linux_pipe.c revision 1.20
1/* $NetBSD: linux_pipe.c,v 1.20 1995/09/19 22:37:33 thorpej Exp $ */ 2 3/* 4 * Copyright (c) 1995 Frank van der Linden 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed for the NetBSD Project 18 * by Frank van der Linden 19 * 4. The name of the author may not be used to endorse or promote products 20 * derived from this software without specific prior written permission 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34/* 35 * Linux compatibility module. Try to deal with various Linux system calls. 36 */ 37 38#include <sys/param.h> 39#include <sys/systm.h> 40#include <sys/namei.h> 41#include <sys/proc.h> 42#include <sys/dir.h> 43#include <sys/file.h> 44#include <sys/stat.h> 45#include <sys/filedesc.h> 46#include <sys/ioctl.h> 47#include <sys/kernel.h> 48#include <sys/malloc.h> 49#include <sys/mbuf.h> 50#include <sys/mman.h> 51#include <sys/mount.h> 52#include <sys/ptrace.h> 53#include <sys/resource.h> 54#include <sys/resourcevar.h> 55#include <sys/signal.h> 56#include <sys/signalvar.h> 57#include <sys/socket.h> 58#include <sys/time.h> 59#include <sys/times.h> 60#include <sys/vnode.h> 61#include <sys/uio.h> 62#include <sys/wait.h> 63#include <sys/utsname.h> 64#include <sys/unistd.h> 65 66#include <sys/syscallargs.h> 67 68#include <vm/vm.h> 69#include <vm/vm_param.h> 70 71#include <compat/linux/linux_types.h> 72#include <compat/linux/linux_fcntl.h> 73#include <compat/linux/linux_mmap.h> 74#include <compat/linux/linux_signal.h> 75#include <compat/linux/linux_syscallargs.h> 76#include <compat/linux/linux_util.h> 77#include <compat/linux/linux_dirent.h> 78 79/* 80 * The information on a terminated (or stopped) process needs 81 * to be converted in order for Linux binaries to get a valid signal 82 * number out of it. 83 */ 84static int 85bsd_to_linux_wstat(status) 86 int *status; 87{ 88 if (WIFSIGNALED(*status)) 89 *status = (*status & ~0177) | 90 bsd_to_linux_sig[WTERMSIG(*status)]; 91 else if (WIFSTOPPED(*status)) 92 *status = (*status & ~0xff00) | 93 (bsd_to_linux_sig[WSTOPSIG(*status)] << 8); 94} 95 96/* 97 * waitpid(2). Passed on to the NetBSD call, surrounded by code to 98 * reserve some space for a NetBSD-style wait status, and converting 99 * it to what Linux wants. 100 */ 101int 102linux_waitpid(p, v, retval) 103 struct proc *p; 104 void *v; 105 register_t *retval; 106{ 107 struct linux_waitpid_args /* { 108 syscallarg(int) pid; 109 syscallarg(int *) status; 110 syscallarg(int) options; 111 } */ *uap = v; 112 struct wait4_args w4a; 113 int error, *status, tstat; 114 caddr_t sg; 115 116 if (SCARG(uap, status) != NULL) { 117 sg = stackgap_init(p->p_emul); 118 status = (int *) stackgap_alloc(&sg, sizeof status); 119 } else 120 status = NULL; 121 122 SCARG(&w4a, pid) = SCARG(uap, pid); 123 SCARG(&w4a, status) = status; 124 SCARG(&w4a, options) = SCARG(uap, options); 125 SCARG(&w4a, rusage) = NULL; 126 127 if ((error = wait4(p, &w4a, retval))) 128 return error; 129 130 p->p_siglist &= ~sigmask(SIGCHLD); 131 132 if (status != NULL) { 133 if ((error = copyin(status, &tstat, sizeof tstat))) 134 return error; 135 136 bsd_to_linux_wstat(&tstat); 137 138 return copyout(&tstat, SCARG(uap, status), sizeof tstat); 139 } 140 141 return 0; 142} 143 144/* 145 * This is very much the same as waitpid() 146 */ 147int 148linux_wait4(p, v, retval) 149 struct proc *p; 150 void *v; 151 register_t *retval; 152{ 153 struct linux_wait4_args /* { 154 syscallarg(int) pid; 155 syscallarg(int *) status; 156 syscallarg(int) options; 157 syscallarg(struct rusage *) rusage; 158 } */ *uap = v; 159 struct wait4_args w4a; 160 int error, *status, tstat; 161 caddr_t sg; 162 163 if (SCARG(uap, status) != NULL) { 164 sg = stackgap_init(p->p_emul); 165 status = (int *) stackgap_alloc(&sg, sizeof status); 166 } else 167 status = NULL; 168 169 SCARG(&w4a, pid) = SCARG(uap, pid); 170 SCARG(&w4a, status) = status; 171 SCARG(&w4a, options) = SCARG(uap, options); 172 SCARG(&w4a, rusage) = SCARG(uap, rusage); 173 174 if ((error = wait4(p, &w4a, retval))) 175 return error; 176 177 p->p_siglist &= ~sigmask(SIGCHLD); 178 179 if (status != NULL) { 180 if ((error = copyin(status, &tstat, sizeof tstat))) 181 return error; 182 183 bsd_to_linux_wstat(&tstat); 184 185 return copyout(&tstat, SCARG(uap, status), sizeof tstat); 186 } 187 188 return 0; 189} 190 191/* 192 * This is the old brk(2) call. I don't think anything in the Linux 193 * world uses this anymore 194 */ 195int 196linux_break(p, v, retval) 197 struct proc *p; 198 void *v; 199 register_t *retval; 200{ 201 struct linux_brk_args /* { 202 syscallarg(char *) nsize; 203 } */ *uap = v; 204 205 return ENOSYS; 206} 207 208/* 209 * Linux brk(2). The check if the new address is >= the old one is 210 * done in the kernel in Linux. NetBSD does it in the library. 211 */ 212int 213linux_brk(p, v, retval) 214 struct proc *p; 215 void *v; 216 register_t *retval; 217{ 218 struct linux_brk_args /* { 219 syscallarg(char *) nsize; 220 } */ *uap = v; 221 char *nbrk = SCARG(uap, nsize); 222 struct obreak_args oba; 223 struct vmspace *vm = p->p_vmspace; 224 int error = 0; 225 caddr_t oldbrk, newbrk; 226 227 oldbrk = vm->vm_daddr + ctob(vm->vm_dsize); 228 /* 229 * XXX inconsistent.. Linux always returns at least the old 230 * brk value, but it will be page-aligned if this fails, 231 * and possibly not page aligned if it succeeds (the user 232 * supplied pointer is returned). 233 */ 234 SCARG(&oba, nsize) = nbrk; 235 236 if ((caddr_t) nbrk > vm->vm_daddr && obreak(p, &oba, retval) == 0) 237 retval[0] = (register_t) nbrk; 238 else 239 retval[0] = (register_t) oldbrk; 240 241 return 0; 242} 243 244/* 245 * I wonder why Linux has gettimeofday() _and_ time().. Still, we 246 * need to deal with it. 247 */ 248int 249linux_time(p, v, retval) 250 struct proc *p; 251 void *v; 252 register_t *retval; 253{ 254 struct linux_time_args /* { 255 linux_time_t *t; 256 } */ *uap = v; 257 struct timeval atv; 258 linux_time_t tt; 259 int error; 260 261 microtime(&atv); 262 263 tt = atv.tv_sec; 264 if (SCARG(uap, t) && (error = copyout(&tt, SCARG(uap, t), sizeof tt))) 265 return error; 266 267 retval[0] = tt; 268 return 0; 269} 270 271/* 272 * Convert BSD statfs structure to Linux statfs structure. 273 * The Linux structure has less fields, and it also wants 274 * the length of a name in a dir entry in a field, which 275 * we fake (probably the wrong way). 276 */ 277static void 278bsd_to_linux_statfs(bsp, lsp) 279 struct statfs *bsp; 280 struct linux_statfs *lsp; 281{ 282 lsp->l_ftype = bsp->f_type; 283 lsp->l_fbsize = bsp->f_bsize; 284 lsp->l_fblocks = bsp->f_blocks; 285 lsp->l_fbfree = bsp->f_bfree; 286 lsp->l_fbavail = bsp->f_bavail; 287 lsp->l_ffiles = bsp->f_files; 288 lsp->l_fffree = bsp->f_ffree; 289 lsp->l_ffsid.val[0] = bsp->f_fsid.val[0]; 290 lsp->l_ffsid.val[1] = bsp->f_fsid.val[1]; 291 lsp->l_fnamelen = MAXNAMLEN; /* XXX */ 292} 293 294/* 295 * Implement the fs stat functions. Straightforward. 