Feature #10351
[PATCH] prevent CVE-2014-6277
Added by shyouhei (Shyouhei Urabe) over 6 years ago. Updated about 3 years ago.
Description
From 4636ca0308f1933c9b191f36e808a8d3bcf5e88e Mon Sep 17 00:00:00 2001 From: "Urabe, Shyouhei" <shyouhei@ruby-lang.org> Date: Wed, 8 Oct 2014 15:44:27 +0900 Subject: [PATCH] prevent CVE-2014-6277 ShellShock was about bash. I think ruby is torellant for that kind of attack. However the concept was that environment variables can potentially contain malicious data. To pass them to victim subprocecsses can sometimes cause catastrophic situation, like arbitrary code execution. Even though ruby itself is not affected, it can be an accomplice in attacks by blindly passing through any uncontrolled info through. Let's just change this, more secure by default. This patch does not add a new feature, nor delete anything. It just changes the default behaviour when ruby spawns subprocesses. Process.spawn('/usr/bin/printenv') # -> prints nothing You can explicitly pass what you need: Process.spawn({'FOO'=>'BAR'}, '/usr/bin/printenv') Or if you are 128% sure what you are doing, can pass everything. ENV['FOO'] = 'BAR' Process.spawn('/usr/bin/printenv', unsetenv_others: false) I don't intend to make things impossible; just give it a better default. It doesn't ultimately solve everything but it should prevent casual faults. --- process.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/process.c b/process.c index f9bc01c..395333d 100644 --- a/process.c +++ b/process.c @@ -2295,7 +2295,7 @@ rb_execarg_fixup(VALUE execarg_obj) eargp->dup2_tmpbuf = tmpbuf; } - unsetenv_others = eargp->unsetenv_others_given && eargp->unsetenv_others_do; + unsetenv_others = !eargp->unsetenv_others_given || eargp->unsetenv_others_do; envopts = eargp->env_modification; if (unsetenv_others || envopts != Qfalse) { VALUE envtbl, envp_str, envp_buf; @@ -2936,7 +2936,7 @@ rb_execarg_run_options(const struct rb_execarg *eargp, struct rb_execarg *sargp, #endif #if !defined(HAVE_WORKING_FORK) - if (eargp->unsetenv_others_given && eargp->unsetenv_others_do) { + if (!eargp->unsetenv_others_given || eargp->unsetenv_others_do) { save_env(sargp); rb_env_clear(); } @@ -3998,8 +3998,8 @@ rb_f_system(int argc, VALUE *argv) * [cmdname, argv0], arg1, ... : command name, argv[0] and zero or more arguments (no shell) * options: hash * clearing environment variables: - * :unsetenv_others => true : clear environment variables except specified by env - * :unsetenv_others => false : don't clear (default) + * :unsetenv_others => true : clear environment variables except specified by env (default) + * :unsetenv_others => false : don't clear * process group: * :pgroup => true or 0 : make a new process group * :pgroup => pgid : join to specified process group -- 1.9.1
Updated by akr (Akira Tanaka) over 6 years ago
It is very aggressive.
At least, you should try "make check".
% make check ... ./miniruby -I./lib -I. -I.ext/common ./tool/runruby.rb --extout=.ext -- --disable-gems "./test/runner.rb" --ruby="./miniruby -I./lib -I. -I.ext/common ./tool/runruby.rb --extout=.ext -- --disable-gems" mkmf.rb can't find header files for ruby at /home/ruby/tst1/lib/ruby/include/ruby.h uncommon.mk:549: recipe for target 'yes-test-all' failed make: *** [yes-test-all] Error 1
Updated by normalperson (Eric Wong) over 6 years ago
shyouhei@ruby-lang.org wrote:
This patch does not add a new feature, nor delete anything. It just
changes the default behaviour when ruby spawns subprocesses.Process.spawn('/usr/bin/printenv') # -> prints nothing
The potential for breakage is way too high. Losing some envs (e.g.
PATH, TMPDIR, SHELL or HOME) can be disastrous and introduce new
security problems.
