rump.c revision 1.157
1 /* $NetBSD: rump.c,v 1.157 2010/03/31 12:16:15 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 1.157 2010/03/31 12:16:15 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/device.h> 40 #include <sys/evcnt.h> 41 #include <sys/event.h> 42 #include <sys/exec_elf.h> 43 #include <sys/filedesc.h> 44 #include <sys/iostat.h> 45 #include <sys/kauth.h> 46 #include <sys/kernel.h> 47 #include <sys/kmem.h> 48 #include <sys/kprintf.h> 49 #include <sys/ksyms.h> 50 #include <sys/msgbuf.h> 51 #include <sys/module.h> 52 #include <sys/once.h> 53 #include <sys/percpu.h> 54 #include <sys/pipe.h> 55 #include <sys/queue.h> 56 #include <sys/reboot.h> 57 #include <sys/resourcevar.h> 58 #include <sys/select.h> 59 #include <sys/sysctl.h> 60 #include <sys/syscall.h> 61 #include <sys/tty.h> 62 #include <sys/uidinfo.h> 63 #include <sys/vmem.h> 64 #include <sys/xcall.h> 65 66 #include <rump/rumpuser.h> 67 68 #include <secmodel/suser/suser.h> 69 70 #include <prop/proplib.h> 71 72 #include <uvm/uvm_readahead.h> 73 74 #include "rump_private.h" 75 #include "rump_net_private.h" 76 #include "rump_vfs_private.h" 77 #include "rump_dev_private.h" 78 79 struct proc proc0; 80 struct session rump_session = { 81 .s_count = 1, 82 .s_flags = 0, 83 .s_leader = &proc0, 84 .s_login = "rumphobo", 85 .s_sid = 0, 86 }; 87 struct pgrp rump_pgrp = { 88 .pg_members = LIST_HEAD_INITIALIZER(pg_members), 89 .pg_session = &rump_session, 90 .pg_jobc = 1, 91 }; 92 struct pstats rump_stats; 93 struct plimit rump_limits; 94 struct filedesc rump_filedesc0; 95 struct proclist allproc; 96 char machine[] = "rump"; 97 static kauth_cred_t rump_susercred; 98 99 /* pretend the master rump proc is init */ 100 struct proc *initproc = &proc0; 101 102 struct rumpuser_mtx *rump_giantlock; 103 104 sigset_t sigcantmask; 105 106 struct device rump_rootdev = { 107 .dv_class = DV_VIRTUAL 108 }; 109 110 #ifdef RUMP_WITHOUT_THREADS 111 int rump_threads = 0; 112 #else 113 int rump_threads = 1; 114 #endif 115 116 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */ 117 118 static void 119 rump_aiodone_worker(struct work *wk, void *dummy) 120 { 121 struct buf *bp = (struct buf *)wk; 122 123 KASSERT(&bp->b_work == wk); 124 bp->b_iodone(bp); 125 } 126 127 static int rump_inited; 128 static struct emul emul_rump = { 129 .e_vm_default_addr = uvm_default_mapaddr, 130 }; 131 132 int rump__unavailable(void); 133 int rump__unavailable() {return EOPNOTSUPP;} 134 __weak_alias(rump_net_init,rump__unavailable); 135 __weak_alias(rump_vfs_init,rump__unavailable); 136 __weak_alias(rump_dev_init,rump__unavailable); 137 138 __weak_alias(rump_vfs_fini,rump__unavailable); 139 140 __weak_alias(biodone,rump__unavailable); 141 __weak_alias(sopoll,rump__unavailable); 142 143 void rump__unavailable_vfs_panic(void); 144 void rump__unavailable_vfs_panic() {panic("vfs component not available");} 145 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic); 146 147 rump_proc_vfs_init_fn rump_proc_vfs_init; 148 rump_proc_vfs_release_fn rump_proc_vfs_release; 149 150 static void add_linkedin_modules(const struct modinfo *const *, size_t); 151 152 static void __noinline 153 messthestack(void) 