1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1997, 1998 Kenneth D. Merry. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The name of the author may not be used to endorse or promote products 16 * derived from this software without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31 #include <sys/cdefs.h> 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 #include <nlist.h> 48 49 #include "devstat.h" 50 51 int 52 compute_stats(struct devstat *current, struct devstat *previous, 53 long double etime, u_int64_t *total_bytes, 54 u_int64_t *total_transfers, u_int64_t *total_blocks, 55 long double *kb_per_transfer, long double *transfers_per_second, 56 long double *mb_per_second, long double *blocks_per_second, 57 long double *ms_per_transaction); 58 59 typedef enum { 60 DEVSTAT_ARG_NOTYPE, 61 DEVSTAT_ARG_UINT64, 62 DEVSTAT_ARG_LD, 63 DEVSTAT_ARG_SKIP 64 } devstat_arg_type; 65 66 char devstat_errbuf[DEVSTAT_ERRBUF_SIZE]; 67 68 /* 69 * Table to match descriptive strings with device types. These are in 70 * order from most common to least common to speed search time. 71 */ 72 struct devstat_match_table match_table[] = { 73 {"da", DEVSTAT_TYPE_DIRECT, DEVSTAT_MATCH_TYPE}, 74 {"cd", DEVSTAT_TYPE_CDROM, DEVSTAT_MATCH_TYPE}, 75 {"scsi", DEVSTAT_TYPE_IF_SCSI, DEVSTAT_MATCH_IF}, 76 {"ide", DEVSTAT_TYPE_IF_IDE, DEVSTAT_MATCH_IF}, 77 {"other", DEVSTAT_TYPE_IF_OTHER, DEVSTAT_MATCH_IF}, 78 {"nvme", DEVSTAT_TYPE_IF_NVME, DEVSTAT_MATCH_IF}, 79 {"worm", DEVSTAT_TYPE_WORM, DEVSTAT_MATCH_TYPE}, 80 {"sa", DEVSTAT_TYPE_SEQUENTIAL,DEVSTAT_MATCH_TYPE}, 81 {"pass", DEVSTAT_TYPE_PASS, DEVSTAT_MATCH_PASS}, 82 {"optical", DEVSTAT_TYPE_OPTICAL, DEVSTAT_MATCH_TYPE}, 83 {"array", DEVSTAT_TYPE_STORARRAY, DEVSTAT_MATCH_TYPE}, 84 {"changer", DEVSTAT_TYPE_CHANGER, DEVSTAT_MATCH_TYPE}, 85 {"scanner", DEVSTAT_TYPE_SCANNER, DEVSTAT_MATCH_TYPE}, 86 {"printer", DEVSTAT_TYPE_PRINTER, DEVSTAT_MATCH_TYPE}, 87 {"floppy", DEVSTAT_TYPE_FLOPPY, DEVSTAT_MATCH_TYPE}, 88 {"proc", DEVSTAT_TYPE_PROCESSOR, DEVSTAT_MATCH_TYPE}, 89 {"comm", DEVSTAT_TYPE_COMM, DEVSTAT_MATCH_TYPE}, 90 {"enclosure", DEVSTAT_TYPE_ENCLOSURE, DEVSTAT_MATCH_TYPE}, 91 {NULL, 0, 0} 92 }; 93 94 struct devstat_args { 95 devstat_metric metric; 96 devstat_arg_type argtype; 97 } devstat_arg_list[] = { 98 { DSM_NONE, DEVSTAT_ARG_NOTYPE }, 99 { DSM_TOTAL_BYTES, DEVSTAT_ARG_UINT64 }, 100 { DSM_TOTAL_BYTES_READ, DEVSTAT_ARG_UINT64 }, 101 { DSM_TOTAL_BYTES_WRITE, DEVSTAT_ARG_UINT64 }, 102 { DSM_TOTAL_TRANSFERS, DEVSTAT_ARG_UINT64 }, 103 { DSM_TOTAL_TRANSFERS_READ, DEVSTAT_ARG_UINT64 }, 104 { DSM_TOTAL_TRANSFERS_WRITE, DEVSTAT_ARG_UINT64 }, 105 { DSM_TOTAL_TRANSFERS_OTHER, DEVSTAT_ARG_UINT64 }, 106 { DSM_TOTAL_BLOCKS, DEVSTAT_ARG_UINT64 }, 107 { DSM_TOTAL_BLOCKS_READ, DEVSTAT_ARG_UINT64 }, 108 { DSM_TOTAL_BLOCKS_WRITE, DEVSTAT_ARG_UINT64 }, 109 { DSM_KB_PER_TRANSFER, DEVSTAT_ARG_LD }, 110 { DSM_KB_PER_TRANSFER_READ, DEVSTAT_ARG_LD }, 111 { DSM_KB_PER_TRANSFER_WRITE, DEVSTAT_ARG_LD }, 112 { DSM_TRANSFERS_PER_SECOND, DEVSTAT_ARG_LD }, 113 { DSM_TRANSFERS_PER_SECOND_READ, DEVSTAT_ARG_LD }, 114 { DSM_TRANSFERS_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, 115 { DSM_TRANSFERS_PER_SECOND_OTHER, DEVSTAT_ARG_LD }, 116 { DSM_MB_PER_SECOND, DEVSTAT_ARG_LD }, 117 { DSM_MB_PER_SECOND_READ, DEVSTAT_ARG_LD }, 118 { DSM_MB_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, 119 { DSM_BLOCKS_PER_SECOND, DEVSTAT_ARG_LD }, 120 { DSM_BLOCKS_PER_SECOND_READ, DEVSTAT_ARG_LD }, 121 { DSM_BLOCKS_PER_SECOND_WRITE, DEVSTAT_ARG_LD }, 122 { DSM_MS_PER_TRANSACTION, DEVSTAT_ARG_LD }, 123 { DSM_MS_PER_TRANSACTION_READ, DEVSTAT_ARG_LD }, 124 { DSM_MS_PER_TRANSACTION_WRITE, DEVSTAT_ARG_LD }, 125 { DSM_SKIP, DEVSTAT_ARG_SKIP }, 126 { DSM_TOTAL_BYTES_FREE, DEVSTAT_ARG_UINT64 }, 127 { DSM_TOTAL_TRANSFERS_FREE, DEVSTAT_ARG_UINT64 }, 128 { DSM_TOTAL_BLOCKS_FREE, DEVSTAT_ARG_UINT64 }, 129 { DSM_KB_PER_TRANSFER_FREE, DEVSTAT_ARG_LD }, 130 { DSM_MB_PER_SECOND_FREE, DEVSTAT_ARG_LD }, 131 { DSM_TRANSFERS_PER_SECOND_FREE, DEVSTAT_ARG_LD }, 132 { DSM_BLOCKS_PER_SECOND_FREE, DEVSTAT_ARG_LD }, 133 { DSM_MS_PER_TRANSACTION_OTHER, DEVSTAT_ARG_LD }, 134 { DSM_MS_PER_TRANSACTION_FREE, DEVSTAT_ARG_LD }, 135 { DSM_BUSY_PCT, DEVSTAT_ARG_LD }, 136 { DSM_QUEUE_LENGTH, DEVSTAT_ARG_UINT64 }, 137 { DSM_TOTAL_DURATION, DEVSTAT_ARG_LD }, 138 { DSM_TOTAL_DURATION_READ, DEVSTAT_ARG_LD }, 139 { DSM_TOTAL_DURATION_WRITE, DEVSTAT_ARG_LD }, 140 { DSM_TOTAL_DURATION_FREE, DEVSTAT_ARG_LD }, 141 { DSM_TOTAL_DURATION_OTHER, DEVSTAT_ARG_LD }, 142 { DSM_TOTAL_BUSY_TIME, DEVSTAT_ARG_LD }, 143 }; 144 145 static const char *namelist[] = { 146 #define X_NUMDEVS 0 147 "_devstat_num_devs", 148 #define X_GENERATION 1 149 "_devstat_generation", 150 #define X_VERSION 2 151 "_devstat_version", 152 #define X_DEVICE_STATQ 3 153 "_device_statq", 154 #define X_TIME_UPTIME 4 155 "_time_uptime", 156 #define X_END 5 157 }; 158 159 /* 160 * Local function declarations. 161 */ 162 static int compare_select(const void *arg1, const void *arg2); 163 static int readkmem(kvm_t *kd, unsigned long addr, void *buf, size_t nbytes); 164 static int readkmem_nl(kvm_t *kd, const char *name, void *buf, size_t nbytes); 165 static char *get_devstat_kvm(kvm_t *kd); 166 167 #define KREADNL(kd, var, val) \ 168 readkmem_nl(kd, namelist[var], &val, sizeof(val)) 169 170 int 171 devstat_getnumdevs(kvm_t *kd) 172 { 173 size_t numdevsize; 174 int numdevs; 175 176 numdevsize = sizeof(int); 177 178 /* 179 * Find out how many devices we have in the system. 180 */ 181 if (kd == NULL) { 182 if (sysctlbyname("kern.devstat.numdevs", &numdevs, 183 &numdevsize, NULL, 0) == -1) { 184 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 185 "%s: error getting number of devices\n" 186 "%s: %s", __func__, __func__, 187 strerror(errno)); 188 return(-1); 189 } else 190 return(numdevs); 191 } else { 192 193 if (KREADNL(kd, X_NUMDEVS, numdevs) == -1) 194 return(-1); 195 else 196 return(numdevs); 197 } 198 } 199 200 /* 201 * This is an easy way to get the generation number, but the generation is 202 * supplied in a more atmoic manner by the kern.devstat.all sysctl. 203 * Because this generation sysctl is separate from the statistics sysctl, 204 * the device list and the generation could change between the time that 205 * this function is called and the device list is retrieved. 206 */ 207 long 208 devstat_getgeneration(kvm_t *kd) 209 { 210 size_t gensize; 211 long generation; 212 213 gensize = sizeof(long); 214 215 /* 216 * Get the current generation number. 