[Xen-devel] [RFC PATCHv1] xen/privcmd: add IOCTL_PRIVCMD_RESTRICT_DOMID

[David Vrabel <david.vrabel@xxxxxxxxxx>]

This restricts the file descriptor to only being able map foreign
memory belonging to a specific domain.  Once a file descriptor has
been restricted its restriction cannot be removed or changed.

A device model (e.g., QEMU) or similar can make use of this before
dropping privileges to prevent the file descriptor being used to
escalate privleges if the process is compromised.

FIXME: This is not good enough (yet) as it does not restrict what
hypercalls may be performed.  Fixing this requires a hypervisor ABI
change.

Signed-off-by: David Vrabel <david.vrabel@xxxxxxxxxx>
---
 drivers/xen/privcmd.c      | 75 ++++++++++++++++++++++++++++++++++++++++++----
 include/uapi/xen/privcmd.h | 26 ++++++++++++++++
 2 files changed, 96 insertions(+), 5 deletions(-)

diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
index df2e6f7..513d1c5 100644
--- a/drivers/xen/privcmd.c
+++ b/drivers/xen/privcmd.c
@@ -43,6 +43,18 @@ MODULE_LICENSE("GPL");
 
 #define PRIV_VMA_LOCKED ((void *)1)
 
+#define UNRESTRICTED_DOMID ((domid_t)-1)
+
+struct privcmd_data {
+       domid_t restrict_domid;
+};
+
+static bool privcmd_is_allowed(struct privcmd_data *priv, domid_t domid)
+{
+       return priv->restrict_domid == UNRESTRICTED_DOMID
+               || priv->restrict_domid == domid;
+}
+
 static int privcmd_vma_range_is_mapped(
                struct vm_area_struct *vma,
                unsigned long addr,
@@ -229,7 +241,7 @@ static int mmap_gfn_range(void *data, void *state)
        return 0;
 }
 
-static long privcmd_ioctl_mmap(void __user *udata)
+static long privcmd_ioctl_mmap(struct privcmd_data *priv, void __user *udata)
 {
        struct privcmd_mmap mmapcmd;
        struct mm_struct *mm = current->mm;
@@ -245,6 +257,9 @@ static long privcmd_ioctl_mmap(void __user *udata)
        if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
                return -EFAULT;
 
+       if (!privcmd_is_allowed(priv, mmapcmd.dom))
+               return -EACCES;
+
        rc = gather_array(&pagelist,
                          mmapcmd.num, sizeof(struct privcmd_mmap_entry),
                          mmapcmd.entry);
@@ -416,7 +431,8 @@ static int alloc_empty_pages(struct vm_area_struct *vma, 
int numpgs)
 
 static const struct vm_operations_struct privcmd_vm_ops;
 
-static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
+static long privcmd_ioctl_mmap_batch(struct privcmd_data *priv, void __user 
*udata,
+                                    int version)
 {
        int ret;
        struct privcmd_mmapbatch_v2 m;
@@ -446,6 +462,9 @@ static long privcmd_ioctl_mmap_batch(void __user *udata, 
int version)
                return -EINVAL;
        }
 
+       if (!privcmd_is_allowed(priv, m.dom))
+               return -EACCES;
+
        nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
        if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
                return -EINVAL;
@@ -548,9 +567,28 @@ out_unlock:
        goto out;
 }
 
+static int privcmd_ioctl_restrict_domid(struct privcmd_data *priv,
+                                       void __user *udata)
+{
+       struct privcmd_restrict_domid prd;
+
+       if (copy_from_user(&prd, udata, sizeof(prd)))
+               return -EFAULT;
+
+       if (prd.domid >= DOMID_FIRST_RESERVED)
+               return -EINVAL;
+       if (priv->restrict_domid != UNRESTRICTED_DOMID)
+               return -EACCES;
+
+       priv->restrict_domid = prd.domid;
+
+       return 0;
+}
+
 static long privcmd_ioctl(struct file *file,
                          unsigned int cmd, unsigned long data)
 {
+       struct privcmd_data *priv = file->private_data;
        int ret = -ENOSYS;
        void __user *udata = (void __user *) data;
 
