2 * Copyright (c) 2003-2007 Tim Kientzle
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer
10 * in this position and unchanged.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 #include "archive_platform.h"
28 __FBSDID("$FreeBSD: src/lib/libarchive/archive_write_disk.c,v 1.24 2008/03/15 04:20:50 kientzle Exp $");
30 #ifdef HAVE_SYS_TYPES_H
31 #include <sys/types.h>
36 #ifdef HAVE_ATTR_XATTR_H
37 #include <attr/xattr.h>
39 #ifdef HAVE_SYS_IOCTL_H
40 #include <sys/ioctl.h>
42 #ifdef HAVE_SYS_STAT_H
45 #ifdef HAVE_SYS_TIME_H
48 #ifdef HAVE_SYS_UTIME_H
49 #include <sys/utime.h>
52 #ifdef HAVE_EXT2FS_EXT2_FS_H
53 #include <ext2fs/ext2_fs.h> /* for Linux file flags */
64 #ifdef HAVE_LINUX_FS_H
65 #include <linux/fs.h> /* for Linux file flags */
67 #ifdef HAVE_LINUX_EXT2_FS_H
68 #include <linux/ext2_fs.h> /* for Linux file flags */
91 #include "archive_string.h"
92 #include "archive_entry.h"
93 #include "archive_private.h"
100 struct fixup_entry
*next
;
104 unsigned long mtime_nanos
;
105 unsigned long atime_nanos
;
106 unsigned long fflags_set
;
107 int fixup
; /* bitmask of what needs fixing */
112 * We use a bitmask to track which operations remain to be done for
113 * this file. In particular, this helps us avoid unnecessary
114 * operations when it's possible to take care of one step as a
115 * side-effect of another. For example, mkdir() can specify the mode
116 * for the newly-created object but symlink() cannot. This means we
117 * can skip chmod() if mkdir() succeeded, but we must explicitly
118 * chmod() if we're trying to create a directory that already exists
119 * (mkdir() failed) or if we're restoring a symlink. Similarly, we
120 * need to verify UID/GID before trying to restore SUID/SGID bits;
121 * that verification can occur explicitly through a stat() call or
122 * implicitly because of a successful chown() call.
124 #define TODO_MODE_FORCE 0x40000000
125 #define TODO_MODE_BASE 0x20000000
126 #define TODO_SUID 0x10000000
127 #define TODO_SUID_CHECK 0x08000000
128 #define TODO_SGID 0x04000000
129 #define TODO_SGID_CHECK 0x02000000
130 #define TODO_MODE (TODO_MODE_BASE|TODO_SUID|TODO_SGID)
131 #define TODO_TIMES ARCHIVE_EXTRACT_TIME
132 #define TODO_OWNER ARCHIVE_EXTRACT_OWNER
133 #define TODO_FFLAGS ARCHIVE_EXTRACT_FFLAGS
134 #define TODO_ACLS ARCHIVE_EXTRACT_ACL
135 #define TODO_XATTR ARCHIVE_EXTRACT_XATTR
137 struct archive_write_disk
{
138 struct archive archive
;
141 struct fixup_entry
*fixup_list
;
142 struct fixup_entry
*current_fixup
;
147 gid_t (*lookup_gid
)(void *private, const char *gname
, gid_t gid
);
148 void (*cleanup_gid
)(void *private);
149 void *lookup_gid_data
;
150 uid_t (*lookup_uid
)(void *private, const char *gname
, gid_t gid
);
151 void (*cleanup_uid
)(void *private);
152 void *lookup_uid_data
;
155 * Full path of last file to satisfy symlink checks.
157 struct archive_string path_safe
;
160 * Cached stat data from disk for the current entry.
161 * If this is valid, pst points to st. Otherwise,
167 /* Information about the object being restored right now. */
168 struct archive_entry
*entry
; /* Entry being extracted. */
169 char *name
; /* Name of entry, possibly edited. */
170 struct archive_string _name_data
; /* backing store for 'name' */
171 /* Tasks remaining for this object. */
173 /* Tasks deferred until end-of-archive. */
175 /* Options requested by the client. */
177 /* Handle for the file we're restoring. */
179 /* Current offset for writing data to the file. */
181 /* Maximum size of file. */
183 /* Dir we were in before this restore; only for deep paths. */
185 /* Mode we should use for this entry; affected by _PERM and umask. */
187 /* UID/GID to use in restoring this entry. */
190 /* Last offset written to disk. */
195 * Default mode for dirs created automatically (will be modified by umask).
196 * Note that POSIX specifies 0777 for implicity-created dirs, "modified
197 * by the process' file creation mask."
199 #define DEFAULT_DIR_MODE 0777
201 * Dir modes are restored in two steps: During the extraction, the permissions
202 * in the archive are modified to match the following limits. During
203 * the post-extract fixup pass, the permissions from the archive are
206 #define MINIMUM_DIR_MODE 0700
207 #define MAXIMUM_DIR_MODE 0775
209 static int check_symlinks(struct archive_write_disk
*);
210 static int create_filesystem_object(struct archive_write_disk
*);
211 static struct fixup_entry
*current_fixup(struct archive_write_disk
*, const char *pathname
);
213 static void edit_deep_directories(struct archive_write_disk
*ad
);
215 static int cleanup_pathname(struct archive_write_disk
*);
216 static int create_dir(struct archive_write_disk
*, char *);
217 static int create_parent_dir(struct archive_write_disk
*, char *);
218 static int older(struct stat
*, struct archive_entry
*);
219 static int restore_entry(struct archive_write_disk
*);
220 #ifdef HAVE_POSIX_ACL
221 static int set_acl(struct archive_write_disk
*, int fd
, struct archive_entry
*,
222 acl_type_t
, int archive_entry_acl_type
, const char *tn
);
224 static int set_acls(struct archive_write_disk
*);
225 static int set_xattrs(struct archive_write_disk
*);
226 static int set_fflags(struct archive_write_disk
*);
227 static int set_fflags_platform(struct archive_write_disk
*, int fd
,
228 const char *name
, mode_t mode
,
229 unsigned long fflags_set
, unsigned long fflags_clear
);
230 static int set_ownership(struct archive_write_disk
*);
231 static int set_mode(struct archive_write_disk
*, int mode
);
232 static int set_time(struct archive_write_disk
*);
233 static struct fixup_entry
*sort_dir_list(struct fixup_entry
*p
);
234 static gid_t
trivial_lookup_gid(void *, const char *, gid_t
);
235 static uid_t
trivial_lookup_uid(void *, const char *, uid_t
);
238 static struct archive_vtable
*archive_write_disk_vtable(void);
240 static int _archive_write_close(struct archive
*);
241 static int _archive_write_finish(struct archive
*);
242 static int _archive_write_header(struct archive
*, struct archive_entry
*);
243 static int _archive_write_finish_entry(struct archive
*);
244 static ssize_t
_archive_write_data(struct archive
*, const void *, size_t);
245 static ssize_t
_archive_write_data_block(struct archive
*, const void *, size_t, off_t
);
248 _archive_write_disk_lazy_stat(struct archive_write_disk
*a
)
250 if (a
->pst
!= NULL
) {
251 /* Already have stat() data available. */
255 if (a
->fd
>= 0 && fstat(a
->fd
, &a
->st
) == 0) {
261 * XXX At this point, symlinks should not be hit, otherwise
262 * XXX a race occured. Do we want to check explicitly for that?
264 if (lstat(a
->name
, &a
->st
) == 0) {
268 archive_set_error(&a
->archive
, errno
, "Couldn't stat file");
269 return (ARCHIVE_WARN
);
272 static struct archive_vtable
*
273 archive_write_disk_vtable(void)
275 static struct archive_vtable av
;
276 static int inited
= 0;
279 av
.archive_write_close
= _archive_write_close
;
280 av
.archive_write_finish
= _archive_write_finish
;
281 av
.archive_write_header
= _archive_write_header
;
282 av
.archive_write_finish_entry
= _archive_write_finish_entry
;
283 av
.archive_write_data
= _archive_write_data
;
284 av
.archive_write_data_block
= _archive_write_data_block
;
291 archive_write_disk_set_options(struct archive
*_a
, int flags
)
293 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
301 * Extract this entry to disk.
303 * TODO: Validate hardlinks. According to the standards, we're
304 * supposed to check each extracted hardlink and squawk if it refers
305 * to a file that we didn't restore. I'm not entirely convinced this
306 * is a good idea, but more importantly: Is there any way to validate
307 * hardlinks without keeping a complete list of filenames from the
308 * entire archive?? Ugh.
