rumpuser.c revision 1.49
1 /* $NetBSD: rumpuser.c,v 1.49 2013/05/01 17:17:54 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 2007-2010 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 "rumpuser_port.h" 29 30 #if !defined(lint) 31 __RCSID("$NetBSD: rumpuser.c,v 1.49 2013/05/01 17:17:54 pooka Exp $"); 32 #endif /* !lint */ 33 34 #include <sys/ioctl.h> 35 #include <sys/mman.h> 36 #include <sys/uio.h> 37 #include <sys/stat.h> 38 #include <sys/time.h> 39 40 #ifdef __NetBSD__ 41 #include <sys/disk.h> 42 #include <sys/disklabel.h> 43 #include <sys/dkio.h> 44 #endif 45 46 #if defined(__NetBSD__) || defined(__FreeBSD__) || defined(__DragonFly__) 47 #include <sys/sysctl.h> 48 #endif 49 50 #include <assert.h> 51 #include <errno.h> 52 #include <fcntl.h> 53 #include <netdb.h> 54 #include <signal.h> 55 #include <stdarg.h> 56 #include <stdint.h> 57 #include <stdio.h> 58 #include <stdlib.h> 59 #include <string.h> 60 #include <time.h> 61 #include <unistd.h> 62 63 #include <rump/rumpuser.h> 64 65 #include "rumpuser_int.h" 66 67 struct rumpuser_hyperup rumpuser__hyp; 68 69 int 70 rumpuser_init(int version, const struct rumpuser_hyperup *hyp) 71 { 72 73 if (version != RUMPUSER_VERSION) { 74 fprintf(stderr, "rumpuser mismatch, kern: %d, hypervisor %d\n", 75 version, RUMPUSER_VERSION); 76 return 1; 77 } 78 79 #ifdef RUMPUSER_USE_DEVRANDOM 80 uint32_t rv; 81 int fd; 82 83 if ((fd = open("/dev/urandom", O_RDONLY)) == -1) { 84 srandom(time(NULL)); 85 } else { 86 if (read(fd, &rv, sizeof(rv)) != sizeof(rv)) 87 srandom(time(NULL)); 88 else 89 srandom(rv); 90 close(fd); 91 } 92 #endif 93 94 rumpuser__thrinit(); 95 rumpuser__hyp = *hyp; 96 97 return 0; 98 } 99 100 int 101 rumpuser_getfileinfo(const char *path, uint64_t *sizep, int *ftp) 102 { 103 struct stat sb; 104 uint64_t size = 0; 105 int needsdev = 0, rv = 0, ft = 0; 106 int fd = -1; 107 108 if (stat(path, &sb) == -1) { 109 rv = errno; 110 goto out; 111 } 112 113 switch (sb.st_mode & S_IFMT) { 114 case S_IFDIR: 115 ft = RUMPUSER_FT_DIR; 116 break; 117 case S_IFREG: 118 ft = RUMPUSER_FT_REG; 119 break; 120 case S_IFBLK: 121 ft = RUMPUSER_FT_BLK; 122 needsdev = 1; 123 break; 124 case S_IFCHR: 125 ft = RUMPUSER_FT_CHR; 126 needsdev = 1; 127 break; 128 default: 129 ft = RUMPUSER_FT_OTHER; 130 break; 131 } 132 133 if (!needsdev) { 134 size = sb.st_size; 135 } else if (sizep) { 136 /* 137 * Welcome to the jungle. Of course querying the kernel 138 * for a device partition size is supposed to be far from 139 * trivial. On NetBSD we use ioctl. On $other platform 140 * we have a problem. We try "the lseek trick" and just 141 * fail if that fails. Platform specific code can later 142 * be written here if appropriate. 143 * 144 * On NetBSD we hope and pray that for block devices nobody 145 * else is holding them open, because otherwise the kernel 146 * will not permit us to open it. Thankfully, this is 147 * usually called only in bootstrap and then we can 148 * forget about it. 149 */ 150 #ifndef __NetBSD__ 151 off_t off; 152 153 fd = open(path, O_RDONLY); 154 if (fd == -1) { 155 rv = errno; 156 goto out; 157 } 158 159 off = lseek(fd, 0, SEEK_END); 160 if (off != 0) { 161 size = off; 162 