xref: /freebsd/sys/netinet/tcp_log_buf.h (revision 2e3f49888ec8851bafb22011533217487764fdb0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2016-2020 Netflix, Inc.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #ifndef __tcp_log_buf_h__
29 #define __tcp_log_buf_h__
30 
31 #define	TCP_LOG_REASON_LEN	32
32 #define	TCP_LOG_TAG_LEN		32
33 #define	TCP_LOG_BUF_VER		(9)
34 
35 /*
36  * Because the (struct tcp_log_buffer) includes 8-byte uint64_t's, it requires
37  * 8-byte alignment to work properly on all platforms. Therefore, we will
38  * enforce 8-byte alignment for all the structures that may appear by
39  * themselves (instead of being embedded in another structure) in a data
40  * stream.
41  */
42 #define	ALIGN_TCP_LOG		__aligned(8)
43 
44 /* Information about the socketbuffer state. */
45 struct tcp_log_sockbuf
46 {
47 	uint32_t	tls_sb_acc;	/* available chars (sb->sb_acc) */
48 	uint32_t	tls_sb_ccc;	/* claimed chars (sb->sb_ccc) */
49 	uint32_t	tls_sb_spare;	/* spare */
50 };
51 
52 /* Optional, verbose information that may be appended to an event log. */
53 struct tcp_log_verbose
54 {
55 #define	TCP_FUNC_LEN	32
56 	char		tlv_snd_frm[TCP_FUNC_LEN]; /* tcp_output() caller */
57 	char		tlv_trace_func[TCP_FUNC_LEN]; /* Function that
58 							 generated trace */
59 	uint32_t	tlv_trace_line;	/* Line number that generated trace */
60 	uint8_t		_pad[4];
61 } ALIGN_TCP_LOG;
62 
63 /* Internal RACK state variables. */
64 struct tcp_log_rack
65 {
66 	uint32_t	tlr_rack_rtt;		/* rc_rack_rtt */
67 	uint8_t		tlr_state;		/* Internal RACK state */
68 	uint8_t		_pad[3];		/* Padding */
69 };
70 
71 struct tcp_log_bbr {
72 	uint64_t cur_del_rate;
73 	uint64_t delRate;
74 	uint64_t rttProp;
75 	uint64_t bw_inuse;
76 	uint32_t inflight;
77 	uint32_t applimited;
78 	uint32_t delivered;
79 	uint32_t timeStamp;
80 	uint32_t epoch;
81 	uint32_t lt_epoch;
82 	uint32_t pkts_out;
83 	uint32_t flex1;
84 	uint32_t flex2;
85 	uint32_t flex3;
86 	uint32_t flex4;
87 	uint32_t flex5;
88 	uint32_t flex6;
89 	uint32_t lost;
90 	uint16_t pacing_gain;
91 	uint16_t cwnd_gain;
92 	uint16_t flex7;
93 	uint8_t bbr_state;
94 	uint8_t bbr_substate;
95 	uint8_t inhpts;
96 	uint8_t __spare;
97 	uint8_t use_lt_bw;
98 	uint8_t flex8;
99 	uint32_t pkt_epoch;
100 };
101 
102 /* shadows tcp_log_bbr struct element sizes */
103 struct tcp_log_raw {
104 	uint64_t u64_flex[4];
105 	uint32_t u32_flex[14];
106 	uint16_t u16_flex[3];
107 	uint8_t u8_flex[6];
108 	uint32_t u32_flex2[1];
109 };
110 
111 struct tcp_log_uint64 {
112 	uint64_t u64_flex[13];
113 };
114 
115 struct tcp_log_sendfile {
116 	uint64_t offset;
117 	uint64_t length;
118 	uint32_t flags;
119 };
120 
121 /*
122  * tcp_log_stackspecific is currently being used as "event specific" log
123  * info by all stacks (i.e. struct tcp_log_bbr is used for generic event
124  * logging). Until this is cleaned up more generically and throughout,
125  * allow events to use the same space in the union.
