rump.c revision 1.327 1 1.327 pooka /* $NetBSD: rump.c,v 1.327 2016/01/26 23:12:17 pooka 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 1.327 pooka __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.327 2016/01/26 23:12:17 pooka 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 1.234 pooka #include <sys/namei.h>
56 1.71 pooka #include <sys/once.h>
57 1.65 pooka #include <sys/percpu.h>
58 1.139 pooka #include <sys/pipe.h>
59 1.162 pooka #include <sys/pool.h>
60 1.244 pgoyette #include <sys/pserialize.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 1.160 pooka #include <sys/timetc.h>
69 1.87 pooka #include <sys/tty.h>
70 1.68 pooka #include <sys/uidinfo.h>
71 1.89 pooka #include <sys/vmem.h>
72 1.143 pooka #include <sys/xcall.h>
73 1.235 tls #include <sys/cprng.h>
74 1.317 riastrad #include <sys/rnd.h>
75 1.279 pooka #include <sys/ktrace.h>
76 1.1 pooka
77 1.327 pooka #include <rump-sys/kern.h>
78 1.327 pooka #include <rump-sys/dev.h>
79 1.327 pooka #include <rump-sys/net.h>
80 1.327 pooka #include <rump-sys/vfs.h>
81 1.327 pooka
82 1.48 pooka #include <rump/rumpuser.h>
83 1.48 pooka
84 1.117 elad #include <secmodel/suser/suser.h>
85 1.101 pooka
86 1.114 pooka #include <prop/proplib.h>
87 1.114 pooka
88 1.174 pooka #include <uvm/uvm_extern.h>
89 1.136 pooka #include <uvm/uvm_readahead.h>
90 1.136 pooka
91 1.164 pooka char machine[] = MACHINE;
92 1.322 justin char machine_arch[] = MACHINE_ARCH;
93 1.1 pooka
94 1.193 pooka struct proc *initproc;
95 1.101 pooka
96 1.144 pooka struct device rump_rootdev = {
97 1.144 pooka .dv_class = DV_VIRTUAL
98 1.144 pooka };
99 1.144 pooka
100 1.54 pooka #ifdef RUMP_WITHOUT_THREADS
101 1.54 pooka int rump_threads = 0;
102 1.54 pooka #else
103 1.54 pooka int rump_threads = 1;
104 1.54 pooka #endif
105 1.54 pooka
106 1.265 pooka static void rump_component_addlocal(void);
107 1.255 pooka static struct lwp *bootlwp;
108 1.255 pooka
109 1.156 pooka static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */
110 1.156 pooka
111 1.218 pooka bool rump_ttycomponent = false;
112 1.217 pooka
113 1.14 pooka static void
114 1.14 pooka rump_aiodone_worker(struct work *wk, void *dummy)
115 1.14 pooka {
116 1.14 pooka struct buf *bp = (struct buf *)wk;
117 1.14 pooka
118 1.14 pooka KASSERT(&bp->b_work == wk);
119 1.14 pooka bp->b_iodone(bp);
120 1.14 pooka }
121 1.14 pooka
122 1.51 pooka static int rump_inited;
123 1.27 pooka
124 1.299 pooka void (*rump_vfs_drainbufs)(int) = (void *)nullop;
125 1.254 pooka int (*rump_vfs_makeonedevnode)(dev_t, const char *,
126 1.254 pooka devmajor_t, devminor_t) = (void *)nullop;
127 1.254 pooka int (*rump_vfs_makedevnodes)(dev_t, const char *, char,
128 1.254 pooka devmajor_t, devminor_t, int) = (void *)nullop;
129 1.321 pooka int (*rump_vfs_makesymlink)(const char *, const char *) = (void *)nullop;
130 1.162 pooka
131 1.299 pooka rump_proc_vfs_init_fn rump_proc_vfs_init = (void *)nullop;
132 1.299 pooka rump_proc_vfs_release_fn rump_proc_vfs_release = (void *)nullop;
133 1.71 pooka
134 1.155 pooka static void add_linkedin_modules(const struct modinfo *const *, size_t);
135 1.155 pooka
136 1.315 pooka static pid_t rspo_wrap_getpid(void) {
137 1.315 pooka return rump_sysproxy_hyp_getpid();
138 1.315 pooka }
