rump.c revision 1.233 1 /* $NetBSD: rump.c,v 1.233 2011/03/21 16:41:08 pooka Exp $ */
2
3 /*
4 * Copyright (c) 2007-2011 Antti Kantee. All Rights Reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
16 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
17 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
18 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
21 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28 #include <sys/cdefs.h>
29 __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.233 2011/03/21 16:41:08 pooka Exp $");
30
31 #include <sys/systm.h>
32 #define ELFSIZE ARCH_ELFSIZE
33
34 #include <sys/param.h>
35 #include <sys/atomic.h>
36 #include <sys/buf.h>
37 #include <sys/callout.h>
38 #include <sys/conf.h>
39 #include <sys/cpu.h>
40 #include <sys/device.h>
41 #include <sys/evcnt.h>
42 #include <sys/event.h>
43 #include <sys/exec_elf.h>
44 #include <sys/filedesc.h>
45 #include <sys/iostat.h>
46 #include <sys/kauth.h>
47 #include <sys/kernel.h>
48 #include <sys/kmem.h>
49 #include <sys/kprintf.h>
50 #include <sys/kthread.h>
51 #include <sys/ksyms.h>
52 #include <sys/msgbuf.h>
53 #include <sys/module.h>
54 #include <sys/once.h>
55 #include <sys/percpu.h>
56 #include <sys/pipe.h>
57 #include <sys/pool.h>
58 #include <sys/queue.h>
59 #include <sys/reboot.h>
60 #include <sys/resourcevar.h>
61 #include <sys/select.h>
62 #include <sys/sysctl.h>
63 #include <sys/syscall.h>
64 #include <sys/syscallvar.h>
65 #include <sys/timetc.h>
66 #include <sys/tty.h>
67 #include <sys/uidinfo.h>
68 #include <sys/vmem.h>
69 #include <sys/xcall.h>
70 #include <sys/simplelock.h>
71
72 #include <rump/rumpuser.h>
73
74 #include <secmodel/suser/suser.h>
75
76 #include <prop/proplib.h>
77
78 #include <uvm/uvm_extern.h>
79 #include <uvm/uvm_readahead.h>
80
81 #include "rump_private.h"
82 #include "rump_net_private.h"
83 #include "rump_vfs_private.h"
84 #include "rump_dev_private.h"
85
86 char machine[] = MACHINE;
87
88 struct proc *initproc;
89
90 struct device rump_rootdev = {
91 .dv_class = DV_VIRTUAL
92 };
93
94 #ifdef RUMP_WITHOUT_THREADS
95 int rump_threads = 0;
96 #else
97 int rump_threads = 1;
98 #endif
99
100 static int rump_proxy_syscall(int, void *, register_t *);
101 static int rump_proxy_rfork(void *, int, const char *);
102 static void rump_proxy_lwpexit(void);
103 static void rump_proxy_execnotify(const char *);
104
105 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */
106
107 #ifdef LOCKDEBUG
108 const int rump_lockdebug = 1;
109 #else
110 const int rump_lockdebug = 0;
111 #endif
112 bool rump_ttycomponent = false;
113
114 static void
115 rump_aiodone_worker(struct work *wk, void *dummy)
116 {
117 struct buf *bp = (struct buf *)wk;
118
119 KASSERT(&bp->b_work == wk);
120 bp->b_iodone(bp);
121 }
122
123 static int rump_inited;
124
125 /*
126 * Make sure pnbuf_cache is available even without vfs
