ae591bbef6f33cf41d9b5d407b054de35a871b58
[wimlib] / src / extract.c
1 /*
2  * extract.c
3  *
4  * Support for extracting WIM images, or files or directories contained in a WIM
5  * image.
6  */
7
8 /*
9  * Copyright (C) 2012, 2013 Eric Biggers
10  *
11  * This file is part of wimlib, a library for working with WIM files.
12  *
13  * wimlib is free software; you can redistribute it and/or modify it under the
14  * terms of the GNU General Public License as published by the Free
15  * Software Foundation; either version 3 of the License, or (at your option)
16  * any later version.
17  *
18  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20  * A PARTICULAR PURPOSE. See the GNU General Public License for more
21  * details.
22  *
23  * You should have received a copy of the GNU General Public License
24  * along with wimlib; if not, see http://www.gnu.org/licenses/.
25  */
26
27 /*
28  * This file provides the API functions wimlib_extract_image(),
29  * wimlib_extract_files(), and wimlib_extract_image_from_pipe().  Internally,
30  * all end up calling extract_tree() zero or more times to extract a tree of
31  * files from the currently selected WIM image to the specified target directory
32  * or NTFS volume.
33  *
34  * Although wimlib supports multiple extraction modes/backends (NTFS-3g, UNIX,
35  * Win32), this file does not itself have code to extract files or directories
36  * to any specific target; instead, it handles generic functionality and relies
37  * on lower-level callback functions declared in `struct apply_operations' to do
38  * the actual extraction.
39  */
40
41 #ifdef HAVE_CONFIG_H
42 #  include "config.h"
43 #endif
44
45 #include "wimlib/apply.h"
46 #include "wimlib/dentry.h"
47 #include "wimlib/encoding.h"
48 #include "wimlib/endianness.h"
49 #include "wimlib/error.h"
50 #include "wimlib/lookup_table.h"
51 #include "wimlib/metadata.h"
52 #include "wimlib/paths.h"
53 #include "wimlib/reparse.h"
54 #include "wimlib/resource.h"
55 #include "wimlib/security.h"
56 #ifdef __WIN32__
57 #  include "wimlib/win32.h" /* for realpath() equivalent */
58 #endif
59 #include "wimlib/xml.h"
60 #include "wimlib/wim.h"
61
62 #include <errno.h>
63 #include <fcntl.h>
64 #include <stdlib.h>
65 #include <sys/stat.h>
66 #include <unistd.h>
67
68 #define WIMLIB_EXTRACT_FLAG_MULTI_IMAGE 0x80000000
69 #define WIMLIB_EXTRACT_FLAG_FROM_PIPE   0x40000000
70 #define WIMLIB_EXTRACT_MASK_PUBLIC      0x3fffffff
71
72 /* Given a WIM dentry in the tree to be extracted, resolve all streams in the
73  * corresponding inode and set 'out_refcnt' in each to 0.  */
74 static int
75 dentry_resolve_and_zero_lte_refcnt(struct wim_dentry *dentry, void *_ctx)
76 {
77         struct apply_ctx *ctx = _ctx;
78         struct wim_inode *inode = dentry->d_inode;
79         struct wim_lookup_table_entry *lte;
80         int ret;
81         bool force = false;
82
83         if (dentry->extraction_skipped)
84                 return 0;
85
86         /* Special case:  when extracting from a pipe, the WIM lookup table is
87          * initially empty, so "resolving" an inode's streams is initially not
88          * possible.  However, we still need to keep track of which streams,
89          * identified by SHA1 message digests, need to be extracted, so we
90          * "resolve" the inode's streams anyway by allocating new entries.  */
91         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE)
92                 force = true;
93         ret = inode_resolve_ltes(inode, ctx->wim->lookup_table, force);
94         if (ret)
95                 return ret;
96         for (unsigned i = 0; i <= inode->i_num_ads; i++) {
97                 lte = inode_stream_lte_resolved(inode, i);
98                 if (lte)
99                         lte->out_refcnt = 0;
100         }
101         return 0;
102 }
103
104 static inline bool
105 is_linked_extraction(const struct apply_ctx *ctx)
106 {
107         return 0 != (ctx->extract_flags & (WIMLIB_EXTRACT_FLAG_HARDLINK |
108                                            WIMLIB_EXTRACT_FLAG_SYMLINK));
109 }
110
111 static inline bool
112 can_extract_named_data_streams(const struct apply_ctx *ctx)
113 {
114         return ctx->supported_features.named_data_streams &&
115                 !is_linked_extraction(ctx);
116 }
117
118 static int
119 ref_stream_to_extract(struct wim_lookup_table_entry *lte,
120                       struct wim_dentry *dentry, struct apply_ctx *ctx)
121 {
122         if (!lte)
123                 return 0;
124
125         /* Tally the size only for each extraction of the stream (not hard
126          * links).  */
127         if (!(dentry->d_inode->i_visited &&
128               ctx->supported_features.hard_links) &&
129             (!is_linked_extraction(ctx) || (lte->out_refcnt == 0 &&
130                                             lte->extracted_file == NULL)))
131         {
132                 ctx->progress.extract.total_bytes += wim_resource_size(lte);
133                 ctx->progress.extract.num_streams++;
134         }
135
136         /* Add stream to the extraction_list only one time, even if it's going
137          * to be extracted to multiple locations.  */
138         if (lte->out_refcnt == 0) {
139                 list_add_tail(&lte->extraction_list, &ctx->stream_list);
140                 ctx->num_streams_remaining++;
141         }
142
143         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_SEQUENTIAL) {
144                 struct wim_dentry **lte_dentries;
145
146                 /* Append dentry to this stream's array of dentries referencing
147                  * it.  Use inline array to avoid memory allocation until the
148                  * number of dentries becomes too large.  */
149                 if (lte->out_refcnt < ARRAY_LEN(lte->inline_lte_dentries)) {
150                         lte_dentries = lte->inline_lte_dentries;
151                 } else {
152                         struct wim_dentry **prev_lte_dentries;
153                         size_t alloc_lte_dentries;
154
155                         if (lte->out_refcnt == ARRAY_LEN(lte->inline_lte_dentries)) {
156                                 prev_lte_dentries = NULL;
157                                 alloc_lte_dentries = ARRAY_LEN(lte->inline_lte_dentries);
158                         } else {
159                                 prev_lte_dentries = lte->lte_dentries;
160                                 alloc_lte_dentries = lte->alloc_lte_dentries;
161                         }
162
163                         if (lte->out_refcnt == alloc_lte_dentries) {
164                                 alloc_lte_dentries *= 2;
165                                 lte_dentries = REALLOC(prev_lte_dentries,
166                                                        alloc_lte_dentries *
167                                                         sizeof(lte_dentries[0]));
168                                 if (!lte_dentries)
169                                         return WIMLIB_ERR_NOMEM;
170                                 if (prev_lte_dentries == NULL) {
171                                         memcpy(lte_dentries,
172                                                lte->inline_lte_dentries,
173                                                sizeof(lte->inline_lte_dentries));
174                                 }
175                                 lte->lte_dentries = lte_dentries;
176                                 lte->alloc_lte_dentries = alloc_lte_dentries;
177                         }
178                         lte_dentries = lte->lte_dentries;
179                 }
180                 lte_dentries[lte->out_refcnt] = dentry;
181         }
182         lte->out_refcnt++;
183         return 0;
184 }
185
186 /* Given a WIM dentry in the tree to be extracted, iterate through streams that
187  * need to be extracted.  For each one, add it to the list of streams to be
188  * extracted (ctx->stream_list) if not already done so, and also update the
189  * progress information (ctx->progress) with the stream.  Furthermore, if doing
190  * a sequential extraction, build a mapping from each the stream to the dentries
191  * referencing it.
192  *
193  * This uses the i_visited member of the inodes (assumed to be 0 initially).  */
194 static int
195 dentry_add_streams_to_extract(struct wim_dentry *dentry, void *_ctx)
196 {
197         struct apply_ctx *ctx = _ctx;
198         struct wim_inode *inode = dentry->d_inode;
199         int ret;
200
201         /* Don't process dentries marked as skipped.  */
202         if (dentry->extraction_skipped)
203                 return 0;
204
205         /* The unnamed data stream will always be extracted, except in an
206          * unlikely case.  */
207         if (!inode_is_encrypted_directory(inode)) {
208                 ret = ref_stream_to_extract(inode_unnamed_lte_resolved(inode),
209                                             dentry, ctx);
210                 if (ret)
211                         return ret;
212         }
213
214         /* Named data streams will be extracted only if supported in the current
215          * extraction mode and volume, and to avoid complications, if not doing
216          * a linked extraction.  */
217         if (can_extract_named_data_streams(ctx)) {
218                 for (u16 i = 0; i < inode->i_num_ads; i++) {
219                         if (!ads_entry_is_named_stream(&inode->i_ads_entries[i]))
220                                 continue;
221                         ret = ref_stream_to_extract(inode->i_ads_entries[i].lte,
222                                                     dentry, ctx);
223                         if (ret)
224                                 return ret;
225                 }
226         }
227         inode->i_visited = 1;
228         return 0;
229 }
230
231 /* Inform library user of progress of stream extraction following the successful
232  * extraction of a copy of the stream specified by @lte.  */
233 static void
234 update_extract_progress(struct apply_ctx *ctx,
235                         const struct wim_lookup_table_entry *lte)
236 {
237         wimlib_progress_func_t progress_func = ctx->progress_func;
238         union wimlib_progress_info *progress = &ctx->progress;
239
240         progress->extract.completed_bytes += wim_resource_size(lte);
241         if (progress_func &&
242             progress->extract.completed_bytes >= ctx->next_progress)
243         {
244                 progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS, progress);
245                 if (progress->extract.completed_bytes >=
246                     progress->extract.total_bytes)
247                 {
248                         ctx->next_progress = ~0ULL;
249                 } else {
250                         ctx->next_progress += progress->extract.total_bytes / 128;
251                         if (ctx->next_progress > progress->extract.total_bytes)
252                                 ctx->next_progress = progress->extract.total_bytes;
253                 }
254         }
255 }
256
257 #ifndef __WIN32__
258 /* Extract a symbolic link (not directly as reparse data), handling fixing up
259  * the target of absolute symbolic links and updating the extract progress.
260  *
261  * @inode must specify the WIM inode for a symbolic link or junction reparse
262  * point.
