Generalized support for referencing resources in external WIMs
[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                         {
1148                                 DEBUG("Creating %"TS" before %"TS" "
1149                                       "to guarantee correct DOS name extraction",
1150                                       dentry_full_path(other_dentry),
1151                                       dentry_full_path(dentry));
1152                                 orig_dentry = dentry;
1153                                 dentry = other_dentry;
1154                                 break;
1155                         }
1156                 }
1157         }
1158 again:
1159         if (!build_extraction_path(path, dentry, ctx))
1160                 return 0;
1161         ret = do_dentry_extract_skeleton(path, dentry, ctx);
1162         if (ret)
1163                 return ret;
1164
1165         dentry->skeleton_extracted = 1;
1166
1167         if (orig_dentry) {
1168                 dentry = orig_dentry;
1169                 orig_dentry = NULL;
1170                 goto again;
1171         }
1172         dentry->d_inode->i_dos_name_extracted = 1;
1173         return 0;
1174 }
1175
1176 static int
1177 dentry_extract_dir_skeleton(struct wim_dentry *dentry, void *_ctx)
1178 {
1179         if (dentry->d_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
1180                 return dentry_extract_skeleton(dentry, _ctx);
1181         return 0;
1182 }
1183
1184 /* Create a file or directory, then immediately extract all streams.  This
1185  * assumes that WIMLIB_EXTRACT_FLAG_SEQUENTIAL is not specified, since the WIM
1186  * may not be read sequentially by this function.  */
1187 static int
1188 dentry_extract(struct wim_dentry *dentry, void *_ctx)
1189 {
1190         struct apply_ctx *ctx = _ctx;
1191         tchar path[ctx->ops->path_max];
1192         int ret;
1193
1194         ret = dentry_extract_skeleton(dentry, ctx);
1195         if (ret)
1196                 return ret;
1197
1198         if (!build_extraction_path(path, dentry, ctx))
1199                 return 0;
1200
1201         return extract_streams(path, ctx, dentry, NULL, NULL);
1202 }
1203
1204 /* Extract all instances of the stream @lte that are being extracted in this
1205  * call of extract_tree().  @can_seek specifies whether the WIM file descriptor
1206  * is seekable or not (e.g. is a pipe).  If not and the stream needs to be
1207  * extracted multiple times, it is extracted to a temporary file first.
1208  *
1209  * This is intended for use with sequential extraction of a WIM image
1210  * (WIMLIB_EXTRACT_FLAG_SEQUENTIAL specified).  */
1211 static int
1212 extract_stream_instances(struct wim_lookup_table_entry *lte,
1213                          struct apply_ctx *ctx, bool can_seek)
1214 {
1215         struct wim_dentry **lte_dentries;
1216         struct wim_lookup_table_entry *lte_tmp = NULL;
1217         struct wim_lookup_table_entry *lte_override;
1218         tchar *stream_tmp_filename = NULL;
1219         tchar path[ctx->ops->path_max];
1220         unsigned i;
1221         int ret;
1222
1223         if (lte->out_refcnt <= ARRAY_LEN(lte->inline_lte_dentries))
1224                 lte_dentries = lte->inline_lte_dentries;
1225         else
1226                 lte_dentries = lte->lte_dentries;
1227
1228         if (likely(can_seek || lte->out_refcnt < 2)) {
1229                 lte_override = lte;
1230         } else {
1231                 /* Need to extract stream to temporary file.  */
1232                 struct filedes fd;
1233                 int raw_fd;
1234
1235                 stream_tmp_filename = ttempnam(NULL, T("wimlib"));
1236                 if (!stream_tmp_filename) {
1237                         ERROR_WITH_ERRNO("Failed to create temporary filename");
1238                         ret = WIMLIB_ERR_OPEN;
1239                         goto out;
1240                 }
1241
1242                 lte_tmp = memdup(lte, sizeof(struct wim_lookup_table_entry));
1243                 if (!lte_tmp) {
1244                         ret = WIMLIB_ERR_NOMEM;
1245                         goto out_free_stream_tmp_filename;
1246                 }
1247                 lte_tmp->resource_location = RESOURCE_IN_FILE_ON_DISK;
1248                 lte_tmp->file_on_disk = stream_tmp_filename;
1249                 lte_override = lte_tmp;
1250
1251                 raw_fd = topen(stream_tmp_filename,
1252                                O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0600);
1253                 if (raw_fd < 0) {
1254                         ERROR_WITH_ERRNO("Failed to open temporary file");
1255                         ret = WIMLIB_ERR_OPEN;
1256                         goto out_free_lte_tmp;
1257                 }
1258                 filedes_init(&fd, raw_fd);
1259                 ret = extract_wim_resource_to_fd(lte, &fd,
1260                                                  wim_resource_size(lte));
1261                 if (filedes_close(&fd) && !ret)
1262                         ret = WIMLIB_ERR_WRITE;
1263                 if (ret)
1264                         goto out_unlink_stream_tmp_file;
1265         }
1266
1267         /* Extract all instances of the stream, reading either from the stream
1268          * in the WIM file or from the temporary file containing the stream.
1269          * dentry->tmp_flag is used to ensure that each dentry is processed only
1270          * once regardless of how many times this stream appears in the streams
1271          * of the corresponding inode.  */
1272         for (i = 0; i < lte->out_refcnt; i++) {
1273                 struct wim_dentry *dentry = lte_dentries[i];
1274
1275                 if (dentry->tmp_flag)
1276                         continue;
1277                 if (!build_extraction_path(path, dentry, ctx))
1278                         continue;
1279                 ret = extract_streams(path, ctx, dentry,
1280                                       lte, lte_override);
1281                 if (ret)
1282                         goto out_clear_tmp_flags;
1283                 dentry->tmp_flag = 1;
1284         }
1285         ret = 0;
1286 out_clear_tmp_flags:
1287         for (i = 0; i < lte->out_refcnt; i++)
1288                 lte_dentries[i]->tmp_flag = 0;
1289 out_unlink_stream_tmp_file:
1290         if (stream_tmp_filename)
1291                 tunlink(stream_tmp_filename);
1292 out_free_lte_tmp:
1293         FREE(lte_tmp);
1294 out_free_stream_tmp_filename:
1295         FREE(stream_tmp_filename);
1296 out:
1297         return ret;
1298 }
1299
1300 /* Extracts a list of streams (ctx.stream_list), assuming that the directory
1301  * structure and empty files were already created.  This relies on the
1302  * per-`struct wim_lookup_table_entry' list of dentries that reference each
1303  * stream that was constructed earlier.  Streams are extracted exactly in the
1304  * order of the stream list; however, unless the WIM's file descriptor is
1305  * detected to be non-seekable, streams may be read from the WIM file more than
1306  * one time if multiple copies need to be extracted.  */
1307 static int
1308 extract_stream_list(struct apply_ctx *ctx)
1309 {
1310         struct wim_lookup_table_entry *lte;
1311         bool can_seek;
1312         int ret;
1313
1314         can_seek = (lseek(ctx->wim->in_fd.fd, 0, SEEK_CUR) != -1);
1315         list_for_each_entry(lte, &ctx->stream_list, extraction_list) {
1316                 ret = extract_stream_instances(lte, ctx, can_seek);
1317                 if (ret)
1318                         return ret;
1319         }
1320         return 0;
1321 }
1322
1323 #define PWM_ALLOW_WIM_HDR 0x00001
1324 #define PWM_SILENT_EOF    0x00002
1325
1326 /* Read the header from a stream in a pipable WIM.  */
1327 static int
1328 read_pwm_stream_header(WIMStruct *pwm, struct wim_lookup_table_entry *lte,
1329                        int flags, struct wim_header_disk *hdr_ret)
1330 {
1331         union {
1332                 struct pwm_stream_hdr stream_hdr;
1333                 struct wim_header_disk pwm_hdr;
1334         } buf;
1335         int ret;
1336
1337         ret = full_read(&pwm->in_fd, &buf.stream_hdr, sizeof(buf.stream_hdr));
1338         if (ret)
1339                 goto read_error;
1340
1341         if ((flags & PWM_ALLOW_WIM_HDR) && buf.stream_hdr.magic == PWM_MAGIC) {
1342                 BUILD_BUG_ON(sizeof(buf.pwm_hdr) < sizeof(buf.stream_hdr));
1343                 ret = full_read(&pwm->in_fd, &buf.stream_hdr + 1,
1344                                 sizeof(buf.pwm_hdr) - sizeof(buf.stream_hdr));
1345
1346                 if (ret)
1347                         goto read_error;
1348                 lte->resource_location = RESOURCE_NONEXISTENT;
1349                 memcpy(hdr_ret, &buf.pwm_hdr, sizeof(buf.pwm_hdr));
1350                 return 0;
1351         }
1352
1353         if (buf.stream_hdr.magic != PWM_STREAM_MAGIC) {
1354                 ERROR("Data read on pipe is invalid (expected stream header).");
1355                 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1356         }
1357
1358         lte->resource_entry.original_size = le64_to_cpu(buf.stream_hdr.uncompressed_size);
