Home | History | Annotate | Line # | Download | only in kern
kern_proc.c revision 1.100
      1 /*	$NetBSD: kern_proc.c,v 1.100 2007/02/09 21:55:31 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 1.100 2007/02/09 21:55:31 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/kauth.h>
     99 #include <sys/sleepq.h>
    100 
    101 #include <uvm/uvm.h>
    102 #include <uvm/uvm_extern.h>
    103 
    104 /*
    105  * Other process lists
    106  */
    107 
    108 struct proclist allproc;
    109 struct proclist zombproc;	/* resources have been freed */
    110 
    111 /*
    112  * There are two locks on global process state.
    113  *
    114  * 1. proclist_lock is a reader/writer lock and is used when modifying or
    115  * examining process state from a process context.  It protects our internal
    116  * tables, all of the process lists, and a number of members of struct lwp
    117  * and struct proc.
    118 
    119  * 2. proclist_mutex is used when allproc must be traversed from an
    120  * interrupt context, or when we must signal processes from an interrupt
    121  * context.  The proclist_lock should always be used in preference.
    122  *
    123  *	proclist_lock	proclist_mutex	structure
    124  *	--------------- --------------- -----------------
    125  *	x				zombproc
    126  *	x		x		pid_table
    127  *	x				proc::p_pptr
    128  *	x				proc::p_sibling
    129  *	x				proc::p_children
    130  *	x		x		allproc
    131  *	x		x		proc::p_pgrp
    132  *	x		x		proc::p_pglist
    133  *	x		x		proc::p_session
    134  *	x		x		proc::p_list
    135  *			x		alllwp
    136  *			x		lwp::l_list
    137  *
    138  * The lock order for processes and LWPs is approximately as following:
    139  *
    140  * kernel_mutex
    141  * -> proclist_lock
    142  *    -> proclist_mutex
    143  *	-> proc::p_mutex
    144  *         -> proc::p_smutex
    145  */
    146 krwlock_t	proclist_lock;
    147 kmutex_t	proclist_mutex;
    148 
    149 /*
    150  * pid to proc lookup is done by indexing the pid_table array.
    151  * Since pid numbers are only allocated when an empty slot
    152  * has been found, there is no need to search any lists ever.
    153  * (an orphaned pgrp will lock the slot, a session will lock
    154  * the pgrp with the same number.)
    155  * If the table is too small it is reallocated with twice the
    156  * previous size and the entries 'unzipped' into the two halves.
    157  * A linked list of free entries is passed through the pt_proc
    158  * field of 'free' items - set odd to be an invalid ptr.
    159  */
    160 
    161 struct pid_table {
    162 	struct proc	*pt_proc;
    163 	struct pgrp	*pt_pgrp;
    164 };
    165 #if 1	/* strongly typed cast - should be a noop */
    166 static inline uint p2u(struct proc *p) { return (uint)(uintptr_t)p; }
    167 #else
    168 #define p2u(p) ((uint)p)
    169 #endif
    170 #define P_VALID(p) (!(p2u(p) & 1))
    171 #define P_NEXT(p) (p2u(p) >> 1)
    172 #define P_FREE(pid) ((struct proc *)(uintptr_t)((pid) << 1 | 1))
    173 
    174 #define INITIAL_PID_TABLE_SIZE	(1 << 5)
    175 static struct pid_table *pid_table;
    176 static uint pid_tbl_mask = INITIAL_PID_TABLE_SIZE - 1;
    177 static uint pid_alloc_lim;	/* max we allocate before growing table */
    178 static uint pid_alloc_cnt;	/* number of allocated pids */
    179 
    180 /* links through free slots - never empty! */
    181 static uint next_free_pt, last_free_pt;
    182 static pid_t pid_max = PID_MAX;		/* largest value we allocate */
    183 
    184 /* Components of the first process -- never freed. */
    185 struct session session0;
    186 struct pgrp pgrp0;
    187 struct proc proc0;
    188 struct lwp lwp0 __aligned(MIN_LWP_ALIGNMENT);
    189 kauth_cred_t cred0;
    190 struct filedesc0 filedesc0;
    191 struct cwdinfo cwdi0;
    192 struct plimit limit0;
    193 struct pstats pstat0;
    194 struct vmspace vmspace0;
    195 struct sigacts sigacts0;
    196 struct turnstile turnstile0;
    197 
    198 extern struct user *proc0paddr;
    199 
    200 extern const struct emul emul_netbsd;	/* defined in kern_exec.c */
    201 
    202 int nofile = NOFILE;
    203 int maxuprc = MAXUPRC;
    204 int cmask = CMASK;
    205 
    206 POOL_INIT(proc_pool, sizeof(struct proc), 0, 0, 0, "procpl",
    207     &pool_allocator_nointr);
    208 POOL_INIT(pgrp_pool, sizeof(struct pgrp), 0, 0, 0, "pgrppl",
    209     &pool_allocator_nointr);
    210 POOL_INIT(plimit_pool, sizeof(struct plimit), 0, 0, 0, "plimitpl",
    211     &pool_allocator_nointr);
    212 POOL_INIT(pstats_pool, sizeof(struct pstats), 0, 0, 0, "pstatspl",
    213     &pool_allocator_nointr);
    214 POOL_INIT(rusage_pool, sizeof(struct rusage), 0, 0, 0, "rusgepl",
    215     &pool_allocator_nointr);
    216 POOL_INIT(session_pool, sizeof(struct session), 0, 0, 0, "sessionpl",
    217     &pool_allocator_nointr);
    218 
    219 MALLOC_DEFINE(M_EMULDATA, "emuldata", "Per-process emulation data");
    220 MALLOC_DEFINE(M_PROC, "proc", "Proc structures");
    221 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures");
    222 
    223 /*
    224  * The process list descriptors, used during pid allocation and
    225  * by sysctl.  No locking on this data structure is needed since
    226  * it is completely static.
    227  */
    228 const struct proclist_desc proclists[] = {
    229 	{ &allproc	},
    230 	{ &zombproc	},
    231 	{ NULL		},
    232 };
    233 
    234 static void orphanpg(struct pgrp *);
    235 static void pg_delete(pid_t);
    236 
    237 static specificdata_domain_t proc_specificdata_domain;
    238 
    239 /*
    240  * Initialize global process hashing structures.
    241  */
    242 void
    243 procinit(void)
    244 {
    245 	const struct proclist_desc *pd;
    246 	int i;
    247 #define	LINK_EMPTY ((PID_MAX + INITIAL_PID_TABLE_SIZE) & ~(INITIAL_PID_TABLE_SIZE - 1))
    248 
    249 	for (pd = proclists; pd->pd_list != NULL; pd++)
    250 		LIST_INIT(pd->pd_list);
    251 
    252 	/*
    253 	 * XXX p_smutex can be IPL_VM except for audio drivers
    254 	 * XXX proclist_lock must die
    255 	 */
    256 	rw_init(&proclist_lock);
    257 	mutex_init(&proclist_mutex, MUTEX_SPIN, IPL_SCHED);
    258 
    259 	pid_table = malloc(INITIAL_PID_TABLE_SIZE * sizeof *pid_table,
    260 			    M_PROC, M_WAITOK);
    261 	/* Set free list running through table...