263  *
264  * @lte_override overrides the resource used as the reparse data for the
265  * symbolic link.  */
266 static int
267 extract_symlink(const tchar *path, struct apply_ctx *ctx,
268                 struct wim_inode *inode,
269                 struct wim_lookup_table_entry *lte_override)
270 {
271         ssize_t bufsize = ctx->ops->path_max;
272         tchar target[bufsize];
273         tchar *buf = target;
274         tchar *fixed_target;
275         ssize_t sret;
276         int ret;
277
278         /* If absolute symbolic link fixups requested, reserve space in the link
279          * target buffer for the absolute path of the target directory.  */
280         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX)
281         {
282                 buf += ctx->realtarget_nchars;
283                 bufsize -= ctx->realtarget_nchars;
284         }
285
286         /* Translate the WIM inode's reparse data into the link target.  */
287         sret = wim_inode_readlink(inode, buf, bufsize - 1, lte_override);
288         if (sret < 0) {
289                 errno = -sret;
290                 return WIMLIB_ERR_READLINK;
291         }
292         buf[sret] = '\0';
293
294         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
295             buf[0] == '/')
296         {
297                 /* Fix absolute symbolic link target to point into the
298                  * actual extraction destination.  */
299                 tmemcpy(target, ctx->realtarget, ctx->realtarget_nchars);
300                 fixed_target = target;
301         } else {
302                 /* Keep same link target.  */
303                 fixed_target = buf;
304         }
305
306         /* Call into the apply_operations to create the symbolic link.  */
307         DEBUG("Creating symlink \"%"TS"\" => \"%"TS"\"",
308               path, fixed_target);
309         ret = ctx->ops->create_symlink(fixed_target, path, ctx);
310         if (ret) {
311                 ERROR_WITH_ERRNO("Failed to create symlink "
312                                  "\"%"TS"\" => \"%"TS"\"", path, fixed_target);
313                 return ret;
314         }
315
316         /* Account for reparse data consumed.  */
317         update_extract_progress(ctx,
318                                 (lte_override ? lte_override :
319                                       inode_unnamed_lte_resolved(inode)));
320         return 0;
321 }
322 #endif /* !__WIN32__ */
323
324 /* Create a file, directory, or symbolic link.  */
325 static int
326 extract_inode(const tchar *path, struct apply_ctx *ctx, struct wim_inode *inode)
327 {
328         int ret;
329
330 #ifndef __WIN32__
331         if (ctx->supported_features.symlink_reparse_points &&
332             !ctx->supported_features.reparse_points &&
333             inode_is_symlink(inode))
334         {
335                 ret = extract_symlink(path, ctx, inode, NULL);
336         } else
337 #endif /* !__WIN32__ */
338         if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) {
339                 ret = ctx->ops->create_directory(path, ctx, &inode->extract_cookie);
340                 if (ret) {
341                         ERROR_WITH_ERRNO("Failed to create the directory "
342                                          "\"%"TS"\"", path);
343                 }
344         } else if ((inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) &&
345                     ctx->ops->extract_encrypted_stream_creates_file &&
346                     ctx->supported_features.encrypted_files) {
347                 ret = ctx->ops->extract_encrypted_stream(
348                                 path, inode_unnamed_lte_resolved(inode), ctx);
349                 if (ret) {
350                         ERROR_WITH_ERRNO("Failed to create and extract "
351                                          "encrypted file \"%"TS"\"", path);
352                 }
353         } else {
354                 ret = ctx->ops->create_file(path, ctx, &inode->extract_cookie);
355                 if (ret) {
356                         ERROR_WITH_ERRNO("Failed to create the file "
357                                          "\"%"TS"\"", path);
358                 }
359         }
360         return ret;
361 }
362
363 static int
364 extract_hardlink(const tchar *oldpath, const tchar *newpath,
365                  struct apply_ctx *ctx)
366 {
367         int ret;
368
369         DEBUG("Creating hardlink \"%"TS"\" => \"%"TS"\"", newpath, oldpath);
370         ret = ctx->ops->create_hardlink(oldpath, newpath, ctx);
371         if (ret) {
372                 ERROR_WITH_ERRNO("Failed to create hardlink "
373                                  "\"%"TS"\" => \"%"TS"\"",
374                                  newpath, oldpath);
375         }
376         return ret;
377 }
378
379 #ifdef __WIN32__
380 static int
381 try_extract_rpfix(u8 *rpbuf,
382                   u16 *rpbuflen_p,
383                   const wchar_t *extract_root_realpath,
384                   unsigned extract_root_realpath_nchars)
385 {
386         struct reparse_data rpdata;
387         wchar_t *target;
388         size_t target_nchars;
389         size_t stripped_nchars;
390         wchar_t *stripped_target;
391         wchar_t stripped_target_nchars;
392         int ret;
393
394         utf16lechar *new_target;
395         utf16lechar *new_print_name;
396         size_t new_target_nchars;
397         size_t new_print_name_nchars;
398         utf16lechar *p;
399
400         ret = parse_reparse_data(rpbuf, *rpbuflen_p, &rpdata);
401         if (ret)
402                 return ret;
403
404         if (extract_root_realpath[0] == L'\0' ||
405             extract_root_realpath[1] != L':' ||
406             extract_root_realpath[2] != L'\\')
407                 return WIMLIB_ERR_REPARSE_POINT_FIXUP_FAILED;
408
409         ret = parse_substitute_name(rpdata.substitute_name,
410                                     rpdata.substitute_name_nbytes,
411                                     rpdata.rptag);
412         if (ret < 0)
413                 return 0;
414         stripped_nchars = ret;
415         target = rpdata.substitute_name;
416         target_nchars = rpdata.substitute_name_nbytes / sizeof(utf16lechar);
417         stripped_target = target + stripped_nchars;
418         stripped_target_nchars = target_nchars - stripped_nchars;
419
420         new_target = alloca((6 + extract_root_realpath_nchars +
421                              stripped_target_nchars) * sizeof(utf16lechar));
422
423         p = new_target;
424         if (stripped_nchars == 6) {
425                 /* Include \??\ prefix if it was present before */
426                 p = wmempcpy(p, L"\\??\\", 4);
427         }
428
429         /* Print name excludes the \??\ if present. */
430         new_print_name = p;
431         if (stripped_nchars != 0) {
432                 /* Get drive letter from real path to extract root, if a drive
433                  * letter was present before. */
434                 *p++ = extract_root_realpath[0];
435                 *p++ = extract_root_realpath[1];
436         }
437         /* Copy the rest of the extract root */
438         p = wmempcpy(p, extract_root_realpath + 2, extract_root_realpath_nchars - 2);
439
440         /* Append the stripped target */
441         p = wmempcpy(p, stripped_target, stripped_target_nchars);
442         new_target_nchars = p - new_target;
443         new_print_name_nchars = p - new_print_name;
444
445         if (new_target_nchars * sizeof(utf16lechar) >= REPARSE_POINT_MAX_SIZE ||
446             new_print_name_nchars * sizeof(utf16lechar) >= REPARSE_POINT_MAX_SIZE)
447                 return WIMLIB_ERR_REPARSE_POINT_FIXUP_FAILED;
448
449         rpdata.substitute_name = new_target;
450         rpdata.substitute_name_nbytes = new_target_nchars * sizeof(utf16lechar);
451         rpdata.print_name = new_print_name;
452         rpdata.print_name_nbytes = new_print_name_nchars * sizeof(utf16lechar);
453         return make_reparse_buffer(&rpdata, rpbuf, rpbuflen_p);
454 }
455 #endif /* __WIN32__ */
456
457 /* Set reparse data on extracted file or directory that has
458  * FILE_ATTRIBUTE_REPARSE_POINT set.  */
459 static int
460 extract_reparse_data(const tchar *path, struct apply_ctx *ctx,
461                      struct wim_inode *inode,
462                      struct wim_lookup_table_entry *lte_override)
463 {
464         int ret;
465         u8 rpbuf[REPARSE_POINT_MAX_SIZE];
466         u16 rpbuflen;
467
468         ret = wim_inode_get_reparse_data(inode, rpbuf, &rpbuflen, lte_override);
469         if (ret)
470                 goto error;
471
472 #ifdef __WIN32__
473         /* Fix up target of absolute symbolic link or junction points so
474          * that they point into the actual extraction target.  */
475         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
476             (inode->i_reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK ||
477              inode->i_reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT) &&
478             !inode->i_not_rpfixed)
479         {
480                 ret = try_extract_rpfix(rpbuf, &rpbuflen, ctx->realtarget,
481                                         ctx->realtarget_nchars);
482                 if (ret && !(ctx->extract_flags &
483                              WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS))
484                 {
485                         WARNING("Reparse point fixup of \"%"TS"\" "
486                                 "failed", path);
487                         ret = 0;
488                 }
489                 if (ret)
490                         goto error;
491         }
492 #endif
493
494         ret = ctx->ops->set_reparse_data(path, rpbuf, rpbuflen, ctx);
495
496         /* On Windows, the SeCreateSymbolicLink privilege is required to create
497          * symbolic links.  To be more friendly towards non-Administrator users,
498          * we merely warn the user if symbolic links cannot be created due to
499          * insufficient permissions or privileges, unless
500          * WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS was provided.  */
501 #ifdef __WIN32__
502         if (ret && inode_is_symlink(inode) &&
503             (errno == EACCES || errno == EPERM) &&
504             !(ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS))
505         {
506                 WARNING("Can't set reparse data on \"%"TS"\": "
507                         "Access denied!\n"
508                         "          You may be trying to "
509                         "extract a symbolic link without the\n"
510                         "          SeCreateSymbolicLink privilege, "
511                         "which by default non-Administrator\n"
512                         "          accounts do not have.",
513                         path);
514                 ret = 0;
515         }
516 #endif
517         if (ret)
518                 goto error;
519
520         /* Account for reparse data consumed.  */
521         update_extract_progress(ctx,
522                                 (lte_override ? lte_override :
523                                       inode_unnamed_lte_resolved(inode)));
524         return 0;
525
526 error:
527         ERROR_WITH_ERRNO("Failed to set reparse data on \"%"TS"\"", path);
528         return ret;
529 }
530
531 /*
532  * Extract zero or more streams to a file.
533  *
534  * This function operates slightly differently depending on whether @lte_spec is
535  * NULL or not.  When @lte_spec is NULL, the behavior is to extract the default
536  * file contents (unnamed stream), and, if named data streams are supported in
537  * the extract mode and volume, any named data streams.  When @lte_spec is NULL,
538  * the behavior is to extract only all copies of the stream @lte_spec, and in
539  * addition use @lte_spec to set the reparse data or create the symbolic link if
540  * appropriate.
541  *
542  * @path
543  *      Path to file to extract (as can be passed to apply_operations
544  *      functions).
545  * @ctx
546  *      Apply context.
547  * @dentry
548  *      WIM dentry that corresponds to the file being extracted.
549  * @lte_spec
550  *      If non-NULL, specifies the lookup table entry for a stream to extract,
551  *      and only that stream will be extracted (although there may be more than
552  *      one instance of it).
553  * @lte_override
554  *      Used only if @lte_spec != NULL; it is passed to the extraction functions
555  *      rather than @lte_spec, allowing the location of the stream to be
556  *      overridden.  (This is used when the WIM is being read from a nonseekable
557  *      file, such as a pipe, when streams need to be used more than once; each
558  *      such stream is extracted to a temporary file.)
559  */
560 static int
561 extract_streams(const tchar *path, struct apply_ctx *ctx,
562                 struct wim_dentry *dentry,
563                 struct wim_lookup_table_entry *lte_spec,
564                 struct wim_lookup_table_entry *lte_override)
565 {
566         struct wim_inode *inode = dentry->d_inode;
567         struct wim_lookup_table_entry *lte;
568         file_spec_t file_spec;
569         int ret;
570
571         if (dentry->was_hardlinked)
572                 return 0;
573
574 #ifdef ENABLE_DEBUG
575         if (lte_spec) {
576                 char sha1_str[100];
577                 char *p = sha1_str;
578                 for (unsigned i = 0; i < SHA1_HASH_SIZE; i++)
579                         p += sprintf(p, "%02x", lte_override->hash[i]);
580                 DEBUG("Extracting stream SHA1=%s to \"%"TS"\"",
581                       sha1_str, path, inode->i_ino);
582         } else {
583                 DEBUG("Extracting streams to \"%"TS"\"", path, inode->i_ino);
584         }
585 #endif
586
587         if (ctx->ops->uses_cookies)
588                 file_spec.cookie = inode->extract_cookie;
589         else
590                 file_spec.path = path;
591
592         /* Unnamed data stream.  */
593         lte = inode_unnamed_lte_resolved(inode);
594         if (lte && (!lte_spec || lte == lte_spec)) {
595                 if (lte_spec)
596                         lte = lte_override;
597                 if (!(inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY |
598                                              FILE_ATTRIBUTE_REPARSE_POINT)))
599                 {
600                         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED &&
601                             ctx->supported_features.encrypted_files) {
602                                 if (!ctx->ops->extract_encrypted_stream_creates_file) {
603                                         ret = ctx->ops->extract_encrypted_stream(
604                                                                 path, lte, ctx);
605                                         if (ret)
606                                                 goto error;
607                                 }
608                         } else {
609                                 ret = ctx->ops->extract_unnamed_stream(
610                                                         file_spec, lte, ctx);
611                                 if (ret)
612                                         goto error;
613                         }
614                         update_extract_progress(ctx, lte);
615                 }
616                 else if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)
617                 {
618                         ret = 0;
619                         if (ctx->supported_features.reparse_points)
620                                 ret = extract_reparse_data(path, ctx, inode, lte);
621                 #ifndef __WIN32__
622                         else if ((inode_is_symlink(inode) &&
623                                   ctx->supported_features.symlink_reparse_points))
624                                 ret = extract_symlink(path, ctx, inode, lte);
625                 #endif
626                         if (ret)
627                                 return ret;
628                 }
629         }
630
631         /* Named data streams.  */
632         if (can_extract_named_data_streams(ctx)) {
633                 for (u16 i = 0; i < inode->i_num_ads; i++) {
634                         struct wim_ads_entry *entry = &inode->i_ads_entries[i];
635
636                         if (!ads_entry_is_named_stream(entry))
637                                 continue;
638                         lte = entry->lte;
639                         if (!lte)
640                                 continue;
641                         if (lte_spec && lte_spec != lte)
642                                 continue;
643                         if (lte_spec)
644                                 lte = lte_override;
645                         ret = ctx->ops->extract_named_stream(file_spec, entry->stream_name,
646                                                              entry->stream_name_nbytes / 2,
647                                                              lte, ctx);
648                         if (ret)
649                                 goto error;
650                         update_extract_progress(ctx, lte);
651                 }
652         }
653         return 0;
654
655 error:
656         ERROR_WITH_ERRNO("Failed to extract data of \"%"TS"\"", path);
657         return ret;
658 }
659
660 /* Set attributes on an extracted file or directory if supported by the
661  * extraction mode.  */
662 static int
663 extract_file_attributes(const tchar *path, struct apply_ctx *ctx,
664                         struct wim_dentry *dentry, unsigned pass)
665 {
666         int ret;
667
668         if (ctx->ops->set_file_attributes &&
669             !(dentry == ctx->extract_root && ctx->root_dentry_is_special)) {
670                 u32 attributes = dentry->d_inode->i_attributes;
671
672                 /* Clear unsupported attributes.  */
673                 attributes &= ctx->supported_attributes_mask;
674
675                 if ((attributes & FILE_ATTRIBUTE_DIRECTORY &&
676                      !ctx->supported_features.encrypted_directories) ||
677                     (!(attributes & FILE_ATTRIBUTE_DIRECTORY) &&
678                      !ctx->supported_features.encrypted_files))
679                 {
680                         attributes &= ~FILE_ATTRIBUTE_ENCRYPTED;
681                 }
682
683                 if (attributes == 0)
684                         attributes = FILE_ATTRIBUTE_NORMAL;
685
686                 ret = ctx->ops->set_file_attributes(path, attributes, ctx, pass);
687                 if (ret) {
688                         ERROR_WITH_ERRNO("Failed to set attributes on "
689                                          "\"%"TS"\"", path);
690                         return ret;
691                 }
692         }
693         return 0;
694 }
695
696
697 /* Set or remove the short (DOS) name on an extracted file or directory if
698  * supported by the extraction mode.  Since DOS names are unimportant and it's
699  * easy to run into problems setting them on Windows (SetFileShortName()
700  * requires SE_RESTORE privilege, which only the Administrator can request, and
701  * also requires DELETE access to the file), failure is ignored unless
702  * WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES is set.  */
703 static int
704 extract_short_name(const tchar *path, struct apply_ctx *ctx,
705                    struct wim_dentry *dentry)
706 {
707         int ret;
708
709         /* The root of the dentry tree being extracted may not be extracted to
710          * its original name, so its short name should be ignored.  */
711         if (dentry == ctx->extract_root)
712                 return 0;
713
714         if (ctx->supported_features.short_names) {
715                 ret = ctx->ops->set_short_name(path,
716                                                dentry->short_name,
717                                                dentry->short_name_nbytes / 2,
718                                                ctx);
719                 if (ret && (ctx->extract_flags &
720                             WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES))
721                 {
722                         ERROR_WITH_ERRNO("Failed to set short name of "
723                                          "\"%"TS"\"", path);
724                         return ret;
725                 }
726         }
727         return 0;
728 }
729
730 /* Set security descriptor, UNIX data, or neither on an extracted file, taking
731  * into account the current extraction mode and flags.  */
732 static int
733 extract_security(const tchar *path, struct apply_ctx *ctx,
734                  struct wim_dentry *dentry)
735 {
736         int ret;
737         struct wim_inode *inode = dentry->d_inode;
738
739         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_NO_ACLS)
740                 return 0;
741
742         if ((ctx->extract_root == dentry) && ctx->root_dentry_is_special)
743                 return 0;
744
745 #ifndef __WIN32__
746         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) {
747                 struct wimlib_unix_data data;
748
749                 ret = inode_get_unix_data(inode, &data, NULL);
750                 if (ret < 0)
751                         ret = 0;
752                 else if (ret == 0)
753                         ret = ctx->ops->set_unix_data(path, &data, ctx);
754                 if (ret) {
755                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_ACLS) {
756                                 ERROR_WITH_ERRNO("Failed to set UNIX owner, "
757                                                  "group, and/or mode on "
758                                                  "\"%"TS"\"", path);
759                                 return ret;
760                         } else {
761                                 WARNING_WITH_ERRNO("Failed to set UNIX owner, "
762                                                    "group, and/or/mode on "
763                                                    "\"%"TS"\"", path);
764                         }
765                 }
766         }
767         else
768 #endif /* __WIN32__ */
769         if (ctx->supported_features.security_descriptors &&
770             inode->i_security_id != -1)
771         {
772                 const struct wim_security_data *sd;
773                 const u8 *desc;
774                 size_t desc_size;
775
776                 sd = wim_const_security_data(ctx->wim);
777                 desc = sd->descriptors[inode->i_security_id];
778                 desc_size = sd->sizes[inode->i_security_id];
779
780                 ret = ctx->ops->set_security_descriptor(path, desc,
781                                                         desc_size, ctx);
782                 if (ret) {
783                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_ACLS) {
784                                 ERROR_WITH_ERRNO("Failed to set security "
785                                                  "descriptor on \"%"TS"\"", path);
786                                 return ret;
787                         } else {
788                         #if 0
789                                 if (errno != EACCES) {
790                                         WARNING_WITH_ERRNO("Failed to set "
791                                                            "security descriptor "
792                                                            "on \"%"TS"\"", path);
793                                 }
794                         #endif
795                                 ctx->no_security_descriptors++;
796                         }
797                 }
798         }
799         return 0;
800 }
801
802 /* Set timestamps on an extracted file.  Failure is warning-only unless
803  * WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS is set.  */
804 static int
805 extract_timestamps(const tchar *path, struct apply_ctx *ctx,
806                    struct wim_dentry *dentry)
807 {
808         struct wim_inode *inode = dentry->d_inode;
809         int ret;
810
811         if ((ctx->extract_root == dentry) && ctx->root_dentry_is_special)
812                 return 0;
813
814         if (ctx->ops->set_timestamps) {
815                 ret = ctx->ops->set_timestamps(path,
816                                                inode->i_creation_time,
817                                                inode->i_last_write_time,
818                                                inode->i_last_access_time,
819                                                ctx);
820                 if (ret) {
821                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS) {
822                                 ERROR_WITH_ERRNO("Failed to set timestamps "
823                                                  "on \"%"TS"\"", path);
824                                 return ret;
825                         } else {
826                                 WARNING_WITH_ERRNO("Failed to set timestamps "
827                                                    "on \"%"TS"\"", path);
828                         }
829                 }
830         }
831         return 0;
832 }
833
834 /* Check whether the extraction of a dentry should be skipped completely.  */
835 static bool
836 dentry_is_supported(struct wim_dentry *dentry,
837                     const struct wim_features *supported_features)
838 {
839         struct wim_inode *inode = dentry->d_inode;
840
841         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
842                 return supported_features->reparse_points ||
843                         (inode_is_symlink(inode) &&
844                          supported_features->symlink_reparse_points);
845         }
846         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
847                 if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
848                         return supported_features->encrypted_directories != 0;
849                 else
850                         return supported_features->encrypted_files != 0;
851         }
852         return true;
853 }
854
855 /* Given a WIM dentry to extract, build the path to which to extract it, in the
856  * format understood by the callbacks in the apply_operations being used.
