| 1 | /* |
| 2 | * Copyright (c) 2005 The DragonFly Project. All rights reserved. |
| 3 | * |
| 4 | * This code is derived from software contributed to The DragonFly Project |
| 5 | * by Jeffrey Hsu. |
| 6 | * |
| 7 | * Redistribution and use in source and binary forms, with or without |
| 8 | * modification, are permitted provided that the following conditions |
| 9 | * are met: |
| 10 | * |
| 11 | * 1. Redistributions of source code must retain the above copyright |
| 12 | * notice, this list of conditions and the following disclaimer. |
| 13 | * 2. Redistributions in binary form must reproduce the above copyright |
| 14 | * notice, this list of conditions and the following disclaimer in |
| 15 | * the documentation and/or other materials provided with the |
| 16 | * distribution. |
| 17 | * 3. Neither the name of The DragonFly Project nor the names of its |
| 18 | * contributors may be used to endorse or promote products derived |
| 19 | * from this software without specific, prior written permission. |
| 20 | * |
| 21 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 22 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 23 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 24 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 25 | * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 26 | * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 27 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 28 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
| 29 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 30 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
| 31 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 32 | * SUCH DAMAGE. |
| 33 | * |
| 34 | * |
| 35 | * Copyright (c) 1982, 1986, 1989, 1991, 1993 |
| 36 | * The Regents of the University of California. All rights reserved. |
| 37 | * (c) UNIX System Laboratories, Inc. |
| 38 | * All or some portions of this file are derived from material licensed |
| 39 | * to the University of California by American Telephone and Telegraph |
| 40 | * Co. or Unix System Laboratories, Inc. and are reproduced herein with |
| 41 | * the permission of UNIX System Laboratories, Inc. |
| 42 | * |
| 43 | * Redistribution and use in source and binary forms, with or without |
| 44 | * modification, are permitted provided that the following conditions |
| 45 | * are met: |
| 46 | * 1. Redistributions of source code must retain the above copyright |
| 47 | * notice, this list of conditions and the following disclaimer. |
| 48 | * 2. Redistributions in binary form must reproduce the above copyright |
| 49 | * notice, this list of conditions and the following disclaimer in the |
| 50 | * documentation and/or other materials provided with the distribution. |
| 51 | * 3. All advertising materials mentioning features or use of this software |
| 52 | * must display the following acknowledgement: |
| 53 | * This product includes software developed by the University of |
| 54 | * California, Berkeley and its contributors. |
| 55 | * 4. Neither the name of the University nor the names of its contributors |
| 56 | * may be used to endorse or promote products derived from this software |
| 57 | * without specific prior written permission. |
| 58 | * |
| 59 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 60 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 61 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 62 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 63 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 64 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 65 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 66 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 67 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 68 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 69 | * SUCH DAMAGE. |
| 70 | * |
| 71 | * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 |
| 72 | * $FreeBSD: src/sys/kern/kern_descrip.c,v 1.81.2.19 2004/02/28 00:43:31 tegge Exp $ |
| 73 | */ |
| 74 | |
| 75 | #include "opt_compat.h" |
| 76 | #include <sys/param.h> |
| 77 | #include <sys/systm.h> |
| 78 | #include <sys/malloc.h> |
| 79 | #include <sys/sysproto.h> |
| 80 | #include <sys/conf.h> |
| 81 | #include <sys/device.h> |
| 82 | #include <sys/file.h> |
| 83 | #include <sys/filedesc.h> |
| 84 | #include <sys/kernel.h> |
| 85 | #include <sys/sysctl.h> |
| 86 | #include <sys/vnode.h> |
| 87 | #include <sys/proc.h> |
| 88 | #include <sys/nlookup.h> |
| 89 | #include <sys/file.h> |
| 90 | #include <sys/stat.h> |
| 91 | #include <sys/filio.h> |
| 92 | #include <sys/fcntl.h> |
| 93 | #include <sys/unistd.h> |
| 94 | #include <sys/resourcevar.h> |
| 95 | #include <sys/event.h> |
| 96 | #include <sys/kern_syscall.h> |
| 97 | #include <sys/kcore.h> |
| 98 | #include <sys/kinfo.h> |
| 99 | #include <sys/un.h> |
| 100 | |
| 101 | #include <vm/vm.h> |
| 102 | #include <vm/vm_extern.h> |
| 103 | |
| 104 | #include <sys/thread2.h> |
| 105 | #include <sys/file2.h> |
| 106 | #include <sys/spinlock2.h> |
| 107 | #include <sys/mplock2.h> |
| 108 | |
| 109 | static void fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd); |
| 110 | static void fdreserve_locked (struct filedesc *fdp, int fd0, int incr); |
| 111 | static struct file *funsetfd_locked (struct filedesc *fdp, int fd); |
| 112 | static void ffree(struct file *fp); |
| 113 | |
| 114 | static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table"); |
| 115 | static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader", |
| 116 | "file desc to leader structures"); |
| 117 | MALLOC_DEFINE(M_FILE, "file", "Open file structure"); |
| 118 | static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); |
| 119 | |
| 120 | static struct krate krate_uidinfo = { .freq = 1 }; |
| 121 | |
| 122 | static d_open_t fdopen; |
| 123 | #define NUMFDESC 64 |
| 124 | |
| 125 | #define CDEV_MAJOR 22 |
| 126 | static struct dev_ops fildesc_ops = { |
| 127 | { "FD", 0, 0 }, |
| 128 | .d_open = fdopen, |
| 129 | }; |
| 130 | |
| 131 | /* |
| 132 | * Descriptor management. |
| 133 | */ |
| 134 | static struct filelist filehead = LIST_HEAD_INITIALIZER(&filehead); |
| 135 | static struct spinlock filehead_spin = SPINLOCK_INITIALIZER(&filehead_spin); |
| 136 | static int nfiles; /* actual number of open files */ |
| 137 | extern int cmask; |
| 138 | |
| 139 | /* |
| 140 | * Fixup fd_freefile and fd_lastfile after a descriptor has been cleared. |
| 141 | * |
| 142 | * MPSAFE - must be called with fdp->fd_spin exclusively held |
| 143 | */ |
| 144 | static __inline |
| 145 | void |
| 146 | fdfixup_locked(struct filedesc *fdp, int fd) |
| 147 | { |
| 148 | if (fd < fdp->fd_freefile) { |
| 149 | fdp->fd_freefile = fd; |
| 150 | } |
| 151 | while (fdp->fd_lastfile >= 0 && |
| 152 | fdp->fd_files[fdp->fd_lastfile].fp == NULL && |
| 153 | fdp->fd_files[fdp->fd_lastfile].reserved == 0 |
| 154 | ) { |
| 155 | --fdp->fd_lastfile; |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * System calls on descriptors. |
| 161 | * |
| 162 | * MPSAFE |
| 163 | */ |
| 164 | int |
| 165 | sys_getdtablesize(struct getdtablesize_args *uap) |
| 166 | { |
| 167 | struct proc *p = curproc; |
| 168 | struct plimit *limit = p->p_limit; |
| 169 | int dtsize; |
| 170 | |
| 171 | spin_lock(&limit->p_spin); |
| 172 | if (limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) |
| 173 | dtsize = INT_MAX; |
| 174 | else |
| 175 | dtsize = (int)limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur; |
| 176 | spin_unlock(&limit->p_spin); |
| 177 | |
| 178 | if (dtsize > maxfilesperproc) |
| 179 | dtsize = maxfilesperproc; |
| 180 | if (dtsize < minfilesperproc) |
| 181 | dtsize = minfilesperproc; |
| 182 | if (p->p_ucred->cr_uid && dtsize > maxfilesperuser) |
| 183 | dtsize = maxfilesperuser; |
| 184 | uap->sysmsg_result = dtsize; |
| 185 | return (0); |
| 186 | } |
| 187 | |
| 188 | /* |
| 189 | * Duplicate a file descriptor to a particular value. |
| 190 | * |
| 191 | * note: keep in mind that a potential race condition exists when closing |
| 192 | * descriptors from a shared descriptor table (via rfork). |
| 193 | * |
| 194 | * MPSAFE |
| 195 | */ |
| 196 | int |
| 197 | sys_dup2(struct dup2_args *uap) |
| 198 | { |
| 199 | int error; |
| 200 | int fd = 0; |
| 201 | |
| 202 | error = kern_dup(DUP_FIXED, uap->from, uap->to, &fd); |
| 203 | uap->sysmsg_fds[0] = fd; |
| 204 | |
| 205 | return (error); |
| 206 | } |
| 207 | |
| 208 | /* |
| 209 | * Duplicate a file descriptor. |
| 210 | * |
| 211 | * MPSAFE |
| 212 | */ |
| 213 | int |
| 214 | sys_dup(struct dup_args *uap) |
| 215 | { |
| 216 | int error; |
| 217 | int fd = 0; |
| 218 | |
| 219 | error = kern_dup(DUP_VARIABLE, uap->fd, 0, &fd); |
| 220 | uap->sysmsg_fds[0] = fd; |
| 221 | |
| 222 | return (error); |
| 223 | } |
| 224 | |
| 225 | /* |
| 226 | * MPALMOSTSAFE - acquires mplock for fp operations |
| 227 | */ |
| 228 | int |
| 229 | kern_fcntl(int fd, int cmd, union fcntl_dat *dat, struct ucred *cred) |
| 230 | { |
| 231 | struct thread *td = curthread; |
| 232 | struct proc *p = td->td_proc; |
| 233 | struct file *fp; |
| 234 | struct vnode *vp; |
| 235 | u_int newmin; |
| 236 | u_int oflags; |
| 237 | u_int nflags; |
| 238 | int tmp, error, flg = F_POSIX; |
| 239 | |
| 240 | KKASSERT(p); |
| 241 | |
| 242 | /* |
| 243 | * Operations on file descriptors that do not require a file pointer. |
| 244 | */ |
| 245 | switch (cmd) { |
| 246 | case F_GETFD: |
| 247 | error = fgetfdflags(p->p_fd, fd, &tmp); |
| 248 | if (error == 0) |
| 249 | dat->fc_cloexec = (tmp & UF_EXCLOSE) ? FD_CLOEXEC : 0; |
| 250 | return (error); |
| 251 | |
| 252 | case F_SETFD: |
| 253 | if (dat->fc_cloexec & FD_CLOEXEC) |
| 254 | error = fsetfdflags(p->p_fd, fd, UF_EXCLOSE); |
| 255 | else |
| 256 | error = fclrfdflags(p->p_fd, fd, UF_EXCLOSE); |
| 257 | return (error); |
| 258 | case F_DUPFD: |
| 259 | newmin = dat->fc_fd; |
| 260 | error = kern_dup(DUP_VARIABLE, fd, newmin, &dat->fc_fd); |
| 261 | return (error); |
| 262 | default: |
| 263 | break; |
| 264 | } |
| 265 | |
| 266 | /* |
| 267 | * Operations on file pointers |
| 268 | */ |
| 269 | if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) |
| 270 | return (EBADF); |
| 271 | |
| 272 | get_mplock(); |
| 273 | switch (cmd) { |
| 274 | case F_GETFL: |
| 275 | dat->fc_flags = OFLAGS(fp->f_flag); |
| 276 | error = 0; |
| 277 | break; |
| 278 | |
| 279 | case F_SETFL: |
| 280 | oflags = fp->f_flag; |
| 281 | nflags = FFLAGS(dat->fc_flags & ~O_ACCMODE) & FCNTLFLAGS; |
| 282 | nflags |= oflags & ~FCNTLFLAGS; |
| 283 | |
| 284 | error = 0; |
| 285 | if (((nflags ^ oflags) & O_APPEND) && (oflags & FAPPENDONLY)) |
| 286 | error = EINVAL; |
| 287 | if (error == 0 && ((nflags ^ oflags) & FASYNC)) { |
| 288 | tmp = nflags & FASYNC; |
| 289 | error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, |
| 290 | cred, NULL); |
| 291 | } |
| 292 | if (error == 0) |
| 293 | fp->f_flag = nflags; |
| 294 | break; |
| 295 | |
| 296 | case F_GETOWN: |
| 297 | error = fo_ioctl(fp, FIOGETOWN, (caddr_t)&dat->fc_owner, |
| 298 | cred, NULL); |
| 299 | break; |
| 300 | |
| 301 | case F_SETOWN: |
| 302 | error = fo_ioctl(fp, FIOSETOWN, (caddr_t)&dat->fc_owner, |
| 303 | cred, NULL); |
| 304 | break; |
| 305 | |
| 306 | case F_SETLKW: |
| 307 | flg |= F_WAIT; |
| 308 | /* Fall into F_SETLK */ |
| 309 | |
| 310 | case F_SETLK: |
| 311 | if (fp->f_type != DTYPE_VNODE) { |
| 312 | error = EBADF; |
| 313 | break; |
| 314 | } |
| 315 | vp = (struct vnode *)fp->f_data; |
| 316 | |
| 317 | /* |
| 318 | * copyin/lockop may block |
| 319 | */ |
