1 /*        $NetBSD: kern_history.c,v 1.19 2019/10/09 05:59:51 skrll Exp $         */
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  *
27  * from: NetBSD: uvm_stat.c,v 1.36 2011/02/02 15:13:34 chuck Exp
28  * from: Id: uvm_stat.c,v 1.1.2.3 1997/12/19 15:01:00 mrg Exp
29  */
30 
31 /*
32  * subr_kernhist.c
33  */
34 
35 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: kern_history.c,v 1.19 2019/10/09 05:59:51 skrll Exp $");
37 
38 #include "opt_ddb.h"
39 #include "opt_kernhist.h"
40 #include "opt_syscall_debug.h"
41 #include "opt_usb.h"
42 #include "opt_uvmhist.h"
43 #include "opt_biohist.h"
44 #include "opt_sysctl.h"
45 
46 #include <sys/atomic.h>
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/cpu.h>
50 #include <sys/sysctl.h>
51 #include <sys/kernhist.h>
52 #include <sys/kmem.h>
53 
54 #ifdef UVMHIST
55 #include <uvm/uvm.h>
56 #endif
57 
58 #ifdef USB_DEBUG
59 #include <dev/usb/usbhist.h>
60 #endif
61 
62 #ifdef BIOHIST
63 #include <sys/biohist.h>
64 #endif
65 
66 #ifdef SYSCALL_DEBUG
67 KERNHIST_DECL(scdebughist);
68 #endif
69 
70 struct addr_xlt {
71           const char *addr;
72           size_t len;
73           uint32_t offset;
74 };
75 
76 /*
77  * globals
78  */
79 
80 struct kern_history_head kern_histories;
81 bool kernhist_sysctl_ready = 0;
82 
83 int kernhist_print_enabled = 1;
84 
85 int sysctl_hist_node;
86 
87 static int sysctl_kernhist_helper(SYSCTLFN_PROTO);
88 
89 #ifdef DDB
90 
91 /*
92  * prototypes
93  */
94 
95 void kernhist_dump(struct kern_history *, size_t count,
96     void (*)(const char *, ...) __printflike(1, 2));
97 static void kernhist_info(struct kern_history *,
98     void (*)(const char *, ...));
99 void kernhist_dumpmask(uint32_t);
100 static void kernhist_dump_histories(struct kern_history *[], size_t count,
101     void (*)(const char *, ...) __printflike(1, 2));
102 
103 /* display info about one kernhist */
104 static void
kernhist_info(struct kern_history * l,void (* pr)(const char *,...))105 kernhist_info(struct kern_history *l, void (*pr)(const char *, ...))
106 {
107 
108           pr("kernhist '%s': at %p total %u next free %u\n",
109               l->name, l, l->n, l->f);
110 }
111 
112 /*
113  * call this from ddb
114  *
115  * expects the system to be quiesced, no locking
116  */
117 void
kernhist_dump(struct kern_history * l,size_t count,void (* pr)(const char *,...))118 kernhist_dump(struct kern_history *l, size_t count,
119     void (*pr)(const char *, ...))
120 {
121           int lcv;
122 
123           lcv = l->f;
124           if (count > l->n)
125                     pr("%s: count %zu > size %u\n", __func__, count, l->n);
126           else if (count)
127                     lcv = (lcv - count) % l->n;
128 
129           do {
130                     if (l->e[lcv].fmt)
131                               kernhist_entry_print(&l->e[lcv], pr);
132                     lcv = (lcv + 1) % l->n;
133           } while (lcv != l->f);
134 }
135 
136 /*
137  * print a merged list of kern_history structures.  count is unused so far.
138  */
139 static void
kernhist_dump_histories(struct kern_history * hists[],size_t count,void (* pr)(const char *,...))140 kernhist_dump_histories(struct kern_history *hists[], size_t count,
141     void (*pr)(const char *, ...))
142 {
143           struct bintime      bt;
144           int       cur[MAXHISTS];
145           int       lcv, hi;
146 
147           /* find the first of each list */
148           for (lcv = 0; hists[lcv]; lcv++)
149                      cur[lcv] = hists[lcv]->f;
150 
151           /*
152            * here we loop "forever", finding the next earliest
153            * history entry and printing it.  cur[X] is the current
154            * entry to test for the history in hists[X].  if it is
155            * -1, then this history is finished.
156            */
157           for (;;) {
158                     hi = -1;
159                     bt.sec = 0; bt.frac = 0;
160 
161                     /* loop over each history */
162                     for (lcv = 0; hists[lcv]; lcv++) {
163 restart:
164                               if (cur[lcv] == -1)
165                                         continue;
166                               if (!hists[lcv]->e)
167                                         continue;
168 
169                               /*
170                                * if the format is empty, go to the next entry
171                                * and retry.
