xref: /netbsd-src/sys/kern/kern_proc.c (revision 1ca06f9c9235889e2ff6dc77279d01d151d70a9a)
1 /*	$NetBSD: kern_proc.c,v 1.158 2009/11/26 00:19:11 matt 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 1.158 2009/11/26 00:19:11 matt 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_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 int nofile = NOFILE;
211 int maxuprc = MAXUPRC;
212 int cmask = CMASK;
213 
214 MALLOC_DEFINE(M_EMULDATA, "emuldata", "Per-process emulation data");
215 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures");
216 
217 /*
218  * The process list descriptors, used during pid allocation and
219  * by sysctl.  No locking on this data structure is needed since
220  * it is completely static.
221  */
222 const struct proclist_desc proclists[] = {
223 	{ &allproc	},
224 	{ &zombproc	},
225 	{ NULL		},
226 };
227 
228 static struct pgrp *	pg_remove(pid_t);
229 static void		pg_delete(pid_t);
230 static void		orphanpg(struct pgrp *);
231 
232 static specificdata_domain_t proc_specificdata_domain;
233 
234 static pool_cache_t proc_cache;
235 
236 static kauth_listener_t proc_listener;
237 
238 static int
239 proc_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
240     void *arg0, void *arg1, void *arg2, void *arg3)
241 {
242 	struct proc *p;
243 	int result;
244 
245 	result = KAUTH_RESULT_DEFER;
246 	p = arg0;
247 
248 	switch (action) {
249 	case KAUTH_PROCESS_CANSEE: {
250 		enum kauth_process_req req;
251 
252 		req = (enum kauth_process_req)arg1;
253 
254 		switch (req) {
255 		case KAUTH_REQ_PROCESS_CANSEE_ARGS:
256 		case KAUTH_REQ_PROCESS_CANSEE_ENTRY:
257 		case KAUTH_REQ_PROCESS_CANSEE_OPENFILES:
258 			result = KAUTH_RESULT_ALLOW;
259 
260 			break;
261 
262 		case KAUTH_REQ_PROCESS_CANSEE_ENV:
263 			if (kauth_cred_getuid(cred) !=
264 			    kauth_cred_getuid(p->p_cred) ||
265 			    kauth_cred_getuid(cred) !=
266 			    kauth_cred_getsvuid(p->p_cred))
267 				break;
268 
269 			result = KAUTH_RESULT_ALLOW;
270 
271 			break;
272 
273 		default:
274 			break;
275 		}
276 
277 		break;
278 		}
279 
280 	case KAUTH_PROCESS_FORK: {
281 		int lnprocs = (int)(unsigned long)arg2;
282 
283 		/*
284 		 * Don't allow a nonprivileged user to use the last few
285 		 * processes. The variable lnprocs is the current number of
286 		 * processes, maxproc is the limit.
287 		 */
288 		if (__predict_false((lnprocs >= maxproc - 5)))
289 			break;
290 
291 		result = KAUTH_RESULT_ALLOW;
292 
293 		break;
294 		}
295 
296 	case KAUTH_PROCESS_CORENAME:
297 	case KAUTH_PROCESS_STOPFLAG:
298 		if (proc_uidmatch(cred, p->p_cred) == 0)
299 			result = KAUTH_RESULT_ALLOW;
300 
301 		break;
302 
303 	default:
304 		break;
305 	}
306 
307 	return result;
308 }
309 
310 /*
311  * Initialize global process hashing structures.
312  */
313 void
314 procinit(void)
315 {
316 	const struct proclist_desc *pd;
317 	u_int i;
318 #define	LINK_EMPTY ((PID_MAX + INITIAL_PID_TABLE_SIZE) & ~(INITIAL_PID_TABLE_SIZE - 1))
319 
320 	for (pd = proclists; pd->pd_list != NULL; pd++)
321 		LIST_INIT(pd->pd_list);
322 
323 	proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
324 	pid_table = kmem_alloc(INITIAL_PID_TABLE_SIZE
325 	    * sizeof(struct pid_table), KM_SLEEP);
326 
327 	/* Set free list running through table...
328 	   Preset 'use count' above PID_MAX so we allocate pid 1 next. */
329 	for (i = 0; i <= pid_tbl_mask; i++) {
330 		pid_table[i].pt_proc = P_FREE(LINK_EMPTY + i + 1);
331 		pid_table[i].pt_pgrp = 0;
332 	}
333 	/* slot 0 is just grabbed */
334 	next_free_pt = 1;
335 	/* Need to fix last entry. */
336 	last_free_pt = pid_tbl_mask;
337 	pid_table[last_free_pt].pt_proc = P_FREE(LINK_EMPTY);
338 	/* point at which we grow table - to avoid reusing pids too often */
339 	pid_alloc_lim = pid_tbl_mask - 1;
340 #undef LINK_EMPTY
341 
342 	proc_specificdata_domain = specificdata_domain_create();
343 	KASSERT(proc_specificdata_domain != NULL);
344 
345 	proc_cache = pool_cache_init(sizeof(struct proc), 0, 0, 0,
346 	    "procpl", NULL, IPL_NONE, NULL, NULL, NULL);
347 
348 	proc_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS,
349 	    proc_listener_cb, NULL);
350 }
351 
352 /*
353  * Initialize process 0.
