selftests/bpf: Implement sample tcp/tcp6 bpf_iter programs
authorYonghong Song <yhs@fb.com>
Tue, 23 Jun 2020 23:08:20 +0000 (16:08 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Thu, 25 Jun 2020 01:37:59 +0000 (18:37 -0700)
In my VM, I got identical result compared to /proc/net/{tcp,tcp6}.
For tcp6:
  $ cat /proc/net/tcp6
    sl  local_address                         remote_address                        st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode
     0: 00000000000000000000000000000000:0016 00000000000000000000000000000000:0000 0A 00000000:00000000 00:00000001 00000000     0        0 17955 1 000000003eb3102e 100 0 0 10 0

  $ cat /sys/fs/bpf/p1
    sl  local_address                         remote_address                        st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode
     0: 00000000000000000000000000000000:0016 00000000000000000000000000000000:0000 0A 00000000:00000000 00:00000000 00000000     0        0 17955 1 000000003eb3102e 100 0 0 10 0

For tcp:
  $ cat /proc/net/tcp
  sl  local_address rem_address   st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode
   0: 00000000:0016 00000000:0000 0A 00000000:00000000 00:00000000 00000000     0        0 2666 1 000000007152e43f 100 0 0 10 0
  $ cat /sys/fs/bpf/p2
  sl  local_address                         remote_address                        st tx_queue rx_queue tr tm->when retrnsmt   uid  timeout inode
   1: 00000000:0016 00000000:0000 0A 00000000:00000000 00:00000000 00000000     0        0 2666 1 000000007152e43f 100 0 0 10 0

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200623230820.3989165-1-yhs@fb.com
tools/testing/selftests/bpf/progs/bpf_iter.h
tools/testing/selftests/bpf/progs/bpf_iter_tcp4.c [new file with mode: 0644]
tools/testing/selftests/bpf/progs/bpf_iter_tcp6.c [new file with mode: 0644]

index 3757e88..bde23e1 100644 (file)
@@ -7,6 +7,8 @@
 #define bpf_iter__netlink bpf_iter__netlink___not_used
 #define bpf_iter__task bpf_iter__task___not_used
 #define bpf_iter__task_file bpf_iter__task_file___not_used
+#define bpf_iter__tcp bpf_iter__tcp___not_used
+#define tcp6_sock tcp6_sock___not_used
 #include "vmlinux.h"
 #undef bpf_iter_meta
 #undef bpf_iter__bpf_map
@@ -14,6 +16,8 @@
 #undef bpf_iter__netlink
 #undef bpf_iter__task
 #undef bpf_iter__task_file
+#undef bpf_iter__tcp
+#undef tcp6_sock
 
 struct bpf_iter_meta {
        struct seq_file *seq;
@@ -47,3 +51,14 @@ struct bpf_iter__bpf_map {
        struct bpf_iter_meta *meta;
        struct bpf_map *map;
 } __attribute__((preserve_access_index));
+
+struct bpf_iter__tcp {
+       struct bpf_iter_meta *meta;
+       struct sock_common *sk_common;
+       uid_t uid;
+} __attribute__((preserve_access_index));
+
+struct tcp6_sock {
+       struct tcp_sock tcp;
+       struct ipv6_pinfo inet6;
+} __attribute__((preserve_access_index));
diff --git a/tools/testing/selftests/bpf/progs/bpf_iter_tcp4.c b/tools/testing/selftests/bpf/progs/bpf_iter_tcp4.c
new file mode 100644 (file)
index 0000000..30fd587
--- /dev/null
@@ -0,0 +1,234 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2020 Facebook */
+#include "bpf_iter.h"
+#include "bpf_tracing_net.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include <bpf/bpf_endian.h>
+
+char _license[] SEC("license") = "GPL";
+
+static int hlist_unhashed_lockless(const struct hlist_node *h)
+{
+        return !(h->pprev);
+}
+
+static int timer_pending(const struct timer_list * timer)
+{
+       return !hlist_unhashed_lockless(&timer->entry);
+}
+
+extern unsigned CONFIG_HZ __kconfig;
+
+#define USER_HZ                100
+#define NSEC_PER_SEC   1000000000ULL
+static clock_t jiffies_to_clock_t(unsigned long x)
+{
+       /* The implementation here tailored to a particular
+        * setting of USER_HZ.
