* [PATCH net-next RFC 1/2] tun: enable NAPI for TUN/TAP driver
2017-09-05 22:35 [PATCH net-next RFC 0/2] Improve code coverage of syzkaller Petar Penkov
@ 2017-09-05 22:35 ` Petar Penkov
2017-09-05 22:51 ` Stephen Hemminger
2017-09-05 22:35 ` [PATCH net-next RFC 2/2] tun: enable napi_gro_frags() " Petar Penkov
1 sibling, 1 reply; 6+ messages in thread
From: Petar Penkov @ 2017-09-05 22:35 UTC (permalink / raw)
To: netdev
Cc: Petar Penkov, Eric Dumazet, Mahesh Bandewar, Willem de Bruijn,
davem, ppenkov
Changes TUN driver to use napi_gro_receive() upon receiving packets
rather than netif_rx_ni(). Adds flag CONFIG_TUN_NAPI that enables
these changes and operation is not affected if the flag is disabled.
SKBs are constructed upon packet arrival and are queued to be
processed later.
The new path was evaluated with a benchmark with the following setup:
Open two tap devices and a receiver thread that reads in a loop for
each device. Start one sender thread and pin all threads to different
CPUs. Send 1M minimum UDP packets to each device and measure sending
time for each of the sending methods:
napi_gro_receive(): 4.90s
netif_rx_ni(): 4.90s
netif_receive_skb(): 7.20s
Signed-off-by: Petar Penkov <ppenkov@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: davem@davemloft.net
Cc: ppenkov@stanford.edu
---
drivers/net/Kconfig | 8 ++++
drivers/net/tun.c | 120 +++++++++++++++++++++++++++++++++++++++++++++++-----
2 files changed, 118 insertions(+), 10 deletions(-)
diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index 83a1616903f8..34850d71ddd1 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -307,6 +307,14 @@ config TAP
This option is selected by any driver implementing tap user space
interface for a virtual interface to re-use core tap functionality.
+config TUN_NAPI
+ bool "NAPI support on tx path for TUN/TAP driver"
+ default n
+ depends on TUN
+ ---help---
+ This option allows the TUN/TAP driver to use NAPI to pass packets to
+ the kernel when receiving packets from user space via write()/send().
+
config TUN_VNET_CROSS_LE
bool "Support for cross-endian vnet headers on little-endian kernels"
default n
diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 06e8f0bb2dab..d5c824e3ec42 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -172,6 +172,7 @@ struct tun_file {
u16 queue_index;
unsigned int ifindex;
};
+ struct napi_struct napi;
struct list_head next;
struct tun_struct *detached;
struct skb_array tx_array;
@@ -229,6 +230,67 @@ struct tun_struct {
struct bpf_prog __rcu *xdp_prog;
};
+static int tun_napi_receive(struct napi_struct *napi, int budget)
+{
+ struct tun_file *tfile = container_of(napi, struct tun_file, napi);
+ struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
+ struct sk_buff_head process_queue;
+ struct sk_buff *skb;
+ int received = 0;
+
+ __skb_queue_head_init(&process_queue);
+
+ spin_lock(&queue->lock);
+ skb_queue_splice_tail_init(queue, &process_queue);
+ spin_unlock(&queue->lock);
+
+ while (received < budget && (skb = __skb_dequeue(&process_queue))) {
+ napi_gro_receive(napi, skb);
+ ++received;
+ }
+
+ if (!skb_queue_empty(&process_queue)) {
+ spin_lock(&queue->lock);
+ skb_queue_splice(&process_queue, queue);
+ spin_unlock(&queue->lock);
+ }
+
+ return received;
+}
+
+static int tun_napi_poll(struct napi_struct *napi, int budget)
+{
+ unsigned int received;
+
+ received = tun_napi_receive(napi, budget);
+
+ if (received < budget)
+ napi_complete_done(napi, received);
+
+ return received;
+}
+
+static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile)
+{
+ if (IS_ENABLED(CONFIG_TUN_NAPI)) {
+ netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
+ NAPI_POLL_WEIGHT);
+ napi_enable(&tfile->napi);
+ }
+}
+
+static void tun_napi_disable(struct tun_file *tfile)
+{
+ if (IS_ENABLED(CONFIG_TUN_NAPI))
+ napi_disable(&tfile->napi);
+}
+
+static void tun_napi_del(struct tun_file *tfile)
+{
+ if (IS_ENABLED(CONFIG_TUN_NAPI))
+ netif_napi_del(&tfile->napi);
