newlib/winsup/doc/pathnames.xml
David Macek bbc87a02fd Fix inconsistencies in docs regarding fstab and executable file detection
The inline list of mount options seemed redundant, so the paragraph now points
to the list below it.

List of executable extensions updated according to fhandler_disk_file.cc. List
of executable magic numbers updated according to path.h (has_exec_chars).

	* pathnames.xml: Fix inconsistencies in docs regarding fstab and
	executable file detection

Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
2015-04-23 22:02:47 +02:00

570 lines
26 KiB
XML

<?xml version="1.0" encoding='UTF-8'?>
<!DOCTYPE sect1 PUBLIC "-//OASIS//DTD DocBook V4.5//EN"
"http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd">
<sect1 id="using-pathnames"><title>Mapping path names</title>
<sect2 id="pathnames-intro"><title>Introduction</title>
<para>Cygwin supports both POSIX- and Win32-style paths. Directory
delimiters may be either forward slashes or backslashes. Paths using
backslashes or starting with a drive letter are always handled as
Win32 paths. POSIX paths must only use forward slashes as delimiter,
otherwise they are treated as Win32 paths and file access might fail
in surprising ways.</para>
<note><para>The usage of Win32 paths, though possible, is deprecated,
since it circumvents important internal path handling mechanisms.
See <xref linkend="pathnames-win32"></xref> and
<xref linkend="pathnames-win32-api"></xref> for more information.
</para></note>
<para>POSIX operating systems (such as Linux) do not have the concept
of drive letters. Instead, all absolute paths begin with a
slash (instead of a drive letter such as "c:") and all file systems
appear as subdirectories (for example, you might buy a new disk and
make it be the <filename>/disk2</filename> directory).</para>
<para>Because many programs written to run on UNIX systems assume
the existence of a single unified POSIX file system structure, Cygwin
maintains a special internal POSIX view of the Win32 file system
that allows these programs to successfully run under Windows. Cygwin
uses this mapping to translate from POSIX to Win32 paths as
necessary.</para>
</sect2>
<sect2 id="mount-table"><title>The Cygwin Mount Table</title>
<para>The <filename>/etc/fstab</filename> file is used to map Win32
drives and network shares into Cygwin's internal POSIX directory tree.
This is a similar concept to the typical UNIX fstab file. The mount
points stored in <filename>/etc/fstab</filename> are globally set for
all users. Sometimes there's a requirement to have user specific
mount points. The Cygwin DLL supports user specific fstab files.
These are stored in the directory <filename>/etc/fstab.d</filename>
and the name of the file is the Cygwin username of the user, as it's
created from the Windows account database or stored in the
<filename>/etc/passwd</filename> file (see
<xref linkend="ntsec-mapping"></xref>). The structure of the
user specific file is identical to the system-wide
<filename>fstab</filename> file.</para>
<para>The file fstab contains descriptive information about the various file
systems. fstab is only read by programs, and not written; it is the
duty of the system administrator to properly create and maintain this
file. Each filesystem is described on a separate line; fields on each
line are separated by tabs or spaces. Lines starting with '#' are
comments.</para>
<para>The first field describes the block special device or
remote filesystem to be mounted. On Cygwin, this is the native Windows
path which the mount point links in. As path separator you MUST use a
slash. Usage of a backslash might lead to unexpected results. UNC
paths (using slashes, not backslashes) are allowed. If the path
contains spaces these can be escaped as <literal>'\040'</literal>.</para>
<para>The second field describes the mount point for the filesystem.
If the name of the mount point contains spaces these can be
escaped as '\040'.</para>
<para>The third field describes the type of the filesystem. Cygwin supports
any string here, since the file system type is usually not evaluated. So it
doesn't matter if you write <literal>FAT</literal> into this field even if
the filesystem is NTFS. Cygwin figures out the filesystem type and its
capabilities by itself.</para>
<para>The only exception is the file system type cygdrive. This type is
used to set the cygdrive prefix. For a description of the cygdrive prefix
see <xref linkend="cygdrive"></xref></para>
<para>The fourth field describes the mount options associated
with the filesystem. It is formatted as a comma separated list of
options. It contains at least the type of mount (binary or text) plus
any additional options appropriate to the filesystem type. The list of
the options, including their meaning, follows.</para>
<screen>
acl - Cygwin uses the filesystem's access control lists (ACLs) to
implement real POSIX permissions (default). This flag only
affects filesystems supporting ACLs (NTFS, for instance) and
is ignored otherwise.
auto - Ignored.
binary - Files default to binary mode (default).
bind - Allows to remount part of the file hierarchy somewhere else.