    262 	   Preset 'use count' above PID_MAX so we allocate pid 1 next. */
    263 	for (i = 0; i <= pid_tbl_mask; i++) {
    264 		pid_table[i].pt_proc = P_FREE(LINK_EMPTY + i + 1);
    265 		pid_table[i].pt_pgrp = 0;
    266 	}
    267 	/* slot 0 is just grabbed */
    268 	next_free_pt = 1;
    269 	/* Need to fix last entry. */
    270 	last_free_pt = pid_tbl_mask;
    271 	pid_table[last_free_pt].pt_proc = P_FREE(LINK_EMPTY);
    272 	/* point at which we grow table - to avoid reusing pids too often */
    273 	pid_alloc_lim = pid_tbl_mask - 1;
    274 #undef LINK_EMPTY
    275 
    276 	LIST_INIT(&alllwp);
    277 
    278 	uihashtbl =
    279 	    hashinit(maxproc / 16, HASH_LIST, M_PROC, M_WAITOK, &uihash);
    280 
    281 	proc_specificdata_domain = specificdata_domain_create();
    282 	KASSERT(proc_specificdata_domain != NULL);
    283 }
    284 
    285 /*
    286  * Initialize process 0.
    287  */
    288 void
    289 proc0_init(void)
    290 {
    291 	struct proc *p;
    292 	struct pgrp *pg;
    293 	struct session *sess;
    294 	struct lwp *l;
    295 	u_int i;
    296 	rlim_t lim;
    297 
    298 	p = &proc0;
    299 	pg = &pgrp0;
    300 	sess = &session0;
    301 	l = &lwp0;
    302 
    303 	/* XXX p_smutex can be IPL_VM except for audio drivers */
    304 	mutex_init(&p->p_smutex, MUTEX_SPIN, IPL_SCHED);
    305 	mutex_init(&p->p_stmutex, MUTEX_SPIN, IPL_STATCLOCK);
    306 	mutex_init(&p->p_rasmutex, MUTEX_SPIN, IPL_NONE);
    307 	mutex_init(&p->p_mutex, MUTEX_DEFAULT, IPL_NONE);
    308 	cv_init(&p->p_refcv, "drainref");
    309 	cv_init(&p->p_waitcv, "wait");
    310 	cv_init(&p->p_lwpcv, "lwpwait");
    311 
    312 	LIST_INIT(&p->p_lwps);
    313 	LIST_INIT(&p->p_sigwaiters);
    314 	LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling);
    315 
    316 	p->p_nlwps = 1;
    317 	p->p_nrlwps = 1;
    318 	p->p_refcnt = 1;
    319 
    320 	pid_table[0].pt_proc = p;
    321 	LIST_INSERT_HEAD(&allproc, p, p_list);
    322 	LIST_INSERT_HEAD(&alllwp, l, l_list);
    323 
    324 	p->p_pgrp = pg;
    325 	pid_table[0].pt_pgrp = pg;
    326 	LIST_INIT(&pg->pg_members);
    327 	LIST_INSERT_HEAD(&pg->pg_members, p, p_pglist);
    328 
    329 	pg->pg_session = sess;
    330 	sess->s_count = 1;
    331 	sess->s_sid = 0;
    332 	sess->s_leader = p;
    333 
    334 	/*
    335 	 * Set P_NOCLDWAIT so that kernel threads are reparented to
    336 	 * init(8) when they exit.  init(8) can easily wait them out
    337 	 * for us.
    338 	 */
    339 	p->p_flag = P_SYSTEM | P_NOCLDWAIT;
    340 	p->p_stat = SACTIVE;
    341 	p->p_nice = NZERO;
    342 	p->p_emul = &emul_netbsd;
    343 #ifdef __HAVE_SYSCALL_INTERN
    344 	(*p->p_emul->e_syscall_intern)(p);
    345 #endif
    346 	strncpy(p->p_comm, "swapper", MAXCOMLEN);
    347 
    348 	l->l_mutex = &sched_mutex;
    349 	l->l_flag = L_INMEM | L_SYSTEM;
    350 	l->l_stat = LSONPROC;
    351 	l->l_ts = &turnstile0;
    352 	l->l_syncobj = &sched_syncobj;
    353 	l->l_refcnt = 1;
    354 	l->l_cpu = curcpu();
    355 	l->l_priority = PRIBIO;
    356 	l->l_usrpri = PRIBIO;
    357 
    358 	callout_init(&l->l_tsleep_ch);
    359 	cv_init(&l->l_sigcv, "sigwait");
    360 
    361 	/* Create credentials. */
    362 	cred0 = kauth_cred_alloc();
    363 	p->p_cred = cred0;
    364 	kauth_cred_hold(cred0);
    365 	l->l_cred = cred0;
    366 
    367 	/* Create the CWD info. */
    368 	p->p_cwdi = &cwdi0;
    369 	cwdi0.cwdi_cmask = cmask;
    370 	cwdi0.cwdi_refcnt = 1;
    371 	simple_lock_init(&cwdi0.cwdi_slock);
    372 
    373 	/* Create the limits structures. */
    374 	p->p_limit = &limit0;
    375 	simple_lock_init(&limit0.p_slock);
    376 	for (i = 0; i < sizeof(p->p_rlimit)/sizeof(p->p_rlimit[0]); i++)
    377 		limit0.pl_rlimit[i].rlim_cur =
    378 		    limit0.pl_rlimit[i].rlim_max = RLIM_INFINITY;
    379 
    380 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
    381 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_cur =
    382 	    maxfiles < nofile ? maxfiles : nofile;
    383 
    384 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
    385 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_cur =
    386 	    maxproc < maxuprc ? maxproc : maxuprc;
    387 
    388 	lim = ptoa(uvmexp.free);
    389 	limit0.pl_rlimit[RLIMIT_RSS].rlim_max = lim;
    390 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim;
    391 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3;
    392 	limit0.pl_corename = defcorename;
    393 	limit0.p_refcnt = 1;
    394 
    395 	/* Configure virtual memory system, set vm rlimits. */
    396 	uvm_init_limits(p);
    397 
    398 	/* Initialize file descriptor table for proc0. */
    399 	p->p_fd = &filedesc0.fd_fd;
    400 	fdinit1(&filedesc0);
    401 
    402 	/*
    403 	 * Initialize proc0's vmspace, which uses the kernel pmap.
