rump.c revision 1.353 1 1.353 chs /* $NetBSD: rump.c,v 1.353 2021/01/17 22:32:25 chs 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 1.353 chs __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.353 2021/01/17 22:32:25 chs 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 1.245 pgoyette #include <sys/kcpuset.h>
48 1.80 pooka #include <sys/kernel.h>
49 1.14 pooka #include <sys/kmem.h>
50 1.81 pooka #include <sys/kprintf.h>
51 1.175 pooka #include <sys/kthread.h>
52 1.82 pooka #include <sys/ksyms.h>
53 1.81 pooka #include <sys/msgbuf.h>
54 1.49 simonb #include <sys/module.h>
55 1.338 pgoyette #include <sys/module_hook.h>
56 1.234 pooka #include <sys/namei.h>
57 1.71 pooka #include <sys/once.h>
58 1.65 pooka #include <sys/percpu.h>
59 1.139 pooka #include <sys/pipe.h>
60 1.162 pooka #include <sys/pool.h>
61 1.1 pooka #include <sys/queue.h>
62 1.121 pooka #include <sys/reboot.h>
63 1.1 pooka #include <sys/resourcevar.h>
64 1.27 pooka #include <sys/select.h>
65 1.87 pooka #include <sys/sysctl.h>
66 1.96 pooka #include <sys/syscall.h>
67 1.169 pooka #include <sys/syscallvar.h>
68 1.332 thorpej #include <sys/threadpool.h>
69 1.160 pooka #include <sys/timetc.h>
70 1.87 pooka #include <sys/tty.h>
71 1.68 pooka #include <sys/uidinfo.h>
72 1.89 pooka #include <sys/vmem.h>
73 1.143 pooka #include <sys/xcall.h>
74 1.235 tls #include <sys/cprng.h>
75 1.317 riastrad #include <sys/rnd.h>
76 1.279 pooka #include <sys/ktrace.h>
77 1.337 riastrad #include <sys/pserialize.h>
78 1.334 ozaki #include <sys/psref.h>
79 1.1 pooka
80 1.327 pooka #include <rump-sys/kern.h>
81 1.327 pooka #include <rump-sys/dev.h>
82 1.327 pooka #include <rump-sys/net.h>
83 1.327 pooka #include <rump-sys/vfs.h>
84 1.327 pooka
85 1.48 pooka #include <rump/rumpuser.h>
86 1.48 pooka
87 1.117 elad #include <secmodel/suser/suser.h>
88 1.101 pooka
89 1.114 pooka #include <prop/proplib.h>
90 1.114 pooka
91 1.174 pooka #include <uvm/uvm_extern.h>
92 1.136 pooka #include <uvm/uvm_readahead.h>
93 1.136 pooka
94 1.164 pooka char machine[] = MACHINE;
95 1.322 justin char machine_arch[] = MACHINE_ARCH;
96 1.1 pooka
97 1.193 pooka struct proc *initproc;
98 1.101 pooka
99 1.144 pooka struct device rump_rootdev = {
100 1.144 pooka .dv_class = DV_VIRTUAL
101 1.144 pooka };
102 1.144 pooka
103 1.54 pooka #ifdef RUMP_WITHOUT_THREADS
104 1.54 pooka int rump_threads = 0;
105 1.54 pooka #else
106 1.54 pooka int rump_threads = 1;
107 1.54 pooka #endif
108 1.54 pooka
109 1.265 pooka static void rump_component_addlocal(void);
110 1.255 pooka static struct lwp *bootlwp;
111 1.255 pooka
112 1.329 joerg /* 16k should be enough for std rump needs */
113 1.329 joerg static char rump_msgbuf[16*1024] __aligned(256);
114 1.156 pooka
115 1.218 pooka bool rump_ttycomponent = false;
116 1.217 pooka
117 1.345 joerg extern pool_cache_t pnbuf_cache;
118 1.333 christos
119 1.51 pooka static int rump_inited;
120 1.27 pooka
121 1.299 pooka void (*rump_vfs_drainbufs)(int) = (void *)nullop;
122 1.254 pooka int (*rump_vfs_makeonedevnode)(dev_t, const char *,
123 1.254 pooka devmajor_t, devminor_t) = (void *)nullop;
124 1.254 pooka int (*rump_vfs_makedevnodes)(dev_t, const char *, char,
125 1.254 pooka devmajor_t, devminor_t, int) = (void *)nullop;
126 1.321 pooka int (*rump_vfs_makesymlink)(const char *, const char *) = (void *)nullop;
127 1.162 pooka
128 1.299 pooka rump_proc_vfs_init_fn rump_proc_vfs_init = (void *)nullop;
129 1.299 pooka rump_proc_vfs_release_fn rump_proc_vfs_release = (void *)nullop;
130 1.71 pooka
