Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/fs/devfs/devfs_vnops.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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/limits.h> 55 #include <sys/lock.h> 56 #include <sys/malloc.h> 57 #include <sys/mman.h> 58 #include <sys/mount.h> 59 #include <sys/namei.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/stat.h> 63 #include <sys/sx.h> 64 #include <sys/sysctl.h> 65 #include <sys/time.h> 66 #include <sys/ttycom.h> 67 #include <sys/unistd.h> 68 #include <sys/vnode.h> 69 70 static struct vop_vector devfs_vnodeops; 71 static struct vop_vector devfs_specops; 72 static struct fileops devfs_ops_f; 73 74 #include <fs/devfs/devfs.h> 75 #include <fs/devfs/devfs_int.h> 76 77 #include <security/mac/mac_framework.h> 78 79 #include <vm/vm.h> 80 #include <vm/vm_extern.h> 81 #include <vm/vm_object.h> 82 83 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data"); 84 85 struct mtx devfs_de_interlock; 86 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); 87 struct sx clone_drain_lock; 88 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock"); 89 struct mtx cdevpriv_mtx; 90 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF); 91 92 SYSCTL_DECL(_vfs_devfs); 93 94 static int devfs_dotimes; 95 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW, 96 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision"); 97 98 /* 99 * Update devfs node timestamp. Note that updates are unlocked and 100 * stat(2) could see partially updated times. 101 */ 102 static void 103 devfs_timestamp(struct timespec *tsp) 104 { 105 time_t ts; 106 107 if (devfs_dotimes) { 108 vfs_timestamp(tsp); 109 } else { 110 ts = time_second; 111 if (tsp->tv_sec != ts) { 112 tsp->tv_sec = ts; 113 tsp->tv_nsec = 0; 114 } 115 } 116 } 117 118 static int 119 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp, 120 int *ref) 121 { 122 123 *dswp = devvn_refthread(fp->f_vnode, devp, ref); 124 if (*devp != fp->f_data) { 125 if (*dswp != NULL) 126 dev_relthread(*devp, *ref); 127 return (ENXIO); 128 } 129 KASSERT((*devp)->si_refcount > 0, 130 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); 131 if (*dswp == NULL) 132 return (ENXIO); 133 curthread->td_fpop = fp; 134 return (0); 135 } 136 137 int 138 devfs_get_cdevpriv(void **datap) 139 { 140 struct file *fp; 141 struct cdev_privdata *p; 142 int error; 143 144 fp = curthread->td_fpop; 145 if (fp == NULL) 146 return (EBADF); 147 p = fp->f_cdevpriv; 148 if (p != NULL) { 149 error = 0; 150 *datap = p->cdpd_data; 151 } else 152 error = ENOENT; 153 return (error); 154 } 155 156 int 157 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr) 158 { 159 struct file *fp; 160 struct cdev_priv *cdp; 161 struct cdev_privdata *p; 162 int error; 163 164 fp = curthread->td_fpop; 165 if (fp == NULL) 166 return (ENOENT); 167 cdp = cdev2priv((struct cdev *)fp->f_data); 168 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK); 169 p->cdpd_data = priv; 170 p->cdpd_dtr = priv_dtr; 171 p->cdpd_fp = fp; 172 mtx_lock(&cdevpriv_mtx); 173 if (fp->f_cdevpriv == NULL) { 174 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list); 175 fp->f_cdevpriv = p; 176 mtx_unlock(&cdevpriv_mtx); 177 error = 0; 178 } else { 179 mtx_unlock(&cdevpriv_mtx); 180 free(p, M_CDEVPDATA); 181 error = EBUSY; 182 } 183 return (error); 184 } 185 186 void 187 devfs_destroy_cdevpriv(struct cdev_privdata *p) 188 { 189 190 mtx_assert(&cdevpriv_mtx, MA_OWNED); 191 KASSERT(p->cdpd_fp->f_cdevpriv == p, 192 ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p)); 193 p->cdpd_fp->f_cdevpriv = NULL; 194 LIST_REMOVE(p, cdpd_list); 195 mtx_unlock(&cdevpriv_mtx); 196 (p->cdpd_dtr)(p->cdpd_data); 197 free(p, M_CDEVPDATA); 198 } 199 200 static void 201 devfs_fpdrop(struct file *fp) 202 { 203 struct cdev_privdata *p; 204 205 mtx_lock(&cdevpriv_mtx); 206 if ((p = fp->f_cdevpriv) == NULL) { 207 mtx_unlock(&cdevpriv_mtx); 208 return; 209 } 210 devfs_destroy_cdevpriv(p); 211 } 212 213 void 214 devfs_clear_cdevpriv(void) 215 { 216 struct file *fp; 217 218 fp = curthread->td_fpop; 219 if (fp == NULL) 220 return; 221 devfs_fpdrop(fp); 222 } 223 224 /* 225 * On success devfs_populate_vp() returns with dmp->dm_lock held. 226 */ 227 static int 228 devfs_populate_vp(struct vnode *vp) 229 { 230 struct devfs_dirent *de; 231 struct devfs_mount *dmp; 232 int locked; 233 234 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp"); 235 236 dmp = VFSTODEVFS(vp->v_mount); 237 locked = VOP_ISLOCKED(vp); 238 239 sx_xlock(&dmp->dm_lock); 240 DEVFS_DMP_HOLD(dmp); 241 242 /* Can't call devfs_populate() with the vnode lock held. */ 243 VOP_UNLOCK(vp, 0); 244 devfs_populate(dmp); 245 246 sx_xunlock(&dmp->dm_lock); 247 vn_lock(vp, locked | LK_RETRY); 248 sx_xlock(&dmp->dm_lock); 249 if (DEVFS_DMP_DROP(dmp)) { 250 sx_xunlock(&dmp->dm_lock); 251 devfs_unmount_final(dmp); 252 return (ERESTART); 253 } 254 if ((vp->v_iflag & VI_DOOMED) != 0) { 255 sx_xunlock(&dmp->dm_lock); 256 return (ERESTART); 257 } 258 de = vp->v_data; 259 KASSERT(de != NULL, 260 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed")); 261 if ((de->de_flags & DE_DOOMED) != 0) { 262 sx_xunlock(&dmp->dm_lock); 263 return (ERESTART); 264 } 265 266 return (0); 267 } 268 269 static int 270 devfs_vptocnp(struct vop_vptocnp_args *ap) 271 { 272 struct vnode *vp = ap->a_vp; 273 struct vnode **dvp = ap->a_vpp; 274 struct devfs_mount *dmp; 275 char *buf = ap->a_buf; 276 int *buflen = ap->a_buflen; 277 struct devfs_dirent *dd, *de; 278 int i, error; 279 280 dmp = VFSTODEVFS(vp->v_mount); 281 282 error = devfs_populate_vp(vp); 283 if (error != 0) 284 return (error); 285 286 if (vp->v_type != VCHR && vp->v_type != VDIR) { 287 error = ENOENT; 288 goto finished; 289 } 290 291 dd = vp->v_data; 292 if (vp->v_type == VDIR && dd == dmp->dm_rootdir) { 293 *dvp = vp; 294 vref(*dvp); 295 goto finished; 296 } 297 298 i = *buflen; 299 i -= dd->de_dirent->d_namlen; 300 if (i < 0) { 301 error = ENOMEM; 302 goto finished; 303 } 304 bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen); 305 *buflen = i; 306 de = devfs_parent_dirent(dd); 307 if (de == NULL) { 308 error = ENOENT; 309 goto finished; 310 } 311 mtx_lock(&devfs_de_interlock); 312 *dvp = de->de_vnode; 313 if (*dvp != NULL) { 314 VI_LOCK(*dvp); 315 mtx_unlock(&devfs_de_interlock); 316 vholdl(*dvp); 317 VI_UNLOCK(*dvp); 318 vref(*dvp); 319 vdrop(*dvp); 320 } else { 321 mtx_unlock(&devfs_de_interlock); 322 error = ENOENT; 323 } 324 finished: 325 sx_xunlock(&dmp->dm_lock); 326 return (error); 327 } 328 329 /* 330 * Construct the fully qualified path name relative to the mountpoint. 