xref: /dragonfly/sys/kern/kern_proc.c (revision 31c7ac8b)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/sysctl.h>
34 #include <sys/malloc.h>
35 #include <sys/proc.h>
36 #include <sys/vnode.h>
37 #include <sys/jail.h>
38 #include <sys/filedesc.h>
39 #include <sys/tty.h>
40 #include <sys/dsched.h>
41 #include <sys/signalvar.h>
42 #include <sys/spinlock.h>
43 #include <sys/random.h>
44 #include <sys/vnode.h>
45 #include <vm/vm.h>
46 #include <sys/lock.h>
47 #include <vm/pmap.h>
48 #include <vm/vm_map.h>
49 #include <sys/user.h>
50 #include <machine/smp.h>
51 
52 #include <sys/refcount.h>
53 #include <sys/spinlock2.h>
54 #include <sys/mplock2.h>
55 
56 /*
57  * Hash table size must be a power of two and is not currently dynamically
58  * sized.  There is a trade-off between the linear scans which must iterate
59  * all HSIZE elements and the number of elements which might accumulate
60  * within each hash chain.
61  */
62 #define ALLPROC_HSIZE	256
63 #define ALLPROC_HMASK	(ALLPROC_HSIZE - 1)
64 #define ALLPROC_HASH(pid)	(pid & ALLPROC_HMASK)
65 #define PGRP_HASH(pid)	(pid & ALLPROC_HMASK)
66 #define SESS_HASH(pid)	(pid & ALLPROC_HMASK)
67 
68 LIST_HEAD(pidhashhead, proc);
69 
70 static MALLOC_DEFINE(M_PGRP, "pgrp", "process group header");
71 MALLOC_DEFINE(M_SESSION, "session", "session header");
72 MALLOC_DEFINE(M_PROC, "proc", "Proc structures");
73 MALLOC_DEFINE(M_LWP, "lwp", "lwp structures");
74 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures");
75 
76 int ps_showallprocs = 1;
77 static int ps_showallthreads = 1;
78 SYSCTL_INT(_security, OID_AUTO, ps_showallprocs, CTLFLAG_RW,
79     &ps_showallprocs, 0,
80     "Unprivileged processes can see processes with different UID/GID");
81 SYSCTL_INT(_security, OID_AUTO, ps_showallthreads, CTLFLAG_RW,
82     &ps_showallthreads, 0,
83     "Unprivileged processes can see kernel threads");
84 
85 static void orphanpg(struct pgrp *pg);
86 static void proc_makepid(struct proc *p, int random_offset);
87 
88 /*
89  * Other process lists
90  */
91 static struct lwkt_token proc_tokens[ALLPROC_HSIZE];
92 static struct proclist allprocs[ALLPROC_HSIZE];	/* locked by proc_tokens */
93 static struct pgrplist allpgrps[ALLPROC_HSIZE];	/* locked by proc_tokens */
94 static struct sesslist allsessn[ALLPROC_HSIZE];	/* locked by proc_tokens */
95 
96 /*
97  * Random component to nextpid generation.  We mix in a random factor to make
98  * it a little harder to predict.  We sanity check the modulus value to avoid
99  * doing it in critical paths.  Don't let it be too small or we pointlessly
100  * waste randomness entropy, and don't let it be impossibly large.  Using a
101  * modulus that is too big causes a LOT more process table scans and slows
102  * down fork processing as the pidchecked caching is defeated.
103  */
104 static int randompid = 0;
105 
106 /*
107  * No requirements.
108  */
109 static int
110 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
111 {
112 	int error, pid;
113 
114 	pid = randompid;
115 	error = sysctl_handle_int(oidp, &pid, 0, req);
116 	if (error || !req->newptr)
117 		return (error);
118 	if (pid < 0 || pid > PID_MAX - 100)     /* out of range */
119 		pid = PID_MAX - 100;
120 	else if (pid < 2)                       /* NOP */
121 		pid = 0;
122 	else if (pid < 100)                     /* Make it reasonable */
123 		pid = 100;
124 	randompid = pid;
125 	return (error);
126 }
127 
128 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
129 	    0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
130 
131 /*
132  * Initialize global process hashing structures.
133  *
134  * These functions are ONLY called from the low level boot code and do
135  * not lock their operations.
136  */
137 void
138 procinit(void)
139 {
140 	u_long i;
141 
142 	for (i = 0; i < ALLPROC_HSIZE; ++i) {
143 		LIST_INIT(&allprocs[i]);
144 		LIST_INIT(&allsessn[i]);
145 		LIST_INIT(&allpgrps[i]);
146 		lwkt_token_init(&proc_tokens[i], "allproc");
147 	}
148 	lwkt_init();
149 	uihashinit();
150 }
151 
152 void
153 procinsertinit(struct proc *p)
154 {
155 	LIST_INSERT_HEAD(&allprocs[ALLPROC_HASH(p->p_pid)], p, p_list);
156 }
157 
158 void
159 pgrpinsertinit(struct pgrp *pg)
160 {
161 	LIST_INSERT_HEAD(&allpgrps[ALLPROC_HASH(pg->pg_id)], pg, pg_list);
162 }
163 
164 void
165 sessinsertinit(struct session *sess)
166 {
167 	LIST_INSERT_HEAD(&allsessn[ALLPROC_HASH(sess->s_sid)], sess, s_list);
168 }
169 
170 /*
171  * Process hold/release support functions.  Called via the PHOLD(),
172  * PRELE(), and PSTALL() macros.
173  *
174  * p->p_lock is a simple hold count with a waiting interlock.  No wakeup()
175  * is issued unless someone is actually waiting for the process.
176  *
177  * Most holds are short-term, allowing a process scan or other similar
178  * operation to access a proc structure without it getting ripped out from
179  * under us.  procfs and process-list sysctl ops also use the hold function
180  * interlocked with various p_flags to keep the vmspace intact when reading
181  * or writing a user process's address space.
182  *
183  * There are two situations where a hold count can be longer.  Exiting lwps
184  * hold the process until the lwp is reaped, and the parent will hold the
185  * child during vfork()/exec() sequences while the child is marked P_PPWAIT.
186  *
187  * The kernel waits for the hold count to drop to 0 (or 1 in some cases) at
188  * various critical points in the fork/exec and exit paths before proceeding.
189  */
190 #define PLOCK_ZOMB	0x20000000
191 #define PLOCK_WAITING	0x40000000
192 #define PLOCK_MASK	0x1FFFFFFF
193 
194 void
195 pstall(struct proc *p, const char *wmesg, int count)
196 {
197 	int o;
198 	int n;
199 
200 	for (;;) {
201 		o = p->p_lock;
202 		cpu_ccfence();
203 		if ((o & PLOCK_MASK) <= count)
204 			break;
205 		n = o | PLOCK_WAITING;
206 		tsleep_interlock(&p->p_lock, 0);
207 
208 		/*
209 		 * If someone is trying to single-step the process during
210 		 * an exec or an exit they can deadlock us because procfs
211 		 * sleeps with the process held.
