Mel Gorman wrote: > On Thu, Mar 25, 2010 at 12:18:05AM -0500, Andrew Hastings wrote: >> It seems to me that hugetlbfs ought to take an extra reference on the vma >> or vm_file or f_mapping or _something_ if vma->vm_file->f_mapping is needed >> by free_huge_page(). > > Again, I haven't looked closely at this but a reference count on the VMA > wouldn't help. After all, the VMAs have already been cleared up and the > page tables. As far as the code is concerned, that file is no longer in > use. I'd also not try reference counting during get_user_pages and > someohw releasing that count later. Too much mess. > > The most likely avenue is to store a reference to the superblock instead > of the mapping in page->private which is what put_quota is really > interested in. There might still be a race there if hugetlbfs managed to > get unmounted before the pages were freed though - not 100% sure. The hugetlbfs_sb_info struct that holds the quota is allocated separately from the superblock. Would it make sense for page->private to point directly to hugetlbfs_sb_info, and reference count hugetlbfs_sb_info instead? Seems like this would avoid the unmount race. -Andrew Hastings Cray Inc. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>