    404 	 * All kernel processes (which never have user space mappings)
    405 	 * share proc0's vmspace, and thus, the kernel pmap.
    406 	 */
    407 	uvmspace_init(&vmspace0, pmap_kernel(), round_page(VM_MIN_ADDRESS),
    408 	    trunc_page(VM_MAX_ADDRESS));
    409 	p->p_vmspace = &vmspace0;
    410 
    411 	l->l_addr = proc0paddr;				/* XXX */
    412 
    413 	p->p_stats = &pstat0;
    414 
    415 	/* Initialize signal state for proc0. */
    416 	p->p_sigacts = &sigacts0;
    417 	mutex_init(&p->p_sigacts->sa_mutex, MUTEX_SPIN, IPL_NONE);
    418 	siginit(p);
    419 
    420 	proc_initspecific(p);
    421 	lwp_initspecific(l);
    422 }
    423 
    424 /*
    425  * Check that the specified process group is in the session of the
    426  * specified process.
    427  * Treats -ve ids as process ids.
    428  * Used to validate TIOCSPGRP requests.
    429  */
    430 int
    431 pgid_in_session(struct proc *p, pid_t pg_id)
    432 {
    433 	struct pgrp *pgrp;
    434 
    435 	if (pg_id < 0) {
    436 		struct proc *p1 = pfind(-pg_id);
    437 		if (p1 == NULL)
    438 			return EINVAL;
    439 		pgrp = p1->p_pgrp;
    440 	} else {
    441 		pgrp = pgfind(pg_id);
    442 		if (pgrp == NULL)
    443 			return EINVAL;
    444 	}
    445 	if (pgrp->pg_session != p->p_pgrp->pg_session)
    446 		return EPERM;
    447 	return 0;
    448 }
    449 
    450 /*
    451  * Is p an inferior of q?
    452  *
    453  * Call with the proclist_lock held.
    454  */
    455 int
    456 inferior(struct proc *p, struct proc *q)
    457 {
    458 
    459 	for (; p != q; p = p->p_pptr)
    460 		if (p->p_pid == 0)
    461 			return 0;
    462 	return 1;
    463 }
    464 
    465 /*
    466  * Locate a process by number
    467  */
    468 struct proc *
    469 p_find(pid_t pid, uint flags)
    470 {
    471 	struct proc *p;
    472 	char stat;
    473 
    474 	if (!(flags & PFIND_LOCKED))
    475 		rw_enter(&proclist_lock, RW_READER);
    476 
    477 	p = pid_table[pid & pid_tbl_mask].pt_proc;
    478 
    479 	/* Only allow live processes to be found by pid. */
    480 	/* XXXSMP p_stat */
    481 	if (P_VALID(p) && p->p_pid == pid && ((stat = p->p_stat) == SACTIVE ||
    482 	    stat == SSTOP || ((flags & PFIND_ZOMBIE) &&
    483 	    (stat == SZOMB || stat == SDEAD || stat == SDYING)))) {
    484 		if (flags & PFIND_UNLOCK_OK)
    485 			 rw_exit(&proclist_lock);
    486 		return p;
    487 	}
    488 	if (flags & PFIND_UNLOCK_FAIL)
    489 		 rw_exit(&proclist_lock);
    490 	return NULL;
    491 }
    492 
    493 
    494 /*
    495  * Locate a process group by number
    496  */
    497 struct pgrp *
    498 pg_find(pid_t pgid, uint flags)
    499 {
    500 	struct pgrp *pg;
    501 
    502 	if (!(flags & PFIND_LOCKED))
    503 		rw_enter(&proclist_lock, RW_READER);
    504 	pg = pid_table[pgid & pid_tbl_mask].pt_pgrp;
    505 	/*
    506 	 * Can't look up a pgrp that only exists because the session
    507 	 * hasn't died yet (traditional)
    508 	 */
    509 	if (pg == NULL || pg->pg_id != pgid || LIST_EMPTY(&pg->pg_members)) {
    510 		if (flags & PFIND_UNLOCK_FAIL)
    511 			 rw_exit(&proclist_lock);
    512 		return NULL;
    513 	}
    514 
    515 	if (flags & PFIND_UNLOCK_OK)
    516 		rw_exit(&proclist_lock);
    517 	return pg;
    518 }
    519 
    520 static void
    521 expand_pid_table(void)
    522 {
    523 	uint pt_size = pid_tbl_mask + 1;
    524 	struct pid_table *n_pt, *new_pt;
    525 	struct proc *proc;
    526 	struct pgrp *pgrp;
    527 	int i;
    528 	pid_t pid;
    529 
    530 	new_pt = malloc(pt_size * 2 * sizeof *new_pt, M_PROC, M_WAITOK);
    531 
    532 	rw_enter(&proclist_lock, RW_WRITER);
    533 	if (pt_size != pid_tbl_mask + 1) {
    534 		/* Another process beat us to it... */
    535 		rw_exit(&proclist_lock);
    536 		FREE(new_pt, M_PROC);
    537 		return;
    538 	}
    539 
    540 	/*
    541 	 * Copy entries from old table into new one.
    542 	 * If 'pid' is 'odd' we need to place in the upper half,
    543 	 * even pid's to the lower half.
    544 	 * Free items stay in the low half so we don't have to
    545 	 * fixup the reference to them.
    546 	 * We stuff free items on the front of the freelist
    547 	 * because we can't write to unmodified entries.
    548 	 * Processing the table backwards maintains a semblance
    549 	 * of issueing pid numbers that increase with time.
    550 	 */
    551 	i = pt_size - 1;
    552 	n_pt = new_pt + i;
    553 	for (; ; i--, n_pt--) {
    554 		proc = pid_table[i].pt_proc;
    555 		pgrp = pid_table[i].pt_pgrp;
    556 		if (!P_VALID(proc)) {
    557 			/* Up 'use count' so that link is valid */
    558 			pid = (P_NEXT(proc) + pt_size) & ~pt_size;
    559 			proc = P_FREE(pid);
    560 			if (pgrp)
    561 				pid = pgrp->pg_id;
    562 		} else
    563 			pid = proc->p_pid;
    564 
    565 		/* Save entry in appropriate half of table */
    566 		n_pt[pid & pt_size].pt_proc = proc;
    567 		n_pt[pid & pt_size].pt_pgrp = pgrp;
    568 
    569 		/* Put other piece on start of free list */
    570 		pid = (pid ^ pt_size) & ~pid_tbl_mask;
    571 		n_pt[pid & pt_size].pt_proc =
    572 				    P_FREE((pid & ~pt_size) | next_free_pt);
    573 		n_pt[pid & pt_size].pt_pgrp = 0;
    574 		next_free_pt = i | (pid & pt_size);
    575 		if (i == 0)
    576 			break;
    577 	}
    578 
    579 	/* Switch tables */
    580 	mutex_enter(&proclist_mutex);
    581 	n_pt = pid_table;
    582 	pid_table = new_pt;
    583 	mutex_exit(&proclist_mutex);
    584 	pid_tbl_mask = pt_size * 2 - 1;
    585 
    586 	/*
    587 	 * pid_max starts as PID_MAX (= 30000), once we have 16384
    588 	 * allocated pids we need it to be larger!