| 320 | if (dat->fc_flock.l_whence == SEEK_CUR) |
| 321 | dat->fc_flock.l_start += fp->f_offset; |
| 322 | |
| 323 | switch (dat->fc_flock.l_type) { |
| 324 | case F_RDLCK: |
| 325 | if ((fp->f_flag & FREAD) == 0) { |
| 326 | error = EBADF; |
| 327 | break; |
| 328 | } |
| 329 | p->p_leader->p_flag |= P_ADVLOCK; |
| 330 | error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, |
| 331 | &dat->fc_flock, flg); |
| 332 | break; |
| 333 | case F_WRLCK: |
| 334 | if ((fp->f_flag & FWRITE) == 0) { |
| 335 | error = EBADF; |
| 336 | break; |
| 337 | } |
| 338 | p->p_leader->p_flag |= P_ADVLOCK; |
| 339 | error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, |
| 340 | &dat->fc_flock, flg); |
| 341 | break; |
| 342 | case F_UNLCK: |
| 343 | error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, |
| 344 | &dat->fc_flock, F_POSIX); |
| 345 | break; |
| 346 | default: |
| 347 | error = EINVAL; |
| 348 | break; |
| 349 | } |
| 350 | |
| 351 | /* |
| 352 | * It is possible to race a close() on the descriptor while |
| 353 | * we were blocked getting the lock. If this occurs the |
| 354 | * close might not have caught the lock. |
| 355 | */ |
| 356 | if (checkfdclosed(p->p_fd, fd, fp)) { |
| 357 | dat->fc_flock.l_whence = SEEK_SET; |
| 358 | dat->fc_flock.l_start = 0; |
| 359 | dat->fc_flock.l_len = 0; |
| 360 | dat->fc_flock.l_type = F_UNLCK; |
| 361 | (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, |
| 362 | F_UNLCK, &dat->fc_flock, F_POSIX); |
| 363 | } |
| 364 | break; |
| 365 | |
| 366 | case F_GETLK: |
| 367 | if (fp->f_type != DTYPE_VNODE) { |
| 368 | error = EBADF; |
| 369 | break; |
| 370 | } |
| 371 | vp = (struct vnode *)fp->f_data; |
| 372 | /* |
| 373 | * copyin/lockop may block |
| 374 | */ |
| 375 | if (dat->fc_flock.l_type != F_RDLCK && |
| 376 | dat->fc_flock.l_type != F_WRLCK && |
| 377 | dat->fc_flock.l_type != F_UNLCK) { |
| 378 | error = EINVAL; |
| 379 | break; |
| 380 | } |
| 381 | if (dat->fc_flock.l_whence == SEEK_CUR) |
| 382 | dat->fc_flock.l_start += fp->f_offset; |
| 383 | error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, |
| 384 | &dat->fc_flock, F_POSIX); |
| 385 | break; |
| 386 | default: |
| 387 | error = EINVAL; |
| 388 | break; |
| 389 | } |
| 390 | rel_mplock(); |
| 391 | |
| 392 | fdrop(fp); |
| 393 | return (error); |
| 394 | } |
| 395 | |
| 396 | /* |
| 397 | * The file control system call. |
| 398 | * |
| 399 | * MPSAFE |
| 400 | */ |
| 401 | int |
| 402 | sys_fcntl(struct fcntl_args *uap) |
| 403 | { |
| 404 | union fcntl_dat dat; |
| 405 | int error; |
| 406 | |
| 407 | switch (uap->cmd) { |
| 408 | case F_DUPFD: |
| 409 | dat.fc_fd = uap->arg; |
| 410 | break; |
| 411 | case F_SETFD: |
| 412 | dat.fc_cloexec = uap->arg; |
| 413 | break; |
| 414 | case F_SETFL: |
| 415 | dat.fc_flags = uap->arg; |
| 416 | break; |
| 417 | case F_SETOWN: |
| 418 | dat.fc_owner = uap->arg; |
| 419 | break; |
| 420 | case F_SETLKW: |
| 421 | case F_SETLK: |
| 422 | case F_GETLK: |
| 423 | error = copyin((caddr_t)uap->arg, &dat.fc_flock, |
| 424 | sizeof(struct flock)); |
| 425 | if (error) |
| 426 | return (error); |
| 427 | break; |
| 428 | } |
| 429 | |
| 430 | error = kern_fcntl(uap->fd, uap->cmd, &dat, curthread->td_ucred); |
| 431 | |
| 432 | if (error == 0) { |
| 433 | switch (uap->cmd) { |
| 434 | case F_DUPFD: |
| 435 | uap->sysmsg_result = dat.fc_fd; |
| 436 | break; |
| 437 | case F_GETFD: |
| 438 | uap->sysmsg_result = dat.fc_cloexec; |
| 439 | break; |
| 440 | case F_GETFL: |
| 441 | uap->sysmsg_result = dat.fc_flags; |
| 442 | break; |
| 443 | case F_GETOWN: |
| 444 | uap->sysmsg_result = dat.fc_owner; |
| 445 | case F_GETLK: |
| 446 | error = copyout(&dat.fc_flock, (caddr_t)uap->arg, |
| 447 | sizeof(struct flock)); |
| 448 | break; |
| 449 | } |
| 450 | } |
| 451 | |
| 452 | return (error); |
| 453 | } |
| 454 | |
| 455 | /* |
| 456 | * Common code for dup, dup2, and fcntl(F_DUPFD). |
| 457 | * |
| 458 | * The type flag can be either DUP_FIXED or DUP_VARIABLE. DUP_FIXED tells |
| 459 | * kern_dup() to destructively dup over an existing file descriptor if new |
| 460 | * is already open. DUP_VARIABLE tells kern_dup() to find the lowest |
| 461 | * unused file descriptor that is greater than or equal to new. |
| 462 | * |
| 463 | * MPSAFE |
| 464 | */ |
| 465 | int |
| 466 | kern_dup(enum dup_type type, int old, int new, int *res) |
| 467 | { |
| 468 | struct thread *td = curthread; |
| 469 | struct proc *p = td->td_proc; |
| 470 | struct filedesc *fdp = p->p_fd; |
| 471 | struct file *fp; |
| 472 | struct file *delfp; |
| 473 | int oldflags; |
| 474 | int holdleaders; |
| 475 | int dtsize; |
| 476 | int error, newfd; |
| 477 | |
| 478 | /* |
| 479 | * Verify that we have a valid descriptor to dup from and |
| 480 | * possibly to dup to. |
| 481 | * |
| 482 | * NOTE: maxfilesperuser is not applicable to dup() |
| 483 | */ |
| 484 | retry: |
| 485 | if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) |
| 486 | dtsize = INT_MAX; |
| 487 | else |
| 488 | dtsize = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; |
| 489 | if (dtsize > maxfilesperproc) |
| 490 | dtsize = maxfilesperproc; |
| 491 | if (dtsize < minfilesperproc) |
| 492 | dtsize = minfilesperproc; |
| 493 | |
| 494 | if (new < 0 || new > dtsize) |
| 495 | return (EINVAL); |
| 496 | |
| 497 | spin_lock(&fdp->fd_spin); |
| 498 | if ((unsigned)old >= fdp->fd_nfiles || fdp->fd_files[old].fp == NULL) { |
| 499 | spin_unlock(&fdp->fd_spin); |
| 500 | return (EBADF); |
| 501 | } |
| 502 | if (type == DUP_FIXED && old == new) { |
| 503 | *res = new; |
| 504 | spin_unlock(&fdp->fd_spin); |
| 505 | return (0); |
| 506 | } |
| 507 | fp = fdp->fd_files[old].fp; |
| 508 | oldflags = fdp->fd_files[old].fileflags; |
| 509 | fhold(fp); /* MPSAFE - can be called with a spinlock held */ |
| 510 | |
| 511 | /* |
| 512 | * Allocate a new descriptor if DUP_VARIABLE, or expand the table |
| 513 | * if the requested descriptor is beyond the current table size. |
| 514 | * |
| 515 | * This can block. Retry if the source descriptor no longer matches |
| 516 | * or if our expectation in the expansion case races. |
| 517 | * |
| 518 | * If we are not expanding or allocating a new decriptor, then reset |
| 519 | * the target descriptor to a reserved state so we have a uniform |
| 520 | * setup for the next code block. |
| 521 | */ |
| 522 | if (type == DUP_VARIABLE || new >= fdp->fd_nfiles) { |
| 523 | spin_unlock(&fdp->fd_spin); |
| 524 | error = fdalloc(p, new, &newfd); |
| 525 | spin_lock(&fdp->fd_spin); |
| 526 | if (error) { |
| 527 | spin_unlock(&fdp->fd_spin); |
| 528 | fdrop(fp); |
| 529 | return (error); |
| 530 | } |
| 531 | /* |
| 532 | * Check for ripout |
| 533 | */ |
| 534 | if (old >= fdp->fd_nfiles || fdp->fd_files[old].fp != fp) { |
| 535 | fsetfd_locked(fdp, NULL, newfd); |
| 536 | spin_unlock(&fdp->fd_spin); |
| 537 | fdrop(fp); |
| 538 | goto retry; |
| 539 | } |
| 540 | /* |
| 541 | * Check for expansion race |
| 542 | */ |
| 543 | if (type != DUP_VARIABLE && new != newfd) { |
| 544 | fsetfd_locked(fdp, NULL, newfd); |
| 545 | spin_unlock(&fdp->fd_spin); |
| 546 | fdrop(fp); |
| 547 | goto retry; |
| 548 | } |
| 549 | /* |
| 550 | * Check for ripout, newfd reused old (this case probably |
| 551 | * can't occur). |
| 552 | */ |
| 553 | if (old == newfd) { |
| 554 | fsetfd_locked(fdp, NULL, newfd); |
| 555 | spin_unlock(&fdp->fd_spin); |
| 556 | fdrop(fp); |
| 557 | goto retry; |
| 558 | } |
| 559 | new = newfd; |
| 560 | delfp = NULL; |
| 561 | } else { |
| 562 | if (fdp->fd_files[new].reserved) { |
| 563 | spin_unlock(&fdp->fd_spin); |
| 564 | fdrop(fp); |
| 565 | kprintf("Warning: dup(): target descriptor %d is reserved, waiting for it to be resolved\n", new); |
| 566 | tsleep(fdp, 0, "fdres", hz); |
| 567 | goto retry; |
| 568 | } |
| 569 | |
| 570 | /* |
| 571 | * If the target descriptor was never allocated we have |
| 572 | * to allocate it. If it was we have to clean out the |
| 573 | * old descriptor. delfp inherits the ref from the |
| 574 | * descriptor table. |
| 575 | */ |
| 576 | delfp = fdp->fd_files[new].fp; |
| 577 | fdp->fd_files[new].fp = NULL; |
| 578 | fdp->fd_files[new].reserved = 1; |
| 579 | if (delfp == NULL) { |
| 580 | fdreserve_locked(fdp, new, 1); |
| 581 | if (new > fdp->fd_lastfile) |
| 582 | fdp->fd_lastfile = new; |
| 583 | } |
| 584 | |
| 585 | } |
| 586 | |
| 587 | /* |
| 588 | * NOTE: still holding an exclusive spinlock |
| 589 | */ |
| 590 | |
| 591 | /* |
| 592 | * If a descriptor is being overwritten we may hve to tell |
| 593 | * fdfree() to sleep to ensure that all relevant process |
| 594 | * leaders can be traversed in closef(). |
| 595 | */ |
| 596 | if (delfp != NULL && p->p_fdtol != NULL) { |
| 597 | fdp->fd_holdleaderscount++; |
| 598 | holdleaders = 1; |
| 599 | } else { |
| 600 | holdleaders = 0; |
| 601 | } |
| 602 | KASSERT(delfp == NULL || type == DUP_FIXED, |
| 603 | ("dup() picked an open file")); |
| 604 | |
| 605 | /* |
| 606 | * Duplicate the source descriptor, update lastfile. If the new |
| 607 | * descriptor was not allocated and we aren't replacing an existing |
| 608 | * descriptor we have to mark the descriptor as being in use. |
| 609 | * |
| 610 | * The fd_files[] array inherits fp's hold reference. |
| 611 | */ |
| 612 | fsetfd_locked(fdp, fp, new); |
| 613 | fdp->fd_files[new].fileflags = oldflags & ~UF_EXCLOSE; |
| 614 | spin_unlock(&fdp->fd_spin); |
| 615 | fdrop(fp); |
| 616 | *res = new; |
| 617 | |
| 618 | /* |
| 619 | * If we dup'd over a valid file, we now own the reference to it |
| 620 | * and must dispose of it using closef() semantics (as if a |
| 621 | * close() were performed on it). |
| 622 | */ |
| 623 | if (delfp) { |
| 624 | if (SLIST_FIRST(&delfp->f_klist)) |
| 625 | knote_fdclose(delfp, fdp, new); |
| 626 | closef(delfp, p); |
| 627 | if (holdleaders) { |
| 628 | spin_lock(&fdp->fd_spin); |
| 629 | fdp->fd_holdleaderscount--; |
| 630 | if (fdp->fd_holdleaderscount == 0 && |
| 631 | fdp->fd_holdleaderswakeup != 0) { |
| 632 | fdp->fd_holdleaderswakeup = 0; |
| 633 | spin_unlock(&fdp->fd_spin); |
| 634 | wakeup(&fdp->fd_holdleaderscount); |
| 635 | } else { |
| 636 | spin_unlock(&fdp->fd_spin); |
| 637 | } |
| 638 | } |
| 639 | } |
| 640 | return (0); |
| 641 | } |
| 642 | |
| 643 | /* |
| 644 | * If sigio is on the list associated with a process or process group, |
| 645 | * disable signalling from the device, remove sigio from the list and |
| 646 | * free sigio. |
| 647 | */ |
| 648 | void |
| 649 | funsetown(struct sigio *sigio) |
| 650 | { |
| 651 | if (sigio == NULL) |
| 652 | return; |
| 653 | crit_enter(); |
| 654 | *(sigio->sio_myref) = NULL; |
| 655 | crit_exit(); |
| 656 | if (sigio->sio_pgid < 0) { |
| 657 | SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, |
| 658 | sigio, sio_pgsigio); |
| 659 | } else /* if ((*sigiop)->sio_pgid > 0) */ { |
| 660 | SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, |
| 661 | sigio, sio_pgsigio); |
| 662 | } |
| 663 | crfree(sigio->sio_ucred); |
| 664 | kfree(sigio, M_SIGIO); |
| 665 | } |
| 666 | |
| 667 | /* Free a list of sigio structures. */ |
| 668 | void |
| 669 | funsetownlst(struct sigiolst *sigiolst) |
| 670 | { |
| 671 | struct sigio *sigio; |
| 672 | |
| 673 | while ((sigio = SLIST_FIRST(sigiolst)) != NULL) |
| 674 | funsetown(sigio); |
| 675 | } |
| 676 | |
| 677 | /* |
| 678 | * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). |
| 679 | * |
| 680 | * After permission checking, add a sigio structure to the sigio list for |
| 681 | * the process or process group. |
| 682 | */ |
| 683 | int |
| 684 | fsetown(pid_t pgid, struct sigio **sigiop) |
