]> wimlib.net Git - wimlib/blobdiff - src/unix_apply.c
unix_apply.c: fix applying security xattrs
[wimlib] / src / unix_apply.c
index 026042bce7219fe1baecfcf5d726b7d40790d73f..daadfc9ec15d31c86a90251fbe699c7a2eb40668 100644 (file)
  */
 
 /*
- * Copyright (C) 2012, 2013 Eric Biggers
+ * Copyright (C) 2012-2016 Eric Biggers
  *
- * This file is part of wimlib, a library for working with WIM files.
+ * This file is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the Free
+ * Software Foundation; either version 3 of the License, or (at your option) any
+ * later version.
  *
- * wimlib is free software; you can redistribute it and/or modify it under the
- * terms of the GNU General Public License as published by the Free
- * Software Foundation; either version 3 of the License, or (at your option)
- * any later version.
- *
- * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
- * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
- * A PARTICULAR PURPOSE. See the GNU General Public License for more
+ * This file is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
  * details.
  *
- * You should have received a copy of the GNU General Public License
- * along with wimlib; if not, see http://www.gnu.org/licenses/.
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this file; if not, see http://www.gnu.org/licenses/.
  */
 
-#ifndef __WIN32__
-
 #ifdef HAVE_CONFIG_H
 #  include "config.h"
 #endif
 
-#include "wimlib/apply.h"
-#include "wimlib/error.h"
-#include "wimlib/lookup_table.h"
-#include "wimlib/resource.h"
-#include "wimlib/timestamp.h"
-
 #include <errno.h>
 #include <fcntl.h>
-#include <limits.h>
+#include <stdlib.h>
 #include <sys/stat.h>
+#include <sys/time.h>
 #include <sys/types.h>
+#ifdef HAVE_SYS_XATTR_H
+#  include <sys/xattr.h>
+#endif
 #include <unistd.h>
 
-#ifdef HAVE_UTIME_H
-#  include <utime.h>
+#include "wimlib/apply.h"
+#include "wimlib/assert.h"
+#include "wimlib/blob_table.h"
+#include "wimlib/dentry.h"
+#include "wimlib/error.h"
+#include "wimlib/file_io.h"
+#include "wimlib/reparse.h"
+#include "wimlib/timestamp.h"
+#include "wimlib/unix_data.h"
+#include "wimlib/xattr.h"
+
+/* We don't require O_NOFOLLOW, but the advantage of having it is that if we
+ * need to extract a file to a location at which there exists a symbolic link,
+ * open(..., O_NOFOLLOW | ...) recognizes the symbolic link rather than
+ * following it and creating the file somewhere else.  (Equivalent to
+ * FILE_OPEN_REPARSE_POINT on Windows.)  */
+#ifndef O_NOFOLLOW
+#  define O_NOFOLLOW 0
 #endif
 
 static int
-unix_start_extract(const char *target, struct apply_ctx *ctx)
+unix_get_supported_features(const char *target,
+                           struct wim_features *supported_features)
 {
-       ctx->supported_features.hard_links = 1;
-       ctx->supported_features.symlink_reparse_points = 1;
-       ctx->supported_features.unix_data = 1;
+       supported_features->sparse_files = 1;
+       supported_features->hard_links = 1;
+       supported_features->symlink_reparse_points = 1;
+       supported_features->unix_data = 1;
+       supported_features->timestamps = 1;
+       supported_features->case_sensitive_filenames = 1;
+#ifdef HAVE_XATTR_SUPPORT
+       supported_features->linux_xattrs = 1;
+#endif
        return 0;
 }
 
