Copyright (c) 2007, Sun Microsystems, Inc. All Rights Reserved.
The contents of this file are subject to the terms of the Common Development and Distribution License (the "License"). You may not use this file except in compliance with the License.
You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. See the License for the specific language governing permissions and limitations under the License.
When distributing Covered Code, include this CDDL HEADER in each file and include the License file at usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your own identifying information: Portions Copyright [yyyy] [name of copyright owner]
Attributes are logically supported as files within the file system. The file system is therefore augmented with an orthogonal name space of file attributes. Any file (including attribute files) can have an arbitrarily deep attribute tree associated with it. Attribute values are accessed by file descriptors obtained through a special attribute interface. This logical view of "attributes as files" allows the leveraging of existing file system interface functionality to support the construction, deletion, and manipulation of attributes.
The special files "." and ".\|." retain their accustomed semantics within the attribute hierarchy. The "." attribute file refers to the current directory and the ".\|." attribute file refers to the parent directory. The unnamed directory at the head of each attribute tree is considered the "child" of the file it is associated with and the ".\|." file refers to the associated file. For any non-directory file with attributes, the ".\|." entry in the unnamed directory refers to a file that is not a directory.
Conceptually, the attribute model is fully general. Extended attributes can be any type of file (doors, links, directories, and so forth) and can even have their own attributes (fully recursive). As a result, the attributes associated with a file could be an arbitrarily deep directory hierarchy where each attribute could have an equally complex attribute tree associated with it. Not all implementations are able to, or want to, support the full model. Implementation are therefore permitted to reject operations that are not supported. For example, the implementation for the UFS file system allows only regular files as attributes (for example, no sub-directories) and rejects attempts to place attributes on attributes.
The following list details the operations that are rejected in the current implementation: link
Any attempt to create links between attribute and non-attribute space is rejected to prevent security-related or otherwise sensitive attributes from being exposed, and therefore manipulable, as regular files.
Any attempt to rename between attribute and non-attribute space is rejected to prevent an already linked file from being renamed and thereby circumventing the link restriction above.
Any attempt to create a "non-regular" file in attribute space is rejected to reduce the functionality, and therefore exposure and risk, of the initial implementation.
The entire available name space has been allocated to "general use" to bring the implementation in line with the NFSv4 draft standard [NFSv4]. That standard defines "named attributes" (equivalent to Solaris Extended Attributes) with no naming restrictions. All Sun applications making use of opaque extended attributes will use the prefix "SUNW".
The command interface for extended attributes is the set of applications provided by Solaris for the manipulation of attributes from the command line. This interface consists of a set of existing utilities that have been extended to be "attribute-aware", plus the runat utility designed to "expose" the extended attribute space so that extended attributes can be manipulated as regular files.
The -@ option enable utilities to manipulate extended attributes. As a rule, this option enables the utility to enter into attribute space when the utility is performing a recursive traversal of file system space. This is a fully recursive concept. If the underlying file system supports recursive attributes and directory structures, the -@ option opens these spaces to the file tree-walking algorithms.
The following utilities accommodate extended attributes (see the individual manual pages for details): cp
By default, cp ignores attributes and copies only file data. This is intended to maintain the semantics implied by cp currently, where attributes (such as owner and mode) are not copied unless the -p option is specified. With the -@ (or -p) option, cp attempts to copy all attributes along with the file data.
The -@ option informs cpio to archive attributes, but by default cpio ignores extended attributes. See Extended Archive Formats below for a description of the new archive records.
File sizes computed include the space allocated for any extended attributes present.
By default, find ignores attributes. The -xattr expression provides support for searches involving attribute space. It returns true if extended attributes are present on the current file.
The fsck utility manages extended attribute data on the disk. A file system with extended attributes can be mounted on versions of Solaris that are not attribute-aware (versions prior to Solaris 9), but the attributes will not be accessible and fsck will strip them from the files and place them in lost+found. Once the attributes have been stripped the file system is completely stable on Solaris versions that are not attribute-aware, but would now be considered corrupted on attribute-aware versions of Solaris. The attribute-aware fsck utility should be run to stabilize the file system before using it in an attribute-aware environment.
