1 /* $NetBSD: rump.c,v 1.331 2018/01/09 04:55:43 msaitoh 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 1.331 2018/01/09 04:55:43 msaitoh 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/cprng.h> 74 #include <sys/rnd.h> 75 #include <sys/ktrace.h> 76 77 #include <rump-sys/kern.h> 78 #include <rump-sys/dev.h> 79 #include <rump-sys/net.h> 80 #include <rump-sys/vfs.h> 81 82 #include <rump/rumpuser.h> 83 84 #include <secmodel/suser/suser.h> 85 86 #include <prop/proplib.h> 87 88 #include <uvm/uvm_extern.h> 89 #include <uvm/uvm_readahead.h> 90 91 char machine[] = MACHINE; 92 char machine_arch[] = MACHINE_ARCH; 93 94 struct proc *initproc; 95 96 struct device rump_rootdev = { 97 .dv_class = DV_VIRTUAL 98 }; 99 100 #ifdef RUMP_WITHOUT_THREADS 101 int rump_threads = 0; 102 #else 103 int rump_threads = 1; 104 #endif 105 106 static void rump_component_addlocal(void); 107 static struct lwp *bootlwp; 108 109 /* 16k should be enough for std rump needs */ 110 static char rump_msgbuf[16*1024] __aligned(256); 111 112 bool rump_ttycomponent = false; 113 114 static void 115 rump_aiodone_worker(struct work *wk, void *dummy) 116 { 117 struct buf *bp = (struct buf *)wk; 118 119 KASSERT(&bp->b_work == wk); 120 bp->b_iodone(bp); 121 } 122 123 static int rump_inited; 124 125 void (*rump_vfs_drainbufs)(int) = (void *)nullop; 126 int (*rump_vfs_makeonedevnode)(dev_t, const char *, 127 devmajor_t, devminor_t) = (void *)nullop; 128 int (*rump_vfs_makedevnodes)(dev_t, const char *, char, 129 devmajor_t, devminor_t, int) = (void *)nullop; 130 int (*rump_vfs_makesymlink)(const char *, const char *) = (void *)nullop; 131 132 rump_proc_vfs_init_fn rump_proc_vfs_init = (void *)nullop; 133 rump_proc_vfs_release_fn rump_proc_vfs_release = (void *)nullop; 134 135 static void add_linkedin_modules(const struct modinfo *const *, size_t); 136 137 static pid_t rspo_wrap_getpid(void) { 138 return rump_sysproxy_hyp_getpid(); 139 } 140 static int rspo_wrap_syscall(int num, void *arg, long *retval) { 141 return rump_sysproxy_hyp_syscall(num, arg, retval); 142 } 143 static int rspo_wrap_rfork(void *priv, int flag, const char *comm) { 144 return rump_sysproxy_hyp_rfork(priv, flag, comm); 145 } 146 static void rspo_wrap_lwpexit(void) { 147 rump_sysproxy_hyp_lwpexit(); 148 } 149 static void rspo_wrap_execnotify(const char *comm) { 150 rump_sysproxy_hyp_execnotify(comm); 151 } 152 static const struct rumpuser_hyperup hyp = { 153 .hyp_schedule = rump_schedule, 154 .hyp_unschedule = rump_unschedule, 155 .hyp_backend_unschedule = rump_user_unschedule, 156 .hyp_backend_schedule = rump_user_schedule, 157 .hyp_lwproc_switch = rump_lwproc_switch, 158 .hyp_lwproc_release = rump_lwproc_releaselwp, 159 .hyp_lwproc_newlwp = rump_lwproc_newlwp, 160 .hyp_lwproc_curlwp = rump_lwproc_curlwp, 161 162 .hyp_getpid = rspo_wrap_getpid, 163 .hyp_syscall = rspo_wrap_syscall, 164 .hyp_lwproc_rfork = rspo_wrap_rfork, 165 .hyp_lwpexit = rspo_wrap_lwpexit, 166 .hyp_execnotify = rspo_wrap_execnotify, 167 }; 168 struct rump_sysproxy_ops rump_sysproxy_ops = { 169 .rspo_copyin = (void *)enxio, 170 .rspo_copyinstr = (void *)enxio, 171 .rspo_copyout = (void *)enxio, 172 .rspo_copyoutstr = (void *)enxio, 173 .rspo_anonmmap = (void *)enxio, 174 .rspo_raise = (void *)enxio, 175 .rspo_fini = (void *)enxio, 176 .rspo_hyp_getpid = (void *)enxio, 177 .rspo_hyp_syscall = (void *)enxio, 178 .rspo_hyp_rfork = (void *)enxio, 179 .rspo_hyp_lwpexit = (void *)enxio, 180 .rspo_hyp_execnotify = (void *)enxio, 181 }; 182 183 int 184 rump_daemonize_begin(void) 185 { 186 187 if (rump_inited) 188 return EALREADY; 189 190 return rumpuser_daemonize_begin(); 191 } 192 193 int 194 rump_daemonize_done(int error) 195 { 196 197 return rumpuser_daemonize_done(error); 198 } 199 200 #ifdef RUMP_USE_CTOR 201 202 /* sysctl bootstrap handling */ 203 struct sysctl_boot_chain sysctl_boot_chain \ 204 = LIST_HEAD_INITIALIZER(sysctl_boot_chain); 205 __link_set_add_text(sysctl_funcs,voidop); /* ensure linkset is non-empty */ 206 207 #else /* RUMP_USE_CTOR */ 208 209 RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT) 210 { 211 __link_set_decl(rump_components, struct rump_component); 212 213 /* 214 * Trick compiler into generating references so that statically 215 * linked rump kernels are generated with the link set symbols. 216 */ 217 asm("" :: "r"(__start_link_set_rump_components)); 218 asm("" :: "r"(__stop_link_set_rump_components)); 219 } 220 221 #endif /* RUMP_USE_CTOR */ 222 223 int 224 rump_init(void) 225 { 226 char buf[256]; 227 struct timespec ts; 228 int64_t sec; 229 long nsec; 230 struct lwp *l, *initlwp; 231 int i, numcpu; 232 233 /* not reentrant */ 234 if (rump_inited) 235 return 0; 236 else if (rump_inited == -1) 237 panic("rump_init: host process restart required"); 238 else 239 rump_inited = 1; 240 241 /* initialize hypervisor */ 242 if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) { 243 rumpuser_dprintf("rumpuser init failed\n"); 244 return EINVAL; 245 } 246 247 /* init minimal lwp/cpu context */ 248 rump_lwproc_init(); 249 l = &lwp0; 250 l->l_cpu = l->l_target_cpu = &rump_bootcpu; 251 rump_lwproc_curlwp_set(l); 252 253 /* retrieve env vars which affect the early stage of bootstrap */ 254 if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) { 255 rump_threads = *buf != '0'; 256 } 257 if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) { 258 if (*buf != '0') 259 boothowto = AB_VERBOSE; 260 } 261 262 if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0) 263 panic("mandatory hypervisor configuration (NCPU) missing"); 264 numcpu = strtoll(buf, NULL, 10); 265 if (numcpu < 1) { 266 panic("rump kernels are not lightweight enough for \"%d\" CPUs", 267 numcpu); 268 } 269 270 rump_thread_init(); 271 rump_cpus_bootstrap(&numcpu); 272 273 rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec); 274 boottime.tv_sec = sec; 275 boottime.tv_nsec = nsec; 276 277 initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf)); 278 aprint_verbose("%s%s", copyright, version); 279 280 rump_intr_init(numcpu); 281 282 rump_tsleep_init(); 283 284 rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN); 285 ksyms_init(); 286 uvm_init(); 287 evcnt_init(); 288 289 kcpuset_sysinit(); 290 once_init(); 291 kernconfig_lock_init(); 292 prop_kern_init(); 293 294 kmem_init(); 295 296 uvm_ra_init(); 297 uao_init(); 298 299 mutex_obj_init(); 300 rw_obj_init(); 301 callout_startup(); 302 303 kprintf_init(); 304 percpu_init(); 305 pserialize_init(); 306 307 kauth_init(); 308 309 secmodel_init(); 310 sysctl_init(); 311 /* 312 * The above call to sysctl_init() only initializes sysctl nodes 313 * from link sets. Initialize sysctls in case we used ctors. 