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_fsync(struct vop_fsync_args *ap) 681 { 682 int error; 683 struct bufobj *bo; 684 struct devfs_dirent *de; 685 686 if (!vn_isdisk(ap->a_vp, &error)) { 687 bo = &ap->a_vp->v_bufobj; 688 de = ap->a_vp->v_data; 689 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) { 690 printf("Device %s went missing before all of the data " 691 "could be written to it; expect data loss.\n", 692 de->de_dirent->d_name); 693 694 error = vop_stdfsync(ap); 695 if (bo->bo_dirty.bv_cnt != 0 || error != 0) 696 panic("devfs_fsync: vop_stdfsync failed."); 697 } 698 699 return (0); 700 } 701 702 return (vop_stdfsync(ap)); 703 } 704 705 static int 706 devfs_getattr(struct vop_getattr_args *ap) 707 { 708 struct vnode *vp = ap->a_vp; 709 struct vattr *vap = ap->a_vap; 710 int error; 711 struct devfs_dirent *de; 712 struct devfs_mount *dmp; 713 struct cdev *dev; 714 715 error = devfs_populate_vp(vp); 716 if (error != 0) 717 return (error); 718 719 dmp = VFSTODEVFS(vp->v_mount); 720 sx_xunlock(&dmp->dm_lock); 721 722 de = vp->v_data; 723 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 724 if (vp->v_type == VDIR) { 725 de = de->de_dir; 726 KASSERT(de != NULL, 727 ("Null dir dirent in devfs_getattr vp=%p", vp)); 728 } 729 vap->va_uid = de->de_uid; 730 vap->va_gid = de->de_gid; 731 vap->va_mode = de->de_mode; 732 if (vp->v_type == VLNK) 733 vap->va_size = strlen(de->de_symlink); 734 else if (vp->v_type == VDIR) 735 vap->va_size = vap->va_bytes = DEV_BSIZE; 736 else 737 vap->va_size = 0; 738 if (vp->v_type != VDIR) 739 vap->va_bytes = 0; 740 vap->va_blocksize = DEV_BSIZE; 741 vap->va_type = vp->v_type; 742 743 #define fix(aa) \ 744 do { \ 745 if ((aa).tv_sec <= 3600) { \ 746 (aa).tv_sec = boottime.tv_sec; \ 747 (aa).tv_nsec = boottime.tv_usec * 1000; \ 748 } \ 749 } while (0) 750 751 if (vp->v_type != VCHR) { 752 fix(de->de_atime); 753 vap->va_atime = de->de_atime; 754 fix(de->de_mtime); 755 vap->va_mtime = de->de_mtime; 756 fix(de->de_ctime); 757 vap->va_ctime = de->de_ctime; 758 } else { 759 dev = vp->v_rdev; 760 fix(dev->si_atime); 761 vap->va_atime = dev->si_atime; 762 fix(dev->si_mtime); 763 vap->va_mtime = dev->si_mtime; 764 fix(dev->si_ctime); 765 vap->va_ctime = dev->si_ctime; 766 767 vap->va_rdev = cdev2priv(dev)->cdp_inode; 768 } 769 vap->va_gen = 0; 770 vap->va_flags = 0; 771 vap->va_filerev = 0; 772 vap->va_nlink = de->de_links; 773 vap->va_fileid = de->de_inode; 774 775 return (error); 776 } 777 778 /* ARGSUSED */ 779 static int 780 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 781 { 782 struct cdev *dev; 783 struct cdevsw *dsw; 784 struct vnode *vp; 785 struct vnode *vpold; 786 int error, i, ref; 787 const char *p; 788 struct fiodgname_arg *fgn; 789 struct file *fpop; 790 791 fpop = td->td_fpop; 792 error = devfs_fp_check(fp, &dev, &dsw, &ref); 793 if (error != 0) { 794 error = vnops.fo_ioctl(fp, com, data, cred, td); 795 return (error); 796 } 797 798 if (com == FIODTYPE) { 799 *(int *)data = dsw->d_flags & D_TYPEMASK; 800 td->td_fpop = fpop; 801 dev_relthread(dev, ref); 802 return (0); 803 } else if (com == FIODGNAME) { 804 fgn = data; 805 p = devtoname(dev); 806 i = strlen(p) + 1; 807 if (i > fgn->len) 808 error = EINVAL; 809 else 810 error = copyout(p, fgn->buf, i); 811 td->td_fpop = fpop; 812 dev_relthread(dev, ref); 813 return (error); 814 } 815 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td); 816 td->td_fpop = NULL; 817 dev_relthread(dev, ref); 818 if (error == ENOIOCTL) 819 error = ENOTTY; 820 if (error == 0 && com == TIOCSCTTY) { 821 vp = fp->f_vnode; 822 823 /* Do nothing if reassigning same control tty */ 824 sx_slock(&proctree_lock); 825 if (td->td_proc->p_session->s_ttyvp == vp) { 826 sx_sunlock(&proctree_lock); 827 return (0); 828 } 829 830 vpold = td->td_proc->p_session->s_ttyvp; 831 VREF(vp); 832 SESS_LOCK(td->td_proc->p_session); 833 td->td_proc->p_session->s_ttyvp = vp; 834 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 835 SESS_UNLOCK(td->td_proc->p_session); 836 837 sx_sunlock(&proctree_lock); 838 839 /* Get rid of reference to old control tty */ 840 if (vpold) 841 vrele(vpold); 842 } 843 return (error); 844 } 845 846 /* ARGSUSED */ 847 static int 848 devfs_kqfilter_f(struct file *fp, struct knote *kn) 849 { 850 struct cdev *dev; 851 struct cdevsw *dsw; 852 int error, ref; 853 struct file *fpop; 854 struct thread *td; 855 856 td = curthread; 857 fpop = td->td_fpop; 858 error = devfs_fp_check(fp, &dev, &dsw, &ref); 859 if (error) 860 return (error); 861 error = dsw->d_kqfilter(dev, kn); 862 td->td_fpop = fpop; 863 dev_relthread(dev, ref); 864 return (error); 865 } 866 867 static inline int 868 devfs_prison_check(struct devfs_dirent *de, struct thread *td) 869 { 870 struct cdev_priv *cdp; 871 struct ucred *dcr; 872 struct proc *p; 873 int error; 874 875 cdp = de->de_cdp; 876 if (cdp == NULL) 877 return (0); 878 dcr = cdp->cdp_c.si_cred; 879 if (dcr == NULL) 880 return (0); 881 882 error = prison_check(td->td_ucred, dcr); 883 if (error == 0) 884 return (0); 885 /* We do, however, allow access to the controlling terminal */ 886 p = td->td_proc; 887 PROC_LOCK(p); 888 if (!(p->p_flag & P_CONTROLT)) { 889 PROC_UNLOCK(p); 890 return (error); 891 } 892 if (p->p_session->s_ttydp == cdp) 893 error = 0; 894 PROC_UNLOCK(p); 895 return (error); 896 } 897 898 static int 899 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 900 { 901 struct componentname *cnp; 902 struct vnode *dvp, **vpp; 903 struct thread *td; 904 struct devfs_dirent *de, *dd; 905 struct devfs_dirent **dde; 906 struct devfs_mount *dmp; 907 struct cdev *cdev; 908 int error, flags, nameiop, dvplocked; 909 char specname[SPECNAMELEN + 1], *pname; 910 911 cnp = ap->a_cnp; 912 vpp = ap->a_vpp; 913 dvp = ap->a_dvp; 914 pname = cnp->cn_nameptr; 915 td = cnp->cn_thread; 916 flags = cnp->cn_flags; 917 nameiop = cnp->cn_nameiop; 918 dmp = VFSTODEVFS(dvp->v_mount); 919 dd = dvp->v_data; 920 *vpp = NULLVP; 921 922 if ((flags & ISLASTCN) && nameiop == RENAME) 923 return (EOPNOTSUPP); 924 925 if (dvp->v_type != VDIR) 926 return (ENOTDIR); 927 928 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 929 return (EIO); 930 931 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 932 if (error) 933 return (error); 934 935 if (cnp->cn_namelen == 1 && *pname == '.') { 936 if ((flags & ISLASTCN) && nameiop != LOOKUP) 937 return (EINVAL); 938 *vpp = dvp; 939 VREF(dvp); 940 return (0); 941 } 942 943 if (flags & ISDOTDOT) { 944 if ((flags & ISLASTCN) && nameiop != LOOKUP) 945 return (EINVAL); 946 de = devfs_parent_dirent(dd); 947 if (de == NULL) 948 return (ENOENT); 949 dvplocked = VOP_ISLOCKED(dvp); 950 VOP_UNLOCK(dvp, 0); 951 error = devfs_allocv(de, dvp->v_mount, 952 cnp->cn_lkflags & LK_TYPE_MASK, vpp); 953 *dm_unlock = 0; 954 vn_lock(dvp, dvplocked | LK_RETRY); 955 return (error); 956 } 957 958 dd = dvp->v_data; 959 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 960 while (de == NULL) { /* While(...) so we can use break */ 961 962 if (nameiop == DELETE) 963 return (ENOENT); 964 965 /* 966 * OK, we didn't have an entry for the name we were asked for 967 * so we try to see if anybody can create it on demand. 