sys/rump/librump/rumpkern/rump.c

   1 /*      $NetBSD: rump.c,v 1.147 2009/12/09 00:11:21 pooka Exp $ */
   2
   3 /*
   4  * Copyright (c) 2007 Antti Kantee.  All Rights Reserved.
   5  *
   6  * Development of this software was supported by Google Summer of Code.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
  18  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  19  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  20  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
  23  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.147 2009/12/09 00:11:21 pooka Exp $");
  32
  33 #include <sys/param.h>
  34 #include <sys/atomic.h>
  35 #include <sys/buf.h>
  36 #include <sys/callout.h>
  37 #include <sys/conf.h>
  38 #include <sys/cpu.h>
  39 #include <sys/device.h>
  40 #include <sys/evcnt.h>
  41 #include <sys/event.h>
  42 #include <sys/exec_elf.h>
  43 #include <sys/filedesc.h>
  44 #include <sys/iostat.h>
  45 #include <sys/kauth.h>
  46 #include <sys/kernel.h>
  47 #include <sys/kmem.h>
  48 #include <sys/kprintf.h>
  49 #include <sys/ksyms.h>
  50 #include <sys/msgbuf.h>
  51 #include <sys/module.h>
  52 #include <sys/once.h>
  53 #include <sys/percpu.h>
  54 #include <sys/pipe.h>
  55 #include <sys/queue.h>
  56 #include <sys/reboot.h>
  57 #include <sys/resourcevar.h>
  58 #include <sys/select.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/syscall.h>
  61 #include <sys/tty.h>
  62 #include <sys/uidinfo.h>
  63 #include <sys/vmem.h>
  64 #include <sys/xcall.h>
  65
  66 #include <rump/rumpuser.h>
  67
  68 #include <secmodel/suser/suser.h>
  69
  70 #include <prop/proplib.h>
  71
  72 #include <uvm/uvm_readahead.h>
  73
  74 #include "rump_private.h"
  75 #include "rump_net_private.h"
  76 #include "rump_vfs_private.h"
  77 #include "rump_dev_private.h"
  78
  79 struct proc proc0;
  80 struct session rump_session = {
  81         .s_count = 1,
  82         .s_flags = 0,
  83         .s_leader = &proc0,
  84         .s_login = "rumphobo",
  85         .s_sid = 0,
  86 };
  87 struct pgrp rump_pgrp = {
  88         .pg_members = LIST_HEAD_INITIALIZER(pg_members),
  89         .pg_session = &rump_session,
  90         .pg_jobc = 1,
  91 };
  92 struct pstats rump_stats;
  93 struct plimit rump_limits;
  94 struct filedesc rump_filedesc0;
  95 struct proclist allproc;
  96 char machine[] = "rump";
  97 static kauth_cred_t rump_susercred;
  98
  99 /* pretend the master rump proc is init */
 100 struct proc *initproc = &proc0;
 101
 102 struct rumpuser_mtx *rump_giantlock;
 103
 104 sigset_t sigcantmask;
 105
 106 struct device rump_rootdev = {
 107         .dv_class = DV_VIRTUAL
 108 };
 109
 110 #ifdef RUMP_WITHOUT_THREADS
 111 int rump_threads = 0;
 112 #else
 113 int rump_threads = 1;
 114 #endif
 115
 116 static void
 117 rump_aiodone_worker(struct work *wk, void *dummy)
 118 {
 119         struct buf *bp = (struct buf *)wk;
 120
 121         KASSERT(&bp->b_work == wk);
 122         bp->b_iodone(bp);
 123 }
 124
 125 static int rump_inited;
 126 static struct emul emul_rump = {
 127         .e_vm_default_addr = uvm_default_mapaddr,
 128 };
 129
 130 int rump__unavailable(void);
 131 int rump__unavailable() {return EOPNOTSUPP;}
 132 __weak_alias(rump_net_init,rump__unavailable);
 133 __weak_alias(rump_vfs_init,rump__unavailable);
 134 __weak_alias(rump_dev_init,rump__unavailable);
 135
 136 __weak_alias(rump_vfs_fini,rump__unavailable);
 137
 138 __weak_alias(biodone,rump__unavailable);
 139 __weak_alias(sopoll,rump__unavailable);
 140
 141 void rump__unavailable_vfs_panic(void);
 142 void rump__unavailable_vfs_panic() {panic("vfs component not available");}
 143 __weak_alias(usermount_common_policy,rump__unavailable_vfs_panic);
 144
 145 rump_proc_vfs_init_fn rump_proc_vfs_init;
 146 rump_proc_vfs_release_fn rump_proc_vfs_release;
 147
 148 /*
 149  * Stir up the stack a bit.  These are exported functions to help
 150  * convince the compiler that we don't want these routines completely
 151  * optimized out or inlined.  Is there an easier way to do this?
