Bug #7312: test_str_crypt(TestM17NComb) fails

• Status: Closed
• Priority: Normal
• Assignee: naruse (Yui NARUSE)
• Target version: 2.0.0
• ruby -v: ruby 2.0.0dev (2012-11-19 trunk 37735) [x86_64-linux]

[vo.x (Vit Ondruch)]

=begin
test_str_crypt(TestM17NComb):
ArgumentError: NULL pointer given
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/test_m17n_comb.rb:728:in crypt'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/test_m17n_comb.rb:728:inblock in test_str_crypt'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:83:in block in each'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:75:inblock in each_index'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:46:in block in make_large_block'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:26:inblock (2 levels) in make_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:21:in times'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:21:inblock in make_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:20:in times'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:20:inmake_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:45:in make_large_block'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:71:ineach_index'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/allpairs.rb:82:in each'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/test_m17n_comb.rb:60:incombination'
/builddir/build/BUILD/ruby-2.0.0-r37564/test/ruby/test_m17n_comb.rb:723:in `test_str_crypt'
=end

[mame (Yusuke Endoh)]

• Target version set to 2.0.0

I cannot reproduce. Anyone?

--
Yusuke Endoh [e-mail]

[vo.x (Vit Ondruch)]

I observe the test error on Fedora Rawhide. Could it be because of some "too new" external library?

[naruse (Yui NARUSE)]

• Status changed from Open to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r37572.
Vit, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

• string.c (rb_str_crypt): crypt(3) may return NULL. Latest glibc (2.16?) crypt(3) actually returns NULL. [Bug #7312]

[naruse (Yui NARUSE)]

• Status changed from Closed to Feedback
• Assignee set to naruse (Yui NARUSE)

vo.x (Vit Ondruch) wrote:

I observe the test error on Fedora Rawhide. Could it be because of some "too new" external library?

As I wrote in the commit, it seems because of glibc's crypt(3)'s change.
(the fundamental reason is ruby hadn't treat the case crypt(3) returns NULL even if POSIX describe it)
Could you inspect when crypt(3) returns NULL?

[vo.x (Vit Ondruch)]

=begin
This is reduced test case:

def test_str_crypt
str = ""
salt = e("\xa1\xa1")

if a(salt).length < 2
  assert_raise(ArgumentError) { str.crypt(salt) }
end

t = str.crypt(salt)
assert_equal(a(str).crypt(a(salt)), t, "#{encdump(str)}.crypt(#{encdump(salt)})")
assert_encoding('ASCII-8BIT', t.encoding)

end

And GDB session:

6916 StringValue(salt);
(gdb)
6917 if (RSTRING_LEN(salt) < 2)
(gdb)
6920 s = RSTRING_PTR(str);
(gdb)
6922 saltp = RSTRING_PTR(salt);
(gdb)
6931 result = rb_str_new2(crypt(s, saltp));
(gdb) s
rb_str_new_cstr (ptr=0x0) at string.c:435
435 if (!ptr) {

So the crypt(3) definitely returns NULL for e("\xa1\xa1").

BTW this is my glibc version:

$ rpm -q glibc
glibc-2.16.90-28.fc19.x86_64

And this [1] might be the offending patch, particularly the crypt/crypt-entry.c change? It seems that they are now stricter what might be salt now.

[1] ((URL:http://sourceware.org/git/?p=glibc.git;a=commitdiff;h=4ba74a357376c8f8bf49487f96ae71cf2460c3f3))
=end

[vo.x (Vit Ondruch)]

• ruby -v changed from ruby 2.0.0dev (2012-11-08 trunk 37564) [x86_64-linux] to ruby 2.0.0dev (2012-11-09 trunk 37589) [x86_64-linux]

=begin
The unit test still fails:

test_str_crypt(TestM17NComb):
Errno::EINVAL: Invalid argument - crypt
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/test_m17n_comb.rb:728:in crypt'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/test_m17n_comb.rb:728:inblock in test_str_crypt'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:83:in block in each'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:75:inblock in each_index'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:46:in block in make_large_block'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:26:inblock (2 levels) in make_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:21:in times'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:21:inblock in make_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:20:in times'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:20:inmake_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:45:in make_large_block'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:71:ineach_index'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/allpairs.rb:82:in each'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/test_m17n_comb.rb:60:incombination'
/builddir/build/BUILD/ruby-2.0.0-r37589/test/ruby/test_m17n_comb.rb:723:in `test_str_crypt'
=end

