xref: /openbsd/lib/libkvm/kvm_proc.c (revision 6e552b66)
1 /*	$OpenBSD: kvm_proc.c,v 1.64 2024/05/10 06:46:14 asou Exp $	*/
2 /*	$NetBSD: kvm_proc.c,v 1.30 1999/03/24 05:50:50 mrg Exp $	*/
3 /*-
4  * Copyright (c) 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Charles M. Hannum.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 /*-
32  * Copyright (c) 1994, 1995 Charles M. Hannum.  All rights reserved.
33  * Copyright (c) 1989, 1992, 1993
34  *	The Regents of the University of California.  All rights reserved.
35  *
36  * This code is derived from software developed by the Computer Systems
37  * Engineering group at Lawrence Berkeley Laboratory under DARPA contract
38  * BG 91-66 and contributed to Berkeley.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. Neither the name of the University nor the names of its contributors
49  *    may be used to endorse or promote products derived from this software
50  *    without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  */
64 
65 /*
66  * Proc traversal interface for kvm.  ps and w are (probably) the exclusive
67  * users of this code, so we've factored it out into a separate module.
68  * Thus, we keep this grunge out of the other kvm applications (i.e.,
69  * most other applications are interested only in open/close/read/nlist).
70  */
71 
72 #define __need_process
73 #include <sys/param.h>	/* VM_MIN_ADDRESS PAGE_SIZE */
74 #include <sys/types.h>
75 #include <sys/signal.h>
76 #include <sys/proc.h>
77 #include <sys/exec.h>
78 #include <sys/stat.h>
79 #include <sys/ioctl.h>
80 #include <sys/tty.h>
81 #include <stddef.h>
82 #include <stdlib.h>
83 #include <string.h>
84 #include <unistd.h>
85 #include <nlist.h>
86 #include <kvm.h>
87 #include <errno.h>
88 
89 #include <uvm/uvm_extern.h>
90 #include <uvm/uvm_amap.h>
91 #include <machine/vmparam.h>
92 #include <machine/pmap.h>
93 
94 #include <sys/sysctl.h>
95 
96 #include <limits.h>
97 #include <db.h>
98 #include <paths.h>
99 
100 #include "kvm_private.h"
101 
102 #define MINIMUM(a, b)	(((a) < (b)) ? (a) : (b))
103 #define MAXIMUM(a, b)	(((a) > (b)) ? (a) : (b))
104 
105 static char	*_kvm_ureadm(kvm_t *, const struct kinfo_proc *, u_long, u_long *);
106 static ssize_t	kvm_ureadm(kvm_t *, const struct kinfo_proc *, u_long, char *, size_t);
107 
108 static char	**kvm_argv(kvm_t *, const struct kinfo_proc *, u_long, int, int, int);
109 
110 static char	**kvm_doargv(kvm_t *, const struct kinfo_proc *, int, int,
111 		    void (*)(struct ps_strings *, u_long *, int *));
112 static int	proc_verify(kvm_t *, const struct kinfo_proc *);
113 static void	ps_str_a(struct ps_strings *, u_long *, int *);
114 static void	ps_str_e(struct ps_strings *, u_long *, int *);
115 
116 static struct vm_anon *
_kvm_findanon(kvm_t * kd,struct vm_amap * amapp,int slot)117 _kvm_findanon(kvm_t *kd, struct vm_amap *amapp, int slot)
118 {
119 	u_long addr;
120 	int bucket;
121 	struct vm_amap amap;
122 	struct vm_amap_chunk chunk, *chunkp;
123 	struct vm_anon *anonp;
124 
125 	addr = (u_long)amapp;
126 	if (KREAD(kd, addr, &amap))
127 		return (NULL);
128 
129 	/* sanity-check slot number */
130 	if (slot > amap.am_nslot)
131 		return (NULL);
132 
133 	if (UVM_AMAP_SMALL(&amap))
134 		chunkp = &amapp->am_small;
135 	else {
136 		bucket = UVM_AMAP_BUCKET(&amap, slot);
