1 /*- 2 * Copyright (c) 2000-2004 3 * Poul-Henning Kamp. All rights reserved. 4 * Copyright (c) 1989, 1992-1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software donated to Berkeley by 8 * Jan-Simon Pendry. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95 32 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43 33 * 34 * $FreeBSD$ 35 */ 36 37 /* 38 * TODO: 39 * remove empty directories 40 * mkdir: want it ? 41 */ 42 43 #include "opt_mac.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/conf.h> 48 #include <sys/dirent.h> 49 #include <sys/fcntl.h> 50 #include <sys/file.h> 51 #include <sys/filedesc.h> 52 #include <sys/filio.h> 53 #include <sys/kernel.h> 54 #include <sys/lock.h> 55 #include <sys/mac.h> 56 #include <sys/malloc.h> 57 #include <sys/mount.h> 58 #include <sys/namei.h> 59 #include <sys/proc.h> 60 #include <sys/stat.h> 61 #include <sys/sx.h> 62 #include <sys/time.h> 63 #include <sys/ttycom.h> 64 #include <sys/unistd.h> 65 #include <sys/vnode.h> 66 67 static struct vop_vector devfs_vnodeops; 68 static struct vop_vector devfs_specops; 69 static struct fileops devfs_ops_f; 70 71 #include <fs/devfs/devfs.h> 72 #include <fs/devfs/devfs_int.h> 73 74 static struct mtx devfs_de_interlock; 75 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); 76 77 static int 78 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp) 79 { 80 81 *devp = fp->f_vnode->v_rdev; 82 if (*devp != fp->f_data) 83 return (ENXIO); 84 KASSERT((*devp)->si_refcount > 0, 85 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); 86 *dswp = dev_refthread(*devp); 87 if (*dswp == NULL) 88 return (ENXIO); 89 return (0); 90 } 91 92 /* 93 * Construct the fully qualified path name relative to the mountpoint 94 */ 95 static char * 96 devfs_fqpn(char *buf, struct vnode *dvp, struct componentname *cnp) 97 { 98 int i; 99 struct devfs_dirent *de, *dd; 100 struct devfs_mount *dmp; 101 102 dmp = VFSTODEVFS(dvp->v_mount); 103 dd = dvp->v_data; 104 i = SPECNAMELEN; 105 buf[i] = '\0'; 106 i -= cnp->cn_namelen; 107 if (i < 0) 108 return (NULL); 109 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); 110 de = dd; 111 while (de != dmp->dm_rootdir) { 112 i--; 113 if (i < 0) 114 return (NULL); 115 buf[i] = '/'; 116 i -= de->de_dirent->d_namlen; 117 if (i < 0) 118 return (NULL); 119 bcopy(de->de_dirent->d_name, buf + i, 120 de->de_dirent->d_namlen); 121 de = TAILQ_FIRST(&de->de_dlist); /* "." */ 122 de = TAILQ_NEXT(de, de_list); /* ".." */ 123 de = de->de_dir; 124 } 125 return (buf + i); 126 } 127 128 int 129 devfs_allocv(struct devfs_dirent *de, struct mount *mp, struct vnode **vpp, struct thread *td) 130 { 131 int error; 132 struct vnode *vp; 133 struct cdev *dev; 134 135 KASSERT(td == curthread, ("devfs_allocv: td != curthread")); 136 loop: 137 138 mtx_lock(&devfs_de_interlock); 139 vp = de->de_vnode; 140 if (vp != NULL) { 141 VI_LOCK(vp); 142 mtx_unlock(&devfs_de_interlock); 143 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) 144 goto loop; 145 *vpp = vp; 146 return (0); 147 } 148 mtx_unlock(&devfs_de_interlock); 149 if (de->de_dirent->d_type == DT_CHR) { 150 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) 151 return (ENOENT); 152 dev = &de->de_cdp->cdp_c; 153 } else { 154 dev = NULL; 155 } 156 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); 157 if (error != 0) { 158 printf("devfs_allocv: failed to allocate new vnode\n"); 159 return (error); 160 } 161 162 if (de->de_dirent->d_type == DT_CHR) { 163 vp->v_type = VCHR; 164 VI_LOCK(vp); 165 dev_lock(); 166 dev_refl(dev); 167 vp->v_rdev = dev; 168 KASSERT(vp->v_usecount == 1, 169 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount)); 170 dev->si_usecount += vp->v_usecount; 171 dev_unlock(); 172 VI_UNLOCK(vp); 173 vp->v_op = &devfs_specops; 174 } else if (de->de_dirent->d_type == DT_DIR) { 175 vp->v_type = VDIR; 176 } else if (de->de_dirent->d_type == DT_LNK) { 177 vp->v_type = VLNK; 178 } else { 179 