139 1.315 pooka static int rspo_wrap_syscall(int num, void *arg, long *retval) {
140 1.315 pooka return rump_sysproxy_hyp_syscall(num, arg, retval);
141 1.315 pooka }
142 1.315 pooka static int rspo_wrap_rfork(void *priv, int flag, const char *comm) {
143 1.315 pooka return rump_sysproxy_hyp_rfork(priv, flag, comm);
144 1.315 pooka }
145 1.315 pooka static void rspo_wrap_lwpexit(void) {
146 1.315 pooka rump_sysproxy_hyp_lwpexit();
147 1.315 pooka }
148 1.315 pooka static void rspo_wrap_execnotify(const char *comm) {
149 1.315 pooka rump_sysproxy_hyp_execnotify(comm);
150 1.315 pooka }
151 1.263 pooka static const struct rumpuser_hyperup hyp = {
152 1.263 pooka .hyp_schedule = rump_schedule,
153 1.263 pooka .hyp_unschedule = rump_unschedule,
154 1.263 pooka .hyp_backend_unschedule = rump_user_unschedule,
155 1.263 pooka .hyp_backend_schedule = rump_user_schedule,
156 1.263 pooka .hyp_lwproc_switch = rump_lwproc_switch,
157 1.263 pooka .hyp_lwproc_release = rump_lwproc_releaselwp,
158 1.263 pooka .hyp_lwproc_newlwp = rump_lwproc_newlwp,
159 1.263 pooka .hyp_lwproc_curlwp = rump_lwproc_curlwp,
160 1.315 pooka
161 1.315 pooka .hyp_getpid = rspo_wrap_getpid,
162 1.315 pooka .hyp_syscall = rspo_wrap_syscall,
163 1.315 pooka .hyp_lwproc_rfork = rspo_wrap_rfork,
164 1.315 pooka .hyp_lwpexit = rspo_wrap_lwpexit,
165 1.315 pooka .hyp_execnotify = rspo_wrap_execnotify,
166 1.315 pooka };
167 1.315 pooka struct rump_sysproxy_ops rump_sysproxy_ops = {
168 1.315 pooka .rspo_copyin = (void *)enxio,
169 1.315 pooka .rspo_copyinstr = (void *)enxio,
170 1.315 pooka .rspo_copyout = (void *)enxio,
171 1.315 pooka .rspo_copyoutstr = (void *)enxio,
172 1.315 pooka .rspo_anonmmap = (void *)enxio,
173 1.315 pooka .rspo_raise = (void *)enxio,
174 1.315 pooka .rspo_fini = (void *)enxio,
175 1.315 pooka .rspo_hyp_getpid = (void *)enxio,
176 1.315 pooka .rspo_hyp_syscall = (void *)enxio,
177 1.315 pooka .rspo_hyp_rfork = (void *)enxio,
178 1.315 pooka .rspo_hyp_lwpexit = (void *)enxio,
179 1.315 pooka .rspo_hyp_execnotify = (void *)enxio,
180 1.197 pooka };
181 1.197 pooka
182 1.61 pooka int
183 1.207 pooka rump_daemonize_begin(void)
184 1.207 pooka {
185 1.207 pooka
186 1.207 pooka if (rump_inited)
187 1.207 pooka return EALREADY;
188 1.207 pooka
189 1.207 pooka return rumpuser_daemonize_begin();
190 1.207 pooka }
191 1.207 pooka
192 1.207 pooka int
193 1.207 pooka rump_daemonize_done(int error)
194 1.207 pooka {
195 1.207 pooka
196 1.207 pooka return rumpuser_daemonize_done(error);
197 1.207 pooka }
198 1.207 pooka
199 1.314 pooka #ifdef RUMP_USE_CTOR
200 1.314 pooka
201 1.314 pooka /* sysctl bootstrap handling */
202 1.314 pooka struct sysctl_boot_chain sysctl_boot_chain \
203 1.314 pooka = LIST_HEAD_INITIALIZER(sysctl_boot_chain);
204 1.314 pooka __link_set_add_text(sysctl_funcs,voidop); /* ensure linkset is non-empty */
205 1.314 pooka
206 1.314 pooka #else /* RUMP_USE_CTOR */
207 1.314 pooka
208 1.252 pooka RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT)
209 1.252 pooka {
210 1.265 pooka __link_set_decl(rump_components, struct rump_component);
211 1.252 pooka
212 1.252 pooka /*
213 1.253 pooka * Trick compiler into generating references so that statically
214 1.253 pooka * linked rump kernels are generated with the link set symbols.