@@ -560,15 +598,19 @@ static long privcmd_ioctl(struct file *file,
                break;
 
        case IOCTL_PRIVCMD_MMAP:
-               ret = privcmd_ioctl_mmap(udata);
+               ret = privcmd_ioctl_mmap(priv, udata);
                break;
 
        case IOCTL_PRIVCMD_MMAPBATCH:
-               ret = privcmd_ioctl_mmap_batch(udata, 1);
+               ret = privcmd_ioctl_mmap_batch(priv, udata, 1);
                break;
 
        case IOCTL_PRIVCMD_MMAPBATCH_V2:
-               ret = privcmd_ioctl_mmap_batch(udata, 2);
+               ret = privcmd_ioctl_mmap_batch(priv, udata, 2);
+               break;
+
+       case IOCTL_PRIVCMD_RESTRICT_DOMID:
+               ret = privcmd_ioctl_restrict_domid(priv, udata);
                break;
 
        default:
@@ -644,10 +686,33 @@ static int privcmd_vma_range_is_mapped(
                                   is_mapped_fn, NULL) != 0;
 }
 
+static int privcmd_open(struct inode *ino, struct file *filp)
+{
+       struct privcmd_data *priv;
+
+       priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+       if (!priv)
+               return -ENOMEM;
+
+       priv->restrict_domid = UNRESTRICTED_DOMID;
+
+       return 0;
+}
+
+static int privcmd_release(struct inode *inode, struct file *file)
+{
+       struct privcmd_data *priv = file->private_data;
+
+       kfree(priv);
+       return 0;
+}
+
 const struct file_operations xen_privcmd_fops = {
        .owner = THIS_MODULE,
        .unlocked_ioctl = privcmd_ioctl,
        .mmap = privcmd_mmap,
+       .open = privcmd_open,
+       .release = privcmd_release,
 };
 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 
diff --git a/include/uapi/xen/privcmd.h b/include/uapi/xen/privcmd.h
index 7ddeeda..e2cea4b 100644
--- a/include/uapi/xen/privcmd.h
+++ b/include/uapi/xen/privcmd.h
@@ -77,6 +77,10 @@ struct privcmd_mmapbatch_v2 {
        int __user *err;  /* array of error codes */
 };
 
+struct privcmd_restrict_domid {
+       domid_t domid;
+};
+
 /*
  * @cmd: IOCTL_PRIVCMD_HYPERCALL
  * @arg: &privcmd_hypercall_t
@@ -99,4 +103,26 @@ struct privcmd_mmapbatch_v2 {
 #define IOCTL_PRIVCMD_MMAPBATCH_V2                             \
        _IOC(_IOC_NONE, 'P', 4, sizeof(struct privcmd_mmapbatch_v2))
 
+/*
+ * @cmd: IOCTL_PRIVCMD_RESTRICT_DOMID
+ * @arg: struct privcmd_restrict_domid *
+ * Return: 0 on success, or -1 (with errno set).
+ *
+ * This restricts the file descriptor to only being able map foreign
+ * memory belonging to a specific domain.  Once a file descriptor has
+ * been restricted its restriction cannot be removed or changed.
+ *
+ * A device model (e.g., QEMU) or similar can make use of this before
+ * dropping privileges to prevent the file descriptor being used to
+ * escalate privleges if the process is compromised.
+ *
+ * FIXME: This is not good enough (yet) as it does not restrict what
+ * hypercalls may be performed.
+ *
+ * EINVAL - the specified domid isn't valid.
+ * EACCES - the file descriptor has already been restricted.
+ */
+#define IOCTL_PRIVCMD_RESTRICT_DOMID                           \
+       _IOC(_IOC_NONE, 'P', 5, sizeof(struct privcmd_restrict_domid))
+
 #endif /* __LINUX_PUBLIC_PRIVCMD_H__ */
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel