devfs_vnops.c revision 304843
1/*- 2 * Copyright (c) 2000-2004 3 * Poul-Henning Kamp. All rights reserved. 4 * Copyright (c) 1989, 1992-1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software donated to Berkeley by 8 * Jan-Simon Pendry. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95 32 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43 33 * 34 * $FreeBSD: stable/11/sys/fs/devfs/devfs_vnops.c 304843 2016-08-26 10:04:10Z kib $ 35 */ 36 37/* 38 * TODO: 39 * mkdir: want it ? 40 */ 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/conf.h> 45#include <sys/dirent.h> 46#include <sys/fcntl.h> 47#include <sys/file.h> 48#include <sys/filedesc.h> 49#include <sys/filio.h> 50#include <sys/jail.h> 51#include <sys/kernel.h> 52#include <sys/lock.h> 53#include <sys/malloc.h> 54#include <sys/mman.h> 55#include <sys/mount.h> 56#include <sys/namei.h> 57#include <sys/priv.h> 58#include <sys/proc.h> 59#include <sys/stat.h> 60#include <sys/sx.h> 61#include <sys/sysctl.h> 62#include <sys/time.h> 63#include <sys/ttycom.h> 64#include <sys/unistd.h> 65#include <sys/vnode.h> 66 67static struct vop_vector devfs_vnodeops; 68static struct vop_vector devfs_specops; 69static struct fileops devfs_ops_f; 70 71#include <fs/devfs/devfs.h> 72#include <fs/devfs/devfs_int.h> 73 74#include <security/mac/mac_framework.h> 75 76#include <vm/vm.h> 77#include <vm/vm_extern.h> 78#include <vm/vm_object.h> 79 80static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data"); 81 82struct mtx devfs_de_interlock; 83MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); 84struct sx clone_drain_lock; 85SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock"); 86struct mtx cdevpriv_mtx; 87MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF); 88 89SYSCTL_DECL(_vfs_devfs); 90 91static int devfs_dotimes; 92SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW, 93 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision"); 94 95/* 96 * Update devfs node timestamp. Note that updates are unlocked and 97 * stat(2) could see partially updated times. 98 */ 99static void 100devfs_timestamp(struct timespec *tsp) 101{ 102 time_t ts; 103 104 if (devfs_dotimes) { 105 vfs_timestamp(tsp); 106 } else { 107 ts = time_second; 108 if (tsp->tv_sec != ts) { 109 tsp->tv_sec = ts; 110 tsp->tv_nsec = 0; 111 } 112 } 113} 114 115static int 116devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp, 117 int *ref) 118{ 119 120 *dswp = devvn_refthread(fp->f_vnode, devp, ref); 121 if (*devp != fp->f_data) { 122 if (*dswp != NULL) 123 dev_relthread(*devp, *ref); 124 return (ENXIO); 125 } 126 KASSERT((*devp)->si_refcount > 0, 127 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); 128 if (*dswp == NULL) 129 return (ENXIO); 130 curthread->td_fpop = fp; 131 return (0); 132} 133 134int 135devfs_get_cdevpriv(void **datap) 136{ 137 struct file *fp; 138 struct cdev_privdata *p; 139 int error; 140 141 fp = curthread->td_fpop; 142 if (fp == NULL) 143 return (EBADF); 144 p = fp->f_cdevpriv; 145 if (p != NULL) { 146 error = 0; 147 *datap = p->cdpd_data; 148 } else 149 error = ENOENT; 150 return (error); 151} 152 153int 154devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr) 155{ 156 struct file *fp; 157 struct cdev_priv *cdp; 158 struct cdev_privdata *p; 159 int error; 160 161 fp = curthread->td_fpop; 162 if (fp == NULL) 163 return (ENOENT); 164 cdp = cdev2priv((struct cdev *)fp->f_data); 165 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK); 166 p->cdpd_data = priv; 167 p->cdpd_dtr = priv_dtr; 168 p->cdpd_fp = fp; 169 mtx_lock(&cdevpriv_mtx); 170 if (fp->f_cdevpriv == NULL) { 171 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list); 172 fp->f_cdevpriv = p; 173 mtx_unlock(&cdevpriv_mtx); 174 error = 0; 175 } else { 176 mtx_unlock(&cdevpriv_mtx); 177 free(p, M_CDEVPDATA); 178 error = EBUSY; 179 } 180 return (error); 181} 182 183void 184devfs_destroy_cdevpriv(struct cdev_privdata *p) 185{ 186 187 mtx_assert(&cdevpriv_mtx, MA_OWNED); 188 KASSERT(p->cdpd_fp->f_cdevpriv == p, 189 ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p)); 190 p->cdpd_fp->f_cdevpriv = NULL; 191 LIST_REMOVE(p, cdpd_list); 192 mtx_unlock(&cdevpriv_mtx); 193 (p->cdpd_dtr)(p->cdpd_data); 194 free(p, M_CDEVPDATA); 195} 196 197static void 198devfs_fpdrop(struct file *fp) 199{ 200 struct cdev_privdata *p; 201 202 mtx_lock(&cdevpriv_mtx); 203 if ((p = fp->f_cdevpriv) == NULL) { 204 mtx_unlock(&cdevpriv_mtx); 205 return; 206 } 207 devfs_destroy_cdevpriv(p); 208} 209 210void 211devfs_clear_cdevpriv(void) 212{ 213 struct file *fp; 214 215 fp = curthread->td_fpop; 216 if (fp == NULL) 217 return; 218 devfs_fpdrop(fp); 219} 220 221/* 222 * On success devfs_populate_vp() returns with dmp->dm_lock held. 223 */ 224static int 225devfs_populate_vp(struct vnode *vp) 226{ 227 struct devfs_dirent *de; 228 struct devfs_mount *dmp; 229 int locked; 230 231 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp"); 232 233 dmp = VFSTODEVFS(vp->v_mount); 234 locked = VOP_ISLOCKED(vp); 235 236 sx_xlock(&dmp->dm_lock); 237 DEVFS_DMP_HOLD(dmp); 238 239 /* Can't call devfs_populate() with the vnode lock held. */ 240 VOP_UNLOCK(vp, 0); 241 devfs_populate(dmp); 242 243 sx_xunlock(&dmp->dm_lock); 244 vn_lock(vp, locked | LK_RETRY); 245 sx_xlock(&dmp->dm_lock); 246 if (DEVFS_DMP_DROP(dmp)) { 247 sx_xunlock(&dmp->dm_lock); 248 devfs_unmount_final(dmp); 249 return (ERESTART); 250 } 251 if ((vp->v_iflag & VI_DOOMED) != 0) { 252 sx_xunlock(&dmp->dm_lock); 253 return (ERESTART); 254 } 255 de = vp->v_data; 256 KASSERT(de != NULL, 257 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed")); 258 if ((de->de_flags & DE_DOOMED) != 0) { 259 sx_xunlock(&dmp->dm_lock); 260 return (ERESTART); 261 } 262 263 return (0); 264} 265 266static int 267devfs_vptocnp(struct vop_vptocnp_args *ap) 268{ 269 struct vnode *vp = ap->a_vp; 270 struct vnode **dvp = ap->a_vpp; 271 struct devfs_mount *dmp; 272 char *buf = ap->a_buf; 273 int *buflen = ap->a_buflen; 274 struct devfs_dirent *dd, *de; 275 int i, error; 276 277 dmp = VFSTODEVFS(vp->v_mount); 278 279 error = devfs_populate_vp(vp); 280 if (error != 0) 281 return (error); 282 283 i = *buflen; 284 dd = vp->v_data; 285 286 if (vp->v_type == VCHR) { 287 i -= strlen(dd->de_cdp->cdp_c.si_name); 288 if (i < 0) { 289 error = ENOMEM; 290 goto finished; 291 } 292 bcopy(dd->de_cdp->cdp_c.si_name, buf + i, 293 strlen(dd->de_cdp->cdp_c.si_name)); 294 de = dd->de_dir; 295 } else if (vp->v_type == VDIR) { 296 if (dd == dmp->dm_rootdir) { 297 *dvp = vp; 298 vref(*dvp); 299 goto finished; 300 } 301 i -= dd->de_dirent->d_namlen; 302 if (i < 0) { 303 error = ENOMEM; 304 goto finished; 305 } 306 bcopy(dd->de_dirent->d_name, buf + i, 307 dd->de_dirent->d_namlen); 308 de = dd; 309 } else { 310 error = ENOENT; 311 goto finished; 312 } 313 *buflen = i; 314 de = devfs_parent_dirent(de); 315 if (de == NULL) { 316 error = ENOENT; 317 goto finished; 318 } 319 mtx_lock(&devfs_de_interlock); 320 *dvp = de->de_vnode; 321 if (*dvp != NULL) { 322 VI_LOCK(*dvp); 323 mtx_unlock(&devfs_de_interlock); 324 vholdl(*dvp); 325 VI_UNLOCK(*dvp); 326 vref(*dvp); 327 vdrop(*dvp); 328 } else { 329 mtx_unlock(&devfs_de_interlock); 330 error = ENOENT; 331 } 332finished: 333 sx_xunlock(&dmp->dm_lock); 334 return (error); 335} 336 337/* 338 * Construct the fully qualified path name relative to the mountpoint. 339 * If a NULL cnp is provided, no '/' is appended to the resulting path. 340 */ 341char * 342devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd, 343 struct componentname *cnp) 344{ 345 int i; 346 struct devfs_dirent *de; 347 348 sx_assert(&dmp->dm_lock, SA_LOCKED); 349 350 i = SPECNAMELEN; 351 buf[i] = '\0'; 352 if (cnp != NULL) 353 i -= cnp->cn_namelen; 354 if (i < 0) 355 return (NULL); 356 if (cnp != NULL) 357 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); 358 de = dd; 359 while (de != dmp->dm_rootdir) { 360 if (cnp != NULL || i < SPECNAMELEN) { 361 i--; 362 if (i < 0) 363 return (NULL); 364 buf[i] = '/'; 365 } 366 i -= de->de_dirent->d_namlen; 367 if (i < 0) 368 return (NULL); 369 bcopy(de->de_dirent->d_name, buf + i, 370 de->de_dirent->d_namlen); 371 de = devfs_parent_dirent(de); 372 if (de == NULL) 373 return (NULL); 374 } 375 return (buf + i); 376} 377 378static int 379devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp, 380 struct devfs_dirent *de) 381{ 382 int not_found; 383 384 not_found = 0; 385 if (de->de_flags & DE_DOOMED) 386 not_found = 1; 387 if (DEVFS_DE_DROP(de)) { 388 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed")); 389 devfs_dirent_free(de); 390 } 391 if (DEVFS_DMP_DROP(dmp)) { 392 KASSERT(not_found == 1, 393 ("DEVFS mount struct freed before dirent")); 394 not_found = 2; 395 sx_xunlock(&dmp->dm_lock); 396 devfs_unmount_final(dmp); 397 } 398 if (not_found == 1 || (drop_dm_lock && not_found != 2)) 399 sx_unlock(&dmp->dm_lock); 400 return (not_found); 401} 402 403static void 404devfs_insmntque_dtr(struct vnode *vp, void *arg) 405{ 406 struct devfs_dirent *de; 407 408 de = (struct devfs_dirent *)arg; 409 mtx_lock(&devfs_de_interlock); 410 vp->v_data = NULL; 411 de->de_vnode = NULL; 412 mtx_unlock(&devfs_de_interlock); 413 vgone(vp); 414 vput(vp); 415} 416 417/* 418 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops 419 * it on return. 420 */ 421int 422devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode, 423 struct vnode **vpp) 424{ 425 int error; 426 struct vnode *vp; 427 struct cdev *dev; 428 struct devfs_mount *dmp; 429 struct cdevsw *dsw; 430 431 dmp = VFSTODEVFS(mp); 432 if (de->de_flags & DE_DOOMED) { 433 sx_xunlock(&dmp->dm_lock); 434 return (ENOENT); 435 } 436loop: 437 DEVFS_DE_HOLD(de); 438 DEVFS_DMP_HOLD(dmp); 439 mtx_lock(&devfs_de_interlock); 440 vp = de->de_vnode; 441 if (vp != NULL) { 442 VI_LOCK(vp); 443 mtx_unlock(&devfs_de_interlock); 444 sx_xunlock(&dmp->dm_lock); 445 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread); 446 sx_xlock(&dmp->dm_lock); 447 if (devfs_allocv_drop_refs(0, dmp, de)) { 448 vput(vp); 449 return (ENOENT); 450 } 451 else if ((vp->v_iflag & VI_DOOMED) != 0) { 452 mtx_lock(&devfs_de_interlock); 453 if (de->de_vnode == vp) { 454 de->de_vnode = NULL; 455 vp->v_data = NULL; 456 } 457 mtx_unlock(&devfs_de_interlock); 458 vput(vp); 459 goto loop; 460 } 461 sx_xunlock(&dmp->dm_lock); 462 *vpp = vp; 463 return (0); 464 } 465 mtx_unlock(&devfs_de_interlock); 466 if (de->de_dirent->d_type == DT_CHR) { 467 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) { 468 devfs_allocv_drop_refs(1, dmp, de); 469 return (ENOENT); 470 } 471 dev = &de->de_cdp->cdp_c; 472 } else { 473 dev = NULL; 474 } 475 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); 476 if (error != 0) { 477 devfs_allocv_drop_refs(1, dmp, de); 478 printf("devfs_allocv: failed to allocate new vnode\n"); 479 return (error); 480 } 481 482 if (de->de_dirent->d_type == DT_CHR) { 483 vp->v_type = VCHR; 484 VI_LOCK(vp); 485 dev_lock(); 486 dev_refl(dev); 487 /* XXX: v_rdev should be protect by vnode lock */ 488 vp->v_rdev = dev; 489 KASSERT(vp->v_usecount == 1, 490 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount)); 491 dev->si_usecount += vp->v_usecount; 492 /* Special casing of ttys for deadfs. Probably redundant. */ 493 dsw = dev->si_devsw; 494 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) 495 vp->v_vflag |= VV_ISTTY; 496 dev_unlock(); 497 VI_UNLOCK(vp); 498 if ((dev->si_flags & SI_ETERNAL) != 0) 499 vp->v_vflag |= VV_ETERNALDEV; 500 vp->v_op = &devfs_specops; 501 } else if (de->de_dirent->d_type == DT_DIR) { 502 vp->v_type = VDIR; 503 } else if (de->de_dirent->d_type == DT_LNK) { 504 vp->v_type = VLNK; 505 } else { 506 vp->v_type = VBAD; 507 } 508 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); 509 VN_LOCK_ASHARE(vp); 510 mtx_lock(&devfs_de_interlock); 511 vp->v_data = de; 512 de->de_vnode = vp; 513 mtx_unlock(&devfs_de_interlock); 514 error = insmntque1(vp, mp, devfs_insmntque_dtr, de); 515 if (error != 0) { 516 (void) devfs_allocv_drop_refs(1, dmp, de); 517 return (error); 518 } 519 if (devfs_allocv_drop_refs(0, dmp, de)) { 520 vput(vp); 521 return (ENOENT); 522 } 523#ifdef MAC 524 mac_devfs_vnode_associate(mp, de, vp); 525#endif 526 sx_xunlock(&dmp->dm_lock); 527 *vpp = vp; 528 return (0); 529} 530 531static int 532devfs_access(struct vop_access_args *ap) 533{ 534 struct vnode *vp = ap->a_vp; 535 struct devfs_dirent *de; 536 struct proc *p; 537 int error; 538 539 de = vp->v_data; 540 if (vp->v_type == VDIR) 541 de = de->de_dir; 542 543 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, 544 ap->a_accmode, ap->a_cred, NULL); 545 if (error == 0) 546 return (0); 547 if (error != EACCES) 548 return (error); 549 p = ap->a_td->td_proc; 550 /* We do, however, allow access to the controlling terminal */ 551 PROC_LOCK(p); 552 if (!