xref: /titanic_50/usr/src/cmd/mdb/common/modules/dtrace/dtrace.c (revision 902686d13f8a39cfdc57f5605470369eafa51e1f)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2013 by Delphix. All rights reserved.
25  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26  */
27 
28 /*
29  * explicitly define DTRACE_ERRDEBUG to pull in definition of dtrace_errhash_t
30  * explicitly define _STDARG_H to avoid stdarg.h/varargs.h u/k defn conflict
31  */
32 #define	DTRACE_ERRDEBUG
33 #define	_STDARG_H
34 
35 #include <mdb/mdb_param.h>
36 #include <mdb/mdb_modapi.h>
37 #include <mdb/mdb_ks.h>
38 #include <sys/dtrace_impl.h>
39 #include <sys/vmem_impl.h>
40 #include <sys/ddi_impldefs.h>
41 #include <sys/sysmacros.h>
42 #include <sys/kobj.h>
43 #include <dtrace.h>
44 #include <alloca.h>
45 #include <ctype.h>
46 #include <errno.h>
47 #include <math.h>
48 #include <stdio.h>
49 #include <unistd.h>
50 
51 /*ARGSUSED*/
52 int
id2probe(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)53 id2probe(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
54 {
55 	uintptr_t probe = NULL;
56 	uintptr_t probes;
57 
58 	if (!(flags & DCMD_ADDRSPEC))
59 		return (DCMD_USAGE);
60 
61 	if (addr == DTRACE_IDNONE || addr > UINT32_MAX)
62 		goto out;
63 
64 	if (mdb_readvar(&probes, "dtrace_probes") == -1) {
65 		mdb_warn("failed to read 'dtrace_probes'");
66 		return (DCMD_ERR);
67 	}
68 
69 	probes += (addr - 1) * sizeof (dtrace_probe_t *);
70 
71 	if (mdb_vread(&probe, sizeof (uintptr_t), probes) == -1) {
72 		mdb_warn("failed to read dtrace_probes[%d]", addr - 1);
73 		return (DCMD_ERR);
74 	}
75 
76 out:
77 	mdb_printf("%p\n", probe);
78 	return (DCMD_OK);
79 }
80 
81 void
dtrace_help(void)82 dtrace_help(void)
83 {
84 
85 	mdb_printf("Given a dtrace_state_t structure that represents a "
86 	    "DTrace consumer, prints\n"
87 	    "dtrace(1M)-like output for in-kernel DTrace data.  (The "
88 	    "dtrace_state_t\n"
89 	    "structures for all DTrace consumers may be obtained by running "
90 	    "the \n"
91 	    "::dtrace_state dcmd.)   When data is present on multiple CPUs, "
92 	    "data are\n"
93 	    "presented in CPU order, with records within each CPU ordered "
94 	    "oldest to \n"
95 	    "youngest.  Options:\n\n"
96 	    "-c cpu     Only provide output for specified CPU.\n");
97 }
98 
99 static int
dtracemdb_eprobe(dtrace_state_t * state,dtrace_eprobedesc_t * epd)100 dtracemdb_eprobe(dtrace_state_t *state, dtrace_eprobedesc_t *epd)
101 {
102 	dtrace_epid_t epid = epd->dtepd_epid;
103 	dtrace_probe_t probe;
104 	dtrace_ecb_t ecb;
105 	uintptr_t addr, paddr, ap;
106 	dtrace_action_t act;
107 	int nactions, nrecs;
108 
109 	addr = (uintptr_t)state->dts_ecbs +
110 	    (epid - 1) * sizeof (dtrace_ecb_t *);
111 
112 	if (mdb_vread(&addr, sizeof (addr), addr) == -1) {
113 		mdb_warn("failed to read ecb for epid %d", epid);
114 		return (-1);
115 	}
116 
117 	if (addr == NULL) {
118 		mdb_warn("epid %d doesn't match an ecb\n", epid);
119 		return (-1);
120 	}
121 
122 	if (mdb_vread(&ecb, sizeof (ecb), addr) == -1) {
123 		mdb_warn("failed to read ecb at %p", addr);
124 		return (-1);
125 	}
126 
127 	paddr = (uintptr_t)ecb.dte_probe;
128 
129 	if (mdb_vread(&probe, sizeof (probe), paddr) == -1) {
130 		mdb_warn("failed to read probe for ecb %p", addr);
131 		return (-1);
132 	}
133 
134 	/*
135 	 * This is a little painful:  in order to find the number of actions,
136 	 * we need to first walk through them.
137 	 */
138 	for (ap = (uintptr_t)ecb.dte_action, nactions = 0; ap != NULL; ) {
139 		if (mdb_vread(&act, sizeof (act), ap) == -1) {
140 			mdb_warn("failed to read action %p on ecb %p",
141 			    ap, addr);
142 			return (-1);
143 		}
144 
145 		if (!DTRACEACT_ISAGG(act.dta_kind) && !act.dta_intuple)
146 			nactions++;
147 
148 		ap = (uintptr_t)act.dta_next;
149 	}
150 
151 	nrecs = epd->dtepd_nrecs;
152 	epd->dtepd_nrecs = nactions;
153 	epd->dtepd_probeid = probe.dtpr_id;
154 	epd->dtepd_uarg = ecb.dte_uarg;
155 	epd->dtepd_size = ecb.dte_size;
156 
157 	for (ap = (uintptr_t)ecb.dte_action, nactions = 0; ap != NULL; ) {
158 		if (mdb_vread(&act, sizeof (act), ap) == -1) {
159 			mdb_warn("failed to read action %p on ecb %p",
160 			    ap, addr);
161 			return (-1);
162 		}
163 
164 		if (!DTRACEACT_ISAGG(act.dta_kind) && !act.dta_intuple) {
165 			if (nrecs-- == 0)
166 				break;
167 
168 			epd->dtepd_rec[nactions++] = act.dta_rec;
169 		}
170 
171 		ap = (uintptr_t)act.dta_next;
172 	}
173 
174 	return (0);
175 }
176 
177 /*ARGSUSED*/
178 static int
dtracemdb_probe(dtrace_state_t * state,dtrace_probedesc_t * pd)179 dtracemdb_probe(dtrace_state_t *state, dtrace_probedesc_t *pd)
180 {
181 	uintptr_t base, addr, paddr, praddr;
182 	int nprobes, i;
183 	dtrace_probe_t probe;
184 	dtrace_provider_t prov;
185 
186 	if (pd->dtpd_id == DTRACE_IDNONE)
187 		pd->dtpd_id++;
188 
189 	if (mdb_readvar(&base, "dtrace_probes") == -1) {
190 		mdb_warn("failed to read 'dtrace_probes'");
191 		return (-1);
192 	}
193 
194 	if (mdb_readvar(&nprobes, "dtrace_nprobes") == -1) {
195 		mdb_warn("failed to read 'dtrace_nprobes'");
196 		return (-1);
197 	}
198 
199 	for (i = pd->dtpd_id; i <= nprobes; i++) {
200 		addr = base + (i - 1) * sizeof (dtrace_probe_t *);
201 
202 		if (mdb_vread(&paddr, sizeof (paddr), addr) == -1) {
203 			mdb_warn("couldn't read probe pointer at %p", addr);
204 			return (-1);
205 		}
206 
207 		if (paddr != NULL)
208 			break;
209 	}
210 
211 	if (paddr == NULL) {
212 		errno = ESRCH;
213 		return (-1);
214 	}
215 
216 	if (mdb_vread(&probe, sizeof (probe), paddr) == -1) {
217 		mdb_warn("couldn't read probe at %p", paddr);
218 		return (-1);
219 	}
220 
221 	pd->dtpd_id = probe.dtpr_id;
222 
223 	if (mdb_vread(pd->dtpd_name, DTRACE_NAMELEN,
224 	    (uintptr_t)probe.dtpr_name) == -1) {
225 		mdb_warn("failed to read probe name for probe %p", paddr);
226 		return (-1);
227 	}
228 
229 	if (mdb_vread(pd->dtpd_func, DTRACE_FUNCNAMELEN,
230 	    (uintptr_t)probe.dtpr_func) == -1) {
231 		mdb_warn("failed to read function name for probe %p", paddr);
232 		return (-1);
233 	}
234 
235 	if (mdb_vread(pd->dtpd_mod, DTRACE_MODNAMELEN,
236 	    (uintptr_t)probe.dtpr_mod) == -1) {
237 		mdb_warn("failed to read module name for probe %p", paddr);
238 		return (-1);
239 	}
240 
241 	praddr = (uintptr_t)probe.dtpr_provider;
242 
243 	if (mdb_vread(&prov, sizeof (prov), praddr) == -1) {
244 		mdb_warn("failed to read provider for probe %p", paddr);
245 		return (-1);
246 	}
247 
248 	if (mdb_vread(pd->dtpd_provider, DTRACE_PROVNAMELEN,
249 	    (uintptr_t)prov.dtpv_name) == -1) {
250 		mdb_warn("failed to read provider name for probe %p", paddr);
251 		return (-1);
252 	}
253 
254 	return (0);
255 }
256 
257 /*ARGSUSED*/
258 static int
dtracemdb_aggdesc(dtrace_state_t * state,dtrace_aggdesc_t * agd)259 dtracemdb_aggdesc(dtrace_state_t *state, dtrace_aggdesc_t *agd)
260 {
261 	dtrace_aggid_t aggid = agd->dtagd_id;
262 	dtrace_aggregation_t agg;
263 	dtrace_ecb_t ecb;
264 	uintptr_t addr, eaddr, ap, last;
265 	dtrace_action_t act;
266 	dtrace_recdesc_t *lrec;
267 	int nactions, nrecs;
268 
269 	addr = (uintptr_t)state->dts_aggregations +
270 	    (aggid - 1) * sizeof (dtrace_aggregation_t *);
271 
272 	if (mdb_vread(&addr, sizeof (addr), addr) == -1) {
273 		mdb_warn("failed to read aggregation for aggid %d", aggid);
274 		return (-1);
275 	}
276 
277 	if (addr == NULL) {
278 		mdb_warn("aggid %d doesn't match an aggregation\n", aggid);
279 		return (-1);
280 	}
281 
282 	if (mdb_vread(&agg, sizeof (agg), addr) == -1) {
283 		mdb_warn("failed to read aggregation at %p", addr);
284 		return (-1);
285 	}
286 
287 	eaddr = (uintptr_t)agg.dtag_ecb;
288 
289 	if (mdb_vread(&ecb, sizeof (ecb), eaddr) == -1) {
290 		mdb_warn("failed to read ecb for aggregation %p", addr);
291 		return (-1);
292 	}
293 
294 	last = (uintptr_t)addr + offsetof(dtrace_aggregation_t, dtag_action);
295 
296 	/*
297 	 * This is a little painful:  in order to find the number of actions,
298 	 * we need to first walk through them.
299 	 */
300 	ap = (uintptr_t)agg.dtag_first;
301 	nactions = 0;
302 
303 	for (;;) {
304 		if (mdb_vread(&act, sizeof (act), ap) == -1) {
305 			mdb_warn("failed to read action %p on aggregation %p",
306 			    ap, addr);
307 			return (-1);
308 		}
309 
310 		nactions++;
311 
312 		if (ap == last)
313 			break;
314 
315 		ap = (uintptr_t)act.dta_next;
316 	}
317 
318 	lrec = &act.dta_rec;
319 	agd->dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - agg.dtag_base;
320 
321 	nrecs = agd->dtagd_nrecs;
322 	agd->dtagd_nrecs = nactions;
323 	agd->dtagd_epid = ecb.dte_epid;
324 
325 	ap = (uintptr_t)agg.dtag_first;
326 	nactions = 0;
327 
328 	for (;;) {
329 		dtrace_recdesc_t rec;
330 
331 		if (mdb_vread(&act, sizeof (act), ap) == -1) {
332 			mdb_warn("failed to read action %p on aggregation %p",
333 			    ap, addr);
334 			return (-1);
335 		}
336 
337 		if (nrecs-- == 0)
338 			break;
339 
340 		rec = act.dta_rec;
341 		rec.dtrd_offset -= agg.dtag_base;
342 		rec.dtrd_uarg = 0;
343 		agd->dtagd_rec[nactions++] = rec;
344 
345 		if (ap == last)
346 			break;
347 
348 		ap = (uintptr_t)act.dta_next;
349 	}
350 
351 	return (0);
352 }
353 
354 static int
dtracemdb_bufsnap(dtrace_buffer_t * which,dtrace_bufdesc_t * desc)355 dtracemdb_bufsnap(dtrace_buffer_t *which, dtrace_bufdesc_t *desc)
356 {
357 	uintptr_t addr;
358 	size_t bufsize;
359 	dtrace_buffer_t buf;
360 	caddr_t data = desc->dtbd_data;
361 	processorid_t max_cpuid, cpu = desc->dtbd_cpu;
362 
363 	if (mdb_readvar(&max_cpuid, "max_cpuid") == -1) {
364 		mdb_warn("failed to read 'max_cpuid'");
365 		errno = EIO;
366 		return (-1);
367 	}
368 
369 	if (cpu < 0 || cpu > max_cpuid) {
370 		errno = EINVAL;
371 		return (-1);
372 	}
373 
374 	addr = (uintptr_t)which + cpu * sizeof (dtrace_buffer_t);
375 
376 	if (mdb_vread(&buf, sizeof (buf), addr) == -1) {
377 		mdb_warn("failed to read buffer description at %p", addr);
378 		errno = EIO;
379 		return (-1);
380 	}
381 
382 	if (buf.dtb_tomax == NULL) {
383 		errno = ENOENT;
384 		return (-1);
385 	}
386 
387 	if (buf.dtb_flags & DTRACEBUF_WRAPPED) {
388 		bufsize = buf.dtb_size;
389 	} else {
390 		bufsize = buf.dtb_offset;
391 	}
392 
393 	if (mdb_vread(data, bufsize, (uintptr_t)buf.dtb_tomax) == -1) {
394 		mdb_warn("couldn't read buffer for CPU %d", cpu);
395 		errno = EIO;
396 		return (-1);
397 	}
398 
399 	if (buf.dtb_offset > buf.dtb_size) {
400 		mdb_warn("buffer for CPU %d has corrupt offset\n", cpu);
401 		errno = EIO;
402 		return (-1);
403 	}
404 
405 	if (buf.dtb_flags & DTRACEBUF_WRAPPED) {
406 		if (buf.dtb_xamot_offset > buf.dtb_size) {
407 			mdb_warn("ringbuffer for CPU %d has corrupt "
408 			    "wrapped offset\n", cpu);
409 			errno = EIO;
410 			return (-1);
411 		}
412 
413 		/*
414 		 * If the ring buffer has wrapped, it needs to be polished.
