2 * unix_apply.c - Code to apply files from a WIM image on UNIX.
6 * Copyright (C) 2012-2016 Eric Biggers
8 * This file is free software; you can redistribute it and/or modify it under
9 * the terms of the GNU Lesser General Public License as published by the Free
10 * Software Foundation; either version 3 of the License, or (at your option) any
13 * This file is distributed in the hope that it will be useful, but WITHOUT
14 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
15 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
18 * You should have received a copy of the GNU Lesser General Public License
19 * along with this file; if not, see http://www.gnu.org/licenses/.
31 #include <sys/types.h>
34 #include "wimlib/apply.h"
35 #include "wimlib/assert.h"
36 #include "wimlib/blob_table.h"
37 #include "wimlib/dentry.h"
38 #include "wimlib/error.h"
39 #include "wimlib/file_io.h"
40 #include "wimlib/reparse.h"
41 #include "wimlib/timestamp.h"
42 #include "wimlib/unix_data.h"
44 /* We don't require O_NOFOLLOW, but the advantage of having it is that if we
45 * need to extract a file to a location at which there exists a symbolic link,
46 * open(..., O_NOFOLLOW | ...) recognizes the symbolic link rather than
47 * following it and creating the file somewhere else. (Equivalent to
48 * FILE_OPEN_REPARSE_POINT on Windows.) */
54 unix_get_supported_features(const char *target,
55 struct wim_features *supported_features)
57 supported_features->sparse_files = 1;
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 */
68 struct unix_apply_ctx {
69 /* Extract flags, the pointer to the WIMStruct, etc. */
70 struct apply_ctx common;
72 /* Buffers for building extraction paths (allocated). */
73 char *pathbufs[NUM_PATHBUFS];
75 /* Index of next pathbuf to use */
76 unsigned which_pathbuf;
78 /* Currently open file descriptors for extraction */
79 struct filedes open_fds[MAX_OPEN_FILES];
81 /* Number of currently open file descriptors in open_fds, starting from
82 * the beginning of the array. */
83 unsigned num_open_fds;
85 /* For each currently open file, whether we're writing to it in "sparse"
87 bool is_sparse_file[MAX_OPEN_FILES];
89 /* Whether is_sparse_file[] is true for any currently open file */
90 bool any_sparse_files;
92 /* Buffer for reading reparse point data into memory */
93 u8 reparse_data[REPARSE_DATA_MAX_SIZE];
95 /* Pointer to the next byte in @reparse_data to fill */
98 /* Absolute path to the target directory (allocated buffer). Only set
99 * if needed for absolute symbolic link fixups. */
100 char *target_abspath;
102 /* Number of characters in target_abspath. */
103 size_t target_abspath_nchars;
105 /* Number of special files we couldn't create due to EPERM */
106 unsigned long num_special_files_ignored;
109 /* Returns the number of characters needed to represent the path to the
110 * specified @dentry when extracted, not including the null terminator or the
111 * path to the target directory itself. */
113 unix_dentry_path_length(const struct wim_dentry *dentry)
116 const struct wim_dentry *d;
120 len += d->d_extraction_name_nchars + 1;
122 } while (!dentry_is_root(d) && will_extract_dentry(d));
127 /* Returns the maximum number of characters needed to represent the path to any
128 * dentry in @dentry_list when extracted, including the null terminator and the
129 * path to the target directory itself. */
131 unix_compute_path_max(const struct list_head *dentry_list,
132 const struct unix_apply_ctx *ctx)
136 const struct wim_dentry *dentry;
138 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
139 len = unix_dentry_path_length(dentry);
144 /* Account for target and null terminator. */
145 return ctx->common.target_nchars + max + 1;
148 /* Builds and returns the filesystem path to which to extract @dentry.
