Home | History | Annotate | Line # | Download | only in rumpkern
rump.c revision 1.133 
      1 /*	$NetBSD: rump.c,v 1.133 2009/11/04 13:32:39 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 1.133 2009/11/04 13:32:39 pooka Exp $");
     32 
     33 #include <sys/param.h>
     34 #include <sys/atomic.h>
     35 #include <sys/buf.h>
     36 #include <sys/callout.h>
     37 #include <sys/conf.h>
     38 #include <sys/cpu.h>
     39 #include <sys/evcnt.h>
     40 #include <sys/event.h>
     41 #include <sys/filedesc.h>
     42 #include <sys/iostat.h>
     43 #include <sys/kauth.h>
     44 #include <sys/kernel.h>
     45 #include <sys/kmem.h>
     46 #include <sys/kprintf.h>
     47 #include <sys/ksyms.h>
     48 #include <sys/msgbuf.h>
     49 #include <sys/module.h>
     50 #include <sys/once.h>
     51 #include <sys/percpu.h>
     52 #include <sys/queue.h>
     53 #include <sys/reboot.h>
     54 #include <sys/resourcevar.h>
     55 #include <sys/select.h>
     56 #include <sys/sysctl.h>
     57 #include <sys/syscall.h>
     58 #include <sys/tty.h>
     59 #include <sys/uidinfo.h>
     60 #include <sys/vmem.h>
     61 
     62 #include <rump/rumpuser.h>
     63 
     64 #include <secmodel/suser/suser.h>
     65 
     66 #include <prop/proplib.h>
     67 
     68 #include "rump_private.h"
     69 #include "rump_net_private.h"
     70 #include "rump_vfs_private.h"
     71 #include "rump_dev_private.h"
     72 
     73 struct proc proc0;
     74 struct session rump_session = {
     75 	.s_count = 1,
     76 	.s_flags = 0,
     77 	.s_leader = &proc0,
     78 	.s_login = "rumphobo",
     79 	.s_sid = 0,
     80 };
     81 struct pgrp rump_pgrp = {
     82 	.pg_members = LIST_HEAD_INITIALIZER(pg_members),
     83 	.pg_session = &rump_session,
     84 	.pg_jobc = 1,
     85 };
     86 struct pstats rump_stats;
     87 struct plimit rump_limits;
     88 struct filedesc rump_filedesc0;
     89 struct proclist allproc;
     90 char machine[] = "rump";
     91 static kauth_cred_t rump_susercred;
     92 
     93 /* pretend the master rump proc is init */
     94 struct proc *initproc = &proc0;
     95 
     96 struct rumpuser_mtx *rump_giantlock;
     97 
     98 sigset_t sigcantmask;
     99 
    100 #ifdef RUMP_WITHOUT_THREADS
    101 int rump_threads = 0;
    102 #else
    103 int rump_threads = 1;
    104 #endif
    105 
    106 static void
    107 rump_aiodone_worker(struct work *wk, void *dummy)
    108 {
    109 	struct buf *bp = (struct buf *)wk;
    110 
    111 	KASSERT(&bp->b_work == wk);
    112 	bp->b_iodone(bp);
    113 }
    114 
    115 static int rump_inited;
    116 static struct emul emul_rump;
    117 
    118 int rump__unavailable(void);
    119 int rump__unavailable() {return EOPNOTSUPP;}
    120 __weak_alias(rump_net_init,rump__unavailable);
    121 __weak_alias(rump_vfs_init,rump__unavailable);
    122 __weak_alias(rump_dev_init,rump__unavailable);
    123 
    124 __weak_alias(rump_vfs_fini,rump__unavailable);
    125 
    126 __weak_alias(biodone,rump__unavailable);
    127 __weak_alias(sopoll,rump__unavailable);
    128 
    129 void rump__unavailable_vfs_panic(void);
    130 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
    131 __weak_alias(vn_open,rump__unavailable_vfs_panic);
    132 __weak_alias(vn_rdwr,rump__unavailable_vfs_panic);
    133 __weak_alias(vn_stat,rump__unavailable_vfs_panic);
    134 __weak_alias(vn_close,rump__unavailable_vfs_panic);
    135 __weak_alias(namei,rump__unavailable_vfs_panic);
    136 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
    137 
    138 static void
    139 pvfsinit_nop(struct proc *p)
    140 {
    141 
    142 	return;
    143 }
    144 
    145 static void
    146 pvfsrele_nop(struct proc *p)
    147 {
    148 
    149 	return;
    150 }
    151 
    152 rump_proc_vfs_init_fn rump_proc_vfs_init = pvfsinit_nop;
    153 rump_proc_vfs_release_fn rump_proc_vfs_release = pvfsrele_nop;
    154 
    155 /*
    156  * Stir up the stack a bit.  These are exported functions to help
    157  * convince the compiler that we don't want these routines completely
    158  * optimized out or inlined.  Is there an easier way to do this?
