linux/net/ipv4/tcp_offload.c

360 lines
8.6 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* IPV4 GSO/GRO offload support
* Linux INET implementation
*
* TCPv4 GSO/GRO support
*/
#include <linux/indirect_call_wrapper.h>
#include <linux/skbuff.h>
#include <net/gro.h>
#include <net/tcp.h>
#include <net/protocol.h>
static void tcp_gso_tstamp(struct sk_buff *skb, unsigned int ts_seq,
unsigned int seq, unsigned int mss)
net-timestamp: TCP timestamping TCP timestamping extends SO_TIMESTAMPING to bytestreams. Bytestreams do not have a 1:1 relationship between send() buffers and network packets. The feature interprets a send call on a bytestream as a request for a timestamp for the last byte in that send() buffer. The choice corresponds to a request for a timestamp when all bytes in the buffer have been sent. That assumption depends on in-order kernel transmission. This is the common case. That said, it is possible to construct a traffic shaping tree that would result in reordering. The guarantee is strong, then, but not ironclad. This implementation supports send and sendpages (splice). GSO replaces one large packet with multiple smaller packets. This patch also copies the option into the correct smaller packet. This patch does not yet support timestamping on data in an initial TCP Fast Open SYN, because that takes a very different data path. If ID generation in ee_data is enabled, bytestream timestamps return a byte offset, instead of the packet counter for datagrams. The implementation supports a single timestamp per packet. It silenty replaces requests for previous timestamps. To avoid missing tstamps, flush the tcp queue by disabling Nagle, cork and autocork. Missing tstamps can be detected by offset when the ee_data ID is enabled. Implementation details: - On GSO, the timestamping code can be included in the main loop. I moved it into its own loop to reduce the impact on the common case to a single branch. - To avoid leaking the absolute seqno to userspace, the offset returned in ee_data must always be relative. It is an offset between an skb and sk field. The first is always set (also for GSO & ACK). The second must also never be uninitialized. Only allow the ID option on sockets in the ESTABLISHED state, for which the seqno is available. Never reset it to zero (instead, move it to the current seqno when reenabling the option). Signed-off-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-05 10:11:49 +08:00
{
while (skb) {
if (before(ts_seq, seq + mss)) {
skb_shinfo(skb)->tx_flags |= SKBTX_SW_TSTAMP;
net-timestamp: TCP timestamping TCP timestamping extends SO_TIMESTAMPING to bytestreams. Bytestreams do not have a 1:1 relationship between send() buffers and network packets. The feature interprets a send call on a bytestream as a request for a timestamp for the last byte in that send() buffer. The choice corresponds to a request for a timestamp when all bytes in the buffer have been sent. That assumption depends on in-order kernel transmission. This is the common case. That said, it is possible to construct a traffic shaping tree that would result in reordering. The guarantee is strong, then, but not ironclad. This implementation supports send and sendpages (splice). GSO replaces one large packet with multiple smaller packets. This patch also copies the option into the correct smaller packet. This patch does not yet support timestamping on data in an initial TCP Fast Open SYN, because that takes a very different data path. If ID generation in ee_data is enabled, bytestream timestamps return a byte offset, instead of the packet counter for datagrams. The implementation supports a single timestamp per packet. It silenty replaces requests for previous timestamps. To avoid missing tstamps, flush the tcp queue by disabling Nagle, cork and autocork. Missing tstamps can be detected by offset when the ee_data ID is enabled. Implementation details: - On GSO, the timestamping code can be included in the main loop. I moved it into its own loop to reduce the impact on the common case to a single branch. - To avoid leaking the absolute seqno to userspace, the offset returned in ee_data must always be relative. It is an offset between an skb and sk field. The first is always set (also for GSO & ACK). The second must also never be uninitialized. Only allow the ID option on sockets in the ESTABLISHED state, for which the seqno is available. Never reset it to zero (instead, move it to the current seqno when reenabling the option). Signed-off-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-05 10:11:49 +08:00
skb_shinfo(skb)->tskey = ts_seq;
return;
}
skb = skb->next;
seq += mss;
}
}
static struct sk_buff *tcp4_gso_segment(struct sk_buff *skb,
netdev_features_t features)
{
if (!(skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4))
return ERR_PTR(-EINVAL);
if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
return ERR_PTR(-EINVAL);
if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
const struct iphdr *iph = ip_hdr(skb);
struct tcphdr *th = tcp_hdr(skb);
/* Set up checksum pseudo header, usually expect stack to
* have done this already.
