rump.c revision 1.274
1 /* $NetBSD: rump.c,v 1.274 2013/09/07 17:58:00 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 1.274 2013/09/07 17:58:00 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 /* retrieve env vars which affect the early stage of bootstrap */ 272 if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) { 273 rump_threads = *buf != '0'; 274 } 275 if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) { 276 if (*buf != '0') 277 boothowto = AB_VERBOSE; 278 } 279 280 if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0) 281 panic("mandatory hypervisor configuration (NCPU) missing"); 282 numcpu = strtoll(buf, NULL, 10); 283 if (numcpu < 1) { 284 panic("rump kernels are not lightweight enough for \"%d\" CPUs", 285 numcpu); 286 } 287 288 rump_thread_init(); 289 rump_cpus_bootstrap(&numcpu); 290 291 rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec); 292 boottime.tv_sec = sec; 293 boottime.tv_nsec = nsec; 294 295 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf)); 296 aprint_verbose("%s%s", copyright, version); 297 298 rump_intr_init(numcpu); 299 300 /* init minimal lwp/cpu context */ 301 l = &lwp0; 302 l->l_lid = 1; 303 l->l_cpu = l->l_target_cpu = rump_cpu; 304 l->l_fd = &filedesc0; 305 306 /* lwp0 isn't created like other threads, so notify hypervisor here */ 307 rumpuser_curlwpop(RUMPUSER_LWP_CREATE, l); 308 rumpuser_curlwpop(RUMPUSER_LWP_SET, l); 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 ts = boottime; 374 tc_setclock(&ts); 375 376 /* we are mostly go. do per-cpu subsystem init */ 377 for (i = 0; i < numcpu; i++) { 378 struct cpu_info *ci = cpu_lookup(i); 379 380 /* attach non-bootstrap CPUs */ 381 if (i > 0) { 382 rump_cpu_attach(ci); 383 ncpu++; 384 } 385 386 callout_init_cpu(ci); 387 softint_init(ci); 388 xc_init_cpu(ci); 389 pool_cache_cpu_init(ci); 390 selsysinit(ci); 391 percpu_init_cpu(ci); 392 393 TAILQ_INIT(&ci->ci_data.cpu_ld_locks); 394 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock); 395 396 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i); 397 } 398 399 /* CPUs are up. allow kernel threads to run */ 400 rump_thread_allow(); 401 402 mksysctls(); 403 kqueue_init(); 404 iostat_init(); 405 fd_sys_init(); 406 module_init(); 407 devsw_init(); 408 pipe_init(); 409 resource_init(); 410 procinit_sysctl(); 411 412 /* start page baroness */ 413 if (rump_threads) { 414 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL, 415 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0) 416 panic("pagedaemon create failed"); 417 } else 418 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */ 419 420 /* process dso's */ 421 rumpuser_dl_bootstrap(add_linkedin_modules, 422 rump_kernelfsym_load, rump_component_load); 423 424 rump_component_addlocal(); 425 rump_component_init(RUMP_COMPONENT_KERN); 426 427 /* initialize factions, if present */ 428 rump_component_init(RUMP__FACTION_VFS); 429 /* pnbuf_cache is used even without vfs */ 430 if (rump_component_count(RUMP__FACTION_VFS) == 0) { 431 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl", 432 NULL, IPL_NONE, NULL, NULL, NULL); 433 } 434 rump_component_init(RUMP__FACTION_NET); 435 rump_component_init(RUMP__FACTION_DEV); 436 KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1 437 && rump_component_count(RUMP__FACTION_NET) <= 1 438 && rump_component_count(RUMP__FACTION_DEV) <= 1); 439 440 rump_component_init(RUMP_COMPONENT_KERN_VFS); 441 