xref: /illumos-gate/usr/src/cmd/mdb/common/modules/mdb_ks/mdb_ks.c (revision 044da6e1a8d008f558b8a0a17f173f2483276900)
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  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2019 Joyent, Inc.
24  * Copyright 2023 RackTop Systems, Inc.
25  * Copyright 2025 Oxide Computer Company
26  */
27 
28 /*
29  * Mdb kernel support module.  This module is loaded automatically when the
30  * kvm target is initialized.  Any global functions declared here are exported
31  * for the resolution of symbols in subsequently loaded modules.
32  *
33  * WARNING: Do not assume that static variables in mdb_ks will be initialized
34  * to zero.
35  */
36 
37 #include <mdb/mdb_target.h>
38 #include <mdb/mdb_param.h>
39 #include <mdb/mdb_modapi.h>
40 #include <mdb/mdb_ctf.h>
41 #include <mdb/mdb_ks.h>
42 
43 #include <sys/types.h>
44 #include <sys/procfs.h>
45 #include <sys/proc.h>
46 #include <sys/dnlc.h>
47 #include <sys/autoconf.h>
48 #include <sys/machelf.h>
49 #include <sys/modctl.h>
50 #include <sys/hwconf.h>
51 #include <sys/kobj.h>
52 #include <sys/fs/autofs.h>
53 #include <sys/ddi_impldefs.h>
54 #include <sys/refstr_impl.h>
55 #include <sys/cpuvar.h>
56 #include <sys/dlpi.h>
57 #include <sys/clock_impl.h>
58 #include <sys/swap.h>
59 #include <errno.h>
60 
61 #include <vm/seg_vn.h>
62 #include <vm/page.h>
63 
64 #define	MDB_PATH_NELEM	256			/* Maximum path components */
65 
66 typedef struct mdb_path {
67 	size_t mdp_nelem;			/* Number of components */
68 	uint_t mdp_complete;			/* Path completely resolved? */
69 	uintptr_t mdp_vnode[MDB_PATH_NELEM];	/* Array of vnode_t addresses */
70 	char *mdp_name[MDB_PATH_NELEM];		/* Array of name components */
71 } mdb_path_t;
72 
73 static int mdb_autonode2path(uintptr_t, mdb_path_t *);
74 static int mdb_sprintpath(char *, size_t, mdb_path_t *);
75 
76 /*
77  * Kernel parameters from <sys/param.h> which we keep in-core:
78  */
79 unsigned long _mdb_ks_pagesize;
80 unsigned int _mdb_ks_pageshift;
81 unsigned long _mdb_ks_pageoffset;
82 unsigned long long _mdb_ks_pagemask;
83 unsigned long _mdb_ks_mmu_pagesize;
84 unsigned int _mdb_ks_mmu_pageshift;
85 unsigned long _mdb_ks_mmu_pageoffset;
86 unsigned long _mdb_ks_mmu_pagemask;
87 uintptr_t _mdb_ks_kernelbase;
88 uintptr_t _mdb_ks_userlimit;
89 uintptr_t _mdb_ks_userlimit32;
90 unsigned long _mdb_ks_msg_bsize;
91 unsigned long _mdb_ks_defaultstksz;
92 int _mdb_ks_ncpu;
93 int _mdb_ks_ncpu_log2;
94 int _mdb_ks_ncpu_p2;
95 
96 /*
97  * In-core copy of DNLC information:
98  */
99 #define	MDB_DNLC_HSIZE	1024
100 #define	MDB_DNLC_HASH(vp)	(((uintptr_t)(vp) >> 3) & (MDB_DNLC_HSIZE - 1))
101 #define	MDB_DNLC_NCACHE_SZ(ncp) (sizeof (ncache_t) + (ncp)->namlen)
102 #define	MDB_DNLC_MAX_RETRY 4
103 
104 static ncache_t **dnlc_hash;	/* mdbs hash array of dnlc entries */
105 
106 /*
107  * copy of page_hash-related data
108  */
109 static int page_hash_loaded;
110 static long mdb_page_hashsz;
111 static uint_t mdb_page_hashsz_shift;	/* Needed for PAGE_HASH_FUNC */
112 static uintptr_t mdb_page_hash;		/* base address of page hash */
113 #define	page_hashsz		mdb_page_hashsz
114 #define	page_hashsz_shift	mdb_page_hashsz_shift
115 
116 /*
117  * This will be the location of the vnodeops pointer for "autofs_vnodeops"
118  * The pointer still needs to be read with mdb_vread() to get the location
119  * of the vnodeops structure for autofs.
120  */
121 static struct vnodeops *autofs_vnops_ptr;
122 
123 /*
124  * STREAMS queue registrations:
125  */
126 typedef struct mdb_qinfo {
127 	const mdb_qops_t *qi_ops;	/* Address of ops vector */
128 	uintptr_t qi_addr;		/* Address of qinit structure (key) */
129 	struct mdb_qinfo *qi_next;	/* Next qinfo in list */
130 } mdb_qinfo_t;
131 
132 static mdb_qinfo_t *qi_head;		/* Head of qinfo chain */
133 
134 /*
135  * Device naming callback structure:
136  */
137 typedef struct nm_query {
138 	const char *nm_name;		/* Device driver name [in/out] */
139 	major_t nm_major;		/* Device major number [in/out] */
140 	ushort_t nm_found;		/* Did we find a match? [out] */
141 } nm_query_t;
142 
143 /*
144  * Address-to-modctl callback structure:
145  */
146 typedef struct a2m_query {
147 	uintptr_t a2m_addr;		/* Virtual address [in] */
148 	uintptr_t a2m_where;		/* Modctl address [out] */
149 } a2m_query_t;
150 
151 /*
152  * Segment-to-mdb_map callback structure:
153  */
154 typedef struct {
155 	struct seg_ops *asm_segvn_ops;	/* Address of segvn ops [in] */
156 	void (*asm_callback)(const struct mdb_map *, void *); /* Callb [in] */
157 	void *asm_cbdata;		/* Callback data [in] */
158 } asmap_arg_t;
159 
160 static void
dnlc_free(void)161 dnlc_free(void)
162 {
163 	ncache_t *ncp, *next;
164 	int i;
165 
166 	if (dnlc_hash == NULL) {
167 		return;
168 	}
169 
170 	/*
171 	 * Free up current dnlc entries
172 	 */
173 	for (i = 0; i < MDB_DNLC_HSIZE; i++) {
174 		for (ncp = dnlc_hash[i]; ncp; ncp = next) {
175 			next = ncp->hash_next;
176 			mdb_free(ncp, MDB_DNLC_NCACHE_SZ(ncp));
177 		}
178 	}
179 	mdb_free(dnlc_hash, MDB_DNLC_HSIZE * sizeof (ncache_t *));
180 	dnlc_hash = NULL;
181 }
182 
183 char bad_dnlc[] = "inconsistent dnlc chain: %d, ncache va: %p"
184 	" - continuing with the rest\n";
185 
186 static int
dnlc_load(void)187 dnlc_load(void)
188 {
189 	int i; /* hash index */
190 	int retry_cnt = 0;
191 	int skip_bad_chains = 0;
192 	int nc_hashsz; /* kernel hash array size */
193 	uintptr_t nc_hash_addr; /* kernel va of ncache hash array */
194 	uintptr_t head; /* kernel va of head of hash chain */
195 
196 	/*
197 	 * If we've already cached the DNLC and we're looking at a dump,
198 	 * our cache is good forever, so don't bother re-loading.
199 	 */
200 	if (dnlc_hash && mdb_prop_postmortem) {
201 		return (0);
202 	}
203 
204 	/*
205 	 * For a core dump, retries wont help.
206 	 * Just print and skip any bad chains.
207 	 */
208 	if (mdb_prop_postmortem) {
209 		skip_bad_chains = 1;
210 	}
211 retry:
212 	if (retry_cnt++ >= MDB_DNLC_MAX_RETRY) {
213 		/*
214 		 * Give up retrying the rapidly changing dnlc.
215 		 * Just print and skip any bad chains
216 		 */
217 		skip_bad_chains = 1;
218 	}
219 
220 	dnlc_free(); /* Free up the mdb hashed dnlc - if any */
221 
222 	/*
223 	 * Although nc_hashsz and the location of nc_hash doesn't currently
224 	 * change, it may do in the future with a more dynamic dnlc.
225 	 * So always read these values afresh.
226 	 */
227 	if (mdb_readvar(&nc_hashsz, "nc_hashsz") == -1) {
228 		mdb_warn("failed to read nc_hashsz");
229 		return (-1);
230 	}
231 	if (mdb_readvar(&nc_hash_addr, "nc_hash") == -1) {
232 		mdb_warn("failed to read nc_hash");
233 		return (-1);
234 	}
235 
236 	/*
237 	 * Allocate the mdb dnlc hash array
238 	 */
239 	dnlc_hash = mdb_zalloc(MDB_DNLC_HSIZE * sizeof (ncache_t *), UM_SLEEP);
240 
241 	/* for each kernel hash chain */
242 	for (i = 0, head = nc_hash_addr; i < nc_hashsz;
243 	    i++, head += sizeof (nc_hash_t)) {
244 		nc_hash_t nch; /* kernel hash chain header */
245 		ncache_t *ncp; /* name cache pointer */
246 		int hash; /* mdb hash value */
247 		uintptr_t nc_va; /* kernel va of next ncache */
248 		uintptr_t ncprev_va; /* kernel va of previous ncache */
249 		int khash; /* kernel dnlc hash value */
250 		uchar_t namelen; /* name length */
251 		ncache_t nc; /* name cache entry */
252 		int nc_size; /* size of a name cache entry */
253 
254 		/*
255 		 * We read each element of the nc_hash array individually
256 		 * just before we process the entries in its chain. This is
257 		 * because the chain can change so rapidly on a running system.
