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); 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 (VN_IS_DOOMED(vp)) { 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 size_t *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 (VN_IS_DOOMED(vp)) { 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 VNPASS(vp->v_usecount == 1, vp); 483 dev->si_usecount++; 484 /* Special casing of ttys for deadfs. Probably redundant. */ 485 dsw = dev->si_devsw; 486 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) 487 vp->v_vflag |= VV_ISTTY; 488 dev_unlock(); 489 VI_UNLOCK(vp); 490 if ((dev->si_flags & SI_ETERNAL) != 0) 491 vp->v_vflag |= VV_ETERNALDEV; 492 vp->v_op = &devfs_specops; 493 } else if (de->de_dirent->d_type == DT_DIR) { 494 vp->v_type = VDIR; 495 } else if (de->de_dirent->d_type == DT_LNK) { 496 vp->v_type = VLNK; 497 } else { 498 vp->v_type = VBAD; 499 } 500 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); 501 VN_LOCK_ASHARE(vp); 502 mtx_lock(&devfs_de_interlock); 503 vp->v_data = de; 504 de->de_vnode = vp; 505 mtx_unlock(&devfs_de_interlock); 506 error = insmntque1(vp, mp, devfs_insmntque_dtr, de); 507 if (error != 0) { 508 (void) devfs_allocv_drop_refs(1, dmp, de); 509 return (error); 510 } 511 if (devfs_allocv_drop_refs(0, dmp, de)) { 512 vput(vp); 513 return (ENOENT); 514 } 515 #ifdef MAC 516 mac_devfs_vnode_associate(mp, de, vp); 517 #endif 518 sx_xunlock(&dmp->dm_lock); 519 *vpp = vp; 520 return (0); 521 } 522 523 static int 524 devfs_access(struct vop_access_args *ap) 525 { 526 struct vnode *vp = ap->a_vp; 527 struct devfs_dirent *de; 528 struct proc *p; 529 int error; 530 531 de = vp->v_data; 532 if (vp->v_type == VDIR) 533 de = de->de_dir; 534 535 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, 536 ap->a_accmode, ap->a_cred, NULL); 537 if (error == 0) 538 return (0); 539 if (error != EACCES) 540 return (error); 541 p = ap->a_td->td_proc; 542 /* We do, however, allow access to the controlling terminal */ 543 PROC_LOCK(p); 544 if (!(p->p_flag & P_CONTROLT)) { 545 PROC_UNLOCK(p); 546 return (error); 547 } 548 if (p->p_session->s_ttydp == de->de_cdp) 549 error = 0; 550 PROC_UNLOCK(p); 551 return (error); 552 } 553 554 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0, 555 "devfs-only flag reuse failed"); 556 557 static int 558 devfs_close(struct vop_close_args *ap) 559 { 560 struct vnode *vp = ap->a_vp, *oldvp; 561 struct thread *td = ap->a_td; 562 struct proc *p; 563 struct cdev *dev = vp->v_rdev; 564 struct cdevsw *dsw; 565 int dflags, error, ref, vp_locked; 566 567 /* 568 * XXX: Don't call d_close() if we were called because of 569 * XXX: insmntque1() failure. 570 */ 571 if (vp->v_data == NULL) 572 return (0); 573 574 /* 575 * Hack: a tty device that is a controlling terminal 576 * has a reference from the session structure. 577 * We cannot easily tell that a character device is 578 * a controlling terminal, unless it is the closing 579 * process' controlling terminal. In that case, 580 * if the reference count is 2 (this last descriptor 581 * plus the session), release the reference from the session. 582 */ 583 if (vp->v_usecount == 2 && td != NULL) { 584 p = td->td_proc; 585 PROC_LOCK(p); 586 if (vp == p->p_session->s_ttyvp) { 587 PROC_UNLOCK(p); 588 oldvp = NULL; 589 sx_xlock(&proctree_lock); 590 if (vp == p->p_session->s_ttyvp) { 591 SESS_LOCK(p->p_session); 592 VI_LOCK(vp); 593 if (vp->v_usecount == 2 && vcount(vp) == 1 && 594 !VN_IS_DOOMED(vp)) { 595 p->p_session->s_ttyvp = NULL; 596 p->p_session->s_ttydp = NULL; 597 oldvp = vp; 598 } 599 VI_UNLOCK(vp); 600 SESS_UNLOCK(p->p_session); 601 } 602 sx_xunlock(&proctree_lock); 603 if (oldvp != NULL) 604 vrele(oldvp); 605 } else 606 PROC_UNLOCK(p); 607 } 608 /* 609 * We do not want to really close the device if it 610 * is still in use unless we are trying to close it 611 * forcibly. Since every use (buffer, vnode, swap, cmap) 612 * holds a reference to the vnode, and because we mark 613 * any other vnodes that alias this device, when the 614 * sum of the reference counts on all the aliased 615 * vnodes descends to one, we are on last close. 616 */ 617 dsw = dev_refthread(dev, &ref); 618 if (dsw == NULL) 619 return (ENXIO); 620 dflags = 0; 621 VI_LOCK(vp); 622 if (vp->v_usecount == 1 && vcount(vp) == 1) 623 dflags |= FLASTCLOSE; 624 if (VN_IS_DOOMED(vp)) { 625 /* Forced close. */ 626 dflags |= FREVOKE | FNONBLOCK; 627 } else if (dsw->d_flags & D_TRACKCLOSE) { 628 /* Keep device updated on status. */ 629 } else if ((dflags & FLASTCLOSE) == 0) { 630 VI_UNLOCK(vp); 631 dev_relthread(dev, ref); 632 return (0); 633 } 634 vholdnz(vp); 635 VI_UNLOCK(vp); 636 vp_locked = VOP_ISLOCKED(vp); 637 VOP_UNLOCK(vp); 638 KASSERT(dev->si_refcount > 0, 639 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 640 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td); 641 dev_relthread(dev, ref); 642 vn_lock(vp, vp_locked | LK_RETRY); 643 vdrop(vp); 644 return (error); 645 } 646 647 static int 648 devfs_close_f(struct file *fp, struct thread *td) 649 { 650 int error; 651 struct file *fpop; 652 653 /* 654 * NB: td may be NULL if this descriptor is closed due to 655 * garbage collection from a closed UNIX domain socket. 