xref: /netbsd-src/sys/kern/kern_proc.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: kern_proc.c,v 1.127 2007/12/05 07:06:53 ad 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 1.127 2007/12/05 07:06:53 ad 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 #include <sys/atomic.h>
102 
103 #include <uvm/uvm.h>
104 #include <uvm/uvm_extern.h>
105 
106 /*
107  * Other process lists
108  */
109 
110 struct proclist allproc;
111 struct proclist zombproc;	/* resources have been freed */
112 
113 /*
114  * There are two locks on global process state.
115  *
116  * 1. proclist_lock is an adaptive mutex and is used when modifying
117  * or examining process state from a process context.  It protects
118  * the internal tables, all of the process lists, and a number of
119  * members of struct proc.
120  *
121  * 2. proclist_mutex is used when allproc must be traversed from an
122  * interrupt context, or when changing the state of processes.  The
123  * proclist_lock should always be used in preference.  In some cases,
124  * both locks need to be held.
125  *
126  *	proclist_lock	proclist_mutex	structure
127  *	--------------- --------------- -----------------
128  *	x				zombproc
129  *	x		x		pid_table
130  *	x				proc::p_pptr
131  *	x				proc::p_sibling
132  *	x				proc::p_children
133  *	x				alllwp
134  *	x		x		allproc
135  *	x		x		proc::p_pgrp
136  *	x		x		proc::p_pglist
137  *	x		x		proc::p_session
138  *	x		x		proc::p_list
139  *			x		lwp::l_list
140  *
141  * The lock order for processes and LWPs is approximately as following:
142  *
143  * kernel_lock
144  * -> proclist_lock
145  *   -> proc::p_mutex
146  *      -> proclist_mutex
147  *         -> proc::p_smutex
148  *           -> proc::p_stmutex
149  *
150  * XXX p_smutex can be run at IPL_VM once audio drivers on the x86
151  * platform are made MP safe.  Currently it blocks interrupts at
152  * IPL_SCHED and below.
153  *
154  * XXX The two process locks (p_smutex + p_mutex), and the two global
155  * state locks (proclist_lock + proclist_mutex) should be merged
156  * together.  However, to do so requires interrupts that interrupts
157  * be run with LWP context.
158  */
159 kmutex_t	proclist_lock;
160 kmutex_t	proclist_mutex;
161 
162 /*
163  * pid to proc lookup is done by indexing the pid_table array.
164  * Since pid numbers are only allocated when an empty slot
165  * has been found, there is no need to search any lists ever.
166  * (an orphaned pgrp will lock the slot, a session will lock
167  * the pgrp with the same number.)
168  * If the table is too small it is reallocated with twice the
169  * previous size and the entries 'unzipped' into the two halves.
170  * A linked list of free entries is passed through the pt_proc
171  * field of 'free' items - set odd to be an invalid ptr.
172  */
173 
174 struct pid_table {
175 	struct proc	*pt_proc;
176 	struct pgrp	*pt_pgrp;
177 };
178 #if 1	/* strongly typed cast - should be a noop */
179 static inline uint p2u(struct proc *p) { return (uint)(uintptr_t)p; }
180 #else
181 #define p2u(p) ((uint)p)
182 #endif
183 #define P_VALID(p) (!(p2u(p) & 1))
184 #define P_NEXT(p) (p2u(p) >> 1)
185 #define P_FREE(pid) ((struct proc *)(uintptr_t)((pid) << 1 | 1))
186 
187 #define INITIAL_PID_TABLE_SIZE	(1 << 5)
188 static struct pid_table *pid_table;
189 static uint pid_tbl_mask = INITIAL_PID_TABLE_SIZE - 1;
190 static uint pid_alloc_lim;	/* max we allocate before growing table */
191 static uint pid_alloc_cnt;	/* number of allocated pids */
192 
193 /* links through free slots - never empty! */
194 static uint next_free_pt, last_free_pt;
195 static pid_t pid_max = PID_MAX;		/* largest value we allocate */
196 
197 /* Components of the first process -- never freed. */
198 
199 extern const struct emul emul_netbsd;	/* defined in kern_exec.c */
200 
201 struct session session0 = {
202 	.s_count = 1,
203 	.s_sid = 0,
204 };
205 struct pgrp pgrp0 = {
206 	.pg_members = LIST_HEAD_INITIALIZER(&pgrp0.pg_members),
207 	.pg_session = &session0,
208 };
209 struct filedesc0 filedesc0;
210 struct cwdinfo cwdi0 = {
211 	.cwdi_cmask = CMASK,		/* see cmask below */
212 	.cwdi_refcnt = 1,
213 };
214 struct plimit limit0 = {
215 	.pl_corename = defcorename,
216 	.pl_refcnt = 1,
217 	.pl_rlimit = {
218 		[0 ... __arraycount(limit0.pl_rlimit) - 1] = {
219 			.rlim_cur = RLIM_INFINITY,
220 			.rlim_max = RLIM_INFINITY,
221 		},
222 	},
223 };
224 struct pstats pstat0;
225 struct vmspace vmspace0;
226 struct sigacts sigacts0;
227 struct turnstile turnstile0;
228 struct proc proc0 = {
229 	.p_lwps = LIST_HEAD_INITIALIZER(&proc0.p_lwps),
230 	.p_sigwaiters = LIST_HEAD_INITIALIZER(&proc0.p_sigwaiters),
231 	.p_nlwps = 1,
232 	.p_nrlwps = 1,
233 	.p_nlwpid = 1,		/* must match lwp0.l_lid */
234 	.p_pgrp = &pgrp0,
235 	.p_comm = "system",
236 	/*
237 	 * Set P_NOCLDWAIT so that kernel threads are reparented to init(8)
238 	 * when they exit.  init(8) can easily wait them out for us.