137 		addr = (u_long)(amap.am_buckets + bucket);
138 		if (KREAD(kd, addr, &chunkp))
139 			return (NULL);
140 
141 		while (chunkp != NULL) {
142 			addr = (u_long)chunkp;
143 			if (KREAD(kd, addr, &chunk))
144 				return (NULL);
145 
146 			if (UVM_AMAP_BUCKET(&amap, chunk.ac_baseslot) !=
147 			    bucket)
148 				return (NULL);
149 			if (slot >= chunk.ac_baseslot &&
150 			    slot < chunk.ac_baseslot + chunk.ac_nslot)
151 				break;
152 
153 			chunkp = TAILQ_NEXT(&chunk, ac_list);
154 		}
155 		if (chunkp == NULL)
156 			return (NULL);
157 	}
158 
159 	addr = (u_long)&chunkp->ac_anon[UVM_AMAP_SLOTIDX(slot)];
160 	if (KREAD(kd, addr, &anonp))
161 		return (NULL);
162 
163 	return (anonp);
164 }
165 
166 static char *
_kvm_ureadm(kvm_t * kd,const struct kinfo_proc * p,u_long va,u_long * cnt)167 _kvm_ureadm(kvm_t *kd, const struct kinfo_proc *p, u_long va, u_long *cnt)
168 {
169 	u_long addr, offset, slot;
170 	struct vmspace vm;
171 	struct vm_anon *anonp, anon;
172 	struct vm_map_entry vme;
173 	struct vm_page pg;
174 	unsigned long rboff;
175 
176 	if (kd->swapspc == 0) {
177 		kd->swapspc = _kvm_malloc(kd, kd->nbpg);
178 		if (kd->swapspc == 0)
179 			return (NULL);
180 	}
181 
182 	rboff = (unsigned long)&vme.daddrs.addr_entry - (unsigned long)&vme;
183 
184 	/*
185 	 * Look through the address map for the memory object
186 	 * that corresponds to the given virtual address.
187 	 */
188 	if (KREAD(kd, (u_long)p->p_vmspace, &vm))
189 		return (NULL);
190 	addr = (u_long)vm.vm_map.addr.rbh_root.rbt_root;
191 	while (1) {
192 		if (addr == 0)
193 			return (NULL);
194 		addr -= rboff;
195 		if (KREAD(kd, addr, &vme))
196 			return (NULL);
197 
198 		if (va < vme.start)
199 			addr = (u_long)vme.daddrs.addr_entry.rbt_left;
200 		else if (va >= vme.end + vme.guard + vme.fspace)
201 			addr = (u_long)vme.daddrs.addr_entry.rbt_right;
202 		else if (va >= vme.end)
203 			return (NULL);
204 		else
205 			break;
206 	}
207 
208 	/*
209 	 * we found the map entry, now to find the object...
210 	 */
211 	if (vme.aref.ar_amap == NULL)
212 		return (NULL);
213 
214 	offset = va - vme.start;
215 	slot = offset / kd->nbpg + vme.aref.ar_pageoff;
216 
217 	anonp = _kvm_findanon(kd, vme.aref.ar_amap, slot);
218 	if (anonp == NULL)
219 		return (NULL);
220 
221 	addr = (u_long)anonp;
222 	if (KREAD(kd, addr, &anon))
223 		return (NULL);
224 
225 	addr = (u_long)anon.an_page;
226 	if (addr) {
227 		if (KREAD(kd, addr, &pg))
228 			return (NULL);
229 
230 		if (_kvm_pread(kd, kd->pmfd, (void *)kd->swapspc,
231 		    (size_t)kd->nbpg, _kvm_pa2off(kd, pg.phys_addr)) != kd->nbpg)
232 			return (NULL);
233 	} else {
234 		if (kd->swfd == -1 ||
235 		    _kvm_pread(kd, kd->swfd, (void *)kd->swapspc,
236 		    (size_t)kd->nbpg,
237 		    (off_t)(anon.an_swslot * kd->nbpg)) != kd->nbpg)
238 			return (NULL);
239 	}
240 
241 	/* Found the page. */
242 	offset %= kd->nbpg;
243 	*cnt = kd->nbpg - offset;
244 	return (&kd->swapspc[offset]);
245 }
246 
247 void *
_kvm_reallocarray(kvm_t * kd,void * p,size_t i,size_t n)248 _kvm_reallocarray(kvm_t *kd, void *p, size_t i, size_t n)
249 {
250 	void *np = reallocarray(p, i, n);
251 
252 	if (np == 0)
253 		_kvm_err(kd, kd->program, "out of memory");
254 	return (np);
255 }
256 
257 /*
258  * Read in an argument vector from the user address space of process p.