+        */
+       u64 tick_nsec = (NSEC_PER_SEC + CONFIG_HZ/2) / CONFIG_HZ;
+       u64 user_hz_nsec = NSEC_PER_SEC / USER_HZ;
+
+       if ((tick_nsec % user_hz_nsec) == 0) {
+               if (CONFIG_HZ < USER_HZ)
+                       return x * (USER_HZ / CONFIG_HZ);
+               else
+                       return x / (CONFIG_HZ / USER_HZ);
+       }
+       return x * tick_nsec/user_hz_nsec;
+}
+
+static clock_t jiffies_delta_to_clock_t(long delta)
+{
+       if (delta <= 0)
+               return 0;
+
+       return jiffies_to_clock_t(delta);
+}
+
+static long sock_i_ino(const struct sock *sk)
+{
+       const struct socket *sk_socket = sk->sk_socket;
+       const struct inode *inode;
+       unsigned long ino;
+
+       if (!sk_socket)
+               return 0;
+
+       inode = &container_of(sk_socket, struct socket_alloc, socket)->vfs_inode;
+       bpf_probe_read(&ino, sizeof(ino), &inode->i_ino);
+       return ino;
+}
+
+static bool
+inet_csk_in_pingpong_mode(const struct inet_connection_sock *icsk)
+{
+       return icsk->icsk_ack.pingpong >= TCP_PINGPONG_THRESH;
+}
+
+static bool tcp_in_initial_slowstart(const struct tcp_sock *tcp)
+{
+       return tcp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
+}
+
+static int dump_tcp_sock(struct seq_file *seq, struct tcp_sock *tp,
+                        uid_t uid, __u32 seq_num)
+{
+       const struct inet_connection_sock *icsk;
+       const struct fastopen_queue *fastopenq;
+       const struct inet_sock *inet;
+       unsigned long timer_expires;
+       const struct sock *sp;
+       __u16 destp, srcp;
+       __be32 dest, src;
+       int timer_active;
+       int rx_queue;
+       int state;
+
+       icsk = &tp->inet_conn;
+       inet = &icsk->icsk_inet;
+       sp = &inet->sk;
+       fastopenq = &icsk->icsk_accept_queue.fastopenq;
+
+       dest = inet->inet_daddr;
+       src = inet->inet_rcv_saddr;
+       destp = bpf_ntohs(inet->inet_dport);
+       srcp = bpf_ntohs(inet->inet_sport);
+
+       if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
+           icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
+           icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
+               timer_active = 1;
+               timer_expires = icsk->icsk_timeout;
+       } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
+               timer_active = 4;
+               timer_expires = icsk->icsk_timeout;
+       } else if (timer_pending(&sp->sk_timer)) {
+               timer_active = 2;
+               timer_expires = sp->sk_timer.expires;
+       } else {
+               timer_active = 0;
+               timer_expires = bpf_jiffies64();
+       }
+
+       state = sp->sk_state;
+       if (state == TCP_LISTEN) {
+               rx_queue = sp->sk_ack_backlog;
+       } else {
+               rx_queue = tp->rcv_nxt - tp->copied_seq;
+               if (rx_queue < 0)
+                       rx_queue = 0;
+       }
+
+       BPF_SEQ_PRINTF(seq, "%4d: %08X:%04X %08X:%04X ",
+                      seq_num, src, srcp, destp, destp);
+       BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d ",
+                      state,
+                      tp->write_seq - tp->snd_una, rx_queue,
+                      timer_active,
+                      jiffies_delta_to_clock_t(timer_expires - bpf_jiffies64()),
+                      icsk->icsk_retransmits, uid,
+                      icsk->icsk_probes_out,
+                      sock_i_ino(sp),
+                      sp->sk_refcnt.refs.counter);
+       BPF_SEQ_PRINTF(seq, "%pK %lu %lu %u %u %d\n",
+                      tp,
+                      jiffies_to_clock_t(icsk->icsk_rto),
+                      jiffies_to_clock_t(icsk->icsk_ack.ato),
+                      (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(icsk),
+                      tp->snd_cwnd,
+                      state == TCP_LISTEN ? fastopenq->max_qlen
