xref: /freebsd/lib/libkvm/kvm_proc.c (revision aa0a1e58)
1 /*-
2  * Copyright (c) 1989, 1992, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software developed by the Computer Systems
6  * Engineering group at Lawrence Berkeley Laboratory under DARPA contract
7  * BG 91-66 and contributed to Berkeley.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 #if 0
35 #if defined(LIBC_SCCS) && !defined(lint)
36 static char sccsid[] = "@(#)kvm_proc.c	8.3 (Berkeley) 9/23/93";
37 #endif /* LIBC_SCCS and not lint */
38 #endif
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 /*
44  * Proc traversal interface for kvm.  ps and w are (probably) the exclusive
45  * users of this code, so we've factored it out into a separate module.
46  * Thus, we keep this grunge out of the other kvm applications (i.e.,
47  * most other applications are interested only in open/close/read/nlist).
48  */
49 
50 #include <sys/param.h>
51 #define	_WANT_UCRED	/* make ucred.h give us 'struct ucred' */
52 #include <sys/ucred.h>
53 #include <sys/queue.h>
54 #include <sys/_lock.h>
55 #include <sys/_mutex.h>
56 #include <sys/_task.h>
57 #include <sys/cpuset.h>
58 #include <sys/user.h>
59 #include <sys/proc.h>
60 #define	_WANT_PRISON	/* make jail.h give us 'struct prison' */
61 #include <sys/jail.h>
62 #include <sys/exec.h>
63 #include <sys/stat.h>
64 #include <sys/sysent.h>
65 #include <sys/ioctl.h>
66 #include <sys/tty.h>
67 #include <sys/file.h>
68 #include <sys/conf.h>
69 #include <stdio.h>
70 #include <stdlib.h>
71 #include <unistd.h>
72 #include <nlist.h>
73 #include <kvm.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_param.h>
77 
78 #include <sys/sysctl.h>
79 
80 #include <limits.h>
81 #include <memory.h>
82 #include <paths.h>
83 
84 #include "kvm_private.h"
85 
86 #define KREAD(kd, addr, obj) \
87 	(kvm_read(kd, addr, (char *)(obj), sizeof(*obj)) != sizeof(*obj))
88 
89 static int ticks;
90 static int hz;
91 static uint64_t cpu_tick_frequency;
92 
93 /*
94  * From sys/kern/kern_tc.c. Depends on cpu_tick_frequency, which is
95  * read/initialized before this function is ever called.
96  */
97 static uint64_t
98 cputick2usec(uint64_t tick)
99 {
100 
101 	if (cpu_tick_frequency == 0)
102 		return (0);
103 	if (tick > 18446744073709551)		/* floor(2^64 / 1000) */
104 		return (tick / (cpu_tick_frequency / 1000000));
105 	else if (tick > 18446744073709)	/* floor(2^64 / 1000000) */
106 		return ((tick * 1000) / (cpu_tick_frequency / 1000));
107 	else
108 		return ((tick * 1000000) / cpu_tick_frequency);
109 }
110 
111 /*
112  * Read proc's from memory file into buffer bp, which has space to hold
113  * at most maxcnt procs.
114  */
115 static int
116 kvm_proclist(kvm_t *kd, int what, int arg, struct proc *p,
117     struct kinfo_proc *bp, int maxcnt)
118 {
119 	int cnt = 0;
120 	struct kinfo_proc kinfo_proc, *kp;
121 	struct pgrp pgrp;
122 	struct session sess;
123 	struct cdev t_cdev;
124 	struct tty tty;
125 	struct vmspace vmspace;
126 	struct sigacts sigacts;
127 #if 0
128 	struct pstats pstats;
129 #endif
130 	struct ucred ucred;
131 	struct prison pr;
132 	struct thread mtd;
133 	struct proc proc;
134 	struct proc pproc;
135 	struct sysentvec sysent;
136 	char svname[KI_EMULNAMELEN];
137 
138 	kp = &kinfo_proc;
139 	kp->ki_structsize = sizeof(kinfo_proc);
140 	/*
141 	 * Loop on the processes. this is completely broken because we need to be
142 	 * able to loop on the threads and merge the ones that are the same process some how.
