Home | History | Annotate | Line # | Download | only in rumpkern
rump.c revision 1.175 
      1 /*	$NetBSD: rump.c,v 1.175 2010/06/03 10:56:20 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 1.175 2010/06/03 10:56:20 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 
     73 #include <rump/rumpuser.h>
     74 
     75 #include <secmodel/suser/suser.h>
     76 
     77 #include <prop/proplib.h>
     78 
     79 #include <uvm/uvm_extern.h>
     80 #include <uvm/uvm_readahead.h>
     81 
     82 #include "rump_private.h"
     83 #include "rump_net_private.h"
     84 #include "rump_vfs_private.h"
     85 #include "rump_dev_private.h"
     86 
     87 struct proc proc0;
     88 struct session rump_session = {
     89 	.s_count = 1,
     90 	.s_flags = 0,
     91 	.s_leader = &proc0,
     92 	.s_login = "rumphobo",
     93 	.s_sid = 0,
     94 };
     95 struct pgrp rump_pgrp = {
     96 	.pg_members = LIST_HEAD_INITIALIZER(pg_members),
     97 	.pg_session = &rump_session,
     98 	.pg_jobc = 1,
     99 };
    100 struct pstats rump_stats;
    101 struct plimit rump_limits;
    102 struct filedesc rump_filedesc0;
    103 struct proclist allproc;
    104 char machine[] = MACHINE;
    105 static kauth_cred_t rump_susercred;
    106 
    107 /* pretend the master rump proc is init */
    108 struct proc *initproc = &proc0;
    109 
    110 struct rumpuser_mtx *rump_giantlock;
    111 
    112 struct device rump_rootdev = {
    113 	.dv_class = DV_VIRTUAL
    114 };
    115 
    116 #ifdef RUMP_WITHOUT_THREADS
    117 int rump_threads = 0;
    118 #else
    119 int rump_threads = 1;
    120 #endif
    121 
    122 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */
    123 
    124 static void
    125 rump_aiodone_worker(struct work *wk, void *dummy)
    126 {
    127 	struct buf *bp = (struct buf *)wk;
    128 
    129 	KASSERT(&bp->b_work == wk);
    130 	bp->b_iodone(bp);
    131 }
    132 
    133 static int rump_inited;
    134 
    135 /*
    136  * Make sure pnbuf_cache is available even without vfs
    137  */
    138 struct pool_cache *pnbuf_cache;
    139 int rump_initpnbufpool(void);
    140 int rump_initpnbufpool(void)
    141 {
    142 
    143         pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
    144 	    NULL, IPL_NONE, NULL, NULL, NULL);
    145 	return EOPNOTSUPP;
    146 }
    147 
    148 int rump__unavailable(void);
    149 int rump__unavailable() {return EOPNOTSUPP;}
    150 __weak_alias(rump_net_init,rump__unavailable);
    151 __weak_alias(rump_vfs_init,rump_initpnbufpool);
    152 __weak_alias(rump_dev_init,rump__unavailable);
    153 
    154 __weak_alias(rump_vfs_fini,rump__unavailable);
    155 
    156 __weak_alias(biodone,rump__unavailable);
    157 __weak_alias(sopoll,rump__unavailable);
    158 
    159 void rump__unavailable_vfs_panic(void);
    160 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
    161 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
    162 
    163 rump_proc_vfs_init_fn rump_proc_vfs_init;
    164 rump_proc_vfs_release_fn rump_proc_vfs_release;
    165 
    166 static void add_linkedin_modules(const struct modinfo *const *, size_t);
    167 
    168 static void __noinline
    169 messthestack(void)
    170 {
    171 	volatile uint32_t mess[64];
    172 	uint64_t d1, d2;
    173 	int i, error;
    174 
    175 	for (i = 0; i < 64; i++) {
    176 		rumpuser_gettime(&d1, &d2, &error);
    177 		mess[i] = d2;
    178 	}
    179 }
    180 
    181 /*
    182  * Create kern.hostname.  why only this you ask.  well, init_sysctl
    183  * is a kitchen sink in need of some gardening.  but i want to use
    184  * kern.hostname today.
