rump.c revision 1.227 1 /* $NetBSD: rump.c,v 1.227 2011/01/30 16:31:42 bouyer 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 1.227 2011/01/30 16:31:42 bouyer 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 #include <sys/simplelock.h>
73
74 #include <rump/rumpuser.h>
75
76 #include <secmodel/suser/suser.h>
77
78 #include <prop/proplib.h>
79
80 #include <uvm/uvm_extern.h>
81 #include <uvm/uvm_readahead.h>
82
83 #include "rump_private.h"
84 #include "rump_net_private.h"
85 #include "rump_vfs_private.h"
86 #include "rump_dev_private.h"
87
88 char machine[] = MACHINE;
89
90 struct proc *initproc;
91
92 struct device rump_rootdev = {
93 .dv_class = DV_VIRTUAL
94 };
95
96 #ifdef RUMP_WITHOUT_THREADS
97 int rump_threads = 0;
98 #else
99 int rump_threads = 1;
100 #endif
101
102 static int rump_proxy_syscall(int, void *, register_t *);
103 static int rump_proxy_rfork(void *, int, const char *);
104 static void rump_proxy_procexit(void);
105
106 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */
107
108 #ifdef LOCKDEBUG
109 const int rump_lockdebug = 1;
110 #else
111 const int rump_lockdebug = 0;
112 #endif
113 bool rump_ttycomponent = false;
114
115 static void
116 rump_aiodone_worker(struct work *wk, void *dummy)
117 {
118 struct buf *bp = (struct buf *)wk;
119
120 KASSERT(&bp->b_work == wk);
121 bp->b_iodone(bp);
122 }
123
124 static int rump_inited;
125
126 /*
127 * Make sure pnbuf_cache is available even without vfs
128 */
129 struct pool_cache *pnbuf_cache;
130 int rump_initpnbufpool(void);
131 int rump_initpnbufpool(void)
132 {
133
134 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
135 NULL, IPL_NONE, NULL, NULL, NULL);
136 return EOPNOTSUPP;
137 }
138
139 int rump__unavailable(void);
140 int rump__unavailable() {return EOPNOTSUPP;}
141 __weak_alias(rump_net_init,rump__unavailable);
142 __weak_alias(rump_vfs_init,rump_initpnbufpool);
143 __weak_alias(rump_dev_init,rump__unavailable);
144
145 __weak_alias(rump_vfs_fini,rump__unavailable);
146
147 __weak_alias(biodone,rump__unavailable);
148 __weak_alias(sopoll,rump__unavailable);
149
150 __weak_alias(rump_vfs_drainbufs,rump__unavailable);
151
152 void rump__unavailable_vfs_panic(void);
153 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
154 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
155
156 /* easier to write vfs-less clients */
157 __weak_alias(rump_pub_etfs_register,rump__unavailable);
158 __weak_alias(rump_pub_etfs_remove,rump__unavailable);
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 /* there's no convenient kernel entry point for this, so just craft out own */
199 static pid_t
200 spgetpid(void)
201 {
202
203 return curproc->p_pid;
204 }
205
206 static const struct rumpuser_sp_ops spops = {
207 .spop_schedule = rump_schedule,
208 .spop_unschedule = rump_unschedule,
209 .spop_lwproc_switch = rump_lwproc_switch,
210 .spop_lwproc_release = rump_lwproc_releaselwp,
211 .spop_lwproc_rfork = rump_proxy_rfork,
212 .spop_lwproc_newlwp = rump_lwproc_newlwp,
213 .spop_lwproc_curlwp = rump_lwproc_curlwp,
214 .spop_procexit = rump_proxy_procexit,
215 .spop_syscall = rump_proxy_syscall,
216 .spop_getpid = spgetpid,
217 };
218
219 int
220 rump_daemonize_begin(void)
221 {
222
223 if (rump_inited)
224 return EALREADY;
225
226 return rumpuser_daemonize_begin();
227 }
228
229 int
230 rump_daemonize_done(int error)
231 {
232
233 return rumpuser_daemonize_done(error);
234 }
235
236 int
237 rump__init(int rump_version)
238 {
239 char buf[256];
240 struct timespec ts;
241 uint64_t sec, nsec;
242 struct lwp *l;
243 int i, numcpu;
244 int error;
245
246 /* not reentrant */
247 if (rump_inited)
248 return 0;
249 else if (rump_inited == -1)
250 panic("rump_init: host process restart required");
251 else
252 rump_inited = 1;
253
254 if (rumpuser_getversion() != RUMPUSER_VERSION) {
255 /* let's hope the ABI of rumpuser_dprintf is the same ;) */
256 rumpuser_dprintf("rumpuser version mismatch: %d vs. %d\n",
257 rumpuser_getversion(), RUMPUSER_VERSION);
258 return EPROGMISMATCH;
259 }
260
261 if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) {
262 if (*buf != '0')
263 boothowto = AB_VERBOSE;
264 }
265
266 if (rumpuser_getenv("RUMP_NCPU", buf, sizeof(buf), &error) == 0)
267 error = 0;
268 if (error == 0) {
269 numcpu = strtoll(buf, NULL, 10);
270 if (numcpu < 1)
271 numcpu = 1;
272 } else {
273 numcpu = rumpuser_getnhostcpu();
274 }
275 rump_cpus_bootstrap(&numcpu);
276
277 rumpuser_gettime(&sec, &nsec, &error);
278 boottime.tv_sec = sec;
279 boottime.tv_nsec = nsec;
280
281 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
282 aprint_verbose("%s%s", copyright, version);
283
284 /*
285 * Seed arc4random() with a "reasonable" amount of randomness.