212 		 */
213 		if (p->p_stops) {
214 			if (p->p_flags & P_INEXEC) {
215 				wakeup(&p->p_stype);
216 			} else if (p->p_flags & P_POSTEXIT) {
217 				spin_lock(&p->p_spin);
218 				p->p_stops = 0;
219 				p->p_step = 0;
220 				spin_unlock(&p->p_spin);
221 				wakeup(&p->p_stype);
222 			}
223 		}
224 
225 		if (atomic_cmpset_int(&p->p_lock, o, n)) {
226 			tsleep(&p->p_lock, PINTERLOCKED, wmesg, 0);
227 		}
228 	}
229 }
230 
231 void
232 phold(struct proc *p)
233 {
234 	atomic_add_int(&p->p_lock, 1);
235 }
236 
237 /*
238  * WARNING!  On last release (p) can become instantly invalid due to
239  *	     MP races.
240  */
241 void
242 prele(struct proc *p)
243 {
244 	int o;
245 	int n;
246 
247 	/*
248 	 * Fast path
249 	 */
250 	if (atomic_cmpset_int(&p->p_lock, 1, 0))
251 		return;
252 
253 	/*
254 	 * Slow path
255 	 */
256 	for (;;) {
257 		o = p->p_lock;
258 		KKASSERT((o & PLOCK_MASK) > 0);
259 		cpu_ccfence();
260 		n = (o - 1) & ~PLOCK_WAITING;
261 		if (atomic_cmpset_int(&p->p_lock, o, n)) {
262 			if (o & PLOCK_WAITING)
263 				wakeup(&p->p_lock);
264 			break;
265 		}
266 	}
267 }
268 
269 /*
270  * Hold and flag serialized for zombie reaping purposes.
271  *
272  * This function will fail if it has to block, returning non-zero with
273  * neither the flag set or the hold count bumped.  Note that we must block
274  * without holding a ref, meaning that the caller must ensure that (p)
275  * remains valid through some other interlock (typically on its parent
276  * process's p_token).
277  *
278  * Zero is returned on success.  The hold count will be incremented and
279  * the serialization flag acquired.  Note that serialization is only against
280  * other pholdzomb() calls, not against phold() calls.
281  */
282 int
283 pholdzomb(struct proc *p)
284 {
285 	int o;
286 	int n;
287 
288 	/*
289 	 * Fast path
290 	 */
291 	if (atomic_cmpset_int(&p->p_lock, 0, PLOCK_ZOMB | 1))
292 		return(0);
293 
294 	/*
295 	 * Slow path
296 	 */
297 	for (;;) {
298 		o = p->p_lock;
299 		cpu_ccfence();
300 		if ((o & PLOCK_ZOMB) == 0) {
301 			n = (o + 1) | PLOCK_ZOMB;
302 			if (atomic_cmpset_int(&p->p_lock, o, n))
303 				return(0);
304 		} else {
305 			KKASSERT((o & PLOCK_MASK) > 0);
306 			n = o | PLOCK_WAITING;
307 			tsleep_interlock(&p->p_lock, 0);
308 			if (atomic_cmpset_int(&p->p_lock, o, n)) {
309 				tsleep(&p->p_lock, PINTERLOCKED, "phldz", 0);
310 				/* (p) can be ripped out at this point */
311 				return(1);
312 			}
313 		}
314 	}
315 }
316 
317 /*
318  * Release PLOCK_ZOMB and the hold count, waking up any waiters.
319  *
320  * WARNING!  On last release (p) can become instantly invalid due to
321  *	     MP races.
322  */
323 void
324 prelezomb(struct proc *p)
325 {
326 	int o;
327 	int n;
328 
329 	/*
330 	 * Fast path
331 	 */
332 	if (atomic_cmpset_int(&p->p_lock, PLOCK_ZOMB | 1, 0))
333 		return;
334 
335 	/*
336 	 * Slow path
337 	 */
338 	KKASSERT(p->p_lock & PLOCK_ZOMB);
339 	for (;;) {
340 		o = p->p_lock;
341 		KKASSERT((o & PLOCK_MASK) > 0);
342 		cpu_ccfence();
343 		n = (o - 1) & ~(PLOCK_ZOMB | PLOCK_WAITING);
344 		if (atomic_cmpset_int(&p->p_lock, o, n)) {
345 			if (o & PLOCK_WAITING)
346 				wakeup(&p->p_lock);
347 			break;
348 		}
349 	}
350 }
351 
352 /*
353  * Is p an inferior of the current process?
354  *
355  * No requirements.
356  */
357 int
358 inferior(struct proc *p)
359 {
360 	struct proc *p2;
361 
362 	PHOLD(p);
363 	lwkt_gettoken_shared(&p->p_token);
364 	while (p != curproc) {
365 		if (p->p_pid == 0) {
366 			lwkt_reltoken(&p->p_token);
367 			return (0);
368 		}
369 		p2 = p->p_pptr;
370 		PHOLD(p2);
371 		lwkt_reltoken(&p->p_token);
372 		PRELE(p);
373 		lwkt_gettoken_shared(&p2->p_token);
374 		p = p2;
375 	}
376 	lwkt_reltoken(&p->p_token);
377 	PRELE(p);
378 
379 	return (1);
380 }
381 
382 /*
383  * Locate a process by number.  The returned process will be referenced and
384  * must be released with PRELE().
385  *
386  * No requirements.
387  */
388 struct proc *
389 pfind(pid_t pid)
390 {
391 	struct proc *p = curproc;
392 	int n;
393 
394 	/*
395 	 * Shortcut the current process
396 	 */
397 	if (p && p->p_pid == pid) {
398 		PHOLD(p);
399 		return (p);
400 	}
401 
402 	/*
403 	 * Otherwise find it in the hash table.
404 	 */
405 	n = ALLPROC_HASH(pid);
406 
407 	lwkt_gettoken_shared(&proc_tokens[n]);
408 	LIST_FOREACH(p, &allprocs[n], p_list) {
409 		if (p->p_stat == SZOMB)
410 			continue;
411 		if (p->p_pid == pid) {
412 			PHOLD(p);
413 			lwkt_reltoken(&proc_tokens[n]);
414 			return (p);
415 		}
416 	}
417 	lwkt_reltoken(&proc_tokens[n]);
418 
419 	return (NULL);
420 }
421 
422 /*
423  * Locate a process by number.  The returned process is NOT referenced.
424  * The result will not be stable and is typically only used to validate
425  * against a process that the caller has in-hand.
426  *
427  * No requirements.
428  */
429 struct proc *
430 pfindn(pid_t pid)
431 {
432 	struct proc *p = curproc;
433 	int n;
434 
435 	/*
436 	 * Shortcut the current process
437 	 */
438 	if (p && p->p_pid == pid)
439 		return (p);
440 
441 	/*
442 	 * Otherwise find it in the hash table.
443 	 */
444 	n = ALLPROC_HASH(pid);
445 
446 	lwkt_gettoken_shared(&proc_tokens[n]);
447 	LIST_FOREACH(p, &allprocs[n], p_list) {
448 		if (p->p_stat == SZOMB)
449 			continue;
450 		if (p->p_pid == pid) {
451 			lwkt_reltoken(&proc_tokens[n]);
452 			return (p);
453 		}
454 	}
455 	lwkt_reltoken(&proc_tokens[n]);
456 
457 	return (NULL);
458 }
459 
460 /*
461  * Locate a process on the zombie list.  Return a process or NULL.
462  * The returned process will be referenced and the caller must release
463  * it with PRELE().
464  *
465  * No other requirements.
