xref: /netbsd-src/sys/kern/kern_proc.c (revision 0df165c04d0a9ca1adde9ed2b890344c937954a6)
1 /*	$NetBSD: kern_proc.c,v 1.124 2007/11/11 23:22:24 matt Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2006, 2007 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  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1989, 1991, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. Neither the name of the University nor the names of its contributors
53  *    may be used to endorse or promote products derived from this software
54  *    without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
57  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
58  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
59  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
60  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
61  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
62  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
63  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
64  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
65  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66  * SUCH DAMAGE.
67  *
68  *	@(#)kern_proc.c	8.7 (Berkeley) 2/14/95
69  */
70 
71 #include <sys/cdefs.h>
72 __KERNEL_RCSID(0, "$NetBSD: kern_proc.c,v 1.124 2007/11/11 23:22:24 matt Exp $");
73 
74 #include "opt_kstack.h"
75 #include "opt_maxuprc.h"
76 #include "opt_multiprocessor.h"
77 #include "opt_lockdebug.h"
78 
79 #include <sys/param.h>
80 #include <sys/systm.h>
81 #include <sys/kernel.h>
82 #include <sys/proc.h>
83 #include <sys/resourcevar.h>
84 #include <sys/buf.h>
85 #include <sys/acct.h>
86 #include <sys/wait.h>
87 #include <sys/file.h>
88 #include <ufs/ufs/quota.h>
89 #include <sys/uio.h>
90 #include <sys/malloc.h>
91 #include <sys/pool.h>
92 #include <sys/mbuf.h>
93 #include <sys/ioctl.h>
94 #include <sys/tty.h>
95 #include <sys/signalvar.h>
96 #include <sys/ras.h>
97 #include <sys/filedesc.h>
98 #include "sys/syscall_stats.h"
99 #include <sys/kauth.h>
100 #include <sys/sleepq.h>
101 
102 #include <uvm/uvm.h>
103 #include <uvm/uvm_extern.h>
104 
105 /*
106  * Other process lists
107  */
108 
109 struct proclist allproc;
110 struct proclist zombproc;	/* resources have been freed */
111 
112 /*
113  * There are two locks on global process state.
114  *
115  * 1. proclist_lock is an adaptive mutex and is used when modifying
116  * or examining process state from a process context.  It protects
117  * the internal tables, all of the process lists, and a number of
118  * members of struct proc.
119  *
120  * 2. proclist_mutex is used when allproc must be traversed from an
121  * interrupt context, or when changing the state of processes.  The
122  * proclist_lock should always be used in preference.  In some cases,
123  * both locks need to be held.
124  *
125  *	proclist_lock	proclist_mutex	structure
126  *	--------------- --------------- -----------------
127  *	x				zombproc
128  *	x		x		pid_table
129  *	x				proc::p_pptr
130  *	x				proc::p_sibling
131  *	x				proc::p_children
132  *	x		x		allproc
133  *	x		x		proc::p_pgrp
134  *	x		x		proc::p_pglist
135  *	x		x		proc::p_session
136  *	x		x		proc::p_list
137  *			x		alllwp
138  *			x		lwp::l_list
139  *
140  * The lock order for processes and LWPs is approximately as following:
141  *
142  * kernel_lock
143  * -> proclist_lock
144  *   -> proc::p_mutex
145  *      -> proclist_mutex
146  *         -> proc::p_smutex
147  *           -> proc::p_stmutex
148  *
149  * XXX p_smutex can be run at IPL_VM once audio drivers on the x86
150  * platform are made MP safe.  Currently it blocks interrupts at
151  * IPL_SCHED and below.
152  *
153  * XXX The two process locks (p_smutex + p_mutex), and the two global
154  * state locks (proclist_lock + proclist_mutex) should be merged
155  * together.  However, to do so requires interrupts that interrupts
156  * be run with LWP context.
157  */
158 kmutex_t	proclist_lock;
159 kmutex_t	proclist_mutex;
160 
161 /*
162  * pid to proc lookup is done by indexing the pid_table array.
163  * Since pid numbers are only allocated when an empty slot
164  * has been found, there is no need to search any lists ever.
165  * (an orphaned pgrp will lock the slot, a session will lock
166  * the pgrp with the same number.)
167  * If the table is too small it is reallocated with twice the
168  * previous size and the entries 'unzipped' into the two halves.
169  * A linked list of free entries is passed through the pt_proc
170  * field of 'free' items - set odd to be an invalid ptr.
171  */
172 
173 struct pid_table {
174 	struct proc	*pt_proc;
175 	struct pgrp	*pt_pgrp;
176 };
177 #if 1	/* strongly typed cast - should be a noop */
178 static inline uint p2u(struct proc *p) { return (uint)(uintptr_t)p; }
179 #else
180 #define p2u(p) ((uint)p)
181 #endif
182 #define P_VALID(p) (!(p2u(p) & 1))
183 #define P_NEXT(p) (p2u(p) >> 1)
184 #define P_FREE(pid) ((struct proc *)(uintptr_t)((pid) << 1 | 1))
185 
186 #define INITIAL_PID_TABLE_SIZE	(1 << 5)
187 static struct pid_table *pid_table;
188 static uint pid_tbl_mask = INITIAL_PID_TABLE_SIZE - 1;
189 static uint pid_alloc_lim;	/* max we allocate before growing table */
190 static uint pid_alloc_cnt;	/* number of allocated pids */
191 
192 /* links through free slots - never empty! */
193 static uint next_free_pt, last_free_pt;
194 static pid_t pid_max = PID_MAX;		/* largest value we allocate */
195 
196 /* Components of the first process -- never freed. */
197 
198 extern const struct emul emul_netbsd;	/* defined in kern_exec.c */
199 
200 struct session session0 = {
201 	.s_count = 1,
202 	.s_sid = 0,
203 };
204 struct pgrp pgrp0 = {
205 	.pg_members = LIST_HEAD_INITIALIZER(&pgrp0.pg_members),
206 	.pg_session = &session0,
207 };
208 struct filedesc0 filedesc0;
209 struct cwdinfo cwdi0 = {
210 	.cwdi_cmask = CMASK,		/* see cmask below */
211 	.cwdi_refcnt = 1,
212 };
213 struct plimit limit0 = {
214 	.pl_corename = defcorename,
215 	.pl_refcnt = 1,
216 	.pl_rlimit = {
217 		[0 ... __arraycount(limit0.pl_rlimit) - 1] = {
218 			.rlim_cur = RLIM_INFINITY,
219 			.rlim_max = RLIM_INFINITY,
220 		},
221 	},
222 };
223 struct pstats pstat0;
224 struct vmspace vmspace0;
225 struct sigacts sigacts0;
226 struct turnstile turnstile0;
227 struct proc proc0 = {
228 	.p_lwps = LIST_HEAD_INITIALIZER(&proc0.p_lwps),
229 	.p_sigwaiters = LIST_HEAD_INITIALIZER(&proc0.p_sigwaiters),
230 	.p_nlwps = 1,
231 	.p_nrlwps = 1,
232 	.p_nlwpid = 1,		/* must match lwp0.l_lid */
233 	.p_pgrp = &pgrp0,
234 	.p_comm = "system",
235 	/*
236 	 * Set P_NOCLDWAIT so that kernel threads are reparented to init(8)
237 	 * when they exit.  init(8) can easily wait them out for us.