1359         copy_hash(lte->hash, buf.stream_hdr.hash);
1360         lte->resource_entry.flags = le32_to_cpu(buf.stream_hdr.flags);
1361         lte->resource_entry.offset = pwm->in_fd.offset;
1362         lte->resource_location = RESOURCE_IN_WIM;
1363         lte->wim = pwm;
1364         if (lte->resource_entry.flags & WIM_RESHDR_FLAG_COMPRESSED) {
1365                 lte->compression_type = pwm->compression_type;
1366                 lte->resource_entry.size = 0;
1367         } else {
1368                 lte->compression_type = WIMLIB_COMPRESSION_TYPE_NONE;
1369                 lte->resource_entry.size = lte->resource_entry.original_size;
1370         }
1371         lte->is_pipable = 1;
1372         return 0;
1373
1374 read_error:
1375         if (ret != WIMLIB_ERR_UNEXPECTED_END_OF_FILE || !(flags & PWM_SILENT_EOF))
1376                 ERROR_WITH_ERRNO("Error reading pipable WIM from pipe");
1377         return ret;
1378 }
1379
1380 /* Skip over an unneeded stream in a pipable WIM being read from a pipe.  */
1381 static int
1382 skip_pwm_stream(struct wim_lookup_table_entry *lte)
1383 {
1384         return read_partial_wim_resource(lte, wim_resource_size(lte),
1385                                          NULL, NULL,
1386                                          WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY,
1387                                          0);
1388 }
1389
1390 static int
1391 extract_streams_from_pipe(struct apply_ctx *ctx)
1392 {
1393         struct wim_lookup_table_entry *found_lte;
1394         struct wim_lookup_table_entry *needed_lte;
1395         struct wim_lookup_table *lookup_table;
1396         struct wim_header_disk pwm_hdr;
1397         int ret;
1398         int pwm_flags;
1399
1400         ret = WIMLIB_ERR_NOMEM;
1401         found_lte = new_lookup_table_entry();
1402         if (!found_lte)
1403                 goto out;
1404
1405         lookup_table = ctx->wim->lookup_table;
1406         pwm_flags = PWM_ALLOW_WIM_HDR;
1407         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1408                 pwm_flags |= PWM_SILENT_EOF;
1409         memcpy(ctx->progress.extract.guid, ctx->wim->hdr.guid, WIM_GID_LEN);
1410         ctx->progress.extract.part_number = ctx->wim->hdr.part_number;
1411         ctx->progress.extract.total_parts = ctx->wim->hdr.total_parts;
1412         if (ctx->progress_func)
1413                 ctx->progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1414                                    &ctx->progress);
1415         while (ctx->num_streams_remaining) {
1416                 ret = read_pwm_stream_header(ctx->wim, found_lte, pwm_flags,
1417                                              &pwm_hdr);
1418                 if (ret) {
1419                         if (ret == WIMLIB_ERR_UNEXPECTED_END_OF_FILE &&
1420                             (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1421                         {
1422                                 goto resume_done;
1423                         }
1424                         goto out_free_found_lte;
1425                 }
1426
1427                 if ((found_lte->resource_location != RESOURCE_NONEXISTENT)
1428                     && !(found_lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)
1429                     && (needed_lte = __lookup_resource(lookup_table, found_lte->hash))
1430                     && (needed_lte->out_refcnt))
1431                 {
1432                         copy_resource_entry(&needed_lte->resource_entry,
1433                                             &found_lte->resource_entry);
1434                         needed_lte->resource_location = found_lte->resource_location;
1435                         needed_lte->wim               = found_lte->wim;
1436                         needed_lte->compression_type  = found_lte->compression_type;
1437                         needed_lte->is_pipable        = found_lte->is_pipable;
1438
1439                         ret = extract_stream_instances(needed_lte, ctx, false);
1440                         if (ret)
1441                                 goto out_free_found_lte;
1442                         ctx->num_streams_remaining--;
1443                 } else if (found_lte->resource_location != RESOURCE_NONEXISTENT) {
1444                         ret = skip_pwm_stream(found_lte);
1445                         if (ret)
1446                                 goto out_free_found_lte;
1447                 } else {
1448                         u16 part_number = le16_to_cpu(pwm_hdr.part_number);
1449                         u16 total_parts = le16_to_cpu(pwm_hdr.total_parts);
1450
1451                         if (part_number != ctx->progress.extract.part_number ||
1452                             total_parts != ctx->progress.extract.total_parts ||
1453                             memcmp(pwm_hdr.guid, ctx->progress.extract.guid,
1454                                    WIM_GID_LEN))
1455                         {
1456                                 ctx->progress.extract.part_number = part_number;
1457                                 ctx->progress.extract.total_parts = total_parts;
1458                                 memcpy(ctx->progress.extract.guid,
1459                                        pwm_hdr.guid, WIM_GID_LEN);
1460                                 if (ctx->progress_func) {
1461                                         ctx->progress_func(
1462                                                 WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1463                                                            &ctx->progress);
1464                                 }
1465
1466                         }
1467                 }
1468         }
1469         ret = 0;
1470 out_free_found_lte:
1471         free_lookup_table_entry(found_lte);
1472 out:
1473         return ret;
1474
1475 resume_done:
1476         /* TODO */
1477         return 0;
1478 }
1479
1480 /* Finish extracting a file, directory, or symbolic link by setting file
1481  * security and timestamps.  */
1482 static int
1483 dentry_extract_final(struct wim_dentry *dentry, void *_ctx)
1484 {
1485         struct apply_ctx *ctx = _ctx;
1486         int ret;
1487         tchar path[ctx->ops->path_max];
1488
1489         if (!build_extraction_path(path, dentry, ctx))
1490                 return 0;
1491
1492         ret = extract_security(path, ctx, dentry);
1493         if (ret)
1494                 return ret;
1495
1496         if (ctx->ops->requires_final_set_attributes_pass) {
1497                 /* Set file attributes (if supported).  */
1498                 ret = extract_file_attributes(path, ctx, dentry, 1);
1499                 if (ret)
1500                         return ret;
1501         }
1502
1503         return extract_timestamps(path, ctx, dentry);
1504 }
1505
1506 /*
1507  * Extract a WIM dentry to standard output.
1508  *
1509  * This obviously doesn't make sense in all cases.  We return an error if the
1510  * dentry does not correspond to a regular file.  Otherwise we extract the
1511  * unnamed data stream only.
1512  */
1513 static int
1514 extract_dentry_to_stdout(struct wim_dentry *dentry)
1515 {
1516         int ret = 0;
1517         if (dentry->d_inode->i_attributes & (FILE_ATTRIBUTE_REPARSE_POINT |
1518                                              FILE_ATTRIBUTE_DIRECTORY))
1519         {
1520                 ERROR("\"%"TS"\" is not a regular file and therefore cannot be "
1521                       "extracted to standard output", dentry_full_path(dentry));
1522                 ret = WIMLIB_ERR_NOT_A_REGULAR_FILE;
1523         } else {
1524                 struct wim_lookup_table_entry *lte;
1525
1526                 lte = inode_unnamed_lte_resolved(dentry->d_inode);
1527                 if (lte) {
1528                         struct filedes _stdout;
1529                         filedes_init(&_stdout, STDOUT_FILENO);
1530                         ret = extract_wim_resource_to_fd(lte, &_stdout,
1531                                                          wim_resource_size(lte));
1532                 }
1533         }
1534         return ret;
1535 }
1536
1537 #ifdef __WIN32__
1538 static const utf16lechar replacement_char = cpu_to_le16(0xfffd);
1539 #else
1540 static const utf16lechar replacement_char = cpu_to_le16('?');
1541 #endif
1542
1543 static bool
1544 file_name_valid(utf16lechar *name, size_t num_chars, bool fix)
1545 {
1546         size_t i;
1547
1548         if (num_chars == 0)
1549                 return true;
1550         for (i = 0; i < num_chars; i++) {
1551                 switch (name[i]) {
1552         #ifdef __WIN32__
1553                 case cpu_to_le16('\\'):
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         #endif
1562                 case cpu_to_le16('/'):
1563                 case cpu_to_le16('\0'):
1564                         if (fix)
1565                                 name[i] = replacement_char;
1566                         else
1567                                 return false;
1568                 }
1569         }
1570
1571 #ifdef __WIN32__
1572         if (name[num_chars - 1] == cpu_to_le16(' ') ||
1573             name[num_chars - 1] == cpu_to_le16('.'))