vp->v_type = VBAD; 180 } 181 mtx_lock(&devfs_de_interlock); 182 vp->v_data = de; 183 de->de_vnode = vp; 184 mtx_unlock(&devfs_de_interlock); 185 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 186 #ifdef MAC 187 mac_associate_vnode_devfs(mp, de, vp); 188 #endif 189 *vpp = vp; 190 return (0); 191 } 192 193 static int 194 devfs_access(struct vop_access_args *ap) 195 { 196 struct vnode *vp = ap->a_vp; 197 struct devfs_dirent *de; 198 int error; 199 200 de = vp->v_data; 201 if (vp->v_type == VDIR) 202 de = de->de_dir; 203 204 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, 205 ap->a_mode, ap->a_cred, NULL); 206 if (!error) 207 return (error); 208 if (error != EACCES) 209 return (error); 210 /* We do, however, allow access to the controlling terminal */ 211 if (!(ap->a_td->td_proc->p_flag & P_CONTROLT)) 212 return (error); 213 if (ap->a_td->td_proc->p_session->s_ttyvp == de->de_vnode) 214 return (0); 215 return (error); 216 } 217 218 /* ARGSUSED */ 219 static int 220 devfs_advlock(struct vop_advlock_args *ap) 221 { 222 223 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL); 224 } 225 226 /* ARGSUSED */ 227 static int 228 devfs_close(struct vop_close_args *ap) 229 { 230 struct vnode *vp = ap->a_vp, *oldvp; 231 struct thread *td = ap->a_td; 232 struct cdev *dev = vp->v_rdev; 233 struct cdevsw *dsw; 234 int error; 235 236 /* 237 * Hack: a tty device that is a controlling terminal 238 * has a reference from the session structure. 239 * We cannot easily tell that a character device is 240 * a controlling terminal, unless it is the closing 241 * process' controlling terminal. In that case, 242 * if the reference count is 2 (this last descriptor 243 * plus the session), release the reference from the session. 244 */ 245 oldvp = NULL; 246 sx_xlock(&proctree_lock); 247 if (td && vp == td->td_proc->p_session->s_ttyvp) { 248 SESS_LOCK(td->td_proc->p_session); 249 VI_LOCK(vp); 250 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) { 251 td->td_proc->p_session->s_ttyvp = NULL; 252 oldvp = vp; 253 } 254 VI_UNLOCK(vp); 255 SESS_UNLOCK(td->td_proc->p_session); 256 } 257 sx_xunlock(&proctree_lock); 258 if (oldvp != NULL) 259 vrele(oldvp); 260 /* 261 * We do not want to really close the device if it 262 * is still in use unless we are trying to close it 263 * forcibly. Since every use (buffer, vnode, swap, cmap) 264 * holds a reference to the vnode, and because we mark 265 * any other vnodes that alias this device, when the 266 * sum of the reference counts on all the aliased 267 * vnodes descends to one, we are on last close. 268 */ 269 dsw = dev_refthread(dev); 270 if (dsw == NULL) 271 return (ENXIO); 272 VI_LOCK(vp); 273 if (vp->v_iflag & VI_DOOMED) { 274 /* Forced close. */ 275 } else if (dsw->d_flags & D_TRACKCLOSE) { 276 /* Keep device updated on status. */ 277 } else if (count_dev(dev) > 1) { 278 VI_UNLOCK(vp); 279 dev_relthread(dev); 280 return (0); 281 } 282 VI_UNLOCK(vp); 283 KASSERT(dev->si_refcount > 0, 284 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 285 if (!(dsw->d_flags & D_NEEDGIANT)) { 286 DROP_GIANT(); 287 error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td); 288 PICKUP_GIANT(); 289 } else { 290 error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td); 291 } 292 dev_relthread(dev); 293 return (error); 294 } 295 296 static int 297 devfs_close_f(struct file *fp, struct thread *td) 298 { 299 300 return (vnops.fo_close(fp, td)); 301 } 302 303 /* ARGSUSED */ 304 static int 305 devfs_fsync(struct vop_fsync_args *ap) 306 { 307 if (!vn_isdisk(ap->a_vp, NULL)) 308 return (0); 309 310 return (vop_stdfsync(ap)); 311 } 312 313 static int 314 devfs_getattr(struct vop_getattr_args *ap) 315 { 316 struct vnode *vp = ap->a_vp; 317 struct vattr *vap = ap->a_vap; 318 int error = 0; 319 struct devfs_dirent *de; 320 struct cdev *dev; 321 322 de = vp->v_data; 323 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 324 if (vp->v_type == VDIR) { 325 de = de->de_dir; 326 KASSERT(de != NULL, 327 ("Null dir dirent in devfs_getattr vp=%p", vp)); 328 } 329 bzero((caddr_t) vap, sizeof(*vap)); 330 vattr_null(vap); 331 vap->va_uid = de->de_uid; 332 vap->va_gid = de->de_gid; 333 vap->va_mode = de->de_mode; 334 if (vp->v_type == VLNK) 335 vap->va_size = strlen(de->de_symlink); 336 else if (vp->v_type == VDIR) 337 vap->va_size = vap->va_bytes = DEV_BSIZE; 338 else 339 vap->va_size = 0; 340 if (vp->v_type != VDIR) 341 vap->va_bytes = 0; 342 vap->va_blocksize = DEV_BSIZE; 343 vap->va_type = vp->v_type; 344 345 #define fix(aa) \ 346 do { \ 347 if ((aa).tv_sec == 0) { \ 348 (aa).tv_sec = boottime.tv_sec; \ 349 (aa).tv_nsec = boottime.tv_usec * 1000; \ 350 } \ 351 } while (0) 352 353 if (vp->v_type != VCHR) { 354 fix(de->de_atime); 355 vap->va_atime = de->de_atime; 356 fix(de->de_mtime); 357 vap->va_mtime = de->de_mtime; 358 fix(de->de_ctime); 359 vap->va_ctime = de->de_ctime; 360 } else { 361 dev = vp->v_rdev; 362 fix(dev->si_atime); 363 vap->va_atime = dev->si_atime; 364 fix(dev->si_mtime); 365 vap->va_mtime = dev->si_mtime; 366 fix(dev->si_ctime); 367 vap->va_ctime = dev->si_ctime; 368 369 vap->va_rdev = dev->si_priv->cdp_inode; 370 } 371 vap->va_gen = 0; 372 vap->va_flags = 0; 373 vap->va_nlink = de->de_links; 374 vap->va_fileid = de->de_inode; 375 376 return (error); 377 } 378 379 /* ARGSUSED */ 380 static int 381 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 382 { 383 struct cdev *dev; 384 struct cdevsw *dsw; 385 struct vnode *vp; 386 struct vnode *vpold; 387 int error, i; 388 const char *p; 389 struct fiodgname_arg *fgn; 390 391 error = devfs_fp_check(fp, &dev, &dsw); 392 if (error) 393 return (error); 394 395 if (com == FIODTYPE) { 396 *(int *)data = dsw->d_flags & D_TYPEMASK; 397 dev_relthread(dev); 398 return (0); 399 } else if (com == FIODGNAME) { 400 fgn = data; 401 p = devtoname(dev); 402 i = strlen(p) + 1; 403 if (i > fgn->len) 404 error = EINVAL; 405 else 406 error = copyout(p, fgn->buf, i); 407 dev_relthread(dev); 408 return (error); 409 } 410 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td); 411 dev_relthread(dev); 412 if (error == ENOIOCTL) 413 error = ENOTTY; 414 if (error == 0 && com == TIOCSCTTY) { 415 vp = fp->f_vnode; 416 417 /* Do nothing if reassigning same control tty */ 418 sx_slock(&proctree_lock); 419 if (td->td_proc->p_session->s_ttyvp == vp) { 420 sx_sunlock(&proctree_lock); 421 return (0); 422 } 423 424 mtx_lock(&Giant); 425 426 vpold = td->td_proc->p_session->s_ttyvp; 427 VREF(vp); 428 SESS_LOCK(td->td_proc->p_session); 429 td->td_proc->p_session->s_ttyvp = vp; 430 SESS_UNLOCK(td->td_proc->p_session); 431 432 sx_sunlock(&proctree_lock); 433 434 /* Get rid of reference to old control tty */ 435 if (vpold) 436 vrele(vpold); 437 mtx_unlock(&Giant); 438 } 439 return (error); 440 } 441 442 /* ARGSUSED */ 443 static int 444 devfs_kqfilter_f(struct file *fp, struct knote *kn) 445 { 446 struct cdev *dev; 447 struct cdevsw *dsw; 448 int error; 449 450 error = devfs_fp_check(fp, &dev, &dsw); 451 if (error) 452 return (error); 453 error = dsw->d_kqfilter(dev, kn); 454 dev_relthread(dev); 455 return (error); 456 } 457 458 static int 459 devfs_lookupx(struct vop_lookup_args *ap) 460 { 461 struct componentname *cnp; 462 struct vnode *dvp, **vpp; 463 struct thread *td; 464 struct devfs_dirent *de, *dd; 465 struct devfs_dirent **dde; 466 struct devfs_mount *dmp; 467 struct cdev *cdev; 468 int error, flags, nameiop; 469 char specname[SPECNAMELEN + 1], *pname; 470 471 cnp = ap->a_cnp; 472 vpp = ap->a_vpp; 473 dvp = ap->a_dvp; 474 pname = cnp->cn_nameptr; 475 td = cnp->cn_thread; 476 flags = cnp->cn_flags; 477 nameiop = cnp->cn_nameiop; 478 dmp = VFSTODEVFS(dvp->v_mount); 479 dd = dvp->v_data; 480 *vpp = NULLVP; 481 482 if ((flags & ISLASTCN) && nameiop == RENAME) 483 return (EOPNOTSUPP); 484 485 if (dvp->v_type != VDIR) 486 return (ENOTDIR); 487 488 