| 685 | { |
| 686 | struct proc *proc; |
| 687 | struct pgrp *pgrp; |
| 688 | struct sigio *sigio; |
| 689 | |
| 690 | if (pgid == 0) { |
| 691 | funsetown(*sigiop); |
| 692 | return (0); |
| 693 | } |
| 694 | if (pgid > 0) { |
| 695 | proc = pfind(pgid); |
| 696 | if (proc == NULL) |
| 697 | return (ESRCH); |
| 698 | |
| 699 | /* |
| 700 | * Policy - Don't allow a process to FSETOWN a process |
| 701 | * in another session. |
| 702 | * |
| 703 | * Remove this test to allow maximum flexibility or |
| 704 | * restrict FSETOWN to the current process or process |
| 705 | * group for maximum safety. |
| 706 | */ |
| 707 | if (proc->p_session != curproc->p_session) |
| 708 | return (EPERM); |
| 709 | |
| 710 | pgrp = NULL; |
| 711 | } else /* if (pgid < 0) */ { |
| 712 | pgrp = pgfind(-pgid); |
| 713 | if (pgrp == NULL) |
| 714 | return (ESRCH); |
| 715 | |
| 716 | /* |
| 717 | * Policy - Don't allow a process to FSETOWN a process |
| 718 | * in another session. |
| 719 | * |
| 720 | * Remove this test to allow maximum flexibility or |
| 721 | * restrict FSETOWN to the current process or process |
| 722 | * group for maximum safety. |
| 723 | */ |
| 724 | if (pgrp->pg_session != curproc->p_session) |
| 725 | return (EPERM); |
| 726 | |
| 727 | proc = NULL; |
| 728 | } |
| 729 | funsetown(*sigiop); |
| 730 | sigio = kmalloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); |
| 731 | if (pgid > 0) { |
| 732 | SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); |
| 733 | sigio->sio_proc = proc; |
| 734 | } else { |
| 735 | SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); |
| 736 | sigio->sio_pgrp = pgrp; |
| 737 | } |
| 738 | sigio->sio_pgid = pgid; |
| 739 | sigio->sio_ucred = crhold(curthread->td_ucred); |
| 740 | /* It would be convenient if p_ruid was in ucred. */ |
| 741 | sigio->sio_ruid = sigio->sio_ucred->cr_ruid; |
| 742 | sigio->sio_myref = sigiop; |
| 743 | crit_enter(); |
| 744 | *sigiop = sigio; |
| 745 | crit_exit(); |
| 746 | return (0); |
| 747 | } |
| 748 | |
| 749 | /* |
| 750 | * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). |
| 751 | */ |
| 752 | pid_t |
| 753 | fgetown(struct sigio *sigio) |
| 754 | { |
| 755 | return (sigio != NULL ? sigio->sio_pgid : 0); |
| 756 | } |
| 757 | |
| 758 | /* |
| 759 | * Close many file descriptors. |
| 760 | * |
| 761 | * MPSAFE |
| 762 | */ |
| 763 | int |
| 764 | sys_closefrom(struct closefrom_args *uap) |
| 765 | { |
| 766 | return(kern_closefrom(uap->fd)); |
| 767 | } |
| 768 | |
| 769 | /* |
| 770 | * Close all file descriptors greater then or equal to fd |
| 771 | * |
| 772 | * MPSAFE |
| 773 | */ |
| 774 | int |
| 775 | kern_closefrom(int fd) |
| 776 | { |
| 777 | struct thread *td = curthread; |
| 778 | struct proc *p = td->td_proc; |
| 779 | struct filedesc *fdp; |
| 780 | |
| 781 | KKASSERT(p); |
| 782 | fdp = p->p_fd; |
| 783 | |
| 784 | if (fd < 0) |
| 785 | return (EINVAL); |
| 786 | |
| 787 | /* |
| 788 | * NOTE: This function will skip unassociated descriptors and |
| 789 | * reserved descriptors that have not yet been assigned. |
| 790 | * fd_lastfile can change as a side effect of kern_close(). |
| 791 | */ |
| 792 | spin_lock(&fdp->fd_spin); |
| 793 | while (fd <= fdp->fd_lastfile) { |
| 794 | if (fdp->fd_files[fd].fp != NULL) { |
| 795 | spin_unlock(&fdp->fd_spin); |
| 796 | /* ok if this races another close */ |
| 797 | if (kern_close(fd) == EINTR) |
| 798 | return (EINTR); |
| 799 | spin_lock(&fdp->fd_spin); |
| 800 | } |
| 801 | ++fd; |
| 802 | } |
| 803 | spin_unlock(&fdp->fd_spin); |
| 804 | return (0); |
| 805 | } |
| 806 | |
| 807 | /* |
| 808 | * Close a file descriptor. |
| 809 | * |
| 810 | * MPSAFE |
| 811 | */ |
| 812 | int |
| 813 | sys_close(struct close_args *uap) |
| 814 | { |
| 815 | return(kern_close(uap->fd)); |
| 816 | } |
| 817 | |
| 818 | /* |
| 819 | * MPALMOSTSAFE - acquires mplock around knote_fdclose() calls |
| 820 | */ |
| 821 | int |
| 822 | kern_close(int fd) |
| 823 | { |
| 824 | struct thread *td = curthread; |
| 825 | struct proc *p = td->td_proc; |
| 826 | struct filedesc *fdp; |
| 827 | struct file *fp; |
| 828 | int error; |
| 829 | int holdleaders; |
| 830 | |
| 831 | KKASSERT(p); |
| 832 | fdp = p->p_fd; |
| 833 | |
| 834 | spin_lock(&fdp->fd_spin); |
| 835 | if ((fp = funsetfd_locked(fdp, fd)) == NULL) { |
| 836 | spin_unlock(&fdp->fd_spin); |
| 837 | return (EBADF); |
| 838 | } |
| 839 | holdleaders = 0; |
| 840 | if (p->p_fdtol != NULL) { |
| 841 | /* |
| 842 | * Ask fdfree() to sleep to ensure that all relevant |
| 843 | * process leaders can be traversed in closef(). |
| 844 | */ |
| 845 | fdp->fd_holdleaderscount++; |
| 846 | holdleaders = 1; |
| 847 | } |
| 848 | |
| 849 | /* |
| 850 | * we now hold the fp reference that used to be owned by the descriptor |
| 851 | * array. |
| 852 | */ |
| 853 | spin_unlock(&fdp->fd_spin); |
| 854 | if (SLIST_FIRST(&fp->f_klist)) |
| 855 | knote_fdclose(fp, fdp, fd); |
| 856 | error = closef(fp, p); |
| 857 | if (holdleaders) { |
| 858 | spin_lock(&fdp->fd_spin); |
| 859 | fdp->fd_holdleaderscount--; |
| 860 | if (fdp->fd_holdleaderscount == 0 && |
| 861 | fdp->fd_holdleaderswakeup != 0) { |
| 862 | fdp->fd_holdleaderswakeup = 0; |
| 863 | spin_unlock(&fdp->fd_spin); |
| 864 | wakeup(&fdp->fd_holdleaderscount); |
| 865 | } else { |
| 866 | spin_unlock(&fdp->fd_spin); |
| 867 | } |
| 868 | } |
| 869 | return (error); |
| 870 | } |
| 871 | |
| 872 | /* |
| 873 | * shutdown_args(int fd, int how) |
| 874 | */ |
| 875 | int |
| 876 | kern_shutdown(int fd, int how) |
| 877 | { |
| 878 | struct thread *td = curthread; |
| 879 | struct proc *p = td->td_proc; |
| 880 | struct file *fp; |
| 881 | int error; |
| 882 | |
| 883 | KKASSERT(p); |
| 884 | |
| 885 | if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) |
| 886 | return (EBADF); |
| 887 | error = fo_shutdown(fp, how); |
| 888 | fdrop(fp); |
| 889 | |
| 890 | return (error); |
| 891 | } |
| 892 | |
| 893 | /* |
| 894 | * MPALMOSTSAFE |
| 895 | */ |
| 896 | int |
| 897 | sys_shutdown(struct shutdown_args *uap) |
| 898 | { |
| 899 | int error; |
| 900 | |
| 901 | get_mplock(); |
| 902 | error = kern_shutdown(uap->s, uap->how); |
| 903 | rel_mplock(); |
| 904 | |
| 905 | return (error); |
| 906 | } |
| 907 | |
| 908 | /* |
| 909 | * MPSAFE |
| 910 | */ |
| 911 | int |
| 912 | kern_fstat(int fd, struct stat *ub) |
| 913 | { |
| 914 | struct thread *td = curthread; |
| 915 | struct proc *p = td->td_proc; |
| 916 | struct file *fp; |
| 917 | int error; |
| 918 | |
| 919 | KKASSERT(p); |
| 920 | |
| 921 | if ((fp = holdfp(p->p_fd, fd, -1)) == NULL) |
| 922 | return (EBADF); |
| 923 | error = fo_stat(fp, ub, td->td_ucred); |
| 924 | fdrop(fp); |
| 925 | |
| 926 | return (error); |
| 927 | } |
| 928 | |
| 929 | /* |
| 930 | * Return status information about a file descriptor. |
| 931 | * |
| 932 | * MPSAFE |
| 933 | */ |
| 934 | int |
| 935 | sys_fstat(struct fstat_args *uap) |
| 936 | { |
| 937 | struct stat st; |
| 938 | int error; |
| 939 | |
| 940 | error = kern_fstat(uap->fd, &st); |
| 941 | |
| 942 | if (error == 0) |
| 943 | error = copyout(&st, uap->sb, sizeof(st)); |
| 944 | return (error); |
| 945 | } |
| 946 | |
| 947 | /* |
| 948 | * Return pathconf information about a file descriptor. |
| 949 | * |
| 950 | * MPALMOSTSAFE |
| 951 | */ |
| 952 | int |
| 953 | sys_fpathconf(struct fpathconf_args *uap) |
| 954 | { |
| 955 | struct thread *td = curthread; |
| 956 | struct proc *p = td->td_proc; |
| 957 | struct file *fp; |
| 958 | struct vnode *vp; |
| 959 | int error = 0; |
| 960 | |
| 961 | if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) |
| 962 | return (EBADF); |
| 963 | |
| 964 | switch (fp->f_type) { |
| 965 | case DTYPE_PIPE: |
| 966 | case DTYPE_SOCKET: |
| 967 | if (uap->name != _PC_PIPE_BUF) { |
| 968 | error = EINVAL; |
| 969 | } else { |
| 970 | uap->sysmsg_result = PIPE_BUF; |
| 971 | error = 0; |
| 972 | } |
| 973 | break; |
| 974 | case DTYPE_FIFO: |
| 975 | case DTYPE_VNODE: |
| 976 | vp = (struct vnode *)fp->f_data; |
| 977 | get_mplock(); |
| 978 | error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); |
| 979 | rel_mplock(); |
| 980 | break; |
| 981 | default: |
| 982 | error = EOPNOTSUPP; |
| 983 | break; |
| 984 | } |
| 985 | fdrop(fp); |
| 986 | return(error); |
| 987 | } |
| 988 | |
| 989 | static int fdexpand; |
| 990 | SYSCTL_INT(_debug, OID_AUTO, fdexpand, CTLFLAG_RD, &fdexpand, 0, |
| 991 | "Number of times a file table has been expanded"); |
| 992 | |
| 993 | /* |
| 994 | * Grow the file table so it can hold through descriptor (want). |
| 995 | * |
| 996 | * The fdp's spinlock must be held exclusively on entry and may be held |
| 997 | * exclusively on return. The spinlock may be cycled by the routine. |
| 998 | * |
| 999 | * MPSAFE |
| 1000 | */ |
| 1001 | static void |
| 1002 | fdgrow_locked(struct filedesc *fdp, int want) |
| 1003 | { |
| 1004 | struct fdnode *newfiles; |
| 1005 | struct fdnode *oldfiles; |
| 1006 | int nf, extra; |
| 1007 | |
| 1008 | nf = fdp->fd_nfiles; |
| 1009 | do { |
| 1010 | /* nf has to be of the form 2^n - 1 */ |
| 1011 | nf = 2 * nf + 1; |
| 1012 | } while (nf <= want); |
| 1013 | |
| 1014 | spin_unlock(&fdp->fd_spin); |
| 1015 | newfiles = kmalloc(nf * sizeof(struct fdnode), M_FILEDESC, M_WAITOK); |
| 1016 | spin_lock(&fdp->fd_spin); |
| 1017 | |
| 1018 | /* |
| 1019 | * We could have raced another extend while we were not holding |
| 1020 | * the spinlock. |
| 1021 | */ |
| 1022 | if (fdp->fd_nfiles >= nf) { |
| 1023 | spin_unlock(&fdp->fd_spin); |
| 1024 | kfree(newfiles, M_FILEDESC); |
| 1025 | spin_lock(&fdp->fd_spin); |
| 1026 | return; |
| 1027 | } |
| 1028 | /* |
| 1029 | * Copy the existing ofile and ofileflags arrays |
| 1030 | * and zero the new portion of each array. |
| 1031 | */ |
| 1032 | extra = nf - fdp->fd_nfiles; |
| 1033 | bcopy(fdp->fd_files, newfiles, fdp->fd_nfiles * sizeof(struct fdnode)); |
| 1034 | bzero(&newfiles[fdp->fd_nfiles], extra * sizeof(struct fdnode)); |
| 1035 | |
| 1036 | oldfiles = fdp->fd_files; |
| 1037 | fdp->fd_files = newfiles; |
| 1038 | fdp->fd_nfiles = nf; |
| 1039 | |
| 1040 | if (oldfiles != fdp->fd_builtin_files) { |
| 1041 | spin_unlock(&fdp->fd_spin); |
| 1042 | kfree(oldfiles, M_FILEDESC); |
| 1043 | spin_lock(&fdp->fd_spin); |
| 1044 | } |
| 1045 | fdexpand++; |
| 1046 | } |
| 1047 | |
| 1048 | /* |
| 1049 | * Number of nodes in right subtree, including the root. |
| 1050 | */ |
| 1051 | static __inline int |
| 1052 | right_subtree_size(int n) |
| 1053 | { |
| 1054 | return (n ^ (n | (n + 1))); |
| 1055 | } |
| 1056 | |
| 1057 | /* |
| 1058 | * Bigger ancestor. |
| 1059 | */ |
| 1060 | static __inline int |
| 1061 | right_ancestor(int n) |
| 1062 | { |
| 1063 | return (n | (n + 1)); |
| 1064 | } |
| 1065 | |
| 1066 | /* |
| 1067 | * Smaller ancestor. |
| 1068 | */ |
| 1069 | static __inline int |
| 1070 | left_ancestor(int n) |
| 1071 | { |
| 1072 | return ((n & (n + 1)) - 1); |
| 1073 | } |
| 1074 | |
| 1075 | /* |
| 1076 | * Traverse the in-place binary tree buttom-up adjusting the allocation |
| 1077 | * count so scans can determine where free descriptors are located. |
| 1078 | * |
| 1079 | * MPSAFE - caller must be holding an exclusive spinlock on fdp |
| 1080 | */ |
| 1081 | static |
| 1082 | void |
| 1083 | fdreserve_locked(struct filedesc *fdp, int fd, int incr) |
| 1084 | { |
| 1085 | while (fd >= 0) { |
| 1086 | fdp->fd_files[fd].allocated += incr; |
| 1087 | KKASSERT(fdp->fd_files[fd].allocated >= 0); |
| 1088 | fd = left_ancestor(fd); |
| 1089 | } |
| 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * Reserve a file descriptor for the process. If no error occurs, the |
