extract.c: Fix extraction of streams of hard-linked files
[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 #include "wimlib/swm.h"
57 #ifdef __WIN32__
58 #  include "wimlib/win32.h" /* for realpath() equivalent */
59 #endif
60 #include "wimlib/xml.h"
61 #include "wimlib/wim.h"
62
63 #include <errno.h>
64 #include <fcntl.h>
65 #include <stdlib.h>
66 #include <sys/stat.h>
67 #include <unistd.h>
68
69 #define WIMLIB_EXTRACT_FLAG_MULTI_IMAGE 0x80000000
70 #define WIMLIB_EXTRACT_FLAG_FROM_PIPE   0x40000000
71 #define WIMLIB_EXTRACT_MASK_PUBLIC      0x3fffffff
72
73 /* Given a WIM dentry in the tree to be extracted, resolve all streams in the
74  * corresponding inode and set 'out_refcnt' in each to 0.  */
75 static int
76 dentry_resolve_and_zero_lte_refcnt(struct wim_dentry *dentry, void *_ctx)
77 {
78         struct apply_ctx *ctx = _ctx;
79         struct wim_inode *inode = dentry->d_inode;
80         struct wim_lookup_table_entry *lte;
81         int ret;
82         bool force = false;
83
84         if (dentry->extraction_skipped)
85                 return 0;
86
87         /* Special case:  when extracting from a pipe, the WIM lookup table is
88          * initially empty, so "resolving" an inode's streams is initially not
89          * possible.  However, we still need to keep track of which streams,
90          * identified by SHA1 message digests, need to be extracted, so we
91          * "resolve" the inode's streams anyway by allocating new entries.  */
92         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE)
93                 force = true;
94         ret = inode_resolve_ltes(inode, ctx->wim->lookup_table, force);
95         if (ret)
96                 return ret;
97         for (unsigned i = 0; i <= inode->i_num_ads; i++) {
98                 lte = inode_stream_lte_resolved(inode, i);
99                 if (lte)
100                         lte->out_refcnt = 0;
101         }
102         return 0;
103 }
104
105 static inline bool
106 is_linked_extraction(const struct apply_ctx *ctx)
107 {
108         return 0 != (ctx->extract_flags & (WIMLIB_EXTRACT_FLAG_HARDLINK |
109                                            WIMLIB_EXTRACT_FLAG_SYMLINK));
110 }
111
112 static inline bool
113 can_extract_named_data_streams(const struct apply_ctx *ctx)
114 {
115         return ctx->supported_features.named_data_streams &&
116                 !is_linked_extraction(ctx);
117 }
118
119 static int
120 ref_stream_to_extract(struct wim_lookup_table_entry *lte,
121                       struct wim_dentry *dentry, struct apply_ctx *ctx)
122 {
123         if (!lte)
124                 return 0;
125
126         /* Tally the size only for each extraction of the stream (not hard
127          * links).  */
128         if (!(dentry->d_inode->i_visited &&
129               ctx->supported_features.hard_links) &&
130             (!is_linked_extraction(ctx) || (lte->out_refcnt == 0 &&
131                                             lte->extracted_file == NULL)))
132         {
133                 ctx->progress.extract.total_bytes += wim_resource_size(lte);
134                 ctx->progress.extract.num_streams++;
135         }
136
137         /* Add stream to the extraction_list only one time, even if it's going
138          * to be extracted to multiple locations.  */
139         if (lte->out_refcnt == 0) {
140                 list_add_tail(&lte->extraction_list, &ctx->stream_list);
141                 ctx->num_streams_remaining++;
142         }
143
144         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_SEQUENTIAL) {
145                 struct wim_dentry **lte_dentries;
146
147                 /* Append dentry to this stream's array of dentries referencing
148                  * it.  Use inline array to avoid memory allocation until the
149                  * number of dentries becomes too large.  */
150                 if (lte->out_refcnt < ARRAY_LEN(lte->inline_lte_dentries)) {
151                         lte_dentries = lte->inline_lte_dentries;
152                 } else {
153                         struct wim_dentry **prev_lte_dentries;
154                         size_t alloc_lte_dentries;
155
156                         if (lte->out_refcnt == ARRAY_LEN(lte->inline_lte_dentries)) {
157                                 prev_lte_dentries = NULL;
158                                 alloc_lte_dentries = ARRAY_LEN(lte->inline_lte_dentries);
159                         } else {
160                                 prev_lte_dentries = lte->lte_dentries;
161                                 alloc_lte_dentries = lte->alloc_lte_dentries;
162                         }
163
164                         if (lte->out_refcnt == alloc_lte_dentries) {
165                                 alloc_lte_dentries *= 2;
166                                 lte_dentries = REALLOC(prev_lte_dentries,
167                                                        alloc_lte_dentries *
168                                                         sizeof(lte_dentries[0]));
169                                 if (!lte_dentries)
170                                         return WIMLIB_ERR_NOMEM;
171                                 if (prev_lte_dentries == NULL) {
172                                         memcpy(lte_dentries,
173                                                lte->inline_lte_dentries,
174                                                sizeof(lte->inline_lte_dentries));
175                                 }
176                                 lte->lte_dentries = lte_dentries;
177                                 lte->alloc_lte_dentries = alloc_lte_dentries;
178                         }
179                         lte_dentries = lte->lte_dentries;
180                 }
181                 lte_dentries[lte->out_refcnt] = dentry;
182         }
183         lte->out_refcnt++;
184         return 0;
185 }
186
187 /* Given a WIM dentry in the tree to be extracted, iterate through streams that
188  * need to be extracted.  For each one, add it to the list of streams to be
189  * extracted (ctx->stream_list) if not already done so, and also update the
190  * progress information (ctx->progress) with the stream.  Furthermore, if doing
191  * a sequential extraction, build a mapping from each the stream to the dentries
192  * referencing it.
193  *
194  * This uses the i_visited member of the inodes (assumed to be 0 initially).  */
195 static int
196 dentry_add_streams_to_extract(struct wim_dentry *dentry, void *_ctx)
197 {
198         struct apply_ctx *ctx = _ctx;
199         struct wim_inode *inode = dentry->d_inode;
200         int ret;
201
202         /* Don't process dentries marked as skipped.  */
203         if (dentry->extraction_skipped)
204                 return 0;
205
206         /* The unnamed data stream will always be extracted, except in an
207          * unlikely case.  */
208         if (!inode_is_encrypted_directory(inode)) {
209                 ret = ref_stream_to_extract(inode_unnamed_lte_resolved(inode),
210                                             dentry, ctx);
211                 if (ret)
212                         return ret;
213         }
214
215         /* Named data streams will be extracted only if supported in the current
216          * extraction mode and volume, and to avoid complications, if not doing
217          * a linked extraction.  */
218         if (can_extract_named_data_streams(ctx)) {
219                 for (u16 i = 0; i < inode->i_num_ads; i++) {
220                         if (!ads_entry_is_named_stream(&inode->i_ads_entries[i]))
221                                 continue;
222                         ret = ref_stream_to_extract(inode->i_ads_entries[i].lte,
223                                                     dentry, ctx);
224                         if (ret)
225                                 return ret;
226                 }
227         }
228         inode->i_visited = 1;
229         return 0;
230 }
231
232 /* Inform library user of progress of stream extraction following the successful
233  * extraction of a copy of the stream specified by @lte.  */
234 static void
235 update_extract_progress(struct apply_ctx *ctx,
236                         const struct wim_lookup_table_entry *lte)
237 {
238         wimlib_progress_func_t progress_func = ctx->progress_func;
239         union wimlib_progress_info *progress = &ctx->progress;
240
241         progress->extract.completed_bytes += wim_resource_size(lte);
242         if (progress_func &&
243             progress->extract.completed_bytes >= ctx->next_progress)
244         {
245                 progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS, progress);
246                 if (progress->extract.completed_bytes >=
247                     progress->extract.total_bytes)
248                 {
249                         ctx->next_progress = ~0ULL;
250                 } else {
251                         ctx->next_progress += progress->extract.total_bytes / 128;
252                         if (ctx->next_progress > progress->extract.total_bytes)
253                                 ctx->next_progress = progress->extract.total_bytes;
254                 }
255         }
256 }
257
258 #ifndef __WIN32__
259 /* Extract a symbolic link (not directly as reparse data), handling fixing up
260  * the target of absolute symbolic links and updating the extract progress.
261  *
262  * @inode must specify the WIM inode for a symbolic link or junction reparse
263  * point.
264  *
265  * @lte_override overrides the resource used as the reparse data for the
266  * symbolic link.  */
267 static int
268 extract_symlink(const tchar *path, struct apply_ctx *ctx,
269                 struct wim_inode *inode,
270                 struct wim_lookup_table_entry *lte_override)
271 {
272         ssize_t bufsize = ctx->ops->path_max;
273         tchar target[bufsize];
274         tchar *buf = target;
275         tchar *fixed_target;
276         ssize_t sret;
277         int ret;
278
279         /* If absolute symbolic link fixups requested, reserve space in the link
280          * target buffer for the absolute path of the target directory.  */
281         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX)
282         {
283                 buf += ctx->realtarget_nchars;
284                 bufsize -= ctx->realtarget_nchars;
285         }
286
287         /* Translate the WIM inode's reparse data into the link target.  */
288         sret = wim_inode_readlink(inode, buf, bufsize - 1, lte_override);
289         if (sret < 0) {
290                 errno = -sret;
291                 return WIMLIB_ERR_READLINK;
292         }
293         buf[sret] = '\0';
294
295         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
296             buf[0] == '/')
297         {
298                 /* Fix absolute symbolic link target to point into the
299                  * actual extraction destination.  */
300                 tmemcpy(target, ctx->realtarget, ctx->realtarget_nchars);
301                 fixed_target = target;
302         } else {
303                 /* Keep same link target.  */
304                 fixed_target = buf;
305         }
306
307         /* Call into the apply_operations to create the symbolic link.  */
308         DEBUG("Creating symlink \"%"TS"\" => \"%"TS"\"",
309               path, fixed_target);
310         ret = ctx->ops->create_symlink(fixed_target, path, ctx);
311         if (ret) {
312                 ERROR_WITH_ERRNO("Failed to create symlink "
313                                  "\"%"TS"\" => \"%"TS"\"", path, fixed_target);
314                 return ret;
315         }
316
317         /* Account for reparse data consumed.  */
318         update_extract_progress(ctx,
319                                 (lte_override ? lte_override :
320                                       inode_unnamed_lte_resolved(inode)));
321         return 0;
322 }
323 #endif /* !__WIN32__ */
324
325 /* Create a file, directory, or symbolic link.  */
326 static int
327 extract_inode(const tchar *path, struct apply_ctx *ctx, struct wim_inode *inode)
328 {
329         int ret;
330
331 #ifndef __WIN32__
332         if (ctx->supported_features.symlink_reparse_points &&
333             !ctx->supported_features.reparse_points &&
334             inode_is_symlink(inode))
335         {
336                 ret = extract_symlink(path, ctx, inode, NULL);
337         } else
338 #endif /* !__WIN32__ */
339         if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) {
340                 ret = ctx->ops->create_directory(path, ctx, &inode->extract_cookie);
341                 if (ret) {
342                         ERROR_WITH_ERRNO("Failed to create the directory "
343                                          "\"%"TS"\"", path);
344                 }
345         } else if ((inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) &&
346                     ctx->ops->extract_encrypted_stream_creates_file &&
347                     ctx->supported_features.encrypted_files) {
348                 ret = ctx->ops->extract_encrypted_stream(
349                                 path, inode_unnamed_lte_resolved(inode), ctx);
350                 if (ret) {
351                         ERROR_WITH_ERRNO("Failed to create and extract "
352                                          "encrypted file \"%"TS"\"", path);
353                 }
354         } else {
355                 ret = ctx->ops->create_file(path, ctx, &inode->extract_cookie);
356                 if (ret) {
357                         ERROR_WITH_ERRNO("Failed to create the file "
358                                          "\"%"TS"\"", path);
359                 }
360         }
361         return ret;
362 }
363
364 static int
365 extract_hardlink(const tchar *oldpath, const tchar *newpath,
366                  struct apply_ctx *ctx)
367 {
368         int ret;
369
370         DEBUG("Creating hardlink \"%"TS"\" => \"%"TS"\"", newpath, oldpath);
371         ret = ctx->ops->create_hardlink(oldpath, newpath, ctx);
372         if (ret) {
373                 ERROR_WITH_ERRNO("Failed to create hardlink "
374                                  "\"%"TS"\" => \"%"TS"\"",
375                                  newpath, oldpath);
376         }
377         return ret;
378 }
379
380 #ifdef __WIN32__
381 static int
382 try_extract_rpfix(u8 *rpbuf,
383                   u16 *rpbuflen_p,
384                   const wchar_t *extract_root_realpath,
385                   unsigned extract_root_realpath_nchars)
386 {
387         struct reparse_data rpdata;
388         wchar_t *target;
389         size_t target_nchars;
390         size_t stripped_nchars;
391         wchar_t *stripped_target;
392         wchar_t stripped_target_nchars;
393         int ret;
394
395         utf16lechar *new_target;
396         utf16lechar *new_print_name;
397         size_t new_target_nchars;
398         size_t new_print_name_nchars;
399         utf16lechar *p;
400
401         ret = parse_reparse_data(rpbuf, *rpbuflen_p, &rpdata);
402         if (ret)
403                 return ret;
404
405         if (extract_root_realpath[0] == L'\0' ||
406             extract_root_realpath[1] != L':' ||
407             extract_root_realpath[2] != L'\\')
408                 return WIMLIB_ERR_REPARSE_POINT_FIXUP_FAILED;
409
410         ret = parse_substitute_name(rpdata.substitute_name,
411                                     rpdata.substitute_name_nbytes,
412                                     rpdata.rptag);
413         if (ret < 0)
414                 return 0;
415         stripped_nchars = ret;
416         target = rpdata.substitute_name;
417         target_nchars = rpdata.substitute_name_nbytes / sizeof(utf16lechar);
418         stripped_target = target + stripped_nchars;
419         stripped_target_nchars = target_nchars - stripped_nchars;
420
421         new_target = alloca((6 + extract_root_realpath_nchars +
422                              stripped_target_nchars) * sizeof(utf16lechar));
423
424         p = new_target;
425         if (stripped_nchars == 6) {
426                 /* Include \??\ prefix if it was present before */
427                 p = wmempcpy(p, L"\\??\\", 4);
428         }
429
430         /* Print name excludes the \??\ if present. */
431         new_print_name = p;
432         if (stripped_nchars != 0) {
433                 /* Get drive letter from real path to extract root, if a drive
434                  * letter was present before. */
435                 *p++ = extract_root_realpath[0];
436                 *p++ = extract_root_realpath[1];
437         }
438         /* Copy the rest of the extract root */
439         p = wmempcpy(p, extract_root_realpath + 2, extract_root_realpath_nchars - 2);
440
441         /* Append the stripped target */
442         p = wmempcpy(p, stripped_target, stripped_target_nchars);
443         new_target_nchars = p - new_target;
444         new_print_name_nchars = p - new_print_name;
445
446         if (new_target_nchars * sizeof(utf16lechar) >= REPARSE_POINT_MAX_SIZE ||
447             new_print_name_nchars * sizeof(utf16lechar) >= REPARSE_POINT_MAX_SIZE)
448                 return WIMLIB_ERR_REPARSE_POINT_FIXUP_FAILED;
449
450         rpdata.substitute_name = new_target;
451         rpdata.substitute_name_nbytes = new_target_nchars * sizeof(utf16lechar);
452         rpdata.print_name = new_print_name;
453         rpdata.print_name_nbytes = new_print_name_nchars * sizeof(utf16lechar);
454         return make_reparse_buffer(&rpdata, rpbuf, rpbuflen_p);
455 }
456 #endif /* __WIN32__ */
457
458 /* Set reparse data on extracted file or directory that has
459  * FILE_ATTRIBUTE_REPARSE_POINT set.  */
460 static int
461 extract_reparse_data(const tchar *path, struct apply_ctx *ctx,
462                      struct wim_inode *inode,
463                      struct wim_lookup_table_entry *lte_override)
464 {
465         int ret;
466         u8 rpbuf[REPARSE_POINT_MAX_SIZE];
467         u16 rpbuflen;
468
469         ret = wim_inode_get_reparse_data(inode, rpbuf, &rpbuflen, lte_override);
470         if (ret)
471                 goto error;
472
473 #ifdef __WIN32__
474         /* Fix up target of absolute symbolic link or junction points so
475          * that they point into the actual extraction target.  */
476         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
477             (inode->i_reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK ||
478              inode->i_reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT) &&
479             !inode->i_not_rpfixed)
480         {
481                 ret = try_extract_rpfix(rpbuf, &rpbuflen, ctx->realtarget,
482                                         ctx->realtarget_nchars);
483                 if (ret && !(ctx->extract_flags &
484                              WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS))
485                 {
486                         WARNING("Reparse point fixup of \"%"TS"\" "
487                                 "failed", path);
488                         ret = 0;
489                 }
490                 if (ret)
491                         goto error;
492         }
493 #endif
494
495         ret = ctx->ops->set_reparse_data(path, rpbuf, rpbuflen, ctx);
496
497         /* On Windows, the SeCreateSymbolicLink privilege is required to create
498          * symbolic links.  To be more friendly towards non-Administrator users,
499          * we merely warn the user if symbolic links cannot be created due to
500          * insufficient permissions or privileges, unless
501          * WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS was provided.  */
502 #ifdef __WIN32__
503         if (ret && inode_is_symlink(inode) &&
504             (errno == EACCES || errno == EPERM) &&
505             !(ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS))
506         {
507                 WARNING("Can't set reparse data on \"%"TS"\": "
508                         "Access denied!\n"
509                         "          You may be trying to "
510                         "extract a symbolic link without the\n"
511                         "          SeCreateSymbolicLink privilege, "
512                         "which by default non-Administrator\n"
513                         "          accounts do not have.",
514                         path);
515                 ret = 0;
516         }
517 #endif
518         if (ret)
519                 goto error;
520
521         /* Account for reparse data consumed.  */
522         update_extract_progress(ctx,
523                                 (lte_override ? lte_override :
524                                       inode_unnamed_lte_resolved(inode)));
525         return 0;
526
527 error:
528         ERROR_WITH_ERRNO("Failed to set reparse data on \"%"TS"\"", path);
529         return ret;
530 }
531
532 /*
533  * Extract zero or more streams to a file.
