Skip to contentSkip to conversation

Restrict procfs permissions

PATCH In review Rene Scharfe 2 participants 3 messages Jan 29, 2005 Proc filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Restrict procfs permissions

Hi all,

this patch adds a umask option to the proc filesystem. It can be used to restrict the permission of users to view each others process information. E.g. on a multi-user shell server one could use a setting of umask=077 to allow all users to view info about their own processes, only. It should prevent "command line snooping" and generally increases privacy on the server.

Top and ps can cope with such restrictions, they simply are quiet about files they cannot access.

The umask option affects permissions of the numerical directories in /proc, only (the process info). And root can see everything, of course, even with a umask setting of 0777. Default umask is 0, i.e. unchanged permissions.

The patch is inspired by the /proc restriction parts of the GrSecurity patch. The main difference is the ability to configure the restrictions dynamically. You can change the umask setting by running

   # mount -o remount,umask=007 /proc

Testing has been *very* light so far -- it compiles and boots. Patch is against 2.6.11-rc2-bk6.

Comments are very welcome.

Thanks,
Rene

fs/proc/base.c +3 -3
diff -rup linux-2.6.11-rc2-bk6/fs/proc/base.c l/fs/proc/base.c
@@ -1222,7 +1222,7 @@ static struct dentry *proc_pident_lookup
 		goto out;
 
 	ei = PROC_I(inode);
-	inode->i_mode = p->mode;
+	inode->i_mode = p->mode & ~proc_umask;
 	/*
 	 * Yes, it does not scale. And it should not. Don't add
 	 * new entries into /proc/<tgid>/ without very good reasons.
@@ -1537,7 +1537,7 @@ struct dentry *proc_pid_lookup(struct in
 		put_task_struct(task);
 		goto out;
 	}
-	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
+	inode->i_mode = (S_IFDIR|S_IRUGO|S_IXUGO) & ~proc_umask;
 	inode->i_op = &proc_tgid_base_inode_operations;
 	inode->i_fop = &proc_tgid_base_operations;
 	inode->i_nlink = 3;
@@ -1592,7 +1592,7 @@ static struct dentry *proc_task_lookup(s
 
 	if (!inode)
 		goto out_drop_task;
-	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
+	inode->i_mode = (S_IFDIR|S_IRUGO|S_IXUGO) & ~proc_umask;
 	inode->i_op = &proc_tid_base_inode_operations;
 	inode->i_fop = &proc_tid_base_operations;
 	inode->i_nlink = 3;
fs/proc/inode.c +14 -1
diff -rup linux-2.6.11-rc2-bk6/fs/proc/inode.c l/fs/proc/inode.c
@@ -22,6 +22,8 @@
 
 extern void free_proc_entry(struct proc_dir_entry *);
 
+umode_t proc_umask = 0;
+
 static inline struct proc_dir_entry * de_get(struct proc_dir_entry *de)
 {
 	if (de)
@@ -127,9 +129,14 @@ int __init proc_init_inodecache(void)
 	return 0;
 }
 
+static int parse_options(char *, uid_t *, gid_t *);
 static int proc_remount(struct super_block *sb, int *flags, char *data)
 {
+	uid_t dummy_uid;
+	gid_t dummy_gid;
+
 	*flags |= MS_NODIRATIME;
+	parse_options(data, &dummy_uid, &dummy_gid);
 	return 0;
 }
 
@@ -144,12 +151,13 @@ static struct super_operations proc_sops
 };
 
 enum {
-	Opt_uid, Opt_gid, Opt_err
+	Opt_uid, Opt_gid, Opt_umask, Opt_err
 };
 
 static match_table_t tokens = {
 	{Opt_uid, "uid=%u"},
 	{Opt_gid, "gid=%u"},
+	{Opt_umask, "umask=%o"},
 	{Opt_err, NULL}
 };
 
@@ -181,6 +189,11 @@ static int parse_options(char *options,u
 				return 0;
 			*gid = option;
 			break;
+		case Opt_umask:
+			if (match_octal(args, &option))
+				return 0;
+			proc_umask = option;
+			break;
 		default:
 			return 0;
 		}
fs/proc/internal.h +2 -0
diff -rup linux-2.6.11-rc2-bk6/fs/proc/internal.h l/fs/proc/internal.h
@@ -16,6 +16,8 @@ struct vmalloc_info {
 	unsigned long	largest_chunk;
 };
 
+extern umode_t proc_umask;
+
 #ifdef CONFIG_MMU
 #define VMALLOC_TOTAL (VMALLOC_END - VMALLOC_START)
 extern void get_vmalloc_info(struct vmalloc_info *vmi);
#1 lore raw
Quoted 10 lines · Rene Scharfe

The patch is inspired by the /proc restriction parts of the GrSecurity patch. The main difference is the ability to configure the restrictions dynamically. You can change the umask setting by running

   # mount -o remount,umask=007 /proc

Testing has been *very* light so far -- it compiles and boots. Patch is against 2.6.11-rc2-bk6.

Comments are very welcome.

It leaves already existing inodes with whatever mode they used to have. _IF_ you want to do that sort of things, do it right - add ->permission() that would apply that umask before checks and if you want it to be seen in results of stat(2) - add ->gettattr() and do the same there.

#2 lore raw
Quoted 9 lines · Al Viro
Quoted 7 lines · Rene Scharfe

The patch is inspired by the /proc restriction parts of the GrSecurity patch. The main difference is the ability to configure the restrictions dynamically. You can change the umask setting by running

# mount -o remount,umask=007 /proc
Quoted 8 lines
Quoted 4 lines

Testing has been *very* light so far -- it compiles and boots. Patch is against 2.6.11-rc2-bk6.

Comments are very welcome.

It leaves already existing inodes with whatever mode they used to have.

I said "configure the restrictions dynamically" but I meant "doesn't need a recompile to change settings". I expect the umask to be specified in /etc/fstab and rarely changed in a running system. With that in mind I think the patch is useful as-is, especially because it's so small. But I agree, that thing is a dirty hack. :]

Quoted 4 lines

_IF_ you want to do that sort of things, do it right - add ->permission() that would apply that umask before checks and if you want it to be seen in results of stat(2) - add ->gettattr() and do the same there.

Aww, that sounds expensive. My favourite solution would be to only allow the umask to be changed at mount time, not when remounting.

Calling parse_options from proc_fill_super, only and not from proc_remount does not help very much because proc_fill_super is only called at boot (or proc module load time). Is there another way?

While we are here: how would one change the uid or gid parameter? With a built-in proc fs the mount -a -t proc in the init scripts only results in a proc_remount call which (in mainline) doesn't bother looking at parameters at all. The same is true for a unmount, mount sequence.

Thanks,
Rene