linux_pipe.c revision 1.27
1/* $NetBSD: linux_pipe.c,v 1.27 1996/05/20 01:59:21 fvdl 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/* linux_misc.c */ 80static void bsd_to_linux_wstat __P((int *)); 81static void bsd_to_linux_statfs __P((struct statfs *, struct linux_statfs *)); 82int linux_select1 __P((struct proc *, register_t *, int, fd_set *, fd_set *, 83 fd_set *, struct timeval *)); 84 85/* 86 * The information on a terminated (or stopped) process needs 87 * to be converted in order for Linux binaries to get a valid signal 88 * number out of it. 89 */ 90static void 91bsd_to_linux_wstat(status) 92 int *status; 93{ 94 95 if (WIFSIGNALED(*status)) 96 *status = (*status & ~0177) | 97 bsd_to_linux_sig[WTERMSIG(*status)]; 98 else if (WIFSTOPPED(*status)) 99 *status = (*status & ~0xff00) | 100 (bsd_to_linux_sig[WSTOPSIG(*status)] << 8); 101} 102 103/* 104 * waitpid(2). Passed on to the NetBSD call, surrounded by code to 105 * reserve some space for a NetBSD-style wait status, and converting 106 * it to what Linux wants. 107 */ 108int 109linux_sys_waitpid(p, v, retval) 110 struct proc *p; 111 void *v; 112 register_t *retval; 113{ 114 struct linux_sys_waitpid_args /* { 115 syscallarg(int) pid; 116 syscallarg(int *) status; 117 syscallarg(int) options; 118 } */ *uap = v; 119 struct sys_wait4_args w4a; 120 int error, *status, tstat; 121 caddr_t sg; 122 123 if (SCARG(uap, status) != NULL) { 124 sg = stackgap_init(p->p_emul); 125 status = (int *) stackgap_alloc(&sg, sizeof status); 126 } else 127 status = NULL; 128 129 SCARG(&w4a, pid) = SCARG(uap, pid); 130 SCARG(&w4a, status) = status; 131 SCARG(&w4a, options) = SCARG(uap, options); 132 SCARG(&w4a, rusage) = NULL; 133 134 if ((error = sys_wait4(p, &w4a, retval))) 135 return error; 136 137 p->p_siglist &= ~sigmask(SIGCHLD); 138 139 if (status != NULL) { 140 if ((error = copyin(status, &tstat, sizeof tstat))) 141 return error; 142 143 bsd_to_linux_wstat(&tstat); 144 return copyout(&tstat, SCARG(uap, status), sizeof tstat); 145 } 146 147 return 0; 148} 149 150/* 151 * This is very much the same as waitpid() 152 */ 153int 154linux_sys_wait4(p, v, retval) 155 struct proc *p; 156 void *v; 157 register_t *retval; 158{ 159 struct linux_sys_wait4_args /* { 160 syscallarg(int) pid; 161 syscallarg(int *) status; 162 syscallarg(int) options; 163 syscallarg(struct rusage *) rusage; 164 } */ *uap = v; 165 struct sys_wait4_args w4a; 166 int error, *status, tstat; 167 caddr_t sg; 168 169 if (SCARG(uap, status) != NULL) { 170 sg = stackgap_init(p->p_emul); 171 status = (int *) stackgap_alloc(&sg, sizeof status); 172 } else 173 status = NULL; 174 175 SCARG(&w4a, pid) = SCARG(uap, pid); 176 SCARG(&w4a, status) = status; 177 SCARG(&w4a, options) = SCARG(uap, options); 178 SCARG(&w4a, rusage) = SCARG(uap, rusage); 179 180 if ((error = sys_wait4(p, &w4a, retval))) 181 return error; 182 183 p->p_siglist &= ~sigmask(SIGCHLD); 184 185 if (status != NULL) { 186 if ((error = copyin(status, &tstat, sizeof tstat))) 187 return error; 188 189 bsd_to_linux_wstat(&tstat); 190 return copyout(&tstat, SCARG(uap, status), sizeof tstat); 191 } 192 193 return 0; 194} 195 196/* 197 * This is the old brk(2) call. I don't think anything in the Linux 198 * world uses this anymore 199 */ 200int 201linux_sys_break(p, v, retval) 202 struct proc *p; 203 void *v; 204 register_t *retval; 205{ 206#if 0 207 struct linux_sys_brk_args /* { 208 syscallarg(char *) nsize; 209 } */ *uap = v; 210#endif 211 212 return ENOSYS; 213} 214 215/* 216 * Linux brk(2). The check if the new address is >= the old one is 217 * done in the kernel in Linux. NetBSD does it in the library. 