Right now, everybody knows about shellshock and patching bash.
This is an over-reaction which causes needless breakage.
(Especially since your example never even spawns a shell)
Updated by akr (Akira Tanaka) over 6 years ago
- Status changed from Open to Feedback
Updated by shyouhei (Shyouhei Urabe) over 6 years ago
Akira, isn't it obvious this patch introduce a backward incompatibility?
I'd be surprised if it could passed the tests, because that means our tests are insufficient. Is it that important to pass make check at this point?
I don't think it's an "at least" thing for this rather conceptual patch. Is it your only concern?
Of course tests shall be modified to fit this when it is going to land.
On 10/10/2014 02:02 AM, akr@fsij.org wrote:
Issue #10351 has been updated by Akira Tanaka.
It is very aggressive.
At least, you should try "make check".
% make check ... ./miniruby -I./lib -I. -I.ext/common ./tool/runruby.rb --extout=.ext -- --disable-gems "./test/runner.rb" --ruby="./miniruby -I./lib -I. -I.ext/common ./tool/runruby.rb --extout=.ext -- --disable-gems" mkmf.rb can't find header files for ruby at /home/ruby/tst1/lib/ruby/include/ruby.h uncommon.mk:549: recipe for target 'yes-test-all' failed make: *** [yes-test-all] Error 1
Feature #10351: [PATCH] prevent CVE-2014-6277
https://bugs.ruby-lang.org/issues/10351#change-49322
- Author: Shyouhei Urabe
- Status: Open
- Priority: Normal
- Assignee:
- Category: core
* Target version: current: 2.2.0¶
>From 4636ca0308f1933c9b191f36e808a8d3bcf5e88e Mon Sep 17 00:00:00 2001 From: "Urabe, Shyouhei" <shyouhei@ruby-lang.org> Date: Wed, 8 Oct 2014 15:44:27 +0900 Subject: [PATCH] prevent CVE-2014-6277 ShellShock was about bash. I think ruby is torellant for that kind of attack. However the concept was that environment variables can potentially contain malicious data. To pass them to victim subprocecsses can sometimes cause catastrophic situation, like arbitrary code execution. Even though ruby itself is not affected, it can be an accomplice in attacks by blindly passing through any uncontrolled info through. Let's just change this, more secure by default. This patch does not add a new feature, nor delete anything. It just changes the default behaviour when ruby spawns subprocesses. Process.spawn('/usr/bin/printenv') # -> prints nothing You can explicitly pass what you need: Process.spawn({'FOO'=>'BAR'}, '/usr/bin/printenv') Or if you are 128% sure what you are doing, can pass everything. ENV['FOO'] = 'BAR' Process.spawn('/usr/bin/printenv', unsetenv_others: false) I don't intend to make things impossible; just give it a better default. It doesn't ultimately solve everything but it should prevent casual faults. --- process.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/process.c b/process.c index f9bc01c..395333d 100644 --- a/process.c +++ b/process.c @@ -2295,7 +2295,7 @@ rb_execarg_fixup(VALUE execarg_obj) eargp->dup2_tmpbuf = tmpbuf; } - unsetenv_others = eargp->unsetenv_others_given && eargp->unsetenv_others_do; + unsetenv_others = !eargp->unsetenv_others_given || eargp->unsetenv_others_do; envopts = eargp->env_modification; if (unsetenv_others || envopts != Qfalse) { VALUE envtbl, envp_str, envp_buf; @@ -2936,7 +2936,7 @@ rb_execarg_run_options(const struct rb_execarg *eargp, struct rb_execarg *sargp, #endif #if !defined(HAVE_WORKING_FORK) - if (eargp->unsetenv_others_given && eargp->unsetenv_others_do) { + if (!eargp->unsetenv_others_given || eargp->unsetenv_others_do) { save_env(sargp); rb_env_clear(); } @@ -3998,8 +3998,8 @@ rb_f_system(int argc, VALUE *argv) * [cmdname, argv0], arg1, ... : command name, argv[0] and zero or more arguments (no shell) * options: hash * clearing environment variables: - * :unsetenv_others => true : clear environment variables except specified by env - * :unsetenv_others => false : don't clear (default) + * :unsetenv_others => true : clear environment variables except specified by env (default) + * :unsetenv_others => false : don't clear * process group: * :pgroup => true or 0 : make a new process group * :pgroup => pgid : join to specified process group
Updated by shyouhei (Shyouhei Urabe) over 6 years ago
On 10/10/2014 02:16 AM, Eric Wong wrote:
shyouhei@ruby-lang.org wrote:
This patch does not add a new feature, nor delete anything. It just
changes the default behaviour when ruby spawns subprocesses.Process.spawn('/usr/bin/printenv') # -> prints nothingThe potential for breakage is way too high.