149 * This cycles through NUM_PATHBUFS different buffers. */
151 unix_build_extraction_path(const struct wim_dentry *dentry,
152 struct unix_apply_ctx *ctx)
156 const struct wim_dentry *d;
158 pathbuf = ctx->pathbufs[ctx->which_pathbuf];
159 ctx->which_pathbuf = (ctx->which_pathbuf + 1) % NUM_PATHBUFS;
161 p = &pathbuf[ctx->common.target_nchars +
162 unix_dentry_path_length(dentry)];
166 p -= d->d_extraction_name_nchars;
167 if (d->d_extraction_name_nchars)
168 memcpy(p, d->d_extraction_name,
169 d->d_extraction_name_nchars);
172 } while (!dentry_is_root(d) && will_extract_dentry(d));
177 /* This causes the next call to unix_build_extraction_path() to use the same
178 * path buffer as the previous call. */
180 unix_reuse_pathbuf(struct unix_apply_ctx *ctx)
182 ctx->which_pathbuf = (ctx->which_pathbuf - 1) % NUM_PATHBUFS;
185 /* Builds and returns the filesystem path to which to extract an unspecified
186 * alias of the @inode. This cycles through NUM_PATHBUFS different buffers. */
188 unix_build_inode_extraction_path(const struct wim_inode *inode,
189 struct unix_apply_ctx *ctx)
191 return unix_build_extraction_path(inode_first_extraction_dentry(inode), ctx);
194 /* Should the specified file be extracted as a directory on UNIX? We extract
195 * the file as a directory if FILE_ATTRIBUTE_DIRECTORY is set and the file does
196 * not have a symlink or junction reparse point. It *may* have a different type
197 * of reparse point. */
199 should_extract_as_directory(const struct wim_inode *inode)
201 return (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) &&
202 !inode_is_symlink(inode);
205 /* Sets the timestamps on a file being extracted.
207 * Either @fd or @path must be specified (not -1 and not NULL, respectively).
210 unix_set_timestamps(int fd, const char *path, u64 atime, u64 mtime)
213 struct timespec times[2];
215 times[0] = wim_timestamp_to_timespec(atime);
216 times[1] = wim_timestamp_to_timespec(mtime);
220 if (fd >= 0 && !futimens(fd, times))
223 #ifdef HAVE_UTIMENSAT
224 if (fd < 0 && !utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW))
228 return WIMLIB_ERR_SET_TIMESTAMPS;
231 struct timeval times[2];
233 times[0] = wim_timestamp_to_timeval(atime);
234 times[1] = wim_timestamp_to_timeval(mtime);
236 if (fd >= 0 && !futimes(fd, times))
238 if (fd < 0 && !lutimes(path, times))
240 return WIMLIB_ERR_SET_TIMESTAMPS;
245 unix_set_owner_and_group(int fd, const char *path, uid_t uid, gid_t gid)
247 if (fd >= 0 && !fchown(fd, uid, gid))
249 if (fd < 0 && !lchown(path, uid, gid))
251 return WIMLIB_ERR_SET_SECURITY;
255 unix_set_mode(int fd, const char *path, mode_t mode)
257 if (fd >= 0 && !fchmod(fd, mode))
259 if (fd < 0 && !chmod(path, mode))
261 return WIMLIB_ERR_SET_SECURITY;
265 * Set metadata on an extracted file.
267 * @fd is an open file descriptor to the extracted file, or -1. @path is the
268 * path to the extracted file, or NULL. If valid, this function uses @fd.
269 * Otherwise, if valid, it uses @path. Otherwise, it calculates the path to one
270 * alias of the extracted file and uses it.