656 */ 657 fpop = curthread->td_fpop; 658 curthread->td_fpop = fp; 659 error = vnops.fo_close(fp, td); 660 curthread->td_fpop = fpop; 661 662 /* 663 * The f_cdevpriv cannot be assigned non-NULL value while we 664 * are destroying the file. 665 */ 666 if (fp->f_cdevpriv != NULL) 667 devfs_fpdrop(fp); 668 return (error); 669 } 670 671 static int 672 devfs_getattr(struct vop_getattr_args *ap) 673 { 674 struct vnode *vp = ap->a_vp; 675 struct vattr *vap = ap->a_vap; 676 struct devfs_dirent *de; 677 struct devfs_mount *dmp; 678 struct cdev *dev; 679 struct timeval boottime; 680 int error; 681 682 error = devfs_populate_vp(vp); 683 if (error != 0) 684 return (error); 685 686 dmp = VFSTODEVFS(vp->v_mount); 687 sx_xunlock(&dmp->dm_lock); 688 689 de = vp->v_data; 690 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 691 if (vp->v_type == VDIR) { 692 de = de->de_dir; 693 KASSERT(de != NULL, 694 ("Null dir dirent in devfs_getattr vp=%p", vp)); 695 } 696 vap->va_uid = de->de_uid; 697 vap->va_gid = de->de_gid; 698 vap->va_mode = de->de_mode; 699 if (vp->v_type == VLNK) 700 vap->va_size = strlen(de->de_symlink); 701 else if (vp->v_type == VDIR) 702 vap->va_size = vap->va_bytes = DEV_BSIZE; 703 else 704 vap->va_size = 0; 705 if (vp->v_type != VDIR) 706 vap->va_bytes = 0; 707 vap->va_blocksize = DEV_BSIZE; 708 vap->va_type = vp->v_type; 709 710 getboottime(&boottime); 711 #define fix(aa) \ 712 do { \ 713 if ((aa).tv_sec <= 3600) { \ 714 (aa).tv_sec = boottime.tv_sec; \ 715 (aa).tv_nsec = boottime.tv_usec * 1000; \ 716 } \ 717 } while (0) 718 719 if (vp->v_type != VCHR) { 720 fix(de->de_atime); 721 vap->va_atime = de->de_atime; 722 fix(de->de_mtime); 723 vap->va_mtime = de->de_mtime; 724 fix(de->de_ctime); 725 vap->va_ctime = de->de_ctime; 726 } else { 727 dev = vp->v_rdev; 728 fix(dev->si_atime); 729 vap->va_atime = dev->si_atime; 730 fix(dev->si_mtime); 731 vap->va_mtime = dev->si_mtime; 732 fix(dev->si_ctime); 733 vap->va_ctime = dev->si_ctime; 734 735 vap->va_rdev = cdev2priv(dev)->cdp_inode; 736 } 737 vap->va_gen = 0; 738 vap->va_flags = 0; 739 vap->va_filerev = 0; 740 vap->va_nlink = de->de_links; 741 vap->va_fileid = de->de_inode; 742 743 return (error); 744 } 745 746 /* ARGSUSED */ 747 static int 748 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 749 { 750 struct file *fpop; 751 int error; 752 753 fpop = td->td_fpop; 754 td->td_fpop = fp; 755 error = vnops.fo_ioctl(fp, com, data, cred, td); 756 td->td_fpop = fpop; 757 return (error); 758 } 759 760 void * 761 fiodgname_buf_get_ptr(void *fgnp, u_long com) 762 { 763 union { 764 struct fiodgname_arg fgn; 765 #ifdef COMPAT_FREEBSD32 766 struct fiodgname_arg32 fgn32; 767 #endif 768 } *fgnup; 769 770 fgnup = fgnp; 771 switch (com) { 772 case FIODGNAME: 773 return (fgnup->fgn.buf); 774 #ifdef COMPAT_FREEBSD32 775 case FIODGNAME_32: 776 return ((void *)(uintptr_t)fgnup->fgn32.buf); 777 #endif 778 default: 779 panic("Unhandled ioctl command %ld", com); 780 } 781 } 782 783 static int 784 devfs_ioctl(struct vop_ioctl_args *ap) 785 { 786 struct fiodgname_arg *fgn; 787 struct vnode *vpold, *vp; 788 struct cdevsw *dsw; 789 struct thread *td; 790 struct session *sess; 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 /* 833 * Do nothing if reassigning same control tty, or if the 834 * control tty has already disappeared. If it disappeared, 835 * it's because we were racing with TIOCNOTTY. TIOCNOTTY 836 * already took care of releasing the old vnode and we have 837 * nothing left to do. 838 */ 839 sx_slock(&proctree_lock); 840 sess = td->td_proc->p_session; 841 if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) { 842 sx_sunlock(&proctree_lock); 843 return (0); 844 } 845 846 vrefact(vp); 847 SESS_LOCK(sess); 848 vpold = sess->s_ttyvp; 849 sess->s_ttyvp = vp; 850 sess->s_ttydp = cdev2priv(dev); 851 SESS_UNLOCK(sess); 852 853 sx_sunlock(&proctree_lock); 854 855 /* Get rid of reference to old control tty */ 856 if (vpold) 857 vrele(vpold); 858 } 859 return (error); 860 } 861 862 /* ARGSUSED */ 863 static int 864 devfs_kqfilter_f(struct file *fp, struct knote *kn) 865 { 866 struct cdev *dev; 867 struct cdevsw *dsw; 868 int error, ref; 869 struct file *fpop; 870 struct thread *td; 871 872 td = curthread; 873 fpop = td->td_fpop; 874 error = devfs_fp_check(fp, &dev, &dsw, &ref); 875 if (error) 876 return (error); 877 error = dsw->d_kqfilter(dev, kn); 878 td->td_fpop = fpop; 879 dev_relthread(dev, ref); 880 return (error); 881 } 882 883 static inline int 884 devfs_prison_check(struct devfs_dirent *de, struct thread *td) 885 { 886 struct cdev_priv *cdp; 887 struct ucred *dcr; 888 struct proc *p; 889 int error; 890 891 cdp = de->de_cdp; 892 if (cdp == NULL) 893 return (0); 894 dcr = cdp->cdp_c.si_cred; 895 if (dcr == NULL) 896 return (0); 897 898 error = prison_check(td->td_ucred, dcr); 899 if (error == 0) 900 return (0); 901 /* We do, however, allow access to the controlling terminal */ 902 p = td->td_proc; 903 PROC_LOCK(p); 904 if (!