ftp: NetBSD uses __dead, and we use __dead2.
[dragonfly.git] / lib / libdevstat / devstat.c
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  * $FreeBSD: src/lib/libdevstat/devstat.c,v 1.6 1999/08/28 00:04:26 peter Exp $
29  * $DragonFly: src/lib/libdevstat/devstat.c,v 1.5 2005/01/08 19:19:26 joerg Exp $
30  */
31
32 #include <sys/types.h>
33 #include <sys/sysctl.h>
34 #include <sys/errno.h>
35
36 #include <ctype.h>
37 #include <err.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41
42 #include "devstat.h"
43
44 char devstat_errbuf[DEVSTAT_ERRBUF_SIZE];
45
46 /*
47  * Table to match descriptive strings with device types.  These are in
48  * order from most common to least common to speed search time.
49  */
50 struct devstat_match_table match_table[] = {
51         {"da",          DEVSTAT_TYPE_DIRECT,    DEVSTAT_MATCH_TYPE},
52         {"cd",          DEVSTAT_TYPE_CDROM,     DEVSTAT_MATCH_TYPE},
53         {"scsi",        DEVSTAT_TYPE_IF_SCSI,   DEVSTAT_MATCH_IF},
54         {"ide",         DEVSTAT_TYPE_IF_IDE,    DEVSTAT_MATCH_IF},
55         {"other",       DEVSTAT_TYPE_IF_OTHER,  DEVSTAT_MATCH_IF},
56         {"worm",        DEVSTAT_TYPE_WORM,      DEVSTAT_MATCH_TYPE},
57         {"sa",          DEVSTAT_TYPE_SEQUENTIAL,DEVSTAT_MATCH_TYPE},
58         {"pass",        DEVSTAT_TYPE_PASS,      DEVSTAT_MATCH_PASS},
59         {"optical",     DEVSTAT_TYPE_OPTICAL,   DEVSTAT_MATCH_TYPE},
60         {"array",       DEVSTAT_TYPE_STORARRAY, DEVSTAT_MATCH_TYPE},
61         {"changer",     DEVSTAT_TYPE_CHANGER,   DEVSTAT_MATCH_TYPE},
62         {"scanner",     DEVSTAT_TYPE_SCANNER,   DEVSTAT_MATCH_TYPE},
63         {"printer",     DEVSTAT_TYPE_PRINTER,   DEVSTAT_MATCH_TYPE},
64         {"floppy",      DEVSTAT_TYPE_FLOPPY,    DEVSTAT_MATCH_TYPE},
65         {"proc",        DEVSTAT_TYPE_PROCESSOR, DEVSTAT_MATCH_TYPE},
66         {"comm",        DEVSTAT_TYPE_COMM,      DEVSTAT_MATCH_TYPE},
67         {"enclosure",   DEVSTAT_TYPE_ENCLOSURE, DEVSTAT_MATCH_TYPE},
68         {NULL,          0,                      0}
69 };
70
71 /*
72  * Local function declarations.
73  */
74 static int compare_select(const void *arg1, const void *arg2);
75
76 int
77 getnumdevs(void)
78 {
79         size_t numdevsize;
80         int numdevs;
81         const char *func_name = "getnumdevs";
82
83         numdevsize = sizeof(int);
84
85         /*
86          * Find out how many devices we have in the system.
87          */
88         if (sysctlbyname("kern.devstat.numdevs", &numdevs,
89                          &numdevsize, NULL, 0) == -1) {
90                 sprintf(devstat_errbuf, "%s: error getting number of devices\n"
91                         "%s: %s", func_name, func_name, strerror(errno));
92                 return(-1);
93         } else
94                 return(numdevs);
95 }
96
97 /*
98  * This is an easy way to get the generation number, but the generation is
99  * supplied in a more atmoic manner by the kern.devstat.all sysctl.
100  * Because this generation sysctl is separate from the statistics sysctl,
101  * the device list and the generation could change between the time that
102  * this function is called and the device list is retreived.
103  */
104 long
105 getgeneration(void)
106 {
107         size_t gensize;
108         long generation;
109         const char *func_name = "getgeneration";
110
111         gensize = sizeof(long);
112
113         /*
114          * Get the current generation number.
115          */
116         if (sysctlbyname("kern.devstat.generation", &generation, 
117                          &gensize, NULL, 0) == -1) {
118                 sprintf(devstat_errbuf,"%s: error getting devstat generation\n"
119                         "%s: %s", func_name, func_name, strerror(errno));
120                 return(-1);
121         } else
122                 return(generation);
123 }
124
125 /*
126  * Get the current devstat version.  The return value of this function
127  * should be compared with DEVSTAT_VERSION, which is defined in
128  * sys/devicestat.h.  This will enable userland programs to determine
129  * whether they are out of sync with the kernel.
130  */
131 int
132 getversion(void)
133 {
134         size_t versize;
135         int version;
136         const char *func_name = "getversion";
137
138         versize = sizeof(int);
139
140         /*
141          * Get the current devstat version.
142          */
143         if (sysctlbyname("kern.devstat.version", &version, &versize,
144                          NULL, 0) == -1) {
145                 sprintf(devstat_errbuf, "%s: error getting devstat version\n"
146                         "%s: %s", func_name, func_name, strerror(errno));
147                 return(-1);
148         } else
149                 return(version);
150 }
151
152 /*
153  * Check the devstat version we know about against the devstat version the
154  * kernel knows about.  If they don't match, print an error into the
155  * devstat error buffer, and return -1.  If they match, return 0.
156  */
157 int
158 checkversion(void)
159 {
160         int retval = 0;
161         int errlen = 0;
162         const char *func_name = "checkversion";
163         int version;
164
165         version = getversion();
166
167         if (version != DEVSTAT_VERSION) {
168                 int buflen = 0;
169                 char tmpstr[256];
170
171                 /*
172                  * This is really pretty silly, but basically the idea is
173                  * that if getversion() returns an error (i.e. -1), then it
174                  * has printed an error message in the buffer.  Therefore,
175                  * we need to add a \n to the end of that message before we
176                  * print our own message in the buffer.