217 */ 218 if (kd == NULL) { 219 if (sysctlbyname("kern.devstat.generation", &generation, 220 &gensize, NULL, 0) == -1) { 221 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 222 "%s: error getting devstat generation\n%s: %s", 223 __func__, __func__, strerror(errno)); 224 return(-1); 225 } else 226 return(generation); 227 } else { 228 if (KREADNL(kd, X_GENERATION, generation) == -1) 229 return(-1); 230 else 231 return(generation); 232 } 233 } 234 235 /* 236 * Get the current devstat version. The return value of this function 237 * should be compared with DEVSTAT_VERSION, which is defined in 238 * sys/devicestat.h. This will enable userland programs to determine 239 * whether they are out of sync with the kernel. 240 */ 241 int 242 devstat_getversion(kvm_t *kd) 243 { 244 size_t versize; 245 int version; 246 247 versize = sizeof(int); 248 249 /* 250 * Get the current devstat version. 251 */ 252 if (kd == NULL) { 253 if (sysctlbyname("kern.devstat.version", &version, &versize, 254 NULL, 0) == -1) { 255 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 256 "%s: error getting devstat version\n%s: %s", 257 __func__, __func__, strerror(errno)); 258 return(-1); 259 } else 260 return(version); 261 } else { 262 if (KREADNL(kd, X_VERSION, version) == -1) 263 return(-1); 264 else 265 return(version); 266 } 267 } 268 269 /* 270 * Check the devstat version we know about against the devstat version the 271 * kernel knows about. If they don't match, print an error into the 272 * devstat error buffer, and return -1. If they match, return 0. 273 */ 274 int 275 devstat_checkversion(kvm_t *kd) 276 { 277 int buflen, res, retval = 0, version; 278 279 version = devstat_getversion(kd); 280 281 if (version != DEVSTAT_VERSION) { 282 /* 283 * If getversion() returns an error (i.e. -1), then it 284 * has printed an error message in the buffer. Therefore, 285 * we need to add a \n to the end of that message before we 286 * print our own message in the buffer. 287 */ 288 if (version == -1) 289 buflen = strlen(devstat_errbuf); 290 else 291 buflen = 0; 292 293 res = snprintf(devstat_errbuf + buflen, 294 DEVSTAT_ERRBUF_SIZE - buflen, 295 "%s%s: userland devstat version %d is not " 296 "the same as the kernel\n%s: devstat " 297 "version %d\n", version == -1 ? "\n" : "", 298 __func__, DEVSTAT_VERSION, __func__, version); 299 300 if (res < 0) 301 devstat_errbuf[buflen] = '\0'; 302 303 buflen = strlen(devstat_errbuf); 304 if (version < DEVSTAT_VERSION) 305 res = snprintf(devstat_errbuf + buflen, 306 DEVSTAT_ERRBUF_SIZE - buflen, 307 "%s: libdevstat newer than kernel\n", 308 __func__); 309 else 310 res = snprintf(devstat_errbuf + buflen, 311 DEVSTAT_ERRBUF_SIZE - buflen, 312 "%s: kernel newer than libdevstat\n", 313 __func__); 314 315 if (res < 0) 316 devstat_errbuf[buflen] = '\0'; 317 318 retval = -1; 319 } 320 321 return(retval); 322 } 323 324 /* 325 * Get the current list of devices and statistics, and the current 326 * generation number. 327 * 328 * Return values: 329 * -1 -- error 330 * 0 -- device list is unchanged 331 * 1 -- device list has changed 332 */ 333 int 334 devstat_getdevs(kvm_t *kd, struct statinfo *stats) 335 { 336 int error; 337 size_t dssize; 338 long oldgeneration; 339 int retval = 0; 340 struct devinfo *dinfo; 341 struct timespec ts; 342 343 dinfo = stats->dinfo; 344 345 if (dinfo == NULL) { 346 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 347 "%s: stats->dinfo was NULL", __func__); 348 return(-1); 349 } 350 351 oldgeneration = dinfo->generation; 352 353 if (kd == NULL) { 354 clock_gettime(CLOCK_MONOTONIC, &ts); 355 stats->snap_time = ts.tv_sec + ts.tv_nsec * 1e-9; 356 357 /* If this is our first time through, mem_ptr will be null. */ 358 if (dinfo->mem_ptr == NULL) { 359 /* 360 * Get the number of devices. If it's negative, it's an 361 * error. Don't bother setting the error string, since 362 * getnumdevs() has already done that for us. 363 */ 364 if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) 365 return(-1); 366 367 /* 368 * The kern.devstat.all sysctl returns the current 369 * generation number, as well as all the devices. 370 * So we need four bytes more. 371 */ 372 dssize = (dinfo->numdevs * sizeof(struct devstat)) + 373 sizeof(long); 374 dinfo->mem_ptr = (u_int8_t *)malloc(dssize); 375 if (dinfo->mem_ptr == NULL) { 376 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 377 "%s: Cannot allocate memory for mem_ptr element", 378 __func__); 379 return(-1); 380 } 381 } else 382 dssize = (dinfo->numdevs * sizeof(struct devstat)) + 383 sizeof(long); 384 385 /* 386 * Request all of the devices. We only really allow for one 387 * ENOMEM failure. It would, of course, be possible to just go 388 * in a loop and keep reallocing the device structure until we 389 * don't get ENOMEM back. I'm not sure it's worth it, though. 390 * If devices are being added to the system that quickly, maybe 391 * the user can just wait until all devices are added. 392 */ 393 for (;;) { 394 error = sysctlbyname("kern.devstat.all", 395 dinfo->mem_ptr, 396 &dssize, NULL, 0); 397 if (error != -1 || errno != EBUSY) 398 break; 399 } 400 if (error == -1) { 401 /* 402 * If we get ENOMEM back, that means that there are 403 * more devices now, so we need to allocate more 404 * space for the device array. 405 */ 406 if (errno == ENOMEM) { 407 /* 408 * No need to set the error string here, 409 * devstat_getnumdevs() will do that if it fails. 410 */ 411 if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) 412 return(-1); 413 414 dssize = (dinfo->numdevs * 415 sizeof(struct devstat)) + sizeof(long); 416 dinfo->mem_ptr = (u_int8_t *) 417 realloc(dinfo->mem_ptr, dssize); 418 if ((error = sysctlbyname("kern.devstat.all", 419 dinfo->mem_ptr, &dssize, NULL, 0)) == -1) { 420 snprintf(devstat_errbuf, 421 sizeof(devstat_errbuf), 422 "%s: error getting device " 423 "stats\n%s: %s", __func__, 424 __func__, strerror(errno)); 425 return(-1); 426 } 427 } else { 428 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 429 "%s: error getting device stats\n" 430 "%s: %s", __func__, __func__, 431 strerror(errno)); 432 return(-1); 433 } 434 } 435 436 } else { 437 if (KREADNL(kd, X_TIME_UPTIME, ts.tv_sec) == -1) 438 return(-1); 439 else 440 stats->snap_time = ts.tv_sec; 441 442 /* 443 * This is of course non-atomic, but since we are working 444 * on a core dump, the generation is unlikely to change 445 */ 446 if ((dinfo->numdevs = devstat_getnumdevs(kd)) == -1) 447 return(-1); 448 if ((dinfo->mem_ptr = (u_int8_t *)get_devstat_kvm(kd)) == NULL) 449 return(-1); 450 } 451 /* 452 * The sysctl spits out the generation as the first four bytes, 453 * then all of the device statistics structures. 