(p->p_flag & P_CONTROLT)) { 905 PROC_UNLOCK(p); 906 return (error); 907 } 908 if (p->p_session->s_ttydp == cdp) 909 error = 0; 910 PROC_UNLOCK(p); 911 return (error); 912 } 913 914 static int 915 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 916 { 917 struct componentname *cnp; 918 struct vnode *dvp, **vpp; 919 struct thread *td; 920 struct devfs_dirent *de, *dd; 921 struct devfs_dirent **dde; 922 struct devfs_mount *dmp; 923 struct mount *mp; 924 struct cdev *cdev; 925 int error, flags, nameiop, dvplocked; 926 char specname[SPECNAMELEN + 1], *pname; 927 928 cnp = ap->a_cnp; 929 vpp = ap->a_vpp; 930 dvp = ap->a_dvp; 931 pname = cnp->cn_nameptr; 932 td = cnp->cn_thread; 933 flags = cnp->cn_flags; 934 nameiop = cnp->cn_nameiop; 935 mp = dvp->v_mount; 936 dmp = VFSTODEVFS(mp); 937 dd = dvp->v_data; 938 *vpp = NULLVP; 939 940 if ((flags & ISLASTCN) && nameiop == RENAME) 941 return (EOPNOTSUPP); 942 943 if (dvp->v_type != VDIR) 944 return (ENOTDIR); 945 946 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 947 return (EIO); 948 949 error = vn_dir_check_exec(dvp, cnp); 950 if (error != 0) 951 return (error); 952 953 if (cnp->cn_namelen == 1 && *pname == '.') { 954 if ((flags & ISLASTCN) && nameiop != LOOKUP) 955 return (EINVAL); 956 *vpp = dvp; 957 VREF(dvp); 958 return (0); 959 } 960 961 if (flags & ISDOTDOT) { 962 if ((flags & ISLASTCN) && nameiop != LOOKUP) 963 return (EINVAL); 964 de = devfs_parent_dirent(dd); 965 if (de == NULL) 966 return (ENOENT); 967 dvplocked = VOP_ISLOCKED(dvp); 968 VOP_UNLOCK(dvp); 969 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, 970 vpp); 971 *dm_unlock = 0; 972 vn_lock(dvp, dvplocked | LK_RETRY); 973 return (error); 974 } 975 976 dd = dvp->v_data; 977 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 978 while (de == NULL) { /* While(...) so we can use break */ 979 980 if (nameiop == DELETE) 981 return (ENOENT); 982 983 /* 984 * OK, we didn't have an entry for the name we were asked for 985 * so we try to see if anybody can create it on demand. 986 */ 987 pname = devfs_fqpn(specname, dmp, dd, cnp); 988 if (pname == NULL) 989 break; 990 991 cdev = NULL; 992 DEVFS_DMP_HOLD(dmp); 993 sx_xunlock(&dmp->dm_lock); 994 sx_slock(&clone_drain_lock); 995 EVENTHANDLER_INVOKE(dev_clone, 996 td->td_ucred, pname, strlen(pname), &cdev); 997 sx_sunlock(&clone_drain_lock); 998 999 if (cdev == NULL) 1000 sx_xlock(&dmp->dm_lock); 1001 else if (devfs_populate_vp(dvp) != 0) { 1002 *dm_unlock = 0; 1003 sx_xlock(&dmp->dm_lock); 1004 if (DEVFS_DMP_DROP(dmp)) { 1005 sx_xunlock(&dmp->dm_lock); 1006 devfs_unmount_final(dmp); 1007 } else 1008 sx_xunlock(&dmp->dm_lock); 1009 dev_rel(cdev); 1010 return (ENOENT); 1011 } 1012 if (DEVFS_DMP_DROP(dmp)) { 1013 *dm_unlock = 0; 1014 sx_xunlock(&dmp->dm_lock); 1015 devfs_unmount_final(dmp); 1016 if (cdev != NULL) 1017 dev_rel(cdev); 1018 return (ENOENT); 1019 } 1020 1021 if (cdev == NULL) 1022 break; 1023 1024 dev_lock(); 1025 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 1026 if (dde != NULL && *dde != NULL) 1027 de = *dde; 1028 dev_unlock(); 1029 dev_rel(cdev); 1030 break; 1031 } 1032 1033 if (de == NULL || de->de_flags & DE_WHITEOUT) { 1034 if ((nameiop == CREATE || nameiop == RENAME) && 1035 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 1036 cnp->cn_flags |= SAVENAME; 1037 return (EJUSTRETURN); 1038 } 1039 return (ENOENT); 1040 } 1041 1042 if (devfs_prison_check(de, td)) 1043 return (ENOENT); 1044 1045 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 1046 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 1047 if (error) 1048 return (error); 1049 if (*vpp == dvp) { 1050 VREF(dvp); 1051 *vpp = dvp; 1052 return (0); 1053 } 1054 } 1055 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp); 1056 *dm_unlock = 0; 1057 return (error); 1058 } 1059 1060 static int 1061 devfs_lookup(struct vop_lookup_args *ap) 1062 { 1063 int j; 1064 struct devfs_mount *dmp; 1065 int dm_unlock; 1066 1067 if (devfs_populate_vp(ap->a_dvp) != 0) 1068 return (ENOTDIR); 1069 1070 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1071 dm_unlock = 1; 1072 j = devfs_lookupx(ap, &dm_unlock); 1073 if (dm_unlock == 1) 1074 sx_xunlock(&dmp->dm_lock); 1075 return (j); 1076 } 1077 1078 static int 1079 devfs_mknod(struct vop_mknod_args *ap) 1080 { 1081 struct componentname *cnp; 1082 struct vnode *dvp, **vpp; 1083 struct devfs_dirent *dd, *de; 1084 struct devfs_mount *dmp; 1085 int error; 1086 1087 /* 1088 * The only type of node we should be creating here is a 1089 * character device, for anything else return EOPNOTSUPP. 