177                  */
178                 if (version == -1) {
179                         buflen = strlen(devstat_errbuf);
180                         errlen = snprintf(tmpstr, sizeof(tmpstr), "\n");
181                         strncat(devstat_errbuf, tmpstr,
182                                 DEVSTAT_ERRBUF_SIZE - buflen - 1);
183                         buflen += errlen;
184                 }
185
186                 errlen = snprintf(tmpstr, sizeof(tmpstr),
187                                   "%s: userland devstat version %d is not "
188                                   "the same as the kernel\n%s: devstat "
189                                   "version %d\n", func_name, DEVSTAT_VERSION,
190                                   func_name, version);
191
192                 if (version == -1) {
193                         strncat(devstat_errbuf, tmpstr,
194                                 DEVSTAT_ERRBUF_SIZE - buflen - 1);
195                         buflen += errlen;
196                 } else {
197                         strncpy(devstat_errbuf, tmpstr, DEVSTAT_ERRBUF_SIZE);
198                         devstat_errbuf[DEVSTAT_ERRBUF_SIZE - 1] = '\0';
199                 }
200
201                 if (version < DEVSTAT_VERSION)
202                         snprintf(tmpstr, sizeof(tmpstr),
203                                  "%s: libdevstat newer than kernel\n",
204                                  func_name);
205                 else
206                         snprintf(tmpstr, sizeof(tmpstr),
207                                  "%s: kernel newer than libdevstat\n",
208                                  func_name);
209
210                 strncat(devstat_errbuf, tmpstr,
211                         DEVSTAT_ERRBUF_SIZE - buflen - 1);
212
213                 retval = -1;
214         }
215
216         return(retval);
217 }
218
219 /*
220  * Get the current list of devices and statistics, and the current
221  * generation number.
222  * 
223  * Return values:
224  * -1  -- error
225  *  0  -- device list is unchanged
226  *  1  -- device list has changed
227  */
228 int
229 getdevs(struct statinfo *stats)
230 {
231         int error;
232         size_t dssize;
233         int oldnumdevs;
234         long oldgeneration;
235         int retval = 0;
236         struct devinfo *dinfo;
237         const char *func_name = "getdevs";
238
239         dinfo = stats->dinfo;
240
241         if (dinfo == NULL) {
242                 sprintf(devstat_errbuf, "%s: stats->dinfo was NULL", func_name);
243                 return(-1);
244         }
245
246         oldnumdevs = dinfo->numdevs;
247         oldgeneration = dinfo->generation;
248
249         /*
250          * If this is our first time through, mem_ptr will be null.  
251          */
252         if (dinfo->mem_ptr == NULL) {
253                 /*
254                  * Get the number of devices.  If it's negative, it's an
255                  * error.  Don't bother setting the error string, since
256                  * getnumdevs() has already done that for us.
257                  */
258                 if ((dinfo->numdevs = getnumdevs()) < 0)
259                         return(-1);
260
261                 /*
262                  * The kern.devstat.all sysctl returns the current generation
263                  * number, as well as all the devices.  So we need four
264                  * bytes more.
265                  */
266                 dssize =(dinfo->numdevs * sizeof(struct devstat)) +sizeof(long);
267                 dinfo->mem_ptr = (u_int8_t *)malloc(dssize);
268         } else
269                 dssize =(dinfo->numdevs * sizeof(struct devstat)) +sizeof(long);
270
271         /* Get the current time when we get the stats */
272         gettimeofday(&stats->busy_time, NULL);
273
274         /*
275          * Request all of the devices.  We only really allow for one
276          * ENOMEM failure.  It would, of course, be possible to just go in
277          * a loop and keep reallocing the device structure until we don't
278          * get ENOMEM back.  I'm not sure it's worth it, though.  If
279          * devices are being added to the system that quickly, maybe the
280          * user can just wait until all devices are added.
281          */
282         if ((error = sysctlbyname("kern.devstat.all", dinfo->mem_ptr, 
283              &dssize, NULL, 0)) == -1) {
284                 /*
285                  * If we get ENOMEM back, that means that there are 
286                  * more devices now, so we need to allocate more 
287                  * space for the device array.
288                  */
289                 if (errno == ENOMEM) {
290                         /*
291                          * No need to set the error string here, getnumdevs()
292                          * will do that if it fails.
293                          */
294                         if ((dinfo->numdevs = getnumdevs()) < 0)
295                                 return(-1);
296
297                         dssize = (dinfo->numdevs * sizeof(struct devstat)) +
298                                 sizeof(long);
299                         dinfo->mem_ptr = (u_int8_t *)realloc(dinfo->mem_ptr,
300                                                              dssize);
301                         if ((error = sysctlbyname("kern.devstat.all", 
302                             dinfo->mem_ptr, &dssize, NULL, 0)) == -1) {
303                                 sprintf(devstat_errbuf,
304                                         "%s: error getting device stats\n"
305                                         "%s: %s", func_name, func_name,
306                                         strerror(errno));
307                                 return(-1);
308                         }
309                 } else {
310                         sprintf(devstat_errbuf,
311                                 "%s: error getting device stats\n"
312                                 "%s: %s", func_name, func_name,
313                                 strerror(errno));
314                         return(-1);
315                 }
316         } 
317
318         /*
319          * The sysctl spits out the generation as the first four bytes,
320          * then all of the device statistics structures.
321          */
322         dinfo->generation = *(long *)dinfo->mem_ptr;
323
324         /*
325          * If the generation has changed, and if the current number of
326          * devices is not the same as the number of devices recorded in the
327          * devinfo structure, it is likely that the device list has shrunk.
328          * The reason that it is likely that the device list has shrunk in
329          * this case is that if the device list has grown, the sysctl above
330          * will return an ENOMEM error, and we will reset the number of
331          * devices and reallocate the device array.  If the second sysctl
332          * fails, we will return an error and therefore never get to this
333          * point.  If the device list has shrunk, the sysctl will not
334          * return an error since we have more space allocated than is
335          * necessary.  So, in the shrinkage case, we catch it here and
336          * reallocate the array so that we don't use any more space than is
337          * necessary.
