devfs_vnops.c revision 327060
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/10/sys/fs/devfs/devfs_vnops.c 327060 2017-12-21 13:32:49Z 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/mount.h> 55#include <sys/namei.h> 56#include <sys/priv.h> 57#include <sys/proc.h> 58#include <sys/stat.h> 59#include <sys/sx.h> 60#include <sys/sysctl.h> 61#include <sys/time.h> 62#include <sys/ttycom.h> 63#include <sys/unistd.h> 64#include <sys/vnode.h> 65 66static struct vop_vector devfs_vnodeops; 67static struct vop_vector devfs_specops; 68static struct fileops devfs_ops_f; 69 70#include <fs/devfs/devfs.h> 71#include <fs/devfs/devfs_int.h> 72 73#include <security/mac/mac_framework.h> 74 75static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data"); 76 77struct mtx devfs_de_interlock; 78MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); 79struct sx clone_drain_lock; 80SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock"); 81struct mtx cdevpriv_mtx; 82MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF); 83 84SYSCTL_DECL(_vfs_devfs); 85 86static int devfs_dotimes; 87SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW, 88 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision"); 89 90/* 91 * Update devfs node timestamp. Note that updates are unlocked and 92 * stat(2) could see partially updated times. 93 */ 94static void 95devfs_timestamp(struct timespec *tsp) 96{ 97 time_t ts; 98 99 if (devfs_dotimes) { 100 vfs_timestamp(tsp); 101 } else { 102 ts = time_second; 103 if (tsp->tv_sec != ts) { 104 tsp->tv_sec = ts; 105 tsp->tv_nsec = 0; 106 } 107 } 108} 109 110static int 111devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp, 112 int *ref) 113{ 114 115 *dswp = devvn_refthread(fp->f_vnode, devp, ref); 116 if (*devp != fp->f_data) { 117 if (*dswp != NULL) 118 dev_relthread(*devp, *ref); 119 return (ENXIO); 120 } 121 KASSERT((*devp)->si_refcount > 0, 122 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); 123 if (*dswp == NULL) 124 return (ENXIO); 125 curthread->td_fpop = fp; 126 return (0); 127} 128 129int 130devfs_get_cdevpriv(void **datap) 131{ 132 struct file *fp; 133 struct cdev_privdata *p; 134 int error; 135 136 fp = curthread->td_fpop; 137 if (fp == NULL) 138 return (EBADF); 139 p = fp->f_cdevpriv; 140 if (p != NULL) { 141 error = 0; 142 *datap = p->cdpd_data; 143 } else 144 error = ENOENT; 145 return (error); 146} 147 148int 149devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr) 150{ 151 struct file *fp; 152 struct cdev_priv *cdp; 153 struct cdev_privdata *p; 154 int error; 155 156 fp = curthread->td_fpop; 157 if (fp == NULL) 158 return (ENOENT); 159 cdp = cdev2priv((struct cdev *)fp->f_data); 160 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK); 161 p->cdpd_data = priv; 162 p->cdpd_dtr = priv_dtr; 163 p->cdpd_fp = fp; 164 mtx_lock(&cdevpriv_mtx); 165 if (fp->f_cdevpriv == NULL) { 166 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list); 167 fp->f_cdevpriv = p; 168 mtx_unlock(&cdevpriv_mtx); 169 error = 0; 170 } else { 171 mtx_unlock(&cdevpriv_mtx); 172 free(p, M_CDEVPDATA); 173 error = EBUSY; 174 } 175 return (error); 176} 177 178void 179devfs_destroy_cdevpriv(struct cdev_privdata *p) 180{ 181 182 mtx_assert(&cdevpriv_mtx, MA_OWNED); 183 p->cdpd_fp->f_cdevpriv = NULL; 184 LIST_REMOVE(p, cdpd_list); 185 mtx_unlock(&cdevpriv_mtx); 186 (p->cdpd_dtr)(p->cdpd_data); 187 free(p, M_CDEVPDATA); 188} 189 190void 191devfs_fpdrop(struct file *fp) 192{ 193 struct cdev_privdata *p; 194 195 mtx_lock(&cdevpriv_mtx); 196 if ((p = fp->f_cdevpriv) == NULL) { 197 mtx_unlock(&cdevpriv_mtx); 198 return; 199 } 200 devfs_destroy_cdevpriv(p); 201} 202 203void 204devfs_clear_cdevpriv(void) 205{ 206 struct file *fp; 207 208 fp = curthread->td_fpop; 209 if (fp == NULL) 210 return; 211 devfs_fpdrop(fp); 212} 213 214/* 215 * On success devfs_populate_vp() returns with dmp->dm_lock held. 216 */ 217static int 218devfs_populate_vp(struct vnode *vp) 219{ 220 struct devfs_dirent *de; 221 struct devfs_mount *dmp; 222 int locked; 223 224 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp"); 225 226 dmp = VFSTODEVFS(vp->v_mount); 227 locked = VOP_ISLOCKED(vp); 228 229 sx_xlock(&dmp->dm_lock); 230 DEVFS_DMP_HOLD(dmp); 231 232 /* Can't call devfs_populate() with the vnode lock held. */ 233 VOP_UNLOCK(vp, 0); 234 devfs_populate(dmp); 235 236 sx_xunlock(&dmp->dm_lock); 237 vn_lock(vp, locked | LK_RETRY); 238 sx_xlock(&dmp->dm_lock); 239 if (DEVFS_DMP_DROP(dmp)) { 240 sx_xunlock(&dmp->dm_lock); 241 devfs_unmount_final(dmp); 242 return (ERESTART); 243 } 244 if ((vp->v_iflag & VI_DOOMED) != 0) { 245 sx_xunlock(&dmp->dm_lock); 246 return (ERESTART); 247 } 248 de = vp->v_data; 249 KASSERT(de != NULL, 250 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed")); 251 if ((de->de_flags & DE_DOOMED) != 0) { 252 sx_xunlock(&dmp->dm_lock); 253 return (ERESTART); 254 } 255 256 return (0); 257} 258 259static int 260devfs_vptocnp(struct vop_vptocnp_args *ap) 261{ 262 struct vnode *vp = ap->a_vp; 263 struct vnode **dvp = ap->a_vpp; 264 struct devfs_mount *dmp; 265 char *buf = ap->a_buf; 266 int *buflen = ap->a_buflen; 267 struct devfs_dirent *dd, *de; 268 int i, error; 269 270 dmp = VFSTODEVFS(vp->v_mount); 271 272 error = devfs_populate_vp(vp); 273 if (error != 0) 274 return (error); 275 276 i = *buflen; 277 dd = vp->v_data; 278 279 if (vp->v_type == VCHR) { 280 i -= strlen(dd->de_cdp->cdp_c.si_name); 281 if (i < 0) { 282 error = ENOMEM; 283 goto finished; 284 } 285 bcopy(dd->de_cdp->cdp_c.si_name, buf + i, 286 strlen(dd->de_cdp->cdp_c.si_name)); 287 de = dd->de_dir; 288 } else if (vp->v_type == VDIR) { 289 if (dd == dmp->dm_rootdir) { 290 *dvp = vp; 291 vref(*dvp); 292 goto finished; 293 } 294 i -= dd->de_dirent->d_namlen; 295 if (i < 0) { 296 error = ENOMEM; 297 goto finished; 298 } 299 bcopy(dd->de_dirent->d_name, buf + i, 300 dd->de_dirent->d_namlen); 301 de = dd; 302 } else { 303 error = ENOENT; 304 goto finished; 305 } 306 *buflen = i; 307 de = devfs_parent_dirent(de); 308 if (de == NULL) { 309 error = ENOENT; 310 goto finished; 311 } 312 mtx_lock(&devfs_de_interlock); 313 *dvp = de->de_vnode; 314 if (*dvp != NULL) { 315 VI_LOCK(*dvp); 316 mtx_unlock(&devfs_de_interlock); 317 vholdl(*dvp); 318 VI_UNLOCK(*dvp); 319 vref(*dvp); 320 vdrop(*dvp); 321 } else { 322 mtx_unlock(&devfs_de_interlock); 323 error = ENOENT; 324 } 325finished: 326 sx_xunlock(&dmp->dm_lock); 327 return (error); 328} 329 330/* 331 * Construct the fully qualified path name relative to the mountpoint. 332 * If a NULL cnp is provided, no '/' is appended to the resulting path. 333 */ 334char * 335devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd, 336 struct componentname *cnp) 337{ 338 int i; 339 struct devfs_dirent *de; 340 341 sx_assert(&dmp->dm_lock, SA_LOCKED); 342 343 i = SPECNAMELEN; 344 buf[i] = '\0'; 345 if (cnp != NULL) 346 i -= cnp->cn_namelen; 347 if (i < 0) 348 return (NULL); 349 if (cnp != NULL) 350 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); 351 de = dd; 352 while (de != dmp->dm_rootdir) { 353 if (cnp != NULL || i < SPECNAMELEN) { 354 i--; 355 if (i < 0) 356 return (NULL); 357 buf[i] = '/'; 358 } 359 i -= de->de_dirent->d_namlen; 360 if (i < 0) 361 return (NULL); 362 bcopy(de->de_dirent->d_name, buf + i, 363 de->de_dirent->d_namlen); 364 de = devfs_parent_dirent(de); 365 if (de == NULL) 366 return (NULL); 367 } 368 return (buf + i); 369} 370 371static int 372devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp, 373 struct devfs_dirent *de) 374{ 375 int not_found; 376 377 not_found = 0; 378 if (de->de_flags & DE_DOOMED) 379 not_found = 1; 380 if (DEVFS_DE_DROP(de)) { 381 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed")); 382 devfs_dirent_free(de); 383 } 384 if (DEVFS_DMP_DROP(dmp)) { 385 KASSERT(not_found == 1, 386 ("DEVFS mount struct freed before dirent")); 387 not_found = 2; 388 sx_xunlock(&dmp->dm_lock); 389 devfs_unmount_final(dmp); 390 } 391 if (not_found == 1 || (drop_dm_lock && not_found != 2)) 392 sx_unlock(&dmp->dm_lock); 393 return (not_found); 394} 395 396static void 397devfs_insmntque_dtr(struct vnode *vp, void *arg) 398{ 399 struct devfs_dirent *de; 400 401 de = (struct devfs_dirent *)arg; 402 mtx_lock(&devfs_de_interlock); 403 vp->v_data = NULL; 404 de->de_vnode = NULL; 405 mtx_unlock(&devfs_de_interlock); 406 vgone(vp); 407 vput(vp); 408} 409 410/* 411 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops 412 * it on return. 413 */ 414int 415devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode, 416 struct vnode **vpp) 417{ 418 int error; 419 struct vnode *vp; 420 struct cdev *dev; 421 struct devfs_mount *dmp; 422 struct cdevsw *dsw; 423 424 dmp = VFSTODEVFS(mp); 425 if (de->de_flags & DE_DOOMED) { 426 sx_xunlock(&dmp->dm_lock); 427 return (ENOENT); 428 } 429loop: 430 DEVFS_DE_HOLD(de); 431 DEVFS_DMP_HOLD(dmp); 432 mtx_lock(&devfs_de_interlock); 433 vp = de->de_vnode; 434 if (vp != NULL) { 435 VI_LOCK(vp); 436 mtx_unlock(&devfs_de_interlock); 437 sx_xunlock(&dmp->dm_lock); 438 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread); 439 sx_xlock(&dmp->dm_lock); 440 if (devfs_allocv_drop_refs(0, dmp, de)) { 441 vput(vp); 442 return (ENOENT); 443 } 444 else if ((vp->v_iflag & VI_DOOMED) != 0) { 445 mtx_lock(&devfs_de_interlock); 446 if (de->de_vnode == vp) { 447 de->de_vnode = NULL; 448 vp->v_data = NULL; 449 } 450 mtx_unlock(&devfs_de_interlock); 451 vput(vp); 452 goto loop; 453 } 454 sx_xunlock(&dmp->dm_lock); 455 *vpp = vp; 456 return (0); 457 } 458 mtx_unlock(&devfs_de_interlock); 459 if (de->de_dirent->d_type == DT_CHR) { 460 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) { 461 devfs_allocv_drop_refs(1, dmp, de); 462 return (ENOENT); 463 } 464 dev = &de->de_cdp->cdp_c; 465 } else { 466 dev = NULL; 467 } 468 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); 469 if (error != 0) { 470 devfs_allocv_drop_refs(1, dmp, de); 471 printf("devfs_allocv: failed to allocate new vnode\n"); 472 return (error); 473 } 474 475 if (de->de_dirent->d_type == DT_CHR) { 476 vp->v_type = VCHR; 477 VI_LOCK(vp); 478 dev_lock(); 479 dev_refl(dev); 480 /* XXX: v_rdev should be protect by vnode lock */ 481 vp->v_rdev = dev; 482 KASSERT(vp->v_usecount == 1, 483 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount)); 484 dev->si_usecount += vp->v_usecount; 485 /* Special casing of ttys for deadfs. Probably redundant. */ 486 dsw = dev->si_devsw; 487 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) 488 vp->v_vflag |= VV_ISTTY; 489 dev_unlock(); 490 VI_UNLOCK(vp); 491 if ((dev->si_flags & SI_ETERNAL) != 0) 492 vp->v_vflag |= VV_ETERNALDEV; 493 vp->v_op = &devfs_specops; 494 } else if (de->de_dirent->d_type == DT_DIR) { 495 vp->v_type = VDIR; 496 } else if (de->de_dirent->d_type == DT_LNK) { 497 vp->v_type = VLNK; 498 } else { 499 vp->v_type = VBAD; 500 } 501 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); 502 VN_LOCK_ASHARE(vp); 503 mtx_lock(&devfs_de_interlock); 504 vp->v_data = de; 505 de->de_vnode = vp; 506 mtx_unlock(&devfs_de_interlock); 507 error = insmntque1(vp, mp, devfs_insmntque_dtr, de); 508 if (error != 0) { 509 (void) devfs_allocv_drop_refs(1, dmp, de); 510 return (error); 511 } 512 if (devfs_allocv_drop_refs(0, dmp, de)) { 513 vput(vp); 514 return (ENOENT); 515 } 516#ifdef MAC 517 mac_devfs_vnode_associate(mp, de, vp); 518#endif 519 sx_xunlock(&dmp->dm_lock); 520 *vpp = vp; 521 return (0); 522} 523 524static int 525devfs_access(struct vop_access_args *ap) 526{ 527 struct vnode *vp = ap->a_vp; 528 struct devfs_dirent *de; 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 /* We do, however, allow access to the controlling terminal */ 542 if (!