126  */
127 union tcp_log_stackspecific
128 {
129 	struct tcp_log_rack u_rack;
130 	struct tcp_log_bbr u_bbr;
131 	struct tcp_log_sendfile u_sf;
132 	struct tcp_log_raw u_raw;	/* "raw" log access */
133 	struct tcp_log_uint64 u64_raw;	/* just u64's - used by process info */
134 };
135 
136 typedef union tcp_log_stackspecific tcp_log_eventspecific_t;
137 
138 struct tcp_log_buffer
139 {
140 	/* Event basics */
141 	struct timeval	tlb_tv;		/* Timestamp of trace */
142 	uint32_t	tlb_ticks;	/* Timestamp of trace */
143 	uint32_t	tlb_sn;		/* Serial number */
144 	uint8_t		tlb_stackid;	/* Stack ID */
145 	uint8_t		tlb_eventid;	/* Event ID */
146 	uint16_t	tlb_eventflags;	/* Flags for the record */
147 #define	TLB_FLAG_RXBUF		0x0001	/* Includes receive buffer info */
148 #define	TLB_FLAG_TXBUF		0x0002	/* Includes send buffer info */
149 #define	TLB_FLAG_HDR		0x0004	/* Includes a TCP header */
150 #define	TLB_FLAG_VERBOSE	0x0008	/* Includes function/line numbers */
151 #define	TLB_FLAG_STACKINFO	0x0010	/* Includes stack-specific info */
152 	int		tlb_errno;	/* Event error (if any) */
153 
154 	/* Internal session state */
155 	struct tcp_log_sockbuf tlb_rxbuf; /* Receive buffer */
156 	struct tcp_log_sockbuf tlb_txbuf; /* Send buffer */
157 
158 	int		tlb_state;	/* TCPCB t_state */
159 	uint32_t	tlb_starttime;	/* TCPCB t_starttime */
160 	uint32_t	tlb_iss;	/* TCPCB iss */
161 	uint32_t	tlb_flags;	/* TCPCB flags */
162 	uint32_t	tlb_snd_una;	/* TCPCB snd_una */
163 	uint32_t	tlb_snd_max;	/* TCPCB snd_max */
164 	uint32_t	tlb_snd_cwnd;	/* TCPCB snd_cwnd */
165 	uint32_t	tlb_snd_nxt;	/* TCPCB snd_nxt */
166 	uint32_t	tlb_snd_recover;/* TCPCB snd_recover */
167 	uint32_t	tlb_snd_wnd;	/* TCPCB snd_wnd */
168 	uint32_t	tlb_snd_ssthresh; /* TCPCB snd_ssthresh */
169 	uint32_t	tlb_srtt;	/* TCPCB t_srtt */
170 	uint32_t	tlb_rttvar;	/* TCPCB t_rttvar */
171 	uint32_t	tlb_rcv_up;	/* TCPCB rcv_up */
172 	uint32_t	tlb_rcv_adv;	/* TCPCB rcv_adv */
173 	uint32_t	tlb_flags2;	/* TCPCB t_flags2 */
174 	uint32_t	tlb_rcv_nxt;	/* TCPCB rcv_nxt */
175 	uint32_t	tlb_rcv_wnd;	/* TCPCB rcv_wnd */
176 	uint32_t	tlb_dupacks;	/* TCPCB t_dupacks */
177 	int		tlb_segqlen;	/* TCPCB segqlen */
178 	int		tlb_snd_numholes; /* TCPCB snd_numholes */
179 	uint32_t	tlb_flex1;	/* Event specific information */
180 	uint32_t	tlb_flex2;	/* Event specific information */
181 	uint32_t	tlb_fbyte_in;	/* TCPCB first byte in time */
182 	uint32_t	tlb_fbyte_out;	/* TCPCB first byte out time */
183 	uint8_t		tlb_snd_scale:4, /* TCPCB snd_scale */
184 			tlb_rcv_scale:4; /* TCPCB rcv_scale */
185 	uint8_t		_pad[3];	/* Padding */
186 	/* Per-stack info */
187 	union tcp_log_stackspecific tlb_stackinfo;
188 #define	tlb_rack	tlb_stackinfo.u_rack
189 
190 	/* The packet */
191 	uint32_t	tlb_len;	/* The packet's data length */
192 	struct tcphdr	tlb_th;		/* The TCP header */
193 	uint8_t		tlb_opts[TCP_MAXOLEN]; /* The TCP options */
194 
195 	/* Verbose information (optional) */
196 	struct tcp_log_verbose tlb_verbose[0];