534  *
535  * This function operates slightly differently depending on whether @lte_spec is
536  * NULL or not.  When @lte_spec is NULL, the behavior is to extract the default
537  * file contents (unnamed stream), and, if named data streams are supported in
538  * the extract mode and volume, any named data streams.  When @lte_spec is NULL,
539  * the behavior is to extract only all copies of the stream @lte_spec, and in
540  * addition use @lte_spec to set the reparse data or create the symbolic link if
541  * appropriate.
542  *
543  * @path
544  *      Path to file to extract (as can be passed to apply_operations
545  *      functions).
546  * @ctx
547  *      Apply context.
548  * @dentry
549  *      WIM dentry that corresponds to the file being extracted.
550  * @lte_spec
551  *      If non-NULL, specifies the lookup table entry for a stream to extract,
552  *      and only that stream will be extracted (although there may be more than
553  *      one instance of it).
554  * @lte_override
555  *      Used only if @lte_spec != NULL; it is passed to the extraction functions
556  *      rather than @lte_spec, allowing the location of the stream to be
557  *      overridden.  (This is used when the WIM is being read from a nonseekable
558  *      file, such as a pipe, when streams need to be used more than once; each
559  *      such stream is extracted to a temporary file.)
560  */
561 static int
562 extract_streams(const tchar *path, struct apply_ctx *ctx,
563                 struct wim_dentry *dentry,
564                 struct wim_lookup_table_entry *lte_spec,
565                 struct wim_lookup_table_entry *lte_override)
566 {
567         struct wim_inode *inode = dentry->d_inode;
568         struct wim_lookup_table_entry *lte;
569         file_spec_t file_spec;
570         int ret;
571
572         if (dentry->was_hardlinked)
573                 return 0;
574
575 #ifdef ENABLE_DEBUG
576         if (lte_spec) {
577                 char sha1_str[100];
578                 char *p = sha1_str;
579                 for (unsigned i = 0; i < SHA1_HASH_SIZE; i++)
580                         p += sprintf(p, "%02x", lte_override->hash[i]);
581                 DEBUG("Extracting stream SHA1=%s to \"%"TS"\"",
582                       sha1_str, path, inode->i_ino);
583         } else {
584                 DEBUG("Extracting streams to \"%"TS"\"", path, inode->i_ino);
585         }
586 #endif
587
588         if (ctx->ops->uses_cookies)
589                 file_spec.cookie = inode->extract_cookie;
590         else
591                 file_spec.path = path;
592
593         /* Unnamed data stream.  */
594         lte = inode_unnamed_lte_resolved(inode);
595         if (lte && (!lte_spec || lte == lte_spec)) {
596                 if (lte_spec)
597                         lte = lte_override;
598                 if (!(inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY |
599                                              FILE_ATTRIBUTE_REPARSE_POINT)))
600                 {
601                         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED &&
602                             ctx->supported_features.encrypted_files) {
603                                 if (!ctx->ops->extract_encrypted_stream_creates_file) {
604                                         ret = ctx->ops->extract_encrypted_stream(
605                                                                 path, lte, ctx);
606                                         if (ret)
607                                                 goto error;
608                                 }
609                         } else {
610                                 ret = ctx->ops->extract_unnamed_stream(
611                                                         file_spec, lte, ctx);
612                                 if (ret)
613                                         goto error;
614                         }
615                         update_extract_progress(ctx, lte);
616                 }
617                 else if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)
618                 {
619                         ret = 0;
620                         if (ctx->supported_features.reparse_points)
621                                 ret = extract_reparse_data(path, ctx, inode, lte);
622                 #ifndef __WIN32__
623                         else if ((inode_is_symlink(inode) &&
624                                   ctx->supported_features.symlink_reparse_points))
625                                 ret = extract_symlink(path, ctx, inode, lte);
626                 #endif
627                         if (ret)
628                                 return ret;
629                 }
630         }
631
632         /* Named data streams.  */
633         if (can_extract_named_data_streams(ctx)) {
634                 for (u16 i = 0; i < inode->i_num_ads; i++) {
635                         struct wim_ads_entry *entry = &inode->i_ads_entries[i];
636
637                         if (!ads_entry_is_named_stream(entry))
638                                 continue;
639                         lte = entry->lte;
640                         if (!lte)
641                                 continue;
642                         if (lte_spec && lte_spec != lte)
643                                 continue;
644                         if (lte_spec)
645                                 lte = lte_override;
646                         ret = ctx->ops->extract_named_stream(file_spec, entry->stream_name,
647                                                              entry->stream_name_nbytes / 2,
648                                                              lte, ctx);
649                         if (ret)
650                                 goto error;
651                         update_extract_progress(ctx, lte);
652                 }
653         }
654         return 0;
655
656 error:
657         ERROR_WITH_ERRNO("Failed to extract data of \"%"TS"\"", path);
658         return ret;
659 }
660
661 /* Set attributes on an extracted file or directory if supported by the
662  * extraction mode.  */
663 static int
664 extract_file_attributes(const tchar *path, struct apply_ctx *ctx,
665                         struct wim_dentry *dentry, unsigned pass)
666 {
667         int ret;
668
669         if (ctx->ops->set_file_attributes &&
670             !(dentry == ctx->extract_root && ctx->root_dentry_is_special)) {
671                 u32 attributes = dentry->d_inode->i_attributes;
672
673                 /* Clear unsupported attributes.  */
674                 attributes &= ctx->supported_attributes_mask;
675
676                 if ((attributes & FILE_ATTRIBUTE_DIRECTORY &&
677                      !ctx->supported_features.encrypted_directories) ||
678                     (!(attributes & FILE_ATTRIBUTE_DIRECTORY) &&
679                      !ctx->supported_features.encrypted_files))
680                 {
681                         attributes &= ~FILE_ATTRIBUTE_ENCRYPTED;
682                 }
683
684                 if (attributes == 0)
685                         attributes = FILE_ATTRIBUTE_NORMAL;
686
687                 ret = ctx->ops->set_file_attributes(path, attributes, ctx, pass);
688                 if (ret) {
689                         ERROR_WITH_ERRNO("Failed to set attributes on "
690                                          "\"%"TS"\"", path);
691                         return ret;
692                 }
693         }
694         return 0;
695 }
696
697
698 /* Set or remove the short (DOS) name on an extracted file or directory if
699  * supported by the extraction mode.  Since DOS names are unimportant and it's
700  * easy to run into problems setting them on Windows (SetFileShortName()
701  * requires SE_RESTORE privilege, which only the Administrator can request, and
702  * also requires DELETE access to the file), failure is ignored unless
703  * WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES is set.  */
704 static int
705 extract_short_name(const tchar *path, struct apply_ctx *ctx,
706                    struct wim_dentry *dentry)
707 {
708         int ret;
709
710         /* The root of the dentry tree being extracted may not be extracted to
711          * its original name, so its short name should be ignored.  */
712         if (dentry == ctx->extract_root)
713                 return 0;
714
715         if (ctx->supported_features.short_names) {
716                 ret = ctx->ops->set_short_name(path,
717                                                dentry->short_name,
718                                                dentry->short_name_nbytes / 2,
719                                                ctx);
720                 if (ret && (ctx->extract_flags &
721                             WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES))
722                 {
723                         ERROR_WITH_ERRNO("Failed to set short name of "
724                                          "\"%"TS"\"", path);
725                         return ret;
726                 }
727         }
728         return 0;
729 }
730
731 /* Set security descriptor, UNIX data, or neither on an extracted file, taking
732  * into account the current extraction mode and flags.  */
733 static int
734 extract_security(const tchar *path, struct apply_ctx *ctx,
735                  struct wim_dentry *dentry)
736 {
737         int ret;
738         struct wim_inode *inode = dentry->d_inode;
739
740         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_NO_ACLS)
741                 return 0;
742
743         if ((ctx->extract_root == dentry) && ctx->root_dentry_is_special)
744                 return 0;
745
746 #ifndef __WIN32__
747         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) {
748                 struct wimlib_unix_data data;
749
750                 ret = inode_get_unix_data(inode, &data, NULL);
751                 if (ret < 0)
752                         ret = 0;
753                 else if (ret == 0)
754                         ret = ctx->ops->set_unix_data(path, &data, ctx);
755                 if (ret) {
756                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_ACLS) {
757                                 ERROR_WITH_ERRNO("Failed to set UNIX owner, "
758                                                  "group, and/or mode on "
759                                                  "\"%"TS"\"", path);
760                                 return ret;
761                         } else {
762                                 WARNING_WITH_ERRNO("Failed to set UNIX owner, "
763                                                    "group, and/or/mode on "
764                                                    "\"%"TS"\"", path);
765                         }
766                 }
767         }
768         else
769 #endif /* __WIN32__ */
770         if (ctx->supported_features.security_descriptors &&
771             inode->i_security_id != -1)
772         {
773                 const struct wim_security_data *sd;
774                 const u8 *desc;
775                 size_t desc_size;
776
777                 sd = wim_const_security_data(ctx->wim);
778                 desc = sd->descriptors[inode->i_security_id];
779                 desc_size = sd->sizes[inode->i_security_id];
780
781                 ret = ctx->ops->set_security_descriptor(path, desc,
782                                                         desc_size, ctx);
783                 if (ret) {
784                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_ACLS) {
785                                 ERROR_WITH_ERRNO("Failed to set security "
786                                                  "descriptor on \"%"TS"\"", path);
787                                 return ret;
788                         } else {
789                         #if 0
790                                 if (errno != EACCES) {
791                                         WARNING_WITH_ERRNO("Failed to set "
792                                                            "security descriptor "
793                                                            "on \"%"TS"\"", path);
794                                 }
795                         #endif
796                                 ctx->no_security_descriptors++;
797                         }
798                 }
799         }
800         return 0;
801 }
802
803 /* Set timestamps on an extracted file.  Failure is warning-only unless
804  * WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS is set.  */
805 static int
806 extract_timestamps(const tchar *path, struct apply_ctx *ctx,
807                    struct wim_dentry *dentry)
808 {
809         struct wim_inode *inode = dentry->d_inode;
810         int ret;
811
812         if ((ctx->extract_root == dentry) && ctx->root_dentry_is_special)
813                 return 0;
814
815         if (ctx->ops->set_timestamps) {
816                 ret = ctx->ops->set_timestamps(path,
817                                                inode->i_creation_time,
818                                                inode->i_last_write_time,
819                                                inode->i_last_access_time,
820                                                ctx);
821                 if (ret) {
822                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS) {
823                                 ERROR_WITH_ERRNO("Failed to set timestamps "
824                                                  "on \"%"TS"\"", path);
825                                 return ret;
826                         } else {
827                                 WARNING_WITH_ERRNO("Failed to set timestamps "
828                                                    "on \"%"TS"\"", path);
829                         }
830                 }
831         }
832         return 0;
833 }
834
835 /* Check whether the extraction of a dentry should be skipped completely.  */
836 static bool
837 dentry_is_supported(struct wim_dentry *dentry,
838                     const struct wim_features *supported_features)
839 {
840         struct wim_inode *inode = dentry->d_inode;
841
842         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
843                 return supported_features->reparse_points ||
844                         (inode_is_symlink(inode) &&
845                          supported_features->symlink_reparse_points);
846         }
847         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
848                 if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
849                         return supported_features->encrypted_directories != 0;
850                 else
851                         return supported_features->encrypted_files != 0;
852         }
853         return true;
854 }
855
856 /* Given a WIM dentry to extract, build the path to which to extract it, in the
857  * format understood by the callbacks in the apply_operations being used.
