Add implementation for capabilities.

Code to actually implement Capsicum capabilities, including fileops and
kern_capwrap(), which creates a capability to wrap an existing file
descriptor.

We also modify kern_close() and closef() to handle capabilities.

Finally, remove cap_filelist from struct capability, since we don't
actually need it.

Approved by: mentor (rwatson), re (Capsicum blanket)
Sponsored by: Google Inc

3 files changed, 240 insertions, 10 deletions

diff --git a/sys/kern/kern_descrip.c b/sys/kern/kern_descrip.c
index f8eaa84..829ece2 100644
--- a/sys/kern/kern_descrip.c
+++ b/sys/kern/kern_descrip.c
@@ -1160,7 +1160,7 @@ kern_close(td, fd)
 	int fd;
 {
 	struct filedesc *fdp;
-	struct file *fp;
+	struct file *fp, *fp_object;
 	int error;
 	int holdleaders;
 
@@ -1195,8 +1195,14 @@ kern_close(td, fd)
 	 * added, and deleteing a knote for the new fd.
 	 */
 	knote_fdclose(td, fd);
-	if (fp->f_type == DTYPE_MQUEUE)
-		mq_fdclose(td, fd, fp);
+
+	/*
+	 * When we're closing an fd with a capability, we need to notify
+	 * mqueue if the underlying object is of type mqueue.
+	 */
+	(void)cap_funwrap(fp, 0, &fp_object);
+	if (fp_object->f_type == DTYPE_MQUEUE)
+		mq_fdclose(td, fd, fp_object);
 	FILEDESC_XUNLOCK(fdp);
 
 	error = closef(fp, td);
@@ -2146,6 +2152,7 @@ closef(struct file *fp, struct thread *td)
 	struct flock lf;
 	struct filedesc_to_leader *fdtol;
 	struct filedesc *fdp;
+	struct file *fp_object;
 
 	/*
 	 * POSIX record locking dictates that any close releases ALL
@@ -2158,11 +2165,15 @@ closef(struct file *fp, struct thread *td)
 	 * NULL thread pointer when there really is no owning
 	 * context that might have locks, or the locks will be
 	 * leaked.
+	 *
+	 * If this is a capability, we do lock processing under the underlying
+	 * node, not the capability itself.
 	 */
-	if (fp->f_type == DTYPE_VNODE && td != NULL) {
+	(void)cap_funwrap(fp, 0, &fp_object);
+	if ((fp_object->f_type == DTYPE_VNODE) && (td != NULL)) {
 		int vfslocked;
 
-		vp = fp->f_vnode;
+		vp = fp_object->f_vnode;
 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
 			lf.l_whence = SEEK_SET;
@@ -2192,7 +2203,7 @@ closef(struct file *fp, struct thread *td)
 				lf.l_start = 0;
 				lf.l_len = 0;
 				lf.l_type = F_UNLCK;
-				vp = fp->f_vnode;
+				vp = fp_object->f_vnode;
 				(void) VOP_ADVLOCK(vp,
 						   (caddr_t)fdtol->fdl_leader,
 						   F_UNLCK, &lf, F_POSIX);
@@ -2497,6 +2508,9 @@ fputsock(struct socket *so)
 
 /*
  * Handle the last reference to a file being closed.
+ *
+ * No special capability handling here, as the capability's fo_close will run
+ * instead of the object here, and perform any necessary drop on the object.
  */
 int
 _fdrop(struct file *fp, struct thread *td)
diff --git a/sys/kern/sys_capability.c b/sys/kern/sys_capability.c
index 8216bea..9cbe462 100644
--- a/sys/kern/sys_capability.c
+++ b/sys/kern/sys_capability.c
@@ -123,18 +123,66 @@ cap_getmode(struct thread *td, struct cap_getmode_args *uap)
  * struct capability describes a capability, and is hung off of its struct
  * file f_data field.  cap_file and cap_rightss are static once hooked up, as
  * neither the object it references nor the rights it encapsulates are
- * permitted to change.  cap_filelist may change when other capabilites are
- * added or removed from the same file, and is currently protected by the
- * pool mutex for the object file descriptor.
+ * permitted to change.
  */
 struct capability {
 	struct file	*cap_object;	/* Underlying object's file. */
 	struct file	*cap_file;	/* Back-pointer to cap's file. */
 	cap_rights_t	 cap_rights;	/* Mask of rights on object. */
-	LIST_ENTRY(capability)	cap_filelist; /* Object's cap list. */
 };
 