+#define NUM_PATHBUFS 2  /* We need 2 when creating hard links  */
+
+struct unix_apply_ctx {
+       /* Extract flags, the pointer to the WIMStruct, etc.  */
+       struct apply_ctx common;
+
+       /* Buffers for building extraction paths (allocated).  */
+       char *pathbufs[NUM_PATHBUFS];
+
+       /* Index of next pathbuf to use  */
+       unsigned which_pathbuf;
+
+       /* Currently open file descriptors for extraction  */
+       struct filedes open_fds[MAX_OPEN_FILES];
+
+       /* Number of currently open file descriptors in open_fds, starting from
+        * the beginning of the array.  */
+       unsigned num_open_fds;
+
+       /* For each currently open file, whether we're writing to it in "sparse"
+        * mode or not.  */
+       bool is_sparse_file[MAX_OPEN_FILES];
+
+       /* Whether is_sparse_file[] is true for any currently open file  */
+       bool any_sparse_files;
+
+       /* Buffer for reading reparse point data into memory  */
+       u8 reparse_data[REPARSE_DATA_MAX_SIZE];
+
+       /* Pointer to the next byte in @reparse_data to fill  */
+       u8 *reparse_ptr;
+
+       /* Absolute path to the target directory (allocated buffer).  Only set
+        * if needed for absolute symbolic link fixups.  */
+       char *target_abspath;
+
+       /* Number of characters in target_abspath.  */
+       size_t target_abspath_nchars;
+
+       /* Number of special files we couldn't create due to EPERM  */
+       unsigned long num_special_files_ignored;
+};
+
+/* Returns the number of characters needed to represent the path to the
+ * specified @dentry when extracted, not including the null terminator or the
+ * path to the target directory itself.  */
+static size_t
+unix_dentry_path_length(const struct wim_dentry *dentry)
+{
+       size_t len = 0;
+       const struct wim_dentry *d;
+
+       d = dentry;
+       do {
+               len += d->d_extraction_name_nchars + 1;
+               d = d->d_parent;
+       } while (!dentry_is_root(d) && will_extract_dentry(d));
+
+       return len;
+}
+
+/* Returns the maximum number of characters needed to represent the path to any
+ * dentry in @dentry_list when extracted, including the null terminator and the
+ * path to the target directory itself.  */
+static size_t
+unix_compute_path_max(const struct list_head *dentry_list,
+                     const struct unix_apply_ctx *ctx)
+{
+       size_t max = 0;
+       size_t len;
+       const struct wim_dentry *dentry;
+
+       list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
+               len = unix_dentry_path_length(dentry);
+               if (len > max)
+                       max = len;
+       }
+
+       /* Account for target and null terminator.  */
+       return ctx->common.target_nchars + max + 1;
+}
+
+/* Builds and returns the filesystem path to which to extract @dentry.
+ * This cycles through NUM_PATHBUFS different buffers.  */
+static const char *
+unix_build_extraction_path(const struct wim_dentry *dentry,
+                          struct unix_apply_ctx *ctx)
+{
+       char *pathbuf;
+       char *p;
+       const struct wim_dentry *d;
+
+       pathbuf = ctx->pathbufs[ctx->which_pathbuf];
+       ctx->which_pathbuf = (ctx->which_pathbuf + 1) % NUM_PATHBUFS;
+
+       p = &pathbuf[ctx->common.target_nchars +
+                    unix_dentry_path_length(dentry)];
+       *p = '\0';
+       d = dentry;
+       do {
+               p -= d->d_extraction_name_nchars;
+               if (d->d_extraction_name_nchars)
+                       memcpy(p, d->d_extraction_name,
+                              d->d_extraction_name_nchars);
+               *--p = '/';
+               d = d->d_parent;
+       } while (!dentry_is_root(d) && will_extract_dentry(d));
+
+       return pathbuf;
+}
+
+/* This causes the next call to unix_build_extraction_path() to use the same
+ * path buffer as the previous call.  */
+static void
+unix_reuse_pathbuf(struct unix_apply_ctx *ctx)
+{
+       ctx->which_pathbuf = (ctx->which_pathbuf - 1) % NUM_PATHBUFS;
+}
+
+/* Builds and returns the filesystem path to which to extract an unspecified
+ * alias of the @inode.  This cycles through NUM_PATHBUFS different buffers.  */
+static const char *
+unix_build_inode_extraction_path(const struct wim_inode *inode,
+                                struct unix_apply_ctx *ctx)
+{
+       return unix_build_extraction_path(inode_first_extraction_dentry(inode), ctx);
+}
+
+/* Should the specified file be extracted as a directory on UNIX?  We extract
+ * the file as a directory if FILE_ATTRIBUTE_DIRECTORY is set and the file does
+ * not have a symlink or junction reparse point.  It *may* have a different type
+ * of reparse point.  */
+static inline bool
+should_extract_as_directory(const struct wim_inode *inode)
+{
+       return (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) &&
+               !inode_is_symlink(inode);
+}
+
+/* Sets the timestamps on a file being extracted.
+ *
+ * Either @fd or @path must be specified (not -1 and not NULL, respectively).
+ */
 static int
-unix_create_file(const char *path, struct apply_ctx *ctx, u64 *cookie_ret)
+unix_set_timestamps(int fd, const char *path, u64 atime, u64 mtime)
 {
-       int fd = open(path, O_TRUNC | O_CREAT | O_WRONLY, 0644);
-       if (fd < 0)
-               return WIMLIB_ERR_OPEN;
-       close(fd);
+       {
+               struct timespec times[2];
+
+               times[0] = wim_timestamp_to_timespec(atime);
+               times[1] = wim_timestamp_to_timespec(mtime);
+
+               errno = ENOSYS;
+#ifdef HAVE_FUTIMENS
+               if (fd >= 0 && !futimens(fd, times))
+                       return 0;
+#endif
+#ifdef HAVE_UTIMENSAT
+               if (fd < 0 && !utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW))
+                       return 0;
+#endif
+               if (errno != ENOSYS)
+                       return WIMLIB_ERR_SET_TIMESTAMPS;
+       }
+       {
+               struct timeval times[2];
+
+               times[0] = wim_timestamp_to_timeval(atime);
+               times[1] = wim_timestamp_to_timeval(mtime);
+
+               if (fd >= 0 && !futimes(fd, times))
+                       return 0;
+               if (fd < 0 && !lutimes(path, times))
+                       return 0;
+               return WIMLIB_ERR_SET_TIMESTAMPS;
+       }
+}
+
+static int
+unix_set_owner_and_group(int fd, const char *path, uid_t uid, gid_t gid)
+{
+       if (fd >= 0 && !fchown(fd, uid, gid))
+               return 0;
+       if (fd < 0 && !lchown(path, uid, gid))
+               return 0;
+       return WIMLIB_ERR_SET_SECURITY;
+}
+
+static int
+unix_set_mode(int fd, const char *path, mode_t mode)
+{
+       if (fd >= 0 && !fchmod(fd, mode))
+               return 0;
+       if (fd < 0 && !chmod(path, mode))
+               return 0;
+       return WIMLIB_ERR_SET_SECURITY;
+}
+
+#ifdef HAVE_XATTR_SUPPORT
+/* Apply extended attributes to a file */
+static int
+apply_linux_xattrs(int fd, const struct wim_inode *inode,
+                  const char *path, struct unix_apply_ctx *ctx,
+                  const void *entries, size_t entries_size)
+{
+       const void * const entries_end = entries + entries_size;
+       char name[XATTR_NAME_MAX + 1];
+
+       for (const struct wimlib_xattr_entry *entry = entries;
+            (void *)entry < entries_end; entry = xattr_entry_next(entry))
+       {
+               u16 name_len;
+               const void *value;
+               u32 value_len;
+               int res;
+
+               if (!valid_xattr_entry(entry, entries_end - (void *)entry)) {
+                       if (!path) {
+                               path = unix_build_inode_extraction_path(inode,
+                                                                       ctx);
+                       }
+                       ERROR("\"%s\": extended attribute is corrupt", path);
+                       return WIMLIB_ERR_INVALID_XATTR;
+               }
+               name_len = le16_to_cpu(entry->name_len);
+               memcpy(name, entry->name, name_len);
+               name[name_len] = '\0';
+
+               value = entry->name + name_len;
+               value_len = le32_to_cpu(entry->value_len);
+
+               if (fd >= 0)
+                       res = fsetxattr(fd, name, value, value_len, 0);
+               else
+                       res = lsetxattr(path, name, value, value_len, 0);
+
+               if (unlikely(res != 0)) {
+                       if (!path) {
+                               path = unix_build_inode_extraction_path(inode,
+                                                                       ctx);
+                       }
+                       if (is_security_xattr(name) &&
+                           (ctx->common.extract_flags &
+                            WIMLIB_EXTRACT_FLAG_STRICT_ACLS))
+                       {
+                               ERROR_WITH_ERRNO("\"%s\": unable to set extended attribute \"%s\"",
+                                                path, name);
+                               return WIMLIB_ERR_SET_XATTR;
+                       }
+                       WARNING_WITH_ERRNO("\"%s\": unable to set extended attribute \"%s\"",
+                                          path, name);
+               }
+       }
        return 0;
 }
+#endif /* HAVE_XATTR_SUPPORT */
 
