2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2005-2007 Joseph Koshy
5 * Copyright (c) 2007 The FreeBSD Foundation
8 * Portions of this software were developed by A. Joseph Koshy under
9 * sponsorship from the FreeBSD Foundation and Google, Inc.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include <sys/param.h>
38 #include <sys/pmclog.h>
50 #include <machine/pmc_mdep.h>
52 #include "libpmcinternal.h"
54 #define PMCLOG_BUFFER_SIZE 4096
59 * The pmclog(3) API is oriented towards parsing an event stream in
60 * "realtime", i.e., from an data source that may or may not preserve
61 * record boundaries -- for example when the data source is elsewhere
62 * on a network. The API allows data to be fed into the parser zero
63 * or more bytes at a time.
65 * The state for a log file parser is maintained in a 'struct
66 * pmclog_parse_state'. Parser invocations are done by calling
67 * 'pmclog_read()'; this function will inform the caller when a
68 * complete event is parsed.
70 * The parser first assembles a complete log file event in an internal
71 * work area (see "ps_saved" below). Once a complete log file event
72 * is read, the parser then parses it and converts it to an event
73 * descriptor usable by the client. We could possibly avoid this two
74 * step process by directly parsing the input log to set fields in the
75 * event record. However the parser's state machine would get
76 * insanely complicated, and this code is unlikely to be used in
77 * performance critical paths.
80 enum pmclog_parser_state {
81 PL_STATE_NEW_RECORD, /* in-between records */
82 PL_STATE_EXPECTING_HEADER, /* header being read */
83 PL_STATE_PARTIAL_RECORD, /* header present but not the record */
84 PL_STATE_ERROR /* parsing error encountered */
87 struct pmclog_parse_state {
88 enum pmclog_parser_state ps_state;
89 enum pmc_cputype ps_arch; /* log file architecture */
90 uint32_t ps_version; /* hwpmc version */
91 int ps_initialized; /* whether initialized */
92 int ps_count; /* count of records processed */
93 off_t ps_offset; /* stream byte offset */
94 union pmclog_entry ps_saved; /* saved partial log entry */
95 int ps_svcount; /* #bytes saved */
96 int ps_fd; /* active fd or -1 */
97 char *ps_buffer; /* scratch buffer if fd != -1 */
98 char *ps_data; /* current parse pointer */
99 size_t ps_len; /* length of buffered data */
102 #define PMCLOG_HEADER_FROM_SAVED_STATE(PS) \
103 (* ((uint32_t *) &(PS)->ps_saved))
105 #define PMCLOG_INITIALIZE_READER(LE,A) LE = (uint32_t *) &(A)
106 #define PMCLOG_READ32(LE,V) do { \
109 #define PMCLOG_READ64(LE,V) do { \
111 _v = (uint64_t) *(LE)++; \
112 _v |= ((uint64_t) *(LE)++) << 32; \
116 #define PMCLOG_READSTRING(LE,DST,LEN) strlcpy((DST), (char *) (LE), (LEN))
119 * Assemble a log record from '*len' octets starting from address '*data'.
120 * Update 'data' and 'len' to reflect the number of bytes consumed.
122 * '*data' is potentially an unaligned address and '*len' octets may
123 * not be enough to complete a event record.
126 static enum pmclog_parser_state
127 pmclog_get_record(struct pmclog_parse_state *ps, char **data, ssize_t *len)
129 int avail, copylen, recordsize, used;
131 const int HEADERSIZE = sizeof(uint32_t);
134 if ((avail = *len) <= 0)
135 return (ps->ps_state = PL_STATE_ERROR);
140 if (ps->ps_state == PL_STATE_NEW_RECORD)
143 dst = (char *) &ps->ps_saved + ps->ps_svcount;
145 switch (ps->ps_state) {
146 case PL_STATE_NEW_RECORD:
151 * Case A: avail < headersize
152 * -> 'expecting header'
154 * Case B: avail >= headersize
155 * B.1: avail < recordsize
156 * -> 'partial record'
157 * B.2: avail >= recordsize
161 copylen = avail < HEADERSIZE ? avail : HEADERSIZE;
162 bcopy(src, dst, copylen);
163 ps->ps_svcount = used = copylen;
165 if (copylen < HEADERSIZE) {
166 ps->ps_state = PL_STATE_EXPECTING_HEADER;
173 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
174 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
179 if (recordsize <= avail) { /* full record available */
180 bcopy(src, dst, recordsize - copylen);
181 ps->ps_svcount = used = recordsize;
185 /* header + a partial record is available */
186 bcopy(src, dst, avail - copylen);
187 ps->ps_svcount = used = avail;
188 ps->ps_state = PL_STATE_PARTIAL_RECORD;
192 case PL_STATE_EXPECTING_HEADER:
197 * Case C: avail+saved < headersize
198 * -> 'expecting header'