454 */ 455 dinfo->generation = *(long *)dinfo->mem_ptr; 456 457 /* 458 * If the generation has changed, and if the current number of 459 * devices is not the same as the number of devices recorded in the 460 * devinfo structure, it is likely that the device list has shrunk. 461 * The reason that it is likely that the device list has shrunk in 462 * this case is that if the device list has grown, the sysctl above 463 * will return an ENOMEM error, and we will reset the number of 464 * devices and reallocate the device array. If the second sysctl 465 * fails, we will return an error and therefore never get to this 466 * point. If the device list has shrunk, the sysctl will not 467 * return an error since we have more space allocated than is 468 * necessary. So, in the shrinkage case, we catch it here and 469 * reallocate the array so that we don't use any more space than is 470 * necessary. 471 */ 472 if (oldgeneration != dinfo->generation) { 473 if (devstat_getnumdevs(kd) != dinfo->numdevs) { 474 if ((dinfo->numdevs = devstat_getnumdevs(kd)) < 0) 475 return(-1); 476 dssize = (dinfo->numdevs * sizeof(struct devstat)) + 477 sizeof(long); 478 dinfo->mem_ptr = (u_int8_t *)realloc(dinfo->mem_ptr, 479 dssize); 480 } 481 retval = 1; 482 } 483 484 dinfo->devices = (struct devstat *)(dinfo->mem_ptr + sizeof(long)); 485 486 return(retval); 487 } 488 489 /* 490 * selectdevs(): 491 * 492 * Devices are selected/deselected based upon the following criteria: 493 * - devices specified by the user on the command line 494 * - devices matching any device type expressions given on the command line 495 * - devices with the highest I/O, if 'top' mode is enabled 496 * - the first n unselected devices in the device list, if maxshowdevs 497 * devices haven't already been selected and if the user has not 498 * specified any devices on the command line and if we're in "add" mode. 499 * 500 * Input parameters: 501 * - device selection list (dev_select) 502 * - current number of devices selected (num_selected) 503 * - total number of devices in the selection list (num_selections) 504 * - devstat generation as of the last time selectdevs() was called 505 * (select_generation) 506 * - current devstat generation (current_generation) 507 * - current list of devices and statistics (devices) 508 * - number of devices in the current device list (numdevs) 509 * - compiled version of the command line device type arguments (matches) 510 * - This is optional. If the number of devices is 0, this will be ignored. 511 * - The matching code pays attention to the current selection mode. So 512 * if you pass in a matching expression, it will be evaluated based 513 * upon the selection mode that is passed in. See below for details. 514 * - number of device type matching expressions (num_matches) 515 * - Set to 0 to disable the matching code. 516 * - list of devices specified on the command line by the user (dev_selections) 517 * - number of devices selected on the command line by the user 518 * (num_dev_selections) 519 * - Our selection mode. There are four different selection modes: 520 * - add mode. (DS_SELECT_ADD) Any devices matching devices explicitly 521 * selected by the user or devices matching a pattern given by the 522 * user will be selected in addition to devices that are already 523 * selected. Additional devices will be selected, up to maxshowdevs 524 * number of devices. 525 * - only mode. (DS_SELECT_ONLY) Only devices matching devices 526 * explicitly given by the user or devices matching a pattern 527 * given by the user will be selected. No other devices will be 528 * selected. 529 * - addonly mode. (DS_SELECT_ADDONLY) This is similar to add and 530 * only. Basically, this will not de-select any devices that are 531 * current selected, as only mode would, but it will also not 532 * gratuitously select up to maxshowdevs devices as add mode would. 533 * - remove mode. (DS_SELECT_REMOVE) Any devices matching devices 534 * explicitly selected by the user or devices matching a pattern 535 * given by the user will be de-selected. 536 * - maximum number of devices we can select (maxshowdevs) 537 * - flag indicating whether or not we're in 'top' mode (perf_select) 538 * 539 * Output data: 540 * - the device selection list may be modified and passed back out 541 * - the number of devices selected and the total number of items in the 542 * device selection list may be changed 543 * - the selection generation may be changed to match the current generation 544 * 545 * Return values: 546 * -1 -- error 547 * 0 -- selected devices are unchanged 548 * 1 -- selected devices changed 549 */ 550 int 551 devstat_selectdevs(struct device_selection **dev_select, int *num_selected, 552 int *num_selections, long *select_generation, 553 long current_generation, struct devstat *devices, 554 int numdevs, struct devstat_match *matches, int num_matches, 555 char **dev_selections, int num_dev_selections, 556 devstat_select_mode select_mode, int maxshowdevs, 557 int perf_select) 558 { 559 int i, j, k; 560 int init_selections = 0, init_selected_var = 0; 561 struct device_selection *old_dev_select = NULL; 562 int old_num_selections = 0, old_num_selected; 563 int selection_number = 0; 564 int changed = 0, found = 0; 565 566 if ((dev_select == NULL) || (devices == NULL) || (numdevs < 0)) 567 return(-1); 568 569 /* 570 * We always want to make sure that we have as many dev_select 571 * entries as there are devices. 572 */ 573 /* 574 * In this case, we haven't selected devices before. 575 */ 576 if (*dev_select == NULL) { 577 *dev_select = (struct device_selection *)malloc(numdevs * 578 sizeof(struct device_selection)); 579 *select_generation = current_generation; 580 init_selections = 1; 581 changed = 1; 582 /* 583 * In this case, we have selected devices before, but the device 584 * list has changed since we last selected devices, so we need to 585 * either enlarge or reduce the size of the device selection list. 586 * But delay the resizing until after copying the data to old_dev_select 587 * as to not lose any data in the case of reducing the size. 588 */ 589 } else if (*num_selections != numdevs) { 590 *select_generation = current_generation; 591 init_selections = 1; 592 /* 593 * In this case, we've selected devices before, and the selection 594 * list is the same size as it was the last time, but the device 595 * list has changed. 