wimlib-imagex, wimlib: Optimize appending new image as update of existing
[wimlib] / src / extract.c
1 /*
2  * extract.c
3  *
4  * Support for extracting WIM images, or files or directories contained in a WIM
5  * image.
6  */
7
8 /*
9  * Copyright (C) 2012, 2013 Eric Biggers
10  *
11  * This file is part of wimlib, a library for working with WIM files.
12  *
13  * wimlib is free software; you can redistribute it and/or modify it under the
14  * terms of the GNU General Public License as published by the Free
15  * Software Foundation; either version 3 of the License, or (at your option)
16  * any later version.
17  *
18  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20  * A PARTICULAR PURPOSE. See the GNU General Public License for more
21  * details.
22  *
23  * You should have received a copy of the GNU General Public License
24  * along with wimlib; if not, see http://www.gnu.org/licenses/.
25  */
26
27 /*
28  * This file provides the API functions wimlib_extract_image(),
29  * wimlib_extract_files(), and wimlib_extract_image_from_pipe().  Internally,
30  * all end up calling extract_tree() zero or more times to extract a tree of
31  * files from the currently selected WIM image to the specified target directory
32  * or NTFS volume.
33  *
34  * Although wimlib supports multiple extraction modes/backends (NTFS-3g, UNIX,
35  * Win32), this file does not itself have code to extract files or directories
36  * to any specific target; instead, it handles generic functionality and relies
37  * on lower-level callback functions declared in `struct apply_operations' to do
38  * the actual extraction.
39  */
40
41 #ifdef HAVE_CONFIG_H
42 #  include "config.h"
43 #endif
44
45 #include "wimlib/apply.h"
46 #include "wimlib/dentry.h"
47 #include "wimlib/encoding.h"
48 #include "wimlib/endianness.h"
49 #include "wimlib/error.h"
50 #include "wimlib/lookup_table.h"
51 #include "wimlib/metadata.h"
52 #include "wimlib/paths.h"
53 #include "wimlib/reparse.h"
54 #include "wimlib/resource.h"
55 #include "wimlib/security.h"
56 #include "wimlib/swm.h"
57 #ifdef __WIN32__
58 #  include "wimlib/win32.h" /* for realpath() equivalent */
59 #endif
60 #include "wimlib/xml.h"
61 #include "wimlib/wim.h"
62
63 #include <errno.h>
64 #include <fcntl.h>
65 #include <stdlib.h>
66 #include <sys/stat.h>
67 #include <unistd.h>
68
69 #define WIMLIB_EXTRACT_FLAG_MULTI_IMAGE 0x80000000
70 #define WIMLIB_EXTRACT_FLAG_FROM_PIPE   0x40000000
71 #define WIMLIB_EXTRACT_MASK_PUBLIC      0x3fffffff
72
73 /* Given a WIM dentry in the tree to be extracted, resolve all streams in the
74  * corresponding inode and set 'out_refcnt' in each to 0.  */
75 static int
76 dentry_resolve_and_zero_lte_refcnt(struct wim_dentry *dentry, void *_ctx)
77 {
78         struct apply_ctx *ctx = _ctx;
79         struct wim_inode *inode = dentry->d_inode;
80         struct wim_lookup_table_entry *lte;
81         int ret;
82         bool force = false;
83
84         if (dentry->extraction_skipped)
85                 return 0;
86
87         /* Special case:  when extracting from a pipe, the WIM lookup table is
88          * initially empty, so "resolving" an inode's streams is initially not
89          * possible.  However, we still need to keep track of which streams,
90          * identified by SHA1 message digests, need to be extracted, so we
91          * "resolve" the inode's streams anyway by allocating new entries.  */
92         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE)
93                 force = true;
94         ret = inode_resolve_ltes(inode, ctx->wim->lookup_table, force);
95         if (ret)
96                 return ret;
97         for (unsigned i = 0; i <= inode->i_num_ads; i++) {
98                 lte = inode_stream_lte_resolved(inode, i);
99                 if (lte)
100                         lte->out_refcnt = 0;
101         }
102         return 0;
103 }
104
105 static inline bool
106 is_linked_extraction(const struct apply_ctx *ctx)
107 {
108         return 0 != (ctx->extract_flags & (WIMLIB_EXTRACT_FLAG_HARDLINK |
109                                            WIMLIB_EXTRACT_FLAG_SYMLINK));
110 }
111
112 static inline bool
113 can_extract_named_data_streams(const struct apply_ctx *ctx)
114 {
115         return ctx->supported_features.named_data_streams &&
116                 !is_linked_extraction(ctx);
117 }
118
119 static int
120 ref_stream_to_extract(struct wim_lookup_table_entry *lte,
121                       struct wim_dentry *dentry, struct apply_ctx *ctx)
122 {
123         if (!lte)
124                 return 0;
125
126         /* Tally the size only for each extraction of the stream (not hard
127          * links).  */
128         if (!(dentry->d_inode->i_visited &&
129               ctx->supported_features.hard_links) &&
130             (!is_linked_extraction(ctx) || (lte->out_refcnt == 0 &&
131                                             lte->extracted_file == NULL)))
132         {
133                 ctx->progress.extract.total_bytes += wim_resource_size(lte);
134                 ctx->progress.extract.num_streams++;
135         }
136
137         /* Add stream to the extraction_list only one time, even if it's going
138          * to be extracted to multiple locations.  */
139         if (lte->out_refcnt == 0) {
140                 list_add_tail(&lte->extraction_list, &ctx->stream_list);
141                 ctx->num_streams_remaining++;
142         }
143
144         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_SEQUENTIAL) {
145                 struct wim_dentry **lte_dentries;
146
147                 /* Append dentry to this stream's array of dentries referencing
148                  * it.  Use inline array to avoid memory allocation until the
149                  * number of dentries becomes too large.  */
150                 if (lte->out_refcnt < ARRAY_LEN(lte->inline_lte_dentries)) {
151                         lte_dentries = lte->inline_lte_dentries;
152                 } else {
153                         struct wim_dentry **prev_lte_dentries;
154                         size_t alloc_lte_dentries;
155
156                         if (lte->out_refcnt == ARRAY_LEN(lte->inline_lte_dentries)) {
157                                 prev_lte_dentries = NULL;
158                                 alloc_lte_dentries = ARRAY_LEN(lte->inline_lte_dentries);
159                         } else {
160                                 prev_lte_dentries = lte->lte_dentries;
161                                 alloc_lte_dentries = lte->alloc_lte_dentries;
162                         }
163
164                         if (lte->out_refcnt == alloc_lte_dentries) {
165                                 alloc_lte_dentries *= 2;
166                                 lte_dentries = REALLOC(prev_lte_dentries,
167                                                        alloc_lte_dentries *
168                                                         sizeof(lte_dentries[0]));
169                                 if (!lte_dentries)
170                                         return WIMLIB_ERR_NOMEM;
171                                 if (prev_lte_dentries == NULL) {
172                                         memcpy(lte_dentries,
173                                                lte->inline_lte_dentries,
174                                                sizeof(lte->inline_lte_dentries));
175                                 }
176                                 lte->lte_dentries = lte_dentries;
177                                 lte->alloc_lte_dentries = alloc_lte_dentries;
178                         }
179                         lte_dentries = lte->lte_dentries;
180                 }
181                 lte_dentries[lte->out_refcnt] = dentry;
182         }
183         lte->out_refcnt++;
184         return 0;
185 }
186
187 /* Given a WIM dentry in the tree to be extracted, iterate through streams that
188  * need to be extracted.  For each one, add it to the list of streams to be
189  * extracted (ctx->stream_list) if not already done so, and also update the
190  * progress information (ctx->progress) with the stream.  Furthermore, if doing
191  * a sequential extraction, build a mapping from each the stream to the dentries
192  * referencing it.
193  *
194  * This uses the i_visited member of the inodes (assumed to be 0 initially).  */
195 static int
196 dentry_add_streams_to_extract(struct wim_dentry *dentry, void *_ctx)
197 {
198         struct apply_ctx *ctx = _ctx;
199         struct wim_inode *inode = dentry->d_inode;
200         int ret;
201
202         /* Don't process dentries marked as skipped.  */
203         if (dentry->extraction_skipped)
204                 return 0;
205
206         /* The unnamed data stream will always be extracted, except in an
207          * unlikely case.  */
208         if (!inode_is_encrypted_directory(inode)) {
209                 ret = ref_stream_to_extract(inode_unnamed_lte_resolved(inode),
210                                             dentry, ctx);
211                 if (ret)
212                         return ret;
213         }
214
215         /* Named data streams will be extracted only if supported in the current
216          * extraction mode and volume, and to avoid complications, if not doing
217          * a linked extraction.  */
218         if (can_extract_named_data_streams(ctx)) {
219                 for (u16 i = 0; i < inode->i_num_ads; i++) {
220                         if (!ads_entry_is_named_stream(&inode->i_ads_entries[i]))
221                                 continue;
222                         ret = ref_stream_to_extract(inode->i_ads_entries[i].lte,
223                                                     dentry, ctx);
224                         if (ret)
225                                 return ret;
226                 }
227         }
228         inode->i_visited = 1;
229         return 0;
230 }
231
232 /* Inform library user of progress of stream extraction following the successful
233  * extraction of a copy of the stream specified by @lte.  */
234 static void
235 update_extract_progress(struct apply_ctx *ctx,
236                         const struct wim_lookup_table_entry *lte)
237 {
238         wimlib_progress_func_t progress_func = ctx->progress_func;
239         union wimlib_progress_info *progress = &ctx->progress;
240
241         progress->extract.completed_bytes += wim_resource_size(lte);
242         if (progress_func &&
243             progress->extract.completed_bytes >= ctx->next_progress)
244         {
245                 progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS, progress);
246                 if (progress->extract.completed_bytes >=
247                     progress->extract.total_bytes)
248                 {
249                         ctx->next_progress = ~0ULL;
250                 } else {
251                         ctx->next_progress += progress->extract.total_bytes / 128;
252                         if (ctx->next_progress > progress->extract.total_bytes)
253                                 ctx->next_progress = progress->extract.total_bytes;
254                 }
255         }
256 }
257
258 #ifndef __WIN32__
259 /* Extract a symbolic link (not directly as reparse data), handling fixing up
260  * the target of absolute symbolic links and updating the extract progress.
261  *
262  * @inode must specify the WIM inode for a symbolic link or junction reparse
263  * point.