172                                */
173                               if (hists[lcv]->e[cur[lcv]].fmt == NULL) {
174                                         cur[lcv] = (cur[lcv] + 1) % (hists[lcv]->n);
175                                         if (cur[lcv] == hists[lcv]->f)
176                                                   cur[lcv] = -1;
177                                         goto restart;
178                               }
179 
180                               /*
181                                * if the time hasn't been set yet, or this entry is
182                                * earlier than the current bt, set the time and history
183                                * index.
184                                */
185                               if (bt.sec == 0 ||
186                                   bintimecmp(&hists[lcv]->e[cur[lcv]].bt, &bt, <)) {
187                                         bt = hists[lcv]->e[cur[lcv]].bt;
188                                         hi = lcv;
189                               }
190                     }
191 
192                     /* if we didn't find any entries, we must be done */
193                     if (hi == -1)
194                               break;
195 
196                     /* print and move to the next entry */
197                     kernhist_entry_print(&hists[hi]->e[cur[hi]], pr);
198 
199                     cur[hi] = (cur[hi] + 1) % (hists[hi]->n);
200                     if (cur[hi] == hists[hi]->f)
201                               cur[hi] = -1;
202           }
203 }
204 
205 /*
206  * call this from ddb.  `bitmask' is from <sys/kernhist.h>.  it
207  * merges the named histories.
208  *
209  * expects the system to be quiesced, no locking
210  */
211 void
kernhist_dumpmask(uint32_t bitmask)212 kernhist_dumpmask(uint32_t bitmask)     /* XXX only support 32 hists */
213 {
214           struct kern_history *hists[MAXHISTS + 1];
215           int i = 0;
216 
217 #ifdef UVMHIST
218           if ((bitmask & KERNHIST_UVMMAPHIST) || bitmask == 0)
219                     hists[i++] = &maphist;
220 
221           if ((bitmask & KERNHIST_UVMPDHIST) || bitmask == 0)
222                     hists[i++] = &pdhist;
223 
224           if ((bitmask & KERNHIST_UVMUBCHIST) || bitmask == 0)
225                     hists[i++] = &ubchist;
226 
227           if ((bitmask & KERNHIST_UVMLOANHIST) || bitmask == 0)
228                     hists[i++] = &loanhist;
229 #endif
230 
231 #ifdef USB_DEBUG
232           if ((bitmask & KERNHIST_USBHIST) || bitmask == 0)
233                     hists[i++] = &usbhist;
234 #endif
235 
236 #ifdef SYSCALL_DEBUG
237           if ((bitmask & KERNHIST_SCDEBUGHIST) || bitmask == 0)
238                     hists[i++] = &scdebughist;
239 #endif
240 
241 #ifdef BIOHIST
242           if ((bitmask & KERNHIST_BIOHIST) || bitmask == 0)
243                     hists[i++] = &biohist;
244 #endif
245 
246           hists[i] = NULL;
247 
248           kernhist_dump_histories(hists, 0, printf);
249 }
250 
251 /*
252  * kernhist_print: ddb hook to print kern history.
253  */
254 void
255 kernhist_print(void *addr, size_t count, const char *modif,
256     void (*pr)(const char *, ...) __printflike(1,2))
257 {
258           struct kern_history *h;
259 
260           LIST_FOREACH(h, &kern_histories, list) {
261                     if (h == addr)
262                               break;
263           }
264 
265           if (h == NULL) {
266                     struct kern_history *hists[MAXHISTS + 1];
267                     int i = 0;
268 #ifdef UVMHIST
269                     hists[i++] = &maphist;
270                     hists[i++] = &pdhist;
271                     hists[i++] = &ubchist;
272                     hists[i++] = &loanhist;
273 #endif
274 #ifdef USB_DEBUG
275                     hists[i++] = &usbhist;
276 #endif
277 
278 #ifdef SYSCALL_DEBUG
279                     hists[i++] = &scdebughist;
280 #endif
281 #ifdef BIOHIST
282                     hists[i++] = &biohist;
283 #endif
284                     hists[i] = NULL;
285 
286                     if (*modif == 'i') {
287                               int lcv;
288 
289                               for (lcv = 0; hists[lcv]; lcv++)
290                                         kernhist_info(hists[lcv], pr);
291                     } else {
292                               kernhist_dump_histories(hists, count, pr);
293                     }
294           } else {
295                     if (*modif == 'i')
296                               kernhist_info(h, pr);
297                     else
298                               kernhist_dump(h, count, pr);
299           }
300 }
301 
302 #endif
303 
304 /*
305  * sysctl interface
306  */
307 
308 /*
309  * sysctl_kernhist_new()
310  *
311  *        If the specified history (or, if no history is specified, any
312  *        history) does not already have a sysctl node (under kern.hist)
313  *        we create a new one and record it's node number.
