devfs_vnops.c revision 302078
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 302078 2016-06-22 09:08:18Z 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 cdev *cdev; 873 int error, flags, nameiop, dvplocked; 874 char specname[SPECNAMELEN + 1], *pname; 875 876 cnp = ap->a_cnp; 877 vpp = ap->a_vpp; 878 dvp = ap->a_dvp; 879 pname = cnp->cn_nameptr; 880 td = cnp->cn_thread; 881 flags = cnp->cn_flags; 882 nameiop = cnp->cn_nameiop; 883 dmp = VFSTODEVFS(dvp->v_mount); 884 dd = dvp->v_data; 885 *vpp = NULLVP; 886 887 if ((flags & ISLASTCN) && nameiop == RENAME) 888 return (EOPNOTSUPP); 889 890 if (dvp->v_type != VDIR) 891 return (ENOTDIR); 892 893 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 894 return (EIO); 895 896 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 897 if (error) 898 return (error); 899 900 if (cnp->cn_namelen == 1 && *pname == '.') { 901 if ((flags & ISLASTCN) && nameiop != LOOKUP) 902 return (EINVAL); 903 *vpp = dvp; 904 VREF(dvp); 905 return (0); 906 } 907 908 if (flags & ISDOTDOT) { 909 if ((flags & ISLASTCN) && nameiop != LOOKUP) 910 return (EINVAL); 911 de = devfs_parent_dirent(dd); 912 if (de == NULL) 913 return (ENOENT); 914 dvplocked = VOP_ISLOCKED(dvp); 915 VOP_UNLOCK(dvp, 0); 916 error = devfs_allocv(de, dvp->v_mount, 917 cnp->cn_lkflags & LK_TYPE_MASK, vpp); 918 *dm_unlock = 0; 919 vn_lock(dvp, dvplocked | LK_RETRY); 920 return (error); 921 } 922 923 dd = dvp->v_data; 924 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 925 while (de == NULL) { /* While(...) so we can use break */ 926 927 if (nameiop == DELETE) 928 return (ENOENT); 929 930 /* 931 * OK, we didn't have an entry for the name we were asked for 932 * so we try to see if anybody can create it on demand. 933 */ 934 pname = devfs_fqpn(specname, dmp, dd, cnp); 935 if (pname == NULL) 936 break; 937 938 cdev = NULL; 939 DEVFS_DMP_HOLD(dmp); 940 sx_xunlock(&dmp->dm_lock); 941 sx_slock(&clone_drain_lock); 942 EVENTHANDLER_INVOKE(dev_clone, 943 td->td_ucred, pname, strlen(pname), &cdev); 944 sx_sunlock(&clone_drain_lock); 945 946 if (cdev == NULL) 947 sx_xlock(&dmp->dm_lock); 948 else if (devfs_populate_vp(dvp) != 0) { 949 *dm_unlock = 0; 950 sx_xlock(&dmp->dm_lock); 951 if (DEVFS_DMP_DROP(dmp)) { 952 sx_xunlock(&dmp->dm_lock); 953 devfs_unmount_final(dmp); 954 } else 955 sx_xunlock(&dmp->dm_lock); 956 dev_rel(cdev); 957 return (ENOENT); 958 } 959 if (DEVFS_DMP_DROP(dmp)) { 960 *dm_unlock = 0; 961 sx_xunlock(&dmp->dm_lock); 962 devfs_unmount_final(dmp); 963 if (cdev != NULL) 964 dev_rel(cdev); 965 return (ENOENT); 966 } 967 968 if (cdev == NULL) 969 break; 970 971 dev_lock(); 972 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 973 if (dde != NULL && *dde != NULL) 974 de = *dde; 975 dev_unlock(); 976 dev_rel(cdev); 977 break; 978 } 979 980 if (de == NULL || de->de_flags & DE_WHITEOUT) { 981 if ((nameiop == CREATE || nameiop == RENAME) && 982 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 983 cnp->cn_flags |= SAVENAME; 984 return (EJUSTRETURN); 985 } 986 return (ENOENT); 987 } 988 989 if (devfs_prison_check(de, td)) 990 return (ENOENT); 991 992 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 993 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 994 if (error) 995 return (error); 996 if (*vpp == dvp) { 997 VREF(dvp); 998 *vpp = dvp; 999 return (0); 1000 } 1001 } 1002 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK, 1003 vpp); 1004 *dm_unlock = 0; 1005 return (error); 1006} 1007 1008static int 1009devfs_lookup(struct vop_lookup_args *ap) 1010{ 1011 int j; 1012 struct devfs_mount *dmp; 1013 int dm_unlock; 1014 1015 if (devfs_populate_vp(ap->a_dvp) != 0) 1016 return (ENOTDIR); 1017 1018 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1019 dm_unlock = 1; 1020 j = devfs_lookupx(ap, &dm_unlock); 1021 if (dm_unlock == 1) 1022 sx_xunlock(&dmp->dm_lock); 1023 return (j); 1024} 1025 1026static int 1027devfs_mknod(struct vop_mknod_args *ap) 1028{ 1029 struct componentname *cnp; 1030 struct vnode *dvp, **vpp; 1031 struct devfs_dirent *dd, *de; 1032 struct devfs_mount *dmp; 1033 int error; 1034 1035 /* 1036 * The only type of node we should be creating here is a 1037 * character device, for anything else return EOPNOTSUPP. 1038 */ 1039 if (ap->a_vap->va_type != VCHR) 1040 return (EOPNOTSUPP); 1041 dvp = ap->a_dvp; 1042 dmp = VFSTODEVFS(dvp->v_mount); 1043 1044 cnp = ap->a_cnp; 1045 vpp = ap->a_vpp; 1046 dd = dvp->v_data; 1047 1048 error = ENOENT; 1049 sx_xlock(&dmp->dm_lock); 1050 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1051 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1052 continue; 1053 if (de->de_dirent->d_type == DT_CHR && 1054 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1055 continue; 1056 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1057 de->de_dirent->d_namlen) != 0) 1058 continue; 1059 if (de->de_flags & DE_WHITEOUT) 1060 break; 1061 goto notfound; 1062 } 1063 if (de == NULL) 1064 goto notfound; 1065 de->de_flags &= ~DE_WHITEOUT; 1066 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1067 return (error); 1068notfound: 1069 sx_xunlock(&dmp->dm_lock); 1070 return (error); 1071} 1072 1073/* ARGSUSED */ 1074static int 1075devfs_open(struct vop_open_args *ap) 1076{ 1077 struct thread *td = ap->a_td; 1078 struct vnode *vp = ap->a_vp; 1079 struct cdev *dev = vp->v_rdev; 1080 struct file *fp = ap->a_fp; 1081 int error, ref, vlocked; 1082 struct cdevsw *dsw; 1083 struct file *fpop; 1084 struct mtx *mtxp; 1085 1086 if (vp->v_type == VBLK) 1087 return (ENXIO); 1088 1089 if (dev == NULL) 1090 return (ENXIO); 1091 1092 /* Make this field valid before any I/O in d_open. */ 1093 if (dev->si_iosize_max == 0) 1094 dev->si_iosize_max = DFLTPHYS; 1095 1096 dsw = dev_refthread(dev, &ref); 1097 if (dsw == NULL) 1098 return (ENXIO); 1099 if (fp == NULL && dsw->d_fdopen != NULL) { 1100 dev_relthread(dev, ref); 1101 return (ENXIO); 1102 } 1103 1104 vlocked = VOP_ISLOCKED(vp); 1105 VOP_UNLOCK(vp, 0); 1106 1107 fpop = td->td_fpop; 1108 td->td_fpop = fp; 1109 if (fp != NULL) { 1110 fp->f_data = dev; 1111 fp->f_vnode = vp; 1112 } 1113 if (dsw->d_fdopen != NULL) 1114 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1115 else 1116 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1117 /* Clean up any cdevpriv upon error. */ 1118 if (error != 0) 1119 devfs_clear_cdevpriv(); 1120 td->td_fpop = fpop; 1121 1122 vn_lock(vp, vlocked | LK_RETRY); 1123 dev_relthread(dev, ref); 1124 if (error != 0) { 1125 if (error == ERESTART) 1126 error = EINTR; 1127 return (error); 1128 } 1129 1130#if 0 /* /dev/console */ 1131 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1132#else 1133 if (fp == NULL) 1134 return (error); 1135#endif 1136 if (fp->f_ops == &badfileops) 1137 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1138 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1139 1140 /* 1141 * Hint to the dofilewrite() to not force the buffer draining 1142 * on the writer to the file. Most likely, the write would 1143 * not need normal buffers. 