2 * unix_apply.c - Code to apply files from a WIM image on UNIX.
6 * Copyright (C) 2012, 2013, 2014 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/apply.h"
29 #include "wimlib/dentry.h"
30 #include "wimlib/error.h"
31 #include "wimlib/file_io.h"
32 #include "wimlib/reparse.h"
33 #include "wimlib/timestamp.h"
34 #include "wimlib/unix_data.h"
42 #include <sys/types.h>
45 /* We don't require O_NOFOLLOW, but the advantage of having it is that if we
46 * need to extract a file to a location at which there exists a symbolic link,
47 * open(..., O_NOFOLLOW | ...) recognizes the symbolic link rather than
48 * following it and creating the file somewhere else. (Equivalent to
49 * FILE_OPEN_REPARSE_POINT on Windows.) */
55 unix_get_supported_features(const char *target,
56 struct wim_features *supported_features)
58 supported_features->hard_links = 1;
59 supported_features->symlink_reparse_points = 1;
60 supported_features->unix_data = 1;
61 supported_features->timestamps = 1;
62 supported_features->case_sensitive_filenames = 1;
66 #define NUM_PATHBUFS 2 /* We need 2 when creating hard links */
67 #define MAX_OPEN_FDS 1024 /* TODO: Add special case for when the number of
68 identical streams exceeds this number. */
70 struct unix_apply_ctx {
71 /* Extract flags, the pointer to the WIMStruct, etc. */
72 struct apply_ctx common;
74 /* Buffers for building extraction paths (allocated). */
75 char *pathbufs[NUM_PATHBUFS];
77 /* Index of next pathbuf to use */
78 unsigned which_pathbuf;
80 /* Currently open file descriptors for extraction */
81 struct filedes open_fds[MAX_OPEN_FDS];
83 /* Number of currently open file descriptors in open_fds, starting from
84 * the beginning of the array. */
85 unsigned num_open_fds;
87 /* Buffer for reading reparse data streams into memory */
88 u8 reparse_data[REPARSE_DATA_MAX_SIZE];
90 /* Pointer to the next byte in @reparse_data to fill */
93 /* Absolute path to the target directory (allocated buffer). Only set
94 * if needed for absolute symbolic link fixups. */
97 /* Number of characters in target_abspath. */
98 size_t target_abspath_nchars;
101 /* Returns the number of characters needed to represent the path to the
102 * specified @dentry when extracted, not including the null terminator or the
103 * path to the target directory itself. */
105 unix_dentry_path_length(const struct wim_dentry *dentry)
108 const struct wim_dentry *d;
112 len += d->d_extraction_name_nchars + 1;
114 } while (!dentry_is_root(d) && will_extract_dentry(d));
119 /* Returns the maximum number of characters needed to represent the path to any
120 * dentry in @dentry_list when extracted, including the null terminator and the
121 * path to the target directory itself. */
123 unix_compute_path_max(const struct list_head *dentry_list,
124 const struct unix_apply_ctx *ctx)
128 const struct wim_dentry *dentry;
130 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
131 len = unix_dentry_path_length(dentry);
136 /* Account for target and null terminator. */
137 return ctx->common.target_nchars + max + 1;
140 /* Builds and returns the filesystem path to which to extract @dentry.
141 * This cycles through NUM_PATHBUFS different buffers. */
143 unix_build_extraction_path(const struct wim_dentry *dentry,
144 struct unix_apply_ctx *ctx)
148 const struct wim_dentry *d;
150 pathbuf = ctx->pathbufs[ctx->which_pathbuf];
151 ctx->which_pathbuf = (ctx->which_pathbuf + 1) % NUM_PATHBUFS;
153 p = &pathbuf[ctx->common.target_nchars +
154 unix_dentry_path_length(dentry)];
158 p -= d->d_extraction_name_nchars;
159 memcpy(p, d->d_extraction_name, d->d_extraction_name_nchars);
162 } while (!dentry_is_root(d) && will_extract_dentry(d));
167 /* This causes the next call to unix_build_extraction_path() to use the same
168 * path buffer as the previous call. */
170 unix_reuse_pathbuf(struct unix_apply_ctx *ctx)
172 ctx->which_pathbuf = (ctx->which_pathbuf - 1) % NUM_PATHBUFS;
175 /* Builds and returns the filesystem path to which to extract an unspecified
176 * alias of the @inode. This cycles through NUM_PATHBUFS different buffers. */
178 unix_build_inode_extraction_path(const struct wim_inode *inode,
179 struct unix_apply_ctx *ctx)
181 return unix_build_extraction_path(inode_first_extraction_dentry(inode), ctx);
184 /* Sets the timestamps on a file being extracted.