312 _archive_write_header(struct archive
*_a
, struct archive_entry
*entry
)
314 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
315 struct fixup_entry
*fe
;
318 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
319 ARCHIVE_STATE_HEADER
| ARCHIVE_STATE_DATA
,
320 "archive_write_disk_header");
321 archive_clear_error(&a
->archive
);
322 if (a
->archive
.state
& ARCHIVE_STATE_DATA
) {
323 r
= _archive_write_finish_entry(&a
->archive
);
324 if (r
== ARCHIVE_FATAL
)
328 /* Set up for this particular entry. */
330 a
->current_fixup
= NULL
;
333 archive_entry_free(a
->entry
);
336 a
->entry
= archive_entry_clone(entry
);
340 a
->uid
= a
->user_uid
;
341 a
->mode
= archive_entry_mode(a
->entry
);
342 a
->filesize
= archive_entry_size(a
->entry
);
343 archive_strcpy(&(a
->_name_data
), archive_entry_pathname(a
->entry
));
344 a
->name
= a
->_name_data
.s
;
345 archive_clear_error(&a
->archive
);
348 * Clean up the requested path. This is necessary for correct
349 * dir restores; the dir restore logic otherwise gets messed
350 * up by nonsense like "dir/.".
352 ret
= cleanup_pathname(a
);
353 if (ret
!= ARCHIVE_OK
)
357 * Set the umask to zero so we get predictable mode settings.
358 * This gets done on every call to _write_header in case the
359 * user edits their umask during the extraction for some
360 * reason. This will be reset before we return. Note that we
361 * don't need to do this in _finish_entry, as the chmod(), etc,
362 * system calls don't obey umask.
364 a
->user_umask
= umask(0);
365 /* From here on, early exit requires "goto done" to clean up. */
367 /* Figure out what we need to do for this entry. */
368 a
->todo
= TODO_MODE_BASE
;
369 if (a
->flags
& ARCHIVE_EXTRACT_PERM
) {
370 a
->todo
|= TODO_MODE_FORCE
; /* Be pushy about permissions. */
372 * SGID requires an extra "check" step because we
373 * cannot easily predict the GID that the system will
374 * assign. (Different systems assign GIDs to files
375 * based on a variety of criteria, including process
376 * credentials and the gid of the enclosing
377 * directory.) We can only restore the SGID bit if
378 * the file has the right GID, and we only know the
379 * GID if we either set it (see set_ownership) or if
380 * we've actually called stat() on the file after it
381 * was restored. Since there are several places at
382 * which we might verify the GID, we need a TODO bit
385 if (a
->mode
& S_ISGID
)
386 a
->todo
|= TODO_SGID
| TODO_SGID_CHECK
;
388 * Verifying the SUID is simpler, but can still be
389 * done in multiple ways, hence the separate "check" bit.
391 if (a
->mode
& S_ISUID
)
392 a
->todo
|= TODO_SUID
| TODO_SUID_CHECK
;
395 * User didn't request full permissions, so don't
396 * restore SUID, SGID bits and obey umask.
401 a
->mode
&= ~a
->user_umask
;
403 if (a
->flags
& ARCHIVE_EXTRACT_OWNER
)
404 a
->todo
|= TODO_OWNER
;
405 if (a
->flags
& ARCHIVE_EXTRACT_TIME
)
406 a
->todo
|= TODO_TIMES
;
407 if (a
->flags
& ARCHIVE_EXTRACT_ACL
)
408 a
->todo
|= TODO_ACLS
;
409 if (a
->flags
& ARCHIVE_EXTRACT_FFLAGS
)
410 a
->todo
|= TODO_FFLAGS
;
411 if (a
->flags
& ARCHIVE_EXTRACT_SECURE_SYMLINKS
) {
412 ret
= check_symlinks(a
);
413 if (ret
!= ARCHIVE_OK
)
417 /* If path exceeds PATH_MAX, shorten the path. */
418 edit_deep_directories(a
);
421 ret
= restore_entry(a
);
424 /* If we changed directory above, restore it here. */
425 if (a
->restore_pwd
>= 0) {
426 fchdir(a
->restore_pwd
);
427 close(a
->restore_pwd
);
433 * Fixup uses the unedited pathname from archive_entry_pathname(),
434 * because it is relative to the base dir and the edited path
435 * might be relative to some intermediate dir as a result of the
436 * deep restore logic.
438 if (a
->deferred
& TODO_MODE
) {
439 fe
= current_fixup(a
, archive_entry_pathname(entry
));
440 fe
->fixup
|= TODO_MODE_BASE
;
444 if (a
->deferred
& TODO_TIMES
) {
445 fe
= current_fixup(a
, archive_entry_pathname(entry
));
446 fe
->fixup
|= TODO_TIMES
;
447 fe
->mtime
= archive_entry_mtime(entry
);
448 fe
->mtime_nanos
= archive_entry_mtime_nsec(entry
);
449 fe
->atime
= archive_entry_atime(entry
);
450 fe
->atime_nanos
= archive_entry_atime_nsec(entry
);
453 if (a
->deferred
& TODO_FFLAGS
) {
454 fe
= current_fixup(a
, archive_entry_pathname(entry
));
455 fe
->fixup
|= TODO_FFLAGS
;
456 /* TODO: Complete this.. defer fflags from below. */
459 /* We've created the object and are ready to pour data into it. */
460 if (ret
== ARCHIVE_OK
)
461 a
->archive
.state
= ARCHIVE_STATE_DATA
;
463 * If it's not open, tell our client not to try writing.
464 * In particular, dirs, links, etc, don't get written to.
467 archive_entry_set_size(entry
, 0);
471 /* Restore the user's umask before returning. */
472 umask(a
->user_umask
);
478 archive_write_disk_set_skip_file(struct archive
*_a
, dev_t d
, ino_t i
)
480 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
481 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
482 ARCHIVE_STATE_ANY
, "archive_write_disk_set_skip_file");
483 a
->skip_file_dev
= d
;
484 a
->skip_file_ino
= i
;
489 _archive_write_data_block(struct archive
*_a
,
490 const void *buff
, size_t size
, off_t offset
)
492 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
493 ssize_t bytes_written
= 0;
494 ssize_t block_size
, bytes_to_write
;
497 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
498 ARCHIVE_STATE_DATA
, "archive_write_disk_block");
500 archive_set_error(&a
->archive
, 0, "File not open");
501 return (ARCHIVE_WARN
);
503 archive_clear_error(&a
->archive
);
505 if (a
->flags
& ARCHIVE_EXTRACT_SPARSE
) {
506 if ((r
= _archive_write_disk_lazy_stat(a
)) != ARCHIVE_OK
)
508 block_size
= a
->pst
->st_blksize
;
512 if ((off_t
)(offset
+ size
) > a
->filesize
) {
513 size
= (size_t)(a
->filesize
- a
->offset
);
514 archive_set_error(&a
->archive
, 0,
515 "Write request too large");
519 /* Write the data. */
521 if (block_size
!= -1) {
524 for (buf
= buff
; size
; ++buf
, --size
, ++offset
) {
530 bytes_to_write
= block_size
- offset
% block_size
;
533 bytes_to_write
= size
;
534 /* Seek if necessary to the specified offset. */
535 if (offset
!= a
->last_offset
) {
536 if (lseek(a
->fd
, offset
, SEEK_SET
) < 0) {
537 archive_set_error(&a
->archive
, errno
, "Seek failed");
538 return (ARCHIVE_FATAL
);
541 bytes_written
= write(a
->fd
, buff
, size
);
542 if (bytes_written
< 0) {
543 archive_set_error(&a
->archive
, errno
, "Write failed");
544 return (ARCHIVE_WARN
);
546 buff
= (const char *)buff
+ bytes_written
;
547 size
-= bytes_written
;
548 offset
+= bytes_written
;
549 a
->last_offset
= a
->offset
= offset
;
556 _archive_write_data(struct archive
*_a
, const void *buff
, size_t size
)
558 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
561 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
562 ARCHIVE_STATE_DATA
, "archive_write_data");
566 r
= _archive_write_data_block(_a
, buff
, size
, a
->offset
);
573 _archive_write_finish_entry(struct archive
*_a
)
575 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
576 int ret
= ARCHIVE_OK
;
578 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
579 ARCHIVE_STATE_HEADER
| ARCHIVE_STATE_DATA
,
580 "archive_write_finish_entry");
581 if (a
->archive
.state
& ARCHIVE_STATE_HEADER
)
583 archive_clear_error(&a
->archive
);
585 if (a
->last_offset
!= a
->filesize
&& a
->fd
>= 0) {
586 if (ftruncate(a
->fd
, a
->filesize
) == -1 &&
588 archive_set_error(&a
->archive
, errno
,
589 "File size could not be restored");
590 return (ARCHIVE_FAILED
);
593 * Explicitly stat the file as some platforms might not
594 * implement the XSI option to extend files via ftruncate.
