Re: + fs-epoll-add-procfs-fdinfo-helper.patch added to -mm tree

[Oleg Nesterov <oleg@redhat.com>]

On 11/13, Cyrill Gorcunov wrote:
>
>  struct signalfd_ctx {
> +	rwlock_t lock;
>  	sigset_t sigmask;

Oh, I don't think.

rwlock_t is horrible in general, and what it can buy for signalfd?
A plain spinlock would be better. Or seqlock_t.

Whatever you do, you are trying to introduce the lock which should
serialize the access to ->sigmask correctly. In this case I think
you should split this change into 2 patches. The first one should
fix the locking, imo. sys_signalfd4() should not use ->siglock at
all, and the users which take ->siglock to read ->sigmask should be
updated.

Or,

> +#ifdef CONFIG_PROC_FS
> +static int signalfd_show_fdinfo(struct seq_file *m, struct file *f)
> +{
> +	struct signalfd_ctx *ctx = f->private_data;
> +	sigset_t sigmask;
> +
> +	read_lock(&ctx->lock);
> +	sigmask = ctx->sigmask;
> +	read_unlock(&ctx->lock);

Just read ctx->sigmask lockless. Do we really care if show_fdinfo()
reads the value "in between" ?

Oleg.