    589 	 */
    590 	if (pid_tbl_mask > PID_MAX) {
    591 		pid_max = pid_tbl_mask * 2 + 1;
    592 		pid_alloc_lim |= pid_alloc_lim << 1;
    593 	} else
    594 		pid_alloc_lim <<= 1;	/* doubles number of free slots... */
    595 
    596 	rw_exit(&proclist_lock);
    597 	FREE(n_pt, M_PROC);
    598 }
    599 
    600 struct proc *
    601 proc_alloc(void)
    602 {
    603 	struct proc *p;
    604 	int nxt;
    605 	pid_t pid;
    606 	struct pid_table *pt;
    607 
    608 	p = pool_get(&proc_pool, PR_WAITOK);
    609 	p->p_stat = SIDL;			/* protect against others */
    610 
    611 	proc_initspecific(p);
    612 	/* allocate next free pid */
    613 
    614 	for (;;expand_pid_table()) {
    615 		if (__predict_false(pid_alloc_cnt >= pid_alloc_lim))
    616 			/* ensure pids cycle through 2000+ values */
    617 			continue;
    618 		rw_enter(&proclist_lock, RW_WRITER);
    619 		pt = &pid_table[next_free_pt];
    620 #ifdef DIAGNOSTIC
    621 		if (__predict_false(P_VALID(pt->pt_proc) || pt->pt_pgrp))
    622 			panic("proc_alloc: slot busy");
    623 #endif
    624 		nxt = P_NEXT(pt->pt_proc);
    625 		if (nxt & pid_tbl_mask)
    626 			break;
    627 		/* Table full - expand (NB last entry not used....) */
    628 		rw_exit(&proclist_lock);
    629 	}
    630 
    631 	/* pid is 'saved use count' + 'size' + entry */
    632 	pid = (nxt & ~pid_tbl_mask) + pid_tbl_mask + 1 + next_free_pt;
    633 	if ((uint)pid > (uint)pid_max)
    634 		pid &= pid_tbl_mask;
    635 	p->p_pid = pid;
    636 	next_free_pt = nxt & pid_tbl_mask;
    637 
    638 	/* Grab table slot */
    639 	mutex_enter(&proclist_mutex);
    640 	pt->pt_proc = p;
    641 	mutex_exit(&proclist_mutex);
    642 	pid_alloc_cnt++;
    643 
    644 	rw_exit(&proclist_lock);
    645 
    646 	return p;
    647 }
    648 
    649 /*
    650  * Free last resources of a process - called from proc_free (in kern_exit.c)
    651  *
    652  * Called with the proclist_lock write held, and releases upon exit.
    653  */
    654 void
    655 proc_free_mem(struct proc *p)
    656 {
    657 	pid_t pid = p->p_pid;
    658 	struct pid_table *pt;
    659 
    660 	LOCK_ASSERT(rw_write_held(&proclist_lock));
    661 
    662 	pt = &pid_table[pid & pid_tbl_mask];
    663 #ifdef DIAGNOSTIC
    664 	if (__predict_false(pt->pt_proc != p))
    665 		panic("proc_free: pid_table mismatch, pid %x, proc %p",
    666 			pid, p);
    667 #endif
    668 	mutex_enter(&proclist_mutex);
    669 	/* save pid use count in slot */
    670 	pt->pt_proc = P_FREE(pid & ~pid_tbl_mask);
    671 
    672 	if (pt->pt_pgrp == NULL) {
    673 		/* link last freed entry onto ours */
    674 		pid &= pid_tbl_mask;
    675 		pt = &pid_table[last_free_pt];
    676 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pid);
    677 		last_free_pt = pid;
    678 		pid_alloc_cnt--;
    679 	}
    680 	mutex_exit(&proclist_mutex);
    681 
    682 	nprocs--;
    683 	rw_exit(&proclist_lock);
    684 
    685 	pool_put(&proc_pool, p);
    686 }
    687 
    688 /*
    689  * Move p to a new or existing process group (and session)
    690  *
    691  * If we are creating a new pgrp, the pgid should equal
    692  * the calling process' pid.
    693  * If is only valid to enter a process group that is in the session
    694  * of the process.
    695  * Also mksess should only be set if we are creating a process group
    696  *
    697  * Only called from sys_setsid, sys_setpgid/sys_setpgrp and the
    698  * SYSV setpgrp support for hpux.
    699  */
    700 int
    701 enterpgrp(struct proc *curp, pid_t pid, pid_t pgid, int mksess)
    702 {
    703 	struct pgrp *new_pgrp, *pgrp;
    704 	struct session *sess;
    705 	struct proc *p;
    706 	int rval;
    707 	pid_t pg_id = NO_PGID;
    708 
    709 	/* Allocate data areas we might need before doing any validity checks */
    710 	rw_enter(&proclist_lock, RW_READER);		/* Because pid_table might change */
    711 	if (pid_table[pgid & pid_tbl_mask].pt_pgrp == 0) {
    712 		rw_exit(&proclist_lock);
    713 		new_pgrp = pool_get(&pgrp_pool, PR_WAITOK);
    714 	} else {
    715 		rw_exit(&proclist_lock);
    716 		new_pgrp = NULL;
    717 	}
    718 	if (mksess)
    719 		sess = pool_get(&session_pool, PR_WAITOK);
    720 	else
    721 		sess = NULL;
    722 
    723 	rw_enter(&proclist_lock, RW_WRITER);
    724 	rval = EPERM;	/* most common error (to save typing) */
    725 
    726 	/* Check pgrp exists or can be created */
    727 	pgrp = pid_table[pgid & pid_tbl_mask].pt_pgrp;
    728 	if (pgrp != NULL && pgrp->pg_id != pgid)
    729 		goto done;
    730 
    731 	/* Can only set another process under restricted circumstances. */
    732 	if (pid != curp->p_pid) {
    733 		/* must exist and be one of our children... */
    734 		if ((p = p_find(pid, PFIND_LOCKED)) == NULL ||
    735 		    !inferior(p, curp)) {
    736 			rval = ESRCH;
    737 			goto done;
    738 		}
    739 		/* ... in the same session... */
    740 		if (sess != NULL || p->p_session != curp->p_session)
    741 			goto done;
    742 		/* ... existing pgid must be in same session ... */
    743 		if (pgrp != NULL && pgrp->pg_session != p->p_session)
    744 			goto done;
    745 		/* ... and not done an exec. */
    746 		if (p->p_flag & P_EXEC) {
    747 			rval = EACCES;
