1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)kern_descrip.c 7.24 (Berkeley) 05/30/91 8 */ 9 10 #include "param.h" 11 #include "systm.h" 12 #include "filedesc.h" 13 #include "kernel.h" 14 #include "vnode.h" 15 #include "proc.h" 16 #include "file.h" 17 #include "socket.h" 18 #include "socketvar.h" 19 #include "stat.h" 20 #include "ioctl.h" 21 #include "fcntl.h" 22 #include "malloc.h" 23 #include "syslog.h" 24 #include "resourcevar.h" 25 26 /* 27 * Descriptor management. 28 */ 29 30 /* 31 * System calls on descriptors. 32 */ 33 /* ARGSUSED */ 34 getdtablesize(p, uap, retval) 35 struct proc *p; 36 struct args *uap; 37 int *retval; 38 { 39 40 *retval = p->p_rlimit[RLIMIT_OFILE].rlim_cur; 41 return (0); 42 } 43 44 /* 45 * Duplicate a file descriptor. 46 */ 47 /* ARGSUSED */ 48 dup(p, uap, retval) 49 struct proc *p; 50 struct args { 51 int i; 52 } *uap; 53 int *retval; 54 { 55 register struct filedesc *fdp = p->p_fd; 56 struct file *fp; 57 int fd, error; 58 59 /* 60 * XXX Compatibility 61 */ 62 if (uap->i &~ 077) { uap->i &= 077; return (dup2(p, uap, retval)); } 63 64 if ((unsigned)uap->i >= fdp->fd_nfiles || 65 (fp = fdp->fd_ofiles[uap->i]) == NULL) 66 return (EBADF); 67 if (error = fdalloc(p, 0, &fd)) 68 return (error); 69 fdp->fd_ofiles[fd] = fp; 70 fdp->fd_ofileflags[fd] = fdp->fd_ofileflags[uap->i] &~ UF_EXCLOSE; 71 fp->f_count++; 72 if (fd > fdp->fd_lastfile) 73 fdp->fd_lastfile = fd; 74 *retval = fd; 75 return (0); 76 } 77 78 /* 79 * Duplicate a file descriptor to a particular value. 80 */ 81 /* ARGSUSED */ 82 dup2(p, uap, retval) 83 struct proc *p; 84 struct args { 85 u_int from; 86 u_int to; 87 } *uap; 88 int *retval; 89 { 90 register struct filedesc *fdp = p->p_fd; 91 register struct file *fp; 92 register u_int old = uap->from, new = uap->to; 93 int i, error; 94 95 if (old >= fdp->fd_nfiles || 96 (fp = fdp->fd_ofiles[old]) == NULL || 97 new >= p->p_rlimit[RLIMIT_OFILE].rlim_cur) 98 return (EBADF); 99 *retval = new; 100 if (old == new) 101 return (0); 102 if (new >= fdp->fd_nfiles) { 103 if (error = fdalloc(p, new, &i)) 104 return (error); 105 if (new != i) 106 panic("dup2: fdalloc"); 107 } else if (fdp->fd_ofiles[new]) { 108 if (fdp->fd_ofileflags[new] & UF_MAPPED) 109 (void) munmapfd(p, new); 110 /* 111 * dup2() must succeed even if the close has an error. 112 */ 113 (void) closef(fdp->fd_ofiles[new], p); 114 } 115 fdp->fd_ofiles[new] = fp; 116 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE; 117 fp->f_count++; 118 if (new > fdp->fd_lastfile) 119 fdp->fd_lastfile = new; 120 return (0); 121 } 122 123 /* 124 * The file control system call. 125 */ 126 /* ARGSUSED */ 127 fcntl(p, uap, retval) 128 struct proc *p; 129 register struct args { 130 int fd; 131 int cmd; 132 int arg; 133 } *uap; 134 int *retval; 135 { 136 register struct filedesc *fdp = p->p_fd; 137 register struct file *fp; 138 register char *pop; 139 struct vnode *vp; 140 int i, tmp, error, flg = F_POSIX; 141 struct flock fl; 142 143 if ((unsigned)uap->fd >= fdp->fd_nfiles || 144 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 145 return (EBADF); 146 pop = &fdp->fd_ofileflags[uap->fd]; 147 switch(uap->cmd) { 148 case F_DUPFD: 149 if ((unsigned)uap->arg >= p->p_rlimit[RLIMIT_OFILE].rlim_cur) 150 return (EINVAL); 151 if (error = fdalloc(p, uap->arg, &i)) 152 return (error); 153 fdp->fd_ofiles[i] = fp; 154 fdp->fd_ofileflags[i] = *pop &~ UF_EXCLOSE; 155 fp->f_count++; 156 if (i > fdp->fd_lastfile) 157 fdp->fd_lastfile = i; 158 *retval = i; 159 return (0); 160 161 case F_GETFD: 162 *retval = *pop & 1; 