215 1.252 pooka */
216 1.253 pooka asm("" :: "r"(__start_link_set_rump_components));
217 1.253 pooka asm("" :: "r"(__stop_link_set_rump_components));
218 1.252 pooka }
219 1.314 pooka
220 1.314 pooka #endif /* RUMP_USE_CTOR */
221 1.252 pooka
222 1.207 pooka int
223 1.259 pooka rump_init(void)
224 1.1 pooka {
225 1.36 pooka char buf[256];
226 1.160 pooka struct timespec ts;
227 1.268 pooka int64_t sec;
228 1.268 pooka long nsec;
229 1.274 pooka struct lwp *l, *initlwp;
230 1.166 pooka int i, numcpu;
231 1.1 pooka
232 1.121 pooka /* not reentrant */
233 1.27 pooka if (rump_inited)
234 1.61 pooka return 0;
235 1.121 pooka else if (rump_inited == -1)
236 1.121 pooka panic("rump_init: host process restart required");
237 1.121 pooka else
238 1.121 pooka rump_inited = 1;
239 1.27 pooka
240 1.258 pooka /* initialize hypervisor */
241 1.263 pooka if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) {
242 1.258 pooka rumpuser_dprintf("rumpuser init failed\n");
243 1.258 pooka return EINVAL;
244 1.177 pooka }
245 1.177 pooka
246 1.276 njoly /* init minimal lwp/cpu context */
247 1.290 pooka rump_lwproc_init();
248 1.276 njoly l = &lwp0;
249 1.276 njoly l->l_cpu = l->l_target_cpu = rump_cpu;
250 1.290 pooka rump_lwproc_curlwp_set(l);
251 1.276 njoly
252 1.258 pooka /* retrieve env vars which affect the early stage of bootstrap */
253 1.264 pooka if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) {
254 1.258 pooka rump_threads = *buf != '0';
255 1.258 pooka }
256 1.264 pooka if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) {
257 1.149 pooka if (*buf != '0')
258 1.149 pooka boothowto = AB_VERBOSE;
259 1.149 pooka }
260 1.264 pooka
261 1.264 pooka if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0)
262 1.264 pooka panic("mandatory hypervisor configuration (NCPU) missing");
263 1.264 pooka numcpu = strtoll(buf, NULL, 10);
264 1.264 pooka if (numcpu < 1) {
265 1.264 pooka panic("rump kernels are not lightweight enough for \"%d\" CPUs",
266 1.264 pooka 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 1.157 pooka boottime.tv_sec = sec;
274 1.157 pooka boottime.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 1.243 pgoyette 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 cprng_init();
326 1.237 tls kern_cprng = cprng_strong_create("kernel", IPL_VM,
327 1.282 pooka CPRNG_INIT_ANY|CPRNG_REKEY_ANY);
328 1.308 ozaki
329 1.282 pooka rump_hyperentropy_init();
330 1.236 njoly
331 1.184 pooka procinit();
332 1.184 pooka proc0_init();
333 1.192 pooka uid_init();
334 1.192 pooka chgproccnt(0, 1);
335 1.184 pooka
336 1.178 pooka l->l_proc = &proc0;
337 1.184 pooka lwp_update_creds(l);
338 1.176 pooka
339 1.159 pooka lwpinit_specificdata();
340 1.176 pooka lwp_initspecific(&lwp0);
341 1.36 pooka
342 1.320 pooka loginit();
343 1.320 pooka
344 1.247 pooka rump_biglock_init();
345 1.247 pooka
346 1.186 pooka rump_scheduler_init(numcpu);
347 1.193 pooka /* revert temporary context and schedule a semireal context */
348 1.290 pooka rump_lwproc_curlwp_clear(l);
349 1.193 pooka initproc = &proc0; /* borrow proc0 before we get initproc started */
350 1.124 pooka rump_schedule();
351 1.255 pooka bootlwp = curlwp;
352 1.124 pooka
353 1.310 riastrad percpu_init();
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 1.262 pooka ts = boottime;