    185  */
    186 static void
    187 mksysctls(void)
    188 {
    189 
    190 	sysctl_createv(NULL, 0, NULL, NULL,
    191 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "kern", NULL,
    192 	    NULL, 0, NULL, 0, CTL_KERN, CTL_EOL);
    193 
    194 	/* XXX: setting hostnamelen is missing */
    195 	sysctl_createv(NULL, 0, NULL, NULL,
    196 	    CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_STRING, "hostname",
    197 	    SYSCTL_DESCR("System hostname"), NULL, 0,
    198 	    &hostname, MAXHOSTNAMELEN, CTL_KERN, KERN_HOSTNAME, CTL_EOL);
    199 }
    200 
    201 int
    202 rump__init(int rump_version)
    203 {
    204 	char buf[256];
    205 	struct timespec ts;
    206 	uint64_t sec, nsec;
    207 	struct proc *p;
    208 	struct lwp *l;
    209 	int i, numcpu;
    210 	int error;
    211 
    212 	/* not reentrant */
    213 	if (rump_inited)
    214 		return 0;
    215 	else if (rump_inited == -1)
    216 		panic("rump_init: host process restart required");
    217 	else
    218 		rump_inited = 1;
    219 
    220 	if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) {
    221 		if (*buf != '0')
    222 			boothowto = AB_VERBOSE;
    223 	}
    224 
    225 	if (rumpuser_getenv("RUMP_NCPU", buf, sizeof(buf), &error) == 0)
    226 		error = 0;
    227 	/* non-x86 is missing CPU_INFO_FOREACH() support */
    228 #if defined(__i386__) || defined(__x86_64__)
    229 	if (error == 0) {
    230 		numcpu = strtoll(buf, NULL, 10);
    231 		if (numcpu < 1)
    232 			numcpu = 1;
    233 	} else {
    234 		numcpu = rumpuser_getnhostcpu();
    235 	}
    236 #else
    237 	if (error == 0)
    238 		printf("NCPU limited to 1 on this host\n");
    239 	numcpu = 1;
    240 #endif
    241 	rump_cpus_bootstrap(numcpu);
    242 
    243 	rumpuser_gettime(&sec, &nsec, &error);
    244 	boottime.tv_sec = sec;
    245 	boottime.tv_nsec = nsec;
    246 
    247 	initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
    248 	aprint_verbose("%s%s", copyright, version);
    249 
    250 	/*
    251 	 * Seed arc4random() with a "reasonable" amount of randomness.
    252 	 * Yes, this is a quick kludge which depends on the arc4random
    253 	 * implementation.
    254 	 */
    255 	messthestack();
    256 	arc4random();
    257 
    258 	if (rump_version != RUMP_VERSION) {
    259 		printf("rump version mismatch, %d vs. %d\n",
    260 		    rump_version, RUMP_VERSION);
    261 		return EPROGMISMATCH;
    262 	}
    263 
    264 	if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
    265 		rump_threads = *buf != '0';
    266 	}
    267 	rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
    268 	    rump_threads);
    269 	rump_intr_init();
    270 	rump_tsleep_init();
    271 
    272 	/* init minimal lwp/cpu context */
    273 	l = &lwp0;
    274 	l->l_lid = 1;
    275 	l->l_cpu = l->l_target_cpu = rump_cpu;
    276 	rumpuser_set_curlwp(l);
    277 
    278 	mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE);
    279 	rumpuser_mutex_recursive_init(&rump_giantlock);
    280 	ksyms_init();
    281 	uvm_init();
    282 	evcnt_init();
    283 
    284 	once_init();
    285 	prop_kern_init();
    286 
    287 	pool_subsystem_init();
    288 	kmem_init();
    289 
    290 	uvm_ra_init();
    291 
    292 	mutex_obj_init();
    293 	callout_startup();
    294 
    295 	kprintf_init();
    296 	loginit();
    297 
