xref: /netbsd-src/sys/rump/librump/rumpkern/rump.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d) 
1 /*	$NetBSD: rump.c,v 1.221 2011/01/22 18:33:25 pooka Exp $	*/
2 
3 /*
4  * Copyright (c) 2007 Antti Kantee.  All Rights Reserved.
5  *
6  * Development of this software was supported by Google Summer of Code.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
18  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
19  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
20  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
23  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.221 2011/01/22 18:33:25 pooka Exp $");
32 
33 #include <sys/systm.h>
34 #define ELFSIZE ARCH_ELFSIZE
35 
36 #include <sys/param.h>
37 #include <sys/atomic.h>
38 #include <sys/buf.h>
39 #include <sys/callout.h>
40 #include <sys/conf.h>
41 #include <sys/cpu.h>
42 #include <sys/device.h>
43 #include <sys/evcnt.h>
44 #include <sys/event.h>
45 #include <sys/exec_elf.h>
46 #include <sys/filedesc.h>
47 #include <sys/iostat.h>
48 #include <sys/kauth.h>
49 #include <sys/kernel.h>
50 #include <sys/kmem.h>
51 #include <sys/kprintf.h>
52 #include <sys/kthread.h>
53 #include <sys/ksyms.h>
54 #include <sys/msgbuf.h>
55 #include <sys/module.h>
56 #include <sys/once.h>
57 #include <sys/percpu.h>
58 #include <sys/pipe.h>
59 #include <sys/pool.h>
60 #include <sys/queue.h>
61 #include <sys/reboot.h>
62 #include <sys/resourcevar.h>
63 #include <sys/select.h>
64 #include <sys/sysctl.h>
65 #include <sys/syscall.h>
66 #include <sys/syscallvar.h>
67 #include <sys/timetc.h>
68 #include <sys/tty.h>
69 #include <sys/uidinfo.h>
70 #include <sys/vmem.h>
71 #include <sys/xcall.h>
72 #include <sys/simplelock.h>
73 
74 #include <rump/rumpuser.h>
75 
76 #include <secmodel/suser/suser.h>
77 
78 #include <prop/proplib.h>
79 
80 #include <uvm/uvm_extern.h>
81 #include <uvm/uvm_readahead.h>
82 
83 #include "rump_private.h"
84 #include "rump_net_private.h"
85 #include "rump_vfs_private.h"
86 #include "rump_dev_private.h"
87 
88 char machine[] = MACHINE;
89 
90 struct proc *initproc;
91 
92 struct device rump_rootdev = {
93 	.dv_class = DV_VIRTUAL
94 };
95 
96 #ifdef RUMP_WITHOUT_THREADS
97 int rump_threads = 0;
98 #else
99 int rump_threads = 1;
100 #endif
101 
102 static int rump_proxy_syscall(int, void *, register_t *);
103 static int rump_proxy_rfork(void *, int);
104 static void rump_proxy_procexit(void);
105 
106 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */
107 
108 #ifdef LOCKDEBUG
109 const int rump_lockdebug = 1;
110 #else
111 const int rump_lockdebug = 0;
112 #endif
113 bool rump_ttycomponent = false;
114 
115 static void
116 rump_aiodone_worker(struct work *wk, void *dummy)
117 {
118 	struct buf *bp = (struct buf *)wk;
119 
120 	KASSERT(&bp->b_work == wk);
121 	bp->b_iodone(bp);
122 }
123 
124 static int rump_inited;
125 
126 /*
127  * Make sure pnbuf_cache is available even without vfs
128  */
129 struct pool_cache *pnbuf_cache;
130 int rump_initpnbufpool(void);
131 int rump_initpnbufpool(void)
132 {
133 
134         pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
135 	    NULL, IPL_NONE, NULL, NULL, NULL);
136 	return EOPNOTSUPP;
137 }
138 
139 int rump__unavailable(void);
140 int rump__unavailable() {return EOPNOTSUPP;}
141 __weak_alias(rump_net_init,rump__unavailable);
142 __weak_alias(rump_vfs_init,rump_initpnbufpool);
143 __weak_alias(rump_dev_init,rump__unavailable);
144 
145 __weak_alias(rump_vfs_fini,rump__unavailable);
146 
147 __weak_alias(biodone,rump__unavailable);
148 __weak_alias(sopoll,rump__unavailable);
149 
150 __weak_alias(rump_vfs_drainbufs,rump__unavailable);
151 
152 void rump__unavailable_vfs_panic(void);
