[Devel] [PATCH rh7 13/14] Port diff-ve-fs-allow-to-remove-swapfile-hardlinks

Vladimir Davydov vdavydov at parallels.com
Tue Jun 23 09:29:49 PDT 2015


Author: Maxim Patlasov
Email: MPatlasov at parallels.com
Subject: fs: allow to remove swapfile hardlinks
Date: Tue, 11 Mar 2014 20:33:12 +0400

There is nothing wrong to allow unlink(2) on a swapfile unless this is the last
link (i_nlink == 1).

The patch is useful for sharing a ploop image (which uses the same
S_SWAPFILE flag for protection against truncate as swapfiles) among several
containers by hardlinks. Otherwise, CT destroy would fail due to inability
to unlink image file.

>>>> Nice.
>>>> Why it's not done in upstream?
>>>> I mean, is there any reason against it?
>>>
>>> No, I think. They had solved the problem of lost swapfile by the
>>> simplest patch: disable any unlink.
>>>
>>
>> Are you going to suggest to patch to the upstream?
>> Or it's just our local usage scenario?
>
> It's our own invention -- to use S_SWAPFILE for protection ploop images
> from truncation. Ordinary loop block-device doesn't use it. Nobody else
> (except swapfiles) uses it. So its our local usage from which nobody
> else can get a benefit (until we didn't push ploop upstream in some form).

https://jira.sw.ru/browse/PSBM-25310

Signed-off-by: Maxim Patlasov <MPatlasov at parallels.com>
=============================================================================

Related to https://jira.sw.ru/browse/PSBM-33650

Signed-off-by: Vladimir Davydov <vdavydov at parallels.com>
---
 fs/namei.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/namei.c b/fs/namei.c
index 7bea210fb872..dc156ed16567 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -2458,7 +2458,8 @@ static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
 	if (IS_APPEND(dir))
 		return -EPERM;
 	if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
-	    IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
+	    IS_IMMUTABLE(victim->d_inode) ||
+	    (IS_SWAPFILE(victim->d_inode) && victim->d_inode->i_nlink == 1))
 		return -EPERM;
 	if (isdir) {
 		if (!S_ISDIR(victim->d_inode->i_mode))
-- 
2.1.4




More information about the Devel mailing list