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 struct devfs_dirent *de; 711 struct devfs_mount *dmp; 712 struct cdev *dev; 713 struct timeval boottime; 714 int error; 715 716 error = devfs_populate_vp(vp); 717 if (error != 0) 718 return (error); 719 720 dmp = VFSTODEVFS(vp->v_mount); 721 sx_xunlock(&dmp->dm_lock); 722 723 de = vp->v_data; 724 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 725 if (vp->v_type == VDIR) { 726 de = de->de_dir; 727 KASSERT(de != NULL, 728 ("Null dir dirent in devfs_getattr vp=%p", vp)); 729 } 730 vap->va_uid = de->de_uid; 731 vap->va_gid = de->de_gid; 732 vap->va_mode = de->de_mode; 733 if (vp->v_type == VLNK) 734 vap->va_size = strlen(de->de_symlink); 735 else if (vp->v_type == VDIR) 736 vap->va_size = vap->va_bytes = DEV_BSIZE; 737 else 738 vap->va_size = 0; 739 if (vp->v_type != VDIR) 740 vap->va_bytes = 0; 741 vap->va_blocksize = DEV_BSIZE; 742 vap->va_type = vp->v_type; 743 744 getboottime(&boottime); 745 #define fix(aa) \ 746 do { \ 747 if ((aa).tv_sec <= 3600) { \ 748 (aa).tv_sec = boottime.tv_sec; \ 749 (aa).tv_nsec = boottime.tv_usec * 1000; \ 750 } \ 751 } while (0) 752 753 if (vp->v_type != VCHR) { 754 fix(de->de_atime); 755 vap->va_atime = de->de_atime; 756 fix(de->de_mtime); 757 vap->va_mtime = de->de_mtime; 758 fix(de->de_ctime); 759 vap->va_ctime = de->de_ctime; 760 } else { 761 dev = vp->v_rdev; 762 fix(dev->si_atime); 763 vap->va_atime = dev->si_atime; 764 fix(dev->si_mtime); 765 vap->va_mtime = dev->si_mtime; 766 fix(dev->si_ctime); 767 vap->va_ctime = dev->si_ctime; 768 769 vap->va_rdev = cdev2priv(dev)->cdp_inode; 770 } 771 vap->va_gen = 0; 772 vap->va_flags = 0; 773 vap->va_filerev = 0; 774 vap->va_nlink = de->de_links; 775 vap->va_fileid = de->de_inode; 776 777 return (error); 778 } 779 780 /* ARGSUSED */ 781 static int 782 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 783 { 784 struct file *fpop; 785 int error; 786 787 fpop = td->td_fpop; 788 td->td_fpop = fp; 789 error = vnops.fo_ioctl(fp, com, data, cred, td); 790 td->td_fpop = fpop; 791 return (error); 792 } 793 794 static int 795 devfs_ioctl(struct vop_ioctl_args *ap) 796 { 797 struct fiodgname_arg *fgn; 798 struct vnode *vpold, *vp; 799 struct cdevsw *dsw; 800 struct thread *td; 801 struct cdev *dev; 802 int error, ref, i; 803 const char *p; 804 u_long com; 805 806 vp = ap->a_vp; 807 com = ap->a_command; 808 td = ap->a_td; 809 810 dsw = devvn_refthread(vp, &dev, &ref); 811 if (dsw == NULL) 812 return (ENXIO); 813 KASSERT(dev->si_refcount > 0, 814 ("devfs: un-referenced struct cdev *(%s)", devtoname(dev))); 815 816 if (com == FIODTYPE) { 817 *(int *)ap->a_data = dsw->d_flags & D_TYPEMASK; 818 error = 0; 819 goto out; 820 } else if (com == FIODGNAME) { 821 fgn = ap->a_data; 822 p = devtoname(dev); 823 i = strlen(p) + 1; 824 if (i > fgn->len) 825 error = EINVAL; 826 else 827 error = copyout(p, fgn->buf, i); 828 goto out; 829 } 830 831 error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td); 832 833 out: 834 dev_relthread(dev, ref); 835 if (error == ENOIOCTL) 836 error = ENOTTY; 837 838 if (error == 0 && com == TIOCSCTTY) { 839 /* Do nothing if reassigning same control tty */ 840 sx_slock(&proctree_lock); 841 if (td->td_proc->p_session->s_ttyvp == vp) { 842 sx_sunlock(&proctree_lock); 843 return (0); 844 } 845 846 vpold = td->td_proc->p_session->s_ttyvp; 847 VREF(vp); 848 SESS_LOCK(td->td_proc->p_session); 849 td->td_proc->p_session->s_ttyvp = vp; 850 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 851 SESS_UNLOCK(td->td_proc->p_session); 852 853 sx_sunlock(&proctree_lock); 854 855 /* Get rid of reference to old control tty */ 856 if (vpold) 857 vrele(vpold); 858 } 859 return (error); 860 } 861 862 /* ARGSUSED */ 863 static int 864 devfs_kqfilter_f(struct file *fp, struct knote *kn) 865 { 866 struct cdev *dev; 867 struct cdevsw *dsw; 868 int error, ref; 869 struct file *fpop; 870 struct thread *td; 871 872 td = curthread; 873 fpop = td->td_fpop; 874 error = devfs_fp_check(fp, &dev, &dsw, &ref); 875 if (error) 876 return (error); 