143 	 */
144 	for (; cnt < maxcnt && p != NULL; p = LIST_NEXT(&proc, p_list)) {
145 		memset(kp, 0, sizeof *kp);
146 		if (KREAD(kd, (u_long)p, &proc)) {
147 			_kvm_err(kd, kd->program, "can't read proc at %p", p);
148 			return (-1);
149 		}
150 		if (proc.p_state != PRS_ZOMBIE) {
151 			if (KREAD(kd, (u_long)TAILQ_FIRST(&proc.p_threads),
152 			    &mtd)) {
153 				_kvm_err(kd, kd->program,
154 				    "can't read thread at %p",
155 				    TAILQ_FIRST(&proc.p_threads));
156 				return (-1);
157 			}
158 		}
159 		if (KREAD(kd, (u_long)proc.p_ucred, &ucred) == 0) {
160 			kp->ki_ruid = ucred.cr_ruid;
161 			kp->ki_svuid = ucred.cr_svuid;
162 			kp->ki_rgid = ucred.cr_rgid;
163 			kp->ki_svgid = ucred.cr_svgid;
164 			kp->ki_cr_flags = ucred.cr_flags;
165 			if (ucred.cr_ngroups > KI_NGROUPS) {
166 				kp->ki_ngroups = KI_NGROUPS;
167 				kp->ki_cr_flags |= KI_CRF_GRP_OVERFLOW;
168 			} else
169 				kp->ki_ngroups = ucred.cr_ngroups;
170 			kvm_read(kd, (u_long)ucred.cr_groups, kp->ki_groups,
171 			    kp->ki_ngroups * sizeof(gid_t));
172 			kp->ki_uid = ucred.cr_uid;
173 			if (ucred.cr_prison != NULL) {
174 				if (KREAD(kd, (u_long)ucred.cr_prison, &pr)) {
175 					_kvm_err(kd, kd->program,
176 					    "can't read prison at %p",
177 					    ucred.cr_prison);
178 					return (-1);
179 				}
180 				kp->ki_jid = pr.pr_id;
181 			}
182 		}
183 
184 		switch(what & ~KERN_PROC_INC_THREAD) {
185 
186 		case KERN_PROC_GID:
187 			if (kp->ki_groups[0] != (gid_t)arg)
188 				continue;
189 			break;
190 
191 		case KERN_PROC_PID:
192 			if (proc.p_pid != (pid_t)arg)
193 				continue;
194 			break;
195 
196 		case KERN_PROC_RGID:
197 			if (kp->ki_rgid != (gid_t)arg)
198 				continue;
199 			break;
200 
201 		case KERN_PROC_UID:
202 			if (kp->ki_uid != (uid_t)arg)
203 				continue;
204 			break;
205 
206 		case KERN_PROC_RUID:
207 			if (kp->ki_ruid != (uid_t)arg)
208 				continue;
209 			break;
210 		}
211 		/*
212 		 * We're going to add another proc to the set.  If this
213 		 * will overflow the buffer, assume the reason is because
214 		 * nprocs (or the proc list) is corrupt and declare an error.
215 		 */
216 		if (cnt >= maxcnt) {
217 			_kvm_err(kd, kd->program, "nprocs corrupt");
218 			return (-1);
219 		}
220 		/*
221 		 * gather kinfo_proc
222 		 */
223 		kp->ki_paddr = p;
224 		kp->ki_addr = 0;	/* XXX uarea */
225 		/* kp->ki_kstack = proc.p_thread.td_kstack; XXXKSE */
226 		kp->ki_args = proc.p_args;
227 		kp->ki_tracep = proc.p_tracevp;
228 		kp->ki_textvp = proc.p_textvp;
229 		kp->ki_fd = proc.p_fd;
230 		kp->ki_vmspace = proc.p_vmspace;
231 		if (proc.p_sigacts != NULL) {
232 			if (KREAD(kd, (u_long)proc.p_sigacts, &sigacts)) {
233 				_kvm_err(kd, kd->program,
234 				    "can't read sigacts at %p", proc.p_sigacts);
235 				return (-1);
236 			}
237 			kp->ki_sigignore = sigacts.ps_sigignore;
238 			kp->ki_sigcatch = sigacts.ps_sigcatch;
239 		}
240 #if 0
241 		if ((proc.p_flag & P_INMEM) && proc.p_stats != NULL) {
242 			if (KREAD(kd, (u_long)proc.p_stats, &pstats)) {
243 				_kvm_err(kd, kd->program,
244 				    "can't read stats at %x", proc.p_stats);
245 				return (-1);
246 			}
247 			kp->ki_start = pstats.p_start;
248 
249 			/*
250 			 * XXX: The times here are probably zero and need
251 			 * to be calculated from the raw data in p_rux and
252 			 * p_crux.