296 */ 297int 298linux_statfs(p, v, retval) 299 struct proc *p; 300 void *v; 301 register_t *retval; 302{ 303 struct linux_statfs_args /* { 304 syscallarg(char *) path; 305 syscallarg(struct linux_statfs *) sp; 306 } */ *uap = v; 307 struct statfs btmp, *bsp; 308 struct linux_statfs ltmp; 309 struct statfs_args bsa; 310 caddr_t sg; 311 int error; 312 313 sg = stackgap_init(p->p_emul); 314 bsp = (struct statfs *) stackgap_alloc(&sg, sizeof (struct statfs)); 315 316 LINUX_CHECK_ALT_EXIST(p, &sg, SCARG(uap, path)); 317 318 SCARG(&bsa, path) = SCARG(uap, path); 319 SCARG(&bsa, buf) = bsp; 320 321 if ((error = statfs(p, &bsa, retval))) 322 return error; 323 324 if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp))) 325 return error; 326 327 bsd_to_linux_statfs(&btmp, <mp); 328 329 return copyout((caddr_t) <mp, (caddr_t) SCARG(uap, sp), sizeof ltmp); 330} 331 332int 333linux_fstatfs(p, v, retval) 334 struct proc *p; 335 void *v; 336 register_t *retval; 337{ 338 struct linux_fstatfs_args /* { 339 syscallarg(int) fd; 340 syscallarg(struct linux_statfs *) sp; 341 } */ *uap = v; 342 struct statfs btmp, *bsp; 343 struct linux_statfs ltmp; 344 struct fstatfs_args bsa; 345 caddr_t sg; 346 int error; 347 348 sg = stackgap_init(p->p_emul); 349 bsp = (struct statfs *) stackgap_alloc(&sg, sizeof (struct statfs)); 350 351 SCARG(&bsa, fd) = SCARG(uap, fd); 352 SCARG(&bsa, buf) = bsp; 353 354 if ((error = statfs(p, &bsa, retval))) 355 return error; 356 357 if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp))) 358 return error; 359 360 bsd_to_linux_statfs(&btmp, <mp); 361 362 return copyout((caddr_t) <mp, (caddr_t) SCARG(uap, sp), sizeof ltmp); 363} 364 365/* 366 * uname(). Just copy the info from the various strings stored in the 367 * kernel, and put it in the Linux utsname structure. That structure 368 * is almost the same as the NetBSD one, only it has fields 65 characters 369 * long, and an extra domainname field. 370 */ 371int 372linux_uname(p, v, retval) 373 struct proc *p; 374 void *v; 375 register_t *retval; 376{ 377 struct linux_uname_args /* { 378 syscallarg(struct linux_utsname *) up; 379 } */ *uap = v; 380 extern char ostype[], hostname[], osrelease[], version[], machine[], 381 domainname[]; 382 struct linux_utsname luts; 383 int len; 384 char *cp; 385 386 strncpy(luts.l_sysname, ostype, sizeof(luts.l_sysname)); 387 strncpy(luts.l_nodename, hostname, sizeof(luts.l_nodename)); 388 strncpy(luts.l_release, osrelease, sizeof(luts.l_release)); 389 strncpy(luts.l_version, version, sizeof(luts.l_version)); 390 strncpy(luts.l_machine, machine, sizeof(luts.l_machine)); 391 strncpy(luts.l_domainname, domainname, sizeof(luts.l_domainname)); 392 393 /* This part taken from the the uname() in libc */ 394 len = sizeof(luts.l_version); 395 for (cp = luts.l_version; len--; ++cp) 396 if (*cp == '\n' || *cp == '\t') 397 if (len > 1) 398 *cp = ' '; 399 else 400 *cp = '\0'; 401 402 return copyout(&luts, SCARG(uap, up), sizeof(luts)); 403} 404 405int 406linux_olduname(p, v, retval) 407 struct proc *p; 408 void *v; 409 register_t *retval; 410{ 411 struct linux_uname_args /* { 412 syscallarg(struct linux_oldutsname *) up; 413 } */ *uap = v; 414 extern char ostype[], hostname[], osrelease[], version[], machine[]; 415 struct linux_oldutsname luts; 416 int len; 417 char *cp; 418 419 strncpy(luts.l_sysname, ostype, sizeof(luts.l_sysname)); 420 strncpy(luts.l_nodename, hostname, sizeof(luts.l_nodename)); 421 strncpy(luts.l_release, osrelease, sizeof(luts.l_release)); 422 strncpy(luts.l_version, version, sizeof(luts.l_version)); 423 strncpy(luts.l_machine, machine, sizeof(luts.l_machine)); 424 425 /* This part taken