1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_capsicum.h" 41 #include "opt_compat.h" 42 #include "opt_ddb.h" 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 48 #include <sys/capsicum.h> 49 #include <sys/conf.h> 50 #include <sys/domain.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/filedesc.h> 54 #include <sys/filio.h> 55 #include <sys/jail.h> 56 #include <sys/kernel.h> 57 #include <sys/ksem.h> 58 #include <sys/limits.h> 59 #include <sys/lock.h> 60 #include <sys/malloc.h> 61 #include <sys/mman.h> 62 #include <sys/mount.h> 63 #include <sys/mqueue.h> 64 #include <sys/mutex.h> 65 #include <sys/namei.h> 66 #include <sys/selinfo.h> 67 #include <sys/pipe.h> 68 #include <sys/priv.h> 69 #include <sys/proc.h> 70 #include <sys/procdesc.h> 71 #include <sys/protosw.h> 72 #include <sys/racct.h> 73 #include <sys/resourcevar.h> 74 #include <sys/sbuf.h> 75 #include <sys/signalvar.h> 76 #include <sys/socketvar.h> 77 #include <sys/stat.h> 78 #include <sys/sx.h> 79 #include <sys/syscallsubr.h> 80 #include <sys/sysctl.h> 81 #include <sys/sysproto.h> 82 #include <sys/tty.h> 83 #include <sys/unistd.h> 84 #include <sys/un.h> 85 #include <sys/unpcb.h> 86 #include <sys/user.h> 87 #include <sys/vnode.h> 88 #ifdef KTRACE 89 #include <sys/ktrace.h> 90 #endif 91 92 #include <net/vnet.h> 93 94 #include <netinet/in.h> 95 #include <netinet/in_pcb.h> 96 97 #include <security/audit/audit.h> 98 99 #include <vm/uma.h> 100 #include <vm/vm.h> 101 102 #include <ddb/ddb.h> 103 104 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); 105 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", 106 "file desc to leader structures"); 107 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); 108 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); 109 110 MALLOC_DECLARE(M_FADVISE); 111 112 static uma_zone_t file_zone; 113 114 void (*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value); 115 116 static int closefp(struct filedesc *fdp, int fd, struct file *fp, 117 struct thread *td, int holdleaders); 118 static int fd_first_free(struct filedesc *fdp, int low, int size); 119 static int fd_last_used(struct filedesc *fdp, int size); 120 static void fdgrowtable(struct filedesc *fdp, int nfd); 121 static void fdgrowtable_exp(struct filedesc *fdp, int nfd); 122 static void fdunused(struct filedesc *fdp, int fd); 123 static void fdused(struct filedesc *fdp, int fd); 124 static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif); 125 static int fill_procdesc_info(struct procdesc *pdp, 126 struct kinfo_file *kif); 127 static int fill_pts_info(struct tty *tp, struct kinfo_file *kif); 128 static int fill_sem_info(struct file *fp, struct kinfo_file *kif); 129 static int fill_shm_info(struct file *fp, struct kinfo_file *kif); 130 static int fill_socket_info(struct socket *so, struct kinfo_file *kif); 131 static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif); 132 static int getmaxfd(struct proc *p); 133 134 /* 135 * Each process has: 136 * 137 * - An array of open file descriptors (fd_ofiles) 138 * - An array of file flags (fd_ofileflags) 139 * - A bitmap recording which descriptors are in use (fd_map) 140 * 141 * A process starts out with NDFILE descriptors. The value of NDFILE has 142 * been selected based the historical limit of 20 open files, and an 143 * assumption that the majority of processes, especially short-lived 144 * processes like shells, will never need more. 145 * 146 * If this initial allocation is exhausted, a larger descriptor table and 147 * map are allocated dynamically, and the pointers in the process's struct 148 * filedesc are updated to point to those. This is repeated every time 149 * the process runs out of file descriptors (provided it hasn't hit its 150 * resource limit). 151 * 152 * Since threads may hold references to individual descriptor table 153 * entries, the tables are never freed. Instead, they are placed on a 154 * linked list and freed only when the struct filedesc is released. 155 */ 156 #define NDFILE 20 157 #define NDSLOTSIZE sizeof(NDSLOTTYPE) 158 #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) 159 #define NDSLOT(x) ((x) / NDENTRIES) 160 #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) 161 #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) 162 163 /* 164 * SLIST entry used to keep track of ofiles which must be reclaimed when 165 * the process exits. 166 */ 167 struct freetable { 168 struct filedescent *ft_table; 169 SLIST_ENTRY(freetable) ft_next; 170 }; 171 172 /* 173 * Initial allocation: a filedesc structure + the head of SLIST used to 174 * keep track of old ofiles + enough space for NDFILE descriptors. 175 */ 176 struct filedesc0 { 177 struct filedesc fd_fd; 178 SLIST_HEAD(, freetable) fd_free; 179 struct filedescent fd_dfiles[NDFILE]; 180 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; 181 }; 182 183 /* 184 * Descriptor management. 185 */ 186 volatile int openfiles; /* actual number of open files */ 187 struct mtx sigio_lock; /* mtx to protect pointers to sigio */ 188 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp); 189 190 /* A mutex to protect the association between a proc and filedesc. */ 191 static struct mtx fdesc_mtx; 192 193 /* 194 * If low >= size, just return low. Otherwise find the first zero bit in the 195 * given bitmap, starting at low and not exceeding size - 1. Return size if 196 * not found. 197 */ 198 static int 199 fd_first_free(struct filedesc *fdp, int low, int size) 200 { 201 NDSLOTTYPE *map = fdp->fd_map; 202 NDSLOTTYPE mask; 203 int off, maxoff; 204 205 if (low >= size) 206 return (low); 207 208 off = NDSLOT(low); 209 if (low % NDENTRIES) { 210 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); 211 if ((mask &= ~map[off]) != 0UL) 212 return (off * NDENTRIES + ffsl(mask) - 1); 213 ++off; 214 } 215 for (maxoff = NDSLOTS(size); off < maxoff; ++off) 216 if (map[off] != ~0UL) 217 return (off * NDENTRIES + ffsl(~map[off]) - 1); 218 return (size); 219 } 220 221 /* 222 * Find the highest non-zero bit in the given bitmap, starting at 0 and 223 * not exceeding size - 1. Return -1 if not found. 224 */ 225 static int 226 fd_last_used(struct filedesc *fdp, int size) 227 { 228 NDSLOTTYPE *map = fdp->fd_map; 229 NDSLOTTYPE mask; 230 int off, minoff; 231 232 off = NDSLOT(size); 233 if (size % NDENTRIES) { 234 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES)); 235 if ((mask &= map[off]) != 0) 236 return (off * NDENTRIES + flsl(mask) - 1); 237 --off; 238 } 239 for (minoff = NDSLOT(0); off >= minoff; --off) 240 if (map[off] != 0) 241 return (off * NDENTRIES + flsl(map[off]) - 1); 242 return (-1); 243 } 244 245 static int 246 fdisused(struct filedesc *fdp, int fd) 247 { 248 249 FILEDESC_LOCK_ASSERT(fdp); 250 251 KASSERT(fd >= 0 && fd < fdp->fd_nfiles, 252 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); 253 254 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); 255 } 256 257 /* 258 * Mark a file descriptor as used. 259 */ 260 static void 261 fdused(struct filedesc *fdp, int fd) 262 { 263 264 FILEDESC_XLOCK_ASSERT(fdp); 265 266 KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); 267 268 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); 269 if (fd > fdp->fd_lastfile) 270 fdp->fd_lastfile = fd; 271 if (fd == fdp->fd_freefile) 272 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles); 273 } 274 275 /* 276 * Mark a file descriptor as unused. 277 */ 278 static void 279 fdunused(struct filedesc *fdp, int fd) 280 { 281 282 FILEDESC_XLOCK_ASSERT(fdp); 283 284 KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); 285 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 286 ("fd=%d is still in use", fd)); 287 288 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); 289 if (fd < fdp->fd_freefile) 290 fdp->fd_freefile = fd; 291 if (fd == fdp->fd_lastfile) 292 fdp->fd_lastfile = fd_last_used(fdp, fd); 293 } 294 295 /* 296 * Free a file descriptor. 297 */ 298 static inline void 299 fdfree(struct filedesc *fdp, int fd) 300 { 301 struct filedescent *fde; 302 303 fde = &fdp->fd_ofiles[fd]; 304 filecaps_free(&fde->fde_caps); 305 bzero(fde, sizeof(*fde)); 306 fdunused(fdp, fd); 307 } 308 309 /* 310 * System calls on descriptors. 311 */ 312 #ifndef _SYS_SYSPROTO_H_ 313 struct getdtablesize_args { 314 int dummy; 315 }; 316 #endif 317 /* ARGSUSED */ 318 int 319 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) 320 { 321 struct proc *p = td->td_proc; 322 uint64_t lim; 323 324 PROC_LOCK(p); 325 td->td_retval[0] = 326 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 327 lim = racct_get_limit(td->td_proc, RACCT_NOFILE); 328 PROC_UNLOCK(p); 329 if (lim < td->td_retval[0]) 330 td->td_retval[0] = lim; 331 return (0); 332 } 333 334 /* 335 * Duplicate a file descriptor to a particular value. 336 * 337 * Note: keep in mind that a potential race condition exists when closing 338 * descriptors from a shared descriptor table (via rfork). 339 */ 340 #ifndef _SYS_SYSPROTO_H_ 341 struct dup2_args { 342 u_int from; 343 u_int to; 344 }; 345 #endif 346 /* ARGSUSED */ 347 int 348 sys_dup2(struct thread *td, struct dup2_args *uap) 349 { 350 351 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to, 352 td->td_retval)); 353 } 354 355 /* 356 * Duplicate a file descriptor. 357 */ 358 #ifndef _SYS_SYSPROTO_H_ 359 struct dup_args { 360 u_int fd; 361 }; 362 #endif 363 /* ARGSUSED */ 364 int 365 sys_dup(struct thread *td, struct dup_args *uap) 366 { 367 368 return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval)); 369 } 370 371 /* 372 * The file control system call. 373 */ 374 #ifndef _SYS_SYSPROTO_H_ 375 struct fcntl_args { 376 int fd; 377 int cmd; 378 long arg; 379 }; 380 #endif 381 /* ARGSUSED */ 382 int 383 sys_fcntl(struct thread *td, struct fcntl_args *uap) 384 { 385 struct flock fl; 386 struct __oflock ofl; 387 intptr_t arg; 388 int error; 389 int cmd; 390 391 error = 0; 392 cmd = uap->cmd; 393 switch (uap->cmd) { 394 case F_OGETLK: 395 case F_OSETLK: 396 case F_OSETLKW: 397 /* 398 * Convert old flock structure to new. 399 */ 400 error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl)); 401 fl.l_start = ofl.l_start; 402 fl.l_len = ofl.l_len; 403 fl.l_pid = ofl.l_pid; 404 fl.l_type = ofl.l_type; 405 fl.l_whence = ofl.l_whence; 406 fl.l_sysid = 0; 407 408 switch (uap->cmd) { 409 case F_OGETLK: 410 cmd = F_GETLK; 411 break; 412 case F_OSETLK: 413 cmd = F_SETLK; 414 break; 415 case F_OSETLKW: 416 cmd = F_SETLKW; 417 break; 418 } 419 arg = (intptr_t)&fl; 420 break; 421 case F_GETLK: 422 case F_SETLK: 423 case F_SETLKW: 424 case F_SETLK_REMOTE: 425 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl)); 426 arg = (intptr_t)&fl; 427 break; 428 default: 429 arg = uap->arg; 430 break; 431 } 432 if (error) 433 return (error); 434 error = kern_fcntl(td, uap->fd, cmd, arg); 435 if (error) 436 return (error); 437 if (uap->cmd == F_OGETLK) { 438 ofl.l_start = fl.l_start; 439 ofl.l_len = fl.l_len; 440 ofl.l_pid = fl.l_pid; 441 ofl.l_type = fl.l_type; 442 ofl.l_whence = fl.l_whence; 443 error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl)); 444 } else if (uap->cmd == F_GETLK) { 445 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl)); 446 } 447 return (error); 448 } 449 450 int 451 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) 452 { 453 struct filedesc *fdp; 454 struct flock *flp; 455 struct file *fp, *fp2; 456 struct filedescent *fde; 457 struct proc *p; 458 struct vnode *vp; 459 cap_rights_t rights; 460 int error, flg, tmp; 461 u_int old, new; 462 uint64_t bsize; 463 off_t foffset; 464 465 error = 0; 466 flg = F_POSIX; 467 p = td->td_proc; 468 fdp = p->p_fd; 469 470 switch (cmd) { 471 case F_DUPFD: 472 tmp = arg; 473 error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval); 474 break; 475 476 case F_DUPFD_CLOEXEC: 477 tmp = arg; 478 error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp, 479 td->td_retval); 480 break; 481 482 case F_DUP2FD: 483 tmp = arg; 484 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval); 485 break; 486 487 case F_DUP2FD_CLOEXEC: 488 tmp = arg; 489 error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp, 490 td->td_retval); 491 break; 492 493 case F_GETFD: 494 FILEDESC_SLOCK(fdp); 495 if ((fp = fget_locked(fdp, fd)) == NULL) { 496 FILEDESC_SUNLOCK(fdp); 497 error = EBADF; 498 break; 499 } 500 fde = &fdp->fd_ofiles[fd]; 501 td->td_retval[0] = 502 (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; 503 FILEDESC_SUNLOCK(fdp); 504 break; 505 506 case F_SETFD: 507 FILEDESC_XLOCK(fdp); 508 if ((fp = fget_locked(fdp, fd)) == NULL) { 509 FILEDESC_XUNLOCK(fdp); 510 error = EBADF; 511 break; 512 } 513 fde = &fdp->fd_ofiles[fd]; 514 fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | 515 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); 516 FILEDESC_XUNLOCK(fdp); 517 break; 518 519 case F_GETFL: 520 error = fget_unlocked(fdp, fd, 521 cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL); 522 if (error != 0) 523 break; 524 td->td_retval[0] = OFLAGS(fp->f_flag); 525 fdrop(fp, td); 526 break; 527 528 case F_SETFL: 529 error = fget_unlocked(fdp, fd, 530 cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL); 531 if (error != 0) 532 break; 533 do { 534 tmp = flg = fp->f_flag; 535 tmp &= ~FCNTLFLAGS; 536 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; 537 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); 538 tmp = fp->f_flag & FNONBLOCK; 539 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 540 if (error != 0) { 541 fdrop(fp, td); 542 break; 543 } 544 tmp = fp->f_flag & FASYNC; 545 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); 546 if (error == 0) { 547 fdrop(fp, td); 548 break; 549 } 550 atomic_clear_int(&fp->f_flag, FNONBLOCK); 551 tmp = 0; 552 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); 553 fdrop(fp, td); 554 break; 555 556 case F_GETOWN: 557 error = fget_unlocked(fdp, fd, 558 cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL); 559 if (error != 0) 560 break; 561 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); 562 if (error == 0) 563 td->td_retval[0] = tmp; 564 fdrop(fp, td); 565 break; 566 567 case F_SETOWN: 568 error = fget_unlocked(fdp, fd, 569 cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL); 570 if (error != 0) 571 break; 572 tmp = arg; 573 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); 574 fdrop(fp, td); 575 break; 576 577 case F_SETLK_REMOTE: 578 error = priv_check(td, PRIV_NFS_LOCKD); 579 if (error) 580 return (error); 581 flg = F_REMOTE; 582 goto do_setlk; 583 584 case F_SETLKW: 585 flg |= F_WAIT; 586 /* FALLTHROUGH F_SETLK */ 587 588 case F_SETLK: 589 do_setlk: 590 cap_rights_init(&rights, CAP_FLOCK); 591 error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL); 592 if (error != 0) 593 break; 594 if (fp->f_type != DTYPE_VNODE) { 595 error = EBADF; 596 fdrop(fp, td); 597 break; 598 } 599 600 flp = (struct flock *)arg; 601 if (flp->l_whence == SEEK_CUR) { 602 foffset = foffset_get(fp); 603 if (foffset < 0 || 604 (flp->l_start > 0 && 605 foffset > OFF_MAX - flp->l_start)) { 606 FILEDESC_SUNLOCK(fdp); 607 error = EOVERFLOW; 608 fdrop(fp, td); 609 break; 610 } 611 flp->l_start += foffset; 612 } 613 614 vp = fp->f_vnode; 615 switch (flp->l_type) { 616 case F_RDLCK: 617 if ((fp->f_flag & FREAD) == 0) { 618 error = EBADF; 619 break; 620 } 621 PROC_LOCK(p->p_leader); 622 p->p_leader->p_flag |= P_ADVLOCK; 623 PROC_UNLOCK(p->p_leader); 624 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 625 flp, flg); 626 break; 627 case F_WRLCK: 628 if ((fp->f_flag & FWRITE) == 0) { 629 error = EBADF; 630 break; 631 } 632 PROC_LOCK(p->p_leader); 633 p->p_leader->p_flag |= P_ADVLOCK; 634 PROC_UNLOCK(p->p_leader); 635 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, 636 flp, flg); 637 break; 638 case F_UNLCK: 639 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, 640 flp, flg); 641 break; 642 case F_UNLCKSYS: 643 /* 644 * Temporary api for testing remote lock 645 * infrastructure. 646 */ 647 if (flg != F_REMOTE) { 648 error = EINVAL; 649 break; 650 } 651 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 652 F_UNLCKSYS, flp, flg); 653 break; 654 default: 655 error = EINVAL; 656 break; 657 } 658 if (error != 0 || flp->l_type == F_UNLCK || 659 flp->l_type == F_UNLCKSYS) { 660 fdrop(fp, td); 661 break; 662 } 663 664 /* 665 * Check for a race with close. 666 * 667 * The vnode is now advisory locked (or unlocked, but this case 668 * is not really important) as the caller requested. 669 * We had to drop the filedesc lock, so we need to recheck if 670 * the descriptor is still valid, because if it was closed 671 * in the meantime we need to remove advisory lock from the 672 * vnode - close on any descriptor leading to an advisory 673 * locked vnode, removes that lock. 674 * We will return 0 on purpose in that case, as the result of 675 * successful advisory lock might have been externally visible 676 * already. This is fine - effectively we pretend to the caller 677 * that the closing thread was a bit slower and that the 678 * advisory lock succeeded before the close. 679 */ 680 error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL); 681 if (error != 0) { 682 fdrop(fp, td); 683 break; 684 } 685 if (fp != fp2) { 686 flp->l_whence = SEEK_SET; 687 flp->l_start = 0; 688 flp->l_len = 0; 689 flp->l_type = F_UNLCK; 690 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, 691 F_UNLCK, flp, F_POSIX); 692 } 693 fdrop(fp, td); 694 fdrop(fp2, td); 695 break; 696 697 case F_GETLK: 698 error = fget_unlocked(fdp, fd, 699 cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL); 700 if (error != 0) 701 break; 702 if (fp->f_type != DTYPE_VNODE) { 703 error = EBADF; 704 fdrop(fp, td); 705 break; 706 } 707 flp = (struct flock *)arg; 708 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && 709 flp->l_type != F_UNLCK) { 710 error = EINVAL; 711 fdrop(fp, td); 712 break; 713 } 714 if (flp->l_whence == SEEK_CUR) { 715 foffset = foffset_get(fp); 716 if ((flp->l_start > 0 && 717 foffset > OFF_MAX - flp->l_start) || 718 (flp->l_start < 0 && 719 foffset < OFF_MIN - flp->l_start)) { 720 FILEDESC_SUNLOCK(fdp); 721 error = EOVERFLOW; 722 fdrop(fp, td); 723 break; 724 } 725 flp->l_start += foffset; 726 } 727 vp = fp->f_vnode; 728 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, 729 F_POSIX); 730 fdrop(fp, td); 731 break; 732 733 case F_RDAHEAD: 734 arg = arg ? 128 * 1024: 0; 735 /* FALLTHROUGH */ 736 case F_READAHEAD: 737 error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL); 738 if (error != 0) 739 break; 740 if (fp->f_type != DTYPE_VNODE) { 741 fdrop(fp, td); 742 error = EBADF; 743 break; 744 } 745 if (arg >= 0) { 746 vp = fp->f_vnode; 747 error = vn_lock(vp, LK_SHARED); 748 if (error != 0) { 749 fdrop(fp, td); 750 break; 751 } 752 bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; 753 VOP_UNLOCK(vp, 0); 754 fp->f_seqcount = (arg + bsize - 1) / bsize; 755 do { 756 new = old = fp->f_flag; 757 new |= FRDAHEAD; 758 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 759 } else { 760 do { 761 new = old = fp->f_flag; 762 new &= ~FRDAHEAD; 763 } while (!atomic_cmpset_rel_int(&fp->f_flag, old, new)); 764 } 765 fdrop(fp, td); 766 break; 767 768 default: 769 error = EINVAL; 770 break; 771 } 772 return (error); 773 } 774 775 static int 776 getmaxfd(struct proc *p) 777 { 778 int maxfd; 779 780 PROC_LOCK(p); 781 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc); 782 PROC_UNLOCK(p); 783 784 return (maxfd); 785 } 786 787 /* 788 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). 789 */ 790 int 791 do_dup(struct thread *td, int flags, int old, int new, 792 register_t *retval) 793 { 794 struct filedesc *fdp; 795 struct filedescent *oldfde, *newfde; 796 struct proc *p; 797 struct file *fp; 798 struct file *delfp; 799 int error, maxfd; 800 801 p = td->td_proc; 802 fdp = p->p_fd; 803 804 /* 805 * Verify we have a valid descriptor to dup from and possibly to 806 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should 807 * return EINVAL when the new descriptor is out of bounds. 808 */ 809 if (old < 0) 810 return (EBADF); 811 if (new < 0) 812 return (flags & DUP_FCNTL ? EINVAL : EBADF); 813 maxfd = getmaxfd(p); 814 if (new >= maxfd) 815 return (flags & DUP_FCNTL ? EINVAL : EBADF); 816 817 FILEDESC_XLOCK(fdp); 818 if (fget_locked(fdp, old) == NULL) { 819 FILEDESC_XUNLOCK(fdp); 820 return (EBADF); 821 } 822 oldfde = &fdp->fd_ofiles[old]; 823 if (flags & DUP_FIXED && old == new) { 824 *retval = new; 825 if (flags & DUP_CLOEXEC) 826 fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; 827 FILEDESC_XUNLOCK(fdp); 828 return (0); 829 } 830 fp = oldfde->fde_file; 831 fhold(fp); 832 833 /* 834 * If the caller specified a file descriptor, make sure the file 835 * table is large enough to hold it, and grab it. Otherwise, just 836 * allocate a new descriptor the usual way. 837 */ 838 if (flags & DUP_FIXED) { 839 if (new >= fdp->fd_nfiles) { 840 /* 841 * The resource limits are here instead of e.g. 842 * fdalloc(), because the file descriptor table may be 843 * shared between processes, so we can't really use 844 * racct_add()/racct_sub(). Instead of counting the 845 * number of actually allocated descriptors, just put 846 * the limit on the size of the file descriptor table. 847 */ 848 #ifdef RACCT 849 PROC_LOCK(p); 850 error = racct_set(p, RACCT_NOFILE, new + 1); 851 PROC_UNLOCK(p); 852 if (error != 0) { 853 FILEDESC_XUNLOCK(fdp); 854 fdrop(fp, td); 855 return (EMFILE); 856 } 857 #endif 858 fdgrowtable_exp(fdp, new + 1); 859 oldfde = &fdp->fd_ofiles[old]; 860 } 861 newfde = &fdp->fd_ofiles[new]; 862 if (newfde->fde_file == NULL) 863 fdused(fdp, new); 864 } else { 865 if ((error = fdalloc(td, new, &new)) != 0) { 866 FILEDESC_XUNLOCK(fdp); 867 fdrop(fp, td); 868 return (error); 869 } 870 newfde = &fdp->fd_ofiles[new]; 871 } 872 873 KASSERT(fp == oldfde->fde_file, ("old fd has been modified")); 874 KASSERT(old != new, ("new fd is same as old")); 875 876 delfp = newfde->fde_file; 877 878 /* 879 * Duplicate the source descriptor. 880 */ 881 filecaps_free(&newfde->fde_caps); 882 *newfde = *oldfde; 883 filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps); 884 if ((flags & DUP_CLOEXEC) != 0) 885 newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; 886 else 887 newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; 888 if (new > fdp->fd_lastfile) 889 fdp->fd_lastfile = new; 890 *retval = new; 891 892 if (delfp != NULL) { 893 (void) closefp(fdp, new, delfp, td, 1); 894 /* closefp() drops the FILEDESC lock for us. */ 895 } else { 896 FILEDESC_XUNLOCK(fdp); 897 } 898 899 return (0); 900 } 901 902 /* 903 * If sigio is on the list associated with a process or process group, 904 * disable signalling from the device, remove sigio from the list and 905 * free sigio. 906 */ 907 void 908 funsetown(struct sigio **sigiop) 909 { 910 struct sigio *sigio; 911 912 SIGIO_LOCK(); 913 sigio = *sigiop; 914 if (sigio == NULL) { 915 SIGIO_UNLOCK(); 916 return; 917 } 918 *(sigio->sio_myref) = NULL; 919 if ((sigio)->sio_pgid < 0) { 920 struct pgrp *pg = (sigio)->sio_pgrp; 921 PGRP_LOCK(pg); 922 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, 923 sigio, sio_pgsigio); 924 PGRP_UNLOCK(pg); 925 } else { 926 struct proc *p = (sigio)->sio_proc; 927 PROC_LOCK(p); 928 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, 929 sigio, sio_pgsigio); 930 PROC_UNLOCK(p); 931 } 932 SIGIO_UNLOCK(); 933 crfree(sigio->sio_ucred); 934 free(sigio, M_SIGIO); 935 } 936 937 /* 938 * Free a list of sigio structures. 939 * We only need to lock the SIGIO_LOCK because we have made ourselves 940 * inaccessible to callers of fsetown and therefore do not need to lock 941 * the proc or pgrp struct for the list manipulation. 942 */ 943 void 944 funsetownlst(struct sigiolst *sigiolst) 945 { 946 struct proc *p; 947 struct pgrp *pg; 948 struct sigio *sigio; 949 950 sigio = SLIST_FIRST(sigiolst); 951 if (sigio == NULL) 952 return; 953 p = NULL; 954 pg = NULL; 955 956 /* 957 * Every entry of the list should belong 958 * to a single proc or pgrp. 959 */ 960 if (sigio->sio_pgid < 0) { 961 pg = sigio->sio_pgrp; 962 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); 963 } else /* if (sigio->sio_pgid > 0) */ { 964 p = sigio->sio_proc; 965 PROC_LOCK_ASSERT(p, MA_NOTOWNED); 966 } 967 968 SIGIO_LOCK(); 969 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { 970 *(sigio->sio_myref) = NULL; 971 if (pg != NULL) { 972 KASSERT(sigio->sio_pgid < 0, 973 ("Proc sigio in pgrp sigio list")); 974 KASSERT(sigio->sio_pgrp == pg, 975 ("Bogus pgrp in sigio list")); 976 PGRP_LOCK(pg); 977 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, 978 sio_pgsigio); 979 PGRP_UNLOCK(pg); 980 } else /* if (p != NULL) */ { 981 KASSERT(sigio->sio_pgid > 0, 982 ("Pgrp sigio in proc sigio list")); 983 KASSERT(sigio->sio_proc == p, 984 ("Bogus proc in sigio list")); 985 PROC_LOCK(p); 986 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, 987 sio_pgsigio); 988 PROC_UNLOCK(p); 989 } 990 SIGIO_UNLOCK(); 991 crfree(sigio->sio_ucred); 992 free(sigio, M_SIGIO); 993 SIGIO_LOCK(); 994 } 995 SIGIO_UNLOCK(); 996 } 997 998 /* 999 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). 1000 * 1001 * After permission checking, add a sigio structure to the sigio list for 1002 * the process or process group. 1003 */ 1004 int 1005 fsetown(pid_t pgid, struct sigio **sigiop) 1006 { 1007 struct proc *proc; 1008 struct pgrp *pgrp; 1009 struct sigio *sigio; 1010 int ret; 1011 1012 if (pgid == 0) { 1013 funsetown(sigiop); 1014 return (0); 1015 } 1016 1017 ret = 0; 1018 1019 /* Allocate and fill in the new sigio out of locks. */ 1020 sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); 1021 sigio->sio_pgid = pgid; 1022 sigio->sio_ucred = crhold(curthread->td_ucred); 1023 sigio->sio_myref = sigiop; 1024 1025 sx_slock(&proctree_lock); 1026 if (pgid > 0) { 1027 proc = pfind(pgid); 1028 if (proc == NULL) { 1029 ret = ESRCH; 1030 goto fail; 1031 } 1032 1033 /* 1034 * Policy - Don't allow a process to FSETOWN a process 1035 * in another session. 1036 * 1037 * Remove this test to allow maximum flexibility or 1038 * restrict FSETOWN to the current process or process 1039 * group for maximum safety. 