127 */
128 struct pool_cache *pnbuf_cache;
129 int rump_initpnbufpool(void);
130 int rump_initpnbufpool(void)
131 {
132
133 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
134 NULL, IPL_NONE, NULL, NULL, NULL);
135 return EOPNOTSUPP;
136 }
137
138 int rump__unavailable(void);
139 int rump__unavailable() {return EOPNOTSUPP;}
140 __weak_alias(rump_net_init,rump__unavailable);
141 __weak_alias(rump_vfs_init,rump_initpnbufpool);
142 __weak_alias(rump_dev_init,rump__unavailable);
143
144 __weak_alias(rump_vfs_fini,rump__unavailable);
145
146 __weak_alias(biodone,rump__unavailable);
147 __weak_alias(sopoll,rump__unavailable);
148
149 __weak_alias(rump_vfs_drainbufs,rump__unavailable);
150
151 void rump__unavailable_vfs_panic(void);
152 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
153 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
154
155 /* easier to write vfs-less clients */
156 __weak_alias(rump_pub_etfs_register,rump__unavailable);
157 __weak_alias(rump_pub_etfs_register_withsize,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_lwpexit = rump_proxy_lwpexit,
215 .spop_syscall = rump_proxy_syscall,
216 .spop_execnotify = rump_proxy_execnotify,
217 .spop_getpid = spgetpid,
218 };
219
220 int
221 rump_daemonize_begin(void)
222 {
223
224 if (rump_inited)
225 return EALREADY;
226
227 return rumpuser_daemonize_begin();
228 }
229
230 int
231 rump_daemonize_done(int error)
232 {
233
234 return rumpuser_daemonize_done(error);
235 }
236
237 int
238 rump__init(int rump_version)
239 {
240 char buf[256];
241 struct timespec ts;
242 uint64_t sec, nsec;
243 struct lwp *l;
244 int i, numcpu;
245 int error;
246
247 /* not reentrant */
248 if (rump_inited)
249 return 0;
250 else if (rump_inited == -1)
251 panic("rump_init: host process restart required");
252 else
253 rump_inited = 1;
254
255 if (rumpuser_getversion() != RUMPUSER_VERSION) {
256 /* let's hope the ABI of rumpuser_dprintf is the same ;) */
257 rumpuser_dprintf("rumpuser version mismatch: %d vs. %d\n",
258 rumpuser_getversion(), RUMPUSER_VERSION);
259 return EPROGMISMATCH;
260 }
261
262 if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) {
263 if (*buf != '0')
264 boothowto = AB_VERBOSE;
265 }
266
267 if (rumpuser_getenv("RUMP_NCPU", buf, sizeof(buf), &error) == 0)
268 error = 0;
269 if (error == 0) {
270 numcpu = strtoll(buf, NULL, 10);
271 if (numcpu < 1)
272 numcpu = 1;
273 } else {
274 numcpu = rumpuser_getnhostcpu();
275 }
276 rump_cpus_bootstrap(&numcpu);
277
278 rumpuser_gettime(&sec, &nsec, &error);
279 boottime.tv_sec = sec;
280 boottime.tv_nsec = nsec;
281
282 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
283 aprint_verbose("%s%s", copyright, version);
284
285 /*
286 * Seed arc4random() with a "reasonable" amount of randomness.
287 * Yes, this is a quick kludge which depends on the arc4random
288 * implementation.