200 * Case D: avail+saved >= headersize
201 * D.1: avail+saved < recordsize
202 * -> 'partial record'
203 * D.2: avail+saved >= recordsize
205 * (see PARTIAL_RECORD handling below)
208 if (avail + ps->ps_svcount < HEADERSIZE) {
209 bcopy(src, dst, avail);
210 ps->ps_svcount += avail;
215 used = copylen = HEADERSIZE - ps->ps_svcount;
216 bcopy(src, dst, copylen);
220 ps->ps_svcount += copylen;
224 case PL_STATE_PARTIAL_RECORD:
229 * Case E: avail+saved < recordsize
230 * -> 'partial record'
232 * Case F: avail+saved >= recordsize
236 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
237 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
242 if (avail + ps->ps_svcount < recordsize) {
244 ps->ps_state = PL_STATE_PARTIAL_RECORD;
246 copylen = recordsize - ps->ps_svcount;
247 ps->ps_state = PL_STATE_NEW_RECORD;
250 bcopy(src, dst, copylen);
251 ps->ps_svcount += copylen;
265 ps->ps_state = PL_STATE_ERROR;
270 * Get an event from the stream pointed to by '*data'. '*len'
271 * indicates the number of bytes available to parse. Arguments
272 * '*data' and '*len' are updated to indicate the number of bytes
277 pmclog_get_event(void *cookie, char **data, ssize_t *len,
278 struct pmclog_ev *ev)
281 uint32_t h, *le, npc;
282 enum pmclog_parser_state e;
283 struct pmclog_parse_state *ps;
285 ps = (struct pmclog_parse_state *) cookie;
287 assert(ps->ps_state != PL_STATE_ERROR);
289 if ((e = pmclog_get_record(ps,data,len)) == PL_STATE_ERROR) {
290 ev->pl_state = PMCLOG_ERROR;
294 if (e != PL_STATE_NEW_RECORD) {
295 ev->pl_state = PMCLOG_REQUIRE_DATA;
299 PMCLOG_INITIALIZE_READER(le, ps->ps_saved);
303 if (!PMCLOG_HEADER_CHECK_MAGIC(h)) {
304 ps->ps_state = PL_STATE_ERROR;
305 ev->pl_state = PMCLOG_ERROR;
309 /* copy out the time stamp */
310 PMCLOG_READ32(le,ev->pl_ts.tv_sec);
311 PMCLOG_READ32(le,ev->pl_ts.tv_nsec);
313 evlen = PMCLOG_HEADER_TO_LENGTH(h);
315 #define PMCLOG_GET_PATHLEN(P,E,TYPE) do { \
316 (P) = (E) - offsetof(struct TYPE, pl_pathname); \
317 if ((P) > PATH_MAX || (P) < 0) \
321 #define PMCLOG_GET_CALLCHAIN_SIZE(SZ,E) do { \
322 (SZ) = ((E) - offsetof(struct pmclog_callchain, pl_pc)) \
323 / sizeof(uintfptr_t); \
326 switch (ev->pl_type = PMCLOG_HEADER_TO_TYPE(h)) {
327 case PMCLOG_TYPE_CALLCHAIN:
328 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pid);
329 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pmcid);
330 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_cpuflags);
331 PMCLOG_GET_CALLCHAIN_SIZE(ev->pl_u.pl_cc.pl_npc,evlen);
332 for (npc = 0; npc < ev->pl_u.pl_cc.pl_npc; npc++)
333 PMCLOG_READADDR(le,ev->pl_u.pl_cc.pl_pc[npc]);
334 for (;npc < PMC_CALLCHAIN_DEPTH_MAX; npc++)
335 ev->pl_u.pl_cc.pl_pc[npc] = (uintfptr_t) 0;
337 case PMCLOG_TYPE_CLOSELOG:
338 ev->pl_state = PMCLOG_EOF;
340 case PMCLOG_TYPE_DROPNOTIFY:
343 case PMCLOG_TYPE_INITIALIZE:
344 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_version);
345 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_arch);
346 ps->ps_version = ev->pl_u.pl_i.pl_version;
347 ps->ps_arch = ev->pl_u.pl_i.pl_arch;
348 ps->ps_initialized = 1;
350 case PMCLOG_TYPE_MAP_IN:
351 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_map_in);
352 PMCLOG_READ32(le,ev->pl_u.pl_mi.pl_pid);
353 PMCLOG_READADDR(le,ev->pl_u.pl_mi.pl_start);
354 PMCLOG_READSTRING(le, ev->pl_u.pl_mi.pl_pathname, pathlen);
356 case PMCLOG_TYPE_MAP_OUT:
357 PMCLOG_READ32(le,ev->pl_u.pl_mo.pl_pid);
358 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_start);
359 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_end);
361 case PMCLOG_TYPE_PCSAMPLE:
362 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pid);
363 PMCLOG_READADDR(le,ev->pl_u.pl_s.pl_pc);
364 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pmcid);
365 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_usermode);
367 case PMCLOG_TYPE_PMCALLOCATE:
368 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_pmcid);
369 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_event);
370 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_flags);
371 if ((ev->pl_u.pl_a.pl_evname =
372 _pmc_name_of_event(ev->pl_u.pl_a.pl_event, ps->ps_arch))
376 case PMCLOG_TYPE_PMCALLOCATEDYN:
377 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_pmcid);
378 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_event);
379 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_flags);