259  * addr if the user-space base address of narg null-terminated contiguous
260  * strings.  This is used to read in both the command arguments and
261  * environment strings.  Read at most maxcnt characters of strings.
262  */
263 static char **
kvm_argv(kvm_t * kd,const struct kinfo_proc * p,u_long addr,int narg,int maxcnt,int isenv)264 kvm_argv(kvm_t *kd, const struct kinfo_proc *p, u_long addr, int narg,
265     int maxcnt, int isenv)
266 {
267 	char *np, *cp, *ep, *ap, **argv, ***pargv, **pargspc, **pargbuf;
268 	u_long oaddr = -1;
269 	int len, cc, *parglen, *pargc;
270 	size_t argc;
271 
272 	/*
273 	 * Check that there aren't an unreasonable number of arguments,
274 	 * and that the address is in user space.
275 	 */
276 	if (narg > ARG_MAX || addr < VM_MIN_ADDRESS || addr >= VM_MAXUSER_ADDRESS)
277 		return (0);
278 
279 	if (isenv) {
280 		pargspc = &kd->envspc;
281 		pargbuf = &kd->envbuf;
282 		parglen = &kd->envlen;
283 		pargv = &kd->envp;
284 		pargc = &kd->envc;
285 	} else {
286 		pargspc = &kd->argspc;
287 		pargbuf = &kd->argbuf;
288 		parglen = &kd->arglen;
289 		pargv = &kd->argv;
290 		pargc = &kd->argc;
291 	}
292 
293 	if (*pargv == 0)
294 		argc = MAXIMUM(narg + 1, 32);
295 	else if (narg + 1 > *pargc)
296 		argc = MAXIMUM(2 * (*pargc), narg + 1);
297 	else
298 		goto argv_allocated;
299 	argv = _kvm_reallocarray(kd, *pargv, argc, sizeof(**pargv));
300 	if (argv == 0)
301 		return (0);
302 	*pargv = argv;
303 	*pargc = argc;
304 
305 argv_allocated:
306 	if (*pargspc == 0) {
307 		*pargspc = _kvm_malloc(kd, kd->nbpg);
308 		if (*pargspc == 0)
309 			return (0);
310 		*parglen = kd->nbpg;
311 	}
312 	if (*pargbuf == 0) {
313 		*pargbuf = _kvm_malloc(kd, kd->nbpg);
314 		if (*pargbuf == 0)
315 			return (0);
316 	}
317 	cc = sizeof(char *) * narg;
318 	if (kvm_ureadm(kd, p, addr, (char *)*pargv, cc) != cc)
319 		return (0);
320 	ap = np = *pargspc;
321 	argv = *pargv;
322 	len = 0;
323 
324 	/*
325 	 * Loop over pages, filling in the argument vector.
326 	 */
327 	while (argv < *pargv + narg && *argv != 0) {
328 		addr = (u_long)*argv & ~(kd->nbpg - 1);
329 		if (addr != oaddr) {
330 			if (kvm_ureadm(kd, p, addr, *pargbuf, kd->nbpg) !=
331 			    kd->nbpg)
332 				return (0);
333 			oaddr = addr;
334 		}
335 		addr = (u_long)*argv & (kd->nbpg - 1);
336 		cp = *pargbuf + addr;
337 		cc = kd->nbpg - addr;
338 		if (maxcnt > 0 && cc > maxcnt - len)
339 			cc = maxcnt - len;
340 		ep = memchr(cp, '\0', cc);
341 		if (ep != 0)
342 			cc = ep - cp + 1;
343 		if (len + cc > *parglen) {
344 			ptrdiff_t off;
345 			char **pp;
346 			char *op = *pargspc;
347 			char *newp;
348 
349 			newp = _kvm_reallocarray(kd, *pargspc,
350 			    *parglen, 2);
351 			if (newp == 0)
352 				return (0);
353 			*pargspc = newp;
354 			*parglen *= 2;
355 			/*
356 			 * Adjust argv pointers in case realloc moved
357 			 * the string space.