218 */ 219int 220linux_sys_brk(p, v, retval) 221 struct proc *p; 222 void *v; 223 register_t *retval; 224{ 225 struct linux_sys_brk_args /* { 226 syscallarg(char *) nsize; 227 } */ *uap = v; 228 char *nbrk = SCARG(uap, nsize); 229 struct sys_obreak_args oba; 230 struct vmspace *vm = p->p_vmspace; 231 caddr_t oldbrk; 232 233 oldbrk = vm->vm_daddr + ctob(vm->vm_dsize); 234 /* 235 * XXX inconsistent.. Linux always returns at least the old 236 * brk value, but it will be page-aligned if this fails, 237 * and possibly not page aligned if it succeeds (the user 238 * supplied pointer is returned). 239 */ 240 SCARG(&oba, nsize) = nbrk; 241 242 if ((caddr_t) nbrk > vm->vm_daddr && sys_obreak(p, &oba, retval) == 0) 243 retval[0] = (register_t)nbrk; 244 else 245 retval[0] = (register_t)oldbrk; 246 247 return 0; 248} 249 250/* 251 * I wonder why Linux has gettimeofday() _and_ time().. Still, we 252 * need to deal with it. 253 */ 254int 255linux_sys_time(p, v, retval) 256 struct proc *p; 257 void *v; 258 register_t *retval; 259{ 260 struct linux_sys_time_args /* { 261 linux_time_t *t; 262 } */ *uap = v; 263 struct timeval atv; 264 linux_time_t tt; 265 int error; 266 267 microtime(&atv); 268 269 tt = atv.tv_sec; 270 if (SCARG(uap, t) && (error = copyout(&tt, SCARG(uap, t), sizeof tt))) 271 return error; 272 273 retval[0] = tt; 274 return 0; 275} 276 277/* 278 * Convert BSD statfs structure to Linux statfs structure. 279 * The Linux structure has less fields, and it also wants 280 * the length of a name in a dir entry in a field, which 281 * we fake (probably the wrong way). 282 */ 283static void 284bsd_to_linux_statfs(bsp, lsp) 285 struct statfs *bsp; 286 struct linux_statfs *lsp; 287{ 288 289 lsp->l_ftype = bsp->f_type; 290 lsp->l_fbsize = bsp->f_bsize; 291 lsp->l_fblocks = bsp->f_blocks; 292 lsp->l_fbfree = bsp->f_bfree; 293 lsp->l_fbavail = bsp->f_bavail; 294 lsp->l_ffiles = bsp->f_files; 295 lsp->l_fffree = bsp->f_ffree; 296 lsp->l_ffsid.val[0] = bsp->f_fsid.val[0]; 297 lsp->l_ffsid.val[1] = bsp->f_fsid.val[1]; 298 lsp->l_fnamelen = MAXNAMLEN; /* XXX */ 299} 300 301/* 302 * Implement the fs stat functions. Straightforward. 303 */ 304int 305linux_sys_statfs(p, v, retval) 306 struct proc *p; 307 void *v; 308 register_t *retval; 309{ 310 struct linux_sys_statfs_args /* { 311 syscallarg(char *) path; 312 syscallarg(struct linux_statfs *) sp; 313 } */ *uap = v; 314 struct statfs btmp, *bsp; 315 struct linux_statfs ltmp; 316 struct sys_statfs_args bsa; 317 caddr_t sg; 318 int error; 319 320 sg = stackgap_init(p->p_emul); 321 bsp = (struct statfs *) stackgap_alloc(&sg, sizeof (struct statfs)); 322 323 LINUX_CHECK_ALT_EXIST(p, &sg, SCARG(uap, path)); 324 325 SCARG(&bsa, path) = SCARG(uap, path); 326 SCARG(&bsa, buf) = bsp; 327 328 if ((error = sys_statfs(p, &bsa, retval))) 329 return error; 330 331 if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp))) 332 return error; 333 334 bsd_to_linux_statfs(&btmp, <mp); 335 336 return copyout((caddr_t) <mp, (caddr_t) SCARG(uap, sp), sizeof ltmp); 337} 338 339int 340linux_sys_fstatfs(p, v, retval) 341 struct proc *p; 342 void *v; 343 register_t *retval; 344{ 345 struct linux_sys_fstatfs_args /* { 346 syscallarg(int) fd; 347 syscallarg(struct linux_statfs *) sp; 348 } */ *uap = v; 349 struct statfs btmp, *bsp; 350 struct linux_statfs ltmp; 351 struct sys_fstatfs_args bsa; 352 caddr_t sg; 353 int error; 354 355 sg = stackgap_init(p->p_emul); 356 bsp = (struct statfs *) stackgap_alloc(&sg, sizeof (struct statfs)); 357 358 SCARG(&bsa, fd) = SCARG(uap, fd); 359 SCARG(&bsa, buf) = bsp; 360 361 if ((error = sys_fstatfs(p, &bsa, retval))) 362 return error; 363 364 if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp))) 365 return error; 366 367 bsd_to_linux_statfs(&btmp, <mp); 368 369 return copyout((caddr_t) <mp, (caddr_t) SCARG(uap, sp), sizeof ltmp); 370} 371 372/* 373 * uname(). Just copy the info from the various strings stored in the 374 * kernel, and put it in the Linux utsname structure. That structure 375 * is almost the same as the NetBSD one, only it has fields 65 characters 376 * long, and an extra domainname field. 