238 	 */
239 	.p_flag = PK_SYSTEM | PK_NOCLDWAIT,
240 	.p_stat = SACTIVE,
241 	.p_nice = NZERO,
242 	.p_emul = &emul_netbsd,
243 	.p_cwdi = &cwdi0,
244 	.p_limit = &limit0,
245 	.p_fd = &filedesc0.fd_fd,
246 	.p_vmspace = &vmspace0,
247 	.p_stats = &pstat0,
248 	.p_sigacts = &sigacts0,
249 };
250 struct lwp lwp0 __aligned(MIN_LWP_ALIGNMENT) = {
251 #ifdef LWP0_CPU_INFO
252 	.l_cpu = LWP0_CPU_INFO,
253 #endif
254 	.l_proc = &proc0,
255 	.l_lid = 1,
256 	.l_flag = LW_INMEM | LW_SYSTEM,
257 	.l_stat = LSONPROC,
258 	.l_ts = &turnstile0,
259 	.l_syncobj = &sched_syncobj,
260 	.l_refcnt = 1,
261 	.l_priority = PRI_USER + NPRI_USER - 1,
262 	.l_inheritedprio = -1,
263 	.l_class = SCHED_OTHER,
264 	.l_pi_lenders = SLIST_HEAD_INITIALIZER(&lwp0.l_pi_lenders),
265 	.l_name = __UNCONST("swapper"),
266 };
267 kauth_cred_t cred0;
268 
269 extern struct user *proc0paddr;
270 
271 int nofile = NOFILE;
272 int maxuprc = MAXUPRC;
273 int cmask = CMASK;
274 
275 POOL_INIT(proc_pool, sizeof(struct proc), 0, 0, 0, "procpl",
276     &pool_allocator_nointr, IPL_NONE);
277 POOL_INIT(pgrp_pool, sizeof(struct pgrp), 0, 0, 0, "pgrppl",
278     &pool_allocator_nointr, IPL_NONE);
279 POOL_INIT(plimit_pool, sizeof(struct plimit), 0, 0, 0, "plimitpl",
280     &pool_allocator_nointr, IPL_NONE);
281 POOL_INIT(pstats_pool, sizeof(struct pstats), 0, 0, 0, "pstatspl",
282     &pool_allocator_nointr, IPL_NONE);
283 POOL_INIT(session_pool, sizeof(struct session), 0, 0, 0, "sessionpl",
284     &pool_allocator_nointr, IPL_NONE);
285 
286 MALLOC_DEFINE(M_EMULDATA, "emuldata", "Per-process emulation data");
287 MALLOC_DEFINE(M_PROC, "proc", "Proc structures");
288 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures");
289 
290 /*
291  * The process list descriptors, used during pid allocation and
292  * by sysctl.  No locking on this data structure is needed since
293  * it is completely static.
294  */
295 const struct proclist_desc proclists[] = {
296 	{ &allproc	},
297 	{ &zombproc	},
298 	{ NULL		},
299 };
300 
301 static void orphanpg(struct pgrp *);
302 static void pg_delete(pid_t);
303 
304 static specificdata_domain_t proc_specificdata_domain;
305 
306 /*
307  * Initialize global process hashing structures.
308  */
309 void
310 procinit(void)
311 {
312 	const struct proclist_desc *pd;
313 	int i;
314 #define	LINK_EMPTY ((PID_MAX + INITIAL_PID_TABLE_SIZE) & ~(INITIAL_PID_TABLE_SIZE - 1))
315 
316 	for (pd = proclists; pd->pd_list != NULL; pd++)
317 		LIST_INIT(pd->pd_list);
318 
319 	mutex_init(&proclist_lock, MUTEX_DEFAULT, IPL_NONE);
320 	mutex_init(&proclist_mutex, MUTEX_SPIN, IPL_SCHED);
321 
322 	pid_table = malloc(INITIAL_PID_TABLE_SIZE * sizeof *pid_table,
323 			    M_PROC, M_WAITOK);
324 	/* Set free list running through table...
325 	   Preset 'use count' above PID_MAX so we allocate pid 1 next. */
326 	for (i = 0; i <= pid_tbl_mask; i++) {
327 		pid_table[i].pt_proc = P_FREE(LINK_EMPTY + i + 1);
328 		pid_table[i].pt_pgrp = 0;
329 	}
330 	/* slot 0 is just grabbed */
331 	next_free_pt = 1;
332 	/* Need to fix last entry. */
333 	last_free_pt = pid_tbl_mask;
334 	pid_table[last_free_pt].pt_proc = P_FREE(LINK_EMPTY);
335 	/* point at which we grow table - to avoid reusing pids too often */
336 	pid_alloc_lim = pid_tbl_mask - 1;
337 #undef LINK_EMPTY
338 
339 	uihashtbl =
340 	    hashinit(maxproc / 16, HASH_LIST, M_PROC, M_WAITOK, &uihash);
341 
342 	proc_specificdata_domain = specificdata_domain_create();
343 	KASSERT(proc_specificdata_domain != NULL);
344 }
345 
346 /*
347  * Initialize process 0.