877 error = dsw->d_kqfilter(dev, kn); 878 td->td_fpop = fpop; 879 dev_relthread(dev, ref); 880 return (error); 881 } 882 883 static inline int 884 devfs_prison_check(struct devfs_dirent *de, struct thread *td) 885 { 886 struct cdev_priv *cdp; 887 struct ucred *dcr; 888 struct proc *p; 889 int error; 890 891 cdp = de->de_cdp; 892 if (cdp == NULL) 893 return (0); 894 dcr = cdp->cdp_c.si_cred; 895 if (dcr == NULL) 896 return (0); 897 898 error = prison_check(td->td_ucred, dcr); 899 if (error == 0) 900 return (0); 901 /* We do, however, allow access to the controlling terminal */ 902 p = td->td_proc; 903 PROC_LOCK(p); 904 if (!(p->p_flag & P_CONTROLT)) { 905 PROC_UNLOCK(p); 906 return (error); 907 } 908 if (p->p_session->s_ttydp == cdp) 909 error = 0; 910 PROC_UNLOCK(p); 911 return (error); 912 } 913 914 static int 915 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 916 { 917 struct componentname *cnp; 918 struct vnode *dvp, **vpp; 919 struct thread *td; 920 struct devfs_dirent *de, *dd; 921 struct devfs_dirent **dde; 922 struct devfs_mount *dmp; 923 struct cdev *cdev; 924 int error, flags, nameiop, dvplocked; 925 char specname[SPECNAMELEN + 1], *pname; 926 927 cnp = ap->a_cnp; 928 vpp = ap->a_vpp; 929 dvp = ap->a_dvp; 930 pname = cnp->cn_nameptr; 931 td = cnp->cn_thread; 932 flags = cnp->cn_flags; 933 nameiop = cnp->cn_nameiop; 934 dmp = VFSTODEVFS(dvp->v_mount); 935 dd = dvp->v_data; 936 *vpp = NULLVP; 937 938 if ((flags & ISLASTCN) && nameiop == RENAME) 939 return (EOPNOTSUPP); 940 941 if (dvp->v_type != VDIR) 942 return (ENOTDIR); 943 944 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 945 return (EIO); 946 947 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 948 if (error) 949 return (error); 950 951 if (cnp->cn_namelen == 1 && *pname == '.') { 952 if ((flags & ISLASTCN) && nameiop != LOOKUP) 953 return (EINVAL); 954 *vpp = dvp; 955 VREF(dvp); 956 return (0); 957 } 958 959 if (flags & ISDOTDOT) { 960 if ((flags & ISLASTCN) && nameiop != LOOKUP) 961 return (EINVAL); 962 de = devfs_parent_dirent(dd); 963 if (de == NULL) 964 return (ENOENT); 965 dvplocked = VOP_ISLOCKED(dvp); 966 VOP_UNLOCK(dvp, 0); 967 error = devfs_allocv(de, dvp->v_mount, 968 cnp->cn_lkflags & LK_TYPE_MASK, vpp); 969 *dm_unlock = 0; 970 vn_lock(dvp, dvplocked | LK_RETRY); 971 return (error); 972 } 973 974 dd = dvp->v_data; 975 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 976 while (de == NULL) { /* While(...) so we can use break */ 977 978 if (nameiop == DELETE) 979 return (ENOENT); 980 981 /* 982 * OK, we didn't have an entry for the name we were asked for 983 * so we try to see if anybody can create it on demand. 984 */ 985 pname = devfs_fqpn(specname, dmp, dd, cnp); 986 if (pname == NULL) 987 break; 988 989 cdev = NULL; 990 DEVFS_DMP_HOLD(dmp); 991 sx_xunlock(&dmp->dm_lock); 992 sx_slock(&clone_drain_lock); 993 EVENTHANDLER_INVOKE(dev_clone, 994 td->td_ucred, pname, strlen(pname), &cdev); 995 sx_sunlock(&clone_drain_lock); 996 997 if (cdev == NULL) 998 sx_xlock(&dmp->dm_lock); 999 else if (devfs_populate_vp(dvp) != 0) { 1000 *dm_unlock = 0; 1001 sx_xlock(&dmp->dm_lock); 1002 if (DEVFS_DMP_DROP(dmp)) { 1003 sx_xunlock(&dmp->dm_lock); 1004 devfs_unmount_final(dmp); 1005 } else 1006 sx_xunlock(&dmp->dm_lock); 1007 dev_rel(cdev); 1008 return (ENOENT); 1009 } 1010 if (DEVFS_DMP_DROP(dmp)) { 1011 *dm_unlock = 0; 1012 sx_xunlock(&dmp->dm_lock); 1013 devfs_unmount_final(dmp); 1014 if (cdev != NULL) 1015 dev_rel(cdev); 1016 return (ENOENT); 1017 } 1018 1019 if (cdev == NULL) 1020 break; 1021 1022 dev_lock(); 1023 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 1024 if (dde != NULL && *dde != NULL) 1025 de = *dde; 1026 dev_unlock(); 1027 dev_rel(cdev); 1028 break; 1029 } 1030 1031 if (de == NULL || de->de_flags & DE_WHITEOUT) { 1032 if ((nameiop == CREATE || nameiop == RENAME) && 1033 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 1034 cnp->cn_flags |= SAVENAME; 1035 return (EJUSTRETURN); 1036 } 1037 return (ENOENT); 1038 } 1039 1040 if (devfs_prison_check(de, td)) 