[PATCH 2/5] /dev/vring: simple userspace-kernel ringbuffer interface.

From: Rusty Russell
Date: Fri Apr 18 2008 - 00:40:49 EST


virtio introduced a ring structure ABI for guest-host communications
(currently used by lguest and kvm). Using this same ABI, we can
create a nice fd version.

This is useful for efficiently passing packets to and from the tun,
for example.

Signed-off-by: Rusty Russell <rusty@xxxxxxxxxxxxxxx>
---
drivers/char/Kconfig | 9 +
drivers/char/Makefile | 2
drivers/char/vring.c | 400 ++++++++++++++++++++++++++++++++++++++++++++++++++
include/linux/vring.h | 58 +++++++
4 files changed, 469 insertions(+)

diff -r b2d9869d338f drivers/char/Kconfig
--- a/drivers/char/Kconfig Fri Apr 18 10:33:58 2008 +1000
+++ b/drivers/char/Kconfig Fri Apr 18 13:35:16 2008 +1000
@@ -1049,5 +1049,14 @@ config DEVPORT

source "drivers/s390/char/Kconfig"

+config VRING
+ tristate "/dev/vring 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, but there's a part of you that wants to say M.
endmenu

diff -r b2d9869d338f drivers/char/Makefile
--- a/drivers/char/Makefile Fri Apr 18 10:33:58 2008 +1000
+++ b/drivers/char/Makefile Fri Apr 18 13:35:16 2008 +1000
@@ -112,6 +112,8 @@ obj-$(CONFIG_JS_RTC) += js-rtc.o
obj-$(CONFIG_JS_RTC) += js-rtc.o
js-rtc-y = rtc.o

+obj-$(CONFIG_VRING) += vring.o
+
# Files generated that shall be removed upon make clean
clean-files := consolemap_deftbl.c defkeymap.c