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 489 return (EIO); 490 491 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 492 if (error) 493 return (error); 494 495 if (cnp->cn_namelen == 1 && *pname == '.') { 496 if ((flags & ISLASTCN) && nameiop != LOOKUP) 497 return (EINVAL); 498 *vpp = dvp; 499 VREF(dvp); 500 return (0); 501 } 502 503 if (flags & ISDOTDOT) { 504 if ((flags & ISLASTCN) && nameiop != LOOKUP) 505 return (EINVAL); 506 VOP_UNLOCK(dvp, 0, td); 507 de = TAILQ_FIRST(&dd->de_dlist); /* "." */ 508 de = TAILQ_NEXT(de, de_list); /* ".." */ 509 de = de->de_dir; 510 error = devfs_allocv(de, dvp->v_mount, vpp, td); 511 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 512 return (error); 513 } 514 515 devfs_populate(dmp); 516 dd = dvp->v_data; 517 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen); 518 while (de == NULL) { /* While(...) so we can use break */ 519 520 if (nameiop == DELETE) 521 return (ENOENT); 522 523 /* 524 * OK, we didn't have an entry for the name we were asked for 525 * so we try to see if anybody can create it on demand. 526 */ 527 pname = devfs_fqpn(specname, dvp, cnp); 528 if (pname == NULL) 529 break; 530 531 cdev = NULL; 532 EVENTHANDLER_INVOKE(dev_clone, 533 td->td_ucred, pname, strlen(pname), &cdev); 534 if (cdev == NULL) 535 break; 536 537 devfs_populate(dmp); 538 539 dev_lock(); 540 dde = &cdev->si_priv->cdp_dirents[dmp->dm_idx]; 541 if (dde != NULL && *dde != NULL) 542 de = *dde; 543 dev_unlock(); 544 dev_rel(cdev); 545 break; 546 } 547 548 if (de == NULL || de->de_flags & DE_WHITEOUT) { 549 if ((nameiop == CREATE || nameiop == RENAME) && 550 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 551 cnp->cn_flags |= SAVENAME; 552 return (EJUSTRETURN); 553 } 554 return (ENOENT); 555 } 556 557 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 558 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 559 if (error) 560 return (error); 561 if (*vpp == dvp) { 562 VREF(dvp); 563 *vpp = dvp; 564 return (0); 565 } 566 } 567 error = devfs_allocv(de, dvp->v_mount, vpp, td); 568 return (error); 569 } 570 571 static int 572 devfs_lookup(struct vop_lookup_args *ap) 573 { 574 int j; 575 struct devfs_mount *dmp; 576 577 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 578 sx_xlock(&dmp->dm_lock); 579 j = devfs_lookupx(ap); 580 sx_xunlock(&dmp->dm_lock); 581 return (j); 582 } 583 584 static int 585 devfs_mknod(struct vop_mknod_args *ap) 586 { 587 struct componentname *cnp; 588 struct vnode *dvp, **vpp; 589 struct thread *td; 590 struct devfs_dirent *dd, *de; 591 struct devfs_mount *dmp; 592 int error; 593 594 /* 595 * The only type of node we should be creating here is a 596 * character device, for anything else return EOPNOTSUPP. 597 */ 598 if (ap->a_vap->va_type != VCHR) 599 return (EOPNOTSUPP); 600 dvp = ap->a_dvp; 601 dmp = VFSTODEVFS(dvp->v_mount); 602 sx_xlock(&dmp->dm_lock); 603 604 cnp = ap->a_cnp; 605 vpp = ap->a_vpp; 606 td = cnp->cn_thread; 607 dd = dvp->v_data; 608 609 error = ENOENT; 610 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 611 if (cnp->cn_namelen != de->de_dirent->d_namlen) 612 continue; 613 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 614 de->de_dirent->d_namlen) != 0) 615 continue; 616 if (de->de_flags & DE_WHITEOUT) 617 break; 618 goto notfound; 619 } 620 if (de == NULL) 621 goto notfound; 622 de->de_flags &= ~DE_WHITEOUT; 623 error = devfs_allocv(de, dvp->v_mount, vpp, td); 624 notfound: 625 sx_xunlock(&dmp->dm_lock); 626 return (error); 627 } 628 629 /* ARGSUSED */ 630 static int 631 devfs_open(struct vop_open_args *ap) 632 { 633 struct thread *td = ap->a_td; 634 struct vnode *vp = ap->a_vp; 635 struct cdev *dev = vp->v_rdev; 636 struct file *fp; 637 int error; 638 struct cdevsw *dsw; 639 640 if (vp->v_type == VBLK) 641 return (ENXIO); 642 643 if (dev == NULL) 644 return (ENXIO); 645 646 /* Make this field valid before any I/O in d_open. */ 647 if (dev->si_iosize_max == 0) 648 dev->si_iosize_max = DFLTPHYS; 649 650 dsw = dev_refthread(dev); 651 if (dsw == NULL) 652 return (ENXIO); 653 654 /* XXX: Special casing of ttys for deadfs. Probably redundant. */ 655 if (dsw->d_flags & D_TTY) 656 vp->v_vflag |= VV_ISTTY; 657 658 VOP_UNLOCK(vp, 0, td); 659 660 if(!