597 if ((ret
= _archive_write_disk_lazy_stat(a
)) != ARCHIVE_OK
)
599 if (a
->st
.st_size
!= a
->filesize
) {
600 const char nul
= '\0';
601 if (lseek(a
->fd
, a
->st
.st_size
- 1, SEEK_SET
) < 0) {
602 archive_set_error(&a
->archive
, errno
, "Seek failed");
603 return (ARCHIVE_FATAL
);
605 if (write(a
->fd
, &nul
, 1) < 0) {
606 archive_set_error(&a
->archive
, errno
,
607 "Write to restore size failed");
608 return (ARCHIVE_FATAL
);
613 /* Restore metadata. */
616 * Look up the "real" UID only if we're going to need it. We
617 * need this for TODO_SGID because chown() requires both.
619 if (a
->todo
& (TODO_OWNER
| TODO_SUID
| TODO_SGID
)) {
620 a
->uid
= a
->lookup_uid(a
->lookup_uid_data
,
621 archive_entry_uname(a
->entry
),
622 archive_entry_uid(a
->entry
));
624 /* Look up the "real" GID only if we're going to need it. */
625 if (a
->todo
& (TODO_OWNER
| TODO_SGID
| TODO_SUID
)) {
626 a
->gid
= a
->lookup_gid(a
->lookup_gid_data
,
627 archive_entry_gname(a
->entry
),
628 archive_entry_gid(a
->entry
));
631 * If restoring ownership, do it before trying to restore suid/sgid
632 * bits. If we set the owner, we know what it is and can skip
633 * a stat() call to examine the ownership of the file on disk.
635 if (a
->todo
& TODO_OWNER
)
636 ret
= set_ownership(a
);
637 if (a
->todo
& TODO_MODE
) {
638 int r2
= set_mode(a
, a
->mode
);
639 if (r2
< ret
) ret
= r2
;
641 if (a
->todo
& TODO_TIMES
) {
642 int r2
= set_time(a
);
643 if (r2
< ret
) ret
= r2
;
645 if (a
->todo
& TODO_ACLS
) {
646 int r2
= set_acls(a
);
647 if (r2
< ret
) ret
= r2
;
649 if (a
->todo
& TODO_XATTR
) {
650 int r2
= set_xattrs(a
);
651 if (r2
< ret
) ret
= r2
;
653 if (a
->todo
& TODO_FFLAGS
) {
654 int r2
= set_fflags(a
);
655 if (r2
< ret
) ret
= r2
;
658 /* If there's an fd, we can close it now. */
663 /* If there's an entry, we can release it now. */
665 archive_entry_free(a
->entry
);
668 a
->archive
.state
= ARCHIVE_STATE_HEADER
;
673 archive_write_disk_set_group_lookup(struct archive
*_a
,
675 gid_t (*lookup_gid
)(void *private, const char *gname
, gid_t gid
),
676 void (*cleanup_gid
)(void *private))
678 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
679 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
680 ARCHIVE_STATE_ANY
, "archive_write_disk_set_group_lookup");
682 a
->lookup_gid
= lookup_gid
;
683 a
->cleanup_gid
= cleanup_gid
;
684 a
->lookup_gid_data
= private_data
;
689 archive_write_disk_set_user_lookup(struct archive
*_a
,
691 uid_t (*lookup_uid
)(void *private, const char *uname
, uid_t uid
),
692 void (*cleanup_uid
)(void *private))
694 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
695 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
696 ARCHIVE_STATE_ANY
, "archive_write_disk_set_user_lookup");
698 a
->lookup_uid
= lookup_uid
;
699 a
->cleanup_uid
= cleanup_uid
;
700 a
->lookup_uid_data
= private_data
;
706 * Create a new archive_write_disk object and initialize it with global state.
709 archive_write_disk_new(void)
711 struct archive_write_disk
*a
;
713 a
= (struct archive_write_disk
*)malloc(sizeof(*a
));
716 memset(a
, 0, sizeof(*a
));
717 a
->archive
.magic
= ARCHIVE_WRITE_DISK_MAGIC
;
718 /* We're ready to write a header immediately. */
719 a
->archive
.state
= ARCHIVE_STATE_HEADER
;
720 a
->archive
.vtable
= archive_write_disk_vtable();
721 a
->lookup_uid
= trivial_lookup_uid
;
722 a
->lookup_gid
= trivial_lookup_gid
;
724 a
->user_uid
= geteuid();
725 #endif /* HAVE_GETEUID */
726 if (archive_string_ensure(&a
->path_safe
, 512) == NULL
) {
730 return (&a
->archive
);
735 * If pathname is longer than PATH_MAX, chdir to a suitable
736 * intermediate dir and edit the path down to a shorter suffix. Note
737 * that this routine never returns an error; if the chdir() attempt
738 * fails for any reason, we just go ahead with the long pathname. The
739 * object creation is likely to fail, but any error will get handled
744 edit_deep_directories(struct archive_write_disk
*a
)
747 char *tail
= a
->name
;
751 /* If path is short, avoid the open() below. */
752 if (strlen(tail
) <= PATH_MAX
)
755 /* Try to record our starting dir. */
756 a
->restore_pwd
= open(".", O_RDONLY
| O_BINARY
);
757 if (a
->restore_pwd
< 0)
760 /* As long as the path is too long... */
761 while (strlen(tail
) > PATH_MAX
) {
762 /* Locate a dir prefix shorter than PATH_MAX. */
763 tail
+= PATH_MAX
- 8;
764 while (tail
> a
->name
&& *tail
!= '/')
766 /* Exit if we find a too-long path component. */
769 /* Create the intermediate dir and chdir to it. */
770 *tail
= '\0'; /* Terminate dir portion */
771 ret
= create_dir(a
, a
->name
);
772 if (ret
== ARCHIVE_OK
&& chdir(a
->name
) != 0)
774 *tail
= '/'; /* Restore the / we removed. */
775 if (ret
!= ARCHIVE_OK
)
778 /* The chdir() succeeded; we've now shortened the path. */
786 * The main restore function.
789 restore_entry(struct archive_write_disk
*a
)
791 int ret
= ARCHIVE_OK
, en
;
793 if (a
->flags
& ARCHIVE_EXTRACT_UNLINK
&& !S_ISDIR(a
->mode
)) {
795 * TODO: Fix this. Apparently, there are platforms
796 * that still allow root to hose the entire filesystem
797 * by unlinking a dir. The S_ISDIR() test above
798 * prevents us from using unlink() here if the new
799 * object is a dir, but that doesn't mean the old
800 * object isn't a dir.
802 if (unlink(a
->name
) == 0) {
803 /* We removed it, reset cached stat. */
805 } else if (errno
== ENOENT
) {
806 /* File didn't exist, that's just as good. */
807 } else if (rmdir(a
->name
) == 0) {
808 /* It was a dir, but now it's gone. */
811 /* We tried, but couldn't get rid of it. */
812 archive_set_error(&a
->archive
, errno
,
814 return(ARCHIVE_WARN
);
818 /* Try creating it first; if this fails, we'll try to recover. */
819 en
= create_filesystem_object(a
);
821 if ((en
== ENOTDIR
|| en
== ENOENT
)
822 && !(a
->flags
& ARCHIVE_EXTRACT_NO_AUTODIR
)) {
823 /* If the parent dir doesn't exist, try creating it. */
824 create_parent_dir(a
, a
->name
);
825 /* Now try to create the object again. */
826 en
= create_filesystem_object(a
);
829 if ((en
== EISDIR
|| en
== EEXIST
)
830 && (a
->flags
& ARCHIVE_EXTRACT_NO_OVERWRITE
)) {
831 /* If we're not overwriting, we're done. */
832 archive_set_error(&a
->archive
, en
, "Already exists");
833 return (ARCHIVE_WARN
);
837 * Some platforms return EISDIR if you call
838 * open(O_WRONLY | O_EXCL | O_CREAT) on a directory, some
839 * return EEXIST. POSIX is ambiguous, requiring EISDIR
840 * for open(O_WRONLY) on a dir and EEXIST for open(O_EXCL | O_CREAT)
841 * on an existing item.
844 /* A dir is in the way of a non-dir, rmdir it. */
845 if (rmdir(a
->name
) != 0) {
846 archive_set_error(&a
->archive
, errno
,
847 "Can't remove already-existing dir");
848 return (ARCHIVE_WARN
);
852 en
= create_filesystem_object(a
);
853 } else if (en
== EEXIST
) {
855 * We know something is in the way, but we don't know what;
856 * we need to find out before we go any further.