I understand this.
Losing some envs (e.g.
PATH, TMPDIR, SHELL or HOME) can be disastrous and introduce new
security problems.
After shellshock I started thinking that every environment variables shall be inspected before passing to another process. There can be various ways, like introducing "the value is sane" flag to each env vars (default false) and let programms check them explicitly, for instance. The approach in this patch is to force programmers write what to pass. For instance if you think PATH shall not be clobbered you should add {'PATH'=ENV['PATH']}.
Right now, everybody knows about shellshock and patching bash.
(I know at least one example who doesn't... the problem is that machine will not update ruby either)
This is an over-reaction which causes needless breakage.
(Especially since your example never even spawns a shell)
What if the spawned subprocess then spawns its own shell? Like I said ruby itself is immune to shellshock. That doesn't mean all the subprocess that we spawn are. Same discussion goes to our subprocesses as well. When they are not shells, that doesn't always mean they don't spawn a shell.
Updated by akr (Akira Tanaka) over 6 years ago
Shyouhei Urabe wrote:
Akira, isn't it obvious this patch introduce a backward incompatibility?
I'd be surprised if it could passed the tests, because that means our tests are insufficient. Is it that important to pass make check at this point?
I don't think it's an "at least" thing for this rather conceptual patch. Is it your only concern?
Of course tests shall be modified to fit this when it is going to land.
The problem is how easy/difficult to modify software to follow this change.
Trying to modify test-all is a way to estimate the difficulty.
I recommend it because it should be done anyway if this issue is accepted.
Updated by shyouhei (Shyouhei Urabe) over 6 years ago
Akira Tanaka wrote:
Shyouhei Urabe wrote:
Akira, isn't it obvious this patch introduce a backward incompatibility?
I'd be surprised if it could passed the tests, because that means our tests are insufficient. Is it that important to pass make check at this point?
I don't think it's an "at least" thing for this rather conceptual patch. Is it your only concern?
Of course tests shall be modified to fit this when it is going to land.
The problem is how easy/difficult to modify software to follow this change.
Trying to modify test-all is a way to estimate the difficulty.
I recommend it because it should be done anyway if this issue is accepted.
I don't understand why you think it is a problem. Difficult? is it?
