Merge git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml

Pull UML updates from Richard Weinberger:

 - a new and faster epoll based IRQ controller and NIC driver

 - misc fixes and janitorial updates

* git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml:
  Fix vector raw inintialization logic
  Migrate vector timers to new timer API
  um: Compile with modern headers
  um: vector: Fix an error handling path in 'vector_parse()'
  um: vector: Fix a memory allocation check
  um: vector: fix missing unlock on error in vector_net_open()
  um: Add missing EXPORT for free_irq_by_fd()
  High Performance UML Vector Network Driver
  Epoll based IRQ controller
  um: Use POSIX ucontext_t instead of struct ucontext
  um: time: Use timespec64 for persistent clock
  um: Restore symbol versions for __memcpy and memcpy

arch/um/Kconfig.net

@@ -109,6 +109,17 @@ config UML_NET_DAEMON
         more than one without conflict.  If you don't need UML networking,
         say N.
 
+config UML_NET_VECTOR
+	bool "Vector I/O high performance network devices"
+	depends on UML_NET
+	help
+	This User-Mode Linux network driver uses multi-message send
+	and receive functions. The host running the UML guest must have
+	a linux kernel version above 3.0 and a libc version > 2.13.
+	This driver provides tap, raw, gre and l2tpv3 network transports
+	with up to 4 times higher network throughput than the UML network
+	drivers.
+
 config UML_NET_VDE
 	bool "VDE transport"
 	depends on UML_NET

arch/um/drivers/Makefile

@@ -9,6 +9,7 @@
 slip-objs := slip_kern.o slip_user.o
 slirp-objs := slirp_kern.o slirp_user.o
 daemon-objs := daemon_kern.o daemon_user.o
+vector-objs := vector_kern.o vector_user.o vector_transports.o
 umcast-objs := umcast_kern.o umcast_user.o
 net-objs := net_kern.o net_user.o
 mconsole-objs := mconsole_kern.o mconsole_user.o
@@ -43,6 +44,7 @@ obj-$(CONFIG_STDERR_CONSOLE) += stderr_console.o
 obj-$(CONFIG_UML_NET_SLIP) += slip.o slip_common.o
 obj-$(CONFIG_UML_NET_SLIRP) += slirp.o slip_common.o
 obj-$(CONFIG_UML_NET_DAEMON) += daemon.o 
+obj-$(CONFIG_UML_NET_VECTOR) += vector.o
 obj-$(CONFIG_UML_NET_VDE) += vde.o
 obj-$(CONFIG_UML_NET_MCAST) += umcast.o
 obj-$(CONFIG_UML_NET_PCAP) += pcap.o
@@ -61,7 +63,7 @@ obj-$(CONFIG_BLK_DEV_COW_COMMON) += cow_user.o
 obj-$(CONFIG_UML_RANDOM) += random.o
 
 # pcap_user.o must be added explicitly.
-USER_OBJS := fd.o null.o pty.o tty.o xterm.o slip_common.o pcap_user.o vde_user.o
+USER_OBJS := fd.o null.o pty.o tty.o xterm.o slip_common.o pcap_user.o vde_user.o vector_user.o
 CFLAGS_null.o = -DDEV_NULL=$(DEV_NULL_PATH)
 
 include arch/um/scripts/Makefile.rules

arch/um/drivers/chan_kern.c

@@ -171,56 +171,19 @@ int enable_chan(struct line *line)
 	return err;
 }
 
-/* Items are added in IRQ context, when free_irq can't be called, and
- * removed in process context, when it can.
- * This handles interrupt sources which disappear, and which need to
- * be permanently disabled.  This is discovered in IRQ context, but
- * the freeing of the IRQ must be done later.
- */
-static DEFINE_SPINLOCK(irqs_to_free_lock);
-static LIST_HEAD(irqs_to_free);
-
-void free_irqs(void)
-{
-	struct chan *chan;
-	LIST_HEAD(list);
-	struct list_head *ele;
-	unsigned long flags;
-
-	spin_lock_irqsave(&irqs_to_free_lock, flags);
-	list_splice_init(&irqs_to_free, &list);
-	spin_unlock_irqrestore(&irqs_to_free_lock, flags);
-
-	list_for_each(ele, &list) {
-		chan = list_entry(ele, struct chan, free_list);
-
-		if (chan->input && chan->enabled)
-			um_free_irq(chan->line->driver->read_irq, chan);
-		if (chan->output && chan->enabled)
-			um_free_irq(chan->line->driver->write_irq, chan);
-		chan->enabled = 0;
-	}
-}
-
 static void close_one_chan(struct chan *chan, int delay_free_irq)
 {
-	unsigned long flags;
-
 	if (!chan->opened)
 		return;
 
-	if (delay_free_irq) {
-		spin_lock_irqsave(&irqs_to_free_lock, flags);
-		list_add(&chan->free_list, &irqs_to_free);
-		spin_unlock_irqrestore(&irqs_to_free_lock, flags);
-	}
-	else {
-		if (chan->input && chan->enabled)
-			um_free_irq(chan->line->driver->read_irq, chan);
-		if (chan->output && chan->enabled)
-			um_free_irq(chan->line->driver->write_irq, chan);
-		chan->enabled = 0;
-	}
+    /* we can safely call free now - it will be marked
+     *  as free and freed once the IRQ stopped processing
+     */
+	if (chan->input && chan->enabled)
+		um_free_irq(chan->line->driver->read_irq, chan);
+	if (chan->output && chan->enabled)
+		um_free_irq(chan->line->driver->write_irq, chan);
+	chan->enabled = 0;
 	if (chan->ops->close != NULL)
 		(*chan->ops->close)(chan->fd, chan->data);
 

arch/um/drivers/line.c

@@ -284,7 +284,7 @@ int line_setup_irq(int fd, int input, int output, struct line *line, void *data)
 	if (err)
 		return err;
 	if (output)
-		err = um_request_irq(driver->write_irq, fd, IRQ_WRITE,
+		err = um_request_irq(driver->write_irq, fd, IRQ_NONE,
 				     line_write_interrupt, IRQF_SHARED,
 				     driver->write_irq_name, data);
 	return err;

arch/um/drivers/net_kern.c

@@ -288,7 +288,7 @@ static void uml_net_user_timer_expire(struct timer_list *t)
 #endif
 }
 
-static void setup_etheraddr(struct net_device *dev, char *str)
+void uml_net_setup_etheraddr(struct net_device *dev, char *str)
 {
 	unsigned char *addr = dev->dev_addr;
 	char *end;
@@ -412,7 +412,7 @@ static void eth_configure(int n, void *init, char *mac,
 	 */
 	snprintf(dev->name, sizeof(dev->name), "eth%d", n);
 
-	setup_etheraddr(dev, mac);
+	uml_net_setup_etheraddr(dev, mac);
 
 	printk(KERN_INFO "Netdevice %d (%pM) : ", n, dev->dev_addr);
 

arch/um/drivers/random.c

@@ -13,6 +13,7 @@
 #include <linux/miscdevice.h>
 #include <linux/delay.h>
 #include <linux/uaccess.h>
+#include <init.h>
 #include <irq_kern.h>
 #include <os.h>
 
@@ -154,7 +155,14 @@ static int __init rng_init (void)
 /*
  * rng_cleanup - shutdown RNG module
  */
-static void __exit rng_cleanup (void)
+
+static void cleanup(void)
+{
+	free_irq_by_fd(random_fd);
+	os_close_file(random_fd);
+}
+
+static void __exit rng_cleanup(void)
 {
 	os_close_file(random_fd);
 	misc_deregister (&rng_miscdev);
@@ -162,6 +170,7 @@ static void __exit rng_cleanup (void)
 
 module_init (rng_init);
 module_exit (rng_cleanup);
+__uml_exitcall(cleanup);
 
 MODULE_DESCRIPTION("UML Host Random Number Generator (RNG) driver");
 MODULE_LICENSE("GPL");

arch/um/drivers/ubd_kern.c

@@ -1587,11 +1587,11 @@ int io_thread(void *arg)
 
 		do {
 			res = os_write_file(kernel_fd, ((char *) io_req_buffer) + written, n);
-			if (res > 0) {
+			if (res >= 0) {
 				written += res;
 			} else {
 				if (res != -EAGAIN) {
-					printk("io_thread - read failed, fd = %d, "
+					printk("io_thread - write failed, fd = %d, "
 					       "err = %d\n", kernel_fd, -n);
 				}
 			}