466  */
467 struct proc *
468 zpfind(pid_t pid)
469 {
470 	struct proc *p = curproc;
471 	int n;
472 
473 	/*
474 	 * Shortcut the current process
475 	 */
476 	if (p && p->p_pid == pid) {
477 		PHOLD(p);
478 		return (p);
479 	}
480 
481 	/*
482 	 * Otherwise find it in the hash table.
483 	 */
484 	n = ALLPROC_HASH(pid);
485 
486 	lwkt_gettoken_shared(&proc_tokens[n]);
487 	LIST_FOREACH(p, &allprocs[n], p_list) {
488 		if (p->p_stat != SZOMB)
489 			continue;
490 		if (p->p_pid == pid) {
491 			PHOLD(p);
492 			lwkt_reltoken(&proc_tokens[n]);
493 			return (p);
494 		}
495 	}
496 	lwkt_reltoken(&proc_tokens[n]);
497 
498 	return (NULL);
499 }
500 
501 
502 void
503 pgref(struct pgrp *pgrp)
504 {
505 	refcount_acquire(&pgrp->pg_refs);
506 }
507 
508 void
509 pgrel(struct pgrp *pgrp)
510 {
511 	int count;
512 	int n;
513 
514 	n = PGRP_HASH(pgrp->pg_id);
515 	for (;;) {
516 		count = pgrp->pg_refs;
517 		cpu_ccfence();
518 		KKASSERT(count > 0);
519 		if (count == 1) {
520 			lwkt_gettoken(&proc_tokens[n]);
521 			if (atomic_cmpset_int(&pgrp->pg_refs, 1, 0))
522 				break;
523 			lwkt_reltoken(&proc_tokens[n]);
524 			/* retry */
525 		} else {
526 			if (atomic_cmpset_int(&pgrp->pg_refs, count, count - 1))
527 				return;
528 			/* retry */
529 		}
530 	}
531 
532 	/*
533 	 * Successful 1->0 transition, pghash_spin is held.
534 	 */
535 	LIST_REMOVE(pgrp, pg_list);
536 
537 	/*
538 	 * Reset any sigio structures pointing to us as a result of
539 	 * F_SETOWN with our pgid.
540 	 */
541 	funsetownlst(&pgrp->pg_sigiolst);
542 
543 	if (pgrp->pg_session->s_ttyp != NULL &&
544 	    pgrp->pg_session->s_ttyp->t_pgrp == pgrp) {
545 		pgrp->pg_session->s_ttyp->t_pgrp = NULL;
546 	}
547 	lwkt_reltoken(&proc_tokens[n]);
548 
549 	sess_rele(pgrp->pg_session);
550 	kfree(pgrp, M_PGRP);
551 }
552 
553 /*
554  * Locate a process group by number.  The returned process group will be
555  * referenced w/pgref() and must be released with pgrel() (or assigned
556  * somewhere if you wish to keep the reference).
557  *
558  * No requirements.
559  */
560 struct pgrp *
561 pgfind(pid_t pgid)
562 {
563 	struct pgrp *pgrp;
564 	int n;
565 
566 	n = PGRP_HASH(pgid);
567 	lwkt_gettoken_shared(&proc_tokens[n]);
568 
569 	LIST_FOREACH(pgrp, &allpgrps[n], pg_list) {
570 		if (pgrp->pg_id == pgid) {
571 			refcount_acquire(&pgrp->pg_refs);
572 			lwkt_reltoken(&proc_tokens[n]);
573 			return (pgrp);
574 		}
575 	}
576 	lwkt_reltoken(&proc_tokens[n]);
577 	return (NULL);
578 }
579 
580 /*
581  * Move p to a new or existing process group (and session)
582  *
583  * No requirements.
584  */
585 int
586 enterpgrp(struct proc *p, pid_t pgid, int mksess)
587 {
588 	struct pgrp *pgrp;
589 	struct pgrp *opgrp;
590 	int error;
591 
592 	pgrp = pgfind(pgid);
593 
594 	KASSERT(pgrp == NULL || !mksess,
595 		("enterpgrp: setsid into non-empty pgrp"));
596 	KASSERT(!SESS_LEADER(p),
597 		("enterpgrp: session leader attempted setpgrp"));
598 
599 	if (pgrp == NULL) {
600 		pid_t savepid = p->p_pid;
601 		struct proc *np;
602 		int n;
603 
604 		/*
605 		 * new process group
606 		 */
607 		KASSERT(p->p_pid == pgid,
608 			("enterpgrp: new pgrp and pid != pgid"));
609 		pgrp = kmalloc(sizeof(struct pgrp), M_PGRP, M_WAITOK | M_ZERO);
610 		pgrp->pg_id = pgid;
611 		LIST_INIT(&pgrp->pg_members);
612 		pgrp->pg_jobc = 0;
613 		SLIST_INIT(&pgrp->pg_sigiolst);
614 		lwkt_token_init(&pgrp->pg_token, "pgrp_token");
615 		refcount_init(&pgrp->pg_refs, 1);
616 		lockinit(&pgrp->pg_lock, "pgwt", 0, 0);
617 
618 		n = PGRP_HASH(pgid);
619 
620 		if ((np = pfindn(savepid)) == NULL || np != p) {
621 			lwkt_reltoken(&proc_tokens[n]);
622 			error = ESRCH;
623 			kfree(pgrp, M_PGRP);
624 			goto fatal;
625 		}
626 
627 		lwkt_gettoken(&proc_tokens[n]);
628 		if (mksess) {
629 			struct session *sess;
630 
631 			/*
632 			 * new session
633 			 */
634 			sess = kmalloc(sizeof(struct session), M_SESSION,
635 				       M_WAITOK | M_ZERO);
636 			lwkt_gettoken(&p->p_token);
637 			sess->s_leader = p;
638 			sess->s_sid = p->p_pid;
639 			sess->s_count = 1;
640 			sess->s_ttyvp = NULL;
641 			sess->s_ttyp = NULL;
642 			bcopy(p->p_session->s_login, sess->s_login,
643 			      sizeof(sess->s_login));
644 			pgrp->pg_session = sess;
645 			KASSERT(p == curproc,
646 				("enterpgrp: mksession and p != curproc"));
647 			p->p_flags &= ~P_CONTROLT;
648 			LIST_INSERT_HEAD(&allsessn[n], sess, s_list);
649 			lwkt_reltoken(&p->p_token);
650 		} else {
651 			lwkt_gettoken(&p->p_token);
652 			pgrp->pg_session = p->p_session;
653 			sess_hold(pgrp->pg_session);
654 			lwkt_reltoken(&p->p_token);
655 		}
656 		LIST_INSERT_HEAD(&allpgrps[n], pgrp, pg_list);
657 
658 		lwkt_reltoken(&proc_tokens[n]);
659 	} else if (pgrp == p->p_pgrp) {
660 		pgrel(pgrp);
661 		goto done;
662 	} /* else pgfind() referenced the pgrp */
663 
664 	lwkt_gettoken(&pgrp->pg_token);
665 	lwkt_gettoken(&p->p_token);
666 
667 	/*
668 	 * Replace p->p_pgrp, handling any races that occur.
669 	 */
670 	while ((opgrp = p->p_pgrp) != NULL) {
671 		pgref(opgrp);
672 		lwkt_gettoken(&opgrp->pg_token);
673 		if (opgrp != p->p_pgrp) {
674 			lwkt_reltoken(&opgrp->pg_token);
675 			pgrel(opgrp);
676 			continue;
677 		}
678 		LIST_REMOVE(p, p_pglist);
679 		break;
680 	}
681 	p->p_pgrp = pgrp;
682 	LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
683 
684 	/*
685 	 * Adjust eligibility of affected pgrps to participate in job control.