377 */ 378int 379linux_sys_uname(p, v, retval) 380 struct proc *p; 381 void *v; 382 register_t *retval; 383{ 384 struct linux_sys_uname_args /* { 385 syscallarg(struct linux_utsname *) up; 386 } */ *uap = v; 387 extern char ostype[], hostname[], osrelease[], version[], machine[], 388 domainname[]; 389 struct linux_utsname luts; 390 int len; 391 char *cp; 392 393 strncpy(luts.l_sysname, ostype, sizeof(luts.l_sysname)); 394 strncpy(luts.l_nodename, hostname, sizeof(luts.l_nodename)); 395 strncpy(luts.l_release, osrelease, sizeof(luts.l_release)); 396 strncpy(luts.l_version, version, sizeof(luts.l_version)); 397 strncpy(luts.l_machine, machine, sizeof(luts.l_machine)); 398 strncpy(luts.l_domainname, domainname, sizeof(luts.l_domainname)); 399 400 /* This part taken from the the uname() in libc */ 401 len = sizeof(luts.l_version); 402 for (cp = luts.l_version; len--; ++cp) 403 if (*cp == '\n' || *cp == '\t') 404 if (len > 1) 405 *cp = ' '; 406 else 407 *cp = '\0'; 408 409 return copyout(&luts, SCARG(uap, up), sizeof(luts)); 410} 411 412int 413linux_sys_olduname(p, v, retval) 414 struct proc *p; 415 void *v; 416 register_t *retval; 417{ 418 struct linux_sys_uname_args /* { 419 syscallarg(struct linux_oldutsname *) up; 420 } */ *uap = v; 421 extern char ostype[], hostname[], osrelease[], version[], machine[]; 422 struct linux_oldutsname luts; 423 int len; 424 char *cp; 425 426 strncpy(luts.l_sysname, ostype, sizeof(luts.l_sysname)); 427 strncpy(luts.l_nodename, hostname, sizeof(luts.l_nodename)); 428 strncpy(luts.l_release, osrelease, sizeof(luts.l_release)); 429 strncpy(luts.l_version, version, sizeof(luts.l_version)); 430 strncpy(luts.l_machine, machine, sizeof(luts.l_machine)); 431 432 /* This part taken from the the uname() in libc */ 433 len = sizeof(luts.l_version); 434 for (cp = luts.l_version; len--; ++cp) 435 if (*cp == '\n' || *cp == '\t') 436 if (len > 1) 437 *cp = ' '; 438 else 439 *cp = '\0'; 440 441 return copyout(&luts, SCARG(uap, up), sizeof(luts)); 442} 443 444int 445linux_sys_oldolduname(p, v, retval) 446 struct proc *p; 447 void *v; 448 register_t *retval; 449{ 450 struct linux_sys_uname_args /* { 451 syscallarg(struct linux_oldoldutsname *) up; 452 } */ *uap = v; 453 extern char ostype[], hostname[], osrelease[], version[], machine[]; 454 struct linux_oldoldutsname luts; 455 int len; 456 char *cp; 457 458 strncpy(luts.l_sysname, ostype, sizeof(luts.l_sysname)); 459 strncpy(luts.l_nodename, hostname, sizeof(luts.l_nodename)); 460 strncpy(luts.l_release, osrelease, sizeof(luts.l_release)); 461 strncpy(luts.l_version, version, sizeof(luts.l_version)); 462 strncpy(luts.l_machine, machine, sizeof(luts.l_machine)); 463 464 /* This part taken from the the uname() in libc */ 465 len = sizeof(luts.l_version); 466 for (cp = luts.l_version; len--; ++cp) 467 if (*cp == '\n' || *cp == '\t') 468 if (len > 1) 469 *cp = ' '; 470 else 471 *cp = '\0'; 472 473 return copyout(&luts, SCARG(uap, up), sizeof(luts)); 474} 475 476/* 477 * Linux wants to pass everything to a syscall in registers. However, 478 * mmap() has 6 of them. Oops: out of register error. They just pass 479 * everything in a structure. 