(dsw->d_flags & D_NEEDGIANT)) { 661 DROP_GIANT(); 662 if (dsw->d_fdopen != NULL) 663 error = dsw->d_fdopen(dev, ap->a_mode, td, ap->a_fdidx); 664 else 665 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 666 PICKUP_GIANT(); 667 } else { 668 if (dsw->d_fdopen != NULL) 669 error = dsw->d_fdopen(dev, ap->a_mode, td, ap->a_fdidx); 670 else 671 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 672 } 673 674 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 675 676 dev_relthread(dev); 677 678 if (error) 679 return (error); 680 681 #if 0 /* /dev/console */ 682 KASSERT(ap->a_fdidx >= 0, 683 ("Could not vnode bypass device on fd %d", ap->a_fdidx)); 684 #else 685 if(ap->a_fdidx < 0) 686 return (error); 687 #endif 688 /* 689 * This is a pretty disgustingly long chain, but I am not 690 * sure there is any better way. Passing the fdidx into 691 * VOP_OPEN() offers us more information than just passing 692 * the file *. 693 */ 694 fp = ap->a_td->td_proc->p_fd->fd_ofiles[ap->a_fdidx]; 695 KASSERT(fp->f_ops == &badfileops, 696 ("Could not vnode bypass device on fdops %p", fp->f_ops)); 697 fp->f_ops = &devfs_ops_f; 698 fp->f_data = dev; 699 return (error); 700 } 701 702 static int 703 devfs_pathconf(struct vop_pathconf_args *ap) 704 { 705 706 switch (ap->a_name) { 707 case _PC_MAC_PRESENT: 708 #ifdef MAC 709 /* 710 * If MAC is enabled, devfs automatically supports 711 * trivial non-persistant label storage. 712 */ 713 *ap->a_retval = 1; 714 #else 715 *ap->a_retval = 0; 716 #endif 717 return (0); 718 default: 719 return (vop_stdpathconf(ap)); 720 } 721 /* NOTREACHED */ 722 } 723 724 /* ARGSUSED */ 725 static int 726 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 727 { 728 struct cdev *dev; 729 struct cdevsw *dsw; 730 int error; 731 732 error = devfs_fp_check(fp, &dev, &dsw); 733 if (error) 734 return (error); 735 error = dsw->d_poll(dev, events, td); 736 dev_relthread(dev); 737 return(error); 738 } 739 740 /* 741 * Print out the contents of a special device vnode. 742 */ 743 static int 744 devfs_print(struct vop_print_args *ap) 745 { 746 747 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 748 return (0); 749 } 750 751 /* ARGSUSED */ 752 static int 753 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td) 754 { 755 struct cdev *dev; 756 int ioflag, error, resid; 757 struct cdevsw *dsw; 758 759 error = devfs_fp_check(fp, &dev, &dsw); 760 if (error) 761 return (error); 762 resid = uio->uio_resid; 763 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 764 if (ioflag & O_DIRECT) 765 ioflag |= IO_DIRECT; 766 767 if ((flags & FOF_OFFSET) == 0) 768 uio->uio_offset = fp->f_offset; 769 770 error = dsw->d_read(dev, uio, ioflag); 771 dev_relthread(dev); 772 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 773 vfs_timestamp(&dev->si_atime); 774 775 if ((flags & FOF_OFFSET) == 0) 776 fp->f_offset = uio->uio_offset; 777 fp->f_nextoff = uio->uio_offset; 778 return (error); 779 } 780 781 static int 782 devfs_readdir(struct vop_readdir_args *ap) 783 { 784 int error; 785 struct uio *uio; 786 struct dirent *dp; 787 struct devfs_dirent *dd; 788 struct devfs_dirent *de; 789 struct devfs_mount *dmp; 790 off_t off, oldoff; 791 int *tmp_ncookies = NULL; 792 793 if (ap->a_vp->v_type != VDIR) 794 return (ENOTDIR); 795 796 uio = ap->a_uio; 797 if (uio->uio_offset < 0) 798 return (EINVAL); 799 800 /* 801 * XXX: This is a temporary hack to get around this filesystem not 802 * supporting cookies. We store the location of the ncookies pointer 803 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 804 * and set the number of cookies to 0. We then set the pointer to 805 * NULL so that vfs_read_dirent doesn't try to call realloc() on 806 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 807 * pointer to its original location before returning to the caller. 