+/*
+ * Apply UNIX-specific metadata to a file if available.  This includes standard
+ * UNIX permissions (uid, gid, and mode) and possibly extended attributes too.
+ *
+ * Note that some xattrs which grant privileges, e.g. security.capability, are
+ * cleared by Linux on chown(), even when running as root.  Also, when running
+ * as non-root, if we need to chmod() the file to readonly, we can't do that
+ * before setting xattrs because setxattr() requires write permission.  These
+ * restrictions result in the following ordering which we follow: chown(),
+ * setxattr(), then chmod().
+ *
+ * N.B. the file may be specified by either 'fd' (for regular files) or 'path',
+ * and it may be a symlink.  For symlinks we need lchown() and lsetxattr() but
+ * need to skip the chmod(), since mode bits are not meaningful for symlinks.
+ */
 static int
-unix_create_directory(const tchar *path, struct apply_ctx *ctx, u64 *cookie_ret)
+apply_unix_metadata(int fd, const struct wim_inode *inode,
+                   const char *path, struct unix_apply_ctx *ctx)
 {
-       struct stat stbuf;
+       bool have_dat;
+       struct wimlib_unix_data dat;
+#ifdef HAVE_XATTR_SUPPORT
+       const void *entries;
+       u32 entries_size;
+#endif
+       int ret;
+
+       have_dat = inode_get_unix_data(inode, &dat);
 
-       if (mkdir(path, 0755)) {
-               if (errno != EEXIST)
-                       return WIMLIB_ERR_MKDIR;
-               if (lstat(path, &stbuf))
-                       return WIMLIB_ERR_STAT;
-               errno = EEXIST;
-               if (!S_ISDIR(stbuf.st_mode))
-                       return WIMLIB_ERR_NOTDIR;
+       if (have_dat) {
+               ret = unix_set_owner_and_group(fd, path, dat.uid, dat.gid);
+               if (ret) {
+                       if (!path)
+                               path = unix_build_inode_extraction_path(inode, ctx);
+                       if (ctx->common.extract_flags &
+                           WIMLIB_EXTRACT_FLAG_STRICT_ACLS)
+                       {
+                               ERROR_WITH_ERRNO("\"%s\": unable to set uid=%"PRIu32" and gid=%"PRIu32,
+                                                path, dat.uid, dat.gid);
+                               return ret;
+                       }
+                       WARNING_WITH_ERRNO("\"%s\": unable to set uid=%"PRIu32" and gid=%"PRIu32,
+                                          path, dat.uid, dat.gid);
+               }
+       }
+
+#if HAVE_XATTR_SUPPORT
+       entries = inode_get_linux_xattrs(inode, &entries_size);
+       if (entries) {
+               ret = apply_linux_xattrs(fd, inode, path, ctx,
+                                        entries, entries_size);
+               if (ret)
+                       return ret;
+       }
+#endif
+
+       if (have_dat && !inode_is_symlink(inode)) {
+               ret = unix_set_mode(fd, path, dat.mode);
+               if (ret) {
+                       if (!path)
+                               path = unix_build_inode_extraction_path(inode, ctx);
+                       if (ctx->common.extract_flags &
+                           WIMLIB_EXTRACT_FLAG_STRICT_ACLS)
+                       {
+                               ERROR_WITH_ERRNO("\"%s\": unable to set mode=0%"PRIo32,
+                                                path, dat.mode);
+                               return ret;
+                       }
+                       WARNING_WITH_ERRNO("\"%s\": unable to set mode=0%"PRIo32,
+                                          path, dat.mode);
+               }
        }
+
        return 0;
 }
 