(p->p_flag & P_CONTROLT)) { 553 PROC_UNLOCK(p); 554 return (error); 555 } 556 if (p->p_session->s_ttydp == de->de_cdp) 557 error = 0; 558 PROC_UNLOCK(p); 559 return (error); 560} 561 562_Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0, 563 "devfs-only flag reuse failed"); 564 565static int 566devfs_close(struct vop_close_args *ap) 567{ 568 struct vnode *vp = ap->a_vp, *oldvp; 569 struct thread *td = ap->a_td; 570 struct proc *p; 571 struct cdev *dev = vp->v_rdev; 572 struct cdevsw *dsw; 573 int dflags, error, ref, vp_locked; 574 575 /* 576 * XXX: Don't call d_close() if we were called because of 577 * XXX: insmntque1() failure. 578 */ 579 if (vp->v_data == NULL) 580 return (0); 581 582 /* 583 * Hack: a tty device that is a controlling terminal 584 * has a reference from the session structure. 585 * We cannot easily tell that a character device is 586 * a controlling terminal, unless it is the closing 587 * process' controlling terminal. In that case, 588 * if the reference count is 2 (this last descriptor 589 * plus the session), release the reference from the session. 590 */ 591 if (td != NULL) { 592 p = td->td_proc; 593 PROC_LOCK(p); 594 if (vp == p->p_session->s_ttyvp) { 595 PROC_UNLOCK(p); 596 oldvp = NULL; 597 sx_xlock(&proctree_lock); 598 if (vp == p->p_session->s_ttyvp) { 599 SESS_LOCK(p->p_session); 600 VI_LOCK(vp); 601 if (count_dev(dev) == 2 && 602 (vp->v_iflag & VI_DOOMED) == 0) { 603 p->p_session->s_ttyvp = NULL; 604 p->p_session->s_ttydp = NULL; 605 oldvp = vp; 606 } 607 VI_UNLOCK(vp); 608 SESS_UNLOCK(p->p_session); 609 } 610 sx_xunlock(&proctree_lock); 611 if (oldvp != NULL) 612 vrele(oldvp); 613 } else 614 PROC_UNLOCK(p); 615 } 616 /* 617 * We do not want to really close the device if it 618 * is still in use unless we are trying to close it 619 * forcibly. Since every use (buffer, vnode, swap, cmap) 620 * holds a reference to the vnode, and because we mark 621 * any other vnodes that alias this device, when the 622 * sum of the reference counts on all the aliased 623 * vnodes descends to one, we are on last close. 624 */ 625 dsw = dev_refthread(dev, &ref); 626 if (dsw == NULL) 627 return (ENXIO); 628 dflags = 0; 629 VI_LOCK(vp); 630 if (vp->v_iflag & VI_DOOMED) { 631 /* Forced close. */ 632 dflags |= FREVOKE | FNONBLOCK; 633 } else if (dsw->d_flags & D_TRACKCLOSE) { 634 /* Keep device updated on status. */ 635 } else if (count_dev(dev) > 1) { 636 VI_UNLOCK(vp); 637 dev_relthread(dev, ref); 638 return (0); 639 } 640 if (count_dev(dev) == 1) 641 dflags |= FLASTCLOSE; 642 vholdl(vp); 643 VI_UNLOCK(vp); 644 vp_locked = VOP_ISLOCKED(vp); 645 VOP_UNLOCK(vp, 0); 646 KASSERT(dev->si_refcount > 0, 647 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 648 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td); 649 dev_relthread(dev, ref); 650 vn_lock(vp, vp_locked | LK_RETRY); 651 vdrop(vp); 652 return (error); 653} 654 655static int 656devfs_close_f(struct file *fp, struct thread *td) 657{ 658 int error; 659 struct file *fpop; 660 661 /* 662 * NB: td may be NULL if this descriptor is closed due to 663 * garbage collection from a closed UNIX domain socket. 664 */ 665 fpop = curthread->td_fpop; 666 curthread->td_fpop = fp; 667 error = vnops.fo_close(fp, td); 668 curthread->td_fpop = fpop; 669 670 /* 671 * The f_cdevpriv cannot be assigned non-NULL value while we 672 * are destroying the file. 673 */ 674 if (fp->f_cdevpriv != NULL) 675 devfs_fpdrop(fp); 676 return (error); 677} 678 679static int 680devfs_fsync(struct vop_fsync_args *ap) 681{ 682 int error; 683 struct bufobj *bo; 684 struct devfs_dirent *de; 685 686 if (!vn_isdisk(ap->a_vp, &error)) { 687 bo = &ap->a_vp->v_bufobj; 688 de = ap->a_vp->v_data; 689 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) { 690 printf("Device %s went missing before all of the data " 691 "could be written to it; expect data loss.\n", 692 de->de_dirent->d_name); 693 694 error = vop_stdfsync(ap); 695 if (bo->bo_dirty.bv_cnt != 0 || error != 0) 696 panic("devfs_fsync: vop_stdfsync failed."); 697 } 698 699 return (0); 700 } 701 702 return (vop_stdfsync(ap)); 703} 704 705static int 706devfs_getattr(struct vop_getattr_args *ap) 707{ 708 struct vnode *vp = ap->a_vp; 709 struct vattr *vap = ap->a_vap; 710 struct devfs_dirent *de; 711 struct devfs_mount *dmp; 712 struct cdev *dev; 713 struct timeval boottime; 714 int error; 715 716 error = devfs_populate_vp(vp); 717 if (error != 0) 718 return (error); 719 720 dmp = VFSTODEVFS(vp->v_mount); 721 sx_xunlock(&dmp->dm_lock); 722 723 de = vp->v_data; 724 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 725 if (vp->v_type == VDIR) { 726 de = de->de_dir; 727 KASSERT(de != NULL, 728 ("Null dir dirent in devfs_getattr vp=%p", vp)); 729 } 730 vap->va_uid = de->de_uid; 731 vap->va_gid = de->de_gid; 732 vap->va_mode = de->de_mode; 733 if (vp->v_type == VLNK) 734 vap->va_size = strlen(de->de_symlink); 735 else if (vp->v_type == VDIR) 736 vap->va_size = vap->va_bytes = DEV_BSIZE; 737 else 738 vap->va_size = 0; 739 if (vp->v_type != VDIR) 740 vap->va_bytes = 0; 741 vap->va_blocksize = DEV_BSIZE; 742 vap->va_type = vp->v_type; 743 744 getboottime(&boottime); 745#define fix(aa) \ 746 do { \ 747 if ((aa).tv_sec <= 3600) { \ 748 (aa).tv_sec = boottime.tv_sec; \ 749 (aa).tv_nsec = boottime.tv_usec * 1000; \ 750 } \ 751 } while (0) 752 753 if (vp->v_type != VCHR) { 754 fix(de->de_atime); 755 vap->va_atime = de->de_atime; 756 fix(de->de_mtime); 757 vap->va_mtime = de->de_mtime; 758 fix(de->de_ctime); 759 vap->va_ctime = de->de_ctime; 760 } else { 761 dev = vp->v_rdev; 762 fix(dev->si_atime); 763 vap->va_atime = dev->si_atime; 764 fix(dev->si_mtime); 765 vap->va_mtime = dev->si_mtime; 766 fix(dev->si_ctime); 767 vap->va_ctime = dev->si_ctime; 768 769 vap->va_rdev = cdev2priv(dev)->cdp_inode; 770 } 771 vap->va_gen = 0; 772 vap->va_flags = 0; 773 vap->va_filerev = 0; 774 vap->va_nlink = de->de_links; 775 vap->va_fileid = de->de_inode; 776 777 return (error); 778} 779 780/* ARGSUSED */ 781static int 782devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 783{ 784 struct cdev *dev; 785 struct cdevsw *dsw; 786 struct vnode *vp; 787 struct vnode *vpold; 788 int error, i, ref; 789 const char *p; 790 struct fiodgname_arg *fgn; 791 struct file *fpop; 792 793 fpop = td->td_fpop; 794 error = devfs_fp_check(fp, &dev, &dsw, &ref); 795 if (error != 0) { 796 error = vnops.fo_ioctl(fp, com, data, cred, td); 797 return (error); 798 } 799 800 if (com == FIODTYPE) { 801 *(int *)data = dsw->d_flags & D_TYPEMASK; 802 td->td_fpop = fpop; 803 dev_relthread(dev, ref); 804 return (0); 805 } else if (com == FIODGNAME) { 806 fgn = data; 807 p = devtoname(dev); 808 i = strlen(p) + 1; 809 if (i > fgn->len) 810 error = EINVAL; 811 else 812 error = copyout(p, fgn->buf, i); 813 td->td_fpop = fpop; 814 dev_relthread(dev, ref); 815 return (error); 816 } 817 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td); 818 td->td_fpop = NULL; 819 dev_relthread(dev, ref); 820 if (error == ENOIOCTL) 821 error = ENOTTY; 822 if (error == 0 && com == TIOCSCTTY) { 823 vp = fp->f_vnode; 824 825 /* Do nothing if reassigning same control tty */ 826 sx_slock(&proctree_lock); 827 if (td->td_proc->p_session->s_ttyvp == vp) { 828 sx_sunlock(&proctree_lock); 829 return (0); 830 } 831 832 vpold = td->td_proc->p_session->s_ttyvp; 833 VREF(vp); 834 SESS_LOCK(td->td_proc->p_session); 835 td->td_proc->p_session->s_ttyvp = vp; 836 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 837 SESS_UNLOCK(td->td_proc->p_session); 838 839 sx_sunlock(&proctree_lock); 840 841 /* Get rid of reference to old control tty */ 842 if (vpold) 843 vrele(vpold); 844 } 845 return (error); 846} 847 848/* ARGSUSED */ 849static int 850devfs_kqfilter_f(struct file *fp, struct knote *kn) 851{ 852 struct cdev *dev; 853 struct cdevsw *dsw; 854 int error, ref; 855 struct file *fpop; 856 struct thread *td; 857 858 td = curthread; 859 fpop = td->td_fpop; 860 error = devfs_fp_check(fp, &dev, &dsw, &ref); 861 if (error) 862 return (error); 863 error = dsw->d_kqfilter(dev, kn); 864 td->td_fpop = fpop; 865 dev_relthread(dev, ref); 866 return (error); 867} 868 869static inline int 870devfs_prison_check(struct devfs_dirent *de, struct thread *td) 871{ 872 struct cdev_priv *cdp; 873 struct ucred *dcr; 874 struct proc *p; 875 int error; 876 877 cdp = de->de_cdp; 878 if (cdp == NULL) 879 return (0); 880 dcr = cdp->cdp_c.si_cred; 881 if (dcr == NULL) 882 return (0); 883 884 error = prison_check(td->td_ucred, dcr); 885 if (error == 0) 886 return (0); 887 /* We do, however, allow access to the controlling terminal */ 888 p = td->td_proc; 889 PROC_LOCK(p); 890 if (!(p->p_flag & P_CONTROLT)) { 891 PROC_UNLOCK(p); 892 return (error); 893 } 894 if (p->p_session->s_ttydp == cdp) 895 error = 0; 896 PROC_UNLOCK(p); 897 return (error); 898} 899 900static int 901devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 902{ 903 struct componentname *cnp; 904 struct vnode *dvp, **vpp; 905 struct thread *td; 906 struct devfs_dirent *de, *dd; 907 struct devfs_dirent **dde; 908 struct devfs_mount *dmp; 909 struct cdev *cdev; 910 int error, flags, nameiop, dvplocked; 911 char specname[SPECNAMELEN + 1], *pname; 912 913 cnp = ap->a_cnp; 914 vpp = ap->a_vpp; 915 dvp = ap->a_dvp; 916 pname = cnp->cn_nameptr; 917 td = cnp->cn_thread; 918 flags = cnp->cn_flags; 919 nameiop = cnp->cn_nameiop; 920 dmp = VFSTODEVFS(dvp->v_mount); 921 dd = dvp->v_data; 922 *vpp = NULLVP; 923 924 if ((flags & ISLASTCN) && nameiop == RENAME) 925 return (EOPNOTSUPP); 926 927 if (dvp->v_type != VDIR) 928 return (ENOTDIR); 929 930 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 931 return (EIO); 932 933 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 934 if (error) 935 return (error); 936 937 if (cnp->cn_namelen == 1 && *pname == '.') { 938 if ((flags & ISLASTCN) && nameiop != LOOKUP) 939 return (EINVAL); 940 *vpp = dvp; 941 VREF(dvp); 942 return (0); 943 } 944 945 if (flags & ISDOTDOT) { 946 if ((flags & ISLASTCN) && nameiop != LOOKUP) 947 return (EINVAL); 948 de = devfs_parent_dirent(dd); 949 if (de == NULL) 950 return (ENOENT); 951 dvplocked = VOP_ISLOCKED(dvp); 952 VOP_UNLOCK(dvp, 0); 953 error = devfs_allocv(de, dvp->v_mount, 954 cnp->cn_lkflags & LK_TYPE_MASK, vpp); 955 *dm_unlock = 0; 956 vn_lock(dvp, dvplocked | LK_RETRY); 957 return (error); 958 } 959 960 dd = dvp->v_data; 961 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 962 while (de == NULL) { /* While(...) so we can use break */ 963 964 if (nameiop == DELETE) 965 return (ENOENT); 966 967 /* 968 * OK, we didn't have an entry for the name we were asked for 969 * so we try to see if anybody can create it on demand. 970 */ 971 pname = devfs_fqpn(specname, dmp, dd, cnp); 972 if (pname == NULL) 973 break; 974 975 cdev = NULL; 976 DEVFS_DMP_HOLD(dmp); 977 sx_xunlock(&dmp->dm_lock); 978 sx_slock(&clone_drain_lock); 979 EVENTHANDLER_INVOKE(dev_clone, 980 td->td_ucred, pname, strlen(pname), &cdev); 981 sx_sunlock(&clone_drain_lock); 982 983 if (cdev == NULL) 984 sx_xlock(&dmp->dm_lock); 985 else if (devfs_populate_vp(dvp) != 0) { 986 *dm_unlock = 0; 987 sx_xlock(&dmp->dm_lock); 988 if (DEVFS_DMP_DROP(dmp)) { 989 sx_xunlock(&dmp->dm_lock); 990 devfs_unmount_final(dmp); 991 } else 992 sx_xunlock(&dmp->dm_lock); 993 dev_rel(cdev); 994 return (ENOENT); 995 } 996 if (DEVFS_DMP_DROP(dmp)) { 997 *dm_unlock = 0; 998 sx_xunlock(&dmp->dm_lock); 999 devfs_unmount_final(dmp); 1000 if (cdev != NULL) 1001 dev_rel(cdev); 1002 return (ENOENT); 1003 } 1004 1005 if (cdev == NULL) 1006 break; 1007 1008 dev_lock(); 1009 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 1010 if (dde != NULL && *dde != NULL) 1011 de = *dde; 1012 dev_unlock(); 1013 dev_rel(cdev); 1014 break; 1015 } 1016 1017 if (de == NULL || de->de_flags & DE_WHITEOUT) { 1018 if ((nameiop == CREATE || nameiop == RENAME) && 1019 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 1020 cnp->cn_flags |= SAVENAME; 1021 return (EJUSTRETURN); 1022 } 1023 return (ENOENT); 1024 } 1025 1026 if (devfs_prison_check(de, td)) 1027 return (ENOENT); 1028 1029 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 1030 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 1031 if (error) 1032 return (error); 1033 if (*vpp == dvp) { 1034 VREF(dvp); 1035 *vpp = dvp; 1036 return (0); 1037 } 1038 } 1039 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK, 1040 vpp); 1041 *dm_unlock = 0; 1042 return (error); 1043} 1044 1045static int 1046devfs_lookup(struct vop_lookup_args *ap) 1047{ 1048 int j; 1049 struct devfs_mount *dmp; 1050 int dm_unlock; 1051 1052 if (devfs_populate_vp(ap->a_dvp) != 0) 1053 return (ENOTDIR); 1054 1055 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1056 dm_unlock = 1; 1057 j = devfs_lookupx(ap, &dm_unlock); 1058 if (dm_unlock == 1) 1059 sx_xunlock(&dmp->dm_lock); 1060 return (j); 1061} 1062 1063static int 1064devfs_mknod(struct vop_mknod_args *ap) 1065{ 1066 struct componentname *cnp; 1067 struct vnode *dvp, **vpp; 1068 struct devfs_dirent *dd, *de; 1069 struct devfs_mount *dmp; 1070 int error; 1071 1072 /* 1073 * The only type of node we should be creating here is a 1074 * character device, for anything else return EOPNOTSUPP. 1075 */ 1076 if (ap->a_vap->va_type != VCHR) 1077 return (EOPNOTSUPP); 1078 dvp = ap->a_dvp; 1079 dmp = VFSTODEVFS(dvp->v_mount); 1080 1081 cnp = ap->a_cnp; 1082 vpp = ap->a_vpp; 1083 dd = dvp->v_data; 1084 1085 error = ENOENT; 1086 sx_xlock(&dmp->dm_lock); 1087 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1088 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1089 continue; 1090 if (de->de_dirent->d_type == DT_CHR && 1091 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1092 continue; 1093 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1094 de->de_dirent->d_namlen) != 0) 1095 continue; 1096 if (de->de_flags & DE_WHITEOUT) 1097 break; 1098 goto notfound; 1099 } 1100 if (de == NULL) 1101 goto notfound; 1102 de->de_flags &= ~DE_WHITEOUT; 1103 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1104 return (error); 1105notfound: 1106 sx_xunlock(&dmp->dm_lock); 1107 return (error); 1108} 1109 1110/* ARGSUSED */ 1111static int 1112devfs_open(struct vop_open_args *ap) 1113{ 1114 struct thread *td = ap->a_td; 1115 struct vnode *vp = ap->a_vp; 1116 struct cdev *dev = vp->v_rdev; 1117 struct file *fp = ap->a_fp; 1118 int error, ref, vlocked; 1119 struct cdevsw *dsw; 1120 struct file *fpop; 1121 struct mtx *mtxp; 1122 1123 if (vp->v_type == VBLK) 1124 return (ENXIO); 1125 1126 if (dev == NULL) 1127 return (ENXIO); 1128 1129 /* Make this field valid before any I/O in d_open. */ 1130 if (dev->si_iosize_max == 0) 1131 dev->si_iosize_max = DFLTPHYS; 1132 1133 dsw = dev_refthread(dev, &ref); 1134 if (dsw == NULL) 1135 return (ENXIO); 1136 if (fp == NULL && dsw->d_fdopen != NULL) { 1137 dev_relthread(dev, ref); 1138 return (ENXIO); 1139 } 1140 1141 vlocked = VOP_ISLOCKED(vp); 1142 VOP_UNLOCK(vp, 0); 1143 1144 fpop = td->td_fpop; 1145 td->td_fpop = fp; 1146 if (fp != NULL) { 1147 fp->f_data = dev; 1148 fp->f_vnode = vp; 1149 } 1150 if (dsw->d_fdopen != NULL) 1151 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1152 else 1153 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1154 /* Clean up any cdevpriv upon error. */ 1155 if (error != 0) 1156 devfs_clear_cdevpriv(); 1157 td->td_fpop = fpop; 1158 1159 vn_lock(vp, vlocked | LK_RETRY); 1160 dev_relthread(dev, ref); 1161 if (error != 0) { 1162 if (error == ERESTART) 1163 error = EINTR; 1164 return (error); 1165 } 1166 1167#if 0 /* /dev/console */ 1168 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1169#else 1170 if (fp == NULL) 1171 return (error); 1172#endif 1173 if (fp->f_ops == &badfileops) 1174 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1175 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1176 1177 /* 1178 * Hint to the dofilewrite() to not force the buffer draining 1179 * on the writer to the file. Most likely, the write would 1180 * not need normal buffers. 1181 */ 1182 mtx_lock(mtxp); 1183 fp->f_vnread_flags |= FDEVFS_VNODE; 1184 mtx_unlock(mtxp); 1185 return (error); 1186} 1187 1188static int 1189devfs_pathconf(struct vop_pathconf_args *ap) 1190{ 1191 1192 switch (ap->a_name) { 1193 case _PC_MAC_PRESENT: 1194#ifdef MAC 1195 /* 1196 * If MAC is enabled, devfs automatically supports 1197 * trivial non-persistant label storage. 1198 */ 1199 *ap->a_retval = 1; 1200#else 1201 *ap->a_retval = 0; 1202#endif 1203 return (0); 1204 default: 1205 return (vop_stdpathconf(ap)); 1206 } 1207 /* NOTREACHED */ 1208} 1209 1210/* ARGSUSED */ 1211static int 1212devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1213{ 1214 struct cdev *dev; 1215 struct cdevsw *dsw; 1216 int error, ref; 1217 struct file *fpop; 1218 1219 fpop = td->td_fpop; 1220 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1221 if (error != 0) { 1222 error = vnops.fo_poll(fp, events, cred, td); 1223 return (error); 1224 } 1225 error = dsw->d_poll(dev, events, td); 1226 td->td_fpop = fpop; 1227 dev_relthread(dev, ref); 1228 return(error); 1229} 1230 1231/* 1232 * Print out the contents of a special device vnode. 