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 * mkdir: want it ? 40 */ 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/conf.h> 45 #include <sys/dirent.h> 46 #include <sys/fcntl.h> 47 #include <sys/file.h> 48 #include <sys/filedesc.h> 49 #include <sys/filio.h> 50 #include <sys/jail.h> 51 #include <sys/kernel.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mount.h> 55 #include <sys/namei.h> 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/stat.h> 59 #include <sys/sx.h> 60 #include <sys/time.h> 61 #include <sys/ttycom.h> 62 #include <sys/unistd.h> 63 #include <sys/vnode.h> 64 65 static struct vop_vector devfs_vnodeops; 66 static struct vop_vector devfs_specops; 67 static struct fileops devfs_ops_f; 68 69 #include <fs/devfs/devfs.h> 70 #include <fs/devfs/devfs_int.h> 71 72 #include <security/mac/mac_framework.h> 73 74 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data"); 75 76 struct mtx devfs_de_interlock; 77 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); 78 struct sx clone_drain_lock; 79 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock"); 80 struct mtx cdevpriv_mtx; 81 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF); 82 83 static int 84 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp, 85 int *ref) 86 { 87 88 *dswp = devvn_refthread(fp->f_vnode, devp, ref); 89 if (*devp != fp->f_data) { 90 if (*dswp != NULL) 91 dev_relthread(*devp, *ref); 92 return (ENXIO); 93 } 94 KASSERT((*devp)->si_refcount > 0, 95 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); 96 if (*dswp == NULL) 97 return (ENXIO); 98 curthread->td_fpop = fp; 99 return (0); 100 } 101 102 int 103 devfs_get_cdevpriv(void **datap) 104 { 105 struct file *fp; 106 struct cdev_privdata *p; 107 int error; 108 109 fp = curthread->td_fpop; 110 if (fp == NULL) 111 return (EBADF); 112 p = fp->f_cdevpriv; 113 if (p != NULL) { 114 error = 0; 115 *datap = p->cdpd_data; 116 } else 117 error = ENOENT; 118 return (error); 119 } 120 121 int 122 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr) 123 { 124 struct file *fp; 125 struct cdev_priv *cdp; 126 struct cdev_privdata *p; 127 int error; 128 129 fp = curthread->td_fpop; 130 if (fp == NULL) 131 return (ENOENT); 132 cdp = cdev2priv((struct cdev *)fp->f_data); 133 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK); 134 p->cdpd_data = priv; 135 p->cdpd_dtr = priv_dtr; 136 p->cdpd_fp = fp; 137 mtx_lock(&cdevpriv_mtx); 138 if (fp->f_cdevpriv == NULL) { 139 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list); 140 fp->f_cdevpriv = p; 141 mtx_unlock(&cdevpriv_mtx); 142 error = 0; 143 } else { 144 mtx_unlock(&cdevpriv_mtx); 145 free(p, M_CDEVPDATA); 146 error = EBUSY; 147 } 148 return (error); 149 } 150 151 void 152 devfs_destroy_cdevpriv(struct cdev_privdata *p) 153 { 154 155 mtx_assert(&cdevpriv_mtx, MA_OWNED); 156 p->cdpd_fp->f_cdevpriv = NULL; 157 LIST_REMOVE(p, cdpd_list); 158 mtx_unlock(&cdevpriv_mtx); 159 (p->cdpd_dtr)(p->cdpd_data); 160 free(p, M_CDEVPDATA); 161 } 162 163 void 164 devfs_fpdrop(struct file *fp) 165 { 166 struct cdev_privdata *p; 167 168 mtx_lock(&cdevpriv_mtx); 169 if ((p = fp->f_cdevpriv) == NULL) { 170 mtx_unlock(&cdevpriv_mtx); 171 return; 172 } 173 devfs_destroy_cdevpriv(p); 174 } 175 176 void 177 devfs_clear_cdevpriv(void) 178 { 179 struct file *fp; 180 181 fp = curthread->td_fpop; 182 if (fp == NULL) 183 return; 184 devfs_fpdrop(fp); 185 } 186 187 /* 188 * On success devfs_populate_vp() returns with dmp->dm_lock held. 189 */ 190 static int 191 devfs_populate_vp(struct vnode *vp) 192 { 193 struct devfs_dirent *de; 194 struct devfs_mount *dmp; 195 int locked; 196 197 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp"); 198 199 dmp = VFSTODEVFS(vp->v_mount); 200 locked = VOP_ISLOCKED(vp); 201 202 sx_xlock(&dmp->dm_lock); 203 DEVFS_DMP_HOLD(dmp); 204 205 /* Can't call devfs_populate() with the vnode lock held. */ 206 VOP_UNLOCK(vp, 0); 207 devfs_populate(dmp); 208 209 sx_xunlock(&dmp->dm_lock); 210 vn_lock(vp, locked | LK_RETRY); 211 sx_xlock(&dmp->dm_lock); 212 if (DEVFS_DMP_DROP(dmp)) { 213 sx_xunlock(&dmp->dm_lock); 214 devfs_unmount_final(dmp); 215 return (EBADF); 216 } 217 if ((vp->v_iflag & VI_DOOMED) != 0) { 218 sx_xunlock(&dmp->dm_lock); 219 return (EBADF); 220 } 221 de = vp->v_data; 222 KASSERT(de != NULL, 223 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed")); 224 if ((de->de_flags & DE_DOOMED) != 0) { 225 sx_xunlock(&dmp->dm_lock); 226 return (EBADF); 227 } 228 229 return (0); 230 } 231 232 static int 233 devfs_vptocnp(struct vop_vptocnp_args *ap) 234 { 235 struct vnode *vp = ap->a_vp; 236 struct vnode **dvp = ap->a_vpp; 237 struct devfs_mount *dmp; 238 char *buf = ap->a_buf; 239 int *buflen = ap->a_buflen; 240 struct devfs_dirent *dd, *de; 241 int i, error; 242 243 dmp = VFSTODEVFS(vp->v_mount); 244 245 error = devfs_populate_vp(vp); 246 if (error != 0) 247 return (error); 248 249 i = *buflen; 250 dd = vp->v_data; 251 252 if (vp->v_type == VCHR) { 253 i -= strlen(dd->de_cdp->cdp_c.si_name); 254 if (i < 0) { 255 error = ENOMEM; 256 goto finished; 257 } 258 bcopy(dd->de_cdp->cdp_c.si_name, buf + i, 259 strlen(dd->de_cdp->cdp_c.si_name)); 260 de = dd->de_dir; 261 } else if (vp->v_type == VDIR) { 262 if (dd == dmp->dm_rootdir) { 263 *dvp = vp; 264 vhold(*dvp); 265 goto finished; 266 } 267 i -= dd->de_dirent->d_namlen; 268 if (i < 0) { 269 error = ENOMEM; 270 goto finished; 271 } 272 bcopy(dd->de_dirent->d_name, buf + i, 273 dd->de_dirent->d_namlen); 274 de = dd; 275 } else { 276 error = ENOENT; 277 goto finished; 278 } 279 *buflen = i; 280 de = devfs_parent_dirent(de); 281 if (de == NULL) { 282 error = ENOENT; 283 goto finished; 284 } 285 mtx_lock(&devfs_de_interlock); 286 *dvp = de->de_vnode; 287 if (*dvp != NULL) { 288 VI_LOCK(*dvp); 289 mtx_unlock(&devfs_de_interlock); 290 vholdl(*dvp); 291 VI_UNLOCK(*dvp); 292 } else { 293 mtx_unlock(&devfs_de_interlock); 294 error = ENOENT; 295 } 296 finished: 297 sx_xunlock(&dmp->dm_lock); 298 return (error); 299 } 300 301 /* 302 * Construct the fully qualified path name relative to the mountpoint. 