153 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
154 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
155 
156 /* easier to write vfs-less clients */
157 __weak_alias(rump_pub_etfs_register,rump__unavailable);
158 __weak_alias(rump_pub_etfs_remove,rump__unavailable);
159 
160 rump_proc_vfs_init_fn rump_proc_vfs_init;
161 rump_proc_vfs_release_fn rump_proc_vfs_release;
162 
163 static void add_linkedin_modules(const struct modinfo *const *, size_t);
164 
165 static void __noinline
166 messthestack(void)
167 {
168 	volatile uint32_t mess[64];
169 	uint64_t d1, d2;
170 	int i, error;
171 
172 	for (i = 0; i < 64; i++) {
173 		rumpuser_gettime(&d1, &d2, &error);
174 		mess[i] = d2;
175 	}
176 }
177 
178 /*
179  * Create kern.hostname.  why only this you ask.  well, init_sysctl
180  * is a kitchen sink in need of some gardening.  but i want to use
181  * kern.hostname today.
182  */
183 static void
184 mksysctls(void)
185 {
186 
187 	sysctl_createv(NULL, 0, NULL, NULL,
188 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "kern", NULL,
189 	    NULL, 0, NULL, 0, CTL_KERN, CTL_EOL);
190 
191 	/* XXX: setting hostnamelen is missing */
192 	sysctl_createv(NULL, 0, NULL, NULL,
193 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_STRING, "hostname",
194 	    SYSCTL_DESCR("System hostname"), NULL, 0,
195 	    &hostname, MAXHOSTNAMELEN, CTL_KERN, KERN_HOSTNAME, CTL_EOL);
196 }
197 
198 /* there's no convenient kernel entry point for this, so just craft out own */
199 static pid_t
200 spgetpid(void)
201 {
202 
203 	return curproc->p_pid;
204 }
205 
206 static const struct rumpuser_sp_ops spops = {
207 	.spop_schedule		= rump_schedule,
208 	.spop_unschedule	= rump_unschedule,
209 	.spop_lwproc_switch	= rump_lwproc_switch,
210 	.spop_lwproc_release	= rump_lwproc_releaselwp,
211 	.spop_lwproc_rfork	= rump_proxy_rfork,
212 	.spop_lwproc_newlwp	= rump_lwproc_newlwp,
213 	.spop_lwproc_curlwp	= rump_lwproc_curlwp,
214 	.spop_procexit		= rump_proxy_procexit,
215 	.spop_syscall		= rump_proxy_syscall,
216 	.spop_getpid		= spgetpid,
217 };
218 
219 int
220 rump_daemonize_begin(void)
221 {
222 
223 	if (rump_inited)
224 		return EALREADY;
225 
226 	return rumpuser_daemonize_begin();
227 }
228 
229 int
230 rump_daemonize_done(int error)
231 {
232 
233 	return rumpuser_daemonize_done(error);
234 }
235 
236 int
237 rump__init(int rump_version)
238 {
239 	char buf[256];
240 	struct timespec ts;
241 	uint64_t sec, nsec;
242 	struct lwp *l;
243 	int i, numcpu;
244 	int error;
245 
246 	/* not reentrant */
247 	if (rump_inited)
248 		return 0;
249 	else if (rump_inited == -1)
250 		panic("rump_init: host process restart required");
251 	else
252 		rump_inited = 1;
253 
254 	if (rumpuser_getversion() != RUMPUSER_VERSION) {
255 		/* let's hope the ABI of rumpuser_dprintf is the same ;) */
256 		rumpuser_dprintf("rumpuser version mismatch: %d vs. %d\n",
257 		    rumpuser_getversion(), RUMPUSER_VERSION);
258 		return EPROGMISMATCH;
259 	}
260 
261 	if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) {
262 		if (*buf != '0')
263 			boothowto = AB_VERBOSE;
264 	}
265 
266 	if (rumpuser_getenv("RUMP_NCPU", buf, sizeof(buf), &error) == 0)
267 		error = 0;
268 	if (error == 0) {
269 		numcpu = strtoll(buf, NULL, 10);
270 		if (numcpu < 1)
271 			numcpu = 1;
272 	} else {
273 		numcpu = rumpuser_getnhostcpu();
274 	}
275 	rump_cpus_bootstrap(&numcpu);
276 
277 	rumpuser_gettime(&sec, &nsec, &error);
278 	boottime.tv_sec = sec;
279 	boottime.tv_nsec = nsec;
280 
281 	initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
282 	aprint_verbose("%s%s", copyright, version);
283 
284 	/*
285 	 * Seed arc4random() with a "reasonable" amount of randomness.
