1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2000-2004 5 * Poul-Henning Kamp. All rights reserved. 6 * Copyright (c) 1989, 1992-1993, 1995 7 * The Regents of the University of California. All rights reserved. 8 * 9 * This code is derived from software donated to Berkeley by 10 * Jan-Simon Pendry. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95 34 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43 35 * 36 * $FreeBSD$ 37 */ 38 39 /* 40 * TODO: 41 * mkdir: want it ? 42 */ 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/conf.h> 47 #include <sys/dirent.h> 48 #include <sys/eventhandler.h> 49 #include <sys/fcntl.h> 50 #include <sys/file.h> 51 #include <sys/filedesc.h> 52 #include <sys/filio.h> 53 #include <sys/jail.h> 54 #include <sys/kernel.h> 55 #include <sys/limits.h> 56 #include <sys/lock.h> 57 #include <sys/malloc.h> 58 #include <sys/mman.h> 59 #include <sys/mount.h> 60 #include <sys/namei.h> 61 #include <sys/priv.h> 62 #include <sys/proc.h> 63 #include <sys/stat.h> 64 #include <sys/sx.h> 65 #include <sys/sysctl.h> 66 #include <sys/time.h> 67 #include <sys/ttycom.h> 68 #include <sys/unistd.h> 69 #include <sys/vnode.h> 70 71 static struct vop_vector devfs_vnodeops; 72 static struct vop_vector devfs_specops; 73 static struct fileops devfs_ops_f; 74 75 #include <fs/devfs/devfs.h> 76 #include <fs/devfs/devfs_int.h> 77 78 #include <security/mac/mac_framework.h> 79 80 #include <vm/vm.h> 81 #include <vm/vm_extern.h> 82 #include <vm/vm_object.h> 83 84 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data"); 85 86 struct mtx devfs_de_interlock; 87 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); 88 struct sx clone_drain_lock; 89 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock"); 90 struct mtx cdevpriv_mtx; 91 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF); 92 93 SYSCTL_DECL(_vfs_devfs); 94 95 static int devfs_dotimes; 96 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW, 97 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision"); 98 99 /* 100 * Update devfs node timestamp. Note that updates are unlocked and 101 * stat(2) could see partially updated times. 102 */ 103 static void 104 devfs_timestamp(struct timespec *tsp) 105 { 106 time_t ts; 107 108 if (devfs_dotimes) { 109 vfs_timestamp(tsp); 110 } else { 111 ts = time_second; 112 if (tsp->tv_sec != ts) { 113 tsp->tv_sec = ts; 114 tsp->tv_nsec = 0; 115 } 116 } 117 } 118 119 static int 120 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp, 121 int *ref) 122 { 123 124 *dswp = devvn_refthread(fp->f_vnode, devp, ref); 125 if (*devp != fp->f_data) { 126 if (*dswp != NULL) 127 dev_relthread(*devp, *ref); 128 return (ENXIO); 129 } 130 KASSERT((*devp)->si_refcount > 0, 131 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); 132 if (*dswp == NULL) 133 return (ENXIO); 134 curthread->td_fpop = fp; 135 return (0); 136 } 137 138 int 139 devfs_get_cdevpriv(void **datap) 140 { 141 struct file *fp; 142 struct cdev_privdata *p; 143 int error; 144 145 fp = curthread->td_fpop; 146 if (fp == NULL) 147 return (EBADF); 148 p = fp->f_cdevpriv; 149 if (p != NULL) { 150 error = 0; 151 *datap = p->cdpd_data; 152 } else 153 error = ENOENT; 154 return (error); 155 } 156 157 int 158 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr) 159 { 160 struct file *fp; 161 struct cdev_priv *cdp; 162 struct cdev_privdata *p; 163 int error; 164 165 fp = curthread->td_fpop; 166 if (fp == NULL) 167 return (ENOENT); 168 cdp = cdev2priv((struct cdev *)fp->f_data); 169 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK); 170 p->cdpd_data = priv; 171 p->cdpd_dtr = priv_dtr; 172 p->cdpd_fp = fp; 173 mtx_lock(&cdevpriv_mtx); 174 if (fp->f_cdevpriv == NULL) { 175 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list); 176 fp->f_cdevpriv = p; 177 mtx_unlock(&cdevpriv_mtx); 178 error = 0; 179 } else { 180 mtx_unlock(&cdevpriv_mtx); 181 free(p, M_CDEVPDATA); 182 error = EBUSY; 183 } 184 return (error); 185 } 186 187 void 188 devfs_destroy_cdevpriv(struct cdev_privdata *p) 189 { 190 191 mtx_assert(&cdevpriv_mtx, MA_OWNED); 192 KASSERT(p->cdpd_fp->f_cdevpriv == p, 193 ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p)); 194 p->cdpd_fp->f_cdevpriv = NULL; 195 LIST_REMOVE(p, cdpd_list); 196 mtx_unlock(&cdevpriv_mtx); 197 (p->cdpd_dtr)(p->cdpd_data); 198 free(p, M_CDEVPDATA); 199 } 200 201 static void 202 devfs_fpdrop(struct file *fp) 203 { 204 struct cdev_privdata *p; 205 206 mtx_lock(&cdevpriv_mtx); 207 if ((p = fp->f_cdevpriv) == NULL) { 208 mtx_unlock(&cdevpriv_mtx); 209 return; 210 } 211 devfs_destroy_cdevpriv(p); 212 } 213 214 void 215 devfs_clear_cdevpriv(void) 216 { 217 struct file *fp; 218 219 fp = curthread->td_fpop; 220 if (fp == NULL) 221 return; 222 devfs_fpdrop(fp); 223 } 224 225 /* 226 * On success devfs_populate_vp() returns with dmp->dm_lock held. 227 */ 228 static int 229 devfs_populate_vp(struct vnode *vp) 230 { 231 struct devfs_dirent *de; 232 struct devfs_mount *dmp; 233 int locked; 234 235 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp"); 236 237 dmp = VFSTODEVFS(vp->v_mount); 238 locked = VOP_ISLOCKED(vp); 239 240 sx_xlock(&dmp->dm_lock); 241 DEVFS_DMP_HOLD(dmp); 242 243 /* Can't call devfs_populate() with the vnode lock held. */ 244 VOP_UNLOCK(vp, 0); 245 devfs_populate(dmp); 246 247 sx_xunlock(&dmp->dm_lock); 248 vn_lock(vp, locked | LK_RETRY); 249 sx_xlock(&dmp->dm_lock); 250 if (DEVFS_DMP_DROP(dmp)) { 251 sx_xunlock(&dmp->dm_lock); 252 devfs_unmount_final(dmp); 253 return (ERESTART); 254 } 255 if ((vp->v_iflag & VI_DOOMED) != 0) { 256 sx_xunlock(&dmp->dm_lock); 257 return (ERESTART); 258 } 259 de = vp->v_data; 260 KASSERT(de != NULL, 261 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed")); 262 if ((de->de_flags & DE_DOOMED) != 0) { 263 sx_xunlock(&dmp->dm_lock); 264 return (ERESTART); 265 } 266 267 return (0); 268 } 269 270 static int 271 devfs_vptocnp(struct vop_vptocnp_args *ap) 272 { 273 struct vnode *vp = ap->a_vp; 274 struct vnode **dvp = ap->a_vpp; 275 struct devfs_mount *dmp; 276 char *buf = ap->a_buf; 277 int *buflen = ap->a_buflen; 278 struct devfs_dirent *dd, *de; 279 int i, error; 280 281 dmp = VFSTODEVFS(vp->v_mount); 282 283 error = devfs_populate_vp(vp); 284 if (error != 0) 285 return (error); 286 287 i = *buflen; 288 dd = vp->v_data; 289 290 