1 /* $NetBSD: rump.c,v 1.329 2016/03/08 14:30:48 joerg 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 1.329 2016/03/08 14:30:48 joerg 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/kcpuset.h>
48 #include <sys/kernel.h>
49 #include <sys/kmem.h>
50 #include <sys/kprintf.h>
51 #include <sys/kthread.h>
52 #include <sys/ksyms.h>
53 #include <sys/msgbuf.h>
54 #include <sys/module.h>
55 #include <sys/namei.h>
56 #include <sys/once.h>
57 #include <sys/percpu.h>
58 #include <sys/pipe.h>
59 #include <sys/pool.h>
60 #include <sys/pserialize.h>
61 #include <sys/queue.h>
62 #include <sys/reboot.h>
63 #include <sys/resourcevar.h>
64 #include <sys/select.h>
65 #include <sys/sysctl.h>
66 #include <sys/syscall.h>
67 #include <sys/syscallvar.h>
68 #include <sys/timetc.h>
69 #include <sys/tty.h>
70 #include <sys/uidinfo.h>
71 #include <sys/vmem.h>
72 #include <sys/xcall.h>
73 #include <sys/cprng.h>
74 #include <sys/rnd.h>
75 #include <sys/ktrace.h>
76
77 #include <rump-sys/kern.h>
78 #include <rump-sys/dev.h>
79 #include <rump-sys/net.h>
80 #include <rump-sys/vfs.h>
81
82 #include <rump/rumpuser.h>
83
84 #include <secmodel/suser/suser.h>
85
86 #include <prop/proplib.h>
87
88 #include <uvm/uvm_extern.h>
89 #include <uvm/uvm_readahead.h>
90
91 char machine[] = MACHINE;
92 char machine_arch[] = MACHINE_ARCH;
93
94 struct proc *initproc;
95
96 struct device rump_rootdev = {
97 .dv_class = DV_VIRTUAL
98 };
99
100 #ifdef RUMP_WITHOUT_THREADS
101 int rump_threads = 0;
102 #else
103 int rump_threads = 1;
104 #endif
105
106 static void rump_component_addlocal(void);
107 static struct lwp *bootlwp;
108
109 /* 16k should be enough for std rump needs */
110 static char rump_msgbuf[16*1024] __aligned(256);
111
112 bool rump_ttycomponent = false;
113
114 static void
rump_aiodone_worker(struct work * wk,void * dummy)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 void (*rump_vfs_drainbufs)(int) = (void *)nullop;
126 int (*rump_vfs_makeonedevnode)(dev_t, const char *,
127 devmajor_t, devminor_t) = (void *)nullop;
128 int (*rump_vfs_makedevnodes)(dev_t, const char *, char,
129 devmajor_t, devminor_t, int) = (void *)nullop;
130 int (*rump_vfs_makesymlink)(const char *, const char *) = (void *)nullop;
131
132 rump_proc_vfs_init_fn rump_proc_vfs_init = (void *)nullop;
133 rump_proc_vfs_release_fn rump_proc_vfs_release = (void *)nullop;
134
135 static void add_linkedin_modules(const struct modinfo *const *, size_t);
136
rspo_wrap_getpid(void)137 static pid_t rspo_wrap_getpid(void) {
138 return rump_sysproxy_hyp_getpid();
139 }
rspo_wrap_syscall(int num,void * arg,long * retval)140 static int rspo_wrap_syscall(int num, void *arg, long *retval) {
141 return rump_sysproxy_hyp_syscall(num, arg, retval);
142 }
rspo_wrap_rfork(void * priv,int flag,const char * comm)143 static int rspo_wrap_rfork(void *priv, int flag, const char *comm) {
144 return rump_sysproxy_hyp_rfork(priv, flag, comm);
145 }
rspo_wrap_lwpexit(void)146 static void rspo_wrap_lwpexit(void) {
147 rump_sysproxy_hyp_lwpexit();
148 }
rspo_wrap_execnotify(const char * comm)149 static void rspo_wrap_execnotify(const char *comm) {
150 rump_sysproxy_hyp_execnotify(comm);
151 }
152 static const struct rumpuser_hyperup hyp = {
153 .hyp_schedule = rump_schedule,
154 .hyp_unschedule = rump_unschedule,
155 .hyp_backend_unschedule = rump_user_unschedule,
156 .hyp_backend_schedule = rump_user_schedule,
157 .hyp_lwproc_switch = rump_lwproc_switch,
158 .hyp_lwproc_release = rump_lwproc_releaselwp,
159 .hyp_lwproc_newlwp = rump_lwproc_newlwp,
160 .hyp_lwproc_curlwp = rump_lwproc_curlwp,
161
162 .hyp_getpid = rspo_wrap_getpid,
163 .hyp_syscall = rspo_wrap_syscall,
164 .hyp_lwproc_rfork = rspo_wrap_rfork,
165 .hyp_lwpexit = rspo_wrap_lwpexit,
166 .hyp_execnotify = rspo_wrap_execnotify,