239 	 */
240 	.p_flag = PK_SYSTEM | PK_NOCLDWAIT,
241 	.p_stat = SACTIVE,
242 	.p_nice = NZERO,
243 	.p_emul = &emul_netbsd,
244 	.p_cwdi = &cwdi0,
245 	.p_limit = &limit0,
246 	.p_fd = &filedesc0.fd_fd,
247 	.p_vmspace = &vmspace0,
248 	.p_stats = &pstat0,
249 	.p_sigacts = &sigacts0,
250 };
251 struct lwp lwp0 __aligned(MIN_LWP_ALIGNMENT) = {
252 #ifdef LWP0_CPU_INFO
253 	.l_cpu = LWP0_CPU_INFO,
254 #endif
255 	.l_proc = &proc0,
256 	.l_lid = 1,
257 	.l_flag = LW_INMEM | LW_SYSTEM,
258 	.l_stat = LSONPROC,
259 	.l_ts = &turnstile0,
260 	.l_syncobj = &sched_syncobj,
261 	.l_refcnt = 1,
262 	.l_priority = PRI_USER + NPRI_USER - 1,
263 	.l_inheritedprio = -1,
264 	.l_class = SCHED_OTHER,
265 	.l_pi_lenders = SLIST_HEAD_INITIALIZER(&lwp0.l_pi_lenders),
266 	.l_name = __UNCONST("swapper"),
267 };
268 kauth_cred_t cred0;
269 
270 extern struct user *proc0paddr;
271 
272 int nofile = NOFILE;
273 int maxuprc = MAXUPRC;
274 int cmask = CMASK;
275 
276 POOL_INIT(proc_pool, sizeof(struct proc), 0, 0, 0, "procpl",
277     &pool_allocator_nointr, IPL_NONE);
278 POOL_INIT(pgrp_pool, sizeof(struct pgrp), 0, 0, 0, "pgrppl",
279     &pool_allocator_nointr, IPL_NONE);
280 POOL_INIT(plimit_pool, sizeof(struct plimit), 0, 0, 0, "plimitpl",
281     &pool_allocator_nointr, IPL_NONE);
282 POOL_INIT(pstats_pool, sizeof(struct pstats), 0, 0, 0, "pstatspl",
283     &pool_allocator_nointr, IPL_NONE);
284 POOL_INIT(session_pool, sizeof(struct session), 0, 0, 0, "sessionpl",
285     &pool_allocator_nointr, IPL_NONE);
286 
287 MALLOC_DEFINE(M_EMULDATA, "emuldata", "Per-process emulation data");
288 MALLOC_DEFINE(M_PROC, "proc", "Proc structures");
289 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures");
290 
291 /*
292  * The process list descriptors, used during pid allocation and
293  * by sysctl.  No locking on this data structure is needed since
294  * it is completely static.
295  */
296 const struct proclist_desc proclists[] = {
297 	{ &allproc	},
298 	{ &zombproc	},
299 	{ NULL		},
300 };
301 
302 static void orphanpg(struct pgrp *);
303 static void pg_delete(pid_t);
304 
305 static specificdata_domain_t proc_specificdata_domain;
306 
307 /*
308  * Initialize global process hashing structures.
309  */
310 void
311 procinit(void)
312 {
313 	const struct proclist_desc *pd;
314 	int i;
315 #define	LINK_EMPTY ((PID_MAX + INITIAL_PID_TABLE_SIZE) & ~(INITIAL_PID_TABLE_SIZE - 1))
316 
317 	for (pd = proclists; pd->pd_list != NULL; pd++)
318 		LIST_INIT(pd->pd_list);
319 
320 	mutex_init(&proclist_lock, MUTEX_DEFAULT, IPL_NONE);
321 	mutex_init(&proclist_mutex, MUTEX_DEFAULT, IPL_SCHED);
322 
323 	pid_table = malloc(INITIAL_PID_TABLE_SIZE * sizeof *pid_table,
324 			    M_PROC, M_WAITOK);
325 	/* Set free list running through table...
326 	   Preset 'use count' above PID_MAX so we allocate pid 1 next. */
327 	for (i = 0; i <= pid_tbl_mask; i++) {
328 		pid_table[i].pt_proc = P_FREE(LINK_EMPTY + i + 1);
329 		pid_table[i].pt_pgrp = 0;
330 	}
331 	/* slot 0 is just grabbed */
332 	next_free_pt = 1;
333 	/* Need to fix last entry. */
334 	last_free_pt = pid_tbl_mask;
335 	pid_table[last_free_pt].pt_proc = P_FREE(LINK_EMPTY);
336 	/* point at which we grow table - to avoid reusing pids too often */
337 	pid_alloc_lim = pid_tbl_mask - 1;
338 #undef LINK_EMPTY
339 
340 	uihashtbl =
341 	    hashinit(maxproc / 16, HASH_LIST, M_PROC, M_WAITOK, &uihash);
342 
343 	proc_specificdata_domain = specificdata_domain_create();
344 	KASSERT(proc_specificdata_domain != NULL);
345 }
346 
347 /*
348  * Initialize process 0.
349  */
350 void
351 proc0_init(void)
352 {
353 	struct proc *p;
354 	struct pgrp *pg;
355 	struct session *sess;
356 	struct lwp *l;
357 	rlim_t lim;
358 
359 	p = &proc0;
360 	pg = &pgrp0;
361 	sess = &session0;
362 	l = &lwp0;
363 
364 	KASSERT(l->l_lid == p->p_nlwpid);
365 
366 	mutex_init(&p->p_smutex, MUTEX_DEFAULT, IPL_SCHED);
367 	mutex_init(&p->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
368 	mutex_init(&p->p_raslock, MUTEX_DEFAULT, IPL_NONE);
369 	mutex_init(&p->p_mutex, MUTEX_DEFAULT, IPL_NONE);
370 	mutex_init(&l->l_swaplock, MUTEX_DEFAULT, IPL_NONE);
371 
372 	rw_init(&p->p_reflock);
373 	cv_init(&p->p_waitcv, "wait");
374 	cv_init(&p->p_lwpcv, "lwpwait");
375 
376 	LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling);
377 
378 	pid_table[0].pt_proc = p;
379 	LIST_INSERT_HEAD(&allproc, p, p_list);
380 	LIST_INSERT_HEAD(&alllwp, l, l_list);
381 
382 	pid_table[0].pt_pgrp = pg;
383 	LIST_INSERT_HEAD(&pg->pg_members, p, p_pglist);
384 
385 #ifdef __HAVE_SYSCALL_INTERN
386 	(*p->p_emul->e_syscall_intern)(p);
387 #endif
388 
389 	callout_init(&l->l_timeout_ch, CALLOUT_MPSAFE);
390 	callout_setfunc(&l->l_timeout_ch, sleepq_timeout, l);
391 	cv_init(&l->l_sigcv, "sigwait");
392 
393 	/* Create credentials. */
394 	cred0 = kauth_cred_alloc();
395 	p->p_cred = cred0;
396 	kauth_cred_hold(cred0);
397 	l->l_cred = cred0;
398 
399 	/* Create the CWD info. */
400 	rw_init(&cwdi0.cwdi_lock);
401 
402 	/* Create the limits structures. */
403 	mutex_init(&limit0.pl_lock, MUTEX_DEFAULT, IPL_NONE);
404 
405 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
406 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_cur =
407 	    maxfiles < nofile ? maxfiles : nofile;
408 
409 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
410 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_cur =
411 	    maxproc < maxuprc ? maxproc : maxuprc;
412 
413 	lim = ptoa(uvmexp.free);
414 	limit0.pl_rlimit[RLIMIT_RSS].rlim_max = lim;
415 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim;
416 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3;
417 
418 	/* Configure virtual memory system, set vm rlimits. */
419 	uvm_init_limits(p);
420 
421 	/* Initialize file descriptor table for proc0. */
422 	fdinit1(&filedesc0);
423 
424 	/*
425 	 * Initialize proc0's vmspace, which uses the kernel pmap.
