\input texinfo @c -*-texinfo-*-
@c %**start of header
@setfilename find-maint.info
@include versionmaint.texi
@settitle Maintaining GNU Findutils @value{VERSION}
@c For double-sided printing, uncomment:
@c @setchapternewpage odd
@c %**end of header
@iftex
@finalout
@end iftex
@dircategory GNU organization
@direntry
* Maintaining Findutils: (find-maint). Maintaining GNU findutils
@end direntry
@copying
This manual explains how GNU findutils is maintained, how changes should
be made and tested, and what resources exist to help developers.
This document corresponds to version @value{VERSION} of the GNU findutils.
Copyright @copyright{} 2007--2022 Free Software Foundation, Inc.
@quotation
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with no
Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts.
A copy of the license is included in the section entitled
``GNU Free Documentation License''.
@end quotation
@end copying
@titlepage
@title Maintaining GNU Findutils
@subtitle version @value{VERSION}, @value{UPDATED}
@author by James Youngman
@page
@vskip 0pt plus 1filll
@insertcopying
@end titlepage
@contents
@ifnottex
@node Top, Introduction, (dir), (dir)
@top Maintaining GNU Findutils
@insertcopying
@end ifnottex
@menu
* Introduction::
* Maintaining GNU Programs::
* Design Issues::
* Coding Conventions::
* Tools::
* Using the GNU Portability Library::
* Documentation::
* Testing::
* Bugs::
* Distributions::
* Internationalisation::
* Security::
* Making Releases::
* GNU Free Documentation License::
@end menu
@node Introduction
@chapter Introduction
This document explains how to contribute to and maintain GNU
Findutils. It concentrates on developer-specific issues. For
information about how to use the software please refer to
@xref{Introduction, ,Introduction,find,The Findutils manual}.
This manual aims to be useful without necessarily being verbose. It's
also a recent document, so there will be a many areas in which
improvements can be made. If you find that the document misses out
important information or any part of the document is be so terse as to
be unuseful, please ask for help on the @email{bug-findutils@@gnu.org}
mailing list. We'll try to improve this document too.
@node Maintaining GNU Programs
@chapter Maintaining GNU Programs
GNU Findutils is part of the GNU Project and so there are a number of
documents which set out standards for the maintenance of GNU
software.
@table @file
@item standards.texi
GNU Project Coding Standards. All changes to findutils should comply
with these standards. In some areas we go somewhat beyond the
requirements of the standards, but these cases are explained in this
manual.
@item maintain.texi
Information for Maintainers of GNU Software. This document provides
guidance for GNU maintainers. Everybody with commit access should
read this document. Everybody else is welcome to do so too, of
course.
@end table
@node Design Issues
@chapter Design Issues
The findutils package is installed on many many systems, usually as a
fundamental component. The programs in the package are often used in
order to successfully boot or fix the system.
This fact means that for findutils we bear in mind considerations that
may not apply so much as for other packages. For example, the fact
that findutils is often a base component motivates us to
@itemize
@item Limit dependencies on libraries
@item Avoid dependencies on other large packages (for example, interpreters)
@item Be conservative when making changes to the 'stable' release branch
@end itemize
All those considerations come before functionality. Functional
enhancements are still made to findutils, but these are almost
exclusively introduced in the 'development' release branch, to allow
extensive testing and proving.
Sometimes it is useful to have a priority list to provide guidance
when making design trade-offs. For findutils, that priority list is:
@enumerate
@item Correctness
@item Standards compliance
@item Security
@item Backward compatibility
@item Performance
@item Functionality
@end enumerate
For example, we support the @code{-exec} action because POSIX
compliance requires this, even though there are security problems with
it and we would otherwise prefer people to use @code{-execdir}. There
are also cases where some performance is sacrificed in the name of
security. For example, the sanity checks that @code{find} performs
while traversing a directory tree may slow it down. We adopt
functional changes, and functional changes are allowed to make
@code{find} slower, but only if there is no detectable impact on users
who don't use the feature.
Backward-incompatible changes do get made in order to comply with
standards (for example the behaviour of @code{-perm -...} changed in
order to comply with POSIX). However, they don't get made in order to
provide better ease of use; for example the semantics of @code{-size
-2G} are almost always unexpected by users, but we retain the current
behaviour because of backward compatibility and for its similarity to
the block-rounding behaviour of @code{-size -30}. We might introduce
a change which does not have the unfortunate rounding behaviour, but
we would choose another syntax (for example @code{-size '<2G'}) for
this.
In a general sense, we try to do test-driven development of the
findutils code; that is, we try to implement test cases for new
features and bug fixes before modifying the code to make the test
pass. Some features of the code are tested well, but the test
coverage for other features is less good. If you are about to modify
the code for a predicate and aren't sure about the test coverage, use
@code{grep} on the test directories and measure the coverage with
@code{lcov} or another test coverage tool.
You should be able to use the @code{coverage} Makefile target (it's
defined in @code{maint.mk} to generate a test coverage report for
findutils. Due to limitations in @code{lcov}, this only works if
your build directory is the same asthe source directory (that is,
you're not using a VPATH build configuration).