264  *
265  * @lte_override overrides the resource used as the reparse data for the
266  * symbolic link.  */
267 static int
268 extract_symlink(const tchar *path, struct apply_ctx *ctx,
269                 struct wim_inode *inode,
270                 struct wim_lookup_table_entry *lte_override)
271 {
272         ssize_t bufsize = ctx->ops->path_max;
273         tchar target[bufsize];
274         tchar *buf = target;
275         tchar *fixed_target;
276         ssize_t sret;
277         int ret;
278
279         /* If absolute symbolic link fixups requested, reserve space in the link
280          * target buffer for the absolute path of the target directory.  */
281         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX)
282         {
283                 buf += ctx->realtarget_nchars;
284                 bufsize -= ctx->realtarget_nchars;
285         }
286
287         /* Translate the WIM inode's reparse data into the link target.  */
288         sret = wim_inode_readlink(inode, buf, bufsize - 1, lte_override);
289         if (sret < 0) {
290                 errno = -sret;
291                 return WIMLIB_ERR_READLINK;
292         }
293         buf[sret] = '\0';
294
295         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
296             buf[0] == '/')
297         {
298                 /* Fix absolute symbolic link target to point into the
299                  * actual extraction destination.  */
300                 tmemcpy(target, ctx->realtarget, ctx->realtarget_nchars);
301                 fixed_target = target;
302         } else {
303                 /* Keep same link target.  */
304                 fixed_target = buf;
305         }
306
307         /* Call into the apply_operations to create the symbolic link.  */
308         DEBUG("Creating symlink \"%"TS"\" => \"%"TS"\"",
309               path, fixed_target);
310         ret = ctx->ops->create_symlink(fixed_target, path, ctx);
311         if (ret) {
312                 ERROR_WITH_ERRNO("Failed to create symlink "
313                                  "\"%"TS"\" => \"%"TS"\"", path, fixed_target);
314                 return ret;
315         }
316
317         /* Account for reparse data consumed.  */
318         update_extract_progress(ctx,
319                                 (lte_override ? lte_override :
320                                       inode_unnamed_lte_resolved(inode)));
321         return 0;
322 }
323 #endif /* !__WIN32__ */
324
325 /* Create a file, directory, or symbolic link.  */
326 static int
327 extract_inode(const tchar *path, struct apply_ctx *ctx, struct wim_inode *inode)
328 {
329         int ret;
330
331 #ifndef __WIN32__
332         if (ctx->supported_features.symlink_reparse_points &&
333             !ctx->supported_features.reparse_points &&
334             inode_is_symlink(inode))
335         {
336                 ret = extract_symlink(path, ctx, inode, NULL);
337         } else
338 #endif /* !__WIN32__ */
339         if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) {
340                 ret = ctx->ops->create_directory(path, ctx, &inode->extract_cookie);
341                 if (ret) {
342                         ERROR_WITH_ERRNO("Failed to create the directory "
343                                          "\"%"TS"\"", path);
344                 }
345         } else if ((inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) &&
346                     ctx->ops->extract_encrypted_stream_creates_file &&
347                     ctx->supported_features.encrypted_files) {
348                 ret = ctx->ops->extract_encrypted_stream(
349                                 path, inode_unnamed_lte_resolved(inode), ctx);
350                 if (ret) {
351                         ERROR_WITH_ERRNO("Failed to create and extract "
352                                          "encrypted file \"%"TS"\"", path);
353                 }
354         } else {
355                 ret = ctx->ops->create_file(path, ctx, &inode->extract_cookie);
356                 if (ret) {
357                         ERROR_WITH_ERRNO("Failed to create the file "
358                                          "\"%"TS"\"", path);
359                 }
360         }
361         return ret;
362 }
363
364 static int
365 extract_hardlink(const tchar *oldpath, const tchar *newpath,
366                  struct apply_ctx *ctx)
367 {
368         int ret;
369
370         DEBUG("Creating hardlink \"%"TS"\" => \"%"TS"\"", newpath, oldpath);
371         ret = ctx->ops->create_hardlink(oldpath, newpath, ctx);
372         if (ret) {
373                 ERROR_WITH_ERRNO("Failed to create hardlink "
374                                  "\"%"TS"\" => \"%"TS"\"",
375                                  newpath, oldpath);
376         }
377         return ret;
378 }
379
380 #ifdef __WIN32__
381 static int
382 try_extract_rpfix(u8 *rpbuf,
383                   u16 *rpbuflen_p,
384                   const wchar_t *extract_root_realpath,
385                   unsigned extract_root_realpath_nchars)
386 {
387         struct reparse_data rpdata;
388         wchar_t *target;
389         size_t target_nchars;
390         size_t stripped_nchars;
391         wchar_t *stripped_target;
392         wchar_t stripped_target_nchars;
393         int ret;
394
395         utf16lechar *new_target;
396         utf16lechar *new_print_name;
397         size_t new_target_nchars;
398         size_t new_print_name_nchars;
399         utf16lechar *p;
400
401         ret = parse_reparse_data(rpbuf, *rpbuflen_p, &rpdata);
402         if (ret)
403                 return ret;
404
405         if (extract_root_realpath[0] == L'\0' ||
406             extract_root_realpath[1] != L':' ||
407             extract_root_realpath[2] != L'\\')
408                 return WIMLIB_ERR_REPARSE_POINT_FIXUP_FAILED;
409
410         ret = parse_substitute_name(rpdata.substitute_name,
411                                     rpdata.substitute_name_nbytes,
412                                     rpdata.rptag);
413         if (ret < 0)
414                 return 0;
415         stripped_nchars = ret;
416         target = rpdata.substitute_name;
417         target_nchars = rpdata.substitute_name_nbytes / sizeof(utf16lechar);
418         stripped_target = target + stripped_nchars;
419         stripped_target_nchars = target_nchars - stripped_nchars;
420
421         new_target = alloca((6 + extract_root_realpath_nchars +
422                              stripped_target_nchars) * sizeof(utf16lechar));
423
424         p = new_target;
425         if (stripped_nchars == 6) {
426                 /* Include \??\ prefix if it was present before */
427                 p = wmempcpy(p, L"\\??\\", 4);
428         }
429
430         /* Print name excludes the \??\ if present. */
431         new_print_name = p;
432         if (stripped_nchars != 0) {
433                 /* Get drive letter from real path to extract root, if a drive
434                  * letter was present before. */
435                 *p++ = extract_root_realpath[0];
436                 *p++ = extract_root_realpath[1];
437         }
438         /* Copy the rest of the extract root */
439         p = wmempcpy(p, extract_root_realpath + 2, extract_root_realpath_nchars - 2);
440
441         /* Append the stripped target */
442         p = wmempcpy(p, stripped_target, stripped_target_nchars);
443         new_target_nchars = p - new_target;
444         new_print_name_nchars = p - new_print_name;
445
446         if (new_target_nchars * sizeof(utf16lechar) >= REPARSE_POINT_MAX_SIZE ||
447             new_print_name_nchars * sizeof(utf16lechar) >= REPARSE_POINT_MAX_SIZE)
448                 return WIMLIB_ERR_REPARSE_POINT_FIXUP_FAILED;
449
450         rpdata.substitute_name = new_target;
451         rpdata.substitute_name_nbytes = new_target_nchars * sizeof(utf16lechar);
452         rpdata.print_name = new_print_name;
453         rpdata.print_name_nbytes = new_print_name_nchars * sizeof(utf16lechar);
454         return make_reparse_buffer(&rpdata, rpbuf, rpbuflen_p);
455 }
456 #endif /* __WIN32__ */
457
458 /* Set reparse data on extracted file or directory that has
459  * FILE_ATTRIBUTE_REPARSE_POINT set.  */
460 static int
461 extract_reparse_data(const tchar *path, struct apply_ctx *ctx,
462                      struct wim_inode *inode,
463                      struct wim_lookup_table_entry *lte_override)
464 {
465         int ret;
466         u8 rpbuf[REPARSE_POINT_MAX_SIZE];
467         u16 rpbuflen;
468
469         ret = wim_inode_get_reparse_data(inode, rpbuf, &rpbuflen, lte_override);
470         if (ret)
471                 goto error;
472
473 #ifdef __WIN32__
474         /* Fix up target of absolute symbolic link or junction points so
475          * that they point into the actual extraction target.  */
476         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
477             (inode->i_reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK ||
478              inode->i_reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT) &&
479             !inode->i_not_rpfixed)
480         {
481                 ret = try_extract_rpfix(rpbuf, &rpbuflen, ctx->realtarget,
482                                         ctx->realtarget_nchars);
483                 if (ret && !(ctx->extract_flags &
484                              WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS))
485                 {
486                         WARNING("Reparse point fixup of \"%"TS"\" "
487                                 "failed", path);
488                         ret = 0;
489                 }
490                 if (ret)
491                         goto error;
492         }
493 #endif
494
495         ret = ctx->ops->set_reparse_data(path, rpbuf, rpbuflen, ctx);
496
497         /* On Windows, the SeCreateSymbolicLink privilege is required to create
498          * symbolic links.  To be more friendly towards non-Administrator users,
499          * we merely warn the user if symbolic links cannot be created due to
500          * insufficient permissions or privileges, unless
501          * WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS was provided.  */
502 #ifdef __WIN32__
503         if (ret && inode_is_symlink(inode) &&
504             (errno == EACCES || errno == EPERM) &&
505             !(ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS))
506         {
507                 WARNING("Can't set reparse data on \"%"TS"\": "
508                         "Access denied!\n"
509                         "          You may be trying to "
510                         "extract a symbolic link without the\n"
511                         "          SeCreateSymbolicLink privilege, "
512                         "which by default non-Administrator\n"
513                         "          accounts do not have.",
514                         path);
515                 ret = 0;
516         }
517 #endif
518         if (ret)
519                 goto error;
520
521         /* Account for reparse data consumed.  */
522         update_extract_progress(ctx,
523                                 (lte_override ? lte_override :
524                                       inode_unnamed_lte_resolved(inode)));
525         return 0;
526
527 error:
528         ERROR_WITH_ERRNO("Failed to set reparse data on \"%"TS"\"", path);
529         return ret;
530 }
531
532 /*
533  * Extract zero or more streams to a file.
534  *
535  * This function operates slightly differently depending on whether @lte_spec is
536  * NULL or not.  When @lte_spec is NULL, the behavior is to extract the default
537  * file contents (unnamed stream), and, if named data streams are supported in
538  * the extract mode and volume, any named data streams.  When @lte_spec is NULL,
539  * the behavior is to extract only all copies of the stream @lte_spec, and in
540  * addition use @lte_spec to set the reparse data or create the symbolic link if
541  * appropriate.
542  *
543  * @path
544  *      Path to file to extract (as can be passed to apply_operations
545  *      functions).
546  * @ctx
547  *      Apply context.
548  * @dentry
549  *      WIM dentry that corresponds to the file being extracted.
550  * @lte_spec
551  *      If non-NULL, specifies the lookup table entry for a stream to extract,
552  *      and only that stream will be extracted (although there may be more than
553  *      one instance of it).
554  * @lte_override
555  *      Used only if @lte_spec != NULL; it is passed to the extraction functions
556  *      rather than @lte_spec, allowing the location of the stream to be
557  *      overridden.  (This is used when the WIM is being read from a nonseekable
558  *      file, such as a pipe, when streams need to be used more than once; each
559  *      such stream is extracted to a temporary file.)
560  */
561 static int
562 extract_streams(const tchar *path, struct apply_ctx *ctx,
563                 struct wim_dentry *dentry,
564                 struct wim_lookup_table_entry *lte_spec,
565                 struct wim_lookup_table_entry *lte_override)
566 {
567         struct wim_inode *inode = dentry->d_inode;
568         struct wim_lookup_table_entry *lte;
569         file_spec_t file_spec;
570         int ret;
571
572         if (dentry->was_hardlinked)
573                 return 0;
574
575 #ifdef ENABLE_DEBUG
576         if (lte_spec) {
577                 char sha1_str[100];
578                 char *p = sha1_str;
579                 for (unsigned i = 0; i < SHA1_HASH_SIZE; i++)
580                         p += sprintf(p, "%02x", lte_override->hash[i]);
581                 DEBUG("Extracting stream SHA1=%s to \"%"TS"\"",
582                       sha1_str, path, inode->i_ino);
583         } else {
584                 DEBUG("Extracting streams to \"%"TS"\"", path, inode->i_ino);
585         }
586 #endif
587
588         if (ctx->ops->uses_cookies)
589                 file_spec.cookie = inode->extract_cookie;
590         else
591                 file_spec.path = path;
592
593         /* Unnamed data stream.  */
594         lte = inode_unnamed_lte_resolved(inode);
595         if (lte && (!lte_spec || lte == lte_spec)) {
596                 if (lte_spec)
597                         lte = lte_override;
598                 if (!(inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY |
599                                              FILE_ATTRIBUTE_REPARSE_POINT)))
600                 {
601                         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED &&
602                             ctx->supported_features.encrypted_files) {
603                                 if (!ctx->ops->extract_encrypted_stream_creates_file) {
604                                         ret = ctx->ops->extract_encrypted_stream(
605                                                                 path, lte, ctx);
606                                         if (ret)
607                                                 goto error;
608                                 }
609                         } else {
610                                 ret = ctx->ops->extract_unnamed_stream(
611                                                         file_spec, lte, ctx);
612                                 if (ret)
613                                         goto error;
614                         }
615                         update_extract_progress(ctx, lte);
616                 }
617                 else if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)
618                 {
619                         ret = 0;
620                         if (ctx->supported_features.reparse_points)
621                                 ret = extract_reparse_data(path, ctx, inode, lte);
622                 #ifndef __WIN32__
623                         else if ((inode_is_symlink(inode) &&
624                                   ctx->supported_features.symlink_reparse_points))
625                                 ret = extract_symlink(path, ctx, inode, lte);
626                 #endif
627                         if (ret)
628                                 return ret;
629                 }
630         }
631
632         /* Named data streams.  */
633         if (can_extract_named_data_streams(ctx)) {
634                 for (u16 i = 0; i < inode->i_num_ads; i++) {
635                         struct wim_ads_entry *entry = &inode->i_ads_entries[i];
636
637                         if (!ads_entry_is_named_stream(entry))
638                                 continue;
639                         lte = entry->lte;
640                         if (!lte)
641                                 continue;
642                         if (lte_spec && lte_spec != lte)
643                                 continue;
644                         if (lte_spec)
645                                 lte = lte_override;
646                         ret = ctx->ops->extract_named_stream(file_spec, entry->stream_name,
647                                                              entry->stream_name_nbytes / 2,
648                                                              lte, ctx);
649                         if (ret)
650                                 goto error;
651                         update_extract_progress(ctx, lte);
652                 }
653         }
654         return 0;
655
656 error:
657         ERROR_WITH_ERRNO("Failed to extract data of \"%"TS"\"", path);
658         return ret;
659 }
660
661 /* Set attributes on an extracted file or directory if supported by the
662  * extraction mode.  */
663 static int
664 extract_file_attributes(const tchar *path, struct apply_ctx *ctx,
665                         struct wim_dentry *dentry, unsigned pass)
666 {
667         int ret;
668
669         if (ctx->ops->set_file_attributes &&
670             !(dentry == ctx->extract_root && ctx->root_dentry_is_special)) {
671                 u32 attributes = dentry->d_inode->i_attributes;
672
673                 /* Clear unsupported attributes.  */
674                 attributes &= ctx->supported_attributes_mask;
675
676                 if ((attributes & FILE_ATTRIBUTE_DIRECTORY &&
677                      !ctx->supported_features.encrypted_directories) ||
678                     (!(attributes & FILE_ATTRIBUTE_DIRECTORY) &&
679                      !ctx->supported_features.encrypted_files))
680                 {
681                         attributes &= ~FILE_ATTRIBUTE_ENCRYPTED;
682                 }
683
684                 if (attributes == 0)
685                         attributes = FILE_ATTRIBUTE_NORMAL;
686
687                 ret = ctx->ops->set_file_attributes(path, attributes, ctx, pass);
688                 if (ret) {
689                         ERROR_WITH_ERRNO("Failed to set attributes on "
690                                          "\"%"TS"\"", path);
691                         return ret;
692                 }
693         }
694         return 0;
695 }
696
697
698 /* Set or remove the short (DOS) name on an extracted file or directory if
699  * supported by the extraction mode.  Since DOS names are unimportant and it's
700  * easy to run into problems setting them on Windows (SetFileShortName()
701  * requires SE_RESTORE privilege, which only the Administrator can request, and
702  * also requires DELETE access to the file), failure is ignored unless
703  * WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES is set.  */
704 static int
705 extract_short_name(const tchar *path, struct apply_ctx *ctx,
706                    struct wim_dentry *dentry)
707 {
708         int ret;
709
710         /* The root of the dentry tree being extracted may not be extracted to
711          * its original name, so its short name should be ignored.  */
712         if (dentry == ctx->extract_root)
713                 return 0;
714
715         if (ctx->supported_features.short_names) {
716                 ret = ctx->ops->set_short_name(path,
717                                                dentry->short_name,
718                                                dentry->short_name_nbytes / 2,
719                                                ctx);
720                 if (ret && (ctx->extract_flags &
721                             WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES))
722                 {
723                         ERROR_WITH_ERRNO("Failed to set short name of "
724                                          "\"%"TS"\"", path);
725                         return ret;
726                 }
727         }
728         return 0;
729 }
730
731 /* Set security descriptor, UNIX data, or neither on an extracted file, taking
732  * into account the current extraction mode and flags.  */
733 static int
734 extract_security(const tchar *path, struct apply_ctx *ctx,
735                  struct wim_dentry *dentry)
736 {
737         int ret;
738         struct wim_inode *inode = dentry->d_inode;
739
740         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_NO_ACLS)
741                 return 0;
742
743         if ((ctx->extract_root == dentry) && ctx->root_dentry_is_special)
744                 return 0;
745
746 #ifndef __WIN32__
747         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) {
748                 struct wimlib_unix_data data;
749
750                 ret = inode_get_unix_data(inode, &data, NULL);
751                 if (ret < 0)
752                         ret = 0;
753                 else if (ret == 0)
754                         ret = ctx->ops->set_unix_data(path, &data, ctx);
755                 if (ret) {
756                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_ACLS) {
757                                 ERROR_WITH_ERRNO("Failed to set UNIX owner, "
758                                                  "group, and/or mode on "
759                                                  "\"%"TS"\"", path);
760                                 return ret;
761                         } else {
762                                 WARNING_WITH_ERRNO("Failed to set UNIX owner, "
763                                                    "group, and/or/mode on "
764                                                    "\"%"TS"\"", path);
765                         }
766                 }
767         }
768         else
769 #endif /* __WIN32__ */
770         if (ctx->supported_features.security_descriptors &&
771             inode->i_security_id != -1)
772         {
773                 const struct wim_security_data *sd;
774                 const u8 *desc;
775                 size_t desc_size;
776
777                 sd = wim_const_security_data(ctx->wim);
778                 desc = sd->descriptors[inode->i_security_id];
779                 desc_size = sd->sizes[inode->i_security_id];
780
781                 ret = ctx->ops->set_security_descriptor(path, desc,
782                                                         desc_size, ctx);
783                 if (ret) {
784                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_ACLS) {
785                                 ERROR_WITH_ERRNO("Failed to set security "
786                                                  "descriptor on \"%"TS"\"", path);
787                                 return ret;
788                         } else {
789                         #if 0
790                                 if (errno != EACCES) {
791                                         WARNING_WITH_ERRNO("Failed to set "
792                                                            "security descriptor "
793                                                            "on \"%"TS"\"", path);
794                                 }
795                         #endif
796                                 ctx->no_security_descriptors++;
797                         }
798                 }
799         }
800         return 0;
801 }
802
803 /* Set timestamps on an extracted file.  Failure is warning-only unless
804  * WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS is set.  */
805 static int
806 extract_timestamps(const tchar *path, struct apply_ctx *ctx,
807                    struct wim_dentry *dentry)
808 {
809         struct wim_inode *inode = dentry->d_inode;
810         int ret;
811
812         if ((ctx->extract_root == dentry) && ctx->root_dentry_is_special)
813                 return 0;
814
815         if (ctx->ops->set_timestamps) {
816                 ret = ctx->ops->set_timestamps(path,
817                                                inode->i_creation_time,
818                                                inode->i_last_write_time,
819                                                inode->i_last_access_time,
820                                                ctx);
821                 if (ret) {
822                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS) {
823                                 ERROR_WITH_ERRNO("Failed to set timestamps "
824                                                  "on \"%"TS"\"", path);
825                                 return ret;
826                         } else {
827                                 WARNING_WITH_ERRNO("Failed to set timestamps "
828                                                    "on \"%"TS"\"", path);
829                         }
830                 }
831         }
832         return 0;
833 }
834
835 /* Check whether the extraction of a dentry should be skipped completely.  */
836 static bool
837 dentry_is_supported(struct wim_dentry *dentry,
838                     const struct wim_features *supported_features)
839 {
840         struct wim_inode *inode = dentry->d_inode;
841
842         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
843                 return supported_features->reparse_points ||
844                         (inode_is_symlink(inode) &&
845                          supported_features->symlink_reparse_points);
846         }
847         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
848                 if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
849                         return supported_features->encrypted_directories != 0;
850                 else
851                         return supported_features->encrypted_files != 0;
852         }
853         return true;
854 }
855
856 /* Given a WIM dentry to extract, build the path to which to extract it, in the
857  * format understood by the callbacks in the apply_operations being used.
