path: root/ChangeLog

# ChangeLog for Path Sandbox
# Copyright 1999-2004 Gentoo Foundation; Distributed under the GPL v2
# $Header$

* sandbox-1.2.1

  23 Apr 2005; Martin Schlemmer <azarah@gentoo.org> Makefile.am, canonicalize.c,
  getcwd.c, libsandbox.c, localdecls.h, sandbox.h, sandbox_futils.c:
  Make sure all functions used in libsandbox.c is declared static.  Define
  SB_STATIC in localdecls.h for this.  Include sandbox_futils.c rather than
  linking with its object.  Hopefully this will fix bug #90153.

  22 Mar 2005; Martin Schlemmer <azarah@gentoo.org> libsandbox.c:
  Allow lchown a symlink in write-allowed path pointing to write-denied
  target.

  21 Mar 2005; Marius Mauch <genone@gentoo.org> libsandbox.c:
  Also show resolved symlink names in the log.

  14 Mar 2005; Martin Schlemmer <azarah@gentoo.org> Makefile.am, libsandbox.c:
  Seems -nostdlib was the problem with the constructor/destructor - remove it
  from Makefile.am, and change the constructor/destructor names again.

  14 Mar 2005; Martin Schlemmer <azarah@gentoo.org> libsandbox.c:
  Also rename the _init() and _fini() declarations.

  14 Mar 2005; Martin Schlemmer <azarah@gentoo.org> libsandbox.c, sandbox.c:
  Fixup the constructor/destructor function names again (they should be _init()
  and _fini() it seems, and not being called caused sandbox_lib_path to be
  unset, and thus breaking the execve() wrapper's LD_PRELOAD protection).
  Add both the path in given SANDBOX_x variable, as well as its symlink
  resolved path in init_env_entries().  Modify filter_path() to be able to
  resolve paths without resolving symlinks, as well as to be able to resolve
  symlinks.  Fix a possible segfault in check_access().  Add symlink resolving
  to check_access() resolving bug #31019.  Add 'hack' for unlink, as the fix
  for bug #31019 cause access violations if we try to remove a symlink that is
  not in protected path, but points to a protected path.  Fix a memory leak in
  sandbox.c (sandbox_pids_file in main()).  Fix the realpath() calls in main()
  (sandbox.c) being unchecked.  Fix the debug logname not having the pid in it
  (pid_string was uninitialized).  General syntax cleanups.

  09 Mar 2005; Brian Harring <ferringb@gentoo.org> sandbox.c: Fixed the 
  infamous "pids file is not a regular file" w/out newline bug.

  09 Mar 2005; Brian Harring <ferringb@gentoo.org> Makefile.am, configure.in:
  Correct libc_version path detection, since it was screwing up if libdir != 
  "/lib/".

  02 Mar 2005; Martin Schlemmer <azarah@gentoo.org> libsandbox.c:
  Hack to make sure sandboxed process cannot remove a device node, bug #79836.

  02 Mar 2005; Martin Schlemmer <azarah@gentoo.org> Makefile.am:
  Fix symbols.in not added to dist.

  02 Mar 2005; Martin Schlemmer <azarah@gentoo.org> Makefile.am, canonicalize.c,
  getcwd.c, libsandbox.c, sandbox.c, sandbox.h, sandbox_futils.c:
  White space fixes.

  02 Mar 2005; Martin Schlemmer <azarah@gentoo.org> Makefile.am, canonicalize.c,
  configure.in, getcwd.c, libsandbox.c, symbols.in:
  Fix inverse test logic in canonicalize.c, use a strncpy.  Fix gcc warning in
  getcwd.c.  Add symbols.in and logic to Makefile.am to generate symbol versions
  for glibc and other libc's that use this.  Update libsandbox.c to use these
  symbol versions if available.  Fix exec wrapper to re-export LD_PRELOAD if the
  process unset it.

  01 Mar 2005; Brian Harring <ferringb@gentoo.org> libsandbox.c: 
  killed off _init and _fini in favor of 
  void __attribute__ ((constructor)) init_func and
  void __attribute__ ((destructor)) closing_func.  _(init|func) were deprecated.

  06 Dec 2004; Brian Harring <ferringb@gentoo.org> Makefile.am, libsandbox.c,
  canonicalize.c, getcwd.c: Fixed compilation *again*.  Hopefully cvs is done
  having the hick-ups.

  04 Dec 2004; Brian Harring <ferringb@gentoo.org> libsandbox.c, getcwd.c,
  Makefile.am:  Fixed compilation.

  01 Dec 2004; Brian Harring <ferringb@gentoo.org> aclocal.m4:
  Gutted the bugger so it stops checking for a c++ and fortran compiler.
  Do *not* regenerate aclocal.m4 for making a release until a better 
  solution is created.