426 	 * All kernel processes (which never have user space mappings)
427 	 * share proc0's vmspace, and thus, the kernel pmap.
428 	 */
429 	uvmspace_init(&vmspace0, pmap_kernel(), round_page(VM_MIN_ADDRESS),
430 	    trunc_page(VM_MAX_ADDRESS));
431 
432 	l->l_addr = proc0paddr;				/* XXX */
433 
434 	/* Initialize signal state for proc0. XXX IPL_SCHED */
435 	mutex_init(&p->p_sigacts->sa_mutex, MUTEX_DEFAULT, IPL_SCHED);
436 	siginit(p);
437 
438 	proc_initspecific(p);
439 	lwp_initspecific(l);
440 
441 	SYSCALL_TIME_LWP_INIT(l);
442 }
443 
444 /*
445  * Check that the specified process group is in the session of the
446  * specified process.
447  * Treats -ve ids as process ids.
448  * Used to validate TIOCSPGRP requests.
449  */
450 int
451 pgid_in_session(struct proc *p, pid_t pg_id)
452 {
453 	struct pgrp *pgrp;
454 	struct session *session;
455 	int error;
456 
457 	mutex_enter(&proclist_lock);
458 	if (pg_id < 0) {
459 		struct proc *p1 = p_find(-pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
460 		if (p1 == NULL)
461 			return EINVAL;
462 		pgrp = p1->p_pgrp;
463 	} else {
464 		pgrp = pg_find(pg_id, PFIND_LOCKED | PFIND_UNLOCK_FAIL);
465 		if (pgrp == NULL)
466 			return EINVAL;
467 	}
468 	session = pgrp->pg_session;
469 	if (session != p->p_pgrp->pg_session)
470 		error = EPERM;
471 	else
472 		error = 0;
473 	mutex_exit(&proclist_lock);
474 
475 	return error;
476 }
477 
478 /*
479  * Is p an inferior of q?
480  *
481  * Call with the proclist_lock held.
482  */
483 int
484 inferior(struct proc *p, struct proc *q)
485 {
486 
487 	for (; p != q; p = p->p_pptr)
488 		if (p->p_pid == 0)
489 			return 0;
490 	return 1;
491 }
492 
493 /*
494  * Locate a process by number
495  */
496 struct proc *
497 p_find(pid_t pid, uint flags)
498 {
499 	struct proc *p;
500 	char stat;
501 
502 	if (!(flags & PFIND_LOCKED))
503 		mutex_enter(&proclist_lock);
504 
505 	p = pid_table[pid & pid_tbl_mask].pt_proc;
506 
507 	/* Only allow live processes to be found by pid. */
508 	/* XXXSMP p_stat */
509 	if (P_VALID(p) && p->p_pid == pid && ((stat = p->p_stat) == SACTIVE ||
510 	    stat == SSTOP || ((flags & PFIND_ZOMBIE) &&
511 	    (stat == SZOMB || stat == SDEAD || stat == SDYING)))) {
512 		if (flags & PFIND_UNLOCK_OK)
513 			 mutex_exit(&proclist_lock);
514 		return p;
515 	}
516 	if (flags & PFIND_UNLOCK_FAIL)
517 		mutex_exit(&proclist_lock);
518 	return NULL;
519 }
520 
521 
522 /*
523  * Locate a process group by number
524  */
525 struct pgrp *
526 pg_find(pid_t pgid, uint flags)
527 {
528 	struct pgrp *pg;
529 
530 	if (!(flags & PFIND_LOCKED))
531 		mutex_enter(&proclist_lock);
532 	pg = pid_table[pgid & pid_tbl_mask].pt_pgrp;
533 	/*
534 	 * Can't look up a pgrp that only exists because the session
535 	 * hasn't died yet (traditional)
536 	 */
537 	if (pg == NULL || pg->pg_id != pgid || LIST_EMPTY(&pg->pg_members)) {
538 		if (flags & PFIND_UNLOCK_FAIL)
539 			 mutex_exit(&proclist_lock);
540 		return NULL;
541 	}
542 
543 	if (flags & PFIND_UNLOCK_OK)
544 		mutex_exit(&proclist_lock);
545 	return pg;
546 }
547 
548 static void
549 expand_pid_table(void)
550 {
551 	uint pt_size = pid_tbl_mask + 1;
552 	struct pid_table *n_pt, *new_pt;
553 	struct proc *proc;
554 	struct pgrp *pgrp;
555 	int i;
556 	pid_t pid;
557 
558 	new_pt = malloc(pt_size * 2 * sizeof *new_pt, M_PROC, M_WAITOK);
559 
560 	mutex_enter(&proclist_lock);
561 	if (pt_size != pid_tbl_mask + 1) {
562 		/* Another process beat us to it... */
563 		mutex_exit(&proclist_lock);
564 		FREE(new_pt, M_PROC);
565 		return;
566 	}
567 
568 	/*
569 	 * Copy entries from old table into new one.
570 	 * If 'pid' is 'odd' we need to place in the upper half,
571 	 * even pid's to the lower half.
572 	 * Free items stay in the low half so we don't have to
573 	 * fixup the reference to them.
574 	 * We stuff free items on the front of the freelist
575 	 * because we can't write to unmodified entries.
576 	 * Processing the table backwards maintains a semblance
577 	 * of issueing pid numbers that increase with time.