arch/um/drivers/vector_kern.h

+/*
+ * Copyright (C) 2002 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
+ * Licensed under the GPL
+ */
+
+#ifndef __UM_VECTOR_KERN_H
+#define __UM_VECTOR_KERN_H
+
+#include <linux/netdevice.h>
+#include <linux/platform_device.h>
+#include <linux/skbuff.h>
+#include <linux/socket.h>
+#include <linux/list.h>
+#include <linux/ctype.h>
+#include <linux/workqueue.h>
+#include <linux/interrupt.h>
+#include "vector_user.h"
+
+/* Queue structure specially adapted for multiple enqueue/dequeue
+ * in a mmsgrecv/mmsgsend context
+ */
+
+/* Dequeue method */
+
+#define QUEUE_SENDMSG 0
+#define QUEUE_SENDMMSG 1
+
+#define VECTOR_RX 1
+#define VECTOR_TX (1 << 1)
+#define VECTOR_BPF (1 << 2)
+#define VECTOR_QDISC_BYPASS (1 << 3)
+
+#define ETH_MAX_PACKET 1500
+#define ETH_HEADER_OTHER 32 /* just in case someone decides to go mad on QnQ */
+
+struct vector_queue {
+	struct mmsghdr *mmsg_vector;
+	void **skbuff_vector;
+	 /* backlink to device which owns us */
+	struct net_device *dev;
+	spinlock_t head_lock;
+	spinlock_t tail_lock;
+	int queue_depth, head, tail, max_depth, max_iov_frags;
+	short options;
+};
+
+struct vector_estats {
+	uint64_t rx_queue_max;
+	uint64_t rx_queue_running_average;
+	uint64_t tx_queue_max;
+	uint64_t tx_queue_running_average;
+	uint64_t rx_encaps_errors;
+	uint64_t tx_timeout_count;
+	uint64_t tx_restart_queue;
+	uint64_t tx_kicks;
+	uint64_t tx_flow_control_xon;
+	uint64_t tx_flow_control_xoff;
+	uint64_t rx_csum_offload_good;
+	uint64_t rx_csum_offload_errors;
+	uint64_t sg_ok;
+	uint64_t sg_linearized;
+};
+
+#define VERIFY_HEADER_NOK -1
+#define VERIFY_HEADER_OK 0
+#define VERIFY_CSUM_OK 1
+
+struct vector_private {
+	struct list_head list;
+	spinlock_t lock;
+	struct net_device *dev;
+
+	int unit;
+
+	/* Timeout timer in TX */
+
+	struct timer_list tl;
+
+	/* Scheduled "remove device" work */
+	struct work_struct reset_tx;
+	struct vector_fds *fds;
+
+	struct vector_queue *rx_queue;
+	struct vector_queue *tx_queue;
+
+	int rx_irq;
+	int tx_irq;
+
+	struct arglist *parsed;
+
+	void *transport_data; /* transport specific params if needed */
+
+	int max_packet;
+	int req_size; /* different from max packet - used for TSO */
+	int headroom;
+
+	int options;
+
+	/* remote address if any - some transports will leave this as null */
+
+	int header_size;
+	int rx_header_size;
+	int coalesce;
+
+	void *header_rxbuffer;
+	void *header_txbuffer;
+
+	int (*form_header)(uint8_t *header,
+		struct sk_buff *skb, struct vector_private *vp);
+	int (*verify_header)(uint8_t *header,
+		struct sk_buff *skb, struct vector_private *vp);
+
+	spinlock_t stats_lock;
+
+	struct tasklet_struct tx_poll;
+	bool rexmit_scheduled;
+	bool opened;
+	bool in_write_poll;
+
+	/* ethtool stats */
+
+	struct vector_estats estats;
+	void *bpf;
+
+	char user[0];
+};
+
+extern int build_transport_data(struct vector_private *vp);
+
+#endif

arch/um/drivers/vector_transports.c

+/*
+ * Copyright (C) 2017 - Cambridge Greys Limited
+ * Copyright (C) 2011 - 2014 Cisco Systems Inc
+ * Licensed under the GPL.
+ */
+
+#include <linux/etherdevice.h>
+#include <linux/netdevice.h>
+#include <linux/skbuff.h>
+#include <linux/slab.h>
+#include <asm/byteorder.h>
+#include <uapi/linux/ip.h>
+#include <uapi/linux/virtio_net.h>
+#include <linux/virtio_net.h>
+#include <linux/virtio_byteorder.h>
+#include <linux/netdev_features.h>
+#include "vector_user.h"
+#include "vector_kern.h"
+
+#define GOOD_LINEAR 512
+#define GSO_ERROR "Incoming GSO frames and GRO disabled on the interface"
+
+struct gre_minimal_header {
+	uint16_t header;
+	uint16_t arptype;
+};
+
+
+struct uml_gre_data {
+	uint32_t rx_key;
+	uint32_t tx_key;
+	uint32_t sequence;
+
+	bool ipv6;
+	bool has_sequence;
+	bool pin_sequence;
+	bool checksum;
+	bool key;
+	struct gre_minimal_header expected_header;
+
+	uint32_t checksum_offset;
+	uint32_t key_offset;
+	uint32_t sequence_offset;
+
+};
+
+struct uml_l2tpv3_data {
+	uint64_t rx_cookie;
+	uint64_t tx_cookie;
+	uint64_t rx_session;
+	uint64_t tx_session;
+	uint32_t counter;
+
+	bool udp;
+	bool ipv6;
+	bool has_counter;
+	bool pin_counter;
+	bool cookie;
+	bool cookie_is_64;
+
+	uint32_t cookie_offset;
+	uint32_t session_offset;
+	uint32_t counter_offset;
+};
+
+static int l2tpv3_form_header(uint8_t *header,
+	struct sk_buff *skb, struct vector_private *vp)
+{
+	struct uml_l2tpv3_data *td = vp->transport_data;
+	uint32_t *counter;
+
+	if (td->udp)
+		*(uint32_t *) header = cpu_to_be32(L2TPV3_DATA_PACKET);
+	(*(uint32_t *) (header + td->session_offset)) = td->tx_session;
+
+	if (td->cookie) {
+		if (td->cookie_is_64)
+			(*(uint64_t *)(header + td->cookie_offset)) =
+				td->tx_cookie;
+		else
+			(*(uint32_t *)(header + td->cookie_offset)) =
+				td->tx_cookie;
+	}
+	if (td->has_counter) {
+		counter = (uint32_t *)(header + td->counter_offset);
+		if (td->pin_counter) {
+			*counter = 0;
+		} else {
+			td->counter++;
+			*counter = cpu_to_be32(td->counter);
+		}
+	}
+	return 0;
+}
+
+static int gre_form_header(uint8_t *header,
+		struct sk_buff *skb, struct vector_private *vp)
+{
+	struct uml_gre_data *td = vp->transport_data;
+	uint32_t *sequence;
+	*((uint32_t *) header) = *((uint32_t *) &td->expected_header);
+	if (td->key)
+		(*(uint32_t *) (header + td->key_offset)) = td->tx_key;
+	if (td->has_sequence) {
+		sequence = (uint32_t *)(header + td->sequence_offset);
+		if (td->pin_sequence)
+			*sequence = 0;
+		else
+			*sequence = cpu_to_be32(++td->sequence);
+	}
+	return 0;
+}
+
+static int raw_form_header(uint8_t *header,
+		struct sk_buff *skb, struct vector_private *vp)
+{
+	struct virtio_net_hdr *vheader = (struct virtio_net_hdr *) header;
+
+	virtio_net_hdr_from_skb(
+		skb,
+		vheader,
+		virtio_legacy_is_little_endian(),
+		false
+	);
+
+	return 0;
+}
+
+static int l2tpv3_verify_header(
+	uint8_t *header, struct sk_buff *skb, struct vector_private *vp)
+{
+	struct uml_l2tpv3_data *td = vp->transport_data;
+	uint32_t *session;
+	uint64_t cookie;
+
+	if ((!td->udp) && (!td->ipv6))
+		header += sizeof(struct iphdr) /* fix for ipv4 raw */;
+
+	/* we do not do a strict check for "data" packets as per
+	 * the RFC spec because the pure IP spec does not have
+	 * that anyway.
+	 */
+
+	if (td->cookie) {
+		if (td->cookie_is_64)
+			cookie = *(uint64_t *)(header + td->cookie_offset);
+		else
+			cookie = *(uint32_t *)(header + td->cookie_offset);
+		if (cookie != td->rx_cookie) {
+			if (net_ratelimit())
+				netdev_err(vp->dev, "uml_l2tpv3: unknown cookie id");
+			return -1;
+		}
+	}
+	session = (uint32_t *) (header + td->session_offset);
+	if (*session != td->rx_session) {
+		if (net_ratelimit())
+			netdev_err(vp->dev, "uml_l2tpv3: session mismatch");
+		return -1;
+	}
+	return 0;
+}
+
+static int gre_verify_header(
+	uint8_t *header, struct sk_buff *skb, struct vector_private *vp)
+{
+
+	uint32_t key;
+	struct uml_gre_data *td = vp->transport_data;
+
+	if (!td->ipv6)
+		header += sizeof(struct iphdr) /* fix for ipv4 raw */;
+
+	if (*((uint32_t *) header) != *((uint32_t *) &td->expected_header)) {
+		if (net_ratelimit())
+			netdev_err(vp->dev, "header type disagreement, expecting %0x, got %0x",
+				*((uint32_t *) &td->expected_header),
+				*((uint32_t *) header)
+			);
+		return -1;
+	}
+
+	if (td->key) {
+		key = (*(uint32_t *)(header + td->key_offset));
+		if (key != td->rx_key) {
+			if (net_ratelimit())
+				netdev_err(vp->dev, "unknown key id %0x, expecting %0x",
+						key, td->rx_key);
+			return -1;
+		}
+	}
+	return 0;
+}
+
+static int raw_verify_header(
+	uint8_t *header, struct sk_buff *skb, struct vector_private *vp)
+{
+	struct virtio_net_hdr *vheader = (struct virtio_net_hdr *) header;
+
+	if ((vheader->gso_type != VIRTIO_NET_HDR_GSO_NONE) &&
+		(vp->req_size != 65536)) {
+		if (net_ratelimit())
+			netdev_err(
+				vp->dev,
+				GSO_ERROR
+		);
+	}
+	if ((vheader->flags & VIRTIO_NET_HDR_F_DATA_VALID) > 0)
+		return 1;
+
+	virtio_net_hdr_to_skb(skb, vheader, virtio_legacy_is_little_endian());
+	return 0;
+}
+
+static bool get_uint_param(
+	struct arglist *def, char *param, unsigned int *result)
+{
+	char *arg = uml_vector_fetch_arg(def, param);
+
+	if (arg != NULL) {
+		if (kstrtoint(arg, 0, result) == 0)
+			return true;
+	}
+	return false;
+}
+
+static bool get_ulong_param(
+	struct arglist *def, char *param, unsigned long *result)
+{
+	char *arg = uml_vector_fetch_arg(def, param);
+
+	if (arg != NULL) {
+		if (kstrtoul(arg, 0, result) == 0)
+			return true;
+		return true;
+	}
+	return false;
+}
+
+static int build_gre_transport_data(struct vector_private *vp)
+{
+	struct uml_gre_data *td;
+	int temp_int;
+	int temp_rx;
+	int temp_tx;
+
+	vp->transport_data = kmalloc(sizeof(struct uml_gre_data), GFP_KERNEL);
+	if (vp->transport_data == NULL)
+		return -ENOMEM;
+	td = vp->transport_data;
+	td->sequence = 0;
+
+	td->expected_header.arptype = GRE_IRB;
+	td->expected_header.header = 0;
+
+	vp->form_header = &gre_form_header;
+	vp->verify_header = &gre_verify_header;
+	vp->header_size = 4;
+	td->key_offset = 4;
+	td->sequence_offset = 4;
+	td->checksum_offset = 4;
+
+	td->ipv6 = false;
+	if (get_uint_param(vp->parsed, "v6", &temp_int)) {
+		if (temp_int > 0)
+			td->ipv6 = true;
+	}
+	td->key = false;
+	if (get_uint_param(vp->parsed, "rx_key", &temp_rx)) {
+		if (get_uint_param(vp->parsed, "tx_key", &temp_tx)) {
+			td->key = true;
+			td->expected_header.header |= GRE_MODE_KEY;
+			td->rx_key = cpu_to_be32(temp_rx);
+			td->tx_key = cpu_to_be32(temp_tx);
+			vp->header_size += 4;
+			td->sequence_offset += 4;
+		} else {
+			return -EINVAL;
+		}
+	}
+
+	td->sequence = false;
+	if (get_uint_param(vp->parsed, "sequence", &temp_int)) {
+		if (temp_int > 0) {
+			vp->header_size += 4;
+			td->has_sequence = true;
+			td->expected_header.header |= GRE_MODE_SEQUENCE;
+			if (get_uint_param(
+				vp->parsed, "pin_sequence", &temp_int)) {
+				if (temp_int > 0)
+					td->pin_sequence = true;
+			}
+		}
+	}
+	vp->rx_header_size = vp->header_size;
+	if (!td->ipv6)
+		vp->rx_header_size += sizeof(struct iphdr);
+	return 0;
+}
+
+static int build_l2tpv3_transport_data(struct vector_private *vp)
+{
+
+	struct uml_l2tpv3_data *td;
+	int temp_int, temp_rxs, temp_txs;
+	unsigned long temp_rx;
+	unsigned long temp_tx;
+
+	vp->transport_data = kmalloc(
+		sizeof(struct uml_l2tpv3_data), GFP_KERNEL);
+
+	if (vp->transport_data == NULL)
+		return -ENOMEM;
+
+	td = vp->transport_data;
+
+	vp->form_header = &l2tpv3_form_header;
+	vp->verify_header = &l2tpv3_verify_header;
+	td->counter = 0;
+
+	vp->header_size = 4;
+	td->session_offset = 0;
+	td->cookie_offset = 4;
+	td->counter_offset = 4;
+
+
+	td->ipv6 = false;
+	if (get_uint_param(vp->parsed, "v6", &temp_int)) {
+		if (temp_int > 0)
+			td->ipv6 = true;
+	}
+
+	if (get_uint_param(vp->parsed, "rx_session", &temp_rxs)) {
+		if (get_uint_param(vp->parsed, "tx_session", &temp_txs)) {
+			td->tx_session = cpu_to_be32(temp_txs);
+			td->rx_session = cpu_to_be32(temp_rxs);
+		} else {
+			return -EINVAL;
+		}
+	} else {
+		return -EINVAL;
+	}
+
+	td->cookie_is_64  = false;
+	if (get_uint_param(vp->parsed, "cookie64", &temp_int)) {
+		if (temp_int > 0)
+			td->cookie_is_64  = true;
+	}
+	td->cookie = false;
+	if (get_ulong_param(vp->parsed, "rx_cookie", &temp_rx)) {
+		if (get_ulong_param(vp->parsed, "tx_cookie", &temp_tx)) {
+			td->cookie = true;
+			if (td->cookie_is_64) {
+				td->rx_cookie = cpu_to_be64(temp_rx);
+				td->tx_cookie = cpu_to_be64(temp_tx);
+				vp->header_size += 8;
+				td->counter_offset += 8;
+			} else {
+				td->rx_cookie = cpu_to_be32(temp_rx);
+				td->tx_cookie = cpu_to_be32(temp_tx);
+				vp->header_size += 4;
+				td->counter_offset += 4;
+			}
+		} else {
+			return -EINVAL;
+		}
+	}
+
+	td->has_counter = false;
+	if (get_uint_param(vp->parsed, "counter", &temp_int)) {
+		if (temp_int > 0) {
+			td->has_counter = true;
+			vp->header_size += 4;
+			if (get_uint_param(
+				vp->parsed, "pin_counter", &temp_int)) {
+				if (temp_int > 0)
+					td->pin_counter = true;
+			}
+		}
+	}
+
+	if (get_uint_param(vp->parsed, "udp", &temp_int)) {
+		if (temp_int > 0) {
+			td->udp = true;
+			vp->header_size += 4;
+			td->counter_offset += 4;
+			td->session_offset += 4;
+			td->cookie_offset += 4;
+		}
+	}
+
+	vp->rx_header_size = vp->header_size;
+	if ((!td->ipv6) && (!td->udp))
+		vp->rx_header_size += sizeof(struct iphdr);
+
+	return 0;
+}
+
+static int build_raw_transport_data(struct vector_private *vp)
+{
+	if (uml_raw_enable_vnet_headers(vp->fds->rx_fd)) {
+		if (!uml_raw_enable_vnet_headers(vp->fds->tx_fd))
+			return -1;
+		vp->form_header = &raw_form_header;
+		vp->verify_header = &raw_verify_header;
+		vp->header_size = sizeof(struct virtio_net_hdr);
+		vp->rx_header_size = sizeof(struct virtio_net_hdr);
+		vp->dev->hw_features |= (NETIF_F_TSO | NETIF_F_GRO);
+		vp->dev->features |=
+			(NETIF_F_RXCSUM | NETIF_F_HW_CSUM |
+				NETIF_F_TSO | NETIF_F_GRO);
+		netdev_info(
+			vp->dev,
+			"raw: using vnet headers for tso and tx/rx checksum"
+		);
+	}
+	return 0;
+}
+
+static int build_tap_transport_data(struct vector_private *vp)
+{
+	if (uml_raw_enable_vnet_headers(vp->fds->rx_fd)) {
+		vp->form_header = &raw_form_header;
+		vp->verify_header = &raw_verify_header;
+		vp->header_size = sizeof(struct virtio_net_hdr);
+		vp->rx_header_size = sizeof(struct virtio_net_hdr);
+		vp->dev->hw_features |=
+			(NETIF_F_TSO | NETIF_F_GSO | NETIF_F_GRO);
+		vp->dev->features |=
+			(NETIF_F_RXCSUM | NETIF_F_HW_CSUM |
+				NETIF_F_TSO | NETIF_F_GSO | NETIF_F_GRO);
+		netdev_info(
+			vp->dev,
+			"tap/raw: using vnet headers for tso and tx/rx checksum"
+		);
+	} else {
+		return 0; /* do not try to enable tap too if raw failed */
+	}
+	if (uml_tap_enable_vnet_headers(vp->fds->tx_fd))
+		return 0;
+	return -1;
+}
+
+int build_transport_data(struct vector_private *vp)
+{
+	char *transport = uml_vector_fetch_arg(vp->parsed, "transport");
+
+	if (strncmp(transport, TRANS_GRE, TRANS_GRE_LEN) == 0)
+		return build_gre_transport_data(vp);
+	if (strncmp(transport, TRANS_L2TPV3, TRANS_L2TPV3_LEN) == 0)
+		return build_l2tpv3_transport_data(vp);
+	if (strncmp(transport, TRANS_RAW, TRANS_RAW_LEN) == 0)
+		return build_raw_transport_data(vp);
+	if (strncmp(transport, TRANS_TAP, TRANS_TAP_LEN) == 0)
+		return build_tap_transport_data(vp);
+	return 0;
+}
+