415 		 * See the comment in dtrace_buffer_polish() for details.
416 		 */
417 		if (buf.dtb_offset < buf.dtb_xamot_offset) {
418 			bzero(data + buf.dtb_offset,
419 			    buf.dtb_xamot_offset - buf.dtb_offset);
420 		}
421 
422 		if (buf.dtb_offset > buf.dtb_xamot_offset) {
423 			bzero(data + buf.dtb_offset,
424 			    buf.dtb_size - buf.dtb_offset);
425 			bzero(data, buf.dtb_xamot_offset);
426 		}
427 
428 		desc->dtbd_oldest = buf.dtb_xamot_offset;
429 	} else {
430 		desc->dtbd_oldest = 0;
431 	}
432 
433 	desc->dtbd_size = bufsize;
434 	desc->dtbd_drops = buf.dtb_drops;
435 	desc->dtbd_errors = buf.dtb_errors;
436 
437 	return (0);
438 }
439 
440 /*
441  * This is essentially identical to its cousin in the kernel -- with the
442  * notable exception that we automatically set DTRACEOPT_GRABANON if this
443  * state is an anonymous enabling.
444  */
445 static dof_hdr_t *
dtracemdb_dof_create(dtrace_state_t * state,int isanon)446 dtracemdb_dof_create(dtrace_state_t *state, int isanon)
447 {
448 	dof_hdr_t *dof;
449 	dof_sec_t *sec;
450 	dof_optdesc_t *opt;
451 	int i, len = sizeof (dof_hdr_t) +
452 	    roundup(sizeof (dof_sec_t), sizeof (uint64_t)) +
453 	    sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
454 
455 	dof = mdb_zalloc(len, UM_SLEEP);
456 	dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0;
457 	dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1;
458 	dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2;
459 	dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3;
460 
461 	dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE;
462 	dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE;
463 	dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION;
464 	dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION;
465 	dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS;
466 	dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS;
467 
468 	dof->dofh_flags = 0;
469 	dof->dofh_hdrsize = sizeof (dof_hdr_t);
470 	dof->dofh_secsize = sizeof (dof_sec_t);
471 	dof->dofh_secnum = 1;	/* only DOF_SECT_OPTDESC */
472 	dof->dofh_secoff = sizeof (dof_hdr_t);
473 	dof->dofh_loadsz = len;
474 	dof->dofh_filesz = len;
475 	dof->dofh_pad = 0;
476 
477 	/*
478 	 * Fill in the option section header...
479 	 */
480 	sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t));
481 	sec->dofs_type = DOF_SECT_OPTDESC;
482 	sec->dofs_align = sizeof (uint64_t);
483 	sec->dofs_flags = DOF_SECF_LOAD;
484 	sec->dofs_entsize = sizeof (dof_optdesc_t);
485 
486 	opt = (dof_optdesc_t *)((uintptr_t)sec +
487 	    roundup(sizeof (dof_sec_t), sizeof (uint64_t)));
488 
489 	sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof;
490 	sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX;
491 
492 	for (i = 0; i < DTRACEOPT_MAX; i++) {
493 		opt[i].dofo_option = i;
494 		opt[i].dofo_strtab = DOF_SECIDX_NONE;
495 		opt[i].dofo_value = state->dts_options[i];
496 	}
497 
498 	if (isanon)
499 		opt[DTRACEOPT_GRABANON].dofo_value = 1;
500 
501 	return (dof);
502 }
503 
504 static int
dtracemdb_format(dtrace_state_t * state,dtrace_fmtdesc_t * desc)505 dtracemdb_format(dtrace_state_t *state, dtrace_fmtdesc_t *desc)
506 {
507 	uintptr_t addr, faddr;
508 	char c;
509 	int len = 0;
510 
511 	if (desc->dtfd_format == 0 || desc->dtfd_format > state->dts_nformats) {
512 		errno = EINVAL;
513 		return (-1);
514 	}
515 
516 	faddr = (uintptr_t)state->dts_formats +
517 	    (desc->dtfd_format - 1) * sizeof (char *);
518 
519 	if (mdb_vread(&addr, sizeof (addr), faddr) == -1) {
520 		mdb_warn("failed to read format string pointer at %p", faddr);
521 		return (-1);
522 	}
523 
524 	do {
525 		if (mdb_vread(&c, sizeof (c), addr + len++) == -1) {
526 			mdb_warn("failed to read format string at %p", addr);
527 			return (-1);
528 		}
529 	} while (c != '\0');
530 
531 	if (len > desc->dtfd_length) {
532 		desc->dtfd_length = len;
533 		return (0);
534 	}
535 
536 	if (mdb_vread(desc->dtfd_string, len, addr) == -1) {
537 		mdb_warn("failed to reread format string at %p", addr);
538 		return (-1);
539 	}
540 
541 	return (0);
542 }
543 
544 static int
dtracemdb_status(dtrace_state_t * state,dtrace_status_t * status)545 dtracemdb_status(dtrace_state_t *state, dtrace_status_t *status)
546 {
547 	dtrace_dstate_t *dstate;
548 	int i, j;
549 	uint64_t nerrs;
550 	uintptr_t addr;
551 	int ncpu;
552 
553 	if (mdb_readvar(&ncpu, "_ncpu") == -1) {
554 		mdb_warn("failed to read '_ncpu'");
555 		return (DCMD_ERR);
556 	}
557 
558 	bzero(status, sizeof (dtrace_status_t));
559 
560 	if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) {
561 		errno = ENOENT;
562 		return (-1);
563 	}
564 
565 	/*
566 	 * For the MDB backend, we never set dtst_exiting or dtst_filled.  This
567 	 * is by design:  we don't want the library to try to stop tracing,
568 	 * because it doesn't particularly mean anything.
569 	 */
570 	nerrs = state->dts_errors;
571 	dstate = &state->dts_vstate.dtvs_dynvars;
572 
573 	for (i = 0; i < ncpu; i++) {
574 		dtrace_dstate_percpu_t dcpu;
575 		dtrace_buffer_t buf;
576 
577 		addr = (uintptr_t)&dstate->dtds_percpu[i];
578 
579 		if (mdb_vread(&dcpu, sizeof (dcpu), addr) == -1) {
580 			mdb_warn("failed to read per-CPU dstate at %p", addr);
581 			return (-1);
582 		}
583 
584 		status->dtst_dyndrops += dcpu.dtdsc_drops;
585 		status->dtst_dyndrops_dirty += dcpu.dtdsc_dirty_drops;
586 		status->dtst_dyndrops_rinsing += dcpu.dtdsc_rinsing_drops;
587 
588 		addr = (uintptr_t)&state->dts_buffer[i];
589 
590 		if (mdb_vread(&buf, sizeof (buf), addr) == -1) {
591 			mdb_warn("failed to read per-CPU buffer at %p", addr);
592 			return (-1);
593 		}
594 
595 		nerrs += buf.dtb_errors;
596 
597 		for (j = 0; j < state->dts_nspeculations; j++) {
598 			dtrace_speculation_t spec;
599 
600 			addr = (uintptr_t)&state->dts_speculations[j];
601 
602 			if (mdb_vread(&spec, sizeof (spec), addr) == -1) {
603 				mdb_warn("failed to read "
604 				    "speculation at %p", addr);
605 				return (-1);
606 			}
607 
608 			addr = (uintptr_t)&spec.dtsp_buffer[i];
609 
610 			if (mdb_vread(&buf, sizeof (buf), addr) == -1) {
611 				mdb_warn("failed to read "
612 				    "speculative buffer at %p", addr);
613 				return (-1);
614 			}
615 
616 			status->dtst_specdrops += buf.dtb_xamot_drops;
617 		}
618 	}
619 
620 	status->dtst_specdrops_busy = state->dts_speculations_busy;
621 	status->dtst_specdrops_unavail = state->dts_speculations_unavail;
622 	status->dtst_errors = nerrs;
623 
624 	return (0);
625 }
626 
627 typedef struct dtracemdb_data {
628 	dtrace_state_t *dtmd_state;
629 	char *dtmd_symstr;
630 	char *dtmd_modstr;
631 	uintptr_t dtmd_addr;
632 	int dtmd_isanon;
633 } dtracemdb_data_t;
634 
635 static int
dtracemdb_ioctl(void * varg,int cmd,void * arg)636 dtracemdb_ioctl(void *varg, int cmd, void *arg)
637 {
638 	dtracemdb_data_t *data = varg;
639 	dtrace_state_t *state = data->dtmd_state;
640 
641 	switch (cmd) {
642 	case DTRACEIOC_CONF: {
643 		dtrace_conf_t *conf = arg;
644 
645 		bzero(conf, sizeof (conf));
646 		conf->dtc_difversion = DIF_VERSION;
647 		conf->dtc_difintregs = DIF_DIR_NREGS;
648 		conf->dtc_diftupregs = DIF_DTR_NREGS;
649 		conf->dtc_ctfmodel = CTF_MODEL_NATIVE;
650 
651 		return (0);
652 	}
653 
654 	case DTRACEIOC_DOFGET: {
655 		dof_hdr_t *hdr = arg, *dof;
656 
657 		dof = dtracemdb_dof_create(state, data->dtmd_isanon);
658 		bcopy(dof, hdr, MIN(hdr->dofh_loadsz, dof->dofh_loadsz));
659 		mdb_free(dof, dof->dofh_loadsz);
660 
661 		return (0);
662 	}
663 
664 	case DTRACEIOC_BUFSNAP:
665 		return (dtracemdb_bufsnap(state->dts_buffer, arg));
666 
667 	case DTRACEIOC_AGGSNAP:
668 		return (dtracemdb_bufsnap(state->dts_aggbuffer, arg));
669 
670 	case DTRACEIOC_AGGDESC:
671 		return (dtracemdb_aggdesc(state, arg));
672 
673 	case DTRACEIOC_EPROBE:
674 		return (dtracemdb_eprobe(state, arg));
675 
676 	case DTRACEIOC_PROBES:
677 		return (dtracemdb_probe(state, arg));
678 
679 	case DTRACEIOC_FORMAT:
680 		return (dtracemdb_format(state, arg));
681 
682 	case DTRACEIOC_STATUS:
683 		return (dtracemdb_status(state, arg));
684 
685 	case DTRACEIOC_GO:
686 		*(processorid_t *)arg = -1;
687 		return (0);
688 
689 	case DTRACEIOC_ENABLE:
690 		errno = ENOTTY; /* see dt_open.c:dtrace_go() */
691 		return (-1);
692 
693 	case DTRACEIOC_PROVIDER:
694 	case DTRACEIOC_PROBEMATCH:
695 		errno = ESRCH;
696 		return (-1);
697 
698 	default:
699 		mdb_warn("unexpected ioctl 0x%x (%s)\n", cmd,
700 		    cmd == DTRACEIOC_PROVIDER	? "DTRACEIOC_PROVIDER" :
701 		    cmd == DTRACEIOC_PROBES	? "DTRACEIOC_PROBES" :
702 		    cmd == DTRACEIOC_BUFSNAP	? "DTRACEIOC_BUFSNAP" :
703 		    cmd == DTRACEIOC_PROBEMATCH	? "DTRACEIOC_PROBEMATCH" :
704 		    cmd == DTRACEIOC_ENABLE	? "DTRACEIOC_ENABLE" :
705 		    cmd == DTRACEIOC_AGGSNAP	? "DTRACEIOC_AGGSNAP" :
706 		    cmd == DTRACEIOC_EPROBE	? "DTRACEIOC_EPROBE" :
707 		    cmd == DTRACEIOC_PROBEARG	? "DTRACEIOC_PROBEARG" :
708 		    cmd == DTRACEIOC_CONF	? "DTRACEIOC_CONF" :
709 		    cmd == DTRACEIOC_STATUS	? "DTRACEIOC_STATUS" :
710 		    cmd == DTRACEIOC_GO		? "DTRACEIOC_GO" :
711 		    cmd == DTRACEIOC_STOP	? "DTRACEIOC_STOP" :
712 		    cmd == DTRACEIOC_AGGDESC	? "DTRACEIOC_AGGDESC" :
713 		    cmd == DTRACEIOC_FORMAT	? "DTRACEIOC_FORMAT" :
714 		    cmd == DTRACEIOC_DOFGET	? "DTRACEIOC_DOFGET" :
715 		    cmd == DTRACEIOC_REPLICATE	? "DTRACEIOC_REPLICATE" :
716 		    "???");
717 		errno = ENXIO;
718 		return (-1);
719 	}
720 }
721 
722 static int