253 			 */
254 			kp->ki_rusage = pstats.p_ru;
255 			kp->ki_childstime = pstats.p_cru.ru_stime;
256 			kp->ki_childutime = pstats.p_cru.ru_utime;
257 			/* Some callers want child-times in a single value */
258 			timeradd(&kp->ki_childstime, &kp->ki_childutime,
259 			    &kp->ki_childtime);
260 		}
261 #endif
262 		if (proc.p_oppid)
263 			kp->ki_ppid = proc.p_oppid;
264 		else if (proc.p_pptr) {
265 			if (KREAD(kd, (u_long)proc.p_pptr, &pproc)) {
266 				_kvm_err(kd, kd->program,
267 				    "can't read pproc at %p", proc.p_pptr);
268 				return (-1);
269 			}
270 			kp->ki_ppid = pproc.p_pid;
271 		} else
272 			kp->ki_ppid = 0;
273 		if (proc.p_pgrp == NULL)
274 			goto nopgrp;
275 		if (KREAD(kd, (u_long)proc.p_pgrp, &pgrp)) {
276 			_kvm_err(kd, kd->program, "can't read pgrp at %p",
277 				 proc.p_pgrp);
278 			return (-1);
279 		}
280 		kp->ki_pgid = pgrp.pg_id;
281 		kp->ki_jobc = pgrp.pg_jobc;
282 		if (KREAD(kd, (u_long)pgrp.pg_session, &sess)) {
283 			_kvm_err(kd, kd->program, "can't read session at %p",
284 				pgrp.pg_session);
285 			return (-1);
286 		}
287 		kp->ki_sid = sess.s_sid;
288 		(void)memcpy(kp->ki_login, sess.s_login,
289 						sizeof(kp->ki_login));
290 		kp->ki_kiflag = sess.s_ttyvp ? KI_CTTY : 0;
291 		if (sess.s_leader == p)
292 			kp->ki_kiflag |= KI_SLEADER;
293 		if ((proc.p_flag & P_CONTROLT) && sess.s_ttyp != NULL) {
294 			if (KREAD(kd, (u_long)sess.s_ttyp, &tty)) {
295 				_kvm_err(kd, kd->program,
296 					 "can't read tty at %p", sess.s_ttyp);
297 				return (-1);
298 			}
299 			if (tty.t_dev != NULL) {
300 				if (KREAD(kd, (u_long)tty.t_dev, &t_cdev)) {
301 					_kvm_err(kd, kd->program,
302 						 "can't read cdev at %p",
303 						tty.t_dev);
304 					return (-1);
305 				}
306 #if 0
307 				kp->ki_tdev = t_cdev.si_udev;
308 #else
309 				kp->ki_tdev = NODEV;
310 #endif
311 			}
312 			if (tty.t_pgrp != NULL) {
313 				if (KREAD(kd, (u_long)tty.t_pgrp, &pgrp)) {
314 					_kvm_err(kd, kd->program,
315 						 "can't read tpgrp at %p",
316 						tty.t_pgrp);
317 					return (-1);
318 				}
319 				kp->ki_tpgid = pgrp.pg_id;
320 			} else
321 				kp->ki_tpgid = -1;
322 			if (tty.t_session != NULL) {
323 				if (KREAD(kd, (u_long)tty.t_session, &sess)) {
324 					_kvm_err(kd, kd->program,
325 					    "can't read session at %p",
326 					    tty.t_session);
327 					return (-1);
328 				}
329 				kp->ki_tsid = sess.s_sid;
330 			}
331 		} else {
332 nopgrp:
333 			kp->ki_tdev = NODEV;
334 		}
335 		if ((proc.p_state != PRS_ZOMBIE) && mtd.td_wmesg)
336 			(void)kvm_read(kd, (u_long)mtd.td_wmesg,
337 			    kp->ki_wmesg, WMESGLEN);
338 
339 		(void)kvm_read(kd, (u_long)proc.p_vmspace,
340 		    (char *)&vmspace, sizeof(vmspace));
341 		kp->ki_size = vmspace.vm_map.size;
342 		/*
343 		 * Approximate the kernel's method of calculating
344 		 * this field.