348  */
349 void
350 proc0_init(void)
351 {
352 	struct proc *p;
353 	struct pgrp *pg;
354 	struct session *sess;
355 	struct lwp *l;
356 	rlim_t lim;
357 
358 	p = &proc0;
359 	pg = &pgrp0;
360 	sess = &session0;
361 	l = &lwp0;
362 
363 	KASSERT(l->l_lid == p->p_nlwpid);
364 
365 	mutex_init(&p->p_smutex, MUTEX_SPIN, IPL_SCHED);
366 	mutex_init(&p->p_stmutex, MUTEX_SPIN, IPL_HIGH);
367 	mutex_init(&p->p_raslock, MUTEX_DEFAULT, IPL_NONE);
368 	mutex_init(&p->p_mutex, MUTEX_DEFAULT, IPL_NONE);
369 	mutex_init(&l->l_swaplock, MUTEX_DEFAULT, IPL_NONE);
370 
371 	rw_init(&p->p_reflock);
372 	cv_init(&p->p_waitcv, "wait");
373 	cv_init(&p->p_lwpcv, "lwpwait");
374 
375 	LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling);
376 
377 	pid_table[0].pt_proc = p;
378 	LIST_INSERT_HEAD(&allproc, p, p_list);
379 	LIST_INSERT_HEAD(&alllwp, l, l_list);
380 
381 	pid_table[0].pt_pgrp = pg;
382 	LIST_INSERT_HEAD(&pg->pg_members, p, p_pglist);
383 
384 #ifdef __HAVE_SYSCALL_INTERN
385 	(*p->p_emul->e_syscall_intern)(p);
386 #endif
387 
388 	callout_init(&l->l_timeout_ch, CALLOUT_MPSAFE);
389 	callout_setfunc(&l->l_timeout_ch, sleepq_timeout, l);
390 	cv_init(&l->l_sigcv, "sigwait");
391 
392 	/* Create credentials. */
393 	cred0 = kauth_cred_alloc();
394 	p->p_cred = cred0;
395 	kauth_cred_hold(cred0);
396 	l->l_cred = cred0;
397 
398 	/* Create the CWD info. */
399 	rw_init(&cwdi0.cwdi_lock);
400 
401 	/* Create the limits structures. */
402 	mutex_init(&limit0.pl_lock, MUTEX_DEFAULT, IPL_NONE);
403 
404 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
405 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_cur =
406 	    maxfiles < nofile ? maxfiles : nofile;
407 
408 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
409 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_cur =
410 	    maxproc < maxuprc ? maxproc : maxuprc;
411 
412 	lim = ptoa(uvmexp.free);
413 	limit0.pl_rlimit[RLIMIT_RSS].rlim_max = lim;
414 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim;
415 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3;
416 
417 	/* Configure virtual memory system, set vm rlimits. */
418 	uvm_init_limits(p);
419 
420 	/* Initialize file descriptor table for proc0. */
421 	fdinit1(&filedesc0);
422 
423 	/*
424 	 * Initialize proc0's vmspace, which uses the kernel pmap.
425 	 * All kernel processes (which never have user space mappings)
426 	 * share proc0's vmspace, and thus, the kernel pmap.
427 	 */
428 	uvmspace_init(&vmspace0, pmap_kernel(), round_page(VM_MIN_ADDRESS),
429 	    trunc_page(VM_MAX_ADDRESS));
430 
431 	l->l_addr = proc0paddr;				/* XXX */
432 
433 	/* Initialize signal state for proc0. */
434 	mutex_init(&p->p_sigacts->sa_mutex, MUTEX_SPIN, IPL_NONE);
435 	siginit(p);
436 
437 	proc_initspecific(p);
438 	lwp_initspecific(l);
439 
440 	SYSCALL_TIME_LWP_INIT(l);
441 }
442 
443 /*
444  * Check that the specified process group is in the session of the
445  * specified process.
446  * Treats -ve ids as process ids.
447  * Used to validate TIOCSPGRP requests.
448  */
449 int
450 pgid_in_session(struct proc *p, pid_t pg_id)
451 {
452 	struct pgrp *pgrp;
453 	struct session *session;
454 	int error;
455 
456 	mutex_enter(&proclist_lock);
457 	if (pg_id < 0) {
458 		struct proc *p1 = p_find(-pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
459 		if (p1 == NULL)
460 			return EINVAL;
461 		pgrp = p1->p_pgrp;
462 	} else {
463 		pgrp = pg_find(pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
464 		if (pgrp == NULL)
465 			return EINVAL;
466 	}
467 	session = pgrp->pg_session;
468 	if (session != p->p_pgrp->pg_session)
469 		error = EPERM;
470 	else
471 		error = 0;
472 	mutex_exit(&proclist_lock);
473 
474 	return error;
475 }
476 
477 /*
478  * Is p an inferior of q?
479  *
480  * Call with the proclist_lock held.
481  */
482 int
483 inferior(struct proc *p, struct proc *q)
484 {
485 
486 	for (; p != q; p = p->p_pptr)
487 		if (p->p_pid == 0)
488 			return 0;
489 	return 1;
490 }
491 
492 /*
493  * Locate a process by number
494  */
495 struct proc *
496 p_find(pid_t pid, uint flags)
497 {
498 	struct proc *p;
499 	char stat;
500 
501 	if (!(flags & PFIND_LOCKED))
502 		mutex_enter(&proclist_lock);
503 
504 	p = pid_table[pid & pid_tbl_mask].pt_proc;
505 
506 	/* Only allow live processes to be found by pid. */
507 	/* XXXSMP p_stat */
508 	if (P_VALID(p) && p->p_pid == pid && ((stat = p->p_stat) == SACTIVE ||
509 	    stat == SSTOP || ((flags & PFIND_ZOMBIE) &&
510 	    (stat == SZOMB || stat == SDEAD || stat == SDYING)))) {
511 		if (flags & PFIND_UNLOCK_OK)
512 			 mutex_exit(&proclist_lock);
513 		return p;
514 	}
515 	if (flags & PFIND_UNLOCK_FAIL)
516 		mutex_exit(&proclist_lock);
517 	return NULL;
518 }
519 
520 
521 /*
522  * Locate a process group by number
523  */
524 struct pgrp *
525 pg_find(pid_t pgid, uint flags)
526 {
527 	struct pgrp *pg;
528 
529 	if (!(flags & PFIND_LOCKED))
530 		mutex_enter(&proclist_lock);
531 	pg = pid_table[pgid & pid_tbl_mask].pt_pgrp;
532 	/*
533 	 * Can't look up a pgrp that only exists because the session
534 	 * hasn't died yet (traditional)
535 	 */
536 	if (pg == NULL || pg->pg_id != pgid || LIST_EMPTY(&pg->pg_members)) {
537 		if (flags & PFIND_UNLOCK_FAIL)
538 			 mutex_exit(&proclist_lock);
539 		return NULL;
540 	}
541 
542 	if (flags & PFIND_UNLOCK_OK)
543 		mutex_exit(&proclist_lock);
544 	return pg;
545 }
546 
547 static void
548 expand_pid_table(void)
549 {
550 	uint pt_size = pid_tbl_mask + 1;
551 	struct pid_table *n_pt, *new_pt;
552 	struct proc *proc;
553 	struct pgrp *pgrp;
554 	int i;
555 	pid_t pid;
556 
557 	new_pt = malloc(pt_size * 2 * sizeof *new_pt, M_PROC, M_WAITOK);
558 
559 	mutex_enter(&proclist_lock);
560 	if (pt_size != pid_tbl_mask + 1) {
561 		/* Another process beat us to it... */
562 		mutex_exit(&proclist_lock);
563 		FREE(new_pt, M_PROC);
564 		return;
565 	}
566 
567 	/*
568 	 * Copy entries from old table into new one.