258 		 */
259 		if (mdb_vread(&nch, sizeof (nc_hash_t), head) == -1) {
260 			mdb_warn("failed to read nc_hash chain header %d", i);
261 			dnlc_free();
262 			return (-1);
263 		}
264 
265 		ncprev_va = head;
266 		nc_va = (uintptr_t)(nch.hash_next);
267 		/* for each entry in the chain */
268 		while (nc_va != head) {
269 			/*
270 			 * The size of the ncache entries varies
271 			 * because the name is appended to the structure.
272 			 * So we read in the structure then re-read
273 			 * for the structure plus name.
274 			 */
275 			if (mdb_vread(&nc, sizeof (ncache_t), nc_va) == -1) {
276 				if (skip_bad_chains) {
277 					mdb_warn(bad_dnlc, i, nc_va);
278 					break;
279 				}
280 				goto retry;
281 			}
282 			nc_size = MDB_DNLC_NCACHE_SZ(&nc);
283 			ncp = mdb_alloc(nc_size, UM_SLEEP);
284 			if (mdb_vread(ncp, nc_size - 1, nc_va) == -1) {
285 				mdb_free(ncp, nc_size);
286 				if (skip_bad_chains) {
287 					mdb_warn(bad_dnlc, i, nc_va);
288 					break;
289 				}
290 				goto retry;
291 			}
292 
293 			/*
294 			 * Check for chain consistency
295 			 */
296 			if ((uintptr_t)ncp->hash_prev != ncprev_va) {
297 				mdb_free(ncp, nc_size);
298 				if (skip_bad_chains) {
299 					mdb_warn(bad_dnlc, i, nc_va);
300 					break;
301 				}
302 				goto retry;
303 			}
304 			/*
305 			 * Terminate the new name with a null.
306 			 * Note, we allowed space for this null when
307 			 * allocating space for the entry.
308 			 */
309 			ncp->name[ncp->namlen] = '\0';
310 
311 			/*
312 			 * Validate new entry by re-hashing using the
313 			 * kernel dnlc hash function and comparing the hash
314 			 */
315 			DNLCHASH(ncp->name, ncp->dp, khash, namelen);
316 			if ((namelen != ncp->namlen) ||
317 			    (khash != ncp->hash)) {
318 				mdb_free(ncp, nc_size);
319 				if (skip_bad_chains) {
320 					mdb_warn(bad_dnlc, i, nc_va);
321 					break;
322 				}
323 				goto retry;
324 			}
325 
326 			/*
327 			 * Finally put the validated entry into the mdb
328 			 * hash chains. Reuse the kernel next hash field
329 			 * for the mdb hash chain pointer.
330 			 */
331 			hash = MDB_DNLC_HASH(ncp->vp);
332 			ncprev_va = nc_va;
333 			nc_va = (uintptr_t)(ncp->hash_next);
334 			ncp->hash_next = dnlc_hash[hash];
335 			dnlc_hash[hash] = ncp;
336 		}
337 	}
338 	return (0);
339 }
340 
341 /*ARGSUSED*/
342 int
dnlcdump(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)343 dnlcdump(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
344 {
345 	ncache_t *ent;
346 	int i;
347 
348 	if ((flags & DCMD_ADDRSPEC) || argc != 0)
349 		return (DCMD_USAGE);
350 
351 	if (dnlc_load() == -1)
352 		return (DCMD_ERR);
353 
354 	mdb_printf("%<u>%-?s %-?s %-32s%</u>\n", "VP", "DVP", "NAME");
355 
356 	for (i = 0; i < MDB_DNLC_HSIZE; i++) {
357 		for (ent = dnlc_hash[i]; ent != NULL; ent = ent->hash_next) {
358 			mdb_printf("%0?p %0?p %s\n",
359 			    ent->vp, ent->dp, ent->name);
360 		}
361 	}
362 
363 	return (DCMD_OK);
364 }
365 
366 static int
mdb_sprintpath(char * buf,size_t len,mdb_path_t * path)367 mdb_sprintpath(char *buf, size_t len, mdb_path_t *path)
368 {
369 	char *s = buf;
370 	int i;
371 
372 	if (len < sizeof ("/..."))
373 		return (-1);
374 
375 	if (!path->mdp_complete) {
376 		(void) strcpy(s, "??");
377 		s += 2;
378 
379 		if (path->mdp_nelem == 0)
380 			return (-1);
381 	}
382 
383 	if (path->mdp_nelem == 0) {
384 		(void) strcpy(s, "/");
385 		return (0);
386 	}
387 
388 	for (i = path->mdp_nelem - 1; i >= 0; i--) {
389 		/*
390 		 * Number of bytes left is the distance from where we
391 		 * are to the end, minus 2 for '/' and '\0'
392 		 */
393 		ssize_t left = (ssize_t)(&buf[len] - s) - 2;
394 
395 		if (left <= 0)
396 			break;
397 
398 		*s++ = '/';
399 		(void) strncpy(s, path->mdp_name[i], left);
400 		s[left - 1] = '\0';
401 		s += strlen(s);
402 
403 		if (left < strlen(path->mdp_name[i]))
404 			break;
405 	}
406 
407 	if (i >= 0)
408 		(void) strcpy(&buf[len - 4], "...");
409 
410 	return (0);
411 }
412 
413 static int
mdb_autonode2path(uintptr_t addr,mdb_path_t * path)414 mdb_autonode2path(uintptr_t addr, mdb_path_t *path)
415 {
416 	fninfo_t fni;
417 	fnnode_t fn;
418 
419 	vnode_t vn;
420 	vfs_t vfs;
421 	struct vnodeops *autofs_vnops = NULL;
422 
423 	/*
424 	 * "autofs_vnops_ptr" is the address of the pointer to the vnodeops
425 	 * structure for autofs.  We want to read it each time we access
426 	 * it since autofs could (in theory) be unloaded and reloaded.
