rump.c revision 1.278
1 /* $NetBSD: rump.c,v 1.278 2013/12/09 16:54:20 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 2007-2011 Antti Kantee. All Rights Reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS 16 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 17 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 18 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 21 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.278 2013/12/09 16:54:20 pooka Exp $"); 30 31 #include <sys/systm.h> 32 #define ELFSIZE ARCH_ELFSIZE 33 34 #include <sys/param.h> 35 #include <sys/atomic.h> 36 #include <sys/buf.h> 37 #include <sys/callout.h> 38 #include <sys/conf.h> 39 #include <sys/cpu.h> 40 #include <sys/device.h> 41 #include <sys/evcnt.h> 42 #include <sys/event.h> 43 #include <sys/exec_elf.h> 44 #include <sys/filedesc.h> 45 #include <sys/iostat.h> 46 #include <sys/kauth.h> 47 #include <sys/kcpuset.h> 48 #include <sys/kernel.h> 49 #include <sys/kmem.h> 50 #include <sys/kprintf.h> 51 #include <sys/kthread.h> 52 #include <sys/ksyms.h> 53 #include <sys/msgbuf.h> 54 #include <sys/module.h> 55 #include <sys/namei.h> 56 #include <sys/once.h> 57 #include <sys/percpu.h> 58 #include <sys/pipe.h> 59 #include <sys/pool.h> 60 #include <sys/pserialize.h> 61 #include <sys/queue.h> 62 #include <sys/reboot.h> 63 #include <sys/resourcevar.h> 64 #include <sys/select.h> 65 #include <sys/sysctl.h> 66 #include <sys/syscall.h> 67 #include <sys/syscallvar.h> 68 #include <sys/timetc.h> 69 #include <sys/tty.h> 70 #include <sys/uidinfo.h> 71 #include <sys/vmem.h> 72 #include <sys/xcall.h> 73 #include <sys/simplelock.h> 74 #include <sys/cprng.h> 75 76 #include <rump/rumpuser.h> 77 78 #include <secmodel/suser/suser.h> 79 80 #include <prop/proplib.h> 81 82 #include <uvm/uvm_extern.h> 83 #include <uvm/uvm_readahead.h> 84 85 #include "rump_private.h" 86 #include "rump_net_private.h" 87 #include "rump_vfs_private.h" 88 #include "rump_dev_private.h" 89 90 char machine[] = MACHINE; 91 92 struct proc *initproc; 93 94 struct device rump_rootdev = { 95 .dv_class = DV_VIRTUAL 96 }; 97 98 #ifdef RUMP_WITHOUT_THREADS 99 int rump_threads = 0; 100 #else 101 int rump_threads = 1; 102 #endif 103 104 static int rump_hyp_syscall(int, void *, long *); 105 static int rump_hyp_rfork(void *, int, const char *); 106 static void rump_hyp_lwpexit(void); 107 static void rump_hyp_execnotify(const char *); 108 109 static void rump_component_addlocal(void); 110 static void rump_component_load(const struct rump_component *); 111 static struct lwp *bootlwp; 112 113 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */ 114 115 #ifdef LOCKDEBUG 116 const int rump_lockdebug = 1; 117 #else 118 const int rump_lockdebug = 0; 119 #endif 120 bool rump_ttycomponent = false; 121 122 static void 123 rump_aiodone_worker(struct work *wk, void *dummy) 124 { 125 struct buf *bp = (struct buf *)wk; 126 127 KASSERT(&bp->b_work == wk); 128 bp->b_iodone(bp); 129 } 130 131 static int rump_inited; 132 133 void (*rump_vfs_drainbufs)(int); 134 void (*rump_vfs_fini)(void); 135 int (*rump_vfs_makeonedevnode)(dev_t, const char *, 136 devmajor_t, devminor_t) = (void *)nullop; 137 int (*rump_vfs_makedevnodes)(dev_t, const char *, char, 138 devmajor_t, devminor_t, int) = (void *)nullop; 139 140 int rump__unavailable(void); 141 int rump__unavailable() {return EOPNOTSUPP;} 142 143 __weak_alias(biodone,rump__unavailable); 144 __weak_alias(sopoll,rump__unavailable); 145 146 void rump__unavailable_vfs_panic(void); 147 void rump__unavailable_vfs_panic() {panic("vfs component not available");} 148 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic); 149 150 /* easier to write vfs-less clients */ 151 __weak_alias(rump_pub_etfs_register,rump__unavailable); 152 __weak_alias(rump_pub_etfs_register_withsize,rump__unavailable); 153 __weak_alias(rump_pub_etfs_remove,rump__unavailable); 154 155 rump_proc_vfs_init_fn rump_proc_vfs_init; 156 rump_proc_vfs_release_fn rump_proc_vfs_release; 157 158 static void add_linkedin_modules(const struct modinfo *const *, size_t); 159 160 /* 161 * Create some sysctl nodes. why only this you ask. well, init_sysctl 162 * is a kitchen sink in need of some gardening. but i want to use 163 * others today. Furthermore, creating a whole kitchen sink full of 164 * sysctl nodes is a waste of cycles for rump kernel bootstrap. 165 */ 166 static void 167 mksysctls(void) 168 { 169 170 /* kern.hostname */ 171 sysctl_createv(NULL, 0, NULL, NULL, 172 CTLFLAG_PERMANENT, CTLTYPE_NODE, "kern", NULL, 173 NULL, 0, NULL, 0, CTL_KERN, CTL_EOL); 174 /* XXX: setting hostnamelen is missing */ 175 sysctl_createv(NULL, 0, NULL, NULL, 176 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_STRING, "hostname", 177 SYSCTL_DESCR("System hostname"), NULL, 0, 178 hostname, MAXHOSTNAMELEN, CTL_KERN, KERN_HOSTNAME, CTL_EOL); 179 180 /* hw.pagesize */ 181 sysctl_createv(NULL, 0, NULL, NULL, 182 CTLFLAG_PERMANENT, 183 CTLTYPE_NODE, "hw", NULL, 184 NULL, 0, NULL, 0, 185 CTL_HW, CTL_EOL); 186 sysctl_createv(NULL, 0, NULL, NULL, 187 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE, 188 CTLTYPE_INT, "pagesize", 189 SYSCTL_DESCR("Software page size"), 190 NULL, PAGE_SIZE, NULL, 0, 191 CTL_HW, HW_PAGESIZE, CTL_EOL); 192 } 193 194 /* there's no convenient kernel entry point for this, so just craft out own */ 195 static pid_t 196 spgetpid(void) 197 { 198 199 return curproc->p_pid; 200 } 201 202 static const struct rumpuser_hyperup hyp = { 203 .hyp_schedule = rump_schedule, 204 .hyp_unschedule = rump_unschedule, 205 .hyp_backend_unschedule = rump_user_unschedule, 206 .hyp_backend_schedule = rump_user_schedule, 207 .hyp_lwproc_switch = rump_lwproc_switch, 208 .hyp_lwproc_release = rump_lwproc_releaselwp, 209 .hyp_lwproc_rfork = rump_hyp_rfork, 210 .hyp_lwproc_newlwp = rump_lwproc_newlwp, 211 .hyp_lwproc_curlwp = rump_lwproc_curlwp, 212 .hyp_lwpexit = rump_hyp_lwpexit, 213 .hyp_syscall = rump_hyp_syscall, 214 .hyp_execnotify = rump_hyp_execnotify, 215 .hyp_getpid = spgetpid, 216 }; 217 218 int 219 rump_daemonize_begin(void) 220 { 221 222 if (rump_inited) 223 return EALREADY; 224 225 return rumpuser_daemonize_begin(); 226 } 227 228 int 229 rump_daemonize_done(int error) 230 { 231 232 return rumpuser_daemonize_done(error); 233 } 234 235 RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT) 236 { 237 __link_set_decl(rump_components, struct rump_component); 238 239 /* 240 * Trick compiler into generating references so that statically 241 * linked rump kernels are generated with the link set symbols. 