131 1.155 pooka static void add_linkedin_modules(const struct modinfo *const *, size_t);
132 1.343 pgoyette static void add_static_evcnt(struct evcnt *);
133 1.155 pooka
134 1.315 pooka static pid_t rspo_wrap_getpid(void) {
135 1.315 pooka return rump_sysproxy_hyp_getpid();
136 1.315 pooka }
137 1.315 pooka static int rspo_wrap_syscall(int num, void *arg, long *retval) {
138 1.315 pooka return rump_sysproxy_hyp_syscall(num, arg, retval);
139 1.315 pooka }
140 1.315 pooka static int rspo_wrap_rfork(void *priv, int flag, const char *comm) {
141 1.315 pooka return rump_sysproxy_hyp_rfork(priv, flag, comm);
142 1.315 pooka }
143 1.315 pooka static void rspo_wrap_lwpexit(void) {
144 1.315 pooka rump_sysproxy_hyp_lwpexit();
145 1.315 pooka }
146 1.315 pooka static void rspo_wrap_execnotify(const char *comm) {
147 1.315 pooka rump_sysproxy_hyp_execnotify(comm);
148 1.315 pooka }
149 1.263 pooka static const struct rumpuser_hyperup hyp = {
150 1.263 pooka .hyp_schedule = rump_schedule,
151 1.263 pooka .hyp_unschedule = rump_unschedule,
152 1.263 pooka .hyp_backend_unschedule = rump_user_unschedule,
153 1.263 pooka .hyp_backend_schedule = rump_user_schedule,
154 1.263 pooka .hyp_lwproc_switch = rump_lwproc_switch,
155 1.263 pooka .hyp_lwproc_release = rump_lwproc_releaselwp,
156 1.263 pooka .hyp_lwproc_newlwp = rump_lwproc_newlwp,
157 1.263 pooka .hyp_lwproc_curlwp = rump_lwproc_curlwp,
158 1.315 pooka
159 1.315 pooka .hyp_getpid = rspo_wrap_getpid,
160 1.315 pooka .hyp_syscall = rspo_wrap_syscall,
161 1.315 pooka .hyp_lwproc_rfork = rspo_wrap_rfork,
162 1.315 pooka .hyp_lwpexit = rspo_wrap_lwpexit,
163 1.315 pooka .hyp_execnotify = rspo_wrap_execnotify,
164 1.315 pooka };
165 1.315 pooka struct rump_sysproxy_ops rump_sysproxy_ops = {
166 1.315 pooka .rspo_copyin = (void *)enxio,
167 1.315 pooka .rspo_copyinstr = (void *)enxio,
168 1.315 pooka .rspo_copyout = (void *)enxio,
169 1.315 pooka .rspo_copyoutstr = (void *)enxio,
170 1.315 pooka .rspo_anonmmap = (void *)enxio,
171 1.315 pooka .rspo_raise = (void *)enxio,
172 1.315 pooka .rspo_fini = (void *)enxio,
173 1.315 pooka .rspo_hyp_getpid = (void *)enxio,
174 1.315 pooka .rspo_hyp_syscall = (void *)enxio,
175 1.315 pooka .rspo_hyp_rfork = (void *)enxio,
176 1.315 pooka .rspo_hyp_lwpexit = (void *)enxio,
177 1.315 pooka .rspo_hyp_execnotify = (void *)enxio,
178 1.197 pooka };
179 1.197 pooka
180 1.61 pooka int
181 1.207 pooka rump_daemonize_begin(void)
182 1.207 pooka {
183 1.207 pooka
184 1.207 pooka if (rump_inited)
185 1.207 pooka return EALREADY;
186 1.207 pooka
187 1.207 pooka return rumpuser_daemonize_begin();
188 1.207 pooka }
189 1.207 pooka
190 1.207 pooka int
191 1.207 pooka rump_daemonize_done(int error)
192 1.207 pooka {
193 1.207 pooka
194 1.207 pooka return rumpuser_daemonize_done(error);
195 1.207 pooka }
196 1.207 pooka
197 1.314 pooka #ifdef RUMP_USE_CTOR
198 1.314 pooka
199 1.314 pooka /* sysctl bootstrap handling */
200 1.314 pooka struct sysctl_boot_chain sysctl_boot_chain \
201 1.314 pooka = LIST_HEAD_INITIALIZER(sysctl_boot_chain);
202 1.314 pooka __link_set_add_text(sysctl_funcs,voidop); /* ensure linkset is non-empty */
203 1.314 pooka
204 1.314 pooka #else /* RUMP_USE_CTOR */
205 1.314 pooka
206 1.252 pooka RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT)
207 1.252 pooka {
208 1.265 pooka __link_set_decl(rump_components, struct rump_component);
209 1.252 pooka
210 1.252 pooka /*
211 1.253 pooka * Trick compiler into generating references so that statically
212 1.253 pooka * linked rump kernels are generated with the link set symbols.