596 */ 597 } else if (*select_generation < current_generation) { 598 *select_generation = current_generation; 599 init_selections = 1; 600 } 601 602 if (*dev_select == NULL) { 603 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 604 "%s: Cannot (re)allocate memory for dev_select argument", 605 __func__); 606 return(-1); 607 } 608 609 /* 610 * If we're in "only" mode, we want to clear out the selected 611 * variable since we're going to select exactly what the user wants 612 * this time through. 613 */ 614 if (select_mode == DS_SELECT_ONLY) 615 init_selected_var = 1; 616 617 /* 618 * In all cases, we want to back up the number of selected devices. 619 * It is a quick and accurate way to determine whether the selected 620 * devices have changed. 621 */ 622 old_num_selected = *num_selected; 623 624 /* 625 * We want to make a backup of the current selection list if 626 * the list of devices has changed, or if we're in performance 627 * selection mode. In both cases, we don't want to make a backup 628 * if we already know for sure that the list will be different. 629 * This is certainly the case if this is our first time through the 630 * selection code. 631 */ 632 if (((init_selected_var != 0) || (init_selections != 0) 633 || (perf_select != 0)) && (changed == 0)){ 634 old_dev_select = (struct device_selection *)malloc( 635 *num_selections * sizeof(struct device_selection)); 636 if (old_dev_select == NULL) { 637 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 638 "%s: Cannot allocate memory for selection list backup", 639 __func__); 640 return(-1); 641 } 642 old_num_selections = *num_selections; 643 bcopy(*dev_select, old_dev_select, 644 sizeof(struct device_selection) * *num_selections); 645 } 646 647 if (!changed && *num_selections != numdevs) { 648 *dev_select = (struct device_selection *)reallocf(*dev_select, 649 numdevs * sizeof(struct device_selection)); 650 } 651 652 if (init_selections != 0) { 653 bzero(*dev_select, sizeof(struct device_selection) * numdevs); 654 655 for (i = 0; i < numdevs; i++) { 656 (*dev_select)[i].device_number = 657 devices[i].device_number; 658 strncpy((*dev_select)[i].device_name, 659 devices[i].device_name, 660 DEVSTAT_NAME_LEN); 661 (*dev_select)[i].device_name[DEVSTAT_NAME_LEN - 1]='\0'; 662 (*dev_select)[i].unit_number = devices[i].unit_number; 663 (*dev_select)[i].position = i; 664 } 665 *num_selections = numdevs; 666 } else if (init_selected_var != 0) { 667 for (i = 0; i < numdevs; i++) 668 (*dev_select)[i].selected = 0; 669 } 670 671 /* we haven't gotten around to selecting anything yet.. */ 672 if ((select_mode == DS_SELECT_ONLY) || (init_selections != 0) 673 || (init_selected_var != 0)) 674 *num_selected = 0; 675 676 /* 677 * Look through any devices the user specified on the command line 678 * and see if they match known devices. If so, select them. 679 */ 680 for (i = 0; (i < *num_selections) && (num_dev_selections > 0); i++) { 681 char tmpstr[80]; 682 683 snprintf(tmpstr, sizeof(tmpstr), "%s%d", 684 (*dev_select)[i].device_name, 685 (*dev_select)[i].unit_number); 686 for (j = 0; j < num_dev_selections; j++) { 687 if (strcmp(tmpstr, dev_selections[j]) == 0) { 688 /* 689 * Here we do different things based on the 690 * mode we're in. If we're in add or 691 * addonly mode, we only select this device 692 * if it hasn't already been selected. 693 * Otherwise, we would be unnecessarily 694 * changing the selection order and 695 * incrementing the selection count. If 696 * we're in only mode, we unconditionally 697 * select this device, since in only mode 698 * any previous selections are erased and 699 * manually specified devices are the first 700 * ones to be selected. If we're in remove 701 * mode, we de-select the specified device and 702 * decrement the selection count. 703 */ 704 switch(select_mode) { 705 case DS_SELECT_ADD: 706 case DS_SELECT_ADDONLY: 707 if ((*dev_select)[i].selected) 708 break; 709 /* FALLTHROUGH */ 710 case DS_SELECT_ONLY: 711 (*dev_select)[i].selected = 712 ++selection_number; 713 (*num_selected)++; 714 break; 715 case DS_SELECT_REMOVE: 716 (*dev_select)[i].selected = 0; 717 (*num_selected)--; 718 /* 719 * This isn't passed back out, we 720 * just use it to keep track of 721 * how many devices we've removed. 722 */ 723 num_dev_selections--; 724 break; 725 } 726 break; 727 } 728 } 729 } 730 731 /* 732 * Go through the user's device type expressions and select devices 733 * accordingly. We only do this if the number of devices already 734 * selected is less than the maximum number we can show. 735 */ 736 for (i = 0; (i < num_matches) && (*num_selected < maxshowdevs); i++) { 737 /* We should probably indicate some error here */ 738 if ((matches[i].match_fields == DEVSTAT_MATCH_NONE) 739 || (matches[i].num_match_categories <= 0)) 740 continue; 741 742 for (j = 0; j < numdevs; j++) { 743 int num_match_categories; 744 745 num_match_categories = matches[i].num_match_categories; 746 747 /* 748 * Determine whether or not the current device 749 * matches the given matching expression. This if 750 * statement consists of three components: 751 * - the device type check 752 * - the device interface check 753 * - the passthrough check 754 * If a the matching test is successful, it 755 * decrements the number of matching categories, 756 * and if we've reached the last element that 757 * needed to be matched, the if statement succeeds. 758 * 759 */ 760 if ((((matches[i].match_fields & DEVSTAT_MATCH_TYPE)!=0) 761 && ((devices[j].device_type & DEVSTAT_TYPE_MASK) == 762 (matches[i].device_type & DEVSTAT_TYPE_MASK)) 763 &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) 764 || (((matches[i].match_fields & 765 DEVSTAT_MATCH_PASS) == 0) 766 && ((devices[j].device_type & 767 DEVSTAT_TYPE_PASS) == 0))) 768 && (--num_match_categories == 0)) 769 || (((matches[i].match_fields & DEVSTAT_MATCH_IF) != 0) 770 && ((devices[j].device_type & DEVSTAT_TYPE_IF_MASK) == 771 (matches[i].device_type & DEVSTAT_TYPE_IF_MASK)) 772 &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) 773 || (((matches[i].match_fields & 774 DEVSTAT_MATCH_PASS) == 0) 775 && ((devices[j].device_type & 776 DEVSTAT_TYPE_PASS) == 0))) 777 && (--num_match_categories == 0)) 778 || (((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0) 779 && ((devices[j].device_type & DEVSTAT_TYPE_PASS) != 0) 780 && (--num_match_categories == 0))) { 781 782 /* 783 * This is probably a non-optimal solution 784 * to the problem that the devices in the 785 * device list will not be in the same 786 * order as the devices in the selection 787 * array. 