rump.c revision 1.244 1 1.244 pgoyette /* $NetBSD: rump.c,v 1.244 2012/08/16 17:47:47 pgoyette Exp $ */
2 1.1 pooka
3 1.1 pooka /*
4 1.233 pooka * Copyright (c) 2007-2011 Antti Kantee. All Rights Reserved.
5 1.1 pooka *
6 1.1 pooka * Redistribution and use in source and binary forms, with or without
7 1.1 pooka * modification, are permitted provided that the following conditions
8 1.1 pooka * are met:
9 1.1 pooka * 1. Redistributions of source code must retain the above copyright
10 1.1 pooka * notice, this list of conditions and the following disclaimer.
11 1.1 pooka * 2. Redistributions in binary form must reproduce the above copyright
12 1.1 pooka * notice, this list of conditions and the following disclaimer in the
13 1.1 pooka * documentation and/or other materials provided with the distribution.
14 1.1 pooka *
15 1.1 pooka * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
16 1.1 pooka * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
17 1.1 pooka * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
18 1.1 pooka * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 1.1 pooka * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 1.1 pooka * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
21 1.1 pooka * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 1.1 pooka * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 1.1 pooka * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 1.1 pooka * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 1.1 pooka * SUCH DAMAGE.
26 1.1 pooka */
27 1.1 pooka
28 1.78 pooka #include <sys/cdefs.h>
29 1.244 pgoyette __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.244 2012/08/16 17:47:47 pgoyette Exp $");
30 1.165 pooka
31 1.165 pooka #include <sys/systm.h>
32 1.165 pooka #define ELFSIZE ARCH_ELFSIZE
33 1.78 pooka
34 1.1 pooka #include <sys/param.h>
35 1.64 pooka #include <sys/atomic.h>
36 1.75 pooka #include <sys/buf.h>
37 1.66 pooka #include <sys/callout.h>
38 1.79 pooka #include <sys/conf.h>
39 1.27 pooka #include <sys/cpu.h>
40 1.144 pooka #include <sys/device.h>
41 1.100 pooka #include <sys/evcnt.h>
42 1.98 pooka #include <sys/event.h>
43 1.137 pooka #include <sys/exec_elf.h>
44 1.1 pooka #include <sys/filedesc.h>
45 1.72 pooka #include <sys/iostat.h>
46 1.1 pooka #include <sys/kauth.h>
47 1.80 pooka #include <sys/kernel.h>
48 1.14 pooka #include <sys/kmem.h>
49 1.81 pooka #include <sys/kprintf.h>
50 1.175 pooka #include <sys/kthread.h>
51 1.82 pooka #include <sys/ksyms.h>
52 1.81 pooka #include <sys/msgbuf.h>
53 1.49 simonb #include <sys/module.h>
54 1.234 pooka #include <sys/namei.h>
55 1.71 pooka #include <sys/once.h>
56 1.65 pooka #include <sys/percpu.h>
57 1.139 pooka #include <sys/pipe.h>
58 1.162 pooka #include <sys/pool.h>
59 1.244 pgoyette #include <sys/pserialize.h>
60 1.1 pooka #include <sys/queue.h>
61 1.121 pooka #include <sys/reboot.h>
62 1.1 pooka #include <sys/resourcevar.h>
63 1.27 pooka #include <sys/select.h>
64 1.87 pooka #include <sys/sysctl.h>
65 1.96 pooka #include <sys/syscall.h>
66 1.169 pooka #include <sys/syscallvar.h>
67 1.160 pooka #include <sys/timetc.h>
68 1.87 pooka #include <sys/tty.h>
69 1.68 pooka #include <sys/uidinfo.h>
70 1.89 pooka #include <sys/vmem.h>
71 1.143 pooka #include <sys/xcall.h>
72 1.216 pooka #include <sys/simplelock.h>
73 1.235 tls #include <sys/cprng.h>
74 1.1 pooka
75 1.48 pooka #include <rump/rumpuser.h>
76 1.48 pooka
77 1.117 elad #include <secmodel/suser/suser.h>
78 1.101 pooka
79 1.114 pooka #include <prop/proplib.h>
80 1.114 pooka
81 1.174 pooka #include <uvm/uvm_extern.h>
82 1.136 pooka #include <uvm/uvm_readahead.h>
83 1.136 pooka
84 1.8 pooka #include "rump_private.h"
85 1.71 pooka #include "rump_net_private.h"
86 1.75 pooka #include "rump_vfs_private.h"
87 1.112 pooka #include "rump_dev_private.h"
88 1.1 pooka
89 1.164 pooka char machine[] = MACHINE;
90 1.1 pooka
91 1.193 pooka struct proc *initproc;
92 1.101 pooka
93 1.144 pooka struct device rump_rootdev = {
94 1.144 pooka .dv_class = DV_VIRTUAL
95 1.144 pooka };
96 1.144 pooka
97 1.54 pooka #ifdef RUMP_WITHOUT_THREADS
98 1.54 pooka int rump_threads = 0;
99 1.54 pooka #else
100 1.54 pooka int rump_threads = 1;
101 1.54 pooka #endif
102 1.54 pooka