+/*
+ * Set metadata on an extracted file.
+ *
+ * @fd is an open file descriptor to the extracted file, or -1.  @path is the
+ * path to the extracted file, or NULL.  If valid, this function uses @fd.
+ * Otherwise, if valid, it uses @path.  Otherwise, it calculates the path to one
+ * alias of the extracted file and uses it.
+ */
 static int
-unix_makelink(const tchar *oldpath, const tchar *newpath,
-             int (*makelink)(const tchar *oldpath, const tchar *newpath))
+unix_set_metadata(int fd, const struct wim_inode *inode,
+                 const char *path, struct unix_apply_ctx *ctx)
 {
-       if ((*makelink)(oldpath, newpath)) {
-               if (errno == EEXIST) {
-                       if (unlink(newpath))
-                               return WIMLIB_ERR_LINK;
-                       if ((*makelink)(oldpath, newpath))
-                               return WIMLIB_ERR_LINK;
-                       return 0;
+       int ret;
+
+       if (fd < 0 && !path)
+               path = unix_build_inode_extraction_path(inode, ctx);
+
+       if (ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) {
+               ret = apply_unix_metadata(fd, inode, path, ctx);
+               if (ret)
+                       return ret;
+       }
+
+       ret = unix_set_timestamps(fd, path, inode->i_last_access_time,
+                                 inode->i_last_write_time);
+       if (ret) {
+               if (!path)
+                       path = unix_build_inode_extraction_path(inode, ctx);
+               if (ctx->common.extract_flags &
+                   WIMLIB_EXTRACT_FLAG_STRICT_TIMESTAMPS)
+               {
+                       ERROR_WITH_ERRNO("\"%s\": unable to set timestamps", path);
+                       return ret;
                }
-               return WIMLIB_ERR_LINK;
+               WARNING_WITH_ERRNO("\"%s\": unable to set timestamps", path);
+       }
+
+       return 0;
+}
+
+/* Extract all needed aliases of the @inode, where one alias, corresponding to
+ * @first_dentry, has already been extracted to @first_path.  */
+static int
+unix_create_hardlinks(const struct wim_inode *inode,
+                     const struct wim_dentry *first_dentry,
+                     const char *first_path, struct unix_apply_ctx *ctx)
+{
+       const struct wim_dentry *dentry;
+       const char *newpath;
+
+       inode_for_each_extraction_alias(dentry, inode) {
+               if (dentry == first_dentry)
+                       continue;
+
+               newpath = unix_build_extraction_path(dentry, ctx);
+       retry_link:
+               if (link(first_path, newpath)) {
+                       if (errno == EEXIST && !unlink(newpath))
+                               goto retry_link;
+                       ERROR_WITH_ERRNO("Can't create hard link "
+                                        "\"%s\" => \"%s\"", newpath, first_path);
+                       return WIMLIB_ERR_LINK;
+               }
+               unix_reuse_pathbuf(ctx);
        }
        return 0;
 }
+
+/* If @dentry represents a directory, create it.  */
+static int
+unix_create_if_directory(const struct wim_dentry *dentry,
+                        struct unix_apply_ctx *ctx)
+{
+       const char *path;
+       struct stat stbuf;
+
+       if (!should_extract_as_directory(dentry->d_inode))
+               return 0;
+
+       path = unix_build_extraction_path(dentry, ctx);
+       if (mkdir(path, 0755) &&
+           /* It's okay if the path already exists, as long as it's a
+            * directory.  */
+           !(errno == EEXIST && !lstat(path, &stbuf) && S_ISDIR(stbuf.st_mode)))
+       {
+               ERROR_WITH_ERRNO("Can't create directory \"%s\"", path);
+               return WIMLIB_ERR_MKDIR;
+       }
+
+       return report_file_created(&ctx->common);
+}
+
+/* If @dentry represents an empty regular file or a special file, create it, set
+ * its metadata, and create any needed hard links.  */
 static int
-unix_create_hardlink(const tchar *oldpath, const tchar *newpath,
-                    struct apply_ctx *ctx)
+unix_extract_if_empty_file(const struct wim_dentry *dentry,
+                          struct unix_apply_ctx *ctx)
 {
-       return unix_makelink(oldpath, newpath, link);
+       const struct wim_inode *inode;
+       struct wimlib_unix_data unix_data;
+       const char *path;
+       int ret;
+
+       inode = dentry->d_inode;
+
+       /* Extract all aliases only when the "first" comes up.  */
+       if (dentry != inode_first_extraction_dentry(inode))
+               return 0;
+
+       /* Is this a directory, a symbolic link, or any type of nonempty file?
+        */
+       if (should_extract_as_directory(inode) || inode_is_symlink(inode) ||
+           inode_get_blob_for_unnamed_data_stream_resolved(inode))
+               return 0;
+
+       /* Recognize special files in UNIX_DATA mode  */
+       if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_UNIX_DATA) &&
+           inode_get_unix_data(inode, &unix_data) &&
+           !S_ISREG(unix_data.mode))
+       {
+               path = unix_build_extraction_path(dentry, ctx);
+       retry_mknod:
+               if (mknod(path, unix_data.mode, unix_data.rdev)) {
+                       if (errno == EPERM) {
+                               WARNING_WITH_ERRNO("Can't create special "
+                                                  "file \"%s\"", path);
+                               ctx->num_special_files_ignored++;
+                               return 0;
+                       }
+                       if (errno == EEXIST && !unlink(path))
+                               goto retry_mknod;
+                       ERROR_WITH_ERRNO("Can't create special file \"%s\"",
+                                        path);
+                       return WIMLIB_ERR_MKNOD;
+               }
+               /* On special files, we can set timestamps immediately because
+                * we don't need to write any data to them.  */
+               ret = unix_set_metadata(-1, inode, path, ctx);
+       } else {
+               int fd;
+
+               path = unix_build_extraction_path(dentry, ctx);
+       retry_create:
+               fd = open(path, O_EXCL | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
+               if (fd < 0) {
+                       if (errno == EEXIST && !unlink(path))
+                               goto retry_create;
+                       ERROR_WITH_ERRNO("Can't create regular file \"%s\"", path);
+                       return WIMLIB_ERR_OPEN;
+               }
+               /* On empty files, we can set timestamps immediately because we
+                * don't need to write any data to them.  */
+               ret = unix_set_metadata(fd, inode, path, ctx);
+               if (close(fd) && !ret) {
+                       ERROR_WITH_ERRNO("Error closing \"%s\"", path);
+                       ret = WIMLIB_ERR_WRITE;
+               }
+       }
+       if (ret)
+               return ret;
+
+       ret = unix_create_hardlinks(inode, dentry, path, ctx);
+       if (ret)
+               return ret;
+
+       return report_file_created(&ctx->common);
 }
 