197 } ALIGN_TCP_LOG;
198 
199 enum tcp_log_events {
200 	TCP_LOG_IN = 1,		/* Incoming packet                   1 */
201 	TCP_LOG_OUT,		/* Transmit (without other event)    2 */
202 	TCP_LOG_RTO,		/* Retransmit timeout                3 */
203 	TCP_LOG_SB_WAKE,	/* Awaken socket buffer              4 */
204 	TCP_LOG_BAD_RETRAN,	/* Detected bad retransmission       5 */
205 	TCP_LOG_PRR,		/* Doing PRR                         6 */
206 	TCP_LOG_REORDER,	/* Detected reorder                  7 */
207 	TCP_LOG_HPTS,		/* Hpts sending a packet             8 */
208 	BBR_LOG_BBRUPD,		/* We updated BBR info               9 */
209 	BBR_LOG_BBRSND,		/* We did a slot calculation and sending is done 10 */
210 	BBR_LOG_ACKCLEAR,	/* A ack clears all outstanding     11 */
211 	BBR_LOG_INQUEUE,	/* The tcb had a packet input to it 12 */
212 	BBR_LOG_TIMERSTAR,	/* Start a timer                    13 */
213 	BBR_LOG_TIMERCANC,	/* Cancel a timer                   14 */
214 	BBR_LOG_ENTREC,		/* Entered recovery                 15 */
215 	BBR_LOG_EXITREC,	/* Exited recovery                  16 */
216 	BBR_LOG_CWND,		/* Cwnd change                      17 */
217 	BBR_LOG_BWSAMP,		/* LT B/W sample has been made      18 */
218 	BBR_LOG_MSGSIZE,	/* We received a EMSGSIZE error     19 */
219 	BBR_LOG_BBRRTT,		/* BBR RTT is updated               20 */
220 	BBR_LOG_JUSTRET,	/* We just returned out of output   21 */
221 	BBR_LOG_STATE,		/* A BBR state change occurred      22 */
222 	BBR_LOG_PKT_EPOCH,      /* A BBR packet epoch occurred      23 */
223 	BBR_LOG_PERSIST,        /* BBR changed to/from a persists   24 */
224 	TCP_LOG_FLOWEND,        /* End of a flow                    25 */
225 	BBR_LOG_RTO,            /* BBR's timeout includes BBR info  26 */
226 	BBR_LOG_DOSEG_DONE,     /* hpts do_segment completes        27 */
227 	BBR_LOG_EXIT_GAIN,      /* hpts do_segment completes        28 */
228 	BBR_LOG_THRESH_CALC,    /* Doing threshold calculation      29 */
229 	TCP_LOG_MAPCHG,		/* Map Changes to the sendmap       30 */
230 	TCP_LOG_USERSEND, 	/* User level sends data            31 */
231 	BBR_RSM_CLEARED,	/* RSM cleared of ACK flags         32 */
232 	BBR_LOG_STATE_TARGET, 	/* Log of target at state           33 */
233 	BBR_LOG_TIME_EPOCH, 	/* A timed based Epoch occurred     34 */
234 	BBR_LOG_TO_PROCESS,	/* A to was processed               35 */
235 	BBR_LOG_BBRTSO,		/* TSO update                       36 */
236 	BBR_LOG_HPTSDIAG,	/* Hpts diag insert                 37 */
237 	BBR_LOG_LOWGAIN,	/* Low gain accounting              38 */
238 	BBR_LOG_PROGRESS,	/* Progress timer event             39 */
239 	TCP_LOG_SOCKET_OPT,	/* A socket option is set           40 */
240 	BBR_LOG_TIMERPREP,	/* A BBR var to debug out TLP issues  41 */
241 	BBR_LOG_ENOBUF_JMP,	/* We had a enobuf jump             42 */
242 	BBR_LOG_HPTSI_CALC,	/* calc the hptsi time              43 */
243 	BBR_LOG_RTT_SHRINKS,	/* We had a log reduction of rttProp 44 */
244 	BBR_LOG_BW_RED_EV,	/* B/W reduction events             45 */