314 */ 315 #ifdef RUMP_USE_CTOR 316 { 317 struct sysctl_setup_chain *ssc; 318 319 while ((ssc = LIST_FIRST(&sysctl_boot_chain)) != NULL) { 320 LIST_REMOVE(ssc, ssc_entries); 321 ssc->ssc_func(NULL); 322 } 323 } 324 #endif /* RUMP_USE_CTOR */ 325 326 rnd_init(); 327 cprng_init(); 328 kern_cprng = cprng_strong_create("kernel", IPL_VM, 329 CPRNG_INIT_ANY|CPRNG_REKEY_ANY); 330 331 rump_hyperentropy_init(); 332 333 procinit(); 334 proc0_init(); 335 uid_init(); 336 chgproccnt(0, 1); 337 338 l->l_proc = &proc0; 339 lwp_update_creds(l); 340 341 lwpinit_specificdata(); 342 lwp_initspecific(&lwp0); 343 344 loginit(); 345 346 rump_biglock_init(); 347 348 rump_scheduler_init(numcpu); 349 /* revert temporary context and schedule a semireal context */ 350 rump_lwproc_curlwp_clear(l); 351 initproc = &proc0; /* borrow proc0 before we get initproc started */ 352 rump_schedule(); 353 bootlwp = curlwp; 354 355 inittimecounter(); 356 ntp_init(); 357 358 #ifdef KTRACE 359 ktrinit(); 360 #endif 361 362 ts = boottime; 363 tc_setclock(&ts); 364 365 extern krwlock_t exec_lock; 366 rw_init(&exec_lock); 367 368 /* we are mostly go. do per-cpu subsystem init */ 369 for (i = 0; i < numcpu; i++) { 370 struct cpu_info *ci = cpu_lookup(i); 371 372 /* attach non-bootstrap CPUs */ 373 if (i > 0) { 374 rump_cpu_attach(ci); 375 ncpu++; 376 } 377 378 callout_init_cpu(ci); 379 softint_init(ci); 380 xc_init_cpu(ci); 381 pool_cache_cpu_init(ci); 382 selsysinit(ci); 383 percpu_init_cpu(ci); 384 385 TAILQ_INIT(&ci->ci_data.cpu_ld_locks); 386 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock); 387 388 aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i); 389 } 390 ncpuonline = ncpu; 391 392 /* Once all CPUs are detected, initialize the per-CPU cprng_fast. */ 393 cprng_fast_init(); 394 395 mp_online = true; 396 397 /* CPUs are up. allow kernel threads to run */ 398 rump_thread_allow(NULL); 399 400 rnd_init_softint(); 401 402 kqueue_init(); 403 iostat_init(); 404 fd_sys_init(); 405 module_init(); 406 devsw_init(); 407 pipe_init(); 408 resource_init(); 409 procinit_sysctl(); 410 time_init(); 411 time_init2(); 412 413 /* start page baroness */ 414 if (rump_threads) { 415 if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL, 416 uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0) 417 panic("pagedaemon create failed"); 418 } else 419 uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */ 420 421 /* process dso's */ 422 rumpuser_dl_bootstrap(add_linkedin_modules, 423 rump_kernelfsym_load, rump_component_load); 424 425 rump_component_addlocal(); 426 rump_component_init(RUMP_COMPONENT_KERN); 427 428 /* initialize factions, if present */ 429 rump_component_init(RUMP__FACTION_VFS); 430 /* pnbuf_cache is used even without vfs */ 431 if (rump_component_count(RUMP__FACTION_VFS) == 0) { 432 pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl", 433 