 static int
-unix_create_symlink(const tchar *oldpath, const tchar *newpath,
-                   struct apply_ctx *ctx)
+unix_create_dirs_and_empty_files(const struct list_head *dentry_list,
+                                struct unix_apply_ctx *ctx)
 {
-       return unix_makelink(oldpath, newpath, symlink);
+       const struct wim_dentry *dentry;
+       int ret;
+
+       list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
+               ret = unix_create_if_directory(dentry, ctx);
+               if (ret)
+                       return ret;
+       }
+       list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
+               ret = unix_extract_if_empty_file(dentry, ctx);
+               if (ret)
+                       return ret;
+       }
+       return 0;
+}
+
+static void
+unix_count_dentries(const struct list_head *dentry_list,
+                   u64 *dir_count_ret, u64 *empty_file_count_ret)
+{
+       const struct wim_dentry *dentry;
+       u64 dir_count = 0;
+       u64 empty_file_count = 0;
+
+       list_for_each_entry(dentry, dentry_list, d_extraction_list_node) {
+
+               const struct wim_inode *inode = dentry->d_inode;
+
+               if (should_extract_as_directory(inode))
+                       dir_count++;
+               else if ((dentry == inode_first_extraction_dentry(inode)) &&
+                        !inode_is_symlink(inode) &&
+                        !inode_get_blob_for_unnamed_data_stream_resolved(inode))
+                       empty_file_count++;
+       }
+
+       *dir_count_ret = dir_count;
+       *empty_file_count_ret = empty_file_count;
 }
 
 static int
-unix_extract_unnamed_stream(file_spec_t file,
-                           struct wim_lookup_table_entry *lte,
-                           struct apply_ctx *ctx)
+unix_create_symlink(const struct wim_inode *inode, const char *path,
+                   size_t rpdatalen, struct unix_apply_ctx *ctx)
 {
-       const char *path = file.path;
-       struct filedes fd;
-       int raw_fd;
+       char target[REPARSE_POINT_MAX_SIZE];
+       struct blob_descriptor blob_override;
        int ret;
 