167 };
168 struct rump_sysproxy_ops rump_sysproxy_ops = {
169 .rspo_copyin = (void *)enxio,
170 .rspo_copyinstr = (void *)enxio,
171 .rspo_copyout = (void *)enxio,
172 .rspo_copyoutstr = (void *)enxio,
173 .rspo_anonmmap = (void *)enxio,
174 .rspo_raise = (void *)enxio,
175 .rspo_fini = (void *)enxio,
176 .rspo_hyp_getpid = (void *)enxio,
177 .rspo_hyp_syscall = (void *)enxio,
178 .rspo_hyp_rfork = (void *)enxio,
179 .rspo_hyp_lwpexit = (void *)enxio,
180 .rspo_hyp_execnotify = (void *)enxio,
181 };
182
183 int
rump_daemonize_begin(void)184 rump_daemonize_begin(void)
185 {
186
187 if (rump_inited)
188 return EALREADY;
189
190 return rumpuser_daemonize_begin();
191 }
192
193 int
rump_daemonize_done(int error)194 rump_daemonize_done(int error)
195 {
196
197 return rumpuser_daemonize_done(error);
198 }
199
200 #ifdef RUMP_USE_CTOR
201
202 /* sysctl bootstrap handling */
203 struct sysctl_boot_chain sysctl_boot_chain \
204 = LIST_HEAD_INITIALIZER(sysctl_boot_chain);
205 __link_set_add_text(sysctl_funcs,voidop); /* ensure linkset is non-empty */
206
207 #else /* RUMP_USE_CTOR */
208
RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT)209 RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT)
210 {
211 __link_set_decl(rump_components, struct rump_component);
212
213 /*
214 * Trick compiler into generating references so that statically
215 * linked rump kernels are generated with the link set symbols.
216 */
217 asm("" :: "r"(__start_link_set_rump_components));
218 asm("" :: "r"(__stop_link_set_rump_components));
219 }
220
221 #endif /* RUMP_USE_CTOR */
222
223 int
rump_init(void)224 rump_init(void)
225 {
226 char buf[256];
227 struct timespec ts;
228 int64_t sec;
229 long nsec;
230 struct lwp *l, *initlwp;
231 int i, numcpu;
232
233 /* not reentrant */
234 if (rump_inited)
235 return 0;
236 else if (rump_inited == -1)
237 panic("rump_init: host process restart required");
238 else
239 rump_inited = 1;
240
241 /* initialize hypervisor */
242 if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) {
243 rumpuser_dprintf("rumpuser init failed\n");
244 return EINVAL;
245 }
246
247 /* init minimal lwp/cpu context */
248 rump_lwproc_init();
249 l = &lwp0;
250 l->l_cpu = l->l_target_cpu = &rump_bootcpu;
251 rump_lwproc_curlwp_set(l);
252
253 /* retrieve env vars which affect the early stage of bootstrap */
254 if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) {
255 rump_threads = *buf != '0';
256 }
257 if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) {
258 if (*buf != '0')
259 boothowto = AB_VERBOSE;
260 }
261
262 if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0)
263 panic("mandatory hypervisor configuration (NCPU) missing");
264 numcpu = strtoll(buf, NULL, 10);
265 if (numcpu < 1) {
266 panic("rump kernels are not lightweight enough for \"%d\" CPUs",
267 numcpu);
268 }
269
270 rump_thread_init();
271 rump_cpus_bootstrap(&numcpu);
272
273 rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec);
274 boottime.tv_sec = sec;
275 boottime.tv_nsec = nsec;
276
277 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
278 aprint_verbose("%s%s", copyright, version);
279
280 rump_intr_init(numcpu);
281
282 rump_tsleep_init();
283
284 rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN);
285 ksyms_init();
286 uvm_init();
287 evcnt_init();
288
289 kcpuset_sysinit();
290 once_init();
291 kernconfig_lock_init();
292 prop_kern_init();
293
294 kmem_init();
295
296 uvm_ra_init();
297 uao_init();
298
299 mutex_obj_init();
300 rw_obj_init();
301 callout_startup();
302
303 kprintf_init();
304 pserialize_init();
305
306 kauth_init();
307
308 secmodel_init();
309 sysctl_init();
310 /*
311 * The above call to sysctl_init() only initializes sysctl nodes
312 * from link sets. Initialize sysctls in case we used ctors.
