[PATCH][2.6] No swapping on memory backed swapfiles

[PATCH][2.6] No swapping on memory backed swapfiles

[Zwane Mwaikambo]

This fixes an oops when running LTP on a system with a tmpfs /tmp. The 
oops is from the lack of a readpage a_op. This patch simply disables 
enabling of swap for memory backed swapfiles.

Unable to handle kernel NULL pointer dereference at virtual address 00000000
 printing eip:
00000000
*pde = 00000000
Oops: 0000 [#1]
CPU:    0
EIP:    0060:[<00000000>]    Not tainted
EFLAGS: 00010246
EIP is at 0x0
eax: 00000000   ebx: c1076458   ecx: c1651660   edx: c1076458
esi: 00000000   edi: e2c38f68   ebp: dfd93f24   esp: e202bf4c
ds: 007b   es: 007b   ss: 0068
Process swapon02 (pid: 5718, threadinfo=e202a000 task=d6f769d0)
Stack: c0145c0c e2c38f68 c1076458 dfd93e88 dfd93e88 00000000 00000000 c01658c8
       dfd93f24 00000000 00000000 e2c38f68 e3d7fe58 c07430d0 dfd93e88 dfd93e88
       00000000 00000001 00000000 00000000 ffffffea dfd93f24 e2c38f68 00000000
Call Trace:
 [<c0145c0c>] read_cache_page+0x5c/0x360
 [<c01658c8>] sys_swapon+0x528/0xa30
 [<c01096e9>] sysenter_past_esp+0x52/0x79

Code:  Bad EIP value.

(gdb) list  *read_cache_page+0x5c
0xc0145c0c is in read_cache_page (mm/filemap.c:1393).
1388                            page_cache_release(cached_page);
1389                            return ERR_PTR(err);
1390                    }
1391                    page = cached_page;
1392                    cached_page = NULL;
1393                    err = filler(data, page);
1394                    if (err < 0) {
1395                            page_cache_release(page);
1396                            page = ERR_PTR(err);
1397                    }

Index: linux-2.6.0-test7/mm/swapfile.c
===================================================================
RCS file: /build/cvsroot/linux-2.6.0-test7/mm/swapfile.c,v
retrieving revision 1.1.1.1
diff -u -p -B -r1.1.1.1 swapfile.c
--- linux-2.6.0-test7/mm/swapfile.c	9 Oct 2003 05:56:12 -0000	1.1.1.1
+++ linux-2.6.0-test7/mm/swapfile.c	13 Oct 2003 07:27:28 -0000
@@ -1233,6 +1233,7 @@ asmlinkage long sys_swapon(const char __
 	struct page *page = NULL;
 	struct inode *inode;
 	struct inode *downed_inode = NULL;
+	struct backing_dev_info *bdi;
 
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
@@ -1293,6 +1294,10 @@ asmlinkage long sys_swapon(const char __
 	}
 
 	error = -EINVAL;
+	bdi = mapping->backing_dev_info;
+	if (bdi->memory_backed)
+		goto bad_swap;
+
 	if (S_ISBLK(inode->i_mode)) {
 		bdev = inode->i_bdev;
 		error = bd_claim(bdev, sys_swapon);

Re: [PATCH][2.6] No swapping on memory backed swapfiles

[Andrew Morton]

Zwane Mwaikambo <zwane@arm.linux.org.uk> wrote:
>
> +	bdi = mapping->backing_dev_info;
>  +	if (bdi->memory_backed)
>  +		goto bad_swap;
>  +

I guess that makes sense, although someone might want to swap onto a
ramdisk-backed file just for some testing purpose.

Why not simply test for a null ->readpage()?

Re: [PATCH][2.6] No swapping on memory backed swapfiles

[John Bradford]

Quote from Andrew Morton <akpm@osdl.org>:
> Zwane Mwaikambo <zwane@arm.linux.org.uk> wrote:
> >
> > +	bdi = mapping->backing_dev_info;
> >  +	if (bdi->memory_backed)
> >  +		goto bad_swap;
> >  +
> 
> I guess that makes sense, although someone might want to swap onto a
> ramdisk-backed file just for some testing purpose.

Or because some RAM is slower than the rest.  This came up a while ago
on the list.

John.

