xref: /netbsd-src/sys/rump/librump/rumpkern/rump.c (revision 28ef056fc275729e9327322d9102bc5d9934cf36)
1 /*	$NetBSD: rump.c,v 1.155 2010/03/05 18:41:46 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 1.155 2010/03/05 18:41:46 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/device.h>
40 #include <sys/evcnt.h>
41 #include <sys/event.h>
42 #include <sys/exec_elf.h>
43 #include <sys/filedesc.h>
44 #include <sys/iostat.h>
45 #include <sys/kauth.h>
46 #include <sys/kernel.h>
47 #include <sys/kmem.h>
48 #include <sys/kprintf.h>
49 #include <sys/ksyms.h>
50 #include <sys/msgbuf.h>
51 #include <sys/module.h>
52 #include <sys/once.h>
53 #include <sys/percpu.h>
54 #include <sys/pipe.h>
55 #include <sys/queue.h>
56 #include <sys/reboot.h>
57 #include <sys/resourcevar.h>
58 #include <sys/select.h>
59 #include <sys/sysctl.h>
60 #include <sys/syscall.h>
61 #include <sys/tty.h>
62 #include <sys/uidinfo.h>
63 #include <sys/vmem.h>
64 #include <sys/xcall.h>
65 
66 #include <rump/rumpuser.h>
67 
68 #include <secmodel/suser/suser.h>
69 
70 #include <prop/proplib.h>
71 
72 #include <uvm/uvm_readahead.h>
73 
74 #include "rump_private.h"
75 #include "rump_net_private.h"
76 #include "rump_vfs_private.h"
77 #include "rump_dev_private.h"
78 
79 struct proc proc0;
80 struct session rump_session = {
81 	.s_count = 1,
82 	.s_flags = 0,
83 	.s_leader = &proc0,
84 	.s_login = "rumphobo",
85 	.s_sid = 0,
86 };
87 struct pgrp rump_pgrp = {
88 	.pg_members = LIST_HEAD_INITIALIZER(pg_members),
89 	.pg_session = &rump_session,
90 	.pg_jobc = 1,
91 };
92 struct pstats rump_stats;
93 struct plimit rump_limits;
94 struct filedesc rump_filedesc0;
95 struct proclist allproc;
96 char machine[] = "rump";
97 static kauth_cred_t rump_susercred;
98 
99 /* pretend the master rump proc is init */
100 struct proc *initproc = &proc0;
101 
102 struct rumpuser_mtx *rump_giantlock;
103 
104 sigset_t sigcantmask;
105 
106 struct device rump_rootdev = {
107 	.dv_class = DV_VIRTUAL
108 };
109 
110 #ifdef RUMP_WITHOUT_THREADS
111 int rump_threads = 0;
112 #else
113 int rump_threads = 1;
114 #endif
115 
116 static void
117 rump_aiodone_worker(struct work *wk, void *dummy)
118 {
119 	struct buf *bp = (struct buf *)wk;
120 
121 	KASSERT(&bp->b_work == wk);
122 	bp->b_iodone(bp);
123 }
124 
125 static int rump_inited;
126 static struct emul emul_rump = {
127 	.e_vm_default_addr = uvm_default_mapaddr,
128 };
129 
130 int rump__unavailable(void);
131 int rump__unavailable() {return EOPNOTSUPP;}
132 __weak_alias(rump_net_init,rump__unavailable);
133 __weak_alias(rump_vfs_init,rump__unavailable);
134 __weak_alias(rump_dev_init,rump__unavailable);
135 
136 __weak_alias(rump_vfs_fini,rump__unavailable);
137 
138 __weak_alias(biodone,rump__unavailable);
139 __weak_alias(sopoll,rump__unavailable);
140 
141 void rump__unavailable_vfs_panic(void);
142 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
143 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
144 
145 rump_proc_vfs_init_fn rump_proc_vfs_init;
146 rump_proc_vfs_release_fn rump_proc_vfs_release;
147 
148 static void add_linkedin_modules(const struct modinfo *const *, size_t);
149 
150 static void __noinline
151 messthestack(void)
152 {
153 	volatile uint32_t mess[64];
154 	uint64_t d1, d2;
155 	int i, error;
156 
157 	for (i = 0; i < 64; i++) {
158 		rumpuser_gettime(&d1, &d2, &error);
159 		mess[i] = d2;
160 	}
161 }
162 
163 int
164 rump__init(int rump_version)
165 {
166 	char buf[256];
167 	struct proc *p;
168 	struct lwp *l;
169 	int i;
170 	int error;
171 
172 	/* not reentrant */
173 	if (rump_inited)
174 		return 0;
175 	else if (rump_inited == -1)
176 		panic("rump_init: host process restart required");
177 	else
178 		rump_inited = 1;
179 
180 	if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) {
181 		if (*buf != '0')
182 			boothowto = AB_VERBOSE;
183 	}
184 
185 	/* Print some silly banners for spammy bootstrap. */
186 	if (boothowto & AB_VERBOSE) {
187 		printf("%s%s", copyright, version);
188 	}
189 
190 	/*
191 	 * Seed arc4random() with a "reasonable" amount of randomness.