858  *
859  * Write the resulting path into @path, which must have room for at least
860  * ctx->ops->max_path characters including the null-terminator.
861  *
862  * Return %true if successful; %false if this WIM dentry doesn't actually need
863  * to be extracted or if the calculated path exceeds ctx->ops->max_path
864  * characters.
865  *
866  * This function clobbers the tmp_list member of @dentry and its ancestors up
867  * until the extraction root.  */
868 static bool
869 build_extraction_path(tchar path[], struct wim_dentry *dentry,
870                       struct apply_ctx *ctx)
871 {
872         size_t path_nchars;
873         LIST_HEAD(ancestor_list);
874         tchar *p = path;
875         const tchar *target_prefix;
876         size_t target_prefix_nchars;
877         struct wim_dentry *d;
878
879         if (dentry->extraction_skipped)
880                 return false;
881
882         path_nchars = ctx->ops->path_prefix_nchars;
883
884         if (ctx->ops->requires_realtarget_in_paths) {
885                 target_prefix        = ctx->realtarget;
886                 target_prefix_nchars = ctx->realtarget_nchars;
887         } else if (ctx->ops->requires_target_in_paths) {
888                 target_prefix        = ctx->target;
889                 target_prefix_nchars = ctx->target_nchars;
890         } else {
891                 target_prefix        = NULL;
892                 target_prefix_nchars = 0;
893         }
894         path_nchars += target_prefix_nchars;
895
896         for (d = dentry; d != ctx->extract_root; d = d->parent) {
897                 path_nchars += d->extraction_name_nchars + 1;
898                 list_add(&d->tmp_list, &ancestor_list);
899         }
900
901         path_nchars++; /* null terminator */
902
903         if (path_nchars > ctx->ops->path_max) {
904                 WARNING("\"%"TS"\": Path too long to extract",
905                         dentry_full_path(dentry));
906                 return false;
907         }
908
909         p = tmempcpy(p, ctx->ops->path_prefix, ctx->ops->path_prefix_nchars);
910         p = tmempcpy(p, target_prefix, target_prefix_nchars);
911         list_for_each_entry(d, &ancestor_list, tmp_list) {
912                 *p++ = ctx->ops->path_separator;
913                 p = tmempcpy(p, d->extraction_name, d->extraction_name_nchars);
914         }
915         *p++ = T('\0');
916         wimlib_assert(p - path == path_nchars);
917         return true;
918 }
919
920 static unsigned
921 get_num_path_components(const tchar *path, tchar path_separator)
922 {
923         unsigned num_components = 0;
924 #ifdef __WIN32__
925         /* Ignore drive letter.  */
926         if (path[0] != L'\0' && path[1] == L':')
927                 path += 2;
928 #endif
929
930         while (*path) {
931                 while (*path == path_separator)
932                         path++;
933                 if (*path)
934                         num_components++;
935                 while (*path && *path != path_separator)
936                         path++;
937         }
938         return num_components;
939 }
940
941 static int
942 extract_multiimage_symlink(const tchar *oldpath, const tchar *newpath,
943                            struct apply_ctx *ctx, struct wim_dentry *dentry)
944 {
945         size_t num_raw_path_components;
946         const struct wim_dentry *d;
947         size_t num_target_path_components;
948         tchar *p;
949         const tchar *p_old;
950         int ret;
951
952         num_raw_path_components = 0;
953         for (d = dentry; d != ctx->extract_root; d = d->parent)
954                 num_raw_path_components++;
955
956         if (ctx->ops->requires_realtarget_in_paths)
957                 num_target_path_components = get_num_path_components(ctx->realtarget,
958                                                                      ctx->ops->path_separator);
959         else if (ctx->ops->requires_target_in_paths)
960                 num_target_path_components = get_num_path_components(ctx->target,
961                                                                      ctx->ops->path_separator);
962         else
963                 num_target_path_components = 0;
964
965         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_MULTI_IMAGE) {
966                 wimlib_assert(num_target_path_components > 0);
967                 num_raw_path_components++;
968                 num_target_path_components--;
969         }
970
971         p_old = oldpath + ctx->ops->path_prefix_nchars;
972 #ifdef __WIN32__
973         if (p_old[0] != L'\0' && p_old[1] == ':')
974                 p_old += 2;
975 #endif
976         while (*p_old == ctx->ops->path_separator)
977                 p_old++;
978         while (--num_target_path_components) {
979                 while (*p_old != ctx->ops->path_separator)
980                         p_old++;
981                 while (*p_old == ctx->ops->path_separator)
982                         p_old++;
983         }
984
985         tchar symlink_target[tstrlen(p_old) + 3 * num_raw_path_components + 1];
986
987         p = &symlink_target[0];
988         while (num_raw_path_components--) {
989                 *p++ = '.';
990                 *p++ = '.';
991                 *p++ = ctx->ops->path_separator;
992         }
993         tstrcpy(p, p_old);
994         DEBUG("Creating symlink \"%"TS"\" => \"%"TS"\"",
995               newpath, symlink_target);
996         ret = ctx->ops->create_symlink(symlink_target, newpath, ctx);
997         if (ret) {
998                 ERROR_WITH_ERRNO("Failed to create symlink "
999                                  "\"%"TS"\" => \"%"TS"\"",
1000                                  newpath, symlink_target);
1001         }
1002         return ret;
1003 }
1004
1005 /* Create the "skeleton" of an extracted file or directory.  Don't yet extract
1006  * data streams, reparse data (including symbolic links), timestamps, and
1007  * security descriptors.  Basically, everything that doesn't require reading
1008  * non-metadata resources from the WIM file and isn't delayed until the final
1009  * pass.  */
1010 static int
1011 do_dentry_extract_skeleton(tchar path[], struct wim_dentry *dentry,
1012                            struct apply_ctx *ctx)
1013 {
1014         struct wim_inode *inode = dentry->d_inode;
1015         int ret;
1016         const tchar *oldpath;
1017
1018         if (unlikely(is_linked_extraction(ctx))) {
1019                 struct wim_lookup_table_entry *unnamed_lte;
1020
1021                 unnamed_lte = inode_unnamed_lte_resolved(dentry->d_inode);
1022                 if (unnamed_lte && unnamed_lte->extracted_file) {
1023                         oldpath = unnamed_lte->extracted_file;
1024                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_HARDLINK)
1025                                 goto hardlink;
1026                         else
1027                                 goto symlink;
1028                 }
1029         }
1030
1031         /* Create hard link if this dentry corresponds to an already-extracted
1032          * inode.  */
1033         if (inode->i_extracted_file) {
1034                 oldpath = inode->i_extracted_file;
1035                 goto hardlink;
1036         }
1037
1038         /* Skip symlinks unless they can be extracted as reparse points rather
1039          * than created directly.  */
1040         if (inode_is_symlink(inode) && !ctx->supported_features.reparse_points)
1041                 return 0;
1042
1043         /* Create this file or directory unless it's the extraction root, which
1044          * was already created if necessary.  */
1045         if (dentry != ctx->extract_root) {
1046                 ret = extract_inode(path, ctx, inode);
1047                 if (ret)
1048                         return ret;
1049         }
1050
1051         /* Create empty named data streams.  */
1052         if (can_extract_named_data_streams(ctx)) {
1053                 for (u16 i = 0; i < inode->i_num_ads; i++) {
1054                         file_spec_t file_spec;
1055                         struct wim_ads_entry *entry = &inode->i_ads_entries[i];
1056
1057                         if (!ads_entry_is_named_stream(entry))
1058                                 continue;
1059                         if (entry->lte)
1060                                 continue;
1061                         if (ctx->ops->uses_cookies)
1062                                 file_spec.cookie = inode->extract_cookie;
1063                         else
1064                                 file_spec.path = path;
1065                         ret = ctx->ops->extract_named_stream(file_spec,
1066                                                              entry->stream_name,
1067                                                              entry->stream_name_nbytes / 2,
1068                                                              entry->lte, ctx);
1069                         if (ret) {
1070                                 ERROR_WITH_ERRNO("\"%"TS"\": failed to create "
1071                                                  "empty named data stream",
1072                                                  path);
1073                                 return ret;
1074                         }
1075                 }
1076         }
1077
1078         /* Set file attributes (if supported).  */
1079         ret = extract_file_attributes(path, ctx, dentry, 0);
1080         if (ret)
1081                 return ret;
1082
1083         /* Set or remove file short name (if supported).  */
1084         ret = extract_short_name(path, ctx, dentry);
1085         if (ret)
1086                 return ret;
1087
1088         /* If inode has multiple links and hard links are supported in this
1089          * extraction mode and volume, save the path to the extracted file in
1090          * case it's needed to create a hard link.  */
1091         if (unlikely(is_linked_extraction(ctx))) {
1092                 struct wim_lookup_table_entry *unnamed_lte;
1093
1094                 unnamed_lte = inode_unnamed_lte_resolved(dentry->d_inode);
1095                 if (unnamed_lte) {
1096                         unnamed_lte->extracted_file = TSTRDUP(path);
1097                         if (!unnamed_lte->extracted_file)
1098                                 return WIMLIB_ERR_NOMEM;
1099                 }
1100         } else if (inode->i_nlink > 1 && ctx->supported_features.hard_links) {
1101                 inode->i_extracted_file = TSTRDUP(path);
1102                 if (!inode->i_extracted_file)
1103                         return WIMLIB_ERR_NOMEM;
1104         }
1105         return 0;
1106
1107 symlink:
1108         ret = extract_multiimage_symlink(oldpath, path, ctx, dentry);
1109         if (ret)
1110                 return ret;
1111         dentry->was_hardlinked = 1;
1112         return 0;
1113
1114 hardlink:
1115         ret = extract_hardlink(oldpath, path, ctx);
1116         if (ret)
1117                 return ret;
1118         dentry->was_hardlinked = 1;
1119         return 0;
1120 }
1121
1122 static int
1123 dentry_extract_skeleton(struct wim_dentry *dentry, void *_ctx)
1124 {
1125         struct apply_ctx *ctx = _ctx;
1126         tchar path[ctx->ops->path_max];
1127         struct wim_dentry *orig_dentry;
1128         struct wim_dentry *other_dentry;
1129         int ret;
1130
1131         /* Here we may re-order the extraction of multiple names (hard links)
1132          * for the same file in the same directory in order to ensure the short
1133          * (DOS) name is set correctly.  A short name is always associated with
1134          * exactly one long name, and at least on NTFS, only one long name for a
1135          * file can have a short name associated with it.  (More specifically,
1136          * there can be unlimited names in the POSIX namespace, but only one
1137          * name can be in the Win32+DOS namespace, or one name in the Win32
1138          * namespace with a corresponding name in the DOS namespace.) To ensure
1139          * the short name of a file is associated with the correct long name in
1140          * a directory, we extract the long name with a corresponding short name
1141          * before any additional names.  This can affect NTFS-3g extraction
1142          * (which uses ntfs_set_ntfs_dos_name(), which doesn't allow specifying
1143          * the long name to associate with a short name) and may affect Win32
1144          * extraction as well (which uses SetFileShortName()).  */
1145
1146         if (dentry->skeleton_extracted)
1147                 return 0;
1148         orig_dentry = NULL;
1149         if (ctx->supported_features.short_names
1150             && !dentry_has_short_name(dentry)
1151             && !dentry->d_inode->i_dos_name_extracted)
1152         {
1153                 inode_for_each_dentry(other_dentry, dentry->d_inode) {
1154                         if (dentry_has_short_name(other_dentry)
1155                             && !other_dentry->skeleton_extracted
1156                             && other_dentry->parent == dentry->parent)
1157                         {
1158                                 DEBUG("Creating %"TS" before %"TS" "
1159                                       "to guarantee correct DOS name extraction",
1160                                       dentry_full_path(other_dentry),
1161                                       dentry_full_path(dentry));
1162                                 orig_dentry = dentry;
1163                                 dentry = other_dentry;
1164                                 break;
1165                         }
1166                 }
1167         }
1168 again:
1169         if (!build_extraction_path(path, dentry, ctx))
1170                 return 0;
1171         ret = do_dentry_extract_skeleton(path, dentry, ctx);
1172         if (ret)
1173                 return ret;
1174
1175         dentry->skeleton_extracted = 1;
1176
1177         if (orig_dentry) {
1178                 dentry = orig_dentry;
1179                 orig_dentry = NULL;
1180                 goto again;
1181         }
1182         dentry->d_inode->i_dos_name_extracted = 1;
1183         return 0;
1184 }
1185
1186 /* Create a file or directory, then immediately extract all streams.  This
1187  * assumes that WIMLIB_EXTRACT_FLAG_SEQUENTIAL is not specified, since the WIM
1188  * may not be read sequentially by this function.  */
1189 static int
1190 dentry_extract(struct wim_dentry *dentry, void *_ctx)
1191 {
1192         struct apply_ctx *ctx = _ctx;
1193         tchar path[ctx->ops->path_max];
1194         int ret;
1195
1196         ret = dentry_extract_skeleton(dentry, ctx);
1197         if (ret)
1198                 return ret;
1199
1200         if (!build_extraction_path(path, dentry, ctx))
1201                 return 0;
1202
1203         return extract_streams(path, ctx, dentry, NULL, NULL);
1204 }
1205
1206 /* Extract all instances of the stream @lte that are being extracted in this
1207  * call of extract_tree().  @can_seek specifies whether the WIM file descriptor
1208  * is seekable or not (e.g. is a pipe).  If not and the stream needs to be
1209  * extracted multiple times, it is extracted to a temporary file first.