+}
+
#ifdef CONFIG_TUN_VNET_CROSS_LE
static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
{
@@ -541,6 +603,11 @@ static void __tun_detach(struct tun_file *tfile, bool clean)
tun = rtnl_dereference(tfile->tun);
+ if (tun && clean) {
+ tun_napi_disable(tfile);
+ tun_napi_del(tfile);
+ }
+
if (tun && !tfile->detached) {
u16 index = tfile->queue_index;
BUG_ON(index >= tun->numqueues);
@@ -598,6 +665,7 @@ static void tun_detach_all(struct net_device *dev)
for (i = 0; i < n; i++) {
tfile = rtnl_dereference(tun->tfiles[i]);
BUG_ON(!tfile);
+ tun_napi_disable(tfile);
tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
tfile->socket.sk->sk_data_ready(tfile->socket.sk);
RCU_INIT_POINTER(tfile->tun, NULL);
@@ -613,6 +681,7 @@ static void tun_detach_all(struct net_device *dev)
synchronize_net();
for (i = 0; i < n; i++) {
tfile = rtnl_dereference(tun->tfiles[i]);
+ tun_napi_del(tfile);
/* Drop read queue */
tun_queue_purge(tfile);
sock_put(&tfile->sk);
@@ -677,10 +746,12 @@ static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filte
rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
tun->numqueues++;
- if (tfile->detached)
+ if (tfile->detached) {
tun_enable_queue(tfile);
- else
+ } else {
sock_hold(&tfile->sk);
+ tun_napi_init(tun, tfile);
+ }
tun_set_real_num_queues(tun);
@@ -956,13 +1027,28 @@ static void tun_poll_controller(struct net_device *dev)
* Tun only receives frames when:
* 1) the char device endpoint gets data from user space
* 2) the tun socket gets a sendmsg call from user space
- * Since both of those are synchronous operations, we are guaranteed
- * never to have pending data when we poll for it
- * so there is nothing to do here but return.
+ * If NAPI is not enabled, since both of those are synchronous
+ * operations, we are guaranteed never to have pending data when we poll
+ * for it so there is nothing to do here but return.
* We need this though so netpoll recognizes us as an interface that
* supports polling, which enables bridge devices in virt setups to
* still use netconsole
+ * If NAPI is enabled, however, we need to schedule polling for all
+ * queues.
*/
+
+ if (IS_ENABLED(CONFIG_TUN_NAPI)) {
+ struct tun_struct *tun = netdev_priv(dev);
+ struct tun_file *tfile;
+ int i;
+
+ rcu_read_lock();
+ for (i = 0; i < tun->numqueues; i++) {
+ tfile = rcu_dereference(tun->tfiles[i]);
+ napi_schedule(&tfile->napi);
+ }
+ rcu_read_unlock();
+ }
return;
}
#endif
@@ -1535,11 +1621,25 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
}
rxhash = __skb_get_hash_symmetric(skb);
-#ifndef CONFIG_4KSTACKS
- tun_rx_batched(tun, tfile, skb, more);
-#else
- netif_rx_ni(skb);
-#endif
+
+ if (IS_ENABLED(CONFIG_TUN_NAPI)) {
+ struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
+ int queue_len;
+
+ spin_lock_bh(&queue->lock);
+ __skb_queue_tail(queue, skb);
+ queue_len = skb_queue_len(queue);
+ spin_unlock(&queue->lock);
+
+ if (!more || queue_len > NAPI_POLL_WEIGHT)
+ napi_schedule(&tfile->napi);
+
+ local_bh_enable();
+ } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
+ tun_rx_batched(tun, tfile, skb, more);
+ } else {
+ netif_rx_ni(skb);
+ }
stats = get_cpu_ptr(tun->pcpu_stats);
u64_stats_update_begin(&stats->syncp);
--
2.14.1.581.gf28d330327-goog
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH net-next RFC 2/2] tun: enable napi_gro_frags() for TUN/TAP driver
2017-09-05 22:35 [PATCH net-next RFC 0/2] Improve code coverage of syzkaller Petar Penkov
2017-09-05 22:35 ` [PATCH net-next RFC 1/2] tun: enable NAPI for TUN/TAP driver Petar Penkov
@ 2017-09-05 22:35 ` Petar Penkov
2017-09-06 18:01 ` Eric Dumazet
1 sibling, 1 reply; 6+ messages in thread
From: Petar Penkov @ 2017-09-05 22:35 UTC (permalink / raw)
To: netdev
Cc: Petar Penkov, Eric Dumazet, Mahesh Bandewar, Willem de Bruijn,
davem, ppenkov
Add a TUN/TAP receive mode that exercises the napi_gro_frags()
interface. This mode is available only in TAP mode, as the interface
expects packets with Ethernet headers.