1040 */ 1041 PROC_UNLOCK(proc); 1042 if (proc->p_session != curthread->td_proc->p_session) { 1043 ret = EPERM; 1044 goto fail; 1045 } 1046 1047 pgrp = NULL; 1048 } else /* if (pgid < 0) */ { 1049 pgrp = pgfind(-pgid); 1050 if (pgrp == NULL) { 1051 ret = ESRCH; 1052 goto fail; 1053 } 1054 PGRP_UNLOCK(pgrp); 1055 1056 /* 1057 * Policy - Don't allow a process to FSETOWN a process 1058 * in another session. 1059 * 1060 * Remove this test to allow maximum flexibility or 1061 * restrict FSETOWN to the current process or process 1062 * group for maximum safety. 1063 */ 1064 if (pgrp->pg_session != curthread->td_proc->p_session) { 1065 ret = EPERM; 1066 goto fail; 1067 } 1068 1069 proc = NULL; 1070 } 1071 funsetown(sigiop); 1072 if (pgid > 0) { 1073 PROC_LOCK(proc); 1074 /* 1075 * Since funsetownlst() is called without the proctree 1076 * locked, we need to check for P_WEXIT. 1077 * XXX: is ESRCH correct? 1078 */ 1079 if ((proc->p_flag & P_WEXIT) != 0) { 1080 PROC_UNLOCK(proc); 1081 ret = ESRCH; 1082 goto fail; 1083 } 1084 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); 1085 sigio->sio_proc = proc; 1086 PROC_UNLOCK(proc); 1087 } else { 1088 PGRP_LOCK(pgrp); 1089 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); 1090 sigio->sio_pgrp = pgrp; 1091 PGRP_UNLOCK(pgrp); 1092 } 1093 sx_sunlock(&proctree_lock); 1094 SIGIO_LOCK(); 1095 *sigiop = sigio; 1096 SIGIO_UNLOCK(); 1097 return (0); 1098 1099 fail: 1100 sx_sunlock(&proctree_lock); 1101 crfree(sigio->sio_ucred); 1102 free(sigio, M_SIGIO); 1103 return (ret); 1104 } 1105 1106 /* 1107 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). 1108 */ 1109 pid_t 1110 fgetown(sigiop) 1111 struct sigio **sigiop; 1112 { 1113 pid_t pgid; 1114 1115 SIGIO_LOCK(); 1116 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; 1117 SIGIO_UNLOCK(); 1118 return (pgid); 1119 } 1120 1121 /* 1122 * Function drops the filedesc lock on return. 1123 */ 1124 static int 1125 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, 1126 int holdleaders) 1127 { 1128 int error; 1129 1130 FILEDESC_XLOCK_ASSERT(fdp); 1131 1132 if (holdleaders) { 1133 if (td->td_proc->p_fdtol != NULL) { 1134 /* 1135 * Ask fdfree() to sleep to ensure that all relevant 1136 * process leaders can be traversed in closef(). 1137 */ 1138 fdp->fd_holdleaderscount++; 1139 } else { 1140 holdleaders = 0; 1141 } 1142 } 1143 1144 /* 1145 * We now hold the fp reference that used to be owned by the 1146 * descriptor array. We have to unlock the FILEDESC *AFTER* 1147 * knote_fdclose to prevent a race of the fd getting opened, a knote 1148 * added, and deleteing a knote for the new fd. 1149 */ 1150 knote_fdclose(td, fd); 1151 1152 /* 1153 * We need to notify mqueue if the object is of type mqueue. 1154 */ 1155 if (fp->f_type == DTYPE_MQUEUE) 1156 mq_fdclose(td, fd, fp); 1157 FILEDESC_XUNLOCK(fdp); 1158 1159 error = closef(fp, td); 1160 if (holdleaders) { 1161 FILEDESC_XLOCK(fdp); 1162 fdp->fd_holdleaderscount--; 1163 if (fdp->fd_holdleaderscount == 0 && 1164 fdp->fd_holdleaderswakeup != 0) { 1165 fdp->fd_holdleaderswakeup = 0; 1166 wakeup(&fdp->fd_holdleaderscount); 1167 } 1168 FILEDESC_XUNLOCK(fdp); 1169 } 1170 return (error); 1171 } 1172 1173 /* 1174 * Close a file descriptor. 1175 */ 1176 #ifndef _SYS_SYSPROTO_H_ 1177 struct close_args { 1178 int fd; 1179 }; 1180 #endif 1181 /* ARGSUSED */ 1182 int 1183 sys_close(td, uap) 1184 struct thread *td; 1185 struct close_args *uap; 1186 { 1187 1188 return (kern_close(td, uap->fd)); 1189 } 1190 1191 int 1192 kern_close(td, fd) 1193 struct thread *td; 1194 int fd; 1195 { 1196 struct filedesc *fdp; 1197 struct file *fp; 1198 1199 fdp = td->td_proc->p_fd; 1200 1201 AUDIT_SYSCLOSE(td, fd); 1202 1203 FILEDESC_XLOCK(fdp); 1204 if ((fp = fget_locked(fdp, fd)) == NULL) { 1205 FILEDESC_XUNLOCK(fdp); 1206 return (EBADF); 1207 } 1208 fdfree(fdp, fd); 1209 1210 /* closefp() drops the FILEDESC lock for us. */ 1211 return (closefp(fdp, fd, fp, td, 1)); 1212 } 1213 1214 /* 1215 * Close open file descriptors. 1216 */ 1217 #ifndef _SYS_SYSPROTO_H_ 1218 struct closefrom_args { 1219 int lowfd; 1220 }; 1221 #endif 1222 /* ARGSUSED */ 1223 int 1224 sys_closefrom(struct thread *td, struct closefrom_args *uap) 1225 { 1226 struct filedesc *fdp; 1227 int fd; 1228 1229 fdp = td->td_proc->p_fd; 1230 AUDIT_ARG_FD(uap->lowfd); 1231 1232 /* 1233 * Treat negative starting file descriptor values identical to 1234 * closefrom(0) which closes all files. 1235 */ 1236 if (uap->lowfd < 0) 1237 uap->lowfd = 0; 1238 FILEDESC_SLOCK(fdp); 1239 for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) { 1240 if (fdp->fd_ofiles[fd].fde_file != NULL) { 1241 FILEDESC_SUNLOCK(fdp); 1242 (void)kern_close(td, fd); 1243 FILEDESC_SLOCK(fdp); 1244 } 1245 } 1246 FILEDESC_SUNLOCK(fdp); 1247 return (0); 1248 } 1249 1250 #if defined(COMPAT_43) 1251 /* 1252 * Return status information about a file descriptor. 1253 */ 1254 #ifndef _SYS_SYSPROTO_H_ 1255 struct ofstat_args { 1256 int fd; 1257 struct ostat *sb; 1258 }; 1259 #endif 1260 /* ARGSUSED */ 1261 int 1262 ofstat(struct thread *td, struct ofstat_args *uap) 1263 { 1264 struct ostat oub; 1265 struct stat ub; 1266 int error; 1267 1268 error = kern_fstat(td, uap->fd, &ub); 1269 if (error == 0) { 1270 cvtstat(&ub, &oub); 1271 error = copyout(&oub, uap->sb, sizeof(oub)); 1272 } 1273 return (error); 1274 } 1275 #endif /* COMPAT_43 */ 1276 1277 /* 1278 * Return status information about a file descriptor. 1279 */ 1280 #ifndef _SYS_SYSPROTO_H_ 1281 struct fstat_args { 1282 int fd; 1283 struct stat *sb; 1284 }; 1285 #endif 1286 /* ARGSUSED */ 1287 int 1288 sys_fstat(struct thread *td, struct fstat_args *uap) 1289 { 1290 struct stat ub; 1291 int error; 1292 1293 error = kern_fstat(td, uap->fd, &ub); 1294 if (error == 0) 1295 error = copyout(&ub, uap->sb, sizeof(ub)); 1296 return (error); 1297 } 1298 1299 int 1300 kern_fstat(struct thread *td, int fd, struct stat *sbp) 1301 { 1302 struct file *fp; 1303 cap_rights_t rights; 1304 int error; 1305 1306 AUDIT_ARG_FD(fd); 1307 1308 error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp); 1309 if (error != 0) 1310 return (error); 1311 1312 AUDIT_ARG_FILE(td->td_proc, fp); 1313 1314 error = fo_stat(fp, sbp, td->td_ucred, td); 1315 fdrop(fp, td); 1316 #ifdef KTRACE 1317 if (error == 0 && KTRPOINT(td, KTR_STRUCT)) 1318 ktrstat(sbp); 1319 #endif 1320 return (error); 1321 } 1322 1323 /* 1324 * Return status information about a file descriptor. 1325 */ 1326 #ifndef _SYS_SYSPROTO_H_ 1327 struct nfstat_args { 1328 int fd; 1329 struct nstat *sb; 1330 }; 1331 #endif 1332 /* ARGSUSED */ 1333 int 1334 sys_nfstat(struct thread *td, struct nfstat_args *uap) 1335 { 1336 struct nstat nub; 1337 struct stat ub; 1338 int error; 1339 1340 error = kern_fstat(td, uap->fd, &ub); 1341 if (error == 0) { 1342 cvtnstat(&ub, &nub); 1343 error = copyout(&nub, uap->sb, sizeof(nub)); 1344 } 1345 return (error); 1346 } 1347 1348 /* 1349 * Return pathconf information about a file descriptor. 1350 */ 1351 #ifndef _SYS_SYSPROTO_H_ 1352 struct fpathconf_args { 1353 int fd; 1354 int name; 1355 }; 1356 #endif 1357 /* ARGSUSED */ 1358 int 1359 sys_fpathconf(struct thread *td, struct fpathconf_args *uap) 1360 { 1361 struct file *fp; 1362 struct vnode *vp; 1363 cap_rights_t rights; 1364 int error; 1365 1366 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp); 1367 if (error != 0) 1368 return (error); 1369 1370 /* If asynchronous I/O is available, it works for all descriptors. */ 1371 if (uap->name == _PC_ASYNC_IO) { 1372 td->td_retval[0] = async_io_version; 1373 goto out; 1374 } 1375 vp = fp->f_vnode; 1376 if (vp != NULL) { 1377 vn_lock(vp, LK_SHARED | LK_RETRY); 1378 error = VOP_PATHCONF(vp, uap->name, td->td_retval); 1379 VOP_UNLOCK(vp, 0); 1380 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { 1381 if (uap->name != _PC_PIPE_BUF) { 1382 error = EINVAL; 1383 } else { 1384 td->td_retval[0] = PIPE_BUF; 1385 error = 0; 1386 } 1387 } else { 1388 error = EOPNOTSUPP; 1389 } 1390 out: 1391 fdrop(fp, td); 1392 return (error); 1393 } 1394 1395 /* 1396 * Initialize filecaps structure. 1397 */ 1398 void 1399 filecaps_init(struct filecaps *fcaps) 1400 { 1401 1402 bzero(fcaps, sizeof(*fcaps)); 1403 fcaps->fc_nioctls = -1; 1404 } 1405 1406 /* 1407 * Copy filecaps structure allocating memory for ioctls array if needed. 1408 */ 1409 void 1410 filecaps_copy(const struct filecaps *src, struct filecaps *dst) 1411 { 1412 size_t size; 1413 1414 *dst = *src; 1415 if (src->fc_ioctls != NULL) { 1416 KASSERT(src->fc_nioctls > 0, 1417 ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); 1418 1419 size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; 1420 dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); 1421 bcopy(src->fc_ioctls, dst->fc_ioctls, size); 1422 } 1423 } 1424 1425 /* 1426 * Move filecaps structure to the new place and clear the old place. 1427 */ 1428 void 1429 filecaps_move(struct filecaps *src, struct filecaps *dst) 1430 { 1431 1432 *dst = *src; 1433 bzero(src, sizeof(*src)); 1434 } 1435 1436 /* 1437 * Fill the given filecaps structure with full rights. 1438 */ 1439 static void 1440 filecaps_fill(struct filecaps *fcaps) 1441 { 1442 1443 CAP_ALL(&fcaps->fc_rights); 1444 fcaps->fc_ioctls = NULL; 1445 fcaps->fc_nioctls = -1; 1446 fcaps->fc_fcntls = CAP_FCNTL_ALL; 1447 } 1448 1449 /* 1450 * Free memory allocated within filecaps structure. 1451 */ 1452 void 1453 filecaps_free(struct filecaps *fcaps) 1454 { 1455 1456 free(fcaps->fc_ioctls, M_FILECAPS); 1457 bzero(fcaps, sizeof(*fcaps)); 1458 } 1459 1460 /* 1461 * Validate the given filecaps structure. 1462 */ 1463 static void 1464 filecaps_validate(const struct filecaps *fcaps, const char *func) 1465 { 1466 1467 KASSERT(cap_rights_is_valid(&fcaps->fc_rights), 1468 ("%s: invalid rights", func)); 1469 KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, 1470 ("%s: invalid fcntls", func)); 1471 KASSERT(fcaps->fc_fcntls == 0 || 1472 cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), 1473 ("%s: fcntls without CAP_FCNTL", func)); 1474 KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : 1475 (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), 1476 ("%s: invalid ioctls", func)); 1477 KASSERT(fcaps->fc_nioctls == 0 || 1478 cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), 1479 ("%s: ioctls without CAP_IOCTL", func)); 1480 } 1481 1482 static void 1483 fdgrowtable_exp(struct filedesc *fdp, int nfd) 1484 { 1485 int nfd1; 1486 1487 FILEDESC_XLOCK_ASSERT(fdp); 1488 1489 nfd1 = fdp->fd_nfiles * 2; 1490 if (nfd1 < nfd) 1491 nfd1 = nfd; 1492 fdgrowtable(fdp, nfd1); 1493 } 1494 1495 /* 1496 * Grow the file table to accomodate (at least) nfd descriptors. 1497 */ 1498 static void 1499 fdgrowtable(struct filedesc *fdp, int nfd) 1500 { 1501 struct filedesc0 *fdp0; 1502 struct freetable *ft; 1503 struct filedescent *ntable; 1504 struct filedescent *otable; 1505 int nnfiles, onfiles; 1506 NDSLOTTYPE *nmap, *omap; 1507 1508 FILEDESC_XLOCK_ASSERT(fdp); 1509 1510 KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); 1511 1512 /* save old values */ 1513 onfiles = fdp->fd_nfiles; 1514 otable = fdp->fd_ofiles; 1515 omap = fdp->fd_map; 1516 1517 /* compute the size of the new table */ 1518 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ 1519 if (nnfiles <= onfiles) 1520 /* the table is already large enough */ 1521 return; 1522 1523 /* 1524 * Allocate a new table. We need enough space for the 1525 * file entries themselves and the struct freetable we will use 1526 * when we decommission the table and place it on the freelist. 1527 * We place the struct freetable in the middle so we don't have 1528 * to worry about padding. 1529 */ 1530 ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable), 1531 M_FILEDESC, M_ZERO | M_WAITOK); 1532 /* copy the old data over and point at the new tables */ 1533 memcpy(ntable, otable, onfiles * sizeof(*otable)); 1534 fdp->fd_ofiles = ntable; 1535 1536 /* 1537 * Allocate a new map only if the old is not large enough. It will 1538 * grow at a slower rate than the table as it can map more 1539 * entries than the table can hold. 1540 */ 1541 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { 1542 nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, 1543 M_ZERO | M_WAITOK); 1544 /* copy over the old data and update the pointer */ 1545 memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); 1546 fdp->fd_map = nmap; 1547 } 1548 1549 /* 1550 * In order to have a valid pattern for fget_unlocked() 1551 * fdp->fd_nfiles must be the last member to be updated, otherwise 1552 * fget_unlocked() consumers may reference a new, higher value for 1553 * fdp->fd_nfiles before to access the fdp->fd_ofiles array, 1554 * resulting in OOB accesses. 1555 */ 1556 atomic_store_rel_int(&fdp->fd_nfiles, nnfiles); 1557 1558 /* 1559 * Do not free the old file table, as some threads may still 1560 * reference entries within it. Instead, place it on a freelist 1561 * which will be processed when the struct filedesc is released. 