rump.c revision 1.185 1 /* $NetBSD: rump.c,v 1.185 2010/09/06 20:10:20 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 1.185 2010/09/06 20:10:20 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 /* is this still necessary or use kern_proc stuff? */
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
101 char machine[] = MACHINE;
102 static kauth_cred_t rump_susercred;
103
104 /* pretend the master rump proc is init */
105 struct proc *initproc = &proc0;
106
107 struct rumpuser_mtx *rump_giantlock;
108
109 struct device rump_rootdev = {
110 .dv_class = DV_VIRTUAL
111 };
112
113 #ifdef RUMP_WITHOUT_THREADS
114 int rump_threads = 0;
115 #else
116 int rump_threads = 1;
117 #endif
118
119 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */
120
121 static void
122 rump_aiodone_worker(struct work *wk, void *dummy)
123 {
124 struct buf *bp = (struct buf *)wk;
125
126 KASSERT(&bp->b_work == wk);
127 bp->b_iodone(bp);
128 }
129
130 static int rump_inited;
131
132 /*
133 * Make sure pnbuf_cache is available even without vfs
134 */
135 struct pool_cache *pnbuf_cache;
136 int rump_initpnbufpool(void);
137 int rump_initpnbufpool(void)
138 {
139
140 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
141 NULL, IPL_NONE, NULL, NULL, NULL);
142 return EOPNOTSUPP;
143 }
144
145 int rump__unavailable(void);
146 int rump__unavailable() {return EOPNOTSUPP;}
147 __weak_alias(rump_net_init,rump__unavailable);
148 __weak_alias(rump_vfs_init,rump_initpnbufpool);
149 __weak_alias(rump_dev_init,rump__unavailable);
150
151 __weak_alias(rump_vfs_fini,rump__unavailable);
152
153 __weak_alias(biodone,rump__unavailable);
154 __weak_alias(sopoll,rump__unavailable);
155
156 void rump__unavailable_vfs_panic(void);
157 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
158 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
159
160 rump_proc_vfs_init_fn rump_proc_vfs_init;
161 rump_proc_vfs_release_fn rump_proc_vfs_release;
162
163 static void add_linkedin_modules(const struct modinfo *const *, size_t);
164
165 static void __noinline
166 messthestack(void)
167 {
168 volatile uint32_t mess[64];
169 uint64_t d1, d2;
170 int i, error;
171
172 for (i = 0; i < 64; i++) {
173 rumpuser_gettime(&d1, &d2, &error);
174 mess[i] = d2;
175 }
176 }
177
178 /*
179 * Create kern.hostname. why only this you ask. well, init_sysctl
180 * is a kitchen sink in need of some gardening. but i want to use
181 * kern.hostname today.
182 */
183 static void
184 mksysctls(void)
185 {
186
187 sysctl_createv(NULL, 0, NULL, NULL,
188 CTLFLAG_PERMANENT, CTLTYPE_NODE, "kern", NULL,
189 NULL, 0, NULL, 0, CTL_KERN, CTL_EOL);
190
191 /* XXX: setting hostnamelen is missing */
192 sysctl_createv(NULL, 0, NULL, NULL,
193 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_STRING, "hostname",
194 SYSCTL_DESCR("System hostname"), NULL, 0,
195 &hostname, MAXHOSTNAMELEN, CTL_KERN, KERN_HOSTNAME, CTL_EOL);
196 }
197
198 int
199 rump__init(int rump_version)
200 {
201 char buf[256];
202 struct timespec ts;
203 uint64_t sec, nsec;
204 struct lwp *l;
205 int i, numcpu;
206 int error;
207
208 /* not reentrant */
209 if (rump_inited)
210 return 0;
211 else if (rump_inited == -1)
212 panic("rump_init: host process restart required");
213 else
214 rump_inited = 1;
215
216 if (rumpuser_getversion() != RUMPUSER_VERSION) {
217 /* let's hope the ABI of rumpuser_dprintf is the same ;) */
218 rumpuser_dprintf("rumpuser version mismatch: %d vs. %d\n",
219 rumpuser_getversion(), RUMPUSER_VERSION);
220 return EPROGMISMATCH;
221 }
222
223 if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) {
224 if (*buf != '0')
225 boothowto = AB_VERBOSE;
226 }
227