242 */ 243 asm("" :: "r"(__start_link_set_rump_components)); 244 asm("" :: "r"(__stop_link_set_rump_components)); 245 } 246 247 int 248 rump_init(void) 249 { 250 char buf[256]; 251 struct timespec ts; 252 int64_t sec; 253 long nsec; 254 struct lwp *l, *initlwp; 255 int i, numcpu; 256 257 /* not reentrant */ 258 if (rump_inited) 259 return 0; 260 else if (rump_inited == -1) 261 panic("rump_init: host process restart required"); 262 else 263 rump_inited = 1; 264 265 /* initialize hypervisor */ 266 if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) { 267 rumpuser_dprintf("rumpuser init failed\n"); 268 return EINVAL; 269 } 270 271 /* init minimal lwp/cpu context */ 272 l = &lwp0; 273 l->l_lid = 1; 274 l->l_cpu = l->l_target_cpu = rump_cpu; 275 l->l_fd = &filedesc0; 276 277 /* lwp0 isn't created like other threads, so notify hypervisor here */ 278 rumpuser_curlwpop(RUMPUSER_LWP_CREATE, l); 279 rumpuser_curlwpop(RUMPUSER_LWP_SET, l); 280 281 /* retrieve env vars which affect the early stage of bootstrap */ 282 if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) { 283 rump_threads = *buf != '0'; 284 } 285 if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) { 286 if (*buf != '0') 287 boothowto = AB_VERBOSE; 288 } 289 290 if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0) 291 panic("mandatory hypervisor configuration (NCPU) missing"); 292 numcpu = strtoll(buf, NULL, 10); 293 if (numcpu < 1) { 294 panic("rump kernels are not lightweight enough for \"%d\" CPUs", 295 numcpu); 296 } 297 298 rump_thread_init(); 299 rump_cpus_bootstrap(&numcpu); 300 301 rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec); 302 boottime.tv_sec = sec; 303 boottime.tv_nsec = nsec; 304 305 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf)); 306 aprint_verbose("%s%s", copyright, version); 307 308 rump_intr_init(numcpu); 309 310 rump_tsleep_init(); 311 312 rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN); 313 ksyms_init(); 314 uvm_init(); 315 evcnt_init(); 316 317 kcpuset_sysinit(); 318 once_init(); 319 kernconfig_lock_init(); 320 prop_kern_init(); 321 322 kmem_init(); 323 kmeminit(); 324 325 uvm_ra_init(); 326 uao_init(); 327 328 mutex_obj_init(); 329 callout_startup(); 330 331 kprintf_init(); 332 pserialize_init(); 333 loginit(); 334 335 kauth_init(); 336 337 secmodel_init(); 338 339 rnd_init(); 340 341 /* 342 * Create the kernel cprng. Yes, it's currently stubbed out 343 * to arc4random() for RUMP, but this won't always be so. 344 */ 345 kern_cprng = cprng_strong_create("kernel", IPL_VM, 346 CPRNG_INIT_ANY|CPRNG_REKEY_ANY); 347 348 procinit(); 349 proc0_init(); 350 sysctl_init(); 351 uid_init(); 352 chgproccnt(0, 1); 353 354 l->l_proc = &proc0; 355 lwp_update_creds(l); 356 357 lwpinit_specificdata(); 358 lwp_initspecific(&lwp0); 359 360 rump_biglock_init(); 361 362 rump_scheduler_init(numcpu); 363 /* revert temporary context and schedule a semireal context */ 364 rumpuser_curlwpop(RUMPUSER_LWP_CLEAR, l); 365 initproc = &proc0; /* borrow proc0 before we