*/
th->check = 0;
skb->ip_summed = CHECKSUM_PARTIAL;
__tcp_v4_send_check(skb, iph->saddr, iph->daddr);
}
return tcp_gso_segment(skb, features);
}
struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
netdev_features_t features)
{
struct sk_buff *segs = ERR_PTR(-EINVAL);
unsigned int sum_truesize = 0;
struct tcphdr *th;
unsigned int thlen;
unsigned int seq;
__be32 delta;
unsigned int oldlen;
unsigned int mss;
struct sk_buff *gso_skb = skb;
__sum16 newcheck;
bool ooo_okay, copy_destructor;
th = tcp_hdr(skb);
thlen = th->doff * 4;
if (thlen < sizeof(*th))
goto out;
if (!pskb_may_pull(skb, thlen))
goto out;
oldlen = (u16)~skb->len;
__skb_pull(skb, thlen);
mss = skb_shinfo(skb)->gso_size;
if (unlikely(skb->len <= mss))
goto out;
if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
/* Packet is from an untrusted source, reset gso_segs. */
skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
segs = NULL;
goto out;
}
copy_destructor = gso_skb->destructor == tcp_wfree;
ooo_okay = gso_skb->ooo_okay;
/* All segments but the first should have ooo_okay cleared */
skb->ooo_okay = 0;
segs = skb_segment(skb, features);
if (IS_ERR(segs))
goto out;
/* Only first segment might have ooo_okay set */
segs->ooo_okay = ooo_okay;
/* GSO partial and frag_list segmentation only requires splitting
* the frame into an MSS multiple and possibly a remainder, both
* cases return a GSO skb. So update the mss now.
*/
if (skb_is_gso(segs))
mss *= skb_shinfo(segs)->gso_segs;
delta = htonl(oldlen + (thlen + mss));
skb = segs;
th = tcp_hdr(skb);
seq = ntohl(th->seq);
net-timestamp: TCP timestamping TCP timestamping extends SO_TIMESTAMPING to bytestreams. Bytestreams do not have a 1:1 relationship between send() buffers and network packets. The feature interprets a send call on a bytestream as a request for a timestamp for the last byte in that send() buffer. The choice corresponds to a request for a timestamp when all bytes in the buffer have been sent. That assumption depends on in-order kernel transmission. This is the common case. That said, it is possible to construct a traffic shaping tree that would result in reordering. The guarantee is strong, then, but not ironclad. This implementation supports send and sendpages (splice). GSO replaces one large packet with multiple smaller packets. This patch also copies the option into the correct smaller packet. This patch does not yet support timestamping on data in an initial TCP Fast Open SYN, because that takes a very different data path. If ID generation in ee_data is enabled, bytestream timestamps return a byte offset, instead of the packet counter for datagrams. The implementation supports a single timestamp per packet. It silenty replaces requests for previous timestamps. To avoid missing tstamps, flush the tcp queue by disabling Nagle, cork and autocork. Missing tstamps can be detected by offset when the ee_data ID is enabled. Implementation details: - On GSO, the timestamping code can be included in the main loop. I moved it into its own loop to reduce the impact on the common case to a single branch. - To avoid leaking the absolute seqno to userspace, the offset returned in ee_data must always be relative. It is an offset between an skb and sk field. The first is always set (also for GSO & ACK). The second must also never be uninitialized. Only allow the ID option on sockets in the ESTABLISHED state, for which the seqno is available. Never reset it to zero (instead, move it to the current seqno when reenabling the option). Signed-off-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-05 10:11:49 +08:00
if (unlikely(skb_shinfo(gso_skb)->tx_flags & SKBTX_SW_TSTAMP))
tcp_gso_tstamp(segs, skb_shinfo(gso_skb)->tskey, seq, mss);
newcheck = ~csum_fold((__force __wsum)((__force u32)th->check +
(__force u32)delta));
while (skb->next) {
th->fin = th->psh = 0;
th->check = newcheck;
if (skb->ip_summed == CHECKSUM_PARTIAL)
gso_reset_checksum(skb, ~th->check);
else
th->check = gso_make_checksum(skb, ~th->check);
seq += mss;
if (copy_destructor) {
skb->destructor = gso_skb->destructor;
skb->sk = gso_skb->sk;
sum_truesize += skb->truesize;
}
skb = skb->next;
th = tcp_hdr(skb);
th->seq = htonl(seq);
th->cwr = 0;
}
/* Following permits TCP Small Queues to work well with GSO :
