rump.c revision 1.279
1 /* $NetBSD: rump.c,v 1.279 2013/12/09 16:56:11 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 1.279 2013/12/09 16:56:11 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 #include <sys/ktrace.h> 76 77 #include <rump/rumpuser.h> 78 79 #include <secmodel/suser/suser.h> 80 81 #include <prop/proplib.h> 82 83 #include <uvm/uvm_extern.h> 84 #include <uvm/uvm_readahead.h> 85 86 #include "rump_private.h" 87 #include "rump_net_private.h" 88 #include "rump_vfs_private.h" 89 #include "rump_dev_private.h" 90 91 char machine[] = MACHINE; 92 93 struct proc *initproc; 94 95 struct device rump_rootdev = { 96 .dv_class = DV_VIRTUAL 97 }; 98 99 #ifdef RUMP_WITHOUT_THREADS 100 int rump_threads = 0; 101 #else 102 int rump_threads = 1; 103 #endif 104 105 static int rump_hyp_syscall(int, void *, long *); 106 static int rump_hyp_rfork(void *, int, const char *); 107 static void rump_hyp_lwpexit(void); 108 static void rump_hyp_execnotify(const char *); 109 110 static void rump_component_addlocal(void); 111 static void rump_component_load(const struct rump_component *); 112 static struct lwp *bootlwp; 113 114 static char rump_msgbuf[16*1024]; /* 16k should be enough for std rump needs */ 115 116 #ifdef LOCKDEBUG 117 const int rump_lockdebug = 1; 118 #else 119 const int rump_lockdebug = 0; 120 #endif 121 bool rump_ttycomponent = false; 122 123 static void 124 rump_aiodone_worker(struct work *wk, void *dummy) 125 { 126 struct buf *bp = (struct buf *)wk; 127 128 KASSERT(&bp->b_work == wk); 129 bp->b_iodone(bp); 130 } 131 132 static int rump_inited; 133 134 void (*rump_vfs_drainbufs)(int); 135 void (*rump_vfs_fini)(void); 136 int (*rump_vfs_makeonedevnode)(dev_t, const char *, 137 devmajor_t, devminor_t) = (void *)nullop; 138 int (*rump_vfs_makedevnodes)(dev_t, const char *, char, 139 devmajor_t, devminor_t, int) = (void *)nullop; 140 141 int rump__unavailable(void); 142 int rump__unavailable() {return EOPNOTSUPP;} 143 144 __weak_alias(biodone,rump__unavailable); 145 __weak_alias(sopoll,rump__unavailable); 146 147 void rump__unavailable_vfs_panic(void); 148 void rump__unavailable_vfs_panic() {panic("vfs component not available");} 149 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic); 150 151 /* easier to write vfs-less clients */ 152 __weak_alias(rump_pub_etfs_register,rump__unavailable); 153 __weak_alias(rump_pub_etfs_register_withsize,rump__unavailable); 154 __weak_alias(rump_pub_etfs_remove,rump__unavailable); 155 156 rump_proc_vfs_init_fn rump_proc_vfs_init; 157 rump_proc_vfs_release_fn rump_proc_vfs_release; 158 159 static void add_linkedin_modules(const struct modinfo *const *, size_t); 160 161 /* 162 * Create some sysctl nodes. why only this you ask. well, init_sysctl 163 * is a kitchen sink in need of some gardening. but i want to use 164 * others today. Furthermore, creating a whole kitchen sink full of 165 * sysctl nodes is a waste of cycles for rump kernel bootstrap. 