Lastly, we try not to depend on having a ``working system''. The
findutils suite is used for diagnosis of problems, and this applies
especially to @code{find}. We should ensure that @code{find} still
works on relatively broken systems, for example systems with damaged
@file{/etc/passwd} or @code{/etc/fstab} files. Another interesting
example is the case where a system is a client of one or more
unresponsive NFS servers. On such a system, if you try to stat all
mount points, your program will hang indefinitely, waiting for the
remote NFS server to respond.
Another interesting but unusual case is broken NFS servers and corrupt
filesystems; sometimes they return `impossible' file modes. It's
important that find does not entirely fail when encountering such a
file.
@node Coding Conventions
@chapter Coding Conventions
Coding style documents which set out to establish a uniform look and
feel to source code have worthy goals, for example greater ease of
maintenance and readability. However, I do not believe that in
general coding style guide authors can envisage every situation, and
it is always possible that it might on occasion be necessary to break
the letter of the style guide in order to honour its spirit, or to
better achieve the style guide's goals.
I've certainly seen many style guides outside the free software world
which make bald statements such as ``functions shall have exactly one
return statement''. The desire to ensure consistency and obviousness
of control flow is laudable, but it is all too common for such bald
requirements to be followed unthinkingly. Certainly I've seen such
coding standards result in unmaintainable code with terrible
infelicities such as functions containing @code{if} statements nested
nine levels deep. I suppose such coding standards don't survive in
free software projects because they tend to drive away potential
contributors or tend to generate heated discussions on mailing lists.
Equally, a nine-level-deep function in a free software program would
quickly get refactored, assuming it is obvious what the function is
supposed to do...
Be that as it may, the approach I will take for this document is to
explain some idioms and practices in use in the findutils source code,
and leave it up to the reader's engineering judgement to decide which
considerations apply to the code they are working on, and whether or
not there is sufficient reason to ignore the guidance in current
circumstances.
@menu
* Make the Compiler Find the Bugs::
* Factor Out Repeated Code::
* Debugging is For Users Too::
* Don't Trust the File System Contents::
* The File System Is Being Modified::
@end menu
@node Make the Compiler Find the Bugs
@section Make the Compiler Find the Bugs
Finding bugs is tedious. If I have a filesystem containing two
million files, and a find command line should print one million of
them, but in fact it misses out 1%, you can tell the program is
printing the wrong result only if you know the right answer for that
filesystem at that time. If you don't know this, you may just not
find out about that bug. For this reason it is important to have a
comprehensive test suite.
The test suite is of course not the only way to find the bugs. The
findutils source code makes liberal use of the assert macro. While on
the one hand these might be a performance drain, the performance
impact of most of these is negligible compared to the time taken to
fetch even one sector from a disk drive.
Assertions should not be used to check the results of operations which
may be affected by the program's external environment. For example,
never assert that a file could be opened successfully. Errors
relating to problems with the program's execution environment should
be diagnosed with a user-oriented error message. An assertion failure
should always denote a bug in the program.
Avoid using @code{assert} to mark not-fully-implemented features of
your code as such. Finish the implementation, disable the code, or
leave the unfinished version on a local branch.
Several programs in the findutils suite perform self-checks. See for
example the function @code{pred_sanity_check} in @file{find/pred.c}.
This is generally desirable.
There are also a number of small ways in which we can help the
compiler to find the bugs for us.
@subsection Constants in Equality Testing
It's a common error to write @code{=} when @code{==} is meant.
Sometimes this happens in new code and is simply due to finger
trouble. Sometimes it is the result of the inadvertent deletion of a
character. In any case, there is a subset of cases where we can
persuade the compiler to generate an error message when we make this
mistake; this is where the equality test is with a constant.
This is an example of a vulnerable piece of code.
@example
if (x == 2)
...
@end example
A simple typo converts the above into
@example
if (x = 2)
...
@end example
We've introduced a bug; the condition is always true, and the value of
@code{x} has been changed. However, a simple change to our practice
would have made us immune to this problem:
@example
if (2 == x)
...
@end example
Usually, the Emacs keystroke @kbd{M-t} can be used to swap the operands.
@subsection Spelling of ASCII NUL
Strings in C are just sequences of characters terminated by a NUL.
The ASCII NUL character has the numerical value zero. It is normally
represented in C code as @samp{\0}. Here is a typical piece of C
code:
@example
*p = '\0';
@end example
Consider what happens if there is an unfortunate typo:
@example
*p = '0';
@end example
We have changed the meaning of our program and the compiler cannot
diagnose this as an error. Our string is no longer terminated. Bad
things will probably happen. It would be better if the compiler could
help us diagnose this problem.
In C, the type of @code{'\0'} is in fact int, not char. This provides
us with a simple way to avoid this error. The constant @code{0} has
the same value and type as the constant @code{'\0'}. However, it is
not as vulnerable to typos. For this reason I normally prefer to
use this code:
@example
*p = 0;
@end example
@node Factor Out Repeated Code
@section Factor Out Repeated Code
Repeated code imposes a greater maintenance burden and increases the
exposure to bugs. For example, if you discover that something you
want to implement has some similarity with an existing piece of code,
don't cut and paste it. Instead, factor the code out. The risk of
cutting and pasting the code, particularly if you do this several
times, is that you end up with several copies of the same code.