In contrast to other entries, the first field in the fstab
line specifies an absolute POSIX path. This path is remounted
to the POSIX path specified as the second path. The conversion
to a Win32 path is done on the fly. Only the root path and
paths preceding the bind entry in the fstab file are used to
convert the POSIX path in the first field to an absolute Win32
path. Note that symlinks are ignored while performing this path
conversion.
cygexec - Treat all files below mount point as cygwin executables.
dos - Always convert leading spaces and trailing dots and spaces to
characters in the UNICODE private use area. This allows to use
broken filesystems which only allow DOS filenames, even if they
are not recognized as such by Cygwin.
exec - Treat all files below mount point as executable.
ihash - Always fake inode numbers rather than using the ones returned
by the filesystem. This allows to use broken filesystems which
don't return unambiguous inode numbers, even if they are not
recognized as such by Cygwin.
noacl - Cygwin ignores filesystem ACLs and only fakes a subset of
permission bits based on the DOS readonly attribute. This
behaviour is the default on FAT and FAT32. The flag is
ignored on NFS filesystems.
nosuid - No suid files are allowed (currently unimplemented).
notexec - Treat all files below mount point as not executable.
nouser - Mount is a system-wide mount.
override - Force the override of an immutable mount point (currently "/").
posix=0 - Switch off case sensitivity for paths under this mount point
(default for the cygdrive prefix).
posix=1 - Switch on case sensitivity for paths under this mount point
(default for all other mount points).
sparse - Switch on support for sparse files. This option only makes
sense on NTFS and then only if you really need sparse files.
Cygwin does not try to create sparse files by default for
performance reasons.
text - Files default to CRLF text mode line endings.
user - Mount is a user mount.
</screen>
<para>While normally the execute permission bits are used to evaluate
executability, this is not possible on filesystems which don't support
permissions at all (like FAT/FAT32), or if ACLs are ignored on filesystems
supporting them (see the aforementioned <literal>acl</literal> mount option).
In these cases, the following heuristic is used to evaluate if a file is
executable: Files ending in certain extensions (.exe, .com, .lnk) are
assumed to be executable. Files whose first two characters are "#!", "MZ",
or ":\n" are also considered to be executable.
The <literal>exec</literal> option is used to instruct Cygwin that the
mounted file is "executable". If the <literal>exec</literal> option is used
with a directory then all files in the directory are executable.
This option allows other files to be marked as executable and avoids the
overhead of opening each file to check for "magic" bytes. The
<literal>cygexec</literal> option is very similar to <literal>exec</literal>,
but also prevents Cygwin from setting up commands and environment variables
for a normal Windows program, adding another small performance gain. The
opposite of these options is the <literal>notexec</literal> option, which
means that no files should be marked as executable under that mount point.</para>
<para>A correct root directory is quite essential to the operation of
Cygwin. A default root directory is evaluated at startup so a
<filename>fstab</filename> entry for the root directory is not necessary.
If it's wrong, nothing will work as expected. Therefore, the root directory
evaluated by Cygwin itself is treated as an immutable mount point and can't
be overridden in /etc/fstab... unless you think you really know what you're
doing. In this case, use the <literal>override</literal> flag in the options
field in the <filename>/etc/fstab</filename> file. Since this is a dangerous
thing to do, do so at your own risk.</para>
<para><filename>/usr/bin</filename> and <filename>/usr/lib</filename> are
by default also automatic mount points generated by the Cygwin DLL similar
to the way the root directory is evaluated. <filename>/usr/bin</filename>
points to the directory the Cygwin DLL is installed in,
<filename>/usr/lib</filename> is supposed to point to the
<filename>/lib</filename> directory. This choice is safe and usually
shouldn't be changed. An fstab entry for them is not required.</para>
<para><literal>nouser</literal> mount points are not overridable by a later
call to <command>mount</command>.