166 */ 167 static void 168 mksysctls(void) 169 { 170 171 /* kern.hostname */ 172 sysctl_createv(NULL, 0, NULL, NULL, 173 CTLFLAG_PERMANENT, CTLTYPE_NODE, "kern", NULL, 174 NULL, 0, NULL, 0, CTL_KERN, CTL_EOL); 175 /* XXX: setting hostnamelen is missing */ 176 sysctl_createv(NULL, 0, NULL, NULL, 177 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, CTLTYPE_STRING, "hostname", 178 SYSCTL_DESCR("System hostname"), NULL, 0, 179 hostname, MAXHOSTNAMELEN, CTL_KERN, KERN_HOSTNAME, CTL_EOL); 180 181 /* hw.pagesize */ 182 sysctl_createv(NULL, 0, NULL, NULL, 183 CTLFLAG_PERMANENT, 184 CTLTYPE_NODE, "hw", NULL, 185 NULL, 0, NULL, 0, 186 CTL_HW, CTL_EOL); 187 sysctl_createv(NULL, 0, NULL, NULL, 188 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE, 189 CTLTYPE_INT, "pagesize", 190 SYSCTL_DESCR("Software page size"), 191 NULL, PAGE_SIZE, NULL, 0, 192 CTL_HW, HW_PAGESIZE, CTL_EOL); 193 } 194 195 /* there's no convenient kernel entry point for this, so just craft out own */ 196 static pid_t 197 spgetpid(void) 198 { 199 200 return curproc->p_pid; 201 } 202 203 static const struct rumpuser_hyperup hyp = { 204 .hyp_schedule = rump_schedule, 205 .hyp_unschedule = rump_unschedule, 206 .hyp_backend_unschedule = rump_user_unschedule, 207 .hyp_backend_schedule = rump_user_schedule, 208 .hyp_lwproc_switch = rump_lwproc_switch, 209 .hyp_lwproc_release = rump_lwproc_releaselwp, 210 .hyp_lwproc_rfork = rump_hyp_rfork, 211 .hyp_lwproc_newlwp = rump_lwproc_newlwp, 212 .hyp_lwproc_curlwp = rump_lwproc_curlwp, 213 .hyp_lwpexit = rump_hyp_lwpexit, 214 .hyp_syscall = rump_hyp_syscall, 215 .hyp_execnotify = rump_hyp_execnotify, 216 .hyp_getpid = spgetpid, 217 }; 218 219 int 220 rump_daemonize_begin(void) 221 { 222 223 if (rump_inited) 224 return EALREADY; 225 226 return rumpuser_daemonize_begin(); 227 } 228 229 int 230 rump_daemonize_done(int error) 231 { 232 233 return rumpuser_daemonize_done(error); 234 } 235 236 RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT) 237 { 238 __link_set_decl(rump_components, struct rump_component); 239 240 /* 241 * Trick compiler into generating references so that statically 242 * linked rump kernels are generated with the link set symbols. 243 */ 244 asm("" :: "r"(__start_link_set_rump_components)); 245 asm("" :: "r"(__stop_link_set_rump_components)); 246 } 247 248 int 249 rump_init(void) 250 { 251 char buf[256]; 252 struct timespec ts; 253 int64_t sec; 254 long nsec; 255 struct lwp *l, *initlwp; 256 int i, numcpu; 257 258 /* not reentrant */ 259 if (rump_inited) 260 return 0; 261 else if (rump_inited == -1) 262 panic("rump_init: host process restart required"); 263 else 264 rump_inited = 1; 265 266 /* initialize hypervisor */ 267 if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) { 268 rumpuser_dprintf("rumpuser init failed\n"); 269 return EINVAL; 270 } 271 272 /* init minimal lwp/cpu context */ 273 l = &lwp0; 274 l->l_lid = 1; 275 l->l_cpu = l->l_target_cpu = rump_cpu; 276 l->l_fd = &filedesc0; 277 278 /* lwp0 isn't created like other threads, so notify hypervisor here */ 279 rumpuser_curlwpop(RUMPUSER_LWP_CREATE, l); 280 rumpuser_curlwpop(RUMPUSER_LWP_SET, l); 281 282 /* retrieve env vars which affect the early stage of bootstrap */ 283 if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) { 284 rump_threads = *buf != '0'; 285 } 286 if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) { 287 if (*buf != '0') 288 boothowto = AB_VERBOSE; 289 } 290 291 if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0) 292 panic("mandatory hypervisor configuration (NCPU) missing"); 293 numcpu = strtoll(buf, NULL, 10); 294 if (numcpu < 1) { 295 panic("rump kernels are not lightweight enough for \"%d\" CPUs", 296 numcpu); 297 } 298 299 rump_thread_init(); 300 rump_cpus_bootstrap(&numcpu); 301 302 rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec); 303 boottime.tv_sec = sec; 304 boottime.tv_nsec = nsec; 305 306 