1 /* 2 * Copyright (c) 1997, 1998 Kenneth D. Merry. 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 * 3. The name of the author may not be used to endorse or promote products 14 * derived from this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/types.h> 33 #include <sys/sysctl.h> 34 #include <sys/errno.h> 35 #include <sys/resource.h> 36 #include <sys/queue.h> 37 38 #include <ctype.h> 39 #include <err.h> 40 #include <fcntl.h> 41 #include <limits.h> 42 #include <stdio.h> 43 #include <stdlib.h> 44 #include <string.h> 45 #include <stdarg.h> 46 #include <kvm.h> 47 48 #include "devstat.h" 49 50 int 51 compute_stats(struct devstat *current, struct devstat *previous, 52 long double etime, u_int64_t *total_bytes, 53 u_int64_t *total_transfers, u_int64_t *total_blocks, 54 long double *kb_per_transfer, long double *transfers_per_second, 55 long double *mb_per_second, long double *blocks_per_second, 56 long double *ms_per_transaction); 57 58 typedef enum { 59 DEVSTAT_ARG_NOTYPE, 60 DEVSTAT_ARG_UINT64, 61 DEVSTAT_ARG_LD, 62 DEVSTAT_ARG_SKIP 63 } devstat_arg_type; 64 65 char devstat_errbuf[DEVSTAT_ERRBUF_SIZE]; 66 67 /* 68 * Table to match descriptive strings with device types. These are in 69 * order from most common to least common to speed search time. 70 */ 71 struct devstat_match_table match_table[] = { 72 {"da", DEVSTAT_TYPE_DIRECT, DEVSTAT_MATCH_TYPE}, 73 {"cd", DEVSTAT_TYPE_CDROM, DEVSTAT_MATCH_TYPE}, 74 {"scsi", DEVSTAT_TYPE_IF_SCSI, DEVSTAT_MATCH_IF}, 75 {"ide", DEVSTAT_TYPE_IF_IDE, DEVSTAT_MATCH_IF}, 76 {"other", DEVSTAT_TYPE_IF_OTHER, DEVSTAT_MATCH_IF}, 77 {"worm", DEVSTAT_TYPE_WORM, DEVSTAT_MATCH_TYPE}, 78 {"sa", DEVSTAT_TYPE_SEQUENTIAL,DEVSTAT_MATCH_TYPE}, 79 {"pass", DEVSTAT_TYPE_PASS, DEVSTAT_MATCH_PASS}, 80 {"optical", DEVSTAT_TYPE_OPTICAL, DEVSTAT_MATCH_TYPE}, 81 {"array", DEVSTAT_TYPE_STORARRAY, DEVSTAT_MATCH_TYPE}, 82 {"changer", DEVSTAT_TYPE_CHANGER, DEVSTAT_MATCH_TYPE}, 83 {"scanner", DEVSTAT_TYPE_SCANNER, DEVSTAT_MATCH_TYPE}, 84 {"printer", DEVSTAT_TYPE_PRINTER, DEVSTAT_MATCH_TYPE}, 85 {"floppy", DEVSTAT_TYPE_FLOPPY, DEVSTAT_MATCH_TYPE}, 86 {"proc", DEVSTAT_TYPE_PROCESSOR, DEVSTAT_MATCH_TYPE}, 87 {"comm", DEVSTAT_TYPE_COMM, DEVSTAT_MATCH_TYPE}, 88 {"enclosure", DEVSTAT_TYPE_ENCLOSURE, DEVSTAT_MATCH_TYPE}, 89 {NULL, 0, 0} 90 }; 91 92 struct devstat_args { 93 devstat_metric metric; 94 devstat_arg_type argtype; 95 } devstat_arg_list[] = { 96 { DSM_NONE, DEVSTAT_ARG_NOTYPE }, 97 { DSM_TOTAL_BYTES, DEVSTAT_ARG_UINT64 }, 98 { DSM_TOTAL_BYTES_READ, DEVSTAT_ARG_UINT64 }, 99 { DSM_TOTAL_BYTES_WRITE, DEVSTAT_ARG_UINT64 }, 100 { DSM_TOTAL_TRANSFERS, DEVSTAT_ARG_UINT64 }, 101 { DSM_TOTAL_TRANSFERS_READ, DEVSTAT_ARG_UINT64 }, 102 { DSM_TOTAL_TRANSFERS_WRITE, DEVSTAT_ARG_UINT64 }, 103 { DSM_TOTAL_TRANSFERS_OTHER, DEVSTAT_ARG_UINT64 }, 104 { DSM_TOTAL_BLOCKS, DEVSTAT_ARG_UINT64 }, 105 { DSM_TOTAL_BLOCKS_READ, DEVSTAT_ARG_UINT64 }, 106 { DSM_TOTAL_BLOCKS_WRITE, DEVSTAT_ARG_UINT64 }, 107 { DSM_KB_PER_TRANSFER, DEVSTAT_ARG_LD }, 108 { DSM_KB_PER_TRANSFER_READ, DEVSTAT_ARG_LD }, 109 { DSM_KB_PER_TRANSFER_WRITE, DEVSTAT_ARG_LD }, 110 { DSM_TRANSFERS_PER_SECOND, DEVSTAT_ARG_LD }, 111 { DSM_TRANSFERS_PER_SECOND_READ, DEVSTAT_ARG_LD }, 112 { DSM_TRANSFERS_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, 113 { DSM_TRANSFERS_PER_SECOND_OTHER, DEVSTAT_ARG_LD }, 114 { DSM_MB_PER_SECOND, DEVSTAT_ARG_LD }, 115 { DSM_MB_PER_SECOND_READ, DEVSTAT_ARG_LD }, 116 { DSM_MB_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, 117 { DSM_BLOCKS_PER_SECOND, DEVSTAT_ARG_LD }, 118 { DSM_BLOCKS_PER_SECOND_READ, DEVSTAT_ARG_LD }, 119 { DSM_BLOCKS_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, 120 { DSM_MS_PER_TRANSACTION, DEVSTAT_ARG_LD }, 121 { DSM_MS_PER_TRANSACTION_READ, DEVSTAT_ARG_LD }, 122 { DSM_MS_PER_TRANSACTION_WRITE, DEVSTAT_ARG_LD }, 123 { DSM_SKIP, DEVSTAT_ARG_SKIP }, 124 { DSM_TOTAL_BYTES_FREE, DEVSTAT_ARG_UINT64 }, 125 { DSM_TOTAL_TRANSFERS_FREE, DEVSTAT_ARG_UINT64 }, 126 { DSM_TOTAL_BLOCKS_FREE, DEVSTAT_ARG_UINT64 }, 127 { DSM_KB_PER_TRANSFER_FREE, DEVSTAT_ARG_LD }, 128 { DSM_MB_PER_SECOND_FREE, DEVSTAT_ARG_LD }, 129 { DSM_TRANSFERS_PER_SECOND_FREE, DEVSTAT_ARG_LD }, 130 { DSM_BLOCKS_PER_SECOND_FREE, DEVSTAT_ARG_LD }, 131 { DSM_MS_PER_TRANSACTION_OTHER, DEVSTAT_ARG_LD }, 132 { DSM_MS_PER_TRANSACTION_FREE, DEVSTAT_ARG_LD }, 133 { DSM_BUSY_PCT, DEVSTAT_ARG_LD }, 134 { DSM_QUEUE_LENGTH, DEVSTAT_ARG_UINT64 }, 135 }; 136 137 static const char *namelist[] = { 138 #define X_NUMDEVS 0 139 "_devstat_num_devs", 140 #define X_GENERATION 1 141 "_devstat_generation", 142 #define X_VERSION 2 143 "_devstat_version", 144 #define X_DEVICE_STATQ 3 145 "_device_statq", 146 #define X_END 4 147 }; 148 149 /* 150 * Local function declarations. 151 */ 152 static int compare_select(const void *arg1, const void *arg2); 153 static int readkmem(kvm_t *kd, unsigned long addr, void *buf, size_t nbytes); 154 static int readkmem_nl(kvm_t *kd, const char *name, void *buf, size_t nbytes); 155 static char *get_devstat_kvm(kvm_t *kd); 156 157 #define KREADNL(kd, var, val) \ 158 readkmem_nl(kd, namelist[var], &val, sizeof(val)) 159 160 int 161 devstat_getnumdevs(kvm_t *kd) 162 { 163 size_t numdevsize; 164 int numdevs; 165 const char *func_name = "devstat_getnumdevs"; 166 167 numdevsize = sizeof(int); 168 169 /* 170 * Find out how many devices we have in the system. 171 */ 172 if (kd == NULL) { 173 if (sysctlbyname("kern.devstat.numdevs", &numdevs, 174 &numdevsize, NULL, 0) == -1) { 175 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 176 "%s: error getting number of devices\n" 177 "%s: %s", func_name, func_name, 178 strerror(errno)); 179 return(-1); 180 } else 181 return(numdevs); 182 } else { 183 184 if (KREADNL(kd, X_NUMDEVS, numdevs) == -1) 185 return(-1); 186 else 187 return(numdevs); 188 } 189 } 190 191 /* 192 * This is an easy way to get the generation number, but the generation is 193 * supplied in a more atmoic manner by the kern.devstat.all sysctl. 194 * Because this generation sysctl is separate from the statistics sysctl, 195 * the device list and the generation could change between the time that 196 * this function is called and the device list is retreived. 197 */ 198 long 199 devstat_getgeneration(kvm_t *kd) 200 { 201 size_t gensize; 202 long generation; 203 const char *func_name = "devstat_getgeneration"; 204 205 gensize = sizeof(long); 206 207 /* 208 * Get the current generation number. 