427 	 */
428 	if (mdb_vread(&autofs_vnops, sizeof (autofs_vnops),
429 	    (uintptr_t)autofs_vnops_ptr) == -1)
430 		return (-1);
431 
432 	if (mdb_vread(&vn, sizeof (vn), addr) == -1)
433 		return (-1);
434 
435 	if (autofs_vnops == NULL || vn.v_op != autofs_vnops)
436 		return (-1);
437 
438 	addr = (uintptr_t)vn.v_data;
439 
440 	if (mdb_vread(&vfs, sizeof (vfs), (uintptr_t)vn.v_vfsp) == -1 ||
441 	    mdb_vread(&fni, sizeof (fni), (uintptr_t)vfs.vfs_data) == -1 ||
442 	    mdb_vread(&vn, sizeof (vn), (uintptr_t)fni.fi_rootvp) == -1)
443 		return (-1);
444 
445 	for (;;) {
446 		size_t elem = path->mdp_nelem++;
447 		char elemstr[MAXNAMELEN];
448 		char *c, *p;
449 
450 		if (elem == MDB_PATH_NELEM) {
451 			path->mdp_nelem--;
452 			return (-1);
453 		}
454 
455 		if (mdb_vread(&fn, sizeof (fn), addr) != sizeof (fn)) {
456 			path->mdp_nelem--;
457 			return (-1);
458 		}
459 
460 		if (mdb_readstr(elemstr, sizeof (elemstr),
461 		    (uintptr_t)fn.fn_name) <= 0) {
462 			(void) strcpy(elemstr, "?");
463 		}
464 
465 		c = mdb_alloc(strlen(elemstr) + 1, UM_SLEEP | UM_GC);
466 		(void) strcpy(c, elemstr);
467 
468 		path->mdp_vnode[elem] = (uintptr_t)fn.fn_vnode;
469 
470 		if (addr == (uintptr_t)fn.fn_parent) {
471 			path->mdp_name[elem] = &c[1];
472 			path->mdp_complete = TRUE;
473 			break;
474 		}
475 
476 		if ((p = strrchr(c, '/')) != NULL)
477 			path->mdp_name[elem] = p + 1;
478 		else
479 			path->mdp_name[elem] = c;
480 
481 		addr = (uintptr_t)fn.fn_parent;
482 	}
483 
484 	return (0);
485 }
486 
487 int
mdb_vnode2path(uintptr_t addr,char * buf,size_t buflen)488 mdb_vnode2path(uintptr_t addr, char *buf, size_t buflen)
489 {
490 	uintptr_t rootdir;
491 	ncache_t *ent;
492 	vnode_t vp;
493 	mdb_path_t path;
494 
495 	/*
496 	 * Check to see if we have a cached value for this vnode
497 	 */
498 	if (mdb_vread(&vp, sizeof (vp), addr) != -1 &&
499 	    vp.v_path != NULL &&
500 	    mdb_readstr(buf, buflen, (uintptr_t)vp.v_path) != -1)
501 		return (0);
502 
503 	if (dnlc_load() == -1)
504 		return (-1);
505 
506 	if (mdb_readvar(&rootdir, "rootdir") == -1) {
507 		mdb_warn("failed to read 'rootdir'");
508 		return (-1);
509 	}
510 
511 	bzero(&path, sizeof (mdb_path_t));
512 again:
513 	if ((addr == 0) && (path.mdp_nelem == 0)) {
514 		/*
515 		 * 0 elems && complete tells sprintpath to just print "/"
516 		 */
517 		path.mdp_complete = TRUE;
518 		goto out;
519 	}
520 
521 	if (addr == rootdir) {
522 		path.mdp_complete = TRUE;
523 		goto out;
524 	}
525 
526 	for (ent = dnlc_hash[MDB_DNLC_HASH(addr)]; ent; ent = ent->hash_next) {
527 		if ((uintptr_t)ent->vp == addr) {
528 			if (strcmp(ent->name, "..") == 0 ||
529 			    strcmp(ent->name, ".") == 0)
530 				continue;
531 
532 			path.mdp_vnode[path.mdp_nelem] = (uintptr_t)ent->vp;
533 			path.mdp_name[path.mdp_nelem] = ent->name;
534 			path.mdp_nelem++;
535 
536 			if (path.mdp_nelem == MDB_PATH_NELEM) {
537 				path.mdp_nelem--;
538 				mdb_warn("path exceeded maximum expected "
539 				    "elements\n");
540 				return (-1);
541 			}
542 
543 			addr = (uintptr_t)ent->dp;
544 			goto again;
545 		}
546 	}
547 
548 	(void) mdb_autonode2path(addr, &path);
549 
550 out:
551 	return (mdb_sprintpath(buf, buflen, &path));
552 }
553 
554 
555 uintptr_t
mdb_pid2proc(pid_t pid,proc_t * proc)556 mdb_pid2proc(pid_t pid, proc_t *proc)
557 {
558 	int pid_hashsz, hash;
559 	uintptr_t paddr, pidhash, procdir;
560 	struct pid pidp;
561 
562 	if (mdb_readvar(&pidhash, "pidhash") == -1)
563 		return (0);
564 
565 	if (mdb_readvar(&pid_hashsz, "pid_hashsz") == -1)
566 		return (0);
567 
568 	if (mdb_readvar(&procdir, "procdir") == -1)
569 		return (0);
570 
571 	hash = pid & (pid_hashsz - 1);
572 
573 	if (mdb_vread(&paddr, sizeof (paddr),
574 	    pidhash + (hash * sizeof (paddr))) == -1)
575 		return (0);
576 
577 	while (paddr != 0) {
578 		if (mdb_vread(&pidp, sizeof (pidp), paddr) == -1)
579 			return (0);
580 
581 		if (pidp.pid_id == pid) {
582 			uintptr_t procp;
583 
584 			if (mdb_vread(&procp, sizeof (procp), procdir +
585 			    (pidp.pid_prslot * sizeof (procp))) == -1)
586 				return (0);
587 
588 			if (proc != NULL)
589 				(void) mdb_vread(proc, sizeof (proc_t), procp);
590 
591 			return (procp);
592 		}
593 		paddr = (uintptr_t)pidp.pid_link;
594 	}
595 	return (0);
596 }
597 
598 int
mdb_cpu2cpuid(uintptr_t cpup)599 mdb_cpu2cpuid(uintptr_t cpup)
600 {
601 	cpu_t cpu;
602 
603 	if (mdb_vread(&cpu, sizeof (cpu_t), cpup) != sizeof (cpu_t))
604 		return (-1);
605 
606 	return (cpu.cpu_id);
607 }
608 
609 int
mdb_cpuset_find(uintptr_t cpusetp)610 mdb_cpuset_find(uintptr_t cpusetp)
611 {
612 	ulong_t	*cpuset;
613 	size_t nr_words = BT_BITOUL(NCPU);
614 	size_t sz = nr_words * sizeof (ulong_t);
615 	size_t	i;
616 	int cpu = -1;
617 
618 	cpuset = mdb_alloc(sz, UM_SLEEP);
619 
620 	if (mdb_vread((void *)cpuset, sz, cpusetp) != sz)
621 		goto out;
622 
623 	for (i = 0; i < nr_words; i++) {
624 		size_t j;
625 		ulong_t m;
626 
627 		for (j = 0, m = 1; j < BT_NBIPUL; j++, m <<= 1) {
628 			if (cpuset[i] & m) {
629 				cpu = i * BT_NBIPUL + j;
630 				goto out;
631 			}
632 		}
633 	}
634 
635 out:
636 	mdb_free(cpuset, sz);
637 	return (cpu);
638 }
639 
640 static int
page_hash_load(void)641 page_hash_load(void)
642 {
643 	if (page_hash_loaded) {
644 		return (1);
645 	}
646 
647 	if (mdb_readvar(&mdb_page_hashsz, "page_hashsz") == -1) {
648 		mdb_warn("unable to read page_hashsz");
649 		return (0);
650 	}
651 	if (mdb_readvar(&mdb_page_hashsz_shift, "page_hashsz_shift") == -1) {
652 		mdb_warn("unable to read page_hashsz_shift");
653 		return (0);
654 	}
655 	if (mdb_readvar(&mdb_page_hash, "page_hash") == -1) {
656 		mdb_warn("unable to read page_hash");
657 		return (0);
658 	}
659 
660 	page_hash_loaded = 1;	/* zeroed on state change */
661 	return (1);
662 }
663 
664 uintptr_t
mdb_page_lookup(uintptr_t vp,u_offset_t offset)665 mdb_page_lookup(uintptr_t vp, u_offset_t offset)
666 {
667 	size_t ndx;
668 	uintptr_t page_hash_entry, pp;
669 
670 	if (!page_hash_loaded && !page_hash_load()) {
671 		return (0);
672 	}
673 
674 	ndx = PAGE_HASH_FUNC(vp, offset);
675 	page_hash_entry = mdb_page_hash + ndx * sizeof (uintptr_t);
676 
677 	if (mdb_vread(&pp, sizeof (pp), page_hash_entry) < 0) {
678 		mdb_warn("unable to read page_hash[%ld] (%p)", ndx,
679 		    page_hash_entry);
680 		return (0);
681 	}
682 
683 	while (pp != 0) {
684 		page_t page;
685 		long nndx;
686 
687 		if (mdb_vread(&page, sizeof (page), pp) < 0) {
688 			mdb_warn("unable to read page_t at %p", pp);
689 			return (0);
690 		}
691 
692 		if ((uintptr_t)page.p_vnode == vp &&
693 		    (uint64_t)page.p_offset == offset)
694 			return (pp);
695 
696 		/*
697 		 * Double check that the pages actually hash to the
698 		 * bucket we're searching.  If not, our version of
699 		 * PAGE_HASH_FUNC() doesn't match the kernel's, and we're
700 		 * not going to be able to find the page.  The most
701 		 * likely reason for this that mdb_ks doesn't match the
702 		 * kernel we're running against.