273 unix_set_metadata(int fd, const struct wim_inode *inode,
274 const char *path, struct unix_apply_ctx *ctx)
277 struct wimlib_unix_data unix_data;
280 path = unix_build_inode_extraction_path(inode, ctx);
282 if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA)
283 && inode_get_unix_data(inode, &unix_data))
285 u32 uid = unix_data.uid;
286 u32 gid = unix_data.gid;
287 u32 mode = unix_data.mode;
289 ret = unix_set_owner_and_group(fd, path, uid, gid);
292 path = unix_build_inode_extraction_path(inode, ctx);
293 if (ctx->common.extract_flags &
294 WIMLIB_EXTRACT_FLAG_STRICT_ACLS)
296 ERROR_WITH_ERRNO("Can't set uid=%"PRIu32" and "
297 "gid=%"PRIu32" on \"%s\"",
301 WARNING_WITH_ERRNO("Can't set uid=%"PRIu32" and "
302 "gid=%"PRIu32" on \"%s\"",
308 if (!inode_is_symlink(inode))
309 ret = unix_set_mode(fd, path, mode);
312 path = unix_build_inode_extraction_path(inode, ctx);
313 if (ctx->common.extract_flags &
314 WIMLIB_EXTRACT_FLAG_STRICT_ACLS)
316 ERROR_WITH_ERRNO("Can't set mode=0%"PRIo32" "
317 "on \"%s\"", mode, path);
320 WARNING_WITH_ERRNO("Can't set mode=0%"PRIo32" "
321 "on \"%s\"", mode, path);
326 ret = unix_set_timestamps(fd, path,
327 inode->i_last_access_time,
328 inode->i_last_write_time);
331 path = unix_build_inode_extraction_path(inode, ctx);
332 if (ctx->common.extract_flags &
333 WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS)
335 ERROR_WITH_ERRNO("Can't set timestamps on \"%s\"", path);
338 WARNING_WITH_ERRNO("Can't set timestamps on \"%s\"", path);
344 /* Extract all needed aliases of the @inode, where one alias, corresponding to
345 * @first_dentry, has already been extracted to @first_path. */
347 unix_create_hardlinks(const struct wim_inode *inode,
348 const struct wim_dentry *first_dentry,
349 const char *first_path, struct unix_apply_ctx *ctx)
351 const struct wim_dentry *dentry;
354 inode_for_each_extraction_alias(dentry, inode) {
355 if (dentry == first_dentry)
358 newpath = unix_build_extraction_path(dentry, ctx);
360 if (link(first_path, newpath)) {
361 if (errno == EEXIST && !unlink(newpath))
363 ERROR_WITH_ERRNO("Can't create hard link "
364 "\"%s\" => \"%s\"", newpath, first_path);
365 return WIMLIB_ERR_LINK;
367 unix_reuse_pathbuf(ctx);
372 /* If @dentry represents a directory, create it. */
374 unix_create_if_directory(const struct wim_dentry *dentry,
375 struct unix_apply_ctx *ctx)
380 if (!should_extract_as_directory(dentry->d_inode))
383 path = unix_build_extraction_path(dentry, ctx);
384 if (mkdir(path, 0755) &&
385 /* It's okay if the path already exists, as long as it's a
387 !(errno == EEXIST && !lstat(path, &stbuf) && S_ISDIR(stbuf.st_mode)))
389 ERROR_WITH_ERRNO("Can't create directory \"%s\"", path);
390 return WIMLIB_ERR_MKDIR;
393 return report_file_created(&ctx->common);
396 /* If @dentry represents an empty regular file or a special file, create it, set
397 * its metadata, and create any needed hard links. */
399 unix_extract_if_empty_file(const struct wim_dentry *dentry,
400 struct unix_apply_ctx *ctx)
402 const struct wim_inode *inode;
403 struct wimlib_unix_data unix_data;
407 inode = dentry->d_inode;
409 /* Extract all aliases only when the "first" comes up. */
410 if (dentry != inode_first_extraction_dentry(inode))
413 /* Is this a directory, a symbolic link, or any type of nonempty file?
