devfs_vnops.c revision 279685
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 279685 2015-03-06 09:22:05Z 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/time.h> 61#include <sys/ttycom.h> 62#include <sys/unistd.h> 63#include <sys/vnode.h> 64 65static struct vop_vector devfs_vnodeops; 66static struct fileops devfs_ops_f; 67 68#include <fs/devfs/devfs.h> 69#include <fs/devfs/devfs_int.h> 70 71#include <security/mac/mac_framework.h> 72 73static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data"); 74 75struct mtx devfs_de_interlock; 76MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF); 77struct sx clone_drain_lock; 78SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock"); 79struct mtx cdevpriv_mtx; 80MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF); 81 82static int 83devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp, 84 int *ref) 85{ 86 87 *dswp = devvn_refthread(fp->f_vnode, devp, ref); 88 if (*devp != fp->f_data) { 89 if (*dswp != NULL) 90 dev_relthread(*devp, *ref); 91 return (ENXIO); 92 } 93 KASSERT((*devp)->si_refcount > 0, 94 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp))); 95 if (*dswp == NULL) 96 return (ENXIO); 97 curthread->td_fpop = fp; 98 return (0); 99} 100 101int 102devfs_get_cdevpriv(void **datap) 103{ 104 struct file *fp; 105 struct cdev_privdata *p; 106 int error; 107 108 fp = curthread->td_fpop; 109 if (fp == NULL) 110 return (EBADF); 111 p = fp->f_cdevpriv; 112 if (p != NULL) { 113 error = 0; 114 *datap = p->cdpd_data; 115 } else 116 error = ENOENT; 117 return (error); 118} 119 120int 121devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr) 122{ 123 struct file *fp; 124 struct cdev_priv *cdp; 125 struct cdev_privdata *p; 126 int error; 127 128 fp = curthread->td_fpop; 129 if (fp == NULL) 130 return (ENOENT); 131 cdp = cdev2priv((struct cdev *)fp->f_data); 132 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK); 133 p->cdpd_data = priv; 134 p->cdpd_dtr = priv_dtr; 135 p->cdpd_fp = fp; 136 mtx_lock(&cdevpriv_mtx); 137 if (fp->f_cdevpriv == NULL) { 138 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list); 139 fp->f_cdevpriv = p; 140 mtx_unlock(&cdevpriv_mtx); 141 error = 0; 142 } else { 143 mtx_unlock(&cdevpriv_mtx); 144 free(p, M_CDEVPDATA); 145 error = EBUSY; 146 } 147 return (error); 148} 149 150void 151devfs_destroy_cdevpriv(struct cdev_privdata *p) 152{ 153 154 mtx_assert(&cdevpriv_mtx, MA_OWNED); 155 p->cdpd_fp->f_cdevpriv = NULL; 156 LIST_REMOVE(p, cdpd_list); 157 mtx_unlock(&cdevpriv_mtx); 158 (p->cdpd_dtr)(p->cdpd_data); 159 free(p, M_CDEVPDATA); 160} 161 162void 163devfs_fpdrop(struct file *fp) 164{ 165 struct cdev_privdata *p; 166 167 mtx_lock(&cdevpriv_mtx); 168 if ((p = fp->f_cdevpriv) == NULL) { 169 mtx_unlock(&cdevpriv_mtx); 170 return; 171 } 172 devfs_destroy_cdevpriv(p); 173} 174 175void 176devfs_clear_cdevpriv(void) 177{ 178 struct file *fp; 179 180 fp = curthread->td_fpop; 181 if (fp == NULL) 182 return; 183 devfs_fpdrop(fp); 184} 185 186/* 187 * On success devfs_populate_vp() returns with dmp->dm_lock held. 188 */ 189static int 190devfs_populate_vp(struct vnode *vp) 191{ 192 struct devfs_dirent *de; 193 struct devfs_mount *dmp; 194 int locked; 195 196 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp"); 197 198 dmp = VFSTODEVFS(vp->v_mount); 199 locked = VOP_ISLOCKED(vp); 200 201 sx_xlock(&dmp->dm_lock); 202 DEVFS_DMP_HOLD(dmp); 203 204 /* Can't call devfs_populate() with the vnode lock held. */ 205 VOP_UNLOCK(vp, 0); 206 devfs_populate(dmp); 207 208 sx_xunlock(&dmp->dm_lock); 209 vn_lock(vp, locked | LK_RETRY); 210 sx_xlock(&dmp->dm_lock); 211 if (DEVFS_DMP_DROP(dmp)) { 212 sx_xunlock(&dmp->dm_lock); 213 devfs_unmount_final(dmp); 214 return (EBADF); 215 } 216 if ((vp->v_iflag & VI_DOOMED) != 0) { 217 sx_xunlock(&dmp->dm_lock); 218 return (EBADF); 219 } 220 de = vp->v_data; 221 KASSERT(de != NULL, 222 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed")); 223 if ((de->de_flags & DE_DOOMED) != 0) { 224 sx_xunlock(&dmp->dm_lock); 225 return (EBADF); 226 } 227 228 return (0); 229} 230 231static int 232devfs_vptocnp(struct vop_vptocnp_args *ap) 233{ 234 struct vnode *vp = ap->a_vp; 235 struct vnode **dvp = ap->a_vpp; 236 struct devfs_mount *dmp; 237 char *buf = ap->a_buf; 238 int *buflen = ap->a_buflen; 239 struct devfs_dirent *dd, *de; 240 int i, error; 241 242 dmp = VFSTODEVFS(vp->v_mount); 243 244 error = devfs_populate_vp(vp); 245 if (error != 0) 246 return (error); 247 248 