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf)); 307 aprint_verbose("%s%s", copyright, version); 308 309 rump_intr_init(numcpu); 310 311 rump_tsleep_init(); 312 313 rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN); 314 ksyms_init(); 315 uvm_init(); 316 evcnt_init(); 317 318 kcpuset_sysinit(); 319 once_init(); 320 kernconfig_lock_init(); 321 prop_kern_init(); 322 323 kmem_init(); 324 kmeminit(); 325 326 uvm_ra_init(); 327 uao_init(); 328 329 mutex_obj_init(); 330 callout_startup(); 331 332 kprintf_init(); 333 pserialize_init(); 334 loginit(); 335 336 kauth_init(); 337 338 secmodel_init(); 339 340 rnd_init(); 341 342 /* 343 * Create the kernel cprng. Yes, it's currently stubbed out 344 * to arc4random() for RUMP, but this won't always be so. 345 */ 346 kern_cprng = cprng_strong_create("kernel", IPL_VM, 347 CPRNG_INIT_ANY|CPRNG_REKEY_ANY); 348 349 procinit(); 350 proc0_init(); 351 sysctl_init(); 352 uid_init(); 353 chgproccnt(0, 1); 354 355 l->l_proc = &proc0; 356 lwp_update_creds(l); 357 358 lwpinit_specificdata(); 359 lwp_initspecific(&lwp0); 360 361 rump_biglock_init(); 362 363 rump_scheduler_init(numcpu); 364 /* revert temporary context and schedule a semireal context */ 365 rumpuser_curlwpop(RUMPUSER_LWP_CLEAR, l); 366 initproc = &proc0; /* borrow proc0 before we get initproc started */ 367 rump_schedule(); 368 bootlwp = curlwp; 369 370 percpu_init(); 371 inittimecounter(); 372 ntp_init(); 373 374 ktrinit(); 375 376 ts = boottime; 377 tc_setclock(&ts); 378 379 /* we are mostly go. do per-cpu subsystem init */ 380 for (i = 0; i < numcpu; i++) { 381 struct cpu_info *ci = cpu_lookup(i); 382 383 /* attach non-bootstrap CPUs */ 384 if (i > 0) { 385 rump_cpu_attach(ci); 386 ncpu++; 387 } 388 389 callout_init_cpu(ci); 390 softint_init(ci); 391 xc_init_cpu(ci); 392 pool_cache_cpu_init(ci); 393 selsysinit(ci); 394 percpu_init_cpu(ci); 395 396 TAILQ_INIT(&ci->ci_data.cpu_ld_locks); 397 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock); 398 399 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i); 400 } 401 402 /* CPUs are up. allow kernel threads to run */ 403 rump_thread_allow(); 404 405 mksysctls(); 406 kqueue_init(); 407 iostat_init(); 408 fd_sys_init(); 409 module_init(); 410 devsw_init(); 411 pipe_init(); 412 resource_init(); 413 procinit_sysctl(); 414 415 /* start page baroness */ 416 if (rump_threads) { 417 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL, 418 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0) 419 panic("pagedaemon create failed"); 420 } else 421 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */ 422 423 /* process dso's */ 424 rumpuser_dl_bootstrap(add_linkedin_modules, 425 rump_kernelfsym_load, rump_component_load); 426 427 rump_component_addlocal(); 428 rump_component_init(RUMP_COMPONENT_KERN); 429 430 /* initialize factions, if present */ 431 rump_component_init(RUMP__FACTION_VFS); 432 /* pnbuf_cache is used even without vfs */ 433 if (rump_component_count(RUMP__FACTION_VFS) == 0) { 434 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl", 435 NULL, IPL_NONE, NULL, NULL, NULL); 436 } 437 rump_component_init(RUMP__FACTION_NET); 438 rump_component_init(RUMP__FACTION_DEV); 439 KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1 440 && rump_component_count(RUMP__FACTION_NET) <= 1 441 && rump_component_count(RUMP__FACTION_DEV) <= 1); 442 443 rump_component_init(RUMP_COMPONENT_KERN_VFS); 444 445 /* 446 * if we initialized the tty