rump.c revision 1.354 1 1.354 riastrad /* $NetBSD: rump.c,v 1.354 2022/03/28 12:38:59 riastradh 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 1.354 riastrad __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.354 2022/03/28 12:38:59 riastradh 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 1.354 riastrad #include <sys/device_impl.h>
42 1.100 pooka #include <sys/evcnt.h>
43 1.98 pooka #include <sys/event.h>
44 1.137 pooka #include <sys/exec_elf.h>
45 1.1 pooka #include <sys/filedesc.h>
46 1.72 pooka #include <sys/iostat.h>
47 1.1 pooka #include <sys/kauth.h>
48 1.245 pgoyette #include <sys/kcpuset.h>
49 1.80 pooka #include <sys/kernel.h>
50 1.14 pooka #include <sys/kmem.h>
51 1.81 pooka #include <sys/kprintf.h>
52 1.175 pooka #include <sys/kthread.h>
53 1.82 pooka #include <sys/ksyms.h>
54 1.81 pooka #include <sys/msgbuf.h>
55 1.49 simonb #include <sys/module.h>
56 1.338 pgoyette #include <sys/module_hook.h>
57 1.234 pooka #include <sys/namei.h>
58 1.71 pooka #include <sys/once.h>
59 1.65 pooka #include <sys/percpu.h>
60 1.139 pooka #include <sys/pipe.h>
61 1.162 pooka #include <sys/pool.h>
62 1.1 pooka #include <sys/queue.h>
63 1.121 pooka #include <sys/reboot.h>
64 1.1 pooka #include <sys/resourcevar.h>
65 1.27 pooka #include <sys/select.h>
66 1.87 pooka #include <sys/sysctl.h>
67 1.96 pooka #include <sys/syscall.h>
68 1.169 pooka #include <sys/syscallvar.h>
69 1.332 thorpej #include <sys/threadpool.h>
70 1.160 pooka #include <sys/timetc.h>
71 1.87 pooka #include <sys/tty.h>
72 1.68 pooka #include <sys/uidinfo.h>
73 1.89 pooka #include <sys/vmem.h>
74 1.143 pooka #include <sys/xcall.h>
75 1.235 tls #include <sys/cprng.h>
76 1.317 riastrad #include <sys/rnd.h>
77 1.279 pooka #include <sys/ktrace.h>
78 1.337 riastrad #include <sys/pserialize.h>
79 1.334 ozaki #include <sys/psref.h>
80 1.1 pooka
81 1.327 pooka #include <rump-sys/kern.h>
82 1.327 pooka #include <rump-sys/dev.h>
83 1.327 pooka #include <rump-sys/net.h>
84 1.327 pooka #include <rump-sys/vfs.h>
85 1.327 pooka
86 1.48 pooka #include <rump/rumpuser.h>
87 1.48 pooka
88 1.117 elad #include <secmodel/suser/suser.h>
89 1.101 pooka
90 1.114 pooka #include <prop/proplib.h>
91 1.114 pooka
92 1.174 pooka #include <uvm/uvm_extern.h>
93 1.136 pooka #include <uvm/uvm_readahead.h>
94 1.136 pooka
95 1.164 pooka char machine[] = MACHINE;
96 1.322 justin char machine_arch[] = MACHINE_ARCH;
97 1.1 pooka
98 1.193 pooka struct proc *initproc;
99 1.101 pooka
100 1.144 pooka struct device rump_rootdev = {
101 1.144 pooka .dv_class = DV_VIRTUAL
102 1.144 pooka };
103 1.144 pooka
104 1.54 pooka #ifdef RUMP_WITHOUT_THREADS
105 1.54 pooka int rump_threads = 0;
106 1.54 pooka #else
107 1.54 pooka int rump_threads = 1;
108 1.54 pooka #endif
109 1.54 pooka
110 1.265 pooka static void rump_component_addlocal(void);
111 1.255 pooka static struct lwp *bootlwp;
112 1.255 pooka
113 1.329 joerg /* 16k should be enough for std rump needs */
114 1.329 joerg static char rump_msgbuf[16*1024] __aligned(256);
115 1.156 pooka
116 1.218 pooka bool rump_ttycomponent = false;
117 1.217 pooka
118 1.345 joerg extern pool_cache_t pnbuf_cache;
119 1.333 christos
120 1.51 pooka static int rump_inited;
121 1.27 pooka
122 1.299 pooka void (*rump_vfs_drainbufs)(int) = (void *)nullop;
123 1.254 pooka int (*rump_vfs_makeonedevnode)(dev_t, const char *,
124 1.254 pooka devmajor_t, devminor_t) = (void *)nullop;
125 1.254 pooka int (*rump_vfs_makedevnodes)(dev_t, const char *, char,
126 1.254 pooka devmajor_t, devminor_t, int) = (void *)nullop;
127 1.321 pooka int (*rump_vfs_makesymlink)(const char *, const char *) = (void *)nullop;
128 1.162 pooka
129 1.299 pooka rump_proc_vfs_init_fn rump_proc_vfs_init = (void *)nullop;