186 * Either @fd or @path must be specified (not -1 and not NULL, respectively).
189 unix_set_timestamps(int fd, const char *path, u64 atime, u64 mtime)
192 struct timespec times[2];
194 times[0] = wim_timestamp_to_timespec(atime);
195 times[1] = wim_timestamp_to_timespec(mtime);
199 if (fd >= 0 && !futimens(fd, times))
202 #ifdef HAVE_UTIMENSAT
203 if (fd < 0 && !utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW))
207 return WIMLIB_ERR_SET_TIMESTAMPS;
210 struct timeval times[2];
212 times[0] = wim_timestamp_to_timeval(atime);
213 times[1] = wim_timestamp_to_timeval(mtime);
215 if (fd >= 0 && !futimes(fd, times))
217 if (fd < 0 && !lutimes(path, times))
219 return WIMLIB_ERR_SET_TIMESTAMPS;
224 unix_set_owner_and_group(int fd, const char *path, uid_t uid, gid_t gid)
226 if (fd >= 0 && !fchown(fd, uid, gid))
228 if (fd < 0 && !lchown(path, uid, gid))
230 return WIMLIB_ERR_SET_SECURITY;
234 unix_set_mode(int fd, const char *path, mode_t mode)
236 if (fd >= 0 && !fchmod(fd, mode))
238 if (fd < 0 && !chmod(path, mode))
240 return WIMLIB_ERR_SET_SECURITY;
244 * Set metadata on an extracted file.
246 * @fd is an open file descriptor to the extracted file, or -1. @path is the
247 * path to the extracted file, or NULL. If valid, this function uses @fd.
248 * Otherwise, if valid, it uses @path. Otherwise, it calculates the path to one
249 * alias of the extracted file and uses it.
252 unix_set_metadata(int fd, const struct wim_inode *inode,
253 const char *path, struct unix_apply_ctx *ctx)
256 struct wimlib_unix_data unix_data;
259 path = unix_build_inode_extraction_path(inode, ctx);
261 if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA)
262 && inode_get_unix_data(inode, &unix_data))
264 u32 uid = unix_data.uid;
265 u32 gid = unix_data.gid;
266 u32 mode = unix_data.mode;
268 ret = unix_set_owner_and_group(fd, path, uid, gid);
271 path = unix_build_inode_extraction_path(inode, ctx);
272 if (ctx->common.extract_flags &
273 WIMLIB_EXTRACT_FLAG_STRICT_ACLS)
275 ERROR_WITH_ERRNO("Can't set uid=%"PRIu32" and "
276 "gid=%"PRIu32" on \"%s\"",
280 WARNING_WITH_ERRNO("Can't set uid=%"PRIu32" and "
281 "gid=%"PRIu32" on \"%s\"",
287 if (!inode_is_symlink(inode))
288 ret = unix_set_mode(fd, path, mode);
291 path = unix_build_inode_extraction_path(inode, ctx);
292 if (ctx->common.extract_flags &
293 WIMLIB_EXTRACT_FLAG_STRICT_ACLS)
295 ERROR_WITH_ERRNO("Can't set mode=0%"PRIo32" "
296 "on \"%s\"", mode, path);
299 WARNING_WITH_ERRNO("Can't set mode=0%"PRIo32" "
300 "on \"%s\"", mode, path);
305 ret = unix_set_timestamps(fd, path,
306 inode->i_last_access_time,
307 inode->i_last_write_time);
310 path = unix_build_inode_extraction_path(inode, ctx);
311 if (ctx->common.extract_flags &
312 WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS)
314 ERROR_WITH_ERRNO("Can't set timestamps on \"%s\"", path);
317 WARNING_WITH_ERRNO("Can't set timestamps on \"%s\"", path);
323 /* Extract all needed aliases of the @inode, where one alias, corresponding to
324 * @first_dentry, has already been extracted to @first_path. */
326 unix_create_hardlinks(const struct wim_inode *inode,
327 const struct wim_dentry *first_dentry,
328 const char *first_path, struct unix_apply_ctx *ctx)
330 const struct wim_dentry *dentry;
333 list_for_each_entry(dentry, &inode->i_extraction_aliases,
