pmclog.c revision 147191
1/*- 2 * Copyright (c) 2005 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 147191 2005-06-09 19:45:09Z 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_MAPPINGCHANGE: 326 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_mappingchange); 327 PMCLOG_READ32(le,ev->pl_u.pl_m.pl_type); 328 PMCLOG_READADDR(le,ev->pl_u.pl_m.pl_start); 329 PMCLOG_READADDR(le,ev->pl_u.pl_m.pl_end); 330 PMCLOG_READ32(le,ev->pl_u.pl_m.pl_pid); 331 PMCLOG_READSTRING(le, ev->pl_u.pl_m.pl_pathname, pathlen); 332 break; 333 case PMCLOG_TYPE_PCSAMPLE: 334 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pid); 335 PMCLOG_READADDR(le,ev->pl_u.pl_s.pl_pc); 336 PMCLOG_READ32(le,ev->pl_u.pl_s.pl_pmcid); 337 break; 338 case PMCLOG_TYPE_PMCALLOCATE: 339 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_pmcid); 340 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_event); 341 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_flags); 342 if ((ev->pl_u.pl_a.pl_evname = 343 pmc_name_of_event(ev->pl_u.pl_a.pl_event)) == NULL) 344 goto error; 345 break; 346 case PMCLOG_TYPE_PMCATTACH: 347 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_pmcattach); 348 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pmcid); 349 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pid); 350 PMCLOG_READSTRING(le,ev->pl_u.pl_t.pl_pathname,pathlen); 351 break; 352 case PMCLOG_TYPE_PMCDETACH: 353 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pmcid); 354 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pid); 355 break; 356 case PMCLOG_TYPE_PROCCSW: 357 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pmcid); 358 PMCLOG_READ64(le,ev->pl_u.pl_c.pl_value); 359 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pid); 360 break; 361 case PMCLOG_TYPE_PROCEXEC: 362 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_procexec); 363 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pid); 364 PMCLOG_READSTRING(le,ev->pl_u.pl_x.pl_pathname,pathlen); 365 break; 366 case PMCLOG_TYPE_PROCEXIT: 367 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pmcid); 368 PMCLOG_READ64(le,ev->pl_u.pl_e.pl_value); 369 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pid); 370 break; 371 case PMCLOG_TYPE_PROCFORK: 372 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_oldpid); 373 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_newpid); 374 break; 375 case PMCLOG_TYPE_SYSEXIT: 376 PMCLOG_READ32(le,ev->pl_u.pl_se.pl_pid); 377 break; 378 case PMCLOG_TYPE_USERDATA: 379 PMCLOG_READ32(le,ev->pl_u.pl_u.pl_userdata); 380 break; 381 default: /* unknown record type */ 382 ps->ps_state = PL_STATE_ERROR; 383 ev->pl_state = PMCLOG_ERROR; 384 return -1; 385 } 386 387 ev->pl_offset = (ps->ps_offset += evlen); 388 ev->pl_count = (ps->ps_count += 1); 389 ev->pl_state = PMCLOG_OK; 390 return 0; 391 392 error: 393 ev->pl_state = PMCLOG_ERROR; 394 ps->ps_state = PL_STATE_ERROR; 395 return -1; 396} 397 398/* 399 * Extract and return the next event from the byte stream. 400 * 401 * Returns 0 and sets the event's state to PMCLOG_OK in case an event 402 * was successfully parsed. Otherwise this function returns -1 and 403 * sets the event's state to one of PMCLOG_REQUIRE_DATA (if more data 404 * is needed) or PMCLOG_EOF (if an EOF was seen) or PMCLOG_ERROR if 405 * a parse error was encountered. 406 */ 407 408int 409pmclog_read(void *cookie, struct pmclog_ev *ev) 410{ 411 ssize_t nread; 412 struct pmclog_parse_state *ps; 413 414 ps = (struct pmclog_parse_state *) cookie; 415 416 if (ps->ps_state == PL_STATE_ERROR) { 417 ev->pl_state = PMCLOG_ERROR; 418 return -1; 419 } 420 421 /* 422 * If there isn't enough data left for a new event try and get 423 * more data. 424 */ 425 if (ps->ps_len == 0) { 426 ev->pl_state = PMCLOG_REQUIRE_DATA; 427 428 /* 429 * If we have a valid file descriptor to read from, attempt 430 * to read from that. This read may return with an error, 431 * (which may be EAGAIN or other recoverable error), or 432 * can return EOF. 433 */ 434 if (ps->ps_fd != PMCLOG_FD_NONE) { 435 nread = read(ps->ps_fd, ps->ps_buffer, 436 PMCLOG_BUFFER_SIZE); 437 438 if (nread <= 0) { 439 ev->pl_state = nread < 0 ? PMCLOG_ERROR : 440 PMCLOG_EOF; 441 return -1; 442 } 443 444 ps->ps_len = nread; 445 ps->ps_data = ps->ps_buffer; 446 } else 447 return -1; 448 } 449 450 assert(ps->ps_len > 0); 451 452 /* 453 * Retrieve one event from the byte stream. 454 */ 455 return pmclog_get_event(ps, &ps->ps_data, &ps->ps_len, ev); 456} 457 458/* 459 * Feed data to a memory based parser. 460 * 461 * The memory area pointed to by 'data' needs to be valid till the 462 * next error return from pmclog_next_event(). 463 */ 464 465int 466pmclog_feed(void *cookie, char *data, int len) 467{ 468 struct pmclog_parse_state *ps; 469 470 ps = (struct pmclog_parse_state *) cookie; 471 472 if (len < 0 || /* invalid length */ 473 ps->ps_buffer || /* called for a file parser */ 474 ps->ps_len != 0) /* unnecessary call */ 475 return -1; 476 477 ps->ps_data = data; 478 ps->ps_len = len; 479 480 return 0; 481} 482 483/* 484 * Allocate and initialize parser state. 485 */ 486 487void * 488pmclog_open(int fd) 489{ 490 struct pmclog_parse_state *ps; 491 492 if ((ps = (struct pmclog_parse_state *) malloc(sizeof(*ps))) == NULL) 493 return NULL; 494 495 ps->ps_state = PL_STATE_NEW_RECORD; 496 ps->ps_arch = -1; 497 ps->ps_initialized = 0; 498 ps->ps_count = 0; 499 ps->ps_offset = (off_t) 0; 500 bzero(&ps->ps_saved, sizeof(ps->ps_saved)); 501 ps->ps_svcount = 0; 502 ps->ps_fd = fd; 503 ps->ps_data = NULL; 504 ps->ps_buffer = NULL; 505 ps->ps_len = 0; 506 507 /* allocate space for a work area */ 508 if (ps->ps_fd != PMCLOG_FD_NONE) { 509 if ((ps->ps_buffer = malloc(PMCLOG_BUFFER_SIZE)) == NULL) 510 return NULL; 511 } 512 513 return ps; 514} 515 516 517/* 518 * Free up parser state. 519 */ 520 521void 522pmclog_close(void *cookie) 523{ 524 struct pmclog_parse_state *ps; 525 526 ps = (struct pmclog_parse_state *) cookie; 527 528 if (ps->ps_buffer) 529 free(ps->ps_buffer); 530 531 free(ps); 532} 533