rumpuser.c revision 1.21
1 /* $NetBSD: rumpuser.c,v 1.21 2012/10/10 11:15:57 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 1.21 2012/10/10 11:15:57 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 ruflags, int *error) 321 { 322 int flags; 323 324 switch (ruflags & RUMPUSER_OPEN_ACCMODE) { 325 case RUMPUSER_OPEN_RDONLY: 326 flags = O_RDONLY; 327 break; 328 case RUMPUSER_OPEN_WRONLY: 329 flags = O_WRONLY; 330 break; 331 case RUMPUSER_OPEN_RDWR: 332 flags = O_RDWR; 333 break; 334 default: 335 *error = EINVAL; 336 return -1; 337 } 338 339 #define TESTSET(_ru_, _h_) if (ruflags & _ru_) flags |= _h_; 340 TESTSET(RUMPUSER_OPEN_CREATE, O_CREAT); 341 TESTSET(RUMPUSER_OPEN_EXCL, O_EXCL); 342 TESTSET(RUMPUSER_OPEN_DIRECT, O_DIRECT); 343 #undef TESTSET 344 345 DOCALL_KLOCK(int, (open(path, flags, 0644))); 346 } 347 348 int 349 rumpuser_ioctl(int fd, u_long cmd, void *data, int *error) 350 { 351 352 DOCALL_KLOCK(int, (ioctl(fd, cmd, data))); 353 } 354 355 int 356 rumpuser_close(int fd, int *error) 357 { 358 359 DOCALL(int, close(fd)); 360 } 361 362 int 363 rumpuser_fsync(int fd, int *error) 364 { 365 366 DOCALL_KLOCK(int, fsync(fd)); 367 } 368 369 ssize_t 370 rumpuser_read(int fd, void *data, size_t size, int *error) 371 { 372 ssize_t rv; 373 374 KLOCK_WRAP(rv = read(fd, data, size)); 375 if (rv == -1) 376 seterror(errno); 377 378 return rv; 379 } 380 381 ssize_t 382 rumpuser_pread(int fd, void *data, size_t size, off_t offset, int *error) 383 { 384 ssize_t rv; 385 386 KLOCK_WRAP(rv = pread(fd, data, size, offset)); 387 if (rv == -1) 388 seterror(errno); 389 390 return rv; 391 } 392 393 void 394 rumpuser_read_bio(int fd, void *data, size_t size, off_t offset, 395 rump_biodone_fn biodone, void *biodonecookie) 396 { 397 ssize_t rv; 398 int error = 0; 399 400 rv = rumpuser_pread(fd, data, size, offset, &error); 401 /* check against <0 instead of ==-1 to get typing below right */ 402 if (rv < 0) 403 rv = 0; 404 405 /* LINTED: see above */ 406 biodone(biodonecookie, rv, error); 407 } 408 409 ssize_t 410 rumpuser_write(int fd, const void *data, size_t size, int *error) 411 { 412 ssize_t rv; 413 414 KLOCK_WRAP(rv = write(fd, data, size)); 415 if (rv == -1) 416 seterror(errno); 417 418 return rv; 419 } 420 421 ssize_t 422 rumpuser_pwrite(int fd, const void *data, size_t size, off_t offset, int *error) 423 { 424 ssize_t rv; 425 426 KLOCK_WRAP(rv = pwrite(fd, data, size, offset)); 427 if (rv == -1) 428 seterror(errno); 429 430 return rv; 431 } 432 433 void 434 rumpuser_write_bio(int fd, const void *data, size_t size, off_t offset, 435 rump_biodone_fn biodone, void *biodonecookie) 436 { 437 ssize_t rv; 438 int error = 0; 439 440 rv = rumpuser_pwrite(fd, data, size, offset, &error); 441 /* check against <0 instead of ==-1 to get typing below right */ 442 if (rv < 0) 443 rv = 0; 444 445 /* LINTED: see above */ 446 biodone(biodonecookie, rv, error); 447 } 448 449 ssize_t 450 rumpuser_readv(int fd, const struct rumpuser_iovec *riov, int iovcnt, 451 int *error) 452 { 453 