if (vp->v_type == VCHR) { 291 i -= strlen(dd->de_cdp->cdp_c.si_name); 292 if (i < 0) { 293 error = ENOMEM; 294 goto finished; 295 } 296 bcopy(dd->de_cdp->cdp_c.si_name, buf + i, 297 strlen(dd->de_cdp->cdp_c.si_name)); 298 de = dd->de_dir; 299 } else if (vp->v_type == VDIR) { 300 if (dd == dmp->dm_rootdir) { 301 *dvp = vp; 302 vref(*dvp); 303 goto finished; 304 } 305 i -= dd->de_dirent->d_namlen; 306 if (i < 0) { 307 error = ENOMEM; 308 goto finished; 309 } 310 bcopy(dd->de_dirent->d_name, buf + i, 311 dd->de_dirent->d_namlen); 312 de = dd; 313 } else { 314 error = ENOENT; 315 goto finished; 316 } 317 *buflen = i; 318 de = devfs_parent_dirent(de); 319 if (de == NULL) { 320 error = ENOENT; 321 goto finished; 322 } 323 mtx_lock(&devfs_de_interlock); 324 *dvp = de->de_vnode; 325 if (*dvp != NULL) { 326 VI_LOCK(*dvp); 327 mtx_unlock(&devfs_de_interlock); 328 vholdl(*dvp); 329 VI_UNLOCK(*dvp); 330 vref(*dvp); 331 vdrop(*dvp); 332 } else { 333 mtx_unlock(&devfs_de_interlock); 334 error = ENOENT; 335 } 336 finished: 337 sx_xunlock(&dmp->dm_lock); 338 return (error); 339 } 340 341 /* 342 * Construct the fully qualified path name relative to the mountpoint. 343 * If a NULL cnp is provided, no '/' is appended to the resulting path. 344 */ 345 char * 346 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd, 347 struct componentname *cnp) 348 { 349 int i; 350 struct devfs_dirent *de; 351 352 sx_assert(&dmp->dm_lock, SA_LOCKED); 353 354 i = SPECNAMELEN; 355 buf[i] = '\0'; 356 if (cnp != NULL) 357 i -= cnp->cn_namelen; 358 if (i < 0) 359 return (NULL); 360 if (cnp != NULL) 361 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); 362 de = dd; 363 while (de != dmp->dm_rootdir) { 364 if (cnp != NULL || i < SPECNAMELEN) { 365 i--; 366 if (i < 0) 367 return (NULL); 368 buf[i] = '/'; 369 } 370 i -= de->de_dirent->d_namlen; 371 if (i < 0) 372 return (NULL); 373 bcopy(de->de_dirent->d_name, buf + i, 374 de->de_dirent->d_namlen); 375 de = devfs_parent_dirent(de); 376 if (de == NULL) 377 return (NULL); 378 } 379 return (buf + i); 380 } 381 382 static int 383 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp, 384 struct devfs_dirent *de) 385 { 386 int not_found; 387 388 not_found = 0; 389 if (de->de_flags & DE_DOOMED) 390 not_found = 1; 391 if (DEVFS_DE_DROP(de)) { 392 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed")); 393 devfs_dirent_free(de); 394 } 395 if (DEVFS_DMP_DROP(dmp)) { 396 KASSERT(not_found == 1, 397 ("DEVFS mount struct freed before dirent")); 398 not_found = 2; 399 sx_xunlock(&dmp->dm_lock); 400 devfs_unmount_final(dmp); 401 } 402 if (not_found == 1 || (drop_dm_lock && not_found != 2)) 403 sx_unlock(&dmp->dm_lock); 404 return (not_found); 405 } 406 407 static void 408 devfs_insmntque_dtr(struct vnode *vp, void *arg) 409 { 410 struct devfs_dirent *de; 411 412 de = (struct devfs_dirent *)arg; 413 mtx_lock(&devfs_de_interlock); 414 vp->v_data = NULL; 415 de->de_vnode = NULL; 416 mtx_unlock(&devfs_de_interlock); 417 vgone(vp); 418 vput(vp); 419 } 420 421 /* 422 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops 423 * it on return. 424 */ 425 int 426 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode, 427 struct vnode **vpp) 428 { 429 int error; 430 struct vnode *vp; 431 struct cdev *dev; 432 struct devfs_mount *dmp; 433 struct cdevsw *dsw; 434 435 dmp = VFSTODEVFS(mp); 436 if (de->de_flags & DE_DOOMED) { 437 sx_xunlock(&dmp->dm_lock); 438 return (ENOENT); 439 } 440 loop: 441 DEVFS_DE_HOLD(de); 442 DEVFS_DMP_HOLD(dmp); 443 mtx_lock(&devfs_de_interlock); 444 vp = de->de_vnode; 445 if (vp != NULL) { 446 VI_LOCK(vp); 447 mtx_unlock(&devfs_de_interlock); 448 sx_xunlock(&dmp->dm_lock); 449 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread); 450 sx_xlock(&dmp->dm_lock); 451 if (devfs_allocv_drop_refs(0, dmp, de)) { 452 vput(vp); 453 return (ENOENT); 454 } 455 else if ((vp->v_iflag & VI_DOOMED) != 0) { 456 mtx_lock(&devfs_de_interlock); 457 if (de->de_vnode == vp) { 458 de->de_vnode = NULL; 459 vp->v_data = NULL; 460 } 461 mtx_unlock(&devfs_de_interlock); 462 vput(vp); 463 goto loop; 464 } 465 sx_xunlock(&dmp->dm_lock); 466 *vpp = vp; 467 return (0); 468 } 469 mtx_unlock(&devfs_de_interlock); 470 if (de->de_dirent->d_type == DT_CHR) { 471 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) { 472 devfs_allocv_drop_refs(1, dmp, de); 473 return (ENOENT); 474 } 475 dev = &de->de_cdp->cdp_c; 476 } else { 477 dev = NULL; 478 } 479 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); 480 if (error != 0) { 481 devfs_allocv_drop_refs(1, dmp, de); 482 printf("devfs_allocv: failed to allocate new vnode\n"); 483 return (error); 484 } 485 486 if (de->de_dirent->d_type == DT_CHR) { 487 vp->v_type = VCHR; 488 VI_LOCK(vp); 489 dev_lock(); 490 dev_refl(dev); 491 /* XXX: v_rdev should be protect by vnode lock */ 492 vp->v_rdev = dev; 493 KASSERT(vp->v_usecount == 1, 494 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount)); 495 dev->si_usecount += vp->v_usecount; 496 /* Special casing of ttys for deadfs. Probably redundant. */ 497 dsw = dev->si_devsw; 498 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) 499 vp->v_vflag |= VV_ISTTY; 500 dev_unlock(); 501 VI_UNLOCK(vp); 502 if ((dev->si_flags & SI_ETERNAL) != 0) 503 vp->v_vflag |= VV_ETERNALDEV; 504 vp->v_op = &devfs_specops; 505 } else if (de->de_dirent->d_type == DT_DIR) { 506 vp->v_type = VDIR; 507 } else if (de->de_dirent->d_type == DT_LNK) { 508 vp->v_type = VLNK; 509 } else { 510 vp->v_type = VBAD; 511 } 512 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); 513 VN_LOCK_ASHARE(vp); 514 mtx_lock(&devfs_de_interlock); 515 vp->v_data = de; 516 de->de_vnode = vp; 517 mtx_unlock(&devfs_de_interlock); 518 error = insmntque1(vp, mp, devfs_insmntque_dtr, de); 519 if (error != 0) { 520 (void) devfs_allocv_drop_refs(1, dmp, de); 521 return (error); 522 } 523 if (devfs_allocv_drop_refs(0, dmp, de)) { 524 vput(vp); 525 return (ENOENT); 526 } 527 #ifdef MAC 528 mac_devfs_vnode_associate(mp, de, vp); 529 #endif 530 sx_xunlock(&dmp->dm_lock); 531 *vpp = vp; 532 return (0); 533 } 534 535 static int 536 devfs_access(struct vop_access_args *ap) 537 { 538 struct vnode *vp = ap->a_vp; 539 struct devfs_dirent *de; 540 struct proc *p; 541 int error; 542 543 de = vp->v_data; 544 if (vp->v_type == VDIR) 545 de = de->de_dir; 546 547 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, 548 ap->a_accmode, ap->a_cred, NULL); 549 if (error == 0) 550 return (0); 551 if (error != EACCES) 552 return (error); 553 p = ap->a_td->td_proc; 554 /* We do, however, allow access to the controlling terminal */ 555 PROC_LOCK(p); 556 if (!