* The callback to TCP stack will be called at the time last frag
* is freed at TX completion, and not right now when gso_skb
* is freed by GSO engine
*/
if (copy_destructor) {
tcp: gso: avoid refcount_t warning from tcp_gso_segment() When a GSO skb of truesize O is segmented into 2 new skbs of truesize N1 and N2, we want to transfer socket ownership to the new fresh skbs. In order to avoid expensive atomic operations on a cache line subject to cache bouncing, we replace the sequence : refcount_add(N1, &sk->sk_wmem_alloc); refcount_add(N2, &sk->sk_wmem_alloc); // repeated by number of segments refcount_sub(O, &sk->sk_wmem_alloc); by a single refcount_add(sum_of(N) - O, &sk->sk_wmem_alloc); Problem is : In some pathological cases, sum(N) - O might be a negative number, and syzkaller bot was apparently able to trigger this trace [1] atomic_t was ok with this construct, but we need to take care of the negative delta with refcount_t [1] refcount_t: saturated; leaking memory. ------------[ cut here ]------------ WARNING: CPU: 0 PID: 8404 at lib/refcount.c:77 refcount_add_not_zero+0x198/0x200 lib/refcount.c:77 Kernel panic - not syncing: panic_on_warn set ... CPU: 0 PID: 8404 Comm: syz-executor2 Not tainted 4.14.0-rc5-mm1+ #20 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:16 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:52 panic+0x1e4/0x41c kernel/panic.c:183 __warn+0x1c4/0x1e0 kernel/panic.c:546 report_bug+0x211/0x2d0 lib/bug.c:183 fixup_bug+0x40/0x90 arch/x86/kernel/traps.c:177 do_trap_no_signal arch/x86/kernel/traps.c:211 [inline] do_trap+0x260/0x390 arch/x86/kernel/traps.c:260 do_error_trap+0x120/0x390 arch/x86/kernel/traps.c:297 do_invalid_op+0x1b/0x20 arch/x86/kernel/traps.c:310 invalid_op+0x18/0x20 arch/x86/entry/entry_64.S:905 RIP: 0010:refcount_add_not_zero+0x198/0x200 lib/refcount.c:77 RSP: 0018:ffff8801c606e3a0 EFLAGS: 00010282 RAX: 0000000000000026 RBX: 0000000000001401 RCX: 0000000000000000 RDX: 0000000000000026 RSI: ffffc900036fc000 RDI: ffffed0038c0dc68 RBP: ffff8801c606e430 R08: 0000000000000001 R09: 0000000000000000 R10: ffff8801d97f5eba R11: 0000000000000000 R12: ffff8801d5acf73c R13: 1ffff10038c0dc75 R14: 00000000ffffffff R15: 00000000fffff72f refcount_add+0x1b/0x60 lib/refcount.c:101 tcp_gso_segment+0x10d0/0x16b0 net/ipv4/tcp_offload.c:155 tcp4_gso_segment+0xd4/0x310 net/ipv4/tcp_offload.c:51 inet_gso_segment+0x60c/0x11c0 net/ipv4/af_inet.c:1271 skb_mac_gso_segment+0x33f/0x660 net/core/dev.c:2749 __skb_gso_segment+0x35f/0x7f0 net/core/dev.c:2821 skb_gso_segment include/linux/netdevice.h:3971 [inline] validate_xmit_skb+0x4ba/0xb20 net/core/dev.c:3074 __dev_queue_xmit+0xe49/0x2070 net/core/dev.c:3497 dev_queue_xmit+0x17/0x20 net/core/dev.c:3538 neigh_hh_output include/net/neighbour.h:471 [inline] neigh_output include/net/neighbour.h:479 [inline] ip_finish_output2+0xece/0x1460 net/ipv4/ip_output.c:229 ip_finish_output+0x85e/0xd10 net/ipv4/ip_output.c:317 NF_HOOK_COND include/linux/netfilter.h:238 [inline] ip_output+0x1cc/0x860 net/ipv4/ip_output.c:405 dst_output include/net/dst.h:459 [inline] ip_local_out+0x95/0x160 net/ipv4/ip_output.c:124 ip_queue_xmit+0x8c6/0x18e0 net/ipv4/ip_output.c:504 tcp_transmit_skb+0x1ab7/0x3840 net/ipv4/tcp_output.c:1137 tcp_write_xmit+0x663/0x4de0 net/ipv4/tcp_output.c:2341 __tcp_push_pending_frames+0xa0/0x250 net/ipv4/tcp_output.c:2513 tcp_push_pending_frames include/net/tcp.h:1722 [inline] tcp_data_snd_check net/ipv4/tcp_input.c:5050 [inline] tcp_rcv_established+0x8c7/0x18a0 net/ipv4/tcp_input.c:5497 tcp_v4_do_rcv+0x2ab/0x7d0 net/ipv4/tcp_ipv4.c:1460 sk_backlog_rcv include/net/sock.h:909 [inline] __release_sock+0x124/0x360 net/core/sock.c:2264 release_sock+0xa4/0x2a0 net/core/sock.c:2776 tcp_sendmsg+0x3a/0x50 net/ipv4/tcp.c:1462 inet_sendmsg+0x11f/0x5e0 net/ipv4/af_inet.c:763 sock_sendmsg_nosec net/socket.c:632 [inline] sock_sendmsg+0xca/0x110 net/socket.c:642 ___sys_sendmsg+0x31c/0x890 net/socket.c:2048 __sys_sendmmsg+0x1e6/0x5f0 net/socket.c:2138 Fixes: 14afee4b6092 ("net: convert sock.sk_wmem_alloc from atomic_t to refcount_t") Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: syzbot <syzkaller@googlegroups.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-11-08 07:15:04 +08:00
int delta;
