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 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->hard_links = 1;
58 supported_features->symlink_reparse_points = 1;
59 supported_features->unix_data = 1;
60 supported_features->timestamps = 1;
61 supported_features->case_sensitive_filenames = 1;
65 #define NUM_PATHBUFS 2 /* We need 2 when creating hard links */
67 struct unix_apply_ctx {
68 /* Extract flags, the pointer to the WIMStruct, etc. */
69 struct apply_ctx common;
71 /* Buffers for building extraction paths (allocated). */
72 char *pathbufs[NUM_PATHBUFS];
74 /* Index of next pathbuf to use */
75 unsigned which_pathbuf;
77 /* Currently open file descriptors for extraction */
78 struct filedes open_fds[MAX_OPEN_FILES];
80 /* Number of currently open file descriptors in open_fds, starting from
81 * the beginning of the array. */
82 unsigned num_open_fds;
84 /* Buffer for reading reparse point data into memory */
85 u8 reparse_data[REPARSE_DATA_MAX_SIZE];
87 /* Pointer to the next byte in @reparse_data to fill */
90 /* Absolute path to the target directory (allocated buffer). Only set
91 * if needed for absolute symbolic link fixups. */
94 /* Number of characters in target_abspath. */
95 size_t target_abspath_nchars;
97 /* Number of special files we couldn't create due to EPERM */
98 unsigned long num_special_files_ignored;
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 inode_for_each_extraction_alias(dentry, inode) {
334 if (dentry == first_dentry)
337 newpath = unix_build_extraction_path(dentry, ctx);
339 if (link(first_path, newpath)) {
340 if (errno == EEXIST && !unlink(newpath))
342 ERROR_WITH_ERRNO("Can't create hard link "
343 "\"%s\" => \"%s\"", newpath, first_path);
344 return WIMLIB_ERR_LINK;
346 unix_reuse_pathbuf(ctx);
351 /* If @dentry represents a directory, create it. */
353 unix_create_if_directory(const struct wim_dentry *dentry,
354 struct unix_apply_ctx *ctx)
359 if (!dentry_is_directory(dentry))
362 path = unix_build_extraction_path(dentry, ctx);
363 if (mkdir(path, 0755) &&
364 /* It's okay if the path already exists, as long as it's a
366 !(errno == EEXIST && !lstat(path, &stbuf) && S_ISDIR(stbuf.st_mode)))
368 ERROR_WITH_ERRNO("Can't create directory \"%s\"", path);
369 return WIMLIB_ERR_MKDIR;
372 return report_file_created(&ctx->common);
375 /* If @dentry represents an empty regular file or a special file, create it, set
376 * its metadata, and create any needed hard links. */
378 unix_extract_if_empty_file(const struct wim_dentry *dentry,
379 struct unix_apply_ctx *ctx)
381 const struct wim_inode *inode;
382 struct wimlib_unix_data unix_data;
386 inode = dentry->d_inode;
388 /* Extract all aliases only when the "first" comes up. */
389 if (dentry != inode_first_extraction_dentry(inode))
392 /* Is this a directory, a symbolic link, or any type of nonempty file?
394 if (inode_is_directory(inode) || inode_is_symlink(inode) ||
395 inode_get_blob_for_unnamed_data_stream_resolved(inode))
398 /* Recognize special files in UNIX_DATA mode */
399 if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) &&
400 inode_get_unix_data(inode, &unix_data) &&
401 !S_ISREG(unix_data.mode))
403 path = unix_build_extraction_path(dentry, ctx);
405 if (mknod(path, unix_data.mode, unix_data.rdev)) {
406 if (errno == EPERM) {
407 WARNING_WITH_ERRNO("Can't create special "
408 "file \"%s\"", path);
409 ctx->num_special_files_ignored++;
412 if (errno == EEXIST && !unlink(path))
414 ERROR_WITH_ERRNO("Can't create special file \"%s\"",
416 return WIMLIB_ERR_MKNOD;
418 /* On special files, we can set timestamps immediately because
419 * we don't need to write any data to them. */
420 ret = unix_set_metadata(-1, inode, path, ctx);
424 path = unix_build_extraction_path(dentry, ctx);
426 fd = open(path, O_TRUNC | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
428 if (errno == EEXIST && !unlink(path))
430 ERROR_WITH_ERRNO("Can't create regular file \"%s\"", path);
431 return WIMLIB_ERR_OPEN;
433 /* On empty files, we can set timestamps immediately because we
434 * don't need to write any data to them. */
435 ret = unix_set_metadata(fd, inode, path, ctx);
436 if (close(fd) && !ret) {
437 ERROR_WITH_ERRNO("Error closing \"%s\"", path);
438 ret = WIMLIB_ERR_WRITE;
444 ret = unix_create_hardlinks(inode, dentry, path, ctx);
448 return report_file_created(&ctx->common);
452 unix_create_dirs_and_empty_files(const struct list_head *dentry_list,
453 struct unix_apply_ctx *ctx)
455 const struct wim_dentry *dentry;
458 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