362 1.160 pooka tc_setclock(&ts);
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 1.186 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 1.309 riastrad cprng_fast_init();
393 1.309 riastrad
394 1.256 pooka /* CPUs are up. allow kernel threads to run */
395 1.294 pooka rump_thread_allow(NULL);
396 1.256 pooka
397 1.282 pooka rnd_init_softint();
398 1.282 pooka
399 1.98 pooka kqueue_init();
400 1.72 pooka iostat_init();
401 1.43 pooka fd_sys_init();
402 1.44 ad module_init();
403 1.79 pooka devsw_init();
404 1.139 pooka pipe_init();
405 1.162 pooka resource_init();
406 1.224 pooka procinit_sysctl();
407 1.292 pooka time_init();
408 1.292 pooka time_init2();
409 1.1 pooka
410 1.175 pooka /* start page baroness */
411 1.175 pooka if (rump_threads) {
412 1.175 pooka if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
413 1.175 pooka uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
414 1.175 pooka panic("pagedaemon create failed");
415 1.175 pooka } else
416 1.175 pooka uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
417 1.175 pooka
418 1.175 pooka /* process dso's */
419 1.255 pooka rumpuser_dl_bootstrap(add_linkedin_modules,
420 1.255 pooka rump_kernelfsym_load, rump_component_load);
421 1.155 pooka
422 1.265 pooka rump_component_addlocal();
423 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN);
424 1.179 pooka
425 1.249 pooka /* initialize factions, if present */
426 1.249 pooka rump_component_init(RUMP__FACTION_VFS);
427 1.249 pooka /* pnbuf_cache is used even without vfs */
428 1.249 pooka if (rump_component_count(RUMP__FACTION_VFS) == 0) {
429 1.249 pooka pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
430 1.249 pooka NULL, IPL_NONE, NULL, NULL, NULL);
431 1.249 pooka }
432 1.249 pooka rump_component_init(RUMP__FACTION_NET);
433 1.249 pooka rump_component_init(RUMP__FACTION_DEV);
434 1.249 pooka KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1
435 1.249 pooka && rump_component_count(RUMP__FACTION_NET) <= 1
436 1.249 pooka && rump_component_count(RUMP__FACTION_DEV) <= 1);
437 1.179 pooka
438 1.179 pooka rump_component_init(RUMP_COMPONENT_KERN_VFS);
439 1.179 pooka
440 1.218 pooka /*
441 1.218 pooka * if we initialized the tty component above, the tyttymtx is
442 1.218 pooka * now initialized. otherwise, we need to initialize it.
443 1.218 pooka */
444 1.218 pooka if (!rump_ttycomponent)
445 1.218 pooka mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
446 1.218 pooka
447 1.112 pooka cold = 0;
448 1.31 pooka
449 1.14 pooka /* aieeeedondest */
450 1.54 pooka if (rump_threads) {
451 1.54 pooka if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
452 1.110 pooka rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
453 1.54 pooka panic("aiodoned");
454 1.54 pooka }
455 1.14 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 1.264 pooka panic("mandatory hypervisor configuration (HOSTNAME) missing");
463 1.264 pooka }
464 1.1 pooka hostnamelen = strlen(hostname);
465 1.24 pooka
466 1.24 pooka sigemptyset(&sigcantmask);
467 1.27 pooka
468 1.89 pooka if (rump_threads)
469 1.89 pooka vmem_rehash_start();
470 1.89 pooka
471 1.193 pooka /*
472 1.274 pooka * Create init (proc 1), used to attach implicit threads in rump.