289 */
290 messthestack();
291 arc4random();
292
293 if (rump_version != RUMP_VERSION) {
294 printf("rump version mismatch, %d vs. %d\n",
295 rump_version, RUMP_VERSION);
296 return EPROGMISMATCH;
297 }
298
299 if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
300 rump_threads = *buf != '0';
301 }
302 rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
303 rump_threads);
304 rump_intr_init(numcpu);
305 rump_tsleep_init();
306
307 /* init minimal lwp/cpu context */
308 l = &lwp0;
309 l->l_lid = 1;
310 l->l_cpu = l->l_target_cpu = rump_cpu;
311 l->l_fd = &filedesc0;
312 rumpuser_set_curlwp(l);
313
314 rumpuser_mutex_init(&rump_giantlock);
315 ksyms_init();
316 uvm_init();
317 evcnt_init();
318
319 once_init();
320 kernconfig_lock_init();
321 prop_kern_init();
322
323 pool_subsystem_init();
324 kmem_init();
325
326 uvm_ra_init();
327 uao_init();
328
329 mutex_obj_init();
330 callout_startup();
331
332 kprintf_init();
333 loginit();
334
335 kauth_init();
336
337 procinit();
338 proc0_init();
339 uid_init();
340 chgproccnt(0, 1);
341
342 l->l_proc = &proc0;
343 lwp_update_creds(l);
344
345 lwpinit_specificdata();
346 lwp_initspecific(&lwp0);
347
348 rump_scheduler_init(numcpu);
349 /* revert temporary context and schedule a semireal context */
350 rumpuser_set_curlwp(NULL);
351 initproc = &proc0; /* borrow proc0 before we get initproc started */
352 rump_schedule();
353
354 percpu_init();
355 inittimecounter();
356 ntp_init();
357
358 rumpuser_gettime(&sec, &nsec, &error);
359 ts.tv_sec = sec;
360 ts.tv_nsec = nsec;
361 tc_setclock(&ts);
362
363 /* we are mostly go. do per-cpu subsystem init */
364 for (i = 0; i < numcpu; i++) {
365 struct cpu_info *ci = cpu_lookup(i);
366
367 /* attach non-bootstrap CPUs */
368 if (i > 0) {
369 rump_cpu_attach(ci);
370 ncpu++;
371 }
372
373 callout_init_cpu(ci);
374 softint_init(ci);
375 xc_init_cpu(ci);
376 pool_cache_cpu_init(ci);
377 selsysinit(ci);
378 percpu_init_cpu(ci);
379
380 TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
381 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
382
383 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
384 }
385
386 sysctl_init();
387 mksysctls();
388 kqueue_init();
389 iostat_init();
390 fd_sys_init();
391 module_init();
392 devsw_init();
393 pipe_init();
394 resource_init();
395 procinit_sysctl();
396
397 /* start page baroness */
398 if (rump_threads) {
399 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
400 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
401 panic("pagedaemon create failed");
402 } else
403 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
404
405 /* process dso's */
406 rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load);
407
408 rump_component_init(RUMP_COMPONENT_KERN);
409
410 /* these do nothing if not present */
411 rump_vfs_init();
412 rump_net_init();
413 rump_dev_init();
414
415 rump_component_init(RUMP_COMPONENT_KERN_VFS);
416
417 /*
418 * if we initialized the tty component above, the tyttymtx is
419 * now initialized. otherwise, we need to initialize it.
420 */
421 if (!rump_ttycomponent)
422 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
423
424 cold = 0;
425
426 /* aieeeedondest */
427 if (rump_threads) {
428 if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
429 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
430 panic("aiodoned");
431 }
432
433 sysctl_finalize();
434
435 module_init_class(MODULE_CLASS_ANY);
436
437 rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
438 hostnamelen = strlen(hostname);
439
440 sigemptyset(&sigcantmask);
441
442 if (rump_threads)
443 vmem_rehash_start();
444
445 /*
446 * Create init, used to attach implicit threads in rump.
447 * (note: must be done after vfsinit to get cwdi)
448 */
449 (void)rump__lwproc_alloclwp(NULL); /* dummy thread for initproc */
450 mutex_enter(proc_lock);
451 initproc = proc_find_raw(1);
452 mutex_exit(proc_lock);
453 if (initproc == NULL)
454 panic("where in the world is initproc?");
455
456 /*
457 * Adjust syscall vector in case factions were dlopen()'d
458 * before calling rump_init().
459 * (modules will handle dynamic syscalls the usual way)
460 *
461 * Note: this will adjust the function vectors of
462 * syscalls which use a funcalias (getpid etc.), but
463 * it makes no difference.
464 */
465 for (i = 0; i < SYS_NSYSENT; i++) {
466 void *sym;
467
468 if (rump_sysent[i].sy_flags & SYCALL_NOSYS ||
469 *syscallnames[i] == '#' ||
470 rump_sysent[i].sy_call == sys_nomodule)
471 continue;
472
473 /*
474 * deal with compat wrappers. makesyscalls.sh should
475 * generate the necessary info instead of this hack,
476 * though. ugly, fix it later.