1041 return (ENOENT); 1042 1043 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 1044 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 1045 if (error) 1046 return (error); 1047 if (*vpp == dvp) { 1048 VREF(dvp); 1049 *vpp = dvp; 1050 return (0); 1051 } 1052 } 1053 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK, 1054 vpp); 1055 *dm_unlock = 0; 1056 return (error); 1057 } 1058 1059 static int 1060 devfs_lookup(struct vop_lookup_args *ap) 1061 { 1062 int j; 1063 struct devfs_mount *dmp; 1064 int dm_unlock; 1065 1066 if (devfs_populate_vp(ap->a_dvp) != 0) 1067 return (ENOTDIR); 1068 1069 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1070 dm_unlock = 1; 1071 j = devfs_lookupx(ap, &dm_unlock); 1072 if (dm_unlock == 1) 1073 sx_xunlock(&dmp->dm_lock); 1074 return (j); 1075 } 1076 1077 static int 1078 devfs_mknod(struct vop_mknod_args *ap) 1079 { 1080 struct componentname *cnp; 1081 struct vnode *dvp, **vpp; 1082 struct devfs_dirent *dd, *de; 1083 struct devfs_mount *dmp; 1084 int error; 1085 1086 /* 1087 * The only type of node we should be creating here is a 1088 * character device, for anything else return EOPNOTSUPP. 1089 */ 1090 if (ap->a_vap->va_type != VCHR) 1091 return (EOPNOTSUPP); 1092 dvp = ap->a_dvp; 1093 dmp = VFSTODEVFS(dvp->v_mount); 1094 1095 cnp = ap->a_cnp; 1096 vpp = ap->a_vpp; 1097 dd = dvp->v_data; 1098 1099 error = ENOENT; 1100 sx_xlock(&dmp->dm_lock); 1101 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1102 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1103 continue; 1104 if (de->de_dirent->d_type == DT_CHR && 1105 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1106 continue; 1107 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1108 de->de_dirent->d_namlen) != 0) 1109 continue; 1110 if (de->de_flags & DE_WHITEOUT) 1111 break; 1112 goto notfound; 1113 } 1114 if (de == NULL) 1115 goto notfound; 1116 de->de_flags &= ~DE_WHITEOUT; 1117 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1118 return (error); 1119 notfound: 1120 sx_xunlock(&dmp->dm_lock); 1121 return (error); 1122 } 1123 1124 /* ARGSUSED */ 1125 static int 1126 devfs_open(struct vop_open_args *ap) 1127 { 1128 struct thread *td = ap->a_td; 1129 struct vnode *vp = ap->a_vp; 1130 struct cdev *dev = vp->v_rdev; 1131 struct file *fp = ap->a_fp; 1132 int error, ref, vlocked; 1133 struct cdevsw *dsw; 1134 struct file *fpop; 1135 struct mtx *mtxp; 1136 1137 if (vp->v_type == VBLK) 1138 return (ENXIO); 1139 1140 if (dev == NULL) 1141 return (ENXIO); 1142 1143 /* Make this field valid before any I/O in d_open. */ 1144 if (dev->si_iosize_max == 0) 1145 dev->si_iosize_max = DFLTPHYS; 1146 1147 dsw = dev_refthread(dev, &ref); 1148 if (dsw == NULL) 1149 return (ENXIO); 1150 if (fp == NULL && dsw->d_fdopen != NULL) { 1151 dev_relthread(dev, ref); 1152 return (ENXIO); 1153 } 1154 1155 vlocked = VOP_ISLOCKED(vp); 1156 VOP_UNLOCK(vp, 0); 1157 1158 fpop = td->td_fpop; 1159 td->td_fpop = fp; 1160 if (fp != NULL) { 1161 fp->f_data = dev; 1162 fp->f_vnode = vp; 1163 } 1164 if (dsw->d_fdopen != NULL) 1165 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1166 else 1167 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1168 /* Clean up any cdevpriv upon error. */ 1169 if (error != 0) 1170 devfs_clear_cdevpriv(); 1171 td->td_fpop = fpop; 1172 1173 vn_lock(vp, vlocked | LK_RETRY); 1174 dev_relthread(dev, ref); 1175 if (error != 0) { 1176 if (error == ERESTART) 1177 error = EINTR; 1178 return (error); 1179 } 1180 1181 #if 0 /* /dev/console */ 1182 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1183 #else 1184 if (fp == NULL) 1185 return (error); 1186 #endif 1187 if (fp->f_ops == &badfileops) 1188 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1189 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1190 1191 /* 1192 * Hint to the dofilewrite() to not force the buffer draining 1193 * on the writer to the file. Most likely, the write would 1194 * not need normal buffers. 