[mtasaka (Mamoru Tasaka)]

Test case:

#define _XOPEN_SOURCE
#include
#include
#include
#include

int main(void){
const char *key = "";
const char *salt = "\xa1\xa1";

char *ret = crypt(key, salt);
int errno_save = errno;
printf("crypt returned val: %p\n", ret);
if (!ret){
    printf("Errno: %d %s\n", errno_save, strerror(errno_save));
}
while (ret && *ret){
    printf("char: %c\n", *ret++);
}

return 0;

}

With glibc 2.16-20.fc18.i686
$ gcc -Wall -Wextra -O2 -g -o test-crypt test-crypt.c -lcrypt && ./test-crypt
crypt returned val: 0x45b83140
char: �
char: �
char: z
char: U
char: w
char: S
char: p
char: 9
char: H
char: 4
char: n
char: 3
char: Y

With glibc 2.16.90-28.fc19.i686
gcc -Wall -Wextra -O2 -g -o test-crypt test-crypt.c -lcrypt && ./test-crypt
crypt returned val: (nil)
Errno: 22 Invalid argument

Note "Invalid argument". man 3p crypt says:

The salt argument is a string chosen from the set:

          a b c d e f g h i j k l m n o p q r s t u v w x y z
          A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
          0 1 2 3 4 5 6 7 8 9 . /

---

so "STRINGS" in test/ruby/test_m17n_comb.rb seems what should
not passed to salt argument for crypt(3).

[mtasaka (Mamoru Tasaka)]

Note that this test (test/ruby/test_m17n_comb.rb) also fails with
ruby 1.9.3p327 with rawhide glibc.

[naruse (Yui NARUSE)]

• Status changed from Feedback to Closed

This issue was solved with changeset r37622.
Vit, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

glibc 2.16 or later denies salt contained other than [0-9A-Za-z./] [Bug #7312]

[kosaki (Motohiro KOSAKI)]

I guess following discussion is a source of this bug.

http://sourceware-org.1504.n7.nabble.com/RFC-FIPS-compliance-and-other-crypt-3-improvements-td6884.html

[vo.x (Vit Ondruch)]

• Status changed from Closed to Open
• % Done changed from 100 to 30
• ruby -v changed from ruby 2.0.0dev (2012-11-09 trunk 37589) [x86_64-linux] to ruby 2.0.0dev (2012-11-19 trunk 37735) [x86_64-linux]

=begin
The fix does not fully work. There are at lease three issues:

(1) It does not work on 64b system, since libc is located in /usr/lib64
(2) The glibcver <=> [2, 16] compares array of strings with array of integers, which returns nil and therefor raises exception.
(3) Even if I force the strict crypt, it fails with
1) Error:
test_str_crypt(TestM17NComb):
ArgumentError: invalid byte sequence in EUC-JP
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/test_m17n_comb.rb:732:in block in test_str_crypt'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:83:inblock in each'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:75:in block in each_index'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:46:inblock in make_large_block'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:26:in block (2 levels) in make_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:21:intimes'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:21:in block in make_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:20:intimes'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:20:in make_basic_block'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:45:inmake_large_block'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:71:in each_index'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/allpairs.rb:82:ineach'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/test_m17n_comb.rb:60:in combination'
/builddir/build/BUILD/ruby-2.0.0-r37735/test/ruby/test_m17n_comb.rb:730:intest_str_crypt'

=end

[naruse (Yui NARUSE)]

vo.x (Vit Ondruch) wrote:

The fix does not fully work. There are at lease three issues:

(1) It does not work on 64b system, since libc is located in /usr/lib64
(2) The glibcver <=> [2, 16] compares array of strings with array of integers, which returns nil and therefor raises exception.
(3) Even if I force the strict crypt, it fails with

I fix (2) and (3) at r37766.

For (1), I cannot find suitable way.
Could you make a patch?

[vo.x (Vit Ondruch)]

• File 0001-Search-for-arch-specific-libc-location.patch 0001-Search-for-arch-specific-libc-location.patch added

naruse (Yui NARUSE) wrote:

For (1), I cannot find suitable way.
Could you make a patch?

This should work.

[naruse (Yui NARUSE)]

• Status changed from Open to Closed
• % Done changed from 30 to 100

This issue was solved with changeset r37773.
Vit, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

• test/ruby/test_m17n_comb.rb (test_str_crypt): Use RbConfig to get libc's directory. Patched by Vit Ondruch [ruby-core:49763] [Bug #7312]