1233 */ 1234static int 1235devfs_print(struct vop_print_args *ap) 1236{ 1237 1238 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1239 return (0); 1240} 1241 1242static int 1243devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1244 int flags, struct thread *td) 1245{ 1246 struct cdev *dev; 1247 int ioflag, error, ref; 1248 ssize_t resid; 1249 struct cdevsw *dsw; 1250 struct file *fpop; 1251 1252 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1253 return (EINVAL); 1254 fpop = td->td_fpop; 1255 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1256 if (error != 0) { 1257 error = vnops.fo_read(fp, uio, cred, flags, td); 1258 return (error); 1259 } 1260 resid = uio->uio_resid; 1261 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1262 if (ioflag & O_DIRECT) 1263 ioflag |= IO_DIRECT; 1264 1265 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1266 error = dsw->d_read(dev, uio, ioflag); 1267 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1268 devfs_timestamp(&dev->si_atime); 1269 td->td_fpop = fpop; 1270 dev_relthread(dev, ref); 1271 1272 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1273 return (error); 1274} 1275 1276static int 1277devfs_readdir(struct vop_readdir_args *ap) 1278{ 1279 int error; 1280 struct uio *uio; 1281 struct dirent *dp; 1282 struct devfs_dirent *dd; 1283 struct devfs_dirent *de; 1284 struct devfs_mount *dmp; 1285 off_t off; 1286 int *tmp_ncookies = NULL; 1287 1288 if (ap->a_vp->v_type != VDIR) 1289 return (ENOTDIR); 1290 1291 uio = ap->a_uio; 1292 if (uio->uio_offset < 0) 1293 return (EINVAL); 1294 1295 /* 1296 * XXX: This is a temporary hack to get around this filesystem not 1297 * supporting cookies. We store the location of the ncookies pointer 1298 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1299 * and set the number of cookies to 0. We then set the pointer to 1300 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1301 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1302 * pointer to its original location before returning to the caller. 1303 */ 1304 if (ap->a_ncookies != NULL) { 1305 tmp_ncookies = ap->a_ncookies; 1306 *ap->a_ncookies = 0; 1307 ap->a_ncookies = NULL; 1308 } 1309 1310 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1311 if (devfs_populate_vp(ap->a_vp) != 0) { 1312 if (tmp_ncookies != NULL) 1313 ap->a_ncookies = tmp_ncookies; 1314 return (EIO); 1315 } 1316 error = 0; 1317 de = ap->a_vp->v_data; 1318 off = 0; 1319 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1320 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1321 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1322 continue; 1323 if (devfs_prison_check(dd, uio->uio_td)) 1324 continue; 1325 if (dd->de_dirent->d_type == DT_DIR) 1326 de = dd->de_dir; 1327 else 1328 de = dd; 1329 dp = dd->de_dirent; 1330 if (dp->d_reclen > uio->uio_resid) 1331 break; 1332 dp->d_fileno = de->de_inode; 1333 if (off >= uio->uio_offset) { 1334 error = vfs_read_dirent(ap, dp, off); 1335 if (error) 1336 break; 1337 } 1338 off += dp->d_reclen; 1339 } 1340 sx_xunlock(&dmp->dm_lock); 1341 uio->uio_offset = off; 1342 1343 /* 1344 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1345 * place. 1346 */ 1347 if (tmp_ncookies != NULL) 1348 ap->a_ncookies = tmp_ncookies; 1349 1350 return (error); 1351} 1352 1353static int 1354devfs_readlink(struct vop_readlink_args *ap) 1355{ 1356 struct devfs_dirent *de; 1357 1358 de = ap->a_vp->v_data; 1359 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1360} 1361 1362static int 1363devfs_reclaim(struct vop_reclaim_args *ap) 1364{ 1365 struct vnode *vp; 1366 struct devfs_dirent *de; 1367 1368 vp = ap->a_vp; 1369 mtx_lock(&devfs_de_interlock); 1370 de = vp->v_data; 1371 if (de != NULL) { 1372 de->de_vnode = NULL; 1373 vp->v_data = NULL; 1374 } 1375 mtx_unlock(&devfs_de_interlock); 1376 vnode_destroy_vobject(vp); 1377 return (0); 1378} 1379 1380static int 1381devfs_reclaim_vchr(struct vop_reclaim_args *ap) 1382{ 1383 struct vnode *vp; 1384 struct cdev *dev; 1385 1386 vp = ap->a_vp; 1387 MPASS(vp->v_type == VCHR); 1388 1389 devfs_reclaim(ap); 1390 1391 VI_LOCK(vp); 1392 dev_lock(); 1393 dev = vp->v_rdev; 1394 vp->v_rdev = NULL; 1395 if (dev != NULL) 1396 dev->si_usecount -= vp->v_usecount; 1397 dev_unlock(); 1398 VI_UNLOCK(vp); 1399 if (dev != NULL) 1400 dev_rel(dev); 1401 return (0); 1402} 1403 1404static int 1405devfs_remove(struct vop_remove_args *ap) 1406{ 1407 struct vnode *dvp = ap->a_dvp; 1408 struct vnode *vp = ap->a_vp; 1409 struct devfs_dirent *dd; 1410 struct devfs_dirent *de, *de_covered; 1411 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1412 1413 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1414 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1415 1416 sx_xlock(&dmp->dm_lock); 1417 dd = ap->a_dvp->v_data; 1418 de = vp->v_data; 1419 if (de->de_cdp == NULL) { 1420 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1421 if (de->de_dirent->d_type == DT_LNK) { 1422 de_covered = devfs_find(dd, de->de_dirent->d_name, 1423 de->de_dirent->d_namlen, 0); 1424 if (de_covered != NULL) 1425 de_covered->de_flags &= ~DE_COVERED; 1426 } 1427 /* We need to unlock dvp because devfs_delete() may lock it. */ 1428 VOP_UNLOCK(vp, 0); 1429 if (dvp != vp) 1430 VOP_UNLOCK(dvp, 0); 1431 devfs_delete(dmp, de, 0); 1432 sx_xunlock(&dmp->dm_lock); 1433 if (dvp != vp) 1434 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1435 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1436 } else { 1437 de->de_flags |= DE_WHITEOUT; 1438 sx_xunlock(&dmp->dm_lock); 1439 } 1440 return (0); 1441} 1442 1443/* 1444 * Revoke is called on a tty when a terminal session ends. The vnode 1445 * is orphaned by setting v_op to deadfs so we need to let go of it 1446 * as well so that we create a new one next time around. 1447 * 1448 */ 1449static int 1450devfs_revoke(struct vop_revoke_args *ap) 1451{ 1452 struct vnode *vp = ap->a_vp, *vp2; 1453 struct cdev *dev; 1454 struct cdev_priv *cdp; 1455 struct devfs_dirent *de; 1456 u_int i; 1457 1458 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1459 1460 dev = vp->v_rdev; 1461 cdp = cdev2priv(dev); 1462 1463 dev_lock(); 1464 cdp->cdp_inuse++; 1465 dev_unlock(); 1466 1467 vhold(vp); 1468 vgone(vp); 1469 vdrop(vp); 1470 1471 VOP_UNLOCK(vp,0); 1472 loop: 1473 for (;;) { 1474 mtx_lock(&devfs_de_interlock); 1475 dev_lock(); 1476 vp2 = NULL; 1477 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1478 de = cdp->cdp_dirents[i]; 1479 if (de == NULL) 1480 continue; 1481 1482 vp2 = de->de_vnode; 1483 if (vp2 != NULL) { 1484 dev_unlock(); 1485 VI_LOCK(vp2); 1486 mtx_unlock(&devfs_de_interlock); 1487 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1488 curthread)) 1489 goto loop; 1490 