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