(p->p_flag & P_CONTROLT)) { 557 PROC_UNLOCK(p); 558 return (error); 559 } 560 if (p->p_session->s_ttydp == de->de_cdp) 561 error = 0; 562 PROC_UNLOCK(p); 563 return (error); 564 } 565 566 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0, 567 "devfs-only flag reuse failed"); 568 569 static int 570 devfs_close(struct vop_close_args *ap) 571 { 572 struct vnode *vp = ap->a_vp, *oldvp; 573 struct thread *td = ap->a_td; 574 struct proc *p; 575 struct cdev *dev = vp->v_rdev; 576 struct cdevsw *dsw; 577 int dflags, error, ref, vp_locked; 578 579 /* 580 * XXX: Don't call d_close() if we were called because of 581 * XXX: insmntque1() failure. 582 */ 583 if (vp->v_data == NULL) 584 return (0); 585 586 /* 587 * Hack: a tty device that is a controlling terminal 588 * has a reference from the session structure. 589 * We cannot easily tell that a character device is 590 * a controlling terminal, unless it is the closing 591 * process' controlling terminal. In that case, 592 * if the reference count is 2 (this last descriptor 593 * plus the session), release the reference from the session. 594 */ 595 if (td != NULL) { 596 p = td->td_proc; 597 PROC_LOCK(p); 598 if (vp == p->p_session->s_ttyvp) { 599 PROC_UNLOCK(p); 600 oldvp = NULL; 601 sx_xlock(&proctree_lock); 602 if (vp == p->p_session->s_ttyvp) { 603 SESS_LOCK(p->p_session); 604 VI_LOCK(vp); 605 if (count_dev(dev) == 2 && 606 (vp->v_iflag & VI_DOOMED) == 0) { 607 p->p_session->s_ttyvp = NULL; 608 p->p_session->s_ttydp = NULL; 609 oldvp = vp; 610 } 611 VI_UNLOCK(vp); 612 SESS_UNLOCK(p->p_session); 613 } 614 sx_xunlock(&proctree_lock); 615 if (oldvp != NULL) 616 vrele(oldvp); 617 } else 618 PROC_UNLOCK(p); 619 } 620 /* 621 * We do not want to really close the device if it 622 * is still in use unless we are trying to close it 623 * forcibly. Since every use (buffer, vnode, swap, cmap) 624 * holds a reference to the vnode, and because we mark 625 * any other vnodes that alias this device, when the 626 * sum of the reference counts on all the aliased 627 * vnodes descends to one, we are on last close. 628 */ 629 dsw = dev_refthread(dev, &ref); 630 if (dsw == NULL) 631 return (ENXIO); 632 dflags = 0; 633 VI_LOCK(vp); 634 if (vp->v_iflag & VI_DOOMED) { 635 /* Forced close. */ 636 dflags |= FREVOKE | FNONBLOCK; 637 } else if (dsw->d_flags & D_TRACKCLOSE) { 638 /* Keep device updated on status. */ 639 } else if (count_dev(dev) > 1) { 640 VI_UNLOCK(vp); 641 dev_relthread(dev, ref); 642 return (0); 643 } 644 if (count_dev(dev) == 1) 645 dflags |= FLASTCLOSE; 646 vholdl(vp); 647 VI_UNLOCK(vp); 648 vp_locked = VOP_ISLOCKED(vp); 649 VOP_UNLOCK(vp, 0); 650 KASSERT(dev->si_refcount > 0, 651 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 652 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td); 653 dev_relthread(dev, ref); 654 vn_lock(vp, vp_locked | LK_RETRY); 655 vdrop(vp); 656 return (error); 657 } 658 659 static int 660 devfs_close_f(struct file *fp, struct thread *td) 661 { 662 int error; 663 struct file *fpop; 664 665 /* 666 * NB: td may be NULL if this descriptor is closed due to 667 * garbage collection from a closed UNIX domain socket. 668 */ 669 fpop = curthread->td_fpop; 670 curthread->td_fpop = fp; 671 error = vnops.fo_close(fp, td); 672 curthread->td_fpop = fpop; 673 674 /* 675 * The f_cdevpriv cannot be assigned non-NULL value while we 676 * are destroying the file. 677 */ 678 if (fp->f_cdevpriv != NULL) 679 devfs_fpdrop(fp); 680 return (error); 681 } 682 683 static int 684 devfs_getattr(struct vop_getattr_args *ap) 685 { 686 struct vnode *vp = ap->a_vp; 687 struct vattr *vap = ap->a_vap; 688 struct devfs_dirent *de; 689 struct devfs_mount *dmp; 690 struct cdev *dev; 691 struct timeval boottime; 692 int error; 693 694 error = devfs_populate_vp(vp); 695 if (error != 0) 696 return (error); 697 698 dmp = VFSTODEVFS(vp->v_mount); 699 sx_xunlock(&dmp->dm_lock); 700 701 de = vp->v_data; 702 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 703 if (vp->v_type == VDIR) { 704 de = de->de_dir; 705 KASSERT(de != NULL, 706 ("Null dir dirent in devfs_getattr vp=%p", vp)); 707 } 708 vap->va_uid = de->de_uid; 709 vap->va_gid = de->de_gid; 710 vap->va_mode = de->de_mode; 711 if (vp->v_type == VLNK) 712 vap->va_size = strlen(de->de_symlink); 713 else if (vp->v_type == VDIR) 714 vap->va_size = vap->va_bytes = DEV_BSIZE; 715 else 716 vap->va_size = 0; 717 if (vp->v_type != VDIR) 718 vap->va_bytes = 0; 719 vap->va_blocksize = DEV_BSIZE; 720 vap->va_type = vp->v_type; 721 722 getboottime(&boottime); 723 #define fix(aa) \ 724 do { \ 725 if ((aa).tv_sec <= 3600) { \ 726 (aa).tv_sec = boottime.tv_sec; \ 727 (aa).tv_nsec = boottime.tv_usec * 1000; \ 728 } \ 729 } while (0) 730 731 if (vp->v_type != VCHR) { 732 fix(de->de_atime); 733 vap->va_atime = de->de_atime; 734 fix(de->de_mtime); 735 vap->va_mtime = de->de_mtime; 736 fix(de->de_ctime); 737 vap->va_ctime = de->de_ctime; 738 } else { 739 dev = vp->v_rdev; 740 fix(dev->si_atime); 741 vap->va_atime = dev->si_atime; 742 fix(dev->si_mtime); 743 vap->va_mtime = dev->si_mtime; 744 fix(dev->si_ctime); 745 vap->va_ctime = dev->si_ctime; 746 747 vap->va_rdev = cdev2priv(dev)->cdp_inode; 748 } 749 vap->va_gen = 0; 750 vap->va_flags = 0; 751 vap->va_filerev = 0; 752 vap->va_nlink = de->de_links; 753 vap->va_fileid = de->de_inode; 754 755 return (error); 756 } 757 758 /* ARGSUSED */ 759 static int 760 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 761 { 762 struct file *fpop; 763 int error; 764 765 fpop = td->td_fpop; 766 td->td_fpop = fp; 767 error = vnops.fo_ioctl(fp, com, data, cred, td); 768 td->td_fpop = fpop; 769 return (error); 770 } 771 772 void * 773 fiodgname_buf_get_ptr(void *fgnp, u_long com) 774 { 775 union { 776 struct fiodgname_arg fgn; 777 #ifdef COMPAT_FREEBSD32 778 struct fiodgname_arg32 fgn32; 779 #endif 780 } *fgnup; 781 782 fgnup = fgnp; 783 switch (com) { 784 case FIODGNAME: 785 return (fgnup->fgn.buf); 786 #ifdef COMPAT_FREEBSD32 787 case FIODGNAME_32: 788 return ((void *)(uintptr_t)fgnup->fgn32.buf); 789 #endif 790 default: 791 panic("Unhandled ioctl command %ld", com); 792 } 793 } 794 795 static int 796 devfs_ioctl(struct vop_ioctl_args *ap) 797 { 798 struct fiodgname_arg *fgn; 799 struct vnode *vpold, *vp; 800 struct cdevsw *dsw; 801 struct thread *td; 802 struct cdev *dev; 803 int error, ref, i; 804 const char *p; 805 u_long com; 806 807 vp = ap->a_vp; 808 com = ap->a_command; 809 td = ap->a_td; 810 811 dsw = devvn_refthread(vp, &dev, &ref); 812 if (dsw == NULL) 813 return (ENXIO); 814 KASSERT(dev->si_refcount > 0, 815 ("devfs: un-referenced struct cdev *(%s)", devtoname(dev))); 816 817 switch (com) { 818 case FIODTYPE: 819 *(int *)ap->a_data = dsw->d_flags & D_TYPEMASK; 820 error = 0; 821 break; 822 case FIODGNAME: 823 #ifdef COMPAT_FREEBSD32 824 case FIODGNAME_32: 825 #endif 826 fgn = ap->a_data; 827 p = devtoname(dev); 828 i = strlen(p) + 1; 829 if (i > fgn->len) 830 error = EINVAL; 831 else 832 error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i); 833 break; 834 default: 835 error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td); 836 } 837 838 dev_relthread(dev, ref); 839 if (error == ENOIOCTL) 840 error = ENOTTY; 841 842 if (error == 0 && com == TIOCSCTTY) { 843 /* Do nothing if reassigning same control tty */ 844 sx_slock(&proctree_lock); 845 if (td->td_proc->p_session->s_ttyvp == vp) { 846 sx_sunlock(&proctree_lock); 847 return (0); 848 } 849 850 vpold = td->td_proc->p_session->s_ttyvp; 851 VREF(vp); 852 SESS_LOCK(td->td_proc->p_session); 853 td->td_proc->p_session->s_ttyvp = vp; 854 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 855 SESS_UNLOCK(td->td_proc->p_session); 856 857 sx_sunlock(&proctree_lock); 858 859 /* Get rid of reference to old control tty */ 860 if (vpold) 861 vrele(vpold); 862 } 863 return (error); 864 } 865 866 /* ARGSUSED */ 867 static int 868 devfs_kqfilter_f(struct file *fp, struct knote *kn) 869 { 870 struct cdev *dev; 871 struct cdevsw *dsw; 872 int error, ref; 873 struct file *fpop; 874 struct thread *td; 875 876 td = curthread; 877 fpop = td->td_fpop; 878 error = devfs_fp_check(fp, &dev, &dsw, &ref); 879 if (error) 880 return (error); 881 error = dsw->d_kqfilter(dev, kn); 882 td->td_fpop = fpop; 883 dev_relthread(dev, ref); 884 return (error); 885 } 886 887 static inline int 888 devfs_prison_check(struct devfs_dirent *de, struct thread *td) 889 { 890 struct cdev_priv *cdp; 891 struct ucred *dcr; 892 struct proc *p; 893 int error; 894 895 cdp = de->de_cdp; 896 if (cdp == NULL) 897 return (0); 898 dcr = cdp->cdp_c.si_cred; 899 if (dcr == NULL) 900 return (0); 901 902 error = prison_check(td->td_ucred, dcr); 903 if (error == 0) 904 return (0); 905 /* We do, however, allow access to the controlling terminal */ 906 p = td->td_proc; 907 PROC_LOCK(p); 908 if (!(p->p_flag & P_CONTROLT)) { 909 PROC_UNLOCK(p); 910 return (error); 911 } 912 if (p->p_session->s_ttydp == cdp) 913 error = 0; 914 PROC_UNLOCK(p); 915 return (error); 916 } 917 918 static int 919 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 920 { 921 struct componentname *cnp; 922 struct vnode *dvp, **vpp; 923 struct thread *td; 924 struct devfs_dirent *de, *dd; 925 struct devfs_dirent **dde; 926 struct devfs_mount *dmp; 927 struct mount *mp; 928 struct cdev *cdev; 929 int error, flags, nameiop, dvplocked; 930 char specname[SPECNAMELEN + 1], *pname; 931 932 cnp = ap->a_cnp; 933 vpp = ap->a_vpp; 934 dvp = ap->a_dvp; 935 pname = cnp->cn_nameptr; 936 td = cnp->cn_thread; 937 flags = cnp->cn_flags; 938 nameiop = cnp->cn_nameiop; 939 mp = dvp->v_mount; 940 dmp = VFSTODEVFS(mp); 941 dd = dvp->v_data; 942 *vpp = NULLVP; 943 944 if ((flags & ISLASTCN) && nameiop == RENAME) 945 return (EOPNOTSUPP); 946 947 if (dvp->v_type != VDIR) 948 return (ENOTDIR); 949 950 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 951 return (EIO); 952 953 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 954 if (error) 955 return (error); 956 957 if (cnp->cn_namelen == 1 && *pname == '.') { 958 if ((flags & ISLASTCN) && nameiop != LOOKUP) 959 return (EINVAL); 960 *vpp = dvp; 961 VREF(dvp); 962 return (0); 963 } 964 965 if (flags & ISDOTDOT) { 966 if ((flags & ISLASTCN) && nameiop != LOOKUP) 967 return (EINVAL); 968 de = devfs_parent_dirent(dd); 969 if (de == NULL) 970 return (ENOENT); 971 dvplocked = VOP_ISLOCKED(dvp); 972 VOP_UNLOCK(dvp, 0); 973 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, 974 vpp); 975 *dm_unlock = 0; 976 vn_lock(dvp, dvplocked | LK_RETRY); 977 return (error); 978 } 979 980 dd = dvp->v_data; 981 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 982 while (de == NULL) { /* While(...) so we can use break */ 983 984 if (nameiop == DELETE) 985 return (ENOENT); 986 987 /* 988 * OK, we didn't have an entry for the name we were asked for 989 * so we try to see if anybody can create it on demand. 990 */ 991 pname = devfs_fqpn(specname, dmp, dd, cnp); 992 if (pname == NULL) 993 break; 994 995 cdev = NULL; 996 DEVFS_DMP_HOLD(dmp); 997 sx_xunlock(&dmp->dm_lock); 998 sx_slock(&clone_drain_lock); 999 EVENTHANDLER_INVOKE(dev_clone, 1000 td->td_ucred, pname, strlen(pname), &cdev); 1001 sx_sunlock(&clone_drain_lock); 1002 1003 if (cdev == NULL) 1004 sx_xlock(&dmp->dm_lock); 1005 else if (devfs_populate_vp(dvp) != 0) { 1006 *dm_unlock = 0; 1007 sx_xlock(&dmp->dm_lock); 1008 if (DEVFS_DMP_DROP(dmp)) { 1009 sx_xunlock(&dmp->dm_lock); 1010 devfs_unmount_final(dmp); 1011 } else 1012 sx_xunlock(&dmp->dm_lock); 1013 dev_rel(cdev); 1014 return (ENOENT); 1015 } 1016 if (DEVFS_DMP_DROP(dmp)) { 1017 *dm_unlock = 0; 1018 sx_xunlock(&dmp->dm_lock); 1019 devfs_unmount_final(dmp); 1020 if (cdev != NULL) 1021 dev_rel(cdev); 1022 return (ENOENT); 1023 } 1024 1025 if (cdev == NULL) 1026 break; 1027 1028 dev_lock(); 1029 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 1030 if (dde != NULL && *dde != NULL) 1031 de = *dde; 1032 dev_unlock(); 1033 dev_rel(cdev); 1034 break; 1035 } 1036 1037 if (de == NULL || de->de_flags & DE_WHITEOUT) { 1038 if ((nameiop == CREATE || nameiop == RENAME) && 1039 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 1040 cnp->cn_flags |= SAVENAME; 1041 return (EJUSTRETURN); 1042 } 1043 return (ENOENT); 1044 } 1045 1046 if (devfs_prison_check(de, td)) 1047 return (ENOENT); 1048 1049 