Re: [PATCH] Support ia32 exec domains without CONFIG_IA32_SUPPORT

[Arun Sharma]

Arun Sharma wrote:

> 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.

Attached is the promised patch. It addresses Christoph's comments and 
fixes the bug Tony found as well.

	-Arun


Add a new system call sys_altroot. This allows using the 
altroot feature on systems where there is only one exec domain.

Signed-off-by: Zou Nanhai <nanhai.zou@xxxxxxxxx>
Signed-off-by: Gordon Jin <gordon.jin@xxxxxxxxx>
Signed-off-by: Arun Sharma <arun.sharma@xxxxxxxxx>

diff -Nraup a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
--- a/arch/ia64/kernel/entry.S	2004-10-12 09:56:51.408496174 -0700
+++ b/arch/ia64/kernel/entry.S	2004-10-12 09:58:17.362596684 -0700
@@ -1527,7 +1527,7 @@ sys_call_table:
 	data8 sys_mq_getsetattr
 	data8 sys_ni_syscall			// reserved for kexec_load
 	data8 sys_ni_syscall
-	data8 sys_ni_syscall			// 1270
+	data8 sys_setaltroot			// 1270
 	data8 sys_ni_syscall
 	data8 sys_ni_syscall
 	data8 sys_ni_syscall
diff -Nraup a/fs/namei.c b/fs/namei.c
--- a/fs/namei.c	2004-10-12 09:56:56.895800795 -0700
+++ b/fs/namei.c	2004-10-12 09:58:33.524705861 -0700
@@ -897,20 +897,20 @@ static int __emul_lookup_dentry(const ch
 	return 1;
 }
 
-void set_fs_altroot(void)
+int __set_fs_altroot(const char *altroot)
 {
-	char *emul = __emul_prefix();
 	struct nameidata nd;
 	struct vfsmount *mnt = NULL, *oldmnt;
 	struct dentry *dentry = NULL, *olddentry;
 	int err;
-
-	if (!emul)
+	if (!altroot)
 		goto set_it;
-	err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
+	err = path_lookup(altroot, 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 +923,58 @@ set_it:
 		dput(olddentry);
 		mntput(oldmnt);
 	}
+	return 0;
+}
+
+void set_fs_altroot()
+{
+	char *emul = __emul_prefix();
+
+	__set_fs_altroot(emul);
+}
+
+asmlinkage long sys_setaltroot(const char __user * altroot)
+{
+	char *emul = NULL;
+	int ret;
+
+	if (altroot) {
+		emul = getname(altroot);
+		if (IS_ERR(emul)) {
+			ret = PTR_ERR(emul);
+			goto out;
+		}
+	}
+
+	if (atomic_read(&current->fs->count) != 1) {
+		struct fs_struct *fsp, *ofsp;
+
+		fsp = copy_fs_struct(current->fs);
+		if (fsp == NULL) {
+			ret = -ENOMEM;
+			goto out_putname;
+		}
+
+		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.
+	 */
+
+	ret = __set_fs_altroot(emul);
+
+out_putname:
+	if (emul)
+		putname(emul);
+out:
+	return ret;
 }
 
 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
diff -Nraup a/include/linux/syscalls.h b/include/linux/syscalls.h
--- a/include/linux/syscalls.h	2004-10-12 09:56:58.124316405 -0700
+++ b/include/linux/syscalls.h	2004-10-12 09:58:17.362596684 -0700
@@ -489,6 +489,7 @@ asmlinkage long sys_nfsservctl(int cmd,
 				void __user *res);
 asmlinkage long sys_syslog(int type, char __user *buf, int len);
 asmlinkage long sys_uselib(const char __user *library);
+asmlinkage long sys_setaltroot(const char __user *altroot);
 asmlinkage long sys_ni_syscall(void);
 
 #endif