(ap->a_td->td_proc->p_flag & P_CONTROLT)) 543 return (error); 544 if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp) 545 return (0); 546 return (error); 547} 548 549/* ARGSUSED */ 550static int 551devfs_close(struct vop_close_args *ap) 552{ 553 struct vnode *vp = ap->a_vp, *oldvp; 554 struct thread *td = ap->a_td; 555 struct cdev *dev = vp->v_rdev; 556 struct cdevsw *dsw; 557 int vp_locked, error, ref; 558 559 /* 560 * XXX: Don't call d_close() if we were called because of 561 * XXX: insmntque1() failure. 562 */ 563 if (vp->v_data == NULL) 564 return (0); 565 566 /* 567 * Hack: a tty device that is a controlling terminal 568 * has a reference from the session structure. 569 * We cannot easily tell that a character device is 570 * a controlling terminal, unless it is the closing 571 * process' controlling terminal. In that case, 572 * if the reference count is 2 (this last descriptor 573 * plus the session), release the reference from the session. 574 */ 575 oldvp = NULL; 576 sx_xlock(&proctree_lock); 577 if (td && vp == td->td_proc->p_session->s_ttyvp) { 578 SESS_LOCK(td->td_proc->p_session); 579 VI_LOCK(vp); 580 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) { 581 td->td_proc->p_session->s_ttyvp = NULL; 582 td->td_proc->p_session->s_ttydp = NULL; 583 oldvp = vp; 584 } 585 VI_UNLOCK(vp); 586 SESS_UNLOCK(td->td_proc->p_session); 587 } 588 sx_xunlock(&proctree_lock); 589 if (oldvp != NULL) 590 vrele(oldvp); 591 /* 592 * We do not want to really close the device if it 593 * is still in use unless we are trying to close it 594 * forcibly. Since every use (buffer, vnode, swap, cmap) 595 * holds a reference to the vnode, and because we mark 596 * any other vnodes that alias this device, when the 597 * sum of the reference counts on all the aliased 598 * vnodes descends to one, we are on last close. 599 */ 600 dsw = dev_refthread(dev, &ref); 601 if (dsw == NULL) 602 return (ENXIO); 603 VI_LOCK(vp); 604 if (vp->v_iflag & VI_DOOMED) { 605 /* Forced close. */ 606 } else if (dsw->d_flags & D_TRACKCLOSE) { 607 /* Keep device updated on status. */ 608 } else if (count_dev(dev) > 1) { 609 VI_UNLOCK(vp); 610 dev_relthread(dev, ref); 611 return (0); 612 } 613 vholdl(vp); 614 VI_UNLOCK(vp); 615 vp_locked = VOP_ISLOCKED(vp); 616 VOP_UNLOCK(vp, 0); 617 KASSERT(dev->si_refcount > 0, 618 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 619 error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td); 620 dev_relthread(dev, ref); 621 vn_lock(vp, vp_locked | LK_RETRY); 622 vdrop(vp); 623 return (error); 624} 625 626static int 627devfs_close_f(struct file *fp, struct thread *td) 628{ 629 int error; 630 struct file *fpop; 631 632 /* 633 * NB: td may be NULL if this descriptor is closed due to 634 * garbage collection from a closed UNIX domain socket. 635 */ 636 fpop = curthread->td_fpop; 637 curthread->td_fpop = fp; 638 error = vnops.fo_close(fp, td); 639 curthread->td_fpop = fpop; 640 641 /* 642 * The f_cdevpriv cannot be assigned non-NULL value while we 643 * are destroying the file. 644 */ 645 if (fp->f_cdevpriv != NULL) 646 devfs_fpdrop(fp); 647 return (error); 648} 649 650static int 651devfs_fsync(struct vop_fsync_args *ap) 652{ 653 int error; 654 struct bufobj *bo; 655 struct devfs_dirent *de; 656 657 if (!vn_isdisk(ap->a_vp, &error)) { 658 bo = &ap->a_vp->v_bufobj; 659 de = ap->a_vp->v_data; 660 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) { 661 printf("Device %s went missing before all of the data " 662 "could be written to it; expect data loss.\n", 663 de->de_dirent->d_name); 664 665 error = vop_stdfsync(ap); 666 if (bo->bo_dirty.bv_cnt != 0 || error != 0) 667 panic("devfs_fsync: vop_stdfsync failed."); 668 } 669 670 return (0); 671 } 672 673 return (vop_stdfsync(ap)); 674} 675 676static int 677devfs_getattr(struct vop_getattr_args *ap) 678{ 679 struct vnode *vp = ap->a_vp; 680 struct vattr *vap = ap->a_vap; 681 int error; 682 struct devfs_dirent *de; 683 struct devfs_mount *dmp; 684 struct cdev *dev; 685 686 error = devfs_populate_vp(vp); 687 if (error != 0) 688 return (error); 689 690 dmp = VFSTODEVFS(vp->v_mount); 691 sx_xunlock(&dmp->dm_lock); 692 693 de = vp->v_data; 694 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 695 if (vp->v_type == VDIR) { 696 de = de->de_dir; 697 KASSERT(de != NULL, 698 ("Null dir dirent in devfs_getattr vp=%p", vp)); 699 } 700 vap->va_uid = de->de_uid; 701 vap->va_gid = de->de_gid; 702 vap->va_mode = de->de_mode; 703 if (vp->v_type == VLNK) 704 vap->va_size = strlen(de->de_symlink); 705 else if (vp->v_type == VDIR) 706 vap->va_size = vap->va_bytes = DEV_BSIZE; 707 else 708 vap->va_size = 0; 709 if (vp->v_type != VDIR) 710 vap->va_bytes = 0; 711 vap->va_blocksize = DEV_BSIZE; 712 vap->va_type = vp->v_type; 713 714#define fix(aa) \ 715 do { \ 716 if ((aa).tv_sec <= 3600) { \ 717 (aa).tv_sec = boottime.tv_sec; \ 718 (aa).tv_nsec = boottime.tv_usec * 1000; \ 719 } \ 720 } while (0) 721 722 if (vp->v_type != VCHR) { 723 fix(de->de_atime); 724 vap->va_atime = de->de_atime; 725 fix(de->de_mtime); 726 vap->va_mtime = de->de_mtime; 727 fix(de->de_ctime); 728 vap->va_ctime = de->de_ctime; 729 } else { 730 dev = vp->v_rdev; 731 fix(dev->si_atime); 732 vap->va_atime = dev->si_atime; 733 fix(dev->si_mtime); 734 vap->va_mtime = dev->si_mtime; 735 fix(dev->si_ctime); 736 vap->va_ctime = dev->si_ctime; 737 738 vap->va_rdev = cdev2priv(dev)->cdp_inode; 739 } 740 vap->va_gen = 0; 741 vap->va_flags = 0; 742 vap->va_filerev = 0; 743 vap->va_nlink = de->de_links; 744 vap->va_fileid = de->de_inode; 745 746 return (error); 747} 748 749/* ARGSUSED */ 750static int 751devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 752{ 753 struct cdev *dev; 754 struct cdevsw *dsw; 755 struct vnode *vp; 756 struct vnode *vpold; 757 int error, i, ref; 758 const char *p; 759 struct fiodgname_arg *fgn; 760 struct file *fpop; 761 762 fpop = td->td_fpop; 763 error = devfs_fp_check(fp, &dev, &dsw, &ref); 764 if (error != 0) { 765 error = vnops.fo_ioctl(fp, com, data, cred, td); 766 return (error); 767 } 768 769 if (com == FIODTYPE) { 770 *(int *)data = dsw->d_flags & D_TYPEMASK; 771 td->td_fpop = fpop; 772 dev_relthread(dev, ref); 773 return (0); 774 } else if (com == FIODGNAME) { 775 fgn = data; 776 p = devtoname(dev); 777 i = strlen(p) + 1; 778 if (i > fgn->len) 779 error = EINVAL; 780 else 781 error = copyout(p, fgn->buf, i); 782 td->td_fpop = fpop; 783 dev_relthread(dev, ref); 784 return (error); 785 } 786 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td); 787 td->td_fpop = NULL; 788 dev_relthread(dev, ref); 789 if (error == ENOIOCTL) 790 error = ENOTTY; 791 if (error == 0 && com == TIOCSCTTY) { 792 vp = fp->f_vnode; 793 794 /* Do nothing if reassigning same control tty */ 795 sx_slock(&proctree_lock); 796 if (td->td_proc->p_session->s_ttyvp == vp) { 797 sx_sunlock(&proctree_lock); 798 return (0); 799 } 800 801 vpold = td->td_proc->p_session->s_ttyvp; 802 VREF(vp); 803 SESS_LOCK(td->td_proc->p_session); 804 td->td_proc->p_session->s_ttyvp = vp; 805 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 806 SESS_UNLOCK(td->td_proc->p_session); 807 808 sx_sunlock(&proctree_lock); 809 810 /* Get rid of reference to old control tty */ 811 if (vpold) 812 vrele(vpold); 813 } 814 return (error); 815} 816 817/* ARGSUSED */ 818static int 819devfs_kqfilter_f(struct file *fp, struct knote *kn) 820{ 821 struct cdev *dev; 822 struct cdevsw *dsw; 823 int error, ref; 824 struct file *fpop; 825 struct thread *td; 826 827 td = curthread; 828 fpop = td->td_fpop; 829 error = devfs_fp_check(fp, &dev, &dsw, &ref); 830 if (error) 831 return (error); 832 error = dsw->d_kqfilter(dev, kn); 833 td->td_fpop = fpop; 834 dev_relthread(dev, ref); 835 return (error); 836} 837 838static inline int 839devfs_prison_check(struct devfs_dirent *de, struct thread *td) 840{ 841 struct cdev_priv *cdp; 842 struct ucred *dcr; 843 int error; 844 845 cdp = de->de_cdp; 846 if (cdp == NULL) 847 return (0); 848 dcr = cdp->cdp_c.si_cred; 849 if (dcr == NULL) 850 return (0); 851 852 error = prison_check(td->td_ucred, dcr); 853 if (error == 0) 854 return (0); 855 /* We do, however, allow access to the controlling terminal */ 856 if (!(td->td_proc->p_flag & P_CONTROLT)) 857 return (error); 858 if (td->td_proc->p_session->s_ttydp == cdp) 859 return (0); 860 return (error); 861} 862 863static int 864devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 865{ 866 struct componentname *cnp; 867 struct vnode *dvp, **vpp; 868 struct thread *td; 869 struct devfs_dirent *de, *dd; 870 struct devfs_dirent **dde; 871 struct devfs_mount *dmp; 872 struct mount *mp; 873 struct cdev *cdev; 874 int error, flags, nameiop, dvplocked; 875 char specname[SPECNAMELEN + 1], *pname; 876 877 cnp = ap->a_cnp; 878 vpp = ap->a_vpp; 879 dvp = ap->a_dvp; 880 pname = cnp->cn_nameptr; 881 td = cnp->cn_thread; 882 flags = cnp->cn_flags; 883 nameiop = cnp->cn_nameiop; 884 mp = dvp->v_mount; 885 dmp = VFSTODEVFS(mp); 886 dd = dvp->v_data; 887 *vpp = NULLVP; 888 889 if ((flags & ISLASTCN) && nameiop == RENAME) 890 return (EOPNOTSUPP); 891 892 if (dvp->v_type != VDIR) 893 return (ENOTDIR); 894 895 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 896 return (EIO); 897 898 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 899 if (error) 900 return (error); 901 902 if (cnp->cn_namelen == 1 && *pname == '.') { 903 if ((flags & ISLASTCN) && nameiop != LOOKUP) 904 return (EINVAL); 905 *vpp = dvp; 906 VREF(dvp); 907 return (0); 908 } 909 910 if (flags & ISDOTDOT) { 911 if ((flags & ISLASTCN) && nameiop != LOOKUP) 912 return (EINVAL); 913 de = devfs_parent_dirent(dd); 914 if (de == NULL) 915 return (ENOENT); 916 dvplocked = VOP_ISLOCKED(dvp); 917 VOP_UNLOCK(dvp, 0); 918 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, 919 vpp); 920 *dm_unlock = 0; 921 vn_lock(dvp, dvplocked | LK_RETRY); 922 return (error); 923 } 924 925 dd = dvp->v_data; 926 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 927 while (de == NULL) { /* While(...) so we can use break */ 928 929 if (nameiop == DELETE) 930 return (ENOENT); 931 932 /* 933 * OK, we didn't have an entry for the name we were asked for 934 * so we try to see if anybody can create it on demand. 