968 */ 969 pname = devfs_fqpn(specname, dmp, dd, cnp); 970 if (pname == NULL) 971 break; 972 973 cdev = NULL; 974 DEVFS_DMP_HOLD(dmp); 975 sx_xunlock(&dmp->dm_lock); 976 sx_slock(&clone_drain_lock); 977 EVENTHANDLER_INVOKE(dev_clone, 978 td->td_ucred, pname, strlen(pname), &cdev); 979 sx_sunlock(&clone_drain_lock); 980 981 if (cdev == NULL) 982 sx_xlock(&dmp->dm_lock); 983 else if (devfs_populate_vp(dvp) != 0) { 984 *dm_unlock = 0; 985 sx_xlock(&dmp->dm_lock); 986 if (DEVFS_DMP_DROP(dmp)) { 987 sx_xunlock(&dmp->dm_lock); 988 devfs_unmount_final(dmp); 989 } else 990 sx_xunlock(&dmp->dm_lock); 991 dev_rel(cdev); 992 return (ENOENT); 993 } 994 if (DEVFS_DMP_DROP(dmp)) { 995 *dm_unlock = 0; 996 sx_xunlock(&dmp->dm_lock); 997 devfs_unmount_final(dmp); 998 if (cdev != NULL) 999 dev_rel(cdev); 1000 return (ENOENT); 1001 } 1002 1003 if (cdev == NULL) 1004 break; 1005 1006 dev_lock(); 1007 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 1008 if (dde != NULL && *dde != NULL) 1009 de = *dde; 1010 dev_unlock(); 1011 dev_rel(cdev); 1012 break; 1013 } 1014 1015 if (de == NULL || de->de_flags & DE_WHITEOUT) { 1016 if ((nameiop == CREATE || nameiop == RENAME) && 1017 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 1018 cnp->cn_flags |= SAVENAME; 1019 return (EJUSTRETURN); 1020 } 1021 return (ENOENT); 1022 } 1023 1024 if (devfs_prison_check(de, td)) 1025 return (ENOENT); 1026 1027 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 1028 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 1029 if (error) 1030 return (error); 1031 if (*vpp == dvp) { 1032 VREF(dvp); 1033 *vpp = dvp; 1034 return (0); 1035 } 1036 } 1037 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK, 1038 vpp); 1039 *dm_unlock = 0; 1040 return (error); 1041 } 1042 1043 static int 1044 devfs_lookup(struct vop_lookup_args *ap) 1045 { 1046 int j; 1047 struct devfs_mount *dmp; 1048 int dm_unlock; 1049 1050 if (devfs_populate_vp(ap->a_dvp) != 0) 1051 return (ENOTDIR); 1052 1053 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1054 dm_unlock = 1; 1055 j = devfs_lookupx(ap, &dm_unlock); 1056 if (dm_unlock == 1) 1057 sx_xunlock(&dmp->dm_lock); 1058 return (j); 1059 } 1060 1061 static int 1062 devfs_mknod(struct vop_mknod_args *ap) 1063 { 1064 struct componentname *cnp; 1065 struct vnode *dvp, **vpp; 1066 struct devfs_dirent *dd, *de; 1067 struct devfs_mount *dmp; 1068 int error; 1069 1070 /* 1071 * The only type of node we should be creating here is a 1072 * character device, for anything else return EOPNOTSUPP. 1073 */ 1074 if (ap->a_vap->va_type != VCHR) 1075 return (EOPNOTSUPP); 1076 dvp = ap->a_dvp; 1077 dmp = VFSTODEVFS(dvp->v_mount); 1078 1079 cnp = ap->a_cnp; 1080 vpp = ap->a_vpp; 1081 dd = dvp->v_data; 1082 1083 error = ENOENT; 1084 sx_xlock(&dmp->dm_lock); 1085 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1086 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1087 continue; 1088 if (de->de_dirent->d_type == DT_CHR && 1089 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1090 continue; 1091 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1092 de->de_dirent->d_namlen) != 0) 1093 continue; 1094 if (de->de_flags & DE_WHITEOUT) 1095 break; 1096 goto notfound; 1097 } 1098 if (de == NULL) 1099 goto notfound; 1100 de->de_flags &= ~DE_WHITEOUT; 1101 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1102 return (error); 1103 notfound: 1104 sx_xunlock(&dmp->dm_lock); 1105 return (error); 1106 } 1107 1108 /* ARGSUSED */ 1109 static int 1110 devfs_open(struct vop_open_args *ap) 1111 { 1112 struct thread *td = ap->a_td; 1113 struct vnode *vp = ap->a_vp; 1114 struct cdev *dev = vp->v_rdev; 1115 struct file *fp = ap->a_fp; 1116 int error, ref, vlocked; 1117 struct cdevsw *dsw; 1118 struct file *fpop; 1119 struct mtx *mtxp; 1120 1121 if (vp->v_type == VBLK) 1122 return (ENXIO); 1123 1124 if (dev == NULL) 1125 return (ENXIO); 1126 1127 /* Make this field valid before any I/O in d_open. */ 1128 if (dev->si_iosize_max == 0) 1129 dev->si_iosize_max = DFLTPHYS; 1130 1131 dsw = dev_refthread(dev, &ref); 1132 if (dsw == NULL) 1133 return (ENXIO); 1134 if (fp == NULL && dsw->d_fdopen != NULL) { 1135 dev_relthread(dev, ref); 1136 return (ENXIO); 1137 } 1138 1139 vlocked = VOP_ISLOCKED(vp); 1140 VOP_UNLOCK(vp, 0); 1141 1142 fpop = td->td_fpop; 1143 td->td_fpop = fp; 1144 if (fp != NULL) { 1145 fp->f_data = dev; 1146 fp->f_vnode = vp; 1147 } 1148 if (dsw->d_fdopen != NULL) 1149 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1150 else 1151 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1152 /* Clean up any cdevpriv upon error. */ 1153 if (error != 0) 1154 devfs_clear_cdevpriv(); 1155 td->td_fpop = fpop; 1156 1157 vn_lock(vp, vlocked | LK_RETRY); 1158 dev_relthread(dev, ref); 1159 if (error != 0) { 1160 if (error == ERESTART) 1161 error = EINTR; 1162 return (error); 1163 } 1164 1165 #if 0 /* /dev/console */ 1166 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1167 #else 1168 if (fp == NULL) 1169 return (error); 1170 #endif 1171 if (fp->f_ops == &badfileops) 1172 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1173 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1174 1175 /* 1176 * Hint to the dofilewrite() to not force the buffer draining 1177 * on the writer to the file. Most likely, the write would 1178 * not need normal buffers. 1179 */ 1180 mtx_lock(mtxp); 1181 fp->f_vnread_flags |= FDEVFS_VNODE; 1182 mtx_unlock(mtxp); 1183 return (error); 1184 } 1185 1186 static int 1187 devfs_pathconf(struct vop_pathconf_args *ap) 1188 { 1189 1190 switch (ap->a_name) { 1191 case _PC_MAC_PRESENT: 1192 #ifdef MAC 1193 /* 1194 * If MAC is enabled, devfs automatically supports 1195 * trivial non-persistant label storage. 1196 */ 1197 *ap->a_retval = 1; 1198 #else 1199 *ap->a_retval = 0; 1200 #endif 1201 return (0); 1202 default: 1203 return (vop_stdpathconf(ap)); 1204 } 1205 /* NOTREACHED */ 1206 } 1207 1208 /* ARGSUSED */ 1209 static int 1210 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1211 { 1212 struct cdev *dev; 1213 struct cdevsw *dsw; 1214 int error, ref; 1215 struct file *fpop; 1216 1217 fpop = td->td_fpop; 1218 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1219 if (error != 0) { 1220 error = vnops.fo_poll(fp, events, cred, td); 1221 return (error); 1222 } 1223 error = dsw->d_poll(dev, events, td); 1224 td->td_fpop = fpop; 1225 dev_relthread(dev, ref); 1226 return(error); 1227 } 1228 1229 /* 1230 * Print out the contents of a special device vnode. 1231 */ 1232 static int 1233 devfs_print(struct vop_print_args *ap) 1234 { 1235 1236 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1237 return (0); 1238 } 1239 1240 static int 1241 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1242 int flags, struct thread *td) 1243 { 1244 struct cdev *dev; 1245 int ioflag, error, ref; 1246 ssize_t resid; 1247 struct cdevsw *dsw; 1248 struct file *fpop; 1249 1250 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1251 return (EINVAL); 1252 fpop = td->td_fpop; 1253 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1254 if (error != 0) { 1255 error = vnops.fo_read(fp, uio, cred, flags, td); 1256 return (error); 1257 } 1258 resid = uio->uio_resid; 1259 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1260 if (ioflag & O_DIRECT) 1261 ioflag |= IO_DIRECT; 1262 1263 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1264 error = dsw->d_read(dev, uio, ioflag); 1265 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1266 devfs_timestamp(&dev->si_atime); 1267 td->td_fpop = fpop; 1268 dev_relthread(dev, ref); 1269 1270 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1271 return (error); 1272 } 1273 1274 static int 1275 devfs_readdir(struct vop_readdir_args *ap) 1276 { 1277 int error; 1278 struct uio *uio; 1279 struct dirent *dp; 1280 struct devfs_dirent *dd; 1281 struct devfs_dirent *de; 1282 struct devfs_mount *dmp; 1283 off_t off; 1284 int *tmp_ncookies = NULL; 1285 1286 if (ap->a_vp->v_type != VDIR) 1287 return (ENOTDIR); 1288 1289 uio = ap->a_uio; 1290 if (uio->uio_offset < 0) 1291 return (EINVAL); 1292 1293 /* 1294 * XXX: This is a temporary hack to get around this filesystem not 1295 * supporting cookies. We store the location of the ncookies pointer 1296 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1297 * and set the number of cookies to 0. We then set the pointer to 1298 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1299 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1300 * pointer to its original location before returning to the caller. 