130 1.299 pooka rump_proc_vfs_release_fn rump_proc_vfs_release = (void *)nullop;
131 1.71 pooka
132 1.155 pooka static void add_linkedin_modules(const struct modinfo *const *, size_t);
133 1.343 pgoyette static void add_static_evcnt(struct evcnt *);
134 1.155 pooka
135 1.315 pooka static pid_t rspo_wrap_getpid(void) {
136 1.315 pooka return rump_sysproxy_hyp_getpid();
137 1.315 pooka }
138 1.315 pooka static int rspo_wrap_syscall(int num, void *arg, long *retval) {
139 1.315 pooka return rump_sysproxy_hyp_syscall(num, arg, retval);
140 1.315 pooka }
141 1.315 pooka static int rspo_wrap_rfork(void *priv, int flag, const char *comm) {
142 1.315 pooka return rump_sysproxy_hyp_rfork(priv, flag, comm);
143 1.315 pooka }
144 1.315 pooka static void rspo_wrap_lwpexit(void) {
145 1.315 pooka rump_sysproxy_hyp_lwpexit();
146 1.315 pooka }
147 1.315 pooka static void rspo_wrap_execnotify(const char *comm) {
148 1.315 pooka rump_sysproxy_hyp_execnotify(comm);
149 1.315 pooka }
150 1.263 pooka static const struct rumpuser_hyperup hyp = {
151 1.263 pooka .hyp_schedule = rump_schedule,
152 1.263 pooka .hyp_unschedule = rump_unschedule,
153 1.263 pooka .hyp_backend_unschedule = rump_user_unschedule,
154 1.263 pooka .hyp_backend_schedule = rump_user_schedule,
155 1.263 pooka .hyp_lwproc_switch = rump_lwproc_switch,
156 1.263 pooka .hyp_lwproc_release = rump_lwproc_releaselwp,
157 1.263 pooka .hyp_lwproc_newlwp = rump_lwproc_newlwp,
158 1.263 pooka .hyp_lwproc_curlwp = rump_lwproc_curlwp,
159 1.315 pooka
160 1.315 pooka .hyp_getpid = rspo_wrap_getpid,
161 1.315 pooka .hyp_syscall = rspo_wrap_syscall,
162 1.315 pooka .hyp_lwproc_rfork = rspo_wrap_rfork,
163 1.315 pooka .hyp_lwpexit = rspo_wrap_lwpexit,
164 1.315 pooka .hyp_execnotify = rspo_wrap_execnotify,
165 1.315 pooka };
166 1.315 pooka struct rump_sysproxy_ops rump_sysproxy_ops = {
167 1.315 pooka .rspo_copyin = (void *)enxio,
168 1.315 pooka .rspo_copyinstr = (void *)enxio,
169 1.315 pooka .rspo_copyout = (void *)enxio,
170 1.315 pooka .rspo_copyoutstr = (void *)enxio,
171 1.315 pooka .rspo_anonmmap = (void *)enxio,
172 1.315 pooka .rspo_raise = (void *)enxio,
173 1.315 pooka .rspo_fini = (void *)enxio,
174 1.315 pooka .rspo_hyp_getpid = (void *)enxio,
175 1.315 pooka .rspo_hyp_syscall = (void *)enxio,
176 1.315 pooka .rspo_hyp_rfork = (void *)enxio,
177 1.315 pooka .rspo_hyp_lwpexit = (void *)enxio,
178 1.315 pooka .rspo_hyp_execnotify = (void *)enxio,
179 1.197 pooka };
180 1.197 pooka
181 1.61 pooka int
182 1.207 pooka rump_daemonize_begin(void)
183 1.207 pooka {
184 1.207 pooka
185 1.207 pooka if (rump_inited)
186 1.207 pooka return EALREADY;
187 1.207 pooka
188 1.207 pooka return rumpuser_daemonize_begin();
189 1.207 pooka }
190 1.207 pooka
191 1.207 pooka int
192 1.207 pooka rump_daemonize_done(int error)
193 1.207 pooka {
194 1.207 pooka
195 1.207 pooka return rumpuser_daemonize_done(error);
196 1.207 pooka }
197 1.207 pooka
198 1.314 pooka #ifdef RUMP_USE_CTOR
199 1.314 pooka
200 1.314 pooka /* sysctl bootstrap handling */
201 1.314 pooka struct sysctl_boot_chain sysctl_boot_chain \
202 1.314 pooka = LIST_HEAD_INITIALIZER(sysctl_boot_chain);
203 1.314 pooka __link_set_add_text(sysctl_funcs,voidop); /* ensure linkset is non-empty */
204 1.314 pooka
205 1.314 pooka #else /* RUMP_USE_CTOR */
206 1.314 pooka
207 1.252 pooka RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT)
208 1.252 pooka {
209 1.265 pooka __link_set_decl(rump_components, struct rump_component);
210 1.252 pooka
211 1.252 pooka /*
212 1.253 pooka * Trick compiler into generating references so that statically
213 1.253 pooka * linked rump kernels are generated with the link set symbols.