415 if (should_extract_as_directory(inode) || inode_is_symlink(inode) ||
416 inode_get_blob_for_unnamed_data_stream_resolved(inode))
419 /* Recognize special files in UNIX_DATA mode */
420 if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) &&
421 inode_get_unix_data(inode, &unix_data) &&
422 !S_ISREG(unix_data.mode))
424 path = unix_build_extraction_path(dentry, ctx);
426 if (mknod(path, unix_data.mode, unix_data.rdev)) {
427 if (errno == EPERM) {
428 WARNING_WITH_ERRNO("Can't create special "
429 "file \"%s\"", path);
430 ctx->num_special_files_ignored++;
433 if (errno == EEXIST && !unlink(path))
435 ERROR_WITH_ERRNO("Can't create special file \"%s\"",
437 return WIMLIB_ERR_MKNOD;
439 /* On special files, we can set timestamps immediately because
440 * we don't need to write any data to them. */
441 ret = unix_set_metadata(-1, inode, path, ctx);
445 path = unix_build_extraction_path(dentry, ctx);
447 fd = open(path, O_EXCL | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
449 if (errno == EEXIST && !unlink(path))
451 ERROR_WITH_ERRNO("Can't create regular file \"%s\"", path);
452 return WIMLIB_ERR_OPEN;
454 /* On empty files, we can set timestamps immediately because we
455 * don't need to write any data to them. */
456 ret = unix_set_metadata(fd, inode, path, ctx);
457 if (close(fd) && !ret) {
458 ERROR_WITH_ERRNO("Error closing \"%s\"", path);
459 ret = WIMLIB_ERR_WRITE;
465 ret = unix_create_hardlinks(inode, dentry, path, ctx);
469 return report_file_created(&ctx->common);
473 unix_create_dirs_and_empty_files(const struct list_head *dentry_list,
474 struct unix_apply_ctx *ctx)
476 const struct wim_dentry *dentry;
479 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
480 ret = unix_create_if_directory(dentry, ctx);
484 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
485 ret = unix_extract_if_empty_file(dentry, ctx);
493 unix_count_dentries(const struct list_head *dentry_list,
494 u64 *dir_count_ret, u64 *empty_file_count_ret)
496 const struct wim_dentry *dentry;
498 u64 empty_file_count = 0;
500 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
502 const struct wim_inode *inode = dentry->d_inode;
504 if (should_extract_as_directory(inode))
506 else if ((dentry == inode_first_extraction_dentry(inode)) &&
507 !inode_is_symlink(inode) &&
508 !inode_get_blob_for_unnamed_data_stream_resolved(inode))
512 *dir_count_ret = dir_count;
513 *empty_file_count_ret = empty_file_count;
517 unix_create_symlink(const struct wim_inode *inode, const char *path,
518 size_t rpdatalen, struct unix_apply_ctx *ctx)
520 char target[REPARSE_POINT_MAX_SIZE];
521 struct blob_descriptor blob_override;
524 blob_set_is_located_in_attached_buffer(&blob_override,
525 ctx->reparse_data, rpdatalen);
527 ret = wim_inode_readlink(inode, target, sizeof(target) - 1,
530 ctx->target_abspath_nchars);
531 if (unlikely(ret < 0)) {
533 return WIMLIB_ERR_READLINK;
538 if (symlink(target, path)) {
539 if (errno == EEXIST && !unlink(path))
541 return WIMLIB_ERR_LINK;
547 unix_cleanup_open_fds(struct unix_apply_ctx *ctx, unsigned offset)
549 for (unsigned i = offset; i < ctx->num_open_fds; i++)
550 filedes_close(&ctx->open_fds[i]);
551 ctx->num_open_fds = 0;
552 ctx->any_sparse_files = false;
556 unix_begin_extract_blob_instance(const struct blob_descriptor *blob,
557 const struct wim_inode *inode,
558 const struct wim_inode_stream *strm,
559 struct unix_apply_ctx *ctx)
561 const struct wim_dentry *first_dentry;
562 const char *first_path;
