pmclog.c revision 157144
1/*-
2 * Copyright (c) 2005-2006 Joseph Koshy
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/lib/libpmc/pmclog.c 157144 2006-03-26 12:20:54Z jkoshy $");
29
30#include <sys/param.h>
31#include <sys/pmc.h>
32#include <sys/pmclog.h>
33
34#include <assert.h>
35#include <errno.h>
36#include <pmc.h>
37#include <pmclog.h>
38#include <stddef.h>
39#include <stdlib.h>
40#include <string.h>
41#include <strings.h>
42#include <unistd.h>
43
44#include <machine/pmc_mdep.h>
45
46#define	PMCLOG_BUFFER_SIZE			4096
47
48/*
49 * API NOTES
50 *
51 * The pmclog(3) API is oriented towards parsing an event stream in
52 * "realtime", i.e., from an data source that may or may not preserve
53 * record boundaries -- for example when the data source is elsewhere
54 * on a network.  The API allows data to be fed into the parser zero
55 * or more bytes at a time.
56 *
57 * The state for a log file parser is maintained in a 'struct
58 * pmclog_parse_state'.  Parser invocations are done by calling
59 * 'pmclog_read()'; this function will inform the caller when a
60 * complete event is parsed.
61 *
62 * The parser first assembles a complete log file event in an internal
63 * work area (see "ps_saved" below).  Once a complete log file event
64 * is read, the parser then parses it and converts it to an event
65 * descriptor usable by the client.  We could possibly avoid this two
66 * step process by directly parsing the input log to set fields in the
67 * event record.  However the parser's state machine would get
68 * insanely complicated, and this code is unlikely to be used in
69 * performance critical paths.
70 */
71
72enum pmclog_parser_state {
73	PL_STATE_NEW_RECORD,		/* in-between records */
74	PL_STATE_EXPECTING_HEADER,	/* header being read */
75	PL_STATE_PARTIAL_RECORD,	/* header present but not the record */
76	PL_STATE_ERROR			/* parsing error encountered */
77};
78
79struct pmclog_parse_state {
80	enum pmclog_parser_state ps_state;
81	enum pmc_cputype	ps_arch;	/* log file architecture */
82	uint32_t		ps_version;	/* hwpmc version */
83	int			ps_initialized;	/* whether initialized */
84	int			ps_count;	/* count of records processed */
85	off_t			ps_offset;	/* stream byte offset */
86	union pmclog_entry	ps_saved;	/* saved partial log entry */
87	int			ps_svcount;	/* #bytes saved */
88	int			ps_fd;		/* active fd or -1 */
89	char			*ps_buffer;	/* scratch buffer if fd != -1 */
90	char			*ps_data;	/* current parse pointer */
91	size_t			ps_len;		/* length of buffered data */
92};
93
94#define	PMCLOG_HEADER_FROM_SAVED_STATE(PS)				\
95	(* ((uint32_t *) &(PS)->ps_saved))
96
97#define	PMCLOG_INITIALIZE_READER(LE,A)	LE = (uint32_t *) &(A)
98#define	PMCLOG_READ32(LE,V) 		do {				\
99		(V)  = *(LE)++;						\
100	} while (0)
101#define	PMCLOG_READ64(LE,V)		do {				\
102		uint64_t _v;						\
103		_v  = (uint64_t) *(LE)++;				\
104		_v |= ((uint64_t) *(LE)++) << 32;			\
105		(V) = _v;						\
106	} while (0)
107
108#define	PMCLOG_READSTRING(LE,DST,LEN)	strlcpy((DST), (char *) (LE), (LEN))
109
110/*
111 * Assemble a log record from '*len' octets starting from address '*data'.
112 * Update 'data' and 'len' to reflect the number of bytes consumed.
113 *
114 * '*data' is potentially an unaligned address and '*len' octets may
115 * not be enough to complete a event record.