214 1.252 pooka */
215 1.253 pooka asm("" :: "r"(__start_link_set_rump_components));
216 1.253 pooka asm("" :: "r"(__stop_link_set_rump_components));
217 1.252 pooka }
218 1.314 pooka
219 1.314 pooka #endif /* RUMP_USE_CTOR */
220 1.252 pooka
221 1.207 pooka int
222 1.350 christos rump_init_callback(void (*cpuinit_callback) (void))
223 1.1 pooka {
224 1.36 pooka char buf[256];
225 1.339 thorpej struct timespec bts;
226 1.268 pooka int64_t sec;
227 1.268 pooka long nsec;
228 1.274 pooka struct lwp *l, *initlwp;
229 1.166 pooka int i, numcpu;
230 1.1 pooka
231 1.121 pooka /* not reentrant */
232 1.27 pooka if (rump_inited)
233 1.61 pooka return 0;
234 1.121 pooka else if (rump_inited == -1)
235 1.350 christos panic("%s: host process restart required", __func__);
236 1.121 pooka else
237 1.121 pooka rump_inited = 1;
238 1.27 pooka
239 1.258 pooka /* initialize hypervisor */
240 1.263 pooka if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) {
241 1.258 pooka rumpuser_dprintf("rumpuser init failed\n");
242 1.258 pooka return EINVAL;
243 1.177 pooka }
244 1.177 pooka
245 1.276 njoly /* init minimal lwp/cpu context */
246 1.290 pooka rump_lwproc_init();
247 1.276 njoly l = &lwp0;
248 1.328 pooka l->l_cpu = l->l_target_cpu = &rump_bootcpu;
249 1.290 pooka rump_lwproc_curlwp_set(l);
250 1.276 njoly
251 1.258 pooka /* retrieve env vars which affect the early stage of bootstrap */
252 1.264 pooka if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) {
253 1.258 pooka rump_threads = *buf != '0';
254 1.258 pooka }
255 1.264 pooka if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) {
256 1.149 pooka if (*buf != '0')
257 1.149 pooka boothowto = AB_VERBOSE;
258 1.149 pooka }
259 1.264 pooka
260 1.264 pooka if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0)
261 1.350 christos panic("%s: mandatory hypervisor configuration (NCPU) missing",
262 1.350 christos __func__);
263 1.264 pooka numcpu = strtoll(buf, NULL, 10);
264 1.264 pooka if (numcpu < 1) {
265 1.350 christos panic("%s: rump kernels are not lightweight enough for %d CPUs",
266 1.350 christos __func__, numcpu);
267 1.167 pooka }
268 1.258 pooka
269 1.256 pooka rump_thread_init();
270 1.204 pooka rump_cpus_bootstrap(&numcpu);
271 1.165 pooka
272 1.266 pooka rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec);
273 1.339 thorpej bts.tv_sec = sec;
274 1.339 thorpej bts.tv_nsec = nsec;
275 1.157 pooka
276 1.156 pooka initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
277 1.156 pooka aprint_verbose("%s%s", copyright, version);
278 1.148 pooka
279 1.187 pooka rump_intr_init(numcpu);
280 1.36 pooka
281 1.271 njoly rump_tsleep_init();
282 1.271 njoly
283 1.261 pooka rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN);
284 1.82 pooka ksyms_init();
285 1.174 pooka uvm_init();
286 1.100 pooka evcnt_init();
287 1.89 pooka
288 1.246 pgoyette kcpuset_sysinit();
289 1.89 pooka once_init();
290 1.181 pgoyette kernconfig_lock_init();
291 1.114 pooka prop_kern_init();
292 1.63 pooka
293 1.52 pooka kmem_init();
294 1.102 pooka
295 1.136 pooka uvm_ra_init();
296 1.185 pooka uao_init();
297 1.136 pooka
298 1.133 pooka mutex_obj_init();
299 1.325 ozaki rw_obj_init();
300 1.143 pooka callout_startup();
301 1.133 pooka
302 1.81 pooka kprintf_init();
303 1.330 ozaki percpu_init();
304 1.337 riastrad pserialize_init();
305 1.52 pooka
306 1.59 pooka kauth_init();
307 1.59 pooka
308 1.238 jym secmodel_init();
309 1.283 pooka sysctl_init();
310 1.314 pooka /*
311 1.314 pooka * The above call to sysctl_init() only initializes sysctl nodes
312 1.314 pooka * from link sets. Initialize sysctls in case we used ctors.