1210  *
1211  * This is intended for use with sequential extraction of a WIM image
1212  * (WIMLIB_EXTRACT_FLAG_SEQUENTIAL specified).  */
1213 static int
1214 extract_stream_instances(struct wim_lookup_table_entry *lte,
1215                          struct apply_ctx *ctx, bool can_seek)
1216 {
1217         struct wim_dentry **lte_dentries;
1218         struct wim_lookup_table_entry *lte_tmp = NULL;
1219         struct wim_lookup_table_entry *lte_override;
1220         tchar *stream_tmp_filename = NULL;
1221         tchar path[ctx->ops->path_max];
1222         unsigned i;
1223         int ret;
1224
1225         if (lte->out_refcnt <= ARRAY_LEN(lte->inline_lte_dentries))
1226                 lte_dentries = lte->inline_lte_dentries;
1227         else
1228                 lte_dentries = lte->lte_dentries;
1229
1230         if (likely(can_seek || lte->out_refcnt < 2)) {
1231                 lte_override = lte;
1232         } else {
1233                 /* Need to extract stream to temporary file.  */
1234                 struct filedes fd;
1235                 int raw_fd;
1236
1237                 stream_tmp_filename = ttempnam(NULL, T("wimlib"));
1238                 if (!stream_tmp_filename) {
1239                         ERROR_WITH_ERRNO("Failed to create temporary filename");
1240                         ret = WIMLIB_ERR_OPEN;
1241                         goto out;
1242                 }
1243
1244                 lte_tmp = memdup(lte, sizeof(struct wim_lookup_table_entry));
1245                 if (!lte_tmp) {
1246                         ret = WIMLIB_ERR_NOMEM;
1247                         goto out_free_stream_tmp_filename;
1248                 }
1249                 lte_tmp->resource_location = RESOURCE_IN_FILE_ON_DISK;
1250                 lte_tmp->file_on_disk = stream_tmp_filename;
1251                 lte_override = lte_tmp;
1252
1253                 raw_fd = topen(stream_tmp_filename,
1254                                O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0600);
1255                 if (raw_fd < 0) {
1256                         ERROR_WITH_ERRNO("Failed to open temporary file");
1257                         ret = WIMLIB_ERR_OPEN;
1258                         goto out_free_lte_tmp;
1259                 }
1260                 filedes_init(&fd, raw_fd);
1261                 ret = extract_wim_resource_to_fd(lte, &fd,
1262                                                  wim_resource_size(lte));
1263                 if (filedes_close(&fd) && !ret)
1264                         ret = WIMLIB_ERR_WRITE;
1265                 if (ret)
1266                         goto out_unlink_stream_tmp_file;
1267         }
1268
1269         /* Extract all instances of the stream, reading either from the stream
1270          * in the WIM file or from the temporary file containing the stream.
1271          * dentry->tmp_flag is used to ensure that each dentry is processed only
1272          * once regardless of how many times this stream appears in the streams
1273          * of the corresponding inode.  */
1274         for (i = 0; i < lte->out_refcnt; i++) {
1275                 struct wim_dentry *dentry = lte_dentries[i];
1276
1277                 if (dentry->tmp_flag)
1278                         continue;
1279                 if (!build_extraction_path(path, dentry, ctx))
1280                         continue;
1281                 ret = extract_streams(path, ctx, dentry,
1282                                       lte, lte_override);
1283                 if (ret)
1284                         goto out_clear_tmp_flags;
1285                 dentry->tmp_flag = 1;
1286         }
1287         ret = 0;
1288 out_clear_tmp_flags:
1289         for (i = 0; i < lte->out_refcnt; i++)
1290                 lte_dentries[i]->tmp_flag = 0;
1291 out_unlink_stream_tmp_file:
1292         if (stream_tmp_filename)
1293                 tunlink(stream_tmp_filename);
1294 out_free_lte_tmp:
1295         FREE(lte_tmp);
1296 out_free_stream_tmp_filename:
1297         FREE(stream_tmp_filename);
1298 out:
1299         return ret;
1300 }
1301
1302 /* Extracts a list of streams (ctx.stream_list), assuming that the directory
1303  * structure and empty files were already created.  This relies on the
1304  * per-`struct wim_lookup_table_entry' list of dentries that reference each
1305  * stream that was constructed earlier.  Streams are extracted exactly in the
1306  * order of the stream list; however, unless the WIM's file descriptor is
1307  * detected to be non-seekable, streams may be read from the WIM file more than
1308  * one time if multiple copies need to be extracted.  */
1309 static int
1310 extract_stream_list(struct apply_ctx *ctx)
1311 {
1312         struct wim_lookup_table_entry *lte;
1313         bool can_seek;
1314         int ret;
1315
1316         can_seek = (lseek(ctx->wim->in_fd.fd, 0, SEEK_CUR) != -1);
1317         list_for_each_entry(lte, &ctx->stream_list, extraction_list) {
1318                 ret = extract_stream_instances(lte, ctx, can_seek);
1319                 if (ret)
1320                         return ret;
1321         }
1322         return 0;
1323 }
1324
1325 #define PWM_ALLOW_WIM_HDR 0x00001
1326 #define PWM_SILENT_EOF    0x00002
1327
1328 /* Read the header from a stream in a pipable WIM.  */
1329 static int
1330 read_pwm_stream_header(WIMStruct *pwm, struct wim_lookup_table_entry *lte,
1331                        int flags, struct wim_header_disk *hdr_ret)
1332 {
1333         union {
1334                 struct pwm_stream_hdr stream_hdr;
1335                 struct wim_header_disk pwm_hdr;
1336         } buf;
1337         int ret;
1338
1339         ret = full_read(&pwm->in_fd, &buf.stream_hdr, sizeof(buf.stream_hdr));
1340         if (ret)
1341                 goto read_error;
1342
1343         if ((flags & PWM_ALLOW_WIM_HDR) && buf.stream_hdr.magic == PWM_MAGIC) {
1344                 BUILD_BUG_ON(sizeof(buf.pwm_hdr) < sizeof(buf.stream_hdr));
1345                 ret = full_read(&pwm->in_fd, &buf.stream_hdr + 1,
1346                                 sizeof(buf.pwm_hdr) - sizeof(buf.stream_hdr));
1347
1348                 if (ret)
1349                         goto read_error;
1350                 lte->resource_location = RESOURCE_NONEXISTENT;
1351                 memcpy(hdr_ret, &buf.pwm_hdr, sizeof(buf.pwm_hdr));
1352                 return 0;
1353         }
1354
1355         if (buf.stream_hdr.magic != PWM_STREAM_MAGIC) {
1356                 ERROR("Data read on pipe is invalid (expected stream header).");
1357                 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1358         }
1359
1360         lte->resource_entry.original_size = le64_to_cpu(buf.stream_hdr.uncompressed_size);
1361         copy_hash(lte->hash, buf.stream_hdr.hash);
1362         lte->resource_entry.flags = le32_to_cpu(buf.stream_hdr.flags);
1363         lte->resource_entry.offset = pwm->in_fd.offset;
1364         lte->resource_location = RESOURCE_IN_WIM;
1365         lte->wim = pwm;
1366         if (lte->resource_entry.flags & WIM_RESHDR_FLAG_COMPRESSED) {
1367                 lte->compression_type = pwm->compression_type;
1368                 lte->resource_entry.size = 0;
1369         } else {
1370                 lte->compression_type = WIMLIB_COMPRESSION_TYPE_NONE;
1371                 lte->resource_entry.size = lte->resource_entry.original_size;
1372         }
1373         lte->is_pipable = 1;
1374         return 0;
1375
1376 read_error:
1377         if (ret != WIMLIB_ERR_UNEXPECTED_END_OF_FILE || !(flags & PWM_SILENT_EOF))
1378                 ERROR_WITH_ERRNO("Error reading pipable WIM from pipe");
1379         return ret;
1380 }
1381
1382 /* Skip over an unneeded stream in a pipable WIM being read from a pipe.  */
1383 static int
1384 skip_pwm_stream(struct wim_lookup_table_entry *lte)
1385 {
1386         return read_partial_wim_resource(lte, wim_resource_size(lte),
1387                                          NULL, NULL,
1388                                          WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY,
1389                                          0);
1390 }
1391
1392 static int
1393 extract_streams_from_pipe(struct apply_ctx *ctx)
1394 {
1395         struct wim_lookup_table_entry *found_lte;
1396         struct wim_lookup_table_entry *needed_lte;
1397         struct wim_lookup_table *lookup_table;
1398         struct wim_header_disk pwm_hdr;
1399         int ret;
1400         int pwm_flags;
1401
1402         ret = WIMLIB_ERR_NOMEM;
1403         found_lte = new_lookup_table_entry();
1404         if (!found_lte)
1405                 goto out;
1406
1407         lookup_table = ctx->wim->lookup_table;
1408         pwm_flags = PWM_ALLOW_WIM_HDR;
1409         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1410                 pwm_flags |= PWM_SILENT_EOF;
1411         memcpy(ctx->progress.extract.guid, ctx->wim->hdr.guid, WIM_GID_LEN);
1412         ctx->progress.extract.part_number = ctx->wim->hdr.part_number;
1413         ctx->progress.extract.total_parts = ctx->wim->hdr.total_parts;
1414         if (ctx->progress_func)
1415                 ctx->progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1416                                    &ctx->progress);
1417         while (ctx->num_streams_remaining) {
1418                 ret = read_pwm_stream_header(ctx->wim, found_lte, pwm_flags,
1419                                              &pwm_hdr);
1420                 if (ret) {
1421                         if (ret == WIMLIB_ERR_UNEXPECTED_END_OF_FILE &&
1422                             (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1423                         {
1424                                 goto resume_done;
1425                         }
1426                         goto out_free_found_lte;
1427                 }
1428
1429                 if ((found_lte->resource_location != RESOURCE_NONEXISTENT)
1430                     && !(found_lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)
1431                     && (needed_lte = __lookup_resource(lookup_table, found_lte->hash))
1432                     && (needed_lte->out_refcnt))
1433                 {
1434                         copy_resource_entry(&needed_lte->resource_entry,
1435                                             &found_lte->resource_entry);
1436                         needed_lte->resource_location = found_lte->resource_location;
1437                         needed_lte->wim               = found_lte->wim;
1438                         needed_lte->compression_type  = found_lte->compression_type;
1439                         needed_lte->is_pipable        = found_lte->is_pipable;
1440
1441                         ret = extract_stream_instances(needed_lte, ctx, false);
1442                         if (ret)
1443                                 goto out_free_found_lte;
1444                         ctx->num_streams_remaining--;
1445                 } else if (found_lte->resource_location != RESOURCE_NONEXISTENT) {
1446                         ret = skip_pwm_stream(found_lte);
1447                         if (ret)
1448                                 goto out_free_found_lte;
1449                 } else {
1450                         u16 part_number = le16_to_cpu(pwm_hdr.part_number);
1451                         u16 total_parts = le16_to_cpu(pwm_hdr.total_parts);
1452
1453                         if (part_number != ctx->progress.extract.part_number ||
1454                             total_parts != ctx->progress.extract.total_parts ||
1455                             memcmp(pwm_hdr.guid, ctx->progress.extract.guid,
1456                                    WIM_GID_LEN))
1457                         {
1458                                 ctx->progress.extract.part_number = part_number;
1459                                 ctx->progress.extract.total_parts = total_parts;
1460                                 memcpy(ctx->progress.extract.guid,
1461                                        pwm_hdr.guid, WIM_GID_LEN);
1462                                 if (ctx->progress_func) {
1463                                         ctx->progress_func(
1464                                                 WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1465                                                            &ctx->progress);
1466                                 }
1467
1468                         }
1469                 }
1470         }
1471         ret = 0;
1472 out_free_found_lte:
1473         free_lookup_table_entry(found_lte);
1474 out:
1475         return ret;
1476
1477 resume_done:
1478         /* TODO */
1479         return 0;
1480 }
1481
1482 /* Finish extracting a file, directory, or symbolic link by setting file
1483  * security and timestamps.  */
1484 static int
1485 dentry_extract_final(struct wim_dentry *dentry, void *_ctx)
1486 {
1487         struct apply_ctx *ctx = _ctx;
1488         int ret;
1489         tchar path[ctx->ops->path_max];
1490
1491         if (!build_extraction_path(path, dentry, ctx))
1492                 return 0;
1493
1494         ret = extract_security(path, ctx, dentry);
1495         if (ret)
1496                 return ret;
1497
1498         if (ctx->ops->requires_final_set_attributes_pass) {
1499                 /* Set file attributes (if supported).  */
1500                 ret = extract_file_attributes(path, ctx, dentry, 1);
1501                 if (ret)
1502                         return ret;
1503         }
1504
1505         return extract_timestamps(path, ctx, dentry);
1506 }
1507
1508 /*
1509  * Extract a WIM dentry to standard output.
1510  *
1511  * This obviously doesn't make sense in all cases.  We return an error if the
1512  * dentry does not correspond to a regular file.  Otherwise we extract the
1513  * unnamed data stream only.