+                               : (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
+                     );
+
+       return 0;
+}
+
+static int dump_tw_sock(struct seq_file *seq, struct tcp_timewait_sock *ttw,
+                       uid_t uid, __u32 seq_num)
+{
+       struct inet_timewait_sock *tw = &ttw->tw_sk;
+       __u16 destp, srcp;
+       __be32 dest, src;
+       long delta;
+
+       delta = tw->tw_timer.expires - bpf_jiffies64();
+       dest = tw->tw_daddr;
+       src  = tw->tw_rcv_saddr;
+       destp = bpf_ntohs(tw->tw_dport);
+       srcp  = bpf_ntohs(tw->tw_sport);
+
+       BPF_SEQ_PRINTF(seq, "%4d: %08X:%04X %08X:%04X ",
+                      seq_num, src, srcp, dest, destp);
+
+       BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
+                      tw->tw_substate, 0, 0,
+                      3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
+                      tw->tw_refcnt.refs.counter, tw);
+
+       return 0;
+}
+
+static int dump_req_sock(struct seq_file *seq, struct tcp_request_sock *treq,
+                        uid_t uid, __u32 seq_num)
+{
+       struct inet_request_sock *irsk = &treq->req;
+       struct request_sock *req = &irsk->req;
+       long ttd;
+
+       ttd = req->rsk_timer.expires - bpf_jiffies64();
+
+       if (ttd < 0)
+               ttd = 0;
+
+       BPF_SEQ_PRINTF(seq, "%4d: %08X:%04X %08X:%04X ",
+                      seq_num, irsk->ir_loc_addr,
+                      irsk->ir_num, irsk->ir_rmt_addr,
+                      bpf_ntohs(irsk->ir_rmt_port));
+       BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
+                      TCP_SYN_RECV, 0, 0, 1, jiffies_to_clock_t(ttd),
+                      req->num_timeout, uid, 0, 0, 0, req);
+
+       return 0;
+}
+
+SEC("iter/tcp")
+int dump_tcp4(struct bpf_iter__tcp *ctx)
+{
+       struct sock_common *sk_common = ctx->sk_common;
+       struct seq_file *seq = ctx->meta->seq;
+       struct tcp_timewait_sock *tw;
+       struct tcp_request_sock *req;
+       struct tcp_sock *tp;
+       uid_t uid = ctx->uid;
+       __u32 seq_num;
+
+       if (sk_common == (void *)0)
+               return 0;
+
+       seq_num = ctx->meta->seq_num;
+       if (seq_num == 0)
+               BPF_SEQ_PRINTF(seq, "  sl  "
+                                   "local_address "
+                                   "rem_address   "
+                                   "st tx_queue rx_queue tr tm->when retrnsmt"
+                                   "   uid  timeout inode\n");
+
+       if (sk_common->skc_family != AF_INET)
+               return 0;
+
+       tp = bpf_skc_to_tcp_sock(sk_common);
+       if (tp)
+               return dump_tcp_sock(seq, tp, uid, seq_num);
+
+       tw = bpf_skc_to_tcp_timewait_sock(sk_common);
+       if (tw)
+               return dump_tw_sock(seq, tw, uid, seq_num);
+
+       req = bpf_skc_to_tcp_request_sock(sk_common);
+       if (req)
+               return dump_req_sock(seq, req, uid, seq_num);
+
+       return 0;
+}
diff --git a/tools/testing/selftests/bpf/progs/bpf_iter_tcp6.c b/tools/testing/selftests/bpf/progs/bpf_iter_tcp6.c
new file mode 100644 (file)
index 0000000..10dec43
--- /dev/null
@@ -0,0 +1,250 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2020 Facebook */
+#include "bpf_iter.h"
+#include "bpf_tracing_net.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include <bpf/bpf_endian.h>
+
+char _license[] SEC("license") = "GPL";
+
+static int hlist_unhashed_lockless(const struct hlist_node *h)
+{
+        return !(h->pprev);
+}
+
+static int timer_pending(const struct timer_list * timer)
+{
+       return !hlist_unhashed_lockless(&timer->entry);
+}
+
+extern unsigned CONFIG_HZ __kconfig;
+
+#define USER_HZ                100
+#define NSEC_PER_SEC   1000000000ULL
+static clock_t jiffies_to_clock_t(unsigned long x)
+{
+       /* The implementation here tailored to a particular
+        * setting of USER_HZ.