313 1.314 pooka */
314 1.314 pooka #ifdef RUMP_USE_CTOR
315 1.314 pooka {
316 1.314 pooka struct sysctl_setup_chain *ssc;
317 1.314 pooka
318 1.326 pooka while ((ssc = LIST_FIRST(&sysctl_boot_chain)) != NULL) {
319 1.326 pooka LIST_REMOVE(ssc, ssc_entries);
320 1.314 pooka ssc->ssc_func(NULL);
321 1.314 pooka }
322 1.314 pooka }
323 1.314 pooka #endif /* RUMP_USE_CTOR */
324 1.238 jym
325 1.240 njoly rnd_init();
326 1.282 pooka rump_hyperentropy_init();
327 1.236 njoly
328 1.184 pooka procinit();
329 1.184 pooka proc0_init();
330 1.192 pooka uid_init();
331 1.192 pooka chgproccnt(0, 1);
332 1.184 pooka
333 1.178 pooka l->l_proc = &proc0;
334 1.184 pooka lwp_update_creds(l);
335 1.176 pooka
336 1.159 pooka lwpinit_specificdata();
337 1.176 pooka lwp_initspecific(&lwp0);
338 1.36 pooka
339 1.334 ozaki /* Must be called after lwpinit_specificdata */
340 1.334 ozaki psref_init();
341 1.334 ozaki
342 1.332 thorpej threadpools_init();
343 1.332 thorpej
344 1.320 pooka loginit();
345 1.320 pooka
346 1.247 pooka rump_biglock_init();
347 1.247 pooka
348 1.186 pooka rump_scheduler_init(numcpu);
349 1.193 pooka /* revert temporary context and schedule a semireal context */
350 1.290 pooka rump_lwproc_curlwp_clear(l);
351 1.193 pooka initproc = &proc0; /* borrow proc0 before we get initproc started */
352 1.124 pooka rump_schedule();
353 1.255 pooka bootlwp = curlwp;
354 1.124 pooka
355 1.160 pooka inittimecounter();
356 1.160 pooka ntp_init();
357 1.160 pooka
358 1.280 pooka #ifdef KTRACE
359 1.278 pooka ktrinit();
360 1.280 pooka #endif
361 1.278 pooka
362 1.339 thorpej tc_setclock(&bts);
363 1.150 pooka
364 1.293 pooka extern krwlock_t exec_lock;
365 1.293 pooka rw_init(&exec_lock);
366 1.293 pooka
367 1.143 pooka /* we are mostly go. do per-cpu subsystem init */
368 1.186 pooka for (i = 0; i < numcpu; i++) {
369 1.143 pooka struct cpu_info *ci = cpu_lookup(i);
370 1.143 pooka
371 1.186 pooka /* attach non-bootstrap CPUs */
372 1.186 pooka if (i > 0) {
373 1.186 pooka rump_cpu_attach(ci);
374 1.186 pooka ncpu++;
375 1.186 pooka }
376 1.340 riastrad snprintf(ci->ci_cpuname, sizeof ci->ci_cpuname, "cpu%d", i);
377 1.186 pooka
378 1.143 pooka callout_init_cpu(ci);
379 1.143 pooka softint_init(ci);
380 1.143 pooka xc_init_cpu(ci);
381 1.143 pooka pool_cache_cpu_init(ci);
382 1.143 pooka selsysinit(ci);
383 1.150 pooka percpu_init_cpu(ci);
384 1.182 pooka
385 1.216 pooka TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
386 1.216 pooka __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
387 1.216 pooka
388 1.182 pooka aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
389 1.143 pooka }
390 1.324 pooka ncpuonline = ncpu;
391 1.43 pooka
392 1.309 riastrad /* Once all CPUs are detected, initialize the per-CPU cprng_fast. */
393 1.347 riastrad cprng_init();
394 1.309 riastrad cprng_fast_init();
395 1.309 riastrad
396 1.331 msaitoh mp_online = true;
397 1.331 msaitoh
398 1.350 christos if (cpuinit_callback)
399 1.350 christos (*cpuinit_callback)();
400 1.350 christos
401 1.256 pooka /* CPUs are up. allow kernel threads to run */
402 1.294 pooka rump_thread_allow(NULL);
403 1.256 pooka
404 1.282 pooka rnd_init_softint();
405 1.282 pooka
406 1.98 pooka kqueue_init();
407 1.72 pooka iostat_init();
408 1.43 pooka fd_sys_init();
409 1.44 ad module_init();
410 1.338 pgoyette module_hook_init();
411 1.79 pooka devsw_init();
412 1.139 pooka pipe_init();
413 1.162 pooka resource_init();
414 1.224 pooka procinit_sysctl();
415 1.292 pooka time_init();
416 1.342 ad config_init();
417 1.1 pooka
418 1.175 pooka /* start page baroness */
419 1.175 pooka if (rump_threads) {
420 1.175 pooka if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
421 1.175 pooka uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
422 1.350 christos panic("%s: pagedaemon create failed", __func__);
423 1.175 pooka } else
424 1.175 pooka uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
425 1.175 pooka
426 1.175 pooka /* process dso's */
427 1.255 pooka rumpuser_dl_bootstrap(add_linkedin_modules,
428 1.343 pgoyette rump_kernelfsym_load, rump_component_load, add_static_evcnt);
429 1.155 pooka
430 1.265 pooka rump_component_addlocal();
431 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN);
432 1.179 pooka
433 1.249 pooka /* initialize factions, if present */
434 1.249 pooka rump_component_init(RUMP__FACTION_VFS);
435 1.249 pooka /* pnbuf_cache is used even without vfs */
436 1.249 pooka if (rump_component_count(RUMP__FACTION_VFS) == 0) {
437 1.249 pooka pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
438 1.249 pooka NULL, IPL_NONE, NULL, NULL, NULL);
439 1.249 pooka }
440 1.249 pooka rump_component_init(RUMP__FACTION_NET);
441 1.249 pooka rump_component_init(RUMP__FACTION_DEV);
442 1.249 pooka KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1
443 1.249 pooka && rump_component_count(RUMP__FACTION_NET) <= 1
444 1.249 pooka && rump_component_count(RUMP__FACTION_DEV) <= 1);
445 1.179 pooka
446 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN_VFS);
447 1.179 pooka
448 1.218 pooka /*
449 1.218 pooka * if we initialized the tty component above, the tyttymtx is
450 1.218 pooka * now initialized. otherwise, we need to initialize it.