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 1050 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 1051 if (error) 1052 return (error); 1053 if (*vpp == dvp) { 1054 VREF(dvp); 1055 *vpp = dvp; 1056 return (0); 1057 } 1058 } 1059 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp); 1060 *dm_unlock = 0; 1061 return (error); 1062 } 1063 1064 static int 1065 devfs_lookup(struct vop_lookup_args *ap) 1066 { 1067 int j; 1068 struct devfs_mount *dmp; 1069 int dm_unlock; 1070 1071 if (devfs_populate_vp(ap->a_dvp) != 0) 1072 return (ENOTDIR); 1073 1074 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1075 dm_unlock = 1; 1076 j = devfs_lookupx(ap, &dm_unlock); 1077 if (dm_unlock == 1) 1078 sx_xunlock(&dmp->dm_lock); 1079 return (j); 1080 } 1081 1082 static int 1083 devfs_mknod(struct vop_mknod_args *ap) 1084 { 1085 struct componentname *cnp; 1086 struct vnode *dvp, **vpp; 1087 struct devfs_dirent *dd, *de; 1088 struct devfs_mount *dmp; 1089 int error; 1090 1091 /* 1092 * The only type of node we should be creating here is a 1093 * character device, for anything else return EOPNOTSUPP. 1094 */ 1095 if (ap->a_vap->va_type != VCHR) 1096 return (EOPNOTSUPP); 1097 dvp = ap->a_dvp; 1098 dmp = VFSTODEVFS(dvp->v_mount); 1099 1100 cnp = ap->a_cnp; 1101 vpp = ap->a_vpp; 1102 dd = dvp->v_data; 1103 1104 error = ENOENT; 1105 sx_xlock(&dmp->dm_lock); 1106 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1107 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1108 continue; 1109 if (de->de_dirent->d_type == DT_CHR && 1110 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1111 continue; 1112 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1113 de->de_dirent->d_namlen) != 0) 1114 continue; 1115 if (de->de_flags & DE_WHITEOUT) 1116 break; 1117 goto notfound; 1118 } 1119 if (de == NULL) 1120 goto notfound; 1121 de->de_flags &= ~DE_WHITEOUT; 1122 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1123 return (error); 1124 notfound: 1125 sx_xunlock(&dmp->dm_lock); 1126 return (error); 1127 } 1128 1129 /* ARGSUSED */ 1130 static int 1131 devfs_open(struct vop_open_args *ap) 1132 { 1133 struct thread *td = ap->a_td; 1134 struct vnode *vp = ap->a_vp; 1135 struct cdev *dev = vp->v_rdev; 1136 struct file *fp = ap->a_fp; 1137 int error, ref, vlocked; 1138 struct cdevsw *dsw; 1139 struct file *fpop; 1140 struct mtx *mtxp; 1141 1142 if (vp->v_type == VBLK) 1143 return (ENXIO); 1144 1145 if (dev == NULL) 1146 return (ENXIO); 1147 1148 /* Make this field valid before any I/O in d_open. */ 1149 if (dev->si_iosize_max == 0) 1150 dev->si_iosize_max = DFLTPHYS; 1151 1152 dsw = dev_refthread(dev, &ref); 1153 if (dsw == NULL) 1154 return (ENXIO); 1155 if (fp == NULL && dsw->d_fdopen != NULL) { 1156 dev_relthread(dev, ref); 1157 return (ENXIO); 1158 } 1159 1160 vlocked = VOP_ISLOCKED(vp); 1161 VOP_UNLOCK(vp, 0); 1162 1163 fpop = td->td_fpop; 1164 td->td_fpop = fp; 1165 if (fp != NULL) { 1166 fp->f_data = dev; 1167 fp->f_vnode = vp; 1168 } 1169 if (dsw->d_fdopen != NULL) 1170 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1171 else 1172 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1173 /* Clean up any cdevpriv upon error. */ 1174 if (error != 0) 1175 devfs_clear_cdevpriv(); 1176 td->td_fpop = fpop; 1177 1178 vn_lock(vp, vlocked | LK_RETRY); 1179 dev_relthread(dev, ref); 1180 if (error != 0) { 1181 if (error == ERESTART) 1182 error = EINTR; 1183 return (error); 1184 } 1185 1186 #if 0 /* /dev/console */ 1187 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1188 #else 1189 if (fp == NULL) 1190 return (error); 1191 #endif 1192 if (fp->f_ops == &badfileops) 1193 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1194 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1195 1196 /* 1197 * Hint to the dofilewrite() to not force the buffer draining 1198 * on the writer to the file. Most likely, the write would 1199 * not need normal buffers. 1200 */ 1201 mtx_lock(mtxp); 1202 fp->f_vnread_flags |= FDEVFS_VNODE; 1203 mtx_unlock(mtxp); 1204 return (error); 1205 } 1206 1207 static int 1208 devfs_pathconf(struct vop_pathconf_args *ap) 1209 { 1210 1211 switch (ap->a_name) { 1212 case _PC_FILESIZEBITS: 1213 *ap->a_retval = 64; 1214 return (0); 1215 case _PC_NAME_MAX: 1216 *ap->a_retval = NAME_MAX; 1217 return (0); 1218 case _PC_LINK_MAX: 1219 *ap->a_retval = INT_MAX; 1220 return (0); 1221 case _PC_SYMLINK_MAX: 1222 *ap->a_retval = MAXPATHLEN; 1223 return (0); 1224 case _PC_MAX_CANON: 1225 if (ap->a_vp->v_vflag & VV_ISTTY) { 1226 *ap->a_retval = MAX_CANON; 1227 return (0); 1228 } 1229 return (EINVAL); 1230 case _PC_MAX_INPUT: 1231 if (ap->a_vp->v_vflag & VV_ISTTY) { 1232 *ap->a_retval = MAX_INPUT; 1233 return (0); 1234 } 1235 return (EINVAL); 1236 case _PC_VDISABLE: 1237 if (ap->a_vp->v_vflag & VV_ISTTY) { 1238 *ap->a_retval = _POSIX_VDISABLE; 1239 return (0); 1240 } 1241 return (EINVAL); 1242 case _PC_MAC_PRESENT: 1243 #ifdef MAC 1244 /* 1245 * If MAC is enabled, devfs automatically supports 1246 * trivial non-persistant label storage. 1247 */ 1248 *ap->a_retval = 1; 1249 #else 1250 *ap->a_retval = 0; 1251 #endif 1252 return (0); 1253 case _PC_CHOWN_RESTRICTED: 1254 *ap->a_retval = 1; 1255 return (0); 1256 default: 1257 return (vop_stdpathconf(ap)); 1258 } 1259 /* NOTREACHED */ 1260 } 1261 1262 /* ARGSUSED */ 1263 static int 1264 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1265 { 1266 struct cdev *dev; 1267 struct cdevsw *dsw; 1268 int error, ref; 1269 struct file *fpop; 1270 1271 fpop = td->td_fpop; 1272 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1273 if (error != 0) { 1274 error = vnops.fo_poll(fp, events, cred, td); 1275 return (error); 1276 } 1277 error = dsw->d_poll(dev, events, td); 1278 td->td_fpop = fpop; 1279 dev_relthread(dev, ref); 1280 return(error); 1281 } 1282 1283 /* 1284 * Print out the contents of a special device vnode. 