578 	 */
579 	i = pt_size - 1;
580 	n_pt = new_pt + i;
581 	for (; ; i--, n_pt--) {
582 		proc = pid_table[i].pt_proc;
583 		pgrp = pid_table[i].pt_pgrp;
584 		if (!P_VALID(proc)) {
585 			/* Up 'use count' so that link is valid */
586 			pid = (P_NEXT(proc) + pt_size) & ~pt_size;
587 			proc = P_FREE(pid);
588 			if (pgrp)
589 				pid = pgrp->pg_id;
590 		} else
591 			pid = proc->p_pid;
592 
593 		/* Save entry in appropriate half of table */
594 		n_pt[pid & pt_size].pt_proc = proc;
595 		n_pt[pid & pt_size].pt_pgrp = pgrp;
596 
597 		/* Put other piece on start of free list */
598 		pid = (pid ^ pt_size) & ~pid_tbl_mask;
599 		n_pt[pid & pt_size].pt_proc =
600 				    P_FREE((pid & ~pt_size) | next_free_pt);
601 		n_pt[pid & pt_size].pt_pgrp = 0;
602 		next_free_pt = i | (pid & pt_size);
603 		if (i == 0)
604 			break;
605 	}
606 
607 	/* Switch tables */
608 	mutex_enter(&proclist_mutex);
609 	n_pt = pid_table;
610 	pid_table = new_pt;
611 	mutex_exit(&proclist_mutex);
612 	pid_tbl_mask = pt_size * 2 - 1;
613 
614 	/*
615 	 * pid_max starts as PID_MAX (= 30000), once we have 16384
616 	 * allocated pids we need it to be larger!
617 	 */
618 	if (pid_tbl_mask > PID_MAX) {
619 		pid_max = pid_tbl_mask * 2 + 1;
620 		pid_alloc_lim |= pid_alloc_lim << 1;
621 	} else
622 		pid_alloc_lim <<= 1;	/* doubles number of free slots... */
623 
624 	mutex_exit(&proclist_lock);
625 	FREE(n_pt, M_PROC);
626 }
627 
628 struct proc *
629 proc_alloc(void)
630 {
631 	struct proc *p;
632 	int nxt;
633 	pid_t pid;
634 	struct pid_table *pt;
635 
636 	p = pool_get(&proc_pool, PR_WAITOK);
637 	p->p_stat = SIDL;			/* protect against others */
638 
639 	proc_initspecific(p);
640 	/* allocate next free pid */
641 
642 	for (;;expand_pid_table()) {
643 		if (__predict_false(pid_alloc_cnt >= pid_alloc_lim))
644 			/* ensure pids cycle through 2000+ values */
645 			continue;
646 		mutex_enter(&proclist_lock);
647 		pt = &pid_table[next_free_pt];
648 #ifdef DIAGNOSTIC
649 		if (__predict_false(P_VALID(pt->pt_proc) || pt->pt_pgrp))
650 			panic("proc_alloc: slot busy");
651 #endif
652 		nxt = P_NEXT(pt->pt_proc);
653 		if (nxt & pid_tbl_mask)
654 			break;
655 		/* Table full - expand (NB last entry not used....) */
656 		mutex_exit(&proclist_lock);
657 	}
658 
659 	/* pid is 'saved use count' + 'size' + entry */
660 	pid = (nxt & ~pid_tbl_mask) + pid_tbl_mask + 1 + next_free_pt;
661 	if ((uint)pid > (uint)pid_max)
662 		pid &= pid_tbl_mask;
663 	p->p_pid = pid;
664 	next_free_pt = nxt & pid_tbl_mask;
665 
666 	/* Grab table slot */
667 	mutex_enter(&proclist_mutex);
668 	pt->pt_proc = p;
669 	mutex_exit(&proclist_mutex);
670 	pid_alloc_cnt++;
671 
672 	mutex_exit(&proclist_lock);
673 
674 	return p;
675 }
676 
677 /*
678  * Free a process id - called from proc_free (in kern_exit.c)
679  *
680  * Called with the proclist_lock held.
681  */
682 void
683 proc_free_pid(struct proc *p)
684 {
685 	pid_t pid = p->p_pid;
686 	struct pid_table *pt;
687 
688 	KASSERT(mutex_owned(&proclist_lock));
689 
690 	pt = &pid_table[pid & pid_tbl_mask];
691 #ifdef DIAGNOSTIC
692 	if (__predict_false(pt->pt_proc != p))
693 		panic("proc_free: pid_table mismatch, pid %x, proc %p",
694 			pid, p);
695 #endif
696 	mutex_enter(&proclist_mutex);
697 	/* save pid use count in slot */
698 	pt->pt_proc = P_FREE(pid & ~pid_tbl_mask);
699 
700 	if (pt->pt_pgrp == NULL) {
701 		/* link last freed entry onto ours */
702 		pid &= pid_tbl_mask;
703 		pt = &pid_table[last_free_pt];
704 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pid);
705 		last_free_pt = pid;
706 		pid_alloc_cnt--;
707 	}
708 	mutex_exit(&proclist_mutex);
709 
710 	atomic_dec_uint(&nprocs);
711 }
712 
713 /*
714  * Move p to a new or existing process group (and session)
715  *
716  * If we are creating a new pgrp, the pgid should equal
717  * the calling process' pid.
718  * If is only valid to enter a process group that is in the session
719  * of the process.
720  * Also mksess should only be set if we are creating a process group
721  *
722  * Only called from sys_setsid, sys_setpgid/sys_setpgrp and the
723  * SYSV setpgrp support for hpux.
