Skip to contentSkip to conversation

Fix crash when ptrace poking hugepage areas

PATCH David Gibson 3 participants 5 messages Nov 29, 2005

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Bill, does this look like the correct fix for the problem to you? If so, please apply Andrew.

set_page_dirty() will not cope with being handed a page * which is part of a compound page, but not the master page in that compound page. This case can occur via access_process_vm() if you attempt to write to another process's hugepage memory area using ptrace() (causing an oops or hang).

This patch fixes the bug by first resolving the page * to the compound page's master page.

mm/page-writeback.c +6 -1
Index: working-2.6/mm/page-writeback.c
===================================================================
@@ -660,7 +660,12 @@ EXPORT_SYMBOL(redirty_page_for_writepage
  */
 int fastcall set_page_dirty(struct page *page)
 {
-	struct address_space *mapping = page_mapping(page);
+	struct address_space *mapping;
+
+	if (unlikely(PageCompound(page)))
+		page = (struct page *)page_private(page);
+
+	mapping = page_mapping(page);
 
 	if (likely(mapping)) {
 		int (*spd)(struct page *) = mapping->a_ops->set_page_dirty;
Signature · 4 lines
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson
#1 lore raw
Quoted 4 lines · David Gibson

Bill, does this look like the correct fix for the problem to you? If so, please apply Andrew.

Quoted 8 lines

set_page_dirty() will not cope with being handed a page * which is part of a compound page, but not the master page in that compound page. This case can occur via access_process_vm() if you attempt to write to another process's hugepage memory area using ptrace() (causing an oops or hang).

This patch fixes the bug by first resolving the page * to the compound page's master page.

We already have to handle this situation for direct-io read()s into hugepages. bio_set_pages_dirty() does

		if (page && !PageCompound(page))
			set_page_dirty_lock(page);

It's such a rare case that it's probably best to continue to do this in the caller rather than in the callee. That's access_process_vm().

Unless there's a reason why we actually want the compound page to be marked dirty? If there is, then direct-io has a problem.

#2 lore raw
Quoted 14 lines · Andrew Morton
Quoted 4 lines · David Gibson

Bill, does this look like the correct fix for the problem to you? If so, please apply Andrew.

Quoted 8 lines

set_page_dirty() will not cope with being handed a page * which is part of a compound page, but not the master page in that compound page. This case can occur via access_process_vm() if you attempt to write to another process's hugepage memory area using ptrace() (causing an oops or hang).

This patch fixes the bug by first resolving the page * to the compound page's master page.

Quoted 8 lines

We already have to handle this situation for direct-io read()s into hugepages. bio_set_pages_dirty() does

		if (page && !PageCompound(page))
			set_page_dirty_lock(page);

It's such a rare case that it's probably best to continue to do this in the caller rather than in the callee. That's access_process_vm().

Good call, revised patch below.

Quoted 2 lines

Unless there's a reason why we actually want the compound page to be marked dirty? If there is, then direct-io has a problem.

I don't think so. Since hugepages are never disk-backed, I think the PageDirty flag is more or less irrelevant.

Fix crash when ptrace poking hugepage areas

set_page_dirty() will not cope with being handed a page * which is part of a compound page, but not the master page in that compound page. This case can occur via access_process_vm() if you attemp to write to another process's hugepage memory area using ptrace() (causing an oops or hang).

This patch fixes the bug by only calling set_page_dirty() from access_process_vm() if the page is not a compound page. We already use a similar fix in bio_set_pages_dirty() for the case of direct io to hugepages.

kernel/ptrace.c +2 -1
Index: working-2.6/kernel/ptrace.c
===================================================================
@@ -241,7 +241,8 @@ int access_process_vm(struct task_struct
 		if (write) {
 			copy_to_user_page(vma, page, addr,
 					  maddr + offset, buf, bytes);
-			set_page_dirty_lock(page);
+			if (!PageCompound(page))
+				set_page_dirty_lock(page);
 		} else {
 			copy_from_user_page(vma, page, addr,
 					    buf, maddr + offset, bytes);
Signature · 4 lines
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson
#4 lore raw
In reply to David Gibson
Quoted 10 lines · David Gibson

Bill, does this look like the correct fix for the problem to you? If so, please apply Andrew.
set_page_dirty() will not cope with being handed a page * which is part of a compound page, but not the master page in that compound page. This case can occur via access_process_vm() if you attempt to write to another process's hugepage memory area using ptrace() (causing an oops or hang).
This patch fixes the bug by first resolving the page * to the compound page's master page.

akpm had already responded, but my general response would have been "Why on earth would you mark a hugepage dirty?" or similar.

-- wli