354  */
355 void
356 proc0_init(void)
357 {
358 	struct proc *p;
359 	struct pgrp *pg;
360 	struct lwp *l;
361 	rlim_t lim;
362 	int i;
363 
364 	p = &proc0;
365 	pg = &pgrp0;
366 	l = &lwp0;
367 
368 	KASSERT(l->l_addr != NULL);
369 	KASSERT(l->l_lid == p->p_nlwpid);
370 
371 	mutex_init(&p->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
372 	mutex_init(&p->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
373 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
374 
375 	rw_init(&p->p_reflock);
376 	cv_init(&p->p_waitcv, "wait");
377 	cv_init(&p->p_lwpcv, "lwpwait");
378 
379 	LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling);
380 
381 	pid_table[0].pt_proc = p;
382 	LIST_INSERT_HEAD(&allproc, p, p_list);
383 	LIST_INSERT_HEAD(&alllwp, l, l_list);
384 
385 	pid_table[0].pt_pgrp = pg;
386 	LIST_INSERT_HEAD(&pg->pg_members, p, p_pglist);
387 
388 #ifdef __HAVE_SYSCALL_INTERN
389 	(*p->p_emul->e_syscall_intern)(p);
390 #endif
391 
392 	callout_init(&l->l_timeout_ch, CALLOUT_MPSAFE);
393 	callout_setfunc(&l->l_timeout_ch, sleepq_timeout, l);
394 	cv_init(&l->l_sigcv, "sigwait");
395 
396 	/* Create credentials. */
397 	cred0 = kauth_cred_alloc();
398 	p->p_cred = cred0;
399 	kauth_cred_hold(cred0);
400 	l->l_cred = cred0;
401 
402 	/* Create the CWD info. */
403 	rw_init(&cwdi0.cwdi_lock);
404 
405 	/* Create the limits structures. */
406 	mutex_init(&limit0.pl_lock, MUTEX_DEFAULT, IPL_NONE);
407 	for (i = 0; i < __arraycount(limit0.pl_rlimit); i++)
408 		limit0.pl_rlimit[i].rlim_cur =
409 		    limit0.pl_rlimit[i].rlim_max = RLIM_INFINITY;
410 
411 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
412 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_cur =
413 	    maxfiles < nofile ? maxfiles : nofile;
414 
415 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
416 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_cur =
417 	    maxproc < maxuprc ? maxproc : maxuprc;
418 
419 	lim = ptoa(uvmexp.free);
420 	limit0.pl_rlimit[RLIMIT_RSS].rlim_max = lim;
421 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim;
422 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3;
423 	limit0.pl_corename = defcorename;
424 	limit0.pl_refcnt = 1;
425 	limit0.pl_sv_limit = NULL;
426 
427 	/* Configure virtual memory system, set vm rlimits. */
428 	uvm_init_limits(p);
429 
430 	/* Initialize file descriptor table for proc0. */
431 	fd_init(&filedesc0);
432 
433 	/*
434 	 * Initialize proc0's vmspace, which uses the kernel pmap.
435 	 * All kernel processes (which never have user space mappings)
436 	 * share proc0's vmspace, and thus, the kernel pmap.
437 	 */
438 	uvmspace_init(&vmspace0, pmap_kernel(), round_page(VM_MIN_ADDRESS),
439 	    trunc_page(VM_MAX_ADDRESS));
440 
441 	/* Initialize signal state for proc0. XXX IPL_SCHED */
442 	mutex_init(&p->p_sigacts->sa_mutex, MUTEX_DEFAULT, IPL_SCHED);
443 	siginit(p);
444 
445 	proc_initspecific(p);
446 	lwp_initspecific(l);
447 
448 	SYSCALL_TIME_LWP_INIT(l);
449 }
450 
451 /*
452  * Session reference counting.
453  */
454 
455 void
456 proc_sesshold(struct session *ss)
457 {
458 
459 	KASSERT(mutex_owned(proc_lock));
460 	ss->s_count++;
461 }
462 
463 void
464 proc_sessrele(struct session *ss)
465 {
466 
467 	KASSERT(mutex_owned(proc_lock));
468 	/*
469 	 * We keep the pgrp with the same id as the session in order to
470 	 * stop a process being given the same pid.  Since the pgrp holds
471 	 * a reference to the session, it must be a 'zombie' pgrp by now.
472 	 */
473 	if (--ss->s_count == 0) {
474 		struct pgrp *pg;
475 
476 		pg = pg_remove(ss->s_sid);
477 		mutex_exit(proc_lock);
478 
479 		kmem_free(pg, sizeof(struct pgrp));
480 		kmem_free(ss, sizeof(struct session));
481 	} else {
482 		mutex_exit(proc_lock);
483 	}
484 }
485 
486 /*
487  * Check that the specified process group is in the session of the
488  * specified process.
489  * Treats -ve ids as process ids.
490  * Used to validate TIOCSPGRP requests.
491  */
492 int
493 pgid_in_session(struct proc *p, pid_t pg_id)
494 {
495 	struct pgrp *pgrp;
496 	struct session *session;
497 	int error;
498 
499 	mutex_enter(proc_lock);
500 	if (pg_id < 0) {
501 		struct proc *p1 = p_find(-pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
502 		if (p1 == NULL)
503 			return EINVAL;
504 		pgrp = p1->p_pgrp;
505 	} else {
506 		pgrp = pg_find(pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
507 		if (pgrp == NULL)
508 			return EINVAL;
509 	}
510 	session = pgrp->pg_session;
511 	if (session != p->p_pgrp->pg_session)
512 		error = EPERM;
513 	else
514 		error = 0;
515 	mutex_exit(proc_lock);
516 
517 	return error;
518 }
519 
520 /*
521  * p_inferior: is p an inferior of q?
