Home | History | Annotate | Line # | Download | only in rumpkern
rump.c revision 1.111 
      1 /*	$NetBSD: rump.c,v 1.111 2009/09/04 12:27:09 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 1.111 2009/09/04 12:27:09 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/resourcevar.h>
     54 #include <sys/select.h>
     55 #include <sys/sysctl.h>
     56 #include <sys/syscall.h>
     57 #include <sys/tty.h>
     58 #include <sys/uidinfo.h>
     59 #include <sys/vmem.h>
     60 
     61 #include <rump/rumpuser.h>
     62 
     63 #include <secmodel/secmodel.h>
     64 
     65 #include "rump_private.h"
     66 #include "rump_net_private.h"
     67 #include "rump_vfs_private.h"
     68 
     69 struct proc proc0;
     70 struct session rump_session = {
     71 	.s_count = 1,
     72 	.s_flags = 0,
     73 	.s_leader = &proc0,
     74 	.s_login = "rumphobo",
     75 	.s_sid = 0,
     76 };
     77 struct pgrp rump_pgrp = {
     78 	.pg_members = LIST_HEAD_INITIALIZER(pg_members),
     79 	.pg_session = &rump_session,
     80 	.pg_jobc = 1,
     81 };
     82 struct pstats rump_stats;
     83 struct plimit rump_limits;
     84 struct cpu_info rump_cpu;
     85 struct filedesc rump_filedesc0;
     86 struct proclist allproc;
     87 char machine[] = "rump";
     88 static kauth_cred_t rump_susercred;
     89 
     90 /* pretend the master rump proc is init */
     91 struct proc *initproc = &proc0;
     92 
     93 struct rumpuser_mtx *rump_giantlock;
     94 
     95 sigset_t sigcantmask;
     96 
     97 #ifdef RUMP_WITHOUT_THREADS
     98 int rump_threads = 0;
     99 #else
    100 int rump_threads = 1;
    101 #endif
    102 
    103 static void
    104 rump_aiodone_worker(struct work *wk, void *dummy)
    105 {
    106 	struct buf *bp = (struct buf *)wk;
    107 
    108 	KASSERT(&bp->b_work == wk);
    109 	bp->b_iodone(bp);
    110 }
    111 
    112 static int rump_inited;
    113 static struct emul emul_rump;
    114 
    115 void rump__unavailable(void);
    116 void rump__unavailable() {}
    117 __weak_alias(rump_net_init,rump__unavailable);
    118 __weak_alias(rump_vfs_init,rump__unavailable);
    119 
    120 void rump__unavailable_vfs_panic(void);
    121 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
    122 __weak_alias(vn_open,rump__unavailable_vfs_panic);
    123 __weak_alias(vn_rdwr,rump__unavailable_vfs_panic);
    124 __weak_alias(vn_close,rump__unavailable_vfs_panic);
    125 
    126 static void
    127 pvfsinit_nop(struct proc *p)
    128 {
    129 
    130 	return;
    131 }
    132 
    133 static void
    134 pvfsrele_nop(struct proc *p)
    135 {
    136 
    137 	return;
    138 }
    139 
    140 rump_proc_vfs_init_fn rump_proc_vfs_init = pvfsinit_nop;
    141 rump_proc_vfs_release_fn rump_proc_vfs_release = pvfsrele_nop;
    142 
    143 /*
    144  * Stir up the stack a bit.  These are exported functions to help
    145  * convince the compiler that we don't want these routines completely
    146  * optimized out or inlined.  Is there an easier way to do this?
    147  */
    148 void nullfn(uint32_t *);
    149 void nullfn(uint32_t *arg){}
    150 void messthestack(void);
    151 void
    152 messthestack(void)
    153 {
    154 	uint32_t mess[64];
    155 	uint64_t d1, d2;
    156 	int i, error;
    157 
    158 	for (i = 0; i < 64; i++) {
    159 		rumpuser_gettime(&d1, &d2, &error);
    160 		mess[i] = d2;
    161 	}
    162 	nullfn(mess);
    163 }
    164 
    165 int
    166 rump__init(int rump_version)
    167 {
    168 	char buf[256];
    169 	struct proc *p;
    170 	struct lwp *l;
    171 	int error;
    172 
    173 	/* XXX */
    174 	if (rump_inited)
    175 		return 0;
    176 	rump_inited = 1;
    177 
    178 	/*
    179 	 * Seed arc4random() with a "reasonable" amount of randomness.