    298 	kauth_init();
    299 	rump_susercred = rump_cred_create(0, 0, 0, NULL);
    300 
    301 	/* init proc0 and rest of lwp0 now that we can allocate memory */
    302 	p = &proc0;
    303 	p->p_stats = &rump_stats;
    304 	p->p_limit = &rump_limits;
    305 	p->p_pgrp = &rump_pgrp;
    306 	p->p_pid = 0;
    307 	p->p_fd = &rump_filedesc0;
    308 	p->p_vmspace = &rump_vmspace;
    309 	p->p_emul = &emul_netbsd;
    310 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    311 	l->l_cred = rump_cred_suserget();
    312 	l->l_proc = p;
    313 	LIST_INIT(&allproc);
    314 	LIST_INSERT_HEAD(&allproc, &proc0, p_list);
    315 	proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    316 	lwpinit_specificdata();
    317 
    318 	mutex_init(&rump_limits.pl_lock, MUTEX_DEFAULT, IPL_NONE);
    319 	rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
    320 	rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY;
    321 	rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY;
    322 	rump_limits.pl_corename = defcorename;
    323 
    324 	rump_scheduler_init();
    325 	/* revert temporary context and schedule a real context */
    326 	l->l_cpu = NULL;
    327 	rumpuser_set_curlwp(NULL);
    328 	rump_schedule();
    329 
    330 	percpu_init();
    331 	inittimecounter();
    332 	ntp_init();
    333 
    334 	rumpuser_gettime(&sec, &nsec, &error);
    335 	ts.tv_sec = sec;
    336 	ts.tv_nsec = nsec;
    337 	tc_setclock(&ts);
    338 
    339 	/* we are mostly go.  do per-cpu subsystem init */
    340 	for (i = 0; i < ncpu; i++) {
    341 		struct cpu_info *ci = cpu_lookup(i);
    342 
    343 		callout_init_cpu(ci);
    344 		softint_init(ci);
    345 		xc_init_cpu(ci);
    346 		pool_cache_cpu_init(ci);
    347 		selsysinit(ci);
    348 		percpu_init_cpu(ci);
    349 	}
    350 
    351 	sysctl_init();
    352 	kqueue_init();
    353 	iostat_init();
    354 	uid_init();
    355 	fd_sys_init();
    356 	module_init();
    357 	devsw_init();
    358 	pipe_init();
    359 	resource_init();
    360 
    361 	/* start page baroness */
    362 	if (rump_threads) {
    363 		if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
    364 		    uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
    365 			panic("pagedaemon create failed");
    366 	} else
    367 		uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
    368 
    369 	/* process dso's */
    370 	rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load);
    371 
    372 	/* these do nothing if not present */
    373 	rump_vfs_init();
    374 	rump_net_init();
    375 	rump_dev_init();
    376 	cold = 0;
    377 
    378 	/* aieeeedondest */
    379 	if (rump_threads) {
    380 		if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
    381 		    rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
    382 			panic("aiodoned");
    383 	}
    384 
    385 	mksysctls();
    386 	sysctl_finalize();
    387 
    388 	module_init_class(MODULE_CLASS_ANY);
    389 
    390 	rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
    391 	hostnamelen = strlen(hostname);
    392 
    393 	sigemptyset(&sigcantmask);
    394 
    395 	lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0);
    396 
    397 	if (rump_threads)
    398 		vmem_rehash_start();
    399 
    400 	rump_unschedule();
    401 
    402 	return 0;
    403 }
    404 
