On 8/23/18 8:15 AM, Oleg Nesterov wrote:
On 08/22, Srikar Dronamraju wrote:
* Vlastimil Babka <vbabka@xxxxxxx> [2018-08-22 12:55:59]:
On 08/15/2018 08:49 PM, Yang Shi wrote:
We need check if mm or vma has uprobes in the following patch to check
if a vma could be unmapped with holding read mmap_sem.
Confused... why can't we call uprobe_munmap() under read_lock(mmap_sem) ?
I'm not sure if it is safe or not because it is not recommended and not
safe to update vma's vm flags with read mmap_sem. uprobe_munmap() may
update mm flags (MMF_RECALC_UPROBES). So, it sounds safer to not call it
under read mmap_sem.
OK, it can race with find_active_uprobe() but I do not see anything really
wrong, and a false-positive MMF_RECALC_UPROBES is fine.
Thanks for confirming this. If it is ok to have such race, we don't have
to have has_uprobes() helper anymore since it can be just called under
read mmap_sem without any special handling.
Yang
Again, I think we should simply kill uprobe_munmap(), but this needs another
discussion.
Oleg.