dtracemdb_modctl(uintptr_t addr,const struct modctl * m,dtracemdb_data_t * data)723 dtracemdb_modctl(uintptr_t addr, const struct modctl *m, dtracemdb_data_t *data)
724 {
725 	struct module mod;
726 
727 	if (m->mod_mp == NULL)
728 		return (WALK_NEXT);
729 
730 	if (mdb_vread(&mod, sizeof (mod), (uintptr_t)m->mod_mp) == -1) {
731 		mdb_warn("couldn't read modctl %p's module", addr);
732 		return (WALK_NEXT);
733 	}
734 
735 	if ((uintptr_t)mod.text > data->dtmd_addr)
736 		return (WALK_NEXT);
737 
738 	if ((uintptr_t)mod.text + mod.text_size <= data->dtmd_addr)
739 		return (WALK_NEXT);
740 
741 	if (mdb_readstr(data->dtmd_modstr, MDB_SYM_NAMLEN,
742 	    (uintptr_t)m->mod_modname) == -1)
743 		return (WALK_ERR);
744 
745 	return (WALK_DONE);
746 }
747 
748 static int
dtracemdb_lookup_by_addr(void * varg,GElf_Addr addr,GElf_Sym * symp,dtrace_syminfo_t * sip)749 dtracemdb_lookup_by_addr(void *varg, GElf_Addr addr, GElf_Sym *symp,
750     dtrace_syminfo_t *sip)
751 {
752 	dtracemdb_data_t *data = varg;
753 
754 	if (data->dtmd_symstr == NULL) {
755 		data->dtmd_symstr = mdb_zalloc(MDB_SYM_NAMLEN,
756 		    UM_SLEEP | UM_GC);
757 	}
758 
759 	if (data->dtmd_modstr == NULL) {
760 		data->dtmd_modstr = mdb_zalloc(MDB_SYM_NAMLEN,
761 		    UM_SLEEP | UM_GC);
762 	}
763 
764 	if (symp != NULL) {
765 		if (mdb_lookup_by_addr(addr, MDB_SYM_FUZZY, data->dtmd_symstr,
766 		    MDB_SYM_NAMLEN, symp) == -1)
767 			return (-1);
768 	}
769 
770 	if (sip != NULL) {
771 		data->dtmd_addr = addr;
772 
773 		(void) strcpy(data->dtmd_modstr, "???");
774 
775 		if (mdb_walk("modctl",
776 		    (mdb_walk_cb_t)dtracemdb_modctl, varg) == -1) {
777 			mdb_warn("couldn't walk 'modctl'");
778 			return (-1);
779 		}
780 
781 		sip->dts_object = data->dtmd_modstr;
782 		sip->dts_id = 0;
783 		sip->dts_name = symp != NULL ? data->dtmd_symstr : NULL;
784 	}
785 
786 	return (0);
787 }
788 
789 /*ARGSUSED*/
790 static int
dtracemdb_stat(void * varg,processorid_t cpu)791 dtracemdb_stat(void *varg, processorid_t cpu)
792 {
793 	GElf_Sym sym;
794 	cpu_t c;
795 	uintptr_t caddr, addr;
796 
797 	if (mdb_lookup_by_name("cpu", &sym) == -1) {
798 		mdb_warn("failed to find symbol for 'cpu'");
799 		return (-1);
800 	}
801 
802 	if (cpu * sizeof (uintptr_t) > sym.st_size)
803 		return (-1);
804 
805 	addr = (uintptr_t)sym.st_value + cpu * sizeof (uintptr_t);
806 
807 	if (mdb_vread(&caddr, sizeof (caddr), addr) == -1) {
808 		mdb_warn("failed to read cpu[%d]", cpu);
809 		return (-1);
810 	}
811 
812 	if (caddr == NULL)
813 		return (-1);
814 
815 	if (mdb_vread(&c, sizeof (c), caddr) == -1) {
816 		mdb_warn("failed to read cpu at %p", caddr);
817 		return (-1);
818 	}
819 
820 	if (c.cpu_flags & CPU_POWEROFF) {
821 		return (P_POWEROFF);
822 	} else if (c.cpu_flags & CPU_SPARE) {
823 		return (P_SPARE);
824 	} else if (c.cpu_flags & CPU_FAULTED) {
825 		return (P_FAULTED);
826 	} else if ((c.cpu_flags & (CPU_READY | CPU_OFFLINE)) != CPU_READY) {
827 		return (P_OFFLINE);
828 	} else if (c.cpu_flags & CPU_ENABLE) {
829 		return (P_ONLINE);
830 	} else {
831 		return (P_NOINTR);
832 	}
833 }
834 
835 /*ARGSUSED*/
836 static long
dtracemdb_sysconf(void * varg,int name)837 dtracemdb_sysconf(void *varg, int name)
838 {
839 	int max_ncpus;
840 	processorid_t max_cpuid;
841 
842 	switch (name) {
843 	case _SC_CPUID_MAX:
844 		if (mdb_readvar(&max_cpuid, "max_cpuid") == -1) {
845 			mdb_warn("failed to read 'max_cpuid'");
846 			return (-1);
847 		}
848 
849 		return (max_cpuid);
850 
851 	case _SC_NPROCESSORS_MAX:
852 		if (mdb_readvar(&max_ncpus, "max_ncpus") == -1) {
853 			mdb_warn("failed to read 'max_ncpus'");
854 			return (-1);
855 		}
856 
857 		return (max_ncpus);
858 
859 	default:
860 		mdb_warn("unexpected sysconf code %d\n", name);
861 		return (-1);
862 	}
863 }
864 
865 const dtrace_vector_t dtrace_mdbops = {
866 	dtracemdb_ioctl,
867 	dtracemdb_lookup_by_addr,
868 	dtracemdb_stat,
869 	dtracemdb_sysconf
870 };
871 
872 typedef struct dtrace_dcmddata {
873 	dtrace_hdl_t *dtdd_dtp;
874 	int dtdd_cpu;
875 	int dtdd_quiet;
876 	int dtdd_flowindent;
877 	int dtdd_heading;
878 	FILE *dtdd_output;
879 } dtrace_dcmddata_t;
880 
881 /*
882  * Helper to grab all the content from a file, spit it into a string, and erase
883  * and reset the file.
884  */
885 static void
print_and_truncate_file(FILE * fp)886 print_and_truncate_file(FILE *fp)
887 {
888 	long len;
889 	char *out;
890 
891 	/* flush, find length of file, seek to beginning, initialize buffer */
892 	if (fflush(fp) || (len = ftell(fp)) < 0 ||
893 	    fseek(fp, 0, SEEK_SET) < 0) {
894 		mdb_warn("couldn't prepare DTrace output file: %d\n", errno);
895 		return;
896 	}
897 
898 	out = mdb_alloc(len + 1, UM_SLEEP);
899 	out[len] = '\0';
900 
901 	/* read file into buffer, truncate file, and seek to beginning */
902 	if ((fread(out, len + 1, sizeof (char), fp) == 0 && ferror(fp)) ||
903 	    ftruncate(fileno(fp), 0) < 0 || fseek(fp, 0, SEEK_SET) < 0) {
904 		mdb_warn("couldn't read DTrace output file: %d\n", errno);
905 		mdb_free(out, len + 1);
906 		return;
907 	}
908 
909 	mdb_printf("%s", out);
910 	mdb_free(out, len + 1);
911 }
912 
913 /*ARGSUSED*/
914 static int
dtrace_dcmdrec(const dtrace_probedata_t * data,const dtrace_recdesc_t * rec,void * arg)915 dtrace_dcmdrec(const dtrace_probedata_t *data,
916     const dtrace_recdesc_t *rec, void *arg)
917 {
918 	dtrace_dcmddata_t *dd = arg;
919 
920 	print_and_truncate_file(dd->dtdd_output);
921 
922 	if (rec == NULL) {
923 		/*
924 		 * We have processed the final record; output the newline if
925 		 * we're not in quiet mode.
926 		 */
927 		if (!dd->dtdd_quiet)
928 			mdb_printf("\n");
929 
930 		return (DTRACE_CONSUME_NEXT);
931 	}
932 
933 	return (DTRACE_CONSUME_THIS);
934 }
935 
936 /*ARGSUSED*/
937 static int
dtrace_dcmdprobe(const dtrace_probedata_t * data,void * arg)938 dtrace_dcmdprobe(const dtrace_probedata_t *data, void *arg)
939 {
940 	dtrace_probedesc_t *pd = data->dtpda_pdesc;
941 	processorid_t cpu = data->dtpda_cpu;
942 	dtrace_dcmddata_t *dd = arg;
943 	char name[DTRACE_FUNCNAMELEN + DTRACE_NAMELEN + 2];
944 
945 	if (dd->dtdd_cpu != -1UL && dd->dtdd_cpu != cpu)
946 		return (DTRACE_CONSUME_NEXT);
947 
948 	if (dd->dtdd_heading == 0) {
949 		if (!dd->dtdd_flowindent) {
950 			if (!dd->dtdd_quiet) {
951 				mdb_printf("%3s %6s %32s\n",
952 				    "CPU", "ID", "FUNCTION:NAME");
953 			}
954 		} else {
955 			mdb_printf("%3s %-41s\n", "CPU", "FUNCTION");
956 		}
957 		dd->dtdd_heading = 1;
958 	}
959 
960 	if (!dd->dtdd_flowindent) {
961 		if (!dd->dtdd_quiet) {
962 			(void) mdb_snprintf(name, sizeof (name), "%s:%s",
963 			    pd->dtpd_func, pd->dtpd_name);
964 
965 			mdb_printf("%3d %6d %32s ", cpu, pd->dtpd_id, name);
966 		}
967 	} else {
968 		int indent = data->dtpda_indent;
969 
970 		if (data->dtpda_flow == DTRACEFLOW_NONE) {
971 			(void) mdb_snprintf(name, sizeof (name), "%*s%s%s:%s",
972 			    indent, "", data->dtpda_prefix, pd->dtpd_func,
973 			    pd->dtpd_name);
974 		} else {
975 			(void) mdb_snprintf(name, sizeof (name), "%*s%s%s",
976 			    indent, "", data->dtpda_prefix, pd->dtpd_func);
977 		}
978 
979 		mdb_printf("%3d %-41s ", cpu, name);
980 	}
981 
982 	return (DTRACE_CONSUME_THIS);
983 }
984 
985 /*ARGSUSED*/
986 static int
dtrace_dcmderr(const dtrace_errdata_t * data,void * arg)987 dtrace_dcmderr(const dtrace_errdata_t *data, void *arg)
988 {
989 	mdb_warn(data->dteda_msg);
990 	return (DTRACE_HANDLE_OK);
991 }
992 
993 /*ARGSUSED*/
994 static int
dtrace_dcmddrop(const dtrace_dropdata_t * data,void * arg)995 dtrace_dcmddrop(const dtrace_dropdata_t *data, void *arg)
996 {
997 	mdb_warn(data->dtdda_msg);
998 	return (DTRACE_HANDLE_OK);
999 }
1000 
1001 /*ARGSUSED*/
1002 static int
dtrace_dcmdbuffered(const dtrace_bufdata_t * bufdata,void * arg)1003 dtrace_dcmdbuffered(const dtrace_bufdata_t *bufdata, void *arg)
1004 {
1005 	mdb_printf("%s", bufdata->dtbda_buffered);
1006 	return (DTRACE_HANDLE_OK);
1007 }
1008 
1009 /*ARGSUSED*/
1010 int
dtrace(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)1011 dtrace(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1012 {
1013 	dtrace_state_t state;
1014 	dtrace_hdl_t *dtp;
1015 	int ncpu, err;
1016 	uintptr_t c = -1UL;
1017 	dtrace_dcmddata_t dd;
1018 	dtrace_optval_t val;
1019 	dtracemdb_data_t md;
1020 	int rval = DCMD_ERR;
1021 	dtrace_anon_t anon;
1022 
1023 	if (!(flags & DCMD_ADDRSPEC))
1024 		return (DCMD_USAGE);
1025 
1026 	if (mdb_getopts(argc, argv, 'c', MDB_OPT_UINTPTR, &c, NULL) != argc)
1027 		return (DCMD_USAGE);
1028 
1029 	if (mdb_readvar(&ncpu, "_ncpu") == -1) {
1030 		mdb_warn("failed to read '_ncpu'");
1031 		return (DCMD_ERR);
1032 	}
1033 
1034 	if (mdb_vread(&state, sizeof (state), addr) == -1) {
1035 		mdb_warn("couldn't read dtrace_state_t at %p", addr);
1036 		return (DCMD_ERR);
1037 	}
1038 
1039 	if (state.dts_anon != NULL) {
1040 		addr = (uintptr_t)state.dts_anon;
1041 
1042 		if (mdb_vread(&state, sizeof (state), addr) == -1) {
1043 			mdb_warn("couldn't read anonymous state at %p", addr);
1044 			return (DCMD_ERR);
1045 		}
1046 	}
1047 
1048 	bzero(&md, sizeof (md));
1049 	md.dtmd_state = &state;
1050 
1051 	if ((dtp = dtrace_vopen(DTRACE_VERSION, DTRACE_O_NOSYS, &err,
1052 	    &dtrace_mdbops, &md)) == NULL) {
1053 		mdb_warn("failed to initialize dtrace: %s\n",
1054 		    dtrace_errmsg(NULL, err));
1055 		return (DCMD_ERR);
1056 	}
1057 
1058 	/*
1059 	 * If this is the anonymous enabling, we need to set a bit indicating
1060 	 * that DTRACEOPT_GRABANON should be set.