442 /* 443 * if we initialized the tty component above, the tyttymtx is 444 * now initialized. otherwise, we need to initialize it. 445 */ 446 if (!rump_ttycomponent) 447 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM); 448 449 cold = 0; 450 451 /* aieeeedondest */ 452 if (rump_threads) { 453 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 454 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 455 panic("aiodoned"); 456 } 457 458 sysctl_finalize(); 459 460 module_init_class(MODULE_CLASS_ANY); 461 462 if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME, 463 hostname, MAXHOSTNAMELEN) != 0) { 464 panic("mandatory hypervisor configuration (HOSTNAME) missing"); 465 } 466 hostnamelen = strlen(hostname); 467 468 sigemptyset(&sigcantmask); 469 470 if (rump_threads) 471 vmem_rehash_start(); 472 473 /* 474 * Create init (proc 1), used to attach implicit threads in rump. 475 * (note: must be done after vfsinit to get cwdi) 476 */ 477 initlwp = rump__lwproc_alloclwp(NULL); 478 mutex_enter(proc_lock); 479 initproc = proc_find_raw(1); 480 mutex_exit(proc_lock); 481 if (initproc == NULL) 482 panic("where in the world is initproc?"); 483 484 /* 485 * Adjust syscall vector in case factions were dlopen()'d 486 * before calling rump_init(). 487 * (modules will handle dynamic syscalls the usual way) 488 * 489 * Note: this will adjust the function vectors of 490 * syscalls which use a funcalias (getpid etc.), but 491 * it makes no difference. 492 */ 493 for (i = 0; i < SYS_NSYSENT; i++) { 494 void *sym; 495 496 if (rump_sysent[i].sy_flags & SYCALL_NOSYS || 497 *syscallnames[i] == '#' || 498 rump_sysent[i].sy_call == sys_nomodule) 499 continue; 500 501 /* 502 * deal with compat wrappers. makesyscalls.sh should 503 * generate the necessary info instead of this hack, 504 * though. ugly, fix it later. 505 */ 506 #define CPFX "compat_" 507 #define CPFXLEN (sizeof(CPFX)-1) 508 if (strncmp(syscallnames[i], CPFX, CPFXLEN) == 0) { 509 const char *p = syscallnames[i] + CPFXLEN; 510 size_t namelen; 511 512 /* skip version number */ 513 while (*p >= '0' && *p <= '9') 514 p++; 515 if (p == syscallnames[i] + CPFXLEN || *p != '_') 516 panic("invalid syscall name %s\n", 517 syscallnames[i]); 518 519 /* skip over the next underscore */ 520 p++; 521 namelen = p + (sizeof("rumpns_")-1) - syscallnames[i]; 522 523 strcpy(buf, "rumpns_"); 524 strcat(buf, syscallnames[i]); 525 /* XXX: no strncat in the kernel */ 526 strcpy(buf+namelen, "sys_"); 527 strcat(buf, p); 528 #undef CPFX 529 #undef CPFXLEN 530 } else { 531 sprintf(buf, "rumpns_sys_%s", syscallnames[i]); 532 } 533 if ((sym = rumpuser_dl_globalsym(buf)) != NULL 534 && sym != rump_sysent[i].sy_call) { 535 #if 0 536 rumpuser_dprintf("adjusting %s: %p (old %p)\n", 537 syscallnames[i], sym, rump_sysent[i].sy_call); 538 #endif 539 rump_sysent[i].sy_call = sym; 540 } 541 } 542 543 rump_component_init(RUMP_COMPONENT_POSTINIT); 544 545 /* component inits done */ 546 bootlwp = NULL; 547 548 /* open 0/1/2 for init */ 549 rump_lwproc_switch(initlwp); 550 rump_consdev_init(); 551 552 /* release cpu */ 553 rump_unschedule(); 554 555 return 0; 556 } 557 /* historic compat */ 558 __strong_alias(rump__init,rump_init); 559 560 int 561 rump_init_server(const char *url) 562 { 563 564 return rumpuser_sp_init(url, ostype, osrelease, MACHINE); 565 } 566 567 void 568 cpu_reboot(int howto, char *bootstr) 569 { 570 int ruhow = 0; 571 void *finiarg; 572 573 printf("rump kernel halting...