724  */
725 int
726 enterpgrp(struct proc *curp, pid_t pid, pid_t pgid, int mksess)
727 {
728 	struct pgrp *new_pgrp, *pgrp;
729 	struct session *sess;
730 	struct proc *p;
731 	int rval;
732 	pid_t pg_id = NO_PGID;
733 
734 	if (mksess)
735 		sess = pool_get(&session_pool, PR_WAITOK);
736 	else
737 		sess = NULL;
738 
739 	/* Allocate data areas we might need before doing any validity checks */
740 	mutex_enter(&proclist_lock);		/* Because pid_table might change */
741 	if (pid_table[pgid & pid_tbl_mask].pt_pgrp == 0) {
742 		mutex_exit(&proclist_lock);
743 		new_pgrp = pool_get(&pgrp_pool, PR_WAITOK);
744 		mutex_enter(&proclist_lock);
745 	} else
746 		new_pgrp = NULL;
747 	rval = EPERM;	/* most common error (to save typing) */
748 
749 	/* Check pgrp exists or can be created */
750 	pgrp = pid_table[pgid & pid_tbl_mask].pt_pgrp;
751 	if (pgrp != NULL && pgrp->pg_id != pgid)
752 		goto done;
753 
754 	/* Can only set another process under restricted circumstances. */
755 	if (pid != curp->p_pid) {
756 		/* must exist and be one of our children... */
757 		if ((p = p_find(pid, PFIND_LOCKED)) == NULL ||
758 		    !inferior(p, curp)) {
759 			rval = ESRCH;
760 			goto done;
761 		}
762 		/* ... in the same session... */
763 		if (sess != NULL || p->p_session != curp->p_session)
764 			goto done;
765 		/* ... existing pgid must be in same session ... */
766 		if (pgrp != NULL && pgrp->pg_session != p->p_session)
767 			goto done;
768 		/* ... and not done an exec. */
769 		if (p->p_flag & PK_EXEC) {
770 			rval = EACCES;
771 			goto done;
772 		}
773 	} else {
774 		/* ... setsid() cannot re-enter a pgrp */
775 		if (mksess && (curp->p_pgid == curp->p_pid ||
776 		    pg_find(curp->p_pid, PFIND_LOCKED)))
777 			goto done;
778 		p = curp;
779 	}
780 
781 	/* Changing the process group/session of a session
782 	   leader is definitely off limits. */
783 	if (SESS_LEADER(p)) {
784 		if (sess == NULL && p->p_pgrp == pgrp)
785 			/* unless it's a definite noop */
786 			rval = 0;
787 		goto done;
788 	}
789 
790 	/* Can only create a process group with id of process */
791 	if (pgrp == NULL && pgid != pid)
792 		goto done;
793 
794 	/* Can only create a session if creating pgrp */
795 	if (sess != NULL && pgrp != NULL)
796 		goto done;
797 
798 	/* Check we allocated memory for a pgrp... */
799 	if (pgrp == NULL && new_pgrp == NULL)
800 		goto done;
801 
802 	/* Don't attach to 'zombie' pgrp */
803 	if (pgrp != NULL && LIST_EMPTY(&pgrp->pg_members))
804 		goto done;
805 
806 	/* Expect to succeed now */
807 	rval = 0;
808 
809 	if (pgrp == p->p_pgrp)
810 		/* nothing to do */
811 		goto done;
812 
813 	/* Ok all setup, link up required structures */
814 
815 	if (pgrp == NULL) {
816 		pgrp = new_pgrp;
817 		new_pgrp = 0;
818 		if (sess != NULL) {
819 			sess->s_sid = p->p_pid;
820 			sess->s_leader = p;
821 			sess->s_count = 1;
822 			sess->s_ttyvp = NULL;
823 			sess->s_ttyp = NULL;
824 			sess->s_flags = p->p_session->s_flags & ~S_LOGIN_SET;
825 			memcpy(sess->s_login, p->p_session->s_login,
826 			    sizeof(sess->s_login));
827 			p->p_lflag &= ~PL_CONTROLT;
828 		} else {
829 			sess = p->p_pgrp->pg_session;
830 			SESSHOLD(sess);
831 		}
832 		pgrp->pg_session = sess;
833 		sess = 0;
834 
835 		pgrp->pg_id = pgid;
836 		LIST_INIT(&pgrp->pg_members);
837 #ifdef DIAGNOSTIC
838 		if (__predict_false(pid_table[pgid & pid_tbl_mask].pt_pgrp))
839 			panic("enterpgrp: pgrp table slot in use");
840 		if (__predict_false(mksess && p != curp))
841 			panic("enterpgrp: mksession and p != curproc");
842 #endif
843 		mutex_enter(&proclist_mutex);
844 		pid_table[pgid & pid_tbl_mask].pt_pgrp = pgrp;
845 		pgrp->pg_jobc = 0;
846 	} else
847 		mutex_enter(&proclist_mutex);
848 
849 #ifdef notyet
850 	/*
851 	 * If there's a controlling terminal for the current session, we
852 	 * have to interlock with it.  See ttread().
853 	 */
854 	if (p->p_session->s_ttyvp != NULL) {
855 		tp = p->p_session->s_ttyp;
856 		mutex_enter(&tp->t_mutex);
857 	} else
858 		tp = NULL;
859 #endif
860 
861 	/*
862 	 * Adjust eligibility of affected pgrps to participate in job control.
863 	 * Increment eligibility counts before decrementing, otherwise we
864 	 * could reach 0 spuriously during the first call.
865 	 */
866 	fixjobc(p, pgrp, 1);
867 	fixjobc(p, p->p_pgrp, 0);
868 
869 	/* Move process to requested group. */
870 	LIST_REMOVE(p, p_pglist);
871 	if (LIST_EMPTY(&p->p_pgrp->pg_members))
872 		/* defer delete until we've dumped the lock */
873 		pg_id = p->p_pgrp->pg_id;
874 	p->p_pgrp = pgrp;
875 	LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
876 	mutex_exit(&proclist_mutex);
877 
878 #ifdef notyet
879 	/* Done with the swap; we can release the tty mutex. */
880 	if (tp != NULL)
881 		mutex_exit(&tp->t_mutex);
882 #endif
883 
884     done:
885 	if (pg_id != NO_PGID)
886 		pg_delete(pg_id);
887 	mutex_exit(&proclist_lock);
888 	if (sess != NULL)
889 		pool_put(&session_pool, sess);
890 	if (new_pgrp != NULL)
891 		pool_put(&pgrp_pool, new_pgrp);
892 #ifdef DEBUG_PGRP
893 	if (__predict_false(rval))
894 		printf("enterpgrp(%d,%d,%d), curproc %d, rval %d\n",
895 			pid, pgid, mksess, curp->p_pid, rval);
896 #endif
897 	return rval;
898 }
899 
900 /*
901  * Remove a process from its process group.  Must be called with the
902  * proclist_lock held.
903  */
904 void
905 leavepgrp(struct proc *p)
906 {
907 	struct pgrp *pgrp;
908 
909 	KASSERT(mutex_owned(&proclist_lock));
910 
911 	/*
912 	 * If there's a controlling terminal for the session, we have to
913 	 * interlock with it.  See ttread().
