2 * Copyright (c) 2005-2007 Joseph Koshy
3 * Copyright (c) 2007 The FreeBSD Foundation
6 * Portions of this software were developed by A. Joseph Koshy under
7 * sponsorship from the FreeBSD Foundation and Google, Inc.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
36 #include <sys/pmclog.h>
48 #include <machine/pmc_mdep.h>
50 #include "libpmcinternal.h"
52 #define PMCLOG_BUFFER_SIZE 4096
57 * The pmclog(3) API is oriented towards parsing an event stream in
58 * "realtime", i.e., from an data source that may or may not preserve
59 * record boundaries -- for example when the data source is elsewhere
60 * on a network. The API allows data to be fed into the parser zero
61 * or more bytes at a time.
63 * The state for a log file parser is maintained in a 'struct
64 * pmclog_parse_state'. Parser invocations are done by calling
65 * 'pmclog_read()'; this function will inform the caller when a
66 * complete event is parsed.
68 * The parser first assembles a complete log file event in an internal
69 * work area (see "ps_saved" below). Once a complete log file event
70 * is read, the parser then parses it and converts it to an event
71 * descriptor usable by the client. We could possibly avoid this two
72 * step process by directly parsing the input log to set fields in the
73 * event record. However the parser's state machine would get
74 * insanely complicated, and this code is unlikely to be used in
75 * performance critical paths.
78 enum pmclog_parser_state {
79 PL_STATE_NEW_RECORD, /* in-between records */
80 PL_STATE_EXPECTING_HEADER, /* header being read */
81 PL_STATE_PARTIAL_RECORD, /* header present but not the record */
82 PL_STATE_ERROR /* parsing error encountered */
85 struct pmclog_parse_state {
86 enum pmclog_parser_state ps_state;
87 enum pmc_cputype ps_arch; /* log file architecture */
88 uint32_t ps_version; /* hwpmc version */
89 int ps_initialized; /* whether initialized */
90 int ps_count; /* count of records processed */
91 off_t ps_offset; /* stream byte offset */
92 union pmclog_entry ps_saved; /* saved partial log entry */
93 int ps_svcount; /* #bytes saved */
94 int ps_fd; /* active fd or -1 */
95 char *ps_buffer; /* scratch buffer if fd != -1 */
96 char *ps_data; /* current parse pointer */
97 size_t ps_len; /* length of buffered data */
100 #define PMCLOG_HEADER_FROM_SAVED_STATE(PS) \
101 (* ((uint32_t *) &(PS)->ps_saved))
103 #define PMCLOG_INITIALIZE_READER(LE,A) LE = (uint32_t *) &(A)
104 #define PMCLOG_READ32(LE,V) do { \
107 #define PMCLOG_READ64(LE,V) do { \
109 _v = (uint64_t) *(LE)++; \
110 _v |= ((uint64_t) *(LE)++) << 32; \
114 #define PMCLOG_READSTRING(LE,DST,LEN) strlcpy((DST), (char *) (LE), (LEN))
117 * Assemble a log record from '*len' octets starting from address '*data'.
118 * Update 'data' and 'len' to reflect the number of bytes consumed.
120 * '*data' is potentially an unaligned address and '*len' octets may
121 * not be enough to complete a event record.
124 static enum pmclog_parser_state
125 pmclog_get_record(struct pmclog_parse_state *ps, char **data, ssize_t *len)
127 int avail, copylen, recordsize, used;
129 const int HEADERSIZE = sizeof(uint32_t);
132 if ((avail = *len) <= 0)
133 return (ps->ps_state = PL_STATE_ERROR);
138 if (ps->ps_state == PL_STATE_NEW_RECORD)
141 dst = (char *) &ps->ps_saved + ps->ps_svcount;
143 switch (ps->ps_state) {
144 case PL_STATE_NEW_RECORD:
149 * Case A: avail < headersize
150 * -> 'expecting header'
152 * Case B: avail >= headersize
153 * B.1: avail < recordsize
154 * -> 'partial record'
155 * B.2: avail >= recordsize
159 copylen = avail < HEADERSIZE ? avail : HEADERSIZE;
160 bcopy(src, dst, copylen);
161 ps->ps_svcount = used = copylen;
163 if (copylen < HEADERSIZE) {
164 ps->ps_state = PL_STATE_EXPECTING_HEADER;
171 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
172 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
177 if (recordsize <= avail) { /* full record available */
178 bcopy(src, dst, recordsize - copylen);
179 ps->ps_svcount = used = recordsize;
183 /* header + a partial record is available */
184 bcopy(src, dst, avail - copylen);
185 ps->ps_svcount = used = avail;
186 ps->ps_state = PL_STATE_PARTIAL_RECORD;
190 case PL_STATE_EXPECTING_HEADER:
195 * Case C: avail+saved < headersize
196 * -> 'expecting header'
198 * Case D: avail+saved >= headersize
