On Mon, Jul 27, 2020 at 5:30 PM Richard Guy Briggs <rgb(a)redhat.com> wrote:
 Issue ghak120 enabled syscall records to accompany required records when
 no rules are present to trigger the storage of syscall context.  A
 reported issue showed that the cwd was not always initialized.  That
 issue was already resolved ... 
Yes and no.  Yes, it appears to be resolved in v5.8-rc1 and above, but
the problematic commit is in v5.7 and I'm not sure backporting the fix
in v5.8-rcX plus this patch is the right thing to do for a released
kernel.  The lowest risk fix for v5.7 at this point is to do a revert;
regardless of what happens with this patch and v5.8-rcX please post a
revert for the audit/stable-5.7 tree as soon as you can.
 ... but a review of all other records that could
 be triggered at the time of a syscall record revealed other potential
 values that could be missing or misleading.  Initialize them.
 The fds array is reset to -1 after the first syscall to indicate it
 isn't valid any more, but was never set to -1 when the context was
 allocated to indicate it wasn't yet valid.
 The audit_inode* functions can be called without going through
 getname_flags() or getname_kernel() that sets audit_names and cwd, so
 set the cwd if it has not already been done so due to audit_names being
 valid.
 The LSM dump_common_audit_data() LSM_AUDIT_DATA_NET:AF_UNIX case was
 missed with the ghak96 patch, so add that case here.
 Please see issue 
https://github.com/linux-audit/audit-kernel/issues/120
 Please see issue 
https://github.com/linux-audit/audit-kernel/issues/96
 Passes audit-testsuite.
 Signed-off-by: Richard Guy Briggs <rgb(a)redhat.com>
 ---
  kernel/auditsc.c     | 3 +++
  security/lsm_audit.c | 1 +
  2 files changed, 4 insertions(+)
 diff --git a/kernel/auditsc.c b/kernel/auditsc.c
 index 6884b50069d1..2f97618e6a34 100644
 --- a/kernel/auditsc.c
 +++ b/kernel/auditsc.c
 @@ -929,6 +929,7 @@ static inline struct audit_context *audit_alloc_context(enum
audit_state state)
         context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0;
         INIT_LIST_HEAD(&context->killed_trees);
         INIT_LIST_HEAD(&context->names_list);
 +       context->fds[0] = -1;
         return context;
  }
 @@ -2076,6 +2077,7 @@ void __audit_inode(struct filename *name, const struct dentry
*dentry,
         }
         handle_path(dentry);
         audit_copy_inode(n, dentry, inode, flags & AUDIT_INODE_NOEVAL);
 +       _audit_getcwd(context);
  }
  void __audit_file(const struct file *file)
 @@ -2194,6 +2196,7 @@ void __audit_inode_child(struct inode *parent,
                 audit_copy_inode(found_child, dentry, inode, 0);
         else
                 found_child->ino = AUDIT_INO_UNSET;
 +       _audit_getcwd(context);
  }
  EXPORT_SYMBOL_GPL(__audit_inode_child);
 diff --git a/security/lsm_audit.c b/security/lsm_audit.c
 index 53d0d183db8f..e93077612246 100644
 --- a/security/lsm_audit.c
 +++ b/security/lsm_audit.c
 @@ -369,6 +369,7 @@ static void dump_common_audit_data(struct audit_buffer *ab,
                                         audit_log_untrustedstring(ab, p);
                                 else
                                         audit_log_n_hex(ab, p, len);
 +                               audit_getcwd();
                                 break;
                         }
                 } 
I understand the "fds[0] = -1" fix in audit_alloc_context()
(ironically, the kzalloc() which is supposed to help with cases like
this, hurts us with this particular field), but I'm still not quite
seeing why we need to sprinkle audit_getcwd() calls everywhere to fix
this bug (this seems more like a feature add than a bigfix).  Yes,
they may fix the problem but it seems like simply adding a
context->pwd test in audit_log_name() similar to what we do in
audit_log_exit() is the correct fix.
We are currently at -rc7 and this really needs to land before v5.8 is
released, presumably this weekend; this means a small and limited bug
fix patch is what is needed.
-- 
paul moore
www.paul-moore.com