565 if (unlikely(strm->stream_type == STREAM_TYPE_REPARSE_POINT)) {
566 /* On UNIX, symbolic links must be created with symlink(), which
567 * requires that the full link target be available. */
568 if (blob->size > REPARSE_DATA_MAX_SIZE) {
569 ERROR_WITH_ERRNO("Reparse data of \"%s\" has size "
570 "%"PRIu64" bytes (exceeds %u bytes)",
571 inode_any_full_path(inode),
572 blob->size, REPARSE_DATA_MAX_SIZE);
573 return WIMLIB_ERR_INVALID_REPARSE_DATA;
575 ctx->reparse_ptr = ctx->reparse_data;
579 wimlib_assert(stream_is_unnamed_data_stream(strm));
581 /* Unnamed data stream of "regular" file */
583 /* This should be ensured by extract_blob_list() */
584 wimlib_assert(ctx->num_open_fds < MAX_OPEN_FILES);
586 first_dentry = inode_first_extraction_dentry(inode);
587 first_path = unix_build_extraction_path(first_dentry, ctx);
589 fd = open(first_path, O_EXCL | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
591 if (errno == EEXIST && !unlink(first_path))
593 ERROR_WITH_ERRNO("Can't create regular file \"%s\"", first_path);
594 return WIMLIB_ERR_OPEN;
596 if (inode->i_attributes & FILE_ATTRIBUTE_SPARSE_FILE) {
597 ctx->is_sparse_file[ctx->num_open_fds] = true;
598 ctx->any_sparse_files = true;
600 ctx->is_sparse_file[ctx->num_open_fds] = false;
601 #ifdef HAVE_POSIX_FALLOCATE
602 posix_fallocate(fd, 0, blob->size);
605 filedes_init(&ctx->open_fds[ctx->num_open_fds++], fd);
606 return unix_create_hardlinks(inode, first_dentry, first_path, ctx);
609 /* Called when starting to read a blob for extraction */
611 unix_begin_extract_blob(struct blob_descriptor *blob, void *_ctx)
613 struct unix_apply_ctx *ctx = _ctx;
614 const struct blob_extraction_target *targets = blob_extraction_targets(blob);
616 for (u32 i = 0; i < blob->out_refcnt; i++) {
617 int ret = unix_begin_extract_blob_instance(blob,
622 ctx->reparse_ptr = NULL;
623 unix_cleanup_open_fds(ctx, 0);
630 /* Called when the next chunk of a blob has been read for extraction */
632 unix_extract_chunk(const struct blob_descriptor *blob, u64 offset,
633 const void *chunk, size_t size, void *_ctx)
635 struct unix_apply_ctx *ctx = _ctx;
636 const void * const end = chunk + size;
644 * For sparse files, only write nonzero regions. This lets the
645 * filesystem use holes to represent zero regions.
647 for (p = chunk; p != end; p += len, offset += len) {
648 zeroes = maybe_detect_sparse_region(p, end - p, &len,
649 ctx->any_sparse_files);
650 for (i = 0; i < ctx->num_open_fds; i++) {
651 if (!zeroes || !ctx->is_sparse_file[i]) {
652 ret = full_pwrite(&ctx->open_fds[i],
660 if (ctx->reparse_ptr)
661 ctx->reparse_ptr = mempcpy(ctx->reparse_ptr, chunk, size);
665 ERROR_WITH_ERRNO("Error writing data to filesystem");
669 /* Called when a blob has been fully read for extraction */
671 unix_end_extract_blob(struct blob_descriptor *blob, int status, void *_ctx)
673 struct unix_apply_ctx *ctx = _ctx;
676 const struct blob_extraction_target *targets = blob_extraction_targets(blob);
678 ctx->reparse_ptr = NULL;
681 unix_cleanup_open_fds(ctx, 0);
687 for (u32 i = 0; i < blob->out_refcnt; i++) {
688 struct wim_inode *inode = targets[i].inode;
690 if (inode_is_symlink(inode)) {
691 /* We finally have the symlink data, so we can create
695 path = unix_build_inode_extraction_path(inode, ctx);
696 ret = unix_create_symlink(inode, path, blob->size, ctx);
698 ERROR_WITH_ERRNO("Can't create symbolic link "
702 ret = unix_set_metadata(-1, inode, path, ctx);
706 struct filedes *fd = &ctx->open_fds[j];
