rump.c revision 1.116
1 /* $NetBSD: rump.c,v 1.116 2009/09/24 21:30:42 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 2007 Antti Kantee. All Rights Reserved. 5 * 6 * Development of this software was supported by Google Summer of Code. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS 18 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 19 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 20 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 23 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.116 2009/09/24 21:30:42 pooka Exp $"); 32 33 #include <sys/param.h> 34 #include <sys/atomic.h> 35 #include <sys/buf.h> 36 #include <sys/callout.h> 37 #include <sys/conf.h> 38 #include <sys/cpu.h> 39 #include <sys/evcnt.h> 40 #include <sys/event.h> 41 #include <sys/filedesc.h> 42 #include <sys/iostat.h> 43 #include <sys/kauth.h> 44 #include <sys/kernel.h> 45 #include <sys/kmem.h> 46 #include <sys/kprintf.h> 47 #include <sys/ksyms.h> 48 #include <sys/msgbuf.h> 49 #include <sys/module.h> 50 #include <sys/once.h> 51 #include <sys/percpu.h> 52 #include <sys/queue.h> 53 #include <sys/resourcevar.h> 54 #include <sys/select.h> 55 #include <sys/sysctl.h> 56 #include <sys/syscall.h> 57 #include <sys/tty.h> 58 #include <sys/uidinfo.h> 59 #include <sys/vmem.h> 60 61 #include <rump/rumpuser.h> 62 63 #include <secmodel/secmodel.h> 64 65 #include <prop/proplib.h> 66 67 #include "rump_private.h" 68 #include "rump_net_private.h" 69 #include "rump_vfs_private.h" 70 #include "rump_dev_private.h" 71 72 struct proc proc0; 73 struct session rump_session = { 74 .s_count = 1, 75 .s_flags = 0, 76 .s_leader = &proc0, 77 .s_login = "rumphobo", 78 .s_sid = 0, 79 }; 80 struct pgrp rump_pgrp = { 81 .pg_members = LIST_HEAD_INITIALIZER(pg_members), 82 .pg_session = &rump_session, 83 .pg_jobc = 1, 84 }; 85 struct pstats rump_stats; 86 struct plimit rump_limits; 87 struct cpu_info rump_cpu; 88 struct filedesc rump_filedesc0; 89 struct proclist allproc; 90 char machine[] = "rump"; 91 static kauth_cred_t rump_susercred; 92 93 /* pretend the master rump proc is init */ 94 struct proc *initproc = &proc0; 95 96 struct rumpuser_mtx *rump_giantlock; 97 98 sigset_t sigcantmask; 99 100 #ifdef RUMP_WITHOUT_THREADS 101 int rump_threads = 0; 102 #else 103 int rump_threads = 1; 104 #endif 105 106 static void 107 rump_aiodone_worker(struct work *wk, void *dummy) 108 { 109 struct buf *bp = (struct buf *)wk; 110 111 KASSERT(&bp->b_work == wk); 112 bp->b_iodone(bp); 113 } 114 115 static int rump_inited; 116 static struct emul emul_rump; 117 118 void rump__unavailable(void); 119 void rump__unavailable() {} 120 __weak_alias(rump_net_init,rump__unavailable); 121 __weak_alias(rump_vfs_init,rump__unavailable); 122 __weak_alias(rump_dev_init,rump__unavailable); 123 124 __weak_alias(biodone,rump__unavailable); 125 126 void rump__unavailable_vfs_panic(void); 127 void rump__unavailable_vfs_panic() {panic("vfs component not available");} 128 __weak_alias(vn_open,rump__unavailable_vfs_panic); 129 __weak_alias(vn_rdwr,rump__unavailable_vfs_panic); 130 __weak_alias(vn_stat,rump__unavailable_vfs_panic); 131 __weak_alias(vn_close,rump__unavailable_vfs_panic); 132 __weak_alias(namei,rump__unavailable_vfs_panic); 133 134 static void 135 pvfsinit_nop(struct proc *p) 136 { 137 138 return; 139 } 140 141 static void 142 pvfsrele_nop(struct proc *p) 143 { 144 145 return; 146 } 147 148 rump_proc_vfs_init_fn rump_proc_vfs_init = pvfsinit_nop; 149 rump_proc_vfs_release_fn rump_proc_vfs_release = pvfsrele_nop; 150 151 /* 152 * Stir up the stack a bit. These are exported functions to help 153 * convince the compiler that we don't want these routines completely 154 * optimized out or inlined. Is there an easier way to do this? 155 */ 156 void nullfn(uint32_t *); 157 void nullfn(uint32_t *arg){} 158 void messthestack(void); 159 void 160 messthestack(void) 161 { 162 uint32_t mess[64]; 163 uint64_t d1, d2; 164 int i, error; 165 166 for (i = 0; i < 64; i++) { 167 rumpuser_gettime(&d1, &d2, &error); 168 mess[i] = d2; 169 } 170 nullfn(mess); 171 } 172 173 int 174 rump__init(int rump_version) 175 { 176 char buf[256]; 177 struct proc *p; 178 struct lwp *l; 179 int error; 180 181 /* XXX */ 182 if (rump_inited) 183 return 0; 184 rump_inited = 1; 185 186 /* 187 * Seed arc4random() with a "reasonable" amount of randomness. 188 * Yes, this is a quick kludge which depends on the arc4random 189 * implementation. 190 */ 191 messthestack(); 192 arc4random(); 193 194 if (rump_version != RUMP_VERSION) { 195 printf("rump version mismatch, %d vs. %d\n", 196 rump_version, RUMP_VERSION); 197 return EPROGMISMATCH; 198 } 199 200 if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) { 201 rump_threads = *buf != '0'; 202 } 203 rumpuser_thrinit(_kernel_lock, _kernel_unlock, rump_threads); 204 205 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE); 206 rumpuser_mutex_recursive_init(&rump_giantlock); 207 ksyms_init(); 208 rumpvm_init(); 209 evcnt_init(); 210 211 once_init(); 212 prop_kern_init(); 213 214 rump_sleepers_init(); 215 216 pool_subsystem_init(); 217 kmem_init(); 218 219 kprintf_init(); 220 loginit(); 221 222 kauth_init(); 223 rump_susercred = rump_cred_create(0, 0, 0, NULL); 224 225 l = &lwp0; 226 p = &proc0; 227 p->p_stats = &rump_stats; 228 p->p_limit = &rump_limits; 229 p->p_pgrp = &rump_pgrp; 230 p->p_pid = 0; 231 p->p_fd = &rump_filedesc0; 232 p->p_vmspace = &rump_vmspace; 233 p->p_emul = &emul_rump; 234 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 235 l->l_cred = rump_cred_suserget(); 236 l->l_proc = p; 237 l->l_lid = 1; 238 LIST_INIT(&allproc); 239 LIST_INSERT_HEAD(&allproc, &proc0, p_list); 240 proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 241 242 rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 243 rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY; 244 rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY; 245 246 callout_startup(); 247 callout_init_cpu(&rump_cpu); 248 249 sysctl_init(); 250 kqueue_init(); 251 iostat_init(); 252 uid_init(); 253 percpu_init(); 254 fd_sys_init(); 255 module_init(); 256 softint_init(&rump_cpu); 257 devsw_init(); 258 secmodel_start(); 259 260 /* these do nothing if not present */ 261 rump_vfs_init(); 262 rump_net_init(); 263 rump_dev_init(); 264 cold = 0; 265 266 /* aieeeedondest */ 267 if (rump_threads) { 268 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 269 