303 * If a NULL cnp is provided, no '/' is appended to the resulting path. 304 */ 305 char * 306 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd, 307 struct componentname *cnp) 308 { 309 int i; 310 struct devfs_dirent *de; 311 312 sx_assert(&dmp->dm_lock, SA_LOCKED); 313 314 i = SPECNAMELEN; 315 buf[i] = '\0'; 316 if (cnp != NULL) 317 i -= cnp->cn_namelen; 318 if (i < 0) 319 return (NULL); 320 if (cnp != NULL) 321 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); 322 de = dd; 323 while (de != dmp->dm_rootdir) { 324 if (cnp != NULL || i < SPECNAMELEN) { 325 i--; 326 if (i < 0) 327 return (NULL); 328 buf[i] = '/'; 329 } 330 i -= de->de_dirent->d_namlen; 331 if (i < 0) 332 return (NULL); 333 bcopy(de->de_dirent->d_name, buf + i, 334 de->de_dirent->d_namlen); 335 de = devfs_parent_dirent(de); 336 if (de == NULL) 337 return (NULL); 338 } 339 return (buf + i); 340 } 341 342 static int 343 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp, 344 struct devfs_dirent *de) 345 { 346 int not_found; 347 348 not_found = 0; 349 if (de->de_flags & DE_DOOMED) 350 not_found = 1; 351 if (DEVFS_DE_DROP(de)) { 352 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed")); 353 devfs_dirent_free(de); 354 } 355 if (DEVFS_DMP_DROP(dmp)) { 356 KASSERT(not_found == 1, 357 ("DEVFS mount struct freed before dirent")); 358 not_found = 2; 359 sx_xunlock(&dmp->dm_lock); 360 devfs_unmount_final(dmp); 361 } 362 if (not_found == 1 || (drop_dm_lock && not_found != 2)) 363 sx_unlock(&dmp->dm_lock); 364 return (not_found); 365 } 366 367 static void 368 devfs_insmntque_dtr(struct vnode *vp, void *arg) 369 { 370 struct devfs_dirent *de; 371 372 de = (struct devfs_dirent *)arg; 373 mtx_lock(&devfs_de_interlock); 374 vp->v_data = NULL; 375 de->de_vnode = NULL; 376 mtx_unlock(&devfs_de_interlock); 377 vgone(vp); 378 vput(vp); 379 } 380 381 /* 382 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops 383 * it on return. 384 */ 385 int 386 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode, 387 struct vnode **vpp) 388 { 389 int error; 390 struct vnode *vp; 391 struct cdev *dev; 392 struct devfs_mount *dmp; 393 struct cdevsw *dsw; 394 395 dmp = VFSTODEVFS(mp); 396 if (de->de_flags & DE_DOOMED) { 397 sx_xunlock(&dmp->dm_lock); 398 return (ENOENT); 399 } 400 DEVFS_DE_HOLD(de); 401 DEVFS_DMP_HOLD(dmp); 402 mtx_lock(&devfs_de_interlock); 403 vp = de->de_vnode; 404 if (vp != NULL) { 405 VI_LOCK(vp); 406 mtx_unlock(&devfs_de_interlock); 407 sx_xunlock(&dmp->dm_lock); 408 error = vget(vp, lockmode | LK_INTERLOCK, curthread); 409 sx_xlock(&dmp->dm_lock); 410 if (devfs_allocv_drop_refs(0, dmp, de)) { 411 if (error == 0) 412 vput(vp); 413 return (ENOENT); 414 } 415 else if (error) { 416 sx_xunlock(&dmp->dm_lock); 417 return (error); 418 } 419 sx_xunlock(&dmp->dm_lock); 420 *vpp = vp; 421 return (0); 422 } 423 mtx_unlock(&devfs_de_interlock); 424 if (de->de_dirent->d_type == DT_CHR) { 425 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) { 426 devfs_allocv_drop_refs(1, dmp, de); 427 return (ENOENT); 428 } 429 dev = &de->de_cdp->cdp_c; 430 } else { 431 dev = NULL; 432 } 433 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); 434 if (error != 0) { 435 devfs_allocv_drop_refs(1, dmp, de); 436 printf("devfs_allocv: failed to allocate new vnode\n"); 437 return (error); 438 } 439 440 if (de->de_dirent->d_type == DT_CHR) { 441 vp->v_type = VCHR; 442 VI_LOCK(vp); 443 dev_lock(); 444 dev_refl(dev); 445 /* XXX: v_rdev should be protect by vnode lock */ 446 vp->v_rdev = dev; 447 KASSERT(vp->v_usecount == 1, 448 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount)); 449 dev->si_usecount += vp->v_usecount; 450 /* Special casing of ttys for deadfs. Probably redundant. */ 451 dsw = dev->si_devsw; 452 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) 453 vp->v_vflag |= VV_ISTTY; 454 dev_unlock(); 455 VI_UNLOCK(vp); 456 if ((dev->si_flags & SI_ETERNAL) != 0) 457 vp->v_vflag |= VV_ETERNALDEV; 458 vp->v_op = &devfs_specops; 459 } else if (de->de_dirent->d_type == DT_DIR) { 460 vp->v_type = VDIR; 461 } else if (de->de_dirent->d_type == DT_LNK) { 462 vp->v_type = VLNK; 463 } else { 464 vp->v_type = VBAD; 465 } 466 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); 467 VN_LOCK_ASHARE(vp); 468 mtx_lock(&devfs_de_interlock); 469 vp->v_data = de; 470 de->de_vnode = vp; 471 mtx_unlock(&devfs_de_interlock); 472 error = insmntque1(vp, mp, devfs_insmntque_dtr, de); 473 if (error != 0) { 474 (void) devfs_allocv_drop_refs(1, dmp, de); 475 return (error); 476 } 477 if (devfs_allocv_drop_refs(0, dmp, de)) { 478 vput(vp); 479 return (ENOENT); 480 } 481 #ifdef MAC 482 mac_devfs_vnode_associate(mp, de, vp); 483 #endif 484 sx_xunlock(&dmp->dm_lock); 485 *vpp = vp; 486 return (0); 487 } 488 489 static int 490 devfs_access(struct vop_access_args *ap) 491 { 492 struct vnode *vp = ap->a_vp; 493 struct devfs_dirent *de; 494 int error; 495 496 de = vp->v_data; 497 if (vp->v_type == VDIR) 498 de = de->de_dir; 499 500 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, 501 ap->a_accmode, ap->a_cred, NULL); 502 if (error == 0) 503 return (0); 504 if (error != EACCES) 505 return (error); 506 /* We do, however, allow access to the controlling terminal */ 507 if (!