Re: [PATCH RFC 1/5] vringfd syscall

Re: [PATCH RFC 1/5] vringfd syscall

[Anthony Liguori]

Rusty Russell wrote:
> For virtualization, we've developed virtio_ring for efficient communication.
> This would also work well for userspace-kernel communication, particularly
> for things like the tun device.  By using the same ABI, we can join guests
> to the host kernel trivially.
> 
> These patches are fairly alpha; I've seen some network stalls I have to
> track down and there are some fixmes.
> 
> Comments welcome!
> Rusty.
> 
> diff -r 99132ad16999 Documentation/test_vring.c
> --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
> +++ b/Documentation/test_vring.c	Sat Apr 05 21:31:40 2008 +1100
> @@ -0,0 +1,47 @@
> +#include <unistd.h>
> +#include <linux/virtio_ring.h>
> +#include <stdio.h>
> +#include <stdint.h>
> +#include <err.h>
> +#include <poll.h>
> +
> +#ifndef __NR_vringfd
> +#define __NR_vringfd		327
> +#endif
> +
> +int main()
> +{
> +	int fd, r;
> +	struct vring vr;
> +	uint16_t used = 0;
> +	struct pollfd pfd;
> +	void *buf = calloc(vring_size(256, getpagesize()), 0);

Shouldn't this be calloc(1, vring_size(256, getpagesize()));?

> +	vring_init(&vr, 256, buf, getpagesize());
> +
> +	fd = syscall(__NR_vringfd, buf, 256, &used);
> +	if (fd < 0)
> +		err(1, "vringfd gave %i", fd);
> +
> +	pfd.fd = fd;
> +	pfd.events = POLLIN;
> +	r = poll(&pfd, 1, 0);
> +	
> +	if (r != 0)
> +		err(1, "poll gave %i", r);
> +
> +	vr.used->idx++;
> +	r = poll(&pfd, 1, 0);
> +	
> +	if (r != 1)
> +		err(1, "poll after buf used gave %i", r);
> +
> +	used++;
> +	r = poll(&pfd, 1, 0);
> +	
> +	if (r != 0)
> +		err(1, "poll after used incremented gave %i", r);

I have a tough time seeing what you're demonstrating here.  Perhaps some 
comments?

> +	close(fd);
> +	return 0;
> +}
> diff -r 99132ad16999 arch/x86/kernel/syscall_table_32.S
> --- a/arch/x86/kernel/syscall_table_32.S	Sat Apr 05 21:20:32 2008 +1100
> +++ b/arch/x86/kernel/syscall_table_32.S	Sat Apr 05 21:31:40 2008 +1100
> @@ -326,3 +326,4 @@ ENTRY(sys_call_table)
>  	.long sys_fallocate
>  	.long sys_timerfd_settime	/* 325 */
>  	.long sys_timerfd_gettime
> +	.long sys_vringfd
> diff -r 99132ad16999 fs/Kconfig
> --- a/fs/Kconfig	Sat Apr 05 21:20:32 2008 +1100
> +++ b/fs/Kconfig	Sat Apr 05 21:31:40 2008 +1100
> @@ -2135,4 +2135,14 @@ source "fs/nls/Kconfig"
>  source "fs/nls/Kconfig"
>  source "fs/dlm/Kconfig"
>  
> +config VRINGFD
> +       bool "vring fd support (EXPERIMENTAL)"
> +       depends on EXPERIMENTAL
> +       help
> +         vring is a ringbuffer implementation for efficient I/O.  It is
> +	 currently used by virtualization hosts (lguest, kvm) for efficient
> +	 networking using the tun driver.
> +
> +	 If unsure, say N.
> +

Should select VIRTIO && VIRTIO_RING

<snip>

> +/* Returns an error, or 0 (no buffers), or an id for vring_used_buffer() */
> +int vring_get_buffer(struct vring_info *vr,
> +		     struct iovec *in_iov,
> +		     unsigned int *num_in, unsigned long *in_len,
> +		     struct iovec *out_iov,
> +		     unsigned int *num_out, unsigned long *out_len)
> +{

It seems unlikely that a caller could place both in_iov/out_iov on the 
stack since to do it safely, you would have to use vring.num which is 
determined by userspace.  Since you have to kmalloc() the buffers 
anyway, why not just allocate a single data structure within this 
function and return it.

> +void vring_used_buffer_atomic(struct vring_info *vr, int id, u32 len)
> +{
> +	struct vring_used_elem *used;
> +
> +	BUG_ON(id <= 0 || id > vr->ring.num);
> +	BUG_ON(!vr->used);
> +
> +	used = &vr->used->ring[vr->used->idx & vr->mask];
> +	used->id = id - 1;
> +	used->len = len;
> +	/* Make sure buffer is written before we update index. */
> +	wmb();
> +	vr->used->idx++;
> +}
> +EXPORT_SYMBOL_GPL(vring_used_buffer_atomic);

When is this actually safe to use?