 /*
+ * Capabilities have a fileops vector, but in practice none should ever be
+ * called except for fo_close, as the capability will normally not be
+ * returned during a file descriptor lookup in the system call code.
+ */
+static fo_rdwr_t capability_read;
+static fo_rdwr_t capability_write;
+static fo_truncate_t capability_truncate;
+static fo_ioctl_t capability_ioctl;
+static fo_poll_t capability_poll;
+static fo_kqfilter_t capability_kqfilter;
+static fo_stat_t capability_stat;
+static fo_close_t capability_close;
+
+static struct fileops capability_ops = {
+	.fo_read = capability_read,
+	.fo_write = capability_write,
+	.fo_truncate = capability_truncate,
+	.fo_ioctl = capability_ioctl,
+	.fo_poll = capability_poll,
+	.fo_kqfilter = capability_kqfilter,
+	.fo_stat = capability_stat,
+	.fo_close = capability_close,
+	.fo_flags = DFLAG_PASSABLE,
+};
+
+static struct fileops capability_ops_unpassable = {
+	.fo_read = capability_read,
+	.fo_write = capability_write,
+	.fo_truncate = capability_truncate,
+	.fo_ioctl = capability_ioctl,
+	.fo_poll = capability_poll,
+	.fo_kqfilter = capability_kqfilter,
+	.fo_stat = capability_stat,
+	.fo_close = capability_close,
+	.fo_flags = 0,
+};
+
+static uma_zone_t capability_zone;
+
+static void
+capability_init(void *dummy __unused)
+{
+
+	capability_zone = uma_zcreate("capability", sizeof(struct capability),
+	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
+	if (capability_zone == NULL)
+		panic("capability_init: capability_zone not initialized");
+}
+SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, capability_init, NULL);
+
+/*
  * Test whether a capability grants the requested rights.
  */
 static int
@@ -147,6 +195,85 @@ cap_check(struct capability *c, cap_rights_t rights)
 }
 
 /*
+ * Extract rights from a capability for monitoring purposes -- not for use in
+ * any other way, as we want to keep all capability permission evaluation in
+ * this one file.
+ */
+cap_rights_t
+cap_rights(struct file *fp_cap)
+{
+	struct capability *c;
+
+	KASSERT(fp_cap->f_type == DTYPE_CAPABILITY,
+	    ("cap_rights: !capability"));
+
+	c = fp_cap->f_data;
+	return (c->cap_rights);
+}
+
+/*
+ * Create a capability to wrap around an existing file.
+ */
+int
+kern_capwrap(struct thread *td, struct file *fp, cap_rights_t rights,
+    struct file **fcappp, int *capfdp)
+{
+	struct capability *cp, *cp_old;
+	struct file *fp_object;
+	int error;
+
+	if ((rights | CAP_MASK_VALID) != CAP_MASK_VALID)
+		return (EINVAL);
+
+	/*
+	 * If a new capability is being derived from an existing capability,
+	 * then the new capability rights must be a subset of the existing
+	 * rights.
+	 */
+	if (fp->f_type == DTYPE_CAPABILITY) {
+		cp_old = fp->f_data;
+		if ((cp_old->cap_rights | rights) != cp_old->cap_rights)
+			return (ENOTCAPABLE);
+	}
+
+	/*
+	 * Allocate a new file descriptor to hang the capability off of.
+	 */
+	error = falloc(td, fcappp, capfdp, fp->f_flag);
+	if (error)
+		return (error);
+
+	/*
+	 * Rather than nesting capabilities, directly reference the object an
+	 * existing capability references.  There's nothing else interesting
+	 * to preserve for future use, as we've incorporated the previous
+	 * rights mask into the new one.  This prevents us from having to
+	 * deal with capability chains.
+	 */
+	if (fp->f_type == DTYPE_CAPABILITY)
+		fp_object = ((struct capability *)fp->f_data)->cap_object;
+	else
+		fp_object = fp;
+	fhold(fp_object);
+	cp = uma_zalloc(capability_zone, M_WAITOK | M_ZERO);
+	cp->cap_rights = rights;
+	cp->cap_object = fp_object;
+	cp->cap_file = *fcappp;
+	if (fp->f_flag & DFLAG_PASSABLE)
+		finit(*fcappp, fp->f_flag, DTYPE_CAPABILITY, cp,
+		    &capability_ops);
+	else
+		finit(*fcappp, fp->f_flag, DTYPE_CAPABILITY, cp,
+		    &capability_ops_unpassable);
+
+	/*
+	 * Release our private reference (the proc filedesc still has one).
+	 */
+	fdrop(*fcappp, td);
+	return (0);
+}
+
+/*
  * Given a file descriptor, test it against a capability rights mask and then
  * return the file descriptor on which to actually perform the requested
  * operation.  As long as the reference to fp_cap remains valid, the returned
@@ -206,6 +333,89 @@ cap_funwrap_mmap(struct file *fp_cap, cap_rights_t rights, u_char *maxprotp,
 	return (0);
 }
 