    180 	 * Yes, this is a quick kludge which depends on the arc4random
    181 	 * implementation.
    182 	 */
    183 	messthestack();
    184 	arc4random();
    185 
    186 	if (rump_version != RUMP_VERSION) {
    187 		printf("rump version mismatch, %d vs. %d\n",
    188 		    rump_version, RUMP_VERSION);
    189 		return EPROGMISMATCH;
    190 	}
    191 
    192 	if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
    193 		rump_threads = *buf != '0';
    194 	}
    195 	rumpuser_thrinit(_kernel_lock, _kernel_unlock, rump_threads);
    196 
    197 	mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE);
    198 	rumpuser_mutex_recursive_init(&rump_giantlock);
    199 	ksyms_init();
    200 	rumpvm_init();
    201 	evcnt_init();
    202 
    203 	once_init();
    204 
    205 	rump_sleepers_init();
    206 
    207 	pool_subsystem_init();
    208 	kmem_init();
    209 
    210 	kprintf_init();
    211 	loginit();
    212 
    213 	kauth_init();
    214 	rump_susercred = rump_cred_create(0, 0, 0, NULL);
    215 
    216 	l = &lwp0;
    217 	p = &proc0;
    218 	p->p_stats = &rump_stats;
    219 	p->p_limit = &rump_limits;
    220 	p->p_pgrp = &rump_pgrp;
    221 	p->p_pid = 0;
    222 	p->p_fd = &rump_filedesc0;
    223 	p->p_vmspace = &rump_vmspace;
    224 	p->p_emul = &emul_rump;
    225 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    226 	l->l_cred = rump_cred_suserget();
    227 	l->l_proc = p;
    228 	l->l_lid = 1;
    229 	LIST_INIT(&allproc);
    230 	LIST_INSERT_HEAD(&allproc, &proc0, p_list);
    231 	proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    232 
    233 	rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
    234 	rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY;
    235 	rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY;
    236 
    237 	callout_startup();
    238 	callout_init_cpu(&rump_cpu);
    239 
    240 	kqueue_init();
    241 	iostat_init();
    242 	uid_init();
    243 	percpu_init();
    244 	fd_sys_init();
    245 	module_init();
    246 	sysctl_init();
    247 	softint_init(&rump_cpu);
    248 	cold = 0;
    249 	devsw_init();
    250 	secmodel_start();
    251 
    252 	/* these do nothing if not present */
    253 	rump_vfs_init();
    254 	rump_net_init();
    255 
    256 	/* aieeeedondest */
    257 	if (rump_threads) {
    258 		if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
    259 		    rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
    260 			panic("aiodoned");
    261 	}
    262 
    263 	rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
    264 	hostnamelen = strlen(hostname);
    265 
    266 	sigemptyset(&sigcantmask);
    267 
    268 	lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0);
    269 
    270 #ifdef RUMP_USE_REAL_ALLOCATORS
    271 	if (rump_threads)
    272 		vmem_rehash_start();
    273 #endif
    274 
    275 	return 0;
    276 }
    277 
    278 struct uio *
    279 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
    280 {
    281 	struct uio *uio;
    282 	enum uio_rw uiorw;
    283 
    284 	switch (rw) {
    285 	case RUMPUIO_READ:
    286 		uiorw = UIO_READ;
    287 		break;
    288 	case RUMPUIO_WRITE:
    289 		uiorw = UIO_WRITE;
    290 		break;
    291 	default:
    292 		panic("%s: invalid rw %d", __func__, rw);
    293 	}
    294 
    295 	uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