209 */ 210 if (kd == NULL) { 211 if (sysctlbyname("kern.devstat.generation", &generation, 212 &gensize, NULL, 0) == -1) { 213 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 214 "%s: error getting devstat generation\n%s: %s", 215 func_name, func_name, strerror(errno)); 216 return(-1); 217 } else 218 return(generation); 219 } else { 220 if (KREADNL(kd, X_GENERATION, generation) == -1) 221 return(-1); 222 else 223 return(generation); 224 } 225 } 226 227 /* 228 * Get the current devstat version. The return value of this function 229 * should be compared with DEVSTAT_VERSION, which is defined in 230 * sys/devicestat.h. This will enable userland programs to determine 231 * whether they are out of sync with the kernel. 232 */ 233 int 234 devstat_getversion(kvm_t *kd) 235 { 236 size_t versize; 237 int version; 238 const char *func_name = "devstat_getversion"; 239 240 versize = sizeof(int); 241 242 /* 243 * Get the current devstat version. 244 */ 245 if (kd == NULL) { 246 if (sysctlbyname("kern.devstat.version", &version, &versize, 247 NULL, 0) == -1) { 248 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 249 "%s: error getting devstat version\n%s: %s", 250 func_name, func_name, strerror(errno)); 251 return(-1); 252 } else 253 return(version); 254 } else { 255 if (KREADNL(kd, X_VERSION, version) == -1) 256 return(-1); 257 else 258 return(version); 259 } 260 } 261 262 /* 263 * Check the devstat version we know about against the devstat version the 264 * kernel knows about. If they don't match, print an error into the 265 * devstat error buffer, and return -1. If they match, return 0. 266 */ 267 int 268 devstat_checkversion(kvm_t *kd) 269 { 270 const char *func_name = "devstat_checkversion"; 271 int buflen, res, retval = 0, version; 272 273 version = devstat_getversion(kd); 274 275 if (version != DEVSTAT_VERSION) { 276 /* 277 * If getversion() returns an error (i.e. -1), then it 278 * has printed an error message in the buffer. Therefore, 279 * we need to add a \n to the end of that message before we 280 * print our own message in the buffer. 281 */ 282 if (version == -1) 283 buflen = strlen(devstat_errbuf); 284 else 285 buflen = 0; 286 287 res = snprintf(devstat_errbuf + buflen, 288 DEVSTAT_ERRBUF_SIZE - buflen, 289 "%s%s: userland devstat version %d is not " 290 "the same as the kernel\n%s: devstat " 291 "version %d\n", version == -1 ? "\n" : "", 292 func_name, DEVSTAT_VERSION, func_name, version); 293 294 if (res < 0) 295 devstat_errbuf[buflen] = '\0'; 296 297 buflen = strlen(devstat_errbuf); 298 if (version < DEVSTAT_VERSION) 299 res = snprintf(devstat_errbuf + buflen, 300 DEVSTAT_ERRBUF_SIZE - buflen, 301 "%s: libdevstat newer than kernel\n", 302 func_name); 303 else 304 res = snprintf(devstat_errbuf + buflen, 305 DEVSTAT_ERRBUF_SIZE - buflen, 306 "%s: kernel newer than libdevstat\n", 307 func_name); 308 309 if (res < 0) 310 devstat_errbuf[buflen] = '\0'; 311 312 retval = -1; 313 } 314 315 return(retval); 316 } 317 318 /* 319 * Get the current list of devices and statistics, and the current 320 * generation number. 321 * 322 * Return values: 323 * -1 -- error 324 * 0 -- device list is unchanged 325 * 1 -- device list has changed 326 */ 327 int 328 devstat_getdevs(kvm_t *kd, struct statinfo *stats) 329 { 330 int error; 331 size_t dssize; 332 int oldnumdevs; 333 long oldgeneration; 334 int retval = 0; 335 struct devinfo *dinfo; 336 const char *func_name = "devstat_getdevs"; 337 struct timespec ts; 338 339 dinfo = stats->dinfo; 340 341 if (dinfo == NULL) { 342 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 343 "%s: stats->dinfo was NULL", func_name); 344 return(-1); 345 } 346 347 oldnumdevs = dinfo->numdevs; 348 oldgeneration = dinfo->generation; 349 350 clock_gettime(CLOCK_MONOTONIC, &ts); 351 stats->snap_time = ts.tv_sec + ts.tv_nsec * 1e-9; 352 353 if (kd == NULL) { 354 /* If this is our first time through, mem_ptr will be null. */ 355 if (dinfo->mem_ptr == NULL) { 356 /* 357 * Get the number of devices. If it's negative, it's an 358 * error. Don't bother setting the error string, since 359 * getnumdevs() has already done that for us. 360 */ 361 if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) 362 return(-1); 363 364 /* 365 * The kern.devstat.all sysctl returns the current 366 * generation number, as well as all the devices. 367 * So we need four bytes more. 368 */ 369 dssize = (dinfo->numdevs * sizeof(struct devstat)) + 370 sizeof(long); 371 dinfo->mem_ptr = (u_int8_t *)malloc(dssize); 372 } else 373 dssize = (dinfo->numdevs * sizeof(struct devstat)) + 374 sizeof(long); 375 376 /* 377 * Request all of the devices. We only really allow for one 378 * ENOMEM failure. It would, of course, be possible to just go 379 * in a loop and keep reallocing the device structure until we 380 * don't get ENOMEM back. I'm not sure it's worth it, though. 381 * If devices are being added to the system that quickly, maybe 382 * the user can just wait until all devices are added. 383 */ 384 for (;;) { 385 error = sysctlbyname("kern.devstat.all", 386 dinfo->mem_ptr, 387 &dssize, NULL, 0); 388 if (error != -1 || errno != EBUSY) 389 break; 390 } 391 if (error == -1) { 392 /* 393 * If we get ENOMEM back, that means that there are 394 * more devices now, so we need to allocate more 395 * space for the device array. 396 */ 397 if (errno == ENOMEM) { 398 /* 399 * No need to set the error string here, 400 * devstat_getnumdevs() will do that if it fails. 401 */ 402 if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) 403 return(-1); 404 405 dssize = (dinfo->numdevs * 406 sizeof(struct devstat)) + sizeof(long); 407 dinfo->mem_ptr = (u_int8_t *) 408 realloc(dinfo->mem_ptr, dssize); 409 if ((error = sysctlbyname("kern.devstat.all", 410 dinfo->mem_ptr, &dssize, NULL, 0)) == -1) { 411 snprintf(devstat_errbuf, 412 sizeof(devstat_errbuf), 413 "%s: error getting device " 414 "stats\n%s: %s", func_name, 415 func_name, strerror(errno)); 416 return(-1); 417 } 418 } else { 419 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 420 "%s: error getting device stats\n" 421 "%s: %s", func_name, func_name, 422 strerror(errno)); 423 return(-1); 424 } 425 } 426 427 } else { 428 /* 429 * This is of course non-atomic, but since we are working 430 * on a core dump, the generation is unlikely to change 431 */ 432 if ((dinfo->numdevs = devstat_getnumdevs(kd)) == -1) 433 return(-1); 434 if ((dinfo->mem_ptr = get_devstat_kvm(kd)) == NULL) 435 return(-1); 436 } 437 /* 438 * The sysctl spits out the generation as the first four bytes, 439 * then all of the device statistics structures. 