rumpuser.c revision 1.45
1 /* $NetBSD: rumpuser.c,v 1.45 2013/04/30 00:03:52 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 1.45 2013/04/30 00:03:52 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 return 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 return 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 return rv; 269 } 270 271 void 272 rumpuser_unmap(void *addr, size_t len) 273 { 274 int rv; 275 276 rv = munmap(addr, len); 277 assert(rv == 0); 278 } 279 280 int 281 rumpuser_open(const char *path, int ruflags, int *fdp) 282 { 283 int fd, flags, rv; 284 285 switch (ruflags & RUMPUSER_OPEN_ACCMODE) { 286 case RUMPUSER_OPEN_RDONLY: 287 flags = O_RDONLY; 288 break; 289 case RUMPUSER_OPEN_WRONLY: 290 flags = O_WRONLY; 291 break; 292 case RUMPUSER_OPEN_RDWR: 293 flags = O_RDWR; 294 break; 295 default: 296 rv = EINVAL; 297 goto out; 298 } 299 300 #define TESTSET(_ru_, _h_) if (ruflags & _ru_) flags |= _h_; 301 TESTSET(RUMPUSER_OPEN_CREATE, O_CREAT); 302 TESTSET(RUMPUSER_OPEN_EXCL, O_EXCL); 303 #undef TESTSET 304 305 KLOCK_WRAP(fd = open(path, flags, 0644)); 306 if (fd == -1) { 307 rv = errno; 308 } else { 309 *fdp = fd; 310 rv = 0; 311 } 312 313 out: 314 return rv; 315 } 316 317 int 318 rumpuser_close(int fd) 319 { 320 int nlocks; 321 322 rumpkern_unsched(&nlocks, NULL); 323 fsync(fd); 324 close(fd); 325 rumpkern_sched(nlocks, NULL); 326 327 return 0; 328 } 329 330 /* 331 * Assume "struct rumpuser_iovec" and "struct iovec" are the same. 332 * If you encounter POSIX platforms where they aren't, add some 333 * translation for iovlen > 1. 334 */ 335 int 336 rumpuser_iovread(int fd, struct rumpuser_iovec *ruiov, size_t iovlen, 337 off_t off, size_t *retp) 338 { 339 struct iovec *iov = (struct iovec *)ruiov; 340 ssize_t nn; 341 int rv; 342 343 if (off == RUMPUSER_IOV_NOSEEK) 344 KLOCK_WRAP(nn = readv(fd, iov, iovlen)); 345 else 346 KLOCK_WRAP(nn = preadv(fd, iov, iovlen, off)); 347 348 if (nn == -1) { 349 rv = errno; 350 } else { 351 *retp = (size_t)nn; 352 rv = 0; 353 } 354 355 return rv; 356 } 357 358 int 359 rumpuser_iovwrite(int fd, const struct rumpuser_iovec *ruiov, size_t iovlen, 360 off_t off, size_t *retp) 361 { 362 const struct iovec *iov = (const struct iovec *)ruiov; 363 ssize_t nn; 364 int rv; 365 366 if (off == RUMPUSER_IOV_NOSEEK) 367 KLOCK_WRAP(nn = writev(fd, iov, iovlen)); 368 else 369 KLOCK_WRAP(nn = pwritev(fd, iov, iovlen, off)); 370 371 if (nn == -1) { 372 rv = errno; 373 } else { 374 *retp = (size_t)nn; 375 rv = 0; 376 } 377 378 return rv; 379 } 380 381 int 382 rumpuser_clock_gettime(uint64_t *sec, uint64_t *nsec, enum rumpclock rclk) 383 { 384 struct timespec ts; 385 clockid_t clk; 386 int rv; 387 388 switch (rclk) { 389 case RUMPUSER_CLOCK_RELWALL: 390 clk = CLOCK_REALTIME; 391 break; 392 case RUMPUSER_CLOCK_ABSMONO: 393 #ifdef HAVE_CLOCK_NANOSLEEP 394 clk = CLOCK_MONOTONIC; 395 #else 396 clk = CLOCK_REALTIME; 397 #endif 398 break; 399 default: 400 abort(); 401 } 402 403 rv = clock_gettime(clk, &ts); 404 if (rv == -1) { 405 rv = errno; 406 } else { 407 *sec = ts.tv_sec; 408 *nsec = ts.tv_nsec; 409 rv = 0; 410 } 411 412 return rv; 413 } 414 415 int 416 rumpuser_clock_sleep(uint64_t sec, uint64_t nsec, enum rumpclock clk) 417 { 418 struct timespec rqt, rmt; 419 int nlocks; 420 int rv; 421 422 rumpkern_unsched(&nlocks, NULL); 423 424 /*LINTED*/ 425 rqt.tv_sec = sec; 426 /*LINTED*/ 427 rqt.tv_nsec = nsec; 428 429 switch (clk) { 430 case RUMPUSER_CLOCK_RELWALL: 431 do { 432 rv = nanosleep(&rqt, &rmt); 433 rqt = rmt; 434 } while (rv == -1 && errno == EINTR); 435 if (rv == -1) { 436 rv = errno; 437 } 438 break; 439 case RUMPUSER_CLOCK_ABSMONO: 440 do { 441 #ifdef HAVE_CLOCK_NANOSLEEP 442 rv = clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME, 443 &rqt, NULL); 444 #else 445 /* le/la/der/die/das sigh. timevalspec tailspin */ 446 struct timespec ts, tsr; 447 clock_gettime(CLOCK_REALTIME, &ts); 448 if (ts.tv_sec == rqt.tv_sec ? 