swap(gso_skb->sk, skb->sk);
swap(gso_skb->destructor, skb->destructor);
sum_truesize += skb->truesize;
tcp: gso: avoid refcount_t warning from tcp_gso_segment() When a GSO skb of truesize O is segmented into 2 new skbs of truesize N1 and N2, we want to transfer socket ownership to the new fresh skbs. In order to avoid expensive atomic operations on a cache line subject to cache bouncing, we replace the sequence : refcount_add(N1, &sk->sk_wmem_alloc); refcount_add(N2, &sk->sk_wmem_alloc); // repeated by number of segments refcount_sub(O, &sk->sk_wmem_alloc); by a single refcount_add(sum_of(N) - O, &sk->sk_wmem_alloc); Problem is : In some pathological cases, sum(N) - O might be a negative number, and syzkaller bot was apparently able to trigger this trace [1] atomic_t was ok with this construct, but we need to take care of the negative delta with refcount_t [1] refcount_t: saturated; leaking memory. ------------[ cut here ]------------ WARNING: CPU: 0 PID: 8404 at lib/refcount.c:77 refcount_add_not_zero+0x198/0x200 lib/refcount.c:77 Kernel panic - not syncing: panic_on_warn set ... CPU: 0 PID: 8404 Comm: syz-executor2 Not tainted 4.14.0-rc5-mm1+ #20 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:16 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:52 panic+0x1e4/0x41c kernel/panic.c:183 __warn+0x1c4/0x1e0 kernel/panic.c:546 report_bug+0x211/0x2d0 lib/bug.c:183 fixup_bug+0x40/0x90 arch/x86/kernel/traps.c:177 do_trap_no_signal arch/x86/kernel/traps.c:211 [inline] do_trap+0x260/0x390 arch/x86/kernel/traps.c:260 do_error_trap+0x120/0x390 arch/x86/kernel/traps.c:297 do_invalid_op+0x1b/0x20 arch/x86/kernel/traps.c:310 invalid_op+0x18/0x20 arch/x86/entry/entry_64.S:905 RIP: 0010:refcount_add_not_zero+0x198/0x200 lib/refcount.c:77 RSP: 0018:ffff8801c606e3a0 EFLAGS: 00010282 RAX: 0000000000000026 RBX: 0000000000001401 RCX: 0000000000000000 RDX: 0000000000000026 RSI: ffffc900036fc000 RDI: ffffed0038c0dc68 RBP: ffff8801c606e430 R08: 0000000000000001 R09: 0000000000000000 R10: ffff8801d97f5eba R11: 0000000000000000 R12: ffff8801d5acf73c R13: 1ffff10038c0dc75 R14: 00000000ffffffff R15: 00000000fffff72f refcount_add+0x1b/0x60 lib/refcount.c:101 tcp_gso_segment+0x10d0/0x16b0 net/ipv4/tcp_offload.c:155 tcp4_gso_segment+0xd4/0x310 net/ipv4/tcp_offload.c:51 inet_gso_segment+0x60c/0x11c0 net/ipv4/af_inet.c:1271 skb_mac_gso_segment+0x33f/0x660 net/core/dev.c:2749 __skb_gso_segment+0x35f/0x7f0 net/core/dev.c:2821 skb_gso_segment include/linux/netdevice.h:3971 [inline] validate_xmit_skb+0x4ba/0xb20 net/core/dev.c:3074 __dev_queue_xmit+0xe49/0x2070 net/core/dev.c:3497 dev_queue_xmit+0x17/0x20 net/core/dev.c:3538 neigh_hh_output include/net/neighbour.h:471 [inline] neigh_output include/net/neighbour.h:479 [inline] ip_finish_output2+0xece/0x1460 net/ipv4/ip_output.c:229 ip_finish_output+0x85e/0xd10 net/ipv4/ip_output.c:317 NF_HOOK_COND include/linux/netfilter.h:238 [inline] ip_output+0x1cc/0x860 net/ipv4/ip_output.c:405 dst_output include/net/dst.h:459 [inline] ip_local_out+0x95/0x160 net/ipv4/ip_output.c:124 ip_queue_xmit+0x8c6/0x18e0 net/ipv4/ip_output.c:504 tcp_transmit_skb+0x1ab7/0x3840 net/ipv4/tcp_output.c:1137 tcp_write_xmit+0x663/0x4de0 net/ipv4/tcp_output.c:2341 __tcp_push_pending_frames+0xa0/0x250 net/ipv4/tcp_output.c:2513 tcp_push_pending_frames include/net/tcp.h:1722 [inline] tcp_data_snd_check net/ipv4/tcp_input.c:5050 [inline] tcp_rcv_established+0x8c7/0x18a0 net/ipv4/tcp_input.c:5497 tcp_v4_do_rcv+0x2ab/0x7d0 net/ipv4/tcp_ipv4.c:1460 sk_backlog_rcv include/net/sock.h:909 [inline] __release_sock+0x124/0x360 net/core/sock.c:2264 release_sock+0xa4/0x2a0 net/core/sock.c:2776 tcp_sendmsg+0x3a/0x50 net/ipv4/tcp.c:1462 inet_sendmsg+0x11f/0x5e0 net/ipv4/af_inet.c:763 sock_sendmsg_nosec net/socket.c:632 [inline] sock_sendmsg+0xca/0x110 net/socket.c:642 ___sys_sendmsg+0x31c/0x890 net/socket.c:2048 __sys_sendmmsg+0x1e6/0x5f0 net/socket.c:2138 Fixes: 14afee4b6092 ("net: convert sock.sk_wmem_alloc from atomic_t to refcount_t") Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: syzbot <syzkaller@googlegroups.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-11-08 07:15:04 +08:00
delta = sum_truesize - gso_skb->truesize;
/* In some pathological cases, delta can be negative.