703 		 */
704 		nndx = PAGE_HASH_FUNC(page.p_vnode, page.p_offset);
705 		if (page.p_vnode != NULL && nndx != ndx) {
706 			mdb_warn("mdb_page_lookup: mdb_ks PAGE_HASH_FUNC() "
707 			    "mismatch: in bucket %ld, but page %p hashes to "
708 			    "bucket %ld\n", ndx, pp, nndx);
709 			return (0);
710 		}
711 
712 		pp = (uintptr_t)page.p_hash;
713 	}
714 
715 	return (0);
716 }
717 
718 char
mdb_vtype2chr(vtype_t type,mode_t mode)719 mdb_vtype2chr(vtype_t type, mode_t mode)
720 {
721 	static const char vttab[] = {
722 		' ',	/* VNON */
723 		' ',	/* VREG */
724 		'/',	/* VDIR */
725 		' ',	/* VBLK */
726 		' ',	/* VCHR */
727 		'@',	/* VLNK */
728 		'|',	/* VFIFO */
729 		'>',	/* VDOOR */
730 		' ',	/* VPROC */
731 		'=',	/* VSOCK */
732 		' ',	/* VBAD */
733 	};
734 
735 	if (type < 0 || type >= sizeof (vttab) / sizeof (vttab[0]))
736 		return ('?');
737 
738 	if (type == VREG && (mode & 0111) != 0)
739 		return ('*');
740 
741 	return (vttab[type]);
742 }
743 
744 struct pfn2page {
745 	pfn_t pfn;
746 	page_t *pp;
747 };
748 
749 /*ARGSUSED*/
750 static int
pfn2page_cb(uintptr_t addr,const struct memseg * msp,void * data)751 pfn2page_cb(uintptr_t addr, const struct memseg *msp, void *data)
752 {
753 	struct pfn2page *p = data;
754 
755 	if (p->pfn >= msp->pages_base && p->pfn < msp->pages_end) {
756 		p->pp = msp->pages + (p->pfn - msp->pages_base);
757 		return (WALK_DONE);
758 	}
759 
760 	return (WALK_NEXT);
761 }
762 
763 uintptr_t
mdb_pfn2page(pfn_t pfn)764 mdb_pfn2page(pfn_t pfn)
765 {
766 	struct pfn2page	arg;
767 	struct page	page;
768 
769 	arg.pfn = pfn;
770 	arg.pp = NULL;
771 
772 	if (mdb_walk("memseg", (mdb_walk_cb_t)pfn2page_cb, &arg) == -1) {
773 		mdb_warn("pfn2page: can't walk memsegs");
774 		return (0);
775 	}
776 	if (arg.pp == NULL) {
777 		mdb_warn("pfn2page: unable to find page_t for pfn %lx\n",
778 		    pfn);
779 		return (0);
780 	}
781 
782 	if (mdb_vread(&page, sizeof (page_t), (uintptr_t)arg.pp) == -1) {
783 		mdb_warn("pfn2page: can't read page 0x%lx at %p", pfn, arg.pp);
784 		return (0);
785 	}
786 	if (page.p_pagenum != pfn) {
787 		mdb_warn("pfn2page: page_t 0x%p should have PFN 0x%lx, "
788 		    "but actually has 0x%lx\n", arg.pp, pfn, page.p_pagenum);
789 		return (0);
790 	}
791 
792 	return ((uintptr_t)arg.pp);
793 }
794 
795 pfn_t
mdb_page2pfn(uintptr_t addr)796 mdb_page2pfn(uintptr_t addr)
797 {
798 	struct page	page;
799 
800 	if (mdb_vread(&page, sizeof (page_t), addr) == -1) {
801 		mdb_warn("pp2pfn: can't read page at %p", addr);
802 		return ((pfn_t)(-1));
803 	}
804 
805 	return (page.p_pagenum);
806 }
807 
808 struct a2m_ctf_module {
809 	char *text;
810 	char *data;
811 	size_t text_size;
812 	size_t data_size;
813 };
814 
815 static int
a2m_walk_modctl(uintptr_t addr,const struct modctl * m,a2m_query_t * a2m)816 a2m_walk_modctl(uintptr_t addr, const struct modctl *m, a2m_query_t *a2m)
817 {
818 	struct a2m_ctf_module mod;
819 
820 	if (m->mod_mp == NULL)
821 		return (0);
822 
823 	if (mdb_ctf_vread(&mod, "struct module", "struct a2m_ctf_module",
824 	    (uintptr_t)m->mod_mp, 0) == -1) {
825 		mdb_warn("couldn't read modctl %p's module", addr);
826 		return (0);
827 	}
828 
829 	if (a2m->a2m_addr >= (uintptr_t)mod.text &&
830 	    a2m->a2m_addr < (uintptr_t)mod.text + mod.text_size)
831 		goto found;
832 
833 	if (a2m->a2m_addr >= (uintptr_t)mod.data &&
834 	    a2m->a2m_addr < (uintptr_t)mod.data + mod.data_size)
835 		goto found;
836 
837 	return (0);
838 
839 found:
840 	a2m->a2m_where = addr;
841 	return (-1);
842 }
843 
844 uintptr_t
mdb_addr2modctl(uintptr_t addr)845 mdb_addr2modctl(uintptr_t addr)
846 {
847 	a2m_query_t a2m;
848 
849 	a2m.a2m_addr = addr;
850 	a2m.a2m_where = 0;
851 
852 	(void) mdb_walk("modctl", (mdb_walk_cb_t)a2m_walk_modctl, &a2m);
853 	return (a2m.a2m_where);
854 }
855 
856 static mdb_qinfo_t *
qi_lookup(uintptr_t qinit_addr)857 qi_lookup(uintptr_t qinit_addr)
858 {
859 	mdb_qinfo_t *qip;
860 
861 	for (qip = qi_head; qip != NULL; qip = qip->qi_next) {
862 		if (qip->qi_addr == qinit_addr)
863 			return (qip);
864 	}
865 
866 	return (NULL);
867 }
868 
869 void
mdb_qops_install(const mdb_qops_t * qops,uintptr_t qinit_addr)870 mdb_qops_install(const mdb_qops_t *qops, uintptr_t qinit_addr)
871 {
872 	mdb_qinfo_t *qip = qi_lookup(qinit_addr);
873 
874 	if (qip != NULL) {
875 		qip->qi_ops = qops;
876 		return;
877 	}
878 
879 	qip = mdb_alloc(sizeof (mdb_qinfo_t), UM_SLEEP);
880 
881 	qip->qi_ops = qops;
882 	qip->qi_addr = qinit_addr;
883 	qip->qi_next = qi_head;
884 
885 	qi_head = qip;
886 }
887 
888 void
mdb_qops_remove(const mdb_qops_t * qops,uintptr_t qinit_addr)889 mdb_qops_remove(const mdb_qops_t *qops, uintptr_t qinit_addr)
890 {
891 	mdb_qinfo_t *qip, *p = NULL;
892 
893 	for (qip = qi_head; qip != NULL; p = qip, qip = qip->qi_next) {
894 		if (qip->qi_addr == qinit_addr && qip->qi_ops == qops) {
895 			if (qi_head == qip)
896 				qi_head = qip->qi_next;
897 			else
898 				p->qi_next = qip->qi_next;
899 			mdb_free(qip, sizeof (mdb_qinfo_t));
900 			return;
901 		}
902 	}
903 }
904 
905 char *
mdb_qname(const queue_t * q,char * buf,size_t nbytes)906 mdb_qname(const queue_t *q, char *buf, size_t nbytes)
907 {
908 	struct module_info mi;
909 	struct qinit qi;
910 
911 	if (mdb_vread(&qi, sizeof (qi), (uintptr_t)q->q_qinfo) == -1) {
912 		mdb_warn("failed to read qinit at %p", q->q_qinfo);
913 		goto err;
914 	}
915 
916 	if (mdb_vread(&mi, sizeof (mi), (uintptr_t)qi.qi_minfo) == -1) {
917 		mdb_warn("failed to read module_info at %p", qi.qi_minfo);
918 		goto err;
919 	}
920 
921 	if (mdb_readstr(buf, nbytes, (uintptr_t)mi.mi_idname) <= 0) {
922 		mdb_warn("failed to read mi_idname at %p", mi.mi_idname);
923 		goto err;
924 	}
925 
926 	return (buf);
927 
928 err:
929 	(void) mdb_snprintf(buf, nbytes, "???");
930 	return (buf);
931 }
932 
933 void
mdb_qinfo(const queue_t * q,char * buf,size_t nbytes)934 mdb_qinfo(const queue_t *q, char *buf, size_t nbytes)
935 {
936 	mdb_qinfo_t *qip = qi_lookup((uintptr_t)q->q_qinfo);
937 	buf[0] = '\0';
938 
939 	if (qip != NULL)
940 		qip->qi_ops->q_info(q, buf, nbytes);
941 }
942 
943 uintptr_t
mdb_qrnext(const queue_t * q)944 mdb_qrnext(const queue_t *q)
945 {
946 	mdb_qinfo_t *qip = qi_lookup((uintptr_t)q->q_qinfo);
947 
948 	if (qip != NULL)
949 		return (qip->qi_ops->q_rnext(q));
950 
951 	return (0);
952 }
953 
954 uintptr_t
mdb_qwnext(const queue_t * q)955 mdb_qwnext(const queue_t *q)
956 {
957 	mdb_qinfo_t *qip = qi_lookup((uintptr_t)q->q_qinfo);
958 
959 	if (qip != NULL)
960 		return (qip->qi_ops->q_wnext(q));
961 
962 	return (0);
963 }
964 
965 uintptr_t
mdb_qrnext_default(const queue_t * q)966 mdb_qrnext_default(const queue_t *q)
967 {
968 	return ((uintptr_t)q->q_next);
969 }
970 
971 uintptr_t
mdb_qwnext_default(const queue_t * q)972 mdb_qwnext_default(const queue_t *q)
973 {
974 	return ((uintptr_t)q->q_next);
975 }
976 
977 /*
978  * The following three routines borrowed from modsubr.c
979  */
980 static int
nm_hash(const char * name)981 nm_hash(const char *name)
982 {
983 	char c;
984 	int hash = 0;
985 
986 	for (c = *name++; c; c = *name++)
987 		hash ^= c;
988 
989 	return (hash & MOD_BIND_HASHMASK);
990 }
991 
992 static uintptr_t
find_mbind(const char * name,uintptr_t * hashtab)993 find_mbind(const char *name, uintptr_t *hashtab)
994 {
995 	int hashndx;
996 	uintptr_t mb;
997 	struct bind mb_local;
998 	char node_name[MAXPATHLEN + 1];
999 
1000 	hashndx = nm_hash(name);
1001 	mb = hashtab[hashndx];
1002 	while (mb) {
1003 		if (mdb_vread(&mb_local, sizeof (mb_local), mb) == -1) {
1004 			mdb_warn("failed to read struct bind at %p", mb);
1005 			return (0);
1006 		}
1007 		if (mdb_readstr(node_name, sizeof (node_name),
1008 		    (uintptr_t)mb_local.b_name) == -1) {
1009 			mdb_warn("failed to read node name string at %p",
1010 			    mb_local.b_name);
1011 			return (0);
1012 		}
1013 
1014 		if (strcmp(name, node_name) == 0)
1015 			break;
1016 
1017 		mb = (uintptr_t)mb_local.b_next;
1018 	}
1019 	return (mb);
1020 }
1021 
1022 int
mdb_name_to_major(const char * name,major_t * major)1023 mdb_name_to_major(const char *name, major_t *major)
1024 {
1025 	uintptr_t mbind;
1026 	uintptr_t mb_hashtab[MOD_BIND_HASHSIZE];
1027 	struct bind mbind_local;
1028 
1029 
1030 	if (mdb_readsym(mb_hashtab, sizeof (mb_hashtab), "mb_hashtab") == -1) {
1031 		mdb_warn("failed to read symbol 'mb_hashtab'");
1032 		return (-1);
1033 	}
1034 
1035 	if ((mbind = find_mbind(name, mb_hashtab)) != 0) {
1036 		if (mdb_vread(&mbind_local, sizeof (mbind_local), mbind) ==
1037 		    -1) {
1038 			mdb_warn("failed to read mbind struct at %p", mbind);
1039 			return (-1);
1040 		}
1041 
1042 		*major = (major_t)mbind_local.b_num;
1043 		return (0);
1044 	}
1045 	return (-1);
1046 }
1047 
1048 const char *
mdb_major_to_name(major_t major)1049 mdb_major_to_name(major_t major)
1050 {
1051 	static char name[MODMAXNAMELEN];
1052 
1053 	uintptr_t devnamesp;
1054 	struct devnames dn;
1055 	uint_t devcnt;
1056 
1057 	if (mdb_readvar(&devcnt, "devcnt") == -1 || major >= devcnt ||
1058 	    mdb_readvar(&devnamesp, "devnamesp") == -1)
1059 		return (NULL);
1060 
1061 	if (mdb_vread(&dn, sizeof (struct devnames), devnamesp +
1062 	    major * sizeof (struct devnames)) != sizeof (struct devnames))
1063 		return (NULL);
1064 
1065 	if (mdb_readstr(name, sizeof (name), (uintptr_t)dn.dn_name) == -1)
1066 		return (NULL);
1067 
1068 	return ((const char *)name);
1069 }
1070 
1071 /*
1072  * Return the name of the driver attached to the dip in drivername.