Furthermore, packets follow the layout of the iovec_iter that was
received. The first iovec is the linear data, and every one after the
first is a fragment. If there are more fragments than the max number,
drop the packet. Additionally, invoke eth_get_headlen() to exercise flow
dissector code and to verify that the header resides in the linear data.
The napi_gro_frags() mode requires setting the IFF_NAPI_FRAGS option.
This is imposed because this mode is intended for testing via tools like
syzkaller and packetdrill, and the increased flexibility it provides can
introduce security vulnerabilities.
Signed-off-by: Petar Penkov <ppenkov@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: davem@davemloft.net
Cc: ppenkov@stanford.edu
---
drivers/net/tun.c | 135 ++++++++++++++++++++++++++++++++++++++++++--
include/uapi/linux/if_tun.h | 1 +
2 files changed, 130 insertions(+), 6 deletions(-)
diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index d5c824e3ec42..2ba9809ab6cd 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -75,6 +75,7 @@
#include <linux/skb_array.h>
#include <linux/bpf.h>
#include <linux/bpf_trace.h>
+#include <linux/mutex.h>
#include <linux/uaccess.h>
@@ -120,8 +121,15 @@ do { \
#define TUN_VNET_LE 0x80000000
#define TUN_VNET_BE 0x40000000
+#if IS_ENABLED(CONFIG_TUN_NAPI)
+#define TUN_FEATURES_EXTRA IFF_NAPI_FRAGS
+#else
+#define TUN_FEATURES_EXTRA 0
+#endif
+
#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
- IFF_MULTI_QUEUE)
+ IFF_MULTI_QUEUE | TUN_FEATURES_EXTRA)
+
#define GOODCOPY_LEN 128
#define FLT_EXACT_COUNT 8
@@ -173,6 +181,7 @@ struct tun_file {
unsigned int ifindex;
};
struct napi_struct napi;
+ struct mutex napi_mutex; /* Protects access to the above napi */
struct list_head next;
struct tun_struct *detached;
struct skb_array tx_array;
@@ -276,6 +285,7 @@ static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile)
netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
NAPI_POLL_WEIGHT);
napi_enable(&tfile->napi);
+ mutex_init(&tfile->napi_mutex);
}
}
@@ -291,6 +301,11 @@ static void tun_napi_del(struct tun_file *tfile)
netif_napi_del(&tfile->napi);
}
+static bool tun_napi_frags_enabled(const struct tun_struct *tun)
+{
+ return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
+}
+
#ifdef CONFIG_TUN_VNET_CROSS_LE
static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
{
@@ -1034,7 +1049,8 @@ static void tun_poll_controller(struct net_device *dev)
* supports polling, which enables bridge devices in virt setups to
* still use netconsole
* If NAPI is enabled, however, we need to schedule polling for all
- * queues.
+ * queues unless we are using napi_gro_frags(), which we call in
+ * process context and not in NAPI context.
*/
if (IS_ENABLED(CONFIG_TUN_NAPI)) {
@@ -1042,6 +1058,9 @@ static void tun_poll_controller(struct net_device *dev)
struct tun_file *tfile;
int i;
+ if (tun_napi_frags_enabled(tun))
+ return;
+
rcu_read_lock();
for (i = 0; i < tun->numqueues; i++) {
tfile = rcu_dereference(tun->tfiles[i]);
@@ -1264,6 +1283,64 @@ static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
return mask;
}
+static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
+ size_t len,
+ const struct iov_iter *it)
+{
+ struct sk_buff *skb;
+ size_t linear;
+ int err;
+ int i;
+
+ if (it->nr_segs > MAX_SKB_FRAGS + 1)
+ return ERR_PTR(-ENOMEM);
+
+ local_bh_disable();
+ skb = napi_get_frags(&tfile->napi);
+ local_bh_enable();
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
+
+ linear = iov_iter_single_seg_count(it);
+ err = __skb_grow(skb, linear);
+ if (err)
+ goto free;
+
+ skb->len = len;
+ skb->data_len = len - linear;
+ skb->truesize += skb->data_len;
+
+ for (i = 1; i < it->nr_segs; i++) {
+ size_t fragsz = it->iov[i].iov_len;
+ unsigned long offset;
+ struct page *page;
+ void *data;
+
+ if (fragsz == 0 || fragsz > PAGE_SIZE) {
+ err = -EINVAL;
+ goto free;
+ }
+
+ local_bh_disable();
+ data = napi_alloc_frag(fragsz);
+ local_bh_enable();
+ if (!data) {
+ err = -ENOMEM;
+ goto free;
+ }
+
+ page = virt_to_page(data);
+ offset = offset_in_page(data);
+ skb_fill_page_desc(skb, i - 1, page, offset, fragsz);
+ }
+
+ return skb;
+free:
+ /* frees skb and all frags allocated with napi_alloc_frag() */
+ napi_free_frags(&tfile->napi);
+ return ERR_PTR(err);
+}
+
/* prepad is the amount to reserve at front. len is length after that.