1574         {
1575                 if (fix)
1576                         name[num_chars - 1] = replacement_char;
1577                 else
1578                         return false;
1579         }
1580 #endif
1581         return true;
1582 }
1583
1584 static bool
1585 dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
1586 {
1587         const utf16lechar *file_name = dentry->file_name;
1588         return file_name != NULL &&
1589                 file_name[0] == cpu_to_le16('.') &&
1590                 (file_name[1] == cpu_to_le16('\0') ||
1591                  (file_name[1] == cpu_to_le16('.') &&
1592                   file_name[2] == cpu_to_le16('\0')));
1593 }
1594
1595 static int
1596 dentry_mark_skipped(struct wim_dentry *dentry, void *_ignore)
1597 {
1598         dentry->extraction_skipped = 1;
1599         return 0;
1600 }
1601
1602 /*
1603  * dentry_calculate_extraction_path-
1604  *
1605  * Calculate the actual filename component at which a WIM dentry will be
1606  * extracted, handling invalid filenames "properly".
1607  *
1608  * dentry->extraction_name usually will be set the same as dentry->file_name (on
1609  * UNIX, converted into the platform's multibyte encoding).  However, if the
1610  * file name contains characters that are not valid on the current platform or
1611  * has some other format that is not valid, leave dentry->extraction_name as
1612  * NULL and set dentry->extraction_skipped to indicate that this dentry should
1613  * not be extracted, unless the appropriate flag
1614  * WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES is set in the extract flags, in
1615  * which case a substitute filename will be created and set instead.
1616  *
1617  * Conflicts with case-insensitive names on Windows are handled similarly; see
1618  * below.
1619  */
1620 static int
1621 dentry_calculate_extraction_path(struct wim_dentry *dentry, void *_args)
1622 {
1623         struct apply_ctx *ctx = _args;
1624         int ret;
1625
1626         if (dentry == ctx->extract_root || dentry->extraction_skipped)
1627                 return 0;
1628
1629         if (!dentry_is_supported(dentry, &ctx->supported_features))
1630                 goto skip_dentry;
1631
1632         if (dentry_is_dot_or_dotdot(dentry)) {
1633                 /* WIM files shouldn't contain . or .. entries.  But if they are
1634                  * there, don't attempt to extract them. */
1635                 WARNING("Skipping extraction of unexpected . or .. file "
1636                         "\"%"TS"\"", dentry_full_path(dentry));
1637                 goto skip_dentry;
1638         }
1639
1640 #ifdef __WIN32__
1641         if (!ctx->ops->supports_case_sensitive_filenames)
1642         {
1643                 struct wim_dentry *other;
1644                 list_for_each_entry(other, &dentry->case_insensitive_conflict_list,
1645                                     case_insensitive_conflict_list)
1646                 {
1647                         if (ctx->extract_flags &
1648                             WIMLIB_EXTRACT_FLAG_ALL_CASE_CONFLICTS) {
1649                                 WARNING("\"%"TS"\" has the same "
1650                                         "case-insensitive name as "
1651                                         "\"%"TS"\"; extracting "
1652                                         "dummy name instead",
1653                                         dentry_full_path(dentry),
1654                                         dentry_full_path(other));
1655                                 goto out_replace;
1656                         } else {
1657                                 WARNING("Not extracting \"%"TS"\": "
1658                                         "has same case-insensitive "
1659                                         "name as \"%"TS"\"",
1660                                         dentry_full_path(dentry),
1661                                         dentry_full_path(other));
1662                                 goto skip_dentry;
1663                         }
1664                 }
1665         }
1666 #else   /* __WIN32__ */
1667         wimlib_assert(ctx->ops->supports_case_sensitive_filenames);
1668 #endif  /* !__WIN32__ */
1669
1670         if (file_name_valid(dentry->file_name, dentry->file_name_nbytes / 2, false)) {
1671 #ifdef __WIN32__
1672                 dentry->extraction_name = dentry->file_name;
1673                 dentry->extraction_name_nchars = dentry->file_name_nbytes / 2;
1674                 return 0;
1675 #else
1676                 return utf16le_to_tstr(dentry->file_name,
1677                                        dentry->file_name_nbytes,
1678                                        &dentry->extraction_name,
1679                                        &dentry->extraction_name_nchars);
1680 #endif
1681         } else {
1682                 if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES)
1683                 {
1684                         WARNING("\"%"TS"\" has an invalid filename "
1685                                 "that is not supported on this platform; "
1686                                 "extracting dummy name instead",
1687                                 dentry_full_path(dentry));
1688                         goto out_replace;
1689                 } else {
1690                         WARNING("Not extracting \"%"TS"\": has an invalid filename "
1691                                 "that is not supported on this platform",
1692                                 dentry_full_path(dentry));
1693                         goto skip_dentry;
1694                 }
1695         }
1696
1697 out_replace:
1698         {
1699                 utf16lechar utf16_name_copy[dentry->file_name_nbytes / 2];
1700
1701                 memcpy(utf16_name_copy, dentry->file_name, dentry->file_name_nbytes);
1702                 file_name_valid(utf16_name_copy, dentry->file_name_nbytes / 2, true);
1703
1704                 tchar *tchar_name;
1705                 size_t tchar_nchars;
1706         #ifdef __WIN32__
1707                 tchar_name = utf16_name_copy;
1708                 tchar_nchars = dentry->file_name_nbytes / 2;
1709         #else
1710                 ret = utf16le_to_tstr(utf16_name_copy,
1711                                       dentry->file_name_nbytes,
1712                                       &tchar_name, &tchar_nchars);
1713                 if (ret)
1714                         return ret;
1715         #endif
1716                 size_t fixed_name_num_chars = tchar_nchars;
1717                 tchar fixed_name[tchar_nchars + 50];
1718
1719                 tmemcpy(fixed_name, tchar_name, tchar_nchars);
1720                 fixed_name_num_chars += tsprintf(fixed_name + tchar_nchars,
1721                                                  T(" (invalid filename #%lu)"),
1722                                                  ++ctx->invalid_sequence);
1723         #ifndef __WIN32__
1724                 FREE(tchar_name);
1725         #endif
1726                 dentry->extraction_name = memdup(fixed_name,
1727                                                  2 * fixed_name_num_chars + 2);
1728                 if (!dentry->extraction_name)
1729                         return WIMLIB_ERR_NOMEM;
1730                 dentry->extraction_name_nchars = fixed_name_num_chars;
1731         }
1732         return 0;
1733
1734 skip_dentry:
1735         for_dentry_in_tree(dentry, dentry_mark_skipped, NULL);
1736         return 0;
1737 }
1738
1739 /* Clean up dentry and inode structure after extraction.  */
1740 static int
1741 dentry_reset_needs_extraction(struct wim_dentry *dentry, void *_ignore)
1742 {
1743         struct wim_inode *inode = dentry->d_inode;
1744
1745         dentry->extraction_skipped = 0;
1746         dentry->was_hardlinked = 0;
1747         dentry->skeleton_extracted = 0;
1748         inode->i_visited = 0;
1749         FREE(inode->i_extracted_file);
1750         inode->i_extracted_file = NULL;
1751         inode->i_dos_name_extracted = 0;
1752         if ((void*)dentry->extraction_name != (void*)dentry->file_name)
1753                 FREE(dentry->extraction_name);
1754         dentry->extraction_name = NULL;
1755         return 0;
1756 }
1757
1758 /* Tally features necessary to extract a dentry and the corresponding inode.  */
1759 static int
1760 dentry_tally_features(struct wim_dentry *dentry, void *_features)
1761 {
1762         struct wim_features *features = _features;
1763         struct wim_inode *inode = dentry->d_inode;
1764
1765         if (inode->i_attributes & FILE_ATTRIBUTE_ARCHIVE)
1766                 features->archive_files++;
1767         if (inode->i_attributes & FILE_ATTRIBUTE_HIDDEN)
1768                 features->hidden_files++;
1769         if (inode->i_attributes & FILE_ATTRIBUTE_SYSTEM)
1770                 features->system_files++;
1771         if (inode->i_attributes & FILE_ATTRIBUTE_COMPRESSED)
1772                 features->compressed_files++;
1773         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
1774                 if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
1775                         features->encrypted_directories++;
1776                 else
1777                         features->encrypted_files++;
1778         }
1779         if (inode->i_attributes & FILE_ATTRIBUTE_NOT_CONTENT_INDEXED)
1780                 features->not_context_indexed_files++;
1781         if (inode->i_attributes & FILE_ATTRIBUTE_SPARSE_FILE)
1782                 features->sparse_files++;
1783         if (inode_has_named_stream(inode))
1784                 features->named_data_streams++;
1785         if (inode->i_visited)
1786                 features->hard_links++;
1787         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1788                 features->reparse_points++;
1789                 if (inode_is_symlink(inode))
1790                         features->symlink_reparse_points++;
1791                 else
1792                         features->other_reparse_points++;
1793         }
1794         if (inode->i_security_id != -1)
1795                 features->security_descriptors++;
1796         if (dentry->short_name_nbytes)
1797                 features->short_names++;
1798         if (inode_has_unix_data(inode))
1799                 features->unix_data++;
1800         inode->i_visited = 1;
1801         return 0;
1802 }
1803
1804 /* Tally the features necessary to extract a dentry tree.  */
1805 static void
1806 dentry_tree_get_features(struct wim_dentry *root, struct wim_features *features)
1807 {
1808         memset(features, 0, sizeof(struct wim_features));
1809         for_dentry_in_tree(root, dentry_tally_features, features);
1810         dentry_tree_clear_inode_visited(root);
1811 }
1812
1813 static u32
1814 compute_supported_attributes_mask(const struct wim_features *supported_features)
1815 {
1816         u32 mask = ~(u32)0;
1817
1818         if (!supported_features->archive_files)
1819                 mask &= ~FILE_ATTRIBUTE_ARCHIVE;
1820
1821         if (!supported_features->hidden_files)
1822                 mask &= ~FILE_ATTRIBUTE_HIDDEN;
1823
1824         if (!supported_features->system_files)
1825                 mask &= ~FILE_ATTRIBUTE_SYSTEM;
1826
1827         if (!supported_features->not_context_indexed_files)
1828                 mask &= ~FILE_ATTRIBUTE_NOT_CONTENT_INDEXED;
1829
1830         if (!supported_features->compressed_files)
1831                 mask &= ~FILE_ATTRIBUTE_COMPRESSED;
1832
1833         if (!supported_features->sparse_files)