arch/um/drivers/vector_user.c

+/*
+ * Copyright (C) 2001 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
+ * Licensed under the GPL
+ */
+
+#include <stdio.h>
+#include <unistd.h>
+#include <stdarg.h>
+#include <errno.h>
+#include <stddef.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <net/if.h>
+#include <linux/if_tun.h>
+#include <arpa/inet.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <net/ethernet.h>
+#include <netinet/ip.h>
+#include <netinet/ether.h>
+#include <linux/if_ether.h>
+#include <linux/if_packet.h>
+#include <sys/socket.h>
+#include <sys/wait.h>
+#include <linux/virtio_net.h>
+#include <netdb.h>
+#include <stdlib.h>
+#include <os.h>
+#include <um_malloc.h>
+#include "vector_user.h"
+
+#define ID_GRE 0
+#define ID_L2TPV3 1
+#define ID_MAX 1
+
+#define TOKEN_IFNAME "ifname"
+
+#define TRANS_RAW "raw"
+#define TRANS_RAW_LEN strlen(TRANS_RAW)
+
+#define VNET_HDR_FAIL "could not enable vnet headers on fd %d"
+#define TUN_GET_F_FAIL "tapraw: TUNGETFEATURES failed: %s"
+#define L2TPV3_BIND_FAIL "l2tpv3_open : could not bind socket err=%i"
+#define BPF_ATTACH_FAIL "Failed to attach filter size %d to %d, err %d\n"
+
+/* This is very ugly and brute force lookup, but it is done
+ * only once at initialization so not worth doing hashes or
+ * anything more intelligent
+ */
+
+char *uml_vector_fetch_arg(struct arglist *ifspec, char *token)
+{
+	int i;
+
+	for (i = 0; i < ifspec->numargs; i++) {
+		if (strcmp(ifspec->tokens[i], token) == 0)
+			return ifspec->values[i];
+	}
+	return NULL;
+
+}
+
+struct arglist *uml_parse_vector_ifspec(char *arg)
+{
+	struct arglist *result;
+	int pos, len;
+	bool parsing_token = true, next_starts = true;
+
+	if (arg == NULL)
+		return NULL;
+	result = uml_kmalloc(sizeof(struct arglist), UM_GFP_KERNEL);
+	if (result == NULL)
+		return NULL;
+	result->numargs = 0;
+	len = strlen(arg);
+	for (pos = 0; pos < len; pos++) {
+		if (next_starts) {
+			if (parsing_token) {
+				result->tokens[result->numargs] = arg + pos;
+			} else {
+				result->values[result->numargs] = arg + pos;
+				result->numargs++;
+			}
+			next_starts = false;
+		}
+		if (*(arg + pos) == '=') {
+			if (parsing_token)
+				parsing_token = false;
+			else
+				goto cleanup;
+			next_starts = true;
+			(*(arg + pos)) = '\0';
+		}
+		if (*(arg + pos) == ',') {
+			parsing_token = true;
+			next_starts = true;
+			(*(arg + pos)) = '\0';
+		}
+	}
+	return result;
+cleanup:
+	printk(UM_KERN_ERR "vector_setup - Couldn't parse '%s'\n", arg);
+	kfree(result);
+	return NULL;
+}
+
+/*
+ * Socket/FD configuration functions. These return an structure
+ * of rx and tx descriptors to cover cases where these are not
+ * the same (f.e. read via raw socket and write via tap).
+ */
+
+#define PATH_NET_TUN "/dev/net/tun"
+
+static struct vector_fds *user_init_tap_fds(struct arglist *ifspec)
+{
+	struct ifreq ifr;
+	int fd = -1;
+	struct sockaddr_ll sock;
+	int err = -ENOMEM, offload;
+	char *iface;
+	struct vector_fds *result = NULL;
+
+	iface = uml_vector_fetch_arg(ifspec, TOKEN_IFNAME);
+	if (iface == NULL) {
+		printk(UM_KERN_ERR "uml_tap: failed to parse interface spec\n");
+		goto tap_cleanup;
+	}
+
+	result = uml_kmalloc(sizeof(struct vector_fds), UM_GFP_KERNEL);
+	if (result == NULL) {
+		printk(UM_KERN_ERR "uml_tap: failed to allocate file descriptors\n");
+		goto tap_cleanup;
+	}
+	result->rx_fd = -1;
+	result->tx_fd = -1;
+	result->remote_addr = NULL;
+	result->remote_addr_size = 0;
+
+	/* TAP */
+
+	fd = open(PATH_NET_TUN, O_RDWR);
+	if (fd < 0) {
+		printk(UM_KERN_ERR "uml_tap: failed to open tun device\n");
+		goto tap_cleanup;
+	}
+	result->tx_fd = fd;
+	memset(&ifr, 0, sizeof(ifr));
+	ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
+	strncpy((char *)&ifr.ifr_name, iface, sizeof(ifr.ifr_name) - 1);
+
+	err = ioctl(fd, TUNSETIFF, (void *) &ifr);
+	if (err != 0) {
+		printk(UM_KERN_ERR "uml_tap: failed to select tap interface\n");
+		goto tap_cleanup;
+	}
+
+	offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6;
+	ioctl(fd, TUNSETOFFLOAD, offload);
+
+	/* RAW */
+
+	fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
+	if (fd == -1) {
+		printk(UM_KERN_ERR
+			"uml_tap: failed to create socket: %i\n", -errno);
+		goto tap_cleanup;
+	}
+	result->rx_fd = fd;
+	memset(&ifr, 0, sizeof(ifr));
+	strncpy((char *)&ifr.ifr_name, iface, sizeof(ifr.ifr_name) - 1);
+	if (ioctl(fd, SIOCGIFINDEX, (void *) &ifr) < 0) {
+		printk(UM_KERN_ERR
+			"uml_tap: failed to set interface: %i\n", -errno);
+		goto tap_cleanup;
+	}
+
+	sock.sll_family = AF_PACKET;
+	sock.sll_protocol = htons(ETH_P_ALL);
+	sock.sll_ifindex = ifr.ifr_ifindex;
+
+	if (bind(fd,
+		(struct sockaddr *) &sock, sizeof(struct sockaddr_ll)) < 0) {
+		printk(UM_KERN_ERR
+			"user_init_tap: failed to bind raw pair, err %d\n",
+				-errno);
+		goto tap_cleanup;
+	}
+	return result;
+tap_cleanup:
+	printk(UM_KERN_ERR "user_init_tap: init failed, error %d", err);
+	if (result != NULL) {
+		if (result->rx_fd >= 0)
+			os_close_file(result->rx_fd);
+		if (result->tx_fd >= 0)
+			os_close_file(result->tx_fd);
+		kfree(result);
+	}
+	return NULL;
+}
+
+
+static struct vector_fds *user_init_raw_fds(struct arglist *ifspec)
+{
+	struct ifreq ifr;
+	int rxfd = -1, txfd = -1;
+	struct sockaddr_ll sock;
+	int err = -ENOMEM;
+	char *iface;
+	struct vector_fds *result = NULL;
+
+	iface = uml_vector_fetch_arg(ifspec, TOKEN_IFNAME);
+	if (iface == NULL)
+		goto cleanup;
+
+	rxfd = socket(AF_PACKET, SOCK_RAW, ETH_P_ALL);
+	if (rxfd == -1) {
+		err = -errno;
+		goto cleanup;
+	}
+	txfd = socket(AF_PACKET, SOCK_RAW, 0); /* Turn off RX on this fd */
+	if (txfd == -1) {
+		err = -errno;
+		goto cleanup;
+	}
+	memset(&ifr, 0, sizeof(ifr));
+	strncpy((char *)&ifr.ifr_name, iface, sizeof(ifr.ifr_name) - 1);
+	if (ioctl(rxfd, SIOCGIFINDEX, (void *) &ifr) < 0) {
+		err = -errno;
+		goto cleanup;
+	}
+
+	sock.sll_family = AF_PACKET;
+	sock.sll_protocol = htons(ETH_P_ALL);
+	sock.sll_ifindex = ifr.ifr_ifindex;
+
+	if (bind(rxfd,
+		(struct sockaddr *) &sock, sizeof(struct sockaddr_ll)) < 0) {
+		err = -errno;
+		goto cleanup;
+	}
+
+	sock.sll_family = AF_PACKET;
+	sock.sll_protocol = htons(ETH_P_IP);
+	sock.sll_ifindex = ifr.ifr_ifindex;
+
+	if (bind(txfd,
+		(struct sockaddr *) &sock, sizeof(struct sockaddr_ll)) < 0) {
+		err = -errno;
+		goto cleanup;
+	}
+
+	result = uml_kmalloc(sizeof(struct vector_fds), UM_GFP_KERNEL);
+	if (result != NULL) {
+		result->rx_fd = rxfd;
+		result->tx_fd = txfd;
+		result->remote_addr = NULL;
+		result->remote_addr_size = 0;
+	}
+	return result;
+cleanup:
+	printk(UM_KERN_ERR "user_init_raw: init failed, error %d", err);
+	if (rxfd >= 0)
+		os_close_file(rxfd);
+	if (txfd >= 0)
+		os_close_file(txfd);
+	if (result != NULL)
+		kfree(result);
+	return NULL;
+}
+
+
+bool uml_raw_enable_qdisc_bypass(int fd)
+{
+	int optval = 1;
+
+	if (setsockopt(fd,
+		SOL_PACKET, PACKET_QDISC_BYPASS,
+		&optval, sizeof(optval)) != 0) {
+		return false;
+	}
+	return true;
+}
+
+bool uml_raw_enable_vnet_headers(int fd)
+{
+	int optval = 1;
+
+	if (setsockopt(fd,