get initproc started */ 366 rump_schedule(); 367 bootlwp = curlwp; 368 369 percpu_init(); 370 inittimecounter(); 371 ntp_init(); 372 373 ktrinit(); 374 375 ts = boottime; 376 tc_setclock(&ts); 377 378 /* we are mostly go. do per-cpu subsystem init */ 379 for (i = 0; i < numcpu; i++) { 380 struct cpu_info *ci = cpu_lookup(i); 381 382 /* attach non-bootstrap CPUs */ 383 if (i > 0) { 384 rump_cpu_attach(ci); 385 ncpu++; 386 } 387 388 callout_init_cpu(ci); 389 softint_init(ci); 390 xc_init_cpu(ci); 391 pool_cache_cpu_init(ci); 392 selsysinit(ci); 393 percpu_init_cpu(ci); 394 395 TAILQ_INIT(&ci->ci_data.cpu_ld_locks); 396 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock); 397 398 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i); 399 } 400 401 /* CPUs are up. allow kernel threads to run */ 402 rump_thread_allow(); 403 404 mksysctls(); 405 kqueue_init(); 406 iostat_init(); 407 fd_sys_init(); 408 module_init(); 409 devsw_init(); 410 pipe_init(); 411 resource_init(); 412 procinit_sysctl(); 413 414 /* start page baroness */ 415 if (rump_threads) { 416 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL, 417 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0) 418 panic("pagedaemon create failed"); 419 } else 420 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */ 421 422 /* process dso's */ 423 rumpuser_dl_bootstrap(add_linkedin_modules, 424 rump_kernelfsym_load, rump_component_load); 425 426 rump_component_addlocal(); 427 rump_component_init(RUMP_COMPONENT_KERN); 428 429 /* initialize factions, if present */ 430 rump_component_init(RUMP__FACTION_VFS); 431 /* pnbuf_cache is used even without vfs */ 432 if (rump_component_count(RUMP__FACTION_VFS) == 0) { 433 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl", 434 NULL, IPL_NONE, NULL, NULL, NULL); 435 } 436 rump_component_init(RUMP__FACTION_NET); 437 rump_component_init(RUMP__FACTION_DEV); 438 KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1 439 && rump_component_count(RUMP__FACTION_NET) <= 1 440 && rump_component_count(RUMP__FACTION_DEV) <= 1); 441 442 rump_component_init(RUMP_COMPONENT_KERN_VFS); 443 444 /* 445 * if we initialized the tty component above, the tyttymtx is 446 * now initialized. otherwise, we need to initialize it. 447 */ 448 if (!rump_ttycomponent) 449 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM); 450 451 cold = 0; 452 453 /* aieeeedondest */ 454 if (rump_threads) { 455 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 456 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 457 panic("aiodoned"); 458 } 459 460 sysctl_finalize(); 461 462 module_init_class(MODULE_CLASS_ANY); 463 464 if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME, 465 hostname, MAXHOSTNAMELEN) != 0) { 466 panic("mandatory hypervisor configuration (HOSTNAME) missing"); 467 } 468 hostnamelen = strlen(hostname); 469 470 sigemptyset(&sigcantmask); 471 472 if (rump_threads) 473 vmem_rehash_start(); 474 475 /* 476 * Create init (proc 1), used to attach implicit threads in rump. 477 * (note: must be done after vfsinit to get cwdi) 478 */ 479 initlwp = rump__lwproc_alloclwp(NULL); 480 mutex_enter(proc_lock); 481 initproc = proc_find_raw(1); 482 mutex_exit(proc_lock); 483 if (initproc == NULL) 484 panic("where in the world is initproc?"); 485 486 /* 487 * Adjust syscall vector in case factions were dlopen()'d 488 * before calling rump_init(). 