    748 			goto done;
    749 		}
    750 	} else {
    751 		/* ... setsid() cannot re-enter a pgrp */
    752 		if (mksess && (curp->p_pgid == curp->p_pid ||
    753 		    pg_find(curp->p_pid, PFIND_LOCKED)))
    754 			goto done;
    755 		p = curp;
    756 	}
    757 
    758 	/* Changing the process group/session of a session
    759 	   leader is definitely off limits. */
    760 	if (SESS_LEADER(p)) {
    761 		if (sess == NULL && p->p_pgrp == pgrp)
    762 			/* unless it's a definite noop */
    763 			rval = 0;
    764 		goto done;
    765 	}
    766 
    767 	/* Can only create a process group with id of process */
    768 	if (pgrp == NULL && pgid != pid)
    769 		goto done;
    770 
    771 	/* Can only create a session if creating pgrp */
    772 	if (sess != NULL && pgrp != NULL)
    773 		goto done;
    774 
    775 	/* Check we allocated memory for a pgrp... */
    776 	if (pgrp == NULL && new_pgrp == NULL)
    777 		goto done;
    778 
    779 	/* Don't attach to 'zombie' pgrp */
    780 	if (pgrp != NULL && LIST_EMPTY(&pgrp->pg_members))
    781 		goto done;
    782 
    783 	/* Expect to succeed now */
    784 	rval = 0;
    785 
    786 	if (pgrp == p->p_pgrp)
    787 		/* nothing to do */
    788 		goto done;
    789 
    790 	/* Ok all setup, link up required structures */
    791 
    792 	if (pgrp == NULL) {
    793 		pgrp = new_pgrp;
    794 		new_pgrp = 0;
    795 		if (sess != NULL) {
    796 			sess->s_sid = p->p_pid;
    797 			sess->s_leader = p;
    798 			sess->s_count = 1;
    799 			sess->s_ttyvp = NULL;
    800 			sess->s_ttyp = NULL;
    801 			sess->s_flags = p->p_session->s_flags & ~S_LOGIN_SET;
    802 			memcpy(sess->s_login, p->p_session->s_login,
    803 			    sizeof(sess->s_login));
    804 			p->p_lflag &= ~PL_CONTROLT;
    805 		} else {
    806 			sess = p->p_pgrp->pg_session;
    807 			SESSHOLD(sess);
    808 		}
    809 		pgrp->pg_session = sess;
    810 		sess = 0;
    811 
    812 		pgrp->pg_id = pgid;
    813 		LIST_INIT(&pgrp->pg_members);
    814 #ifdef DIAGNOSTIC
    815 		if (__predict_false(pid_table[pgid & pid_tbl_mask].pt_pgrp))
    816 			panic("enterpgrp: pgrp table slot in use");
    817 		if (__predict_false(mksess && p != curp))
    818 			panic("enterpgrp: mksession and p != curproc");
    819 #endif
    820 		mutex_enter(&proclist_mutex);
    821 		pid_table[pgid & pid_tbl_mask].pt_pgrp = pgrp;
    822 		pgrp->pg_jobc = 0;
    823 	} else
    824 		mutex_enter(&proclist_mutex);
    825 
    826 #ifdef notyet
    827 	/*
    828 	 * If there's a controlling terminal for the current session, we
    829 	 * have to interlock with it.  See ttread().
    830 	 */
    831 	if (p->p_session->s_ttyvp != NULL) {
    832 		tp = p->p_session->s_ttyp;
    833 		mutex_enter(&tp->t_mutex);
    834 	} else
    835 		tp = NULL;
    836 #endif
    837 
    838 	/*
    839 	 * Adjust eligibility of affected pgrps to participate in job control.
    840 	 * Increment eligibility counts before decrementing, otherwise we
    841 	 * could reach 0 spuriously during the first call.
    842 	 */
    843 	fixjobc(p, pgrp, 1);
    844 	fixjobc(p, p->p_pgrp, 0);
    845 
    846 	/* Move process to requested group. */
    847 	LIST_REMOVE(p, p_pglist);
    848 	if (LIST_EMPTY(&p->p_pgrp->pg_members))
    849 		/* defer delete until we've dumped the lock */
    850 		pg_id = p->p_pgrp->pg_id;
    851 	p->p_pgrp = pgrp;
    852 	LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
    853 	mutex_exit(&proclist_mutex);
    854 
    855 #ifdef notyet
    856 	/* Done with the swap; we can release the tty mutex. */
    857 	if (tp != NULL)
    858 		mutex_exit(&tp->t_mutex);
    859 #endif
    860 
    861     done:
    862 	if (pg_id != NO_PGID)
    863 		pg_delete(pg_id);
    864 	rw_exit(&proclist_lock);
    865 	if (sess != NULL)
    866 		pool_put(&session_pool, sess);
    867 	if (new_pgrp != NULL)
    868 		pool_put(&pgrp_pool, new_pgrp);
    869 #ifdef DEBUG_PGRP
    870 	if (__predict_false(rval))
    871 		printf("enterpgrp(%d,%d,%d), curproc %d, rval %d\n",
    872 			pid, pgid, mksess, curp->p_pid, rval);
    873 #endif
    874 	return rval;
    875 }
    876 
    877 /*
    878  * Remove a process from its process group.  Must be called with the
    879  * proclist_lock write held.
    880  */
    881 void
    882 leavepgrp(struct proc *p)
    883 {
    884 	struct pgrp *pgrp;
    885 
    886 	LOCK_ASSERT(rw_write_held(&proclist_lock));
    887 
    888 	/*
    889 	 * If there's a controlling terminal for the session, we have to
    890 	 * interlock with it.  See ttread().
    891 	 */
    892 	mutex_enter(&proclist_mutex);
    893 #ifdef notyet
    894 	if (p_>p_session->s_ttyvp != NULL) {
    895 		tp = p->p_session->s_ttyp;
    896 		mutex_enter(&tp->t_mutex);
    897 	} else
    898 		tp = NULL;
    899 #endif
    900 
    901 	pgrp = p->p_pgrp;
    902 	LIST_REMOVE(p, p_pglist);
    903 	p->p_pgrp = NULL;
    904 
    905 #ifdef notyet
    906 	if (tp != NULL)
    907 		mutex_exit(&tp->t_mutex);
    908 #endif
    909 	mutex_exit(&proclist_mutex);
    910 
    911 	if (LIST_EMPTY(&pgrp->pg_members))
    912 		pg_delete(pgrp->pg_id);
    913 }
    914 
    915 /*
    916  * Free a process group.  Must be called with the proclist_lock write held.