857  *
858  * Write the resulting path into @path, which must have room for at least
859  * ctx->ops->max_path characters including the null-terminator.
860  *
861  * Return %true if successful; %false if this WIM dentry doesn't actually need
862  * to be extracted or if the calculated path exceeds ctx->ops->max_path
863  * characters.
864  *
865  * This function clobbers the tmp_list member of @dentry and its ancestors up
866  * until the extraction root.  */
867 static bool
868 build_extraction_path(tchar path[], struct wim_dentry *dentry,
869                       struct apply_ctx *ctx)
870 {
871         size_t path_nchars;
872         LIST_HEAD(ancestor_list);
873         tchar *p = path;
874         const tchar *target_prefix;
875         size_t target_prefix_nchars;
876         struct wim_dentry *d;
877
878         if (dentry->extraction_skipped)
879                 return false;
880
881         path_nchars = ctx->ops->path_prefix_nchars;
882
883         if (ctx->ops->requires_realtarget_in_paths) {
884                 target_prefix        = ctx->realtarget;
885                 target_prefix_nchars = ctx->realtarget_nchars;
886         } else if (ctx->ops->requires_target_in_paths) {
887                 target_prefix        = ctx->target;
888                 target_prefix_nchars = ctx->target_nchars;
889         } else {
890                 target_prefix        = NULL;
891                 target_prefix_nchars = 0;
892         }
893         path_nchars += target_prefix_nchars;
894
895         for (d = dentry; d != ctx->extract_root; d = d->parent) {
896                 path_nchars += d->extraction_name_nchars + 1;
897                 list_add(&d->tmp_list, &ancestor_list);
898         }
899
900         path_nchars++; /* null terminator */
901
902         if (path_nchars > ctx->ops->path_max) {
903                 WARNING("\"%"TS"\": Path too long to extract",
904                         dentry_full_path(dentry));
905                 return false;
906         }
907
908         p = tmempcpy(p, ctx->ops->path_prefix, ctx->ops->path_prefix_nchars);
909         p = tmempcpy(p, target_prefix, target_prefix_nchars);
910         list_for_each_entry(d, &ancestor_list, tmp_list) {
911                 *p++ = ctx->ops->path_separator;
912                 p = tmempcpy(p, d->extraction_name, d->extraction_name_nchars);
913         }
914         *p++ = T('\0');
915         wimlib_assert(p - path == path_nchars);
916         return true;
917 }
918
919 static unsigned
920 get_num_path_components(const tchar *path, tchar path_separator)
921 {
922         unsigned num_components = 0;
923 #ifdef __WIN32__
924         /* Ignore drive letter.  */
925         if (path[0] != L'\0' && path[1] == L':')
926                 path += 2;
927 #endif
928
929         while (*path) {
930                 while (*path == path_separator)
931                         path++;
932                 if (*path)
933                         num_components++;
934                 while (*path && *path != path_separator)
935                         path++;
936         }
937         return num_components;
938 }
939
940 static int
941 extract_multiimage_symlink(const tchar *oldpath, const tchar *newpath,
942                            struct apply_ctx *ctx, struct wim_dentry *dentry)
943 {
944         size_t num_raw_path_components;
945         const struct wim_dentry *d;
946         size_t num_target_path_components;
947         tchar *p;
948         const tchar *p_old;
949         int ret;
950
951         num_raw_path_components = 0;
952         for (d = dentry; d != ctx->extract_root; d = d->parent)
953                 num_raw_path_components++;
954
955         if (ctx->ops->requires_realtarget_in_paths)
956                 num_target_path_components = get_num_path_components(ctx->realtarget,
957                                                                      ctx->ops->path_separator);
958         else if (ctx->ops->requires_target_in_paths)
959                 num_target_path_components = get_num_path_components(ctx->target,
960                                                                      ctx->ops->path_separator);
961         else
962                 num_target_path_components = 0;
963
964         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_MULTI_IMAGE) {
965                 wimlib_assert(num_target_path_components > 0);
966                 num_raw_path_components++;
967                 num_target_path_components--;
968         }
969
970         p_old = oldpath + ctx->ops->path_prefix_nchars;
971 #ifdef __WIN32__
972         if (p_old[0] != L'\0' && p_old[1] == ':')
973                 p_old += 2;
974 #endif
975         while (*p_old == ctx->ops->path_separator)
976                 p_old++;
977         while (--num_target_path_components) {
978                 while (*p_old != ctx->ops->path_separator)
979                         p_old++;
980                 while (*p_old == ctx->ops->path_separator)
981                         p_old++;
982         }
983
984         tchar symlink_target[tstrlen(p_old) + 3 * num_raw_path_components + 1];
985
986         p = &symlink_target[0];
987         while (num_raw_path_components--) {
988                 *p++ = '.';
989                 *p++ = '.';
990                 *p++ = ctx->ops->path_separator;
991         }
992         tstrcpy(p, p_old);
993         DEBUG("Creating symlink \"%"TS"\" => \"%"TS"\"",
994               newpath, symlink_target);
995         ret = ctx->ops->create_symlink(symlink_target, newpath, ctx);
996         if (ret) {
997                 ERROR_WITH_ERRNO("Failed to create symlink "
998                                  "\"%"TS"\" => \"%"TS"\"",
999                                  newpath, symlink_target);
1000         }
1001         return ret;
1002 }
1003
1004 /* Create the "skeleton" of an extracted file or directory.  Don't yet extract
1005  * data streams, reparse data (including symbolic links), timestamps, and
1006  * security descriptors.  Basically, everything that doesn't require reading
1007  * non-metadata resources from the WIM file and isn't delayed until the final
1008  * pass.  */
1009 static int
1010 do_dentry_extract_skeleton(tchar path[], struct wim_dentry *dentry,
1011                            struct apply_ctx *ctx)
1012 {
1013         struct wim_inode *inode = dentry->d_inode;
1014         int ret;
1015         const tchar *oldpath;
1016
1017         if (unlikely(is_linked_extraction(ctx))) {
1018                 struct wim_lookup_table_entry *unnamed_lte;
1019
1020                 unnamed_lte = inode_unnamed_lte_resolved(dentry->d_inode);
1021                 if (unnamed_lte && unnamed_lte->extracted_file) {
1022                         oldpath = unnamed_lte->extracted_file;
1023                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_HARDLINK)
1024                                 goto hardlink;
1025                         else
1026                                 goto symlink;
1027                 }
1028         }
1029
1030         /* Create hard link if this dentry corresponds to an already-extracted
1031          * inode.  */
1032         if (inode->i_extracted_file) {
1033                 oldpath = inode->i_extracted_file;
1034                 goto hardlink;
1035         }
1036
1037         /* Skip symlinks unless they can be extracted as reparse points rather
1038          * than created directly.  */
1039         if (inode_is_symlink(inode) && !ctx->supported_features.reparse_points)
1040                 return 0;
1041
1042         /* Create this file or directory unless it's the extraction root, which
1043          * was already created if necessary.  */
1044         if (dentry != ctx->extract_root) {
1045                 ret = extract_inode(path, ctx, inode);
1046                 if (ret)
1047                         return ret;
1048         }
1049
1050         /* Create empty named data streams.  */
1051         if (can_extract_named_data_streams(ctx)) {
1052                 for (u16 i = 0; i < inode->i_num_ads; i++) {
1053                         file_spec_t file_spec;
1054                         struct wim_ads_entry *entry = &inode->i_ads_entries[i];
1055
1056                         if (!ads_entry_is_named_stream(entry))
1057                                 continue;
1058                         if (entry->lte)
1059                                 continue;
1060                         if (ctx->ops->uses_cookies)
1061                                 file_spec.cookie = inode->extract_cookie;
1062                         else
1063                                 file_spec.path = path;
1064                         ret = ctx->ops->extract_named_stream(file_spec,
1065                                                              entry->stream_name,
1066                                                              entry->stream_name_nbytes / 2,
1067                                                              entry->lte, ctx);
1068                         if (ret) {
1069                                 ERROR_WITH_ERRNO("\"%"TS"\": failed to create "
1070                                                  "empty named data stream",
1071                                                  path);
1072                                 return ret;
1073                         }
1074                 }
1075         }
1076
1077         /* Set file attributes (if supported).  */
1078         ret = extract_file_attributes(path, ctx, dentry, 0);
1079         if (ret)
1080                 return ret;
1081
1082         /* Set or remove file short name (if supported).  */
1083         ret = extract_short_name(path, ctx, dentry);
1084         if (ret)
1085                 return ret;
1086
1087         /* If inode has multiple links and hard links are supported in this
1088          * extraction mode and volume, save the path to the extracted file in
1089          * case it's needed to create a hard link.  */
1090         if (unlikely(is_linked_extraction(ctx))) {
1091                 struct wim_lookup_table_entry *unnamed_lte;
1092
1093                 unnamed_lte = inode_unnamed_lte_resolved(dentry->d_inode);
1094                 if (unnamed_lte) {
1095                         unnamed_lte->extracted_file = TSTRDUP(path);
1096                         if (!unnamed_lte->extracted_file)
1097                                 return WIMLIB_ERR_NOMEM;
1098                 }
1099         } else if (inode->i_nlink > 1 && ctx->supported_features.hard_links) {
1100                 inode->i_extracted_file = TSTRDUP(path);
1101                 if (!inode->i_extracted_file)
1102                         return WIMLIB_ERR_NOMEM;
1103         }
1104         return 0;
1105
1106 symlink:
1107         ret = extract_multiimage_symlink(oldpath, path, ctx, dentry);
1108         if (ret)
1109                 return ret;
1110         dentry->was_hardlinked = 1;
1111         return 0;
1112
1113 hardlink:
1114         ret = extract_hardlink(oldpath, path, ctx);
1115         if (ret)
1116                 return ret;
1117         dentry->was_hardlinked = 1;
1118         return 0;
1119 }
1120
1121 /* This is a wrapper around do_dentry_extract_skeleton() that handles building
1122  * the path, doing short name reordering.  This is also idempotent; dentries
1123  * already processed have skeleton_extracted set and no action is taken.  See
1124  * apply_operations.requires_short_name_reordering for more details about short
1125  * name reordering.  */
1126 static int
1127 dentry_extract_skeleton(struct wim_dentry *dentry, void *_ctx)
1128 {
1129         struct apply_ctx *ctx = _ctx;
1130         tchar path[ctx->ops->path_max];
1131         struct wim_dentry *orig_dentry;
1132         struct wim_dentry *other_dentry;
1133         int ret;
1134
1135         if (dentry->skeleton_extracted)
1136                 return 0;
1137
1138         orig_dentry = NULL;
1139         if (ctx->supported_features.short_names
1140             && ctx->ops->requires_short_name_reordering
1141             && !dentry_has_short_name(dentry)
1142             && !dentry->d_inode->i_dos_name_extracted)
1143         {
1144                 inode_for_each_dentry(other_dentry, dentry->d_inode) {
1145                         if (dentry_has_short_name(other_dentry)
1146                             && !other_dentry->skeleton_extracted
1147                             && other_dentry->in_extraction_tree)
1148                         {
1149                                 DEBUG("Creating %"TS" before %"TS" "
1150                                       "to guarantee correct DOS name extraction",
1151                                       dentry_full_path(other_dentry),
1152                                       dentry_full_path(dentry));
1153                                 orig_dentry = dentry;
1154                                 dentry = other_dentry;
1155                                 break;
1156                         }
1157                 }
1158         }
1159 again:
1160         if (!build_extraction_path(path, dentry, ctx))
1161                 return 0;
1162         ret = do_dentry_extract_skeleton(path, dentry, ctx);
1163         if (ret)
1164                 return ret;
1165
1166         dentry->skeleton_extracted = 1;
1167
1168         if (orig_dentry) {
1169                 dentry = orig_dentry;
1170                 orig_dentry = NULL;
1171                 goto again;
1172         }
1173         dentry->d_inode->i_dos_name_extracted = 1;
1174         return 0;
1175 }
1176
1177 static int
1178 dentry_extract_dir_skeleton(struct wim_dentry *dentry, void *_ctx)
1179 {
1180         if (dentry->d_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
1181                 return dentry_extract_skeleton(dentry, _ctx);
1182         return 0;
1183 }
1184
1185 /* Create a file or directory, then immediately extract all streams.  This
1186  * assumes that WIMLIB_EXTRACT_FLAG_SEQUENTIAL is not specified, since the WIM
1187  * may not be read sequentially by this function.  */
1188 static int
1189 dentry_extract(struct wim_dentry *dentry, void *_ctx)
1190 {
1191         struct apply_ctx *ctx = _ctx;
1192         tchar path[ctx->ops->path_max];
1193         int ret;
1194
1195         ret = dentry_extract_skeleton(dentry, ctx);
1196         if (ret)
1197                 return ret;
1198
1199         if (!build_extraction_path(path, dentry, ctx))
1200                 return 0;
1201
1202         return extract_streams(path, ctx, dentry, NULL, NULL);
1203 }
1204
1205 /* Extract all instances of the stream @lte that are being extracted in this
1206  * call of extract_tree().  @can_seek specifies whether the WIM file descriptor
1207  * is seekable or not (e.g. is a pipe).  If not and the stream needs to be
1208  * extracted multiple times, it is extracted to a temporary file first.