1285 */ 1286 static int 1287 devfs_print(struct vop_print_args *ap) 1288 { 1289 1290 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1291 return (0); 1292 } 1293 1294 static int 1295 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1296 int flags, struct thread *td) 1297 { 1298 struct cdev *dev; 1299 int ioflag, error, ref; 1300 ssize_t resid; 1301 struct cdevsw *dsw; 1302 struct file *fpop; 1303 1304 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1305 return (EINVAL); 1306 fpop = td->td_fpop; 1307 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1308 if (error != 0) { 1309 error = vnops.fo_read(fp, uio, cred, flags, td); 1310 return (error); 1311 } 1312 resid = uio->uio_resid; 1313 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1314 if (ioflag & O_DIRECT) 1315 ioflag |= IO_DIRECT; 1316 1317 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1318 error = dsw->d_read(dev, uio, ioflag); 1319 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1320 devfs_timestamp(&dev->si_atime); 1321 td->td_fpop = fpop; 1322 dev_relthread(dev, ref); 1323 1324 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1325 return (error); 1326 } 1327 1328 static int 1329 devfs_readdir(struct vop_readdir_args *ap) 1330 { 1331 int error; 1332 struct uio *uio; 1333 struct dirent *dp; 1334 struct devfs_dirent *dd; 1335 struct devfs_dirent *de; 1336 struct devfs_mount *dmp; 1337 off_t off; 1338 int *tmp_ncookies = NULL; 1339 1340 if (ap->a_vp->v_type != VDIR) 1341 return (ENOTDIR); 1342 1343 uio = ap->a_uio; 1344 if (uio->uio_offset < 0) 1345 return (EINVAL); 1346 1347 /* 1348 * XXX: This is a temporary hack to get around this filesystem not 1349 * supporting cookies. We store the location of the ncookies pointer 1350 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1351 * and set the number of cookies to 0. We then set the pointer to 1352 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1353 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1354 * pointer to its original location before returning to the caller. 1355 */ 1356 if (ap->a_ncookies != NULL) { 1357 tmp_ncookies = ap->a_ncookies; 1358 *ap->a_ncookies = 0; 1359 ap->a_ncookies = NULL; 1360 } 1361 1362 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1363 if (devfs_populate_vp(ap->a_vp) != 0) { 1364 if (tmp_ncookies != NULL) 1365 ap->a_ncookies = tmp_ncookies; 1366 return (EIO); 1367 } 1368 error = 0; 1369 de = ap->a_vp->v_data; 1370 off = 0; 1371 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1372 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1373 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1374 continue; 1375 if (devfs_prison_check(dd, uio->uio_td)) 1376 continue; 1377 if (dd->de_dirent->d_type == DT_DIR) 1378 de = dd->de_dir; 1379 else 1380 de = dd; 1381 dp = dd->de_dirent; 1382 MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp)); 1383 if (dp->d_reclen > uio->uio_resid) 1384 break; 1385 dp->d_fileno = de->de_inode; 1386 /* NOTE: d_off is the offset for the *next* entry. */ 1387 dp->d_off = off + dp->d_reclen; 1388 if (off >= uio->uio_offset) { 1389 error = vfs_read_dirent(ap, dp, off); 1390 if (error) 1391 break; 1392 } 1393 off += dp->d_reclen; 1394 } 1395 sx_xunlock(&dmp->dm_lock); 1396 uio->uio_offset = off; 1397 1398 /* 1399 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1400 * place. 1401 */ 1402 if (tmp_ncookies != NULL) 1403 ap->a_ncookies = tmp_ncookies; 1404 1405 return (error); 1406 } 1407 1408 static int 1409 devfs_readlink(struct vop_readlink_args *ap) 1410 { 1411 struct devfs_dirent *de; 1412 1413 de = ap->a_vp->v_data; 1414 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1415 } 1416 1417 static int 1418 devfs_reclaim(struct vop_reclaim_args *ap) 1419 { 1420 struct vnode *vp; 1421 struct devfs_dirent *de; 1422 1423 vp = ap->a_vp; 1424 mtx_lock(&devfs_de_interlock); 1425 de = vp->v_data; 1426 if (de != NULL) { 1427 de->de_vnode = NULL; 1428 vp->v_data = NULL; 1429 } 1430 mtx_unlock(&devfs_de_interlock); 1431 vnode_destroy_vobject(vp); 1432 return (0); 1433 } 1434 1435 static int 1436 devfs_reclaim_vchr(struct vop_reclaim_args *ap) 1437 { 1438 struct vnode *vp; 1439 struct cdev *dev; 1440 1441 vp = ap->a_vp; 1442 MPASS(vp->v_type == VCHR); 1443 1444 devfs_reclaim(ap); 1445 1446 VI_LOCK(vp); 1447 dev_lock(); 1448 dev = vp->v_rdev; 1449 vp->v_rdev = NULL; 1450 if (dev != NULL) 1451 dev->si_usecount -= vp->v_usecount; 1452 dev_unlock(); 1453 VI_UNLOCK(vp); 1454 if (dev != NULL) 1455 dev_rel(dev); 1456 return (0); 1457 } 1458 1459 static int 1460 devfs_remove(struct vop_remove_args *ap) 1461 { 1462 struct vnode *dvp = ap->a_dvp; 1463 struct vnode *vp = ap->a_vp; 1464 struct devfs_dirent *dd; 1465 struct devfs_dirent *de, *de_covered; 1466 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1467 1468 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1469 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1470 1471 sx_xlock(&dmp->dm_lock); 1472 dd = ap->a_dvp->v_data; 1473 de = vp->v_data; 1474 if (de->de_cdp == NULL) { 1475 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1476 if (de->de_dirent->d_type == DT_LNK) { 1477 de_covered = devfs_find(dd, de->de_dirent->d_name, 1478 de->de_dirent->d_namlen, 0); 1479 if (de_covered != NULL) 1480 de_covered->de_flags &= ~DE_COVERED; 1481 } 1482 /* We need to unlock dvp because devfs_delete() may lock it. */ 1483 VOP_UNLOCK(vp, 0); 1484 if (dvp != vp) 1485 VOP_UNLOCK(dvp, 0); 1486 devfs_delete(dmp, de, 0); 1487 sx_xunlock(&dmp->dm_lock); 1488 if (dvp != vp) 1489 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1490 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1491 } else { 1492 de->de_flags |= DE_WHITEOUT; 1493 sx_xunlock(&dmp->dm_lock); 1494 } 1495 return (0); 1496 } 1497 1498 /* 1499 * Revoke is called on a tty when a terminal session ends. The vnode 1500 * is orphaned by setting v_op to deadfs so we need to let go of it 1501 * as well so that we create a new one next time around. 