rump.c revision 1.118
1 /* $NetBSD: rump.c,v 1.118 2009/10/05 09:09:29 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 1.118 2009/10/05 09:09:29 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/suser/suser.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 l->l_cpu = &rump_cpu; 239 LIST_INIT(&allproc); 240 LIST_INSERT_HEAD(&allproc, &proc0, p_list); 241 proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 242 243 rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 244 rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY; 245 rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY; 246 247 callout_startup(); 248 callout_init_cpu(&rump_cpu); 249 250 sysctl_init(); 251 kqueue_init(); 252 iostat_init(); 253 uid_init(); 254 percpu_init(); 255 fd_sys_init(); 256 module_init(); 257 softint_init(&rump_cpu); 258 devsw_init(); 259 secmodel_suser_start(); 260 261 /* these do nothing if not present */ 262 rump_vfs_init(); 263 rump_net_init(); 264 rump_dev_init(); 265 cold = 0; 266 267 /* aieeeedondest */ 268 if (rump_threads) { 269 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 270 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 271 panic("aiodoned"); 272 } 273 274 sysctl_finalize(); 275 276 rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error); 277 hostnamelen = strlen(hostname); 278 279 sigemptyset(&sigcantmask); 280 281 lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0); 282 283 #ifdef RUMP_USE_REAL_ALLOCATORS 284 if (rump_threads) 285 vmem_rehash_start(); 286 #endif 287 288 rumpuser_dl_module_bootstrap(); 289 290 return 0; 291 } 292 293 struct uio * 294 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw) 295 { 296 struct uio *uio; 297 enum uio_rw uiorw; 298 299 switch (rw) { 300 case RUMPUIO_READ: 301 uiorw = UIO_READ; 302 break; 303 case RUMPUIO_WRITE: 304 uiorw = UIO_WRITE; 305 break; 306 default: 307 panic("%s: invalid rw %d", __func__, rw); 308 } 309 310 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP); 311 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP); 312 313 uio->uio_iov->iov_base = buf; 314 uio->uio_iov->iov_len = bufsize; 315 316 uio->uio_iovcnt = 1; 317 uio->uio_offset = offset; 318 uio->uio_resid = bufsize; 319 uio->uio_rw = uiorw; 320 uio->uio_vmspace = UIO_VMSPACE_SYS; 321 322 return uio; 323 } 324 325 size_t 326 rump_uio_getresid(struct uio *uio) 327 { 328 329 return uio->uio_resid; 330 } 331 332 off_t 333 rump_uio_getoff(struct uio *uio) 334 { 335 336 return uio->uio_offset; 337 } 338 339 size_t 340 rump_uio_free(struct uio *uio) 341 { 342 size_t resid; 343 344 resid = uio->uio_resid; 345 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov)); 346 kmem_free(uio, sizeof(*uio)); 347 348 return resid; 349 } 350 351 /* public interface */ 352 static pid_t nextpid = 1; 353 struct lwp * 354 rump_newproc_switch() 355 { 356 struct lwp *oldlwp = curlwp; 357 pid_t mypid; 358 359 mypid = atomic_inc_uint_nv(&nextpid); 360 if (__predict_false(mypid == 0)) 361 mypid = atomic_inc_uint_nv(&nextpid); 362 363 rumpuser_set_curlwp(NULL); 364 rump_setup_curlwp(mypid, 0, 1); 365 366 return oldlwp; 367 } 368 369 /* rump private */ 370 struct lwp * 371 rump_setup_curlwp(pid_t pid, lwpid_t lid, int set) 372 { 373 struct lwp *l; 374 struct proc *p; 375 376 l = kmem_zalloc(sizeof(struct lwp), KM_SLEEP); 377 if (pid != 0) { 378 p = kmem_zalloc(sizeof(struct proc), KM_SLEEP); 379 rump_proc_vfs_init(p); 380 p->p_stats = &rump_stats; 381 p->p_limit = &rump_limits; 382 p->p_pid = pid; 383 p->p_vmspace = &rump_vmspace; 384 p->p_fd = fd_init(NULL); 385 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 386 } else { 387 p = &proc0; 388 } 389 390 l->l_cred = rump_cred_suserget(); 391 l->l_proc = p; 392 l->l_lid = lid; 393 l->l_fd = p->p_fd; 