If the original code had a bug, you now have N places where this needs
to be fixed. It's all to easy to miss some out when trying to fix the
bug. Equally, it's quite possible that when pasting the code into
some function, the pasted code was not quite adapted correctly to its
new environment. To pick a contrived example, perhaps it modifies a
global variable which it (that [original] code) shouldn't be touching
in its new home. Worse, perhaps it makes some unstated assumption about
the nature of the input arguments which is in fact not true for the
context of the now duplicated code.
A good example of the use of refactoring in findutils is the
@code{collect_arg} function in @file{find/parser.c}.
The findutils test suite is comprehensive enough that refactoring code
should not generally be a daunting prospect from a testing point of
view. Nevertheless there are some areas which are only
lightly-tested:
@enumerate
@item Tests on the ages of files
@item Code which deals with the values returned by operating system calls (for example handling of ENOENT)
@item Code dealing with OS limits (for example, limits on path length
or exec arguments)
@item Code relating to features not all systems have (for example
Solaris Doors)
@end enumerate
Please exercise caution when working in those areas.
@node Debugging is For Users Too
@section Debugging is For Users Too
Debug and diagnostic code is often used to verify that a program is
working in the way its author thinks it should be. But users are
often uncertain about what a program is doing, too. Exposing them a
little more diagnostic information can help. Much of the diagnostic
code in @code{find}, for example, is controlled by the @samp{-D} flag,
as opposed to C preprocessor directives.
Making diagnostic messages available to users also means that the
phrasing of the diagnostic messages becomes important, too.
@node Don't Trust the File System Contents
@section Don't Trust the File System Contents
People use @code{find} to search in directories created by other
people. Sometimes they do this to check to suspicious activity (for
example to look for new setuid binaries). This means that it would be
bad if @code{find} were vulnerable to, say, a security problem
exploitable by constructing a specially-crafted filename. The same
consideration would apply to @code{locate} and @code{updatedb}.
Henry Spencer said this well in his fifth commandment:
@quotation
Thou shalt check the array bounds of all strings (indeed, all arrays),
for surely where thou typest @samp{foo} someone someday shall type
@samp{supercalifragilisticexpialidocious}.
@end quotation
Symbolic links can often be a problem. If @code{find} calls
@code{lstat} on something and discovers that it is a directory, it's
normal for @code{find} to recurse into it. Even if the @code{chdir}
system call is used immediately, there is still a window of
opportunity between the @code{lstat} and the @code{chdir} in which a
malicious person could rename the directory and substitute a symbolic
link to some other directory.
@node The File System Is Being Modified
@section The File System Is Being Modified
The filesystem gets modified while you are traversing it. For,
example, it's normal for files to get deleted while @code{find} is
traversing a directory. Issuing an error message seems helpful when a
file is deleted from the one directory you are interested in, but if
@code{find} is searching 15000 directories, such a message becomes
less helpful.
Bear in mind also that it is possible for the directory @code{find} is
searching to be concurrently moved elsewhere in the file system,
and that the directory in which @code{find} was started could be
deleted.
Henry Spencer's sixth commandment is also apposite here:
@quotation
If a function be advertised to return an error code in the event of
difficulties, thou shalt check for that code, yea, even though the
checks triple the size of thy code and produce aches in thy typing
fingers, for if thou thinkest ``it cannot happen to me'', the gods
shall surely punish thee for thy arrogance.
@end quotation
There are a lot of files out there. They come in all dates and
sizes. There is a condition out there in the real world to exercise
every bit of the code base. So we try to test that code base before
someone falls over a bug.
@node Tools
@chapter Tools
Most of the tools required to build findutils are mentioned in the
file @file{README-hacking}. We also use some other tools:
@table @asis
@item System call traces
Much of the execution time of find is spent waiting for filesystem
operations. A system call trace (for example, that provided by
@code{strace}) shows what system calls are being made. Using this
information we can work to remove unnecessary file system operations.
@item Valgrind
Valgrind is a tool which dynamically verifies the memory accesses a
program makes to ensure that they are valid (for example, that the
behaviour of the program does not in any way depend on the contents of
uninitialized memory).
@item DejaGnu
DejaGnu is the test framework used to run the findutils test suite
(the @code{runtest} program is part of DejaGnu). It would be ideal if
everybody building @code{findutils} also ran the test suite, but many
people don't have DejaGnu installed. When changes are made to
findutils, DejaGnu is invoked a lot. @xref{Testing}, for more
information.
@end table
@node Using the GNU Portability Library
@chapter Using the GNU Portability Library
The Gnulib library (@url{https://www.gnu.org/software/gnulib/}) makes a
variety of systems look more like a GNU/Linux system and also applies
a bunch of automatic bug fixes and workarounds. Some of these also
apply to GNU/Linux systems too. For example, the Gnulib regex
implementation is used when we determine that we are building on a
GNU libc system with a bug in the regex implementation.