from the the uname() in libc */ 426 len = sizeof(luts.l_version); 427 for (cp = luts.l_version; len--; ++cp) 428 if (*cp == '\n' || *cp == '\t') 429 if (len > 1) 430 *cp = ' '; 431 else 432 *cp = '\0'; 433 434 return copyout(&luts, SCARG(uap, up), sizeof(luts)); 435} 436 437int 438linux_oldolduname(p, v, retval) 439 struct proc *p; 440 void *v; 441 register_t *retval; 442{ 443 struct linux_uname_args /* { 444 syscallarg(struct linux_oldoldutsname *) up; 445 } */ *uap = v; 446 extern char ostype[], hostname[], osrelease[], version[], machine[]; 447 struct linux_oldoldutsname luts; 448 int len; 449 char *cp; 450 451 strncpy(luts.l_sysname, ostype, sizeof(luts.l_sysname)); 452 strncpy(luts.l_nodename, hostname, sizeof(luts.l_nodename)); 453 strncpy(luts.l_release, osrelease, sizeof(luts.l_release)); 454 strncpy(luts.l_version, version, sizeof(luts.l_version)); 455 strncpy(luts.l_machine, machine, sizeof(luts.l_machine)); 456 457 /* This part taken from the the uname() in libc */ 458 len = sizeof(luts.l_version); 459 for (cp = luts.l_version; len--; ++cp) 460 if (*cp == '\n' || *cp == '\t') 461 if (len > 1) 462 *cp = ' '; 463 else 464 *cp = '\0'; 465 466 return copyout(&luts, SCARG(uap, up), sizeof(luts)); 467} 468 469/* 470 * Linux wants to pass everything to a syscall in registers. However, 471 * mmap() has 6 of them. Oops: out of register error. They just pass 472 * everything in a structure. 473 */ 474int 475linux_mmap(p, v, retval) 476 struct proc *p; 477 void *v; 478 register_t *retval; 479{ 480 struct linux_mmap_args /* { 481 syscallarg(struct linux_mmap *) lmp; 482 } */ *uap = v; 483 struct linux_mmap lmap; 484 struct mmap_args cma; 485 int error, flags; 486 487 if ((error = copyin(SCARG(uap, lmp), &lmap, sizeof lmap))) 488 return error; 489 490 flags = 0; 491 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_SHARED, MAP_SHARED); 492 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_PRIVATE, MAP_PRIVATE); 493 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_FIXED, MAP_FIXED); 494 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_ANON, MAP_ANON); 495 496 SCARG(&cma,addr) = lmap.lm_addr; 497 SCARG(&cma,len) = lmap.lm_len; 498 SCARG(&cma,prot) = lmap.lm_prot; 499 SCARG(&cma,flags) = flags; 500 SCARG(&cma,fd) = lmap.lm_fd; 501 SCARG(&cma,pad) = 0; 502 SCARG(&cma,pos) = lmap.lm_pos; 503 504 return mmap(p, &cma, retval); 505} 506 507/* 508 * Linux doesn't use the retval[1] value to determine whether 509 * we are the child or parent. 510 */ 511int 512linux_fork(p, uap, retval) 513 struct proc *p; 514 void *uap; 515 register_t *retval; 516{ 517 int error; 518 519 if ((error = fork(p, uap, retval))) 520 return error; 521 522 if (retval[1] == 1) 523 retval[0] = 0; 524 525 return 0; 526} 527 528/* 529 * This code is partly stolen from src/lib/libc/compat-43/times.c 530 * XXX - CLK_TCK isn't declared in /sys, just in <time.h>, done here 531 */ 532 533#define CLK_TCK 100 534#define CONVTCK(r) (r.tv_sec * CLK_TCK + r.tv_usec / (1000000 / CLK_TCK)) 535 536int 537linux_times(p, v, retval) 538 struct proc *p; 539 void *v; 540 register_t *retval; 541{ 542 struct linux_times_args /* { 543 syscallarg(struct times *) tms; 544 } */ *uap = v; 545 struct timeval t; 546 struct linux_tms ltms; 547 struct rusage ru; 548 int error, s; 549 550 calcru(p, &ru.ru_utime, &ru.ru_stime, NULL); 551 ltms.ltms_utime = CONVTCK(ru.ru_utime); 552 ltms.ltms_stime = CONVTCK(ru.ru_stime); 553 554 ltms.ltms_cutime = CONVTCK(p->p_stats->p_cru.ru_utime); 555 ltms.ltms_cstime = CONVTCK(p->p_stats->p_cru.ru_stime); 556 557 if ((error = copyout(<ms, SCARG(uap, tms), sizeof ltms))) 558 return error; 559 560 s = splclock(); 561 timersub(&time, &boottime, &t); 562 splx(s); 563 564 retval[0] = ((linux_clock_t)(CONVTCK(t))); 565 return 0; 566} 567 568/* 569 * NetBSD passes fd[0] in retval[0], and fd[1] in retval[1]. 