103 1.199 pooka static int rump_proxy_syscall(int, void *, register_t *);
104 1.226 pooka static int rump_proxy_rfork(void *, int, const char *);
105 1.232 pooka static void rump_proxy_lwpexit(void);
106 1.230 pooka static void rump_proxy_execnotify(const char *);
107 1.199 pooka
108 1.156 pooka static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */
109 1.156 pooka
110 1.217 pooka #ifdef LOCKDEBUG
111 1.217 pooka const int rump_lockdebug = 1;
112 1.217 pooka #else
113 1.217 pooka const int rump_lockdebug = 0;
114 1.217 pooka #endif
115 1.218 pooka bool rump_ttycomponent = false;
116 1.217 pooka
117 1.14 pooka static void
118 1.14 pooka rump_aiodone_worker(struct work *wk, void *dummy)
119 1.14 pooka {
120 1.14 pooka struct buf *bp = (struct buf *)wk;
121 1.14 pooka
122 1.14 pooka KASSERT(&bp->b_work == wk);
123 1.14 pooka bp->b_iodone(bp);
124 1.14 pooka }
125 1.14 pooka
126 1.51 pooka static int rump_inited;
127 1.27 pooka
128 1.162 pooka /*
129 1.162 pooka * Make sure pnbuf_cache is available even without vfs
130 1.162 pooka */
131 1.162 pooka int rump_initpnbufpool(void);
132 1.162 pooka int rump_initpnbufpool(void)
133 1.162 pooka {
134 1.162 pooka
135 1.162 pooka pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
136 1.162 pooka NULL, IPL_NONE, NULL, NULL, NULL);
137 1.162 pooka return EOPNOTSUPP;
138 1.162 pooka }
139 1.162 pooka
140 1.126 pooka int rump__unavailable(void);
141 1.126 pooka int rump__unavailable() {return EOPNOTSUPP;}
142 1.86 pooka __weak_alias(rump_net_init,rump__unavailable);
143 1.162 pooka __weak_alias(rump_vfs_init,rump_initpnbufpool);
144 1.112 pooka __weak_alias(rump_dev_init,rump__unavailable);
145 1.75 pooka
146 1.121 pooka __weak_alias(rump_vfs_fini,rump__unavailable);
147 1.121 pooka
148 1.113 pooka __weak_alias(biodone,rump__unavailable);
149 1.126 pooka __weak_alias(sopoll,rump__unavailable);
150 1.113 pooka
151 1.188 pooka __weak_alias(rump_vfs_drainbufs,rump__unavailable);
152 1.188 pooka
153 1.92 pooka void rump__unavailable_vfs_panic(void);
154 1.92 pooka void rump__unavailable_vfs_panic() {panic("vfs component not available");}
155 1.132 pooka __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
156 1.92 pooka
157 1.211 pooka /* easier to write vfs-less clients */
158 1.211 pooka __weak_alias(rump_pub_etfs_register,rump__unavailable);
159 1.228 pooka __weak_alias(rump_pub_etfs_register_withsize,rump__unavailable);
160 1.211 pooka __weak_alias(rump_pub_etfs_remove,rump__unavailable);
161 1.211 pooka
162 1.147 pooka rump_proc_vfs_init_fn rump_proc_vfs_init;
163 1.147 pooka rump_proc_vfs_release_fn rump_proc_vfs_release;
164 1.71 pooka
165 1.155 pooka static void add_linkedin_modules(const struct modinfo *const *, size_t);
166 1.155 pooka
167 1.158 pooka /*
168 1.158 pooka * Create kern.hostname. why only this you ask. well, init_sysctl
169 1.158 pooka * is a kitchen sink in need of some gardening. but i want to use
170 1.158 pooka * kern.hostname today.
171 1.158 pooka */
172 1.158 pooka static void
173 1.158 pooka mksysctls(void)
174 1.158 pooka {
175 1.158 pooka
176 1.158 pooka sysctl_createv(NULL, 0, NULL, NULL,
177 1.158 pooka CTLFLAG_PERMANENT, CTLTYPE_NODE, "kern", NULL,
178 1.158 pooka NULL, 0, NULL, 0, CTL_KERN, CTL_EOL);
179 1.158 pooka
180 1.158 pooka /* XXX: setting hostnamelen is missing */
181 1.158 pooka sysctl_createv(NULL, 0, NULL, NULL,
182 1.158 pooka CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_STRING, "hostname",
183 1.158 pooka SYSCTL_DESCR("System hostname"), NULL, 0,
184 1.241 dsl hostname, MAXHOSTNAMELEN, CTL_KERN, KERN_HOSTNAME, CTL_EOL);
185 1.158 pooka }
186 1.158 pooka
187 1.201 pooka /* there's no convenient kernel entry point for this, so just craft out own */
188 1.201 pooka static pid_t
189 1.201 pooka spgetpid(void)
190 1.201 pooka {
191 1.201 pooka
192 1.201 pooka return curproc->p_pid;
193 1.201 pooka }
194 1.201 pooka
195 1.197 pooka static const struct rumpuser_sp_ops spops = {
196 1.197 pooka .spop_schedule = rump_schedule,
197 1.197 pooka .spop_unschedule = rump_unschedule,
198 1.197 pooka .spop_lwproc_switch = rump_lwproc_switch,
199 1.197 pooka .spop_lwproc_release = rump_lwproc_releaselwp,
200 1.214 pooka .spop_lwproc_rfork = rump_proxy_rfork,
201 1.201 pooka .spop_lwproc_newlwp = rump_lwproc_newlwp,
