xref: /netbsd-src/sys/kern/kern_proc.c (revision 6deb2c22d20de1d75d538e8a5c57b573926fd157)
1 /*	$NetBSD: kern_proc.c,v 1.154 2009/10/04 03:15:08 elad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1989, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)kern_proc.c	8.7 (Berkeley) 2/14/95
62  */
63 
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: kern_proc.c,v 1.154 2009/10/04 03:15:08 elad Exp $");
66 
67 #include "opt_kstack.h"
68 #include "opt_maxuprc.h"
69 
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
73 #include <sys/proc.h>
74 #include <sys/resourcevar.h>
75 #include <sys/buf.h>
76 #include <sys/acct.h>
77 #include <sys/wait.h>
78 #include <sys/file.h>
79 #include <ufs/ufs/quota.h>
80 #include <sys/uio.h>
81 #include <sys/pool.h>
82 #include <sys/pset.h>
83 #include <sys/mbuf.h>
84 #include <sys/ioctl.h>
85 #include <sys/tty.h>
86 #include <sys/signalvar.h>
87 #include <sys/ras.h>
88 #include <sys/sa.h>
89 #include <sys/savar.h>
90 #include <sys/filedesc.h>
91 #include "sys/syscall_stats.h"
92 #include <sys/kauth.h>
93 #include <sys/sleepq.h>
94 #include <sys/atomic.h>
95 #include <sys/kmem.h>
96 
97 #include <uvm/uvm.h>
98 #include <uvm/uvm_extern.h>
99 
100 /*
101  * Other process lists
102  */
103 
104 struct proclist allproc;
105 struct proclist zombproc;	/* resources have been freed */
106 
107 kmutex_t	*proc_lock;
108 
109 /*
110  * pid to proc lookup is done by indexing the pid_table array.
111  * Since pid numbers are only allocated when an empty slot
112  * has been found, there is no need to search any lists ever.
113  * (an orphaned pgrp will lock the slot, a session will lock
114  * the pgrp with the same number.)
115  * If the table is too small it is reallocated with twice the
116  * previous size and the entries 'unzipped' into the two halves.
117  * A linked list of free entries is passed through the pt_proc
118  * field of 'free' items - set odd to be an invalid ptr.
119  */
120 
121 struct pid_table {
122 	struct proc	*pt_proc;
123 	struct pgrp	*pt_pgrp;
124 };
125 #if 1	/* strongly typed cast - should be a noop */
126 static inline uint p2u(struct proc *p) { return (uint)(uintptr_t)p; }
127 #else
128 #define p2u(p) ((uint)p)
129 #endif
130 #define P_VALID(p) (!(p2u(p) & 1))
131 #define P_NEXT(p) (p2u(p) >> 1)
132 #define P_FREE(pid) ((struct proc *)(uintptr_t)((pid) << 1 | 1))
133 
134 #define INITIAL_PID_TABLE_SIZE	(1 << 5)
135 static struct pid_table *pid_table;
136 static uint pid_tbl_mask = INITIAL_PID_TABLE_SIZE - 1;
137 static uint pid_alloc_lim;	/* max we allocate before growing table */
138 static uint pid_alloc_cnt;	/* number of allocated pids */
139 
140 /* links through free slots - never empty! */
141 static uint next_free_pt, last_free_pt;
142 static pid_t pid_max = PID_MAX;		/* largest value we allocate */
143 
144 /* Components of the first process -- never freed. */
145 
146 extern struct emul emul_netbsd;	/* defined in kern_exec.c */
147 
148 struct session session0 = {
149 	.s_count = 1,
150 	.s_sid = 0,
151 };
152 struct pgrp pgrp0 = {
153 	.pg_members = LIST_HEAD_INITIALIZER(&pgrp0.pg_members),
154 	.pg_session = &session0,
155 };
156 filedesc_t filedesc0;
157 struct cwdinfo cwdi0 = {
158 	.cwdi_cmask = CMASK,		/* see cmask below */
159 	.cwdi_refcnt = 1,
160 };
161 struct plimit limit0;
162 struct pstats pstat0;
163 struct vmspace vmspace0;
164 struct sigacts sigacts0;
165 struct turnstile turnstile0;
166 struct proc proc0 = {
167 	.p_lwps = LIST_HEAD_INITIALIZER(&proc0.p_lwps),
168 	.p_sigwaiters = LIST_HEAD_INITIALIZER(&proc0.p_sigwaiters),
169 	.p_nlwps = 1,
170 	.p_nrlwps = 1,
171 	.p_nlwpid = 1,		/* must match lwp0.l_lid */
172 	.p_pgrp = &pgrp0,
173 	.p_comm = "system",
174 	/*
175 	 * Set P_NOCLDWAIT so that kernel threads are reparented to init(8)
176 	 * when they exit.  init(8) can easily wait them out for us.
177 	 */
178 	.p_flag = PK_SYSTEM | PK_NOCLDWAIT,
179 	.p_stat = SACTIVE,
180 	.p_nice = NZERO,
181 	.p_emul = &emul_netbsd,
182 	.p_cwdi = &cwdi0,
183 	.p_limit = &limit0,
184 	.p_fd = &filedesc0,
185 	.p_vmspace = &vmspace0,
186 	.p_stats = &pstat0,
187 	.p_sigacts = &sigacts0,
188 };
189 struct lwp lwp0 __aligned(MIN_LWP_ALIGNMENT) = {
190 #ifdef LWP0_CPU_INFO
191 	.l_cpu = LWP0_CPU_INFO,
192 #endif
193 	.l_proc = &proc0,
194 	.l_lid = 1,
195 	.l_flag = LW_INMEM | LW_SYSTEM,
196 	.l_stat = LSONPROC,
197 	.l_ts = &turnstile0,
198 	.l_syncobj = &sched_syncobj,
199 	.l_refcnt = 1,
200 	.l_priority = PRI_USER + NPRI_USER - 1,
201 	.l_inheritedprio = -1,
202 	.l_class = SCHED_OTHER,
203 	.l_psid = PS_NONE,
204 	.l_pi_lenders = SLIST_HEAD_INITIALIZER(&lwp0.l_pi_lenders),
205 	.l_name = __UNCONST("swapper"),
206 	.l_fd = &filedesc0,
207 };
208 kauth_cred_t cred0;
209 
210 extern struct user *proc0paddr;
211 
212 int nofile = NOFILE;
213 int maxuprc = MAXUPRC;
214 int cmask = CMASK;
215 
216 MALLOC_DEFINE(M_EMULDATA, "emuldata", "Per-process emulation data");
217 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures");
218 
219 /*
220  * The process list descriptors, used during pid allocation and
221  * by sysctl.  No locking on this data structure is needed since
222  * it is completely static.