228 if (rumpuser_getenv("RUMP_NCPU", buf, sizeof(buf), &error) == 0)
229 error = 0;
230 /* non-x86 is missing CPU_INFO_FOREACH() support */
231 #if defined(__i386__) || defined(__x86_64__)
232 if (error == 0) {
233 numcpu = strtoll(buf, NULL, 10);
234 if (numcpu < 1)
235 numcpu = 1;
236 } else {
237 numcpu = rumpuser_getnhostcpu();
238 }
239 #else
240 if (error == 0)
241 printf("NCPU limited to 1 on this host\n");
242 numcpu = 1;
243 #endif
244 rump_cpus_bootstrap(numcpu);
245
246 rumpuser_gettime(&sec, &nsec, &error);
247 boottime.tv_sec = sec;
248 boottime.tv_nsec = nsec;
249
250 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
251 aprint_verbose("%s%s", copyright, version);
252
253 /*
254 * Seed arc4random() with a "reasonable" amount of randomness.
255 * Yes, this is a quick kludge which depends on the arc4random
256 * implementation.
257 */
258 messthestack();
259 arc4random();
260
261 if (rump_version != RUMP_VERSION) {
262 printf("rump version mismatch, %d vs. %d\n",
263 rump_version, RUMP_VERSION);
264 return EPROGMISMATCH;
265 }
266
267 if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
268 rump_threads = *buf != '0';
269 }
270 rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
271 rump_threads);
272 rump_intr_init();
273 rump_tsleep_init();
274
275 /* init minimal lwp/cpu context */
276 l = &lwp0;
277 l->l_lid = 1;
278 l->l_cpu = l->l_target_cpu = rump_cpu;
279 l->l_fd = &filedesc0;
280 rumpuser_set_curlwp(l);
281
282 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE);
283 rumpuser_mutex_recursive_init(&rump_giantlock);
284 ksyms_init();
285 uvm_init();
286 evcnt_init();
287
288 once_init();
289 kernconfig_lock_init();
290 prop_kern_init();
291
292 pool_subsystem_init();
293 kmem_init();
294
295 uvm_ra_init();
296 uao_init();
297
298 mutex_obj_init();
299 callout_startup();
300
301 kprintf_init();
302 loginit();
303
304 kauth_init();
305 rump_susercred = rump_cred_create(0, 0, 0, NULL);
306
307 procinit();
308 proc0_init();
309
310 l->l_proc = &proc0;
311 lwp_update_creds(l);
312
313 lwpinit_specificdata();
314 lwp_initspecific(&lwp0);
315
316 rump_scheduler_init();
317 /* revert temporary context and schedule a real context */
318 l->l_cpu = NULL;
319 rumpuser_set_curlwp(NULL);
320 rump_schedule();
321
322 percpu_init();
323 inittimecounter();
324 ntp_init();
325
326 rumpuser_gettime(&sec, &nsec, &error);
327 ts.tv_sec = sec;
328 ts.tv_nsec = nsec;
329 tc_setclock(&ts);
330
331 /* we are mostly go. do per-cpu subsystem init */
332 for (i = 0; i < ncpu; i++) {
333 struct cpu_info *ci = cpu_lookup(i);
334
335 callout_init_cpu(ci);
336 softint_init(ci);
337 xc_init_cpu(ci);
338 pool_cache_cpu_init(ci);
339 selsysinit(ci);
340 percpu_init_cpu(ci);
341
342 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
343 }
344
345 sysctl_init();
346 kqueue_init();
347 iostat_init();
348 uid_init();
349 fd_sys_init();
350 module_init();
351 devsw_init();
352 pipe_init();
353 resource_init();
354
355 /* start page baroness */
356 if (rump_threads) {
357 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
358 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
359 panic("pagedaemon create failed");
360 } else
361 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
362
363 /* process dso's */
364 rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load);
365
366 rump_component_init(RUMP_COMPONENT_KERN);
367
368 /* these do nothing if not present */
369 rump_vfs_init();
370 rump_net_init();
371 rump_dev_init();
372
373 rump_component_init(RUMP_COMPONENT_KERN_VFS);
374
375 cold = 0;
376
377 /* aieeeedondest */
378 if (rump_threads) {
379 if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