569 	 * If 'pid' is 'odd' we need to place in the upper half,
570 	 * even pid's to the lower half.
571 	 * Free items stay in the low half so we don't have to
572 	 * fixup the reference to them.
573 	 * We stuff free items on the front of the freelist
574 	 * because we can't write to unmodified entries.
575 	 * Processing the table backwards maintains a semblance
576 	 * of issueing pid numbers that increase with time.
577 	 */
578 	i = pt_size - 1;
579 	n_pt = new_pt + i;
580 	for (; ; i--, n_pt--) {
581 		proc = pid_table[i].pt_proc;
582 		pgrp = pid_table[i].pt_pgrp;
583 		if (!P_VALID(proc)) {
584 			/* Up 'use count' so that link is valid */
585 			pid = (P_NEXT(proc) + pt_size) & ~pt_size;
586 			proc = P_FREE(pid);
587 			if (pgrp)
588 				pid = pgrp->pg_id;
589 		} else
590 			pid = proc->p_pid;
591 
592 		/* Save entry in appropriate half of table */
593 		n_pt[pid & pt_size].pt_proc = proc;
594 		n_pt[pid & pt_size].pt_pgrp = pgrp;
595 
596 		/* Put other piece on start of free list */
597 		pid = (pid ^ pt_size) & ~pid_tbl_mask;
598 		n_pt[pid & pt_size].pt_proc =
599 				    P_FREE((pid & ~pt_size) | next_free_pt);
600 		n_pt[pid & pt_size].pt_pgrp = 0;
601 		next_free_pt = i | (pid & pt_size);
602 		if (i == 0)
603 			break;
604 	}
605 
606 	/* Switch tables */
607 	mutex_enter(&proclist_mutex);
608 	n_pt = pid_table;
609 	pid_table = new_pt;
610 	mutex_exit(&proclist_mutex);
611 	pid_tbl_mask = pt_size * 2 - 1;
612 
613 	/*
614 	 * pid_max starts as PID_MAX (= 30000), once we have 16384
615 	 * allocated pids we need it to be larger!
616 	 */
617 	if (pid_tbl_mask > PID_MAX) {
618 		pid_max = pid_tbl_mask * 2 + 1;
619 		pid_alloc_lim |= pid_alloc_lim << 1;
620 	} else
621 		pid_alloc_lim <<= 1;	/* doubles number of free slots... */
622 
623 	mutex_exit(&proclist_lock);
624 	FREE(n_pt, M_PROC);
625 }
626 
627 struct proc *
628 proc_alloc(void)
629 {
630 	struct proc *p;
631 	int nxt;
632 	pid_t pid;
633 	struct pid_table *pt;
634 
635 	p = pool_get(&proc_pool, PR_WAITOK);
636 	p->p_stat = SIDL;			/* protect against others */
637 
638 	proc_initspecific(p);
639 	/* allocate next free pid */
640 
641 	for (;;expand_pid_table()) {
642 		if (__predict_false(pid_alloc_cnt >= pid_alloc_lim))
643 			/* ensure pids cycle through 2000+ values */
644 			continue;
645 		mutex_enter(&proclist_lock);
646 		pt = &pid_table[next_free_pt];
647 #ifdef DIAGNOSTIC
648 		if (__predict_false(P_VALID(pt->pt_proc) || pt->pt_pgrp))
649 			panic("proc_alloc: slot busy");
650 #endif
651 		nxt = P_NEXT(pt->pt_proc);
652 		if (nxt & pid_tbl_mask)
653 			break;
654 		/* Table full - expand (NB last entry not used....) */
655 		mutex_exit(&proclist_lock);
656 	}
657 
658 	/* pid is 'saved use count' + 'size' + entry */
659 	pid = (nxt & ~pid_tbl_mask) + pid_tbl_mask + 1 + next_free_pt;
660 	if ((uint)pid > (uint)pid_max)
661 		pid &= pid_tbl_mask;
662 	p->p_pid = pid;
663 	next_free_pt = nxt & pid_tbl_mask;
664 
665 	/* Grab table slot */
666 	mutex_enter(&proclist_mutex);
667 	pt->pt_proc = p;
668 	mutex_exit(&proclist_mutex);
669 	pid_alloc_cnt++;
670 
671 	mutex_exit(&proclist_lock);
672 
673 	return p;
674 }
675 
676 /*
677  * Free a process id - called from proc_free (in kern_exit.c)
678  *
679  * Called with the proclist_lock held.
680  */
681 void
682 proc_free_pid(struct proc *p)
683 {
684 	pid_t pid = p->p_pid;
685 	struct pid_table *pt;
686 
687 	KASSERT(mutex_owned(&proclist_lock));
688 
689 	pt = &pid_table[pid & pid_tbl_mask];
690 #ifdef DIAGNOSTIC
691 	if (__predict_false(pt->pt_proc != p))
692 		panic("proc_free: pid_table mismatch, pid %x, proc %p",
693 			pid, p);
694 #endif
695 	mutex_enter(&proclist_mutex);
696 	/* save pid use count in slot */
697 	pt->pt_proc = P_FREE(pid & ~pid_tbl_mask);
698 
699 	if (pt->pt_pgrp == NULL) {
700 		/* link last freed entry onto ours */
701 		pid &= pid_tbl_mask;
702 		pt = &pid_table[last_free_pt];
703 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pid);
704 		last_free_pt = pid;
705 		pid_alloc_cnt--;
706 	}
707 	mutex_exit(&proclist_mutex);
708 
709 	nprocs--;
710 }
711 
712 /*
713  * Move p to a new or existing process group (and session)
714  *
715  * If we are creating a new pgrp, the pgid should equal
716  * the calling process' pid.
717  * If is only valid to enter a process group that is in the session
718  * of the process.
719  * Also mksess should only be set if we are creating a process group
720  *
721  * Only called from sys_setsid, sys_setpgid/sys_setpgrp and the
722  * SYSV setpgrp support for hpux.