286 	 * Yes, this is a quick kludge which depends on the arc4random
287 	 * implementation.
288 	 */
289 	messthestack();
290 	arc4random();
291 
292 	if (rump_version != RUMP_VERSION) {
293 		printf("rump version mismatch, %d vs. %d\n",
294 		    rump_version, RUMP_VERSION);
295 		return EPROGMISMATCH;
296 	}
297 
298 	if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
299 		rump_threads = *buf != '0';
300 	}
301 	rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
302 	    rump_threads);
303 	rump_intr_init(numcpu);
304 	rump_tsleep_init();
305 
306 	/* init minimal lwp/cpu context */
307 	l = &lwp0;
308 	l->l_lid = 1;
309 	l->l_cpu = l->l_target_cpu = rump_cpu;
310 	l->l_fd = &filedesc0;
311 	rumpuser_set_curlwp(l);
312 
313 	rumpuser_mutex_init(&rump_giantlock);
314 	ksyms_init();
315 	uvm_init();
316 	evcnt_init();
317 
318 	once_init();
319 	kernconfig_lock_init();
320 	prop_kern_init();
321 
322 	pool_subsystem_init();
323 	kmem_init();
324 
325 	uvm_ra_init();
326 	uao_init();
327 
328 	mutex_obj_init();
329 	callout_startup();
330 
331 	kprintf_init();
332 	loginit();
333 
334 	kauth_init();
335 
336 	procinit();
337 	proc0_init();
338 	uid_init();
339 	chgproccnt(0, 1);
340 
341 	l->l_proc = &proc0;
342 	lwp_update_creds(l);
343 
344 	lwpinit_specificdata();
345 	lwp_initspecific(&lwp0);
346 
347 	rump_scheduler_init(numcpu);
348 	/* revert temporary context and schedule a semireal context */
349 	l->l_cpu = NULL;
350 	rumpuser_set_curlwp(NULL);
351 	initproc = &proc0; /* borrow proc0 before we get initproc started */
352 	rump_schedule();
353 
354 	percpu_init();
355 	inittimecounter();
356 	ntp_init();
357 
358 	rumpuser_gettime(&sec, &nsec, &error);
359 	ts.tv_sec = sec;
360 	ts.tv_nsec = nsec;
361 	tc_setclock(&ts);
362 
363 	/* we are mostly go.  do per-cpu subsystem init */
364 	for (i = 0; i < numcpu; i++) {
365 		struct cpu_info *ci = cpu_lookup(i);
366 
367 		/* attach non-bootstrap CPUs */
368 		if (i > 0) {
369 			rump_cpu_attach(ci);
370 			ncpu++;
371 		}
372 
373 		callout_init_cpu(ci);
374 		softint_init(ci);
375 		xc_init_cpu(ci);
376 		pool_cache_cpu_init(ci);
377 		selsysinit(ci);
378 		percpu_init_cpu(ci);
379 
380 		TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
381 		__cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
382 
383 		aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
384 	}
385 
386 	sysctl_init();
387 	mksysctls();
388 	kqueue_init();
389 	iostat_init();
390 	fd_sys_init();
391 	module_init();
392 	devsw_init();
393 	pipe_init();
394 	resource_init();
395 
396 	/* start page baroness */
397 	if (rump_threads) {
398 		if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
399 		    uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
400 			panic("pagedaemon create failed");
401 	} else
402 		uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
403 
404 	/* process dso's */
405 	rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load);
406 
407 	rump_component_init(RUMP_COMPONENT_KERN);
408 
409 	/* these do nothing if not present */
410 	rump_vfs_init();
411 	rump_net_init();
412 	rump_dev_init();
413 
414 	rump_component_init(RUMP_COMPONENT_KERN_VFS);
415 
416 	/*
417 	 * if we initialized the tty component above, the tyttymtx is
418 	 * now initialized.  otherwise, we need to initialize it.