345 		 */
346 #define		pmap_resident_count(pm) ((pm)->pm_stats.resident_count)
347 		kp->ki_rssize = pmap_resident_count(&vmspace.vm_pmap);
348 		kp->ki_swrss = vmspace.vm_swrss;
349 		kp->ki_tsize = vmspace.vm_tsize;
350 		kp->ki_dsize = vmspace.vm_dsize;
351 		kp->ki_ssize = vmspace.vm_ssize;
352 
353 		switch (what & ~KERN_PROC_INC_THREAD) {
354 
355 		case KERN_PROC_PGRP:
356 			if (kp->ki_pgid != (pid_t)arg)
357 				continue;
358 			break;
359 
360 		case KERN_PROC_SESSION:
361 			if (kp->ki_sid != (pid_t)arg)
362 				continue;
363 			break;
364 
365 		case KERN_PROC_TTY:
366 			if ((proc.p_flag & P_CONTROLT) == 0 ||
367 			     kp->ki_tdev != (dev_t)arg)
368 				continue;
369 			break;
370 		}
371 		if (proc.p_comm[0] != 0)
372 			strlcpy(kp->ki_comm, proc.p_comm, MAXCOMLEN);
373 		(void)kvm_read(kd, (u_long)proc.p_sysent, (char *)&sysent,
374 		    sizeof(sysent));
375 		(void)kvm_read(kd, (u_long)sysent.sv_name, (char *)&svname,
376 		    sizeof(svname));
377 		if (svname[0] != 0)
378 			strlcpy(kp->ki_emul, svname, KI_EMULNAMELEN);
379 		if ((proc.p_state != PRS_ZOMBIE) &&
380 		    (mtd.td_blocked != 0)) {
381 			kp->ki_kiflag |= KI_LOCKBLOCK;
382 			if (mtd.td_lockname)
383 				(void)kvm_read(kd,
384 				    (u_long)mtd.td_lockname,
385 				    kp->ki_lockname, LOCKNAMELEN);
386 			kp->ki_lockname[LOCKNAMELEN] = 0;
387 		}
388 		kp->ki_runtime = cputick2usec(proc.p_rux.rux_runtime);
389 		kp->ki_pid = proc.p_pid;
390 		kp->ki_siglist = proc.p_siglist;
391 		SIGSETOR(kp->ki_siglist, mtd.td_siglist);
392 		kp->ki_sigmask = mtd.td_sigmask;
393 		kp->ki_xstat = proc.p_xstat;
394 		kp->ki_acflag = proc.p_acflag;
395 		kp->ki_lock = proc.p_lock;
396 		if (proc.p_state != PRS_ZOMBIE) {
397 			kp->ki_swtime = (ticks - proc.p_swtick) / hz;
398 			kp->ki_flag = proc.p_flag;
399 			kp->ki_sflag = 0;
400 			kp->ki_nice = proc.p_nice;
401 			kp->ki_traceflag = proc.p_traceflag;
402 			if (proc.p_state == PRS_NORMAL) {
403 				if (TD_ON_RUNQ(&mtd) ||
404 				    TD_CAN_RUN(&mtd) ||
405 				    TD_IS_RUNNING(&mtd)) {
406 					kp->ki_stat = SRUN;
407 				} else if (mtd.td_state ==
408 				    TDS_INHIBITED) {
409 					if (P_SHOULDSTOP(&proc)) {
410 						kp->ki_stat = SSTOP;
411 					} else if (
412 					    TD_IS_SLEEPING(&mtd)) {
413 						kp->ki_stat = SSLEEP;
414 					} else if (TD_ON_LOCK(&mtd)) {
415 						kp->ki_stat = SLOCK;
416 					} else {
417 						kp->ki_stat = SWAIT;
418 					}
419 				}
420 			} else {
421 				kp->ki_stat = SIDL;
422 			}
423 			/* Stuff from the thread */
424 			kp->ki_pri.pri_level = mtd.td_priority;
425 			kp->ki_pri.pri_native = mtd.td_base_pri;
426 			kp->ki_lastcpu = mtd.td_lastcpu;
427 			kp->ki_wchan = mtd.td_wchan;
428 			if (mtd.td_name[0] != 0)
429 				strlcpy(kp->ki_ocomm, mtd.td_name, MAXCOMLEN);
430 			kp->ki_oncpu = mtd.td_oncpu;
431 			if (mtd.td_name[0] != '\0')
432 				strlcpy(kp->ki_ocomm, mtd.td_name, sizeof(kp->ki_ocomm));
433 			kp->ki_pctcpu = 0;
434 			kp->ki_rqindex = 0;
435 		} else {
436 			kp->ki_stat = SZOMB;
437 		}
438 		bcopy(&kinfo_proc, bp, sizeof(kinfo_proc));
439 		++bp;
440 		++cnt;
441 	}
442 	return (cnt);
443 }
444 
445 /*
446  * Build proc info array by reading in proc list from a crash dump.
447  * Return number of procs read.  maxcnt is the max we will read.
448  */
449 static int
450 kvm_deadprocs(kvm_t *kd, int what, int arg, u_long a_allproc,
451     u_long a_zombproc, int maxcnt)
452 {
453 	struct kinfo_proc *bp = kd->procbase;
454 	int acnt, zcnt;
455 	struct proc *p;
456 
457 	if (KREAD(kd, a_allproc, &p)) {
458 		_kvm_err(kd, kd->program, "cannot read allproc");
459 		return (-1);
460 	}
461 	acnt = kvm_proclist(kd, what, arg, p, bp, maxcnt);
462 	if (acnt < 0)
463 		return (acnt);
464 
465 	if (KREAD(kd, a_zombproc, &p)) {
466 		_kvm_err(kd, kd->program, "cannot read zombproc");
467 		return (-1);
468 	}
469 	zcnt = kvm_proclist(kd, what, arg, p, bp + acnt, maxcnt - acnt);
470 	if (zcnt < 0)
471 		zcnt = 0;
472 
473 	return (acnt + zcnt);
474 }
475 
476 struct kinfo_proc *
477 kvm_getprocs(kvm_t *kd, int op, int arg, int *cnt)
478 {
479 	int mib[4], st, nprocs;
480 	size_t size;
481 	int temp_op;
482 
483 	if (kd->procbase != 0) {
484 		free((void *)kd->procbase);
485 		/*
486 		 * Clear this pointer in case this call fails.  Otherwise,
487 		 * kvm_close() will free it again.