> +
> +void vring_wake(struct vring_info *vr)
> +{
> +	wake_up(&vr->poll_wait);
> +}
> +EXPORT_SYMBOL_GPL(vring_wake);
> +
> +struct vring_info *vring_attach(int fd, const struct vring_ops *ops,
> +				void *data, bool atomic_use)
> +{
> +	struct file *filp;
> +	struct vring_info *vr;
> +
> +	/* Must be a valid fd, and must be one of ours. */
> +	filp = fget(fd);
> +	if (!filp) {
> +		vr = ERR_PTR(-EBADF);
> +		goto out;
> +	}
> +
> +	if (filp->f_op != &vring_fops) {
> +		vr = ERR_PTR(-EBADF);
> +		goto fput;
> +	}
> +
> +	/* Mutex simply protects against parallel vring_attach. */
> +	mutex_lock(&vring_lock);
> +	vr = filp->private_data;
> +	if (vr->ops) {
> +		vr = ERR_PTR(-EBUSY);
> +		goto unlock;
> +	}
> +
> +	/* If they want to use atomically, we have to map the page. */
> +	if (atomic_use) {
> +		if (get_user_pages(current, current->mm,
> +				   (unsigned long)vr->ring.used, 1, 1, 1,
> +				   &vr->used_page, NULL) != 1) {
> +			vr = ERR_PTR(-EFAULT);
> +			goto unlock;
> +		}

Oh, this is when it's safe to use.  You don't seem to be acquiring 
current->mm->mmap_sem here.  Also, this assumes the used ring fits 
within a single page which isn't true with a ring > 512 elements.

A consequence of this is that devices that interact with a ring queue 
atomically now have an additional capability: pinning an arbitrary 
amount of physical memory.

I think this means that it's no longer safe to give a tap fd to an 
unprivileged process regardless of how you're routing the traffic.

Regards,

Anthony Liguori

Re: [PATCH RFC 1/5] vringfd syscall

[Rusty Russell]

On Sunday 06 April 2008 03:13:41 Anthony Liguori wrote:
> > +	void *buf = calloc(vring_size(256, getpagesize()), 0);
>
> Shouldn't this be calloc(1, vring_size(256, getpagesize()));?

Heh, yes... spot the last minute change from malloc to calloc.

> > +	if (r != 0)
> > +		err(1, "poll after used incremented gave %i", r);
>
> I have a tough time seeing what you're demonstrating here.  Perhaps some
> comments?

Well, not I have lguest working, I can just blow away the test program.  It
just tests poll.

> > +config VRINGFD
> > +       bool "vring fd support (EXPERIMENTAL)"
> > +       depends on EXPERIMENTAL
> > +       help
> > +         vring is a ringbuffer implementation for efficient I/O.  It is
> > +	 currently used by virtualization hosts (lguest, kvm) for efficient
> > +	 networking using the tun driver.
> > +
> > +	 If unsure, say N.
> > +
>
> Should select VIRTIO && VIRTIO_RING

I don't think so.  It doesn't depend on either.

> > +/* Returns an error, or 0 (no buffers), or an id for vring_used_buffer()
> > */ +int vring_get_buffer(struct vring_info *vr,
> > +		     struct iovec *in_iov,
> > +		     unsigned int *num_in, unsigned long *in_len,
> > +		     struct iovec *out_iov,
> > +		     unsigned int *num_out, unsigned long *out_len)
> > +{
>
> It seems unlikely that a caller could place both in_iov/out_iov on the
> stack since to do it safely, you would have to use vring.num which is
> determined by userspace.  Since you have to kmalloc() the buffers
> anyway, why not just allocate a single data structure within this
> function and return it.

This needs a comment. num_out and num_in are in parameters specifying the 
maximum of each.

> > +	/* If they want to use atomically, we have to map the page. */
> > +	if (atomic_use) {
> > +		if (get_user_pages(current, current->mm,
> > +				   (unsigned long)vr->ring.used, 1, 1, 1,
> > +				   &vr->used_page, NULL) != 1) {
> > +			vr = ERR_PTR(-EFAULT);
> > +			goto unlock;
> > +		}
>
> Oh, this is when it's safe to use.  You don't seem to be acquiring
> current->mm->mmap_sem here.  Also, this assumes the used ring fits
> within a single page which isn't true with a ring > 512 elements.

Yes, this is a hack.  It actually means ring <= 256 for PAGE_SIZE 4096.  I'm 
not entirely comfortable with it.

> A consequence of this is that devices that interact with a ring queue
> atomically now have an additional capability: pinning an arbitrary
> amount of physical memory.

Erk, the size check that was supposed to be here got lost in the reshuffle :(.

One option is to use a sliding window, but better is to do best effort and 
have the tun driver fall back (this is actually possible with a slight 
change).

Thanks,
Rusty.