    159  */
    160 void nullfn(uint32_t *);
    161 void nullfn(uint32_t *arg){}
    162 void messthestack(void);
    163 void
    164 messthestack(void)
    165 {
    166 	uint32_t mess[64];
    167 	uint64_t d1, d2;
    168 	int i, error;
    169 
    170 	for (i = 0; i < 64; i++) {
    171 		rumpuser_gettime(&d1, &d2, &error);
    172 		mess[i] = d2;
    173 	}
    174 	nullfn(mess);
    175 }
    176 
    177 int
    178 rump__init(int rump_version)
    179 {
    180 	char buf[256];
    181 	struct proc *p;
    182 	struct lwp *l;
    183 	int error;
    184 
    185 	/* not reentrant */
    186 	if (rump_inited)
    187 		return 0;
    188 	else if (rump_inited == -1)
    189 		panic("rump_init: host process restart required");
    190 	else
    191 		rump_inited = 1;
    192 
    193 	/*
    194 	 * Seed arc4random() with a "reasonable" amount of randomness.
    195 	 * Yes, this is a quick kludge which depends on the arc4random
    196 	 * implementation.
    197 	 */
    198 	messthestack();
    199 	arc4random();
    200 
    201 	if (rump_version != RUMP_VERSION) {
    202 		printf("rump version mismatch, %d vs. %d\n",
    203 		    rump_version, RUMP_VERSION);
    204 		return EPROGMISMATCH;
    205 	}
    206 
    207 	if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
    208 		rump_threads = *buf != '0';
    209 	}
    210 	rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
    211 	    rump_threads);
    212 
    213 	/* init minimal lwp/cpu context */
    214 	l = &lwp0;
    215 	l->l_lid = 1;
    216 	l->l_cpu = rump_cpu;
    217 	rumpuser_set_curlwp(l);
    218 
    219 	mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE);
    220 	rumpuser_mutex_recursive_init(&rump_giantlock);
    221 	ksyms_init();
    222 	rumpvm_init();
    223 	evcnt_init();
    224 
    225 	once_init();
    226 	prop_kern_init();
    227 
    228 	rump_sleepers_init();
    229 
    230 	pool_subsystem_init();
    231 	kmem_init();
    232 
    233 	mutex_obj_init();
    234 
    235 	kprintf_init();
    236 	loginit();
    237 
    238 	kauth_init();
    239 	rump_susercred = rump_cred_create(0, 0, 0, NULL);
    240 
    241 	/* init proc0 and rest of lwp0 now that we can allocate memory */
    242 	p = &proc0;
    243 	p->p_stats = &rump_stats;
    244 	p->p_limit = &rump_limits;
    245 	p->p_pgrp = &rump_pgrp;
    246 	p->p_pid = 0;
    247 	p->p_fd = &rump_filedesc0;
    248 	p->p_vmspace = &rump_vmspace;
    249 	p->p_emul = &emul_rump;
    250 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    251 	l->l_cred = rump_cred_suserget();
    252 	l->l_proc = p;
    253 	LIST_INIT(&allproc);
    254 	LIST_INSERT_HEAD(&allproc, &proc0, p_list);
    255 	proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    256 
    257 	rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
    258 	rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY;
    259 	rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY;
    260 
    261 	rump_scheduler_init();
    262 	/* revert temporary context and schedule a real context */
    263 	l->l_cpu = NULL;
    264 	rumpuser_set_curlwp(NULL);
    265 	rump_schedule();
    266 
    267 	callout_startup();
    268 	callout_init_cpu(rump_cpu);
    269 	selsysinit(rump_cpu);
    270 
    271 	sysctl_init();