686 	 * Increment eligibility counts before decrementing, otherwise we
687 	 * could reach 0 spuriously during the first call.
688 	 */
689 	fixjobc(p, pgrp, 1);
690 	if (opgrp) {
691 		fixjobc(p, opgrp, 0);
692 		lwkt_reltoken(&opgrp->pg_token);
693 		pgrel(opgrp);	/* manual pgref */
694 		pgrel(opgrp);	/* p->p_pgrp ref */
695 	}
696 	lwkt_reltoken(&p->p_token);
697 	lwkt_reltoken(&pgrp->pg_token);
698 done:
699 	error = 0;
700 fatal:
701 	return (error);
702 }
703 
704 /*
705  * Remove process from process group
706  *
707  * No requirements.
708  */
709 int
710 leavepgrp(struct proc *p)
711 {
712 	struct pgrp *pg = p->p_pgrp;
713 
714 	lwkt_gettoken(&p->p_token);
715 	while ((pg = p->p_pgrp) != NULL) {
716 		pgref(pg);
717 		lwkt_gettoken(&pg->pg_token);
718 		if (p->p_pgrp != pg) {
719 			lwkt_reltoken(&pg->pg_token);
720 			pgrel(pg);
721 			continue;
722 		}
723 		p->p_pgrp = NULL;
724 		LIST_REMOVE(p, p_pglist);
725 		lwkt_reltoken(&pg->pg_token);
726 		pgrel(pg);	/* manual pgref */
727 		pgrel(pg);	/* p->p_pgrp ref */
728 		break;
729 	}
730 	lwkt_reltoken(&p->p_token);
731 
732 	return (0);
733 }
734 
735 /*
736  * Adjust the ref count on a session structure.  When the ref count falls to
737  * zero the tty is disassociated from the session and the session structure
738  * is freed.  Note that tty assocation is not itself ref-counted.
739  *
740  * No requirements.
741  */
742 void
743 sess_hold(struct session *sp)
744 {
745 	atomic_add_int(&sp->s_count, 1);
746 }
747 
748 /*
749  * No requirements.
750  */
751 void
752 sess_rele(struct session *sess)
753 {
754 	struct tty *tp;
755 	int count;
756 	int n;
757 
758 	n = SESS_HASH(sess->s_sid);
759 	for (;;) {
760 		count = sess->s_count;
761 		cpu_ccfence();
762 		KKASSERT(count > 0);
763 		if (count == 1) {
764 			lwkt_gettoken(&tty_token);
765 			lwkt_gettoken(&proc_tokens[n]);
766 			if (atomic_cmpset_int(&sess->s_count, 1, 0))
767 				break;
768 			lwkt_reltoken(&proc_tokens[n]);
769 			lwkt_reltoken(&tty_token);
770 			/* retry */
771 		} else {
772 			if (atomic_cmpset_int(&sess->s_count, count, count - 1))
773 				return;
774 			/* retry */
775 		}
776 	}
777 
778 	/*
779 	 * Successful 1->0 transition and tty_token is held.
780 	 */
781 	LIST_REMOVE(sess, s_list);
782 
783 	if (sess->s_ttyp && sess->s_ttyp->t_session) {
784 #ifdef TTY_DO_FULL_CLOSE
785 		/* FULL CLOSE, see ttyclearsession() */
786 		KKASSERT(sess->s_ttyp->t_session == sess);
787 		sess->s_ttyp->t_session = NULL;
788 #else
789 		/* HALF CLOSE, see ttyclearsession() */
790 		if (sess->s_ttyp->t_session == sess)
791 			sess->s_ttyp->t_session = NULL;
792 #endif
793 	}
794 	if ((tp = sess->s_ttyp) != NULL) {
795 		sess->s_ttyp = NULL;
796 		ttyunhold(tp);
797 	}
798 	lwkt_reltoken(&proc_tokens[n]);
799 	lwkt_reltoken(&tty_token);
800 
801 	kfree(sess, M_SESSION);
802 }
803 
804 /*
805  * Adjust pgrp jobc counters when specified process changes process group.
806  * We count the number of processes in each process group that "qualify"
807  * the group for terminal job control (those with a parent in a different
808  * process group of the same session).  If that count reaches zero, the
809  * process group becomes orphaned.  Check both the specified process'
810  * process group and that of its children.
811  * entering == 0 => p is leaving specified group.
812  * entering == 1 => p is entering specified group.
813  *
814  * No requirements.
815  */
816 void
817 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
818 {
819 	struct pgrp *hispgrp;
820 	struct session *mysession;
821 	struct proc *np;
822 
823 	/*
824 	 * Check p's parent to see whether p qualifies its own process
825 	 * group; if so, adjust count for p's process group.
826 	 */
827 	lwkt_gettoken(&p->p_token);	/* p_children scan */
828 	lwkt_gettoken(&pgrp->pg_token);
829 
830 	mysession = pgrp->pg_session;
831 	if ((hispgrp = p->p_pptr->p_pgrp) != pgrp &&
832 	    hispgrp->pg_session == mysession) {
833 		if (entering)
834 			pgrp->pg_jobc++;
835 		else if (--pgrp->pg_jobc == 0)
836 			orphanpg(pgrp);
837 	}
838 
839 	/*
840 	 * Check this process' children to see whether they qualify
841 	 * their process groups; if so, adjust counts for children's
842 	 * process groups.
843 	 */
844 	LIST_FOREACH(np, &p->p_children, p_sibling) {
845 		PHOLD(np);
846 		lwkt_gettoken(&np->p_token);
847 		if ((hispgrp = np->p_pgrp) != pgrp &&
848 		    hispgrp->pg_session == mysession &&
849 		    np->p_stat != SZOMB) {
850 			pgref(hispgrp);
851 			lwkt_gettoken(&hispgrp->pg_token);
852 			if (entering)
853 				hispgrp->pg_jobc++;
854 			else if (--hispgrp->pg_jobc == 0)
855 				orphanpg(hispgrp);
856 			lwkt_reltoken(&hispgrp->pg_token);
857 			pgrel(hispgrp);
858 		}
859 		lwkt_reltoken(&np->p_token);
860 		PRELE(np);
861 	}
862 	KKASSERT(pgrp->pg_refs > 0);
863 	lwkt_reltoken(&pgrp->pg_token);
864 	lwkt_reltoken(&p->p_token);
865 }
866 
867 /*
868  * A process group has become orphaned;
869  * if there are any stopped processes in the group,
870  * hang-up all process in that group.
871  *
872  * The caller must hold pg_token.
873  */
874 static void
875 orphanpg(struct pgrp *pg)
876 {
877 	struct proc *p;
878 
879 	LIST_FOREACH(p, &pg->pg_members, p_pglist) {
880 		if (p->p_stat == SSTOP) {
881 			LIST_FOREACH(p, &pg->pg_members, p_pglist) {
882 				ksignal(p, SIGHUP);
883 				ksignal(p, SIGCONT);
884 			}
885 			return;
886 		}
887 	}
888 }
889 
890 /*
891  * Add a new process to the allproc list and the PID hash.  This
892  * also assigns a pid to the new process.
893  *
894  * No requirements.