163 return (0); 164 165 case F_SETFD: 166 *pop = (*pop &~ 1) | (uap->arg & 1); 167 return (0); 168 169 case F_GETFL: 170 *retval = OFLAGS(fp->f_flag); 171 return (0); 172 173 case F_SETFL: 174 fp->f_flag &= ~FCNTLFLAGS; 175 fp->f_flag |= FFLAGS(uap->arg) & FCNTLFLAGS; 176 tmp = fp->f_flag & FNONBLOCK; 177 error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p); 178 if (error) 179 return (error); 180 tmp = fp->f_flag & FASYNC; 181 error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, (caddr_t)&tmp, p); 182 if (!error) 183 return (0); 184 fp->f_flag &= ~FNONBLOCK; 185 tmp = 0; 186 (void) (*fp->f_ops->fo_ioctl)(fp, FIONBIO, (caddr_t)&tmp, p); 187 return (error); 188 189 case F_GETOWN: 190 if (fp->f_type == DTYPE_SOCKET) { 191 *retval = ((struct socket *)fp->f_data)->so_pgid; 192 return (0); 193 } 194 error = (*fp->f_ops->fo_ioctl) 195 (fp, (int)TIOCGPGRP, (caddr_t)retval, p); 196 *retval = -*retval; 197 return (error); 198 199 case F_SETOWN: 200 if (fp->f_type == DTYPE_SOCKET) { 201 ((struct socket *)fp->f_data)->so_pgid = uap->arg; 202 return (0); 203 } 204 if (uap->arg <= 0) { 205 uap->arg = -uap->arg; 206 } else { 207 struct proc *p1 = pfind(uap->arg); 208 if (p1 == 0) 209 return (ESRCH); 210 uap->arg = p1->p_pgrp->pg_id; 211 } 212 return ((*fp->f_ops->fo_ioctl) 213 (fp, (int)TIOCSPGRP, (caddr_t)&uap->arg, p)); 214 215 case F_SETLKW: 216 flg |= F_WAIT; 217 /* Fall into F_SETLK */ 218 219 case F_SETLK: 220 if (fp->f_type != DTYPE_VNODE) 221 return (EBADF); 222 vp = (struct vnode *)fp->f_data; 223 /* Copy in the lock structure */ 224 error = copyin((caddr_t)uap->arg, (caddr_t)&fl, sizeof (fl)); 225 if (error) 226 return (error); 227 if (fl.l_whence == SEEK_CUR) 228 fl.l_start += fp->f_offset; 229 switch (fl.l_type) { 230 231 case F_RDLCK: 232 if ((fp->f_flag & FREAD) == 0) 233 return (EBADF); 234 return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg)); 235 236 case F_WRLCK: 237 if ((fp->f_flag & FWRITE) == 0) 238 return (EBADF); 239 return (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &fl, flg)); 240 241 case F_UNLCK: 242 return (VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &fl, 243 F_POSIX)); 244 245 default: 246 return (EINVAL); 247 } 248 249 case F_GETLK: 250 if (fp->f_type != DTYPE_VNODE) 251 return (EBADF); 252 vp = (struct vnode *)fp->f_data; 253 /* Copy in the lock structure */ 254 error = copyin((caddr_t)uap->arg, (caddr_t)&fl, sizeof (fl)); 255 if (error) 256 return (error); 257 if (fl.l_whence == SEEK_CUR) 258 fl.l_start += fp->f_offset; 259 if (error = VOP_ADVLOCK(vp, (caddr_t)p, F_GETLK, &fl, F_POSIX)) 260 return (error); 261 return (copyout((caddr_t)&fl, (caddr_t)uap->arg, sizeof (fl))); 262 263 default: 264 return (EINVAL); 265 } 266 /* NOTREACHED */ 267 } 268 269 /* 270 * Close a file descriptor. 271 */ 272 /* ARGSUSED */ 273 close(p, uap, retval) 274 struct proc *p; 275 struct args { 276 int fd; 277 } *uap; 278 int *retval; 279 { 280 register struct filedesc *fdp = p->p_fd; 281 register struct file *fp; 282 register int fd = uap->fd; 283 register u_char *pf; 284 285 if ((unsigned)fd >= fdp->fd_nfiles || 286 (fp = fdp->fd_ofiles[fd]) == NULL) 287 return (EBADF); 288 pf = (u_char *)&fdp->fd_ofileflags[fd]; 289 if (*pf & UF_MAPPED) 290 (void) munmapfd(p, fd); 291 fdp->fd_ofiles[fd] = NULL; 292 while (fdp->fd_lastfile > 0 && fdp->fd_ofiles[fdp->fd_lastfile] == NULL) 293 fdp->fd_lastfile--; 294 if (fd < fdp->fd_freefile) 295 fdp->fd_freefile = fd; 296 *pf = 0; 297 return (closef(fp, p)); 298 } 299 300 /* 301 * Return status information about a file descriptor. 