[PATCH RFC 1/5] vringfd syscall

[Rusty Russell]

For virtualization, we've developed virtio_ring for efficient communication.
This would also work well for userspace-kernel communication, particularly
for things like the tun device.  By using the same ABI, we can join guests
to the host kernel trivially.

These patches are fairly alpha; I've seen some network stalls I have to
track down and there are some fixmes.

Comments welcome!
Rusty.

diff -r 99132ad16999 Documentation/test_vring.c
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/Documentation/test_vring.c	Sat Apr 05 21:31:40 2008 +1100
@@ -0,0 +1,47 @@
+#include <unistd.h>
+#include <linux/virtio_ring.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <err.h>
+#include <poll.h>
+
+#ifndef __NR_vringfd
+#define __NR_vringfd		327
+#endif
+
+int main()
+{
+	int fd, r;
+	struct vring vr;
+	uint16_t used = 0;
+	struct pollfd pfd;
+	void *buf = calloc(vring_size(256, getpagesize()), 0);
+
+	vring_init(&vr, 256, buf, getpagesize());
+
+	fd = syscall(__NR_vringfd, buf, 256, &used);
+	if (fd < 0)
+		err(1, "vringfd gave %i", fd);
+
+	pfd.fd = fd;
+	pfd.events = POLLIN;
+	r = poll(&pfd, 1, 0);
+	
+	if (r != 0)
+		err(1, "poll gave %i", r);
+
+	vr.used->idx++;
+	r = poll(&pfd, 1, 0);
+	
+	if (r != 1)
+		err(1, "poll after buf used gave %i", r);
+
+	used++;
+	r = poll(&pfd, 1, 0);
+	
+	if (r != 0)
+		err(1, "poll after used incremented gave %i", r);
+
+	close(fd);
+	return 0;
+}
diff -r 99132ad16999 arch/x86/kernel/syscall_table_32.S
--- a/arch/x86/kernel/syscall_table_32.S	Sat Apr 05 21:20:32 2008 +1100
+++ b/arch/x86/kernel/syscall_table_32.S	Sat Apr 05 21:31:40 2008 +1100
@@ -326,3 +326,4 @@ ENTRY(sys_call_table)
 	.long sys_fallocate
 	.long sys_timerfd_settime	/* 325 */
 	.long sys_timerfd_gettime
+	.long sys_vringfd
diff -r 99132ad16999 fs/Kconfig
--- a/fs/Kconfig	Sat Apr 05 21:20:32 2008 +1100
+++ b/fs/Kconfig	Sat Apr 05 21:31:40 2008 +1100
@@ -2135,4 +2135,14 @@ source "fs/nls/Kconfig"
 source "fs/nls/Kconfig"
 source "fs/dlm/Kconfig"
 