diff -r b2d9869d338f drivers/char/vring.c
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/char/vring.c Fri Apr 18 13:35:16 2008 +1000
@@ -0,0 +1,400 @@
+/* Ring-buffer device 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/module.h>
+#include <linux/miscdevice.h>
+
+struct vring_info {
+ struct mutex lock;
+
+ struct vring ring;
+ u16 mask;
+ u16 last_used;
+
+ const struct vring_ops *ops;
+ void *ops_data;
+
+ /* Waitqueue for poll() */
+ wait_queue_head_t poll_wait;
+};
+
+static unsigned int vring_poll(struct file *filp,
+ struct poll_table_struct *poll)
+{
+ struct vring_info *vr = filp->private_data;
+ unsigned int mask;
+ u16 used = 0;
+
+ /* Poll can't error, so let's not go silly here. */
+ get_user(used, &vr->ring.used->idx);
+
+ /* More buffers have been used? It's 'readable'. */
+ if (used != vr->last_used)
+ mask = POLLIN | POLLRDNORM;
+ else {
+ mask = 0;
+ /* If we need to pull, it's also readable. */
+ mutex_lock(&vr->lock);
+ if (vr->ops && vr->ops->needs_pull) {
+ if (vr->ops->needs_pull(vr->ops_data))
+ mask = POLLIN | POLLRDNORM;
+ }
+ mutex_unlock(&vr->lock);
+ }
+
+ poll_wait(filp, &vr->poll_wait, poll);
+
+ return mask;
+}
+
+/* Read may not be necessary for all use cases, in fact. */
+static ssize_t vring_read(struct file *filp, char __user *buf,
+ size_t size, loff_t *off)
+{
+ struct vring_info *vr = filp->private_data;
+ int err;
+
+ /* Some uses of vrings require updating in user context. This
+ * is best done close to the caller, ie. here. */
+ mutex_lock(&vr->lock);
+ if (vr->ops && vr->ops->pull)
+ err = vr->ops->pull(vr->ops_data);
+ else
+ err = 0;
+ mutex_unlock(&vr->lock);
+
+ /* Update our last_used value to clear the poll. */
+ if (!err)
+ err = get_user(vr->last_used, &vr->ring.used->idx);
+
+ return err;
+}
+
+/* Write kicks the other end to say we have buffers. */
+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;
+ int err;
+
+ mutex_lock(&vr->lock);
+ if (vr->ops && vr->ops->push)
+ err = vr->ops->push(vr->ops_data);
+ else
+ err = 0;
+ mutex_unlock(&vr->lock);
+
+ return err;
+}
+
+/* We assume anyone attached holds a reference, so this won't mess them up */
+static int vring_release(struct inode *inode, struct file *filp)
+{
+ struct vring_info *vr = filp->private_data;
+
+ kfree(vr);
+ return 0;
+}
+
+static int vring_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+ unsigned long size, num_descs;
+ struct vring_info *vr = filp->private_data;
+ int err;
+
+ /* We overload mmap's offset to hold the ring number. */
+ num_descs = vma->vm_pgoff;
+
+ /* Must be a power of two, and limit indices to a u16. */
+ if (!num_descs || (num_descs & (num_descs-1)) || num_descs > 65536)
+ return -EINVAL;
+
+ /* mmap size must be what we expect for such a ring. */
+ size = vma->vm_end - vma->vm_start;
+ if (size != ALIGN(vring_size(num_descs, PAGE_SIZE), PAGE_SIZE))
+ return -EINVAL;
+
+ /* We only let them map this in one place. */
+ mutex_lock(&vr->lock);
+ if (vr->ring.num != 0) {
+ err = -EBUSY;
+ goto unlock;
+ }
+
+ vring_init(&vr->ring, num_descs, (void *)vma->vm_start, PAGE_SIZE);
+
+ vr->mask = num_descs - 1;
+ err = 0;
+
+unlock:
+ mutex_unlock(&vr->lock);
+ return err;
+}
+
+static int vring_open(struct inode *in, struct file *filp)
+{
+ struct vring_info *vr;
+
+ filp->private_data = vr = kzalloc(sizeof(*vr), GFP_KERNEL);
+ if (!vr)
+ return -ENOMEM;
+
+ init_waitqueue_head(&vr->poll_wait);
+ mutex_init(&vr->lock);
+ return 0;
+}
+
+static const struct file_operations vring_fops = {
+ .open = vring_open,
+ .release = vring_release,
+ .mmap = vring_mmap,
+ .read = vring_read,
+ .write = vring_write,
+ .poll = vring_poll,
+};
+
+/**
+ * vring_get_buffer - get a buffer from the vring
+ * @vr: the vring
+ * @in_iov: the iovec array for input buffers
+ * @num_in: the size of the in_iov array, updated by this function.
+ * @in_len: the total length of in_iov after this function.
+ * @out_iov: the iovec array for output buffers
+ * @num_out: the size of the ut_iov array, updated by this function.
+ * @out_len: the total length of out_iov after this function.
+ *
+ * A vring buffer is an array of input and output parts. This gets the next
+ * available buffer, and returns a non-zero id which is handed back to
+ * vring_used_buffer() once you're finished with the buffer. A zero return
+ * means no available buffers, negative for error.
+ */
+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 avail, last_avail, head;
+ struct vring_desc d;
+
+ if (unlikely(get_user(avail, &vr->ring.avail->idx)))
+ return -EFAULT;