449 ts.tv_nsec > rqt.tv_nsec : ts.tv_sec > rqt.tv_sec) { 450 rv = 0; 451 } else { 452 tsr.tv_sec = rqt.tv_sec - ts.tv_sec; 453 tsr.tv_nsec = rqt.tv_nsec - ts.tv_nsec; 454 if (tsr.tv_nsec < 0) { 455 tsr.tv_sec--; 456 tsr.tv_nsec += 1000*1000*1000; 457 } 458 rv = nanosleep(&tsr, NULL); 459 } 460 #endif 461 } while (rv == -1 && errno == EINTR); 462 if (rv == -1) { 463 rv = errno; 464 } 465 break; 466 default: 467 abort(); 468 } 469 470 rumpkern_sched(nlocks, NULL); 471 return rv; 472 } 473 474 static int 475 gethostncpu(void) 476 { 477 int ncpu = 1; 478 479 #if defined(__NetBSD__) || defined(__FreeBSD__) || defined(__DragonFly__) 480 size_t sz = sizeof(ncpu); 481 482 sysctlbyname("hw.ncpu", &ncpu, &sz, NULL, 0); 483 #elif defined(__linux__) || defined(__CYGWIN__) 484 FILE *fp; 485 char *line = NULL; 486 size_t n = 0; 487 488 /* If anyone knows a better way, I'm all ears */ 489 if ((fp = fopen("/proc/cpuinfo", "r")) != NULL) { 490 ncpu = 0; 491 while (getline(&line, &n, fp) != -1) { 492 if (strncmp(line, 493 "processor", sizeof("processor")-1) == 0) 494 ncpu++; 495 } 496 if (ncpu == 0) 497 ncpu = 1; 498 free(line); 499 fclose(fp); 500 } 501 #elif __sun__ 502 /* XXX: this is just a rough estimate ... */ 503 ncpu = sysconf(_SC_NPROCESSORS_ONLN); 504 #endif 505 506 return ncpu; 507 } 508 509 int 510 rumpuser_getparam(const char *name, void *buf, size_t blen) 511 { 512 int rv; 513 514 if (strcmp(name, RUMPUSER_PARAM_NCPU) == 0) { 515 int ncpu; 516 517 if (getenv_r("RUMP_NCPU", buf, blen) == -1) { 518 ncpu = gethostncpu(); 519 snprintf(buf, blen, "%d", ncpu); 520 } 521 rv = 0; 522 } else if (strcmp(name, RUMPUSER_PARAM_HOSTNAME) == 0) { 523 char tmp[MAXHOSTNAMELEN]; 524 525 if (gethostname(tmp, sizeof(tmp)) == -1) { 526 snprintf(buf, blen, "rump-%05d", (int)getpid()); 527 } else { 528 snprintf(buf, blen, "rump-%05d.%s", 529 (int)getpid(), tmp); 530 } 531 rv = 0; 532 } else if (*name == '_') { 533 rv = EINVAL; 534 } else { 535 if (getenv_r(name, buf, blen) == -1) 536 rv = errno; 537 else 538 rv = 0; 539 } 540 541 return rv; 542 } 543 544 void 545 rumpuser_putchar(int c) 546 { 547 548 putchar(c); 549 } 550 551 void 552 rumpuser_exit(int rv) 553 { 554 555 if (rv == RUMPUSER_PANIC) 556 abort(); 557 else 558 exit(rv); 559 } 560 561 void 562 rumpuser_seterrno(int error) 563 { 564 565 errno = error; 566 } 567 568 /* 569 * This is meant for safe debugging prints from the kernel. 570 */ 571 void 572 rumpuser_dprintf(const char *format, ...) 573 { 574 va_list ap; 575 576 va_start(ap, format); 577 vfprintf(stderr, format, ap); 578 va_end(ap); 579 } 580 581 int 582 rumpuser_kill(int64_t pid, int sig) 583 { 584 int rv, error; 585 586 #ifdef __NetBSD__ 587 if (pid == RUMPUSER_PID_SELF) { 588 error = raise(sig); 589 } else { 590 error = kill((pid_t)pid, sig); 591 } 592 if (error == -1) 593 rv = errno; 594 else 595 rv = 0; 596 #else 597 /* XXXfixme: signal numbers may not match on non-NetBSD */ 598 rv = EOPNOTSUPP; 599 #endif 600 601 return rv; 602 } 603 604 int 605 rumpuser_getrandom(void *buf, size_t buflen, int flags, size_t *retp) 606 { 607 size_t origlen = buflen; 608 uint32_t *p = buf; 609 uint32_t tmp; 610 int chunk; 611 612 do { 613 chunk = buflen < 4 ? buflen : 4; /* portable MIN ... */ 614 tmp = RUMPUSER_RANDOM(); 615 memcpy(p, &tmp, chunk); 616 p++; 617 buflen -= chunk; 618 } while (chunk); 619 620 *retp = origlen; 621 return 0; 622 } 623
Indexes created Wed Sep 24 15:09:51 GMT 2025