+config VRINGFD
+       bool "vring fd support (EXPERIMENTAL)"
+       depends on EXPERIMENTAL
+       help
+         vring is a ringbuffer implementation for efficient I/O.  It is
+	 currently used by virtualization hosts (lguest, kvm) for efficient
+	 networking using the tun driver.
+
+	 If unsure, say N.
+
 endmenu
diff -r 99132ad16999 fs/Makefile
--- a/fs/Makefile	Sat Apr 05 21:20:32 2008 +1100
+++ b/fs/Makefile	Sat Apr 05 21:31:40 2008 +1100
@@ -119,3 +119,4 @@ obj-$(CONFIG_DEBUG_FS)		+= debugfs/
 obj-$(CONFIG_DEBUG_FS)		+= debugfs/
 obj-$(CONFIG_OCFS2_FS)		+= ocfs2/
 obj-$(CONFIG_GFS2_FS)           += gfs2/
+obj-$(CONFIG_VRINGFD)		+= vring.o
diff -r 99132ad16999 fs/vring.c
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/fs/vring.c	Sat Apr 05 21:31:40 2008 +1100
@@ -0,0 +1,376 @@
+/* Ring-buffer file descriptor implementation.
+ *
+ *  Copyright 2008 Rusty Russell IBM Corporation
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ */
+#include <linux/virtio_ring.h>
+#include <linux/vring.h>
+#include <linux/init.h>
+#include <linux/mutex.h>
+#include <linux/wait.h>
+#include <linux/fs.h>
+#include <linux/poll.h>
+#include <linux/highmem.h>
+#include <linux/file.h>
+#include <linux/mount.h>
+#include <linux/magic.h>
+#include <linux/module.h>
+
+static struct vfsmount *vring_mnt;
+static DEFINE_MUTEX(vring_lock);
+
+struct vring_info
+{
+	struct vring ring;
+	u16 mask;
+	u16 __user *last_used;
+	u16 last_avail;
+
+	const struct vring_ops *ops;
+	void *ops_data;
+
+	/* Waitqueue for poll() */
+	wait_queue_head_t poll_wait;
+
+	/* The mapped used ring. */
+	struct vring_used *used;
+	struct page *used_page;
+};
+
+static unsigned int vring_poll(struct file *filp,
+			       struct poll_table_struct *poll)
+{
+	struct vring_info *vr = filp->private_data;
+	int err;
+	unsigned int mask;
+	u16 used, last_used;
+
+	/* Some uses of vrings require updating in user context.  This
+	 * is best done close to the caller, ie. here. */
+	if (vr->ops && vr->ops->pull) {
+		err = vr->ops->pull(vr->ops_data);
+		if (unlikely(err < 0))
+			return err;
+
+		if (err > 0) {
+			/* Buffers have been used, no need to check indices */
+			mask = POLLIN | POLLRDNORM;
+			goto poll_wait;
+		}
+	}
+
+	err = get_user(used, &vr->ring.used->idx);
+	if (unlikely(err))
+		return err;
+
+	err = get_user(last_used, vr->last_used);
+	if (unlikely(err))
+		return err;
+
+	/* More buffers have been used?  It's 'readable'. */
+	if (used != last_used)
+		mask = POLLIN | POLLRDNORM;
+	else
+		mask = 0;
+
+poll_wait:
+	poll_wait(filp, &vr->poll_wait, poll);
+
+	return mask;
+}
+
+static ssize_t vring_write(struct file *filp, const char __user *buf,
+			   size_t size, loff_t *off)
+{
+	struct vring_info *vr = filp->private_data;
+
+	if (vr->ops && vr->ops->push)
+		return vr->ops->push(vr->ops_data);
+
+	return -EINVAL;
+}
+
+static int vring_release(struct inode *inode, struct file *filp)
+{
+	struct vring_info *vr = filp->private_data;
+
+	/* Callback for other end. */
+	if (vr->ops && vr->ops->destroy)
+		vr->ops->destroy(vr->ops_data);
+
+	if (vr->used) {
+		kunmap(vr->used_page);
+		put_page(vr->used_page);
+	}
+
+	kfree(vr);
+	return 0;
+}
+
+static const struct file_operations vring_fops = {
+	.release	= vring_release,
+	.write		= vring_write,
+	.poll		= vring_poll,
+};
+
+asmlinkage long sys_vringfd(void __user *addr,
+			    unsigned num_descs,
+			    u16 __user *last_used)
+{
+	int fd, err;
+	struct file *filp;
+	struct vring_info *vr;
+
+	/* Must be a power of two, and representable by u16 */
+	if (!num_descs || (num_descs & (num_descs-1)) || num_descs > 65536) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	fd = get_unused_fd();
+	if (fd < 0) {
+		err = fd;
+		goto out;
+	}
+
+	filp = alloc_file(vring_mnt, dget(vring_mnt->mnt_root), FMODE_WRITE,
+			  &vring_fops);
+	if (!filp) {
+		err = -ENFILE;
+		goto put_fd;
+	}
+
+	filp->private_data = vr = kmalloc(sizeof(*vr), GFP_KERNEL);
+	if (!vr) {
+		err = -ENOMEM;
+		goto put_filp;
+	}
+
+	/* Set up pointers into ring. */
+	vring_init(&vr->ring, num_descs, addr, PAGE_SIZE);
+	init_waitqueue_head(&vr->poll_wait);
+	vr->last_used = last_used;
+	vr->mask = num_descs - 1;
+	vr->ops = NULL;
+	vr->used = NULL;
+
+	err = get_user(vr->last_avail, &vr->ring.avail->idx);
+	if (err)
+		goto free_vr;
+	
+	fd_install(fd, filp);
+	return fd;
+
+free_vr:
+	kfree(vr);
+put_filp:
+	put_filp(filp);
+put_fd:
+	put_unused_fd(fd);
+out:
+	return err;
+}
+
+/* Returns an error, or 0 (no buffers), or an id for vring_used_buffer() */