\n"); 574 575 if (!RUMP_LOCALPROC_P(curproc)) 576 finiarg = curproc->p_vmspace->vm_map.pmap; 577 else 578 finiarg = NULL; 579 580 /* dump means we really take the dive here */ 581 if ((howto & RB_DUMP) || panicstr) { 582 ruhow = RUMPUSER_PANIC; 583 goto out; 584 } 585 586 /* try to sync */ 587 if (!((howto & RB_NOSYNC) || panicstr)) { 588 if (rump_vfs_fini) 589 rump_vfs_fini(); 590 } 591 592 doshutdownhooks(); 593 594 /* your wish is my command */ 595 if (howto & RB_HALT) { 596 printf("rump kernel halted\n"); 597 rumpuser_sp_fini(finiarg); 598 for (;;) { 599 rumpuser_clock_sleep(RUMPUSER_CLOCK_RELWALL, 10, 0); 600 } 601 } 602 603 /* this function is __dead, we must exit */ 604 out: 605 printf("halted\n"); 606 rumpuser_sp_fini(finiarg); 607 rumpuser_exit(ruhow); 608 } 609 610 struct uio * 611 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw) 612 { 613 struct uio *uio; 614 enum uio_rw uiorw; 615 616 switch (rw) { 617 case RUMPUIO_READ: 618 uiorw = UIO_READ; 619 break; 620 case RUMPUIO_WRITE: 621 uiorw = UIO_WRITE; 622 break; 623 default: 624 panic("%s: invalid rw %d", __func__, rw); 625 } 626 627 uio = kmem_alloc(sizeof(struct uio), KM_SLEEP); 628 uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP); 629 630 uio->uio_iov->iov_base = buf; 631 uio->uio_iov->iov_len = bufsize; 632 633 uio->uio_iovcnt = 1; 634 uio->uio_offset = offset; 635 uio->uio_resid = bufsize; 636 uio->uio_rw = uiorw; 637 UIO_SETUP_SYSSPACE(uio); 638 639 return uio; 640 } 641 642 size_t 643 rump_uio_getresid(struct uio *uio) 644 { 645 646 return uio->uio_resid; 647 } 648 649 off_t 650 rump_uio_getoff(struct uio *uio) 651 { 652 653 return uio->uio_offset; 654 } 655 656 size_t 657 rump_uio_free(struct uio *uio) 658 { 659 size_t resid; 660 661 resid = uio->uio_resid; 662 kmem_free(uio->uio_iov, sizeof(*uio->uio_iov)); 663 kmem_free(uio, sizeof(*uio)); 664 665 return resid; 666 } 667 668 kauth_cred_t 669 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups) 670 { 671 kauth_cred_t cred; 672 int rv; 673 674 cred = kauth_cred_alloc(); 675 kauth_cred_setuid(cred, uid); 676 kauth_cred_seteuid(cred, uid); 677 kauth_cred_setsvuid(cred, uid); 678 kauth_cred_setgid(cred, gid); 679 kauth_cred_setgid(cred, gid); 680 kauth_cred_setegid(cred, gid); 681 kauth_cred_setsvgid(cred, gid); 682 rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE); 683 /* oh this is silly. and by "this" I mean kauth_cred_setgroups() */ 684 assert(rv == 0); 685 686 return cred; 687 } 688 689 void 690 rump_cred_put(kauth_cred_t cred) 691 { 692 693 kauth_cred_free(cred); 694 } 695 696 static int compcounter[RUMP_COMPONENT_MAX]; 697 static int compinited[RUMP_COMPONENT_MAX]; 698 699 /* 700 * Yea, this is O(n^2), but we're only looking at a handful of components. 701 * Components are always initialized from the thread that called rump_init(). 702 * Could also free these when done with them, but prolly not worth it. 703 */ 704 struct compstore { 705 const struct rump_component *cs_rc; 706 LIST_ENTRY(compstore) cs_entries; 707 }; 708 static LIST_HEAD(, compstore) cshead = LIST_HEAD_INITIALIZER(cshead); 709 710 /* 711 * add components which are visible from the current object. 