1144 */ 1145 mtx_lock(mtxp); 1146 fp->f_vnread_flags |= FDEVFS_VNODE; 1147 mtx_unlock(mtxp); 1148 return (error); 1149} 1150 1151static int 1152devfs_pathconf(struct vop_pathconf_args *ap) 1153{ 1154 1155 switch (ap->a_name) { 1156 case _PC_MAC_PRESENT: 1157#ifdef MAC 1158 /* 1159 * If MAC is enabled, devfs automatically supports 1160 * trivial non-persistant label storage. 1161 */ 1162 *ap->a_retval = 1; 1163#else 1164 *ap->a_retval = 0; 1165#endif 1166 return (0); 1167 default: 1168 return (vop_stdpathconf(ap)); 1169 } 1170 /* NOTREACHED */ 1171} 1172 1173/* ARGSUSED */ 1174static int 1175devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1176{ 1177 struct cdev *dev; 1178 struct cdevsw *dsw; 1179 int error, ref; 1180 struct file *fpop; 1181 1182 fpop = td->td_fpop; 1183 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1184 if (error != 0) { 1185 error = vnops.fo_poll(fp, events, cred, td); 1186 return (error); 1187 } 1188 error = dsw->d_poll(dev, events, td); 1189 td->td_fpop = fpop; 1190 dev_relthread(dev, ref); 1191 return(error); 1192} 1193 1194/* 1195 * Print out the contents of a special device vnode. 1196 */ 1197static int 1198devfs_print(struct vop_print_args *ap) 1199{ 1200 1201 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1202 return (0); 1203} 1204 1205static int 1206devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1207 int flags, struct thread *td) 1208{ 1209 struct cdev *dev; 1210 int ioflag, error, ref; 1211 ssize_t resid; 1212 struct cdevsw *dsw; 1213 struct file *fpop; 1214 1215 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1216 return (EINVAL); 1217 fpop = td->td_fpop; 1218 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1219 if (error != 0) { 1220 error = vnops.fo_read(fp, uio, cred, flags, td); 1221 return (error); 1222 } 1223 resid = uio->uio_resid; 1224 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1225 if (ioflag & O_DIRECT) 1226 ioflag |= IO_DIRECT; 1227 1228 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1229 error = dsw->d_read(dev, uio, ioflag); 1230 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1231 devfs_timestamp(&dev->si_atime); 1232 td->td_fpop = fpop; 1233 dev_relthread(dev, ref); 1234 1235 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1236 return (error); 1237} 1238 1239static int 1240devfs_readdir(struct vop_readdir_args *ap) 1241{ 1242 int error; 1243 struct uio *uio; 1244 struct dirent *dp; 1245 struct devfs_dirent *dd; 1246 struct devfs_dirent *de; 1247 struct devfs_mount *dmp; 1248 off_t off; 1249 int *tmp_ncookies = NULL; 1250 1251 if (ap->a_vp->v_type != VDIR) 1252 return (ENOTDIR); 1253 1254 uio = ap->a_uio; 1255 if (uio->uio_offset < 0) 1256 return (EINVAL); 1257 1258 /* 1259 * XXX: This is a temporary hack to get around this filesystem not 1260 * supporting cookies. We store the location of the ncookies pointer 1261 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1262 * and set the number of cookies to 0. We then set the pointer to 1263 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1264 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1265 * pointer to its original location before returning to the caller. 