1090 */ 1091 if (ap->a_vap->va_type != VCHR) 1092 return (EOPNOTSUPP); 1093 dvp = ap->a_dvp; 1094 dmp = VFSTODEVFS(dvp->v_mount); 1095 1096 cnp = ap->a_cnp; 1097 vpp = ap->a_vpp; 1098 dd = dvp->v_data; 1099 1100 error = ENOENT; 1101 sx_xlock(&dmp->dm_lock); 1102 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1103 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1104 continue; 1105 if (de->de_dirent->d_type == DT_CHR && 1106 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1107 continue; 1108 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1109 de->de_dirent->d_namlen) != 0) 1110 continue; 1111 if (de->de_flags & DE_WHITEOUT) 1112 break; 1113 goto notfound; 1114 } 1115 if (de == NULL) 1116 goto notfound; 1117 de->de_flags &= ~DE_WHITEOUT; 1118 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1119 return (error); 1120 notfound: 1121 sx_xunlock(&dmp->dm_lock); 1122 return (error); 1123 } 1124 1125 /* ARGSUSED */ 1126 static int 1127 devfs_open(struct vop_open_args *ap) 1128 { 1129 struct thread *td = ap->a_td; 1130 struct vnode *vp = ap->a_vp; 1131 struct cdev *dev = vp->v_rdev; 1132 struct file *fp = ap->a_fp; 1133 int error, ref, vlocked; 1134 struct cdevsw *dsw; 1135 struct file *fpop; 1136 1137 if (vp->v_type == VBLK) 1138 return (ENXIO); 1139 1140 if (dev == NULL) 1141 return (ENXIO); 1142 1143 /* Make this field valid before any I/O in d_open. */ 1144 if (dev->si_iosize_max == 0) 1145 dev->si_iosize_max = DFLTPHYS; 1146 1147 dsw = dev_refthread(dev, &ref); 1148 if (dsw == NULL) 1149 return (ENXIO); 1150 if (fp == NULL && dsw->d_fdopen != NULL) { 1151 dev_relthread(dev, ref); 1152 return (ENXIO); 1153 } 1154 1155 vlocked = VOP_ISLOCKED(vp); 1156 VOP_UNLOCK(vp); 1157 1158 fpop = td->td_fpop; 1159 td->td_fpop = fp; 1160 if (fp != NULL) { 1161 fp->f_data = dev; 1162 fp->f_vnode = vp; 1163 } 1164 if (dsw->d_fdopen != NULL) 1165 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1166 else 1167 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1168 /* Clean up any cdevpriv upon error. */ 1169 if (error != 0) 1170 devfs_clear_cdevpriv(); 1171 td->td_fpop = fpop; 1172 1173 vn_lock(vp, vlocked | LK_RETRY); 1174 dev_relthread(dev, ref); 1175 if (error != 0) { 1176 if (error == ERESTART) 1177 error = EINTR; 1178 return (error); 1179 } 1180 1181 #if 0 /* /dev/console */ 1182 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1183 #else 1184 if (fp == NULL) 1185 return (error); 1186 #endif 1187 if (fp->f_ops == &badfileops) 1188 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1189 return (error); 1190 } 1191 1192 static int 1193 devfs_pathconf(struct vop_pathconf_args *ap) 1194 { 1195 1196 switch (ap->a_name) { 1197 case _PC_FILESIZEBITS: 1198 *ap->a_retval = 64; 1199 return (0); 1200 case _PC_NAME_MAX: 1201 *ap->a_retval = NAME_MAX; 1202 return (0); 1203 case _PC_LINK_MAX: 1204 *ap->a_retval = INT_MAX; 1205 return (0); 1206 case _PC_SYMLINK_MAX: 1207 *ap->a_retval = MAXPATHLEN; 1208 return (0); 1209 case _PC_MAX_CANON: 1210 if (ap->a_vp->v_vflag & VV_ISTTY) { 1211 *ap->a_retval = MAX_CANON; 1212 return (0); 1213 } 1214 return (EINVAL); 1215 case _PC_MAX_INPUT: 1216 if (ap->a_vp->v_vflag & VV_ISTTY) { 1217 *ap->a_retval = MAX_INPUT; 1218 return (0); 1219 } 1220 return (EINVAL); 1221 case _PC_VDISABLE: 1222 if (ap->a_vp->v_vflag & VV_ISTTY) { 1223 *ap->a_retval = _POSIX_VDISABLE; 1224 return (0); 1225 } 1226 return (EINVAL); 1227 case _PC_MAC_PRESENT: 1228 #ifdef MAC 1229 /* 1230 * If MAC is enabled, devfs automatically supports 1231 * trivial non-persistant label storage. 1232 */ 1233 *ap->a_retval = 1; 1234 #else 1235 *ap->a_retval = 0; 1236 #endif 1237 return (0); 1238 case _PC_CHOWN_RESTRICTED: 1239 *ap->a_retval = 1; 1240 return (0); 1241 default: 1242 return (vop_stdpathconf(ap)); 1243 } 1244 /* NOTREACHED */ 1245 } 1246 1247 /* ARGSUSED */ 1248 static int 1249 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1250 { 1251 struct cdev *dev; 1252 struct cdevsw *dsw; 1253 int error, ref; 1254 struct file *fpop; 1255 1256 fpop = td->td_fpop; 1257 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1258 if (error != 0) { 1259 error = vnops.fo_poll(fp, events, cred, td); 1260 return (error); 1261 } 1262 error = dsw->d_poll(dev, events, td); 1263 td->td_fpop = fpop; 1264 dev_relthread(dev, ref); 1265 return(error); 1266 } 1267 1268 /* 1269 * Print out the contents of a special device vnode. 