202 1.197 pooka .spop_lwproc_curlwp = rump_lwproc_curlwp,
203 1.232 pooka .spop_lwpexit = rump_proxy_lwpexit,
204 1.199 pooka .spop_syscall = rump_proxy_syscall,
205 1.230 pooka .spop_execnotify = rump_proxy_execnotify,
206 1.201 pooka .spop_getpid = spgetpid,
207 1.197 pooka };
208 1.197 pooka
209 1.61 pooka int
210 1.207 pooka rump_daemonize_begin(void)
211 1.207 pooka {
212 1.207 pooka
213 1.207 pooka if (rump_inited)
214 1.207 pooka return EALREADY;
215 1.207 pooka
216 1.207 pooka return rumpuser_daemonize_begin();
217 1.207 pooka }
218 1.207 pooka
219 1.207 pooka int
220 1.207 pooka rump_daemonize_done(int error)
221 1.207 pooka {
222 1.207 pooka
223 1.207 pooka return rumpuser_daemonize_done(error);
224 1.207 pooka }
225 1.207 pooka
226 1.207 pooka int
227 1.83 pooka rump__init(int rump_version)
228 1.1 pooka {
229 1.36 pooka char buf[256];
230 1.160 pooka struct timespec ts;
231 1.157 pooka uint64_t sec, nsec;
232 1.14 pooka struct lwp *l;
233 1.166 pooka int i, numcpu;
234 1.1 pooka int error;
235 1.1 pooka
236 1.121 pooka /* not reentrant */
237 1.27 pooka if (rump_inited)
238 1.61 pooka return 0;
239 1.121 pooka else if (rump_inited == -1)
240 1.121 pooka panic("rump_init: host process restart required");
241 1.121 pooka else
242 1.121 pooka rump_inited = 1;
243 1.27 pooka
244 1.177 pooka if (rumpuser_getversion() != RUMPUSER_VERSION) {
245 1.177 pooka /* let's hope the ABI of rumpuser_dprintf is the same ;) */
246 1.177 pooka rumpuser_dprintf("rumpuser version mismatch: %d vs. %d\n",
247 1.177 pooka rumpuser_getversion(), RUMPUSER_VERSION);
248 1.177 pooka return EPROGMISMATCH;
249 1.177 pooka }
250 1.177 pooka
251 1.149 pooka if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) {
252 1.149 pooka if (*buf != '0')
253 1.149 pooka boothowto = AB_VERBOSE;
254 1.149 pooka }
255 1.149 pooka
256 1.167 pooka if (rumpuser_getenv("RUMP_NCPU", buf, sizeof(buf), &error) == 0)
257 1.167 pooka error = 0;
258 1.167 pooka if (error == 0) {
259 1.167 pooka numcpu = strtoll(buf, NULL, 10);
260 1.167 pooka if (numcpu < 1)
261 1.167 pooka numcpu = 1;
262 1.167 pooka } else {
263 1.167 pooka numcpu = rumpuser_getnhostcpu();
264 1.167 pooka }
265 1.204 pooka rump_cpus_bootstrap(&numcpu);
266 1.165 pooka
267 1.157 pooka rumpuser_gettime(&sec, &nsec, &error);
268 1.157 pooka boottime.tv_sec = sec;
269 1.157 pooka boottime.tv_nsec = nsec;
270 1.157 pooka
271 1.156 pooka initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
272 1.156 pooka aprint_verbose("%s%s", copyright, version);
273 1.148 pooka
274 1.61 pooka if (rump_version != RUMP_VERSION) {
275 1.61 pooka printf("rump version mismatch, %d vs. %d\n",
276 1.61 pooka rump_version, RUMP_VERSION);
277 1.61 pooka return EPROGMISMATCH;
278 1.61 pooka }
279 1.61 pooka
280 1.54 pooka if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
281 1.54 pooka rump_threads = *buf != '0';
282 1.54 pooka }
283 1.124 pooka rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
284 1.124 pooka rump_threads);
285 1.187 pooka rump_intr_init(numcpu);
286 1.173 pooka rump_tsleep_init();
287 1.36 pooka
288 1.131 pooka /* init minimal lwp/cpu context */
289 1.131 pooka l = &lwp0;
290 1.131 pooka l->l_lid = 1;
291 1.172 pooka l->l_cpu = l->l_target_cpu = rump_cpu;
292 1.178 pooka l->l_fd = &filedesc0;
293 1.131 pooka rumpuser_set_curlwp(l);
294 1.131 pooka
295 1.208 pooka rumpuser_mutex_init(&rump_giantlock);
296 1.82 pooka ksyms_init();
297 1.174 pooka uvm_init();
298 1.100 pooka evcnt_init();
299 1.89 pooka
300 1.89 pooka once_init();
301 1.181 pgoyette kernconfig_lock_init();
302 1.114 pooka prop_kern_init();
303 1.63 pooka
304 1.52 pooka kmem_init();
305 1.102 pooka
306 1.136 pooka uvm_ra_init();
307 1.185 pooka uao_init();
308 1.136 pooka
309 1.133 pooka mutex_obj_init();
310 1.143 pooka callout_startup();
311 1.133 pooka
312 1.81 pooka kprintf_init();
313 1.243 pgoyette pserialize_init();
314 1.81 pooka loginit();
315 1.52 pooka
316 1.59 pooka kauth_init();
317 1.59 pooka
318 1.238 jym secmodel_init();
319 1.238 jym
320 1.240 njoly rnd_init();
321 1.240 njoly
322 1.236 njoly /*
323 1.237 tls * Create the kernel cprng. Yes, it's currently stubbed out
324 1.237 tls * to arc4random() for RUMP, but this won't always be so.