480 */ 481int 482linux_sys_mmap(p, v, retval) 483 struct proc *p; 484 void *v; 485 register_t *retval; 486{ 487 struct linux_sys_mmap_args /* { 488 syscallarg(struct linux_mmap *) lmp; 489 } */ *uap = v; 490 struct linux_mmap lmap; 491 struct sys_mmap_args cma; 492 int error, flags; 493 494 if ((error = copyin(SCARG(uap, lmp), &lmap, sizeof lmap))) 495 return error; 496 497 flags = 0; 498 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_SHARED, MAP_SHARED); 499 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_PRIVATE, MAP_PRIVATE); 500 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_FIXED, MAP_FIXED); 501 flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_ANON, MAP_ANON); 502 503 SCARG(&cma,addr) = lmap.lm_addr; 504 SCARG(&cma,len) = lmap.lm_len; 505 SCARG(&cma,prot) = lmap.lm_prot; 506 SCARG(&cma,flags) = flags; 507 SCARG(&cma,fd) = lmap.lm_fd; 508 SCARG(&cma,pad) = 0; 509 SCARG(&cma,pos) = lmap.lm_pos; 510 511 return sys_mmap(p, &cma, retval); 512} 513 514int 515linux_sys_msync(p, v, retval) 516 struct proc *p; 517 void *v; 518 register_t *retval; 519{ 520 struct linux_sys_msync_args /* { 521 syscallarg(caddr_t) addr; 522 syscallarg(int) len; 523 syscallarg(int) fl; 524 } */ *uap = v; 525 526 struct sys_msync_args bma; 527 528 /* flags are ignored */ 529 SCARG(&bma, addr) = SCARG(uap, addr); 530 SCARG(&bma, len) = SCARG(uap, len); 531 532 return sys_msync(p, &bma, retval); 533} 534 535/* 536 * This code is partly stolen from src/lib/libc/compat-43/times.c 537 * XXX - CLK_TCK isn't declared in /sys, just in <time.h>, done here 538 */ 539 540#define CLK_TCK 100 541#define CONVTCK(r) (r.tv_sec * CLK_TCK + r.tv_usec / (1000000 / CLK_TCK)) 542 543int 544linux_sys_times(p, v, retval) 545 struct proc *p; 546 void *v; 547 register_t *retval; 548{ 549 struct linux_sys_times_args /* { 550 syscallarg(struct times *) tms; 551 } */ *uap = v; 552 struct timeval t; 553 struct linux_tms ltms; 554 struct rusage ru; 555 int error, s; 556 557 calcru(p, &ru.ru_utime, &ru.ru_stime, NULL); 558 ltms.ltms_utime = CONVTCK(ru.ru_utime); 559 ltms.ltms_stime = CONVTCK(ru.ru_stime); 560 561 ltms.ltms_cutime = CONVTCK(p->p_stats->p_cru.ru_utime); 562 ltms.ltms_cstime = CONVTCK(p->p_stats->p_cru.ru_stime); 563 564 if ((error = copyout(<ms, SCARG(uap, tms), sizeof ltms))) 565 return error; 566 567 s = splclock(); 568 timersub(&time, &boottime, &t); 569 splx(s); 570 571 retval[0] = ((linux_clock_t)(CONVTCK(t))); 572 return 0; 573} 574 575/* 576 * NetBSD passes fd[0] in retval[0], and fd[1] in retval[1]. 577 * Linux directly passes the pointer. 578 */ 579int 580linux_sys_pipe(p, v, retval) 581 struct proc *p; 582 void *v; 583 register_t *retval; 584{ 585 struct linux_sys_pipe_args /* { 586 syscallarg(int *) pfds; 587 } */ *uap = v; 588 int error; 589 590 if ((error = sys_pipe(p, 0, retval))) 591 return error; 592 593 /* Assumes register_t is an int */ 594 595 if ((error = copyout(retval, SCARG(uap, pfds), 2 * sizeof (int)))) 596 return error; 597 598 retval[0] = 0; 599 return 0; 600} 601 602/* 603 * Alarm. This is a libc call which uses setitimer(2) in NetBSD. 604 * Fiddle with the timers to make it work. 605 */ 606int 607linux_sys_alarm(p, v, retval) 608 struct proc *p; 609 void *v; 610 register_t *retval; 611{ 612 struct linux_sys_alarm_args /* { 613 syscallarg(unsigned int) secs; 614 } */ *uap = v; 615 int s; 616 struct itimerval *itp, it; 617 618 itp = &p->p_realtimer; 619 s = splclock(); 620 /* 621 * Clear any pending timer alarms. 