522  */
523 static inline bool
524 p_inferior(struct proc *p, struct proc *q)
525 {
526 
527 	KASSERT(mutex_owned(proc_lock));
528 
529 	for (; p != q; p = p->p_pptr)
530 		if (p->p_pid == 0)
531 			return false;
532 	return true;
533 }
534 
535 /*
536  * Locate a process by number
537  */
538 struct proc *
539 p_find(pid_t pid, uint flags)
540 {
541 	struct proc *p;
542 	char stat;
543 
544 	if (!(flags & PFIND_LOCKED))
545 		mutex_enter(proc_lock);
546 
547 	p = pid_table[pid & pid_tbl_mask].pt_proc;
548 
549 	/* Only allow live processes to be found by pid. */
550 	/* XXXSMP p_stat */
551 	if (P_VALID(p) && p->p_pid == pid && ((stat = p->p_stat) == SACTIVE ||
552 	    stat == SSTOP || ((flags & PFIND_ZOMBIE) &&
553 	    (stat == SZOMB || stat == SDEAD || stat == SDYING)))) {
554 		if (flags & PFIND_UNLOCK_OK)
555 			 mutex_exit(proc_lock);
556 		return p;
557 	}
558 	if (flags & PFIND_UNLOCK_FAIL)
559 		mutex_exit(proc_lock);
560 	return NULL;
561 }
562 
563 
564 /*
565  * Locate a process group by number
566  */
567 struct pgrp *
568 pg_find(pid_t pgid, uint flags)
569 {
570 	struct pgrp *pg;
571 
572 	if (!(flags & PFIND_LOCKED))
573 		mutex_enter(proc_lock);
574 	pg = pid_table[pgid & pid_tbl_mask].pt_pgrp;
575 	/*
576 	 * Can't look up a pgrp that only exists because the session
577 	 * hasn't died yet (traditional)
578 	 */
579 	if (pg == NULL || pg->pg_id != pgid || LIST_EMPTY(&pg->pg_members)) {
580 		if (flags & PFIND_UNLOCK_FAIL)
581 			 mutex_exit(proc_lock);
582 		return NULL;
583 	}
584 
585 	if (flags & PFIND_UNLOCK_OK)
586 		mutex_exit(proc_lock);
587 	return pg;
588 }
589 
590 static void
591 expand_pid_table(void)
592 {
593 	size_t pt_size, tsz;
594 	struct pid_table *n_pt, *new_pt;
595 	struct proc *proc;
596 	struct pgrp *pgrp;
597 	pid_t pid;
598 	u_int i;
599 
600 	pt_size = pid_tbl_mask + 1;
601 	tsz = pt_size * 2 * sizeof(struct pid_table);
602 	new_pt = kmem_alloc(tsz, KM_SLEEP);
603 
604 	mutex_enter(proc_lock);
605 	if (pt_size != pid_tbl_mask + 1) {
606 		/* Another process beat us to it... */
607 		mutex_exit(proc_lock);
608 		kmem_free(new_pt, tsz);
609 		return;
610 	}
611 
612 	/*
613 	 * Copy entries from old table into new one.
614 	 * If 'pid' is 'odd' we need to place in the upper half,
615 	 * even pid's to the lower half.
616 	 * Free items stay in the low half so we don't have to
617 	 * fixup the reference to them.
618 	 * We stuff free items on the front of the freelist
619 	 * because we can't write to unmodified entries.
620 	 * Processing the table backwards maintains a semblance
621 	 * of issueing pid numbers that increase with time.
622 	 */
623 	i = pt_size - 1;
624 	n_pt = new_pt + i;
625 	for (; ; i--, n_pt--) {
626 		proc = pid_table[i].pt_proc;
627 		pgrp = pid_table[i].pt_pgrp;
628 		if (!P_VALID(proc)) {
629 			/* Up 'use count' so that link is valid */
630 			pid = (P_NEXT(proc) + pt_size) & ~pt_size;
631 			proc = P_FREE(pid);
632 			if (pgrp)
633 				pid = pgrp->pg_id;
634 		} else
635 			pid = proc->p_pid;
636 
637 		/* Save entry in appropriate half of table */
638 		n_pt[pid & pt_size].pt_proc = proc;
639 		n_pt[pid & pt_size].pt_pgrp = pgrp;
640 
641 		/* Put other piece on start of free list */
642 		pid = (pid ^ pt_size) & ~pid_tbl_mask;
643 		n_pt[pid & pt_size].pt_proc =
644 				    P_FREE((pid & ~pt_size) | next_free_pt);
645 		n_pt[pid & pt_size].pt_pgrp = 0;
646 		next_free_pt = i | (pid & pt_size);
647 		if (i == 0)
648 			break;
649 	}
650 
651 	/* Save old table size and switch tables */
652 	tsz = pt_size * sizeof(struct pid_table);
653 	n_pt = pid_table;
654 	pid_table = new_pt;
655 	pid_tbl_mask = pt_size * 2 - 1;
656 
657 	/*
658 	 * pid_max starts as PID_MAX (= 30000), once we have 16384
659 	 * allocated pids we need it to be larger!
