xref: /netbsd-src/sys/rump/librump/rumpkern/rump.c (revision f75f5aae154fcd0572e8889e4fea2a51d67bbf08)
1 /*	$NetBSD: rump.c,v 1.121 2009/10/09 14:41:36 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 1.121 2009/10/09 14:41:36 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 cpu_info rump_cpu;
89 struct filedesc rump_filedesc0;
90 struct proclist allproc;
91 char machine[] = "rump";
92 static kauth_cred_t rump_susercred;
93 
94 /* pretend the master rump proc is init */
95 struct proc *initproc = &proc0;
96 
97 struct rumpuser_mtx *rump_giantlock;
98 
99 sigset_t sigcantmask;
100 
101 #ifdef RUMP_WITHOUT_THREADS
102 int rump_threads = 0;
103 #else
104 int rump_threads = 1;
105 #endif
106 
107 static void
108 rump_aiodone_worker(struct work *wk, void *dummy)
109 {
110 	struct buf *bp = (struct buf *)wk;
111 
112 	KASSERT(&bp->b_work == wk);
113 	bp->b_iodone(bp);
114 }
115 
116 static int rump_inited;
117 static struct emul emul_rump;
118 
119 void rump__unavailable(void);
120 void rump__unavailable() {}
121 __weak_alias(rump_net_init,rump__unavailable);
122 __weak_alias(rump_vfs_init,rump__unavailable);
123 __weak_alias(rump_dev_init,rump__unavailable);
124 
125 __weak_alias(rump_vfs_fini,rump__unavailable);
126 
127 __weak_alias(biodone,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 
137 static void
138 pvfsinit_nop(struct proc *p)
139 {
140 
141 	return;
142 }
143 
144 static void
145 pvfsrele_nop(struct proc *p)
146 {
147 
148 	return;
149 }
150 
151 rump_proc_vfs_init_fn rump_proc_vfs_init = pvfsinit_nop;
152 rump_proc_vfs_release_fn rump_proc_vfs_release = pvfsrele_nop;
153 
154 /*
155  * Stir up the stack a bit.  These are exported functions to help
156  * convince the compiler that we don't want these routines completely
157  * optimized out or inlined.  Is there an easier way to do this?
158  */
159 void nullfn(uint32_t *);
160 void nullfn(uint32_t *arg){}
161 void messthestack(void);
162 void
163 messthestack(void)
164 {
165 	uint32_t mess[64];
166 	uint64_t d1, d2;
167 	int i, error;
168 
169 	for (i = 0; i < 64; i++) {
170 		rumpuser_gettime(&d1, &d2, &error);
171 		mess[i] = d2;
172 	}
173 	nullfn(mess);
174 }
175 
176 int
177 rump__init(int rump_version)
178 {
179 	char buf[256];
180 	struct proc *p;
181 	struct lwp *l;
182 	int error;
183 
184 	/* not reentrant */
185 	if (rump_inited)
186 		return 0;
187 	else if (rump_inited == -1)
188 		panic("rump_init: host process restart required");
189 	else
190 		rump_inited = 1;
191 
192 	/*
193 	 * Seed arc4random() with a "reasonable" amount of randomness.
194 	 * Yes, this is a quick kludge which depends on the arc4random
195 	 * implementation.
196 	 */
197 	messthestack();
198 	arc4random();
199 
200 	if (rump_version != RUMP_VERSION) {
201 		printf("rump version mismatch, %d vs. %d\n",
202 		    rump_version, RUMP_VERSION);
203 		return EPROGMISMATCH;
204 	}
205 
206 	if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
207 		rump_threads = *buf != '0';
208 	}
209 	rumpuser_thrinit(_kernel_lock, _kernel_unlock, rump_threads);
210 
211 	mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE);
212 	rumpuser_mutex_recursive_init(&rump_giantlock);
213 	ksyms_init();
214 	rumpvm_init();
215 	evcnt_init();
216 
217 	once_init();
218 	prop_kern_init();
219 
220 	rump_sleepers_init();
221 
222 	pool_subsystem_init();
223 	kmem_init();
224 
225 	kprintf_init();
226 	loginit();
227 
228 	kauth_init();
229 	rump_susercred = rump_cred_create(0, 0, 0, NULL);
230 
231 	l = &lwp0;
232 	p = &proc0;
233 	p->p_stats = &rump_stats;
234 	p->p_limit = &rump_limits;
235 	p->p_pgrp = &rump_pgrp;
236 	p->p_pid = 0;
237 	p->p_fd = &rump_filedesc0;
238 	p->p_vmspace = &rump_vmspace;
239 	p->p_emul = &emul_rump;
240 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
241 	l->l_cred = rump_cred_suserget();
242 	l->l_proc = p;
243 	l->l_lid = 1;
244 	l->l_cpu = &rump_cpu;
245 	LIST_INIT(&allproc);
246 	LIST_INSERT_HEAD(&allproc, &proc0, p_list);
247 	proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
248 