331 * If a NULL cnp is provided, no '/' is appended to the resulting path. 332 */ 333 char * 334 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd, 335 struct componentname *cnp) 336 { 337 int i; 338 struct devfs_dirent *de; 339 340 sx_assert(&dmp->dm_lock, SA_LOCKED); 341 342 i = SPECNAMELEN; 343 buf[i] = '\0'; 344 if (cnp != NULL) 345 i -= cnp->cn_namelen; 346 if (i < 0) 347 return (NULL); 348 if (cnp != NULL) 349 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); 350 de = dd; 351 while (de != dmp->dm_rootdir) { 352 if (cnp != NULL || i < SPECNAMELEN) { 353 i--; 354 if (i < 0) 355 return (NULL); 356 buf[i] = '/'; 357 } 358 i -= de->de_dirent->d_namlen; 359 if (i < 0) 360 return (NULL); 361 bcopy(de->de_dirent->d_name, buf + i, 362 de->de_dirent->d_namlen); 363 de = devfs_parent_dirent(de); 364 if (de == NULL) 365 return (NULL); 366 } 367 return (buf + i); 368 } 369 370 static int 371 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp, 372 struct devfs_dirent *de) 373 { 374 int not_found; 375 376 not_found = 0; 377 if (de->de_flags & DE_DOOMED) 378 not_found = 1; 379 if (DEVFS_DE_DROP(de)) { 380 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed")); 381 devfs_dirent_free(de); 382 } 383 if (DEVFS_DMP_DROP(dmp)) { 384 KASSERT(not_found == 1, 385 ("DEVFS mount struct freed before dirent")); 386 not_found = 2; 387 sx_xunlock(&dmp->dm_lock); 388 devfs_unmount_final(dmp); 389 } 390 if (not_found == 1 || (drop_dm_lock && not_found != 2)) 391 sx_unlock(&dmp->dm_lock); 392 return (not_found); 393 } 394 395 static void 396 devfs_insmntque_dtr(struct vnode *vp, void *arg) 397 { 398 struct devfs_dirent *de; 399 400 de = (struct devfs_dirent *)arg; 401 mtx_lock(&devfs_de_interlock); 402 vp->v_data = NULL; 403 de->de_vnode = NULL; 404 mtx_unlock(&devfs_de_interlock); 405 vgone(vp); 406 vput(vp); 407 } 408 409 /* 410 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops 411 * it on return. 412 */ 413 int 414 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode, 415 struct vnode **vpp) 416 { 417 int error; 418 struct vnode *vp; 419 struct cdev *dev; 420 struct devfs_mount *dmp; 421 struct cdevsw *dsw; 422 423 dmp = VFSTODEVFS(mp); 424 if (de->de_flags & DE_DOOMED) { 425 sx_xunlock(&dmp->dm_lock); 426 return (ENOENT); 427 } 428 loop: 429 DEVFS_DE_HOLD(de); 430 DEVFS_DMP_HOLD(dmp); 431 mtx_lock(&devfs_de_interlock); 432 vp = de->de_vnode; 433 if (vp != NULL) { 434 VI_LOCK(vp); 435 mtx_unlock(&devfs_de_interlock); 436 sx_xunlock(&dmp->dm_lock); 437 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread); 438 sx_xlock(&dmp->dm_lock); 439 if (devfs_allocv_drop_refs(0, dmp, de)) { 440 vput(vp); 441 return (ENOENT); 442 } 443 else if ((vp->v_iflag & VI_DOOMED) != 0) { 444 mtx_lock(&devfs_de_interlock); 445 if (de->de_vnode == vp) { 446 de->de_vnode = NULL; 447 vp->v_data = NULL; 448 } 449 mtx_unlock(&devfs_de_interlock); 450 vput(vp); 451 goto loop; 452 } 453 sx_xunlock(&dmp->dm_lock); 454 *vpp = vp; 455 return (0); 456 } 457 mtx_unlock(&devfs_de_interlock); 458 if (de->de_dirent->d_type == DT_CHR) { 459 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) { 460 devfs_allocv_drop_refs(1, dmp, de); 461 return (ENOENT); 462 } 463 dev = &de->de_cdp->cdp_c; 464 } else { 465 dev = NULL; 466 } 467 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); 468 if (error != 0) { 469 devfs_allocv_drop_refs(1, dmp, de); 470 printf("devfs_allocv: failed to allocate new vnode\n"); 471 return (error); 472 } 473 474 if (de->de_dirent->d_type == DT_CHR) { 475 vp->v_type = VCHR; 476 VI_LOCK(vp); 477 dev_lock(); 478 dev_refl(dev); 479 /* XXX: v_rdev should be protect by vnode lock */ 480 vp->v_rdev = dev; 481 KASSERT(vp->v_usecount == 1, 482 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount)); 483 dev->si_usecount += vp->v_usecount; 484 /* Special casing of ttys for deadfs. Probably redundant. */ 485 dsw = dev->si_devsw; 486 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) 487 vp->v_vflag |= VV_ISTTY; 488 dev_unlock(); 489 VI_UNLOCK(vp); 490 if ((dev->si_flags & SI_ETERNAL) != 0) 491 vp->v_vflag |= VV_ETERNALDEV; 492 vp->v_op = &devfs_specops; 493 } else if (de->de_dirent->d_type == DT_DIR) { 494 vp->v_type = VDIR; 495 } else if (de->de_dirent->d_type == DT_LNK) { 496 vp->v_type = VLNK; 497 } else { 498 vp->v_type = VBAD; 499 } 500 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); 501 VN_LOCK_ASHARE(vp); 502 mtx_lock(&devfs_de_interlock); 503 vp->v_data = de; 504 de->de_vnode = vp; 505 mtx_unlock(&devfs_de_interlock); 506 error = insmntque1(vp, mp, devfs_insmntque_dtr, de); 507 if (error != 0) { 508 (void) devfs_allocv_drop_refs(1, dmp, de); 509 return (error); 510 } 511 if (devfs_allocv_drop_refs(0, dmp, de)) { 512 vput(vp); 513 return (ENOENT); 514 } 515 #ifdef MAC 516 mac_devfs_vnode_associate(mp, de, vp); 517 #endif 518 sx_xunlock(&dmp->dm_lock); 519 *vpp = vp; 520 return (0); 521 } 522 523 static int 524 devfs_access(struct vop_access_args *ap) 525 { 526 struct vnode *vp = ap->a_vp; 527 struct devfs_dirent *de; 528 struct proc *p; 529 int error; 530 531 de = vp->v_data; 532 if (vp->v_type == VDIR) 533 de = de->de_dir; 534 535 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, 536 ap->a_accmode, ap->a_cred, NULL); 537 if (error == 0) 538 return (0); 539 if (error != EACCES) 540 return (error); 541 p = ap->a_td->td_proc; 542 /* We do, however, allow access to the controlling terminal */ 543 PROC_LOCK(p); 544 if (!