213 1.252 pooka */
214 1.253 pooka asm("" :: "r"(__start_link_set_rump_components));
215 1.253 pooka asm("" :: "r"(__stop_link_set_rump_components));
216 1.252 pooka }
217 1.314 pooka
218 1.314 pooka #endif /* RUMP_USE_CTOR */
219 1.252 pooka
220 1.207 pooka int
221 1.350 christos rump_init_callback(void (*cpuinit_callback) (void))
222 1.1 pooka {
223 1.36 pooka char buf[256];
224 1.339 thorpej struct timespec bts;
225 1.268 pooka int64_t sec;
226 1.268 pooka long nsec;
227 1.274 pooka struct lwp *l, *initlwp;
228 1.166 pooka int i, numcpu;
229 1.1 pooka
230 1.121 pooka /* not reentrant */
231 1.27 pooka if (rump_inited)
232 1.61 pooka return 0;
233 1.121 pooka else if (rump_inited == -1)
234 1.350 christos panic("%s: host process restart required", __func__);
235 1.121 pooka else
236 1.121 pooka rump_inited = 1;
237 1.27 pooka
238 1.258 pooka /* initialize hypervisor */
239 1.263 pooka if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) {
240 1.258 pooka rumpuser_dprintf("rumpuser init failed\n");
241 1.258 pooka return EINVAL;
242 1.177 pooka }
243 1.177 pooka
244 1.276 njoly /* init minimal lwp/cpu context */
245 1.290 pooka rump_lwproc_init();
246 1.276 njoly l = &lwp0;
247 1.328 pooka l->l_cpu = l->l_target_cpu = &rump_bootcpu;
248 1.290 pooka rump_lwproc_curlwp_set(l);
249 1.276 njoly
250 1.258 pooka /* retrieve env vars which affect the early stage of bootstrap */
251 1.264 pooka if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) {
252 1.258 pooka rump_threads = *buf != '0';
253 1.258 pooka }
254 1.264 pooka if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) {
255 1.149 pooka if (*buf != '0')
256 1.149 pooka boothowto = AB_VERBOSE;
257 1.149 pooka }
258 1.264 pooka
259 1.264 pooka if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0)
260 1.350 christos panic("%s: mandatory hypervisor configuration (NCPU) missing",
261 1.350 christos __func__);
262 1.264 pooka numcpu = strtoll(buf, NULL, 10);
263 1.264 pooka if (numcpu < 1) {
264 1.350 christos panic("%s: rump kernels are not lightweight enough for %d CPUs",
265 1.350 christos __func__, numcpu);
266 1.167 pooka }
267 1.258 pooka
268 1.256 pooka rump_thread_init();
269 1.204 pooka rump_cpus_bootstrap(&numcpu);
270 1.165 pooka
271 1.266 pooka rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec);
272 1.339 thorpej bts.tv_sec = sec;
273 1.339 thorpej bts.tv_nsec = nsec;
274 1.157 pooka
275 1.156 pooka initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
276 1.156 pooka aprint_verbose("%s%s", copyright, version);
277 1.148 pooka
278 1.187 pooka rump_intr_init(numcpu);
279 1.36 pooka
280 1.271 njoly rump_tsleep_init();
281 1.271 njoly
282 1.261 pooka rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN);
283 1.82 pooka ksyms_init();
284 1.174 pooka uvm_init();
285 1.100 pooka evcnt_init();
286 1.89 pooka
287 1.246 pgoyette kcpuset_sysinit();
288 1.89 pooka once_init();
289 1.181 pgoyette kernconfig_lock_init();
290 1.114 pooka prop_kern_init();
291 1.63 pooka
292 1.52 pooka kmem_init();
293 1.102 pooka
294 1.136 pooka uvm_ra_init();
295 1.185 pooka uao_init();
296 1.136 pooka
297 1.133 pooka mutex_obj_init();
298 1.325 ozaki rw_obj_init();
299 1.143 pooka callout_startup();
300 1.133 pooka
301 1.81 pooka kprintf_init();
302 1.330 ozaki percpu_init();
303 1.337 riastrad pserialize_init();
304 1.52 pooka
305 1.59 pooka kauth_init();
306 1.59 pooka
307 1.238 jym secmodel_init();
308 1.283 pooka sysctl_init();
309 1.314 pooka /*
310 1.314 pooka * The above call to sysctl_init() only initializes sysctl nodes
311 1.314 pooka * from link sets. Initialize sysctls in case we used ctors.
