Skip to contentSkip to conversation

Add additional error check to mm/mincore.c

PATCH NAK Bruce Dubbs 2 participants 2 messages Mar 25, 2007

lore
Share & tools
lore

Bruce Dubbs

author
#0 lore raw

PATCH 1/1 Add additional error check to mm/mincore.c

I some circumstances, mincore can succeed when it shouldn't.

Example:
Two files are mmapped to a process and they are adjacent in memory.
If mincore is run with a requested length that is too large, the function does not differentiate between the different file pointers within the different vma structures and inappropriately returns success.

The attached patch, against 2.6.20.3, fixes this behavior.

This behavior was found when running the Linux Test Project's mincore01 on an IA32 system. Test 3 "unexpectedly" succeeds.

-- Bruce

mincore.c renamed from mincore.c.old +17 -3
@@ -43,7 +43,8 @@
  * all the arguments, we hold the mmap semaphore: we should
  * just return the amount of info we're asked for.
  */
-static long do_mincore(unsigned long addr, unsigned char *vec, unsigned long pages)
+static long do_mincore(unsigned long addr, unsigned char *vec, unsigned long pages,
+        struct file** file_struct)
 {
 	unsigned long i, nr, pgoff;
 	struct vm_area_struct *vma = find_vma(current->mm, addr);

@@ -64,7 +65,19 @@
 	 * this is what we've traditionally done, so we'll just
 	 * continue doing it.
 	 */
-	if (!vma->vm_file)
+
+    /* 
+     * Initialize file pointer to the value in the first vma structure
+     */
+
+    if ( *file_struct == NULL && vma->vm_file )
+        *file_struct = vma->vm_file;
+
+    /*
+     * Return an error if the is no file mapped of the file is different
+     */
+ 
+	if (!vma->vm_file || vma->vm_file != *file_struct)
 		return -ENOMEM;
 
 	/*

@@ -115,6 +128,7 @@
 	long retval;
 	unsigned long pages;
 	unsigned char *tmp;
+    static struct file* file = NULL;
 
 	/* Check the start address: needs to be page-aligned.. */
  	if (start & ~PAGE_CACHE_MASK)

@@ -142,7 +156,7 @@
 		 * the temporary buffer size.
 		 */
 		down_read(&current->mm->mmap_sem);
-		retval = do_mincore(start, tmp, min(pages, PAGE_SIZE));
+		retval = do_mincore(start, tmp, min(pages, PAGE_SIZE), &file);
 		up_read(&current->mm->mmap_sem);
 
 		if (retval <= 0)
NAK
#1 lore raw
Quoted 7 lines · Bruce Dubbs

I some circumstances, mincore can succeed when it shouldn't.

Example:
Two files are mmapped to a process and they are adjacent in memory.
If mincore is run with a requested length that is too large, the function does not differentiate between the different file pointers within the different vma structures and inappropriately returns success.

You've got this the wrong way round. mincore should succeed in these cases: just like munmap, mprotect, msync, madvise etc do.

Quoted 2 lines

The attached patch, against 2.6.20.3, fixes this behavior.

NAK, sorry.

Quoted 3 lines

This behavior was found when running the Linux Test Project's mincore01 on an IA32 system. Test 3 "unexpectedly" succeeds.

Please follow that up with the LTP guys: I expect they'll quickly agree that the test is wrong and correct it.

It probably dates from when mmap addresses were commonly allocated in ascending order, whereas now they're commonly allocated in descending
order: neither being a rule, neither to be assumed. (Ah, I see Nick has already suggested this memory layout issue to you.)

Another thing to keep in mind is that 2.6.21-rc fixes mincore to succeed on anonymous areas, where before it failed for no very good reason. That can't be the issue here, since your report is against earlier releases; but it is something for the test fixer to note.

Thanks,
Hugh

Quoted 54 lines

-- Bruce

mincore.c renamed from mincore.c.old +17 -3
@@ -43,7 +43,8 @@
  * all the arguments, we hold the mmap semaphore: we should
  * just return the amount of info we're asked for.
  */
-static long do_mincore(unsigned long addr, unsigned char *vec, unsigned long pages)
+static long do_mincore(unsigned long addr, unsigned char *vec, unsigned long pages,
+        struct file** file_struct)
 {
 	unsigned long i, nr, pgoff;
 	struct vm_area_struct *vma = find_vma(current->mm, addr);

@@ -64,7 +65,19 @@
 	 * this is what we've traditionally done, so we'll just
 	 * continue doing it.
 	 */
-	if (!vma->vm_file)
+
+    /* 
+     * Initialize file pointer to the value in the first vma structure
+     */
+
+    if ( *file_struct == NULL && vma->vm_file )
+        *file_struct = vma->vm_file;
+
+    /*
+     * Return an error if the is no file mapped of the file is different
+     */
+ 
+	if (!vma->vm_file || vma->vm_file != *file_struct)
 		return -ENOMEM;
 
 	/*

@@ -115,6 +128,7 @@
 	long retval;
 	unsigned long pages;
 	unsigned char *tmp;
+    static struct file* file = NULL;
 
 	/* Check the start address: needs to be page-aligned.. */
  	if (start & ~PAGE_CACHE_MASK)

@@ -142,7 +156,7 @@
 		 * the temporary buffer size.
 		 */
 		down_read(&current->mm->mmap_sem);
-		retval = do_mincore(start, tmp, min(pages, PAGE_SIZE));
+		retval = do_mincore(start, tmp, min(pages, PAGE_SIZE), &file);
 		up_read(&current->mm->mmap_sem);
 
 		if (retval <= 0)