935 */ 936 pname = devfs_fqpn(specname, dmp, dd, cnp); 937 if (pname == NULL) 938 break; 939 940 cdev = NULL; 941 DEVFS_DMP_HOLD(dmp); 942 sx_xunlock(&dmp->dm_lock); 943 sx_slock(&clone_drain_lock); 944 EVENTHANDLER_INVOKE(dev_clone, 945 td->td_ucred, pname, strlen(pname), &cdev); 946 sx_sunlock(&clone_drain_lock); 947 948 if (cdev == NULL) 949 sx_xlock(&dmp->dm_lock); 950 else if (devfs_populate_vp(dvp) != 0) { 951 *dm_unlock = 0; 952 sx_xlock(&dmp->dm_lock); 953 if (DEVFS_DMP_DROP(dmp)) { 954 sx_xunlock(&dmp->dm_lock); 955 devfs_unmount_final(dmp); 956 } else 957 sx_xunlock(&dmp->dm_lock); 958 dev_rel(cdev); 959 return (ENOENT); 960 } 961 if (DEVFS_DMP_DROP(dmp)) { 962 *dm_unlock = 0; 963 sx_xunlock(&dmp->dm_lock); 964 devfs_unmount_final(dmp); 965 if (cdev != NULL) 966 dev_rel(cdev); 967 return (ENOENT); 968 } 969 970 if (cdev == NULL) 971 break; 972 973 dev_lock(); 974 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 975 if (dde != NULL && *dde != NULL) 976 de = *dde; 977 dev_unlock(); 978 dev_rel(cdev); 979 break; 980 } 981 982 if (de == NULL || de->de_flags & DE_WHITEOUT) { 983 if ((nameiop == CREATE || nameiop == RENAME) && 984 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 985 cnp->cn_flags |= SAVENAME; 986 return (EJUSTRETURN); 987 } 988 return (ENOENT); 989 } 990 991 if (devfs_prison_check(de, td)) 992 return (ENOENT); 993 994 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 995 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 996 if (error) 997 return (error); 998 if (*vpp == dvp) { 999 VREF(dvp); 1000 *vpp = dvp; 1001 return (0); 1002 } 1003 } 1004 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp); 1005 *dm_unlock = 0; 1006 return (error); 1007} 1008 1009static int 1010devfs_lookup(struct vop_lookup_args *ap) 1011{ 1012 int j; 1013 struct devfs_mount *dmp; 1014 int dm_unlock; 1015 1016 if (devfs_populate_vp(ap->a_dvp) != 0) 1017 return (ENOTDIR); 1018 1019 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1020 dm_unlock = 1; 1021 j = devfs_lookupx(ap, &dm_unlock); 1022 if (dm_unlock == 1) 1023 sx_xunlock(&dmp->dm_lock); 1024 return (j); 1025} 1026 1027static int 1028devfs_mknod(struct vop_mknod_args *ap) 1029{ 1030 struct componentname *cnp; 1031 struct vnode *dvp, **vpp; 1032 struct devfs_dirent *dd, *de; 1033 struct devfs_mount *dmp; 1034 int error; 1035 1036 /* 1037 * The only type of node we should be creating here is a 1038 * character device, for anything else return EOPNOTSUPP. 1039 */ 1040 if (ap->a_vap->va_type != VCHR) 1041 return (EOPNOTSUPP); 1042 dvp = ap->a_dvp; 1043 dmp = VFSTODEVFS(dvp->v_mount); 1044 1045 cnp = ap->a_cnp; 1046 vpp = ap->a_vpp; 1047 dd = dvp->v_data; 1048 1049 error = ENOENT; 1050 sx_xlock(&dmp->dm_lock); 1051 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1052 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1053 continue; 1054 if (de->de_dirent->d_type == DT_CHR && 1055 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1056 continue; 1057 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1058 de->de_dirent->d_namlen) != 0) 1059 continue; 1060 if (de->de_flags & DE_WHITEOUT) 1061 break; 1062 goto notfound; 1063 } 1064 if (de == NULL) 1065 goto notfound; 1066 de->de_flags &= ~DE_WHITEOUT; 1067 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1068 return (error); 1069notfound: 1070 sx_xunlock(&dmp->dm_lock); 1071 return (error); 1072} 1073 1074/* ARGSUSED */ 1075static int 1076devfs_open(struct vop_open_args *ap) 1077{ 1078 struct thread *td = ap->a_td; 1079 struct vnode *vp = ap->a_vp; 1080 struct cdev *dev = vp->v_rdev; 1081 struct file *fp = ap->a_fp; 1082 int error, ref, vlocked; 1083 struct cdevsw *dsw; 1084 struct file *fpop; 1085 struct mtx *mtxp; 1086 1087 if (vp->v_type == VBLK) 1088 return (ENXIO); 1089 1090 if (dev == NULL) 1091 return (ENXIO); 1092 1093 /* Make this field valid before any I/O in d_open. */ 1094 if (dev->si_iosize_max == 0) 1095 dev->si_iosize_max = DFLTPHYS; 1096 1097 dsw = dev_refthread(dev, &ref); 1098 if (dsw == NULL) 1099 return (ENXIO); 1100 if (fp == NULL && dsw->d_fdopen != NULL) { 1101 dev_relthread(dev, ref); 1102 return (ENXIO); 1103 } 1104 1105 vlocked = VOP_ISLOCKED(vp); 1106 VOP_UNLOCK(vp, 0); 1107 1108 fpop = td->td_fpop; 1109 td->td_fpop = fp; 1110 if (fp != NULL) { 1111 fp->f_data = dev; 1112 fp->f_vnode = vp; 1113 } 1114 if (dsw->d_fdopen != NULL) 1115 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1116 else 1117 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1118 /* Clean up any cdevpriv upon error. */ 1119 if (error != 0) 1120 devfs_clear_cdevpriv(); 1121 td->td_fpop = fpop; 1122 1123 vn_lock(vp, vlocked | LK_RETRY); 1124 dev_relthread(dev, ref); 1125 if (error != 0) { 1126 if (error == ERESTART) 1127 error = EINTR; 1128 return (error); 1129 } 1130 1131#if 0 /* /dev/console */ 1132 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1133#else 1134 if (fp == NULL) 1135 return (error); 1136#endif 1137 if (fp->f_ops == &badfileops) 1138 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1139 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1140 1141 /* 1142 * Hint to the dofilewrite() to not force the buffer draining 1143 * on the writer to the file. Most likely, the write would 1144 * not need normal buffers. 1145 */ 1146 mtx_lock(mtxp); 1147 fp->f_vnread_flags |= FDEVFS_VNODE; 1148 mtx_unlock(mtxp); 1149 return (error); 1150} 1151 1152static int 1153devfs_pathconf(struct vop_pathconf_args *ap) 1154{ 1155 1156 switch (ap->a_name) { 1157 case _PC_MAC_PRESENT: 1158#ifdef MAC 1159 /* 1160 * If MAC is enabled, devfs automatically supports 1161 * trivial non-persistant label storage. 1162 */ 1163 *ap->a_retval = 1; 1164#else 1165 *ap->a_retval = 0; 1166#endif 1167 return (0); 1168 default: 1169 return (vop_stdpathconf(ap)); 1170 } 1171 /* NOTREACHED */ 1172} 1173 1174/* ARGSUSED */ 1175static int 1176devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1177{ 1178 struct cdev *dev; 1179 struct cdevsw *dsw; 1180 int error, ref; 1181 struct file *fpop; 1182 1183 fpop = td->td_fpop; 1184 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1185 if (error != 0) { 1186 error = vnops.fo_poll(fp, events, cred, td); 1187 return (error); 1188 } 1189 error = dsw->d_poll(dev, events, td); 1190 td->td_fpop = fpop; 1191 dev_relthread(dev, ref); 1192 return(error); 1193} 1194 1195/* 1196 * Print out the contents of a special device vnode. 