    405 /* maybe support sys_reboot some day for remote shutdown */
    406 void
    407 rump_reboot(int howto)
    408 {
    409 
    410 	/* dump means we really take the dive here */
    411 	if ((howto & RB_DUMP) || panicstr) {
    412 		rumpuser_exit(RUMPUSER_PANIC);
    413 		/*NOTREACHED*/
    414 	}
    415 
    416 	/* try to sync */
    417 	if (!((howto & RB_NOSYNC) || panicstr)) {
    418 		rump_vfs_fini();
    419 	}
    420 
    421 	/* your wish is my command */
    422 	if (howto & RB_HALT) {
    423 		for (;;) {
    424 			uint64_t sec = 5, nsec = 0;
    425 			int error;
    426 
    427 			rumpuser_nanosleep(&sec, &nsec, &error);
    428 		}
    429 	}
    430 	rump_inited = -1;
    431 }
    432 
    433 struct uio *
    434 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
    435 {
    436 	struct uio *uio;
    437 	enum uio_rw uiorw;
    438 
    439 	switch (rw) {
    440 	case RUMPUIO_READ:
    441 		uiorw = UIO_READ;
    442 		break;
    443 	case RUMPUIO_WRITE:
    444 		uiorw = UIO_WRITE;
    445 		break;
    446 	default:
    447 		panic("%s: invalid rw %d", __func__, rw);
    448 	}
    449 
    450 	uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
    451 	uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
    452 
    453 	uio->uio_iov->iov_base = buf;
    454 	uio->uio_iov->iov_len = bufsize;
    455 
    456 	uio->uio_iovcnt = 1;
    457 	uio->uio_offset = offset;
    458 	uio->uio_resid = bufsize;
    459 	uio->uio_rw = uiorw;
    460 	uio->uio_vmspace = UIO_VMSPACE_SYS;
    461 
    462 	return uio;
    463 }
    464 
    465 size_t
    466 rump_uio_getresid(struct uio *uio)
    467 {
    468 
    469 	return uio->uio_resid;
    470 }
    471 
    472 off_t
    473 rump_uio_getoff(struct uio *uio)
    474 {
    475 
    476 	return uio->uio_offset;
    477 }
    478 
    479 size_t
    480 rump_uio_free(struct uio *uio)
    481 {
    482 	size_t resid;
    483 
    484 	resid = uio->uio_resid;
    485 	kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
    486 	kmem_free(uio, sizeof(*uio));
    487 
    488 	return resid;
    489 }
    490 
    491 static pid_t nextpid = 1;
    492 struct lwp *
    493 rump_newproc_switch()
    494 {
    495 	struct lwp *l;
    496 	pid_t mypid;
    497 
    498 	mypid = atomic_inc_uint_nv(&nextpid);
    499 	if (__predict_false(mypid == 0))
    500 		mypid = atomic_inc_uint_nv(&nextpid);
    501 
    502 	l = rump_lwp_alloc(mypid, 0);
    503 	rump_lwp_switch(l);
    504 
    505 	return l;
    506 }
    507 
    508 struct lwp *
    509 rump_lwp_alloc_and_switch(pid_t pid, lwpid_t lid)
    510 {
    511 	struct lwp *l;
    512 
    513 	l = rump_lwp_alloc(pid, lid);
    514 	rump_lwp_switch(l);
    515 
    516 	return l;
    517 }
    518 
    519 struct lwp *
    520 rump_lwp_alloc(pid_t pid, lwpid_t lid)
    521 {
    522 	struct lwp *l;
    523 	struct proc *p;
    524 
    525 	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
    526 	if (pid != 0) {
    527 		p = kmem_zalloc(sizeof(*p), KM_SLEEP);
    528 		if (rump_proc_vfs_init)
    529 			rump_proc_vfs_init(p);
    530 		p->p_stats = &rump_stats;
    531 		p->p_limit = lim_copy(&rump_limits);
    532 		p->p_pid = pid;
    533 		p->p_vmspace = &rump_vmspace;
    534 		p->p_emul = &emul_netbsd;
    535 		p->p_fd = fd_init(NULL);
    536 		p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    537 		p->p_pgrp = &rump_pgrp;