394 l->l_mutex = RUMP_LMUTEX_MAGIC; 395 l->l_cpu = &rump_cpu; 396 397 if (set) 398 rumpuser_set_curlwp(l); 399 400 return l; 401 } 402 403 /* rump private. NEEDS WORK! */ 404 void 405 rump_set_vmspace(struct vmspace *vm) 406 { 407 struct proc *p = curproc; 408 409 p->p_vmspace = vm; 410 } 411 412 void 413 rump_clear_curlwp(void) 414 { 415 struct lwp *l; 416 struct proc *p; 417 418 l = rumpuser_get_curlwp(); 419 p = l->l_proc; 420 if (p->p_pid != 0) { 421 mutex_obj_free(p->p_lock); 422 fd_free(); 423 rump_proc_vfs_release(p); 424 rump_cred_put(l->l_cred); 425 kmem_free(p, sizeof(*p)); 426 } 427 kmem_free(l, sizeof(*l)); 428 rumpuser_set_curlwp(NULL); 429 } 430 431 struct lwp * 432 rump_get_curlwp(void) 433 { 434 struct lwp *l; 435 436 l = rumpuser_get_curlwp(); 437 if (l == NULL) 438 l = &lwp0; 439 440 return l; 441 } 442 443 void 444 rump_set_curlwp(struct lwp *l) 445 { 446 447 /* clear current */ 448 rumpuser_set_curlwp(NULL); 449 /* set new */ 450 rumpuser_set_curlwp(l); 451 } 452 453 kauth_cred_t 454 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups) 455 { 456 kauth_cred_t cred; 457 int rv; 458 459 cred = kauth_cred_alloc(); 460 kauth_cred_setuid(cred, uid); 461 kauth_cred_seteuid(cred, uid); 462 kauth_cred_setsvuid(cred, uid); 463 kauth_cred_setgid(cred, gid); 464 kauth_cred_setgid(cred, gid); 465 kauth_cred_setegid(cred, gid); 466 kauth_cred_setsvgid(cred, gid); 467 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE); 468 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */ 469 assert(rv == 0); 470 471 return cred; 472 } 473 474 void 475 rump_cred_put(kauth_cred_t cred) 476 { 477 478 kauth_cred_free(cred); 479 } 480 481 kauth_cred_t 482 rump_cred_suserget(void) 483 { 484 485 kauth_cred_hold(rump_susercred); 486 return rump_susercred; 487 } 488 489 /* 490 * Return the next system lwpid 491 */ 492 lwpid_t 493 rump_nextlid(void) 494 { 495 lwpid_t retid; 496 497 mutex_enter(proc0.p_lock); 498 /* 499 * Take next one, don't return 0 500 * XXX: most likely we'll have collisions in case this 501 * wraps around. 502 */ 503 if (++proc0.p_nlwpid == 0) 504 ++proc0.p_nlwpid; 505 retid = proc0.p_nlwpid; 506 mutex_exit(proc0.p_lock); 507 508 return retid; 509 } 510 511 int 512 rump_module_init(struct modinfo *mi, prop_dictionary_t props) 513 { 514 515 if (!module_compatible(mi->mi_version, __NetBSD_Version__)) 516 return EPROGMISMATCH; 517 518 return mi->mi_modcmd(MODULE_CMD_INIT, props); 519 } 520 521 int 522 rump_module_fini(struct modinfo *mi) 523 { 524 525 return mi->mi_modcmd(MODULE_CMD_FINI, NULL); 526 } 527 528 int _syspuffs_stub(int, int *); 529 int 530 _syspuffs_stub(int fd, int *newfd) 531 { 532 533 return ENODEV; 534 } 535 __weak_alias(rump_syspuffs_glueinit,_syspuffs_stub); 536 537 static int 538 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen, 539 register_t *retval) 540 { 541 struct lwp *l; 542 struct sysent *callp; 543 544 if (__predict_false(num >= SYS_NSYSENT)) 545 return ENOSYS; 546 547 l = curlwp; 548 callp = rump_sysent + num; 549 return callp->sy_call(l, (void *)data, retval); 550 } 551 552 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local; 553 void *rump_sysproxy_arg; 554 555 /* 556 * This whole syscall-via-rpc is still taking form. For example, it 557 * may be necessary to set syscalls individually instead of lobbing 558 * them all to the same place. So don't think this interface is 559 * set in stone. 560 */ 561 int 562 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg) 563 { 564 565 if (rump_sysproxy_arg) 566 return EBUSY; 567 568 rump_sysproxy_arg = arg; 569 rump_sysproxy = proxy; 570 571 return 0; 572 } 573 574 int 575 rump_getversion() 576 { 577 578 return __NetBSD_Version__; 579 } 580
Indexes created Tue Sep 30 07:10:03 GMT 2025