Re: [PATCH 18/26] mm: Convert process_vm_rw_pages() to use get_user_pages_unlocked()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



(10/2/13 3:36 PM), Jan Kara wrote:
On Wed 02-10-13 12:32:33, KOSAKI Motohiro wrote:
(10/2/13 10:27 AM), Jan Kara wrote:
Signed-off-by: Jan Kara <jack@xxxxxxx>
---
   mm/process_vm_access.c | 8 ++------
   1 file changed, 2 insertions(+), 6 deletions(-)

diff --git a/mm/process_vm_access.c b/mm/process_vm_access.c
index fd26d0433509..c1bc47d8ed90 100644
--- a/mm/process_vm_access.c
+++ b/mm/process_vm_access.c
@@ -64,12 +64,8 @@ static int process_vm_rw_pages(struct task_struct *task,
   	*bytes_copied = 0;

   	/* Get the pages we're interested in */
-	down_read(&mm->mmap_sem);
-	pages_pinned = get_user_pages(task, mm, pa,
-				      nr_pages_to_copy,
-				      vm_write, 0, process_pages, NULL);
-	up_read(&mm->mmap_sem);
-
+	pages_pinned = get_user_pages_unlocked(task, mm, pa, nr_pages_to_copy,
+					       vm_write, 0, process_pages);
   	if (pages_pinned != nr_pages_to_copy) {
   		rc = -EFAULT;
   		goto end;

This is wrong because original code is wrong. In this function, page may
be pointed to anon pages. Then, you should keep to take mmap_sem until
finish to copying. Otherwise concurrent fork() makes nasty COW issue.
   Hum, can you be more specific? I suppose you are speaking about situation
when the remote task we are copying data from/to does fork while
process_vm_rw_pages() runs. If we are copying data from remote task, I
don't see how COW could cause any problem. If we are copying to remote task
and fork happens after get_user_pages() but before copy_to_user() then I
can see we might be having some trouble. copy_to_user() would then copy
data into both original remote process and its child thus essentially
bypassing COW. If the child process manages to COW some of the pages before
copy_to_user() happens, it can even see only some of the pages. Is that what
you mean?

scenario 1: vm_write==0

Process P1 call get_user_pages(pa, process_pages) in process_vm_rw_pages
P1 unlock mmap_sem.
Process P2 call fork(). and make P3.
P2 write memory pa. now the "process_pages" is owned by P3 (the child process)
P3 write memory pa. and then the content of "process_pages" is changed.
P1 read process_pages as P2's page. but actually, it is P3's data. Then, P1 observe garbage, at least unintended, data was read.


scenario 2: vm_write==1

Process P1 call get_user_pages(pa, process_pages) in process_vm_rw_pages. It makes COW break and any anon page sharing broke.
P1 unlock mmap_sem.
P2 call fork(). and make P3. And, now COW page sharing is restored.
P2 write memory pa. now the "process_pages" is owned by P3.
P3 write memory pa. it mean P3 changes "process_pages".
P1 write process_pages as P2's page. but actually, it is P3's. It override above P3's write and then P3 observe data corruption.



The solution is to keep holding mmap_sem until finishing process_pages access
because mmap_sem prevent fork. and then race never be happen. I mean you cann't use
get_user_pages_unlock() if target address point to anon pages.

I'm not sure these story match your explanation.


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]