1197 */ 1198static int 1199devfs_print(struct vop_print_args *ap) 1200{ 1201 1202 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1203 return (0); 1204} 1205 1206static int 1207devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1208 int flags, struct thread *td) 1209{ 1210 struct cdev *dev; 1211 int ioflag, error, ref; 1212 ssize_t resid; 1213 struct cdevsw *dsw; 1214 struct file *fpop; 1215 1216 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1217 return (EINVAL); 1218 fpop = td->td_fpop; 1219 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1220 if (error != 0) { 1221 error = vnops.fo_read(fp, uio, cred, flags, td); 1222 return (error); 1223 } 1224 resid = uio->uio_resid; 1225 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1226 if (ioflag & O_DIRECT) 1227 ioflag |= IO_DIRECT; 1228 1229 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1230 error = dsw->d_read(dev, uio, ioflag); 1231 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1232 devfs_timestamp(&dev->si_atime); 1233 td->td_fpop = fpop; 1234 dev_relthread(dev, ref); 1235 1236 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1237 return (error); 1238} 1239 1240static int 1241devfs_readdir(struct vop_readdir_args *ap) 1242{ 1243 int error; 1244 struct uio *uio; 1245 struct dirent *dp; 1246 struct devfs_dirent *dd; 1247 struct devfs_dirent *de; 1248 struct devfs_mount *dmp; 1249 off_t off; 1250 int *tmp_ncookies = NULL; 1251 1252 if (ap->a_vp->v_type != VDIR) 1253 return (ENOTDIR); 1254 1255 uio = ap->a_uio; 1256 if (uio->uio_offset < 0) 1257 return (EINVAL); 1258 1259 /* 1260 * XXX: This is a temporary hack to get around this filesystem not 1261 * supporting cookies. We store the location of the ncookies pointer 1262 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1263 * and set the number of cookies to 0. We then set the pointer to 1264 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1265 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1266 * pointer to its original location before returning to the caller. 1267 */ 1268 if (ap->a_ncookies != NULL) { 1269 tmp_ncookies = ap->a_ncookies; 1270 *ap->a_ncookies = 0; 1271 ap->a_ncookies = NULL; 1272 } 1273 1274 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1275 if (devfs_populate_vp(ap->a_vp) != 0) { 1276 if (tmp_ncookies != NULL) 1277 ap->a_ncookies = tmp_ncookies; 1278 return (EIO); 1279 } 1280 error = 0; 1281 de = ap->a_vp->v_data; 1282 off = 0; 1283 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1284 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1285 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1286 continue; 1287 if (devfs_prison_check(dd, uio->uio_td)) 1288 continue; 1289 if (dd->de_dirent->d_type == DT_DIR) 1290 de = dd->de_dir; 1291 else 1292 de = dd; 1293 dp = dd->de_dirent; 1294 if (dp->d_reclen > uio->uio_resid) 1295 break; 1296 dp->d_fileno = de->de_inode; 1297 if (off >= uio->uio_offset) { 1298 error = vfs_read_dirent(ap, dp, off); 1299 if (error) 1300 break; 1301 } 1302 off += dp->d_reclen; 1303 } 1304 sx_xunlock(&dmp->dm_lock); 1305 uio->uio_offset = off; 1306 1307 /* 1308 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1309 * place. 1310 */ 1311 if (tmp_ncookies != NULL) 1312 ap->a_ncookies = tmp_ncookies; 1313 1314 return (error); 1315} 1316 1317static int 1318devfs_readlink(struct vop_readlink_args *ap) 1319{ 1320 struct devfs_dirent *de; 1321 1322 de = ap->a_vp->v_data; 1323 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1324} 1325 1326static int 1327devfs_reclaim(struct vop_reclaim_args *ap) 1328{ 1329 struct vnode *vp; 1330 struct devfs_dirent *de; 1331 1332 vp = ap->a_vp; 1333 mtx_lock(&devfs_de_interlock); 1334 de = vp->v_data; 1335 if (de != NULL) { 1336 de->de_vnode = NULL; 1337 vp->v_data = NULL; 1338 } 1339 mtx_unlock(&devfs_de_interlock); 1340 vnode_destroy_vobject(vp); 1341 return (0); 1342} 1343 1344static int 1345devfs_reclaim_vchr(struct vop_reclaim_args *ap) 1346{ 1347 struct vnode *vp; 1348 struct cdev *dev; 1349 1350 vp = ap->a_vp; 1351 MPASS(vp->v_type == VCHR); 1352 1353 devfs_reclaim(ap); 1354 1355 VI_LOCK(vp); 1356 dev_lock(); 1357 dev = vp->v_rdev; 1358 vp->v_rdev = NULL; 1359 if (dev != NULL) 1360 dev->si_usecount -= vp->v_usecount; 1361 dev_unlock(); 1362 VI_UNLOCK(vp); 1363 if (dev != NULL) 1364 dev_rel(dev); 1365 return (0); 1366} 1367 1368static int 1369devfs_remove(struct vop_remove_args *ap) 1370{ 1371 struct vnode *dvp = ap->a_dvp; 1372 struct vnode *vp = ap->a_vp; 1373 struct devfs_dirent *dd; 1374 struct devfs_dirent *de, *de_covered; 1375 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1376 1377 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1378 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1379 1380 sx_xlock(&dmp->dm_lock); 1381 dd = ap->a_dvp->v_data; 1382 de = vp->v_data; 1383 if (de->de_cdp == NULL) { 1384 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1385 if (de->de_dirent->d_type == DT_LNK) { 1386 de_covered = devfs_find(dd, de->de_dirent->d_name, 1387 de->de_dirent->d_namlen, 0); 1388 if (de_covered != NULL) 1389 de_covered->de_flags &= ~DE_COVERED; 1390 } 1391 /* We need to unlock dvp because devfs_delete() may lock it. */ 1392 VOP_UNLOCK(vp, 0); 1393 if (dvp != vp) 1394 VOP_UNLOCK(dvp, 0); 1395 devfs_delete(dmp, de, 0); 1396 sx_xunlock(&dmp->dm_lock); 1397 if (dvp != vp) 1398 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1399 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1400 } else { 1401 de->de_flags |= DE_WHITEOUT; 1402 sx_xunlock(&dmp->dm_lock); 1403 } 1404 return (0); 1405} 1406 1407/* 1408 * Revoke is called on a tty when a terminal session ends. The vnode 1409 * is orphaned by setting v_op to deadfs so we need to let go of it 1410 * as well so that we create a new one next time around. 