rumpuser.c revision 1.18
1 /* $NetBSD: rumpuser.c,v 1.18 2012/07/27 09:09:05 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 1.18 2012/07/27 09:09:05 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 #include <sys/sysctl.h> 45 #include <sys/event.h> 46 #endif 47 48 #include <assert.h> 49 #include <err.h> 50 #include <errno.h> 51 #include <fcntl.h> 52 #include <poll.h> 53 #include <signal.h> 54 #include <stdarg.h> 55 #include <stdint.h> 56 #include <stdio.h> 57 #include <stdlib.h> 58 #include <string.h> 59 #include <time.h> 60 #include <unistd.h> 61 62 #include <rump/rumpuser.h> 63 64 #include "rumpuser_int.h" 65 66 int 67 rumpuser_getversion(void) 68 { 69 70 return RUMPUSER_VERSION; 71 } 72 73 int 74 rumpuser_getfileinfo(const char *path, uint64_t *sizep, int *ftp, int *error) 75 { 76 struct stat sb; 77 uint64_t size; 78 int needsdev = 0, rv = 0, ft; 79 int fd = -1; 80 81 if (stat(path, &sb) == -1) { 82 seterror(errno); 83 return -1; 84 } 85 86 switch (sb.st_mode & S_IFMT) { 87 case S_IFDIR: 88 ft = RUMPUSER_FT_DIR; 89 break; 90 case S_IFREG: 91 ft = RUMPUSER_FT_REG; 92 break; 93 case S_IFBLK: 94 ft = RUMPUSER_FT_BLK; 95 needsdev = 1; 96 break; 97 case S_IFCHR: 98 ft = RUMPUSER_FT_CHR; 99 needsdev = 1; 100 break; 101 default: 102 ft = RUMPUSER_FT_OTHER; 103 break; 104 } 105 106 if (!needsdev) { 107 size = sb.st_size; 108 } else if (sizep) { 109 /* 110 * Welcome to the jungle. Of course querying the kernel 111 * for a device partition size is supposed to be far from 112 * trivial. On NetBSD we use ioctl. On $other platform 113 * we have a problem. We try "the lseek trick" and just 114 * fail if that fails. Platform specific code can later 115 * be written here if appropriate. 116 * 117 * On NetBSD we hope and pray that for block devices nobody 118 * else is holding them open, because otherwise the kernel 119 * will not permit us to open it. Thankfully, this is 120 * usually called only in bootstrap and then we can 121 * forget about it. 122 */ 123 #ifndef __NetBSD__ 124 off_t off; 125 126 fd = open(path, O_RDONLY); 127 if (fd == -1) { 128 seterror(errno); 129 rv = -1; 130 goto out; 131 } 132 133 off = lseek(fd, 0, SEEK_END); 134 if (off != 0) { 135 size = off; 136 goto out; 137 } 138 fprintf(stderr, "error: device size query not implemented on " 139 "this platform\n"); 140 seterror(EOPNOTSUPP); 141 rv = -1; 142 goto out; 143 #else 144 struct disklabel lab; 145 struct partition *parta; 146 struct dkwedge_info dkw; 147 148 fd = open(path, O_RDONLY); 149 if (fd == -1) { 150 seterror(errno); 151 rv = -1; 152 goto out; 153 } 154 155 if (ioctl(fd, DIOCGDINFO, &lab) == 0) { 156 parta = &lab.d_partitions[DISKPART(sb.st_rdev)]; 157 size = (uint64_t)lab.d_secsize * parta->p_size; 158 goto out; 159 } 160 161 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) == 0) { 162 /* 163 * XXX: should use DIOCGDISKINFO to query 164 * sector size, but that requires proplib, 165 * so just don't bother for now. it's nice 166 * that something as difficult as figuring out 167 * a partition's size has been made so easy. 168 */ 169 size = dkw.dkw_size << DEV_BSHIFT; 170 goto out; 171 } 172 173 seterror(errno); 174 rv = -1; 175 #endif /* __NetBSD__ */ 176 } 177 178 out: 179 if (rv == 0 && sizep) 180 *sizep = size; 181 if (rv == 0 && ftp) 182 *ftp = ft; 183 if (fd != -1) 184 close(fd); 185 186 return rv; 187 } 188 189 int 190 rumpuser_nanosleep(uint64_t *sec, uint64_t *nsec, int *error) 191 { 192 struct timespec rqt, rmt; 193 int rv; 194 195 /*LINTED*/ 196 rqt.tv_sec = *sec; 197 /*LINTED*/ 198 rqt.tv_nsec = *nsec; 199 200 KLOCK_WRAP(rv = nanosleep(&rqt, &rmt)); 201 if (rv == -1) 202 seterror(errno); 203 204 *sec = rmt.tv_sec; 205 *nsec = rmt.tv_nsec; 206 207 return rv; 208 } 209 210 void * 211 rumpuser_malloc(size_t howmuch, int alignment) 212 { 213 void *mem; 214 int rv; 215 216 if (alignment == 0) 217 alignment = sizeof(void *); 218 219 rv = posix_memalign(&mem, (size_t)alignment, howmuch); 220 if (__predict_false(rv != 0)) { 221 if (rv == EINVAL) { 222 printf("rumpuser_malloc: invalid alignment %d\n", 223 alignment); 224 abort(); 225 } 226 mem = NULL; 227 } 228 229 return mem; 230 } 231 232 void * 233 rumpuser_realloc(void *ptr, size_t howmuch) 234 { 235 236 return realloc(ptr, howmuch); 237 } 238 239 void 240 rumpuser_free(void *ptr) 241 { 242 243 free(ptr); 244 } 245 246 void * 247 rumpuser_anonmmap(void *prefaddr, size_t size, int alignbit, 248 int exec, int *error) 249 { 250 void *rv; 251 int prot; 252 253 #ifndef MAP_ALIGNED 254 #define MAP_ALIGNED(a) 0 255 if (alignbit) 256 fprintf(stderr, "rumpuser_anonmmap: warning, requested " 257 "alignment not supported by hypervisor\n"); 258 #endif 259 260 prot = PROT_READ|PROT_WRITE; 261 if (exec) 262 prot |= PROT_EXEC; 263 rv = mmap(prefaddr, size, prot, 264 MAP_ANON | MAP_ALIGNED(alignbit), -1, 0); 265 if (rv == MAP_FAILED) { 266 seterror(errno); 267 return NULL; 268 } 269 return rv; 270 } 271 272 void 273 rumpuser_unmap(void *addr, size_t len) 274 { 275 int rv; 276 277 rv = munmap(addr, len); 278 assert(rv == 0); 279 } 280 281 void * 282 rumpuser_filemmap(int fd, off_t offset, size_t len, int flags, int *error) 283 { 284 void *rv; 285 int mmflags, prot; 286 287 if (flags & RUMPUSER_FILEMMAP_TRUNCATE) 288 ftruncate(fd, offset + len); 289 290 mmflags = MAP_FILE; 291 if (flags & RUMPUSER_FILEMMAP_SHARED) 292 mmflags |= MAP_SHARED; 293 else 294 mmflags |= MAP_PRIVATE; 295 296 prot = 0; 297 if (flags & RUMPUSER_FILEMMAP_READ) 298 prot |= PROT_READ; 299 if (flags & RUMPUSER_FILEMMAP_WRITE) 300 prot |= PROT_WRITE; 301 302 rv = mmap(NULL, len, PROT_READ|PROT_WRITE, mmflags, fd, offset); 303 if (rv == MAP_FAILED) { 304 seterror(errno); 305 return NULL; 306 } 307 308 seterror(0); 309 return rv; 310 } 311 312 int 313 rumpuser_memsync(void *addr, size_t len, int *error) 314 { 315 316 DOCALL_KLOCK(int, (msync(addr, len, MS_SYNC))); 317 } 318 319 int 320 rumpuser_open(const char *path, int flags, int *error) 321 { 322 323 DOCALL(int, (open(path, flags, 0644))); 324 } 325 326 int 327 rumpuser_ioctl(int fd, u_long cmd, void *data, int *error) 328 { 329 330 DOCALL_KLOCK(int, (ioctl(fd, cmd, data))); 331 } 332 333 int 334 rumpuser_close(int fd, int *error) 335 { 336 337 DOCALL(int, close(fd)); 338 } 339 340 int 341 rumpuser_fsync(int fd, int *error) 342 { 343 344 DOCALL_KLOCK(int, fsync(fd)); 345 } 346 347 ssize_t 348 rumpuser_read(int fd, void *data, size_t size, int *error) 349 { 350 ssize_t rv; 351 352 KLOCK_WRAP(rv = read(fd, data, size)); 353 if (rv == -1) 354 seterror(errno); 355 356 return rv; 357 } 358 359 ssize_t 360 rumpuser_pread(int fd, void *data, size_t size, off_t offset, int *error) 361 { 362 ssize_t rv; 363 364 KLOCK_WRAP(rv = pread(fd, data, size, offset)); 365 if (rv == -1) 366 seterror(errno); 367 368 return rv; 369 } 370 371 void 372 rumpuser_read_bio(int fd, void *data, size_t size, off_t offset, 373 rump_biodone_fn biodone, void *biodonecookie) 374 { 375 ssize_t rv; 376 int error = 0; 377 378 rv = rumpuser_pread(fd, data, size, offset, &error); 379 /* check against <0 instead of ==-1 to get typing below right */ 380 if (rv < 0) 381 rv = 0; 382 383 /* LINTED: see above */ 384 biodone(biodonecookie, rv, error); 385 } 386 387 ssize_t 388 rumpuser_write(int fd, const void *data, size_t size, int *error) 389 { 390 ssize_t rv; 391 392 KLOCK_WRAP(rv = write(fd, data, size)); 393 if (rv == -1) 394 seterror(errno); 395 396 return rv; 397 } 398 399 ssize_t 400 rumpuser_pwrite(int fd, const void *data, size_t size, off_t offset, int *error) 401 { 402 ssize_t rv; 403 404 KLOCK_WRAP(rv = pwrite(fd, data, size, offset)); 405 if (rv == -1) 406 seterror(errno); 407 408 return rv; 409 } 410 411 void 412 rumpuser_write_bio(int fd, const void *data, size_t size, off_t offset, 413 rump_biodone_fn biodone, void *biodonecookie) 414 { 415 ssize_t rv; 416 int error = 0; 417 418 rv = rumpuser_pwrite(fd, data, size, offset, &error); 419 /* check against <0 instead of ==-1 to get typing below right */ 420 if (rv < 0) 421 rv = 0; 422 423 /* LINTED: see above */ 424 biodone(biodonecookie, rv, error); 425 } 426 427 ssize_t 428 rumpuser_readv(int fd, const struct rumpuser_iovec *riov, int iovcnt, 429 int *error) 430 { 431 struct iovec *iovp; 432 ssize_t rv; 433 int i; 434 435 iovp = malloc(iovcnt * sizeof(struct iovec)); 436 if (iovp == NULL) { 437 seterror(ENOMEM); 438 return -1; 439 } 440 for (i = 0; i < iovcnt; i++) { 441 iovp[i].iov_base = riov[i].iov_base; 442 /*LINTED*/ 443 iovp[i].iov_len = riov[i].iov_len; 444 } 445 446 KLOCK_WRAP(rv = readv(fd, iovp, iovcnt)); 447 if (rv == -1) 448 seterror(errno); 449 free(iovp); 450 451 return rv; 452 } 453 454 ssize_t 455 rumpuser_writev(int fd, const struct rumpuser_iovec *riov, int iovcnt, 456 int *error) 457 { 458 struct iovec *iovp; 459 ssize_t rv; 460 int i; 461 462 iovp = malloc(iovcnt * sizeof(struct iovec)); 463 if (iovp == NULL) { 464 seterror(ENOMEM); 465 return -1; 466 } 467 for (i = 0; i < iovcnt; i++) { 468 iovp[i].iov_base = riov[i].iov_base; 469 /*LINTED*/ 470 iovp[i].iov_len = riov[i].iov_len; 471 } 472 473 KLOCK_WRAP(rv = writev(fd, iovp, iovcnt)); 474 if (rv == -1) 475 seterror(errno); 476 free(iovp); 477 478 return rv; 479 } 480 481 int 482 rumpuser_gettime(uint64_t *sec, uint64_t *nsec, int *error) 483 { 484 struct timeval tv; 485 int rv; 486 487 rv = gettimeofday(&tv, NULL); 488 if (rv == -1) { 489 seterror(errno); 490 return rv; 491 } 492 493 *sec = tv.tv_sec; 494 *nsec = tv.tv_usec * 1000; 495 496 return 0; 497 } 498 499 int 500 rumpuser_getenv(const char *name, char *buf, size_t blen, int *error) 501 { 502 503 #ifdef __linux__ 504 char *tmp; 505 506 *error = 0; 507 if ((tmp = getenv(name)) != NULL) { 508 if (strlen(tmp) >= blen) { 509 *error = ERANGE; 510 return -1; 511 } 512 strcpy(buf, tmp); 513 return 0; 514 } else { 515 *error = ENOENT; 516 return -1; 517 } 518 #else 519 DOCALL(int, getenv_r(name, buf, blen)); 520 #endif 521 } 522 523 int 524 rumpuser_gethostname(char *name, size_t namelen, int *error) 525 { 526 char tmp[MAXHOSTNAMELEN]; 527 528 if (gethostname(tmp, sizeof(tmp)) == -1) { 529 snprintf(name, namelen, "rump-%05d.rumpdomain", getpid()); 530 } else { 531 snprintf(name, namelen, "rump-%05d.