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