(p->p_flag & P_CONTROLT)) { 545 PROC_UNLOCK(p); 546 return (error); 547 } 548 if (p->p_session->s_ttydp == de->de_cdp) 549 error = 0; 550 PROC_UNLOCK(p); 551 return (error); 552 } 553 554 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0, 555 "devfs-only flag reuse failed"); 556 557 static int 558 devfs_close(struct vop_close_args *ap) 559 { 560 struct vnode *vp = ap->a_vp, *oldvp; 561 struct thread *td = ap->a_td; 562 struct proc *p; 563 struct cdev *dev = vp->v_rdev; 564 struct cdevsw *dsw; 565 int dflags, error, ref, vp_locked; 566 567 /* 568 * XXX: Don't call d_close() if we were called because of 569 * XXX: insmntque1() failure. 570 */ 571 if (vp->v_data == NULL) 572 return (0); 573 574 /* 575 * Hack: a tty device that is a controlling terminal 576 * has a reference from the session structure. 577 * We cannot easily tell that a character device is 578 * a controlling terminal, unless it is the closing 579 * process' controlling terminal. In that case, 580 * if the reference count is 2 (this last descriptor 581 * plus the session), release the reference from the session. 582 */ 583 if (td != NULL) { 584 p = td->td_proc; 585 PROC_LOCK(p); 586 if (vp == p->p_session->s_ttyvp) { 587 PROC_UNLOCK(p); 588 oldvp = NULL; 589 sx_xlock(&proctree_lock); 590 if (vp == p->p_session->s_ttyvp) { 591 SESS_LOCK(p->p_session); 592 VI_LOCK(vp); 593 if (count_dev(dev) == 2 && 594 (vp->v_iflag & VI_DOOMED) == 0) { 595 p->p_session->s_ttyvp = NULL; 596 p->p_session->s_ttydp = NULL; 597 oldvp = vp; 598 } 599 VI_UNLOCK(vp); 600 SESS_UNLOCK(p->p_session); 601 } 602 sx_xunlock(&proctree_lock); 603 if (oldvp != NULL) 604 vrele(oldvp); 605 } else 606 PROC_UNLOCK(p); 607 } 608 /* 609 * We do not want to really close the device if it 610 * is still in use unless we are trying to close it 611 * forcibly. Since every use (buffer, vnode, swap, cmap) 612 * holds a reference to the vnode, and because we mark 613 * any other vnodes that alias this device, when the 614 * sum of the reference counts on all the aliased 615 * vnodes descends to one, we are on last close. 616 */ 617 dsw = dev_refthread(dev, &ref); 618 if (dsw == NULL) 619 return (ENXIO); 620 dflags = 0; 621 VI_LOCK(vp); 622 if (vp->v_iflag & VI_DOOMED) { 623 /* Forced close. */ 624 dflags |= FREVOKE | FNONBLOCK; 625 } else if (dsw->d_flags & D_TRACKCLOSE) { 626 /* Keep device updated on status. */ 627 } else if (count_dev(dev) > 1) { 628 VI_UNLOCK(vp); 629 dev_relthread(dev, ref); 630 return (0); 631 } 632 if (count_dev(dev) == 1) 633 dflags |= FLASTCLOSE; 634 vholdl(vp); 635 VI_UNLOCK(vp); 636 vp_locked = VOP_ISLOCKED(vp); 637 VOP_UNLOCK(vp, 0); 638 KASSERT(dev->si_refcount > 0, 639 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 640 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td); 641 dev_relthread(dev, ref); 642 vn_lock(vp, vp_locked | LK_RETRY); 643 vdrop(vp); 644 return (error); 645 } 646 647 static int 648 devfs_close_f(struct file *fp, struct thread *td) 649 { 650 int error; 651 struct file *fpop; 652 653 /* 654 * NB: td may be NULL if this descriptor is closed due to 655 * garbage collection from a closed UNIX domain socket. 656 */ 657 fpop = curthread->td_fpop; 658 curthread->td_fpop = fp; 659 error = vnops.fo_close(fp, td); 660 curthread->td_fpop = fpop; 661 662 /* 663 * The f_cdevpriv cannot be assigned non-NULL value while we 664 * are destroying the file. 665 */ 666 if (fp->f_cdevpriv != NULL) 667 devfs_fpdrop(fp); 668 return (error); 669 } 670 671 static int 672 devfs_getattr(struct vop_getattr_args *ap) 673 { 674 struct vnode *vp = ap->a_vp; 675 struct vattr *vap = ap->a_vap; 676 struct devfs_dirent *de; 677 struct devfs_mount *dmp; 678 struct cdev *dev; 679 struct timeval boottime; 680 int error; 681 682 error = devfs_populate_vp(vp); 683 if (error != 0) 684 return (error); 685 686 dmp = VFSTODEVFS(vp->v_mount); 687 sx_xunlock(&dmp->dm_lock); 688 689 de = vp->v_data; 690 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 691 if (vp->v_type == VDIR) { 692 de = de->de_dir; 693 KASSERT(de != NULL, 694 ("Null dir dirent in devfs_getattr vp=%p", vp)); 695 } 696 vap->va_uid = de->de_uid; 697 vap->va_gid = de->de_gid; 698 vap->va_mode = de->de_mode; 699 if (vp->v_type == VLNK) 700 vap->va_size = strlen(de->de_symlink); 701 else if (vp->v_type == VDIR) 702 vap->va_size = vap->va_bytes = DEV_BSIZE; 703 else 704 vap->va_size = 0; 705 if (vp->v_type != VDIR) 706 vap->va_bytes = 0; 707 vap->va_blocksize = DEV_BSIZE; 708 vap->va_type = vp->v_type; 709 710 getboottime(&boottime); 711 #define fix(aa) \ 712 do { \ 713 if ((aa).tv_sec <= 3600) { \ 714 (aa).tv_sec = boottime.tv_sec; \ 715 (aa).tv_nsec = boottime.tv_usec * 1000; \ 716 } \ 717 } while (0) 718 719 if (vp->v_type != VCHR) { 720 fix(de->de_atime); 721 vap->va_atime = de->de_atime; 722 fix(de->de_mtime); 723 vap->va_mtime = de->de_mtime; 724 fix(de->de_ctime); 725 vap->va_ctime = de->de_ctime; 726 } else { 727 dev = vp->v_rdev; 728 fix(dev->si_atime); 729 vap->va_atime = dev->si_atime; 730 fix(dev->si_mtime); 731 vap->va_mtime = dev->si_mtime; 732 fix(dev->si_ctime); 733 vap->va_ctime = dev->si_ctime; 734 735 vap->va_rdev = cdev2priv(dev)->cdp_inode; 736 } 737 vap->va_gen = 0; 738 vap->va_flags = 0; 739 vap->va_filerev = 0; 740 vap->va_nlink = de->de_links; 741 vap->va_fileid = de->de_inode; 742 743 return (error); 744 } 745 746 /* ARGSUSED */ 747 static int 748 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 749 { 750 struct file *fpop; 751 int error; 752 753 fpop = td->td_fpop; 754 td->td_fpop = fp; 755 error = vnops.fo_ioctl(fp, com, data, cred, td); 756 td->td_fpop = fpop; 757 return (error); 758 } 759 760 static int 761 devfs_ioctl(struct vop_ioctl_args *ap) 762 { 763 struct fiodgname_arg *fgn; 764 struct vnode *vpold, *vp; 765 struct cdevsw *dsw; 766 struct thread *td; 767 struct cdev *dev; 768 int error, ref, i; 769 const char *p; 770 u_long com; 771 772 vp = ap->a_vp; 773 com = ap->a_command; 774 td = ap->a_td; 775 776 dsw = devvn_refthread(vp, &dev, &ref); 777 if (dsw == NULL) 778 return (ENXIO); 779 KASSERT(dev->si_refcount > 0, 780 ("devfs: un-referenced struct cdev *(%s)", devtoname(dev))); 781 782 if (com == FIODTYPE) { 783 *(int *)ap->a_data = dsw->d_flags & D_TYPEMASK; 784 error = 0; 785 goto out; 786 } else if (com == FIODGNAME) { 787 fgn = ap->a_data; 788 p = devtoname(dev); 789 i = strlen(p) + 1; 790 if (i > fgn->len) 791 error = EINVAL; 792 else 793 error = copyout(p, fgn->buf, i); 