@@ -298,6 +298,7 @@ struct sock_common {
* @sk_filter: socket filtering instructions
* @sk_timer: sock cleanup timer
* @sk_stamp: time stamp of last packet received
+ * @sk_stamp_seq: lock for accessing sk_stamp
* @sk_tsflags: SO_TIMESTAMPING socket options
* @sk_tskey: counter to disambiguate concurrent tstamp requests
* @sk_zckey: counter to order MSG_ZEROCOPY notifications
@@ -474,6 +475,7 @@ struct sock {
const struct cred *sk_peer_cred;
long sk_rcvtimeo;
ktime_t sk_stamp;
+ seqlock_t sk_stamp_seq;
u16 sk_tsflags;
u8 sk_shutdown;
u32 sk_tskey;
@@ -2311,8 +2313,11 @@ sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
(hwtstamps->hwtstamp &&
(sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
__sock_recv_timestamp(msg, sk, skb);
- else
+ else {
+ write_seqlock(&sk->sk_stamp_seq);
sk->sk_stamp = kt;
+ write_sequnlock(&sk->sk_stamp_seq);
+ }
if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
__sock_recv_wifi_status(msg, sk, skb);
@@ -2332,10 +2337,15 @@ static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
__sock_recv_ts_and_drops(msg, sk, skb);
- else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
+ else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) {
+ write_seqlock(&sk->sk_stamp_seq);
sk->sk_stamp = skb->tstamp;
- else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
+ write_sequnlock(&sk->sk_stamp_seq);
+ } else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) {
+ write_seqlock(&sk->sk_stamp_seq);
sk->sk_stamp = 0;
+ write_sequnlock(&sk->sk_stamp_seq);
+ }
}
void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
@@ -459,6 +459,7 @@ int compat_sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
{
struct compat_timeval __user *ctv;
int err;
+ unsigned int seq;
struct timeval tv;
if (COMPAT_USE_64BIT_TIME)
@@ -467,12 +468,21 @@ int compat_sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
ctv = (struct compat_timeval __user *) userstamp;
err = -ENOENT;
sock_enable_timestamp(sk, SOCK_TIMESTAMP);
- tv = ktime_to_timeval(sk->sk_stamp);
+
+ do {
+ seq = read_seqbegin(&sk->sk_stamp_seq);
+ tv = ktime_to_timeval(sk->sk_stamp);
+ } while (read_seqretry(&sk->sk_stamp_seq, seq));
+
if (tv.tv_sec == -1)
return err;
if (tv.tv_sec == 0) {
- sk->sk_stamp = ktime_get_real();
- tv = ktime_to_timeval(sk->sk_stamp);
+ ktime_t kt = ktime_get_real();
+
+ write_seqlock(&sk->sk_stamp_seq);
+ sk->sk_stamp = kt;
+ write_sequnlock(&sk->sk_stamp_seq);
+ tv = ktime_to_timeval(kt);
}
err = 0;
if (put_user(tv.tv_sec, &ctv->tv_sec) ||
@@ -486,6 +496,7 @@ int compat_sock_get_timestampns(struct sock *sk, struct timespec __user *usersta
{
struct compat_timespec __user *ctv;
int err;
+ unsigned int seq;
struct timespec ts;
if (COMPAT_USE_64BIT_TIME)
@@ -494,12 +505,21 @@ int compat_sock_get_timestampns(struct sock *sk, struct timespec __user *usersta
ctv = (struct compat_timespec __user *) userstamp;
err = -ENOENT;
sock_enable_timestamp(sk, SOCK_TIMESTAMP);
- ts = ktime_to_timespec(sk->sk_stamp);
+
+ do {
+ seq = read_seqbegin(&sk->sk_stamp_seq);
+ ts = ktime_to_timespec(sk->sk_stamp);
+ } while (read_seqretry(&sk->sk_stamp_seq, seq));
+
if (ts.tv_sec == -1)
return err;
if (ts.tv_sec == 0) {
- sk->sk_stamp = ktime_get_real();
- ts = ktime_to_timespec(sk->sk_stamp);
+ ktime_t kt = ktime_get_real();
+
+ write_seqlock(&sk->sk_stamp_seq);
+ sk->sk_stamp = kt;
+ write_sequnlock(&sk->sk_stamp_seq);
+ ts = ktime_to_timespec(kt);
}
err = 0;
if (put_user(ts.tv_sec, &ctv->tv_sec) ||