302 */ 303 /* ARGSUSED */ 304 fstat(p, uap, retval) 305 struct proc *p; 306 register struct args { 307 int fd; 308 struct stat *sb; 309 } *uap; 310 int *retval; 311 { 312 register struct filedesc *fdp = p->p_fd; 313 register struct file *fp; 314 struct stat ub; 315 int error; 316 317 if ((unsigned)uap->fd >= fdp->fd_nfiles || 318 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 319 return (EBADF); 320 switch (fp->f_type) { 321 322 case DTYPE_VNODE: 323 error = vn_stat((struct vnode *)fp->f_data, &ub, p); 324 break; 325 326 case DTYPE_SOCKET: 327 error = soo_stat((struct socket *)fp->f_data, &ub); 328 break; 329 330 default: 331 panic("fstat"); 332 /*NOTREACHED*/ 333 } 334 if (error == 0) 335 error = copyout((caddr_t)&ub, (caddr_t)uap->sb, sizeof (ub)); 336 return (error); 337 } 338 339 /* 340 * Allocate a file descriptor for the process. 341 */ 342 int fdexpand; 343 344 fdalloc(p, want, result) 345 struct proc *p; 346 int want; 347 int *result; 348 { 349 register struct filedesc *fdp = p->p_fd; 350 register int i; 351 int lim, last, nfiles; 352 struct file **newofile; 353 char *newofileflags; 354 355 /* 356 * Search for a free descriptor starting at the higher 357 * of want or fd_freefile. If that fails, consider 358 * expanding the ofile array. 359 */ 360 lim = p->p_rlimit[RLIMIT_OFILE].rlim_cur; 361 for (;;) { 362 last = min(fdp->fd_nfiles, lim); 363 if ((i = want) < fdp->fd_freefile) 364 i = fdp->fd_freefile; 365 for (; i < last; i++) { 366 if (fdp->fd_ofiles[i] == NULL) { 367 fdp->fd_ofileflags[i] = 0; 368 if (i > fdp->fd_lastfile) 369 fdp->fd_lastfile = i; 370 if (want <= fdp->fd_freefile) 371 fdp->fd_freefile = i; 372 *result = i; 373 return (0); 374 } 375 } 376 377 /* 378 * No space in current array. Expand? 379 */ 380 if (fdp->fd_nfiles >= lim) 381 return (EMFILE); 382 if (fdp->fd_nfiles < NDEXTENT) 383 nfiles = NDEXTENT; 384 else 385 nfiles = 2 * fdp->fd_nfiles; 386 MALLOC(newofile, struct file **, nfiles * OFILESIZE, 387 M_FILEDESC, M_WAITOK); 388 newofileflags = (char *) &newofile[nfiles]; 389 /* 390 * Copy the existing ofile and ofileflags arrays 391 * and zero the new portion of each array. 392 */ 393 bcopy(fdp->fd_ofiles, newofile, 394 (i = sizeof(struct file *) * fdp->fd_nfiles)); 395 bzero((char *)newofile + i, nfiles * sizeof(struct file *) - i); 396 bcopy(fdp->fd_ofileflags, newofileflags, 397 (i = sizeof(char) * fdp->fd_nfiles)); 398 bzero(newofileflags + i, nfiles * sizeof(char) - i); 399 if (fdp->fd_nfiles > NDFILE) 400 FREE(fdp->fd_ofiles, M_FILEDESC); 401 fdp->fd_ofiles = newofile; 402 fdp->fd_ofileflags = newofileflags; 403 fdp->fd_nfiles = nfiles; 404 fdexpand++; 405 } 406 } 407 408 /* 409 * Check to see whether n user file descriptors 410 * are available to the process p. 411 */ 412 fdavail(p, n) 413 struct proc *p; 414 register int n; 415 { 416 register struct filedesc *fdp = p->p_fd; 417 register struct file **fpp; 418 register int i; 419 420 if ((i = p->p_rlimit[RLIMIT_OFILE].rlim_cur - fdp->fd_nfiles) > 0 && 421 (n -= i) <= 0) 422 return (1); 423 fpp = &fdp->fd_ofiles[fdp->fd_freefile]; 424 for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++) 425 if (*fpp == NULL && --n <= 0) 426 return (1); 427 return (0); 428 } 429 430 struct file *lastf; 431 /* 432 * Create a new open file structure and allocate 433 * a file decriptor for the process that refers to it. 434 */ 435 falloc(p, resultfp, resultfd) 436 register struct proc *p; 437 struct file **resultfp; 438 int *resultfd; 439 { 440 register struct file *fp; 441 int