1562 * 1563 * Note that if onfiles == NDFILE, we're dealing with the original 1564 * static allocation contained within (struct filedesc0 *)fdp, 1565 * which must not be freed. 1566 */ 1567 if (onfiles > NDFILE) { 1568 ft = (struct freetable *)&otable[onfiles]; 1569 fdp0 = (struct filedesc0 *)fdp; 1570 ft->ft_table = otable; 1571 SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); 1572 } 1573 /* 1574 * The map does not have the same possibility of threads still 1575 * holding references to it. So always free it as long as it 1576 * does not reference the original static allocation. 1577 */ 1578 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) 1579 free(omap, M_FILEDESC); 1580 } 1581 1582 /* 1583 * Allocate a file descriptor for the process. 1584 */ 1585 int 1586 fdalloc(struct thread *td, int minfd, int *result) 1587 { 1588 struct proc *p = td->td_proc; 1589 struct filedesc *fdp = p->p_fd; 1590 int fd = -1, maxfd, allocfd; 1591 #ifdef RACCT 1592 int error; 1593 #endif 1594 1595 FILEDESC_XLOCK_ASSERT(fdp); 1596 1597 if (fdp->fd_freefile > minfd) 1598 minfd = fdp->fd_freefile; 1599 1600 maxfd = getmaxfd(p); 1601 1602 /* 1603 * Search the bitmap for a free descriptor starting at minfd. 1604 * If none is found, grow the file table. 1605 */ 1606 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); 1607 if (fd >= maxfd) 1608 return (EMFILE); 1609 if (fd >= fdp->fd_nfiles) { 1610 allocfd = min(fd * 2, maxfd); 1611 #ifdef RACCT 1612 PROC_LOCK(p); 1613 error = racct_set(p, RACCT_NOFILE, allocfd); 1614 PROC_UNLOCK(p); 1615 if (error != 0) 1616 return (EMFILE); 1617 #endif 1618 /* 1619 * fd is already equal to first free descriptor >= minfd, so 1620 * we only need to grow the table and we are done. 1621 */ 1622 fdgrowtable_exp(fdp, allocfd); 1623 } 1624 1625 /* 1626 * Perform some sanity checks, then mark the file descriptor as 1627 * used and return it to the caller. 1628 */ 1629 KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), 1630 ("invalid descriptor %d", fd)); 1631 KASSERT(!fdisused(fdp, fd), 1632 ("fd_first_free() returned non-free descriptor")); 1633 KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, 1634 ("file descriptor isn't free")); 1635 KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set")); 1636 fdused(fdp, fd); 1637 *result = fd; 1638 return (0); 1639 } 1640 1641 /* 1642 * Allocate n file descriptors for the process. 1643 */ 1644 int 1645 fdallocn(struct thread *td, int minfd, int *fds, int n) 1646 { 1647 struct proc *p = td->td_proc; 1648 struct filedesc *fdp = p->p_fd; 1649 int i; 1650 1651 FILEDESC_XLOCK_ASSERT(fdp); 1652 1653 for (i = 0; i < n; i++) 1654 if (fdalloc(td, 0, &fds[i]) != 0) 1655 break; 1656 1657 if (i < n) { 1658 for (i--; i >= 0; i--) 1659 fdunused(fdp, fds[i]); 1660 return (EMFILE); 1661 } 1662 1663 return (0); 1664 } 1665 1666 /* 1667 * Create a new open file structure and allocate a file decriptor for the 1668 * process that refers to it. We add one reference to the file for the 1669 * descriptor table and one reference for resultfp. This is to prevent us 1670 * being preempted and the entry in the descriptor table closed after we 1671 * release the FILEDESC lock. 1672 */ 1673 int 1674 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags) 1675 { 1676 struct file *fp; 1677 int error, fd; 1678 1679 error = falloc_noinstall(td, &fp); 1680 if (error) 1681 return (error); /* no reference held on error */ 1682 1683 error = finstall(td, fp, &fd, flags, NULL); 1684 if (error) { 1685 fdrop(fp, td); /* one reference (fp only) */ 1686 return (error); 1687 } 1688 1689 if (resultfp != NULL) 1690 *resultfp = fp; /* copy out result */ 1691 else 1692 fdrop(fp, td); /* release local reference */ 1693 1694 if (resultfd != NULL) 1695 *resultfd = fd; 1696 1697 return (0); 1698 } 1699 1700 /* 1701 * Create a new open file structure without allocating a file descriptor. 1702 */ 1703 int 1704 falloc_noinstall(struct thread *td, struct file **resultfp) 1705 { 1706 struct file *fp; 1707 int maxuserfiles = maxfiles - (maxfiles / 20); 1708 static struct timeval lastfail; 1709 static int curfail; 1710 1711 KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); 1712 1713 if ((openfiles >= maxuserfiles && 1714 priv_check(td, PRIV_MAXFILES) != 0) || 1715 openfiles >= maxfiles) { 1716 if (ppsratecheck(&lastfail, &curfail, 1)) { 1717 printf("kern.maxfiles limit exceeded by uid %i, " 1718 "please see tuning(7).\n", td->td_ucred->cr_ruid); 1719 } 1720 return (ENFILE); 1721 } 1722 atomic_add_int(&openfiles, 1); 1723 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO); 1724 refcount_init(&fp->f_count, 1); 1725 fp->f_cred = crhold(td->td_ucred); 1726 fp->f_ops = &badfileops; 1727 fp->f_data = NULL; 1728 fp->f_vnode = NULL; 1729 *resultfp = fp; 1730 return (0); 1731 } 1732 1733 /* 1734 * Install a file in a file descriptor table. 1735 */ 1736 int 1737 finstall(struct thread *td, struct file *fp, int *fd, int flags, 1738 struct filecaps *fcaps) 1739 { 1740 struct filedesc *fdp = td->td_proc->p_fd; 1741 struct filedescent *fde; 1742 int error; 1743 1744 KASSERT(fd != NULL, ("%s: fd == NULL", __func__)); 1745 KASSERT(fp != NULL, ("%s: fp == NULL", __func__)); 1746 if (fcaps != NULL) 1747 filecaps_validate(fcaps, __func__); 1748 1749 FILEDESC_XLOCK(fdp); 1750 if ((error = fdalloc(td, 0, fd))) { 1751 FILEDESC_XUNLOCK(fdp); 1752 return (error); 1753 } 1754 fhold(fp); 1755 fde = &fdp->fd_ofiles[*fd]; 1756 fde->fde_file = fp; 1757 if ((flags & O_CLOEXEC) != 0) 1758 fde->fde_flags |= UF_EXCLOSE; 1759 if (fcaps != NULL) 1760 filecaps_move(fcaps, &fde->fde_caps); 1761 else 1762 filecaps_fill(&fde->fde_caps); 1763 FILEDESC_XUNLOCK(fdp); 1764 return (0); 1765 } 1766 1767 /* 1768 * Build a new filedesc structure from another. 1769 * Copy the current, root, and jail root vnode references. 1770 */ 1771 struct filedesc * 1772 fdinit(struct filedesc *fdp) 1773 { 1774 struct filedesc0 *newfdp; 1775 1776 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO); 1777 FILEDESC_LOCK_INIT(&newfdp->fd_fd); 1778 if (fdp != NULL) { 1779 FILEDESC_SLOCK(fdp); 1780 newfdp->fd_fd.fd_cdir = fdp->fd_cdir; 1781 if (newfdp->fd_fd.fd_cdir) 1782 VREF(newfdp->fd_fd.fd_cdir); 1783 newfdp->fd_fd.fd_rdir = fdp->fd_rdir; 1784 if (newfdp->fd_fd.fd_rdir) 1785 VREF(newfdp->fd_fd.fd_rdir); 1786 newfdp->fd_fd.fd_jdir = fdp->fd_jdir; 1787 if (newfdp->fd_fd.fd_jdir) 1788 VREF(newfdp->fd_fd.fd_jdir); 1789 FILEDESC_SUNLOCK(fdp); 1790 } 1791 1792 /* Create the file descriptor table. */ 1793 newfdp->fd_fd.fd_refcnt = 1; 1794 newfdp->fd_fd.fd_holdcnt = 1; 1795 newfdp->fd_fd.fd_cmask = CMASK; 1796 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles; 1797 newfdp->fd_fd.fd_nfiles = NDFILE; 1798 newfdp->fd_fd.fd_map = newfdp->fd_dmap; 1799 newfdp->fd_fd.fd_lastfile = -1; 1800 return (&newfdp->fd_fd); 1801 } 1802 1803 static struct filedesc * 1804 fdhold(struct proc *p) 1805 { 1806 struct filedesc *fdp; 1807 1808 mtx_lock(&fdesc_mtx); 1809 fdp = p->p_fd; 1810 if (fdp != NULL) 1811 fdp->fd_holdcnt++; 1812 mtx_unlock(&fdesc_mtx); 1813 return (fdp); 1814 } 1815 1816 static void 1817 fddrop(struct filedesc *fdp) 1818 { 1819 struct filedesc0 *fdp0; 1820 struct freetable *ft; 1821 int i; 1822 1823 mtx_lock(&fdesc_mtx); 1824 i = --fdp->fd_holdcnt; 1825 mtx_unlock(&fdesc_mtx); 1826 if (i > 0) 1827 return; 1828 1829 FILEDESC_LOCK_DESTROY(fdp); 1830 fdp0 = (struct filedesc0 *)fdp; 1831 while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) { 1832 SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next); 1833 free(ft->ft_table, M_FILEDESC); 1834 } 1835 free(fdp, M_FILEDESC); 1836 } 1837 1838 /* 1839 * Share a filedesc structure. 1840 */ 1841 struct filedesc * 1842 fdshare(struct filedesc *fdp) 1843 { 1844 1845 FILEDESC_XLOCK(fdp); 1846 fdp->fd_refcnt++; 1847 FILEDESC_XUNLOCK(fdp); 1848 return (fdp); 1849 } 1850 1851 /* 1852 * Unshare a filedesc structure, if necessary by making a copy 1853 */ 1854 void 1855 fdunshare(struct proc *p, struct thread *td) 1856 { 1857 1858 FILEDESC_XLOCK(p->p_fd); 1859 if (p->p_fd->fd_refcnt > 1) { 1860 struct filedesc *tmp; 1861 1862 FILEDESC_XUNLOCK(p->p_fd); 1863 tmp = fdcopy(p->p_fd); 1864 fdescfree(td); 1865 p->p_fd = tmp; 1866 } else 1867 FILEDESC_XUNLOCK(p->p_fd); 1868 } 1869 1870 /* 1871 * Copy a filedesc structure. A NULL pointer in returns a NULL reference, 1872 * this is to ease callers, not catch errors. 1873 */ 1874 struct filedesc * 1875 fdcopy(struct filedesc *fdp) 1876 { 1877 struct filedesc *newfdp; 1878 struct filedescent *nfde, *ofde; 1879 int i; 1880 1881 /* Certain daemons might not have file descriptors. */ 1882 if (fdp == NULL) 1883 return (NULL); 1884 1885 newfdp = fdinit(fdp); 1886 FILEDESC_SLOCK(fdp); 1887 while (fdp->fd_lastfile >= newfdp->fd_nfiles) { 1888 FILEDESC_SUNLOCK(fdp); 1889 FILEDESC_XLOCK(newfdp); 1890 fdgrowtable(newfdp, fdp->fd_lastfile + 1); 1891 FILEDESC_XUNLOCK(newfdp); 1892 FILEDESC_SLOCK(fdp); 1893 } 1894 /* copy all passable descriptors (i.e. not kqueue) */ 1895 newfdp->fd_freefile = -1; 1896 for (i = 0; i <= fdp->fd_lastfile; ++i) { 1897 ofde = &fdp->fd_ofiles[i]; 1898 if (fdisused(fdp, i) && 1899 (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) && 1900 ofde->fde_file->f_ops != &badfileops) { 1901 nfde = &newfdp->fd_ofiles[i]; 1902 *nfde = *ofde; 1903 filecaps_copy(&ofde->fde_caps, &nfde->fde_caps); 1904 fhold(nfde->fde_file); 1905 newfdp->fd_lastfile = i; 1906 } else { 1907 if (newfdp->fd_freefile == -1) 1908 newfdp->fd_freefile = i; 1909 } 1910 } 1911 newfdp->fd_cmask = fdp->fd_cmask; 1912 FILEDESC_SUNLOCK(fdp); 1913 FILEDESC_XLOCK(newfdp); 1914 for (i = 0; i <= newfdp->fd_lastfile; ++i) { 1915 if (newfdp->fd_ofiles[i].fde_file != NULL) 1916 fdused(newfdp, i); 1917 } 1918 if (newfdp->fd_freefile == -1) 1919 newfdp->fd_freefile = i; 1920 FILEDESC_XUNLOCK(newfdp); 1921 return (newfdp); 1922 } 1923 1924 /* 1925 * Release a filedesc structure. 1926 */ 1927 void 1928 fdescfree(struct thread *td) 1929 { 1930 struct filedesc *fdp; 1931 int i; 1932 struct filedesc_to_leader *fdtol; 1933 struct file *fp; 1934 struct vnode *cdir, *jdir, *rdir, *vp; 1935 struct flock lf; 1936 1937 /* Certain daemons might not have file descriptors. */ 1938 fdp = td->td_proc->p_fd; 1939 if (fdp == NULL) 1940 return; 1941 1942 #ifdef RACCT 1943 PROC_LOCK(td->td_proc); 1944 racct_set(td->td_proc, RACCT_NOFILE, 0); 1945 PROC_UNLOCK(td->td_proc); 1946 #endif 1947 1948 /* Check for special need to clear POSIX style locks */ 1949 fdtol = td->td_proc->p_fdtol; 1950 if (fdtol != NULL) { 1951 FILEDESC_XLOCK(fdp); 1952 KASSERT(fdtol->fdl_refcount > 0, 1953 ("filedesc_to_refcount botch: fdl_refcount=%d", 1954 fdtol->fdl_refcount)); 1955 if (fdtol->fdl_refcount == 1 && 1956 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1957 for (i = 0; i <= fdp->fd_lastfile; i++) { 1958 fp = fdp->fd_ofiles[i].fde_file; 1959 if (fp == NULL || fp->f_type != DTYPE_VNODE) 1960 continue; 1961 fhold(fp); 1962 FILEDESC_XUNLOCK(fdp); 1963 lf.l_whence = SEEK_SET; 1964 lf.l_start = 0; 1965 lf.l_len = 0; 1966 lf.l_type = F_UNLCK; 1967 vp = fp->f_vnode; 1968 (void) VOP_ADVLOCK(vp, 1969 (caddr_t)td->td_proc->p_leader, F_UNLCK, 1970 &lf, F_POSIX); 1971 FILEDESC_XLOCK(fdp); 1972 fdrop(fp, td); 1973 } 1974 } 1975 retry: 1976 if (fdtol->fdl_refcount == 1) { 1977 if (fdp->fd_holdleaderscount > 0 && 1978 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 1979 /* 1980 * close() or do_dup() has cleared a reference 1981 * in a shared file descriptor table. 1982 */ 1983 fdp->fd_holdleaderswakeup = 1; 1984 sx_sleep(&fdp->fd_holdleaderscount, 1985 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); 1986 goto retry; 1987 } 1988 if (fdtol->fdl_holdcount > 0) { 1989 /* 1990 * Ensure that fdtol->fdl_leader remains 1991 * valid in closef(). 1992 */ 1993 fdtol->fdl_wakeup = 1; 1994 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, 1995 "fdlhold", 0); 1996 goto retry; 1997 } 1998 } 1999 fdtol->fdl_refcount--; 2000 if (fdtol->fdl_refcount == 0 && 2001 fdtol->fdl_holdcount == 0) { 2002 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; 2003 fdtol->fdl_prev->fdl_next = fdtol->fdl_next; 2004 } else 2005 fdtol = NULL; 2006 td->td_proc->p_fdtol = NULL; 2007 FILEDESC_XUNLOCK(fdp); 2008 if (fdtol != NULL) 2009 free(fdtol, M_FILEDESC_TO_LEADER); 2010 } 2011 FILEDESC_XLOCK(fdp); 2012 i = --fdp->fd_refcnt; 2013 FILEDESC_XUNLOCK(fdp); 2014 if (i > 0) 2015 return; 2016 2017 for (i = 0; i <= fdp->fd_lastfile; i++) { 2018 fp = fdp->fd_ofiles[i].fde_file; 2019 if (fp != NULL) { 2020 FILEDESC_XLOCK(fdp); 2021 fdfree(fdp, i); 2022 FILEDESC_XUNLOCK(fdp); 2023 (void) closef(fp, td); 2024 } 2025 } 2026 FILEDESC_XLOCK(fdp); 2027 2028 /* XXX This should happen earlier. */ 2029 mtx_lock(&fdesc_mtx); 2030 td->td_proc->p_fd = NULL; 2031 mtx_unlock(&fdesc_mtx); 2032 2033 if (fdp->fd_nfiles > NDFILE) 2034 free(fdp->fd_ofiles, M_FILEDESC); 2035 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) 2036 free(fdp->fd_map, M_FILEDESC); 2037 2038 fdp->fd_nfiles = 0; 2039 2040 cdir = fdp->fd_cdir; 2041 fdp->fd_cdir = NULL; 2042 rdir = fdp->fd_rdir; 2043 fdp->fd_rdir = NULL; 2044 jdir = fdp->fd_jdir; 2045 fdp->fd_jdir = NULL; 2046 FILEDESC_XUNLOCK(fdp); 2047 2048 if (cdir != NULL) 2049 vrele(cdir); 2050 if (rdir != NULL) 2051 vrele(rdir); 2052 if (jdir != NULL) 2053 vrele(jdir); 2054 2055 fddrop(fdp); 2056 } 2057 2058 /* 2059 * For setugid programs, we don't want to people to use that setugidness 2060 * to generate error messages which write to a file which otherwise would 2061 * otherwise be off-limits to the process. We check for filesystems where 2062 * the vnode can change out from under us after execve (like [lin]procfs). 