858 if (lstat(a
->name
, &a
->st
) != 0) {
859 archive_set_error(&a
->archive
, errno
,
860 "Can't stat existing object");
861 return (ARCHIVE_WARN
);
864 /* TODO: if it's a symlink... */
866 if (a
->flags
& ARCHIVE_EXTRACT_NO_OVERWRITE_NEWER
) {
867 if (!older(&(a
->st
), a
->entry
)) {
868 archive_set_error(&a
->archive
, 0,
869 "File on disk is not older; skipping.");
870 return (ARCHIVE_FAILED
);
874 /* If it's our archive, we're done. */
875 if (a
->skip_file_dev
> 0 &&
876 a
->skip_file_ino
> 0 &&
877 a
->st
.st_dev
== a
->skip_file_dev
&&
878 a
->st
.st_ino
== a
->skip_file_ino
) {
879 archive_set_error(&a
->archive
, 0, "Refusing to overwrite archive");
880 return (ARCHIVE_FAILED
);
883 if (!S_ISDIR(a
->st
.st_mode
)) {
884 /* A non-dir is in the way, unlink it. */
885 if (unlink(a
->name
) != 0) {
886 archive_set_error(&a
->archive
, errno
,
887 "Can't unlink already-existing object");
888 return (ARCHIVE_WARN
);
892 en
= create_filesystem_object(a
);
893 } else if (!S_ISDIR(a
->mode
)) {
894 /* A dir is in the way of a non-dir, rmdir it. */
895 if (rmdir(a
->name
) != 0) {
896 archive_set_error(&a
->archive
, errno
,
897 "Can't remove already-existing dir");
898 return (ARCHIVE_WARN
);
901 en
= create_filesystem_object(a
);
904 * There's a dir in the way of a dir. Don't
905 * waste time with rmdir()/mkdir(), just fix
906 * up the permissions on the existing dir.
907 * Note that we don't change perms on existing
908 * dirs unless _EXTRACT_PERM is specified.
910 if ((a
->mode
!= a
->st
.st_mode
)
911 && (a
->todo
& TODO_MODE_FORCE
))
912 a
->deferred
|= (a
->todo
& TODO_MODE
);
913 /* Ownership doesn't need deferred fixup. */
914 en
= 0; /* Forget the EEXIST. */
919 /* Everything failed; give up here. */
920 archive_set_error(&a
->archive
, en
, "Can't create '%s'", a
->name
);
921 return (ARCHIVE_WARN
);
924 a
->pst
= NULL
; /* Cached stat data no longer valid. */
929 * Returns 0 if creation succeeds, or else returns errno value from
930 * the failed system call. Note: This function should only ever perform
931 * a single system call.
934 create_filesystem_object(struct archive_write_disk
*a
)
936 /* Create the entry. */
937 const char *linkname
;
938 mode_t final_mode
, mode
;
941 /* We identify hard/symlinks according to the link names. */
942 /* Since link(2) and symlink(2) don't handle modes, we're done here. */
943 linkname
= archive_entry_hardlink(a
->entry
);
944 if (linkname
!= NULL
) {
945 r
= link(linkname
, a
->name
) ? errno
: 0;
947 * New cpio and pax formats allow hardlink entries
948 * to carry data, so we may have to open the file
949 * for hardlink entries.
951 * If the hardlink was successfully created and
952 * the archive doesn't have carry data for it,
953 * consider it to be non-authoritive for meta data.
954 * This is consistent with GNU tar and BSD pax.
955 * If the hardlink does carry data, let the last
956 * archive entry decide ownership.
958 if (r
== 0 && a
->filesize
== 0) {
961 } if (r
== 0 && a
->filesize
> 0) {
962 a
->fd
= open(a
->name
, O_WRONLY
| O_TRUNC
| O_BINARY
);
968 linkname
= archive_entry_symlink(a
->entry
);
969 if (linkname
!= NULL
)
970 return symlink(linkname
, a
->name
) ? errno
: 0;
973 * The remaining system calls all set permissions, so let's
974 * try to take advantage of that to avoid an extra chmod()
975 * call. (Recall that umask is set to zero right now!)
978 /* Mode we want for the final restored object (w/o file type bits). */
979 final_mode
= a
->mode
& 07777;
981 * The mode that will actually be restored in this step. Note
982 * that SUID, SGID, etc, require additional work to ensure
983 * security, so we never restore them at this point.
985 mode
= final_mode
& 0777;
987 switch (a
->mode
& AE_IFMT
) {
989 /* POSIX requires that we fall through here. */
992 a
->fd
= open(a
->name
,
993 O_WRONLY
| O_CREAT
| O_EXCL
| O_BINARY
, mode
);
998 /* Note: we use AE_IFCHR for the case label, and
999 * S_IFCHR for the mknod() call. This is correct. */
1000 r
= mknod(a
->name
, mode
| S_IFCHR
,
1001 archive_entry_rdev(a
->entry
));
1003 /* TODO: Find a better way to warn about our inability
1004 * to restore a char device node. */
1006 #endif /* HAVE_MKNOD */
1010 r
= mknod(a
->name
, mode
| S_IFBLK
,
1011 archive_entry_rdev(a
->entry
));
1013 /* TODO: Find a better way to warn about our inability
1014 * to restore a block device node. */
1016 #endif /* HAVE_MKNOD */
1019 mode
= (mode
| MINIMUM_DIR_MODE
) & MAXIMUM_DIR_MODE
;
1020 r
= mkdir(a
->name
, mode
);
1022 /* Defer setting dir times. */
1023 a
->deferred
|= (a
->todo
& TODO_TIMES
);
1024 a
->todo
&= ~TODO_TIMES
;
1025 /* Never use an immediate chmod(). */
1026 if (mode
!= final_mode
)
1027 a
->deferred
|= (a
->todo
& TODO_MODE
);
1028 a
->todo
&= ~TODO_MODE
;
1033 r
= mkfifo(a
->name
, mode
);
1035 /* TODO: Find a better way to warn about our inability
1036 * to restore a fifo. */
1038 #endif /* HAVE_MKFIFO */
1042 /* All the system calls above set errno on failure. */
1046 /* If we managed to set the final mode, we've avoided a chmod(). */
1047 if (mode
== final_mode
)
1048 a
->todo
&= ~TODO_MODE
;
1053 * Cleanup function for archive_extract. Mostly, this involves processing
1054 * the fixup list, which is used to address a number of problems:
1055 * * Dir permissions might prevent us from restoring a file in that
1056 * dir, so we restore the dir with minimum 0700 permissions first,
1057 * then correct the mode at the end.
1058 * * Similarly, the act of restoring a file touches the directory
1059 * and changes the timestamp on the dir, so we have to touch-up dir
1060 * timestamps at the end as well.
1061 * * Some file flags can interfere with the restore by, for example,
1062 * preventing the creation of hardlinks to those files.
1064 * Note that tar/cpio do not require that archives be in a particular
1065 * order; there is no way to know when the last file has been restored
1066 * within a directory, so there's no way to optimize the memory usage
1067 * here by fixing up the directory any earlier than the
1070 * XXX TODO: Directory ACLs should be restored here, for the same
1071 * reason we set directory perms here. XXX
1074 _archive_write_close(struct archive
*_a
)
1076 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
1077 struct fixup_entry
*next
, *p
;
1080 __archive_check_magic(&a
->archive
, ARCHIVE_WRITE_DISK_MAGIC
,
1081 ARCHIVE_STATE_HEADER
| ARCHIVE_STATE_DATA
,
1082 "archive_write_disk_close");
1083 ret
= _archive_write_finish_entry(&a
->archive
);
1085 /* Sort dir list so directories are fixed up in depth-first order. */
1086 p
= sort_dir_list(a
->fixup_list
);
1089 a
->pst
= NULL
; /* Mark stat cache as out-of-date. */
1090 if (p
->fixup
& TODO_TIMES
) {
1092 /* {f,l,}utimes() are preferred, when available. */
1093 struct timeval times
[2];
1094 times
[1].tv_sec
= p
->mtime
;
1095 times
[1].tv_usec
= p
->mtime_nanos
/ 1000;
1096 times
[0].tv_sec
= p
->atime
;
1097 times
[0].tv_usec
= p
->atime_nanos
/ 1000;
1099 lutimes(p
->name
, times
);
1101 utimes(p
->name
, times
);
1104 /* utime() is more portable, but less precise. */
1105 struct utimbuf times
;
1106 times
.modtime
= p
->mtime
;
1107 times
.actime
= p
->atime
;
1109 utime(p
->name
, ×
);
1112 if (p
->fixup
& TODO_MODE_BASE
)
1113 chmod(p
->name
, p
->mode
);
1115 if (p
->fixup
& TODO_FFLAGS
)
1116 set_fflags_platform(a
, -1, p
->name
,
1117 p
->mode
, p
->fflags_set
, 0);
1124 a
->fixup_list
= NULL
;
1129 _archive_write_finish(struct archive
*_a
)
1131 struct archive_write_disk
*a
= (struct archive_write_disk
*)_a
;
1133 ret
= _archive_write_close(&a
->archive
);
1134 if (a
->cleanup_gid
!= NULL
&& a
->lookup_gid_data
!= NULL
)
1135 (a
->cleanup_gid
)(a
->lookup_gid_data
);
1136 if (a
->cleanup_uid
!= NULL
&& a
->lookup_uid_data
!= NULL
)
1137 (a
->cleanup_uid
)(a
->lookup_uid_data
);
1138 archive_string_free(&a
->_name_data
);
1139 archive_string_free(&a
->archive
.error_string
);
1140 archive_string_free(&a
->path_safe
);
1146 * Simple O(n log n) merge sort to order the fixup list. In
1147 * particular, we want to restore dir timestamps depth-first.