858  *
859  * Write the resulting path into @path, which must have room for at least
860  * ctx->ops->max_path characters including the null-terminator.
861  *
862  * Return %true if successful; %false if this WIM dentry doesn't actually need
863  * to be extracted or if the calculated path exceeds ctx->ops->max_path
864  * characters.
865  *
866  * This function clobbers the tmp_list member of @dentry and its ancestors up
867  * until the extraction root.  */
868 static bool
869 build_extraction_path(tchar path[], struct wim_dentry *dentry,
870                       struct apply_ctx *ctx)
871 {
872         size_t path_nchars;
873         LIST_HEAD(ancestor_list);
874         tchar *p = path;
875         const tchar *target_prefix;
876         size_t target_prefix_nchars;
877         struct wim_dentry *d;
878
879         if (dentry->extraction_skipped)
880                 return false;
881
882         path_nchars = ctx->ops->path_prefix_nchars;
883
884         if (ctx->ops->requires_realtarget_in_paths) {
885                 target_prefix        = ctx->realtarget;
886                 target_prefix_nchars = ctx->realtarget_nchars;
887         } else if (ctx->ops->requires_target_in_paths) {
888                 target_prefix        = ctx->target;
889                 target_prefix_nchars = ctx->target_nchars;
890         } else {
891                 target_prefix        = NULL;
892                 target_prefix_nchars = 0;
893         }
894         path_nchars += target_prefix_nchars;
895
896         for (d = dentry; d != ctx->extract_root; d = d->parent) {
897                 path_nchars += d->extraction_name_nchars + 1;
898                 list_add(&d->tmp_list, &ancestor_list);
899         }
900
901         path_nchars++; /* null terminator */
902
903         if (path_nchars > ctx->ops->path_max) {
904                 WARNING("\"%"TS"\": Path too long to extract",
905                         dentry_full_path(dentry));
906                 return false;
907         }
908
909         p = tmempcpy(p, ctx->ops->path_prefix, ctx->ops->path_prefix_nchars);
910         p = tmempcpy(p, target_prefix, target_prefix_nchars);
911         list_for_each_entry(d, &ancestor_list, tmp_list) {
912                 *p++ = ctx->ops->path_separator;
913                 p = tmempcpy(p, d->extraction_name, d->extraction_name_nchars);
914         }
915         *p++ = T('\0');
916         wimlib_assert(p - path == path_nchars);
917         return true;
918 }
919
920 static unsigned
921 get_num_path_components(const tchar *path, tchar path_separator)
922 {
923         unsigned num_components = 0;
924 #ifdef __WIN32__
925         /* Ignore drive letter.  */
926         if (path[0] != L'\0' && path[1] == L':')
927                 path += 2;
928 #endif
929
930         while (*path) {
931                 while (*path == path_separator)
932                         path++;
933                 if (*path)
934                         num_components++;
935                 while (*path && *path != path_separator)
936                         path++;
937         }
938         return num_components;
939 }
940
941 static int
942 extract_multiimage_symlink(const tchar *oldpath, const tchar *newpath,
943                            struct apply_ctx *ctx, struct wim_dentry *dentry)
944 {
945         size_t num_raw_path_components;
946         const struct wim_dentry *d;
947         size_t num_target_path_components;
948         tchar *p;
949         const tchar *p_old;
950         int ret;
951
952         num_raw_path_components = 0;
953         for (d = dentry; d != ctx->extract_root; d = d->parent)
954                 num_raw_path_components++;
955
956         if (ctx->ops->requires_realtarget_in_paths)
957                 num_target_path_components = get_num_path_components(ctx->realtarget,
958                                                                      ctx->ops->path_separator);
959         else if (ctx->ops->requires_target_in_paths)
960                 num_target_path_components = get_num_path_components(ctx->target,
961                                                                      ctx->ops->path_separator);
962         else
963                 num_target_path_components = 0;
964
965         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_MULTI_IMAGE) {
966                 wimlib_assert(num_target_path_components > 0);
967                 num_raw_path_components++;
968                 num_target_path_components--;
969         }
970
971         p_old = oldpath + ctx->ops->path_prefix_nchars;
972 #ifdef __WIN32__
973         if (p_old[0] != L'\0' && p_old[1] == ':')
974                 p_old += 2;
975 #endif
976         while (*p_old == ctx->ops->path_separator)
977                 p_old++;
978         while (--num_target_path_components) {
979                 while (*p_old != ctx->ops->path_separator)
980                         p_old++;
981                 while (*p_old == ctx->ops->path_separator)
982                         p_old++;
983         }
984
985         tchar symlink_target[tstrlen(p_old) + 3 * num_raw_path_components + 1];
986
987         p = &symlink_target[0];
988         while (num_raw_path_components--) {
989                 *p++ = '.';
990                 *p++ = '.';
991                 *p++ = ctx->ops->path_separator;
992         }
993         tstrcpy(p, p_old);
994         DEBUG("Creating symlink \"%"TS"\" => \"%"TS"\"",
995               newpath, symlink_target);
996         ret = ctx->ops->create_symlink(symlink_target, newpath, ctx);
997         if (ret) {
998                 ERROR_WITH_ERRNO("Failed to create symlink "
999                                  "\"%"TS"\" => \"%"TS"\"",
1000                                  newpath, symlink_target);
1001         }
1002         return ret;
1003 }
1004
1005 /* Create the "skeleton" of an extracted file or directory.  Don't yet extract
1006  * data streams, reparse data (including symbolic links), timestamps, and
1007  * security descriptors.  Basically, everything that doesn't require reading
1008  * non-metadata resources from the WIM file and isn't delayed until the final
1009  * pass.  */
1010 static int
1011 do_dentry_extract_skeleton(tchar path[], struct wim_dentry *dentry,
1012                            struct apply_ctx *ctx)
1013 {
1014         struct wim_inode *inode = dentry->d_inode;
1015         int ret;
1016         const tchar *oldpath;
1017
1018         if (unlikely(is_linked_extraction(ctx))) {
1019                 struct wim_lookup_table_entry *unnamed_lte;
1020
1021                 unnamed_lte = inode_unnamed_lte_resolved(dentry->d_inode);
1022                 if (unnamed_lte && unnamed_lte->extracted_file) {
1023                         oldpath = unnamed_lte->extracted_file;
1024                         if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_HARDLINK)
1025                                 goto hardlink;
1026                         else
1027                                 goto symlink;
1028                 }
1029         }
1030
1031         /* Create hard link if this dentry corresponds to an already-extracted
1032          * inode.  */
1033         if (inode->i_extracted_file) {
1034                 oldpath = inode->i_extracted_file;
1035                 goto hardlink;
1036         }
1037
1038         /* Skip symlinks unless they can be extracted as reparse points rather
1039          * than created directly.  */
1040         if (inode_is_symlink(inode) && !ctx->supported_features.reparse_points)
1041                 return 0;
1042
1043         /* Create this file or directory unless it's the extraction root, which
1044          * was already created if necessary.  */
1045         if (dentry != ctx->extract_root) {
1046                 ret = extract_inode(path, ctx, inode);
1047                 if (ret)
1048                         return ret;
1049         }
1050
1051         /* Create empty named data streams.  */
1052         if (can_extract_named_data_streams(ctx)) {
1053                 for (u16 i = 0; i < inode->i_num_ads; i++) {
1054                         file_spec_t file_spec;
1055                         struct wim_ads_entry *entry = &inode->i_ads_entries[i];
1056
1057                         if (!ads_entry_is_named_stream(entry))
1058                                 continue;
1059                         if (entry->lte)
1060                                 continue;
1061                         if (ctx->ops->uses_cookies)
1062                                 file_spec.cookie = inode->extract_cookie;
1063                         else
1064                                 file_spec.path = path;
1065                         ret = ctx->ops->extract_named_stream(file_spec,
1066                                                              entry->stream_name,
1067                                                              entry->stream_name_nbytes / 2,
1068                                                              entry->lte, ctx);
1069                         if (ret) {
1070                                 ERROR_WITH_ERRNO("\"%"TS"\": failed to create "
1071                                                  "empty named data stream",
1072                                                  path);
1073                                 return ret;
1074                         }
1075                 }
1076         }
1077
1078         /* Set file attributes (if supported).  */
1079         ret = extract_file_attributes(path, ctx, dentry, 0);
1080         if (ret)
1081                 return ret;
1082
1083         /* Set or remove file short name (if supported).  */
1084         ret = extract_short_name(path, ctx, dentry);
1085         if (ret)
1086                 return ret;
1087
1088         /* If inode has multiple links and hard links are supported in this
1089          * extraction mode and volume, save the path to the extracted file in
1090          * case it's needed to create a hard link.  */
1091         if (unlikely(is_linked_extraction(ctx))) {
1092                 struct wim_lookup_table_entry *unnamed_lte;
1093
1094                 unnamed_lte = inode_unnamed_lte_resolved(dentry->d_inode);
1095                 if (unnamed_lte) {
1096                         unnamed_lte->extracted_file = TSTRDUP(path);
1097                         if (!unnamed_lte->extracted_file)
1098                                 return WIMLIB_ERR_NOMEM;
1099                 }
1100         } else if (inode->i_nlink > 1 && ctx->supported_features.hard_links) {
1101                 inode->i_extracted_file = TSTRDUP(path);
1102                 if (!inode->i_extracted_file)
1103                         return WIMLIB_ERR_NOMEM;
1104         }
1105         return 0;
1106
1107 symlink:
1108         ret = extract_multiimage_symlink(oldpath, path, ctx, dentry);
1109         if (ret)
1110                 return ret;
1111         dentry->was_hardlinked = 1;
1112         return 0;
1113
1114 hardlink:
1115         ret = extract_hardlink(oldpath, path, ctx);
1116         if (ret)
1117                 return ret;
1118         dentry->was_hardlinked = 1;
1119         return 0;
1120 }
1121
1122 /* This is a wrapper around do_dentry_extract_skeleton() that handles building
1123  * the path, doing short name reordering.  This is also idempotent; dentries
1124  * already processed have skeleton_extracted set and no action is taken.  See
1125  * apply_operations.requires_short_name_reordering for more details about short
1126  * name reordering.  */
1127 static int
1128 dentry_extract_skeleton(struct wim_dentry *dentry, void *_ctx)
1129 {
1130         struct apply_ctx *ctx = _ctx;
1131         tchar path[ctx->ops->path_max];
1132         struct wim_dentry *orig_dentry;
1133         struct wim_dentry *other_dentry;
1134         int ret;
1135
1136         if (dentry->skeleton_extracted)
1137                 return 0;
1138
1139         orig_dentry = NULL;
1140         if (ctx->supported_features.short_names
1141             && ctx->ops->requires_short_name_reordering
1142             && !dentry_has_short_name(dentry)
1143             && !dentry->d_inode->i_dos_name_extracted)
1144         {
1145                 inode_for_each_dentry(other_dentry, dentry->d_inode) {
1146                         if (dentry_has_short_name(other_dentry)
1147                             && !other_dentry->skeleton_extracted)
1148                         {
1149                                 DEBUG("Creating %"TS" before %"TS" "
1150                                       "to guarantee correct DOS name extraction",
1151                                       dentry_full_path(other_dentry),
1152                                       dentry_full_path(dentry));
1153                                 orig_dentry = dentry;
1154                                 dentry = other_dentry;
1155                                 break;
1156                         }
1157                 }
1158         }
1159 again:
1160         if (!build_extraction_path(path, dentry, ctx))
1161                 return 0;
1162         ret = do_dentry_extract_skeleton(path, dentry, ctx);
1163         if (ret)
1164                 return ret;
1165
1166         dentry->skeleton_extracted = 1;
1167
1168         if (orig_dentry) {
1169                 dentry = orig_dentry;
1170                 orig_dentry = NULL;
1171                 goto again;
1172         }
1173         dentry->d_inode->i_dos_name_extracted = 1;
1174         return 0;
1175 }
1176
1177 static int
1178 dentry_extract_dir_skeleton(struct wim_dentry *dentry, void *_ctx)
1179 {
1180         if (dentry->d_inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
1181                 return dentry_extract_skeleton(dentry, _ctx);
1182         return 0;
1183 }
1184
1185 /* Create a file or directory, then immediately extract all streams.  This
1186  * assumes that WIMLIB_EXTRACT_FLAG_SEQUENTIAL is not specified, since the WIM
1187  * may not be read sequentially by this function.  */
1188 static int
1189 dentry_extract(struct wim_dentry *dentry, void *_ctx)
1190 {
1191         struct apply_ctx *ctx = _ctx;
1192         tchar path[ctx->ops->path_max];
1193         int ret;
1194
1195         ret = dentry_extract_skeleton(dentry, ctx);
1196         if (ret)
1197                 return ret;
1198
1199         if (!build_extraction_path(path, dentry, ctx))
1200                 return 0;
1201
1202         return extract_streams(path, ctx, dentry, NULL, NULL);
1203 }
1204
1205 /* Extract all instances of the stream @lte that are being extracted in this
1206  * call of extract_tree().  @can_seek specifies whether the WIM file descriptor
1207  * is seekable or not (e.g. is a pipe).  If not and the stream needs to be
1208  * extracted multiple times, it is extracted to a temporary file first.