+int vring_get_buffer(struct vring_info *vr,
+		     struct iovec *in_iov,
+		     unsigned int *num_in, unsigned long *in_len,
+		     struct iovec *out_iov,
+		     unsigned int *num_out, unsigned long *out_len)
+{
+	unsigned int i, in = 0, out = 0;
+	unsigned long dummy;
+	u16 head;
+	struct vring_desc d;
+
+	if (unlikely(get_user(head, &vr->ring.avail->idx) != 0))
+		return -EFAULT;
+
+	if (vr->last_avail == head)
+		return 0;
+
+	if (!in_len)
+		in_len = &dummy;
+	if (!out_len)
+		out_len = &dummy;
+
+	*in_len = *out_len = 0;
+	
+	if (unlikely(get_user(head, &vr->ring.avail->ring[head]) != 0))
+		return -EFAULT;
+
+	i = head;
+	do {
+		if (unlikely(i >= vr->ring.num)) {
+			pr_debug("vring: bad index: %u\n", i);
+			return -EINVAL;
+		}
+
+		if (copy_from_user(&d, &vr->ring.desc[i], sizeof(d)) != 0)
+			return -EFAULT;
+
+		if (d.flags & VRING_DESC_F_WRITE) {
+			/* Check for length and iovec overflows */
+			if (!num_in)
+				return -EINVAL;
+			if (in == *num_in || *in_len + d.len < *in_len)
+				return -E2BIG;
+			in_iov[in].iov_len = d.len;
+			*in_len += d.len;
+			in_iov[in].iov_base = (void __user*)(long)d.addr;
+			in++;
+		} else {
+			if (!num_out)
+				return -EINVAL;
+			if (out == *num_out || *out_len + d.len < *out_len)
+				return -E2BIG;
+			out_iov[out].iov_len = d.len;
+			*out_len += d.len;
+			out_iov[out].iov_base = (void __user*)(long)d.addr;
+			out++;
+		}
+
+		i = d.next;
+	} while (d.flags & VRING_DESC_F_NEXT);
+
+	if (num_in)
+		*num_in = in;
+	if (num_out)
+		*num_out = out;
+
+	/* 0 is a valid head, so add one. */
+	vr->last_avail++;
+	return head + 1;
+}
+EXPORT_SYMBOL_GPL(vring_get_buffer);
+
+void vring_used_buffer(struct vring_info *vr, int id, u32 len)
+{
+	struct vring_used_elem used;
+	u16 used_idx;
+
+	BUG_ON(id <= 0 || id > vr->ring.num);
+
+	used.id = id - 1;
+	used.len = len;
+	if (get_user(used_idx, &vr->ring.used->idx) != 0)
+		return;
+
+	copy_to_user(&vr->ring.used->ring[used_idx & vr->mask], &used,
+		     sizeof(used));
+	wmb();
+	used_idx++;
+	put_user(used_idx, &vr->ring.used->idx);
+}
+EXPORT_SYMBOL_GPL(vring_used_buffer);
+
+void vring_used_buffer_atomic(struct vring_info *vr, int id, u32 len)
+{
+	struct vring_used_elem *used;
+
+	BUG_ON(id <= 0 || id > vr->ring.num);
+	BUG_ON(!vr->used);
+
+	used = &vr->used->ring[vr->used->idx & vr->mask];
+	used->id = id - 1;
+	used->len = len;
+	/* Make sure buffer is written before we update index. */
+	wmb();
+	vr->used->idx++;
+}
+EXPORT_SYMBOL_GPL(vring_used_buffer_atomic);
+
+void vring_wake(struct vring_info *vr)
+{
+	wake_up(&vr->poll_wait);
+}
+EXPORT_SYMBOL_GPL(vring_wake);
+
+struct vring_info *vring_attach(int fd, const struct vring_ops *ops,
+				void *data, bool atomic_use)
+{
+	struct file *filp;
+	struct vring_info *vr;
+
+	/* Must be a valid fd, and must be one of ours. */
+	filp = fget(fd);
+	if (!filp) {
+		vr = ERR_PTR(-EBADF);
+		goto out;
+	}
+
+	if (filp->f_op != &vring_fops) {
+		vr = ERR_PTR(-EBADF);
+		goto fput;
+	}
+
+	/* Mutex simply protects against parallel vring_attach. */
+	mutex_lock(&vring_lock);
+	vr = filp->private_data;
+	if (vr->ops) {
+		vr = ERR_PTR(-EBUSY);
+		goto unlock;
+	}
+
+	/* If they want to use atomically, we have to map the page. */
+	if (atomic_use) {
+		if (get_user_pages(current, current->mm,
+				   (unsigned long)vr->ring.used, 1, 1, 1,
+				   &vr->used_page, NULL) != 1) {
+			vr = ERR_PTR(-EFAULT);
+			goto unlock;
+		}
+		vr->used = kmap(vr->used_page);
+		if (!vr->used) {
+			put_page(vr->used_page);
+			vr = ERR_PTR(-ENOMEM);
+			goto unlock;
+		}
+	}
+
+	vr->ops = ops;
+	vr->ops_data = data;
+
+unlock:
+	mutex_unlock(&vring_lock);
+fput:
+	fput(filp);
+out:
+	return vr;
+}
+EXPORT_SYMBOL_GPL(vring_attach);
+
+static int vringfs_get_sb(struct file_system_type *fs_type,
+			  int flags, const char *dev_name, void *data,
+			  struct vfsmount *mnt)
+{
+	return get_sb_pseudo(fs_type, "vring", NULL, VRINGFS_SUPER_MAGIC, mnt);
+}
+
+static struct file_system_type vring_fs_type = {
+	.name		= "vringfs",
+	.get_sb		= vringfs_get_sb,
+	.kill_sb	= kill_anon_super,
+};
+
+static int init(void)
+{
+	register_filesystem(&vring_fs_type);
+	vring_mnt = kern_mount(&vring_fs_type);
+	return 0;
+}
+
+module_init(init);
diff -r 99132ad16999 include/asm-x86/unistd_32.h
--- a/include/asm-x86/unistd_32.h	Sat Apr 05 21:20:32 2008 +1100
+++ b/include/asm-x86/unistd_32.h	Sat Apr 05 21:31:40 2008 +1100
@@ -332,6 +332,7 @@
 #define __NR_fallocate		324
 #define __NR_timerfd_settime	325
 #define __NR_timerfd_gettime	326
+#define __NR_vringfd		327
 
