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