+		SOL_PACKET, PACKET_VNET_HDR,
+		&optval, sizeof(optval)) != 0) {
+		printk(UM_KERN_INFO VNET_HDR_FAIL, fd);
+		return false;
+	}
+	return true;
+}
+bool uml_tap_enable_vnet_headers(int fd)
+{
+	unsigned int features;
+	int len = sizeof(struct virtio_net_hdr);
+
+	if (ioctl(fd, TUNGETFEATURES, &features) == -1) {
+		printk(UM_KERN_INFO TUN_GET_F_FAIL, strerror(errno));
+		return false;
+	}
+	if ((features & IFF_VNET_HDR) == 0) {
+		printk(UM_KERN_INFO "tapraw: No VNET HEADER support");
+		return false;
+	}
+	ioctl(fd, TUNSETVNETHDRSZ, &len);
+	return true;
+}
+
+static struct vector_fds *user_init_socket_fds(struct arglist *ifspec, int id)
+{
+	int err = -ENOMEM;
+	int fd = -1, gairet;
+	struct addrinfo srchints;
+	struct addrinfo dsthints;
+	bool v6, udp;
+	char *value;
+	char *src, *dst, *srcport, *dstport;
+	struct addrinfo *gairesult = NULL;
+	struct vector_fds *result = NULL;
+
+
+	value = uml_vector_fetch_arg(ifspec, "v6");
+	v6 = false;
+	udp = false;
+	if (value != NULL) {
+		if (strtol((const char *) value, NULL, 10) > 0)
+			v6 = true;
+	}
+
+	value = uml_vector_fetch_arg(ifspec, "udp");
+	if (value != NULL) {
+		if (strtol((const char *) value, NULL, 10) > 0)
+			udp = true;
+	}
+	src = uml_vector_fetch_arg(ifspec, "src");
+	dst = uml_vector_fetch_arg(ifspec, "dst");
+	srcport = uml_vector_fetch_arg(ifspec, "srcport");
+	dstport = uml_vector_fetch_arg(ifspec, "dstport");
+
+	memset(&dsthints, 0, sizeof(dsthints));
+
+	if (v6)
+		dsthints.ai_family = AF_INET6;
+	else
+		dsthints.ai_family = AF_INET;
+
+	switch (id) {
+	case ID_GRE:
+		dsthints.ai_socktype = SOCK_RAW;
+		dsthints.ai_protocol = IPPROTO_GRE;
+		break;
+	case ID_L2TPV3:
+		if (udp) {
+			dsthints.ai_socktype = SOCK_DGRAM;
+			dsthints.ai_protocol = 0;
+		} else {
+			dsthints.ai_socktype = SOCK_RAW;
+			dsthints.ai_protocol = IPPROTO_L2TP;
+		}
+		break;
+	default:
+		printk(KERN_ERR "Unsupported socket type\n");
+		return NULL;
+	}
+	memcpy(&srchints, &dsthints, sizeof(struct addrinfo));
+
+	gairet = getaddrinfo(src, srcport, &dsthints, &gairesult);
+	if ((gairet != 0) || (gairesult == NULL)) {
+		printk(UM_KERN_ERR
+			"socket_open : could not resolve src, error = %s",
+			gai_strerror(gairet)
+		);
+		return NULL;
+	}
+	fd = socket(gairesult->ai_family,
+		gairesult->ai_socktype, gairesult->ai_protocol);
+	if (fd == -1) {
+		printk(UM_KERN_ERR
+			"socket_open : could not open socket, error = %d",
+			-errno
+		);
+		goto cleanup;
+	}
+	if (bind(fd,
+		(struct sockaddr *) gairesult->ai_addr,
+		gairesult->ai_addrlen)) {
+		printk(UM_KERN_ERR L2TPV3_BIND_FAIL, errno);
+		goto cleanup;
+	}
+
+	if (gairesult != NULL)
+		freeaddrinfo(gairesult);
+
+	gairesult = NULL;
+
+	gairet = getaddrinfo(dst, dstport, &dsthints, &gairesult);
+	if ((gairet != 0) || (gairesult == NULL)) {
+		printk(UM_KERN_ERR
+			"socket_open : could not resolve dst, error = %s",
+			gai_strerror(gairet)
+		);
+		return NULL;
+	}
+
+	result = uml_kmalloc(sizeof(struct vector_fds), UM_GFP_KERNEL);
+	if (result != NULL) {
+		result->rx_fd = fd;
+		result->tx_fd = fd;
+		result->remote_addr = uml_kmalloc(
+			gairesult->ai_addrlen, UM_GFP_KERNEL);
+		if (result->remote_addr == NULL)
+			goto cleanup;
+		result->remote_addr_size = gairesult->ai_addrlen;
+		memcpy(
+			result->remote_addr,
+			gairesult->ai_addr,
+			gairesult->ai_addrlen
+		);
+	}
+	freeaddrinfo(gairesult);
+	return result;
+cleanup:
+	if (gairesult != NULL)
+		freeaddrinfo(gairesult);
+	printk(UM_KERN_ERR "user_init_socket: init failed, error %d", err);
+	if (fd >= 0)
+		os_close_file(fd);
+	if (result != NULL) {
+		if (result->remote_addr != NULL)
+			kfree(result->remote_addr);
+		kfree(result);
+	}
+	return NULL;
+}
+
+struct vector_fds *uml_vector_user_open(
+	int unit,
+	struct arglist *parsed
+)
+{
+	char *transport;
+
+	if (parsed == NULL) {
+		printk(UM_KERN_ERR "no parsed config for unit %d\n", unit);
+		return NULL;
+	}
+	transport = uml_vector_fetch_arg(parsed, "transport");
+	if (transport == NULL) {
+		printk(UM_KERN_ERR "missing transport for unit %d\n", unit);
+		return NULL;
+	}
+	if (strncmp(transport, TRANS_RAW, TRANS_RAW_LEN) == 0)
+		return user_init_raw_fds(parsed);
+	if (strncmp(transport, TRANS_TAP, TRANS_TAP_LEN) == 0)
+		return user_init_tap_fds(parsed);
+	if (strncmp(transport, TRANS_GRE, TRANS_GRE_LEN) == 0)
+		return user_init_socket_fds(parsed, ID_GRE);
+	if (strncmp(transport, TRANS_L2TPV3, TRANS_L2TPV3_LEN) == 0)
+		return user_init_socket_fds(parsed, ID_L2TPV3);
+	return NULL;
+}
+
+
+int uml_vector_sendmsg(int fd, void *hdr, int flags)
+{
+	int n;
+
+	CATCH_EINTR(n = sendmsg(fd, (struct msghdr *) hdr,  flags));
+	if ((n < 0) && (errno == EAGAIN))
+		return 0;
+	if (n >= 0)
+		return n;
+	else
+		return -errno;
+}
+
+int uml_vector_recvmsg(int fd, void *hdr, int flags)
+{
+	int n;
+
+	CATCH_EINTR(n = recvmsg(fd, (struct msghdr *) hdr,  flags));
+	if ((n < 0) && (errno == EAGAIN))
+		return 0;
+	if (n >= 0)
+		return n;
+	else
+		return -errno;
+}
+
+int uml_vector_writev(int fd, void *hdr, int iovcount)
+{
+	int n;
+
+	CATCH_EINTR(n = writev(fd, (struct iovec *) hdr,  iovcount));
+	if ((n < 0) && (errno == EAGAIN))
+		return 0;
+	if (n >= 0)
+		return n;
+	else
+		return -errno;
+}
+
+int uml_vector_sendmmsg(
+	int fd,
+	void *msgvec,
+	unsigned int vlen,
+	unsigned int flags)
+{
+	int n;
+
+	CATCH_EINTR(n = sendmmsg(fd, (struct mmsghdr *) msgvec, vlen, flags));
+	if ((n < 0) && (errno == EAGAIN))
+		return 0;
+	if (n >= 0)
+		return n;
+	else
+		return -errno;
+}
+
+int uml_vector_recvmmsg(
+	int fd,
+	void *msgvec,
+	unsigned int vlen,
+	unsigned int flags)
+{
+	int n;
+
+	CATCH_EINTR(
+		n = recvmmsg(fd, (struct mmsghdr *) msgvec, vlen, flags, 0));
+	if ((n < 0) && (errno == EAGAIN))
+		return 0;
+	if (n >= 0)
+		return n;
+	else
+		return -errno;
+}
+int uml_vector_attach_bpf(int fd, void *bpf, int bpf_len)
+{
+	int err = setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, bpf, bpf_len);
+
+	if (err < 0)
+		printk(KERN_ERR BPF_ATTACH_FAIL, bpf_len, fd, -errno);
+	return err;
+}
+
+#define DEFAULT_BPF_LEN 6
+
+void *uml_vector_default_bpf(int fd, void *mac)
+{
+	struct sock_filter *bpf;
+	uint32_t *mac1 = (uint32_t *)(mac + 2);
+	uint16_t *mac2 = (uint16_t *) mac;
+	struct sock_fprog bpf_prog = {
+		.len = 6,
+		.filter = NULL,
+	};
+
+	bpf = uml_kmalloc(
+		sizeof(struct sock_filter) * DEFAULT_BPF_LEN, UM_GFP_KERNEL);
+	if (bpf != NULL) {
+		bpf_prog.filter = bpf;
+		/* ld	[8] */
+		bpf[0] = (struct sock_filter){ 0x20, 0, 0, 0x00000008 };
+		/* jeq	#0xMAC[2-6] jt 2 jf 5*/
+		bpf[1] = (struct sock_filter){ 0x15, 0, 3, ntohl(*mac1)};
+		/* ldh	[6] */
+		bpf[2] = (struct sock_filter){ 0x28, 0, 0, 0x00000006 };
+		/* jeq	#0xMAC[0-1] jt 4 jf 5 */
+		bpf[3] = (struct sock_filter){ 0x15, 0, 1, ntohs(*mac2)};
+		/* ret	#0 */
+		bpf[4] = (struct sock_filter){ 0x6, 0, 0, 0x00000000 };
+		/* ret	#0x40000 */
+		bpf[5] = (struct sock_filter){ 0x6, 0, 0, 0x00040000 };
+		if (uml_vector_attach_bpf(
+			fd, &bpf_prog, sizeof(struct sock_fprog)) < 0) {
+			kfree(bpf);
+			bpf = NULL;
+		}
+	}
+	return bpf;
+}
+