338          */
339         if (oldgeneration != dinfo->generation) {
340                 if (getnumdevs() != dinfo->numdevs) {
341                         if ((dinfo->numdevs = getnumdevs()) < 0)
342                                 return(-1);
343                         dssize = (dinfo->numdevs * sizeof(struct devstat)) +
344                                 sizeof(long);
345                         dinfo->mem_ptr = (u_int8_t *)realloc(dinfo->mem_ptr,
346                                                              dssize);
347                 }
348                 retval = 1;
349         }
350
351         dinfo->devices = (struct devstat *)(dinfo->mem_ptr + sizeof(long));
352
353         return(retval);
354 }
355
356 /*
357  * selectdevs():
358  *
359  * Devices are selected/deselected based upon the following criteria:
360  * - devices specified by the user on the command line
361  * - devices matching any device type expressions given on the command line
362  * - devices with the highest I/O, if 'top' mode is enabled
363  * - the first n unselected devices in the device list, if maxshowdevs
364  *   devices haven't already been selected and if the user has not
365  *   specified any devices on the command line and if we're in "add" mode.
366  *
367  * Input parameters:
368  * - device selection list (dev_select)
369  * - current number of devices selected (num_selected)
370  * - total number of devices in the selection list (num_selections)
371  * - devstat generation as of the last time selectdevs() was called
372  *   (select_generation)
373  * - current devstat generation (current_generation)
374  * - current list of devices and statistics (devices)
375  * - number of devices in the current device list (numdevs)
376  * - compiled version of the command line device type arguments (matches)
377  *   - This is optional.  If the number of devices is 0, this will be ignored.
378  *   - The matching code pays attention to the current selection mode.  So
379  *     if you pass in a matching expression, it will be evaluated based
380  *     upon the selection mode that is passed in.  See below for details.
381  * - number of device type matching expressions (num_matches)
382  *   - Set to 0 to disable the matching code.
383  * - list of devices specified on the command line by the user (dev_selections)
384  * - number of devices selected on the command line by the user
385  *   (num_dev_selections)
386  * - Our selection mode.  There are four different selection modes:
387  *      - add mode.  (DS_SELECT_ADD) Any devices matching devices explicitly
388  *        selected by the user or devices matching a pattern given by the
389  *        user will be selected in addition to devices that are already
390  *        selected.  Additional devices will be selected, up to maxshowdevs
391  *        number of devices. 
392  *      - only mode. (DS_SELECT_ONLY)  Only devices matching devices
393  *        explicitly given by the user or devices matching a pattern
394  *        given by the user will be selected.  No other devices will be
395  *        selected.
396  *      - addonly mode.  (DS_SELECT_ADDONLY)  This is similar to add and
397  *        only.  Basically, this will not de-select any devices that are
398  *        current selected, as only mode would, but it will also not
399  *        gratuitously select up to maxshowdevs devices as add mode would.
400  *      - remove mode.  (DS_SELECT_REMOVE)  Any devices matching devices
401  *        explicitly selected by the user or devices matching a pattern
402  *        given by the user will be de-selected.
403  * - maximum number of devices we can select (maxshowdevs)
404  * - flag indicating whether or not we're in 'top' mode (perf_select)
405  *
406  * Output data:
407  * - the device selection list may be modified and passed back out
408  * - the number of devices selected and the total number of items in the
409  *   device selection list may be changed
410  * - the selection generation may be changed to match the current generation
411  * 
412  * Return values:
413  * -1  -- error
414  *  0  -- selected devices are unchanged
415  *  1  -- selected devices changed
416  */
417 int
418 selectdevs(struct device_selection **dev_select, int *num_selected,
419            int *num_selections, long *select_generation, 
420            long current_generation, struct devstat *devices, int numdevs,
421            struct devstat_match *matches, int num_matches,
422            char **dev_selections, int num_dev_selections,
423            devstat_select_mode select_mode, int maxshowdevs, int perf_select)
424 {
425         int i, j, k;
426         int init_selections = 0, init_selected_var = 0;
427         struct device_selection *old_dev_select = NULL;
428         int old_num_selections = 0, old_num_selected;
429         int selection_number = 0;
430         int changed = 0, found = 0;
431
432         if ((dev_select == NULL) || (devices == NULL) || (numdevs <= 0))
433                 return(-1);
434
435         /*
436          * We always want to make sure that we have as many dev_select
437          * entries as there are devices. 
438          */
439         /*
440          * In this case, we haven't selected devices before.
441          */
442         if (*dev_select == NULL) {
443                 *dev_select = (struct device_selection *)malloc(numdevs *
444                         sizeof(struct device_selection));
445                 *select_generation = current_generation;
446                 init_selections = 1;
447                 changed = 1;
448         /*
449          * In this case, we have selected devices before, but the device
450          * list has changed since we last selected devices, so we need to
451          * either enlarge or reduce the size of the device selection list.
452          */
453         } else if (*num_selections != numdevs) {
454                 *dev_select = (struct device_selection *)realloc(*dev_select,
455                         numdevs * sizeof(struct device_selection));
456                 *select_generation = current_generation;
457                 init_selections = 1;
458         /*
459          * In this case, we've selected devices before, and the selection
460          * list is the same size as it was the last time, but the device
461          * list has changed.
462          */
463         } else if (*select_generation < current_generation) {
464                 *select_generation = current_generation;
465                 init_selections = 1;
466         }
467
468         /*
469          * If we're in "only" mode, we want to clear out the selected
470          * variable since we're going to select exactly what the user wants
471          * this time through.
472          */
473         if (select_mode == DS_SELECT_ONLY)
474                 init_selected_var = 1;
475
476         /*
477          * In all cases, we want to back up the number of selected devices.