477 */
478 #define CPFX "compat_"
479 #define CPFXLEN (sizeof(CPFX)-1)
480 if (strncmp(syscallnames[i], CPFX, CPFXLEN) == 0) {
481 const char *p = syscallnames[i] + CPFXLEN;
482 size_t namelen;
483
484 /* skip version number */
485 while (*p >= '0' && *p <= '9')
486 p++;
487 if (p == syscallnames[i] + CPFXLEN || *p != '_')
488 panic("invalid syscall name %s\n",
489 syscallnames[i]);
490
491 /* skip over the next underscore */
492 p++;
493 namelen = p + (sizeof("rumpns_")-1) - syscallnames[i];
494
495 strcpy(buf, "rumpns_");
496 strcat(buf, syscallnames[i]);
497 /* XXX: no strncat in the kernel */
498 strcpy(buf+namelen, "sys_");
499 strcat(buf, p);
500 #undef CPFX
501 #undef CPFXLEN
502 } else {
503 sprintf(buf, "rumpns_sys_%s", syscallnames[i]);
504 }
505 if ((sym = rumpuser_dl_globalsym(buf)) != NULL
506 && sym != rump_sysent[i].sy_call) {
507 #if 0
508 rumpuser_dprintf("adjusting %s: %p (old %p)\n",
509 syscallnames[i], sym, rump_sysent[i].sy_call);
510 #endif
511 rump_sysent[i].sy_call = sym;
512 }
513 }
514
515 /* release cpu */
516 rump_unschedule();
517
518 return 0;
519 }
520
521 int
522 rump_init_server(const char *url)
523 {
524
525 return rumpuser_sp_init(url, &spops, ostype, osrelease, MACHINE);
526 }
527
528 void
529 cpu_reboot(int howto, char *bootstr)
530 {
531 int ruhow = 0;
532 void *finiarg;
533
534 printf("rump kernel halting...\n");
535
536 if (!RUMP_LOCALPROC_P(curproc))
537 finiarg = curproc->p_vmspace->vm_map.pmap;
538 else
539 finiarg = NULL;
540
541 /* dump means we really take the dive here */
542 if ((howto & RB_DUMP) || panicstr) {
543 ruhow = RUMPUSER_PANIC;
544 goto out;
545 }
546
547 /* try to sync */
548 if (!((howto & RB_NOSYNC) || panicstr)) {
549 rump_vfs_fini();
550 }
551
552 /* your wish is my command */
553 if (howto & RB_HALT) {
554 printf("rump kernel halted\n");
555 rumpuser_sp_fini(finiarg);
556 for (;;) {
557 uint64_t sec = 5, nsec = 0;
558 int error;
559
560 rumpuser_nanosleep(&sec, &nsec, &error);
561 }
562 }
563
564 /* this function is __dead, we must exit */
565 out:
566 printf("halted\n");
567 rumpuser_sp_fini(finiarg);
568 rumpuser_exit(ruhow);
569 }
570
571 struct uio *
572 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
573 {
574 struct uio *uio;
575 enum uio_rw uiorw;
576
577 switch (rw) {
578 case RUMPUIO_READ:
579 uiorw = UIO_READ;
580 break;
581 case RUMPUIO_WRITE:
582 uiorw = UIO_WRITE;
583 break;
584 default:
585 panic("%s: invalid rw %d", __func__, rw);
586 }
587
588 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
589 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
590
591 uio->uio_iov->iov_base = buf;
592 uio->uio_iov->iov_len = bufsize;
593
594 uio->uio_iovcnt = 1;
595 uio->uio_offset = offset;
596 uio->uio_resid = bufsize;
597 uio->uio_rw = uiorw;
598 UIO_SETUP_SYSSPACE(uio);
599
600 return uio;
601 }
602
603 size_t
604 rump_uio_getresid(struct uio *uio)
605 {
606
607 return uio->uio_resid;
608 }
609
610 off_t
611 rump_uio_getoff(struct uio *uio)
612 {
613
614 return uio->uio_offset;
615 }
616
617 size_t
618 rump_uio_free(struct uio *uio)
619 {
620 size_t resid;
621
622 resid = uio->uio_resid;
623 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