1195 */ 1196 mtx_lock(mtxp); 1197 fp->f_vnread_flags |= FDEVFS_VNODE; 1198 mtx_unlock(mtxp); 1199 return (error); 1200 } 1201 1202 static int 1203 devfs_pathconf(struct vop_pathconf_args *ap) 1204 { 1205 1206 switch (ap->a_name) { 1207 case _PC_MAC_PRESENT: 1208 #ifdef MAC 1209 /* 1210 * If MAC is enabled, devfs automatically supports 1211 * trivial non-persistant label storage. 1212 */ 1213 *ap->a_retval = 1; 1214 #else 1215 *ap->a_retval = 0; 1216 #endif 1217 return (0); 1218 default: 1219 return (vop_stdpathconf(ap)); 1220 } 1221 /* NOTREACHED */ 1222 } 1223 1224 /* ARGSUSED */ 1225 static int 1226 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1227 { 1228 struct cdev *dev; 1229 struct cdevsw *dsw; 1230 int error, ref; 1231 struct file *fpop; 1232 1233 fpop = td->td_fpop; 1234 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1235 if (error != 0) { 1236 error = vnops.fo_poll(fp, events, cred, td); 1237 return (error); 1238 } 1239 error = dsw->d_poll(dev, events, td); 1240 td->td_fpop = fpop; 1241 dev_relthread(dev, ref); 1242 return(error); 1243 } 1244 1245 /* 1246 * Print out the contents of a special device vnode. 1247 */ 1248 static int 1249 devfs_print(struct vop_print_args *ap) 1250 { 1251 1252 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1253 return (0); 1254 } 1255 1256 static int 1257 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1258 int flags, struct thread *td) 1259 { 1260 struct cdev *dev; 1261 int ioflag, error, ref; 1262 ssize_t resid; 1263 struct cdevsw *dsw; 1264 struct file *fpop; 1265 1266 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1267 return (EINVAL); 1268 fpop = td->td_fpop; 1269 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1270 if (error != 0) { 1271 error = vnops.fo_read(fp, uio, cred, flags, td); 1272 return (error); 1273 } 1274 resid = uio->uio_resid; 1275 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1276 if (ioflag & O_DIRECT) 1277 ioflag |= IO_DIRECT; 1278 1279 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1280 error = dsw->d_read(dev, uio, ioflag); 1281 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1282 devfs_timestamp(&dev->si_atime); 1283 td->td_fpop = fpop; 1284 dev_relthread(dev, ref); 1285 1286 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1287 return (error); 1288 } 1289 1290 static int 1291 devfs_readdir(struct vop_readdir_args *ap) 1292 { 1293 int error; 1294 struct uio *uio; 1295 struct dirent *dp; 1296 struct devfs_dirent *dd; 1297 struct devfs_dirent *de; 1298 struct devfs_mount *dmp; 1299 off_t off; 1300 int *tmp_ncookies = NULL; 1301 1302 if (ap->a_vp->v_type != VDIR) 1303 return (ENOTDIR); 1304 1305 uio = ap->a_uio; 1306 if (uio->uio_offset < 0) 1307 return (EINVAL); 1308 1309 /* 1310 * XXX: This is a temporary hack to get around this filesystem not 1311 * supporting cookies. We store the location of the ncookies pointer 1312 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1313 * and set the number of cookies to 0. We then set the pointer to 1314 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1315 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1316 * pointer to its original location before returning to the caller. 1317 */ 1318 if (ap->a_ncookies != NULL) { 1319 tmp_ncookies = ap->a_ncookies; 1320 *ap->a_ncookies = 0; 1321 ap->a_ncookies = NULL; 1322 } 1323 1324 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1325 if (devfs_populate_vp(ap->a_vp) != 0) { 1326 if (tmp_ncookies != NULL) 1327 ap->a_ncookies = tmp_ncookies; 1328 return (EIO); 1329 } 1330 error = 0; 1331 de = ap->a_vp->v_data; 1332 off = 0; 1333 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1334 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1335 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1336 continue; 1337 if (devfs_prison_check(dd, uio->uio_td)) 1338 continue; 1339 if (dd->de_dirent->d_type == DT_DIR) 1340 de = dd->de_dir; 1341 else 1342 de = dd; 1343 dp = dd->de_dirent; 1344 if (dp->d_reclen > uio->uio_resid) 1345 break; 1346 dp->d_fileno = de->de_inode; 1347 if (off >= uio->uio_offset) { 1348 error = vfs_read_dirent(ap, dp, off); 1349 if (error) 1350 break; 1351 } 1352 off += dp->d_reclen; 1353 } 1354 sx_xunlock(&dmp->dm_lock); 1355 uio->uio_offset = off; 1356 1357 /* 1358 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1359 * place. 