+        */
+       u64 tick_nsec = (NSEC_PER_SEC + CONFIG_HZ/2) / CONFIG_HZ;
+       u64 user_hz_nsec = NSEC_PER_SEC / USER_HZ;
+
+       if ((tick_nsec % user_hz_nsec) == 0) {
+               if (CONFIG_HZ < USER_HZ)
+                       return x * (USER_HZ / CONFIG_HZ);
+               else
+                       return x / (CONFIG_HZ / USER_HZ);
+       }
+       return x * tick_nsec/user_hz_nsec;
+}
+
+static clock_t jiffies_delta_to_clock_t(long delta)
+{
+       if (delta <= 0)
+               return 0;
+
+       return jiffies_to_clock_t(delta);
+}
+
+static long sock_i_ino(const struct sock *sk)
+{
+       const struct socket *sk_socket = sk->sk_socket;
+       const struct inode *inode;
+       unsigned long ino;
+
+       if (!sk_socket)
+               return 0;
+
+       inode = &container_of(sk_socket, struct socket_alloc, socket)->vfs_inode;
+       bpf_probe_read(&ino, sizeof(ino), &inode->i_ino);
+       return ino;
+}
+
+static bool
+inet_csk_in_pingpong_mode(const struct inet_connection_sock *icsk)
+{
+       return icsk->icsk_ack.pingpong >= TCP_PINGPONG_THRESH;
+}
+
+static bool tcp_in_initial_slowstart(const struct tcp_sock *tcp)
+{
+       return tcp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
+}
+
+static int dump_tcp6_sock(struct seq_file *seq, struct tcp6_sock *tp,
+                        uid_t uid, __u32 seq_num)
+{
+       const struct inet_connection_sock *icsk;
+       const struct fastopen_queue *fastopenq;
+       const struct in6_addr *dest, *src;
+       const struct inet_sock *inet;
+       unsigned long timer_expires;
+       const struct sock *sp;
+       __u16 destp, srcp;
+       int timer_active;
+       int rx_queue;
+       int state;
+
+       icsk = &tp->tcp.inet_conn;
+       inet = &icsk->icsk_inet;
+       sp = &inet->sk;
+       fastopenq = &icsk->icsk_accept_queue.fastopenq;
+
+       dest = &sp->sk_v6_daddr;
+       src = &sp->sk_v6_rcv_saddr;
+       destp = bpf_ntohs(inet->inet_dport);
+       srcp = bpf_ntohs(inet->inet_sport);
+
+       if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
+           icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
+           icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
+               timer_active = 1;
+               timer_expires = icsk->icsk_timeout;
+       } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
+               timer_active = 4;
+               timer_expires = icsk->icsk_timeout;
+       } else if (timer_pending(&sp->sk_timer)) {
+               timer_active = 2;
+               timer_expires = sp->sk_timer.expires;
+       } else {
+               timer_active = 0;
+               timer_expires = bpf_jiffies64();
+       }
+
+       state = sp->sk_state;
+       if (state == TCP_LISTEN) {
+               rx_queue = sp->sk_ack_backlog;
+       } else {
+               rx_queue = tp->tcp.rcv_nxt - tp->tcp.copied_seq;
+               if (rx_queue < 0)
+                       rx_queue = 0;
+       }
+
+       BPF_SEQ_PRINTF(seq, "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X ",
+                      seq_num,
+                      src->s6_addr32[0], src->s6_addr32[1],
+                      src->s6_addr32[2], src->s6_addr32[3], srcp,
+                      dest->s6_addr32[0], dest->s6_addr32[1],
+                      dest->s6_addr32[2], dest->s6_addr32[3], destp);
+       BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d ",
+                      state,
+                      tp->tcp.write_seq - tp->tcp.snd_una, rx_queue,
+                      timer_active,
+                      jiffies_delta_to_clock_t(timer_expires - bpf_jiffies64()),
+                      icsk->icsk_retransmits, uid,
+                      icsk->icsk_probes_out,
+                      sock_i_ino(sp),
+                      sp->sk_refcnt.refs.counter);
+       BPF_SEQ_PRINTF(seq, "%pK %lu %lu %u %u %d\n",
+                      tp,
+                      jiffies_to_clock_t(icsk->icsk_rto),