i = *buflen; 249 dd = vp->v_data; 250 251 if (vp->v_type == VCHR) { 252 i -= strlen(dd->de_cdp->cdp_c.si_name); 253 if (i < 0) { 254 error = ENOMEM; 255 goto finished; 256 } 257 bcopy(dd->de_cdp->cdp_c.si_name, buf + i, 258 strlen(dd->de_cdp->cdp_c.si_name)); 259 de = dd->de_dir; 260 } else if (vp->v_type == VDIR) { 261 if (dd == dmp->dm_rootdir) { 262 *dvp = vp; 263 vref(*dvp); 264 goto finished; 265 } 266 i -= dd->de_dirent->d_namlen; 267 if (i < 0) { 268 error = ENOMEM; 269 goto finished; 270 } 271 bcopy(dd->de_dirent->d_name, buf + i, 272 dd->de_dirent->d_namlen); 273 de = dd; 274 } else { 275 error = ENOENT; 276 goto finished; 277 } 278 *buflen = i; 279 de = devfs_parent_dirent(de); 280 if (de == NULL) { 281 error = ENOENT; 282 goto finished; 283 } 284 mtx_lock(&devfs_de_interlock); 285 *dvp = de->de_vnode; 286 if (*dvp != NULL) { 287 VI_LOCK(*dvp); 288 mtx_unlock(&devfs_de_interlock); 289 vholdl(*dvp); 290 VI_UNLOCK(*dvp); 291 vref(*dvp); 292 vdrop(*dvp); 293 } else { 294 mtx_unlock(&devfs_de_interlock); 295 error = ENOENT; 296 } 297finished: 298 sx_xunlock(&dmp->dm_lock); 299 return (error); 300} 301 302/* 303 * Construct the fully qualified path name relative to the mountpoint. 304 * If a NULL cnp is provided, no '/' is appended to the resulting path. 305 */ 306char * 307devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd, 308 struct componentname *cnp) 309{ 310 int i; 311 struct devfs_dirent *de; 312 313 sx_assert(&dmp->dm_lock, SA_LOCKED); 314 315 i = SPECNAMELEN; 316 buf[i] = '\0'; 317 if (cnp != NULL) 318 i -= cnp->cn_namelen; 319 if (i < 0) 320 return (NULL); 321 if (cnp != NULL) 322 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen); 323 de = dd; 324 while (de != dmp->dm_rootdir) { 325 if (cnp != NULL || i < SPECNAMELEN) { 326 i--; 327 if (i < 0) 328 return (NULL); 329 buf[i] = '/'; 330 } 331 i -= de->de_dirent->d_namlen; 332 if (i < 0) 333 return (NULL); 334 bcopy(de->de_dirent->d_name, buf + i, 335 de->de_dirent->d_namlen); 336 de = devfs_parent_dirent(de); 337 if (de == NULL) 338 return (NULL); 339 } 340 return (buf + i); 341} 342 343static int 344devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp, 345 struct devfs_dirent *de) 346{ 347 int not_found; 348 349 not_found = 0; 350 if (de->de_flags & DE_DOOMED) 351 not_found = 1; 352 if (DEVFS_DE_DROP(de)) { 353 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed")); 354 devfs_dirent_free(de); 355 } 356 if (DEVFS_DMP_DROP(dmp)) { 357 KASSERT(not_found == 1, 358 ("DEVFS mount struct freed before dirent")); 359 not_found = 2; 360 sx_xunlock(&dmp->dm_lock); 361 devfs_unmount_final(dmp); 362 } 363 if (not_found == 1 || (drop_dm_lock && not_found != 2)) 364 sx_unlock(&dmp->dm_lock); 365 return (not_found); 366} 367 368static void 369devfs_insmntque_dtr(struct vnode *vp, void *arg) 370{ 371 struct devfs_dirent *de; 372 373 de = (struct devfs_dirent *)arg; 374 mtx_lock(&devfs_de_interlock); 375 vp->v_data = NULL; 376 de->de_vnode = NULL; 377 mtx_unlock(&devfs_de_interlock); 378 vgone(vp); 379 vput(vp); 380} 381 382/* 383 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops 384 * it on return. 385 */ 386int 387devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode, 388 struct vnode **vpp) 389{ 390 int error; 391 struct vnode *vp; 392 struct cdev *dev; 393 struct devfs_mount *dmp; 394 struct cdevsw *dsw; 395 396 dmp = VFSTODEVFS(mp); 397 if (de->de_flags & DE_DOOMED) { 398 sx_xunlock(&dmp->dm_lock); 399 return (ENOENT); 400 } 401loop: 402 DEVFS_DE_HOLD(de); 403 DEVFS_DMP_HOLD(dmp); 404 mtx_lock(&devfs_de_interlock); 405 vp = de->de_vnode; 406 if (vp != NULL) { 407 VI_LOCK(vp); 408 mtx_unlock(&devfs_de_interlock); 409 sx_xunlock(&dmp->dm_lock); 410 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread); 411 sx_xlock(&dmp->dm_lock); 412 if (devfs_allocv_drop_refs(0, dmp, de)) { 413 vput(vp); 414 return (ENOENT); 415 } 416 else if ((vp->v_iflag & VI_DOOMED) != 0) { 417 mtx_lock(&devfs_de_interlock); 418 if (de->de_vnode == vp) { 419 de->de_vnode = NULL; 420 vp->v_data = NULL; 421 } 422 mtx_unlock(&devfs_de_interlock); 423 vput(vp); 424 goto loop; 425 } 426 sx_xunlock(&dmp->dm_lock); 427 *vpp = vp; 428 return (0); 429 } 430 mtx_unlock(&devfs_de_interlock); 431 if (de->de_dirent->d_type == DT_CHR) { 432 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) { 433 devfs_allocv_drop_refs(1, dmp, de); 434 return (ENOENT); 435 } 436 dev = &de->de_cdp->cdp_c; 437 } else { 438 dev = NULL; 439 } 440 