1149 static struct fixup_entry
*
1150 sort_dir_list(struct fixup_entry
*p
)
1152 struct fixup_entry
*a
, *b
, *t
;
1156 /* A one-item list is already sorted. */
1157 if (p
->next
== NULL
)
1160 /* Step 1: split the list. */
1164 /* Step a twice, t once. */
1170 /* Now, t is at the mid-point, so break the list here. */
1175 /* Step 2: Recursively sort the two sub-lists. */
1176 a
= sort_dir_list(a
);
1177 b
= sort_dir_list(b
);
1179 /* Step 3: Merge the returned lists. */
1180 /* Pick the first element for the merged list. */
1181 if (strcmp(a
->name
, b
->name
) > 0) {
1189 /* Always put the later element on the list first. */
1190 while (a
!= NULL
&& b
!= NULL
) {
1191 if (strcmp(a
->name
, b
->name
) > 0) {
1201 /* Only one list is non-empty, so just splice it on. */
1211 * Returns a new, initialized fixup entry.
1213 * TODO: Reduce the memory requirements for this list by using a tree
1214 * structure rather than a simple list of names.
1216 static struct fixup_entry
*
1217 new_fixup(struct archive_write_disk
*a
, const char *pathname
)
1219 struct fixup_entry
*fe
;
1221 fe
= (struct fixup_entry
*)malloc(sizeof(struct fixup_entry
));
1224 fe
->next
= a
->fixup_list
;
1227 fe
->name
= strdup(pathname
);
1232 * Returns a fixup structure for the current entry.
1234 static struct fixup_entry
*
1235 current_fixup(struct archive_write_disk
*a
, const char *pathname
)
1237 if (a
->current_fixup
== NULL
)
1238 a
->current_fixup
= new_fixup(a
, pathname
);
1239 return (a
->current_fixup
);
1242 /* TODO: Make this work. */
1244 * TODO: The deep-directory support bypasses this; disable deep directory
1245 * support if we're doing symlink checks.
1248 * TODO: Someday, integrate this with the deep dir support; they both
1249 * scan the path and both can be optimized by comparing against other
1253 check_symlinks(struct archive_write_disk
*a
)
1261 * Guard against symlink tricks. Reject any archive entry whose
1262 * destination would be altered by a symlink.
1264 /* Whatever we checked last time doesn't need to be re-checked. */
1267 while ((*pn
!= '\0') && (*p
== *pn
))
1270 /* Keep going until we've checked the entire name. */
1271 while (pn
[0] != '\0' && (pn
[0] != '/' || pn
[1] != '\0')) {
1272 /* Skip the next path element. */
1273 while (*pn
!= '\0' && *pn
!= '/')
1277 /* Check that we haven't hit a symlink. */
1278 r
= lstat(a
->name
, &st
);
1280 /* We've hit a dir that doesn't exist; stop now. */
1281 if (errno
== ENOENT
)
1283 } else if (S_ISLNK(st
.st_mode
)) {
1286 * Last element is symlink; remove it
1287 * so we can overwrite it with the
1288 * item being extracted.
1290 if (unlink(a
->name
)) {
1291 archive_set_error(&a
->archive
, errno
,
1292 "Could not remove symlink %s",
1295 return (ARCHIVE_WARN
);
1299 * Even if we did remove it, a warning
1300 * is in order. The warning is silly,
1301 * though, if we're just replacing one
1302 * symlink with another symlink.
1304 if (!S_ISLNK(a
->mode
)) {
1305 archive_set_error(&a
->archive
, 0,
1306 "Removing symlink %s",
1309 /* Symlink gone. No more problem! */
1312 } else if (a
->flags
& ARCHIVE_EXTRACT_UNLINK
) {
1313 /* User asked us to remove problems. */
1314 if (unlink(a
->name
) != 0) {
1315 archive_set_error(&a
->archive
, 0,
1316 "Cannot remove intervening symlink %s",
1319 return (ARCHIVE_WARN
);
1323 archive_set_error(&a
->archive
, 0,
1324 "Cannot extract through symlink %s",
1327 return (ARCHIVE_WARN
);
1332 /* We've checked and/or cleaned the whole path, so remember it. */
1333 archive_strcpy(&a
->path_safe
, a
->name
);
1334 return (ARCHIVE_OK
);
1338 * Canonicalize the pathname. In particular, this strips duplicate
1339 * '/' characters, '.' elements, and trailing '/'. It also raises an
1340 * error for an empty path, a trailing '..' or (if _SECURE_NODOTDOT is
1341 * set) any '..' in the path.
1344 cleanup_pathname(struct archive_write_disk
*a
)
1347 char separator
= '\0';
1348 int lastdotdot
= 0; /* True if last elt copied was '..' */
1350 dest
= src
= a
->name
;
1352 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_MISC
,
1353 "Invalid empty pathname");
1354 return (ARCHIVE_WARN
);
1357 /* Skip leading '/'. */
1361 /* Scan the pathname one element at a time. */
1363 /* src points to first char after '/' */
1364 if (src
[0] == '\0') {
1366 } else if (src
[0] == '/') {
1367 /* Found '//', ignore second one. */
1370 } else if (src
[0] == '.') {
1371 if (src
[1] == '\0') {
1372 /* Ignore trailing '.' */
1374 } else if (src
[1] == '/') {
1378 } else if (src
[1] == '.') {
1379 if (src
[2] == '/' || src
[2] == '\0') {
1380 /* Conditionally warn about '..' */
1381 if (a
->flags
& ARCHIVE_EXTRACT_SECURE_NODOTDOT
) {
1382 archive_set_error(&a
->archive
,
1384 "Path contains '..'");
1385 return (ARCHIVE_WARN
);
1391 * Note: Under no circumstances do we
1392 * remove '..' elements. In
1393 * particular, restoring
1394 * '/foo/../bar/' should create the
1395 * 'foo' dir as a side-effect.
1402 /* Copy current element, including leading '/'. */
1405 while (*src
!= '\0' && *src
!= '/') {
1412 /* Skip '/' separator. */
1416 * We've just copied zero or more path elements, not including the
1420 /* Trailing '..' is always wrong. */
1421 archive_set_error(&a
->archive
,
1423 "Path contains trailing '..'");
1424 return (ARCHIVE_WARN
);
1426 if (dest
== a
->name
) {
1428 * Nothing got copied. The path must have been something
1429 * like '.' or '/' or './' or '/././././/./'.
1436 /* Terminate the result. */
1438 return (ARCHIVE_OK
);
1442 * Create the parent directory of the specified path, assuming path
1443 * is already in mutable storage.
1446 create_parent_dir(struct archive_write_disk
*a
, char *path
)
1451 /* Remove tail element to obtain parent name. */
1452 slash
= strrchr(path
, '/');
1454 return (ARCHIVE_OK
);
1456 r
= create_dir(a
, path
);
1462 * Create the specified dir, recursing to create parents as necessary.
1464 * Returns ARCHIVE_OK if the path exists when we're done here.
1465 * Otherwise, returns ARCHIVE_WARN.
1466 * Assumes path is in mutable storage; path is unchanged on exit.