314  */
315 void
sysctl_kernhist_new(struct kern_history * hist)316 sysctl_kernhist_new(struct kern_history *hist)
317 {
318           int error;
319           struct kern_history *h;
320           const struct sysctlnode *rnode = NULL;
321 
322           membar_consumer();
323           if (kernhist_sysctl_ready == 0)
324                     return;
325 
326           LIST_FOREACH(h, &kern_histories, list) {
327                     if (hist && h != hist)
328                               continue;
329                     if (h->s != 0)
330                               continue;
331                     error = sysctl_createv(NULL, 0, NULL, &rnode,
332                                   CTLFLAG_PERMANENT,
333                                   CTLTYPE_STRUCT, h->name,
334                                   SYSCTL_DESCR("history data"),
335                                   sysctl_kernhist_helper, 0, NULL, 0,
336                                   CTL_KERN, sysctl_hist_node, CTL_CREATE, CTL_EOL);
337                     if (error == 0)
338                               h->s = rnode->sysctl_num;
339                     if (hist == h)
340                               break;
341           }
342 }
343 
344 /*
345  * sysctl_kerhnist_init()
346  *
347  *        Create the 2nd level "hw.hist" sysctl node
348  */
349 void
sysctl_kernhist_init(void)350 sysctl_kernhist_init(void)
351 {
352           const struct sysctlnode *rnode = NULL;
353 
354           sysctl_createv(NULL, 0, NULL, &rnode,
355                               CTLFLAG_PERMANENT,
356                               CTLTYPE_NODE, "hist",
357                               SYSCTL_DESCR("kernel history tables"),
358                               sysctl_kernhist_helper, 0, NULL, 0,
359                               CTL_KERN, CTL_CREATE, CTL_EOL);
360           sysctl_hist_node = rnode->sysctl_num;
361 
362           kernhist_sysctl_ready = 1;
363           membar_producer();
364 
365           sysctl_kernhist_new(NULL);
366 }
367 
368 /*
369  * find_string()
370  *
371  *        Search the address-to-offset translation table for matching an
372  *        address and len, and return the index of the entry we found.  If
373  *        not found, returns index 0 which points to the "?" entry.  (We
374  *        start matching at index 1, ignoring any matches of the "?" entry
375  *        itself.)
376  */
377 static int
find_string(struct addr_xlt table[],size_t * count,const char * string,size_t len)378 find_string(struct addr_xlt table[], size_t *count, const char *string,
379               size_t len)
380 {
381           int i;
382 
383           for (i = 1; i < *count; i++)
384                     if (string == table[i].addr && len == table[i].len)
385                               return i;
386 
387           return 0;
388 }
389 
390 /*
391  * add_string()
392  *
393  *        If the string and len are unique, add a new address-to-offset
394  *        entry in the translation table and set the offset of the next
395  *        entry.
396  */
397 static void
add_string(struct addr_xlt table[],size_t * count,const char * string,size_t len)398 add_string(struct addr_xlt table[], size_t *count, const char *string,
399              size_t len)
400 {
401 
402           if (find_string(table, count, string, len) == 0) {
403                     table[*count].addr = string;
404                     table[*count].len = len;
405                     table[*count + 1].offset = table[*count].offset + len + 1;
406                     (*count)++;
407           }
408 }
409 
410 /*
411  * sysctl_kernhist_helper
412  *
413  *        This helper routine is called for all accesses to the kern.hist
414  *        hierarchy.
415  */
416 static int
sysctl_kernhist_helper(SYSCTLFN_ARGS)417 sysctl_kernhist_helper(SYSCTLFN_ARGS)
418 {
419           struct kern_history *h;
420           struct kern_history_ent *in_evt;
421           struct sysctl_history_event *out_evt;
422           struct sysctl_history *buf;
423           struct addr_xlt *xlate_t, *xlt;
424           size_t bufsize, xlate_s;
425           size_t xlate_c;
426           const char *strp __diagused;
427           char *next;
428           int i, j;
429           int error;
430 
431           if (namelen == 1 && name[0] == CTL_QUERY)
432                     return sysctl_query(SYSCTLFN_CALL(rnode));
433 
434           /*
435            * Disallow userland updates, verify that we arrived at a
436            * valid history rnode
437            */
438           if (newp)
439                     return EPERM;
440           if (namelen != 1 || name[0] != CTL_EOL)
441                     return EINVAL;
442 
443           /* Find the correct kernhist for this sysctl node */
444           LIST_FOREACH(h, &kern_histories, list) {
445                     if (h->s == rnode->sysctl_num)
446                               break;
447           }
448           if (h == NULL)
449                     return ENOENT;
450 
451           /*
452            * Worst case is two string pointers per history entry, plus
453            * two for the history name and "?" string; allocate an extra
454            * entry since we pre-set the "next" entry's offset member.