723  */
724 int
725 enterpgrp(struct proc *curp, pid_t pid, pid_t pgid, int mksess)
726 {
727 	struct pgrp *new_pgrp, *pgrp;
728 	struct session *sess;
729 	struct proc *p;
730 	int rval;
731 	pid_t pg_id = NO_PGID;
732 
733 	if (mksess)
734 		sess = pool_get(&session_pool, PR_WAITOK);
735 	else
736 		sess = NULL;
737 
738 	/* Allocate data areas we might need before doing any validity checks */
739 	mutex_enter(&proclist_lock);		/* Because pid_table might change */
740 	if (pid_table[pgid & pid_tbl_mask].pt_pgrp == 0) {
741 		mutex_exit(&proclist_lock);
742 		new_pgrp = pool_get(&pgrp_pool, PR_WAITOK);
743 		mutex_enter(&proclist_lock);
744 	} else
745 		new_pgrp = NULL;
746 	rval = EPERM;	/* most common error (to save typing) */
747 
748 	/* Check pgrp exists or can be created */
749 	pgrp = pid_table[pgid & pid_tbl_mask].pt_pgrp;
750 	if (pgrp != NULL && pgrp->pg_id != pgid)
751 		goto done;
752 
753 	/* Can only set another process under restricted circumstances. */
754 	if (pid != curp->p_pid) {
755 		/* must exist and be one of our children... */
756 		if ((p = p_find(pid, PFIND_LOCKED)) == NULL ||
757 		    !inferior(p, curp)) {
758 			rval = ESRCH;
759 			goto done;
760 		}
761 		/* ... in the same session... */
762 		if (sess != NULL || p->p_session != curp->p_session)
763 			goto done;
764 		/* ... existing pgid must be in same session ... */
765 		if (pgrp != NULL && pgrp->pg_session != p->p_session)
766 			goto done;
767 		/* ... and not done an exec. */
768 		if (p->p_flag & PK_EXEC) {
769 			rval = EACCES;
770 			goto done;
771 		}
772 	} else {
773 		/* ... setsid() cannot re-enter a pgrp */
774 		if (mksess && (curp->p_pgid == curp->p_pid ||
775 		    pg_find(curp->p_pid, PFIND_LOCKED)))
776 			goto done;
777 		p = curp;
778 	}
779 
780 	/* Changing the process group/session of a session
781 	   leader is definitely off limits. */
782 	if (SESS_LEADER(p)) {
783 		if (sess == NULL && p->p_pgrp == pgrp)
784 			/* unless it's a definite noop */
785 			rval = 0;
786 		goto done;
787 	}
788 
789 	/* Can only create a process group with id of process */
790 	if (pgrp == NULL && pgid != pid)
791 		goto done;
792 
793 	/* Can only create a session if creating pgrp */
794 	if (sess != NULL && pgrp != NULL)
795 		goto done;
796 
797 	/* Check we allocated memory for a pgrp... */
798 	if (pgrp == NULL && new_pgrp == NULL)
799 		goto done;
800 
801 	/* Don't attach to 'zombie' pgrp */
802 	if (pgrp != NULL && LIST_EMPTY(&pgrp->pg_members))
803 		goto done;
804 
805 	/* Expect to succeed now */
806 	rval = 0;
807 
808 	if (pgrp == p->p_pgrp)
809 		/* nothing to do */
810 		goto done;
811 
812 	/* Ok all setup, link up required structures */
813 
814 	if (pgrp == NULL) {
815 		pgrp = new_pgrp;
816 		new_pgrp = 0;
817 		if (sess != NULL) {
818 			sess->s_sid = p->p_pid;
819 			sess->s_leader = p;
820 			sess->s_count = 1;
821 			sess->s_ttyvp = NULL;
822 			sess->s_ttyp = NULL;
823 			sess->s_flags = p->p_session->s_flags & ~S_LOGIN_SET;
824 			memcpy(sess->s_login, p->p_session->s_login,
825 			    sizeof(sess->s_login));
826 			p->p_lflag &= ~PL_CONTROLT;
827 		} else {
828 			sess = p->p_pgrp->pg_session;
829 			SESSHOLD(sess);
830 		}
831 		pgrp->pg_session = sess;
832 		sess = 0;
833 
834 		pgrp->pg_id = pgid;
835 		LIST_INIT(&pgrp->pg_members);
836 #ifdef DIAGNOSTIC
837 		if (__predict_false(pid_table[pgid & pid_tbl_mask].pt_pgrp))
838 			panic("enterpgrp: pgrp table slot in use");
839 		if (__predict_false(mksess && p != curp))
840 			panic("enterpgrp: mksession and p != curproc");
841 #endif
842 		mutex_enter(&proclist_mutex);
843 		pid_table[pgid & pid_tbl_mask].pt_pgrp = pgrp;
844 		pgrp->pg_jobc = 0;
845 	} else
846 		mutex_enter(&proclist_mutex);
847 
848 #ifdef notyet
849 	/*
850 	 * If there's a controlling terminal for the current session, we
851 	 * have to interlock with it.  See ttread().
852 	 */
853 	if (p->p_session->s_ttyvp != NULL) {
854 		tp = p->p_session->s_ttyp;
855 		mutex_enter(&tp->t_mutex);
856 	} else
857 		tp = NULL;
858 #endif
859 
860 	/*
861 	 * Adjust eligibility of affected pgrps to participate in job control.
862 	 * Increment eligibility counts before decrementing, otherwise we
863 	 * could reach 0 spuriously during the first call.
864 	 */
865 	fixjobc(p, pgrp, 1);
866 	fixjobc(p, p->p_pgrp, 0);
867 
868 	/* Move process to requested group. */
869 	LIST_REMOVE(p, p_pglist);
870 	if (LIST_EMPTY(&p->p_pgrp->pg_members))
871 		/* defer delete until we've dumped the lock */
872 		pg_id = p->p_pgrp->pg_id;
873 	p->p_pgrp = pgrp;
874 	LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
875 	mutex_exit(&proclist_mutex);
876 
877 #ifdef notyet
878 	/* Done with the swap; we can release the tty mutex. */
879 	if (tp != NULL)
880 		mutex_exit(&tp->t_mutex);
881 #endif
882 
883     done:
884 	if (pg_id != NO_PGID)
885 		pg_delete(pg_id);
886 	mutex_exit(&proclist_lock);
887 	if (sess != NULL)
888 		pool_put(&session_pool, sess);
889 	if (new_pgrp != NULL)
890 		pool_put(&pgrp_pool, new_pgrp);
891 #ifdef DEBUG_PGRP
892 	if (__predict_false(rval))
893 		printf("enterpgrp(%d,%d,%d), curproc %d, rval %d\n",
894 			pid, pgid, mksess, curp->p_pid, rval);
895 #endif
896 	return rval;
897 }
898 
899 /*
900  * Remove a process from its process group.  Must be called with the
901  * proclist_lock held.