478          * It is a quick and accurate way to determine whether the selected
479          * devices have changed.
480          */
481         old_num_selected = *num_selected;
482
483         /*
484          * We want to make a backup of the current selection list if 
485          * the list of devices has changed, or if we're in performance 
486          * selection mode.  In both cases, we don't want to make a backup
487          * if we already know for sure that the list will be different.
488          * This is certainly the case if this is our first time through the
489          * selection code.
490          */
491         if (((init_selected_var != 0) || (init_selections != 0)
492          || (perf_select != 0)) && (changed == 0)){
493                 old_dev_select = (struct device_selection *)malloc(
494                     *num_selections * sizeof(struct device_selection));
495                 old_num_selections = *num_selections;
496                 bcopy(*dev_select, old_dev_select, 
497                     sizeof(struct device_selection) * *num_selections);
498         }
499
500         if (init_selections != 0) {
501                 bzero(*dev_select, sizeof(struct device_selection) * numdevs);
502
503                 for (i = 0; i < numdevs; i++) {
504                         (*dev_select)[i].device_number = 
505                                 devices[i].device_number;
506                         strncpy((*dev_select)[i].device_name,
507                                 devices[i].device_name,
508                                 DEVSTAT_NAME_LEN);
509                         (*dev_select)[i].device_name[DEVSTAT_NAME_LEN - 1]='\0';
510                         (*dev_select)[i].unit_number = devices[i].unit_number;
511                         (*dev_select)[i].position = i;
512                 }
513                 *num_selections = numdevs;
514         } else if (init_selected_var != 0) {
515                 for (i = 0; i < numdevs; i++) 
516                         (*dev_select)[i].selected = 0;
517         }
518
519         /* we haven't gotten around to selecting anything yet.. */
520         if ((select_mode == DS_SELECT_ONLY) || (init_selections != 0)
521          || (init_selected_var != 0))
522                 *num_selected = 0;
523
524         /*
525          * Look through any devices the user specified on the command line
526          * and see if they match known devices.  If so, select them.
527          */
528         for (i = 0; (i < *num_selections) && (num_dev_selections > 0); i++) {
529                 char tmpstr[80];
530
531                 snprintf(tmpstr, sizeof(tmpstr), "%s%d",
532                         (*dev_select)[i].device_name,
533                         (*dev_select)[i].unit_number);
534                 for (j = 0; j < num_dev_selections; j++) {
535                         if (strcmp(tmpstr, dev_selections[j]) == 0) {
536                                 /*
537                                  * Here we do different things based on the
538                                  * mode we're in.  If we're in add or
539                                  * addonly mode, we only select this device
540                                  * if it hasn't already been selected.
541                                  * Otherwise, we would be unnecessarily
542                                  * changing the selection order and
543                                  * incrementing the selection count.  If
544                                  * we're in only mode, we unconditionally
545                                  * select this device, since in only mode
546                                  * any previous selections are erased and
547                                  * manually specified devices are the first
548                                  * ones to be selected.  If we're in remove
549                                  * mode, we de-select the specified device and
550                                  * decrement the selection count.
551                                  */
552                                 switch(select_mode) {
553                                 case DS_SELECT_ADD:
554                                 case DS_SELECT_ADDONLY:
555                                         if ((*dev_select)[i].selected)
556                                                 break;
557                                         /* FALLTHROUGH */
558                                 case DS_SELECT_ONLY:
559                                         (*dev_select)[i].selected =
560                                                 ++selection_number;
561                                         (*num_selected)++;
562                                         break;
563                                 case DS_SELECT_REMOVE:
564                                         (*dev_select)[i].selected = 0;
565                                         (*num_selected)--;
566                                         /*
567                                          * This isn't passed back out, we
568                                          * just use it to keep track of
569                                          * how many devices we've removed.
570                                          */
571                                         num_dev_selections--;
572                                         break;
573                                 }
574                                 break;
575                         }
576                 }
577         }
578
579         /*
580          * Go through the user's device type expressions and select devices
581          * accordingly.  We only do this if the number of devices already
582          * selected is less than the maximum number we can show.
583          */
584         for (i = 0; (i < num_matches) && (*num_selected < maxshowdevs); i++) {
585                 /* We should probably indicate some error here */
586                 if ((matches[i].match_fields == DEVSTAT_MATCH_NONE)
587                  || (matches[i].num_match_categories <= 0))
588                         continue;
589
590                 for (j = 0; j < numdevs; j++) {
591                         int num_match_categories;
592
593                         num_match_categories = matches[i].num_match_categories;
594
595                         /*
596                          * Determine whether or not the current device
597                          * matches the given matching expression.  This if
598                          * statement consists of three components:
599                          *   - the device type check
600                          *   - the device interface check
601                          *   - the passthrough check
602                          * If a the matching test is successful, it 
603                          * decrements the number of matching categories,
604                          * and if we've reached the last element that
605                          * needed to be matched, the if statement succeeds.
606                          * 
607                          */
608                         if ((((matches[i].match_fields & DEVSTAT_MATCH_TYPE)!=0)
609                           && ((devices[j].device_type & DEVSTAT_TYPE_MASK) ==
610                                 (matches[i].device_type & DEVSTAT_TYPE_MASK))
611                           &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0)
612                            || (((matches[i].match_fields & 
613                                 DEVSTAT_MATCH_PASS) == 0)
614                             && ((devices[j].device_type &
615                                 DEVSTAT_TYPE_PASS) == 0)))
616                           && (--num_match_categories == 0)) 
617                          || (((matches[i].match_fields & DEVSTAT_MATCH_IF) != 0)
618                           && ((devices[j].device_type & DEVSTAT_TYPE_IF_MASK) ==
619                                 (matches[i].device_type & DEVSTAT_TYPE_IF_MASK))
620                           &&(((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0)
621                            || (((matches[i].match_fields &
622                                 DEVSTAT_MATCH_PASS) == 0)
623                             && ((devices[j].device_type & 
624                                 DEVSTAT_TYPE_PASS) == 0)))
625                           && (--num_match_categories == 0))
626                          || (((matches[i].match_fields & DEVSTAT_MATCH_PASS)!=0)
627                           && ((devices[j].device_type & DEVSTAT_TYPE_PASS) != 0)
628                           && (--num_match_categories == 0))) {
629
630                                 /*
631                                  * This is probably a non-optimal solution
632                                  * to the problem that the devices in the
633                                  * device list will not be in the same
634                                  * order as the devices in the selection
635                                  * array.