| 1094 | * caller MUST at some point call fsetfd() or assign a file pointer |
| 1095 | * or dispose of the reservation. |
| 1096 | * |
| 1097 | * MPSAFE |
| 1098 | */ |
| 1099 | int |
| 1100 | fdalloc(struct proc *p, int want, int *result) |
| 1101 | { |
| 1102 | struct filedesc *fdp = p->p_fd; |
| 1103 | struct uidinfo *uip; |
| 1104 | int fd, rsize, rsum, node, lim; |
| 1105 | |
| 1106 | /* |
| 1107 | * Check dtable size limit |
| 1108 | */ |
| 1109 | spin_lock(&p->p_limit->p_spin); |
| 1110 | if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) |
| 1111 | lim = INT_MAX; |
| 1112 | else |
| 1113 | lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; |
| 1114 | spin_unlock(&p->p_limit->p_spin); |
| 1115 | |
| 1116 | if (lim > maxfilesperproc) |
| 1117 | lim = maxfilesperproc; |
| 1118 | if (lim < minfilesperproc) |
| 1119 | lim = minfilesperproc; |
| 1120 | if (want >= lim) |
| 1121 | return (EMFILE); |
| 1122 | |
| 1123 | /* |
| 1124 | * Check that the user has not run out of descriptors (non-root only). |
| 1125 | * As a safety measure the dtable is allowed to have at least |
| 1126 | * minfilesperproc open fds regardless of the maxfilesperuser limit. |
| 1127 | */ |
| 1128 | if (p->p_ucred->cr_uid && fdp->fd_nfiles >= minfilesperproc) { |
| 1129 | uip = p->p_ucred->cr_uidinfo; |
| 1130 | if (uip->ui_openfiles > maxfilesperuser) { |
| 1131 | krateprintf(&krate_uidinfo, |
| 1132 | "Warning: user %d pid %d (%s) ran out of " |
| 1133 | "file descriptors (%d/%d)\n", |
| 1134 | p->p_ucred->cr_uid, (int)p->p_pid, |
| 1135 | p->p_comm, |
| 1136 | uip->ui_openfiles, maxfilesperuser); |
| 1137 | return(ENFILE); |
| 1138 | } |
| 1139 | } |
| 1140 | |
| 1141 | /* |
| 1142 | * Grow the dtable if necessary |
| 1143 | */ |
| 1144 | spin_lock(&fdp->fd_spin); |
| 1145 | if (want >= fdp->fd_nfiles) |
| 1146 | fdgrow_locked(fdp, want); |
| 1147 | |
| 1148 | /* |
| 1149 | * Search for a free descriptor starting at the higher |
| 1150 | * of want or fd_freefile. If that fails, consider |
| 1151 | * expanding the ofile array. |
| 1152 | * |
| 1153 | * NOTE! the 'allocated' field is a cumulative recursive allocation |
| 1154 | * count. If we happen to see a value of 0 then we can shortcut |
| 1155 | * our search. Otherwise we run through through the tree going |
| 1156 | * down branches we know have free descriptor(s) until we hit a |
| 1157 | * leaf node. The leaf node will be free but will not necessarily |
| 1158 | * have an allocated field of 0. |
| 1159 | */ |
| 1160 | retry: |
| 1161 | /* move up the tree looking for a subtree with a free node */ |
| 1162 | for (fd = max(want, fdp->fd_freefile); fd < min(fdp->fd_nfiles, lim); |
| 1163 | fd = right_ancestor(fd)) { |
| 1164 | if (fdp->fd_files[fd].allocated == 0) |
| 1165 | goto found; |
| 1166 | |
| 1167 | rsize = right_subtree_size(fd); |
| 1168 | if (fdp->fd_files[fd].allocated == rsize) |
| 1169 | continue; /* right subtree full */ |
| 1170 | |
| 1171 | /* |
| 1172 | * Free fd is in the right subtree of the tree rooted at fd. |
| 1173 | * Call that subtree R. Look for the smallest (leftmost) |
| 1174 | * subtree of R with an unallocated fd: continue moving |
| 1175 | * down the left branch until encountering a full left |
| 1176 | * subtree, then move to the right. |
| 1177 | */ |
| 1178 | for (rsum = 0, rsize /= 2; rsize > 0; rsize /= 2) { |
| 1179 | node = fd + rsize; |
| 1180 | rsum += fdp->fd_files[node].allocated; |
| 1181 | if (fdp->fd_files[fd].allocated == rsum + rsize) { |
| 1182 | fd = node; /* move to the right */ |
| 1183 | if (fdp->fd_files[node].allocated == 0) |
| 1184 | goto found; |
| 1185 | rsum = 0; |
| 1186 | } |
| 1187 | } |
| 1188 | goto found; |
| 1189 | } |
| 1190 | |
| 1191 | /* |
| 1192 | * No space in current array. Expand? |
| 1193 | */ |
| 1194 | if (fdp->fd_nfiles >= lim) { |
| 1195 | spin_unlock(&fdp->fd_spin); |
| 1196 | return (EMFILE); |
| 1197 | } |
| 1198 | fdgrow_locked(fdp, want); |
| 1199 | goto retry; |
| 1200 | |
| 1201 | found: |
| 1202 | KKASSERT(fd < fdp->fd_nfiles); |
| 1203 | if (fd > fdp->fd_lastfile) |
| 1204 | fdp->fd_lastfile = fd; |
| 1205 | if (want <= fdp->fd_freefile) |
| 1206 | fdp->fd_freefile = fd; |
| 1207 | *result = fd; |
| 1208 | KKASSERT(fdp->fd_files[fd].fp == NULL); |
| 1209 | KKASSERT(fdp->fd_files[fd].reserved == 0); |
| 1210 | fdp->fd_files[fd].fileflags = 0; |
| 1211 | fdp->fd_files[fd].reserved = 1; |
| 1212 | fdreserve_locked(fdp, fd, 1); |
| 1213 | spin_unlock(&fdp->fd_spin); |
| 1214 | return (0); |
| 1215 | } |
| 1216 | |
| 1217 | /* |
| 1218 | * Check to see whether n user file descriptors |
| 1219 | * are available to the process p. |
| 1220 | * |
| 1221 | * MPSAFE |
| 1222 | */ |
| 1223 | int |
| 1224 | fdavail(struct proc *p, int n) |
| 1225 | { |
| 1226 | struct filedesc *fdp = p->p_fd; |
| 1227 | struct fdnode *fdnode; |
| 1228 | int i, lim, last; |
| 1229 | |
| 1230 | spin_lock(&p->p_limit->p_spin); |
| 1231 | if (p->p_rlimit[RLIMIT_NOFILE].rlim_cur > INT_MAX) |
| 1232 | lim = INT_MAX; |
| 1233 | else |
| 1234 | lim = (int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur; |
| 1235 | spin_unlock(&p->p_limit->p_spin); |
| 1236 | |
| 1237 | if (lim > maxfilesperproc) |
| 1238 | lim = maxfilesperproc; |
| 1239 | if (lim < minfilesperproc) |
| 1240 | lim = minfilesperproc; |
| 1241 | |
| 1242 | spin_lock(&fdp->fd_spin); |
| 1243 | if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) { |
| 1244 | spin_unlock(&fdp->fd_spin); |
| 1245 | return (1); |
| 1246 | } |
| 1247 | last = min(fdp->fd_nfiles, lim); |
| 1248 | fdnode = &fdp->fd_files[fdp->fd_freefile]; |
| 1249 | for (i = last - fdp->fd_freefile; --i >= 0; ++fdnode) { |
| 1250 | if (fdnode->fp == NULL && --n <= 0) { |
| 1251 | spin_unlock(&fdp->fd_spin); |
| 1252 | return (1); |
| 1253 | } |
| 1254 | } |
| 1255 | spin_unlock(&fdp->fd_spin); |
| 1256 | return (0); |
| 1257 | } |
| 1258 | |
| 1259 | /* |
| 1260 | * Revoke open descriptors referencing (f_data, f_type) |
| 1261 | * |
| 1262 | * Any revoke executed within a prison is only able to |
| 1263 | * revoke descriptors for processes within that prison. |
| 1264 | * |
| 1265 | * Returns 0 on success or an error code. |
| 1266 | */ |
| 1267 | struct fdrevoke_info { |
| 1268 | void *data; |
| 1269 | short type; |
| 1270 | short unused; |
| 1271 | int count; |
| 1272 | int intransit; |
| 1273 | struct ucred *cred; |
| 1274 | struct file *nfp; |
| 1275 | }; |
| 1276 | |
| 1277 | static int fdrevoke_check_callback(struct file *fp, void *vinfo); |
| 1278 | static int fdrevoke_proc_callback(struct proc *p, void *vinfo); |
| 1279 | |
| 1280 | int |
| 1281 | fdrevoke(void *f_data, short f_type, struct ucred *cred) |
| 1282 | { |
| 1283 | struct fdrevoke_info info; |
| 1284 | int error; |
| 1285 | |
| 1286 | bzero(&info, sizeof(info)); |
| 1287 | info.data = f_data; |
| 1288 | info.type = f_type; |
| 1289 | info.cred = cred; |
| 1290 | error = falloc(NULL, &info.nfp, NULL); |
| 1291 | if (error) |
| 1292 | return (error); |
| 1293 | |
| 1294 | /* |
| 1295 | * Scan the file pointer table once. dups do not dup file pointers, |
| 1296 | * only descriptors, so there is no leak. Set FREVOKED on the fps |
| 1297 | * being revoked. |
| 1298 | */ |
| 1299 | allfiles_scan_exclusive(fdrevoke_check_callback, &info); |
| 1300 | |
| 1301 | /* |
| 1302 | * If any fps were marked track down the related descriptors |
| 1303 | * and close them. Any dup()s at this point will notice |
| 1304 | * the FREVOKED already set in the fp and do the right thing. |
| 1305 | * |
| 1306 | * Any fps with non-zero msgcounts (aka sent over a unix-domain |
| 1307 | * socket) bumped the intransit counter and will require a |
| 1308 | * scan. Races against fps leaving the socket are closed by |
| 1309 | * the socket code checking for FREVOKED. |
| 1310 | */ |
| 1311 | if (info.count) |
| 1312 | allproc_scan(fdrevoke_proc_callback, &info); |
| 1313 | if (info.intransit) |
| 1314 | unp_revoke_gc(info.nfp); |
| 1315 | fdrop(info.nfp); |
| 1316 | return(0); |
| 1317 | } |
| 1318 | |
| 1319 | /* |
| 1320 | * Locate matching file pointers directly. |
| 1321 | * |
| 1322 | * WARNING: allfiles_scan_exclusive() holds a spinlock through these calls! |
| 1323 | */ |
| 1324 | static int |
| 1325 | fdrevoke_check_callback(struct file *fp, void *vinfo) |
| 1326 | { |
| 1327 | struct fdrevoke_info *info = vinfo; |
| 1328 | |
| 1329 | /* |
| 1330 | * File pointers already flagged for revokation are skipped. |
| 1331 | */ |
| 1332 | if (fp->f_flag & FREVOKED) |
| 1333 | return(0); |
| 1334 | |
| 1335 | /* |
| 1336 | * If revoking from a prison file pointers created outside of |
| 1337 | * that prison, or file pointers without creds, cannot be revoked. |
| 1338 | */ |
| 1339 | if (info->cred->cr_prison && |
| 1340 | (fp->f_cred == NULL || |
| 1341 | info->cred->cr_prison != fp->f_cred->cr_prison)) { |
| 1342 | return(0); |
| 1343 | } |
| 1344 | |
| 1345 | /* |
| 1346 | * If the file pointer matches then mark it for revocation. The |
| 1347 | * flag is currently only used by unp_revoke_gc(). |
| 1348 | * |
| 1349 | * info->count is a heuristic and can race in a SMP environment. |
| 1350 | */ |
| 1351 | if (info->data == fp->f_data && info->type == fp->f_type) { |
| 1352 | atomic_set_int(&fp->f_flag, FREVOKED); |
| 1353 | info->count += fp->f_count; |
| 1354 | if (fp->f_msgcount) |
| 1355 | ++info->intransit; |
| 1356 | } |
| 1357 | return(0); |
| 1358 | } |
| 1359 | |
| 1360 | /* |
| 1361 | * Locate matching file pointers via process descriptor tables. |
| 1362 | */ |
| 1363 | static int |
| 1364 | fdrevoke_proc_callback(struct proc *p, void *vinfo) |
| 1365 | { |
| 1366 | struct fdrevoke_info *info = vinfo; |
| 1367 | struct filedesc *fdp; |
| 1368 | struct file *fp; |
| 1369 | int n; |
| 1370 | |
| 1371 | if (p->p_stat == SIDL || p->p_stat == SZOMB) |
| 1372 | return(0); |
| 1373 | if (info->cred->cr_prison && |
| 1374 | info->cred->cr_prison != p->p_ucred->cr_prison) { |
| 1375 | return(0); |
| 1376 | } |
| 1377 | |
| 1378 | /* |
| 1379 | * If the controlling terminal of the process matches the |
| 1380 | * vnode being revoked we clear the controlling terminal. |
| 1381 | * |
| 1382 | * The normal spec_close() may not catch this because it |
| 1383 | * uses curproc instead of p. |
| 1384 | */ |
| 1385 | if (p->p_session && info->type == DTYPE_VNODE && |
| 1386 | info->data == p->p_session->s_ttyvp) { |
| 1387 | p->p_session->s_ttyvp = NULL; |
| 1388 | vrele(info->data); |
| 1389 | } |
| 1390 | |
| 1391 | /* |
| 1392 | * Softref the fdp to prevent it from being destroyed |
| 1393 | */ |
| 1394 | spin_lock(&p->p_spin); |
| 1395 | if ((fdp = p->p_fd) == NULL) { |
| 1396 | spin_unlock(&p->p_spin); |
| 1397 | return(0); |
| 1398 | } |
| 1399 | atomic_add_int(&fdp->fd_softrefs, 1); |
| 1400 | spin_unlock(&p->p_spin); |
| 1401 | |
| 1402 | /* |
| 1403 | * Locate and close any matching file descriptors. |
| 1404 | */ |
| 1405 | spin_lock(&fdp->fd_spin); |
| 1406 | for (n = 0; n < fdp->fd_nfiles; ++n) { |
| 1407 | if ((fp = fdp->fd_files[n].fp) == NULL) |
| 1408 | continue; |
| 1409 | if (fp->f_flag & FREVOKED) { |
| 1410 | fhold(info->nfp); |
| 1411 | fdp->fd_files[n].fp = info->nfp; |
| 1412 | spin_unlock(&fdp->fd_spin); |
| 1413 | knote_fdclose(fp, fdp, n); /* XXX */ |
| 1414 | closef(fp, p); |
| 1415 | spin_lock(&fdp->fd_spin); |
| 1416 | --info->count; |
| 1417 | } |
| 1418 | } |
| 1419 | spin_unlock(&fdp->fd_spin); |
| 1420 | atomic_subtract_int(&fdp->fd_softrefs, 1); |
| 1421 | return(0); |
| 1422 | } |
| 1423 | |
| 1424 | /* |
| 1425 | * falloc: |