712 */ 713 static void 714 rump_component_addlocal(void) 715 { 716 __link_set_decl(rump_components, struct rump_component); 717 struct rump_component *const *rc; 718 719 __link_set_foreach(rc, rump_components) { 720 rump_component_load(*rc); 721 } 722 } 723 724 static void 725 rump_component_load(const struct rump_component *rc) 726 { 727 struct compstore *cs; 728 729 KASSERT(curlwp == bootlwp); 730 731 LIST_FOREACH(cs, &cshead, cs_entries) { 732 if (rc == cs->cs_rc) 733 return; 734 } 735 736 cs = kmem_alloc(sizeof(*cs), KM_SLEEP); 737 cs->cs_rc = rc; 738 LIST_INSERT_HEAD(&cshead, cs, cs_entries); 739 KASSERT(rc->rc_type < RUMP_COMPONENT_MAX); 740 compcounter[rc->rc_type]++; 741 } 742 743 int 744 rump_component_count(enum rump_component_type type) 745 { 746 747 KASSERT(curlwp == bootlwp); 748 KASSERT(type < RUMP_COMPONENT_MAX); 749 return compcounter[type]; 750 } 751 752 void 753 rump_component_init(enum rump_component_type type) 754 { 755 struct compstore *cs; 756 const struct rump_component *rc; 757 758 KASSERT(curlwp == bootlwp); 759 KASSERT(!compinited[type]); 760 LIST_FOREACH(cs, &cshead, cs_entries) { 761 rc = cs->cs_rc; 762 if (rc->rc_type == type) 763 rc->rc_init(); 764 } 765 compinited[type] = 1; 766 } 767 768 /* 769 * Initialize a module which has already been loaded and linked 770 * with dlopen(). This is fundamentally the same as a builtin module. 771 */ 772 int 773 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo) 774 { 775 776 return module_builtin_add(mip, nmodinfo, true); 777 } 778 779 /* 780 * Finish module (flawless victory, fatality!). 781 */ 782 int 783 rump_module_fini(const struct modinfo *mi) 784 { 785 786 return module_builtin_remove(mi, true); 787 } 788 789 /* 790 * Add loaded and linked module to the builtin list. It will 791 * later be initialized with module_init_class(). 792 */ 793 794 static void 795 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo) 796 { 797 798 module_builtin_add(mip, nmodinfo, false); 799 } 800 801 int 802 rump_kernelfsym_load(void *symtab, uint64_t symsize, 803 char *strtab, uint64_t strsize) 804 { 805 static int inited = 0; 806 Elf64_Ehdr ehdr; 807 808 if (inited) 809 return EBUSY; 810 inited = 1; 811 812 /* 813 * Use 64bit header since it's bigger. Shouldn't make a 814 * difference, since we're passing in all zeroes anyway. 815 */ 816 memset(&ehdr, 0, sizeof(ehdr)); 817 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize); 818 819 return 0; 820 } 821 822 static int 823 rump_hyp_syscall(int num, void *arg, long *retval) 824 { 825 register_t regrv[2] = {0, 0}; 826 struct lwp *l; 827 struct sysent *callp; 828 int rv; 829 830 if (__predict_false(num >= SYS_NSYSENT)) 831 return ENOSYS; 832 833 callp = rump_sysent + num; 834 l = curlwp; 835 rv = sy_call(callp, l, (void *)arg, regrv); 836 retval[0] = regrv[0]; 837 retval[1] = regrv[1]; 838 839 return rv; 840 } 841 842 static int 843 rump_hyp_rfork(void *priv, int flags, const char *comm) 844 { 845 struct vmspace *newspace; 846 struct proc *p; 847 int error; 848 849 if ((error = rump_lwproc_rfork(flags)) != 0) 850 return error; 851 852 /* 853 * Since it's a proxy proc, adjust the vmspace. 854 * Refcount will eternally be 1. 855 */ 856 p = curproc; 857 newspace = kmem_zalloc(sizeof(*newspace), KM_SLEEP); 858 newspace->vm_refcnt = 1; 859 newspace->vm_map.pmap = priv; 860 KASSERT(p->p_vmspace == vmspace_kernel()); 861 p->p_vmspace = newspace; 862 if (comm) 863 strlcpy(p->p_comm, comm, sizeof(p->p_comm)); 864 865 return 0; 866 } 867 868 /* 869 * Order all lwps in a process to exit. does *not* wait for them to drain. 