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