440 */ 441 dinfo->generation = *(long *)dinfo->mem_ptr; 442 443 /* 444 * If the generation has changed, and if the current number of 445 * devices is not the same as the number of devices recorded in the 446 * devinfo structure, it is likely that the device list has shrunk. 447 * The reason that it is likely that the device list has shrunk in 448 * this case is that if the device list has grown, the sysctl above 449 * will return an ENOMEM error, and we will reset the number of 450 * devices and reallocate the device array. If the second sysctl 451 * fails, we will return an error and therefore never get to this 452 * point. If the device list has shrunk, the sysctl will not 453 * return an error since we have more space allocated than is 454 * necessary. So, in the shrinkage case, we catch it here and 455 * reallocate the array so that we don't use any more space than is 456 * necessary. 457 */ 458 if (oldgeneration != dinfo->generation) { 459 if (devstat_getnumdevs(kd) != dinfo->numdevs) { 460 if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) 461 return(-1); 462 dssize = (dinfo->numdevs * sizeof(struct devstat)) + 463 sizeof(long); 464 dinfo->mem_ptr = (u_int8_t *)realloc(dinfo->mem_ptr, 465 dssize); 466 } 467 retval = 1; 468 } 469 470 dinfo->devices = (struct devstat *)(dinfo->mem_ptr + sizeof(long)); 471 472 return(retval); 473 } 474 475 /* 476 * selectdevs(): 477 * 478 * Devices are selected/deselected based upon the following criteria: 479 * - devices specified by the user on the command line 480 * - devices matching any device type expressions given on the command line 481 * - devices with the highest I/O, if 'top' mode is enabled 482 * - the first n unselected devices in the device list, if maxshowdevs 483 * devices haven't already been selected and if the user has not 484 * specified any devices on the command line and if we're in "add" mode. 485 * 486 * Input parameters: 487 * - device selection list (dev_select) 488 * - current number of devices selected (num_selected) 489 * - total number of devices in the selection list (num_selections) 490 * - devstat generation as of the last time selectdevs() was called 491 * (select_generation) 492 * - current devstat generation (current_generation) 493 * - current list of devices and statistics (devices) 494 * - number of devices in the current device list (numdevs) 495 * - compiled version of the command line device type arguments (matches) 496 * - This is optional. If the number of devices is 0, this will be ignored. 497 * - The matching code pays attention to the current selection mode. So 498 * if you pass in a matching expression, it will be evaluated based 499 * upon the selection mode that is passed in. See below for details. 500 * - number of device type matching expressions (num_matches) 501 * - Set to 0 to disable the matching code. 502 * - list of devices specified on the command line by the user (dev_selections) 503 * - number of devices selected on the command line by the user 504 * (num_dev_selections) 505 * - Our selection mode. There are four different selection modes: 506 * - add mode. (DS_SELECT_ADD) Any devices matching devices explicitly 507 * selected by the user or devices matching a pattern given by the 508 * user will be selected in addition to devices that are already 509 * selected. Additional devices will be selected, up to maxshowdevs 510 * number of devices. 511 * - only mode. (DS_SELECT_ONLY) Only devices matching devices 512 * explicitly given by the user or devices matching a pattern 513 * given by the user will be selected. No other devices will be 514 * selected. 515 * - addonly mode. (DS_SELECT_ADDONLY) This is similar to add and 516 * only. Basically, this will not de-select any devices that are 517 * current selected, as only mode would, but it will also not 518 * gratuitously select up to maxshowdevs devices as add mode would. 519 * - remove mode. (DS_SELECT_REMOVE) Any devices matching devices 520 * explicitly selected by the user or devices matching a pattern 521 * given by the user will be de-selected. 522 * - maximum number of devices we can select (maxshowdevs) 523 * - flag indicating whether or not we're in 'top' mode (perf_select) 524 * 525 * Output data: 526 * - the device selection list may be modified and passed back out 527 * - the number of devices selected and the total number of items in the 528 * device selection list may be changed 529 * - the selection generation may be changed to match the current generation 530 * 531 * Return values: 532 * -1 -- error 533 * 0 -- selected devices are unchanged 534 * 1 -- selected devices changed 535 */ 536 int 537 devstat_selectdevs(struct device_selection **dev_select, int *num_selected, 538 int *num_selections, long *select_generation, 539 long current_generation, struct devstat *devices, 540 int numdevs, struct devstat_match *matches, int num_matches, 541 char **dev_selections, int num_dev_selections, 542 devstat_select_mode select_mode, int maxshowdevs, 543 int perf_select) 544 { 545 int i, j, k; 546 int init_selections = 0, init_selected_var = 0; 547 struct device_selection *old_dev_select = NULL; 548 int old_num_selections = 0, old_num_selected; 549 int selection_number = 0; 550 int changed = 0, found = 0; 551 552 if ((dev_select == NULL) || (devices == NULL) || (numdevs < 0)) 553 return(-1); 554 555 /* 556 * We always want to make sure that we have as many dev_select 557 * entries as there are devices. 558 */ 559 /* 560 * In this case, we haven't selected devices before. 561 */ 562 if (*dev_select == NULL) { 563 *dev_select = (struct device_selection *)malloc(numdevs * 564 sizeof(struct device_selection)); 565 *select_generation = current_generation; 566 init_selections = 1; 567 changed = 1; 568 /* 569 * In this case, we have selected devices before, but the device 570 * list has changed since we last selected devices, so we need to 571 * either enlarge or reduce the size of the device selection list. 572 */ 573 } else if (*num_selections != numdevs) { 574 *dev_select = (struct device_selection *)realloc(*dev_select, 575 numdevs * sizeof(struct device_selection)); 576 *select_generation = current_generation; 577 init_selections = 1; 578 /* 579 * In this case, we've selected devices before, and the selection 580 * list is the same size as it was the last time, but the device 581 * list has changed. 