1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2020 Facebook */
3 #include "bpf_iter.h"
4 #include "bpf_tracing_net.h"
5 #include <bpf/bpf_helpers.h>
6 #include <bpf/bpf_tracing.h>
7 #include <bpf/bpf_endian.h>
8
9 char _license[] SEC("license") = "GPL";
10
hlist_unhashed_lockless(const struct hlist_node * h)11 static int hlist_unhashed_lockless(const struct hlist_node *h)
12 {
13 return !(h->pprev);
14 }
15
timer_pending(const struct timer_list * timer)16 static int timer_pending(const struct timer_list * timer)
17 {
18 return !hlist_unhashed_lockless(&timer->entry);
19 }
20
21 extern unsigned CONFIG_HZ __kconfig;
22
23 #define USER_HZ 100
24 #define NSEC_PER_SEC 1000000000ULL
jiffies_to_clock_t(unsigned long x)25 static clock_t jiffies_to_clock_t(unsigned long x)
26 {
27 /* The implementation here tailored to a particular
28 * setting of USER_HZ.
29 */
30 u64 tick_nsec = (NSEC_PER_SEC + CONFIG_HZ/2) / CONFIG_HZ;
31 u64 user_hz_nsec = NSEC_PER_SEC / USER_HZ;
32
33 if ((tick_nsec % user_hz_nsec) == 0) {
34 if (CONFIG_HZ < USER_HZ)
35 return x * (USER_HZ / CONFIG_HZ);
36 else
37 return x / (CONFIG_HZ / USER_HZ);
38 }
39 return x * tick_nsec/user_hz_nsec;
40 }
41
jiffies_delta_to_clock_t(long delta)42 static clock_t jiffies_delta_to_clock_t(long delta)
43 {
44 if (delta <= 0)
45 return 0;
46
47 return jiffies_to_clock_t(delta);
48 }
49
sock_i_ino(const struct sock * sk)50 static long sock_i_ino(const struct sock *sk)
51 {
52 const struct socket *sk_socket = sk->sk_socket;
53 const struct inode *inode;
54 unsigned long ino;
55
56 if (!sk_socket)
57 return 0;
58
59 inode = &container_of(sk_socket, struct socket_alloc, socket)->vfs_inode;
60 bpf_probe_read_kernel(&ino, sizeof(ino), &inode->i_ino);
61 return ino;
62 }
63
64 static bool
inet_csk_in_pingpong_mode(const struct inet_connection_sock * icsk)65 inet_csk_in_pingpong_mode(const struct inet_connection_sock *icsk)
66 {
67 return icsk->icsk_ack.pingpong >= TCP_PINGPONG_THRESH;
68 }
69
tcp_in_initial_slowstart(const struct tcp_sock * tcp)70 static bool tcp_in_initial_slowstart(const struct tcp_sock *tcp)
71 {
72 return tcp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
73 }
74
dump_tcp_sock(struct seq_file * seq,struct tcp_sock * tp,uid_t uid,__u32 seq_num)75 static int dump_tcp_sock(struct seq_file *seq, struct tcp_sock *tp,
76 uid_t uid, __u32 seq_num)
77 {
78 const struct inet_connection_sock *icsk;
79 const struct fastopen_queue *fastopenq;
80 const struct inet_sock *inet;
81 unsigned long timer_expires;
82 const struct sock *sp;
83 __u16 destp, srcp;
84 __be32 dest, src;
85 int timer_active;
86 int rx_queue;
87 int state;
88
89 icsk = &tp->inet_conn;
90 inet = &icsk->icsk_inet;
91 sp = &inet->sk;
92 fastopenq = &icsk->icsk_accept_queue.fastopenq;
93
94 dest = inet->inet_daddr;
95 src = inet->inet_rcv_saddr;
96 destp = bpf_ntohs(inet->inet_dport);
97 srcp = bpf_ntohs(inet->inet_sport);
98
99 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
100 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
101 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
102 timer_active = 1;
103 timer_expires = icsk->icsk_timeout;
104 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
105 timer_active = 4;
106 timer_expires = icsk->icsk_timeout;
107 } else if (timer_pending(&sp->sk_timer)) {
108 timer_active = 2;
109 timer_expires = sp->sk_timer.expires;
110 } else {
111 timer_active = 0;
112 timer_expires = bpf_jiffies64();
113 }
114
115 state = sp->sk_state;
116 if (state == TCP_LISTEN) {
