10. Files

10.1. Binaries

Two different packages must not install programs with different functionality but with the same filenames. (The case of two programs having the same functionality but different implementations is handled via “alternatives” or the “Conflicts” mechanism. See Maintainer Scripts and Conflicting binary packages - Conflicts respectively.) If this case happens, one of the programs must be renamed. The maintainers should report this to the debian-devel mailing list and try to find a consensus about which program will have to be renamed. If a consensus cannot be reached, both programs must be renamed.

To support merged-/usr systems, packages must not install files in both /path and /usr/path. For example, a package must not install both /bin/example and /usr/bin/example.

If a file is moved between /path and /usr/path in revisions of a Debian package, and a compatibility symlink at the old path is needed, the symlink must be managed in a way that will not break when /path and /usr/path are the same underlying directory due to symlinks or other mechanisms.

Binary executables must not be statically linked with the GNU C library, since this prevents the binary from benefiting from fixes and improvements to the C library without being rebuilt and complicates security updates. This requirement may be relaxed for binary executables whose intended purpose is to diagnose and fix the system in situations where the GNU C library may not be usable (such as system recovery shells or utilities like ldconfig) or for binary executables where the security benefits of static linking outweigh the drawbacks.

By default, when a package is being built, any binaries created should include debugging information, as well as being compiled with optimization. You should also turn on as many reasonable compilation warnings as possible (see Main building script: debian/rules). For the C programming language, this means the following compilation parameters should be used:

CC = gcc
CFLAGS = -O2 -g -Wall # sane warning options vary between programs
LDFLAGS = # none

By default all installed binaries should be stripped by calling

strip --strip-unneeded --remove-section=.comment --remove-section=.note binaries

on the binaries after they have been copied into debian/tmp but before the tree is made into a package.

It is not recommended to strip binaries by passing the -s flag to install, because this fails to remove .comment and .note sections, and also prevents the automatic creation of dbgsym binary packages by tools like dh_strip.

Although binaries in the build tree should be compiled with debugging information by default, it can often be difficult to debug programs if they are also subjected to compiler optimization. For this reason, it is recommended to support the standardized environment variable DEB_BUILD_OPTIONS (see debian/rules and DEB_BUILD_OPTIONS). This variable can contain several flags to change how a package is compiled and built.

It is up to the package maintainer to decide what compilation options are best for the package. Certain binaries (such as computationally-intensive programs) will function better with certain flags (-O3, for example); feel free to use them. Please use good judgment here. Don’t use flags for the sake of it; only use them if there is good reason to do so. Feel free to override the upstream author’s ideas about which compilation options are best: they are often inappropriate for our environment.

10.2. Libraries

If the package is architecture: any, then the shared library compilation and linking flags must have -fPIC, or the package shall not build on some of the supported architectures. 1 Any exception to this rule must be discussed on the mailing list debian-devel@lists.debian.org, and a rough consensus obtained. The reasons for not compiling with -fPIC flag must be recorded in the file README.Debian, and care must be taken to either restrict the architecture or arrange for -fPIC to be used on architectures where it is required. 2

As to the static libraries, the common case is not to have relocatable code, since there is no benefit, unless in specific cases; therefore the static version must not be compiled with the -fPIC flag. Any exception to this rule should be discussed on the mailing list debian-devel@lists.debian.org, and the reasons for compiling with the -fPIC flag must be recorded in the file README.Debian. 3

In other words, if both a shared and a static library is being built, each source unit (*.c, for example, for C files) will need to be compiled twice, for the normal case.

Libraries should be built with threading support and to be thread-safe if the library supports this.

Although not enforced by the build tools, shared libraries must be linked against all libraries that they use symbols from in the same way that binaries are. This ensures the correct functioning of the symbols and shlibs systems and guarantees that all libraries can be safely opened with dlopen(). Packagers may wish to use the gcc option -Wl,-z,defs when building a shared library. Since this option enforces symbol resolution at build time, a missing library reference will be caught early as a fatal build error.

All installed shared libraries should be stripped with

strip --strip-unneeded --remove-section=.comment --remove-section=.note your-lib

(The option --strip-unneeded makes strip remove only the symbols which aren’t needed for relocation processing.) Shared libraries can function perfectly well when stripped, since the symbols for dynamic linking are in a separate part of the ELF object file. 4

Note that under some circumstances it may be useful to install a shared library unstripped, for example when building a separate package to support debugging. The debhelper dh_strip` tool can create such packages automatically.

