2 * reparse.c - Handle reparse data.
6 * Copyright (C) 2012, 2013 Eric Biggers
8 * This file is part of wimlib, a library for working with WIM files.
10 * wimlib is free software; you can redistribute it and/or modify it under the
11 * terms of the GNU General Public License as published by the Free
12 * Software Foundation; either version 3 of the License, or (at your option)
15 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
16 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
17 * A PARTICULAR PURPOSE. See the GNU General Public License for more
20 * You should have received a copy of the GNU General Public License
21 * along with wimlib; if not, see http://www.gnu.org/licenses/.
28 #include "wimlib/assert.h"
29 #include "wimlib/compiler.h"
30 #include "wimlib/endianness.h"
31 #include "wimlib/dentry.h"
32 #include "wimlib/encoding.h"
33 #include "wimlib/error.h"
34 #include "wimlib/lookup_table.h"
35 #include "wimlib/reparse.h"
36 #include "wimlib/resource.h"
39 # include "wimlib/win32.h" /* for win32_get_file_and_vol_ids() */
48 /* On-disk format of a symbolic link (WIM_IO_REPARSE_TAG_SYMLINK) or junction
49 * point (WIM_IO_REPARSE_TAG_MOUNT_POINT) reparse data buffer. */
50 struct reparse_buffer_disk {
54 le16 substitute_name_offset;
55 le16 substitute_name_nbytes;
56 le16 print_name_offset;
57 le16 print_name_nbytes;
61 u8 data[REPARSE_POINT_MAX_SIZE - 20];
62 } _packed_attribute symlink;
64 u8 data[REPARSE_POINT_MAX_SIZE - 16];
65 } _packed_attribute junction;
69 static const utf16lechar volume_junction_prefix[11] = {
83 /* Parse the "substitute name" (link target) from a symbolic link or junction
88 * Non-negative integer:
89 * The name is an absolute symbolic link in one of several formats,
90 * and the return value is the number of UTF-16LE characters that need to
91 * be advanced to reach a simple "absolute" path starting with a backslash
92 * (i.e. skip over \??\ and/or drive letter)
94 * SUBST_NAME_IS_VOLUME_JUNCTION:
95 * The name is a volume junction.
96 * SUBST_NAME_IS_RELATIVE_LINK:
97 * The name is a relative symbolic link.
98 * SUBST_NAME_IS_UNKNOWN:
99 * The name does not appear to be a valid symbolic link, junction,
103 parse_substitute_name(const utf16lechar *substitute_name,
104 u16 substitute_name_nbytes, u32 rptag)
106 u16 substitute_name_nchars = substitute_name_nbytes / 2;
108 if (substitute_name_nchars >= 7 &&
109 substitute_name[0] == cpu_to_le16('\\') &&
110 substitute_name[1] == cpu_to_le16('?') &&
111 substitute_name[2] == cpu_to_le16('?') &&
112 substitute_name[3] == cpu_to_le16('\\') &&
113 substitute_name[4] != cpu_to_le16('\0') &&
114 substitute_name[5] == cpu_to_le16(':') &&
115 substitute_name[6] == cpu_to_le16('\\'))
117 /* "Full" symlink or junction (\??\x:\ prefixed path) */
119 } else if (rptag == WIM_IO_REPARSE_TAG_MOUNT_POINT &&
120 substitute_name_nchars >= 12 &&
121 memcmp(substitute_name, volume_junction_prefix,
122 sizeof(volume_junction_prefix)) == 0 &&
123 substitute_name[substitute_name_nchars - 1] == cpu_to_le16('\\'))
125 /* Volume junction. Can't really do anything with it. */
126 return SUBST_NAME_IS_VOLUME_JUNCTION;
127 } else if (rptag == WIM_IO_REPARSE_TAG_SYMLINK &&
128 substitute_name_nchars >= 3 &&
129 substitute_name[0] != cpu_to_le16('\0') &&
130 substitute_name[1] == cpu_to_le16(':') &&
131 substitute_name[2] == cpu_to_le16('\\'))
133 /* "Absolute" symlink, with drive letter */
135 } else if (rptag == WIM_IO_REPARSE_TAG_SYMLINK &&
136 substitute_name_nchars >= 1)
138 if (substitute_name[0] == cpu_to_le16('\\'))
139 /* "Absolute" symlink, without drive letter */
142 /* "Relative" symlink, without drive letter */
143 return SUBST_NAME_IS_RELATIVE_LINK;
145 return SUBST_NAME_IS_UNKNOWN;
150 * Read the data from a symbolic link, junction, or mount point reparse point
151 * buffer into a `struct reparse_data'.