1834                 mask &= ~FILE_ATTRIBUTE_SPARSE_FILE;
1835
1836         if (!supported_features->reparse_points)
1837                 mask &= ~FILE_ATTRIBUTE_REPARSE_POINT;
1838
1839         return mask;
1840 }
1841
1842 static int
1843 do_feature_check(const struct wim_features *required_features,
1844                  const struct wim_features *supported_features,
1845                  int extract_flags,
1846                  const struct apply_operations *ops,
1847                  const tchar *wim_source_path)
1848 {
1849         const tchar *loc;
1850         const tchar *mode = T("this extraction mode");
1851
1852         if (wim_source_path[0] == '\0')
1853                 loc = T("the WIM image");
1854         else
1855                 loc = wim_source_path;
1856
1857         /* We're an archive program, so theoretically we can do what we want
1858          * with FILE_ATTRIBUTE_ARCHIVE (which is a dumb flag anyway).  Don't
1859          * bother the user about it.  */
1860 #if 0
1861         if (required_features->archive_files && !supported_features->archive_files)
1862         {
1863                 WARNING(
1864           "%lu files in %"TS" are marked as archived, but this attribute\n"
1865 "          is not supported in %"TS".",
1866                         required_features->archive_files, loc, mode);
1867         }
1868 #endif
1869
1870         if (required_features->hidden_files && !supported_features->hidden_files)
1871         {
1872                 WARNING(
1873           "%lu files in %"TS" are marked as hidden, but this\n"
1874 "          attribute is not supported in %"TS".",
1875                         required_features->hidden_files, loc, mode);
1876         }
1877
1878         if (required_features->system_files && !supported_features->system_files)
1879         {
1880                 WARNING(
1881           "%lu files in %"TS" are marked as system files,\n"
1882 "          but this attribute is not supported in %"TS".",
1883                         required_features->system_files, loc, mode);
1884         }
1885
1886         if (required_features->compressed_files && !supported_features->compressed_files)
1887         {
1888                 WARNING(
1889           "%lu files in %"TS" are marked as being transparently\n"
1890 "          compressed, but transparent compression is not supported in\n"
1891 "          %"TS".  These files will be extracted as uncompressed.",
1892                         required_features->compressed_files, loc, mode);
1893         }
1894
1895         if (required_features->encrypted_files && !supported_features->encrypted_files)
1896         {
1897                 WARNING(
1898           "%lu files in %"TS" are marked as being encrypted,\n"
1899 "           but encryption is not supported in %"TS".  These files\n"
1900 "           will not be extracted.",
1901                         required_features->encrypted_files, loc, mode);
1902         }
1903
1904         if (required_features->encrypted_directories &&
1905             !supported_features->encrypted_directories)
1906         {
1907                 WARNING(
1908           "%lu directories in %"TS" are marked as being encrypted,\n"
1909 "           but encryption is not supported in %"TS".\n"
1910 "           These directories will be extracted as unencrypted.",
1911                         required_features->encrypted_directories, loc, mode);
1912         }
1913
1914         if (required_features->not_context_indexed_files &&
1915             !supported_features->not_context_indexed_files)
1916         {
1917                 WARNING(
1918           "%lu files in %"TS" are marked as not content indexed,\n"
1919 "          but this attribute is not supported in %"TS".",
1920                         required_features->not_context_indexed_files, loc, mode);
1921         }
1922
1923         if (required_features->sparse_files && !supported_features->sparse_files)
1924         {
1925                 WARNING(
1926           "%lu files in %"TS" are marked as sparse, but creating\n"
1927 "           sparse files is not supported in %"TS".  These files\n"
1928 "           will be extracted as non-sparse.",
1929                         required_features->sparse_files, loc, mode);
1930         }
1931
1932         if (required_features->named_data_streams &&
1933             !supported_features->named_data_streams)
1934         {
1935                 WARNING(
1936           "%lu files in %"TS" contain one or more alternate (named)\n"
1937 "          data streams, which are not supported in %"TS".\n"
1938 "          Alternate data streams will NOT be extracted.",
1939                         required_features->named_data_streams, loc, mode);
1940         }
1941
1942         if (unlikely(extract_flags & (WIMLIB_EXTRACT_FLAG_HARDLINK |
1943                                       WIMLIB_EXTRACT_FLAG_SYMLINK)) &&
1944             required_features->named_data_streams &&
1945             supported_features->named_data_streams)
1946         {
1947                 WARNING(
1948           "%lu files in %"TS" contain one or more alternate (named)\n"
1949 "          data streams, which are not supported in linked extraction mode.\n"
1950 "          Alternate data streams will NOT be extracted.",
1951                         required_features->named_data_streams, loc);
1952         }
1953
1954         if (required_features->hard_links && !supported_features->hard_links)
1955         {
1956                 WARNING(
1957           "%lu files in %"TS" are hard links, but hard links are\n"
1958 "          not supported in %"TS".  Hard links will be extracted as\n"
1959 "          duplicate copies of the linked files.",
1960                         required_features->hard_links, loc, mode);
1961         }
1962
1963         if (required_features->reparse_points && !supported_features->reparse_points)
1964         {
1965                 if (supported_features->symlink_reparse_points) {
1966                         if (required_features->other_reparse_points) {
1967                                 WARNING(
1968           "%lu files in %"TS" are reparse points that are neither\n"
1969 "          symbolic links nor junction points and are not supported in\n"
1970 "          %"TS".  These reparse points will not be extracted.",
1971                                         required_features->other_reparse_points, loc,
1972                                         mode);
1973                         }
1974                 } else {
1975                         WARNING(
1976           "%lu files in %"TS" are reparse points, which are\n"
1977 "          not supported in %"TS" and will not be extracted.",
1978                                 required_features->reparse_points, loc, mode);
1979                 }
1980         }
1981
1982         if (required_features->security_descriptors &&
1983             !supported_features->security_descriptors)
1984         {
1985                 WARNING(
1986           "%lu files in %"TS" have Windows NT security descriptors,\n"
1987 "          but extracting security descriptors is not supported in\n"
1988 "          %"TS".  No security descriptors will be extracted.",
1989                         required_features->security_descriptors, loc, mode);
1990         }
1991
1992         if (required_features->short_names && !supported_features->short_names)
1993         {
1994                 WARNING(
1995           "%lu files in %"TS" have short (DOS) names, but\n"
1996 "          extracting short names is not supported in %"TS".\n"
1997 "          Short names will not be extracted.\n",
1998                         required_features->short_names, loc, mode);
1999         }
2000
2001         if ((extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) &&
2002             required_features->unix_data && !supported_features->unix_data)
2003         {
2004                 ERROR("Extracting UNIX data is not supported in %"TS, mode);
2005                 return WIMLIB_ERR_UNSUPPORTED;
2006         }
2007         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES) &&
2008             required_features->short_names && !supported_features->short_names)
2009         {
2010                 ERROR("Extracting short names is not supported in %"TS"", mode);
2011                 return WIMLIB_ERR_UNSUPPORTED;
2012         }
2013         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS) &&
2014             !ops->set_timestamps)
2015         {
2016                 ERROR("Extracting timestamps is not supported in %"TS"", mode);
2017                 return WIMLIB_ERR_UNSUPPORTED;
2018         }
2019         if (((extract_flags & (WIMLIB_EXTRACT_FLAG_STRICT_ACLS |
2020                                WIMLIB_EXTRACT_FLAG_UNIX_DATA))
2021              == WIMLIB_EXTRACT_FLAG_STRICT_ACLS) &&
2022             required_features->security_descriptors &&
2023             !supported_features->security_descriptors)
2024         {
2025                 ERROR("Extracting security descriptors is not supported in %"TS, mode);
2026                 return WIMLIB_ERR_UNSUPPORTED;
2027         }
2028
2029         if ((extract_flags & WIMLIB_EXTRACT_FLAG_HARDLINK) &&
2030             !supported_features->hard_links)
2031         {
2032                 ERROR("Hard link extraction mode requested, but "
2033                       "%"TS" does not support hard links!", mode);
2034                 return WIMLIB_ERR_UNSUPPORTED;
2035         }
2036
2037         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS) &&
2038             required_features->symlink_reparse_points &&
2039             !(supported_features->symlink_reparse_points ||
2040               supported_features->reparse_points))
2041         {
2042                 ERROR("Extracting symbolic links is not supported in %"TS, mode);
2043                 return WIMLIB_ERR_UNSUPPORTED;
2044         }
2045
2046         if ((extract_flags & WIMLIB_EXTRACT_FLAG_SYMLINK) &&
2047             !supported_features->symlink_reparse_points)
2048         {
2049                 ERROR("Symbolic link extraction mode requested, but "
2050                       "%"TS" does not support symbolic "
2051                       "links!", mode);
2052                 return WIMLIB_ERR_UNSUPPORTED;
2053         }
2054         return 0;
2055 }
2056
2057 static void
2058 do_extract_warnings(struct apply_ctx *ctx)
2059 {
2060         if (ctx->partial_security_descriptors == 0 &&
2061             ctx->no_security_descriptors == 0)
2062                 return;
2063
2064         WARNING("Extraction of \"%"TS"\" complete, but with one or more warnings:",
2065                 ctx->target);
2066         if (ctx->partial_security_descriptors != 0) {
2067                 WARNING("- Could only partially set the security descriptor\n"
2068                         "            on %lu files or directories.",
2069                         ctx->partial_security_descriptors);
2070         }
2071         if (ctx->no_security_descriptors != 0) {
2072                 WARNING("- Could not set security descriptor at all\n"
2073                         "            on %lu files or directories.",
2074                         ctx->no_security_descriptors);
2075         }
2076 #ifdef __WIN32__
2077         WARNING("To fully restore all security descriptors, run the program\n"
2078                 "          with Administrator rights.");
2079 #endif
2080 }
2081
2082 /*
2083  * extract_tree - Extract a file or directory tree from the currently selected
2084  *                WIM image.