902  */
903 void
904 leavepgrp(struct proc *p)
905 {
906 	struct pgrp *pgrp;
907 
908 	KASSERT(mutex_owned(&proclist_lock));
909 
910 	/*
911 	 * If there's a controlling terminal for the session, we have to
912 	 * interlock with it.  See ttread().
913 	 */
914 	mutex_enter(&proclist_mutex);
915 #ifdef notyet
916 	if (p_>p_session->s_ttyvp != NULL) {
917 		tp = p->p_session->s_ttyp;
918 		mutex_enter(&tp->t_mutex);
919 	} else
920 		tp = NULL;
921 #endif
922 
923 	pgrp = p->p_pgrp;
924 	LIST_REMOVE(p, p_pglist);
925 	p->p_pgrp = NULL;
926 
927 #ifdef notyet
928 	if (tp != NULL)
929 		mutex_exit(&tp->t_mutex);
930 #endif
931 	mutex_exit(&proclist_mutex);
932 
933 	if (LIST_EMPTY(&pgrp->pg_members))
934 		pg_delete(pgrp->pg_id);
935 }
936 
937 /*
938  * Free a process group.  Must be called with the proclist_lock held.
939  */
940 static void
941 pg_free(pid_t pg_id)
942 {
943 	struct pgrp *pgrp;
944 	struct pid_table *pt;
945 
946 	KASSERT(mutex_owned(&proclist_lock));
947 
948 	pt = &pid_table[pg_id & pid_tbl_mask];
949 	pgrp = pt->pt_pgrp;
950 #ifdef DIAGNOSTIC
951 	if (__predict_false(!pgrp || pgrp->pg_id != pg_id
952 	    || !LIST_EMPTY(&pgrp->pg_members)))
953 		panic("pg_free: process group absent or has members");
954 #endif
955 	pt->pt_pgrp = 0;
956 
957 	if (!P_VALID(pt->pt_proc)) {
958 		/* orphaned pgrp, put slot onto free list */
959 #ifdef DIAGNOSTIC
960 		if (__predict_false(P_NEXT(pt->pt_proc) & pid_tbl_mask))
961 			panic("pg_free: process slot on free list");
962 #endif
963 		mutex_enter(&proclist_mutex);
964 		pg_id &= pid_tbl_mask;
965 		pt = &pid_table[last_free_pt];
966 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pg_id);
967 		mutex_exit(&proclist_mutex);
968 		last_free_pt = pg_id;
969 		pid_alloc_cnt--;
970 	}
971 	pool_put(&pgrp_pool, pgrp);
972 }
973 
974 /*
975  * Delete a process group.  Must be called with the proclist_lock held.
976  */
977 static void
978 pg_delete(pid_t pg_id)
979 {
980 	struct pgrp *pgrp;
981 	struct tty *ttyp;
982 	struct session *ss;
983 	int is_pgrp_leader;
984 
985 	KASSERT(mutex_owned(&proclist_lock));
986 
987 	pgrp = pid_table[pg_id & pid_tbl_mask].pt_pgrp;
988 	if (pgrp == NULL || pgrp->pg_id != pg_id ||
989 	    !LIST_EMPTY(&pgrp->pg_members))
990 		return;
991 
992 	ss = pgrp->pg_session;
993 
994 	/* Remove reference (if any) from tty to this process group */
995 	ttyp = ss->s_ttyp;
996 	if (ttyp != NULL && ttyp->t_pgrp == pgrp) {
997 		ttyp->t_pgrp = NULL;
998 #ifdef DIAGNOSTIC
999 		if (ttyp->t_session != ss)
1000 			panic("pg_delete: wrong session on terminal");
1001 #endif
1002 	}
1003 
1004 	/*
1005 	 * The leading process group in a session is freed
1006 	 * by sessdelete() if last reference.
1007 	 */
1008 	is_pgrp_leader = (ss->s_sid == pgrp->pg_id);
1009 	SESSRELE(ss);
1010 
1011 	if (is_pgrp_leader)
1012 		return;
1013 
1014 	pg_free(pg_id);
1015 }
1016 
1017 /*
1018  * Delete session - called from SESSRELE when s_count becomes zero.
1019  * Must be called with the proclist_lock held.
1020  */
1021 void
1022 sessdelete(struct session *ss)
1023 {
1024 
1025 	KASSERT(mutex_owned(&proclist_lock));
1026 
1027 	/*
1028 	 * We keep the pgrp with the same id as the session in
1029 	 * order to stop a process being given the same pid.
1030 	 * Since the pgrp holds a reference to the session, it
1031 	 * must be a 'zombie' pgrp by now.
1032 	 */
1033 	pg_free(ss->s_sid);
1034 	pool_put(&session_pool, ss);
1035 }
1036 
1037 /*
1038  * Adjust pgrp jobc counters when specified process changes process group.
1039  * We count the number of processes in each process group that "qualify"
1040  * the group for terminal job control (those with a parent in a different
1041  * process group of the same session).  If that count reaches zero, the
1042  * process group becomes orphaned.  Check both the specified process'
1043  * process group and that of its children.
1044  * entering == 0 => p is leaving specified group.
1045  * entering == 1 => p is entering specified group.
1046  *
1047  * Call with proclist_lock held.
1048  */
1049 void
1050 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
1051 {
1052 	struct pgrp *hispgrp;
1053 	struct session *mysession = pgrp->pg_session;
1054 	struct proc *child;
1055 
1056 	KASSERT(mutex_owned(&proclist_lock));
1057 	KASSERT(mutex_owned(&proclist_mutex));
1058 
1059 	/*
1060 	 * Check p's parent to see whether p qualifies its own process
1061 	 * group; if so, adjust count for p's process group.
1062 	 */
1063 	hispgrp = p->p_pptr->p_pgrp;
1064 	if (hispgrp != pgrp && hispgrp->pg_session == mysession) {
1065 		if (entering) {
1066 			mutex_enter(&p->p_smutex);
1067 			p->p_sflag &= ~PS_ORPHANPG;
1068 			mutex_exit(&p->p_smutex);
1069 			pgrp->pg_jobc++;
1070 		} else if (--pgrp->pg_jobc == 0)
1071 			orphanpg(pgrp);
1072 	}
1073 
1074 	/*
1075 	 * Check this process' children to see whether they qualify
1076 	 * their process groups; if so, adjust counts for children's
1077 	 * process groups.