313 */
314 #ifdef RUMP_USE_CTOR
315 {
316 struct sysctl_setup_chain *ssc;
317
318 while ((ssc = LIST_FIRST(&sysctl_boot_chain)) != NULL) {
319 LIST_REMOVE(ssc, ssc_entries);
320 ssc->ssc_func(NULL);
321 }
322 }
323 #endif /* RUMP_USE_CTOR */
324
325 rnd_init();
326 cprng_init();
327 kern_cprng = cprng_strong_create("kernel", IPL_VM,
328 CPRNG_INIT_ANY|CPRNG_REKEY_ANY);
329
330 rump_hyperentropy_init();
331
332 procinit();
333 proc0_init();
334 uid_init();
335 chgproccnt(0, 1);
336
337 l->l_proc = &proc0;
338 lwp_update_creds(l);
339
340 lwpinit_specificdata();
341 lwp_initspecific(&lwp0);
342
343 loginit();
344
345 rump_biglock_init();
346
347 rump_scheduler_init(numcpu);
348 /* revert temporary context and schedule a semireal context */
349 rump_lwproc_curlwp_clear(l);
350 initproc = &proc0; /* borrow proc0 before we get initproc started */
351 rump_schedule();
352 bootlwp = curlwp;
353
354 percpu_init();
355 inittimecounter();
356 ntp_init();
357
358 #ifdef KTRACE
359 ktrinit();
360 #endif
361
362 ts = boottime;
363 tc_setclock(&ts);
364
365 extern krwlock_t exec_lock;
366 rw_init(&exec_lock);
367
368 /* we are mostly go. do per-cpu subsystem init */
369 for (i = 0; i < numcpu; i++) {
370 struct cpu_info *ci = cpu_lookup(i);
371
372 /* attach non-bootstrap CPUs */
373 if (i > 0) {
374 rump_cpu_attach(ci);
375 ncpu++;
376 }
377
378 callout_init_cpu(ci);
379 softint_init(ci);
380 xc_init_cpu(ci);
381 pool_cache_cpu_init(ci);
382 selsysinit(ci);
383 percpu_init_cpu(ci);
384
385 TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
386 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
387
388 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
389 }
390 ncpuonline = ncpu;
391
392 /* Once all CPUs are detected, initialize the per-CPU cprng_fast. */
393 cprng_fast_init();
394
395 /* CPUs are up. allow kernel threads to run */
396 rump_thread_allow(NULL);
397
398 rnd_init_softint();
399
400 kqueue_init();
401 iostat_init();
402 fd_sys_init();
403 module_init();
404 devsw_init();
405 pipe_init();
406 resource_init();
407 procinit_sysctl();
408 time_init();
409 time_init2();
410
411 /* start page baroness */
412 if (rump_threads) {
413 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
414 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
415 panic("pagedaemon create failed");
416 } else
417 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
418
419 /* process dso's */
420 rumpuser_dl_bootstrap(add_linkedin_modules,
421 rump_kernelfsym_load, rump_component_load);
422
423 rump_component_addlocal();
424 rump_component_init(RUMP_COMPONENT_KERN);
425
426 /* initialize factions, if present */
427 rump_component_init(RUMP__FACTION_VFS);
428 /* pnbuf_cache is used even without vfs */
429 if (rump_component_count(RUMP__FACTION_VFS) == 0) {
430 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
431 NULL, IPL_NONE, NULL, NULL, NULL);
432 }
433 rump_component_init(RUMP__FACTION_NET);
434 rump_component_init(RUMP__FACTION_DEV);
435 KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1
436 && rump_component_count(RUMP__FACTION_NET) <= 1
437 && rump_component_count(RUMP__FACTION_DEV) <= 1);
438
439 rump_component_init(RUMP_COMPONENT_KERN_VFS);
440
441 /*
442 * if we initialized the tty component above, the tyttymtx is
443 * now initialized. otherwise, we need to initialize it.