570 * Linux directly passes the pointer. 571 */ 572int 573linux_pipe(p, v, retval) 574 struct proc *p; 575 void *v; 576 register_t *retval; 577{ 578 struct linux_pipe_args /* { 579 syscallarg(int *) pfds; 580 } */ *uap = v; 581 int error; 582 583 if ((error = pipe(p, 0, retval))) 584 return error; 585 586 /* Assumes register_t is an int */ 587 588 if ((error = copyout(retval, SCARG(uap, pfds), 2 * sizeof (int)))) 589 return error; 590 591 retval[0] = 0; 592 return 0; 593} 594 595/* 596 * Alarm. This is a libc call which used setitimer(2) in NetBSD. 597 * Fiddle with the timers to make it work. 598 */ 599int 600linux_alarm(p, v, retval) 601 struct proc *p; 602 void *v; 603 register_t *retval; 604{ 605 struct linux_alarm_args /* { 606 syscallarg(unsigned int) secs; 607 } */ *uap = v; 608 int error, s; 609 struct itimerval *itp, it; 610 611 itp = &p->p_realtimer; 612 s = splclock(); 613 /* 614 * Clear any pending timer alarms. 615 */ 616 untimeout(realitexpire, p); 617 timerclear(&itp->it_interval); 618 if (timerisset(&itp->it_value) && 619 timercmp(&itp->it_value, &time, >)) 620 timersub(&itp->it_value, &time, &itp->it_value); 621 /* 622 * Return how many seconds were left (rounded up) 623 */ 624 retval[0] = itp->it_value.tv_sec; 625 if (itp->it_value.tv_usec) 626 retval[0]++; 627 628 /* 629 * alarm(0) just resets the timer. 630 */ 631 if (SCARG(uap, secs) == 0) { 632 timerclear(&itp->it_value); 633 splx(s); 634 return 0; 635 } 636 637 /* 638 * Check the new alarm time for sanity, and set it. 639 */ 640 timerclear(&it.it_interval); 641 it.it_value.tv_sec = SCARG(uap, secs); 642 it.it_value.tv_usec = 0; 643 if (itimerfix(&it.it_value) || itimerfix(&it.it_interval)) { 644 splx(s); 645 return (EINVAL); 646 } 647 648 if (timerisset(&it.it_value)) { 649 timeradd(&it.it_value, &time, &it.it_value); 650 timeout(realitexpire, p, hzto(&it.it_value)); 651 } 652 p->p_realtimer = it; 653 splx(s); 654 655 return 0; 656} 657 658/* 659 * utime(). Do conversion to things that utimes() understands, 660 * and pass it on. 661 */ 662int 663linux_utime(p, v, retval) 664 struct proc *p; 665 void *v; 666 register_t *retval; 667{ 668 struct linux_utime_args /* { 669 syscallarg(char *) path; 670 syscallarg(struct linux_utimbuf *)times; 671 } */ *uap = v; 672 caddr_t sg; 673 int error; 674 struct utimes_args ua; 675 struct timeval tv[2], *tvp; 676 struct linux_utimbuf lut; 677 678 sg = stackgap_init(p->p_emul); 679 LINUX_CHECK_ALT_EXIST(p, &sg, SCARG(uap, path)); 680 681 SCARG(&ua, path) = SCARG(uap, path); 682 683 if (SCARG(uap, times) != NULL) { 684 if ((error = copyin(SCARG(uap, times), &lut, sizeof lut))) 685 return error; 686 tv[0].tv_usec = tv[1].tv_usec = 0; 687 tv[0].tv_sec = lut.l_actime; 688 tv[1].tv_sec = lut.l_modtime; 689 tvp = (struct timeval *) stackgap_alloc(&sg, sizeof(tv)); 690 if ((error = copyout(tv, tvp, sizeof tv))) 691 return error; 692 SCARG(&ua, tptr) = tvp; 693 } 694 else 695 SCARG(&ua, tptr) = NULL; 696 697 return utimes(p, uap, retval); 698} 699 700/* 701 * The old Linux readdir was only able to read one entry at a time, 702 * even though it had a 'count' argument. In fact, the emulation 703 * of the old call was better than the original, because it did handle 704 * the count arg properly. Don't bother with it anymore now, and use 705 * it to distinguish between old and new. The difference is that the 706 * newer one actually does multiple entries, and the reclen field 707 * really is the reclen, not the namelength. 