2063 * 2064 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is 2065 * sufficient. We also don't check for setugidness since we know we are. 2066 */ 2067 static int 2068 is_unsafe(struct file *fp) 2069 { 2070 if (fp->f_type == DTYPE_VNODE) { 2071 struct vnode *vp = fp->f_vnode; 2072 2073 if ((vp->v_vflag & VV_PROCDEP) != 0) 2074 return (1); 2075 } 2076 return (0); 2077 } 2078 2079 /* 2080 * Make this setguid thing safe, if at all possible. 2081 */ 2082 void 2083 setugidsafety(struct thread *td) 2084 { 2085 struct filedesc *fdp; 2086 struct file *fp; 2087 int i; 2088 2089 /* Certain daemons might not have file descriptors. */ 2090 fdp = td->td_proc->p_fd; 2091 if (fdp == NULL) 2092 return; 2093 2094 /* 2095 * Note: fdp->fd_ofiles may be reallocated out from under us while 2096 * we are blocked in a close. Be careful! 2097 */ 2098 FILEDESC_XLOCK(fdp); 2099 for (i = 0; i <= fdp->fd_lastfile; i++) { 2100 if (i > 2) 2101 break; 2102 fp = fdp->fd_ofiles[i].fde_file; 2103 if (fp != NULL && is_unsafe(fp)) { 2104 knote_fdclose(td, i); 2105 /* 2106 * NULL-out descriptor prior to close to avoid 2107 * a race while close blocks. 2108 */ 2109 fdfree(fdp, i); 2110 FILEDESC_XUNLOCK(fdp); 2111 (void) closef(fp, td); 2112 FILEDESC_XLOCK(fdp); 2113 } 2114 } 2115 FILEDESC_XUNLOCK(fdp); 2116 } 2117 2118 /* 2119 * If a specific file object occupies a specific file descriptor, close the 2120 * file descriptor entry and drop a reference on the file object. This is a 2121 * convenience function to handle a subsequent error in a function that calls 2122 * falloc() that handles the race that another thread might have closed the 2123 * file descriptor out from under the thread creating the file object. 2124 */ 2125 void 2126 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td) 2127 { 2128 2129 FILEDESC_XLOCK(fdp); 2130 if (fdp->fd_ofiles[idx].fde_file == fp) { 2131 fdfree(fdp, idx); 2132 FILEDESC_XUNLOCK(fdp); 2133 fdrop(fp, td); 2134 } else 2135 FILEDESC_XUNLOCK(fdp); 2136 } 2137 2138 /* 2139 * Close any files on exec? 2140 */ 2141 void 2142 fdcloseexec(struct thread *td) 2143 { 2144 struct filedesc *fdp; 2145 struct filedescent *fde; 2146 struct file *fp; 2147 int i; 2148 2149 /* Certain daemons might not have file descriptors. */ 2150 fdp = td->td_proc->p_fd; 2151 if (fdp == NULL) 2152 return; 2153 2154 /* 2155 * We cannot cache fd_ofiles since operations 2156 * may block and rip them out from under us. 2157 */ 2158 FILEDESC_XLOCK(fdp); 2159 for (i = 0; i <= fdp->fd_lastfile; i++) { 2160 fde = &fdp->fd_ofiles[i]; 2161 fp = fde->fde_file; 2162 if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || 2163 (fde->fde_flags & UF_EXCLOSE))) { 2164 fdfree(fdp, i); 2165 (void) closefp(fdp, i, fp, td, 0); 2166 /* closefp() drops the FILEDESC lock. */ 2167 FILEDESC_XLOCK(fdp); 2168 } 2169 } 2170 FILEDESC_XUNLOCK(fdp); 2171 } 2172 2173 /* 2174 * It is unsafe for set[ug]id processes to be started with file 2175 * descriptors 0..2 closed, as these descriptors are given implicit 2176 * significance in the Standard C library. fdcheckstd() will create a 2177 * descriptor referencing /dev/null for each of stdin, stdout, and 2178 * stderr that is not already open. 2179 */ 2180 int 2181 fdcheckstd(struct thread *td) 2182 { 2183 struct filedesc *fdp; 2184 register_t retval, save; 2185 int i, error, devnull; 2186 2187 fdp = td->td_proc->p_fd; 2188 if (fdp == NULL) 2189 return (0); 2190 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); 2191 devnull = -1; 2192 error = 0; 2193 for (i = 0; i < 3; i++) { 2194 if (fdp->fd_ofiles[i].fde_file != NULL) 2195 continue; 2196 if (devnull < 0) { 2197 save = td->td_retval[0]; 2198 error = kern_open(td, "/dev/null", UIO_SYSSPACE, 2199 O_RDWR, 0); 2200 devnull = td->td_retval[0]; 2201 td->td_retval[0] = save; 2202 if (error) 2203 break; 2204 KASSERT(devnull == i, ("oof, we didn't get our fd")); 2205 } else { 2206 error = do_dup(td, DUP_FIXED, devnull, i, &retval); 2207 if (error != 0) 2208 break; 2209 } 2210 } 2211 return (error); 2212 } 2213 2214 /* 2215 * Internal form of close. Decrement reference count on file structure. 2216 * Note: td may be NULL when closing a file that was being passed in a 2217 * message. 2218 * 2219 * XXXRW: Giant is not required for the caller, but often will be held; this 2220 * makes it moderately likely the Giant will be recursed in the VFS case. 2221 */ 2222 int 2223 closef(struct file *fp, struct thread *td) 2224 { 2225 struct vnode *vp; 2226 struct flock lf; 2227 struct filedesc_to_leader *fdtol; 2228 struct filedesc *fdp; 2229 2230 /* 2231 * POSIX record locking dictates that any close releases ALL 2232 * locks owned by this process. This is handled by setting 2233 * a flag in the unlock to free ONLY locks obeying POSIX 2234 * semantics, and not to free BSD-style file locks. 2235 * If the descriptor was in a message, POSIX-style locks 2236 * aren't passed with the descriptor, and the thread pointer 2237 * will be NULL. Callers should be careful only to pass a 2238 * NULL thread pointer when there really is no owning 2239 * context that might have locks, or the locks will be 2240 * leaked. 2241 */ 2242 if (fp->f_type == DTYPE_VNODE && td != NULL) { 2243 vp = fp->f_vnode; 2244 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { 2245 lf.l_whence = SEEK_SET; 2246 lf.l_start = 0; 2247 lf.l_len = 0; 2248 lf.l_type = F_UNLCK; 2249 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, 2250 F_UNLCK, &lf, F_POSIX); 2251 } 2252 fdtol = td->td_proc->p_fdtol; 2253 if (fdtol != NULL) { 2254 /* 2255 * Handle special case where file descriptor table is 2256 * shared between multiple process leaders. 2257 */ 2258 fdp = td->td_proc->p_fd; 2259 FILEDESC_XLOCK(fdp); 2260 for (fdtol = fdtol->fdl_next; 2261 fdtol != td->td_proc->p_fdtol; 2262 fdtol = fdtol->fdl_next) { 2263 if ((fdtol->fdl_leader->p_flag & 2264 P_ADVLOCK) == 0) 2265 continue; 2266 fdtol->fdl_holdcount++; 2267 FILEDESC_XUNLOCK(fdp); 2268 lf.l_whence = SEEK_SET; 2269 lf.l_start = 0; 2270 lf.l_len = 0; 2271 lf.l_type = F_UNLCK; 2272 vp = fp->f_vnode; 2273 (void) VOP_ADVLOCK(vp, 2274 (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, 2275 F_POSIX); 2276 FILEDESC_XLOCK(fdp); 2277 fdtol->fdl_holdcount--; 2278 if (fdtol->fdl_holdcount == 0 && 2279 fdtol->fdl_wakeup != 0) { 2280 fdtol->fdl_wakeup = 0; 2281 wakeup(fdtol); 2282 } 2283 } 2284 FILEDESC_XUNLOCK(fdp); 2285 } 2286 } 2287 return (fdrop(fp, td)); 2288 } 2289 2290 /* 2291 * Initialize the file pointer with the specified properties. 2292 * 2293 * The ops are set with release semantics to be certain that the flags, type, 2294 * and data are visible when ops is. This is to prevent ops methods from being 2295 * called with bad data. 2296 */ 2297 void 2298 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) 2299 { 2300 fp->f_data = data; 2301 fp->f_flag = flag; 2302 fp->f_type = type; 2303 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); 2304 } 2305 2306 int 2307 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, 2308 int needfcntl, struct file **fpp, cap_rights_t *haverightsp) 2309 { 2310 struct file *fp; 2311 u_int count; 2312 #ifdef CAPABILITIES 2313 cap_rights_t haverights; 2314 int error; 2315 #endif 2316 2317 /* 2318 * Avoid reads reordering and then a first access to the 2319 * fdp->fd_ofiles table which could result in OOB operation. 2320 */ 2321 if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles)) 2322 return (EBADF); 2323 /* 2324 * Fetch the descriptor locklessly. We avoid fdrop() races by 2325 * never raising a refcount above 0. To accomplish this we have 2326 * to use a cmpset loop rather than an atomic_add. The descriptor 2327 * must be re-verified once we acquire a reference to be certain 2328 * that the identity is still correct and we did not lose a race 2329 * due to preemption. 2330 */ 2331 for (;;) { 2332 fp = fdp->fd_ofiles[fd].fde_file; 2333 if (fp == NULL) 2334 return (EBADF); 2335 #ifdef CAPABILITIES 2336 haverights = *cap_rights(fdp, fd); 2337 if (needrightsp != NULL) { 2338 error = cap_check(&haverights, needrightsp); 2339 if (error != 0) 2340 return (error); 2341 if (cap_rights_is_set(needrightsp, CAP_FCNTL)) { 2342 error = cap_fcntl_check(fdp, fd, needfcntl); 2343 if (error != 0) 2344 return (error); 2345 } 2346 } 2347 #endif 2348 count = fp->f_count; 2349 if (count == 0) 2350 continue; 2351 /* 2352 * Use an acquire barrier to prevent caching of fd_ofiles 2353 * so it is refreshed for verification. 2354 */ 2355 if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1) 2356 continue; 2357 if (fp == fdp->fd_ofiles[fd].fde_file) 2358 break; 2359 fdrop(fp, curthread); 2360 } 2361 *fpp = fp; 2362 if (haverightsp != NULL) { 2363 #ifdef CAPABILITIES 2364 *haverightsp = haverights; 2365 #else 2366 CAP_ALL(haverightsp); 2367 #endif 2368 } 2369 return (0); 2370 } 2371 2372 /* 2373 * Extract the file pointer associated with the specified descriptor for the 2374 * current user process. 2375 * 2376 * If the descriptor doesn't exist or doesn't match 'flags', EBADF is 2377 * returned. 2378 * 2379 * File's rights will be checked against the capability rights mask. 2380 * 2381 * If an error occured the non-zero error is returned and *fpp is set to 2382 * NULL. Otherwise *fpp is held and set and zero is returned. Caller is 2383 * responsible for fdrop(). 2384 */ 2385 static __inline int 2386 _fget(struct thread *td, int fd, struct file **fpp, int flags, 2387 cap_rights_t *needrightsp, u_char *maxprotp) 2388 { 2389 struct filedesc *fdp; 2390 struct file *fp; 2391 cap_rights_t haverights, needrights; 2392 int error; 2393 2394 *fpp = NULL; 2395 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2396 return (EBADF); 2397 if (needrightsp != NULL) 2398 needrights = *needrightsp; 2399 else 2400 cap_rights_init(&needrights); 2401 if (maxprotp != NULL) 2402 cap_rights_set(&needrights, CAP_MMAP); 2403 error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights); 2404 if (error != 0) 2405 return (error); 2406 if (fp->f_ops == &badfileops) { 2407 fdrop(fp, td); 2408 return (EBADF); 2409 } 2410 2411 #ifdef CAPABILITIES 2412 /* 2413 * If requested, convert capability rights to access flags. 2414 */ 2415 if (maxprotp != NULL) 2416 *maxprotp = cap_rights_to_vmprot(&haverights); 2417 #else /* !CAPABILITIES */ 2418 if (maxprotp != NULL) 2419 *maxprotp = VM_PROT_ALL; 2420 #endif /* CAPABILITIES */ 2421 2422 /* 2423 * FREAD and FWRITE failure return EBADF as per POSIX. 2424 */ 2425 error = 0; 2426 switch (flags) { 2427 case FREAD: 2428 case FWRITE: 2429 if ((fp->f_flag & flags) == 0) 2430 error = EBADF; 2431 break; 2432 case FEXEC: 2433 if ((fp->f_flag & (FREAD | FEXEC)) == 0 || 2434 ((fp->f_flag & FWRITE) != 0)) 2435 error = EBADF; 2436 break; 2437 case 0: 2438 break; 2439 default: 2440 KASSERT(0, ("wrong flags")); 2441 } 2442 2443 if (error != 0) { 2444 fdrop(fp, td); 2445 return (error); 2446 } 2447 2448 *fpp = fp; 2449 return (0); 2450 } 2451 2452 int 2453 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2454 { 2455 2456 return(_fget(td, fd, fpp, 0, rightsp, NULL)); 2457 } 2458 2459 int 2460 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, 2461 struct file **fpp) 2462 { 2463 2464 return (_fget(td, fd, fpp, 0, rightsp, maxprotp)); 2465 } 2466 2467 int 2468 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2469 { 2470 2471 return(_fget(td, fd, fpp, FREAD, rightsp, NULL)); 2472 } 2473 2474 int 2475 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) 2476 { 2477 2478 return (_fget(td, fd, fpp, FWRITE, rightsp, NULL)); 2479 } 2480 2481 /* 2482 * Like fget() but loads the underlying vnode, or returns an error if the 2483 * descriptor does not represent a vnode. Note that pipes use vnodes but 2484 * never have VM objects. The returned vnode will be vref()'d. 2485 * 2486 * XXX: what about the unused flags ? 