808 */ 809 if (ap->a_ncookies != NULL) { 810 tmp_ncookies = ap->a_ncookies; 811 *ap->a_ncookies = 0; 812 ap->a_ncookies = NULL; 813 } 814 815 dmp = VFSTODEVFS(ap->a_vp->v_mount); 816 sx_xlock(&dmp->dm_lock); 817 devfs_populate(dmp); 818 error = 0; 819 de = ap->a_vp->v_data; 820 off = 0; 821 oldoff = uio->uio_offset; 822 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 823 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 824 if (dd->de_flags & DE_WHITEOUT) 825 continue; 826 if (dd->de_dirent->d_type == DT_DIR) 827 de = dd->de_dir; 828 else 829 de = dd; 830 dp = dd->de_dirent; 831 if (dp->d_reclen > uio->uio_resid) 832 break; 833 dp->d_fileno = de->de_inode; 834 if (off >= uio->uio_offset) { 835 error = vfs_read_dirent(ap, dp, off); 836 if (error) 837 break; 838 } 839 off += dp->d_reclen; 840 } 841 sx_xunlock(&dmp->dm_lock); 842 uio->uio_offset = off; 843 844 /* 845 * Restore ap->a_ncookies if it wasn't originally NULL in the first 846 * place. 847 */ 848 if (tmp_ncookies != NULL) 849 ap->a_ncookies = tmp_ncookies; 850 851 return (error); 852 } 853 854 static int 855 devfs_readlink(struct vop_readlink_args *ap) 856 { 857 struct devfs_dirent *de; 858 859 de = ap->a_vp->v_data; 860 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 861 } 862 863 static int 864 devfs_reclaim(struct vop_reclaim_args *ap) 865 { 866 struct vnode *vp = ap->a_vp; 867 struct devfs_dirent *de; 868 struct cdev *dev; 869 870 mtx_lock(&devfs_de_interlock); 871 de = vp->v_data; 872 if (de != NULL) { 873 de->de_vnode = NULL; 874 vp->v_data = NULL; 875 } 876 mtx_unlock(&devfs_de_interlock); 877 878 vnode_destroy_vobject(vp); 879 880 dev = vp->v_rdev; 881 vp->v_rdev = NULL; 882 883 if (dev == NULL) 884 return (0); 885 886 dev_lock(); 887 dev->si_usecount -= vp->v_usecount; 888 dev_unlock(); 889 dev_rel(dev); 890 return (0); 891 } 892 893 static int 894 devfs_remove(struct vop_remove_args *ap) 895 { 896 struct vnode *vp = ap->a_vp; 897 struct devfs_dirent *dd; 898 struct devfs_dirent *de; 899 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 900 901 sx_xlock(&dmp->dm_lock); 902 dd = ap->a_dvp->v_data; 903 de = vp->v_data; 904 if (de->de_cdp == NULL) { 905 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 906 devfs_delete(dmp, de); 907 } else { 908 de->de_flags |= DE_WHITEOUT; 909 } 910 sx_xunlock(&dmp->dm_lock); 911 return (0); 912 } 913 914 /* 915 * Revoke is called on a tty when a terminal session ends. The vnode 916 * is orphaned by setting v_op to deadfs so we need to let go of it 917 * as well so that we create a new one next time around. 918 * 919 */ 920 static int 921 devfs_revoke(struct vop_revoke_args *ap) 922 { 923 struct vnode *vp = ap->a_vp, *vp2; 924 struct cdev *dev; 925 struct cdev_priv *cdp; 926 struct devfs_dirent *de; 927 int i; 928 929 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 930 931 dev = vp->v_rdev; 932 cdp = dev->si_priv; 933 for (;;) { 934 mtx_lock(&devfs_de_interlock); 935 dev_lock(); 936 vp2 = NULL; 937 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 938 de = cdp->cdp_dirents[i]; 939 if (de == NULL) 940 continue; 941 942 vp2 = de->de_vnode; 943 if (vp2 != NULL) { 944 de->de_vnode = NULL; 945 dev_unlock(); 946 VI_LOCK(vp2); 947 mtx_unlock(&devfs_de_interlock); 948 vholdl(vp2); 949 VI_UNLOCK(vp2); 950 vgone(vp2); 951 vdrop(vp2); 952 break; 953 } 954 } 955 if (vp2 != NULL) { 956 continue; 957 } 958 dev_unlock(); 959 mtx_unlock(&devfs_de_interlock); 960 break; 961 } 962 return (0); 963 } 964 965 static int 966 devfs_rioctl(struct vop_ioctl_args *ap) 967 { 968 int error; 969 struct devfs_mount *dmp; 970 971 dmp = VFSTODEVFS(ap->a_vp->v_mount); 972 sx_xlock(&dmp->dm_lock); 973 devfs_populate(dmp); 974 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 