223  */
224 const struct proclist_desc proclists[] = {
225 	{ &allproc	},
226 	{ &zombproc	},
227 	{ NULL		},
228 };
229 
230 static struct pgrp *	pg_remove(pid_t);
231 static void		pg_delete(pid_t);
232 static void		orphanpg(struct pgrp *);
233 
234 static specificdata_domain_t proc_specificdata_domain;
235 
236 static pool_cache_t proc_cache;
237 
238 static kauth_listener_t proc_listener;
239 
240 static int
241 proc_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
242     void *arg0, void *arg1, void *arg2, void *arg3)
243 {
244 	struct proc *p;
245 	int result;
246 
247 	result = KAUTH_RESULT_DEFER;
248 	p = arg0;
249 
250 	switch (action) {
251 	case KAUTH_PROCESS_CANSEE: {
252 		enum kauth_process_req req;
253 
254 		req = (enum kauth_process_req)arg1;
255 
256 		switch (req) {
257 		case KAUTH_REQ_PROCESS_CANSEE_ARGS:
258 		case KAUTH_REQ_PROCESS_CANSEE_ENTRY:
259 		case KAUTH_REQ_PROCESS_CANSEE_OPENFILES:
260 			result = KAUTH_RESULT_ALLOW;
261 
262 			break;
263 
264 		case KAUTH_REQ_PROCESS_CANSEE_ENV:
265 			if (kauth_cred_getuid(cred) !=
266 			    kauth_cred_getuid(p->p_cred) ||
267 			    kauth_cred_getuid(cred) !=
268 			    kauth_cred_getsvuid(p->p_cred))
269 				break;
270 
271 			result = KAUTH_RESULT_ALLOW;
272 
273 			break;
274 
275 		default:
276 			break;
277 		}
278 
279 		break;
280 		}
281 
282 	case KAUTH_PROCESS_FORK: {
283 		int lnprocs = (int)(unsigned long)arg2;
284 
285 		/*
286 		 * Don't allow a nonprivileged user to use the last few
287 		 * processes. The variable lnprocs is the current number of
288 		 * processes, maxproc is the limit.
289 		 */
290 		if (__predict_false((lnprocs >= maxproc - 5)))
291 			break;
292 
293 		result = KAUTH_RESULT_ALLOW;
294 
295 		break;
296 		}
297 
298 	case KAUTH_PROCESS_CORENAME:
299 	case KAUTH_PROCESS_STOPFLAG:
300 		if (proc_uidmatch(cred, p->p_cred) == 0)
301 			result = KAUTH_RESULT_ALLOW;
302 
303 		break;
304 
305 	default:
306 		break;
307 	}
308 
309 	return result;
310 }
311 
312 /*
313  * Initialize global process hashing structures.
314  */
315 void
316 procinit(void)
317 {
318 	const struct proclist_desc *pd;
319 	u_int i;
320 #define	LINK_EMPTY ((PID_MAX + INITIAL_PID_TABLE_SIZE) & ~(INITIAL_PID_TABLE_SIZE - 1))
321 
322 	for (pd = proclists; pd->pd_list != NULL; pd++)
323 		LIST_INIT(pd->pd_list);
324 
325 	proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
326 	pid_table = kmem_alloc(INITIAL_PID_TABLE_SIZE
327 	    * sizeof(struct pid_table), KM_SLEEP);
328 
329 	/* Set free list running through table...
330 	   Preset 'use count' above PID_MAX so we allocate pid 1 next. */
331 	for (i = 0; i <= pid_tbl_mask; i++) {
332 		pid_table[i].pt_proc = P_FREE(LINK_EMPTY + i + 1);
333 		pid_table[i].pt_pgrp = 0;
334 	}
335 	/* slot 0 is just grabbed */
336 	next_free_pt = 1;
337 	/* Need to fix last entry. */
338 	last_free_pt = pid_tbl_mask;
339 	pid_table[last_free_pt].pt_proc = P_FREE(LINK_EMPTY);
340 	/* point at which we grow table - to avoid reusing pids too often */
341 	pid_alloc_lim = pid_tbl_mask - 1;
342 #undef LINK_EMPTY
343 
344 	proc_specificdata_domain = specificdata_domain_create();
345 	KASSERT(proc_specificdata_domain != NULL);
346 
347 	proc_cache = pool_cache_init(sizeof(struct proc), 0, 0, 0,
348 	    "procpl", NULL, IPL_NONE, NULL, NULL, NULL);
349 
350 	proc_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS,
351 	    proc_listener_cb, NULL);
352 }
353 
354 /*
355  * Initialize process 0.
356  */
357 void
358 proc0_init(void)
359 {
360 	struct proc *p;
361 	struct pgrp *pg;
362 	struct session *sess;
363 	struct lwp *l;
364 	rlim_t lim;
365 	int i;
366 
367 	p = &proc0;
368 	pg = &pgrp0;
369 	sess = &session0;
370 	l = &lwp0;
371 
372 	KASSERT(l->l_lid == p->p_nlwpid);
373 
374 	mutex_init(&p->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
375 	mutex_init(&p->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
376 	mutex_init(&l->l_swaplock, MUTEX_DEFAULT, IPL_NONE);
377 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
378 
379 	rw_init(&p->p_reflock);
380 	cv_init(&p->p_waitcv, "wait");
381 	cv_init(&p->p_lwpcv, "lwpwait");
382 
383 	LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling);
384 
385 	pid_table[0].pt_proc = p;
386 	LIST_INSERT_HEAD(&allproc, p, p_list);
387 	LIST_INSERT_HEAD(&alllwp, l, l_list);
388 
389 	pid_table[0].pt_pgrp = pg;
390 	LIST_INSERT_HEAD(&pg->pg_members, p, p_pglist);
391 
392 #ifdef __HAVE_SYSCALL_INTERN
393 	(*p->p_emul->e_syscall_intern)(p);
394 #endif
395 
396 	callout_init(&l->l_timeout_ch, CALLOUT_MPSAFE);
397 	callout_setfunc(&l->l_timeout_ch, sleepq_timeout, l);
398 	cv_init(&l->l_sigcv, "sigwait");
399 
400 	/* Create credentials. */
401 	cred0 = kauth_cred_alloc();
402 	p->p_cred = cred0;
403 	kauth_cred_hold(cred0);
404 	l->l_cred = cred0;
405 
406 	/* Create the CWD info. */
407 	rw_init(&cwdi0.cwdi_lock);
408 
409 	/* Create the limits structures. */
410 	mutex_init(&limit0.pl_lock, MUTEX_DEFAULT, IPL_NONE);
411 	for (i = 0; i < __arraycount(limit0.pl_rlimit); i++)
412 		limit0.pl_rlimit[i].rlim_cur =
413 		    limit0.pl_rlimit[i].rlim_max = RLIM_INFINITY;
414 
415 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
416 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_cur =
417 	    maxfiles < nofile ? maxfiles : nofile;
418 
419 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
420 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_cur =
421 	    maxproc < maxuprc ? maxproc : maxuprc;
422 
423 	lim = ptoa(uvmexp.free);
424 	limit0.pl_rlimit[RLIMIT_RSS].rlim_max = lim;
425 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim;
426 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3;
427 	limit0.pl_corename = defcorename;
428 	limit0.pl_refcnt = 1;
429 	limit0.pl_sv_limit = NULL;
430 
431 	/* Configure virtual memory system, set vm rlimits. */
432 	uvm_init_limits(p);
433 
434 	/* Initialize file descriptor table for proc0. */
435 	fd_init(&filedesc0);
436 
437 	/*
438 	 * Initialize proc0's vmspace, which uses the kernel pmap.