1514  */
1515 static int
1516 extract_dentry_to_stdout(struct wim_dentry *dentry)
1517 {
1518         int ret = 0;
1519         if (dentry->d_inode->i_attributes & (FILE_ATTRIBUTE_REPARSE_POINT |
1520                                              FILE_ATTRIBUTE_DIRECTORY))
1521         {
1522                 ERROR("\"%"TS"\" is not a regular file and therefore cannot be "
1523                       "extracted to standard output", dentry_full_path(dentry));
1524                 ret = WIMLIB_ERR_NOT_A_REGULAR_FILE;
1525         } else {
1526                 struct wim_lookup_table_entry *lte;
1527
1528                 lte = inode_unnamed_lte_resolved(dentry->d_inode);
1529                 if (lte) {
1530                         struct filedes _stdout;
1531                         filedes_init(&_stdout, STDOUT_FILENO);
1532                         ret = extract_wim_resource_to_fd(lte, &_stdout,
1533                                                          wim_resource_size(lte));
1534                 }
1535         }
1536         return ret;
1537 }
1538
1539 #ifdef __WIN32__
1540 static const utf16lechar replacement_char = cpu_to_le16(0xfffd);
1541 #else
1542 static const utf16lechar replacement_char = cpu_to_le16('?');
1543 #endif
1544
1545 static bool
1546 file_name_valid(utf16lechar *name, size_t num_chars, bool fix)
1547 {
1548         size_t i;
1549
1550         if (num_chars == 0)
1551                 return true;
1552         for (i = 0; i < num_chars; i++) {
1553                 switch (name[i]) {
1554         #ifdef __WIN32__
1555                 case cpu_to_le16('\\'):
1556                 case cpu_to_le16(':'):
1557                 case cpu_to_le16('*'):
1558                 case cpu_to_le16('?'):
1559                 case cpu_to_le16('"'):
1560                 case cpu_to_le16('<'):
1561                 case cpu_to_le16('>'):
1562                 case cpu_to_le16('|'):
1563         #endif
1564                 case cpu_to_le16('/'):
1565                 case cpu_to_le16('\0'):
1566                         if (fix)
1567                                 name[i] = replacement_char;
1568                         else
1569                                 return false;
1570                 }
1571         }
1572
1573 #ifdef __WIN32__
1574         if (name[num_chars - 1] == cpu_to_le16(' ') ||
1575             name[num_chars - 1] == cpu_to_le16('.'))
1576         {
1577                 if (fix)
1578                         name[num_chars - 1] = replacement_char;
1579                 else
1580                         return false;
1581         }
1582 #endif
1583         return true;
1584 }
1585
1586 static bool
1587 dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
1588 {
1589         const utf16lechar *file_name = dentry->file_name;
1590         return file_name != NULL &&
1591                 file_name[0] == cpu_to_le16('.') &&
1592                 (file_name[1] == cpu_to_le16('\0') ||
1593                  (file_name[1] == cpu_to_le16('.') &&
1594                   file_name[2] == cpu_to_le16('\0')));
1595 }
1596
1597 static int
1598 dentry_mark_skipped(struct wim_dentry *dentry, void *_ignore)
1599 {
1600         dentry->extraction_skipped = 1;
1601         return 0;
1602 }
1603
1604 /*
1605  * dentry_calculate_extraction_path-
1606  *
1607  * Calculate the actual filename component at which a WIM dentry will be
1608  * extracted, handling invalid filenames "properly".
1609  *
1610  * dentry->extraction_name usually will be set the same as dentry->file_name (on
1611  * UNIX, converted into the platform's multibyte encoding).  However, if the
1612  * file name contains characters that are not valid on the current platform or
1613  * has some other format that is not valid, leave dentry->extraction_name as
1614  * NULL and set dentry->extraction_skipped to indicate that this dentry should
1615  * not be extracted, unless the appropriate flag
1616  * WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES is set in the extract flags, in
1617  * which case a substitute filename will be created and set instead.
1618  *
1619  * Conflicts with case-insensitive names on Windows are handled similarly; see
1620  * below.
1621  */
1622 static int
1623 dentry_calculate_extraction_path(struct wim_dentry *dentry, void *_args)
1624 {
1625         struct apply_ctx *ctx = _args;
1626         int ret;
1627
1628         if (dentry == ctx->extract_root || dentry->extraction_skipped)
1629                 return 0;
1630
1631         if (!dentry_is_supported(dentry, &ctx->supported_features))
1632                 goto skip_dentry;
1633
1634         if (dentry_is_dot_or_dotdot(dentry)) {
1635                 /* WIM files shouldn't contain . or .. entries.  But if they are
1636                  * there, don't attempt to extract them. */
1637                 WARNING("Skipping extraction of unexpected . or .. file "
1638                         "\"%"TS"\"", dentry_full_path(dentry));
1639                 goto skip_dentry;
1640         }
1641
1642 #ifdef __WIN32__
1643         if (!ctx->ops->supports_case_sensitive_filenames)
1644         {
1645                 struct wim_dentry *other;
1646                 list_for_each_entry(other, &dentry->case_insensitive_conflict_list,
1647                                     case_insensitive_conflict_list)
1648                 {
1649                         if (ctx->extract_flags &
1650                             WIMLIB_EXTRACT_FLAG_ALL_CASE_CONFLICTS) {
1651                                 WARNING("\"%"TS"\" has the same "
1652                                         "case-insensitive name as "
1653                                         "\"%"TS"\"; extracting "
1654                                         "dummy name instead",
1655                                         dentry_full_path(dentry),
1656                                         dentry_full_path(other));
1657                                 goto out_replace;
1658                         } else {
1659                                 WARNING("Not extracting \"%"TS"\": "
1660                                         "has same case-insensitive "
1661                                         "name as \"%"TS"\"",
1662                                         dentry_full_path(dentry),
1663                                         dentry_full_path(other));
1664                                 goto skip_dentry;
1665                         }
1666                 }
1667         }
1668 #else   /* __WIN32__ */
1669         wimlib_assert(ctx->ops->supports_case_sensitive_filenames);
1670 #endif  /* !__WIN32__ */
1671
1672         if (file_name_valid(dentry->file_name, dentry->file_name_nbytes / 2, false)) {
1673 #ifdef __WIN32__
1674                 dentry->extraction_name = dentry->file_name;
1675                 dentry->extraction_name_nchars = dentry->file_name_nbytes / 2;
1676                 return 0;
1677 #else
1678                 return utf16le_to_tstr(dentry->file_name,
1679                                        dentry->file_name_nbytes,
1680                                        &dentry->extraction_name,
1681                                        &dentry->extraction_name_nchars);
1682 #endif
1683         } else {
1684                 if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES)
1685                 {
1686                         WARNING("\"%"TS"\" has an invalid filename "
1687                                 "that is not supported on this platform; "
1688                                 "extracting dummy name instead",
1689                                 dentry_full_path(dentry));
1690                         goto out_replace;
1691                 } else {
1692                         WARNING("Not extracting \"%"TS"\": has an invalid filename "
1693                                 "that is not supported on this platform",
1694                                 dentry_full_path(dentry));
1695                         goto skip_dentry;
1696                 }
1697         }
1698
1699 out_replace:
1700         {
1701                 utf16lechar utf16_name_copy[dentry->file_name_nbytes / 2];
1702
1703                 memcpy(utf16_name_copy, dentry->file_name, dentry->file_name_nbytes);
1704                 file_name_valid(utf16_name_copy, dentry->file_name_nbytes / 2, true);
1705
1706                 tchar *tchar_name;
1707                 size_t tchar_nchars;
1708         #ifdef __WIN32__
1709                 tchar_name = utf16_name_copy;
1710                 tchar_nchars = dentry->file_name_nbytes / 2;
1711         #else
1712                 ret = utf16le_to_tstr(utf16_name_copy,
1713                                       dentry->file_name_nbytes,
1714                                       &tchar_name, &tchar_nchars);
1715                 if (ret)
1716                         return ret;
1717         #endif
1718                 size_t fixed_name_num_chars = tchar_nchars;
1719                 tchar fixed_name[tchar_nchars + 50];
1720
1721                 tmemcpy(fixed_name, tchar_name, tchar_nchars);
1722                 fixed_name_num_chars += tsprintf(fixed_name + tchar_nchars,
1723                                                  T(" (invalid filename #%lu)"),
1724                                                  ++ctx->invalid_sequence);
1725         #ifndef __WIN32__
1726                 FREE(tchar_name);
1727         #endif
1728                 dentry->extraction_name = memdup(fixed_name,
1729                                                  2 * fixed_name_num_chars + 2);
1730                 if (!dentry->extraction_name)
1731                         return WIMLIB_ERR_NOMEM;
1732                 dentry->extraction_name_nchars = fixed_name_num_chars;
1733         }
1734         return 0;
1735
1736 skip_dentry:
1737         for_dentry_in_tree(dentry, dentry_mark_skipped, NULL);
1738         return 0;
1739 }
1740
1741 /* Clean up dentry and inode structure after extraction.  */
1742 static int
1743 dentry_reset_needs_extraction(struct wim_dentry *dentry, void *_ignore)
1744 {
1745         struct wim_inode *inode = dentry->d_inode;
1746
1747         dentry->extraction_skipped = 0;
1748         dentry->was_hardlinked = 0;
1749         dentry->skeleton_extracted = 0;
1750         inode->i_visited = 0;
1751         FREE(inode->i_extracted_file);
1752         inode->i_extracted_file = NULL;
1753         inode->i_dos_name_extracted = 0;
1754         if ((void*)dentry->extraction_name != (void*)dentry->file_name)
1755                 FREE(dentry->extraction_name);
1756         dentry->extraction_name = NULL;
1757         return 0;
1758 }
1759
1760 /* Tally features necessary to extract a dentry and the corresponding inode.  */
1761 static int
1762 dentry_tally_features(struct wim_dentry *dentry, void *_features)
1763 {
1764         struct wim_features *features = _features;
1765         struct wim_inode *inode = dentry->d_inode;
1766
1767         if (inode->i_attributes & FILE_ATTRIBUTE_ARCHIVE)
1768                 features->archive_files++;
1769         if (inode->i_attributes & FILE_ATTRIBUTE_HIDDEN)
1770                 features->hidden_files++;
1771         if (inode->i_attributes & FILE_ATTRIBUTE_SYSTEM)
1772                 features->system_files++;
1773         if (inode->i_attributes & FILE_ATTRIBUTE_COMPRESSED)
1774                 features->compressed_files++;
1775         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
1776                 if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
1777                         features->encrypted_directories++;
1778                 else
1779                         features->encrypted_files++;
1780         }
1781         if (inode->i_attributes & FILE_ATTRIBUTE_NOT_CONTENT_INDEXED)
1782                 features->not_context_indexed_files++;
1783         if (inode->i_attributes & FILE_ATTRIBUTE_SPARSE_FILE)
1784                 features->sparse_files++;
1785         if (inode_has_named_stream(inode))
1786                 features->named_data_streams++;
1787         if (inode->i_visited)
1788                 features->hard_links++;
1789         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1790                 features->reparse_points++;
1791                 if (inode_is_symlink(inode))
1792                         features->symlink_reparse_points++;
1793                 else
1794                         features->other_reparse_points++;
1795         }
1796         if (inode->i_security_id != -1)
1797                 features->security_descriptors++;
1798         if (dentry->short_name_nbytes)
1799                 features->short_names++;
1800         if (inode_has_unix_data(inode))
1801                 features->unix_data++;
1802         inode->i_visited = 1;
1803         return 0;
1804 }
1805
1806 static int
1807 dentry_clear_inode_visited(struct wim_dentry *dentry, void *_ignore)
1808 {
1809         dentry->d_inode->i_visited = 0;
1810         return 0;
1811 }
1812
1813 /* Tally the features necessary to extract a dentry tree.  */
1814 static void
1815 dentry_tree_get_features(struct wim_dentry *root, struct wim_features *features)
1816 {
1817         memset(features, 0, sizeof(struct wim_features));
1818         for_dentry_in_tree(root, dentry_tally_features, features);
1819         for_dentry_in_tree(root, dentry_clear_inode_visited, NULL);
1820 }
1821
1822 static u32
1823 compute_supported_attributes_mask(const struct wim_features *supported_features)
1824 {
1825         u32 mask = ~(u32)0;
1826
1827         if (!supported_features->archive_files)
1828                 mask &= ~FILE_ATTRIBUTE_ARCHIVE;
1829
1830         if (!supported_features->hidden_files)
1831                 mask &= ~FILE_ATTRIBUTE_HIDDEN;
1832
1833         if (!supported_features->system_files)
1834                 mask &= ~FILE_ATTRIBUTE_SYSTEM;
1835
1836         if (!supported_features->not_context_indexed_files)
1837                 mask &= ~FILE_ATTRIBUTE_NOT_CONTENT_INDEXED;
1838
1839         if (!supported_features->compressed_files)
1840                 mask &= ~FILE_ATTRIBUTE_COMPRESSED;
1841
1842         if (!supported_features->sparse_files)
1843                 mask &= ~FILE_ATTRIBUTE_SPARSE_FILE;
1844
1845         if (!supported_features->reparse_points)
1846                 mask &= ~FILE_ATTRIBUTE_REPARSE_POINT;
1847
1848         return mask;
1849 }
1850
1851 static int
1852 do_feature_check(const struct wim_features *required_features,
1853                  const struct wim_features *supported_features,
1854                  int extract_flags,
1855                  const struct apply_operations *ops,
1856                  const tchar *wim_source_path)
1857 {
1858         const tchar *loc;
1859         const tchar *mode = T("this extraction mode");
1860
1861         if (wim_source_path[0] == '\0')
1862                 loc = T("the WIM image");
1863         else
1864                 loc = wim_source_path;
1865
1866         /* We're an archive program, so theoretically we can do what we want
1867          * with FILE_ATTRIBUTE_ARCHIVE (which is a dumb flag anyway).  Don't
1868          * bother the user about it.  */
1869 #if 0
1870         if (required_features->archive_files && !supported_features->archive_files)
1871         {
1872                 WARNING(
1873           "%lu files in %"TS" are marked as archived, but this attribute\n"
1874 "          is not supported in %"TS".",
1875                         required_features->archive_files, loc, mode);
1876         }
1877 #endif
1878
1879         if (required_features->hidden_files && !supported_features->hidden_files)
1880         {
1881                 WARNING(
1882           "%lu files in %"TS" are marked as hidden, but this\n"
1883 "          attribute is not supported in %"TS".",
1884                         required_features->hidden_files, loc, mode);
1885         }
1886
1887         if (required_features->system_files && !supported_features->system_files)
1888         {
1889                 WARNING(
1890           "%lu files in %"TS" are marked as system files,\n"
1891 "          but this attribute is not supported in %"TS".",
1892                         required_features->system_files, loc, mode);
1893         }
1894
1895         if (required_features->compressed_files && !supported_features->compressed_files)
1896         {
1897                 WARNING(
1898           "%lu files in %"TS" are marked as being transparently\n"