1209  *
1210  * This is intended for use with sequential extraction of a WIM image
1211  * (WIMLIB_EXTRACT_FLAG_SEQUENTIAL specified).  */
1212 static int
1213 extract_stream_instances(struct wim_lookup_table_entry *lte,
1214                          struct apply_ctx *ctx, bool can_seek)
1215 {
1216         struct wim_dentry **lte_dentries;
1217         struct wim_lookup_table_entry *lte_tmp = NULL;
1218         struct wim_lookup_table_entry *lte_override;
1219         tchar *stream_tmp_filename = NULL;
1220         tchar path[ctx->ops->path_max];
1221         unsigned i;
1222         int ret;
1223
1224         if (lte->out_refcnt <= ARRAY_LEN(lte->inline_lte_dentries))
1225                 lte_dentries = lte->inline_lte_dentries;
1226         else
1227                 lte_dentries = lte->lte_dentries;
1228
1229         if (likely(can_seek || lte->out_refcnt < 2)) {
1230                 lte_override = lte;
1231         } else {
1232                 /* Need to extract stream to temporary file.  */
1233                 struct filedes fd;
1234                 int raw_fd;
1235
1236                 stream_tmp_filename = ttempnam(NULL, T("wimlib"));
1237                 if (!stream_tmp_filename) {
1238                         ERROR_WITH_ERRNO("Failed to create temporary filename");
1239                         ret = WIMLIB_ERR_OPEN;
1240                         goto out;
1241                 }
1242
1243                 lte_tmp = memdup(lte, sizeof(struct wim_lookup_table_entry));
1244                 if (!lte_tmp) {
1245                         ret = WIMLIB_ERR_NOMEM;
1246                         goto out_free_stream_tmp_filename;
1247                 }
1248                 lte_tmp->resource_location = RESOURCE_IN_FILE_ON_DISK;
1249                 lte_tmp->file_on_disk = stream_tmp_filename;
1250                 lte_override = lte_tmp;
1251
1252                 raw_fd = topen(stream_tmp_filename,
1253                                O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0600);
1254                 if (raw_fd < 0) {
1255                         ERROR_WITH_ERRNO("Failed to open temporary file");
1256                         ret = WIMLIB_ERR_OPEN;
1257                         goto out_free_lte_tmp;
1258                 }
1259                 filedes_init(&fd, raw_fd);
1260                 ret = extract_wim_resource_to_fd(lte, &fd,
1261                                                  wim_resource_size(lte));
1262                 if (filedes_close(&fd) && !ret)
1263                         ret = WIMLIB_ERR_WRITE;
1264                 if (ret)
1265                         goto out_unlink_stream_tmp_file;
1266         }
1267
1268         /* Extract all instances of the stream, reading either from the stream
1269          * in the WIM file or from the temporary file containing the stream.
1270          * dentry->tmp_flag is used to ensure that each dentry is processed only
1271          * once regardless of how many times this stream appears in the streams
1272          * of the corresponding inode.  */
1273         for (i = 0; i < lte->out_refcnt; i++) {
1274                 struct wim_dentry *dentry = lte_dentries[i];
1275
1276                 if (dentry->tmp_flag)
1277                         continue;
1278                 if (!build_extraction_path(path, dentry, ctx))
1279                         continue;
1280                 ret = extract_streams(path, ctx, dentry,
1281                                       lte, lte_override);
1282                 if (ret)
1283                         goto out_clear_tmp_flags;
1284                 dentry->tmp_flag = 1;
1285         }
1286         ret = 0;
1287 out_clear_tmp_flags:
1288         for (i = 0; i < lte->out_refcnt; i++)
1289                 lte_dentries[i]->tmp_flag = 0;
1290 out_unlink_stream_tmp_file:
1291         if (stream_tmp_filename)
1292                 tunlink(stream_tmp_filename);
1293 out_free_lte_tmp:
1294         FREE(lte_tmp);
1295 out_free_stream_tmp_filename:
1296         FREE(stream_tmp_filename);
1297 out:
1298         return ret;
1299 }
1300
1301 /* Extracts a list of streams (ctx.stream_list), assuming that the directory
1302  * structure and empty files were already created.  This relies on the
1303  * per-`struct wim_lookup_table_entry' list of dentries that reference each
1304  * stream that was constructed earlier.  Streams are extracted exactly in the
1305  * order of the stream list; however, unless the WIM's file descriptor is
1306  * detected to be non-seekable, streams may be read from the WIM file more than
1307  * one time if multiple copies need to be extracted.  */
1308 static int
1309 extract_stream_list(struct apply_ctx *ctx)
1310 {
1311         struct wim_lookup_table_entry *lte;
1312         bool can_seek;
1313         int ret;
1314
1315         can_seek = (lseek(ctx->wim->in_fd.fd, 0, SEEK_CUR) != -1);
1316         list_for_each_entry(lte, &ctx->stream_list, extraction_list) {
1317                 ret = extract_stream_instances(lte, ctx, can_seek);
1318                 if (ret)
1319                         return ret;
1320         }
1321         return 0;
1322 }
1323
1324 #define PWM_ALLOW_WIM_HDR 0x00001
1325 #define PWM_SILENT_EOF    0x00002
1326
1327 /* Read the header from a stream in a pipable WIM.  */
1328 static int
1329 read_pwm_stream_header(WIMStruct *pwm, struct wim_lookup_table_entry *lte,
1330                        int flags, struct wim_header_disk *hdr_ret)
1331 {
1332         union {
1333                 struct pwm_stream_hdr stream_hdr;
1334                 struct wim_header_disk pwm_hdr;
1335         } buf;
1336         int ret;
1337
1338         ret = full_read(&pwm->in_fd, &buf.stream_hdr, sizeof(buf.stream_hdr));
1339         if (ret)
1340                 goto read_error;
1341
1342         if ((flags & PWM_ALLOW_WIM_HDR) && buf.stream_hdr.magic == PWM_MAGIC) {
1343                 BUILD_BUG_ON(sizeof(buf.pwm_hdr) < sizeof(buf.stream_hdr));
1344                 ret = full_read(&pwm->in_fd, &buf.stream_hdr + 1,
1345                                 sizeof(buf.pwm_hdr) - sizeof(buf.stream_hdr));
1346
1347                 if (ret)
1348                         goto read_error;
1349                 lte->resource_location = RESOURCE_NONEXISTENT;
1350                 memcpy(hdr_ret, &buf.pwm_hdr, sizeof(buf.pwm_hdr));
1351                 return 0;
1352         }
1353
1354         if (buf.stream_hdr.magic != PWM_STREAM_MAGIC) {
1355                 ERROR("Data read on pipe is invalid (expected stream header).");
1356                 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1357         }
1358
1359         lte->resource_entry.original_size = le64_to_cpu(buf.stream_hdr.uncompressed_size);
1360         copy_hash(lte->hash, buf.stream_hdr.hash);
1361         lte->resource_entry.flags = le32_to_cpu(buf.stream_hdr.flags);
1362         lte->resource_entry.offset = pwm->in_fd.offset;
1363         lte->resource_location = RESOURCE_IN_WIM;
1364         lte->wim = pwm;
1365         if (lte->resource_entry.flags & WIM_RESHDR_FLAG_COMPRESSED) {
1366                 lte->compression_type = pwm->compression_type;
1367                 lte->resource_entry.size = 0;
1368         } else {
1369                 lte->compression_type = WIMLIB_COMPRESSION_TYPE_NONE;
1370                 lte->resource_entry.size = lte->resource_entry.original_size;
1371         }
1372         lte->is_pipable = 1;
1373         return 0;
1374
1375 read_error:
1376         if (ret != WIMLIB_ERR_UNEXPECTED_END_OF_FILE || !(flags & PWM_SILENT_EOF))
1377                 ERROR_WITH_ERRNO("Error reading pipable WIM from pipe");
1378         return ret;
1379 }
1380
1381 /* Skip over an unneeded stream in a pipable WIM being read from a pipe.  */
1382 static int
1383 skip_pwm_stream(struct wim_lookup_table_entry *lte)
1384 {
1385         return read_partial_wim_resource(lte, wim_resource_size(lte),
1386                                          NULL, NULL,
1387                                          WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY,
1388                                          0);
1389 }
1390
1391 static int
1392 extract_streams_from_pipe(struct apply_ctx *ctx)
1393 {
1394         struct wim_lookup_table_entry *found_lte;
1395         struct wim_lookup_table_entry *needed_lte;
1396         struct wim_lookup_table *lookup_table;
1397         struct wim_header_disk pwm_hdr;
1398         int ret;
1399         int pwm_flags;
1400
1401         ret = WIMLIB_ERR_NOMEM;
1402         found_lte = new_lookup_table_entry();
1403         if (!found_lte)
1404                 goto out;
1405
1406         lookup_table = ctx->wim->lookup_table;
1407         pwm_flags = PWM_ALLOW_WIM_HDR;
1408         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1409                 pwm_flags |= PWM_SILENT_EOF;
1410         memcpy(ctx->progress.extract.guid, ctx->wim->hdr.guid, WIM_GID_LEN);
1411         ctx->progress.extract.part_number = ctx->wim->hdr.part_number;
1412         ctx->progress.extract.total_parts = ctx->wim->hdr.total_parts;
1413         if (ctx->progress_func)
1414                 ctx->progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1415                                    &ctx->progress);
1416         while (ctx->num_streams_remaining) {
1417                 ret = read_pwm_stream_header(ctx->wim, found_lte, pwm_flags,
1418                                              &pwm_hdr);
1419                 if (ret) {
1420                         if (ret == WIMLIB_ERR_UNEXPECTED_END_OF_FILE &&
1421                             (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1422                         {
1423                                 goto resume_done;
1424                         }
1425                         goto out_free_found_lte;
1426                 }
1427
1428                 if ((found_lte->resource_location != RESOURCE_NONEXISTENT)
1429                     && !(found_lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)
1430                     && (needed_lte = lookup_resource(lookup_table, found_lte->hash))
1431                     && (needed_lte->out_refcnt))
1432                 {
1433                         copy_resource_entry(&needed_lte->resource_entry,
1434                                             &found_lte->resource_entry);
1435                         needed_lte->resource_location = found_lte->resource_location;
1436                         needed_lte->wim               = found_lte->wim;
1437                         needed_lte->compression_type  = found_lte->compression_type;
1438                         needed_lte->is_pipable        = found_lte->is_pipable;
1439
1440                         ret = extract_stream_instances(needed_lte, ctx, false);
1441                         if (ret)
1442                                 goto out_free_found_lte;
1443                         ctx->num_streams_remaining--;
1444                 } else if (found_lte->resource_location != RESOURCE_NONEXISTENT) {
1445                         ret = skip_pwm_stream(found_lte);
1446                         if (ret)
1447                                 goto out_free_found_lte;
1448                 } else {
1449                         u16 part_number = le16_to_cpu(pwm_hdr.part_number);
1450                         u16 total_parts = le16_to_cpu(pwm_hdr.total_parts);
1451
1452                         if (part_number != ctx->progress.extract.part_number ||
1453                             total_parts != ctx->progress.extract.total_parts ||
1454                             memcmp(pwm_hdr.guid, ctx->progress.extract.guid,
1455                                    WIM_GID_LEN))
1456                         {
1457                                 ctx->progress.extract.part_number = part_number;
1458                                 ctx->progress.extract.total_parts = total_parts;
1459                                 memcpy(ctx->progress.extract.guid,
1460                                        pwm_hdr.guid, WIM_GID_LEN);
1461                                 if (ctx->progress_func) {
1462                                         ctx->progress_func(
1463                                                 WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1464                                                            &ctx->progress);
1465                                 }
1466
1467                         }
1468                 }
1469         }
1470         ret = 0;
1471 out_free_found_lte:
1472         free_lookup_table_entry(found_lte);
1473 out:
1474         return ret;
1475
1476 resume_done:
1477         /* TODO */
1478         return 0;
1479 }
1480
1481 /* Finish extracting a file, directory, or symbolic link by setting file
1482  * security and timestamps.  */
1483 static int
1484 dentry_extract_final(struct wim_dentry *dentry, void *_ctx)
1485 {
1486         struct apply_ctx *ctx = _ctx;
1487         int ret;
1488         tchar path[ctx->ops->path_max];
1489
1490         if (!build_extraction_path(path, dentry, ctx))
1491                 return 0;
1492
1493         ret = extract_security(path, ctx, dentry);
1494         if (ret)
1495                 return ret;
1496
1497         if (ctx->ops->requires_final_set_attributes_pass) {
1498                 /* Set file attributes (if supported).  */
1499                 ret = extract_file_attributes(path, ctx, dentry, 1);
1500                 if (ret)
1501                         return ret;
1502         }
1503
1504         return extract_timestamps(path, ctx, dentry);
1505 }
1506
1507 /*
1508  * Extract a WIM dentry to standard output.