489 * (modules will handle dynamic syscalls the usual way) 490 * 491 * Note: this will adjust the function vectors of 492 * syscalls which use a funcalias (getpid etc.), but 493 * it makes no difference. 494 */ 495 for (i = 0; i < SYS_NSYSENT; i++) { 496 void *sym; 497 498 if (rump_sysent[i].sy_flags & SYCALL_NOSYS || 499 *syscallnames[i] == '#' || 500 rump_sysent[i].sy_call == sys_nomodule) 501 continue; 502 503 /* 504 * deal with compat wrappers. makesyscalls.sh should 505 * generate the necessary info instead of this hack, 506 * though. ugly, fix it later. 507 */ 508 #define CPFX "compat_" 509 #define CPFXLEN (sizeof(CPFX)-1) 510 if (strncmp(syscallnames[i], CPFX, CPFXLEN) == 0) { 511 const char *p = syscallnames[i] + CPFXLEN; 512 size_t namelen; 513 514 /* skip version number */ 515 while (*p >= '0' && *p <= '9') 516 p++; 517 if (p == syscallnames[i] + CPFXLEN || *p != '_') 518 panic("invalid syscall name %s\n", 519 syscallnames[i]); 520 521 /* skip over the next underscore */ 522 p++; 523 namelen = p + (sizeof("rumpns_")-1) - syscallnames[i]; 524 525 strcpy(buf, "rumpns_"); 526 strcat(buf, syscallnames[i]); 527 /* XXX: no strncat in the kernel */ 528 strcpy(buf+namelen, "sys_"); 529 strcat(buf, p); 530 #undef CPFX 531 #undef CPFXLEN 532 } else { 533 sprintf(buf, "rumpns_sys_%s", syscallnames[i]); 534 } 535 if ((sym = rumpuser_dl_globalsym(buf)) != NULL 536 && sym != rump_sysent[i].sy_call) { 537 #if 0 538 rumpuser_dprintf("adjusting %s: %p (old %p)\n", 539 syscallnames[i], sym, rump_sysent[i].sy_call); 540 #endif 541 rump_sysent[i].sy_call = sym; 542 } 543 } 544 545 rump_component_init(RUMP_COMPONENT_POSTINIT); 546 547 /* component inits done */ 548 bootlwp = NULL; 549 550 /* open 0/1/2 for init */ 551 KASSERT(rump_lwproc_curlwp() == NULL); 552 rump_lwproc_switch(initlwp); 553 rump_consdev_init(); 554 rump_lwproc_switch(NULL); 555 556 /* release cpu */ 557 rump_unschedule(); 558 559 return 0; 560 } 561 /* historic compat */ 562 __strong_alias(rump__init,rump_init); 563 564 int 565 rump_init_server(const char *url) 566 { 567 568 return rumpuser_sp_init(url, ostype, osrelease, MACHINE); 569 } 570 571 void 572 cpu_reboot(int howto, char *bootstr) 573 { 574 int ruhow = 0; 575 void *finiarg; 576 577 printf("rump kernel halting...\n"); 578 579 if (!RUMP_LOCALPROC_P(curproc)) 580 finiarg = curproc->p_vmspace->vm_map.pmap; 581 else 582 finiarg = NULL; 583 584 /* dump means we really take the dive here */ 585 if ((howto & RB_DUMP) || panicstr) { 586 ruhow = RUMPUSER_PANIC; 587 goto out; 588 } 589 590 /* try to sync */ 591 if (!