1209  *
1210  * This is intended for use with sequential extraction of a WIM image
1211  * (WIMLIB_EXTRACT_FLAG_SEQUENTIAL specified).  */
1212 static int
1213 extract_stream_instances(struct wim_lookup_table_entry *lte,
1214                          struct apply_ctx *ctx, bool can_seek)
1215 {
1216         struct wim_dentry **lte_dentries;
1217         struct wim_lookup_table_entry *lte_tmp = NULL;
1218         struct wim_lookup_table_entry *lte_override;
1219         tchar *stream_tmp_filename = NULL;
1220         tchar path[ctx->ops->path_max];
1221         unsigned i;
1222         int ret;
1223
1224         if (lte->out_refcnt <= ARRAY_LEN(lte->inline_lte_dentries))
1225                 lte_dentries = lte->inline_lte_dentries;
1226         else
1227                 lte_dentries = lte->lte_dentries;
1228
1229         if (likely(can_seek || lte->out_refcnt < 2)) {
1230                 lte_override = lte;
1231         } else {
1232                 /* Need to extract stream to temporary file.  */
1233                 struct filedes fd;
1234                 int raw_fd;
1235
1236                 stream_tmp_filename = ttempnam(NULL, T("wimlib"));
1237                 if (!stream_tmp_filename) {
1238                         ERROR_WITH_ERRNO("Failed to create temporary filename");
1239                         ret = WIMLIB_ERR_OPEN;
1240                         goto out;
1241                 }
1242
1243                 lte_tmp = memdup(lte, sizeof(struct wim_lookup_table_entry));
1244                 if (!lte_tmp) {
1245                         ret = WIMLIB_ERR_NOMEM;
1246                         goto out_free_stream_tmp_filename;
1247                 }
1248                 lte_tmp->resource_location = RESOURCE_IN_FILE_ON_DISK;
1249                 lte_tmp->file_on_disk = stream_tmp_filename;
1250                 lte_override = lte_tmp;
1251
1252                 raw_fd = topen(stream_tmp_filename,
1253                                O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0600);
1254                 if (raw_fd < 0) {
1255                         ERROR_WITH_ERRNO("Failed to open temporary file");
1256                         ret = WIMLIB_ERR_OPEN;
1257                         goto out_free_lte_tmp;
1258                 }
1259                 filedes_init(&fd, raw_fd);
1260                 ret = extract_wim_resource_to_fd(lte, &fd,
1261                                                  wim_resource_size(lte));
1262                 if (filedes_close(&fd) && !ret)
1263                         ret = WIMLIB_ERR_WRITE;
1264                 if (ret)
1265                         goto out_unlink_stream_tmp_file;
1266         }
1267
1268         /* Extract all instances of the stream, reading either from the stream
1269          * in the WIM file or from the temporary file containing the stream.
1270          * dentry->tmp_flag is used to ensure that each dentry is processed only
1271          * once regardless of how many times this stream appears in the streams
1272          * of the corresponding inode.  */
1273         for (i = 0; i < lte->out_refcnt; i++) {
1274                 struct wim_dentry *dentry = lte_dentries[i];
1275
1276                 if (dentry->tmp_flag)
1277                         continue;
1278                 if (!build_extraction_path(path, dentry, ctx))
1279                         continue;
1280                 ret = extract_streams(path, ctx, dentry,
1281                                       lte, lte_override);
1282                 if (ret)
1283                         goto out_clear_tmp_flags;
1284                 dentry->tmp_flag = 1;
1285         }
1286         ret = 0;
1287 out_clear_tmp_flags:
1288         for (i = 0; i < lte->out_refcnt; i++)
1289                 lte_dentries[i]->tmp_flag = 0;
1290 out_unlink_stream_tmp_file:
1291         if (stream_tmp_filename)
1292                 tunlink(stream_tmp_filename);
1293 out_free_lte_tmp:
1294         FREE(lte_tmp);
1295 out_free_stream_tmp_filename:
1296         FREE(stream_tmp_filename);
1297 out:
1298         return ret;
1299 }
1300
1301 /* Extracts a list of streams (ctx.stream_list), assuming that the directory
1302  * structure and empty files were already created.  This relies on the
1303  * per-`struct wim_lookup_table_entry' list of dentries that reference each
1304  * stream that was constructed earlier.  Streams are extracted exactly in the
1305  * order of the stream list; however, unless the WIM's file descriptor is
1306  * detected to be non-seekable, streams may be read from the WIM file more than
1307  * one time if multiple copies need to be extracted.  */
1308 static int
1309 extract_stream_list(struct apply_ctx *ctx)
1310 {
1311         struct wim_lookup_table_entry *lte;
1312         bool can_seek;
1313         int ret;
1314
1315         can_seek = (lseek(ctx->wim->in_fd.fd, 0, SEEK_CUR) != -1);
1316         list_for_each_entry(lte, &ctx->stream_list, extraction_list) {
1317                 ret = extract_stream_instances(lte, ctx, can_seek);
1318                 if (ret)
1319                         return ret;
1320         }
1321         return 0;
1322 }
1323
1324 #define PWM_ALLOW_WIM_HDR 0x00001
1325 #define PWM_SILENT_EOF    0x00002
1326
1327 /* Read the header from a stream in a pipable WIM.  */
1328 static int
1329 read_pwm_stream_header(WIMStruct *pwm, struct wim_lookup_table_entry *lte,
1330                        int flags, struct wim_header_disk *hdr_ret)
1331 {
1332         union {
1333                 struct pwm_stream_hdr stream_hdr;
1334                 struct wim_header_disk pwm_hdr;
1335         } buf;
1336         int ret;
1337
1338         ret = full_read(&pwm->in_fd, &buf.stream_hdr, sizeof(buf.stream_hdr));
1339         if (ret)
1340                 goto read_error;
1341
1342         if ((flags & PWM_ALLOW_WIM_HDR) && buf.stream_hdr.magic == PWM_MAGIC) {
1343                 BUILD_BUG_ON(sizeof(buf.pwm_hdr) < sizeof(buf.stream_hdr));
1344                 ret = full_read(&pwm->in_fd, &buf.stream_hdr + 1,
1345                                 sizeof(buf.pwm_hdr) - sizeof(buf.stream_hdr));
1346
1347                 if (ret)
1348                         goto read_error;
1349                 lte->resource_location = RESOURCE_NONEXISTENT;
1350                 memcpy(hdr_ret, &buf.pwm_hdr, sizeof(buf.pwm_hdr));
1351                 return 0;
1352         }
1353
1354         if (buf.stream_hdr.magic != PWM_STREAM_MAGIC) {
1355                 ERROR("Data read on pipe is invalid (expected stream header).");
1356                 return WIMLIB_ERR_INVALID_PIPABLE_WIM;
1357         }
1358
1359         lte->resource_entry.original_size = le64_to_cpu(buf.stream_hdr.uncompressed_size);
1360         copy_hash(lte->hash, buf.stream_hdr.hash);
1361         lte->resource_entry.flags = le32_to_cpu(buf.stream_hdr.flags);
1362         lte->resource_entry.offset = pwm->in_fd.offset;
1363         lte->resource_location = RESOURCE_IN_WIM;
1364         lte->wim = pwm;
1365         if (lte->resource_entry.flags & WIM_RESHDR_FLAG_COMPRESSED) {
1366                 lte->compression_type = pwm->compression_type;
1367                 lte->resource_entry.size = 0;
1368         } else {
1369                 lte->compression_type = WIMLIB_COMPRESSION_TYPE_NONE;
1370                 lte->resource_entry.size = lte->resource_entry.original_size;
1371         }
1372         lte->is_pipable = 1;
1373         return 0;
1374
1375 read_error:
1376         if (ret != WIMLIB_ERR_UNEXPECTED_END_OF_FILE || !(flags & PWM_SILENT_EOF))
1377                 ERROR_WITH_ERRNO("Error reading pipable WIM from pipe");
1378         return ret;
1379 }
1380
1381 /* Skip over an unneeded stream in a pipable WIM being read from a pipe.  */
1382 static int
1383 skip_pwm_stream(struct wim_lookup_table_entry *lte)
1384 {
1385         return read_partial_wim_resource(lte, wim_resource_size(lte),
1386                                          NULL, NULL,
1387                                          WIMLIB_READ_RESOURCE_FLAG_SEEK_ONLY,
1388                                          0);
1389 }
1390
1391 static int
1392 extract_streams_from_pipe(struct apply_ctx *ctx)
1393 {
1394         struct wim_lookup_table_entry *found_lte;
1395         struct wim_lookup_table_entry *needed_lte;
1396         struct wim_lookup_table *lookup_table;
1397         struct wim_header_disk pwm_hdr;
1398         int ret;
1399         int pwm_flags;
1400
1401         ret = WIMLIB_ERR_NOMEM;
1402         found_lte = new_lookup_table_entry();
1403         if (!found_lte)
1404                 goto out;
1405
1406         lookup_table = ctx->wim->lookup_table;
1407         pwm_flags = PWM_ALLOW_WIM_HDR;
1408         if ((ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1409                 pwm_flags |= PWM_SILENT_EOF;
1410         memcpy(ctx->progress.extract.guid, ctx->wim->hdr.guid, WIM_GID_LEN);
1411         ctx->progress.extract.part_number = ctx->wim->hdr.part_number;
1412         ctx->progress.extract.total_parts = ctx->wim->hdr.total_parts;
1413         if (ctx->progress_func)
1414                 ctx->progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1415                                    &ctx->progress);
1416         while (ctx->num_streams_remaining) {
1417                 ret = read_pwm_stream_header(ctx->wim, found_lte, pwm_flags,
1418                                              &pwm_hdr);
1419                 if (ret) {
1420                         if (ret == WIMLIB_ERR_UNEXPECTED_END_OF_FILE &&
1421                             (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_RESUME))
1422                         {
1423                                 goto resume_done;
1424                         }
1425                         goto out_free_found_lte;
1426                 }
1427
1428                 if ((found_lte->resource_location != RESOURCE_NONEXISTENT)
1429                     && !(found_lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA)
1430                     && (needed_lte = __lookup_resource(lookup_table, found_lte->hash))
1431                     && (needed_lte->out_refcnt))
1432                 {
1433                         copy_resource_entry(&needed_lte->resource_entry,
1434                                             &found_lte->resource_entry);
1435                         needed_lte->resource_location = found_lte->resource_location;
1436                         needed_lte->wim               = found_lte->wim;
1437                         needed_lte->compression_type  = found_lte->compression_type;
1438                         needed_lte->is_pipable        = found_lte->is_pipable;
1439
1440                         ret = extract_stream_instances(needed_lte, ctx, false);
1441                         if (ret)
1442                                 goto out_free_found_lte;
1443                         ctx->num_streams_remaining--;
1444                 } else if (found_lte->resource_location != RESOURCE_NONEXISTENT) {
1445                         ret = skip_pwm_stream(found_lte);
1446                         if (ret)
1447                                 goto out_free_found_lte;
1448                 } else {
1449                         u16 part_number = le16_to_cpu(pwm_hdr.part_number);
1450                         u16 total_parts = le16_to_cpu(pwm_hdr.total_parts);
1451
1452                         if (part_number != ctx->progress.extract.part_number ||
1453                             total_parts != ctx->progress.extract.total_parts ||
1454                             memcmp(pwm_hdr.guid, ctx->progress.extract.guid,
1455                                    WIM_GID_LEN))
1456                         {
1457                                 ctx->progress.extract.part_number = part_number;
1458                                 ctx->progress.extract.total_parts = total_parts;
1459                                 memcpy(ctx->progress.extract.guid,
1460                                        pwm_hdr.guid, WIM_GID_LEN);
1461                                 if (ctx->progress_func) {
1462                                         ctx->progress_func(
1463                                                 WIMLIB_PROGRESS_MSG_EXTRACT_SPWM_PART_BEGIN,
1464                                                            &ctx->progress);
1465                                 }
1466
1467                         }
1468                 }
1469         }
1470         ret = 0;
1471 out_free_found_lte:
1472         free_lookup_table_entry(found_lte);
1473 out:
1474         return ret;
1475
1476 resume_done:
1477         /* TODO */
1478         return 0;
1479 }
1480
1481 /* Finish extracting a file, directory, or symbolic link by setting file
1482  * security and timestamps.  */
1483 static int
1484 dentry_extract_final(struct wim_dentry *dentry, void *_ctx)
1485 {
1486         struct apply_ctx *ctx = _ctx;
1487         int ret;
1488         tchar path[ctx->ops->path_max];
1489
1490         if (!build_extraction_path(path, dentry, ctx))
1491                 return 0;
1492
1493         ret = extract_security(path, ctx, dentry);
1494         if (ret)
1495                 return ret;
1496
1497         if (ctx->ops->requires_final_set_attributes_pass) {
1498                 /* Set file attributes (if supported).  */
1499                 ret = extract_file_attributes(path, ctx, dentry, 1);
1500                 if (ret)
1501                         return ret;
1502         }
1503
1504         return extract_timestamps(path, ctx, dentry);
1505 }
1506
1507 /*
1508  * Extract a WIM dentry to standard output.
1509  *
1510  * This obviously doesn't make sense in all cases.  We return an error if the
1511  * dentry does not correspond to a regular file.  Otherwise we extract the
1512  * unnamed data stream only.