439 	 * All kernel processes (which never have user space mappings)
440 	 * share proc0's vmspace, and thus, the kernel pmap.
441 	 */
442 	uvmspace_init(&vmspace0, pmap_kernel(), round_page(VM_MIN_ADDRESS),
443 	    trunc_page(VM_MAX_ADDRESS));
444 
445 	l->l_addr = proc0paddr;				/* XXX */
446 
447 	/* Initialize signal state for proc0. XXX IPL_SCHED */
448 	mutex_init(&p->p_sigacts->sa_mutex, MUTEX_DEFAULT, IPL_SCHED);
449 	siginit(p);
450 
451 	proc_initspecific(p);
452 	lwp_initspecific(l);
453 
454 	SYSCALL_TIME_LWP_INIT(l);
455 }
456 
457 /*
458  * Session reference counting.
459  */
460 
461 void
462 proc_sesshold(struct session *ss)
463 {
464 
465 	KASSERT(mutex_owned(proc_lock));
466 	ss->s_count++;
467 }
468 
469 void
470 proc_sessrele(struct session *ss)
471 {
472 
473 	KASSERT(mutex_owned(proc_lock));
474 	/*
475 	 * We keep the pgrp with the same id as the session in order to
476 	 * stop a process being given the same pid.  Since the pgrp holds
477 	 * a reference to the session, it must be a 'zombie' pgrp by now.
478 	 */
479 	if (--ss->s_count == 0) {
480 		struct pgrp *pg;
481 
482 		pg = pg_remove(ss->s_sid);
483 		mutex_exit(proc_lock);
484 
485 		kmem_free(pg, sizeof(struct pgrp));
486 		kmem_free(ss, sizeof(struct session));
487 	} else {
488 		mutex_exit(proc_lock);
489 	}
490 }
491 
492 /*
493  * Check that the specified process group is in the session of the
494  * specified process.
495  * Treats -ve ids as process ids.
496  * Used to validate TIOCSPGRP requests.
497  */
498 int
499 pgid_in_session(struct proc *p, pid_t pg_id)
500 {
501 	struct pgrp *pgrp;
502 	struct session *session;
503 	int error;
504 
505 	mutex_enter(proc_lock);
506 	if (pg_id < 0) {
507 		struct proc *p1 = p_find(-pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
508 		if (p1 == NULL)
509 			return EINVAL;
510 		pgrp = p1->p_pgrp;
511 	} else {
512 		pgrp = pg_find(pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
513 		if (pgrp == NULL)
514 			return EINVAL;
515 	}
516 	session = pgrp->pg_session;
517 	if (session != p->p_pgrp->pg_session)
518 		error = EPERM;
519 	else
520 		error = 0;
521 	mutex_exit(proc_lock);
522 
523 	return error;
524 }
525 
526 /*
527  * p_inferior: is p an inferior of q?
528  */
529 static inline bool
530 p_inferior(struct proc *p, struct proc *q)
531 {
532 
533 	KASSERT(mutex_owned(proc_lock));
534 
535 	for (; p != q; p = p->p_pptr)
536 		if (p->p_pid == 0)
537 			return false;
538 	return true;
539 }
540 
541 /*
542  * Locate a process by number
543  */
544 struct proc *
545 p_find(pid_t pid, uint flags)
546 {
547 	struct proc *p;
548 	char stat;
549 
550 	if (!(flags & PFIND_LOCKED))
551 		mutex_enter(proc_lock);
552 
553 	p = pid_table[pid & pid_tbl_mask].pt_proc;
554 
555 	/* Only allow live processes to be found by pid. */
556 	/* XXXSMP p_stat */
557 	if (P_VALID(p) && p->p_pid == pid && ((stat = p->p_stat) == SACTIVE ||
558 	    stat == SSTOP || ((flags & PFIND_ZOMBIE) &&
559 	    (stat == SZOMB || stat == SDEAD || stat == SDYING)))) {
560 		if (flags & PFIND_UNLOCK_OK)
561 			 mutex_exit(proc_lock);
562 		return p;
563 	}
564 	if (flags & PFIND_UNLOCK_FAIL)
565 		mutex_exit(proc_lock);
566 	return NULL;
567 }
568 
569 
570 /*
571  * Locate a process group by number
572  */
573 struct pgrp *
574 pg_find(pid_t pgid, uint flags)
575 {
576 	struct pgrp *pg;
577 
578 	if (!(flags & PFIND_LOCKED))
579 		mutex_enter(proc_lock);
580 	pg = pid_table[pgid & pid_tbl_mask].pt_pgrp;
581 	/*
582 	 * Can't look up a pgrp that only exists because the session
583 	 * hasn't died yet (traditional)
584 	 */
585 	if (pg == NULL || pg->pg_id != pgid || LIST_EMPTY(&pg->pg_members)) {
586 		if (flags & PFIND_UNLOCK_FAIL)
587 			 mutex_exit(proc_lock);
588 		return NULL;
589 	}
590 
591 	if (flags & PFIND_UNLOCK_OK)
592 		mutex_exit(proc_lock);
593 	return pg;
594 }
595 
596 static void
597 expand_pid_table(void)
598 {
599 	size_t pt_size, tsz;
600 	struct pid_table *n_pt, *new_pt;
601 	struct proc *proc;
602 	struct pgrp *pgrp;
603 	pid_t pid;
604 	u_int i;
605 
606 	pt_size = pid_tbl_mask + 1;
607 	tsz = pt_size * 2 * sizeof(struct pid_table);
608 	new_pt = kmem_alloc(tsz, KM_SLEEP);
609 
610 	mutex_enter(proc_lock);
611 	if (pt_size != pid_tbl_mask + 1) {
612 		/* Another process beat us to it... */
613 		mutex_exit(proc_lock);
614 		kmem_free(new_pt, tsz);
615 		return;
616 	}
617 
618 	/*
619 	 * Copy entries from old table into new one.