1266 */ 1267 if (ap->a_ncookies != NULL) { 1268 tmp_ncookies = ap->a_ncookies; 1269 *ap->a_ncookies = 0; 1270 ap->a_ncookies = NULL; 1271 } 1272 1273 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1274 if (devfs_populate_vp(ap->a_vp) != 0) { 1275 if (tmp_ncookies != NULL) 1276 ap->a_ncookies = tmp_ncookies; 1277 return (EIO); 1278 } 1279 error = 0; 1280 de = ap->a_vp->v_data; 1281 off = 0; 1282 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1283 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1284 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1285 continue; 1286 if (devfs_prison_check(dd, uio->uio_td)) 1287 continue; 1288 if (dd->de_dirent->d_type == DT_DIR) 1289 de = dd->de_dir; 1290 else 1291 de = dd; 1292 dp = dd->de_dirent; 1293 if (dp->d_reclen > uio->uio_resid) 1294 break; 1295 dp->d_fileno = de->de_inode; 1296 if (off >= uio->uio_offset) { 1297 error = vfs_read_dirent(ap, dp, off); 1298 if (error) 1299 break; 1300 } 1301 off += dp->d_reclen; 1302 } 1303 sx_xunlock(&dmp->dm_lock); 1304 uio->uio_offset = off; 1305 1306 /* 1307 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1308 * place. 1309 */ 1310 if (tmp_ncookies != NULL) 1311 ap->a_ncookies = tmp_ncookies; 1312 1313 return (error); 1314} 1315 1316static int 1317devfs_readlink(struct vop_readlink_args *ap) 1318{ 1319 struct devfs_dirent *de; 1320 1321 de = ap->a_vp->v_data; 1322 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1323} 1324 1325static int 1326devfs_reclaim(struct vop_reclaim_args *ap) 1327{ 1328 struct vnode *vp; 1329 struct devfs_dirent *de; 1330 1331 vp = ap->a_vp; 1332 mtx_lock(&devfs_de_interlock); 1333 de = vp->v_data; 1334 if (de != NULL) { 1335 de->de_vnode = NULL; 1336 vp->v_data = NULL; 1337 } 1338 mtx_unlock(&devfs_de_interlock); 1339 vnode_destroy_vobject(vp); 1340 return (0); 1341} 1342 1343static int 1344devfs_reclaim_vchr(struct vop_reclaim_args *ap) 1345{ 1346 struct vnode *vp; 1347 struct cdev *dev; 1348 1349 vp = ap->a_vp; 1350 MPASS(vp->v_type == VCHR); 1351 1352 devfs_reclaim(ap); 1353 1354 VI_LOCK(vp); 1355 dev_lock(); 1356 dev = vp->v_rdev; 1357 vp->v_rdev = NULL; 1358 if (dev != NULL) 1359 dev->si_usecount -= vp->v_usecount; 1360 dev_unlock(); 1361 VI_UNLOCK(vp); 1362 if (dev != NULL) 1363 dev_rel(dev); 1364 return (0); 1365} 1366 1367static int 1368devfs_remove(struct vop_remove_args *ap) 1369{ 1370 struct vnode *dvp = ap->a_dvp; 1371 struct vnode *vp = ap->a_vp; 1372 struct devfs_dirent *dd; 1373 struct devfs_dirent *de, *de_covered; 1374 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1375 1376 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1377 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1378 1379 sx_xlock(&dmp->dm_lock); 1380 dd = ap->a_dvp->v_data; 1381 de = vp->v_data; 1382 if (de->de_cdp == NULL) { 1383 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1384 if (de->de_dirent->d_type == DT_LNK) { 1385 de_covered = devfs_find(dd, de->de_dirent->d_name, 1386 de->de_dirent->d_namlen, 0); 1387 if (de_covered != NULL) 1388 de_covered->de_flags &= ~DE_COVERED; 1389 } 1390 /* We need to unlock dvp because devfs_delete() may lock it. */ 1391 VOP_UNLOCK(vp, 0); 1392 if (dvp != vp) 1393 VOP_UNLOCK(dvp, 0); 1394 devfs_delete(dmp, de, 0); 1395 sx_xunlock(&dmp->dm_lock); 1396 if (dvp != vp) 1397 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1398 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1399 } else { 1400 de->de_flags |= DE_WHITEOUT; 1401 sx_xunlock(&dmp->dm_lock); 1402 } 1403 return (0); 1404} 1405 1406/* 1407 * Revoke is called on a tty when a terminal session ends. The vnode 1408 * is orphaned by setting v_op to deadfs so we need to let go of it 1409 * as well so that we create a new one next time around. 