1513  */
1514 static int
1515 extract_dentry_to_stdout(struct wim_dentry *dentry)
1516 {
1517         int ret = 0;
1518         if (dentry->d_inode->i_attributes & (FILE_ATTRIBUTE_REPARSE_POINT |
1519                                              FILE_ATTRIBUTE_DIRECTORY))
1520         {
1521                 ERROR("\"%"TS"\" is not a regular file and therefore cannot be "
1522                       "extracted to standard output", dentry_full_path(dentry));
1523                 ret = WIMLIB_ERR_NOT_A_REGULAR_FILE;
1524         } else {
1525                 struct wim_lookup_table_entry *lte;
1526
1527                 lte = inode_unnamed_lte_resolved(dentry->d_inode);
1528                 if (lte) {
1529                         struct filedes _stdout;
1530                         filedes_init(&_stdout, STDOUT_FILENO);
1531                         ret = extract_wim_resource_to_fd(lte, &_stdout,
1532                                                          wim_resource_size(lte));
1533                 }
1534         }
1535         return ret;
1536 }
1537
1538 #ifdef __WIN32__
1539 static const utf16lechar replacement_char = cpu_to_le16(0xfffd);
1540 #else
1541 static const utf16lechar replacement_char = cpu_to_le16('?');
1542 #endif
1543
1544 static bool
1545 file_name_valid(utf16lechar *name, size_t num_chars, bool fix)
1546 {
1547         size_t i;
1548
1549         if (num_chars == 0)
1550                 return true;
1551         for (i = 0; i < num_chars; i++) {
1552                 switch (name[i]) {
1553         #ifdef __WIN32__
1554                 case cpu_to_le16('\\'):
1555                 case cpu_to_le16(':'):
1556                 case cpu_to_le16('*'):
1557                 case cpu_to_le16('?'):
1558                 case cpu_to_le16('"'):
1559                 case cpu_to_le16('<'):
1560                 case cpu_to_le16('>'):
1561                 case cpu_to_le16('|'):
1562         #endif
1563                 case cpu_to_le16('/'):
1564                 case cpu_to_le16('\0'):
1565                         if (fix)
1566                                 name[i] = replacement_char;
1567                         else
1568                                 return false;
1569                 }
1570         }
1571
1572 #ifdef __WIN32__
1573         if (name[num_chars - 1] == cpu_to_le16(' ') ||
1574             name[num_chars - 1] == cpu_to_le16('.'))
1575         {
1576                 if (fix)
1577                         name[num_chars - 1] = replacement_char;
1578                 else
1579                         return false;
1580         }
1581 #endif
1582         return true;
1583 }
1584
1585 static bool
1586 dentry_is_dot_or_dotdot(const struct wim_dentry *dentry)
1587 {
1588         const utf16lechar *file_name = dentry->file_name;
1589         return file_name != NULL &&
1590                 file_name[0] == cpu_to_le16('.') &&
1591                 (file_name[1] == cpu_to_le16('\0') ||
1592                  (file_name[1] == cpu_to_le16('.') &&
1593                   file_name[2] == cpu_to_le16('\0')));
1594 }
1595
1596 static int
1597 dentry_mark_skipped(struct wim_dentry *dentry, void *_ignore)
1598 {
1599         dentry->extraction_skipped = 1;
1600         return 0;
1601 }
1602
1603 /*
1604  * dentry_calculate_extraction_path-
1605  *
1606  * Calculate the actual filename component at which a WIM dentry will be
1607  * extracted, handling invalid filenames "properly".
1608  *
1609  * dentry->extraction_name usually will be set the same as dentry->file_name (on
1610  * UNIX, converted into the platform's multibyte encoding).  However, if the
1611  * file name contains characters that are not valid on the current platform or
1612  * has some other format that is not valid, leave dentry->extraction_name as
1613  * NULL and set dentry->extraction_skipped to indicate that this dentry should
1614  * not be extracted, unless the appropriate flag
1615  * WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES is set in the extract flags, in
1616  * which case a substitute filename will be created and set instead.
1617  *
1618  * Conflicts with case-insensitive names on Windows are handled similarly; see
1619  * below.
1620  */
1621 static int
1622 dentry_calculate_extraction_path(struct wim_dentry *dentry, void *_args)
1623 {
1624         struct apply_ctx *ctx = _args;
1625         int ret;
1626
1627         if (dentry == ctx->extract_root || dentry->extraction_skipped)
1628                 return 0;
1629
1630         if (!dentry_is_supported(dentry, &ctx->supported_features))
1631                 goto skip_dentry;
1632
1633         if (dentry_is_dot_or_dotdot(dentry)) {
1634                 /* WIM files shouldn't contain . or .. entries.  But if they are
1635                  * there, don't attempt to extract them. */
1636                 WARNING("Skipping extraction of unexpected . or .. file "
1637                         "\"%"TS"\"", dentry_full_path(dentry));
1638                 goto skip_dentry;
1639         }
1640
1641 #ifdef __WIN32__
1642         if (!ctx->ops->supports_case_sensitive_filenames)
1643         {
1644                 struct wim_dentry *other;
1645                 list_for_each_entry(other, &dentry->case_insensitive_conflict_list,
1646                                     case_insensitive_conflict_list)
1647                 {
1648                         if (ctx->extract_flags &
1649                             WIMLIB_EXTRACT_FLAG_ALL_CASE_CONFLICTS) {
1650                                 WARNING("\"%"TS"\" has the same "
1651                                         "case-insensitive name as "
1652                                         "\"%"TS"\"; extracting "
1653                                         "dummy name instead",
1654                                         dentry_full_path(dentry),
1655                                         dentry_full_path(other));
1656                                 goto out_replace;
1657                         } else {
1658                                 WARNING("Not extracting \"%"TS"\": "
1659                                         "has same case-insensitive "
1660                                         "name as \"%"TS"\"",
1661                                         dentry_full_path(dentry),
1662                                         dentry_full_path(other));
1663                                 goto skip_dentry;
1664                         }
1665                 }
1666         }
1667 #else   /* __WIN32__ */
1668         wimlib_assert(ctx->ops->supports_case_sensitive_filenames);
1669 #endif  /* !__WIN32__ */
1670
1671         if (file_name_valid(dentry->file_name, dentry->file_name_nbytes / 2, false)) {
1672 #ifdef __WIN32__
1673                 dentry->extraction_name = dentry->file_name;
1674                 dentry->extraction_name_nchars = dentry->file_name_nbytes / 2;
1675                 return 0;
1676 #else
1677                 return utf16le_to_tstr(dentry->file_name,
1678                                        dentry->file_name_nbytes,
1679                                        &dentry->extraction_name,
1680                                        &dentry->extraction_name_nchars);
1681 #endif
1682         } else {
1683                 if (ctx->extract_flags & WIMLIB_EXTRACT_FLAG_REPLACE_INVALID_FILENAMES)
1684                 {
1685                         WARNING("\"%"TS"\" has an invalid filename "
1686                                 "that is not supported on this platform; "
1687                                 "extracting dummy name instead",
1688                                 dentry_full_path(dentry));
1689                         goto out_replace;
1690                 } else {
1691                         WARNING("Not extracting \"%"TS"\": has an invalid filename "
1692                                 "that is not supported on this platform",
1693                                 dentry_full_path(dentry));
1694                         goto skip_dentry;
1695                 }
1696         }
1697
1698 out_replace:
1699         {
1700                 utf16lechar utf16_name_copy[dentry->file_name_nbytes / 2];
1701
1702                 memcpy(utf16_name_copy, dentry->file_name, dentry->file_name_nbytes);
1703                 file_name_valid(utf16_name_copy, dentry->file_name_nbytes / 2, true);
1704
1705                 tchar *tchar_name;
1706                 size_t tchar_nchars;
1707         #ifdef __WIN32__
1708                 tchar_name = utf16_name_copy;
1709                 tchar_nchars = dentry->file_name_nbytes / 2;
1710         #else
1711                 ret = utf16le_to_tstr(utf16_name_copy,
1712                                       dentry->file_name_nbytes,
1713                                       &tchar_name, &tchar_nchars);
1714                 if (ret)
1715                         return ret;
1716         #endif
1717                 size_t fixed_name_num_chars = tchar_nchars;
1718                 tchar fixed_name[tchar_nchars + 50];
1719
1720                 tmemcpy(fixed_name, tchar_name, tchar_nchars);
1721                 fixed_name_num_chars += tsprintf(fixed_name + tchar_nchars,
1722                                                  T(" (invalid filename #%lu)"),
1723                                                  ++ctx->invalid_sequence);
1724         #ifndef __WIN32__
1725                 FREE(tchar_name);
1726         #endif
1727                 dentry->extraction_name = memdup(fixed_name,
1728                                                  2 * fixed_name_num_chars + 2);
1729                 if (!dentry->extraction_name)
1730                         return WIMLIB_ERR_NOMEM;
1731                 dentry->extraction_name_nchars = fixed_name_num_chars;
1732         }
1733         return 0;
1734
1735 skip_dentry:
1736         for_dentry_in_tree(dentry, dentry_mark_skipped, NULL);
1737         return 0;
1738 }
1739
1740 /* Clean up dentry and inode structure after extraction.  */
1741 static int
1742 dentry_reset_needs_extraction(struct wim_dentry *dentry, void *_ignore)
1743 {
1744         struct wim_inode *inode = dentry->d_inode;
1745
1746         dentry->extraction_skipped = 0;
1747         dentry->was_hardlinked = 0;
1748         dentry->skeleton_extracted = 0;
1749         inode->i_visited = 0;
1750         FREE(inode->i_extracted_file);
1751         inode->i_extracted_file = NULL;
1752         inode->i_dos_name_extracted = 0;
1753         if ((void*)dentry->extraction_name != (void*)dentry->file_name)
1754                 FREE(dentry->extraction_name);
1755         dentry->extraction_name = NULL;
1756         return 0;
1757 }
1758
1759 /* Tally features necessary to extract a dentry and the corresponding inode.  */
1760 static int
1761 dentry_tally_features(struct wim_dentry *dentry, void *_features)
1762 {
1763         struct wim_features *features = _features;
1764         struct wim_inode *inode = dentry->d_inode;
1765
1766         if (inode->i_attributes & FILE_ATTRIBUTE_ARCHIVE)
1767                 features->archive_files++;
1768         if (inode->i_attributes & FILE_ATTRIBUTE_HIDDEN)
1769                 features->hidden_files++;
1770         if (inode->i_attributes & FILE_ATTRIBUTE_SYSTEM)
1771                 features->system_files++;
1772         if (inode->i_attributes & FILE_ATTRIBUTE_COMPRESSED)
1773                 features->compressed_files++;
1774         if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) {
1775                 if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
1776                         features->encrypted_directories++;
1777                 else
1778                         features->encrypted_files++;
1779         }
1780         if (inode->i_attributes & FILE_ATTRIBUTE_NOT_CONTENT_INDEXED)
1781                 features->not_context_indexed_files++;
1782         if (inode->i_attributes & FILE_ATTRIBUTE_SPARSE_FILE)
1783                 features->sparse_files++;
1784         if (inode_has_named_stream(inode))
1785                 features->named_data_streams++;
1786         if (inode->i_visited)
1787                 features->hard_links++;
1788         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1789                 features->reparse_points++;
1790                 if (inode_is_symlink(inode))
1791                         features->symlink_reparse_points++;
1792                 else
1793                         features->other_reparse_points++;
1794         }
1795         if (inode->i_security_id != -1)
1796                 features->security_descriptors++;
1797         if (dentry->short_name_nbytes)
1798                 features->short_names++;
1799         if (inode_has_unix_data(inode))
1800                 features->unix_data++;
1801         inode->i_visited = 1;
1802         return 0;
1803 }
1804
1805 /* Tally the features necessary to extract a dentry tree.  */
1806 static void
1807 dentry_tree_get_features(struct wim_dentry *root, struct wim_features *features)
1808 {
1809         memset(features, 0, sizeof(struct wim_features));
1810         for_dentry_in_tree(root, dentry_tally_features, features);
1811         dentry_tree_clear_inode_visited(root);
1812 }
1813
1814 static u32
1815 compute_supported_attributes_mask(const struct wim_features *supported_features)
1816 {
1817         u32 mask = ~(u32)0;
1818
1819         if (!supported_features->archive_files)
1820                 mask &= ~FILE_ATTRIBUTE_ARCHIVE;
1821
1822         if (!supported_features->hidden_files)
1823                 mask &= ~FILE_ATTRIBUTE_HIDDEN;
1824
1825         if (!supported_features->system_files)
1826                 mask &= ~FILE_ATTRIBUTE_SYSTEM;
1827
1828         if (!supported_features->not_context_indexed_files)
1829                 mask &= ~FILE_ATTRIBUTE_NOT_CONTENT_INDEXED;
1830
1831         if (!supported_features->compressed_files)
1832                 mask &= ~FILE_ATTRIBUTE_COMPRESSED;
1833
1834         if (!supported_features->sparse_files)
1835                 mask &= ~FILE_ATTRIBUTE_SPARSE_FILE;
1836
1837         if (!supported_features->reparse_points)
1838                 mask &= ~FILE_ATTRIBUTE_REPARSE_POINT;
1839
1840         return mask;
1841 }
1842
1843 static int
1844 do_feature_check(const struct wim_features *required_features,
1845                  const struct wim_features *supported_features,
1846                  int extract_flags,
1847                  const struct apply_operations *ops,
1848                  const tchar *wim_source_path)
1849 {
1850         const tchar *loc;
1851         const tchar *mode = T("this extraction mode");
1852
1853         if (wim_source_path[0] == '\0')
1854                 loc = T("the WIM image");
1855         else
1856                 loc = wim_source_path;
1857
1858         /* We're an archive program, so theoretically we can do what we want
1859          * with FILE_ATTRIBUTE_ARCHIVE (which is a dumb flag anyway).  Don't
1860          * bother the user about it.  */
1861 #if 0
1862         if (required_features->archive_files && !supported_features->archive_files)
1863         {
1864                 WARNING(
1865           "%lu files in %"TS" are marked as archived, but this attribute\n"
1866 "          is not supported in %"TS".",
1867                         required_features->archive_files, loc, mode);
1868         }
1869 #endif
1870
1871         if (required_features->hidden_files && !supported_features->hidden_files)
1872         {
1873                 WARNING(
1874           "%lu files in %"TS" are marked as hidden, but this\n"
1875 "          attribute is not supported in %"TS".",
1876                         required_features->hidden_files, loc, mode);
1877         }
1878
1879         if (required_features->system_files && !supported_features->system_files)
1880         {
1881                 WARNING(
1882           "%lu files in %"TS" are marked as system files,\n"
1883 "          but this attribute is not supported in %"TS".",
1884                         required_features->system_files, loc, mode);
1885         }
1886
1887         if (required_features->compressed_files && !supported_features->compressed_files)
1888         {
1889                 WARNING(
1890           "%lu files in %"TS" are marked as being transparently\n"
1891 "          compressed, but transparent compression is not supported in\n"
1892 "          %"TS".  These files will be extracted as uncompressed.",