286 * Yes, this is a quick kludge which depends on the arc4random
287 * implementation.
288 */
289 messthestack();
290 arc4random();
291
292 if (rump_version != RUMP_VERSION) {
293 printf("rump version mismatch, %d vs. %d\n",
294 rump_version, RUMP_VERSION);
295 return EPROGMISMATCH;
296 }
297
298 if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
299 rump_threads = *buf != '0';
300 }
301 rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
302 rump_threads);
303 rump_intr_init(numcpu);
304 rump_tsleep_init();
305
306 /* init minimal lwp/cpu context */
307 l = &lwp0;
308 l->l_lid = 1;
309 l->l_cpu = l->l_target_cpu = rump_cpu;
310 l->l_fd = &filedesc0;
311 rumpuser_set_curlwp(l);
312
313 rumpuser_mutex_init(&rump_giantlock);
314 ksyms_init();
315 uvm_init();
316 evcnt_init();
317
318 once_init();
319 kernconfig_lock_init();
320 prop_kern_init();
321
322 pool_subsystem_init();
323 kmem_init();
324
325 uvm_ra_init();
326 uao_init();
327
328 mutex_obj_init();
329 callout_startup();
330
331 kprintf_init();
332 loginit();
333
334 kauth_init();
335
336 procinit();
337 proc0_init();
338 uid_init();
339 chgproccnt(0, 1);
340
341 l->l_proc = &proc0;
342 lwp_update_creds(l);
343
344 lwpinit_specificdata();
345 lwp_initspecific(&lwp0);
346
347 rump_scheduler_init(numcpu);
348 /* revert temporary context and schedule a semireal context */
349 rumpuser_set_curlwp(NULL);
350 initproc = &proc0; /* borrow proc0 before we get initproc started */
351 rump_schedule();
352
353 percpu_init();
354 inittimecounter();
355 ntp_init();
356
357 rumpuser_gettime(&sec, &nsec, &error);
358 ts.tv_sec = sec;
359 ts.tv_nsec = nsec;
360 tc_setclock(&ts);
361
362 /* we are mostly go. do per-cpu subsystem init */
363 for (i = 0; i < numcpu; i++) {
364 struct cpu_info *ci = cpu_lookup(i);
365
366 /* attach non-bootstrap CPUs */
367 if (i > 0) {
368 rump_cpu_attach(ci);
369 ncpu++;
370 }
371
372 callout_init_cpu(ci);
373 softint_init(ci);
374 xc_init_cpu(ci);
375 pool_cache_cpu_init(ci);
376 selsysinit(ci);
377 percpu_init_cpu(ci);
378
379 TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
380 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
381
382 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
383 }
384
385 sysctl_init();
386 mksysctls();
387 kqueue_init();
388 iostat_init();
389 fd_sys_init();
390 module_init();
391 devsw_init();
392 pipe_init();
393 resource_init();
394 procinit_sysctl();
395
396 /* start page baroness */
397 if (rump_threads) {
398 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
399 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
400 panic("pagedaemon create failed");
401 } else
402 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
403
404 /* process dso's */
405 rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load);
406
407 rump_component_init(RUMP_COMPONENT_KERN);
408
409 /* these do nothing if not present */
410 rump_vfs_init();
411 rump_net_init();
412 rump_dev_init();
413
414 rump_component_init(RUMP_COMPONENT_KERN_VFS);
415
416 /*
417 * if we initialized the tty component above, the tyttymtx is
418 * now initialized. otherwise, we need to initialize it.
