integrity: mtime patch for mmap files

PATCH In review Mimi Zohar 2 participants 3 messages Mar 23, 2007 – Mar 26, 2007

Subsystems (2)

Memory management Memory mapping

lore

Share & tools

lore

Mimi Zohar

author

2007-03-23 16:11 UTC #0 lore raw

PATCH 5/7 integrity: mtime patch for mmap files

mmaped files are hashed and hmac'ed properly by EVM after being modified when the file's mtime is updated correctly. Peter Staubach's 'memory mapped files not updating timestamps v2' patch properly updates the mmapped file's mtime. In those cases when an application does not first msync the file before calling munmap, the mtime is eventually updated, but not before file_close is called. This patch forces the mtime for mmaped files to be updated in a timely manner.

Signed-off-by Mimi Zohar <[email protected]>

----

mm/mmap.c +5 -0

Index: linux-2.6.21-rc4-mm1/mm/mmap.c
===================================================================
@@ -1790,6 +1790,7 @@ int do_munmap(struct mm_struct *mm, unsi
 {
 	unsigned long end;
 	struct vm_area_struct *vma, *prev, *last;
+	struct file *file;
 
 	if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
 		return -EINVAL;

@@ -1808,6 +1809,10 @@ int do_munmap(struct mm_struct *mm, unsi
 	if (vma->vm_start >= end)
 		return 0;
 
+	file = vma->vm_file;
+	if (file && test_and_clear_bit(AS_MCTIME, &file->f_mapping->flags))
+		file_update_time(file);
+
 	/*
 	 * If we need to split any vma, do it now to save pain later.
 	 *

Andrew Morton

maintainer

2007-03-25 08:27 UTC #1 lore raw

Quoted 7 lines · On Fri, 23 Mar 2007 12:09:58 -0400 Mimi Zohar <[email protected]> wrote:

mmaped files are hashed and hmac'ed properly by EVM after being modified when the file's mtime is updated correctly. Peter Staubach's 'memory mapped files not updating timestamps v2' patch properly updates the mmapped file's mtime. In those cases when an application does not first msync the file before calling munmap, the mtime is eventually updated, but not before file_close is called. This patch forces the mtime for mmaped files to be updated in a timely manner.

I don't think we've ever had that patch in -mm?

Quoted 15 lines

Signed-off-by Mimi Zohar <[email protected]>
----
Index: linux-2.6.21-rc4-mm1/mm/mmap.c
===================================================================
mm/mmap.c +1 -0
@@ -1790,6 +1790,7 @@ int do_munmap(struct mm_struct *mm, unsi
 {
 	unsigned long end;
 	struct vm_area_struct *vma, *prev, *last;
+	struct file *file;
 
 	if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
 		return -EINVAL;

mm/mmap.c int do_munmap(struct mm_struct *mm, unsi · line 1818

Quoted 10 lines

mm/mmap.c +4 -0

@@ -1808,6 +1809,10 @@ int do_munmap(struct mm_struct *mm, unsi
 	if (vma->vm_start >= end)
 		return 0;
 
+	file = vma->vm_file;
+	if (file && test_and_clear_bit(AS_MCTIME, &file->f_mapping->flags))
+		file_update_time(file);
+
 	/*
 	 * If we need to split any vma, do it now to save pain later.
 	 *

Hence this won't compile.

Mimi Zohar

author

2007-03-26 20:50 UTC #2 lore raw

Quoted 11 lines · Andrew Morton

Quoted 7 lines · On Fri, 23 Mar 2007 12:09:58 -0400 Mimi Zohar <[email protected]> wrote:
mmaped files are hashed and hmac'ed properly by EVM after being modified when the file's mtime is updated correctly. Peter Staubach's 'memory mapped files not updating timestamps v2' patch properly updates the mmapped file's mtime. In those cases when an application does not first msync the file before calling munmap, the mtime is eventually updated, but not before file_close is called. This patch forces the mtime for mmaped files to be updated in a timely manner.
I don't think we've ever had that patch in -mm?

We are basing the EVM code on Peter's '[PATCH] memory mapped files not updating timestamps v2' patch http://lkml.org/lkml/2007/2/20/255 on the assumption that it would be accepted. I guess that has not yet happened.

Mimi Zohar