Mount points given in <filename>/etc/fstab</filename> are by default
<literal>nouser</literal> mount points, unless you specify the option
<literal>user</literal>. This allows the administrator to set certain
paths so that they are not overridable by users. In contrast, all mount
points in the user specific fstab file are <literal>user</literal> mount
points.</para>
<para>The fifth and sixth field are ignored. They are
so far only specified to keep a Linux-like fstab file layout.</para>
<para>Note that you don't have to specify an fstab entry for the root dir,
unless you want to have the root dir pointing to somewhere entirely
different (hopefully you know what you're doing), or if you want to
mount the root dir with special options (for instance, as text mount).</para>
<para>Example entries:</para>
<itemizedlist spacing="compact">
<listitem>
<para>Just a normal mount point:</para>
<screen> c:/foo /bar fat32 binary 0 0</screen>
</listitem>
<listitem>
<para>A mount point for a textmode mount with case sensitivity switched off:</para>
<screen> C:/foo /bar/baz ntfs text,posix=0 0 0</screen>
</listitem>
<listitem>
<para>A mount point for a Windows directory with spaces in it:</para>
<screen> C:/Documents\040and\040Settings /docs ext3 binary 0 0</screen>
</listitem>
<listitem>
<para>A mount point for a remote directory, don't store POSIX permissions in ACLs:</para>
<screen> //server/share/subdir /srv/subdir smbfs binary,noacl 0 0</screen>
</listitem>
<listitem>
<para>This is just a comment:</para>
<screen> # This is just a comment</screen>
</listitem>
<listitem>
<para>Set the cygdrive prefix to /mnt:</para>
<screen> none /mnt cygdrive binary 0 0</screen>
</listitem>
<listitem>
<para>Remount /var to /usr/var:</para>
<screen> /var /usr/var none bind</screen>
<para>Assuming <filename>/var</filename> points to
<filename>C:/cygwin/var</filename>, <filename>/usr/var</filename> now
also points to <filename>C:/cygwin/var</filename>. This is equivalent
to the Linux <literal>bind</literal> option available since
Linux 2.4.0.</para>
</listitem>
</itemizedlist>
<para>Whenever Cygwin generates a Win32 path from a POSIX one, it uses
the longest matching prefix in the mount table. Thus, if
<filename>C:</filename> is mounted as <filename>/c</filename> and also
as <filename>/</filename>, then Cygwin would translate
<filename>C:/foo/bar</filename> to <filename>/c/foo/bar</filename>.
This translation is normally only used when trying to derive the
POSIX equivalent current directory. Otherwise, the handling of MS-DOS
filenames bypasses the mount table.
</para>
<para>If you want to see the current set of mount points valid in your
session, you can invoke the Cygwin tool <command>mount</command> without
arguments:</para>
<example id="pathnames-mount-ex">
<title>Displaying the current set of mount points</title>
<screen>
<prompt>bash$</prompt> <userinput>mount</userinput>
f:/cygwin/bin on /usr/bin type ntfs (binary,auto)
f:/cygwin/lib on /usr/lib type ntfs (binary,auto)
f:/cygwin on / type ntfs (binary,auto)
e:/src on /usr/src type vfat (binary)
c: on /cygdrive/c type ntfs (binary,posix=0,user,noumount,auto)
e: on /cygdrive/e type vfat (binary,posix=0,user,noumount,auto)
</screen>
</example>
<para>You can also use the <command>mount</command> command to add
new mount points, and the <command>umount</command> to delete
them. However, since they are only stored in memory, these mount
points will disappear as soon as your last Cygwin process ends.
See <xref linkend="mount"></xref> and <xref linkend="umount"></xref> for more
information.</para>
</sect2>
<sect2 id="unc-paths"><title>UNC paths</title>
<para>Apart from the unified POSIX tree starting at the <filename>/</filename>
directory, UNC pathnames starting with two slashes and a server name
(<filename>//machine/share/...</filename>) are supported as well.