325 1.236 njoly */
326 1.237 tls kern_cprng = cprng_strong_create("kernel", IPL_VM,
327 1.237 tls CPRNG_INIT_ANY|CPRNG_REKEY_ANY);
328 1.236 njoly
329 1.184 pooka procinit();
330 1.184 pooka proc0_init();
331 1.242 christos sysctl_init();
332 1.192 pooka uid_init();
333 1.192 pooka chgproccnt(0, 1);
334 1.184 pooka
335 1.178 pooka l->l_proc = &proc0;
336 1.184 pooka lwp_update_creds(l);
337 1.176 pooka
338 1.159 pooka lwpinit_specificdata();
339 1.176 pooka lwp_initspecific(&lwp0);
340 1.36 pooka
341 1.186 pooka rump_scheduler_init(numcpu);
342 1.193 pooka /* revert temporary context and schedule a semireal context */
343 1.131 pooka rumpuser_set_curlwp(NULL);
344 1.193 pooka initproc = &proc0; /* borrow proc0 before we get initproc started */
345 1.124 pooka rump_schedule();
346 1.124 pooka
347 1.150 pooka percpu_init();
348 1.160 pooka inittimecounter();
349 1.160 pooka ntp_init();
350 1.160 pooka
351 1.160 pooka rumpuser_gettime(&sec, &nsec, &error);
352 1.160 pooka ts.tv_sec = sec;
353 1.160 pooka ts.tv_nsec = nsec;
354 1.160 pooka tc_setclock(&ts);
355 1.150 pooka
356 1.143 pooka /* we are mostly go. do per-cpu subsystem init */
357 1.186 pooka for (i = 0; i < numcpu; i++) {
358 1.143 pooka struct cpu_info *ci = cpu_lookup(i);
359 1.143 pooka
360 1.186 pooka /* attach non-bootstrap CPUs */
361 1.186 pooka if (i > 0) {
362 1.186 pooka rump_cpu_attach(ci);
363 1.186 pooka ncpu++;
364 1.186 pooka }
365 1.186 pooka
366 1.143 pooka callout_init_cpu(ci);
367 1.143 pooka softint_init(ci);
368 1.143 pooka xc_init_cpu(ci);
369 1.143 pooka pool_cache_cpu_init(ci);
370 1.143 pooka selsysinit(ci);
371 1.150 pooka percpu_init_cpu(ci);
372 1.182 pooka
373 1.216 pooka TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
374 1.216 pooka __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
375 1.216 pooka
376 1.182 pooka aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
377 1.143 pooka }
378 1.43 pooka
379 1.191 pooka mksysctls();
380 1.98 pooka kqueue_init();
381 1.72 pooka iostat_init();
382 1.43 pooka fd_sys_init();
383 1.44 ad module_init();
384 1.79 pooka devsw_init();
385 1.139 pooka pipe_init();
386 1.162 pooka resource_init();
387 1.224 pooka procinit_sysctl();
388 1.1 pooka
389 1.175 pooka /* start page baroness */
390 1.175 pooka if (rump_threads) {
391 1.175 pooka if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
392 1.175 pooka uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
393 1.175 pooka panic("pagedaemon create failed");
394 1.175 pooka } else
395 1.175 pooka uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
396 1.175 pooka
397 1.175 pooka /* process dso's */
398 1.155 pooka rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load);
399 1.155 pooka
400 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN);
401 1.179 pooka
402 1.75 pooka /* these do nothing if not present */
403 1.75 pooka rump_vfs_init();
404 1.75 pooka rump_net_init();
405 1.112 pooka rump_dev_init();
406 1.179 pooka
407 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN_VFS);
408 1.179 pooka
409 1.218 pooka /*
410 1.218 pooka * if we initialized the tty component above, the tyttymtx is
411 1.218 pooka * now initialized. otherwise, we need to initialize it.
412 1.218 pooka */
413 1.218 pooka if (!rump_ttycomponent)
414 1.218 pooka mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
415 1.218 pooka
416 1.112 pooka cold = 0;
417 1.31 pooka
418 1.14 pooka /* aieeeedondest */
419 1.54 pooka if (rump_threads) {
420 1.54 pooka if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
421 1.110 pooka rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
422 1.54 pooka panic("aiodoned");
423 1.54 pooka }
424 1.14 pooka
425 1.115 pooka sysctl_finalize();
426 1.115 pooka
427 1.155 pooka module_init_class(MODULE_CLASS_ANY);
428 1.140 pooka
429 1.1 pooka rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
430 1.1 pooka hostnamelen = strlen(hostname);
431 1.24 pooka
432 1.24 pooka sigemptyset(&sigcantmask);
433 1.27 pooka
434 1.89 pooka if (rump_threads)
435 1.89 pooka vmem_rehash_start();
436 1.89 pooka
437 1.193 pooka /*
438 1.193 pooka * Create init, used to attach implicit threads in rump.