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp); 441 if (error != 0) { 442 devfs_allocv_drop_refs(1, dmp, de); 443 printf("devfs_allocv: failed to allocate new vnode\n"); 444 return (error); 445 } 446 447 if (de->de_dirent->d_type == DT_CHR) { 448 vp->v_type = VCHR; 449 VI_LOCK(vp); 450 dev_lock(); 451 dev_refl(dev); 452 /* XXX: v_rdev should be protect by vnode lock */ 453 vp->v_rdev = dev; 454 KASSERT(vp->v_usecount == 1, 455 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount)); 456 dev->si_usecount += vp->v_usecount; 457 /* Special casing of ttys for deadfs. Probably redundant. */ 458 dsw = dev->si_devsw; 459 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0) 460 vp->v_vflag |= VV_ISTTY; 461 dev_unlock(); 462 VI_UNLOCK(vp); 463 if ((dev->si_flags & SI_ETERNAL) != 0) 464 vp->v_vflag |= VV_ETERNALDEV; 465 vp->v_op = &devfs_specops; 466 } else if (de->de_dirent->d_type == DT_DIR) { 467 vp->v_type = VDIR; 468 } else if (de->de_dirent->d_type == DT_LNK) { 469 vp->v_type = VLNK; 470 } else { 471 vp->v_type = VBAD; 472 } 473 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS); 474 VN_LOCK_ASHARE(vp); 475 mtx_lock(&devfs_de_interlock); 476 vp->v_data = de; 477 de->de_vnode = vp; 478 mtx_unlock(&devfs_de_interlock); 479 error = insmntque1(vp, mp, devfs_insmntque_dtr, de); 480 if (error != 0) { 481 (void) devfs_allocv_drop_refs(1, dmp, de); 482 return (error); 483 } 484 if (devfs_allocv_drop_refs(0, dmp, de)) { 485 vput(vp); 486 return (ENOENT); 487 } 488#ifdef MAC 489 mac_devfs_vnode_associate(mp, de, vp); 490#endif 491 sx_xunlock(&dmp->dm_lock); 492 *vpp = vp; 493 return (0); 494} 495 496static int 497devfs_access(struct vop_access_args *ap) 498{ 499 struct vnode *vp = ap->a_vp; 500 struct devfs_dirent *de; 501 int error; 502 503 de = vp->v_data; 504 if (vp->v_type == VDIR) 505 de = de->de_dir; 506 507 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid, 508 ap->a_accmode, ap->a_cred, NULL); 509 if (error == 0) 510 return (0); 511 if (error != EACCES) 512 return (error); 513 /* We do, however, allow access to the controlling terminal */ 514 if (!(ap->a_td->td_proc->p_flag & P_CONTROLT)) 515 return (error); 516 if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp) 517 return (0); 518 return (error); 519} 520 521/* ARGSUSED */ 522static int 523devfs_close(struct vop_close_args *ap) 524{ 525 struct vnode *vp = ap->a_vp, *oldvp; 526 struct thread *td = ap->a_td; 527 struct cdev *dev = vp->v_rdev; 528 struct cdevsw *dsw; 529 int vp_locked, error, ref; 530 531 /* 532 * XXX: Don't call d_close() if we were called because of 533 * XXX: insmntque1() failure. 534 */ 535 if (vp->v_data == NULL) 536 return (0); 537 538 /* 539 * Hack: a tty device that is a controlling terminal 540 * has a reference from the session structure. 541 * We cannot easily tell that a character device is 542 * a controlling terminal, unless it is the closing 543 * process' controlling terminal. In that case, 544 * if the reference count is 2 (this last descriptor 545 * plus the session), release the reference from the session. 546 */ 547 oldvp = NULL; 548 sx_xlock(&proctree_lock); 549 if (td && vp == td->td_proc->p_session->s_ttyvp) { 550 SESS_LOCK(td->td_proc->p_session); 551 VI_LOCK(vp); 552 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) { 553 td->td_proc->p_session->s_ttyvp = NULL; 554 td->td_proc->p_session->s_ttydp = NULL; 555 oldvp = vp; 556 } 557 VI_UNLOCK(vp); 558 SESS_UNLOCK(td->td_proc->p_session); 559 } 560 sx_xunlock(&proctree_lock); 561 if (oldvp != NULL) 562 vrele(oldvp); 563 /* 564 * We do not want to really close the device if it 565 * is still in use unless we are trying to close it 566 * forcibly. Since every use (buffer, vnode, swap, cmap) 567 * holds a reference to the vnode, and because we mark 568 * any other vnodes that alias this device, when the 569 * sum of the reference counts on all the aliased 570 * vnodes descends to one, we are on last close. 571 */ 572 dsw = dev_refthread(dev, &ref); 573 if (dsw == NULL) 574 return (ENXIO); 575 VI_LOCK(vp); 576 if (vp->v_iflag & VI_DOOMED) { 577 /* Forced close. */ 578 } else if (dsw->d_flags & D_TRACKCLOSE) { 579 /* Keep device updated on status. */ 580 } else if (count_dev(dev) > 1) { 581 VI_UNLOCK(vp); 582 dev_relthread(dev, ref); 583 return (0); 584 } 585 vholdl(vp); 586 VI_UNLOCK(vp); 587 vp_locked = VOP_ISLOCKED(vp); 588 VOP_UNLOCK(vp, 0); 589 KASSERT(dev->si_refcount > 0, 590 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev))); 591 error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td); 592 dev_relthread(dev, ref); 593 vn_lock(vp, vp_locked | LK_RETRY); 594 vdrop(vp); 595 return (error); 596} 597 598static int 599devfs_close_f(struct file *fp, struct thread *td) 600{ 601 int error; 602 struct file *fpop; 603 604 /* 605 * NB: td may be NULL if this descriptor is closed due to 606 * garbage collection from a closed UNIX domain socket. 