1073  */
1074 int
mdb_devinfo2driver(uintptr_t dip_addr,char * drivername,size_t namebufsize)1075 mdb_devinfo2driver(uintptr_t dip_addr, char *drivername, size_t namebufsize)
1076 {
1077 	struct dev_info	devinfo;
1078 	char bind_name[MAXPATHLEN + 1];
1079 	major_t	major;
1080 	const char *namestr;
1081 
1082 
1083 	if (mdb_vread(&devinfo, sizeof (devinfo), dip_addr) == -1) {
1084 		mdb_warn("failed to read devinfo at %p", dip_addr);
1085 		return (-1);
1086 	}
1087 
1088 	if (mdb_readstr(bind_name, sizeof (bind_name),
1089 	    (uintptr_t)devinfo.devi_binding_name) == -1) {
1090 		mdb_warn("failed to read binding name at %p",
1091 		    devinfo.devi_binding_name);
1092 		return (-1);
1093 	}
1094 
1095 	/*
1096 	 * Many->one relation: various names to one major number
1097 	 */
1098 	if (mdb_name_to_major(bind_name, &major) == -1) {
1099 		mdb_warn("failed to translate bind name to major number\n");
1100 		return (-1);
1101 	}
1102 
1103 	/*
1104 	 * One->one relation: one major number corresponds to one driver
1105 	 */
1106 	if ((namestr = mdb_major_to_name(major)) == NULL) {
1107 		(void) strncpy(drivername, "???", namebufsize);
1108 		return (-1);
1109 	}
1110 
1111 	(void) strncpy(drivername, namestr, namebufsize);
1112 	return (0);
1113 }
1114 
1115 /*
1116  * Find the name of the driver attached to this dip (if any), given:
1117  * - the address of a dip (in core)
1118  * - the NAME of the global pointer to the driver's i_ddi_soft_state struct
1119  * - pointer to a pointer to receive the address
1120  */
1121 int
mdb_devinfo2statep(uintptr_t dip_addr,char * soft_statep_name,uintptr_t * statep)1122 mdb_devinfo2statep(uintptr_t dip_addr, char *soft_statep_name,
1123     uintptr_t *statep)
1124 {
1125 	struct dev_info	dev_info;
1126 
1127 
1128 	if (mdb_vread(&dev_info, sizeof (dev_info), dip_addr) == -1) {
1129 		mdb_warn("failed to read devinfo at %p", dip_addr);
1130 		return (-1);
1131 	}
1132 
1133 	return (mdb_get_soft_state_byname(soft_statep_name,
1134 	    dev_info.devi_instance, statep, NULL, 0));
1135 }
1136 
1137 /*
1138  * Returns a pointer to the top of the soft state struct for the instance
1139  * specified (in state_addr), given the address of the global soft state
1140  * pointer and size of the struct.  Also fills in the buffer pointed to by
1141  * state_buf_p (if non-NULL) with the contents of the state struct.
1142  */
1143 int
mdb_get_soft_state_byaddr(uintptr_t ssaddr,uint_t instance,uintptr_t * state_addr,void * state_buf_p,size_t sizeof_state)1144 mdb_get_soft_state_byaddr(uintptr_t ssaddr, uint_t instance,
1145     uintptr_t *state_addr, void *state_buf_p, size_t sizeof_state)
1146 {
1147 	struct i_ddi_soft_state ss;
1148 	void *statep;
1149 
1150 
1151 	if (mdb_vread(&ss, sizeof (ss), ssaddr) == -1)
1152 		return (-1);
1153 
1154 	if (instance >= ss.n_items)
1155 		return (-1);
1156 
1157 	if (mdb_vread(&statep, sizeof (statep), (uintptr_t)ss.array +
1158 	    (sizeof (statep) * instance)) == -1)
1159 		return (-1);
1160 
1161 	if (state_addr != NULL)
1162 		*state_addr = (uintptr_t)statep;
1163 
1164 	if (statep == NULL) {
1165 		errno = ENOENT;
1166 		return (-1);
1167 	}
1168 
1169 	if (state_buf_p != NULL) {
1170 
1171 		/* Read the state struct into the buffer in local space. */
1172 		if (mdb_vread(state_buf_p, sizeof_state,
1173 		    (uintptr_t)statep) == -1)
1174 			return (-1);
1175 	}
1176 
1177 	return (0);
1178 }
1179 
1180 
1181 /*
1182  * Returns a pointer to the top of the soft state struct for the instance
1183  * specified (in state_addr), given the name of the global soft state pointer
1184  * and size of the struct.  Also fills in the buffer pointed to by
1185  * state_buf_p (if non-NULL) with the contents of the state struct.
