBUTE_INFORMATION attr_info;
1115         FILE_FS_VOLUME_INFORMATION vol_info;
1116         IO_STATUS_BLOCK iosb;
1117         NTSTATUS status;
1118         u32 vol_flags;
1119
1120         /* Get volume flags  */
1121         status = (*func_NtQueryVolumeInformationFile)(h, &iosb,
1122                                                       &attr_info,
1123                                                       sizeof(attr_info),
1124                                                       FileFsAttributeInformation);
1125         if (likely((NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW) &&
1126                    (iosb.Information >=
1127                         offsetof(FILE_FS_ATTRIBUTE_INFORMATION,
1128                                  FileSystemAttributes) +
1129                         sizeof(attr_info.FileSystemAttributes))))
1130         {
1131                 vol_flags = attr_info.FileSystemAttributes;
1132         } else {
1133                 winnt_warning(status, L"\"%ls\": Can't get volume attributes",
1134                               printable_path(full_path));
1135                 vol_flags = 0;
1136         }
1137
1138         /* Get volume ID.  */
1139         status = (*func_NtQueryVolumeInformationFile)(h, &iosb,
1140                                                       &vol_info,
1141                                                       sizeof(vol_info),
1142                                                       FileFsVolumeInformation);
1143         if (likely((NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW) &&
1144                    (iosb.Information >=
1145                         offsetof(FILE_FS_VOLUME_INFORMATION,
1146                                  VolumeSerialNumber) +
1147                         sizeof(vol_info.VolumeSerialNumber))))
1148         {
1149                 params->capture_root_dev = vol_info.VolumeSerialNumber;
1150         } else {
1151                 winnt_warning(status, L"\"%ls\": Can't get volume ID",
1152                               printable_path(full_path));
1153                 params->capture_root_dev = 0;
1154         }
1155         return vol_flags;
1156 }
1157
1158 struct file_info {
1159         u32 attributes;
1160         u32 num_links;
1161         u64 creation_time;
1162         u64 last_write_time;
1163         u64 last_access_time;
1164         u64 ino;
1165         u64 end_of_file;
1166 };
1167
1168 static noinline_for_stack NTSTATUS
1169 get_file_info(HANDLE h, struct file_info *info)
1170 {
1171         IO_STATUS_BLOCK iosb;
1172         NTSTATUS status;
1173         FILE_ALL_INFORMATION all_info;
1174
1175         status = (*func_NtQueryInformationFile)(h, &iosb, &all_info,
1176                                                 sizeof(all_info),
1177                                                 FileAllInformation);
1178
1179         if (unlikely(!NT_SUCCESS(status) && status != STATUS_BUFFER_OVERFLOW))
1180                 return status;
1181
1182         info->attributes = all_info.BasicInformation.FileAttributes;
1183         info->num_links = all_info.StandardInformation.NumberOfLinks;
1184         info->creation_time = all_info.BasicInformation.CreationTime.QuadPart;
1185         info->last_write_time = all_info.BasicInformation.LastWriteTime.QuadPart;
1186         info->last_access_time = all_info.BasicInformation.LastAccessTime.QuadPart;
1187         info->ino = all_info.InternalInformation.IndexNumber.QuadPart;
1188         info->end_of_file = all_info.StandardInformation.EndOfFile.QuadPart;
1189         return STATUS_SUCCESS;
1190 }
1191
1192 static int
1193 winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret,
1194                                   HANDLE cur_dir,
1195                                   wchar_t *full_path,
1196                                   size_t full_path_nchars,
1197                                   const wchar_t *filename,
1198                                   size_t filename_nchars,
1199                                   struct capture_params *params,
1200                                   struct winnt_scan_stats *stats,
1201                                   u32 vol_flags)
1202 {
1203         struct wim_dentry *root = NULL;
1204         struct wim_inode *inode = NULL;
1205         HANDLE h = NULL;
1206         int ret;
1207         NTSTATUS status;
1208         struct file_info file_info;
1209         ACCESS_MASK requestedPerms;
1210         u64 sort_key;
1211
1212         ret = try_exclude(full_path, params);
1213         if (unlikely(ret < 0)) /* Excluded? */
1214                 goto out_progress;
1215         if (unlikely(ret > 0)) /* Error? */
1216                 goto out;
1217