459 ret = unix_create_if_directory(dentry, ctx);
463 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
464 ret = unix_extract_if_empty_file(dentry, ctx);
472 unix_count_dentries(const struct list_head *dentry_list,
473 u64 *dir_count_ret, u64 *empty_file_count_ret)
475 const struct wim_dentry *dentry;
477 u64 empty_file_count = 0;
479 list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
481 const struct wim_inode *inode = dentry->d_inode;
483 if (inode_is_directory(inode))
485 else if ((dentry == inode_first_extraction_dentry(inode)) &&
486 !inode_is_symlink(inode) &&
487 !inode_get_blob_for_unnamed_data_stream_resolved(inode))
491 *dir_count_ret = dir_count;
492 *empty_file_count_ret = empty_file_count;
496 unix_create_symlink(const struct wim_inode *inode, const char *path,
497 size_t rpdatalen, struct unix_apply_ctx *ctx)
499 char target[REPARSE_POINT_MAX_SIZE];
500 struct blob_descriptor blob_override;
503 blob_set_is_located_in_attached_buffer(&blob_override,
504 ctx->reparse_data, rpdatalen);
506 ret = wim_inode_readlink(inode, target, sizeof(target) - 1,
509 ctx->target_abspath_nchars);
510 if (unlikely(ret < 0)) {
512 return WIMLIB_ERR_READLINK;
517 if (symlink(target, path)) {
518 if (errno == EEXIST && !unlink(path))
520 return WIMLIB_ERR_LINK;
526 unix_cleanup_open_fds(struct unix_apply_ctx *ctx, unsigned offset)
528 for (unsigned i = offset; i < ctx->num_open_fds; i++)
529 filedes_close(&ctx->open_fds[i]);
530 ctx->num_open_fds = 0;
534 unix_begin_extract_blob_instance(const struct blob_descriptor *blob,
535 const struct wim_inode *inode,
536 const struct wim_inode_stream *strm,
537 struct unix_apply_ctx *ctx)
539 const struct wim_dentry *first_dentry;
540 const char *first_path;
543 if (unlikely(strm->stream_type == STREAM_TYPE_REPARSE_POINT)) {
544 /* On UNIX, symbolic links must be created with symlink(), which
545 * requires that the full link target be available. */
546 if (blob->size > REPARSE_DATA_MAX_SIZE) {
547 ERROR_WITH_ERRNO("Reparse data of \"%s\" has size "
548 "%"PRIu64" bytes (exceeds %u bytes)",
549 inode_first_full_path(inode),
550 blob->size, REPARSE_DATA_MAX_SIZE);
551 return WIMLIB_ERR_INVALID_REPARSE_DATA;
553 ctx->reparse_ptr = ctx->reparse_data;
557 wimlib_assert(stream_is_unnamed_data_stream(strm));
559 /* Unnamed data stream of "regular" file */
561 /* This should be ensured by extract_blob_list() */
562 wimlib_assert(ctx->num_open_fds < MAX_OPEN_FILES);
564 first_dentry = inode_first_extraction_dentry(inode);
565 first_path = unix_build_extraction_path(first_dentry, ctx);
567 fd = open(first_path, O_TRUNC | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
569 if (errno == EEXIST && !unlink(first_path))
571 ERROR_WITH_ERRNO("Can't create regular file \"%s\"", first_path);
572 return WIMLIB_ERR_OPEN;
574 filedes_init(&ctx->open_fds[ctx->num_open_fds++], fd);
575 return unix_create_hardlinks(inode, first_dentry, first_path, ctx);
578 /* Called when starting to read a blob for extraction */
580 unix_begin_extract_blob(struct blob_descriptor *blob, void *_ctx)
582 struct unix_apply_ctx *ctx = _ctx;
583 const struct blob_extraction_target *targets = blob_extraction_targets(blob);
585 for (u32 i = 0; i < blob->out_refcnt; i++) {
586 int ret = unix_begin_extract_blob_instance(blob,
591 ctx->reparse_ptr = NULL;
592 unix_cleanup_open_fds(ctx, 0);
599 /* Called when the next chunk of a blob has been read for extraction */
601 unix_extract_chunk(const void *chunk, size_t size, void *_ctx)
603 struct unix_apply_ctx *ctx = _ctx;
606 for (unsigned i = 0; i < ctx->num_open_fds; i++) {
607 ret = full_write(&ctx->open_fds[i], chunk, size);
609 ERROR_WITH_ERRNO("Error writing data to filesystem");
613 if (ctx->reparse_ptr)
614 ctx->reparse_ptr = mempcpy(ctx->reparse_ptr, chunk, size);
618 /* Called when a blob has been fully read for extraction */
620 unix_end_extract_blob(struct blob_descriptor *blob, int status, void *_ctx)
622 struct unix_apply_ctx *ctx = _ctx;
625 const struct blob_extraction_target *targets = blob_extraction_targets(blob);
627 ctx->reparse_ptr = NULL;
630 unix_cleanup_open_fds(ctx, 0);
636 for (u32 i = 0; i < blob->out_refcnt; i++) {
637 struct wim_inode *inode = targets[i].inode;
639 if (inode_is_symlink(inode)) {
640 /* We finally have the symlink data, so we can create
644 path = unix_build_inode_extraction_path(inode, ctx);
645 ret = unix_create_symlink(inode, path, blob->size, ctx);
647 ERROR_WITH_ERRNO("Can't create symbolic link "
651 ret = unix_set_metadata(-1, inode, path, ctx);
655 /* Set metadata on regular file just before closing it.