1360 */ 1361 if (tmp_ncookies != NULL) 1362 ap->a_ncookies = tmp_ncookies; 1363 1364 return (error); 1365 } 1366 1367 static int 1368 devfs_readlink(struct vop_readlink_args *ap) 1369 { 1370 struct devfs_dirent *de; 1371 1372 de = ap->a_vp->v_data; 1373 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1374 } 1375 1376 static int 1377 devfs_reclaim(struct vop_reclaim_args *ap) 1378 { 1379 struct vnode *vp; 1380 struct devfs_dirent *de; 1381 1382 vp = ap->a_vp; 1383 mtx_lock(&devfs_de_interlock); 1384 de = vp->v_data; 1385 if (de != NULL) { 1386 de->de_vnode = NULL; 1387 vp->v_data = NULL; 1388 } 1389 mtx_unlock(&devfs_de_interlock); 1390 vnode_destroy_vobject(vp); 1391 return (0); 1392 } 1393 1394 static int 1395 devfs_reclaim_vchr(struct vop_reclaim_args *ap) 1396 { 1397 struct vnode *vp; 1398 struct cdev *dev; 1399 1400 vp = ap->a_vp; 1401 MPASS(vp->v_type == VCHR); 1402 1403 devfs_reclaim(ap); 1404 1405 VI_LOCK(vp); 1406 dev_lock(); 1407 dev = vp->v_rdev; 1408 vp->v_rdev = NULL; 1409 if (dev != NULL) 1410 dev->si_usecount -= vp->v_usecount; 1411 dev_unlock(); 1412 VI_UNLOCK(vp); 1413 if (dev != NULL) 1414 dev_rel(dev); 1415 return (0); 1416 } 1417 1418 static int 1419 devfs_remove(struct vop_remove_args *ap) 1420 { 1421 struct vnode *dvp = ap->a_dvp; 1422 struct vnode *vp = ap->a_vp; 1423 struct devfs_dirent *dd; 1424 struct devfs_dirent *de, *de_covered; 1425 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1426 1427 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1428 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1429 1430 sx_xlock(&dmp->dm_lock); 1431 dd = ap->a_dvp->v_data; 1432 de = vp->v_data; 1433 if (de->de_cdp == NULL) { 1434 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1435 if (de->de_dirent->d_type == DT_LNK) { 1436 de_covered = devfs_find(dd, de->de_dirent->d_name, 1437 de->de_dirent->d_namlen, 0); 1438 if (de_covered != NULL) 1439 de_covered->de_flags &= ~DE_COVERED; 1440 } 1441 /* We need to unlock dvp because devfs_delete() may lock it. */ 1442 VOP_UNLOCK(vp, 0); 1443 if (dvp != vp) 1444 VOP_UNLOCK(dvp, 0); 1445 devfs_delete(dmp, de, 0); 1446 sx_xunlock(&dmp->dm_lock); 1447 if (dvp != vp) 1448 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1449 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1450 } else { 1451 de->de_flags |= DE_WHITEOUT; 1452 sx_xunlock(&dmp->dm_lock); 1453 } 1454 return (0); 1455 } 1456 1457 /* 1458 * Revoke is called on a tty when a terminal session ends. The vnode 1459 * is orphaned by setting v_op to deadfs so we need to let go of it 1460 * as well so that we create a new one next time around. 1461 * 1462 */ 1463 static int 1464 devfs_revoke(struct vop_revoke_args *ap) 1465 { 1466 struct vnode *vp = ap->a_vp, *vp2; 1467 struct cdev *dev; 1468 struct cdev_priv *cdp; 1469 struct devfs_dirent *de; 1470 u_int i; 1471 1472 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1473 1474 dev = vp->v_rdev; 1475 cdp = cdev2priv(dev); 1476 1477 dev_lock(); 1478 cdp->cdp_inuse++; 1479 dev_unlock(); 1480 1481 vhold(vp); 1482 vgone(vp); 1483 vdrop(vp); 1484 1485 VOP_UNLOCK(vp,0); 1486 loop: 1487 for (;;) { 1488 mtx_lock(&devfs_de_interlock); 1489 dev_lock(); 1490 vp2 = NULL; 1491 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1492 de = cdp->cdp_dirents[i]; 1493 if (de == NULL) 1494 continue; 1495 1496 vp2 = de->de_vnode; 1497 if (vp2 != NULL) { 1498 dev_unlock(); 1499 VI_LOCK(vp2); 1500 mtx_unlock(&devfs_de_interlock); 1501 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1502 curthread)) 1503 goto loop; 1504 vhold(vp2); 1505 vgone(vp2); 1506 vdrop(vp2); 1507 vput(vp2); 1508 break; 1509 } 1510 } 1511 if (vp2 != NULL) { 1512 continue; 1513 } 1514 dev_unlock(); 1515 mtx_unlock(&devfs_de_interlock); 1516 break; 1517 } 1518 dev_lock(); 1519 cdp->cdp_inuse--; 