This fsdb utility is able to find the inode for the "hidden" extended attribute directory.
The ls -@ command displays an "@" following the mode information when extended attributes are present. More precisely, the output line for a given file contains an "@" character following the mode characters if the pathconf(2) variable XATTR_EXISTS is set to true. See the pathconf() section below. The -@ option uses the same general output format as the -l option.
When a file is moved, all attributes are carried along with the file rename. When a file is moved across a file system boundary, the copy command invoked is similar to the cp -p variant described above and extended attributes are "moved". If the extended file attributes cannot be replicated, the move operation fails and the source file is not removed.
The -@ option informs pax to archive attributes, but by default pax ignores extended attributes. The pax(1) utility is a generic replacement for both tar(1) and cpio(1) and is able to produce either output format in its archive. See Extended Archive Formats below for a description of the new archive records.
In the default case, tar does not attempt to place attributes in the archive. If the -@ option is specified, however, tar traverses into the attribute space of all files being placed in the archive and attempts to add the attributes to the archive. A new record type has been introduced for extended attribute entries in tar archive files (the same is true for pax and cpio archives) similar to the way ACLs records were defined. See Extended Archive Formats below for a description of the new archive records.
There is a class of utilities (chmod, chown, chgrp) that one might expect to be modified in a manner similar to those listed above. For example, one might expect that performing chmod on a file would not only affect the file itself but would also affect at least the extended attribute directory if not any existing extended attribute files. This is not the case. The model chosen for extended attributes implies that the attribute directory and the attributes themselves are all file objects in their own right, and can therefore have independent file status attributes associated with them (a given implementation cannot support this, for example, for intrinsic attributes). The relationship is left undefined and a fine-grained control mechanism (runat(1)) is provided to allow manipulation of extended attribute status attributes as necessary.
The runat utility has the following syntax:
runat filename [command]
The runat utility executes the supplied command in the context of the "attribute space" associated with the indicated file. If no command argument is supplied, a shell is invoked. See runat(1) for details.
The primary interface required to access extended attributes at the programmatic level is the openat(2) function. Once a file descriptor has been obtained for an attribute file by an openat() call, all normal file system semantics apply. There is no attempt to place special semantics on read(2), write(2), ftruncate(3C), or other functions when applied to attribute file descriptors relative to "normal" file descriptors.
The set of existing attributes can be browsed by calling openat() with "." as the file name and the O_XATTR flag set, resulting in a file descriptor for the attribute directory. The list of attributes is obtained by calls to getdents(2) on the returned file descriptor. If the target file did not previously have any attributes associated with it, an empty top-level attribute directory is created for the file and subsequent getdents() calls will return only "." and ".\|.". While the owner of the parent file owns the extended attribute directory, it is not charged against its quota if the directory is empty. Attribute files themselves, however, are charged against the user quota as any other regular file.
Additional system calls have been provided as convenience functions. These include the fchownat(2), fstatat(2), futimesat(2), renameat(2), unlinkat(2). These new functions, along with openat(), provide a mechanism to access files relative to an arbitrary point in the file system, rather than only the current working directory. This mechanism is particularly useful in situations when a file descriptor is available with no path. The openat() function, in particular, can be used in many contexts where chdir() or fchdir() is currently required. See chdir(2).
int openat (int fd, const char *path, int oflag [, mode_t mode])
The openat(2) function behaves exactly as open(2) except when given a relative path. Where open() resolves a relative path from the current working directory, openat() resolves the path based on the vnode indicated by the supplied file descriptor. When oflag is O_XATTR, openat() interprets the path argument as an extended attribute reference. The following code fragment uses openat() to examine the attributes of some already opened file:
dfd = openat(fd, ".", O_RDONLY|O_XATTR); (void)getdents(dfd, buf, nbytes);
If openat() is passed the special value AT_FDCWD as its first (fd) argument, its behavior is identical to open() and the relative path arguments are interpreted relative to the current working directory. If the O_XATTR flag is provided to openat() or to open(), the supplied path is interpreted as a reference to an extended attribute on the current working directory.