620 	 * If 'pid' is 'odd' we need to place in the upper half,
621 	 * even pid's to the lower half.
622 	 * Free items stay in the low half so we don't have to
623 	 * fixup the reference to them.
624 	 * We stuff free items on the front of the freelist
625 	 * because we can't write to unmodified entries.
626 	 * Processing the table backwards maintains a semblance
627 	 * of issueing pid numbers that increase with time.
628 	 */
629 	i = pt_size - 1;
630 	n_pt = new_pt + i;
631 	for (; ; i--, n_pt--) {
632 		proc = pid_table[i].pt_proc;
633 		pgrp = pid_table[i].pt_pgrp;
634 		if (!P_VALID(proc)) {
635 			/* Up 'use count' so that link is valid */
636 			pid = (P_NEXT(proc) + pt_size) & ~pt_size;
637 			proc = P_FREE(pid);
638 			if (pgrp)
639 				pid = pgrp->pg_id;
640 		} else
641 			pid = proc->p_pid;
642 
643 		/* Save entry in appropriate half of table */
644 		n_pt[pid & pt_size].pt_proc = proc;
645 		n_pt[pid & pt_size].pt_pgrp = pgrp;
646 
647 		/* Put other piece on start of free list */
648 		pid = (pid ^ pt_size) & ~pid_tbl_mask;
649 		n_pt[pid & pt_size].pt_proc =
650 				    P_FREE((pid & ~pt_size) | next_free_pt);
651 		n_pt[pid & pt_size].pt_pgrp = 0;
652 		next_free_pt = i | (pid & pt_size);
653 		if (i == 0)
654 			break;
655 	}
656 
657 	/* Save old table size and switch tables */
658 	tsz = pt_size * sizeof(struct pid_table);
659 	n_pt = pid_table;
660 	pid_table = new_pt;
661 	pid_tbl_mask = pt_size * 2 - 1;
662 
663 	/*
664 	 * pid_max starts as PID_MAX (= 30000), once we have 16384
665 	 * allocated pids we need it to be larger!
666 	 */
667 	if (pid_tbl_mask > PID_MAX) {
668 		pid_max = pid_tbl_mask * 2 + 1;
669 		pid_alloc_lim |= pid_alloc_lim << 1;
670 	} else
671 		pid_alloc_lim <<= 1;	/* doubles number of free slots... */
672 
673 	mutex_exit(proc_lock);
674 	kmem_free(n_pt, tsz);
675 }
676 
677 struct proc *
678 proc_alloc(void)
679 {
680 	struct proc *p;
681 	int nxt;
682 	pid_t pid;
683 	struct pid_table *pt;
684 
685 	p = pool_cache_get(proc_cache, PR_WAITOK);
686 	p->p_stat = SIDL;			/* protect against others */
687 
688 	proc_initspecific(p);
689 	/* allocate next free pid */
690 
691 	for (;;expand_pid_table()) {
692 		if (__predict_false(pid_alloc_cnt >= pid_alloc_lim))
693 			/* ensure pids cycle through 2000+ values */
694 			continue;
695 		mutex_enter(proc_lock);
696 		pt = &pid_table[next_free_pt];
697 #ifdef DIAGNOSTIC
698 		if (__predict_false(P_VALID(pt->pt_proc) || pt->pt_pgrp))
699 			panic("proc_alloc: slot busy");
700 #endif
701 		nxt = P_NEXT(pt->pt_proc);
702 		if (nxt & pid_tbl_mask)
703 			break;
704 		/* Table full - expand (NB last entry not used....) */
705 		mutex_exit(proc_lock);
706 	}
707 
708 	/* pid is 'saved use count' + 'size' + entry */
709 	pid = (nxt & ~pid_tbl_mask) + pid_tbl_mask + 1 + next_free_pt;
710 	if ((uint)pid > (uint)pid_max)
711 		pid &= pid_tbl_mask;
712 	p->p_pid = pid;
713 	next_free_pt = nxt & pid_tbl_mask;
714 
715 	/* Grab table slot */
716 	pt->pt_proc = p;
717 	pid_alloc_cnt++;
718 
719 	mutex_exit(proc_lock);
720 
721 	return p;
722 }
723 
724 /*
725  * Free a process id - called from proc_free (in kern_exit.c)
726  *
727  * Called with the proc_lock held.
728  */
729 void
730 proc_free_pid(struct proc *p)
731 {
732 	pid_t pid = p->p_pid;
733 	struct pid_table *pt;
734 
735 	KASSERT(mutex_owned(proc_lock));
736 
737 	pt = &pid_table[pid & pid_tbl_mask];
738 #ifdef DIAGNOSTIC
739 	if (__predict_false(pt->pt_proc != p))
740 		panic("proc_free: pid_table mismatch, pid %x, proc %p",
741 			pid, p);
742 #endif
743 	/* save pid use count in slot */
744 	pt->pt_proc = P_FREE(pid & ~pid_tbl_mask);
745 
746 	if (pt->pt_pgrp == NULL) {
747 		/* link last freed entry onto ours */
748 		pid &= pid_tbl_mask;
749 		pt = &pid_table[last_free_pt];
750 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pid);
751 		last_free_pt = pid;
752 		pid_alloc_cnt--;
753 	}
754 
755 	atomic_dec_uint(&nprocs);
756 }
757 
758 void
759 proc_free_mem(struct proc *p)
760 {
761 
762 	pool_cache_put(proc_cache, p);
763 }
764 
765 /*
766  * proc_enterpgrp: move p to a new or existing process group (and session).