334 d_extraction_alias_node)
336 if (dentry == first_dentry)
339 newpath = unix_build_extraction_path(dentry, ctx);
341 if (link(first_path, newpath)) {
342 if (errno == EEXIST && !unlink(newpath))
344 ERROR_WITH_ERRNO("Can't create hard link "
345 "\"%s\" => \"%s\"", newpath, first_path);
346 return WIMLIB_ERR_LINK;
348 unix_reuse_pathbuf(ctx);
353 /* If @dentry represents a directory, create it. */
355 unix_create_if_directory(const struct wim_dentry *dentry,
356 struct unix_apply_ctx *ctx)
361 if (!dentry_is_directory(dentry))
364 path = unix_build_extraction_path(dentry, ctx);
365 if (mkdir(path, 0755) &&
366 /* It's okay if the path already exists, as long as it's a
368 !(errno == EEXIST && !lstat(path, &stbuf) && S_ISDIR(stbuf.st_mode)))
370 ERROR_WITH_ERRNO("Can't create directory \"%s\"", path);
371 return WIMLIB_ERR_MKDIR;
376 /* If @dentry represents an empty regular file, create it, set its metadata, and
377 * create any needed hard links. */
379 unix_extract_if_empty_file(const struct wim_dentry *dentry,
380 struct unix_apply_ctx *ctx)
382 const struct wim_inode *inode;
387 inode = dentry->d_inode;
389 /* Extract all aliases only when the "first" comes up. */
390 if (dentry != inode_first_extraction_dentry(inode))
393 /* Not an empty regular file? */
394 if (inode_is_directory(inode) || inode_is_symlink(inode) ||
395 inode_unnamed_lte_resolved(inode))
398 path = unix_build_extraction_path(dentry, ctx);
400 fd = open(path, O_TRUNC | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
402 if (errno == EEXIST && !unlink(path))
404 ERROR_WITH_ERRNO("Can't create regular file \"%s\"", path);
405 return WIMLIB_ERR_OPEN;
407 /* On empty files, we can set timestamps immediately because we don't
408 * need to write any data to them. */
409 ret = unix_set_metadata(fd, inode, path, ctx);
410 if (close(fd) && !ret) {
411 ERROR_WITH_ERRNO("Error closing \"%s\"", path);
412 ret = WIMLIB_ERR_WRITE;
417 return unix_create_hardlinks(inode, dentry, path, ctx);
421 unix_create_dirs_and_empty_files(const struct list_head *dentry_list,
422 struct unix_apply_ctx *ctx)
424 const struct wim_dentry *dentry;
427 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
428 ret = unix_create_if_directory(dentry, ctx);
432 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
433 ret = unix_extract_if_empty_file(dentry, ctx);
441 unix_create_symlink(const struct wim_inode *inode, const char *path,
442 const u8 *rpdata, u16 rpdatalen, bool rpfix,
443 const char *apply_dir, size_t apply_dir_nchars)
445 char link_target[REPARSE_DATA_MAX_SIZE];
447 struct wim_lookup_table_entry lte_override;
449 lte_override.resource_location = RESOURCE_IN_ATTACHED_BUFFER;
450 lte_override.attached_buffer = (void *)rpdata;
451 lte_override.size = rpdatalen;
453 ret = wim_inode_readlink(inode, link_target,
454 sizeof(link_target) - 1, <e_override);
457 return WIMLIB_ERR_READLINK;
460 link_target[ret] = 0;
462 if (rpfix && link_target[0] == '/') {
464 /* "Fix" the absolute symbolic link by prepending the absolute
465 * path to the target directory. */
467 if (sizeof(link_target) - (ret + 1) < apply_dir_nchars) {
468 errno = ENAMETOOLONG;
469 return WIMLIB_ERR_REPARSE_POINT_FIXUP_FAILED;
471 memmove(link_target + apply_dir_nchars, link_target,
473 memcpy(link_target, apply_dir, apply_dir_nchars);