1893                         required_features->compressed_files, loc, mode);
1894         }
1895
1896         if (required_features->encrypted_files && !supported_features->encrypted_files)
1897         {
1898                 WARNING(
1899           "%lu files in %"TS" are marked as being encrypted,\n"
1900 "           but encryption is not supported in %"TS".  These files\n"
1901 "           will not be extracted.",
1902                         required_features->encrypted_files, loc, mode);
1903         }
1904
1905         if (required_features->encrypted_directories &&
1906             !supported_features->encrypted_directories)
1907         {
1908                 WARNING(
1909           "%lu directories in %"TS" are marked as being encrypted,\n"
1910 "           but encryption is not supported in %"TS".\n"
1911 "           These directories will be extracted as unencrypted.",
1912                         required_features->encrypted_directories, loc, mode);
1913         }
1914
1915         if (required_features->not_context_indexed_files &&
1916             !supported_features->not_context_indexed_files)
1917         {
1918                 WARNING(
1919           "%lu files in %"TS" are marked as not content indexed,\n"
1920 "          but this attribute is not supported in %"TS".",
1921                         required_features->not_context_indexed_files, loc, mode);
1922         }
1923
1924         if (required_features->sparse_files && !supported_features->sparse_files)
1925         {
1926                 WARNING(
1927           "%lu files in %"TS" are marked as sparse, but creating\n"
1928 "           sparse files is not supported in %"TS".  These files\n"
1929 "           will be extracted as non-sparse.",
1930                         required_features->sparse_files, loc, mode);
1931         }
1932
1933         if (required_features->named_data_streams &&
1934             !supported_features->named_data_streams)
1935         {
1936                 WARNING(
1937           "%lu files in %"TS" contain one or more alternate (named)\n"
1938 "          data streams, which are not supported in %"TS".\n"
1939 "          Alternate data streams will NOT be extracted.",
1940                         required_features->named_data_streams, loc, mode);
1941         }
1942
1943         if (unlikely(extract_flags & (WIMLIB_EXTRACT_FLAG_HARDLINK |
1944                                       WIMLIB_EXTRACT_FLAG_SYMLINK)) &&
1945             required_features->named_data_streams &&
1946             supported_features->named_data_streams)
1947         {
1948                 WARNING(
1949           "%lu files in %"TS" contain one or more alternate (named)\n"
1950 "          data streams, which are not supported in linked extraction mode.\n"
1951 "          Alternate data streams will NOT be extracted.",
1952                         required_features->named_data_streams, loc);
1953         }
1954
1955         if (required_features->hard_links && !supported_features->hard_links)
1956         {
1957                 WARNING(
1958           "%lu files in %"TS" are hard links, but hard links are\n"
1959 "          not supported in %"TS".  Hard links will be extracted as\n"
1960 "          duplicate copies of the linked files.",
1961                         required_features->hard_links, loc, mode);
1962         }
1963
1964         if (required_features->reparse_points && !supported_features->reparse_points)
1965         {
1966                 if (supported_features->symlink_reparse_points) {
1967                         if (required_features->other_reparse_points) {
1968                                 WARNING(
1969           "%lu files in %"TS" are reparse points that are neither\n"
1970 "          symbolic links nor junction points and are not supported in\n"
1971 "          %"TS".  These reparse points will not be extracted.",
1972                                         required_features->other_reparse_points, loc,
1973                                         mode);
1974                         }
1975                 } else {
1976                         WARNING(
1977           "%lu files in %"TS" are reparse points, which are\n"
1978 "          not supported in %"TS" and will not be extracted.",
1979                                 required_features->reparse_points, loc, mode);
1980                 }
1981         }
1982
1983         if (required_features->security_descriptors &&
1984             !supported_features->security_descriptors)
1985         {
1986                 WARNING(
1987           "%lu files in %"TS" have Windows NT security descriptors,\n"
1988 "          but extracting security descriptors is not supported in\n"
1989 "          %"TS".  No security descriptors will be extracted.",
1990                         required_features->security_descriptors, loc, mode);
1991         }
1992
1993         if (required_features->short_names && !supported_features->short_names)
1994         {
1995                 WARNING(
1996           "%lu files in %"TS" have short (DOS) names, but\n"
1997 "          extracting short names is not supported in %"TS".\n"
1998 "          Short names will not be extracted.\n",
1999                         required_features->short_names, loc, mode);
2000         }
2001
2002         if ((extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) &&
2003             required_features->unix_data && !supported_features->unix_data)
2004         {
2005                 ERROR("Extracting UNIX data is not supported in %"TS, mode);
2006                 return WIMLIB_ERR_UNSUPPORTED;
2007         }
2008         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SHORT_NAMES) &&
2009             required_features->short_names && !supported_features->short_names)
2010         {
2011                 ERROR("Extracting short names is not supported in %"TS"", mode);
2012                 return WIMLIB_ERR_UNSUPPORTED;
2013         }
2014         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS) &&
2015             !ops->set_timestamps)
2016         {
2017                 ERROR("Extracting timestamps is not supported in %"TS"", mode);
2018                 return WIMLIB_ERR_UNSUPPORTED;
2019         }
2020         if (((extract_flags & (WIMLIB_EXTRACT_FLAG_STRICT_ACLS |
2021                                WIMLIB_EXTRACT_FLAG_UNIX_DATA))
2022              == WIMLIB_EXTRACT_FLAG_STRICT_ACLS) &&
2023             required_features->security_descriptors &&
2024             !supported_features->security_descriptors)
2025         {
2026                 ERROR("Extracting security descriptors is not supported in %"TS, mode);
2027                 return WIMLIB_ERR_UNSUPPORTED;
2028         }
2029
2030         if ((extract_flags & WIMLIB_EXTRACT_FLAG_HARDLINK) &&
2031             !supported_features->hard_links)
2032         {
2033                 ERROR("Hard link extraction mode requested, but "
2034                       "%"TS" does not support hard links!", mode);
2035                 return WIMLIB_ERR_UNSUPPORTED;
2036         }
2037
2038         if ((extract_flags & WIMLIB_EXTRACT_FLAG_STRICT_SYMLINKS) &&
2039             required_features->symlink_reparse_points &&
2040             !(supported_features->symlink_reparse_points ||
2041               supported_features->reparse_points))
2042         {
2043                 ERROR("Extracting symbolic links is not supported in %"TS, mode);
2044                 return WIMLIB_ERR_UNSUPPORTED;
2045         }
2046
2047         if ((extract_flags & WIMLIB_EXTRACT_FLAG_SYMLINK) &&
2048             !supported_features->symlink_reparse_points)
2049         {
2050                 ERROR("Symbolic link extraction mode requested, but "
2051                       "%"TS" does not support symbolic "
2052                       "links!", mode);
2053                 return WIMLIB_ERR_UNSUPPORTED;
2054         }
2055         return 0;
2056 }
2057
2058 static void
2059 do_extract_warnings(struct apply_ctx *ctx)
2060 {
2061         if (ctx->partial_security_descriptors == 0 &&
2062             ctx->no_security_descriptors == 0)
2063                 return;
2064
2065         WARNING("Extraction of \"%"TS"\" complete, but with one or more warnings:",
2066                 ctx->target);
2067         if (ctx->partial_security_descriptors != 0) {
2068                 WARNING("- Could only partially set the security descriptor\n"
2069                         "            on %lu files or directories.",
2070                         ctx->partial_security_descriptors);
2071         }
2072         if (ctx->no_security_descriptors != 0) {
2073                 WARNING("- Could not set security descriptor at all\n"
2074                         "            on %lu files or directories.",
2075                         ctx->no_security_descriptors);
2076         }
2077 #ifdef __WIN32__
2078         WARNING("To fully restore all security descriptors, run the program\n"
2079                 "          with Administrator rights.");
2080 #endif
2081 }
2082
2083 /*
2084  * extract_tree - Extract a file or directory tree from the currently selected
2085  *                WIM image.
2086  *
2087  * @wim:        WIMStruct for the WIM file, with the desired image selected
2088  *              (as wim->current_image).
2089  *
2090  * @wim_source_path:
2091  *              "Canonical" (i.e. no leading or trailing slashes, path
2092  *              separators WIM_PATH_SEPARATOR) path inside the WIM image to
2093  *              extract.  An empty string means the full image.
2094  *
2095  * @target:
2096  *              Filesystem path to extract the file or directory tree to.
2097  *              (Or, with WIMLIB_EXTRACT_FLAG_NTFS: the name of a NTFS volume.)
2098  *
2099  * @extract_flags:
2100  *              WIMLIB_EXTRACT_FLAG_*.  Also, the private flag
2101  *              WIMLIB_EXTRACT_FLAG_MULTI_IMAGE will be set if this is being
2102  *              called through wimlib_extract_image() with WIMLIB_ALL_IMAGES as
2103  *              the image.
2104  *
2105  * @progress_func:
2106  *              If non-NULL, progress function for the extraction.  The messages
2107  *              that may be sent in this function are:
2108  *
2109  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN or
2110  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN;
2111  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN;
2112  *              WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END;
2113  *              WIMLIB_PROGRESS_MSG_EXTRACT_DENTRY;
2114  *              WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS;
2115  *              WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS;
2116  *              WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END or
2117  *                      WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END.
2118  *
2119  * Returns 0 on success; a positive WIMLIB_ERR_* code on failure.
2120  */
2121 static int
2122 extract_tree(WIMStruct *wim, const tchar *wim_source_path, const tchar *target,
2123              int extract_flags, wimlib_progress_func_t progress_func)
2124 {
2125         struct wim_dentry *root;
2126         struct wim_features required_features;
2127         struct apply_ctx ctx;
2128         int ret;
2129         struct wim_lookup_table_entry *lte;
2130
2131         /* Start initializing the apply_ctx.  */
2132         memset(&ctx, 0, sizeof(struct apply_ctx));
2133         ctx.wim = wim;
2134         ctx.extract_flags = extract_flags;
2135         ctx.target = target;
2136         ctx.target_nchars = tstrlen(target);
2137         ctx.progress_func = progress_func;
2138         if (progress_func) {
2139                 ctx.progress.extract.wimfile_name = wim->filename;
2140                 ctx.progress.extract.image = wim->current_image;
2141                 ctx.progress.extract.extract_flags = (extract_flags &
2142                                                       WIMLIB_EXTRACT_MASK_PUBLIC);
2143                 ctx.progress.extract.image_name = wimlib_get_image_name(wim,
2144                                                                         wim->current_image);
2145                 ctx.progress.extract.extract_root_wim_source_path = wim_source_path;
2146                 ctx.progress.extract.target = target;
2147         }
2148         INIT_LIST_HEAD(&ctx.stream_list);
2149
2150         /* Translate the path to extract into the corresponding
2151          * `struct wim_dentry', which will be the root of the
2152          * "dentry tree" to extract.  */
2153         root = get_dentry(wim, wim_source_path);
2154         if (!root) {
2155                 ERROR("Path \"%"TS"\" does not exist in WIM image %d",
2156                       wim_source_path, wim->current_image);
2157                 ret = WIMLIB_ERR_PATH_DOES_NOT_EXIST;
2158                 goto out;
2159         }
2160
2161         ctx.extract_root = root;
2162
2163         /* Select the appropriate apply_operations based on the
2164          * platform and extract_flags.  */
2165 #ifdef __WIN32__
2166         ctx.ops = &win32_apply_ops;
2167 #else
2168         ctx.ops = &unix_apply_ops;
2169 #endif
2170
2171 #ifdef WITH_NTFS_3G
2172         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS)
2173                 ctx.ops = &ntfs_3g_apply_ops;
2174 #endif
2175
2176         /* Call the start_extract() callback.  This gives the apply_operations
2177          * implementation a chance to do any setup needed to access the volume.
2178          * Furthermore, it's expected to set the supported features of this
2179          * extraction mode (ctx.supported_features), which are determined at
2180          * runtime as they may vary depending on the actual volume.  These
2181          * features are then compared with the actual features extracting this
2182          * dentry tree requires.  Some mismatches will merely produce warnings
2183          * and the unsupported data will be ignored; others will produce errors.
2184          */
2185         ret = ctx.ops->start_extract(target, &ctx);
2186         if (ret)
2187                 goto out;
2188
2189         dentry_tree_get_features(root, &required_features);
2190         ret = do_feature_check(&required_features, &ctx.supported_features,
2191                                extract_flags, ctx.ops, wim_source_path);
2192         if (ret)
2193                 goto out_finish_or_abort_extract;
2194
2195         ctx.supported_attributes_mask =
2196                 compute_supported_attributes_mask(&ctx.supported_features);
2197
2198         /* Figure out whether the root dentry is being extracted to the root of
2199          * a volume and therefore needs to be treated "specially", for example
2200          * not being explicitly created and not having attributes set.  */
2201         if (ctx.ops->target_is_root && ctx.ops->root_directory_is_special)
2202                 ctx.root_dentry_is_special = ctx.ops->target_is_root(target);
2203
2204         /* Calculate the actual filename component of each extracted dentry.  In
2205          * the process, set the dentry->extraction_skipped flag on dentries that
2206          * are being skipped for some reason (e.g. invalid filename).  */
2207         ret = for_dentry_in_tree(root, dentry_calculate_extraction_path, &ctx);
2208         if (ret)
2209                 goto out_dentry_reset_needs_extraction;
2210
2211         /* Build the list of the streams that need to be extracted and
2212          * initialize ctx.progress.extract with stream information.  */
2213         ret = for_dentry_in_tree(ctx.extract_root,
2214                                  dentry_resolve_and_zero_lte_refcnt, &ctx);
2215         if (ret)
2216                 goto out_dentry_reset_needs_extraction;
2217
2218         ret = for_dentry_in_tree(ctx.extract_root,
2219                                  dentry_add_streams_to_extract, &ctx);
2220         if (ret)
2221                 goto out_teardown_stream_list;
2222
2223         if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2224                 /* When extracting from a pipe, the number of bytes of data to
2225                  * extract can't be determined in the normal way (examining the
2226                  * lookup table), since at this point all we have is a set of
2227                  * SHA1 message digests of streams that need to be extracted.