444 */
445 if (!rump_ttycomponent)
446 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
447
448 cold = 0;
449
450 /* aieeeedondest */
451 if (rump_threads) {
452 if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
453 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
454 panic("aiodoned");
455 }
456
457 sysctl_finalize();
458
459 module_init_class(MODULE_CLASS_ANY);
460
461 if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME,
462 hostname, MAXHOSTNAMELEN) != 0) {
463 panic("mandatory hypervisor configuration (HOSTNAME) missing");
464 }
465 hostnamelen = strlen(hostname);
466
467 sigemptyset(&sigcantmask);
468
469 if (rump_threads)
470 vmem_rehash_start();
471
472 /*
473 * Create init (proc 1), used to attach implicit threads in rump.
474 * (note: must be done after vfsinit to get cwdi)
475 */
476 initlwp = rump__lwproc_alloclwp(NULL);
477 mutex_enter(proc_lock);
478 initproc = proc_find_raw(1);
479 mutex_exit(proc_lock);
480 if (initproc == NULL)
481 panic("where in the world is initproc?");
482 strlcpy(initproc->p_comm, "rumplocal", sizeof(initproc->p_comm));
483
484 rump_component_init(RUMP_COMPONENT_POSTINIT);
485
486 /* load syscalls */
487 rump_component_init(RUMP_COMPONENT_SYSCALL);
488
489 /* component inits done */
490 bootlwp = NULL;
491
492 /* open 0/1/2 for init */
493 KASSERT(rump_lwproc_curlwp() == NULL);
494 rump_lwproc_switch(initlwp);
495 rump_consdev_init();
496 rump_lwproc_switch(NULL);
497
498 /* release cpu */
499 rump_unschedule();
500
501 return 0;
502 }
503 /* historic compat */
504 __strong_alias(rump__init,rump_init);
505
506 static int compcounter[RUMP_COMPONENT_MAX];
507 static int compinited[RUMP_COMPONENT_MAX];
508
509 /*
510 * Yea, this is O(n^2), but we're only looking at a handful of components.
511 * Components are always initialized from the thread that called rump_init().
512 */
513 static LIST_HEAD(, rump_component) rchead = LIST_HEAD_INITIALIZER(rchead);
514
515 #ifdef RUMP_USE_CTOR
516 struct modinfo_boot_chain modinfo_boot_chain \
517 = LIST_HEAD_INITIALIZER(modinfo_boot_chain);
518
519 static void
rump_component_addlocal(void)520 rump_component_addlocal(void)
521 {
522 struct modinfo_chain *mc;
523
524 while ((mc = LIST_FIRST(&modinfo_boot_chain)) != NULL) {
525 LIST_REMOVE(mc, mc_entries);
526 module_builtin_add(&mc->mc_info, 1, false);
527 }
528 }
529
530 #else /* RUMP_USE_CTOR */
531
532 static void
rump_component_addlocal(void)533 rump_component_addlocal(void)
534 {
535 __link_set_decl(rump_components, struct rump_component);
536 struct rump_component *const *rc;
537
538 __link_set_foreach(rc, rump_components) {
539 rump_component_load(*rc);
540 }
541 }
542 #endif /* RUMP_USE_CTOR */
543
544 void
rump_component_load(const struct rump_component * rc_const)545 rump_component_load(const struct rump_component *rc_const)
546 {
547 struct rump_component *rc, *rc_iter;
548
549 /* time for rump component loading and unloading has passed */
550 if (!cold)
551 return;
552
553 /*
554 * XXX: this is ok since the "const" was removed from the
555 * definition of RUMP_COMPONENT().