767  *
768  * If we are creating a new pgrp, the pgid should equal
769  * the calling process' pid.
770  * If is only valid to enter a process group that is in the session
771  * of the process.
772  * Also mksess should only be set if we are creating a process group
773  *
774  * Only called from sys_setsid and sys_setpgid.
775  */
776 int
777 proc_enterpgrp(struct proc *curp, pid_t pid, pid_t pgid, bool mksess)
778 {
779 	struct pgrp *new_pgrp, *pgrp;
780 	struct session *sess;
781 	struct proc *p;
782 	int rval;
783 	pid_t pg_id = NO_PGID;
784 
785 	sess = mksess ? kmem_alloc(sizeof(*sess), KM_SLEEP) : NULL;
786 
787 	/* Allocate data areas we might need before doing any validity checks */
788 	mutex_enter(proc_lock);		/* Because pid_table might change */
789 	if (pid_table[pgid & pid_tbl_mask].pt_pgrp == 0) {
790 		mutex_exit(proc_lock);
791 		new_pgrp = kmem_alloc(sizeof(*new_pgrp), KM_SLEEP);
792 		mutex_enter(proc_lock);
793 	} else
794 		new_pgrp = NULL;
795 	rval = EPERM;	/* most common error (to save typing) */
796 
797 	/* Check pgrp exists or can be created */
798 	pgrp = pid_table[pgid & pid_tbl_mask].pt_pgrp;
799 	if (pgrp != NULL && pgrp->pg_id != pgid)
800 		goto done;
801 
802 	/* Can only set another process under restricted circumstances. */
803 	if (pid != curp->p_pid) {
804 		/* must exist and be one of our children... */
805 		if ((p = p_find(pid, PFIND_LOCKED)) == NULL ||
806 		    !p_inferior(p, curp)) {
807 			rval = ESRCH;
808 			goto done;
809 		}
810 		/* ... in the same session... */
811 		if (sess != NULL || p->p_session != curp->p_session)
812 			goto done;
813 		/* ... existing pgid must be in same session ... */
814 		if (pgrp != NULL && pgrp->pg_session != p->p_session)
815 			goto done;
816 		/* ... and not done an exec. */
817 		if (p->p_flag & PK_EXEC) {
818 			rval = EACCES;
819 			goto done;
820 		}
821 	} else {
822 		/* ... setsid() cannot re-enter a pgrp */
823 		if (mksess && (curp->p_pgid == curp->p_pid ||
824 		    pg_find(curp->p_pid, PFIND_LOCKED)))
825 			goto done;
826 		p = curp;
827 	}
828 
829 	/* Changing the process group/session of a session
830 	   leader is definitely off limits. */
831 	if (SESS_LEADER(p)) {
832 		if (sess == NULL && p->p_pgrp == pgrp)
833 			/* unless it's a definite noop */
834 			rval = 0;
835 		goto done;
836 	}
837 
838 	/* Can only create a process group with id of process */
839 	if (pgrp == NULL && pgid != pid)
840 		goto done;
841 
842 	/* Can only create a session if creating pgrp */
843 	if (sess != NULL && pgrp != NULL)
844 		goto done;
845 
846 	/* Check we allocated memory for a pgrp... */
847 	if (pgrp == NULL && new_pgrp == NULL)
848 		goto done;
849 
850 	/* Don't attach to 'zombie' pgrp */
851 	if (pgrp != NULL && LIST_EMPTY(&pgrp->pg_members))
852 		goto done;
853 
854 	/* Expect to succeed now */
855 	rval = 0;
856 
857 	if (pgrp == p->p_pgrp)
858 		/* nothing to do */
859 		goto done;
860 
861 	/* Ok all setup, link up required structures */
862 
863 	if (pgrp == NULL) {
864 		pgrp = new_pgrp;
865 		new_pgrp = NULL;
866 		if (sess != NULL) {
867 			sess->s_sid = p->p_pid;
868 			sess->s_leader = p;
869 			sess->s_count = 1;
870 			sess->s_ttyvp = NULL;
871 			sess->s_ttyp = NULL;
872 			sess->s_flags = p->p_session->s_flags & ~S_LOGIN_SET;
873 			memcpy(sess->s_login, p->p_session->s_login,
874 			    sizeof(sess->s_login));
875 			p->p_lflag &= ~PL_CONTROLT;
876 		} else {
877 			sess = p->p_pgrp->pg_session;
878 			proc_sesshold(sess);
879 		}
880 		pgrp->pg_session = sess;
881 		sess = NULL;
882 
883 		pgrp->pg_id = pgid;
884 		LIST_INIT(&pgrp->pg_members);
885 #ifdef DIAGNOSTIC
886 		if (__predict_false(pid_table[pgid & pid_tbl_mask].pt_pgrp))
887 			panic("enterpgrp: pgrp table slot in use");
888 		if (__predict_false(mksess && p != curp))
889 			panic("enterpgrp: mksession and p != curproc");
890 #endif
891 		pid_table[pgid & pid_tbl_mask].pt_pgrp = pgrp;
892 		pgrp->pg_jobc = 0;
893 	}
894 
895 	/*
896 	 * Adjust eligibility of affected pgrps to participate in job control.
897 	 * Increment eligibility counts before decrementing, otherwise we
898 	 * could reach 0 spuriously during the first call.
899 	 */
900 	fixjobc(p, pgrp, 1);
901 	fixjobc(p, p->p_pgrp, 0);
902 
903 	/* Interlock with ttread(). */
904 	mutex_spin_enter(&tty_lock);
905 
906 	/* Move process to requested group. */
907 	LIST_REMOVE(p, p_pglist);
908 	if (LIST_EMPTY(&p->p_pgrp->pg_members))
909 		/* defer delete until we've dumped the lock */
910 		pg_id = p->p_pgrp->pg_id;
911 	p->p_pgrp = pgrp;
912 	LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
913 
914 	/* Done with the swap; we can release the tty mutex. */
915 	mutex_spin_exit(&tty_lock);
916 
917     done:
918 	if (pg_id != NO_PGID) {
919 		/* Releases proc_lock. */
920 		pg_delete(pg_id);
921 	} else {
922 		mutex_exit(proc_lock);
923 	}
924 	if (sess != NULL)
925 		kmem_free(sess, sizeof(*sess));
926 	if (new_pgrp != NULL)
927 		kmem_free(new_pgrp, sizeof(*new_pgrp));
928 #ifdef DEBUG_PGRP
929 	if (__predict_false(rval))
930 		printf("enterpgrp(%d,%d,%d), curproc %d, rval %d\n",
931 			pid, pgid, mksess, curp->p_pid, rval);
932 #endif
933 	return rval;
934 }
935 
936 /*
937  * proc_leavepgrp: remove a process from its process group.