struct iovec *iovp; 454 ssize_t rv; 455 int i; 456 457 iovp = malloc(iovcnt * sizeof(struct iovec)); 458 if (iovp == NULL) { 459 seterror(ENOMEM); 460 return -1; 461 } 462 for (i = 0; i < iovcnt; i++) { 463 iovp[i].iov_base = riov[i].iov_base; 464 /*LINTED*/ 465 iovp[i].iov_len = riov[i].iov_len; 466 } 467 468 KLOCK_WRAP(rv = readv(fd, iovp, iovcnt)); 469 if (rv == -1) 470 seterror(errno); 471 free(iovp); 472 473 return rv; 474 } 475 476 ssize_t 477 rumpuser_writev(int fd, const struct rumpuser_iovec *riov, int iovcnt, 478 int *error) 479 { 480 struct iovec *iovp; 481 ssize_t rv; 482 int i; 483 484 iovp = malloc(iovcnt * sizeof(struct iovec)); 485 if (iovp == NULL) { 486 seterror(ENOMEM); 487 return -1; 488 } 489 for (i = 0; i < iovcnt; i++) { 490 iovp[i].iov_base = riov[i].iov_base; 491 /*LINTED*/ 492 iovp[i].iov_len = riov[i].iov_len; 493 } 494 495 KLOCK_WRAP(rv = writev(fd, iovp, iovcnt)); 496 if (rv == -1) 497 seterror(errno); 498 free(iovp); 499 500 return rv; 501 } 502 503 int 504 rumpuser_gettime(uint64_t *sec, uint64_t *nsec, int *error) 505 { 506 struct timeval tv; 507 int rv; 508 509 rv = gettimeofday(&tv, NULL); 510 if (rv == -1) { 511 seterror(errno); 512 return rv; 513 } 514 515 *sec = tv.tv_sec; 516 *nsec = tv.tv_usec * 1000; 517 518 return 0; 519 } 520 521 int 522 rumpuser_getenv(const char *name, char *buf, size_t blen, int *error) 523 { 524 525 DOCALL(int, getenv_r(name, buf, blen)); 526 } 527 528 int 529 rumpuser_gethostname(char *name, size_t namelen, int *error) 530 { 531 char tmp[MAXHOSTNAMELEN]; 532 533 if (gethostname(tmp, sizeof(tmp)) == -1) { 534 snprintf(name, namelen, "rump-%05d.rumpdomain", getpid()); 535 } else { 536 snprintf(name, namelen, "rump-%05d.%s.rumpdomain", 537 getpid(), tmp); 538 } 539 540 *error = 0; 541 return 0; 542 } 543 544 int 545 rumpuser_poll(struct pollfd *fds, int nfds, int timeout, int *error) 546 { 547 548 DOCALL_KLOCK(int, (poll(fds, (nfds_t)nfds, timeout))); 549 } 550 551 int 552 rumpuser_putchar(int c, int *error) 553 { 554 555 DOCALL(int, (putchar(c))); 556 } 557 558 void 559 rumpuser_exit(int rv) 560 { 561 562 if (rv == RUMPUSER_PANIC) 563 abort(); 564 else 565 exit(rv); 566 } 567 568 void 569 rumpuser_seterrno(int error) 570 { 571 572 errno = error; 573 } 574 575 /* 576 * On NetBSD we use kqueue, on Linux we use inotify. The underlying 577 * interface requirements aren't quite the same, but we have a very 578 * good chance of doing the fd->path mapping on Linux thanks to dcache, 579 * so just keep the existing interfaces for now. 580 */ 581 #if defined(__NetBSD__) 582 int 583 rumpuser_writewatchfile_setup(int kq, int fd, intptr_t opaque, int *error) 584 { 585 struct kevent kev; 586 587 if (kq == -1) { 588 kq = kqueue(); 589 if (kq == -1) { 590 seterror(errno); 591 return -1; 592 } 593 } 594 595 EV_SET(&kev, fd, EVFILT_VNODE, EV_ADD|EV_ENABLE|EV_CLEAR, 596 NOTE_WRITE, 0, opaque); 597 if (kevent(kq, &kev, 1, NULL, 0, NULL) == -1) { 598 seterror(errno); 599 return -1; 600 } 601 602 return kq; 603 } 