(ap->a_td->td_proc->p_flag & P_CONTROLT)) 508 return (error); 509 if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp) 510 return (0); 511 return (error); 512 } 513 514 /* ARGSUSED */ 515 static int 516 devfs_close(struct vop_close_args *ap) 517 { 518 struct vnode *vp = ap->a_vp, *oldvp; 519 struct thread *td = ap->a_td; 520 struct cdev *dev = vp->v_rdev; 521 struct cdevsw *dsw; 522 int vp_locked, error, ref; 523 524 /* 525 * XXX: Don't call d_close() if we were called because of 526 * XXX: insmntque1() failure. 527 */ 528 if (vp->v_data == NULL) 529 return (0); 530 531 /* 532 * Hack: a tty device that is a controlling terminal 533 * has a reference from the session structure. 534 * We cannot easily tell that a character device is 535 * a controlling terminal, unless it is the closing 536 * process' controlling terminal. In that case, 537 * if the reference count is 2 (this last descriptor 538 * plus the session), release the reference from the session. 539 */ 540 oldvp = NULL; 541 sx_xlock(&proctree_lock); 542 if (td && vp == td->td_proc->p_session->s_ttyvp) { 543 SESS_LOCK(td->td_proc->p_session); 544 VI_LOCK(vp); 545 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) { 546 td->td_proc->p_session->s_ttyvp = NULL; 547 td->td_proc->p_session->s_ttydp = NULL; 548 oldvp = vp; 549 } 550 VI_UNLOCK(vp); 551 SESS_UNLOCK(td->td_proc->p_session); 552 } 553 sx_xunlock(&proctree_lock); 554 if (oldvp != NULL) 555 vrele(oldvp); 556 /* 557 * We do not want to really close the device if it 558 * is still in use unless we are trying to close it 559 * forcibly. Since every use (buffer, vnode, swap, cmap) 560 * holds a reference to the vnode, and because we mark 561 * any other vnodes that alias this device, when the 562 * sum of the reference counts on all the aliased 563 * vnodes descends to one, we are on last close. 564 */ 565 dsw = dev_refthread(dev, &ref); 566 if (dsw == NULL) 567 return (ENXIO); 568 VI_LOCK(vp); 569 if (vp->v_iflag & VI_DOOMED) { 570 /* Forced close. */ 571 } else if (dsw->d_flags & D_TRACKCLOSE) { 572 /* Keep device updated on status. */ 573 } else if (count_dev(dev) > 1) { 574 VI_UNLOCK(vp); 575 dev_relthread(dev, ref); 576 return (0); 577 } 578 vholdl(vp); 579 VI_UNLOCK(vp); 580 vp_locked = VOP_ISLOCKED(vp); 581 VOP_UNLOCK(vp, 0); 582 KASSERT(dev->si_refcount > 0, 583 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 584 error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td); 585 dev_relthread(dev, ref); 586 vn_lock(vp, vp_locked | LK_RETRY); 587 vdrop(vp); 588 return (error); 589 } 590 591 static int 592 devfs_close_f(struct file *fp, struct thread *td) 593 { 594 int error; 595 struct file *fpop; 596 597 fpop = td->td_fpop; 598 td->td_fpop = fp; 599 error = vnops.fo_close(fp, td); 600 td->td_fpop = fpop; 601 return (error); 602 } 603 604 static int 605 devfs_fsync(struct vop_fsync_args *ap) 606 { 607 int error; 608 struct bufobj *bo; 609 struct devfs_dirent *de; 610 611 if (!vn_isdisk(ap->a_vp, &error)) { 612 bo = &ap->a_vp->v_bufobj; 613 de = ap->a_vp->v_data; 614 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) { 615 printf("Device %s went missing before all of the data " 616 "could be written to it; expect data loss.\n", 617 de->de_dirent->d_name); 618 619 error = vop_stdfsync(ap); 620 if (bo->bo_dirty.bv_cnt != 0 || error != 0) 621 panic("devfs_fsync: vop_stdfsync failed."); 622 } 623 624 return (0); 625 } 626 627 return (vop_stdfsync(ap)); 628 } 629 630 static int 631 devfs_getattr(struct vop_getattr_args *ap) 632 { 633 struct vnode *vp = ap->a_vp; 634 struct vattr *vap = ap->a_vap; 635 int error; 636 struct devfs_dirent *de; 637 struct devfs_mount *dmp; 638 struct cdev *dev; 639 640 error = devfs_populate_vp(vp); 641 if (error != 0) 642 return (error); 643 644 dmp = VFSTODEVFS(vp->v_mount); 645 sx_xunlock(&dmp->dm_lock); 646 647 de = vp->v_data; 648 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 649 if (vp->v_type == VDIR) { 650 de = de->de_dir; 651 KASSERT(de != NULL, 652 ("Null dir dirent in devfs_getattr vp=%p", vp)); 653 } 654 vap->va_uid = de->de_uid; 655 vap->va_gid = de->de_gid; 656 vap->va_mode = de->de_mode; 657 if (vp->v_type == VLNK) 658 vap->va_size = strlen(de->de_symlink); 659 else if (vp->v_type == VDIR) 660 vap->va_size = vap->va_bytes = DEV_BSIZE; 661 else 662 vap->va_size = 0; 663 if (vp->v_type != VDIR) 664 vap->va_bytes = 0; 665 vap->va_blocksize = DEV_BSIZE; 666 vap->va_type = vp->v_type; 667 668 #define fix(aa) \ 669 do { \ 670 if ((aa).tv_sec <= 3600) { \ 671 (aa).tv_sec = boottime.tv_sec; \ 672 (aa).tv_nsec = boottime.tv_usec * 1000; \ 673 } \ 674 } while (0) 675 676 if (vp->v_type != VCHR) { 677 fix(de->de_atime); 678 vap->va_atime = de->de_atime; 679 fix(de->de_mtime); 680 vap->va_mtime = de->de_mtime; 681 fix(de->de_ctime); 682 vap->va_ctime = de->de_ctime; 683 } else { 684 dev = vp->v_rdev; 685 fix(dev->si_atime); 686 