794 goto out; 795 } 796 797 error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td); 798 799 out: 800 dev_relthread(dev, ref); 801 if (error == ENOIOCTL) 802 error = ENOTTY; 803 804 if (error == 0 && com == TIOCSCTTY) { 805 /* 806 * Do nothing if reassigning same control tty, or if the 807 * control tty has already disappeared. If it disappeared, 808 * it's because we were racing with TIOCNOTTY. TIOCNOTTY 809 * already took care of releasing the old vnode and we have 810 * nothing left to do. 811 */ 812 sx_slock(&proctree_lock); 813 if (td->td_proc->p_session->s_ttyvp == vp || 814 td->td_proc->p_session->s_ttyp == NULL) { 815 sx_sunlock(&proctree_lock); 816 return (0); 817 } 818 819 vpold = td->td_proc->p_session->s_ttyvp; 820 VREF(vp); 821 SESS_LOCK(td->td_proc->p_session); 822 td->td_proc->p_session->s_ttyvp = vp; 823 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 824 SESS_UNLOCK(td->td_proc->p_session); 825 826 sx_sunlock(&proctree_lock); 827 828 /* Get rid of reference to old control tty */ 829 if (vpold) 830 vrele(vpold); 831 } 832 return (error); 833 } 834 835 /* ARGSUSED */ 836 static int 837 devfs_kqfilter_f(struct file *fp, struct knote *kn) 838 { 839 struct cdev *dev; 840 struct cdevsw *dsw; 841 int error, ref; 842 struct file *fpop; 843 struct thread *td; 844 845 td = curthread; 846 fpop = td->td_fpop; 847 error = devfs_fp_check(fp, &dev, &dsw, &ref); 848 if (error) 849 return (error); 850 error = dsw->d_kqfilter(dev, kn); 851 td->td_fpop = fpop; 852 dev_relthread(dev, ref); 853 return (error); 854 } 855 856 static inline int 857 devfs_prison_check(struct devfs_dirent *de, struct thread *td) 858 { 859 struct cdev_priv *cdp; 860 struct ucred *dcr; 861 struct proc *p; 862 int error; 863 864 cdp = de->de_cdp; 865 if (cdp == NULL) 866 return (0); 867 dcr = cdp->cdp_c.si_cred; 868 if (dcr == NULL) 869 return (0); 870 871 error = prison_check(td->td_ucred, dcr); 872 if (error == 0) 873 return (0); 874 /* We do, however, allow access to the controlling terminal */ 875 p = td->td_proc; 876 PROC_LOCK(p); 877 if (!(p->p_flag & P_CONTROLT)) { 878 PROC_UNLOCK(p); 879 return (error); 880 } 881 if (p->p_session->s_ttydp == cdp) 882 error = 0; 883 PROC_UNLOCK(p); 884 return (error); 885 } 886 887 static int 888 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 889 { 890 struct componentname *cnp; 891 struct vnode *dvp, **vpp; 892 struct thread *td; 893 struct devfs_dirent *de, *dd; 894 struct devfs_dirent **dde; 895 struct devfs_mount *dmp; 896 struct mount *mp; 897 struct cdev *cdev; 898 int error, flags, nameiop, dvplocked; 899 char specname[SPECNAMELEN + 1], *pname; 900 901 cnp = ap->a_cnp; 902 vpp = ap->a_vpp; 903 dvp = ap->a_dvp; 904 pname = cnp->cn_nameptr; 905 td = cnp->cn_thread; 906 flags = cnp->cn_flags; 907 nameiop = cnp->cn_nameiop; 908 mp = dvp->v_mount; 909 dmp = VFSTODEVFS(mp); 910 dd = dvp->v_data; 911 *vpp = NULLVP; 912 913 if ((flags & ISLASTCN) && nameiop == RENAME) 914 return (EOPNOTSUPP); 915 916 if (dvp->v_type != VDIR) 917 return (ENOTDIR); 918 919 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 920 return (EIO); 921 922 error = vn_dir_check_exec(dvp, cnp); 923 if (error != 0) 924 return (error); 925 926 if (cnp->cn_namelen == 1 && *pname == '.') { 927 if ((flags & ISLASTCN) && nameiop != LOOKUP) 928 return (EINVAL); 929 *vpp = dvp; 930 VREF(dvp); 931 return (0); 932 } 933 934 if (flags & ISDOTDOT) { 935 if ((flags & ISLASTCN) && nameiop != LOOKUP) 936 return (EINVAL); 937 de = devfs_parent_dirent(dd); 938 if (de == NULL) 939 return (ENOENT); 940 dvplocked = VOP_ISLOCKED(dvp); 941 VOP_UNLOCK(dvp, 0); 942 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, 943 vpp); 944 *dm_unlock = 0; 945 vn_lock(dvp, dvplocked | LK_RETRY); 946 return (error); 947 } 948 949 dd = dvp->v_data; 950 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 951 while (de == NULL) { /* While(...) so we can use break */ 952 953 if (nameiop == DELETE) 954 return (ENOENT); 955 956 /* 957 * OK, we didn't have an entry for the name we were asked for 958 * so we try to see if anybody can create it on demand. 959 */ 960 pname = devfs_fqpn(specname, dmp, dd, cnp); 961 if (pname == NULL) 962 break; 963 964 cdev = NULL; 965 DEVFS_DMP_HOLD(dmp); 966 sx_xunlock(&dmp->dm_lock); 967 sx_slock(&clone_drain_lock); 968 EVENTHANDLER_INVOKE(dev_clone, 969 td->td_ucred, pname, strlen(pname), &cdev); 970 sx_sunlock(&clone_drain_lock); 971 972 if (cdev == NULL) 973 sx_xlock(&dmp->dm_lock); 974 else if (devfs_populate_vp(dvp) != 0) { 975 *dm_unlock = 0; 976 sx_xlock(&dmp->dm_lock); 977 if (DEVFS_DMP_DROP(dmp)) { 978 sx_xunlock(&dmp->dm_lock); 979 devfs_unmount_final(dmp); 980 } else 981 sx_xunlock(&dmp->dm_lock); 982 dev_rel(cdev); 983 return (ENOENT); 984 } 985 if (DEVFS_DMP_DROP(dmp)) { 986 *dm_unlock = 0; 987 sx_xunlock(&dmp->dm_lock); 988 devfs_unmount_final(dmp); 989 if (cdev != NULL) 990 dev_rel(cdev); 991 return (ENOENT); 992 } 993 994 if (cdev == NULL) 995 break; 996 997 dev_lock(); 998 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 999 if (dde != NULL && *dde != NULL) 1000 de = *dde; 1001 dev_unlock(); 1002 dev_rel(cdev); 1003 break; 1004 } 1005 1006 if (de == NULL || de->de_flags & DE_WHITEOUT) { 1007 if ((nameiop == CREATE || nameiop == RENAME) && 1008 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 1009 cnp->cn_flags |= SAVENAME; 1010 return (EJUSTRETURN); 1011 } 1012 return (ENOENT); 1013 } 1014 1015 if (devfs_prison_check(de, td)) 1016 return (ENOENT); 1017 1018 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 1019 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 1020 if (error) 1021 return (error); 1022 if (*vpp == dvp) { 1023 VREF(dvp); 1024 *vpp = dvp; 1025 return (0); 1026 } 1027 } 1028 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp); 1029 *dm_unlock = 0; 1030 return (error); 1031 } 1032 1033 static int 1034 devfs_lookup(struct vop_lookup_args *ap) 1035 { 1036 int j; 1037 struct devfs_mount *dmp; 1038 int dm_unlock; 1039 1040 if (devfs_populate_vp(ap->a_dvp) != 0) 1041 return (ENOTDIR); 1042 1043 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1044 dm_unlock = 1; 1045 j = devfs_lookupx(ap, &dm_unlock); 1046 if (dm_unlock == 1) 1047 sx_xunlock(&dmp->dm_lock); 1048 return (j); 1049 } 1050 1051 static int 1052 devfs_mknod(struct vop_mknod_args *ap) 1053 { 1054 struct componentname *cnp; 1055 struct vnode *dvp, **vpp; 1056 struct devfs_dirent *dd, *de; 1057 struct devfs_mount *dmp; 1058 int error; 1059 1060 /* 1061 * The only type of node we should be creating here is a 1062 * character device, for anything else return EOPNOTSUPP. 