624 kmem_free(uio, sizeof(*uio));
625
626 return resid;
627 }
628
629 kauth_cred_t
630 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
631 {
632 kauth_cred_t cred;
633 int rv;
634
635 cred = kauth_cred_alloc();
636 kauth_cred_setuid(cred, uid);
637 kauth_cred_seteuid(cred, uid);
638 kauth_cred_setsvuid(cred, uid);
639 kauth_cred_setgid(cred, gid);
640 kauth_cred_setgid(cred, gid);
641 kauth_cred_setegid(cred, gid);
642 kauth_cred_setsvgid(cred, gid);
643 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
644 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */
645 assert(rv == 0);
646
647 return cred;
648 }
649
650 void
651 rump_cred_put(kauth_cred_t cred)
652 {
653
654 kauth_cred_free(cred);
655 }
656
657 static int compcounter[RUMP_COMPONENT_MAX];
658
659 static void
660 rump_component_init_cb(struct rump_component *rc, int type)
661 {
662
663 KASSERT(type < RUMP_COMPONENT_MAX);
664 if (rc->rc_type == type) {
665 rc->rc_init();
666 compcounter[type]++;
667 }
668 }
669
670 int
671 rump_component_count(enum rump_component_type type)
672 {
673
674 KASSERT(type <= RUMP_COMPONENT_MAX);
675 return compcounter[type];
676 }
677
678 void
679 rump_component_init(enum rump_component_type type)
680 {
681
682 rumpuser_dl_component_init(type, rump_component_init_cb);
683 }
684
685 /*
686 * Initialize a module which has already been loaded and linked
687 * with dlopen(). This is fundamentally the same as a builtin module.
688 */
689 int
690 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
691 {
692
693 return module_builtin_add(mip, nmodinfo, true);
694 }
695
696 /*
697 * Finish module (flawless victory, fatality!).
698 */
699 int
700 rump_module_fini(const struct modinfo *mi)
701 {
702
703 return module_builtin_remove(mi, true);
704 }
705
706 /*
707 * Add loaded and linked module to the builtin list. It will
708 * later be initialized with module_init_class().
709 */
710
711 static void
712 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
713 {
714
715 module_builtin_add(mip, nmodinfo, false);
716 }
717
718 int
719 rump_kernelfsym_load(void *symtab, uint64_t symsize,
720 char *strtab, uint64_t strsize)
721 {
722 static int inited = 0;
723 Elf64_Ehdr ehdr;
724
725 if (inited)
726 return EBUSY;
727 inited = 1;
728
729 /*
730 * Use 64bit header since it's bigger. Shouldn't make a
731 * difference, since we're passing in all zeroes anyway.
732 */
733 memset(&ehdr, 0, sizeof(ehdr));
734 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
735
736 return 0;
737 }
738
739 static int
740 rump_proxy_syscall(int num, void *arg, register_t *retval)
741 {
742 struct lwp *l;
743 struct sysent *callp;
744 int rv;
745
746 if (__predict_false(num >= SYS_NSYSENT))
747 return ENOSYS;
748
749 callp = rump_sysent + num;
750 l = curlwp;
751 rv = sy_call(callp, l, (void *)arg, retval);
752
753 return rv;
754 }
755
756 static int
757 rump_proxy_rfork(void *priv, int flags, const char *comm)
758 {
759 struct vmspace *newspace;
760 struct proc *p;
761 int error;
762
763 if ((error = rump_lwproc_rfork(flags)) != 0)
764 return error;
765
766 /*
767 * Since it's a proxy proc, adjust the vmspace.