%s.rumpdomain", 532 getpid(), tmp); 533 } 534 535 *error = 0; 536 return 0; 537 } 538 539 int 540 rumpuser_poll(struct pollfd *fds, int nfds, int timeout, int *error) 541 { 542 543 DOCALL_KLOCK(int, (poll(fds, (nfds_t)nfds, timeout))); 544 } 545 546 int 547 rumpuser_putchar(int c, int *error) 548 { 549 550 DOCALL(int, (putchar(c))); 551 } 552 553 void 554 rumpuser_exit(int rv) 555 { 556 557 if (rv == RUMPUSER_PANIC) 558 abort(); 559 else 560 exit(rv); 561 } 562 563 void 564 rumpuser_seterrno(int error) 565 { 566 567 errno = error; 568 } 569 570 #ifdef __NetBSD__ 571 int 572 rumpuser_writewatchfile_setup(int kq, int fd, intptr_t opaque, int *error) 573 { 574 struct kevent kev; 575 576 if (kq == -1) { 577 kq = kqueue(); 578 if (kq == -1) { 579 seterror(errno); 580 return -1; 581 } 582 } 583 584 EV_SET(&kev, fd, EVFILT_VNODE, EV_ADD|EV_ENABLE|EV_CLEAR, 585 NOTE_WRITE, 0, opaque); 586 if (kevent(kq, &kev, 1, NULL, 0, NULL) == -1) { 587 seterror(errno); 588 return -1; 589 } 590 591 return kq; 592 } 593 594 int 595 rumpuser_writewatchfile_wait(int kq, intptr_t *opaque, int *error) 596 { 597 struct kevent kev; 598 int rv; 599 600 again: 601 KLOCK_WRAP(rv = kevent(kq, NULL, 0, &kev, 1, NULL)); 602 if (rv == -1) { 603 if (errno == EINTR) 604 goto again; 605 seterror(errno); 606 return -1; 607 } 608 609 if (opaque) 610 *opaque = kev.udata; 611 return rv; 612 } 613 #endif 614 615 /* 616 * This is meant for safe debugging prints from the kernel. 617 */ 618 int 619 rumpuser_dprintf(const char *format, ...) 620 { 621 va_list ap; 622 int rv; 623 624 va_start(ap, format); 625 rv = vfprintf(stderr, format, ap); 626 va_end(ap); 627 628 return rv; 629 } 630 631 int 632 rumpuser_kill(int64_t pid, int sig, int *error) 633 { 634 635 #ifdef __NetBSD__ 636 if (pid == RUMPUSER_PID_SELF) { 637 DOCALL(int, raise(sig)); 638 } else { 639 DOCALL(int, kill((pid_t)pid, sig)); 640 } 641 #else 642 /* XXXfixme: signal numbers may not match on non-NetBSD */ 643 seterror(EOPNOTSUPP); 644 return -1; 645 #endif 646 } 647 648 int 649 rumpuser_getnhostcpu(void) 650 { 651 int ncpu = 1; 652 653 #ifdef __NetBSD__ 654 size_t sz = sizeof(ncpu); 655 656 sysctlbyname("hw.ncpu", &ncpu, &sz, NULL, 0); 657 #elif __linux__ 658 FILE *fp; 659 char *line = NULL; 660 size_t n = 0; 661 662 /* If anyone knows a better way, I'm all ears */ 663 if ((fp = fopen("/proc/cpuinfo", "r")) != NULL) { 664 ncpu = 0; 665 while (getline(&line, &n, fp) != -1) { 666 if (strncmp(line, 667 "processor", sizeof("processor")-1) == 0) 668 ncpu++; 669 } 670 if (ncpu == 0) 671 ncpu = 1; 672 free(line); 673 fclose(fp); 674 } 675 #endif 676 677 return ncpu; 678 } 679 680 /* XXX: this hypercall needs a better name */ 681 uint32_t 682 rumpuser_arc4random(void) 683 { 684 685 return arc4random(); 686 } 687
Indexes created Fri Sep 26 08:10:20 GMT 2025