622 */ 623 untimeout(realitexpire, p); 624 timerclear(&itp->it_interval); 625 if (timerisset(&itp->it_value) && 626 timercmp(&itp->it_value, &time, >)) 627 timersub(&itp->it_value, &time, &itp->it_value); 628 /* 629 * Return how many seconds were left (rounded up) 630 */ 631 retval[0] = itp->it_value.tv_sec; 632 if (itp->it_value.tv_usec) 633 retval[0]++; 634 635 /* 636 * alarm(0) just resets the timer. 637 */ 638 if (SCARG(uap, secs) == 0) { 639 timerclear(&itp->it_value); 640 splx(s); 641 return 0; 642 } 643 644 /* 645 * Check the new alarm time for sanity, and set it. 646 */ 647 timerclear(&it.it_interval); 648 it.it_value.tv_sec = SCARG(uap, secs); 649 it.it_value.tv_usec = 0; 650 if (itimerfix(&it.it_value) || itimerfix(&it.it_interval)) { 651 splx(s); 652 return (EINVAL); 653 } 654 655 if (timerisset(&it.it_value)) { 656 timeradd(&it.it_value, &time, &it.it_value); 657 timeout(realitexpire, p, hzto(&it.it_value)); 658 } 659 p->p_realtimer = it; 660 splx(s); 661 662 return 0; 663} 664 665/* 666 * utime(). Do conversion to things that utimes() understands, 667 * and pass it on. 668 */ 669int 670linux_sys_utime(p, v, retval) 671 struct proc *p; 672 void *v; 673 register_t *retval; 674{ 675 struct linux_sys_utime_args /* { 676 syscallarg(char *) path; 677 syscallarg(struct linux_utimbuf *)times; 678 } */ *uap = v; 679 caddr_t sg; 680 int error; 681 struct sys_utimes_args ua; 682 struct timeval tv[2], *tvp; 683 struct linux_utimbuf lut; 684 685 sg = stackgap_init(p->p_emul); 686 LINUX_CHECK_ALT_EXIST(p, &sg, SCARG(uap, path)); 687 688 SCARG(&ua, path) = SCARG(uap, path); 689 690 if (SCARG(uap, times) != NULL) { 691 if ((error = copyin(SCARG(uap, times), &lut, sizeof lut))) 692 return error; 693 tv[0].tv_usec = tv[1].tv_usec = 0; 694 tv[0].tv_sec = lut.l_actime; 695 tv[1].tv_sec = lut.l_modtime; 696 tvp = (struct timeval *) stackgap_alloc(&sg, sizeof(tv)); 697 if ((error = copyout(tv, tvp, sizeof tv))) 698 return error; 699 SCARG(&ua, tptr) = tvp; 700 } 701 else 702 SCARG(&ua, tptr) = NULL; 703 704 return sys_utimes(p, uap, retval); 705} 706 707/* 708 * The old Linux readdir was only able to read one entry at a time, 709 * even though it had a 'count' argument. In fact, the emulation 710 * of the old call was better than the original, because it did handle 711 * the count arg properly. Don't bother with it anymore now, and use 712 * it to distinguish between old and new. The difference is that the 713 * newer one actually does multiple entries, and the reclen field 714 * really is the reclen, not the namelength. 715 */ 716int 717linux_sys_readdir(p, v, retval) 718 struct proc *p; 719 void *v; 720 register_t *retval; 721{ 722 struct linux_sys_readdir_args /* { 723 syscallarg(int) fd; 724 syscallarg(struct linux_dirent *) dent; 725 syscallarg(unsigned int) count; 726 } */ *uap = v; 727 728 SCARG(uap, count) = 1; 729 return linux_sys_getdents(p, uap, retval); 730} 731 732/* 733 * Linux 'readdir' call. This code is mostly taken from the 734 * SunOS getdents call (see compat/sunos/sunos_misc.c), though 735 * an attempt has been made to keep it a little cleaner (failing 736 * miserably, because of the cruft needed if count 1 is passed). 737 * 738 * The d_off field should contain the offset of the next valid entry, 739 * but in Linux it has the offset of the entry itself. We emulate 740 * that bug here. 741 * 742 * Read in BSD-style entries, convert them, and copy them out. 743 * 744 * Note that this doesn't handle union-mounted filesystems. 