768 * Refcount will eternally be 1.
769 */
770 p = curproc;
771 newspace = kmem_zalloc(sizeof(*newspace), KM_SLEEP);
772 newspace->vm_refcnt = 1;
773 newspace->vm_map.pmap = priv;
774 KASSERT(p->p_vmspace == vmspace_kernel());
775 p->p_vmspace = newspace;
776 if (comm)
777 strlcpy(p->p_comm, comm, sizeof(p->p_comm));
778
779 return 0;
780 }
781
782 /*
783 * Order all lwps in a process to exit. does *not* wait for them to drain.
784 */
785 static void
786 rump_proxy_lwpexit(void)
787 {
788 struct proc *p = curproc;
789 uint64_t where;
790 struct lwp *l;
791
792 mutex_enter(p->p_lock);
793 /*
794 * First pass: mark all lwps in the process with LW_RUMP_QEXIT
795 * so that they know they should exit.
796 */
797 LIST_FOREACH(l, &p->p_lwps, l_sibling) {
798 if (l == curlwp)
799 continue;
800 l->l_flag |= LW_RUMP_QEXIT;
801 }
802 mutex_exit(p->p_lock);
803
804 /*
805 * Next, make sure everyone on all CPUs sees our status
806 * update. This keeps threads inside cv_wait() and makes
807 * sure we don't access a stale cv pointer later when
808 * we wake up the threads.
809 */
810
811 where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
812 xc_wait(where);
813
814 /*
815 * Ok, all lwps are either:
816 * 1) not in the cv code
817 * 2) sleeping on l->l_private
818 * 3) sleeping on p->p_waitcv
819 *
820 * Either way, l_private is stable until we set PS_RUMP_LWPEXIT
821 * in p->p_sflag.
822 */
823
824 mutex_enter(p->p_lock);
825 LIST_FOREACH(l, &p->p_lwps, l_sibling) {
826 if (l->l_private)
827 cv_broadcast(l->l_private);
828 }
829 p->p_sflag |= PS_RUMP_LWPEXIT;
830 cv_broadcast(&p->p_waitcv);
831 mutex_exit(p->p_lock);
832 }
833
834 /*
835 * Notify process that all threads have been drained and exec is complete.
836 */
837 static void
838 rump_proxy_execnotify(const char *comm)
839 {
840 struct proc *p = curproc;
841
842 fd_closeexec();
843 mutex_enter(p->p_lock);
844 KASSERT(p->p_nlwps == 1 && p->p_sflag & PS_RUMP_LWPEXIT);
845 p->p_sflag &= ~PS_RUMP_LWPEXIT;
846 mutex_exit(p->p_lock);
847 strlcpy(p->p_comm, comm, sizeof(p->p_comm));
848 }
849
850 int
851 rump_boot_gethowto()
852 {
853
854 return boothowto;
855 }
856
857 void
858 rump_boot_sethowto(int howto)
859 {
860
861 boothowto = howto;
862 }
863
864 int
865 rump_getversion(void)
866 {
867
868 return __NetBSD_Version__;
869 }
870
871 /*
872 * Note: may be called unscheduled. Not fully safe since no locking
873 * of allevents (currently that's not even available).
874 */
875 void
876 rump_printevcnts()
877 {
878 struct evcnt *ev;
879
880 TAILQ_FOREACH(ev, &allevents, ev_list)
881 rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
882 ev->ev_group, ev->ev_name, ev->ev_count);
883 }
884
885 /*
886 * If you use this interface ... well ... all bets are off.
887 * The original purpose is for the p2k fs server library to be
888 * able to use the same pid/lid for VOPs as the host kernel.
889 */
890 void
891 rump_allbetsareoff_setid(pid_t pid, int lid)
892 {
893 struct lwp *l = curlwp;
894 struct proc *p = l->l_proc;
895
896 l->l_lid = lid;
897 p->p_pid = pid;
898 }
899