Skip to contentSkip to conversation

mq_open() honor leading slash

PATCH In review Chris Wright 3 participants 5 messages Apr 15, 2004

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mq_open() honor leading slash

SUSv3 requires the following on mq_open() with name containing leading slash:

If name begins with the slash character, then processes calling mq_open() with the same value of name shall refer to the same message queue object, as long as that name has not been removed. If name does not begin with the slash character, the effect is implementation-defined. The interpretation of slash characters other than the leading slash character in name is implementation-defined.

The use of lookup_one_len() to force all lookups relative to the mqueue_mnt root guarantees absolute rather than relative lookup without leading slash, and generates an error on a name that contains any slashes at all. This is inconsitent with the part of the spec that requires leading slash to be effectively an absolute path (unless you consider -EACCES being "the same message queue object" ;-)

Patch below simply eats all leading slashes before passing name to lookup_one_len() in mq_open() and mq_unlink().

thanks,
-chris

Signature · 49 lines
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net

===== ipc/mqueue.c 1.8 vs edited =====
--- 1.8/ipc/mqueue.c	Wed Apr 14 18:37:51 2004
+++ edited/ipc/mqueue.c	Thu Apr 15 10:31:59 2004
@@ -609,7 +609,7 @@
 {
 	struct dentry *dentry;
 	struct file *filp;
-	char *name;
+	char *name, *qname;
 	int fd, error;
 
 	if (IS_ERR(name = getname(u_name)))
@@ -619,8 +619,11 @@
 	if (fd < 0)
 		goto out_putname;
 
+	qname = name;
+	while (*qname == '/')
+		qname++;
 	down(&mqueue_mnt->mnt_root->d_inode->i_sem);
-	dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
+	dentry = lookup_one_len(qname, mqueue_mnt->mnt_root, strlen(qname));
 	if (IS_ERR(dentry)) {
 		error = PTR_ERR(dentry);
 		goto out_err;
@@ -665,7 +668,7 @@
 asmlinkage long sys_mq_unlink(const char __user *u_name)
 {
 	int err;
-	char *name;
+	char *name, *qname;
 	struct dentry *dentry;
 	struct inode *inode = NULL;
 
@@ -673,8 +676,11 @@
 	if (IS_ERR(name))
 		return PTR_ERR(name);
 
+	qname = name;
+	while (*qname == '/')
+		qname++;
 	down(&mqueue_mnt->mnt_root->d_inode->i_sem);
-	dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
+	dentry = lookup_one_len(qname, mqueue_mnt->mnt_root, strlen(qname));
 	if (IS_ERR(dentry)) {
 		err = PTR_ERR(dentry);
 		goto out_unlock;
#1 lore raw
Quoted 18 lines · Chris Wright

SUSv3 requires the following on mq_open() with name containing leading slash:

If name begins with the slash character, then processes calling mq_open() with the same value of name shall refer to the same message queue object, as long as that name has not been removed. If name does not begin with the slash character, the effect is implementation-defined. The interpretation of slash characters other than the leading slash character in name is implementation-defined.

The use of lookup_one_len() to force all lookups relative to the mqueue_mnt root guarantees absolute rather than relative lookup without leading slash, and generates an error on a name that contains any slashes at all. This is inconsitent with the part of the spec that requires leading slash to be effectively an absolute path (unless you consider -EACCES being "the same message queue object" ;-)

Quoted 2 lines

Patch below simply eats all leading slashes before passing name to lookup_one_len() in mq_open() and mq_unlink().

glibc already strips the leading slash in userland.
If you want to do it in the kernel instead, it shouldn't IMHO be silent if it doesn't see a leading slash or sees more than one. I.e.

	error = -EINVAL;
	if (name[0] != '/')
		goto out_err;
...
	dentry = lookup_one_len(name + 1, mqueue_mnt->mnt_root, strlen(name + 1));

	Jakub
#2 lore raw
Quoted 5 lines · * Jakub Jelinek ([email protected]) wrote:
Quoted 2 lines · Chris Wright

Patch below simply eats all leading slashes before passing name to lookup_one_len() in mq_open() and mq_unlink().

glibc already strips the leading slash in userland.

Ah, OK. I'm just using the kernel interfaces directly.

Quoted 5 lines

If you want to do it in the kernel instead, it shouldn't IMHO be silent if it doesn't see a leading slash or sees more than one. I.e.

	error = -EINVAL;
	if (name[0] != '/')
		goto out_err;

Given it's implementation defined what happens w/out leading slash, I see no trouble with allowing names w/out leading slashes to be looked up as well as names with leading slash (in kernel rather than glibc). But I don't feel strongly either way.

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net
#3 lore raw
In reply to Chris Wright
Quoted 4 lines · Chris Wright

Patch below simply eats all leading slashes before passing name to lookup_one_len() in mq_open() and mq_unlink().

Why should we do that in kernel space?
The kernel interface is "no slash at all". User space can add a SuS compatible layer on top of the kernel interface (i.e. fail with -EINVAL if the first character is not a /, then skip the slash and pass "name+1" to sys_mq_open()).

--
Manfred

#4 lore raw
Quoted 9 lines · * Manfred Spraul ([email protected]) wrote:
Quoted 3 lines · Chris Wright

Patch below simply eats all leading slashes before passing name to lookup_one_len() in mq_open() and mq_unlink().

Why should we do that in kernel space?
The kernel interface is "no slash at all". User space can add a SuS compatible layer on top of the kernel interface (i.e. fail with -EINVAL if the first character is not a /, then skip the slash and pass "name+1" to sys_mq_open()).

Jakub said similar, and I missed the fact that glibc was sanitizing before calling kernel. I agree, it's fine as is.

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net