476 if (symlink(link_target, path)) {
477 if (errno == EEXIST && !unlink(path))
479 return WIMLIB_ERR_LINK;
485 unix_cleanup_open_fds(struct unix_apply_ctx *ctx, unsigned offset)
487 for (unsigned i = offset; i < ctx->num_open_fds; i++)
488 filedes_close(&ctx->open_fds[i]);
489 ctx->num_open_fds = 0;
493 unix_begin_extract_stream_instance(const struct wim_lookup_table_entry *stream,
494 const struct wim_inode *inode,
495 struct unix_apply_ctx *ctx)
497 const struct wim_dentry *first_dentry;
498 const char *first_path;
501 if (inode_is_symlink(inode)) {
502 /* On UNIX, symbolic links must be created with symlink(), which
503 * requires that the full link target be available. */
504 if (stream->size > REPARSE_DATA_MAX_SIZE) {
505 ERROR_WITH_ERRNO("Reparse data of \"%s\" has size "
506 "%"PRIu64" bytes (exceeds %u bytes)",
507 inode_first_full_path(inode),
508 stream->size, REPARSE_DATA_MAX_SIZE);
509 return WIMLIB_ERR_INVALID_REPARSE_DATA;
511 ctx->reparse_ptr = ctx->reparse_data;
515 first_dentry = inode_first_extraction_dentry(inode);
516 first_path = unix_build_extraction_path(first_dentry, ctx);
518 fd = open(first_path, O_TRUNC | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
520 if (errno == EEXIST && !unlink(first_path))
522 ERROR_WITH_ERRNO("Can't create regular file \"%s\"", first_path);
523 return WIMLIB_ERR_OPEN;
525 filedes_init(&ctx->open_fds[ctx->num_open_fds++], fd);
526 return unix_create_hardlinks(inode, first_dentry, first_path, ctx);
529 /* Called when starting to read a single-instance stream for extraction */
531 unix_begin_extract_stream(struct wim_lookup_table_entry *stream,
532 u32 flags, void *_ctx)
534 struct unix_apply_ctx *ctx = _ctx;
535 const struct stream_owner *owners = stream_owners(stream);
538 for (u32 i = 0; i < stream->out_refcnt; i++) {
539 const struct wim_inode *inode = owners[i].inode;
541 ret = unix_begin_extract_stream_instance(stream, inode, ctx);
543 ctx->reparse_ptr = NULL;
544 unix_cleanup_open_fds(ctx, 0);
551 /* Called when the next chunk of a single-instance stream has been read for
554 unix_extract_chunk(const void *chunk, size_t size, void *_ctx)
556 struct unix_apply_ctx *ctx = _ctx;
559 for (unsigned i = 0; i < ctx->num_open_fds; i++) {
560 ret = full_write(&ctx->open_fds[i], chunk, size);
562 ERROR_WITH_ERRNO("Error writing data to filesystem");
566 if (ctx->reparse_ptr)
567 ctx->reparse_ptr = mempcpy(ctx->reparse_ptr, chunk, size);
571 /* Called when a single-instance stream has been fully read for extraction */
573 unix_end_extract_stream(struct wim_lookup_table_entry *stream, int status,
576 struct unix_apply_ctx *ctx = _ctx;
579 const struct stream_owner *owners = stream_owners(stream);
581 ctx->reparse_ptr = NULL;
584 unix_cleanup_open_fds(ctx, 0);
590 for (u32 i = 0; i < stream->out_refcnt; i++) {
591 struct wim_inode *inode = owners[i].inode;
593 if (inode_is_symlink(inode)) {
594 /* We finally have the symlink data, so we can create
598 path = unix_build_inode_extraction_path(inode, ctx);
599 ret = unix_create_symlink(inode, path,
602 (ctx->common.extract_flags &
603 WIMLIB_EXTRACT_FLAG_RPFIX),
605 ctx->target_abspath_nchars);
607 ERROR_WITH_ERRNO("Can't create symbolic link "
611 ret = unix_set_metadata(-1, inode, path, ctx);
615 /* Set metadata on regular file just before closing it.