249 	rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
250 	rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY;
251 	rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY;
252 
253 	callout_startup();
254 	callout_init_cpu(&rump_cpu);
255 
256 	sysctl_init();
257 	kqueue_init();
258 	iostat_init();
259 	uid_init();
260 	percpu_init();
261 	fd_sys_init();
262 	module_init();
263 	softint_init(&rump_cpu);
264 	devsw_init();
265 
266 	/* these do nothing if not present */
267 	rump_vfs_init();
268 	rump_net_init();
269 	rump_dev_init();
270 	cold = 0;
271 
272 	/* aieeeedondest */
273 	if (rump_threads) {
274 		if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
275 		    rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
276 			panic("aiodoned");
277 	}
278 
279 	sysctl_finalize();
280 
281 	rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
282 	hostnamelen = strlen(hostname);
283 
284 	sigemptyset(&sigcantmask);
285 
286 	lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0);
287 
288 #ifdef RUMP_USE_REAL_ALLOCATORS
289 	if (rump_threads)
290 		vmem_rehash_start();
291 #endif
292 
293 	rumpuser_dl_module_bootstrap();
294 
295 	return 0;
296 }
297 
298 /* maybe support sys_reboot some day for remote shutdown */
299 void
300 rump_reboot(int howto)
301 {
302 
303 	/* dump means we really take the dive here */
304 	if ((howto & RB_DUMP) || panicstr) {
305 		rumpuser_exit(RUMPUSER_PANIC);
306 		/*NOTREACHED*/
307 	}
308 
309 	/* try to sync */
310 	if (!((howto & RB_NOSYNC) || panicstr)) {
311 		rump_vfs_fini();
312 	}
313 
314 	/* your wish is my command */
315 	if (howto & RB_HALT) {
316 		for (;;) {
317 			uint64_t sec = 5, nsec = 0;
318 			int error;
319 
320 			rumpuser_nanosleep(&sec, &nsec, &error);
321 		}
322 	}
323 	rump_inited = -1;
324 }
325 
326 struct uio *
327 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
328 {
329 	struct uio *uio;
330 	enum uio_rw uiorw;
331 
332 	switch (rw) {
333 	case RUMPUIO_READ:
334 		uiorw = UIO_READ;
335 		break;
336 	case RUMPUIO_WRITE:
337 		uiorw = UIO_WRITE;
338 		break;
339 	default:
340 		panic("%s: invalid rw %d", __func__, rw);
341 	}
342 
343 	uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
344 	uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
345 
346 	uio->uio_iov->iov_base = buf;
347 	uio->uio_iov->iov_len = bufsize;
348 
349 	uio->uio_iovcnt = 1;
350 	uio->uio_offset = offset;
351 	uio->uio_resid = bufsize;
352 	uio->uio_rw = uiorw;
353 	uio->uio_vmspace = UIO_VMSPACE_SYS;
354 
355 	return uio;
356 }
357 
358 size_t
359 rump_uio_getresid(struct uio *uio)
360 {
361 
362 	return uio->uio_resid;
363 }
364 
365 off_t
366 rump_uio_getoff(struct uio *uio)
367 {
368 
369 	return uio->uio_offset;
370 }
371 
372 size_t
373 rump_uio_free(struct uio *uio)
374 {
375 	size_t resid;
376 
377 	resid = uio->uio_resid;
378 	kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
379 	kmem_free(uio, sizeof(*uio));
380 
381 	return resid;
382 }
383 
384 /* public interface */
385 static pid_t nextpid = 1;
386 struct lwp *
387 rump_newproc_switch()
388 {
389 	struct lwp *oldlwp = curlwp;
390 	pid_t mypid;
391 
392 	mypid = atomic_inc_uint_nv(&nextpid);
393 	if (__predict_false(mypid == 0))
394 		mypid = atomic_inc_uint_nv(&nextpid);
395 
396 	rumpuser_set_curlwp(NULL);
397 	rump_setup_curlwp(mypid, 0, 1);
398 
399 	return oldlwp;
400 }
401 
402 /* rump private */
403 struct lwp *
404 rump_setup_curlwp(pid_t pid, lwpid_t lid, int set)
405 {
406 	struct lwp *l;
407 	struct proc *p;
408 
409 	l = kmem_zalloc(sizeof(struct lwp), KM_SLEEP);
410 	if (pid != 0) {
411 		p = kmem_zalloc(sizeof(struct proc), KM_SLEEP);
412 		rump_proc_vfs_init(p);
413 		p->p_stats = &rump_stats;
414 		p->p_limit = &rump_limits;
415 		p->p_pid = pid;
416 		p->p_vmspace = &rump_vmspace;
417 		p->p_fd = fd_init(NULL);
418 		p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
419 	} else {
420 		p = &proc0;
421 	}
422 
423 	l->l_cred = rump_cred_suserget();
424 	l->l_proc = p;
425 	l->l_lid = lid;
426 	l->l_fd = p->p_fd;
427 	l->l_mutex = RUMP_LMUTEX_MAGIC;