1520 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1521 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1522 dev_unlock(); 1523 dev_rel(&cdp->cdp_c); 1524 } else 1525 dev_unlock(); 1526 1527 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1528 return (0); 1529 } 1530 1531 static int 1532 devfs_rioctl(struct vop_ioctl_args *ap) 1533 { 1534 struct vnode *vp; 1535 struct devfs_mount *dmp; 1536 int error; 1537 1538 vp = ap->a_vp; 1539 vn_lock(vp, LK_SHARED | LK_RETRY); 1540 if (vp->v_iflag & VI_DOOMED) { 1541 VOP_UNLOCK(vp, 0); 1542 return (EBADF); 1543 } 1544 dmp = VFSTODEVFS(vp->v_mount); 1545 sx_xlock(&dmp->dm_lock); 1546 VOP_UNLOCK(vp, 0); 1547 DEVFS_DMP_HOLD(dmp); 1548 devfs_populate(dmp); 1549 if (DEVFS_DMP_DROP(dmp)) { 1550 sx_xunlock(&dmp->dm_lock); 1551 devfs_unmount_final(dmp); 1552 return (ENOENT); 1553 } 1554 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1555 sx_xunlock(&dmp->dm_lock); 1556 return (error); 1557 } 1558 1559 static int 1560 devfs_rread(struct vop_read_args *ap) 1561 { 1562 1563 if (ap->a_vp->v_type != VDIR) 1564 return (EINVAL); 1565 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1566 } 1567 1568 static int 1569 devfs_setattr(struct vop_setattr_args *ap) 1570 { 1571 struct devfs_dirent *de; 1572 struct vattr *vap; 1573 struct vnode *vp; 1574 struct thread *td; 1575 int c, error; 1576 uid_t uid; 1577 gid_t gid; 1578 1579 vap = ap->a_vap; 1580 vp = ap->a_vp; 1581 td = curthread; 1582 if ((vap->va_type != VNON) || 1583 (vap->va_nlink != VNOVAL) || 1584 (vap->va_fsid != VNOVAL) || 1585 (vap->va_fileid != VNOVAL) || 1586 (vap->va_blocksize != VNOVAL) || 1587 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1588 (vap->va_rdev != VNOVAL) || 1589 ((int)vap->va_bytes != VNOVAL) || 1590 (vap->va_gen != VNOVAL)) { 1591 return (EINVAL); 1592 } 1593 1594 error = devfs_populate_vp(vp); 1595 if (error != 0) 1596 return (error); 1597 1598 de = vp->v_data; 1599 if (vp->v_type == VDIR) 1600 de = de->de_dir; 1601 1602 c = 0; 1603 if (vap->va_uid == (uid_t)VNOVAL) 1604 uid = de->de_uid; 1605 else 1606 uid = vap->va_uid; 1607 if (vap->va_gid == (gid_t)VNOVAL) 1608 gid = de->de_gid; 1609 else 1610 gid = vap->va_gid; 1611 if (uid != de->de_uid || gid != de->de_gid) { 1612 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1613 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1614 error = priv_check(td, PRIV_VFS_CHOWN); 1615 if (error != 0) 1616 goto ret; 1617 } 1618 de->de_uid = uid; 1619 de->de_gid = gid; 1620 c = 1; 1621 } 1622 1623 if (vap->va_mode != (mode_t)VNOVAL) { 1624 if (ap->a_cred->cr_uid != de->de_uid) { 1625 error = priv_check(td, PRIV_VFS_ADMIN); 1626 if (error != 0) 1627 goto ret; 1628 } 1629 de->de_mode = vap->va_mode; 1630 c = 1; 1631 } 1632 1633 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1634 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1635 if (error != 0) 1636 goto ret; 1637 if (vap->va_atime.tv_sec != VNOVAL) { 1638 if (vp->v_type == VCHR) 1639 vp->v_rdev->si_atime = vap->va_atime; 1640 else 1641 de->de_atime = vap->va_atime; 1642 } 1643 if (vap->va_mtime.tv_sec != VNOVAL) { 1644 if (vp->v_type == VCHR) 1645 vp->v_rdev->si_mtime = vap->va_mtime; 1646 else 1647 de->de_mtime = vap->va_mtime; 1648 } 1649 c = 1; 1650 } 1651 1652 if (c) { 1653 if (vp->v_type == VCHR) 1654 vfs_timestamp(&vp->v_rdev->si_ctime); 1655 else 1656 vfs_timestamp(&de->de_mtime); 1657 } 1658 1659 ret: 1660 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1661 return (error); 1662 } 1663 1664 #ifdef MAC 1665 static int 1666 devfs_setlabel(struct vop_setlabel_args *ap) 1667 { 1668 struct vnode *vp; 1669 struct devfs_dirent *de; 1670 1671 vp = ap->a_vp; 1672 de = vp->v_data; 1673 1674 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1675 mac_devfs_update(vp->v_mount, de, vp); 1676 1677 return (0); 1678 } 1679 #endif 1680 1681 static int 1682 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1683 { 1684 1685 return (vnops.fo_stat(fp, sb, cred, td)); 1686 } 1687 1688 static int 1689 devfs_symlink(struct