657 struct filedes *fd = &ctx->open_fds[j];
659 ret = unix_set_metadata(fd->fd, inode, NULL, ctx);
663 if (filedes_close(fd)) {
664 ERROR_WITH_ERRNO("Error closing \"%s\"",
665 unix_build_inode_extraction_path(inode, ctx));
666 ret = WIMLIB_ERR_WRITE;
672 unix_cleanup_open_fds(ctx, j);
677 unix_set_dir_metadata(struct list_head *dentry_list, struct unix_apply_ctx *ctx)
679 const struct wim_dentry *dentry;
682 list_for_each_entry_reverse(dentry, dentry_list, d_extraction_list_node) {
683 if (dentry_is_directory(dentry)) {
684 ret = unix_set_metadata(-1, dentry->d_inode, NULL, ctx);
687 ret = report_file_metadata_applied(&ctx->common);
696 unix_extract(struct list_head *dentry_list, struct apply_ctx *_ctx)
699 struct unix_apply_ctx *ctx = (struct unix_apply_ctx *)_ctx;
702 u64 empty_file_count;
704 /* Compute the maximum path length that will be needed, then allocate
705 * some path buffers. */
706 path_max = unix_compute_path_max(dentry_list, ctx);
708 for (unsigned i = 0; i < NUM_PATHBUFS; i++) {
709 ctx->pathbufs[i] = MALLOC(path_max);
710 if (!ctx->pathbufs[i]) {
711 ret = WIMLIB_ERR_NOMEM;
714 /* Pre-fill the target in each path buffer. We'll just append
715 * the rest of the paths after this. */
716 memcpy(ctx->pathbufs[i],
717 ctx->common.target, ctx->common.target_nchars);
720 /* Extract directories and empty regular files. Directories are needed
721 * because we can't extract any other files until their directories
722 * exist. Empty files are needed because they don't have
723 * representatives in the blob list. */
725 unix_count_dentries(dentry_list, &dir_count, &empty_file_count);
727 ret = start_file_structure_phase(&ctx->common, dir_count + empty_file_count);
731 ret = unix_create_dirs_and_empty_files(dentry_list, ctx);
735 ret = end_file_structure_phase(&ctx->common);
739 /* Get full path to target if needed for absolute symlink fixups. */
740 if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
741 ctx->common.required_features.symlink_reparse_points)
743 ctx->target_abspath = realpath(ctx->common.target, NULL);
744 if (!ctx->target_abspath) {
745 ret = WIMLIB_ERR_NOMEM;
748 ctx->target_abspath_nchars = strlen(ctx->target_abspath);
751 /* Extract nonempty regular files and symbolic links. */
753 struct read_blob_callbacks cbs = {
754 .begin_blob = unix_begin_extract_blob,
755 .consume_chunk = unix_extract_chunk,
756 .end_blob = unix_end_extract_blob,
759 ret = extract_blob_list(&ctx->common, &cbs);
764 /* Set directory metadata. We do this last so that we get the right
765 * directory timestamps. */
766 ret = start_file_metadata_phase(&ctx->common, dir_count);
770 ret = unix_set_dir_metadata(dentry_list, ctx);
774 ret = end_file_metadata_phase(&ctx->common);
778 if (ctx->num_special_files_ignored) {
779 WARNING("%lu special files were not extracted due to EPERM!",
780 ctx->num_special_files_ignored);
783 for (unsigned i = 0; i < NUM_PATHBUFS; i++)
784 FREE(ctx->pathbufs[i]);
785 FREE(ctx->target_abspath);
789 const struct apply_operations unix_apply_ops = {
791 .get_supported_features = unix_get_supported_features,
792 .extract = unix_extract,
793 .context_size = sizeof(struct unix_apply_ctx),