vap->va_atime = dev->si_atime; 687 fix(dev->si_mtime); 688 vap->va_mtime = dev->si_mtime; 689 fix(dev->si_ctime); 690 vap->va_ctime = dev->si_ctime; 691 692 vap->va_rdev = cdev2priv(dev)->cdp_inode; 693 } 694 vap->va_gen = 0; 695 vap->va_flags = 0; 696 vap->va_filerev = 0; 697 vap->va_nlink = de->de_links; 698 vap->va_fileid = de->de_inode; 699 700 return (error); 701 } 702 703 /* ARGSUSED */ 704 static int 705 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 706 { 707 struct cdev *dev; 708 struct cdevsw *dsw; 709 struct vnode *vp; 710 struct vnode *vpold; 711 int error, i, ref; 712 const char *p; 713 struct fiodgname_arg *fgn; 714 struct file *fpop; 715 716 fpop = td->td_fpop; 717 error = devfs_fp_check(fp, &dev, &dsw, &ref); 718 if (error) 719 return (error); 720 721 if (com == FIODTYPE) { 722 *(int *)data = dsw->d_flags & D_TYPEMASK; 723 td->td_fpop = fpop; 724 dev_relthread(dev, ref); 725 return (0); 726 } else if (com == FIODGNAME) { 727 fgn = data; 728 p = devtoname(dev); 729 i = strlen(p) + 1; 730 if (i > fgn->len) 731 error = EINVAL; 732 else 733 error = copyout(p, fgn->buf, i); 734 td->td_fpop = fpop; 735 dev_relthread(dev, ref); 736 return (error); 737 } 738 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td); 739 td->td_fpop = NULL; 740 dev_relthread(dev, ref); 741 if (error == ENOIOCTL) 742 error = ENOTTY; 743 if (error == 0 && com == TIOCSCTTY) { 744 vp = fp->f_vnode; 745 746 /* Do nothing if reassigning same control tty */ 747 sx_slock(&proctree_lock); 748 if (td->td_proc->p_session->s_ttyvp == vp) { 749 sx_sunlock(&proctree_lock); 750 return (0); 751 } 752 753 vpold = td->td_proc->p_session->s_ttyvp; 754 VREF(vp); 755 SESS_LOCK(td->td_proc->p_session); 756 td->td_proc->p_session->s_ttyvp = vp; 757 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 758 SESS_UNLOCK(td->td_proc->p_session); 759 760 sx_sunlock(&proctree_lock); 761 762 /* Get rid of reference to old control tty */ 763 if (vpold) 764 vrele(vpold); 765 } 766 return (error); 767 } 768 769 /* ARGSUSED */ 770 static int 771 devfs_kqfilter_f(struct file *fp, struct knote *kn) 772 { 773 struct cdev *dev; 774 struct cdevsw *dsw; 775 int error, ref; 776 struct file *fpop; 777 struct thread *td; 778 779 td = curthread; 780 fpop = td->td_fpop; 781 error = devfs_fp_check(fp, &dev, &dsw, &ref); 782 if (error) 783 return (error); 784 error = dsw->d_kqfilter(dev, kn); 785 td->td_fpop = fpop; 786 dev_relthread(dev, ref); 787 return (error); 788 } 789 790 static inline int 791 devfs_prison_check(struct devfs_dirent *de, struct thread *td) 792 { 793 struct cdev_priv *cdp; 794 struct ucred *dcr; 795 int error; 796 797 cdp = de->de_cdp; 798 if (cdp == NULL) 799 return (0); 800 dcr = cdp->cdp_c.si_cred; 801 if (dcr == NULL) 802 return (0); 803 804 error = prison_check(td->td_ucred, dcr); 805 if (error == 0) 806 return (0); 807 /* We do, however, allow access to the controlling terminal */ 808 if (!(td->td_proc->p_flag & P_CONTROLT)) 809 return (error); 810 if (td->td_proc->p_session->s_ttydp == cdp) 811 return (0); 812 return (error); 813 } 814 815 static int 816 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 817 { 818 struct componentname *cnp; 819 struct vnode *dvp, **vpp; 820 struct thread *td; 821 struct devfs_dirent *de, *dd; 822 struct devfs_dirent **dde; 823 struct devfs_mount *dmp; 824 struct cdev *cdev; 825 int error, flags, nameiop, dvplocked; 826 char specname[SPECNAMELEN + 1], *pname; 827 828 cnp = ap->a_cnp; 829 vpp = ap->a_vpp; 830 dvp = ap->a_dvp; 831 pname = cnp->cn_nameptr; 832 td = cnp->cn_thread; 833 flags = cnp->cn_flags; 834 nameiop = cnp->cn_nameiop; 835 dmp = VFSTODEVFS(dvp->v_mount); 836 dd = dvp->v_data; 837 *vpp = NULLVP; 838 839 if ((flags & ISLASTCN) && nameiop == RENAME) 840 return (EOPNOTSUPP); 841 842 if (dvp->v_type != VDIR) 843 return (ENOTDIR); 844 845 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 846 return (EIO); 847 848 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 849 if (error) 850 return (error); 851 852 if (cnp->cn_namelen == 1 && *pname == '.') { 853 if ((flags & ISLASTCN) && nameiop != LOOKUP) 854 return (EINVAL); 855 *vpp = dvp; 856 VREF(dvp); 857 return (0); 858 } 859 860 if (flags & ISDOTDOT) { 861 if ((flags & ISLASTCN) && nameiop != LOOKUP) 862 return (EINVAL); 863 de = devfs_parent_dirent(dd); 864 if (de == NULL) 865 return (ENOENT); 866 dvplocked = VOP_ISLOCKED(dvp); 867 VOP_UNLOCK(dvp, 0); 868 error = devfs_allocv(de, dvp->v_mount, 869 cnp->cn_lkflags & LK_TYPE_MASK, vpp); 870 *dm_unlock = 0; 871 vn_lock(dvp, dvplocked | LK_RETRY); 872 return (error); 873 } 874 875 dd = dvp->v_data; 876 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 877 while (de == NULL) { /* While(...) so we can use break */ 878 879 if (nameiop == DELETE) 880 return (ENOENT); 881 882 /* 883 * OK, we didn't have an entry for the name we were asked for 884 * so we try to see if anybody can create it on demand. 