660 	 */
661 	if (pid_tbl_mask > PID_MAX) {
662 		pid_max = pid_tbl_mask * 2 + 1;
663 		pid_alloc_lim |= pid_alloc_lim << 1;
664 	} else
665 		pid_alloc_lim <<= 1;	/* doubles number of free slots... */
666 
667 	mutex_exit(proc_lock);
668 	kmem_free(n_pt, tsz);
669 }
670 
671 struct proc *
672 proc_alloc(void)
673 {
674 	struct proc *p;
675 	int nxt;
676 	pid_t pid;
677 	struct pid_table *pt;
678 
679 	p = pool_cache_get(proc_cache, PR_WAITOK);
680 	p->p_stat = SIDL;			/* protect against others */
681 
682 	proc_initspecific(p);
683 	/* allocate next free pid */
684 
685 	for (;;expand_pid_table()) {
686 		if (__predict_false(pid_alloc_cnt >= pid_alloc_lim))
687 			/* ensure pids cycle through 2000+ values */
688 			continue;
689 		mutex_enter(proc_lock);
690 		pt = &pid_table[next_free_pt];
691 #ifdef DIAGNOSTIC
692 		if (__predict_false(P_VALID(pt->pt_proc) || pt->pt_pgrp))
693 			panic("proc_alloc: slot busy");
694 #endif
695 		nxt = P_NEXT(pt->pt_proc);
696 		if (nxt & pid_tbl_mask)
697 			break;
698 		/* Table full - expand (NB last entry not used....) */
699 		mutex_exit(proc_lock);
700 	}
701 
702 	/* pid is 'saved use count' + 'size' + entry */
703 	pid = (nxt & ~pid_tbl_mask) + pid_tbl_mask + 1 + next_free_pt;
704 	if ((uint)pid > (uint)pid_max)
705 		pid &= pid_tbl_mask;
706 	p->p_pid = pid;
707 	next_free_pt = nxt & pid_tbl_mask;
708 
709 	/* Grab table slot */
710 	pt->pt_proc = p;
711 	pid_alloc_cnt++;
712 
713 	mutex_exit(proc_lock);
714 
715 	return p;
716 }
717 
718 /*
719  * Free a process id - called from proc_free (in kern_exit.c)
720  *
721  * Called with the proc_lock held.
722  */
723 void
724 proc_free_pid(struct proc *p)
725 {
726 	pid_t pid = p->p_pid;
727 	struct pid_table *pt;
728 
729 	KASSERT(mutex_owned(proc_lock));
730 
731 	pt = &pid_table[pid & pid_tbl_mask];
732 #ifdef DIAGNOSTIC
733 	if (__predict_false(pt->pt_proc != p))
734 		panic("proc_free: pid_table mismatch, pid %x, proc %p",
735 			pid, p);
736 #endif
737 	/* save pid use count in slot */
738 	pt->pt_proc = P_FREE(pid & ~pid_tbl_mask);
739 
740 	if (pt->pt_pgrp == NULL) {
741 		/* link last freed entry onto ours */
742 		pid &= pid_tbl_mask;
743 		pt = &pid_table[last_free_pt];
744 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pid);
745 		last_free_pt = pid;
746 		pid_alloc_cnt--;
747 	}
748 
749 	atomic_dec_uint(&nprocs);
750 }
751 
752 void
753 proc_free_mem(struct proc *p)
754 {
755 
756 	pool_cache_put(proc_cache, p);
757 }
758 
759 /*
760  * proc_enterpgrp: move p to a new or existing process group (and session).
761  *
762  * If we are creating a new pgrp, the pgid should equal
763  * the calling process' pid.
764  * If is only valid to enter a process group that is in the session
765  * of the process.
766  * Also mksess should only be set if we are creating a process group
767  *
768  * Only called from sys_setsid and sys_setpgid.
769  */
770 int
771 proc_enterpgrp(struct proc *curp, pid_t pid, pid_t pgid, bool mksess)
772 {
773 	struct pgrp *new_pgrp, *pgrp;
774 	struct session *sess;
775 	struct proc *p;
776 	int rval;
777 	pid_t pg_id = NO_PGID;
778 
779 	sess = mksess ? kmem_alloc(sizeof(*sess), KM_SLEEP) : NULL;
780 
781 	/* Allocate data areas we might need before doing any validity checks */
782 	mutex_enter(proc_lock);		/* Because pid_table might change */
783 	if (pid_table[pgid & pid_tbl_mask].pt_pgrp == 0) {
784 		mutex_exit(proc_lock);
785 		new_pgrp = kmem_alloc(sizeof(*new_pgrp), KM_SLEEP);
786 		mutex_enter(proc_lock);
787 	} else
788 		new_pgrp = NULL;
789 	rval = EPERM;	/* most common error (to save typing) */
790 
791 	/* Check pgrp exists or can be created */
792 	pgrp = pid_table[pgid & pid_tbl_mask].pt_pgrp;
793 	if (pgrp != NULL && pgrp->pg_id != pgid)
794 		goto done;
795 
796 	/* Can only set another process under restricted circumstances. */
797 	if (pid != curp->p_pid) {
798 		/* must exist and be one of our children... */
799 		if ((p = p_find(pid, PFIND_LOCKED)) == NULL ||
800 		    !p_inferior(p, curp)) {
801 			rval = ESRCH;
802 			goto done;
803 		}
804 		/* ... in the same session... */
805 		if (sess != NULL || p->p_session != curp->p_session)
806 			goto done;
807 		/* ... existing pgid must be in same session ... */
808 		if (pgrp != NULL && pgrp->pg_session != p->p_session)
809 			goto done;
810 		/* ... and not done an exec. */
811 		if (p->p_flag & PK_EXEC) {
812 			rval = EACCES;
813 			goto done;
814 		}
815 	} else {
816 		/* ... setsid() cannot re-enter a pgrp */
817 		if (mksess && (curp->p_pgid == curp->p_pid ||
818 		    pg_find(curp->p_pid, PFIND_LOCKED)))
819 			goto done;
820 		p = curp;
821 	}
822 
823 	/* Changing the process group/session of a session