    272 	kqueue_init();
    273 	iostat_init();
    274 	uid_init();
    275 	percpu_init();
    276 	fd_sys_init();
    277 	module_init();
    278 	softint_init(rump_cpu);
    279 	devsw_init();
    280 
    281 	/* these do nothing if not present */
    282 	rump_vfs_init();
    283 	rump_net_init();
    284 	rump_dev_init();
    285 	cold = 0;
    286 
    287 	/* aieeeedondest */
    288 	if (rump_threads) {
    289 		if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
    290 		    rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
    291 			panic("aiodoned");
    292 	}
    293 
    294 	sysctl_finalize();
    295 
    296 	rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
    297 	hostnamelen = strlen(hostname);
    298 
    299 	sigemptyset(&sigcantmask);
    300 
    301 	lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0);
    302 
    303 #ifdef RUMP_USE_REAL_ALLOCATORS
    304 	if (rump_threads)
    305 		vmem_rehash_start();
    306 #endif
    307 
    308 	rumpuser_dl_module_bootstrap(rump_module_init);
    309 	rump_unschedule();
    310 
    311 	return 0;
    312 }
    313 
    314 /* maybe support sys_reboot some day for remote shutdown */
    315 void
    316 rump_reboot(int howto)
    317 {
    318 
    319 	/* dump means we really take the dive here */
    320 	if ((howto & RB_DUMP) || panicstr) {
    321 		rumpuser_exit(RUMPUSER_PANIC);
    322 		/*NOTREACHED*/
    323 	}
    324 
    325 	/* try to sync */
    326 	if (!((howto & RB_NOSYNC) || panicstr)) {
    327 		rump_vfs_fini();
    328 	}
    329 
    330 	/* your wish is my command */
    331 	if (howto & RB_HALT) {
    332 		for (;;) {
    333 			uint64_t sec = 5, nsec = 0;
    334 			int error;
    335 
    336 			rumpuser_nanosleep(&sec, &nsec, &error);
    337 		}
    338 	}
    339 	rump_inited = -1;
    340 }
    341 
    342 struct uio *
    343 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
    344 {
    345 	struct uio *uio;
    346 	enum uio_rw uiorw;
    347 
    348 	switch (rw) {
    349 	case RUMPUIO_READ:
    350 		uiorw = UIO_READ;
    351 		break;
    352 	case RUMPUIO_WRITE:
    353 		uiorw = UIO_WRITE;
    354 		break;
    355 	default:
    356 		panic("%s: invalid rw %d", __func__, rw);
    357 	}
    358 
    359 	uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
    360 	uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
    361 
    362 	uio->uio_iov->iov_base = buf;
    363 	uio->uio_iov->iov_len = bufsize;
    364 
    365 	uio->uio_iovcnt = 1;
    366 	uio->uio_offset = offset;
    367 	uio->uio_resid = bufsize;
    368 	uio->uio_rw = uiorw;
    369 	uio->uio_vmspace = UIO_VMSPACE_SYS;
    370 
    371 	return uio;
    372 }
    373 
    374 size_t
    375 rump_uio_getresid(struct uio *uio)
    376 {
    377 
    378 	return uio->uio_resid;
    379 }
    380 
    381 off_t
    382 rump_uio_getoff(struct uio *uio)
    383 {
    384 
    385 	return uio->uio_offset;
    386 }
    387 
    388 size_t
    389 rump_uio_free(struct uio *uio)
    390 {
    391 	size_t resid;
    392 
    393 	resid = uio->uio_resid;
    394 	kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
    395 	kmem_free(uio, sizeof(*uio));
    396 
    397 	return resid;
    398 }
    399 
    400 static pid_t nextpid = 1;
    401 struct lwp *