419 	 */
420 	if (!rump_ttycomponent)
421 		mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
422 
423 	cold = 0;
424 
425 	/* aieeeedondest */
426 	if (rump_threads) {
427 		if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
428 		    rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
429 			panic("aiodoned");
430 	}
431 
432 	sysctl_finalize();
433 
434 	module_init_class(MODULE_CLASS_ANY);
435 
436 	rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
437 	hostnamelen = strlen(hostname);
438 
439 	sigemptyset(&sigcantmask);
440 
441 	if (rump_threads)
442 		vmem_rehash_start();
443 
444 	/*
445 	 * Create init, used to attach implicit threads in rump.
446 	 * (note: must be done after vfsinit to get cwdi)
447 	 */
448 	(void)rump__lwproc_alloclwp(NULL); /* dummy thread for initproc */
449 	mutex_enter(proc_lock);
450 	initproc = proc_find_raw(1);
451 	mutex_exit(proc_lock);
452 	if (initproc == NULL)
453 		panic("where in the world is initproc?");
454 
455 	/*
456 	 * Adjust syscall vector in case factions were dlopen()'d
457 	 * before calling rump_init().
458 	 * (modules will handle dynamic syscalls the usual way)
459 	 *
460 	 * Note: this will adjust the function vectors of
461 	 * syscalls which use a funcalias (getpid etc.), but
462 	 * it makes no difference.
463 	 */
464 	for (i = 0; i < SYS_NSYSENT; i++) {
465 		void *sym;
466 
467 		if (rump_sysent[i].sy_flags & SYCALL_NOSYS ||
468 		    *syscallnames[i] == '#' ||
469 		    rump_sysent[i].sy_call == sys_nomodule)
470 			continue;
471 
472 		/* if present, adjust symbol value */
473 		sprintf(buf, "rumpns_sys_%s", syscallnames[i]);
474 		if ((sym = rumpuser_dl_globalsym(buf)) != NULL
475 		    && sym != rump_sysent[i].sy_call) {
476 #if 0
477 			rumpuser_dprintf("adjusting %s: %p (old %p)\n",
478 			    syscallnames[i], sym, rump_sysent[i].sy_call);
479 #endif
480 			rump_sysent[i].sy_call = sym;
481 		}
482 	}
483 
484 	/* release cpu */
485 	rump_unschedule();
486 
487 	return 0;
488 }
489 
490 int
491 rump_init_server(const char *url)
492 {
493 
494 	return rumpuser_sp_init(url, &spops, ostype, osrelease, MACHINE);
495 }
496 
497 void
498 cpu_reboot(int howto, char *bootstr)
499 {
500 	int ruhow = 0;
501 	void *finiarg;
502 
503 	printf("rump kernel halting...\n");
504 
505 	if (!RUMP_LOCALPROC_P(curproc))
506 		finiarg = curproc->p_vmspace->vm_map.pmap;
507 	else
508 		finiarg = NULL;
509 	rumpuser_sp_fini(finiarg);
510 
511 	/* dump means we really take the dive here */
512 	if ((howto & RB_DUMP) || panicstr) {
513 		ruhow = RUMPUSER_PANIC;
514 		goto out;
515 	}
516 
517 	/* try to sync */
518 	if (!((howto & RB_NOSYNC) || panicstr)) {
519 		rump_vfs_fini();
520 	}
521 
522 	/* your wish is my command */
523 	if (howto & RB_HALT) {
524 		printf("rump kernel halted\n");
525 		for (;;) {
526 			uint64_t sec = 5, nsec = 0;
527 			int error;
528 
529 			rumpuser_nanosleep(&sec, &nsec, &error);
530 		}
531 	}
532 
533 	/* this function is __dead, we must exit */
534  out:
535 	printf("halted\n");
536 	rumpuser_exit(ruhow);
537 }
538 
539 struct uio *
540 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
541 {
542 	struct uio *uio;
543 	enum uio_rw uiorw;
544 
545 	switch (rw) {
546 	case RUMPUIO_READ:
547 		uiorw = UIO_READ;
548 		break;
549 	case RUMPUIO_WRITE:
550 		uiorw = UIO_WRITE;