| 1426 | * Create a new open file structure and reserve a file decriptor |
| 1427 | * for the process that refers to it. |
| 1428 | * |
| 1429 | * Root creds are checked using lp, or assumed if lp is NULL. If |
| 1430 | * resultfd is non-NULL then lp must also be non-NULL. No file |
| 1431 | * descriptor is reserved (and no process context is needed) if |
| 1432 | * resultfd is NULL. |
| 1433 | * |
| 1434 | * A file pointer with a refcount of 1 is returned. Note that the |
| 1435 | * file pointer is NOT associated with the descriptor. If falloc |
| 1436 | * returns success, fsetfd() MUST be called to either associate the |
| 1437 | * file pointer or clear the reservation. |
| 1438 | * |
| 1439 | * MPSAFE |
| 1440 | */ |
| 1441 | int |
| 1442 | falloc(struct lwp *lp, struct file **resultfp, int *resultfd) |
| 1443 | { |
| 1444 | static struct timeval lastfail; |
| 1445 | static int curfail; |
| 1446 | struct file *fp; |
| 1447 | struct ucred *cred = lp ? lp->lwp_thread->td_ucred : proc0.p_ucred; |
| 1448 | int error; |
| 1449 | |
| 1450 | fp = NULL; |
| 1451 | |
| 1452 | /* |
| 1453 | * Handle filetable full issues and root overfill. |
| 1454 | */ |
| 1455 | if (nfiles >= maxfiles - maxfilesrootres && |
| 1456 | (cred->cr_ruid != 0 || nfiles >= maxfiles)) { |
| 1457 | if (ppsratecheck(&lastfail, &curfail, 1)) { |
| 1458 | kprintf("kern.maxfiles limit exceeded by uid %d, " |
| 1459 | "please see tuning(7).\n", |
| 1460 | cred->cr_ruid); |
| 1461 | } |
| 1462 | error = ENFILE; |
| 1463 | goto done; |
| 1464 | } |
| 1465 | |
| 1466 | /* |
| 1467 | * Allocate a new file descriptor. |
| 1468 | */ |
| 1469 | fp = kmalloc(sizeof(struct file), M_FILE, M_WAITOK | M_ZERO); |
| 1470 | spin_init(&fp->f_spin); |
| 1471 | SLIST_INIT(&fp->f_klist); |
| 1472 | fp->f_count = 1; |
| 1473 | fp->f_ops = &badfileops; |
| 1474 | fp->f_seqcount = 1; |
| 1475 | fsetcred(fp, cred); |
| 1476 | spin_lock(&filehead_spin); |
| 1477 | nfiles++; |
| 1478 | LIST_INSERT_HEAD(&filehead, fp, f_list); |
| 1479 | spin_unlock(&filehead_spin); |
| 1480 | if (resultfd) { |
| 1481 | if ((error = fdalloc(lp->lwp_proc, 0, resultfd)) != 0) { |
| 1482 | fdrop(fp); |
| 1483 | fp = NULL; |
| 1484 | } |
| 1485 | } else { |
| 1486 | error = 0; |
| 1487 | } |
| 1488 | done: |
| 1489 | *resultfp = fp; |
| 1490 | return (error); |
| 1491 | } |
| 1492 | |
| 1493 | /* |
| 1494 | * Check for races against a file descriptor by determining that the |
| 1495 | * file pointer is still associated with the specified file descriptor, |
| 1496 | * and a close is not currently in progress. |
| 1497 | * |
| 1498 | * MPSAFE |
| 1499 | */ |
| 1500 | int |
| 1501 | checkfdclosed(struct filedesc *fdp, int fd, struct file *fp) |
| 1502 | { |
| 1503 | int error; |
| 1504 | |
| 1505 | spin_lock(&fdp->fd_spin); |
| 1506 | if ((unsigned)fd >= fdp->fd_nfiles || fp != fdp->fd_files[fd].fp) |
| 1507 | error = EBADF; |
| 1508 | else |
| 1509 | error = 0; |
| 1510 | spin_unlock(&fdp->fd_spin); |
| 1511 | return (error); |
| 1512 | } |
| 1513 | |
| 1514 | /* |
| 1515 | * Associate a file pointer with a previously reserved file descriptor. |
| 1516 | * This function always succeeds. |
| 1517 | * |
| 1518 | * If fp is NULL, the file descriptor is returned to the pool. |
| 1519 | */ |
| 1520 | |
| 1521 | /* |
| 1522 | * MPSAFE (exclusive spinlock must be held on call) |
| 1523 | */ |
| 1524 | static void |
| 1525 | fsetfd_locked(struct filedesc *fdp, struct file *fp, int fd) |
| 1526 | { |
| 1527 | KKASSERT((unsigned)fd < fdp->fd_nfiles); |
| 1528 | KKASSERT(fdp->fd_files[fd].reserved != 0); |
| 1529 | if (fp) { |
| 1530 | fhold(fp); |
| 1531 | fdp->fd_files[fd].fp = fp; |
| 1532 | fdp->fd_files[fd].reserved = 0; |
| 1533 | } else { |
| 1534 | fdp->fd_files[fd].reserved = 0; |
| 1535 | fdreserve_locked(fdp, fd, -1); |
| 1536 | fdfixup_locked(fdp, fd); |
| 1537 | } |
| 1538 | } |
| 1539 | |
| 1540 | /* |
| 1541 | * MPSAFE |
| 1542 | */ |
| 1543 | void |
| 1544 | fsetfd(struct filedesc *fdp, struct file *fp, int fd) |
| 1545 | { |
| 1546 | spin_lock(&fdp->fd_spin); |
| 1547 | fsetfd_locked(fdp, fp, fd); |
| 1548 | spin_unlock(&fdp->fd_spin); |
| 1549 | } |
| 1550 | |
| 1551 | /* |
| 1552 | * MPSAFE (exclusive spinlock must be held on call) |
| 1553 | */ |
| 1554 | static |
| 1555 | struct file * |
| 1556 | funsetfd_locked(struct filedesc *fdp, int fd) |
| 1557 | { |
| 1558 | struct file *fp; |
| 1559 | |
| 1560 | if ((unsigned)fd >= fdp->fd_nfiles) |
| 1561 | return (NULL); |
| 1562 | if ((fp = fdp->fd_files[fd].fp) == NULL) |
| 1563 | return (NULL); |
| 1564 | fdp->fd_files[fd].fp = NULL; |
| 1565 | fdp->fd_files[fd].fileflags = 0; |
| 1566 | |
| 1567 | fdreserve_locked(fdp, fd, -1); |
| 1568 | fdfixup_locked(fdp, fd); |
| 1569 | return(fp); |
| 1570 | } |
| 1571 | |
| 1572 | /* |
| 1573 | * MPSAFE |
| 1574 | */ |
| 1575 | int |
| 1576 | fgetfdflags(struct filedesc *fdp, int fd, int *flagsp) |
| 1577 | { |
| 1578 | int error; |
| 1579 | |
| 1580 | spin_lock(&fdp->fd_spin); |
| 1581 | if (((u_int)fd) >= fdp->fd_nfiles) { |
| 1582 | error = EBADF; |
| 1583 | } else if (fdp->fd_files[fd].fp == NULL) { |
| 1584 | error = EBADF; |
| 1585 | } else { |
| 1586 | *flagsp = fdp->fd_files[fd].fileflags; |
| 1587 | error = 0; |
| 1588 | } |
| 1589 | spin_unlock(&fdp->fd_spin); |
| 1590 | return (error); |
| 1591 | } |
| 1592 | |
| 1593 | /* |
| 1594 | * MPSAFE |
| 1595 | */ |
| 1596 | int |
| 1597 | fsetfdflags(struct filedesc *fdp, int fd, int add_flags) |
| 1598 | { |
| 1599 | int error; |
| 1600 | |
| 1601 | spin_lock(&fdp->fd_spin); |
| 1602 | if (((u_int)fd) >= fdp->fd_nfiles) { |
| 1603 | error = EBADF; |
| 1604 | } else if (fdp->fd_files[fd].fp == NULL) { |
| 1605 | error = EBADF; |
| 1606 | } else { |
| 1607 | fdp->fd_files[fd].fileflags |= add_flags; |
| 1608 | error = 0; |
| 1609 | } |
| 1610 | spin_unlock(&fdp->fd_spin); |
| 1611 | return (error); |
| 1612 | } |
| 1613 | |
| 1614 | /* |
| 1615 | * MPSAFE |
| 1616 | */ |
| 1617 | int |
| 1618 | fclrfdflags(struct filedesc *fdp, int fd, int rem_flags) |
| 1619 | { |
| 1620 | int error; |
| 1621 | |
| 1622 | spin_lock(&fdp->fd_spin); |
| 1623 | if (((u_int)fd) >= fdp->fd_nfiles) { |
| 1624 | error = EBADF; |
| 1625 | } else if (fdp->fd_files[fd].fp == NULL) { |
| 1626 | error = EBADF; |
| 1627 | } else { |
| 1628 | fdp->fd_files[fd].fileflags &= ~rem_flags; |
| 1629 | error = 0; |
| 1630 | } |
| 1631 | spin_unlock(&fdp->fd_spin); |
| 1632 | return (error); |
| 1633 | } |
| 1634 | |
| 1635 | /* |
| 1636 | * Set/Change/Clear the creds for a fp and synchronize the uidinfo. |
| 1637 | */ |
| 1638 | void |
| 1639 | fsetcred(struct file *fp, struct ucred *ncr) |
| 1640 | { |
| 1641 | struct ucred *ocr; |
| 1642 | struct uidinfo *uip; |
| 1643 | |
| 1644 | ocr = fp->f_cred; |
| 1645 | if (ocr == NULL || ncr == NULL || ocr->cr_uidinfo != ncr->cr_uidinfo) { |
| 1646 | if (ocr) { |
| 1647 | uip = ocr->cr_uidinfo; |
| 1648 | atomic_add_int(&uip->ui_openfiles, -1); |
| 1649 | } |
| 1650 | if (ncr) { |
| 1651 | uip = ncr->cr_uidinfo; |
| 1652 | atomic_add_int(&uip->ui_openfiles, 1); |
| 1653 | } |
| 1654 | } |
| 1655 | if (ncr) |
| 1656 | crhold(ncr); |
| 1657 | fp->f_cred = ncr; |
| 1658 | if (ocr) |
| 1659 | crfree(ocr); |
| 1660 | } |
| 1661 | |
| 1662 | /* |
| 1663 | * Free a file descriptor. |
| 1664 | */ |
| 1665 | static |
| 1666 | void |
| 1667 | ffree(struct file *fp) |
| 1668 | { |
| 1669 | KASSERT((fp->f_count == 0), ("ffree: fp_fcount not 0!")); |
| 1670 | spin_lock(&filehead_spin); |
| 1671 | LIST_REMOVE(fp, f_list); |
| 1672 | nfiles--; |
| 1673 | spin_unlock(&filehead_spin); |
| 1674 | fsetcred(fp, NULL); |
| 1675 | if (fp->f_nchandle.ncp) |
| 1676 | cache_drop(&fp->f_nchandle); |
| 1677 | kfree(fp, M_FILE); |
| 1678 | } |
| 1679 | |
| 1680 | /* |
| 1681 | * called from init_main, initialize filedesc0 for proc0. |
| 1682 | */ |
| 1683 | void |
| 1684 | fdinit_bootstrap(struct proc *p0, struct filedesc *fdp0, int cmask) |
| 1685 | { |
| 1686 | p0->p_fd = fdp0; |
| 1687 | p0->p_fdtol = NULL; |
| 1688 | fdp0->fd_refcnt = 1; |
| 1689 | fdp0->fd_cmask = cmask; |
| 1690 | fdp0->fd_files = fdp0->fd_builtin_files; |
| 1691 | fdp0->fd_nfiles = NDFILE; |
| 1692 | fdp0->fd_lastfile = -1; |
| 1693 | spin_init(&fdp0->fd_spin); |
| 1694 | } |
| 1695 | |
| 1696 | /* |
| 1697 | * Build a new filedesc structure. |
| 1698 | * |
| 1699 | * NOT MPSAFE (vref) |
| 1700 | */ |
| 1701 | struct filedesc * |
| 1702 | fdinit(struct proc *p) |
| 1703 | { |
| 1704 | struct filedesc *newfdp; |
| 1705 | struct filedesc *fdp = p->p_fd; |
| 1706 | |
| 1707 | newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK|M_ZERO); |
| 1708 | spin_lock(&fdp->fd_spin); |
| 1709 | if (fdp->fd_cdir) { |
| 1710 | newfdp->fd_cdir = fdp->fd_cdir; |
| 1711 | vref(newfdp->fd_cdir); |
| 1712 | cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); |
| 1713 | } |
| 1714 | |
| 1715 | /* |
| 1716 | * rdir may not be set in e.g. proc0 or anything vm_fork'd off of |
| 1717 | * proc0, but should unconditionally exist in other processes. |
| 1718 | */ |
| 1719 | if (fdp->fd_rdir) { |
| 1720 | newfdp->fd_rdir = fdp->fd_rdir; |
| 1721 | vref(newfdp->fd_rdir); |
| 1722 | cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); |
| 1723 | } |
| 1724 | if (fdp->fd_jdir) { |
| 1725 | newfdp->fd_jdir = fdp->fd_jdir; |
| 1726 | vref(newfdp->fd_jdir); |
| 1727 | cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); |
| 1728 | } |
| 1729 | spin_unlock(&fdp->fd_spin); |
| 1730 | |
| 1731 | /* Create the file descriptor table. */ |
| 1732 | newfdp->fd_refcnt = 1; |
| 1733 | newfdp->fd_cmask = cmask; |
| 1734 | newfdp->fd_files = newfdp->fd_builtin_files; |
| 1735 | newfdp->fd_nfiles = NDFILE; |
| 1736 | newfdp->fd_lastfile = -1; |
| 1737 | spin_init(&newfdp->fd_spin); |
| 1738 | |
| 1739 | return (newfdp); |
| 1740 | } |
| 1741 | |
| 1742 | /* |
| 1743 | * Share a filedesc structure. |
| 1744 | * |
| 1745 | * MPSAFE |
| 1746 | */ |
| 1747 | struct filedesc * |
| 1748 | fdshare(struct proc *p) |
| 1749 | { |
| 1750 | struct filedesc *fdp; |
| 1751 | |
| 1752 | fdp = p->p_fd; |
| 1753 | spin_lock(&fdp->fd_spin); |
| 1754 | fdp->fd_refcnt++; |
| 1755 | spin_unlock(&fdp->fd_spin); |
| 1756 | return (fdp); |
| 1757 | } |
| 1758 | |
| 1759 | /* |
| 1760 | * Copy a filedesc structure. |
| 1761 | * |
| 1762 | * MPSAFE |
| 1763 | */ |
| 1764 | struct filedesc * |
| 1765 | fdcopy(struct proc *p) |
| 1766 | { |
| 1767 | struct filedesc *fdp = p->p_fd; |
| 1768 | struct filedesc *newfdp; |
| 1769 | struct fdnode *fdnode; |
| 1770 | int i; |
| 1771 | int ni; |
| 1772 | |
| 1773 | /* |
| 1774 | * Certain daemons might not have file descriptors. |
| 1775 | */ |
| 1776 | if (fdp == NULL) |
| 1777 | return (NULL); |
| 1778 | |
| 1779 | /* |
| 1780 | * Allocate the new filedesc and fd_files[] array. This can race |
| 1781 | * with operations by other threads on the fdp so we have to be |
| 1782 | * careful. |
| 1783 | */ |
| 1784 | newfdp = kmalloc(sizeof(struct filedesc), M_FILEDESC, M_WAITOK | M_ZERO); |
| 1785 | again: |
| 1786 | spin_lock(&fdp->fd_spin); |
| 1787 | if (fdp->fd_lastfile < NDFILE) { |
| 1788 | newfdp->fd_files = newfdp->fd_builtin_files; |
| 1789 | i = NDFILE; |
| 1790 | } else { |
| 1791 | /* |
| 1792 | * We have to allocate (N^2-1) entries for our in-place |
| 1793 | * binary tree. Allow the table to shrink. |
| 1794 | */ |
| 1795 | i = fdp->fd_nfiles; |
| 1796 | ni = (i - 1) / 2; |
| 1797 | while (ni > fdp->fd_lastfile && ni > NDFILE) { |
| 1798 | i = ni; |
| 1799 | ni = (i - 1) / 2; |
| 1800 | } |
| 1801 | spin_unlock(&fdp->fd_spin); |
| 1802 | newfdp->fd_files = kmalloc(i * sizeof(struct fdnode), |
| 1803 | M_FILEDESC, M_WAITOK | M_ZERO); |
