Re: [PATCH] Support ia32 exec domains without CONFIG_IA32_SUPPORT

[Arun Sharma]

David Woodhouse wrote:

> If you make it generic so that qemu can use it for emulating i386 even
> on machines like ppc64, perhaps it would be saner.

There is already an architecture independent solution to the problem - 
set_fs_altroot(). The only problem is that it's not easily accessible 
from userland. Accessing it via set_personality() works only if there 
are multiple exec domains.

I can't see how anyone would accept a patch that adds linux-i386 exec 
domain in generic code.

Christoph doesn't like the idea of adding exec-domains just for this 
purpose and has suggested adding a new system call to set the altroot. A 
prototype patch to do this already exists. I will be cleaning it up and 
 posting it to LKML later this week. The main purpose of moving the 
discussion to LKML was to see how receptive people were to the proposed 
new system call.

So far I haven't heard anyone complaining :)

	-Arun

PS: Prototype patch with known bugs attached.
Index: linux-2.6-cvs/arch/ia64/kernel/entry.S
===================================================================
RCS file: /home/adsharma/disk2/cvs/linux-2.5/arch/ia64/kernel/entry.S,v
retrieving revision 1.49
diff -u -r1.49 entry.S
--- linux-2.6-cvs/arch/ia64/kernel/entry.S	24 Aug 2004 18:27:07 -0000	1.49
+++ linux-2.6-cvs/arch/ia64/kernel/entry.S	8 Oct 2004 17:35:53 -0000
@@ -1527,7 +1527,7 @@
 	data8 sys_mq_getsetattr
 	data8 sys_ni_syscall			// reserved for kexec_load
 	data8 sys_ni_syscall
-	data8 sys_ni_syscall			// 1270
+	data8 sys_altroot			// 1270
 	data8 sys_ni_syscall
 	data8 sys_ni_syscall
 	data8 sys_ni_syscall
Index: linux-2.6-cvs/fs/namei.c
===================================================================
RCS file: /home/adsharma/disk2/cvs/linux-2.5/fs/namei.c,v
retrieving revision 1.110
diff -u -r1.110 namei.c
--- linux-2.6-cvs/fs/namei.c	2 Oct 2004 17:59:55 -0000	1.110
+++ linux-2.6-cvs/fs/namei.c	8 Oct 2004 18:14:11 -0000
@@ -897,7 +897,7 @@
 	return 1;
 }
 
-void set_fs_altroot(void)
+int set_fs_altroot(const char __user *altroot)
 {
 	char *emul = __emul_prefix();
 	struct nameidata nd;
@@ -905,12 +905,20 @@
 	struct dentry *dentry = NULL, *olddentry;
 	int err;
 
+	if (altroot) {
+		emul = getname(altroot);
+		if (IS_ERR(emul))
+			return PTR_ERR(emul);
+	}
+
 	if (!emul)
 		goto set_it;
 	err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
 	if (!err) {
 		mnt = nd.mnt;
 		dentry = nd.dentry;
+	} else {
+		return err;
 	}
 set_it:
 	write_lock(&current->fs->lock);
@@ -923,6 +931,33 @@
 		dput(olddentry);
 		mntput(oldmnt);
 	}
+	return 0;
+}
+
+asmlinkage int sys_altroot(const char __user * altroot)
+{
+	if (atomic_read(&current->fs->count) != 1) {
+		struct fs_struct *fsp, *ofsp;
+
+		fsp = copy_fs_struct(current->fs);
+		if (fsp == NULL) {
+			return -ENOMEM;
+		}
+
+		task_lock(current);
+		ofsp = current->fs;
+		current->fs = fsp;
+		task_unlock(current);
+
+		put_fs_struct(ofsp);
+	}
+
+	/*
+	 * At that point we are guaranteed to be the sole owner of
+	 * current->fs.
+	 */
+
+	return set_fs_altroot(altroot);
 }
 
 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
Index: linux-2.6-cvs/fs/open.c
===================================================================
RCS file: /home/adsharma/disk2/cvs/linux-2.5/fs/open.c,v
retrieving revision 1.71
diff -u -r1.71 open.c
--- linux-2.6-cvs/fs/open.c	8 Aug 2004 01:54:18 -0000	1.71
+++ linux-2.6-cvs/fs/open.c	8 Oct 2004 17:38:03 -0000
@@ -584,7 +584,7 @@
 		goto dput_and_out;
 
 	set_fs_root(current->fs, nd.mnt, nd.dentry);
-	set_fs_altroot();
+	set_fs_altroot(NULL);
 	error = 0;
 dput_and_out:
 	path_release(&nd);
Index: linux-2.6-cvs/include/linux/fs_struct.h
===================================================================
RCS file: /home/adsharma/disk2/cvs/linux-2.5/include/linux/fs_struct.h,v
retrieving revision 1.7
diff -u -r1.7 fs_struct.h
--- linux-2.6-cvs/include/linux/fs_struct.h	17 Nov 2002 03:52:23 -0000	1.7
+++ linux-2.6-cvs/include/linux/fs_struct.h	8 Oct 2004 17:39:16 -0000
@@ -19,7 +19,7 @@
 }
 
 extern void exit_fs(struct task_struct *);
-extern void set_fs_altroot(void);
+extern int set_fs_altroot(const char __user *);
 extern void set_fs_root(struct fs_struct *, struct vfsmount *, struct dentry *);
 extern void set_fs_pwd(struct fs_struct *, struct vfsmount *, struct dentry *);
 extern struct fs_struct *copy_fs_struct(struct fs_struct *);
Index: linux-2.6-cvs/kernel/exec_domain.c
===================================================================
RCS file: /home/adsharma/disk2/cvs/linux-2.5/kernel/exec_domain.c,v
retrieving revision 1.18
diff -u -r1.18 exec_domain.c
--- linux-2.6-cvs/kernel/exec_domain.c	8 Sep 2004 14:50:53 -0000	1.18
+++ linux-2.6-cvs/kernel/exec_domain.c	8 Oct 2004 17:33:00 -0000
@@ -167,7 +167,7 @@
 	current->personality = personality;
 	oep = current_thread_info()->exec_domain;
 	current_thread_info()->exec_domain = ep;
-	set_fs_altroot();
+	set_fs_altroot(NULL);
 
 	module_put(oep->module);
 	return 0;