((howto & RB_NOSYNC) || panicstr)) { 592 if (rump_vfs_fini) 593 rump_vfs_fini(); 594 } 595 596 doshutdownhooks(); 597 598 /* your wish is my command */ 599 if (howto & RB_HALT) { 600 printf("rump kernel halted\n"); 601 rumpuser_sp_fini(finiarg); 602 for (;;) { 603 rumpuser_clock_sleep(RUMPUSER_CLOCK_RELWALL, 10, 0); 604 } 605 } 606 607 /* this function is __dead, we must exit */ 608 out: 609 printf("halted\n"); 610 rumpuser_sp_fini(finiarg); 611 rumpuser_exit(ruhow); 612 } 613 614 struct uio * 615 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw) 616 { 617 struct uio *uio; 618 enum uio_rw uiorw; 619 620 switch (rw) { 621 case RUMPUIO_READ: 622 uiorw = UIO_READ; 623 break; 624 case RUMPUIO_WRITE: 625 uiorw = UIO_WRITE; 626 break; 627 default: 628 panic("%s: invalid rw %d", __func__, rw); 629 } 630 631 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP); 632 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP); 633 634 uio->uio_iov->iov_base = buf; 635 uio->uio_iov->iov_len = bufsize; 636 637 uio->uio_iovcnt = 1; 638 uio->uio_offset = offset; 639 uio->uio_resid = bufsize; 640 uio->uio_rw = uiorw; 641 UIO_SETUP_SYSSPACE(uio); 642 643 return uio; 644 } 645 646 size_t 647 rump_uio_getresid(struct uio *uio) 648 { 649 650 return uio->uio_resid; 651 } 652 653 off_t 654 rump_uio_getoff(struct uio *uio) 655 { 656 657 return uio->uio_offset; 658 } 659 660 size_t 661 rump_uio_free(struct uio *uio) 662 { 663 size_t resid; 664 665 resid = uio->uio_resid; 666 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov)); 667 kmem_free(uio, sizeof(*uio)); 668 669 return resid; 670 } 671 672 kauth_cred_t 673 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups) 674 { 675 kauth_cred_t cred; 676 int rv; 677 678 cred = kauth_cred_alloc(); 679 kauth_cred_setuid(cred, uid); 680 kauth_cred_seteuid(cred, uid); 681 kauth_cred_setsvuid(cred, uid); 682 kauth_cred_setgid(cred, gid); 683 kauth_cred_setgid(cred, gid); 684 kauth_cred_setegid(cred, gid); 685 kauth_cred_setsvgid(cred, gid); 686 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE); 687 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */ 688 assert(rv == 0); 689 690 return cred; 691 } 692 693 void 694 rump_cred_put(kauth_cred_t cred) 695 { 696 697 kauth_cred_free(cred); 698 } 699 700 static int compcounter[RUMP_COMPONENT_MAX]; 701 static int compinited[RUMP_COMPONENT_MAX]; 702 703 /* 704 * Yea, this is O(n^2), but we're only looking at a handful of components. 705 * Components are always initialized from the thread that called rump_init(). 706 * Could also free these when done with them, but prolly not worth it. 707 */ 708 struct compstore { 709 const struct rump_component *cs_rc; 710 LIST_ENTRY(compstore) cs_entries; 711 }; 712 static LIST_HEAD(, compstore) cshead = LIST_HEAD_INITIALIZER(cshead); 713 714 /* 715 * add components which are visible from the current object. 716 */ 717 static void 718 rump_component_addlocal(void) 719 { 720 __link_set_decl(rump_components, struct rump_component); 721 struct rump_component *const *rc; 722 723 __link_set_foreach(rc, rump_components) { 724 rump_component_load(*rc); 725 } 726 } 727 728 static void 729 rump_component_load(const struct rump_component *rc) 730 { 731 struct compstore *cs; 732 733 KASSERT(curlwp == bootlwp); 734 735 LIST_FOREACH(cs, &cshead, cs_entries) { 736 if (rc == cs->cs_rc) 737 return; 738 } 739 740 cs = kmem_alloc(sizeof(*cs), KM_SLEEP); 741 cs->cs_rc = rc; 742 LIST_INSERT_HEAD(&cshead, cs, cs_entries); 743 KASSERT(rc->rc_type < RUMP_COMPONENT_MAX); 744 compcounter[rc->rc_type]++; 745 } 746 747 int 748 