Shared object files (often .so files) that are not public libraries, that is, they are not meant to be linked to by third party executables (binaries of other packages), should be installed in subdirectories of the /usr/lib or /usr/lib/triplet directories (see the FHS for a definition). Such files are exempt from the rules that govern ordinary shared libraries, except that they must not be installed executable and should be stripped. 5

Packages that use libtool to create and install their shared libraries install a file containing additional metadata (ending in .la) alongside the library. For public libraries intended for use by other packages, these files normally should not be included in the Debian package, since the information they include is not necessary to link with the shared library on Debian and can add unnecessary additional dependencies to other programs or libraries. 6 If the .la file is required for that library (if, for instance, it’s loaded via libltdl in a way that requires that meta-information), the dependency_libs setting in the .la file should normally be set to the empty string. If the shared library development package has historically included the .la, it must be retained in the development package (with dependency_libs emptied) until all libraries that depend on it have removed or emptied dependency_libs in their .la files to prevent linking with those other libraries using libtool from failing.

If the .la must be included, it should be included in the development (-dev) package, unless the library will be loaded by libtool’s libltdl library. If it is intended for use with libltdl, the .la files must go in the run-time library package.

These requirements for handling of .la files do not apply to loadable modules or libraries not installed in directories searched by default by the dynamic linker. Packages installing loadable modules will frequently need to install the .la files alongside the modules so that they can be loaded by libltdl. dependency_libs does not need to be modified for libraries or modules that are not installed in directories searched by the dynamic linker by default and not intended for use by other packages.

You must make sure that you use only released versions of shared libraries to build your packages; otherwise other users will not be able to run your binaries properly. Producing source packages that depend on unreleased compilers is also usually a bad idea.

10.3. Shared libraries

This section has moved to Shared libraries.

10.4. Scripts

All command scripts, including the package maintainer scripts inside the package and used by dpkg, should have a #! line naming the shell to be used to interpret them.

In the case of Perl scripts this should be #!/usr/bin/perl.

When scripts are installed into a directory in the system PATH, the script name should not include an extension such as .sh or .pl that denotes the scripting language currently used to implement it.

Shell scripts (sh and bash) other than init.d scripts should almost certainly start with set -e so that errors are detected. init.d scripts are something of a special case, due to how frequently they need to call commands that are allowed to fail, and it may instead be easier to check the exit status of commands directly. See Writing the scripts for more information about writing init.d scripts.

Every script should use set -e or check the exit status of every command.

Scripts may assume that /bin/sh implements the POSIX.1-2017 Shell Command Language 7 plus the following additional features not mandated by POSIX.1-2017.. 8

  • echo -n, if implemented as a shell built-in, must not generate a newline.

  • test, if implemented as a shell built-in, must support -a and -o as binary logical operators.

  • local to create a scoped variable must be supported, including listing multiple variables in a single local command and assigning a value to a variable at the same time as localizing it. local may or may not preserve the variable value from an outer scope if no assignment is present. Uses such as:

    fname () {
        local a b c=delta d
        # ... use a, b, c, d ...
    }
    

    must be supported and must set the value of c to delta.

  • The XSI extension to kill allowing kill -signal, where signal is either the name of a signal or one of the numeric signals listed in the XSI extension (0, 1, 2, 3, 6, 9, 14, and 15), must be supported if kill is implemented as a shell built-in.

  • The XSI extension to trap allowing numeric signals must be supported. In addition to the signal numbers listed in the extension, which are the same as for kill above, 13 (SIGPIPE) must be allowed.

If a shell script requires non-POSIX.1-2017 features from the shell interpreter other than those listed above, the appropriate shell must be specified in the first line of the script (e.g., #!/bin/bash) and the package must depend on the package providing the shell (unless the shell package is marked “Essential”, as in the case of bash).

You may wish to restrict your script to POSIX.1-2017 features plus the above set when possible so that it may use /bin/sh as its interpreter. Checking your script with checkbashisms from the devscripts package or running your script with an alternate shell such as posh may help uncover violations of the above requirements. If in doubt whether a script complies with these requirements, use /bin/bash.