245 	BBR_LOG_REDUCE,		/* old bbr log reduce for 4.1 and earlier 46*/
246 	TCP_LOG_RTT,		/* A rtt (in useconds) is being sampled and applied to the srtt algo 47 */
247 	BBR_LOG_SETTINGS_CHG,	/* Settings changed for loss response 48 */
248 	BBR_LOG_SRTT_GAIN_EVENT, /* SRTT gaining -- now not used    49 */
249 	TCP_LOG_REASS,		/* Reassembly buffer logging        50 */
250 	TCP_HDWR_PACE_SIZE,	/*  TCP pacing size set (rl and rack uses this)  51 */
251 	BBR_LOG_HDWR_PACE,	/* TCP Hardware pacing log          52 */
252 	BBR_LOG_TSTMP_VAL,	/* Temp debug timestamp validation  53 */
253 	TCP_LOG_CONNEND,	/* End of connection                54 */
254 	TCP_LOG_LRO,		/* LRO entry                        55 */
255 	TCP_SACK_FILTER_RES,	/* Results of SACK Filter           56 */
256 	TCP_SAD_DETECT,		/* Sack Attack Detection            57 */
257 	TCP_TIMELY_WORK,	/* Logs regarding Timely CC tweaks  58 */
258 	TCP_LOG_USER_EVENT,	/* User space event data            59 */
259 	TCP_LOG_SENDFILE,	/* sendfile() logging for TCP connections 60 */
260 	TCP_LOG_REQ_T,		/* logging of request tracking 61 */
261 	TCP_LOG_ACCOUNTING,	/* Log of TCP Accounting data 62 */
262 	TCP_LOG_FSB,		/* FSB information 63 */
263 	RACK_DSACK_HANDLING,	/* Handling of DSACK in rack for reordering window 64 */
264 	TCP_HYSTART,		/* TCP Hystart logging 65 */
265 	TCP_CHG_QUERY,		/* Change query during fnc_init() 66 */
266 	TCP_RACK_LOG_COLLAPSE,	/* Window collapse by peer 67 */
267 	TCP_RACK_TP_TRIGGERED,	/* A rack tracepoint is triggered 68 */
268 	TCP_HYBRID_PACING_LOG,	/* Hybrid pacing log 69 */
269 	TCP_LOG_PRU,		/* TCP protocol user request        70 */
270 	TCP_LOG_END		/* End (keep at end)                71 */
271 };
272 
273 enum tcp_log_states {
274 	TCP_LOG_STATE_RATIO_OFF = -2,	/* Log ratio evaluation yielded an OFF
275 					   result. Only used for tlb_logstate */
276 	TCP_LOG_STATE_CLEAR = -1,	/* Deactivate and clear tracing. Passed
277 					   to tcp_log_state_change() but never
278 					   stored in any logstate variable */
279 	TCP_LOG_STATE_OFF = 0,		/* Pause */
280 
281 	/* Positively numbered states represent active logging modes */
282 	TCP_LOG_STATE_TAIL=1,		/* Keep the trailing events */
283 	TCP_LOG_STATE_HEAD=2,		/* Keep the leading events */
284 	TCP_LOG_STATE_HEAD_AUTO=3,	/* Keep the leading events, and
285 					   automatically dump them to the
286 					   device  */
287 	TCP_LOG_STATE_CONTINUAL=4,	/* Continually dump the data when full */
288 	TCP_LOG_STATE_TAIL_AUTO=5,	/* Keep the trailing events, and
289 					   automatically dump them when the
290 					   session ends */
291 	TCP_LOG_VIA_BBPOINTS=6		/* Log only if the BB point has been configured */
292 };
293 
294 /* Use this if we don't know whether the operation succeeded. */
295 #define	ERRNO_UNK	(-1)
296 
297 /*
298  * If the user included dev/tcp_log/tcp_log_dev.h, then include our private
299  * headers. Otherwise, there is no reason to pollute all the files with an
300  * additional include.