1899 "          compressed, but transparent compression is not supported in\n"
1900 "          %"TS".  These files will be extracted as uncompressed.",
1901                         required_features->compressed_files, loc, mode);
1902         }
1903
1904         if (required_features->encrypted_files && !supported_features->encrypted_files)
1905         {
1906                 WARNING(
1907           "%lu files in %"TS" are marked as being encrypted,\n"
1908 "           but encryption is not supported in %"TS".  These files\n"
1909 "           will not be extracted.",
1910                         required_features->encrypted_files, loc, mode);
1911         }
1912
1913         if (required_features->encrypted_directories &&
1914             !supported_features->encrypted_directories)
1915         {
1916                 WARNING(
1917           "%lu directories in %"TS" are marked as being encrypted,\n"
1918 "           but encryption is not supported in %"TS".\n"
1919 "           These directories will be extracted as unencrypted.",
1920                         required_features->encrypted_directories, loc, mode);
1921         }
1922
1923         if (required_features->not_context_indexed_files &&
1924             !supported_features->not_context_indexed_files)
1925         {
1926                 WARNING(
1927           "%lu files in %"TS" are marked as not content indexed,\n"
1928 "          but this attribute is not supported in %"TS".",
1929                         required_features->not_context_indexed_files, loc, mode);
1930         }
1931
1932         if (required_features->sparse_files && !supported_features->sparse_files)
1933         {
1934                 WARNING(
1935           "%lu files in %"TS" are marked as sparse, but creating\n"
1936 "           sparse files is not supported in %"TS".  These files\n"
1937 "           will be extracted as non-sparse.",
1938                         required_features->sparse_files, loc, mode);
1939         }
1940
1941         if (required_features->named_data_streams &&
1942             !supported_features->named_data_streams)
1943         {
1944                 WARNING(
1945           "%lu files in %"TS" contain one or more alternate (named)\n"
1946 "          data streams, which are not supported in %"TS".\n"
1947 "          Alternate data streams will NOT be extracted.",
1948                         required_features->named_data_streams, loc, mode);
1949         }
1950
1951         if (unlikely(extract_flags & (WIMLIB_EXTRACT_FLAG_HARDLINK |
1952                                       WIMLIB_EXTRACT_FLAG_SYMLINK)) &&
1953             required_features->named_data_streams &&
1954             supported_features->named_data_streams)
1955         {
1956                 WARNING(
1957           "%lu files in %"TS" contain one or more alternate (named)\n"
1958 "          data streams, which are not supported in linked extraction mode.\n"
1959 "          Alternate data streams will NOT be extracted.",
1960                         required_features->named_data_streams, loc);
1961         }
1962
1963         if (required_features->hard_links && !supported_features->hard_links)
1964         {
1965                 WARNING(
1966           "%lu files in %"TS" are hard links, but hard links are\n"
1967 "          not supported in %"TS".  Hard links will be extracted as\n"
1968 "          duplicate copies of the linked files.",
1969                         required_features->hard_links, loc, mode);
1970         }
1971
1972         if (required_features->reparse_points && !supported_features->reparse_points)
1973         {
1974                 if (supported_features->symlink_reparse_points) {
1975                         if (required_features->other_reparse_points) {
1976                                 WARNING(
1977           "%lu files in %"TS" are reparse points that are neither\n"
1978 "          symbolic links nor junction points and are not supported in\n"
1979 "          %"TS".  These reparse points will not be extracted.",
1980                                         required_features->other_reparse_points, loc,
1981                                         mode);
1982                         }
1983                 } else {
1984                         WARNING(
1985           "%lu files in %"TS" are reparse points, which are\n"
1986 "          not supported in %"TS" and will not be extracted.",
1987                                 required_features->reparse_points, loc, mode);
1988                 }
1989         }
1990
1991         if (required_features->security_descriptors &&
1992             !supported_features->security_descriptors)
1993         {
1994                 WARNING(
1995           "%lu files in %"TS" have Windows NT security descriptors,\n"
1996 "          but extracting security descriptors is not supported in\n"
1997 "          %"TS".  No security descriptors will be extracted.",
1998                         required_features->security_descriptors, loc, mode);
1999         }
2000
2001         if (required_features->short_names && !supported_features->short_names)
2002         {
2003                 WARNING(
2004           "%lu files in %"TS" have short (DOS) names, but\n"
2005 "          extracting short names is not supported in %"TS".\n"
2006 "          Short names will not be extracted.\n",
2007                         required_features->short_names, loc, mode);
2008         }
2009
2010         if ((extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) &&
2011             required_features->unix_data && !supported_features->unix_data)
2012         {
2013                 ERROR("Extracting UNIX data is not supported in %"TS, mode);
2014                 return WIMLIB_ERR_UNSUPPORTED;
2015         }
2016         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES) &&
2017             required_features->short_names && !supported_features->short_names)
2018         {
2019                 ERROR("Extracting short names is not supported in %"TS"", mode);
2020                 return WIMLIB_ERR_UNSUPPORTED;
2021         }
2022         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS) &&
2023             !ops->set_timestamps)
2024         {
2025                 ERROR("Extracting timestamps is not supported in %"TS"", mode);
2026                 return WIMLIB_ERR_UNSUPPORTED;
2027         }
2028         if (((extract_flags & (WIMLIB_EXTRACT_FLAG_STRICT_ACLS |
2029                                WIMLIB_EXTRACT_FLAG_UNIX_DATA))
2030              == WIMLIB_EXTRACT_FLAG_STRICT_ACLS) &&
2031             required_features->security_descriptors &&
2032             !supported_features->security_descriptors)
2033         {
2034                 ERROR("Extracting security descriptors is not supported in %"TS, mode);
2035                 return WIMLIB_ERR_UNSUPPORTED;
2036         }
2037
2038         if ((extract_flags & WIMLIB_EXTRACT_FLAG_HARDLINK) &&
2039             !supported_features->hard_links)
2040         {
2041                 ERROR("Hard link extraction mode requested, but "
2042                       "%"TS" does not support hard links!", mode);
2043                 return WIMLIB_ERR_UNSUPPORTED;
2044         }
2045
2046         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS) &&
2047             required_features->symlink_reparse_points &&
2048             !(supported_features->symlink_reparse_points ||
2049               supported_features->reparse_points))
2050         {
2051                 ERROR("Extracting symbolic links is not supported in %"TS, mode);
2052                 return WIMLIB_ERR_UNSUPPORTED;
2053         }
2054
2055         if ((extract_flags & WIMLIB_EXTRACT_FLAG_SYMLINK) &&
2056             !supported_features->symlink_reparse_points)
2057         {
2058                 ERROR("Symbolic link extraction mode requested, but "
2059                       "%"TS" does not support symbolic "
2060                       "links!", mode);
2061                 return WIMLIB_ERR_UNSUPPORTED;
2062         }
2063         return 0;
2064 }
2065
2066 static void
2067 do_extract_warnings(struct apply_ctx *ctx)
2068 {
2069         if (ctx->partial_security_descriptors == 0 &&
2070             ctx->no_security_descriptors == 0)
2071                 return;
2072
2073         WARNING("Extraction of \"%"TS"\" complete, but with one or more warnings:",
2074                 ctx->target);
2075         if (ctx->partial_security_descriptors != 0) {
2076                 WARNING("- Could only partially set the security descriptor\n"
2077                         "            on %lu files or directories.",
2078                         ctx->partial_security_descriptors);
2079         }
2080         if (ctx->no_security_descriptors != 0) {
2081                 WARNING("- Could not set security descriptor at all\n"
2082                         "            on %lu files or directories.",
2083                         ctx->no_security_descriptors);
2084         }
2085 #ifdef __WIN32__
2086         WARNING("To fully restore all security descriptors, run the program\n"
2087                 "          with Administrator rights.");
2088 #endif
2089 }
2090
2091 /*
2092  * extract_tree - Extract a file or directory tree from the currently selected
2093  *                WIM image.
2094  *
2095  * @wim:        WIMStruct for the WIM file, with the desired image selected
2096  *              (as wim->current_image).
2097  *
2098  * @wim_source_path:
2099  *              "Canonical" (i.e. no leading or trailing slashes, path
2100  *              separators WIM_PATH_SEPARATOR) path inside the WIM image to
2101  *              extract.  An empty string means the full image.
2102  *
2103  * @target:
2104  *              Filesystem path to extract the file or directory tree to.
2105  *              (Or, with WIMLIB_EXTRACT_FLAG_NTFS: the name of a NTFS volume.)
2106  *
2107  * @extract_flags:
2108  *              WIMLIB_EXTRACT_FLAG_*.  Also, the private flag
2109  *              WIMLIB_EXTRACT_FLAG_MULTI_IMAGE will be set if this is being
2110  *              called through wimlib_extract_image() with WIMLIB_ALL_IMAGES as
2111  *              the image.
2112  *
2113  * @progress_func:
2114  *              If non-NULL, progress function for the extraction.  The messages
2115  *              that may be sent in this function are:
2116  *
2117  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN or
2118  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN;
2119  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN;
2120  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END;
2121  *              WIMLIB_PROGRESS_MSG_EXTRACT_DENTRY;
2122  *              WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS;
2123  *              WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS;
2124  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END or
2125  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END.
2126  *
2127  * Returns 0 on success; a positive WIMLIB_ERR_* code on failure.
2128  */
2129 static int
2130 extract_tree(WIMStruct *wim, const tchar *wim_source_path, const tchar *target,
2131              int extract_flags, wimlib_progress_func_t progress_func)
2132 {
2133         struct wim_dentry *root;
2134         struct wim_features required_features;
2135         struct apply_ctx ctx;
2136         int ret;
2137         struct wim_lookup_table_entry *lte;
2138
2139         /* Start initializing the apply_ctx.  */
2140         memset(&ctx, 0, sizeof(struct apply_ctx));
2141         ctx.wim = wim;
2142         ctx.extract_flags = extract_flags;
2143         ctx.target = target;
2144         ctx.target_nchars = tstrlen(target);
2145         ctx.progress_func = progress_func;
2146         if (progress_func) {
2147                 ctx.progress.extract.wimfile_name = wim->filename;
2148                 ctx.progress.extract.image = wim->current_image;
2149                 ctx.progress.extract.extract_flags = (extract_flags &
2150                                                       WIMLIB_EXTRACT_MASK_PUBLIC);
2151                 ctx.progress.extract.image_name = wimlib_get_image_name(wim,
2152                                                                         wim->current_image);
2153                 ctx.progress.extract.extract_root_wim_source_path = wim_source_path;
2154                 ctx.progress.extract.target = target;
2155         }
2156         INIT_LIST_HEAD(&ctx.stream_list);
2157
2158         /* Translate the path to extract into the corresponding
2159          * `struct wim_dentry', which will be the root of the
2160          * "dentry tree" to extract.  */
2161         root = get_dentry(wim, wim_source_path);
2162         if (!root) {
2163                 ERROR("Path \"%"TS"\" does not exist in WIM image %d",
2164                       wim_source_path, wim->current_image);
2165                 ret = WIMLIB_ERR_PATH_DOES_NOT_EXIST;
2166                 goto out;
2167         }
2168
2169         ctx.extract_root = root;
2170
2171         /* Select the appropriate apply_operations based on the
2172          * platform and extract_flags.  */
2173 #ifdef __WIN32__
2174         ctx.ops = &win32_apply_ops;
2175 #else
2176         ctx.ops = &unix_apply_ops;
2177 #endif
2178
2179 #ifdef WITH_NTFS_3G
2180         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS)
2181                 ctx.ops = &ntfs_3g_apply_ops;
2182 #endif
2183
2184         /* Call the start_extract() callback.  This gives the apply_operations
2185          * implementation a chance to do any setup needed to access the volume.
2186          * Furthermore, it's expected to set the supported features of this
2187          * extraction mode (ctx.supported_features), which are determined at
2188          * runtime as they may vary depending on the actual volume.  These
2189          * features are then compared with the actual features extracting this
2190          * dentry tree requires.  Some mismatches will merely produce warnings
2191          * and the unsupported data will be ignored; others will produce errors.
2192          */
2193         ret = ctx.ops->start_extract(target, &ctx);
2194         if (ret)
2195                 goto out;
2196
2197         dentry_tree_get_features(root, &required_features);
2198         ret = do_feature_check(&required_features, &ctx.supported_features,
2199                                extract_flags, ctx.ops, wim_source_path);
2200         if (ret)
2201                 goto out_finish_or_abort_extract;
2202
2203         ctx.supported_attributes_mask =
2204                 compute_supported_attributes_mask(&ctx.supported_features);
2205
2206         /* Figure out whether the root dentry is being extracted to the root of
2207          * a volume and therefore needs to be treated "specially", for example
2208          * not being explicitly created and not having attributes set.  */
2209         if (ctx.ops->target_is_root && ctx.ops->root_directory_is_special)
2210                 ctx.root_dentry_is_special = ctx.ops->target_is_root(target);
2211
2212         /* Calculate the actual filename component of each extracted dentry.  In
2213          * the process, set the dentry->extraction_skipped flag on dentries that
2214          * are being skipped for some reason (e.g. invalid filename).  */
2215         ret = for_dentry_in_tree(root, dentry_calculate_extraction_path, &ctx);
2216         if (ret)
2217                 goto out_dentry_reset_needs_extraction;
2218
2219         /* Build the list of the streams that need to be extracted and
2220          * initialize ctx.progress.extract with stream information.  */
2221         ret = for_dentry_in_tree(ctx.extract_root,
2222                                  dentry_resolve_and_zero_lte_refcnt, &ctx);
2223         if (ret)
2224                 goto out_dentry_reset_needs_extraction;
2225
2226         ret = for_dentry_in_tree(ctx.extract_root,
2227                                  dentry_add_streams_to_extract, &ctx);
2228         if (ret)
2229                 goto out_teardown_stream_list;
2230
2231         if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2232                 /* When extracting from a pipe, the number of bytes of data to
2233                  * extract can't be determined in the normal way (examining the
2234                  * lookup table), since at this point all we have is a set of
2235                  * SHA1 message digests of streams that need to be extracted.