goto out; 163 } 164 fprintf(stderr, "error: device size query not implemented on " 165 "this platform\n"); 166 rv = EOPNOTSUPP; 167 goto out; 168 #else 169 struct disklabel lab; 170 struct partition *parta; 171 struct dkwedge_info dkw; 172 173 fd = open(path, O_RDONLY); 174 if (fd == -1) { 175 rv = errno; 176 goto out; 177 } 178 179 if (ioctl(fd, DIOCGDINFO, &lab) == 0) { 180 parta = &lab.d_partitions[DISKPART(sb.st_rdev)]; 181 size = (uint64_t)lab.d_secsize * parta->p_size; 182 goto out; 183 } 184 185 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) == 0) { 186 /* 187 * XXX: should use DIOCGDISKINFO to query 188 * sector size, but that requires proplib, 189 * so just don't bother for now. it's nice 190 * that something as difficult as figuring out 191 * a partition's size has been made so easy. 192 */ 193 size = dkw.dkw_size << DEV_BSHIFT; 194 goto out; 195 } 196 197 rv = errno; 198 #endif /* __NetBSD__ */ 199 } 200 201 out: 202 if (rv == 0 && sizep) 203 *sizep = size; 204 if (rv == 0 && ftp) 205 *ftp = ft; 206 if (fd != -1) 207 close(fd); 208 209 ET(rv); 210 } 211 212 int 213 rumpuser_malloc(size_t howmuch, int alignment, void **memp) 214 { 215 void *mem; 216 int rv; 217 218 if (alignment == 0) 219 alignment = sizeof(void *); 220 221 rv = posix_memalign(&mem, (size_t)alignment, howmuch); 222 if (__predict_false(rv != 0)) { 223 if (rv == EINVAL) { 224 printf("rumpuser_malloc: invalid alignment %d\n", 225 alignment); 226 abort(); 227 } 228 } 229 230 *memp = mem; 231 ET(rv); 232 } 233 234 /*ARGSUSED1*/ 235 void 236 rumpuser_free(void *ptr, size_t size) 237 { 238 239 free(ptr); 240 } 241 242 int 243 rumpuser_anonmmap(void *prefaddr, size_t size, int alignbit, 244 int exec, void **memp) 245 { 246 void *mem; 247 int prot, rv; 248 249 #ifndef MAP_ALIGNED 250 #define MAP_ALIGNED(a) 0 251 if (alignbit) 252 fprintf(stderr, "rumpuser_anonmmap: warning, requested " 253 "alignment not supported by hypervisor\n"); 254 #endif 255 256 prot = PROT_READ|PROT_WRITE; 257 if (exec) 258 prot |= PROT_EXEC; 259 mem = mmap(prefaddr, size, prot, 260 MAP_PRIVATE | MAP_ANON | MAP_ALIGNED(alignbit), -1, 0); 261 if (mem == MAP_FAILED) { 262 rv = errno; 263 } else { 264 *memp = mem; 265 rv = 0; 266 } 267 268 ET(rv); 269 } 270 271 void 272 rumpuser_unmap(void *addr, size_t len) 273 { 274 275 munmap(addr, len); 276 } 277 278 int 279 rumpuser_open(const char *path, int ruflags, int *fdp) 280 { 281 int fd, flags, rv; 282 283 switch (ruflags & RUMPUSER_OPEN_ACCMODE) { 284 case RUMPUSER_OPEN_RDONLY: 285 flags = O_RDONLY; 286 break; 287 case RUMPUSER_OPEN_WRONLY: 288 flags = O_WRONLY; 289 break; 290 case RUMPUSER_OPEN_RDWR: 291 flags = O_RDWR; 292 break; 293 default: 294 rv = EINVAL; 295 goto out; 296 } 297 298 #define TESTSET(_ru_, _h_) if (ruflags & _ru_) flags |= _h_; 299 TESTSET(RUMPUSER_OPEN_CREATE, O_CREAT); 300 TESTSET(RUMPUSER_OPEN_EXCL, O_EXCL); 301 #undef TESTSET 302 303 KLOCK_WRAP(fd = open(path, flags, 0644)); 304 if (fd == -1) { 305 rv = errno; 306 } else { 307 *fdp = fd; 308 rv = 0; 309 } 310 311 out: 312 ET(rv); 313 } 314 315 int 316 rumpuser_close(int fd) 317 { 318 int nlocks; 319 320 rumpkern_unsched(&nlocks, NULL); 321 fsync(fd); 322 close(fd); 323 rumpkern_sched(nlocks, NULL); 324 325 ET(0); 326 } 327 328 /* 329 * Assume "struct rumpuser_iovec" and "struct iovec" are the same. 330 * If you encounter POSIX platforms where they aren't, add some 331 * translation for iovlen > 1. 