Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932726AbaJWC0R (ORCPT ); Wed, 22 Oct 2014 22:26:17 -0400 Received: from x35.xmailserver.org ([64.71.152.41]:47704 "EHLO x35.xmailserver.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754175AbaJWC0I (ORCPT ); Wed, 22 Oct 2014 22:26:08 -0400 X-AuthUser: davidel@xmailserver.org Date: Wed, 22 Oct 2014 19:26:07 -0700 (PDT) From: Davide Libenzi X-X-Sender: davide@davide-lnx3 To: Linux Kernel Mailing List cc: Andrew Morton , Hugh Dickins , KOSAKI Motohiro , linux-mm@kvack.org Subject: [patch][resend] MAP_HUGETLB munmap fails with size not 2MB aligned Message-ID: User-Agent: Alpine 2.10 (DEB 1266 2009-07-14) X-GPG-FINGRPRINT: CFAE 5BEE FD36 F65E E640 56FE 0974 BF23 270F 474E X-GPG-PUBLIC_KEY: http://www.xmailserver.org/davidel.asc MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org [Resending with proper CC list suggested by Andrew] Calling munmap on a MAP_HUGETLB area, and a size which is not 2MB aligned, causes munmap to fail. Tested on 3.13.x but tracking back to 3.2.x. In do_munmap() we forcibly want a 4KB default page, and we wrongly calculate the end of the map. Since the calculated end is within the end address of the target vma, we try to do a split with an address right in the middle of a huge page, which would fail with EINVAL. Tentative (untested) patch and test case attached (be sure you have a few huge pages available via /proc/sys/vm/nr_hugepages tinkering). Signed-Off-By: Davide Libenzi - Davide diff --git a/mm/mmap.c b/mm/mmap.c index 7f85520..6dba257 100644 --- a/mm/mmap.c +++ b/mm/mmap.c @@ -2528,10 +2528,6 @@ int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) return -EINVAL; - len = PAGE_ALIGN(len); - if (len == 0) - return -EINVAL; - /* Find the first overlapping VMA */ vma = find_vma(mm, start); if (!vma) @@ -2539,6 +2535,16 @@ int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) prev = vma->vm_prev; /* we have start < vma->vm_end */ + if (likely(!is_vm_hugetlb_page(vma))) + len = PAGE_ALIGN(len); + else { + unsigned long hpage_size = huge_page_size(hstate_vma(vma)); + + len = ALIGN(len, hpage_size); + } + if (unlikely(len == 0)) + return -EINVAL; + /* if it doesn't overlap, we have nothing.. */ end = start + len; if (vma->vm_start >= end) [hugebug.c] #include #include #include #include #include static void test(int flags, size_t size) { void* addr = mmap(NULL, size, PROT_READ | PROT_WRITE, flags | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); if (addr == MAP_FAILED) { perror("mmap"); exit(1); } *(char*) addr = 17; if (munmap(addr, size) != 0) { perror("munmap"); exit(1); } } int main(int ac, const char** av) { static const size_t hugepage_size = 2 * 1024 * 1024; printf("Testing normal pages with 2MB size ...\n"); test(0, hugepage_size); printf("OK\n"); printf("Testing huge pages with 2MB size ...\n"); test(MAP_HUGETLB, hugepage_size); printf("OK\n"); printf("Testing normal pages with 4KB byte size ...\n"); test(0, 4096); printf("OK\n"); printf("Testing huge pages with 4KB byte size ...\n"); test(MAP_HUGETLB, 4096); printf("OK\n"); printf("Testing normal pages with 1 byte size ...\n"); test(0, 1); printf("OK\n"); printf("Testing huge pages with 1 byte size ...\n"); test(MAP_HUGETLB, 1); printf("OK\n"); return 0; } -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/