116 */
117
118static enum pmclog_parser_state
119pmclog_get_record(struct pmclog_parse_state *ps, char **data, ssize_t *len)
120{
121	int avail, copylen, recordsize, used;
122	uint32_t h;
123	const int HEADERSIZE = sizeof(uint32_t);
124	char *src, *dst;
125
126	if ((avail = *len) <= 0)
127		return (ps->ps_state = PL_STATE_ERROR);
128
129	src = *data;
130	h = used = 0;
131
132	if (ps->ps_state == PL_STATE_NEW_RECORD)
133		ps->ps_svcount = 0;
134
135	dst = (char *) &ps->ps_saved + ps->ps_svcount;
136
137	switch (ps->ps_state) {
138	case PL_STATE_NEW_RECORD:
139
140		/*
141		 * Transitions:
142		 *
143		 * Case A: avail < headersize
144		 *	-> 'expecting header'
145		 *
146		 * Case B: avail >= headersize
147		 *    B.1: avail < recordsize
148		 *	   -> 'partial record'
149		 *    B.2: avail >= recordsize
150		 *         -> 'new record'
151		 */
152
153		copylen = avail < HEADERSIZE ? avail : HEADERSIZE;
154		bcopy(src, dst, copylen);
155		ps->ps_svcount = used = copylen;
156
157		if (copylen < HEADERSIZE) {
158			ps->ps_state = PL_STATE_EXPECTING_HEADER;
159			goto done;
160		}
161
162		src += copylen;
163		dst += copylen;
164
165		h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
166		recordsize = PMCLOG_HEADER_TO_LENGTH(h);
167
168		if (recordsize <= 0)
169			goto error;
170
171		if (recordsize <= avail) { /* full record available */
172			bcopy(src, dst, recordsize - copylen);
173			ps->ps_svcount = used = recordsize;
174			goto done;
175		}
176
177		/* header + a partial record is available */
178		bcopy(src, dst, avail - copylen);
179		ps->ps_svcount = used = avail;
180		ps->ps_state = PL_STATE_PARTIAL_RECORD;
181
182		break;
183
184	case PL_STATE_EXPECTING_HEADER:
185
186		/*
187		 * Transitions:
188		 *
189		 * Case C: avail+saved < headersize
190		 * 	-> 'expecting header'
191		 *
192		 * Case D: avail+saved >= headersize
193		 *    D.1: avail+saved < recordsize
194		 *    	-> 'partial record'
195		 *    D.2: avail+saved >= recordsize
196		 *    	-> 'new record'
197		 *    (see PARTIAL_RECORD handling below)
198		 */
199
200		if (avail + ps->ps_svcount < HEADERSIZE) {
201			bcopy(src, dst, avail);
202			ps->ps_svcount += avail;
203			used = avail;
204			break;
205		}
206
207		used = copylen = HEADERSIZE - ps->ps_svcount;
208		bcopy(src, dst, copylen);
209		src += copylen;
210		dst += copylen;
211		avail -= copylen;
212		ps->ps_svcount += copylen;
213
214		/*FALLTHROUGH*/
215
216	case PL_STATE_PARTIAL_RECORD:
217
218		/*
219		 * Transitions:
220		 *
221		 * Case E: avail+saved < recordsize
222		 * 	-> 'partial record'
223		 *
224		 * Case F: avail+saved >= recordsize
225		 * 	-> 'new record'
226		 */
227
228		h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
229		recordsize = PMCLOG_HEADER_TO_LENGTH(h);
230
231		if (recordsize <= 0)
232			goto error;
233
234		if (avail + ps->ps_svcount < recordsize) {
235			copylen = avail;
236			ps->ps_state = PL_STATE_PARTIAL_RECORD;
237		} else {
238			copylen = recordsize - ps->ps_svcount;
239			ps->ps_state = PL_STATE_NEW_RECORD;
240		}
241
242		bcopy(src, dst, copylen);
243		ps->ps_svcount += copylen;
244		used += copylen;
245		break;
246
247	default:
248		goto error;
249	}
250
251 done:
252	*data += used;
253	*len  -= used;
254	return ps->ps_state;
255
256 error:
257	ps->ps_state = PL_STATE_ERROR;
258	return ps->ps_state;
259}
260
261/*
262 * Get an event from the stream pointed to by '*data'.  '*len'
263 * indicates the number of bytes available to parse.  Arguments
264 * '*data' and '*len' are updated to indicate the number of bytes
265 * consumed.