1301 */ 1302 if (ap->a_ncookies != NULL) { 1303 tmp_ncookies = ap->a_ncookies; 1304 *ap->a_ncookies = 0; 1305 ap->a_ncookies = NULL; 1306 } 1307 1308 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1309 if (devfs_populate_vp(ap->a_vp) != 0) { 1310 if (tmp_ncookies != NULL) 1311 ap->a_ncookies = tmp_ncookies; 1312 return (EIO); 1313 } 1314 error = 0; 1315 de = ap->a_vp->v_data; 1316 off = 0; 1317 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1318 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1319 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1320 continue; 1321 if (devfs_prison_check(dd, uio->uio_td)) 1322 continue; 1323 if (dd->de_dirent->d_type == DT_DIR) 1324 de = dd->de_dir; 1325 else 1326 de = dd; 1327 dp = dd->de_dirent; 1328 if (dp->d_reclen > uio->uio_resid) 1329 break; 1330 dp->d_fileno = de->de_inode; 1331 if (off >= uio->uio_offset) { 1332 error = vfs_read_dirent(ap, dp, off); 1333 if (error) 1334 break; 1335 } 1336 off += dp->d_reclen; 1337 } 1338 sx_xunlock(&dmp->dm_lock); 1339 uio->uio_offset = off; 1340 1341 /* 1342 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1343 * place. 1344 */ 1345 if (tmp_ncookies != NULL) 1346 ap->a_ncookies = tmp_ncookies; 1347 1348 return (error); 1349 } 1350 1351 static int 1352 devfs_readlink(struct vop_readlink_args *ap) 1353 { 1354 struct devfs_dirent *de; 1355 1356 de = ap->a_vp->v_data; 1357 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1358 } 1359 1360 static int 1361 devfs_reclaim(struct vop_reclaim_args *ap) 1362 { 1363 struct vnode *vp = ap->a_vp; 1364 struct devfs_dirent *de; 1365 struct cdev *dev; 1366 1367 mtx_lock(&devfs_de_interlock); 1368 de = vp->v_data; 1369 if (de != NULL) { 1370 de->de_vnode = NULL; 1371 vp->v_data = NULL; 1372 } 1373 mtx_unlock(&devfs_de_interlock); 1374 1375 vnode_destroy_vobject(vp); 1376 1377 VI_LOCK(vp); 1378 dev_lock(); 1379 dev = vp->v_rdev; 1380 vp->v_rdev = NULL; 1381 1382 if (dev == NULL) { 1383 dev_unlock(); 1384 VI_UNLOCK(vp); 1385 return (0); 1386 } 1387 1388 dev->si_usecount -= vp->v_usecount; 1389 dev_unlock(); 1390 VI_UNLOCK(vp); 1391 dev_rel(dev); 1392 return (0); 1393 } 1394 1395 static int 1396 devfs_remove(struct vop_remove_args *ap) 1397 { 1398 struct vnode *dvp = ap->a_dvp; 1399 struct vnode *vp = ap->a_vp; 1400 struct devfs_dirent *dd; 1401 struct devfs_dirent *de, *de_covered; 1402 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1403 1404 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1405 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1406 1407 sx_xlock(&dmp->dm_lock); 1408 dd = ap->a_dvp->v_data; 1409 de = vp->v_data; 1410 if (de->de_cdp == NULL) { 1411 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1412 if (de->de_dirent->d_type == DT_LNK) { 1413 de_covered = devfs_find(dd, de->de_dirent->d_name, 1414 de->de_dirent->d_namlen, 0); 1415 if (de_covered != NULL) 1416 de_covered->de_flags &= ~DE_COVERED; 1417 } 1418 /* We need to unlock dvp because devfs_delete() may lock it. */ 1419 VOP_UNLOCK(vp, 0); 1420 if (dvp != vp) 1421 VOP_UNLOCK(dvp, 0); 1422 devfs_delete(dmp, de, 0); 1423 sx_xunlock(&dmp->dm_lock); 1424 if (dvp != vp) 1425 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1426 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1427 } else { 1428 de->de_flags |= DE_WHITEOUT; 1429 sx_xunlock(&dmp->dm_lock); 1430 } 1431 return (0); 1432 } 1433 1434 /* 1435 * Revoke is called on a tty when a terminal session ends. The vnode 1436 * is orphaned by setting v_op to deadfs so we need to let go of it 1437 * as well so that we create a new one next time around. 