895  */
896 void
897 proc_add_allproc(struct proc *p)
898 {
899 	int random_offset;
900 
901 	if ((random_offset = randompid) != 0) {
902 		read_random(&random_offset, sizeof(random_offset));
903 		random_offset = (random_offset & 0x7FFFFFFF) % randompid;
904 	}
905 	proc_makepid(p, random_offset);
906 }
907 
908 /*
909  * Calculate a new process pid.  This function is integrated into
910  * proc_add_allproc() to guarentee that the new pid is not reused before
911  * the new process can be added to the allproc list.
912  *
913  * p_pid is assigned and the process is added to the allproc hash table
914  */
915 static
916 void
917 proc_makepid(struct proc *p, int random_offset)
918 {
919 	static pid_t nextpid;	/* heuristic, allowed to race */
920 	struct pgrp *pg;
921 	struct proc *ps;
922 	struct session *sess;
923 	pid_t base;
924 	int n;
925 
926 	/*
927 	 * Calculate a hash index and find an unused process id within
928 	 * the table, looping if we cannot find one.
929 	 */
930 	if (random_offset)
931 		atomic_add_int(&nextpid, random_offset);
932 retry:
933 	base = atomic_fetchadd_int(&nextpid, 1) + 1;
934 	if (base >= PID_MAX) {
935 		base = base % PID_MAX;
936 		if (base < 100)
937 			base += 100;
938 	}
939 	n = ALLPROC_HASH(base);
940 	lwkt_gettoken(&proc_tokens[n]);
941 
942 	LIST_FOREACH(ps, &allprocs[n], p_list) {
943 		if (ps->p_pid == base) {
944 			base += ALLPROC_HSIZE;
945 			if (base >= PID_MAX) {
946 				lwkt_reltoken(&proc_tokens[n]);
947 				goto retry;
948 			}
949 		}
950 	}
951 	LIST_FOREACH(pg, &allpgrps[n], pg_list) {
952 		if (pg->pg_id == base) {
953 			base += ALLPROC_HSIZE;
954 			if (base >= PID_MAX) {
955 				lwkt_reltoken(&proc_tokens[n]);
956 				goto retry;
957 			}
958 		}
959 	}
960 	LIST_FOREACH(sess, &allsessn[n], s_list) {
961 		if (sess->s_sid == base) {
962 			base += ALLPROC_HSIZE;
963 			if (base >= PID_MAX) {
964 				lwkt_reltoken(&proc_tokens[n]);
965 				goto retry;
966 			}
967 		}
968 	}
969 
970 	/*
971 	 * Assign the pid and insert the process.
972 	 */
973 	p->p_pid = base;
974 	LIST_INSERT_HEAD(&allprocs[n], p, p_list);
975 	lwkt_reltoken(&proc_tokens[n]);
976 }
977 
978 /*
979  * Called from exit1 to place the process into a zombie state.
980  * The process is removed from the pid hash and p_stat is set
981  * to SZOMB.  Normal pfind[n]() calls will not find it any more.
982  *
983  * Caller must hold p->p_token.  We are required to wait until p_lock
984  * becomes zero before we can manipulate the list, allowing allproc
985  * scans to guarantee consistency during a list scan.
986  */
987 void
988 proc_move_allproc_zombie(struct proc *p)
989 {
990 	int n;
991 
992 	n = ALLPROC_HASH(p->p_pid);
993 	PSTALL(p, "reap1", 0);
994 	lwkt_gettoken(&proc_tokens[n]);
995 
996 	PSTALL(p, "reap1a", 0);
997 	p->p_stat = SZOMB;
998 
999 	lwkt_reltoken(&proc_tokens[n]);
1000 	dsched_exit_proc(p);
1001 }
1002 
1003 /*
1004  * This routine is called from kern_wait() and will remove the process
1005  * from the zombie list and the sibling list.  This routine will block
1006  * if someone has a lock on the proces (p_lock).
1007  *
1008  * Caller must hold p->p_token.  We are required to wait until p_lock
1009  * becomes zero before we can manipulate the list, allowing allproc
1010  * scans to guarantee consistency during a list scan.
1011  */
1012 void
1013 proc_remove_zombie(struct proc *p)
1014 {
1015 	int n;
1016 
1017 	n = ALLPROC_HASH(p->p_pid);
1018 
1019 	PSTALL(p, "reap2", 0);
1020 	lwkt_gettoken(&proc_tokens[n]);
1021 	PSTALL(p, "reap2a", 0);
1022 	LIST_REMOVE(p, p_list);		/* from remove master list */
1023 	LIST_REMOVE(p, p_sibling);	/* and from sibling list */
1024 	p->p_pptr = NULL;
1025 	lwkt_reltoken(&proc_tokens[n]);
1026 }
1027 
1028 /*
1029  * Handle various requirements prior to returning to usermode.  Called from
1030  * platform trap and system call code.
1031  */
1032 void
1033 lwpuserret(struct lwp *lp)
1034 {
1035 	struct proc *p = lp->lwp_proc;
1036 
1037 	if (lp->lwp_mpflags & LWP_MP_VNLRU) {
1038 		atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU);
1039 		allocvnode_gc();
1040 	}
1041 	if (lp->lwp_mpflags & LWP_MP_WEXIT) {
1042 		lwkt_gettoken(&p->p_token);
1043 		lwp_exit(0, NULL);
1044 		lwkt_reltoken(&p->p_token);     /* NOT REACHED */
1045 	}
1046 }
1047 
1048 /*
1049  * Kernel threads run from user processes can also accumulate deferred
1050  * actions which need to be acted upon.  Callers include:
1051  *
1052  * nfsd		- Can allocate lots of vnodes
1053  */
1054 void
1055 lwpkthreaddeferred(void)
1056 {
1057 	struct lwp *lp = curthread->td_lwp;
1058 
1059 	if (lp) {
1060 		if (lp->lwp_mpflags & LWP_MP_VNLRU) {
1061 			atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU);
1062 			allocvnode_gc();
1063 		}
1064 	}
1065 }
1066 
1067 /*
1068  * Scan all processes on the allproc list.  The process is automatically
1069  * held for the callback.  A return value of -1 terminates the loop.
1070  * Zombie procs are skipped.
1071  *
1072  * The callback is made with the process held and proc_token held.
1073  *
1074  * We limit the scan to the number of processes as-of the start of
1075  * the scan so as not to get caught up in an endless loop if new processes
1076  * are created more quickly than we can scan the old ones.  Add a little
1077  * slop to try to catch edge cases since nprocs can race.
1078  *
1079  * No requirements.
1080  */
1081 void
1082 allproc_scan(int (*callback)(struct proc *, void *), void *data)
1083 {
1084 	int limit = nprocs + ncpus;
1085 	struct proc *p;
1086 	int r;
1087 	int n;
1088 
1089 	/*
1090 	 * proc_tokens[n] protects the allproc list and PHOLD() prevents the
1091 	 * process from being removed from the allproc list or the zombproc
1092 	 * list.