* We need to either use refcount_add() or refcount_sub_and_test()
*/
if (likely(delta >= 0))
refcount_add(delta, &skb->sk->sk_wmem_alloc);
else
WARN_ON_ONCE(refcount_sub_and_test(-delta, &skb->sk->sk_wmem_alloc));
}
delta = htonl(oldlen + (skb_tail_pointer(skb) -
skb_transport_header(skb)) +
skb->data_len);
th->check = ~csum_fold((__force __wsum)((__force u32)th->check +
(__force u32)delta));
if (skb->ip_summed == CHECKSUM_PARTIAL)
gso_reset_checksum(skb, ~th->check);
else
th->check = gso_make_checksum(skb, ~th->check);
out:
return segs;
}
struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb)
{
struct sk_buff *pp = NULL;
struct sk_buff *p;
struct tcphdr *th;
struct tcphdr *th2;
unsigned int len;
unsigned int thlen;
__be32 flags;
unsigned int mss = 1;
unsigned int hlen;
unsigned int off;
int flush = 1;
int i;
off = skb_gro_offset(skb);
hlen = off + sizeof(*th);
th = skb_gro_header(skb, hlen, off);
if (unlikely(!th))
goto out;
thlen = th->doff * 4;
if (thlen < sizeof(*th))
goto out;
hlen = off + thlen;
if (skb_gro_header_hard(skb, hlen)) {
th = skb_gro_header_slow(skb, hlen, off);
if (unlikely(!th))
goto out;
}
skb_gro_pull(skb, thlen);
len = skb_gro_len(skb);
flags = tcp_flag_word(th);
list_for_each_entry(p, head, list) {
if (!NAPI_GRO_CB(p)->same_flow)
continue;
th2 = tcp_hdr(p);
if (*(u32 *)&th->source ^ *(u32 *)&th2->source) {
NAPI_GRO_CB(p)->same_flow = 0;
continue;
}
goto found;
}
p = NULL;
goto out_check_final;
found:
net-gre-gro: Add GRE support to the GRO stack This patch built on top of Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") to add the support of the standard GRE (RFC1701/RFC2784/RFC2890) to the GRO stack. It also serves as an example for supporting other encapsulation protocols in the GRO stack in the future. The patch supports version 0 and all the flags (key, csum, seq#) but will flush any pkt with the S (seq#) flag. This is because the S flag is not support by GSO, and a GRO pkt may end up in the forwarding path, thus requiring GSO support to break it up correctly. Currently the "packet_offload" structure only contains L3 (ETH_P_IP/ ETH_P_IPV6) GRO offload support so the encapped pkts are limited to IP pkts (i.e., w/o L2 hdr). But support for other protocol type can be easily added, so is the support for GRE variations like NVGRE. The patch also support csum offload. Specifically if the csum flag is on and the h/w is capable of checksumming the payload (CHECKSUM_COMPLETE), the code will take advantage of the csum computed by the h/w when validating the GRE csum. Note that commit 60769a5dcd8755715c7143b4571d5c44f01796f1 "ipv4: gre: add GRO capability" already introduces GRO capability to IPv4 GRE tunnels, using the gro_cells infrastructure. But GRO is done after GRE hdr has been removed (i.e., decapped). The following patch applies GRO when pkts first come in (before hitting the GRE tunnel code). There is some performance advantage for applying GRO as early as possible. Also this approach is transparent to other subsystem like Open vSwitch where GRE decap is handled outside of the IP stack hence making it harder for the gro_cells stuff to apply. On the other hand, some NICs are still not capable of hashing on the inner hdr of a GRE pkt (RSS). In that case the GRO processing of pkts from the same remote host will all happen on the same CPU and the performance may be suboptimal. I'm including some rough preliminary performance numbers below. Note that the performance will be highly dependent on traffic load, mix as usual. Moreover it also depends on NIC offload features hence the following is by no means a comprehesive study. Local testing and tuning will be needed to decide the best setting. All tests spawned 50 copies of netperf TCP_STREAM and ran for 30 secs. (super_netperf 50 -H 192.168.1.18 -l 30) An IP GRE tunnel with only the key flag on (e.g., ip tunnel add gre1 mode gre local 10.246.17.18 remote 10.246.17.17 ttl 255 key 123) is configured. The GRO support for pkts AFTER decap are controlled through the device feature of the GRE device (e.g., ethtool -K gre1 gro on/off). 