708 */ 709int 710linux_readdir(p, v, retval) 711 struct proc *p; 712 void *v; 713 register_t *retval; 714{ 715 struct linux_readdir_args /* { 716 syscallarg(int) fd; 717 syscallarg(struct linux_dirent *) dent; 718 syscallarg(unsigned int) count; 719 } */ *uap = v; 720 721 SCARG(uap, count) = 1; 722 return linux_getdents(p, uap, retval); 723} 724 725/* 726 * Linux 'readdir' call. This code is mostly taken from the 727 * SunOS getdents call (see compat/sunos/sunos_misc.c), though 728 * an attempt has been made to keep it a little cleaner (failing 729 * miserably, because of the cruft needed if count 1 is passed). 730 * 731 * The d_off field should contain the offset of the next valid entry, 732 * but in Linux it has the offset of the entry itself. We emulate 733 * that bug here. 734 * 735 * Read in BSD-style entries, convert them, and copy them out. 736 * 737 * Note that this doesn't handle union-mounted filesystems. 738 */ 739int 740linux_getdents(p, v, retval) 741 struct proc *p; 742 void *v; 743 register_t *retval; 744{ 745 struct linux_readdir_args /* { 746 syscallarg(int) fd; 747 syscallarg(struct linux_dirent *) dent; 748 syscallarg(unsigned int) count; 749 } */ *uap = v; 750 register struct dirent *bdp; 751 struct vnode *vp; 752 caddr_t inp, buf; /* BSD-format */ 753 int len, reclen; /* BSD-format */ 754 caddr_t outp; /* Linux-format */ 755 int resid, linuxreclen; /* Linux-format */ 756 struct file *fp; 757 struct uio auio; 758 struct iovec aiov; 759 struct linux_dirent idb; 760 off_t off; /* true file offset */ 761 linux_off_t soff; /* Linux file offset */ 762 int buflen, error, eofflag, nbytes, oldcall; 763 struct vattr va; 764 765 if ((error = getvnode(p->p_fd, SCARG(uap, fd), &fp)) != 0) 766 return (error); 767 768 if ((fp->f_flag & FREAD) == 0) 769 return (EBADF); 770 771 vp = (struct vnode *)fp->f_data; 772 773 if (vp->v_type != VDIR) /* XXX vnode readdir op should do this */ 774 return (EINVAL); 775 776 if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p))) 777 return error; 778 779 nbytes = SCARG(uap, count); 780 if (nbytes == 1) { /* emulating old, broken behaviour */ 781 nbytes = sizeof (struct linux_dirent); 782 buflen = max(va.va_blocksize, nbytes); 783 oldcall = 1; 784 } else { 785 buflen = min(MAXBSIZE, nbytes); 786 oldcall = 0; 787 } 788 buf = malloc(buflen, M_TEMP, M_WAITOK); 789 VOP_LOCK(vp); 790 off = fp->f_offset; 791again: 792 aiov.iov_base = buf; 793 aiov.iov_len = buflen; 794 auio.uio_iov = &aiov; 795 auio.uio_iovcnt = 1; 796 auio.uio_rw = UIO_READ; 797 auio.uio_segflg = UIO_SYSSPACE; 798 auio.uio_procp = p; 799 auio.uio_resid = buflen; 800 auio.uio_offset = off; 801 /* 802 * First we read into the malloc'ed buffer, then 803 * we massage it into user space, one record at a time. 804 */ 805 error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, (u_long *)0, 0); 806 if (error) 807 goto out; 808 809 inp = buf; 810 outp = (caddr_t) SCARG(uap, dent); 811 resid = nbytes; 812 if ((len = buflen - auio.uio_resid) == 0) 813 goto eof; 814 815 for (; len > 0; len -= reclen) { 816 bdp = (struct dirent *)inp; 817 reclen = bdp->d_reclen; 818 if (reclen & 3) 819 panic("linux_readdir"); 820 off += reclen; 821 if (bdp->d_fileno == 0) { 822 inp += reclen; /* it is a hole; squish it out */ 823 continue; 824 } 825 linuxreclen = LINUX_RECLEN(&idb, bdp->d_namlen); 826 if (reclen > len || resid < linuxreclen) { 827 /* entry too big for buffer, so just stop */ 828 outp++; 829 break; 830 } 831 /* 832 * Massage in place to make a Linux-shaped dirent (otherwise 833 * we have to worry about touching user memory outside of 834 * the copyout() call). 