    917  */
    918 static void
    919 pg_free(pid_t pg_id)
    920 {
    921 	struct pgrp *pgrp;
    922 	struct pid_table *pt;
    923 
    924 	LOCK_ASSERT(rw_write_held(&proclist_lock));
    925 
    926 	pt = &pid_table[pg_id & pid_tbl_mask];
    927 	pgrp = pt->pt_pgrp;
    928 #ifdef DIAGNOSTIC
    929 	if (__predict_false(!pgrp || pgrp->pg_id != pg_id
    930 	    || !LIST_EMPTY(&pgrp->pg_members)))
    931 		panic("pg_free: process group absent or has members");
    932 #endif
    933 	pt->pt_pgrp = 0;
    934 
    935 	if (!P_VALID(pt->pt_proc)) {
    936 		/* orphaned pgrp, put slot onto free list */
    937 #ifdef DIAGNOSTIC
    938 		if (__predict_false(P_NEXT(pt->pt_proc) & pid_tbl_mask))
    939 			panic("pg_free: process slot on free list");
    940 #endif
    941 		mutex_enter(&proclist_mutex);
    942 		pg_id &= pid_tbl_mask;
    943 		pt = &pid_table[last_free_pt];
    944 		pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pg_id);
    945 		mutex_exit(&proclist_mutex);
    946 		last_free_pt = pg_id;
    947 		pid_alloc_cnt--;
    948 	}
    949 	pool_put(&pgrp_pool, pgrp);
    950 }
    951 
    952 /*
    953  * Delete a process group.  Must be called with the proclist_lock write
    954  * held.
    955  */
    956 static void
    957 pg_delete(pid_t pg_id)
    958 {
    959 	struct pgrp *pgrp;
    960 	struct tty *ttyp;
    961 	struct session *ss;
    962 	int is_pgrp_leader;
    963 
    964 	LOCK_ASSERT(rw_write_held(&proclist_lock));
    965 
    966 	pgrp = pid_table[pg_id & pid_tbl_mask].pt_pgrp;
    967 	if (pgrp == NULL || pgrp->pg_id != pg_id ||
    968 	    !LIST_EMPTY(&pgrp->pg_members))
    969 		return;
    970 
    971 	ss = pgrp->pg_session;
    972 
    973 	/* Remove reference (if any) from tty to this process group */
    974 	ttyp = ss->s_ttyp;
    975 	if (ttyp != NULL && ttyp->t_pgrp == pgrp) {
    976 		ttyp->t_pgrp = NULL;
    977 #ifdef DIAGNOSTIC
    978 		if (ttyp->t_session != ss)
    979 			panic("pg_delete: wrong session on terminal");
    980 #endif
    981 	}
    982 
    983 	/*
    984 	 * The leading process group in a session is freed
    985 	 * by sessdelete() if last reference.
    986 	 */
    987 	is_pgrp_leader = (ss->s_sid == pgrp->pg_id);
    988 	SESSRELE(ss);
    989 
    990 	if (is_pgrp_leader)
    991 		return;
    992 
    993 	pg_free(pg_id);
    994 }
    995 
    996 /*
    997  * Delete session - called from SESSRELE when s_count becomes zero.
    998  * Must be called with the proclist_lock write held.
    999  */
   1000 void
   1001 sessdelete(struct session *ss)
   1002 {
   1003 
   1004 	LOCK_ASSERT(rw_write_held(&proclist_lock));
   1005 
   1006 	/*
   1007 	 * We keep the pgrp with the same id as the session in
   1008 	 * order to stop a process being given the same pid.
   1009 	 * Since the pgrp holds a reference to the session, it
   1010 	 * must be a 'zombie' pgrp by now.
   1011 	 */
   1012 	pg_free(ss->s_sid);
   1013 	pool_put(&session_pool, ss);
   1014 }
   1015 
   1016 /*
   1017  * Adjust pgrp jobc counters when specified process changes process group.
   1018  * We count the number of processes in each process group that "qualify"
   1019  * the group for terminal job control (those with a parent in a different
   1020  * process group of the same session).  If that count reaches zero, the
   1021  * process group becomes orphaned.  Check both the specified process'
   1022  * process group and that of its children.
   1023  * entering == 0 => p is leaving specified group.
   1024  * entering == 1 => p is entering specified group.
   1025  *
   1026  * Call with proclist_lock write held.
   1027  */
   1028 void
   1029 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
   1030 {
   1031 	struct pgrp *hispgrp;
   1032 	struct session *mysession = pgrp->pg_session;
   1033 	struct proc *child;
   1034 
   1035 	LOCK_ASSERT(rw_write_held(&proclist_lock));
   1036 	LOCK_ASSERT(mutex_owned(&proclist_mutex));
   1037 
   1038 	/*
   1039 	 * Check p's parent to see whether p qualifies its own process
   1040 	 * group; if so, adjust count for p's process group.
   1041 	 */
   1042 	hispgrp = p->p_pptr->p_pgrp;
   1043 	if (hispgrp != pgrp && hispgrp->pg_session == mysession) {
   1044 		if (entering) {
   1045 			mutex_enter(&p->p_smutex);
   1046 			p->p_sflag &= ~PS_ORPHANPG;
   1047 			mutex_exit(&p->p_smutex);
   1048 			pgrp->pg_jobc++;
   1049 		} else if (--pgrp->pg_jobc == 0)
   1050 			orphanpg(pgrp);
   1051 	}
   1052 
   1053 	/*
   1054 	 * Check this process' children to see whether they qualify
   1055 	 * their process groups; if so, adjust counts for children's
   1056 	 * process groups.
   1057 	 */
   1058 	LIST_FOREACH(child, &p->p_children, p_sibling) {
   1059 		hispgrp = child->p_pgrp;
   1060 		if (hispgrp != pgrp && hispgrp->pg_session == mysession &&
   1061 		    !P_ZOMBIE(child)) {
   1062 			if (entering) {
   1063 				mutex_enter(&child->p_smutex);
   1064 				child->p_sflag &= ~PS_ORPHANPG;
   1065 				mutex_exit(&child->p_smutex);
   1066 				hispgrp->pg_jobc++;
   1067 			} else if (--hispgrp->pg_jobc == 0)
   1068 				orphanpg(hispgrp);
   1069 		}
   1070 	}
   1071 }
   1072 
   1073 /*
   1074  * A process group has become orphaned;
   1075  * if there are any stopped processes in the group,
   1076  * hang-up all process in that group.
   1077  *
   1078  * Call with proclist_lock write held.