636                                  */
637                                 for (k = 0; k < numdevs; k++) {
638                                         if ((*dev_select)[k].position == j) {
639                                                 found = 1;
640                                                 break;
641                                         }
642                                 }
643
644                                 /*
645                                  * There shouldn't be a case where a device
646                                  * in the device list is not in the
647                                  * selection list...but it could happen.
648                                  */
649                                 if (found != 1) {
650                                         fprintf(stderr, "selectdevs: couldn't"
651                                                 " find %s%d in selection "
652                                                 "list\n",
653                                                 devices[j].device_name,
654                                                 devices[j].unit_number);
655                                         break;
656                                 }
657
658                                 /*
659                                  * We do different things based upon the
660                                  * mode we're in.  If we're in add or only
661                                  * mode, we go ahead and select this device
662                                  * if it hasn't already been selected.  If
663                                  * it has already been selected, we leave
664                                  * it alone so we don't mess up the
665                                  * selection ordering.  Manually specified
666                                  * devices have already been selected, and
667                                  * they have higher priority than pattern
668                                  * matched devices.  If we're in remove
669                                  * mode, we de-select the given device and
670                                  * decrement the selected count.
671                                  */
672                                 switch(select_mode) {
673                                 case DS_SELECT_ADD:
674                                 case DS_SELECT_ADDONLY:
675                                 case DS_SELECT_ONLY:
676                                         if ((*dev_select)[k].selected != 0)
677                                                 break;
678                                         (*dev_select)[k].selected =
679                                                 ++selection_number;
680                                         (*num_selected)++;
681                                         break;
682                                 case DS_SELECT_REMOVE:
683                                         (*dev_select)[k].selected = 0;
684                                         (*num_selected)--;
685                                         break;
686                                 }
687                         }
688                 }
689         }
690
691         /*
692          * Here we implement "top" mode.  Devices are sorted in the
693          * selection array based on two criteria:  whether or not they are
694          * selected (not selection number, just the fact that they are
695          * selected!) and the number of bytes in the "bytes" field of the
696          * selection structure.  The bytes field generally must be kept up
697          * by the user.  In the future, it may be maintained by library
698          * functions, but for now the user has to do the work.
699          *
700          * At first glance, it may seem wrong that we don't go through and
701          * select every device in the case where the user hasn't specified
702          * any devices or patterns.  In fact, though, it won't make any
703          * difference in the device sorting.  In that particular case (i.e.
704          * when we're in "add" or "only" mode, and the user hasn't
705          * specified anything) the first time through no devices will be
706          * selected, so the only criterion used to sort them will be their
707          * performance.  The second time through, and every time thereafter,
708          * all devices will be selected, so again selection won't matter.
709          */
710         if (perf_select != 0) {
711
712                 /* Sort the device array by throughput  */
713                 qsort(*dev_select, *num_selections,
714                       sizeof(struct device_selection),
715                       compare_select);
716
717                 if (*num_selected == 0) {
718                         /*
719                          * Here we select every device in the array, if it
720                          * isn't already selected.  Because the 'selected'
721                          * variable in the selection array entries contains
722                          * the selection order, the devstats routine can show
723                          * the devices that were selected first.
724                          */
725                         for (i = 0; i < *num_selections; i++) {
726                                 if ((*dev_select)[i].selected == 0) {
727                                         (*dev_select)[i].selected =
728                                                 ++selection_number;
729                                         (*num_selected)++;
730                                 }
731                         }
732                 } else {
733                         selection_number = 0;
734                         for (i = 0; i < *num_selections; i++) {
735                                 if ((*dev_select)[i].selected != 0) {
736                                         (*dev_select)[i].selected =
737                                                 ++selection_number;
738                                 }
739                         }
740                 }
741         }
742
743         /*
744          * If we're in the "add" selection mode and if we haven't already
745          * selected maxshowdevs number of devices, go through the array and
746          * select any unselected devices.  If we're in "only" mode, we
747          * obviously don't want to select anything other than what the user
748          * specifies.  If we're in "remove" mode, it probably isn't a good
749          * idea to go through and select any more devices, since we might
750          * end up selecting something that the user wants removed.  Through
751          * more complicated logic, we could actually figure this out, but
752          * that would probably require combining this loop with the various
753          * selections loops above.
754          */
755         if ((select_mode == DS_SELECT_ADD) && (*num_selected < maxshowdevs)) {
756                 for (i = 0; i < *num_selections; i++)
757                         if ((*dev_select)[i].selected == 0) {
758                                 (*dev_select)[i].selected = ++selection_number;
759                                 (*num_selected)++;
760                         }
761         }
762
763         /*
764          * Look at the number of devices that have been selected.  If it
765          * has changed, set the changed variable.  Otherwise, if we've
766          * made a backup of the selection list, compare it to the current
767          * selection list to see if the selected devices have changed.
768          */
769         if ((changed == 0) && (old_num_selected != *num_selected))
770                 changed = 1;
771         else if ((changed == 0) && (old_dev_select != NULL)) {
772                 /*
773                  * Now we go through the selection list and we look at
774                  * it three different ways.
775                  */
776                 for (i = 0; (i < *num_selections) && (changed == 0) && 
777                      (i < old_num_selections); i++) {
778                         /*
779                          * If the device at index i in both the new and old
780                          * selection arrays has the same device number and
781                          * selection status, it hasn't changed.  We
782                          * continue on to the next index.