1270 */ 1271 static int 1272 devfs_print(struct vop_print_args *ap) 1273 { 1274 1275 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1276 return (0); 1277 } 1278 1279 static int 1280 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1281 int flags, struct thread *td) 1282 { 1283 struct cdev *dev; 1284 int ioflag, error, ref; 1285 ssize_t resid; 1286 struct cdevsw *dsw; 1287 struct file *fpop; 1288 1289 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1290 return (EINVAL); 1291 fpop = td->td_fpop; 1292 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1293 if (error != 0) { 1294 error = vnops.fo_read(fp, uio, cred, flags, td); 1295 return (error); 1296 } 1297 resid = uio->uio_resid; 1298 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1299 if (ioflag & O_DIRECT) 1300 ioflag |= IO_DIRECT; 1301 1302 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1303 error = dsw->d_read(dev, uio, ioflag); 1304 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1305 devfs_timestamp(&dev->si_atime); 1306 td->td_fpop = fpop; 1307 dev_relthread(dev, ref); 1308 1309 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R); 1310 return (error); 1311 } 1312 1313 static int 1314 devfs_readdir(struct vop_readdir_args *ap) 1315 { 1316 int error; 1317 struct uio *uio; 1318 struct dirent *dp; 1319 struct devfs_dirent *dd; 1320 struct devfs_dirent *de; 1321 struct devfs_mount *dmp; 1322 off_t off; 1323 int *tmp_ncookies = NULL; 1324 1325 if (ap->a_vp->v_type != VDIR) 1326 return (ENOTDIR); 1327 1328 uio = ap->a_uio; 1329 if (uio->uio_offset < 0) 1330 return (EINVAL); 1331 1332 /* 1333 * XXX: This is a temporary hack to get around this filesystem not 1334 * supporting cookies. We store the location of the ncookies pointer 1335 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1336 * and set the number of cookies to 0. We then set the pointer to 1337 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1338 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1339 * pointer to its original location before returning to the caller. 1340 */ 1341 if (ap->a_ncookies != NULL) { 1342 tmp_ncookies = ap->a_ncookies; 1343 *ap->a_ncookies = 0; 1344 ap->a_ncookies = NULL; 1345 } 1346 1347 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1348 if (devfs_populate_vp(ap->a_vp) != 0) { 1349 if (tmp_ncookies != NULL) 1350 ap->a_ncookies = tmp_ncookies; 1351 return (EIO); 1352 } 1353 error = 0; 1354 de = ap->a_vp->v_data; 1355 off = 0; 1356 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1357 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1358 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1359 continue; 1360 if (devfs_prison_check(dd, uio->uio_td)) 1361 continue; 1362 if (dd->de_dirent->d_type == DT_DIR) 1363 de = dd->de_dir; 1364 else 1365 de = dd; 1366 dp = dd->de_dirent; 1367 MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp)); 1368 if (dp->d_reclen > uio->uio_resid) 1369 break; 1370 dp->d_fileno = de->de_inode; 1371 /* NOTE: d_off is the offset for the *next* entry. */ 1372 dp->d_off = off + dp->d_reclen; 1373 if (off >= uio->uio_offset) { 1374 error = vfs_read_dirent(ap, dp, off); 1375 if (error) 1376 break; 1377 } 1378 off += dp->d_reclen; 1379 } 1380 sx_xunlock(&dmp->dm_lock); 1381 uio->uio_offset = off; 1382 1383 /* 1384 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1385 * place. 1386 */ 1387 if (tmp_ncookies != NULL) 1388 ap->a_ncookies = tmp_ncookies; 1389 1390 return (error); 1391 } 1392 1393 static int 1394 devfs_readlink(struct vop_readlink_args *ap) 1395 { 1396 struct devfs_dirent *de; 1397 1398 de = ap->a_vp->v_data; 1399 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1400 } 1401 1402 static int 1403 devfs_reclaim(struct vop_reclaim_args *ap) 1404 { 1405 struct vnode *vp; 1406 struct devfs_dirent *de; 1407 1408 vp = ap->a_vp; 1409 mtx_lock(&devfs_de_interlock); 1410 de = vp->v_data; 1411 if (de != NULL) { 1412 de->de_vnode = NULL; 1413 vp->v_data = NULL; 1414 } 1415 mtx_unlock(&devfs_de_interlock); 1416 return (0); 1417 } 1418 1419 static int 1420 devfs_reclaim_vchr(struct vop_reclaim_args *ap) 1421 { 1422 struct vnode *vp; 1423 struct cdev *dev; 1424 1425 vp = ap->a_vp; 1426 MPASS(vp->v_type == VCHR); 1427 1428 devfs_reclaim(ap); 1429 1430 VI_LOCK(vp); 1431 dev_lock(); 1432 dev = vp->v_rdev; 1433 vp->v_rdev = NULL; 1434 if (dev != NULL) 1435 dev->si_usecount -= (vp->v_usecount > 0); 1436 dev_unlock(); 1437 VI_UNLOCK(vp); 1438 if (dev != NULL) 1439 dev_rel(dev); 1440 return (0); 1441 } 1442 1443 static int 1444 devfs_remove(struct vop_remove_args *ap) 1445 { 1446 struct vnode *dvp = ap->a_dvp; 1447 struct vnode *vp = ap->a_vp; 1448 struct devfs_dirent *dd; 1449 struct devfs_dirent *de, *de_covered; 1450 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1451 1452 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1453 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1454 1455 sx_xlock(&dmp->dm_lock); 1456 