1093 	 */
1094 	for (n = 0; n < ALLPROC_HSIZE; ++n) {
1095 		if (LIST_FIRST(&allprocs[n]) == NULL)
1096 			continue;
1097 		lwkt_gettoken(&proc_tokens[n]);
1098 		LIST_FOREACH(p, &allprocs[n], p_list) {
1099 			if (p->p_stat == SZOMB)
1100 				continue;
1101 			PHOLD(p);
1102 			r = callback(p, data);
1103 			PRELE(p);
1104 			if (r < 0)
1105 				break;
1106 			if (--limit < 0)
1107 				break;
1108 		}
1109 		lwkt_reltoken(&proc_tokens[n]);
1110 
1111 		/*
1112 		 * Check if asked to stop early
1113 		 */
1114 		if (p)
1115 			break;
1116 	}
1117 }
1118 
1119 /*
1120  * Scan all lwps of processes on the allproc list.  The lwp is automatically
1121  * held for the callback.  A return value of -1 terminates the loop.
1122  *
1123  * The callback is made with the proces and lwp both held, and proc_token held.
1124  *
1125  * No requirements.
1126  */
1127 void
1128 alllwp_scan(int (*callback)(struct lwp *, void *), void *data)
1129 {
1130 	struct proc *p;
1131 	struct lwp *lp;
1132 	int r = 0;
1133 	int n;
1134 
1135 	for (n = 0; n < ALLPROC_HSIZE; ++n) {
1136 		if (LIST_FIRST(&allprocs[n]) == NULL)
1137 			continue;
1138 		lwkt_gettoken(&proc_tokens[n]);
1139 		LIST_FOREACH(p, &allprocs[n], p_list) {
1140 			if (p->p_stat == SZOMB)
1141 				continue;
1142 			PHOLD(p);
1143 			lwkt_gettoken(&p->p_token);
1144 			FOREACH_LWP_IN_PROC(lp, p) {
1145 				LWPHOLD(lp);
1146 				r = callback(lp, data);
1147 				LWPRELE(lp);
1148 			}
1149 			lwkt_reltoken(&p->p_token);
1150 			PRELE(p);
1151 			if (r < 0)
1152 				break;
1153 		}
1154 		lwkt_reltoken(&proc_tokens[n]);
1155 
1156 		/*
1157 		 * Asked to exit early
1158 		 */
1159 		if (p)
1160 			break;
1161 	}
1162 }
1163 
1164 /*
1165  * Scan all processes on the zombproc list.  The process is automatically
1166  * held for the callback.  A return value of -1 terminates the loop.
1167  *
1168  * No requirements.
1169  * The callback is made with the proces held and proc_token held.
1170  */
1171 void
1172 zombproc_scan(int (*callback)(struct proc *, void *), void *data)
1173 {
1174 	struct proc *p;
1175 	int r;
1176 	int n;
1177 
1178 	/*
1179 	 * proc_tokens[n] protects the allproc list and PHOLD() prevents the
1180 	 * process from being removed from the allproc list or the zombproc
1181 	 * list.
1182 	 */
1183 	for (n = 0; n < ALLPROC_HSIZE; ++n) {
1184 		if (LIST_FIRST(&allprocs[n]) == NULL)
1185 			continue;
1186 		lwkt_gettoken(&proc_tokens[n]);
1187 		LIST_FOREACH(p, &allprocs[n], p_list) {
1188 			if (p->p_stat != SZOMB)
1189 				continue;
1190 			PHOLD(p);
1191 			r = callback(p, data);
1192 			PRELE(p);
1193 			if (r < 0)
1194 				break;
1195 		}
1196 		lwkt_reltoken(&proc_tokens[n]);
1197 
1198 		/*
1199 		 * Check if asked to stop early
1200 		 */
1201 		if (p)
1202 			break;
1203 	}
1204 }
1205 
1206 #include "opt_ddb.h"
1207 #ifdef DDB
1208 #include <ddb/ddb.h>
1209 
1210 /*
1211  * Debugging only
1212  */
1213 DB_SHOW_COMMAND(pgrpdump, pgrpdump)
1214 {
1215 	struct pgrp *pgrp;
1216 	struct proc *p;
1217 	int i;
1218 
1219 	for (i = 0; i < ALLPROC_HSIZE; ++i) {
1220 		if (LIST_EMPTY(&allpgrps[i]))
1221 			continue;
1222 		kprintf("\tindx %d\n", i);
1223 		LIST_FOREACH(pgrp, &allpgrps[i], pg_list) {
1224 			kprintf("\tpgrp %p, pgid %ld, sess %p, "
1225 				"sesscnt %d, mem %p\n",
1226 				(void *)pgrp, (long)pgrp->pg_id,
1227 				(void *)pgrp->pg_session,
1228 				pgrp->pg_session->s_count,
1229 				(void *)LIST_FIRST(&pgrp->pg_members));
1230 			LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1231 				kprintf("\t\tpid %ld addr %p pgrp %p\n",
1232 					(long)p->p_pid, (void *)p,
1233 					(void *)p->p_pgrp);
1234 			}
1235 		}
1236 	}
1237 }
1238 #endif /* DDB */
1239 
1240 /*
1241  * The caller must hold proc_token.
1242  */
1243 static int
1244 sysctl_out_proc(struct proc *p, struct sysctl_req *req, int flags)
1245 {
1246 	struct kinfo_proc ki;
1247 	struct lwp *lp;
1248 	int skp = 0, had_output = 0;
1249 	int error;
1250 
1251 	bzero(&ki, sizeof(ki));
1252 	lwkt_gettoken_shared(&p->p_token);
1253 	fill_kinfo_proc(p, &ki);
1254 	if ((flags & KERN_PROC_FLAG_LWP) == 0)
1255 		skp = 1;
1256 	error = 0;
1257 	FOREACH_LWP_IN_PROC(lp, p) {
1258 		LWPHOLD(lp);
1259 		fill_kinfo_lwp(lp, &ki.kp_lwp);
1260 		had_output = 1;
1261 		error = SYSCTL_OUT(req, &ki, sizeof(ki));
1262 		LWPRELE(lp);
1263 		if (error)
1264 			break;
1265 		if (skp)
1266 			break;
1267 	}
1268 	lwkt_reltoken(&p->p_token);
1269 	/* We need to output at least the proc, even if there is no lwp. */
1270 	if (had_output == 0) {
1271 		error = SYSCTL_OUT(req, &ki, sizeof(ki));
1272 	}
1273 	return (error);
1274 }
1275 
1276 /*
1277  * The caller must hold proc_token.
1278  */
1279 static int
1280 sysctl_out_proc_kthread(struct thread *td, struct sysctl_req *req)
1281 {
1282 	struct kinfo_proc ki;
1283 	int error;
1284 
1285 	fill_kinfo_proc_kthread(td, &ki);
1286 	error = SYSCTL_OUT(req, &ki, sizeof(ki));
1287 	if (error)
1288 		return error;
1289 	return(0);
1290 }
1291 
1292 /*
1293  * No requirements.
1294  */
1295 static int
1296 sysctl_kern_proc(SYSCTL_HANDLER_ARGS)
1297 {
1298 	int *name = (int *)arg1;
1299 	int oid = oidp->oid_number;
1300 	u_int namelen = arg2;
1301 	struct proc *p;
1302 	struct thread *td;
1303 	struct thread *marker;
1304 	int flags = 0;
1305 	int error = 0;
1306 	int n;
1307 	int origcpu;
1308 	struct ucred *cr1 = curproc->p_ucred;
1309 
1310 	flags = oid & KERN_PROC_FLAGMASK;
1311 	oid &= ~KERN_PROC_FLAGMASK;
1312 
1313 	if ((oid == KERN_PROC_ALL && namelen != 0) ||
1314 	    (oid != KERN_PROC_ALL && namelen != 1)) {
1315 		return (EINVAL);
1316 	}
1317 
1318 	/*
1319 	 * proc_token protects the allproc list and PHOLD() prevents the
1320 	 * process from being removed from the allproc list or the zombproc
1321 	 * list.
