pppCleanup() not called when user lacks ~/.pppauth directory

[GoogleCodeExporter]

What steps will reproduce the problem?
1. Login with user without ~/.pppauth directory

SVN version.

Please provide any additional information below.

When readKeyFile() fails after pppInit() is done there's no call 
to pppCleanup() which cause soft to leak 32 bytes at:
==7209== 32 bytes in 1 blocks are definitely lost in loss record 5 of 5
==7209==    at 0x4022C3B: calloc (in
/usr/lib/valgrind/vgpreload_memcheck-x86-linux.so)
==7209==    by 0x804A008: mp_init_size (mpi.c:284)
==7209==    by 0x8049F2E: mp_init (mpi.c:238)
==7209==    by 0x8049933: readKeyFile (keyfiles.c:352)
==7209==    by 0x8048EFF: pam_sm_authenticate (memtest.c:56)
==7209==    by 0x8048FED: main (memtest.c:110)

UPDATE: it's also readKeyFile() which calls mp_init, but won't call
mp_clear on error. Also it might sometimes not zero buf, which holds the key.

I did a .c file which simulates logins without help of PAM so I can run it
on valgrind; that's the only case of leak I've found so far. I'm not
entirely sure how PAM works, but I think that this leak wouldn't lead to
some big memory usage on long-running systems (in sshd?) therefore this
leak might not be so important, yet it would be nice to patch it up.

Setting user to NULL at exit of pam_sm_authenticate doesn't do anything
important as I think, so possible fix, using common goto-method is in
attached patch. (Also fixed whitespace 4 spaces vs. tab and removed drop of
const keyword on const char *user; will work fine if setUser prototype is
changed to accept const (as in my previous patch) ).

Original issue reported on code.google.com by TomaszFo...@gmail.com on 7 Nov 2009 at 9:45

Attachments:

• leak.patch