419 */
420 if (!rump_ttycomponent)
421 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
422
423 cold = 0;
424
425 /* aieeeedondest */
426 if (rump_threads) {
427 if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
428 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
429 panic("aiodoned");
430 }
431
432 sysctl_finalize();
433
434 module_init_class(MODULE_CLASS_ANY);
435
436 rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
437 hostnamelen = strlen(hostname);
438
439 sigemptyset(&sigcantmask);
440
441 if (rump_threads)
442 vmem_rehash_start();
443
444 /*
445 * Create init, used to attach implicit threads in rump.
446 * (note: must be done after vfsinit to get cwdi)
447 */
448 (void)rump__lwproc_alloclwp(NULL); /* dummy thread for initproc */
449 mutex_enter(proc_lock);
450 initproc = proc_find_raw(1);
451 mutex_exit(proc_lock);
452 if (initproc == NULL)
453 panic("where in the world is initproc?");
454
455 /*
456 * Adjust syscall vector in case factions were dlopen()'d
457 * before calling rump_init().
458 * (modules will handle dynamic syscalls the usual way)
459 *
460 * Note: this will adjust the function vectors of
461 * syscalls which use a funcalias (getpid etc.), but
462 * it makes no difference.
463 */
464 for (i = 0; i < SYS_NSYSENT; i++) {
465 void *sym;
466
467 if (rump_sysent[i].sy_flags & SYCALL_NOSYS ||
468 *syscallnames[i] == '#' ||
469 rump_sysent[i].sy_call == sys_nomodule)
470 continue;
471
472 /*
473 * deal with compat wrappers. makesyscalls.sh should
474 * generate the necessary info instead of this hack,
475 * though. ugly, fix it later.
476 */
477 #define CPFX "compat_"
478 #define CPFXLEN (sizeof(CPFX)-1)
479 if (strncmp(syscallnames[i], CPFX, CPFXLEN) == 0) {
480 const char *p = syscallnames[i] + CPFXLEN;
481 size_t namelen;
482
483 /* skip version number */
484 while (*p >= '0' && *p <= '9')
485 p++;
486 if (p == syscallnames[i] + CPFXLEN || *p != '_')
487 panic("invalid syscall name %s\n",
488 syscallnames[i]);
489
490 /* skip over the next underscore */
491 p++;
492 namelen = p + (sizeof("rumpns_")-1) - syscallnames[i];
493
494 strcpy(buf, "rumpns_");
495 strcat(buf, syscallnames[i]);
496 /* XXX: no strncat in the kernel */
497 strcpy(buf+namelen, "sys_");
498 strcat(buf, p);
499 #undef CPFX
500 #undef CPFXLEN
501 } else {
502 sprintf(buf, "rumpns_sys_%s", syscallnames[i]);
503 }
504 if ((sym = rumpuser_dl_globalsym(buf)) != NULL
505 && sym != rump_sysent[i].sy_call) {
506 #if 0
507 rumpuser_dprintf("adjusting %s: %p (old %p)\n",
508 syscallnames[i], sym, rump_sysent[i].sy_call);
509 #endif
510 rump_sysent[i].sy_call = sym;
511 }
512 }
513
514 /* release cpu */
515 rump_unschedule();
516
517 return 0;
518 }
519
520 int
521 rump_init_server(const char *url)
522 {
523
524 return rumpuser_sp_init(url, &spops, ostype, osrelease, MACHINE);
525 }
526
527 void
528 cpu_reboot(int howto, char *bootstr)
529 {
530 int ruhow = 0;
531 void *finiarg;
532
533 printf("rump kernel halting...\n");
534
535 if (!RUMP_LOCALPROC_P(curproc))
536 finiarg = curproc->p_vmspace->vm_map.pmap;
537 else
538 finiarg = NULL;
539
540 /* dump means we really take the dive here */
541 if ((howto & RB_DUMP) || panicstr) {
542 ruhow = RUMPUSER_PANIC;
543 goto out;
544 }
545
546 /* try to sync */
547 if (!((howto & RB_NOSYNC) || panicstr)) {
548 rump_vfs_fini();
549 }
550
551 /* your wish is my command */
552 if (howto & RB_HALT) {
553 printf("rump kernel halted\n");
554 rumpuser_sp_fini(finiarg);
555 for (;;) {
556 uint64_t sec = 5, nsec = 0;
557 int error;
558
559 rumpuser_nanosleep(&sec, &nsec, &error);
560 }
561 }
