lkml.org 
[lkml]   [2009]   [Jun]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject[PATCH] HWPOISON: fix tasklist_lock/anon_vma locking order
On Tue, Jun 09, 2009 at 06:09:22PM +0800, Nick Piggin wrote:
> On Wed, Jun 03, 2009 at 08:46:47PM +0200, Andi Kleen wrote:
>
> Why not have this in rmap.c and not export the locking?
> I don't know.. does Hugh care?

I don't know either :)

> > +/*
> > + * Collect processes when the error hit an anonymous page.
> > + */
> > +static void collect_procs_anon(struct page *page, struct list_head *to_kill,
> > + struct to_kill **tkc)
> > +{
> > + struct vm_area_struct *vma;
> > + struct task_struct *tsk;
> > + struct anon_vma *av = page_lock_anon_vma(page);
> > +
> > + if (av == NULL) /* Not actually mapped anymore */
> > + return;
> > +
> > + read_lock(&tasklist_lock);
> > + for_each_process (tsk) {
> > + if (!tsk->mm)
> > + continue;
> > + list_for_each_entry (vma, &av->head, anon_vma_node) {
> > + if (vma->vm_mm == tsk->mm)
> > + add_to_kill(tsk, page, vma, to_kill, tkc);
> > + }
> > + }
> > + page_unlock_anon_vma(av);
> > + read_unlock(&tasklist_lock);
> > +}
> > +
> > +/*
> > + * Collect processes when the error hit a file mapped page.
> > + */
> > +static void collect_procs_file(struct page *page, struct list_head *to_kill,
> > + struct to_kill **tkc)
> > +{
> > + struct vm_area_struct *vma;
> > + struct task_struct *tsk;
> > + struct prio_tree_iter iter;
> > + struct address_space *mapping = page_mapping(page);
> > +
> > + /*
> > + * A note on the locking order between the two locks.
> > + * We don't rely on this particular order.
> > + * If you have some other code that needs a different order
> > + * feel free to switch them around. Or add a reverse link
> > + * from mm_struct to task_struct, then this could be all
> > + * done without taking tasklist_lock and looping over all tasks.
> > + */
> > +
> > + read_lock(&tasklist_lock);
> > + spin_lock(&mapping->i_mmap_lock);
>
> This still has my original complaint that it nests tasklist lock inside
> anon vma lock and outside inode mmap lock (and anon_vma nests inside i_mmap).
> I guess the property of our current rw locks means that does not matter,
> but it could if we had "fair" rw locks, or some tree (-rt tree maybe)
> changed rw lock to a plain exclusive lock.

Andi must forgot that - he did change the comment on locking order.
This incremental patch aligns the code with his comment in rmap.c.

---
HWPOISON: fix tasklist_lock/anon_vma locking order

To avoid possible deadlock. Proposed by Nick Piggin:

You have tasklist_lock(R) nesting outside i_mmap_lock, and inside anon_vma
lock. And anon_vma lock nests inside i_mmap_lock.

This seems fragile. If rwlocks ever become FIFO or tasklist_lock changes
type (maybe -rt kernels do it), then you could have a task holding
anon_vma lock and waiting for tasklist_lock, and another holding tasklist
lock and waiting for i_mmap_lock, and another holding i_mmap_lock and
waiting for anon_vma lock.

CC: Nick Piggin <npiggin@suse.de>
CC: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
---
mm/memory-failure.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)

--- sound-2.6.orig/mm/memory-failure.c
+++ sound-2.6/mm/memory-failure.c
@@ -215,12 +215,14 @@ static void collect_procs_anon(struct pa
{
struct vm_area_struct *vma;
struct task_struct *tsk;
- struct anon_vma *av = page_lock_anon_vma(page);
+ struct anon_vma *av;

+ read_lock(&tasklist_lock);
+
+ av = page_lock_anon_vma(page);
if (av == NULL) /* Not actually mapped anymore */
- return;
+ goto out;

- read_lock(&tasklist_lock);
for_each_process (tsk) {
if (!tsk->mm)
continue;
@@ -230,6 +232,7 @@ static void collect_procs_anon(struct pa
}
}
page_unlock_anon_vma(av);
+out:
read_unlock(&tasklist_lock);
}


\
 
 \ /
  Last update: 2009-06-10 05:13    [W:0.194 / U:1.196 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site