    538 		l->l_cred = rump_cred_suserget();
    539 	} else {
    540 		p = &proc0;
    541 		l->l_cred = rump_susercred;
    542 	}
    543 
    544 	l->l_proc = p;
    545 	l->l_lid = lid;
    546 	l->l_fd = p->p_fd;
    547 	l->l_cpu = NULL;
    548 	l->l_target_cpu = rump_cpu;
    549 	lwp_initspecific(l);
    550 	LIST_INSERT_HEAD(&alllwp, l, l_list);
    551 
    552 	return l;
    553 }
    554 
    555 void
    556 rump_lwp_switch(struct lwp *newlwp)
    557 {
    558 	struct lwp *l = curlwp;
    559 
    560 	rumpuser_set_curlwp(NULL);
    561 	newlwp->l_cpu = newlwp->l_target_cpu = l->l_cpu;
    562 	newlwp->l_mutex = l->l_mutex;
    563 	l->l_mutex = NULL;
    564 	l->l_cpu = NULL;
    565 	rumpuser_set_curlwp(newlwp);
    566 	if (l->l_flag & LW_WEXIT)
    567 		rump_lwp_free(l);
    568 }
    569 
    570 /* XXX: this has effect only on non-pid0 lwps */
    571 void
    572 rump_lwp_release(struct lwp *l)
    573 {
    574 	struct proc *p;
    575 
    576 	p = l->l_proc;
    577 	if (p->p_pid != 0) {
    578 		mutex_obj_free(p->p_lock);
    579 		fd_free();
    580 		if (rump_proc_vfs_release)
    581 			rump_proc_vfs_release(p);
    582 		rump_cred_put(l->l_cred);
    583 		limfree(p->p_limit);
    584 		kmem_free(p, sizeof(*p));
    585 	}
    586 	KASSERT((l->l_flag & LW_WEXIT) == 0);
    587 	l->l_flag |= LW_WEXIT;
    588 }
    589 
    590 void
    591 rump_lwp_free(struct lwp *l)
    592 {
    593 
    594 	KASSERT(l->l_flag & LW_WEXIT);
    595 	KASSERT(l->l_mutex == NULL);
    596 	if (l->l_name)
    597 		kmem_free(l->l_name, MAXCOMLEN);
    598 	lwp_finispecific(l);
    599 	LIST_REMOVE(l, l_list);
    600 	kmem_free(l, sizeof(*l));
    601 }
    602 
    603 struct lwp *
    604 rump_lwp_curlwp(void)
    605 {
    606 	struct lwp *l = curlwp;
    607 
    608 	if (l->l_flag & LW_WEXIT)
    609 		return NULL;
    610 	return l;
    611 }
    612 
    613 /* rump private.  NEEDS WORK! */
    614 void
    615 rump_set_vmspace(struct vmspace *vm)
    616 {
    617 	struct proc *p = curproc;
    618 
    619 	p->p_vmspace = vm;
    620 }
    621 
    622 kauth_cred_t
    623 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
    624 {
    625 	kauth_cred_t cred;
    626 	int rv;
    627 
    628 	cred = kauth_cred_alloc();
    629 	kauth_cred_setuid(cred, uid);
    630 	kauth_cred_seteuid(cred, uid);
    631 	kauth_cred_setsvuid(cred, uid);
    632 	kauth_cred_setgid(cred, gid);
    633 	kauth_cred_setgid(cred, gid);
    634 	kauth_cred_setegid(cred, gid);
    635 	kauth_cred_setsvgid(cred, gid);
    636 	rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
    637 	/* oh this is silly.  and by "this" I mean kauth_cred_setgroups() */
    638 	assert(rv == 0);
    639 
    640 	return cred;
    641 }
    642 
    643 void
    644 rump_cred_put(kauth_cred_t cred)
    645 {
    646 
    647 	kauth_cred_free(cred);
    648 }
    649 
    650 kauth_cred_t
    651 rump_cred_suserget(void)
    652 {
    653 
    654 	kauth_cred_hold(rump_susercred);
    655 	return rump_susercred;
    656 }
    657 
    658 /*
    659  * Return the next system lwpid
    660  */
    661 lwpid_t
    662 rump_nextlid(void)
    663 {
    664 	lwpid_t retid;
    665 
    666 	mutex_enter(proc0.p_lock);
    667 	/*
    668 	 * Take next one, don't return 0
    669 	 * XXX: most likely we'll have collisions in case this
    670 	 * wraps around.