914 	 */
915 	mutex_enter(&proclist_mutex);
916 #ifdef notyet
917 	if (p_>p_session->s_ttyvp != NULL) {
918 		tp = p->p_session->s_ttyp;
919 		mutex_enter(&tp->t_mutex);
920 	} else
921 		tp = NULL;
922 #endif
923 
924 	pgrp = p->p_pgrp;
925 	LIST_REMOVE(p, p_pglist);
926 	p->p_pgrp = NULL;
927 
928 #ifdef notyet
929 	if (tp != NULL)
930 		mutex_exit(&tp->t_mutex);
931 #endif
932 	mutex_exit(&proclist_mutex);
933 
934 	if (LIST_EMPTY(&pgrp->pg_members))
935 		pg_delete(pgrp->pg_id);
936 }
937 
938 /*
939  * Free a process group.  Must be called with the proclist_lock held.
940  */
941 static void
942 pg_free(pid_t pg_id)
943 {
944 	struct pgrp *pgrp;
945 	struct pid_table *pt;
946 
947 	KASSERT(mutex_owned(&proclist_lock));
948 
949 	pt = &pid_table[pg_id & pid_tbl_mask];
950 	pgrp = pt->pt_pgrp;
951 #ifdef DIAGNOSTIC
952 	if (__predict_false(!pgrp || pgrp->pg_id != pg_id
953 	    || !LIST_EMPTY(&pgrp->pg_members)))
954 		panic("pg_free: process group absent or has members");
955 #endif
956 	pt->pt_pgrp = 0;
957 
958 	if (!P_VALID(pt->pt_proc)) {
959 		/* orphaned pgrp, put slot onto free list */
960 #ifdef DIAGNOSTIC
961 		if (__predict_false(P_NEXT(pt->pt_proc) & pid_tbl_mask))
962 			panic("pg_free: process slot on free list");
963 #endif
964 		mutex_enter(&proclist_mutex);
965 		pg_id &= pid_tbl_mask;
966 		pt = &pid_table[last_free_pt];
967 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pg_id);
968 		mutex_exit(&proclist_mutex);
969 		last_free_pt = pg_id;
970 		pid_alloc_cnt--;
971 	}
972 	pool_put(&pgrp_pool, pgrp);
973 }
974 
975 /*
976  * Delete a process group.  Must be called with the proclist_lock held.
977  */
978 static void
979 pg_delete(pid_t pg_id)
980 {
981 	struct pgrp *pgrp;
982 	struct tty *ttyp;
983 	struct session *ss;
984 	int is_pgrp_leader;
985 
986 	KASSERT(mutex_owned(&proclist_lock));
987 
988 	pgrp = pid_table[pg_id & pid_tbl_mask].pt_pgrp;
989 	if (pgrp == NULL || pgrp->pg_id != pg_id ||
990 	    !LIST_EMPTY(&pgrp->pg_members))
991 		return;
992 
993 	ss = pgrp->pg_session;
994 
995 	/* Remove reference (if any) from tty to this process group */
996 	ttyp = ss->s_ttyp;
997 	if (ttyp != NULL && ttyp->t_pgrp == pgrp) {
998 		ttyp->t_pgrp = NULL;
999 #ifdef DIAGNOSTIC
1000 		if (ttyp->t_session != ss)
1001 			panic("pg_delete: wrong session on terminal");
1002 #endif
1003 	}
1004 
1005 	/*
1006 	 * The leading process group in a session is freed
1007 	 * by sessdelete() if last reference.
1008 	 */
1009 	is_pgrp_leader = (ss->s_sid == pgrp->pg_id);
1010 	SESSRELE(ss);
1011 
1012 	if (is_pgrp_leader)
1013 		return;
1014 
1015 	pg_free(pg_id);
1016 }
1017 
1018 /*
1019  * Delete session - called from SESSRELE when s_count becomes zero.
1020  * Must be called with the proclist_lock held.
1021  */
1022 void
1023 sessdelete(struct session *ss)
1024 {
1025 
1026 	KASSERT(mutex_owned(&proclist_lock));
1027 
1028 	/*
1029 	 * We keep the pgrp with the same id as the session in
1030 	 * order to stop a process being given the same pid.
1031 	 * Since the pgrp holds a reference to the session, it
1032 	 * must be a 'zombie' pgrp by now.
1033 	 */
1034 	pg_free(ss->s_sid);
1035 	pool_put(&session_pool, ss);
1036 }
1037 
1038 /*
1039  * Adjust pgrp jobc counters when specified process changes process group.
1040  * We count the number of processes in each process group that "qualify"
1041  * the group for terminal job control (those with a parent in a different
1042  * process group of the same session).  If that count reaches zero, the
1043  * process group becomes orphaned.  Check both the specified process'
1044  * process group and that of its children.
1045  * entering == 0 => p is leaving specified group.
1046  * entering == 1 => p is entering specified group.
1047  *
1048  * Call with proclist_lock held.
1049  */
1050 void
1051 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
1052 {
1053 	struct pgrp *hispgrp;
1054 	struct session *mysession = pgrp->pg_session;
1055 	struct proc *child;
1056 
1057 	KASSERT(mutex_owned(&proclist_lock));
1058 	KASSERT(mutex_owned(&proclist_mutex));
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 			mutex_enter(&p->p_smutex);
1068 			p->p_sflag &= ~PS_ORPHANPG;
1069 			mutex_exit(&p->p_smutex);
1070 			pgrp->pg_jobc++;
1071 		} else if (--pgrp->pg_jobc == 0)
1072 			orphanpg(pgrp);
1073 	}
1074 
1075 	/*
1076 	 * Check this process' children to see whether they qualify
1077 	 * their process groups; if so, adjust counts for children's
1078 	 * process groups.
1079 	 */
1080 	LIST_FOREACH(child, &p->p_children, p_sibling) {
1081 		hispgrp = child->p_pgrp;
1082 		if (hispgrp != pgrp && hispgrp->pg_session == mysession &&
1083 		    !P_ZOMBIE(child)) {
1084 			if (entering) {
1085 				mutex_enter(&child->p_smutex);
1086 				child->p_sflag &= ~PS_ORPHANPG;
1087 				mutex_exit(&child->p_smutex);
1088 				hispgrp->pg_jobc++;
1089 			} else if (--hispgrp->pg_jobc == 0)
1090 				orphanpg(hispgrp);
1091 		}
1092 	}
1093 }
1094 
1095 /*
1096  * A process group has become orphaned;
1097  * if there are any stopped processes in the group,
1098  * hang-up all process in that group.