551 		break;
552 	default:
553 		panic("%s: invalid rw %d", __func__, rw);
554 	}
555 
556 	uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
557 	uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
558 
559 	uio->uio_iov->iov_base = buf;
560 	uio->uio_iov->iov_len = bufsize;
561 
562 	uio->uio_iovcnt = 1;
563 	uio->uio_offset = offset;
564 	uio->uio_resid = bufsize;
565 	uio->uio_rw = uiorw;
566 	UIO_SETUP_SYSSPACE(uio);
567 
568 	return uio;
569 }
570 
571 size_t
572 rump_uio_getresid(struct uio *uio)
573 {
574 
575 	return uio->uio_resid;
576 }
577 
578 off_t
579 rump_uio_getoff(struct uio *uio)
580 {
581 
582 	return uio->uio_offset;
583 }
584 
585 size_t
586 rump_uio_free(struct uio *uio)
587 {
588 	size_t resid;
589 
590 	resid = uio->uio_resid;
591 	kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
592 	kmem_free(uio, sizeof(*uio));
593 
594 	return resid;
595 }
596 
597 kauth_cred_t
598 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
599 {
600 	kauth_cred_t cred;
601 	int rv;
602 
603 	cred = kauth_cred_alloc();
604 	kauth_cred_setuid(cred, uid);
605 	kauth_cred_seteuid(cred, uid);
606 	kauth_cred_setsvuid(cred, uid);
607 	kauth_cred_setgid(cred, gid);
608 	kauth_cred_setgid(cred, gid);
609 	kauth_cred_setegid(cred, gid);
610 	kauth_cred_setsvgid(cred, gid);
611 	rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
612 	/* oh this is silly.  and by "this" I mean kauth_cred_setgroups() */
613 	assert(rv == 0);
614 
615 	return cred;
616 }
617 
618 void
619 rump_cred_put(kauth_cred_t cred)
620 {
621 
622 	kauth_cred_free(cred);
623 }
624 
625 static int compcounter[RUMP_COMPONENT_MAX];
626 
627 static void
628 rump_component_init_cb(struct rump_component *rc, int type)
629 {
630 
631 	KASSERT(type < RUMP_COMPONENT_MAX);
632 	if (rc->rc_type == type) {
633 		rc->rc_init();
634 		compcounter[type]++;
635 	}
636 }
637 
638 int
639 rump_component_count(enum rump_component_type type)
640 {
641 
642 	KASSERT(type <= RUMP_COMPONENT_MAX);
643 	return compcounter[type];
644 }
645 
646 void
647 rump_component_init(enum rump_component_type type)
648 {
649 
650 	rumpuser_dl_component_init(type, rump_component_init_cb);
651 }
652 
653 /*
654  * Initialize a module which has already been loaded and linked
655  * with dlopen(). This is fundamentally the same as a builtin module.
656  */
657 int
658 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
659 {
660 
661 	return module_builtin_add(mip, nmodinfo, true);
662 }
663 
664 /*
665  * Finish module (flawless victory, fatality!).
666  */
667 int
668 rump_module_fini(const struct modinfo *mi)
669 {
670 
671 	return module_builtin_remove(mi, true);
672 }
673 
674 /*
675  * Add loaded and linked module to the builtin list.  It will
676  * later be initialized with module_init_class().
677  */
678 
679 static void
680 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
681 {
682 
683 	module_builtin_add(mip, nmodinfo, false);
684 }
685 
686 int
687 rump_kernelfsym_load(void *symtab, uint64_t symsize,
688 	char *strtab, uint64_t strsize)
689 {
690 	static int inited = 0;
691 	Elf64_Ehdr ehdr;
692 
693 	if (inited)
694 		return EBUSY;
695 	inited = 1;
696 
697 	/*
698 	 * Use 64bit header since it's bigger.  Shouldn't make a
699 	 * difference, since we're passing in all zeroes anyway.