1061 	 */
1062 	if (mdb_readvar(&anon, "dtrace_anon") == -1) {
1063 		mdb_warn("failed to read 'dtrace_anon'");
1064 		return (DCMD_ERR);
1065 	}
1066 
1067 	md.dtmd_isanon = ((uintptr_t)anon.dta_state == addr);
1068 
1069 	if (dtrace_go(dtp) != 0) {
1070 		mdb_warn("failed to initialize dtrace: %s\n",
1071 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1072 		goto err;
1073 	}
1074 
1075 	bzero(&dd, sizeof (dd));
1076 	dd.dtdd_dtp = dtp;
1077 	dd.dtdd_cpu = c;
1078 
1079 	if (dtrace_getopt(dtp, "flowindent", &val) == -1) {
1080 		mdb_warn("couldn't get 'flowindent' option: %s\n",
1081 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1082 		goto err;
1083 	}
1084 
1085 	dd.dtdd_flowindent = (val != DTRACEOPT_UNSET);
1086 
1087 	if (dtrace_getopt(dtp, "quiet", &val) == -1) {
1088 		mdb_warn("couldn't get 'quiet' option: %s\n",
1089 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1090 		goto err;
1091 	}
1092 
1093 	dd.dtdd_quiet = (val != DTRACEOPT_UNSET);
1094 
1095 	if (dtrace_handle_err(dtp, dtrace_dcmderr, NULL) == -1) {
1096 		mdb_warn("couldn't add err handler: %s\n",
1097 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1098 		goto err;
1099 	}
1100 
1101 	if (dtrace_handle_drop(dtp, dtrace_dcmddrop, NULL) == -1) {
1102 		mdb_warn("couldn't add drop handler: %s\n",
1103 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1104 		goto err;
1105 	}
1106 
1107 	if (dtrace_handle_buffered(dtp, dtrace_dcmdbuffered, NULL) == -1) {
1108 		mdb_warn("couldn't add buffered handler: %s\n",
1109 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1110 		goto err;
1111 	}
1112 
1113 	if (dtrace_status(dtp) == -1) {
1114 		mdb_warn("couldn't get status: %s\n",
1115 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1116 		goto err;
1117 	}
1118 
1119 	if (dtrace_aggregate_snap(dtp) == -1) {
1120 		mdb_warn("couldn't snapshot aggregation: %s\n",
1121 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1122 		goto err;
1123 	}
1124 
1125 	if ((dd.dtdd_output = tmpfile()) == NULL) {
1126 		mdb_warn("couldn't open DTrace output file: %d\n", errno);
1127 		goto err;
1128 	}
1129 
1130 	if (dtrace_consume(dtp, dd.dtdd_output,
1131 	    dtrace_dcmdprobe, dtrace_dcmdrec, &dd) == -1) {
1132 		mdb_warn("couldn't consume DTrace buffers: %s\n",
1133 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1134 	}
1135 
1136 	if (dtrace_aggregate_print(dtp, NULL, NULL) == -1) {
1137 		mdb_warn("couldn't print aggregation: %s\n",
1138 		    dtrace_errmsg(dtp, dtrace_errno(dtp)));
1139 		goto err;
1140 	}
1141 
1142 	rval = DCMD_OK;
1143 err:
1144 	dtrace_close(dtp);
1145 	fclose(dd.dtdd_output);
1146 	return (rval);
1147 }
1148 
1149 static int
dtrace_errhash_cmp(const void * l,const void * r)1150 dtrace_errhash_cmp(const void *l, const void *r)
1151 {
1152 	uintptr_t lhs = *((uintptr_t *)l);
1153 	uintptr_t rhs = *((uintptr_t *)r);
1154 	dtrace_errhash_t lerr, rerr;
1155 	char lmsg[256], rmsg[256];
1156 
1157 	(void) mdb_vread(&lerr, sizeof (lerr), lhs);
1158 	(void) mdb_vread(&rerr, sizeof (rerr), rhs);
1159 
1160 	if (lerr.dter_msg == NULL)
1161 		return (-1);
1162 
1163 	if (rerr.dter_msg == NULL)
1164 		return (1);
1165 
1166 	(void) mdb_readstr(lmsg, sizeof (lmsg), (uintptr_t)lerr.dter_msg);
1167 	(void) mdb_readstr(rmsg, sizeof (rmsg), (uintptr_t)rerr.dter_msg);
1168 
1169 	return (strcmp(lmsg, rmsg));
1170 }
1171 
1172 int
dtrace_errhash_init(mdb_walk_state_t * wsp)1173 dtrace_errhash_init(mdb_walk_state_t *wsp)
1174 {
1175 	GElf_Sym sym;
1176 	uintptr_t *hash, addr;
1177 	int i;
1178 
1179 	if (wsp->walk_addr != NULL) {
1180 		mdb_warn("dtrace_errhash walk only supports global walks\n");
1181 		return (WALK_ERR);
1182 	}
1183 
1184 	if (mdb_lookup_by_name("dtrace_errhash", &sym) == -1) {
1185 		mdb_warn("couldn't find 'dtrace_errhash' (non-DEBUG kernel?)");
1186 		return (WALK_ERR);
1187 	}
1188 
1189 	addr = (uintptr_t)sym.st_value;
1190 	hash = mdb_alloc(DTRACE_ERRHASHSZ * sizeof (uintptr_t),
1191 	    UM_SLEEP | UM_GC);
1192 
1193 	for (i = 0; i < DTRACE_ERRHASHSZ; i++)
1194 		hash[i] = addr + i * sizeof (dtrace_errhash_t);
1195 
1196 	qsort(hash, DTRACE_ERRHASHSZ, sizeof (uintptr_t), dtrace_errhash_cmp);
1197 
1198 	wsp->walk_addr = 0;
1199 	wsp->walk_data = hash;
1200 
1201 	return (WALK_NEXT);
1202 }
1203 
1204 int
dtrace_errhash_step(mdb_walk_state_t * wsp)1205 dtrace_errhash_step(mdb_walk_state_t *wsp)
1206 {
1207 	int ndx = (int)wsp->walk_addr;
1208 	uintptr_t *hash = wsp->walk_data;
1209 	dtrace_errhash_t err;
1210 	uintptr_t addr;
1211 
1212 	if (ndx >= DTRACE_ERRHASHSZ)
1213 		return (WALK_DONE);
1214 
1215 	wsp->walk_addr = ndx + 1;
1216 	addr = hash[ndx];
1217 
1218 	if (mdb_vread(&err, sizeof (err), addr) == -1) {
1219 		mdb_warn("failed to read dtrace_errhash_t at %p", addr);
1220 		return (WALK_DONE);
1221 	}
1222 
1223 	if (err.dter_msg == NULL)
1224 		return (WALK_NEXT);
1225 
1226 	return (wsp->walk_callback(addr, &err, wsp->walk_cbdata));
1227 }
1228 
1229 /*ARGSUSED*/
1230 int
dtrace_errhash(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)1231 dtrace_errhash(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1232 {
1233 	dtrace_errhash_t err;
1234 	char msg[256];
1235 
1236 	if (!(flags & DCMD_ADDRSPEC)) {
1237 		if (mdb_walk_dcmd("dtrace_errhash", "dtrace_errhash",
1238 		    argc, argv) == -1) {
1239 			mdb_warn("can't walk 'dtrace_errhash'");
1240 			return (DCMD_ERR);
1241 		}
1242 
1243 		return (DCMD_OK);
1244 	}
1245 
1246 	if (DCMD_HDRSPEC(flags))
1247 		mdb_printf("%8s %s\n", "COUNT", "ERROR");
1248 
1249 	if (mdb_vread(&err, sizeof (err), addr) == -1) {
1250 		mdb_warn("failed to read dtrace_errhash_t at %p", addr);
1251 		return (DCMD_ERR);
1252 	}
1253 
1254 	addr = (uintptr_t)err.dter_msg;
1255 
1256 	if (mdb_readstr(msg, sizeof (msg), addr) == -1) {
1257 		mdb_warn("failed to read error msg at %p", addr);
1258 		return (DCMD_ERR);
1259 	}
1260 
1261 	mdb_printf("%8d %s", err.dter_count, msg);
1262 
1263 	/*
1264 	 * Some error messages include a newline -- only print the newline
1265 	 * if the message doesn't have one.
1266 	 */
1267 	if (msg[strlen(msg) - 1] != '\n')
1268 		mdb_printf("\n");
1269 
1270 	return (DCMD_OK);
1271 }
1272 
1273 int
dtrace_helptrace_init(mdb_walk_state_t * wsp)1274 dtrace_helptrace_init(mdb_walk_state_t *wsp)
1275 {
1276 	uint32_t next;
1277 	uintptr_t buffer;
1278 
1279 	if (wsp->walk_addr != NULL) {
1280 		mdb_warn("dtrace_helptrace only supports global walks\n");
1281 		return (WALK_ERR);
1282 	}
1283 
1284 	if (mdb_readvar(&buffer, "dtrace_helptrace_buffer") == -1) {
1285 		mdb_warn("couldn't read 'dtrace_helptrace_buffer'");
1286 		return (WALK_ERR);
1287 	}
1288 
1289 	if (buffer == NULL) {
1290 		mdb_warn("helper tracing is not enabled\n");
1291 		return (WALK_ERR);
1292 	}
1293 
1294 	if (mdb_readvar(&next, "dtrace_helptrace_next") == -1) {
1295 		mdb_warn("couldn't read 'dtrace_helptrace_next'");
1296 		return (WALK_ERR);
1297 	}
1298 
1299 	wsp->walk_addr = next;
1300 
1301 	return (WALK_NEXT);
1302 }
1303 
1304 int
dtrace_helptrace_step(mdb_walk_state_t * wsp)1305 dtrace_helptrace_step(mdb_walk_state_t *wsp)
1306 {
1307 	uint32_t next, size, nlocals, bufsize;
1308 	uintptr_t buffer, addr;
1309 	dtrace_helptrace_t *ht;
1310 	int rval;
1311 
1312 	if (mdb_readvar(&next, "dtrace_helptrace_next") == -1) {
1313 		mdb_warn("couldn't read 'dtrace_helptrace_next'");
1314 		return (WALK_ERR);
1315 	}
1316 
1317 	if (mdb_readvar(&bufsize, "dtrace_helptrace_bufsize") == -1) {
1318 		mdb_warn("couldn't read 'dtrace_helptrace_bufsize'");
1319 		return (WALK_ERR);
1320 	}
1321 
1322 	if (mdb_readvar(&buffer, "dtrace_helptrace_buffer") == -1) {
1323 		mdb_warn("couldn't read 'dtrace_helptrace_buffer'");
1324 		return (WALK_ERR);
1325 	}
1326 
1327 	if (mdb_readvar(&nlocals, "dtrace_helptrace_nlocals") == -1) {
1328 		mdb_warn("couldn't read 'dtrace_helptrace_nlocals'");
1329 		return (WALK_ERR);
1330 	}
1331 
1332 	size = sizeof (dtrace_helptrace_t) +
1333 	    nlocals * sizeof (uint64_t) - sizeof (uint64_t);
1334 
1335 	if (wsp->walk_addr + size > bufsize) {
1336 		if (next == 0)
1337 			return (WALK_DONE);
1338 
1339 		wsp->walk_addr = 0;
1340 	}
1341 
1342 	addr = buffer + wsp->walk_addr;
1343 	ht = alloca(size);
1344 
1345 	if (mdb_vread(ht, size, addr) == -1) {
1346 		mdb_warn("couldn't read entry at %p", addr);
1347 		return (WALK_ERR);
1348 	}
1349 
1350 	if (ht->dtht_helper != NULL) {
1351 		rval = wsp->walk_callback(addr, ht, wsp->walk_cbdata);
1352 
1353 		if (rval != WALK_NEXT)
1354 			return (rval);
1355 	}
1356 
1357 	if (wsp->walk_addr < next && wsp->walk_addr + size >= next)
1358 		return (WALK_DONE);
1359 
1360 	wsp->walk_addr += size;
1361 	return (WALK_NEXT);
1362 }
1363 
1364 int
dtrace_helptrace(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)1365 dtrace_helptrace(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1366 {
1367 	dtrace_helptrace_t help;
1368 	dtrace_helper_action_t helper;
1369 	char where[30];
1370 	uint_t opt_v = FALSE;
1371 	uintptr_t haddr;
1372 
1373 	if (!(flags & DCMD_ADDRSPEC)) {
1374 		if (mdb_walk_dcmd("dtrace_helptrace", "dtrace_helptrace",
1375 		    argc, argv) == -1) {
1376 			mdb_warn("can't walk 'dtrace_helptrace'");
1377 			return (DCMD_ERR);
1378 		}
1379 
1380 		return (DCMD_OK);
1381 	}
1382 
1383 	if (mdb_getopts(argc, argv, 'v',
1384 	    MDB_OPT_SETBITS, TRUE, &opt_v, NULL) != argc)
1385 		return (DCMD_USAGE);
1386 
1387 	if (DCMD_HDRSPEC(flags)) {
1388 		mdb_printf(" %?s %?s %12s %s\n",
1389 		    "ADDR", "HELPER", "WHERE", "DIFO");
1390 	}
1391 
1392 	if (mdb_vread(&help, sizeof (help), addr) == -1) {
1393 		mdb_warn("failed to read dtrace_helptrace_t at %p", addr);
1394 		return (DCMD_ERR);
1395 	}
1396 
1397 	switch (help.dtht_where) {
1398 	case 0:
1399 		(void) mdb_snprintf(where, sizeof (where), "predicate");
1400 		break;
1401 
1402 	case DTRACE_HELPTRACE_NEXT:
1403 		(void) mdb_snprintf(where, sizeof (where), "next");
1404 		break;
1405 
1406 	case DTRACE_HELPTRACE_DONE:
1407 		(void) mdb_snprintf(where, sizeof (where), "done");
1408 		break;
1409 
1410 	case DTRACE_HELPTRACE_ERR:
1411 		(void) mdb_snprintf(where, sizeof (where), "err");
1412 		break;
1413 
1414 	default:
1415 		(void) mdb_snprintf(where, sizeof (where),
1416 		    "action #%d", help.dtht_where);
1417 		break;
1418 	}
1419 
1420 	mdb_printf(" %?p %?p %12s ", addr, help.dtht_helper, where);
1421 
1422 	haddr = (uintptr_t)help.dtht_helper;
1423 
1424 	if (mdb_vread(&helper, sizeof (helper), haddr) == -1) {
1425 		/*
1426 		 * We're not going to warn in this case -- we're just not going
1427 		 * to print anything exciting.