1099  *
1100  * Call with proclist_lock held.
1101  */
1102 static void
1103 orphanpg(struct pgrp *pg)
1104 {
1105 	struct proc *p;
1106 	int doit;
1107 
1108 	KASSERT(mutex_owned(&proclist_lock));
1109 	KASSERT(mutex_owned(&proclist_mutex));
1110 
1111 	doit = 0;
1112 
1113 	LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1114 		mutex_enter(&p->p_smutex);
1115 		if (p->p_stat == SSTOP) {
1116 			doit = 1;
1117 			p->p_sflag |= PS_ORPHANPG;
1118 		}
1119 		mutex_exit(&p->p_smutex);
1120 	}
1121 
1122 	if (doit) {
1123 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1124 			psignal(p, SIGHUP);
1125 			psignal(p, SIGCONT);
1126 		}
1127 	}
1128 }
1129 
1130 #ifdef DDB
1131 #include <ddb/db_output.h>
1132 void pidtbl_dump(void);
1133 void
1134 pidtbl_dump(void)
1135 {
1136 	struct pid_table *pt;
1137 	struct proc *p;
1138 	struct pgrp *pgrp;
1139 	int id;
1140 
1141 	db_printf("pid table %p size %x, next %x, last %x\n",
1142 		pid_table, pid_tbl_mask+1,
1143 		next_free_pt, last_free_pt);
1144 	for (pt = pid_table, id = 0; id <= pid_tbl_mask; id++, pt++) {
1145 		p = pt->pt_proc;
1146 		if (!P_VALID(p) && !pt->pt_pgrp)
1147 			continue;
1148 		db_printf("  id %x: ", id);
1149 		if (P_VALID(p))
1150 			db_printf("proc %p id %d (0x%x) %s\n",
1151 				p, p->p_pid, p->p_pid, p->p_comm);
1152 		else
1153 			db_printf("next %x use %x\n",
1154 				P_NEXT(p) & pid_tbl_mask,
1155 				P_NEXT(p) & ~pid_tbl_mask);
1156 		if ((pgrp = pt->pt_pgrp)) {
1157 			db_printf("\tsession %p, sid %d, count %d, login %s\n",
1158 			    pgrp->pg_session, pgrp->pg_session->s_sid,
1159 			    pgrp->pg_session->s_count,
1160 			    pgrp->pg_session->s_login);
1161 			db_printf("\tpgrp %p, pg_id %d, pg_jobc %d, members %p\n",
1162 			    pgrp, pgrp->pg_id, pgrp->pg_jobc,
1163 			    pgrp->pg_members.lh_first);
1164 			for (p = pgrp->pg_members.lh_first; p != 0;
1165 			    p = p->p_pglist.le_next) {
1166 				db_printf("\t\tpid %d addr %p pgrp %p %s\n",
1167 				    p->p_pid, p, p->p_pgrp, p->p_comm);
1168 			}
1169 		}
1170 	}
1171 }
1172 #endif /* DDB */
1173 
1174 #ifdef KSTACK_CHECK_MAGIC
1175 #include <sys/user.h>
1176 
1177 #define	KSTACK_MAGIC	0xdeadbeaf
1178 
1179 /* XXX should be per process basis? */
1180 int kstackleftmin = KSTACK_SIZE;
1181 int kstackleftthres = KSTACK_SIZE / 8; /* warn if remaining stack is
1182 					  less than this */
1183 
1184 void
1185 kstack_setup_magic(const struct lwp *l)
1186 {
1187 	uint32_t *ip;
1188 	uint32_t const *end;
1189 
1190 	KASSERT(l != NULL);
1191 	KASSERT(l != &lwp0);
1192 
1193 	/*
1194 	 * fill all the stack with magic number
1195 	 * so that later modification on it can be detected.
1196 	 */
1197 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1198 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1199 	for (; ip < end; ip++) {
1200 		*ip = KSTACK_MAGIC;
1201 	}
1202 }
1203 
1204 void
1205 kstack_check_magic(const struct lwp *l)
1206 {
1207 	uint32_t const *ip, *end;
1208 	int stackleft;
1209 
1210 	KASSERT(l != NULL);
1211 
1212 	/* don't check proc0 */ /*XXX*/
1213 	if (l == &lwp0)
1214 		return;
1215 
1216 #ifdef __MACHINE_STACK_GROWS_UP
1217 	/* stack grows upwards (eg. hppa) */
1218 	ip = (uint32_t *)((void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1219 	end = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1220 	for (ip--; ip >= end; ip--)
1221 		if (*ip != KSTACK_MAGIC)
1222 			break;
1223 
1224 	stackleft = (void *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE - (void *)ip;
1225 #else /* __MACHINE_STACK_GROWS_UP */
1226 	/* stack grows downwards (eg. i386) */
1227 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
1228 	end = (uint32_t *)((char *)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
1229 	for (; ip < end; ip++)
1230 		if (*ip != KSTACK_MAGIC)
1231 			break;
1232 
1233 	stackleft = ((const char *)ip) - (const char *)KSTACK_LOWEST_ADDR(l);
1234 #endif /* __MACHINE_STACK_GROWS_UP */
1235 
1236 	if (kstackleftmin > stackleft) {
1237 		kstackleftmin = stackleft;
1238 		if (stackleft < kstackleftthres)
1239 			printf("warning: kernel stack left %d bytes"
1240 			    "(pid %u:lid %u)\n", stackleft,
1241 			    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1242 	}
1243 
1244 	if (stackleft <= 0) {
1245 		panic("magic on the top of kernel stack changed for "
1246 		    "pid %u, lid %u: maybe kernel stack overflow",
1247 		    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
1248 	}
1249 }
1250 #endif /* KSTACK_CHECK_MAGIC */
1251 
1252 /*
1253  * XXXSMP this is bust, it grabs a read lock and then messes about
1254  * with allproc.