  20 Nov 2004; Brian Harring <ferringb@gentoo.org> Makefile.am, sandbox_futils.c:
  Removal of more hardcoded paths.

  20 Nov 2004; Brian Harring <ferringb@gentoo.org> Makefile.am, configure.in,
  sandbox_futils.c: tweaks to install sandbox.bashrc, and use it.

  19 Nov 2004; Brian Harring <ferringb@gentoo.org>:
  Sandbox is now autotooled, create-localdecls needs to be killed and the code
  shifted into configure.in.  Currently builds *one* libsandbox.so- if multiple
  are desired (-m64 and -m32 for amd64), the ebuild should do it (imo).
  To get to a point of testing, automake && autoconf; created requisite files w/
  a(utomake|clocal)-1.8, and autoconf 2.59.  Installs to /usr/, instead of 
  /lib and /usr/lib/portage/bin.

  14 Nov 2004; Brian Harring <ferringb@gentoo.org> libsandbox.c, sandbox.c:
  closing out bug #70225, potential overflow of the sandbox_pids_file var.

  07 Nov 2004; Brian Harring <ferringb@gentoo.org> libsandbox.c: c99 standard, 
  (think it was at least) allows intermixing of code and data segments.  bug #70351
  should be fixed by this.

  03 Nov 2004; Brian Harring <ferringb@gentoo.org> libsandbox.c, sandbox_futils.c:
  futils fix from bug #65201 via solar, and libsandbox log path checks via #69137

  02 Aug 2004; Nicholas Jones <carpaski@gentoo.org> libsandbox.c: Code from
  Seth Robertson that tracked down all adjuct flags for read operations that
  do not invoke a write operation.

  04 Apr 2004; Nicholas Jones <carpaski@gentoo.org> libsandbox.c, sandbox.c:
  Another fix from jstubbs regarding a free() on a stack variable for the
  environment -- tracking now prevents extraneous free()'s segfault.
  
  04 Apr 2004; Nicholas Jones <carpaski@gentoo.org> libsandbox.c, sandbox.c:
  J. Stubbs tracked down a new bug where mkdir was failing to the patch on
  the lstat in mkdir... it now only returns 0 or -1 as documented for mkdir.
  Also remove the errno = ESUCCESS settings as documentation points out that
  a library isn't allowed to do that.

  04 Apr 2004; Nicholas Jones <carpaski@gentoo.org> libsandbox.c: Added a
  file_security_check() function to check random potential exploits on files
  that sandbox is to load and read -- Normally sandboxpids.tmp. This fixes
  the 'system-crippling' exploits (bug 21923) and catches a few other
  potential problems.

  20 Mar 2004; Nicholas Jones <carpaski@gentoo.org> Makefile: Updates for
  32/64 bit sandbox. Made CC and LD '?=' values to allow passed in CC to work.

  20 Mar 2004; Nicholas Jones <carpaski@gentoo.org> libsandbox.c:
  bug 42048 -- Fixed the lstat/errno conditions for mkdir <caleb@g.o>.
  Added the 64/32 bit sandbox patch for AMD64 bug 32963 <brad/azarah>.

  29 Feb 2004; Martin Schlemmer <azarah@gentoo.org> sandbox.c, sandbox_futils.c :
  Fix permissions and group of pids file and logs.  Permissions should be 0664
  and group should be 'portage'.  Bug #34260.

  28 Feb 2004; Martin Schlemmer <azarah@gentoo.org> libsandbox.c :
  Besides a small cleanup, redo how we replace LD_PRELOAD in the environ passed
  to the real execve (in our execve wrapper).  Seems that on some arches (sparc
  among others) do not allow us to tamper with the readonly copy passed to
  execve, so pass our own copy of the environment.  Bug #42290.

  11 Jan 2004; Nicholas Jones <carpaski@gentoo.org> create-decls:
  Changed tail to head and added a notice about duration of glibc check.

  21 Dec 2003; Nicholas Jones <carpaski@gentoo.org> create-decls:
  Changed the glibc subversion check to use /usr/bin/* instead of /bin/sh
  as there isn't a guarentee that it is dynamic.

  02 Nov 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c :
  If 'file' passed to before_syscall(const char *func, const char *file) is
  invalid, we should set errno to ENOENT, and not EINVAL.  This should
  close bug #32238.