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 270 panic("aiodoned"); 271 } 272 273 sysctl_finalize(); 274 275 rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error); 276 hostnamelen = strlen(hostname); 277 278 sigemptyset(&sigcantmask); 279 280 lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0); 281 282 #ifdef RUMP_USE_REAL_ALLOCATORS 283 if (rump_threads) 284 vmem_rehash_start(); 285 #endif 286 287 rumpuser_dl_module_bootstrap(); 288 289 return 0; 290 } 291 292 struct uio * 293 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw) 294 { 295 struct uio *uio; 296 enum uio_rw uiorw; 297 298 switch (rw) { 299 case RUMPUIO_READ: 300 uiorw = UIO_READ; 301 break; 302 case RUMPUIO_WRITE: 303 uiorw = UIO_WRITE; 304 break; 305 default: 306 panic("%s: invalid rw %d", __func__, rw); 307 } 308 309 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP); 310 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP); 311 312 uio->uio_iov->iov_base = buf; 313 uio->uio_iov->iov_len = bufsize; 314 315 uio->uio_iovcnt = 1; 316 uio->uio_offset = offset; 317 uio->uio_resid = bufsize; 318 uio->uio_rw = uiorw; 319 uio->uio_vmspace = UIO_VMSPACE_SYS; 320 321 return uio; 322 } 323 324 size_t 325 rump_uio_getresid(struct uio *uio) 326 { 327 328 return uio->uio_resid; 329 } 330 331 off_t 332 rump_uio_getoff(struct uio *uio) 333 { 334 335 return uio->uio_offset; 336 } 337 338 size_t 339 rump_uio_free(struct uio *uio) 340 { 341 size_t resid; 342 343 resid = uio->uio_resid; 344 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov)); 345 kmem_free(uio, sizeof(*uio)); 346 347 return resid; 348 } 349 350 /* public interface */ 351 static pid_t nextpid = 1; 352 struct lwp * 353 rump_newproc_switch() 354 { 355 struct lwp *oldlwp = curlwp; 356 pid_t mypid; 357 358 mypid = atomic_inc_uint_nv(&nextpid); 359 if (__predict_false(mypid == 0)) 360 mypid = atomic_inc_uint_nv(&nextpid); 361 362 rumpuser_set_curlwp(NULL); 363 rump_setup_curlwp(mypid, 0, 1); 364 365 return oldlwp; 366 } 367 368 /* rump private */ 369 struct lwp * 370 rump_setup_curlwp(pid_t pid, lwpid_t lid, int set) 371 { 372 struct lwp *l; 373 struct proc *p; 374 375 l = kmem_zalloc(sizeof(struct lwp), KM_SLEEP); 376 if (pid != 0) { 377 p = kmem_zalloc(sizeof(struct proc), KM_SLEEP); 378 rump_proc_vfs_init(p); 379 p->p_stats = &rump_stats; 380 p->p_limit = &rump_limits; 381 p->p_pid = pid; 382 p->p_vmspace = &rump_vmspace; 383 p->p_fd = fd_init(NULL); 384 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 385 } else { 386 p = &proc0; 387 } 388 389 l->l_cred = rump_cred_suserget(); 390 l->l_proc = p; 391 l->l_lid = lid; 392 l->l_fd = p->p_fd; 393 l->l_mutex = RUMP_LMUTEX_MAGIC; 394 l->l_cpu = &rump_cpu; 395 396 if (set) 397 rumpuser_set_curlwp(l); 398 399 return l; 400 } 401 402 /* rump private. NEEDS WORK! */ 403 void 404 rump_set_vmspace(struct vmspace *vm) 405 { 406 struct proc *p = curproc; 407 408 p->p_vmspace = vm; 409 } 410 411 void 412 rump_clear_curlwp(void) 413 { 414 struct lwp *l; 415 struct proc *p; 416 417 l = rumpuser_get_curlwp(); 418 p = l->l_proc; 419 if (p->p_pid != 0) { 420 mutex_obj_free(p->p_lock); 421 fd_free(); 422 rump_proc_vfs_release(p); 423 rump_cred_put(l->l_cred); 