vhold(vp2); 1491 vgone(vp2); 1492 vdrop(vp2); 1493 vput(vp2); 1494 break; 1495 } 1496 } 1497 if (vp2 != NULL) { 1498 continue; 1499 } 1500 dev_unlock(); 1501 mtx_unlock(&devfs_de_interlock); 1502 break; 1503 } 1504 dev_lock(); 1505 cdp->cdp_inuse--; 1506 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1507 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1508 dev_unlock(); 1509 dev_rel(&cdp->cdp_c); 1510 } else 1511 dev_unlock(); 1512 1513 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1514 return (0); 1515} 1516 1517static int 1518devfs_rioctl(struct vop_ioctl_args *ap) 1519{ 1520 struct vnode *vp; 1521 struct devfs_mount *dmp; 1522 int error; 1523 1524 vp = ap->a_vp; 1525 vn_lock(vp, LK_SHARED | LK_RETRY); 1526 if (vp->v_iflag & VI_DOOMED) { 1527 VOP_UNLOCK(vp, 0); 1528 return (EBADF); 1529 } 1530 dmp = VFSTODEVFS(vp->v_mount); 1531 sx_xlock(&dmp->dm_lock); 1532 VOP_UNLOCK(vp, 0); 1533 DEVFS_DMP_HOLD(dmp); 1534 devfs_populate(dmp); 1535 if (DEVFS_DMP_DROP(dmp)) { 1536 sx_xunlock(&dmp->dm_lock); 1537 devfs_unmount_final(dmp); 1538 return (ENOENT); 1539 } 1540 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1541 sx_xunlock(&dmp->dm_lock); 1542 return (error); 1543} 1544 1545static int 1546devfs_rread(struct vop_read_args *ap) 1547{ 1548 1549 if (ap->a_vp->v_type != VDIR) 1550 return (EINVAL); 1551 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1552} 1553 1554static int 1555devfs_setattr(struct vop_setattr_args *ap) 1556{ 1557 struct devfs_dirent *de; 1558 struct vattr *vap; 1559 struct vnode *vp; 1560 struct thread *td; 1561 int c, error; 1562 uid_t uid; 1563 gid_t gid; 1564 1565 vap = ap->a_vap; 1566 vp = ap->a_vp; 1567 td = curthread; 1568 if ((vap->va_type != VNON) || 1569 (vap->va_nlink != VNOVAL) || 1570 (vap->va_fsid != VNOVAL) || 1571 (vap->va_fileid != VNOVAL) || 1572 (vap->va_blocksize != VNOVAL) || 1573 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1574 (vap->va_rdev != VNOVAL) || 1575 ((int)vap->va_bytes != VNOVAL) || 1576 (vap->va_gen != VNOVAL)) { 1577 return (EINVAL); 1578 } 1579 1580 error = devfs_populate_vp(vp); 1581 if (error != 0) 1582 return (error); 1583 1584 de = vp->v_data; 1585 if (vp->v_type == VDIR) 1586 de = de->de_dir; 1587 1588 c = 0; 1589 if (vap->va_uid == (uid_t)VNOVAL) 1590 uid = de->de_uid; 1591 else 1592 uid = vap->va_uid; 1593 if (vap->va_gid == (gid_t)VNOVAL) 1594 gid = de->de_gid; 1595 else 1596 gid = vap->va_gid; 1597 if (uid != de->de_uid || gid != de->de_gid) { 1598 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1599 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1600 error = priv_check(td, PRIV_VFS_CHOWN); 1601 if (error != 0) 1602 goto ret; 1603 } 1604 de->de_uid = uid; 1605 de->de_gid = gid; 1606 c = 1; 1607 } 1608 1609 if (vap->va_mode != (mode_t)VNOVAL) { 1610 if (ap->a_cred->cr_uid != de->de_uid) { 1611 error = priv_check(td, PRIV_VFS_ADMIN); 1612 if (error != 0) 1613 goto ret; 1614 } 1615 de->de_mode = vap->va_mode; 1616 c = 1; 1617 } 1618 1619 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1620 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1621 if (error != 0) 1622 goto ret; 1623 if (vap->va_atime.tv_sec != VNOVAL) { 1624 if (vp->v_type == VCHR) 1625 vp->v_rdev->si_atime = vap->va_atime; 1626 else 1627 de->de_atime = vap->va_atime; 1628 } 1629 if (vap->va_mtime.tv_sec != VNOVAL) { 1630 if (vp->v_type == VCHR) 1631 vp->v_rdev->si_mtime = vap->va_mtime; 1632 else 1633 de->de_mtime = vap->va_mtime; 1634 } 1635 c = 1; 1636 } 1637 1638 if (c) { 1639 if (vp->v_type == VCHR) 1640 vfs_timestamp(&vp->v_rdev->si_ctime); 1641 else 1642 vfs_timestamp(&de->de_mtime); 1643 } 1644 1645ret: 1646 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1647 return (error); 1648} 1649 1650#ifdef MAC 1651static int 1652devfs_setlabel(struct vop_setlabel_args *ap) 1653{ 1654 struct vnode *vp; 1655 struct devfs_dirent *de; 1656 1657 vp = ap->a_vp; 1658 de = vp->v_data; 1659 1660 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1661 mac_devfs_update(vp->v_mount, de, vp); 1662 1663 return (0); 1664} 1665#endif 1666 1667static int 1668devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1669{ 1670 1671 return (vnops.fo_stat(fp, sb, cred, td)); 1672} 1673 1674static int 1675devfs_symlink(struct vop_symlink_args *ap) 1676{ 1677 int i, error; 1678 struct devfs_dirent *dd; 1679 struct devfs_dirent *de, *de_covered, *de_dotdot; 1680 struct devfs_mount *dmp; 1681 1682 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1683 if (error) 1684 return(error); 1685 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1686 if (devfs_populate_vp(ap->a_dvp) != 0) 1687 return (ENOENT); 1688 1689 dd = ap->a_dvp->v_data; 1690 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1691 de->de_flags = DE_USER; 1692 de->de_uid = 0; 1693 de->de_gid = 0; 1694 de->de_mode = 0755; 1695 de->de_inode = alloc_unr(devfs_inos); 1696 de->de_dir = dd; 1697 de->de_dirent->d_type = DT_LNK; 1698 i = strlen(ap->a_target) + 1; 1699 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1700 bcopy(ap->a_target, de->de_symlink, i); 1701#ifdef MAC 1702 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1703#endif 1704 de_covered = devfs_find(dd, de->de_dirent->d_name, 1705 de->de_dirent->d_namlen, 0); 1706 if (de_covered != NULL) { 1707 if ((de_covered->de_flags & DE_USER) != 0) { 1708 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1709 sx_xunlock(&dmp->dm_lock); 1710 return (EEXIST); 1711 } 1712 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1713 ("devfs_symlink: entry %p already covered", de_covered)); 1714 de_covered->de_flags |= DE_COVERED; 1715 } 1716 1717 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1718 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1719 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1720 devfs_dir_ref_de(dmp, dd); 1721 devfs_rules_apply(dmp, de); 1722 1723 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1724} 1725 1726static int 1727devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1728{ 1729 1730 return (vnops.fo_truncate(fp, length, cred, td)); 1731} 1732 1733static