192 	 * Yes, this is a quick kludge which depends on the arc4random
193 	 * implementation.
194 	 */
195 	messthestack();
196 	arc4random();
197 
198 	if (rump_version != RUMP_VERSION) {
199 		printf("rump version mismatch, %d vs. %d\n",
200 		    rump_version, RUMP_VERSION);
201 		return EPROGMISMATCH;
202 	}
203 
204 	if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
205 		rump_threads = *buf != '0';
206 	}
207 	rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
208 	    rump_threads);
209 	rump_intr_init();
210 
211 	/* init minimal lwp/cpu context */
212 	l = &lwp0;
213 	l->l_lid = 1;
214 	l->l_cpu = rump_cpu;
215 	rumpuser_set_curlwp(l);
216 
217 	mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE);
218 	rumpuser_mutex_recursive_init(&rump_giantlock);
219 	ksyms_init();
220 	rumpvm_init();
221 	evcnt_init();
222 
223 	once_init();
224 	prop_kern_init();
225 
226 	pool_subsystem_init();
227 	kmem_init();
228 
229 	uvm_ra_init();
230 
231 	mutex_obj_init();
232 	callout_startup();
233 
234 	kprintf_init();
235 	loginit();
236 
237 	kauth_init();
238 	rump_susercred = rump_cred_create(0, 0, 0, NULL);
239 
240 	/* init proc0 and rest of lwp0 now that we can allocate memory */
241 	p = &proc0;
242 	p->p_stats = &rump_stats;
243 	p->p_limit = &rump_limits;
244 	p->p_pgrp = &rump_pgrp;
245 	p->p_pid = 0;
246 	p->p_fd = &rump_filedesc0;
247 	p->p_vmspace = &rump_vmspace;
248 	p->p_emul = &emul_rump;
249 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
250 	l->l_cred = rump_cred_suserget();
251 	l->l_proc = p;
252 	LIST_INIT(&allproc);
253 	LIST_INSERT_HEAD(&allproc, &proc0, p_list);
254 	proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
255 
256 	rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
257 	rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY;
258 	rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY;
259 
260 	rump_scheduler_init();
261 	/* revert temporary context and schedule a real context */
262 	l->l_cpu = NULL;
263 	rumpuser_set_curlwp(NULL);
264 	rump_schedule();
265 
266 	percpu_init();
267 
268 	/* we are mostly go.  do per-cpu subsystem init */
269 	for (i = 0; i < ncpu; i++) {
270 		struct cpu_info *ci = cpu_lookup(i);
271 
272 		callout_init_cpu(ci);
273 		softint_init(ci);
274 		xc_init_cpu(ci);
275 		pool_cache_cpu_init(ci);
276 		selsysinit(ci);
277 		percpu_init_cpu(ci);
278 	}
279 
280 	sysctl_init();
281 	kqueue_init();
282 	iostat_init();
283 	uid_init();
284 	fd_sys_init();
285 	module_init();
286 	devsw_init();
287 	pipe_init();
288 
289 	rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load);
290 
291 	/* these do nothing if not present */
292 	rump_vfs_init();
293 	rump_net_init();
294 	rump_dev_init();
295 	cold = 0;
296 
297 	/* aieeeedondest */
298 	if (rump_threads) {
299 		if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
300 		    rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
301 			panic("aiodoned");
302 	}
303 