562
563 /* this function is __dead, we must exit */
564 out:
565 printf("halted\n");
566 rumpuser_sp_fini(finiarg);
567 rumpuser_exit(ruhow);
568 }
569
570 struct uio *
571 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
572 {
573 struct uio *uio;
574 enum uio_rw uiorw;
575
576 switch (rw) {
577 case RUMPUIO_READ:
578 uiorw = UIO_READ;
579 break;
580 case RUMPUIO_WRITE:
581 uiorw = UIO_WRITE;
582 break;
583 default:
584 panic("%s: invalid rw %d", __func__, rw);
585 }
586
587 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
588 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
589
590 uio->uio_iov->iov_base = buf;
591 uio->uio_iov->iov_len = bufsize;
592
593 uio->uio_iovcnt = 1;
594 uio->uio_offset = offset;
595 uio->uio_resid = bufsize;
596 uio->uio_rw = uiorw;
597 UIO_SETUP_SYSSPACE(uio);
598
599 return uio;
600 }
601
602 size_t
603 rump_uio_getresid(struct uio *uio)
604 {
605
606 return uio->uio_resid;
607 }
608
609 off_t
610 rump_uio_getoff(struct uio *uio)
611 {
612
613 return uio->uio_offset;
614 }
615
616 size_t
617 rump_uio_free(struct uio *uio)
618 {
619 size_t resid;
620
621 resid = uio->uio_resid;
622 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
623 kmem_free(uio, sizeof(*uio));
624
625 return resid;
626 }
627
628 kauth_cred_t
629 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
630 {
631 kauth_cred_t cred;
632 int rv;
633
634 cred = kauth_cred_alloc();
635 kauth_cred_setuid(cred, uid);
636 kauth_cred_seteuid(cred, uid);
637 kauth_cred_setsvuid(cred, uid);
638 kauth_cred_setgid(cred, gid);
639 kauth_cred_setgid(cred, gid);
640 kauth_cred_setegid(cred, gid);
641 kauth_cred_setsvgid(cred, gid);
642 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
643 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */
644 assert(rv == 0);
645
646 return cred;
647 }
648
649 void
650 rump_cred_put(kauth_cred_t cred)
651 {
652
653 kauth_cred_free(cred);
654 }
655
656 static int compcounter[RUMP_COMPONENT_MAX];
657
658 static void
659 rump_component_init_cb(struct rump_component *rc, int type)
660 {
661
662 KASSERT(type < RUMP_COMPONENT_MAX);
663 if (rc->rc_type == type) {
664 rc->rc_init();
665 compcounter[type]++;
666 }
667 }
668
669 int
670 rump_component_count(enum rump_component_type type)
671 {
672
673 KASSERT(type <= RUMP_COMPONENT_MAX);
674 return compcounter[type];
675 }
676
677 void
678 rump_component_init(enum rump_component_type type)
679 {
680
681 rumpuser_dl_component_init(type, rump_component_init_cb);
682 }
683
684 /*
685 * Initialize a module which has already been loaded and linked
686 * with dlopen(). This is fundamentally the same as a builtin module.
687 */
688 int
689 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
690 {
691
692 return module_builtin_add(mip, nmodinfo, true);
693 }
694
695 /*
696 * Finish module (flawless victory, fatality!).
697 */
698 int
699 rump_module_fini(const struct modinfo *mi)
700 {
701
702 return module_builtin_remove(mi, true);
703 }
704
705 /*
706 * Add loaded and linked module to the builtin list. It will
707 * later be initialized with module_init_class().
708 */
709
710 static void
711 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
712 {
713
714 module_builtin_add(mip, nmodinfo, false);
715 }
716
717 int
718 rump_kernelfsym_load(void *symtab, uint64_t symsize,
719 char *strtab, uint64_t strsize)
720 {
721 static int inited = 0;
722 Elf64_Ehdr ehdr;
723
724 if (inited)
725 return EBUSY;
726 inited = 1;
727
728 /*
729 * Use 64bit header since it's bigger. Shouldn't make a
730 * difference, since we're passing in all zeroes anyway.