dd = ap->a_dvp->v_data; 1457 de = vp->v_data; 1458 if (de->de_cdp == NULL) { 1459 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1460 if (de->de_dirent->d_type == DT_LNK) { 1461 de_covered = devfs_find(dd, de->de_dirent->d_name, 1462 de->de_dirent->d_namlen, 0); 1463 if (de_covered != NULL) 1464 de_covered->de_flags &= ~DE_COVERED; 1465 } 1466 /* We need to unlock dvp because devfs_delete() may lock it. */ 1467 VOP_UNLOCK(vp); 1468 if (dvp != vp) 1469 VOP_UNLOCK(dvp); 1470 devfs_delete(dmp, de, 0); 1471 sx_xunlock(&dmp->dm_lock); 1472 if (dvp != vp) 1473 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1474 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1475 } else { 1476 de->de_flags |= DE_WHITEOUT; 1477 sx_xunlock(&dmp->dm_lock); 1478 } 1479 return (0); 1480 } 1481 1482 /* 1483 * Revoke is called on a tty when a terminal session ends. The vnode 1484 * is orphaned by setting v_op to deadfs so we need to let go of it 1485 * as well so that we create a new one next time around. 1486 * 1487 */ 1488 static int 1489 devfs_revoke(struct vop_revoke_args *ap) 1490 { 1491 struct vnode *vp = ap->a_vp, *vp2; 1492 struct cdev *dev; 1493 struct cdev_priv *cdp; 1494 struct devfs_dirent *de; 1495 u_int i; 1496 1497 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1498 1499 dev = vp->v_rdev; 1500 cdp = cdev2priv(dev); 1501 1502 dev_lock(); 1503 cdp->cdp_inuse++; 1504 dev_unlock(); 1505 1506 vhold(vp); 1507 vgone(vp); 1508 vdrop(vp); 1509 1510 VOP_UNLOCK(vp); 1511 loop: 1512 for (;;) { 1513 mtx_lock(&devfs_de_interlock); 1514 dev_lock(); 1515 vp2 = NULL; 1516 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1517 de = cdp->cdp_dirents[i]; 1518 if (de == NULL) 1519 continue; 1520 1521 vp2 = de->de_vnode; 1522 if (vp2 != NULL) { 1523 dev_unlock(); 1524 VI_LOCK(vp2); 1525 mtx_unlock(&devfs_de_interlock); 1526 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1527 curthread)) 1528 goto loop; 1529 vhold(vp2); 1530 vgone(vp2); 1531 vdrop(vp2); 1532 vput(vp2); 1533 break; 1534 } 1535 } 1536 if (vp2 != NULL) { 1537 continue; 1538 } 1539 dev_unlock(); 1540 mtx_unlock(&devfs_de_interlock); 1541 break; 1542 } 1543 dev_lock(); 1544 cdp->cdp_inuse--; 1545 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1546 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1547 dev_unlock(); 1548 dev_rel(&cdp->cdp_c); 1549 } else 1550 dev_unlock(); 1551 1552 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1553 return (0); 1554 } 1555 1556 static int 1557 devfs_rioctl(struct vop_ioctl_args *ap) 1558 { 1559 struct vnode *vp; 1560 struct devfs_mount *dmp; 1561 int error; 1562 1563 vp = ap->a_vp; 1564 vn_lock(vp, LK_SHARED | LK_RETRY); 1565 if (VN_IS_DOOMED(vp)) { 1566 VOP_UNLOCK(vp); 1567 return (EBADF); 1568 } 1569 dmp = VFSTODEVFS(vp->v_mount); 1570 sx_xlock(&dmp->dm_lock); 1571 VOP_UNLOCK(vp); 1572 DEVFS_DMP_HOLD(dmp); 1573 devfs_populate(dmp); 1574 if (DEVFS_DMP_DROP(dmp)) { 1575 sx_xunlock(&dmp->dm_lock); 1576 devfs_unmount_final(dmp); 1577 return (ENOENT); 1578 } 1579 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1580 sx_xunlock(&dmp->dm_lock); 1581 return (error); 1582 } 1583 1584 static int 1585 devfs_rread(struct vop_read_args *ap) 1586 { 1587 1588 if (ap->a_vp->v_type != VDIR) 1589 return (EINVAL); 1590 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1591 } 1592 1593 static int 1594 devfs_setattr(struct vop_setattr_args *ap) 1595 { 1596 struct devfs_dirent *de; 1597 struct vattr *vap; 1598 struct vnode *vp; 1599 struct thread *td; 1600 int c, error; 1601 uid_t uid; 1602 gid_t gid; 1603 1604 vap = ap->a_vap; 1605 vp = ap->a_vp; 1606 td = curthread; 1607 if ((vap->va_type != VNON) || 1608 (vap->va_nlink != VNOVAL) || 1609 (vap->va_fsid != VNOVAL) || 1610 (vap->va_fileid != VNOVAL) || 1611 (vap->va_blocksize != VNOVAL) || 1612 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1613 (vap->va_rdev != VNOVAL) || 1614 ((int)vap->va_bytes != VNOVAL) || 1615 (vap->va_gen != VNOVAL)) { 1616 return (EINVAL); 1617 } 1618 1619 error = devfs_populate_vp(vp); 1620 if (error != 0) 1621 return (error); 1622 1623 de = vp->v_data; 1624 if (vp->v_type == VDIR) 1625 de = de->de_dir; 1626 1627 c = 0; 1628 if (vap->va_uid == (uid_t)VNOVAL) 1629 uid = de->de_uid; 1630 else 1631 uid = vap->va_uid; 1632 if (vap->va_gid == (gid_t)VNOVAL) 1633 gid = de->de_gid; 1634 else 1635 gid = vap->va_gid; 1636 if (uid != de->de_uid || gid != de->de_gid) { 1637 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1638 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1639 error = priv_check(td, PRIV_VFS_CHOWN); 1640 if (error != 0) 1641 goto ret; 1642 } 1643 de->de_uid = uid; 1644 de->de_gid = gid; 1645 c = 1; 1646 } 1647 1648 if (vap->va_mode != (mode_t)VNOVAL) { 1649 if (ap->a_cred->cr_uid != de->de_uid) { 1650 error = priv_check(td, PRIV_VFS_ADMIN); 