    671 	 */
    672 	if (++proc0.p_nlwpid == 0)
    673 		++proc0.p_nlwpid;
    674 	retid = proc0.p_nlwpid;
    675 	mutex_exit(proc0.p_lock);
    676 
    677 	return retid;
    678 }
    679 
    680 static int compcounter[RUMP_COMPONENT_MAX];
    681 
    682 static void
    683 rump_component_init_cb(struct rump_component *rc, int type)
    684 {
    685 
    686 	KASSERT(type < RUMP_COMPONENT_MAX);
    687 	if (rc->rc_type == type) {
    688 		rc->rc_init();
    689 		compcounter[type]++;
    690 	}
    691 }
    692 
    693 int
    694 rump_component_count(enum rump_component_type type)
    695 {
    696 
    697 	KASSERT(type <= RUMP_COMPONENT_MAX);
    698 	return compcounter[type];
    699 }
    700 
    701 void
    702 rump_component_init(enum rump_component_type type)
    703 {
    704 
    705 	rumpuser_dl_component_init(type, rump_component_init_cb);
    706 }
    707 
    708 /*
    709  * Initialize a module which has already been loaded and linked
    710  * with dlopen(). This is fundamentally the same as a builtin module.
    711  */
    712 int
    713 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
    714 {
    715 
    716 	return module_builtin_add(mip, nmodinfo, true);
    717 }
    718 
    719 /*
    720  * Finish module (flawless victory, fatality!).
    721  */
    722 int
    723 rump_module_fini(const struct modinfo *mi)
    724 {
    725 
    726 	return module_builtin_remove(mi, true);
    727 }
    728 
    729 /*
    730  * Add loaded and linked module to the builtin list.  It will
    731  * later be initialized with module_init_class().
    732  */
    733 
    734 static void
    735 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
    736 {
    737 
    738 	module_builtin_add(mip, nmodinfo, false);
    739 }
    740 
    741 int
    742 rump_kernelfsym_load(void *symtab, uint64_t symsize,
    743 	char *strtab, uint64_t strsize)
    744 {
    745 	static int inited = 0;
    746 	Elf64_Ehdr ehdr;
    747 
    748 	if (inited)
    749 		return EBUSY;
    750 	inited = 1;
    751 
    752 	/*
    753 	 * Use 64bit header since it's bigger.  Shouldn't make a
    754 	 * difference, since we're passing in all zeroes anyway.
    755 	 */
    756 	memset(&ehdr, 0, sizeof(ehdr));
    757 	ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
    758 
    759 	return 0;
    760 }
    761 
    762 static int
    763 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen,
    764 	register_t *retval)
    765 {
    766 	struct lwp *l;
    767 	struct sysent *callp;
    768 	int rv;
    769 
    770 	if (__predict_false(num >= SYS_NSYSENT))
    771 		return ENOSYS;
    772 
    773 	callp = rump_sysent + num;
    774 	rump_schedule();
    775 	l = curlwp;
    776 	rv = sy_call(callp, l, (void *)data, retval);
    777 	rump_unschedule();
    778 
    779 	return rv;
    780 }
    781 
    782 int
    783 rump_boot_gethowto()
    784 {
    785 
    786 	return boothowto;
    787 }
    788 
    789 void
    790 rump_boot_sethowto(int howto)
    791 {
    792 
    793 	boothowto = howto;
    794 }
    795 
    796 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local;
    797 void *rump_sysproxy_arg;
    798 
    799 /*
    800  * This whole syscall-via-rpc is still taking form.  For example, it
    801  * may be necessary to set syscalls individually instead of lobbing
    802  * them all to the same place.  So don't think this interface is
    803  * set in stone.
    804  */
    805 int
    806 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg)
    807 {
    808 
    809 	if (rump_sysproxy_arg)
    810 		return EBUSY;
    811 
    812 	rump_sysproxy_arg = arg;
    813 	rump_sysproxy = proxy;
    814 
    815 	return 0;
    816 }
    817 
    818 int
    819 rump_getversion(void)
    820 {
    821 
    822 	return __NetBSD_Version__;
    823 }
    824 
    825 /*
    826  * Note: may be called unscheduled.  Not fully safe since no locking
    827  * of allevents (currently that's not even available).
    828  */
    829 void
    830 rump_printevcnts()
    831 {
    832 	struct evcnt *ev;
    833 
    834 	TAILQ_FOREACH(ev, &allevents, ev_list)
    835 		rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
    836 		    ev->ev_group, ev->ev_name, ev->ev_count);
    837 }
    838