1469 create_dir(struct archive_write_disk
*a
, char *path
)
1472 struct fixup_entry
*le
;
1474 mode_t mode_final
, mode
;
1479 /* Check for special names and just skip them. */
1480 slash
= strrchr(path
, '/');
1486 if (base
[0] == '\0' ||
1487 (base
[0] == '.' && base
[1] == '\0') ||
1488 (base
[0] == '.' && base
[1] == '.' && base
[2] == '\0')) {
1489 /* Don't bother trying to create null path, '.', or '..'. */
1490 if (slash
!= NULL
) {
1492 r
= create_dir(a
, path
);
1496 return (ARCHIVE_OK
);
1500 * Yes, this should be stat() and not lstat(). Using lstat()
1501 * here loses the ability to extract through symlinks. Also note
1502 * that this should not use the a->st cache.
1504 if (stat(path
, &st
) == 0) {
1505 if (S_ISDIR(st
.st_mode
))
1506 return (ARCHIVE_OK
);
1507 if ((a
->flags
& ARCHIVE_EXTRACT_NO_OVERWRITE
)) {
1508 archive_set_error(&a
->archive
, EEXIST
,
1509 "Can't create directory '%s'", path
);
1510 return (ARCHIVE_WARN
);
1512 if (unlink(path
) != 0) {
1513 archive_set_error(&a
->archive
, errno
,
1514 "Can't create directory '%s': "
1515 "Conflicting file cannot be removed");
1516 return (ARCHIVE_WARN
);
1518 } else if (errno
!= ENOENT
&& errno
!= ENOTDIR
) {
1520 archive_set_error(&a
->archive
, errno
, "Can't test directory '%s'", path
);
1521 return (ARCHIVE_WARN
);
1522 } else if (slash
!= NULL
) {
1524 r
= create_dir(a
, path
);
1526 if (r
!= ARCHIVE_OK
)
1531 * Mode we want for the final restored directory. Per POSIX,
1532 * implicitly-created dirs must be created obeying the umask.
1533 * There's no mention whether this is different for privileged
1534 * restores (which the rest of this code handles by pretending
1535 * umask=0). I've chosen here to always obey the user's umask for
1536 * implicit dirs, even if _EXTRACT_PERM was specified.
1538 mode_final
= DEFAULT_DIR_MODE
& ~a
->user_umask
;
1539 /* Mode we want on disk during the restore process. */
1541 mode
|= MINIMUM_DIR_MODE
;
1542 mode
&= MAXIMUM_DIR_MODE
;
1543 if (mkdir(path
, mode
) == 0) {
1544 if (mode
!= mode_final
) {
1545 le
= new_fixup(a
, path
);
1546 le
->fixup
|=TODO_MODE_BASE
;
1547 le
->mode
= mode_final
;
1549 return (ARCHIVE_OK
);
1553 * Without the following check, a/b/../b/c/d fails at the
1554 * second visit to 'b', so 'd' can't be created. Note that we
1555 * don't add it to the fixup list here, as it's already been
1558 if (stat(path
, &st
) == 0 && S_ISDIR(st
.st_mode
))
1559 return (ARCHIVE_OK
);
1561 archive_set_error(&a
->archive
, errno
, "Failed to create dir '%s'", path
);
1562 return (ARCHIVE_WARN
);
1566 * Note: Although we can skip setting the user id if the desired user
1567 * id matches the current user, we cannot skip setting the group, as
1568 * many systems set the gid bit based on the containing directory. So
1569 * we have to perform a chown syscall if we want to restore the SGID
1570 * bit. (The alternative is to stat() and then possibly chown(); it's
1571 * more efficient to skip the stat() and just always chown().) Note
1572 * that a successful chown() here clears the TODO_SGID_CHECK bit, which
1573 * allows set_mode to skip the stat() check for the GID.
1576 set_ownership(struct archive_write_disk
*a
)
1578 /* If we know we can't change it, don't bother trying. */
1579 if (a
->user_uid
!= 0 && a
->user_uid
!= a
->uid
) {
1580 archive_set_error(&a
->archive
, errno
,
1581 "Can't set UID=%d", a
->uid
);
1582 return (ARCHIVE_WARN
);
1586 /* If we have an fd, we can avoid a race. */
1587 if (a
->fd
>= 0 && fchown(a
->fd
, a
->uid
, a
->gid
) == 0) {
1588 /* We've set owner and know uid/gid are correct. */
1589 a
->todo
&= ~(TODO_OWNER
| TODO_SGID_CHECK
| TODO_SUID_CHECK
);
1590 return (ARCHIVE_OK
);
1594 /* We prefer lchown() but will use chown() if that's all we have. */
1595 /* Of course, if we have neither, this will always fail. */
1597 if (lchown(a
->name
, a
->uid
, a
->gid
) == 0) {
1598 /* We've set owner and know uid/gid are correct. */
1599 a
->todo
&= ~(TODO_OWNER
| TODO_SGID_CHECK
| TODO_SUID_CHECK
);
1600 return (ARCHIVE_OK
);
1603 if (!S_ISLNK(a
->mode
) && chown(a
->name
, a
->uid
, a
->gid
) == 0) {
1604 /* We've set owner and know uid/gid are correct. */
1605 a
->todo
&= ~(TODO_OWNER
| TODO_SGID_CHECK
| TODO_SUID_CHECK
);
1606 return (ARCHIVE_OK
);
1610 archive_set_error(&a
->archive
, errno
,
1611 "Can't set user=%d/group=%d for %s", a
->uid
, a
->gid
,
1613 return (ARCHIVE_WARN
);
1618 * The utimes()-family functions provide high resolution and
1619 * a way to set time on an fd or a symlink. We prefer them
1620 * when they're available.
1623 set_time(struct archive_write_disk
*a
)
1625 struct timeval times
[2];
1627 times
[1].tv_sec
= archive_entry_mtime(a
->entry
);
1628 times
[1].tv_usec
= archive_entry_mtime_nsec(a
->entry
) / 1000;
1630 times
[0].tv_sec
= archive_entry_atime(a
->entry
);
1631 times
[0].tv_usec
= archive_entry_atime_nsec(a
->entry
) / 1000;
1634 if (a
->fd
>= 0 && futimes(a
->fd
, times
) == 0) {
1635 return (ARCHIVE_OK
);
1640 if (lutimes(a
->name
, times
) != 0)
1642 if (!S_ISLNK(a
->mode
) && utimes(a
->name
, times
) != 0)
1645 archive_set_error(&a
->archive
, errno
, "Can't update time for %s",
1647 return (ARCHIVE_WARN
);
1651 * Note: POSIX does not provide a portable way to restore ctime.
1652 * (Apart from resetting the system clock, which is distasteful.)
1653 * So, any restoration of ctime will necessarily be OS-specific.
1656 /* XXX TODO: Can FreeBSD restore ctime? XXX */
1657 return (ARCHIVE_OK
);
1659 #elif defined(HAVE_UTIME)
1661 * utime() is an older, more standard interface that we'll use
1662 * if utimes() isn't available.
1665 set_time(struct archive_write_disk
*a
)
1667 struct utimbuf times
;
1669 times
.modtime
= archive_entry_mtime(a
->entry
);
1670 times
.actime
= archive_entry_atime(a
->entry
);
1671 if (!S_ISLNK(a
->mode
) && utime(a
->name
, ×
) != 0) {
1672 archive_set_error(&a
->archive
, errno
,
1673 "Can't update time for %s", a
->name
);
1674 return (ARCHIVE_WARN
);
1676 return (ARCHIVE_OK
);
1679 /* This platform doesn't give us a way to restore the time. */
1681 set_time(struct archive_write_disk
*a
)
1683 (void)a
; /* UNUSED */
1684 archive_set_error(&a
->archive
, errno
,
1685 "Can't update time for %s", a
->name
);
1686 return (ARCHIVE_WARN
);
1692 set_mode(struct archive_write_disk
*a
, int mode
)
1695 mode
&= 07777; /* Strip off file type bits. */
1697 if (a
->todo
& TODO_SGID_CHECK
) {
1699 * If we don't know the GID is right, we must stat()
1700 * to verify it. We can't just check the GID of this
1701 * process, since systems sometimes set GID from
1702 * the enclosing dir or based on ACLs.
1704 if ((r
= _archive_write_disk_lazy_stat(a
)) != ARCHIVE_OK
)
1706 if (a
->pst
->st_gid
!= a
->gid
) {
1708 if (a
->flags
& ARCHIVE_EXTRACT_OWNER
) {
1710 * This is only an error if you
1711 * requested owner restore. If you
1712 * didn't, we'll try to restore
1713 * sgid/suid, but won't consider it a
1714 * problem if we can't.