int 1734devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1735 int flags, struct thread *td) 1736{ 1737 struct cdev *dev; 1738 int error, ioflag, ref; 1739 ssize_t resid; 1740 struct cdevsw *dsw; 1741 struct file *fpop; 1742 1743 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1744 return (EINVAL); 1745 fpop = td->td_fpop; 1746 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1747 if (error != 0) { 1748 error = vnops.fo_write(fp, uio, cred, flags, td); 1749 return (error); 1750 } 1751 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1752 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1753 if (ioflag & O_DIRECT) 1754 ioflag |= IO_DIRECT; 1755 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1756 1757 resid = uio->uio_resid; 1758 1759 error = dsw->d_write(dev, uio, ioflag); 1760 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1761 devfs_timestamp(&dev->si_ctime); 1762 dev->si_mtime = dev->si_ctime; 1763 } 1764 td->td_fpop = fpop; 1765 dev_relthread(dev, ref); 1766 1767 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1768 return (error); 1769} 1770 1771static int 1772devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1773 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1774 struct thread *td) 1775{ 1776 struct cdev *dev; 1777 struct cdevsw *dsw; 1778 struct mount *mp; 1779 struct vnode *vp; 1780 struct file *fpop; 1781 vm_object_t object; 1782 vm_prot_t maxprot; 1783 int error, ref; 1784 1785 vp = fp->f_vnode; 1786 1787 /* 1788 * Ensure that file and memory protections are 1789 * compatible. 1790 */ 1791 mp = vp->v_mount; 1792 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) 1793 maxprot = VM_PROT_NONE; 1794 else 1795 maxprot = VM_PROT_EXECUTE; 1796 if ((fp->f_flag & FREAD) != 0) 1797 maxprot |= VM_PROT_READ; 1798 else if ((prot & VM_PROT_READ) != 0) 1799 return (EACCES); 1800 1801 /* 1802 * If we are sharing potential changes via MAP_SHARED and we 1803 * are trying to get write permission although we opened it 1804 * without asking for it, bail out. 1805 * 1806 * Note that most character devices always share mappings. 1807 * The one exception is that D_MMAP_ANON devices 1808 * (i.e. /dev/zero) permit private writable mappings. 1809 * 1810 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests 1811 * as well as updating maxprot to permit writing for 1812 * D_MMAP_ANON devices rather than doing that here. 1813 */ 1814 if ((flags & MAP_SHARED) != 0) { 1815 if ((fp->f_flag & FWRITE) != 0) 1816 maxprot |= VM_PROT_WRITE; 1817 else if ((prot & VM_PROT_WRITE) != 0) 1818 return (EACCES); 1819 } 1820 maxprot &= cap_maxprot; 1821 1822 fpop = td->td_fpop; 1823 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1824 if (error != 0) 1825 return (error); 1826 1827 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff, 1828 &object); 1829 td->td_fpop = fpop; 1830 dev_relthread(dev, ref); 1831 if (error != 0) 1832 return (error); 1833 1834 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1835 foff, FALSE, td); 1836 if (error != 0) 1837 vm_object_deallocate(object); 1838 return (error); 1839} 1840 1841dev_t 1842dev2udev(struct cdev *x) 1843{ 1844 if (x == NULL) 1845 return (NODEV); 1846 return (cdev2priv(x)->cdp_inode); 1847} 1848 1849static struct fileops devfs_ops_f = { 1850 .fo_read = devfs_read_f, 1851 .fo_write = devfs_write_f, 1852 .fo_truncate = devfs_truncate_f, 1853 .fo_ioctl = devfs_ioctl_f, 1854 .fo_poll = devfs_poll_f, 1855 .fo_kqfilter = devfs_kqfilter_f, 1856 .fo_stat = devfs_stat_f, 1857 .fo_close = devfs_close_f, 1858 .fo_chmod = vn_chmod, 1859 .fo_chown = vn_chown, 1860 .fo_sendfile = vn_sendfile, 1861 .fo_seek = vn_seek, 1862 .fo_fill_kinfo = vn_fill_kinfo, 1863 .fo_mmap = devfs_mmap_f, 1864 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1865}; 1866 1867static struct vop_vector devfs_vnodeops = { 1868 .vop_default = &default_vnodeops, 1869 1870 .vop_access = devfs_access, 1871 .vop_getattr = devfs_getattr, 1872 .vop_ioctl = devfs_rioctl, 1873 .vop_lookup = devfs_lookup, 1874 .vop_mknod = devfs_mknod, 1875 .vop_pathconf = devfs_pathconf, 1876 .vop_read = devfs_rread, 1877 .vop_readdir = devfs_readdir, 1878 .vop_readlink = devfs_readlink, 1879 .vop_reclaim = devfs_reclaim, 1880 .vop_remove = devfs_remove, 1881 .vop_revoke = devfs_revoke, 1882 .vop_setattr = devfs_setattr, 1883#ifdef MAC 1884 .vop_setlabel = devfs_setlabel, 1885#endif 1886 .vop_symlink = devfs_symlink, 1887 .vop_vptocnp = devfs_vptocnp, 1888}; 1889 1890static struct vop_vector devfs_specops = { 1891 .vop_default = &default_vnodeops, 1892 1893 .vop_access = devfs_access, 1894 .vop_bmap = VOP_PANIC, 1895 .vop_close = devfs_close, 1896 .vop_create = VOP_PANIC, 1897 .vop_fsync = devfs_fsync, 1898 .vop_getattr = devfs_getattr, 1899 .vop_link = VOP_PANIC, 1900 .vop_mkdir = VOP_PANIC, 1901 .vop_mknod = VOP_PANIC, 1902 .vop_open = devfs_open, 1903 .vop_pathconf = devfs_pathconf, 1904 .vop_poll = dead_poll, 1905 .vop_print = devfs_print, 1906 .vop_read = dead_read, 1907 .vop_readdir = VOP_PANIC, 1908 .vop_readlink = VOP_PANIC, 1909 .vop_reallocblks = VOP_PANIC, 1910 .vop_reclaim = devfs_reclaim_vchr, 1911 .vop_remove = devfs_remove, 1912 .vop_rename = VOP_PANIC, 1913 .vop_revoke = devfs_revoke, 1914 .vop_rmdir = VOP_PANIC, 1915 .vop_setattr = devfs_setattr, 1916#ifdef MAC 1917 .vop_setlabel = devfs_setlabel, 1918#endif 1919 .vop_strategy = VOP_PANIC, 1920 .vop_symlink = VOP_PANIC, 1921 .vop_vptocnp = devfs_vptocnp, 1922 .vop_write = dead_write, 1923}; 1924 1925/* 1926 * Our calling convention to the device drivers used to be that we passed 1927 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1928 * flags instead since that's what open(), close() and ioctl() takes and 1929 * we don't really want vnode.h in device drivers. 1930 * We solved the source compatibility by redefining some vnode flags to 1931 * be the same as the fcntl ones and by sending down the bitwise OR of 1932 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1933 * pulls the rug out under this. 1934 */ 1935CTASSERT(O_NONBLOCK == IO_NDELAY); 1936CTASSERT(O_FSYNC == IO_SYNC); 1937