vop_symlink_args *ap) 1690 { 1691 int i, error; 1692 struct devfs_dirent *dd; 1693 struct devfs_dirent *de, *de_covered, *de_dotdot; 1694 struct devfs_mount *dmp; 1695 1696 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1697 if (error) 1698 return(error); 1699 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1700 if (devfs_populate_vp(ap->a_dvp) != 0) 1701 return (ENOENT); 1702 1703 dd = ap->a_dvp->v_data; 1704 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1705 de->de_flags = DE_USER; 1706 de->de_uid = 0; 1707 de->de_gid = 0; 1708 de->de_mode = 0755; 1709 de->de_inode = alloc_unr(devfs_inos); 1710 de->de_dir = dd; 1711 de->de_dirent->d_type = DT_LNK; 1712 i = strlen(ap->a_target) + 1; 1713 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1714 bcopy(ap->a_target, de->de_symlink, i); 1715 #ifdef MAC 1716 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1717 #endif 1718 de_covered = devfs_find(dd, de->de_dirent->d_name, 1719 de->de_dirent->d_namlen, 0); 1720 if (de_covered != NULL) { 1721 if ((de_covered->de_flags & DE_USER) != 0) { 1722 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1723 sx_xunlock(&dmp->dm_lock); 1724 return (EEXIST); 1725 } 1726 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1727 ("devfs_symlink: entry %p already covered", de_covered)); 1728 de_covered->de_flags |= DE_COVERED; 1729 } 1730 1731 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1732 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1733 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1734 devfs_dir_ref_de(dmp, dd); 1735 devfs_rules_apply(dmp, de); 1736 1737 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1738 } 1739 1740 static int 1741 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1742 { 1743 1744 return (vnops.fo_truncate(fp, length, cred, td)); 1745 } 1746 1747 static int 1748 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1749 int flags, struct thread *td) 1750 { 1751 struct cdev *dev; 1752 int error, ioflag, ref; 1753 ssize_t resid; 1754 struct cdevsw *dsw; 1755 struct file *fpop; 1756 1757 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1758 return (EINVAL); 1759 fpop = td->td_fpop; 1760 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1761 if (error != 0) { 1762 error = vnops.fo_write(fp, uio, cred, flags, td); 1763 return (error); 1764 } 1765 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1766 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1767 if (ioflag & O_DIRECT) 1768 ioflag |= IO_DIRECT; 1769 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1770 1771 resid = uio->uio_resid; 1772 1773 error = dsw->d_write(dev, uio, ioflag); 1774 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1775 devfs_timestamp(&dev->si_ctime); 1776 dev->si_mtime = dev->si_ctime; 1777 } 1778 td->td_fpop = fpop; 1779 dev_relthread(dev, ref); 1780 1781 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1782 return (error); 1783 } 1784 1785 static int 1786 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1787 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1788 struct thread *td) 1789 { 1790 struct cdev *dev; 1791 struct cdevsw *dsw; 1792 struct mount *mp; 1793 struct vnode *vp; 1794 struct file *fpop; 1795 vm_object_t object; 1796 vm_prot_t maxprot; 1797 int error, ref; 1798 1799 vp = fp->f_vnode; 1800 1801 /* 1802 * Ensure that file and memory protections are 1803 * compatible. 1804 */ 1805 mp = vp->v_mount; 1806 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) 1807 maxprot = VM_PROT_NONE; 1808 else 1809 maxprot = VM_PROT_EXECUTE; 1810 if ((fp->f_flag & FREAD) != 0) 1811 maxprot |= VM_PROT_READ; 1812 else if ((prot & VM_PROT_READ) != 0) 1813 return (EACCES); 1814 1815 /* 1816 * If we are sharing potential changes via MAP_SHARED and we 1817 * are trying to get write permission although we opened it 1818 * without asking for it, bail out. 1819 * 1820 * Note that most character devices always share mappings. 1821 * The one exception is that D_MMAP_ANON devices 1822 * (i.e. /dev/zero) permit private writable mappings. 