[PATCH 3/8] mm/swap: prevent concurrent swapon on the same S_ISBLK blockdev

[PATCH 3/8] mm/swap: prevent concurrent swapon on the same S_ISBLK blockdev

[Weijie Yang]

When swapon the same S_ISBLK blockdev concurrent, the allocated two
swap_info could hold the same block_device, because claim_swapfile()
allow the same holder(here, it is sys_swapon function).

To prevent this situation, This patch adds swap_lock protect to ensure
we can find this situation and return -EBUSY for one swapon call.

As for S_ISREG swapfile, claim_swapfile() already prevent this scenario
by holding inode->i_mutex.

This patch is just for a rare scenario, aim to correct of code.

Signed-off-by: Weijie Yang <weijie.yang@samsung.com>
---
 mm/swapfile.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/mm/swapfile.c b/mm/swapfile.c
index 4d24158..413c213 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -2459,9 +2459,10 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
 		goto bad_swap;
 	}
 
+	/* prevent concurrent swapon on the same S_ISBLK blockdev */
+	spin_lock(&swap_lock);
 	p->swap_file = swap_file;
 	mapping = swap_file->f_mapping;
-
 	for (i = 0; i < nr_swapfiles; i++) {
 		struct swap_info_struct *q = swap_info[i];
 
@@ -2472,6 +2473,7 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
 			goto bad_swap;
 		}
 	}
+	spin_unlock(&swap_lock);
 
 	inode = mapping->host;
 	/* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
-- 
1.7.10.4

Re: [PATCH 3/8] mm/swap: prevent concurrent swapon on the same S_ISBLK blockdev

[Andrew Morton]

On Mon, 27 Jan 2014 18:03:04 +0800 Weijie Yang <weijie.yang@samsung.com> wrote:

> When swapon the same S_ISBLK blockdev concurrent, the allocated two
> swap_info could hold the same block_device, because claim_swapfile()
> allow the same holder(here, it is sys_swapon function).
> 
> To prevent this situation, This patch adds swap_lock protect to ensure
> we can find this situation and return -EBUSY for one swapon call.
> 
> As for S_ISREG swapfile, claim_swapfile() already prevent this scenario
> by holding inode->i_mutex.
> 
> This patch is just for a rare scenario, aim to correct of code.
> 

hm, OK.  Would it be saner to pass a unique `holder' to
claim_swapfile()?  Say, `p'?

Truly, I am fed up with silly swapon/swapoff races.  How often does
anyone call these things?  Let's slap a huge lock around the whole
thing and be done with it?

Re: [PATCH 3/8] mm/swap: prevent concurrent swapon on the same S_ISBLK blockdev

[Hugh Dickins]

On Mon, 3 Feb 2014, Andrew Morton wrote:
> On Mon, 27 Jan 2014 18:03:04 +0800 Weijie Yang <weijie.yang@samsung.com> wrote:
> 
> > When swapon the same S_ISBLK blockdev concurrent, the allocated two
> > swap_info could hold the same block_device, because claim_swapfile()
> > allow the same holder(here, it is sys_swapon function).
> > 
> > To prevent this situation, This patch adds swap_lock protect to ensure
> > we can find this situation and return -EBUSY for one swapon call.
> > 
> > As for S_ISREG swapfile, claim_swapfile() already prevent this scenario
> > by holding inode->i_mutex.
> > 
> > This patch is just for a rare scenario, aim to correct of code.
> > 
> 
> hm, OK.  Would it be saner to pass a unique `holder' to
> claim_swapfile()?  Say, `p'?
> 
> Truly, I am fed up with silly swapon/swapoff races.  How often does
> anyone call these things?  Let's slap a huge lock around the whole
> thing and be done with it?

That answer makes me sad: we can't be bothered to get it right,
even when Weijie goes to the trouble of presenting a series to do so.
But I sure don't deserve a vote until I've actually looked through it.

Hugh