    296 	uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
    297 
    298 	uio->uio_iov->iov_base = buf;
    299 	uio->uio_iov->iov_len = bufsize;
    300 
    301 	uio->uio_iovcnt = 1;
    302 	uio->uio_offset = offset;
    303 	uio->uio_resid = bufsize;
    304 	uio->uio_rw = uiorw;
    305 	uio->uio_vmspace = UIO_VMSPACE_SYS;
    306 
    307 	return uio;
    308 }
    309 
    310 size_t
    311 rump_uio_getresid(struct uio *uio)
    312 {
    313 
    314 	return uio->uio_resid;
    315 }
    316 
    317 off_t
    318 rump_uio_getoff(struct uio *uio)
    319 {
    320 
    321 	return uio->uio_offset;
    322 }
    323 
    324 size_t
    325 rump_uio_free(struct uio *uio)
    326 {
    327 	size_t resid;
    328 
    329 	resid = uio->uio_resid;
    330 	kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
    331 	kmem_free(uio, sizeof(*uio));
    332 
    333 	return resid;
    334 }
    335 
    336 /* public interface */
    337 static pid_t nextpid = 1;
    338 struct lwp *
    339 rump_newproc_switch()
    340 {
    341 	struct lwp *oldlwp = curlwp;
    342 	pid_t mypid;
    343 
    344 	mypid = atomic_inc_uint_nv(&nextpid);
    345 	if (__predict_false(mypid == 0))
    346 		mypid = atomic_inc_uint_nv(&nextpid);
    347 
    348 	rumpuser_set_curlwp(NULL);
    349 	rump_setup_curlwp(mypid, 0, 1);
    350 
    351 	return oldlwp;
    352 }
    353 
    354 /* rump private */
    355 struct lwp *
    356 rump_setup_curlwp(pid_t pid, lwpid_t lid, int set)
    357 {
    358 	struct lwp *l;
    359 	struct proc *p;
    360 
    361 	l = kmem_zalloc(sizeof(struct lwp), KM_SLEEP);
    362 	if (pid != 0) {
    363 		p = kmem_zalloc(sizeof(struct proc), KM_SLEEP);
    364 		rump_proc_vfs_init(p);
    365 		p->p_stats = &rump_stats;
    366 		p->p_limit = &rump_limits;
    367 		p->p_pid = pid;
    368 		p->p_vmspace = &rump_vmspace;
    369 		p->p_fd = fd_init(NULL);
    370 		p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    371 	} else {
    372 		p = &proc0;
    373 	}
    374 
    375 	l->l_cred = rump_cred_suserget();
    376 	l->l_proc = p;
    377 	l->l_lid = lid;
    378 	l->l_fd = p->p_fd;
    379 	l->l_mutex = RUMP_LMUTEX_MAGIC;
    380 	l->l_cpu = &rump_cpu;
    381 
    382 	if (set)
    383 		rumpuser_set_curlwp(l);
    384 
    385 	return l;
    386 }
    387 
    388 /* rump private.  NEEDS WORK! */
    389 void
    390 rump_set_vmspace(struct vmspace *vm)
    391 {
    392 	struct proc *p = curproc;
    393 
    394 	p->p_vmspace = vm;
    395 }
    396 
    397 void
    398 rump_clear_curlwp(void)
    399 {
    400 	struct lwp *l;
    401 	struct proc *p;
    402 
    403 	l = rumpuser_get_curlwp();
    404 	p = l->l_proc;
    405 	if (p->p_pid != 0) {
    406 		mutex_obj_free(p->p_lock);
    407 		fd_free();
    408 		rump_proc_vfs_release(p);
    409 		rump_cred_put(l->l_cred);
    410 		kmem_free(p, sizeof(*p));
    411 	}
    412 	kmem_free(l, sizeof(*l));
    413 	rumpuser_set_curlwp(NULL);
    414 }
    415 
    416 struct lwp *
    417 rump_get_curlwp(void)
    418 {
    419 	struct lwp *l;
    420 
    421 	l = rumpuser_get_curlwp();
    422 	if (l == NULL)
    423 		l = &lwp0;
    424 
    425 	return l;
    426 }
    427 
    428 void
    429 rump_set_curlwp(struct lwp *l)
    430 {
    431 
    432 	/* clear current */
    433 	rumpuser_set_curlwp(NULL);
