The patch titled Subject: mempolicy: fix show_numa_map() vs exec() + do_set_mempolicy() race has been added to the -mm tree. Its filename is mempolicy-fix-show_numa_map-vs-exec-do_set_mempolicy-race.patch This patch should soon appear at http://ozlabs.org/~akpm/mmots/broken-out/mempolicy-fix-show_numa_map-vs-exec-do_set_mempolicy-race.patch and later at http://ozlabs.org/~akpm/mmotm/broken-out/mempolicy-fix-show_numa_map-vs-exec-do_set_mempolicy-race.patch Before you just go and hit "reply", please: a) Consider who else should be cc'ed b) Prefer to cc a suitable mailing list as well c) Ideally: find the original patch on the mailing list and do a reply-to-all to that, adding suitable additional cc's *** Remember to use Documentation/SubmitChecklist when testing your code *** The -mm tree is included into linux-next and is updated there every 3-4 working days ------------------------------------------------------ From: Oleg Nesterov <oleg@xxxxxxxxxx> Subject: mempolicy: fix show_numa_map() vs exec() + do_set_mempolicy() race 9e7814404b77 "hold task->mempolicy while numa_maps scans." fixed the race with the exiting task but this is not enough. The current code assumes that get_vma_policy(task) should either see task->mempolicy == NULL or it should be equal to ->task_mempolicy saved by hold_task_mempolicy(), so we can never race with __mpol_put(). But this can only work if we can't race with do_set_mempolicy(), and thus we can't race with another do_set_mempolicy() or do_exit() after that. However, do_set_mempolicy()->down_write(mmap_sem) can not prevent this race. This task can exec, change it's ->mm, and call do_set_mempolicy() after that; in this case they take 2 different locks. Change hold_task_mempolicy() to use get_task_policy(), it never returns NULL, and change show_numa_map() to use __get_vma_policy() or fall back to proc_priv->task_mempolicy. Note: this is the minimal fix, we will cleanup this code later. I think hold_task_mempolicy() and release_task_mempolicy() should die, we can move this logic into show_numa_map(). Or we can move get_task_policy() outside of ->mmap_sem and !CONFIG_NUMA code at least. Signed-off-by: Oleg Nesterov <oleg@xxxxxxxxxx> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> Cc: David Rientjes <rientjes@xxxxxxxxxx> Cc: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx> Cc: Alexander Viro <viro@xxxxxxxxxxxxxxxxxx> Cc: Cyrill Gorcunov <gorcunov@xxxxxxxxxx> Cc: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx> Cc: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx> Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx> Cc: Hugh Dickins <hughd@xxxxxxxxxx> Cc: Andi Kleen <andi@xxxxxxxxxxxxxx> Cc: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- fs/proc/task_mmu.c | 33 +++++++++------------------------ 1 file changed, 9 insertions(+), 24 deletions(-) diff -puN fs/proc/task_mmu.c~mempolicy-fix-show_numa_map-vs-exec-do_set_mempolicy-race fs/proc/task_mmu.c --- a/fs/proc/task_mmu.c~mempolicy-fix-show_numa_map-vs-exec-do_set_mempolicy-race +++ a/fs/proc/task_mmu.c @@ -87,32 +87,14 @@ unsigned long task_statm(struct mm_struc #ifdef CONFIG_NUMA /* - * These functions are for numa_maps but called in generic **maps seq_file - * ->start(), ->stop() ops. - * - * numa_maps scans all vmas under mmap_sem and checks their mempolicy. - * Each mempolicy object is controlled by reference counting. The problem here - * is how to avoid accessing dead mempolicy object. - * - * Because we're holding mmap_sem while reading seq_file, it's safe to access - * each vma's mempolicy, no vma objects will never drop refs to mempolicy. - * - * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy - * is set and replaced under mmap_sem but unrefed and cleared