607 */ 608 fpop = curthread->td_fpop; 609 curthread->td_fpop = fp; 610 error = vnops.fo_close(fp, td); 611 curthread->td_fpop = fpop; 612 613 /* 614 * The f_cdevpriv cannot be assigned non-NULL value while we 615 * are destroying the file. 616 */ 617 if (fp->f_cdevpriv != NULL) 618 devfs_fpdrop(fp); 619 return (error); 620} 621 622static int 623devfs_fsync(struct vop_fsync_args *ap) 624{ 625 int error; 626 struct bufobj *bo; 627 struct devfs_dirent *de; 628 629 if (!vn_isdisk(ap->a_vp, &error)) { 630 bo = &ap->a_vp->v_bufobj; 631 de = ap->a_vp->v_data; 632 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) { 633 printf("Device %s went missing before all of the data " 634 "could be written to it; expect data loss.\n", 635 de->de_dirent->d_name); 636 637 error = vop_stdfsync(ap); 638 if (bo->bo_dirty.bv_cnt != 0 || error != 0) 639 panic("devfs_fsync: vop_stdfsync failed."); 640 } 641 642 return (0); 643 } 644 645 return (vop_stdfsync(ap)); 646} 647 648static int 649devfs_getattr(struct vop_getattr_args *ap) 650{ 651 struct vnode *vp = ap->a_vp; 652 struct vattr *vap = ap->a_vap; 653 int error; 654 struct devfs_dirent *de; 655 struct devfs_mount *dmp; 656 struct cdev *dev; 657 658 error = devfs_populate_vp(vp); 659 if (error != 0) 660 return (error); 661 662 dmp = VFSTODEVFS(vp->v_mount); 663 sx_xunlock(&dmp->dm_lock); 664 665 de = vp->v_data; 666 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp)); 667 if (vp->v_type == VDIR) { 668 de = de->de_dir; 669 KASSERT(de != NULL, 670 ("Null dir dirent in devfs_getattr vp=%p", vp)); 671 } 672 vap->va_uid = de->de_uid; 673 vap->va_gid = de->de_gid; 674 vap->va_mode = de->de_mode; 675 if (vp->v_type == VLNK) 676 vap->va_size = strlen(de->de_symlink); 677 else if (vp->v_type == VDIR) 678 vap->va_size = vap->va_bytes = DEV_BSIZE; 679 else 680 vap->va_size = 0; 681 if (vp->v_type != VDIR) 682 vap->va_bytes = 0; 683 vap->va_blocksize = DEV_BSIZE; 684 vap->va_type = vp->v_type; 685 686#define fix(aa) \ 687 do { \ 688 if ((aa).tv_sec <= 3600) { \ 689 (aa).tv_sec = boottime.tv_sec; \ 690 (aa).tv_nsec = boottime.tv_usec * 1000; \ 691 } \ 692 } while (0) 693 694 if (vp->v_type != VCHR) { 695 fix(de->de_atime); 696 vap->va_atime = de->de_atime; 697 fix(de->de_mtime); 698 vap->va_mtime = de->de_mtime; 699 fix(de->de_ctime); 700 vap->va_ctime = de->de_ctime; 701 } else { 702 dev = vp->v_rdev; 703 fix(dev->si_atime); 704 vap->va_atime = dev->si_atime; 705 fix(dev->si_mtime); 706 vap->va_mtime = dev->si_mtime; 707 fix(dev->si_ctime); 708 vap->va_ctime = dev->si_ctime; 709 710 vap->va_rdev = cdev2priv(dev)->cdp_inode; 711 } 712 vap->va_gen = 0; 713 vap->va_flags = 0; 714 vap->va_filerev = 0; 715 vap->va_nlink = de->de_links; 716 vap->va_fileid = de->de_inode; 717 718 return (error); 719} 720 721/* ARGSUSED */ 722static int 723devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td) 724{ 725 struct cdev *dev; 726 struct cdevsw *dsw; 727 struct vnode *vp; 728 struct vnode *vpold; 729 int error, i, ref; 730 const char *p; 731 struct fiodgname_arg *fgn; 732 struct file *fpop; 733 734 fpop = td->td_fpop; 735 error = devfs_fp_check(fp, &dev, &dsw, &ref); 736 if (error != 0) { 737 error = vnops.fo_ioctl(fp, com, data, cred, td); 738 return (error); 739 } 740 741 if (com == FIODTYPE) { 742 *(int *)data = dsw->d_flags & D_TYPEMASK; 743 td->td_fpop = fpop; 744 dev_relthread(dev, ref); 745 return (0); 746 } else if (com == FIODGNAME) { 747 fgn = data; 748 p = devtoname(dev); 749 i = strlen(p) + 1; 750 if (i > fgn->len) 751 error = EINVAL; 752 else 753 error = copyout(p, fgn->buf, i); 754 td->td_fpop = fpop; 755 dev_relthread(dev, ref); 756 return (error); 757 } 758 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td); 759 td->td_fpop = NULL; 760 dev_relthread(dev, ref); 761 if (error == ENOIOCTL) 762 error = ENOTTY; 763 if (error == 0 && com == TIOCSCTTY) { 764 vp = fp->f_vnode; 765 766 /* Do nothing if reassigning same control tty */ 767 sx_slock(&proctree_lock); 768 if (td->td_proc->p_session->s_ttyvp == vp) { 769 