488 		 */
489 		kd->procbase = 0;
490 	}
491 	if (ISALIVE(kd)) {
492 		size = 0;
493 		mib[0] = CTL_KERN;
494 		mib[1] = KERN_PROC;
495 		mib[2] = op;
496 		mib[3] = arg;
497 		temp_op = op & ~KERN_PROC_INC_THREAD;
498 		st = sysctl(mib,
499 		    temp_op == KERN_PROC_ALL || temp_op == KERN_PROC_PROC ?
500 		    3 : 4, NULL, &size, NULL, 0);
501 		if (st == -1) {
502 			_kvm_syserr(kd, kd->program, "kvm_getprocs");
503 			return (0);
504 		}
505 		/*
506 		 * We can't continue with a size of 0 because we pass
507 		 * it to realloc() (via _kvm_realloc()), and passing 0
508 		 * to realloc() results in undefined behavior.
509 		 */
510 		if (size == 0) {
511 			/*
512 			 * XXX: We should probably return an invalid,
513 			 * but non-NULL, pointer here so any client
514 			 * program trying to dereference it will
515 			 * crash.  However, _kvm_freeprocs() calls
516 			 * free() on kd->procbase if it isn't NULL,
517 			 * and free()'ing a junk pointer isn't good.
518 			 * Then again, _kvm_freeprocs() isn't used
519 			 * anywhere . . .
520 			 */
521 			kd->procbase = _kvm_malloc(kd, 1);
522 			goto liveout;
523 		}
524 		do {
525 			size += size / 10;
526 			kd->procbase = (struct kinfo_proc *)
527 			    _kvm_realloc(kd, kd->procbase, size);
528 			if (kd->procbase == 0)
529 				return (0);
530 			st = sysctl(mib, temp_op == KERN_PROC_ALL ||
531 			    temp_op == KERN_PROC_PROC ? 3 : 4,
532 			    kd->procbase, &size, NULL, 0);
533 		} while (st == -1 && errno == ENOMEM);
534 		if (st == -1) {
535 			_kvm_syserr(kd, kd->program, "kvm_getprocs");
536 			return (0);
537 		}
538 		/*
539 		 * We have to check the size again because sysctl()
540 		 * may "round up" oldlenp if oldp is NULL; hence it
541 		 * might've told us that there was data to get when
542 		 * there really isn't any.
543 		 */
544 		if (size > 0 &&
545 		    kd->procbase->ki_structsize != sizeof(struct kinfo_proc)) {
546 			_kvm_err(kd, kd->program,
547 			    "kinfo_proc size mismatch (expected %zu, got %d)",
548 			    sizeof(struct kinfo_proc),
549 			    kd->procbase->ki_structsize);
550 			return (0);
551 		}
552 liveout:
553 		nprocs = size == 0 ? 0 : size / kd->procbase->ki_structsize;
554 	} else {
555 		struct nlist nl[7], *p;
556 
557 		nl[0].n_name = "_nprocs";
558 		nl[1].n_name = "_allproc";
559 		nl[2].n_name = "_zombproc";
560 		nl[3].n_name = "_ticks";
561 		nl[4].n_name = "_hz";
562 		nl[5].n_name = "_cpu_tick_frequency";
563 		nl[6].n_name = 0;
564 
565 		if (kvm_nlist(kd, nl) != 0) {
566 			for (p = nl; p->n_type != 0; ++p)
567 				;
568 			_kvm_err(kd, kd->program,
569 				 "%s: no such symbol", p->n_name);
570 			return (0);
571 		}
572 		if (KREAD(kd, nl[0].n_value, &nprocs)) {
573 			_kvm_err(kd, kd->program, "can't read nprocs");
574 			return (0);
575 		}
576 		if (KREAD(kd, nl[3].n_value, &ticks)) {
577 			_kvm_err(kd, kd->program, "can't read ticks");
578 			return (0);
579 		}
580 		if (KREAD(kd, nl[4].n_value, &hz)) {
581 			_kvm_err(kd, kd->program, "can't read hz");
582 			return (0);
583 		}
584 		if (KREAD(kd, nl[5].n_value, &cpu_tick_frequency)) {
585 			_kvm_err(kd, kd->program,
586 			    "can't read cpu_tick_frequency");
587 			return (0);
588 		}
589 		size = nprocs * sizeof(struct kinfo_proc);
590 		kd->procbase = (struct kinfo_proc *)_kvm_malloc(kd, size);
591 		if (kd->procbase == 0)
592 			return (0);
593 
594 		nprocs = kvm_deadprocs(kd, op, arg, nl[1].n_value,
595 				      nl[2].n_value, nprocs);
596 #ifdef notdef
597 		size = nprocs * sizeof(struct kinfo_proc);
598 		(void)realloc(kd->procbase, size);
599 #endif
600 	}
601 	*cnt = nprocs;
602 	return (kd->procbase);
603 }
604 
605 void
606 _kvm_freeprocs(kvm_t *kd)
607 {
608 	if (kd->procbase) {
609 		free(kd->procbase);
610 		kd->procbase = 0;
611 	}
612 }
613 
614 void *
615 _kvm_realloc(kvm_t *kd, void *p, size_t n)
616 {
617 	void *np = (void *)realloc(p, n);
618 
619 	if (np == 0) {
620 		free(p);
621 		_kvm_err(kd, kd->program, "out of memory");
622 	}
623 	return (np);
624 }
625 
626 #ifndef MAX
627 #define MAX(a, b) ((a) > (b) ? (a) : (b))
628 #endif
629 
630 /*
631  * Read in an argument vector from the user address space of process kp.