1716 archive_set_error(&a
->archive
, -1,
1717 "Can't restore SGID bit");
1721 /* While we're here, double-check the UID. */
1722 if (a
->pst
->st_uid
!= a
->uid
1723 && (a
->todo
& TODO_SUID
)) {
1725 if (a
->flags
& ARCHIVE_EXTRACT_OWNER
) {
1726 archive_set_error(&a
->archive
, -1,
1727 "Can't restore SUID bit");
1731 a
->todo
&= ~TODO_SGID_CHECK
;
1732 a
->todo
&= ~TODO_SUID_CHECK
;
1733 } else if (a
->todo
& TODO_SUID_CHECK
) {
1735 * If we don't know the UID is right, we can just check
1736 * the user, since all systems set the file UID from
1739 if (a
->user_uid
!= a
->uid
) {
1741 if (a
->flags
& ARCHIVE_EXTRACT_OWNER
) {
1742 archive_set_error(&a
->archive
, -1,
1743 "Can't make file SUID");
1747 a
->todo
&= ~TODO_SUID_CHECK
;
1750 if (S_ISLNK(a
->mode
)) {
1753 * If this is a symlink, use lchmod(). If the
1754 * platform doesn't support lchmod(), just skip it. A
1755 * platform that doesn't provide a way to set
1756 * permissions on symlinks probably ignores
1757 * permissions on symlinks, so a failure here has no
1760 if (lchmod(a
->name
, mode
) != 0) {
1761 archive_set_error(&a
->archive
, errno
,
1762 "Can't set permissions to 0%o", (int)mode
);
1766 } else if (!S_ISDIR(a
->mode
)) {
1768 * If it's not a symlink and not a dir, then use
1769 * fchmod() or chmod(), depending on whether we have
1770 * an fd. Dirs get their perms set during the
1771 * post-extract fixup, which is handled elsewhere.
1775 if (fchmod(a
->fd
, mode
) != 0) {
1776 archive_set_error(&a
->archive
, errno
,
1777 "Can't set permissions to 0%o", (int)mode
);
1782 /* If this platform lacks fchmod(), then
1783 * we'll just use chmod(). */
1784 if (chmod(a
->name
, mode
) != 0) {
1785 archive_set_error(&a
->archive
, errno
,
1786 "Can't set permissions to 0%o", (int)mode
);
1794 set_fflags(struct archive_write_disk
*a
)
1796 struct fixup_entry
*le
;
1797 unsigned long set
, clear
;
1800 mode_t mode
= archive_entry_mode(a
->entry
);
1803 * Make 'critical_flags' hold all file flags that can't be
1804 * immediately restored. For example, on BSD systems,
1805 * SF_IMMUTABLE prevents hardlinks from being created, so
1806 * should not be set until after any hardlinks are created. To
1807 * preserve some semblance of portability, this uses #ifdef
1808 * extensively. Ugly, but it works.
1810 * Yes, Virginia, this does create a security race. It's mitigated
1811 * somewhat by the practice of creating dirs 0700 until the extract
1812 * is done, but it would be nice if we could do more than that.
1813 * People restoring critical file systems should be wary of
1814 * other programs that might try to muck with files as they're
1817 /* Hopefully, the compiler will optimize this mess into a constant. */
1820 critical_flags
|= SF_IMMUTABLE
;
1823 critical_flags
|= UF_IMMUTABLE
;
1826 critical_flags
|= SF_APPEND
;
1829 critical_flags
|= UF_APPEND
;
1831 #ifdef EXT2_APPEND_FL
1832 critical_flags
|= EXT2_APPEND_FL
;
1834 #ifdef EXT2_IMMUTABLE_FL
1835 critical_flags
|= EXT2_IMMUTABLE_FL
;
1838 if (a
->todo
& TODO_FFLAGS
) {
1839 archive_entry_fflags(a
->entry
, &set
, &clear
);
1842 * The first test encourages the compiler to eliminate
1843 * all of this if it's not necessary.
1845 if ((critical_flags
!= 0) && (set
& critical_flags
)) {
1846 le
= current_fixup(a
, a
->name
);
1847 le
->fixup
|= TODO_FFLAGS
;
1848 le
->fflags_set
= set
;
1849 /* Store the mode if it's not already there. */
1850 if ((le
->fixup
& TODO_MODE
) == 0)
1853 r
= set_fflags_platform(a
, a
->fd
,
1854 a
->name
, mode
, set
, clear
);
1855 if (r
!= ARCHIVE_OK
)
1859 return (ARCHIVE_OK
);
1863 #if ( defined(HAVE_LCHFLAGS) || defined(HAVE_CHFLAGS) || defined(HAVE_FCHFLAGS) ) && !defined(__linux)
1865 set_fflags_platform(struct archive_write_disk
*a
, int fd
, const char *name
,
1866 mode_t mode
, unsigned long set
, unsigned long clear
)
1870 (void)mode
; /* UNUSED */
1871 if (set
== 0 && clear
== 0)
1872 return (ARCHIVE_OK
);
1875 * XXX Is the stat here really necessary? Or can I just use
1876 * the 'set' flags directly? In particular, I'm not sure
1877 * about the correct approach if we're overwriting an existing
1878 * file that already has flags on it. XXX
1880 if ((r
= _archive_write_disk_lazy_stat(a
)) != ARCHIVE_OK
)
1883 a
->st
.st_flags
&= ~clear
;
1884 a
->st
.st_flags
|= set
;
1885 #ifdef HAVE_FCHFLAGS
1886 /* If platform has fchflags() and we were given an fd, use it. */
1887 if (fd
>= 0 && fchflags(fd
, a
->st
.st_flags
) == 0)
1888 return (ARCHIVE_OK
);
1891 * If we can't use the fd to set the flags, we'll use the
1892 * pathname to set flags. We prefer lchflags() but will use
1893 * chflags() if we must.
1895 #ifdef HAVE_LCHFLAGS
1896 if (lchflags(name
, a
->st
.st_flags
) == 0)
1897 return (ARCHIVE_OK
);
1898 #elif defined(HAVE_CHFLAGS)
1899 if (S_ISLNK(a
->st
.st_mode
)) {
1900 archive_set_error(&a
->archive
, errno
,
1901 "Can't set file flags on symlink.");
1902 return (ARCHIVE_WARN
);
1904 if (chflags(name
, a
->st
.st_flags
) == 0)
1905 return (ARCHIVE_OK
);
1907 archive_set_error(&a
->archive
, errno
,
1908 "Failed to set file flags");
1909 return (ARCHIVE_WARN
);
1912 #elif defined(__linux) && defined(EXT2_IOC_GETFLAGS) && defined(EXT2_IOC_SETFLAGS)
1915 * Linux has flags too, but uses ioctl() to access them instead of
1916 * having a separate chflags() system call.
1919 set_fflags_platform(struct archive_write_disk
*a
, int fd
, const char *name
,
1920 mode_t mode
, unsigned long set
, unsigned long clear
)
1924 unsigned long newflags
, oldflags
;
1925 unsigned long sf_mask
= 0;
1927 if (set
== 0 && clear
== 0)
1928 return (ARCHIVE_OK
);
1929 /* Only regular files and dirs can have flags. */
1930 if (!S_ISREG(mode
) && !S_ISDIR(mode
))
1931 return (ARCHIVE_OK
);
1933 /* If we weren't given an fd, open it ourselves. */
1935 myfd
= open(name
, O_RDONLY
| O_NONBLOCK
| O_BINARY
);
1937 return (ARCHIVE_OK
);
1940 * Linux has no define for the flags that are only settable by
1941 * the root user. This code may seem a little complex, but
1942 * there seem to be some Linux systems that lack these
1943 * defines. (?) The code below degrades reasonably gracefully
1944 * if sf_mask is incomplete.
1946 #ifdef EXT2_IMMUTABLE_FL
1947 sf_mask
|= EXT2_IMMUTABLE_FL
;
1949 #ifdef EXT2_APPEND_FL
1950 sf_mask
|= EXT2_APPEND_FL
;
1953 * XXX As above, this would be way simpler if we didn't have
1954 * to read the current flags from disk. XXX
1957 /* Try setting the flags as given. */
1958 if (ioctl(myfd
, EXT2_IOC_GETFLAGS
, &oldflags
) >= 0) {
1959 newflags
= (oldflags
& ~clear
) | set
;
1960 if (ioctl(myfd
, EXT2_IOC_SETFLAGS
, &newflags
) >= 0)
1965 /* If we couldn't set all the flags, try again with a subset. */
1966 if (ioctl(myfd
, EXT2_IOC_GETFLAGS
, &oldflags
) >= 0) {
1967 newflags
&= ~sf_mask
;
1968 oldflags
&= sf_mask
;
1969 newflags
|= oldflags
;
1970 if (ioctl(myfd
, EXT2_IOC_SETFLAGS
, &newflags
) >= 0)
1973 /* We couldn't set the flags, so report the failure. */
1975 archive_set_error(&a
->archive
, errno
,
1976 "Failed to set file flags");
1984 #else /* Not HAVE_CHFLAGS && Not __linux */
1987 * Of course, some systems have neither BSD chflags() nor Linux' flags
1988 * support through ioctl().