error, i; 442 443 if (error = fdalloc(p, 0, &i)) 444 return (error); 445 if (lastf == 0) 446 lastf = file; 447 for (fp = lastf; fp < fileNFILE; fp++) 448 if (fp->f_count == 0) 449 goto slot; 450 for (fp = file; fp < lastf; fp++) 451 if (fp->f_count == 0) 452 goto slot; 453 tablefull("file"); 454 return (ENFILE); 455 slot: 456 p->p_fd->fd_ofiles[i] = fp; 457 fp->f_count = 1; 458 fp->f_data = 0; 459 fp->f_offset = 0; 460 fp->f_cred = p->p_ucred; 461 crhold(fp->f_cred); 462 lastf = fp + 1; 463 if (resultfp) 464 *resultfp = fp; 465 if (resultfd) 466 *resultfd = i; 467 return (0); 468 } 469 470 /* 471 * Copy a filedesc structure. 472 */ 473 struct filedesc * 474 fdcopy(p) 475 struct proc *p; 476 { 477 register struct filedesc *newfdp, *fdp = p->p_fd; 478 register struct file **fpp; 479 register int i; 480 481 MALLOC(newfdp, struct filedesc *, sizeof(struct filedesc0), 482 M_FILEDESC, M_WAITOK); 483 bcopy(fdp, newfdp, sizeof(struct filedesc)); 484 VREF(newfdp->fd_cdir); 485 if (newfdp->fd_rdir) 486 VREF(newfdp->fd_rdir); 487 newfdp->fd_refcnt = 1; 488 489 /* 490 * If the number of open files fits in the internal arrays 491 * of the open file structure, use them, otherwise allocate 492 * additional memory for the number of descriptors currently 493 * in use. 494 */ 495 if (newfdp->fd_lastfile < NDFILE) { 496 newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles; 497 newfdp->fd_ofileflags = 498 ((struct filedesc0 *) newfdp)->fd_dfileflags; 499 i = NDFILE; 500 } else { 501 /* 502 * Compute the smallest multiple of NDEXTENT needed 503 * for the file descriptors currently in use, 504 * allowing the table to shrink. 505 */ 506 i = newfdp->fd_nfiles; 507 while (i > 2 * NDEXTENT && i >= newfdp->fd_lastfile * 2) 508 i /= 2; 509 MALLOC(newfdp->fd_ofiles, struct file **, i * OFILESIZE, 510 M_FILEDESC, M_WAITOK); 511 newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i]; 512 } 513 newfdp->fd_nfiles = i; 514 bcopy(fdp->fd_ofiles, newfdp->fd_ofiles, i * sizeof(struct file **)); 515 bcopy(fdp->fd_ofileflags, newfdp->fd_ofileflags, i * sizeof(char)); 516 fpp = newfdp->fd_ofiles; 517 for (i = newfdp->fd_lastfile; i-- >= 0; fpp++) 518 if (*fpp != NULL) 519 (*fpp)->f_count++; 520 return (newfdp); 521 } 522 523 /* 524 * Release a filedesc structure. 525 */ 526 void 527 fdfree(p) 528 struct proc *p; 529 { 530 register struct filedesc *fdp = p->p_fd; 531 struct file **fpp; 532 register int i; 533 534 if (--fdp->fd_refcnt > 0) 535 return; 536 fpp = fdp->fd_ofiles; 537 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) 538 if (*fpp) 539 (void) closef(*fpp, p); 540 if (fdp->fd_nfiles > NDFILE) 541 FREE(fdp->fd_ofiles, M_FILEDESC); 542 vrele(fdp->fd_cdir); 543 if (fdp->fd_rdir) 544 vrele(fdp->fd_rdir); 545 FREE(fdp, M_FILEDESC); 546 } 547 548 /* 549 * Internal form of close. 550 * Decrement reference count on file structure. 551 */ 552 closef(fp, p) 553 register struct file *fp; 554 struct proc *p; 555 { 556 struct vnode *vp; 557 struct flock lf; 558 int error; 559 560 if (fp == NULL) 561 return (0); 562 /* 563 * POSIX record locking dictates that any close releases ALL 564 * locks owned by this process. This is handled by setting 565 * a flag in the unlock to free ONLY locks obeying POSIX 566 * semantics, and not to free BSD-style file locks. 