--- lib/mkmf.rb | 7 +++++-- lib/rake/file_utils.rb | 5 +++-- lib/webrick/httpservlet/cgi_runner.rb | 10 ++++------ test/lib/test/unit.rb | 5 ++++- test/rake/support/ruby_runner.rb | 4 +++- test/rake/test_rake_file_utils.rb | 13 ++++--------- test/ruby/envutil.rb | 5 ++++- test/ruby/test_process.rb | 11 ++++++----- test/ruby/test_require.rb | 32 +++++++++++++++++++++----------- test/ruby/test_rubyoptions.rb | 28 ++++++++++++++-------------- test/rubygems/test_gem.rb | 10 ++-------- tool/runruby.rb | 5 ++--- 12 files changed, 72 insertions(+), 63 deletions(-) diff --git a/lib/mkmf.rb b/lib/mkmf.rb index cb66d88..d5b5fe7 100644 --- a/lib/mkmf.rb +++ b/lib/mkmf.rb @@ -384,15 +384,18 @@ module MakeMakefile end Logging::open do puts command.quote + envp = { + "PATH"=>ENV["PATH"], + } if opts and opts[:werror] result = nil Logging.postpone do |log| - result = (system(libpath_env, command) and File.zero?(log.path)) + result = (system(envp.merge(libpath_env), command) and File.zero?(log.path)) "" end result else - system(libpath_env, command) + system(envp.merge(libpath_env), command) end end end diff --git a/lib/rake/file_utils.rb b/lib/rake/file_utils.rb index 27f4e2e..14e8686 100644 --- a/lib/rake/file_utils.rb +++ b/lib/rake/file_utils.rb @@ -90,10 +90,11 @@ module FileUtils # def ruby(*args, &block) options = (Hash === args.last) ? args.pop : {} + envp = (Hash === args.first) ? args.shift : {} if args.length > 1 - sh(*([RUBY] + args + [options]), &block) + sh(envp, *([RUBY] + args + [options]), &block) else - sh("#{RUBY} #{args.first}", options, &block) + sh(envp, "#{RUBY} #{args.first}", options, &block) end end diff --git a/lib/webrick/httpservlet/cgi_runner.rb b/lib/webrick/httpservlet/cgi_runner.rb index 32ecb6f..85722d5 100644 --- a/lib/webrick/httpservlet/cgi_runner.rb +++ b/lib/webrick/httpservlet/cgi_runner.rb @@ -31,16 +31,14 @@ STDERR.reopen(open(err, "w")) len = sysread(STDIN, 8).to_i dump = sysread(STDIN, len) hash = Marshal.restore(dump) -ENV.keys.each{|name| ENV.delete(name) } -hash.each{|k, v| ENV[k] = v if v } -dir = File::dirname(ENV["SCRIPT_FILENAME"]) +dir = File::dirname(hash["SCRIPT_FILENAME"]) Dir::chdir dir if ARGV[0] argv = ARGV.dup - argv << ENV["SCRIPT_FILENAME"] - exec(*argv) + argv << hash["SCRIPT_FILENAME"] + exec(hash, *argv) # NOTREACHED end -exec ENV["SCRIPT_FILENAME"] +exec hash, ENV["SCRIPT_FILENAME"] diff --git a/test/lib/test/unit.rb b/test/lib/test/unit.rb index c4fdf67..e527a1b 100644 --- a/test/lib/test/unit.rb +++ b/test/lib/test/unit.rb @@ -275,7 +275,10 @@ module Test class Worker def self.launch(ruby,args=[]) - io = IO.popen([*ruby, + envp = { + "PATH" => ENV["PATH"], + } + io = IO.popen(envp, [*ruby, "#{File.dirname(__FILE__)}/unit/parallel.rb", *args], "rb+") new(io, io.pid, :waiting) diff --git a/test/rake/support/ruby_runner.rb b/test/rake/support/ruby_runner.rb index d51dd24..d35f9bd 100644 --- a/test/rake/support/ruby_runner.rb +++ b/test/rake/support/ruby_runner.rb @@ -18,7 +18,9 @@ module RubyRunner def run_ruby(option_list) puts "COMMAND: [#{RUBY} #{option_list.join ' '}]" if @verbose - Open3.popen3(RUBY, *option_list) {|inn, out, err, wait| + # It seems vital for some tests to pass