301  *
302  * This structure is aligned to an 8-byte boundary to match the alignment
303  * requirements of (struct tcp_log_buffer).
304  */
305 #ifdef __tcp_log_dev_h__
306 struct tcp_log_header {
307 	struct tcp_log_common_header tlh_common;
308 #define	tlh_version	tlh_common.tlch_version
309 #define	tlh_type	tlh_common.tlch_type
310 #define	tlh_length	tlh_common.tlch_length
311 	struct in_endpoints	tlh_ie;
312 	struct timeval		tlh_offset;	/* Uptime -> UTC offset */
313 	char			tlh_id[TCP_LOG_ID_LEN];
314 	char			tlh_reason[TCP_LOG_REASON_LEN];
315 	char			tlh_tag[TCP_LOG_TAG_LEN];
316 	uint8_t		tlh_af;
317 	uint8_t		_pad[7];
318 } ALIGN_TCP_LOG;
319 
320 #ifdef _KERNEL
321 struct tcp_log_dev_log_queue {
322 	struct tcp_log_dev_queue tldl_common;
323 	char			tldl_id[TCP_LOG_ID_LEN];
324 	char			tldl_reason[TCP_LOG_REASON_LEN];
325 	char			tldl_tag[TCP_LOG_TAG_LEN];
326 	struct in_endpoints	tldl_ie;
327 	struct tcp_log_stailq	tldl_entries;
328 	int			tldl_count;
329 	uint8_t			tldl_af;
330 };
331 #endif /* _KERNEL */
332 #endif /* __tcp_log_dev_h__ */
333 
334 /*
335  * Defined BBPOINTS that can be used
336  * with TCP_LOG_VIA_BBPOINTS.
337  */
338 #define TCP_BBPOINT_NONE		0
339 #define TCP_BBPOINT_REQ_LEVEL_LOGGING	1
340 
341 /*********************/
342 /* TCP Trace points */
343 /*********************/
344 /*
345  * TCP trace points are interesting points within
346  * the TCP code that the author/debugger may want
347  * to have BB logging enabled if we hit that point.
348  * In order to enable a trace point you set the
349  * sysctl var net.inet.tcp.bb.tp.number to
350  * one of the numbers listed below. You also
351  * must make sure net.inet.tcp.bb.tp.bbmode is
352  * non-zero, the default is 4 for continuous tracing.
353  * You also set in the number of connections you want
354  * have get BB logs in net.inet.tcp.bb.tp.count.
355  *
356  * Count will decrement every time BB logging is assigned
357  * to a connection that hit your tracepoint.
358  *
359  * You can enable all trace points by setting the number
360  * to 0xffffffff. You can disable all trace points by
361  * setting number to zero (or count to 0).
362  *
363  * Below are the enumerated list of tracepoints that
364  * have currently been defined in the code. Add more
365  * as you add a call to rack_trace_point(rack, <name>);
366  * where <name> is defined below.