473 1.193 pooka * (note: must be done after vfsinit to get cwdi)
474 1.193 pooka */
475 1.274 pooka initlwp = rump__lwproc_alloclwp(NULL);
476 1.193 pooka mutex_enter(proc_lock);
477 1.193 pooka initproc = proc_find_raw(1);
478 1.193 pooka mutex_exit(proc_lock);
479 1.193 pooka if (initproc == NULL)
480 1.193 pooka panic("where in the world is initproc?");
481 1.318 pooka strlcpy(initproc->p_comm, "rumplocal", sizeof(initproc->p_comm));
482 1.193 pooka
483 1.304 pooka rump_component_init(RUMP_COMPONENT_POSTINIT);
484 1.213 pooka
485 1.304 pooka /* load syscalls */
486 1.304 pooka rump_component_init(RUMP_COMPONENT_SYSCALL);
487 1.249 pooka
488 1.274 pooka /* component inits done */
489 1.274 pooka bootlwp = NULL;
490 1.274 pooka
491 1.274 pooka /* open 0/1/2 for init */
492 1.275 pooka KASSERT(rump_lwproc_curlwp() == NULL);
493 1.274 pooka rump_lwproc_switch(initlwp);
494 1.274 pooka rump_consdev_init();
495 1.275 pooka rump_lwproc_switch(NULL);
496 1.274 pooka
497 1.193 pooka /* release cpu */
498 1.124 pooka rump_unschedule();
499 1.116 pooka
500 1.2 pooka return 0;
501 1.2 pooka }
502 1.259 pooka /* historic compat */
503 1.259 pooka __strong_alias(rump__init,rump_init);
504 1.2 pooka
505 1.154 pooka static int compcounter[RUMP_COMPONENT_MAX];
506 1.248 pooka static int compinited[RUMP_COMPONENT_MAX];
507 1.154 pooka
508 1.255 pooka /*
509 1.255 pooka * Yea, this is O(n^2), but we're only looking at a handful of components.
510 1.255 pooka * Components are always initialized from the thread that called rump_init().
511 1.255 pooka */
512 1.295 pooka static LIST_HEAD(, rump_component) rchead = LIST_HEAD_INITIALIZER(rchead);
513 1.255 pooka
514 1.296 pooka #ifdef RUMP_USE_CTOR
515 1.296 pooka struct modinfo_boot_chain modinfo_boot_chain \
516 1.296 pooka = LIST_HEAD_INITIALIZER(modinfo_boot_chain);
517 1.296 pooka
518 1.296 pooka static void
519 1.296 pooka rump_component_addlocal(void)
520 1.296 pooka {
521 1.296 pooka struct modinfo_chain *mc;
522 1.296 pooka
523 1.296 pooka while ((mc = LIST_FIRST(&modinfo_boot_chain)) != NULL) {
524 1.296 pooka LIST_REMOVE(mc, mc_entries);
525 1.296 pooka module_builtin_add(&mc->mc_info, 1, false);
526 1.296 pooka }
527 1.296 pooka }
528 1.296 pooka
529 1.296 pooka #else /* RUMP_USE_CTOR */
530 1.296 pooka
531 1.265 pooka static void
532 1.265 pooka rump_component_addlocal(void)
533 1.265 pooka {
534 1.265 pooka __link_set_decl(rump_components, struct rump_component);
535 1.265 pooka struct rump_component *const *rc;
536 1.265 pooka
537 1.265 pooka __link_set_foreach(rc, rump_components) {
538 1.265 pooka rump_component_load(*rc);
539 1.265 pooka }
540 1.265 pooka }
541 1.296 pooka #endif /* RUMP_USE_CTOR */
542 1.265 pooka
543 1.296 pooka void
544 1.295 pooka rump_component_load(const struct rump_component *rc_const)
545 1.154 pooka {
546 1.295 pooka struct rump_component *rc, *rc_iter;
547 1.295 pooka
548 1.326 pooka /* time for rump component loading and unloading has passed */
549 1.326 pooka if (!cold)
550 1.326 pooka return;
551 1.326 pooka
552 1.295 pooka /*
553 1.295 pooka * XXX: this is ok since the "const" was removed from the
554 1.295 pooka * definition of RUMP_COMPONENT().