439 1.193 pooka * (note: must be done after vfsinit to get cwdi)
440 1.193 pooka */
441 1.193 pooka (void)rump__lwproc_alloclwp(NULL); /* dummy thread for initproc */
442 1.193 pooka mutex_enter(proc_lock);
443 1.193 pooka initproc = proc_find_raw(1);
444 1.193 pooka mutex_exit(proc_lock);
445 1.193 pooka if (initproc == NULL)
446 1.193 pooka panic("where in the world is initproc?");
447 1.193 pooka
448 1.213 pooka /*
449 1.213 pooka * Adjust syscall vector in case factions were dlopen()'d
450 1.213 pooka * before calling rump_init().
451 1.213 pooka * (modules will handle dynamic syscalls the usual way)
452 1.213 pooka *
453 1.213 pooka * Note: this will adjust the function vectors of
454 1.213 pooka * syscalls which use a funcalias (getpid etc.), but
455 1.213 pooka * it makes no difference.
456 1.213 pooka */
457 1.213 pooka for (i = 0; i < SYS_NSYSENT; i++) {
458 1.213 pooka void *sym;
459 1.213 pooka
460 1.213 pooka if (rump_sysent[i].sy_flags & SYCALL_NOSYS ||
461 1.213 pooka *syscallnames[i] == '#' ||
462 1.213 pooka rump_sysent[i].sy_call == sys_nomodule)
463 1.213 pooka continue;
464 1.213 pooka
465 1.222 pooka /*
466 1.222 pooka * deal with compat wrappers. makesyscalls.sh should
467 1.222 pooka * generate the necessary info instead of this hack,
468 1.222 pooka * though. ugly, fix it later.
469 1.222 pooka */
470 1.222 pooka #define CPFX "compat_"
471 1.222 pooka #define CPFXLEN (sizeof(CPFX)-1)
472 1.222 pooka if (strncmp(syscallnames[i], CPFX, CPFXLEN) == 0) {
473 1.222 pooka const char *p = syscallnames[i] + CPFXLEN;
474 1.222 pooka size_t namelen;
475 1.222 pooka
476 1.222 pooka /* skip version number */
477 1.222 pooka while (*p >= '0' && *p <= '9')
478 1.222 pooka p++;
479 1.222 pooka if (p == syscallnames[i] + CPFXLEN || *p != '_')
480 1.222 pooka panic("invalid syscall name %s\n",
481 1.222 pooka syscallnames[i]);
482 1.222 pooka
483 1.222 pooka /* skip over the next underscore */
484 1.222 pooka p++;
485 1.222 pooka namelen = p + (sizeof("rumpns_")-1) - syscallnames[i];
486 1.222 pooka
487 1.222 pooka strcpy(buf, "rumpns_");
488 1.222 pooka strcat(buf, syscallnames[i]);
489 1.222 pooka /* XXX: no strncat in the kernel */
490 1.222 pooka strcpy(buf+namelen, "sys_");
491 1.222 pooka strcat(buf, p);
492 1.222 pooka #undef CPFX
493 1.222 pooka #undef CPFXLEN
494 1.222 pooka } else {
495 1.222 pooka sprintf(buf, "rumpns_sys_%s", syscallnames[i]);
496 1.222 pooka }
497 1.213 pooka if ((sym = rumpuser_dl_globalsym(buf)) != NULL
498 1.213 pooka && sym != rump_sysent[i].sy_call) {
499 1.213 pooka #if 0
500 1.213 pooka rumpuser_dprintf("adjusting %s: %p (old %p)\n",
501 1.213 pooka syscallnames[i], sym, rump_sysent[i].sy_call);
502 1.213 pooka #endif
503 1.213 pooka rump_sysent[i].sy_call = sym;
504 1.213 pooka }
505 1.213 pooka }
506 1.213 pooka
507 1.193 pooka /* release cpu */
508 1.124 pooka rump_unschedule();
509 1.116 pooka
510 1.2 pooka return 0;
511 1.2 pooka }
512 1.2 pooka
513 1.207 pooka int
514 1.207 pooka rump_init_server(const char *url)
515 1.207 pooka {
516 1.207 pooka
517 1.212 pooka return rumpuser_sp_init(url, &spops, ostype, osrelease, MACHINE);
518 1.207 pooka }
519 1.207 pooka
520 1.121 pooka void
521 1.203 pooka cpu_reboot(int howto, char *bootstr)
522 1.121 pooka {
523 1.209 pooka int ruhow = 0;
524 1.220 pooka void *finiarg;
525 1.209 pooka
526 1.209 pooka printf("rump kernel halting...\n");
527 1.220 pooka
528 1.220 pooka if (!RUMP_LOCALPROC_P(curproc))
529 1.220 pooka finiarg = curproc->p_vmspace->vm_map.pmap;
530 1.220 pooka else
531 1.220 pooka finiarg = NULL;
532 1.121 pooka
533 1.121 pooka /* dump means we really take the dive here */
534 1.121 pooka if ((howto & RB_DUMP) || panicstr) {
535 1.209 pooka ruhow = RUMPUSER_PANIC;
536 1.209 pooka goto out;
537 1.121 pooka }
538 1.121 pooka
539 1.121 pooka /* try to sync */
540 1.121 pooka if (!((howto & RB_NOSYNC) || panicstr)) {
541 1.121 pooka rump_vfs_fini();
542 1.121 pooka }
543 1.121 pooka
544 1.121 pooka /* your wish is my command */
545 1.121 pooka if (howto & RB_HALT) {
546 1.203 pooka printf("rump kernel halted\n");
547 1.227 bouyer rumpuser_sp_fini(finiarg);
548 1.121 pooka for (;;) {
549 1.121 pooka uint64_t sec = 5, nsec = 0;