1509  *
1510  * This obviously doesn't make sense in all cases.  We return an error if the
1511  * dentry does not correspond to a regular file.  Otherwise we extract the
1512  * unnamed data stream only.
1513  */
1514 static int
1515 extract_dentry_to_stdout(struct wim_dentry *dentry)
1516 {
1517         int ret = 0;
1518         if (dentry->d_inode->i_attributes & (FILE_ATTRIBUTE_REPARSE_POINT |
1519                                              FILE_ATTRIBUTE_DIRECTORY))
1520         {
1521                 ERROR("\"%"TS"\" is not a regular file and therefore cannot be "
1522                       "extracted to standard output", dentry_full_path(dentry));
1523                 ret = WIMLIB_ERR_NOT_A_REGULAR_FILE;
1524         } else {
1525                 struct wim_lookup_table_entry *lte;
1526
1527                 lte = inode_unnamed_lte_resolved(dentry->d_inode);
1528                 if (lte) {
1529                         struct filedes _stdout;
1530                         filedes_init(&_stdout, STDOUT_FILENO);
1531                         ret = extract_wim_resource_to_fd(lte, &_stdout,
1532                                                          wim_resource_size(lte));
1533                 }
1534         }
1535         return ret;
1536 }
1537
1538 #ifdef __WIN32__
1539 static const utf16lechar replacement_char = cpu_to_le16(0xfffd);
1540 #else
1541 static const utf16lechar replacement_char = cpu_to_le16('?');
1542 #endif
1543
1544 static bool
1545 file_name_valid(utf16lechar *name, size_t num_chars, bool fix)
1546 {
1547         size_t i;
1548
1549         if (num_chars == 0)
1550                 return true;
1551         for (i = 0; i < num_chars; i++) {
1552                 switch (name[i]) {
1553         #ifdef __WIN32__
1554                 case cpu_to_le16('\\'):
1555                 case cpu_to_le16(':'):
1556                 case cpu_to_le16('*'):
1557                 case cpu_to_le16('?'):
1558                 case cpu_to_le16('"'):
1559                 case cpu_to_le16('<'):
1560                 case cpu_to_le16('>'):
1561                 case cpu_to_le16('|'):
1562         #endif
1563                 case cpu_to_le16('/'):
1564                 case cpu_to_le16('\0'):
1565                         if (fix)
1566                                 name[i] = replacement_char;
1567                         else
1568                                 return false;
1569                 }
1570         }
1571
1572 #ifdef __WIN32__
1573         if (name[num_chars - 1] == cpu_to_le16(' ') ||
1574             name[num_chars - 1] == cpu_to_le16('.'))
1575         {
1576                 if (fix)
1577                         name[num_chars - 1] = replacement_char;
1578                 else
1579                         return false;
1580         }
1581 #endif
1582         return true;
1583 }
1584
1585 static bool
1586 dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
1587 {
1588         const utf16lechar *file_name = dentry->file_name;
1589         return file_name != NULL &&
1590                 file_name[0] == cpu_to_le16('.') &&
1591                 (file_name[1] == cpu_to_le16('\0') ||
1592                  (file_name[1] == cpu_to_le16('.') &&
1593                   file_name[2] == cpu_to_le16('\0')));
1594 }
1595
1596 static int
1597 dentry_mark_skipped(struct wim_dentry *dentry, void *_ignore)
1598 {
1599         dentry->extraction_skipped = 1;
1600         return 0;
1601 }
1602
1603 /*
1604  * dentry_calculate_extraction_path-
1605  *
1606  * Calculate the actual filename component at which a WIM dentry will be
1607  * extracted, handling invalid filenames "properly".
1608  *
1609  * dentry->extraction_name usually will be set the same as dentry->file_name (on
1610  * UNIX, converted into the platform's multibyte encoding).  However, if the
1611  * file name contains characters that are not valid on the current platform or
1612  * has some other format that is not valid, leave dentry->extraction_name as
1613  * NULL and set dentry->extraction_skipped to indicate that this dentry should
1614  * not be extracted, unless the appropriate flag
1615  * WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES is set in the extract flags, in
1616  * which case a substitute filename will be created and set instead.
1617  *
1618  * Conflicts with case-insensitive names on Windows are handled similarly; see
1619  * below.
1620  */
1621 static int
1622 dentry_calculate_extraction_path(struct wim_dentry *dentry, void *_args)
1623 {
1624         struct apply_ctx *ctx = _args;
1625         int ret;
1626
1627         dentry->in_extraction_tree = 1;
1628
1629         if (dentry == ctx->extract_root || dentry->extraction_skipped)
1630                 return 0;
1631
1632         if (!dentry_is_supported(dentry, &ctx->supported_features))
1633                 goto skip_dentry;
1634
1635         if (dentry_is_dot_or_dotdot(dentry)) {
1636                 /* WIM files shouldn't contain . or .. entries.  But if they are
1637                  * there, don't attempt to extract them. */
1638                 WARNING("Skipping extraction of unexpected . or .. file "
1639                         "\"%"TS"\"", dentry_full_path(dentry));
1640                 goto skip_dentry;
1641         }
1642
1643 #ifdef __WIN32__
1644         if (!ctx->ops->supports_case_sensitive_filenames)
1645         {
1646                 struct wim_dentry *other;
1647                 list_for_each_entry(other, &dentry->case_insensitive_conflict_list,
1648                                     case_insensitive_conflict_list)
1649                 {
1650                         if (ctx->extract_flags &
1651                             WIMLIB_EXTRACT_FLAG_ALL_CASE_CONFLICTS) {
1652                                 WARNING("\"%"TS"\" has the same "
1653                                         "case-insensitive name as "
1654                                         "\"%"TS"\"; extracting "
1655                                         "dummy name instead",
1656                                         dentry_full_path(dentry),
1657                                         dentry_full_path(other));
1658                                 goto out_replace;
1659                         } else {
1660                                 WARNING("Not extracting \"%"TS"\": "
1661                                         "has same case-insensitive "
1662                                         "name as \"%"TS"\"",
1663                                         dentry_full_path(dentry),
1664                                         dentry_full_path(other));
1665                                 goto skip_dentry;
1666                         }
1667                 }
1668         }
1669 #else   /* __WIN32__ */
1670         wimlib_assert(ctx->ops->supports_case_sensitive_filenames);
1671 #endif  /* !__WIN32__ */
1672
1673         if (file_name_valid(dentry->file_name, dentry->file_name_nbytes / 2, false)) {
1674 #ifdef __WIN32__
1675                 dentry->extraction_name = dentry->file_name;
1676                 dentry->extraction_name_nchars = dentry->file_name_nbytes / 2;
1677                 return 0;
1678 #else
1679                 return utf16le_to_tstr(dentry->file_name,
1680                                        dentry->file_name_nbytes,
1681                                        &dentry->extraction_name,
1682                                        &dentry->extraction_name_nchars);
1683 #endif
1684         } else {
1685                 if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES)
1686                 {
1687                         WARNING("\"%"TS"\" has an invalid filename "
1688                                 "that is not supported on this platform; "
1689                                 "extracting dummy name instead",
1690                                 dentry_full_path(dentry));
1691                         goto out_replace;
1692                 } else {
1693                         WARNING("Not extracting \"%"TS"\": has an invalid filename "
1694                                 "that is not supported on this platform",
1695                                 dentry_full_path(dentry));
1696                         goto skip_dentry;
1697                 }
1698         }
1699
1700 out_replace:
1701         {
1702                 utf16lechar utf16_name_copy[dentry->file_name_nbytes / 2];
1703
1704                 memcpy(utf16_name_copy, dentry->file_name, dentry->file_name_nbytes);
1705                 file_name_valid(utf16_name_copy, dentry->file_name_nbytes / 2, true);
1706
1707                 tchar *tchar_name;
1708                 size_t tchar_nchars;
1709         #ifdef __WIN32__
1710                 tchar_name = utf16_name_copy;
1711                 tchar_nchars = dentry->file_name_nbytes / 2;
1712         #else
1713                 ret = utf16le_to_tstr(utf16_name_copy,
1714                                       dentry->file_name_nbytes,
1715                                       &tchar_name, &tchar_nchars);
1716                 if (ret)
1717                         return ret;
1718         #endif
1719                 size_t fixed_name_num_chars = tchar_nchars;
1720                 tchar fixed_name[tchar_nchars + 50];
1721
1722                 tmemcpy(fixed_name, tchar_name, tchar_nchars);
1723                 fixed_name_num_chars += tsprintf(fixed_name + tchar_nchars,
1724                                                  T(" (invalid filename #%lu)"),
1725                                                  ++ctx->invalid_sequence);
1726         #ifndef __WIN32__
1727                 FREE(tchar_name);
1728         #endif
1729                 dentry->extraction_name = memdup(fixed_name,
1730                                                  2 * fixed_name_num_chars + 2);
1731                 if (!dentry->extraction_name)
1732                         return WIMLIB_ERR_NOMEM;
1733                 dentry->extraction_name_nchars = fixed_name_num_chars;
1734         }
1735         return 0;
1736
1737 skip_dentry:
1738         for_dentry_in_tree(dentry, dentry_mark_skipped, NULL);
1739         return 0;
1740 }
1741
1742 /* Clean up dentry and inode structure after extraction.  */
1743 static int
1744 dentry_reset_needs_extraction(struct wim_dentry *dentry, void *_ignore)
1745 {
1746         struct wim_inode *inode = dentry->d_inode;
1747
1748         dentry->in_extraction_tree = 0;
1749         dentry->extraction_skipped = 0;
1750         dentry->was_hardlinked = 0;
1751         dentry->skeleton_extracted = 0;
1752         inode->i_visited = 0;
1753         FREE(inode->i_extracted_file);
1754         inode->i_extracted_file = NULL;
1755         inode->i_dos_name_extracted = 0;
1756         if ((void*)dentry->extraction_name != (void*)dentry->file_name)
1757                 FREE(dentry->extraction_name);
1758         dentry->extraction_name = NULL;
1759         return 0;
1760 }
1761
1762 /* Tally features necessary to extract a dentry and the corresponding inode.  */
1763 static int
1764 dentry_tally_features(struct wim_dentry *dentry, void *_features)
1765 {
1766         struct wim_features *features = _features;
1767         struct wim_inode *inode = dentry->d_inode;
1768
1769         if (inode->i_attributes & FILE_ATTRIBUTE_ARCHIVE)
1770                 features->archive_files++;
1771         if (inode->i_attributes & FILE_ATTRIBUTE_HIDDEN)
1772                 features->hidden_files++;
1773         if (inode->i_attributes & FILE_ATTRIBUTE_SYSTEM)
1774                 features->system_files++;
1775         if (inode->i_attributes & FILE_ATTRIBUTE_COMPRESSED)
1776                 features->compressed_files++;
1777         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
1778                 if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
1779                         features->encrypted_directories++;
1780                 else
1781                         features->encrypted_files++;
1782         }
1783         if (inode->i_attributes & FILE_ATTRIBUTE_NOT_CONTENT_INDEXED)
1784                 features->not_context_indexed_files++;
1785         if (inode->i_attributes & FILE_ATTRIBUTE_SPARSE_FILE)
1786                 features->sparse_files++;
1787         if (inode_has_named_stream(inode))
1788                 features->named_data_streams++;
1789         if (inode->i_visited)
1790                 features->hard_links++;
1791         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1792                 features->reparse_points++;
1793                 if (inode_is_symlink(inode))
1794                         features->symlink_reparse_points++;
1795                 else
1796                         features->other_reparse_points++;
1797         }
1798         if (inode->i_security_id != -1)
1799                 features->security_descriptors++;
1800         if (dentry->short_name_nbytes)
1801                 features->short_names++;
1802         if (inode_has_unix_data(inode))
1803                 features->unix_data++;
1804         inode->i_visited = 1;
1805         return 0;
1806 }
1807
1808 /* Tally the features necessary to extract a dentry tree.  */
1809 static void
1810 dentry_tree_get_features(struct wim_dentry *root, struct wim_features *features)
1811 {
1812         memset(features, 0, sizeof(struct wim_features));
1813         for_dentry_in_tree(root, dentry_tally_features, features);
1814         dentry_tree_clear_inode_visited(root);
1815 }
1816
1817 static u32
1818 compute_supported_attributes_mask(const struct wim_features *supported_features)
1819 {
1820         u32 mask = ~(u32)0;
1821
1822         if (!supported_features->archive_files)
1823                 mask &= ~FILE_ATTRIBUTE_ARCHIVE;
1824
1825         if (!supported_features->hidden_files)
1826                 mask &= ~FILE_ATTRIBUTE_HIDDEN;
1827
1828         if (!supported_features->system_files)
1829                 mask &= ~FILE_ATTRIBUTE_SYSTEM;
1830
1831         if (!supported_features->not_context_indexed_files)
1832                 mask &= ~FILE_ATTRIBUTE_NOT_CONTENT_INDEXED;
1833
1834         if (!supported_features->compressed_files)
1835                 mask &= ~FILE_ATTRIBUTE_COMPRESSED;
1836
1837         if (!supported_features->sparse_files)
1838                 mask &= ~FILE_ATTRIBUTE_SPARSE_FILE;
1839
1840         if (!supported_features->reparse_points)
1841                 mask &= ~FILE_ATTRIBUTE_REPARSE_POINT;
1842
1843         return mask;
1844 }
1845
1846 static int
1847 do_feature_check(const struct wim_features *required_features,
1848                  const struct wim_features *supported_features,
1849                  int extract_flags,
1850                  const struct apply_operations *ops,
1851                  const tchar *wim_source_path)
1852 {
1853         const tchar *loc;
1854         const tchar *mode = T("this extraction mode");
1855
1856         if (wim_source_path[0] == '\0')
1857                 loc = T("the WIM image");
1858         else
1859                 loc = wim_source_path;
1860
1861         /* We're an archive program, so theoretically we can do what we want
1862          * with FILE_ATTRIBUTE_ARCHIVE (which is a dumb flag anyway).  Don't
1863          * bother the user about it.  */
1864 #if 0
1865         if (required_features->archive_files && !supported_features->archive_files)