824 	   leader is definitely off limits. */
825 	if (SESS_LEADER(p)) {
826 		if (sess == NULL && p->p_pgrp == pgrp)
827 			/* unless it's a definite noop */
828 			rval = 0;
829 		goto done;
830 	}
831 
832 	/* Can only create a process group with id of process */
833 	if (pgrp == NULL && pgid != pid)
834 		goto done;
835 
836 	/* Can only create a session if creating pgrp */
837 	if (sess != NULL && pgrp != NULL)
838 		goto done;
839 
840 	/* Check we allocated memory for a pgrp... */
841 	if (pgrp == NULL && new_pgrp == NULL)
842 		goto done;
843 
844 	/* Don't attach to 'zombie' pgrp */
845 	if (pgrp != NULL && LIST_EMPTY(&pgrp->pg_members))
846 		goto done;
847 
848 	/* Expect to succeed now */
849 	rval = 0;
850 
851 	if (pgrp == p->p_pgrp)
852 		/* nothing to do */
853 		goto done;
854 
855 	/* Ok all setup, link up required structures */
856 
857 	if (pgrp == NULL) {
858 		pgrp = new_pgrp;
859 		new_pgrp = NULL;
860 		if (sess != NULL) {
861 			sess->s_sid = p->p_pid;
862 			sess->s_leader = p;
863 			sess->s_count = 1;
864 			sess->s_ttyvp = NULL;
865 			sess->s_ttyp = NULL;
866 			sess->s_flags = p->p_session->s_flags & ~S_LOGIN_SET;
867 			memcpy(sess->s_login, p->p_session->s_login,
868 			    sizeof(sess->s_login));
869 			p->p_lflag &= ~PL_CONTROLT;
870 		} else {
871 			sess = p->p_pgrp->pg_session;
872 			proc_sesshold(sess);
873 		}
874 		pgrp->pg_session = sess;
875 		sess = NULL;
876 
877 		pgrp->pg_id = pgid;
878 		LIST_INIT(&pgrp->pg_members);
879 #ifdef DIAGNOSTIC
880 		if (__predict_false(pid_table[pgid & pid_tbl_mask].pt_pgrp))
881 			panic("enterpgrp: pgrp table slot in use");
882 		if (__predict_false(mksess && p != curp))
883 			panic("enterpgrp: mksession and p != curproc");
884 #endif
885 		pid_table[pgid & pid_tbl_mask].pt_pgrp = pgrp;
886 		pgrp->pg_jobc = 0;
887 	}
888 
889 	/*
890 	 * Adjust eligibility of affected pgrps to participate in job control.
891 	 * Increment eligibility counts before decrementing, otherwise we
892 	 * could reach 0 spuriously during the first call.
893 	 */
894 	fixjobc(p, pgrp, 1);
895 	fixjobc(p, p->p_pgrp, 0);
896 
897 	/* Interlock with ttread(). */
898 	mutex_spin_enter(&tty_lock);
899 
900 	/* Move process to requested group. */
901 	LIST_REMOVE(p, p_pglist);
902 	if (LIST_EMPTY(&p->p_pgrp->pg_members))
903 		/* defer delete until we've dumped the lock */
904 		pg_id = p->p_pgrp->pg_id;
905 	p->p_pgrp = pgrp;
906 	LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
907 
908 	/* Done with the swap; we can release the tty mutex. */
909 	mutex_spin_exit(&tty_lock);
910 
911     done:
912 	if (pg_id != NO_PGID) {
913 		/* Releases proc_lock. */
914 		pg_delete(pg_id);
915 	} else {
916 		mutex_exit(proc_lock);
917 	}
918 	if (sess != NULL)
919 		kmem_free(sess, sizeof(*sess));
920 	if (new_pgrp != NULL)
921 		kmem_free(new_pgrp, sizeof(*new_pgrp));
922 #ifdef DEBUG_PGRP
923 	if (__predict_false(rval))
924 		printf("enterpgrp(%d,%d,%d), curproc %d, rval %d\n",
925 			pid, pgid, mksess, curp->p_pid, rval);
926 #endif
927 	return rval;
928 }
929 
930 /*
931  * proc_leavepgrp: remove a process from its process group.
932  *  => must be called with the proc_lock held, which will be released;
933  */
934 void
935 proc_leavepgrp(struct proc *p)
936 {
937 	struct pgrp *pgrp;
938 
939 	KASSERT(mutex_owned(proc_lock));
940 
941 	/* Interlock with ttread() */
942 	mutex_spin_enter(&tty_lock);
943 	pgrp = p->p_pgrp;
944 	LIST_REMOVE(p, p_pglist);
945 	p->p_pgrp = NULL;
946 	mutex_spin_exit(&tty_lock);
947 
948 	if (LIST_EMPTY(&pgrp->pg_members)) {
949 		/* Releases proc_lock. */
950 		pg_delete(pgrp->pg_id);
951 	} else {
952 		mutex_exit(proc_lock);
953 	}
954 }
955 
956 /*
957  * pg_remove: remove a process group from the table.
958  *  => must be called with the proc_lock held;
959  *  => returns process group to free;
960  */
961 static struct pgrp *
962 pg_remove(pid_t pg_id)
963 {
964 	struct pgrp *pgrp;
965 	struct pid_table *pt;
966 
967 	KASSERT(mutex_owned(proc_lock));
968 
969 	pt = &pid_table[pg_id & pid_tbl_mask];
970 	pgrp = pt->pt_pgrp;
971 
972 	KASSERT(pgrp != NULL);
973 	KASSERT(pgrp->pg_id == pg_id);
974 	KASSERT(LIST_EMPTY(&pgrp->pg_members));
975 
976 	pt->pt_pgrp = NULL;
977 
978 	if (!P_VALID(pt->pt_proc)) {
979 		/* Orphaned pgrp, put slot onto free list. */
980 		KASSERT((P_NEXT(pt->pt_proc) & pid_tbl_mask) == 0);
981 		pg_id &= pid_tbl_mask;
982 		pt = &pid_table[last_free_pt];
983 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pg_id);
984 		last_free_pt = pg_id;
985 		pid_alloc_cnt--;
986 	}
987 	return pgrp;
988 }
989 
990 /*
991  * pg_delete: delete and free a process group.