555 1.295 pooka *
556 1.295 pooka * However, to preserve the hypercall interface, the const
557 1.295 pooka * remains here. This can be fixed in the next hypercall revision.
558 1.295 pooka */
559 1.295 pooka rc = __UNCONST(rc_const);
560 1.154 pooka
561 1.307 justin KASSERT(!rump_inited || curlwp == bootlwp);
562 1.248 pooka
563 1.295 pooka LIST_FOREACH(rc_iter, &rchead, rc_entries) {
564 1.295 pooka if (rc_iter == rc)
565 1.255 pooka return;
566 1.154 pooka }
567 1.255 pooka
568 1.295 pooka LIST_INSERT_HEAD(&rchead, rc, rc_entries);
569 1.255 pooka KASSERT(rc->rc_type < RUMP_COMPONENT_MAX);
570 1.255 pooka compcounter[rc->rc_type]++;
571 1.154 pooka }
572 1.154 pooka
573 1.326 pooka void
574 1.326 pooka rump_component_unload(struct rump_component *rc)
575 1.326 pooka {
576 1.326 pooka
577 1.326 pooka /*
578 1.326 pooka * Checking for cold is enough because rump_init() both
579 1.326 pooka * flips it and handles component loading.
580 1.326 pooka */
581 1.326 pooka if (!cold)
582 1.326 pooka return;
583 1.326 pooka
584 1.326 pooka LIST_REMOVE(rc, rc_entries);
585 1.326 pooka }
586 1.326 pooka
587 1.154 pooka int
588 1.154 pooka rump_component_count(enum rump_component_type type)
589 1.154 pooka {
590 1.154 pooka
591 1.255 pooka KASSERT(curlwp == bootlwp);
592 1.255 pooka KASSERT(type < RUMP_COMPONENT_MAX);
593 1.154 pooka return compcounter[type];
594 1.154 pooka }
595 1.154 pooka
596 1.154 pooka void
597 1.154 pooka rump_component_init(enum rump_component_type type)
598 1.154 pooka {
599 1.296 pooka const struct rump_component *rc, *rc_safe;
600 1.154 pooka
601 1.255 pooka KASSERT(curlwp == bootlwp);
602 1.248 pooka KASSERT(!compinited[type]);
603 1.296 pooka LIST_FOREACH_SAFE(rc, &rchead, rc_entries, rc_safe) {
604 1.296 pooka if (rc->rc_type == type) {
605 1.255 pooka rc->rc_init();
606 1.296 pooka LIST_REMOVE(rc, rc_entries);
607 1.296 pooka }
608 1.255 pooka }
609 1.248 pooka compinited[type] = 1;
610 1.154 pooka }
611 1.154 pooka
612 1.155 pooka /*
613 1.155 pooka * Initialize a module which has already been loaded and linked
614 1.155 pooka * with dlopen(). This is fundamentally the same as a builtin module.
615 1.296 pooka *
616 1.296 pooka * XXX: this interface does not really work in the RUMP_USE_CTOR case,
617 1.296 pooka * but I'm not sure it's anything to cry about. In feeling blue,
618 1.296 pooka * things could somehow be handled via modinfo_boot_chain.
619 1.155 pooka */
620 1.76 pooka int
621 1.155 pooka rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
622 1.76 pooka {
623 1.76 pooka
624 1.155 pooka return module_builtin_add(mip, nmodinfo, true);
625 1.106 pooka }
626 1.106 pooka
627 1.155 pooka /*
628 1.155 pooka * Finish module (flawless victory, fatality!).
629 1.155 pooka */
630 1.106 pooka int
631 1.155 pooka rump_module_fini(const struct modinfo *mi)
632 1.106 pooka {
633 1.120 pooka
634 1.155 pooka return module_builtin_remove(mi, true);
635 1.155 pooka }
636 1.155 pooka
637 1.155 pooka /*
638 1.155 pooka * Add loaded and linked module to the builtin list. It will
639 1.155 pooka * later be initialized with module_init_class().