783                          */
784                         if (((*dev_select)[i].device_number ==
785                              old_dev_select[i].device_number)
786                          && ((*dev_select)[i].selected == 
787                              old_dev_select[i].selected))
788                                 continue;
789
790                         /*
791                          * Now, if we're still going through the if
792                          * statement, the above test wasn't true.  So we
793                          * check here to see if the device at index i in
794                          * the current array is the same as the device at
795                          * index i in the old array.  If it is, that means
796                          * that its selection number has changed.  Set
797                          * changed to 1 and exit the loop.
798                          */
799                         else if ((*dev_select)[i].device_number ==
800                                   old_dev_select[i].device_number) {
801                                 changed = 1;
802                                 break;
803                         }
804                         /*
805                          * If we get here, then the device at index i in
806                          * the current array isn't the same device as the
807                          * device at index i in the old array.
808                          */
809                         else {
810                                 found = 0;
811
812                                 /*
813                                  * Search through the old selection array
814                                  * looking for a device with the same
815                                  * device number as the device at index i
816                                  * in the current array.  If the selection
817                                  * status is the same, then we mark it as
818                                  * found.  If the selection status isn't
819                                  * the same, we break out of the loop.
820                                  * Since found isn't set, changed will be
821                                  * set to 1 below.
822                                  */
823                                 for (j = 0; j < old_num_selections; j++) {
824                                         if (((*dev_select)[i].device_number ==
825                                               old_dev_select[j].device_number)
826                                          && ((*dev_select)[i].selected ==
827                                               old_dev_select[j].selected)){
828                                                 found = 1;
829                                                 break;
830                                         }
831                                         else if ((*dev_select)[i].device_number
832                                             == old_dev_select[j].device_number)
833                                                 break;
834                                 }
835                                 if (found == 0)
836                                         changed = 1;
837                         }
838                 }
839         }
840         if (old_dev_select != NULL)
841                 free(old_dev_select);
842
843         return(changed);
844 }
845
846 /*
847  * Comparison routine for qsort() above.  Note that the comparison here is
848  * backwards -- generally, it should return a value to indicate whether
849  * arg1 is <, =, or > arg2.  Instead, it returns the opposite.  The reason
850  * it returns the opposite is so that the selection array will be sorted in
851  * order of decreasing performance.  We sort on two parameters.  The first
852  * sort key is whether or not one or the other of the devices in question
853  * has been selected.  If one of them has, and the other one has not, the
854  * selected device is automatically more important than the unselected
855  * device.  If neither device is selected, we judge the devices based upon
856  * performance.
857  */
858 static int
859 compare_select(const void *arg1, const void *arg2)
860 {
861         if ((((const struct device_selection *)arg1)->selected)
862          && (((const struct device_selection *)arg2)->selected == 0))
863                 return(-1);
864         else if ((((const struct device_selection *)arg1)->selected == 0)
865               && (((const struct device_selection *)arg2)->selected))
866                 return(1);
867         else if (((const struct device_selection *)arg2)->bytes <
868                  ((const struct device_selection *)arg1)->bytes)
869                 return(-1);
870         else if (((const struct device_selection *)arg2)->bytes >
871                  ((const struct device_selection *)arg1)->bytes)
872                 return(1);
873         else
874                 return(0);
875 }
876
877 /*
878  * Take a string with the general format "arg1,arg2,arg3", and build a
879  * device matching expression from it.
880  */
881 int
882 buildmatch(const char *match_str, struct devstat_match **matches,
883            int *num_matches)
884 {
885         char *tstr[5];
886         char **tempstr;
887         char *matchbuf_orig;    /* strdup of match_str */
888         char *matchbuf;         /* allow strsep to clobber */
889         int num_args;
890         int i, j;
891         int retval = -1;
892
893         /* We can't do much without a string to parse */
894         if (match_str == NULL) {
895                 sprintf(devstat_errbuf, "%s: no match expression", __func__);
896                 return(-1);
897         }
898
899         /*
900          * Break the (comma delimited) input string out into separate strings.
901          * strsep is destructive, so copy the string first.
902          */
903         matchbuf = matchbuf_orig = strdup(match_str);
904         if (matchbuf == NULL) {
905                 sprintf(devstat_errbuf, "%s: out of memory", __func__);
906                 return(-1);
907         }
908         for (tempstr = tstr, num_args  = 0; 
909              (*tempstr = strsep(&matchbuf, ",")) != NULL && (num_args < 5); 
910              num_args++)
911                 if (**tempstr != '\0')
912                         if (++tempstr >= &tstr[5])
913                                 break;
914
915         /* The user gave us too many type arguments */
916         if (num_args > 3) {
917                 sprintf(devstat_errbuf, "%s: too many type arguments",
918                         __func__);
919                 goto cleanup;
920         }
921
922         /*
923          * Since you can't realloc a pointer that hasn't been malloced
924          * first, we malloc first and then realloc.
925          */
926         if (*num_matches == 0)
927                 *matches = (struct devstat_match *)malloc(
928                            sizeof(struct devstat_match));
929         else
930                 *matches = (struct devstat_match *)realloc(*matches,
931                           sizeof(struct devstat_match) * (*num_matches + 1));
932                           
933         /* Make sure the current entry is clear */
934         bzero(&matches[0][*num_matches], sizeof(struct devstat_match));
935
936         /*
937          * Step through the arguments the user gave us and build a device
938          * matching expression from them.
939          */
940         for (i = 0; i < num_args; i++) {
941                 char *tempstr2, *tempstr3;
942
943                 /*
944                  * Get rid of leading white space.
945                  */
946                 tempstr2 = tstr[i];
947                 while (isspace(*tempstr2) && (*tempstr2 != '\0'))
948                         tempstr2++;
949
950                 /*
951                  * Get rid of trailing white space.
952                  */
953                 tempstr3 = &tempstr2[strlen(tempstr2) - 1];
954
955                 while ((*tempstr3 != '\0') && (tempstr3 > tempstr2)
956                     && (isspace(*tempstr3))) {
957                         *tempstr3 = '\0';
958                         tempstr3--;
959                 }
960
961                 /*
962                  * Go through the match table comparing the user's
963                  * arguments to known device types, interfaces, etc.  