885 */ 886 pname = devfs_fqpn(specname, dmp, dd, cnp); 887 if (pname == NULL) 888 break; 889 890 cdev = NULL; 891 DEVFS_DMP_HOLD(dmp); 892 sx_xunlock(&dmp->dm_lock); 893 sx_slock(&clone_drain_lock); 894 EVENTHANDLER_INVOKE(dev_clone, 895 td->td_ucred, pname, strlen(pname), &cdev); 896 sx_sunlock(&clone_drain_lock); 897 898 if (cdev == NULL) 899 sx_xlock(&dmp->dm_lock); 900 else if (devfs_populate_vp(dvp) != 0) { 901 *dm_unlock = 0; 902 sx_xlock(&dmp->dm_lock); 903 if (DEVFS_DMP_DROP(dmp)) { 904 sx_xunlock(&dmp->dm_lock); 905 devfs_unmount_final(dmp); 906 } else 907 sx_xunlock(&dmp->dm_lock); 908 dev_rel(cdev); 909 return (ENOENT); 910 } 911 if (DEVFS_DMP_DROP(dmp)) { 912 *dm_unlock = 0; 913 sx_xunlock(&dmp->dm_lock); 914 devfs_unmount_final(dmp); 915 if (cdev != NULL) 916 dev_rel(cdev); 917 return (ENOENT); 918 } 919 920 if (cdev == NULL) 921 break; 922 923 dev_lock(); 924 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 925 if (dde != NULL && *dde != NULL) 926 de = *dde; 927 dev_unlock(); 928 dev_rel(cdev); 929 break; 930 } 931 932 if (de == NULL || de->de_flags & DE_WHITEOUT) { 933 if ((nameiop == CREATE || nameiop == RENAME) && 934 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 935 cnp->cn_flags |= SAVENAME; 936 return (EJUSTRETURN); 937 } 938 return (ENOENT); 939 } 940 941 if (devfs_prison_check(de, td)) 942 return (ENOENT); 943 944 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 945 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 946 if (error) 947 return (error); 948 if (*vpp == dvp) { 949 VREF(dvp); 950 *vpp = dvp; 951 return (0); 952 } 953 } 954 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK, 955 vpp); 956 *dm_unlock = 0; 957 return (error); 958 } 959 960 static int 961 devfs_lookup(struct vop_lookup_args *ap) 962 { 963 int j; 964 struct devfs_mount *dmp; 965 int dm_unlock; 966 967 if (devfs_populate_vp(ap->a_dvp) != 0) 968 return (ENOTDIR); 969 970 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 971 dm_unlock = 1; 972 j = devfs_lookupx(ap, &dm_unlock); 973 if (dm_unlock == 1) 974 sx_xunlock(&dmp->dm_lock); 975 return (j); 976 } 977 978 static int 979 devfs_mknod(struct vop_mknod_args *ap) 980 { 981 struct componentname *cnp; 982 struct vnode *dvp, **vpp; 983 struct devfs_dirent *dd, *de; 984 struct devfs_mount *dmp; 985 int error; 986 987 /* 988 * The only type of node we should be creating here is a 989 * character device, for anything else return EOPNOTSUPP. 990 */ 991 if (ap->a_vap->va_type != VCHR) 992 return (EOPNOTSUPP); 993 dvp = ap->a_dvp; 994 dmp = VFSTODEVFS(dvp->v_mount); 995 996 cnp = ap->a_cnp; 997 vpp = ap->a_vpp; 998 dd = dvp->v_data; 999 1000 error = ENOENT; 1001 sx_xlock(&dmp->dm_lock); 1002 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1003 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1004 continue; 1005 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1006 de->de_dirent->d_namlen) != 0) 1007 continue; 1008 if (de->de_flags & DE_WHITEOUT) 1009 break; 1010 goto notfound; 1011 } 1012 if (de == NULL) 1013 goto notfound; 1014 de->de_flags &= ~DE_WHITEOUT; 1015 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1016 return (error); 1017 notfound: 1018 sx_xunlock(&dmp->dm_lock); 1019 return (error); 1020 } 1021 1022 /* ARGSUSED */ 1023 static int 1024 devfs_open(struct vop_open_args *ap) 1025 { 1026 struct thread *td = ap->a_td; 1027 struct vnode *vp = ap->a_vp; 1028 struct cdev *dev = vp->v_rdev; 1029 struct file *fp = ap->a_fp; 1030 int error, ref, vlocked; 1031 struct cdevsw *dsw; 1032 struct file *fpop; 1033 1034 if (vp->v_type == VBLK) 1035 return (ENXIO); 1036 1037 if (dev == NULL) 1038 return (ENXIO); 1039 1040 /* Make this field valid before any I/O in d_open. */ 1041 if (dev->si_iosize_max == 0) 1042 dev->si_iosize_max = DFLTPHYS; 1043 1044 dsw = dev_refthread(dev, &ref); 1045 if (dsw == NULL) 1046 return (ENXIO); 1047 1048 vlocked = VOP_ISLOCKED(vp); 1049 VOP_UNLOCK(vp, 0); 1050 1051 fpop = td->td_fpop; 1052 td->td_fpop = fp; 1053 if (fp != NULL) { 1054 fp->f_data = dev; 1055 fp->f_vnode = vp; 1056 } 1057 if (dsw->d_fdopen != NULL) 1058 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1059 else 1060 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1061 td->td_fpop = fpop; 1062 1063 vn_lock(vp, vlocked | LK_RETRY); 1064 dev_relthread(dev, ref); 1065 if (error) 1066 return (error); 1067 1068 #if 0 /* /dev/console */ 1069 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1070 #else 1071 if (fp == NULL) 1072 return (error); 1073 #endif 1074 if (fp->f_ops == &badfileops) 1075 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1076 return (error); 1077 } 1078 1079 static int 1080 devfs_pathconf(struct vop_pathconf_args *ap) 1081 { 1082 1083 switch (ap->a_name) { 1084 case _PC_MAC_PRESENT: 1085 #ifdef MAC 1086 /* 1087 * If MAC is enabled, devfs automatically supports 1088 * trivial non-persistant label storage. 1089 */ 1090 *ap->a_retval = 1; 1091 #else 1092 *ap->a_retval = 0; 1093 #endif 1094 return (0); 1095 default: 1096 return (vop_stdpathconf(ap)); 1097 } 1098 /* NOTREACHED */ 1099 } 1100 1101 /* ARGSUSED */ 1102 static int 1103 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1104 { 1105 struct cdev *dev; 1106 struct cdevsw *dsw; 1107 int error, ref; 1108 struct file *fpop; 1109 1110 fpop = td->td_fpop; 1111 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1112 if (error) 1113 return (poll_no_poll(events)); 1114 error = dsw->d_poll(dev, events, td); 1115 td->td_fpop = fpop; 1116 dev_relthread(dev, ref); 1117 return(error); 1118 } 1119 1120 /* 1121 * Print out the contents of a special device vnode. 