2085  *
2086  * @wim:        WIMStruct for the WIM file, with the desired image selected
2087  *              (as wim->current_image).
2088  *
2089  * @wim_source_path:
2090  *              "Canonical" (i.e. no leading or trailing slashes, path
2091  *              separators WIM_PATH_SEPARATOR) path inside the WIM image to
2092  *              extract.  An empty string means the full image.
2093  *
2094  * @target:
2095  *              Filesystem path to extract the file or directory tree to.
2096  *              (Or, with WIMLIB_EXTRACT_FLAG_NTFS: the name of a NTFS volume.)
2097  *
2098  * @extract_flags:
2099  *              WIMLIB_EXTRACT_FLAG_*.  Also, the private flag
2100  *              WIMLIB_EXTRACT_FLAG_MULTI_IMAGE will be set if this is being
2101  *              called through wimlib_extract_image() with WIMLIB_ALL_IMAGES as
2102  *              the image.
2103  *
2104  * @progress_func:
2105  *              If non-NULL, progress function for the extraction.  The messages
2106  *              that may be sent in this function are:
2107  *
2108  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN or
2109  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN;
2110  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN;
2111  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END;
2112  *              WIMLIB_PROGRESS_MSG_EXTRACT_DENTRY;
2113  *              WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS;
2114  *              WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS;
2115  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END or
2116  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END.
2117  *
2118  * Returns 0 on success; a positive WIMLIB_ERR_* code on failure.
2119  */
2120 static int
2121 extract_tree(WIMStruct *wim, const tchar *wim_source_path, const tchar *target,
2122              int extract_flags, wimlib_progress_func_t progress_func)
2123 {
2124         struct wim_dentry *root;
2125         struct wim_features required_features;
2126         struct apply_ctx ctx;
2127         int ret;
2128         struct wim_lookup_table_entry *lte;
2129
2130         /* Start initializing the apply_ctx.  */
2131         memset(&ctx, 0, sizeof(struct apply_ctx));
2132         ctx.wim = wim;
2133         ctx.extract_flags = extract_flags;
2134         ctx.target = target;
2135         ctx.target_nchars = tstrlen(target);
2136         ctx.progress_func = progress_func;
2137         if (progress_func) {
2138                 ctx.progress.extract.wimfile_name = wim->filename;
2139                 ctx.progress.extract.image = wim->current_image;
2140                 ctx.progress.extract.extract_flags = (extract_flags &
2141                                                       WIMLIB_EXTRACT_MASK_PUBLIC);
2142                 ctx.progress.extract.image_name = wimlib_get_image_name(wim,
2143                                                                         wim->current_image);
2144                 ctx.progress.extract.extract_root_wim_source_path = wim_source_path;
2145                 ctx.progress.extract.target = target;
2146         }
2147         INIT_LIST_HEAD(&ctx.stream_list);
2148
2149         /* Translate the path to extract into the corresponding
2150          * `struct wim_dentry', which will be the root of the
2151          * "dentry tree" to extract.  */
2152         root = get_dentry(wim, wim_source_path);
2153         if (!root) {
2154                 ERROR("Path \"%"TS"\" does not exist in WIM image %d",
2155                       wim_source_path, wim->current_image);
2156                 ret = WIMLIB_ERR_PATH_DOES_NOT_EXIST;
2157                 goto out;
2158         }
2159
2160         ctx.extract_root = root;
2161
2162         /* Select the appropriate apply_operations based on the
2163          * platform and extract_flags.  */
2164 #ifdef __WIN32__
2165         ctx.ops = &win32_apply_ops;
2166 #else
2167         ctx.ops = &unix_apply_ops;
2168 #endif
2169
2170 #ifdef WITH_NTFS_3G
2171         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS)
2172                 ctx.ops = &ntfs_3g_apply_ops;
2173 #endif
2174
2175         /* Call the start_extract() callback.  This gives the apply_operations
2176          * implementation a chance to do any setup needed to access the volume.
2177          * Furthermore, it's expected to set the supported features of this
2178          * extraction mode (ctx.supported_features), which are determined at
2179          * runtime as they may vary depending on the actual volume.  These
2180          * features are then compared with the actual features extracting this
2181          * dentry tree requires.  Some mismatches will merely produce warnings
2182          * and the unsupported data will be ignored; others will produce errors.
2183          */
2184         ret = ctx.ops->start_extract(target, &ctx);
2185         if (ret)
2186                 goto out;
2187
2188         dentry_tree_get_features(root, &required_features);
2189         ret = do_feature_check(&required_features, &ctx.supported_features,
2190                                extract_flags, ctx.ops, wim_source_path);
2191         if (ret)
2192                 goto out_finish_or_abort_extract;
2193
2194         ctx.supported_attributes_mask =
2195                 compute_supported_attributes_mask(&ctx.supported_features);
2196
2197         /* Figure out whether the root dentry is being extracted to the root of
2198          * a volume and therefore needs to be treated "specially", for example
2199          * not being explicitly created and not having attributes set.  */
2200         if (ctx.ops->target_is_root && ctx.ops->root_directory_is_special)
2201                 ctx.root_dentry_is_special = ctx.ops->target_is_root(target);
2202
2203         /* Calculate the actual filename component of each extracted dentry.  In
2204          * the process, set the dentry->extraction_skipped flag on dentries that
2205          * are being skipped for some reason (e.g. invalid filename).  */
2206         ret = for_dentry_in_tree(root, dentry_calculate_extraction_path, &ctx);
2207         if (ret)
2208                 goto out_dentry_reset_needs_extraction;
2209
2210         /* Build the list of the streams that need to be extracted and
2211          * initialize ctx.progress.extract with stream information.  */
2212         ret = for_dentry_in_tree(ctx.extract_root,
2213                                  dentry_resolve_and_zero_lte_refcnt, &ctx);
2214         if (ret)
2215                 goto out_dentry_reset_needs_extraction;
2216
2217         ret = for_dentry_in_tree(ctx.extract_root,
2218                                  dentry_add_streams_to_extract, &ctx);
2219         if (ret)
2220                 goto out_teardown_stream_list;
2221
2222         if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2223                 /* When extracting from a pipe, the number of bytes of data to
2224                  * extract can't be determined in the normal way (examining the
2225                  * lookup table), since at this point all we have is a set of
2226                  * SHA1 message digests of streams that need to be extracted.