938  *  => must be called with the proc_lock held, which will be released;
939  */
940 void
941 proc_leavepgrp(struct proc *p)
942 {
943 	struct pgrp *pgrp;
944 
945 	KASSERT(mutex_owned(proc_lock));
946 
947 	/* Interlock with ttread() */
948 	mutex_spin_enter(&tty_lock);
949 	pgrp = p->p_pgrp;
950 	LIST_REMOVE(p, p_pglist);
951 	p->p_pgrp = NULL;
952 	mutex_spin_exit(&tty_lock);
953 
954 	if (LIST_EMPTY(&pgrp->pg_members)) {
955 		/* Releases proc_lock. */
956 		pg_delete(pgrp->pg_id);
957 	} else {
958 		mutex_exit(proc_lock);
959 	}
960 }
961 
962 /*
963  * pg_remove: remove a process group from the table.
964  *  => must be called with the proc_lock held;
965  *  => returns process group to free;
966  */
967 static struct pgrp *
968 pg_remove(pid_t pg_id)
969 {
970 	struct pgrp *pgrp;
971 	struct pid_table *pt;
972 
973 	KASSERT(mutex_owned(proc_lock));
974 
975 	pt = &pid_table[pg_id & pid_tbl_mask];
976 	pgrp = pt->pt_pgrp;
977 
978 	KASSERT(pgrp != NULL);
979 	KASSERT(pgrp->pg_id == pg_id);
980 	KASSERT(LIST_EMPTY(&pgrp->pg_members));
981 
982 	pt->pt_pgrp = NULL;
983 
984 	if (!P_VALID(pt->pt_proc)) {
985 		/* Orphaned pgrp, put slot onto free list. */
986 		KASSERT((P_NEXT(pt->pt_proc) & pid_tbl_mask) == 0);
987 		pg_id &= pid_tbl_mask;
988 		pt = &pid_table[last_free_pt];
989 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pg_id);
990 		last_free_pt = pg_id;
991 		pid_alloc_cnt--;
992 	}
993 	return pgrp;
994 }
995 
996 /*
997  * pg_delete: delete and free a process group.
998  *  => must be called with the proc_lock held, which will be released.
999  */
1000 static void
1001 pg_delete(pid_t pg_id)
1002 {
1003 	struct pgrp *pg;
1004 	struct tty *ttyp;
1005 	struct session *ss;
1006 
1007 	KASSERT(mutex_owned(proc_lock));
1008 
1009 	pg = pid_table[pg_id & pid_tbl_mask].pt_pgrp;
1010 	if (pg == NULL || pg->pg_id != pg_id || !LIST_EMPTY(&pg->pg_members)) {
1011 		mutex_exit(proc_lock);
1012 		return;
1013 	}
1014 
1015 	ss = pg->pg_session;
1016 
1017 	/* Remove reference (if any) from tty to this process group */
1018 	mutex_spin_enter(&tty_lock);
1019 	ttyp = ss->s_ttyp;
1020 	if (ttyp != NULL && ttyp->t_pgrp == pg) {
1021 		ttyp->t_pgrp = NULL;
1022 		KASSERT(ttyp->t_session == ss);
1023 	}
1024 	mutex_spin_exit(&tty_lock);
1025 
1026 	/*
1027 	 * The leading process group in a session is freed by proc_sessrele(),
1028 	 * if last reference.  Note: proc_sessrele() releases proc_lock.
1029 	 */
1030 	pg = (ss->s_sid != pg->pg_id) ? pg_remove(pg_id) : NULL;
1031 	proc_sessrele(ss);
1032 
1033 	if (pg != NULL) {
1034 		/* Free it, if was not done by proc_sessrele(). */
1035 		kmem_free(pg, sizeof(struct pgrp));
1036 	}
1037 }
1038 
1039 /*
1040  * Adjust pgrp jobc counters when specified process changes process group.
1041  * We count the number of processes in each process group that "qualify"
1042  * the group for terminal job control (those with a parent in a different
1043  * process group of the same session).  If that count reaches zero, the
1044  * process group becomes orphaned.  Check both the specified process'
1045  * process group and that of its children.
1046  * entering == 0 => p is leaving specified group.
1047  * entering == 1 => p is entering specified group.
1048  *
1049  * Call with proc_lock held.
1050  */
1051 void
1052 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
1053 {
1054 	struct pgrp *hispgrp;
1055 	struct session *mysession = pgrp->pg_session;
1056 	struct proc *child;
1057 
1058 	KASSERT(mutex_owned(proc_lock));
1059 
1060 	/*
1061 	 * Check p's parent to see whether p qualifies its own process
1062 	 * group; if so, adjust count for p's process group.
1063 	 */
1064 	hispgrp = p->p_pptr->p_pgrp;
1065 	if (hispgrp != pgrp && hispgrp->pg_session == mysession) {
1066 		if (entering) {
1067 			pgrp->pg_jobc++;
1068 			p->p_lflag &= ~PL_ORPHANPG;
1069 		} else if (--pgrp->pg_jobc == 0)
1070 			orphanpg(pgrp);
1071 	}
1072 
1073 	/*
1074 	 * Check this process' children to see whether they qualify
1075 	 * their process groups; if so, adjust counts for children's
1076 	 * process groups.
1077 	 */
1078 	LIST_FOREACH(child, &p->p_children, p_sibling) {
1079 		hispgrp = child->p_pgrp;
1080 		if (hispgrp != pgrp && hispgrp->pg_session == mysession &&
1081 		    !P_ZOMBIE(child)) {
1082 			if (entering) {
1083 				child->p_lflag &= ~PL_ORPHANPG;
1084 				hispgrp->pg_jobc++;
1085 			} else if (--hispgrp->pg_jobc == 0)
1086 				orphanpg(hispgrp);
1087 		}
1088 	}
1089 }
1090 
1091 /*
1092  * A process group has become orphaned;
1093  * if there are any stopped processes in the group,
1094  * hang-up all process in that group.
1095  *
1096  * Call with proc_lock held.