Perl scripts should check for errors when making any system calls, including open, print, close, rename and system.

csh and tcsh should be avoided as scripting languages. See Csh Programming Considered Harmful, one of the comp.unix.* FAQs, which can be found at http://www.faqs.org/faqs/unix-faq/shell/csh-whynot/. If an upstream package comes with csh scripts then you must make sure that they start with #!/bin/csh and make your package depend on the c-shell virtual package.

Any scripts which create files in world-writeable directories (e.g., in /tmp) must use a mechanism which will fail atomically if a file with the same name already exists.

The Debian base system provides the tempfile and mktemp utilities for use by scripts for this purpose.

10.6. Device files

Packages must not include device files or named pipes in the package file tree.

Debian packages should assume that device files in /dev are dynamically managed by the kernel or some other system facility and do not have to be explicitly created or managed by the package. Debian packages other than those whose purpose is to manage the /dev device file tree must not attempt to create or remove device files in /dev when a dynamic device management facility is in use.

If named pipes or device files outside of /dev are required by a package, they should normally be created when necessary by the programs in the package, by init scripts or systemd unit files, or by similar on-demand mechanisms. If such files need to be created during package installation, they must be created in the postinst maintainer script 10 and removed in either the prerm or the postrm maintainer script.

10.7. Configuration files

10.7.1. Definitions

configuration file

A file that affects the operation of a program, or provides site- or host-specific information, or otherwise customizes the behavior of a program. Typically, configuration files are intended to be modified by the system administrator (if needed or desired) to conform to local policy or to provide more useful site-specific behavior.

conffile

A file listed in a package’s conffiles file, and is treated specially by dpkg (see Details of configuration).

The distinction between these two is important; they are not interchangeable concepts. Almost all conffiles are configuration files, but many configuration files are not conffiles.

As noted elsewhere, /etc/init.d scripts, /etc/default files, scripts installed in /etc/cron.{hourly,daily,weekly,monthly}, and cron configuration installed in /etc/cron.d must be treated as configuration files. In general, any script that embeds configuration information is de-facto a configuration file and should be treated as such.

10.7.2. Location

Any configuration files created or used by your package must reside in /etc. If there are several, consider creating a subdirectory of /etc named after your package.

If your package creates or uses configuration files outside of /etc, and it is not feasible to modify the package to use /etc directly, put the files in /etc and create symbolic links to those files from the location that the package requires.

10.7.3. Behavior

Configuration file handling must conform to the following behavior:

  • local changes must be preserved during a package upgrade, and

  • configuration files must be preserved when the package is removed, and only deleted when the package is purged.

Obsolete configuration files without local changes should be removed by the package during upgrade. 11

The easy way to achieve this behavior is to make the configuration file a conffile. This is appropriate only if it is possible to distribute a default version that will work for most installations, although some system administrators may choose to modify it. This implies that the default version will be part of the package distribution, and must not be modified by the maintainer scripts during installation (or at any other time).

In order to ensure that local changes are preserved correctly, packages must not contain or make hard links to conffiles. 12

The other way to do it is via the maintainer scripts. In this case, the configuration file must not be listed as a conffile and must not be part of the package distribution. If the existence of a file is required for the package to be sensibly configured it is the responsibility of the package maintainer to provide maintainer scripts which correctly create, update and maintain the file and remove it on purge. (See Package maintainer scripts and installation procedure for more information.) These scripts must be idempotent (i.e., must work correctly if dpkg needs to re-run them due to errors during installation or removal), must cope with all the variety of ways dpkg can call maintainer scripts, must not overwrite or otherwise mangle the user’s configuration without asking, must not ask unnecessary questions (particularly during upgrades), and must otherwise be good citizens.

The scripts are not required to configure every possible option for the package, but only those necessary to get the package running on a given system. Ideally the sysadmin should not have to do any configuration other than that done (semi-)automatically by the postinst script.

A common practice is to create a script called package-configure and have the package’s postinst call it if and only if the configuration file does not already exist. In certain cases it is useful for there to be an example or template file which the maintainer scripts use. Such files should be in /usr/share/package or /usr/lib/package (depending on whether they are architecture-independent or not). There should be symbolic links to them from /usr/share/doc/package/examples if they are examples, and should be perfectly ordinary dpkg-handled files (not configuration files).