154 { 155 volatile uint32_t mess[64]; 156 uint64_t d1, d2; 157 int i, error; 158 159 for (i = 0; i < 64; i++) { 160 rumpuser_gettime(&d1, &d2, &error); 161 mess[i] = d2; 162 } 163 } 164 165 int 166 rump__init(int rump_version) 167 { 168 char buf[256]; 169 uint64_t sec, nsec; 170 struct proc *p; 171 struct lwp *l; 172 int i; 173 int error; 174 175 /* not reentrant */ 176 if (rump_inited) 177 return 0; 178 else if (rump_inited == -1) 179 panic("rump_init: host process restart required"); 180 else 181 rump_inited = 1; 182 183 if (rumpuser_getenv("RUMP_VERBOSE", buf, sizeof(buf), &error) == 0) { 184 if (*buf != '0') 185 boothowto = AB_VERBOSE; 186 } 187 188 rumpuser_gettime(&sec, &nsec, &error); 189 boottime.tv_sec = sec; 190 boottime.tv_nsec = nsec; 191 192 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf)); 193 aprint_verbose("%s%s", copyright, version); 194 195 /* 196 * Seed arc4random() with a "reasonable" amount of randomness. 197 * Yes, this is a quick kludge which depends on the arc4random 198 * implementation. 199 */ 200 messthestack(); 201 arc4random(); 202 203 if (rump_version != RUMP_VERSION) { 204 printf("rump version mismatch, %d vs. %d\n", 205 rump_version, RUMP_VERSION); 206 return EPROGMISMATCH; 207 } 208 209 if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) { 210 rump_threads = *buf != '0'; 211 } 212 rumpuser_thrinit(rump_user_schedule, rump_user_unschedule, 213 rump_threads); 214 rump_intr_init(); 215 216 /* init minimal lwp/cpu context */ 217 l = &lwp0; 218 l->l_lid = 1; 219 l->l_cpu = rump_cpu; 220 rumpuser_set_curlwp(l); 221 222 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE); 223 rumpuser_mutex_recursive_init(&rump_giantlock); 224 ksyms_init(); 225 rumpvm_init(); 226 evcnt_init(); 227 228 once_init(); 229 prop_kern_init(); 230 231 pool_subsystem_init(); 232 kmem_init(); 233 234 uvm_ra_init(); 235 236 mutex_obj_init(); 237 callout_startup(); 238 239 kprintf_init(); 240 loginit(); 241 242 kauth_init(); 243 rump_susercred = rump_cred_create(0, 0, 0, NULL); 244 245 /* init proc0 and rest of lwp0 now that we can allocate memory */ 246 p = &proc0; 247 p->p_stats = &rump_stats; 248 p->p_limit = &rump_limits; 249 p->p_pgrp = &rump_pgrp; 250 p->p_pid = 0; 251 p->p_fd = &rump_filedesc0; 252 p->p_vmspace = &rump_vmspace; 253 p->p_emul = &emul_rump; 254 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 255 l->l_cred = rump_cred_suserget(); 256 l->l_proc = p; 257 LIST_INIT(&allproc); 258 LIST_INSERT_HEAD(&allproc, &proc0, p_list); 259 proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 260 261 rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; 262 rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY; 263 rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY; 264 265 rump_scheduler_init(); 266 /* revert temporary context and schedule a real context */ 267 l->l_cpu = NULL; 268 rumpuser_set_curlwp(NULL); 269 rump_schedule(); 270 271 percpu_init(); 272 273 /* we are mostly go. do per-cpu subsystem init */ 274 for (i = 0; i < ncpu; i++) { 275 struct cpu_info *ci = cpu_lookup(i); 276 277 callout_init_cpu(ci); 278 softint_init(ci); 279 xc_init_cpu(ci); 280 