199 * D.1: avail+saved < recordsize
200 * -> 'partial record'
201 * D.2: avail+saved >= recordsize
203 * (see PARTIAL_RECORD handling below)
206 if (avail + ps->ps_svcount < HEADERSIZE) {
207 bcopy(src, dst, avail);
208 ps->ps_svcount += avail;
213 used = copylen = HEADERSIZE - ps->ps_svcount;
214 bcopy(src, dst, copylen);
218 ps->ps_svcount += copylen;
222 case PL_STATE_PARTIAL_RECORD:
227 * Case E: avail+saved < recordsize
228 * -> 'partial record'
230 * Case F: avail+saved >= recordsize
234 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
235 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
240 if (avail + ps->ps_svcount < recordsize) {
242 ps->ps_state = PL_STATE_PARTIAL_RECORD;
244 copylen = recordsize - ps->ps_svcount;
245 ps->ps_state = PL_STATE_NEW_RECORD;
248 bcopy(src, dst, copylen);
249 ps->ps_svcount += copylen;
263 ps->ps_state = PL_STATE_ERROR;
268 * Get an event from the stream pointed to by '*data'. '*len'
269 * indicates the number of bytes available to parse. Arguments
270 * '*data' and '*len' are updated to indicate the number of bytes
275 pmclog_get_event(void *cookie, char **data, ssize_t *len,
276 struct pmclog_ev *ev)
279 uint32_t h, *le, npc;
280 enum pmclog_parser_state e;
281 struct pmclog_parse_state *ps;
283 ps = (struct pmclog_parse_state *) cookie;
285 assert(ps->ps_state != PL_STATE_ERROR);
287 if ((e = pmclog_get_record(ps,data,len)) == PL_STATE_ERROR) {
288 ev->pl_state = PMCLOG_ERROR;
292 if (e != PL_STATE_NEW_RECORD) {
293 ev->pl_state = PMCLOG_REQUIRE_DATA;
297 PMCLOG_INITIALIZE_READER(le, ps->ps_saved);
301 if (!PMCLOG_HEADER_CHECK_MAGIC(h)) {
302 ps->ps_state = PL_STATE_ERROR;
303 ev->pl_state = PMCLOG_ERROR;
307 /* copy out the time stamp */
308 PMCLOG_READ32(le,ev->pl_ts.tv_sec);
309 PMCLOG_READ32(le,ev->pl_ts.tv_nsec);
311 evlen = PMCLOG_HEADER_TO_LENGTH(h);
313 #define PMCLOG_GET_PATHLEN(P,E,TYPE) do { \
314 (P) = (E) - offsetof(struct TYPE, pl_pathname); \
315 if ((P) > PATH_MAX || (P) < 0) \
319 #define PMCLOG_GET_CALLCHAIN_SIZE(SZ,E) do { \
320 (SZ) = ((E) - offsetof(struct pmclog_callchain, pl_pc)) \
321 / sizeof(uintfptr_t); \
324 switch (ev->pl_type = PMCLOG_HEADER_TO_TYPE(h)) {
325 case PMCLOG_TYPE_CALLCHAIN:
326 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pid);
327 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pmcid);
328 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_cpuflags);
329 PMCLOG_GET_CALLCHAIN_SIZE(ev->pl_u.pl_cc.pl_npc,evlen);
330 for (npc = 0; npc < ev->pl_u.pl_cc.pl_npc; npc++)
331 PMCLOG_READADDR(le,ev->pl_u.pl_cc.pl_pc[npc]);
332 for (;npc < PMC_CALLCHAIN_DEPTH_MAX; npc++)
333 ev->pl_u.pl_cc.pl_pc[npc] = (uintfptr_t) 0;
335 case PMCLOG_TYPE_CLOSELOG:
336 case PMCLOG_TYPE_DROPNOTIFY:
339 case PMCLOG_TYPE_INITIALIZE:
340 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_version);
341 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_arch);
342 ps->ps_version = ev->pl_u.pl_i.pl_version;
343 ps->ps_arch = ev->pl_u.pl_i.pl_arch;
344 ps->ps_initialized = 1;
346 case PMCLOG_TYPE_MAP_IN:
347 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_map_in);
348 PMCLOG_READ32(le,ev->pl_u.pl_mi.pl_pid);
349 PMCLOG_READADDR(le,ev->pl_u.pl_mi.pl_start);
350 PMCLOG_READSTRING(le, ev->pl_u.pl_mi.pl_pathname, pathlen);
352 case PMCLOG_TYPE_MAP_OUT:
353 PMCLOG_READ32(le,ev->pl_u.pl_mo.pl_pid);
354 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_start);
355 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_end);
357 case PMCLOG_TYPE_PCSAMPLE:
358 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pid);
359 PMCLOG_READADDR(le,ev->pl_u.pl_s.pl_pc);
360 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pmcid);
361 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_usermode);
363 case PMCLOG_TYPE_PMCALLOCATE:
364 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_pmcid);
365 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_event);
366 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_flags);
367 if ((ev->pl_u.pl_a.pl_evname =
368 _pmc_name_of_event(ev->pl_u.pl_a.pl_event, ps->ps_arch))
372 case PMCLOG_TYPE_PMCALLOCATEDYN:
373 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_pmcid);
374 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_event);
375 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_flags);