582 */ 583 } else if (*select_generation < current_generation) { 584 *select_generation = current_generation; 585 init_selections = 1; 586 } 587 588 /* 589 * If we're in "only" mode, we want to clear out the selected 590 * variable since we're going to select exactly what the user wants 591 * this time through. 592 */ 593 if (select_mode == DS_SELECT_ONLY) 594 init_selected_var = 1; 595 596 /* 597 * In all cases, we want to back up the number of selected devices. 598 * It is a quick and accurate way to determine whether the selected 599 * devices have changed. 600 */ 601 old_num_selected = *num_selected; 602 603 /* 604 * We want to make a backup of the current selection list if 605 * the list of devices has changed, or if we're in performance 606 * selection mode. In both cases, we don't want to make a backup 607 * if we already know for sure that the list will be different. 608 * This is certainly the case if this is our first time through the 609 * selection code. 610 */ 611 if (((init_selected_var != 0) || (init_selections != 0) 612 || (perf_select != 0)) && (changed == 0)){ 613 old_dev_select = (struct device_selection *)malloc( 614 *num_selections * sizeof(struct device_selection)); 615 old_num_selections = *num_selections; 616 bcopy(*dev_select, old_dev_select, 617 sizeof(struct device_selection) * *num_selections); 618 } 619 620 if (init_selections != 0) { 621 bzero(*dev_select, sizeof(struct device_selection) * numdevs); 622 623 for (i = 0; i < numdevs; i++) { 624 (*dev_select)[i].device_number = 625 devices[i].device_number; 626 strncpy((*dev_select)[i].device_name, 627 devices[i].device_name, 628 DEVSTAT_NAME_LEN); 629 (*dev_select)[i].device_name[DEVSTAT_NAME_LEN - 1]='\0'; 630 (*dev_select)[i].unit_number = devices[i].unit_number; 631 (*dev_select)[i].position = i; 632 } 633 *num_selections = numdevs; 634 } else if (init_selected_var != 0) { 635 for (i = 0; i < numdevs; i++) 636 (*dev_select)[i].selected = 0; 637 } 638 639 /* we haven't gotten around to selecting anything yet.. */ 640 if ((select_mode == DS_SELECT_ONLY) || (init_selections != 0) 641 || (init_selected_var != 0)) 642 *num_selected = 0; 643 644 /* 645 * Look through any devices the user specified on the command line 646 * and see if they match known devices. If so, select them. 647 */ 648 for (i = 0; (i < *num_selections) && (num_dev_selections > 0); i++) { 649 char tmpstr[80]; 650 651 snprintf(tmpstr, sizeof(tmpstr), "%s%d", 652 (*dev_select)[i].device_name, 653 (*dev_select)[i].unit_number); 654 for (j = 0; j < num_dev_selections; j++) { 655 if (strcmp(tmpstr, dev_selections[j]) == 0) { 656 /* 657 * Here we do different things based on the 658 * mode we're in. If we're in add or 659 * addonly mode, we only select this device 660 * if it hasn't already been selected. 661 * Otherwise, we would be unnecessarily 662 * changing the selection order and 663 * incrementing the selection count. If 664 * we're in only mode, we unconditionally 665 * select this device, since in only mode 666 * any previous selections are erased and 667 * manually specified devices are the first 668 * ones to be selected. If we're in remove 669 * mode, we de-select the specified device and 670 * decrement the selection count. 671 */ 672 switch(select_mode) { 673 case DS_SELECT_ADD: 674 case DS_SELECT_ADDONLY: 675 if ((*dev_select)[i].selected) 676 break; 677 /* FALLTHROUGH */ 678 case DS_SELECT_ONLY: 679 (*dev_select)[i].selected = 680 ++selection_number; 681 (*num_selected)++; 682 break; 683 case DS_SELECT_REMOVE: 684 (*dev_select)[i].selected = 0; 685 (*num_selected)--; 686 /* 687 * This isn't passed back out, we 688 * just use it to keep track of 689 * how many devices we've removed. 690 */ 691 num_dev_selections--; 692 break; 693 } 694 break; 695 } 696 } 697 } 698 699 /* 700 * Go through the user's device type expressions and select devices 701 * accordingly. We only do this if the number of devices already 702 * selected is less than the maximum number we can show. 703 */ 704 for (i = 0; (i < num_matches) && (*num_selected < maxshowdevs); i++) { 705 /* We should probably indicate some error here */ 706 if ((matches[i].match_fields == DEVSTAT_MATCH_NONE) 707 || (matches[i].num_match_categories <= 0)) 708 continue; 709 710 for (j = 0; j < numdevs; j++) { 711 int num_match_categories; 712 713 num_match_categories = matches[i].num_match_categories; 714 715 /* 716 * Determine whether or not the current device 717 * matches the given matching expression. This if 718 * statement consists of three components: 719 * - the device type check 720 * - the device interface check 721 * - the passthrough check 722 * If a the matching test is successful, it 723 * decrements the number of matching categories, 724 * and if we've reached the last element that 725 * needed to be matched, the if statement succeeds. 726 * 727 */ 728 if ((((matches[i].match_fields & DEVSTAT_MATCH_TYPE)!=0) 729 && ((devices[j].device_type & DEVSTAT_TYPE_MASK) == 730 (matches[i].device_type & DEVSTAT_TYPE_MASK)) 731 &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) 732 || (((matches[i].match_fields & 733 DEVSTAT_MATCH_PASS) == 0) 734 && ((devices[j].device_type & 735 DEVSTAT_TYPE_PASS) == 0))) 736 && (--num_match_categories == 0)) 737 || (((matches[i].match_fields & DEVSTAT_MATCH_IF) != 0) 738 && ((devices[j].device_type & DEVSTAT_TYPE_IF_MASK) == 739 (matches[i].device_type & DEVSTAT_TYPE_IF_MASK)) 740 &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) 741 || (((matches[i].match_fields & 742 DEVSTAT_MATCH_PASS) == 0) 743 && ((devices[j].device_type & 744 DEVSTAT_TYPE_PASS) == 0))) 745 && (--num_match_categories == 0)) 746 || (((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) 747 && ((devices[j].device_type & DEVSTAT_TYPE_PASS) != 0) 748 && (--num_match_categories == 0))) { 749 750 /* 751 * This is probably a non-optimal solution 752 * to the problem that the devices in the 753 * device list will not be in the same 754 * order as the devices in the selection 755 * array. 756 */ 757 for (k = 0; k < numdevs; k++) { 758 if ((*dev_select)[k].position == j) { 759 found = 1; 760 break; 761 } 762 } 763 764 /* 765 * There shouldn't be a case where a device 766 * in the device list is not in the 767 * selection list...but it could happen. 