Re: [PATCH RFC v2] pidns: introduce syscall getvpid
From: Oleg Nesterov <hidden>
Date: 2015-09-24 17:37:41
Also in:
lkml
On 09/24, Konstantin Khlebnikov wrote:
On Thu, Sep 24, 2015 at 5:56 PM, Oleg Nesterov [off-list ref] wrote:quoted
On 09/24, Konstantin Khlebnikov wrote:quoted
+SYSCALL_DEFINE3(getvpid, pid_t, pid, int, source, int, target) +{ + struct file *source_file = NULL, *target_file = NULL; + struct pid_namespace *source_ns, *target_ns; + struct pid *struct_pid; + struct ns_common *ns; + pid_t result; + + if (source >= 0) { + source_file = proc_ns_fget(source); + result = PTR_ERR(source_file); + if (IS_ERR(source_file)) + goto out; + ns = get_proc_ns(file_inode(source_file)); + result = -EINVAL; + if (ns->ops->type != CLONE_NEWPID) + goto out; + source_ns = container_of(ns, struct pid_namespace, ns); + } else + source_ns = task_active_pid_ns(current); + + if (target >= 0) { + target_file = proc_ns_fget(target); + result = PTR_ERR(target_file); + if (IS_ERR(target_file)) + goto out; + ns = get_proc_ns(file_inode(target_file)); + result = -EINVAL; + if (ns->ops->type != CLONE_NEWPID) + goto out; + target_ns = container_of(ns, struct pid_namespace, ns); + } else + target_ns = task_active_pid_ns(current); +Hmm. Eric, Konstantin, how about (uncompiled/untested) patch below in a preparation? The code above doesn't look very readable.I've tried to do something like that but that comes too far so send patch as is.
OK, I won't insist. The code above asks for cleanup/factorization, but we can do this later.
Actually we can go deeper and replace struct file* with struct fd: this saves couple atomic ops for singlethreaded task.
it's not about performance... But yes, we can do more and create get_pid_ns_by_fd() similar to get_net_ns_by_fd(). We do not need file at all, we need pid_ns. Oleg.