1410 * 1411 */ 1412static int 1413devfs_revoke(struct vop_revoke_args *ap) 1414{ 1415 struct vnode *vp = ap->a_vp, *vp2; 1416 struct cdev *dev; 1417 struct cdev_priv *cdp; 1418 struct devfs_dirent *de; 1419 u_int i; 1420 1421 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1422 1423 dev = vp->v_rdev; 1424 cdp = cdev2priv(dev); 1425 1426 dev_lock(); 1427 cdp->cdp_inuse++; 1428 dev_unlock(); 1429 1430 vhold(vp); 1431 vgone(vp); 1432 vdrop(vp); 1433 1434 VOP_UNLOCK(vp,0); 1435 loop: 1436 for (;;) { 1437 mtx_lock(&devfs_de_interlock); 1438 dev_lock(); 1439 vp2 = NULL; 1440 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1441 de = cdp->cdp_dirents[i]; 1442 if (de == NULL) 1443 continue; 1444 1445 vp2 = de->de_vnode; 1446 if (vp2 != NULL) { 1447 dev_unlock(); 1448 VI_LOCK(vp2); 1449 mtx_unlock(&devfs_de_interlock); 1450 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1451 curthread)) 1452 goto loop; 1453 vhold(vp2); 1454 vgone(vp2); 1455 vdrop(vp2); 1456 vput(vp2); 1457 break; 1458 } 1459 } 1460 if (vp2 != NULL) { 1461 continue; 1462 } 1463 dev_unlock(); 1464 mtx_unlock(&devfs_de_interlock); 1465 break; 1466 } 1467 dev_lock(); 1468 cdp->cdp_inuse--; 1469 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1470 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1471 dev_unlock(); 1472 dev_rel(&cdp->cdp_c); 1473 } else 1474 dev_unlock(); 1475 1476 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1477 return (0); 1478} 1479 1480static int 1481devfs_rioctl(struct vop_ioctl_args *ap) 1482{ 1483 struct vnode *vp; 1484 struct devfs_mount *dmp; 1485 int error; 1486 1487 vp = ap->a_vp; 1488 vn_lock(vp, LK_SHARED | LK_RETRY); 1489 if (vp->v_iflag & VI_DOOMED) { 1490 VOP_UNLOCK(vp, 0); 1491 return (EBADF); 1492 } 1493 dmp = VFSTODEVFS(vp->v_mount); 1494 sx_xlock(&dmp->dm_lock); 1495 VOP_UNLOCK(vp, 0); 1496 DEVFS_DMP_HOLD(dmp); 1497 devfs_populate(dmp); 1498 if (DEVFS_DMP_DROP(dmp)) { 1499 sx_xunlock(&dmp->dm_lock); 1500 devfs_unmount_final(dmp); 1501 return (ENOENT); 1502 } 1503 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1504 sx_xunlock(&dmp->dm_lock); 1505 return (error); 1506} 1507 1508static int 1509devfs_rread(struct vop_read_args *ap) 1510{ 1511 1512 if (ap->a_vp->v_type != VDIR) 1513 return (EINVAL); 1514 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1515} 1516 1517static int 1518devfs_setattr(struct vop_setattr_args *ap) 1519{ 1520 struct devfs_dirent *de; 1521 struct vattr *vap; 1522 struct vnode *vp; 1523 struct thread *td; 1524 int c, error; 1525 uid_t uid; 1526 gid_t gid; 1527 1528 vap = ap->a_vap; 1529 vp = ap->a_vp; 1530 td = curthread; 1531 if ((vap->va_type != VNON) || 1532 (vap->va_nlink != VNOVAL) || 1533 (vap->va_fsid != VNOVAL) || 1534 (vap->va_fileid != VNOVAL) || 1535 (vap->va_blocksize != VNOVAL) || 1536 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1537 (vap->va_rdev != VNOVAL) || 1538 ((int)vap->va_bytes != VNOVAL) || 1539 (vap->va_gen != VNOVAL)) { 1540 return (EINVAL); 1541 } 1542 1543 error = devfs_populate_vp(vp); 1544 if (error != 0) 1545 return (error); 1546 1547 de = vp->v_data; 1548 if (vp->v_type == VDIR) 1549 de = de->de_dir; 1550 1551 c = 0; 1552 if (vap->va_uid == (uid_t)VNOVAL) 1553 uid = de->de_uid; 1554 else 1555 uid = vap->va_uid; 1556 if (vap->va_gid == (gid_t)VNOVAL) 1557 gid = de->de_gid; 1558 else 1559 gid = vap->va_gid; 1560 if (uid != de->de_uid || gid != de->de_gid) { 1561 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1562 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1563 error = priv_check(td, PRIV_VFS_CHOWN); 1564 if (error != 0) 1565 goto ret; 1566 } 1567 de->de_uid = uid; 1568 de->de_gid = gid; 1569 c = 1; 1570 } 1571 1572 if (vap->va_mode != (mode_t)VNOVAL) { 1573 if (ap->a_cred->cr_uid != de->de_uid) { 1574 error = priv_check(td, PRIV_VFS_ADMIN); 1575 if (error != 0) 1576 goto ret; 1577 } 1578 de->de_mode = vap->va_mode; 1579 c = 1; 1580 } 1581 