835 */ 836 idb.d_ino = (long)bdp->d_fileno; 837 idb.d_off = off - reclen; 838 /* 839 * The old readdir() call used the reclen field as namlen. 840 */ 841 idb.d_reclen = oldcall ? (u_short)bdp->d_namlen : linuxreclen; 842 strcpy(idb.d_name, bdp->d_name); 843 if ((error = copyout((caddr_t)&idb, outp, linuxreclen))) 844 goto out; 845 /* advance past this real entry */ 846 inp += reclen; 847 /* advance output past Linux-shaped entry */ 848 outp += linuxreclen; 849 resid -= linuxreclen; 850 if (oldcall) 851 break; 852 } 853 854 /* if we squished out the whole block, try again */ 855 if (outp == (caddr_t) SCARG(uap, dent)) 856 goto again; 857 fp->f_offset = off; /* update the vnode offset */ 858 859 if (oldcall) 860 nbytes = resid + linuxreclen; 861 862eof: 863 *retval = nbytes - resid; 864out: 865 VOP_UNLOCK(vp); 866 free(buf, M_TEMP); 867 return error; 868} 869 870/* 871 * Not sure why the arguments to this older version of select() were put 872 * into a structure, because there are 5, and that can all be handled 873 * in registers on the i386 like Linux wants to. 874 */ 875int 876linux_oldselect(p, v, retval) 877 struct proc *p; 878 void *v; 879 register_t *retval; 880{ 881 struct linux_oldselect_args /* { 882 syscallarg(struct linux_select *) lsp; 883 } */ *uap = v; 884 struct linux_select ls; 885 int error; 886 887 if ((error = copyin(SCARG(uap, lsp), &ls, sizeof(ls)))) 888 return error; 889 890 return linux_select1(p, retval, ls.nfds, ls.readfds, ls.writefds, 891 ls.exceptfds, ls.timeout); 892} 893 894/* 895 * Even when just using registers to pass arguments to syscalls you can 896 * have 5 of them on the i386. So this newer version of select() does 897 * this. 898 */ 899int 900linux_select(p, v, retval) 901 struct proc *p; 902 void *v; 903 register_t *retval; 904{ 905 struct linux_select_args /* { 906 syscallarg(int) nfds; 907 syscallarg(fd_set *) readfds; 908 syscallarg(fd_set *) writefds; 909 syscallarg(fd_set *) exceptfds; 910 syscallarg(struct timeval *) timeout; 911 } */ *uap = v; 912 913 return linux_select1(p, retval, SCARG(uap, nfds), SCARG(uap, readfds), 914 SCARG(uap, writefds), SCARG(uap, exceptfds), SCARG(uap, timeout)); 915} 916 917/* 918 * Common code for the old and new versions of select(). A couple of 919 * things are important: 920 * 1) return the amount of time left in the 'timeout' parameter 921 * 2) select never returns ERESTART on Linux, always return EINTR 922 */ 923int 924linux_select1(p, retval, nfds, readfds, writefds, exceptfds, timeout) 925 struct proc *p; 926 register_t *retval; 927 int nfds; 928 fd_set *readfds, *writefds, *exceptfds; 929 struct timeval *timeout; 930{ 931 struct select_args bsa; 932 struct timeval tv0, tv1, utv, *tvp; 933 caddr_t sg; 934 int error; 935 936 SCARG(&bsa, nd) = nfds; 937 SCARG(&bsa, in) = readfds; 938 SCARG(&bsa, ou) = writefds; 939 SCARG(&bsa, ex) = exceptfds; 940 SCARG(&bsa, tv) = timeout; 941 942 /* 943 * Store current time for computation of the amount of 944 * time left. 945 */ 946 if (timeout) { 947 if ((error = copyin(timeout, &utv, sizeof(utv)))) 948 return error; 949 if (itimerfix(&utv)) { 950 /* 951 * The timeval was invalid. Convert it to something 952 * valid that will act as it does under Linux. 