+                      jiffies_to_clock_t(icsk->icsk_ack.ato),
+                      (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(icsk),
+                      tp->tcp.snd_cwnd,
+                      state == TCP_LISTEN ? fastopenq->max_qlen
+                               : (tcp_in_initial_slowstart(&tp->tcp) ? -1
+                                                                     : tp->tcp.snd_ssthresh)
+                     );
+
+       return 0;
+}
+
+static int dump_tw_sock(struct seq_file *seq, struct tcp_timewait_sock *ttw,
+                       uid_t uid, __u32 seq_num)
+{
+       struct inet_timewait_sock *tw = &ttw->tw_sk;
+       const struct in6_addr *dest, *src;
+       __u16 destp, srcp;
+       long delta;
+
+       delta = tw->tw_timer.expires - bpf_jiffies64();
+       dest = &tw->tw_v6_daddr;
+       src  = &tw->tw_v6_rcv_saddr;
+       destp = bpf_ntohs(tw->tw_dport);
+       srcp  = bpf_ntohs(tw->tw_sport);
+
+       BPF_SEQ_PRINTF(seq, "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X ",
+                      seq_num,
+                      src->s6_addr32[0], src->s6_addr32[1],
+                      src->s6_addr32[2], src->s6_addr32[3], srcp,
+                      dest->s6_addr32[0], dest->s6_addr32[1],
+                      dest->s6_addr32[2], dest->s6_addr32[3], destp);
+
+       BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
+                      tw->tw_substate, 0, 0,
+                      3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
+                      tw->tw_refcnt.refs.counter, tw);
+
+       return 0;
+}
+
+static int dump_req_sock(struct seq_file *seq, struct tcp_request_sock *treq,
+                        uid_t uid, __u32 seq_num)
+{
+       struct inet_request_sock *irsk = &treq->req;
+       struct request_sock *req = &irsk->req;
+       struct in6_addr *src, *dest;
+       long ttd;
+
+       ttd = req->rsk_timer.expires - bpf_jiffies64();
+       src = &irsk->ir_v6_loc_addr;
+       dest = &irsk->ir_v6_rmt_addr;
+
+       if (ttd < 0)
+               ttd = 0;
+
+       BPF_SEQ_PRINTF(seq, "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X ",
+                      seq_num,
+                      src->s6_addr32[0], src->s6_addr32[1],
+                      src->s6_addr32[2], src->s6_addr32[3],
+                      irsk->ir_num,
+                      dest->s6_addr32[0], dest->s6_addr32[1],
+                      dest->s6_addr32[2], dest->s6_addr32[3],
+                      bpf_ntohs(irsk->ir_rmt_port));
+       BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
+                      TCP_SYN_RECV, 0, 0, 1, jiffies_to_clock_t(ttd),
+                      req->num_timeout, uid, 0, 0, 0, req);
+
+       return 0;
+}
+
+SEC("iter/tcp")
+int dump_tcp6(struct bpf_iter__tcp *ctx)
+{
+       struct sock_common *sk_common = ctx->sk_common;
+       struct seq_file *seq = ctx->meta->seq;
+       struct tcp_timewait_sock *tw;
+       struct tcp_request_sock *req;
+       struct tcp6_sock *tp;
+       uid_t uid = ctx->uid;
+       __u32 seq_num;
+
+       if (sk_common == (void *)0)
+               return 0;
+
+       seq_num = ctx->meta->seq_num;
+       if (seq_num == 0)
+               BPF_SEQ_PRINTF(seq, "  sl  "
+                                   "local_address                         "
+                                   "remote_address                        "
+                                   "st tx_queue rx_queue tr tm->when retrnsmt"
+                                   "   uid  timeout inode\n");
+
+       if (sk_common->skc_family != AF_INET6)
+               return 0;
+
+       tp = bpf_skc_to_tcp6_sock(sk_common);
+       if (tp)
+               return dump_tcp6_sock(seq, tp, uid, seq_num);
+
+       tw = bpf_skc_to_tcp_timewait_sock(sk_common);
+       if (tw)
+               return dump_tw_sock(seq, tw, uid, seq_num);
+
+       req = bpf_skc_to_tcp_request_sock(sk_common);
+       if (req)
+               return dump_req_sock(seq, req, uid, seq_num);
+
+       return 0;
+}