1122 */ 1123 static int 1124 devfs_print(struct vop_print_args *ap) 1125 { 1126 1127 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1128 return (0); 1129 } 1130 1131 /* ARGSUSED */ 1132 static int 1133 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td) 1134 { 1135 struct cdev *dev; 1136 int ioflag, error, ref, resid; 1137 struct cdevsw *dsw; 1138 struct file *fpop; 1139 1140 fpop = td->td_fpop; 1141 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1142 if (error) 1143 return (error); 1144 resid = uio->uio_resid; 1145 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1146 if (ioflag & O_DIRECT) 1147 ioflag |= IO_DIRECT; 1148 1149 if ((flags & FOF_OFFSET) == 0) 1150 uio->uio_offset = fp->f_offset; 1151 1152 error = dsw->d_read(dev, uio, ioflag); 1153 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1154 vfs_timestamp(&dev->si_atime); 1155 td->td_fpop = fpop; 1156 dev_relthread(dev, ref); 1157 1158 if ((flags & FOF_OFFSET) == 0) 1159 fp->f_offset = uio->uio_offset; 1160 fp->f_nextoff = uio->uio_offset; 1161 return (error); 1162 } 1163 1164 static int 1165 devfs_readdir(struct vop_readdir_args *ap) 1166 { 1167 int error; 1168 struct uio *uio; 1169 struct dirent *dp; 1170 struct devfs_dirent *dd; 1171 struct devfs_dirent *de; 1172 struct devfs_mount *dmp; 1173 off_t off; 1174 int *tmp_ncookies = NULL; 1175 1176 if (ap->a_vp->v_type != VDIR) 1177 return (ENOTDIR); 1178 1179 uio = ap->a_uio; 1180 if (uio->uio_offset < 0) 1181 return (EINVAL); 1182 1183 /* 1184 * XXX: This is a temporary hack to get around this filesystem not 1185 * supporting cookies. We store the location of the ncookies pointer 1186 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1187 * and set the number of cookies to 0. We then set the pointer to 1188 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1189 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1190 * pointer to its original location before returning to the caller. 1191 */ 1192 if (ap->a_ncookies != NULL) { 1193 tmp_ncookies = ap->a_ncookies; 1194 *ap->a_ncookies = 0; 1195 ap->a_ncookies = NULL; 1196 } 1197 1198 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1199 if (devfs_populate_vp(ap->a_vp) != 0) { 1200 if (tmp_ncookies != NULL) 1201 ap->a_ncookies = tmp_ncookies; 1202 return (EIO); 1203 } 1204 error = 0; 1205 de = ap->a_vp->v_data; 1206 off = 0; 1207 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1208 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1209 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1210 continue; 1211 if (devfs_prison_check(dd, uio->uio_td)) 1212 continue; 1213 if (dd->de_dirent->d_type == DT_DIR) 1214 de = dd->de_dir; 1215 else 1216 de = dd; 1217 dp = dd->de_dirent; 1218 if (dp->d_reclen > uio->uio_resid) 1219 break; 1220 dp->d_fileno = de->de_inode; 1221 if (off >= uio->uio_offset) { 1222 error = vfs_read_dirent(ap, dp, off); 1223 if (error) 1224 break; 1225 } 1226 off += dp->d_reclen; 1227 } 1228 sx_xunlock(&dmp->dm_lock); 1229 uio->uio_offset = off; 1230 1231 /* 1232 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1233 * place. 1234 */ 1235 if (tmp_ncookies != NULL) 1236 ap->a_ncookies = tmp_ncookies; 1237 1238 return (error); 1239 } 1240 1241 static int 1242 devfs_readlink(struct vop_readlink_args *ap) 1243 { 1244 struct devfs_dirent *de; 1245 1246 de = ap->a_vp->v_data; 1247 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1248 } 1249 1250 static int 1251 devfs_reclaim(struct vop_reclaim_args *ap) 1252 { 1253 struct vnode *vp = ap->a_vp; 1254 struct devfs_dirent *de; 1255 struct cdev *dev; 1256 1257 mtx_lock(&devfs_de_interlock); 1258 de = vp->v_data; 1259 if (de != NULL) { 1260 de->de_vnode = NULL; 1261 vp->v_data = NULL; 1262 } 1263 mtx_unlock(&devfs_de_interlock); 1264 1265 vnode_destroy_vobject(vp); 1266 1267 VI_LOCK(vp); 1268 dev_lock(); 1269 dev = vp->v_rdev; 1270 vp->v_rdev = NULL; 1271 1272 if (dev == NULL) { 1273 dev_unlock(); 1274 VI_UNLOCK(vp); 1275 return (0); 1276 } 1277 1278 dev->si_usecount -= vp->v_usecount; 1279 dev_unlock(); 1280 VI_UNLOCK(vp); 1281 dev_rel(dev); 1282 return (0); 1283 } 1284 1285 static int 1286 devfs_remove(struct vop_remove_args *ap) 1287 { 1288 struct vnode *dvp = ap->a_dvp; 1289 struct vnode *vp = ap->a_vp; 1290 struct devfs_dirent *dd; 1291 struct devfs_dirent *de, *de_covered; 1292 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1293 1294 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1295 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1296 1297 sx_xlock(&dmp->dm_lock); 1298 dd = ap->a_dvp->v_data; 1299 de = vp->v_data; 1300 if (de->de_cdp == NULL) { 1301 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1302 if (de->de_dirent->d_type == DT_LNK) { 1303 de_covered = devfs_find(dd, de->de_dirent->d_name, 1304 de->de_dirent->d_namlen, 0); 1305 if (de_covered != NULL) 1306 de_covered->de_flags &= ~DE_COVERED; 1307 } 1308 /* We need to unlock dvp because devfs_delete() may lock it. */ 1309 VOP_UNLOCK(vp, 0); 1310 if (dvp != vp) 1311 VOP_UNLOCK(dvp, 0); 1312 devfs_delete(dmp, de, 0); 1313 sx_xunlock(&dmp->dm_lock); 1314 if (dvp != vp) 1315 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1316 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1317 } else { 1318 de->de_flags |= DE_WHITEOUT; 1319 sx_xunlock(&dmp->dm_lock); 1320 } 1321 return (0); 1322 } 1323 1324 /* 1325 * Revoke is called on a tty when a terminal session ends. The vnode 1326 * is orphaned by setting v_op to deadfs so we need to let go of it 1327 * as well so that we create a new one next time around. 