    402 rump_newproc_switch()
    403 {
    404 	struct lwp *l;
    405 	pid_t mypid;
    406 
    407 	mypid = atomic_inc_uint_nv(&nextpid);
    408 	if (__predict_false(mypid == 0))
    409 		mypid = atomic_inc_uint_nv(&nextpid);
    410 
    411 	l = rump_lwp_alloc(mypid, 0);
    412 	rump_lwp_switch(l);
    413 
    414 	return l;
    415 }
    416 
    417 struct lwp *
    418 rump_lwp_alloc_and_switch(pid_t pid, lwpid_t lid)
    419 {
    420 	struct lwp *l;
    421 
    422 	l = rump_lwp_alloc(pid, lid);
    423 	rump_lwp_switch(l);
    424 
    425 	return l;
    426 }
    427 
    428 struct lwp *
    429 rump_lwp_alloc(pid_t pid, lwpid_t lid)
    430 {
    431 	struct lwp *l;
    432 	struct proc *p;
    433 
    434 	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
    435 	if (pid != 0) {
    436 		p = kmem_zalloc(sizeof(*p), KM_SLEEP);
    437 		rump_proc_vfs_init(p);
    438 		p->p_stats = &rump_stats;
    439 		p->p_limit = &rump_limits;
    440 		p->p_pid = pid;
    441 		p->p_vmspace = &rump_vmspace;
    442 		p->p_fd = fd_init(NULL);
    443 		p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    444 	} else {
    445 		p = &proc0;
    446 	}
    447 
    448 	l->l_cred = rump_cred_suserget();
    449 	l->l_proc = p;
    450 	l->l_lid = lid;
    451 	l->l_fd = p->p_fd;
    452 	l->l_cpu = NULL;
    453 
    454 	return l;
    455 }
    456 
    457 void
    458 rump_lwp_switch(struct lwp *newlwp)
    459 {
    460 	struct lwp *l = curlwp;
    461 
    462 	rumpuser_set_curlwp(NULL);
    463 	newlwp->l_cpu = l->l_cpu;
    464 	newlwp->l_mutex = l->l_mutex;
    465 	l->l_mutex = NULL;
    466 	l->l_cpu = NULL;
    467 	rumpuser_set_curlwp(newlwp);
    468 	if (l->l_flag & LW_WEXIT)
    469 		rump_lwp_free(l);
    470 }
    471 
    472 /* XXX: this has effect only on non-pid0 lwps */
    473 void
    474 rump_lwp_release(struct lwp *l)
    475 {
    476 	struct proc *p;
    477 
    478 	p = l->l_proc;
    479 	if (p->p_pid != 0) {
    480 		mutex_obj_free(p->p_lock);
    481 		fd_free();
    482 		rump_proc_vfs_release(p);
    483 		rump_cred_put(l->l_cred);
    484 		kmem_free(p, sizeof(*p));
    485 	}
    486 	KASSERT((l->l_flag & LW_WEXIT) == 0);
    487 	l->l_flag |= LW_WEXIT;
    488 }
    489 
    490 void
    491 rump_lwp_free(struct lwp *l)
    492 {
    493 
    494 	KASSERT(l->l_flag & LW_WEXIT);
    495 	KASSERT(l->l_mutex == NULL);
    496 	rump_cred_put(l->l_cred);
    497 	kmem_free(l, sizeof(*l));
    498 }
    499 
    500 struct lwp *
    501 rump_lwp_curlwp(void)
    502 {
    503 	struct lwp *l = curlwp;
    504 
    505 	if (l->l_flag & LW_WEXIT)
    506 		return NULL;
    507 	return l;
    508 }
    509 
    510 /* rump private.  NEEDS WORK! */
    511 void
    512 rump_set_vmspace(struct vmspace *vm)
    513 {
    514 	struct proc *p = curproc;
    515 
    516 	p->p_vmspace = vm;
    517 }
    518 
    519 kauth_cred_t
    520 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
    521 {
    522 	kauth_cred_t cred;
    523 	int rv;
    524 
    525 	cred = kauth_cred_alloc();
    526 	kauth_cred_setuid(cred, uid);
    527 	kauth_cred_seteuid(cred, uid);
    528 	kauth_cred_setsvuid(cred, uid);
    529 	kauth_cred_setgid(cred, gid);
    530 	kauth_cred_setgid(cred, gid);