1823 * 1824 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests 1825 * as well as updating maxprot to permit writing for 1826 * D_MMAP_ANON devices rather than doing that here. 1827 */ 1828 if ((flags & MAP_SHARED) != 0) { 1829 if ((fp->f_flag & FWRITE) != 0) 1830 maxprot |= VM_PROT_WRITE; 1831 else if ((prot & VM_PROT_WRITE) != 0) 1832 return (EACCES); 1833 } 1834 maxprot &= cap_maxprot; 1835 1836 fpop = td->td_fpop; 1837 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1838 if (error != 0) 1839 return (error); 1840 1841 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff, 1842 &object); 1843 td->td_fpop = fpop; 1844 dev_relthread(dev, ref); 1845 if (error != 0) 1846 return (error); 1847 1848 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1849 foff, FALSE, td); 1850 if (error != 0) 1851 vm_object_deallocate(object); 1852 return (error); 1853 } 1854 1855 dev_t 1856 dev2udev(struct cdev *x) 1857 { 1858 if (x == NULL) 1859 return (NODEV); 1860 return (cdev2priv(x)->cdp_inode); 1861 } 1862 1863 static struct fileops devfs_ops_f = { 1864 .fo_read = devfs_read_f, 1865 .fo_write = devfs_write_f, 1866 .fo_truncate = devfs_truncate_f, 1867 .fo_ioctl = devfs_ioctl_f, 1868 .fo_poll = devfs_poll_f, 1869 .fo_kqfilter = devfs_kqfilter_f, 1870 .fo_stat = devfs_stat_f, 1871 .fo_close = devfs_close_f, 1872 .fo_chmod = vn_chmod, 1873 .fo_chown = vn_chown, 1874 .fo_sendfile = vn_sendfile, 1875 .fo_seek = vn_seek, 1876 .fo_fill_kinfo = vn_fill_kinfo, 1877 .fo_mmap = devfs_mmap_f, 1878 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1879 }; 1880 1881 /* Vops for non-CHR vnodes in /dev. */ 1882 static struct vop_vector devfs_vnodeops = { 1883 .vop_default = &default_vnodeops, 1884 1885 .vop_access = devfs_access, 1886 .vop_getattr = devfs_getattr, 1887 .vop_ioctl = devfs_rioctl, 1888 .vop_lookup = devfs_lookup, 1889 .vop_mknod = devfs_mknod, 1890 .vop_pathconf = devfs_pathconf, 1891 .vop_read = devfs_rread, 1892 .vop_readdir = devfs_readdir, 1893 .vop_readlink = devfs_readlink, 1894 .vop_reclaim = devfs_reclaim, 1895 .vop_remove = devfs_remove, 1896 .vop_revoke = devfs_revoke, 1897 .vop_setattr = devfs_setattr, 1898 #ifdef MAC 1899 .vop_setlabel = devfs_setlabel, 1900 #endif 1901 .vop_symlink = devfs_symlink, 1902 .vop_vptocnp = devfs_vptocnp, 1903 }; 1904 1905 /* Vops for VCHR vnodes in /dev. */ 1906 static struct vop_vector devfs_specops = { 1907 .vop_default = &default_vnodeops, 1908 1909 .vop_access = devfs_access, 1910 .vop_bmap = VOP_PANIC, 1911 .vop_close = devfs_close, 1912 .vop_create = VOP_PANIC, 1913 .vop_fsync = devfs_fsync, 1914 .vop_getattr = devfs_getattr, 1915 .vop_ioctl = devfs_ioctl, 1916 .vop_link = VOP_PANIC, 1917 .vop_mkdir = VOP_PANIC, 1918 .vop_mknod = VOP_PANIC, 1919 .vop_open = devfs_open, 1920 .vop_pathconf = devfs_pathconf, 1921 .vop_poll = dead_poll, 1922 .vop_print = devfs_print, 1923 .vop_read = dead_read, 1924 .vop_readdir = VOP_PANIC, 1925 .vop_readlink = VOP_PANIC, 1926 .vop_reallocblks = VOP_PANIC, 1927 .vop_reclaim = devfs_reclaim_vchr, 1928 .vop_remove = devfs_remove, 1929 .vop_rename = VOP_PANIC, 1930 .vop_revoke = devfs_revoke, 1931 .vop_rmdir = VOP_PANIC, 1932 .vop_setattr = devfs_setattr, 1933 #ifdef MAC 1934 .vop_setlabel = devfs_setlabel, 1935 #endif 1936 .vop_strategy = VOP_PANIC, 1937 .vop_symlink = VOP_PANIC, 1938 .vop_vptocnp = devfs_vptocnp, 1939 .vop_write = dead_write, 1940 }; 1941 1942 /* 1943 * Our calling convention to the device drivers used to be that we passed 1944 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1945 * flags instead since that's what open(), close() and ioctl() takes and 1946 * we don't really want vnode.h in device drivers. 1947 * We solved the source compatibility by redefining some vnode flags to 1948 * be the same as the fcntl ones and by sending down the bitwise OR of 1949 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1950 * pulls the rug out under this. 1951 */ 1952 CTASSERT(O_NONBLOCK == IO_NDELAY); 1953 CTASSERT(O_FSYNC == IO_SYNC); 1954