int unlinkat (int dirfd, const char *pathflag, int flagflag)
The unlinkat(2) function deletes an entry from a directory. The path argument indicates the name of the entry to remove. If path an absolute path, the dirfd argument is ignored. If it is a relative path, it is interpreted relative to the directory indicated by the dirfd argument. If dirfd does not refer to a valid directory, the function returns ENOTDIR. If the special value AT_FDCWD is specified for dirfd, a relative path argument is resolved relative to the current working directory. If the flag argument is 0, all other semantics of this function are equivalent to unlink(2). If flag is set to AT_REMOVEDIR, all other semantics of this function are equivalent to rmdir(2).
int renameat (int fromfd, const char *old, int tofd, const char *new)
The renameat(2) function renames an entry in a directory, possibly moving the entry into a different directory. The old argument indicates the name of the entry to rename. If this argument is a relative path, it is interpreted relative to the directory indicated by the fd argument. If it is an absolute path, the fromfd argument is ignored. The new argument indicates the new name for the entry. If this argument is a relative path, it is interpreted relative to the directory indicated by the tofd argument. If it is an absolute path, the tofd argument is ignored.
In the relative path cases, if the directory file descriptor arguments do not refer to a valid directory, the function returns ENOTDIR. All other semantics of this function are equivalent to rename(2).
If a special value AT_FDCWD is specified for either the fromfd or tofd arguments, their associated path arguments (old and new) are interpreted relative to the current working directory if they are not specified as absolute paths. Any attempt to use renameat() to move a file that is not an extended attribute into an extended attribute directory (so that it becomes an extended attribute) will fail. The same is true for an attempt to move a file that is an extended attribute into a directory that is not an extended attribute directory.
int fstatat (int fd, const char *path, struct stat* buf, int flag)
The fstatat(2) function obtains information about a file. If the path argument is relative, it is resolved relative to the fd argument file descriptor, otherwise the fd argument is ignored. If the fd argument is a special value AT_FDCWD the path is resolved relative to the current working directory. If the path argument is a null pointer, the function returns information about the file referenced by the fd argument. In all other relative path cases, if the fd argument does not refer to a valid directory, the function returns ENOTDIR. If AT_SYMLINK_NOFOLLOW is set in the flag argument, the function will not automatically traverse a symbolic link at the position of the path. If _AT_TRIGGER is set in the flag argument and the vnode is a trigger mount point, the mount is performed and the function returns the attributes of the root of the mounted filesystem. The fstatat() function is a multipurpose function that can be used in place of stat(), lstat(), or fstat(). See stat(2)
The function call stat(path, buf) is identical to fstatat(AT_FDCWD, path, buf, 0).
The function call lstat(path, buf) is identical to fstatat(AT_FDCWD, path, buf, AT_SYMLINK_NOFOLLOW)
The function call fstat(fildes, buf) is identical to fstatat(fildes, NULL, buf, 0).
int fchownat (int fd, const char *path, uid_t owner, gid_t group, \e int flag)
The fchownat(2) function sets the owner ID and group ID for a file. If the path argument is relative, it is resolved relative to the fd argument file descriptor, otherwise the fd argument is ignored. If the fd argument is a special value AT_FDCWD the path is resolved relative to the current working directory. If the path argument is a null pointer, the function sets the owner and group ID of the file referenced by the fd argument. In all other relative path cases, if the fd argument does not refer to a valid directory, the function returns ENOTDIR. If the flag argument is set to AT_SYMLINK_NOFOLLOW, the function will not automatically traverse a symbolic link at the position of the path. The fchownat() function is a multi-purpose function that can be used in place of chown(), lchown(), or fchown(). See chown(2).
The function call chown(path, owner, group) is equivalent to fchownat(AT_FDCWD, path, owner, group, 0).
