Skip to content

Commit 7eeba17

Browse files
arjunroykuba-moo
arjunroy
authored and
kuba-moo
committed
tcp: Add receive timestamp support for receive zerocopy.
tcp_recvmsg() uses the CMSG mechanism to receive control information like packet receive timestamps. This patch adds CMSG fields to struct tcp_zerocopy_receive, and provides receive timestamps if available to the user. Signed-off-by: Arjun Roy <[email protected]> Signed-off-by: Jakub Kicinski <[email protected]>
1 parent 925bba2 commit 7eeba17

File tree

2 files changed

+88
-32
lines changed

2 files changed

+88
-32
lines changed

‎include/uapi/linux/tcp.h‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -354,5 +354,9 @@ struct tcp_zerocopy_receive {
354354
__u64 copybuf_address; /* in: copybuf address (small reads) */
355355
__s32 copybuf_len; /* in/out: copybuf bytes avail/used or error */
356356
__u32 flags; /* in: flags */
357+
__u64 msg_control; /* ancillary data */
358+
__u64 msg_controllen;
359+
__u32 msg_flags;
360+
/* __u32 hole; Next we must add >1 u32 otherwise length checks fail. */
357361
};
358362
#endif /* _UAPI_LINUX_TCP_H */

‎net/ipv4/tcp.c‎

Lines changed: 84 additions & 32 deletions
Original file line numberDiff line numberDiff line change
@@ -1745,6 +1745,20 @@ int tcp_set_rcvlowat(struct sock *sk, int val)
17451745
}
17461746
EXPORT_SYMBOL(tcp_set_rcvlowat);
17471747

1748+
static void tcp_update_recv_tstamps(struct sk_buff *skb,
1749+
struct scm_timestamping_internal *tss)
1750+
{
1751+
if (skb->tstamp)
1752+
tss->ts[0] = ktime_to_timespec64(skb->tstamp);
1753+
else
1754+
tss->ts[0] = (struct timespec64) {0};
1755+
1756+
if (skb_hwtstamps(skb)->hwtstamp)
1757+
tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp);
1758+
else
1759+
tss->ts[2] = (struct timespec64) {0};
1760+
}
1761+
17481762
#ifdef CONFIG_MMU
17491763
static const struct vm_operations_struct tcp_vm_ops = {
17501764
};
@@ -1848,13 +1862,13 @@ static int tcp_recvmsg_locked(struct sock *sk, struct msghdr *msg, size_t len,
18481862
struct scm_timestamping_internal *tss,
18491863
int *cmsg_flags);
18501864
static int receive_fallback_to_copy(struct sock *sk,
1851-
struct tcp_zerocopy_receive *zc, int inq)
1865+
struct tcp_zerocopy_receive *zc, int inq,
1866+
struct scm_timestamping_internal *tss)
18521867
{
18531868
unsigned long copy_address = (unsigned long)zc->copybuf_address;
1854-
struct scm_timestamping_internal tss_unused;
1855-
int err, cmsg_flags_unused;
18561869
struct msghdr msg = {};
18571870
struct iovec iov;
1871+
int err;
18581872

18591873
zc->length = 0;
18601874
zc->recv_skip_hint = 0;
@@ -1868,7 +1882,7 @@ static int receive_fallback_to_copy(struct sock *sk,
18681882
return err;
18691883

18701884
err = tcp_recvmsg_locked(sk, &msg, inq, /*nonblock=*/1, /*flags=*/0,
1871-
&tss_unused, &cmsg_flags_unused);
1885+
tss, &zc->msg_flags);
18721886
if (err < 0)
18731887
return err;
18741888

@@ -1909,21 +1923,27 @@ static int tcp_copy_straggler_data(struct tcp_zerocopy_receive *zc,
19091923
return (__s32)copylen;
19101924
}
19111925

1912-
static int tcp_zerocopy_handle_leftover_data(struct tcp_zerocopy_receive *zc,
1913-
struct sock *sk,
1914-
struct sk_buff *skb,
1915-
u32 *seq,
1916-
s32 copybuf_len)
1926+
static int tcp_zc_handle_leftover(struct tcp_zerocopy_receive *zc,
1927+
struct sock *sk,
1928+
struct sk_buff *skb,
1929+
u32 *seq,
1930+
s32 copybuf_len,
1931+
struct scm_timestamping_internal *tss)
19171932
{
19181933
u32 offset, copylen = min_t(u32, copybuf_len, zc->recv_skip_hint);
19191934