| 1804 | |
| 1805 | /* |
| 1806 | * Check for race, retry |
| 1807 | */ |
| 1808 | spin_lock(&fdp->fd_spin); |
| 1809 | if (i <= fdp->fd_lastfile) { |
| 1810 | spin_unlock(&fdp->fd_spin); |
| 1811 | kfree(newfdp->fd_files, M_FILEDESC); |
| 1812 | goto again; |
| 1813 | } |
| 1814 | } |
| 1815 | |
| 1816 | /* |
| 1817 | * Dup the remaining fields. vref() and cache_hold() can be |
| 1818 | * safely called while holding the read spinlock on fdp. |
| 1819 | * |
| 1820 | * The read spinlock on fdp is still being held. |
| 1821 | * |
| 1822 | * NOTE: vref and cache_hold calls for the case where the vnode |
| 1823 | * or cache entry already has at least one ref may be called |
| 1824 | * while holding spin locks. |
| 1825 | */ |
| 1826 | if ((newfdp->fd_cdir = fdp->fd_cdir) != NULL) { |
| 1827 | vref(newfdp->fd_cdir); |
| 1828 | cache_copy(&fdp->fd_ncdir, &newfdp->fd_ncdir); |
| 1829 | } |
| 1830 | /* |
| 1831 | * We must check for fd_rdir here, at least for now because |
| 1832 | * the init process is created before we have access to the |
| 1833 | * rootvode to take a reference to it. |
| 1834 | */ |
| 1835 | if ((newfdp->fd_rdir = fdp->fd_rdir) != NULL) { |
| 1836 | vref(newfdp->fd_rdir); |
| 1837 | cache_copy(&fdp->fd_nrdir, &newfdp->fd_nrdir); |
| 1838 | } |
| 1839 | if ((newfdp->fd_jdir = fdp->fd_jdir) != NULL) { |
| 1840 | vref(newfdp->fd_jdir); |
| 1841 | cache_copy(&fdp->fd_njdir, &newfdp->fd_njdir); |
| 1842 | } |
| 1843 | newfdp->fd_refcnt = 1; |
| 1844 | newfdp->fd_nfiles = i; |
| 1845 | newfdp->fd_lastfile = fdp->fd_lastfile; |
| 1846 | newfdp->fd_freefile = fdp->fd_freefile; |
| 1847 | newfdp->fd_cmask = fdp->fd_cmask; |
| 1848 | spin_init(&newfdp->fd_spin); |
| 1849 | |
| 1850 | /* |
| 1851 | * Copy the descriptor table through (i). This also copies the |
| 1852 | * allocation state. Then go through and ref the file pointers |
| 1853 | * and clean up any KQ descriptors. |
| 1854 | * |
| 1855 | * kq descriptors cannot be copied. Since we haven't ref'd the |
| 1856 | * copied files yet we can ignore the return value from funsetfd(). |
| 1857 | * |
| 1858 | * The read spinlock on fdp is still being held. |
| 1859 | */ |
| 1860 | bcopy(fdp->fd_files, newfdp->fd_files, i * sizeof(struct fdnode)); |
| 1861 | for (i = 0 ; i < newfdp->fd_nfiles; ++i) { |
| 1862 | fdnode = &newfdp->fd_files[i]; |
| 1863 | if (fdnode->reserved) { |
| 1864 | fdreserve_locked(newfdp, i, -1); |
| 1865 | fdnode->reserved = 0; |
| 1866 | fdfixup_locked(newfdp, i); |
| 1867 | } else if (fdnode->fp) { |
| 1868 | if (fdnode->fp->f_type == DTYPE_KQUEUE) { |
| 1869 | (void)funsetfd_locked(newfdp, i); |
| 1870 | } else { |
| 1871 | fhold(fdnode->fp); |
| 1872 | } |
| 1873 | } |
| 1874 | } |
| 1875 | spin_unlock(&fdp->fd_spin); |
| 1876 | return (newfdp); |
| 1877 | } |
| 1878 | |
| 1879 | /* |
| 1880 | * Release a filedesc structure. |
| 1881 | * |
| 1882 | * NOT MPSAFE (MPSAFE for refs > 1, but the final cleanup code is not MPSAFE) |
| 1883 | */ |
| 1884 | void |
| 1885 | fdfree(struct proc *p, struct filedesc *repl) |
| 1886 | { |
| 1887 | struct filedesc *fdp; |
| 1888 | struct fdnode *fdnode; |
| 1889 | int i; |
| 1890 | struct filedesc_to_leader *fdtol; |
| 1891 | struct file *fp; |
| 1892 | struct vnode *vp; |
| 1893 | struct flock lf; |
| 1894 | |
| 1895 | /* |
| 1896 | * Certain daemons might not have file descriptors. |
| 1897 | */ |
| 1898 | fdp = p->p_fd; |
| 1899 | if (fdp == NULL) { |
| 1900 | p->p_fd = repl; |
| 1901 | return; |
| 1902 | } |
| 1903 | |
| 1904 | /* |
| 1905 | * Severe messing around to follow. |
| 1906 | */ |
| 1907 | spin_lock(&fdp->fd_spin); |
| 1908 | |
| 1909 | /* Check for special need to clear POSIX style locks */ |
| 1910 | fdtol = p->p_fdtol; |
| 1911 | if (fdtol != NULL) { |
| 1912 | KASSERT(fdtol->fdl_refcount > 0, |
| 1913 | ("filedesc_to_refcount botch: fdl_refcount=%d", |
| 1914 | fdtol->fdl_refcount)); |
| 1915 | if (fdtol->fdl_refcount == 1 && |
| 1916 | (p->p_leader->p_flag & P_ADVLOCK) != 0) { |
| 1917 | for (i = 0; i <= fdp->fd_lastfile; ++i) { |
| 1918 | fdnode = &fdp->fd_files[i]; |
| 1919 | if (fdnode->fp == NULL || |
| 1920 | fdnode->fp->f_type != DTYPE_VNODE) { |
| 1921 | continue; |
| 1922 | } |
| 1923 | fp = fdnode->fp; |
| 1924 | fhold(fp); |
| 1925 | spin_unlock(&fdp->fd_spin); |
| 1926 | |
| 1927 | lf.l_whence = SEEK_SET; |
| 1928 | lf.l_start = 0; |
| 1929 | lf.l_len = 0; |
| 1930 | lf.l_type = F_UNLCK; |
| 1931 | vp = (struct vnode *)fp->f_data; |
| 1932 | (void) VOP_ADVLOCK(vp, |
| 1933 | (caddr_t)p->p_leader, |
| 1934 | F_UNLCK, |
| 1935 | &lf, |
| 1936 | F_POSIX); |
| 1937 | fdrop(fp); |
| 1938 | spin_lock(&fdp->fd_spin); |
| 1939 | } |
| 1940 | } |
| 1941 | retry: |
| 1942 | if (fdtol->fdl_refcount == 1) { |
| 1943 | if (fdp->fd_holdleaderscount > 0 && |
| 1944 | (p->p_leader->p_flag & P_ADVLOCK) != 0) { |
| 1945 | /* |
| 1946 | * close() or do_dup() has cleared a reference |
| 1947 | * in a shared file descriptor table. |
| 1948 | */ |
| 1949 | fdp->fd_holdleaderswakeup = 1; |
| 1950 | ssleep(&fdp->fd_holdleaderscount, |
| 1951 | &fdp->fd_spin, 0, "fdlhold", 0); |
| 1952 | goto retry; |
| 1953 | } |
| 1954 | if (fdtol->fdl_holdcount > 0) { |
| 1955 | /* |
| 1956 | * Ensure that fdtol->fdl_leader |
| 1957 | * remains valid in closef(). |
| 1958 | */ |
| 1959 | fdtol->fdl_wakeup = 1; |
| 1960 | ssleep(fdtol, &fdp->fd_spin, 0, "fdlhold", 0); |
| 1961 | goto retry; |
| 1962 | } |
| 1963 | } |
| 1964 | fdtol->fdl_refcount--; |
| 1965 | if (fdtol->fdl_refcount == 0 && |
| 1966 | fdtol->fdl_holdcount == 0) { |
| 1967 | fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; |
| 1968 | fdtol->fdl_prev->fdl_next = fdtol->fdl_next; |
| 1969 | } else { |
| 1970 | fdtol = NULL; |
| 1971 | } |
| 1972 | p->p_fdtol = NULL; |
| 1973 | if (fdtol != NULL) { |
| 1974 | spin_unlock(&fdp->fd_spin); |
| 1975 | kfree(fdtol, M_FILEDESC_TO_LEADER); |
| 1976 | spin_lock(&fdp->fd_spin); |
| 1977 | } |
| 1978 | } |
| 1979 | if (--fdp->fd_refcnt > 0) { |
| 1980 | spin_unlock(&fdp->fd_spin); |
| 1981 | spin_lock(&p->p_spin); |
| 1982 | p->p_fd = repl; |
| 1983 | spin_unlock(&p->p_spin); |
| 1984 | return; |
| 1985 | } |
| 1986 | |
| 1987 | /* |
| 1988 | * Even though we are the last reference to the structure allproc |
| 1989 | * scans may still reference the structure. Maintain proper |
| 1990 | * locks until we can replace p->p_fd. |
| 1991 | * |
| 1992 | * Also note that kqueue's closef still needs to reference the |
| 1993 | * fdp via p->p_fd, so we have to close the descriptors before |
| 1994 | * we replace p->p_fd. |
| 1995 | */ |
| 1996 | for (i = 0; i <= fdp->fd_lastfile; ++i) { |
| 1997 | if (fdp->fd_files[i].fp) { |
| 1998 | fp = funsetfd_locked(fdp, i); |
| 1999 | if (fp) { |
| 2000 | spin_unlock(&fdp->fd_spin); |
| 2001 | if (SLIST_FIRST(&fp->f_klist)) |
| 2002 | knote_fdclose(fp, fdp, i); |
| 2003 | closef(fp, p); |
| 2004 | spin_lock(&fdp->fd_spin); |
| 2005 | } |
| 2006 | } |
| 2007 | } |
| 2008 | spin_unlock(&fdp->fd_spin); |
| 2009 | |
| 2010 | /* |
| 2011 | * Interlock against an allproc scan operations (typically frevoke). |
| 2012 | */ |
| 2013 | spin_lock(&p->p_spin); |
| 2014 | p->p_fd = repl; |
| 2015 | spin_unlock(&p->p_spin); |
| 2016 | |
| 2017 | /* |
| 2018 | * Wait for any softrefs to go away. This race rarely occurs so |
| 2019 | * we can use a non-critical-path style poll/sleep loop. The |
| 2020 | * race only occurs against allproc scans. |
| 2021 | * |
| 2022 | * No new softrefs can occur with the fdp disconnected from the |
| 2023 | * process. |
| 2024 | */ |
| 2025 | if (fdp->fd_softrefs) { |
| 2026 | kprintf("pid %d: Warning, fdp race avoided\n", p->p_pid); |
| 2027 | while (fdp->fd_softrefs) |
| 2028 | tsleep(&fdp->fd_softrefs, 0, "fdsoft", 1); |
| 2029 | } |
| 2030 | |
| 2031 | if (fdp->fd_files != fdp->fd_builtin_files) |
| 2032 | kfree(fdp->fd_files, M_FILEDESC); |
| 2033 | if (fdp->fd_cdir) { |
| 2034 | cache_drop(&fdp->fd_ncdir); |
| 2035 | vrele(fdp->fd_cdir); |
| 2036 | } |
| 2037 | if (fdp->fd_rdir) { |
| 2038 | cache_drop(&fdp->fd_nrdir); |
| 2039 | vrele(fdp->fd_rdir); |
| 2040 | } |
| 2041 | if (fdp->fd_jdir) { |
| 2042 | cache_drop(&fdp->fd_njdir); |
| 2043 | vrele(fdp->fd_jdir); |
| 2044 | } |
| 2045 | kfree(fdp, M_FILEDESC); |
| 2046 | } |
| 2047 | |
| 2048 | /* |
| 2049 | * Retrieve and reference the file pointer associated with a descriptor. |
| 2050 | * |
| 2051 | * MPSAFE |
| 2052 | */ |
| 2053 | struct file * |
| 2054 | holdfp(struct filedesc *fdp, int fd, int flag) |
| 2055 | { |
| 2056 | struct file* fp; |
| 2057 | |
| 2058 | spin_lock(&fdp->fd_spin); |
| 2059 | if (((u_int)fd) >= fdp->fd_nfiles) { |
| 2060 | fp = NULL; |
| 2061 | goto done; |
| 2062 | } |
| 2063 | if ((fp = fdp->fd_files[fd].fp) == NULL) |
| 2064 | goto done; |
| 2065 | if ((fp->f_flag & flag) == 0 && flag != -1) { |
| 2066 | fp = NULL; |
| 2067 | goto done; |
| 2068 | } |
| 2069 | fhold(fp); |
| 2070 | done: |
| 2071 | spin_unlock(&fdp->fd_spin); |
| 2072 | return (fp); |
| 2073 | } |
| 2074 | |
| 2075 | /* |
| 2076 | * holdsock() - load the struct file pointer associated |
| 2077 | * with a socket into *fpp. If an error occurs, non-zero |
| 2078 | * will be returned and *fpp will be set to NULL. |
| 2079 | * |
| 2080 | * MPSAFE |
| 2081 | */ |
| 2082 | int |
| 2083 | holdsock(struct filedesc *fdp, int fd, struct file **fpp) |
| 2084 | { |
| 2085 | struct file *fp; |
| 2086 | int error; |
| 2087 | |
| 2088 | spin_lock(&fdp->fd_spin); |
| 2089 | if ((unsigned)fd >= fdp->fd_nfiles) { |
| 2090 | error = EBADF; |
| 2091 | fp = NULL; |
| 2092 | goto done; |
| 2093 | } |
| 2094 | if ((fp = fdp->fd_files[fd].fp) == NULL) { |
| 2095 | error = EBADF; |
| 2096 | goto done; |
| 2097 | } |
| 2098 | if (fp->f_type != DTYPE_SOCKET) { |
| 2099 | error = ENOTSOCK; |
| 2100 | goto done; |
| 2101 | } |
| 2102 | fhold(fp); |
| 2103 | error = 0; |
| 2104 | done: |
| 2105 | spin_unlock(&fdp->fd_spin); |
| 2106 | *fpp = fp; |
| 2107 | return (error); |
| 2108 | } |
| 2109 | |
| 2110 | /* |
| 2111 | * Convert a user file descriptor to a held file pointer. |
| 2112 | * |
| 2113 | * MPSAFE |
| 2114 | */ |
| 2115 | int |
| 2116 | holdvnode(struct filedesc *fdp, int fd, struct file **fpp) |
| 2117 | { |
| 2118 | struct file *fp; |
| 2119 | int error; |
| 2120 | |
| 2121 | spin_lock(&fdp->fd_spin); |
| 2122 | if ((unsigned)fd >= fdp->fd_nfiles) { |
| 2123 | error = EBADF; |
| 2124 | fp = NULL; |
| 2125 | goto done; |
| 2126 | } |
| 2127 | if ((fp = fdp->fd_files[fd].fp) == NULL) { |
| 2128 | error = EBADF; |
| 2129 | goto done; |
| 2130 | } |
| 2131 | if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) { |
| 2132 | fp = NULL; |
| 2133 | error = EINVAL; |
| 2134 | goto done; |
| 2135 | } |
| 2136 | fhold(fp); |
| 2137 | error = 0; |
| 2138 | done: |
| 2139 | spin_unlock(&fdp->fd_spin); |
| 2140 | *fpp = fp; |
| 2141 | return (error); |
| 2142 | } |
| 2143 | |
| 2144 | /* |
| 2145 | * For setugid programs, we don't want to people to use that setugidness |
| 2146 | * to generate error messages which write to a file which otherwise would |
| 2147 | * otherwise be off-limits to the process. |
| 2148 | * |
| 2149 | * This is a gross hack to plug the hole. A better solution would involve |
| 2150 | * a special vop or other form of generalized access control mechanism. We |
| 2151 | * go ahead and just reject all procfs file systems accesses as dangerous. |
| 2152 | * |
| 2153 | * Since setugidsafety calls this only for fd 0, 1 and 2, this check is |
| 2154 | * sufficient. We also don't for check setugidness since we know we are. |
| 2155 | */ |
| 2156 | static int |