They are handled as POSIX paths if only containing forward slashes. There's
also a virtual directory <filename>//</filename> which allows to enumerate
the fileservers known to the local machine with <command>ls</command>.
Same goes for the UNC paths of the type <filename>//machine</filename>,
which allow to enumerate the shares provided by the server
<literal>machine</literal>. For often used UNC paths it makes sense to
add them to the mount table (see <xref linkend="mount-table"></xref> so
they are included in the unified POSIX path tree.</para>
</sect2>
<sect2 id="cygdrive"><title>The cygdrive path prefix</title>
<para>As already outlined in <xref linkend="ov-hi-files"></xref>, you can
access arbitary drives on your system by using the cygdrive path prefix.
The default value for this prefix is <filename>/cygdrive</filename>, and
a path to any drive can be constructed by using the cygdrive prefix and
appending the drive letter as subdirectory, like this:</para>
<screen>
bash$ ls -l /cygdrive/f/somedir
</screen>
<para>This lists the content of the directory F:\somedir.</para>
<para>The cygdrive prefix is a virtual directory under which all drives
on a system are subsumed. The mount options of the cygdrive prefix is
used for all file access through the cygdrive prefixed drives. For instance,
assuming the cygdrive mount options are <literal>binary,posix=0</literal>,
then any file <filename>/cygdrive/x/file</filename> will be opened in
binary mode by default (mount option <literal>binary</literal>), and the case
of the filename doesn't matter (mount option <literal>posix=0</literal>).
</para>
<para>The cygdrive prefix flags are also used for all UNC paths starting with
two slashes, unless they are accessed through a mount point. For instance,
consider these <filename>/etc/fstab</filename> entries:</para>
<screen>
//server/share /mysrv ntfs posix=1,acl 0 0
none /cygdrive cygdrive posix=0,noacl 0 0
</screen>
<para>Assume there's a file <filename>\\server\share\foo</filename> on the
share. When accessing it as <filename>/mysrv/foo</filename>, then the flags
<literal>posix=1,acl</literal> of the /mysrv mount point are used. When
accessing it as <filename>//server/share/foo</filename>, then the flags
for the cygdrive prefix, <literal>posix=0,noacl</literal> are used.</para>
<note><para>This only applies to UNC paths using forward slashes. When
using backslashes the flags for native paths are used. See
<xref linkend="pathnames-win32"></xref>.</para></note>
<para>The cygdrive prefix may be changed in the fstab file as outlined above.
Please note that you must not use the cygdrive prefix for any other mount
point. For instance this:</para>
<screen>
none /cygdrive cygdrive binary 0 0
D: /cygdrive/d somefs text 0 0
</screen>
<para>will not make file access using the /mnt/d path prefix suddenly using
textmode. If you want to mount any drive explicitly in another mode than
the cygdrive prefix, use a distinct path prefix:</para>
<screen>
none /cygdrive cygdrive binary 0 0
D: /mnt/d somefs text 0 0
</screen>
<para>To simplify scripting, Cygwin also provides a
<filename>/proc/cygdrive</filename> symlink, which allows to use a fixed path
in scripts, even if the actual cygdrive prefix has been changed, or is different
between different users. So, in scripts, conveniently use the
<filename>/proc/cygdrive</filename> symlink to successfully access files
independently from the current cygdrive prefix:</para>
<screen>
$ mount -p
Prefix Type Flags
/mnt user binmode
$ cat &gt; x.sh &lt;&lt;EOF
cd /proc/cygdrive/c/Windows/System32/Drivers/etc
ls -l hosts
EOF
$ sh -c ./x.sh
-rwxrwx---+ 1 SYSTEM SYSTEM 826 Sep 4 22:43 hosts
</screen>
</sect2>
<sect2 id="pathnames-symlinks"><title>Symbolic links</title>
<para>Symbolic links are not present and supported on Windows until Windows
Vista/Server 2008, and then only on some filesystems. Since POSIX applications
are rightfully expecting to use symlinks and the
<literal>symlink(2)</literal> system call, Cygwin had to find a
workaround for this Windows flaw.</para>
<para>Cygwin creates symbolic links potentially in multiple different
ways:</para>
<itemizedlist mark="bullet">
<listitem>
<para>The default symlinks are plain files containing a magic cookie
followed by the path to which the link points. They are marked with the
DOS SYSTEM attribute so that only files with that attribute have to be
read to determine whether or not the file is a symbolic link.</para>
<note><para>Starting with Cygwin 1.7, symbolic links are using UTF-16 to encode
the filename of the target file, to better support internationalization.