153 * See http://msdn.microsoft.com/en-us/library/cc232006(v=prot.10).aspx for a
154 * description of the format of the reparse point buffers.
157 parse_reparse_data(const u8 * restrict rpbuf, u16 rpbuflen,
158 struct reparse_data * restrict rpdata)
160 u16 substitute_name_offset;
161 u16 print_name_offset;
162 const struct reparse_buffer_disk *rpbuf_disk =
163 (const struct reparse_buffer_disk*)rpbuf;
166 memset(rpdata, 0, sizeof(*rpdata));
169 rpdata->rptag = le32_to_cpu(rpbuf_disk->rptag);
170 wimlib_assert(rpdata->rptag == WIM_IO_REPARSE_TAG_SYMLINK ||
171 rpdata->rptag == WIM_IO_REPARSE_TAG_MOUNT_POINT);
172 rpdata->rpdatalen = le16_to_cpu(rpbuf_disk->rpdatalen);
173 rpdata->rpreserved = le16_to_cpu(rpbuf_disk->rpreserved);
174 substitute_name_offset = le16_to_cpu(rpbuf_disk->substitute_name_offset);
175 rpdata->substitute_name_nbytes = le16_to_cpu(rpbuf_disk->substitute_name_nbytes);
176 print_name_offset = le16_to_cpu(rpbuf_disk->print_name_offset);
177 rpdata->print_name_nbytes = le16_to_cpu(rpbuf_disk->print_name_nbytes);
179 if ((substitute_name_offset & 1) | (print_name_offset & 1) |
180 (rpdata->substitute_name_nbytes & 1) | (rpdata->print_name_nbytes & 1))
182 /* Names would be unaligned... */
186 if (rpdata->rptag == WIM_IO_REPARSE_TAG_SYMLINK) {
189 rpdata->rpflags = le32_to_cpu(rpbuf_disk->symlink.rpflags);
190 data = rpbuf_disk->symlink.data;
192 data = rpbuf_disk->junction.data;
194 if ((size_t)substitute_name_offset + rpdata->substitute_name_nbytes +
195 (data - rpbuf) > rpbuflen)
197 if ((size_t)print_name_offset + rpdata->print_name_nbytes +
198 (data - rpbuf) > rpbuflen)
200 rpdata->substitute_name = (utf16lechar*)&data[substitute_name_offset];
201 rpdata->print_name = (utf16lechar*)&data[print_name_offset];
204 ERROR("Invalid reparse data");
205 return WIMLIB_ERR_INVALID_REPARSE_DATA;
209 * Create a reparse point data buffer.
211 * @rpdata: Structure that contains the data we need.
213 * @rpbuf: Buffer into which to write the reparse point data buffer. Must be
214 * at least REPARSE_POINT_MAX_SIZE bytes long.