975 sx_xunlock(&dmp->dm_lock); 976 return (error); 977 } 978 979 static int 980 devfs_rread(struct vop_read_args *ap) 981 { 982 983 if (ap->a_vp->v_type != VDIR) 984 return (EINVAL); 985 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 986 } 987 988 static int 989 devfs_setattr(struct vop_setattr_args *ap) 990 { 991 struct devfs_dirent *de; 992 struct vattr *vap; 993 struct vnode *vp; 994 int c, error; 995 uid_t uid; 996 gid_t gid; 997 998 vap = ap->a_vap; 999 vp = ap->a_vp; 1000 if ((vap->va_type != VNON) || 1001 (vap->va_nlink != VNOVAL) || 1002 (vap->va_fsid != VNOVAL) || 1003 (vap->va_fileid != VNOVAL) || 1004 (vap->va_blocksize != VNOVAL) || 1005 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1006 (vap->va_rdev != VNOVAL) || 1007 ((int)vap->va_bytes != VNOVAL) || 1008 (vap->va_gen != VNOVAL)) { 1009 return (EINVAL); 1010 } 1011 1012 de = vp->v_data; 1013 if (vp->v_type == VDIR) 1014 de = de->de_dir; 1015 1016 error = c = 0; 1017 if (vap->va_uid == (uid_t)VNOVAL) 1018 uid = de->de_uid; 1019 else 1020 uid = vap->va_uid; 1021 if (vap->va_gid == (gid_t)VNOVAL) 1022 gid = de->de_gid; 1023 else 1024 gid = vap->va_gid; 1025 if (uid != de->de_uid || gid != de->de_gid) { 1026 if (((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1027 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) && 1028 (error = suser_cred(ap->a_td->td_ucred, SUSER_ALLOWJAIL)) != 0) 1029 return (error); 1030 de->de_uid = uid; 1031 de->de_gid = gid; 1032 c = 1; 1033 } 1034 1035 if (vap->va_mode != (mode_t)VNOVAL) { 1036 if ((ap->a_cred->cr_uid != de->de_uid) && 1037 (error = suser_cred(ap->a_td->td_ucred, SUSER_ALLOWJAIL))) 1038 return (error); 1039 de->de_mode = vap->va_mode; 1040 c = 1; 1041 } 1042 1043 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1044 /* See the comment in ufs_vnops::ufs_setattr(). */ 1045 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) && 1046 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 1047 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td)))) 1048 return (error); 1049 if (vap->va_atime.tv_sec != VNOVAL) { 1050 if (vp->v_type == VCHR) 1051 vp->v_rdev->si_atime = vap->va_atime; 1052 else 1053 de->de_atime = vap->va_atime; 1054 } 1055 if (vap->va_mtime.tv_sec != VNOVAL) { 1056 if (vp->v_type == VCHR) 1057 vp->v_rdev->si_mtime = vap->va_mtime; 1058 else 1059 de->de_mtime = vap->va_mtime; 1060 } 1061 c = 1; 1062 } 1063 1064 if (c) { 1065 if (vp->v_type == VCHR) 1066 vfs_timestamp(&vp->v_rdev->si_ctime); 1067 else 1068 vfs_timestamp(&de->de_mtime); 1069 } 1070 return (0); 1071 } 1072 1073 #ifdef MAC 1074 static int 1075 devfs_setlabel(struct vop_setlabel_args *ap) 1076 { 1077 struct vnode *vp; 1078 struct devfs_dirent *de; 1079 1080 vp = ap->a_vp; 1081 de = vp->v_data; 1082 1083 mac_relabel_vnode(ap->a_cred, vp, ap->a_label); 1084 mac_update_devfsdirent(vp->v_mount, de, vp); 1085 1086 return (0); 1087 } 1088 #endif 1089 1090 static int 1091 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1092 { 1093 1094 return (vnops.fo_stat(fp, sb, cred, td)); 1095 } 1096 1097 static int 1098 devfs_symlink(struct vop_symlink_args *ap) 1099 { 1100 int i, error; 1101 struct devfs_dirent *dd; 1102 struct devfs_dirent *de; 1103 struct devfs_mount *dmp; 1104 struct thread *td; 1105 1106 td = ap->a_cnp->cn_thread; 1107 KASSERT(td == curthread, ("devfs_symlink: td != curthread")); 1108 error = suser(td); 1109 if (error) 1110 return(error); 1111 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1112 dd = ap->a_dvp->v_data; 1113 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1114 de->de_uid = 0; 1115 de->de_gid = 0; 1116 de->de_mode = 0755; 1117 de->de_inode = alloc_unr(devfs_inos); 1118 de->de_dirent->d_type = DT_LNK; 1119 i = strlen(ap->a_target) + 1; 1120 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1121 bcopy(ap->a_target, de->de_symlink, i); 1122 sx_xlock(&dmp->dm_lock); 1123 #ifdef MAC 1124 