These two styles of configuration file handling must not be mixed, for that way lies madness: dpkg will ask about overwriting the file every time the package is upgraded.

10.7.4. Sharing configuration files

If two or more packages use the same configuration file and it is reasonable for both to be installed at the same time, one of these packages must be defined as owner of the configuration file, i.e., it will be the package which handles that file as a configuration file. Other packages that use the configuration file must depend on the owning package if they require the configuration file to operate. If the other package will use the configuration file if present, but is capable of operating without it, no dependency need be declared.

If it is desirable for two or more related packages to share a configuration file and for all of the related packages to be able to modify that configuration file, then the following should be done:

  1. One of the related packages (the “owning” package) will manage the configuration file with maintainer scripts as described in the previous section.

  2. The owning package should also provide a program that the other packages may use to modify the configuration file.

  3. The related packages must use the provided program to make any desired modifications to the configuration file. They should either depend on the core package to guarantee that the configuration modifier program is available or accept gracefully that they cannot modify the configuration file if it is not. (This is in addition to the fact that the configuration file may not even be present in the latter scenario.)

Sometimes it’s appropriate to create a new package which provides the basic infrastructure for the other packages and which manages the shared configuration files. (The sgml-base package is a good example.)

If the configuration file cannot be shared as described above, the packages must be marked as conflicting with each other. Two packages that specify the same file as a conffile must conflict. This is an instance of the general rule about not sharing files. Neither alternatives nor diversions are likely to be appropriate in this case; in particular, dpkg does not handle diverted conffiles well.

When two packages both declare the same conffile, they may see left-over configuration files from each other even though they conflict with each other. If a user removes (without purging) one of the packages and installs the other, the new package will take over the conffile from the old package. If the file was modified by the user, it will be treated the same as any other locally modified conffile during an upgrade.

The maintainer scripts must not alter a conffile of any package, including the one the scripts belong to.

10.7.5. User configuration files (“dotfiles”)

The files in /etc/skel will automatically be copied into new user accounts by adduser. No other program should reference the files in /etc/skel.

Therefore, if a program needs a dotfile to exist in advance in $HOME to work sensibly, that dotfile should be installed in /etc/skel and treated as a configuration file.

However, programs that require dotfiles in order to operate sensibly are a bad thing, unless they do create the dotfiles themselves automatically.

Furthermore, programs should be configured by the Debian default installation to behave as closely to the upstream default behavior as possible.

Therefore, if a program in a Debian package needs to be configured in some way in order to operate sensibly, that should be done using a site-wide configuration file placed in /etc. Only if the program doesn’t support a site-wide default configuration and the package maintainer doesn’t have time to add it may a default per-user file be placed in /etc/skel.

/etc/skel should be as empty as we can make it. This is particularly true because there is no easy (or necessarily desirable) mechanism for ensuring that the appropriate dotfiles are copied into the accounts of existing users when a package is installed.

10.8. Log files

Log files should usually be named /var/log/package.log. If you have many log files, or need a separate directory for permission reasons (/var/log is writable only by root), you should usually create a directory named /var/log/package and place your log files there.

Log files must be rotated occasionally so that they don’t grow indefinitely. The best way to do this is to install a log rotation configuration file in the directory /etc/logrotate.d, normally named /etc/logrotate.d/package, and use the facilities provided by logrotate. 13 Here is a good example for a logrotate config file (for more information see logrotate(8)):

/var/log/foo/*.log {
    rotate 12
    weekly
    compress
    missingok
    postrotate
        start-stop-daemon -K -p /var/run/foo.pid -s HUP -x /usr/sbin/foo -q
    endscript
}

This rotates all files under /var/log/foo, saves 12 compressed generations, and tells the daemon to reopen its log files after the log rotation. It skips this log rotation (via missingok) if no such log file is present, which avoids errors if the package is removed but not purged.

Log files should be removed when the package is purged (but not when it is only removed). This should be done by the postrm script when it is called with the argument purge (see Details of removal and/or configuration purging).

