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