2487 */ 2488 static __inline int 2489 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, 2490 struct vnode **vpp) 2491 { 2492 struct file *fp; 2493 int error; 2494 2495 *vpp = NULL; 2496 error = _fget(td, fd, &fp, flags, needrightsp, NULL); 2497 if (error != 0) 2498 return (error); 2499 if (fp->f_vnode == NULL) { 2500 error = EINVAL; 2501 } else { 2502 *vpp = fp->f_vnode; 2503 vref(*vpp); 2504 } 2505 fdrop(fp, td); 2506 2507 return (error); 2508 } 2509 2510 int 2511 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2512 { 2513 2514 return (_fgetvp(td, fd, 0, rightsp, vpp)); 2515 } 2516 2517 int 2518 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, 2519 struct filecaps *havecaps, struct vnode **vpp) 2520 { 2521 struct filedesc *fdp; 2522 struct file *fp; 2523 #ifdef CAPABILITIES 2524 int error; 2525 #endif 2526 2527 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL) 2528 return (EBADF); 2529 2530 fp = fget_locked(fdp, fd); 2531 if (fp == NULL || fp->f_ops == &badfileops) 2532 return (EBADF); 2533 2534 #ifdef CAPABILITIES 2535 if (needrightsp != NULL) { 2536 error = cap_check(cap_rights(fdp, fd), needrightsp); 2537 if (error != 0) 2538 return (error); 2539 } 2540 #endif 2541 2542 if (fp->f_vnode == NULL) 2543 return (EINVAL); 2544 2545 *vpp = fp->f_vnode; 2546 vref(*vpp); 2547 filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps); 2548 2549 return (0); 2550 } 2551 2552 int 2553 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2554 { 2555 2556 return (_fgetvp(td, fd, FREAD, rightsp, vpp)); 2557 } 2558 2559 int 2560 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) 2561 { 2562 2563 return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); 2564 } 2565 2566 #ifdef notyet 2567 int 2568 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, 2569 struct vnode **vpp) 2570 { 2571 2572 return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); 2573 } 2574 #endif 2575 2576 /* 2577 * Like fget() but loads the underlying socket, or returns an error if the 2578 * descriptor does not represent a socket. 2579 * 2580 * We bump the ref count on the returned socket. XXX Also obtain the SX lock 2581 * in the future. 2582 * 2583 * Note: fgetsock() and fputsock() are deprecated, as consumers should rely 2584 * on their file descriptor reference to prevent the socket from being free'd 2585 * during use. 2586 */ 2587 int 2588 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, 2589 u_int *fflagp) 2590 { 2591 struct file *fp; 2592 int error; 2593 2594 *spp = NULL; 2595 if (fflagp != NULL) 2596 *fflagp = 0; 2597 if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0) 2598 return (error); 2599 if (fp->f_type != DTYPE_SOCKET) { 2600 error = ENOTSOCK; 2601 } else { 2602 *spp = fp->f_data; 2603 if (fflagp) 2604 *fflagp = fp->f_flag; 2605 SOCK_LOCK(*spp); 2606 soref(*spp); 2607 SOCK_UNLOCK(*spp); 2608 } 2609 fdrop(fp, td); 2610 2611 return (error); 2612 } 2613 2614 /* 2615 * Drop the reference count on the socket and XXX release the SX lock in the 2616 * future. The last reference closes the socket. 2617 * 2618 * Note: fputsock() is deprecated, see comment for fgetsock(). 2619 */ 2620 void 2621 fputsock(struct socket *so) 2622 { 2623 2624 ACCEPT_LOCK(); 2625 SOCK_LOCK(so); 2626 CURVNET_SET(so->so_vnet); 2627 sorele(so); 2628 CURVNET_RESTORE(); 2629 } 2630 2631 /* 2632 * Handle the last reference to a file being closed. 2633 */ 2634 int 2635 _fdrop(struct file *fp, struct thread *td) 2636 { 2637 int error; 2638 2639 error = 0; 2640 if (fp->f_count != 0) 2641 panic("fdrop: count %d", fp->f_count); 2642 if (fp->f_ops != &badfileops) 2643 error = fo_close(fp, td); 2644 atomic_subtract_int(&openfiles, 1); 2645 crfree(fp->f_cred); 2646 free(fp->f_advice, M_FADVISE); 2647 uma_zfree(file_zone, fp); 2648 2649 return (error); 2650 } 2651 2652 /* 2653 * Apply an advisory lock on a file descriptor. 2654 * 2655 * Just attempt to get a record lock of the requested type on the entire file 2656 * (l_whence = SEEK_SET, l_start = 0, l_len = 0). 2657 */ 2658 #ifndef _SYS_SYSPROTO_H_ 2659 struct flock_args { 2660 int fd; 2661 int how; 2662 }; 2663 #endif 2664 /* ARGSUSED */ 2665 int 2666 sys_flock(struct thread *td, struct flock_args *uap) 2667 { 2668 struct file *fp; 2669 struct vnode *vp; 2670 struct flock lf; 2671 cap_rights_t rights; 2672 int error; 2673 2674 error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp); 2675 if (error != 0) 2676 return (error); 2677 if (fp->f_type != DTYPE_VNODE) { 2678 fdrop(fp, td); 2679 return (EOPNOTSUPP); 2680 } 2681 2682 vp = fp->f_vnode; 2683 lf.l_whence = SEEK_SET; 2684 lf.l_start = 0; 2685 lf.l_len = 0; 2686 if (uap->how & LOCK_UN) { 2687 lf.l_type = F_UNLCK; 2688 atomic_clear_int(&fp->f_flag, FHASLOCK); 2689 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); 2690 goto done2; 2691 } 2692 if (uap->how & LOCK_EX) 2693 lf.l_type = F_WRLCK; 2694 else if (uap->how & LOCK_SH) 2695 lf.l_type = F_RDLCK; 2696 else { 2697 error = EBADF; 2698 goto done2; 2699 } 2700 atomic_set_int(&fp->f_flag, FHASLOCK); 2701 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 2702 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); 2703 done2: 2704 fdrop(fp, td); 2705 return (error); 2706 } 2707 /* 2708 * Duplicate the specified descriptor to a free descriptor. 2709 */ 2710 int 2711 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, 2712 int openerror, int *indxp) 2713 { 2714 struct file *fp; 2715 int error, indx; 2716 2717 KASSERT(openerror == ENODEV || openerror == ENXIO, 2718 ("unexpected error %d in %s", openerror, __func__)); 2719 2720 /* 2721 * If the to-be-dup'd fd number is greater than the allowed number 2722 * of file descriptors, or the fd to be dup'd has already been 2723 * closed, then reject. 2724 */ 2725 FILEDESC_XLOCK(fdp); 2726 if ((fp = fget_locked(fdp, dfd)) == NULL) { 2727 FILEDESC_XUNLOCK(fdp); 2728 return (EBADF); 2729 } 2730 2731 error = fdalloc(td, 0, &indx); 2732 if (error != 0) { 2733 FILEDESC_XUNLOCK(fdp); 2734 return (error); 2735 } 2736 2737 /* 2738 * There are two cases of interest here. 2739 * 2740 * For ENODEV simply dup (dfd) to file descriptor (indx) and return. 2741 * 2742 * For ENXIO steal away the file structure from (dfd) and store it in 2743 * (indx). (dfd) is effectively closed by this operation. 2744 */ 2745 switch (openerror) { 2746 case ENODEV: 2747 /* 2748 * Check that the mode the file is being opened for is a 2749 * subset of the mode of the existing descriptor. 2750 */ 2751 if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { 2752 fdunused(fdp, indx); 2753 FILEDESC_XUNLOCK(fdp); 2754 return (EACCES); 2755 } 2756 fhold(fp); 2757 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2758 filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps, 2759 &fdp->fd_ofiles[indx].fde_caps); 2760 break; 2761 case ENXIO: 2762 /* 2763 * Steal away the file pointer from dfd and stuff it into indx. 2764 */ 2765 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd]; 2766 bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd])); 2767 fdunused(fdp, dfd); 2768 break; 2769 } 2770 FILEDESC_XUNLOCK(fdp); 2771 *indxp = indx; 2772 return (0); 2773 } 2774 2775 /* 2776 * Scan all active processes and prisons to see if any of them have a current 2777 * or root directory of `olddp'. If so, replace them with the new mount point. 2778 */ 2779 void 2780 mountcheckdirs(struct vnode *olddp, struct vnode *newdp) 2781 { 2782 struct filedesc *fdp; 2783 struct prison *pr; 2784 struct proc *p; 2785 int nrele; 2786 2787 if (vrefcnt(olddp) == 1) 2788 return; 2789 nrele = 0; 2790 sx_slock(&allproc_lock); 2791 FOREACH_PROC_IN_SYSTEM(p) { 2792 fdp = fdhold(p); 2793 if (fdp == NULL) 2794 continue; 2795 FILEDESC_XLOCK(fdp); 2796 if (fdp->fd_cdir == olddp) { 2797 vref(newdp); 2798 fdp->fd_cdir = newdp; 2799 nrele++; 2800 } 2801 if (fdp->fd_rdir == olddp) { 2802 vref(newdp); 2803 fdp->fd_rdir = newdp; 2804 nrele++; 2805 } 2806 if (fdp->fd_jdir == olddp) { 2807 vref(newdp); 2808 fdp->fd_jdir = newdp; 2809 nrele++; 2810 } 2811 FILEDESC_XUNLOCK(fdp); 2812 fddrop(fdp); 2813 } 2814 sx_sunlock(&allproc_lock); 2815 if (rootvnode == olddp) { 2816 vref(newdp); 2817 rootvnode = newdp; 2818 nrele++; 2819 } 2820 mtx_lock(&prison0.pr_mtx); 2821 if (prison0.pr_root == olddp) { 2822 vref(newdp); 2823 prison0.pr_root = newdp; 2824 nrele++; 2825 } 2826 mtx_unlock(&prison0.pr_mtx); 2827 sx_slock(&allprison_lock); 2828 TAILQ_FOREACH(pr, &allprison, pr_list) { 2829 mtx_lock(&pr->pr_mtx); 2830 if (pr->pr_root == olddp) { 2831 vref(newdp); 2832 pr->pr_root = newdp; 2833 nrele++; 2834 } 2835 mtx_unlock(&pr->pr_mtx); 2836 } 2837 sx_sunlock(&allprison_lock); 2838 while (nrele--) 2839 vrele(olddp); 2840 } 2841 2842 struct filedesc_to_leader * 2843 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) 2844 { 2845 struct filedesc_to_leader *fdtol; 2846 2847 fdtol = malloc(sizeof(struct filedesc_to_leader), 2848 M_FILEDESC_TO_LEADER, 2849 M_WAITOK); 2850 fdtol->fdl_refcount = 1; 2851 fdtol->fdl_holdcount = 0; 2852 fdtol->fdl_wakeup = 0; 2853 fdtol->fdl_leader = leader; 2854 if (old != NULL) { 2855 FILEDESC_XLOCK(fdp); 2856 fdtol->fdl_next = old->fdl_next; 2857 fdtol->fdl_prev = old; 2858 old->fdl_next = fdtol; 2859 fdtol->fdl_next->fdl_prev = fdtol; 2860 FILEDESC_XUNLOCK(fdp); 2861 } else { 2862 fdtol->fdl_next = fdtol; 2863 fdtol->fdl_prev = fdtol; 2864 } 2865 return (fdtol); 2866 } 2867 2868 /* 2869 * Get file structures globally. 2870 */ 2871 static int 2872 sysctl_kern_file(SYSCTL_HANDLER_ARGS) 2873 { 2874 struct xfile xf; 2875 struct filedesc *fdp; 2876 struct file *fp; 2877 struct proc *p; 2878 int error, n; 2879 2880 error = sysctl_wire_old_buffer(req, 0); 2881 if (error != 0) 2882 return (error); 2883 if (req->oldptr == NULL) { 2884 n = 0; 2885 sx_slock(&allproc_lock); 2886 FOREACH_PROC_IN_SYSTEM(p) { 2887 if (p->p_state == PRS_NEW) 2888 continue; 2889 fdp = fdhold(p); 2890 if (fdp == NULL) 2891 continue; 2892 /* overestimates sparse tables. */ 2893 if (fdp->fd_lastfile > 0) 2894 n += fdp->fd_lastfile; 2895 fddrop(fdp); 2896 } 2897 sx_sunlock(&allproc_lock); 2898 return (SYSCTL_OUT(req, 0, n * sizeof(xf))); 2899 } 2900 error = 0; 2901 bzero(&xf, sizeof(xf)); 2902 xf.xf_size = sizeof(xf); 2903 sx_slock(&allproc_lock); 2904 FOREACH_PROC_IN_SYSTEM(p) { 2905 PROC_LOCK(p); 2906 if (p->p_state == PRS_NEW) { 2907 PROC_UNLOCK(p); 2908 continue; 2909 } 2910 if (p_cansee(req->td, p) != 0) { 2911 PROC_UNLOCK(p); 2912 continue; 2913 } 2914 xf.xf_pid = p->p_pid; 2915 xf.xf_uid = p->p_ucred->cr_uid; 2916 PROC_UNLOCK(p); 2917 fdp = fdhold(p); 2918 if (fdp == NULL) 2919 continue; 2920 FILEDESC_SLOCK(fdp); 2921 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) { 2922 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 2923 continue; 2924 xf.xf_fd = n; 2925 xf.xf_file = fp; 2926 xf.xf_data = fp->f_data; 2927 xf.xf_vnode = fp->f_vnode; 2928 xf.xf_type = fp->f_type; 2929 xf.xf_count = fp->f_count; 2930 xf.xf_msgcount = 0; 2931 xf.xf_offset = foffset_get(fp); 2932 xf.xf_flag = fp->f_flag; 2933 error = SYSCTL_OUT(req, &xf, sizeof(xf)); 2934 if (error) 2935 break; 2936 } 2937 FILEDESC_SUNLOCK(fdp); 2938 fddrop(fdp); 2939 if (error) 2940 break; 2941 } 2942 sx_sunlock(&allproc_lock); 2943 return (error); 2944 } 2945 2946 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 2947 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); 2948 2949 #ifdef KINFO_OFILE_SIZE 2950 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); 2951 #endif 2952 2953 #ifdef COMPAT_FREEBSD7 2954 static int 2955 export_vnode_for_osysctl(struct vnode *vp, int type, 2956 struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req) 2957 { 2958 int error; 2959 char *fullpath, *freepath; 2960 2961 bzero(kif, sizeof(*kif)); 2962 kif->kf_structsize = sizeof(*kif); 2963 2964 vref(vp); 2965 kif->kf_fd = type; 2966 kif->kf_type = KF_TYPE_VNODE; 2967 /* This function only handles directories. */ 2968 if (vp->v_type != VDIR) { 2969 vrele(vp); 2970 return (ENOTDIR); 2971 } 2972 kif->kf_vnode_type = KF_VTYPE_VDIR; 2973 2974 /* 2975 * This is not a true file descriptor, so we set a bogus refcount 2976 * and offset to indicate these fields should be ignored. 