1218         /* Open the file.  */
1219         requestedPerms = FILE_READ_DATA |
1220                          FILE_READ_ATTRIBUTES |
1221                          READ_CONTROL |
1222                          ACCESS_SYSTEM_SECURITY |
1223                          SYNCHRONIZE;
1224 retry_open:
1225         status = winnt_openat(cur_dir,
1226                               (cur_dir ? filename : full_path),
1227                               (cur_dir ? filename_nchars : full_path_nchars),
1228                               requestedPerms,
1229                               &h);
1230         if (unlikely(!NT_SUCCESS(status))) {
1231                 if (status == STATUS_DELETE_PENDING) {
1232                         WARNING("\"%ls\": Deletion pending; skipping file",
1233                                 printable_path(full_path));
1234                         ret = 0;
1235                         goto out;
1236                 }
1237                 if (status == STATUS_ACCESS_DENIED &&
1238                     (requestedPerms & FILE_READ_DATA)) {
1239                         /* This happens on encrypted files.  */
1240                         requestedPerms &= ~FILE_READ_DATA;
1241                         goto retry_open;
1242                 }
1243
1244                 winnt_error(status, L"\"%ls\": Can't open file",
1245                             printable_path(full_path));
1246                 if (status == STATUS_FVE_LOCKED_VOLUME)
1247                         ret = WIMLIB_ERR_FVE_LOCKED_VOLUME;
1248                 else
1249                         ret = WIMLIB_ERR_OPEN;
1250                 goto out;
1251         }
1252
1253         /* Get information about the file.  */
1254         status = get_file_info(h, &file_info);
1255         if (!NT_SUCCESS(status)) {
1256                 winnt_error(status, L"\"%ls\": Can't get file information",
1257                             printable_path(full_path));
1258                 ret = WIMLIB_ERR_STAT;
1259                 goto out;
1260         }
1261
1262         if (unlikely(!(requestedPerms & FILE_READ_DATA)) &&
1263             !(file_info.attributes & FILE_ATTRIBUTE_ENCRYPTED))
1264         {
1265                 ERROR("\"%ls\": Permission to read data was denied",
1266                       printable_path(full_path));
1267                 ret = WIMLIB_ERR_OPEN;
1268                 goto out;
1269         }
1270
1271         if (unlikely(!cur_dir)) {
1272                 /* Root of tree being captured; get volume information.  */
1273                 vol_flags = get_volume_information(h, full_path, params);
1274                 params->capture_root_ino = file_info.ino;
1275         }
1276
1277
1278         /* Create a WIM dentry with an associated inode, which may be shared.
1279          *
1280          * However, we need to explicitly check for directories and files with
1281          * only 1 link and refuse to hard link them.  This is because Windows
1282          * has a bug where it can return duplicate File IDs for files and
1283          * directories on the FAT filesystem.
1284          *
1285          * Since we don't follow mount points on Windows, we don't need to query
1286          * the volume ID per-file.  Just once, for the root, is enough.  But we
1287          * can't simply pass 0, because then there could be inode collisions
1288          * among multiple calls to win32_build_dentry_tree() that are scanning
1289          * files on different volumes.  */
1290         ret = inode_table_new_dentry(params->inode_table,
1291                                      filename,
1292                                      file_info.ino,
1293                                      params->capture_root_dev,
1294                                      (file_info.num_links <= 1 ||
1295                                         (file_info.attributes & FILE_ATTRIBUTE_DIRECTORY)),
1296                                      &root);
1297         if (ret)
1298                 goto out;
1299
1300         /* Get the short (DOS) name of the file.  */
1301         status = winnt_get_short_name(h, root);
1302
1303         /* If we can't read the short filename for any reason other than
1304          * out-of-memory, just ignore the error and assume the file has no short
1305          * name.  This shouldn't be an issue, since the short names are
1306          * essentially obsolete anyway.  */
1307         if (unlikely(status == STATUS_NO_MEMORY)) {
1308                 ret = WIMLIB_ERR_NOMEM;
1309                 goto out;
1310         }
1311
1312         inode = root->d_inode;
1313
1314         if (inode->i_nlink > 1) {
1315                 /* Shared inode (hard link); skip reading per-inode information.