376 PMCLOG_READSTRING(le,ev->pl_u.pl_ad.pl_evname,PMC_NAME_MAX);
378 case PMCLOG_TYPE_PMCATTACH:
379 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_pmcattach);
380 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pmcid);
381 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pid);
382 PMCLOG_READSTRING(le,ev->pl_u.pl_t.pl_pathname,pathlen);
384 case PMCLOG_TYPE_PMCDETACH:
385 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pmcid);
386 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pid);
388 case PMCLOG_TYPE_PROCCSW:
389 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pmcid);
390 PMCLOG_READ64(le,ev->pl_u.pl_c.pl_value);
391 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pid);
393 case PMCLOG_TYPE_PROCEXEC:
394 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_procexec);
395 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pid);
396 PMCLOG_READADDR(le,ev->pl_u.pl_x.pl_entryaddr);
397 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pmcid);
398 PMCLOG_READSTRING(le,ev->pl_u.pl_x.pl_pathname,pathlen);
400 case PMCLOG_TYPE_PROCEXIT:
401 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pmcid);
402 PMCLOG_READ64(le,ev->pl_u.pl_e.pl_value);
403 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pid);
405 case PMCLOG_TYPE_PROCFORK:
406 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_oldpid);
407 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_newpid);
409 case PMCLOG_TYPE_SYSEXIT:
410 PMCLOG_READ32(le,ev->pl_u.pl_se.pl_pid);
412 case PMCLOG_TYPE_USERDATA:
413 PMCLOG_READ32(le,ev->pl_u.pl_u.pl_userdata);
415 default: /* unknown record type */
416 ps->ps_state = PL_STATE_ERROR;
417 ev->pl_state = PMCLOG_ERROR;
421 ev->pl_offset = (ps->ps_offset += evlen);
422 ev->pl_count = (ps->ps_count += 1);
423 ev->pl_state = PMCLOG_OK;
427 ev->pl_state = PMCLOG_ERROR;
428 ps->ps_state = PL_STATE_ERROR;
433 * Extract and return the next event from the byte stream.
435 * Returns 0 and sets the event's state to PMCLOG_OK in case an event
436 * was successfully parsed. Otherwise this function returns -1 and
437 * sets the event's state to one of PMCLOG_REQUIRE_DATA (if more data
438 * is needed) or PMCLOG_EOF (if an EOF was seen) or PMCLOG_ERROR if
439 * a parse error was encountered.
443 pmclog_read(void *cookie, struct pmclog_ev *ev)
447 struct pmclog_parse_state *ps;
449 ps = (struct pmclog_parse_state *) cookie;
451 if (ps->ps_state == PL_STATE_ERROR) {
452 ev->pl_state = PMCLOG_ERROR;
457 * If there isn't enough data left for a new event try and get
460 if (ps->ps_len == 0) {
461 ev->pl_state = PMCLOG_REQUIRE_DATA;
464 * If we have a valid file descriptor to read from, attempt
465 * to read from that. This read may return with an error,
466 * (which may be EAGAIN or other recoverable error), or
469 if (ps->ps_fd != PMCLOG_FD_NONE) {
471 nread = read(ps->ps_fd, ps->ps_buffer,
476 ev->pl_state = PMCLOG_EOF;
477 else if (errno != EAGAIN) /* not restartable */
478 ev->pl_state = PMCLOG_ERROR;
483 ps->ps_data = ps->ps_buffer;
488 assert(ps->ps_len > 0);
491 /* Retrieve one event from the byte stream. */
492 retval = pmclog_get_event(ps, &ps->ps_data, &ps->ps_len, ev);
495 * If we need more data and we have a configured fd, try read
498 if (retval < 0 && ev->pl_state == PMCLOG_REQUIRE_DATA &&
500 assert(ps->ps_len == 0);
508 * Feed data to a memory based parser.
510 * The memory area pointed to by 'data' needs to be valid till the
511 * next error return from pmclog_next_event().
515 pmclog_feed(void *cookie, char *data, int len)
517 struct pmclog_parse_state *ps;
519 ps = (struct pmclog_parse_state *) cookie;
521 if (len < 0 || /* invalid length */
522 ps->ps_buffer || /* called for a file parser */
523 ps->ps_len != 0) /* unnecessary call */
533 * Allocate and initialize parser state.
539 struct pmclog_parse_state *ps;
541 if ((ps = (struct pmclog_parse_state *) malloc(sizeof(*ps))) == NULL)
544 ps->ps_state = PL_STATE_NEW_RECORD;
546 ps->ps_initialized = 0;
548 ps->ps_offset = (off_t) 0;
549 bzero(&ps->ps_saved, sizeof(ps->ps_saved));
553 ps->ps_buffer = NULL;
556 /* allocate space for a work area */
557 if (ps->ps_fd != PMCLOG_FD_NONE) {
558 if ((ps->ps_buffer = malloc(PMCLOG_BUFFER_SIZE)) == NULL) {
569 * Free up parser state.
573 pmclog_close(void *cookie)
575 struct pmclog_parse_state *ps;
577 ps = (struct pmclog_parse_state *) cookie;