312 1.314 pooka */
313 1.314 pooka #ifdef RUMP_USE_CTOR
314 1.314 pooka {
315 1.314 pooka struct sysctl_setup_chain *ssc;
316 1.314 pooka
317 1.326 pooka while ((ssc = LIST_FIRST(&sysctl_boot_chain)) != NULL) {
318 1.326 pooka LIST_REMOVE(ssc, ssc_entries);
319 1.314 pooka ssc->ssc_func(NULL);
320 1.314 pooka }
321 1.314 pooka }
322 1.314 pooka #endif /* RUMP_USE_CTOR */
323 1.238 jym
324 1.240 njoly rnd_init();
325 1.282 pooka rump_hyperentropy_init();
326 1.236 njoly
327 1.184 pooka procinit();
328 1.184 pooka proc0_init();
329 1.192 pooka uid_init();
330 1.192 pooka chgproccnt(0, 1);
331 1.184 pooka
332 1.178 pooka l->l_proc = &proc0;
333 1.184 pooka lwp_update_creds(l);
334 1.176 pooka
335 1.159 pooka lwpinit_specificdata();
336 1.176 pooka lwp_initspecific(&lwp0);
337 1.36 pooka
338 1.334 ozaki /* Must be called after lwpinit_specificdata */
339 1.334 ozaki psref_init();
340 1.334 ozaki
341 1.332 thorpej threadpools_init();
342 1.332 thorpej
343 1.320 pooka loginit();
344 1.320 pooka
345 1.247 pooka rump_biglock_init();
346 1.247 pooka
347 1.186 pooka rump_scheduler_init(numcpu);
348 1.193 pooka /* revert temporary context and schedule a semireal context */
349 1.290 pooka rump_lwproc_curlwp_clear(l);
350 1.193 pooka initproc = &proc0; /* borrow proc0 before we get initproc started */
351 1.124 pooka rump_schedule();
352 1.255 pooka bootlwp = curlwp;
353 1.124 pooka
354 1.160 pooka inittimecounter();
355 1.160 pooka ntp_init();
356 1.160 pooka
357 1.280 pooka #ifdef KTRACE
358 1.278 pooka ktrinit();
359 1.280 pooka #endif
360 1.278 pooka
361 1.339 thorpej tc_setclock(&bts);
362 1.150 pooka
363 1.293 pooka extern krwlock_t exec_lock;
364 1.293 pooka rw_init(&exec_lock);
365 1.293 pooka
366 1.143 pooka /* we are mostly go. do per-cpu subsystem init */
367 1.186 pooka for (i = 0; i < numcpu; i++) {
368 1.143 pooka struct cpu_info *ci = cpu_lookup(i);
369 1.143 pooka
370 1.186 pooka /* attach non-bootstrap CPUs */
371 1.186 pooka if (i > 0) {
372 1.186 pooka rump_cpu_attach(ci);
373 1.186 pooka ncpu++;
374 1.186 pooka }
375 1.340 riastrad snprintf(ci->ci_cpuname, sizeof ci->ci_cpuname, "cpu%d", i);
376 1.186 pooka
377 1.143 pooka callout_init_cpu(ci);
378 1.143 pooka softint_init(ci);
379 1.143 pooka xc_init_cpu(ci);
380 1.143 pooka pool_cache_cpu_init(ci);
381 1.143 pooka selsysinit(ci);
382 1.150 pooka percpu_init_cpu(ci);
383 1.182 pooka
384 1.216 pooka TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
385 1.216 pooka __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
386 1.216 pooka
387 1.182 pooka aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
388 1.143 pooka }
389 1.324 pooka ncpuonline = ncpu;
390 1.43 pooka
391 1.309 riastrad /* Once all CPUs are detected, initialize the per-CPU cprng_fast. */
392 1.347 riastrad cprng_init();
393 1.309 riastrad cprng_fast_init();
394 1.309 riastrad
395 1.331 msaitoh mp_online = true;
396 1.331 msaitoh
397 1.350 christos if (cpuinit_callback)
398 1.350 christos (*cpuinit_callback)();
399 1.350 christos
400 1.256 pooka /* CPUs are up. allow kernel threads to run */
401 1.294 pooka rump_thread_allow(NULL);
402 1.256 pooka
403 1.282 pooka rnd_init_softint();
404 1.282 pooka
405 1.98 pooka kqueue_init();
406 1.72 pooka iostat_init();
407 1.43 pooka fd_sys_init();
408 1.44 ad module_init();
409 1.338 pgoyette module_hook_init();
410 1.79 pooka devsw_init();
411 1.139 pooka pipe_init();
412 1.162 pooka resource_init();
413 1.224 pooka procinit_sysctl();
414 1.292 pooka time_init();
415 1.342 ad config_init();
416 1.1 pooka
417 1.175 pooka /* start page baroness */
418 1.175 pooka if (rump_threads) {
419 1.175 pooka if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
420 1.175 pooka uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
421 1.350 christos panic("%s: pagedaemon create failed", __func__);
422 1.175 pooka } else
423 1.175 pooka uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
424 1.175 pooka
425 1.175 pooka /* process dso's */
426 1.255 pooka rumpuser_dl_bootstrap(add_linkedin_modules,
427 1.343 pgoyette rump_kernelfsym_load, rump_component_load, add_static_evcnt);
428 1.155 pooka
429 1.265 pooka rump_component_addlocal();
430 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN);
431 1.179 pooka
432 1.249 pooka /* initialize factions, if present */
433 1.249 pooka rump_component_init(RUMP__FACTION_VFS);
434 1.249 pooka /* pnbuf_cache is used even without vfs */
435 1.249 pooka if (rump_component_count(RUMP__FACTION_VFS) == 0) {
436 1.249 pooka pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
437 1.249 pooka NULL, IPL_NONE, NULL, NULL, NULL);
438 1.249 pooka }
439 1.249 pooka rump_component_init(RUMP__FACTION_NET);
440 1.249 pooka rump_component_init(RUMP__FACTION_DEV);
441 1.249 pooka KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1
442 1.249 pooka && rump_component_count(RUMP__FACTION_NET) <= 1
443 1.249 pooka && rump_component_count(RUMP__FACTION_DEV) <= 1);
444 1.179 pooka
445 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN_VFS);
446 1.179 pooka
447 1.218 pooka /*
448 1.218 pooka * if we initialized the tty component above, the tyttymtx is
449 1.218 pooka * now initialized. otherwise, we need to initialize it.