1428 		 */
1429 		mdb_printf("???\n");
1430 	} else {
1431 		switch (help.dtht_where) {
1432 		case 0:
1433 			mdb_printf("%p\n", helper.dtha_predicate);
1434 			break;
1435 
1436 		case DTRACE_HELPTRACE_NEXT:
1437 		case DTRACE_HELPTRACE_DONE:
1438 		case DTRACE_HELPTRACE_ERR:
1439 			mdb_printf("-\n");
1440 			break;
1441 
1442 		default:
1443 			haddr = (uintptr_t)helper.dtha_actions +
1444 			    (help.dtht_where - 1) * sizeof (uintptr_t);
1445 
1446 			if (mdb_vread(&haddr, sizeof (haddr), haddr) == -1) {
1447 				mdb_printf("???\n");
1448 			} else {
1449 				mdb_printf("%p\n", haddr);
1450 			}
1451 		}
1452 	}
1453 
1454 	if (opt_v) {
1455 		int i;
1456 
1457 		if (help.dtht_where == DTRACE_HELPTRACE_ERR) {
1458 			int f = help.dtht_fault;
1459 
1460 			mdb_printf("%?s| %?s %10s |\n", "", "", "");
1461 			mdb_printf("%?s| %?s %10s +->  fault: %s\n", "", "", "",
1462 			    f == DTRACEFLT_BADADDR ? "BADADDR" :
1463 			    f == DTRACEFLT_BADALIGN ? "BADALIGN" :
1464 			    f == DTRACEFLT_ILLOP ? "ILLOP" :
1465 			    f == DTRACEFLT_DIVZERO ? "DIVZERO" :
1466 			    f == DTRACEFLT_NOSCRATCH ? "NOSCRATCH" :
1467 			    f == DTRACEFLT_KPRIV ? "KPRIV" :
1468 			    f == DTRACEFLT_UPRIV ? "UPRIV" :
1469 			    f == DTRACEFLT_TUPOFLOW ? "TUPOFLOW" :
1470 			    f == DTRACEFLT_BADSTACK ? "BADSTACK" :
1471 			    "DTRACEFLT_UNKNOWN");
1472 			mdb_printf("%?s| %?s %12s     addr: 0x%x\n", "", "", "",
1473 			    help.dtht_illval);
1474 			mdb_printf("%?s| %?s %12s   offset: %d\n", "", "", "",
1475 			    help.dtht_fltoffs);
1476 		}
1477 
1478 		mdb_printf("%?s|\n%?s+--> %?s %4s %s\n", "", "",
1479 		    "ADDR", "NDX", "VALUE");
1480 		addr += sizeof (help) - sizeof (uint64_t);
1481 
1482 		for (i = 0; i < help.dtht_nlocals; i++) {
1483 			uint64_t val;
1484 
1485 			if (mdb_vread(&val, sizeof (val), addr) == -1) {
1486 				mdb_warn("couldn't read local at %p", addr);
1487 				continue;
1488 			}
1489 
1490 			mdb_printf("%?s     %?p %4d %p\n", "", addr, i, val);
1491 			addr += sizeof (uint64_t);
1492 		}
1493 
1494 		mdb_printf("\n");
1495 	}
1496 
1497 	return (DCMD_OK);
1498 }
1499 
1500 /*ARGSUSED*/
1501 static int
dtrace_state_walk(uintptr_t addr,const vmem_seg_t * seg,minor_t * highest)1502 dtrace_state_walk(uintptr_t addr, const vmem_seg_t *seg, minor_t *highest)
1503 {
1504 	if (seg->vs_end > *highest)
1505 		*highest = seg->vs_end;
1506 
1507 	return (WALK_NEXT);
1508 }
1509 
1510 typedef struct dtrace_state_walk {
1511 	uintptr_t dtsw_softstate;
1512 	minor_t dtsw_max;
1513 	minor_t dtsw_current;
1514 } dtrace_state_walk_t;
1515 
1516 int
dtrace_state_init(mdb_walk_state_t * wsp)1517 dtrace_state_init(mdb_walk_state_t *wsp)
1518 {
1519 	uintptr_t dtrace_minor;
1520 	minor_t max = 0;
1521 	dtrace_state_walk_t *dw;
1522 
1523 	if (wsp->walk_addr != NULL) {
1524 		mdb_warn("dtrace_state only supports global walks\n");
1525 		return (WALK_ERR);
1526 	}
1527 
1528 	/*
1529 	 * Find the dtrace_minor vmem arena and walk it to get the maximum
1530 	 * minor number.
1531 	 */
1532 	if (mdb_readvar(&dtrace_minor, "dtrace_minor") == -1) {
1533 		mdb_warn("failed to read 'dtrace_minor'");
1534 		return (WALK_ERR);
1535 	}
1536 
1537 	if (mdb_pwalk("vmem_alloc", (mdb_walk_cb_t)dtrace_state_walk,
1538 	    &max, dtrace_minor) == -1) {
1539 		mdb_warn("couldn't walk 'vmem_alloc'");
1540 		return (WALK_ERR);
1541 	}
1542 
1543 	dw = mdb_zalloc(sizeof (dtrace_state_walk_t), UM_SLEEP | UM_GC);
1544 	dw->dtsw_current = 0;
1545 	dw->dtsw_max = max;
1546 
1547 	if (mdb_readvar(&dw->dtsw_softstate, "dtrace_softstate") == -1) {
1548 		mdb_warn("failed to read 'dtrace_softstate'");
1549 		return (DCMD_ERR);
1550 	}
1551 
1552 	wsp->walk_data = dw;
1553 
1554 	return (WALK_NEXT);
1555 }
1556 
1557 int
dtrace_state_step(mdb_walk_state_t * wsp)1558 dtrace_state_step(mdb_walk_state_t *wsp)
1559 {
1560 	dtrace_state_walk_t *dw = wsp->walk_data;
1561 	uintptr_t statep;
1562 	dtrace_state_t state;
1563 	int rval;
1564 
1565 	while (mdb_get_soft_state_byaddr(dw->dtsw_softstate, dw->dtsw_current,
1566 	    &statep, NULL, 0) == -1) {
1567 		if (dw->dtsw_current >= dw->dtsw_max)
1568 			return (WALK_DONE);
1569 
1570 		dw->dtsw_current++;
1571 	}
1572 
1573 	if (mdb_vread(&state, sizeof (state), statep) == -1) {
1574 		mdb_warn("couldn't read dtrace_state_t at %p", statep);
1575 		return (WALK_NEXT);
1576 	}
1577 
1578 	rval = wsp->walk_callback(statep, &state, wsp->walk_cbdata);
1579 	dw->dtsw_current++;
1580 
1581 	return (rval);
1582 }
1583 
1584 typedef struct dtrace_state_data {
1585 	int dtsd_major;
1586 	uintptr_t dtsd_proc;
1587 	uintptr_t dtsd_softstate;
1588 	uintptr_t dtsd_state;
1589 } dtrace_state_data_t;
1590 
1591 static int
dtrace_state_file(uintptr_t addr,struct file * f,dtrace_state_data_t * data)1592 dtrace_state_file(uintptr_t addr, struct file *f, dtrace_state_data_t *data)
1593 {
1594 	vnode_t vnode;
1595 	proc_t proc;
1596 	minor_t minor;
1597 	uintptr_t statep;
1598 
1599 	if (mdb_vread(&vnode, sizeof (vnode), (uintptr_t)f->f_vnode) == -1) {
1600 		mdb_warn("couldn't read vnode at %p", (uintptr_t)f->f_vnode);
1601 		return (WALK_NEXT);
1602 	}
1603 
1604 	if (getmajor(vnode.v_rdev) != data->dtsd_major)
1605 		return (WALK_NEXT);
1606 
1607 	minor = getminor(vnode.v_rdev);
1608 
1609 	if (mdb_vread(&proc, sizeof (proc), data->dtsd_proc) == -1) {
1610 		mdb_warn("failed to read proc at %p", data->dtsd_proc);
1611 		return (WALK_NEXT);
1612 	}
1613 
1614 	if (mdb_get_soft_state_byaddr(data->dtsd_softstate, minor,
1615 	    &statep, NULL, 0) == -1) {
1616 		mdb_warn("failed to read softstate for minor %d", minor);
1617 		return (WALK_NEXT);
1618 	}
1619 
1620 	if (statep != data->dtsd_state)
1621 		return (WALK_NEXT);
1622 
1623 	mdb_printf("%?p %5d %?p %-*s %?p\n", statep, minor,
1624 	    data->dtsd_proc, MAXCOMLEN, proc.p_user.u_comm, addr);
1625 
1626 	return (WALK_NEXT);
1627 }
1628 
1629 /*ARGSUSED*/
1630 static int
dtrace_state_proc(uintptr_t addr,void * ignored,dtrace_state_data_t * data)1631 dtrace_state_proc(uintptr_t addr, void *ignored, dtrace_state_data_t *data)
1632 {
1633 	data->dtsd_proc = addr;
1634 
1635 	if (mdb_pwalk("file",
1636 	    (mdb_walk_cb_t)dtrace_state_file, data, addr) == -1) {
1637 		mdb_warn("couldn't walk 'file' for proc %p", addr);
1638 		return (WALK_ERR);
1639 	}
1640 
1641 	return (WALK_NEXT);
1642 }
1643 
1644 void
dtrace_state_help(void)1645 dtrace_state_help(void)
1646 {
1647 	mdb_printf("Given a dtrace_state_t structure, displays all "
1648 	    /*CSTYLED*/
1649 	    "consumers, or \"<anonymous>\"\nif the consumer is anonymous.  If "
1650 	    "no state structure is provided, iterates\nover all state "
1651 	    "structures.\n\n"
1652 	    "Addresses in ADDR column may be provided to ::dtrace to obtain\n"
1653 	    "dtrace(1M)-like output for in-kernel DTrace data.\n");
1654 }
1655 
1656 int
dtrace_state(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)1657 dtrace_state(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1658 {
1659 	uintptr_t devi;
1660 	struct dev_info info;
1661 	dtrace_state_data_t data;
1662 	dtrace_anon_t anon;
1663 	dtrace_state_t state;
1664 
1665 	if (!(flags & DCMD_ADDRSPEC)) {
1666 		if (mdb_walk_dcmd("dtrace_state",
1667 		    "dtrace_state", argc, argv) == -1) {
1668 			mdb_warn("can't walk dtrace_state");
1669 			return (DCMD_ERR);
1670 		}
1671 		return (DCMD_OK);
1672 	}
1673 
1674 	if (DCMD_HDRSPEC(flags)) {
1675 		mdb_printf("%?s %5s %?s %-*s %?s\n", "ADDR", "MINOR", "PROC",
1676 		    MAXCOMLEN, "NAME", "FILE");
1677 	}
1678 
1679 	/*
1680 	 * First determine if this is anonymous state.
1681 	 */
1682 	if (mdb_readvar(&anon, "dtrace_anon") == -1) {
1683 		mdb_warn("failed to read 'dtrace_anon'");
1684 		return (DCMD_ERR);
1685 	}
1686 
1687 	if ((uintptr_t)anon.dta_state == addr) {
1688 		if (mdb_vread(&state, sizeof (state), addr) == -1) {
1689 			mdb_warn("failed to read anon at %p", addr);
1690 			return (DCMD_ERR);
1691 		}
1692 
1693 		mdb_printf("%?p %5d %?s %-*s %?s\n", addr,
1694 		    getminor(state.dts_dev), "-", MAXCOMLEN,
1695 		    "<anonymous>", "-");
1696 
1697 		return (DCMD_OK);
1698 	}
1699 
1700 	if (mdb_readvar(&devi, "dtrace_devi") == -1) {
1701 		mdb_warn("failed to read 'dtrace_devi'");
1702 		return (DCMD_ERR);
1703 	}
1704 
1705 	if (mdb_vread(&info, sizeof (struct dev_info), devi) == -1) {
1706 		mdb_warn("failed to read 'dev_info'");
1707 		return (DCMD_ERR);
1708 	}
1709 
1710 	data.dtsd_major = info.devi_major;
1711 
1712 	if (mdb_readvar(&data.dtsd_softstate, "dtrace_softstate") == -1) {
1713 		mdb_warn("failed to read 'dtrace_softstate'");
1714 		return (DCMD_ERR);
1715 	}
1716 
1717 	data.dtsd_state = addr;
1718 
1719 	/*
1720 	 * Walk through all processes and all open files looking for this
1721 	 * state.  It must be open somewhere...