@section How and Why we Import the Gnulib Code
Gnulib does not have a release process which results in a source
tarball you can download. Instead, the code is simply made available
by GIT, so we import gnulib via the submodule feature. The bootstrap
script performs the necessary steps.
Findutils does not use all the Gnulib code. The modules we need are
listed in the file @file{bootstrap.conf}.
The upshot of all this is that we can use the findutils git repository
to track which version of Gnulib every findutils release uses.
A small number of files are installed by automake and will therefore
vary according to which version of automake was used to generate a
release. This includes for example boiler-plate GNU files such as
@file{ABOUT-NLS}, @file{INSTALL} and @file{COPYING}.
@section How We Fix Gnulib Bugs
Gnulib is used by quite a number of GNU projects, and this means that
it gets plenty of testing. Therefore there are relatively few bugs in
the Gnulib code, but it does happen from time to time.
However, since there is no waiting around for a Gnulib source release
tarball, Gnulib bugs are generally fixed quickly. Here is an outline
of the way we would contribute a fix to Gnulib (assuming you know it
is not already fixed in the current Gnulib git tree):
@table @asis
@item Check you already completed a copyright assignment for Gnulib
@item Begin with a vanilla git tree
Download the Findutils source code from git (or use the tree you have
already)
@item Run the bootstrap script
@item Run configure
@item Build findutils
Build findutils and run the test suite, which should pass. In our
example we assume you have just noticed a bug in Gnulib, not that
recent Gnulib changes broke the findutils regression tests.
@item Write a test case
If in fact Gnulib did break the findutils regression tests, you can probably
skip this step, since you already have a test case demonstrating the problem.
Otherwise, write a findutils test case for the bug and/or a Gnulib test case.
@item Fix the Gnulib bug
Make sure your editor follows symbolic links so that your changes to
@file{gnulib/...} actually affect the files in the git working
directory you checked out earlier. Observe that your test now passes.
@item Prepare a Gnulib patch
In the gnulib subdirectory, use @code{git format-patch} to prepare the
patch. Follow the normal usage for checkin comments (take a look at
the output of @code{git log}). Check that the patch conforms with the
GNU coding standards, and email it to the Gnulib mailing list.
@item Wait for the patch to be applied
Once your bug fix has been applied, you can update your gnulib
directory from git, and then check in the change to the submodule as
normal (you can check @code{git help submodule} for details).
@end table
There is an alternative to the method above; it is possible to store
local diffs to be patched into gnulib beneath the
@file{gnulib-local}. Normally however, there is no need for this,
since gnulib updates are very prompt.
@section How to update Gnulib to latest
With a non-dirty working tree, the command @code{make update-gnulib-to-latest}
(or the shorter alias @code{make gnulib-sync} allows, well, to update the
gnulib submodule. In detail, that is:
@enumerate
@item Fetching the latest upstream gnulib reference.
@item Copying the files which should stay in sync like
@file{bootstrap} from gnulib into the findutils working tree.
@item And finally showing the @code{git status} for the gnulib submodule
and the above copied files.
@end enumerate
After that, the maintainer compares if all is correct, if the findutils build
and run correct, and finally commits with the new gnulib version, e.g. via
@code{git gui}.
The @code{gnulib-sync} target can be run any time - after a @code{configure}
run -, and only rejects to run if the working tree is dirty.
@node Documentation
@chapter Documentation
The findutils git tree includes several different types of
documentation.
@section git change log
The git change log for the source tree contains check-in messages
which describe each check-in. These have a standard format:
@smallexample
Summary of the change.
(ChangeLog-style detail)
@end smallexample
Here, the format of the detail part follows the standard GNU ChangeLog
style, but without whitespace in the left margin and without
author/date headers. Take a look at the output of @code{git log} to
see some examples. The README-hacking file also contains an example
with an explanation.
@section User Documentation
User-oriented documentation is provided as manual pages and in
Texinfo. See
@ref{Introduction,,Introduction,find,The Findutils manual}.
Please make sure both sets of documentation are updated if you make a
change to the code. The GNU coding standards do not normally call for
maintaining manual pages on the grounds of effort duplication.
However, the manual page format is more convenient for quick
reference, and so it's worth maintaining both types of documentation.
However, the manual pages are normally rather more terse than the
Texinfo documentation. The manual pages are suitable for reference
use, but the Texinfo manual should also include introductory and
tutorial material.
We make the user documentation available on the web, on the GNU
project web site. These web pages are source-controlled via CVS
(still!). If you are a member of the @samp{findutils} project on
Savannah you should be able to check the web pages out like this
(@samp{$USER} is a placeholder for your Savannah username):
@smallexample
cvs -d :ext:$USER@@cvs.savannah.gnu.org:/web/findutils checkout findutils/manual
@end smallexample
You can automatically update the documentation in this repository
by using the script @samp{build-aux/update-online-manual.sh} with the
path to the findutils Git repository as parameter.