870 */ 871 static void 872 rump_hyp_lwpexit(void) 873 { 874 struct proc *p = curproc; 875 uint64_t where; 876 struct lwp *l; 877 878 mutex_enter(p->p_lock); 879 /* 880 * First pass: mark all lwps in the process with LW_RUMP_QEXIT 881 * so that they know they should exit. 882 */ 883 LIST_FOREACH(l, &p->p_lwps, l_sibling) { 884 if (l == curlwp) 885 continue; 886 l->l_flag |= LW_RUMP_QEXIT; 887 } 888 mutex_exit(p->p_lock); 889 890 /* 891 * Next, make sure everyone on all CPUs sees our status 892 * update. This keeps threads inside cv_wait() and makes 893 * sure we don't access a stale cv pointer later when 894 * we wake up the threads. 895 */ 896 897 where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL); 898 xc_wait(where); 899 900 /* 901 * Ok, all lwps are either: 902 * 1) not in the cv code 903 * 2) sleeping on l->l_private 904 * 3) sleeping on p->p_waitcv 905 * 906 * Either way, l_private is stable until we set PS_RUMP_LWPEXIT 907 * in p->p_sflag. 908 */ 909 910 mutex_enter(p->p_lock); 911 LIST_FOREACH(l, &p->p_lwps, l_sibling) { 912 if (l->l_private) 913 cv_broadcast(l->l_private); 914 } 915 p->p_sflag |= PS_RUMP_LWPEXIT; 916 cv_broadcast(&p->p_waitcv); 917 mutex_exit(p->p_lock); 918 } 919 920 /* 921 * Notify process that all threads have been drained and exec is complete. 922 */ 923 static void 924 rump_hyp_execnotify(const char *comm) 925 { 926 struct proc *p = curproc; 927 928 fd_closeexec(); 929 mutex_enter(p->p_lock); 930 KASSERT(p->p_nlwps == 1 && p->p_sflag & PS_RUMP_LWPEXIT); 931 p->p_sflag &= ~PS_RUMP_LWPEXIT; 932 mutex_exit(p->p_lock); 933 strlcpy(p->p_comm, comm, sizeof(p->p_comm)); 934 } 935 936 int 937 rump_boot_gethowto() 938 { 939 940 return boothowto; 941 } 942 943 void 944 rump_boot_sethowto(int howto) 945 { 946 947 boothowto = howto; 948 } 949 950 int 951 rump_getversion(void) 952 { 953 954 return __NetBSD_Version__; 955 } 956 957 /* 958 * Note: may be called unscheduled. Not fully safe since no locking 959 * of allevents (currently that's not even available). 960 */ 961 void 962 rump_printevcnts() 963 { 964 struct evcnt *ev; 965 966 TAILQ_FOREACH(ev, &allevents, ev_list) 967 rumpuser_dprintf("%s / %s: %" PRIu64 "\n", 968 ev->ev_group, ev->ev_name, ev->ev_count); 969 } 970 971 /* 972 * If you use this interface ... well ... all bets are off. 973 * The original purpose is for the p2k fs server library to be 974 * able to use the same pid/lid for VOPs as the host kernel. 975 */ 976 void 977 rump_allbetsareoff_setid(pid_t pid, int lid) 978 { 979 struct lwp *l = curlwp; 980 struct proc *p = l->l_proc; 981 982 l->l_lid = lid; 983 p->p_pid = pid; 984 } 985 986 #include <sys/pserialize.h> 987 988 static void 989 ipiemu(void *a1, void *a2) 990 { 991 992 xc__highpri_intr(NULL); 993 pserialize_switchpoint(); 994 } 995 996 void 997 rump_xc_highpri(struct cpu_info *ci) 998 { 999 1000 if (ci) 1001 xc_unicast(0, ipiemu, NULL, NULL, ci); 1002 else 1003 xc_broadcast(0, ipiemu, NULL, NULL); 1004 } 1005 1006 int 1007 rump_syscall(int num, void *data, size_t dlen, register_t *retval) 1008 { 1009 struct proc *p; 1010 struct emul *e; 1011 struct sysent *callp; 1012 int rv; 1013 1014 rump_schedule(); 1015 p = curproc; 1016 e = p->p_emul; 1017 #ifndef __HAVE_MINIMAL_EMUL 1018 KASSERT(num > 0 && num < e->e_nsysent); 1019 #endif 1020 callp = e->e_sysent + num; 1021 1022 rv = sy_call(callp, curlwp, data, retval); 1023 rump_unschedule(); 1024 1025 return rv; 1026 } 1027
Indexes created Mon Sep 29 21:09:56 GMT 2025