1438 * 1439 */ 1440 static int 1441 devfs_revoke(struct vop_revoke_args *ap) 1442 { 1443 struct vnode *vp = ap->a_vp, *vp2; 1444 struct cdev *dev; 1445 struct cdev_priv *cdp; 1446 struct devfs_dirent *de; 1447 u_int i; 1448 1449 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1450 1451 dev = vp->v_rdev; 1452 cdp = cdev2priv(dev); 1453 1454 dev_lock(); 1455 cdp->cdp_inuse++; 1456 dev_unlock(); 1457 1458 vhold(vp); 1459 vgone(vp); 1460 vdrop(vp); 1461 1462 VOP_UNLOCK(vp,0); 1463 loop: 1464 for (;;) { 1465 mtx_lock(&devfs_de_interlock); 1466 dev_lock(); 1467 vp2 = NULL; 1468 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1469 de = cdp->cdp_dirents[i]; 1470 if (de == NULL) 1471 continue; 1472 1473 vp2 = de->de_vnode; 1474 if (vp2 != NULL) { 1475 dev_unlock(); 1476 VI_LOCK(vp2); 1477 mtx_unlock(&devfs_de_interlock); 1478 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1479 curthread)) 1480 goto loop; 1481 vhold(vp2); 1482 vgone(vp2); 1483 vdrop(vp2); 1484 vput(vp2); 1485 break; 1486 } 1487 } 1488 if (vp2 != NULL) { 1489 continue; 1490 } 1491 dev_unlock(); 1492 mtx_unlock(&devfs_de_interlock); 1493 break; 1494 } 1495 dev_lock(); 1496 cdp->cdp_inuse--; 1497 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1498 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1499 dev_unlock(); 1500 dev_rel(&cdp->cdp_c); 1501 } else 1502 dev_unlock(); 1503 1504 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1505 return (0); 1506 } 1507 1508 static int 1509 devfs_rioctl(struct vop_ioctl_args *ap) 1510 { 1511 struct vnode *vp; 1512 struct devfs_mount *dmp; 1513 int error; 1514 1515 vp = ap->a_vp; 1516 vn_lock(vp, LK_SHARED | LK_RETRY); 1517 if (vp->v_iflag & VI_DOOMED) { 1518 VOP_UNLOCK(vp, 0); 1519 return (EBADF); 1520 } 1521 dmp = VFSTODEVFS(vp->v_mount); 1522 sx_xlock(&dmp->dm_lock); 1523 VOP_UNLOCK(vp, 0); 1524 DEVFS_DMP_HOLD(dmp); 1525 devfs_populate(dmp); 1526 if (DEVFS_DMP_DROP(dmp)) { 1527 sx_xunlock(&dmp->dm_lock); 1528 devfs_unmount_final(dmp); 1529 return (ENOENT); 1530 } 1531 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1532 sx_xunlock(&dmp->dm_lock); 1533 return (error); 1534 } 1535 1536 static int 1537 devfs_rread(struct vop_read_args *ap) 1538 { 1539 1540 if (ap->a_vp->v_type != VDIR) 1541 return (EINVAL); 1542 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1543 } 1544 1545 static int 1546 devfs_setattr(struct vop_setattr_args *ap) 1547 { 1548 struct devfs_dirent *de; 1549 struct vattr *vap; 1550 struct vnode *vp; 1551 struct thread *td; 1552 int c, error; 1553 uid_t uid; 1554 gid_t gid; 1555 1556 vap = ap->a_vap; 1557 vp = ap->a_vp; 1558 td = curthread; 1559 if ((vap->va_type != VNON) || 1560 (vap->va_nlink != VNOVAL) || 1561 (vap->va_fsid != VNOVAL) || 1562 (vap->va_fileid != VNOVAL) || 1563 (vap->va_blocksize != VNOVAL) || 1564 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1565 (vap->va_rdev != VNOVAL) || 1566 ((int)vap->va_bytes != VNOVAL) || 1567 (vap->va_gen != VNOVAL)) { 1568 return (EINVAL); 1569 } 1570 1571 error = devfs_populate_vp(vp); 1572 if (error != 0) 1573 return (error); 1574 1575 de = vp->v_data; 1576 if (vp->v_type == VDIR) 1577 de = de->de_dir; 1578 1579 c = 0; 1580 if (vap->va_uid == (uid_t)VNOVAL) 1581 uid = de->de_uid; 1582 else 1583 uid = vap->va_uid; 1584 if (vap->va_gid == (gid_t)VNOVAL) 1585 gid = de->de_gid; 1586 else 1587 gid = vap->va_gid; 1588 if (uid != de->de_uid || gid != de->de_gid) { 1589 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1590 