Re: [PATCH v4 3.0-rc2-tip 7/22] 7: uprobes: mmap and fork hooks.

[Peter Zijlstra]

On Fri, 2011-06-17 at 10:20 +0530, Srikar Dronamraju wrote:
> > 
> > void __unregister_uprobe(...)
> > {
> >   uprobe = find_uprobe(); // ref++
> >   if (delete_consumer(...)); // includes tree removal on last consumer
> >                              // implies we own the last ref
> >      return; // consumers
> > 
> >   vma_prio_tree_foreach() {
> >      // create list
> >   }
> > 
> >   list_for_each_entry_safe() {
> >     // remove from list
> >     remove_breakpoint(); // unconditional, if it wasn't there
> >                          // its a nop anyway, can't get any new
> >                          // new probes on account of holding
> >                          // uprobes_mutex and mmap() doesn't see
> >                          // it due to tree removal.
> >   }
> > }
> > 
> 
> This would have a bigger race.
> A breakpoint might be hit by which time the node is removed and we
> have no way to find out the uprobe. So we deliver an extra TRAP to the
> app.

Gah indeed. Back to the drawing board for me.

> > int mmap_uprobe(...)
> > {
> >   spin_lock(&uprobes_treelock);
> >   for_each_probe_in_inode() {
> >     // create list;
> >   }
> >   spin_unlock(..);
> > 
> >   list_for_each_entry_safe() {
> >     // remove from list
> >     ret = install_breakpoint();
> >     if (ret)
> >       goto fail;
> >     if (!uprobe_still_there()) // takes treelock
> >       remove_breakpoint();
> >   }
> > 
> >   return 0;
> > 
> > fail:
> >   list_for_each_entry_safe() {
> >     // destroy list
> >   }
> >   return ret;
> > }
> > 
> 
> 
> register_uprobe will race with mmap_uprobe's first pass.
> So we might end up with a vma that doesnot have a breakpoint inserted
> but inserted in all other vma that map to the same inode.

I'm not seeing this though, if mmap_uprobe() is before register_uprobe()
inserts the probe in the tree, the vma is already in the rmap and
register_uprobe() will find it in its vma walk. If its after,
mmap_uprobe() will find it and install, if a concurrent
register_uprobe()'s vma walk also finds it, it will -EEXISTS and ignore
the error.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/