617 struct filedes *fd = &ctx->open_fds[j];
619 ret = unix_set_metadata(fd->fd, inode, NULL, ctx);
623 if (filedes_close(fd)) {
624 ERROR_WITH_ERRNO("Error closing \"%s\"",
625 unix_build_inode_extraction_path(inode, ctx));
626 ret = WIMLIB_ERR_WRITE;
632 unix_cleanup_open_fds(ctx, j);
637 unix_set_dir_metadata(struct list_head *dentry_list, struct unix_apply_ctx *ctx)
639 const struct wim_dentry *dentry;
642 list_for_each_entry_reverse(dentry, dentry_list, d_extraction_list_node) {
643 if (dentry_is_directory(dentry)) {
644 ret = unix_set_metadata(-1, dentry->d_inode, NULL, ctx);
653 unix_extract(struct list_head *dentry_list, struct apply_ctx *_ctx)
656 struct unix_apply_ctx *ctx = (struct unix_apply_ctx *)_ctx;
659 /* Compute the maximum path length that will be needed, then allocate
660 * some path buffers. */
661 path_max = unix_compute_path_max(dentry_list, ctx);
663 for (unsigned i = 0; i < NUM_PATHBUFS; i++) {
664 ctx->pathbufs[i] = MALLOC(path_max);
665 if (!ctx->pathbufs[i]) {
666 ret = WIMLIB_ERR_NOMEM;
669 /* Pre-fill the target in each path buffer. We'll just append
670 * the rest of the paths after this. */
671 memcpy(ctx->pathbufs[i],
672 ctx->common.target, ctx->common.target_nchars);
675 /* Extract directories and empty regular files. Directories are needed
676 * because we can't extract any other files until their directories
677 * exist. Empty files are needed because they don't have
678 * representatives in the stream list. */
679 ret = unix_create_dirs_and_empty_files(dentry_list, ctx);
683 /* Get full path to target if needed for absolute symlink fixups. */
684 if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
685 ctx->common.required_features.symlink_reparse_points)
687 ctx->target_abspath = realpath(ctx->common.target, NULL);
688 if (!ctx->target_abspath) {
689 ret = WIMLIB_ERR_NOMEM;
692 ctx->target_abspath_nchars = strlen(ctx->target_abspath);
695 /* Extract nonempty regular files and symbolic links. */
697 struct read_stream_list_callbacks cbs = {
698 .begin_stream = unix_begin_extract_stream,
699 .begin_stream_ctx = ctx,
700 .consume_chunk = unix_extract_chunk,
701 .consume_chunk_ctx = ctx,
702 .end_stream = unix_end_extract_stream,
703 .end_stream_ctx = ctx,
705 ret = extract_stream_list(&ctx->common, &cbs);
709 /* Set directory metadata. We do this last so that we get the right
710 * directory timestamps. */
711 ret = unix_set_dir_metadata(dentry_list, ctx);
713 for (unsigned i = 0; i < NUM_PATHBUFS; i++)
714 FREE(ctx->pathbufs[i]);
715 FREE(ctx->target_abspath);
719 const struct apply_operations unix_apply_ops = {
721 .get_supported_features = unix_get_supported_features,
722 .extract = unix_extract,
723 .context_size = sizeof(struct unix_apply_ctx),