component above, the tyttymtx is 447 * now initialized. otherwise, we need to initialize it. 448 */ 449 if (!rump_ttycomponent) 450 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM); 451 452 cold = 0; 453 454 /* aieeeedondest */ 455 if (rump_threads) { 456 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 457 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 458 panic("aiodoned"); 459 } 460 461 sysctl_finalize(); 462 463 module_init_class(MODULE_CLASS_ANY); 464 465 if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME, 466 hostname, MAXHOSTNAMELEN) != 0) { 467 panic("mandatory hypervisor configuration (HOSTNAME) missing"); 468 } 469 hostnamelen = strlen(hostname); 470 471 sigemptyset(&sigcantmask); 472 473 if (rump_threads) 474 vmem_rehash_start(); 475 476 /* 477 * Create init (proc 1), used to attach implicit threads in rump. 478 * (note: must be done after vfsinit to get cwdi) 479 */ 480 initlwp = rump__lwproc_alloclwp(NULL); 481 mutex_enter(proc_lock); 482 initproc = proc_find_raw(1); 483 mutex_exit(proc_lock); 484 if (initproc == NULL) 485 panic("where in the world is initproc?"); 486 487 /* 488 * Adjust syscall vector in case factions were dlopen()'d 489 * before calling rump_init(). 490 * (modules will handle dynamic syscalls the usual way) 491 * 492 * Note: this will adjust the function vectors of 493 * syscalls which use a funcalias (getpid etc.), but 494 * it makes no difference. 495 */ 496 for (i = 0; i < SYS_NSYSENT; i++) { 497 void *sym; 498 499 if (rump_sysent[i].sy_flags & SYCALL_NOSYS || 500 *syscallnames[i] == '#' || 501 rump_sysent[i].sy_call == sys_nomodule) 502 continue; 503 504 /* 505 * deal with compat wrappers. makesyscalls.sh should 506 * generate the necessary info instead of this hack, 507 * though. ugly, fix it later. 508 */ 509 #define CPFX "compat_" 510 #define CPFXLEN (sizeof(CPFX)-1) 511 if (strncmp(syscallnames[i], CPFX, CPFXLEN) == 0) { 512 const char *p = syscallnames[i] + CPFXLEN; 513 size_t namelen; 514 515 /* skip version number */ 516 while (*p >= '0' && *p <= '9') 517 p++; 518 if (p == syscallnames[i] + CPFXLEN || *p != '_') 519 panic("invalid syscall name %s\n", 520 syscallnames[i]); 521 522 /* skip over the next underscore */ 523 p++; 524 namelen = p + (sizeof("rumpns_")-1) - syscallnames[i]; 525 526 strcpy(buf, "rumpns_"); 527 strcat(buf, syscallnames[i]); 528 /* XXX: no strncat in the kernel */ 529 strcpy(buf+namelen, "sys_"); 530 strcat(buf, p); 531 #undef CPFX 532 #undef CPFXLEN 533 } else { 534 sprintf(buf, "rumpns_sys_%s", syscallnames[i]); 535 } 536 if ((sym = rumpuser_dl_globalsym(buf)) != NULL 537 && sym != rump_sysent[i].sy_call) { 538 #if 0 539 rumpuser_dprintf("adjusting %s: %p (old %p)\n", 540 syscallnames[i], sym, rump_sysent[i].sy_call); 541 #endif 542 rump_sysent[i].sy_call = sym; 543 } 544 } 545 546 rump_component_init(RUMP_COMPONENT_POSTINIT); 547 548 /* component inits done */ 549 bootlwp = NULL; 550 551 /* open 0/1/2 for init */ 552 KASSERT(rump_lwproc_curlwp() == NULL); 553 rump_lwproc_switch(initlwp); 554 rump_consdev_init(); 555 rump_lwproc_switch(NULL); 556 557 /* release cpu */ 558 rump_unschedule(); 559 560 return 0; 561 } 562 /* historic compat */ 563 __strong_alias(rump__init,rump_init); 564 565 int 566 rump_init_server(const char *url) 567 { 568 569 return rumpuser_sp_init(url, ostype, osrelease, MACHINE); 570 } 571 572 void 573 cpu_reboot(int howto, char *bootstr) 574 { 575 int ruhow = 0; 576 void *finiarg; 577 578 printf("rump kernel halting...