+ if (unlikely(get_user(last_avail, &vring_last_avail(&vr->ring))))
+ return -EFAULT;
+
+ if (last_avail == avail)
+ 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[last_avail
+ & vr->mask])))
+ 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) {
+ pr_debug("vring: writable desc %u in ring %p\n",
+ i, vr->ring.desc);
+ 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) {
+ pr_debug("vring: readable desc %u in ring %p\n",
+ i, vr->ring.desc);
+ 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;
+
+ last_avail++;
+ put_user(last_avail, &vring_last_avail(&vr->ring));
+
+ /* 0 is a valid head, so add one. */
+ return head + 1;
+}
+EXPORT_SYMBOL_GPL(vring_get_buffer);
+
+/**
+ * vring_used_buffer - return a used buffer to the vring
+ * @vr: the vring
+ * @id: the id returned from vring_get_buffer
+ * @len: the total bytes *written* to the 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;
+
+ if (copy_to_user(&vr->ring.used->ring[used_idx & vr->mask], &used,
+ sizeof(used)))
+ return;
+
+ wmb();
+ used_idx++;
+ put_user(used_idx, &vr->ring.used->idx);
+}
+EXPORT_SYMBOL_GPL(vring_used_buffer);
+
+void vring_wake(struct vring_info *vr)
+{
+ wake_up(&vr->poll_wait);
+}
+EXPORT_SYMBOL_GPL(vring_wake);
+
+/**
+ * vring_get - check out a vring file descriptor
+ * @filp: the file structure to attach to (eg. from fget()).
+ *
+ * Userspace opens /dev/vring and mmaps it, then hands that fd to the
+ * kernel subsystem it wants to communicate with. That subsystem uses
+ * this routine and vring_set_ops() to attach to it.
+ *
+ * This simply checks that it really is a vring fd (otherwise it
+ * returns NULL), the other routine checks that it's not already
+ * attached.
+ */
+struct vring_info *vring_get(struct file *filp)
+{
+ /* Must be one of ours. */
+ if (filp->f_op != &vring_fops)
+ return NULL;
+
+ return filp->private_data;
+}
+EXPORT_SYMBOL_GPL(vring_get);
+
+/**
+ * vring_set_ops - attach operations to a vring file descriptor.
+ * @vr: the vring_info returned from vring_get.
+ * @ops: the operations to attach.
+ * @ops_data: the argument to the ops callbacks.
+ *
+ * This is called after vring_get(): the reason for the two-part
+ * process is that the ops can be called before vring_set_ops returns
+ * (we don't do locking), so you really need to set things up before
+ * this call.
+ *
+ * This simply checks that the ring is not already attached to something,
+ * then sets the ops.
+ */
+int vring_set_ops(struct vring_info *vr,
+ const struct vring_ops *ops, void *ops_data)
+{
+ int err;
+
+ mutex_lock(&vr->lock);
+ if (vr->ops) {
+ err = -EBUSY;
+ goto unlock;
+ }
+
+ /* We don't lock, so make sure we get this in the right order. */
+ vr->ops_data = ops_data;
+ wmb();
+ vr->ops = ops;
+
+ err = 0;
+unlock:
+ mutex_unlock(&vr->lock);
+ local_irq_enable();
+ return err;
+}
+EXPORT_SYMBOL_GPL(vring_set_ops);
+
+/**
+ * vring_unset_ops - remove operations to a vring file descriptor.
+ * @vr: the vring_info previously successfully vring_set_ops'd
+ */
+void vring_unset_ops(struct vring_info *vr)
+{
+ BUG_ON(!vr->ops);
+ mutex_lock(&vr->lock);
+ vr->ops = NULL;
+ mutex_unlock(&vr->lock);
+}
+EXPORT_SYMBOL_GPL(vring_unset_ops);
+
+static struct miscdevice vring_dev = {
+ .minor = MISC_DYNAMIC_MINOR,
+ .name = KBUILD_MODNAME,
+ .fops = &vring_fops,
+};
+
+static int __init init(void)
+{
+ return misc_register(&vring_dev);
+}
+
+static void __exit fini(void)
+{
+ misc_deregister(&vring_dev);
+}
+
+module_init(init);
+module_exit(fini);
diff -r b2d9869d338f include/linux/vring.h
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/include/linux/vring.h Fri Apr 18 13:35:16 2008 +1000
@@ -0,0 +1,58 @@
+/* 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
+
+/**
+ * vring_ops - operations for a vring fd.
+ * @needs_pull: more data is pending, need to call pull.
+ * @pull: callback when read() is called to report used buffers.
+ * @push: callback when write() is called to notify of added buffers.
+ *
+ * Any of these callbacks can be NULL, if you don't need them.
+ */
+struct vring_ops {
+ bool (*needs_pull)(void *ops_data);
+
+ /* Returns 0 or negative errno. */
+ int (*pull)(void *ops_data);
+
+ /* Returns 0 or negative errno. */
+ int (*push)(void *ops_data);
+};
+
+struct file;
+
+struct vring_info *vring_get(struct file *filp);
+int vring_set_ops(struct vring_info *,
+ const struct vring_ops *ops, void *ops_data);
+void vring_unset_ops(struct vring_info *vr);
+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_wake(struct vring_info *vr);
+#endif /* _LINUX_VRING_H */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/