964                  */
965                 for (j = 0; match_table[j].match_str != NULL; j++) {
966                         /*
967                          * We do case-insensitive matching, in case someone
968                          * wants to enter "SCSI" instead of "scsi" or
969                          * something like that.  Only compare as many 
970                          * characters as are in the string in the match 
971                          * table.  This should help if someone tries to use 
972                          * a super-long match expression.  
973                          */
974                         if (strncasecmp(tempstr2, match_table[j].match_str,
975                             strlen(match_table[j].match_str)) == 0) {
976                                 /*
977                                  * Make sure the user hasn't specified two
978                                  * items of the same type, like "da" and
979                                  * "cd".  One device cannot be both.
980                                  */
981                                 if (((*matches)[*num_matches].match_fields &
982                                     match_table[j].match_field) != 0) {
983                                         sprintf(devstat_errbuf,
984                                                 "%s: cannot have more than "
985                                                 "one match item in a single "
986                                                 "category", __func__);
987                                         goto cleanup;
988                                 }
989                                 /*
990                                  * If we've gotten this far, we have a
991                                  * winner.  Set the appropriate fields in
992                                  * the match entry.
993                                  */
994                                 (*matches)[*num_matches].match_fields |=
995                                         match_table[j].match_field;
996                                 (*matches)[*num_matches].device_type |=
997                                         match_table[j].type;
998                                 (*matches)[*num_matches].num_match_categories++;
999                                 break;
1000                         }
1001                 }
1002                 /*
1003                  * We should have found a match in the above for loop.  If
1004                  * not, that means the user entered an invalid device type
1005                  * or interface.
1006                  */
1007                 if ((*matches)[*num_matches].num_match_categories != (i + 1)) {
1008                         snprintf(devstat_errbuf, sizeof(devstat_errbuf),
1009                                 "%s: unknown match item \"%s\"", __func__,
1010                                 tstr[i]);
1011                         goto cleanup;
1012                 }
1013         }
1014
1015         (*num_matches)++;
1016         retval = 0;
1017 cleanup:
1018         free(matchbuf_orig);
1019         return(retval);
1020 }
1021
1022 /*
1023  * Compute a number of device statistics.  Only one field is mandatory, and
1024  * that is "current".  Everything else is optional.  The caller passes in
1025  * pointers to variables to hold the various statistics he desires.  If he
1026  * doesn't want a particular staistic, he should pass in a NULL pointer.
1027  * Return values:
1028  * 0   -- success
1029  * -1  -- failure
1030  */
1031 int
1032 compute_stats(struct devstat *current, struct devstat *previous,
1033               long double etime, u_int64_t *total_bytes,
1034               u_int64_t *total_transfers, u_int64_t *total_blocks,
1035               long double *kb_per_transfer, long double *transfers_per_second,
1036               long double *mb_per_second, long double *blocks_per_second,
1037               long double *ms_per_transaction)
1038 {
1039         u_int64_t totalbytes, totaltransfers, totalblocks;
1040
1041         /*
1042          * current is the only mandatory field.
1043          */
1044         if (current == NULL) {
1045                 sprintf(devstat_errbuf, "%s: current stats structure was NULL",
1046                         __func__);
1047                 return(-1);
1048         }
1049
1050         totalbytes = (current->bytes_written + current->bytes_read) -
1051                      ((previous) ? (previous->bytes_written +
1052                                     previous->bytes_read) : 0);
1053
1054         if (total_bytes)
1055                 *total_bytes = totalbytes;
1056
1057         totaltransfers = (current->num_reads +
1058                           current->num_writes +
1059                           current->num_other) -
1060                          ((previous) ?
1061                           (previous->num_reads +
1062                            previous->num_writes +
1063                            previous->num_other) : 0);
1064         if (total_transfers)
1065                 *total_transfers = totaltransfers;
1066
1067         if (transfers_per_second) {
1068                 if (etime > 0.0) {
1069                         *transfers_per_second = totaltransfers;
1070                         *transfers_per_second /= etime;
1071                 } else
1072                         *transfers_per_second = 0.0;
1073         }
1074
1075         if (kb_per_transfer) {
1076                 *kb_per_transfer = totalbytes;
1077                 *kb_per_transfer /= 1024;
1078                 if (totaltransfers > 0)
1079                         *kb_per_transfer /= totaltransfers;
1080                 else
1081                         *kb_per_transfer = 0.0;
1082         }
1083
1084         if (mb_per_second) {
1085                 *mb_per_second = totalbytes;
1086                 *mb_per_second /= 1024 * 1024;
1087                 if (etime > 0.0)
1088                         *mb_per_second /= etime;
1089                 else
1090                         *mb_per_second = 0.0;
1091         }
1092
1093         totalblocks = totalbytes;
1094         if (current->block_size > 0)
1095                 totalblocks /= current->block_size;
1096         else
1097                 totalblocks /= 512;
1098
1099         if (total_blocks)
1100                 *total_blocks = totalblocks;
1101
1102         if (blocks_per_second) {
1103                 *blocks_per_second = totalblocks;
1104                 if (etime > 0.0)
1105                         *blocks_per_second /= etime;
1106                 else
1107                         *blocks_per_second = 0.0;
1108         }
1109
1110         if (ms_per_transaction) {
1111                 if (totaltransfers > 0) {
1112                         *ms_per_transaction = etime;
1113                         *ms_per_transaction /= totaltransfers;
1114                         *ms_per_transaction *= 1000;
1115                 } else
1116                         *ms_per_transaction = 0.0;
1117         }
1118
1119         return(0);
1120 }
1121
1122 int
1123 compute_stats_read(struct devstat *current, struct devstat *previous,
1124               long double etime, u_int64_t *total_bytes,
1125               u_int64_t *total_transfers, u_int64_t *total_blocks,
1126               long double *kb_per_transfer, long double *transfers_per_second,
1127               long double *mb_per_second, long double *blocks_per_second,
1128               long double *ms_per_transaction)
1129 {
1130         u_int64_t totalbytes, totaltransfers, totalblocks;
1131
1132         /*
1133          * current is the only mandatory field.