   1079  */
   1080 static void
   1081 orphanpg(struct pgrp *pg)
   1082 {
   1083 	struct proc *p;
   1084 	int doit;
   1085 
   1086 	LOCK_ASSERT(rw_write_held(&proclist_lock));
   1087 	LOCK_ASSERT(mutex_owned(&proclist_mutex));
   1088 
   1089 	doit = 0;
   1090 
   1091 	LIST_FOREACH(p, &pg->pg_members, p_pglist) {
   1092 		mutex_enter(&p->p_smutex);
   1093 		if (p->p_stat == SSTOP) {
   1094 			doit = 1;
   1095 			p->p_sflag |= PS_ORPHANPG;
   1096 		}
   1097 		mutex_exit(&p->p_smutex);
   1098 	}
   1099 
   1100 	if (doit) {
   1101 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
   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 			    pgrp->pg_members.lh_first);
   1142 			for (p = pgrp->pg_members.lh_first; p != 0;
   1143 			    p = p->p_pglist.le_next) {
   1144 				db_printf("\t\tpid %d addr %p pgrp %p %s\n",
   1145 				    p->p_pid, p, p->p_pgrp, p->p_comm);
   1146 			}
   1147 		}
   1148 	}
   1149 }
   1150 #endif /* DDB */
   1151 
   1152 #ifdef KSTACK_CHECK_MAGIC
   1153 #include <sys/user.h>
   1154 
   1155 #define	KSTACK_MAGIC	0xdeadbeaf
   1156 
   1157 /* XXX should be per process basis? */
   1158 int kstackleftmin = KSTACK_SIZE;
   1159 int kstackleftthres = KSTACK_SIZE / 8; /* warn if remaining stack is
   1160 					  less than this */
   1161 
   1162 void
   1163 kstack_setup_magic(const struct lwp *l)
   1164 {
   1165 	uint32_t *ip;
   1166 	uint32_t const *end;
   1167 
   1168 	KASSERT(l != NULL);
   1169 	KASSERT(l != &lwp0);
   1170 
   1171 	/*
   1172 	 * fill all the stack with magic number
   1173 	 * so that later modification on it can be detected.
   1174 	 */
   1175 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
   1176 	end = (uint32_t *)((caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
   1177 	for (; ip < end; ip++) {
   1178 		*ip = KSTACK_MAGIC;
   1179 	}
   1180 }
   1181 
   1182 void
   1183 kstack_check_magic(const struct lwp *l)
   1184 {
   1185 	uint32_t const *ip, *end;
   1186 	int stackleft;
   1187 
   1188 	KASSERT(l != NULL);
   1189 
   1190 	/* don't check proc0 */ /*XXX*/
   1191 	if (l == &lwp0)
   1192 		return;
   1193 
   1194 #ifdef __MACHINE_STACK_GROWS_UP
   1195 	/* stack grows upwards (eg. hppa) */
   1196 	ip = (uint32_t *)((caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
   1197 	end = (uint32_t *)KSTACK_LOWEST_ADDR(l);
   1198 	for (ip--; ip >= end; ip--)
   1199 		if (*ip != KSTACK_MAGIC)
   1200 			break;
   1201 
   1202 	stackleft = (caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE - (caddr_t)ip;
   1203 #else /* __MACHINE_STACK_GROWS_UP */
   1204 	/* stack grows downwards (eg. i386) */
   1205 	ip = (uint32_t *)KSTACK_LOWEST_ADDR(l);
   1206 	end = (uint32_t *)((caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE);
   1207 	for (; ip < end; ip++)
   1208 		if (*ip != KSTACK_MAGIC)
   1209 			break;
   1210 
   1211 	stackleft = ((const char *)ip) - (const char *)KSTACK_LOWEST_ADDR(l);
   1212 #endif /* __MACHINE_STACK_GROWS_UP */
   1213 
   1214 	if (kstackleftmin > stackleft) {
   1215 		kstackleftmin = stackleft;
   1216 		if (stackleft < kstackleftthres)
   1217 			printf("warning: kernel stack left %d bytes"
   1218 			    "(pid %u:lid %u)\n", stackleft,
   1219 			    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
   1220 	}
   1221 
   1222 	if (stackleft <= 0) {
   1223 		panic("magic on the top of kernel stack changed for "
   1224 		    "pid %u, lid %u: maybe kernel stack overflow",
   1225 		    (u_int)l->l_proc->p_pid, (u_int)l->l_lid);
   1226 	}
   1227 }
   1228 #endif /* KSTACK_CHECK_MAGIC */
   1229 
   1230 /*
   1231  * XXXSMP this is bust, it grabs a read lock and then messes about
   1232  * with allproc.
   1233  */
   1234 int
   1235 proclist_foreach_call(struct proclist *list,
   1236     int (*callback)(struct proc *, void *arg), void *arg)
   1237 {
   1238 	struct proc marker;
   1239 	struct proc *p;
   1240 	struct lwp * const l = curlwp;
   1241 	int ret = 0;
   1242 
   1243 	marker.p_flag = P_MARKER;
   1244 	PHOLD(l);
   1245 	rw_enter(&proclist_lock, RW_READER);
   1246 	for (p = LIST_FIRST(list); ret == 0 && p != NULL;) {
   1247 		if (p->p_flag & P_MARKER) {
   1248 			p = LIST_NEXT(p, p_list);
   1249 			continue;
   1250 		}
   1251 		LIST_INSERT_AFTER(p, &marker, p_list);
   1252 		ret = (*callback)(p, arg);
   1253 		KASSERT(rw_read_held(&proclist_lock));
   1254 		p = LIST_NEXT(&marker, p_list);
   1255 		LIST_REMOVE(&marker, p_list);
   1256 	}
   1257 	rw_exit(&proclist_lock);
   1258 	PRELE(l);
   1259 
   1260 	return ret;
   1261 }
   1262 
   1263 int
   1264 proc_vmspace_getref(struct proc *p, struct vmspace **vm)
   1265 {
   1266 
   1267 	/* XXXCDC: how should locking work here? */
   1268 
   1269 	/* curproc exception is for coredump. */
   1270 
   1271 	if ((p != curproc && (p->p_sflag & PS_WEXIT) != 0) ||
   1272 	    (p->p_vmspace->vm_refcnt < 1)) { /* XXX */
   1273 		return EFAULT;
   1274 	}
   1275 
   1276 	uvmspace_addref(p->p_vmspace);
   1277 	*vm = p->p_vmspace;
   1278 
   1279 	return 0;
   1280 }
   1281 
   1282 /*
   1283  * Acquire a write lock on the process credential.