rump_component_count(enum rump_component_type type) 749 { 750 751 KASSERT(curlwp == bootlwp); 752 KASSERT(type < RUMP_COMPONENT_MAX); 753 return compcounter[type]; 754 } 755 756 void 757 rump_component_init(enum rump_component_type type) 758 { 759 struct compstore *cs; 760 const struct rump_component *rc; 761 762 KASSERT(curlwp == bootlwp); 763 KASSERT(!compinited[type]); 764 LIST_FOREACH(cs, &cshead, cs_entries) { 765 rc = cs->cs_rc; 766 if (rc->rc_type == type) 767 rc->rc_init(); 768 } 769 compinited[type] = 1; 770 } 771 772 /* 773 * Initialize a module which has already been loaded and linked 774 * with dlopen(). This is fundamentally the same as a builtin module. 775 */ 776 int 777 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo) 778 { 779 780 return module_builtin_add(mip, nmodinfo, true); 781 } 782 783 /* 784 * Finish module (flawless victory, fatality!). 785 */ 786 int 787 rump_module_fini(const struct modinfo *mi) 788 { 789 790 return module_builtin_remove(mi, true); 791 } 792 793 /* 794 * Add loaded and linked module to the builtin list. It will 795 * later be initialized with module_init_class(). 796 */ 797 798 static void 799 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo) 800 { 801 802 module_builtin_add(mip, nmodinfo, false); 803 } 804 805 int 806 rump_kernelfsym_load(void *symtab, uint64_t symsize, 807 char *strtab, uint64_t strsize) 808 { 809 static int inited = 0; 810 Elf64_Ehdr ehdr; 811 812 if (inited) 813 return EBUSY; 814 inited = 1; 815 816 /* 817 * Use 64bit header since it's bigger. Shouldn't make a 818 * difference, since we're passing in all zeroes anyway. 819 */ 820 memset(&ehdr, 0, sizeof(ehdr)); 821 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize); 822 823 return 0; 824 } 825 826 static int 827 rump_hyp_syscall(int num, void *arg, long *retval) 828 { 829 register_t regrv[2] = {0, 0}; 830 struct lwp *l; 831 struct sysent *callp; 832 int rv; 833 834 if (__predict_false(num >= SYS_NSYSENT)) 835 return ENOSYS; 836 837 callp = rump_sysent + num; 838 l = curlwp; 839 rv = sy_invoke(callp, l, (void *)arg, regrv, num); 840 retval[0] = regrv[0]; 841 retval[1] = regrv[1]; 842 843 return rv; 844 } 845 846 static int 847 rump_hyp_rfork(void *priv, int flags, const char *comm) 848 { 849 struct vmspace *newspace; 850 struct proc *p; 851 int error; 852 853 if ((error = rump_lwproc_rfork(flags)) != 0) 854 return error; 855 856 /* 857 * Since it's a proxy proc, adjust the vmspace. 858 * Refcount will eternally be 1. 859 */ 860 p = curproc; 861 newspace = kmem_zalloc(sizeof(*newspace), KM_SLEEP); 862 newspace->vm_refcnt = 1; 863 newspace->vm_map.pmap = priv; 864 KASSERT(p->p_vmspace == vmspace_kernel()); 865 p->p_vmspace = newspace; 866 if (comm) 867 strlcpy(p->p_comm, comm, sizeof(p->p_comm)); 868 869 return 0; 870 } 871 872 /* 873 * Order all lwps in a process to exit. does *not* wait for them to drain. 