1866         {
1867                 WARNING(
1868           "%lu files in %"TS" are marked as archived, but this attribute\n"
1869 "          is not supported in %"TS".",
1870                         required_features->archive_files, loc, mode);
1871         }
1872 #endif
1873
1874         if (required_features->hidden_files && !supported_features->hidden_files)
1875         {
1876                 WARNING(
1877           "%lu files in %"TS" are marked as hidden, but this\n"
1878 "          attribute is not supported in %"TS".",
1879                         required_features->hidden_files, loc, mode);
1880         }
1881
1882         if (required_features->system_files && !supported_features->system_files)
1883         {
1884                 WARNING(
1885           "%lu files in %"TS" are marked as system files,\n"
1886 "          but this attribute is not supported in %"TS".",
1887                         required_features->system_files, loc, mode);
1888         }
1889
1890         if (required_features->compressed_files && !supported_features->compressed_files)
1891         {
1892                 WARNING(
1893           "%lu files in %"TS" are marked as being transparently\n"
1894 "          compressed, but transparent compression is not supported in\n"
1895 "          %"TS".  These files will be extracted as uncompressed.",
1896                         required_features->compressed_files, loc, mode);
1897         }
1898
1899         if (required_features->encrypted_files && !supported_features->encrypted_files)
1900         {
1901                 WARNING(
1902           "%lu files in %"TS" are marked as being encrypted,\n"
1903 "           but encryption is not supported in %"TS".  These files\n"
1904 "           will not be extracted.",
1905                         required_features->encrypted_files, loc, mode);
1906         }
1907
1908         if (required_features->encrypted_directories &&
1909             !supported_features->encrypted_directories)
1910         {
1911                 WARNING(
1912           "%lu directories in %"TS" are marked as being encrypted,\n"
1913 "           but encryption is not supported in %"TS".\n"
1914 "           These directories will be extracted as unencrypted.",
1915                         required_features->encrypted_directories, loc, mode);
1916         }
1917
1918         if (required_features->not_context_indexed_files &&
1919             !supported_features->not_context_indexed_files)
1920         {
1921                 WARNING(
1922           "%lu files in %"TS" are marked as not content indexed,\n"
1923 "          but this attribute is not supported in %"TS".",
1924                         required_features->not_context_indexed_files, loc, mode);
1925         }
1926
1927         if (required_features->sparse_files && !supported_features->sparse_files)
1928         {
1929                 WARNING(
1930           "%lu files in %"TS" are marked as sparse, but creating\n"
1931 "           sparse files is not supported in %"TS".  These files\n"
1932 "           will be extracted as non-sparse.",
1933                         required_features->sparse_files, loc, mode);
1934         }
1935
1936         if (required_features->named_data_streams &&
1937             !supported_features->named_data_streams)
1938         {
1939                 WARNING(
1940           "%lu files in %"TS" contain one or more alternate (named)\n"
1941 "          data streams, which are not supported in %"TS".\n"
1942 "          Alternate data streams will NOT be extracted.",
1943                         required_features->named_data_streams, loc, mode);
1944         }
1945
1946         if (unlikely(extract_flags & (WIMLIB_EXTRACT_FLAG_HARDLINK |
1947                                       WIMLIB_EXTRACT_FLAG_SYMLINK)) &&
1948             required_features->named_data_streams &&
1949             supported_features->named_data_streams)
1950         {
1951                 WARNING(
1952           "%lu files in %"TS" contain one or more alternate (named)\n"
1953 "          data streams, which are not supported in linked extraction mode.\n"
1954 "          Alternate data streams will NOT be extracted.",
1955                         required_features->named_data_streams, loc);
1956         }
1957
1958         if (required_features->hard_links && !supported_features->hard_links)
1959         {
1960                 WARNING(
1961           "%lu files in %"TS" are hard links, but hard links are\n"
1962 "          not supported in %"TS".  Hard links will be extracted as\n"
1963 "          duplicate copies of the linked files.",
1964                         required_features->hard_links, loc, mode);
1965         }
1966
1967         if (required_features->reparse_points && !supported_features->reparse_points)
1968         {
1969                 if (supported_features->symlink_reparse_points) {
1970                         if (required_features->other_reparse_points) {
1971                                 WARNING(
1972           "%lu files in %"TS" are reparse points that are neither\n"
1973 "          symbolic links nor junction points and are not supported in\n"
1974 "          %"TS".  These reparse points will not be extracted.",
1975                                         required_features->other_reparse_points, loc,
1976                                         mode);
1977                         }
1978                 } else {
1979                         WARNING(
1980           "%lu files in %"TS" are reparse points, which are\n"
1981 "          not supported in %"TS" and will not be extracted.",
1982                                 required_features->reparse_points, loc, mode);
1983                 }
1984         }
1985
1986         if (required_features->security_descriptors &&
1987             !supported_features->security_descriptors)
1988         {
1989                 WARNING(
1990           "%lu files in %"TS" have Windows NT security descriptors,\n"
1991 "          but extracting security descriptors is not supported in\n"
1992 "          %"TS".  No security descriptors will be extracted.",
1993                         required_features->security_descriptors, loc, mode);
1994         }
1995
1996         if (required_features->short_names && !supported_features->short_names)
1997         {
1998                 WARNING(
1999           "%lu files in %"TS" have short (DOS) names, but\n"
2000 "          extracting short names is not supported in %"TS".\n"
2001 "          Short names will not be extracted.\n",
2002                         required_features->short_names, loc, mode);
2003         }
2004
2005         if ((extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) &&
2006             required_features->unix_data && !supported_features->unix_data)
2007         {
2008                 ERROR("Extracting UNIX data is not supported in %"TS, mode);
2009                 return WIMLIB_ERR_UNSUPPORTED;
2010         }
2011         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES) &&
2012             required_features->short_names && !supported_features->short_names)
2013         {
2014                 ERROR("Extracting short names is not supported in %"TS"", mode);
2015                 return WIMLIB_ERR_UNSUPPORTED;
2016         }
2017         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS) &&
2018             !ops->set_timestamps)
2019         {
2020                 ERROR("Extracting timestamps is not supported in %"TS"", mode);
2021                 return WIMLIB_ERR_UNSUPPORTED;
2022         }
2023         if (((extract_flags & (WIMLIB_EXTRACT_FLAG_STRICT_ACLS |
2024                                WIMLIB_EXTRACT_FLAG_UNIX_DATA))
2025              == WIMLIB_EXTRACT_FLAG_STRICT_ACLS) &&
2026             required_features->security_descriptors &&
2027             !supported_features->security_descriptors)
2028         {
2029                 ERROR("Extracting security descriptors is not supported in %"TS, mode);
2030                 return WIMLIB_ERR_UNSUPPORTED;
2031         }
2032
2033         if ((extract_flags & WIMLIB_EXTRACT_FLAG_HARDLINK) &&
2034             !supported_features->hard_links)
2035         {
2036                 ERROR("Hard link extraction mode requested, but "
2037                       "%"TS" does not support hard links!", mode);
2038                 return WIMLIB_ERR_UNSUPPORTED;
2039         }
2040
2041         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS) &&
2042             required_features->symlink_reparse_points &&
2043             !(supported_features->symlink_reparse_points ||
2044               supported_features->reparse_points))
2045         {
2046                 ERROR("Extracting symbolic links is not supported in %"TS, mode);
2047                 return WIMLIB_ERR_UNSUPPORTED;
2048         }
2049
2050         if ((extract_flags & WIMLIB_EXTRACT_FLAG_SYMLINK) &&
2051             !supported_features->symlink_reparse_points)
2052         {
2053                 ERROR("Symbolic link extraction mode requested, but "
2054                       "%"TS" does not support symbolic "
2055                       "links!", mode);
2056                 return WIMLIB_ERR_UNSUPPORTED;
2057         }
2058         return 0;
2059 }
2060
2061 static void
2062 do_extract_warnings(struct apply_ctx *ctx)
2063 {
2064         if (ctx->partial_security_descriptors == 0 &&
2065             ctx->no_security_descriptors == 0)
2066                 return;
2067
2068         WARNING("Extraction of \"%"TS"\" complete, but with one or more warnings:",
2069                 ctx->target);
2070         if (ctx->partial_security_descriptors != 0) {
2071                 WARNING("- Could only partially set the security descriptor\n"
2072                         "            on %lu files or directories.",
2073                         ctx->partial_security_descriptors);
2074         }
2075         if (ctx->no_security_descriptors != 0) {
2076                 WARNING("- Could not set security descriptor at all\n"
2077                         "            on %lu files or directories.",
2078                         ctx->no_security_descriptors);
2079         }
2080 #ifdef __WIN32__
2081         WARNING("To fully restore all security descriptors, run the program\n"
2082                 "          with Administrator rights.");
2083 #endif
2084 }
2085
2086 /*
2087  * extract_tree - Extract a file or directory tree from the currently selected
2088  *                WIM image.
2089  *
2090  * @wim:        WIMStruct for the WIM file, with the desired image selected
2091  *              (as wim->current_image).
2092  *
2093  * @wim_source_path:
2094  *              "Canonical" (i.e. no leading or trailing slashes, path
2095  *              separators WIM_PATH_SEPARATOR) path inside the WIM image to
2096  *              extract.  An empty string means the full image.
2097  *
2098  * @target:
2099  *              Filesystem path to extract the file or directory tree to.
2100  *              (Or, with WIMLIB_EXTRACT_FLAG_NTFS: the name of a NTFS volume.)
2101  *
2102  * @extract_flags:
2103  *              WIMLIB_EXTRACT_FLAG_*.  Also, the private flag
2104  *              WIMLIB_EXTRACT_FLAG_MULTI_IMAGE will be set if this is being
2105  *              called through wimlib_extract_image() with WIMLIB_ALL_IMAGES as
2106  *              the image.
2107  *
2108  * @progress_func:
2109  *              If non-NULL, progress function for the extraction.  The messages
2110  *              that may be sent in this function are:
2111  *
2112  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN or
2113  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN;
2114  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN;
2115  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END;
2116  *              WIMLIB_PROGRESS_MSG_EXTRACT_DENTRY;
2117  *              WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS;
2118  *              WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS;
2119  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END or
2120  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END.
2121  *
2122  * Returns 0 on success; a positive WIMLIB_ERR_* code on failure.
2123  */
2124 static int
2125 extract_tree(WIMStruct *wim, const tchar *wim_source_path, const tchar *target,
2126              int extract_flags, wimlib_progress_func_t progress_func)
2127 {
2128         struct wim_dentry *root;
2129         struct wim_features required_features;
2130         struct apply_ctx ctx;
2131         int ret;
2132         struct wim_lookup_table_entry *lte;
2133
2134         /* Start initializing the apply_ctx.  */
2135         memset(&ctx, 0, sizeof(struct apply_ctx));
2136         ctx.wim = wim;
2137         ctx.extract_flags = extract_flags;
2138         ctx.target = target;
2139         ctx.target_nchars = tstrlen(target);
2140         ctx.progress_func = progress_func;
2141         if (progress_func) {
2142                 ctx.progress.extract.wimfile_name = wim->filename;
2143                 ctx.progress.extract.image = wim->current_image;
2144                 ctx.progress.extract.extract_flags = (extract_flags &
2145                                                       WIMLIB_EXTRACT_MASK_PUBLIC);
2146                 ctx.progress.extract.image_name = wimlib_get_image_name(wim,
2147                                                                         wim->current_image);
2148                 ctx.progress.extract.extract_root_wim_source_path = wim_source_path;
2149                 ctx.progress.extract.target = target;
2150         }
2151         INIT_LIST_HEAD(&ctx.stream_list);
2152
2153         /* Translate the path to extract into the corresponding
2154          * `struct wim_dentry', which will be the root of the
2155          * "dentry tree" to extract.  */
2156         root = get_dentry(wim, wim_source_path);
2157         if (!root) {
2158                 ERROR("Path \"%"TS"\" does not exist in WIM image %d",
2159                       wim_source_path, wim->current_image);
2160                 ret = WIMLIB_ERR_PATH_DOES_NOT_EXIST;
2161                 goto out;
2162         }
2163
2164         ctx.extract_root = root;
2165
2166         /* Select the appropriate apply_operations based on the
2167          * platform and extract_flags.  */
2168 #ifdef __WIN32__
2169         ctx.ops = &win32_apply_ops;
2170 #else
2171         ctx.ops = &unix_apply_ops;
2172 #endif
2173
2174 #ifdef WITH_NTFS_3G
2175         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS)
2176                 ctx.ops = &ntfs_3g_apply_ops;
2177 #endif
2178
2179         /* Call the start_extract() callback.  This gives the apply_operations
2180          * implementation a chance to do any setup needed to access the volume.
2181          * Furthermore, it's expected to set the supported features of this
2182          * extraction mode (ctx.supported_features), which are determined at
2183          * runtime as they may vary depending on the actual volume.  These
2184          * features are then compared with the actual features extracting this
2185          * dentry tree requires.  Some mismatches will merely produce warnings
2186          * and the unsupported data will be ignored; others will produce errors.