19201935
if (!copylen)
19211936
return 0;
19221937
/* skb is null if inq < PAGE_SIZE. */
1923-
if (skb)
1938+
if (skb) {
19241939
offset = *seq - TCP_SKB_CB(skb)->seq;
1925-
else
1940+
} else {
19261941
skb = tcp_recv_skb(sk, *seq, &offset);
1942+
if (TCP_SKB_CB(skb)->has_rxtstamp) {
1943+
tcp_update_recv_tstamps(skb, tss);
1944+
zc->msg_flags |= TCP_CMSG_TS;
1945+
}
1946+
}
19271947

19281948
zc->copybuf_len = tcp_copy_straggler_data(zc, skb, copylen, &offset,
19291949
seq);
@@ -2010,9 +2030,37 @@ static int tcp_zerocopy_vm_insert_batch(struct vm_area_struct *vma,
20102030
err);
20112031
}
20122032

2033+
static void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
2034+
struct scm_timestamping_internal *tss);
2035+
static void tcp_zc_finalize_rx_tstamp(struct sock *sk,
2036+
struct tcp_zerocopy_receive *zc,
2037+
struct scm_timestamping_internal *tss)
2038+
{
2039+
unsigned long msg_control_addr;
2040+
struct msghdr cmsg_dummy;
2041+
2042+
msg_control_addr = (unsigned long)zc->msg_control;
2043+
cmsg_dummy.msg_control = (void *)msg_control_addr;
2044+
cmsg_dummy.msg_controllen =
2045+
(__kernel_size_t)zc->msg_controllen;
2046+
cmsg_dummy.msg_flags = in_compat_syscall()
2047+
? MSG_CMSG_COMPAT : 0;
2048+
zc->msg_flags = 0;
2049+
if (zc->msg_control == msg_control_addr &&
2050+
zc->msg_controllen == cmsg_dummy.msg_controllen) {
2051+
tcp_recv_timestamp(&cmsg_dummy, sk, tss);
2052+
zc->msg_control = (__u64)
2053+
((uintptr_t)cmsg_dummy.msg_control);
2054+
zc->msg_controllen =
2055+
(__u64)cmsg_dummy.msg_controllen;
2056+
zc->msg_flags = (__u32)cmsg_dummy.msg_flags;
2057+
}
2058+
}
2059+
20132060
#define TCP_ZEROCOPY_PAGE_BATCH_SIZE 32
20142061
static int tcp_zerocopy_receive(struct sock *sk,
2015-
struct tcp_zerocopy_receive *zc)
2062+
struct tcp_zerocopy_receive *zc,
2063+
struct scm_timestamping_internal *tss)
20162064
{
20172065
u32 length = 0, offset, vma_len, avail_len, copylen = 0;
20182066
unsigned long address = (unsigned long)zc->address;
@@ -2029,6 +2077,7 @@ static int tcp_zerocopy_receive(struct sock *sk,
20292077
int ret;
20302078

20312079
zc->copybuf_len = 0;
2080+
zc->msg_flags = 0;
20322081

20332082
if (address & (PAGE_SIZE - 1) || address != zc->address)
20342083
return -EINVAL;
@@ -2039,7 +2088,7 @@ static int tcp_zerocopy_receive(struct sock *sk,
20392088
sock_rps_record_flow(sk);
20402089

20412090
if (inq && inq <= copybuf_len)
2042-
return receive_fallback_to_copy(sk, zc, inq);
2091+
return receive_fallback_to_copy(sk, zc, inq, tss);
20432092