mac_create_devfs_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1125 #endif 1126 TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list); 1127 devfs_allocv(de, ap->a_dvp->v_mount, ap->a_vpp, td); 1128 sx_xunlock(&dmp->dm_lock); 1129 return (0); 1130 } 1131 1132 /* ARGSUSED */ 1133 static int 1134 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td) 1135 { 1136 struct cdev *dev; 1137 int error, ioflag, resid; 1138 struct cdevsw *dsw; 1139 1140 error = devfs_fp_check(fp, &dev, &dsw); 1141 if (error) 1142 return (error); 1143 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1144 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1145 if (ioflag & O_DIRECT) 1146 ioflag |= IO_DIRECT; 1147 if ((flags & FOF_OFFSET) == 0) 1148 uio->uio_offset = fp->f_offset; 1149 1150 resid = uio->uio_resid; 1151 1152 error = dsw->d_write(dev, uio, ioflag); 1153 dev_relthread(dev); 1154 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1155 vfs_timestamp(&dev->si_ctime); 1156 dev->si_mtime = dev->si_ctime; 1157 } 1158 1159 if ((flags & FOF_OFFSET) == 0) 1160 fp->f_offset = uio->uio_offset; 1161 fp->f_nextoff = uio->uio_offset; 1162 return (error); 1163 } 1164 1165 dev_t 1166 dev2udev(struct cdev *x) 1167 { 1168 if (x == NULL) 1169 return (NODEV); 1170 return (x->si_priv->cdp_inode); 1171 } 1172 1173 static struct fileops devfs_ops_f = { 1174 .fo_read = devfs_read_f, 1175 .fo_write = devfs_write_f, 1176 .fo_ioctl = devfs_ioctl_f, 1177 .fo_poll = devfs_poll_f, 1178 .fo_kqfilter = devfs_kqfilter_f, 1179 .fo_stat = devfs_stat_f, 1180 .fo_close = devfs_close_f, 1181 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1182 }; 1183 1184 static struct vop_vector devfs_vnodeops = { 1185 .vop_default = &default_vnodeops, 1186 1187 .vop_access = devfs_access, 1188 .vop_getattr = devfs_getattr, 1189 .vop_ioctl = devfs_rioctl, 1190 .vop_lookup = devfs_lookup, 1191 .vop_mknod = devfs_mknod, 1192 .vop_pathconf = devfs_pathconf, 1193 .vop_read = devfs_rread, 1194 .vop_readdir = devfs_readdir, 1195 .vop_readlink = devfs_readlink, 1196 .vop_reclaim = devfs_reclaim, 1197 .vop_remove = devfs_remove, 1198 .vop_revoke = devfs_revoke, 1199 .vop_setattr = devfs_setattr, 1200 #ifdef MAC 1201 .vop_setlabel = devfs_setlabel, 1202 #endif 1203 .vop_symlink = devfs_symlink, 1204 }; 1205 1206 static struct vop_vector devfs_specops = { 1207 .vop_default = &default_vnodeops, 1208 1209 .vop_access = devfs_access, 1210 .vop_advlock = devfs_advlock, 1211 .vop_bmap = VOP_PANIC, 1212 .vop_close = devfs_close, 1213 .vop_create = VOP_PANIC, 1214 .vop_fsync = devfs_fsync, 1215 .vop_getattr = devfs_getattr, 1216 .vop_lease = VOP_NULL, 1217 .vop_link = VOP_PANIC, 1218 .vop_mkdir = VOP_PANIC, 1219 .vop_mknod = VOP_PANIC, 1220 .vop_open = devfs_open, 1221 .vop_pathconf = devfs_pathconf, 1222 .vop_print = devfs_print, 1223 .vop_read = VOP_PANIC, 1224 .vop_readdir = VOP_PANIC, 1225 .vop_readlink = VOP_PANIC, 1226 .vop_reallocblks = VOP_PANIC, 1227 .vop_reclaim = devfs_reclaim, 1228 .vop_remove = devfs_remove, 1229 .vop_rename = VOP_PANIC, 1230 .vop_revoke = devfs_revoke, 1231 .vop_rmdir = VOP_PANIC, 1232 .vop_setattr = devfs_setattr, 1233 #ifdef MAC 1234 .vop_setlabel = devfs_setlabel, 1235 #endif 1236 .vop_strategy = VOP_PANIC, 1237 .vop_symlink = VOP_PANIC, 1238 .vop_write = VOP_PANIC, 1239 }; 1240 1241 /* 1242 * Our calling convention to the device drivers used to be that we passed 1243 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1244 * flags instead since that's what open(), close() and ioctl() takes and 1245 * we don't really want vnode.h in device drivers. 1246 * We solved the source compatibility by redefining some vnode flags to 1247 * be the same as the fcntl ones and by sending down the bitwise OR of 1248 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1249 * pulls the rug out under this. 1250 */ 1251 CTASSERT(O_NONBLOCK == IO_NDELAY); 1252 CTASSERT(O_FSYNC == IO_SYNC); 1253