358 			 */
359 			off = *pargspc - op;
360 			for (pp = *pargv; pp < argv; pp++)
361 				*pp += off;
362 			ap += off;
363 			np += off;
364 		}
365 		memcpy(np, cp, cc);
366 		np += cc;
367 		len += cc;
368 		if (ep != 0) {
369 			*argv++ = ap;
370 			ap = np;
371 		} else
372 			*argv += cc;
373 		if (maxcnt > 0 && len >= maxcnt) {
374 			/*
375 			 * We're stopping prematurely.  Terminate the
376 			 * current string.
377 			 */
378 			if (ep == 0) {
379 				*np = '\0';
380 				*argv++ = ap;
381 			}
382 			break;
383 		}
384 	}
385 	/* Make sure argv is terminated. */
386 	*argv = 0;
387 	return (*pargv);
388 }
389 
390 static void
ps_str_a(struct ps_strings * p,u_long * addr,int * n)391 ps_str_a(struct ps_strings *p, u_long *addr, int *n)
392 {
393 	*addr = (u_long)p->ps_argvstr;
394 	*n = p->ps_nargvstr;
395 }
396 
397 static void
ps_str_e(struct ps_strings * p,u_long * addr,int * n)398 ps_str_e(struct ps_strings *p, u_long *addr, int *n)
399 {
400 	*addr = (u_long)p->ps_envstr;
401 	*n = p->ps_nenvstr;
402 }
403 
404 /*
405  * Determine if the proc indicated by p is still active.
406  * This test is not 100% foolproof in theory, but chances of
407  * being wrong are very low.
408  */
409 static int
proc_verify(kvm_t * kd,const struct kinfo_proc * p)410 proc_verify(kvm_t *kd, const struct kinfo_proc *p)
411 {
412 	struct proc kernproc;
413 	struct process kernprocess;
414 
415 	if (p->p_psflags & (PS_EMBRYO | PS_ZOMBIE))
416 		return (0);
417 
418 	/*
419 	 * Just read in the whole proc.  It's not that big relative
420 	 * to the cost of the read system call.
421 	 */
422 	if (KREAD(kd, (u_long)p->p_paddr, &kernproc))
423 		return (0);
424 	if (KREAD(kd, (u_long)kernproc.p_p, &kernprocess))
425 		return (0);
426 	if (p->p_pid != kernprocess.ps_pid)
427 		return (0);
428 	return ((kernprocess.ps_flags & (PS_EMBRYO | PS_ZOMBIE)) == 0);
429 }
430 
431 static char **
kvm_doargv(kvm_t * kd,const struct kinfo_proc * p,int nchr,int isenv,void (* info)(struct ps_strings *,u_long *,int *))432 kvm_doargv(kvm_t *kd, const struct kinfo_proc *p, int nchr, int isenv,
433     void (*info)(struct ps_strings *, u_long *, int *))
434 {
435 	struct proc pp;
436 	struct process pr;
437 	struct ps_strings *ps;
438 	struct ps_strings arginfo;
439 	u_long addr;
440 	char **ap;
441 	int cnt;
442 
443 	/* get ps_strings address */
444 	if (KREAD(kd, (u_long)p->p_paddr, &pp))
445 		return (0);
446 	if (KREAD(kd, (u_long)pp.p_p, &pr))
447 		return (0);
448 	ps = (struct ps_strings *)pr.ps_strings;
449 
450 	/*
451 	 * Pointers are stored at the top of the user stack.
452 	 */
453 	if (p->p_psflags & (PS_EMBRYO | PS_ZOMBIE) ||
454 	    kvm_ureadm(kd, p, (u_long)ps, (char *)&arginfo,
455 	    sizeof(arginfo)) != sizeof(arginfo))
456 		return (0);
457 
458 	(*info)(&arginfo, &addr, &cnt);
459 	if (cnt == 0)
460 		return (0);
461 	ap = kvm_argv(kd, p, addr, cnt, nchr, isenv);
462 	/*
463 	 * For live kernels, make sure this process didn't go away.