2228                  * However, we can get a reasonably accurate estimate by taking
2229                  * <TOTALBYTES> from the corresponding <IMAGE> in the WIM XML
2230                  * data.  This does assume that a full image is being extracted,
2231                  * but currently there is no API for doing otherwise.  (Also,
2232                  * subtract <HARDLINKBYTES> from this if hard links are
2233                  * supported by the extraction mode.)  */
2234                 ctx.progress.extract.total_bytes =
2235                         wim_info_get_image_total_bytes(wim->wim_info,
2236                                                        wim->current_image);
2237                 if (ctx.supported_features.hard_links) {
2238                         ctx.progress.extract.total_bytes -=
2239                                 wim_info_get_image_hard_link_bytes(wim->wim_info,
2240                                                                    wim->current_image);
2241                 }
2242         }
2243
2244         /* Handle the special case of extracting a file to standard
2245          * output.  In that case, "root" should be a single file, not a
2246          * directory tree.  (If not, extract_dentry_to_stdout() will
2247          * return an error.)  */
2248         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT) {
2249                 ret = extract_dentry_to_stdout(root);
2250                 goto out_teardown_stream_list;
2251         }
2252
2253         /* If a sequential extraction was specified, sort the streams to be
2254          * extracted by their position in the WIM file so that the WIM file can
2255          * be read sequentially.  */
2256         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2257                               WIMLIB_EXTRACT_FLAG_FROM_PIPE))
2258                                         == WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2259         {
2260                 ret = sort_stream_list_by_sequential_order(
2261                                 &ctx.stream_list,
2262                                 offsetof(struct wim_lookup_table_entry,
2263                                          extraction_list));
2264                 if (ret)
2265                         goto out_teardown_stream_list;
2266         }
2267
2268         if (ctx.ops->realpath_works_on_nonexisting_files &&
2269             ((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) ||
2270              ctx.ops->requires_realtarget_in_paths))
2271         {
2272                 ctx.realtarget = realpath(target, NULL);
2273                 if (!ctx.realtarget) {
2274                         ret = WIMLIB_ERR_NOMEM;
2275                         goto out_teardown_stream_list;
2276                 }
2277                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2278         }
2279
2280         if (progress_func) {
2281                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_BEGIN :
2282                                                  WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_BEGIN,
2283                               &ctx.progress);
2284         }
2285
2286         if (!ctx.root_dentry_is_special)
2287         {
2288                 tchar path[ctx.ops->path_max];
2289                 if (build_extraction_path(path, root, &ctx))
2290                 {
2291                         ret = extract_inode(path, &ctx, root->d_inode);
2292                         if (ret)
2293                                 goto out_free_realtarget;
2294                 }
2295         }
2296
2297         /* If we need to fix up the targets of absolute symbolic links
2298          * (WIMLIB_EXTRACT_FLAG_RPFIX) or the extraction mode requires paths to
2299          * be absolute, use realpath() (or its replacement on Windows) to get
2300          * the absolute path to the extraction target.  Note that this requires
2301          * the target directory to exist, unless
2302          * realpath_works_on_nonexisting_files is set in the apply_operations.
2303          * */
2304         if (!ctx.realtarget &&
2305             (((extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
2306               required_features.symlink_reparse_points) ||
2307              ctx.ops->requires_realtarget_in_paths))
2308         {
2309                 ctx.realtarget = realpath(target, NULL);
2310                 if (!ctx.realtarget) {
2311                         ret = WIMLIB_ERR_NOMEM;
2312                         goto out_free_realtarget;
2313                 }
2314                 ctx.realtarget_nchars = tstrlen(ctx.realtarget);
2315         }
2316
2317         if (ctx.ops->requires_short_name_reordering) {
2318                 ret = for_dentry_in_tree(root, dentry_extract_dir_skeleton,
2319                                          &ctx);
2320                 if (ret)
2321                         goto out_free_realtarget;
2322         }
2323
2324         /* Finally, the important part: extract the tree of files.  */
2325         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SEQUENTIAL |
2326                              WIMLIB_EXTRACT_FLAG_FROM_PIPE)) {
2327                 /* Sequential extraction requested, so two passes are needed
2328                  * (one for directory structure, one for streams.)  */
2329                 if (progress_func)
2330                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2331                                       &ctx.progress);
2332
2333                 if (!(extract_flags & WIMLIB_EXTRACT_FLAG_RESUME)) {
2334                         ret = for_dentry_in_tree(root, dentry_extract_skeleton, &ctx);
2335                         if (ret)
2336                                 goto out_free_realtarget;
2337                 }
2338                 if (progress_func)
2339                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2340                                       &ctx.progress);
2341                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE)
2342                         ret = extract_streams_from_pipe(&ctx);
2343                 else
2344                         ret = extract_stream_list(&ctx);
2345                 if (ret)
2346                         goto out_free_realtarget;
2347         } else {
2348                 /* Sequential extraction was not requested, so we can make do
2349                  * with one pass where we both create the files and extract
2350                  * streams.   */
2351                 if (progress_func)
2352                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_BEGIN,
2353                                       &ctx.progress);
2354                 ret = for_dentry_in_tree(root, dentry_extract, &ctx);
2355                 if (ret)
2356                         goto out_free_realtarget;
2357                 if (progress_func)
2358                         progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_DIR_STRUCTURE_END,
2359                                       &ctx.progress);
2360         }
2361
2362         /* If the total number of bytes to extract was miscalculated, just jump
2363          * to the calculated number in order to avoid confusing the progress
2364          * function.  This should only occur when extracting from a pipe.  */
2365         if (ctx.progress.extract.completed_bytes != ctx.progress.extract.total_bytes)
2366         {
2367                 DEBUG("Calculated %"PRIu64" bytes to extract, but actually "
2368                       "extracted %"PRIu64,
2369                       ctx.progress.extract.total_bytes,
2370                       ctx.progress.extract.completed_bytes);
2371         }
2372         if (progress_func &&
2373             ctx.progress.extract.completed_bytes < ctx.progress.extract.total_bytes)
2374         {
2375                 ctx.progress.extract.completed_bytes = ctx.progress.extract.total_bytes;
2376                 progress_func(WIMLIB_PROGRESS_MSG_EXTRACT_STREAMS, &ctx.progress);
2377         }
2378
2379         /* Apply security descriptors and timestamps.  This is done at the end,
2380          * and in a depth-first manner, to prevent timestamps from getting
2381          * changed by subsequent extract operations and to minimize the chance
2382          * of the restored security descriptors getting in our way.  */
2383         if (progress_func)
2384                 progress_func(WIMLIB_PROGRESS_MSG_APPLY_TIMESTAMPS,
2385                               &ctx.progress);
2386         ret = for_dentry_in_tree_depth(root, dentry_extract_final, &ctx);
2387         if (ret)
2388                 goto out_free_realtarget;
2389
2390         if (progress_func) {
2391                 progress_func(*wim_source_path ? WIMLIB_PROGRESS_MSG_EXTRACT_TREE_END :
2392                               WIMLIB_PROGRESS_MSG_EXTRACT_IMAGE_END,
2393                               &ctx.progress);
2394         }
2395
2396         do_extract_warnings(&ctx);
2397
2398         ret = 0;
2399 out_free_realtarget:
2400         FREE(ctx.realtarget);
2401 out_teardown_stream_list:
2402         /* Free memory allocated as part of the mapping from each
2403          * wim_lookup_table_entry to the dentries that reference it.  */
2404         if (ctx.extract_flags & WIMLIB_EXTRACT_FLAG_SEQUENTIAL)
2405                 list_for_each_entry(lte, &ctx.stream_list, extraction_list)
2406                         if (lte->out_refcnt > ARRAY_LEN(lte->inline_lte_dentries))
2407                                 FREE(lte->lte_dentries);
2408 out_dentry_reset_needs_extraction:
2409         for_dentry_in_tree(root, dentry_reset_needs_extraction, NULL);
2410 out_finish_or_abort_extract:
2411         if (ret) {
2412                 if (ctx.ops->abort_extract)
2413                         ctx.ops->abort_extract(&ctx);
2414         } else {
2415                 if (ctx.ops->finish_extract)
2416                         ret = ctx.ops->finish_extract(&ctx);
2417         }
2418 out:
2419         return ret;
2420 }
2421
2422 /* Validates a single wimlib_extract_command, mostly checking to make sure the
2423  * extract flags make sense. */
2424 static int
2425 check_extract_command(struct wimlib_extract_command *cmd, int wim_header_flags)
2426 {
2427         int extract_flags;
2428
2429         /* Empty destination path? */
2430         if (cmd->fs_dest_path[0] == T('\0'))
2431                 return WIMLIB_ERR_INVALID_PARAM;
2432
2433         extract_flags = cmd->extract_flags;
2434
2435         /* Check for invalid flag combinations  */
2436         if ((extract_flags &
2437              (WIMLIB_EXTRACT_FLAG_SYMLINK |
2438               WIMLIB_EXTRACT_FLAG_HARDLINK)) == (WIMLIB_EXTRACT_FLAG_SYMLINK |
2439                                                  WIMLIB_EXTRACT_FLAG_HARDLINK))
2440                 return WIMLIB_ERR_INVALID_PARAM;
2441
2442         if ((extract_flags &
2443              (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2444               WIMLIB_EXTRACT_FLAG_STRICT_ACLS)) == (WIMLIB_EXTRACT_FLAG_NO_ACLS |
2445                                                     WIMLIB_EXTRACT_FLAG_STRICT_ACLS))
2446                 return WIMLIB_ERR_INVALID_PARAM;
2447
2448         if ((extract_flags &
2449              (WIMLIB_EXTRACT_FLAG_RPFIX |
2450               WIMLIB_EXTRACT_FLAG_NORPFIX)) == (WIMLIB_EXTRACT_FLAG_RPFIX |
2451                                                 WIMLIB_EXTRACT_FLAG_NORPFIX))
2452                 return WIMLIB_ERR_INVALID_PARAM;
2453
2454         if ((extract_flags &
2455              (WIMLIB_EXTRACT_FLAG_RESUME |
2456               WIMLIB_EXTRACT_FLAG_FROM_PIPE)) == WIMLIB_EXTRACT_FLAG_RESUME)
2457                 return WIMLIB_ERR_INVALID_PARAM;
2458
2459         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2460 #ifndef WITH_NTFS_3G
2461                 ERROR("wimlib was compiled without support for NTFS-3g, so\n"
2462                       "        we cannot apply a WIM image directly to a NTFS volume.");
2463                 return WIMLIB_ERR_UNSUPPORTED;
2464 #endif
2465         }
2466
2467         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_RPFIX |
2468                               WIMLIB_EXTRACT_FLAG_NORPFIX)) == 0)
2469         {
2470                 /* Do reparse point fixups by default if the WIM header says
2471                  * they are enabled and we are extracting a full image. */
2472                 if (wim_header_flags & WIM_HDR_FLAG_RP_FIX)
2473                         extract_flags |= WIMLIB_EXTRACT_FLAG_RPFIX;
2474         }
2475
2476         /* TODO: Since UNIX data entries are stored in the file resources, in a
2477          * completely sequential extraction they may come up before the
2478          * corresponding file or symbolic link data.  This needs to be handled
2479          * better.  */
2480         if ((extract_flags & (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2481                               WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2482                                     == (WIMLIB_EXTRACT_FLAG_UNIX_DATA |
2483                                         WIMLIB_EXTRACT_FLAG_SEQUENTIAL))
2484         {
2485                 if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2486                         WARNING("Setting UNIX file/owner group may "
2487                                 "be impossible on some\n"
2488                                 "          symbolic links "
2489                                 "when applying from a pipe.");
2490                 } else {
2491                         extract_flags &= ~WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2492                         WARNING("Disabling sequential extraction for "
2493                                 "UNIX data mode");
2494                 }
2495         }
2496
2497         cmd->extract_flags = extract_flags;
2498         return 0;
2499 }
2500
2501
2502 /* Internal function to execute extraction commands for a WIM image.  The paths
2503  * in the extract commands are expected to be already "canonicalized".  */
2504 static int
2505 do_wimlib_extract_files(WIMStruct *wim,
2506                         int image,
2507                         struct wimlib_extract_command *cmds,
2508                         size_t num_cmds,
2509                         wimlib_progress_func_t progress_func)
2510 {
2511         int ret;
2512         bool found_link_cmd = false;
2513         bool found_nolink_cmd = false;
2514
2515         /* Select the image from which we are extracting files */
2516         ret = select_wim_image(wim, image);
2517         if (ret)
2518                 return ret;
2519
2520         /* Make sure there are no streams in the WIM that have not been
2521          * checksummed yet.  */
2522         ret = wim_checksum_unhashed_streams(wim);
2523         if (ret)
2524                 return ret;
2525
2526         /* Check for problems with the extraction commands */
2527         for (size_t i = 0; i < num_cmds; i++) {
2528                 ret = check_extract_command(&cmds[i], wim->hdr.flags);
2529                 if (ret)
2530                         return ret;
2531                 if (cmds[i].extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2532                                              WIMLIB_EXTRACT_FLAG_HARDLINK)) {
2533                         found_link_cmd = true;
2534                 } else {
2535                         found_nolink_cmd = true;
2536                 }
2537                 if (found_link_cmd && found_nolink_cmd) {
2538                         ERROR("Symlink or hardlink extraction mode must "
2539                               "be set on all extraction commands");
2540                         return WIMLIB_ERR_INVALID_PARAM;
2541                 }
2542         }
2543
2544         /* Execute the extraction commands */
2545         for (size_t i = 0; i < num_cmds; i++) {
2546                 ret = extract_tree(wim,
2547                                    cmds[i].wim_source_path,
2548                                    cmds[i].fs_dest_path,
2549                                    cmds[i].extract_flags,
2550                                    progress_func);
2551                 if (ret)
2552                         return ret;
2553         }
2554         return 0;
2555 }
2556
2557 /* API function documented in wimlib.h  */
2558 WIMLIBAPI int
2559 wimlib_extract_files(WIMStruct *wim,
2560                      int image,
2561                      const struct wimlib_extract_command *cmds,
2562                      size_t num_cmds,
2563                      int default_extract_flags,
2564                      WIMStruct **additional_swms,
2565                      unsigned num_additional_swms,
2566                      wimlib_progress_func_t progress_func)
2567 {
2568         int ret;
2569         struct wimlib_extract_command *cmds_copy;
2570         int all_flags = 0;
2571
2572         default_extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2573
2574         ret = verify_swm_set(wim, additional_swms, num_additional_swms);
2575         if (ret)
2576                 goto out;
2577
2578         if (num_cmds == 0)
2579                 goto out;
2580
2581         if (num_additional_swms)
2582                 merge_lookup_tables(wim, additional_swms, num_additional_swms);
2583
2584         cmds_copy = CALLOC(num_cmds, sizeof(cmds[0]));
2585         if (!cmds_copy) {
2586                 ret = WIMLIB_ERR_NOMEM;
2587                 goto out_restore_lookup_table;
2588         }
2589
2590         for (size_t i = 0; i < num_cmds; i++) {
2591                 cmds_copy[i].extract_flags = (default_extract_flags |
2592                                                  cmds[i].extract_flags)
2593                                                 & WIMLIB_EXTRACT_MASK_PUBLIC;
2594                 all_flags |= cmds_copy[i].extract_flags;
2595
2596                 cmds_copy[i].wim_source_path = canonicalize_wim_path(cmds[i].wim_source_path);
2597                 if (!cmds_copy[i].wim_source_path) {
2598                         ret = WIMLIB_ERR_NOMEM;
2599                         goto out_free_cmds_copy;
2600                 }
2601
2602                 cmds_copy[i].fs_dest_path = canonicalize_fs_path(cmds[i].fs_dest_path);
2603                 if (!cmds_copy[i].fs_dest_path) {
2604                         ret = WIMLIB_ERR_NOMEM;
2605                         goto out_free_cmds_copy;
2606                 }
2607
2608         }
2609         ret = do_wimlib_extract_files(wim, image,
2610                                       cmds_copy, num_cmds,
2611                                       progress_func);
2612
2613         if (all_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2614                          WIMLIB_EXTRACT_FLAG_HARDLINK))
2615         {
2616                 for_lookup_table_entry(wim->lookup_table,
2617                                        lte_free_extracted_file, NULL);
2618         }
2619 out_free_cmds_copy:
2620         for (size_t i = 0; i < num_cmds; i++) {
2621                 FREE(cmds_copy[i].wim_source_path);
2622                 FREE(cmds_copy[i].fs_dest_path);
2623         }
2624         FREE(cmds_copy);
2625 out_restore_lookup_table:
2626         if (num_additional_swms)
2627                 unmerge_lookup_table(wim);
2628 out:
2629         return ret;
2630 }
2631
2632 /*
2633  * Extracts an image from a WIM file.