| 2157 | is_unsafe(struct file *fp) |
| 2158 | { |
| 2159 | if (fp->f_type == DTYPE_VNODE && |
| 2160 | ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS) |
| 2161 | return (1); |
| 2162 | return (0); |
| 2163 | } |
| 2164 | |
| 2165 | /* |
| 2166 | * Make this setguid thing safe, if at all possible. |
| 2167 | * |
| 2168 | * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() |
| 2169 | */ |
| 2170 | void |
| 2171 | setugidsafety(struct proc *p) |
| 2172 | { |
| 2173 | struct filedesc *fdp = p->p_fd; |
| 2174 | int i; |
| 2175 | |
| 2176 | /* Certain daemons might not have file descriptors. */ |
| 2177 | if (fdp == NULL) |
| 2178 | return; |
| 2179 | |
| 2180 | /* |
| 2181 | * note: fdp->fd_files may be reallocated out from under us while |
| 2182 | * we are blocked in a close. Be careful! |
| 2183 | */ |
| 2184 | for (i = 0; i <= fdp->fd_lastfile; i++) { |
| 2185 | if (i > 2) |
| 2186 | break; |
| 2187 | if (fdp->fd_files[i].fp && is_unsafe(fdp->fd_files[i].fp)) { |
| 2188 | struct file *fp; |
| 2189 | |
| 2190 | /* |
| 2191 | * NULL-out descriptor prior to close to avoid |
| 2192 | * a race while close blocks. |
| 2193 | */ |
| 2194 | if ((fp = funsetfd_locked(fdp, i)) != NULL) { |
| 2195 | knote_fdclose(fp, fdp, i); |
| 2196 | closef(fp, p); |
| 2197 | } |
| 2198 | } |
| 2199 | } |
| 2200 | } |
| 2201 | |
| 2202 | /* |
| 2203 | * Close any files on exec? |
| 2204 | * |
| 2205 | * NOT MPSAFE - scans fdp without spinlocks, calls knote_fdclose() |
| 2206 | */ |
| 2207 | void |
| 2208 | fdcloseexec(struct proc *p) |
| 2209 | { |
| 2210 | struct filedesc *fdp = p->p_fd; |
| 2211 | int i; |
| 2212 | |
| 2213 | /* Certain daemons might not have file descriptors. */ |
| 2214 | if (fdp == NULL) |
| 2215 | return; |
| 2216 | |
| 2217 | /* |
| 2218 | * We cannot cache fd_files since operations may block and rip |
| 2219 | * them out from under us. |
| 2220 | */ |
| 2221 | for (i = 0; i <= fdp->fd_lastfile; i++) { |
| 2222 | if (fdp->fd_files[i].fp != NULL && |
| 2223 | (fdp->fd_files[i].fileflags & UF_EXCLOSE)) { |
| 2224 | struct file *fp; |
| 2225 | |
| 2226 | /* |
| 2227 | * NULL-out descriptor prior to close to avoid |
| 2228 | * a race while close blocks. |
| 2229 | */ |
| 2230 | if ((fp = funsetfd_locked(fdp, i)) != NULL) { |
| 2231 | knote_fdclose(fp, fdp, i); |
| 2232 | closef(fp, p); |
| 2233 | } |
| 2234 | } |
| 2235 | } |
| 2236 | } |
| 2237 | |
| 2238 | /* |
| 2239 | * It is unsafe for set[ug]id processes to be started with file |
| 2240 | * descriptors 0..2 closed, as these descriptors are given implicit |
| 2241 | * significance in the Standard C library. fdcheckstd() will create a |
| 2242 | * descriptor referencing /dev/null for each of stdin, stdout, and |
| 2243 | * stderr that is not already open. |
| 2244 | * |
| 2245 | * NOT MPSAFE - calls falloc, vn_open, etc |
| 2246 | */ |
| 2247 | int |
| 2248 | fdcheckstd(struct lwp *lp) |
| 2249 | { |
| 2250 | struct nlookupdata nd; |
| 2251 | struct filedesc *fdp; |
| 2252 | struct file *fp; |
| 2253 | int retval; |
| 2254 | int i, error, flags, devnull; |
| 2255 | |
| 2256 | fdp = lp->lwp_proc->p_fd; |
| 2257 | if (fdp == NULL) |
| 2258 | return (0); |
| 2259 | devnull = -1; |
| 2260 | error = 0; |
| 2261 | for (i = 0; i < 3; i++) { |
| 2262 | if (fdp->fd_files[i].fp != NULL) |
| 2263 | continue; |
| 2264 | if (devnull < 0) { |
| 2265 | if ((error = falloc(lp, &fp, &devnull)) != 0) |
| 2266 | break; |
| 2267 | |
| 2268 | error = nlookup_init(&nd, "/dev/null", UIO_SYSSPACE, |
| 2269 | NLC_FOLLOW|NLC_LOCKVP); |
| 2270 | flags = FREAD | FWRITE; |
| 2271 | if (error == 0) |
| 2272 | error = vn_open(&nd, fp, flags, 0); |
| 2273 | if (error == 0) |
| 2274 | fsetfd(fdp, fp, devnull); |
| 2275 | else |
| 2276 | fsetfd(fdp, NULL, devnull); |
| 2277 | fdrop(fp); |
| 2278 | nlookup_done(&nd); |
| 2279 | if (error) |
| 2280 | break; |
| 2281 | KKASSERT(i == devnull); |
| 2282 | } else { |
| 2283 | error = kern_dup(DUP_FIXED, devnull, i, &retval); |
| 2284 | if (error != 0) |
| 2285 | break; |
| 2286 | } |
| 2287 | } |
| 2288 | return (error); |
| 2289 | } |
| 2290 | |
| 2291 | /* |
| 2292 | * Internal form of close. |
| 2293 | * Decrement reference count on file structure. |
| 2294 | * Note: td and/or p may be NULL when closing a file |
| 2295 | * that was being passed in a message. |
| 2296 | * |
| 2297 | * MPALMOSTSAFE - acquires mplock for VOP operations |
| 2298 | */ |
| 2299 | int |
| 2300 | closef(struct file *fp, struct proc *p) |
| 2301 | { |
| 2302 | struct vnode *vp; |
| 2303 | struct flock lf; |
| 2304 | struct filedesc_to_leader *fdtol; |
| 2305 | |
| 2306 | if (fp == NULL) |
| 2307 | return (0); |
| 2308 | |
| 2309 | /* |
| 2310 | * POSIX record locking dictates that any close releases ALL |
| 2311 | * locks owned by this process. This is handled by setting |
| 2312 | * a flag in the unlock to free ONLY locks obeying POSIX |
| 2313 | * semantics, and not to free BSD-style file locks. |
| 2314 | * If the descriptor was in a message, POSIX-style locks |
| 2315 | * aren't passed with the descriptor. |
| 2316 | */ |
| 2317 | if (p != NULL && fp->f_type == DTYPE_VNODE && |
| 2318 | (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) |
| 2319 | ) { |
| 2320 | get_mplock(); |
| 2321 | if ((p->p_leader->p_flag & P_ADVLOCK) != 0) { |
| 2322 | lf.l_whence = SEEK_SET; |
| 2323 | lf.l_start = 0; |
| 2324 | lf.l_len = 0; |
| 2325 | lf.l_type = F_UNLCK; |
| 2326 | vp = (struct vnode *)fp->f_data; |
| 2327 | (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, |
| 2328 | &lf, F_POSIX); |
| 2329 | } |
| 2330 | fdtol = p->p_fdtol; |
| 2331 | if (fdtol != NULL) { |
| 2332 | /* |
| 2333 | * Handle special case where file descriptor table |
| 2334 | * is shared between multiple process leaders. |
| 2335 | */ |
| 2336 | for (fdtol = fdtol->fdl_next; |
| 2337 | fdtol != p->p_fdtol; |
| 2338 | fdtol = fdtol->fdl_next) { |
| 2339 | if ((fdtol->fdl_leader->p_flag & |
| 2340 | P_ADVLOCK) == 0) |
| 2341 | continue; |
| 2342 | fdtol->fdl_holdcount++; |
| 2343 | lf.l_whence = SEEK_SET; |
| 2344 | lf.l_start = 0; |
| 2345 | lf.l_len = 0; |
| 2346 | lf.l_type = F_UNLCK; |
| 2347 | vp = (struct vnode *)fp->f_data; |
| 2348 | (void) VOP_ADVLOCK(vp, |
| 2349 | (caddr_t)fdtol->fdl_leader, |
| 2350 | F_UNLCK, &lf, F_POSIX); |
| 2351 | fdtol->fdl_holdcount--; |
| 2352 | if (fdtol->fdl_holdcount == 0 && |
| 2353 | fdtol->fdl_wakeup != 0) { |
| 2354 | fdtol->fdl_wakeup = 0; |
| 2355 | wakeup(fdtol); |
| 2356 | } |
| 2357 | } |
| 2358 | } |
| 2359 | rel_mplock(); |
| 2360 | } |
| 2361 | return (fdrop(fp)); |
| 2362 | } |
| 2363 | |
| 2364 | /* |
| 2365 | * MPSAFE |
| 2366 | * |
| 2367 | * fhold() can only be called if f_count is already at least 1 (i.e. the |
| 2368 | * caller of fhold() already has a reference to the file pointer in some |
| 2369 | * manner or other). |
| 2370 | * |
| 2371 | * f_count is not spin-locked. Instead, atomic ops are used for |
| 2372 | * incrementing, decrementing, and handling the 1->0 transition. |
| 2373 | */ |
| 2374 | void |
| 2375 | fhold(struct file *fp) |
| 2376 | { |
| 2377 | atomic_add_int(&fp->f_count, 1); |
| 2378 | } |
| 2379 | |
| 2380 | /* |
| 2381 | * fdrop() - drop a reference to a descriptor |
| 2382 | * |
| 2383 | * MPALMOSTSAFE - acquires mplock for final close sequence |
| 2384 | */ |
| 2385 | int |
| 2386 | fdrop(struct file *fp) |
| 2387 | { |
| 2388 | struct flock lf; |
| 2389 | struct vnode *vp; |
| 2390 | int error; |
| 2391 | |
| 2392 | /* |
| 2393 | * A combined fetch and subtract is needed to properly detect |
| 2394 | * 1->0 transitions, otherwise two cpus dropping from a ref |
| 2395 | * count of 2 might both try to run the 1->0 code. |
| 2396 | */ |
| 2397 | if (atomic_fetchadd_int(&fp->f_count, -1) > 1) |
| 2398 | return (0); |
| 2399 | |
| 2400 | KKASSERT(SLIST_FIRST(&fp->f_klist) == NULL); |
| 2401 | get_mplock(); |
| 2402 | |
| 2403 | /* |
| 2404 | * The last reference has gone away, we own the fp structure free |
| 2405 | * and clear. |
| 2406 | */ |
| 2407 | if (fp->f_count < 0) |
| 2408 | panic("fdrop: count < 0"); |
| 2409 | if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE && |
| 2410 | (((struct vnode *)fp->f_data)->v_flag & VMAYHAVELOCKS) |
| 2411 | ) { |
| 2412 | lf.l_whence = SEEK_SET; |
| 2413 | lf.l_start = 0; |
| 2414 | lf.l_len = 0; |
| 2415 | lf.l_type = F_UNLCK; |
| 2416 | vp = (struct vnode *)fp->f_data; |
| 2417 | (void) VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); |
| 2418 | } |
| 2419 | if (fp->f_ops != &badfileops) |
| 2420 | error = fo_close(fp); |
| 2421 | else |
| 2422 | error = 0; |
| 2423 | ffree(fp); |
| 2424 | rel_mplock(); |
| 2425 | return (error); |
| 2426 | } |
| 2427 | |
| 2428 | /* |
| 2429 | * Apply an advisory lock on a file descriptor. |
| 2430 | * |
| 2431 | * Just attempt to get a record lock of the requested type on |
| 2432 | * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0). |
| 2433 | * |
| 2434 | * MPALMOSTSAFE |
| 2435 | */ |
| 2436 | int |
| 2437 | sys_flock(struct flock_args *uap) |
| 2438 | { |
| 2439 | struct proc *p = curproc; |
| 2440 | struct file *fp; |
| 2441 | struct vnode *vp; |
| 2442 | struct flock lf; |
| 2443 | int error; |
| 2444 | |
| 2445 | if ((fp = holdfp(p->p_fd, uap->fd, -1)) == NULL) |
| 2446 | return (EBADF); |
| 2447 | get_mplock(); |
| 2448 | if (fp->f_type != DTYPE_VNODE) { |
| 2449 | error = EOPNOTSUPP; |
| 2450 | goto done; |
| 2451 | } |
| 2452 | vp = (struct vnode *)fp->f_data; |
| 2453 | lf.l_whence = SEEK_SET; |
| 2454 | lf.l_start = 0; |
| 2455 | lf.l_len = 0; |
| 2456 | if (uap->how & LOCK_UN) { |
| 2457 | lf.l_type = F_UNLCK; |
| 2458 | fp->f_flag &= ~FHASLOCK; |
| 2459 | error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, 0); |
| 2460 | goto done; |
| 2461 | } |
| 2462 | if (uap->how & LOCK_EX) |
| 2463 | lf.l_type = F_WRLCK; |
| 2464 | else if (uap->how & LOCK_SH) |
| 2465 | lf.l_type = F_RDLCK; |
| 2466 | else { |
| 2467 | error = EBADF; |
| 2468 | goto done; |
| 2469 | } |
| 2470 | fp->f_flag |= FHASLOCK; |
| 2471 | if (uap->how & LOCK_NB) |
| 2472 | error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, 0); |
| 2473 | else |
| 2474 | error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, F_WAIT); |
| 2475 | done: |
| 2476 | rel_mplock(); |
| 2477 | fdrop(fp); |
| 2478 | return (error); |
| 2479 | } |
| 2480 | |
| 2481 | /* |
| 2482 | * File Descriptor pseudo-device driver (/dev/fd/). |
| 2483 | * |
| 2484 | * Opening minor device N dup()s the file (if any) connected to file |
| 2485 | * descriptor N belonging to the calling process. Note that this driver |
| 2486 | * consists of only the ``open()'' routine, because all subsequent |
| 2487 | * references to this file will be direct to the other driver. |
| 2488 | */ |
| 2489 | static int |
| 2490 | fdopen(struct dev_open_args *ap) |
| 2491 | { |
| 2492 | thread_t td = curthread; |
| 2493 | |
| 2494 | KKASSERT(td->td_lwp != NULL); |
| 2495 | |
| 2496 | /* |
| 2497 | * XXX Kludge: set curlwp->lwp_dupfd to contain the value of the |
| 2498 | * the file descriptor being sought for duplication. The error |
| 2499 | * return ensures that the vnode for this device will be released |
| 2500 | * by vn_open. Open will detect this special error and take the |
| 2501 | * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN |
| 2502 | * will simply report the error. |
| 2503 | */ |
| 2504 | td->td_lwp->lwp_dupfd = minor(ap->a_head.a_dev); |
| 2505 | return (ENODEV); |
| 2506 | } |
| 2507 | |
| 2508 | /* |
| 2509 | * The caller has reserved the file descriptor dfd for us. On success we |
| 2510 | * must fsetfd() it. On failure the caller will clean it up. |