745 */ 746int 747linux_sys_getdents(p, v, retval) 748 struct proc *p; 749 void *v; 750 register_t *retval; 751{ 752 struct linux_sys_readdir_args /* { 753 syscallarg(int) fd; 754 syscallarg(caddr_t) dent; 755 syscallarg(unsigned int) count; 756 } */ *uap = v; 757 register struct dirent *bdp; 758 struct vnode *vp; 759 caddr_t inp, buf; /* BSD-format */ 760 int len, reclen; /* BSD-format */ 761 caddr_t outp; /* Linux-format */ 762 int resid, linux_reclen = 0; /* Linux-format */ 763 struct file *fp; 764 struct uio auio; 765 struct iovec aiov; 766 struct linux_dirent idb; 767 off_t off; /* true file offset */ 768 int buflen, error, eofflag, nbytes, oldcall; 769 struct vattr va; 770 u_long *cookiebuf, *cookie; 771 int ncookies; 772 773 if ((error = getvnode(p->p_fd, SCARG(uap, fd), &fp)) != 0) 774 return (error); 775 776 if ((fp->f_flag & FREAD) == 0) 777 return (EBADF); 778 779 vp = (struct vnode *)fp->f_data; 780 781 if (vp->v_type != VDIR) /* XXX vnode readdir op should do this */ 782 return (EINVAL); 783 784 if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p))) 785 return error; 786 787 nbytes = SCARG(uap, count); 788 if (nbytes == 1) { /* emulating old, broken behaviour */ 789 nbytes = sizeof (struct linux_dirent); 790 buflen = max(va.va_blocksize, nbytes); 791 oldcall = 1; 792 } else { 793 buflen = min(MAXBSIZE, nbytes); 794 oldcall = 0; 795 } 796 buf = malloc(buflen, M_TEMP, M_WAITOK); 797 ncookies = buflen / 16; 798 cookiebuf = malloc(ncookies * sizeof(*cookiebuf), M_TEMP, M_WAITOK); 799 VOP_LOCK(vp); 800 off = fp->f_offset; 801again: 802 aiov.iov_base = buf; 803 aiov.iov_len = buflen; 804 auio.uio_iov = &aiov; 805 auio.uio_iovcnt = 1; 806 auio.uio_rw = UIO_READ; 807 auio.uio_segflg = UIO_SYSSPACE; 808 auio.uio_procp = p; 809 auio.uio_resid = buflen; 810 auio.uio_offset = off; 811 /* 812 * First we read into the malloc'ed buffer, then 813 * we massage it into user space, one record at a time. 814 */ 815 error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, cookiebuf, 816 ncookies); 817 if (error) 818 goto out; 819 820 inp = buf; 821 outp = SCARG(uap, dent); 822 resid = nbytes; 823 if ((len = buflen - auio.uio_resid) == 0) 824 goto eof; 825 826 for (cookie = cookiebuf; len > 0; len -= reclen) { 827 bdp = (struct dirent *)inp; 828 reclen = bdp->d_reclen; 829 if (reclen & 3) 830 panic("linux_readdir"); 831 if (bdp->d_fileno == 0) { 832 inp += reclen; /* it is a hole; squish it out */ 833 off = *cookie++; 834 continue; 835 } 836 linux_reclen = LINUX_RECLEN(&idb, bdp->d_namlen); 837 if (reclen > len || resid < linux_reclen) { 838 /* entry too big for buffer, so just stop */ 839 outp++; 840 off = *cookie++; 841 break; 842 } 843 /* 844 * Massage in place to make a Linux-shaped dirent (otherwise 845 * we have to worry about touching user memory outside of 846 * the copyout() call). 847 */ 848 idb.d_ino = (linux_ino_t)bdp->d_fileno; 849 /* 850 * The old readdir() call misuses the offset and reclen fields. 851 */ 852 if (oldcall) { 853 idb.d_off = (linux_off_t)linux_reclen; 854 idb.d_reclen = (u_short)bdp->d_namlen; 855 } else { 856 idb.d_off = (linux_off_t)off; 857 idb.d_reclen = (u_short)linux_reclen; 858 } 859 strcpy(idb.d_name, bdp->d_name); 860 if ((error = copyout((caddr_t)&idb, outp, linux_reclen))) 861 goto out; 862 /* advance past this real entry */ 863 inp += reclen; 864 off = *cookie++; /* each entry points to itself */ 865 /* advance output past Linux-shaped entry */ 866 outp += linux_reclen; 867 resid -= linux_reclen; 868 if (oldcall) 869 break; 870 } 871 872 /* if we squished out the whole block, try again */ 873 if (outp == SCARG(uap, dent)) 874 goto again; 875 fp->f_offset = off; /* update the vnode offset */ 876 877 if (oldcall) 878 nbytes = resid + linux_reclen; 879 880eof: 881 *retval = nbytes - resid; 882out: 883 VOP_UNLOCK(vp); 884 free(cookiebuf, M_TEMP); 885 free(buf, M_TEMP); 886 return error; 887} 888 889/* 890 * Not sure why the arguments to this older version of select() were put 891 * into a structure, because there are 5, and that can all be handled 892 * in registers on the i386 like Linux wants to. 893 */ 894int 895linux_sys_oldselect(p, v, retval) 896 struct proc *p; 897 void *v; 898 register_t *retval; 899{ 900 struct linux_sys_oldselect_args /* { 901 syscallarg(struct linux_select *) lsp; 902 } */ *uap = v; 903 struct linux_select ls; 904 int error; 905 906 if ((error = copyin(SCARG(uap, lsp), &ls, sizeof(ls)))) 907 return error; 908 909 return linux_select1(p, retval, ls.nfds, ls.readfds, ls.writefds, 910 ls.exceptfds, ls.timeout); 911} 912 913/* 914 * Even when just using registers to pass arguments to syscalls you can 915 * have 5 of them on the i386. So this newer version of select() does 916 * this. 917 */ 918int 919linux_sys_select(p, v, retval) 920 struct proc *p; 921 void *v; 922 register_t *retval; 923{ 924 struct linux_sys_select_args /* { 925 syscallarg(int) nfds; 926 syscallarg(fd_set *) readfds; 927 syscallarg(fd_set *) writefds; 928 syscallarg(fd_set *) exceptfds; 929 syscallarg(struct timeval *) timeout; 930 } */ *uap = v; 931 932 return linux_select1(p, retval, SCARG(uap, nfds), SCARG(uap, readfds), 933 SCARG(uap, writefds), SCARG(uap, exceptfds), SCARG(uap, timeout)); 934} 935 936/* 937 * Common code for the old and new versions of select(). A couple of 938 * things are important: 939 * 1) return the amount of time left in the 'timeout' parameter 940 * 2) select never returns ERESTART on Linux, always return EINTR 941 */ 942int 943linux_select1(p, retval, nfds, readfds, writefds, exceptfds, timeout) 944 struct proc *p; 945 register_t *retval; 946 int nfds; 947 fd_set *readfds, *writefds, *exceptfds; 948 struct timeval *timeout; 949{ 950 struct sys_select_args bsa; 951 struct timeval tv0, tv1, utv, *tvp; 952 caddr_t sg; 953 int error; 954 955 SCARG(&bsa, nd) = nfds; 956 SCARG(&bsa, in) = readfds; 957 SCARG(&bsa, ou) = writefds; 958 SCARG(&bsa, ex) = exceptfds; 959 SCARG(&bsa, tv) = timeout; 960 961 /* 962 * Store current time for computation of the amount of 963 * time left. 964 */ 965 if (timeout) { 966 if ((error = copyin(timeout, &utv, sizeof(utv)))) 967 return error; 968 if (itimerfix(&utv)) { 969 /* 970 * The timeval was invalid. Convert it to something 971 * valid that will act as it does under Linux. 972 */ 973 sg = stackgap_init(p->p_emul); 974 tvp = stackgap_alloc(&sg, sizeof(utv)); 975 utv.tv_sec += utv.tv_usec / 1000000; 976 utv.tv_usec %= 1000000; 977 if (utv.tv_usec < 0) { 978 utv.tv_sec -= 1; 979 utv.tv_usec += 1000000; 980 } 981 if (utv.tv_sec < 0) 982 timerclear(&utv); 983 if ((error = copyout(&utv, tvp, sizeof(utv)))) 984 return error; 985 SCARG(&bsa, tv) = tvp; 986 } 987 microtime(&tv0); 988 } 989 990 error = sys_select(p, &bsa, retval); 991 if (error) { 992 /* 993 * See fs/select.c in the Linux kernel. Without this, 994 * Maelstrom doesn't work. 995 */ 996 if (error == ERESTART) 997 error = EINTR; 998 return error; 999 } 1000 1001 if (timeout) { 1002 if (*retval) { 1003 /* 1004 * Compute how much time was left of the timeout, 1005 * by subtracting the current time and the time 1006 * before we started the call, and subtracting 1007 * that result from the user-supplied value. 1008 */ 1009 microtime(&tv1); 1010 timersub(&tv1, &tv0, &tv1); 1011 timersub(&utv, &tv1, &utv); 1012 if (utv.tv_sec < 0) 1013 timerclear(&utv); 1014 } else 1015 timerclear(&utv); 1016 if ((error = copyout(&utv, timeout, sizeof(utv)))) 1017 return error; 1018 } 1019 1020 return 0; 1021} 1022 1023/* 1024 * Get the process group of a certain process. Look it up 1025 * and return the value. 