1722 	 */
1723 	if (mdb_walk("proc", (mdb_walk_cb_t)dtrace_state_proc, &data) == -1) {
1724 		mdb_warn("couldn't walk 'proc'");
1725 		return (DCMD_ERR);
1726 	}
1727 
1728 	return (DCMD_OK);
1729 }
1730 
1731 typedef struct dtrace_aggkey_data {
1732 	uintptr_t *dtakd_hash;
1733 	uintptr_t dtakd_hashsize;
1734 	uintptr_t dtakd_next;
1735 	uintptr_t dtakd_ndx;
1736 } dtrace_aggkey_data_t;
1737 
1738 int
dtrace_aggkey_init(mdb_walk_state_t * wsp)1739 dtrace_aggkey_init(mdb_walk_state_t *wsp)
1740 {
1741 	dtrace_buffer_t buf;
1742 	uintptr_t addr;
1743 	dtrace_aggbuffer_t agb;
1744 	dtrace_aggkey_data_t *data;
1745 	size_t hsize;
1746 
1747 	if ((addr = wsp->walk_addr) == NULL) {
1748 		mdb_warn("dtrace_aggkey walk needs aggregation buffer\n");
1749 		return (WALK_ERR);
1750 	}
1751 
1752 	if (mdb_vread(&buf, sizeof (buf), addr) == -1) {
1753 		mdb_warn("failed to read aggregation buffer at %p", addr);
1754 		return (WALK_ERR);
1755 	}
1756 
1757 	addr = (uintptr_t)buf.dtb_tomax +
1758 	    buf.dtb_size - sizeof (dtrace_aggbuffer_t);
1759 
1760 	if (mdb_vread(&agb, sizeof (agb), addr) == -1) {
1761 		mdb_warn("failed to read dtrace_aggbuffer_t at %p", addr);
1762 		return (WALK_ERR);
1763 	}
1764 
1765 	data = mdb_zalloc(sizeof (dtrace_aggkey_data_t), UM_SLEEP);
1766 
1767 	data->dtakd_hashsize = agb.dtagb_hashsize;
1768 	hsize = agb.dtagb_hashsize * sizeof (dtrace_aggkey_t *);
1769 	data->dtakd_hash = mdb_alloc(hsize, UM_SLEEP);
1770 
1771 	if (mdb_vread(data->dtakd_hash, hsize,
1772 	    (uintptr_t)agb.dtagb_hash) == -1) {
1773 		mdb_warn("failed to read hash at %p",
1774 		    (uintptr_t)agb.dtagb_hash);
1775 		mdb_free(data->dtakd_hash, hsize);
1776 		mdb_free(data, sizeof (dtrace_aggkey_data_t));
1777 		return (WALK_ERR);
1778 	}
1779 
1780 	wsp->walk_data = data;
1781 	return (WALK_NEXT);
1782 }
1783 
1784 int
dtrace_aggkey_step(mdb_walk_state_t * wsp)1785 dtrace_aggkey_step(mdb_walk_state_t *wsp)
1786 {
1787 	dtrace_aggkey_data_t *data = wsp->walk_data;
1788 	dtrace_aggkey_t key;
1789 	uintptr_t addr;
1790 
1791 	while ((addr = data->dtakd_next) == NULL) {
1792 		if (data->dtakd_ndx == data->dtakd_hashsize)
1793 			return (WALK_DONE);
1794 
1795 		data->dtakd_next = data->dtakd_hash[data->dtakd_ndx++];
1796 	}
1797 
1798 	if (mdb_vread(&key, sizeof (key), addr) == -1) {
1799 		mdb_warn("failed to read dtrace_aggkey_t at %p", addr);
1800 		return (WALK_ERR);
1801 	}
1802 
1803 	data->dtakd_next = (uintptr_t)key.dtak_next;
1804 
1805 	return (wsp->walk_callback(addr, &key, wsp->walk_cbdata));
1806 }
1807 
1808 void
dtrace_aggkey_fini(mdb_walk_state_t * wsp)1809 dtrace_aggkey_fini(mdb_walk_state_t *wsp)
1810 {
1811 	dtrace_aggkey_data_t *data = wsp->walk_data;
1812 	size_t hsize;
1813 
1814 	hsize = data->dtakd_hashsize * sizeof (dtrace_aggkey_t *);
1815 	mdb_free(data->dtakd_hash, hsize);
1816 	mdb_free(data, sizeof (dtrace_aggkey_data_t));
1817 }
1818 
1819 typedef struct dtrace_dynvar_data {
1820 	dtrace_dynhash_t *dtdvd_hash;
1821 	uintptr_t dtdvd_hashsize;
1822 	uintptr_t dtdvd_next;
1823 	uintptr_t dtdvd_ndx;
1824 	uintptr_t dtdvd_sink;
1825 } dtrace_dynvar_data_t;
1826 
1827 int
dtrace_dynvar_init(mdb_walk_state_t * wsp)1828 dtrace_dynvar_init(mdb_walk_state_t *wsp)
1829 {
1830 	uintptr_t addr;
1831 	dtrace_dstate_t dstate;
1832 	dtrace_dynvar_data_t *data;
1833 	size_t hsize;
1834 	GElf_Sym sym;
1835 
1836 	if ((addr = wsp->walk_addr) == NULL) {
1837 		mdb_warn("dtrace_dynvar walk needs dtrace_dstate_t\n");
1838 		return (WALK_ERR);
1839 	}
1840 
1841 	if (mdb_vread(&dstate, sizeof (dstate), addr) == -1) {
1842 		mdb_warn("failed to read dynamic state at %p", addr);
1843 		return (WALK_ERR);
1844 	}
1845 
1846 	if (mdb_lookup_by_name("dtrace_dynhash_sink", &sym) == -1) {
1847 		mdb_warn("couldn't find 'dtrace_dynhash_sink'");
1848 		return (WALK_ERR);
1849 	}
1850 
1851 	data = mdb_zalloc(sizeof (dtrace_dynvar_data_t), UM_SLEEP);
1852 
1853 	data->dtdvd_hashsize = dstate.dtds_hashsize;
1854 	hsize = dstate.dtds_hashsize * sizeof (dtrace_dynhash_t);
1855 	data->dtdvd_hash = mdb_alloc(hsize, UM_SLEEP);
1856 	data->dtdvd_sink = (uintptr_t)sym.st_value;
1857 
1858 	if (mdb_vread(data->dtdvd_hash, hsize,
1859 	    (uintptr_t)dstate.dtds_hash) == -1) {
1860 		mdb_warn("failed to read hash at %p",
1861 		    (uintptr_t)dstate.dtds_hash);
1862 		mdb_free(data->dtdvd_hash, hsize);
1863 		mdb_free(data, sizeof (dtrace_dynvar_data_t));
1864 		return (WALK_ERR);
1865 	}
1866 
1867 	data->dtdvd_next = (uintptr_t)data->dtdvd_hash[0].dtdh_chain;
1868 
1869 	wsp->walk_data = data;
1870 	return (WALK_NEXT);
1871 }
1872 
1873 int
dtrace_dynvar_step(mdb_walk_state_t * wsp)1874 dtrace_dynvar_step(mdb_walk_state_t *wsp)
1875 {
1876 	dtrace_dynvar_data_t *data = wsp->walk_data;
1877 	dtrace_dynvar_t dynvar, *dvar;
1878 	size_t dvarsize;
1879 	uintptr_t addr;
1880 	int nkeys;
1881 
1882 	while ((addr = data->dtdvd_next) == data->dtdvd_sink) {
1883 		if (data->dtdvd_ndx == data->dtdvd_hashsize)
1884 			return (WALK_DONE);
1885 
1886 		data->dtdvd_next =
1887 		    (uintptr_t)data->dtdvd_hash[data->dtdvd_ndx++].dtdh_chain;
1888 	}
1889 
1890 	if (mdb_vread(&dynvar, sizeof (dynvar), addr) == -1) {
1891 		mdb_warn("failed to read dtrace_dynvar_t at %p", addr);
1892 		return (WALK_ERR);
1893 	}
1894 
1895 	/*
1896 	 * Now we need to allocate the correct size.
1897 	 */
1898 	nkeys = dynvar.dtdv_tuple.dtt_nkeys;
1899 	dvarsize = (uintptr_t)&dynvar.dtdv_tuple.dtt_key[nkeys] -
1900 	    (uintptr_t)&dynvar;
1901 
1902 	dvar = alloca(dvarsize);
1903 
1904 	if (mdb_vread(dvar, dvarsize, addr) == -1) {
1905 		mdb_warn("failed to read dtrace_dynvar_t at %p", addr);
1906 		return (WALK_ERR);
1907 	}
1908 
1909 	data->dtdvd_next = (uintptr_t)dynvar.dtdv_next;
1910 
1911 	return (wsp->walk_callback(addr, dvar, wsp->walk_cbdata));
1912 }
1913 
1914 void
dtrace_dynvar_fini(mdb_walk_state_t * wsp)1915 dtrace_dynvar_fini(mdb_walk_state_t *wsp)
1916 {
1917 	dtrace_dynvar_data_t *data = wsp->walk_data;
1918 	size_t hsize;
1919 
1920 	hsize = data->dtdvd_hashsize * sizeof (dtrace_dynvar_t *);
1921 	mdb_free(data->dtdvd_hash, hsize);
1922 	mdb_free(data, sizeof (dtrace_dynvar_data_t));
1923 }
1924 
1925 typedef struct dtrace_hashstat_data {
1926 	size_t *dthsd_counts;
1927 	size_t dthsd_hashsize;
1928 	char *dthsd_data;
1929 	size_t dthsd_size;
1930 	int dthsd_header;
1931 } dtrace_hashstat_data_t;
1932 
1933 typedef void (*dtrace_hashstat_func_t)(dtrace_hashstat_data_t *);
1934 
1935 static void
dtrace_hashstat_additive(dtrace_hashstat_data_t * data)1936 dtrace_hashstat_additive(dtrace_hashstat_data_t *data)
1937 {
1938 	int i;
1939 	int hval = 0;
1940 
1941 	for (i = 0; i < data->dthsd_size; i++)
1942 		hval += data->dthsd_data[i];
1943 
1944 	data->dthsd_counts[hval % data->dthsd_hashsize]++;
1945 }
1946 
1947 static void
dtrace_hashstat_shifty(dtrace_hashstat_data_t * data)1948 dtrace_hashstat_shifty(dtrace_hashstat_data_t *data)
1949 {
1950 	uint64_t hval = 0;
1951 	int i;
1952 
1953 	if (data->dthsd_size < sizeof (uint64_t)) {
1954 		dtrace_hashstat_additive(data);
1955 		return;
1956 	}
1957 
1958 	for (i = 0; i < data->dthsd_size; i += sizeof (uint64_t)) {
1959 		/* LINTED - alignment */
1960 		uint64_t val = *((uint64_t *)&data->dthsd_data[i]);
1961 
1962 		hval += (val & ((1 << NBBY) - 1)) +
1963 		    ((val >> NBBY) & ((1 << NBBY) - 1)) +
1964 		    ((val >> (NBBY << 1)) & ((1 << NBBY) - 1)) +
1965 		    ((val >> (NBBY << 2)) & ((1 << NBBY) - 1)) +
1966 		    (val & USHRT_MAX) + (val >> (NBBY << 1) & USHRT_MAX);
1967 	}
1968 
1969 	data->dthsd_counts[hval % data->dthsd_hashsize]++;
1970 }
1971 
1972 static void
dtrace_hashstat_knuth(dtrace_hashstat_data_t * data)1973 dtrace_hashstat_knuth(dtrace_hashstat_data_t *data)
1974 {
1975 	int i;
1976 	int hval = data->dthsd_size;
1977 
1978 	for (i = 0; i < data->dthsd_size; i++)
1979 		hval = (hval << 4) ^ (hval >> 28) ^ data->dthsd_data[i];
1980 
1981 	data->dthsd_counts[hval % data->dthsd_hashsize]++;
1982 }
1983 
1984 static void
dtrace_hashstat_oneatatime(dtrace_hashstat_data_t * data)1985 dtrace_hashstat_oneatatime(dtrace_hashstat_data_t *data)
1986 {
1987 	int i;
1988 	uint32_t hval = 0;
1989 
1990 	for (i = 0; i < data->dthsd_size; i++) {
1991 		hval += data->dthsd_data[i];
1992 		hval += (hval << 10);
1993 		hval ^= (hval >> 6);
1994 	}
1995 
1996 	hval += (hval << 3);
1997 	hval ^= (hval >> 11);
1998 	hval += (hval << 15);
1999 
2000 	data->dthsd_counts[hval % data->dthsd_hashsize]++;
2001 }
2002 
2003 static void
dtrace_hashstat_fnv(dtrace_hashstat_data_t * data)2004 dtrace_hashstat_fnv(dtrace_hashstat_data_t *data)
2005 {
2006 	static const uint32_t prime = 0x01000193;
2007 	uint32_t hval = 0;
2008 	int i;
2009 
2010 	for (i = 0; i < data->dthsd_size; i++) {
2011 		hval *= prime;
2012 		hval ^= data->dthsd_data[i];
2013 	}
2014 
2015 	data->dthsd_counts[hval % data->dthsd_hashsize]++;
2016 }
2017 
2018 static void
dtrace_hashstat_stats(char * name,dtrace_hashstat_data_t * data)2019 dtrace_hashstat_stats(char *name, dtrace_hashstat_data_t *data)
2020 {
2021 	size_t nz = 0, i;
2022 	int longest = 0;
2023 	size_t ttl = 0;
2024 	double sum = 0.0;
2025 	double avg;
2026 	uint_t util, stddev;
2027 
2028 	if (!data->dthsd_header) {
2029 		mdb_printf("%15s %11s %11s %11s %11s %11s\n", "NAME",
2030 		    "HASHSIZE", "%UTIL", "LONGEST", "AVERAGE", "STDDEV");
2031 		data->dthsd_header = 1;
2032 	}
2033 
2034 	for (i = 0; i < data->dthsd_hashsize; i++) {
2035 		if (data->dthsd_counts[i] != 0) {
2036 			nz++;
2037 
2038 			if (data->dthsd_counts[i] > longest)
2039 				longest = data->dthsd_counts[i];
2040 
2041 			ttl += data->dthsd_counts[i];
2042 		}
2043 	}
2044 
2045 	if (nz == 0) {
2046 		mdb_printf("%15s %11d %11s %11s %11s %11s\n", name,
2047 		    data->dthsd_hashsize, "-", "-", "-", "-");
2048 		return;
2049 	}
2050 
2051 	avg = (double)ttl / (double)nz;
2052 
2053 	for (i = 0; i < data->dthsd_hashsize; i++) {
2054 		double delta = (double)data->dthsd_counts[i] - avg;
2055 
2056 		if (data->dthsd_counts[i] == 0)
2057 			continue;
2058 
2059 		sum += delta * delta;
2060 	}
2061 
2062 	util = (nz * 1000) / data->dthsd_hashsize;
2063 	stddev = (uint_t)sqrt(sum / (double)nz) * 10;
2064 
2065 	mdb_printf("%15s %11d %9u.%1u %11d %11d %9u.%1u\n", name,
2066 	    data->dthsd_hashsize, util / 10, util % 10, longest, ttl / nz,
2067 	    stddev / 10, stddev % 10);
2068 }
2069 
2070 static struct dtrace_hashstat {
2071 	char *dths_name;
2072 	dtrace_hashstat_func_t dths_func;
2073 } _dtrace_hashstat[] = {
2074 	{ "<actual>", NULL },
2075 	{ "additive", dtrace_hashstat_additive },
2076 	{ "shifty", dtrace_hashstat_shifty },
2077 	{ "knuth", dtrace_hashstat_knuth },
2078 	{ "one-at-a-time", dtrace_hashstat_oneatatime },
2079 	{ "fnv", dtrace_hashstat_fnv },
2080 	{ NULL, 0 }
2081 };
2082 
2083 typedef struct dtrace_aggstat_data {
2084 	dtrace_hashstat_data_t dtagsd_hash;
2085 	dtrace_hashstat_func_t dtagsd_func;
2086 } dtrace_aggstat_data_t;
2087 
2088 static int
dtrace_aggstat_walk(uintptr_t addr,dtrace_aggkey_t * key,dtrace_aggstat_data_t * data)2089 dtrace_aggstat_walk(uintptr_t addr, dtrace_aggkey_t *key,
2090     dtrace_aggstat_data_t *data)
2091 {
2092 	dtrace_hashstat_data_t *hdata = &data->dtagsd_hash;
2093 	size_t size;
2094 
2095 	if (data->dtagsd_func == NULL) {
2096 		size_t bucket = key->dtak_hashval % hdata->dthsd_hashsize;
2097 
2098 		hdata->dthsd_counts[bucket]++;
2099 		return (WALK_NEXT);
2100 	}
2101 
2102 	/*
2103 	 * We need to read the data.
2104 	 */
2105 	size = key->dtak_size - sizeof (dtrace_aggid_t);
2106 	addr = (uintptr_t)key->dtak_data + sizeof (dtrace_aggid_t);
2107 	hdata->dthsd_data = alloca(size);
2108 	hdata->dthsd_size = size;
2109 
2110 	if (mdb_vread(hdata->dthsd_data, size, addr) == -1) {
2111 		mdb_warn("couldn't read data at %p", addr);
2112 		return (WALK_ERR);
2113 	}
2114 
2115 	data->dtagsd_func(hdata);
2116 
2117 	return (WALK_NEXT);
2118 }
2119 
2120 /*ARGSUSED*/
2121 int
dtrace_aggstat(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)2122 dtrace_aggstat(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2123 {
2124 	dtrace_buffer_t buf;
2125 	uintptr_t aaddr;
2126 	dtrace_aggbuffer_t agb;
2127 	size_t hsize, i, actual, prime, evenpow;
2128 	dtrace_aggstat_data_t data;
2129 	dtrace_hashstat_data_t *hdata = &data.dtagsd_hash;
2130 
2131 	bzero(&data, sizeof (data));
2132 
2133 	if (!(flags & DCMD_ADDRSPEC))
2134 		return (DCMD_USAGE);
2135 
2136 	if (mdb_vread(&buf, sizeof (buf), addr) == -1) {
2137 		mdb_warn("failed to read aggregation buffer at %p", addr);
2138 		return (DCMD_ERR);
2139 	}
2140 
2141 	aaddr = (uintptr_t)buf.dtb_tomax +
2142 	    buf.dtb_size - sizeof (dtrace_aggbuffer_t);
2143 
2144 	if (mdb_vread(&agb, sizeof (agb), aaddr) == -1) {
2145 		mdb_warn("failed to read dtrace_aggbuffer_t at %p", aaddr);
2146 		return (DCMD_ERR);
2147 	}
2148 
2149 	hsize = (actual = agb.dtagb_hashsize) * sizeof (size_t);
2150 	hdata->dthsd_counts = mdb_alloc(hsize, UM_SLEEP | UM_GC);
2151 
2152 	/*
2153 	 * Now pick the largest prime smaller than the hash size.  (If the
2154 	 * existing size is prime, we'll pick a smaller prime just for the
2155 	 * hell of it.)