992  *  => must be called with the proc_lock held, which will be released.
993  */
994 static void
995 pg_delete(pid_t pg_id)
996 {
997 	struct pgrp *pg;
998 	struct tty *ttyp;
999 	struct session *ss;
1000 
1001 	KASSERT(mutex_owned(proc_lock));
1002 
1003 	pg = pid_table[pg_id & pid_tbl_mask].pt_pgrp;
1004 	if (pg == NULL || pg->pg_id != pg_id || !LIST_EMPTY(&pg->pg_members)) {
1005 		mutex_exit(proc_lock);
1006 		return;
1007 	}
1008 
1009 	ss = pg->pg_session;
1010 
1011 	/* Remove reference (if any) from tty to this process group */
1012 	mutex_spin_enter(&tty_lock);
1013 	ttyp = ss->s_ttyp;
1014 	if (ttyp != NULL && ttyp->t_pgrp == pg) {
1015 		ttyp->t_pgrp = NULL;
1016 		KASSERT(ttyp->t_session == ss);
1017 	}
1018 	mutex_spin_exit(&tty_lock);
1019 
1020 	/*
1021 	 * The leading process group in a session is freed by proc_sessrele(),
1022 	 * if last reference.  Note: proc_sessrele() releases proc_lock.
1023 	 */
1024 	pg = (ss->s_sid != pg->pg_id) ? pg_remove(pg_id) : NULL;
1025 	proc_sessrele(ss);
1026 
1027 	if (pg != NULL) {
1028 		/* Free it, if was not done by proc_sessrele(). */
1029 		kmem_free(pg, sizeof(struct pgrp));
1030 	}
1031 }
1032 
1033 /*
1034  * Adjust pgrp jobc counters when specified process changes process group.
1035  * We count the number of processes in each process group that "qualify"
1036  * the group for terminal job control (those with a parent in a different
1037  * process group of the same session).  If that count reaches zero, the
1038  * process group becomes orphaned.  Check both the specified process'
1039  * process group and that of its children.
1040  * entering == 0 => p is leaving specified group.
1041  * entering == 1 => p is entering specified group.
1042  *
1043  * Call with proc_lock held.
1044  */
1045 void
1046 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
1047 {
1048 	struct pgrp *hispgrp;
1049 	struct session *mysession = pgrp->pg_session;
1050 	struct proc *child;
1051 
1052 	KASSERT(mutex_owned(proc_lock));
1053 
1054 	/*
1055 	 * Check p's parent to see whether p qualifies its own process
1056 	 * group; if so, adjust count for p's process group.
1057 	 */
1058 	hispgrp = p->p_pptr->p_pgrp;
1059 	if (hispgrp != pgrp && hispgrp->pg_session == mysession) {
1060 		if (entering) {
1061 			pgrp->pg_jobc++;
1062 			p->p_lflag &= ~PL_ORPHANPG;
1063 		} else if (--pgrp->pg_jobc == 0)
1064 			orphanpg(pgrp);
1065 	}
1066 
1067 	/*
1068 	 * Check this process' children to see whether they qualify
1069 	 * their process groups; if so, adjust counts for children's
1070 	 * process groups.
1071 	 */
1072 	LIST_FOREACH(child, &p->p_children, p_sibling) {
1073 		hispgrp = child->p_pgrp;
1074 		if (hispgrp != pgrp && hispgrp->pg_session == mysession &&
1075 		    !P_ZOMBIE(child)) {
1076 			if (entering) {
1077 				child->p_lflag &= ~PL_ORPHANPG;
1078 				hispgrp->pg_jobc++;
1079 			} else if (--hispgrp->pg_jobc == 0)
1080 				orphanpg(hispgrp);
1081 		}
1082 	}
1083 }
1084 
1085 /*
1086  * A process group has become orphaned;
1087  * if there are any stopped processes in the group,
1088  * hang-up all process in that group.
1089  *
1090  * Call with proc_lock held.
1091  */
1092 static void
1093 orphanpg(struct pgrp *pg)
1094 {
1095 	struct proc *p;
1096 
1097 	KASSERT(mutex_owned(proc_lock));
1098 
1099 	LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1100 		if (p->p_stat == SSTOP) {
1101 			p->p_lflag |= PL_ORPHANPG;
1102 			psignal(p, SIGHUP);
1103 			psignal(p, SIGCONT);
1104 		}
1105 	}
1106 }
1107 
1108 #ifdef DDB
1109 #include <ddb/db_output.h>
1110 void pidtbl_dump(void);
1111 void
1112 pidtbl_dump(void)
1113 {
1114 	struct pid_table *pt;
1115 	struct proc *p;
1116 	struct pgrp *pgrp;
1117 	int id;
1118 
1119 	db_printf("pid table %p size %x, next %x, last %x\n",
1120 		pid_table, pid_tbl_mask+1,
1121 		next_free_pt, last_free_pt);
1122 	for (pt = pid_table, id = 0; id <= pid_tbl_mask; id++, pt++) {
1123 		p = pt->pt_proc;
1124 		if (!P_VALID(p) && !pt->pt_pgrp)
1125 			continue;
1126 		db_printf("  id %x: ", id);
1127 		if (P_VALID(p))
1128 			db_printf("proc %p id %d (0x%x) %s\n",
1129 				p, p->p_pid, p->p_pid, p->p_comm);
1130 		else
1131 			db_printf("next %x use %x\n",
1132 				P_NEXT(p) & pid_tbl_mask,
1133 				P_NEXT(p) & ~pid_tbl_mask);
1134 		if ((pgrp = pt->pt_pgrp)) {
1135 			db_printf("\tsession %p, sid %d, count %d, login %s\n",
1136 			    pgrp->pg_session, pgrp->pg_session->s_sid,
1137 			    pgrp->pg_session->s_count,
1138 			    pgrp->pg_session->s_login);
1139 			db_printf("\tpgrp %p, pg_id %d, pg_jobc %d, members %p\n",
1140 			    pgrp, pgrp->pg_id, pgrp->pg_jobc,
1141 			    LIST_FIRST(&pgrp->pg_members));
1142 			LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1143 				db_printf("\t\tpid %d addr %p pgrp %p %s\n",
1144 				    p->p_pid, p, p->p_pgrp, p->p_comm);
1145 			}
1146 		}
1147 	}
1148 }
1149 #endif /* DDB */
1150 
1151 #ifdef KSTACK_CHECK_MAGIC
1152 #include <sys/user.h>
1153 
1154 #define	KSTACK_MAGIC	0xdeadbeaf
1155 
1156 /* XXX should be per process basis? */
1157 static int	kstackleftmin = KSTACK_SIZE;
1158 static int	kstackleftthres = KSTACK_SIZE / 8;
1159 
1160 void
1161 kstack_setup_magic(const struct lwp *l)
1162 {
1163 	uint32_t *ip;
1164 	uint32_t const *end;
1165 
1166 	KASSERT(l != NULL);
1167 	KASSERT(l != &lwp0);
1168 
1169 	/*
1170 	 * fill all the stack with magic number
1171 	 * so that later modification on it can be detected.