367  */
368 #define TCP_TP_HWENOBUF		0x00000001	/* When we are doing hardware pacing and hit enobufs */
369 #define TCP_TP_ENOBUF		0x00000002	/* When we hit enobufs with software pacing */
370 #define TCP_TP_COLLAPSED_WND	0x00000003	/* When a peer to collapses its rwnd on us */
371 #define TCP_TP_COLLAPSED_RXT	0x00000004	/* When we actually retransmit a collapsed window rsm */
372 #define TCP_TP_REQ_LOG_FAIL	0x00000005	/* We tried to allocate a Request log but had no space */
373 #define TCP_TP_RESET_RCV	0x00000006	/* Triggers when we receive a RST */
374 #define TCP_TP_EXCESS_RXT	0x00000007	/* When we get excess RXT's clamping the cwnd */
375 #define TCP_TP_SAD_TRIGGERED	0x00000008	/* Sack Attack Detection triggers */
376 
377 #define TCP_TP_SAD_SUSPECT	0x0000000a	/* A sack has supicious information in it */
378 
379 #ifdef _KERNEL
380 
381 extern uint32_t tcp_trace_point_config;
382 extern uint32_t tcp_trace_point_bb_mode;
383 extern int32_t tcp_trace_point_count;
384 
385 /*
386  * Returns true if any sort of BB logging is enabled,
387  * commonly used throughout the codebase.
388  */
389 static inline int
390 tcp_bblogging_on(struct tcpcb *tp)
391 {
392 	if (tp->_t_logstate <= TCP_LOG_STATE_OFF)
393 		return (0);
394 	if (tp->_t_logstate == TCP_LOG_VIA_BBPOINTS)
395 		return (0);
396 	return (1);
397 }
398 
399 /*
400  * Returns true if we match a specific bbpoint when
401  * in TCP_LOG_VIA_BBPOINTS, but also returns true
402  * for all the other logging states.
403  */
404 static inline int
405 tcp_bblogging_point_on(struct tcpcb *tp, uint8_t bbpoint)
406 {
407 	if (tp->_t_logstate <= TCP_LOG_STATE_OFF)
408 		return (0);
409 	if ((tp->_t_logstate == TCP_LOG_VIA_BBPOINTS) &&
410 	    (tp->_t_logpoint == bbpoint))
411 		return (1);
412 	else if (tp->_t_logstate == TCP_LOG_VIA_BBPOINTS)
413 		return (0);
414 	return (1);
415 }
416 
417 static inline void
418 tcp_set_bblog_state(struct tcpcb *tp, uint8_t ls, uint8_t bbpoint)
419 {
420 	if ((ls == TCP_LOG_VIA_BBPOINTS) &&
421 	    (tp->_t_logstate <= TCP_LOG_STATE_OFF)){
422 		/*
423 		 * We don't allow a BBPOINTS set to override
424 		 * other types of BB logging set by other means such
425 		 * as the bb_ratio/bb_state URL parameters. In other
426 		 * words BBlogging must be *off* in order to turn on
427 		 * a BBpoint.
428 		 */
429 		tp->_t_logpoint = bbpoint;
430 		tp->_t_logstate = ls;
431 	} else if (ls != TCP_LOG_VIA_BBPOINTS) {
432 		tp->_t_logpoint = 0;
433 		if ((ls >= TCP_LOG_STATE_OFF) &&
434 		    (ls < TCP_LOG_VIA_BBPOINTS))
435 			tp->_t_logstate = ls;
436 	}
437 }
438 
439 static inline uint32_t
440 tcp_get_bblog_state(struct tcpcb *tp)
441 {
442 	return (tp->_t_logstate);
443 }
444 
445 static inline void
446 tcp_trace_point(struct tcpcb *tp, int num)
447 {
448 #ifdef TCP_BLACKBOX
449 	if (((tcp_trace_point_config == num)  ||
450 	     (tcp_trace_point_config == 0xffffffff)) &&
451 	    (tcp_trace_point_bb_mode != 0) &&
452 	    (tcp_trace_point_count > 0) &&
453 	    (tcp_bblogging_on(tp) == 0)) {
454 		int res;
455 		res = atomic_fetchadd_int(&tcp_trace_point_count, -1);
456 		if (res > 0) {
457 			tcp_set_bblog_state(tp, tcp_trace_point_bb_mode, TCP_BBPOINT_NONE);
458 		} else {
459 			/* Loss a race assure its zero now */
460 			tcp_trace_point_count = 0;
461 		}
462 	}
463 #endif
464 }
465 
466 #define	TCP_LOG_BUF_DEFAULT_SESSION_LIMIT	5000
467 #define	TCP_LOG_BUF_DEFAULT_GLOBAL_LIMIT	5000000
468 
469 /*
470  * TCP_LOG_EVENT_VERBOSE: The same as TCP_LOG_EVENT, except it always
471  * tries to record verbose information.