788 */ 789 for (k = 0; k < numdevs; k++) { 790 if ((*dev_select)[k].position == j) { 791 found = 1; 792 break; 793 } 794 } 795 796 /* 797 * There shouldn't be a case where a device 798 * in the device list is not in the 799 * selection list...but it could happen. 800 */ 801 if (found != 1) { 802 fprintf(stderr, "selectdevs: couldn't" 803 " find %s%d in selection " 804 "list\n", 805 devices[j].device_name, 806 devices[j].unit_number); 807 break; 808 } 809 810 /* 811 * We do different things based upon the 812 * mode we're in. If we're in add or only 813 * mode, we go ahead and select this device 814 * if it hasn't already been selected. If 815 * it has already been selected, we leave 816 * it alone so we don't mess up the 817 * selection ordering. Manually specified 818 * devices have already been selected, and 819 * they have higher priority than pattern 820 * matched devices. If we're in remove 821 * mode, we de-select the given device and 822 * decrement the selected count. 823 */ 824 switch(select_mode) { 825 case DS_SELECT_ADD: 826 case DS_SELECT_ADDONLY: 827 case DS_SELECT_ONLY: 828 if ((*dev_select)[k].selected != 0) 829 break; 830 (*dev_select)[k].selected = 831 ++selection_number; 832 (*num_selected)++; 833 break; 834 case DS_SELECT_REMOVE: 835 (*dev_select)[k].selected = 0; 836 (*num_selected)--; 837 break; 838 } 839 } 840 } 841 } 842 843 /* 844 * Here we implement "top" mode. Devices are sorted in the 845 * selection array based on two criteria: whether or not they are 846 * selected (not selection number, just the fact that they are 847 * selected!) and the number of bytes in the "bytes" field of the 848 * selection structure. The bytes field generally must be kept up 849 * by the user. In the future, it may be maintained by library 850 * functions, but for now the user has to do the work. 851 * 852 * At first glance, it may seem wrong that we don't go through and 853 * select every device in the case where the user hasn't specified 854 * any devices or patterns. In fact, though, it won't make any 855 * difference in the device sorting. In that particular case (i.e. 856 * when we're in "add" or "only" mode, and the user hasn't 857 * specified anything) the first time through no devices will be 858 * selected, so the only criterion used to sort them will be their 859 * performance. The second time through, and every time thereafter, 860 * all devices will be selected, so again selection won't matter. 861 */ 862 if (perf_select != 0) { 863 864 /* Sort the device array by throughput */ 865 qsort(*dev_select, *num_selections, 866 sizeof(struct device_selection), 867 compare_select); 868 869 if (*num_selected == 0) { 870 /* 871 * Here we select every device in the array, if it 872 * isn't already selected. Because the 'selected' 873 * variable in the selection array entries contains 874 * the selection order, the devstats routine can show 875 * the devices that were selected first. 876 */ 877 for (i = 0; i < *num_selections; i++) { 878 if ((*dev_select)[i].selected == 0) { 879 (*dev_select)[i].selected = 880 ++selection_number; 881 (*num_selected)++; 882 } 883 } 884 } else { 885 selection_number = 0; 886 for (i = 0; i < *num_selections; i++) { 887 if ((*dev_select)[i].selected != 0) { 888 (*dev_select)[i].selected = 889 ++selection_number; 890 } 891 } 892 } 893 } 894 895 /* 896 * If we're in the "add" selection mode and if we haven't already 897 * selected maxshowdevs number of devices, go through the array and 898 * select any unselected devices. If we're in "only" mode, we 899 * obviously don't want to select anything other than what the user 900 * specifies. If we're in "remove" mode, it probably isn't a good 901 * idea to go through and select any more devices, since we might 902 * end up selecting something that the user wants removed. Through 903 * more complicated logic, we could actually figure this out, but 904 * that would probably require combining this loop with the various 905 * selections loops above. 906 */ 907 if ((select_mode == DS_SELECT_ADD) && (*num_selected < maxshowdevs)) { 908 for (i = 0; i < *num_selections; i++) 909 if ((*dev_select)[i].selected == 0) { 910 (*dev_select)[i].selected = ++selection_number; 911 (*num_selected)++; 912 } 913 } 914 915 /* 916 * Look at the number of devices that have been selected. If it 917 * has changed, set the changed variable. Otherwise, if we've 918 * made a backup of the selection list, compare it to the current 919 * selection list to see if the selected devices have changed. 920 */ 921 if ((changed == 0) && (old_num_selected != *num_selected)) 922 changed = 1; 923 else if ((changed == 0) && (old_dev_select != NULL)) { 924 /* 925 * Now we go through the selection list and we look at 926 * it three different ways. 927 */ 928 for (i = 0; (i < *num_selections) && (changed == 0) && 929 (i < old_num_selections); i++) { 930 /* 931 * If the device at index i in both the new and old 932 * selection arrays has the same device number and 933 * selection status, it hasn't changed. We 934 * continue on to the next index. 935 */ 936 if (((*dev_select)[i].device_number == 937 old_dev_select[i].device_number) 938 && ((*dev_select)[i].selected == 939 old_dev_select[i].selected)) 940 continue; 941 942 /* 943 * Now, if we're still going through the if 944 * statement, the above test wasn't true. So we 945 * check here to see if the device at index i in 946 * the current array is the same as the device at 947 * index i in the old array. If it is, that means 948 * that its selection number has changed. Set 949 * changed to 1 and exit the loop. 950 */ 951 else if ((*dev_select)[i].device_number == 952 old_dev_select[i].device_number) { 953 changed = 1; 954 break; 955 } 956 /* 957 * If we get here, then the device at index i in 958 * the current array isn't the same device as the 959 * device at index i in the old array. 960 */ 961 else { 962 found = 0; 963 964 /* 965 * Search through the old selection array 966 * looking for a device with the same 967 * device number as the device at index i 968 * in the current array. If the selection 969 * status is the same, then we mark it as 970 * found. If the selection status isn't 971 * the same, we break out of the loop. 972 * Since found isn't set, changed will be 973 * set to 1 below. 