640 1.155 pooka */
641 1.155 pooka
642 1.155 pooka static void
643 1.155 pooka add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
644 1.155 pooka {
645 1.106 pooka
646 1.155 pooka module_builtin_add(mip, nmodinfo, false);
647 1.76 pooka }
648 1.76 pooka
649 1.137 pooka int
650 1.137 pooka rump_kernelfsym_load(void *symtab, uint64_t symsize,
651 1.137 pooka char *strtab, uint64_t strsize)
652 1.137 pooka {
653 1.137 pooka static int inited = 0;
654 1.137 pooka Elf64_Ehdr ehdr;
655 1.137 pooka
656 1.137 pooka if (inited)
657 1.137 pooka return EBUSY;
658 1.137 pooka inited = 1;
659 1.137 pooka
660 1.137 pooka /*
661 1.137 pooka * Use 64bit header since it's bigger. Shouldn't make a
662 1.137 pooka * difference, since we're passing in all zeroes anyway.
663 1.137 pooka */
664 1.137 pooka memset(&ehdr, 0, sizeof(ehdr));
665 1.137 pooka ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
666 1.137 pooka
667 1.137 pooka return 0;
668 1.137 pooka }
669 1.137 pooka
670 1.191 pooka int
671 1.130 pooka rump_boot_gethowto()
672 1.130 pooka {
673 1.130 pooka
674 1.130 pooka return boothowto;
675 1.130 pooka }
676 1.130 pooka
677 1.130 pooka void
678 1.130 pooka rump_boot_sethowto(int howto)
679 1.130 pooka {
680 1.130 pooka
681 1.130 pooka boothowto = howto;
682 1.130 pooka }
683 1.130 pooka
684 1.109 pooka int
685 1.124 pooka rump_getversion(void)
686 1.109 pooka {
687 1.109 pooka
688 1.109 pooka return __NetBSD_Version__;
689 1.109 pooka }
690 1.289 pooka /* compat */
691 1.289 pooka __strong_alias(rump_pub_getversion,rump_getversion);
692 1.170 pooka
693 1.170 pooka /*
694 1.170 pooka * Note: may be called unscheduled. Not fully safe since no locking
695 1.170 pooka * of allevents (currently that's not even available).
696 1.170 pooka */
697 1.170 pooka void
698 1.170 pooka rump_printevcnts()
699 1.170 pooka {
700 1.170 pooka struct evcnt *ev;
701 1.170 pooka
702 1.170 pooka TAILQ_FOREACH(ev, &allevents, ev_list)
703 1.170 pooka rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
704 1.170 pooka ev->ev_group, ev->ev_name, ev->ev_count);
705 1.170 pooka }
706 1.184 pooka
707 1.184 pooka /*
708 1.184 pooka * If you use this interface ... well ... all bets are off.
709 1.184 pooka * The original purpose is for the p2k fs server library to be
710 1.184 pooka * able to use the same pid/lid for VOPs as the host kernel.