332 */ 333 int 334 rumpuser_iovread(int fd, struct rumpuser_iovec *ruiov, size_t iovlen, 335 off_t off, size_t *retp) 336 { 337 struct iovec *iov = (struct iovec *)ruiov; 338 ssize_t nn; 339 int rv; 340 341 if (off == RUMPUSER_IOV_NOSEEK) { 342 KLOCK_WRAP(nn = readv(fd, iov, iovlen)); 343 } else { 344 int nlocks; 345 346 rumpkern_unsched(&nlocks, NULL); 347 if (lseek(fd, off, SEEK_SET) == off) { 348 nn = readv(fd, iov, iovlen); 349 } else { 350 nn = -1; 351 } 352 rumpkern_sched(nlocks, NULL); 353 } 354 355 if (nn == -1) { 356 rv = errno; 357 } else { 358 *retp = (size_t)nn; 359 rv = 0; 360 } 361 362 ET(rv); 363 } 364 365 int 366 rumpuser_iovwrite(int fd, const struct rumpuser_iovec *ruiov, size_t iovlen, 367 off_t off, size_t *retp) 368 { 369 const struct iovec *iov = (const struct iovec *)ruiov; 370 ssize_t nn; 371 int rv; 372 373 if (off == RUMPUSER_IOV_NOSEEK) { 374 KLOCK_WRAP(nn = writev(fd, iov, iovlen)); 375 } else { 376 int nlocks; 377 378 rumpkern_unsched(&nlocks, NULL); 379 if (lseek(fd, off, SEEK_SET) == off) { 380 nn = writev(fd, iov, iovlen); 381 } else { 382 nn = -1; 383 } 384 rumpkern_sched(nlocks, NULL); 385 } 386 387 if (nn == -1) { 388 rv = errno; 389 } else { 390 *retp = (size_t)nn; 391 rv = 0; 392 } 393 394 ET(rv); 395 } 396 397 int 398 rumpuser_clock_gettime(enum rumpclock rclk, uint64_t *sec, uint64_t *nsec) 399 { 400 struct timespec ts; 401 clockid_t clk; 402 int rv; 403 404 switch (rclk) { 405 case RUMPUSER_CLOCK_RELWALL: 406 clk = CLOCK_REALTIME; 407 break; 408 case RUMPUSER_CLOCK_ABSMONO: 409 #ifdef HAVE_CLOCK_NANOSLEEP 410 clk = CLOCK_MONOTONIC; 411 #else 412 clk = CLOCK_REALTIME; 413 #endif 414 break; 415 default: 416 abort(); 417 } 418 419 if (clock_gettime(clk, &ts) == -1) { 420 rv = errno; 421 } else { 422 *sec = ts.tv_sec; 423 *nsec = ts.tv_nsec; 424 rv = 0; 425 } 426 427 ET(rv); 428 } 429 430 int 431 rumpuser_clock_sleep(enum rumpclock clk, uint64_t sec, uint64_t nsec) 432 { 433 struct timespec rqt, rmt; 434 int nlocks; 435 int rv; 436 437 rumpkern_unsched(&nlocks, NULL); 438 439 /*LINTED*/ 440 rqt.tv_sec = sec; 441 /*LINTED*/ 442 rqt.tv_nsec = nsec; 443 444 switch (clk) { 445 case RUMPUSER_CLOCK_RELWALL: 446 do { 447 rv = nanosleep(&rqt, &rmt); 448 rqt = rmt; 449 } while (rv == -1 && errno == EINTR); 450 if (rv == -1) { 451 rv = errno; 452 } 453 break; 454 case RUMPUSER_CLOCK_ABSMONO: 455 do { 456 #ifdef HAVE_CLOCK_NANOSLEEP 457 rv = clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME, 458 &rqt, NULL); 459 #else 460 /* le/la/der/die/das sigh. timevalspec tailspin */ 461 struct timespec ts, tsr; 462 clock_gettime(CLOCK_REALTIME, &ts); 463 if (ts.tv_sec == rqt.tv_sec ? 