2227                  * However, we can get a reasonably accurate estimate by taking
2228                  * <TOTALBYTES> from the corresponding <IMAGE> in the WIM XML
2229                  * data.  This does assume that a full image is being extracted,
2230                  * but currently there is no API for doing otherwise.  (Also,
2231                  * subtract <HARDLINKBYTES> from this if hard links are
2232                  * supported by the extraction mode.)  */
2233                 ctx.progress.extract.total_bytes =
2234                         wim_info_get_image_total_bytes(wim->wim_info,
2235                                                        wim->current_image);
2236                 if (ctx.supported_features.hard_links) {
2237                         ctx.progress.extract.total_bytes -=
2238                                 wim_info_get_image_hard_link_bytes(wim->wim_info,
2239                                                                    wim->current_image);
2240                 }
2241         }
2242
2243         /* Handle the special case of extracting a file to standard
2244          * output.  In that case, "root" should be a single file, not a
2245          * directory tree.  (If not, extract_dentry_to_stdout() will
2246          * return an error.)  */
2247         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT) {
2248                 ret = extract_dentry_to_stdout(root);
2249                 goto out_teardown_stream_list;
2250         }
2251
2252         /* If a sequential extraction was specified, sort the streams to be
2253          * extracted by their position in the WIM file so that the WIM file can
2254          * be read sequentially.  */
2255         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2256                               WIMLIB_EXTRACT_FLAG_FROM_PIPE))
2257                                         == WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2258         {
2259                 ret = sort_stream_list_by_sequential_order(
2260                                 &ctx.stream_list,
2261                                 offsetof(struct wim_lookup_table_entry,
2262                                          extraction_list));
2263                 if (ret)
2264                         goto out_teardown_stream_list;
2265         }
2266
2267         if (ctx.ops->realpath_works_on_nonexisting_files &&
2268             ((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) ||
2269              ctx.ops->requires_realtarget_in_paths))
2270         {
2271                 ctx.realtarget = realpath(target, NULL);
2272                 if (!ctx.realtarget) {
2273                         ret = WIMLIB_ERR_NOMEM;
2274                         goto out_teardown_stream_list;
2275                 }
2276                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2277         }
2278
2279         if (progress_func) {
2280                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN :
2281                                                  WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN,
2282                               &ctx.progress);
2283         }
2284
2285         if (!ctx.root_dentry_is_special)
2286         {
2287                 tchar path[ctx.ops->path_max];
2288                 if (build_extraction_path(path, root, &ctx))
2289                 {
2290                         ret = extract_inode(path, &ctx, root->d_inode);
2291                         if (ret)
2292                                 goto out_free_realtarget;
2293                 }
2294         }
2295
2296         /* If we need to fix up the targets of absolute symbolic links
2297          * (WIMLIB_EXTRACT_FLAG_RPFIX) or the extraction mode requires paths to
2298          * be absolute, use realpath() (or its replacement on Windows) to get
2299          * the absolute path to the extraction target.  Note that this requires
2300          * the target directory to exist, unless
2301          * realpath_works_on_nonexisting_files is set in the apply_operations.
2302          * */
2303         if (!ctx.realtarget &&
2304             (((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
2305               required_features.symlink_reparse_points) ||
2306              ctx.ops->requires_realtarget_in_paths))
2307         {
2308                 ctx.realtarget = realpath(target, NULL);
2309                 if (!ctx.realtarget) {
2310                         ret = WIMLIB_ERR_NOMEM;
2311                         goto out_free_realtarget;
2312                 }
2313                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2314         }
2315
2316         if (ctx.ops->requires_short_name_reordering) {
2317                 ret = for_dentry_in_tree(root, dentry_extract_dir_skeleton,
2318                                          &ctx);
2319                 if (ret)
2320                         goto out_free_realtarget;
2321         }
2322
2323         /* Finally, the important part: extract the tree of files.  */
2324         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2325                              WIMLIB_EXTRACT_FLAG_FROM_PIPE)) {
2326                 /* Sequential extraction requested, so two passes are needed
2327                  * (one for directory structure, one for streams.)  */
2328                 if (progress_func)
2329                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2330                                       &ctx.progress);
2331
2332                 if (!(extract_flags & WIMLIB_EXTRACT_FLAG_RESUME)) {
2333                         ret = for_dentry_in_tree(root, dentry_extract_skeleton, &ctx);
2334                         if (ret)
2335                                 goto out_free_realtarget;
2336                 }
2337                 if (progress_func)
2338                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2339                                       &ctx.progress);
2340                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE)
2341                         ret = extract_streams_from_pipe(&ctx);
2342                 else
2343                         ret = extract_stream_list(&ctx);
2344                 if (ret)
2345                         goto out_free_realtarget;
2346         } else {
2347                 /* Sequential extraction was not requested, so we can make do
2348                  * with one pass where we both create the files and extract
2349                  * streams.   */
2350                 if (progress_func)
2351                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2352                                       &ctx.progress);
2353                 ret = for_dentry_in_tree(root, dentry_extract, &ctx);
2354                 if (ret)
2355                         goto out_free_realtarget;
2356                 if (progress_func)
2357                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2358                                       &ctx.progress);
2359         }
2360
2361         /* If the total number of bytes to extract was miscalculated, just jump
2362          * to the calculated number in order to avoid confusing the progress
2363          * function.  This should only occur when extracting from a pipe.  */
2364         if (ctx.progress.extract.completed_bytes != ctx.progress.extract.total_bytes)
2365         {
2366                 DEBUG("Calculated %"PRIu64" bytes to extract, but actually "
2367                       "extracted %"PRIu64,
2368                       ctx.progress.extract.total_bytes,
2369                       ctx.progress.extract.completed_bytes);
2370         }
2371         if (progress_func &&
2372             ctx.progress.extract.completed_bytes < ctx.progress.extract.total_bytes)
2373         {
2374                 ctx.progress.extract.completed_bytes = ctx.progress.extract.total_bytes;
2375                 progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS, &ctx.progress);
2376         }
2377
2378         /* Apply security descriptors and timestamps.  This is done at the end,
2379          * and in a depth-first manner, to prevent timestamps from getting
2380          * changed by subsequent extract operations and to minimize the chance
2381          * of the restored security descriptors getting in our way.  */
2382         if (progress_func)
2383                 progress_func(WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS,
2384                               &ctx.progress);
2385         ret = for_dentry_in_tree_depth(root, dentry_extract_final, &ctx);
2386         if (ret)
2387                 goto out_free_realtarget;
2388
2389         if (progress_func) {
2390                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END :
2391                               WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END,
2392                               &ctx.progress);
2393         }
2394
2395         do_extract_warnings(&ctx);
2396
2397         ret = 0;
2398 out_free_realtarget:
2399         FREE(ctx.realtarget);
2400 out_teardown_stream_list:
2401         /* Free memory allocated as part of the mapping from each
2402          * wim_lookup_table_entry to the dentries that reference it.  */
2403         if (ctx.extract_flags & WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2404                 list_for_each_entry(lte, &ctx.stream_list, extraction_list)
2405                         if (lte->out_refcnt > ARRAY_LEN(lte->inline_lte_dentries))
2406                                 FREE(lte->lte_dentries);
2407 out_dentry_reset_needs_extraction:
2408         for_dentry_in_tree(root, dentry_reset_needs_extraction, NULL);
2409 out_finish_or_abort_extract:
2410         if (ret) {
2411                 if (ctx.ops->abort_extract)
2412                         ctx.ops->abort_extract(&ctx);
2413         } else {
2414                 if (ctx.ops->finish_extract)
2415                         ret = ctx.ops->finish_extract(&ctx);
2416         }
2417 out:
2418         return ret;
2419 }
2420
2421 /* Validates a single wimlib_extract_command, mostly checking to make sure the
2422  * extract flags make sense. */
2423 static int
2424 check_extract_command(struct wimlib_extract_command *cmd, int wim_header_flags)
2425 {
2426         int extract_flags;
2427
2428         /* Empty destination path? */
2429         if (cmd->fs_dest_path[0] == T('\0'))
2430                 return WIMLIB_ERR_INVALID_PARAM;
2431
2432         extract_flags = cmd->extract_flags;
2433
2434         /* Check for invalid flag combinations  */
2435         if ((extract_flags &
2436              (WIMLIB_EXTRACT_FLAG_SYMLINK |
2437               WIMLIB_EXTRACT_FLAG_HARDLINK)) == (WIMLIB_EXTRACT_FLAG_SYMLINK |
2438                                                  WIMLIB_EXTRACT_FLAG_HARDLINK))
2439                 return WIMLIB_ERR_INVALID_PARAM;
2440
2441         if ((extract_flags &
2442              (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2443               WIMLIB_EXTRACT_FLAG_STRICT_ACLS)) == (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2444                                                     WIMLIB_EXTRACT_FLAG_STRICT_ACLS))
2445                 return WIMLIB_ERR_INVALID_PARAM;
2446
2447         if ((extract_flags &
2448              (WIMLIB_EXTRACT_FLAG_RPFIX |
2449               WIMLIB_EXTRACT_FLAG_NORPFIX)) == (WIMLIB_EXTRACT_FLAG_RPFIX |
2450                                                 WIMLIB_EXTRACT_FLAG_NORPFIX))