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1591 error = priv_check(td, PRIV_VFS_CHOWN); 1592 if (error != 0) 1593 goto ret; 1594 } 1595 de->de_uid = uid; 1596 de->de_gid = gid; 1597 c = 1; 1598 } 1599 1600 if (vap->va_mode != (mode_t)VNOVAL) { 1601 if (ap->a_cred->cr_uid != de->de_uid) { 1602 error = priv_check(td, PRIV_VFS_ADMIN); 1603 if (error != 0) 1604 goto ret; 1605 } 1606 de->de_mode = vap->va_mode; 1607 c = 1; 1608 } 1609 1610 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1611 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1612 if (error != 0) 1613 goto ret; 1614 if (vap->va_atime.tv_sec != VNOVAL) { 1615 if (vp->v_type == VCHR) 1616 vp->v_rdev->si_atime = vap->va_atime; 1617 else 1618 de->de_atime = vap->va_atime; 1619 } 1620 if (vap->va_mtime.tv_sec != VNOVAL) { 1621 if (vp->v_type == VCHR) 1622 vp->v_rdev->si_mtime = vap->va_mtime; 1623 else 1624 de->de_mtime = vap->va_mtime; 1625 } 1626 c = 1; 1627 } 1628 1629 if (c) { 1630 if (vp->v_type == VCHR) 1631 vfs_timestamp(&vp->v_rdev->si_ctime); 1632 else 1633 vfs_timestamp(&de->de_mtime); 1634 } 1635 1636 ret: 1637 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1638 return (error); 1639 } 1640 1641 #ifdef MAC 1642 static int 1643 devfs_setlabel(struct vop_setlabel_args *ap) 1644 { 1645 struct vnode *vp; 1646 struct devfs_dirent *de; 1647 1648 vp = ap->a_vp; 1649 de = vp->v_data; 1650 1651 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1652 mac_devfs_update(vp->v_mount, de, vp); 1653 1654 return (0); 1655 } 1656 #endif 1657 1658 static int 1659 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1660 { 1661 1662 return (vnops.fo_stat(fp, sb, cred, td)); 1663 } 1664 1665 static int 1666 devfs_symlink(struct vop_symlink_args *ap) 1667 { 1668 int i, error; 1669 struct devfs_dirent *dd; 1670 struct devfs_dirent *de, *de_covered, *de_dotdot; 1671 struct devfs_mount *dmp; 1672 1673 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1674 if (error) 1675 return(error); 1676 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1677 if (devfs_populate_vp(ap->a_dvp) != 0) 1678 return (ENOENT); 1679 1680 dd = ap->a_dvp->v_data; 1681 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1682 de->de_flags = DE_USER; 1683 de->de_uid = 0; 1684 de->de_gid = 0; 1685 de->de_mode = 0755; 1686 de->de_inode = alloc_unr(devfs_inos); 1687 de->de_dir = dd; 1688 de->de_dirent->d_type = DT_LNK; 1689 i = strlen(ap->a_target) + 1; 1690 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1691 bcopy(ap->a_target, de->de_symlink, i); 1692 #ifdef MAC 1693 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1694 #endif 1695 de_covered = devfs_find(dd, de->de_dirent->d_name, 1696 de->de_dirent->d_namlen, 0); 1697 if (de_covered != NULL) { 1698 if ((de_covered->de_flags & DE_USER) != 0) { 1699 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1700 sx_xunlock(&dmp->dm_lock); 1701 return (EEXIST); 1702 } 1703 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1704 ("devfs_symlink: entry %p already covered", de_covered)); 1705 de_covered->de_flags |= DE_COVERED; 1706 } 1707 1708 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1709 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1710 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1711 devfs_dir_ref_de(dmp, dd); 1712 devfs_rules_apply(dmp, de); 1713 1714 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1715 } 1716 1717 static int 1718 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1719 { 1720 1721 return (vnops.fo_truncate(fp, length, cred, td)); 1722 } 1723 1724 static int 1725 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1726 int flags, struct thread *td) 1727 { 1728 struct cdev *dev; 1729 int error, ioflag, ref; 1730 ssize_t resid; 1731 struct cdevsw *dsw; 1732 struct file *fpop; 1733 1734 