428 	l->l_cpu = &rump_cpu;
429 
430 	if (set)
431 		rumpuser_set_curlwp(l);
432 
433 	return l;
434 }
435 
436 /* rump private.  NEEDS WORK! */
437 void
438 rump_set_vmspace(struct vmspace *vm)
439 {
440 	struct proc *p = curproc;
441 
442 	p->p_vmspace = vm;
443 }
444 
445 void
446 rump_clear_curlwp(void)
447 {
448 	struct lwp *l;
449 	struct proc *p;
450 
451 	l = rumpuser_get_curlwp();
452 	p = l->l_proc;
453 	if (p->p_pid != 0) {
454 		mutex_obj_free(p->p_lock);
455 		fd_free();
456 		rump_proc_vfs_release(p);
457 		rump_cred_put(l->l_cred);
458 		kmem_free(p, sizeof(*p));
459 	}
460 	kmem_free(l, sizeof(*l));
461 	rumpuser_set_curlwp(NULL);
462 }
463 
464 struct lwp *
465 rump_get_curlwp(void)
466 {
467 	struct lwp *l;
468 
469 	l = rumpuser_get_curlwp();
470 	if (l == NULL)
471 		l = &lwp0;
472 
473 	return l;
474 }
475 
476 void
477 rump_set_curlwp(struct lwp *l)
478 {
479 
480 	/* clear current */
481 	rumpuser_set_curlwp(NULL);
482 	/* set new */
483 	rumpuser_set_curlwp(l);
484 }
485 
486 kauth_cred_t
487 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
488 {
489 	kauth_cred_t cred;
490 	int rv;
491 
492 	cred = kauth_cred_alloc();
493 	kauth_cred_setuid(cred, uid);
494 	kauth_cred_seteuid(cred, uid);
495 	kauth_cred_setsvuid(cred, uid);
496 	kauth_cred_setgid(cred, gid);
497 	kauth_cred_setgid(cred, gid);
498 	kauth_cred_setegid(cred, gid);
499 	kauth_cred_setsvgid(cred, gid);
500 	rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
501 	/* oh this is silly.  and by "this" I mean kauth_cred_setgroups() */
502 	assert(rv == 0);
503 
504 	return cred;
505 }
506 
507 void
508 rump_cred_put(kauth_cred_t cred)
509 {
510 
511 	kauth_cred_free(cred);
512 }
513 
514 kauth_cred_t
515 rump_cred_suserget(void)
516 {
517 
518 	kauth_cred_hold(rump_susercred);
519 	return rump_susercred;
520 }
521 
522 /*
523  * Return the next system lwpid
524  */
525 lwpid_t
526 rump_nextlid(void)
527 {
528 	lwpid_t retid;
529 
530 	mutex_enter(proc0.p_lock);
531 	/*
532 	 * Take next one, don't return 0
533 	 * XXX: most likely we'll have collisions in case this
534 	 * wraps around.
535 	 */
536 	if (++proc0.p_nlwpid == 0)
537 		++proc0.p_nlwpid;
538 	retid = proc0.p_nlwpid;
539 	mutex_exit(proc0.p_lock);
540 
541 	return retid;
542 }
543 
544 int
545 rump_module_init(struct modinfo *mi, prop_dictionary_t props)
546 {
547 	int rv;
548 
549 	if (!module_compatible(mi->mi_version, __NetBSD_Version__))
550 		return EPROGMISMATCH;
551 
552 	rv = mi->mi_modcmd(MODULE_CMD_INIT, props);
553 	if (rv == 0 && mi->mi_class == MODULE_CLASS_SECMODEL)
554 		secmodel_register();
555 
556 	return rv;
557 }
558 
559 int
560 rump_module_fini(struct modinfo *mi)
561 {
562 	int rv;
563 
564 	rv = mi->mi_modcmd(MODULE_CMD_FINI, NULL);
565 	if (rv == 0 && mi->mi_class == MODULE_CLASS_SECMODEL)
566 		secmodel_deregister();
567 
568 	return rv;
569 }
570 
571 int _syspuffs_stub(int, int *);
572 int
573 _syspuffs_stub(int fd, int *newfd)
574 {
575 
576 	return ENODEV;
577 }
578 __weak_alias(rump_syspuffs_glueinit,_syspuffs_stub);
579 
580 static int
581 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen,
582 	register_t *retval)
583 {
584 	struct lwp *l;
585 	struct sysent *callp;
586 
587 	if (__predict_false(num >= SYS_NSYSENT))
588 		return ENOSYS;
589 
590 	l = curlwp;
591 	callp = rump_sysent + num;
592 	return callp->sy_call(l, (void *)data, retval);
593 }
594 
595 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local;
596 void *rump_sysproxy_arg;
597 
598 /*
599  * This whole syscall-via-rpc is still taking form.  For example, it
600  * may be necessary to set syscalls individually instead of lobbing
601  * them all to the same place.  So don't think this interface is
602  * set in stone.
603  */
604 int
605 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg)
606 {
607 
608 	if (rump_sysproxy_arg)
609 		return EBUSY;
610 
611 	rump_sysproxy_arg = arg;
612 	rump_sysproxy = proxy;
613 
614 	return 0;
615 }
616 
617 int
618 rump_getversion()
619 {
620 
621 	return __NetBSD_Version__;
622 }
623