  14 Oct 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c :
  Fix a bug that occurs mainly on 64bit arch, where the file passed to
  the functions we wrap, is invalid, and then cause canonicalize to pass
  garbage to before_syscall(), thanks to great detective work from
  Andrea Luzzardi <al@sig11.org> (bug #29846).

  13 Oct 2003; Martin Schlemmer <azarah@gentoo.org> create-localdecls :
  Add a uClibc detection patch from Peter S. Mazinger <ps.m@gmx.net>.

  13 Oct 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c :
  Fix a bug in libsandbox.c 's checking in the rename wrapper - it basically
  only checked the destination patch, and not the source, so we could move
  a protected file to a unprotected directory, and then delete/modify it.
  Thanks to Andrea Luzzardi (scox) <al@sig11.org>, bug #30992, for this fix.

  12 Oct 2003; Nicholas Jones <carpaski@gentoo.org> sandbox.c :
  Added python2.3 to the predict section/variable.

  28 Sep 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c, sandbox.c,
  sandbox.h, sandbox_futils.c :
  Add support to set the pids file via SANDBOX_PIDS_FILE at startup.  If
  it is not set, it will revert to its old value.

  27 Sep 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c :
  Fix our mkdir wrapper to check if the dir exist, and return EEXIST if so,
  rather than failing with a violation, bug #29748.

  27 Jul 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c :
  Fix canonicalize() to ignore calls with path = "".

  27 Jul 2003; Martin Schlemmer <azarah@gentoo.org> getcwd.c, libsandbox.c,
  sandbox_futils.c, canonicalize.c :
  Once again coreutils fails, as my systems had 2.5 kernel, the getcwd system
  call handled strings larger than PATH_MAX (bug #21766).  It however does not
  work the same on 2.4 kernels.

  To fix, I added the posix implementation of getcwd() (from glibc cvs) that
  do not need the system call.  We use the default getcwd() function via a
  wrapper (egetcwd), and then lstat the returned path.  If lstat fails, it
  means the current directory was removed, OR that the the system call for
  getcwd failed (curious is that it do not fail and return NULL or set
  errno, but rather just truncate the retured directory - usually from the
  start), and if so, we use the generic getcwd() function (__egetcwd).  Note
  that we do not use the generic version all the time, as it calls lstat()
  a great number of times, and performance degrade much.

  29 Jun 2003; Martin Schlemmer <azarah@gentoo.org> create-localdecls,
  libsandbox.c :
  Make sure SB_PATH_MAX will not wrap.  Fix two possible memory leaks.

  22 Jun 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c, canonicalize.c
  create-localdecls :
  When checking path names of files accessed, we need to canonicalize it, else
  it may be a symlink in a 'write allowed' directory pointing to a file in a
  directory we should not have write access to.

  With something like coreutils-5.0, we have two problems:
  1)  One of the tests checks if getcwd() can return a path longer than
      PATH_MAX.  This test then tries to create a dir which even while
      created local (mkdir("conftest2")), it ends up being resolved with
      a name that is much larger than PATH_MAX.  The problem now is that
      canonicalize() have undefined behaviour when the path was too long
      (returned wrongly truncated paths, etc), and pass the wrong path to
      before_syscall() (causing the bogus sandbox violations).
  2)  The ecanonicalize() function we used, along with the canonicalize()
      function did not support longer than PATH_MAX.  This is partly a
      cause for 1), but the error checking (rather lack of it) of calls
      to erealpath() in canonicalize() was the prime reason for 1).

  As we do not use this canonicalized name to call the function, we resolve this
  by fixing canonicalize() to do better error checking, and ecanonicalize() as
  well as all functions in libsandbox.c to use a PATH_MAX of 'PATH_MAX * 2'.
  While they will resolve paths properly now, and can check if a write/read is
  allowed, the functions called from the sandboxed environment will still work
  as expected.

  This should resolve bug #21766.

  06 Apr 2003; Martin Schlemmer <azarah@gentoo.org> libsandbox.c :
  For some reason sandbox fails with a 'open_wr' if you run 'locale -a' under
  it (bug #16298).
  
  Problem is that for some reason locale fopen's locale.alias with mode "rm".

  -------------------------------------------------------
  nosferatu root # grep fopen locale.log 
  fopen("/usr/share/locale/locale.alias", "rm"ACCESS DENIED  open_wr:   /usr/share/locale/locale.alias
  nosferatu root # 
  --------------------------------------------------------

  I checked the source of locale, but it have fopen with mode 'r', so
  not sure where the "rm" mode comes from.  Anyhow, changed the check in
  before_syscall_open_char() to also see mode "rm" as readonly.

  23 Feb 2003; Martin Schlemmer <azarah@gentoo.org> create-localdecls :

  Add glibc-2.3 support.