1502 * 1503 */ 1504 static int 1505 devfs_revoke(struct vop_revoke_args *ap) 1506 { 1507 struct vnode *vp = ap->a_vp, *vp2; 1508 struct cdev *dev; 1509 struct cdev_priv *cdp; 1510 struct devfs_dirent *de; 1511 u_int i; 1512 1513 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1514 1515 dev = vp->v_rdev; 1516 cdp = cdev2priv(dev); 1517 1518 dev_lock(); 1519 cdp->cdp_inuse++; 1520 dev_unlock(); 1521 1522 vhold(vp); 1523 vgone(vp); 1524 vdrop(vp); 1525 1526 VOP_UNLOCK(vp,0); 1527 loop: 1528 for (;;) { 1529 mtx_lock(&devfs_de_interlock); 1530 dev_lock(); 1531 vp2 = NULL; 1532 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1533 de = cdp->cdp_dirents[i]; 1534 if (de == NULL) 1535 continue; 1536 1537 vp2 = de->de_vnode; 1538 if (vp2 != NULL) { 1539 dev_unlock(); 1540 VI_LOCK(vp2); 1541 mtx_unlock(&devfs_de_interlock); 1542 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1543 curthread)) 1544 goto loop; 1545 vhold(vp2); 1546 vgone(vp2); 1547 vdrop(vp2); 1548 vput(vp2); 1549 break; 1550 } 1551 } 1552 if (vp2 != NULL) { 1553 continue; 1554 } 1555 dev_unlock(); 1556 mtx_unlock(&devfs_de_interlock); 1557 break; 1558 } 1559 dev_lock(); 1560 cdp->cdp_inuse--; 1561 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1562 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1563 dev_unlock(); 1564 dev_rel(&cdp->cdp_c); 1565 } else 1566 dev_unlock(); 1567 1568 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1569 return (0); 1570 } 1571 1572 static int 1573 devfs_rioctl(struct vop_ioctl_args *ap) 1574 { 1575 struct vnode *vp; 1576 struct devfs_mount *dmp; 1577 int error; 1578 1579 vp = ap->a_vp; 1580 vn_lock(vp, LK_SHARED | LK_RETRY); 1581 if (vp->v_iflag & VI_DOOMED) { 1582 VOP_UNLOCK(vp, 0); 1583 return (EBADF); 1584 } 1585 dmp = VFSTODEVFS(vp->v_mount); 1586 sx_xlock(&dmp->dm_lock); 1587 VOP_UNLOCK(vp, 0); 1588 DEVFS_DMP_HOLD(dmp); 1589 devfs_populate(dmp); 1590 if (DEVFS_DMP_DROP(dmp)) { 1591 sx_xunlock(&dmp->dm_lock); 1592 devfs_unmount_final(dmp); 1593 return (ENOENT); 1594 } 1595 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1596 sx_xunlock(&dmp->dm_lock); 1597 return (error); 1598 } 1599 1600 static int 1601 devfs_rread(struct vop_read_args *ap) 1602 { 1603 1604 if (ap->a_vp->v_type != VDIR) 1605 return (EINVAL); 1606 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1607 } 1608 1609 static int 1610 devfs_setattr(struct vop_setattr_args *ap) 1611 { 1612 struct devfs_dirent *de; 1613 struct vattr *vap; 1614 struct vnode *vp; 1615 struct thread *td; 1616 int c, error; 1617 uid_t uid; 1618 gid_t gid; 1619 1620 vap = ap->a_vap; 1621 vp = ap->a_vp; 1622 td = curthread; 1623 if ((vap->va_type != VNON) || 1624 (vap->va_nlink != VNOVAL) || 1625 (vap->va_fsid != VNOVAL) || 1626 (vap->va_fileid != VNOVAL) || 1627 (vap->va_blocksize != VNOVAL) || 1628 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1629 (vap->va_rdev != VNOVAL) || 1630 ((int)vap->va_bytes != VNOVAL) || 1631 (vap->va_gen != VNOVAL)) { 1632 return (EINVAL); 1633 } 1634 1635 error = devfs_populate_vp(vp); 1636 if (error != 0) 1637 return (error); 1638 1639 de = vp->v_data; 1640 if (vp->v_type == VDIR) 1641 de = de->de_dir; 1642 1643 c = 0; 1644 if (vap->va_uid == (uid_t)VNOVAL) 1645 uid = de->de_uid; 1646 else 1647 uid = vap->va_uid; 1648 if (vap->va_gid == (gid_t)VNOVAL) 1649 gid = de->de_gid; 1650 else 1651 gid = vap->va_gid; 1652 if (uid != de->de_uid || gid != de->de_gid) { 1653 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1654 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1655 error = priv_check(td, PRIV_VFS_CHOWN); 1656 if (error != 0) 1657 goto ret; 1658 } 1659 de->de_uid = uid; 1660 de->de_gid = gid; 1661 c = 1; 1662 } 1663 1664 if (vap->va_mode != (mode_t)VNOVAL) { 1665 if (ap->a_cred->cr_uid != de->de_uid) { 1666 error = priv_check(td, PRIV_VFS_ADMIN); 1667 if (error != 0) 1668 goto ret; 1669 } 1670 de->de_mode = vap->va_mode; 1671 c = 1; 1672 } 1673 1674 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1675 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1676 if (error != 0) 1677 goto ret; 1678 if (vap->va_atime.tv_sec != VNOVAL) { 1679 if (vp->v_type == VCHR) 1680 vp->v_rdev->si_atime = vap->va_atime; 1681 else 1682 de->de_atime = vap->va_atime; 1683 } 1684 if (vap->va_mtime.tv_sec != VNOVAL) { 1685 if (vp->v_type == VCHR) 1686 vp->v_rdev->si_mtime = vap->va_mtime; 1687 else 1688 de->de_mtime = vap->va_mtime; 1689 } 1690 c = 1; 1691 } 1692 1693 if (c) { 1694 if (vp->v_type == VCHR) 1695 vfs_timestamp(&vp->v_rdev->si_ctime); 1696 else 1697 vfs_timestamp(&de->de_mtime); 1698 } 1699 1700 ret: 1701 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1702 return (error); 1703 } 1704 1705 #ifdef MAC 1706 static int 1707 devfs_setlabel(struct vop_setlabel_args *ap) 1708 { 1709 struct vnode *vp; 1710 struct devfs_dirent *de; 1711 1712 vp = ap->a_vp; 1713 de = vp->v_data; 1714 1715 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1716 mac_devfs_update(vp->v_mount, de, vp); 1717 1718 return (0); 1719 } 1720 #endif 1721 1722 static int 1723 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1724 { 1725 1726 return (vnops.fo_stat(fp, sb, cred, td)); 1727 } 1728 1729 static int 1730 devfs_symlink(struct vop_symlink_args *ap) 1731 { 1732 int i, error; 1733 struct devfs_dirent *dd; 1734 struct devfs_dirent *de, *de_covered, *de_dotdot; 1735 struct devfs_mount *dmp; 1736 1737 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1738 if (error) 1739 return(error); 1740 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1741 if (devfs_populate_vp(ap->a_dvp) != 0) 1742 return (ENOENT); 1743 1744 dd = ap->a_dvp->v_data; 1745 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1746 de->de_flags = DE_USER; 1747 de->de_uid = 0; 1748 de->de_gid = 0; 1749 de->de_mode = 0755; 1750 de->de_inode = alloc_unr(devfs_inos); 1751 de->de_dir = dd; 1752 de->de_dirent->d_type = DT_LNK; 1753 i = strlen(ap->a_target) + 1; 1754 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1755 bcopy(ap->a_target, de->de_symlink, i); 1756 #ifdef MAC 1757 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1758 #endif 1759 de_covered = devfs_find(dd, de->de_dirent->d_name, 1760 de->de_dirent->d_namlen, 0); 1761 if (de_covered != NULL) { 1762 if ((de_covered->de_flags & DE_USER) != 0) { 1763 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1764 sx_xunlock(&dmp->dm_lock); 1765 return (EEXIST); 1766 } 1767 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1768 ("devfs_symlink: entry %p already covered", de_covered)); 1769 de_covered->de_flags |= DE_COVERED; 1770 } 1771 1772 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1773 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1774 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1775 devfs_dir_ref_de(dmp, dd); 1776 devfs_rules_apply(dmp, de); 1777 1778 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1779 } 1780 1781 static int 1782 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1783 { 1784 1785 return (vnops.fo_truncate(fp, length, cred, td)); 1786 } 1787 1788 static