1411 * 1412 */ 1413static int 1414devfs_revoke(struct vop_revoke_args *ap) 1415{ 1416 struct vnode *vp = ap->a_vp, *vp2; 1417 struct cdev *dev; 1418 struct cdev_priv *cdp; 1419 struct devfs_dirent *de; 1420 u_int i; 1421 1422 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1423 1424 dev = vp->v_rdev; 1425 cdp = cdev2priv(dev); 1426 1427 dev_lock(); 1428 cdp->cdp_inuse++; 1429 dev_unlock(); 1430 1431 vhold(vp); 1432 vgone(vp); 1433 vdrop(vp); 1434 1435 VOP_UNLOCK(vp,0); 1436 loop: 1437 for (;;) { 1438 mtx_lock(&devfs_de_interlock); 1439 dev_lock(); 1440 vp2 = NULL; 1441 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1442 de = cdp->cdp_dirents[i]; 1443 if (de == NULL) 1444 continue; 1445 1446 vp2 = de->de_vnode; 1447 if (vp2 != NULL) { 1448 dev_unlock(); 1449 VI_LOCK(vp2); 1450 mtx_unlock(&devfs_de_interlock); 1451 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1452 curthread)) 1453 goto loop; 1454 vhold(vp2); 1455 vgone(vp2); 1456 vdrop(vp2); 1457 vput(vp2); 1458 break; 1459 } 1460 } 1461 if (vp2 != NULL) { 1462 continue; 1463 } 1464 dev_unlock(); 1465 mtx_unlock(&devfs_de_interlock); 1466 break; 1467 } 1468 dev_lock(); 1469 cdp->cdp_inuse--; 1470 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1471 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1472 dev_unlock(); 1473 dev_rel(&cdp->cdp_c); 1474 } else 1475 dev_unlock(); 1476 1477 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1478 return (0); 1479} 1480 1481static int 1482devfs_rioctl(struct vop_ioctl_args *ap) 1483{ 1484 struct vnode *vp; 1485 struct devfs_mount *dmp; 1486 int error; 1487 1488 vp = ap->a_vp; 1489 vn_lock(vp, LK_SHARED | LK_RETRY); 1490 if (vp->v_iflag & VI_DOOMED) { 1491 VOP_UNLOCK(vp, 0); 1492 return (EBADF); 1493 } 1494 dmp = VFSTODEVFS(vp->v_mount); 1495 sx_xlock(&dmp->dm_lock); 1496 VOP_UNLOCK(vp, 0); 1497 DEVFS_DMP_HOLD(dmp); 1498 devfs_populate(dmp); 1499 if (DEVFS_DMP_DROP(dmp)) { 1500 sx_xunlock(&dmp->dm_lock); 1501 devfs_unmount_final(dmp); 1502 return (ENOENT); 1503 } 1504 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1505 sx_xunlock(&dmp->dm_lock); 1506 return (error); 1507} 1508 1509static int 1510devfs_rread(struct vop_read_args *ap) 1511{ 1512 1513 if (ap->a_vp->v_type != VDIR) 1514 return (EINVAL); 1515 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1516} 1517 1518static int 1519devfs_setattr(struct vop_setattr_args *ap) 1520{ 1521 struct devfs_dirent *de; 1522 struct vattr *vap; 1523 struct vnode *vp; 1524 struct thread *td; 1525 int c, error; 1526 uid_t uid; 1527 gid_t gid; 1528 1529 vap = ap->a_vap; 1530 vp = ap->a_vp; 1531 td = curthread; 1532 if ((vap->va_type != VNON) || 1533 (vap->va_nlink != VNOVAL) || 1534 (vap->va_fsid != VNOVAL) || 1535 (vap->va_fileid != VNOVAL) || 1536 (vap->va_blocksize != VNOVAL) || 1537 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1538 (vap->va_rdev != VNOVAL) || 1539 ((int)vap->va_bytes != VNOVAL) || 1540 (vap->va_gen != VNOVAL)) { 1541 return (EINVAL); 1542 } 1543 1544 error = devfs_populate_vp(vp); 1545 if (error != 0) 1546 return (error); 1547 1548 de = vp->v_data; 1549 if (vp->v_type == VDIR) 1550 de = de->de_dir; 1551 1552 c = 0; 1553 if (vap->va_uid == (uid_t)VNOVAL) 1554 uid = de->de_uid; 1555 else 1556 uid = vap->va_uid; 1557 if (vap->va_gid == (gid_t)VNOVAL) 1558 gid = de->de_gid; 1559 else 1560 gid = vap->va_gid; 1561 if (uid != de->de_uid || gid != de->de_gid) { 1562 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1563 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1564 error = priv_check(td, PRIV_VFS_CHOWN); 1565 if (error != 0) 1566 goto ret; 1567 } 1568 de->de_uid = uid; 1569 de->de_gid = gid; 1570 c = 1; 1571 } 1572 1573 if (vap->va_mode != (mode_t)VNOVAL) { 1574 if (ap->a_cred->cr_uid != de->de_uid) { 1575 error = priv_check(td, PRIV_VFS_ADMIN); 1576 if (error != 0) 1577 goto ret; 1578 } 1579 de->de_mode = vap->va_mode; 1580 c = 1; 1581 } 1582 1583 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1584 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1585 if (error != 0) 1586 goto ret; 1587 if (vap->va_atime.tv_sec != VNOVAL) { 1588 if (vp->v_type == VCHR) 1589 vp->v_rdev->si_atime = vap->va_atime; 1590 else 1591 de->de_atime = vap->va_atime; 1592 } 1593 if (vap->va_mtime.tv_sec != VNOVAL) { 1594 if (vp->v_type == VCHR) 1595 vp->v_rdev->si_mtime = vap->va_mtime; 1596 else 1597 de->de_mtime = vap->va_mtime; 1598 } 1599 c = 1; 1600 } 1601 1602 if (c) { 1603 if (vp->v_type == VCHR) 1604 vfs_timestamp(&vp->v_rdev->si_ctime); 1605 else 1606 vfs_timestamp(&de->de_mtime); 1607 } 1608 1609ret: 1610 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1611 return (error); 1612} 1613 1614#ifdef MAC 1615static int 1616devfs_setlabel(struct vop_setlabel_args *ap) 1617{ 1618 struct vnode *vp; 1619 struct devfs_dirent *de; 1620 1621 vp = ap->a_vp; 1622 de = vp->v_data; 1623 1624 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1625 mac_devfs_update(vp->v_mount, de, vp); 1626 1627 return (0); 1628} 1629#endif 1630 1631static int 1632devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1633{ 1634 1635 return (vnops.fo_stat(fp, sb, cred, td)); 1636} 1637 1638static int 1639devfs_symlink(struct vop_symlink_args *ap) 1640{ 1641 int i, error; 1642 struct devfs_dirent *dd; 1643 struct devfs_dirent *de, *de_covered, *de_dotdot; 1644 struct devfs_mount *dmp; 1645 1646 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1647 