472  */
473 #define	TCP_LOG_EVENT_VERBOSE(tp, th, rxbuf, txbuf, eventid, errornum, len, stackinfo, th_hostorder, tv) \
474 	do {								\
475 		if (tcp_bblogging_on(tp)) \
476 			tcp_log_event(tp, th, rxbuf, txbuf, eventid,	\
477 			    errornum, len, stackinfo, th_hostorder,	\
478 			    tp->t_output_caller, __func__, __LINE__, tv);\
479 	} while (0)
480 
481 /*
482  * TCP_LOG_EVENT: This is a macro so we can capture function/line
483  * information when needed. You can use the macro when you are not
484  * doing a lot of prep in the stack specific information i.e. you
485  * don't add extras (stackinfo). If you are adding extras which
486  * means filling out a stack variable instead use the tcp_log_event()
487  * function but enclose the call to the log (and all the setup) in a
488  * if (tcp_bblogging_on(tp)) {
489  *   ... setup and logging call ...
490  * }
491  *
492  * Always use the macro tcp_bblogging_on() since sometimes the defintions
493  * do change.
494  *
495  * BBlogging also supports the concept of a BBpoint. The idea behind this
496  * is that when you set a specific BBpoint on and turn the logging into
497  * the BBpoint mode (TCP_LOG_VIA_BBPOINTS) you will be defining very very
498  * few of these points to come out. The point is specific to a code you
499  * want tied to that one BB logging. This allows you to turn on a much broader
500  * scale set of limited logging on more connections without overwhelming the
501  * I/O system with too much BBlogs. This of course means you need to be quite
502  * careful on how many BBlogs go with each point, but you can have multiple points
503  * only one of which is active at a time.
504  *
505  * To define a point you add it above under the define for TCP_BBPOINT_NONE (which
506  * is the default i.e. no point is defined. You then, for your point use the
507  * tcp_bblogging_point_on(struct tcpcb *tp, uint8_t bbpoint) inline to enclose
508  * your call to tcp_log_event.  Do not use one of the TCP_LOGGING macros else
509  * your point will never come out. You specify your defined point in the bbpoint
510  * side of the inline. An example of this you can find in rack where the
511  * TCP_BBPOINT_REQ_LEVEL_LOGGING is used. There a specific set of logs are generated
512  * for each request that tcp is tracking.
513  *
514  * When turning on BB logging use the inline:
515  * tcp_set_bblog_state(struct tcpcb *tp, uint8_t ls, uint8_t bbpoint)
516  * the ls field is the logging state TCP_LOG_STATE_CONTINUAL etc. The
517  * bbpoint field is ignored unless the ls field is set to TCP_LOG_VIA_BBPOINTS.
518  * Currently there is only a socket option that turns on the non-BBPOINT
519  * logging.
520  *
521  * Prototype:
522  * TCP_LOG_EVENT(struct tcpcb *tp, struct tcphdr *th, struct sockbuf *rxbuf,
523  *     struct sockbuf *txbuf, uint8_t eventid, int errornum,
524  *     union tcp_log_stackspecific *stackinfo)
525  *
526  * tp is mandatory and must be write locked.
527  * th is optional; if present, it will appear in the record.
528  * rxbuf and txbuf are optional; if present, they will appear in the record.
529  * eventid is mandatory.
530  * errornum is mandatory (it indicates the success or failure of the
531  *     operation associated with the event).
532  * len indicates the length of the packet. If no packet, use 0.