Symlinks created by older Cygwin releases can be read just fine. However,
you could run into problems with them if you're now using another character
set than the one you used when creating these symlinks
(see <xref linkend="setup-locale-problems"></xref>). Please note that this
new UTF-16 style of symlinks is not compatible with older Cygwin release,
which can't read the target filename correctly.</para></note>
</listitem>
<listitem>
<para>The shortcut style symlinks are Windows <literal>.lnk</literal>
shortcut files with a special header and the DOS READONLY attribute set.
This symlink type is created if the environment variable
<literal>CYGWIN</literal> (see <xref linkend="using-cygwinenv"></xref>)
is set to contain the string <literal>winsymlinks</literal> or
<literal>winsymlinks:lnk</literal>. On the MVFS filesystem, which does
not support the DOS SYSTEM attribute, this is the one and only supported
symlink type, independently from the <literal>winsymlinks</literal>
setting.</para>
</listitem>
<listitem>
<para>Native Windows symlinks are only created on Windows Vista/2008 and later,
and only on filesystems supporting reparse points. Due to to their weird
restrictions and behaviour, they are only created if the user
explicitely requests creating them. This is done by setting the
environment variable <literal>CYGWIN</literal> to contain the string
<literal>winsymlinks:native</literal> or
<literal>winsymlinks:nativestrict</literal>. For the difference between
these two settings, see <xref linkend="using-cygwinenv"></xref>.
On AFS, native symlinks are the only supported type of symlink due to
AFS lacking support for DOS attributes. This is independent from the
<literal>winsymlinks</literal> setting.</para>
</listitem>
<listitem>
<para>On the NFS filesystem, Cygwin always creates real NFS symlinks.</para>
</listitem>
</itemizedlist>
<para>All of the above four symlink types are recognized and used as symlinks
under all circumstances. However, if the default plain file symlink type
is lacking its DOS SYSTEM bit, or if the shortcut file is lacking the DOS
READONLY attribute, they are not recognized as symlink.</para>
<para>Apart from these four types, there's also a fifth type, which is
recognized as symlink but never generated by Cygwin, directory
junctions. This is an older reparse point type, supported by Windows
since Windows 2000. Filesystem junctions on the other hand are not
handled as symlinks, since otherwise they would not be recognized as
filesystem borders by commands like <command>find -xdev</command>.</para>
</sect2>
<sect2 id="pathnames-win32"><title>Using native Win32 paths</title>
<para>Using native Win32 paths in Cygwin, while possible, is generally
inadvisable. Those paths circumvent all internal integrity checking and
bypass the information given in the Cygwin mount table.</para>
<para>The following paths are treated as native Win32 paths in Cygwin:</para>
<itemizedlist spacing="compact">
<listitem>
<para>All paths starting with a drive specifier</para>
<screen>
C:\foo
C:/foo
</screen>
</listitem>
<listitem>
<para>All paths containing at least one backslash as path component</para>
<screen>
C:/foo/bar<emphasis role='bold'>\</emphasis>baz/...
</screen>
</listitem>
<listitem>
<para>UNC paths using backslashes</para>
<screen>
\\server\share\...
</screen>
</listitem>
</itemizedlist>
<para>When accessing files using native Win32 paths as above, Cygwin uses a
default setting for the mount flags. All paths using DOS notation will be
treated as case insensitive, and permissions are just faked as if the
underlying drive is a FAT drive. This also applies to NTFS and other
filesystems which usually are capable of case sensitivity and storing
permissions.</para>
</sect2>
<sect2 id="pathnames-win32-api"><title>Using the Win32 file API in Cygwin applications</title>
<para>Special care must be taken if your application uses Win32 file API
functions like <function>CreateFile</function> to access files using
relative pathnames, or if your application uses functions like
<function>CreateProcess</function> or <function>ShellExecute</function>
to start other applications.</para>
<para>When a Cygwin application is started, the Windows idea of the current
working directory (CWD) is not necessarily the same as the Cygwin CWD.