under task_lock(). - * So, without task_lock(), we cannot trust get_vma_policy() because we cannot - * gurantee the task never exits under us. But taking task_lock() around - * get_vma_plicy() causes lock order problem. - * - * To access task->mempolicy without lock, we hold a reference count of an - * object pointed by task->mempolicy and remember it. This will guarantee - * that task->mempolicy points to an alive object or NULL in numa_maps accesses. + * Save get_task_policy() for show_numa_map(). */ static void hold_task_mempolicy(struct proc_maps_private *priv) { struct task_struct *task = priv->task; task_lock(task); - priv->task_mempolicy = task->mempolicy; + priv->task_mempolicy = get_task_policy(task); mpol_get(priv->task_mempolicy); task_unlock(task); } @@ -1406,7 +1388,6 @@ static int show_numa_map(struct seq_file struct vm_area_struct *vma = v; struct numa_maps *md = &numa_priv->md; struct file *file = vma->vm_file; - struct task_struct *task = proc_priv->task; struct mm_struct *mm = vma->vm_mm; struct mm_walk walk = {}; struct mempolicy *pol; @@ -1426,9 +1407,13 @@ static int show_numa_map(struct seq_file walk.private = md; walk.mm = mm; - pol = get_vma_policy(task, vma, vma->vm_start); - mpol_to_str(buffer, sizeof(buffer), pol); - mpol_cond_put(pol); + pol = __get_vma_policy(vma, vma->vm_start); + if (pol) { + mpol_to_str(buffer, sizeof(buffer), pol); + mpol_cond_put(pol); + } else { + mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); + } seq_printf(m, "%08lx %s", vma->vm_start, buffer); _ Patches currently in -mm which might be from oleg@xxxxxxxxxx are prctl-pr_set_mm-introduce-pr_set_mm_map-operation-v4.patch mempolicy-change-alloc_pages_vma-to-use-mpol_cond_put.patch mempolicy-change-get_task_policy-to-return-default_policy-rather-than-null.patch mempolicy-sanitize-the-usage-of-get_task_policy.patch mempolicy-remove-the-task-arg-of-vma_policy_mof-and-simplify-it.patch mempolicy-introduce-__get_vma_policy-export-get_task_policy.patch mempolicy-fix-show_numa_map-vs-exec-do_set_mempolicy-race.patch mempolicy-kill-do_set_mempolicy-down_writemm-mmap_sem.patch mempolicy-unexport-get_vma_policy-and-remove-its-task-arg.patch earlyprintk-re-enable-earlyprintk-calling-early_param.patch fs-proc-task_mmuc-dont-use-task-mm-in-m_start-and-show_map.patch fs-proc-task_mmuc-unify-simplify-do_maps_open-and-numa_maps_open.patch proc-introduce-proc_mem_open.patch fs-proc-task_mmuc-shift-mm_access-from-m_start-to-proc_maps_open.patch fs-proc-task_mmuc-shift-mm_access-from-m_start-to-proc_maps_open-checkpatch-fixes.patch fs-proc-task_mmuc-simplify-the-vma_stop-logic.patch fs-proc-task_mmuc-simplify-the-vma_stop-logic-checkpatch-fixes.patch fs-proc-task_mmuc-cleanup-the-tail_vma-horror-in-m_next.patch fs-proc-task_mmuc-shift-priv-task-=-null-from-m_start-to-m_stop.patch fs-proc-task_mmuc-kill-the-suboptimal-and-confusing-m-version-logic.patch fs-proc-task_mmuc-simplify-m_start-to-make-it-readable.patch fs-proc-task_mmuc-introduce-m_next_vma-helper.patch fs-proc-task_mmuc-reintroduce-m-version-logic.patch fs-proc-task_mmuc-update-m-version-in-the-main-loop-in-m_start.patch fs-proc-task_nommuc-change-maps_open-to-use-__seq_open_private.patch fs-proc-task_nommuc-shift-mm_access-from-m_start-to-proc_maps_open.patch fs-proc-task_nommuc-shift-mm_access-from-m_start-to-proc_maps_open-checkpatch-fixes.patch fs-proc-task_nommuc-dont-use-priv-task-mm.patch proc-maps-replace-proc_maps_private-pid-with-struct-inode-inode.patch proc-maps-make-vm_is_stack-logic-namespace-friendly.patch ipc-shm-kill-the-historical-wrong-mm-start_stack-check.patch linux-next.patch -- To unsubscribe from this list: send the line "unsubscribe mm-commits" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html