768 */ 769 if (found != 1) { 770 fprintf(stderr, "selectdevs: couldn't" 771 " find %s%d in selection " 772 "list\n", 773 devices[j].device_name, 774 devices[j].unit_number); 775 break; 776 } 777 778 /* 779 * We do different things based upon the 780 * mode we're in. If we're in add or only 781 * mode, we go ahead and select this device 782 * if it hasn't already been selected. If 783 * it has already been selected, we leave 784 * it alone so we don't mess up the 785 * selection ordering. Manually specified 786 * devices have already been selected, and 787 * they have higher priority than pattern 788 * matched devices. If we're in remove 789 * mode, we de-select the given device and 790 * decrement the selected count. 791 */ 792 switch(select_mode) { 793 case DS_SELECT_ADD: 794 case DS_SELECT_ADDONLY: 795 case DS_SELECT_ONLY: 796 if ((*dev_select)[k].selected != 0) 797 break; 798 (*dev_select)[k].selected = 799 ++selection_number; 800 (*num_selected)++; 801 break; 802 case DS_SELECT_REMOVE: 803 (*dev_select)[k].selected = 0; 804 (*num_selected)--; 805 break; 806 } 807 } 808 } 809 } 810 811 /* 812 * Here we implement "top" mode. Devices are sorted in the 813 * selection array based on two criteria: whether or not they are 814 * selected (not selection number, just the fact that they are 815 * selected!) and the number of bytes in the "bytes" field of the 816 * selection structure. The bytes field generally must be kept up 817 * by the user. In the future, it may be maintained by library 818 * functions, but for now the user has to do the work. 819 * 820 * At first glance, it may seem wrong that we don't go through and 821 * select every device in the case where the user hasn't specified 822 * any devices or patterns. In fact, though, it won't make any 823 * difference in the device sorting. In that particular case (i.e. 824 * when we're in "add" or "only" mode, and the user hasn't 825 * specified anything) the first time through no devices will be 826 * selected, so the only criterion used to sort them will be their 827 * performance. The second time through, and every time thereafter, 828 * all devices will be selected, so again selection won't matter. 829 */ 830 if (perf_select != 0) { 831 832 /* Sort the device array by throughput */ 833 qsort(*dev_select, *num_selections, 834 sizeof(struct device_selection), 835 compare_select); 836 837 if (*num_selected == 0) { 838 /* 839 * Here we select every device in the array, if it 840 * isn't already selected. Because the 'selected' 841 * variable in the selection array entries contains 842 * the selection order, the devstats routine can show 843 * the devices that were selected first. 844 */ 845 for (i = 0; i < *num_selections; i++) { 846 if ((*dev_select)[i].selected == 0) { 847 (*dev_select)[i].selected = 848 ++selection_number; 849 (*num_selected)++; 850 } 851 } 852 } else { 853 selection_number = 0; 854 for (i = 0; i < *num_selections; i++) { 855 if ((*dev_select)[i].selected != 0) { 856 (*dev_select)[i].selected = 857 ++selection_number; 858 } 859 } 860 } 861 } 862 863 /* 864 * If we're in the "add" selection mode and if we haven't already 865 * selected maxshowdevs number of devices, go through the array and 866 * select any unselected devices. If we're in "only" mode, we 867 * obviously don't want to select anything other than what the user 868 * specifies. If we're in "remove" mode, it probably isn't a good 869 * idea to go through and select any more devices, since we might 870 * end up selecting something that the user wants removed. Through 871 * more complicated logic, we could actually figure this out, but 872 * that would probably require combining this loop with the various 873 * selections loops above. 874 */ 875 if ((select_mode == DS_SELECT_ADD) && (*num_selected < maxshowdevs)) { 876 for (i = 0; i < *num_selections; i++) 877 if ((*dev_select)[i].selected == 0) { 878 (*dev_select)[i].selected = ++selection_number; 879 (*num_selected)++; 880 } 881 } 882 883 /* 884 * Look at the number of devices that have been selected. If it 885 * has changed, set the changed variable. Otherwise, if we've 886 * made a backup of the selection list, compare it to the current 887 * selection list to see if the selected devices have changed. 888 */ 889 if ((changed == 0) && (old_num_selected != *num_selected)) 890 changed = 1; 891 else if ((changed == 0) && (old_dev_select != NULL)) { 892 /* 893 * Now we go through the selection list and we look at 894 * it three different ways. 895 */ 896 for (i = 0; (i < *num_selections) && (changed == 0) && 897 (i < old_num_selections); i++) { 898 /* 899 * If the device at index i in both the new and old 900 * selection arrays has the same device number and 901 * selection status, it hasn't changed. We 902 * continue on to the next index. 903 */ 904 if (((*dev_select)[i].device_number == 905 old_dev_select[i].device_number) 906 && ((*dev_select)[i].selected == 907 old_dev_select[i].selected)) 908 continue; 909 910 /* 911 * Now, if we're still going through the if 912 * statement, the above test wasn't true. So we 913 * check here to see if the device at index i in 914 * the current array is the same as the device at 915 * index i in the old array. If it is, that means 916 * that its selection number has changed. Set 917 * changed to 1 and exit the loop. 918 */ 919 else if ((*dev_select)[i].device_number == 920 old_dev_select[i].device_number) { 921 changed = 1; 922 break; 923 } 924 /* 925 * If we get here, then the device at index i in 926 * the current array isn't the same device as the 927 * device at index i in the old array. 928 */ 929 else { 930 found = 0; 931 932 /* 933 * Search through the old selection array 934 * looking for a device with the same 935 * device number as the device at index i 936 * in the current array. If the selection 937 * status is the same, then we mark it as 938 * found. If the selection status isn't 939 * the same, we break out of the loop. 940 * Since found isn't set, changed will be 941 * set to 1 below. 942 */ 943 for (j = 0; j < old_num_selections; j++) { 944 if (((*dev_select)[i].device_number == 945 old_dev_select[j].device_number) 946 && ((*dev_select)[i].selected == 947 old_dev_select[j].selected)){ 948 found = 1; 949 break; 950 } 951 else if ((*dev_select)[i].device_number 952 == old_dev_select[j].device_number) 953 break; 954 } 955 if (found == 0) 956 changed = 1; 957 } 958 } 959 } 960 if (old_dev_select != NULL) 961 free(old_dev_select); 962 963 return(changed); 964 } 965 966 /* 967 * Comparison routine for qsort() above. Note that the comparison here is 968 * backwards -- generally, it should return a value to indicate whether 969 * arg1 is <, =, or > arg2. Instead, it returns the opposite. The reason 970 * it returns the opposite is so that the selection array will be sorted in 971 * order of decreasing performance. We sort on two parameters. The first 972 * sort key is whether or not one or the other of the devices in question 973 * has been selected. If one of them has, and the other one has not, the 974 * selected device is automatically more important than the unselected 975 * device. If neither device is selected, we judge the devices based upon 976 * performance. 