gru: fix potential use-after-free when purging GRU tlbs

Fix potential SGI GRU bug that could cause a use-after-free.  If one
thread in a task is flushing the GRU and another thread destroys the GRU
context, there is the potential to access a table after it has been freed.

Copy the gms pointer to a local variable before unlocking the gts table.
Note that no refcnt is needed for the gms - the reference is held
indirectly by the task's mm_struct.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This commit is contained in:
Jack Steiner 2009-06-17 16:28:33 -07:00 committed by Linus Torvalds
parent b1b19fcfa4
commit 1926ee85a9

View file

@ -702,6 +702,7 @@ int gru_user_flush_tlb(unsigned long arg)
{ {
struct gru_thread_state *gts; struct gru_thread_state *gts;
struct gru_flush_tlb_req req; struct gru_flush_tlb_req req;
struct gru_mm_struct *gms;
STAT(user_flush_tlb); STAT(user_flush_tlb);
if (copy_from_user(&req, (void __user *)arg, sizeof(req))) if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
@ -714,8 +715,9 @@ int gru_user_flush_tlb(unsigned long arg)
if (!gts) if (!gts)
return -EINVAL; return -EINVAL;
gru_flush_tlb_range(gts->ts_gms, req.vaddr, req.len); gms = gts->ts_gms;
gru_unlock_gts(gts); gru_unlock_gts(gts);
gru_flush_tlb_range(gms, req.vaddr, req.len);
return 0; return 0;
} }