Re: [PATCH][2.6] No swapping on memory backed swapfiles

[Zwane Mwaikambo]

On Mon, 13 Oct 2003, John Bradford wrote:

> > I guess that makes sense, although someone might want to swap onto a
> > ramdisk-backed file just for some testing purpose.
> 
> Or because some RAM is slower than the rest.  This came up a while ago
> on the list.

Could you possibly be referring to something else? How would this slower 
RAM be setup?

Re: [PATCH][2.6] No swapping on memory backed swapfiles

[bill davidsen]

In article <200310130832.h9D8WJ4g000157@81-2-122-30.bradfords.org.uk>,
John Bradford  <john@grabjohn.com> wrote:
| Quote from Andrew Morton <akpm@osdl.org>:
| > Zwane Mwaikambo <zwane@arm.linux.org.uk> wrote:
| > >
| > > +	bdi = mapping->backing_dev_info;
| > >  +	if (bdi->memory_backed)
| > >  +		goto bad_swap;
| > >  +
| > 
| > I guess that makes sense, although someone might want to swap onto a
| > ramdisk-backed file just for some testing purpose.
| 
| Or because some RAM is slower than the rest.  This came up a while ago
| on the list.

Something on my "learn how to..." list, I have some systems which are
setup to cache only the first 64MB, and I bet they would run a lot
faster if the rest were used as swap. It is definitely faster with
mem=64 than letting the CPU beat the whole memory.
-- 
bill davidsen <davidsen@tmr.com>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.

Re: [PATCH][2.6] No swapping on memory backed swapfiles

[Zwane Mwaikambo]

On Mon, 13 Oct 2003, Andrew Morton wrote:

> Zwane Mwaikambo <zwane@arm.linux.org.uk> wrote:
> >
> > +	bdi = mapping->backing_dev_info;
> >  +	if (bdi->memory_backed)
> >  +		goto bad_swap;
> >  +
> 
> I guess that makes sense, although someone might want to swap onto a
> ramdisk-backed file just for some testing purpose.
> 
> Why not simply test for a null ->readpage()?

That was my initial intention, but then i wondered why someone would be 
swapping to a memory backed filesystem in the first place. Also things 
like ramfs have ->readpage().

Re: [PATCH][2.6] No swapping on memory backed swapfiles

[Andrew Morton]

Zwane Mwaikambo <zwane@arm.linux.org.uk> wrote:
>
> On Mon, 13 Oct 2003, Andrew Morton wrote:
> 
> > Zwane Mwaikambo <zwane@arm.linux.org.uk> wrote:
> > >
> > > +	bdi = mapping->backing_dev_info;
> > >  +	if (bdi->memory_backed)
> > >  +		goto bad_swap;
> > >  +
> > 
> > I guess that makes sense, although someone might want to swap onto a
> > ramdisk-backed file just for some testing purpose.
> > 
> > Why not simply test for a null ->readpage()?
> 
> That was my initial intention, but then i wondered why someone would be 
> swapping to a memory backed filesystem in the first place. Also things 
> like ramfs have ->readpage().

A non-memory-backed fs may still have a NULL ->readpage though (ncpfs?).

I do think it best to just test for ->readpage, and give the user the extra
rope.

Re: [PATCH][2.6] No swapping on memory backed swapfiles

[Zwane Mwaikambo]

On Mon, 13 Oct 2003, Andrew Morton wrote:

> A non-memory-backed fs may still have a NULL ->readpage though (ncpfs?).
> 
> I do think it best to just test for ->readpage, and give the user the extra
> rope.

Ok, here we go;

Index: linux-2.6.0-test7/mm/swapfile.c
===================================================================
RCS file: /build/cvsroot/linux-2.6.0-test7/mm/swapfile.c,v
retrieving revision 1.1.1.1
diff -u -p -B -r1.1.1.1 swapfile.c
--- linux-2.6.0-test7/mm/swapfile.c	9 Oct 2003 05:56:12 -0000	1.1.1.1
+++ linux-2.6.0-test7/mm/swapfile.c	13 Oct 2003 09:23:31 -0000
@@ -1318,6 +1318,10 @@ asmlinkage long sys_swapon(const char __
 	/*
 	 * Read the swap header.
 	 */
+	if (!mapping->a_ops->readpage) {
+		error = -EIO;
+		goto bad_swap;
+	}
 	page = read_cache_page(mapping, 0,
 			(filler_t *)mapping->a_ops->readpage, swap_file);
 	if (IS_ERR(page)) {