556 *
557 * However, to preserve the hypercall interface, the const
558 * remains here. This can be fixed in the next hypercall revision.
559 */
560 rc = __UNCONST(rc_const);
561
562 KASSERT(!rump_inited || curlwp == bootlwp);
563
564 LIST_FOREACH(rc_iter, &rchead, rc_entries) {
565 if (rc_iter == rc)
566 return;
567 }
568
569 LIST_INSERT_HEAD(&rchead, rc, rc_entries);
570 KASSERT(rc->rc_type < RUMP_COMPONENT_MAX);
571 compcounter[rc->rc_type]++;
572 }
573
574 void
rump_component_unload(struct rump_component * rc)575 rump_component_unload(struct rump_component *rc)
576 {
577
578 /*
579 * Checking for cold is enough because rump_init() both
580 * flips it and handles component loading.
581 */
582 if (!cold)
583 return;
584
585 LIST_REMOVE(rc, rc_entries);
586 }
587
588 int
rump_component_count(enum rump_component_type type)589 rump_component_count(enum rump_component_type type)
590 {
591
592 KASSERT(curlwp == bootlwp);
593 KASSERT(type < RUMP_COMPONENT_MAX);
594 return compcounter[type];
595 }
596
597 void
rump_component_init(enum rump_component_type type)598 rump_component_init(enum rump_component_type type)
599 {
600 const struct rump_component *rc, *rc_safe;
601
602 KASSERT(curlwp == bootlwp);
603 KASSERT(!compinited[type]);
604 LIST_FOREACH_SAFE(rc, &rchead, rc_entries, rc_safe) {
605 if (rc->rc_type == type) {
606 rc->rc_init();
607 LIST_REMOVE(rc, rc_entries);
608 }
609 }
610 compinited[type] = 1;
611 }
612
613 /*
614 * Initialize a module which has already been loaded and linked
615 * with dlopen(). This is fundamentally the same as a builtin module.
616 *
617 * XXX: this interface does not really work in the RUMP_USE_CTOR case,
618 * but I'm not sure it's anything to cry about. In feeling blue,
619 * things could somehow be handled via modinfo_boot_chain.
620 */
621 int
rump_module_init(const struct modinfo * const * mip,size_t nmodinfo)622 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
623 {
624
625 return module_builtin_add(mip, nmodinfo, true);
626 }
627
628 /*
629 * Finish module (flawless victory, fatality!).
630 */
631 int
rump_module_fini(const struct modinfo * mi)632 rump_module_fini(const struct modinfo *mi)
633 {
634
635 return module_builtin_remove(mi, true);
636 }
637
638 /*
639 * Add loaded and linked module to the builtin list. It will
640 * later be initialized with module_init_class().
641 */
642
643 static void
add_linkedin_modules(const struct modinfo * const * mip,size_t nmodinfo)644 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
645 {
646
647 module_builtin_add(mip, nmodinfo, false);
648 }
649
650 int
rump_kernelfsym_load(void * symtab,uint64_t symsize,char * strtab,uint64_t strsize)651 rump_kernelfsym_load(void *symtab, uint64_t symsize,
652 char *strtab, uint64_t strsize)
653 {
654 static int inited = 0;
655 Elf64_Ehdr ehdr;
656
657 if (inited)
658 return EBUSY;
659 inited = 1;
660
661 /*
662 * Use 64bit header since it's bigger. Shouldn't make a
663 * difference, since we're passing in all zeroes anyway.
664 */
665 memset(&ehdr, 0, sizeof(ehdr));
666 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
667
668 return 0;
669 }
670
671 int
rump_boot_gethowto()672 rump_boot_gethowto()
673 {
674
675 return boothowto;
676 }
677
678 void
rump_boot_sethowto(int howto)679 rump_boot_sethowto(int howto)
680 {
681
682 boothowto = howto;
683 }
684
685 int
rump_getversion(void)686 rump_getversion(void)
687 {
688
689 return __NetBSD_Version__;
690 }
691 /* compat */
692 __strong_alias(rump_pub_getversion,rump_getversion);
693
694 /*
695 * Note: may be called unscheduled. Not fully safe since no locking
696 * of allevents (currently that's not even available).