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1735 return (EINVAL); 1736 fpop = td->td_fpop; 1737 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1738 if (error != 0) { 1739 error = vnops.fo_write(fp, uio, cred, flags, td); 1740 return (error); 1741 } 1742 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1743 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1744 if (ioflag & O_DIRECT) 1745 ioflag |= IO_DIRECT; 1746 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1747 1748 resid = uio->uio_resid; 1749 1750 error = dsw->d_write(dev, uio, ioflag); 1751 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1752 devfs_timestamp(&dev->si_ctime); 1753 dev->si_mtime = dev->si_ctime; 1754 } 1755 td->td_fpop = fpop; 1756 dev_relthread(dev, ref); 1757 1758 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1759 return (error); 1760 } 1761 1762 static int 1763 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1764 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1765 struct thread *td) 1766 { 1767 struct cdev *dev; 1768 struct cdevsw *dsw; 1769 struct mount *mp; 1770 struct vnode *vp; 1771 struct file *fpop; 1772 vm_object_t object; 1773 vm_prot_t maxprot; 1774 int error, ref; 1775 1776 vp = fp->f_vnode; 1777 1778 /* 1779 * Ensure that file and memory protections are 1780 * compatible. 1781 */ 1782 mp = vp->v_mount; 1783 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) 1784 maxprot = VM_PROT_NONE; 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 static struct vop_vector devfs_vnodeops = { 1859 .vop_default = &default_vnodeops, 1860 1861 .vop_access = devfs_access, 1862 .vop_getattr = devfs_getattr, 1863 .vop_ioctl = devfs_rioctl, 1864 .vop_lookup = devfs_lookup, 1865 .vop_mknod = devfs_mknod, 1866 .vop_pathconf = devfs_pathconf, 1867 .vop_read = devfs_rread, 1868 .vop_readdir = devfs_readdir, 1869 .vop_readlink = devfs_readlink, 1870 .vop_reclaim = devfs_reclaim, 1871 .vop_remove = devfs_remove, 1872 .vop_revoke = devfs_revoke, 1873 .vop_setattr = devfs_setattr, 1874 #ifdef MAC 1875 .vop_setlabel = devfs_setlabel, 1876 #endif 1877 .vop_symlink = devfs_symlink, 1878 .vop_vptocnp = devfs_vptocnp, 1879 }; 1880 1881 static struct vop_vector devfs_specops = { 1882 .vop_default = &default_vnodeops, 1883 1884 .vop_access = devfs_access, 1885 .vop_bmap = VOP_PANIC, 1886 .vop_close = devfs_close, 1887 .vop_create = VOP_PANIC, 1888 .vop_fsync = devfs_fsync, 1889 .vop_getattr = devfs_getattr, 1890 .vop_link = VOP_PANIC, 1891 .vop_mkdir = VOP_PANIC, 1892 .vop_mknod = VOP_PANIC, 1893 .vop_open = devfs_open, 1894 .vop_pathconf = devfs_pathconf, 1895 .vop_poll = dead_poll, 1896 .vop_print = devfs_print, 1897 .vop_read = dead_read, 1898 .vop_readdir = VOP_PANIC, 1899 .vop_readlink = VOP_PANIC, 1900 .vop_reallocblks = VOP_PANIC, 1901 .vop_reclaim = devfs_reclaim, 1902 .vop_remove = devfs_remove, 1903 .vop_rename = VOP_PANIC, 1904 .vop_revoke = devfs_revoke, 1905 .vop_rmdir = VOP_PANIC, 1906 .vop_setattr = devfs_setattr, 1907 #ifdef MAC 1908 .vop_setlabel = devfs_setlabel, 1909 #endif 1910 .vop_strategy = VOP_PANIC, 1911 .vop_symlink = VOP_PANIC, 1912 .vop_vptocnp = devfs_vptocnp, 1913 .vop_write = dead_write, 1914 }; 1915 1916 /* 1917 * Our calling convention to the device drivers used to be that we passed 1918 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1919 * flags instead since that's what open(), close() and ioctl() takes and 1920 * we don't really want vnode.h in device drivers. 1921 * We solved the source compatibility by redefining some vnode flags to 1922 * be the same as the fcntl ones and by sending down the bitwise OR of 1923 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1924 * pulls the rug out under this. 1925 */ 1926 CTASSERT(O_NONBLOCK == IO_NDELAY); 1927 CTASSERT(O_FSYNC == IO_SYNC); 1928