int 1789 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1790 int flags, struct thread *td) 1791 { 1792 struct cdev *dev; 1793 int error, ioflag, ref; 1794 ssize_t resid; 1795 struct cdevsw *dsw; 1796 struct file *fpop; 1797 1798 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1799 return (EINVAL); 1800 fpop = td->td_fpop; 1801 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1802 if (error != 0) { 1803 error = vnops.fo_write(fp, uio, cred, flags, td); 1804 return (error); 1805 } 1806 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1807 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1808 if (ioflag & O_DIRECT) 1809 ioflag |= IO_DIRECT; 1810 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1811 1812 resid = uio->uio_resid; 1813 1814 error = dsw->d_write(dev, uio, ioflag); 1815 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1816 devfs_timestamp(&dev->si_ctime); 1817 dev->si_mtime = dev->si_ctime; 1818 } 1819 td->td_fpop = fpop; 1820 dev_relthread(dev, ref); 1821 1822 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1823 return (error); 1824 } 1825 1826 static int 1827 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1828 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1829 struct thread *td) 1830 { 1831 struct cdev *dev; 1832 struct cdevsw *dsw; 1833 struct mount *mp; 1834 struct vnode *vp; 1835 struct file *fpop; 1836 vm_object_t object; 1837 vm_prot_t maxprot; 1838 int error, ref; 1839 1840 vp = fp->f_vnode; 1841 1842 /* 1843 * Ensure that file and memory protections are 1844 * compatible. 1845 */ 1846 mp = vp->v_mount; 1847 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 1848 maxprot = VM_PROT_NONE; 1849 if ((prot & VM_PROT_EXECUTE) != 0) 1850 return (EACCES); 1851 } else 1852 maxprot = VM_PROT_EXECUTE; 1853 if ((fp->f_flag & FREAD) != 0) 1854 maxprot |= VM_PROT_READ; 1855 else if ((prot & VM_PROT_READ) != 0) 1856 return (EACCES); 1857 1858 /* 1859 * If we are sharing potential changes via MAP_SHARED and we 1860 * are trying to get write permission although we opened it 1861 * without asking for it, bail out. 1862 * 1863 * Note that most character devices always share mappings. 1864 * The one exception is that D_MMAP_ANON devices 1865 * (i.e. /dev/zero) permit private writable mappings. 1866 * 1867 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests 1868 * as well as updating maxprot to permit writing for 1869 * D_MMAP_ANON devices rather than doing that here. 1870 */ 1871 if ((flags & MAP_SHARED) != 0) { 1872 if ((fp->f_flag & FWRITE) != 0) 1873 maxprot |= VM_PROT_WRITE; 1874 else if ((prot & VM_PROT_WRITE) != 0) 1875 return (EACCES); 1876 } 1877 maxprot &= cap_maxprot; 1878 1879 fpop = td->td_fpop; 1880 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1881 if (error != 0) 1882 return (error); 1883 1884 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff, 1885 &object); 1886 td->td_fpop = fpop; 1887 dev_relthread(dev, ref); 1888 if (error != 0) 1889 return (error); 1890 1891 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1892 foff, FALSE, td); 1893 if (error != 0) 1894 vm_object_deallocate(object); 1895 return (error); 1896 } 1897 1898 dev_t 1899 dev2udev(struct cdev *x) 1900 { 1901 if (x == NULL) 1902 return (NODEV); 1903 return (cdev2priv(x)->cdp_inode); 1904 } 1905 1906 static struct fileops devfs_ops_f = { 1907 .fo_read = devfs_read_f, 1908 .fo_write = devfs_write_f, 1909 .fo_truncate = devfs_truncate_f, 1910 .fo_ioctl = devfs_ioctl_f, 1911 .fo_poll = devfs_poll_f, 1912 .fo_kqfilter = devfs_kqfilter_f, 1913 .fo_stat = devfs_stat_f, 1914 .fo_close = devfs_close_f, 1915 .fo_chmod = vn_chmod, 1916 .fo_chown = vn_chown, 1917 .fo_sendfile = vn_sendfile, 1918 .fo_seek = vn_seek, 1919 .fo_fill_kinfo = vn_fill_kinfo, 1920 .fo_mmap = devfs_mmap_f, 1921 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1922 }; 1923 1924 /* Vops for non-CHR vnodes in /dev. */ 1925 static struct vop_vector devfs_vnodeops = { 1926 .vop_default = &default_vnodeops, 1927 1928 .vop_access = devfs_access, 1929 .vop_getattr = devfs_getattr, 1930 .vop_ioctl = devfs_rioctl, 1931 .vop_lookup = devfs_lookup, 1932 .vop_mknod = devfs_mknod, 1933 .vop_pathconf = devfs_pathconf, 1934 .vop_read = devfs_rread, 1935 .vop_readdir = devfs_readdir, 1936 .vop_readlink = devfs_readlink, 1937 .vop_reclaim = devfs_reclaim, 1938 .vop_remove = devfs_remove, 1939 .vop_revoke = devfs_revoke, 1940 .vop_setattr = devfs_setattr, 1941 #ifdef MAC 1942 .vop_setlabel = devfs_setlabel, 1943 #endif 1944 .vop_symlink = devfs_symlink, 1945 .vop_vptocnp = devfs_vptocnp, 1946 }; 1947 1948 /* Vops for VCHR vnodes in /dev. */ 1949 static struct vop_vector devfs_specops = { 1950 .vop_default = &default_vnodeops, 1951 1952 .vop_access = devfs_access, 1953 .vop_bmap = VOP_PANIC, 1954 .vop_close = devfs_close, 1955 .vop_create = VOP_PANIC, 1956 .vop_fsync = vop_stdfsync, 1957 .vop_getattr = devfs_getattr, 1958 .vop_ioctl = devfs_ioctl, 1959 .vop_link = VOP_PANIC, 1960 .vop_mkdir = VOP_PANIC, 1961 .vop_mknod = VOP_PANIC, 1962 .vop_open = devfs_open, 1963 .vop_pathconf = devfs_pathconf, 1964 .vop_poll = dead_poll, 1965 .vop_print = devfs_print, 1966 .vop_read = dead_read, 1967 .vop_readdir = VOP_PANIC, 1968 .vop_readlink = VOP_PANIC, 1969 .vop_reallocblks = VOP_PANIC, 1970 .vop_reclaim = devfs_reclaim_vchr, 1971 .vop_remove = devfs_remove, 1972 .vop_rename = VOP_PANIC, 1973 .vop_revoke = devfs_revoke, 1974 .vop_rmdir = VOP_PANIC, 1975 .vop_setattr = devfs_setattr, 1976 #ifdef MAC 1977 .vop_setlabel = devfs_setlabel, 1978 #endif 1979 .vop_strategy = VOP_PANIC, 1980 .vop_symlink = VOP_PANIC, 1981 .vop_vptocnp = devfs_vptocnp, 1982 .vop_write = dead_write, 1983 }; 1984 1985 /* 1986 * Our calling convention to the device drivers used to be that we passed 1987 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1988 * flags instead since that's what open(), close() and ioctl() takes and 1989 * we don't really want vnode.h in device drivers. 1990 * We solved the source compatibility by redefining some vnode flags to 1991 * be the same as the fcntl ones and by sending down the bitwise OR of 1992 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1993 * pulls the rug out under this. 1994 */ 1995 CTASSERT(O_NONBLOCK == IO_NDELAY); 1996 CTASSERT(O_FSYNC == IO_SYNC); 1997