2236                  * However, we can get a reasonably accurate estimate by taking
2237                  * <TOTALBYTES> from the corresponding <IMAGE> in the WIM XML
2238                  * data.  This does assume that a full image is being extracted,
2239                  * but currently there is no API for doing otherwise.  (Also,
2240                  * subtract <HARDLINKBYTES> from this if hard links are
2241                  * supported by the extraction mode.)  */
2242                 ctx.progress.extract.total_bytes =
2243                         wim_info_get_image_total_bytes(wim->wim_info,
2244                                                        wim->current_image);
2245                 if (ctx.supported_features.hard_links) {
2246                         ctx.progress.extract.total_bytes -=
2247                                 wim_info_get_image_hard_link_bytes(wim->wim_info,
2248                                                                    wim->current_image);
2249                 }
2250         }
2251
2252         /* Handle the special case of extracting a file to standard
2253          * output.  In that case, "root" should be a single file, not a
2254          * directory tree.  (If not, extract_dentry_to_stdout() will
2255          * return an error.)  */
2256         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT) {
2257                 ret = extract_dentry_to_stdout(root);
2258                 goto out_teardown_stream_list;
2259         }
2260
2261         /* If a sequential extraction was specified, sort the streams to be
2262          * extracted by their position in the WIM file so that the WIM file can
2263          * be read sequentially.  */
2264         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2265                               WIMLIB_EXTRACT_FLAG_FROM_PIPE))
2266                                         == WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2267         {
2268                 ret = sort_stream_list_by_sequential_order(
2269                                 &ctx.stream_list,
2270                                 offsetof(struct wim_lookup_table_entry,
2271                                          extraction_list));
2272                 if (ret)
2273                         goto out_teardown_stream_list;
2274         }
2275
2276         if (ctx.ops->realpath_works_on_nonexisting_files &&
2277             ((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) ||
2278              ctx.ops->requires_realtarget_in_paths))
2279         {
2280                 ctx.realtarget = realpath(target, NULL);
2281                 if (!ctx.realtarget) {
2282                         ret = WIMLIB_ERR_NOMEM;
2283                         goto out_teardown_stream_list;
2284                 }
2285                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2286         }
2287
2288         if (progress_func) {
2289                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN :
2290                                                  WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN,
2291                               &ctx.progress);
2292         }
2293
2294         if (!ctx.root_dentry_is_special)
2295         {
2296                 tchar path[ctx.ops->path_max];
2297                 if (build_extraction_path(path, root, &ctx))
2298                 {
2299                         ret = extract_inode(path, &ctx, root->d_inode);
2300                         if (ret)
2301                                 goto out_free_realtarget;
2302                 }
2303         }
2304
2305         /* If we need to fix up the targets of absolute symbolic links
2306          * (WIMLIB_EXTRACT_FLAG_RPFIX) or the extraction mode requires paths to
2307          * be absolute, use realpath() (or its replacement on Windows) to get
2308          * the absolute path to the extraction target.  Note that this requires
2309          * the target directory to exist, unless
2310          * realpath_works_on_nonexisting_files is set in the apply_operations.
2311          * */
2312         if (!ctx.realtarget &&
2313             (((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
2314               required_features.symlink_reparse_points) ||
2315              ctx.ops->requires_realtarget_in_paths))
2316         {
2317                 ctx.realtarget = realpath(target, NULL);
2318                 if (!ctx.realtarget) {
2319                         ret = WIMLIB_ERR_NOMEM;
2320                         goto out_free_realtarget;
2321                 }
2322                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2323         }
2324
2325         /* Finally, the important part: extract the tree of files.  */
2326         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2327                              WIMLIB_EXTRACT_FLAG_FROM_PIPE)) {
2328                 /* Sequential extraction requested, so two passes are needed
2329                  * (one for directory structure, one for streams.)  */
2330                 if (progress_func)
2331                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2332                                       &ctx.progress);
2333
2334                 if (!(extract_flags & WIMLIB_EXTRACT_FLAG_RESUME)) {
2335                         ret = for_dentry_in_tree(root, dentry_extract_skeleton, &ctx);
2336                         if (ret)
2337                                 goto out_free_realtarget;
2338                 }
2339                 if (progress_func)
2340                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2341                                       &ctx.progress);
2342                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE)
2343                         ret = extract_streams_from_pipe(&ctx);
2344                 else
2345                         ret = extract_stream_list(&ctx);
2346                 if (ret)
2347                         goto out_free_realtarget;
2348         } else {
2349                 /* Sequential extraction was not requested, so we can make do
2350                  * with one pass where we both create the files and extract
2351                  * streams.   */
2352                 if (progress_func)
2353                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2354                                       &ctx.progress);
2355                 ret = for_dentry_in_tree(root, dentry_extract, &ctx);
2356                 if (ret)
2357                         goto out_free_realtarget;
2358                 if (progress_func)
2359                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2360                                       &ctx.progress);
2361         }
2362
2363         /* If the total number of bytes to extract was miscalculated, just jump
2364          * to the calculated number in order to avoid confusing the progress
2365          * function.  This should only occur when extracting from a pipe.  */
2366         if (ctx.progress.extract.completed_bytes != ctx.progress.extract.total_bytes)
2367         {
2368                 DEBUG("Calculated %"PRIu64" bytes to extract, but actually "
2369                       "extracted %"PRIu64,
2370                       ctx.progress.extract.total_bytes,
2371                       ctx.progress.extract.completed_bytes);
2372         }
2373         if (progress_func &&
2374             ctx.progress.extract.completed_bytes < ctx.progress.extract.total_bytes)
2375         {
2376                 ctx.progress.extract.completed_bytes = ctx.progress.extract.total_bytes;
2377                 progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS, &ctx.progress);
2378         }
2379
2380         /* Apply security descriptors and timestamps.  This is done at the end,
2381          * and in a depth-first manner, to prevent timestamps from getting
2382          * changed by subsequent extract operations and to minimize the chance
2383          * of the restored security descriptors getting in our way.  */
2384         if (progress_func)
2385                 progress_func(WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS,
2386                               &ctx.progress);
2387         ret = for_dentry_in_tree_depth(root, dentry_extract_final, &ctx);
2388         if (ret)
2389                 goto out_free_realtarget;
2390
2391         if (progress_func) {
2392                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END :
2393                               WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END,
2394                               &ctx.progress);
2395         }
2396
2397         do_extract_warnings(&ctx);
2398
2399         ret = 0;
2400 out_free_realtarget:
2401         FREE(ctx.realtarget);
2402 out_teardown_stream_list:
2403         /* Free memory allocated as part of the mapping from each
2404          * wim_lookup_table_entry to the dentries that reference it.  */
2405         if (ctx.extract_flags & WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2406                 list_for_each_entry(lte, &ctx.stream_list, extraction_list)
2407                         if (lte->out_refcnt > ARRAY_LEN(lte->inline_lte_dentries))
2408                                 FREE(lte->lte_dentries);
2409 out_dentry_reset_needs_extraction:
2410         for_dentry_in_tree(root, dentry_reset_needs_extraction, NULL);
2411 out_finish_or_abort_extract:
2412         if (ret) {
2413                 if (ctx.ops->abort_extract)
2414                         ctx.ops->abort_extract(&ctx);
2415         } else {
2416                 if (ctx.ops->finish_extract)
2417                         ret = ctx.ops->finish_extract(&ctx);
2418         }
2419 out:
2420         return ret;
2421 }
2422
2423 /* Validates a single wimlib_extract_command, mostly checking to make sure the
2424  * extract flags make sense. */
2425 static int
2426 check_extract_command(struct wimlib_extract_command *cmd, int wim_header_flags)
2427 {
2428         int extract_flags;
2429
2430         /* Empty destination path? */
2431         if (cmd->fs_dest_path[0] == T('\0'))
2432                 return WIMLIB_ERR_INVALID_PARAM;
2433
2434         extract_flags = cmd->extract_flags;
2435
2436         /* Check for invalid flag combinations  */
2437         if ((extract_flags &
2438              (WIMLIB_EXTRACT_FLAG_SYMLINK |
2439               WIMLIB_EXTRACT_FLAG_HARDLINK)) == (WIMLIB_EXTRACT_FLAG_SYMLINK |
2440                                                  WIMLIB_EXTRACT_FLAG_HARDLINK))
2441                 return WIMLIB_ERR_INVALID_PARAM;
2442
2443         if ((extract_flags &
2444              (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2445               WIMLIB_EXTRACT_FLAG_STRICT_ACLS)) == (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2446                                                     WIMLIB_EXTRACT_FLAG_STRICT_ACLS))
2447                 return WIMLIB_ERR_INVALID_PARAM;
2448
2449         if ((extract_flags &
2450              (WIMLIB_EXTRACT_FLAG_RPFIX |
2451               WIMLIB_EXTRACT_FLAG_NORPFIX)) == (WIMLIB_EXTRACT_FLAG_RPFIX |
2452                                                 WIMLIB_EXTRACT_FLAG_NORPFIX))
2453                 return WIMLIB_ERR_INVALID_PARAM;
2454
2455         if ((extract_flags &
2456              (WIMLIB_EXTRACT_FLAG_RESUME |
2457               WIMLIB_EXTRACT_FLAG_FROM_PIPE)) == WIMLIB_EXTRACT_FLAG_RESUME)
2458                 return WIMLIB_ERR_INVALID_PARAM;
2459
2460         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2461 #ifndef WITH_NTFS_3G
2462                 ERROR("wimlib was compiled without support for NTFS-3g, so\n"
2463                       "        we cannot apply a WIM image directly to a NTFS volume.");
2464                 return WIMLIB_ERR_UNSUPPORTED;
2465 #endif
2466         }
2467
2468         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_RPFIX |
2469                               WIMLIB_EXTRACT_FLAG_NORPFIX)) == 0)
2470         {
2471                 /* Do reparse point fixups by default if the WIM header says
2472                  * they are enabled and we are extracting a full image. */
2473                 if (wim_header_flags & WIM_HDR_FLAG_RP_FIX)
2474                         extract_flags |= WIMLIB_EXTRACT_FLAG_RPFIX;
2475         }
2476
2477         /* TODO: Since UNIX data entries are stored in the file resources, in a
2478          * completely sequential extraction they may come up before the
2479          * corresponding file or symbolic link data.  This needs to be handled
2480          * better.  */
2481         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2482                               WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2483                                     == (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2484                                         WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2485         {
2486                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2487                         WARNING("Setting UNIX file/owner group may "
2488                                 "be impossible on some\n"
2489                                 "          symbolic links "
2490                                 "when applying from a pipe.");
2491                 } else {
2492                         extract_flags &= ~WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2493                         WARNING("Disabling sequential extraction for "
2494                                 "UNIX data mode");
2495                 }
2496         }
2497
2498         cmd->extract_flags = extract_flags;
2499         return 0;
2500 }
2501
2502
2503 /* Internal function to execute extraction commands for a WIM image.  The paths
2504  * in the extract commands are expected to be already "canonicalized".  */
2505 static int
2506 do_wimlib_extract_files(WIMStruct *wim,
2507                         int image,
2508                         struct wimlib_extract_command *cmds,
2509                         size_t num_cmds,
2510                         wimlib_progress_func_t progress_func)
2511 {
2512         int ret;
2513         bool found_link_cmd = false;
2514         bool found_nolink_cmd = false;
2515
2516         /* Select the image from which we are extracting files */
2517         ret = select_wim_image(wim, image);
2518         if (ret)
2519                 return ret;
2520
2521         /* Make sure there are no streams in the WIM that have not been
2522          * checksummed yet.  */
2523         ret = wim_checksum_unhashed_streams(wim);
2524         if (ret)
2525                 return ret;
2526
2527         /* Check for problems with the extraction commands */
2528         for (size_t i = 0; i < num_cmds; i++) {
2529                 ret = check_extract_command(&cmds[i], wim->hdr.flags);
2530                 if (ret)
2531                         return ret;
2532                 if (cmds[i].extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2533                                              WIMLIB_EXTRACT_FLAG_HARDLINK)) {
2534                         found_link_cmd = true;
2535                 } else {
2536                         found_nolink_cmd = true;
2537                 }
2538                 if (found_link_cmd && found_nolink_cmd) {
2539                         ERROR("Symlink or hardlink extraction mode must "
2540                               "be set on all extraction commands");
2541                         return WIMLIB_ERR_INVALID_PARAM;
2542                 }
2543         }
2544
2545         /* Execute the extraction commands */
2546         for (size_t i = 0; i < num_cmds; i++) {
2547                 ret = extract_tree(wim,
2548                                    cmds[i].wim_source_path,
2549                                    cmds[i].fs_dest_path,
2550                                    cmds[i].extract_flags,
2551                                    progress_func);
2552                 if (ret)
2553                         return ret;
2554         }
2555         return 0;
2556 }
2557
2558 /* API function documented in wimlib.h  */
2559 WIMLIBAPI int
2560 wimlib_extract_files(WIMStruct *wim,
2561                      int image,
2562                      const struct wimlib_extract_command *cmds,
2563                      size_t num_cmds,
2564                      int default_extract_flags,
2565                      WIMStruct **additional_swms,
2566                      unsigned num_additional_swms,
2567                      wimlib_progress_func_t progress_func)
2568 {
2569         int ret;
2570         struct wimlib_extract_command *cmds_copy;
2571         int all_flags = 0;
2572
2573         default_extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2574
2575         ret = verify_swm_set(wim, additional_swms, num_additional_swms);
2576         if (ret)
2577                 goto out;
2578
2579         if (num_cmds == 0)
2580                 goto out;
2581
2582         if (num_additional_swms)
2583                 merge_lookup_tables(wim, additional_swms, num_additional_swms);
2584
2585         cmds_copy = CALLOC(num_cmds, sizeof(cmds[0]));
2586         if (!cmds_copy) {
2587                 ret = WIMLIB_ERR_NOMEM;
2588                 goto out_restore_lookup_table;
2589         }
2590
2591         for (size_t i = 0; i < num_cmds; i++) {
2592                 cmds_copy[i].extract_flags = (default_extract_flags |
2593                                                  cmds[i].extract_flags)
2594                                                 & WIMLIB_EXTRACT_MASK_PUBLIC;
2595                 all_flags |= cmds_copy[i].extract_flags;
2596
2597                 cmds_copy[i].wim_source_path = canonicalize_wim_path(cmds[i].wim_source_path);
2598                 if (!cmds_copy[i].wim_source_path) {
2599                         ret = WIMLIB_ERR_NOMEM;
2600                         goto out_free_cmds_copy;
2601                 }
2602
2603                 cmds_copy[i].fs_dest_path = canonicalize_fs_path(cmds[i].fs_dest_path);
2604                 if (!cmds_copy[i].fs_dest_path) {
2605                         ret = WIMLIB_ERR_NOMEM;
2606                         goto out_free_cmds_copy;
2607                 }
2608
2609         }
2610         ret = do_wimlib_extract_files(wim, image,
2611                                       cmds_copy, num_cmds,
2612                                       progress_func);
2613
2614         if (all_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2615                          WIMLIB_EXTRACT_FLAG_HARDLINK))
2616         {
2617                 for_lookup_table_entry(wim->lookup_table,
2618                                        lte_free_extracted_file, NULL);
2619         }
2620 out_free_cmds_copy:
2621         for (size_t i = 0; i < num_cmds; i++) {
2622                 FREE(cmds_copy[i].wim_source_path);
2623                 FREE(cmds_copy[i].fs_dest_path);
2624         }
2625         FREE(cmds_copy);
2626 out_restore_lookup_table:
2627         if (num_additional_swms)
2628                 unmerge_lookup_table(wim);
2629 out:
2630         return ret;
2631 }
2632
2633 /*
2634  * Extracts an image from a WIM file.