451 1.218 pooka */
452 1.218 pooka if (!rump_ttycomponent)
453 1.218 pooka mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
454 1.218 pooka
455 1.112 pooka cold = 0;
456 1.31 pooka
457 1.115 pooka sysctl_finalize();
458 1.115 pooka
459 1.155 pooka module_init_class(MODULE_CLASS_ANY);
460 1.140 pooka
461 1.264 pooka if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME,
462 1.264 pooka hostname, MAXHOSTNAMELEN) != 0) {
463 1.350 christos panic(
464 1.350 christos "%s: mandatory hypervisor configuration (HOSTNAME) missing",
465 1.350 christos __func__);
466 1.264 pooka }
467 1.1 pooka hostnamelen = strlen(hostname);
468 1.24 pooka
469 1.24 pooka sigemptyset(&sigcantmask);
470 1.27 pooka
471 1.89 pooka if (rump_threads)
472 1.89 pooka vmem_rehash_start();
473 1.89 pooka
474 1.193 pooka /*
475 1.274 pooka * Create init (proc 1), used to attach implicit threads in rump.
476 1.193 pooka * (note: must be done after vfsinit to get cwdi)
477 1.193 pooka */
478 1.274 pooka initlwp = rump__lwproc_alloclwp(NULL);
479 1.348 ad mutex_enter(&proc_lock);
480 1.193 pooka initproc = proc_find_raw(1);
481 1.348 ad mutex_exit(&proc_lock);
482 1.193 pooka if (initproc == NULL)
483 1.350 christos panic("%s: where in the world is initproc?", __func__);
484 1.318 pooka strlcpy(initproc->p_comm, "rumplocal", sizeof(initproc->p_comm));
485 1.193 pooka
486 1.304 pooka rump_component_init(RUMP_COMPONENT_POSTINIT);
487 1.213 pooka
488 1.304 pooka /* load syscalls */
489 1.304 pooka rump_component_init(RUMP_COMPONENT_SYSCALL);
490 1.249 pooka
491 1.274 pooka /* component inits done */
492 1.274 pooka bootlwp = NULL;
493 1.274 pooka
494 1.274 pooka /* open 0/1/2 for init */
495 1.275 pooka KASSERT(rump_lwproc_curlwp() == NULL);
496 1.274 pooka rump_lwproc_switch(initlwp);
497 1.274 pooka rump_consdev_init();
498 1.275 pooka rump_lwproc_switch(NULL);
499 1.274 pooka
500 1.193 pooka /* release cpu */
501 1.124 pooka rump_unschedule();
502 1.116 pooka
503 1.2 pooka return 0;
504 1.2 pooka }
505 1.350 christos
506 1.350 christos int
507 1.350 christos rump_init(void)
508 1.350 christos {
509 1.350 christos return rump_init_callback(NULL);
510 1.350 christos }
511 1.350 christos
512 1.259 pooka /* historic compat */
513 1.259 pooka __strong_alias(rump__init,rump_init);
514 1.2 pooka
515 1.154 pooka static int compcounter[RUMP_COMPONENT_MAX];
516 1.248 pooka static int compinited[RUMP_COMPONENT_MAX];
517 1.154 pooka
518 1.255 pooka /*
519 1.255 pooka * Yea, this is O(n^2), but we're only looking at a handful of components.
520 1.255 pooka * Components are always initialized from the thread that called rump_init().