1991 set_fflags_platform(struct archive_write_disk
*a
, int fd
, const char *name
,
1992 mode_t mode
, unsigned long set
, unsigned long clear
)
1994 (void)a
; /* UNUSED */
1995 (void)fd
; /* UNUSED */
1996 (void)name
; /* UNUSED */
1997 (void)mode
; /* UNUSED */
1998 (void)set
; /* UNUSED */
1999 (void)clear
; /* UNUSED */
2000 return (ARCHIVE_OK
);
2003 #endif /* __linux */
2005 #ifndef HAVE_POSIX_ACL
2006 /* Default empty function body to satisfy mainline code. */
2008 set_acls(struct archive_write_disk
*a
)
2010 (void)a
; /* UNUSED */
2011 return (ARCHIVE_OK
);
2017 * XXX TODO: What about ACL types other than ACCESS and DEFAULT?
2020 set_acls(struct archive_write_disk
*a
)
2024 ret
= set_acl(a
, a
->fd
, a
->entry
, ACL_TYPE_ACCESS
,
2025 ARCHIVE_ENTRY_ACL_TYPE_ACCESS
, "access");
2026 if (ret
!= ARCHIVE_OK
)
2028 ret
= set_acl(a
, a
->fd
, a
->entry
, ACL_TYPE_DEFAULT
,
2029 ARCHIVE_ENTRY_ACL_TYPE_DEFAULT
, "default");
2035 set_acl(struct archive_write_disk
*a
, int fd
, struct archive_entry
*entry
,
2036 acl_type_t acl_type
, int ae_requested_type
, const char *tname
)
2039 acl_entry_t acl_entry
;
2040 acl_permset_t acl_permset
;
2042 int ae_type
, ae_permset
, ae_tag
, ae_id
;
2045 const char *ae_name
;
2050 entries
= archive_entry_acl_reset(entry
, ae_requested_type
);
2052 return (ARCHIVE_OK
);
2053 acl
= acl_init(entries
);
2054 while (archive_entry_acl_next(entry
, ae_requested_type
, &ae_type
,
2055 &ae_permset
, &ae_tag
, &ae_id
, &ae_name
) == ARCHIVE_OK
) {
2056 acl_create_entry(&acl
, &acl_entry
);
2059 case ARCHIVE_ENTRY_ACL_USER
:
2060 acl_set_tag_type(acl_entry
, ACL_USER
);
2061 ae_uid
= a
->lookup_uid(a
->lookup_uid_data
,
2063 acl_set_qualifier(acl_entry
, &ae_uid
);
2065 case ARCHIVE_ENTRY_ACL_GROUP
:
2066 acl_set_tag_type(acl_entry
, ACL_GROUP
);
2067 ae_gid
= a
->lookup_gid(a
->lookup_gid_data
,
2069 acl_set_qualifier(acl_entry
, &ae_gid
);
2071 case ARCHIVE_ENTRY_ACL_USER_OBJ
:
2072 acl_set_tag_type(acl_entry
, ACL_USER_OBJ
);
2074 case ARCHIVE_ENTRY_ACL_GROUP_OBJ
:
2075 acl_set_tag_type(acl_entry
, ACL_GROUP_OBJ
);
2077 case ARCHIVE_ENTRY_ACL_MASK
:
2078 acl_set_tag_type(acl_entry
, ACL_MASK
);
2080 case ARCHIVE_ENTRY_ACL_OTHER
:
2081 acl_set_tag_type(acl_entry
, ACL_OTHER
);
2088 acl_get_permset(acl_entry
, &acl_permset
);
2089 acl_clear_perms(acl_permset
);
2090 if (ae_permset
& ARCHIVE_ENTRY_ACL_EXECUTE
)
2091 acl_add_perm(acl_permset
, ACL_EXECUTE
);
2092 if (ae_permset
& ARCHIVE_ENTRY_ACL_WRITE
)
2093 acl_add_perm(acl_permset
, ACL_WRITE
);
2094 if (ae_permset
& ARCHIVE_ENTRY_ACL_READ
)
2095 acl_add_perm(acl_permset
, ACL_READ
);
2098 name
= archive_entry_pathname(entry
);
2100 /* Try restoring the ACL through 'fd' if we can. */
2102 if (fd
>= 0 && acl_type
== ACL_TYPE_ACCESS
&& acl_set_fd(fd
, acl
) == 0)
2106 #if HAVE_ACL_SET_FD_NP
2107 if (fd
>= 0 && acl_set_fd_np(fd
, acl
, acl_type
) == 0)
2112 if (acl_set_file(name
, acl_type
, acl
) != 0) {
2113 archive_set_error(&a
->archive
, errno
, "Failed to set %s acl", tname
);
2123 * Restore extended attributes - Linux implementation
2126 set_xattrs(struct archive_write_disk
*a
)
2128 struct archive_entry
*entry
= a
->entry
;
2129 static int warning_done
= 0;
2130 int ret
= ARCHIVE_OK
;
2131 int i
= archive_entry_xattr_reset(entry
);
2137 archive_entry_xattr_next(entry
, &name
, &value
, &size
);
2139 strncmp(name
, "xfsroot.", 8) != 0 &&
2140 strncmp(name
, "system.", 7) != 0) {
2144 e
= fsetxattr(a
->fd
, name
, value
, size
, 0);
2148 e
= lsetxattr(archive_entry_pathname(entry
),
2149 name
, value
, size
, 0);
2152 if (errno
== ENOTSUP
) {
2153 if (!warning_done
) {
2155 archive_set_error(&a
->archive
, errno
,
2156 "Cannot restore extended "
2157 "attributes on this file "
2161 archive_set_error(&a
->archive
, errno
,
2162 "Failed to set extended attribute");
2166 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_FILE_FORMAT
,
2167 "Invalid extended attribute encountered");
2175 * Restore extended attributes - stub implementation for unsupported systems
2178 set_xattrs(struct archive_write_disk
*a
)
2180 static int warning_done
= 0;
2182 /* If there aren't any extended attributes, then it's okay not
2183 * to extract them, otherwise, issue a single warning. */
2184 if (archive_entry_xattr_count(a
->entry
) != 0 && !warning_done
) {
2186 archive_set_error(&a
->archive
, ARCHIVE_ERRNO_FILE_FORMAT
,
2187 "Cannot restore extended attributes on this system");
2188 return (ARCHIVE_WARN
);
2190 /* Warning was already emitted; suppress further warnings. */
2191 return (ARCHIVE_OK
);
2197 * Trivial implementations of gid/uid lookup functions.
2198 * These are normally overridden by the client, but these stub
2199 * versions ensure that we always have something that works.
2202 trivial_lookup_gid(void *private_data
, const char *gname
, gid_t gid
)
2204 (void)private_data
; /* UNUSED */
2205 (void)gname
; /* UNUSED */
2210 trivial_lookup_uid(void *private_data
, const char *uname
, uid_t uid
)
2212 (void)private_data
; /* UNUSED */
2213 (void)uname
; /* UNUSED */
2218 * Test if file on disk is older than entry.
2221 older(struct stat
*st
, struct archive_entry
*entry
)
2223 /* First, test the seconds and return if we have a definite answer. */
2224 /* Definitely older. */
2225 if (st
->st_mtime
< archive_entry_mtime(entry
))
2227 /* Definitely younger. */
2228 if (st
->st_mtime
> archive_entry_mtime(entry
))
2230 /* If this platform supports fractional seconds, try those. */
2231 #if HAVE_STRUCT_STAT_ST_MTIMESPEC_TV_NSEC
2232 /* Definitely older. */
2233 if (st
->st_mtimespec
.tv_nsec
< archive_entry_mtime_nsec(entry
))
2235 /* Definitely younger. */
2236 if (st
->st_mtimespec
.tv_nsec
> archive_entry_mtime_nsec(entry
))
2238 #elif HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
2239 /* Definitely older. */
2240 if (st
->st_mtim
.tv_nsec
< archive_entry_mtime_nsec(entry
))
2242 /* Definitely older. */
2243 if (st
->st_mtim
.tv_nsec
> archive_entry_mtime_nsec(entry
))
2246 /* This system doesn't have high-res timestamps. */
2248 /* Same age, so not older. */