  22 Feb 2003; Martin Schlemmer <azarah@gentoo.org> sandbox.c :

  Some /etc/ld.so.preload fixes.  Just changed the #if defines to cover all
  operations releated to preload, as well as only try to modify ld.so.preload
  if we can.  Also modify to write the pid to /tmp/sandboxpids.tmp even when
  not using ld.so.preload.  Fix to not write this instance of sandbox's pid
  to /tmp/sandboxpids.tmp on exit if this is not the last sandbox running.

  22 Feb 2003; Nicholas Jones <carpaski@gentoo.org> Makefile :
  
  Changed the LD to CC for hppa.

  22 Feb 2003; Nicholas Jones <carpaski@gentoo.org> create-localdecls :
  
  Killed the previous changes I made.

  17 Feb 2003; Nicholas Jones <carpaski@gentoo.org> create-localdecls :
  
  Added parisc to BROKEN_RTLD_ARCHLIST to see if it we can fix the relocation probs.

  09 Jan 2003; J Robert Ray <jrray@gentoo.org> sandbox.c :
  
  Don't segfault if $HOME isn't set, set $HOME to "/" instead.  Fixes bug 10868.

  16 Dec 2002; Martin Schlemmer <azarah@gentoo.org> create-localdecls :

  Fix memory leak for mips, bug #12236.  Thanks to Torgeir Hansen <torgeir@trenger.ro>
  for this fix.

  4 Dec 2002; J Robert Ray <jrray@gentoo.org> sandbox.h sandbox_futils.c :
  
  sandbox_futils defined a dirname() function that was masking the same
  function in glibc and was broken (e.g.: SANDBOX_DIR was being set to
  '/usr/lib/portage/bi/').  Fixed function to return expected results and
  renamed it to sb_dirname() to no longer mask the glibc function.  Closes bug
  11231.

  4 Dec 2002; Martin Schlemmer <azarah@gentoo.org> :

  Fix a segfault in libsandbox.c if canonicalize() was called with
  first parameter = NULL.

  1 Sep 2002; Martin Schlemmer <azarah@gentoo.org> :

  Fix my braindead 'return 1;' in a void function.  Updated sandbox.c,
  cleanup() for this.

  Change cleanup() in sandbox.c not to exit with fail status if
  the pidsfile is missing.  We really should still display sandbox
  violations if they occured.

  31 Aug 2002; Martin Schlemmer <azarah@gentoo.org> :

  Update cleanup() in sandbox.c to remove the PIDSFILE if this is
  the last sandbox running.

  25 Aug 2002; Martin Schlemmer <azarah@gentoo.org> :

  Major cleanups to mainly libsandbox.c again.

  22 Aug 2002; Martin Schlemmer <azarah@gentoo.org> :

  Add copyrights to sandbox.h and sandbox_futils.h.  If wrong, the
  parties involved should please contact me so that we can fix it.

  Add opendir wrapper to libsandbox.c.

  21 Aug 2002; Martin Schlemmer <azarah@gentoo.org> :

  Do some more cleanups to ecanonicalize(), as it dropped filenames in
  rare cases (after my symlink cleanups), and caused glibc to bork.
  These fixes went into canonicalize.c.

  20 Aug 2002; Martin Schlemmer <azarah@gentoo.org> :

  Fix spawn_shell() and main() in sandbox.c to properly return fail
  status.

  19 Aug 2002; Martin Schlemmer <azarah@gentoo.org> :

  The new canonicalize() function in libsandbox.c also resolved symlinks,
  which caused on cleaning sandbox errors if the symlink pointed to a
  file in the live root.   Ripped out canonicalize() and realpath() from
  glibc; removed the symlink stuff, and changed them to ecanonicalize()
  and erealpath().

  18 Aug 2002; Martin Schlemmer <azarah@gentoo.org> :

  Ripped out all the wrappers, and implemented those of InstallWatch.
  Losts of cleanups and bugfixes.  Implement a execve that forces
  $LIBSANDBOX in $LD_PRELOAD.  We can now thus do away with the feared
  /etc/ld.so.preload (*g*) ... Made the needed changes to sandbox.c,
  sandbox.h and sandbox_futils.c.  Rewrote the Makefile for most
  parts; it now have an install target.
  
  Reformat the whole thing to look somewhat like the reworked sandbox.c
  and new sandbox.h and sandbox_futils.c from:
  
    Brad House <brad@mainstreetsoftworks.com>.

  Additional Copyrights now due to the InstallWatch code:
  
    Copyright (C) 1998-9 Pancrazio `Ezio' de Mauro <p@demauro.net>