464 	 */
465 	if (ap != 0 && ISALIVE(kd) && !proc_verify(kd, p))
466 		ap = 0;
467 	return (ap);
468 }
469 
470 static char **
kvm_arg_sysctl(kvm_t * kd,pid_t pid,int nchr,int isenv)471 kvm_arg_sysctl(kvm_t *kd, pid_t pid, int nchr, int isenv)
472 {
473 	size_t len, orglen;
474 	int mib[4], ret;
475 	char *buf, **pargbuf;
476 
477 	if (isenv) {
478 		pargbuf = &kd->envbuf;
479 		orglen = kd->nbpg;
480 	} else {
481 		pargbuf = &kd->argbuf;
482 		orglen = 8 * kd->nbpg;	/* XXX - should be ARG_MAX */
483 	}
484 	if (*pargbuf == NULL &&
485 	    (*pargbuf = _kvm_malloc(kd, orglen)) == NULL)
486 		return (NULL);
487 
488 again:
489 	mib[0] = CTL_KERN;
490 	mib[1] = KERN_PROC_ARGS;
491 	mib[2] = (int)pid;
492 	mib[3] = isenv ? KERN_PROC_ENV : KERN_PROC_ARGV;
493 
494 	len = orglen;
495 	ret = (sysctl(mib, 4, *pargbuf, &len, NULL, 0) == -1);
496 	if (ret && errno == ENOMEM) {
497 		buf = _kvm_reallocarray(kd, *pargbuf, orglen, 2);
498 		if (buf == NULL)
499 			return (NULL);
500 		orglen *= 2;
501 		*pargbuf = buf;
502 		goto again;
503 	}
504 
505 	if (ret) {
506 		free(*pargbuf);
507 		*pargbuf = NULL;
508 		_kvm_syserr(kd, kd->program, "kvm_arg_sysctl");
509 		return (NULL);
510 	}
511 #if 0
512 	for (argv = (char **)*pargbuf; *argv != NULL; argv++)
513 		if (strlen(*argv) > nchr)
514 			*argv[nchr] = '\0';
515 #endif
516 
517 	return (char **)(*pargbuf);
518 }
519 
520 /*
521  * Get the command args.  This code is now machine independent.
522  */
523 char **
kvm_getargv(kvm_t * kd,const struct kinfo_proc * kp,int nchr)524 kvm_getargv(kvm_t *kd, const struct kinfo_proc *kp, int nchr)
525 {
526 	if (ISALIVE(kd))
527 		return (kvm_arg_sysctl(kd, kp->p_pid, nchr, 0));
528 	return (kvm_doargv(kd, kp, nchr, 0, ps_str_a));
529 }
530 
531 char **
kvm_getenvv(kvm_t * kd,const struct kinfo_proc * kp,int nchr)532 kvm_getenvv(kvm_t *kd, const struct kinfo_proc *kp, int nchr)
533 {
534 	if (ISALIVE(kd))
535 		return (kvm_arg_sysctl(kd, kp->p_pid, nchr, 1));
536 	return (kvm_doargv(kd, kp, nchr, 1, ps_str_e));
537 }
538 
539 /*
540  * Read from user space.  The user context is given by p.
541  */
542 static ssize_t
kvm_ureadm(kvm_t * kd,const struct kinfo_proc * p,u_long uva,char * buf,size_t len)543 kvm_ureadm(kvm_t *kd, const struct kinfo_proc *p, u_long uva, char *buf,
544     size_t len)
545 {
546 	char *cp = buf;
547 
548 	while (len > 0) {
549 		u_long cnt;
550 		size_t cc;
551 		char *dp;
552 
553 		dp = _kvm_ureadm(kd, p, uva, &cnt);
554 		if (dp == 0) {
555 			_kvm_err(kd, 0, "invalid address (%lx)", uva);
556 			return (0);
557 		}
558 		cc = (size_t)MINIMUM(cnt, len);
559 		memcpy(cp, dp, cc);
560 		cp += cc;
561 		uva += cc;
562 		len -= cc;
563 	}
564 	return (ssize_t)(cp - buf);
565 }
566