1328 * 1329 */ 1330 static int 1331 devfs_revoke(struct vop_revoke_args *ap) 1332 { 1333 struct vnode *vp = ap->a_vp, *vp2; 1334 struct cdev *dev; 1335 struct cdev_priv *cdp; 1336 struct devfs_dirent *de; 1337 int i; 1338 1339 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1340 1341 dev = vp->v_rdev; 1342 cdp = cdev2priv(dev); 1343 1344 dev_lock(); 1345 cdp->cdp_inuse++; 1346 dev_unlock(); 1347 1348 vhold(vp); 1349 vgone(vp); 1350 vdrop(vp); 1351 1352 VOP_UNLOCK(vp,0); 1353 loop: 1354 for (;;) { 1355 mtx_lock(&devfs_de_interlock); 1356 dev_lock(); 1357 vp2 = NULL; 1358 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1359 de = cdp->cdp_dirents[i]; 1360 if (de == NULL) 1361 continue; 1362 1363 vp2 = de->de_vnode; 1364 if (vp2 != NULL) { 1365 dev_unlock(); 1366 VI_LOCK(vp2); 1367 mtx_unlock(&devfs_de_interlock); 1368 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1369 curthread)) 1370 goto loop; 1371 vhold(vp2); 1372 vgone(vp2); 1373 vdrop(vp2); 1374 vput(vp2); 1375 break; 1376 } 1377 } 1378 if (vp2 != NULL) { 1379 continue; 1380 } 1381 dev_unlock(); 1382 mtx_unlock(&devfs_de_interlock); 1383 break; 1384 } 1385 dev_lock(); 1386 cdp->cdp_inuse--; 1387 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1388 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1389 dev_unlock(); 1390 dev_rel(&cdp->cdp_c); 1391 } else 1392 dev_unlock(); 1393 1394 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1395 return (0); 1396 } 1397 1398 static int 1399 devfs_rioctl(struct vop_ioctl_args *ap) 1400 { 1401 struct vnode *vp; 1402 struct devfs_mount *dmp; 1403 int error; 1404 1405 vp = ap->a_vp; 1406 vn_lock(vp, LK_SHARED | LK_RETRY); 1407 if (vp->v_iflag & VI_DOOMED) { 1408 VOP_UNLOCK(vp, 0); 1409 return (EBADF); 1410 } 1411 dmp = VFSTODEVFS(vp->v_mount); 1412 sx_xlock(&dmp->dm_lock); 1413 VOP_UNLOCK(vp, 0); 1414 DEVFS_DMP_HOLD(dmp); 1415 devfs_populate(dmp); 1416 if (DEVFS_DMP_DROP(dmp)) { 1417 sx_xunlock(&dmp->dm_lock); 1418 devfs_unmount_final(dmp); 1419 return (ENOENT); 1420 } 1421 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1422 sx_xunlock(&dmp->dm_lock); 1423 return (error); 1424 } 1425 1426 static int 1427 devfs_rread(struct vop_read_args *ap) 1428 { 1429 1430 if (ap->a_vp->v_type != VDIR) 1431 return (EINVAL); 1432 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1433 } 1434 1435 static int 1436 devfs_setattr(struct vop_setattr_args *ap) 1437 { 1438 struct devfs_dirent *de; 1439 struct vattr *vap; 1440 struct vnode *vp; 1441 struct thread *td; 1442 int c, error; 1443 uid_t uid; 1444 gid_t gid; 1445 1446 vap = ap->a_vap; 1447 vp = ap->a_vp; 1448 td = curthread; 1449 if ((vap->va_type != VNON) || 1450 (vap->va_nlink != VNOVAL) || 1451 (vap->va_fsid != VNOVAL) || 1452 (vap->va_fileid != VNOVAL) || 1453 (vap->va_blocksize != VNOVAL) || 1454 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1455 (vap->va_rdev != VNOVAL) || 1456 ((int)vap->va_bytes != VNOVAL) || 1457 (vap->va_gen != VNOVAL)) { 1458 return (EINVAL); 1459 } 1460 1461 de = vp->v_data; 1462 if (vp->v_type == VDIR) 1463 de = de->de_dir; 1464 1465 error = c = 0; 1466 if (vap->va_uid == (uid_t)VNOVAL) 1467 uid = de->de_uid; 1468 else 1469 uid = vap->va_uid; 1470 if (vap->va_gid == (gid_t)VNOVAL) 1471 gid = de->de_gid; 1472 else 1473 gid = vap->va_gid; 1474 if (uid != de->de_uid || gid != de->de_gid) { 1475 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1476 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1477 error = priv_check(td, PRIV_VFS_CHOWN); 1478 if (error) 1479 return (error); 1480 } 1481 de->de_uid = uid; 1482 de->de_gid = gid; 1483 c = 1; 1484 } 1485 1486 if (vap->va_mode != (mode_t)VNOVAL) { 1487 if (ap->a_cred->cr_uid != de->de_uid) { 1488 error = priv_check(td, PRIV_VFS_ADMIN); 1489 if (error) 1490 return (error); 1491 } 1492 de->de_mode = vap->va_mode; 1493 c = 1; 1494 } 1495 1496 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1497 /* See the comment in ufs_vnops::ufs_setattr(). */ 1498 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) && 1499 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 1500 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td)))) 1501 return (error); 1502 if (vap->va_atime.tv_sec != VNOVAL) { 1503 if (vp->v_type == VCHR) 1504 vp->v_rdev->si_atime = vap->va_atime; 1505 else 1506 de->de_atime = vap->va_atime; 1507 } 1508 if (vap->va_mtime.tv_sec != VNOVAL) { 1509 if (vp->v_type == VCHR) 1510 vp->v_rdev->si_mtime = vap->va_mtime; 1511 else 1512 de->de_mtime = vap->va_mtime; 1513 } 1514 c = 1; 1515 } 1516 1517 if (c) { 1518 if (vp->v_type == VCHR) 1519 vfs_timestamp(&vp->v_rdev->si_ctime); 1520 else 1521 vfs_timestamp(&de->de_mtime); 1522 } 1523 return (0); 1524 } 1525 1526 #ifdef MAC 1527 static int 1528 devfs_setlabel(struct vop_setlabel_args *ap) 1529 { 1530 struct vnode *vp; 1531 struct devfs_dirent *de; 1532 1533 vp = ap->a_vp; 1534 de = vp->v_data; 1535 1536 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1537 mac_devfs_update(vp->v_mount, de, vp); 1538 1539 return (0); 1540 } 1541 #endif 1542 1543 static int 1544 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1545 { 1546 1547 return (vnops.fo_stat(fp, sb, cred, td)); 1548 } 1549 1550 static int 1551 devfs_symlink(struct vop_symlink_args *ap) 1552 { 1553 int i, error; 1554 struct devfs_dirent *dd; 1555 struct devfs_dirent *de, *de_covered, *de_dotdot; 1556 struct devfs_mount *dmp; 1557 1558 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1559 if (error) 