\n"); 579 580 if (!RUMP_LOCALPROC_P(curproc)) 581 finiarg = curproc->p_vmspace->vm_map.pmap; 582 else 583 finiarg = NULL; 584 585 /* dump means we really take the dive here */ 586 if ((howto & RB_DUMP) || panicstr) { 587 ruhow = RUMPUSER_PANIC; 588 goto out; 589 } 590 591 /* try to sync */ 592 if (!((howto & RB_NOSYNC) || panicstr)) { 593 if (rump_vfs_fini) 594 rump_vfs_fini(); 595 } 596 597 doshutdownhooks(); 598 599 /* your wish is my command */ 600 if (howto & RB_HALT) { 601 printf("rump kernel halted\n"); 602 rumpuser_sp_fini(finiarg); 603 for (;;) { 604 rumpuser_clock_sleep(RUMPUSER_CLOCK_RELWALL, 10, 0); 605 } 606 } 607 608 /* this function is __dead, we must exit */ 609 out: 610 printf("halted\n"); 611 rumpuser_sp_fini(finiarg); 612 rumpuser_exit(ruhow); 613 } 614 615 struct uio * 616 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw) 617 { 618 struct uio *uio; 619 enum uio_rw uiorw; 620 621 switch (rw) { 622 case RUMPUIO_READ: 623 uiorw = UIO_READ; 624 break; 625 case RUMPUIO_WRITE: 626 uiorw = UIO_WRITE; 627 break; 628 default: 629 panic("%s: invalid rw %d", __func__, rw); 630 } 631 632 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP); 633 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP); 634 635 uio->uio_iov->iov_base = buf; 636 uio->uio_iov->iov_len = bufsize; 637 638 uio->uio_iovcnt = 1; 639 uio->uio_offset = offset; 640 uio->uio_resid = bufsize; 641 uio->uio_rw = uiorw; 642 UIO_SETUP_SYSSPACE(uio); 643 644 return uio; 645 } 646 647 size_t 648 rump_uio_getresid(struct uio *uio) 649 { 650 651 return uio->uio_resid; 652 } 653 654 off_t 655 rump_uio_getoff(struct uio *uio) 656 { 657 658 return uio->uio_offset; 659 } 660 661 size_t 662 rump_uio_free(struct uio *uio) 663 { 664 size_t resid; 665 666 resid = uio->uio_resid; 667 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov)); 668 kmem_free(uio, sizeof(*uio)); 669 670 return resid; 671 } 672 673 kauth_cred_t 674 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups) 675 { 676 kauth_cred_t cred; 677 int rv; 678 679 cred = kauth_cred_alloc(); 680 kauth_cred_setuid(cred, uid); 681 kauth_cred_seteuid(cred, uid); 682 kauth_cred_setsvuid(cred, uid); 683 kauth_cred_setgid(cred, gid); 684 kauth_cred_setgid(cred, gid); 685 kauth_cred_setegid(cred, gid); 686 kauth_cred_setsvgid(cred, gid); 687 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE); 688 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */ 689 assert(rv == 0); 690 691 return cred; 692 } 693 694 void 695 rump_cred_put(kauth_cred_t cred) 696 { 697 698 kauth_cred_free(cred); 699 } 700 701 static int compcounter[RUMP_COMPONENT_MAX]; 702 static int compinited[RUMP_COMPONENT_MAX]; 703 704 /* 705 * Yea, this is O(n^2), but we're only looking at a handful of components. 706 * Components are always initialized from the thread that called rump_init(). 707 * Could also free these when done with them, but prolly not worth it. 708 */ 709 struct compstore { 710 const struct rump_component *cs_rc; 711 LIST_ENTRY(compstore) cs_entries; 712 }; 713 static LIST_HEAD(, compstore) cshead = LIST_HEAD_INITIALIZER(cshead); 714 715 /* 716 * add components which are visible from the current object. 