977 */ 978 static int 979 compare_select(const void *arg1, const void *arg2) 980 { 981 if ((((const struct device_selection *)arg1)->selected) 982 && (((const struct device_selection *)arg2)->selected == 0)) 983 return(-1); 984 else if ((((const struct device_selection *)arg1)->selected == 0) 985 && (((const struct device_selection *)arg2)->selected)) 986 return(1); 987 else if (((const struct device_selection *)arg2)->bytes < 988 ((const struct device_selection *)arg1)->bytes) 989 return(-1); 990 else if (((const struct device_selection *)arg2)->bytes > 991 ((const struct device_selection *)arg1)->bytes) 992 return(1); 993 else 994 return(0); 995 } 996 997 /* 998 * Take a string with the general format "arg1,arg2,arg3", and build a 999 * device matching expression from it. 1000 */ 1001 int 1002 devstat_buildmatch(char *match_str, struct devstat_match **matches, 1003 int *num_matches) 1004 { 1005 char *tstr[5]; 1006 char **tempstr; 1007 int num_args; 1008 int i, j; 1009 const char *func_name = "devstat_buildmatch"; 1010 1011 /* We can't do much without a string to parse */ 1012 if (match_str == NULL) { 1013 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1014 "%s: no match expression", func_name); 1015 return(-1); 1016 } 1017 1018 /* 1019 * Break the (comma delimited) input string out into separate strings. 1020 */ 1021 for (tempstr = tstr, num_args = 0; 1022 (*tempstr = strsep(&match_str, ",")) != NULL && (num_args < 5); 1023 num_args++) 1024 if (**tempstr != '\0') 1025 if (++tempstr >= &tstr[5]) 1026 break; 1027 1028 /* The user gave us too many type arguments */ 1029 if (num_args > 3) { 1030 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1031 "%s: too many type arguments", func_name); 1032 return(-1); 1033 } 1034 1035 /* 1036 * Since you can't realloc a pointer that hasn't been malloced 1037 * first, we malloc first and then realloc. 1038 */ 1039 if (*num_matches == 0) 1040 *matches = (struct devstat_match *)malloc( 1041 sizeof(struct devstat_match)); 1042 else 1043 *matches = (struct devstat_match *)realloc(*matches, 1044 sizeof(struct devstat_match) * (*num_matches + 1)); 1045 1046 /* Make sure the current entry is clear */ 1047 bzero(&matches[0][*num_matches], sizeof(struct devstat_match)); 1048 1049 /* 1050 * Step through the arguments the user gave us and build a device 1051 * matching expression from them. 1052 */ 1053 for (i = 0; i < num_args; i++) { 1054 char *tempstr2, *tempstr3; 1055 1056 /* 1057 * Get rid of leading white space. 1058 */ 1059 tempstr2 = tstr[i]; 1060 while (isspace(*tempstr2) && (*tempstr2 != '\0')) 1061 tempstr2++; 1062 1063 /* 1064 * Get rid of trailing white space. 1065 */ 1066 tempstr3 = &tempstr2[strlen(tempstr2) - 1]; 1067 1068 while ((*tempstr3 != '\0') && (tempstr3 > tempstr2) 1069 && (isspace(*tempstr3))) { 1070 *tempstr3 = '\0'; 1071 tempstr3--; 1072 } 1073 1074 /* 1075 * Go through the match table comparing the user's 1076 * arguments to known device types, interfaces, etc. 1077 */ 1078 for (j = 0; match_table[j].match_str != NULL; j++) { 1079 /* 1080 * We do case-insensitive matching, in case someone 1081 * wants to enter "SCSI" instead of "scsi" or 1082 * something like that. Only compare as many 1083 * characters as are in the string in the match 1084 * table. This should help if someone tries to use 1085 * a super-long match expression. 1086 */ 1087 if (strncasecmp(tempstr2, match_table[j].match_str, 1088 strlen(match_table[j].match_str)) == 0) { 1089 /* 1090 * Make sure the user hasn't specified two 1091 * items of the same type, like "da" and 1092 * "cd". One device cannot be both. 1093 */ 1094 if (((*matches)[*num_matches].match_fields & 1095 match_table[j].match_field) != 0) { 1096 snprintf(devstat_errbuf, 1097 sizeof(devstat_errbuf), 1098 "%s: cannot have more than " 1099 "one match item in a single " 1100 "category", func_name); 1101 return(-1); 1102 } 1103 /* 1104 * If we've gotten this far, we have a 1105 * winner. Set the appropriate fields in 1106 * the match entry. 1107 */ 1108 (*matches)[*num_matches].match_fields |= 1109 match_table[j].match_field; 1110 (*matches)[*num_matches].device_type |= 1111 match_table[j].type; 1112 (*matches)[*num_matches].num_match_categories++; 1113 break; 1114 } 1115 } 1116 /* 1117 * We should have found a match in the above for loop. If 1118 * not, that means the user entered an invalid device type 1119 * or interface. 1120 */ 1121 if ((*matches)[*num_matches].num_match_categories != (i + 1)) { 1122 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1123 "%s: unknown match item \"%s\"", func_name, 1124 tstr[i]); 1125 return(-1); 1126 } 1127 } 1128 1129 (*num_matches)++; 1130 1131 return(0); 1132 } 1133 1134 /* 1135 * Compute a number of device statistics. Only one field is mandatory, and 1136 * that is "current". Everything else is optional. The caller passes in 1137 * pointers to variables to hold the various statistics he desires. If he 1138 * doesn't want a particular staistic, he should pass in a NULL pointer. 1139 * Return values: 1140 * 0 -- success 1141 * -1 -- failure 1142 */ 1143 int 1144 compute_stats(struct devstat *current, struct devstat *previous, 1145 long double etime, u_int64_t *total_bytes, 1146 u_int64_t *total_transfers, u_int64_t *total_blocks, 1147 long double *kb_per_transfer, long double *transfers_per_second, 1148 long double *mb_per_second, long double *blocks_per_second, 1149 long double *ms_per_transaction) 1150 { 1151 return(devstat_compute_statistics(current, previous, etime, 1152 total_bytes ? DSM_TOTAL_BYTES : DSM_SKIP, 1153 total_bytes, 1154 total_transfers ? DSM_TOTAL_TRANSFERS : DSM_SKIP, 1155 total_transfers, 1156 total_blocks ? DSM_TOTAL_BLOCKS : DSM_SKIP, 1157 total_blocks, 1158 kb_per_transfer ? DSM_KB_PER_TRANSFER : DSM_SKIP, 1159 kb_per_transfer, 1160 transfers_per_second ? DSM_TRANSFERS_PER_SECOND : DSM_SKIP, 1161 transfers_per_second, 1162 mb_per_second ? DSM_MB_PER_SECOND : DSM_SKIP, 1163 mb_per_second, 1164 blocks_per_second ? DSM_BLOCKS_PER_SECOND : DSM_SKIP, 1165 blocks_per_second, 1166 ms_per_transaction ? DSM_MS_PER_TRANSACTION : DSM_SKIP, 1167 ms_per_transaction, 1168 DSM_NONE)); 1169 } 1170 1171 1172 /* This is 1/2^64 */ 1173 #define BINTIME_SCALE 5.42101086242752217003726400434970855712890625e-20 1174 1175 long double 1176 devstat_compute_etime(struct bintime *cur_time, struct bintime *prev_time) 1177 { 1178 long double etime; 1179 1180 etime = cur_time->sec; 1181 etime += cur_time->frac * BINTIME_SCALE; 1182 if (prev_time != NULL) { 1183 etime -= prev_time->sec; 1184 etime -= prev_time->frac * BINTIME_SCALE; 1185 } 1186 return(etime); 1187 } 1188 1189 #define DELTA(field, index) \ 1190 (current->field[(index)] - (previous ? previous->field[(index)] : 0)) 1191 1192 #define DELTA_T(field) \ 1193 devstat_compute_etime(¤t->field, \ 1194 (previous ? &previous->field : NULL)) 1195 1196 int 1197 devstat_compute_statistics(struct devstat *current, struct devstat *previous, 1198 long double etime, ...) 