708 /* If the file is sparse, extend it to its final size. */
709 if (ctx->is_sparse_file[j] && ftruncate(fd->fd, blob->size)) {
710 ERROR_WITH_ERRNO("Error extending \"%s\" to final size",
711 unix_build_inode_extraction_path(inode, ctx));
712 ret = WIMLIB_ERR_WRITE;
716 /* Set metadata on regular file just before closing. */
717 ret = unix_set_metadata(fd->fd, inode, NULL, ctx);
721 if (filedes_close(fd)) {
722 ERROR_WITH_ERRNO("Error closing \"%s\"",
723 unix_build_inode_extraction_path(inode, ctx));
724 ret = WIMLIB_ERR_WRITE;
730 unix_cleanup_open_fds(ctx, j);
735 unix_set_dir_metadata(struct list_head *dentry_list, struct unix_apply_ctx *ctx)
737 const struct wim_dentry *dentry;
740 list_for_each_entry_reverse(dentry, dentry_list, d_extraction_list_node) {
741 if (should_extract_as_directory(dentry->d_inode)) {
742 ret = unix_set_metadata(-1, dentry->d_inode, NULL, ctx);
745 ret = report_file_metadata_applied(&ctx->common);
754 unix_extract(struct list_head *dentry_list, struct apply_ctx *_ctx)
757 struct unix_apply_ctx *ctx = (struct unix_apply_ctx *)_ctx;
760 u64 empty_file_count;
762 /* Compute the maximum path length that will be needed, then allocate
763 * some path buffers. */
764 path_max = unix_compute_path_max(dentry_list, ctx);
766 for (unsigned i = 0; i < NUM_PATHBUFS; i++) {
767 ctx->pathbufs[i] = MALLOC(path_max);
768 if (!ctx->pathbufs[i]) {
769 ret = WIMLIB_ERR_NOMEM;
772 /* Pre-fill the target in each path buffer. We'll just append
773 * the rest of the paths after this. */
774 memcpy(ctx->pathbufs[i],
775 ctx->common.target, ctx->common.target_nchars);
778 /* Extract directories and empty regular files. Directories are needed
779 * because we can't extract any other files until their directories
780 * exist. Empty files are needed because they don't have
781 * representatives in the blob list. */
783 unix_count_dentries(dentry_list, &dir_count, &empty_file_count);
785 ret = start_file_structure_phase(&ctx->common, dir_count + empty_file_count);
789 ret = unix_create_dirs_and_empty_files(dentry_list, ctx);
793 ret = end_file_structure_phase(&ctx->common);
797 /* Get full path to target if needed for absolute symlink fixups. */
798 if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
799 ctx->common.required_features.symlink_reparse_points)
801 ctx->target_abspath = realpath(ctx->common.target, NULL);
802 if (!ctx->target_abspath) {
803 ret = WIMLIB_ERR_NOMEM;
806 ctx->target_abspath_nchars = strlen(ctx->target_abspath);
809 /* Extract nonempty regular files and symbolic links. */
811 struct read_blob_callbacks cbs = {
812 .begin_blob = unix_begin_extract_blob,
813 .continue_blob = unix_extract_chunk,
814 .end_blob = unix_end_extract_blob,
817 ret = extract_blob_list(&ctx->common, &cbs);
822 /* Set directory metadata. We do this last so that we get the right
823 * directory timestamps. */
824 ret = start_file_metadata_phase(&ctx->common, dir_count);
828 ret = unix_set_dir_metadata(dentry_list, ctx);
832 ret = end_file_metadata_phase(&ctx->common);
836 if (ctx->num_special_files_ignored) {
837 WARNING("%lu special files were not extracted due to EPERM!",
838 ctx->num_special_files_ignored);
841 for (unsigned i = 0; i < NUM_PATHBUFS; i++)
842 FREE(ctx->pathbufs[i]);
843 FREE(ctx->target_abspath);
847 const struct apply_operations unix_apply_ops = {
849 .get_supported_features = unix_get_supported_features,
850 .extract = unix_extract,
851 .context_size = sizeof(struct unix_apply_ctx),