There are a couple of restrictions in the Win32 API, which disallow certain
directories as Win32 CWD:</para>
<itemizedlist spacing="compact">
<listitem>
<para>The Windows subsystem only supports CWD paths of up to 258 chars.
This restriction doesn't apply for Cygwin processes, at least not as
long as they use the POSIX API (chdir, getcwd). This means, if a Cygwin
process has a CWD using an absolute path longer than 258 characters, the
Cygwin CWD and the Windows CWD differ.</para>
</listitem>
<listitem>
<para>The Win32 API call to set the current directory,
<function>SetCurrentDirectory</function>, fails for directories for which
the user has no permissions, even if the user is an administrator. This
restriction doesn't apply for Cygwin processes, if they are running under
an administrator account.</para>
</listitem>
<listitem>
<para><function>SetCurrentDirectory</function> does not support
case-sensitive filenames.
</para>
</listitem>
<listitem>
<para>Last, but not least, <function>SetCurrentDirectory</function> can't
work on virtual Cygwin paths like /proc or /cygdrive. These paths only
exists in the Cygwin realm so they have no meaning to a native Win32
process.</para>
</listitem>
</itemizedlist>
<para>As long as the Cygwin CWD is usable as Windows CWD, the Cygwin and
Windows CWDs are in sync within a process. However, if the Cygwin process
changes its working directory into one of the directories which are
unusable as Windows CWD, we're in trouble. If the process uses the
Win32 API to access a file using a relative pathname, the resulting
absolute path would not match the expectations of the process. In the
worst case, the wrong files are deleted.</para>
<para>To workaround this problem, Cygwin sets the Windows CWD to a special
directory in this case. This special directory points to a virtual
filesystem within the native NT namespace (<filename>\??\PIPE\</filename>).
Since it's not a real filesystem, the deliberate effect is that a call to,
for instance, <function>CreateFile ("foo", ...);</function> will fail,
as long as the processes CWD doesn't work as Windows CWD.</para>
<para>So, in general, don't use the Win32 file API in Cygwin applications.
If you <emphasis role='bold'>really</emphasis> need to access files using
the Win32 API, or if you <emphasis role='bold'>really</emphasis> have to use
<function>CreateProcess</function> to start applications, rather than
the POSIX <function>exec(3)</function> family of functions, you have to
make sure that the Cygwin CWD is set to some directory which is valid as
Win32 CWD.</para>
</sect2>
<sect2 id="pathnames-additional"><title>Additional Path-related Information</title>
<para>The <command>cygpath</command> program provides the ability to
translate between Win32 and POSIX pathnames in shell scripts. See
<xref linkend="cygpath"></xref> for the details.</para>
<para>The <envar>HOME</envar>, <envar>PATH</envar>, and
<envar>LD_LIBRARY_PATH</envar> environment variables are automatically
converted from Win32 format to POSIX format (e.g. from
<filename>c:/cygwin\bin</filename> to <filename>/bin</filename>, if
there was a mount from that Win32 path to that POSIX path) when a Cygwin
process first starts.</para>
<para>Symbolic links can also be used to map Win32 pathnames to POSIX.
For example, the command
<command>ln -s //pollux/home/joe/data /data</command> would have about
the same effect as creating a mount point from
<filename>//pollux/home/joe/data</filename> to <filename>/data</filename>
using <command>mount</command>, except that symbolic links cannot set
the default file access mode. Other differences are that the mapping is
distributed throughout the file system and proceeds by iteratively
walking the directory tree instead of matching the longest prefix in a
kernel table. Note that symbolic links will only work on network
drives that are properly configured to support the "system" file
attribute. Many do not do so by default (the Unix Samba server does
not by default, for example).</para>
</sect2>
</sect1>