1097  */
1098 static void
1099 orphanpg(struct pgrp *pg)
1100 {
1101 	struct proc *p;
1102 	int doit;
1103 
1104 	KASSERT(mutex_owned(proc_lock));
1105 
1106 	doit = 0;
1107 
1108 	LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1109 		if (p->p_stat == SSTOP) {
1110 			p->p_lflag |= PL_ORPHANPG;
1111 			psignal(p, SIGHUP);
1112 			psignal(p, SIGCONT);
1113 		}
1114 	}
1115 }
1116 
1117 #ifdef DDB
1118 #include <ddb/db_output.h>
1119 void pidtbl_dump(void);
1120 void
1121 pidtbl_dump(void)
1122 {
1123 	struct pid_table *pt;
1124 	struct proc *p;
1125 	struct pgrp *pgrp;
1126 	int id;
1127 
1128 	db_printf("pid table %p size %x, next %x, last %x\n",
1129 		pid_table, pid_tbl_mask+1,
1130 		next_free_pt, last_free_pt);
1131 	for (pt = pid_table, id = 0; id <= pid_tbl_mask; id++, pt++) {
1132 		p = pt->pt_proc;
1133 		if (!P_VALID(p) && !pt->pt_pgrp)
1134 			continue;
1135 		db_printf("  id %x: ", id);
1136 		if (P_VALID(p))
1137 			db_printf("proc %p id %d (0x%x) %s\n",
1138 				p, p->p_pid, p->p_pid, p->p_comm);
1139 		else
1140 			db_printf("next %x use %x\n",
1141 				P_NEXT(p) & pid_tbl_mask,
1142 				P_NEXT(p) & ~pid_tbl_mask);
1143 		if ((pgrp = pt->pt_pgrp)) {
1144 			db_printf("\tsession %p, sid %d, count %d, login %s\n",
1145 			    pgrp->pg_session, pgrp->pg_session->s_sid,
1146 			    pgrp->pg_session->s_count,
1147 			    pgrp->pg_session->s_login);
1148 			db_printf("\tpgrp %p, pg_id %d, pg_jobc %d, members %p\n",
1149 			    pgrp, pgrp->pg_id, pgrp->pg_jobc,
1150 			    LIST_FIRST(&pgrp->pg_members));
1151 			LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1152 				db_printf("\t\tpid %d addr %p pgrp %p %s\n",
1153 				    p->p_pid, p, p->p_pgrp, p->p_comm);
1154 			}
1155 		}
1156 	}
1157 }
1158 #endif /* DDB */
1159 
1160 #ifdef KSTACK_CHECK_MAGIC
1161 #include <sys/user.h>
1162 
1163 #define	KSTACK_MAGIC	0xdeadbeaf
1164 
1165 /* XXX should be per process basis? */
1166 static int	kstackleftmin = KSTACK_SIZE;
1167 static int	kstackleftthres = KSTACK_SIZE / 8;
1168 
1169 void
1170 kstack_setup_magic(const struct lwp *l)
1171 {
1172 	uint32_t *ip;
1173 	uint32_t const *end;
1174 
1175 	KASSERT(l != NULL);
1176 	KASSERT(l != &lwp0);
1177 
1178 	/*
1179 	 * fill all the stack with magic number
1180 	 * so that later modification on it can be detected.
1181 	 */
1182 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1183 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1184 	for (; ip < end; ip++) {
1185 		*ip = KSTACK_MAGIC;
1186 	}
1187 }
1188 
1189 void
1190 kstack_check_magic(const struct lwp *l)
1191 {
1192 	uint32_t const *ip, *end;
1193 	int stackleft;
1194 
1195 	KASSERT(l != NULL);
1196 
1197 	/* don't check proc0 */ /*XXX*/
1198 	if (l == &lwp0)
1199 		return;
1200 
1201 #ifdef __MACHINE_STACK_GROWS_UP
1202 	/* stack grows upwards (eg. hppa) */
1203 	ip = (uint32_t *)((void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1204 	end = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1205 	for (ip--; ip >= end; ip--)
1206 		if (*ip != KSTACK_MAGIC)
1207 			break;
1208 
1209 	stackleft = (void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE - (void *)ip;
1210 #else /* __MACHINE_STACK_GROWS_UP */
1211 	/* stack grows downwards (eg. i386) */
1212 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1213 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1214 	for (; ip < end; ip++)
1215 		if (*ip != KSTACK_MAGIC)
1216 			break;
1217 
1218 	stackleft = ((const char *)ip) - (const char *)KSTACK_LOWEST_ADDR(l);
1219 #endif /* __MACHINE_STACK_GROWS_UP */
1220 
1221 	if (kstackleftmin > stackleft) {
1222 		kstackleftmin = stackleft;
1223 		if (stackleft < kstackleftthres)
1224 			printf("warning: kernel stack left %d bytes"
1225 			    "(pid %u:lid %u)\n", stackleft,
1226 			    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1227 	}
1228 
1229 	if (stackleft <= 0) {
1230 		panic("magic on the top of kernel stack changed for "
1231 		    "pid %u, lid %u: maybe kernel stack overflow",
1232 		    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1233 	}
1234 }
1235 #endif /* KSTACK_CHECK_MAGIC */
1236 
1237 int
1238 proclist_foreach_call(struct proclist *list,
1239     int (*callback)(struct proc *, void *arg), void *arg)
1240 {
1241 	struct proc marker;
1242 	struct proc *p;
1243 	struct lwp * const l = curlwp;
1244 	int ret = 0;
1245 
1246 	marker.p_flag = PK_MARKER;
1247 	uvm_lwp_hold(l);
1248 	mutex_enter(proc_lock);
1249 	for (p = LIST_FIRST(list); ret == 0 && p != NULL;) {
1250 		if (p->p_flag & PK_MARKER) {
1251 			p = LIST_NEXT(p, p_list);
1252 			continue;
1253 		}
1254 		LIST_INSERT_AFTER(p, &marker, p_list);
1255 		ret = (*callback)(p, arg);
1256 		KASSERT(mutex_owned(proc_lock));
1257 		p = LIST_NEXT(&marker, p_list);
1258 		LIST_REMOVE(&marker, p_list);
1259 	}
1260 	mutex_exit(proc_lock);
1261 	uvm_lwp_rele(l);
1262 
1263 	return ret;
1264 }
1265 
1266 int
1267 proc_vmspace_getref(struct proc *p, struct vmspace **vm)
1268 {
1269 
1270 	/* XXXCDC: how should locking work here? */
1271 
1272 	/* curproc exception is for coredump. */
1273 
1274 	if ((p != curproc && (p->p_sflag & PS_WEXIT) != 0) ||
1275 	    (p->p_vmspace->vm_refcnt < 1)) { /* XXX */
1276 		return EFAULT;
1277 	}
1278 
1279 	uvmspace_addref(p->p_vmspace);
1280 	*vm = p->p_vmspace;
1281 
1282 	return 0;
1283 }
1284 
1285 /*
1286  * Acquire a write lock on the process credential.
