cgroup: restructure the failure path in cgroup_write_event_control()
authorLi Zefan <lizefan@huawei.com>
Thu, 1 Aug 2013 01:51:47 +0000 (09:51 +0800)
committerTejun Heo <tj@kernel.org>
Thu, 1 Aug 2013 13:29:41 +0000 (09:29 -0400)
It uses a single label and checks the validity of each pointer. This
is err-prone, and actually we had a bug because one of the check was
insufficient.

Use multi lables as we do in other places.

v2:
- drop initializations of local variables.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/cgroup.c

index 3f65933335253d8be3e4dfd2b59839018f8ee3ad..9f6dab22289efce9c6e826ece301331f437ebf1c 100644 (file)
@@ -3934,11 +3934,11 @@ static void cgroup_event_ptable_queue_proc(struct file *file,
 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
                                      const char *buffer)
 {
-       struct cgroup_event *event = NULL;
+       struct cgroup_event *event;
        struct cgroup *cgrp_cfile;
        unsigned int efd, cfd;
-       struct file *efile = NULL;
-       struct file *cfile = NULL;
+       struct file *efile;
+       struct file *cfile;
        char *endp;
        int ret;
 
@@ -3964,31 +3964,31 @@ static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
        efile = eventfd_fget(efd);
        if (IS_ERR(efile)) {
                ret = PTR_ERR(efile);
-               goto fail;
+               goto out_kfree;
        }
 
        event->eventfd = eventfd_ctx_fileget(efile);
        if (IS_ERR(event->eventfd)) {
                ret = PTR_ERR(event->eventfd);
-               goto fail;
+               goto out_put_efile;
        }
 
        cfile = fget(cfd);
        if (!cfile) {
                ret = -EBADF;
-               goto fail;
+               goto out_put_eventfd;
        }
 
        /* the process need read permission on control file */
        /* AV: shouldn't we check that it's been opened for read instead? */
        ret = inode_permission(file_inode(cfile), MAY_READ);
        if (ret < 0)
-               goto fail;
+               goto out_put_cfile;
 
        event->cft = __file_cft(cfile);
        if (IS_ERR(event->cft)) {
                ret = PTR_ERR(event->cft);
-               goto fail;
+               goto out_put_cfile;
        }
 
        /*
@@ -3998,18 +3998,18 @@ static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
        cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
        if (cgrp_cfile != cgrp) {
                ret = -EINVAL;
-               goto fail;
+               goto out_put_cfile;
        }
 
        if (!event->cft->register_event || !event->cft->unregister_event) {
                ret = -EINVAL;
-               goto fail;
+               goto out_put_cfile;
        }
 
        ret = event->cft->register_event(cgrp, event->cft,
                        event->eventfd, buffer);
        if (ret)
-               goto fail;
+               goto out_put_cfile;
 
        efile->f_op->poll(efile, &event->pt);
 
@@ -4029,16 +4029,13 @@ static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
 
        return 0;
 
-fail:
-       if (cfile)
-               fput(cfile);
-
-       if (event && event->eventfd && !IS_ERR(event->eventfd))
-               eventfd_ctx_put(event->eventfd);
-
-       if (!IS_ERR_OR_NULL(efile))
-               fput(efile);
-
+out_put_cfile:
+       fput(cfile);
+out_put_eventfd:
+       eventfd_ctx_put(event->eventfd);
+out_put_efile:
+       fput(efile);
+out_kfree:
        kfree(event);
 
        return ret;