1582 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1583 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1584 if (error != 0) 1585 goto ret; 1586 if (vap->va_atime.tv_sec != VNOVAL) { 1587 if (vp->v_type == VCHR) 1588 vp->v_rdev->si_atime = vap->va_atime; 1589 else 1590 de->de_atime = vap->va_atime; 1591 } 1592 if (vap->va_mtime.tv_sec != VNOVAL) { 1593 if (vp->v_type == VCHR) 1594 vp->v_rdev->si_mtime = vap->va_mtime; 1595 else 1596 de->de_mtime = vap->va_mtime; 1597 } 1598 c = 1; 1599 } 1600 1601 if (c) { 1602 if (vp->v_type == VCHR) 1603 vfs_timestamp(&vp->v_rdev->si_ctime); 1604 else 1605 vfs_timestamp(&de->de_mtime); 1606 } 1607 1608ret: 1609 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock); 1610 return (error); 1611} 1612 1613#ifdef MAC 1614static int 1615devfs_setlabel(struct vop_setlabel_args *ap) 1616{ 1617 struct vnode *vp; 1618 struct devfs_dirent *de; 1619 1620 vp = ap->a_vp; 1621 de = vp->v_data; 1622 1623 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1624 mac_devfs_update(vp->v_mount, de, vp); 1625 1626 return (0); 1627} 1628#endif 1629 1630static int 1631devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1632{ 1633 1634 return (vnops.fo_stat(fp, sb, cred, td)); 1635} 1636 1637static int 1638devfs_symlink(struct vop_symlink_args *ap) 1639{ 1640 int i, error; 1641 struct devfs_dirent *dd; 1642 struct devfs_dirent *de, *de_covered, *de_dotdot; 1643 struct devfs_mount *dmp; 1644 1645 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1646 if (error) 1647 return(error); 1648 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1649 if (devfs_populate_vp(ap->a_dvp) != 0) 1650 return (ENOENT); 1651 1652 dd = ap->a_dvp->v_data; 1653 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1654 de->de_flags = DE_USER; 1655 de->de_uid = 0; 1656 de->de_gid = 0; 1657 de->de_mode = 0755; 1658 de->de_inode = alloc_unr(devfs_inos); 1659 de->de_dir = dd; 1660 de->de_dirent->d_type = DT_LNK; 1661 i = strlen(ap->a_target) + 1; 1662 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1663 bcopy(ap->a_target, de->de_symlink, i); 1664#ifdef MAC 1665 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1666#endif 1667 de_covered = devfs_find(dd, de->de_dirent->d_name, 1668 de->de_dirent->d_namlen, 0); 1669 if (de_covered != NULL) { 1670 if ((de_covered->de_flags & DE_USER) != 0) { 1671 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1672 sx_xunlock(&dmp->dm_lock); 1673 return (EEXIST); 1674 } 1675 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1676 ("devfs_symlink: entry %p already covered", de_covered)); 1677 de_covered->de_flags |= DE_COVERED; 1678 } 1679 1680 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1681 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1682 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1683 devfs_dir_ref_de(dmp, dd); 1684 devfs_rules_apply(dmp, de); 1685 1686 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1687} 1688 1689static int 1690devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1691{ 1692 1693 return (vnops.fo_truncate(fp, length, cred, td)); 1694} 1695 1696static int 1697devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1698 int flags, struct thread *td) 1699{ 1700 struct cdev *dev; 1701 int error, ioflag, ref; 1702 ssize_t resid; 1703 struct cdevsw *dsw; 1704 struct file *fpop; 1705 1706 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1707 return (EINVAL); 1708 fpop = td->td_fpop; 1709 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1710 if (error != 0) { 1711 error = vnops.fo_write(fp, uio, cred, flags, td); 