974 */ 975 for (j = 0; j < old_num_selections; j++) { 976 if (((*dev_select)[i].device_number == 977 old_dev_select[j].device_number) 978 && ((*dev_select)[i].selected == 979 old_dev_select[j].selected)){ 980 found = 1; 981 break; 982 } 983 else if ((*dev_select)[i].device_number 984 == old_dev_select[j].device_number) 985 break; 986 } 987 if (found == 0) 988 changed = 1; 989 } 990 } 991 } 992 if (old_dev_select != NULL) 993 free(old_dev_select); 994 995 return(changed); 996 } 997 998 /* 999 * Comparison routine for qsort() above. Note that the comparison here is 1000 * backwards -- generally, it should return a value to indicate whether 1001 * arg1 is <, =, or > arg2. Instead, it returns the opposite. The reason 1002 * it returns the opposite is so that the selection array will be sorted in 1003 * order of decreasing performance. We sort on two parameters. The first 1004 * sort key is whether or not one or the other of the devices in question 1005 * has been selected. If one of them has, and the other one has not, the 1006 * selected device is automatically more important than the unselected 1007 * device. If neither device is selected, we judge the devices based upon 1008 * performance. 1009 */ 1010 static int 1011 compare_select(const void *arg1, const void *arg2) 1012 { 1013 if ((((const struct device_selection *)arg1)->selected) 1014 && (((const struct device_selection *)arg2)->selected == 0)) 1015 return(-1); 1016 else if ((((const struct device_selection *)arg1)->selected == 0) 1017 && (((const struct device_selection *)arg2)->selected)) 1018 return(1); 1019 else if (((const struct device_selection *)arg2)->bytes < 1020 ((const struct device_selection *)arg1)->bytes) 1021 return(-1); 1022 else if (((const struct device_selection *)arg2)->bytes > 1023 ((const struct device_selection *)arg1)->bytes) 1024 return(1); 1025 else 1026 return(0); 1027 } 1028 1029 /* 1030 * Take a string with the general format "arg1,arg2,arg3", and build a 1031 * device matching expression from it. 1032 */ 1033 int 1034 devstat_buildmatch(char *match_str, struct devstat_match **matches, 1035 int *num_matches) 1036 { 1037 char *tstr[5]; 1038 char **tempstr; 1039 int num_args; 1040 int i, j; 1041 1042 /* We can't do much without a string to parse */ 1043 if (match_str == NULL) { 1044 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1045 "%s: no match expression", __func__); 1046 return(-1); 1047 } 1048 1049 /* 1050 * Break the (comma delimited) input string out into separate strings. 1051 */ 1052 for (tempstr = tstr, num_args = 0; 1053 (*tempstr = strsep(&match_str, ",")) != NULL && (num_args < 5);) 1054 if (**tempstr != '\0') { 1055 num_args++; 1056 if (++tempstr >= &tstr[5]) 1057 break; 1058 } 1059 1060 /* The user gave us too many type arguments */ 1061 if (num_args > 3) { 1062 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1063 "%s: too many type arguments", __func__); 1064 return(-1); 1065 } 1066 1067 if (*num_matches == 0) 1068 *matches = NULL; 1069 1070 *matches = (struct devstat_match *)reallocf(*matches, 1071 sizeof(struct devstat_match) * (*num_matches + 1)); 1072 1073 if (*matches == NULL) { 1074 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1075 "%s: Cannot allocate memory for matches list", __func__); 1076 return(-1); 1077 } 1078 1079 /* Make sure the current entry is clear */ 1080 bzero(&matches[0][*num_matches], sizeof(struct devstat_match)); 1081 1082 /* 1083 * Step through the arguments the user gave us and build a device 1084 * matching expression from them. 1085 */ 1086 for (i = 0; i < num_args; i++) { 1087 char *tempstr2, *tempstr3; 1088 1089 /* 1090 * Get rid of leading white space. 1091 */ 1092 tempstr2 = tstr[i]; 1093 while (isspace(*tempstr2) && (*tempstr2 != '\0')) 1094 tempstr2++; 1095 1096 /* 1097 * Get rid of trailing white space. 1098 */ 1099 tempstr3 = &tempstr2[strlen(tempstr2) - 1]; 1100 1101 while ((*tempstr3 != '\0') && (tempstr3 > tempstr2) 1102 && (isspace(*tempstr3))) { 1103 *tempstr3 = '\0'; 1104 tempstr3--; 1105 } 1106 1107 /* 1108 * Go through the match table comparing the user's 1109 * arguments to known device types, interfaces, etc. 1110 */ 1111 for (j = 0; match_table[j].match_str != NULL; j++) { 1112 /* 1113 * We do case-insensitive matching, in case someone 1114 * wants to enter "SCSI" instead of "scsi" or 1115 * something like that. Only compare as many 1116 * characters as are in the string in the match 1117 * table. This should help if someone tries to use 1118 * a super-long match expression. 1119 */ 1120 if (strncasecmp(tempstr2, match_table[j].match_str, 1121 strlen(match_table[j].match_str)) == 0) { 1122 /* 1123 * Make sure the user hasn't specified two 1124 * items of the same type, like "da" and 1125 * "cd". One device cannot be both. 1126 */ 1127 if (((*matches)[*num_matches].match_fields & 1128 match_table[j].match_field) != 0) { 1129 snprintf(devstat_errbuf, 1130 sizeof(devstat_errbuf), 1131 "%s: cannot have more than " 1132 "one match item in a single " 1133 "category", __func__); 1134 return(-1); 1135 } 1136 /* 1137 * If we've gotten this far, we have a 1138 * winner. Set the appropriate fields in 1139 * the match entry. 1140 */ 1141 (*matches)[*num_matches].match_fields |= 1142 match_table[j].match_field; 1143 (*matches)[*num_matches].device_type |= 1144 match_table[j].type; 1145 (*matches)[*num_matches].num_match_categories++; 1146 break; 1147 } 1148 } 1149 /* 1150 * We should have found a match in the above for loop. If 1151 * not, that means the user entered an invalid device type 1152 * or interface. 1153 */ 1154 if ((*matches)[*num_matches].num_match_categories != (i + 1)) { 1155 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1156 "%s: unknown match item \"%s\"", __func__, 1157 tstr[i]); 1158 return(-1); 1159 } 1160 } 1161 1162 (*num_matches)++; 1163 1164 return(0); 1165 } 1166 1167 /* 1168 * Compute a number of device statistics. Only one field is mandatory, and 1169 * that is "current". Everything else is optional. The caller passes in 1170 * pointers to variables to hold the various statistics he desires. If he 1171 * doesn't want a particular staistic, he should pass in a NULL pointer. 