1651 if (error != 0) 1652 goto ret; 1653 } 1654 de->de_mode = vap->va_mode; 1655 c = 1; 1656 } 1657 1658 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1659 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1660 if (error != 0) 1661 goto ret; 1662 if (vap->va_atime.tv_sec != VNOVAL) { 1663 if (vp->v_type == VCHR) 1664 vp->v_rdev->si_atime = vap->va_atime; 1665 else 1666 de->de_atime = vap->va_atime; 1667 } 1668 if (vap->va_mtime.tv_sec != VNOVAL) { 1669 if (vp->v_type == VCHR) 1670 vp->v_rdev->si_mtime = vap->va_mtime; 1671 else 1672 de->de_mtime = vap->va_mtime; 1673 } 1674 c = 1; 1675 } 1676 1677 if (c) { 1678 if (vp->v_type == VCHR) 1679 vfs_timestamp(&vp->v_rdev->si_ctime); 1680 else 1681 vfs_timestamp(&de->de_mtime); 1682 } 1683 1684 ret: 1685 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1686 return (error); 1687 } 1688 1689 #ifdef MAC 1690 static int 1691 devfs_setlabel(struct vop_setlabel_args *ap) 1692 { 1693 struct vnode *vp; 1694 struct devfs_dirent *de; 1695 1696 vp = ap->a_vp; 1697 de = vp->v_data; 1698 1699 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1700 mac_devfs_update(vp->v_mount, de, vp); 1701 1702 return (0); 1703 } 1704 #endif 1705 1706 static int 1707 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1708 { 1709 1710 return (vnops.fo_stat(fp, sb, cred, td)); 1711 } 1712 1713 static int 1714 devfs_symlink(struct vop_symlink_args *ap) 1715 { 1716 int i, error; 1717 struct devfs_dirent *dd; 1718 struct devfs_dirent *de, *de_covered, *de_dotdot; 1719 struct devfs_mount *dmp; 1720 1721 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1722 if (error) 1723 return(error); 1724 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1725 if (devfs_populate_vp(ap->a_dvp) != 0) 1726 return (ENOENT); 1727 1728 dd = ap->a_dvp->v_data; 1729 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1730 de->de_flags = DE_USER; 1731 de->de_uid = 0; 1732 de->de_gid = 0; 1733 de->de_mode = 0755; 1734 de->de_inode = alloc_unr(devfs_inos); 1735 de->de_dir = dd; 1736 de->de_dirent->d_type = DT_LNK; 1737 i = strlen(ap->a_target) + 1; 1738 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1739 bcopy(ap->a_target, de->de_symlink, i); 1740 #ifdef MAC 1741 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1742 #endif 1743 de_covered = devfs_find(dd, de->de_dirent->d_name, 1744 de->de_dirent->d_namlen, 0); 1745 if (de_covered != NULL) { 1746 if ((de_covered->de_flags & DE_USER) != 0) { 1747 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1748 sx_xunlock(&dmp->dm_lock); 1749 return (EEXIST); 1750 } 1751 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1752 ("devfs_symlink: entry %p already covered", de_covered)); 1753 de_covered->de_flags |= DE_COVERED; 1754 } 1755 1756 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1757 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1758 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1759 devfs_dir_ref_de(dmp, dd); 1760 devfs_rules_apply(dmp, de); 1761 1762 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1763 } 1764 1765 static int 1766 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1767 { 1768 1769 return (vnops.fo_truncate(fp, length, cred, td)); 1770 } 1771 1772 static int 1773 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1774 int flags, struct thread *td) 1775 { 1776 struct cdev *dev; 1777 int error, ioflag, ref; 1778 ssize_t resid; 1779 struct cdevsw *dsw; 1780 struct file *fpop; 1781 1782 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1783 return (EINVAL); 1784 fpop = td->td_fpop; 1785 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1786 if (error != 0) { 1787 error = vnops.fo_write(fp, uio, cred, flags, td); 1788 return (error); 1789 } 1790 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1791 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1792 if (ioflag & O_DIRECT) 1793 ioflag |= IO_DIRECT; 1794 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1795 1796 resid = uio->uio_resid; 1797 1798 error = dsw->d_write(dev, uio, ioflag); 1799 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1800 devfs_timestamp(&dev->si_ctime); 1801 dev->si_mtime = dev->si_ctime; 1802 } 1803 td->td_fpop = fpop; 1804 dev_relthread(dev, ref); 1805 1806 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W); 1807 return (error); 1808 } 1809 1810 static int 1811 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1812 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1813 struct thread *td) 1814 { 1815 struct cdev *dev; 1816 struct cdevsw *dsw; 1817 struct mount *mp; 1818 struct vnode *vp; 1819 struct file *fpop; 1820 vm_object_t object; 1821 vm_prot_t maxprot; 1822 int error, ref; 1823 1824 vp = fp->f_vnode; 1825 1826 /* 1827 * Ensure that file and memory protections are 1828 * compatible. 