1255  */
1256 int
1257 proclist_foreach_call(struct proclist *list,
1258     int (*callback)(struct proc *, void *arg), void *arg)
1259 {
1260 	struct proc marker;
1261 	struct proc *p;
1262 	struct lwp * const l = curlwp;
1263 	int ret = 0;
1264 
1265 	marker.p_flag = PK_MARKER;
1266 	uvm_lwp_hold(l);
1267 	mutex_enter(&proclist_lock);
1268 	for (p = LIST_FIRST(list); ret == 0 && p != NULL;) {
1269 		if (p->p_flag & PK_MARKER) {
1270 			p = LIST_NEXT(p, p_list);
1271 			continue;
1272 		}
1273 		LIST_INSERT_AFTER(p, &marker, p_list);
1274 		ret = (*callback)(p, arg);
1275 		KASSERT(mutex_owned(&proclist_lock));
1276 		p = LIST_NEXT(&marker, p_list);
1277 		LIST_REMOVE(&marker, p_list);
1278 	}
1279 	mutex_exit(&proclist_lock);
1280 	uvm_lwp_rele(l);
1281 
1282 	return ret;
1283 }
1284 
1285 int
1286 proc_vmspace_getref(struct proc *p, struct vmspace **vm)
1287 {
1288 
1289 	/* XXXCDC: how should locking work here? */
1290 
1291 	/* curproc exception is for coredump. */
1292 
1293 	if ((p != curproc && (p->p_sflag & PS_WEXIT) != 0) ||
1294 	    (p->p_vmspace->vm_refcnt < 1)) { /* XXX */
1295 		return EFAULT;
1296 	}
1297 
1298 	uvmspace_addref(p->p_vmspace);
1299 	*vm = p->p_vmspace;
1300 
1301 	return 0;
1302 }
1303 
1304 /*
1305  * Acquire a write lock on the process credential.
1306  */
1307 void
1308 proc_crmod_enter(void)
1309 {
1310 	struct lwp *l = curlwp;
1311 	struct proc *p = l->l_proc;
1312 	struct plimit *lim;
1313 	kauth_cred_t oc;
1314 	char *cn;
1315 
1316 	/* Reset what needs to be reset in plimit. */
1317 	if (p->p_limit->pl_corename != defcorename) {
1318 		lim_privatise(p, false);
1319 		lim = p->p_limit;
1320 		mutex_enter(&lim->pl_lock);
1321 		cn = lim->pl_corename;
1322 		lim->pl_corename = defcorename;
1323 		mutex_exit(&lim->pl_lock);
1324 		if (cn != defcorename)
1325 			free(cn, M_TEMP);
1326 	}
1327 
1328 	mutex_enter(&p->p_mutex);
1329 
1330 	/* Ensure the LWP cached credentials are up to date. */
1331 	if ((oc = l->l_cred) != p->p_cred) {
1332 		kauth_cred_hold(p->p_cred);
1333 		l->l_cred = p->p_cred;
1334 		kauth_cred_free(oc);
1335 	}
1336 
1337 }
1338 
1339 /*
1340  * Set in a new process credential, and drop the write lock.  The credential
1341  * must have a reference already.  Optionally, free a no-longer required
1342  * credential.  The scheduler also needs to inspect p_cred, so we also
1343  * briefly acquire the sched state mutex.
1344  */
1345 void
1346 proc_crmod_leave(kauth_cred_t scred, kauth_cred_t fcred, bool sugid)
1347 {
1348 	struct lwp *l = curlwp;
1349 	struct proc *p = l->l_proc;
1350 	kauth_cred_t oc;
1351 
1352 	/* Is there a new credential to set in? */
1353 	if (scred != NULL) {
1354 		mutex_enter(&p->p_smutex);
1355 		p->p_cred = scred;
1356 		mutex_exit(&p->p_smutex);
1357 
1358 		/* Ensure the LWP cached credentials are up to date. */
1359 		if ((oc = l->l_cred) != scred) {
1360 			kauth_cred_hold(scred);
1361 			l->l_cred = scred;
1362 		}
1363 	} else
1364 		oc = NULL;	/* XXXgcc */
1365 
1366 	if (sugid) {
1367 		/*
1368 		 * Mark process as having changed credentials, stops
1369 		 * tracing etc.
1370 		 */
1371 		p->p_flag |= PK_SUGID;
1372 	}
1373 
1374 	mutex_exit(&p->p_mutex);
1375 
1376 	/* If there is a credential to be released, free it now. */
1377 	if (fcred != NULL) {
1378 		KASSERT(scred != NULL);
1379 		kauth_cred_free(fcred);
1380 		if (oc != scred)
1381 			kauth_cred_free(oc);
1382 	}
1383 }
1384 
1385 /*
1386  * proc_specific_key_create --
1387  *	Create a key for subsystem proc-specific data.
1388  */
1389 int
1390 proc_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1391 {
1392 
1393 	return (specificdata_key_create(proc_specificdata_domain, keyp, dtor));
1394 }
1395 
1396 /*
1397  * proc_specific_key_delete --
1398  *	Delete a key for subsystem proc-specific data.
1399  */
1400 void
1401 proc_specific_key_delete(specificdata_key_t key)
1402 {
1403 
1404 	specificdata_key_delete(proc_specificdata_domain, key);
1405 }
1406 
1407 /*
1408  * proc_initspecific --
1409  *	Initialize a proc's specificdata container.
1410  */
1411 void
1412 proc_initspecific(struct proc *p)
1413 {
1414 	int error;
1415 
1416 	error = specificdata_init(proc_specificdata_domain, &p->p_specdataref);
1417 	KASSERT(error == 0);
1418 }
1419 
1420 /*
1421  * proc_finispecific --
1422  *	Finalize a proc's specificdata container.
1423  */
1424 void
1425 proc_finispecific(struct proc *p)
1426 {
1427 
1428 	specificdata_fini(proc_specificdata_domain, &p->p_specdataref);
1429 }
1430 
1431 /*
1432  * proc_getspecific --
1433  *	Return proc-specific data corresponding to the specified key.
1434  */
1435 void *
1436 proc_getspecific(struct proc *p, specificdata_key_t key)
1437 {
1438 
1439 	return (specificdata_getspecific(proc_specificdata_domain,
1440 					 &p->p_specdataref, key));
1441 }
1442 
1443 /*
1444  * proc_setspecific --
1445  *	Set proc-specific data corresponding to the specified key.
1446  */
1447 void
1448 proc_setspecific(struct proc *p, specificdata_key_t key, void *data)
1449 {
1450 
1451 	specificdata_setspecific(proc_specificdata_domain,
1452 				 &p->p_specdataref, key, data);
1453 }
1454