604 605 int 606 rumpuser_writewatchfile_wait(int kq, intptr_t *opaque, int *error) 607 { 608 struct kevent kev; 609 int rv; 610 611 again: 612 KLOCK_WRAP(rv = kevent(kq, NULL, 0, &kev, 1, NULL)); 613 if (rv == -1) { 614 if (errno == EINTR) 615 goto again; 616 seterror(errno); 617 return -1; 618 } 619 620 if (opaque) 621 *opaque = kev.udata; 622 return rv; 623 } 624 625 #elif defined(__linux__) 626 #include <sys/inotify.h> 627 628 int 629 rumpuser_writewatchfile_setup(int inotify, int fd, intptr_t notused, int *error) 630 { 631 char procbuf[PATH_MAX], linkbuf[PATH_MAX]; 632 ssize_t nn; 633 634 if (inotify == -1) { 635 inotify = inotify_init(); 636 if (inotify == -1) { 637 seterror(errno); 638 return -1; 639 } 640 } 641 642 /* ok, need to map fd into path for inotify */ 643 snprintf(procbuf, sizeof(procbuf), "/proc/self/fd/%d", fd); 644 nn = readlink(procbuf, linkbuf, sizeof(linkbuf)); 645 if (nn >= (ssize_t)sizeof(linkbuf)) { 646 nn = -1; 647 errno = E2BIG; /* pick something */ 648 } 649 if (nn == -1) { 650 seterror(errno); 651 close(inotify); 652 return -1; 653 } 654 655 if (inotify_add_watch(inotify, linkbuf, IN_MODIFY) == -1) { 656 seterror(errno); 657 close(inotify); 658 return -1; 659 } 660 661 return inotify; 662 } 663 664 int 665 rumpuser_writewatchfile_wait(int kq, intptr_t *opaque, int *error) 666 { 667 struct inotify_event iev; 668 ssize_t nn; 669 670 do { 671 KLOCK_WRAP(nn = read(kq, &iev, sizeof(iev))); 672 } while (errno == EINTR); 673 674 if (nn == -1) { 675 seterror(errno); 676 return -1; 677 } 678 return (nn/sizeof(iev)); 679 } 680 #endif 681 682 /* 683 * This is meant for safe debugging prints from the kernel. 684 */ 685 int 686 rumpuser_dprintf(const char *format, ...) 687 { 688 va_list ap; 689 int rv; 690 691 va_start(ap, format); 692 rv = vfprintf(stderr, format, ap); 693 va_end(ap); 694 695 return rv; 696 } 697 698 int 699 rumpuser_kill(int64_t pid, int sig, int *error) 700 { 701 702 #ifdef __NetBSD__ 703 if (pid == RUMPUSER_PID_SELF) { 704 DOCALL(int, raise(sig)); 705 } else { 706 DOCALL(int, kill((pid_t)pid, sig)); 707 } 708 #else 709 /* XXXfixme: signal numbers may not match on non-NetBSD */ 710 seterror(EOPNOTSUPP); 711 return -1; 712 #endif 713 } 714 715 int 716 rumpuser_getnhostcpu(void) 717 { 718 int ncpu = 1; 719 720 #ifdef __NetBSD__ 721 size_t sz = sizeof(ncpu); 722 723 sysctlbyname("hw.ncpu", &ncpu, &sz, NULL, 0); 724 #elif __linux__ 725 FILE *fp; 726 char *line = NULL; 727 size_t n = 0; 728 729 /* If anyone knows a better way, I'm all ears */ 730 if ((fp = fopen("/proc/cpuinfo", "r")) != NULL) { 731 ncpu = 0; 732 while (getline(&line, &n, fp) != -1) { 733 if (strncmp(line, 734 "processor", sizeof("processor")-1) == 0) 735 ncpu++; 736 } 737 if (ncpu == 0) 738 ncpu = 1; 739 free(line); 740 fclose(fp); 741 } 742 #endif 743 744 return ncpu; 745 } 746 747 /* XXX: this hypercall needs a better name */ 748 uint32_t 749 rumpuser_arc4random(void) 750 { 751 752 return arc4random(); 753 } 754
Indexes created Wed Sep 24 05:09:52 GMT 2025