@smallexample
build-aux/update-online-manual.sh $HOME/git/findutils
@end smallexample
That script will generate the documentation in the directory
@samp{doc/manual/}
by calling the @code{make} target @samp{web-manual};
then it will copy over the files into the CVS checkout.
There you can check the documentation once again before committing
to CVS. The Savannah CVS server will automatically initiate the
transfer to the web server.
@section Build Guidance
@table @file
@item ABOUT-NLS
Describes the Free Translation Project, the translation status of
various GNU projects, and how to participate by translating an
application.
@item AUTHORS
Lists the authors of findutils.
@item COPYING
The copyright license covering findutils; currently, the GNU GPL,
version 3.
@item INSTALL
Generic installation instructions for installing GNU programs.
@item README
Information about how to compile findutils in particular
@item README-hacking
Describes how to build findutils from the code in git.
@item THANKS
Thanks for people who contributed to findutils. Generally, if
someone's contribution was significant enough to need a copyright
assignment, their name should go in here.
@item TODO
Mainly obsolete. Please add bugs to the Savannah bug tracker instead
of adding entries to this file.
@end table
@section Release Information
@table @file
@item NEWS
Enumerates the user-visible change in each release. Typical changes
are fixed bugs, functionality changes and documentation changes.
Include the date when a release is made.
@item ChangeLog
This file enumerates all changes to the findutils source code (with
the possible exception of @file{.cvsignore} and @code{.gitignore}
changes). The level of detail used for this file should be sufficient
to answer the questions ``what changed?'' and ``why was it changed?''.
The file is generated from the git commit messages during @code{make dist}.
If a change fixes a bug, always give the bug reference number in the
@file{NEWS} file and of course also in the checkin message.
In general, it should be possible to enumerate all
material changes to a function by searching for its name in
@file{ChangeLog}. Mention when each release is made.
@end table
@node Testing
@chapter Testing
This chapter will explain the general procedures for adding tests to
the test suite, and the functions defined in the findutils-specific
DejaGnu configuration. Where appropriate references will be made to
the DejaGnu documentation.
@node Bugs
@chapter Bugs
Bugs are logged in the Savannah bug tracker
@url{https://savannah.gnu.org/bugs/?group=findutils}. The tracker
offers several fields but their use is largely obvious. The
life-cycle of a bug is like this:
@table @asis
@item Open
Someone, usually a maintainer, a distribution maintainer or a user,
creates a bug by filling in the form. They fill in field values as
they see fit. This will generate an email to
@email{bug-findutils@@gnu.org}.
@item Triage
The bug hangs around with @samp{Status=None} until someone begins to
work on it. At that point they set the ``Assigned To'' field and will
sometimes set the status to @samp{In Progress}, especially if the bug
will take a while to fix.
@item Non-bugs
Quite a lot of reports are not actually bugs; for these the usual
procedure is to explain why the problem is not a bug, set the status
to @samp{Invalid} and close the bug. Make sure you set the
@samp{Assigned to} field to yourself before closing the bug.
@item Fixing
When you commit a bug fix into git (or in the case of a contributed
patch, commit the change), mark the bug as @samp{Fixed}. Make sure
you include a new test case where this is relevant. If you can figure
out which releases are affected, please also set the @samp{Release}
field to the earliest release which is affected by the bug.
Indicate which source branch the fix is included in (for example,
4.2.x or 4.3.x). Don't close the bug yet.
@item Release
When a release is made which includes the bug fix, make sure the bug
is listed in the NEWS file. Once the release is made, fill in the
@samp{Fixed Release} field and close the bug.
@end table
@node Distributions
@chapter Distributions
Almost all GNU/Linux distributions include findutils, but only some of
them have a package maintainer who is a member of the mailing list.
Distributions don't often feed back patches to the
@email{bug-findutils@@gnu.org} list, but on the other hand many of
their patches relate only to standards for file locations and so
forth, and are therefore distribution specific. On an irregular basis
I check the current patches being used by one or two distributions,
but the total number of GNU/Linux distributions is large enough that
we could not hope to cover them all.
Often, bugs are raised against a distribution's bug tracker instead of
GNU's. Periodically (about every six months) I take a look at some
of the more accessible bug trackers to indicate which bugs have been
fixed upstream.
Many distributions include both findutils and the slocate package,
which provides a replacement @code{locate}.
@node Internationalisation
@chapter Internationalisation
Translation is essentially automated from the maintainer's point of
view. The TP mails the maintainer when a new PO file is available,
and we just download it and check it in. The @file{bootstrap} script
copies @file{.po} files into the working tree. For more information,
please see
@url{https://translationproject.org/domain/findutils.html}.
@node Security
@chapter Security
See @ref{Security Considerations, ,Security Considerations,find,The
Findutils manual}, for a full description of the findutils approach to
security considerations and discussion of particular tools.
If someone reports a security bug publicly, we should fix this as
rapidly as possible. If necessary, this can mean issuing a fixed
release containing just the one bug fix. We try to avoid issuing
releases which include both significant security fixes and functional
changes.
Where someone reports a security problem privately, we generally try
to construct and test a patch without pushing the intermediate code to
the public repository.
