On Mon, Aug 22, 2016 at 11:41:34AM -0400, Richard Guy Briggs wrote: > Fix original naive attempt to get/lock access to task->mm->exe_file by > using get_mm_exe_file and checking for NULL. > > See: https://lkml.org/lkml/2016/7/30/97 > > Signed-off-by: Richard Guy Briggs <r...@redhat.com> > --- > kernel/audit_watch.c | 13 ++++++++++--- > 1 files changed, 10 insertions(+), 3 deletions(-) > > diff --git a/kernel/audit_watch.c b/kernel/audit_watch.c > index d6709eb..0b29279 100644 > --- a/kernel/audit_watch.c > +++ b/kernel/audit_watch.c > @@ -19,6 +19,7 @@ > * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA > */ > > +#include <linux/file.h> > #include <linux/kernel.h> > #include <linux/audit.h> > #include <linux/kthread.h> > @@ -540,14 +541,20 @@ int audit_dupe_exe(struct audit_krule *new, struct > audit_krule *old) > > int audit_exe_compare(struct task_struct *tsk, struct audit_fsnotify_mark > *mark) > { > + struct mm_struct *mm; > struct file *exe_file; > unsigned long ino; > dev_t dev; > > - rcu_read_lock(); > - exe_file = rcu_dereference(tsk->mm->exe_file); > + mm = get_task_mm(tsk); > + if (!mm) > + return 0; > + exe_file = get_mm_exe_file(mm); > + mmput(mm); > + if (!exe_file) > + return 0; > ino = exe_file->f_inode->i_ino; > dev = exe_file->f_inode->i_sb->s_dev; > - rcu_read_unlock(); > + fput(exe_file); > return audit_mark_compare(mark, ino, dev); > }
I think this works but I have a better idea. This basically duplicates a part of proc_exe_link and does unnecessary work. Instead, one can introduce get_task_exe_file. task_lock held guarantees the stability of mm struct itself, and then we can we can try to grab the file without mm refcount dance. I'll implement this later today. -- Mateusz Guzik