arch/um/drivers/vector_user.h

+/*
+ * Copyright (C) 2002 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
+ * Licensed under the GPL
+ */
+
+#ifndef __UM_VECTOR_USER_H
+#define __UM_VECTOR_USER_H
+
+#define MAXVARGS	20
+
+#define TOKEN_IFNAME "ifname"
+
+#define TRANS_RAW "raw"
+#define TRANS_RAW_LEN strlen(TRANS_RAW)
+
+#define TRANS_TAP "tap"
+#define TRANS_TAP_LEN strlen(TRANS_TAP)
+
+
+#define TRANS_GRE "gre"
+#define TRANS_GRE_LEN strlen(TRANS_RAW)
+
+#define TRANS_L2TPV3 "l2tpv3"
+#define TRANS_L2TPV3_LEN strlen(TRANS_L2TPV3)
+
+#ifndef IPPROTO_GRE
+#define IPPROTO_GRE 0x2F
+#endif
+
+#define GRE_MODE_CHECKSUM	cpu_to_be16(8 << 12)	/* checksum */
+#define GRE_MODE_RESERVED	cpu_to_be16(4 << 12)	/* unused */
+#define GRE_MODE_KEY		cpu_to_be16(2 << 12)	/* KEY present */
+#define GRE_MODE_SEQUENCE	cpu_to_be16(1 << 12)	/* sequence */
+
+#define GRE_IRB cpu_to_be16(0x6558)
+
+#define L2TPV3_DATA_PACKET 0x30000
+
+/* IANA-assigned IP protocol ID for L2TPv3 */
+
+#ifndef IPPROTO_L2TP
+#define IPPROTO_L2TP 0x73
+#endif
+
+struct arglist {
+	int	numargs;
+	char	*tokens[MAXVARGS];
+	char	*values[MAXVARGS];
+};
+
+/* Separating read and write FDs allows us to have different
+ * rx and tx method. Example - read tap via raw socket using
+ * recvmmsg, write using legacy tap write calls
+ */
+
+struct vector_fds {
+	int rx_fd;
+	int tx_fd;
+	void *remote_addr;
+	int remote_addr_size;
+};
+
+#define VECTOR_READ	1
+#define VECTOR_WRITE	(1 < 1)
+#define VECTOR_HEADERS	(1 < 2)
+
+extern struct arglist *uml_parse_vector_ifspec(char *arg);
+
+extern struct vector_fds *uml_vector_user_open(
+	int unit,
+	struct arglist *parsed
+);
+
+extern char *uml_vector_fetch_arg(
+	struct arglist *ifspec,
+	char *token
+);
+
+extern int uml_vector_recvmsg(int fd, void *hdr, int flags);
+extern int uml_vector_sendmsg(int fd, void *hdr, int flags);
+extern int uml_vector_writev(int fd, void *hdr, int iovcount);
+extern int uml_vector_sendmmsg(
+	int fd, void *msgvec,
+	unsigned int vlen,
+	unsigned int flags
+);
+extern int uml_vector_recvmmsg(
+	int fd,
+	void *msgvec,
+	unsigned int vlen,
+	unsigned int flags
+);
+extern void *uml_vector_default_bpf(int fd, void *mac);
+extern int uml_vector_attach_bpf(int fd, void *bpf, int bpf_len);
+extern bool uml_raw_enable_qdisc_bypass(int fd);
+extern bool uml_raw_enable_vnet_headers(int fd);
+extern bool uml_tap_enable_vnet_headers(int fd);
+
+
+#endif

arch/um/include/asm/asm-prototypes.h

+#include <asm-generic/asm-prototypes.h>

arch/um/include/asm/irq.h

@@ -18,7 +18,19 @@
 #define XTERM_IRQ 		13
 #define RANDOM_IRQ 		14
 
+#ifdef CONFIG_UML_NET_VECTOR
+
+#define VECTOR_BASE_IRQ		15
+#define VECTOR_IRQ_SPACE	8
+
+#define LAST_IRQ (VECTOR_IRQ_SPACE + VECTOR_BASE_IRQ)
+
+#else
+
 #define LAST_IRQ RANDOM_IRQ
+
+#endif
+
 #define NR_IRQS (LAST_IRQ + 1)
 