    531 	kauth_cred_setegid(cred, gid);
    532 	kauth_cred_setsvgid(cred, gid);
    533 	rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
    534 	/* oh this is silly.  and by "this" I mean kauth_cred_setgroups() */
    535 	assert(rv == 0);
    536 
    537 	return cred;
    538 }
    539 
    540 void
    541 rump_cred_put(kauth_cred_t cred)
    542 {
    543 
    544 	kauth_cred_free(cred);
    545 }
    546 
    547 kauth_cred_t
    548 rump_cred_suserget(void)
    549 {
    550 
    551 	kauth_cred_hold(rump_susercred);
    552 	return rump_susercred;
    553 }
    554 
    555 /*
    556  * Return the next system lwpid
    557  */
    558 lwpid_t
    559 rump_nextlid(void)
    560 {
    561 	lwpid_t retid;
    562 
    563 	mutex_enter(proc0.p_lock);
    564 	/*
    565 	 * Take next one, don't return 0
    566 	 * XXX: most likely we'll have collisions in case this
    567 	 * wraps around.
    568 	 */
    569 	if (++proc0.p_nlwpid == 0)
    570 		++proc0.p_nlwpid;
    571 	retid = proc0.p_nlwpid;
    572 	mutex_exit(proc0.p_lock);
    573 
    574 	return retid;
    575 }
    576 
    577 int
    578 rump_module_init(struct modinfo *mi, prop_dictionary_t props)
    579 {
    580 	int rv;
    581 
    582 	if (!module_compatible(mi->mi_version, __NetBSD_Version__))
    583 		return EPROGMISMATCH;
    584 
    585 	rv = mi->mi_modcmd(MODULE_CMD_INIT, props);
    586 	if (rv == 0 && mi->mi_class == MODULE_CLASS_SECMODEL)
    587 		secmodel_register();
    588 
    589 	return rv;
    590 }
    591 
    592 int
    593 rump_module_fini(struct modinfo *mi)
    594 {
    595 	int rv;
    596 
    597 	rv = mi->mi_modcmd(MODULE_CMD_FINI, NULL);
    598 	if (rv == 0 && mi->mi_class == MODULE_CLASS_SECMODEL)
    599 		secmodel_deregister();
    600 
    601 	return rv;
    602 }
    603 
    604 static int
    605 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen,
    606 	register_t *retval)
    607 {
    608 	struct lwp *l;
    609 	struct sysent *callp;
    610 	int rv;
    611 
    612 	if (__predict_false(num >= SYS_NSYSENT))
    613 		return ENOSYS;
    614 
    615 	callp = rump_sysent + num;
    616 	rump_schedule();
    617 	l = curlwp;
    618 	rv = callp->sy_call(l, (void *)data, retval);
    619 	rump_unschedule();
    620 
    621 	return rv;
    622 }
    623 
    624 int
    625 rump_boot_gethowto()
    626 {
    627 
    628 	return boothowto;
    629 }
    630 
    631 void
    632 rump_boot_sethowto(int howto)
    633 {
    634 
    635 	boothowto = howto;
    636 }
    637 
    638 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local;
    639 void *rump_sysproxy_arg;
    640 
    641 /*
    642  * This whole syscall-via-rpc is still taking form.  For example, it
    643  * may be necessary to set syscalls individually instead of lobbing
    644  * them all to the same place.  So don't think this interface is
    645  * set in stone.
    646  */
    647 int
    648 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg)
    649 {
    650 
    651 	if (rump_sysproxy_arg)
    652 		return EBUSY;
    653 
    654 	rump_sysproxy_arg = arg;
    655 	rump_sysproxy = proxy;
    656 
    657 	return 0;
    658 }
    659 
    660 int
    661 rump_getversion(void)
    662 {
    663 
    664 	return __NetBSD_Version__;
    665 }
    666