 152  */
 153 void nullfn(uint32_t *);
 154 void nullfn(uint32_t *arg){}
 155 void messthestack(void);
 156 void
 157 messthestack(void)
 158 {
 159         uint32_t mess[64];
 160         uint64_t d1, d2;
 161         int i, error;
 162
 163         for (i = 0; i < 64; i++) {
 164                 rumpuser_gettime(&d1, &d2, &error);
 165                 mess[i] = d2;
 166         }
 167         nullfn(mess);
 168 }
 169
 170 int
 171 rump__init(int rump_version)
 172 {
 173         char buf[256];
 174         struct proc *p;
 175         struct lwp *l;
 176         int i;
 177         int error;
 178
 179         /* not reentrant */
 180         if (rump_inited)
 181                 return 0;
 182         else if (rump_inited == -1)
 183                 panic("rump_init: host process restart required");
 184         else
 185                 rump_inited = 1;
 186
 187         /* Print some silly banners for spammy bootstrap. */
 188         if (boothowto & AB_VERBOSE) {
 189                 printf("%s%s", copyright, version);
 190         }
 191
 192         /*
 193          * Seed arc4random() with a "reasonable" amount of randomness.
 194          * Yes, this is a quick kludge which depends on the arc4random
 195          * implementation.
 196          */
 197         messthestack();
 198         arc4random();
 199
 200         if (rump_version != RUMP_VERSION) {
 201                 printf("rump version mismatch, %d vs. %d\n",
 202                     rump_version, RUMP_VERSION);
 203                 return EPROGMISMATCH;
 204         }
 205
 206         if (rumpuser_getenv("RUMP_THREADS", buf, sizeof(buf), &error) == 0) {
 207                 rump_threads = *buf != '0';
 208         }
 209         rumpuser_thrinit(rump_user_schedule, rump_user_unschedule,
 210             rump_threads);
 211         rump_intr_init();
 212
 213         /* init minimal lwp/cpu context */
 214         l = &lwp0;
 215         l->l_lid = 1;
 216         l->l_cpu = rump_cpu;
 217         rumpuser_set_curlwp(l);
 218
 219         mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_NONE);
 220         rumpuser_mutex_recursive_init(&rump_giantlock);
 221         ksyms_init();
 222         rumpvm_init();
 223         evcnt_init();
 224
 225         once_init();
 226         prop_kern_init();
 227
 228         pool_subsystem_init();
 229         kmem_init();
 230
 231         uvm_ra_init();
 232
 233         mutex_obj_init();
 234         callout_startup();
 235
 236         kprintf_init();
 237         loginit();
 238
 239         kauth_init();
 240         rump_susercred = rump_cred_create(0, 0, 0, NULL);
 241
 242         /* init proc0 and rest of lwp0 now that we can allocate memory */
 243         p = &proc0;
 244         p->p_stats = &rump_stats;
 245         p->p_limit = &rump_limits;
 246         p->p_pgrp = &rump_pgrp;
 247         p->p_pid = 0;
 248         p->p_fd = &rump_filedesc0;
 249         p->p_vmspace = &rump_vmspace;
 250         p->p_emul = &emul_rump;
 251         p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
 252         l->l_cred = rump_cred_suserget();
 253         l->l_proc = p;
 254         LIST_INIT(&allproc);
 255         LIST_INSERT_HEAD(&allproc, &proc0, p_list);
 256         proc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
 257
 258         rump_limits.pl_rlimit[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
 259         rump_limits.pl_rlimit[RLIMIT_NOFILE].rlim_cur = RLIM_INFINITY;
 260         rump_limits.pl_rlimit[RLIMIT_SBSIZE].rlim_cur = RLIM_INFINITY;
 261
 262         rump_scheduler_init();
 263         /* revert temporary context and schedule a real context */
 264         l->l_cpu = NULL;
 265         rumpuser_set_curlwp(NULL);
 266         rump_schedule();
 267
 268         /* we are mostly go.  do per-cpu subsystem init */
 269         for (i = 0; i < ncpu; i++) {
 270                 struct cpu_info *ci = cpu_lookup(i);
 271
 272                 callout_init_cpu(ci);
 273                 softint_init(ci);
 274                 xc_init_cpu(ci);
 275                 pool_cache_cpu_init(ci);