20442093
if (inq < PAGE_SIZE) {
20452094
zc->length = 0;
@@ -2084,6 +2133,11 @@ static int tcp_zerocopy_receive(struct sock *sk,
20842133
} else {
20852134
skb = tcp_recv_skb(sk, seq, &offset);
20862135
}
2136+
2137+
if (TCP_SKB_CB(skb)->has_rxtstamp) {
2138+
tcp_update_recv_tstamps(skb, tss);
2139+
zc->msg_flags |= TCP_CMSG_TS;
2140+
}
20872141
zc->recv_skip_hint = skb->len - offset;
20882142
frags = skb_advance_to_frag(skb, offset, &offset_frag);
20892143
if (!frags || offset_frag)
@@ -2126,8 +2180,7 @@ static int tcp_zerocopy_receive(struct sock *sk,
21262180
mmap_read_unlock(current->mm);
21272181
/* Try to copy straggler data. */
21282182
if (!ret)
2129-
copylen = tcp_zerocopy_handle_leftover_data(zc, sk, skb, &seq,
2130-
copybuf_len);
2183+
copylen = tcp_zc_handle_leftover(zc, sk, skb, &seq, copybuf_len, tss);
21312184

21322185
if (length + copylen) {
21332186
WRITE_ONCE(tp->copied_seq, seq);
@@ -2148,20 +2201,6 @@ static int tcp_zerocopy_receive(struct sock *sk,
21482201
}
21492202
#endif
21502203

2151-
static void tcp_update_recv_tstamps(struct sk_buff *skb,
2152-
struct scm_timestamping_internal *tss)
2153-
{
2154-
if (skb->tstamp)
2155-
tss->ts[0] = ktime_to_timespec64(skb->tstamp);
2156-
else
2157-
tss->ts[0] = (struct timespec64) {0};
2158-
2159-
if (skb_hwtstamps(skb)->hwtstamp)
2160-
tss->ts[2] = ktime_to_timespec64(skb_hwtstamps(skb)->hwtstamp);
2161-
else
2162-
tss->ts[2] = (struct timespec64) {0};
2163-
}
2164-
21652204
/* Similar to __sock_recv_timestamp, but does not require an skb */
21662205
static void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
21672206
struct scm_timestamping_internal *tss)
@@ -4105,6 +4144,7 @@ static int do_tcp_getsockopt(struct sock *sk, int level,
41054144
}
41064145
#ifdef CONFIG_MMU
41074146
case TCP_ZEROCOPY_RECEIVE: {
4147+
struct scm_timestamping_internal tss;
41084148
struct tcp_zerocopy_receive zc = {};
41094149
int err;
41104150

@@ -4120,11 +4160,18 @@ static int do_tcp_getsockopt(struct sock *sk, int level,
41204160
if (copy_from_user(&zc, optval, len))
41214161
return -EFAULT;
41224162
lock_sock(sk);
4123-
err = tcp_zerocopy_receive(sk, &zc);
4163+
err = tcp_zerocopy_receive(sk, &zc, &tss);
41244164
release_sock(sk);
4125-
if (len >= offsetofend(struct tcp_zerocopy_receive, err))
4126-
goto zerocopy_rcv_sk_err;
4165+
if (len >= offsetofend(struct tcp_zerocopy_receive, msg_flags))
4166+
goto zerocopy_rcv_cmsg;
41274167
switch (len) {
4168+
case offsetofend(struct tcp_zerocopy_receive, msg_flags):
4169+
goto zerocopy_rcv_cmsg;
4170+
case offsetofend(struct tcp_zerocopy_receive, msg_controllen):
4171+
case offsetofend(struct tcp_zerocopy_receive, msg_control):
4172+
case offsetofend(struct tcp_zerocopy_receive, flags):
4173+
case offsetofend(struct tcp_zerocopy_receive, copybuf_len):
4174+
case offsetofend(struct tcp_zerocopy_receive, copybuf_address):
41284175
case offsetofend(struct tcp_zerocopy_receive, err):
41294176
goto zerocopy_rcv_sk_err;
41304177
case offsetofend(struct tcp_zerocopy_receive, inq):
@@ -4133,6 +4180,11 @@ static int do_tcp_getsockopt(struct sock *sk, int level,
41334180
default:
41344181
goto zerocopy_rcv_out;
41354182
}
4183+
zerocopy_rcv_cmsg:
4184+
if (zc.msg_flags & TCP_CMSG_TS)
4185+
tcp_zc_finalize_rx_tstamp(sk, &zc, &tss);
4186+
else
4187+
zc.msg_flags = 0;
41364188
zerocopy_rcv_sk_err:
41374189
if (!err)
41384190
zc.err = sock_error(sk);

0 commit comments

Comments
 (0)