10.9. Permissions and owners

The rules in this section are guidelines for general use. If necessary you may deviate from the details below. However, if you do so you must make sure that what is done is secure and you should try to be as consistent as possible with the rest of the system. You are also encouraged to discuss it on debian-devel first.

Files should be owned by root:root, and made writable only by the owner and universally readable (and executable, if appropriate), that is mode 644 or 755.

Directories should be mode 755 or (for group-writability) mode 2775. The ownership of the directory should be consistent with its mode: if a directory is mode 2775, it should be owned by the group that needs write access to it. 14

Control information files should be owned by root:root and either mode 644 (for most files) or mode 755 (for executables such as maintainer scripts).

Setuid and setgid executables should be mode 4755 or 2755 respectively, and owned by the appropriate user or group. They should not be made unreadable (modes like 4711 or 2711 or even 4111); doing so achieves no extra security, because anyone can find the binary in the freely available Debian package; it is merely inconvenient. For the same reason you should not restrict read or execute permissions on non-set-id executables.

Some setuid programs need to be restricted to particular sets of users, using file permissions. In this case they should be owned by the uid to which they are set-id, and by the group which should be allowed to execute them. They should have mode 4754; again there is no point in making them unreadable to those users who must not be allowed to execute them.

It is possible to arrange that the system administrator can reconfigure the package to correspond to their local security policy by changing the permissions on a binary: they can do this by using dpkg-statoverride, as described below. 15 Another method you should consider is to create a group for people allowed to use the program(s) and make any setuid executables executable only by that group.

If you need to create a new user or group for your package there are two possibilities. Firstly, you may need to make some files in the binary package be owned by this user or group, or you may need to compile the user or group id (rather than just the name) into the binary (though this latter should be avoided if possible, as in this case you need a statically allocated id).

If you need a statically allocated id, you must ask for a user or group id from the base-passwd maintainer, and must not release the package until you have been allocated one. Once you have been allocated one you must either make the package depend on a version of the base-passwd package with the id present in /etc/passwd or /etc/group, or arrange for your package to create the user or group itself with the correct id (using adduser) in its preinst or postinst. (Doing it in the postinst is to be preferred if it is possible, otherwise a pre-dependency will be needed on the adduser package.)

On the other hand, the program might be able to determine the uid or gid from the user or group name at runtime, so that a dynamically allocated id can be used. In this case you should choose an appropriate user or group name, discussing this on debian-devel and checking that it is unique. When this has been checked you must arrange for your package to create the user or group if necessary using adduser in the preinst or postinst script (again, the latter is to be preferred if it is possible).

Note that changing the numeric value of an id associated with a name is very difficult, and involves searching the file system for all appropriate files. You need to think carefully whether a static or dynamic id is required, since changing your mind later will cause problems.

10.9.1. The use of dpkg-statoverride

This section is not intended as policy, but as a description of the use of dpkg-statoverride.

If a system administrator wishes to have a file (or directory or other such thing) installed with owner and permissions different from those in the distributed Debian package, they can use the dpkg-statoverride program to instruct dpkg to use the different settings every time the file is installed. Thus the package maintainer should distribute the files with their normal permissions, and leave it for the system administrator to make any desired changes. For example, a daemon which is normally required to be setuid root, but in certain situations could be used without being setuid, should be installed setuid in the .deb. Then the local system administrator can change this if they wish. If there are two standard ways of doing it, the package maintainer can use debconf to find out the preference, and call dpkg-statoverride in the maintainer script if necessary to accommodate the system administrator’s choice. Care must be taken during upgrades to not override an existing setting.

Given the above, dpkg-statoverride is essentially a tool for system administrators and would not normally be needed in the maintainer scripts. There is one type of situation, though, where calls to dpkg-statoverride would be needed in the maintainer scripts, and that involves packages which use dynamically allocated user or group ids. In such a situation, something like the following idiom can be very helpful in the package’s postinst, where sysuser is a dynamically allocated id:

for i in /usr/bin/foo /usr/sbin/bar; do
    # only do something when no setting exists
    if ! dpkg-statoverride --list $i >/dev/null 2>&1; then
        #include: debconf processing, question about foo and bar
        if [ "$RET" = "true" ] ; then
            dpkg-statoverride --update --add sysuser root 4755 $i
        fi
    fi
done

The corresponding code to remove the override when the package is purged would be:

for i in /usr/bin/foo /usr/sbin/bar; do
    if dpkg-statoverride --list $i >/dev/null 2>&1; then
        dpkg-statoverride --remove $i
    fi
done

10.10. File names

The name of the files installed by binary packages in the system PATH (namely /bin, /sbin, /usr/bin, /usr/sbin and /usr/games) must be encoded in ASCII.