1172 * Return values: 1173 * 0 -- success 1174 * -1 -- failure 1175 */ 1176 int 1177 compute_stats(struct devstat *current, struct devstat *previous, 1178 long double etime, u_int64_t *total_bytes, 1179 u_int64_t *total_transfers, u_int64_t *total_blocks, 1180 long double *kb_per_transfer, long double *transfers_per_second, 1181 long double *mb_per_second, long double *blocks_per_second, 1182 long double *ms_per_transaction) 1183 { 1184 return(devstat_compute_statistics(current, previous, etime, 1185 total_bytes ? DSM_TOTAL_BYTES : DSM_SKIP, 1186 total_bytes, 1187 total_transfers ? DSM_TOTAL_TRANSFERS : DSM_SKIP, 1188 total_transfers, 1189 total_blocks ? DSM_TOTAL_BLOCKS : DSM_SKIP, 1190 total_blocks, 1191 kb_per_transfer ? DSM_KB_PER_TRANSFER : DSM_SKIP, 1192 kb_per_transfer, 1193 transfers_per_second ? DSM_TRANSFERS_PER_SECOND : DSM_SKIP, 1194 transfers_per_second, 1195 mb_per_second ? DSM_MB_PER_SECOND : DSM_SKIP, 1196 mb_per_second, 1197 blocks_per_second ? DSM_BLOCKS_PER_SECOND : DSM_SKIP, 1198 blocks_per_second, 1199 ms_per_transaction ? DSM_MS_PER_TRANSACTION : DSM_SKIP, 1200 ms_per_transaction, 1201 DSM_NONE)); 1202 } 1203 1204 1205 /* This is 1/2^64 */ 1206 #define BINTIME_SCALE 5.42101086242752217003726400434970855712890625e-20 1207 1208 long double 1209 devstat_compute_etime(struct bintime *cur_time, struct bintime *prev_time) 1210 { 1211 long double etime; 1212 1213 etime = cur_time->sec; 1214 etime += cur_time->frac * BINTIME_SCALE; 1215 if (prev_time != NULL) { 1216 etime -= prev_time->sec; 1217 etime -= prev_time->frac * BINTIME_SCALE; 1218 } 1219 return(etime); 1220 } 1221 1222 #define DELTA(field, index) \ 1223 (current->field[(index)] - (previous ? previous->field[(index)] : 0)) 1224 1225 #define DELTA_T(field) \ 1226 devstat_compute_etime(¤t->field, \ 1227 (previous ? &previous->field : NULL)) 1228 1229 int 1230 devstat_compute_statistics(struct devstat *current, struct devstat *previous, 1231 long double etime, ...) 1232 { 1233 u_int64_t totalbytes, totalbytesread, totalbyteswrite, totalbytesfree; 1234 u_int64_t totaltransfers, totaltransfersread, totaltransferswrite; 1235 u_int64_t totaltransfersother, totalblocks, totalblocksread; 1236 u_int64_t totalblockswrite, totaltransfersfree, totalblocksfree; 1237 long double totalduration, totaldurationread, totaldurationwrite; 1238 long double totaldurationfree, totaldurationother; 1239 va_list ap; 1240 devstat_metric metric; 1241 u_int64_t *destu64; 1242 long double *destld; 1243 int retval; 1244 1245 retval = 0; 1246 1247 /* 1248 * current is the only mandatory field. 1249 */ 1250 if (current == NULL) { 1251 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1252 "%s: current stats structure was NULL", __func__); 1253 return(-1); 1254 } 1255 1256 totalbytesread = DELTA(bytes, DEVSTAT_READ); 1257 totalbyteswrite = DELTA(bytes, DEVSTAT_WRITE); 1258 totalbytesfree = DELTA(bytes, DEVSTAT_FREE); 1259 totalbytes = totalbytesread + totalbyteswrite + totalbytesfree; 1260 1261 totaltransfersread = DELTA(operations, DEVSTAT_READ); 1262 totaltransferswrite = DELTA(operations, DEVSTAT_WRITE); 1263 totaltransfersother = DELTA(operations, DEVSTAT_NO_DATA); 1264 totaltransfersfree = DELTA(operations, DEVSTAT_FREE); 1265 totaltransfers = totaltransfersread + totaltransferswrite + 1266 totaltransfersother + totaltransfersfree; 1267 1268 totalblocks = totalbytes; 1269 totalblocksread = totalbytesread; 1270 totalblockswrite = totalbyteswrite; 1271 totalblocksfree = totalbytesfree; 1272 1273 if (current->block_size > 0) { 1274 totalblocks /= current->block_size; 1275 totalblocksread /= current->block_size; 1276 totalblockswrite /= current->block_size; 1277 totalblocksfree /= current->block_size; 1278 } else { 1279 totalblocks /= 512; 1280 totalblocksread /= 512; 1281 totalblockswrite /= 512; 1282 totalblocksfree /= 512; 1283 } 1284 1285 totaldurationread = DELTA_T(duration[DEVSTAT_READ]); 1286 totaldurationwrite = DELTA_T(duration[DEVSTAT_WRITE]); 1287 totaldurationfree = DELTA_T(duration[DEVSTAT_FREE]); 1288 totaldurationother = DELTA_T(duration[DEVSTAT_NO_DATA]); 1289 totalduration = totaldurationread + totaldurationwrite + 1290 totaldurationfree + totaldurationother; 1291 1292 va_start(ap, etime); 1293 1294 while ((metric = (devstat_metric)va_arg(ap, devstat_metric)) != 0) { 1295 1296 if (metric == DSM_NONE) 1297 break; 1298 1299 if (metric >= DSM_MAX) { 1300 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1301 "%s: metric %d is out of range", __func__, 1302 metric); 1303 retval = -1; 1304 goto bailout; 1305 } 1306 1307 switch (devstat_arg_list[metric].argtype) { 1308 case DEVSTAT_ARG_UINT64: 1309 destu64 = (u_int64_t *)va_arg(ap, u_int64_t *); 1310 break; 1311 case DEVSTAT_ARG_LD: 1312 destld = (long double *)va_arg(ap, long double *); 1313 break; 1314 case DEVSTAT_ARG_SKIP: 1315 destld = (long double *)va_arg(ap, long double *); 1316 break; 1317 default: 1318 retval = -1; 1319 goto bailout; 1320 break; /* NOTREACHED */ 1321 } 1322 1323 if (devstat_arg_list[metric].argtype == DEVSTAT_ARG_SKIP) 1324 continue; 1325 1326 switch (metric) { 1327 case DSM_TOTAL_BYTES: 1328 *destu64 = totalbytes; 1329 break; 1330 case DSM_TOTAL_BYTES_READ: 1331 *destu64 = totalbytesread; 1332 break; 1333 case DSM_TOTAL_BYTES_WRITE: 1334 *destu64 = totalbyteswrite; 1335 break; 1336 case DSM_TOTAL_BYTES_FREE: 1337 *destu64 = totalbytesfree; 1338 break; 1339 case DSM_TOTAL_TRANSFERS: 1340 *destu64 = totaltransfers; 1341 break; 1342 case DSM_TOTAL_TRANSFERS_READ: 1343 *destu64 = totaltransfersread; 1344 break; 1345 case DSM_TOTAL_TRANSFERS_WRITE: 1346 *destu64 = totaltransferswrite; 1347 break; 1348 case DSM_TOTAL_TRANSFERS_FREE: 1349 *destu64 = totaltransfersfree; 1350 break; 1351 case DSM_TOTAL_TRANSFERS_OTHER: 1352 *destu64 = totaltransfersother; 1353 break; 1354 case DSM_TOTAL_BLOCKS: 1355 *destu64 = totalblocks; 1356 break; 1357 case DSM_TOTAL_BLOCKS_READ: 1358 *destu64 = totalblocksread; 1359 break; 1360 case DSM_TOTAL_BLOCKS_WRITE: 1361 *destu64 = totalblockswrite; 1362 break; 1363 case DSM_TOTAL_BLOCKS_FREE: 1364 *destu64 = totalblocksfree; 1365 break; 1366 case DSM_KB_PER_TRANSFER: 1367 *destld = totalbytes; 1368 *destld /= 1024; 1369 if (totaltransfers > 0) 1370 *destld /= totaltransfers; 1371 else 1372 *destld = 0.0; 1373 break; 1374 case DSM_KB_PER_TRANSFER_READ: 1375 *destld = totalbytesread; 1376 *destld /= 1024; 1377 if (totaltransfersread > 0) 1378 *destld /= totaltransfersread; 1379 else 1380 *destld = 0.0; 1381 break; 1382 case DSM_KB_PER_TRANSFER_WRITE: 1383 *destld = totalbyteswrite; 1384 *destld /= 1024; 1385 if (totaltransferswrite > 0) 1386 *destld /= totaltransferswrite; 1387 else 1388 *destld = 0.0; 1389 break; 1390 case DSM_KB_PER_TRANSFER_FREE: 1391 *destld = totalbytesfree; 1392 *destld /= 1024; 1393 if (totaltransfersfree > 0) 1394 *destld /= totaltransfersfree; 1395 else 1396 *destld = 0.0; 1397 break; 1398 case DSM_TRANSFERS_PER_SECOND: 1399 if (etime > 0.0) { 1400 *destld = totaltransfers; 1401 *destld /= etime; 1402 } else 1403 *destld = 0.0; 1404 break; 1405 case DSM_TRANSFERS_PER_SECOND_READ: 1406 if (etime > 0.0) { 1407 *destld = totaltransfersread; 1408 *destld /= etime; 1409 } else 1410 *destld = 0.0; 1411 break; 1412 