380 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
381 panic("aiodoned");
382 }
383
384 mksysctls();
385 sysctl_finalize();
386
387 module_init_class(MODULE_CLASS_ANY);
388
389 rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
390 hostnamelen = strlen(hostname);
391
392 sigemptyset(&sigcantmask);
393
394 if (rump_threads)
395 vmem_rehash_start();
396
397 rump_unschedule();
398
399 return 0;
400 }
401
402 /* maybe support sys_reboot some day for remote shutdown */
403 void
404 rump_reboot(int howto)
405 {
406
407 /* dump means we really take the dive here */
408 if ((howto & RB_DUMP) || panicstr) {
409 rumpuser_exit(RUMPUSER_PANIC);
410 /*NOTREACHED*/
411 }
412
413 /* try to sync */
414 if (!((howto & RB_NOSYNC) || panicstr)) {
415 rump_vfs_fini();
416 }
417
418 /* your wish is my command */
419 if (howto & RB_HALT) {
420 for (;;) {
421 uint64_t sec = 5, nsec = 0;
422 int error;
423
424 rumpuser_nanosleep(&sec, &nsec, &error);
425 }
426 }
427 rump_inited = -1;
428 }
429
430 struct uio *
431 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
432 {
433 struct uio *uio;
434 enum uio_rw uiorw;
435
436 switch (rw) {
437 case RUMPUIO_READ:
438 uiorw = UIO_READ;
439 break;
440 case RUMPUIO_WRITE:
441 uiorw = UIO_WRITE;
442 break;
443 default:
444 panic("%s: invalid rw %d", __func__, rw);
445 }
446
447 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
448 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
449
450 uio->uio_iov->iov_base = buf;
451 uio->uio_iov->iov_len = bufsize;
452
453 uio->uio_iovcnt = 1;
454 uio->uio_offset = offset;
455 uio->uio_resid = bufsize;
456 uio->uio_rw = uiorw;
457 uio->uio_vmspace = UIO_VMSPACE_SYS;
458
459 return uio;
460 }
461
462 size_t
463 rump_uio_getresid(struct uio *uio)
464 {
465
466 return uio->uio_resid;
467 }
468
469 off_t
470 rump_uio_getoff(struct uio *uio)
471 {
472
473 return uio->uio_offset;
474 }
475
476 size_t
477 rump_uio_free(struct uio *uio)
478 {
479 size_t resid;
480
481 resid = uio->uio_resid;
482 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
483 kmem_free(uio, sizeof(*uio));
484
485 return resid;
486 }
487
488 /* rump private. NEEDS WORK! */
489 void
490 rump_set_vmspace(struct vmspace *vm)
491 {
492 struct proc *p = curproc;
493
494 p->p_vmspace = vm;
495 }
496
497 kauth_cred_t
498 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
499 {
500 kauth_cred_t cred;
501 int rv;
502
503 cred = kauth_cred_alloc();
504 kauth_cred_setuid(cred, uid);
505 kauth_cred_seteuid(cred, uid);
506 kauth_cred_setsvuid(cred, uid);
507 kauth_cred_setgid(cred, gid);
508 kauth_cred_setgid(cred, gid);
509 kauth_cred_setegid(cred, gid);
510 kauth_cred_setsvgid(cred, gid);
511 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
512 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */
513 assert(rv == 0);
514
515 return cred;
516 }
517
518 void
519 rump_cred_put(kauth_cred_t cred)
520 {
521
522 kauth_cred_free(cred);
523 }
524
525 static int compcounter[RUMP_COMPONENT_MAX];
526
527 static void
528 rump_component_init_cb(struct rump_component *rc, int type)
529 {
530
531 KASSERT(type < RUMP_COMPONENT_MAX);
532 if (rc->rc_type == type) {
533 rc->rc_init();
534 compcounter[type]++;
535 }
536 }
537
538 int
539 rump_component_count(enum rump_component_type type)
540 {
541
542 KASSERT(type <= RUMP_COMPONENT_MAX);
543 return compcounter[type];
544 }
545
546 void
547 rump_component_init(enum rump_component_type type)
548 {
549
550 rumpuser_dl_component_init(type, rump_component_init_cb);
551 }
552
553 /*
554 * Initialize a module which has already been loaded and linked
555 * with dlopen(). This is fundamentally the same as a builtin module.