1172 	 */
1173 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1174 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1175 	for (; ip < end; ip++) {
1176 		*ip = KSTACK_MAGIC;
1177 	}
1178 }
1179 
1180 void
1181 kstack_check_magic(const struct lwp *l)
1182 {
1183 	uint32_t const *ip, *end;
1184 	int stackleft;
1185 
1186 	KASSERT(l != NULL);
1187 
1188 	/* don't check proc0 */ /*XXX*/
1189 	if (l == &lwp0)
1190 		return;
1191 
1192 #ifdef __MACHINE_STACK_GROWS_UP
1193 	/* stack grows upwards (eg. hppa) */
1194 	ip = (uint32_t *)((void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1195 	end = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1196 	for (ip--; ip >= end; ip--)
1197 		if (*ip != KSTACK_MAGIC)
1198 			break;
1199 
1200 	stackleft = (void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE - (void *)ip;
1201 #else /* __MACHINE_STACK_GROWS_UP */
1202 	/* stack grows downwards (eg. i386) */
1203 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1204 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1205 	for (; ip < end; ip++)
1206 		if (*ip != KSTACK_MAGIC)
1207 			break;
1208 
1209 	stackleft = ((const char *)ip) - (const char *)KSTACK_LOWEST_ADDR(l);
1210 #endif /* __MACHINE_STACK_GROWS_UP */
1211 
1212 	if (kstackleftmin > stackleft) {
1213 		kstackleftmin = stackleft;
1214 		if (stackleft < kstackleftthres)
1215 			printf("warning: kernel stack left %d bytes"
1216 			    "(pid %u:lid %u)\n", stackleft,
1217 			    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1218 	}
1219 
1220 	if (stackleft <= 0) {
1221 		panic("magic on the top of kernel stack changed for "
1222 		    "pid %u, lid %u: maybe kernel stack overflow",
1223 		    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1224 	}
1225 }
1226 #endif /* KSTACK_CHECK_MAGIC */
1227 
1228 int
1229 proclist_foreach_call(struct proclist *list,
1230     int (*callback)(struct proc *, void *arg), void *arg)
1231 {
1232 	struct proc marker;
1233 	struct proc *p;
1234 	int ret = 0;
1235 
1236 	marker.p_flag = PK_MARKER;
1237 	mutex_enter(proc_lock);
1238 	for (p = LIST_FIRST(list); ret == 0 && p != NULL;) {
1239 		if (p->p_flag & PK_MARKER) {
1240 			p = LIST_NEXT(p, p_list);
1241 			continue;
1242 		}
1243 		LIST_INSERT_AFTER(p, &marker, p_list);
1244 		ret = (*callback)(p, arg);
1245 		KASSERT(mutex_owned(proc_lock));
1246 		p = LIST_NEXT(&marker, p_list);
1247 		LIST_REMOVE(&marker, p_list);
1248 	}
1249 	mutex_exit(proc_lock);
1250 
1251 	return ret;
1252 }
1253 
1254 int
1255 proc_vmspace_getref(struct proc *p, struct vmspace **vm)
1256 {
1257 
1258 	/* XXXCDC: how should locking work here? */
1259 
1260 	/* curproc exception is for coredump. */
1261 
1262 	if ((p != curproc && (p->p_sflag & PS_WEXIT) != 0) ||
1263 	    (p->p_vmspace->vm_refcnt < 1)) { /* XXX */
1264 		return EFAULT;
1265 	}
1266 
1267 	uvmspace_addref(p->p_vmspace);
1268 	*vm = p->p_vmspace;
1269 
1270 	return 0;
1271 }
1272 
1273 /*
1274  * Acquire a write lock on the process credential.