450 1.218 pooka */
451 1.218 pooka if (!rump_ttycomponent)
452 1.218 pooka mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
453 1.218 pooka
454 1.112 pooka cold = 0;
455 1.31 pooka
456 1.115 pooka sysctl_finalize();
457 1.115 pooka
458 1.155 pooka module_init_class(MODULE_CLASS_ANY);
459 1.140 pooka
460 1.264 pooka if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME,
461 1.264 pooka hostname, MAXHOSTNAMELEN) != 0) {
462 1.350 christos panic(
463 1.350 christos "%s: mandatory hypervisor configuration (HOSTNAME) missing",
464 1.350 christos __func__);
465 1.264 pooka }
466 1.1 pooka hostnamelen = strlen(hostname);
467 1.24 pooka
468 1.24 pooka sigemptyset(&sigcantmask);
469 1.27 pooka
470 1.89 pooka if (rump_threads)
471 1.89 pooka vmem_rehash_start();
472 1.89 pooka
473 1.193 pooka /*
474 1.274 pooka * Create init (proc 1), used to attach implicit threads in rump.
475 1.193 pooka * (note: must be done after vfsinit to get cwdi)
476 1.193 pooka */
477 1.274 pooka initlwp = rump__lwproc_alloclwp(NULL);
478 1.348 ad mutex_enter(&proc_lock);
479 1.193 pooka initproc = proc_find_raw(1);
480 1.348 ad mutex_exit(&proc_lock);
481 1.193 pooka if (initproc == NULL)
482 1.350 christos panic("%s: where in the world is initproc?", __func__);
483 1.318 pooka strlcpy(initproc->p_comm, "rumplocal", sizeof(initproc->p_comm));
484 1.193 pooka
485 1.304 pooka rump_component_init(RUMP_COMPONENT_POSTINIT);
486 1.213 pooka
487 1.304 pooka /* load syscalls */
488 1.304 pooka rump_component_init(RUMP_COMPONENT_SYSCALL);
489 1.249 pooka
490 1.274 pooka /* component inits done */
491 1.274 pooka bootlwp = NULL;
492 1.274 pooka
493 1.274 pooka /* open 0/1/2 for init */
494 1.275 pooka KASSERT(rump_lwproc_curlwp() == NULL);
495 1.274 pooka rump_lwproc_switch(initlwp);
496 1.274 pooka rump_consdev_init();
497 1.275 pooka rump_lwproc_switch(NULL);
498 1.274 pooka
499 1.193 pooka /* release cpu */
500 1.124 pooka rump_unschedule();
501 1.116 pooka
502 1.2 pooka return 0;
503 1.2 pooka }
504 1.350 christos
505 1.350 christos int
506 1.350 christos rump_init(void)
507 1.350 christos {
508 1.350 christos return rump_init_callback(NULL);
509 1.350 christos }
510 1.350 christos
511 1.259 pooka /* historic compat */
512 1.259 pooka __strong_alias(rump__init,rump_init);
513 1.2 pooka
514 1.154 pooka static int compcounter[RUMP_COMPONENT_MAX];
515 1.248 pooka static int compinited[RUMP_COMPONENT_MAX];
516 1.154 pooka
517 1.255 pooka /*
518 1.255 pooka * Yea, this is O(n^2), but we're only looking at a handful of components.
519 1.255 pooka * Components are always initialized from the thread that called rump_init().