1316                  */
1317                 goto out_progress;
1318         }
1319
1320         inode->i_attributes = file_info.attributes;
1321         inode->i_creation_time = file_info.creation_time;
1322         inode->i_last_write_time = file_info.last_write_time;
1323         inode->i_last_access_time = file_info.last_access_time;
1324
1325         /* Get the file's security descriptor, unless we are capturing in
1326          * NO_ACLS mode or the volume does not support security descriptors.  */
1327         if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS)
1328             && (vol_flags & FILE_PERSISTENT_ACLS))
1329         {
1330                 status = winnt_get_security_descriptor(h, inode,
1331                                                        params->sd_set, stats,
1332                                                        params->add_flags);
1333                 if (!NT_SUCCESS(status)) {
1334                         winnt_error(status,
1335                                     L"\"%ls\": Can't read security descriptor",
1336                                     printable_path(full_path));
1337                         ret = WIMLIB_ERR_STAT;
1338                         goto out;
1339                 }
1340         }
1341
1342         /* If this is a reparse point, load the reparse data.  */
1343         if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)) {
1344                 ret = winnt_load_reparse_data(h, inode, full_path, params);
1345                 if (ret)
1346                         goto out;
1347         }
1348
1349         sort_key = get_sort_key(h);
1350
1351         if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) {
1352                 /* Load information about the raw encrypted data.  This is
1353                  * needed for any directory or non-directory that has
1354                  * FILE_ATTRIBUTE_ENCRYPTED set.
1355                  *
1356                  * Note: since OpenEncryptedFileRaw() fails with
1357                  * ERROR_SHARING_VIOLATION if there are any open handles to the
1358                  * file, we have to close the file and re-open it later if
1359                  * needed.  */
1360                 (*func_NtClose)(h);
1361                 h = NULL;
1362                 ret = winnt_scan_efsrpc_raw_data(inode, full_path,
1363                                                  params->unhashed_blobs);
1364                 if (ret)
1365                         goto out;
1366         } else {
1367                 /*
1368                  * Load information about data streams (unnamed and named).
1369                  *
1370                  * Skip this step for encrypted files, since the data from
1371                  * ReadEncryptedFileRaw() already contains all data streams (and
1372                  * they do in fact all get restored by WriteEncryptedFileRaw().)
1373                  *
1374                  * Note: WIMGAPI (as of Windows 8.1) gets wrong and stores both
1375                  * the EFSRPC data and the named data stream(s)...!
1376                  */
1377                 ret = winnt_scan_data_streams(h,
1378                                               full_path,
1379                                               full_path_nchars,
1380                                               inode,
1381                                               params->unhashed_blobs,
1382                                               file_info.end_of_file,
1383                                               vol_flags);
1384                 if (ret)
1385                         goto out;
1386         }
1387
1388         set_sort_key(inode, sort_key);
1389
1390         if (inode_is_directory(inode)) {
1391
1392                 /* Directory: recurse to children.  */
1393
1394                 if (unlikely(!h)) {
1395                         /* Re-open handle that was closed to read raw encrypted
1396                          * data.  */
1397                         status = winnt_openat(cur_dir,
1398                                               (cur_dir ?
1399                                                filename : full_path),
1400                                               (cur_dir ?