pool_cache_cpu_init(ci); 281 selsysinit(ci); 282 percpu_init_cpu(ci); 283 } 284 285 sysctl_init(); 286 kqueue_init(); 287 iostat_init(); 288 uid_init(); 289 fd_sys_init(); 290 module_init(); 291 devsw_init(); 292 pipe_init(); 293 294 rumpuser_dl_bootstrap(add_linkedin_modules, rump_kernelfsym_load); 295 296 /* these do nothing if not present */ 297 rump_vfs_init(); 298 rump_net_init(); 299 rump_dev_init(); 300 cold = 0; 301 302 /* aieeeedondest */ 303 if (rump_threads) { 304 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 305 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 306 panic("aiodoned"); 307 } 308 309 sysctl_finalize(); 310 311 module_init_class(MODULE_CLASS_ANY); 312 313 rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error); 314 hostnamelen = strlen(hostname); 315 316 sigemptyset(&sigcantmask); 317 318 lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0); 319 320 if (rump_threads) 321 vmem_rehash_start(); 322 323 rump_unschedule(); 324 325 return 0; 326 } 327 328 /* maybe support sys_reboot some day for remote shutdown */ 329 void 330 rump_reboot(int howto) 331 { 332 333 /* dump means we really take the dive here */ 334 if ((howto & RB_DUMP) || panicstr) { 335 rumpuser_exit(RUMPUSER_PANIC); 336 /*NOTREACHED*/ 337 } 338 339 /* try to sync */ 340 if (!((howto & RB_NOSYNC) || panicstr)) { 341 rump_vfs_fini(); 342 } 343 344 /* your wish is my command */ 345 if (howto & RB_HALT) { 346 for (;;) { 347 uint64_t sec = 5, nsec = 0; 348 int error; 349 350 rumpuser_nanosleep(&sec, &nsec, &error); 351 } 352 } 353 rump_inited = -1; 354 } 355 356 struct uio * 357 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw) 358 { 359 struct uio *uio; 360 enum uio_rw uiorw; 361 362 switch (rw) { 363 case RUMPUIO_READ: 364 uiorw = UIO_READ; 365 break; 366 case RUMPUIO_WRITE: 367 uiorw = UIO_WRITE; 368 break; 369 default: 370 panic("%s: invalid rw %d", __func__, rw); 371 } 372 373 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP); 374 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP); 375 376 uio->uio_iov->iov_base = buf; 377 uio->uio_iov->iov_len = bufsize; 378 379 uio->uio_iovcnt = 1; 380 uio->uio_offset = offset; 381 uio->uio_resid = bufsize; 382 uio->uio_rw = uiorw; 383 uio->uio_vmspace = UIO_VMSPACE_SYS; 384 385 return uio; 386 } 387 388 size_t 389 rump_uio_getresid(struct uio *uio) 390 { 391 392 return uio->uio_resid; 393 } 394 395 off_t 396 rump_uio_getoff(struct uio *uio) 397 { 398 399 return uio->uio_offset; 400 } 401 402 size_t 403 rump_uio_free(struct uio *uio) 404 { 405 size_t resid; 406 407 resid = uio->uio_resid; 408 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov)); 409 kmem_free(uio, sizeof(*uio)); 410 411 return resid; 412 } 413 414 static pid_t nextpid = 1; 415 struct lwp * 416 rump_newproc_switch() 417 { 418 struct lwp *l; 419 pid_t mypid; 420 421 mypid = atomic_inc_uint_nv(&nextpid); 422 if (__predict_false(mypid == 0)) 423 mypid = atomic_inc_uint_nv(&nextpid); 424 425 l = rump_lwp_alloc(mypid, 0); 426 rump_lwp_switch(l); 427 428 return l; 429 } 430 431 struct lwp * 432 rump_lwp_alloc_and_switch(pid_t pid, lwpid_t lid) 433 { 434 struct lwp *l; 435 436 l = rump_lwp_alloc(pid, lid); 437 rump_lwp_switch(l); 438 439 return l; 440 } 441 442 struct lwp * 443 rump_lwp_alloc(pid_t pid, lwpid_t lid) 444 { 445 struct lwp *l; 446 struct proc *p; 447 448 l = kmem_zalloc(sizeof(*l), KM_SLEEP); 449 if (pid != 0) { 450 p = kmem_zalloc(sizeof(*p), KM_SLEEP); 451 if (rump_proc_vfs_init) 452 rump_proc_vfs_init(p); 453 p->p_stats = &rump_stats; 454 p->p_limit = &rump_limits; 455 p->p_pid = pid; 456 p->p_vmspace = &rump_vmspace; 457 p->p_emul = &emul_rump; 458 p->p_fd = fd_init(NULL); 459 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 460 l->l_cred = rump_cred_suserget(); 461 } else { 462 p = &proc0; 463 l->l_cred = rump_susercred; 464 } 465 466 l->l_proc = p; 467 l->l_lid = lid; 468 l->l_fd = p->p_fd; 469 l->l_cpu = NULL; 470 471 return l; 472 } 473 474 void 475 rump_lwp_switch(struct lwp *newlwp) 476 { 477 struct lwp *l = curlwp; 478 479 rumpuser_set_curlwp(NULL); 480 newlwp->l_cpu = l->l_cpu; 481 newlwp->l_mutex = l->l_mutex; 482 l->l_mutex = NULL; 483 l->l_cpu = NULL; 484 rumpuser_set_curlwp(newlwp); 485 if (l->l_flag & LW_WEXIT) 486 rump_lwp_free(l); 487 } 488 489 /* XXX: this has effect only on non-pid0 lwps */ 490 void 491 rump_lwp_release(struct lwp *l) 492 { 493 struct proc *p; 494 495 p = l->l_proc; 496 if (p->p_pid != 0) { 497 mutex_obj_free(p->p_lock); 498 fd_free(); 499 if (rump_proc_vfs_release) 500 rump_proc_vfs_release(p); 501 rump_cred_put(l->l_cred); 502 kmem_free(p, sizeof(*p)); 503 } 504 KASSERT((l->l_flag & LW_WEXIT) == 0); 505 l->l_flag |= LW_WEXIT; 506 } 507 508 void 509 rump_lwp_free(struct lwp *l) 510 { 511 512 KASSERT(l->l_flag & LW_WEXIT); 513 KASSERT(l->l_mutex == NULL); 514 if (l->l_name) 515 kmem_free(l->l_name, MAXCOMLEN); 516 kmem_free(l, sizeof(*l)); 517 } 518 519 struct lwp * 520 rump_lwp_curlwp(void) 521 { 522 struct lwp *l = curlwp; 523 524 if (l->l_flag & LW_WEXIT) 525 return NULL; 526 return l; 527 } 528 529 /* rump private. NEEDS WORK! */ 530 void 531 rump_set_vmspace(struct vmspace *vm) 532 { 533 struct proc *p = curproc; 534 535 p->p_vmspace = vm; 536 } 537 538 kauth_cred_t 539 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups) 540 { 541 kauth_cred_t cred; 542 int rv; 543 544 cred = kauth_cred_alloc(); 545 kauth_cred_setuid(cred, uid); 546 kauth_cred_seteuid(cred, uid); 547 kauth_cred_setsvuid(cred, uid); 548 kauth_cred_setgid(cred, gid); 549 kauth_cred_setgid(cred, gid); 550 kauth_cred_setegid(cred, gid); 551 kauth_cred_setsvgid(cred, gid); 552 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE); 553 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */ 554 assert(rv == 0); 555 556 return cred; 557 } 558 559 void 560 rump_cred_put(kauth_cred_t cred) 561 { 562 563 kauth_cred_free(cred); 564 } 565 566 kauth_cred_t 567 rump_cred_suserget(void) 568 { 569 570 kauth_cred_hold(rump_susercred); 571 return rump_susercred; 572 } 573 574 /* 575 * Return the next system lwpid 576 */ 577 lwpid_t 578 rump_nextlid(void) 579 { 580 lwpid_t retid; 581 582 mutex_enter(proc0.p_lock); 583 /* 584 * Take next one, don't return 0 585 * XXX: most likely we'll have collisions in case this 586 * wraps around. 