 #ifdef __KERNEL__
 
diff -r 99132ad16999 include/linux/magic.h
--- a/include/linux/magic.h	Sat Apr 05 21:20:32 2008 +1100
+++ b/include/linux/magic.h	Sat Apr 05 21:31:40 2008 +1100
@@ -41,5 +41,6 @@
 
 #define FUTEXFS_SUPER_MAGIC	0xBAD1DEA
 #define INOTIFYFS_SUPER_MAGIC	0x2BAD1DEA
+#define VRINGFS_SUPER_MAGIC	0xB1BBAD
 
 #endif /* __LINUX_MAGIC_H__ */
diff -r 99132ad16999 include/linux/syscalls.h
--- a/include/linux/syscalls.h	Sat Apr 05 21:20:32 2008 +1100
+++ b/include/linux/syscalls.h	Sat Apr 05 21:31:40 2008 +1100
@@ -614,6 +614,7 @@ asmlinkage long sys_timerfd_gettime(int 
 asmlinkage long sys_timerfd_gettime(int ufd, struct itimerspec __user *otmr);
 asmlinkage long sys_eventfd(unsigned int count);
 asmlinkage long sys_fallocate(int fd, int mode, loff_t offset, loff_t len);
+asmlinkage long sys_vringfd(void __user *, unsigned num, u16 __user *);
 
 int kernel_execve(const char *filename, char *const argv[], char *const envp[]);
 
diff -r 99132ad16999 include/linux/vring.h
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/include/linux/vring.h	Sat Apr 05 21:31:40 2008 +1100
@@ -0,0 +1,54 @@
+/* Ring-buffer file descriptor implementation.
+ *
+ *  Copyright 2008 Rusty Russell IBM Corporation
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ */
+#ifndef _LINUX_VRING_H
+#define _LINUX_VRING_H
+
+/* All members are optional */
+struct vring_ops
+{
+	/* Cleanup */
+	void (*destroy)(void *);
+
+	/* Returns number of used buffers, or negative errno. */
+	int (*pull)(void *);
+
+	/* Returns 0 or negative errno. */
+	int (*push)(void *);
+};
+
+/* If they want to call vring_used_buffer_atomic(), set atomic_use.
+ * This currently means that the userspace used buffer must fit in a page. */
+struct vring_info *vring_attach(int fd, const struct vring_ops *ops,
+				void *data, bool atomic_use);
+
+struct iovec;
+
+/* Returns an error, or 0 (no buffers), or an id for vring_used_buffer() */
+int vring_get_buffer(struct vring_info *vr,
+		     struct iovec *in_iov,
+		     unsigned int *num_in, unsigned long *in_len,
+		     struct iovec *out_iov,
+		     unsigned int *num_out, unsigned long *out_len);
+
+void vring_used_buffer(struct vring_info *vr, int id, u32 len);
+
+void vring_used_buffer_atomic(struct vring_info *vr, int id, u32 len);
+
+void vring_wake(struct vring_info *vr);
+#endif /* _LINUX_VRING_H */
diff -r 99132ad16999 kernel/sys_ni.c
--- a/kernel/sys_ni.c	Sat Apr 05 21:20:32 2008 +1100
+++ b/kernel/sys_ni.c	Sat Apr 05 21:31:40 2008 +1100
@@ -161,3 +161,4 @@ cond_syscall(compat_sys_timerfd_settime)
 cond_syscall(compat_sys_timerfd_settime);
 cond_syscall(compat_sys_timerfd_gettime);
 cond_syscall(sys_eventfd);
+cond_syscall(sys_vringfd);

Re: [PATCH RFC 1/5] vringfd syscall

[Jonathan Corbet]

Hey, Rusty,

> For virtualization, we've developed virtio_ring for efficient communication.
> This would also work well for userspace-kernel communication, particularly
> for things like the tun device.  By using the same ABI, we can join guests
> to the host kernel trivially.

I'm *sure* you meant to document that somewhat non-trivial proposed new
kernel API as soon as you got a moment.  But, since I went to the
trouble of reverse engineering it, I decided not to wait and to make a
little unreliable guide of my own.  For those who are curious about how
vringfd() is meant to work (or about how badly I misunderstood it), the
info is at:

	http://lwn.net/SubscriberLink/276856/8c927025c53ad7ce/

Hopefully it will be helpful,

jon

Re: [PATCH RFC 1/5] vringfd syscall

[Rusty Russell]

On Tuesday 08 April 2008 03:54:34 Jonathan Corbet wrote:
> Hey, Rusty,
>
> > For virtualization, we've developed virtio_ring for efficient
> > communication. This would also work well for userspace-kernel
> > communication, particularly for things like the tun device.  By using the
> > same ABI, we can join guests to the host kernel trivially.
>
> I'm *sure* you meant to document that somewhat non-trivial proposed new
> kernel API as soon as you got a moment.