533  * stackinfo is optional; if present, it will appear in the record.
534  */
535 struct tcpcb;
536 #ifdef TCP_LOG_FORCEVERBOSE
537 #define	TCP_LOG_EVENT	TCP_LOG_EVENT_VERBOSE
538 #else
539 #define	TCP_LOG_EVENT(tp, th, rxbuf, txbuf, eventid, errornum, len, stackinfo, th_hostorder) \
540 	do {								\
541 		if (tcp_log_verbose)					\
542 			TCP_LOG_EVENT_VERBOSE(tp, th, rxbuf, txbuf,	\
543 			    eventid, errornum, len, stackinfo,		\
544 			    th_hostorder, NULL);			\
545 		else if (tcp_bblogging_on(tp))				\
546 			tcp_log_event(tp, th, rxbuf, txbuf, eventid,	\
547 			    errornum, len, stackinfo, th_hostorder,	\
548 			    NULL, NULL, 0, NULL);			\
549 	} while (0)
550 #endif /* TCP_LOG_FORCEVERBOSE */
551 #define	TCP_LOG_EVENTP(tp, th, rxbuf, txbuf, eventid, errornum, len, stackinfo, th_hostorder, tv) \
552 	do {								\
553 		if (tcp_bblogging_on(tp))				\
554 			tcp_log_event(tp, th, rxbuf, txbuf, eventid,	\
555 			    errornum, len, stackinfo, th_hostorder,	\
556 			    NULL, NULL, 0, tv);				\
557 	} while (0)
558 
559 #ifdef TCP_BLACKBOX
560 extern bool tcp_log_verbose;
561 void tcp_log_drain(struct tcpcb *tp);
562 int tcp_log_dump_tp_logbuf(struct tcpcb *tp, char *reason, int how, bool force);
563 void tcp_log_dump_tp_bucket_logbufs(struct tcpcb *tp, char *reason);
564 struct tcp_log_buffer *tcp_log_event(struct tcpcb *tp, struct tcphdr *th, struct sockbuf *rxbuf,
565     struct sockbuf *txbuf, uint8_t eventid, int errornum, uint32_t len,
566     union tcp_log_stackspecific *stackinfo, int th_hostorder,
567     const char *output_caller, const char *func, int line, const struct timeval *tv);
568 size_t tcp_log_get_id(struct tcpcb *tp, char *buf);
569 size_t tcp_log_get_tag(struct tcpcb *tp, char *buf);
570 u_int tcp_log_get_id_cnt(struct tcpcb *tp);
571 int tcp_log_getlogbuf(struct sockopt *sopt, struct tcpcb *tp);
572 void tcp_log_init(void);
573 int tcp_log_set_id(struct tcpcb *tp, char *id);
574 int tcp_log_set_tag(struct tcpcb *tp, char *tag);
575 int tcp_log_state_change(struct tcpcb *tp, int state);
576 void tcp_log_tcpcbinit(struct tcpcb *tp);
577 void tcp_log_tcpcbfini(struct tcpcb *tp);
578 void tcp_log_flowend(struct tcpcb *tp);
579 void tcp_log_sendfile(struct socket *so, off_t offset, size_t nbytes,
580     int flags);
581 int tcp_log_apply_ratio(struct tcpcb *tp, int ratio);
582 #else /* !TCP_BLACKBOX */
583 #define tcp_log_verbose	(false)
584 
585 static inline struct tcp_log_buffer *
586 tcp_log_event(struct tcpcb *tp, struct tcphdr *th, struct sockbuf *rxbuf,
587     struct sockbuf *txbuf, uint8_t eventid, int errornum, uint32_t len,
588     union tcp_log_stackspecific *stackinfo, int th_hostorder,
589     const char *output_caller, const char *func, int line,
590     const struct timeval *tv)
591 {
592 
593 	return (NULL);
594 }
595 #endif /* TCP_BLACKBOX */
596 
597 #endif	/* _KERNEL */
598 #endif	/* __tcp_log_buf_h__ */
599