217 make_reparse_buffer(const struct reparse_data * restrict rpdata,
219 u16 * restrict rpbuflen_ret)
221 struct reparse_buffer_disk *rpbuf_disk =
222 (struct reparse_buffer_disk*)rpbuf;
225 rpbuf_disk->rptag = cpu_to_le32(rpdata->rptag);
226 rpbuf_disk->rpreserved = cpu_to_le16(rpdata->rpreserved);
227 rpbuf_disk->substitute_name_offset = cpu_to_le16(0);
228 rpbuf_disk->substitute_name_nbytes = cpu_to_le16(rpdata->substitute_name_nbytes);
229 rpbuf_disk->print_name_offset = cpu_to_le16(rpdata->substitute_name_nbytes + 2);
230 rpbuf_disk->print_name_nbytes = cpu_to_le16(rpdata->print_name_nbytes);
232 if (rpdata->rptag == WIM_IO_REPARSE_TAG_SYMLINK) {
233 rpbuf_disk->symlink.rpflags = cpu_to_le32(rpdata->rpflags);
234 data = rpbuf_disk->symlink.data;
236 data = rpbuf_disk->junction.data;
239 /* We null-terminate the substitute and print names, although this may
240 * not be strictly necessary. Note that the byte counts should not
241 * include the null terminators. */
242 if (data + rpdata->substitute_name_nbytes +
243 rpdata->print_name_nbytes +
244 2 * sizeof(utf16lechar) - rpbuf > REPARSE_POINT_MAX_SIZE)
246 ERROR("Reparse data is too long!");
247 return WIMLIB_ERR_INVALID_REPARSE_DATA;
249 data = mempcpy(data, rpdata->substitute_name, rpdata->substitute_name_nbytes);
250 *(utf16lechar*)data = cpu_to_le16(0);
252 data = mempcpy(data, rpdata->print_name, rpdata->print_name_nbytes);
253 *(utf16lechar*)data = cpu_to_le16(0);
255 rpbuf_disk->rpdatalen = cpu_to_le16(data - rpbuf - 8);
256 *rpbuflen_ret = data - rpbuf;
261 * Read the reparse data from a WIM inode that is a reparse point.
263 * @rpbuf points to a buffer at least REPARSE_POINT_MAX_SIZE bytes into which
264 * the reparse point data buffer will be reconstructed.
266 * Note: in the WIM format, the first 8 bytes of the reparse point data buffer
267 * are omitted, presumably because we already know the reparse tag from the
268 * dentry, and we already know the reparse tag length from the lookup table
269 * entry resource length. However, we reconstruct the first 8 bytes in the
270 * buffer returned by this function.
273 wim_inode_get_reparse_data(const struct wim_inode * restrict inode,
275 u16 * restrict rpbuflen_ret)
277 struct wim_lookup_table_entry *lte;
279 struct reparse_buffer_disk *rpbuf_disk;
282 wimlib_assert(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT);
284 lte = inode_unnamed_lte_resolved(inode);
286 ERROR("Reparse point has no reparse data!");
287 return WIMLIB_ERR_INVALID_REPARSE_DATA;
290 if (wim_resource_size(lte) > REPARSE_POINT_MAX_SIZE - 8) {
291 ERROR("Reparse data is too long!");
292 return WIMLIB_ERR_INVALID_REPARSE_DATA;
294 rpdatalen = wim_resource_size(lte);
296 /* Read the data from the WIM file */
297 ret = read_full_resource_into_buf(lte, rpbuf + 8);
301 /* Reconstruct the first 8 bytes of the reparse point buffer */
302 rpbuf_disk = (struct reparse_buffer_disk*)rpbuf;
305 rpbuf_disk->rptag = cpu_to_le32(inode->i_reparse_tag);
307 /* ReparseDataLength */
308 rpbuf_disk->rpdatalen = cpu_to_le16(rpdatalen);
311 * XXX this could be one of the unknown fields in the WIM dentry. */
312 rpbuf_disk->rpreserved = cpu_to_le16(0);
314 *rpbuflen_ret = rpdatalen + 8;
318 /* UNIX version of getting and setting the data in reparse points */
319 #if !defined(__WIN32__)
321 /* Get the UNIX symlink target from a WIM inode. The inode may be either a
322 * "real" symlink (reparse tag WIM_IO_REPARSE_TAG_SYMLINK), or it may be a
323 * junction point (reparse tag WIM_IO_REPARSE_TAG_MOUNT_POINT).