953 */ 954 sg = stackgap_init(p->p_emul); 955 tvp = stackgap_alloc(&sg, sizeof(utv)); 956 utv.tv_sec += utv.tv_usec / 1000000; 957 utv.tv_usec %= 1000000; 958 if (utv.tv_usec < 0) { 959 utv.tv_sec -= 1; 960 utv.tv_usec += 1000000; 961 } 962 if (utv.tv_sec < 0) 963 timerclear(&utv); 964 if ((error = copyout(&utv, tvp, sizeof(utv)))) 965 return error; 966 SCARG(&bsa, tv) = tvp; 967 } 968 microtime(&tv0); 969 } 970 971 error = select(p, &bsa, retval); 972 if (error) { 973 /* 974 * See fs/select.c in the Linux kernel. Without this, 975 * Maelstrom doesn't work. 976 */ 977 if (error == ERESTART) 978 error = EINTR; 979 return error; 980 } 981 982 if (timeout) { 983 if (*retval) { 984 /* 985 * Compute how much time was left of the timeout, 986 * by subtracting the current time and the time 987 * before we started the call, and subtracting 988 * that result from the user-supplied value. 989 */ 990 microtime(&tv1); 991 timersub(&tv1, &tv0, &tv1); 992 timersub(&utv, &tv1, &utv); 993 if (utv.tv_sec < 0) 994 timerclear(&utv); 995 } else 996 timerclear(&utv); 997 if ((error = copyout(&utv, timeout, sizeof(utv)))) 998 return error; 999 } 1000 1001 return 0; 1002} 1003 1004/* 1005 * Get the process group of a certain process. Look it up 1006 * and return the value. 1007 */ 1008int 1009linux_getpgid(p, v, retval) 1010 struct proc *p; 1011 void *v; 1012 register_t *retval; 1013{ 1014 struct linux_getpgid_args /* { 1015 syscallarg(int) pid; 1016 } */ *uap = v; 1017 struct proc *targp; 1018 1019 if (SCARG(uap, pid) != 0 && SCARG(uap, pid) != p->p_pid) 1020 if ((targp = pfind(SCARG(uap, pid))) == 0) 1021 return ESRCH; 1022 else 1023 targp = p; 1024 1025 retval[0] = targp->p_pgid; 1026 return 0; 1027} 1028 1029/* 1030 * Set the 'personality' (emulation mode) for the current process. Only 1031 * accept the Linux personality here (0). This call is needed because 1032 * the Linux ELF crt0 issues it in an ugly kludge to make sure that 1033 * ELF binaries run in Linux mode, not SVR4 mode. 1034 */ 1035int 1036linux_personality(p, v, retval) 1037 struct proc *p; 1038 void *v; 1039 register_t *retval; 1040{ 1041 struct linux_personality_args /* { 1042 syscallarg(int) per; 1043 } */ *uap = v; 1044 1045 if (SCARG(uap, per) != 0) 1046 return EINVAL; 1047 retval[0] = 0; 1048 return 0; 1049} 1050 1051/* 1052 * The calls are here because of type conversions. 1053 */ 1054int 1055linux_setreuid(p, v, retval) 1056 struct proc *p; 1057 void *v; 1058 register_t *retval; 1059{ 1060 struct linux_setreuid_args /* { 1061 syscallarg(int) ruid; 1062 syscallarg(int) euid; 1063 } */ *uap = v; 1064 struct compat_43_setreuid_args bsa; 1065 1066 SCARG(&bsa, ruid) = ((linux_uid_t)SCARG(uap, ruid) == (linux_uid_t)-1) ? 1067 (uid_t)-1 : SCARG(uap, ruid); 1068 SCARG(&bsa, euid) = ((linux_uid_t)SCARG(uap, euid) == (linux_uid_t)-1) ? 1069 (uid_t)-1 : SCARG(uap, euid); 1070 1071 return compat_43_setreuid(p, &bsa, retval); 1072} 1073 1074int 1075linux_setregid(p, v, retval) 1076 struct proc *p; 1077 void *v; 1078 register_t *retval; 1079{ 1080 struct linux_setregid_args /* { 1081 syscallarg(int) rgid; 1082 syscallarg(int) egid; 1083 } */ *uap = v; 1084 struct compat_43_setregid_args bsa; 1085 1086 SCARG(&bsa, rgid) = ((linux_gid_t)SCARG(uap, rgid) == (linux_gid_t)-1) ? 1087 (uid_t)-1 : SCARG(uap, rgid); 1088 SCARG(&bsa, egid) = ((linux_gid_t)SCARG(uap, egid) == (linux_gid_t)-1) ? 1089 (uid_t)-1 : SCARG(uap, egid); 1090 1091 return compat_43_setregid(p, &bsa, retval); 1092} 1093