From 3fa30460ea502133a18a07b14452cd660906f16f Mon Sep 17 00:00:00 2001 From: David Howells Date: Tue, 23 Mar 2010 13:35:21 -0700 Subject: nommu: fix an incorrect comment in the do_mmap_shared_file() Fix an incorrect comment in the do_mmap_shared_file(). If a mapping is requested MAP_SHARED, then a private copy cannot be made and still provide correct semantics. Signed-off-by: David Howells Reported-by: Dave Hudson Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/nommu.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'mm/nommu.c') diff --git a/mm/nommu.c b/mm/nommu.c index 605ace8982a8..e4b8f4d28a3f 100644 --- a/mm/nommu.c +++ b/mm/nommu.c @@ -1040,10 +1040,9 @@ static int do_mmap_shared_file(struct vm_area_struct *vma) if (ret != -ENOSYS) return ret; - /* getting an ENOSYS error indicates that direct mmap isn't - * possible (as opposed to tried but failed) so we'll fall - * through to making a private copy of the data and mapping - * that if we can */ + /* getting -ENOSYS indicates that direct mmap isn't possible (as + * opposed to tried but failed) so we can only give a suitable error as + * it's not possible to make a private copy if MAP_SHARED was given */ return -ENODEV; } -- cgit v1.2.2 From 7561e8ca0dfaf6fca3feef982830de3b65300e5b Mon Sep 17 00:00:00 2001 From: David Howells Date: Thu, 25 Mar 2010 16:48:38 +0000 Subject: NOMMU: Revert 'nommu: get_user_pages(): pin last page on non-page-aligned start' Revert the following patch: commit c08c6e1f54c85fc299cf9f88cf330d6dd28a9a1d Author: Steven J. Magnani Date: Fri Mar 5 13:42:24 2010 -0800 nommu: get_user_pages(): pin last page on non-page-aligned start As it assumes that the mappings begin at the start of pages - something that isn't necessarily true on NOMMU systems. On NOMMU systems, it is possible for a mapping to only occupy part of the page, and not necessarily touch either end of it; in fact it's also possible for multiple non-overlapping mappings to coexist on one page (consider direct mappings of ROMFS files, for example). Signed-off-by: David Howells Acked-by: Steven J. Magnani Signed-off-by: Linus Torvalds --- mm/nommu.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'mm/nommu.c') diff --git a/mm/nommu.c b/mm/nommu.c index e4b8f4d28a3f..089982f5a4cf 100644 --- a/mm/nommu.c +++ b/mm/nommu.c @@ -146,7 +146,7 @@ int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); for (i = 0; i < nr_pages; i++) { - vma = find_extend_vma(mm, start); + vma = find_vma(mm, start); if (!vma) goto finish_or_fault; @@ -764,7 +764,7 @@ EXPORT_SYMBOL(find_vma); */ struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) { - return find_vma(mm, addr & PAGE_MASK); + return find_vma(mm, addr); } /* -- cgit v1.2.2 From e1ee65d85904c5dd4b9cea1b15d5e85e20eae8a1 Mon Sep 17 00:00:00 2001 From: David Howells Date: Thu, 25 Mar 2010 16:48:44 +0000 Subject: NOMMU: Fix __get_user_pages() to pin last page on offset buffers Fix __get_user_pages() to make it pin the last page on a buffer that doesn't begin at the start of a page, but is a multiple of PAGE_SIZE in size. The problem is that __get_user_pages() advances the pointer too much when it iterates to the next page if the page it's currently looking at isn't used from the first byte. This can cause the end of a short VMA to be reached prematurely, resulting in the last page being lost. Signed-off-by: Steven J. Magnani Signed-off-by: David Howells Signed-off-by: Linus Torvalds --- mm/nommu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'mm/nommu.c') diff --git a/mm/nommu.c b/mm/nommu.c index 089982f5a4cf..63fa17d121f0 100644 --- a/mm/nommu.c +++ b/mm/nommu.c @@ -162,7 +162,7 @@ int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, } if (vmas) vmas[i] = vma; - start += PAGE_SIZE; + start = (start + PAGE_SIZE) & PAGE_MASK; } return i; -- cgit v1.2.2