@@ -2783,6 +2783,7 @@ void sock_init_data(struct socket *sock, struct sock *sk)
sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
sk->sk_stamp = SK_DEFAULT_STAMP;
+ seqlock_init(&sk->sk_stamp_seq);
atomic_set(&sk->sk_zckey, 0);
#ifdef CONFIG_NET_RX_BUSY_POLL
@@ -2880,15 +2881,27 @@ EXPORT_SYMBOL(lock_sock_fast);
int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
{
struct timeval tv;
+ unsigned int seq;
sock_enable_timestamp(sk, SOCK_TIMESTAMP);
- tv = ktime_to_timeval(sk->sk_stamp);
+
+ do {
+ seq = read_seqbegin(&sk->sk_stamp_seq);
+ tv = ktime_to_timeval(sk->sk_stamp);
+ } while (read_seqretry(&sk->sk_stamp_seq, seq));
+
if (tv.tv_sec == -1)
return -ENOENT;
if (tv.tv_sec == 0) {
- sk->sk_stamp = ktime_get_real();
- tv = ktime_to_timeval(sk->sk_stamp);
+ ktime_t kt = ktime_get_real();
+
+ write_seqlock(&sk->sk_stamp_seq);
+ sk->sk_stamp = kt;
+ write_sequnlock(&sk->sk_stamp_seq);
+
+ tv = ktime_to_timeval(kt);
}
+
return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
}
EXPORT_SYMBOL(sock_get_timestamp);
@@ -2896,13 +2909,24 @@ EXPORT_SYMBOL(sock_get_timestamp);
int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
{
struct timespec ts;
+ unsigned int seq;
sock_enable_timestamp(sk, SOCK_TIMESTAMP);
- ts = ktime_to_timespec(sk->sk_stamp);
+
+ do {
+ seq = read_seqbegin(&sk->sk_stamp_seq);
+ ts = ktime_to_timespec(sk->sk_stamp);
+ } while (read_seqretry(&sk->sk_stamp_seq, seq));
+
if (ts.tv_sec == -1)
return -ENOENT;
if (ts.tv_sec == 0) {
- sk->sk_stamp = ktime_get_real();
+ ktime_t kt = ktime_get_real();
+
+ write_seqlock(&sk->sk_stamp_seq);
+ sk->sk_stamp = kt;
+ write_sequnlock(&sk->sk_stamp_seq);
+
ts = ktime_to_timespec(sk->sk_stamp);
}
return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
@@ -549,7 +549,9 @@ static int svc_udp_recvfrom(struct svc_rqst *rqstp)
/* Don't enable netstamp, sunrpc doesn't
need that much accuracy */
}
+ write_seqlock(&svsk->sk_sk->sk_stamp_seq);
svsk->sk_sk->sk_stamp = skb->tstamp;
+ write_sequnlock(&svsk->sk_sk->sk_stamp_seq);
set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
len = skb->len;
Al Viro mentioned (Message-ID <20170626041334.GZ10672@ZenIV.linux.org.uk>) that there is probably a race condition lurking in accesses of sk_stamp on 32-bit machines. sock->sk_stamp is of type ktime_t which is always an s64. On a 32 bit architecture, we might run into situations of unsafe access as the access to the field becomes non atomic. Use seqlocks for synchronization. This allows us to avoid using spinlocks for readers as readers do not need mutual exclusion. Another approach to solve this is to require sk_lock for all modifications of the timestamps. The current approach allows for timestamps to have their own lock: sk_stamp_lock. This allows for the patch to not compete with already existing critical sections, and side effects are limited to the paths in the patch. The addition of the new field maintains the data locality optimizations from commit 9115e8cd2a0c ("net: reorganize struct sock for better data locality") Note that all the instances of the sk_stamp accesses are either through the ioctl or the syscall recvmsg. Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com> --- Changes since v1: * fixed sunrpc sk_stamp update include/net/sock.h | 16 +++++++++++++--- net/compat.c | 32 ++++++++++++++++++++++++++------ net/core/sock.c | 34 +++++++++++++++++++++++++++++----- net/sunrpc/svcsock.c | 2 ++ 4 files changed, 70 insertions(+), 14 deletions(-)