The function call lchown(path, owner, group) is equivalent to fchownat(AT_FDCWD, path, owner, group, AT_SYMLINK_NOFOLLOW).
int futimesat (int fd, const char *path, const struct timeval \e times[2])
The futimesat(2) function sets the access and modification times for a file. If the path argument is relative, it is resolved relative to the fd argument file descriptor; otherwise the fd argument is ignored. If the fd argument is the special value AT_FDCWD, the path is resolved relative to the current working directory. If the path argument is a null pointer, the function sets the access and modification times of the file referenced by the fd argument. In all other relative path cases, if the fd argument does not refer to a valid directory, the function returns ENOTDIR. The futimesat() function can be used in place of utimes(2).
The function call utimes(path, times) is equivalent to futimesat(AT_FDCWD, path, times).
long int pathconf(const char *path, int name)
Two variables have been added to pathconf(2) to provide enhanced support for extended attribute manipulation. The XATTR_ENABLED variable allows an application to determine if attribute support is currently enabled for the file in question. The XATTR_EXISTS variable allows an application to determine whether there are any extended attributes associated with the supplied path.
int attropen (const char *path, const char *attrpath, int oflag \e [, mode_t mode])
The attropen(3C) function returns a file descriptor for the named attribute, attrpath, of the file indicated by path. The oflag and mode arguments are identical to the open(2) arguments and are applied to the open operation on the attribute file (for example, using the O_CREAT flag creates a new attribute). Once opened, all normal file system operations can be used on the attribute file descriptor. The attropen() function is a convenience function and is equivalent to the following sequence of operations:
fd = open (path, O_RDONLY); attrfd = openat(fd, attrpath, oflag|O_XATTR, mode); close(fd);
The set of existing attributes can be browsed by calling attropen() with "." as the attribute name. The list of attributes is obtained by calling getdents(2) (or fdopendir(3C) followed by readdir(3C), see below) on the returned file descriptor.
DIR * fdopendir (const int fd)
The fdopendir(3C) function promotes a file descriptor for a directory to a directory pointer suitable for use with the readdir(3C) function. The originating file descriptor should not be used again following the call to fdopendir(). The directory pointer should be closed with a call to closedir(3C). If the provided file descriptor does not reference a directory, the function returns ENOTDIR. This function is useful in circumstances where the only available handle on a directory is a file descriptor. See attropen(3C) and openat(2).
The following examples demonstrate how the API might be used to perform basic operations on extended attributes:
Example 1 List extended attributes on a file.
attrdirfd = attropen("test", ".", O_RDONLY);
dirp = fdopendir(attrdirfd);
while (dp = readdir(dirp)) {
...
Example 2 Open an extended attribute.
attrfd = attropen("test", dp->d_name, O_RDONLY);
or
attrfd = openat(attrdirfd, dp->d_name, O_RDONLY);
Example 3 Read from an extended attribute.
while (read(attrfd, buf, 512) > 0) {
...
Example 4 Create an extended attribute.
newfd = attropen("test", "attr", O_CREAT|O_RDWR);
or
newfd = openat(attrdirfd, "attr", O_CREAT|O_RDWR);
Example 5 Write to an extended attribute.
count = write(newfd, buf, length);
Example 6 Delete an extended attribute.
error = unlinkat(attrdirfd, "attr");
Applications intending to access the interfaces defined here as well as the POSIX and X/Open specification-conforming interfaces should define the macro _ATFILE_SOURCE to be 1 and set whichever feature test macros are appropriate to obtain the desired environment. See standards(7).
As noted above in the description of command utilities modified to provide support for extended attributes, the archive formats for tar(1) and cpio(1) have been extended to provide support for archiving extended attributes. This section describes the specifics of the archive format extensions.