1026 */ 1027int 1028linux_sys_getpgid(p, v, retval) 1029 struct proc *p; 1030 void *v; 1031 register_t *retval; 1032{ 1033 struct linux_sys_getpgid_args /* { 1034 syscallarg(int) pid; 1035 } */ *uap = v; 1036 struct proc *targp; 1037 1038 if (SCARG(uap, pid) != 0 && SCARG(uap, pid) != p->p_pid) { 1039 if ((targp = pfind(SCARG(uap, pid))) == 0) 1040 return ESRCH; 1041 } 1042 else 1043 targp = p; 1044 1045 retval[0] = targp->p_pgid; 1046 return 0; 1047} 1048 1049/* 1050 * Set the 'personality' (emulation mode) for the current process. Only 1051 * accept the Linux personality here (0). This call is needed because 1052 * the Linux ELF crt0 issues it in an ugly kludge to make sure that 1053 * ELF binaries run in Linux mode, not SVR4 mode. 1054 */ 1055int 1056linux_sys_personality(p, v, retval) 1057 struct proc *p; 1058 void *v; 1059 register_t *retval; 1060{ 1061 struct linux_sys_personality_args /* { 1062 syscallarg(int) per; 1063 } */ *uap = v; 1064 1065 if (SCARG(uap, per) != 0) 1066 return EINVAL; 1067 retval[0] = 0; 1068 return 0; 1069} 1070 1071/* 1072 * The calls are here because of type conversions. 1073 */ 1074int 1075linux_sys_setreuid(p, v, retval) 1076 struct proc *p; 1077 void *v; 1078 register_t *retval; 1079{ 1080 struct linux_sys_setreuid_args /* { 1081 syscallarg(int) ruid; 1082 syscallarg(int) euid; 1083 } */ *uap = v; 1084 struct compat_43_sys_setreuid_args bsa; 1085 1086 SCARG(&bsa, ruid) = ((linux_uid_t)SCARG(uap, ruid) == (linux_uid_t)-1) ? 1087 (uid_t)-1 : SCARG(uap, ruid); 1088 SCARG(&bsa, euid) = ((linux_uid_t)SCARG(uap, euid) == (linux_uid_t)-1) ? 1089 (uid_t)-1 : SCARG(uap, euid); 1090 1091 return compat_43_sys_setreuid(p, &bsa, retval); 1092} 1093 1094int 1095linux_sys_setregid(p, v, retval) 1096 struct proc *p; 1097 void *v; 1098 register_t *retval; 1099{ 1100 struct linux_sys_setregid_args /* { 1101 syscallarg(int) rgid; 1102 syscallarg(int) egid; 1103 } */ *uap = v; 1104 struct compat_43_sys_setregid_args bsa; 1105 1106 SCARG(&bsa, rgid) = ((linux_gid_t)SCARG(uap, rgid) == (linux_gid_t)-1) ? 1107 (uid_t)-1 : SCARG(uap, rgid); 1108 SCARG(&bsa, egid) = ((linux_gid_t)SCARG(uap, egid) == (linux_gid_t)-1) ? 1109 (uid_t)-1 : SCARG(uap, egid); 1110 1111 return compat_43_sys_setregid(p, &bsa, retval); 1112} 1113 1114int 1115linux_sys_getsid(p, v, retval) 1116 struct proc *p; 1117 void *v; 1118 register_t *retval; 1119{ 1120 struct linux_sys_getsid_args /* { 1121 syscallarg(int) pid; 1122 } */ *uap = v; 1123 struct proc *p1; 1124 pid_t pid; 1125 1126 pid = (pid_t)SCARG(uap, pid); 1127 1128 if (pid == 0) { 1129 retval[0] = (int)p->p_session; /* XXX Oh well */ 1130 return 0; 1131 } 1132 1133 p1 = pfind((int)pid); 1134 if (p1 == NULL) 1135 return ESRCH; 1136 1137 retval[0] = (int)p1->p_session; 1138 return 0; 1139} 1140 1141int 1142linux_sys___sysctl(p, v, retval) 1143 struct proc *p; 1144 void *v; 1145 register_t *retval; 1146{ 1147 struct linux_sys___sysctl_args /* { 1148 syscallarg(struct linux___sysctl *) lsp; 1149 } */ *uap = v; 1150 struct linux___sysctl ls; 1151 struct sys___sysctl_args bsa; 1152 int error; 1153 1154 if ((error = copyin(SCARG(uap, lsp), &ls, sizeof ls))) 1155 return error; 1156 SCARG(&bsa, name) = ls.name; 1157 SCARG(&bsa, namelen) = ls.namelen; 1158 SCARG(&bsa, old) = ls.old; 1159 SCARG(&bsa, oldlenp) = ls.oldlenp; 1160 SCARG(&bsa, new) = ls.new; 1161 SCARG(&bsa, newlen) = ls.newlen; 1162 1163 return sys___sysctl(p, &bsa, retval); 1164} 1165