117 rx_queue = sp->sk_ack_backlog;
118 } else {
119 rx_queue = tp->rcv_nxt - tp->copied_seq;
120 if (rx_queue < 0)
121 rx_queue = 0;
122 }
123
124 BPF_SEQ_PRINTF(seq, "%4d: %08X:%04X %08X:%04X ",
125 seq_num, src, srcp, dest, destp);
126 BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d ",
127 state,
128 tp->write_seq - tp->snd_una, rx_queue,
129 timer_active,
130 jiffies_delta_to_clock_t(timer_expires - bpf_jiffies64()),
131 icsk->icsk_retransmits, uid,
132 icsk->icsk_probes_out,
133 sock_i_ino(sp),
134 sp->sk_refcnt.refs.counter);
135 BPF_SEQ_PRINTF(seq, "%pK %lu %lu %u %u %d\n",
136 tp,
137 jiffies_to_clock_t(icsk->icsk_rto),
138 jiffies_to_clock_t(icsk->icsk_ack.ato),
139 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(icsk),
140 tp->snd_cwnd,
141 state == TCP_LISTEN ? fastopenq->max_qlen
142 : (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
143 );
144
145 return 0;
146 }
147
dump_tw_sock(struct seq_file * seq,struct tcp_timewait_sock * ttw,uid_t uid,__u32 seq_num)148 static int dump_tw_sock(struct seq_file *seq, struct tcp_timewait_sock *ttw,
149 uid_t uid, __u32 seq_num)
150 {
151 struct inet_timewait_sock *tw = &ttw->tw_sk;
152 __u16 destp, srcp;
153 __be32 dest, src;
154 long delta;
155
156 delta = tw->tw_timer.expires - bpf_jiffies64();
157 dest = tw->tw_daddr;
158 src = tw->tw_rcv_saddr;
159 destp = bpf_ntohs(tw->tw_dport);
160 srcp = bpf_ntohs(tw->tw_sport);
161
162 BPF_SEQ_PRINTF(seq, "%4d: %08X:%04X %08X:%04X ",
163 seq_num, src, srcp, dest, destp);
164
165 BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
166 tw->tw_substate, 0, 0,
167 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
168 tw->tw_refcnt.refs.counter, tw);
169
170 return 0;
171 }
172
dump_req_sock(struct seq_file * seq,struct tcp_request_sock * treq,uid_t uid,__u32 seq_num)173 static int dump_req_sock(struct seq_file *seq, struct tcp_request_sock *treq,
174 uid_t uid, __u32 seq_num)
175 {
176 struct inet_request_sock *irsk = &treq->req;
177 struct request_sock *req = &irsk->req;
178 long ttd;
179
180 ttd = req->rsk_timer.expires - bpf_jiffies64();
181
182 if (ttd < 0)
183 ttd = 0;
184
185 BPF_SEQ_PRINTF(seq, "%4d: %08X:%04X %08X:%04X ",
186 seq_num, irsk->ir_loc_addr,
187 irsk->ir_num, irsk->ir_rmt_addr,
188 bpf_ntohs(irsk->ir_rmt_port));
189 BPF_SEQ_PRINTF(seq, "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
190 TCP_SYN_RECV, 0, 0, 1, jiffies_to_clock_t(ttd),
191 req->num_timeout, uid, 0, 0, 0, req);
192
193 return 0;
194 }
195
196 SEC("iter/tcp")
dump_tcp4(struct bpf_iter__tcp * ctx)197 int dump_tcp4(struct bpf_iter__tcp *ctx)
198 {
199 struct sock_common *sk_common = ctx->sk_common;
200 struct seq_file *seq = ctx->meta->seq;
201 struct tcp_timewait_sock *tw;
202 struct tcp_request_sock *req;
203 struct tcp_sock *tp;
204 uid_t uid = ctx->uid;
205 __u32 seq_num;
206
207 if (sk_common == (void *)0)
208 return 0;
209
210 seq_num = ctx->meta->seq_num;
211 if (seq_num == 0)
212 BPF_SEQ_PRINTF(seq, " sl "
213 "local_address "
214 "rem_address "
215 "st tx_queue rx_queue tr tm->when retrnsmt"
216 " uid timeout inode\n");
217
218 if (sk_common->skc_family != AF_INET)
219 return 0;
220
221 tp = bpf_skc_to_tcp_sock(sk_common);
222 if (tp)
223 return dump_tcp_sock(seq, tp, uid, seq_num);
224
225 tw = bpf_skc_to_tcp_timewait_sock(sk_common);
226 if (tw)
227 return dump_tw_sock(seq, tw, uid, seq_num);
228
229 req = bpf_skc_to_tcp_request_sock(sk_common);
230 if (req)
231 return dump_req_sock(seq, req, uid, seq_num);
232
233 return 0;
234 }
235