266 */
267
268static int
269pmclog_get_event(void *cookie, char **data, ssize_t *len,
270    struct pmclog_ev *ev)
271{
272	int evlen, pathlen;
273	uint32_t h, *le;
274	enum pmclog_parser_state e;
275	struct pmclog_parse_state *ps;
276
277	ps = (struct pmclog_parse_state *) cookie;
278
279	assert(ps->ps_state != PL_STATE_ERROR);
280
281	if ((e = pmclog_get_record(ps,data,len)) == PL_STATE_ERROR) {
282		ev->pl_state = PMCLOG_ERROR;
283		return -1;
284	}
285
286	if (e != PL_STATE_NEW_RECORD) {
287		ev->pl_state = PMCLOG_REQUIRE_DATA;
288		return -1;
289	}
290
291	PMCLOG_INITIALIZE_READER(le, ps->ps_saved);
292
293	PMCLOG_READ32(le,h);
294
295	if (!PMCLOG_HEADER_CHECK_MAGIC(h)) {
296		ps->ps_state = PL_STATE_ERROR;
297		ev->pl_state = PMCLOG_ERROR;
298		return -1;
299	}
300
301	/* copy out the time stamp */
302	PMCLOG_READ32(le,ev->pl_ts.tv_sec);
303	PMCLOG_READ32(le,ev->pl_ts.tv_nsec);
304
305	evlen = PMCLOG_HEADER_TO_LENGTH(h);
306
307#define	PMCLOG_GET_PATHLEN(P,E,TYPE) do {				\
308		(P) = (E) - offsetof(struct TYPE, pl_pathname);		\
309		if ((P) > PATH_MAX || (P) < 0)				\
310			goto error;					\
311	} while (0)
312
313	switch (ev->pl_type = PMCLOG_HEADER_TO_TYPE(h)) {
314	case PMCLOG_TYPE_CLOSELOG:
315	case PMCLOG_TYPE_DROPNOTIFY:
316		/* nothing to do */
317		break;
318	case PMCLOG_TYPE_INITIALIZE:
319		PMCLOG_READ32(le,ev->pl_u.pl_i.pl_version);
320		PMCLOG_READ32(le,ev->pl_u.pl_i.pl_arch);
321		ps->ps_version = ev->pl_u.pl_i.pl_version;
322		ps->ps_arch = ev->pl_u.pl_i.pl_arch;
323		ps->ps_initialized = 1;
324		break;
325	case PMCLOG_TYPE_MAP_IN:
326		PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_map_in);
327		PMCLOG_READ32(le,ev->pl_u.pl_mi.pl_pid);
328		PMCLOG_READADDR(le,ev->pl_u.pl_mi.pl_start);
329		PMCLOG_READSTRING(le, ev->pl_u.pl_mi.pl_pathname, pathlen);
330		break;
331	case PMCLOG_TYPE_MAP_OUT:
332		PMCLOG_READ32(le,ev->pl_u.pl_mo.pl_pid);
333		PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_start);
334		PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_end);
335		break;
336	case PMCLOG_TYPE_PCSAMPLE:
337		PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pid);
338		PMCLOG_READADDR(le,ev->pl_u.pl_s.pl_pc);
339		PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pmcid);
340		PMCLOG_READ32(le,ev->pl_u.pl_s.pl_usermode);
341		break;
342	case PMCLOG_TYPE_PMCALLOCATE:
343		PMCLOG_READ32(le,ev->pl_u.pl_a.pl_pmcid);
344		PMCLOG_READ32(le,ev->pl_u.pl_a.pl_event);
345		PMCLOG_READ32(le,ev->pl_u.pl_a.pl_flags);
346		if ((ev->pl_u.pl_a.pl_evname =
347		    pmc_name_of_event(ev->pl_u.pl_a.pl_event)) == NULL)
348			goto error;
349		break;
350	case PMCLOG_TYPE_PMCATTACH:
351		PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_pmcattach);
352		PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pmcid);
353		PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pid);
354		PMCLOG_READSTRING(le,ev->pl_u.pl_t.pl_pathname,pathlen);
355		break;
356	case PMCLOG_TYPE_PMCDETACH:
357		PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pmcid);
358		PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pid);
359		break;
360	case PMCLOG_TYPE_PROCCSW:
361		PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pmcid);
362		PMCLOG_READ64(le,ev->pl_u.pl_c.pl_value);
363		PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pid);
364		break;
365	case PMCLOG_TYPE_PROCEXEC:
366		PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_procexec);
367		PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pid);
368		PMCLOG_READADDR(le,ev->pl_u.pl_x.pl_entryaddr);
369		PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pmcid);
370		PMCLOG_READSTRING(le,ev->pl_u.pl_x.pl_pathname,pathlen);
371		break;
372	case PMCLOG_TYPE_PROCEXIT:
373		PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pmcid);
374		PMCLOG_READ64(le,ev->pl_u.pl_e.pl_value);
375		PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pid);
376		break;
377	case PMCLOG_TYPE_PROCFORK:
378		PMCLOG_READ32(le,ev->pl_u.pl_f.pl_oldpid);
379		PMCLOG_READ32(le,ev->pl_u.pl_f.pl_newpid);
380		break;
381	case PMCLOG_TYPE_SYSEXIT:
382		PMCLOG_READ32(le,ev->pl_u.pl_se.pl_pid);
383		break;
384	case PMCLOG_TYPE_USERDATA:
385		PMCLOG_READ32(le,ev->pl_u.pl_u.pl_userdata);
386		break;
387	default:	/* unknown record type */
388		ps->ps_state = PL_STATE_ERROR;
389		ev->pl_state = PMCLOG_ERROR;
390		return -1;
391	}
392
393	ev->pl_offset = (ps->ps_offset += evlen);
394	ev->pl_count  = (ps->ps_count += 1);
395	ev->pl_state = PMCLOG_OK;
396	return 0;
397
398 error:
399	ev->pl_state = PMCLOG_ERROR;
400	ps->ps_state = PL_STATE_ERROR;
401	return -1;
402}
403
404/*
405 * Extract and return the next event from the byte stream.