874 */ 875 static void 876 rump_hyp_lwpexit(void) 877 { 878 struct proc *p = curproc; 879 uint64_t where; 880 struct lwp *l; 881 882 mutex_enter(p->p_lock); 883 /* 884 * First pass: mark all lwps in the process with LW_RUMP_QEXIT 885 * so that they know they should exit. 886 */ 887 LIST_FOREACH(l, &p->p_lwps, l_sibling) { 888 if (l == curlwp) 889 continue; 890 l->l_flag |= LW_RUMP_QEXIT; 891 } 892 mutex_exit(p->p_lock); 893 894 /* 895 * Next, make sure everyone on all CPUs sees our status 896 * update. This keeps threads inside cv_wait() and makes 897 * sure we don't access a stale cv pointer later when 898 * we wake up the threads. 899 */ 900 901 where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 902 xc_wait(where); 903 904 /* 905 * Ok, all lwps are either: 906 * 1) not in the cv code 907 * 2) sleeping on l->l_private 908 * 3) sleeping on p->p_waitcv 909 * 910 * Either way, l_private is stable until we set PS_RUMP_LWPEXIT 911 * in p->p_sflag. 912 */ 913 914 mutex_enter(p->p_lock); 915 LIST_FOREACH(l, &p->p_lwps, l_sibling) { 916 if (l->l_private) 917 cv_broadcast(l->l_private); 918 } 919 p->p_sflag |= PS_RUMP_LWPEXIT; 920 cv_broadcast(&p->p_waitcv); 921 mutex_exit(p->p_lock); 922 } 923 924 /* 925 * Notify process that all threads have been drained and exec is complete. 926 */ 927 static void 928 rump_hyp_execnotify(const char *comm) 929 { 930 struct proc *p = curproc; 931 932 fd_closeexec(); 933 mutex_enter(p->p_lock); 934 KASSERT(p->p_nlwps == 1 && p->p_sflag & PS_RUMP_LWPEXIT); 935 p->p_sflag &= ~PS_RUMP_LWPEXIT; 936 mutex_exit(p->p_lock); 937 strlcpy(p->p_comm, comm, sizeof(p->p_comm)); 938 } 939 940 int 941 rump_boot_gethowto() 942 { 943 944 return boothowto; 945 } 946 947 void 948 rump_boot_sethowto(int howto) 949 { 950 951 boothowto = howto; 952 } 953 954 int 955 rump_getversion(void) 956 { 957 958 return __NetBSD_Version__; 959 } 960 961 /* 962 * Note: may be called unscheduled. Not fully safe since no locking 963 * of allevents (currently that's not even available). 964 */ 965 void 966 rump_printevcnts() 967 { 968 struct evcnt *ev; 969 970 TAILQ_FOREACH(ev, &allevents, ev_list) 971 rumpuser_dprintf("%s / %s: %" PRIu64 "\n", 972 ev->ev_group, ev->ev_name, ev->ev_count); 973 } 974 975 /* 976 * If you use this interface ... well ... all bets are off. 977 * The original purpose is for the p2k fs server library to be 978 * able to use the same pid/lid for VOPs as the host kernel. 979 */ 980 void 981 rump_allbetsareoff_setid(pid_t pid, int lid) 982 { 983 struct lwp *l = curlwp; 984 struct proc *p = l->l_proc; 985 986 l->l_lid = lid; 987 p->p_pid = pid; 988 } 989 990 #include <sys/pserialize.h> 991 992 static void 993 ipiemu(void *a1, void *a2) 994 { 995 996 xc__highpri_intr(NULL); 997 pserialize_switchpoint(); 998 } 999 1000 void 1001 rump_xc_highpri(struct cpu_info *ci) 1002 { 1003 1004 if (ci) 1005 xc_unicast(0, ipiemu, NULL, NULL, ci); 1006 else 1007 xc_broadcast(0, ipiemu, NULL, NULL); 1008 } 1009 1010 int 1011 rump_syscall(int num, void *data, size_t dlen, register_t *retval) 1012 { 1013 struct proc *p; 1014 struct emul *e; 1015 struct sysent *callp; 1016 int rv; 1017 1018 rump_schedule(); 1019 p = curproc; 1020 e = p->p_emul; 1021 #ifndef __HAVE_MINIMAL_EMUL 1022 KASSERT(num > 0 && num < e->e_nsysent); 1023 #endif 1024 callp = e->e_sysent + num; 1025 1026 rv = sy_invoke(callp, curlwp, data, retval, num); 1027 rump_unschedule(); 1028 1029 return rv; 1030 } 1031
Indexes created Mon Sep 29 21:09:56 GMT 2025