1199 { 1200 const char *func_name = "devstat_compute_statistics"; 1201 u_int64_t totalbytes, totalbytesread, totalbyteswrite, totalbytesfree; 1202 u_int64_t totaltransfers, totaltransfersread, totaltransferswrite; 1203 u_int64_t totaltransfersother, totalblocks, totalblocksread; 1204 u_int64_t totalblockswrite, totaltransfersfree, totalblocksfree; 1205 va_list ap; 1206 devstat_metric metric; 1207 u_int64_t *destu64; 1208 long double *destld; 1209 int retval, i; 1210 1211 retval = 0; 1212 1213 /* 1214 * current is the only mandatory field. 1215 */ 1216 if (current == NULL) { 1217 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1218 "%s: current stats structure was NULL", func_name); 1219 return(-1); 1220 } 1221 1222 totalbytesread = DELTA(bytes, DEVSTAT_READ); 1223 totalbyteswrite = DELTA(bytes, DEVSTAT_WRITE); 1224 totalbytesfree = DELTA(bytes, DEVSTAT_FREE); 1225 totalbytes = totalbytesread + totalbyteswrite + totalbytesfree; 1226 1227 totaltransfersread = DELTA(operations, DEVSTAT_READ); 1228 totaltransferswrite = DELTA(operations, DEVSTAT_WRITE); 1229 totaltransfersother = DELTA(operations, DEVSTAT_NO_DATA); 1230 totaltransfersfree = DELTA(operations, DEVSTAT_FREE); 1231 totaltransfers = totaltransfersread + totaltransferswrite + 1232 totaltransfersother + totaltransfersfree; 1233 1234 totalblocks = totalbytes; 1235 totalblocksread = totalbytesread; 1236 totalblockswrite = totalbyteswrite; 1237 totalblocksfree = totalbytesfree; 1238 1239 if (current->block_size > 0) { 1240 totalblocks /= current->block_size; 1241 totalblocksread /= current->block_size; 1242 totalblockswrite /= current->block_size; 1243 totalblocksfree /= current->block_size; 1244 } else { 1245 totalblocks /= 512; 1246 totalblocksread /= 512; 1247 totalblockswrite /= 512; 1248 totalblocksfree /= 512; 1249 } 1250 1251 va_start(ap, etime); 1252 1253 while ((metric = (devstat_metric)va_arg(ap, devstat_metric)) != 0) { 1254 1255 if (metric == DSM_NONE) 1256 break; 1257 1258 if (metric >= DSM_MAX) { 1259 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1260 "%s: metric %d is out of range", func_name, 1261 metric); 1262 retval = -1; 1263 goto bailout; 1264 } 1265 1266 switch (devstat_arg_list[metric].argtype) { 1267 case DEVSTAT_ARG_UINT64: 1268 destu64 = (u_int64_t *)va_arg(ap, u_int64_t *); 1269 break; 1270 case DEVSTAT_ARG_LD: 1271 destld = (long double *)va_arg(ap, long double *); 1272 break; 1273 case DEVSTAT_ARG_SKIP: 1274 destld = (long double *)va_arg(ap, long double *); 1275 break; 1276 default: 1277 retval = -1; 1278 goto bailout; 1279 break; /* NOTREACHED */ 1280 } 1281 1282 if (devstat_arg_list[metric].argtype == DEVSTAT_ARG_SKIP) 1283 continue; 1284 1285 switch (metric) { 1286 case DSM_TOTAL_BYTES: 1287 *destu64 = totalbytes; 1288 break; 1289 case DSM_TOTAL_BYTES_READ: 1290 *destu64 = totalbytesread; 1291 break; 1292 case DSM_TOTAL_BYTES_WRITE: 1293 *destu64 = totalbyteswrite; 1294 break; 1295 case DSM_TOTAL_BYTES_FREE: 1296 *destu64 = totalbytesfree; 1297 break; 1298 case DSM_TOTAL_TRANSFERS: 1299 *destu64 = totaltransfers; 1300 break; 1301 case DSM_TOTAL_TRANSFERS_READ: 1302 *destu64 = totaltransfersread; 1303 break; 1304 case DSM_TOTAL_TRANSFERS_WRITE: 1305 *destu64 = totaltransferswrite; 1306 break; 1307 case DSM_TOTAL_TRANSFERS_FREE: 1308 *destu64 = totaltransfersfree; 1309 break; 1310 case DSM_TOTAL_TRANSFERS_OTHER: 1311 *destu64 = totaltransfersother; 1312 break; 1313 case DSM_TOTAL_BLOCKS: 1314 *destu64 = totalblocks; 1315 break; 1316 case DSM_TOTAL_BLOCKS_READ: 1317 *destu64 = totalblocksread; 1318 break; 1319 case DSM_TOTAL_BLOCKS_WRITE: 1320 *destu64 = totalblockswrite; 1321 break; 1322 case DSM_TOTAL_BLOCKS_FREE: 1323 *destu64 = totalblocksfree; 1324 break; 1325 case DSM_KB_PER_TRANSFER: 1326 *destld = totalbytes; 1327 *destld /= 1024; 1328 if (totaltransfers > 0) 1329 *destld /= totaltransfers; 1330 else 1331 *destld = 0.0; 1332 break; 1333 case DSM_KB_PER_TRANSFER_READ: 1334 *destld = totalbytesread; 1335 *destld /= 1024; 1336 if (totaltransfersread > 0) 1337 *destld /= totaltransfersread; 1338 else 1339 *destld = 0.0; 1340 break; 1341 case DSM_KB_PER_TRANSFER_WRITE: 1342 *destld = totalbyteswrite; 1343 *destld /= 1024; 1344 if (totaltransferswrite > 0) 1345 *destld /= totaltransferswrite; 1346 else 1347 *destld = 0.0; 1348 break; 1349 case DSM_KB_PER_TRANSFER_FREE: 1350 *destld = totalbytesfree; 1351 *destld /= 1024; 1352 if (totaltransfersfree > 0) 1353 *destld /= totaltransfersfree; 1354 else 1355 *destld = 0.0; 1356 break; 1357 case DSM_TRANSFERS_PER_SECOND: 1358 if (etime > 0.0) { 1359 *destld = totaltransfers; 1360 *destld /= etime; 1361 } else 1362 *destld = 0.0; 1363 break; 1364 case DSM_TRANSFERS_PER_SECOND_READ: 1365 if (etime > 0.0) { 1366 *destld = totaltransfersread; 1367 *destld /= etime; 1368 } else 1369 *destld = 0.0; 1370 break; 1371 case DSM_TRANSFERS_PER_SECOND_WRITE: 1372 if (etime > 0.0) { 1373 *destld = totaltransferswrite; 1374 *destld /= etime; 1375 } else 1376 *destld = 0.0; 1377 break; 1378 case DSM_TRANSFERS_PER_SECOND_FREE: 1379 if (etime > 0.0) { 1380 *destld = totaltransfersfree; 1381 *destld /= etime; 1382 } else 1383 *destld = 0.0; 1384 break; 1385 case DSM_TRANSFERS_PER_SECOND_OTHER: 1386 if (etime > 0.0) { 1387 *destld = totaltransfersother; 1388 *destld /= etime; 1389 } else 1390 *destld = 0.0; 1391 break; 1392 case DSM_MB_PER_SECOND: 1393 *destld = totalbytes; 1394 *destld /= 1024 * 1024; 1395 if (etime > 0.0) 1396 *destld /= etime; 1397 else 1398 *destld = 0.0; 1399 break; 1400 case DSM_MB_PER_SECOND_READ: 1401 *destld = totalbytesread; 1402 *destld /= 1024 * 1024; 1403 if (etime > 0.0) 1404 *destld /= etime; 1405 else 1406 *destld = 0.0; 1407 break; 1408 case DSM_MB_PER_SECOND_WRITE: 1409 *destld = totalbyteswrite; 1410 *destld /= 1024 * 1024; 1411 if (etime > 0.0) 1412 *destld /= etime; 1413 else 1414 *destld = 0.0; 1415 break; 1416 case DSM_MB_PER_SECOND_FREE: 1417 *destld = totalbytesfree; 1418 *destld /= 1024 * 1024; 1419 if (etime > 0.0) 1420 *destld /= etime; 1421 else 1422 *destld = 0.0; 1423 break; 1424 case DSM_BLOCKS_PER_SECOND: 1425 *destld = totalblocks; 1426 if (etime > 0.0) 1427 *destld /= etime; 1428 else 1429 *destld = 0.0; 1430 break; 1431 case DSM_BLOCKS_PER_SECOND_READ: 1432 *destld = totalblocksread; 1433 if (etime > 0.0) 1434 *destld /= etime; 1435 else 1436 *destld = 0.0; 1437 break; 1438 case DSM_BLOCKS_PER_SECOND_WRITE: 1439 *destld = totalblockswrite; 1440 if (etime > 0.0) 1441 *destld /= etime; 1442 else 1443 *destld = 0.0; 1444 break; 1445 case DSM_BLOCKS_PER_SECOND_FREE: 1446 *destld = totalblocksfree; 1447 if (etime > 0.0) 1448 *destld /= etime; 1449 else 1450 *destld = 0.0; 1451 break; 1452 /* 1453 * This calculation is somewhat bogus. It simply divides 1454 * the elapsed time by the total number of transactions 1455 * completed. While that does give the caller a good 1456 * picture of the average rate of transaction completion, 1457 * it doesn't necessarily give the caller a good view of 1458 * how long transactions took to complete on average. 1459 * Those two numbers will be different for a device that 1460 * can handle more than one transaction at a time. e.g. 1461 * SCSI disks doing tagged queueing. 1462 * 1463 * The only way to accurately determine the real average 1464 * time per transaction would be to compute and store the 1465 * time on a per-transaction basis. That currently isn't 1466 * done in the kernel, and would only be desireable if it 1467 * could be implemented in a somewhat non-intrusive and high 1468 * performance way. 1469 */ 1470 case DSM_MS_PER_TRANSACTION: 1471 if (totaltransfers > 0) { 1472 *destld = 0; 1473 for (i = 0; i < DEVSTAT_N_TRANS_FLAGS; i++) 1474 *destld += DELTA_T(duration[i]); 1475 *destld /= totaltransfers; 1476 *destld *= 1000; 1477 } else 1478 *destld = 0.0; 1479 break; 1480 /* 1481 * As above, these next two really only give the average 1482 * rate of completion for read and write transactions, not 1483 * the average time the transaction took to complete. 1484 */ 1485 case DSM_MS_PER_TRANSACTION_READ: 1486 if (totaltransfersread > 0) { 1487 *destld = DELTA_T(duration[DEVSTAT_READ]); 1488 *destld /= totaltransfersread; 1489 *destld *= 1000; 1490 } else 1491 *destld = 0.0; 1492 break; 1493 case DSM_MS_PER_TRANSACTION_WRITE: 1494 if (totaltransferswrite > 0) { 1495 *destld = DELTA_T(duration[DEVSTAT_WRITE]); 1496 *destld /= totaltransferswrite; 1497 *destld *= 1000; 1498 } else 1499 *destld = 0.0; 1500 break; 1501 case DSM_MS_PER_TRANSACTION_FREE: 1502 if (totaltransfersfree > 0) { 1503 *destld = DELTA_T(duration[DEVSTAT_FREE]); 1504 *destld /= totaltransfersfree; 1505 *destld *= 1000; 1506 } else 1507 *destld = 0.0; 1508 break; 1509 case DSM_MS_PER_TRANSACTION_OTHER: 1510 if (totaltransfersother > 0) { 1511 *destld = DELTA_T(duration[DEVSTAT_NO_DATA]); 1512 *destld /= totaltransfersother; 1513 *destld *= 1000; 1514 } else 1515 *destld = 0.0; 1516 break; 1517 case DSM_BUSY_PCT: 1518 *destld = DELTA_T(busy_time); 1519 if (*destld < 0) 1520 *destld = 0; 1521 *destld /= etime; 1522 *destld *= 100; 1523 if (*destld < 0) 1524 *destld = 0; 1525 break; 1526 case DSM_QUEUE_LENGTH: 1527 *destu64 = current->start_count - current->end_count; 1528 break; 1529 /* 1530 * XXX: comment out the default block to see if any case's are missing. 1531 */ 1532 #if 1 1533 default: 1534 /* 1535 * This shouldn't happen, since we should have 1536 * caught any out of range metrics at the top of 1537 * the loop. 1538 */ 1539 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1540 "%s: unknown metric %d", func_name, metric); 1541 retval = -1; 1542 goto bailout; 1543 break; /* NOTREACHED */ 1544 #endif 1545 } 1546 } 1547 1548 bailout: 1549 1550 va_end(ap); 1551 return(retval); 1552 } 1553 1554 static int 1555 readkmem(kvm_t *kd, unsigned long addr, void *buf, size_t nbytes) 1556 { 1557 const char *func_name = "readkmem"; 1558 1559 if (kvm_read(kd, addr, buf, nbytes) == -1) { 1560 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1561 "%s: error reading value (kvm_read): %s", func_name, 1562 kvm_geterr(kd)); 1563 return(-1); 1564 } 1565 return(0); 1566 } 1567 1568 static int 1569 readkmem_nl(kvm_t *kd, const char *name, void *buf, size_t nbytes) 1570 { 1571 const char *func_name = "readkmem_nl"; 1572 struct nlist nl[2]; 1573 1574 (const char *)nl[0].n_name = name; 1575 nl[1].n_name = NULL; 1576 1577 if (kvm_nlist(kd, nl) == -1) { 1578 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1579 "%s: error getting name list (kvm_nlist): %s", 1580 func_name, kvm_geterr(kd)); 1581 return(-1); 1582 } 1583 return(readkmem(kd, nl[0].n_value, buf, nbytes)); 1584 } 1585 1586 /* 1587 * This duplicates the functionality of the kernel sysctl handler for poking 1588 * through crash dumps. 1589 */ 1590 static char * 1591 get_devstat_kvm(kvm_t *kd) 1592 { 1593 int error, i, wp; 1594 long gen; 1595 struct devstat *nds; 1596 struct devstat ds; 1597 struct devstatlist dhead; 1598 int num_devs; 1599 char *rv = NULL; 1600 const char *func_name = "get_devstat_kvm"; 1601 1602 if ((num_devs = devstat_getnumdevs(kd)) <= 0) 1603 return(NULL); 1604 error = 0; 1605 if (KREADNL(kd, X_DEVICE_STATQ, dhead) == -1) 1606 return(NULL); 1607 1608 nds = STAILQ_FIRST(&dhead); 1609 1610 if ((rv = malloc(sizeof(gen))) == NULL) { 1611 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1612 "%s: out of memory (initial malloc failed)", 1613 func_name); 1614 return(NULL); 1615 } 1616 gen = devstat_getgeneration(kd); 1617 memcpy(rv, &gen, sizeof(gen)); 1618 wp = sizeof(gen); 1619 /* 1620 * Now push out all the devices. 1621 */ 1622 for (i = 0; (nds != NULL) && (i < num_devs); 1623 nds = STAILQ_NEXT(nds, dev_links), i++) { 1624 if (readkmem(kd, (long)nds, &ds, sizeof(ds)) == -1) { 1625 free(rv); 1626 return(NULL); 1627 } 1628 nds = &ds; 1629 rv = (char *)reallocf(rv, sizeof(gen) + 1630 sizeof(ds) * (i + 1)); 1631 if (rv == NULL) { 1632 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1633 "%s: out of memory (malloc failed)", 1634 func_name); 1635 return(NULL); 1636 } 1637 memcpy(rv + wp, &ds, sizeof(ds)); 1638 wp += sizeof(ds); 1639 } 1640 return(rv); 1641 } 1642