1186  */
1187 int
mdb_get_soft_state_byname(char * softstatep_name,uint_t instance,uintptr_t * state_addr,void * state_buf_p,size_t sizeof_state)1188 mdb_get_soft_state_byname(char *softstatep_name, uint_t instance,
1189     uintptr_t *state_addr, void *state_buf_p, size_t sizeof_state)
1190 {
1191 	uintptr_t ssaddr;
1192 
1193 	if (mdb_readvar((void *)&ssaddr, softstatep_name) == -1)
1194 		return (-1);
1195 
1196 	return (mdb_get_soft_state_byaddr(ssaddr, instance, state_addr,
1197 	    state_buf_p, sizeof_state));
1198 }
1199 
1200 static void
tdelta_help(void)1201 tdelta_help(void)
1202 {
1203 	mdb_printf(
1204 "Prints the difference between two nanosecond (hrtime_t) timestamps.\n\n"
1205 "The time used for comparison depends on the context in which mdb is invoked:\n"
1206 "	Live system	The current time (gethrtime()).\n"
1207 "	Crash dump	The time the dump started writing to the dump device.\n"
1208 "	KMDB		The most recent time the system entered KMDB.\n"
1209 "\n"
1210 "Note that with crash dumps, the dump time is often a few milliseconds\n"
1211 "after the time of the panic (panic_hrtime).\n"
1212 "\n"
1213 "The following options are supported:\n"
1214 "	-h	Print the delta in human readable form.\n"
1215 "	-t %<u>ts%</u>	Use %<u>ts%</u> as the comparison timestamp.\n");
1216 }
1217 
1218 static int
tdelta(uintptr_t addr,uint_t flags,int argc,const mdb_arg_t * argv)1219 tdelta(uintptr_t addr, uint_t flags, int argc, const mdb_arg_t *argv)
1220 {
1221 	int64_t	val = 0;
1222 	int64_t	delta = 0;
1223 	int	do_nice = FALSE;
1224 
1225 	if (!(flags & DCMD_ADDRSPEC)) {
1226 		mdb_warn("dcmd requires an input hrtime_t");
1227 		return (DCMD_ERR);
1228 	}
1229 
1230 	val = (int64_t)mdb_gethrtime();
1231 
1232 	if (mdb_getopts(argc, argv,
1233 	    'h', MDB_OPT_SETBITS, TRUE, &do_nice,
1234 	    't', MDB_OPT_UINT64, &val,
1235 	    NULL) != argc) {
1236 		return (DCMD_USAGE);
1237 	}
1238 
1239 	delta = (int64_t)addr - val;
1240 
1241 	if (do_nice) {
1242 		char buf[32] = { 0 };
1243 
1244 		mdb_nicetime(delta, buf, sizeof (buf));
1245 		mdb_printf("%s\n", buf);
1246 	} else {
1247 		mdb_printf("%ll#R\n", delta);
1248 	}
1249 
1250 	return (DCMD_OK);
1251 }
1252 
1253 static const mdb_dcmd_t dcmds[] = {
1254 	{ "dnlc", NULL, "print DNLC contents", dnlcdump },
1255 	{
1256 	    "tdelta", "?[-h][-t ts]", "compute ns time delta", tdelta,
1257 	    tdelta_help },
1258 	{ NULL }
1259 };
1260 
1261 static const mdb_modinfo_t modinfo = { MDB_API_VERSION, dcmds };
1262 
1263 /*ARGSUSED*/
1264 static void
update_vars(void * arg)1265 update_vars(void *arg)
1266 {
1267 	GElf_Sym sym;
1268 
1269 	if (mdb_lookup_by_name("auto_vnodeops", &sym) == 0)
1270 		autofs_vnops_ptr = (struct vnodeops *)(uintptr_t)sym.st_value;
1271 	else
1272 		autofs_vnops_ptr = NULL;
1273 
1274 	(void) mdb_readvar(&_mdb_ks_pagesize, "_pagesize");
1275 	(void) mdb_readvar(&_mdb_ks_pageshift, "_pageshift");
1276 	(void) mdb_readvar(&_mdb_ks_pageoffset, "_pageoffset");
1277 	(void) mdb_readvar(&_mdb_ks_pagemask, "_pagemask");
1278 	(void) mdb_readvar(&_mdb_ks_mmu_pagesize, "_mmu_pagesize");
1279 	(void) mdb_readvar(&_mdb_ks_mmu_pageshift, "_mmu_pageshift");
1280 	(void) mdb_readvar(&_mdb_ks_mmu_pageoffset, "_mmu_pageoffset");
1281 	(void) mdb_readvar(&_mdb_ks_mmu_pagemask, "_mmu_pagemask");
1282 	(void) mdb_readvar(&_mdb_ks_kernelbase, "_kernelbase");
1283 
1284 	(void) mdb_readvar(&_mdb_ks_userlimit, "_userlimit");
1285 	(void) mdb_readvar(&_mdb_ks_userlimit32, "_userlimit32");
1286 	(void) mdb_readvar(&_mdb_ks_msg_bsize, "_msg_bsize");
1287 	(void) mdb_readvar(&_mdb_ks_defaultstksz, "_defaultstksz");
1288 	(void) mdb_readvar(&_mdb_ks_ncpu, "_ncpu");
1289 	(void) mdb_readvar(&_mdb_ks_ncpu_log2, "_ncpu_log2");
1290 	(void) mdb_readvar(&_mdb_ks_ncpu_p2, "_ncpu_p2");
1291 
1292 	page_hash_loaded = 0;	/* invalidate cached page_hash state */
1293 }
1294 
1295 const mdb_modinfo_t *
_mdb_init(void)1296 _mdb_init(void)
1297 {
1298 	/*
1299 	 * When used with mdb, mdb_ks is a separate dmod.  With kmdb, however,
1300 	 * mdb_ks is compiled into the debugger module.  kmdb cannot
1301 	 * automatically modunload itself when it exits.  If it restarts after
1302 	 * debugger fault, static variables may not be initialized to zero.
1303 	 * They must be manually reinitialized here.
1304 	 */
1305 	dnlc_hash = NULL;
1306 	qi_head = NULL;
1307 
1308 	mdb_callback_add(MDB_CALLBACK_STCHG, update_vars, NULL);
1309 
1310 	update_vars(NULL);
1311 
1312 	return (&modinfo);
1313 }
1314 
1315 void
_mdb_fini(void)1316 _mdb_fini(void)
1317 {
1318 	dnlc_free();
1319 	while (qi_head != NULL) {
1320 		mdb_qinfo_t *qip = qi_head;
1321 		qi_head = qip->qi_next;
1322 		mdb_free(qip, sizeof (mdb_qinfo_t));
1323 	}
1324 }
1325 
1326 /*
1327  * Interface between MDB kproc target and mdb_ks.  The kproc target relies
1328  * on looking up and invoking these functions in mdb_ks so that dependencies
1329  * on the current kernel implementation are isolated in mdb_ks.
1330  */
1331 
1332 /*
1333  * Given the address of a proc_t, return the p.p_as pointer; return NULL
1334  * if we were unable to read a proc structure from the given address.
1335  */
1336 uintptr_t
mdb_kproc_as(uintptr_t proc_addr)1337 mdb_kproc_as(uintptr_t proc_addr)
1338 {
1339 	proc_t p;
1340 
1341 	if (mdb_vread(&p, sizeof (p), proc_addr) == sizeof (p))
1342 		return ((uintptr_t)p.p_as);
1343 
1344 	return (0);
1345 }
1346 
1347 /*
1348  * Given the address of a proc_t, return the p.p_model value; return
1349  * PR_MODEL_UNKNOWN if we were unable to read a proc structure or if
1350  * the model value does not match one of the two known values.
1351  */
1352 uint_t
mdb_kproc_model(uintptr_t proc_addr)1353 mdb_kproc_model(uintptr_t proc_addr)
1354 {
1355 	proc_t p;
1356 
1357 	if (mdb_vread(&p, sizeof (p), proc_addr) == sizeof (p)) {
1358 		switch (p.p_model) {
1359 		case DATAMODEL_ILP32:
1360 			return (PR_MODEL_ILP32);
1361 		case DATAMODEL_LP64:
1362 			return (PR_MODEL_LP64);
1363 		}
1364 	}
1365 
1366 	return (PR_MODEL_UNKNOWN);
1367 }
1368 
1369 /*
1370  * Callback function for walking process's segment list.  For each segment,
1371  * we fill in an mdb_map_t describing its properties, and then invoke
1372  * the callback function provided by the kproc target.
1373  */
1374 static int
asmap_step(uintptr_t addr,const struct seg * seg,asmap_arg_t * asmp)1375 asmap_step(uintptr_t addr, const struct seg *seg, asmap_arg_t *asmp)
1376 {
1377 	struct segvn_data svd;
1378 	mdb_map_t map;
1379 
1380 	if (seg->s_ops == asmp->asm_segvn_ops && mdb_vread(&svd,
1381 	    sizeof (svd), (uintptr_t)seg->s_data) == sizeof (svd)) {
1382 
1383 		if (svd.vp != NULL) {
1384 			if (mdb_vnode2path((uintptr_t)svd.vp, map.map_name,
1385 			    MDB_TGT_MAPSZ) != 0) {
1386 				(void) mdb_snprintf(map.map_name,
1387 				    MDB_TGT_MAPSZ, "[ vnode %p ]", svd.vp);
1388 			}
1389 		} else
1390 			(void) strcpy(map.map_name, "[ anon ]");
1391 
1392 	} else {
1393 		(void) mdb_snprintf(map.map_name, MDB_TGT_MAPSZ,
1394 		    "[ seg %p ]", addr);
1395 	}
1396 
1397 	map.map_base = (uintptr_t)seg->s_base;
1398 	map.map_size = seg->s_size;
1399 	map.map_flags = 0;
1400 
1401 	asmp->asm_callback((const struct mdb_map *)&map, asmp->asm_cbdata);
1402 	return (WALK_NEXT);
1403 }
1404 
1405 /*
1406  * Given a process address space, walk its segment list using the seg walker,
1407  * convert the segment data to an mdb_map_t, and pass this information
1408  * back to the kproc target via the given callback function.
1409  */
1410 int
mdb_kproc_asiter(uintptr_t as,void (* func)(const struct mdb_map *,void *),void * p)1411 mdb_kproc_asiter(uintptr_t as,
1412     void (*func)(const struct mdb_map *, void *), void *p)
1413 {
1414 	asmap_arg_t arg;
1415 	GElf_Sym sym;
1416 
1417 	arg.asm_segvn_ops = NULL;
1418 	arg.asm_callback = func;
1419 	arg.asm_cbdata = p;
1420 
1421 	if (mdb_lookup_by_name("segvn_ops", &sym) == 0)
1422 		arg.asm_segvn_ops = (struct seg_ops *)(uintptr_t)sym.st_value;
1423 
1424 	return (mdb_pwalk("seg", (mdb_walk_cb_t)asmap_step, &arg, as));
1425 }
1426 
1427 /*
1428  * Copy the auxv array from the given process's u-area into the provided
1429  * buffer.  If the buffer is NULL, only return the size of the auxv array
1430  * so the caller knows how much space will be required.