RAKE_* envitonment variables + env = ENV.keys.grep(/^RAKE/).each_with_object({}){|k, r| r[k] = ENV[k] } + Open3.popen3(env, RUBY, *option_list) {|inn, out, err, wait| inn.close @exit = wait ? wait.value : $? diff --git a/test/rake/test_rake_file_utils.rb b/test/rake/test_rake_file_utils.rb index 37d33dc..80971d4 100644 --- a/test/rake/test_rake_file_utils.rb +++ b/test/rake/test_rake_file_utils.rb @@ -130,18 +130,16 @@ class TestRakeFileUtils < Rake::TestCase def test_sh_with_a_single_string_argument check_expansion - ENV['RAKE_TEST_SH'] = 'someval' verbose(false) { - sh %{#{RUBY} check_expansion.rb #{env_var} someval} + sh({'RAKE_TEST_SH'=>'someval'}, %{#{RUBY} check_expansion.rb #{env_var} someval}) } end def test_sh_with_multiple_arguments check_no_expansion - ENV['RAKE_TEST_SH'] = 'someval' verbose(false) { - sh RUBY, 'check_no_expansion.rb', env_var, 'someval' + sh({'RAKE_TEST_SH'=>'someval'}, RUBY, 'check_no_expansion.rb', env_var, 'someval') } end @@ -225,11 +223,9 @@ class TestRakeFileUtils < Rake::TestCase def test_ruby_with_a_single_string_argument check_expansion - ENV['RAKE_TEST_SH'] = 'someval' - verbose(false) { replace_ruby { - ruby %{check_expansion.rb #{env_var} someval} + ruby({'RAKE_TEST_SH'=>'someval'}, %{check_expansion.rb #{env_var} someval}) } } end @@ -237,10 +233,9 @@ class TestRakeFileUtils < Rake::TestCase def test_ruby_with_multiple_arguments check_no_expansion - ENV['RAKE_TEST_SH'] = 'someval' verbose(false) { replace_ruby { - ruby 'check_no_expansion.rb', env_var, 'someval' + ruby({'RAKE_TEST_SH'=>'someval'}, 'check_no_expansion.rb', env_var, 'someval') } } end diff --git a/test/ruby/envutil.rb b/test/ruby/envutil.rb index f5fbb7c..b02bed3 100644 --- a/test/ruby/envutil.rb +++ b/test/ruby/envutil.rb @@ -47,7 +47,10 @@ module EnvUtil err_p.set_encoding(encoding) if err_p end c = "C" - child_env = {} + # some defaults + child_env = { + 'RUBYLIB' => ENV['RUBYLIB'], + } LANG_ENVS.each {|lc| child_env[lc] = c} if Array === args and Hash === args.first child_env.update(args.shift) diff --git a/test/ruby/test_process.rb b/test/ruby/test_process.rb index b314b7f..56700d1 100644 --- a/test/ruby/test_process.rb +++ b/test/ruby/test_process.rb @@ -282,7 +282,8 @@ class TestProcess < Test::Unit::TestCase end end cmd << '-e' << 'puts ENV.keys.map{|e|e.upcase}' - IO.popen(cmd) {|io| + # @shyouhei's note: it seems this test intended to pass all envs. + IO.popen(cmd, unsetenv_others: false) {|io| assert_equal("PATH\n", io.read) } @@ -298,11 +299,11 @@ class TestProcess < Test::Unit::TestCase old = ENV["hmm"] begin ENV["hmm"] = "fufu" - IO.popen(ENVCOMMAND) {|io| assert_match(/^hmm=fufu$/, io.read) } + IO.popen(ENVCOMMAND) {|io| assert_match(/^$/, io.read) } IO.popen([{"hmm"=>""}, *ENVCOMMAND]) {|io| assert_match(/^hmm=$/, io.read) } IO.popen([{"hmm"=>nil}, *ENVCOMMAND]) {|io| assert_not_match(/^hmm=/, io.read) } ENV["hmm"] = "" - IO.popen(ENVCOMMAND) {|io| assert_match(/^hmm=$/, io.read) } + IO.popen(ENVCOMMAND) {|io| assert_match(/^$/, io.read) } IO.popen([{"hmm"=>""}, *ENVCOMMAND]) {|io| assert_match(/^hmm=$/, io.read) } IO.popen([{"hmm"=>nil}, *ENVCOMMAND]) {|io| assert_not_match(/^hmm=/, io.read) } ENV["hmm"] = nil @@ -323,11 +324,11 @@ class TestProcess < Test::Unit::TestCase old = ENV["hmm"] begin ENV["hmm"] = "fufu" - IO.popen(cmd) {|io| assert_match(/^hmm=fufu$/, io.read, message)} + IO.popen(cmd) {|io| assert_match(/^$/, io.read, message)} IO.popen({"hmm"=>""}, cmd) {|io| assert_match(/^hmm=$/, io.read, message)} IO.popen({"hmm"=>nil}, cmd) {|io| assert_not_match(/^hmm=/, io.read, message)} ENV["hmm"] = "" - IO.popen(cmd) {|io| assert_match(/^hmm=$/, io.read, message)} + IO.popen(cmd) {|io| assert_match(/^$/, io.read, message)} IO.popen({"hmm"=>""}, cmd) {|io| assert_match(/^hmm=$/, io.read, message)} IO.popen({"hmm"=>nil}, cmd) {|io| assert_not_match(/^hmm=/, io.read, message)} ENV["hmm"] = nil diff --git a/test/ruby/test_require.rb b/test/ruby/test_require.rb index 3bd327c..9e76bbf 100644 --- a/test/ruby/test_require.rb +++ b/test/ruby/test_require.rb @@ -118,9 +118,11 @@ class TestRequire < Test::Unit::TestCase env_rubypath, env_home = ENV["RUBYPATH"], ENV["HOME"] pathname_too_long = /pathname too long \(ignored\).*\(LoadError\)/m - ENV["RUBYPATH"] = "~" - ENV["HOME"] = "/foo" * 1024 - assert_in_out_err(%w(-S -w test_ruby_test_require), "", [], pathname_too_long) + env = { + "RUBYPATH" => "~", + "HOME" => "/foo" * 1024, + } + assert_in_out_err([env] + %w(-S -w test_ruby_test_require), "", [], pathname_too_long) ensure env_rubypath ? ENV["RUBYPATH"] = env_rubypath : ENV.delete("RUBYPATH") @@ -131,9 +133,11 @@ class TestRequire < Test::Unit::TestCase env_rubypath, env_home = ENV["RUBYPATH"], ENV["HOME"] pathname_too_long = /pathname too long \(ignored\).*\(LoadError\)/m - ENV["RUBYPATH"] = "~" + "/foo" * 1024 - ENV["HOME"] = "/foo" - assert_in_out_err(%w(-S -w test_ruby_test_require), "", [], pathname_too_long) + env = { + "RUBYPATH" => "~" + "/foo" * 1024, + "HOME" => "/foo", + } + assert_in_out_err([env] + %w(-S -w test_ruby_test_require), "", [], pathname_too_long) ensure env_rubypath ? ENV["RUBYPATH"] = env_rubypath : ENV.delete("RUBYPATH") @@ -147,12 +151,18 @@ class TestRequire < Test::Unit::TestCase t.puts "p :ok" t.close - ENV["RUBYPATH"] = "~" - ENV["HOME"] = t.path - assert_in_out_err(%w(-S test_ruby_test_require), "", [], /\(LoadError\)/) + env = { + "RUBYPATH" => "~", + "HOME" => t.path, + } + assert_in_out_err([env] + %w(-S test_ruby_test_require), "", [], /\(LoadError\)/) - ENV["HOME"], name = File.split(t.path) - assert_in_out_err(["-S", name], "", %w(:ok), []) + home, name = File.split(t.path) + env = { + "RUBYPATH" => "~", + "HOME" => home, + } + assert_in_out_err([env, "-S", name], "", %w(:ok), []) } ensure env_rubypath ? ENV["RUBYPATH"] = env_rubypath : ENV.delete("RUBYPATH") diff --git a/test/ruby/test_rubyoptions.rb b/test/ruby/test_rubyoptions.rb index d4e5b2a..ba62969 100644 --- a/test/ruby/test_rubyoptions.rb +++ b/test/ruby/test_rubyoptions.rb @@ -242,20 +242,20 @@ class TestRubyOptions < Test::Unit::TestCase def test_rubyopt rubyopt_orig = ENV['RUBYOPT'] - ENV['RUBYOPT'] = ' - -' - assert_in_out_err([], "", [], []) + env = {'RUBYOPT' => ' - -'} + assert_in_out_err([env], "", [], []) - ENV['RUBYOPT'] = '-e "p 1"' - assert_in_out_err([], "", [], /invalid switch in RUBYOPT: -e \(RuntimeError\)/) + env['RUBYOPT'] = '-e "p 1"' + assert_in_out_err([env], "", [], /invalid switch in RUBYOPT: -e \(RuntimeError\)/) - ENV['RUBYOPT'] = '-T1' - assert_in_out_err(["--disable-gems"], "", [], /no program input from stdin allowed in tainted mode \(SecurityError\)/) + env['RUBYOPT'] = '-T1' + assert_in_out_err([env, "--disable-gems"], "", [], /no program input from stdin allowed in tainted mode \(SecurityError\)/) - ENV['RUBYOPT'] = '-T4' - assert_in_out_err(["--disable-gems"], "", [], /no program input from stdin allowed in tainted mode \(SecurityError\)/) + env['RUBYOPT'] = '-T4' + assert_in_out_err([env, "--disable-gems"], "", [], /no program input from stdin allowed in tainted mode \(SecurityError\)/) - ENV['RUBYOPT'] = '-Eus-ascii -KN' - assert_in_out_err(%w(-Eutf-8 -KU), "p '\u3042'") do |r, e| + env['RUBYOPT'] = '-Eus-ascii -KN' + assert_in_out_err([env]+%w(-Eutf-8 -KU), "p '\u3042'") do |r, e| assert_equal("\"\u3042\"", r.join.force_encoding(Encoding::UTF_8)) assert_equal([], e) end @@ -279,13 +279,13 @@ class TestRubyOptions < Test::Unit::TestCase @verbose = $VERBOSE $VERBOSE = nil - ENV['PATH'] = File.dirname(t.path) + env={'PATH' => File.dirname(t.path)} - assert_in_out_err(%w(-S) + [File.basename(t.path)], "", %w(1), []) + assert_in_out_err([env] + %w(-S) + [File.basename(t.path)], "", %w(1), []) - ENV['RUBYPATH'] = File.dirname(t.path) + env['RUBYPATH'] = File.dirname(t.path) - assert_in_out_err(%w(-S) + [File.basename(t.path)], "", %w(1), []) + assert_in_out_err([env] + %w(-S) + [File.basename(t.path)], "", %w(1), []) } ensure diff --git a/test/rubygems/test_gem.rb b/test/rubygems/test_gem.rb index 0da28bb..85556ae 100644 --- a/test/rubygems/test_gem.rb +++ b/test/rubygems/test_gem.rb @@ -1260,10 +1260,7 @@ class TestGem < Gem::TestCase install_gem b, :install_dir => path install_gem c, :install_dir => path - ENV['GEM_PATH'] = path - ENV['RUBYGEMS_GEMDEPS'] = "-" - - out = `#{Gem.ruby.dup.untaint} -I #{LIB_PATH.untaint} -rubygems -e "p Gem.loaded_specs.values.map(&:full_name).sort"` + out = `GEM_PATH=#{path} RUBYGEMS_GEMDEPS=- #{Gem.ruby.dup.untaint} -I #{LIB_PATH.untaint} -rubygems -e "p Gem.loaded_specs.values.map(&:full_name).sort"` assert_equal '["a-1", "b-1", "c-1"]', out.strip end @@ -1290,12 +1287,9 @@ class TestGem < Gem::TestCase install_gem b, :install_dir => path install_gem c, :install_dir => path - ENV['GEM_PATH'] = path - ENV['RUBYGEMS_GEMDEPS'] = "-" - Dir.mkdir "sub1" out = Dir.chdir "sub1" do - `#{Gem.ruby.dup.untaint} -I #{LIB_PATH.untaint} -rubygems -e "p Gem.loaded_specs.values.map(&:full_name).sort"` + `GEM_PATH=#{path} RUBYGEMS_GEMDEPS=- #{Gem.ruby.dup.untaint} -I #{LIB_PATH.untaint} -rubygems -e "p Gem.loaded_specs.values.map(&:full_name).sort"` end Dir.rmdir "sub1" diff --git a/tool/runruby.rb b/tool/runruby.rb index 86b9327..cf63055 100755 --- a/tool/runruby.rb +++ b/tool/runruby.rb @@ -66,6 +66,7 @@ config["bindir"] = abs_archdir env = {} +env["HOME"] = ENV["HOME"] # RDoc relies on it env["RUBY"] = File.expand_path(ruby) env["PATH"] = [abs_archdir, ENV["PATH"]].compact.join(File::PATH_SEPARATOR) @@ -84,8 +85,6 @@ if File.file?(libruby_so) end end -ENV.update env - cmd = [ruby] cmd.concat(ARGV) cmd.unshift(*precommand) unless precommand.empty? @@ -96,4 +95,4 @@ if show puts Shellwords.join(cmd) end -exec(*cmd) +exec(env, *cmd) -- 1.9.1