 #endif

arch/um/include/shared/irq_user.h

@@ -7,6 +7,7 @@
 #define __IRQ_USER_H__
 
 #include <sysdep/ptrace.h>
+#include <stdbool.h>
 
 struct irq_fd {
 	struct irq_fd *next;
@@ -15,10 +16,17 @@ struct irq_fd {
 	int type;
 	int irq;
 	int events;
-	int current_events;
+	bool active;
+	bool pending;
+	bool purge;
 };
 
-enum { IRQ_READ, IRQ_WRITE };
+#define IRQ_READ  0
+#define IRQ_WRITE 1
+#define IRQ_NONE 2
+#define MAX_IRQ_TYPE (IRQ_NONE + 1)
+
+
 
 struct siginfo;
 extern void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs);

arch/um/include/shared/net_kern.h

@@ -65,5 +65,7 @@ extern int tap_setup_common(char *str, char *type, char **dev_name,
 			    char **mac_out, char **gate_addr);
 extern void register_transport(struct transport *new);
 extern unsigned short eth_protocol(struct sk_buff *skb);
+extern void uml_net_setup_etheraddr(struct net_device *dev, char *str);
+
 
 #endif

arch/um/include/shared/os.h

@@ -290,15 +290,16 @@ extern void halt_skas(void);
 extern void reboot_skas(void);
 
 /* irq.c */
-extern int os_waiting_for_events(struct irq_fd *active_fds);
-extern int os_create_pollfd(int fd, int events, void *tmp_pfd, int size_tmpfds);
-extern void os_free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg,
-		struct irq_fd *active_fds, struct irq_fd ***last_irq_ptr2);
-extern void os_free_irq_later(struct irq_fd *active_fds,
-		int irq, void *dev_id);
-extern int os_get_pollfd(int i);
-extern void os_set_pollfd(int i, int fd);
+extern int os_waiting_for_events_epoll(void);
+extern void *os_epoll_get_data_pointer(int index);
+extern int os_epoll_triggered(int index, int events);
+extern int os_event_mask(int irq_type);
+extern int os_setup_epoll(void);
+extern int os_add_epoll_fd(int events, int fd, void *data);
+extern int os_mod_epoll_fd(int events, int fd, void *data);
+extern int os_del_epoll_fd(int fd);
 extern void os_set_ioignore(void);
+extern void os_close_epoll_fd(void);
 
 /* sigio.c */
 extern int add_sigio_fd(int fd);

arch/um/kernel/irq.c

 /*
+ * Copyright (C) 2017 - Cambridge Greys Ltd
+ * Copyright (C) 2011 - 2014 Cisco Systems Inc
  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
  * Derived (i.e. mostly copied) from arch/i386/kernel/irq.c:
@@ -16,243 +18,362 @@
 #include <as-layout.h>
 #include <kern_util.h>
 #include <os.h>
+#include <irq_user.h>
 
-/*
- * This list is accessed under irq_lock, except in sigio_handler,
- * where it is safe from being modified.  IRQ handlers won't change it -
- * if an IRQ source has vanished, it will be freed by free_irqs just
- * before returning from sigio_handler.  That will process a separate
- * list of irqs to free, with its own locking, coming back here to
- * remove list elements, taking the irq_lock to do so.
+
+/* When epoll triggers we do not know why it did so
+ * we can also have different IRQs for read and write.
+ * This is why we keep a small irq_fd array for each fd -
+ * one entry per IRQ type
  */
-static struct irq_fd *active_fds = NULL;
-static struct irq_fd **last_irq_ptr = &active_fds;
 
-extern void free_irqs(void);
+struct irq_entry {
+	struct irq_entry *next;
+	int fd;
+	struct irq_fd *irq_array[MAX_IRQ_TYPE + 1];
+};
+
+static struct irq_entry *active_fds;
+
+static DEFINE_SPINLOCK(irq_lock);
+
+static void irq_io_loop(struct irq_fd *irq, struct uml_pt_regs *regs)
+{
+/*
+ * irq->active guards against reentry
+ * irq->pending accumulates pending requests
+ * if pending is raised the irq_handler is re-run
+ * until pending is cleared
+ */
+	if (irq->active) {
+		irq->active = false;
+		do {
+			irq->pending = false;
+			do_IRQ(irq->irq, regs);
+		} while (irq->pending && (!irq->purge));
+		if (!irq->purge)
+			irq->active = true;
+	} else {
+		irq->pending = true;
+	}
+}
 
 void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 {
-	struct irq_fd *irq_fd;
-	int n;
+	struct irq_entry *irq_entry;
+	struct irq_fd *irq;
+
+	int n, i, j;
 
 	while (1) {
-		n = os_waiting_for_events(active_fds);
+		/* This is now lockless - epoll keeps back-referencesto the irqs
+		 * which have trigger it so there is no need to walk the irq
+		 * list and lock it every time. We avoid locking by turning off
+		 * IO for a specific fd by executing os_del_epoll_fd(fd) before
+		 * we do any changes to the actual data structures
+		 */
+		n = os_waiting_for_events_epoll();
+
 		if (n <= 0) {
 			if (n == -EINTR)
 				continue;
-			else break;
+			else
+				break;
 		}
 
-		for (irq_fd = active_fds; irq_fd != NULL;
-		     irq_fd = irq_fd->next) {
-			if (irq_fd->current_events != 0) {
-				irq_fd->current_events = 0;
-				do_IRQ(irq_fd->irq, regs);
+		for (i = 0; i < n ; i++) {
+			/* Epoll back reference is the entry with 3 irq_fd
+			 * leaves - one for each irq type.
+			 */
+			irq_entry = (struct irq_entry *)
+				os_epoll_get_data_pointer(i);
+			for (j = 0; j < MAX_IRQ_TYPE ; j++) {
+				irq = irq_entry->irq_array[j];
+				if (irq == NULL)
+					continue;
+				if (os_epoll_triggered(i, irq->events) > 0)
+					irq_io_loop(irq, regs);
+				if (irq->purge) {
+					irq_entry->irq_array[j] = NULL;
+					kfree(irq);
+				}
 			}
 		}
 	}
+}
+
+static int assign_epoll_events_to_irq(struct irq_entry *irq_entry)
+{
+	int i;
+	int events = 0;
+	struct irq_fd *irq;
 
-	free_irqs();
+	for (i = 0; i < MAX_IRQ_TYPE ; i++) {
+		irq = irq_entry->irq_array[i];
+		if (irq != NULL)
+			events = irq->events | events;
+	}
+	if (events > 0) {
+	/* os_add_epoll will call os_mod_epoll if this already exists */
+		return os_add_epoll_fd(events, irq_entry->fd, irq_entry);
+	}
+	/* No events - delete */
+	return os_del_epoll_fd(irq_entry->fd);
 }
 
-static DEFINE_SPINLOCK(irq_lock);
+
 
 static int activate_fd(int irq, int fd, int type, void *dev_id)
 {
-	struct pollfd *tmp_pfd;
-	struct irq_fd *new_fd, *irq_fd;
+	struct irq_fd *new_fd;
+	struct irq_entry *irq_entry;
+	int i, err, events;
 	unsigned long flags;
-	int events, err, n;
 
 	err = os_set_fd_async(fd);
 	if (err < 0)
 		goto out;
 
-	err = -ENOMEM;
-	new_fd = kmalloc(sizeof(struct irq_fd), GFP_KERNEL);
-	if (new_fd == NULL)
-		goto out;
+	spin_lock_irqsave(&irq_lock, flags);
 
-	if (type == IRQ_READ)
-		events = UM_POLLIN | UM_POLLPRI;
-	else events = UM_POLLOUT;
-	*new_fd = ((struct irq_fd) { .next  		= NULL,
-				     .id 		= dev_id,
-				     .fd 		= fd,
-				     .type 		= type,
-				     .irq 		= irq,
-				     .events 		= events,
-				     .current_events 	= 0 } );
+	/* Check if we have an entry for this fd */
 
 	err = -EBUSY;
-	spin_lock_irqsave(&irq_lock, flags);
-	for (irq_fd = active_fds; irq_fd != NULL; irq_fd = irq_fd->next) {
-		if ((irq_fd->fd == fd) && (irq_fd->type == type)) {
-			printk(KERN_ERR "Registering fd %d twice\n", fd);
-			printk(KERN_ERR "Irqs : %d, %d\n", irq_fd->irq, irq);
-			printk(KERN_ERR "Ids : 0x%p, 0x%p\n", irq_fd->id,
-			       dev_id);
+	for (irq_entry = active_fds;
+		irq_entry != NULL; irq_entry = irq_entry->next) {
+		if (irq_entry->fd == fd)
+			break;
+	}
+
+	if (irq_entry == NULL) {
+		/* This needs to be atomic as it may be called from an
+		 * IRQ context.
+		 */
+		irq_entry = kmalloc(sizeof(struct irq_entry), GFP_ATOMIC);
+		if (irq_entry == NULL) {
+			printk(KERN_ERR
+				"Failed to allocate new IRQ entry\n");
 			goto out_unlock;
 		}
+		irq_entry->fd = fd;
+		for (i = 0; i < MAX_IRQ_TYPE; i++)
+			irq_entry->irq_array[i] = NULL;
+		irq_entry->next = active_fds;
+		active_fds = irq_entry;
 	}
 
-	if (type == IRQ_WRITE)
-		fd = -1;
-
-	tmp_pfd = NULL;
-	n = 0;
+	/* Check if we are trying to re-register an interrupt for a
+	 * particular fd
+	 */
 
-	while (1) {
-		n = os_create_pollfd(fd, events, tmp_pfd, n);
-		if (n == 0)
-			break;
+	if (irq_entry->irq_array[type] != NULL) {
+		printk(KERN_ERR
+			"Trying to reregister IRQ %d FD %d TYPE %d ID %p\n",
+			irq, fd, type, dev_id
+		);
+		goto out_unlock;
+	} else {
+		/* New entry for this fd */
+
+		err = -ENOMEM;
+		new_fd = kmalloc(sizeof(struct irq_fd), GFP_ATOMIC);
+		if (new_fd == NULL)
+			goto out_unlock;
 