| 2511 | * |
| 2512 | * MPSAFE |
| 2513 | */ |
| 2514 | int |
| 2515 | dupfdopen(struct filedesc *fdp, int dfd, int sfd, int mode, int error) |
| 2516 | { |
| 2517 | struct file *wfp; |
| 2518 | struct file *xfp; |
| 2519 | int werror; |
| 2520 | |
| 2521 | if ((wfp = holdfp(fdp, sfd, -1)) == NULL) |
| 2522 | return (EBADF); |
| 2523 | |
| 2524 | /* |
| 2525 | * Close a revoke/dup race. Duping a descriptor marked as revoked |
| 2526 | * will dup a dummy descriptor instead of the real one. |
| 2527 | */ |
| 2528 | if (wfp->f_flag & FREVOKED) { |
| 2529 | kprintf("Warning: attempt to dup() a revoked descriptor\n"); |
| 2530 | fdrop(wfp); |
| 2531 | wfp = NULL; |
| 2532 | werror = falloc(NULL, &wfp, NULL); |
| 2533 | if (werror) |
| 2534 | return (werror); |
| 2535 | } |
| 2536 | |
| 2537 | /* |
| 2538 | * There are two cases of interest here. |
| 2539 | * |
| 2540 | * For ENODEV simply dup sfd to file descriptor dfd and return. |
| 2541 | * |
| 2542 | * For ENXIO steal away the file structure from sfd and store it |
| 2543 | * dfd. sfd is effectively closed by this operation. |
| 2544 | * |
| 2545 | * Any other error code is just returned. |
| 2546 | */ |
| 2547 | switch (error) { |
| 2548 | case ENODEV: |
| 2549 | /* |
| 2550 | * Check that the mode the file is being opened for is a |
| 2551 | * subset of the mode of the existing descriptor. |
| 2552 | */ |
| 2553 | if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) { |
| 2554 | error = EACCES; |
| 2555 | break; |
| 2556 | } |
| 2557 | spin_lock(&fdp->fd_spin); |
| 2558 | fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; |
| 2559 | fsetfd_locked(fdp, wfp, dfd); |
| 2560 | spin_unlock(&fdp->fd_spin); |
| 2561 | error = 0; |
| 2562 | break; |
| 2563 | case ENXIO: |
| 2564 | /* |
| 2565 | * Steal away the file pointer from dfd, and stuff it into indx. |
| 2566 | */ |
| 2567 | spin_lock(&fdp->fd_spin); |
| 2568 | fdp->fd_files[dfd].fileflags = fdp->fd_files[sfd].fileflags; |
| 2569 | fsetfd(fdp, wfp, dfd); |
| 2570 | if ((xfp = funsetfd_locked(fdp, sfd)) != NULL) { |
| 2571 | spin_unlock(&fdp->fd_spin); |
| 2572 | fdrop(xfp); |
| 2573 | } else { |
| 2574 | spin_unlock(&fdp->fd_spin); |
| 2575 | } |
| 2576 | error = 0; |
| 2577 | break; |
| 2578 | default: |
| 2579 | break; |
| 2580 | } |
| 2581 | fdrop(wfp); |
| 2582 | return (error); |
| 2583 | } |
| 2584 | |
| 2585 | /* |
| 2586 | * NOT MPSAFE - I think these refer to a common file descriptor table |
| 2587 | * and we need to spinlock that to link fdtol in. |
| 2588 | */ |
| 2589 | struct filedesc_to_leader * |
| 2590 | filedesc_to_leader_alloc(struct filedesc_to_leader *old, |
| 2591 | struct proc *leader) |
| 2592 | { |
| 2593 | struct filedesc_to_leader *fdtol; |
| 2594 | |
| 2595 | fdtol = kmalloc(sizeof(struct filedesc_to_leader), |
| 2596 | M_FILEDESC_TO_LEADER, M_WAITOK); |
| 2597 | fdtol->fdl_refcount = 1; |
| 2598 | fdtol->fdl_holdcount = 0; |
| 2599 | fdtol->fdl_wakeup = 0; |
| 2600 | fdtol->fdl_leader = leader; |
| 2601 | if (old != NULL) { |
| 2602 | fdtol->fdl_next = old->fdl_next; |
| 2603 | fdtol->fdl_prev = old; |
| 2604 | old->fdl_next = fdtol; |
| 2605 | fdtol->fdl_next->fdl_prev = fdtol; |
| 2606 | } else { |
| 2607 | fdtol->fdl_next = fdtol; |
| 2608 | fdtol->fdl_prev = fdtol; |
| 2609 | } |
| 2610 | return fdtol; |
| 2611 | } |
| 2612 | |
| 2613 | /* |
| 2614 | * Scan all file pointers in the system. The callback is made with |
| 2615 | * the master list spinlock held exclusively. |
| 2616 | * |
| 2617 | * MPSAFE |
| 2618 | */ |
| 2619 | void |
| 2620 | allfiles_scan_exclusive(int (*callback)(struct file *, void *), void *data) |
| 2621 | { |
| 2622 | struct file *fp; |
| 2623 | int res; |
| 2624 | |
| 2625 | spin_lock(&filehead_spin); |
| 2626 | LIST_FOREACH(fp, &filehead, f_list) { |
| 2627 | res = callback(fp, data); |
| 2628 | if (res < 0) |
| 2629 | break; |
| 2630 | } |
| 2631 | spin_unlock(&filehead_spin); |
| 2632 | } |
| 2633 | |
| 2634 | /* |
| 2635 | * Get file structures. |
| 2636 | * |
| 2637 | * NOT MPSAFE - process list scan, SYSCTL_OUT (probably not mpsafe) |
| 2638 | */ |
| 2639 | |
| 2640 | struct sysctl_kern_file_info { |
| 2641 | int count; |
| 2642 | int error; |
| 2643 | struct sysctl_req *req; |
| 2644 | }; |
| 2645 | |
| 2646 | static int sysctl_kern_file_callback(struct proc *p, void *data); |
| 2647 | |
| 2648 | static int |
| 2649 | sysctl_kern_file(SYSCTL_HANDLER_ARGS) |
| 2650 | { |
| 2651 | struct sysctl_kern_file_info info; |
| 2652 | |
| 2653 | /* |
| 2654 | * Note: because the number of file descriptors is calculated |
| 2655 | * in different ways for sizing vs returning the data, |
| 2656 | * there is information leakage from the first loop. However, |
| 2657 | * it is of a similar order of magnitude to the leakage from |
| 2658 | * global system statistics such as kern.openfiles. |
| 2659 | * |
| 2660 | * When just doing a count, note that we cannot just count |
| 2661 | * the elements and add f_count via the filehead list because |
| 2662 | * threaded processes share their descriptor table and f_count might |
| 2663 | * still be '1' in that case. |
| 2664 | * |
| 2665 | * Since the SYSCTL op can block, we must hold the process to |
| 2666 | * prevent it being ripped out from under us either in the |
| 2667 | * file descriptor loop or in the greater LIST_FOREACH. The |
| 2668 | * process may be in varying states of disrepair. If the process |
| 2669 | * is in SZOMB we may have caught it just as it is being removed |
| 2670 | * from the allproc list, we must skip it in that case to maintain |
| 2671 | * an unbroken chain through the allproc list. |
| 2672 | */ |
| 2673 | info.count = 0; |
| 2674 | info.error = 0; |
| 2675 | info.req = req; |
| 2676 | allproc_scan(sysctl_kern_file_callback, &info); |
| 2677 | |
| 2678 | /* |
| 2679 | * When just calculating the size, overestimate a bit to try to |
| 2680 | * prevent system activity from causing the buffer-fill call |
| 2681 | * to fail later on. |
| 2682 | */ |
| 2683 | if (req->oldptr == NULL) { |
| 2684 | info.count = (info.count + 16) + (info.count / 10); |
| 2685 | info.error = SYSCTL_OUT(req, NULL, |
| 2686 | info.count * sizeof(struct kinfo_file)); |
| 2687 | } |
| 2688 | return (info.error); |
| 2689 | } |
| 2690 | |
| 2691 | static int |
| 2692 | sysctl_kern_file_callback(struct proc *p, void *data) |
| 2693 | { |
| 2694 | struct sysctl_kern_file_info *info = data; |
| 2695 | struct kinfo_file kf; |
| 2696 | struct filedesc *fdp; |
| 2697 | struct file *fp; |
| 2698 | uid_t uid; |
| 2699 | int n; |
| 2700 | |
| 2701 | if (p->p_stat == SIDL || p->p_stat == SZOMB) |
| 2702 | return(0); |
| 2703 | if (!PRISON_CHECK(info->req->td->td_ucred, p->p_ucred) != 0) |
| 2704 | return(0); |
| 2705 | |
| 2706 | /* |
| 2707 | * Softref the fdp to prevent it from being destroyed |
| 2708 | */ |
| 2709 | spin_lock(&p->p_spin); |
| 2710 | if ((fdp = p->p_fd) == NULL) { |
| 2711 | spin_unlock(&p->p_spin); |
| 2712 | return(0); |
| 2713 | } |
| 2714 | atomic_add_int(&fdp->fd_softrefs, 1); |
| 2715 | spin_unlock(&p->p_spin); |
| 2716 | |
| 2717 | /* |
| 2718 | * The fdp's own spinlock prevents the contents from being |
| 2719 | * modified. |
| 2720 | */ |
| 2721 | spin_lock(&fdp->fd_spin); |
| 2722 | for (n = 0; n < fdp->fd_nfiles; ++n) { |
| 2723 | if ((fp = fdp->fd_files[n].fp) == NULL) |
| 2724 | continue; |
| 2725 | if (info->req->oldptr == NULL) { |
| 2726 | ++info->count; |
| 2727 | } else { |
| 2728 | uid = p->p_ucred ? p->p_ucred->cr_uid : -1; |
| 2729 | kcore_make_file(&kf, fp, p->p_pid, uid, n); |
| 2730 | spin_unlock(&fdp->fd_spin); |
| 2731 | info->error = SYSCTL_OUT(info->req, &kf, sizeof(kf)); |
| 2732 | spin_lock(&fdp->fd_spin); |
| 2733 | if (info->error) |
| 2734 | break; |
| 2735 | } |
| 2736 | } |
| 2737 | spin_unlock(&fdp->fd_spin); |
| 2738 | atomic_subtract_int(&fdp->fd_softrefs, 1); |
| 2739 | if (info->error) |
| 2740 | return(-1); |
| 2741 | return(0); |
| 2742 | } |
| 2743 | |
| 2744 | SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD, |
| 2745 | 0, 0, sysctl_kern_file, "S,file", "Entire file table"); |
| 2746 | |
| 2747 | SYSCTL_INT(_kern, OID_AUTO, minfilesperproc, CTLFLAG_RW, |
| 2748 | &minfilesperproc, 0, "Minimum files allowed open per process"); |
| 2749 | SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, |
| 2750 | &maxfilesperproc, 0, "Maximum files allowed open per process"); |
| 2751 | SYSCTL_INT(_kern, OID_AUTO, maxfilesperuser, CTLFLAG_RW, |
| 2752 | &maxfilesperuser, 0, "Maximum files allowed open per user"); |
| 2753 | |
| 2754 | SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, |
| 2755 | &maxfiles, 0, "Maximum number of files"); |
| 2756 | |
| 2757 | SYSCTL_INT(_kern, OID_AUTO, maxfilesrootres, CTLFLAG_RW, |
| 2758 | &maxfilesrootres, 0, "Descriptors reserved for root use"); |
| 2759 | |
| 2760 | SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, |
| 2761 | &nfiles, 0, "System-wide number of open files"); |
| 2762 | |
| 2763 | static void |
| 2764 | fildesc_drvinit(void *unused) |
| 2765 | { |
| 2766 | int fd; |
| 2767 | |
| 2768 | for (fd = 0; fd < NUMFDESC; fd++) { |
| 2769 | make_dev(&fildesc_ops, fd, |
| 2770 | UID_BIN, GID_BIN, 0666, "fd/%d", fd); |
| 2771 | } |
| 2772 | |
| 2773 | make_dev(&fildesc_ops, 0, UID_ROOT, GID_WHEEL, 0666, "stdin"); |
| 2774 | make_dev(&fildesc_ops, 1, UID_ROOT, GID_WHEEL, 0666, "stdout"); |
| 2775 | make_dev(&fildesc_ops, 2, UID_ROOT, GID_WHEEL, 0666, "stderr"); |
| 2776 | } |
| 2777 | |
| 2778 | /* |
| 2779 | * MPSAFE |
| 2780 | */ |
| 2781 | struct fileops badfileops = { |
| 2782 | .fo_read = badfo_readwrite, |
| 2783 | .fo_write = badfo_readwrite, |
| 2784 | .fo_ioctl = badfo_ioctl, |
| 2785 | .fo_poll = badfo_poll, |
| 2786 | .fo_kqfilter = badfo_kqfilter, |
| 2787 | .fo_stat = badfo_stat, |
| 2788 | .fo_close = badfo_close, |
| 2789 | .fo_shutdown = badfo_shutdown |
| 2790 | }; |
| 2791 | |
| 2792 | int |
| 2793 | badfo_readwrite( |
| 2794 | struct file *fp, |
| 2795 | struct uio *uio, |
| 2796 | struct ucred *cred, |
| 2797 | int flags |
| 2798 | ) { |
| 2799 | return (EBADF); |
| 2800 | } |
| 2801 | |
| 2802 | int |
| 2803 | badfo_ioctl(struct file *fp, u_long com, caddr_t data, |
| 2804 | struct ucred *cred, struct sysmsg *msgv) |
| 2805 | { |
| 2806 | return (EBADF); |
| 2807 | } |
| 2808 | |
| 2809 | int |
| 2810 | badfo_poll(struct file *fp, int events, struct ucred *cred) |
| 2811 | { |
| 2812 | return (0); |
| 2813 | } |
| 2814 | |
| 2815 | /* |
| 2816 | * Must return an error to prevent registration, typically |
| 2817 | * due to a revoked descriptor (file_filtops assigned). |
| 2818 | */ |
| 2819 | int |
| 2820 | badfo_kqfilter(struct file *fp, struct knote *kn) |
| 2821 | { |
| 2822 | return (EOPNOTSUPP); |
| 2823 | } |
| 2824 | |
| 2825 | /* |
| 2826 | * MPSAFE |
| 2827 | */ |
| 2828 | int |
| 2829 | badfo_stat(struct file *fp, struct stat *sb, struct ucred *cred) |
| 2830 | { |
| 2831 | return (EBADF); |
| 2832 | } |
| 2833 | |
| 2834 | /* |
| 2835 | * MPSAFE |
| 2836 | */ |
| 2837 | int |
| 2838 | badfo_close(struct file *fp) |
| 2839 | { |
| 2840 | return (EBADF); |
| 2841 | } |
| 2842 | |
| 2843 | /* |
| 2844 | * MPSAFE |
| 2845 | */ |
| 2846 | int |
| 2847 | badfo_shutdown(struct file *fp, int how) |
| 2848 | { |
| 2849 | return (EBADF); |
| 2850 | } |
| 2851 | |
| 2852 | /* |
| 2853 | * MPSAFE |
| 2854 | */ |
| 2855 | int |
| 2856 | nofo_shutdown(struct file *fp, int how) |
| 2857 | { |
| 2858 | return (EOPNOTSUPP); |
| 2859 | } |
| 2860 | |
| 2861 | SYSINIT(fildescdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR, |
| 2862 | fildesc_drvinit,NULL) |