717 */ 718 static void 719 rump_component_addlocal(void) 720 { 721 __link_set_decl(rump_components, struct rump_component); 722 struct rump_component *const *rc; 723 724 __link_set_foreach(rc, rump_components) { 725 rump_component_load(*rc); 726 } 727 } 728 729 static void 730 rump_component_load(const struct rump_component *rc) 731 { 732 struct compstore *cs; 733 734 KASSERT(curlwp == bootlwp); 735 736 LIST_FOREACH(cs, &cshead, cs_entries) { 737 if (rc == cs->cs_rc) 738 return; 739 } 740 741 cs = kmem_alloc(sizeof(*cs), KM_SLEEP); 742 cs->cs_rc = rc; 743 LIST_INSERT_HEAD(&cshead, cs, cs_entries); 744 KASSERT(rc->rc_type < RUMP_COMPONENT_MAX); 745 compcounter[rc->rc_type]++; 746 } 747 748 int 749 rump_component_count(enum rump_component_type type) 750 { 751 752 KASSERT(curlwp == bootlwp); 753 KASSERT(type < RUMP_COMPONENT_MAX); 754 return compcounter[type]; 755 } 756 757 void 758 rump_component_init(enum rump_component_type type) 759 { 760 struct compstore *cs; 761 const struct rump_component *rc; 762 763 KASSERT(curlwp == bootlwp); 764 KASSERT(!compinited[type]); 765 LIST_FOREACH(cs, &cshead, cs_entries) { 766 rc = cs->cs_rc; 767 if (rc->rc_type == type) 768 rc->rc_init(); 769 } 770 compinited[type] = 1; 771 } 772 773 /* 774 * Initialize a module which has already been loaded and linked 775 * with dlopen(). This is fundamentally the same as a builtin module. 776 */ 777 int 778 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo) 779 { 780 781 return module_builtin_add(mip, nmodinfo, true); 782 } 783 784 /* 785 * Finish module (flawless victory, fatality!). 786 */ 787 int 788 rump_module_fini(const struct modinfo *mi) 789 { 790 791 return module_builtin_remove(mi, true); 792 } 793 794 /* 795 * Add loaded and linked module to the builtin list. It will 796 * later be initialized with module_init_class(). 797 */ 798 799 static void 800 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo) 801 { 802 803 module_builtin_add(mip, nmodinfo, false); 804 } 805 806 int 807 rump_kernelfsym_load(void *symtab, uint64_t symsize, 808 char *strtab, uint64_t strsize) 809 { 810 static int inited = 0; 811 Elf64_Ehdr ehdr; 812 813 if (inited) 814 return EBUSY; 815 inited = 1; 816 817 /* 818 * Use 64bit header since it's bigger. Shouldn't make a 819 * difference, since we're passing in all zeroes anyway. 820 */ 821 memset(&ehdr, 0, sizeof(ehdr)); 822 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize); 823 824 return 0; 825 } 826 827 static int 828 rump_hyp_syscall(int num, void *arg, long *retval) 829 { 830 register_t regrv[2] = {0, 0}; 831 struct lwp *l; 832 struct sysent *callp; 833 int rv; 834 835 if (__predict_false(num >= SYS_NSYSENT)) 836 return ENOSYS; 837 838 callp = rump_sysent + num; 839 l = curlwp; 840 rv = sy_invoke(callp, l, (void *)arg, regrv, num); 841 retval[0] = regrv[0]; 842 retval[1] = regrv[1]; 843 844 return rv; 845 } 846 847 static int 848 rump_hyp_rfork(void *priv, int flags, const char *comm) 849 { 850 struct vmspace *newspace; 851 struct proc *p; 852 int error; 853 854 if ((error = rump_lwproc_rfork(flags)) != 0) 855 return error; 856 857 /* 858 * Since it's a proxy proc, adjust the vmspace. 859 * Refcount will eternally be 1. 860 */ 861 p = curproc; 862 newspace = kmem_zalloc(sizeof(*newspace), KM_SLEEP); 863 newspace->vm_refcnt = 1; 864 newspace->vm_map.pmap = priv; 865 KASSERT(p->p_vmspace == vmspace_kernel()); 866 p->p_vmspace = newspace; 867 if (comm) 868 strlcpy(p->p_comm, comm, sizeof(p->p_comm)); 869 870 return 0; 871 } 872 873 /* 874 * Order all lwps in a process to exit. does *not* wait for them to drain. 