2451                 return WIMLIB_ERR_INVALID_PARAM;
2452
2453         if ((extract_flags &
2454              (WIMLIB_EXTRACT_FLAG_RESUME |
2455               WIMLIB_EXTRACT_FLAG_FROM_PIPE)) == WIMLIB_EXTRACT_FLAG_RESUME)
2456                 return WIMLIB_ERR_INVALID_PARAM;
2457
2458         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2459 #ifndef WITH_NTFS_3G
2460                 ERROR("wimlib was compiled without support for NTFS-3g, so\n"
2461                       "        we cannot apply a WIM image directly to a NTFS volume.");
2462                 return WIMLIB_ERR_UNSUPPORTED;
2463 #endif
2464         }
2465
2466         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_RPFIX |
2467                               WIMLIB_EXTRACT_FLAG_NORPFIX)) == 0)
2468         {
2469                 /* Do reparse point fixups by default if the WIM header says
2470                  * they are enabled and we are extracting a full image. */
2471                 if (wim_header_flags & WIM_HDR_FLAG_RP_FIX)
2472                         extract_flags |= WIMLIB_EXTRACT_FLAG_RPFIX;
2473         }
2474
2475         /* TODO: Since UNIX data entries are stored in the file resources, in a
2476          * completely sequential extraction they may come up before the
2477          * corresponding file or symbolic link data.  This needs to be handled
2478          * better.  */
2479         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2480                               WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2481                                     == (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2482                                         WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2483         {
2484                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2485                         WARNING("Setting UNIX file/owner group may "
2486                                 "be impossible on some\n"
2487                                 "          symbolic links "
2488                                 "when applying from a pipe.");
2489                 } else {
2490                         extract_flags &= ~WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2491                         WARNING("Disabling sequential extraction for "
2492                                 "UNIX data mode");
2493                 }
2494         }
2495
2496         cmd->extract_flags = extract_flags;
2497         return 0;
2498 }
2499
2500
2501 /* Internal function to execute extraction commands for a WIM image.  The paths
2502  * in the extract commands are expected to be already "canonicalized".  */
2503 static int
2504 do_wimlib_extract_files(WIMStruct *wim,
2505                         int image,
2506                         struct wimlib_extract_command *cmds,
2507                         size_t num_cmds,
2508                         wimlib_progress_func_t progress_func)
2509 {
2510         int ret;
2511         bool found_link_cmd = false;
2512         bool found_nolink_cmd = false;
2513
2514         /* Select the image from which we are extracting files */
2515         ret = select_wim_image(wim, image);
2516         if (ret)
2517                 return ret;
2518
2519         /* Make sure there are no streams in the WIM that have not been
2520          * checksummed yet.  */
2521         ret = wim_checksum_unhashed_streams(wim);
2522         if (ret)
2523                 return ret;
2524
2525         /* Check for problems with the extraction commands */
2526         for (size_t i = 0; i < num_cmds; i++) {
2527                 ret = check_extract_command(&cmds[i], wim->hdr.flags);
2528                 if (ret)
2529                         return ret;
2530                 if (cmds[i].extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2531                                              WIMLIB_EXTRACT_FLAG_HARDLINK)) {
2532                         found_link_cmd = true;
2533                 } else {
2534                         found_nolink_cmd = true;
2535                 }
2536                 if (found_link_cmd && found_nolink_cmd) {
2537                         ERROR("Symlink or hardlink extraction mode must "
2538                               "be set on all extraction commands");
2539                         return WIMLIB_ERR_INVALID_PARAM;
2540                 }
2541         }
2542
2543         /* Execute the extraction commands */
2544         for (size_t i = 0; i < num_cmds; i++) {
2545                 ret = extract_tree(wim,
2546                                    cmds[i].wim_source_path,
2547                                    cmds[i].fs_dest_path,
2548                                    cmds[i].extract_flags,
2549                                    progress_func);
2550                 if (ret)
2551                         return ret;
2552         }
2553         return 0;
2554 }
2555
2556 /* API function documented in wimlib.h  */
2557 WIMLIBAPI int
2558 wimlib_extract_files(WIMStruct *wim,
2559                      int image,
2560                      const struct wimlib_extract_command *cmds,
2561                      size_t num_cmds,
2562                      int default_extract_flags,
2563                      wimlib_progress_func_t progress_func)
2564 {
2565         int ret;
2566         struct wimlib_extract_command *cmds_copy;
2567         int all_flags = 0;
2568
2569         default_extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2570
2571         if (num_cmds == 0)
2572                 return 0;
2573
2574         cmds_copy = CALLOC(num_cmds, sizeof(cmds[0]));
2575         if (!cmds_copy)
2576                 return WIMLIB_ERR_NOMEM;
2577
2578         for (size_t i = 0; i < num_cmds; i++) {
2579                 cmds_copy[i].extract_flags = (default_extract_flags |
2580                                                  cmds[i].extract_flags)
2581                                                 & WIMLIB_EXTRACT_MASK_PUBLIC;
2582                 all_flags |= cmds_copy[i].extract_flags;
2583
2584                 cmds_copy[i].wim_source_path = canonicalize_wim_path(cmds[i].wim_source_path);
2585                 if (!cmds_copy[i].wim_source_path) {
2586                         ret = WIMLIB_ERR_NOMEM;
2587                         goto out_free_cmds_copy;
2588                 }
2589
2590                 cmds_copy[i].fs_dest_path = canonicalize_fs_path(cmds[i].fs_dest_path);
2591                 if (!cmds_copy[i].fs_dest_path) {
2592                         ret = WIMLIB_ERR_NOMEM;
2593                         goto out_free_cmds_copy;
2594                 }
2595
2596         }
2597         ret = do_wimlib_extract_files(wim, image,
2598                                       cmds_copy, num_cmds,
2599                                       progress_func);
2600
2601         if (all_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2602                          WIMLIB_EXTRACT_FLAG_HARDLINK))
2603         {
2604                 for_lookup_table_entry(wim->lookup_table,
2605                                        lte_free_extracted_file, NULL);
2606         }
2607 out_free_cmds_copy:
2608         for (size_t i = 0; i < num_cmds; i++) {
2609                 FREE(cmds_copy[i].wim_source_path);
2610                 FREE(cmds_copy[i].fs_dest_path);
2611         }
2612         FREE(cmds_copy);
2613         return ret;
2614 }
2615
2616 /*
2617  * Extracts an image from a WIM file.
2618  *
2619  * @wim:                WIMStruct for the WIM file.
2620  *
2621  * @image:              Number of the single image to extract.
2622  *
2623  * @target:             Directory or NTFS volume to extract the image to.
2624  *
2625  * @extract_flags:      Bitwise or of WIMLIB_EXTRACT_FLAG_*.
2626  *
2627  * @progress_func:      If non-NULL, a progress function to be called
2628  *                      periodically.
2629  *
2630  * Returns 0 on success; nonzero on failure.
2631  */
2632 static int
2633 extract_single_image(WIMStruct *wim, int image,
2634                      const tchar *target, int extract_flags,
2635                      wimlib_progress_func_t progress_func)
2636 {
2637         int ret;
2638         tchar *target_copy = canonicalize_fs_path(target);
2639         if (!target_copy)
2640                 return WIMLIB_ERR_NOMEM;
2641         struct wimlib_extract_command cmd = {
2642                 .wim_source_path = T(""),
2643                 .fs_dest_path = target_copy,
2644                 .extract_flags = extract_flags,
2645         };
2646         ret = do_wimlib_extract_files(wim, image, &cmd, 1, progress_func);
2647         FREE(target_copy);
2648         return ret;
2649 }
2650
2651 static const tchar * const filename_forbidden_chars =
2652 T(
2653 #ifdef __WIN32__
2654 "<>:\"/\\|?*"
2655 #else
2656 "/"
2657 #endif
2658 );
2659
2660 /* This function checks if it is okay to use a WIM image's name as a directory
2661  * name.  */
2662 static bool
2663 image_name_ok_as_dir(const tchar *image_name)
2664 {
2665         return image_name && *image_name &&
2666                 !tstrpbrk(image_name, filename_forbidden_chars) &&
2667                 tstrcmp(image_name, T(".")) &&
2668                 tstrcmp(image_name, T(".."));
2669 }
2670
2671 /* Extracts all images from the WIM to the directory @target, with the images
2672  * placed in subdirectories named by their image names. */
2673 static int
2674 extract_all_images(WIMStruct *wim,
2675                    const tchar *target,
2676                    int extract_flags,
2677                    wimlib_progress_func_t progress_func)
2678 {
2679         size_t image_name_max_len = max(xml_get_max_image_name_len(wim), 20);
2680         size_t output_path_len = tstrlen(target);
2681         tchar buf[output_path_len + 1 + image_name_max_len + 1];
2682         int ret;
2683         int image;
2684         const tchar *image_name;
2685         struct stat stbuf;
2686
2687         extract_flags |= WIMLIB_EXTRACT_FLAG_MULTI_IMAGE;
2688
2689         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2690                 ERROR("Cannot extract multiple images in NTFS extraction mode.");
2691                 return WIMLIB_ERR_INVALID_PARAM;
2692         }
2693
2694         if (tstat(target, &stbuf)) {
2695                 if (errno == ENOENT) {
2696                         if (tmkdir(target, 0755)) {
2697                                 ERROR_WITH_ERRNO("Failed to create directory \"%"TS"\"", target);
2698                                 return WIMLIB_ERR_MKDIR;
2699                         }
2700                 } else {
2701                         ERROR_WITH_ERRNO("Failed to stat \"%"TS"\"", target);
2702                         return WIMLIB_ERR_STAT;
2703                 }
2704         } else if (!S_ISDIR(stbuf.st_mode)) {
2705                 ERROR("\"%"TS"\" is not a directory", target);
2706                 return WIMLIB_ERR_NOTDIR;
2707         }
2708
2709         tmemcpy(buf, target, output_path_len);
2710         buf[output_path_len] = OS_PREFERRED_PATH_SEPARATOR;
2711         for (image = 1; image <= wim->hdr.image_count; image++) {
2712                 image_name = wimlib_get_image_name(wim, image);
2713                 if (image_name_ok_as_dir(image_name)) {
2714                         tstrcpy(buf + output_path_len + 1, image_name);
2715                 } else {
2716                         /* Image name is empty or contains forbidden characters.
