rump.c revision 1.113
1 /* $NetBSD: rump.c,v 1.113 2009/09/08 20:10:44 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 1.113 2009/09/08 20:10:44 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 "rump_private.h" 66 #include "rump_net_private.h" 67 #include "rump_vfs_private.h" 68 #include "rump_dev_private.h" 69 70 struct proc proc0; 71 struct session rump_session = { 72 .s_count = 1, 73 .s_flags = 0, 74 .s_leader = &proc0, 75 .s_login = "rumphobo", 76 .s_sid = 0, 77 }; 78 struct pgrp rump_pgrp = { 79 .pg_members = LIST_HEAD_INITIALIZER(pg_members), 80 .pg_session = &rump_session, 81 .pg_jobc = 1, 82 }; 83 struct pstats rump_stats; 84 struct plimit rump_limits; 85 struct cpu_info rump_cpu; 86 struct filedesc rump_filedesc0; 87 struct proclist allproc; 88 char machine[] = "rump"; 89 static kauth_cred_t rump_susercred; 90 91 /* pretend the master rump proc is init */ 92 struct proc *initproc = &proc0; 93 94 struct rumpuser_mtx *rump_giantlock; 95 96 sigset_t sigcantmask; 97 98 #ifdef RUMP_WITHOUT_THREADS 99 int rump_threads = 0; 100 #else 101 int rump_threads = 1; 102 #endif 103 104 static void 105 rump_aiodone_worker(struct work *wk, void *dummy) 106 { 107 struct buf *bp = (struct buf *)wk; 108 109 KASSERT(&bp->b_work == wk); 110 bp->b_iodone(bp); 111 } 112 113 static int rump_inited; 114 static struct emul emul_rump; 115 116 void rump__unavailable(void); 117 void rump__unavailable() {} 118 __weak_alias(rump_net_init,rump__unavailable); 119 __weak_alias(rump_vfs_init,rump__unavailable); 120 __weak_alias(rump_dev_init,rump__unavailable); 121 122 __weak_alias(biodone,rump__unavailable); 123 124 void rump__unavailable_vfs_panic(void); 125 void rump__unavailable_vfs_panic() {panic("vfs component not available");} 126 __weak_alias(vn_open,rump__unavailable_vfs_panic); 127 __weak_alias(vn_rdwr,rump__unavailable_vfs_panic); 128 __weak_alias(vn_stat,rump__unavailable_vfs_panic); 129 __weak_alias(vn_close,rump__unavailable_vfs_panic); 130 __weak_alias(namei,rump__unavailable_vfs_panic); 131 132 static void 133 pvfsinit_nop(struct proc *p) 134 { 135 136 return; 137 } 138 139 static void 140 pvfsrele_nop(struct proc *p) 141 { 142 143 return; 144 } 145 146 rump_proc_vfs_init_fn rump_proc_vfs_init = pvfsinit_nop; 147 rump_proc_vfs_release_fn rump_proc_vfs_release = pvfsrele_nop; 148 149 /* 150 * Stir up the stack a bit. These are exported functions to help 151 * convince the compiler that we don't want these routines completely 152 * optimized out or inlined. Is there an easier way to do this? 153 */ 154 void nullfn(uint32_t *); 155 void nullfn(uint32_t *arg){} 156 void messthestack(void); 157 void 158 messthestack(void) 159 { 160 uint32_t mess[64]; 161 uint64_t d1, d2; 162 int i, error; 163 164 for (i = 0; i < 64; i++) { 165 rumpuser_gettime(&d1, &d2, &error); 166 mess[i] = d2; 167 } 168 nullfn(mess); 169 } 170 171 int 172 rump__init(int rump_version) 173 { 174 char buf[256]; 175 struct proc *p; 176 struct lwp *l; 177 int error; 178 179 /* XXX */ 180 if (rump_inited) 181 return 0; 182 rump_inited = 1; 183 184 /* 185 * Seed arc4random() with a "reasonable" amount of randomness. 186 * Yes, this is a quick kludge which depends on the arc4random 187 * implementation. 188 */ 189 messthestack(); 190 arc4random(); 191 192 if (rump_version != RUMP_VERSION) { 193 printf("rump version mismatch, %d vs. %d\n", 194 rump_version, RUMP_VERSION); 195 return EPROGMISMATCH; 196 } 197 198 if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) { 199 rump_threads = *buf != '0'; 200 } 201 rumpuser_thrinit(_kernel_lock, _kernel_unlock, rump_threads); 202 203 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE); 204 rumpuser_mutex_recursive_init(&rump_giantlock); 205 ksyms_init(); 206 rumpvm_init(); 207 evcnt_init(); 208 209 once_init(); 210 211 rump_sleepers_init(); 212 213 