The tar archive is made up of a series of 512 byte blocks. Each archived file is represented by a header block and zero or more data blocks containing the file contents. The header block is structured as shown in the following table.
| Field Name Length (in Octets) Description |
| Name 100 File name string |
| Mode 8 12 file mode bits |
| Uid 8 User ID of file owner |
| Gid 8 Group ID of file owner |
| Size 12 Size of file |
| Mtime 12 File modification time |
| Chksum 8 File contents checksum |
| Typeflag 1 File type flag |
| Linkname 100 Link target name if file linked |
| Magic 6 "ustar" |
| Version 2 "00" |
| Uname 32 User name of file owner |
| Gname 32 Group name of file owner |
| Devmajor 8 Major device ID if special file |
| Devminor 8 Minor device ID if special file |
| Prefix 155 Path prefix string for file |
The extended attribute project extends the above header format by defining a new header type (for the Typeflag field). The type 'E' is defined to be used for all extended attribute files. Attribute files are stored in the tar archive as a sequence of two <header ,data> pairs. The first file contains the data necessary to locate and name the extended attribute in the file system. The second file contains the actual attribute file data. Both files use an 'E' type header. The prefix and name fields in extended attribute headers are ignored, though they should be set to meaningful values for the benefit of archivers that do not process these headers. Solaris archivers set the prefix field to "/dev/null" to prevent archivers that do not understand the type 'E' header from trying to restore extended attribute files in inappropriate places.
The cpio archive format is octet-oriented rather than block-oriented. Each file entry in the archive includes a header that describes the file, followed by the file name, followed by the contents of the file. These data are arranged as described in the following table.
| Field Name Length (in Octets) Description |
| c_magic 6 70707 |
| c_dev 6 First half of unique file ID |
| c_ino 6 Second half of unique file ID |
| c_mode 6 File mode bits |
| c_uid 6 User ID of file owner |
| c_gid 6 Group ID of file owner |
| c_nlink 6 Number of links referencing file |
| c_rdev 6 Information for special files |
| c_mtime 11 Modification time of file |
| c_namesize 6 Length of file pathname |
| c_filesize 11 Length of file content |
| c_name c_namesize File pathname |
| c_filedata c_filesize File content |
The basic archive file structure is not changed for extended attributes. The file type bits stored in the c_mode field for an attribute file are set to 0xB000. As with the tar archive format, extended attributes are stored in cpio archives as two consecutive file entries. The first file describes the location/name for the extended attribute. The second file contains the actual attribute file content. The c_name field in extended attribute headers is ignored, though it should be set to a meaningful value for the benefit of archivers that do not process these headers. Solaris archivers start the pathname with "/dev/null/"to prevent archivers that do not understand the type 'E' header from trying to restore extended attribute files in inappropriate places.
Both the tar and cpio archive formats can contain the special files described above, always paired with the extended attribute data record, for identifying the precise location of the extended attribute. These special data files are necessary because there is no simple naming mechanism for extended attribute files. Extended attributes are not visible in the file system name space. The extended attribute name space must be "tunneled into" using the openat() function. The attribute identification data must support not only the flat naming structure for extended attributes, but also the possibility of future extensions allowing for attribute directory hierarchies and recursive attributes. The data file is therefore composed of a sequence of records. It begins with a fixed length header describing the content. The following table describes the format of this data file.
| Field Name Length (in Octets) Description |
| h_version 7 Name file version |
| h_size 10 Length of data file |
| h_component_len 10 Total length of all path segments |
| h_link_comp_len 10 Total length of all link segments |
| path h_component_len Complex path |
| link_path h_link_comp_len Complex link path |
As demonstrated above, the header is followed by a record describing the "path" to the attribute file. This path is composed of two or more path segments separated by a null character. Each segment describes a path rooted at the hidden extended attribute directory of the leaf file of the previous segment, making it possible to name attributes on attributes. The first segment is always the path to the parent file that roots the entire sequence in the normal name space. The following table describes the format of each segment.
| Field Name Length (in Octets) Description |
| h_namesz 7 Length of segment path |
| h_typeflag 1 Actual file type of attribute file |
| h_names h_namesz Parent path + segment path |
If the attribute file is linked to another file, the path record is followed by a second record describing the location of the referencing file. The structure of this record is identical to the record described above.
cp (1), cpio (1), du (1), find (1), ls (1), mv (1), pax (1), runat (1), tar (1), chown (2), link (2), open (2), pathconf (2), rename (2), stat (2), unlink (2), utimes (2), attropen (3C), standards (7), fsck (8)