1063 */ 1064 if (ap->a_vap->va_type != VCHR) 1065 return (EOPNOTSUPP); 1066 dvp = ap->a_dvp; 1067 dmp = VFSTODEVFS(dvp->v_mount); 1068 1069 cnp = ap->a_cnp; 1070 vpp = ap->a_vpp; 1071 dd = dvp->v_data; 1072 1073 error = ENOENT; 1074 sx_xlock(&dmp->dm_lock); 1075 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1076 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1077 continue; 1078 if (de->de_dirent->d_type == DT_CHR && 1079 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1080 continue; 1081 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1082 de->de_dirent->d_namlen) != 0) 1083 continue; 1084 if (de->de_flags & DE_WHITEOUT) 1085 break; 1086 goto notfound; 1087 } 1088 if (de == NULL) 1089 goto notfound; 1090 de->de_flags &= ~DE_WHITEOUT; 1091 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1092 return (error); 1093 notfound: 1094 sx_xunlock(&dmp->dm_lock); 1095 return (error); 1096 } 1097 1098 /* ARGSUSED */ 1099 static int 1100 devfs_open(struct vop_open_args *ap) 1101 { 1102 struct thread *td = ap->a_td; 1103 struct vnode *vp = ap->a_vp; 1104 struct cdev *dev = vp->v_rdev; 1105 struct file *fp = ap->a_fp; 1106 int error, ref, vlocked; 1107 struct cdevsw *dsw; 1108 struct file *fpop; 1109 struct mtx *mtxp; 1110 1111 if (vp->v_type == VBLK) 1112 return (ENXIO); 1113 1114 if (dev == NULL) 1115 return (ENXIO); 1116 1117 /* Make this field valid before any I/O in d_open. */ 1118 if (dev->si_iosize_max == 0) 1119 dev->si_iosize_max = DFLTPHYS; 1120 1121 dsw = dev_refthread(dev, &ref); 1122 if (dsw == NULL) 1123 return (ENXIO); 1124 if (fp == NULL && dsw->d_fdopen != NULL) { 1125 dev_relthread(dev, ref); 1126 return (ENXIO); 1127 } 1128 1129 vlocked = VOP_ISLOCKED(vp); 1130 VOP_UNLOCK(vp, 0); 1131 1132 fpop = td->td_fpop; 1133 td->td_fpop = fp; 1134 if (fp != NULL) { 1135 fp->f_data = dev; 1136 fp->f_vnode = vp; 1137 } 1138 if (dsw->d_fdopen != NULL) 1139 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1140 else 1141 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1142 /* Clean up any cdevpriv upon error. */ 1143 if (error != 0) 1144 devfs_clear_cdevpriv(); 1145 td->td_fpop = fpop; 1146 1147 vn_lock(vp, vlocked | LK_RETRY); 1148 dev_relthread(dev, ref); 1149 if (error != 0) { 1150 if (error == ERESTART) 1151 error = EINTR; 1152 return (error); 1153 } 1154 1155 #if 0 /* /dev/console */ 1156 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1157 #else 1158 if (fp == NULL) 1159 return (error); 1160 #endif 1161 if (fp->f_ops == &badfileops) 1162 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1163 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1164 1165 /* 1166 * Hint to the dofilewrite() to not force the buffer draining 1167 * on the writer to the file. Most likely, the write would 1168 * not need normal buffers. 1169 */ 1170 mtx_lock(mtxp); 1171 fp->f_vnread_flags |= FDEVFS_VNODE; 1172 mtx_unlock(mtxp); 1173 return (error); 1174 } 1175 1176 static int 1177 devfs_pathconf(struct vop_pathconf_args *ap) 1178 { 1179 1180 switch (ap->a_name) { 1181 case _PC_FILESIZEBITS: 1182 *ap->a_retval = 64; 1183 return (0); 1184 case _PC_NAME_MAX: 1185 *ap->a_retval = NAME_MAX; 1186 return (0); 1187 case _PC_LINK_MAX: 1188 *ap->a_retval = INT_MAX; 1189 return (0); 1190 case _PC_SYMLINK_MAX: 1191 *ap->a_retval = MAXPATHLEN; 1192 return (0); 1193 case _PC_MAX_CANON: 1194 if (ap->a_vp->v_vflag & VV_ISTTY) { 1195 *ap->a_retval = MAX_CANON; 1196 return (0); 1197 } 1198 return (EINVAL); 1199 case _PC_MAX_INPUT: 1200 if (ap->a_vp->v_vflag & VV_ISTTY) { 1201 *ap->a_retval = MAX_INPUT; 1202 return (0); 1203 } 1204 return (EINVAL); 1205 case _PC_VDISABLE: 1206 if (ap->a_vp->v_vflag & VV_ISTTY) { 1207 *ap->a_retval = _POSIX_VDISABLE; 1208 return (0); 1209 } 1210 return (EINVAL); 1211 case _PC_MAC_PRESENT: 1212 #ifdef MAC 1213 /* 1214 * If MAC is enabled, devfs automatically supports 1215 * trivial non-persistent label storage. 1216 */ 1217 *ap->a_retval = 1; 1218 #else 1219 *ap->a_retval = 0; 1220 #endif 1221 return (0); 1222 case _PC_CHOWN_RESTRICTED: 1223 *ap->a_retval = 1; 1224 return (0); 1225 default: 1226 return (vop_stdpathconf(ap)); 1227 } 1228 /* NOTREACHED */ 1229 } 1230 1231 /* ARGSUSED */ 1232 static int 1233 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1234 { 1235 struct cdev *dev; 1236 struct cdevsw *dsw; 1237 int error, ref; 1238 struct file *fpop; 1239 1240 fpop = td->td_fpop; 1241 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1242 if (error != 0) { 1243 error = vnops.fo_poll(fp, events, cred, td); 1244 return (error); 1245 } 1246 error = dsw->d_poll(dev, events, td); 1247 td->td_fpop = fpop; 1248 dev_relthread(dev, ref); 1249 return(error); 1250 } 1251 1252 /* 1253 * Print out the contents of a special device vnode. 1254 */ 1255 static int 1256 devfs_print(struct vop_print_args *ap) 1257 { 1258 1259 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1260 return (0); 1261 } 1262 1263 static int 1264 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1265 int flags, struct thread *td) 1266 { 1267 struct cdev *dev; 1268 int ioflag, error, ref; 1269 ssize_t resid; 1270 struct cdevsw *dsw; 1271 struct file *fpop; 1272 1273 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1274 return (EINVAL); 1275 fpop = td->td_fpop; 1276 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1277 if (error != 0) { 1278 error = vnops.fo_read(fp, uio, cred, flags, td); 1279 return (error); 1280 } 1281 resid = uio->uio_resid; 1282 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1283 if (ioflag & O_DIRECT) 1284 ioflag |= IO_DIRECT; 1285 1286 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1287 error = dsw->d_read(dev, uio, ioflag); 1288 