if (error) 1648 return(error); 1649 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1650 if (devfs_populate_vp(ap->a_dvp) != 0) 1651 return (ENOENT); 1652 1653 dd = ap->a_dvp->v_data; 1654 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1655 de->de_flags = DE_USER; 1656 de->de_uid = 0; 1657 de->de_gid = 0; 1658 de->de_mode = 0755; 1659 de->de_inode = alloc_unr(devfs_inos); 1660 de->de_dir = dd; 1661 de->de_dirent->d_type = DT_LNK; 1662 i = strlen(ap->a_target) + 1; 1663 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1664 bcopy(ap->a_target, de->de_symlink, i); 1665#ifdef MAC 1666 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1667#endif 1668 de_covered = devfs_find(dd, de->de_dirent->d_name, 1669 de->de_dirent->d_namlen, 0); 1670 if (de_covered != NULL) { 1671 if ((de_covered->de_flags & DE_USER) != 0) { 1672 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1673 sx_xunlock(&dmp->dm_lock); 1674 return (EEXIST); 1675 } 1676 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1677 ("devfs_symlink: entry %p already covered", de_covered)); 1678 de_covered->de_flags |= DE_COVERED; 1679 } 1680 1681 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1682 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1683 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1684 devfs_dir_ref_de(dmp, dd); 1685 devfs_rules_apply(dmp, de); 1686 1687 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1688} 1689 1690static int 1691devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1692{ 1693 1694 return (vnops.fo_truncate(fp, length, cred, td)); 1695} 1696 1697static int 1698devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1699 int flags, struct thread *td) 1700{ 1701 struct cdev *dev; 1702 int error, ioflag, ref; 1703 ssize_t resid; 1704 struct cdevsw *dsw; 1705 struct file *fpop; 1706 1707 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1708 return (EINVAL); 1709 fpop = td->td_fpop; 1710 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1711 if (error != 0) { 1712 error = vnops.fo_write(fp, uio, cred, flags, td); 1713 return (error); 1714 } 1715 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1716 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1717 if (ioflag & O_DIRECT) 1718 ioflag |= IO_DIRECT; 1719 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1720 1721 resid = uio->uio_resid; 1722 1723 error = dsw->d_write(dev, uio, ioflag); 1724 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1725 devfs_timestamp(&dev->si_ctime); 1726 dev->si_mtime = dev->si_ctime; 1727 } 1728 td->td_fpop = fpop; 1729 dev_relthread(dev, ref); 1730 1731 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1732 return (error); 1733} 1734 1735dev_t 1736dev2udev(struct cdev *x) 1737{ 1738 if (x == NULL) 1739 return (NODEV); 1740 return (cdev2priv(x)->cdp_inode); 1741} 1742 1743static struct fileops devfs_ops_f = { 1744 .fo_read = devfs_read_f, 1745 .fo_write = devfs_write_f, 1746 .fo_truncate = devfs_truncate_f, 1747 .fo_ioctl = devfs_ioctl_f, 1748 .fo_poll = devfs_poll_f, 1749 .fo_kqfilter = devfs_kqfilter_f, 1750 .fo_stat = devfs_stat_f, 1751 .fo_close = devfs_close_f, 1752 .fo_chmod = vn_chmod, 1753 .fo_chown = vn_chown, 1754 .fo_sendfile = vn_sendfile, 1755 .fo_seek = vn_seek, 1756 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1757}; 1758 1759static struct vop_vector devfs_vnodeops = { 1760 .vop_default = &default_vnodeops, 1761 1762 .vop_access = devfs_access, 1763 .vop_getattr = devfs_getattr, 1764 .vop_ioctl = devfs_rioctl, 1765 .vop_lookup = devfs_lookup, 1766 .vop_mknod = devfs_mknod, 1767 .vop_pathconf = devfs_pathconf, 1768 .vop_read = devfs_rread, 1769 .vop_readdir = devfs_readdir, 1770 .vop_readlink = devfs_readlink, 1771 .vop_reclaim = devfs_reclaim, 1772 .vop_remove = devfs_remove, 1773 .vop_revoke = devfs_revoke, 1774 .vop_setattr = devfs_setattr, 1775#ifdef MAC 1776 .vop_setlabel = devfs_setlabel, 1777#endif 1778 .vop_symlink = devfs_symlink, 1779 .vop_vptocnp = devfs_vptocnp, 1780}; 1781 1782static struct vop_vector devfs_specops = { 1783 .vop_default = &default_vnodeops, 1784 1785 .vop_access = devfs_access, 1786 .vop_bmap = VOP_PANIC, 1787 .vop_close = devfs_close, 1788 .vop_create = VOP_PANIC, 1789 .vop_fsync = devfs_fsync, 1790 .vop_getattr = devfs_getattr, 1791 .vop_link = VOP_PANIC, 1792 .vop_mkdir = VOP_PANIC, 1793 .vop_mknod = VOP_PANIC, 1794 .vop_open = devfs_open, 1795 .vop_pathconf = devfs_pathconf, 1796 .vop_poll = dead_poll, 1797 .vop_print = devfs_print, 1798 .vop_read = dead_read, 1799 .vop_readdir = VOP_PANIC, 1800 .vop_readlink = VOP_PANIC, 1801 .vop_reallocblks = VOP_PANIC, 1802 .vop_reclaim = devfs_reclaim_vchr, 1803 .vop_remove = devfs_remove, 1804 .vop_rename = VOP_PANIC, 1805 .vop_revoke = devfs_revoke, 1806 .vop_rmdir = VOP_PANIC, 1807 .vop_setattr = devfs_setattr, 1808#ifdef MAC 1809 .vop_setlabel = devfs_setlabel, 1810#endif 1811 .vop_strategy = VOP_PANIC, 1812 .vop_symlink = VOP_PANIC, 1813 .vop_vptocnp = devfs_vptocnp, 1814 .vop_write = dead_write, 1815}; 1816 1817/* 1818 * Our calling convention to the device drivers used to be that we passed 1819 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1820 * flags instead since that's what open(), close() and ioctl() takes and 1821 * we don't really want vnode.h in device drivers. 1822 * We solved the source compatibility by redefining some vnode flags to 1823 * be the same as the fcntl ones and by sending down the bitwise OR of 1824 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1825 * pulls the rug out under this. 1826 */ 1827CTASSERT(O_NONBLOCK == IO_NDELAY); 1828CTASSERT(O_FSYNC == IO_SYNC); 1829