304 	sysctl_finalize();
305 
306 	module_init_class(MODULE_CLASS_ANY);
307 
308 	rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
309 	hostnamelen = strlen(hostname);
310 
311 	sigemptyset(&sigcantmask);
312 
313 	lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0);
314 
315 	if (rump_threads)
316 		vmem_rehash_start();
317 
318 	rump_unschedule();
319 
320 	return 0;
321 }
322 
323 /* maybe support sys_reboot some day for remote shutdown */
324 void
325 rump_reboot(int howto)
326 {
327 
328 	/* dump means we really take the dive here */
329 	if ((howto & RB_DUMP) || panicstr) {
330 		rumpuser_exit(RUMPUSER_PANIC);
331 		/*NOTREACHED*/
332 	}
333 
334 	/* try to sync */
335 	if (!((howto & RB_NOSYNC) || panicstr)) {
336 		rump_vfs_fini();
337 	}
338 
339 	/* your wish is my command */
340 	if (howto & RB_HALT) {
341 		for (;;) {
342 			uint64_t sec = 5, nsec = 0;
343 			int error;
344 
345 			rumpuser_nanosleep(&sec, &nsec, &error);
346 		}
347 	}
348 	rump_inited = -1;
349 }
350 
351 struct uio *
352 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
353 {
354 	struct uio *uio;
355 	enum uio_rw uiorw;
356 
357 	switch (rw) {
358 	case RUMPUIO_READ:
359 		uiorw = UIO_READ;
360 		break;
361 	case RUMPUIO_WRITE:
362 		uiorw = UIO_WRITE;
363 		break;
364 	default:
365 		panic("%s: invalid rw %d", __func__, rw);
366 	}
367 
368 	uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
369 	uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
370 
371 	uio->uio_iov->iov_base = buf;
372 	uio->uio_iov->iov_len = bufsize;
373 
374 	uio->uio_iovcnt = 1;
375 	uio->uio_offset = offset;
376 	uio->uio_resid = bufsize;
377 	uio->uio_rw = uiorw;
378 	uio->uio_vmspace = UIO_VMSPACE_SYS;
379 
380 	return uio;
381 }
382 
383 size_t
384 rump_uio_getresid(struct uio *uio)
385 {
386 
387 	return uio->uio_resid;
388 }
389 
390 off_t
391 rump_uio_getoff(struct uio *uio)
392 {
393 
394 	return uio->uio_offset;
395 }
396 
397 size_t
398 rump_uio_free(struct uio *uio)
399 {
400 	size_t resid;
401 
402 	resid = uio->uio_resid;
403 	kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
404 	kmem_free(uio, sizeof(*uio));
405 
406 	return resid;
407 }
408 
409 static pid_t nextpid = 1;
410 struct lwp *
411 rump_newproc_switch()
412 {
413 	struct lwp *l;
414 	pid_t mypid;
415 
416 	mypid = atomic_inc_uint_nv(&nextpid);
417 	if (__predict_false(mypid == 0))
418 		mypid = atomic_inc_uint_nv(&nextpid);
419 
420 	l = rump_lwp_alloc(mypid, 0);
421 	rump_lwp_switch(l);
422 
423 	return l;
424 }
425 
426 struct lwp *
427 rump_lwp_alloc_and_switch(pid_t pid, lwpid_t lid)
428 {
429 	struct lwp *l;
430 
431 	l = rump_lwp_alloc(pid, lid);
432 	rump_lwp_switch(l);
433 
434 	return l;
435 }
436 
437 struct lwp *
438 rump_lwp_alloc(pid_t pid, lwpid_t lid)
439 {
440 	struct lwp *l;
441 	struct proc *p;
442 
443 	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
444 	if (pid != 0) {
445 		p = kmem_zalloc(sizeof(*p), KM_SLEEP);