-		/*
-		 * n > 0
-		 * It means we couldn't put new pollfd to current pollfds
-		 * and tmp_fds is NULL or too small for new pollfds array.
-		 * Needed size is equal to n as minimum.
-		 *
-		 * Here we have to drop the lock in order to call
-		 * kmalloc, which might sleep.
-		 * If something else came in and changed the pollfds array
-		 * so we will not be able to put new pollfd struct to pollfds
-		 * then we free the buffer tmp_fds and try again.
+		events = os_event_mask(type);
+
+		*new_fd = ((struct irq_fd) {
+			.id		= dev_id,
+			.irq		= irq,
+			.type		= type,
+			.events		= events,
+			.active		= true,
+			.pending	= false,
+			.purge		= false
+		});
+		/* Turn off any IO on this fd - allows us to
+		 * avoid locking the IRQ loop
 		 */
-		spin_unlock_irqrestore(&irq_lock, flags);
-		kfree(tmp_pfd);
-
-		tmp_pfd = kmalloc(n, GFP_KERNEL);
-		if (tmp_pfd == NULL)
-			goto out_kfree;
-
-		spin_lock_irqsave(&irq_lock, flags);
+		os_del_epoll_fd(irq_entry->fd);
+		irq_entry->irq_array[type] = new_fd;
 	}
 
-	*last_irq_ptr = new_fd;
-	last_irq_ptr = &new_fd->next;
-
+	/* Turn back IO on with the correct (new) IO event mask */
+	assign_epoll_events_to_irq(irq_entry);
 	spin_unlock_irqrestore(&irq_lock, flags);
-
-	/*
-	 * This calls activate_fd, so it has to be outside the critical
-	 * section.
-	 */
-	maybe_sigio_broken(fd, (type == IRQ_READ));
+	maybe_sigio_broken(fd, (type != IRQ_NONE));
 
 	return 0;
-
- out_unlock:
+out_unlock:
 	spin_unlock_irqrestore(&irq_lock, flags);
- out_kfree:
-	kfree(new_fd);
- out:
+out:
 	return err;
 }
 
-static void free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg)
+/*
+ * Walk the IRQ list and dispose of any unused entries.
+ * Should be done under irq_lock.
+ */
+
+static void garbage_collect_irq_entries(void)
 {
-	unsigned long flags;
+	int i;
+	bool reap;
+	struct irq_entry *walk;
+	struct irq_entry *previous = NULL;
+	struct irq_entry *to_free;
 
-	spin_lock_irqsave(&irq_lock, flags);
-	os_free_irq_by_cb(test, arg, active_fds, &last_irq_ptr);
-	spin_unlock_irqrestore(&irq_lock, flags);
+	if (active_fds == NULL)
+		return;
+	walk = active_fds;
+	while (walk != NULL) {
+		reap = true;
+		for (i = 0; i < MAX_IRQ_TYPE ; i++) {
+			if (walk->irq_array[i] != NULL) {
+				reap = false;
+				break;
+			}
+		}
+		if (reap) {
+			if (previous == NULL)
+				active_fds = walk->next;
+			else
+				previous->next = walk->next;
+			to_free = walk;
+		} else {
+			to_free = NULL;
+		}
+		walk = walk->next;
+		if (to_free != NULL)
+			kfree(to_free);
+	}
 }
 
-struct irq_and_dev {
-	int irq;
-	void *dev;
-};
+/*
+ * Walk the IRQ list and get the descriptor for our FD
+ */
 
-static int same_irq_and_dev(struct irq_fd *irq, void *d)
+static struct irq_entry *get_irq_entry_by_fd(int fd)
 {
-	struct irq_and_dev *data = d;
+	struct irq_entry *walk = active_fds;
 
-	return ((irq->irq == data->irq) && (irq->id == data->dev));
+	while (walk != NULL) {
+		if (walk->fd == fd)
+			return walk;
+		walk = walk->next;
+	}
+	return NULL;
 }
 
-static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
-{
-	struct irq_and_dev data = ((struct irq_and_dev) { .irq  = irq,
-							  .dev  = dev });
 
-	free_irq_by_cb(same_irq_and_dev, &data);
-}
+/*
+ * Walk the IRQ list and dispose of an entry for a specific
+ * device, fd and number. Note - if sharing an IRQ for read
+ * and writefor the same FD it will be disposed in either case.
+ * If this behaviour is undesirable use different IRQ ids.
+ */
 
-static int same_fd(struct irq_fd *irq, void *fd)
-{
-	return (irq->fd == *((int *)fd));
-}
+#define IGNORE_IRQ 1
+#define IGNORE_DEV (1<<1)
 
-void free_irq_by_fd(int fd)
+static void do_free_by_irq_and_dev(
+	struct irq_entry *irq_entry,
+	unsigned int irq,
+	void *dev,
+	int flags
+)
 {
-	free_irq_by_cb(same_fd, &fd);
+	int i;
+	struct irq_fd *to_free;
+
+	for (i = 0; i < MAX_IRQ_TYPE ; i++) {
+		if (irq_entry->irq_array[i] != NULL) {
+			if (
+			((flags & IGNORE_IRQ) ||
+				(irq_entry->irq_array[i]->irq == irq)) &&
+			((flags & IGNORE_DEV) ||
+				(irq_entry->irq_array[i]->id == dev))
+			) {
+				/* Turn off any IO on this fd - allows us to
+				 * avoid locking the IRQ loop
+				 */
+				os_del_epoll_fd(irq_entry->fd);
+				to_free = irq_entry->irq_array[i];
+				irq_entry->irq_array[i] = NULL;
+				assign_epoll_events_to_irq(irq_entry);
+				if (to_free->active)
+					to_free->purge = true;
+				else
+					kfree(to_free);
+			}
+		}
+	}
 }
 
-/* Must be called with irq_lock held */
-static struct irq_fd *find_irq_by_fd(int fd, int irqnum, int *index_out)
+void free_irq_by_fd(int fd)
 {
-	struct irq_fd *irq;
-	int i = 0;
-	int fdi;
+	struct irq_entry *to_free;
+	unsigned long flags;
 
-	for (irq = active_fds; irq != NULL; irq = irq->next) {
-		if ((irq->fd == fd) && (irq->irq == irqnum))
-			break;
-		i++;
-	}
-	if (irq == NULL) {
-		printk(KERN_ERR "find_irq_by_fd doesn't have descriptor %d\n",
-		       fd);
-		goto out;
-	}
-	fdi = os_get_pollfd(i);
-	if ((fdi != -1) && (fdi != fd)) {
-		printk(KERN_ERR "find_irq_by_fd - mismatch between active_fds "
-		       "and pollfds, fd %d vs %d, need %d\n", irq->fd,
-		       fdi, fd);
-		irq = NULL;
-		goto out;
+	spin_lock_irqsave(&irq_lock, flags);
+	to_free = get_irq_entry_by_fd(fd);
+	if (to_free != NULL) {
+		do_free_by_irq_and_dev(
+			to_free,
+			-1,
+			NULL,
+			IGNORE_IRQ | IGNORE_DEV
+		);
 	}
-	*index_out = i;
- out:
-	return irq;
+	garbage_collect_irq_entries();
+	spin_unlock_irqrestore(&irq_lock, flags);
 }
+EXPORT_SYMBOL(free_irq_by_fd);
 
-void reactivate_fd(int fd, int irqnum)
+static void free_irq_by_irq_and_dev(unsigned int irq, void *dev)
 {
-	struct irq_fd *irq;
+	struct irq_entry *to_free;
 	unsigned long flags;
-	int i;
 
 	spin_lock_irqsave(&irq_lock, flags);
-	irq = find_irq_by_fd(fd, irqnum, &i);
-	if (irq == NULL) {
-		spin_unlock_irqrestore(&irq_lock, flags);
-		return;
+	to_free = active_fds;
+	while (to_free != NULL) {
+		do_free_by_irq_and_dev(
+			to_free,
+			irq,
+			dev,
+			0
+		);
+		to_free = to_free->next;
 	}
-	os_set_pollfd(i, irq->fd);
+	garbage_collect_irq_entries();
 	spin_unlock_irqrestore(&irq_lock, flags);
+}
 
-	add_sigio_fd(fd);
+
+void reactivate_fd(int fd, int irqnum)
+{
+	/** NOP - we do auto-EOI now **/
 }
 
 void deactivate_fd(int fd, int irqnum)
 {
-	struct irq_fd *irq;
+	struct irq_entry *to_free;
 	unsigned long flags;
-	int i;
 
+	os_del_epoll_fd(fd);
 	spin_lock_irqsave(&irq_lock, flags);
-	irq = find_irq_by_fd(fd, irqnum, &i);
-	if (irq == NULL) {
-		spin_unlock_irqrestore(&irq_lock, flags);
-		return;
+	to_free = get_irq_entry_by_fd(fd);
+	if (to_free != NULL) {
+		do_free_by_irq_and_dev(
+			to_free,
+			irqnum,
+			NULL,
+			IGNORE_DEV
+		);
 	}
-
-	os_set_pollfd(i, -1);
+	garbage_collect_irq_entries();
 	spin_unlock_irqrestore(&irq_lock, flags);
-
 	ignore_sigio_fd(fd);
 }
 EXPORT_SYMBOL(deactivate_fd);
@@ -265,17 +386,28 @@ EXPORT_SYMBOL(deactivate_fd);
  */
 int deactivate_all_fds(void)
 {
-	struct irq_fd *irq;
-	int err;
+	unsigned long flags;
+	struct irq_entry *to_free;
 
-	for (irq = active_fds; irq != NULL; irq = irq->next) {
-		err = os_clear_fd_async(irq->fd);
-		if (err)
-			return err;
-	}
-	/* If there is a signal already queued, after unblocking ignore it */
+	spin_lock_irqsave(&irq_lock, flags);
+	/* Stop IO. The IRQ loop has no lock so this is our
+	 * only way of making sure we are safe to dispose
+	 * of all IRQ handlers
+	 */
 	os_set_ioignore();
-
+	to_free = active_fds;
+	while (to_free != NULL) {
+		do_free_by_irq_and_dev(
+			to_free,
+			-1,
+			NULL,
+			IGNORE_IRQ | IGNORE_DEV
+		);
+		to_free = to_free->next;
+	}
+	garbage_collect_irq_entries();
+	spin_unlock_irqrestore(&irq_lock, flags);
+	os_close_epoll_fd();
 	return 0;
 }
 