2717                          * Use image number instead. */
2718                         tsprintf(buf + output_path_len + 1, T("%d"), image);
2719                 }
2720                 ret = extract_single_image(wim, image, buf, extract_flags,
2721                                            progress_func);
2722                 if (ret)
2723                         return ret;
2724         }
2725         return 0;
2726 }
2727
2728 static int
2729 do_wimlib_extract_image(WIMStruct *wim,
2730                         int image,
2731                         const tchar *target,
2732                         int extract_flags,
2733                         wimlib_progress_func_t progress_func)
2734 {
2735         int ret;
2736
2737         if (image == WIMLIB_ALL_IMAGES) {
2738                 ret = extract_all_images(wim, target, extract_flags,
2739                                          progress_func);
2740         } else {
2741                 ret = extract_single_image(wim, image, target, extract_flags,
2742                                            progress_func);
2743         }
2744
2745         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2746                              WIMLIB_EXTRACT_FLAG_HARDLINK))
2747         {
2748                 for_lookup_table_entry(wim->lookup_table,
2749                                        lte_free_extracted_file,
2750                                        NULL);
2751         }
2752         return ret;
2753 }
2754
2755 /* API function documented in wimlib.h  */
2756 WIMLIBAPI int
2757 wimlib_extract_image_from_pipe(int pipe_fd, const tchar *image_num_or_name,
2758                                const tchar *target, int extract_flags,
2759                                wimlib_progress_func_t progress_func)
2760 {
2761         int ret;
2762         WIMStruct *pwm;
2763         struct filedes *in_fd;
2764         int image;
2765         unsigned i;
2766
2767         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2768
2769         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT)
2770                 return WIMLIB_ERR_INVALID_PARAM;
2771
2772         extract_flags |= WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2773
2774         /* Read the WIM header from the pipe and get a WIMStruct to represent
2775          * the pipable WIM.  Caveats:  Unlike getting a WIMStruct with
2776          * wimlib_open_wim(), getting a WIMStruct in this way will result in
2777          * an empty lookup table, no XML data read, and no filename set.  */
2778         ret = open_wim_as_WIMStruct(&pipe_fd,
2779                                     WIMLIB_OPEN_FLAG_FROM_PIPE |
2780                                                 WIMLIB_OPEN_FLAG_SPLIT_OK,
2781                                     &pwm, progress_func);
2782         if (ret)
2783                 return ret;
2784
2785         /* Sanity check to make sure this is a pipable WIM.  */
2786         if (pwm->hdr.magic != PWM_MAGIC) {
2787                 ERROR("The WIM being read from file descriptor %d "
2788                       "is not pipable!", pipe_fd);
2789                 ret = WIMLIB_ERR_NOT_PIPABLE;
2790                 goto out_wimlib_free;
2791         }
2792
2793         /* Sanity check to make sure the first part of a pipable split WIM is
2794          * sent over the pipe first.  */
2795         if (pwm->hdr.part_number != 1) {
2796                 ERROR("The first part of the split WIM must be "
2797                       "sent over the pipe first.");
2798                 ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2799                 goto out_wimlib_free;
2800         }
2801
2802         in_fd = &pwm->in_fd;
2803         wimlib_assert(in_fd->offset == WIM_HEADER_DISK_SIZE);
2804
2805         /* As mentioned, the WIMStruct we created from the pipe does not have
2806          * XML data yet.  Fix this by reading the extra copy of the XML data
2807          * that directly follows the header in pipable WIMs.  (Note: see
2808          * write_pipable_wim() for more details about the format of pipable
2809          * WIMs.)  */
2810         {
2811                 struct wim_lookup_table_entry xml_lte;
2812                 ret = read_pwm_stream_header(pwm, &xml_lte, 0, NULL);
2813                 if (ret)
2814                         goto out_wimlib_free;
2815
2816                 if (!(xml_lte.resource_entry.flags & WIM_RESHDR_FLAG_METADATA))
2817                 {
2818                         ERROR("Expected XML data, but found non-metadata "
2819                               "stream.");
2820                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2821                         goto out_wimlib_free;
2822                 }
2823
2824                 copy_resource_entry(&pwm->hdr.xml_res_entry,
2825                                     &xml_lte.resource_entry);
2826
2827                 ret = read_wim_xml_data(pwm);
2828                 if (ret)
2829                         goto out_wimlib_free;
2830                 if (wim_info_get_num_images(pwm->wim_info) != pwm->hdr.image_count) {
2831                         ERROR("Image count in XML data is not the same as in WIM header.");
2832                         ret = WIMLIB_ERR_XML;
2833                         goto out_wimlib_free;
2834                 }
2835         }
2836
2837         /* Get image index (this may use the XML data that was just read to
2838          * resolve an image name).  */
2839         if (image_num_or_name) {
2840                 image = wimlib_resolve_image(pwm, image_num_or_name);
2841                 if (image == WIMLIB_NO_IMAGE) {
2842                         ERROR("\"%"TS"\" is not a valid image in the pipable WIM!",
2843                               image_num_or_name);
2844                         ret = WIMLIB_ERR_INVALID_IMAGE;
2845                         goto out_wimlib_free;
2846                 } else if (image == WIMLIB_ALL_IMAGES) {
2847                         ERROR("Applying all images from a pipe is not supported.");
2848                         ret = WIMLIB_ERR_INVALID_IMAGE;
2849                         goto out_wimlib_free;
2850                 }
2851         } else {
2852                 if (pwm->hdr.image_count != 1) {
2853                         ERROR("No image was specified, but the pipable WIM "
2854                               "did not contain exactly 1 image");
2855                         ret = WIMLIB_ERR_INVALID_IMAGE;
2856                         goto out_wimlib_free;
2857                 }
2858                 image = 1;
2859         }
2860
2861         /* Load the needed metadata resource.  */
2862         for (i = 1; i <= pwm->hdr.image_count; i++) {
2863                 struct wim_lookup_table_entry *metadata_lte;
2864                 struct wim_image_metadata *imd;
2865
2866                 metadata_lte = new_lookup_table_entry();
2867                 if (!metadata_lte) {
2868                         ret = WIMLIB_ERR_NOMEM;
2869                         goto out_wimlib_free;
2870                 }
2871
2872                 ret = read_pwm_stream_header(pwm, metadata_lte, 0, NULL);
2873                 imd = pwm->image_metadata[i - 1];
2874                 imd->metadata_lte = metadata_lte;
2875                 if (ret)
2876                         goto out_wimlib_free;
2877
2878                 if (!(metadata_lte->resource_entry.flags &
2879                       WIM_RESHDR_FLAG_METADATA))
2880                 {
2881                         ERROR("Expected metadata resource, but found "
2882                               "non-metadata stream.");
2883                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2884                         goto out_wimlib_free;
2885                 }
2886
2887                 if (i == image) {
2888                         /* Metadata resource is for the images being extracted.
2889                          * Parse it and save the metadata in memory.  */
2890                         ret = read_metadata_resource(pwm, imd);
2891                         if (ret)
2892                                 goto out_wimlib_free;
2893                         imd->modified = 1;
2894                 } else {
2895                         /* Metadata resource is not for the image being
2896                          * extracted.  Skip over it.  */
2897                         ret = skip_pwm_stream(metadata_lte);
2898                         if (ret)
2899                                 goto out_wimlib_free;
2900                 }
2901         }
2902         /* Extract the image.  */
2903         extract_flags |= WIMLIB_EXTRACT_FLAG_FROM_PIPE;
2904         ret = do_wimlib_extract_image(pwm, image, target,
2905                                       extract_flags, progress_func);
2906         /* Clean up and return.  */
2907 out_wimlib_free:
2908         wimlib_free(pwm);
2909         return ret;
2910 }
2911
2912 /* API function documented in wimlib.h  */
2913 WIMLIBAPI int
2914 wimlib_extract_image(WIMStruct *wim,
2915                      int image,
2916                      const tchar *target,
2917                      int extract_flags,
2918                      wimlib_progress_func_t progress_func)
2919 {
2920         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2921         return do_wimlib_extract_image(wim, image, target, extract_flags,
2922                                        progress_func);
2923 }