Actually, yes.  But I wanted to get it out there before I start the treck 
across to the virtualization summit.

A few points:
'The page alignment for the used array is important - that array might be 
mapped separately into kernel space.'
   Well, the used array is written by one side only, so it's possible to split 
the ring here and make each part r/o to the other side.  More importantly, a 
page boundary is almost certainly a cacheline boundary, and we already have a 
userspace interface for it.

'Note that the flags fields in the vring_avail and vring_used structures 
appear to be unused.'
   virtio uses these for wakeup/interrupt suppression.  It's a cheap way to 
avoid hypercalls, and we can use them the same way to avoid system calls (you 
set the suppression bit while you're actually looking at the ring).

The need for the kmap (and hence the atomic horror) has now been alleviated: I 
changed the shinfo destructor code to allow the destructor to hold onto the 
skb data so it can queue it and free it later.

BTW, the only place currently where both output and input buffers are used is 
the virtio_blk driver doing a read, where the header describes the operation, 
and the other buffers are overwritten with the data.

Thanks!
Rusty.

Re: [PATCH RFC 1/5] vringfd syscall

[Arnd Bergmann]

On Saturday 05 April 2008, Rusty Russell wrote:
> +asmlinkage long sys_vringfd(void __user *addr,
> +                           unsigned num_descs,
> +                           u16 __user *last_used)
> +{
> +       int fd, err;
> +       struct file *filp;
> +       struct vring_info *vr;
> +
> +       /* Must be a power of two, and representable by u16 */
> +       if (!num_descs || (num_descs & (num_descs-1)) || num_descs > 65536) {
> +               err = -EINVAL;
> +               goto out;
> +       }
> +
> +       fd = get_unused_fd();
> +       if (fd < 0) {
> +               err = fd;
> +               goto out;
> +       }
> +
> +       filp = alloc_file(vring_mnt, dget(vring_mnt->mnt_root), FMODE_WRITE,
> +                         &vring_fops);
> +       if (!filp) {
> +               err = -ENFILE;
> +               goto put_fd;
> +       }

This looks like a candidate for anon_inode_getfd(), which would let you
get rid of the code for registering your own file system.

	Arnd <><

Re: [PATCH RFC 1/5] vringfd syscall

[Jeremy Fitzhardinge]

Rusty Russell wrote:
> +asmlinkage long sys_vringfd(void __user *addr,
> +			    unsigned num_descs,
> +			    u16 __user *last_used)
> +{
> +	int fd, err;
> +	struct file *filp;
> +	struct vring_info *vr;
> +
> +	/* Must be a power of two, and representable by u16 */
> +	if (!num_descs || (num_descs & (num_descs-1)) || num_descs > 65536) {
>   

65536 doesn't fit in u16.

    J

Re: [PATCH RFC 1/5] vringfd syscall

[Marcelo Tosatti]

Hi Rusty,

A couple comments below.

On Sat, Apr 05, 2008 at 10:02:08PM +1000, Rusty Russell wrote:

> +static unsigned int vring_poll(struct file *filp,
> +			       struct poll_table_struct *poll)
> +{
> +	struct vring_info *vr = filp->private_data;
> +	int err;
> +	unsigned int mask;
> +	u16 used, last_used;
> +
> +	/* Some uses of vrings require updating in user context.  This
> +	 * is best done close to the caller, ie. here. */
> +	if (vr->ops && vr->ops->pull) {
> +		err = vr->ops->pull(vr->ops_data);
> +		if (unlikely(err < 0))
> +			return err;
> +
> +		if (err > 0) {
> +			/* Buffers have been used, no need to check indices */
> +			mask = POLLIN | POLLRDNORM;
> +			goto poll_wait;
> +		}
> +	}
> +
> +	err = get_user(used, &vr->ring.used->idx);
> +	if (unlikely(err))
> +		return err;
> +
> +	err = get_user(last_used, vr->last_used);
> +	if (unlikely(err))
> +		return err;
> +
> +	/* More buffers have been used?  It's 'readable'. */
> +	if (used != last_used)
> +		mask = POLLIN | POLLRDNORM;
> +	else
> +		mask = 0;
> +
> +poll_wait:
> +	poll_wait(filp, &vr->poll_wait, poll);
> +
> +	return mask;
> +}

I suppose you are doing data copy in ->poll instead of ->read to save
a system call? This is weird, not conformant to what the interface is
supposed to do.