NULL, IPL_NONE, NULL, NULL, NULL); 434 } 435 rump_component_init(RUMP__FACTION_NET); 436 rump_component_init(RUMP__FACTION_DEV); 437 KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1 438 && rump_component_count(RUMP__FACTION_NET) <= 1 439 && rump_component_count(RUMP__FACTION_DEV) <= 1); 440 441 rump_component_init(RUMP_COMPONENT_KERN_VFS); 442 443 /* 444 * if we initialized the tty component above, the tyttymtx is 445 * now initialized. otherwise, we need to initialize it. 446 */ 447 if (!rump_ttycomponent) 448 mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM); 449 450 cold = 0; 451 452 /* aieeeedondest */ 453 if (rump_threads) { 454 if (workqueue_create(&uvm.aiodone_queue, "aiodoned", 455 rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE)) 456 panic("aiodoned"); 457 } 458 459 sysctl_finalize(); 460 461 module_init_class(MODULE_CLASS_ANY); 462 463 if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME, 464 hostname, MAXHOSTNAMELEN) != 0) { 465 panic("mandatory hypervisor configuration (HOSTNAME) missing"); 466 } 467 hostnamelen = strlen(hostname); 468 469 sigemptyset(&sigcantmask); 470 471 if (rump_threads) 472 vmem_rehash_start(); 473 474 /* 475 * Create init (proc 1), used to attach implicit threads in rump. 476 * (note: must be done after vfsinit to get cwdi) 477 */ 478 initlwp = rump__lwproc_alloclwp(NULL); 479 mutex_enter(proc_lock); 480 initproc = proc_find_raw(1); 481 mutex_exit(proc_lock); 482 if (initproc == NULL) 483 panic("where in the world is initproc?"); 484 strlcpy(initproc->p_comm, "rumplocal", sizeof(initproc->p_comm)); 485 486 rump_component_init(RUMP_COMPONENT_POSTINIT); 487 488 /* load syscalls */ 489 rump_component_init(RUMP_COMPONENT_SYSCALL); 490 491 /* component inits done */ 492 bootlwp = NULL; 493 494 /* open 0/1/2 for init */ 495 KASSERT(rump_lwproc_curlwp() == NULL); 496 rump_lwproc_switch(initlwp); 497 rump_consdev_init(); 498 rump_lwproc_switch(NULL); 499 500 /* release cpu */ 501 rump_unschedule(); 502 503 return 0; 504 } 505 /* historic compat */ 506 __strong_alias(rump__init,rump_init); 507 508 static int compcounter[RUMP_COMPONENT_MAX]; 509 static int compinited[RUMP_COMPONENT_MAX]; 510 511 /* 512 * Yea, this is O(n^2), but we're only looking at a handful of components. 513 * Components are always initialized from the thread that called rump_init(). 514 */ 515 static LIST_HEAD(, rump_component) rchead = LIST_HEAD_INITIALIZER(rchead); 516 517 #ifdef RUMP_USE_CTOR 518 struct modinfo_boot_chain modinfo_boot_chain \ 519 = LIST_HEAD_INITIALIZER(modinfo_boot_chain); 520 521 static void 522 rump_component_addlocal(void) 523 { 524 struct modinfo_chain *mc; 525 526 while ((mc = LIST_FIRST(&modinfo_boot_chain)) != NULL) { 527 LIST_REMOVE(mc, mc_entries); 528 module_builtin_add(&mc->mc_info, 1, false); 529 } 530 } 531 532 #else /* RUMP_USE_CTOR */ 533 534 static void 535 rump_component_addlocal(void) 536 { 537 __link_set_decl(rump_components, struct rump_component); 538 struct rump_component *const *rc; 539 540 __link_set_foreach(rc, rump_components) { 541 rump_component_load(*rc); 542 } 543 } 544 #endif /* RUMP_USE_CTOR */ 545 546 void 547 rump_component_load(const struct rump_component *rc_const) 548 { 549 struct rump_component *rc, *rc_iter; 550 551 /* time for rump component loading and unloading has passed */ 552 if (!cold) 553 return; 554 555 /* 556 * XXX: this is ok since the "const" was removed from the 557 * definition of RUMP_COMPONENT(). 