1712 return (error); 1713 } 1714 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1715 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1716 if (ioflag & O_DIRECT) 1717 ioflag |= IO_DIRECT; 1718 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1719 1720 resid = uio->uio_resid; 1721 1722 error = dsw->d_write(dev, uio, ioflag); 1723 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1724 devfs_timestamp(&dev->si_ctime); 1725 dev->si_mtime = dev->si_ctime; 1726 } 1727 td->td_fpop = fpop; 1728 dev_relthread(dev, ref); 1729 1730 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1731 return (error); 1732} 1733 1734dev_t 1735dev2udev(struct cdev *x) 1736{ 1737 if (x == NULL) 1738 return (NODEV); 1739 return (cdev2priv(x)->cdp_inode); 1740} 1741 1742static struct fileops devfs_ops_f = { 1743 .fo_read = devfs_read_f, 1744 .fo_write = devfs_write_f, 1745 .fo_truncate = devfs_truncate_f, 1746 .fo_ioctl = devfs_ioctl_f, 1747 .fo_poll = devfs_poll_f, 1748 .fo_kqfilter = devfs_kqfilter_f, 1749 .fo_stat = devfs_stat_f, 1750 .fo_close = devfs_close_f, 1751 .fo_chmod = vn_chmod, 1752 .fo_chown = vn_chown, 1753 .fo_sendfile = vn_sendfile, 1754 .fo_seek = vn_seek, 1755 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1756}; 1757 1758static struct vop_vector devfs_vnodeops = { 1759 .vop_default = &default_vnodeops, 1760 1761 .vop_access = devfs_access, 1762 .vop_getattr = devfs_getattr, 1763 .vop_ioctl = devfs_rioctl, 1764 .vop_lookup = devfs_lookup, 1765 .vop_mknod = devfs_mknod, 1766 .vop_pathconf = devfs_pathconf, 1767 .vop_read = devfs_rread, 1768 .vop_readdir = devfs_readdir, 1769 .vop_readlink = devfs_readlink, 1770 .vop_reclaim = devfs_reclaim, 1771 .vop_remove = devfs_remove, 1772 .vop_revoke = devfs_revoke, 1773 .vop_setattr = devfs_setattr, 1774#ifdef MAC 1775 .vop_setlabel = devfs_setlabel, 1776#endif 1777 .vop_symlink = devfs_symlink, 1778 .vop_vptocnp = devfs_vptocnp, 1779}; 1780 1781static struct vop_vector devfs_specops = { 1782 .vop_default = &default_vnodeops, 1783 1784 .vop_access = devfs_access, 1785 .vop_bmap = VOP_PANIC, 1786 .vop_close = devfs_close, 1787 .vop_create = VOP_PANIC, 1788 .vop_fsync = devfs_fsync, 1789 .vop_getattr = devfs_getattr, 1790 .vop_link = VOP_PANIC, 1791 .vop_mkdir = VOP_PANIC, 1792 .vop_mknod = VOP_PANIC, 1793 .vop_open = devfs_open, 1794 .vop_pathconf = devfs_pathconf, 1795 .vop_poll = dead_poll, 1796 .vop_print = devfs_print, 1797 .vop_read = dead_read, 1798 .vop_readdir = VOP_PANIC, 1799 .vop_readlink = VOP_PANIC, 1800 .vop_reallocblks = VOP_PANIC, 1801 .vop_reclaim = devfs_reclaim_vchr, 1802 .vop_remove = devfs_remove, 1803 .vop_rename = VOP_PANIC, 1804 .vop_revoke = devfs_revoke, 1805 .vop_rmdir = VOP_PANIC, 1806 .vop_setattr = devfs_setattr, 1807#ifdef MAC 1808 .vop_setlabel = devfs_setlabel, 1809#endif 1810 .vop_strategy = VOP_PANIC, 1811 .vop_symlink = VOP_PANIC, 1812 .vop_vptocnp = devfs_vptocnp, 1813 .vop_write = dead_write, 1814}; 1815 1816/* 1817 * Our calling convention to the device drivers used to be that we passed 1818 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1819 * flags instead since that's what open(), close() and ioctl() takes and 1820 * we don't really want vnode.h in device drivers. 1821 * We solved the source compatibility by redefining some vnode flags to 1822 * be the same as the fcntl ones and by sending down the bitwise OR of 1823 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1824 * pulls the rug out under this. 1825 */ 1826CTASSERT(O_NONBLOCK == IO_NDELAY); 1827CTASSERT(O_FSYNC == IO_SYNC); 1828