 276                 selsysinit(ci);
 277         }
 278
 279         sysctl_init();
 280         kqueue_init();
 281         iostat_init();
 282         uid_init();
 283         percpu_init();
 284         fd_sys_init();
 285         module_init();
 286         devsw_init();
 287         pipe_init();
 288
 289         /* these do nothing if not present */
 290         rump_vfs_init();
 291         rump_net_init();
 292         rump_dev_init();
 293         cold = 0;
 294
 295         /* aieeeedondest */
 296         if (rump_threads) {
 297                 if (workqueue_create(&uvm.aiodone_queue, "aiodoned",
 298                     rump_aiodone_worker, NULL, 0, 0, WQ_MPSAFE))
 299                         panic("aiodoned");
 300         }
 301
 302         sysctl_finalize();
 303
 304         rumpuser_dl_module_bootstrap(rump_module_init, rump_kernelfsym_load);
 305
 306         rumpuser_gethostname(hostname, MAXHOSTNAMELEN, &error);
 307         hostnamelen = strlen(hostname);
 308
 309         sigemptyset(&sigcantmask);
 310
 311         lwp0.l_fd = proc0.p_fd = fd_init(&rump_filedesc0);
 312
 313         if (rump_threads)
 314                 vmem_rehash_start();
 315
 316         rump_unschedule();
 317
 318         return 0;
 319 }
 320
 321 /* maybe support sys_reboot some day for remote shutdown */
 322 void
 323 rump_reboot(int howto)
 324 {
 325
 326         /* dump means we really take the dive here */
 327         if ((howto & RB_DUMP) || panicstr) {
 328                 rumpuser_exit(RUMPUSER_PANIC);
 329                 /*NOTREACHED*/
 330         }
 331
 332         /* try to sync */
 333         if (!((howto & RB_NOSYNC) || panicstr)) {
 334                 rump_vfs_fini();
 335         }
 336
 337         /* your wish is my command */
 338         if (howto & RB_HALT) {
 339                 for (;;) {
 340                         uint64_t sec = 5, nsec = 0;
 341                         int error;
 342
 343                         rumpuser_nanosleep(&sec, &nsec, &error);
 344                 }
 345         }
 346         rump_inited = -1;
 347 }
 348
 349 struct uio *
 350 rump_uio_setup(void *buf, size_t bufsize, off_t offset, enum rump_uiorw rw)
 351 {
 352         struct uio *uio;
 353         enum uio_rw uiorw;
 354
 355         switch (rw) {
 356         case RUMPUIO_READ:
 357                 uiorw = UIO_READ;
 358                 break;
 359         case RUMPUIO_WRITE:
 360                 uiorw = UIO_WRITE;
 361                 break;
 362         default:
 363                 panic("%s: invalid rw %d", __func__, rw);
 364         }
 365
 366         uio = kmem_alloc(sizeof(struct uio), KM_SLEEP);
 367         uio->uio_iov = kmem_alloc(sizeof(struct iovec), KM_SLEEP);
 368
 369         uio->uio_iov->iov_base = buf;
 370         uio->uio_iov->iov_len = bufsize;
 371
 372         uio->uio_iovcnt = 1;
 373         uio->uio_offset = offset;
 374         uio->uio_resid = bufsize;
 375         uio->uio_rw = uiorw;
 376         uio->uio_vmspace = UIO_VMSPACE_SYS;
 377
 378         return uio;
 379 }
 380
 381 size_t
 382 rump_uio_getresid(struct uio *uio)
 383 {
 384
 385         return uio->uio_resid;
 386 }
 387
 388 off_t
 389 rump_uio_getoff(struct uio *uio)
 390 {
 391
 392         return uio->uio_offset;
 393 }
 394
 395 size_t
 396 rump_uio_free(struct uio *uio)
 397 {
 398         size_t resid;
 399
 400         resid = uio->uio_resid;
 401         kmem_free(uio->uio_iov, sizeof(*uio->uio_iov));
 402         kmem_free(uio, sizeof(*uio));
 403
 404         return resid;
 405 }
 406
 407 static pid_t nextpid = 1;
 408 struct lwp *
 409 rump_newproc_switch()
 410 {
 411         struct lwp *l;
 412         pid_t mypid;
 413
 414         mypid = atomic_inc_uint_nv(&nextpid);
 415         if (__predict_false(mypid == 0))
 416                 mypid = atomic_inc_uint_nv(&nextpid);
 417
 418         l = rump_lwp_alloc(mypid, 0);
 419         rump_lwp_switch(l);
 420
 421         return l;
 422 }
 423
 424 struct lwp *
 425 rump_lwp_alloc_and_switch(pid_t pid, lwpid_t lid)
 426 {
 427         struct lwp *l;