464 ts.tv_nsec > rqt.tv_nsec : ts.tv_sec > rqt.tv_sec) { 465 rv = 0; 466 } else { 467 tsr.tv_sec = rqt.tv_sec - ts.tv_sec; 468 tsr.tv_nsec = rqt.tv_nsec - ts.tv_nsec; 469 if (tsr.tv_nsec < 0) { 470 tsr.tv_sec--; 471 tsr.tv_nsec += 1000*1000*1000; 472 } 473 rv = nanosleep(&tsr, NULL); 474 } 475 #endif 476 } while (rv == -1 && errno == EINTR); 477 if (rv == -1) { 478 rv = errno; 479 } 480 break; 481 default: 482 abort(); 483 } 484 485 rumpkern_sched(nlocks, NULL); 486 487 ET(rv); 488 } 489 490 static int 491 gethostncpu(void) 492 { 493 int ncpu = 1; 494 495 #if defined(__NetBSD__) || defined(__FreeBSD__) || defined(__DragonFly__) 496 size_t sz = sizeof(ncpu); 497 498 sysctlbyname("hw.ncpu", &ncpu, &sz, NULL, 0); 499 #elif defined(__linux__) || defined(__CYGWIN__) 500 FILE *fp; 501 char *line = NULL; 502 size_t n = 0; 503 504 /* If anyone knows a better way, I'm all ears */ 505 if ((fp = fopen("/proc/cpuinfo", "r")) != NULL) { 506 ncpu = 0; 507 while (getline(&line, &n, fp) != -1) { 508 if (strncmp(line, 509 "processor", sizeof("processor")-1) == 0) 510 ncpu++; 511 } 512 if (ncpu == 0) 513 ncpu = 1; 514 free(line); 515 fclose(fp); 516 } 517 #elif __sun__ 518 /* XXX: this is just a rough estimate ... */ 519 ncpu = sysconf(_SC_NPROCESSORS_ONLN); 520 #endif 521 522 return ncpu; 523 } 524 525 int 526 rumpuser_getparam(const char *name, void *buf, size_t blen) 527 { 528 int rv; 529 530 if (strcmp(name, RUMPUSER_PARAM_NCPU) == 0) { 531 int ncpu; 532 533 if (getenv_r("RUMP_NCPU", buf, blen) == -1) { 534 ncpu = gethostncpu(); 535 snprintf(buf, blen, "%d", ncpu); 536 } 537 rv = 0; 538 } else if (strcmp(name, RUMPUSER_PARAM_HOSTNAME) == 0) { 539 char tmp[MAXHOSTNAMELEN]; 540 541 if (gethostname(tmp, sizeof(tmp)) == -1) { 542 snprintf(buf, blen, "rump-%05d", (int)getpid()); 543 } else { 544 snprintf(buf, blen, "rump-%05d.%s", 545 (int)getpid(), tmp); 546 } 547 rv = 0; 548 } else if (*name == '_') { 549 rv = EINVAL; 550 } else { 551 if (getenv_r(name, buf, blen) == -1) 552 rv = errno; 553 else 554 rv = 0; 555 } 556 557 ET(rv); 558 } 559 560 void 561 rumpuser_putchar(int c) 562 { 563 564 putchar(c); 565 } 566 567 void 568 rumpuser_exit(int rv) 569 { 570 571 if (rv == RUMPUSER_PANIC) 572 abort(); 573 else 574 exit(rv); 575 } 576 577 void 578 rumpuser_seterrno(int error) 579 { 580 581 errno = error; 582 } 583 584 /* 585 * This is meant for safe debugging prints from the kernel. 586 */ 587 void 588 rumpuser_dprintf(const char *format, ...) 589 { 590 va_list ap; 591 592 va_start(ap, format); 593 vfprintf(stderr, format, ap); 594 va_end(ap); 595 } 596 597 int 598 rumpuser_kill(int64_t pid, int sig) 599 { 600 int rv; 601 602 #ifdef __NetBSD__ 603 int error; 604 605 if (pid == RUMPUSER_PID_SELF) { 606 error = raise(sig); 607 } else { 608 error = kill((pid_t)pid, sig); 609 } 610 if (error == -1) 611 rv = errno; 612 else 613 rv = 0; 614 #else 615 /* XXXfixme: signal numbers may not match on non-NetBSD */ 616 rv = EOPNOTSUPP; 617 #endif 618 619 ET(rv); 620 } 621 622 int 623 rumpuser_getrandom(void *buf, size_t buflen, int flags, size_t *retp) 624 { 625 size_t origlen = buflen; 626 uint32_t *p = buf; 627 uint32_t tmp; 628 int chunk; 629 630 do { 631 chunk = buflen < 4 ? buflen : 4; /* portable MIN ... */ 632 tmp = RUMPUSER_RANDOM(); 633 memcpy(p, &tmp, chunk); 634 p++; 635 buflen -= chunk; 636 } while (chunk); 637 638 *retp = origlen; 639 ET(0); 640 } 641
Indexes created Wed Sep 24 05:09:52 GMT 2025