697 */
698 void
rump_printevcnts()699 rump_printevcnts()
700 {
701 struct evcnt *ev;
702
703 TAILQ_FOREACH(ev, &allevents, ev_list)
704 rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
705 ev->ev_group, ev->ev_name, ev->ev_count);
706 }
707
708 /*
709 * If you use this interface ... well ... all bets are off.
710 * The original purpose is for the p2k fs server library to be
711 * able to use the same pid/lid for VOPs as the host kernel.
712 */
713 void
rump_allbetsareoff_setid(pid_t pid,int lid)714 rump_allbetsareoff_setid(pid_t pid, int lid)
715 {
716 struct lwp *l = curlwp;
717 struct proc *p = l->l_proc;
718
719 l->l_lid = lid;
720 p->p_pid = pid;
721 }
722
723 #include <sys/pserialize.h>
724
725 static void
ipiemu(void * a1,void * a2)726 ipiemu(void *a1, void *a2)
727 {
728
729 xc__highpri_intr(NULL);
730 pserialize_switchpoint();
731 }
732
733 void
rump_xc_highpri(struct cpu_info * ci)734 rump_xc_highpri(struct cpu_info *ci)
735 {
736
737 if (ci)
738 xc_unicast(0, ipiemu, NULL, NULL, ci);
739 else
740 xc_broadcast(0, ipiemu, NULL, NULL);
741 }
742
743 int
rump_syscall(int num,void * data,size_t dlen,register_t * retval)744 rump_syscall(int num, void *data, size_t dlen, register_t *retval)
745 {
746 struct proc *p;
747 struct emul *e;
748 struct sysent *callp;
749 const int *etrans = NULL;
750 int rv;
751
752 rump_schedule();
753 p = curproc;
754 e = p->p_emul;
755 #ifndef __HAVE_MINIMAL_EMUL
756 KASSERT(num > 0 && num < e->e_nsysent);
757 #endif
758 callp = e->e_sysent + num;
759
760 rv = sy_invoke(callp, curlwp, data, retval, num);
761
762 /*
763 * I hope that (!__HAVE_MINIMAL_EMUL || __HAVE_SYSCALL_INTERN) is
764 * an invariant ...
765 */
766 #if !defined(__HAVE_MINIMAL_EMUL)
767 etrans = e->e_errno;
768 #elif defined(__HAVE_SYSCALL_INTERN)
769 etrans = p->p_emuldata;
770 #endif
771
772 if (etrans) {
773 rv = etrans[rv];
774 /*
775 * XXX: small hack since Linux etrans vectors on some
776 * archs contain negative errnos, but rump_syscalls
777 * uses the -1 + errno ABI. Note that these
778 * negative values are always the result of translation,
779 * otherwise the above translation method would not
780 * work very well.
781 */
782 if (rv < 0)
783 rv = -rv;
784 }
785 rump_unschedule();
786
787 return rv;
788 }
789
790 void
rump_syscall_boot_establish(const struct rump_onesyscall * calls,size_t ncall)791 rump_syscall_boot_establish(const struct rump_onesyscall *calls, size_t ncall)
792 {
793 struct sysent *callp;
794 size_t i;
795
796 for (i = 0; i < ncall; i++) {
797 callp = rump_sysent + calls[i].ros_num;
798 KASSERT(bootlwp != NULL
799 && callp->sy_call == (sy_call_t *)enosys);
800 callp->sy_call = calls[i].ros_handler;
801 }
802 }
803
804 struct rump_boot_etfs *ebstart;
805 void
rump_boot_etfs_register(struct rump_boot_etfs * eb)806 rump_boot_etfs_register(struct rump_boot_etfs *eb)
807 {
808
809 /*
810 * Could use atomics, but, since caller would need to synchronize
811 * against calling rump_init() anyway, easier to just specify the
812 * interface as "caller serializes". This solve-by-specification
813 * approach avoids the grey area of using atomics before rump_init()
814 * runs.
815 */
816 eb->_eb_next = ebstart;
817 eb->eb_status = -1;
818 ebstart = eb;
819 }
820