520 1.255 pooka */
521 1.295 pooka static LIST_HEAD(, rump_component) rchead = LIST_HEAD_INITIALIZER(rchead);
522 1.255 pooka
523 1.296 pooka #ifdef RUMP_USE_CTOR
524 1.296 pooka struct modinfo_boot_chain modinfo_boot_chain \
525 1.296 pooka = LIST_HEAD_INITIALIZER(modinfo_boot_chain);
526 1.296 pooka
527 1.296 pooka static void
528 1.296 pooka rump_component_addlocal(void)
529 1.296 pooka {
530 1.296 pooka struct modinfo_chain *mc;
531 1.296 pooka
532 1.296 pooka while ((mc = LIST_FIRST(&modinfo_boot_chain)) != NULL) {
533 1.296 pooka LIST_REMOVE(mc, mc_entries);
534 1.296 pooka module_builtin_add(&mc->mc_info, 1, false);
535 1.296 pooka }
536 1.296 pooka }
537 1.296 pooka
538 1.296 pooka #else /* RUMP_USE_CTOR */
539 1.296 pooka
540 1.265 pooka static void
541 1.265 pooka rump_component_addlocal(void)
542 1.265 pooka {
543 1.265 pooka __link_set_decl(rump_components, struct rump_component);
544 1.265 pooka struct rump_component *const *rc;
545 1.265 pooka
546 1.265 pooka __link_set_foreach(rc, rump_components) {
547 1.265 pooka rump_component_load(*rc);
548 1.265 pooka }
549 1.265 pooka }
550 1.296 pooka #endif /* RUMP_USE_CTOR */
551 1.265 pooka
552 1.296 pooka void
553 1.295 pooka rump_component_load(const struct rump_component *rc_const)
554 1.154 pooka {
555 1.295 pooka struct rump_component *rc, *rc_iter;
556 1.295 pooka
557 1.326 pooka /* time for rump component loading and unloading has passed */
558 1.326 pooka if (!cold)
559 1.326 pooka return;
560 1.326 pooka
561 1.295 pooka /*
562 1.295 pooka * XXX: this is ok since the "const" was removed from the
563 1.295 pooka * definition of RUMP_COMPONENT().
564 1.295 pooka *
565 1.295 pooka * However, to preserve the hypercall interface, the const
566 1.295 pooka * remains here. This can be fixed in the next hypercall revision.
567 1.295 pooka */
568 1.295 pooka rc = __UNCONST(rc_const);
569 1.154 pooka
570 1.307 justin KASSERT(!rump_inited || curlwp == bootlwp);
571 1.248 pooka
572 1.295 pooka LIST_FOREACH(rc_iter, &rchead, rc_entries) {
573 1.295 pooka if (rc_iter == rc)
574 1.255 pooka return;
575 1.154 pooka }
576 1.255 pooka
577 1.295 pooka LIST_INSERT_HEAD(&rchead, rc, rc_entries);
578 1.255 pooka KASSERT(rc->rc_type < RUMP_COMPONENT_MAX);
579 1.255 pooka compcounter[rc->rc_type]++;
580 1.154 pooka }
581 1.154 pooka
582 1.326 pooka void
583 1.326 pooka rump_component_unload(struct rump_component *rc)
584 1.326 pooka {
585 1.326 pooka
586 1.326 pooka /*
587 1.326 pooka * Checking for cold is enough because rump_init() both
588 1.326 pooka * flips it and handles component loading.
589 1.326 pooka */
590 1.326 pooka if (!cold)
591 1.326 pooka return;
592 1.326 pooka
593 1.326 pooka LIST_REMOVE(rc, rc_entries);
594 1.326 pooka }
595 1.326 pooka
596 1.154 pooka int
597 1.154 pooka rump_component_count(enum rump_component_type type)
598 1.154 pooka {
599 1.154 pooka
600 1.255 pooka KASSERT(curlwp == bootlwp);
601 1.255 pooka KASSERT(type < RUMP_COMPONENT_MAX);
602 1.154 pooka return compcounter[type];
603 1.154 pooka }
604 1.154 pooka
605 1.154 pooka void
606 1.154 pooka rump_component_init(enum rump_component_type type)
607 1.154 pooka {
608 1.353 chs struct rump_component *rc, *rc_next, rc_marker;
609 1.154 pooka
610 1.255 pooka KASSERT(curlwp == bootlwp);
611 1.248 pooka KASSERT(!compinited[type]);
612 1.353 chs
613 1.353 chs rc_marker.rc_type = RUMP_COMPONENT_MAX;
614 1.353 chs rc_marker.rc_init = NULL;
615 1.353 chs for (rc = LIST_FIRST(&rchead); rc != NULL; rc = rc_next) {
616 1.296 pooka if (rc->rc_type == type) {
617 1.353 chs LIST_INSERT_AFTER(rc, &rc_marker, rc_entries);
618 1.255 pooka rc->rc_init();
619 1.296 pooka LIST_REMOVE(rc, rc_entries);
620 1.353 chs rc_next = LIST_NEXT(&rc_marker, rc_entries);
621 1.353 chs LIST_REMOVE(&rc_marker, rc_entries);
622 1.353 chs } else {
623 1.353 chs rc_next = LIST_NEXT(rc, rc_entries);
624 1.296 pooka }
625 1.255 pooka }
626 1.248 pooka compinited[type] = 1;
627 1.154 pooka }
628 1.154 pooka
629 1.155 pooka /*
630 1.155 pooka * Initialize a module which has already been loaded and linked
631 1.155 pooka * with dlopen(). This is fundamentally the same as a builtin module.