1560 return(error); 1561 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1562 if (devfs_populate_vp(ap->a_dvp) != 0) 1563 return (ENOENT); 1564 1565 dd = ap->a_dvp->v_data; 1566 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1567 de->de_flags = DE_USER; 1568 de->de_uid = 0; 1569 de->de_gid = 0; 1570 de->de_mode = 0755; 1571 de->de_inode = alloc_unr(devfs_inos); 1572 de->de_dir = dd; 1573 de->de_dirent->d_type = DT_LNK; 1574 i = strlen(ap->a_target) + 1; 1575 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1576 bcopy(ap->a_target, de->de_symlink, i); 1577 #ifdef MAC 1578 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1579 #endif 1580 de_covered = devfs_find(dd, de->de_dirent->d_name, 1581 de->de_dirent->d_namlen, 0); 1582 if (de_covered != NULL) { 1583 if ((de_covered->de_flags & DE_USER) != 0) { 1584 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1585 sx_xunlock(&dmp->dm_lock); 1586 return (EEXIST); 1587 } 1588 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1589 ("devfs_symlink: entry %p already covered", de_covered)); 1590 de_covered->de_flags |= DE_COVERED; 1591 } 1592 1593 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1594 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1595 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1596 devfs_dir_ref_de(dmp, dd); 1597 devfs_rules_apply(dmp, de); 1598 1599 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1600 } 1601 1602 static int 1603 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1604 { 1605 1606 return (vnops.fo_truncate(fp, length, cred, td)); 1607 } 1608 1609 /* ARGSUSED */ 1610 static int 1611 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td) 1612 { 1613 struct cdev *dev; 1614 int error, ioflag, ref, resid; 1615 struct cdevsw *dsw; 1616 struct file *fpop; 1617 1618 fpop = td->td_fpop; 1619 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1620 if (error) 1621 return (error); 1622 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1623 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1624 if (ioflag & O_DIRECT) 1625 ioflag |= IO_DIRECT; 1626 if ((flags & FOF_OFFSET) == 0) 1627 uio->uio_offset = fp->f_offset; 1628 1629 resid = uio->uio_resid; 1630 1631 error = dsw->d_write(dev, uio, ioflag); 1632 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1633 vfs_timestamp(&dev->si_ctime); 1634 dev->si_mtime = dev->si_ctime; 1635 } 1636 td->td_fpop = fpop; 1637 dev_relthread(dev, ref); 1638 1639 if ((flags & FOF_OFFSET) == 0) 1640 fp->f_offset = uio->uio_offset; 1641 fp->f_nextoff = uio->uio_offset; 1642 return (error); 1643 } 1644 1645 dev_t 1646 dev2udev(struct cdev *x) 1647 { 1648 if (x == NULL) 1649 return (NODEV); 1650 return (cdev2priv(x)->cdp_inode); 1651 } 1652 1653 static struct fileops devfs_ops_f = { 1654 .fo_read = devfs_read_f, 1655 .fo_write = devfs_write_f, 1656 .fo_truncate = devfs_truncate_f, 1657 .fo_ioctl = devfs_ioctl_f, 1658 .fo_poll = devfs_poll_f, 1659 .fo_kqfilter = devfs_kqfilter_f, 1660 .fo_stat = devfs_stat_f, 1661 .fo_close = devfs_close_f, 1662 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1663 }; 1664 1665 static struct vop_vector devfs_vnodeops = { 1666 .vop_default = &default_vnodeops, 1667 1668 .vop_access = devfs_access, 1669 .vop_getattr = devfs_getattr, 1670 .vop_ioctl = devfs_rioctl, 1671 .vop_lookup = devfs_lookup, 1672 .vop_mknod = devfs_mknod, 1673 .vop_pathconf = devfs_pathconf, 1674 .vop_read = devfs_rread, 1675 .vop_readdir = devfs_readdir, 1676 .vop_readlink = devfs_readlink, 1677 .vop_reclaim = devfs_reclaim, 1678 .vop_remove = devfs_remove, 1679 .vop_revoke = devfs_revoke, 1680 .vop_setattr = devfs_setattr, 1681 #ifdef MAC 1682 .vop_setlabel = devfs_setlabel, 1683 #endif 1684 .vop_symlink = devfs_symlink, 1685 .vop_vptocnp = devfs_vptocnp, 1686 }; 1687 1688 static struct vop_vector devfs_specops = { 1689 .vop_default = &default_vnodeops, 1690 1691 .vop_access = devfs_access, 1692 .vop_bmap = VOP_PANIC, 1693 .vop_close = devfs_close, 1694 .vop_create = VOP_PANIC, 1695 .vop_fsync = devfs_fsync, 1696 .vop_getattr = devfs_getattr, 1697 .vop_link = VOP_PANIC, 1698 .vop_mkdir = VOP_PANIC, 1699 .vop_mknod = VOP_PANIC, 1700 .vop_open = devfs_open, 1701 .vop_pathconf = devfs_pathconf, 1702 .vop_print = devfs_print, 1703 .vop_read = VOP_PANIC, 1704 .vop_readdir = VOP_PANIC, 1705 .vop_readlink = VOP_PANIC, 1706 .vop_reallocblks = VOP_PANIC, 1707 .vop_reclaim = devfs_reclaim, 1708 .vop_remove = devfs_remove, 1709 .vop_rename = VOP_PANIC, 1710 .vop_revoke = devfs_revoke, 1711 .vop_rmdir = VOP_PANIC, 1712 .vop_setattr = devfs_setattr, 1713 #ifdef MAC 1714 .vop_setlabel = devfs_setlabel, 1715 #endif 1716 .vop_strategy = VOP_PANIC, 1717 .vop_symlink = VOP_PANIC, 1718 .vop_vptocnp = devfs_vptocnp, 1719 .vop_write = VOP_PANIC, 1720 }; 1721 1722 /* 1723 * Our calling convention to the device drivers used to be that we passed 1724 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1725 * flags instead since that's what open(), close() and ioctl() takes and 1726 * we don't really want vnode.h in device drivers. 1727 * We solved the source compatibility by redefining some vnode flags to 1728 * be the same as the fcntl ones and by sending down the bitwise OR of 1729 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1730 * pulls the rug out under this. 1731 */ 1732 CTASSERT(O_NONBLOCK == IO_NDELAY); 1733 CTASSERT(O_FSYNC == IO_SYNC); 1734