1.1 ethtool -K gre1 gro off; ethtool -K eth0 gro off thruput: 9.16Gbps CPU utilization: 19% 1.2 ethtool -K gre1 gro on; ethtool -K eth0 gro off thruput: 5.9Gbps CPU utilization: 15% 1.3 ethtool -K gre1 gro off; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 12-13% 1.4 ethtool -K gre1 gro on; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 10% The following tests were performed on a different NIC that is capable of csum offload. I.e., the h/w is capable of computing IP payload csum (CHECKSUM_COMPLETE). 2.1 ethtool -K gre1 gro on (hence will use gro_cells) 2.1.1 ethtool -K eth0 gro off; csum offload disabled thruput: 8.53Gbps CPU utilization: 9% 2.1.2 ethtool -K eth0 gro off; csum offload enabled thruput: 8.97Gbps CPU utilization: 7-8% 2.1.3 ethtool -K eth0 gro on; csum offload disabled thruput: 8.83Gbps CPU utilization: 5-6% 2.1.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.98Gbps CPU utilization: 5% 2.2 ethtool -K gre1 gro off 2.2.1 ethtool -K eth0 gro off; csum offload disabled thruput: 5.93Gbps CPU utilization: 9% 2.2.2 ethtool -K eth0 gro off; csum offload enabled thruput: 5.62Gbps CPU utilization: 8% 2.2.3 ethtool -K eth0 gro on; csum offload disabled thruput: 7.69Gbps CPU utilization: 8% 2.2.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.96Gbps CPU utilization: 5-6% Signed-off-by: H.K. Jerry Chu <hkchu@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-08 02:23:19 +08:00
/* Include the IP ID check below from the inner most IP hdr */
flush = NAPI_GRO_CB(p)->flush;
flush |= (__force int)(flags & TCP_FLAG_CWR);
flush |= (__force int)((flags ^ tcp_flag_word(th2)) &
~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH));
flush |= (__force int)(th->ack_seq ^ th2->ack_seq);
for (i = sizeof(*th); i < thlen; i += 4)
flush |= *(u32 *)((u8 *)th + i) ^
*(u32 *)((u8 *)th2 + i);
/* When we receive our second frame we can made a decision on if we
* continue this flow as an atomic flow with a fixed ID or if we use
* an incrementing ID.
*/
if (NAPI_GRO_CB(p)->flush_id != 1 ||
NAPI_GRO_CB(p)->count != 1 ||
!NAPI_GRO_CB(p)->is_atomic)
flush |= NAPI_GRO_CB(p)->flush_id;
else
NAPI_GRO_CB(p)->is_atomic = false;
mss = skb_shinfo(p)->gso_size;
/* If skb is a GRO packet, make sure its gso_size matches prior packet mss.
* If it is a single frame, do not aggregate it if its length
* is bigger than our mss.
*/
if (unlikely(skb_is_gso(skb)))
flush |= (mss != skb_shinfo(skb)->gso_size);
else
flush |= (len - 1) >= mss;
flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq);
#ifdef CONFIG_TLS_DEVICE
flush |= p->decrypted ^ skb->decrypted;
#endif
if (flush || skb_gro_receive(p, skb)) {
mss = 1;
goto out_check_final;
}
tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH);
out_check_final:
/* Force a flush if last segment is smaller than mss. */
if (unlikely(skb_is_gso(skb)))
flush = len != NAPI_GRO_CB(skb)->count * skb_shinfo(skb)->gso_size;
else
flush = len < mss;
flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH |
TCP_FLAG_RST | TCP_FLAG_SYN |
TCP_FLAG_FIN));
if (p && (!NAPI_GRO_CB(skb)->same_flow || flush))
pp = p;
out:
net-gre-gro: Add GRE support to the GRO stack This patch built on top of Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") to add the support of the standard GRE (RFC1701/RFC2784/RFC2890) to the GRO stack. It also serves as an example for supporting other encapsulation protocols in the GRO stack in the future. The patch supports version 0 and all the flags (key, csum, seq#) but will flush any pkt with the S (seq#) flag. This is because the S flag is not support by GSO, and a GRO pkt may end up in the forwarding path, thus requiring GSO support to break it up correctly. Currently the "packet_offload" structure only contains L3 (ETH_P_IP/ ETH_P_IPV6) GRO offload support so the encapped pkts are limited to IP pkts (i.e., w/o L2 hdr). But support for other protocol type can be easily added, so is the support for GRE variations like NVGRE. The patch also support csum offload. Specifically if the csum flag is on and the h/w is capable of checksumming the payload (CHECKSUM_COMPLETE), the code will take advantage of the csum computed by the h/w when validating the GRE csum. Note that commit 60769a5dcd8755715c7143b4571d5c44f01796f1 "ipv4: gre: add GRO capability" already introduces