2156 	 */
2157 	for (prime = agb.dtagb_hashsize - 1; prime > 7; prime--) {
2158 		size_t limit = prime / 7;
2159 
2160 		for (i = 2; i < limit; i++) {
2161 			if ((prime % i) == 0)
2162 				break;
2163 		}
2164 
2165 		if (i == limit)
2166 			break;
2167 	}
2168 
2169 	/*
2170 	 * And now we want to pick the largest power of two smaller than the
2171 	 * hashsize.
2172 	 */
2173 	for (i = 0; (1 << i) < agb.dtagb_hashsize; i++)
2174 		continue;
2175 
2176 	evenpow = (1 << (i - 1));
2177 
2178 	for (i = 0; _dtrace_hashstat[i].dths_name != NULL; i++) {
2179 		data.dtagsd_func = _dtrace_hashstat[i].dths_func;
2180 
2181 		hdata->dthsd_hashsize = actual;
2182 		hsize = hdata->dthsd_hashsize * sizeof (size_t);
2183 		bzero(hdata->dthsd_counts, hsize);
2184 
2185 		if (mdb_pwalk("dtrace_aggkey",
2186 		    (mdb_walk_cb_t)dtrace_aggstat_walk, &data, addr) == -1) {
2187 			mdb_warn("failed to walk dtrace_aggkey at %p", addr);
2188 			return (DCMD_ERR);
2189 		}
2190 
2191 		dtrace_hashstat_stats(_dtrace_hashstat[i].dths_name, hdata);
2192 
2193 		/*
2194 		 * If we were just printing the actual value, we won't try
2195 		 * any of the sizing experiments.
2196 		 */
2197 		if (data.dtagsd_func == NULL)
2198 			continue;
2199 
2200 		hdata->dthsd_hashsize = prime;
2201 		hsize = hdata->dthsd_hashsize * sizeof (size_t);
2202 		bzero(hdata->dthsd_counts, hsize);
2203 
2204 		if (mdb_pwalk("dtrace_aggkey",
2205 		    (mdb_walk_cb_t)dtrace_aggstat_walk, &data, addr) == -1) {
2206 			mdb_warn("failed to walk dtrace_aggkey at %p", addr);
2207 			return (DCMD_ERR);
2208 		}
2209 
2210 		dtrace_hashstat_stats(_dtrace_hashstat[i].dths_name, hdata);
2211 
2212 		hdata->dthsd_hashsize = evenpow;
2213 		hsize = hdata->dthsd_hashsize * sizeof (size_t);
2214 		bzero(hdata->dthsd_counts, hsize);
2215 
2216 		if (mdb_pwalk("dtrace_aggkey",
2217 		    (mdb_walk_cb_t)dtrace_aggstat_walk, &data, addr) == -1) {
2218 			mdb_warn("failed to walk dtrace_aggkey at %p", addr);
2219 			return (DCMD_ERR);
2220 		}
2221 
2222 		dtrace_hashstat_stats(_dtrace_hashstat[i].dths_name, hdata);
2223 	}
2224 
2225 	return (DCMD_OK);
2226 }
2227 
2228 /*ARGSUSED*/
2229 static int
dtrace_dynstat_walk(uintptr_t addr,dtrace_dynvar_t * dynvar,dtrace_aggstat_data_t * data)2230 dtrace_dynstat_walk(uintptr_t addr, dtrace_dynvar_t *dynvar,
2231     dtrace_aggstat_data_t *data)
2232 {
2233 	dtrace_hashstat_data_t *hdata = &data->dtagsd_hash;
2234 	dtrace_tuple_t *tuple = &dynvar->dtdv_tuple;
2235 	dtrace_key_t *key = tuple->dtt_key;
2236 	size_t size = 0, offs = 0;
2237 	int i, nkeys = tuple->dtt_nkeys;
2238 	char *buf;
2239 
2240 	if (data->dtagsd_func == NULL) {
2241 		size_t bucket = dynvar->dtdv_hashval % hdata->dthsd_hashsize;
2242 
2243 		hdata->dthsd_counts[bucket]++;
2244 		return (WALK_NEXT);
2245 	}
2246 
2247 	/*
2248 	 * We want to hand the hashing algorithm a contiguous buffer.  First
2249 	 * run through the tuple and determine the size.
2250 	 */
2251 	for (i = 0; i < nkeys; i++) {
2252 		if (key[i].dttk_size == 0) {
2253 			size += sizeof (uint64_t);
2254 		} else {
2255 			size += key[i].dttk_size;
2256 		}
2257 	}
2258 
2259 	buf = alloca(size);
2260 
2261 	/*
2262 	 * Now go back through the tuple and copy the data into the buffer.
2263 	 */
2264 	for (i = 0; i < nkeys; i++) {
2265 		if (key[i].dttk_size == 0) {
2266 			bcopy(&key[i].dttk_value, &buf[offs],
2267 			    sizeof (uint64_t));
2268 			offs += sizeof (uint64_t);
2269 		} else {
2270 			if (mdb_vread(&buf[offs], key[i].dttk_size,
2271 			    key[i].dttk_value) == -1) {
2272 				mdb_warn("couldn't read tuple data at %p",
2273 				    key[i].dttk_value);
2274 				return (WALK_ERR);
2275 			}
2276 
2277 			offs += key[i].dttk_size;
2278 		}
2279 	}
2280 
2281 	hdata->dthsd_data = buf;
2282 	hdata->dthsd_size = size;
2283 
2284 	data->dtagsd_func(hdata);
2285 
2286 	return (WALK_NEXT);
2287 }
2288 
2289 /*ARGSUSED*/
2290 int
dtrace_dynstat(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)2291 dtrace_dynstat(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2292 {
2293 	dtrace_dstate_t dstate;
2294 	size_t hsize, i, actual, prime;
2295 	dtrace_aggstat_data_t data;
2296 	dtrace_hashstat_data_t *hdata = &data.dtagsd_hash;
2297 
2298 	bzero(&data, sizeof (data));
2299 
2300 	if (!(flags & DCMD_ADDRSPEC))
2301 		return (DCMD_USAGE);
2302 
2303 	if (mdb_vread(&dstate, sizeof (dstate), addr) == -1) {
2304 		mdb_warn("failed to read dynamic variable state at %p", addr);
2305 		return (DCMD_ERR);
2306 	}
2307 
2308 	hsize = (actual = dstate.dtds_hashsize) * sizeof (size_t);
2309 	hdata->dthsd_counts = mdb_alloc(hsize, UM_SLEEP | UM_GC);
2310 
2311 	/*
2312 	 * Now pick the largest prime smaller than the hash size.  (If the
2313 	 * existing size is prime, we'll pick a smaller prime just for the
2314 	 * hell of it.)
2315 	 */
2316 	for (prime = dstate.dtds_hashsize - 1; prime > 7; prime--) {
2317 		size_t limit = prime / 7;
2318 
2319 		for (i = 2; i < limit; i++) {
2320 			if ((prime % i) == 0)
2321 				break;
2322 		}
2323 
2324 		if (i == limit)
2325 			break;
2326 	}
2327 
2328 	for (i = 0; _dtrace_hashstat[i].dths_name != NULL; i++) {
2329 		data.dtagsd_func = _dtrace_hashstat[i].dths_func;
2330 
2331 		hdata->dthsd_hashsize = actual;
2332 		hsize = hdata->dthsd_hashsize * sizeof (size_t);
2333 		bzero(hdata->dthsd_counts, hsize);
2334 
2335 		if (mdb_pwalk("dtrace_dynvar",
2336 		    (mdb_walk_cb_t)dtrace_dynstat_walk, &data, addr) == -1) {
2337 			mdb_warn("failed to walk dtrace_dynvar at %p", addr);
2338 			return (DCMD_ERR);
2339 		}
2340 
2341 		dtrace_hashstat_stats(_dtrace_hashstat[i].dths_name, hdata);
2342 
2343 		/*
2344 		 * If we were just printing the actual value, we won't try
2345 		 * any of the sizing experiments.
2346 		 */
2347 		if (data.dtagsd_func == NULL)
2348 			continue;
2349 
2350 		hdata->dthsd_hashsize = prime;
2351 		hsize = hdata->dthsd_hashsize * sizeof (size_t);
2352 		bzero(hdata->dthsd_counts, hsize);
2353 
2354 		if (mdb_pwalk("dtrace_dynvar",
2355 		    (mdb_walk_cb_t)dtrace_dynstat_walk, &data, addr) == -1) {
2356 			mdb_warn("failed to walk dtrace_aggkey at %p", addr);
2357 			return (DCMD_ERR);
2358 		}
2359 
2360 		dtrace_hashstat_stats(_dtrace_hashstat[i].dths_name, hdata);
2361 	}
2362 
2363 	return (DCMD_OK);
2364 }
2365 
2366 typedef struct dtrace_ecb_walk {
2367 	dtrace_ecb_t **dtew_ecbs;
2368 	int dtew_necbs;
2369 	int dtew_curecb;
2370 } dtrace_ecb_walk_t;
2371 
2372 static int
dtrace_ecb_init(mdb_walk_state_t * wsp)2373 dtrace_ecb_init(mdb_walk_state_t *wsp)
2374 {
2375 	uintptr_t addr;
2376 	dtrace_state_t state;
2377 	dtrace_ecb_walk_t *ecbwp;
2378 
2379 	if ((addr = wsp->walk_addr) == NULL) {
2380 		mdb_warn("dtrace_ecb walk needs dtrace_state_t\n");
2381 		return (WALK_ERR);
2382 	}
2383 
2384 	if (mdb_vread(&state, sizeof (state), addr) == -1) {
2385 		mdb_warn("failed to read dtrace state pointer at %p", addr);
2386 		return (WALK_ERR);
2387 	}
2388 
2389 	ecbwp = mdb_zalloc(sizeof (dtrace_ecb_walk_t), UM_SLEEP | UM_GC);
2390 
2391 	ecbwp->dtew_ecbs = state.dts_ecbs;
2392 	ecbwp->dtew_necbs = state.dts_necbs;
2393 	ecbwp->dtew_curecb = 0;
2394 
2395 	wsp->walk_data = ecbwp;
2396 
2397 	return (WALK_NEXT);
2398 }
2399 
2400 static int
dtrace_ecb_step(mdb_walk_state_t * wsp)2401 dtrace_ecb_step(mdb_walk_state_t *wsp)
2402 {
2403 	uintptr_t ecbp, addr;
2404 	dtrace_ecb_walk_t *ecbwp = wsp->walk_data;
2405 
2406 	addr = (uintptr_t)ecbwp->dtew_ecbs +
2407 	    ecbwp->dtew_curecb * sizeof (dtrace_ecb_t *);
2408 
2409 	if (ecbwp->dtew_curecb++ == ecbwp->dtew_necbs)
2410 		return (WALK_DONE);
2411 
2412 	if (mdb_vread(&ecbp, sizeof (addr), addr) == -1) {
2413 		mdb_warn("failed to read ecb at entry %d\n",
2414 		    ecbwp->dtew_curecb);
2415 		return (WALK_ERR);
2416 	}
2417 
2418 	if (ecbp == NULL)
2419 		return (WALK_NEXT);
2420 
2421 	return (wsp->walk_callback(ecbp, NULL, wsp->walk_cbdata));
2422 }
2423 
2424 static void
dtrace_options_numtostr(uint64_t num,char * buf,size_t len)2425 dtrace_options_numtostr(uint64_t num, char *buf, size_t len)
2426 {
2427 	uint64_t n = num;
2428 	int index = 0;
2429 	char u;
2430 
2431 	while (n >= 1024) {
2432 		n = (n + (1024 / 2)) / 1024; /* Round up or down */
2433 		index++;
2434 	}
2435 
2436 	u = " KMGTPE"[index];
2437 
2438 	if (index == 0) {
2439 		(void) mdb_snprintf(buf, len, "%llu", (u_longlong_t)n);
2440 	} else if (n < 10 && (num & (num - 1)) != 0) {
2441 		(void) mdb_snprintf(buf, len, "%.2f%c",
2442 		    (double)num / (1ULL << 10 * index), u);
2443 	} else if (n < 100 && (num & (num - 1)) != 0) {
2444 		(void) mdb_snprintf(buf, len, "%.1f%c",
2445 		    (double)num / (1ULL << 10 * index), u);
2446 	} else {
2447 		(void) mdb_snprintf(buf, len, "%llu%c", (u_longlong_t)n, u);
2448 	}
2449 }
2450 
2451 static void
dtrace_options_numtohz(uint64_t num,char * buf,size_t len)2452 dtrace_options_numtohz(uint64_t num, char *buf, size_t len)
2453 {
2454 	(void) mdb_snprintf(buf, len, "%dhz", NANOSEC/num);
2455 }
2456 
2457 static void
dtrace_options_numtobufpolicy(uint64_t num,char * buf,size_t len)2458 dtrace_options_numtobufpolicy(uint64_t num, char *buf, size_t len)
2459 {
2460 	char *policy = "unknown";
2461 
2462 	switch (num) {
2463 		case DTRACEOPT_BUFPOLICY_RING:
2464 			policy = "ring";
2465 			break;
2466 
2467 		case DTRACEOPT_BUFPOLICY_FILL:
2468 			policy = "fill";
2469 			break;
2470 
2471 		case DTRACEOPT_BUFPOLICY_SWITCH:
2472 			policy = "switch";
2473 			break;
2474 	}
2475 
2476 	(void) mdb_snprintf(buf, len, "%s", policy);
2477 }
2478 
2479 static void
dtrace_options_numtocpu(uint64_t cpu,char * buf,size_t len)2480 dtrace_options_numtocpu(uint64_t cpu, char *buf, size_t len)
2481 {
2482 	if (cpu == DTRACE_CPUALL)
2483 		(void) mdb_snprintf(buf, len, "%7s", "unbound");
2484 	else
2485 		(void) mdb_snprintf(buf, len, "%d", cpu);
2486 }
2487 
2488 typedef void (*dtrace_options_func_t)(uint64_t, char *, size_t);
2489 
2490 static struct dtrace_options {
2491 	char *dtop_optstr;
2492 	dtrace_options_func_t dtop_func;