-       raw_fd = open(path, O_WRONLY | O_TRUNC);
-       if (raw_fd < 0)
+       blob_set_is_located_in_attached_buffer(&blob_override,
+                                              ctx->reparse_data, rpdatalen);
+
+       ret = wim_inode_readlink(inode, target, sizeof(target) - 1,
+                                &blob_override,
+                                ctx->target_abspath,
+                                ctx->target_abspath_nchars);
+       if (unlikely(ret < 0)) {
+               errno = -ret;
+               return WIMLIB_ERR_READLINK;
+       }
+       target[ret] = '\0';
+
+retry_symlink:
+       if (symlink(target, path)) {
+               if (errno == EEXIST && !unlink(path))
+                       goto retry_symlink;
+               return WIMLIB_ERR_LINK;
+       }
+       return 0;
+}
+
+static void
+unix_cleanup_open_fds(struct unix_apply_ctx *ctx, unsigned offset)
+{
+       for (unsigned i = offset; i < ctx->num_open_fds; i++)
+               filedes_close(&ctx->open_fds[i]);
+       ctx->num_open_fds = 0;
+       ctx->any_sparse_files = false;
+}
+
+static int
+unix_begin_extract_blob_instance(const struct blob_descriptor *blob,
+                                const struct wim_inode *inode,
+                                const struct wim_inode_stream *strm,
+                                struct unix_apply_ctx *ctx)
+{
+       const struct wim_dentry *first_dentry;
+       const char *first_path;
+       int fd;
+
+       if (unlikely(strm->stream_type == STREAM_TYPE_REPARSE_POINT)) {
+               /* On UNIX, symbolic links must be created with symlink(), which
+                * requires that the full link target be available.  */
+               if (blob->size > REPARSE_DATA_MAX_SIZE) {
+                       ERROR_WITH_ERRNO("Reparse data of \"%s\" has size "
+                                        "%"PRIu64" bytes (exceeds %u bytes)",
+                                        inode_any_full_path(inode),
+                                        blob->size, REPARSE_DATA_MAX_SIZE);
+                       return WIMLIB_ERR_INVALID_REPARSE_DATA;
+               }
+               ctx->reparse_ptr = ctx->reparse_data;
+               return 0;
+       }
+
+       wimlib_assert(stream_is_unnamed_data_stream(strm));
+
+       /* Unnamed data stream of "regular" file  */
+
+       /* This should be ensured by extract_blob_list()  */
+       wimlib_assert(ctx->num_open_fds < MAX_OPEN_FILES);
+
+       first_dentry = inode_first_extraction_dentry(inode);
+       first_path = unix_build_extraction_path(first_dentry, ctx);
+retry_create:
+       fd = open(first_path, O_EXCL | O_CREAT | O_WRONLY | O_NOFOLLOW, 0644);
+       if (fd < 0) {
+               if (errno == EEXIST && !unlink(first_path))
+                       goto retry_create;
+               ERROR_WITH_ERRNO("Can't create regular file \"%s\"", first_path);
                return WIMLIB_ERR_OPEN;
-       filedes_init(&fd, raw_fd);
-       ret = extract_wim_resource_to_fd(lte, &fd, wim_resource_size(lte));
-       if (filedes_close(&fd) && !ret)
-               ret = WIMLIB_ERR_WRITE;
+       }
+       if (inode->i_attributes & FILE_ATTRIBUTE_SPARSE_FILE) {
+               ctx->is_sparse_file[ctx->num_open_fds] = true;
+               ctx->any_sparse_files = true;
+       } else {
+               ctx->is_sparse_file[ctx->num_open_fds] = false;
+#ifdef HAVE_POSIX_FALLOCATE
+               posix_fallocate(fd, 0, blob->size);
+#endif
+       }
+       filedes_init(&ctx->open_fds[ctx->num_open_fds++], fd);
+       return unix_create_hardlinks(inode, first_dentry, first_path, ctx);
+}
+
+/* Called when starting to read a blob for extraction  */
+static int
+unix_begin_extract_blob(struct blob_descriptor *blob, void *_ctx)
+{
+       struct unix_apply_ctx *ctx = _ctx;
+       const struct blob_extraction_target *targets = blob_extraction_targets(blob);
+
+       for (u32 i = 0; i < blob->out_refcnt; i++) {
+               int ret = unix_begin_extract_blob_instance(blob,
+                                                          targets[i].inode,
+                                                          targets[i].stream,
+                                                          ctx);
+               if (ret) {
+                       ctx->reparse_ptr = NULL;
+                       unix_cleanup_open_fds(ctx, 0);
+                       return ret;
+               }
+       }
+       return 0;
+}
+
+/* Called when the next chunk of a blob has been read for extraction  */
+static int
+unix_extract_chunk(const struct blob_descriptor *blob, u64 offset,
+                  const void *chunk, size_t size, void *_ctx)
+{
+       struct unix_apply_ctx *ctx = _ctx;
+       const void * const end = chunk + size;
+       const void *p;
+       bool zeroes;
+       size_t len;
+       unsigned i;
+       int ret;
+
+       /*
+        * For sparse files, only write nonzero regions.  This lets the
+        * filesystem use holes to represent zero regions.
+        */
+       for (p = chunk; p != end; p += len, offset += len) {
+               zeroes = maybe_detect_sparse_region(p, end - p, &len,
+                                                   ctx->any_sparse_files);
+               for (i = 0; i < ctx->num_open_fds; i++) {
+                       if (!zeroes || !ctx->is_sparse_file[i]) {
+                               ret = full_pwrite(&ctx->open_fds[i],
+                                                 p, len, offset);
+                               if (ret)
+                                       goto err;
+                       }
+               }
+       }
+
+       if (ctx->reparse_ptr)
+               ctx->reparse_ptr = mempcpy(ctx->reparse_ptr, chunk, size);
+       return 0;
+
+err:
+       ERROR_WITH_ERRNO("Error writing data to filesystem");
        return ret;
 }
 