   1284  */
   1285 void
   1286 proc_crmod_enter(void)
   1287 {
   1288 	struct lwp *l = curlwp;
   1289 	struct proc *p = l->l_proc;
   1290 	struct plimit *lim;
   1291 	kauth_cred_t oc;
   1292 	char *cn;
   1293 
   1294 	mutex_enter(&p->p_mutex);
   1295 
   1296 	/* Ensure the LWP cached credentials are up to date. */
   1297 	if ((oc = l->l_cred) != p->p_cred) {
   1298 		kauth_cred_hold(p->p_cred);
   1299 		l->l_cred = p->p_cred;
   1300 		kauth_cred_free(oc);
   1301 	}
   1302 
   1303 	/* Reset what needs to be reset in plimit. */
   1304 	lim = p->p_limit;
   1305 	if (lim->pl_corename != defcorename) {
   1306 		if (lim->p_refcnt > 1 &&
   1307 		    (lim->p_lflags & PL_SHAREMOD) == 0) {
   1308 			p->p_limit = limcopy(p);
   1309 			limfree(lim);
   1310 			lim = p->p_limit;
   1311 		}
   1312 		simple_lock(&lim->p_slock);
   1313 		cn = lim->pl_corename;
   1314 		lim->pl_corename = defcorename;
   1315 		simple_unlock(&lim->p_slock);
   1316 		if (cn != defcorename)
   1317 			free(cn, M_TEMP);
   1318 	}
   1319 }
   1320 
   1321 /*
   1322  * Set in a new process credential, and drop the write lock.  The credential
   1323  * must have a reference already.  Optionally, free a no-longer required
   1324  * credential.  The scheduler also needs to inspect p_cred, so we also
   1325  * briefly acquire the sched state mutex.
   1326  */
   1327 void
   1328 proc_crmod_leave(kauth_cred_t scred, kauth_cred_t fcred, boolean_t sugid)
   1329 {
   1330 	struct lwp *l = curlwp;
   1331 	struct proc *p = l->l_proc;
   1332 	kauth_cred_t oc;
   1333 
   1334 	/* Is there a new credential to set in? */
   1335 	if (scred != NULL) {
   1336 		mutex_enter(&p->p_smutex);
   1337 		p->p_cred = scred;
   1338 		mutex_exit(&p->p_smutex);
   1339 
   1340 		/* Ensure the LWP cached credentials are up to date. */
   1341 		if ((oc = l->l_cred) != scred) {
   1342 			kauth_cred_hold(scred);
   1343 			l->l_cred = scred;
   1344 		}
   1345 	} else
   1346 		oc = NULL;	/* XXXgcc */
   1347 
   1348 	if (sugid) {
   1349 		/*
   1350 		 * Mark process as having changed credentials, stops
   1351 		 * tracing etc.
   1352 		 */
   1353 		p->p_flag |= P_SUGID;
   1354 	}
   1355 
   1356 	mutex_exit(&p->p_mutex);
   1357 
   1358 	/* If there is a credential to be released, free it now. */
   1359 	if (fcred != NULL) {
   1360 		KASSERT(scred != NULL);
   1361 		kauth_cred_free(fcred);
   1362 		if (oc != scred)
   1363 			kauth_cred_free(oc);
   1364 	}
   1365 }
   1366 
   1367 /*
   1368  * Acquire a reference on a process, to prevent it from exiting or execing.
   1369  */
   1370 int
   1371 proc_addref(struct proc *p)
   1372 {
   1373 
   1374 	LOCK_ASSERT(mutex_owned(&p->p_mutex));
   1375 
   1376 	if (p->p_refcnt <= 0)
   1377 		return EAGAIN;
   1378 	p->p_refcnt++;
   1379 
   1380 	return 0;
   1381 }
   1382 
   1383 /*
   1384  * Release a reference on a process.
   1385  */
   1386 void
   1387 proc_delref(struct proc *p)
   1388 {
   1389 
   1390 	LOCK_ASSERT(mutex_owned(&p->p_mutex));
   1391 
   1392 	if (p->p_refcnt < 0) {
   1393 		if (++p->p_refcnt == 0)
   1394 			cv_broadcast(&p->p_refcv);
   1395 	} else {
   1396 		p->p_refcnt--;
   1397 		KASSERT(p->p_refcnt != 0);
   1398 	}
   1399 }
   1400 
   1401 /*
   1402  * Wait for all references on the process to drain, and prevent new
   1403  * references from being acquired.
   1404  */
   1405 void
   1406 proc_drainrefs(struct proc *p)
   1407 {
   1408 
   1409 	LOCK_ASSERT(mutex_owned(&p->p_mutex));
   1410 	KASSERT(p->p_refcnt > 0);
   1411 
   1412 	/*
   1413 	 * The process itself holds the last reference.  Once it's released,
   1414 	 * no new references will be granted.  If we have already locked out
   1415 	 * new references (refcnt <= 0), potentially due to a failed exec,
   1416 	 * there is nothing more to do.
   1417 	 */
   1418 	p->p_refcnt = 1 - p->p_refcnt;
   1419 	while (p->p_refcnt != 0)
   1420 		cv_wait(&p->p_refcv, &p->p_mutex);
   1421 }
   1422 
   1423 /*
   1424  * proc_specific_key_create --
   1425  *	Create a key for subsystem proc-specific data.
   1426  */
   1427 int
   1428 proc_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
   1429 {
   1430 
   1431 	return (specificdata_key_create(proc_specificdata_domain, keyp, dtor));
   1432 }
   1433 
   1434 /*
   1435  * proc_specific_key_delete --
   1436  *	Delete a key for subsystem proc-specific data.
   1437  */
   1438 void
   1439 proc_specific_key_delete(specificdata_key_t key)
   1440 {
   1441 
   1442 	specificdata_key_delete(proc_specificdata_domain, key);
   1443 }
   1444 
   1445 /*
   1446  * proc_initspecific --
   1447  *	Initialize a proc's specificdata container.
   1448  */
   1449 void
   1450 proc_initspecific(struct proc *p)
   1451 {
   1452 	int error;
   1453 
   1454 	error = specificdata_init(proc_specificdata_domain, &p->p_specdataref);
   1455 	KASSERT(error == 0);
   1456 }
   1457 
   1458 /*
   1459  * proc_finispecific --
   1460  *	Finalize a proc's specificdata container.
   1461  */
   1462 void
   1463 proc_finispecific(struct proc *p)
   1464 {
   1465 
   1466 	specificdata_fini(proc_specificdata_domain, &p->p_specdataref);
   1467 }
   1468 
   1469 /*
   1470  * proc_getspecific --
   1471  *	Return proc-specific data corresponding to the specified key.
   1472  */
   1473 void *
   1474 proc_getspecific(struct proc *p, specificdata_key_t key)
   1475 {
   1476 
   1477 	return (specificdata_getspecific(proc_specificdata_domain,
   1478 					 &p->p_specdataref, key));
   1479 }
   1480 
   1481 /*
   1482  * proc_setspecific --
   1483  *	Set proc-specific data corresponding to the specified key.
   1484  */
   1485 void
   1486 proc_setspecific(struct proc *p, specificdata_key_t key, void *data)
   1487 {
   1488 
   1489 	specificdata_setspecific(proc_specificdata_domain,
   1490 				 &p->p_specdataref, key, data);
   1491 }
   1492