446 		if (rump_proc_vfs_init)
447 			rump_proc_vfs_init(p);
448 		p->p_stats = &rump_stats;
449 		p->p_limit = &rump_limits;
450 		p->p_pid = pid;
451 		p->p_vmspace = &rump_vmspace;
452 		p->p_emul = &emul_rump;
453 		p->p_fd = fd_init(NULL);
454 		p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
455 		l->l_cred = rump_cred_suserget();
456 	} else {
457 		p = &proc0;
458 		l->l_cred = rump_susercred;
459 	}
460 
461 	l->l_proc = p;
462 	l->l_lid = lid;
463 	l->l_fd = p->p_fd;
464 	l->l_cpu = NULL;
465 
466 	return l;
467 }
468 
469 void
470 rump_lwp_switch(struct lwp *newlwp)
471 {
472 	struct lwp *l = curlwp;
473 
474 	rumpuser_set_curlwp(NULL);
475 	newlwp->l_cpu = l->l_cpu;
476 	newlwp->l_mutex = l->l_mutex;
477 	l->l_mutex = NULL;
478 	l->l_cpu = NULL;
479 	rumpuser_set_curlwp(newlwp);
480 	if (l->l_flag & LW_WEXIT)
481 		rump_lwp_free(l);
482 }
483 
484 /* XXX: this has effect only on non-pid0 lwps */
485 void
486 rump_lwp_release(struct lwp *l)
487 {
488 	struct proc *p;
489 
490 	p = l->l_proc;
491 	if (p->p_pid != 0) {
492 		mutex_obj_free(p->p_lock);
493 		fd_free();
494 		if (rump_proc_vfs_release)
495 			rump_proc_vfs_release(p);
496 		rump_cred_put(l->l_cred);
497 		kmem_free(p, sizeof(*p));
498 	}
499 	KASSERT((l->l_flag & LW_WEXIT) == 0);
500 	l->l_flag |= LW_WEXIT;
501 }
502 
503 void
504 rump_lwp_free(struct lwp *l)
505 {
506 
507 	KASSERT(l->l_flag & LW_WEXIT);
508 	KASSERT(l->l_mutex == NULL);
509 	if (l->l_name)
510 		kmem_free(l->l_name, MAXCOMLEN);
511 	kmem_free(l, sizeof(*l));
512 }
513 
514 struct lwp *
515 rump_lwp_curlwp(void)
516 {
517 	struct lwp *l = curlwp;
518 
519 	if (l->l_flag & LW_WEXIT)
520 		return NULL;
521 	return l;
522 }
523 
524 /* rump private.  NEEDS WORK! */
525 void
526 rump_set_vmspace(struct vmspace *vm)
527 {
528 	struct proc *p = curproc;
529 
530 	p->p_vmspace = vm;
531 }
532 
533 kauth_cred_t
534 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
535 {
536 	kauth_cred_t cred;
537 	int rv;
538 
539 	cred = kauth_cred_alloc();
540 	kauth_cred_setuid(cred, uid);
541 	kauth_cred_seteuid(cred, uid);
542 	kauth_cred_setsvuid(cred, uid);
543 	kauth_cred_setgid(cred, gid);
544 	kauth_cred_setgid(cred, gid);
545 	kauth_cred_setegid(cred, gid);
546 	kauth_cred_setsvgid(cred, gid);
547 	rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
548 	/* oh this is silly.  and by "this" I mean kauth_cred_setgroups() */
549 	assert(rv == 0);
550 
551 	return cred;
552 }
553 
554 void
555 rump_cred_put(kauth_cred_t cred)
556 {
557 
558 	kauth_cred_free(cred);
559 }
560 
561 kauth_cred_t
562 rump_cred_suserget(void)
563 {
564 
565 	kauth_cred_hold(rump_susercred);
566 	return rump_susercred;
567 }
568 
569 /*
570  * Return the next system lwpid
571  */
572 lwpid_t
573 rump_nextlid(void)
574 {
575 	lwpid_t retid;
576 
577 	mutex_enter(proc0.p_lock);
578 	/*
579 	 * Take next one, don't return 0
580 	 * XXX: most likely we'll have collisions in case this
581 	 * wraps around.