521 1.255 pooka */
522 1.295 pooka static LIST_HEAD(, rump_component) rchead = LIST_HEAD_INITIALIZER(rchead);
523 1.255 pooka
524 1.296 pooka #ifdef RUMP_USE_CTOR
525 1.296 pooka struct modinfo_boot_chain modinfo_boot_chain \
526 1.296 pooka = LIST_HEAD_INITIALIZER(modinfo_boot_chain);
527 1.296 pooka
528 1.296 pooka static void
529 1.296 pooka rump_component_addlocal(void)
530 1.296 pooka {
531 1.296 pooka struct modinfo_chain *mc;
532 1.296 pooka
533 1.296 pooka while ((mc = LIST_FIRST(&modinfo_boot_chain)) != NULL) {
534 1.296 pooka LIST_REMOVE(mc, mc_entries);
535 1.296 pooka module_builtin_add(&mc->mc_info, 1, false);
536 1.296 pooka }
537 1.296 pooka }
538 1.296 pooka
539 1.296 pooka #else /* RUMP_USE_CTOR */
540 1.296 pooka
541 1.265 pooka static void
542 1.265 pooka rump_component_addlocal(void)
543 1.265 pooka {
544 1.265 pooka __link_set_decl(rump_components, struct rump_component);
545 1.265 pooka struct rump_component *const *rc;
546 1.265 pooka
547 1.265 pooka __link_set_foreach(rc, rump_components) {
548 1.265 pooka rump_component_load(*rc);
549 1.265 pooka }
550 1.265 pooka }
551 1.296 pooka #endif /* RUMP_USE_CTOR */
552 1.265 pooka
553 1.296 pooka void
554 1.295 pooka rump_component_load(const struct rump_component *rc_const)
555 1.154 pooka {
556 1.295 pooka struct rump_component *rc, *rc_iter;
557 1.295 pooka
558 1.326 pooka /* time for rump component loading and unloading has passed */
559 1.326 pooka if (!cold)
560 1.326 pooka return;
561 1.326 pooka
562 1.295 pooka /*
563 1.295 pooka * XXX: this is ok since the "const" was removed from the
564 1.295 pooka * definition of RUMP_COMPONENT().
565 1.295 pooka *
566 1.295 pooka * However, to preserve the hypercall interface, the const
567 1.295 pooka * remains here. This can be fixed in the next hypercall revision.
568 1.295 pooka */
569 1.295 pooka rc = __UNCONST(rc_const);
570 1.154 pooka
571 1.307 justin KASSERT(!rump_inited || curlwp == bootlwp);
572 1.248 pooka
573 1.295 pooka LIST_FOREACH(rc_iter, &rchead, rc_entries) {
574 1.295 pooka if (rc_iter == rc)
575 1.255 pooka return;
576 1.154 pooka }
577 1.255 pooka
578 1.295 pooka LIST_INSERT_HEAD(&rchead, rc, rc_entries);
579 1.255 pooka KASSERT(rc->rc_type < RUMP_COMPONENT_MAX);
580 1.255 pooka compcounter[rc->rc_type]++;
581 1.154 pooka }
582 1.154 pooka
583 1.326 pooka void
584 1.326 pooka rump_component_unload(struct rump_component *rc)
585 1.326 pooka {
586 1.326 pooka
587 1.326 pooka /*
588 1.326 pooka * Checking for cold is enough because rump_init() both
589 1.326 pooka * flips it and handles component loading.
590 1.326 pooka */
591 1.326 pooka if (!cold)
592 1.326 pooka return;
593 1.326 pooka
594 1.326 pooka LIST_REMOVE(rc, rc_entries);
595 1.326 pooka }
596 1.326 pooka
597 1.154 pooka int
598 1.154 pooka rump_component_count(enum rump_component_type type)
599 1.154 pooka {
600 1.154 pooka
601 1.255 pooka KASSERT(curlwp == bootlwp);
602 1.255 pooka KASSERT(type < RUMP_COMPONENT_MAX);
603 1.154 pooka return compcounter[type];
604 1.154 pooka }
605 1.154 pooka
606 1.154 pooka void
607 1.154 pooka rump_component_init(enum rump_component_type type)
608 1.154 pooka {
609 1.353 chs struct rump_component *rc, *rc_next, rc_marker;
610 1.154 pooka
611 1.255 pooka KASSERT(curlwp == bootlwp);
612 1.248 pooka KASSERT(!compinited[type]);
613 1.353 chs
614 1.353 chs rc_marker.rc_type = RUMP_COMPONENT_MAX;
615 1.353 chs rc_marker.rc_init = NULL;
616 1.353 chs for (rc = LIST_FIRST(&rchead); rc != NULL; rc = rc_next) {
617 1.296 pooka if (rc->rc_type == type) {
618 1.353 chs LIST_INSERT_AFTER(rc, &rc_marker, rc_entries);
619 1.255 pooka rc->rc_init();
620 1.296 pooka LIST_REMOVE(rc, rc_entries);
621 1.353 chs rc_next = LIST_NEXT(&rc_marker, rc_entries);
622 1.353 chs LIST_REMOVE(&rc_marker, rc_entries);
623 1.353 chs } else {
624 1.353 chs rc_next = LIST_NEXT(rc, rc_entries);
625 1.296 pooka }
626 1.255 pooka }
627 1.248 pooka compinited[type] = 1;
628 1.154 pooka }
629 1.154 pooka
630 1.155 pooka /*
631 1.155 pooka * Initialize a module which has already been loaded and linked
632 1.155 pooka * with dlopen(). This is fundamentally the same as a builtin module.