Once everything has been tested, this allows us to make a release and
push the patch. The advantage of doing things this way is that we
avoid situations where people watching for git commits can figure out
and exploit a security problem before a fixed release is available.
It's important that security problems be fixed promptly, but don't
rush so much that things go wrong. Make sure the new release really
fixes the problem. It's usually best not to include functional
changes in your security-fix release.
If the security problem is serious, send an alert to
@email{vendor-sec@@lst.de}. The members of the list include most
GNU/Linux distributions. The point of doing this is to allow them to
prepare to release your security fix to their customers, once the fix
becomes available. Here is an example alert:-
@smallexample
GNU findutils heap buffer overrun (potential privilege escalation)
I. BACKGROUND
=============
GNU findutils is a set of programs which search for files on Unix-like
systems. It is maintained by the GNU Project of the Free Software
Foundation. For more information, see
@url{https://www.gnu.org/software/findutils}.
II. DESCRIPTION
===============
When GNU locate reads filenames from an old-format locate database,
they are read into a fixed-length buffer allocated on the heap.
Filenames longer than the 1026-byte buffer can cause a buffer overrun.
The overrunning data can be chosen by any person able to control the
names of filenames created on the local system. This will normally
include all local users, but in many cases also remote users (for
example in the case of FTP servers allowing uploads).
III. ANALYSIS
=============
Findutils supports three different formats of locate database, its
native format "LOCATE02", the slocate variant of LOCATE02, and a
traditional ("old") format that locate uses on other Unix systems.
When locate reads filenames from a LOCATE02 database (the default
format), the buffer into which data is read is automatically extended
to accommodate the length of the filenames.
This automatic buffer extension does not happen for old-format
databases. Instead a 1026-byte buffer is used. When a longer
pathname appears in the locate database, the end of this buffer is
overrun. The buffer is allocated on the heap (not the stack).
If the locate database is in the default LOCATE02 format, the locate
program does perform automatic buffer extension, and the program is
not vulnerable to this problem. The software used to build the
old-format locate database is not itself vulnerable to the same
attack.
Most installations of GNU findutils do not use the old database
format, and so will not be vulnerable.
IV. DETECTION
=============
Software
--------
All existing releases of findutils are affected.
Installations
-------------
To discover the longest path name on a given system, you can use the
following command (requires GNU findutils and GNU coreutils):
@verbatim
find / -print0 | tr -c '\0' 'x' | tr '\0' '\n' | wc -L
@end verbatim
V. EXAMPLE
==========
This section includes a shell script which determines which of a list
of locate binaries is vulnerable to the problem. The shell script has
been tested only on glibc based systems having a mktemp binary.
NOTE: This script deliberately overruns the buffer in order to
determine if a binary is affected. Therefore running it on your
system may have undesirable effects. We recommend that you read the
script before running it.
@verbatim
#! /bin/sh
set +m
if vanilla_db="$(mktemp nicedb.XXXXXX)" ; then
if updatedb --prunepaths="" --old-format --localpaths="/tmp" \
--output="$@{vanilla_db@}" ; then
true
else
rm -f "$@{vanilla_db@}"
vanilla_db=""
echo "Failed to create old-format locate database; skipping the sanity checks" >&2
fi
fi
make_overrun_db() @{
# Start with a valid database
cat "$@{vanilla_db@}"
# Make the final entry really long
dd if=/dev/zero bs=1 count=1500 2>/dev/null | tr '\000' 'x'
@}
ulimit -c 0
usage() @{ echo "usage: $0 binary [binary...]" >&2; exit $1; @}
[ $# -eq 0 ] && usage 1
bad=""
good=""
ugly=""
if dbfile="$(mktemp nasty.XXXXXX)"
then
make_overrun_db > "$dbfile"
for locate ; do
ver="$locate = $("$locate" --version | head -1)"
if [ -z "$vanilla_db" ] || "$locate" -d "$vanilla_db" "" >/dev/null ; then
"$locate" -d "$dbfile" "" >/dev/null
if [ $? -gt 128 ] ; then
bad="$bad
vulnerable: $ver"
else
good="$good
good: $ver"
fi
else
# the regular locate failed
ugly="$ugly
buggy, may or may not be vulnerable: $ver"
fi
done
rm -f "$@{dbfile@}" "$@{vanilla_db@}"
# good: unaffected. bad: affected (vulnerable).
# ugly: doesn't even work for a normal old-format database.
echo "$good"
echo "$bad"
echo "$ugly"
else
exit 1
fi
@end verbatim
VI. VENDOR RESPONSE
===================
The GNU project discovered the problem while 'locate' was being worked
on; this is the first public announcement of the problem.
The GNU findutils mantainer has issued a patch as p[art of this
announcement. The patch appears below.
A source release of findutils-4.2.31 will be issued on 2007-05-30.
That release will of course include the patch. The patch will be
committed to the public CVS repository at the same time. Public
announcements of the release, including a description of the bug, will
be made at the same time as the release.
A release of findutils-4.3.x will follow and will also include the
patch.
VII. PATCH
==========
This patch should apply to findutils-4.2.23 and later.