582 	 */
583 	if (++proc0.p_nlwpid == 0)
584 		++proc0.p_nlwpid;
585 	retid = proc0.p_nlwpid;
586 	mutex_exit(proc0.p_lock);
587 
588 	return retid;
589 }
590 
591 static int compcounter[RUMP_COMPONENT_MAX];
592 
593 static void
594 rump_component_init_cb(struct rump_component *rc, int type)
595 {
596 
597 	KASSERT(type < RUMP_COMPONENT_MAX);
598 	if (rc->rc_type == type) {
599 		rc->rc_init();
600 		compcounter[type]++;
601 	}
602 }
603 
604 int
605 rump_component_count(enum rump_component_type type)
606 {
607 
608 	KASSERT(type <= RUMP_COMPONENT_MAX);
609 	return compcounter[type];
610 }
611 
612 void
613 rump_component_init(enum rump_component_type type)
614 {
615 
616 	rumpuser_dl_component_init(type, rump_component_init_cb);
617 }
618 
619 /*
620  * Initialize a module which has already been loaded and linked
621  * with dlopen(). This is fundamentally the same as a builtin module.
622  */
623 int
624 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
625 {
626 
627 	return module_builtin_add(mip, nmodinfo, true);
628 }
629 
630 /*
631  * Finish module (flawless victory, fatality!).
632  */
633 int
634 rump_module_fini(const struct modinfo *mi)
635 {
636 
637 	return module_builtin_remove(mi, true);
638 }
639 
640 /*
641  * Add loaded and linked module to the builtin list.  It will
642  * later be initialized with module_init_class().
643  */
644 
645 static void
646 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
647 {
648 
649 	module_builtin_add(mip, nmodinfo, false);
650 }
651 
652 int
653 rump_kernelfsym_load(void *symtab, uint64_t symsize,
654 	char *strtab, uint64_t strsize)
655 {
656 	static int inited = 0;
657 	Elf64_Ehdr ehdr;
658 
659 	if (inited)
660 		return EBUSY;
661 	inited = 1;
662 
663 	/*
664 	 * Use 64bit header since it's bigger.  Shouldn't make a
665 	 * difference, since we're passing in all zeroes anyway.
666 	 */
667 	memset(&ehdr, 0, sizeof(ehdr));
668 	ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
669 
670 	return 0;
671 }
672 
673 static int
674 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen,
675 	register_t *retval)
676 {
677 	struct lwp *l;
678 	struct sysent *callp;
679 	int rv;
680 
681 	if (__predict_false(num >= SYS_NSYSENT))
682 		return ENOSYS;
683 
684 	callp = rump_sysent + num;
685 	rump_schedule();
686 	l = curlwp;
687 	rv = callp->sy_call(l, (void *)data, retval);
688 	rump_unschedule();
689 
690 	return rv;
691 }
692 
693 int
694 rump_boot_gethowto()
695 {
696 
697 	return boothowto;
698 }
699 
700 void
701 rump_boot_sethowto(int howto)
702 {
703 
704 	boothowto = howto;
705 }
706 
707 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local;
708 void *rump_sysproxy_arg;
709 
710 /*
711  * This whole syscall-via-rpc is still taking form.  For example, it
712  * may be necessary to set syscalls individually instead of lobbing
713  * them all to the same place.  So don't think this interface is
714  * set in stone.
715  */
716 int
717 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg)
718 {
719 
720 	if (rump_sysproxy_arg)
721 		return EBUSY;
722 
723 	rump_sysproxy_arg = arg;
724 	rump_sysproxy = proxy;
725 
726 	return 0;
727 }
728 
729 int
730 rump_getversion(void)
731 {
732 
733 	return __NetBSD_Version__;
734 }
735