632 1.296 pooka *
633 1.296 pooka * XXX: this interface does not really work in the RUMP_USE_CTOR case,
634 1.296 pooka * but I'm not sure it's anything to cry about. In feeling blue,
635 1.296 pooka * things could somehow be handled via modinfo_boot_chain.
636 1.155 pooka */
637 1.76 pooka int
638 1.155 pooka rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
639 1.76 pooka {
640 1.76 pooka
641 1.155 pooka return module_builtin_add(mip, nmodinfo, true);
642 1.106 pooka }
643 1.106 pooka
644 1.155 pooka /*
645 1.155 pooka * Finish module (flawless victory, fatality!).
646 1.155 pooka */
647 1.106 pooka int
648 1.155 pooka rump_module_fini(const struct modinfo *mi)
649 1.106 pooka {
650 1.120 pooka
651 1.155 pooka return module_builtin_remove(mi, true);
652 1.155 pooka }
653 1.155 pooka
654 1.155 pooka /*
655 1.155 pooka * Add loaded and linked module to the builtin list. It will
656 1.155 pooka * later be initialized with module_init_class().
657 1.155 pooka */
658 1.155 pooka
659 1.155 pooka static void
660 1.155 pooka add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
661 1.155 pooka {
662 1.106 pooka
663 1.155 pooka module_builtin_add(mip, nmodinfo, false);
664 1.76 pooka }
665 1.76 pooka
666 1.343 pgoyette /*
667 1.344 pgoyette * Add an evcnt.
668 1.343 pgoyette */
669 1.343 pgoyette static void
670 1.343 pgoyette add_static_evcnt(struct evcnt *ev)
671 1.343 pgoyette {
672 1.343 pgoyette
673 1.343 pgoyette evcnt_attach_static(ev);
674 1.343 pgoyette }
675 1.343 pgoyette
676 1.137 pooka int
677 1.137 pooka rump_kernelfsym_load(void *symtab, uint64_t symsize,
678 1.137 pooka char *strtab, uint64_t strsize)
679 1.137 pooka {
680 1.137 pooka static int inited = 0;
681 1.137 pooka Elf64_Ehdr ehdr;
682 1.137 pooka
683 1.137 pooka if (inited)
684 1.137 pooka return EBUSY;
685 1.137 pooka inited = 1;
686 1.137 pooka
687 1.137 pooka /*
688 1.137 pooka * Use 64bit header since it's bigger. Shouldn't make a
689 1.137 pooka * difference, since we're passing in all zeroes anyway.
690 1.137 pooka */
691 1.137 pooka memset(&ehdr, 0, sizeof(ehdr));
692 1.137 pooka ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
693 1.137 pooka
694 1.137 pooka return 0;
695 1.137 pooka }
696 1.137 pooka
697 1.191 pooka int
698 1.130 pooka rump_boot_gethowto()
699 1.130 pooka {
700 1.130 pooka
701 1.130 pooka return boothowto;
702 1.130 pooka }
703 1.130 pooka
704 1.130 pooka void
705 1.130 pooka rump_boot_sethowto(int howto)
706 1.130 pooka {
707 1.130 pooka
708 1.130 pooka boothowto = howto;
709 1.130 pooka }
710 1.130 pooka
711 1.109 pooka int
712 1.124 pooka rump_getversion(void)
713 1.109 pooka {
714 1.109 pooka
715 1.109 pooka return __NetBSD_Version__;
716 1.109 pooka }
717 1.289 pooka /* compat */
718 1.289 pooka __strong_alias(rump_pub_getversion,rump_getversion);
719 1.170 pooka
720 1.170 pooka /*
721 1.170 pooka * Note: may be called unscheduled. Not fully safe since no locking
722 1.170 pooka * of allevents (currently that's not even available).
723 1.170 pooka */
724 1.170 pooka void
725 1.170 pooka rump_printevcnts()
726 1.170 pooka {
727 1.170 pooka struct evcnt *ev;
728 1.170 pooka
729 1.170 pooka TAILQ_FOREACH(ev, &allevents, ev_list)
730 1.170 pooka rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
731 1.170 pooka ev->ev_group, ev->ev_name, ev->ev_count);
732 1.170 pooka }
733 1.184 pooka
734 1.184 pooka /*
735 1.184 pooka * If you use this interface ... well ... all bets are off.
736 1.184 pooka * The original purpose is for the p2k fs server library to be
737 1.184 pooka * able to use the same pid/lid for VOPs as the host kernel.
