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