GRO capability to IPv4 GRE tunnels, using the gro_cells infrastructure. But GRO is done after GRE hdr has been removed (i.e., decapped). The following patch applies GRO when pkts first come in (before hitting the GRE tunnel code). There is some performance advantage for applying GRO as early as possible. Also this approach is transparent to other subsystem like Open vSwitch where GRE decap is handled outside of the IP stack hence making it harder for the gro_cells stuff to apply. On the other hand, some NICs are still not capable of hashing on the inner hdr of a GRE pkt (RSS). In that case the GRO processing of pkts from the same remote host will all happen on the same CPU and the performance may be suboptimal. I'm including some rough preliminary performance numbers below. Note that the performance will be highly dependent on traffic load, mix as usual. Moreover it also depends on NIC offload features hence the following is by no means a comprehesive study. Local testing and tuning will be needed to decide the best setting. All tests spawned 50 copies of netperf TCP_STREAM and ran for 30 secs. (super_netperf 50 -H 192.168.1.18 -l 30) An IP GRE tunnel with only the key flag on (e.g., ip tunnel add gre1 mode gre local 10.246.17.18 remote 10.246.17.17 ttl 255 key 123) is configured. The GRO support for pkts AFTER decap are controlled through the device feature of the GRE device (e.g., ethtool -K gre1 gro on/off). 1.1 ethtool -K gre1 gro off; ethtool -K eth0 gro off thruput: 9.16Gbps CPU utilization: 19% 1.2 ethtool -K gre1 gro on; ethtool -K eth0 gro off thruput: 5.9Gbps CPU utilization: 15% 1.3 ethtool -K gre1 gro off; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 12-13% 1.4 ethtool -K gre1 gro on; ethtool -K eth0 gro on thruput: 9.26Gbps CPU utilization: 10% The following tests were performed on a different NIC that is capable of csum offload. I.e., the h/w is capable of computing IP payload csum (CHECKSUM_COMPLETE). 2.1 ethtool -K gre1 gro on (hence will use gro_cells) 2.1.1 ethtool -K eth0 gro off; csum offload disabled thruput: 8.53Gbps CPU utilization: 9% 2.1.2 ethtool -K eth0 gro off; csum offload enabled thruput: 8.97Gbps CPU utilization: 7-8% 2.1.3 ethtool -K eth0 gro on; csum offload disabled thruput: 8.83Gbps CPU utilization: 5-6% 2.1.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.98Gbps CPU utilization: 5% 2.2 ethtool -K gre1 gro off 2.2.1 ethtool -K eth0 gro off; csum offload disabled thruput: 5.93Gbps CPU utilization: 9% 2.2.2 ethtool -K eth0 gro off; csum offload enabled thruput: 5.62Gbps CPU utilization: 8% 2.2.3 ethtool -K eth0 gro on; csum offload disabled thruput: 7.69Gbps CPU utilization: 8% 2.2.4 ethtool -K eth0 gro on; csum offload enabled thruput: 8.96Gbps CPU utilization: 5-6% Signed-off-by: H.K. Jerry Chu <hkchu@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-08 02:23:19 +08:00
NAPI_GRO_CB(skb)->flush |= (flush != 0);
return pp;
}
int tcp_gro_complete(struct sk_buff *skb)
{
struct tcphdr *th = tcp_hdr(skb);
skb->csum_start = (unsigned char *)th - skb->head;
skb->csum_offset = offsetof(struct tcphdr, check);
skb->ip_summed = CHECKSUM_PARTIAL;
skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
if (th->cwr)
skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
net, gro: Set inner transport header offset in tcp/udp GRO hook GSO expects inner transport header offset to be valid when skb->encapsulation flag is set. GSO uses this value to calculate the length of an individual segment of a GSO packet in skb_gso_transport_seglen(). However, tcp/udp gro_complete callbacks don't update the skb->inner_transport_header when processing an encapsulated TCP/UDP segment. As a result a GRO skb has ->inner_transport_header set to a value carried over from earlier skb processing. This can have mild to tragic consequences. From miscalculating the GSO segment length to triggering a page fault [1], when trying to read TCP/UDP header at an address past the skb->data page. The latter scenario leads to an oops report like so: BUG: unable to handle page fault for address: ffff9fa7ec00d008 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 123f201067 P4D 123f201067 PUD 123f209067 PMD 0 Oops: 