1322 	 */
1323 	if (oid == KERN_PROC_PID) {
1324 		p = pfind((pid_t)name[0]);
1325 		if (p) {
1326 			if (PRISON_CHECK(cr1, p->p_ucred))
1327 				error = sysctl_out_proc(p, req, flags);
1328 			PRELE(p);
1329 		}
1330 		goto post_threads;
1331 	}
1332 	p = NULL;
1333 
1334 	if (!req->oldptr) {
1335 		/* overestimate by 5 procs */
1336 		error = SYSCTL_OUT(req, 0, sizeof (struct kinfo_proc) * 5);
1337 		if (error)
1338 			goto post_threads;
1339 	}
1340 
1341 	for (n = 0; n < ALLPROC_HSIZE; ++n) {
1342 		if (LIST_EMPTY(&allprocs[n]))
1343 			continue;
1344 		lwkt_gettoken_shared(&proc_tokens[n]);
1345 		LIST_FOREACH(p, &allprocs[n], p_list) {
1346 			/*
1347 			 * Show a user only their processes.
1348 			 */
1349 			if ((!ps_showallprocs) && p_trespass(cr1, p->p_ucred))
1350 				continue;
1351 			/*
1352 			 * Skip embryonic processes.
1353 			 */
1354 			if (p->p_stat == SIDL)
1355 				continue;
1356 			/*
1357 			 * TODO - make more efficient (see notes below).
1358 			 * do by session.
1359 			 */
1360 			switch (oid) {
1361 			case KERN_PROC_PGRP:
1362 				/* could do this by traversing pgrp */
1363 				if (p->p_pgrp == NULL ||
1364 				    p->p_pgrp->pg_id != (pid_t)name[0])
1365 					continue;
1366 				break;
1367 
1368 			case KERN_PROC_TTY:
1369 				if ((p->p_flags & P_CONTROLT) == 0 ||
1370 				    p->p_session == NULL ||
1371 				    p->p_session->s_ttyp == NULL ||
1372 				    dev2udev(p->p_session->s_ttyp->t_dev) !=
1373 					(udev_t)name[0])
1374 					continue;
1375 				break;
1376 
1377 			case KERN_PROC_UID:
1378 				if (p->p_ucred == NULL ||
1379 				    p->p_ucred->cr_uid != (uid_t)name[0])
1380 					continue;
1381 				break;
1382 
1383 			case KERN_PROC_RUID:
1384 				if (p->p_ucred == NULL ||
1385 				    p->p_ucred->cr_ruid != (uid_t)name[0])
1386 					continue;
1387 				break;
1388 			}
1389 
1390 			if (!PRISON_CHECK(cr1, p->p_ucred))
1391 				continue;
1392 			PHOLD(p);
1393 			error = sysctl_out_proc(p, req, flags);
1394 			PRELE(p);
1395 			if (error) {
1396 				lwkt_reltoken(&proc_tokens[n]);
1397 				goto post_threads;
1398 			}
1399 		}
1400 		lwkt_reltoken(&proc_tokens[n]);
1401 	}
1402 
1403 	/*
1404 	 * Iterate over all active cpus and scan their thread list.  Start
1405 	 * with the next logical cpu and end with our original cpu.  We
1406 	 * migrate our own thread to each target cpu in order to safely scan
1407 	 * its thread list.  In the last loop we migrate back to our original
1408 	 * cpu.
1409 	 */
1410 	origcpu = mycpu->gd_cpuid;
1411 	if (!ps_showallthreads || jailed(cr1))
1412 		goto post_threads;
1413 
1414 	marker = kmalloc(sizeof(struct thread), M_TEMP, M_WAITOK|M_ZERO);
1415 	marker->td_flags = TDF_MARKER;
1416 	error = 0;
1417 
1418 	for (n = 1; n <= ncpus; ++n) {
1419 		globaldata_t rgd;
1420 		int nid;
1421 
1422 		nid = (origcpu + n) % ncpus;
1423 		if ((smp_active_mask & CPUMASK(nid)) == 0)
1424 			continue;
1425 		rgd = globaldata_find(nid);
1426 		lwkt_setcpu_self(rgd);
1427 
1428 		crit_enter();
1429 		TAILQ_INSERT_TAIL(&rgd->gd_tdallq, marker, td_allq);
1430 
1431 		while ((td = TAILQ_PREV(marker, lwkt_queue, td_allq)) != NULL) {
1432 			TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq);
1433 			TAILQ_INSERT_BEFORE(td, marker, td_allq);
1434 			if (td->td_flags & TDF_MARKER)
1435 				continue;
1436 			if (td->td_proc)
1437 				continue;
1438 
1439 			lwkt_hold(td);
1440 			crit_exit();
1441 
1442 			switch (oid) {
1443 			case KERN_PROC_PGRP:
1444 			case KERN_PROC_TTY:
1445 			case KERN_PROC_UID:
1446 			case KERN_PROC_RUID:
1447 				break;
1448 			default:
1449 				error = sysctl_out_proc_kthread(td, req);
1450 				break;
1451 			}
1452 			lwkt_rele(td);
1453 			crit_enter();
1454 			if (error)
1455 				break;
1456 		}
1457 		TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq);
1458 		crit_exit();
1459 
1460 		if (error)
1461 			break;
1462 	}
1463 
1464 	/*
1465 	 * Userland scheduler expects us to return on the same cpu we
1466 	 * started on.
1467 	 */
1468 	if (mycpu->gd_cpuid != origcpu)
1469 		lwkt_setcpu_self(globaldata_find(origcpu));
1470 
1471 	kfree(marker, M_TEMP);
1472 
1473 post_threads:
1474 	return (error);
1475 }
1476 
1477 /*
1478  * This sysctl allows a process to retrieve the argument list or process
1479  * title for another process without groping around in the address space
1480  * of the other process.  It also allow a process to set its own "process
1481  * title to a string of its own choice.
1482  *
1483  * No requirements.
1484  */
1485 static int
1486 sysctl_kern_proc_args(SYSCTL_HANDLER_ARGS)
1487 {
1488 	int *name = (int*) arg1;
1489 	u_int namelen = arg2;
1490 	struct proc *p;
1491 	struct pargs *opa;
1492 	struct pargs *pa;
1493 	int error = 0;
1494 	struct ucred *cr1 = curproc->p_ucred;
1495 
1496 	if (namelen != 1)
1497 		return (EINVAL);
1498 
1499 	p = pfind((pid_t)name[0]);
1500 	if (p == NULL)
1501 		goto done;
1502 	lwkt_gettoken(&p->p_token);
1503 
1504 	if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred))
1505 		goto done;
1506 
1507 	if (req->newptr && curproc != p) {
1508 		error = EPERM;
1509 		goto done;
1510 	}
1511 	if (req->oldptr && (pa = p->p_args) != NULL) {
1512 		refcount_acquire(&pa->ar_ref);
1513 		error = SYSCTL_OUT(req, pa->ar_args, pa->ar_length);
1514 		if (refcount_release(&pa->ar_ref))
1515 			kfree(pa, M_PARGS);
1516 	}
1517 	if (req->newptr == NULL)
1518 		goto done;
1519 
1520 	if (req->newlen + sizeof(struct pargs) > ps_arg_cache_limit) {
1521 		goto done;
1522 	}
1523 
1524 	pa = kmalloc(sizeof(struct pargs) + req->newlen, M_PARGS, M_WAITOK);
1525 	refcount_init(&pa->ar_ref, 1);
1526 	pa->ar_length = req->newlen;
1527 	error = SYSCTL_IN(req, pa->ar_args, req->newlen);
1528 	if (error) {
1529 		kfree(pa, M_PARGS);
1530 		goto done;
1531 	}
1532 
1533 
1534 	/*
1535 	 * Replace p_args with the new pa.  p_args may have previously
1536 	 * been NULL.