2977 */ 2978 kif->kf_ref_count = -1; 2979 kif->kf_offset = -1; 2980 2981 freepath = NULL; 2982 fullpath = "-"; 2983 FILEDESC_SUNLOCK(fdp); 2984 vn_fullpath(curthread, vp, &fullpath, &freepath); 2985 vrele(vp); 2986 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 2987 if (freepath != NULL) 2988 free(freepath, M_TEMP); 2989 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 2990 FILEDESC_SLOCK(fdp); 2991 return (error); 2992 } 2993 2994 /* 2995 * Get per-process file descriptors for use by procstat(1), et al. 2996 */ 2997 static int 2998 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) 2999 { 3000 char *fullpath, *freepath; 3001 struct kinfo_ofile *kif; 3002 struct filedesc *fdp; 3003 int error, i, *name; 3004 struct shmfd *shmfd; 3005 struct socket *so; 3006 struct vnode *vp; 3007 struct ksem *ks; 3008 struct file *fp; 3009 struct proc *p; 3010 struct tty *tp; 3011 3012 name = (int *)arg1; 3013 error = pget((pid_t)name[0], PGET_CANDEBUG, &p); 3014 if (error != 0) 3015 return (error); 3016 fdp = fdhold(p); 3017 PROC_UNLOCK(p); 3018 if (fdp == NULL) 3019 return (ENOENT); 3020 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); 3021 FILEDESC_SLOCK(fdp); 3022 if (fdp->fd_cdir != NULL) 3023 export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif, 3024 fdp, req); 3025 if (fdp->fd_rdir != NULL) 3026 export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif, 3027 fdp, req); 3028 if (fdp->fd_jdir != NULL) 3029 export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif, 3030 fdp, req); 3031 for (i = 0; fdp->fd_refcnt > 0 && i < fdp->fd_nfiles; i++) { 3032 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3033 continue; 3034 bzero(kif, sizeof(*kif)); 3035 kif->kf_structsize = sizeof(*kif); 3036 ks = NULL; 3037 vp = NULL; 3038 so = NULL; 3039 tp = NULL; 3040 shmfd = NULL; 3041 kif->kf_fd = i; 3042 3043 switch (fp->f_type) { 3044 case DTYPE_VNODE: 3045 kif->kf_type = KF_TYPE_VNODE; 3046 vp = fp->f_vnode; 3047 break; 3048 3049 case DTYPE_SOCKET: 3050 kif->kf_type = KF_TYPE_SOCKET; 3051 so = fp->f_data; 3052 break; 3053 3054 case DTYPE_PIPE: 3055 kif->kf_type = KF_TYPE_PIPE; 3056 break; 3057 3058 case DTYPE_FIFO: 3059 kif->kf_type = KF_TYPE_FIFO; 3060 vp = fp->f_vnode; 3061 break; 3062 3063 case DTYPE_KQUEUE: 3064 kif->kf_type = KF_TYPE_KQUEUE; 3065 break; 3066 3067 case DTYPE_CRYPTO: 3068 kif->kf_type = KF_TYPE_CRYPTO; 3069 break; 3070 3071 case DTYPE_MQUEUE: 3072 kif->kf_type = KF_TYPE_MQUEUE; 3073 break; 3074 3075 case DTYPE_SHM: 3076 kif->kf_type = KF_TYPE_SHM; 3077 shmfd = fp->f_data; 3078 break; 3079 3080 case DTYPE_SEM: 3081 kif->kf_type = KF_TYPE_SEM; 3082 ks = fp->f_data; 3083 break; 3084 3085 case DTYPE_PTS: 3086 kif->kf_type = KF_TYPE_PTS; 3087 tp = fp->f_data; 3088 break; 3089 3090 case DTYPE_PROCDESC: 3091 kif->kf_type = KF_TYPE_PROCDESC; 3092 break; 3093 3094 default: 3095 kif->kf_type = KF_TYPE_UNKNOWN; 3096 break; 3097 } 3098 kif->kf_ref_count = fp->f_count; 3099 if (fp->f_flag & FREAD) 3100 kif->kf_flags |= KF_FLAG_READ; 3101 if (fp->f_flag & FWRITE) 3102 kif->kf_flags |= KF_FLAG_WRITE; 3103 if (fp->f_flag & FAPPEND) 3104 kif->kf_flags |= KF_FLAG_APPEND; 3105 if (fp->f_flag & FASYNC) 3106 kif->kf_flags |= KF_FLAG_ASYNC; 3107 if (fp->f_flag & FFSYNC) 3108 kif->kf_flags |= KF_FLAG_FSYNC; 3109 if (fp->f_flag & FNONBLOCK) 3110 kif->kf_flags |= KF_FLAG_NONBLOCK; 3111 if (fp->f_flag & O_DIRECT) 3112 kif->kf_flags |= KF_FLAG_DIRECT; 3113 if (fp->f_flag & FHASLOCK) 3114 kif->kf_flags |= KF_FLAG_HASLOCK; 3115 kif->kf_offset = foffset_get(fp); 3116 if (vp != NULL) { 3117 vref(vp); 3118 switch (vp->v_type) { 3119 case VNON: 3120 kif->kf_vnode_type = KF_VTYPE_VNON; 3121 break; 3122 case VREG: 3123 kif->kf_vnode_type = KF_VTYPE_VREG; 3124 break; 3125 case VDIR: 3126 kif->kf_vnode_type = KF_VTYPE_VDIR; 3127 break; 3128 case VBLK: 3129 kif->kf_vnode_type = KF_VTYPE_VBLK; 3130 break; 3131 case VCHR: 3132 kif->kf_vnode_type = KF_VTYPE_VCHR; 3133 break; 3134 case VLNK: 3135 kif->kf_vnode_type = KF_VTYPE_VLNK; 3136 break; 3137 case VSOCK: 3138 kif->kf_vnode_type = KF_VTYPE_VSOCK; 3139 break; 3140 case VFIFO: 3141 kif->kf_vnode_type = KF_VTYPE_VFIFO; 3142 break; 3143 case VBAD: 3144 kif->kf_vnode_type = KF_VTYPE_VBAD; 3145 break; 3146 default: 3147 kif->kf_vnode_type = KF_VTYPE_UNKNOWN; 3148 break; 3149 } 3150 /* 3151 * It is OK to drop the filedesc lock here as we will 3152 * re-validate and re-evaluate its properties when 3153 * the loop continues. 3154 */ 3155 freepath = NULL; 3156 fullpath = "-"; 3157 FILEDESC_SUNLOCK(fdp); 3158 vn_fullpath(curthread, vp, &fullpath, &freepath); 3159 vrele(vp); 3160 strlcpy(kif->kf_path, fullpath, 3161 sizeof(kif->kf_path)); 3162 if (freepath != NULL) 3163 free(freepath, M_TEMP); 3164 FILEDESC_SLOCK(fdp); 3165 } 3166 if (so != NULL) { 3167 struct sockaddr *sa; 3168 3169 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa) 3170 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3171 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3172 free(sa, M_SONAME); 3173 } 3174 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa) 3175 == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3176 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3177 free(sa, M_SONAME); 3178 } 3179 kif->kf_sock_domain = 3180 so->so_proto->pr_domain->dom_family; 3181 kif->kf_sock_type = so->so_type; 3182 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3183 } 3184 if (tp != NULL) { 3185 strlcpy(kif->kf_path, tty_devname(tp), 3186 sizeof(kif->kf_path)); 3187 } 3188 if (shmfd != NULL) 3189 shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path)); 3190 if (ks != NULL && ksem_info != NULL) 3191 ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL); 3192 error = SYSCTL_OUT(req, kif, sizeof(*kif)); 3193 if (error) 3194 break; 3195 } 3196 FILEDESC_SUNLOCK(fdp); 3197 fddrop(fdp); 3198 free(kif, M_TEMP); 3199 return (0); 3200 } 3201 3202 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, 3203 CTLFLAG_RD||CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, 3204 "Process ofiledesc entries"); 3205 #endif /* COMPAT_FREEBSD7 */ 3206 3207 #ifdef KINFO_FILE_SIZE 3208 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); 3209 #endif 3210 3211 struct export_fd_buf { 3212 struct filedesc *fdp; 3213 struct sbuf *sb; 3214 ssize_t remainder; 3215 struct kinfo_file kif; 3216 }; 3217 3218 static int 3219 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt, 3220 int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf) 3221 { 3222 struct { 3223 int fflag; 3224 int kf_fflag; 3225 } fflags_table[] = { 3226 { FAPPEND, KF_FLAG_APPEND }, 3227 { FASYNC, KF_FLAG_ASYNC }, 3228 { FFSYNC, KF_FLAG_FSYNC }, 3229 { FHASLOCK, KF_FLAG_HASLOCK }, 3230 { FNONBLOCK, KF_FLAG_NONBLOCK }, 3231 { FREAD, KF_FLAG_READ }, 3232 { FWRITE, KF_FLAG_WRITE }, 3233 { O_CREAT, KF_FLAG_CREAT }, 3234 { O_DIRECT, KF_FLAG_DIRECT }, 3235 { O_EXCL, KF_FLAG_EXCL }, 3236 { O_EXEC, KF_FLAG_EXEC }, 3237 { O_EXLOCK, KF_FLAG_EXLOCK }, 3238 { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, 3239 { O_SHLOCK, KF_FLAG_SHLOCK }, 3240 { O_TRUNC, KF_FLAG_TRUNC } 3241 }; 3242 #define NFFLAGS (sizeof(fflags_table) / sizeof(*fflags_table)) 3243 struct kinfo_file *kif; 3244 struct vnode *vp; 3245 int error, locked; 3246 unsigned int i; 3247 3248 if (efbuf->remainder == 0) 3249 return (0); 3250 kif = &efbuf->kif; 3251 bzero(kif, sizeof(*kif)); 3252 locked = efbuf->fdp != NULL; 3253 switch (type) { 3254 case KF_TYPE_FIFO: 3255 case KF_TYPE_VNODE: 3256 if (locked) { 3257 FILEDESC_SUNLOCK(efbuf->fdp); 3258 locked = 0; 3259 } 3260 vp = (struct vnode *)data; 3261 error = fill_vnode_info(vp, kif); 3262 vrele(vp); 3263 break; 3264 case KF_TYPE_SOCKET: 3265 error = fill_socket_info((struct socket *)data, kif); 3266 break; 3267 case KF_TYPE_PIPE: 3268 error = fill_pipe_info((struct pipe *)data, kif); 3269 break; 3270 case KF_TYPE_PTS: 3271 error = fill_pts_info((struct tty *)data, kif); 3272 break; 3273 case KF_TYPE_PROCDESC: 3274 error = fill_procdesc_info((struct procdesc *)data, kif); 3275 break; 3276 case KF_TYPE_SEM: 3277 error = fill_sem_info((struct file *)data, kif); 3278 break; 3279 case KF_TYPE_SHM: 3280 error = fill_shm_info((struct file *)data, kif); 3281 break; 3282 default: 3283 error = 0; 3284 } 3285 if (error == 0) 3286 kif->kf_status |= KF_ATTR_VALID; 3287 3288 /* 3289 * Translate file access flags. 3290 */ 3291 for (i = 0; i < NFFLAGS; i++) 3292 if (fflags & fflags_table[i].fflag) 3293 kif->kf_flags |= fflags_table[i].kf_fflag; 3294 if (rightsp != NULL) 3295 kif->kf_cap_rights = *rightsp; 3296 else 3297 cap_rights_init(&kif->kf_cap_rights); 3298 kif->kf_fd = fd; 3299 kif->kf_type = type; 3300 kif->kf_ref_count = refcnt; 3301 kif->kf_offset = offset; 3302 /* Pack record size down */ 3303 kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + 3304 strlen(kif->kf_path) + 1; 3305 kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); 3306 if (efbuf->remainder != -1) { 3307 if (efbuf->remainder < kif->kf_structsize) { 3308 /* Terminate export. */ 3309 efbuf->remainder = 0; 3310 if (efbuf->fdp != NULL && !locked) 3311 FILEDESC_SLOCK(efbuf->fdp); 3312 return (0); 3313 } 3314 efbuf->remainder -= kif->kf_structsize; 3315 } 3316 if (locked) 3317 FILEDESC_SUNLOCK(efbuf->fdp); 3318 error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize); 3319 if (efbuf->fdp != NULL) 3320 FILEDESC_SLOCK(efbuf->fdp); 3321 return (error); 3322 } 3323 3324 /* 3325 * Store a process file descriptor information to sbuf. 3326 * 3327 * Takes a locked proc as argument, and returns with the proc unlocked. 3328 */ 3329 int 3330 kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) 3331 { 3332 struct file *fp; 3333 struct filedesc *fdp; 3334 struct export_fd_buf *efbuf; 3335 struct vnode *cttyvp, *textvp, *tracevp; 3336 int64_t offset; 3337 void *data; 3338 int error, i; 3339 int type, refcnt, fflags; 3340 cap_rights_t rights; 3341 3342 PROC_LOCK_ASSERT(p, MA_OWNED); 3343 3344 /* ktrace vnode */ 3345 tracevp = p->p_tracevp; 3346 if (tracevp != NULL) 3347 vref(tracevp); 3348 /* text vnode */ 3349 textvp = p->p_textvp; 3350 if (textvp != NULL) 3351 vref(textvp); 3352 /* Controlling tty. */ 3353 cttyvp = NULL; 3354 if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { 3355 cttyvp = p->p_pgrp->pg_session->s_ttyvp; 3356 if (cttyvp != NULL) 3357 vref(cttyvp); 3358 } 3359 fdp = fdhold(p); 3360 PROC_UNLOCK(p); 3361 efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); 3362 efbuf->fdp = NULL; 3363 efbuf->sb = sb; 3364 efbuf->remainder = maxlen; 3365 if (tracevp != NULL) 3366 export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE, 3367 FREAD | FWRITE, -1, -1, NULL, efbuf); 3368 if (textvp != NULL) 3369 export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT, 3370 FREAD, -1, -1, NULL, efbuf); 3371 if (cttyvp != NULL) 3372 export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY, 3373 FREAD | FWRITE, -1, -1, NULL, efbuf); 3374 error = 0; 3375 if (fdp == NULL) 3376 goto fail; 3377 efbuf->fdp = fdp; 3378 FILEDESC_SLOCK(fdp); 3379 /* working directory */ 3380 if (fdp->fd_cdir != NULL) { 3381 vref(fdp->fd_cdir); 3382 data = fdp->fd_cdir; 3383 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD, 3384 FREAD, -1, -1, NULL, efbuf); 3385 } 3386 /* root directory */ 3387 if (fdp->fd_rdir != NULL) { 3388 vref(fdp->fd_rdir); 3389 data = fdp->fd_rdir; 3390 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT, 3391 FREAD, -1, -1, NULL, efbuf); 3392 } 3393 /* jail directory */ 3394 if (fdp->fd_jdir != NULL) { 3395 vref(fdp->fd_jdir); 3396 data = fdp->fd_jdir; 3397 export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL, 3398 FREAD, -1, -1, NULL, efbuf); 3399 } 3400 for (i = 0; fdp->fd_refcnt > 0 && i < fdp->fd_nfiles; i++) { 3401 if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) 3402 continue; 3403 data = NULL; 3404 #ifdef CAPABILITIES 3405 rights = *cap_rights(fdp, i); 3406 #else /* !CAPABILITIES */ 3407 cap_rights_init(&rights); 3408 #endif 3409 switch (fp->f_type) { 3410 case DTYPE_VNODE: 3411 type = KF_TYPE_VNODE; 3412 vref(fp->f_vnode); 3413 data = fp->f_vnode; 3414 break; 3415 3416 case DTYPE_SOCKET: 3417 type = KF_TYPE_SOCKET; 3418 data = fp->f_data; 3419 break; 3420 3421 case DTYPE_PIPE: 3422 type = KF_TYPE_PIPE; 3423 data = fp->f_data; 3424 break; 3425 3426 case DTYPE_FIFO: 3427 type = KF_TYPE_FIFO; 3428 vref(fp->f_vnode); 3429 data = fp->f_vnode; 3430 break; 3431 3432 case DTYPE_KQUEUE: 3433 type = KF_TYPE_KQUEUE; 3434 break; 3435 3436 case DTYPE_CRYPTO: 3437 type = KF_TYPE_CRYPTO; 3438 break; 3439 3440 case DTYPE_MQUEUE: 3441 type = KF_TYPE_MQUEUE; 3442 break; 3443 3444 case DTYPE_SHM: 3445 type = KF_TYPE_SHM; 3446 data = fp; 3447 break; 3448 3449 case DTYPE_SEM: 3450 type = KF_TYPE_SEM; 3451 data = fp; 3452 break; 3453 3454 case DTYPE_PTS: 3455 type = KF_TYPE_PTS; 3456 data = fp->f_data; 3457 break; 3458 3459 case DTYPE_PROCDESC: 3460 type = KF_TYPE_PROCDESC; 3461 data = fp->f_data; 3462 break; 3463 3464 default: 3465 type = KF_TYPE_UNKNOWN; 3466 break; 3467 } 3468 refcnt = fp->f_count; 3469 fflags = fp->f_flag; 3470 offset = foffset_get(fp); 3471 3472 /* 3473 * Create sysctl entry. 3474 * It is OK to drop the filedesc lock here as we will 3475 * re-validate and re-evaluate its properties when 3476 * the loop continues. 3477 */ 3478 error = export_fd_to_sb(data, type, i, fflags, refcnt, 3479 offset, &rights, efbuf); 3480 if (error != 0) 3481 break; 3482 } 3483 FILEDESC_SUNLOCK(fdp); 3484 fddrop(fdp); 3485 fail: 3486 free(efbuf, M_TEMP); 3487 return (error); 3488 } 3489 3490 #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) 3491 3492 /* 3493 * Get per-process file descriptors for use by procstat(1), et al. 3494 */ 3495 static int 3496 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) 3497 { 3498 struct sbuf sb; 3499 struct proc *p; 3500 ssize_t maxlen; 3501 int error, error2, *name; 3502 3503 name = (int *)arg1; 3504 3505 sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); 3506 error = pget((pid_t)name[0], PGET_CANDEBUG, &p); 3507 if (error != 0) { 3508 sbuf_delete(&sb); 3509 return (error); 3510 } 3511 maxlen = req->oldptr != NULL ? req->oldlen : -1; 3512 error = kern_proc_filedesc_out(p, &sb, maxlen); 3513 error2 = sbuf_finish(&sb); 3514 sbuf_delete(&sb); 3515 return (error != 0 ? error : error2); 3516 } 3517 3518 int 3519 vntype_to_kinfo(int vtype) 3520 { 3521 struct { 3522 int vtype; 3523 int kf_vtype; 3524 } vtypes_table[] = { 3525 { VBAD, KF_VTYPE_VBAD }, 3526 { VBLK, KF_VTYPE_VBLK }, 3527 { VCHR, KF_VTYPE_VCHR }, 3528 { VDIR, KF_VTYPE_VDIR }, 3529 { VFIFO, KF_VTYPE_VFIFO }, 3530 { VLNK, KF_VTYPE_VLNK }, 3531 { VNON, KF_VTYPE_VNON }, 3532 { VREG, KF_VTYPE_VREG }, 3533 { VSOCK, KF_VTYPE_VSOCK } 3534 }; 3535 #define NVTYPES (sizeof(vtypes_table) / sizeof(*vtypes_table)) 3536 unsigned int i; 3537 3538 /* 3539 * Perform vtype translation. 