550 1.121 pooka int error;
551 1.121 pooka
552 1.121 pooka rumpuser_nanosleep(&sec, &nsec, &error);
553 1.121 pooka }
554 1.121 pooka }
555 1.203 pooka
556 1.203 pooka /* this function is __dead, we must exit */
557 1.209 pooka out:
558 1.209 pooka printf("halted\n");
559 1.227 bouyer rumpuser_sp_fini(finiarg);
560 1.209 pooka rumpuser_exit(ruhow);
561 1.121 pooka }
562 1.121 pooka
563 1.8 pooka struct uio *
564 1.123 pooka rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
565 1.8 pooka {
566 1.8 pooka struct uio *uio;
567 1.8 pooka enum uio_rw uiorw;
568 1.8 pooka
569 1.8 pooka switch (rw) {
570 1.8 pooka case RUMPUIO_READ:
571 1.8 pooka uiorw = UIO_READ;
572 1.8 pooka break;
573 1.8 pooka case RUMPUIO_WRITE:
574 1.8 pooka uiorw = UIO_WRITE;
575 1.8 pooka break;
576 1.11 pooka default:
577 1.11 pooka panic("%s: invalid rw %d", __func__, rw);
578 1.8 pooka }
579 1.8 pooka
580 1.28 pooka uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
581 1.28 pooka uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
582 1.8 pooka
583 1.8 pooka uio->uio_iov->iov_base = buf;
584 1.8 pooka uio->uio_iov->iov_len = bufsize;
585 1.8 pooka
586 1.8 pooka uio->uio_iovcnt = 1;
587 1.8 pooka uio->uio_offset = offset;
588 1.8 pooka uio->uio_resid = bufsize;
589 1.8 pooka uio->uio_rw = uiorw;
590 1.198 pooka UIO_SETUP_SYSSPACE(uio);
591 1.8 pooka
592 1.8 pooka return uio;
593 1.8 pooka }
594 1.8 pooka
595 1.8 pooka size_t
596 1.123 pooka rump_uio_getresid(struct uio *uio)
597 1.8 pooka {
598 1.8 pooka
599 1.8 pooka return uio->uio_resid;
600 1.8 pooka }
601 1.8 pooka
602 1.8 pooka off_t
603 1.123 pooka rump_uio_getoff(struct uio *uio)
604 1.8 pooka {
605 1.8 pooka
606 1.8 pooka return uio->uio_offset;
607 1.8 pooka }
608 1.8 pooka
609 1.8 pooka size_t
610 1.123 pooka rump_uio_free(struct uio *uio)
611 1.8 pooka {
612 1.8 pooka size_t resid;
613 1.8 pooka
614 1.8 pooka resid = uio->uio_resid;
615 1.28 pooka kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
616 1.28 pooka kmem_free(uio, sizeof(*uio));
617 1.8 pooka
618 1.8 pooka return resid;
619 1.8 pooka }
620 1.8 pooka
621 1.59 pooka kauth_cred_t
622 1.123 pooka rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
623 1.59 pooka {
624 1.59 pooka kauth_cred_t cred;
625 1.59 pooka int rv;
626 1.59 pooka
627 1.59 pooka cred = kauth_cred_alloc();
628 1.59 pooka kauth_cred_setuid(cred, uid);
629 1.59 pooka kauth_cred_seteuid(cred, uid);
630 1.59 pooka kauth_cred_setsvuid(cred, uid);
631 1.59 pooka kauth_cred_setgid(cred, gid);
632 1.59 pooka kauth_cred_setgid(cred, gid);
633 1.59 pooka kauth_cred_setegid(cred, gid);
634 1.59 pooka kauth_cred_setsvgid(cred, gid);
635 1.59 pooka rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
636 1.59 pooka /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */
637 1.59 pooka assert(rv == 0);
638 1.59 pooka
639 1.59 pooka return cred;
640 1.59 pooka }
641 1.59 pooka
642 1.59 pooka void
643 1.123 pooka rump_cred_put(kauth_cred_t cred)
644 1.59 pooka {
645 1.59 pooka
646 1.59 pooka kauth_cred_free(cred);
647 1.59 pooka }
648 1.59 pooka
649 1.154 pooka static int compcounter[RUMP_COMPONENT_MAX];
650 1.154 pooka
651 1.154 pooka static void
652 1.154 pooka rump_component_init_cb(struct rump_component *rc, int type)
653 1.154 pooka {
654 1.154 pooka
655 1.154 pooka KASSERT(type < RUMP_COMPONENT_MAX);
656 1.154 pooka if (rc->rc_type == type) {
657 1.154 pooka rc->rc_init();
658 1.154 pooka compcounter[type]++;
659 1.154 pooka }
660 1.154 pooka }
661 1.154 pooka
662 1.154 pooka int
663 1.154 pooka rump_component_count(enum rump_component_type type)
664 1.154 pooka {
665 1.154 pooka
666 1.154 pooka KASSERT(type <= RUMP_COMPONENT_MAX);
667 1.154 pooka return compcounter[type];
668 1.154 pooka }
669 1.154 pooka
670 1.154 pooka void
671 1.154 pooka rump_component_init(enum rump_component_type type)
672 1.154 pooka {
673 1.154 pooka
674 1.154 pooka rumpuser_dl_component_init(type, rump_component_init_cb);
675 1.154 pooka }
676 1.154 pooka
677 1.155 pooka /*
678 1.155 pooka * Initialize a module which has already been loaded and linked
679 1.155 pooka * with dlopen(). This is fundamentally the same as a builtin module.