sx_sunlock(&proctree_lock); 770 return (0); 771 } 772 773 vpold = td->td_proc->p_session->s_ttyvp; 774 VREF(vp); 775 SESS_LOCK(td->td_proc->p_session); 776 td->td_proc->p_session->s_ttyvp = vp; 777 td->td_proc->p_session->s_ttydp = cdev2priv(dev); 778 SESS_UNLOCK(td->td_proc->p_session); 779 780 sx_sunlock(&proctree_lock); 781 782 /* Get rid of reference to old control tty */ 783 if (vpold) 784 vrele(vpold); 785 } 786 return (error); 787} 788 789/* ARGSUSED */ 790static int 791devfs_kqfilter_f(struct file *fp, struct knote *kn) 792{ 793 struct cdev *dev; 794 struct cdevsw *dsw; 795 int error, ref; 796 struct file *fpop; 797 struct thread *td; 798 799 td = curthread; 800 fpop = td->td_fpop; 801 error = devfs_fp_check(fp, &dev, &dsw, &ref); 802 if (error) 803 return (error); 804 error = dsw->d_kqfilter(dev, kn); 805 td->td_fpop = fpop; 806 dev_relthread(dev, ref); 807 return (error); 808} 809 810static inline int 811devfs_prison_check(struct devfs_dirent *de, struct thread *td) 812{ 813 struct cdev_priv *cdp; 814 struct ucred *dcr; 815 int error; 816 817 cdp = de->de_cdp; 818 if (cdp == NULL) 819 return (0); 820 dcr = cdp->cdp_c.si_cred; 821 if (dcr == NULL) 822 return (0); 823 824 error = prison_check(td->td_ucred, dcr); 825 if (error == 0) 826 return (0); 827 /* We do, however, allow access to the controlling terminal */ 828 if (!(td->td_proc->p_flag & P_CONTROLT)) 829 return (error); 830 if (td->td_proc->p_session->s_ttydp == cdp) 831 return (0); 832 return (error); 833} 834 835static int 836devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock) 837{ 838 struct componentname *cnp; 839 struct vnode *dvp, **vpp; 840 struct thread *td; 841 struct devfs_dirent *de, *dd; 842 struct devfs_dirent **dde; 843 struct devfs_mount *dmp; 844 struct cdev *cdev; 845 int error, flags, nameiop, dvplocked; 846 char specname[SPECNAMELEN + 1], *pname; 847 848 cnp = ap->a_cnp; 849 vpp = ap->a_vpp; 850 dvp = ap->a_dvp; 851 pname = cnp->cn_nameptr; 852 td = cnp->cn_thread; 853 flags = cnp->cn_flags; 854 nameiop = cnp->cn_nameiop; 855 dmp = VFSTODEVFS(dvp->v_mount); 856 dd = dvp->v_data; 857 *vpp = NULLVP; 858 859 if ((flags & ISLASTCN) && nameiop == RENAME) 860 return (EOPNOTSUPP); 861 862 if (dvp->v_type != VDIR) 863 return (ENOTDIR); 864 865 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT)) 866 return (EIO); 867 868 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 869 if (error) 870 return (error); 871 872 if (cnp->cn_namelen == 1 && *pname == '.') { 873 if ((flags & ISLASTCN) && nameiop != LOOKUP) 874 return (EINVAL); 875 *vpp = dvp; 876 VREF(dvp); 877 return (0); 878 } 879 880 if (flags & ISDOTDOT) { 881 if ((flags & ISLASTCN) && nameiop != LOOKUP) 882 return (EINVAL); 883 de = devfs_parent_dirent(dd); 884 if (de == NULL) 885 return (ENOENT); 886 dvplocked = VOP_ISLOCKED(dvp); 887 VOP_UNLOCK(dvp, 0); 888 error = devfs_allocv(de, dvp->v_mount, 889 cnp->cn_lkflags & LK_TYPE_MASK, vpp); 890 *dm_unlock = 0; 891 vn_lock(dvp, dvplocked | LK_RETRY); 892 return (error); 893 } 894 895 dd = dvp->v_data; 896 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0); 897 while (de == NULL) { /* While(...) so we can use break */ 898 899 if (nameiop == DELETE) 900 return (ENOENT); 901 902 /* 903 * OK, we didn't have an entry for the name we were asked for 904 * so we try to see if anybody can create it on demand. 905 */ 906 pname = devfs_fqpn(specname, dmp, dd, cnp); 907 if (pname == NULL) 908 break; 909 910 cdev = NULL; 911 DEVFS_DMP_HOLD(dmp); 912 sx_xunlock(&dmp->dm_lock); 913 sx_slock(&clone_drain_lock); 914 EVENTHANDLER_INVOKE(dev_clone, 915 td->td_ucred, pname, strlen(pname), &cdev); 916 sx_sunlock(&clone_drain_lock); 917 918 if (cdev == NULL) 919 sx_xlock(&dmp->dm_lock); 920 else if (devfs_populate_vp(dvp) != 0) { 921 *dm_unlock = 0; 922 sx_xlock(&dmp->dm_lock); 923 if (DEVFS_DMP_DROP(dmp)) { 924 sx_xunlock(&dmp->dm_lock); 925 devfs_unmount_final(dmp); 926 } else 927 sx_xunlock(&dmp->dm_lock); 928 dev_rel(cdev); 929 return (ENOENT); 930 } 931 if (DEVFS_DMP_DROP(dmp)) { 932 *dm_unlock = 0; 933 sx_xunlock(&dmp->dm_lock); 934 devfs_unmount_final(dmp); 935 if (cdev != NULL) 936 dev_rel(cdev); 937 return (ENOENT); 938 } 939 940 if (cdev == NULL) 941 break; 942 943 dev_lock(); 944 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx]; 945 if (dde != NULL && *dde != NULL) 946 de = *dde; 947 dev_unlock(); 948 dev_rel(cdev); 949 break; 950 } 951 952 if (de == NULL || de->de_flags & DE_WHITEOUT) { 953 if ((nameiop == CREATE || nameiop == RENAME) && 954 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) { 955 cnp->cn_flags |= SAVENAME; 956 return (EJUSTRETURN); 957 } 958 return (ENOENT); 959 } 960 961 if (devfs_prison_check(de, td)) 962 return (ENOENT); 963 964 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) { 965 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 966 if (error) 967 return (error); 968 if (*vpp == dvp) { 969 VREF(dvp); 970 *vpp = dvp; 971 return (0); 972 } 973 } 974 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK, 975 vpp); 976 *dm_unlock = 0; 977 return (error); 978} 979 980static int 981devfs_lookup(struct vop_lookup_args *ap) 982{ 983 int j; 984 struct devfs_mount *dmp; 985 int dm_unlock; 986 987 if (devfs_populate_vp(ap->a_dvp) != 0) 988 return (ENOTDIR); 989 990 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 991 dm_unlock = 1; 992 j = devfs_lookupx(ap, &dm_unlock); 993 if (dm_unlock == 1) 994 sx_xunlock(&dmp->dm_lock); 995 return (j); 996} 997 998static int 999devfs_mknod(struct vop_mknod_args *ap) 1000{ 1001 struct componentname *cnp; 1002 struct vnode *dvp, **vpp; 1003 struct devfs_dirent *dd, *de; 1004 struct devfs_mount *dmp; 1005 int error; 1006 1007 /* 1008 * The only type of node we should be creating here is a 1009 * character device, for anything else return EOPNOTSUPP. 1010 */ 1011 if (ap->a_vap->va_type != VCHR) 1012 return (EOPNOTSUPP); 1013 dvp = ap->a_dvp; 1014 dmp = VFSTODEVFS(dvp->v_mount); 1015 1016 cnp = ap->a_cnp; 1017 vpp = ap->a_vpp; 1018 dd = dvp->v_data; 1019 1020 error = ENOENT; 1021 sx_xlock(&dmp->dm_lock); 1022 TAILQ_FOREACH(de, &dd->de_dlist, de_list) { 1023 if (cnp->cn_namelen != de->de_dirent->d_namlen) 1024 continue; 1025 if (de->de_dirent->d_type == DT_CHR && 1026 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0) 1027 continue; 1028 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name, 1029 de->de_dirent->d_namlen) != 0) 1030 continue; 1031 if (de->de_flags & DE_WHITEOUT) 1032 break; 1033 goto notfound; 1034 } 1035 if (de == NULL) 1036 goto notfound; 1037 de->de_flags &= ~DE_WHITEOUT; 1038 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp); 1039 return (error); 1040notfound: 1041 sx_xunlock(&dmp->dm_lock); 1042 return (error); 1043} 1044 1045/* ARGSUSED */ 1046static int 1047devfs_open(struct vop_open_args *ap) 1048{ 1049 struct thread *td = ap->a_td; 1050 struct vnode *vp = ap->a_vp; 1051 struct cdev *dev = vp->v_rdev; 1052 struct file *fp = ap->a_fp; 1053 int error, ref, vlocked; 1054 struct cdevsw *dsw; 1055 struct file *fpop; 1056 struct mtx *mtxp; 1057 1058 if (vp->v_type == VBLK) 1059 return (ENXIO); 1060 1061 if (dev == NULL) 1062 return (ENXIO); 1063 1064 /* Make this field valid before any I/O in d_open. */ 1065 if (dev->si_iosize_max == 0) 1066 dev->si_iosize_max = DFLTPHYS; 1067 1068 dsw = dev_refthread(dev, &ref); 1069 if (dsw == NULL) 1070 return (ENXIO); 1071 if (fp == NULL && dsw->d_fdopen != NULL) { 1072 dev_relthread(dev, ref); 1073 return (ENXIO); 1074 } 1075 1076 vlocked = VOP_ISLOCKED(vp); 1077 VOP_UNLOCK(vp, 0); 1078 1079 fpop = td->td_fpop; 1080 td->td_fpop = fp; 1081 if (fp != NULL) { 1082 fp->f_data = dev; 1083 fp->f_vnode = vp; 1084 } 1085 if (dsw->d_fdopen != NULL) 1086 error = dsw->d_fdopen(dev, ap->a_mode, td, fp); 1087 else 1088 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td); 1089 /* cleanup any cdevpriv upon error */ 1090 if (error != 0) 1091 devfs_clear_cdevpriv(); 1092 td->td_fpop = fpop; 1093 1094 vn_lock(vp, vlocked | LK_RETRY); 1095 dev_relthread(dev, ref); 1096 if (error != 0) { 1097 if (error == ERESTART) 1098 error = EINTR; 1099 return (error); 1100 } 1101 1102#if 0 /* /dev/console */ 1103 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp")); 1104#else 1105 if (fp == NULL) 1106 return (error); 1107#endif 1108 if (fp->f_ops == &badfileops) 1109 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f); 1110 mtxp = mtx_pool_find(mtxpool_sleep, fp); 1111 1112 /* 1113 * Hint to the dofilewrite() to not force the buffer draining 1114 * on the writer to the file. Most likely, the write would 1115 * not need normal buffers. 1116 */ 1117 mtx_lock(mtxp); 1118 fp->f_vnread_flags |= FDEVFS_VNODE; 1119 mtx_unlock(mtxp); 1120 return (error); 1121} 1122 1123static int 1124devfs_pathconf(struct vop_pathconf_args *ap) 1125{ 1126 1127 switch (ap->a_name) { 1128 case _PC_MAC_PRESENT: 1129#ifdef MAC 1130 /* 1131 * If MAC is enabled, devfs automatically supports 1132 * trivial non-persistant label storage. 