1829 */ 1830 mp = vp->v_mount; 1831 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 1832 maxprot = VM_PROT_NONE; 1833 if ((prot & VM_PROT_EXECUTE) != 0) 1834 return (EACCES); 1835 } else 1836 maxprot = VM_PROT_EXECUTE; 1837 if ((fp->f_flag & FREAD) != 0) 1838 maxprot |= VM_PROT_READ; 1839 else if ((prot & VM_PROT_READ) != 0) 1840 return (EACCES); 1841 1842 /* 1843 * If we are sharing potential changes via MAP_SHARED and we 1844 * are trying to get write permission although we opened it 1845 * without asking for it, bail out. 1846 * 1847 * Note that most character devices always share mappings. 1848 * The one exception is that D_MMAP_ANON devices 1849 * (i.e. /dev/zero) permit private writable mappings. 1850 * 1851 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests 1852 * as well as updating maxprot to permit writing for 1853 * D_MMAP_ANON devices rather than doing that here. 1854 */ 1855 if ((flags & MAP_SHARED) != 0) { 1856 if ((fp->f_flag & FWRITE) != 0) 1857 maxprot |= VM_PROT_WRITE; 1858 else if ((prot & VM_PROT_WRITE) != 0) 1859 return (EACCES); 1860 } 1861 maxprot &= cap_maxprot; 1862 1863 fpop = td->td_fpop; 1864 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1865 if (error != 0) 1866 return (error); 1867 1868 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff, 1869 &object); 1870 td->td_fpop = fpop; 1871 dev_relthread(dev, ref); 1872 if (error != 0) 1873 return (error); 1874 1875 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1876 foff, FALSE, td); 1877 if (error != 0) 1878 vm_object_deallocate(object); 1879 return (error); 1880 } 1881 1882 dev_t 1883 dev2udev(struct cdev *x) 1884 { 1885 if (x == NULL) 1886 return (NODEV); 1887 return (cdev2priv(x)->cdp_inode); 1888 } 1889 1890 static struct fileops devfs_ops_f = { 1891 .fo_read = devfs_read_f, 1892 .fo_write = devfs_write_f, 1893 .fo_truncate = devfs_truncate_f, 1894 .fo_ioctl = devfs_ioctl_f, 1895 .fo_poll = devfs_poll_f, 1896 .fo_kqfilter = devfs_kqfilter_f, 1897 .fo_stat = devfs_stat_f, 1898 .fo_close = devfs_close_f, 1899 .fo_chmod = vn_chmod, 1900 .fo_chown = vn_chown, 1901 .fo_sendfile = vn_sendfile, 1902 .fo_seek = vn_seek, 1903 .fo_fill_kinfo = vn_fill_kinfo, 1904 .fo_mmap = devfs_mmap_f, 1905 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1906 }; 1907 1908 /* Vops for non-CHR vnodes in /dev. */ 1909 static struct vop_vector devfs_vnodeops = { 1910 .vop_default = &default_vnodeops, 1911 1912 .vop_access = devfs_access, 1913 .vop_getattr = devfs_getattr, 1914 .vop_ioctl = devfs_rioctl, 1915 .vop_lookup = devfs_lookup, 1916 .vop_mknod = devfs_mknod, 1917 .vop_pathconf = devfs_pathconf, 1918 .vop_read = devfs_rread, 1919 .vop_readdir = devfs_readdir, 1920 .vop_readlink = devfs_readlink, 1921 .vop_reclaim = devfs_reclaim, 1922 .vop_remove = devfs_remove, 1923 .vop_revoke = devfs_revoke, 1924 .vop_setattr = devfs_setattr, 1925 #ifdef MAC 1926 .vop_setlabel = devfs_setlabel, 1927 #endif 1928 .vop_symlink = devfs_symlink, 1929 .vop_vptocnp = devfs_vptocnp, 1930 }; 1931 VFS_VOP_VECTOR_REGISTER(devfs_vnodeops); 1932 1933 /* Vops for VCHR vnodes in /dev. */ 1934 static struct vop_vector devfs_specops = { 1935 .vop_default = &default_vnodeops, 1936 1937 .vop_access = devfs_access, 1938 .vop_bmap = VOP_PANIC, 1939 .vop_close = devfs_close, 1940 .vop_create = VOP_PANIC, 1941 .vop_fsync = vop_stdfsync, 1942 .vop_getattr = devfs_getattr, 1943 .vop_ioctl = devfs_ioctl, 1944 .vop_link = VOP_PANIC, 1945 .vop_mkdir = VOP_PANIC, 1946 .vop_mknod = VOP_PANIC, 1947 .vop_open = devfs_open, 1948 .vop_pathconf = devfs_pathconf, 1949 .vop_poll = dead_poll, 1950 .vop_print = devfs_print, 1951 .vop_read = dead_read, 1952 .vop_readdir = VOP_PANIC, 1953 .vop_readlink = VOP_PANIC, 1954 .vop_reallocblks = VOP_PANIC, 1955 .vop_reclaim = devfs_reclaim_vchr, 1956 .vop_remove = devfs_remove, 1957 .vop_rename = VOP_PANIC, 1958 .vop_revoke = devfs_revoke, 1959 .vop_rmdir = VOP_PANIC, 1960 .vop_setattr = devfs_setattr, 1961 #ifdef MAC 1962 .vop_setlabel = devfs_setlabel, 1963 #endif 1964 .vop_strategy = VOP_PANIC, 1965 .vop_symlink = VOP_PANIC, 1966 .vop_vptocnp = devfs_vptocnp, 1967 .vop_write = dead_write, 1968 }; 1969 VFS_VOP_VECTOR_REGISTER(devfs_specops); 1970 1971 /* 1972 * Our calling convention to the device drivers used to be that we passed 1973 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1974 * flags instead since that's what open(), close() and ioctl() takes and 1975 * we don't really want vnode.h in device drivers. 1976 * We solved the source compatibility by redefining some vnode flags to 1977 * be the same as the fcntl ones and by sending down the bitwise OR of 1978 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1979 * pulls the rug out under this. 1980 */ 1981 CTASSERT(O_NONBLOCK == IO_NDELAY); 1982 CTASSERT(O_FSYNC == IO_SYNC); 1983