1537 	 */
1538 	opa = p->p_args;
1539 	p->p_args = pa;
1540 
1541 	if (opa) {
1542 		KKASSERT(opa->ar_ref > 0);
1543 		if (refcount_release(&opa->ar_ref)) {
1544 			kfree(opa, M_PARGS);
1545 			/* opa = NULL; */
1546 		}
1547 	}
1548 done:
1549 	if (p) {
1550 		lwkt_reltoken(&p->p_token);
1551 		PRELE(p);
1552 	}
1553 	return (error);
1554 }
1555 
1556 static int
1557 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
1558 {
1559 	int *name = (int*) arg1;
1560 	u_int namelen = arg2;
1561 	struct proc *p;
1562 	int error = 0;
1563 	char *fullpath, *freepath;
1564 	struct ucred *cr1 = curproc->p_ucred;
1565 
1566 	if (namelen != 1)
1567 		return (EINVAL);
1568 
1569 	p = pfind((pid_t)name[0]);
1570 	if (p == NULL)
1571 		goto done;
1572 	lwkt_gettoken_shared(&p->p_token);
1573 
1574 	/*
1575 	 * If we are not allowed to see other args, we certainly shouldn't
1576 	 * get the cwd either. Also check the usual trespassing.
1577 	 */
1578 	if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred))
1579 		goto done;
1580 
1581 	if (req->oldptr && p->p_fd != NULL && p->p_fd->fd_ncdir.ncp) {
1582 		struct nchandle nch;
1583 
1584 		cache_copy(&p->p_fd->fd_ncdir, &nch);
1585 		error = cache_fullpath(p, &nch, NULL,
1586 				       &fullpath, &freepath, 0);
1587 		cache_drop(&nch);
1588 		if (error)
1589 			goto done;
1590 		error = SYSCTL_OUT(req, fullpath, strlen(fullpath) + 1);
1591 		kfree(freepath, M_TEMP);
1592 	}
1593 
1594 done:
1595 	if (p) {
1596 		lwkt_reltoken(&p->p_token);
1597 		PRELE(p);
1598 	}
1599 	return (error);
1600 }
1601 
1602 /*
1603  * This sysctl allows a process to retrieve the path of the executable for
1604  * itself or another process.
1605  */
1606 static int
1607 sysctl_kern_proc_pathname(SYSCTL_HANDLER_ARGS)
1608 {
1609 	pid_t *pidp = (pid_t *)arg1;
1610 	unsigned int arglen = arg2;
1611 	struct proc *p;
1612 	struct vnode *vp;
1613 	char *retbuf, *freebuf;
1614 	int error;
1615 
1616 	if (arglen != 1)
1617 		return (EINVAL);
1618 	if (*pidp == -1) {	/* -1 means this process */
1619 		p = curproc;
1620 	} else {
1621 		p = pfind(*pidp);
1622 		if (p == NULL)
1623 			return (ESRCH);
1624 	}
1625 
1626 	vp = p->p_textvp;
1627 	if (vp == NULL) {
1628 		return (0);
1629 	}
1630 	vref(vp);
1631 	error = vn_fullpath(p, vp, &retbuf, &freebuf, 0);
1632 	vrele(vp);
1633 	if (error)
1634 		return (error);
1635 	error = SYSCTL_OUT(req, retbuf, strlen(retbuf) + 1);
1636 	kfree(freebuf, M_TEMP);
1637 	return (error);
1638 }
1639 
1640 SYSCTL_NODE(_kern, KERN_PROC, proc, CTLFLAG_RD,  0, "Process table");
1641 
1642 SYSCTL_PROC(_kern_proc, KERN_PROC_ALL, all, CTLFLAG_RD|CTLTYPE_STRUCT,
1643 	0, 0, sysctl_kern_proc, "S,proc", "Return entire process table");
1644 
1645 SYSCTL_NODE(_kern_proc, KERN_PROC_PGRP, pgrp, CTLFLAG_RD,
1646 	sysctl_kern_proc, "Process table");
1647 
1648 SYSCTL_NODE(_kern_proc, KERN_PROC_TTY, tty, CTLFLAG_RD,
1649 	sysctl_kern_proc, "Process table");
1650 
1651 SYSCTL_NODE(_kern_proc, KERN_PROC_UID, uid, CTLFLAG_RD,
1652 	sysctl_kern_proc, "Process table");
1653 
1654 SYSCTL_NODE(_kern_proc, KERN_PROC_RUID, ruid, CTLFLAG_RD,
1655 	sysctl_kern_proc, "Process table");
1656 
1657 SYSCTL_NODE(_kern_proc, KERN_PROC_PID, pid, CTLFLAG_RD,
1658 	sysctl_kern_proc, "Process table");
1659 
1660 SYSCTL_NODE(_kern_proc, (KERN_PROC_ALL | KERN_PROC_FLAG_LWP), all_lwp, CTLFLAG_RD,
1661 	sysctl_kern_proc, "Process table");
1662 
1663 SYSCTL_NODE(_kern_proc, (KERN_PROC_PGRP | KERN_PROC_FLAG_LWP), pgrp_lwp, CTLFLAG_RD,
1664 	sysctl_kern_proc, "Process table");
1665 
1666 SYSCTL_NODE(_kern_proc, (KERN_PROC_TTY | KERN_PROC_FLAG_LWP), tty_lwp, CTLFLAG_RD,
1667 	sysctl_kern_proc, "Process table");
1668 
1669 SYSCTL_NODE(_kern_proc, (KERN_PROC_UID | KERN_PROC_FLAG_LWP), uid_lwp, CTLFLAG_RD,
1670 	sysctl_kern_proc, "Process table");
1671 
1672 SYSCTL_NODE(_kern_proc, (KERN_PROC_RUID | KERN_PROC_FLAG_LWP), ruid_lwp, CTLFLAG_RD,
1673 	sysctl_kern_proc, "Process table");
1674 
1675 SYSCTL_NODE(_kern_proc, (KERN_PROC_PID | KERN_PROC_FLAG_LWP), pid_lwp, CTLFLAG_RD,
1676 	sysctl_kern_proc, "Process table");
1677 
1678 SYSCTL_NODE(_kern_proc, KERN_PROC_ARGS, args, CTLFLAG_RW | CTLFLAG_ANYBODY,
1679 	sysctl_kern_proc_args, "Process argument list");
1680 
1681 SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD | CTLFLAG_ANYBODY,
1682 	sysctl_kern_proc_cwd, "Process argument list");
1683 
1684 static SYSCTL_NODE(_kern_proc, KERN_PROC_PATHNAME, pathname, CTLFLAG_RD,
1685 	sysctl_kern_proc_pathname, "Process executable path");
1686