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