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