567 */ 568 if (fp->f_type == DTYPE_VNODE) { 569 lf.l_whence = SEEK_SET; 570 lf.l_start = 0; 571 lf.l_len = 0; 572 lf.l_type = F_UNLCK; 573 vp = (struct vnode *)fp->f_data; 574 (void) VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX); 575 } 576 if (--fp->f_count > 0) 577 return (0); 578 if (fp->f_count < 0) 579 panic("closef: count < 0"); 580 if (fp->f_type == DTYPE_VNODE) 581 (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 582 error = (*fp->f_ops->fo_close)(fp, p); 583 crfree(fp->f_cred); 584 fp->f_count = 0; 585 return (error); 586 } 587 588 /* 589 * Apply an advisory lock on a file descriptor. 590 * 591 * Just attempt to get a record lock of the requested type on 592 * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). 593 */ 594 595 /* ARGSUSED */ 596 flock(p, uap, retval) 597 struct proc *p; 598 register struct args { 599 int fd; 600 int how; 601 } *uap; 602 int *retval; 603 { 604 register struct filedesc *fdp = p->p_fd; 605 register struct file *fp; 606 struct vnode *vp; 607 struct flock lf; 608 int error; 609 610 if ((unsigned)uap->fd >= fdp->fd_nfiles || 611 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 612 return (EBADF); 613 if (fp->f_type != DTYPE_VNODE) 614 return (EOPNOTSUPP); 615 vp = (struct vnode *)fp->f_data; 616 lf.l_whence = SEEK_SET; 617 lf.l_start = 0; 618 lf.l_len = 0; 619 if (uap->how & LOCK_UN) { 620 lf.l_type = F_UNLCK; 621 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK)); 622 } 623 if (uap->how & LOCK_EX) 624 lf.l_type = F_WRLCK; 625 else if (uap->how & LOCK_SH) 626 lf.l_type = F_RDLCK; 627 else 628 return (EBADF); 629 if (uap->how & LOCK_NB) 630 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK)); 631 return (VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_FLOCK|F_WAIT)); 632 } 633 634 /* 635 * File Descriptor pseudo-device driver (/dev/fd/). 636 * 637 * Opening minor device N dup()s the file (if any) connected to file 638 * descriptor N belonging to the calling process. Note that this driver 639 * consists of only the ``open()'' routine, because all subsequent 640 * references to this file will be direct to the other driver. 641 */ 642 /* ARGSUSED */ 643 fdopen(dev, mode, type) 644 dev_t dev; 645 int mode, type; 646 { 647 648 /* 649 * XXX Kludge: set curproc->p_dupfd to contain the value of the 650 * the file descriptor being sought for duplication. The error 651 * return ensures that the vnode for this device will be released 652 * by vn_open. Open will detect this special error and take the 653 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 654 * will simply report the error. 655 */ 656 curproc->p_dupfd = minor(dev); /* XXX */ 657 return (ENODEV); 658 } 659 660 /* 661 * Duplicate the specified descriptor to a free descriptor. 662 */ 663 dupfdopen(fdp, indx, dfd, mode) 664 register struct filedesc *fdp; 665 register int indx, dfd; 666 int mode; 667 { 668 register struct file *wfp; 669 struct file *fp; 670 671 /* 672 * If the to-be-dup'd fd number is greater than the allowed number 673 * of file descriptors, or the fd to be dup'd has already been 674 * closed, reject. Note, check for new == old is necessary as 675 * falloc could allocate an already closed to-be-dup'd descriptor 676 * as the new descriptor. 677 */ 678 fp = fdp->fd_ofiles[indx]; 679 if ((u_int)dfd >= fdp->fd_nfiles || 680 (wfp = fdp->fd_ofiles[dfd]) == NULL || fp == wfp) 681 return (EBADF); 682 683 /* 684 * Check that the mode the file is being opened for is a subset 685 * of the mode of the existing descriptor. 686 */ 687 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) 688 return (EACCES); 689 fdp->fd_ofiles[indx] = wfp; 690 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd]; 691 wfp->f_count++; 692 if (indx > fdp->fd_lastfile) 693 fdp->fd_lastfile = indx; 694 return (0); 695 } 696