The name of the files and directories installed by binary packages outside the system PATH must be encoded in UTF-8 and should be restricted to ASCII when it is possible to do so.

1

If you are using GCC, -fPIC produces code with relocatable position independent code, which is required for most architectures to create a shared library, with i386 and perhaps some others where non position independent code is permitted in a shared library.

Position independent code may have a performance penalty, especially on i386. However, in most cases the speed penalty must be measured against the memory wasted on the few architectures where non position independent code is even possible.

2

Some of the reasons why this might be required is if the library contains hand crafted assembly code that is not relocatable, the speed penalty is excessive for compute intensive libs, and similar reasons.

3

Some of the reasons for linking static libraries with the -fPIC flag are if, for example, one needs a Perl API for a library that is under rapid development, and has an unstable API, so shared libraries are pointless at this phase of the library’s development. In that case, since Perl needs a library with relocatable code, it may make sense to create a static library with relocatable code. Another reason cited is if you are distilling various libraries into a common shared library, like mklibs does in the Debian installer project.

4

You might want to replace --strip-unneeded with --strip-debug for static libraries, as dh_strip does. When stripping static libraries, you should also pass --enable-deterministic-archives to ensure that your package build is reproducible.

5

A common example are the so-called “plug-ins”, internal shared objects that are dynamically loaded by programs using dlopen(3).

6

These files store, among other things, all libraries on which that shared library depends. Unfortunately, if the .la file is present and contains that dependency information, using libtool when linking against that library will cause the resulting program or library to be linked against those dependencies as well, even if this is unnecessary. This can create unneeded dependencies on shared library packages that would otherwise be hidden behind the library ABI, and can make library transitions to new SONAMEs unnecessarily complicated and difficult to manage.

7

The Open Group Base Specifications Issue 7, 2018 Edition, which is also known as POSIX.1-2017 and as IEEE Std 1003.1-2017 and is available on the World Wide Web from The Open Group.

8

These features are in widespread use in the Linux community and are implemented in all of bash, dash, and ksh, the most common shells users may wish to use as /bin/sh.

9

This is necessary to allow top-level directories to be symlinks. If linking /var/run to /run were done with the relative symbolic link ../run, but /var were a symbolic link to /srv/disk1, the symbolic link would point to /srv/run rather than the intended target.

10

It’s better to use mkfifo rather than mknod to create named pipes to avoid false positives from automated checks for packages incorrectly creating device files.

11

The dpkg-maintscript-helper tool, available from the dpkg package, can help for this task.

12

Rationale: There are two problems with hard links. The first is that some editors break the link while editing one of the files, so that the two files may unwittingly become unlinked and different. The second is that dpkg might break the hard link while upgrading conffiles.

13

The traditional approach to log files has been to set up ad hoc log rotation schemes using simple shell scripts and cron. While this approach is highly customizable, it requires quite a lot of sysadmin work. Even though the original Debian system helped a little by automatically installing a system which can be used as a template, this was deemed not enough.

The use of logrotate, a program developed by Red Hat, is better, as it centralizes log management. It has both a configuration file (/etc/logrotate.conf) and a directory where packages can drop their individual log rotation configurations (/etc/logrotate.d).

14

When a package is upgraded, and the owner or permissions of a file included in the package has changed, dpkg arranges for the ownership and permissions to be correctly set upon installation. However, this does not extend to directories; the permissions and ownership of directories already on the system does not change on install or upgrade of packages. This makes sense, since otherwise common directories like /usr would always be in flux. To correctly change permissions of a directory the package owns, explicit action is required, usually in the postinst script. Care must be taken to handle downgrades as well, in that case.

15

Ordinary files installed by dpkg (as opposed to conffiles and other similar objects) normally have their permissions reset to the distributed permissions when the package is reinstalled. However, the use of dpkg-statoverride overrides this default behavior.