633 1.296 pooka *
634 1.296 pooka * XXX: this interface does not really work in the RUMP_USE_CTOR case,
635 1.296 pooka * but I'm not sure it's anything to cry about. In feeling blue,
636 1.296 pooka * things could somehow be handled via modinfo_boot_chain.
637 1.155 pooka */
638 1.76 pooka int
639 1.155 pooka rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
640 1.76 pooka {
641 1.76 pooka
642 1.155 pooka return module_builtin_add(mip, nmodinfo, true);
643 1.106 pooka }
644 1.106 pooka
645 1.155 pooka /*
646 1.155 pooka * Finish module (flawless victory, fatality!).
647 1.155 pooka */
648 1.106 pooka int
649 1.155 pooka rump_module_fini(const struct modinfo *mi)
650 1.106 pooka {
651 1.120 pooka
652 1.155 pooka return module_builtin_remove(mi, true);
653 1.155 pooka }
654 1.155 pooka
655 1.155 pooka /*
656 1.155 pooka * Add loaded and linked module to the builtin list. It will
657 1.155 pooka * later be initialized with module_init_class().
658 1.155 pooka */
659 1.155 pooka
660 1.155 pooka static void
661 1.155 pooka add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
662 1.155 pooka {
663 1.106 pooka
664 1.155 pooka module_builtin_add(mip, nmodinfo, false);
665 1.76 pooka }
666 1.76 pooka
667 1.343 pgoyette /*
668 1.344 pgoyette * Add an evcnt.
669 1.343 pgoyette */
670 1.343 pgoyette static void
671 1.343 pgoyette add_static_evcnt(struct evcnt *ev)
672 1.343 pgoyette {
673 1.343 pgoyette
674 1.343 pgoyette evcnt_attach_static(ev);
675 1.343 pgoyette }
676 1.343 pgoyette
677 1.137 pooka int
678 1.137 pooka rump_kernelfsym_load(void *symtab, uint64_t symsize,
679 1.137 pooka char *strtab, uint64_t strsize)
680 1.137 pooka {
681 1.137 pooka static int inited = 0;
682 1.137 pooka Elf64_Ehdr ehdr;
683 1.137 pooka
684 1.137 pooka if (inited)
685 1.137 pooka return EBUSY;
686 1.137 pooka inited = 1;
687 1.137 pooka
688 1.137 pooka /*
689 1.137 pooka * Use 64bit header since it's bigger. Shouldn't make a
690 1.137 pooka * difference, since we're passing in all zeroes anyway.
691 1.137 pooka */
692 1.137 pooka memset(&ehdr, 0, sizeof(ehdr));
693 1.137 pooka ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
694 1.137 pooka
695 1.137 pooka return 0;
696 1.137 pooka }
697 1.137 pooka
698 1.191 pooka int
699 1.130 pooka rump_boot_gethowto()
700 1.130 pooka {
701 1.130 pooka
702 1.130 pooka return boothowto;
703 1.130 pooka }
704 1.130 pooka
705 1.130 pooka void
706 1.130 pooka rump_boot_sethowto(int howto)
707 1.130 pooka {
708 1.130 pooka
709 1.130 pooka boothowto = howto;
710 1.130 pooka }
711 1.130 pooka
712 1.109 pooka int
713 1.124 pooka rump_getversion(void)
714 1.109 pooka {
715 1.109 pooka
716 1.109 pooka return __NetBSD_Version__;
717 1.109 pooka }
718 1.289 pooka /* compat */
719 1.289 pooka __strong_alias(rump_pub_getversion,rump_getversion);
720 1.170 pooka
721 1.170 pooka /*
722 1.170 pooka * Note: may be called unscheduled. Not fully safe since no locking
723 1.170 pooka * of allevents (currently that's not even available).
724 1.170 pooka */
725 1.170 pooka void
726 1.170 pooka rump_printevcnts()
727 1.170 pooka {
728 1.170 pooka struct evcnt *ev;
729 1.170 pooka
730 1.170 pooka TAILQ_FOREACH(ev, &allevents, ev_list)
731 1.170 pooka rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
732 1.170 pooka ev->ev_group, ev->ev_name, ev->ev_count);
733 1.170 pooka }
734 1.184 pooka
735 1.184 pooka /*
736 1.184 pooka * If you use this interface ... well ... all bets are off.
737 1.184 pooka * The original purpose is for the p2k fs server library to be
738 1.184 pooka * able to use the same pid/lid for VOPs as the host kernel.