680 1.155 pooka */
681 1.76 pooka int
682 1.155 pooka rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
683 1.76 pooka {
684 1.76 pooka
685 1.155 pooka return module_builtin_add(mip, nmodinfo, true);
686 1.106 pooka }
687 1.106 pooka
688 1.155 pooka /*
689 1.155 pooka * Finish module (flawless victory, fatality!).
690 1.155 pooka */
691 1.106 pooka int
692 1.155 pooka rump_module_fini(const struct modinfo *mi)
693 1.106 pooka {
694 1.120 pooka
695 1.155 pooka return module_builtin_remove(mi, true);
696 1.155 pooka }
697 1.155 pooka
698 1.155 pooka /*
699 1.155 pooka * Add loaded and linked module to the builtin list. It will
700 1.155 pooka * later be initialized with module_init_class().
701 1.155 pooka */
702 1.155 pooka
703 1.155 pooka static void
704 1.155 pooka add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
705 1.155 pooka {
706 1.106 pooka
707 1.155 pooka module_builtin_add(mip, nmodinfo, false);
708 1.76 pooka }
709 1.76 pooka
710 1.137 pooka int
711 1.137 pooka rump_kernelfsym_load(void *symtab, uint64_t symsize,
712 1.137 pooka char *strtab, uint64_t strsize)
713 1.137 pooka {
714 1.137 pooka static int inited = 0;
715 1.137 pooka Elf64_Ehdr ehdr;
716 1.137 pooka
717 1.137 pooka if (inited)
718 1.137 pooka return EBUSY;
719 1.137 pooka inited = 1;
720 1.137 pooka
721 1.137 pooka /*
722 1.137 pooka * Use 64bit header since it's bigger. Shouldn't make a
723 1.137 pooka * difference, since we're passing in all zeroes anyway.
724 1.137 pooka */
725 1.137 pooka memset(&ehdr, 0, sizeof(ehdr));
726 1.137 pooka ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
727 1.137 pooka
728 1.137 pooka return 0;
729 1.137 pooka }
730 1.137 pooka
731 1.199 pooka static int
732 1.199 pooka rump_proxy_syscall(int num, void *arg, register_t *retval)
733 1.95 pooka {
734 1.95 pooka struct lwp *l;
735 1.95 pooka struct sysent *callp;
736 1.124 pooka int rv;
737 1.95 pooka
738 1.95 pooka if (__predict_false(num >= SYS_NSYSENT))
739 1.95 pooka return ENOSYS;
740 1.95 pooka
741 1.95 pooka callp = rump_sysent + num;
742 1.125 pooka l = curlwp;
743 1.191 pooka rv = sy_call(callp, l, (void *)arg, retval);
744 1.191 pooka
745 1.191 pooka return rv;
746 1.191 pooka }
747 1.191 pooka
748 1.206 pooka static int
749 1.226 pooka rump_proxy_rfork(void *priv, int flags, const char *comm)
750 1.206 pooka {
751 1.206 pooka struct vmspace *newspace;
752 1.226 pooka struct proc *p;
753 1.206 pooka int error;
754 1.206 pooka
755 1.214 pooka if ((error = rump_lwproc_rfork(flags)) != 0)
756 1.206 pooka return error;
757 1.206 pooka
758 1.206 pooka /*
759 1.206 pooka * Since it's a proxy proc, adjust the vmspace.
760 1.206 pooka * Refcount will eternally be 1.
761 1.206 pooka */
762 1.226 pooka p = curproc;
763 1.229 pooka newspace = kmem_zalloc(sizeof(*newspace), KM_SLEEP);
764 1.206 pooka newspace->vm_refcnt = 1;
765 1.206 pooka newspace->vm_map.pmap = priv;
766 1.226 pooka KASSERT(p->p_vmspace == vmspace_kernel());
767 1.226 pooka p->p_vmspace = newspace;
768 1.226 pooka if (comm)
769 1.226 pooka strlcpy(p->p_comm, comm, sizeof(p->p_comm));
770 1.206 pooka
771 1.206 pooka return 0;
772 1.206 pooka }
773 1.206 pooka
774 1.232 pooka /*
775 1.232 pooka * Order all lwps in a process to exit. does *not* wait for them to drain.