632  * addr if the user-space base address of narg null-terminated contiguous
633  * strings.  This is used to read in both the command arguments and
634  * environment strings.  Read at most maxcnt characters of strings.
635  */
636 static char **
637 kvm_argv(kvm_t *kd, const struct kinfo_proc *kp, u_long addr, int narg,
638     int maxcnt)
639 {
640 	char *np, *cp, *ep, *ap;
641 	u_long oaddr = -1;
642 	int len, cc;
643 	char **argv;
644 
645 	/*
646 	 * Check that there aren't an unreasonable number of arguments,
647 	 * and that the address is in user space.  Special test for
648 	 * VM_MIN_ADDRESS as it evaluates to zero, but is not a simple zero
649 	 * constant for some archs.  We cannot use the pre-processor here and
650 	 * for some archs the compiler would trigger a signedness warning.
651 	 */
652 	if (narg > 512 || addr + 1 < VM_MIN_ADDRESS + 1 || addr >= VM_MAXUSER_ADDRESS)
653 		return (0);
654 
655 	/*
656 	 * kd->argv : work space for fetching the strings from the target
657 	 *            process's space, and is converted for returning to caller
658 	 */
659 	if (kd->argv == 0) {
660 		/*
661 		 * Try to avoid reallocs.
662 		 */
663 		kd->argc = MAX(narg + 1, 32);
664 		kd->argv = (char **)_kvm_malloc(kd, kd->argc *
665 						sizeof(*kd->argv));
666 		if (kd->argv == 0)
667 			return (0);
668 	} else if (narg + 1 > kd->argc) {
669 		kd->argc = MAX(2 * kd->argc, narg + 1);
670 		kd->argv = (char **)_kvm_realloc(kd, kd->argv, kd->argc *
671 						sizeof(*kd->argv));
672 		if (kd->argv == 0)
673 			return (0);
674 	}
675 	/*
676 	 * kd->argspc : returned to user, this is where the kd->argv
677 	 *              arrays are left pointing to the collected strings.
678 	 */
679 	if (kd->argspc == 0) {
680 		kd->argspc = (char *)_kvm_malloc(kd, PAGE_SIZE);
681 		if (kd->argspc == 0)
682 			return (0);
683 		kd->arglen = PAGE_SIZE;
684 	}
685 	/*
686 	 * kd->argbuf : used to pull in pages from the target process.
687 	 *              the strings are copied out of here.
688 	 */
689 	if (kd->argbuf == 0) {
690 		kd->argbuf = (char *)_kvm_malloc(kd, PAGE_SIZE);
691 		if (kd->argbuf == 0)
692 			return (0);
693 	}
694 
695 	/* Pull in the target process'es argv vector */
696 	cc = sizeof(char *) * narg;
697 	if (kvm_uread(kd, kp, addr, (char *)kd->argv, cc) != cc)
698 		return (0);
699 	/*
700 	 * ap : saved start address of string we're working on in kd->argspc
701 	 * np : pointer to next place to write in kd->argspc
702 	 * len: length of data in kd->argspc
703 	 * argv: pointer to the argv vector that we are hunting around the
704 	 *       target process space for, and converting to addresses in
705 	 *       our address space (kd->argspc).
706 	 */
707 	ap = np = kd->argspc;
708 	argv = kd->argv;
709 	len = 0;
710 	/*
711 	 * Loop over pages, filling in the argument vector.
712 	 * Note that the argv strings could be pointing *anywhere* in
713 	 * the user address space and are no longer contiguous.
714 	 * Note that *argv is modified when we are going to fetch a string
715 	 * that crosses a page boundary.  We copy the next part of the string
716 	 * into to "np" and eventually convert the pointer.