875 */ 876 static void 877 rump_hyp_lwpexit(void) 878 { 879 struct proc *p = curproc; 880 uint64_t where; 881 struct lwp *l; 882 883 mutex_enter(p->p_lock); 884 /* 885 * First pass: mark all lwps in the process with LW_RUMP_QEXIT 886 * so that they know they should exit. 887 */ 888 LIST_FOREACH(l, &p->p_lwps, l_sibling) { 889 if (l == curlwp) 890 continue; 891 l->l_flag |= LW_RUMP_QEXIT; 892 } 893 mutex_exit(p->p_lock); 894 895 /* 896 * Next, make sure everyone on all CPUs sees our status 897 * update. This keeps threads inside cv_wait() and makes 898 * sure we don't access a stale cv pointer later when 899 * we wake up the threads. 900 */ 901 902 where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 903 xc_wait(where); 904 905 /* 906 * Ok, all lwps are either: 907 * 1) not in the cv code 908 * 2) sleeping on l->l_private 909 * 3) sleeping on p->p_waitcv 910 * 911 * Either way, l_private is stable until we set PS_RUMP_LWPEXIT 912 * in p->p_sflag. 913 */ 914 915 mutex_enter(p->p_lock); 916 LIST_FOREACH(l, &p->p_lwps, l_sibling) { 917 if (l->l_private) 918 cv_broadcast(l->l_private); 919 } 920 p->p_sflag |= PS_RUMP_LWPEXIT; 921 cv_broadcast(&p->p_waitcv); 922 mutex_exit(p->p_lock); 923 } 924 925 /* 926 * Notify process that all threads have been drained and exec is complete. 927 */ 928 static void 929 rump_hyp_execnotify(const char *comm) 930 { 931 struct proc *p = curproc; 932 933 fd_closeexec(); 934 mutex_enter(p->p_lock); 935 KASSERT(p->p_nlwps == 1 && p->p_sflag & PS_RUMP_LWPEXIT); 936 p->p_sflag &= ~PS_RUMP_LWPEXIT; 937 mutex_exit(p->p_lock); 938 strlcpy(p->p_comm, comm, sizeof(p->p_comm)); 939 } 940 941 int 942 rump_boot_gethowto() 943 { 944 945 return boothowto; 946 } 947 948 void 949 rump_boot_sethowto(int howto) 950 { 951 952 boothowto = howto; 953 } 954 955 int 956 rump_getversion(void) 957 { 958 959 return __NetBSD_Version__; 960 } 961 962 /* 963 * Note: may be called unscheduled. Not fully safe since no locking 964 * of allevents (currently that's not even available). 965 */ 966 void 967 rump_printevcnts() 968 { 969 struct evcnt *ev; 970 971 TAILQ_FOREACH(ev, &allevents, ev_list) 972 rumpuser_dprintf("%s / %s: %" PRIu64 "\n", 973 ev->ev_group, ev->ev_name, ev->ev_count); 974 } 975 976 /* 977 * If you use this interface ... well ... all bets are off. 978 * The original purpose is for the p2k fs server library to be 979 * able to use the same pid/lid for VOPs as the host kernel. 980 */ 981 void 982 rump_allbetsareoff_setid(pid_t pid, int lid) 983 { 984 struct lwp *l = curlwp; 985 struct proc *p = l->l_proc; 986 987 l->l_lid = lid; 988 p->p_pid = pid; 989 } 990 991 #include <sys/pserialize.h> 992 993 static void 994 ipiemu(void *a1, void *a2) 995 { 996 997 xc__highpri_intr(NULL); 998 pserialize_switchpoint(); 999 } 1000 1001 void 1002 rump_xc_highpri(struct cpu_info *ci) 1003 { 1004 1005 if (ci) 1006 xc_unicast(0, ipiemu, NULL, NULL, ci); 1007 else 1008 xc_broadcast(0, ipiemu, NULL, NULL); 1009 } 1010 1011 int 1012 rump_syscall(int num, void *data, size_t dlen, register_t *retval) 1013 { 1014 struct proc *p; 1015 struct emul *e; 1016 struct sysent *callp; 1017 int rv; 1018 1019 rump_schedule(); 1020 p = curproc; 1021 e = p->p_emul; 1022 #ifndef __HAVE_MINIMAL_EMUL 1023 KASSERT(num > 0 && num < e->e_nsysent); 1024 #endif 1025 callp = e->e_sysent + num; 1026 1027 rv = sy_invoke(callp, curlwp, data, retval, num); 1028 rump_unschedule(); 1029 1030 return rv; 1031 } 1032
Indexes created Mon Sep 29 21:09:56 GMT 2025