1431  */
1432 int
mdb_kproc_auxv(uintptr_t proc,auxv_t * auxv)1433 mdb_kproc_auxv(uintptr_t proc, auxv_t *auxv)
1434 {
1435 	if (auxv != NULL) {
1436 		proc_t p;
1437 
1438 		if (mdb_vread(&p, sizeof (p), proc) != sizeof (p))
1439 			return (-1);
1440 
1441 		bcopy(p.p_user.u_auxv, auxv,
1442 		    sizeof (auxv_t) * __KERN_NAUXV_IMPL);
1443 	}
1444 
1445 	return (__KERN_NAUXV_IMPL);
1446 }
1447 
1448 /*
1449  * Given a process address, return the PID.
1450  */
1451 pid_t
mdb_kproc_pid(uintptr_t proc_addr)1452 mdb_kproc_pid(uintptr_t proc_addr)
1453 {
1454 	struct pid pid;
1455 	proc_t p;
1456 
1457 	if (mdb_vread(&p, sizeof (p), proc_addr) == sizeof (p) &&
1458 	    mdb_vread(&pid, sizeof (pid), (uintptr_t)p.p_pidp) == sizeof (pid))
1459 		return (pid.pid_id);
1460 
1461 	return (-1);
1462 }
1463 
1464 /*
1465  * Interface between the MDB kvm target and mdb_ks.  The kvm target relies
1466  * on looking up and invoking these functions in mdb_ks so that dependencies
1467  * on the current kernel implementation are isolated in mdb_ks.
1468  */
1469 
1470 /*
1471  * Determine whether or not the thread that panicked the given kernel was a
1472  * kernel thread (panic_thread->t_procp == &p0).
1473  */
1474 void
mdb_dump_print_content(dumphdr_t * dh,pid_t content)1475 mdb_dump_print_content(dumphdr_t *dh, pid_t content)
1476 {
1477 	GElf_Sym sym;
1478 	uintptr_t pt;
1479 	uintptr_t procp;
1480 	int expcont = 0;
1481 	int actcont;
1482 
1483 	(void) mdb_readvar(&expcont, "dump_conflags");
1484 	actcont = dh->dump_flags & DF_CONTENT;
1485 
1486 	if (actcont == DF_ALL) {
1487 		mdb_printf("dump content: all kernel and user pages\n");
1488 		return;
1489 	} else if (actcont == DF_CURPROC) {
1490 		mdb_printf("dump content: kernel pages and pages from "
1491 		    "PID %d", content);
1492 		return;
1493 	}
1494 
1495 	mdb_printf("dump content: kernel pages only\n");
1496 	if (!(expcont & DF_CURPROC))
1497 		return;
1498 
1499 	if (mdb_readvar(&pt, "panic_thread") != sizeof (pt) || pt == 0)
1500 		goto kthreadpanic_err;
1501 
1502 	if (mdb_vread(&procp, sizeof (procp), pt + OFFSETOF(kthread_t,
1503 	    t_procp)) == -1 || procp == 0)
1504 		goto kthreadpanic_err;
1505 
1506 	if (mdb_lookup_by_name("p0", &sym) != 0)
1507 		goto kthreadpanic_err;
1508 
1509 	if (procp == (uintptr_t)sym.st_value) {
1510 		mdb_printf("  (curproc requested, but a kernel thread "
1511 		    "panicked)\n");
1512 	} else {
1513 		mdb_printf("  (curproc requested, but the process that "
1514 		    "panicked could not be dumped)\n");
1515 	}
1516 
1517 	return;
1518 
1519 kthreadpanic_err:
1520 	mdb_printf("  (curproc requested, but the process that panicked could "
1521 	    "not be found)\n");
1522 }
1523 
1524 /*
1525  * Determine the process that was saved in a `curproc' dump.  This process will
1526  * be recorded as the first element in dump_pids[].
1527  */
1528 int
mdb_dump_find_curproc(void)1529 mdb_dump_find_curproc(void)
1530 {
1531 	uintptr_t pidp;
1532 	pid_t pid = -1;
1533 
1534 	if (mdb_readvar(&pidp, "dump_pids") == sizeof (pidp) &&
1535 	    mdb_vread(&pid, sizeof (pid), pidp) == sizeof (pid) &&
1536 	    pid > 0)
1537 		return (pid);
1538 	else
1539 		return (-1);
1540 }
1541 
1542 
1543 /*
1544  * Following three funcs extracted from sunddi.c
1545  */
1546 
1547 /*
1548  * Return core address of root node of devinfo tree
1549  */
1550 static uintptr_t
mdb_ddi_root_node(void)1551 mdb_ddi_root_node(void)
1552 {
1553 	uintptr_t	top_devinfo_addr;
1554 
1555 	/* return (top_devinfo);   */
1556 	if (mdb_readvar(&top_devinfo_addr, "top_devinfo") == -1) {
1557 		mdb_warn("failed to read top_devinfo");
1558 		return (0);
1559 	}
1560 	return (top_devinfo_addr);
1561 }
1562 
1563 /*
1564  * Return the name of the devinfo node pointed at by 'dip_addr' in the buffer
1565  * pointed at by 'name.'
1566  *
1567  * - dip_addr is a pointer to a dev_info struct in core.
1568  */
1569 static char *
mdb_ddi_deviname(uintptr_t dip_addr,char * name,size_t name_size)1570 mdb_ddi_deviname(uintptr_t dip_addr, char *name, size_t name_size)
1571 {
1572 	uintptr_t addrname;
1573 	ssize_t	length;
1574 	char *local_namep = name;
1575 	size_t local_name_size = name_size;
1576 	struct dev_info	local_dip;
1577 
1578 
1579 	if (dip_addr == mdb_ddi_root_node()) {
1580 		if (name_size < 1) {
1581 			mdb_warn("failed to get node name: buf too small\n");
1582 			return (NULL);
1583 		}
1584 
1585 		*name = '\0';
1586 		return (name);
1587 	}
1588 
1589 	if (name_size < 2) {
1590 		mdb_warn("failed to get node name: buf too small\n");
1591 		return (NULL);
1592 	}
1593 
1594 	local_namep = name;
1595 	*local_namep++ = '/';
1596 	*local_namep = '\0';
1597 	local_name_size--;
1598 
1599 	if (mdb_vread(&local_dip, sizeof (struct dev_info), dip_addr) == -1) {
1600 		mdb_warn("failed to read devinfo struct");
1601 	}
1602 
1603 	length = mdb_readstr(local_namep, local_name_size,
1604 	    (uintptr_t)local_dip.devi_node_name);
1605 	if (length == -1) {
1606 		mdb_warn("failed to read node name");
1607 		return (NULL);
1608 	}
1609 	local_namep += length;
1610 	local_name_size -= length;
1611 	addrname = (uintptr_t)local_dip.devi_addr;
1612 
1613 	if (addrname != 0) {
1614 
1615 		if (local_name_size < 2) {
1616 			mdb_warn("not enough room for node address string");
1617 			return (name);
1618 		}
1619 		*local_namep++ = '@';
1620 		*local_namep = '\0';
1621 		local_name_size--;
1622 
1623 		length = mdb_readstr(local_namep, local_name_size, addrname);
1624 		if (length == -1) {
1625 			mdb_warn("failed to read name");
1626 			return (NULL);
1627 		}
1628 	}
1629 
1630 	return (name);
1631 }
1632 
1633 /*
1634  * Generate the full path under the /devices dir to the device entry.
1635  *
1636  * dip is a pointer to a devinfo struct in core (not in local memory).
1637  */
1638 char *
mdb_ddi_pathname(uintptr_t dip_addr,char * path,size_t pathlen)1639 mdb_ddi_pathname(uintptr_t dip_addr, char *path, size_t pathlen)
1640 {
1641 	struct dev_info local_dip;
1642 	uintptr_t	parent_dip;
1643 	char		*bp;
1644 	size_t		buf_left;
1645 
1646 
1647 	if (dip_addr == mdb_ddi_root_node()) {
1648 		*path = '\0';
1649 		return (path);
1650 	}
1651 
1652 
1653 	if (mdb_vread(&local_dip, sizeof (struct dev_info), dip_addr) == -1) {
1654 		mdb_warn("failed to read devinfo struct");
1655 	}
1656 
1657 	parent_dip = (uintptr_t)local_dip.devi_parent;
1658 	(void) mdb_ddi_pathname(parent_dip, path, pathlen);
1659 
1660 	bp = path + strlen(path);
1661 	buf_left = pathlen - strlen(path);
1662 	(void) mdb_ddi_deviname(dip_addr, bp, buf_left);
1663 	return (path);
1664 }
1665 
1666 
1667 /*
1668  * Read in the string value of a refstr, which is appended to the end of
1669  * the structure.
1670  */
1671 ssize_t
mdb_read_refstr(uintptr_t refstr_addr,char * str,size_t nbytes)1672 mdb_read_refstr(uintptr_t refstr_addr, char *str, size_t nbytes)
1673 {
1674 	struct refstr *r = (struct refstr *)refstr_addr;
1675 
1676 	return (mdb_readstr(str, nbytes, (uintptr_t)r->rs_string));
1677 }
1678 
1679 /*
1680  * Chase an mblk list by b_next and return the length.