558 * 559 * However, to preserve the hypercall interface, the const 560 * remains here. This can be fixed in the next hypercall revision. 561 */ 562 rc = __UNCONST(rc_const); 563 564 KASSERT(!rump_inited || curlwp == bootlwp); 565 566 LIST_FOREACH(rc_iter, &rchead, rc_entries) { 567 if (rc_iter == rc) 568 return; 569 } 570 571 LIST_INSERT_HEAD(&rchead, rc, rc_entries); 572 KASSERT(rc->rc_type < RUMP_COMPONENT_MAX); 573 compcounter[rc->rc_type]++; 574 } 575 576 void 577 rump_component_unload(struct rump_component *rc) 578 { 579 580 /* 581 * Checking for cold is enough because rump_init() both 582 * flips it and handles component loading. 583 */ 584 if (!cold) 585 return; 586 587 LIST_REMOVE(rc, rc_entries); 588 } 589 590 int 591 rump_component_count(enum rump_component_type type) 592 { 593 594 KASSERT(curlwp == bootlwp); 595 KASSERT(type < RUMP_COMPONENT_MAX); 596 return compcounter[type]; 597 } 598 599 void 600 rump_component_init(enum rump_component_type type) 601 { 602 const struct rump_component *rc, *rc_safe; 603 604 KASSERT(curlwp == bootlwp); 605 KASSERT(!compinited[type]); 606 LIST_FOREACH_SAFE(rc, &rchead, rc_entries, rc_safe) { 607 if (rc->rc_type == type) { 608 rc->rc_init(); 609 LIST_REMOVE(rc, rc_entries); 610 } 611 } 612 compinited[type] = 1; 613 } 614 615 /* 616 * Initialize a module which has already been loaded and linked 617 * with dlopen(). This is fundamentally the same as a builtin module. 618 * 619 * XXX: this interface does not really work in the RUMP_USE_CTOR case, 620 * but I'm not sure it's anything to cry about. In feeling blue, 621 * things could somehow be handled via modinfo_boot_chain. 622 */ 623 int 624 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo) 625 { 626 627 return module_builtin_add(mip, nmodinfo, true); 628 } 629 630 /* 631 * Finish module (flawless victory, fatality!). 632 */ 633 int 634 rump_module_fini(const struct modinfo *mi) 635 { 636 637 return module_builtin_remove(mi, true); 638 } 639 640 /* 641 * Add loaded and linked module to the builtin list. It will 642 * later be initialized with module_init_class(). 643 */ 644 645 static void 646 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo) 647 { 648 649 module_builtin_add(mip, nmodinfo, false); 650 } 651 652 int 653 rump_kernelfsym_load(void *symtab, uint64_t symsize, 654 char *strtab, uint64_t strsize) 655 { 656 static int inited = 0; 657 Elf64_Ehdr ehdr; 658 659 if (inited) 660 return EBUSY; 661 inited = 1; 662 663 /* 664 * Use 64bit header since it's bigger. Shouldn't make a 665 * difference, since we're passing in all zeroes anyway. 666 */ 667 memset(&ehdr, 0, sizeof(ehdr)); 668 ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize); 669 670 return 0; 671 } 672 673 int 674 rump_boot_gethowto() 675 { 676 677 return boothowto; 678 } 679 680 void 681 rump_boot_sethowto(int howto) 682 { 683 684 boothowto = howto; 685 } 686 687 int 688 rump_getversion(void) 689 { 690 691 return __NetBSD_Version__; 692 } 693 /* compat */ 694 __strong_alias(rump_pub_getversion,rump_getversion); 695 696 /* 697 * Note: may be called unscheduled. Not fully safe since no locking 698 * of allevents (currently that's not even available). 