 428
 429         l = rump_lwp_alloc(pid, lid);
 430         rump_lwp_switch(l);
 431
 432         return l;
 433 }
 434
 435 struct lwp *
 436 rump_lwp_alloc(pid_t pid, lwpid_t lid)
 437 {
 438         struct lwp *l;
 439         struct proc *p;
 440
 441         l = kmem_zalloc(sizeof(*l), KM_SLEEP);
 442         if (pid != 0) {
 443                 p = kmem_zalloc(sizeof(*p), KM_SLEEP);
 444                 if (rump_proc_vfs_init)
 445                         rump_proc_vfs_init(p);
 446                 p->p_stats = &rump_stats;
 447                 p->p_limit = &rump_limits;
 448                 p->p_pid = pid;
 449                 p->p_vmspace = &rump_vmspace;
 450                 p->p_emul = &emul_rump;
 451                 p->p_fd = fd_init(NULL);
 452                 p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
 453                 l->l_cred = rump_cred_suserget();
 454         } else {
 455                 p = &proc0;
 456                 l->l_cred = rump_susercred;
 457         }
 458
 459         l->l_proc = p;
 460         l->l_lid = lid;
 461         l->l_fd = p->p_fd;
 462         l->l_cpu = NULL;
 463
 464         return l;
 465 }
 466
 467 void
 468 rump_lwp_switch(struct lwp *newlwp)
 469 {
 470         struct lwp *l = curlwp;
 471
 472         rumpuser_set_curlwp(NULL);
 473         newlwp->l_cpu = l->l_cpu;
 474         newlwp->l_mutex = l->l_mutex;
 475         l->l_mutex = NULL;
 476         l->l_cpu = NULL;
 477         rumpuser_set_curlwp(newlwp);
 478         if (l->l_flag & LW_WEXIT)
 479                 rump_lwp_free(l);
 480 }
 481
 482 /* XXX: this has effect only on non-pid0 lwps */
 483 void
 484 rump_lwp_release(struct lwp *l)
 485 {
 486         struct proc *p;
 487
 488         p = l->l_proc;
 489         if (p->p_pid != 0) {
 490                 mutex_obj_free(p->p_lock);
 491                 fd_free();
 492                 if (rump_proc_vfs_release)
 493                         rump_proc_vfs_release(p);
 494                 rump_cred_put(l->l_cred);
 495                 kmem_free(p, sizeof(*p));
 496         }
 497         KASSERT((l->l_flag & LW_WEXIT) == 0);
 498         l->l_flag |= LW_WEXIT;
 499 }
 500
 501 void
 502 rump_lwp_free(struct lwp *l)
 503 {
 504
 505         KASSERT(l->l_flag & LW_WEXIT);
 506         KASSERT(l->l_mutex == NULL);
 507         kmem_free(l, sizeof(*l));
 508 }
 509
 510 struct lwp *
 511 rump_lwp_curlwp(void)
 512 {
 513         struct lwp *l = curlwp;
 514
 515         if (l->l_flag & LW_WEXIT)
 516                 return NULL;
 517         return l;
 518 }
 519
 520 /* rump private.  NEEDS WORK! */
 521 void
 522 rump_set_vmspace(struct vmspace *vm)
 523 {
 524         struct proc *p = curproc;
 525
 526         p->p_vmspace = vm;
 527 }
 528
 529 kauth_cred_t
 530 rump_cred_create(uid_t uid, gid_t gid, size_t ngroups, gid_t *groups)
 531 {
 532         kauth_cred_t cred;
 533         int rv;
 534
 535         cred = kauth_cred_alloc();
 536         kauth_cred_setuid(cred, uid);
 537         kauth_cred_seteuid(cred, uid);
 538         kauth_cred_setsvuid(cred, uid);
 539         kauth_cred_setgid(cred, gid);
 540         kauth_cred_setgid(cred, gid);
 541         kauth_cred_setegid(cred, gid);
 542         kauth_cred_setsvgid(cred, gid);
 543         rv = kauth_cred_setgroups(cred, groups, ngroups, 0, UIO_SYSSPACE);
 544         /* oh this is silly.  and by "this" I mean kauth_cred_setgroups() */
 545         assert(rv == 0);
 546
 547         return cred;
 548 }
 549
 550 void
 551 rump_cred_put(kauth_cred_t cred)
 552 {
 553
 554         kauth_cred_free(cred);
 555 }
 556
 557 kauth_cred_t
 558 rump_cred_suserget(void)
 559 {
 560
 561         kauth_cred_hold(rump_susercred);
 562         return rump_susercred;
 563 }
 564
 565 /*
 566  * Return the next system lwpid
 567  */
 568 lwpid_t
 569 rump_nextlid(void)
 570 {
 571         lwpid_t retid;
 572
 573         mutex_enter(proc0.p_lock);
 574         /*
 575          * Take next one, don't return 0
 576          * XXX: most likely we'll have collisions in case this
 577          * wraps around.