@@ -353,8 +485,11 @@ void __init init_IRQ(void)
 
 	irq_set_chip_and_handler(TIMER_IRQ, &SIGVTALRM_irq_type, handle_edge_irq);
 
+
 	for (i = 1; i < NR_IRQS; i++)
 		irq_set_chip_and_handler(i, &normal_irq_type, handle_edge_irq);
+	/* Initialize EPOLL Loop */
+	os_setup_epoll();
 }
 
 /*

arch/um/kernel/time.c

@@ -121,12 +121,12 @@ static void __init um_timer_setup(void)
 	clockevents_register_device(&timer_clockevent);
 }
 
-void read_persistent_clock(struct timespec *ts)
+void read_persistent_clock64(struct timespec64 *ts)
 {
 	long long nsecs = os_persistent_clock_emulation();
 
-	set_normalized_timespec(ts, nsecs / NSEC_PER_SEC,
-				nsecs % NSEC_PER_SEC);
+	set_normalized_timespec64(ts, nsecs / NSEC_PER_SEC,
+				  nsecs % NSEC_PER_SEC);
 }
 
 void __init time_init(void)

arch/um/os-Linux/file.c

@@ -12,6 +12,7 @@
 #include <sys/mount.h>
 #include <sys/socket.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include <sys/un.h>
 #include <sys/types.h>
 #include <os.h>

arch/um/os-Linux/irq.c

 /*
+ * Copyright (C) 2017 - Cambridge Greys Ltd
+ * Copyright (C) 2011 - 2014 Cisco Systems Inc
  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
  */
 
 #include <stdlib.h>
 #include <errno.h>
-#include <poll.h>
+#include <sys/epoll.h>
 #include <signal.h>
 #include <string.h>
 #include <irq_user.h>
 #include <os.h>
 #include <um_malloc.h>
 
+/* Epoll support */
+
+static int epollfd = -1;
+
+#define MAX_EPOLL_EVENTS 64
+
+static struct epoll_event epoll_events[MAX_EPOLL_EVENTS];
+
+/* Helper to return an Epoll data pointer from an epoll event structure.
+ * We need to keep this one on the userspace side to keep includes separate
+ */
+
+void *os_epoll_get_data_pointer(int index)
+{
+	return epoll_events[index].data.ptr;
+}
+
+/* Helper to compare events versus the events in the epoll structure.
+ * Same as above - needs to be on the userspace side
+ */
+
+
+int os_epoll_triggered(int index, int events)
+{
+	return epoll_events[index].events & events;
+}
+/* Helper to set the event mask.
+ * The event mask is opaque to the kernel side, because it does not have
+ * access to the right includes/defines for EPOLL constants.
+ */
+
+int os_event_mask(int irq_type)
+{
+	if (irq_type == IRQ_READ)
+		return EPOLLIN | EPOLLPRI;
+	if (irq_type == IRQ_WRITE)
+		return EPOLLOUT;
+	return 0;
+}
+
 /*
- * Locked by irq_lock in arch/um/kernel/irq.c.  Changed by os_create_pollfd
- * and os_free_irq_by_cb, which are called under irq_lock.
+ * Initial Epoll Setup
  */
-static struct pollfd *pollfds = NULL;
-static int pollfds_num = 0;
-static int pollfds_size = 0;
+int os_setup_epoll(void)
+{
+	epollfd = epoll_create(MAX_EPOLL_EVENTS);
+	return epollfd;
+}
 
-int os_waiting_for_events(struct irq_fd *active_fds)
+/*
+ * Helper to run the actual epoll_wait
+ */
+int os_waiting_for_events_epoll(void)
 {
-	struct irq_fd *irq_fd;
-	int i, n, err;
+	int n, err;
 
-	n = poll(pollfds, pollfds_num, 0);
+	n = epoll_wait(epollfd,
+		(struct epoll_event *) &epoll_events, MAX_EPOLL_EVENTS, 0);
 	if (n < 0) {
 		err = -errno;
 		if (errno != EINTR)
-			printk(UM_KERN_ERR "os_waiting_for_events:"
-			       " poll returned %d, errno = %d\n", n, errno);
+			printk(
+				UM_KERN_ERR "os_waiting_for_events:"
+				" epoll returned %d, error = %s\n", n,
+				strerror(errno)
+			);
 		return err;
 	}
-
-	if (n == 0)
-		return 0;
-
-	irq_fd = active_fds;
-
-	for (i = 0; i < pollfds_num; i++) {
-		if (pollfds[i].revents != 0) {
-			irq_fd->current_events = pollfds[i].revents;
-			pollfds[i].fd = -1;
-		}
-		irq_fd = irq_fd->next;
-	}
 	return n;
 }
 
-int os_create_pollfd(int fd, int events, void *tmp_pfd, int size_tmpfds)
-{
-	if (pollfds_num == pollfds_size) {
-		if (size_tmpfds <= pollfds_size * sizeof(pollfds[0])) {
-			/* return min size needed for new pollfds area */
-			return (pollfds_size + 1) * sizeof(pollfds[0]);
-		}
-
-		if (pollfds != NULL) {
-			memcpy(tmp_pfd, pollfds,
-			       sizeof(pollfds[0]) * pollfds_size);
-			/* remove old pollfds */
-			kfree(pollfds);
-		}
-		pollfds = tmp_pfd;
-		pollfds_size++;
-	} else
-		kfree(tmp_pfd);	/* remove not used tmp_pfd */
-
-	pollfds[pollfds_num] = ((struct pollfd) { .fd		= fd,
-						  .events	= events,
-						  .revents	= 0 });
-	pollfds_num++;
-
-	return 0;
-}
 
-void os_free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg,
-		struct irq_fd *active_fds, struct irq_fd ***last_irq_ptr2)
+/*
+ * Helper to add a fd to epoll
+ */
+int os_add_epoll_fd(int events, int fd, void *data)
 {
-	struct irq_fd **prev;
-	int i = 0;
-
-	prev = &active_fds;
-	while (*prev != NULL) {
-		if ((*test)(*prev, arg)) {
-			struct irq_fd *old_fd = *prev;
-			if ((pollfds[i].fd != -1) &&
-			    (pollfds[i].fd != (*prev)->fd)) {
-				printk(UM_KERN_ERR "os_free_irq_by_cb - "
-				       "mismatch between active_fds and "
-				       "pollfds, fd %d vs %d\n",
-				       (*prev)->fd, pollfds[i].fd);
-				goto out;
-			}
-
-			pollfds_num--;
-
-			/*
-			 * This moves the *whole* array after pollfds[i]
-			 * (though it doesn't spot as such)!
-			 */
-			memmove(&pollfds[i], &pollfds[i + 1],
-			       (pollfds_num - i) * sizeof(pollfds[0]));
-			if (*last_irq_ptr2 == &old_fd->next)
-				*last_irq_ptr2 = prev;
-
-			*prev = (*prev)->next;
-			if (old_fd->type == IRQ_WRITE)
-				ignore_sigio_fd(old_fd->fd);
-			kfree(old_fd);
-			continue;
-		}
-		prev = &(*prev)->next;
-		i++;
-	}
- out:
-	return;
+	struct epoll_event event;
+	int result;
+
+	event.data.ptr = data;
+	event.events = events | EPOLLET;
+	result = epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &event);
+	if ((result) && (errno == EEXIST))
+		result = os_mod_epoll_fd(events, fd, data);
+	if (result)
+		printk("epollctl add err fd %d, %s\n", fd, strerror(errno));
+	return result;
 }
 
-int os_get_pollfd(int i)
+/*
+ * Helper to mod the fd event mask and/or data backreference
+ */
+int os_mod_epoll_fd(int events, int fd, void *data)
 {
-	return pollfds[i].fd;
+	struct epoll_event event;
+	int result;
+
+	event.data.ptr = data;
+	event.events = events;
+	result = epoll_ctl(epollfd, EPOLL_CTL_MOD, fd, &event);
+	if (result)
+		printk(UM_KERN_ERR
+			"epollctl mod err fd %d, %s\n", fd, strerror(errno));
+	return result;
 }
 
-void os_set_pollfd(int i, int fd)
+/*
+ * Helper to delete the epoll fd
+ */
+int os_del_epoll_fd(int fd)
 {
-	pollfds[i].fd = fd;
+	struct epoll_event event;
+	int result;
+	/* This is quiet as we use this as IO ON/OFF - so it is often
+	 * invoked on a non-existent fd
+	 */
+	result = epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, &event);
+	return result;
 }
 
 void os_set_ioignore(void)
 {
 	signal(SIGIO, SIG_IGN);
 }
+
+void os_close_epoll_fd(void)
+{
+	/* Needed so we do not leak an fd when rebooting */
+	os_close_file(epollfd);
+}

arch/um/os-Linux/signal.c

@@ -16,6 +16,7 @@
 #include <os.h>
 #include <sysdep/mcontext.h>
 #include <um_malloc.h>
+#include <sys/ucontext.h>
 
 void (*sig_info[NSIG])(int, struct siginfo *, struct uml_pt_regs *) = {
 	[SIGTRAP]	= relay_signal,
@@ -159,7 +160,7 @@ static void (*handlers[_NSIG])(int sig, struct siginfo *si, mcontext_t *mc) = {
 
 static void hard_handler(int sig, siginfo_t *si, void *p)
 {
-	struct ucontext *uc = p;
+	ucontext_t *uc = p;
 	mcontext_t *mc = &uc->uc_mcontext;
 	unsigned long pending = 1UL << sig;
 

arch/x86/um/stub_segv.c

@@ -6,11 +6,12 @@
 #include <sysdep/stub.h>
 #include <sysdep/faultinfo.h>
 #include <sysdep/mcontext.h>
+#include <sys/ucontext.h>
 
 void __attribute__ ((__section__ (".__syscall_stub")))
 stub_segv_handler(int sig, siginfo_t *info, void *p)
 {
-	struct ucontext *uc = p;
+	ucontext_t *uc = p;
 
 	GET_FAULTINFO_FROM_MC(*((struct faultinfo *) STUB_DATA),
 			      &uc->uc_mcontext);