738 1.184 pooka */
739 1.184 pooka void
740 1.184 pooka rump_allbetsareoff_setid(pid_t pid, int lid)
741 1.184 pooka {
742 1.184 pooka struct lwp *l = curlwp;
743 1.184 pooka struct proc *p = l->l_proc;
744 1.184 pooka
745 1.184 pooka l->l_lid = lid;
746 1.184 pooka p->p_pid = pid;
747 1.184 pooka }
748 1.251 martin
749 1.251 martin static void
750 1.251 martin ipiemu(void *a1, void *a2)
751 1.251 martin {
752 1.251 martin
753 1.251 martin xc__highpri_intr(NULL);
754 1.251 martin }
755 1.251 martin
756 1.251 martin void
757 1.251 martin rump_xc_highpri(struct cpu_info *ci)
758 1.251 martin {
759 1.251 martin
760 1.251 martin if (ci)
761 1.251 martin xc_unicast(0, ipiemu, NULL, NULL, ci);
762 1.251 martin else
763 1.251 martin xc_broadcast(0, ipiemu, NULL, NULL);
764 1.251 martin }
765 1.272 pooka
766 1.272 pooka int
767 1.272 pooka rump_syscall(int num, void *data, size_t dlen, register_t *retval)
768 1.272 pooka {
769 1.272 pooka struct proc *p;
770 1.272 pooka struct emul *e;
771 1.272 pooka struct sysent *callp;
772 1.281 pooka const int *etrans = NULL;
773 1.272 pooka int rv;
774 1.272 pooka
775 1.272 pooka rump_schedule();
776 1.272 pooka p = curproc;
777 1.272 pooka e = p->p_emul;
778 1.272 pooka #ifndef __HAVE_MINIMAL_EMUL
779 1.349 kamil num &= e->e_nsysent - 1;
780 1.349 kamil #else
781 1.349 kamil num &= SYS_NSYSENT - 1;
782 1.272 pooka #endif
783 1.272 pooka callp = e->e_sysent + num;
784 1.272 pooka
785 1.277 pooka rv = sy_invoke(callp, curlwp, data, retval, num);
786 1.281 pooka
787 1.281 pooka /*
788 1.281 pooka * I hope that (!__HAVE_MINIMAL_EMUL || __HAVE_SYSCALL_INTERN) is
789 1.281 pooka * an invariant ...
790 1.281 pooka */
791 1.281 pooka #if !defined(__HAVE_MINIMAL_EMUL)
792 1.281 pooka etrans = e->e_errno;
793 1.281 pooka #elif defined(__HAVE_SYSCALL_INTERN)
794 1.281 pooka etrans = p->p_emuldata;
795 1.281 pooka #endif
796 1.281 pooka
797 1.281 pooka if (etrans) {
798 1.281 pooka rv = etrans[rv];
799 1.281 pooka /*
800 1.281 pooka * XXX: small hack since Linux etrans vectors on some
801 1.281 pooka * archs contain negative errnos, but rump_syscalls
802 1.281 pooka * uses the -1 + errno ABI. Note that these
803 1.281 pooka * negative values are always the result of translation,
804 1.281 pooka * otherwise the above translation method would not
805 1.281 pooka * work very well.
806 1.281 pooka */
807 1.281 pooka if (rv < 0)
808 1.281 pooka rv = -rv;
809 1.281 pooka }
810 1.272 pooka rump_unschedule();
811 1.272 pooka
812 1.272 pooka return rv;
813 1.272 pooka }
814 1.292 pooka
815 1.304 pooka void
816 1.304 pooka rump_syscall_boot_establish(const struct rump_onesyscall *calls, size_t ncall)
817 1.304 pooka {
818 1.304 pooka struct sysent *callp;
819 1.304 pooka size_t i;
820 1.304 pooka
821 1.304 pooka for (i = 0; i < ncall; i++) {
822 1.304 pooka callp = rump_sysent + calls[i].ros_num;
823 1.304 pooka KASSERT(bootlwp != NULL
824 1.335 christos && callp->sy_call == (sy_call_t *)(void *)enosys);
825 1.304 pooka callp->sy_call = calls[i].ros_handler;
826 1.304 pooka }
827 1.304 pooka }
828 1.304 pooka
829 1.306 pooka struct rump_boot_etfs *ebstart;
830 1.306 pooka void
831 1.306 pooka rump_boot_etfs_register(struct rump_boot_etfs *eb)
832 1.306 pooka {
833 1.306 pooka
834 1.306 pooka /*
835 1.306 pooka * Could use atomics, but, since caller would need to synchronize
836 1.306 pooka * against calling rump_init() anyway, easier to just specify the
837 1.306 pooka * interface as "caller serializes". This solve-by-specification
838 1.306 pooka * approach avoids the grey area of using atomics before rump_init()
839 1.306 pooka * runs.
840 1.306 pooka */
841 1.306 pooka eb->_eb_next = ebstart;
842 1.306 pooka eb->eb_status = -1;
843 1.306 pooka ebstart = eb;
844 1.306 pooka }
845