pool_subsystem_init(); 214 kmem_init(); 215 216 kprintf_init(); 217 loginit(); 218 219 kauth_init(); 220 rump_susercred = rump_cred_create(0, 0, 0, NULL); 221 222 l = &lwp0; 223 p = &proc0; 224 p->p_stats = &rump_stats; 225 p->p_limit = &rump_limits; 226 p->p_pgrp = &rump_pgrp; 227 p->p_pid = 0; 228 p->p_fd = &rump_filedesc0; 229 p->p_vmspace = &rump_vmspace; 230 p->p_emul = &emul_rump; 231 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 232 l->l_cred = rump_cred_suserget(); 233 l->l_proc = p; 234 l->l_lid = 1; 235 LIST_INIT(&allproc); 236 LIST_INSERT_HEAD(&allproc, &proc0, p_list); 237 proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 238 239 rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 240 rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY; 241 rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY; 242 243 callout_startup(); 244 callout_init_cpu(&rump_cpu); 245 246 kqueue_init(); 247 iostat_init(); 248 uid_init(); 249 percpu_init(); 250 fd_sys_init(); 251 module_init(); 252 sysctl_init(); 253 softint_init(&rump_cpu); 254 devsw_init(); 255 secmodel_start(); 256 257 /* these do nothing if not present */ 258 rump_vfs_init(); 259 rump_net_init(); 260 rump_dev_init(); 261 cold = 0; 262 263 /* aieeeedondest */ 264 if (rump_threads) { 265 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 266 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 267 panic("aiodoned"); 268 } 269 270 rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error); 271 hostnamelen = strlen(hostname); 272 273 sigemptyset(&sigcantmask); 274 275 lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0); 276 277 #ifdef RUMP_USE_REAL_ALLOCATORS 278 if (rump_threads) 279 vmem_rehash_start(); 280 #endif 281 282 return 0; 283 } 284 285 struct uio * 286 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw) 287 { 288 struct uio *uio; 289 enum uio_rw uiorw; 290 291 switch (rw) { 292 case RUMPUIO_READ: 293 uiorw = UIO_READ; 294 break; 295 case RUMPUIO_WRITE: 296 uiorw = UIO_WRITE; 297 break; 298 default: 299 panic("%s: invalid rw %d", __func__, rw); 300 } 301 302 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP); 303 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP); 304 305 uio->uio_iov->iov_base = buf; 306 uio->uio_iov->iov_len = bufsize; 307 308 uio->uio_iovcnt = 1; 309 uio->uio_offset = offset; 310 uio->uio_resid = bufsize; 311 uio->uio_rw = uiorw; 312 uio->uio_vmspace = UIO_VMSPACE_SYS; 313 314 return uio; 315 } 316 317 size_t 318 rump_uio_getresid(struct uio *uio) 319 { 320 321 return uio->uio_resid; 322 } 323 324 off_t 325 rump_uio_getoff(struct uio *uio) 326 { 327 328 return uio->uio_offset; 329 } 330 331 size_t 332 rump_uio_free(struct uio *uio) 333 { 334 size_t resid; 335 336 resid = uio->uio_resid; 337 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov)); 338 kmem_free(uio, sizeof(*uio)); 339 340 return resid; 341 } 342 343 /* public interface */ 344 static pid_t nextpid = 1; 345 struct lwp * 346 rump_newproc_switch() 347 { 348 struct lwp *oldlwp = curlwp; 349 pid_t mypid; 350 351 mypid = atomic_inc_uint_nv(&nextpid); 352 if (__predict_false(mypid == 0)) 353 mypid = atomic_inc_uint_nv(&nextpid); 354 355 rumpuser_set_curlwp(NULL); 356 rump_setup_curlwp(mypid, 0, 1); 357 358 return oldlwp; 359 } 360 361 /* rump private */ 362 struct lwp * 363 rump_setup_curlwp(pid_t pid, lwpid_t lid, int set) 364 { 365 struct lwp *l; 366 struct proc *p; 367 368 l = kmem_zalloc(sizeof(struct lwp), KM_SLEEP); 369 if (pid != 0) { 370 p = kmem_zalloc(sizeof(struct proc), KM_SLEEP); 371 rump_proc_vfs_init(p); 372 p->p_stats = &rump_stats; 373 p->p_limit = &rump_limits; 374 p->p_pid = pid; 375 p->p_vmspace = &rump_vmspace; 376 p->p_fd = fd_init(NULL); 377 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 378 } else { 379 p = &proc0; 380 } 381 382 l->l_cred = rump_cred_suserget(); 383 l->l_proc = p; 384 l->l_lid = lid; 385 l->l_fd = p->p_fd; 386 l->l_mutex = RUMP_LMUTEX_MAGIC; 387 l->l_cpu = &rump_cpu; 388 389 if (set) 390 rumpuser_set_curlwp(l); 391 392 return l; 393 } 394 395 /* rump private. NEEDS WORK! */ 396 void 397 rump_set_vmspace(struct vmspace *vm) 398 { 399 struct proc *p = curproc; 400 401 p->p_vmspace = vm; 402 } 403 404 void 405 rump_clear_curlwp(void) 406 { 407 struct lwp *l; 408 struct proc *p; 409 410 l = rumpuser_get_curlwp(); 411 p = l->l_proc; 412 if (p->p_pid != 0) { 413 mutex_obj_free(p->p_lock); 414 fd_free(); 415 rump_proc_vfs_release(p); 416 rump_cred_put(l->l_cred); 417 kmem_free(p, sizeof(*p)); 418 } 419 kmem_free(l, sizeof(*l)); 420 rumpuser_set_curlwp(NULL); 421 } 422 423 struct lwp * 424 rump_get_curlwp(void) 425 { 426 struct lwp *l; 427 428 l = rumpuser_get_curlwp(); 429 if (l == NULL) 430 l = &lwp0; 431 432 return l; 433 } 434 435 void 436 rump_set_curlwp(struct lwp *l) 437 { 438 439 /* clear current */ 440 rumpuser_set_curlwp(NULL); 441 /* set new */ 442 rumpuser_set_curlwp(l); 443 } 444 445 kauth_cred_t 446 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups) 447 { 448 kauth_cred_t cred; 449 int rv; 450 451 cred = kauth_cred_alloc(); 452 kauth_cred_setuid(cred, uid); 453 kauth_cred_seteuid(cred, uid); 454 kauth_cred_setsvuid(cred, uid); 455 kauth_cred_setgid(cred, gid); 456 kauth_cred_setgid(cred, gid); 457 kauth_cred_setegid(cred, gid); 458 kauth_cred_setsvgid(cred, gid); 459 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE); 460 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */ 461 assert(rv == 0); 462 463 return cred; 464 } 465 466 void 467 rump_cred_put(kauth_cred_t cred) 468 { 469 470 kauth_cred_free(cred); 471 } 472 473 kauth_cred_t 474 rump_cred_suserget(void) 475 { 476 477 kauth_cred_hold(rump_susercred); 478 return rump_susercred; 479 } 480 481 /* 482 * Return the next system lwpid 483 */ 484 lwpid_t 485 rump_nextlid(void) 486 { 487 lwpid_t retid; 488 489 mutex_enter(proc0.p_lock); 490 /* 491 * Take next one, don't return 0 492 * XXX: most likely we'll have collisions in case this 493 * wraps around. 494 */ 495 if (++proc0.p_nlwpid == 0) 496 ++proc0.p_nlwpid; 497 retid = proc0.p_nlwpid; 498 mutex_exit(proc0.p_lock); 499 500 return retid; 501 } 502 503 int 504 rump_module_init(struct modinfo *mi, prop_dictionary_t props) 505 { 506 507 if (!module_compatible(mi->mi_version, __NetBSD_Version__)) 508 return EPROGMISMATCH; 509 510 return mi->mi_modcmd(MODULE_CMD_INIT, props); 511 } 512 513 int 514 rump_module_fini(struct modinfo *mi) 515 { 516 517 return mi->mi_modcmd(MODULE_CMD_FINI, NULL); 518 } 519 520 int _syspuffs_stub(int, int *); 521 int 522 _syspuffs_stub(int fd, int *newfd) 523 { 524 525 return ENODEV; 526 } 527 __weak_alias(rump_syspuffs_glueinit,_syspuffs_stub); 528 529 static int 530 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen, 531 register_t *retval) 532 { 533 struct lwp *l; 534 struct sysent *callp; 535 536 if (__predict_false(num >= SYS_NSYSENT)) 537 return ENOSYS; 538 539 l = curlwp; 540 callp = rump_sysent + num; 541 return callp->sy_call(l, (void *)data, retval); 542 } 543 544 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local; 545 void *rump_sysproxy_arg; 546 547 /* 548 * This whole syscall-via-rpc is still taking form. For example, it 549 * may be necessary to set syscalls individually instead of lobbing 550 * them all to the same place. So don't think this interface is 551 * set in stone. 552 */ 553 int 554 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg) 555 { 556 557 if (rump_sysproxy_arg) 558 return EBUSY; 559 560 rump_sysproxy_arg = arg; 561 rump_sysproxy = proxy; 562 563 return 0; 564 } 565 566 int 567 rump_getversion() 568 { 569 570 return __NetBSD_Version__; 571 } 572
Indexes created Tue Sep 30 06:09:59 GMT 2025