739 1.184 pooka */
740 1.184 pooka void
741 1.184 pooka rump_allbetsareoff_setid(pid_t pid, int lid)
742 1.184 pooka {
743 1.184 pooka struct lwp *l = curlwp;
744 1.184 pooka struct proc *p = l->l_proc;
745 1.184 pooka
746 1.184 pooka l->l_lid = lid;
747 1.184 pooka p->p_pid = pid;
748 1.184 pooka }
749 1.251 martin
750 1.251 martin static void
751 1.251 martin ipiemu(void *a1, void *a2)
752 1.251 martin {
753 1.251 martin
754 1.251 martin xc__highpri_intr(NULL);
755 1.251 martin }
756 1.251 martin
757 1.251 martin void
758 1.251 martin rump_xc_highpri(struct cpu_info *ci)
759 1.251 martin {
760 1.251 martin
761 1.251 martin if (ci)
762 1.251 martin xc_unicast(0, ipiemu, NULL, NULL, ci);
763 1.251 martin else
764 1.251 martin xc_broadcast(0, ipiemu, NULL, NULL);
765 1.251 martin }
766 1.272 pooka
767 1.272 pooka int
768 1.272 pooka rump_syscall(int num, void *data, size_t dlen, register_t *retval)
769 1.272 pooka {
770 1.272 pooka struct proc *p;
771 1.272 pooka struct emul *e;
772 1.272 pooka struct sysent *callp;
773 1.281 pooka const int *etrans = NULL;
774 1.272 pooka int rv;
775 1.272 pooka
776 1.272 pooka rump_schedule();
777 1.272 pooka p = curproc;
778 1.272 pooka e = p->p_emul;
779 1.272 pooka #ifndef __HAVE_MINIMAL_EMUL
780 1.349 kamil num &= e->e_nsysent - 1;
781 1.349 kamil #else
782 1.349 kamil num &= SYS_NSYSENT - 1;
783 1.272 pooka #endif
784 1.272 pooka callp = e->e_sysent + num;
785 1.272 pooka
786 1.277 pooka rv = sy_invoke(callp, curlwp, data, retval, num);
787 1.281 pooka
788 1.281 pooka /*
789 1.281 pooka * I hope that (!__HAVE_MINIMAL_EMUL || __HAVE_SYSCALL_INTERN) is
790 1.281 pooka * an invariant ...
791 1.281 pooka */
792 1.281 pooka #if !defined(__HAVE_MINIMAL_EMUL)
793 1.281 pooka etrans = e->e_errno;
794 1.281 pooka #elif defined(__HAVE_SYSCALL_INTERN)
795 1.281 pooka etrans = p->p_emuldata;
796 1.281 pooka #endif
797 1.281 pooka
798 1.281 pooka if (etrans) {
799 1.281 pooka rv = etrans[rv];
800 1.281 pooka /*
801 1.281 pooka * XXX: small hack since Linux etrans vectors on some
802 1.281 pooka * archs contain negative errnos, but rump_syscalls
803 1.281 pooka * uses the -1 + errno ABI. Note that these
804 1.281 pooka * negative values are always the result of translation,
805 1.281 pooka * otherwise the above translation method would not
806 1.281 pooka * work very well.
807 1.281 pooka */
808 1.281 pooka if (rv < 0)
809 1.281 pooka rv = -rv;
810 1.281 pooka }
811 1.272 pooka rump_unschedule();
812 1.272 pooka
813 1.272 pooka return rv;
814 1.272 pooka }
815 1.292 pooka
816 1.304 pooka void
817 1.304 pooka rump_syscall_boot_establish(const struct rump_onesyscall *calls, size_t ncall)
818 1.304 pooka {
819 1.304 pooka struct sysent *callp;
820 1.304 pooka size_t i;
821 1.304 pooka
822 1.304 pooka for (i = 0; i < ncall; i++) {
823 1.304 pooka callp = rump_sysent + calls[i].ros_num;
824 1.304 pooka KASSERT(bootlwp != NULL
825 1.335 christos && callp->sy_call == (sy_call_t *)(void *)enosys);
826 1.304 pooka callp->sy_call = calls[i].ros_handler;
827 1.304 pooka }
828 1.304 pooka }
829 1.304 pooka
830 1.306 pooka struct rump_boot_etfs *ebstart;
831 1.306 pooka void
832 1.306 pooka rump_boot_etfs_register(struct rump_boot_etfs *eb)
833 1.306 pooka {
834 1.306 pooka
835 1.306 pooka /*
836 1.306 pooka * Could use atomics, but, since caller would need to synchronize
837 1.306 pooka * against calling rump_init() anyway, easier to just specify the
838 1.306 pooka * interface as "caller serializes". This solve-by-specification
839 1.306 pooka * approach avoids the grey area of using atomics before rump_init()
840 1.306 pooka * runs.
841 1.306 pooka */
842 1.306 pooka eb->_eb_next = ebstart;
843 1.306 pooka eb->eb_status = -1;
844 1.306 pooka ebstart = eb;
845 1.306 pooka }
846