776 1.232 pooka */
777 1.219 pooka static void
778 1.232 pooka rump_proxy_lwpexit(void)
779 1.219 pooka {
780 1.219 pooka struct proc *p = curproc;
781 1.219 pooka uint64_t where;
782 1.219 pooka struct lwp *l;
783 1.219 pooka
784 1.219 pooka mutex_enter(p->p_lock);
785 1.219 pooka /*
786 1.232 pooka * First pass: mark all lwps in the process with LW_RUMP_QEXIT
787 1.219 pooka * so that they know they should exit.
788 1.219 pooka */
789 1.219 pooka LIST_FOREACH(l, &p->p_lwps, l_sibling) {
790 1.219 pooka if (l == curlwp)
791 1.219 pooka continue;
792 1.232 pooka l->l_flag |= LW_RUMP_QEXIT;
793 1.219 pooka }
794 1.219 pooka mutex_exit(p->p_lock);
795 1.219 pooka
796 1.219 pooka /*
797 1.219 pooka * Next, make sure everyone on all CPUs sees our status
798 1.219 pooka * update. This keeps threads inside cv_wait() and makes
799 1.219 pooka * sure we don't access a stale cv pointer later when
800 1.219 pooka * we wake up the threads.
801 1.219 pooka */
802 1.219 pooka
803 1.219 pooka where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
804 1.219 pooka xc_wait(where);
805 1.219 pooka
806 1.219 pooka /*
807 1.219 pooka * Ok, all lwps are either:
808 1.219 pooka * 1) not in the cv code
809 1.219 pooka * 2) sleeping on l->l_private
810 1.219 pooka * 3) sleeping on p->p_waitcv
811 1.219 pooka *
812 1.232 pooka * Either way, l_private is stable until we set PS_RUMP_LWPEXIT
813 1.232 pooka * in p->p_sflag.
814 1.219 pooka */
815 1.219 pooka
816 1.219 pooka mutex_enter(p->p_lock);
817 1.219 pooka LIST_FOREACH(l, &p->p_lwps, l_sibling) {
818 1.219 pooka if (l->l_private)
819 1.219 pooka cv_broadcast(l->l_private);
820 1.219 pooka }
821 1.232 pooka p->p_sflag |= PS_RUMP_LWPEXIT;
822 1.219 pooka cv_broadcast(&p->p_waitcv);
823 1.219 pooka mutex_exit(p->p_lock);
824 1.232 pooka }
825 1.219 pooka
826 1.232 pooka /*
827 1.232 pooka * Notify process that all threads have been drained and exec is complete.
828 1.232 pooka */
829 1.232 pooka static void
830 1.232 pooka rump_proxy_execnotify(const char *comm)
831 1.232 pooka {
832 1.232 pooka struct proc *p = curproc;
833 1.232 pooka
834 1.232 pooka fd_closeexec();
835 1.232 pooka mutex_enter(p->p_lock);
836 1.232 pooka KASSERT(p->p_nlwps == 1 && p->p_sflag & PS_RUMP_LWPEXIT);
837 1.232 pooka p->p_sflag &= ~PS_RUMP_LWPEXIT;
838 1.232 pooka mutex_exit(p->p_lock);
839 1.232 pooka strlcpy(p->p_comm, comm, sizeof(p->p_comm));
840 1.219 pooka }
841 1.219 pooka
842 1.191 pooka int
843 1.130 pooka rump_boot_gethowto()
844 1.130 pooka {
845 1.130 pooka
846 1.130 pooka return boothowto;
847 1.130 pooka }
848 1.130 pooka
849 1.130 pooka void
850 1.130 pooka rump_boot_sethowto(int howto)
851 1.130 pooka {
852 1.130 pooka
853 1.130 pooka boothowto = howto;
854 1.130 pooka }
855 1.130 pooka
856 1.109 pooka int
857 1.124 pooka rump_getversion(void)
858 1.109 pooka {
859 1.109 pooka
860 1.109 pooka return __NetBSD_Version__;
861 1.109 pooka }
862 1.170 pooka
863 1.170 pooka /*
864 1.170 pooka * Note: may be called unscheduled. Not fully safe since no locking
865 1.170 pooka * of allevents (currently that's not even available).
866 1.170 pooka */
867 1.170 pooka void
868 1.170 pooka rump_printevcnts()
869 1.170 pooka {
870 1.170 pooka struct evcnt *ev;
871 1.170 pooka
872 1.170 pooka TAILQ_FOREACH(ev, &allevents, ev_list)
873 1.170 pooka rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
874 1.170 pooka ev->ev_group, ev->ev_name, ev->ev_count);
875 1.170 pooka }
876 1.184 pooka
877 1.184 pooka /*
878 1.184 pooka * If you use this interface ... well ... all bets are off.
879 1.184 pooka * The original purpose is for the p2k fs server library to be
880 1.184 pooka * able to use the same pid/lid for VOPs as the host kernel.
881 1.184 pooka */
882 1.184 pooka void
883 1.184 pooka rump_allbetsareoff_setid(pid_t pid, int lid)
884 1.184 pooka {
885 1.184 pooka struct lwp *l = curlwp;
886 1.184 pooka struct proc *p = l->l_proc;
887 1.184 pooka
888 1.184 pooka l->l_lid = lid;
889 1.184 pooka p->p_pid = pid;
890 1.184 pooka }
891