    434 	/* set new */
    435 	rumpuser_set_curlwp(l);
    436 }
    437 
    438 kauth_cred_t
    439 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
    440 {
    441 	kauth_cred_t cred;
    442 	int rv;
    443 
    444 	cred = kauth_cred_alloc();
    445 	kauth_cred_setuid(cred, uid);
    446 	kauth_cred_seteuid(cred, uid);
    447 	kauth_cred_setsvuid(cred, uid);
    448 	kauth_cred_setgid(cred, gid);
    449 	kauth_cred_setgid(cred, gid);
    450 	kauth_cred_setegid(cred, gid);
    451 	kauth_cred_setsvgid(cred, gid);
    452 	rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
    453 	/* oh this is silly.  and by "this" I mean kauth_cred_setgroups() */
    454 	assert(rv == 0);
    455 
    456 	return cred;
    457 }
    458 
    459 void
    460 rump_cred_put(kauth_cred_t cred)
    461 {
    462 
    463 	kauth_cred_free(cred);
    464 }
    465 
    466 kauth_cred_t
    467 rump_cred_suserget(void)
    468 {
    469 
    470 	kauth_cred_hold(rump_susercred);
    471 	return rump_susercred;
    472 }
    473 
    474 /*
    475  * Return the next system lwpid
    476  */
    477 lwpid_t
    478 rump_nextlid(void)
    479 {
    480 	lwpid_t retid;
    481 
    482 	mutex_enter(proc0.p_lock);
    483 	/*
    484 	 * Take next one, don't return 0
    485 	 * XXX: most likely we'll have collisions in case this
    486 	 * wraps around.
    487 	 */
    488 	if (++proc0.p_nlwpid == 0)
    489 		++proc0.p_nlwpid;
    490 	retid = proc0.p_nlwpid;
    491 	mutex_exit(proc0.p_lock);
    492 
    493 	return retid;
    494 }
    495 
    496 int
    497 rump_module_init(struct modinfo *mi, prop_dictionary_t props)
    498 {
    499 
    500 	if (!module_compatible(mi->mi_version, __NetBSD_Version__))
    501 		return EPROGMISMATCH;
    502 
    503 	return mi->mi_modcmd(MODULE_CMD_INIT, props);
    504 }
    505 
    506 int
    507 rump_module_fini(struct modinfo *mi)
    508 {
    509 
    510 	return mi->mi_modcmd(MODULE_CMD_FINI, NULL);
    511 }
    512 
    513 int _syspuffs_stub(int, int *);
    514 int
    515 _syspuffs_stub(int fd, int *newfd)
    516 {
    517 
    518 	return ENODEV;
    519 }
    520 __weak_alias(rump_syspuffs_glueinit,_syspuffs_stub);
    521 
    522 static int
    523 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen,
    524 	register_t *retval)
    525 {
    526 	struct lwp *l;
    527 	struct sysent *callp;
    528 
    529 	if (__predict_false(num >= SYS_NSYSENT))
    530 		return ENOSYS;
    531 
    532 	l = curlwp;
    533 	callp = rump_sysent + num;
    534 	return callp->sy_call(l, (void *)data, retval);
    535 }
    536 
    537 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local;
    538 void *rump_sysproxy_arg;
    539 
    540 /*
    541  * This whole syscall-via-rpc is still taking form.  For example, it
    542  * may be necessary to set syscalls individually instead of lobbing
    543  * them all to the same place.  So don't think this interface is
    544  * set in stone.
    545  */
    546 int
    547 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg)
    548 {
    549 
    550 	if (rump_sysproxy_arg)
    551 		return EBUSY;
    552 
    553 	rump_sysproxy_arg = arg;
    554 	rump_sysproxy = proxy;
    555 
    556 	return 0;
    557 }
    558 
    559 int
    560 rump_getversion()
    561 {
    562 
    563 	return __NetBSD_Version__;
    564 }
    565