This way select/poll syscalls might block in userspace datacopy. The
application might have a higher priority fd in the fdset to be informed
of, for example.

So why not change this to the common arrangement, with vring_poll adding
the waitqueue with poll_wait() and vring_read doing the actual data copy ?

> +struct vring_info *vring_attach(int fd, const struct vring_ops *ops,
> +				void *data, bool atomic_use)
> +{
> +	struct file *filp;
> +	struct vring_info *vr;
> +
> +	/* Must be a valid fd, and must be one of ours. */
> +	filp = fget(fd);
> +	if (!filp) {
> +		vr = ERR_PTR(-EBADF);
> +		goto out;
> +	}
> +
> +	if (filp->f_op != &vring_fops) {
> +		vr = ERR_PTR(-EBADF);
> +		goto fput;
> +	}
> +
> +	/* Mutex simply protects against parallel vring_attach. */
> +	mutex_lock(&vring_lock);
> +	vr = filp->private_data;
> +	if (vr->ops) {
> +		vr = ERR_PTR(-EBUSY);
> +		goto unlock;
> +	}
> +
> +	/* If they want to use atomically, we have to map the page. */
> +	if (atomic_use) {
> +		if (get_user_pages(current, current->mm,
> +				   (unsigned long)vr->ring.used, 1, 1, 1,
> +				   &vr->used_page, NULL) != 1) {

Can't the same be achieved by the app mlocking the vring pages, which
then goes through standard rlimit checking ?

Re: [PATCH RFC 1/5] vringfd syscall

[Marcelo Tosatti]

On Sat, Apr 12, 2008 at 02:18:20PM -0300, Marcelo Tosatti wrote:

> > +		mask = POLLIN | POLLRDNORM;
> > +	else
> > +		mask = 0;
> > +
> > +poll_wait:
> > +	poll_wait(filp, &vr->poll_wait, poll);
> > +
> > +	return mask;
> > +}
> 
> I suppose you are doing data copy in ->poll instead of ->read to save
> a system call? This is weird, not conformant to what the interface is
> supposed to do.
> 
> This way select/poll syscalls might block in userspace datacopy. The
> application might have a higher priority fd in the fdset to be informed
> of, for example.
> 
> So why not change this to the common arrangement, with vring_poll adding
> the waitqueue with poll_wait() and vring_read doing the actual data copy ?
> 
> > +struct vring_info *vring_attach(int fd, const struct vring_ops *ops,
> > +				void *data, bool atomic_use)
> > +{
> > +	struct file *filp;
> > +	struct vring_info *vr;
> > +
> > +	/* Must be a valid fd, and must be one of ours. */
> > +	filp = fget(fd);
> > +	if (!filp) {
> > +		vr = ERR_PTR(-EBADF);
> > +		goto out;
> > +	}
> > +
> > +	if (filp->f_op != &vring_fops) {
> > +		vr = ERR_PTR(-EBADF);
> > +		goto fput;
> > +	}
> > +
> > +	/* Mutex simply protects against parallel vring_attach. */
> > +	mutex_lock(&vring_lock);
> > +	vr = filp->private_data;
> > +	if (vr->ops) {
> > +		vr = ERR_PTR(-EBUSY);
> > +		goto unlock;
> > +	}
> > +
> > +	/* If they want to use atomically, we have to map the page. */
> > +	if (atomic_use) {
> > +		if (get_user_pages(current, current->mm,
> > +				   (unsigned long)vr->ring.used, 1, 1, 1,
> > +				   &vr->used_page, NULL) != 1) {
> 
> Can't the same be achieved by the app mlocking the vring pages, which
> then goes through standard rlimit checking ?

Oh, this is a driver API to allow the copy to take place in atomic
contexes. You might want to add some sort of limit enforcement.

Also forgot mmap_sem there.

Re: [PATCH RFC 1/5] vringfd syscall

[Rusty Russell]

On Sunday 13 April 2008 03:18:20 Marcelo Tosatti wrote:
> I suppose you are doing data copy in ->poll instead of ->read to save
> a system call? This is weird, not conformant to what the interface is
> supposed to do.

Hi Marcelo,

   Thanks for reading this.

   Yeah, I dislike it too.  The main motivation was that not all rings need to 
be "pulled" in this way, but it turns out that userspace can detect this 
(poll, check ring, if nothing, try read, check ring).  So next version will 
unscrew the poll callback.

> > +	/* If they want to use atomically, we have to map the page. */
> > +	if (atomic_use) {
> > +		if (get_user_pages(current, current->mm,
> > +				   (unsigned long)vr->ring.used, 1, 1, 1,
> > +				   &vr->used_page, NULL) != 1) {
>
> Can't the same be achieved by the app mlocking the vring pages, which
> then goes through standard rlimit checking ?

Tun/tap writes to this from any context.  Fortunately, this too is changing, 
as I came up with a sane way of queueing to userspace context for this case.

Expect updated patches Monday.

Thanks!
Rusty.