2635  *
2636  * @wim:                WIMStruct for the WIM file.
2637  *
2638  * @image:              Number of the single image to extract.
2639  *
2640  * @target:             Directory or NTFS volume to extract the image to.
2641  *
2642  * @extract_flags:      Bitwise or of WIMLIB_EXTRACT_FLAG_*.
2643  *
2644  * @progress_func:      If non-NULL, a progress function to be called
2645  *                      periodically.
2646  *
2647  * Returns 0 on success; nonzero on failure.
2648  */
2649 static int
2650 extract_single_image(WIMStruct *wim, int image,
2651                      const tchar *target, int extract_flags,
2652                      wimlib_progress_func_t progress_func)
2653 {
2654         int ret;
2655         tchar *target_copy = canonicalize_fs_path(target);
2656         if (!target_copy)
2657                 return WIMLIB_ERR_NOMEM;
2658         struct wimlib_extract_command cmd = {
2659                 .wim_source_path = T(""),
2660                 .fs_dest_path = target_copy,
2661                 .extract_flags = extract_flags,
2662         };
2663         ret = do_wimlib_extract_files(wim, image, &cmd, 1, progress_func);
2664         FREE(target_copy);
2665         return ret;
2666 }
2667
2668 static const tchar * const filename_forbidden_chars =
2669 T(
2670 #ifdef __WIN32__
2671 "<>:\"/\\|?*"
2672 #else
2673 "/"
2674 #endif
2675 );
2676
2677 /* This function checks if it is okay to use a WIM image's name as a directory
2678  * name.  */
2679 static bool
2680 image_name_ok_as_dir(const tchar *image_name)
2681 {
2682         return image_name && *image_name &&
2683                 !tstrpbrk(image_name, filename_forbidden_chars) &&
2684                 tstrcmp(image_name, T(".")) &&
2685                 tstrcmp(image_name, T(".."));
2686 }
2687
2688 /* Extracts all images from the WIM to the directory @target, with the images
2689  * placed in subdirectories named by their image names. */
2690 static int
2691 extract_all_images(WIMStruct *wim,
2692                    const tchar *target,
2693                    int extract_flags,
2694                    wimlib_progress_func_t progress_func)
2695 {
2696         size_t image_name_max_len = max(xml_get_max_image_name_len(wim), 20);
2697         size_t output_path_len = tstrlen(target);
2698         tchar buf[output_path_len + 1 + image_name_max_len + 1];
2699         int ret;
2700         int image;
2701         const tchar *image_name;
2702         struct stat stbuf;
2703
2704         extract_flags |= WIMLIB_EXTRACT_FLAG_MULTI_IMAGE;
2705
2706         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2707                 ERROR("Cannot extract multiple images in NTFS extraction mode.");
2708                 return WIMLIB_ERR_INVALID_PARAM;
2709         }
2710
2711         if (tstat(target, &stbuf)) {
2712                 if (errno == ENOENT) {
2713                         if (tmkdir(target, 0755)) {
2714                                 ERROR_WITH_ERRNO("Failed to create directory \"%"TS"\"", target);
2715                                 return WIMLIB_ERR_MKDIR;
2716                         }
2717                 } else {
2718                         ERROR_WITH_ERRNO("Failed to stat \"%"TS"\"", target);
2719                         return WIMLIB_ERR_STAT;
2720                 }
2721         } else if (!S_ISDIR(stbuf.st_mode)) {
2722                 ERROR("\"%"TS"\" is not a directory", target);
2723                 return WIMLIB_ERR_NOTDIR;
2724         }
2725
2726         tmemcpy(buf, target, output_path_len);
2727         buf[output_path_len] = OS_PREFERRED_PATH_SEPARATOR;
2728         for (image = 1; image <= wim->hdr.image_count; image++) {
2729                 image_name = wimlib_get_image_name(wim, image);
2730                 if (image_name_ok_as_dir(image_name)) {
2731                         tstrcpy(buf + output_path_len + 1, image_name);
2732                 } else {
2733                         /* Image name is empty or contains forbidden characters.
2734                          * Use image number instead. */
2735                         tsprintf(buf + output_path_len + 1, T("%d"), image);
2736                 }
2737                 ret = extract_single_image(wim, image, buf, extract_flags,
2738                                            progress_func);
2739                 if (ret)
2740                         return ret;
2741         }
2742         return 0;
2743 }
2744
2745 static int
2746 do_wimlib_extract_image(WIMStruct *wim,
2747                         int image,
2748                         const tchar *target,
2749                         int extract_flags,
2750                         WIMStruct **additional_swms,
2751                         unsigned num_additional_swms,
2752                         wimlib_progress_func_t progress_func)
2753 {
2754         int ret;
2755
2756         if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2757                 wimlib_assert(wim->hdr.part_number == 1);
2758                 wimlib_assert(num_additional_swms == 0);
2759         } else {
2760                 ret = verify_swm_set(wim, additional_swms, num_additional_swms);
2761                 if (ret)
2762                         return ret;
2763
2764                 if (num_additional_swms)
2765                         merge_lookup_tables(wim, additional_swms, num_additional_swms);
2766         }
2767
2768         if (image == WIMLIB_ALL_IMAGES) {
2769                 ret = extract_all_images(wim, target, extract_flags,
2770                                          progress_func);
2771         } else {
2772                 ret = extract_single_image(wim, image, target, extract_flags,
2773                                            progress_func);
2774         }
2775
2776         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2777                              WIMLIB_EXTRACT_FLAG_HARDLINK))
2778         {
2779                 for_lookup_table_entry(wim->lookup_table,
2780                                        lte_free_extracted_file,
2781                                        NULL);
2782         }
2783         if (num_additional_swms)
2784                 unmerge_lookup_table(wim);
2785         return ret;
2786 }
2787
2788 /* API function documented in wimlib.h  */
2789 WIMLIBAPI int
2790 wimlib_extract_image_from_pipe(int pipe_fd, const tchar *image_num_or_name,
2791                                const tchar *target, int extract_flags,
2792                                wimlib_progress_func_t progress_func)
2793 {
2794         int ret;
2795         WIMStruct *pwm;
2796         struct filedes *in_fd;
2797         int image;
2798         unsigned i;
2799
2800         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2801
2802         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT)
2803                 return WIMLIB_ERR_INVALID_PARAM;
2804
2805         extract_flags |= WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2806
2807         /* Read the WIM header from the pipe and get a WIMStruct to represent
2808          * the pipable WIM.  Caveats:  Unlike getting a WIMStruct with
2809          * wimlib_open_wim(), getting a WIMStruct in this way will result in
2810          * an empty lookup table, no XML data read, and no filename set.  */
2811         ret = open_wim_as_WIMStruct(&pipe_fd,
2812                                     WIMLIB_OPEN_FLAG_FROM_PIPE |
2813                                                 WIMLIB_OPEN_FLAG_SPLIT_OK,
2814                                     &pwm, progress_func);
2815         if (ret)
2816                 return ret;
2817
2818         /* Sanity check to make sure this is a pipable WIM.  */
2819         if (pwm->hdr.magic != PWM_MAGIC) {
2820                 ERROR("The WIM being read from file descriptor %d "
2821                       "is not pipable!", pipe_fd);
2822                 ret = WIMLIB_ERR_NOT_PIPABLE;
2823                 goto out_wimlib_free;
2824         }
2825
2826         /* Sanity check to make sure the first part of a pipable split WIM is
2827          * sent over the pipe first.  */
2828         if (pwm->hdr.part_number != 1) {
2829                 ERROR("The first part of the split WIM must be "
2830                       "sent over the pipe first.");
2831                 ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2832                 goto out_wimlib_free;
2833         }
2834
2835         in_fd = &pwm->in_fd;
2836         wimlib_assert(in_fd->offset == WIM_HEADER_DISK_SIZE);
2837
2838         /* As mentioned, the WIMStruct we created from the pipe does not have
2839          * XML data yet.  Fix this by reading the extra copy of the XML data
2840          * that directly follows the header in pipable WIMs.  (Note: see
2841          * write_pipable_wim() for more details about the format of pipable
2842          * WIMs.)  */
2843         {
2844                 struct wim_lookup_table_entry xml_lte;
2845                 ret = read_pwm_stream_header(pwm, &xml_lte, 0, NULL);
2846                 if (ret)
2847                         goto out_wimlib_free;
2848
2849                 if (!(xml_lte.resource_entry.flags & WIM_RESHDR_FLAG_METADATA))
2850                 {
2851                         ERROR("Expected XML data, but found non-metadata "
2852                               "stream.");
2853                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2854                         goto out_wimlib_free;
2855                 }
2856
2857                 copy_resource_entry(&pwm->hdr.xml_res_entry,
2858                                     &xml_lte.resource_entry);
2859
2860                 ret = read_wim_xml_data(pwm);
2861                 if (ret)
2862                         goto out_wimlib_free;
2863                 if (wim_info_get_num_images(pwm->wim_info) != pwm->hdr.image_count) {
2864                         ERROR("Image count in XML data is not the same as in WIM header.");
2865                         ret = WIMLIB_ERR_XML;
2866                         goto out_wimlib_free;
2867                 }
2868         }
2869
2870         /* Get image index (this may use the XML data that was just read to
2871          * resolve an image name).  */
2872         if (image_num_or_name) {
2873                 image = wimlib_resolve_image(pwm, image_num_or_name);
2874                 if (image == WIMLIB_NO_IMAGE) {
2875                         ERROR("\"%"TS"\" is not a valid image in the pipable WIM!",
2876                               image_num_or_name);
2877                         ret = WIMLIB_ERR_INVALID_IMAGE;
2878                         goto out_wimlib_free;
2879                 } else if (image == WIMLIB_ALL_IMAGES) {
2880                         ERROR("Applying all images from a pipe is not supported.");
2881                         ret = WIMLIB_ERR_INVALID_IMAGE;
2882                         goto out_wimlib_free;
2883                 }
2884         } else {
2885                 if (pwm->hdr.image_count != 1) {
2886                         ERROR("No image was specified, but the pipable WIM "
2887                               "did not contain exactly 1 image");
2888                         ret = WIMLIB_ERR_INVALID_IMAGE;
2889                         goto out_wimlib_free;
2890                 }
2891                 image = 1;
2892         }
2893
2894         /* Load the needed metadata resource.  */
2895         for (i = 1; i <= pwm->hdr.image_count; i++) {
2896                 struct wim_lookup_table_entry *metadata_lte;
2897                 struct wim_image_metadata *imd;
2898
2899                 metadata_lte = new_lookup_table_entry();
2900                 if (!metadata_lte) {
2901                         ret = WIMLIB_ERR_NOMEM;
2902                         goto out_wimlib_free;
2903                 }
2904
2905                 ret = read_pwm_stream_header(pwm, metadata_lte, 0, NULL);
2906                 imd = pwm->image_metadata[i - 1];
2907                 imd->metadata_lte = metadata_lte;
2908                 if (ret)
2909                         goto out_wimlib_free;
2910
2911                 if (!(metadata_lte->resource_entry.flags &
2912                       WIM_RESHDR_FLAG_METADATA))
2913                 {
2914                         ERROR("Expected metadata resource, but found "
2915                               "non-metadata stream.");
2916                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2917                         goto out_wimlib_free;
2918                 }
2919
2920                 if (i == image) {
2921                         /* Metadata resource is for the images being extracted.
2922                          * Parse it and save the metadata in memory.  */
2923                         ret = read_metadata_resource(pwm, imd);
2924                         if (ret)
2925                                 goto out_wimlib_free;
2926                         imd->modified = 1;
2927                 } else {
2928                         /* Metadata resource is not for the image being
2929                          * extracted.  Skip over it.  */
2930                         ret = skip_pwm_stream(metadata_lte);
2931                         if (ret)
2932                                 goto out_wimlib_free;
2933                 }
2934         }
2935         /* Extract the image.  */
2936         extract_flags |= WIMLIB_EXTRACT_FLAG_FROM_PIPE;
2937         ret = do_wimlib_extract_image(pwm, image, target,
2938                                       extract_flags, NULL, 0, progress_func);
2939         /* Clean up and return.  */
2940 out_wimlib_free:
2941         wimlib_free(pwm);
2942         return ret;
2943 }
2944
2945 /* API function documented in wimlib.h  */
2946 WIMLIBAPI int
2947 wimlib_extract_image(WIMStruct *wim,
2948                      int image,
2949                      const tchar *target,
2950                      int extract_flags,
2951                      WIMStruct **additional_swms,
2952                      unsigned num_additional_swms,
2953                      wimlib_progress_func_t progress_func)
2954 {
2955         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2956         return do_wimlib_extract_image(wim, image, target, extract_flags,
2957                                        additional_swms, num_additional_swms,
2958                                        progress_func);
2959 }