699 */ 700 void 701 rump_printevcnts() 702 { 703 struct evcnt *ev; 704 705 TAILQ_FOREACH(ev, &allevents, ev_list) 706 rumpuser_dprintf("%s / %s: %" PRIu64 "\n", 707 ev->ev_group, ev->ev_name, ev->ev_count); 708 } 709 710 /* 711 * If you use this interface ... well ... all bets are off. 712 * The original purpose is for the p2k fs server library to be 713 * able to use the same pid/lid for VOPs as the host kernel. 714 */ 715 void 716 rump_allbetsareoff_setid(pid_t pid, int lid) 717 { 718 struct lwp *l = curlwp; 719 struct proc *p = l->l_proc; 720 721 l->l_lid = lid; 722 p->p_pid = pid; 723 } 724 725 #include <sys/pserialize.h> 726 727 static void 728 ipiemu(void *a1, void *a2) 729 { 730 731 xc__highpri_intr(NULL); 732 pserialize_switchpoint(); 733 } 734 735 void 736 rump_xc_highpri(struct cpu_info *ci) 737 { 738 739 if (ci) 740 xc_unicast(0, ipiemu, NULL, NULL, ci); 741 else 742 xc_broadcast(0, ipiemu, NULL, NULL); 743 } 744 745 int 746 rump_syscall(int num, void *data, size_t dlen, register_t *retval) 747 { 748 struct proc *p; 749 struct emul *e; 750 struct sysent *callp; 751 const int *etrans = NULL; 752 int rv; 753 754 rump_schedule(); 755 p = curproc; 756 e = p->p_emul; 757 #ifndef __HAVE_MINIMAL_EMUL 758 KASSERT(num > 0 && num < e->e_nsysent); 759 #endif 760 callp = e->e_sysent + num; 761 762 rv = sy_invoke(callp, curlwp, data, retval, num); 763 764 /* 765 * I hope that (!__HAVE_MINIMAL_EMUL || __HAVE_SYSCALL_INTERN) is 766 * an invariant ... 767 */ 768 #if !defined(__HAVE_MINIMAL_EMUL) 769 etrans = e->e_errno; 770 #elif defined(__HAVE_SYSCALL_INTERN) 771 etrans = p->p_emuldata; 772 #endif 773 774 if (etrans) { 775 rv = etrans[rv]; 776 /* 777 * XXX: small hack since Linux etrans vectors on some 778 * archs contain negative errnos, but rump_syscalls 779 * uses the -1 + errno ABI. Note that these 780 * negative values are always the result of translation, 781 * otherwise the above translation method would not 782 * work very well. 783 */ 784 if (rv < 0) 785 rv = -rv; 786 } 787 rump_unschedule(); 788 789 return rv; 790 } 791 792 void 793 rump_syscall_boot_establish(const struct rump_onesyscall *calls, size_t ncall) 794 { 795 struct sysent *callp; 796 size_t i; 797 798 for (i = 0; i < ncall; i++) { 799 callp = rump_sysent + calls[i].ros_num; 800 KASSERT(bootlwp != NULL 801 && callp->sy_call == (sy_call_t *)enosys); 802 callp->sy_call = calls[i].ros_handler; 803 } 804 } 805 806 struct rump_boot_etfs *ebstart; 807 void 808 rump_boot_etfs_register(struct rump_boot_etfs *eb) 809 { 810 811 /* 812 * Could use atomics, but, since caller would need to synchronize 813 * against calling rump_init() anyway, easier to just specify the 814 * interface as "caller serializes". This solve-by-specification 815 * approach avoids the grey area of using atomics before rump_init() 816 * runs. 817 */ 818 eb->_eb_next = ebstart; 819 eb->eb_status = -1; 820 ebstart = eb; 821 } 822