2634  *
2635  * @wim:                WIMStruct for the WIM file.
2636  *
2637  * @image:              Number of the single image to extract.
2638  *
2639  * @target:             Directory or NTFS volume to extract the image to.
2640  *
2641  * @extract_flags:      Bitwise or of WIMLIB_EXTRACT_FLAG_*.
2642  *
2643  * @progress_func:      If non-NULL, a progress function to be called
2644  *                      periodically.
2645  *
2646  * Returns 0 on success; nonzero on failure.
2647  */
2648 static int
2649 extract_single_image(WIMStruct *wim, int image,
2650                      const tchar *target, int extract_flags,
2651                      wimlib_progress_func_t progress_func)
2652 {
2653         int ret;
2654         tchar *target_copy = canonicalize_fs_path(target);
2655         if (!target_copy)
2656                 return WIMLIB_ERR_NOMEM;
2657         struct wimlib_extract_command cmd = {
2658                 .wim_source_path = T(""),
2659                 .fs_dest_path = target_copy,
2660                 .extract_flags = extract_flags,
2661         };
2662         ret = do_wimlib_extract_files(wim, image, &cmd, 1, progress_func);
2663         FREE(target_copy);
2664         return ret;
2665 }
2666
2667 static const tchar * const filename_forbidden_chars =
2668 T(
2669 #ifdef __WIN32__
2670 "<>:\"/\\|?*"
2671 #else
2672 "/"
2673 #endif
2674 );
2675
2676 /* This function checks if it is okay to use a WIM image's name as a directory
2677  * name.  */
2678 static bool
2679 image_name_ok_as_dir(const tchar *image_name)
2680 {
2681         return image_name && *image_name &&
2682                 !tstrpbrk(image_name, filename_forbidden_chars) &&
2683                 tstrcmp(image_name, T(".")) &&
2684                 tstrcmp(image_name, T(".."));
2685 }
2686
2687 /* Extracts all images from the WIM to the directory @target, with the images
2688  * placed in subdirectories named by their image names. */
2689 static int
2690 extract_all_images(WIMStruct *wim,
2691                    const tchar *target,
2692                    int extract_flags,
2693                    wimlib_progress_func_t progress_func)
2694 {
2695         size_t image_name_max_len = max(xml_get_max_image_name_len(wim), 20);
2696         size_t output_path_len = tstrlen(target);
2697         tchar buf[output_path_len + 1 + image_name_max_len + 1];
2698         int ret;
2699         int image;
2700         const tchar *image_name;
2701         struct stat stbuf;
2702
2703         extract_flags |= WIMLIB_EXTRACT_FLAG_MULTI_IMAGE;
2704
2705         if (extract_flags & WIMLIB_EXTRACT_FLAG_NTFS) {
2706                 ERROR("Cannot extract multiple images in NTFS extraction mode.");
2707                 return WIMLIB_ERR_INVALID_PARAM;
2708         }
2709
2710         if (tstat(target, &stbuf)) {
2711                 if (errno == ENOENT) {
2712                         if (tmkdir(target, 0755)) {
2713                                 ERROR_WITH_ERRNO("Failed to create directory \"%"TS"\"", target);
2714                                 return WIMLIB_ERR_MKDIR;
2715                         }
2716                 } else {
2717                         ERROR_WITH_ERRNO("Failed to stat \"%"TS"\"", target);
2718                         return WIMLIB_ERR_STAT;
2719                 }
2720         } else if (!S_ISDIR(stbuf.st_mode)) {
2721                 ERROR("\"%"TS"\" is not a directory", target);
2722                 return WIMLIB_ERR_NOTDIR;
2723         }
2724
2725         tmemcpy(buf, target, output_path_len);
2726         buf[output_path_len] = OS_PREFERRED_PATH_SEPARATOR;
2727         for (image = 1; image <= wim->hdr.image_count; image++) {
2728                 image_name = wimlib_get_image_name(wim, image);
2729                 if (image_name_ok_as_dir(image_name)) {
2730                         tstrcpy(buf + output_path_len + 1, image_name);
2731                 } else {
2732                         /* Image name is empty or contains forbidden characters.
2733                          * Use image number instead. */
2734                         tsprintf(buf + output_path_len + 1, T("%d"), image);
2735                 }
2736                 ret = extract_single_image(wim, image, buf, extract_flags,
2737                                            progress_func);
2738                 if (ret)
2739                         return ret;
2740         }
2741         return 0;
2742 }
2743
2744 static int
2745 do_wimlib_extract_image(WIMStruct *wim,
2746                         int image,
2747                         const tchar *target,
2748                         int extract_flags,
2749                         WIMStruct **additional_swms,
2750                         unsigned num_additional_swms,
2751                         wimlib_progress_func_t progress_func)
2752 {
2753         int ret;
2754
2755         if (extract_flags & WIMLIB_EXTRACT_FLAG_FROM_PIPE) {
2756                 wimlib_assert(wim->hdr.part_number == 1);
2757                 wimlib_assert(num_additional_swms == 0);
2758         } else {
2759                 ret = verify_swm_set(wim, additional_swms, num_additional_swms);
2760                 if (ret)
2761                         return ret;
2762
2763                 if (num_additional_swms)
2764                         merge_lookup_tables(wim, additional_swms, num_additional_swms);
2765         }
2766
2767         if (image == WIMLIB_ALL_IMAGES) {
2768                 ret = extract_all_images(wim, target, extract_flags,
2769                                          progress_func);
2770         } else {
2771                 ret = extract_single_image(wim, image, target, extract_flags,
2772                                            progress_func);
2773         }
2774
2775         if (extract_flags & (WIMLIB_EXTRACT_FLAG_SYMLINK |
2776                              WIMLIB_EXTRACT_FLAG_HARDLINK))
2777         {
2778                 for_lookup_table_entry(wim->lookup_table,
2779                                        lte_free_extracted_file,
2780                                        NULL);
2781         }
2782         if (num_additional_swms)
2783                 unmerge_lookup_table(wim);
2784         return ret;
2785 }
2786
2787 /* API function documented in wimlib.h  */
2788 WIMLIBAPI int
2789 wimlib_extract_image_from_pipe(int pipe_fd, const tchar *image_num_or_name,
2790                                const tchar *target, int extract_flags,
2791                                wimlib_progress_func_t progress_func)
2792 {
2793         int ret;
2794         WIMStruct *pwm;
2795         struct filedes *in_fd;
2796         int image;
2797         unsigned i;
2798
2799         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2800
2801         if (extract_flags & WIMLIB_EXTRACT_FLAG_TO_STDOUT)
2802                 return WIMLIB_ERR_INVALID_PARAM;
2803
2804         extract_flags |= WIMLIB_EXTRACT_FLAG_SEQUENTIAL;
2805
2806         /* Read the WIM header from the pipe and get a WIMStruct to represent
2807          * the pipable WIM.  Caveats:  Unlike getting a WIMStruct with
2808          * wimlib_open_wim(), getting a WIMStruct in this way will result in
2809          * an empty lookup table, no XML data read, and no filename set.  */
2810         ret = open_wim_as_WIMStruct(&pipe_fd,
2811                                     WIMLIB_OPEN_FLAG_FROM_PIPE |
2812                                                 WIMLIB_OPEN_FLAG_SPLIT_OK,
2813                                     &pwm, progress_func);
2814         if (ret)
2815                 return ret;
2816
2817         /* Sanity check to make sure this is a pipable WIM.  */
2818         if (pwm->hdr.magic != PWM_MAGIC) {
2819                 ERROR("The WIM being read from file descriptor %d "
2820                       "is not pipable!", pipe_fd);
2821                 ret = WIMLIB_ERR_NOT_PIPABLE;
2822                 goto out_wimlib_free;
2823         }
2824
2825         /* Sanity check to make sure the first part of a pipable split WIM is
2826          * sent over the pipe first.  */
2827         if (pwm->hdr.part_number != 1) {
2828                 ERROR("The first part of the split WIM must be "
2829                       "sent over the pipe first.");
2830                 ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2831                 goto out_wimlib_free;
2832         }
2833
2834         in_fd = &pwm->in_fd;
2835         wimlib_assert(in_fd->offset == WIM_HEADER_DISK_SIZE);
2836
2837         /* As mentioned, the WIMStruct we created from the pipe does not have
2838          * XML data yet.  Fix this by reading the extra copy of the XML data
2839          * that directly follows the header in pipable WIMs.  (Note: see
2840          * write_pipable_wim() for more details about the format of pipable
2841          * WIMs.)  */
2842         {
2843                 struct wim_lookup_table_entry xml_lte;
2844                 ret = read_pwm_stream_header(pwm, &xml_lte, 0, NULL);
2845                 if (ret)
2846                         goto out_wimlib_free;
2847
2848                 if (!(xml_lte.resource_entry.flags & WIM_RESHDR_FLAG_METADATA))
2849                 {
2850                         ERROR("Expected XML data, but found non-metadata "
2851                               "stream.");
2852                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2853                         goto out_wimlib_free;
2854                 }
2855
2856                 copy_resource_entry(&pwm->hdr.xml_res_entry,
2857                                     &xml_lte.resource_entry);
2858
2859                 ret = read_wim_xml_data(pwm);
2860                 if (ret)
2861                         goto out_wimlib_free;
2862                 if (wim_info_get_num_images(pwm->wim_info) != pwm->hdr.image_count) {
2863                         ERROR("Image count in XML data is not the same as in WIM header.");
2864                         ret = WIMLIB_ERR_XML;
2865                         goto out_wimlib_free;
2866                 }
2867         }
2868
2869         /* Get image index (this may use the XML data that was just read to
2870          * resolve an image name).  */
2871         if (image_num_or_name) {
2872                 image = wimlib_resolve_image(pwm, image_num_or_name);
2873                 if (image == WIMLIB_NO_IMAGE) {
2874                         ERROR("\"%"TS"\" is not a valid image in the pipable WIM!",
2875                               image_num_or_name);
2876                         ret = WIMLIB_ERR_INVALID_IMAGE;
2877                         goto out_wimlib_free;
2878                 } else if (image == WIMLIB_ALL_IMAGES) {
2879                         ERROR("Applying all images from a pipe is not supported.");
2880                         ret = WIMLIB_ERR_INVALID_IMAGE;
2881                         goto out_wimlib_free;
2882                 }
2883         } else {
2884                 if (pwm->hdr.image_count != 1) {
2885                         ERROR("No image was specified, but the pipable WIM "
2886                               "did not contain exactly 1 image");
2887                         ret = WIMLIB_ERR_INVALID_IMAGE;
2888                         goto out_wimlib_free;
2889                 }
2890                 image = 1;
2891         }
2892
2893         /* Load the needed metadata resource.  */
2894         for (i = 1; i <= pwm->hdr.image_count; i++) {
2895                 struct wim_lookup_table_entry *metadata_lte;
2896                 struct wim_image_metadata *imd;
2897
2898                 metadata_lte = new_lookup_table_entry();
2899                 if (!metadata_lte) {
2900                         ret = WIMLIB_ERR_NOMEM;
2901                         goto out_wimlib_free;
2902                 }
2903
2904                 ret = read_pwm_stream_header(pwm, metadata_lte, 0, NULL);
2905                 imd = pwm->image_metadata[i - 1];
2906                 imd->metadata_lte = metadata_lte;
2907                 if (ret)
2908                         goto out_wimlib_free;
2909
2910                 if (!(metadata_lte->resource_entry.flags &
2911                       WIM_RESHDR_FLAG_METADATA))
2912                 {
2913                         ERROR("Expected metadata resource, but found "
2914                               "non-metadata stream.");
2915                         ret = WIMLIB_ERR_INVALID_PIPABLE_WIM;
2916                         goto out_wimlib_free;
2917                 }
2918
2919                 if (i == image) {
2920                         /* Metadata resource is for the images being extracted.
2921                          * Parse it and save the metadata in memory.  */
2922                         ret = read_metadata_resource(pwm, imd);
2923                         if (ret)
2924                                 goto out_wimlib_free;
2925                         imd->modified = 1;
2926                 } else {
2927                         /* Metadata resource is not for the image being
2928                          * extracted.  Skip over it.  */
2929                         ret = skip_pwm_stream(metadata_lte);
2930                         if (ret)
2931                                 goto out_wimlib_free;
2932                 }
2933         }
2934         /* Extract the image.  */
2935         extract_flags |= WIMLIB_EXTRACT_FLAG_FROM_PIPE;
2936         ret = do_wimlib_extract_image(pwm, image, target,
2937                                       extract_flags, NULL, 0, progress_func);
2938         /* Clean up and return.  */
2939 out_wimlib_free:
2940         wimlib_free(pwm);
2941         return ret;
2942 }
2943
2944 /* API function documented in wimlib.h  */
2945 WIMLIBAPI int
2946 wimlib_extract_image(WIMStruct *wim,
2947                      int image,
2948                      const tchar *target,
2949                      int extract_flags,
2950                      WIMStruct **additional_swms,
2951                      unsigned num_additional_swms,
2952                      wimlib_progress_func_t progress_func)
2953 {
2954         extract_flags &= WIMLIB_EXTRACT_MASK_PUBLIC;
2955         return do_wimlib_extract_image(wim, image, target, extract_flags,
2956                                        additional_swms, num_additional_swms,
2957                                        progress_func);
2958 }