+/*
+ * When a capability is closed, simply drop the reference on the underlying
+ * object and free the capability.  fdrop() will handle the case where the
+ * underlying object also needs to close, and the caller will have already
+ * performed any object-specific lock or mqueue handling.
+ */
+static int
+capability_close(struct file *fp, struct thread *td)
+{
+	struct capability *c;
+	struct file *fp_object;
+
+	KASSERT(fp->f_type == DTYPE_CAPABILITY,
+	    ("capability_close: !capability"));
+
+	c = fp->f_data;
+	fp->f_ops = &badfileops;
+	fp->f_data = NULL;
+	fp_object = c->cap_object;
+	uma_zfree(capability_zone, c);
+	return (fdrop(fp_object, td));
+}
+
+/*
+ * In general, file descriptor operations should never make it to the
+ * capability, only the underlying file descriptor operation vector, so panic
+ * if we do turn up here.
+ */
+static int
+capability_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
+    int flags, struct thread *td)
+{
+
+	panic("capability_read");
+}
+
+static int
+capability_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
+    int flags, struct thread *td)
+{
+
+	panic("capability_write");
+}
+
+static int
+capability_truncate(struct file *fp, off_t length, struct ucred *active_cred,
+    struct thread *td)
+{
+
+	panic("capability_truncate");
+}
+
+static int
+capability_ioctl(struct file *fp, u_long com, void *data,
+    struct ucred *active_cred, struct thread *td)
+{
+
+	panic("capability_ioctl");
+}
+
+static int
+capability_poll(struct file *fp, int events, struct ucred *active_cred,
+    struct thread *td)
+{
+
+	panic("capability_poll");
+}
+
+static int
+capability_kqfilter(struct file *fp, struct knote *kn)
+{
+
+	panic("capability_kqfilter");
+}
+
+static int
+capability_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
+    struct thread *td)
+{
+
+	panic("capability_stat");
+}
+
 #else /* !CAPABILITIES */
 
 /*
diff --git a/sys/sys/capability.h b/sys/sys/capability.h
index b56aa8b..5f5c35e 100644
--- a/sys/sys/capability.h
+++ b/sys/sys/capability.h
@@ -64,6 +64,12 @@
 #define IN_CAPABILITY_MODE(td) (td->td_ucred->cr_flags & CRED_FLAG_CAPMODE)
 
 /*
+ * Create a capability to wrap a file object.
+ */
+int	kern_capwrap(struct thread *td, struct file *fp, cap_rights_t rights,
+	    struct file **cap, int *capfd);
+
+/*
  * Unwrap a capability if its rights mask is a superset of 'rights'.
  *
  * Unwrapping a non-capability is effectively a no-op; the value of fp_cap