+/* Called when a blob has been fully read for extraction  */
 static int
-unix_set_unix_data(const tchar *path, const struct wimlib_unix_data *data,
-                  struct apply_ctx *ctx)
+unix_end_extract_blob(struct blob_descriptor *blob, int status, void *_ctx)
 {
-       struct stat stbuf;
+       struct unix_apply_ctx *ctx = _ctx;
+       int ret;
+       unsigned j;
+       const struct blob_extraction_target *targets = blob_extraction_targets(blob);
+
+       ctx->reparse_ptr = NULL;
+
+       if (status) {
+               unix_cleanup_open_fds(ctx, 0);
+               return status;
+       }
+
+       j = 0;
+       ret = 0;
+       for (u32 i = 0; i < blob->out_refcnt; i++) {
+               struct wim_inode *inode = targets[i].inode;
+
+               if (inode_is_symlink(inode)) {
+                       /* We finally have the symlink data, so we can create
+                        * the symlink.  */
+                       const char *path;
+
+                       path = unix_build_inode_extraction_path(inode, ctx);
+                       ret = unix_create_symlink(inode, path, blob->size, ctx);
+                       if (ret) {
+                               ERROR_WITH_ERRNO("Can't create symbolic link "
+                                                "\"%s\"", path);
+                               break;
+                       }
+                       ret = unix_set_metadata(-1, inode, path, ctx);
+                       if (ret)
+                               break;
+               } else {
+                       struct filedes *fd = &ctx->open_fds[j];
+
+                       /* If the file is sparse, extend it to its final size. */
+                       if (ctx->is_sparse_file[j] && ftruncate(fd->fd, blob->size)) {
+                               ERROR_WITH_ERRNO("Error extending \"%s\" to final size",
+                                                unix_build_inode_extraction_path(inode, ctx));
+                               ret = WIMLIB_ERR_WRITE;
+                               break;
+                       }
+
+                       /* Set metadata on regular file just before closing.  */
+                       ret = unix_set_metadata(fd->fd, inode, NULL, ctx);
+                       if (ret)
+                               break;
+
+                       if (filedes_close(fd)) {
+                               ERROR_WITH_ERRNO("Error closing \"%s\"",
+                                                unix_build_inode_extraction_path(inode, ctx));
+                               ret = WIMLIB_ERR_WRITE;
+                               break;
+                       }
+                       j++;
+               }
+       }
+       unix_cleanup_open_fds(ctx, j);
+       return ret;
+}
+
+static int
+unix_set_dir_metadata(struct list_head *dentry_list, struct unix_apply_ctx *ctx)
+{
+       const struct wim_dentry *dentry;
+       int ret;
 
-       if (lstat(path, &stbuf))
-               return WIMLIB_ERR_SET_SECURITY;
-       if (!S_ISLNK(stbuf.st_mode))
-               if (chmod(path, data->mode))
-                       return WIMLIB_ERR_SET_SECURITY;
-       if (lchown(path, data->uid, data->gid))
-               return WIMLIB_ERR_SET_SECURITY;
+       list_for_each_entry_reverse(dentry, dentry_list, d_extraction_list_node) {
+               if (should_extract_as_directory(dentry->d_inode)) {
+                       ret = unix_set_metadata(-1, dentry->d_inode, NULL, ctx);
+                       if (ret)
+                               return ret;
+                       ret = report_file_metadata_applied(&ctx->common);
+                       if (ret)
+                               return ret;
+               }
+       }
        return 0;
 }
 
 static int
-unix_set_timestamps(const tchar *path, u64 creation_time, u64 last_write_time,
-                   u64 last_access_time, struct apply_ctx *ctx)
+unix_extract(struct list_head *dentry_list, struct apply_ctx *_ctx)
 {
        int ret;
+       struct unix_apply_ctx *ctx = (struct unix_apply_ctx *)_ctx;
+       size_t path_max;
+       u64 dir_count;
+       u64 empty_file_count;
 
-#ifdef HAVE_UTIMENSAT
-       /* Convert the WIM timestamps, which are accurate to 100 nanoseconds,
-        * into `struct timespec's for passing to utimensat(), which is accurate
-        * to 1 nanosecond. */
-
-       struct timespec ts[2];
-       ts[0] = wim_timestamp_to_timespec(last_access_time);
-       ts[1] = wim_timestamp_to_timespec(last_write_time);
-       ret = utimensat(AT_FDCWD, path, ts, AT_SYMLINK_NOFOLLOW);
+       /* Compute the maximum path length that will be needed, then allocate
+        * some path buffers.  */
+       path_max = unix_compute_path_max(dentry_list, ctx);
+
+       for (unsigned i = 0; i < NUM_PATHBUFS; i++) {
+               ctx->pathbufs[i] = MALLOC(path_max);
+               if (!ctx->pathbufs[i]) {
+                       ret = WIMLIB_ERR_NOMEM;
+                       goto out;
+               }
+               /* Pre-fill the target in each path buffer.  We'll just append
+                * the rest of the paths after this.  */
+               memcpy(ctx->pathbufs[i],
+                      ctx->common.target, ctx->common.target_nchars);
+       }
+
+       /* Extract directories and empty regular files.  Directories are needed
+        * because we can't extract any other files until their directories
+        * exist.  Empty files are needed because they don't have
+        * representatives in the blob list.  */
+
+       unix_count_dentries(dentry_list, &dir_count, &empty_file_count);
+
+       ret = start_file_structure_phase(&ctx->common, dir_count + empty_file_count);
        if (ret)
-               ret = errno;
-#else
-       ret = ENOSYS;
-#endif
+               goto out;
 
-       if (ret == ENOSYS) {
-               /* utimensat() not implemented or not available */
-       #ifdef HAVE_LUTIMES
-               /* Convert the WIM timestamps, which are accurate to 100
-                * nanoseconds, into `struct timeval's for passing to lutimes(),
-                * which is accurate to 1 microsecond. */
-               struct timeval tv[2];
-               tv[0] = wim_timestamp_to_timeval(last_access_time);
-               tv[1] = wim_timestamp_to_timeval(last_write_time);
-               ret = lutimes(path, tv);
-               if (ret)
-                       ret = errno;
-       #endif
-       }
-
-       if (ret == ENOSYS) {
-               /* utimensat() and lutimes() both not implemented or not
-                * available */
-       #ifdef HAVE_UTIME
-               /* Convert the WIM timestamps, which are accurate to 100
-                * nanoseconds, into a `struct utimbuf's for passing to
-                * utime(), which is accurate to 1 second. */
-               struct utimbuf buf;
-               buf.actime = wim_timestamp_to_unix(last_access_time);
-               buf.modtime = wim_timestamp_to_unix(last_write_time);
-               ret = utime(path, &buf);
-       #endif
+       ret = unix_create_dirs_and_empty_files(dentry_list, ctx);
+       if (ret)
+               goto out;
+
+       ret = end_file_structure_phase(&ctx->common);
+       if (ret)
+               goto out;
+
+       /* Get full path to target if needed for absolute symlink fixups.  */
+       if ((ctx->common.extract_flags & WIMLIB_EXTRACT_FLAG_RPFIX) &&
+           ctx->common.required_features.symlink_reparse_points)
+       {
+               ctx->target_abspath = realpath(ctx->common.target, NULL);
+               if (!ctx->target_abspath) {
+                       ret = WIMLIB_ERR_NOMEM;
+                       goto out;
+               }
+               ctx->target_abspath_nchars = strlen(ctx->target_abspath);
        }
+
+       /* Extract nonempty regular files and symbolic links.  */
+
+       struct read_blob_callbacks cbs = {
+               .begin_blob     = unix_begin_extract_blob,
+               .continue_blob  = unix_extract_chunk,
+               .end_blob       = unix_end_extract_blob,
+               .ctx            = ctx,
+       };
+       ret = extract_blob_list(&ctx->common, &cbs);
        if (ret)
-               return WIMLIB_ERR_SET_TIMESTAMPS;
-       return 0;
+               goto out;
+
+
+       /* Set directory metadata.  We do this last so that we get the right
+        * directory timestamps.  */
+       ret = start_file_metadata_phase(&ctx->common, dir_count);
+       if (ret)
+               goto out;
+
+       ret = unix_set_dir_metadata(dentry_list, ctx);
+       if (ret)
+               goto out;
+
+       ret = end_file_metadata_phase(&ctx->common);
+       if (ret)
+               goto out;
+
+       if (ctx->num_special_files_ignored) {
+               WARNING("%lu special files were not extracted due to EPERM!",
+                       ctx->num_special_files_ignored);
+       }
+out:
+       for (unsigned i = 0; i < NUM_PATHBUFS; i++)
+               FREE(ctx->pathbufs[i]);
+       FREE(ctx->target_abspath);
+       return ret;
 }
 
 const struct apply_operations unix_apply_ops = {
-       .name = "UNIX",
-
-       .start_extract          = unix_start_extract,
-       .create_file            = unix_create_file,
-       .create_directory       = unix_create_directory,
-       .create_hardlink        = unix_create_hardlink,
-       .create_symlink         = unix_create_symlink,
-       .extract_unnamed_stream = unix_extract_unnamed_stream,
-       .set_unix_data          = unix_set_unix_data,
-       .set_timestamps         = unix_set_timestamps,
-
-       .path_separator = '/',
-       .path_max = PATH_MAX,
-
-       .requires_target_in_paths = 1,
-       .supports_case_sensitive_filenames = 1,
+       .name                   = "UNIX",
+       .get_supported_features = unix_get_supported_features,
+       .extract                = unix_extract,
+       .context_size           = sizeof(struct unix_apply_ctx),
 };
-
-#endif /* !__WIN32__ */