1134          */
1135         if (current == NULL) {
1136                 sprintf(devstat_errbuf, "%s: current stats structure was NULL",
1137                         __func__);
1138                 return(-1);
1139         }
1140
1141         totalbytes = current->bytes_read -
1142                      (previous ? previous->bytes_read : 0);
1143
1144         if (total_bytes)
1145                 *total_bytes = totalbytes;
1146
1147         totaltransfers = current->num_reads -
1148                          (previous ? previous->num_reads : 0);
1149         if (total_transfers)
1150                 *total_transfers = totaltransfers;
1151
1152         if (transfers_per_second) {
1153                 if (etime > 0.0) {
1154                         *transfers_per_second = totaltransfers;
1155                         *transfers_per_second /= etime;
1156                 } else
1157                         *transfers_per_second = 0.0;
1158         }
1159
1160         if (kb_per_transfer) {
1161                 *kb_per_transfer = totalbytes;
1162                 *kb_per_transfer /= 1024;
1163                 if (totaltransfers > 0)
1164                         *kb_per_transfer /= totaltransfers;
1165                 else
1166                         *kb_per_transfer = 0.0;
1167         }
1168
1169         if (mb_per_second) {
1170                 *mb_per_second = totalbytes;
1171                 *mb_per_second /= 1024 * 1024;
1172                 if (etime > 0.0)
1173                         *mb_per_second /= etime;
1174                 else
1175                         *mb_per_second = 0.0;
1176         }
1177
1178         totalblocks = totalbytes;
1179         if (current->block_size > 0)
1180                 totalblocks /= current->block_size;
1181         else
1182                 totalblocks /= 512;
1183
1184         if (total_blocks)
1185                 *total_blocks = totalblocks;
1186
1187         if (blocks_per_second) {
1188                 *blocks_per_second = totalblocks;
1189                 if (etime > 0.0)
1190                         *blocks_per_second /= etime;
1191                 else
1192                         *blocks_per_second = 0.0;
1193         }
1194
1195         if (ms_per_transaction) {
1196                 if (totaltransfers > 0) {
1197                         *ms_per_transaction = etime;
1198                         *ms_per_transaction /= totaltransfers;
1199                         *ms_per_transaction *= 1000;
1200                 } else
1201                         *ms_per_transaction = 0.0;
1202         }
1203
1204         return(0);
1205 }
1206
1207 int
1208 compute_stats_write(struct devstat *current, struct devstat *previous,
1209               long double etime, u_int64_t *total_bytes,
1210               u_int64_t *total_transfers, u_int64_t *total_blocks,
1211               long double *kb_per_transfer, long double *transfers_per_second,
1212               long double *mb_per_second, long double *blocks_per_second,
1213               long double *ms_per_transaction)
1214 {
1215         u_int64_t totalbytes, totaltransfers, totalblocks;
1216
1217         /*
1218          * current is the only mandatory field.
1219          */
1220         if (current == NULL) {
1221                 sprintf(devstat_errbuf, "%s: current stats structure was NULL",
1222                         __func__);
1223                 return(-1);
1224         }
1225
1226         totalbytes = current->bytes_written -
1227                      (previous ? previous->bytes_written : 0);
1228
1229         if (total_bytes)
1230                 *total_bytes = totalbytes;
1231
1232         totaltransfers = current->num_writes -
1233                          (previous ? previous->num_writes : 0);
1234         if (total_transfers)
1235                 *total_transfers = totaltransfers;
1236
1237         if (transfers_per_second) {
1238                 if (etime > 0.0) {
1239                         *transfers_per_second = totaltransfers;
1240                         *transfers_per_second /= etime;
1241                 } else
1242                         *transfers_per_second = 0.0;
1243         }
1244
1245         if (kb_per_transfer) {
1246                 *kb_per_transfer = totalbytes;
1247                 *kb_per_transfer /= 1024;
1248                 if (totaltransfers > 0)
1249                         *kb_per_transfer /= totaltransfers;
1250                 else
1251                         *kb_per_transfer = 0.0;
1252         }
1253
1254         if (mb_per_second) {
1255                 *mb_per_second = totalbytes;
1256                 *mb_per_second /= 1024 * 1024;
1257                 if (etime > 0.0)
1258                         *mb_per_second /= etime;
1259                 else
1260                         *mb_per_second = 0.0;
1261         }
1262
1263         totalblocks = totalbytes;
1264         if (current->block_size > 0)
1265                 totalblocks /= current->block_size;
1266         else
1267                 totalblocks /= 512;
1268
1269         if (total_blocks)
1270                 *total_blocks = totalblocks;
1271
1272         if (blocks_per_second) {
1273                 *blocks_per_second = totalblocks;
1274                 if (etime > 0.0)
1275                         *blocks_per_second /= etime;
1276                 else
1277                         *blocks_per_second = 0.0;
1278         }
1279
1280         if (ms_per_transaction) {
1281                 if (totaltransfers > 0) {
1282                         *ms_per_transaction = etime;
1283                         *ms_per_transaction /= totaltransfers;
1284                         *ms_per_transaction *= 1000;
1285                 } else
1286                         *ms_per_transaction = 0.0;
1287         }
1288
1289         return(0);
1290 }
1291
1292 long double
1293 compute_etime(struct timeval cur_time, struct timeval prev_time)
1294 {
1295         struct timeval busy_time;
1296         u_int64_t busy_usec;
1297         long double etime;
1298
1299         timersub(&cur_time, &prev_time, &busy_time);
1300
1301         busy_usec = busy_time.tv_sec;  
1302         busy_usec *= 1000000;          
1303         busy_usec += busy_time.tv_usec;
1304         etime = busy_usec;
1305         etime /= 1000000;
1306
1307         return(etime);
1308 }