325 * This has similar semantics to the UNIX readlink() function, except the path
326 * argument is swapped out with the `struct wim_inode' for a reparse point, and
327 * on failure a negated error code is returned rather than -1 with errno set. */
329 wim_inode_readlink(const struct wim_inode * restrict inode,
330 char * restrict buf, size_t bufsize)
333 struct reparse_buffer_disk rpbuf_disk _aligned_attribute(8);
334 struct reparse_data rpdata;
336 char *translated_target;
337 size_t link_target_len;
340 wimlib_assert(inode_is_symlink(inode));
342 if (wim_inode_get_reparse_data(inode, (u8*)&rpbuf_disk, &rpbuflen))
345 if (parse_reparse_data((const u8*)&rpbuf_disk, rpbuflen, &rpdata))
348 ret = utf16le_to_tstr(rpdata.substitute_name,
349 rpdata.substitute_name_nbytes,
350 &link_target, &link_target_len);
354 translated_target = link_target;
355 ret = parse_substitute_name(rpdata.substitute_name,
356 rpdata.substitute_name_nbytes,
359 case SUBST_NAME_IS_RELATIVE_LINK:
360 goto out_translate_slashes;
361 case SUBST_NAME_IS_VOLUME_JUNCTION:
363 case SUBST_NAME_IS_UNKNOWN:
364 ERROR("Can't understand reparse point "
365 "substitute name \"%s\"", link_target);
367 goto out_free_link_target;
369 translated_target += ret;
370 link_target_len -= ret;
374 out_translate_slashes:
375 for (size_t i = 0; i < link_target_len; i++)
376 if (translated_target[i] == '\\')
377 translated_target[i] = '/';
379 if (link_target_len > bufsize) {
380 link_target_len = bufsize;
383 ret = link_target_len;
385 memcpy(buf, translated_target, link_target_len);
386 out_free_link_target:
392 wim_inode_set_symlink(struct wim_inode *inode,
394 struct wim_lookup_table *lookup_table)
397 struct reparse_buffer_disk rpbuf_disk _aligned_attribute(8);
398 struct reparse_data rpdata;
399 static const char abs_subst_name_prefix[12] = "\\\0?\0?\0\\\0C\0:\0";
400 static const char abs_print_name_prefix[4] = "C\0:\0";
401 utf16lechar *name_utf16le;
402 size_t name_utf16le_nbytes;
406 DEBUG("Creating reparse point data buffer for UNIX "
407 "symlink target \"%s\"", target);
408 memset(&rpdata, 0, sizeof(rpdata));
409 ret = tstr_to_utf16le(target, strlen(target),
410 &name_utf16le, &name_utf16le_nbytes);
414 for (size_t i = 0; i < name_utf16le_nbytes / 2; i++)
415 if (name_utf16le[i] == cpu_to_le16('/'))
416 name_utf16le[i] = cpu_to_le16('\\');
418 /* Compatability notes:
420 * On UNIX, an absolute symbolic link begins with '/'; everything else
421 * is a relative symbolic link. (Quite simple compared to the various
422 * ways to provide Windows paths.)
424 * To change a UNIX relative symbolic link to Windows format, we only
425 * need to translate it to UTF-16LE and replace backslashes with forward
426 * slashes. We do not make any attempt to handle filename character
427 * problems, such as a link target that itself contains backslashes on
428 * UNIX. Then, for these relative links, we set the reparse header
429 * @flags field to SYMBOLIC_LINK_RELATIVE.
431 * For UNIX absolute symbolic links, we must set the @flags field to 0.
432 * Then, there are multiple options as to actually represent the
433 * absolute link targets:
435 * (1) An absolute path beginning with one backslash character. similar
436 * to UNIX-style, just with a different path separator. Print name same
437 * as substitute name.
439 * (2) Absolute path beginning with drive letter followed by a
440 * backslash. Print name same as substitute name.
442 * (3) Absolute path beginning with drive letter followed by a
443 * backslash; substitute name prefixed with \??\, otherwise same as
446 * We choose option (3) here, and we just assume C: for the drive
447 * letter. The reasoning for this is:
449 * (1) Microsoft imagex.exe has a bug where it does not attempt to do
450 * reparse point fixups for these links, even though they are valid
451 * absolute links. (Note: in this case prefixing the substitute name
452 * with \??\ does not work; it just makes the data unable to be restored
454 * (2) Microsoft imagex.exe will fail when doing reparse point fixups
455 * for these. It apparently contains a bug that causes it to create an
456 * invalid reparse point, which then cannot be restored.