1275  */
1276 void
1277 proc_crmod_enter(void)
1278 {
1279 	struct lwp *l = curlwp;
1280 	struct proc *p = l->l_proc;
1281 	struct plimit *lim;
1282 	kauth_cred_t oc;
1283 	char *cn;
1284 
1285 	/* Reset what needs to be reset in plimit. */
1286 	if (p->p_limit->pl_corename != defcorename) {
1287 		lim_privatise(p, false);
1288 		lim = p->p_limit;
1289 		mutex_enter(&lim->pl_lock);
1290 		cn = lim->pl_corename;
1291 		lim->pl_corename = defcorename;
1292 		mutex_exit(&lim->pl_lock);
1293 		if (cn != defcorename)
1294 			free(cn, M_TEMP);
1295 	}
1296 
1297 	mutex_enter(p->p_lock);
1298 
1299 	/* Ensure the LWP cached credentials are up to date. */
1300 	if ((oc = l->l_cred) != p->p_cred) {
1301 		kauth_cred_hold(p->p_cred);
1302 		l->l_cred = p->p_cred;
1303 		kauth_cred_free(oc);
1304 	}
1305 
1306 }
1307 
1308 /*
1309  * Set in a new process credential, and drop the write lock.  The credential
1310  * must have a reference already.  Optionally, free a no-longer required
1311  * credential.  The scheduler also needs to inspect p_cred, so we also
1312  * briefly acquire the sched state mutex.
1313  */
1314 void
1315 proc_crmod_leave(kauth_cred_t scred, kauth_cred_t fcred, bool sugid)
1316 {
1317 	struct lwp *l = curlwp, *l2;
1318 	struct proc *p = l->l_proc;
1319 	kauth_cred_t oc;
1320 
1321 	KASSERT(mutex_owned(p->p_lock));
1322 
1323 	/* Is there a new credential to set in? */
1324 	if (scred != NULL) {
1325 		p->p_cred = scred;
1326 		LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
1327 			if (l2 != l)
1328 				l2->l_prflag |= LPR_CRMOD;
1329 		}
1330 
1331 		/* Ensure the LWP cached credentials are up to date. */
1332 		if ((oc = l->l_cred) != scred) {
1333 			kauth_cred_hold(scred);
1334 			l->l_cred = scred;
1335 		}
1336 	} else
1337 		oc = NULL;	/* XXXgcc */
1338 
1339 	if (sugid) {
1340 		/*
1341 		 * Mark process as having changed credentials, stops
1342 		 * tracing etc.
1343 		 */
1344 		p->p_flag |= PK_SUGID;
1345 	}
1346 
1347 	mutex_exit(p->p_lock);
1348 
1349 	/* If there is a credential to be released, free it now. */
1350 	if (fcred != NULL) {
1351 		KASSERT(scred != NULL);
1352 		kauth_cred_free(fcred);
1353 		if (oc != scred)
1354 			kauth_cred_free(oc);
1355 	}
1356 }
1357 
1358 /*
1359  * proc_specific_key_create --
1360  *	Create a key for subsystem proc-specific data.
1361  */
1362 int
1363 proc_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1364 {
1365 
1366 	return (specificdata_key_create(proc_specificdata_domain, keyp, dtor));
1367 }
1368 
1369 /*
1370  * proc_specific_key_delete --
1371  *	Delete a key for subsystem proc-specific data.
1372  */
1373 void
1374 proc_specific_key_delete(specificdata_key_t key)
1375 {
1376 
1377 	specificdata_key_delete(proc_specificdata_domain, key);
1378 }
1379 
1380 /*
1381  * proc_initspecific --
1382  *	Initialize a proc's specificdata container.
1383  */
1384 void
1385 proc_initspecific(struct proc *p)
1386 {
1387 	int error;
1388 
1389 	error = specificdata_init(proc_specificdata_domain, &p->p_specdataref);
1390 	KASSERT(error == 0);
1391 }
1392 
1393 /*
1394  * proc_finispecific --
1395  *	Finalize a proc's specificdata container.
1396  */
1397 void
1398 proc_finispecific(struct proc *p)
1399 {
1400 
1401 	specificdata_fini(proc_specificdata_domain, &p->p_specdataref);
1402 }
1403 
1404 /*
1405  * proc_getspecific --
1406  *	Return proc-specific data corresponding to the specified key.
1407  */
1408 void *
1409 proc_getspecific(struct proc *p, specificdata_key_t key)
1410 {
1411 
1412 	return (specificdata_getspecific(proc_specificdata_domain,
1413 					 &p->p_specdataref, key));
1414 }
1415 
1416 /*
1417  * proc_setspecific --
1418  *	Set proc-specific data corresponding to the specified key.
1419  */
1420 void
1421 proc_setspecific(struct proc *p, specificdata_key_t key, void *data)
1422 {
1423 
1424 	specificdata_setspecific(proc_specificdata_domain,
1425 				 &p->p_specdataref, key, data);
1426 }
1427 
1428 int
1429 proc_uidmatch(kauth_cred_t cred, kauth_cred_t target)
1430 {
1431 	int r = 0;
1432 
1433 	if (kauth_cred_getuid(cred) != kauth_cred_getuid(target) ||
1434 	    kauth_cred_getuid(cred) != kauth_cred_getsvuid(target)) {
1435 		/*
1436 		 * suid proc of ours or proc not ours
1437 		 */
1438 		r = EPERM;
1439 	} else if (kauth_cred_getgid(target) != kauth_cred_getsvgid(target)) {
1440 		/*
1441 		 * sgid proc has sgid back to us temporarily
1442 		 */
1443 		r = EPERM;
1444 	} else {
1445 		/*
1446 		 * our rgid must be in target's group list (ie,
1447 		 * sub-processes started by a sgid process)
1448 		 */
1449 		int ismember = 0;
1450 
1451 		if (kauth_cred_ismember_gid(cred,
1452 		    kauth_cred_getgid(target), &ismember) != 0 ||
1453 		    !ismember)
1454 			r = EPERM;
1455 	}
1456 
1457 	return (r);
1458 }
1459 
1460