1133 */ 1134 *ap->a_retval = 1; 1135#else 1136 *ap->a_retval = 0; 1137#endif 1138 return (0); 1139 default: 1140 return (vop_stdpathconf(ap)); 1141 } 1142 /* NOTREACHED */ 1143} 1144 1145/* ARGSUSED */ 1146static int 1147devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td) 1148{ 1149 struct cdev *dev; 1150 struct cdevsw *dsw; 1151 int error, ref; 1152 struct file *fpop; 1153 1154 fpop = td->td_fpop; 1155 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1156 if (error != 0) { 1157 error = vnops.fo_poll(fp, events, cred, td); 1158 return (error); 1159 } 1160 error = dsw->d_poll(dev, events, td); 1161 td->td_fpop = fpop; 1162 dev_relthread(dev, ref); 1163 return(error); 1164} 1165 1166/* 1167 * Print out the contents of a special device vnode. 1168 */ 1169static int 1170devfs_print(struct vop_print_args *ap) 1171{ 1172 1173 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev)); 1174 return (0); 1175} 1176 1177static int 1178devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, 1179 int flags, struct thread *td) 1180{ 1181 struct cdev *dev; 1182 int ioflag, error, ref; 1183 ssize_t resid; 1184 struct cdevsw *dsw; 1185 struct file *fpop; 1186 1187 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1188 return (EINVAL); 1189 fpop = td->td_fpop; 1190 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1191 if (error != 0) { 1192 error = vnops.fo_read(fp, uio, cred, flags, td); 1193 return (error); 1194 } 1195 resid = uio->uio_resid; 1196 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT); 1197 if (ioflag & O_DIRECT) 1198 ioflag |= IO_DIRECT; 1199 1200 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1201 error = dsw->d_read(dev, uio, ioflag); 1202 if (uio->uio_resid != resid || (error == 0 && resid != 0)) 1203 vfs_timestamp(&dev->si_atime); 1204 td->td_fpop = fpop; 1205 dev_relthread(dev, ref); 1206 1207 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1208 return (error); 1209} 1210 1211static int 1212devfs_readdir(struct vop_readdir_args *ap) 1213{ 1214 int error; 1215 struct uio *uio; 1216 struct dirent *dp; 1217 struct devfs_dirent *dd; 1218 struct devfs_dirent *de; 1219 struct devfs_mount *dmp; 1220 off_t off; 1221 int *tmp_ncookies = NULL; 1222 1223 if (ap->a_vp->v_type != VDIR) 1224 return (ENOTDIR); 1225 1226 uio = ap->a_uio; 1227 if (uio->uio_offset < 0) 1228 return (EINVAL); 1229 1230 /* 1231 * XXX: This is a temporary hack to get around this filesystem not 1232 * supporting cookies. We store the location of the ncookies pointer 1233 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent() 1234 * and set the number of cookies to 0. We then set the pointer to 1235 * NULL so that vfs_read_dirent doesn't try to call realloc() on 1236 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies 1237 * pointer to its original location before returning to the caller. 1238 */ 1239 if (ap->a_ncookies != NULL) { 1240 tmp_ncookies = ap->a_ncookies; 1241 *ap->a_ncookies = 0; 1242 ap->a_ncookies = NULL; 1243 } 1244 1245 dmp = VFSTODEVFS(ap->a_vp->v_mount); 1246 if (devfs_populate_vp(ap->a_vp) != 0) { 1247 if (tmp_ncookies != NULL) 1248 ap->a_ncookies = tmp_ncookies; 1249 return (EIO); 1250 } 1251 error = 0; 1252 de = ap->a_vp->v_data; 1253 off = 0; 1254 TAILQ_FOREACH(dd, &de->de_dlist, de_list) { 1255 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__)); 1256 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT)) 1257 continue; 1258 if (devfs_prison_check(dd, uio->uio_td)) 1259 continue; 1260 if (dd->de_dirent->d_type == DT_DIR) 1261 de = dd->de_dir; 1262 else 1263 de = dd; 1264 dp = dd->de_dirent; 1265 if (dp->d_reclen > uio->uio_resid) 1266 break; 1267 dp->d_fileno = de->de_inode; 1268 if (off >= uio->uio_offset) { 1269 error = vfs_read_dirent(ap, dp, off); 1270 if (error) 1271 break; 1272 } 1273 off += dp->d_reclen; 1274 } 1275 sx_xunlock(&dmp->dm_lock); 1276 uio->uio_offset = off; 1277 1278 /* 1279 * Restore ap->a_ncookies if it wasn't originally NULL in the first 1280 * place. 