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1289 devfs_timestamp(&dev->si_atime); 1290 td->td_fpop = fpop; 1291 dev_relthread(dev, ref); 1292 1293 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1294 return (error); 1295 } 1296 1297 static int 1298 devfs_readdir(struct vop_readdir_args *ap) 1299 { 1300 int error; 1301 struct uio *uio; 1302 struct dirent *dp; 1303 struct devfs_dirent *dd; 1304 struct devfs_dirent *de; 1305 struct devfs_mount *dmp; 1306 off_t off; 1307 int *tmp_ncookies = NULL; 1308 1309 if (ap->a_vp->v_type != VDIR) 1310 return (ENOTDIR); 1311 1312 uio = ap->a_uio; 1313 if (uio->uio_offset < 0) 1314 return (EINVAL); 1315 1316 /* 1317 * XXX: This is a temporary hack to get around this filesystem not 1318 * supporting cookies. We store the location of the ncookies pointer 1319 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1320 * and set the number of cookies to 0. We then set the pointer to 1321 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1322 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1323 * pointer to its original location before returning to the caller. 1324 */ 1325 if (ap->a_ncookies != NULL) { 1326 tmp_ncookies = ap->a_ncookies; 1327 *ap->a_ncookies = 0; 1328 ap->a_ncookies = NULL; 1329 } 1330 1331 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1332 if (devfs_populate_vp(ap->a_vp) != 0) { 1333 if (tmp_ncookies != NULL) 1334 ap->a_ncookies = tmp_ncookies; 1335 return (EIO); 1336 } 1337 error = 0; 1338 de = ap->a_vp->v_data; 1339 off = 0; 1340 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1341 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1342 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1343 continue; 1344 if (devfs_prison_check(dd, uio->uio_td)) 1345 continue; 1346 if (dd->de_dirent->d_type == DT_DIR) 1347 de = dd->de_dir; 1348 else 1349 de = dd; 1350 dp = dd->de_dirent; 1351 MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp)); 1352 if (dp->d_reclen > uio->uio_resid) 1353 break; 1354 dp->d_fileno = de->de_inode; 1355 /* NOTE: d_off is the offset for the *next* entry. */ 1356 dp->d_off = off + dp->d_reclen; 1357 if (off >= uio->uio_offset) { 1358 error = vfs_read_dirent(ap, dp, off); 1359 if (error) 1360 break; 1361 } 1362 off += dp->d_reclen; 1363 } 1364 sx_xunlock(&dmp->dm_lock); 1365 uio->uio_offset = off; 1366 1367 /* 1368 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1369 * place. 1370 */ 1371 if (tmp_ncookies != NULL) 1372 ap->a_ncookies = tmp_ncookies; 1373 1374 return (error); 1375 } 1376 1377 static int 1378 devfs_readlink(struct vop_readlink_args *ap) 1379 { 1380 struct devfs_dirent *de; 1381 1382 de = ap->a_vp->v_data; 1383 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1384 } 1385 1386 static int 1387 devfs_reclaim(struct vop_reclaim_args *ap) 1388 { 1389 struct vnode *vp; 1390 struct devfs_dirent *de; 1391 1392 vp = ap->a_vp; 1393 mtx_lock(&devfs_de_interlock); 1394 de = vp->v_data; 1395 if (de != NULL) { 1396 de->de_vnode = NULL; 1397 vp->v_data = NULL; 1398 } 1399 mtx_unlock(&devfs_de_interlock); 1400 vnode_destroy_vobject(vp); 1401 return (0); 1402 } 1403 1404 static int 1405 devfs_reclaim_vchr(struct vop_reclaim_args *ap) 1406 { 1407 struct vnode *vp; 1408 struct cdev *dev; 1409 1410 vp = ap->a_vp; 1411 MPASS(vp->v_type == VCHR); 1412 1413 devfs_reclaim(ap); 1414 1415 VI_LOCK(vp); 1416 dev_lock(); 1417 dev = vp->v_rdev; 1418 vp->v_rdev = NULL; 1419 if (dev != NULL) 1420 dev->si_usecount -= vp->v_usecount; 1421 dev_unlock(); 1422 VI_UNLOCK(vp); 1423 if (dev != NULL) 1424 dev_rel(dev); 1425 return (0); 1426 } 1427 1428 static int 1429 devfs_remove(struct vop_remove_args *ap) 1430 { 1431 struct vnode *dvp = ap->a_dvp; 1432 struct vnode *vp = ap->a_vp; 1433 struct devfs_dirent *dd; 1434 struct devfs_dirent *de, *de_covered; 1435 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1436 1437 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1438 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1439 1440 sx_xlock(&dmp->dm_lock); 1441 dd = ap->a_dvp->v_data; 1442 de = vp->v_data; 1443 if (de->de_cdp == NULL) { 1444 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1445 if (de->de_dirent->d_type == DT_LNK) { 1446 de_covered = devfs_find(dd, de->de_dirent->d_name, 1447 de->de_dirent->d_namlen, 0); 1448 if (de_covered != NULL) 1449 de_covered->de_flags &= ~DE_COVERED; 1450 } 1451 /* We need to unlock dvp because devfs_delete() may lock it. */ 1452 VOP_UNLOCK(vp, 0); 1453 if (dvp != vp) 1454 VOP_UNLOCK(dvp, 0); 1455 devfs_delete(dmp, de, 0); 1456 sx_xunlock(&dmp->dm_lock); 1457 if (dvp != vp) 1458 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1459 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1460 } else { 1461 de->de_flags |= DE_WHITEOUT; 1462 sx_xunlock(&dmp->dm_lock); 1463 } 1464 return (0); 1465 } 1466 1467 /* 1468 * Revoke is called on a tty when a terminal session ends. The vnode 1469 * is orphaned by setting v_op to deadfs so we need to let go of it 1470 * as well so that we create a new one next time around. 1471 * 1472 */ 1473 static int 1474 devfs_revoke(struct vop_revoke_args *ap) 1475 { 1476 struct vnode *vp = ap->a_vp, *vp2; 1477 struct cdev *dev; 1478 struct cdev_priv *cdp; 1479 struct devfs_dirent *de; 1480 u_int i; 1481 1482 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1483 1484 dev = vp->v_rdev; 1485 cdp = cdev2priv(dev); 1486 1487 dev_lock(); 1488 cdp->cdp_inuse++; 1489 dev_unlock(); 1490 1491 vhold(vp); 1492 vgone(vp); 1493 vdrop(vp); 1494 1495 VOP_UNLOCK(vp,0); 1496 loop: 1497 for (;;) { 1498 mtx_lock(&devfs_de_interlock); 1499 dev_lock(); 1500 vp2 = NULL; 1501 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1502 de = cdp->cdp_dirents[i]; 1503 if (de == NULL) 1504 continue; 1505 1506 vp2 = de->de_vnode; 1507 if (vp2 != NULL) { 1508 dev_unlock(); 1509 VI_LOCK(vp2); 1510 mtx_unlock(&devfs_de_interlock); 1511 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1512 curthread)) 1513 goto loop; 1514 vhold(vp2); 1515 vgone(vp2); 1516 vdrop(vp2); 1517 vput(vp2); 1518 break; 1519 } 1520 } 1521 if (vp2 != NULL) { 1522 continue; 1523 } 1524 dev_unlock(); 1525 mtx_unlock(&devfs_de_interlock); 1526 break; 