587 */ 588 if (++proc0.p_nlwpid == 0) 589 ++proc0.p_nlwpid; 590 retid = proc0.p_nlwpid; 591 mutex_exit(proc0.p_lock); 592 593 return retid; 594 } 595 596 static int compcounter[RUMP_COMPONENT_MAX]; 597 598 static void 599 rump_component_init_cb(struct rump_component *rc, int type) 600 { 601 602 KASSERT(type < RUMP_COMPONENT_MAX); 603 if (rc->rc_type == type) { 604 rc->rc_init(); 605 compcounter[type]++; 606 } 607 } 608 609 int 610 rump_component_count(enum rump_component_type type) 611 { 612 613 KASSERT(type <= RUMP_COMPONENT_MAX); 614 return compcounter[type]; 615 } 616 617 void 618 rump_component_init(enum rump_component_type type) 619 { 620 621 rumpuser_dl_component_init(type, rump_component_init_cb); 622 } 623 624 /* 625 * Initialize a module which has already been loaded and linked 626 * with dlopen(). This is fundamentally the same as a builtin module. 627 */ 628 int 629 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo) 630 { 631 632 return module_builtin_add(mip, nmodinfo, true); 633 } 634 635 /* 636 * Finish module (flawless victory, fatality!). 637 */ 638 int 639 rump_module_fini(const struct modinfo *mi) 640 { 641 642 return module_builtin_remove(mi, true); 643 } 644 645 /* 646 * Add loaded and linked module to the builtin list. It will 647 * later be initialized with module_init_class(). 648 */ 649 650 static void 651 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo) 652 { 653 654 module_builtin_add(mip, nmodinfo, false); 655 } 656 657 int 658 rump_kernelfsym_load(void *symtab, uint64_t symsize, 659 char *strtab, uint64_t strsize) 660 { 661 static int inited = 0; 662 Elf64_Ehdr ehdr; 663 664 if (inited) 665 return EBUSY; 666 inited = 1; 667 668 /* 669 * Use 64bit header since it's bigger. Shouldn't make a 670 * difference, since we're passing in all zeroes anyway. 671 */ 672 memset(&ehdr, 0, sizeof(ehdr)); 673 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize); 674 675 return 0; 676 } 677 678 static int 679 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen, 680 register_t *retval) 681 { 682 struct lwp *l; 683 struct sysent *callp; 684 int rv; 685 686 if (__predict_false(num >= SYS_NSYSENT)) 687 return ENOSYS; 688 689 callp = rump_sysent + num; 690 rump_schedule(); 691 l = curlwp; 692 rv = callp->sy_call(l, (void *)data, retval); 693 rump_unschedule(); 694 695 return rv; 696 } 697 698 int 699 rump_boot_gethowto() 700 { 701 702 return boothowto; 703 } 704 705 void 706 rump_boot_sethowto(int howto) 707 { 708 709 boothowto = howto; 710 } 711 712 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local; 713 void *rump_sysproxy_arg; 714 715 /* 716 * This whole syscall-via-rpc is still taking form. For example, it 717 * may be necessary to set syscalls individually instead of lobbing 718 * them all to the same place. So don't think this interface is 719 * set in stone. 720 */ 721 int 722 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg) 723 { 724 725 if (rump_sysproxy_arg) 726 return EBUSY; 727 728 rump_sysproxy_arg = arg; 729 rump_sysproxy = proxy; 730 731 return 0; 732 } 733 734 int 735 rump_getversion(void) 736 { 737 738 return __NetBSD_Version__; 739 } 740
Indexes created Sun Sep 28 07:09:56 GMT 2025