0000 [#1] SMP NOPTI CPU: 44 PID: 0 Comm: swapper/44 Not tainted 5.4.53-cloudflare-2020.7.21 #1 Hardware name: HYVE EDGE-METAL-GEN10/HS-1811DLite1, BIOS V2.15 02/21/2020 RIP: 0010:skb_gso_transport_seglen+0x44/0xa0 Code: c0 41 83 e0 11 f6 87 81 00 00 00 20 74 30 0f b7 87 aa 00 00 00 0f [...] RSP: 0018:ffffad8640bacbb8 EFLAGS: 00010202 RAX: 000000000000feda RBX: ffff9fcc8d31bc00 RCX: ffff9fa7ec00cffc RDX: ffff9fa7ebffdec0 RSI: 000000000000feda RDI: 0000000000000122 RBP: 00000000000005c4 R08: 0000000000000001 R09: 0000000000000000 R10: ffff9fe588ae3800 R11: ffff9fe011fc92f0 R12: ffff9fcc8d31bc00 R13: ffff9fe0119d4300 R14: 00000000000005c4 R15: ffff9fba57d70900 FS: 0000000000000000(0000) GS:ffff9fe68df00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: ffff9fa7ec00d008 CR3: 0000003e99b1c000 CR4: 0000000000340ee0 Call Trace: <IRQ> skb_gso_validate_network_len+0x11/0x70 __ip_finish_output+0x109/0x1c0 ip_sublist_rcv_finish+0x57/0x70 ip_sublist_rcv+0x2aa/0x2d0 ? ip_rcv_finish_core.constprop.0+0x390/0x390 ip_list_rcv+0x12b/0x14f __netif_receive_skb_list_core+0x2a9/0x2d0 netif_receive_skb_list_internal+0x1b5/0x2e0 napi_complete_done+0x93/0x140 veth_poll+0xc0/0x19f [veth] ? mlx5e_napi_poll+0x221/0x610 [mlx5_core] net_rx_action+0x1f8/0x790 __do_softirq+0xe1/0x2bf irq_exit+0x8e/0xc0 do_IRQ+0x58/0xe0 common_interrupt+0xf/0xf </IRQ> The bug can be observed in a simple setup where we send IP/GRE/IP/TCP packets into a netns over a veth pair. Inside the netns, packets are forwarded to dummy device: trafgen -> [veth A]--[veth B] -forward-> [dummy] For veth B to GRO aggregate packets on receive, it needs to have an XDP program attached (for example, a trivial XDP_PASS). Additionally, for UDP, we need to enable GSO_UDP_L4 feature on the device: ip netns exec A ethtool -K AB rx-udp-gro-forwarding on The last component is an artificial delay to increase the chances of GRO batching happening: ip netns exec A tc qdisc add dev AB root \ netem delay 200us slot 5ms 10ms packets 2 bytes 64k With such a setup in place, the bug can be observed by tracing the skb outer and inner offsets when GSO skb is transmitted from the dummy device: tcp: FUNC DEV SKB_LEN NH TH ENC INH ITH GSO_SIZE GSO_TYPE ip_finish_output dumB 2830 270 290 1 294 254 1383 (tcpv4,gre,) ^^^ udp: FUNC DEV SKB_LEN NH TH ENC INH ITH GSO_SIZE GSO_TYPE ip_finish_output dumB 2818 270 290 1 294 254 1383 (gre,udp_l4,) ^^^ Fix it by updating the inner transport header offset in tcp/udp gro_complete callbacks, similar to how {inet,ipv6}_gro_complete callbacks update the inner network header offset, when skb->encapsulation flag is set. [1] https://lore.kernel.org/netdev/CAKxSbF01cLpZem2GFaUaifh0S-5WYViZemTicAg7FCHOnh6kug@mail.gmail.com/ Fixes: bf296b125b21 ("tcp: Add GRO support") Fixes: f993bc25e519 ("net: core: handle encapsulation offloads when computing segment lengths") Fixes: e20cf8d3f1f7 ("udp: implement GRO for plain UDP sockets.") Reported-by: Alex Forster <aforster@cloudflare.com> Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2021-07-29 21:48:20 +08:00
if (skb->encapsulation)
skb->inner_transport_header = skb->transport_header;
return 0;
}
EXPORT_SYMBOL(tcp_gro_complete);
INDIRECT_CALLABLE_SCOPE
struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb)
{
/* Don't bother verifying checksum if we're going to flush anyway. */
if (!NAPI_GRO_CB(skb)->flush &&
skb_gro_checksum_validate(skb, IPPROTO_TCP,
inet_gro_compute_pseudo)) {
NAPI_GRO_CB(skb)->flush = 1;
return NULL;
}
return tcp_gro_receive(head, skb);
}
INDIRECT_CALLABLE_SCOPE int tcp4_gro_complete(struct sk_buff *skb, int thoff)
{
const struct iphdr *iph = ip_hdr(skb);
struct tcphdr *th = tcp_hdr(skb);
th->check = ~tcp_v4_check(skb->len - thoff, iph->saddr,
iph->daddr, 0);
skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4;
if (NAPI_GRO_CB(skb)->is_atomic)
skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
return tcp_gro_complete(skb);
}
static const struct net_offload tcpv4_offload = {
.callbacks = {
.gso_segment = tcp4_gso_segment,
.gro_receive = tcp4_gro_receive,
.gro_complete = tcp4_gro_complete,
},
};
int __init tcpv4_offload_init(void)
{
return inet_add_offload(&tcpv4_offload, IPPROTO_TCP);
}