1287  */
1288 void
1289 proc_crmod_enter(void)
1290 {
1291 	struct lwp *l = curlwp;
1292 	struct proc *p = l->l_proc;
1293 	struct plimit *lim;
1294 	kauth_cred_t oc;
1295 	char *cn;
1296 
1297 	/* Reset what needs to be reset in plimit. */
1298 	if (p->p_limit->pl_corename != defcorename) {
1299 		lim_privatise(p, false);
1300 		lim = p->p_limit;
1301 		mutex_enter(&lim->pl_lock);
1302 		cn = lim->pl_corename;
1303 		lim->pl_corename = defcorename;
1304 		mutex_exit(&lim->pl_lock);
1305 		if (cn != defcorename)
1306 			free(cn, M_TEMP);
1307 	}
1308 
1309 	mutex_enter(p->p_lock);
1310 
1311 	/* Ensure the LWP cached credentials are up to date. */
1312 	if ((oc = l->l_cred) != p->p_cred) {
1313 		kauth_cred_hold(p->p_cred);
1314 		l->l_cred = p->p_cred;
1315 		kauth_cred_free(oc);
1316 	}
1317 
1318 }
1319 
1320 /*
1321  * Set in a new process credential, and drop the write lock.  The credential
1322  * must have a reference already.  Optionally, free a no-longer required
1323  * credential.  The scheduler also needs to inspect p_cred, so we also
1324  * briefly acquire the sched state mutex.
1325  */
1326 void
1327 proc_crmod_leave(kauth_cred_t scred, kauth_cred_t fcred, bool sugid)
1328 {
1329 	struct lwp *l = curlwp, *l2;
1330 	struct proc *p = l->l_proc;
1331 	kauth_cred_t oc;
1332 
1333 	KASSERT(mutex_owned(p->p_lock));
1334 
1335 	/* Is there a new credential to set in? */
1336 	if (scred != NULL) {
1337 		p->p_cred = scred;
1338 		LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
1339 			if (l2 != l)
1340 				l2->l_prflag |= LPR_CRMOD;
1341 		}
1342 
1343 		/* Ensure the LWP cached credentials are up to date. */
1344 		if ((oc = l->l_cred) != scred) {
1345 			kauth_cred_hold(scred);
1346 			l->l_cred = scred;
1347 		}
1348 	} else
1349 		oc = NULL;	/* XXXgcc */
1350 
1351 	if (sugid) {
1352 		/*
1353 		 * Mark process as having changed credentials, stops
1354 		 * tracing etc.
1355 		 */
1356 		p->p_flag |= PK_SUGID;
1357 	}
1358 
1359 	mutex_exit(p->p_lock);
1360 
1361 	/* If there is a credential to be released, free it now. */
1362 	if (fcred != NULL) {
1363 		KASSERT(scred != NULL);
1364 		kauth_cred_free(fcred);
1365 		if (oc != scred)
1366 			kauth_cred_free(oc);
1367 	}
1368 }
1369 
1370 /*
1371  * proc_specific_key_create --
1372  *	Create a key for subsystem proc-specific data.
1373  */
1374 int
1375 proc_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1376 {
1377 
1378 	return (specificdata_key_create(proc_specificdata_domain, keyp, dtor));
1379 }
1380 
1381 /*
1382  * proc_specific_key_delete --
1383  *	Delete a key for subsystem proc-specific data.
1384  */
1385 void
1386 proc_specific_key_delete(specificdata_key_t key)
1387 {
1388 
1389 	specificdata_key_delete(proc_specificdata_domain, key);
1390 }
1391 
1392 /*
1393  * proc_initspecific --
1394  *	Initialize a proc's specificdata container.
1395  */
1396 void
1397 proc_initspecific(struct proc *p)
1398 {
1399 	int error;
1400 
1401 	error = specificdata_init(proc_specificdata_domain, &p->p_specdataref);
1402 	KASSERT(error == 0);
1403 }
1404 
1405 /*
1406  * proc_finispecific --
1407  *	Finalize a proc's specificdata container.
1408  */
1409 void
1410 proc_finispecific(struct proc *p)
1411 {
1412 
1413 	specificdata_fini(proc_specificdata_domain, &p->p_specdataref);
1414 }
1415 
1416 /*
1417  * proc_getspecific --
1418  *	Return proc-specific data corresponding to the specified key.
1419  */
1420 void *
1421 proc_getspecific(struct proc *p, specificdata_key_t key)
1422 {
1423 
1424 	return (specificdata_getspecific(proc_specificdata_domain,
1425 					 &p->p_specdataref, key));
1426 }
1427 
1428 /*
1429  * proc_setspecific --
1430  *	Set proc-specific data corresponding to the specified key.
1431  */
1432 void
1433 proc_setspecific(struct proc *p, specificdata_key_t key, void *data)
1434 {
1435 
1436 	specificdata_setspecific(proc_specificdata_domain,
1437 				 &p->p_specdataref, key, data);
1438 }
1439 
1440 int
1441 proc_uidmatch(kauth_cred_t cred, kauth_cred_t target)
1442 {
1443 	int r = 0;
1444 
1445 	if (kauth_cred_getuid(cred) != kauth_cred_getuid(target) ||
1446 	    kauth_cred_getuid(cred) != kauth_cred_getsvuid(target)) {
1447 		/*
1448 		 * suid proc of ours or proc not ours
1449 		 */
1450 		r = EPERM;
1451 	} else if (kauth_cred_getgid(target) != kauth_cred_getsvgid(target)) {
1452 		/*
1453 		 * sgid proc has sgid back to us temporarily
1454 		 */
1455 		r = EPERM;
1456 	} else {
1457 		/*
1458 		 * our rgid must be in target's group list (ie,
1459 		 * sub-processes started by a sgid process)
1460 		 */
1461 		int ismember = 0;
1462 
1463 		if (kauth_cred_ismember_gid(cred,
1464 		    kauth_cred_getgid(target), &ismember) != 0 ||
1465 		    !ismember)
1466 			r = EPERM;
1467 	}
1468 
1469 	return (r);
1470 }
1471 
1472