406 *
407 * Returns 0 and sets the event's state to PMCLOG_OK in case an event
408 * was successfully parsed.  Otherwise this function returns -1 and
409 * sets the event's state to one of PMCLOG_REQUIRE_DATA (if more data
410 * is needed) or PMCLOG_EOF (if an EOF was seen) or PMCLOG_ERROR if
411 * a parse error was encountered.
412 */
413
414int
415pmclog_read(void *cookie, struct pmclog_ev *ev)
416{
417	int retval;
418	ssize_t nread;
419	struct pmclog_parse_state *ps;
420
421	ps = (struct pmclog_parse_state *) cookie;
422
423	if (ps->ps_state == PL_STATE_ERROR) {
424		ev->pl_state = PMCLOG_ERROR;
425		return -1;
426	}
427
428	/*
429	 * If there isn't enough data left for a new event try and get
430	 * more data.
431	 */
432	if (ps->ps_len == 0) {
433		ev->pl_state = PMCLOG_REQUIRE_DATA;
434
435		/*
436		 * If we have a valid file descriptor to read from, attempt
437		 * to read from that.  This read may return with an error,
438		 * (which may be EAGAIN or other recoverable error), or
439		 * can return EOF.
440		 */
441		if (ps->ps_fd != PMCLOG_FD_NONE) {
442		refill:
443			nread = read(ps->ps_fd, ps->ps_buffer,
444			    PMCLOG_BUFFER_SIZE);
445
446			if (nread <= 0) {
447				if (nread == 0)
448					ev->pl_state = PMCLOG_EOF;
449				else if (errno != EAGAIN) /* not restartable */
450					ev->pl_state = PMCLOG_ERROR;
451				return -1;
452			}
453
454			ps->ps_len = nread;
455			ps->ps_data = ps->ps_buffer;
456		} else
457			return -1;
458	}
459
460	assert(ps->ps_len > 0);
461
462
463	 /* Retrieve one event from the byte stream. */
464	retval = pmclog_get_event(ps, &ps->ps_data, &ps->ps_len, ev);
465
466	/*
467	 * If we need more data and we have a configured fd, try read
468	 * from it.
469	 */
470	if (retval < 0 && ev->pl_state == PMCLOG_REQUIRE_DATA &&
471	    ps->ps_fd != -1) {
472		assert(ps->ps_len == 0);
473		goto refill;
474	}
475
476	return retval;
477}
478
479/*
480 * Feed data to a memory based parser.
481 *
482 * The memory area pointed to by 'data' needs to be valid till the
483 * next error return from pmclog_next_event().
484 */
485
486int
487pmclog_feed(void *cookie, char *data, int len)
488{
489	struct pmclog_parse_state *ps;
490
491	ps = (struct pmclog_parse_state *) cookie;
492
493	if (len < 0 ||		/* invalid length */
494	    ps->ps_buffer ||	/* called for a file parser */
495	    ps->ps_len != 0)	/* unnecessary call */
496		return -1;
497
498	ps->ps_data = data;
499	ps->ps_len  = len;
500
501	return 0;
502}
503
504/*
505 * Allocate and initialize parser state.
506 */
507
508void *
509pmclog_open(int fd)
510{
511	struct pmclog_parse_state *ps;
512
513	if ((ps = (struct pmclog_parse_state *) malloc(sizeof(*ps))) == NULL)
514		return NULL;
515
516	ps->ps_state = PL_STATE_NEW_RECORD;
517	ps->ps_arch = -1;
518	ps->ps_initialized = 0;
519	ps->ps_count = 0;
520	ps->ps_offset = (off_t) 0;
521	bzero(&ps->ps_saved, sizeof(ps->ps_saved));
522	ps->ps_svcount = 0;
523	ps->ps_fd    = fd;
524	ps->ps_data  = NULL;
525	ps->ps_buffer = NULL;
526	ps->ps_len   = 0;
527
528	/* allocate space for a work area */
529	if (ps->ps_fd != PMCLOG_FD_NONE) {
530		if ((ps->ps_buffer = malloc(PMCLOG_BUFFER_SIZE)) == NULL)
531			return NULL;
532	}
533
534	return ps;
535}
536
537
538/*
539 * Free up parser state.
540 */
541
542void
543pmclog_close(void *cookie)
544{
545	struct pmclog_parse_state *ps;
546
547	ps = (struct pmclog_parse_state *) cookie;
548
549	if (ps->ps_buffer)
550		free(ps->ps_buffer);
551
552	free(ps);
553}
554