From: David Woodhouse dwmw@amazon.co.uk
commit 6c26bd4384da24841bac4f067741bbca18b0fb74 upstream,
If mas_store_gfp() in the gather loop failed, the 'error' variable that ultimately gets returned was not being set. In many cases, its original value of -ENOMEM was still in place, and that was fine. But if VMAs had been split at the start or end of the range, then 'error' could be zero.
Change to the 'error = foo(); if (error) goto â¦' idiom to fix the bug.
Also clean up a later case which avoided the same bug by *explicitly* setting error = -ENOMEM right before calling the function that might return -ENOMEM.
In a final cosmetic change, move the 'Point of no return' comment to *after* the goto. That's been in the wrong place since the preallocation was removed, and this new error path was added.
Fixes: 606c812eb1d5 ("mm/mmap: Fix error path in do_vmi_align_munmap()") Signed-off-by: David Woodhouse dwmw@amazon.co.uk Cc: stable@vger.kernel.org Reviewed-by: Greg Kroah-Hartman gregkh@linuxfoundation.org Reviewed-by: Liam R. Howlett Liam.Howlett@oracle.com Signed-off-by: David Woodhouse dwmw@amazon.co.uk Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- mm/mmap.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
--- a/mm/mmap.c +++ b/mm/mmap.c @@ -2404,7 +2404,8 @@ do_mas_align_munmap(struct ma_state *mas break; } mas_set_range(&mas_detach, next->vm_start, next->vm_end - 1); - if (mas_store_gfp(&mas_detach, next, GFP_KERNEL)) + error = mas_store_gfp(&mas_detach, next, GFP_KERNEL); + if (error) goto munmap_gather_failed; if (next->vm_flags & VM_LOCKED) locked_vm += vma_pages(next); @@ -2456,6 +2457,7 @@ do_mas_align_munmap(struct ma_state *mas mas_set_range(mas, start, end - 1); } #endif + /* Point of no return */ mas_store_prealloc(mas, NULL);
mm->locked_vm -= locked_vm;