1078 	 */
1079 	LIST_FOREACH(child, &p->p_children, p_sibling) {
1080 		hispgrp = child->p_pgrp;
1081 		if (hispgrp != pgrp && hispgrp->pg_session == mysession &&
1082 		    !P_ZOMBIE(child)) {
1083 			if (entering) {
1084 				mutex_enter(&child->p_smutex);
1085 				child->p_sflag &= ~PS_ORPHANPG;
1086 				mutex_exit(&child->p_smutex);
1087 				hispgrp->pg_jobc++;
1088 			} else if (--hispgrp->pg_jobc == 0)
1089 				orphanpg(hispgrp);
1090 		}
1091 	}
1092 }
1093 
1094 /*
1095  * A process group has become orphaned;
1096  * if there are any stopped processes in the group,
1097  * hang-up all process in that group.
1098  *
1099  * Call with proclist_lock held.
1100  */
1101 static void
1102 orphanpg(struct pgrp *pg)
1103 {
1104 	struct proc *p;
1105 	int doit;
1106 
1107 	KASSERT(mutex_owned(&proclist_lock));
1108 	KASSERT(mutex_owned(&proclist_mutex));
1109 
1110 	doit = 0;
1111 
1112 	LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1113 		mutex_enter(&p->p_smutex);
1114 		if (p->p_stat == SSTOP) {
1115 			doit = 1;
1116 			p->p_sflag |= PS_ORPHANPG;
1117 		}
1118 		mutex_exit(&p->p_smutex);
1119 	}
1120 
1121 	if (doit) {
1122 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1123 			psignal(p, SIGHUP);
1124 			psignal(p, SIGCONT);
1125 		}
1126 	}
1127 }
1128 
1129 #ifdef DDB
1130 #include <ddb/db_output.h>
1131 void pidtbl_dump(void);
1132 void
1133 pidtbl_dump(void)
1134 {
1135 	struct pid_table *pt;
1136 	struct proc *p;
1137 	struct pgrp *pgrp;
1138 	int id;
1139 
1140 	db_printf("pid table %p size %x, next %x, last %x\n",
1141 		pid_table, pid_tbl_mask+1,
1142 		next_free_pt, last_free_pt);
1143 	for (pt = pid_table, id = 0; id <= pid_tbl_mask; id++, pt++) {
1144 		p = pt->pt_proc;
1145 		if (!P_VALID(p) && !pt->pt_pgrp)
1146 			continue;
1147 		db_printf("  id %x: ", id);
1148 		if (P_VALID(p))
1149 			db_printf("proc %p id %d (0x%x) %s\n",
1150 				p, p->p_pid, p->p_pid, p->p_comm);
1151 		else
1152 			db_printf("next %x use %x\n",
1153 				P_NEXT(p) & pid_tbl_mask,
1154 				P_NEXT(p) & ~pid_tbl_mask);
1155 		if ((pgrp = pt->pt_pgrp)) {
1156 			db_printf("\tsession %p, sid %d, count %d, login %s\n",
1157 			    pgrp->pg_session, pgrp->pg_session->s_sid,
1158 			    pgrp->pg_session->s_count,
1159 			    pgrp->pg_session->s_login);
1160 			db_printf("\tpgrp %p, pg_id %d, pg_jobc %d, members %p\n",
1161 			    pgrp, pgrp->pg_id, pgrp->pg_jobc,
1162 			    pgrp->pg_members.lh_first);
1163 			for (p = pgrp->pg_members.lh_first; p != 0;
1164 			    p = p->p_pglist.le_next) {
1165 				db_printf("\t\tpid %d addr %p pgrp %p %s\n",
1166 				    p->p_pid, p, p->p_pgrp, p->p_comm);
1167 			}
1168 		}
1169 	}
1170 }
1171 #endif /* DDB */
1172 
1173 #ifdef KSTACK_CHECK_MAGIC
1174 #include <sys/user.h>
1175 
1176 #define	KSTACK_MAGIC	0xdeadbeaf
1177 
1178 /* XXX should be per process basis? */
1179 int kstackleftmin = KSTACK_SIZE;
1180 int kstackleftthres = KSTACK_SIZE / 8; /* warn if remaining stack is
1181 					  less than this */
1182 
1183 void
1184 kstack_setup_magic(const struct lwp *l)
1185 {
1186 	uint32_t *ip;
1187 	uint32_t const *end;
1188 
1189 	KASSERT(l != NULL);
1190 	KASSERT(l != &lwp0);
1191 
1192 	/*
1193 	 * fill all the stack with magic number
1194 	 * so that later modification on it can be detected.
1195 	 */
1196 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1197 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1198 	for (; ip < end; ip++) {
1199 		*ip = KSTACK_MAGIC;
1200 	}
1201 }
1202 
1203 void
1204 kstack_check_magic(const struct lwp *l)
1205 {
1206 	uint32_t const *ip, *end;
1207 	int stackleft;
1208 
1209 	KASSERT(l != NULL);
1210 
1211 	/* don't check proc0 */ /*XXX*/
1212 	if (l == &lwp0)
1213 		return;
1214 
1215 #ifdef __MACHINE_STACK_GROWS_UP
1216 	/* stack grows upwards (eg. hppa) */
1217 	ip = (uint32_t *)((void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1218 	end = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1219 	for (ip--; ip >= end; ip--)
1220 		if (*ip != KSTACK_MAGIC)
1221 			break;
1222 
1223 	stackleft = (void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE - (void *)ip;
1224 #else /* __MACHINE_STACK_GROWS_UP */
1225 	/* stack grows downwards (eg. i386) */
1226 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1227 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1228 	for (; ip < end; ip++)
1229 		if (*ip != KSTACK_MAGIC)
1230 			break;
1231 
1232 	stackleft = ((const char *)ip) - (const char *)KSTACK_LOWEST_ADDR(l);
1233 #endif /* __MACHINE_STACK_GROWS_UP */
1234 
1235 	if (kstackleftmin > stackleft) {
1236 		kstackleftmin = stackleft;
1237 		if (stackleft < kstackleftthres)
1238 			printf("warning: kernel stack left %d bytes"
1239 			    "(pid %u:lid %u)\n", stackleft,
1240 			    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1241 	}
1242 
1243 	if (stackleft <= 0) {
1244 		panic("magic on the top of kernel stack changed for "
1245 		    "pid %u, lid %u: maybe kernel stack overflow",
1246 		    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1247 	}
1248 }
1249 #endif /* KSTACK_CHECK_MAGIC */
1250 
1251 /*
1252  * XXXSMP this is bust, it grabs a read lock and then messes about
1253  * with allproc.