700 	 */
701 	memset(&ehdr, 0, sizeof(ehdr));
702 	ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
703 
704 	return 0;
705 }
706 
707 static int
708 rump_proxy_syscall(int num, void *arg, register_t *retval)
709 {
710 	struct lwp *l;
711 	struct sysent *callp;
712 	int rv;
713 
714 	if (__predict_false(num >= SYS_NSYSENT))
715 		return ENOSYS;
716 
717 	callp = rump_sysent + num;
718 	l = curlwp;
719 	rv = sy_call(callp, l, (void *)arg, retval);
720 
721 	return rv;
722 }
723 
724 static int
725 rump_proxy_rfork(void *priv, int flags)
726 {
727 	struct vmspace *newspace;
728 	int error;
729 
730 	if ((error = rump_lwproc_rfork(flags)) != 0)
731 		return error;
732 
733 	/*
734 	 * Since it's a proxy proc, adjust the vmspace.
735 	 * Refcount will eternally be 1.
736 	 */
737 	newspace = kmem_alloc(sizeof(*newspace), KM_SLEEP);
738 	newspace->vm_refcnt = 1;
739 	newspace->vm_map.pmap = priv;
740 	KASSERT(curproc->p_vmspace == vmspace_kernel());
741 	curproc->p_vmspace = newspace;
742 
743 	return 0;
744 }
745 
746 static void
747 rump_proxy_procexit(void)
748 {
749 	struct proc *p = curproc;
750 	uint64_t where;
751 	struct lwp *l;
752 
753 	mutex_enter(p->p_lock);
754 	/*
755 	 * First pass: mark all lwps in the process with LSDEAD
756 	 * so that they know they should exit.
757 	 */
758 	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
759 		if (l == curlwp)
760 			continue;
761 		l->l_stat = LSDEAD;
762 	}
763 	mutex_exit(p->p_lock);
764 
765 	/*
766 	 * Next, make sure everyone on all CPUs sees our status
767 	 * update.  This keeps threads inside cv_wait() and makes
768 	 * sure we don't access a stale cv pointer later when
769 	 * we wake up the threads.
770 	 */
771 
772 	where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
773 	xc_wait(where);
774 
775 	/*
776 	 * Ok, all lwps are either:
777 	 *  1) not in the cv code
778 	 *  2) sleeping on l->l_private
779 	 *  3) sleeping on p->p_waitcv
780 	 *
781 	 * Either way, l_private is stable until we change the lwps
782 	 * state to LSZOMB.
783 	 */
784 
785 	mutex_enter(p->p_lock);
786 	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
787 		if (l->l_private)
788 			cv_broadcast(l->l_private);
789 		l->l_stat = LSZOMB;
790 	}
791 	cv_broadcast(&p->p_waitcv);
792 	mutex_exit(p->p_lock);
793 
794 	/*
795 	 * Don't wait for lwps to exit.  There's refcounting in the
796 	 * rumpuser sp code which makes this safe.  Also, this routine
797 	 * should not sleep for a long time.
798 	 */
799 }
800 
801 int
802 rump_boot_gethowto()
803 {
804 
805 	return boothowto;
806 }
807 
808 void
809 rump_boot_sethowto(int howto)
810 {
811 
812 	boothowto = howto;
813 }
814 
815 int
816 rump_getversion(void)
817 {
818 
819 	return __NetBSD_Version__;
820 }
821 
822 /*
823  * Note: may be called unscheduled.  Not fully safe since no locking
824  * of allevents (currently that's not even available).
825  */
826 void
827 rump_printevcnts()
828 {
829 	struct evcnt *ev;
830 
831 	TAILQ_FOREACH(ev, &allevents, ev_list)
832 		rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
833 		    ev->ev_group, ev->ev_name, ev->ev_count);
834 }
835 
836 /*
837  * If you use this interface ... well ... all bets are off.
838  * The original purpose is for the p2k fs server library to be
839  * able to use the same pid/lid for VOPs as the host kernel.
840  */
841 void
842 rump_allbetsareoff_setid(pid_t pid, int lid)
843 {
844 	struct lwp *l = curlwp;
845 	struct proc *p = l->l_proc;
846 
847 	l->l_lid = lid;
848 	p->p_pid = pid;
849 }
850