424 kmem_free(p, sizeof(*p)); 425 } 426 kmem_free(l, sizeof(*l)); 427 rumpuser_set_curlwp(NULL); 428 } 429 430 struct lwp * 431 rump_get_curlwp(void) 432 { 433 struct lwp *l; 434 435 l = rumpuser_get_curlwp(); 436 if (l == NULL) 437 l = &lwp0; 438 439 return l; 440 } 441 442 void 443 rump_set_curlwp(struct lwp *l) 444 { 445 446 /* clear current */ 447 rumpuser_set_curlwp(NULL); 448 /* set new */ 449 rumpuser_set_curlwp(l); 450 } 451 452 kauth_cred_t 453 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups) 454 { 455 kauth_cred_t cred; 456 int rv; 457 458 cred = kauth_cred_alloc(); 459 kauth_cred_setuid(cred, uid); 460 kauth_cred_seteuid(cred, uid); 461 kauth_cred_setsvuid(cred, uid); 462 kauth_cred_setgid(cred, gid); 463 kauth_cred_setgid(cred, gid); 464 kauth_cred_setegid(cred, gid); 465 kauth_cred_setsvgid(cred, gid); 466 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE); 467 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */ 468 assert(rv == 0); 469 470 return cred; 471 } 472 473 void 474 rump_cred_put(kauth_cred_t cred) 475 { 476 477 kauth_cred_free(cred); 478 } 479 480 kauth_cred_t 481 rump_cred_suserget(void) 482 { 483 484 kauth_cred_hold(rump_susercred); 485 return rump_susercred; 486 } 487 488 /* 489 * Return the next system lwpid 490 */ 491 lwpid_t 492 rump_nextlid(void) 493 { 494 lwpid_t retid; 495 496 mutex_enter(proc0.p_lock); 497 /* 498 * Take next one, don't return 0 499 * XXX: most likely we'll have collisions in case this 500 * wraps around. 501 */ 502 if (++proc0.p_nlwpid == 0) 503 ++proc0.p_nlwpid; 504 retid = proc0.p_nlwpid; 505 mutex_exit(proc0.p_lock); 506 507 return retid; 508 } 509 510 int 511 rump_module_init(struct modinfo *mi, prop_dictionary_t props) 512 { 513 514 if (!module_compatible(mi->mi_version, __NetBSD_Version__)) 515 return EPROGMISMATCH; 516 517 return mi->mi_modcmd(MODULE_CMD_INIT, props); 518 } 519 520 int 521 rump_module_fini(struct modinfo *mi) 522 { 523 524 return mi->mi_modcmd(MODULE_CMD_FINI, NULL); 525 } 526 527 int _syspuffs_stub(int, int *); 528 int 529 _syspuffs_stub(int fd, int *newfd) 530 { 531 532 return ENODEV; 533 } 534 __weak_alias(rump_syspuffs_glueinit,_syspuffs_stub); 535 536 static int 537 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen, 538 register_t *retval) 539 { 540 struct lwp *l; 541 struct sysent *callp; 542 543 if (__predict_false(num >= SYS_NSYSENT)) 544 return ENOSYS; 545 546 l = curlwp; 547 callp = rump_sysent + num; 548 return callp->sy_call(l, (void *)data, retval); 549 } 550 551 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local; 552 void *rump_sysproxy_arg; 553 554 /* 555 * This whole syscall-via-rpc is still taking form. For example, it 556 * may be necessary to set syscalls individually instead of lobbing 557 * them all to the same place. So don't think this interface is 558 * set in stone. 559 */ 560 int 561 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg) 562 { 563 564 if (rump_sysproxy_arg) 565 return EBUSY; 566 567 rump_sysproxy_arg = arg; 568 rump_sysproxy = proxy; 569 570 return 0; 571 } 572 573 int 574 rump_getversion() 575 { 576 577 return __NetBSD_Version__; 578 } 579
Indexes created Fri Sep 26 20:09:58 GMT 2025