Updated by jeremyevans0 (Jeremy Evans) over 6 years ago
I agree with Eric, this change is too likely to break things, including the principle of least surprise. No other program language I'm aware of spawns system commands with an empty environment by default. Here are a few I tried:
awk 'BEGIN {system("env")}'
perl -e "system('env')"
ruby -e "system('env')"
python -c 'import os; os.system("env")'
php -r "system('env');"
lua -e 'os.execute("env")'
clojure -e "(use '[clojure.java.shell :only [sh]]) (println (:out (sh \"env\"))) (System/exit 0)"
gforth -e 'sh "env"' -e bye
The costs of this proposed change are much higher than the benefits. This just makes interaction with the operating system in general more difficult. It is likely this patch introduces more vulnerabilities than it fixes, in addition to breaking things.
Updated by shyouhei (Shyouhei Urabe) over 6 years ago
Jeremy Evans wrote:
The costs of this proposed change are much higher than the benefits. This just makes interaction with the operating system in general more difficult. It is likely this patch introduces more vulnerabilities than it fixes, in addition to breaking things.
I'm not a fundamentalist here. If anyone could propose a moderate way to samitize malicious environment variables, that should be taken seriously.
After shellshock we have to realize that environment variables are not as safe as we thought them to be. I don't think it's a good thing to neglect this fact to blame bash only. I'm proposing what I can do.
Updated by normalperson (Eric Wong) over 6 years ago
How about supporting ENV.freeze instead?
Currently ENV.freeze is a no-op, this patch changes that:
http://80x24.org/spew/m/400e216159e74b65608f2f0b296817cc9823e3bb.txt
ENV.freeze should behave like Hash#freeze and not allow future modifications to the ENV from pure Ruby libraries. Users may call ENV.freeze to prevent 3rd-party (pure) Ruby libraries from modifying the process environment any further. This cannot not defend against users who modify the environment using 3rd-party C extensions, Fiddle, or FFI RubyGem
Updated by shyouhei (Shyouhei Urabe) over 6 years ago
I see this patch is (basically) good to have.
One thing sprung in my mind is that perhaps we should also freeze each environment variables, not only the ENV object. That is not the same way Hash#freeze works, but ENV keys are already frozen as-is so also freezing values might be an opiton. That should prevent modifying already-existent environment variables, such as RUBYOPT.
On 10/21/2014 05:24 AM, Eric Wong wrote:
How about supporting ENV.freeze instead?
Currently ENV.freeze is a no-op, this patch changes that:http://80x24.org/spew/m/400e216159e74b65608f2f0b296817cc9823e3bb.txt
ENV.freeze should behave like Hash#freeze and not allow future modifications to the ENV from pure Ruby libraries. Users may call ENV.freeze to prevent 3rd-party (pure) Ruby libraries from modifying the process environment any further. This cannot not defend against users who modify the environment using 3rd-party C extensions, Fiddle, or FFI RubyGem