711 1.184 pooka */
712 1.184 pooka void
713 1.184 pooka rump_allbetsareoff_setid(pid_t pid, int lid)
714 1.184 pooka {
715 1.184 pooka struct lwp *l = curlwp;
716 1.184 pooka struct proc *p = l->l_proc;
717 1.184 pooka
718 1.184 pooka l->l_lid = lid;
719 1.184 pooka p->p_pid = pid;
720 1.184 pooka }
721 1.251 martin
722 1.251 martin #include <sys/pserialize.h>
723 1.251 martin
724 1.251 martin static void
725 1.251 martin ipiemu(void *a1, void *a2)
726 1.251 martin {
727 1.251 martin
728 1.251 martin xc__highpri_intr(NULL);
729 1.251 martin pserialize_switchpoint();
730 1.251 martin }
731 1.251 martin
732 1.251 martin void
733 1.251 martin rump_xc_highpri(struct cpu_info *ci)
734 1.251 martin {
735 1.251 martin
736 1.251 martin if (ci)
737 1.251 martin xc_unicast(0, ipiemu, NULL, NULL, ci);
738 1.251 martin else
739 1.251 martin xc_broadcast(0, ipiemu, NULL, NULL);
740 1.251 martin }
741 1.272 pooka
742 1.272 pooka int
743 1.272 pooka rump_syscall(int num, void *data, size_t dlen, register_t *retval)
744 1.272 pooka {
745 1.272 pooka struct proc *p;
746 1.272 pooka struct emul *e;
747 1.272 pooka struct sysent *callp;
748 1.281 pooka const int *etrans = NULL;
749 1.272 pooka int rv;
750 1.272 pooka
751 1.272 pooka rump_schedule();
752 1.272 pooka p = curproc;
753 1.272 pooka e = p->p_emul;
754 1.272 pooka #ifndef __HAVE_MINIMAL_EMUL
755 1.272 pooka KASSERT(num > 0 && num < e->e_nsysent);
756 1.272 pooka #endif
757 1.272 pooka callp = e->e_sysent + num;
758 1.272 pooka
759 1.277 pooka rv = sy_invoke(callp, curlwp, data, retval, num);
760 1.281 pooka
761 1.281 pooka /*
762 1.281 pooka * I hope that (!__HAVE_MINIMAL_EMUL || __HAVE_SYSCALL_INTERN) is
763 1.281 pooka * an invariant ...
764 1.281 pooka */
765 1.281 pooka #if !defined(__HAVE_MINIMAL_EMUL)
766 1.281 pooka etrans = e->e_errno;
767 1.281 pooka #elif defined(__HAVE_SYSCALL_INTERN)
768 1.281 pooka etrans = p->p_emuldata;
769 1.281 pooka #endif
770 1.281 pooka
771 1.281 pooka if (etrans) {
772 1.281 pooka rv = etrans[rv];
773 1.281 pooka /*
774 1.281 pooka * XXX: small hack since Linux etrans vectors on some
775 1.281 pooka * archs contain negative errnos, but rump_syscalls
776 1.281 pooka * uses the -1 + errno ABI. Note that these
777 1.281 pooka * negative values are always the result of translation,
778 1.281 pooka * otherwise the above translation method would not
779 1.281 pooka * work very well.
780 1.281 pooka */
781 1.281 pooka if (rv < 0)
782 1.281 pooka rv = -rv;
783 1.281 pooka }
784 1.272 pooka rump_unschedule();
785 1.272 pooka
786 1.272 pooka return rv;
787 1.272 pooka }
788 1.292 pooka
789 1.304 pooka void
790 1.304 pooka rump_syscall_boot_establish(const struct rump_onesyscall *calls, size_t ncall)
791 1.304 pooka {
792 1.304 pooka struct sysent *callp;
793 1.304 pooka size_t i;
794 1.304 pooka
795 1.304 pooka for (i = 0; i < ncall; i++) {
796 1.304 pooka callp = rump_sysent + calls[i].ros_num;
797 1.304 pooka KASSERT(bootlwp != NULL
798 1.304 pooka && callp->sy_call == (sy_call_t *)enosys);
799 1.304 pooka callp->sy_call = calls[i].ros_handler;
800 1.304 pooka }
801 1.304 pooka }
802 1.304 pooka
803 1.306 pooka struct rump_boot_etfs *ebstart;
804 1.306 pooka void
805 1.306 pooka rump_boot_etfs_register(struct rump_boot_etfs *eb)
806 1.306 pooka {
807 1.306 pooka
808 1.306 pooka /*
809 1.306 pooka * Could use atomics, but, since caller would need to synchronize
810 1.306 pooka * against calling rump_init() anyway, easier to just specify the
811 1.306 pooka * interface as "caller serializes". This solve-by-specification
812 1.306 pooka * approach avoids the grey area of using atomics before rump_init()
813 1.306 pooka * runs.
814 1.306 pooka */
815 1.306 pooka eb->_eb_next = ebstart;
816 1.306 pooka eb->eb_status = -1;
817 1.306 pooka ebstart = eb;
818 1.306 pooka }
819