457 * (3) This is the only option I tested for which reparse point fixups
458 * worked properly in Microsoft imagex.exe.
460 * So option (3) it is.
463 rpdata.rptag = inode->i_reparse_tag;
464 if (target[0] == '/') {
465 rpdata.substitute_name_nbytes = name_utf16le_nbytes +
466 sizeof(abs_subst_name_prefix);
467 rpdata.print_name_nbytes = name_utf16le_nbytes +
468 sizeof(abs_print_name_prefix);
469 rpdata.substitute_name = alloca(rpdata.substitute_name_nbytes);
470 rpdata.print_name = alloca(rpdata.print_name_nbytes);
471 memcpy(rpdata.substitute_name, abs_subst_name_prefix,
472 sizeof(abs_subst_name_prefix));
473 memcpy(rpdata.print_name, abs_print_name_prefix,
474 sizeof(abs_print_name_prefix));
475 memcpy((void*)rpdata.substitute_name + sizeof(abs_subst_name_prefix),
476 name_utf16le, name_utf16le_nbytes);
477 memcpy((void*)rpdata.print_name + sizeof(abs_print_name_prefix),
478 name_utf16le, name_utf16le_nbytes);
480 rpdata.substitute_name_nbytes = name_utf16le_nbytes;
481 rpdata.print_name_nbytes = name_utf16le_nbytes;
482 rpdata.substitute_name = name_utf16le;
483 rpdata.print_name = name_utf16le;
484 rpdata.rpflags = SYMBOLIC_LINK_RELATIVE;
487 ret = make_reparse_buffer(&rpdata, (u8*)&rpbuf_disk, &rpbuflen);
489 ret = inode_set_unnamed_stream(inode,
490 (u8*)&rpbuf_disk + 8,
498 #include <sys/stat.h>
501 unix_get_ino_and_dev(const char *path, u64 *ino_ret, u64 *dev_ret)
504 if (stat(path, &stbuf)) {
506 WARNING_WITH_ERRNO("Failed to stat \"%s\"", path);
507 /* Treat as a link pointing outside the capture root (it
508 * most likely is). */
509 return WIMLIB_ERR_STAT;
511 *ino_ret = stbuf.st_ino;
512 *dev_ret = stbuf.st_dev;
517 #endif /* !defined(__WIN32__) */
519 /* is_rp_path_separator() - characters treated as path separators in absolute
520 * symbolic link targets */
523 # define is_rp_path_separator(c) ((c) == L'\\' || (c) == L'/')
524 # define os_get_ino_and_dev win32_get_file_and_vol_ids
526 # define is_rp_path_separator(c) ((c) == '/')
527 # define os_get_ino_and_dev unix_get_ino_and_dev
530 /* Fix up absolute symbolic link targets--- mostly shared between UNIX and
533 capture_fixup_absolute_symlink(tchar *dest,
534 u64 capture_root_ino, u64 capture_root_dev)
539 /* Skip drive letter */
540 if (!is_rp_path_separator(*dest))
544 DEBUG("Fixing symlink or junction \"%"TS"\"", dest);
551 while (is_rp_path_separator(*p))
556 ret = os_get_ino_and_dev(dest, &ino, &dev);
559 if (ret) /* stat() failed before we got to the capture root---
560 assume the link points outside it. */
563 if (ino == capture_root_ino && dev == capture_root_dev) {
564 /* Link points inside capture root. Return abbreviated
567 *(p - 1) = OS_PREFERRED_PATH_SEPARATOR;
568 while (p - 1 >= dest && is_rp_path_separator(*(p - 1)))
571 if (!is_rp_path_separator(dest[0])) {
576 wimlib_assert(p >= dest);
581 /* Link points outside capture root. */
587 } while (!is_rp_path_separator(*p) && *p != T('\0'));