1527 } 1528 dev_lock(); 1529 cdp->cdp_inuse--; 1530 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1531 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1532 dev_unlock(); 1533 dev_rel(&cdp->cdp_c); 1534 } else 1535 dev_unlock(); 1536 1537 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1538 return (0); 1539 } 1540 1541 static int 1542 devfs_rioctl(struct vop_ioctl_args *ap) 1543 { 1544 struct vnode *vp; 1545 struct devfs_mount *dmp; 1546 int error; 1547 1548 vp = ap->a_vp; 1549 vn_lock(vp, LK_SHARED | LK_RETRY); 1550 if (vp->v_iflag & VI_DOOMED) { 1551 VOP_UNLOCK(vp, 0); 1552 return (EBADF); 1553 } 1554 dmp = VFSTODEVFS(vp->v_mount); 1555 sx_xlock(&dmp->dm_lock); 1556 VOP_UNLOCK(vp, 0); 1557 DEVFS_DMP_HOLD(dmp); 1558 devfs_populate(dmp); 1559 if (DEVFS_DMP_DROP(dmp)) { 1560 sx_xunlock(&dmp->dm_lock); 1561 devfs_unmount_final(dmp); 1562 return (ENOENT); 1563 } 1564 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1565 sx_xunlock(&dmp->dm_lock); 1566 return (error); 1567 } 1568 1569 static int 1570 devfs_rread(struct vop_read_args *ap) 1571 { 1572 1573 if (ap->a_vp->v_type != VDIR) 1574 return (EINVAL); 1575 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1576 } 1577 1578 static int 1579 devfs_setattr(struct vop_setattr_args *ap) 1580 { 1581 struct devfs_dirent *de; 1582 struct vattr *vap; 1583 struct vnode *vp; 1584 struct thread *td; 1585 int c, error; 1586 uid_t uid; 1587 gid_t gid; 1588 1589 vap = ap->a_vap; 1590 vp = ap->a_vp; 1591 td = curthread; 1592 if ((vap->va_type != VNON) || 1593 (vap->va_nlink != VNOVAL) || 1594 (vap->va_fsid != VNOVAL) || 1595 (vap->va_fileid != VNOVAL) || 1596 (vap->va_blocksize != VNOVAL) || 1597 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1598 (vap->va_rdev != VNOVAL) || 1599 ((int)vap->va_bytes != VNOVAL) || 1600 (vap->va_gen != VNOVAL)) { 1601 return (EINVAL); 1602 } 1603 1604 error = devfs_populate_vp(vp); 1605 if (error != 0) 1606 return (error); 1607 1608 de = vp->v_data; 1609 if (vp->v_type == VDIR) 1610 de = de->de_dir; 1611 1612 c = 0; 1613 if (vap->va_uid == (uid_t)VNOVAL) 1614 uid = de->de_uid; 1615 else 1616 uid = vap->va_uid; 1617 if (vap->va_gid == (gid_t)VNOVAL) 1618 gid = de->de_gid; 1619 else 1620 gid = vap->va_gid; 1621 if (uid != de->de_uid || gid != de->de_gid) { 1622 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1623 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1624 error = priv_check(td, PRIV_VFS_CHOWN); 1625 if (error != 0) 1626 goto ret; 1627 } 1628 de->de_uid = uid; 1629 de->de_gid = gid; 1630 c = 1; 1631 } 1632 1633 if (vap->va_mode != (mode_t)VNOVAL) { 1634 if (ap->a_cred->cr_uid != de->de_uid) { 1635 error = priv_check(td, PRIV_VFS_ADMIN); 1636 if (error != 0) 1637 goto ret; 1638 } 1639 de->de_mode = vap->va_mode; 1640 c = 1; 1641 } 1642 1643 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1644 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1645 if (error != 0) 1646 goto ret; 1647 if (vap->va_atime.tv_sec != VNOVAL) { 1648 if (vp->v_type == VCHR) 1649 vp->v_rdev->si_atime = vap->va_atime; 1650 else 1651 de->de_atime = vap->va_atime; 1652 } 1653 if (vap->va_mtime.tv_sec != VNOVAL) { 1654 if (vp->v_type == VCHR) 1655 vp->v_rdev->si_mtime = vap->va_mtime; 1656 else 1657 de->de_mtime = vap->va_mtime; 1658 } 1659 c = 1; 1660 } 1661 1662 if (c) { 1663 if (vp->v_type == VCHR) 1664 vfs_timestamp(&vp->v_rdev->si_ctime); 1665 else 1666 vfs_timestamp(&de->de_mtime); 1667 } 1668 1669 ret: 1670 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1671 return (error); 1672 } 1673 1674 #ifdef MAC 1675 static int 1676 devfs_setlabel(struct vop_setlabel_args *ap) 1677 { 1678 struct vnode *vp; 1679 struct devfs_dirent *de; 1680 1681 vp = ap->a_vp; 1682 de = vp->v_data; 1683 1684 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1685 mac_devfs_update(vp->v_mount, de, vp); 1686 1687 return (0); 1688 } 1689 #endif 1690 1691 static int 1692 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1693 { 1694 1695 return (vnops.fo_stat(fp, sb, cred, td)); 1696 } 1697 1698 static int 1699 devfs_symlink(struct vop_symlink_args *ap) 1700 { 1701 int i, error; 1702 struct devfs_dirent *dd; 1703 struct devfs_dirent *de, *de_covered, *de_dotdot; 1704 struct devfs_mount *dmp; 1705 1706 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1707 if (error) 1708 return(error); 1709 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1710 if (devfs_populate_vp(ap->a_dvp) != 0) 1711 return (ENOENT); 1712 1713 dd = ap->a_dvp->v_data; 1714 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1715 de->de_flags = DE_USER; 1716 de->de_uid = 0; 1717 de->de_gid = 0; 1718 de->de_mode = 0755; 1719 de->de_inode = alloc_unr(devfs_inos); 1720 de->de_dir = dd; 1721 de->de_dirent->d_type = DT_LNK; 1722 i = strlen(ap->a_target) + 1; 1723 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1724 bcopy(ap->a_target, de->de_symlink, i); 1725 #ifdef MAC 1726 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1727 #endif 1728 de_covered = devfs_find(dd, de->de_dirent->d_name, 1729 de->de_dirent->d_namlen, 0); 1730 if (de_covered != NULL) { 1731 if ((de_covered->de_flags & DE_USER) != 0) { 1732 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1733 sx_xunlock(&dmp->dm_lock); 1734 return (EEXIST); 1735 } 1736 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1737 ("devfs_symlink: entry %p already covered", de_covered)); 1738 de_covered->de_flags |= DE_COVERED; 1739 } 1740 1741 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1742 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1743 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1744 devfs_dir_ref_de(dmp, dd); 1745 devfs_rules_apply(dmp, de); 1746 1747 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1748 } 1749 1750 static int 1751 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1752 { 1753 1754 return (vnops.fo_truncate(fp, length, cred, td)); 1755 } 1756 1757 static int 1758 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1759 int flags, struct thread *td) 1760 { 1761 struct cdev *dev; 1762 int error, ioflag, ref; 1763 ssize_t resid; 1764 struct cdevsw *dsw; 1765 struct file *fpop; 1766 