1254  */
1255 int
1256 proclist_foreach_call(struct proclist *list,
1257     int (*callback)(struct proc *, void *arg), void *arg)
1258 {
1259 	struct proc marker;
1260 	struct proc *p;
1261 	struct lwp * const l = curlwp;
1262 	int ret = 0;
1263 
1264 	marker.p_flag = PK_MARKER;
1265 	uvm_lwp_hold(l);
1266 	mutex_enter(&proclist_lock);
1267 	for (p = LIST_FIRST(list); ret == 0 && p != NULL;) {
1268 		if (p->p_flag & PK_MARKER) {
1269 			p = LIST_NEXT(p, p_list);
1270 			continue;
1271 		}
1272 		LIST_INSERT_AFTER(p, &marker, p_list);
1273 		ret = (*callback)(p, arg);
1274 		KASSERT(mutex_owned(&proclist_lock));
1275 		p = LIST_NEXT(&marker, p_list);
1276 		LIST_REMOVE(&marker, p_list);
1277 	}
1278 	mutex_exit(&proclist_lock);
1279 	uvm_lwp_rele(l);
1280 
1281 	return ret;
1282 }
1283 
1284 int
1285 proc_vmspace_getref(struct proc *p, struct vmspace **vm)
1286 {
1287 
1288 	/* XXXCDC: how should locking work here? */
1289 
1290 	/* curproc exception is for coredump. */
1291 
1292 	if ((p != curproc && (p->p_sflag & PS_WEXIT) != 0) ||
1293 	    (p->p_vmspace->vm_refcnt < 1)) { /* XXX */
1294 		return EFAULT;
1295 	}
1296 
1297 	uvmspace_addref(p->p_vmspace);
1298 	*vm = p->p_vmspace;
1299 
1300 	return 0;
1301 }
1302 
1303 /*
1304  * Acquire a write lock on the process credential.
1305  */
1306 void
1307 proc_crmod_enter(void)
1308 {
1309 	struct lwp *l = curlwp;
1310 	struct proc *p = l->l_proc;
1311 	struct plimit *lim;
1312 	kauth_cred_t oc;
1313 	char *cn;
1314 
1315 	/* Reset what needs to be reset in plimit. */
1316 	if (p->p_limit->pl_corename != defcorename) {
1317 		lim_privatise(p, false);
1318 		lim = p->p_limit;
1319 		mutex_enter(&lim->pl_lock);
1320 		cn = lim->pl_corename;
1321 		lim->pl_corename = defcorename;
1322 		mutex_exit(&lim->pl_lock);
1323 		if (cn != defcorename)
1324 			free(cn, M_TEMP);
1325 	}
1326 
1327 	mutex_enter(&p->p_mutex);
1328 
1329 	/* Ensure the LWP cached credentials are up to date. */
1330 	if ((oc = l->l_cred) != p->p_cred) {
1331 		kauth_cred_hold(p->p_cred);
1332 		l->l_cred = p->p_cred;
1333 		kauth_cred_free(oc);
1334 	}
1335 
1336 }
1337 
1338 /*
1339  * Set in a new process credential, and drop the write lock.  The credential
1340  * must have a reference already.  Optionally, free a no-longer required
1341  * credential.  The scheduler also needs to inspect p_cred, so we also
1342  * briefly acquire the sched state mutex.
1343  */
1344 void
1345 proc_crmod_leave(kauth_cred_t scred, kauth_cred_t fcred, bool sugid)
1346 {
1347 	struct lwp *l = curlwp;
1348 	struct proc *p = l->l_proc;
1349 	kauth_cred_t oc;
1350 
1351 	/* Is there a new credential to set in? */
1352 	if (scred != NULL) {
1353 		mutex_enter(&p->p_smutex);
1354 		p->p_cred = scred;
1355 		mutex_exit(&p->p_smutex);
1356 
1357 		/* Ensure the LWP cached credentials are up to date. */
1358 		if ((oc = l->l_cred) != scred) {
1359 			kauth_cred_hold(scred);
1360 			l->l_cred = scred;
1361 		}
1362 	} else
1363 		oc = NULL;	/* XXXgcc */
1364 
1365 	if (sugid) {
1366 		/*
1367 		 * Mark process as having changed credentials, stops
1368 		 * tracing etc.
1369 		 */
1370 		p->p_flag |= PK_SUGID;
1371 	}
1372 
1373 	mutex_exit(&p->p_mutex);
1374 
1375 	/* If there is a credential to be released, free it now. */
1376 	if (fcred != NULL) {
1377 		KASSERT(scred != NULL);
1378 		kauth_cred_free(fcred);
1379 		if (oc != scred)
1380 			kauth_cred_free(oc);
1381 	}
1382 }
1383 
1384 /*
1385  * proc_specific_key_create --
1386  *	Create a key for subsystem proc-specific data.
1387  */
1388 int
1389 proc_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1390 {
1391 
1392 	return (specificdata_key_create(proc_specificdata_domain, keyp, dtor));
1393 }
1394 
1395 /*
1396  * proc_specific_key_delete --
1397  *	Delete a key for subsystem proc-specific data.
1398  */
1399 void
1400 proc_specific_key_delete(specificdata_key_t key)
1401 {
1402 
1403 	specificdata_key_delete(proc_specificdata_domain, key);
1404 }
1405 
1406 /*
1407  * proc_initspecific --
1408  *	Initialize a proc's specificdata container.
1409  */
1410 void
1411 proc_initspecific(struct proc *p)
1412 {
1413 	int error;
1414 
1415 	error = specificdata_init(proc_specificdata_domain, &p->p_specdataref);
1416 	KASSERT(error == 0);
1417 }
1418 
1419 /*
1420  * proc_finispecific --
1421  *	Finalize a proc's specificdata container.
1422  */
1423 void
1424 proc_finispecific(struct proc *p)
1425 {
1426 
1427 	specificdata_fini(proc_specificdata_domain, &p->p_specdataref);
1428 }
1429 
1430 /*
1431  * proc_getspecific --
1432  *	Return proc-specific data corresponding to the specified key.
1433  */
1434 void *
1435 proc_getspecific(struct proc *p, specificdata_key_t key)
1436 {
1437 
1438 	return (specificdata_getspecific(proc_specificdata_domain,
1439 					 &p->p_specdataref, key));
1440 }
1441 
1442 /*
1443  * proc_setspecific --
1444  *	Set proc-specific data corresponding to the specified key.
1445  */
1446 void
1447 proc_setspecific(struct proc *p, specificdata_key_t key, void *data)
1448 {
1449 
1450 	specificdata_setspecific(proc_specificdata_domain,
1451 				 &p->p_specdataref, key, data);
1452 }
1453