2187          */
2188         ret = ctx.ops->start_extract(target, &ctx);
2189         if (ret)
2190                 goto out;
2191
2192         dentry_tree_get_features(root, &required_features);
2193         ret = do_feature_check(&required_features, &ctx.supported_features,
2194                                extract_flags, ctx.ops, wim_source_path);
2195         if (ret)
2196                 goto out_finish_or_abort_extract;
2197
2198         ctx.supported_attributes_mask =
2199                 compute_supported_attributes_mask(&ctx.supported_features);
2200
2201         /* Figure out whether the root dentry is being extracted to the root of
2202          * a volume and therefore needs to be treated "specially", for example
2203          * not being explicitly created and not having attributes set.  */
2204         if (ctx.ops->target_is_root && ctx.ops->root_directory_is_special)
2205                 ctx.root_dentry_is_special = ctx.ops->target_is_root(target);
2206
2207         /* Calculate the actual filename component of each extracted dentry.  In
2208          * the process, set the dentry->extraction_skipped flag on dentries that
2209          * are being skipped for some reason (e.g. invalid filename).  */
2210         ret = for_dentry_in_tree(root, dentry_calculate_extraction_path, &ctx);
2211         if (ret)
2212                 goto out_dentry_reset_needs_extraction;
2213
2214         /* Build the list of the streams that need to be extracted and
2215          * initialize ctx.progress.extract with stream information.  */
2216         ret = for_dentry_in_tree(ctx.extract_root,
2217                                  dentry_resolve_and_zero_lte_refcnt, &ctx);
2218         if (ret)
2219                 goto out_dentry_reset_needs_extraction;
2220
2221         ret = for_dentry_in_tree(ctx.extract_root,
2222                                  dentry_add_streams_to_extract, &ctx);
2223         if (ret)
2224                 goto out_teardown_stream_list;
2225
2226         if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2227                 /* When extracting from a pipe, the number of bytes of data to
2228                  * extract can't be determined in the normal way (examining the
2229                  * lookup table), since at this point all we have is a set of
2230                  * SHA1 message digests of streams that need to be extracted.
2231                  * However, we can get a reasonably accurate estimate by taking
2232                  * <TOTALBYTES> from the corresponding <IMAGE> in the WIM XML
2233                  * data.  This does assume that a full image is being extracted,
2234                  * but currently there is no API for doing otherwise.  (Also,
2235                  * subtract <HARDLINKBYTES> from this if hard links are
2236                  * supported by the extraction mode.)  */
2237                 ctx.progress.extract.total_bytes =
2238                         wim_info_get_image_total_bytes(wim->wim_info,
2239                                                        wim->current_image);
2240                 if (ctx.supported_features.hard_links) {
2241                         ctx.progress.extract.total_bytes -=
2242                                 wim_info_get_image_hard_link_bytes(wim->wim_info,
2243                                                                    wim->current_image);
2244                 }
2245         }
2246
2247         /* Handle the special case of extracting a file to standard
2248          * output.  In that case, "root" should be a single file, not a
2249          * directory tree.  (If not, extract_dentry_to_stdout() will
2250          * return an error.)  */
2251         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT) {
2252                 ret = extract_dentry_to_stdout(root);
2253                 goto out_teardown_stream_list;
2254         }
2255
2256         /* If a sequential extraction was specified, sort the streams to be
2257          * extracted by their position in the WIM file so that the WIM file can
2258          * be read sequentially.  */
2259         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2260                               WIMLIB_EXTRACT_FLAG_FROM_PIPE))
2261                                         == WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2262         {
2263                 ret = sort_stream_list_by_sequential_order(
2264                                 &ctx.stream_list,
2265                                 offsetof(struct wim_lookup_table_entry,
2266                                          extraction_list));
2267                 if (ret)
2268                         goto out_teardown_stream_list;
2269         }
2270
2271         if (ctx.ops->realpath_works_on_nonexisting_files &&
2272             ((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) ||
2273              ctx.ops->requires_realtarget_in_paths))
2274         {
2275                 ctx.realtarget = realpath(target, NULL);
2276                 if (!ctx.realtarget) {
2277                         ret = WIMLIB_ERR_NOMEM;
2278                         goto out_teardown_stream_list;
2279                 }
2280                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2281         }
2282
2283         if (progress_func) {
2284                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN :
2285                                                  WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN,
2286                               &ctx.progress);
2287         }
2288
2289         if (!ctx.root_dentry_is_special)
2290         {
2291                 tchar path[ctx.ops->path_max];
2292                 if (build_extraction_path(path, root, &ctx))
2293                 {
2294                         ret = extract_inode(path, &ctx, root->d_inode);
2295                         if (ret)
2296                                 goto out_free_realtarget;
2297                 }
2298         }
2299
2300         /* If we need to fix up the targets of absolute symbolic links
2301          * (WIMLIB_EXTRACT_FLAG_RPFIX) or the extraction mode requires paths to
2302          * be absolute, use realpath() (or its replacement on Windows) to get
2303          * the absolute path to the extraction target.  Note that this requires
2304          * the target directory to exist, unless
2305          * realpath_works_on_nonexisting_files is set in the apply_operations.
2306          * */
2307         if (!ctx.realtarget &&
2308             (((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
2309               required_features.symlink_reparse_points) ||
2310              ctx.ops->requires_realtarget_in_paths))
2311         {
2312                 ctx.realtarget = realpath(target, NULL);
2313                 if (!ctx.realtarget) {
2314                         ret = WIMLIB_ERR_NOMEM;
2315                         goto out_free_realtarget;
2316                 }
2317                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2318         }
2319
2320         if (ctx.ops->requires_short_name_reordering) {
2321                 ret = for_dentry_in_tree(root, dentry_extract_dir_skeleton,
2322                                          &ctx);
2323                 if (ret)
2324                         goto out_free_realtarget;
2325         }
2326
2327         /* Finally, the important part: extract the tree of files.  */
2328         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2329                              WIMLIB_EXTRACT_FLAG_FROM_PIPE)) {
2330                 /* Sequential extraction requested, so two passes are needed
2331                  * (one for directory structure, one for streams.)  */
2332                 if (progress_func)
2333                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2334                                       &ctx.progress);
2335
2336                 if (!(extract_flags & WIMLIB_EXTRACT_FLAG_RESUME)) {
2337                         ret = for_dentry_in_tree(root, dentry_extract_skeleton, &ctx);
2338                         if (ret)
2339                                 goto out_free_realtarget;
2340                 }
2341                 if (progress_func)
2342                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2343                                       &ctx.progress);
2344                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE)
2345                         ret = extract_streams_from_pipe(&ctx);
2346                 else
2347                         ret = extract_stream_list(&ctx);
2348                 if (ret)
2349                         goto out_free_realtarget;
2350         } else {
2351                 /* Sequential extraction was not requested, so we can make do
2352                  * with one pass where we both create the files and extract
2353                  * streams.   */
2354                 if (progress_func)
2355                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2356                                       &ctx.progress);
2357                 ret = for_dentry_in_tree(root, dentry_extract, &ctx);
2358                 if (ret)
2359                         goto out_free_realtarget;
2360                 if (progress_func)
2361                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2362                                       &ctx.progress);
2363         }
2364
2365         /* If the total number of bytes to extract was miscalculated, just jump
2366          * to the calculated number in order to avoid confusing the progress
2367          * function.  This should only occur when extracting from a pipe.  */
2368         if (ctx.progress.extract.completed_bytes != ctx.progress.extract.total_bytes)
2369         {
2370                 DEBUG("Calculated %"PRIu64" bytes to extract, but actually "
2371                       "extracted %"PRIu64,
2372                       ctx.progress.extract.total_bytes,
2373                       ctx.progress.extract.completed_bytes);
2374         }
2375         if (progress_func &&
2376             ctx.progress.extract.completed_bytes < ctx.progress.extract.total_bytes)
2377         {
2378                 ctx.progress.extract.completed_bytes = ctx.progress.extract.total_bytes;
2379                 progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS, &ctx.progress);
2380         }
2381
2382         /* Apply security descriptors and timestamps.  This is done at the end,
2383          * and in a depth-first manner, to prevent timestamps from getting
2384          * changed by subsequent extract operations and to minimize the chance
2385          * of the restored security descriptors getting in our way.  */
2386         if (progress_func)
2387                 progress_func(WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS,
2388                               &ctx.progress);
2389         ret = for_dentry_in_tree_depth(root, dentry_extract_final, &ctx);
2390         if (ret)
2391                 goto out_free_realtarget;
2392
2393         if (progress_func) {
2394                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END :
2395                               WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END,
2396                               &ctx.progress);
2397         }
2398
2399         do_extract_warnings(&ctx);
2400
2401         ret = 0;
2402 out_free_realtarget:
2403         FREE(ctx.realtarget);
2404 out_teardown_stream_list:
2405         /* Free memory allocated as part of the mapping from each
2406          * wim_lookup_table_entry to the dentries that reference it.  */
2407         if (ctx.extract_flags & WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2408                 list_for_each_entry(lte, &ctx.stream_list, extraction_list)
2409                         if (lte->out_refcnt > ARRAY_LEN(lte->inline_lte_dentries))
2410                                 FREE(lte->lte_dentries);
2411 out_dentry_reset_needs_extraction:
2412         for_dentry_in_tree(root, dentry_reset_needs_extraction, NULL);
2413 out_finish_or_abort_extract:
2414         if (ret) {
2415                 if (ctx.ops->abort_extract)
2416                         ctx.ops->abort_extract(&ctx);
2417         } else {
2418                 if (ctx.ops->finish_extract)
2419                         ret = ctx.ops->finish_extract(&ctx);
2420         }
2421 out:
2422         return ret;
2423 }
2424
2425 /* Validates a single wimlib_extract_command, mostly checking to make sure the
2426  * extract flags make sense. */
2427 static int
2428 check_extract_command(struct wimlib_extract_command *cmd, int wim_header_flags)
2429 {
2430         int extract_flags;
2431
2432         /* Empty destination path? */
2433         if (cmd->fs_dest_path[0] == T('\0'))
2434                 return WIMLIB_ERR_INVALID_PARAM;
2435
2436         extract_flags = cmd->extract_flags;
2437
2438         /* Check for invalid flag combinations  */
2439         if ((extract_flags &
2440              (WIMLIB_EXTRACT_FLAG_SYMLINK |
2441               WIMLIB_EXTRACT_FLAG_HARDLINK)) == (WIMLIB_EXTRACT_FLAG_SYMLINK |
2442                                                  WIMLIB_EXTRACT_FLAG_HARDLINK))
2443                 return WIMLIB_ERR_INVALID_PARAM;
2444
2445         if ((extract_flags &
2446              (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2447               WIMLIB_EXTRACT_FLAG_STRICT_ACLS)) == (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2448                                                     WIMLIB_EXTRACT_FLAG_STRICT_ACLS))
2449                 return WIMLIB_ERR_INVALID_PARAM;
2450
2451         if ((extract_flags &
2452              (WIMLIB_EXTRACT_FLAG_RPFIX |
2453               WIMLIB_EXTRACT_FLAG_NORPFIX)) == (WIMLIB_EXTRACT_FLAG_RPFIX |
2454                                                 WIMLIB_EXTRACT_FLAG_NORPFIX))
2455                 return WIMLIB_ERR_INVALID_PARAM;
2456
2457         if ((extract_flags &
2458              (WIMLIB_EXTRACT_FLAG_RESUME |
2459               WIMLIB_EXTRACT_FLAG_FROM_PIPE)) == WIMLIB_EXTRACT_FLAG_RESUME)
2460                 return WIMLIB_ERR_INVALID_PARAM;
2461
2462         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2463 #ifndef WITH_NTFS_3G
2464                 ERROR("wimlib was compiled without support for NTFS-3g, so\n"
2465                       "        we cannot apply a WIM image directly to a NTFS volume.");
2466                 return WIMLIB_ERR_UNSUPPORTED;
2467 #endif
2468         }
2469
2470         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_RPFIX |
2471                               WIMLIB_EXTRACT_FLAG_NORPFIX)) == 0)
2472         {
2473                 /* Do reparse point fixups by default if the WIM header says
2474                  * they are enabled and we are extracting a full image. */
2475                 if (wim_header_flags & WIM_HDR_FLAG_RP_FIX)
2476                         extract_flags |= WIMLIB_EXTRACT_FLAG_RPFIX;
2477         }
2478
2479         /* TODO: Since UNIX data entries are stored in the file resources, in a
2480          * completely sequential extraction they may come up before the
2481          * corresponding file or symbolic link data.  This needs to be handled
2482          * better.  */
2483         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2484                               WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2485                                     == (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2486                                         WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2487         {
2488                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2489                         WARNING("Setting UNIX file/owner group may "
2490                                 "be impossible on some\n"
2491                                 "          symbolic links "
2492                                 "when applying from a pipe.");
2493                 } else {
2494                         extract_flags &= ~WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2495                         WARNING("Disabling sequential extraction for "
2496                                 "UNIX data mode");
2497                 }
2498         }
2499
2500         cmd->extract_flags = extract_flags;
2501         return 0;
2502 }
2503
2504
2505 /* Internal function to execute extraction commands for a WIM image.  The paths
2506  * in the extract commands are expected to be already "canonicalized".  */
2507 static int
2508 do_wimlib_extract_files(WIMStruct *wim,
2509                         int image,
2510                         struct wimlib_extract_command *cmds,
2511                         size_t num_cmds,
2512                         wimlib_progress_func_t progress_func)
2513 {
2514         int ret;
2515         bool found_link_cmd = false;
2516         bool found_nolink_cmd = false;
2517
2518         /* Select the image from which we are extracting files */
2519         ret = select_wim_image(wim, image);
2520         if (ret)
2521                 return ret;
2522
2523         /* Make sure there are no streams in the WIM that have not been
2524          * checksummed yet.  */
2525         ret = wim_checksum_unhashed_streams(wim);
2526         if (ret)
2527                 return ret;
2528
2529         /* Check for problems with the extraction commands */
2530         for (size_t i = 0; i < num_cmds; i++) {
2531                 ret = check_extract_command(&cmds[i], wim->hdr.flags);
2532                 if (ret)
2533                         return ret;
2534                 if (cmds[i].extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2535                                              WIMLIB_EXTRACT_FLAG_HARDLINK)) {
2536                         found_link_cmd = true;
2537                 } else {
2538                         found_nolink_cmd = true;
2539                 }
2540                 if (found_link_cmd && found_nolink_cmd) {
2541                         ERROR("Symlink or hardlink extraction mode must "
2542                               "be set on all extraction commands");
2543                         return WIMLIB_ERR_INVALID_PARAM;
2544                 }
2545         }
2546
2547         /* Execute the extraction commands */
2548         for (size_t i = 0; i < num_cmds; i++) {
2549                 ret = extract_tree(wim,
2550                                    cmds[i].wim_source_path,
2551                                    cmds[i].fs_dest_path,
2552                                    cmds[i].extract_flags,
2553                                    progress_func);
2554                 if (ret)
2555                         return ret;
2556         }
2557         return 0;
2558 }
2559
2560 /* API function documented in wimlib.h  */
2561 WIMLIBAPI int
2562 wimlib_extract_files(WIMStruct *wim,
2563                      int image,
2564                      const struct wimlib_extract_command *cmds,
2565                      size_t num_cmds,
2566                      int default_extract_flags,
2567                      wimlib_progress_func_t progress_func)
2568 {
2569         int ret;
2570         struct wimlib_extract_command *cmds_copy;
2571         int all_flags = 0;
2572
2573         default_extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2574
2575         if (num_cmds == 0)
2576                 return 0;
2577
2578         cmds_copy = CALLOC(num_cmds, sizeof(cmds[0]));
2579         if (!cmds_copy)
2580                 return WIMLIB_ERR_NOMEM;
2581
2582         for (size_t i = 0; i < num_cmds; i++) {
2583                 cmds_copy[i].extract_flags = (default_extract_flags |
2584                                                  cmds[i].extract_flags)
2585                                                 & WIMLIB_EXTRACT_MASK_PUBLIC;
2586                 all_flags |= cmds_copy[i].extract_flags;
2587
2588                 cmds_copy[i].wim_source_path = canonicalize_wim_path(cmds[i].wim_source_path);
2589                 if (!cmds_copy[i].wim_source_path) {
2590                         ret = WIMLIB_ERR_NOMEM;
2591                         goto out_free_cmds_copy;
2592                 }
2593
2594                 cmds_copy[i].fs_dest_path = canonicalize_fs_path(cmds[i].fs_dest_path);
2595                 if (!cmds_copy[i].fs_dest_path) {
2596                         ret = WIMLIB_ERR_NOMEM;
2597                         goto out_free_cmds_copy;
2598                 }
2599
2600         }
2601         ret = do_wimlib_extract_files(wim, image,
2602                                       cmds_copy, num_cmds,
2603                                       progress_func);
2604
2605         if (all_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2606                          WIMLIB_EXTRACT_FLAG_HARDLINK))
2607         {
2608                 for_lookup_table_entry(wim->lookup_table,
2609                                        lte_free_extracted_file, NULL);
2610         }
2611 out_free_cmds_copy:
2612         for (size_t i = 0; i < num_cmds; i++) {
2613                 FREE(cmds_copy[i].wim_source_path);
2614                 FREE(cmds_copy[i].fs_dest_path);
2615         }
2616         FREE(cmds_copy);
2617         return ret;
2618 }
2619
2620 /*
2621  * Extracts an image from a WIM file.