717 	 */
718 	while (argv < kd->argv + narg && *argv != 0) {
719 
720 		/* get the address that the current argv string is on */
721 		addr = (u_long)*argv & ~(PAGE_SIZE - 1);
722 
723 		/* is it the same page as the last one? */
724 		if (addr != oaddr) {
725 			if (kvm_uread(kd, kp, addr, kd->argbuf, PAGE_SIZE) !=
726 			    PAGE_SIZE)
727 				return (0);
728 			oaddr = addr;
729 		}
730 
731 		/* offset within the page... kd->argbuf */
732 		addr = (u_long)*argv & (PAGE_SIZE - 1);
733 
734 		/* cp = start of string, cc = count of chars in this chunk */
735 		cp = kd->argbuf + addr;
736 		cc = PAGE_SIZE - addr;
737 
738 		/* dont get more than asked for by user process */
739 		if (maxcnt > 0 && cc > maxcnt - len)
740 			cc = maxcnt - len;
741 
742 		/* pointer to end of string if we found it in this page */
743 		ep = memchr(cp, '\0', cc);
744 		if (ep != 0)
745 			cc = ep - cp + 1;
746 		/*
747 		 * at this point, cc is the count of the chars that we are
748 		 * going to retrieve this time. we may or may not have found
749 		 * the end of it.  (ep points to the null if the end is known)
750 		 */
751 
752 		/* will we exceed the malloc/realloced buffer? */
753 		if (len + cc > kd->arglen) {
754 			int off;
755 			char **pp;
756 			char *op = kd->argspc;
757 
758 			kd->arglen *= 2;
759 			kd->argspc = (char *)_kvm_realloc(kd, kd->argspc,
760 							  kd->arglen);
761 			if (kd->argspc == 0)
762 				return (0);
763 			/*
764 			 * Adjust argv pointers in case realloc moved
765 			 * the string space.
766 			 */
767 			off = kd->argspc - op;
768 			for (pp = kd->argv; pp < argv; pp++)
769 				*pp += off;
770 			ap += off;
771 			np += off;
772 		}
773 		/* np = where to put the next part of the string in kd->argspc*/
774 		/* np is kinda redundant.. could use "kd->argspc + len" */
775 		memcpy(np, cp, cc);
776 		np += cc;	/* inc counters */
777 		len += cc;
778 
779 		/*
780 		 * if end of string found, set the *argv pointer to the
781 		 * saved beginning of string, and advance. argv points to
782 		 * somewhere in kd->argv..  This is initially relative
783 		 * to the target process, but when we close it off, we set
784 		 * it to point in our address space.
785 		 */
786 		if (ep != 0) {
787 			*argv++ = ap;
788 			ap = np;
789 		} else {
790 			/* update the address relative to the target process */
791 			*argv += cc;
792 		}
793 
794 		if (maxcnt > 0 && len >= maxcnt) {
795 			/*
796 			 * We're stopping prematurely.  Terminate the
797 			 * current string.
798 			 */
799 			if (ep == 0) {
800 				*np = '\0';
801 				*argv++ = ap;
802 			}
803 			break;
804 		}
805 	}
806 	/* Make sure argv is terminated. */
807 	*argv = 0;
808 	return (kd->argv);
809 }
810 
811 static void
812 ps_str_a(struct ps_strings *p, u_long *addr, int *n)
813 {
814 	*addr = (u_long)p->ps_argvstr;
815 	*n = p->ps_nargvstr;
816 }
817 
818 static void
819 ps_str_e (struct ps_strings *p, u_long *addr, int *n)
820 {
821 	*addr = (u_long)p->ps_envstr;
822 	*n = p->ps_nenvstr;
823 }
824 
825 /*
826  * Determine if the proc indicated by p is still active.
827  * This test is not 100% foolproof in theory, but chances of
828  * being wrong are very low.
829  */
830 static int
831 proc_verify(const struct kinfo_proc *curkp)
832 {
833 	struct kinfo_proc newkp;
834 	int mib[4];
835 	size_t len;
836 
837 	mib[0] = CTL_KERN;
838 	mib[1] = KERN_PROC;
839 	mib[2] = KERN_PROC_PID;
840 	mib[3] = curkp->ki_pid;
841 	len = sizeof(newkp);
842 	if (sysctl(mib, 4, &newkp, &len, NULL, 0) == -1)
843 		return (0);
844 	return (curkp->ki_pid == newkp.ki_pid &&
845 	    (newkp.ki_stat != SZOMB || curkp->ki_stat == SZOMB));
846 }
847 
848 static char **
849 kvm_doargv(kvm_t *kd, const struct kinfo_proc *kp, int nchr,
850     void (*info)(struct ps_strings *, u_long *, int *))
851 {
852 	char **ap;
853 	u_long addr;
854 	int cnt;
855 	static struct ps_strings arginfo;
856 	static u_long ps_strings;
857 	size_t len;
858 
859 	if (ps_strings == 0) {
860 		len = sizeof(ps_strings);
861 		if (sysctlbyname("kern.ps_strings", &ps_strings, &len, NULL,
862 		    0) == -1)
863 			ps_strings = PS_STRINGS;
864 	}
865 
866 	/*
867 	 * Pointers are stored at the top of the user stack.