731 */
732 memset(&ehdr, 0, sizeof(ehdr));
733 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
734
735 return 0;
736 }
737
738 static int
739 rump_proxy_syscall(int num, void *arg, register_t *retval)
740 {
741 struct lwp *l;
742 struct sysent *callp;
743 int rv;
744
745 if (__predict_false(num >= SYS_NSYSENT))
746 return ENOSYS;
747
748 callp = rump_sysent + num;
749 l = curlwp;
750 rv = sy_call(callp, l, (void *)arg, retval);
751
752 return rv;
753 }
754
755 static int
756 rump_proxy_rfork(void *priv, int flags, const char *comm)
757 {
758 struct vmspace *newspace;
759 struct proc *p;
760 int error;
761
762 if ((error = rump_lwproc_rfork(flags)) != 0)
763 return error;
764
765 /*
766 * Since it's a proxy proc, adjust the vmspace.
767 * Refcount will eternally be 1.
768 */
769 p = curproc;
770 newspace = kmem_alloc(sizeof(*newspace), KM_SLEEP);
771 newspace->vm_refcnt = 1;
772 newspace->vm_map.pmap = priv;
773 KASSERT(p->p_vmspace == vmspace_kernel());
774 p->p_vmspace = newspace;
775 if (comm)
776 strlcpy(p->p_comm, comm, sizeof(p->p_comm));
777
778 return 0;
779 }
780
781 static void
782 rump_proxy_procexit(void)
783 {
784 struct proc *p = curproc;
785 uint64_t where;
786 struct lwp *l;
787
788 mutex_enter(p->p_lock);
789 /*
790 * First pass: mark all lwps in the process with LSDEAD
791 * so that they know they should exit.
792 */
793 LIST_FOREACH(l, &p->p_lwps, l_sibling) {
794 if (l == curlwp)
795 continue;
796 l->l_flag |= LW_RUMP_DYING;
797 }
798 mutex_exit(p->p_lock);
799
800 /*
801 * Next, make sure everyone on all CPUs sees our status
802 * update. This keeps threads inside cv_wait() and makes
803 * sure we don't access a stale cv pointer later when
804 * we wake up the threads.
805 */
806
807 where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
808 xc_wait(where);
809
810 /*
811 * Ok, all lwps are either:
812 * 1) not in the cv code
813 * 2) sleeping on l->l_private
814 * 3) sleeping on p->p_waitcv
815 *
816 * Either way, l_private is stable until we change the lwps
817 * state to LSZOMB.
818 */
819
820 mutex_enter(p->p_lock);
821 LIST_FOREACH(l, &p->p_lwps, l_sibling) {
822 if (l->l_private)
823 cv_broadcast(l->l_private);
824 }
825 p->p_stat = SDYING;
826 cv_broadcast(&p->p_waitcv);
827 mutex_exit(p->p_lock);
828
829 /*
830 * Don't wait for lwps to exit. There's refcounting in the
831 * rumpuser sp code which makes this safe. Also, this routine
832 * should not sleep for a long time.
833 */
834 }
835
836 int
837 rump_boot_gethowto()
838 {
839
840 return boothowto;
841 }
842
843 void
844 rump_boot_sethowto(int howto)
845 {
846
847 boothowto = howto;
848 }
849
850 int
851 rump_getversion(void)
852 {
853
854 return __NetBSD_Version__;
855 }
856
857 /*
858 * Note: may be called unscheduled. Not fully safe since no locking
859 * of allevents (currently that's not even available).
860 */
861 void
862 rump_printevcnts()
863 {
864 struct evcnt *ev;
865
866 TAILQ_FOREACH(ev, &allevents, ev_list)
867 rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
868 ev->ev_group, ev->ev_name, ev->ev_count);
869 }
870
871 /*
872 * If you use this interface ... well ... all bets are off.
873 * The original purpose is for the p2k fs server library to be
874 * able to use the same pid/lid for VOPs as the host kernel.
875 */
876 void
877 rump_allbetsareoff_setid(pid_t pid, int lid)
878 {
879 struct lwp *l = curlwp;
880 struct proc *p = l->l_proc;
881
882 l->l_lid = lid;
883 p->p_pid = pid;
884 }
885