2622  *
2623  * @wim:                WIMStruct for the WIM file.
2624  *
2625  * @image:              Number of the single image to extract.
2626  *
2627  * @target:             Directory or NTFS volume to extract the image to.
2628  *
2629  * @extract_flags:      Bitwise or of WIMLIB_EXTRACT_FLAG_*.
2630  *
2631  * @progress_func:      If non-NULL, a progress function to be called
2632  *                      periodically.
2633  *
2634  * Returns 0 on success; nonzero on failure.
2635  */
2636 static int
2637 extract_single_image(WIMStruct *wim, int image,
2638                      const tchar *target, int extract_flags,
2639                      wimlib_progress_func_t progress_func)
2640 {
2641         int ret;
2642         tchar *target_copy = canonicalize_fs_path(target);
2643         if (!target_copy)
2644                 return WIMLIB_ERR_NOMEM;
2645         struct wimlib_extract_command cmd = {
2646                 .wim_source_path = T(""),
2647                 .fs_dest_path = target_copy,
2648                 .extract_flags = extract_flags,
2649         };
2650         ret = do_wimlib_extract_files(wim, image, &cmd, 1, progress_func);
2651         FREE(target_copy);
2652         return ret;
2653 }
2654
2655 static const tchar * const filename_forbidden_chars =
2656 T(
2657 #ifdef __WIN32__
2658 "<>:\"/\\|?*"
2659 #else
2660 "/"
2661 #endif
2662 );
2663
2664 /* This function checks if it is okay to use a WIM image's name as a directory
2665  * name.  */
2666 static bool
2667 image_name_ok_as_dir(const tchar *image_name)
2668 {
2669         return image_name && *image_name &&
2670                 !tstrpbrk(image_name, filename_forbidden_chars) &&
2671                 tstrcmp(image_name, T(".")) &&
2672                 tstrcmp(image_name, T(".."));
2673 }
2674
2675 /* Extracts all images from the WIM to the directory @target, with the images
2676  * placed in subdirectories named by their image names. */
2677 static int
2678 extract_all_images(WIMStruct *wim,
2679                    const tchar *target,
2680                    int extract_flags,
2681                    wimlib_progress_func_t progress_func)
2682 {
2683         size_t image_name_max_len = max(xml_get_max_image_name_len(wim), 20);
2684         size_t output_path_len = tstrlen(target);
2685         tchar buf[output_path_len + 1 + image_name_max_len + 1];
2686         int ret;
2687         int image;
2688         const tchar *image_name;
2689         struct stat stbuf;
2690
2691         extract_flags |= WIMLIB_EXTRACT_FLAG_MULTI_IMAGE;
2692
2693         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2694                 ERROR("Cannot extract multiple images in NTFS extraction mode.");
2695                 return WIMLIB_ERR_INVALID_PARAM;
2696         }
2697
2698         if (tstat(target, &stbuf)) {
2699                 if (errno == ENOENT) {
2700                         if (tmkdir(target, 0755)) {
2701                                 ERROR_WITH_ERRNO("Failed to create directory \"%"TS"\"", target);
2702                                 return WIMLIB_ERR_MKDIR;
2703                         }
2704                 } else {
2705                         ERROR_WITH_ERRNO("Failed to stat \"%"TS"\"", target);
2706                         return WIMLIB_ERR_STAT;
2707                 }
2708         } else if (!S_ISDIR(stbuf.st_mode)) {
2709                 ERROR("\"%"TS"\" is not a directory", target);
2710                 return WIMLIB_ERR_NOTDIR;
2711         }
2712
2713         tmemcpy(buf, target, output_path_len);
2714         buf[output_path_len] = OS_PREFERRED_PATH_SEPARATOR;
2715         for (image = 1; image <= wim->hdr.image_count; image++) {
2716                 image_name = wimlib_get_image_name(wim, image);
2717                 if (image_name_ok_as_dir(image_name)) {
2718                         tstrcpy(buf + output_path_len + 1, image_name);
2719                 } else {
2720                         /* Image name is empty or contains forbidden characters.
2721                          * Use image number instead. */
2722                         tsprintf(buf + output_path_len + 1, T("%d"), image);
2723                 }
2724                 ret = extract_single_image(wim, image, buf, extract_flags,
2725                                            progress_func);
2726                 if (ret)
2727                         return ret;
2728         }
2729         return 0;
2730 }
2731
2732 static int
2733 do_wimlib_extract_image(WIMStruct *wim,
2734                         int image,
2735                         const tchar *target,
2736                         int extract_flags,
2737                         wimlib_progress_func_t progress_func)
2738 {
2739         int ret;
2740
2741         if (image == WIMLIB_ALL_IMAGES) {
2742                 ret = extract_all_images(wim, target, extract_flags,
2743                                          progress_func);
2744         } else {
2745                 ret = extract_single_image(wim, image, target, extract_flags,
2746                                            progress_func);
2747         }
2748
2749         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2750                              WIMLIB_EXTRACT_FLAG_HARDLINK))
2751         {
2752                 for_lookup_table_entry(wim->lookup_table,
2753                                        lte_free_extracted_file,
2754                                        NULL);
2755         }
2756         return ret;
2757 }
2758
2759 /* API function documented in wimlib.h  */
2760 WIMLIBAPI int
2761 wimlib_extract_image_from_pipe(int pipe_fd, const tchar *image_num_or_name,
2762                                const tchar *target, int extract_flags,
2763                                wimlib_progress_func_t progress_func)
2764 {
2765         int ret;
2766         WIMStruct *pwm;
2767         struct filedes *in_fd;
2768         int image;
2769         unsigned i;
2770
2771         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2772
2773         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT)
2774                 return WIMLIB_ERR_INVALID_PARAM;
2775
2776         extract_flags |= WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2777
2778         /* Read the WIM header from the pipe and get a WIMStruct to represent
2779          * the pipable WIM.  Caveats:  Unlike getting a WIMStruct with
2780          * wimlib_open_wim(), getting a WIMStruct in this way will result in
2781          * an empty lookup table, no XML data read, and no filename set.  */
2782         ret = open_wim_as_WIMStruct(&pipe_fd,
2783                                     WIMLIB_OPEN_FLAG_FROM_PIPE,
2784                                     &pwm, progress_func);
2785         if (ret)
2786                 return ret;
2787
2788         /* Sanity check to make sure this is a pipable WIM.  */
2789         if (pwm->hdr.magic != PWM_MAGIC) {
2790                 ERROR("The WIM being read from file descriptor %d "
2791                       "is not pipable!", pipe_fd);
2792                 ret = WIMLIB_ERR_NOT_PIPABLE;
2793                 goto out_wimlib_free;
2794         }
2795
2796         /* Sanity check to make sure the first part of a pipable split WIM is
2797          * sent over the pipe first.  */
2798         if (pwm->hdr.part_number != 1) {
2799                 ERROR("The first part of the split WIM must be "
2800                       "sent over the pipe first.");
2801                 ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2802                 goto out_wimlib_free;
2803         }
2804
2805         in_fd = &pwm->in_fd;
2806         wimlib_assert(in_fd->offset == WIM_HEADER_DISK_SIZE);
2807
2808         /* As mentioned, the WIMStruct we created from the pipe does not have
2809          * XML data yet.  Fix this by reading the extra copy of the XML data
2810          * that directly follows the header in pipable WIMs.  (Note: see
2811          * write_pipable_wim() for more details about the format of pipable
2812          * WIMs.)  */
2813         {
2814                 struct wim_lookup_table_entry xml_lte;
2815                 ret = read_pwm_stream_header(pwm, &xml_lte, 0, NULL);
2816                 if (ret)
2817                         goto out_wimlib_free;
2818
2819                 if (!(xml_lte.resource_entry.flags & WIM_RESHDR_FLAG_METADATA))
2820                 {
2821                         ERROR("Expected XML data, but found non-metadata "
2822                               "stream.");
2823                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2824                         goto out_wimlib_free;
2825                 }
2826
2827                 copy_resource_entry(&pwm->hdr.xml_res_entry,
2828                                     &xml_lte.resource_entry);
2829
2830                 ret = read_wim_xml_data(pwm);
2831                 if (ret)
2832                         goto out_wimlib_free;
2833                 if (wim_info_get_num_images(pwm->wim_info) != pwm->hdr.image_count) {
2834                         ERROR("Image count in XML data is not the same as in WIM header.");
2835                         ret = WIMLIB_ERR_XML;
2836                         goto out_wimlib_free;
2837                 }
2838         }
2839
2840         /* Get image index (this may use the XML data that was just read to
2841          * resolve an image name).  */
2842         if (image_num_or_name) {
2843                 image = wimlib_resolve_image(pwm, image_num_or_name);
2844                 if (image == WIMLIB_NO_IMAGE) {
2845                         ERROR("\"%"TS"\" is not a valid image in the pipable WIM!",
2846                               image_num_or_name);
2847                         ret = WIMLIB_ERR_INVALID_IMAGE;
2848                         goto out_wimlib_free;
2849                 } else if (image == WIMLIB_ALL_IMAGES) {
2850                         ERROR("Applying all images from a pipe is not supported.");
2851                         ret = WIMLIB_ERR_INVALID_IMAGE;
2852                         goto out_wimlib_free;
2853                 }
2854         } else {
2855                 if (pwm->hdr.image_count != 1) {
2856                         ERROR("No image was specified, but the pipable WIM "
2857                               "did not contain exactly 1 image");
2858                         ret = WIMLIB_ERR_INVALID_IMAGE;
2859                         goto out_wimlib_free;
2860                 }
2861                 image = 1;
2862         }
2863
2864         /* Load the needed metadata resource.  */
2865         for (i = 1; i <= pwm->hdr.image_count; i++) {
2866                 struct wim_lookup_table_entry *metadata_lte;
2867                 struct wim_image_metadata *imd;
2868
2869                 metadata_lte = new_lookup_table_entry();
2870                 if (!metadata_lte) {
2871                         ret = WIMLIB_ERR_NOMEM;
2872                         goto out_wimlib_free;
2873                 }
2874
2875                 ret = read_pwm_stream_header(pwm, metadata_lte, 0, NULL);
2876                 imd = pwm->image_metadata[i - 1];
2877                 imd->metadata_lte = metadata_lte;
2878                 if (ret)
2879                         goto out_wimlib_free;
2880
2881                 if (!(metadata_lte->resource_entry.flags &
2882                       WIM_RESHDR_FLAG_METADATA))
2883                 {
2884                         ERROR("Expected metadata resource, but found "
2885                               "non-metadata stream.");
2886                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2887                         goto out_wimlib_free;
2888                 }
2889
2890                 if (i == image) {
2891                         /* Metadata resource is for the images being extracted.
2892                          * Parse it and save the metadata in memory.  */
2893                         ret = read_metadata_resource(pwm, imd);
2894                         if (ret)
2895                                 goto out_wimlib_free;
2896                         imd->modified = 1;
2897                 } else {
2898                         /* Metadata resource is not for the image being
2899                          * extracted.  Skip over it.  */
2900                         ret = skip_pwm_stream(metadata_lte);
2901                         if (ret)
2902                                 goto out_wimlib_free;
2903                 }
2904         }
2905         /* Extract the image.  */
2906         extract_flags |= WIMLIB_EXTRACT_FLAG_FROM_PIPE;
2907         ret = do_wimlib_extract_image(pwm, image, target,
2908                                       extract_flags, progress_func);
2909         /* Clean up and return.  */
2910 out_wimlib_free:
2911         wimlib_free(pwm);
2912         return ret;
2913 }
2914
2915 /* API function documented in wimlib.h  */
2916 WIMLIBAPI int
2917 wimlib_extract_image(WIMStruct *wim,
2918                      int image,
2919                      const tchar *target,
2920                      int extract_flags,
2921                      wimlib_progress_func_t progress_func)
2922 {
2923         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2924         return do_wimlib_extract_image(wim, image, target, extract_flags,
2925                                        progress_func);
2926 }