3540 */ 3541 for (i = 0; i < NVTYPES; i++) 3542 if (vtypes_table[i].vtype == vtype) 3543 break; 3544 if (i < NVTYPES) 3545 return (vtypes_table[i].kf_vtype); 3546 3547 return (KF_VTYPE_UNKNOWN); 3548 } 3549 3550 static int 3551 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif) 3552 { 3553 struct vattr va; 3554 char *fullpath, *freepath; 3555 int error; 3556 3557 if (vp == NULL) 3558 return (1); 3559 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); 3560 freepath = NULL; 3561 fullpath = "-"; 3562 error = vn_fullpath(curthread, vp, &fullpath, &freepath); 3563 if (error == 0) { 3564 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); 3565 } 3566 if (freepath != NULL) 3567 free(freepath, M_TEMP); 3568 3569 /* 3570 * Retrieve vnode attributes. 3571 */ 3572 va.va_fsid = VNOVAL; 3573 va.va_rdev = NODEV; 3574 vn_lock(vp, LK_SHARED | LK_RETRY); 3575 error = VOP_GETATTR(vp, &va, curthread->td_ucred); 3576 VOP_UNLOCK(vp, 0); 3577 if (error != 0) 3578 return (error); 3579 if (va.va_fsid != VNOVAL) 3580 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; 3581 else 3582 kif->kf_un.kf_file.kf_file_fsid = 3583 vp->v_mount->mnt_stat.f_fsid.val[0]; 3584 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; 3585 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); 3586 kif->kf_un.kf_file.kf_file_size = va.va_size; 3587 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; 3588 return (0); 3589 } 3590 3591 static int 3592 fill_socket_info(struct socket *so, struct kinfo_file *kif) 3593 { 3594 struct sockaddr *sa; 3595 struct inpcb *inpcb; 3596 struct unpcb *unpcb; 3597 int error; 3598 3599 if (so == NULL) 3600 return (1); 3601 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 3602 kif->kf_sock_type = so->so_type; 3603 kif->kf_sock_protocol = so->so_proto->pr_protocol; 3604 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 3605 switch(kif->kf_sock_domain) { 3606 case AF_INET: 3607 case AF_INET6: 3608 if (kif->kf_sock_protocol == IPPROTO_TCP) { 3609 if (so->so_pcb != NULL) { 3610 inpcb = (struct inpcb *)(so->so_pcb); 3611 kif->kf_un.kf_sock.kf_sock_inpcb = 3612 (uintptr_t)inpcb->inp_ppcb; 3613 } 3614 } 3615 break; 3616 case AF_UNIX: 3617 if (so->so_pcb != NULL) { 3618 unpcb = (struct unpcb *)(so->so_pcb); 3619 if (unpcb->unp_conn) { 3620 kif->kf_un.kf_sock.kf_sock_unpconn = 3621 (uintptr_t)unpcb->unp_conn; 3622 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 3623 so->so_rcv.sb_state; 3624 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 3625 so->so_snd.sb_state; 3626 } 3627 } 3628 break; 3629 } 3630 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 3631 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 3632 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 3633 free(sa, M_SONAME); 3634 } 3635 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 3636 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 3637 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 3638 free(sa, M_SONAME); 3639 } 3640 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 3641 sizeof(kif->kf_path)); 3642 return (0); 3643 } 3644 3645 static int 3646 fill_pts_info(struct tty *tp, struct kinfo_file *kif) 3647 { 3648 3649 if (tp == NULL) 3650 return (1); 3651 kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp); 3652 strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path)); 3653 return (0); 3654 } 3655 3656 static int 3657 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif) 3658 { 3659 3660 if (pi == NULL) 3661 return (1); 3662 kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi; 3663 kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer; 3664 kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt; 3665 return (0); 3666 } 3667 3668 static int 3669 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif) 3670 { 3671 3672 if (pdp == NULL) 3673 return (1); 3674 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 3675 return (0); 3676 } 3677 3678 static int 3679 fill_sem_info(struct file *fp, struct kinfo_file *kif) 3680 { 3681 struct thread *td; 3682 struct stat sb; 3683 3684 td = curthread; 3685 if (fp->f_data == NULL) 3686 return (1); 3687 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3688 return (1); 3689 if (ksem_info == NULL) 3690 return (1); 3691 ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path), 3692 &kif->kf_un.kf_sem.kf_sem_value); 3693 kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode; 3694 return (0); 3695 } 3696 3697 static int 3698 fill_shm_info(struct file *fp, struct kinfo_file *kif) 3699 { 3700 struct thread *td; 3701 struct stat sb; 3702 3703 td = curthread; 3704 if (fp->f_data == NULL) 3705 return (1); 3706 if (fo_stat(fp, &sb, td->td_ucred, td) != 0) 3707 return (1); 3708 shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path)); 3709 kif->kf_un.kf_file.kf_file_mode = sb.st_mode; 3710 kif->kf_un.kf_file.kf_file_size = sb.st_size; 3711 return (0); 3712 } 3713 3714 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, 3715 CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, 3716 "Process filedesc entries"); 3717 3718 #ifdef DDB 3719 /* 3720 * For the purposes of debugging, generate a human-readable string for the 3721 * file type. 3722 */ 3723 static const char * 3724 file_type_to_name(short type) 3725 { 3726 3727 switch (type) { 3728 case 0: 3729 return ("zero"); 3730 case DTYPE_VNODE: 3731 return ("vnod"); 3732 case DTYPE_SOCKET: 3733 return ("sock"); 3734 case DTYPE_PIPE: 3735 return ("pipe"); 3736 case DTYPE_FIFO: 3737 return ("fifo"); 3738 case DTYPE_KQUEUE: 3739 return ("kque"); 3740 case DTYPE_CRYPTO: 3741 return ("crpt"); 3742 case DTYPE_MQUEUE: 3743 return ("mque"); 3744 case DTYPE_SHM: 3745 return ("shm"); 3746 case DTYPE_SEM: 3747 return ("ksem"); 3748 default: 3749 return ("unkn"); 3750 } 3751 } 3752 3753 /* 3754 * For the purposes of debugging, identify a process (if any, perhaps one of 3755 * many) that references the passed file in its file descriptor array. Return 3756 * NULL if none. 3757 */ 3758 static struct proc * 3759 file_to_first_proc(struct file *fp) 3760 { 3761 struct filedesc *fdp; 3762 struct proc *p; 3763 int n; 3764 3765 FOREACH_PROC_IN_SYSTEM(p) { 3766 if (p->p_state == PRS_NEW) 3767 continue; 3768 fdp = p->p_fd; 3769 if (fdp == NULL) 3770 continue; 3771 for (n = 0; n < fdp->fd_nfiles; n++) { 3772 if (fp == fdp->fd_ofiles[n].fde_file) 3773 return (p); 3774 } 3775 } 3776 return (NULL); 3777 } 3778 3779 static void 3780 db_print_file(struct file *fp, int header) 3781 { 3782 struct proc *p; 3783 3784 if (header) 3785 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", 3786 "File", "Type", "Data", "Flag", "GCFl", "Count", 3787 "MCount", "Vnode", "FPID", "FCmd"); 3788 p = file_to_first_proc(fp); 3789 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp, 3790 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag, 3791 0, fp->f_count, 0, fp->f_vnode, 3792 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); 3793 } 3794 3795 DB_SHOW_COMMAND(file, db_show_file) 3796 { 3797 struct file *fp; 3798 3799 if (!have_addr) { 3800 db_printf("usage: show file <addr>\n"); 3801 return; 3802 } 3803 fp = (struct file *)addr; 3804 db_print_file(fp, 1); 3805 } 3806 3807 DB_SHOW_COMMAND(files, db_show_files) 3808 { 3809 struct filedesc *fdp; 3810 struct file *fp; 3811 struct proc *p; 3812 int header; 3813 int n; 3814 3815 header = 1; 3816 FOREACH_PROC_IN_SYSTEM(p) { 3817 if (p->p_state == PRS_NEW) 3818 continue; 3819 if ((fdp = p->p_fd) == NULL) 3820 continue; 3821 for (n = 0; n < fdp->fd_nfiles; ++n) { 3822 if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) 3823 continue; 3824 db_print_file(fp, header); 3825 header = 0; 3826 } 3827 } 3828 } 3829 #endif 3830 3831 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, 3832 &maxfilesperproc, 0, "Maximum files allowed open per process"); 3833 3834 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, 3835 &maxfiles, 0, "Maximum number of files"); 3836 3837 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, 3838 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files"); 3839 3840 /* ARGSUSED*/ 3841 static void 3842 filelistinit(void *dummy) 3843 { 3844 3845 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, 3846 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 3847 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); 3848 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF); 3849 } 3850 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); 3851 3852 /*-------------------------------------------------------------------*/ 3853 3854 static int 3855 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, 3856 int flags, struct thread *td) 3857 { 3858 3859 return (EBADF); 3860 } 3861 3862 static int 3863 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, 3864 struct thread *td) 3865 { 3866 3867 return (EINVAL); 3868 } 3869 3870 static int 3871 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, 3872 struct thread *td) 3873 { 3874 3875 return (EBADF); 3876 } 3877 3878 static int 3879 badfo_poll(struct file *fp, int events, struct ucred *active_cred, 3880 struct thread *td) 3881 { 3882 3883 return (0); 3884 } 3885 3886 static int 3887 badfo_kqfilter(struct file *fp, struct knote *kn) 3888 { 3889 3890 return (EBADF); 3891 } 3892 3893 static int 3894 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 3895 struct thread *td) 3896 { 3897 3898 return (EBADF); 3899 } 3900 3901 static int 3902 badfo_close(struct file *fp, struct thread *td) 3903 { 3904 3905 return (EBADF); 3906 } 3907 3908 static int 3909 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3910 struct thread *td) 3911 { 3912 3913 return (EBADF); 3914 } 3915 3916 static int 3917 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3918 struct thread *td) 3919 { 3920 3921 return (EBADF); 3922 } 3923 3924 static int 3925 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3926 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3927 int kflags, struct sendfile_sync *sfs, struct thread *td) 3928 { 3929 3930 return (EBADF); 3931 } 3932 3933 struct fileops badfileops = { 3934 .fo_read = badfo_readwrite, 3935 .fo_write = badfo_readwrite, 3936 .fo_truncate = badfo_truncate, 3937 .fo_ioctl = badfo_ioctl, 3938 .fo_poll = badfo_poll, 3939 .fo_kqfilter = badfo_kqfilter, 3940 .fo_stat = badfo_stat, 3941 .fo_close = badfo_close, 3942 .fo_chmod = badfo_chmod, 3943 .fo_chown = badfo_chown, 3944 .fo_sendfile = badfo_sendfile, 3945 }; 3946 3947 int 3948 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, 3949 struct thread *td) 3950 { 3951 3952 return (EINVAL); 3953 } 3954 3955 int 3956 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, 3957 struct thread *td) 3958 { 3959 3960 return (EINVAL); 3961 } 3962 3963 int 3964 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, 3965 struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, 3966 int kflags, struct sendfile_sync *sfs, struct thread *td) 3967 { 3968 3969 return (EINVAL); 3970 } 3971 3972 /*-------------------------------------------------------------------*/ 3973 3974 /* 3975 * File Descriptor pseudo-device driver (/dev/fd/). 3976 * 3977 * Opening minor device N dup()s the file (if any) connected to file 3978 * descriptor N belonging to the calling process. Note that this driver 3979 * consists of only the ``open()'' routine, because all subsequent 3980 * references to this file will be direct to the other driver. 3981 * 3982 * XXX: we could give this one a cloning event handler if necessary. 3983 */ 3984 3985 /* ARGSUSED */ 3986 static int 3987 fdopen(struct cdev *dev, int mode, int type, struct thread *td) 3988 { 3989 3990 /* 3991 * XXX Kludge: set curthread->td_dupfd to contain the value of the 3992 * the file descriptor being sought for duplication. The error 3993 * return ensures that the vnode for this device will be released 3994 * by vn_open. Open will detect this special error and take the 3995 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN 3996 * will simply report the error. 3997 */ 3998 td->td_dupfd = dev2unit(dev); 3999 return (ENODEV); 4000 } 4001 4002 static struct cdevsw fildesc_cdevsw = { 4003 .d_version = D_VERSION, 4004 .d_open = fdopen, 4005 .d_name = "FD", 4006 }; 4007 4008 static void 4009 fildesc_drvinit(void *unused) 4010 { 4011 struct cdev *dev; 4012 4013 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, 4014 UID_ROOT, GID_WHEEL, 0666, "fd/0"); 4015 make_dev_alias(dev, "stdin"); 4016 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, 4017 UID_ROOT, GID_WHEEL, 0666, "fd/1"); 4018 make_dev_alias(dev, "stdout"); 4019 dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, 4020 UID_ROOT, GID_WHEEL, 0666, "fd/2"); 4021 make_dev_alias(dev, "stderr"); 4022 } 4023 4024 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); 4025