* linear is a hint as to how much to copy (usually headers). */
static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
@@ -1466,6 +1543,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
int err;
u32 rxhash;
int generic_xdp = 1;
+ bool frags = tun_napi_frags_enabled(tun);
if (!(tun->dev->flags & IFF_UP))
return -EIO;
@@ -1523,7 +1601,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
zerocopy = true;
}
- if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
+ if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
skb = tun_build_skb(tun, tfile, from, &gso, len, &generic_xdp);
if (IS_ERR(skb)) {
this_cpu_inc(tun->pcpu_stats->rx_dropped);
@@ -1540,10 +1618,24 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
linear = tun16_to_cpu(tun, gso.hdr_len);
}
- skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
+ if (frags) {
+ mutex_lock(&tfile->napi_mutex);
+ skb = tun_napi_alloc_frags(tfile, copylen, from);
+ /* tun_napi_alloc_frags() enforces a layout for the skb.
+ * If zerocopy is enabled, then this layout will be
+ * overwritten by zerocopy_sg_from_iter().
+ */
+ zerocopy = false;
+ } else {
+ skb = tun_alloc_skb(tfile, align, copylen, linear,
+ noblock);
+ }
+
if (IS_ERR(skb)) {
if (PTR_ERR(skb) != -EAGAIN)
this_cpu_inc(tun->pcpu_stats->rx_dropped);
+ if (frags)
+ mutex_unlock(&tfile->napi_mutex);
return PTR_ERR(skb);
}
@@ -1555,6 +1647,11 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
if (err) {
this_cpu_inc(tun->pcpu_stats->rx_dropped);
kfree_skb(skb);
+ if (frags) {
+ tfile->napi.skb = NULL;
+ mutex_unlock(&tfile->napi_mutex);
+ }
+
return -EFAULT;
}
}
@@ -1562,6 +1659,11 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
kfree_skb(skb);
+ if (frags) {
+ tfile->napi.skb = NULL;
+ mutex_unlock(&tfile->napi_mutex);
+ }
+
return -EINVAL;
}
@@ -1587,7 +1689,8 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
skb->dev = tun->dev;
break;
case IFF_TAP:
- skb->protocol = eth_type_trans(skb, tun->dev);
+ if (!frags)
+ skb->protocol = eth_type_trans(skb, tun->dev);
break;
}
@@ -1622,7 +1725,23 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
rxhash = __skb_get_hash_symmetric(skb);
- if (IS_ENABLED(CONFIG_TUN_NAPI)) {
+ if (frags) {
+ /* Exercise flow dissector code path. */
+ u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
+
+ if (headlen > skb_headlen(skb) || headlen < ETH_HLEN) {
+ this_cpu_inc(tun->pcpu_stats->rx_dropped);
+ napi_free_frags(&tfile->napi);
+ mutex_unlock(&tfile->napi_mutex);
+ WARN_ON(1);
+ return -ENOMEM;
+ }
+
+ local_bh_disable();
+ napi_gro_frags(&tfile->napi);
+ local_bh_enable();
+ mutex_unlock(&tfile->napi_mutex);
+ } else if (IS_ENABLED(CONFIG_TUN_NAPI)) {
struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
int queue_len;
@@ -2168,6 +2287,10 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
tun->flags = (tun->flags & ~TUN_FEATURES) |
(ifr->ifr_flags & TUN_FEATURES);
+ if (!IS_ENABLED(CONFIG_TUN_NAPI) ||
+ (tun->flags & TUN_TYPE_MASK) != IFF_TAP)
+ tun->flags = tun->flags & ~IFF_NAPI_FRAGS;
+
/* Make sure persistent devices do not get stuck in
* xoff state.
*/
diff --git a/include/uapi/linux/if_tun.h b/include/uapi/linux/if_tun.h
index 3cb5e1d85ddd..1eb1eb42f151 100644
--- a/include/uapi/linux/if_tun.h
+++ b/include/uapi/linux/if_tun.h
@@ -60,6 +60,7 @@
/* TUNSETIFF ifr flags */
#define IFF_TUN 0x0001
#define IFF_TAP 0x0002
+#define IFF_NAPI_FRAGS 0x0010
#define IFF_NO_PI 0x1000
/* This flag has no real effect */
#define IFF_ONE_QUEUE 0x2000
--
2.14.1.581.gf28d330327-goog
^ permalink raw reply related [flat|nested] 6+ messages in thread