1767 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1768 return (EINVAL); 1769 fpop = td->td_fpop; 1770 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1771 if (error != 0) { 1772 error = vnops.fo_write(fp, uio, cred, flags, td); 1773 return (error); 1774 } 1775 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1776 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1777 if (ioflag & O_DIRECT) 1778 ioflag |= IO_DIRECT; 1779 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1780 1781 resid = uio->uio_resid; 1782 1783 error = dsw->d_write(dev, uio, ioflag); 1784 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1785 devfs_timestamp(&dev->si_ctime); 1786 dev->si_mtime = dev->si_ctime; 1787 } 1788 td->td_fpop = fpop; 1789 dev_relthread(dev, ref); 1790 1791 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1792 return (error); 1793 } 1794 1795 static int 1796 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1797 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1798 struct thread *td) 1799 { 1800 struct cdev *dev; 1801 struct cdevsw *dsw; 1802 struct mount *mp; 1803 struct vnode *vp; 1804 struct file *fpop; 1805 vm_object_t object; 1806 vm_prot_t maxprot; 1807 int error, ref; 1808 1809 vp = fp->f_vnode; 1810 1811 /* 1812 * Ensure that file and memory protections are 1813 * compatible. 1814 */ 1815 mp = vp->v_mount; 1816 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 1817 maxprot = VM_PROT_NONE; 1818 if ((prot & VM_PROT_EXECUTE) != 0) 1819 return (EACCES); 1820 } else 1821 maxprot = VM_PROT_EXECUTE; 1822 if ((fp->f_flag & FREAD) != 0) 1823 maxprot |= VM_PROT_READ; 1824 else if ((prot & VM_PROT_READ) != 0) 1825 return (EACCES); 1826 1827 /* 1828 * If we are sharing potential changes via MAP_SHARED and we 1829 * are trying to get write permission although we opened it 1830 * without asking for it, bail out. 1831 * 1832 * Note that most character devices always share mappings. 1833 * The one exception is that D_MMAP_ANON devices 1834 * (i.e. /dev/zero) permit private writable mappings. 1835 * 1836 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests 1837 * as well as updating maxprot to permit writing for 1838 * D_MMAP_ANON devices rather than doing that here. 1839 */ 1840 if ((flags & MAP_SHARED) != 0) { 1841 if ((fp->f_flag & FWRITE) != 0) 1842 maxprot |= VM_PROT_WRITE; 1843 else if ((prot & VM_PROT_WRITE) != 0) 1844 return (EACCES); 1845 } 1846 maxprot &= cap_maxprot; 1847 1848 fpop = td->td_fpop; 1849 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1850 if (error != 0) 1851 return (error); 1852 1853 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff, 1854 &object); 1855 td->td_fpop = fpop; 1856 dev_relthread(dev, ref); 1857 if (error != 0) 1858 return (error); 1859 1860 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1861 foff, FALSE, td); 1862 if (error != 0) 1863 vm_object_deallocate(object); 1864 return (error); 1865 } 1866 1867 dev_t 1868 dev2udev(struct cdev *x) 1869 { 1870 if (x == NULL) 1871 return (NODEV); 1872 return (cdev2priv(x)->cdp_inode); 1873 } 1874 1875 static struct fileops devfs_ops_f = { 1876 .fo_read = devfs_read_f, 1877 .fo_write = devfs_write_f, 1878 .fo_truncate = devfs_truncate_f, 1879 .fo_ioctl = devfs_ioctl_f, 1880 .fo_poll = devfs_poll_f, 1881 .fo_kqfilter = devfs_kqfilter_f, 1882 .fo_stat = devfs_stat_f, 1883 .fo_close = devfs_close_f, 1884 .fo_chmod = vn_chmod, 1885 .fo_chown = vn_chown, 1886 .fo_sendfile = vn_sendfile, 1887 .fo_seek = vn_seek, 1888 .fo_fill_kinfo = vn_fill_kinfo, 1889 .fo_mmap = devfs_mmap_f, 1890 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1891 }; 1892 1893 /* Vops for non-CHR vnodes in /dev. */ 1894 static struct vop_vector devfs_vnodeops = { 1895 .vop_default = &default_vnodeops, 1896 1897 .vop_access = devfs_access, 1898 .vop_getattr = devfs_getattr, 1899 .vop_ioctl = devfs_rioctl, 1900 .vop_lookup = devfs_lookup, 1901 .vop_mknod = devfs_mknod, 1902 .vop_pathconf = devfs_pathconf, 1903 .vop_read = devfs_rread, 1904 .vop_readdir = devfs_readdir, 1905 .vop_readlink = devfs_readlink, 1906 .vop_reclaim = devfs_reclaim, 1907 .vop_remove = devfs_remove, 1908 .vop_revoke = devfs_revoke, 1909 .vop_setattr = devfs_setattr, 1910 #ifdef MAC 1911 .vop_setlabel = devfs_setlabel, 1912 #endif 1913 .vop_symlink = devfs_symlink, 1914 .vop_vptocnp = devfs_vptocnp, 1915 }; 1916 1917 /* Vops for VCHR vnodes in /dev. */ 1918 static struct vop_vector devfs_specops = { 1919 .vop_default = &default_vnodeops, 1920 1921 .vop_access = devfs_access, 1922 .vop_bmap = VOP_PANIC, 1923 .vop_close = devfs_close, 1924 .vop_create = VOP_PANIC, 1925 .vop_fsync = vop_stdfsync, 1926 .vop_getattr = devfs_getattr, 1927 .vop_ioctl = devfs_ioctl, 1928 .vop_link = VOP_PANIC, 1929 .vop_mkdir = VOP_PANIC, 1930 .vop_mknod = VOP_PANIC, 1931 .vop_open = devfs_open, 1932 .vop_pathconf = devfs_pathconf, 1933 .vop_poll = dead_poll, 1934 .vop_print = devfs_print, 1935 .vop_read = dead_read, 1936 .vop_readdir = VOP_PANIC, 1937 .vop_readlink = VOP_PANIC, 1938 .vop_reallocblks = VOP_PANIC, 1939 .vop_reclaim = devfs_reclaim_vchr, 1940 .vop_remove = devfs_remove, 1941 .vop_rename = VOP_PANIC, 1942 .vop_revoke = devfs_revoke, 1943 .vop_rmdir = VOP_PANIC, 1944 .vop_setattr = devfs_setattr, 1945 #ifdef MAC 1946 .vop_setlabel = devfs_setlabel, 1947 #endif 1948 .vop_strategy = VOP_PANIC, 1949 .vop_symlink = VOP_PANIC, 1950 .vop_vptocnp = devfs_vptocnp, 1951 .vop_write = dead_write, 1952 }; 1953 1954 /* 1955 * Our calling convention to the device drivers used to be that we passed 1956 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1957 * flags instead since that's what open(), close() and ioctl() takes and 1958 * we don't really want vnode.h in device drivers. 1959 * We solved the source compatibility by redefining some vnode flags to 1960 * be the same as the fcntl ones and by sending down the bitwise OR of 1961 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1962 * pulls the rug out under this. 1963 */ 1964 CTASSERT(O_NONBLOCK == IO_NDELAY); 1965 CTASSERT(O_FSYNC == IO_SYNC);
Cache object: 2bf97085a6e370eea21422243ec772b6
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]