1401                                                filename_nchars : full_path_nchars),
1402                                               FILE_LIST_DIRECTORY | SYNCHRONIZE,
1403                                               &h);
1404                         if (!NT_SUCCESS(status)) {
1405                                 winnt_error(status,
1406                                             L"\"%ls\": Can't re-open file",
1407                                             printable_path(full_path));
1408                                 ret = WIMLIB_ERR_OPEN;
1409                                 goto out;
1410                         }
1411                 }
1412                 ret = winnt_recurse_directory(h,
1413                                               full_path,
1414                                               full_path_nchars,
1415                                               root,
1416                                               params,
1417                                               stats,
1418                                               vol_flags);
1419                 if (ret)
1420                         goto out;
1421         }
1422
1423 out_progress:
1424         params->progress.scan.cur_path = printable_path(full_path);
1425         if (likely(root))
1426                 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK, inode);
1427         else
1428                 ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL);
1429 out:
1430         if (likely(h))
1431                 (*func_NtClose)(h);
1432         if (unlikely(ret)) {
1433                 free_dentry_tree(root, params->blob_table);
1434                 root = NULL;
1435                 ret = report_capture_error(params, ret, full_path);
1436         }
1437         *root_ret = root;
1438         return ret;
1439 }
1440
1441 static void
1442 winnt_do_scan_warnings(const wchar_t *path, const struct winnt_scan_stats *stats)
1443 {
1444         if (likely(stats->num_get_sacl_priv_notheld == 0 &&
1445                    stats->num_get_sd_access_denied == 0))
1446                 return;
1447
1448         WARNING("Scan of \"%ls\" complete, but with one or more warnings:", path);
1449         if (stats->num_get_sacl_priv_notheld != 0) {
1450                 WARNING("- Could not capture SACL (System Access Control List)\n"
1451                         "            on %lu files or directories.",
1452                         stats->num_get_sacl_priv_notheld);
1453         }
1454         if (stats->num_get_sd_access_denied != 0) {
1455                 WARNING("- Could not capture security descriptor at all\n"
1456                         "            on %lu files or directories.",
1457                         stats->num_get_sd_access_denied);
1458         }
1459         WARNING("To fully capture all security descriptors, run the program\n"
1460                 "          with Administrator rights.");
1461 }
1462
1463 #define WINDOWS_NT_MAX_PATH 32768
1464
1465 /* Win32 version of capturing a directory tree.  */
1466 int
1467 win32_build_dentry_tree(struct wim_dentry **root_ret,
1468                         const wchar_t *root_disk_path,
1469                         struct capture_params *params)
1470 {
1471         wchar_t *path;
1472         int ret;
1473         UNICODE_STRING ntpath;
1474         struct winnt_scan_stats stats;
1475         size_t ntpath_nchars;
1476
1477         /* WARNING: There is no check for overflow later when this buffer is
1478          * being used!  But it's as long as the maximum path length understood
1479          * by Windows NT (which is NOT the same as MAX_PATH).  */
1480         path = MALLOC((WINDOWS_NT_MAX_PATH + 1) * sizeof(wchar_t));
1481         if (!path)
1482                 return WIMLIB_ERR_NOMEM;
1483
1484         ret = win32_path_to_nt_path(root_disk_path, &ntpath);
1485         if (ret)
1486                 goto out_free_path;
1487
1488         if (ntpath.Length < 4 * sizeof(wchar_t) ||
1489             ntpath.Length > WINDOWS_NT_MAX_PATH * sizeof(wchar_t) ||
1490             wmemcmp(ntpath.Buffer, L"\\??\\", 4))
1491         {
1492                 ERROR("\"%ls\": unrecognized path format", root_disk_path);
1493                 ret = WIMLIB_ERR_INVALID_PARAM;
1494         } else {
1495                 ntpath_nchars = ntpath.Length / sizeof(wchar_t);
1496                 wmemcpy(path, ntpath.Buffer, ntpath_nchars);
1497                 path[ntpath_nchars] = L'\0';
1498
1499                 params->capture_root_nchars = ntpath_nchars;
1500                 if (path[ntpath_nchars - 1] == L'\\')
1501                         params->capture_root_nchars--;
1502                 ret = 0;
1503         }
1504         HeapFree(GetProcessHeap(), 0, ntpath.Buffer);
1505         if (ret)
1506                 goto out_free_path;
1507
1508         memset(&stats, 0, sizeof(stats));
1509
1510         ret = winnt_build_dentry_tree_recursive(root_ret, NULL,
1511                                                 path, ntpath_nchars,
1512                                                 L"", 0, params, &stats, 0);
1513 out_free_path:
1514         FREE(path);
1515         if (ret == 0)
1516                 winnt_do_scan_warnings(root_disk_path, &stats);
1517         return ret;
1518 }
1519
1520 #endif /* __WIN32__ */