case DSM_TRANSFERS_PER_SECOND_WRITE: 1413 if (etime > 0.0) { 1414 *destld = totaltransferswrite; 1415 *destld /= etime; 1416 } else 1417 *destld = 0.0; 1418 break; 1419 case DSM_TRANSFERS_PER_SECOND_FREE: 1420 if (etime > 0.0) { 1421 *destld = totaltransfersfree; 1422 *destld /= etime; 1423 } else 1424 *destld = 0.0; 1425 break; 1426 case DSM_TRANSFERS_PER_SECOND_OTHER: 1427 if (etime > 0.0) { 1428 *destld = totaltransfersother; 1429 *destld /= etime; 1430 } else 1431 *destld = 0.0; 1432 break; 1433 case DSM_MB_PER_SECOND: 1434 *destld = totalbytes; 1435 *destld /= 1024 * 1024; 1436 if (etime > 0.0) 1437 *destld /= etime; 1438 else 1439 *destld = 0.0; 1440 break; 1441 case DSM_MB_PER_SECOND_READ: 1442 *destld = totalbytesread; 1443 *destld /= 1024 * 1024; 1444 if (etime > 0.0) 1445 *destld /= etime; 1446 else 1447 *destld = 0.0; 1448 break; 1449 case DSM_MB_PER_SECOND_WRITE: 1450 *destld = totalbyteswrite; 1451 *destld /= 1024 * 1024; 1452 if (etime > 0.0) 1453 *destld /= etime; 1454 else 1455 *destld = 0.0; 1456 break; 1457 case DSM_MB_PER_SECOND_FREE: 1458 *destld = totalbytesfree; 1459 *destld /= 1024 * 1024; 1460 if (etime > 0.0) 1461 *destld /= etime; 1462 else 1463 *destld = 0.0; 1464 break; 1465 case DSM_BLOCKS_PER_SECOND: 1466 *destld = totalblocks; 1467 if (etime > 0.0) 1468 *destld /= etime; 1469 else 1470 *destld = 0.0; 1471 break; 1472 case DSM_BLOCKS_PER_SECOND_READ: 1473 *destld = totalblocksread; 1474 if (etime > 0.0) 1475 *destld /= etime; 1476 else 1477 *destld = 0.0; 1478 break; 1479 case DSM_BLOCKS_PER_SECOND_WRITE: 1480 *destld = totalblockswrite; 1481 if (etime > 0.0) 1482 *destld /= etime; 1483 else 1484 *destld = 0.0; 1485 break; 1486 case DSM_BLOCKS_PER_SECOND_FREE: 1487 *destld = totalblocksfree; 1488 if (etime > 0.0) 1489 *destld /= etime; 1490 else 1491 *destld = 0.0; 1492 break; 1493 /* 1494 * Some devstat callers update the duration and some don't. 1495 * So this will only be accurate if they provide the 1496 * duration. 1497 */ 1498 case DSM_MS_PER_TRANSACTION: 1499 if (totaltransfers > 0) { 1500 *destld = totalduration; 1501 *destld /= totaltransfers; 1502 *destld *= 1000; 1503 } else 1504 *destld = 0.0; 1505 break; 1506 case DSM_MS_PER_TRANSACTION_READ: 1507 if (totaltransfersread > 0) { 1508 *destld = totaldurationread; 1509 *destld /= totaltransfersread; 1510 *destld *= 1000; 1511 } else 1512 *destld = 0.0; 1513 break; 1514 case DSM_MS_PER_TRANSACTION_WRITE: 1515 if (totaltransferswrite > 0) { 1516 *destld = totaldurationwrite; 1517 *destld /= totaltransferswrite; 1518 *destld *= 1000; 1519 } else 1520 *destld = 0.0; 1521 break; 1522 case DSM_MS_PER_TRANSACTION_FREE: 1523 if (totaltransfersfree > 0) { 1524 *destld = totaldurationfree; 1525 *destld /= totaltransfersfree; 1526 *destld *= 1000; 1527 } else 1528 *destld = 0.0; 1529 break; 1530 case DSM_MS_PER_TRANSACTION_OTHER: 1531 if (totaltransfersother > 0) { 1532 *destld = totaldurationother; 1533 *destld /= totaltransfersother; 1534 *destld *= 1000; 1535 } else 1536 *destld = 0.0; 1537 break; 1538 case DSM_BUSY_PCT: 1539 *destld = DELTA_T(busy_time); 1540 if (*destld < 0) 1541 *destld = 0; 1542 *destld /= etime; 1543 *destld *= 100; 1544 if (*destld < 0) 1545 *destld = 0; 1546 break; 1547 case DSM_QUEUE_LENGTH: 1548 *destu64 = current->start_count - current->end_count; 1549 break; 1550 case DSM_TOTAL_DURATION: 1551 *destld = totalduration; 1552 break; 1553 case DSM_TOTAL_DURATION_READ: 1554 *destld = totaldurationread; 1555 break; 1556 case DSM_TOTAL_DURATION_WRITE: 1557 *destld = totaldurationwrite; 1558 break; 1559 case DSM_TOTAL_DURATION_FREE: 1560 *destld = totaldurationfree; 1561 break; 1562 case DSM_TOTAL_DURATION_OTHER: 1563 *destld = totaldurationother; 1564 break; 1565 case DSM_TOTAL_BUSY_TIME: 1566 *destld = DELTA_T(busy_time); 1567 break; 1568 /* 1569 * XXX: comment out the default block to see if any case's are missing. 1570 */ 1571 #if 1 1572 default: 1573 /* 1574 * This shouldn't happen, since we should have 1575 * caught any out of range metrics at the top of 1576 * the loop. 1577 */ 1578 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1579 "%s: unknown metric %d", __func__, metric); 1580 retval = -1; 1581 goto bailout; 1582 break; /* NOTREACHED */ 1583 #endif 1584 } 1585 } 1586 1587 bailout: 1588 1589 va_end(ap); 1590 return(retval); 1591 } 1592 1593 static int 1594 readkmem(kvm_t *kd, unsigned long addr, void *buf, size_t nbytes) 1595 { 1596 1597 if (kvm_read(kd, addr, buf, nbytes) == -1) { 1598 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1599 "%s: error reading value (kvm_read): %s", __func__, 1600 kvm_geterr(kd)); 1601 return(-1); 1602 } 1603 return(0); 1604 } 1605 1606 static int 1607 readkmem_nl(kvm_t *kd, const char *name, void *buf, size_t nbytes) 1608 { 1609 struct nlist nl[2]; 1610 1611 nl[0].n_name = (char *)name; 1612 nl[1].n_name = NULL; 1613 1614 if (kvm_nlist(kd, nl) == -1) { 1615 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1616 "%s: error getting name list (kvm_nlist): %s", 1617 __func__, kvm_geterr(kd)); 1618 return(-1); 1619 } 1620 return(readkmem(kd, nl[0].n_value, buf, nbytes)); 1621 } 1622 1623 /* 1624 * This duplicates the functionality of the kernel sysctl handler for poking 1625 * through crash dumps. 1626 */ 1627 static char * 1628 get_devstat_kvm(kvm_t *kd) 1629 { 1630 int i, wp; 1631 long gen; 1632 struct devstat *nds; 1633 struct devstat ds; 1634 struct devstatlist dhead; 1635 int num_devs; 1636 char *rv = NULL; 1637 1638 if ((num_devs = devstat_getnumdevs(kd)) <= 0) 1639 return(NULL); 1640 if (KREADNL(kd, X_DEVICE_STATQ, dhead) == -1) 1641 return(NULL); 1642 1643 nds = STAILQ_FIRST(&dhead); 1644 1645 if ((rv = malloc(sizeof(gen))) == NULL) { 1646 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1647 "%s: out of memory (initial malloc failed)", 1648 __func__); 1649 return(NULL); 1650 } 1651 gen = devstat_getgeneration(kd); 1652 memcpy(rv, &gen, sizeof(gen)); 1653 wp = sizeof(gen); 1654 /* 1655 * Now push out all the devices. 1656 */ 1657 for (i = 0; (nds != NULL) && (i < num_devs); 1658 nds = STAILQ_NEXT(nds, dev_links), i++) { 1659 if (readkmem(kd, (long)nds, &ds, sizeof(ds)) == -1) { 1660 free(rv); 1661 return(NULL); 1662 } 1663 nds = &ds; 1664 rv = (char *)reallocf(rv, sizeof(gen) + 1665 sizeof(ds) * (i + 1)); 1666 if (rv == NULL) { 1667 snprintf(devstat_errbuf, sizeof(devstat_errbuf), 1668 "%s: out of memory (malloc failed)", 1669 __func__); 1670 return(NULL); 1671 } 1672 memcpy(rv + wp, &ds, sizeof(ds)); 1673 wp += sizeof(ds); 1674 } 1675 return(rv); 1676 } 1677