Findutils-4.2.23 was released almost two years ago.
@verbatim
Index: locate/locate.c
===================================================================
RCS file: /cvsroot/findutils/findutils/locate/locate.c,v
retrieving revision 1.58.2.2
diff -u -p -r1.58.2.2 locate.c
--- locate/locate.c 22 Apr 2007 16:57:42 -0000 1.58.2.2
+++ locate/locate.c 28 May 2007 10:18:16 -0000
@@@@ -124,9 +124,9 @@@@ extern int errno;
#include "locatedb.h"
#include <getline.h>
-#include "../gnulib/lib/xalloc.h"
-#include "../gnulib/lib/error.h"
-#include "../gnulib/lib/human.h"
+#include "xalloc.h"
+#include "error.h"
+#include "human.h"
#include "dirname.h"
#include "closeout.h"
#include "nextelem.h"
@@@@ -468,10 +468,36 @@@@ visit_justprint_unquoted(struct process_
return VISIT_CONTINUE;
@}
+static void
+toolong (struct process_data *procdata)
+@{
+ error (EXIT_FAILURE, 0,
+ _("locate database %s contains a "
+ "filename longer than locate can handle"),
+ procdata->dbfile);
+@}
+
+static void
+extend (struct process_data *procdata, size_t siz1, size_t siz2)
+@{
+ /* Figure out if the addition operation is safe before performing it. */
+ if (SIZE_MAX - siz1 < siz2)
+ @{
+ toolong (procdata);
+ @}
+ else if (procdata->pathsize < (siz1+siz2))
+ @{
+ procdata->pathsize = siz1+siz2;
+ procdata->original_filename = x2nrealloc (procdata->original_filename,
+ &procdata->pathsize,
+ 1);
+ @}
+@}
+
static int
visit_old_format(struct process_data *procdata, void *context)
@{
- register char *s;
+ register size_t i;
(void) context;
/* Get the offset in the path where this path info starts. */
@@@@ -479,20 +505,35 @@@@ visit_old_format(struct process_data *pr
procdata->count += getw (procdata->fp) - LOCATEDB_OLD_OFFSET;
else
procdata->count += procdata->c - LOCATEDB_OLD_OFFSET;
+ assert(procdata->count > 0);
- /* Overlay the old path with the remainder of the new. */
- for (s = procdata->original_filename + procdata->count;
+ /* Overlay the old path with the remainder of the new. Read
+ * more data until we get to the next filename.
+ */
+ for (i=procdata->count;
(procdata->c = getc (procdata->fp)) > LOCATEDB_OLD_ESCAPE;)
- if (procdata->c < 0200)
- *s++ = procdata->c; /* An ordinary character. */
- else
- @{
- /* Bigram markers have the high bit set. */
- procdata->c &= 0177;
- *s++ = procdata->bigram1[procdata->c];
- *s++ = procdata->bigram2[procdata->c];
- @}
- *s-- = '\0';
+ @{
+ if (procdata->c < 0200)
+ @{
+ /* An ordinary character. */
+ extend (procdata, i, 1u);
+ procdata->original_filename[i++] = procdata->c;
+ @}
+ else
+ @{
+ /* Bigram markers have the high bit set. */
+ extend (procdata, i, 2u);
+ procdata->c &= 0177;
+ procdata->original_filename[i++] = procdata->bigram1[procdata->c];
+ procdata->original_filename[i++] = procdata->bigram2[procdata->c];
+ @}
+ @}
+
+ /* Consider the case where we executed the loop body zero times; we
+ * still need space for the terminating null byte.
+ */
+ extend (procdata, i, 1u);
+ procdata->original_filename[i] = 0;
procdata->munged_filename = procdata->original_filename;
@end verbatim
VIII. THANKS
============
Thanks to Rob Holland <rob@@inversepath.com> and Tavis Ormandy.
VIII. CVE INFORMATION
=====================
No CVE candidate number has yet been assigned for this vulnerability.
If someone provides one, I will include it in the public announcement
and change logs.
@end smallexample
The original announcement above was sent out with a cleartext PGP
signature, of course, but that has been omitted from the example.
Once a fixed release is available, announce the new release using the
normal channels. Any CVE number assigned for the problem should be
included in the @file{ChangeLog} and @file{NEWS} entries. See
@url{https://cve.mitre.org/} for an explanation of CVE numbers.
@node Making Releases
@chapter Making Releases
This section will explain how to make a findutils release. For the
time being here is a terse description of the main steps:
@set RELEASE X.Y.Z
@set RELTAG v@value{RELEASE}
@enumerate
@item Commit changes; make sure your working directory has no
uncommitted changes.
@item Update translation files; re-run bootstrap to download the
newest @samp{.po} files.
@item Make sure compiler warnings would block the release; re-run
@samp{configure} with the options
@code{--enable-compiler-warnings --enable-compiler-warnings-are-errors}.
@item Test; make sure that all changes you have made have tests, and
that the tests pass.
Verify this with @code{env RUN_EXPENSIVE_TESTS=yes make distcheck}.