380 PMCLOG_READSTRING(le,ev->pl_u.pl_ad.pl_evname,PMC_NAME_MAX);
382 case PMCLOG_TYPE_PMCATTACH:
383 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_pmcattach);
384 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pmcid);
385 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pid);
386 PMCLOG_READSTRING(le,ev->pl_u.pl_t.pl_pathname,pathlen);
388 case PMCLOG_TYPE_PMCDETACH:
389 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pmcid);
390 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pid);
392 case PMCLOG_TYPE_PROCCSW:
393 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pmcid);
394 PMCLOG_READ64(le,ev->pl_u.pl_c.pl_value);
395 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pid);
397 case PMCLOG_TYPE_PROCEXEC:
398 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_procexec);
399 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pid);
400 PMCLOG_READADDR(le,ev->pl_u.pl_x.pl_entryaddr);
401 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pmcid);
402 PMCLOG_READSTRING(le,ev->pl_u.pl_x.pl_pathname,pathlen);
404 case PMCLOG_TYPE_PROCEXIT:
405 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pmcid);
406 PMCLOG_READ64(le,ev->pl_u.pl_e.pl_value);
407 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pid);
409 case PMCLOG_TYPE_PROCFORK:
410 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_oldpid);
411 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_newpid);
413 case PMCLOG_TYPE_SYSEXIT:
414 PMCLOG_READ32(le,ev->pl_u.pl_se.pl_pid);
416 case PMCLOG_TYPE_USERDATA:
417 PMCLOG_READ32(le,ev->pl_u.pl_u.pl_userdata);
419 default: /* unknown record type */
420 ps->ps_state = PL_STATE_ERROR;
421 ev->pl_state = PMCLOG_ERROR;
425 ev->pl_offset = (ps->ps_offset += evlen);
426 ev->pl_count = (ps->ps_count += 1);
427 ev->pl_state = PMCLOG_OK;
431 ev->pl_state = PMCLOG_ERROR;
432 ps->ps_state = PL_STATE_ERROR;
437 * Extract and return the next event from the byte stream.
439 * Returns 0 and sets the event's state to PMCLOG_OK in case an event
440 * was successfully parsed. Otherwise this function returns -1 and
441 * sets the event's state to one of PMCLOG_REQUIRE_DATA (if more data
442 * is needed) or PMCLOG_EOF (if an EOF was seen) or PMCLOG_ERROR if
443 * a parse error was encountered.
447 pmclog_read(void *cookie, struct pmclog_ev *ev)
451 struct pmclog_parse_state *ps;
453 ps = (struct pmclog_parse_state *) cookie;
455 if (ps->ps_state == PL_STATE_ERROR) {
456 ev->pl_state = PMCLOG_ERROR;
461 * If there isn't enough data left for a new event try and get
464 if (ps->ps_len == 0) {
465 ev->pl_state = PMCLOG_REQUIRE_DATA;
468 * If we have a valid file descriptor to read from, attempt
469 * to read from that. This read may return with an error,
470 * (which may be EAGAIN or other recoverable error), or
473 if (ps->ps_fd != PMCLOG_FD_NONE) {
475 nread = read(ps->ps_fd, ps->ps_buffer,
480 ev->pl_state = PMCLOG_EOF;
481 else if (errno != EAGAIN) /* not restartable */
482 ev->pl_state = PMCLOG_ERROR;
487 ps->ps_data = ps->ps_buffer;
492 assert(ps->ps_len > 0);
495 /* Retrieve one event from the byte stream. */
496 retval = pmclog_get_event(ps, &ps->ps_data, &ps->ps_len, ev);
499 * If we need more data and we have a configured fd, try read
502 if (retval < 0 && ev->pl_state == PMCLOG_REQUIRE_DATA &&
504 assert(ps->ps_len == 0);
512 * Feed data to a memory based parser.
514 * The memory area pointed to by 'data' needs to be valid till the
515 * next error return from pmclog_next_event().
519 pmclog_feed(void *cookie, char *data, int len)
521 struct pmclog_parse_state *ps;
523 ps = (struct pmclog_parse_state *) cookie;
525 if (len < 0 || /* invalid length */
526 ps->ps_buffer || /* called for a file parser */
527 ps->ps_len != 0) /* unnecessary call */
537 * Allocate and initialize parser state.
543 struct pmclog_parse_state *ps;
545 if ((ps = (struct pmclog_parse_state *) malloc(sizeof(*ps))) == NULL)
548 ps->ps_state = PL_STATE_NEW_RECORD;
550 ps->ps_initialized = 0;
552 ps->ps_offset = (off_t) 0;
553 bzero(&ps->ps_saved, sizeof(ps->ps_saved));
557 ps->ps_buffer = NULL;
560 /* allocate space for a work area */
561 if (ps->ps_fd != PMCLOG_FD_NONE) {
562 if ((ps->ps_buffer = malloc(PMCLOG_BUFFER_SIZE)) == NULL) {
573 * Free up parser state.
577 pmclog_close(void *cookie)
579 struct pmclog_parse_state *ps;
581 ps = (struct pmclog_parse_state *) cookie;