868 	 */
869 	if (kp->ki_stat == SZOMB ||
870 	    kvm_uread(kd, kp, ps_strings, (char *)&arginfo,
871 		      sizeof(arginfo)) != sizeof(arginfo))
872 		return (0);
873 
874 	(*info)(&arginfo, &addr, &cnt);
875 	if (cnt == 0)
876 		return (0);
877 	ap = kvm_argv(kd, kp, addr, cnt, nchr);
878 	/*
879 	 * For live kernels, make sure this process didn't go away.
880 	 */
881 	if (ap != 0 && ISALIVE(kd) && !proc_verify(kp))
882 		ap = 0;
883 	return (ap);
884 }
885 
886 /*
887  * Get the command args.  This code is now machine independent.
888  */
889 char **
890 kvm_getargv(kvm_t *kd, const struct kinfo_proc *kp, int nchr)
891 {
892 	int oid[4];
893 	int i;
894 	size_t bufsz;
895 	static unsigned long buflen;
896 	static char *buf, *p;
897 	static char **bufp;
898 	static int argc;
899 
900 	if (!ISALIVE(kd)) {
901 		_kvm_err(kd, kd->program,
902 		    "cannot read user space from dead kernel");
903 		return (0);
904 	}
905 
906 	if (!buflen) {
907 		bufsz = sizeof(buflen);
908 		i = sysctlbyname("kern.ps_arg_cache_limit",
909 		    &buflen, &bufsz, NULL, 0);
910 		if (i == -1) {
911 			buflen = 0;
912 		} else {
913 			buf = malloc(buflen);
914 			if (buf == NULL)
915 				buflen = 0;
916 			argc = 32;
917 			bufp = malloc(sizeof(char *) * argc);
918 		}
919 	}
920 	if (buf != NULL) {
921 		oid[0] = CTL_KERN;
922 		oid[1] = KERN_PROC;
923 		oid[2] = KERN_PROC_ARGS;
924 		oid[3] = kp->ki_pid;
925 		bufsz = buflen;
926 		i = sysctl(oid, 4, buf, &bufsz, 0, 0);
927 		if (i == 0 && bufsz > 0) {
928 			i = 0;
929 			p = buf;
930 			do {
931 				bufp[i++] = p;
932 				p += strlen(p) + 1;
933 				if (i >= argc) {
934 					argc += argc;
935 					bufp = realloc(bufp,
936 					    sizeof(char *) * argc);
937 				}
938 			} while (p < buf + bufsz);
939 			bufp[i++] = 0;
940 			return (bufp);
941 		}
942 	}
943 	if (kp->ki_flag & P_SYSTEM)
944 		return (NULL);
945 	return (kvm_doargv(kd, kp, nchr, ps_str_a));
946 }
947 
948 char **
949 kvm_getenvv(kvm_t *kd, const struct kinfo_proc *kp, int nchr)
950 {
951 	return (kvm_doargv(kd, kp, nchr, ps_str_e));
952 }
953 
954 /*
955  * Read from user space.  The user context is given by p.
956  */
957 ssize_t
958 kvm_uread(kvm_t *kd, const struct kinfo_proc *kp, u_long uva, char *buf,
959 	size_t len)
960 {
961 	char *cp;
962 	char procfile[MAXPATHLEN];
963 	ssize_t amount;
964 	int fd;
965 
966 	if (!ISALIVE(kd)) {
967 		_kvm_err(kd, kd->program,
968 		    "cannot read user space from dead kernel");
969 		return (0);
970 	}
971 
972 	sprintf(procfile, "/proc/%d/mem", kp->ki_pid);
973 	fd = open(procfile, O_RDONLY, 0);
974 	if (fd < 0) {
975 		_kvm_err(kd, kd->program, "cannot open %s", procfile);
976 		return (0);
977 	}
978 
979 	cp = buf;
980 	while (len > 0) {
981 		errno = 0;
982 		if (lseek(fd, (off_t)uva, 0) == -1 && errno != 0) {
983 			_kvm_err(kd, kd->program, "invalid address (%lx) in %s",
984 			    uva, procfile);
985 			break;
986 		}
987 		amount = read(fd, cp, len);
988 		if (amount < 0) {
989 			_kvm_syserr(kd, kd->program, "error reading %s",
990 			    procfile);
991 			break;
992 		}
993 		if (amount == 0) {
994 			_kvm_err(kd, kd->program, "EOF reading %s", procfile);
995 			break;
996 		}
997 		cp += amount;
998 		uva += amount;
999 		len -= amount;
1000 	}
1001 
1002 	close(fd);
1003 	return ((ssize_t)(cp - buf));
1004 }
1005