556 */
557 int
558 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
559 {
560
561 return module_builtin_add(mip, nmodinfo, true);
562 }
563
564 /*
565 * Finish module (flawless victory, fatality!).
566 */
567 int
568 rump_module_fini(const struct modinfo *mi)
569 {
570
571 return module_builtin_remove(mi, true);
572 }
573
574 /*
575 * Add loaded and linked module to the builtin list. It will
576 * later be initialized with module_init_class().
577 */
578
579 static void
580 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
581 {
582
583 module_builtin_add(mip, nmodinfo, false);
584 }
585
586 int
587 rump_kernelfsym_load(void *symtab, uint64_t symsize,
588 char *strtab, uint64_t strsize)
589 {
590 static int inited = 0;
591 Elf64_Ehdr ehdr;
592
593 if (inited)
594 return EBUSY;
595 inited = 1;
596
597 /*
598 * Use 64bit header since it's bigger. Shouldn't make a
599 * difference, since we're passing in all zeroes anyway.
600 */
601 memset(&ehdr, 0, sizeof(ehdr));
602 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
603
604 return 0;
605 }
606
607 static int
608 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen,
609 register_t *retval)
610 {
611 struct lwp *l;
612 struct sysent *callp;
613 int rv;
614
615 if (__predict_false(num >= SYS_NSYSENT))
616 return ENOSYS;
617
618 callp = rump_sysent + num;
619 rump_schedule();
620 l = curlwp;
621 rv = sy_call(callp, l, (void *)data, retval);
622 rump_unschedule();
623
624 return rv;
625 }
626
627 int
628 rump_boot_gethowto()
629 {
630
631 return boothowto;
632 }
633
634 void
635 rump_boot_sethowto(int howto)
636 {
637
638 boothowto = howto;
639 }
640
641 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local;
642 void *rump_sysproxy_arg;
643
644 /*
645 * This whole syscall-via-rpc is still taking form. For example, it
646 * may be necessary to set syscalls individually instead of lobbing
647 * them all to the same place. So don't think this interface is
648 * set in stone.
649 */
650 int
651 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg)
652 {
653
654 if (rump_sysproxy_arg)
655 return EBUSY;
656
657 rump_sysproxy_arg = arg;
658 rump_sysproxy = proxy;
659
660 return 0;
661 }
662
663 int
664 rump_getversion(void)
665 {
666
667 return __NetBSD_Version__;
668 }
669
670 /*
671 * Note: may be called unscheduled. Not fully safe since no locking
672 * of allevents (currently that's not even available).
673 */
674 void
675 rump_printevcnts()
676 {
677 struct evcnt *ev;
678
679 TAILQ_FOREACH(ev, &allevents, ev_list)
680 rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
681 ev->ev_group, ev->ev_name, ev->ev_count);
682 }
683
684 /*
685 * If you use this interface ... well ... all bets are off.
686 * The original purpose is for the p2k fs server library to be
687 * able to use the same pid/lid for VOPs as the host kernel.
688 */
689 void
690 rump_allbetsareoff_setid(pid_t pid, int lid)
691 {
692 struct lwp *l = curlwp;
693 struct proc *p = l->l_proc;
694
695 l->l_lid = lid;
696 p->p_pid = pid;
697 }
698