1681  */
1682 int
mdb_mblk_count(const mblk_t * mb)1683 mdb_mblk_count(const mblk_t *mb)
1684 {
1685 	int count;
1686 	mblk_t mblk;
1687 
1688 	if (mb == NULL)
1689 		return (0);
1690 
1691 	count = 1;
1692 	while (mb->b_next != NULL) {
1693 		count++;
1694 		if (mdb_vread(&mblk, sizeof (mblk), (uintptr_t)mb->b_next) ==
1695 		    -1)
1696 			break;
1697 		mb = &mblk;
1698 	}
1699 	return (count);
1700 }
1701 
1702 /*
1703  * Write the given MAC address as a printable string in the usual colon-
1704  * separated format.  Assumes that buflen is at least 2.
1705  */
1706 void
mdb_mac_addr(const uint8_t * addr,size_t alen,char * buf,size_t buflen)1707 mdb_mac_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
1708 {
1709 	int slen;
1710 
1711 	if (alen == 0 || buflen < 4) {
1712 		(void) strcpy(buf, "?");
1713 		return;
1714 	}
1715 	for (;;) {
1716 		/*
1717 		 * If there are more MAC address bytes available, but we won't
1718 		 * have any room to print them, then add "..." to the string
1719 		 * instead.  See below for the 'magic number' explanation.
1720 		 */
1721 		if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
1722 			(void) strcpy(buf, "...");
1723 			break;
1724 		}
1725 		slen = mdb_snprintf(buf, buflen, "%02x", *addr++);
1726 		buf += slen;
1727 		if (--alen == 0)
1728 			break;
1729 		*buf++ = ':';
1730 		buflen -= slen + 1;
1731 		/*
1732 		 * At this point, based on the first 'if' statement above,
1733 		 * either alen == 1 and buflen >= 3, or alen > 1 and
1734 		 * buflen >= 4.  The first case leaves room for the final "xx"
1735 		 * number and trailing NUL byte.  The second leaves room for at
1736 		 * least "...".  Thus the apparently 'magic' numbers chosen for
1737 		 * that statement.
1738 		 */
1739 	}
1740 }
1741 
1742 /*
1743  * Produce a string that represents a DLPI primitive, or NULL if no such string
1744  * is possible.
1745  */
1746 const char *
mdb_dlpi_prim(int prim)1747 mdb_dlpi_prim(int prim)
1748 {
1749 	switch (prim) {
1750 	case DL_INFO_REQ:	return ("DL_INFO_REQ");
1751 	case DL_INFO_ACK:	return ("DL_INFO_ACK");
1752 	case DL_ATTACH_REQ:	return ("DL_ATTACH_REQ");
1753 	case DL_DETACH_REQ:	return ("DL_DETACH_REQ");
1754 	case DL_BIND_REQ:	return ("DL_BIND_REQ");
1755 	case DL_BIND_ACK:	return ("DL_BIND_ACK");
1756 	case DL_UNBIND_REQ:	return ("DL_UNBIND_REQ");
1757 	case DL_OK_ACK:		return ("DL_OK_ACK");
1758 	case DL_ERROR_ACK:	return ("DL_ERROR_ACK");
1759 	case DL_ENABMULTI_REQ:	return ("DL_ENABMULTI_REQ");
1760 	case DL_DISABMULTI_REQ:	return ("DL_DISABMULTI_REQ");
1761 	case DL_PROMISCON_REQ:	return ("DL_PROMISCON_REQ");
1762 	case DL_PROMISCOFF_REQ:	return ("DL_PROMISCOFF_REQ");
1763 	case DL_UNITDATA_REQ:	return ("DL_UNITDATA_REQ");
1764 	case DL_UNITDATA_IND:	return ("DL_UNITDATA_IND");
1765 	case DL_UDERROR_IND:	return ("DL_UDERROR_IND");
1766 	case DL_PHYS_ADDR_REQ:	return ("DL_PHYS_ADDR_REQ");
1767 	case DL_PHYS_ADDR_ACK:	return ("DL_PHYS_ADDR_ACK");
1768 	case DL_SET_PHYS_ADDR_REQ:	return ("DL_SET_PHYS_ADDR_REQ");
1769 	case DL_NOTIFY_REQ:	return ("DL_NOTIFY_REQ");
1770 	case DL_NOTIFY_ACK:	return ("DL_NOTIFY_ACK");
1771 	case DL_NOTIFY_IND:	return ("DL_NOTIFY_IND");
1772 	case DL_NOTIFY_CONF:	return ("DL_NOTIFY_CONF");
1773 	case DL_CAPABILITY_REQ:	return ("DL_CAPABILITY_REQ");
1774 	case DL_CAPABILITY_ACK:	return ("DL_CAPABILITY_ACK");
1775 	case DL_CONTROL_REQ:	return ("DL_CONTROL_REQ");
1776 	case DL_CONTROL_ACK:	return ("DL_CONTROL_ACK");
1777 	case DL_PASSIVE_REQ:	return ("DL_PASSIVE_REQ");
1778 	default:		return (NULL);
1779 	}
1780 }
1781 
1782 /*
1783  * mdb_gethrtime() returns the hires system time. This will be the timestamp at
1784  * which we dropped into, if called from, kmdb(1); the core dump's hires time
1785  * if inspecting one; or the running system's hires time if we're inspecting
1786  * a live kernel.
1787  */
1788 hrtime_t
mdb_gethrtime(void)1789 mdb_gethrtime(void)
1790 {
1791 	uintptr_t ptr;
1792 	GElf_Sym sym;
1793 	lbolt_info_t lbi;
1794 	hrtime_t ts;
1795 
1796 	/*
1797 	 * We first check whether the lbolt info structure has been allocated
1798 	 * and initialized. If not, lbolt_hybrid will be pointing at
1799 	 * lbolt_bootstrap.
1800 	 */
1801 	if (mdb_lookup_by_name("lbolt_bootstrap", &sym) == -1)
1802 		return (0);
1803 
1804 	if (mdb_readvar(&ptr, "lbolt_hybrid") == -1)
1805 		return (0);
1806 
1807 	if (ptr == (uintptr_t)sym.st_value)
1808 		return (0);
1809 
1810 #ifdef _KMDB
1811 	if (mdb_readvar(&ptr, "lb_info") == -1)
1812 		return (0);
1813 
1814 	if (mdb_vread(&lbi, sizeof (lbolt_info_t), ptr) !=
1815 	    sizeof (lbolt_info_t))
1816 		return (0);
1817 
1818 	ts = lbi.lbi_debug_ts;
1819 #else
1820 	if (mdb_prop_postmortem) {
1821 		if (mdb_readvar(&ptr, "lb_info") == -1)
1822 			return (0);
1823 
1824 		if (mdb_vread(&lbi, sizeof (lbolt_info_t), ptr) !=
1825 		    sizeof (lbolt_info_t))
1826 			return (0);
1827 
1828 		ts = lbi.lbi_debug_ts;
1829 	} else {
1830 		ts = gethrtime();
1831 	}
1832 #endif
1833 	return (ts);
1834 }
1835 
1836 /*
1837  * mdb_get_lbolt() returns the number of clock ticks since system boot.
1838  * Depending on the context in which it's called, the value will be derived
1839  * from different sources per mdb_gethrtime(). If inspecting a panicked
1840  * system, the routine returns the 'panic_lbolt64' variable from the core file.
1841  */
1842 int64_t
mdb_get_lbolt(void)1843 mdb_get_lbolt(void)
1844 {
1845 	lbolt_info_t lbi;
1846 	uintptr_t ptr;
1847 	int64_t pl;
1848 	hrtime_t ts;
1849 	int nsec;
1850 
1851 	if (mdb_readvar(&pl, "panic_lbolt64") != -1 && pl > 0)
1852 		return (pl);
1853 
1854 	/*
1855 	 * mdb_gethrtime() will return zero if the lbolt info structure hasn't
1856 	 * been allocated and initialized yet, or if it fails to read it.
1857 	 */
1858 	if ((ts = mdb_gethrtime()) <= 0)
1859 		return (0);
1860 
1861 	/*
1862 	 * Load the time spent in kmdb, if any.
1863 	 */
1864 	if (mdb_readvar(&ptr, "lb_info") == -1)
1865 		return (0);
1866 
1867 	if (mdb_vread(&lbi, sizeof (lbolt_info_t), ptr) !=
1868 	    sizeof (lbolt_info_t))
1869 		return (0);
1870 
1871 	if (mdb_readvar(&nsec, "nsec_per_tick") == -1 || nsec == 0) {
1872 		mdb_warn("failed to read 'nsec_per_tick'");
1873 		return (-1);
1874 	}
1875 
1876 	return ((ts/nsec) - lbi.lbi_debug_time);
1877 }
1878 
1879 void
mdb_print_buildversion(void)1880 mdb_print_buildversion(void)
1881 {
1882 	GElf_Sym sym;
1883 
1884 	if (mdb_lookup_by_name("buildversion", &sym) != 0)
1885 		return;
1886 
1887 	char *str = mdb_zalloc(4096, UM_SLEEP | UM_GC);
1888 
1889 	if (mdb_readstr(str, 4096, sym.st_value) < 1)
1890 		return;
1891 
1892 	mdb_printf("build version: %s\n", str);
1893 }
1894