2493 } _dtrace_options[] = {
2494 	{ "bufsize", dtrace_options_numtostr },
2495 	{ "bufpolicy", dtrace_options_numtobufpolicy },
2496 	{ "dynvarsize", dtrace_options_numtostr },
2497 	{ "aggsize", dtrace_options_numtostr },
2498 	{ "specsize", dtrace_options_numtostr },
2499 	{ "nspec", dtrace_options_numtostr },
2500 	{ "strsize", dtrace_options_numtostr },
2501 	{ "cleanrate", dtrace_options_numtohz },
2502 	{ "cpu", dtrace_options_numtocpu },
2503 	{ "bufresize", dtrace_options_numtostr },
2504 	{ "grabanon", dtrace_options_numtostr },
2505 	{ "flowindent", dtrace_options_numtostr },
2506 	{ "quiet", dtrace_options_numtostr },
2507 	{ "stackframes", dtrace_options_numtostr },
2508 	{ "ustackframes", dtrace_options_numtostr },
2509 	{ "aggrate", dtrace_options_numtohz },
2510 	{ "switchrate", dtrace_options_numtohz },
2511 	{ "statusrate", dtrace_options_numtohz },
2512 	{ "destructive", dtrace_options_numtostr },
2513 	{ "stackindent", dtrace_options_numtostr },
2514 	{ "rawbytes", dtrace_options_numtostr },
2515 	{ "jstackframes", dtrace_options_numtostr },
2516 	{ "jstackstrsize", dtrace_options_numtostr },
2517 	{ "aggsortkey", dtrace_options_numtostr },
2518 	{ "aggsortrev", dtrace_options_numtostr },
2519 	{ "aggsortpos", dtrace_options_numtostr },
2520 	{ "aggsortkeypos", dtrace_options_numtostr }
2521 };
2522 
2523 static void
dtrace_options_help(void)2524 dtrace_options_help(void)
2525 {
2526 	mdb_printf("Given a dtrace_state_t structure, displays the "
2527 	    "current tunable option\nsettings.\n");
2528 }
2529 
2530 /*ARGSUSED*/
2531 static int
dtrace_options(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)2532 dtrace_options(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2533 {
2534 	dtrace_state_t state;
2535 	int i = 0;
2536 	dtrace_optval_t *options;
2537 	char val[32];
2538 
2539 	if (!(flags & DCMD_ADDRSPEC))
2540 		return (DCMD_USAGE);
2541 
2542 	if (mdb_vread(&state, sizeof (dtrace_state_t), (uintptr_t)addr) == -1) {
2543 		mdb_warn("failed to read state pointer at %p\n", addr);
2544 		return (DCMD_ERR);
2545 	}
2546 
2547 	options = &state.dts_options[0];
2548 
2549 	mdb_printf("%<u>%-25s %s%</u>\n", "OPTION", "VALUE");
2550 	for (i = 0; i < DTRACEOPT_MAX; i++) {
2551 		if (options[i] == DTRACEOPT_UNSET) {
2552 			mdb_printf("%-25s %s\n",
2553 			    _dtrace_options[i].dtop_optstr, "UNSET");
2554 		} else {
2555 			(void) _dtrace_options[i].dtop_func(options[i],
2556 			    val, 32);
2557 			mdb_printf("%-25s %s\n",
2558 			    _dtrace_options[i].dtop_optstr, val);
2559 		}
2560 	}
2561 
2562 	return (DCMD_OK);
2563 }
2564 
2565 static int
pid2state_init(mdb_walk_state_t * wsp)2566 pid2state_init(mdb_walk_state_t *wsp)
2567 {
2568 	dtrace_state_data_t *data;
2569 	uintptr_t devi;
2570 	uintptr_t proc;
2571 	struct dev_info info;
2572 	pid_t pid = (pid_t)wsp->walk_addr;
2573 
2574 	if (wsp->walk_addr == NULL) {
2575 		mdb_warn("pid2state walk requires PID\n");
2576 		return (WALK_ERR);
2577 	}
2578 
2579 	data = mdb_zalloc(sizeof (dtrace_state_data_t), UM_SLEEP | UM_GC);
2580 
2581 	if (mdb_readvar(&data->dtsd_softstate, "dtrace_softstate") == -1) {
2582 		mdb_warn("failed to read 'dtrace_softstate'");
2583 		return (DCMD_ERR);
2584 	}
2585 
2586 	if ((proc = mdb_pid2proc(pid, NULL)) == NULL) {
2587 		mdb_warn("PID 0t%d not found\n", pid);
2588 		return (DCMD_ERR);
2589 	}
2590 
2591 	if (mdb_readvar(&devi, "dtrace_devi") == -1) {
2592 		mdb_warn("failed to read 'dtrace_devi'");
2593 		return (DCMD_ERR);
2594 	}
2595 
2596 	if (mdb_vread(&info, sizeof (struct dev_info), devi) == -1) {
2597 		mdb_warn("failed to read 'dev_info'");
2598 		return (DCMD_ERR);
2599 	}
2600 
2601 	data->dtsd_major = info.devi_major;
2602 	data->dtsd_proc = proc;
2603 
2604 	wsp->walk_data = data;
2605 
2606 	return (WALK_NEXT);
2607 }
2608 
2609 /*ARGSUSED*/
2610 static int
pid2state_file(uintptr_t addr,struct file * f,dtrace_state_data_t * data)2611 pid2state_file(uintptr_t addr, struct file *f, dtrace_state_data_t *data)
2612 {
2613 	vnode_t vnode;
2614 	minor_t minor;
2615 	uintptr_t statep;
2616 
2617 	/* Get the vnode for this file */
2618 	if (mdb_vread(&vnode, sizeof (vnode), (uintptr_t)f->f_vnode) == -1) {
2619 		mdb_warn("couldn't read vnode at %p", (uintptr_t)f->f_vnode);
2620 		return (WALK_NEXT);
2621 	}
2622 
2623 
2624 	/* Is this the dtrace device? */
2625 	if (getmajor(vnode.v_rdev) != data->dtsd_major)
2626 		return (WALK_NEXT);
2627 
2628 	/* Get the minor number for this device entry */
2629 	minor = getminor(vnode.v_rdev);
2630 
2631 	if (mdb_get_soft_state_byaddr(data->dtsd_softstate, minor,
2632 	    &statep, NULL, 0) == -1) {
2633 		mdb_warn("failed to read softstate for minor %d", minor);
2634 		return (WALK_NEXT);
2635 	}
2636 
2637 	mdb_printf("%p\n", statep);
2638 
2639 	return (WALK_NEXT);
2640 }
2641 
2642 static int
pid2state_step(mdb_walk_state_t * wsp)2643 pid2state_step(mdb_walk_state_t *wsp)
2644 {
2645 	dtrace_state_data_t *ds = wsp->walk_data;
2646 
2647 	if (mdb_pwalk("file",
2648 	    (mdb_walk_cb_t)pid2state_file, ds, ds->dtsd_proc) == -1) {
2649 		mdb_warn("couldn't walk 'file' for proc %p", ds->dtsd_proc);
2650 		return (WALK_ERR);
2651 	}
2652 
2653 	return (WALK_DONE);
2654 }
2655 
2656 /*ARGSUSED*/
2657 static int
dtrace_probes_walk(uintptr_t addr,void * ignored,uintptr_t * target)2658 dtrace_probes_walk(uintptr_t addr, void *ignored, uintptr_t *target)
2659 {
2660 	dtrace_ecb_t ecb;
2661 	dtrace_probe_t probe;
2662 	dtrace_probedesc_t pd;
2663 
2664 	if (addr == NULL)
2665 		return (WALK_ERR);
2666 
2667 	if (mdb_vread(&ecb, sizeof (dtrace_ecb_t), addr) == -1) {
2668 		mdb_warn("failed to read ecb %p\n", addr);
2669 		return (WALK_ERR);
2670 	}
2671 
2672 	if (ecb.dte_probe == NULL)
2673 		return (WALK_ERR);
2674 
2675 	if (mdb_vread(&probe, sizeof (dtrace_probe_t),
2676 	    (uintptr_t)ecb.dte_probe) == -1) {
2677 		mdb_warn("failed to read probe %p\n", ecb.dte_probe);
2678 		return (WALK_ERR);
2679 	}
2680 
2681 	pd.dtpd_id = probe.dtpr_id;
2682 	dtracemdb_probe(NULL, &pd);
2683 
2684 	mdb_printf("%5d %10s %17s %33s %s\n", pd.dtpd_id, pd.dtpd_provider,
2685 	    pd.dtpd_mod, pd.dtpd_func, pd.dtpd_name);
2686 
2687 	return (WALK_NEXT);
2688 }
2689 
2690 static void
dtrace_probes_help(void)2691 dtrace_probes_help(void)
2692 {
2693 	mdb_printf("Given a dtrace_state_t structure, displays all "
2694 	    "its active enablings.  If no\nstate structure is provided, "
2695 	    "all available probes are listed.\n");
2696 }
2697 
2698 /*ARGSUSED*/
2699 static int
dtrace_probes(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)2700 dtrace_probes(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
2701 {
2702 	dtrace_probedesc_t pd;
2703 	uintptr_t caddr, base, paddr;
2704 	int nprobes, i;
2705 
2706 	mdb_printf("%5s %10s %17s %33s %s\n",
2707 	    "ID", "PROVIDER", "MODULE", "FUNCTION", "NAME");
2708 
2709 	if (!(flags & DCMD_ADDRSPEC)) {
2710 		/*
2711 		 * If no argument is provided just display all available
2712 		 * probes.
2713 		 */
2714 		if (mdb_readvar(&base, "dtrace_probes") == -1) {
2715 			mdb_warn("failed to read 'dtrace_probes'");
2716 			return (-1);
2717 		}
2718 
2719 		if (mdb_readvar(&nprobes, "dtrace_nprobes") == -1) {
2720 			mdb_warn("failed to read 'dtrace_nprobes'");
2721 			return (-1);
2722 		}
2723 
2724 		for (i = 0; i < nprobes; i++) {
2725 			caddr = base + i  * sizeof (dtrace_probe_t *);
2726 
2727 			if (mdb_vread(&paddr, sizeof (paddr), caddr) == -1) {
2728 				mdb_warn("couldn't read probe pointer at %p",
2729 				    caddr);
2730 				continue;
2731 			}
2732 
2733 			if (paddr == NULL)
2734 				continue;
2735 
2736 			pd.dtpd_id = i + 1;
2737 			if (dtracemdb_probe(NULL, &pd) == 0) {
2738 				mdb_printf("%5d %10s %17s %33s %s\n",
2739 				    pd.dtpd_id, pd.dtpd_provider,
2740 				    pd.dtpd_mod, pd.dtpd_func, pd.dtpd_name);
2741 			}
2742 		}
2743 	} else {
2744 		if (mdb_pwalk("dtrace_ecb", (mdb_walk_cb_t)dtrace_probes_walk,
2745 		    NULL, addr) == -1) {
2746 			mdb_warn("couldn't walk 'dtrace_ecb'");
2747 			return (DCMD_ERR);
2748 		}
2749 	}
2750 
2751 	return (DCMD_OK);
2752 }
2753 
2754 const mdb_dcmd_t kernel_dcmds[] = {
2755 	{ "id2probe", ":", "translate a dtrace_id_t to a dtrace_probe_t",
2756 	    id2probe },
2757 	{ "dtrace", ":[-c cpu]", "print dtrace(1M)-like output",
2758 	    dtrace, dtrace_help },
2759 	{ "dtrace_errhash", ":", "print DTrace error hash", dtrace_errhash },
2760 	{ "dtrace_helptrace", ":", "print DTrace helper trace",
2761 	    dtrace_helptrace },
2762 	{ "dtrace_state", ":", "print active DTrace consumers", dtrace_state,
2763 	    dtrace_state_help },
2764 	{ "dtrace_aggstat", ":",
2765 	    "print DTrace aggregation hash statistics", dtrace_aggstat },
2766 	{ "dtrace_dynstat", ":",
2767 	    "print DTrace dynamic variable hash statistics", dtrace_dynstat },
2768 	{ "dtrace_options", ":",
2769 	    "print a DTrace consumer's current tuneable options",
2770 	    dtrace_options, dtrace_options_help },
2771 	{ "dtrace_probes", "?", "print a DTrace consumer's enabled probes",
2772 	    dtrace_probes, dtrace_probes_help },
2773 	{ NULL }
2774 };
2775 
2776 const mdb_walker_t kernel_walkers[] = {
2777 	{ "dtrace_errhash", "walk hash of DTrace error messasges",
2778 		dtrace_errhash_init, dtrace_errhash_step },
2779 	{ "dtrace_helptrace", "walk DTrace helper trace entries",
2780 		dtrace_helptrace_init, dtrace_helptrace_step },
2781 	{ "dtrace_state", "walk DTrace per-consumer softstate",
2782 		dtrace_state_init, dtrace_state_step },
2783 	{ "dtrace_aggkey", "walk DTrace aggregation keys",
2784 		dtrace_aggkey_init, dtrace_aggkey_step, dtrace_aggkey_fini },
2785 	{ "dtrace_dynvar", "walk DTrace dynamic variables",
2786 		dtrace_dynvar_init, dtrace_dynvar_step, dtrace_dynvar_fini },
2787 	{ "dtrace_ecb", "walk a DTrace consumer's enabling control blocks",
2788 		dtrace_ecb_init, dtrace_ecb_step },
2789 	{ "pid2state", "walk a processes dtrace_state structures",
2790 	    pid2state_init, pid2state_step },
2791 	{ NULL }
2792 };
2793