@c The RUN_EXPENSIVE_TESTS environment variable is checked in init.cfg.
@item Bugs; make sure all Savannah bug entries fixed in this release
are marked as fixed in Savannah. Optionally close them too to save
duplicate work (otherwise, close them after the release is uploaded).
@item Add new release in Savannah field values; see the @code{Bugs >
Edit Field Values} menu item. Add a field value for the release you
are about to make so that users can report bugs in it.
@item Update version; make sure that the NEWS file
is updated with the new release number (and checked in).
@c There is no longer any need to update configure.ac, since it no
@c longer contains version information.
@item Tag the release; findutils releases are tagged like this for
example: v4.5.5. You can create a tag with the a command like this:
@c we use @example here because @value will not work within @code or @samp.
@example
git tag -s -m "Findutils release @value{RELEASE}" @value{RELTAG}
@end example
@noindent
@item Build the release tarball; do this with @code{make distcheck}.
Copy the tarball somewhere safe.
@item Merge; if the release (and signed tag) were made on a
local branch, merge the branch to your local master.
@item Push; push your master to origin/master.
@item Push the new release tag; assuming that the name of your remote is
@samp{origin}, this is:
@example
git push origin tag @value{RELTAG}
@end example
@item Prepare the upload and upload it.
You can do this with
@c we use @example here because @value will not work within @code or @samp.
@example
build-aux/gnupload --to ftp.gnu.org:findutils findutils-@value{RELEASE}.tar.xz
@end example
@noindent
Use @code{alpha.gnu.org:findutils} for an alpha or beta release.
@xref{Automated FTP Uploads, ,Automated FTP
Uploads, maintain, Information for Maintainers of GNU Software},
for detailed upload instructions.
@item Check the FTP upload worked; you can look for an email from the
robot or check the contents of the actual FTP site.
@item Make a release announcement; include an extract from the NEWS
file which explains what's changed. Announcements for test releases
should just go to @email{bug-findutils@@gnu.org}. Announcements for
stable releases should go to @email{info-gnu@@gnu.org} as well.
@item Post-release administrativa: add a new dummy release header in NEWS:
@code{* Major changes in release ?.?.?, YYYY-MM-DD}
and update the @code{old_NEWS_hash} in @file{cfg.mk} with
@code{make update-NEWS-hash}.
Commit both changes.
@c make update-NEWS-hash supports make news-check but we normally
@c don't do that (and I'm not sure that the current NEWS file would
@c pass the check anyway).
@item Close bugs; any bugs recorded on Savannah which were fixed in this
release should now be marked as closed if there were not already.
Update the @samp{Fixed Release} field of these bugs appropriately and
make sure the @samp{Assigned to} field is populated.
@end enumerate
@node GNU Free Documentation License
@appendix GNU Free Documentation License
@include fdl.texi
@bye
@comment texi related words used by Emacs' spell checker ispell.el
@comment LocalWords: texinfo setfilename settitle setchapternewpage
@comment LocalWords: iftex finalout ifinfo DIR titlepage vskip pt
@comment LocalWords: filll dir samp dfn noindent xref pxref
@comment LocalWords: var deffn texi deffnx itemx emph asis
@comment LocalWords: findex smallexample subsubsection cindex
@comment LocalWords: dircategory direntry itemize
@comment other words used by Emacs' spell checker ispell.el
@comment LocalWords: README fred updatedb xargs Plett Rendell akefile
@comment LocalWords: args grep Filesystems fo foo fOo wildcards iname
@comment LocalWords: ipath regex iregex expr fubar regexps
@comment LocalWords: metacharacters macs sr sc inode lname ilname
@comment LocalWords: sysdep noleaf ls inum xdev filesystems usr atime
@comment LocalWords: ctime mtime amin cmin mmin al daystart Sladkey rm
@comment LocalWords: anewer cnewer bckw rf xtype uname gname uid gid
@comment LocalWords: nouser nogroup chown chgrp perm ch maxdepth
@comment LocalWords: mindepth cpio src CD AFS statted stat fstype ufs
@comment LocalWords: nfs tmp mfs printf fprint dils rw djm Nov lwall
@comment LocalWords: POSIXLY fls fprintf strftime locale's EDT GMT AP
@comment LocalWords: EST diff perl backquotes sprintf Falstad Oct cron
@comment LocalWords: eg vmunix mkdir afs allexec allwrite ARG bigram
@comment LocalWords: bigrams cd chmod comp crc CVS dbfile eof
@comment LocalWords: fileserver filesystem fn frcode Ghazi Hnewc iXX
@comment LocalWords: joeuser Kaveh localpaths localuser LOGNAME
@comment LocalWords: Meyering mv netpaths netuser nonblank nonblanks
@comment LocalWords: ois ok Pinard printindex proc procs prunefs
@comment LocalWords: prunepaths pwd RFS rmadillo rmdir rsh sbins str
@comment LocalWords: su Timar ubins ug unstripped vf VM Weitzel
@comment LocalWords: wildcard zlogout basename execdir wholename iwholename
@comment LocalWords: timestamp timestamps Solaris FreeBSD OpenBSD POSIX