 578          */
 579         if (++proc0.p_nlwpid == 0)
 580                 ++proc0.p_nlwpid;
 581         retid = proc0.p_nlwpid;
 582         mutex_exit(proc0.p_lock);
 583
 584         return retid;
 585 }
 586
 587 #define ERROUT(err) do { rv = err; goto out; } while (/*CONSTCOND*/0)
 588 int
 589 rump_module_init(struct modinfo *mi, prop_dictionary_t props)
 590 {
 591         struct module *mod;
 592         int rv;
 593
 594         /* module_dummy */
 595         if (mi->mi_name == NULL)
 596                 return EINVAL;
 597
 598         mutex_enter(&module_lock);
 599         if (module_lookup(mi->mi_name))
 600                 ERROUT(EEXIST);
 601
 602         if (!module_compatible(mi->mi_version, __NetBSD_Version__))
 603                 ERROUT(EPROGMISMATCH);
 604
 605         rv = mi->mi_modcmd(MODULE_CMD_INIT, props);
 606         if (rv == 0) {
 607                 mod = kmem_zalloc(sizeof(*mod), KM_SLEEP);
 608                 mod->mod_info = mi;
 609                 module_enqueue(mod);
 610                 if (mi->mi_class == MODULE_CLASS_SECMODEL)
 611                         secmodel_register();
 612         }
 613
 614  out:
 615         mutex_exit(&module_lock);
 616         return rv;
 617 }
 618
 619 int
 620 rump_module_fini(struct modinfo *mi)
 621 {
 622         int rv;
 623
 624         rv = mi->mi_modcmd(MODULE_CMD_FINI, NULL);
 625         if (rv == 0 && mi->mi_class == MODULE_CLASS_SECMODEL)
 626                 secmodel_deregister();
 627
 628         return rv;
 629 }
 630
 631 int
 632 rump_kernelfsym_load(void *symtab, uint64_t symsize,
 633         char *strtab, uint64_t strsize)
 634 {
 635         static int inited = 0;
 636         Elf64_Ehdr ehdr;
 637
 638         if (inited)
 639                 return EBUSY;
 640         inited = 1;
 641
 642         /*
 643          * Use 64bit header since it's bigger.  Shouldn't make a
 644          * difference, since we're passing in all zeroes anyway.
 645          */
 646         memset(&ehdr, 0, sizeof(ehdr));
 647         ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
 648
 649         return 0;
 650 }
 651
 652 static int
 653 rump_sysproxy_local(int num, void *arg, uint8_t *data, size_t dlen,
 654         register_t *retval)
 655 {
 656         struct lwp *l;
 657         struct sysent *callp;
 658         int rv;
 659
 660         if (__predict_false(num >= SYS_NSYSENT))
 661                 return ENOSYS;
 662
 663         callp = rump_sysent + num;
 664         rump_schedule();
 665         l = curlwp;
 666         rv = callp->sy_call(l, (void *)data, retval);
 667         rump_unschedule();
 668
 669         return rv;
 670 }
 671
 672 int
 673 rump_boot_gethowto()
 674 {
 675
 676         return boothowto;
 677 }
 678
 679 void
 680 rump_boot_sethowto(int howto)
 681 {
 682
 683         boothowto = howto;
 684 }
 685
 686 rump_sysproxy_t rump_sysproxy = rump_sysproxy_local;
 687 void *rump_sysproxy_arg;
 688
 689 /*
 690  * This whole syscall-via-rpc is still taking form.  For example, it
 691  * may be necessary to set syscalls individually instead of lobbing
 692  * them all to the same place.  So don't think this interface is
 693  * set in stone.
 694  */
 695 int
 696 rump_sysproxy_set(rump_sysproxy_t proxy, void *arg)
 697 {
 698
 699         if (rump_sysproxy_arg)
 700                 return EBUSY;
 701
 702         rump_sysproxy_arg = arg;
 703         rump_sysproxy = proxy;
 704
 705         return 0;
 706 }
 707
 708 int
 709 rump_getversion(void)
 710 {
 711
 712         return __NetBSD_Version__;
 713 }