455            */
456           xlate_s = sizeof(struct addr_xlt) * h->n * 2 + 3;
457           xlate_t = kmem_alloc(xlate_s, KM_SLEEP);
458           xlate_c = 0;
459 
460           /* offset 0 reserved for NULL pointer, ie unused history entry */
461           xlate_t[0].offset = 1;
462 
463           /*
464            * If the history gets updated and an unexpected string is
465            * found later, we'll point it here.  Otherwise, we'd have to
466            * repeat this process iteratively, and it could take multiple
467            * iterations before terminating.
468            */
469           add_string(xlate_t, &xlate_c, "?", 0);
470 
471           /* Copy the history name itself to the export structure */
472           add_string(xlate_t, &xlate_c, h->name, h->namelen);
473 
474           /*
475            * Loop through all used history entries to find the unique
476            * fn and fmt strings
477            */
478           for (i = 0, in_evt = h->e; i < h->n; i++, in_evt++) {
479                     if (in_evt->fn == NULL)
480                               continue;
481                     add_string(xlate_t, &xlate_c, in_evt->fn, in_evt->fnlen);
482                     add_string(xlate_t, &xlate_c, in_evt->fmt, in_evt->fmtlen);
483           }
484 
485           /* Total buffer size includes header, events, and string table */
486           bufsize = sizeof(struct sysctl_history) +
487               h->n * sizeof(struct sysctl_history_event) +
488               xlate_t[xlate_c].offset;
489           buf = kmem_alloc(bufsize, KM_SLEEP);
490 
491           /*
492            * Copy history header info to the export structure
493            */
494           j = find_string(xlate_t, &xlate_c, h->name, h->namelen);
495           buf->sh_nameoffset = xlate_t[j].offset;
496           buf->sh_numentries = h->n;
497           buf->sh_nextfree = h->f;
498 
499           /*
500            * Loop through the history events again, copying the data to
501            * the export structure
502            */
503           for (i = 0, in_evt = h->e, out_evt = buf->sh_events; i < h->n;
504               i++, in_evt++, out_evt++) {
505                     if (in_evt->fn == NULL) {     /* skip unused entries */
506                               out_evt->she_funcoffset = 0;
507                               out_evt->she_fmtoffset = 0;
508                               continue;
509                     }
510                     out_evt->she_bintime = in_evt->bt;
511                     out_evt->she_callnumber = in_evt->call;
512                     out_evt->she_cpunum = in_evt->cpunum;
513                     out_evt->she_values[0] = in_evt->v[0];
514                     out_evt->she_values[1] = in_evt->v[1];
515                     out_evt->she_values[2] = in_evt->v[2];
516                     out_evt->she_values[3] = in_evt->v[3];
517                     j = find_string(xlate_t, &xlate_c, in_evt->fn, in_evt->fnlen);
518                     out_evt->she_funcoffset = xlate_t[j].offset;
519                     j = find_string(xlate_t, &xlate_c, in_evt->fmt, in_evt->fmtlen);
520                     out_evt->she_fmtoffset = xlate_t[j].offset;
521           }
522 
523           /*
524            * Finally, fill the text string area with all the unique
525            * strings we found earlier.
526            *
527            * Skip the initial byte, since we use an offset of 0 to mean
528            * a NULL pointer (which means an unused history event).
529            */
530           strp = next = (char *)(&buf->sh_events[h->n]);
531           *next++ = '\0';
532 
533           /*
534            * Then copy each string into the export structure, making
535            * sure to terminate each string with a '\0' character
536            */
537           for (i = 0, xlt = xlate_t; i < xlate_c; i++, xlt++) {
538                     KASSERTMSG((next - strp) == xlt->offset,
539                         "entry %d at wrong offset %"PRIu32, i, xlt->offset);
540                     memcpy(next, xlt->addr, xlt->len);
541                     next += xlt->len;
542                     *next++ = '\0';
543           }
544 
545           /* Copy data to userland */
546           error = copyout(buf, oldp, uimin(bufsize, *oldlenp));
547 
548           /* If copyout was successful but only partial, report ENOMEM */
549           if (error == 0 && *oldlenp < bufsize)
550                     error = ENOMEM;
551 
552           *oldlenp = bufsize; /* inform userland of space requirements */
553 
554           /* Free up the stuff we allocated */
555           kmem_free(buf, bufsize);
556           kmem_free(xlate_t, xlate_s);
557 
558           return error;
559 }
560