1281 */ 1282 if (tmp_ncookies != NULL) 1283 ap->a_ncookies = tmp_ncookies; 1284 1285 return (error); 1286} 1287 1288static int 1289devfs_readlink(struct vop_readlink_args *ap) 1290{ 1291 struct devfs_dirent *de; 1292 1293 de = ap->a_vp->v_data; 1294 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio)); 1295} 1296 1297static int 1298devfs_reclaim(struct vop_reclaim_args *ap) 1299{ 1300 struct vnode *vp = ap->a_vp; 1301 struct devfs_dirent *de; 1302 struct cdev *dev; 1303 1304 mtx_lock(&devfs_de_interlock); 1305 de = vp->v_data; 1306 if (de != NULL) { 1307 de->de_vnode = NULL; 1308 vp->v_data = NULL; 1309 } 1310 mtx_unlock(&devfs_de_interlock); 1311 1312 vnode_destroy_vobject(vp); 1313 1314 VI_LOCK(vp); 1315 dev_lock(); 1316 dev = vp->v_rdev; 1317 vp->v_rdev = NULL; 1318 1319 if (dev == NULL) { 1320 dev_unlock(); 1321 VI_UNLOCK(vp); 1322 return (0); 1323 } 1324 1325 dev->si_usecount -= vp->v_usecount; 1326 dev_unlock(); 1327 VI_UNLOCK(vp); 1328 dev_rel(dev); 1329 return (0); 1330} 1331 1332static int 1333devfs_remove(struct vop_remove_args *ap) 1334{ 1335 struct vnode *dvp = ap->a_dvp; 1336 struct vnode *vp = ap->a_vp; 1337 struct devfs_dirent *dd; 1338 struct devfs_dirent *de, *de_covered; 1339 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount); 1340 1341 ASSERT_VOP_ELOCKED(dvp, "devfs_remove"); 1342 ASSERT_VOP_ELOCKED(vp, "devfs_remove"); 1343 1344 sx_xlock(&dmp->dm_lock); 1345 dd = ap->a_dvp->v_data; 1346 de = vp->v_data; 1347 if (de->de_cdp == NULL) { 1348 TAILQ_REMOVE(&dd->de_dlist, de, de_list); 1349 if (de->de_dirent->d_type == DT_LNK) { 1350 de_covered = devfs_find(dd, de->de_dirent->d_name, 1351 de->de_dirent->d_namlen, 0); 1352 if (de_covered != NULL) 1353 de_covered->de_flags &= ~DE_COVERED; 1354 } 1355 /* We need to unlock dvp because devfs_delete() may lock it. */ 1356 VOP_UNLOCK(vp, 0); 1357 if (dvp != vp) 1358 VOP_UNLOCK(dvp, 0); 1359 devfs_delete(dmp, de, 0); 1360 sx_xunlock(&dmp->dm_lock); 1361 if (dvp != vp) 1362 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 1363 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1364 } else { 1365 de->de_flags |= DE_WHITEOUT; 1366 sx_xunlock(&dmp->dm_lock); 1367 } 1368 return (0); 1369} 1370 1371/* 1372 * Revoke is called on a tty when a terminal session ends. The vnode 1373 * is orphaned by setting v_op to deadfs so we need to let go of it 1374 * as well so that we create a new one next time around. 1375 * 1376 */ 1377static int 1378devfs_revoke(struct vop_revoke_args *ap) 1379{ 1380 struct vnode *vp = ap->a_vp, *vp2; 1381 struct cdev *dev; 1382 struct cdev_priv *cdp; 1383 struct devfs_dirent *de; 1384 int i; 1385 1386 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL")); 1387 1388 dev = vp->v_rdev; 1389 cdp = cdev2priv(dev); 1390 1391 dev_lock(); 1392 cdp->cdp_inuse++; 1393 dev_unlock(); 1394 1395 vhold(vp); 1396 vgone(vp); 1397 vdrop(vp); 1398 1399 VOP_UNLOCK(vp,0); 1400 loop: 1401 for (;;) { 1402 mtx_lock(&devfs_de_interlock); 1403 dev_lock(); 1404 vp2 = NULL; 1405 for (i = 0; i <= cdp->cdp_maxdirent; i++) { 1406 de = cdp->cdp_dirents[i]; 1407 if (de == NULL) 1408 continue; 1409 1410 vp2 = de->de_vnode; 1411 if (vp2 != NULL) { 1412 dev_unlock(); 1413 VI_LOCK(vp2); 1414 mtx_unlock(&devfs_de_interlock); 1415 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK, 1416 curthread)) 1417 goto loop; 1418 vhold(vp2); 1419 vgone(vp2); 1420 vdrop(vp2); 1421 vput(vp2); 1422 break; 1423 } 1424 } 1425 if (vp2 != NULL) { 1426 continue; 1427 } 1428 dev_unlock(); 1429 mtx_unlock(&devfs_de_interlock); 1430 break; 1431 } 1432 dev_lock(); 1433 cdp->cdp_inuse--; 1434 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) { 1435 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list); 1436 dev_unlock(); 1437 dev_rel(&cdp->cdp_c); 1438 } else 1439 dev_unlock(); 1440 1441 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1442 return (0); 1443} 1444 1445static int 1446devfs_rioctl(struct vop_ioctl_args *ap) 1447{ 1448 struct vnode *vp; 1449 struct devfs_mount *dmp; 1450 int error; 1451 1452 vp = ap->a_vp; 1453 vn_lock(vp, LK_SHARED | LK_RETRY); 1454 if (vp->v_iflag & VI_DOOMED) { 1455 VOP_UNLOCK(vp, 0); 1456 return (EBADF); 1457 } 1458 dmp = VFSTODEVFS(vp->v_mount); 1459 sx_xlock(&dmp->dm_lock); 1460 VOP_UNLOCK(vp, 0); 1461 DEVFS_DMP_HOLD(dmp); 1462 devfs_populate(dmp); 1463 if (DEVFS_DMP_DROP(dmp)) { 1464 sx_xunlock(&dmp->dm_lock); 1465 devfs_unmount_final(dmp); 1466 return (ENOENT); 1467 } 1468 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td); 1469 sx_xunlock(&dmp->dm_lock); 1470 return (error); 1471} 1472 1473static int 1474devfs_rread(struct vop_read_args *ap) 1475{ 1476 1477 if (ap->a_vp->v_type != VDIR) 1478 return (EINVAL); 1479 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL)); 1480} 1481 1482static int 1483devfs_setattr(struct vop_setattr_args *ap) 1484{ 1485 struct devfs_dirent *de; 1486 struct vattr *vap; 1487 struct vnode *vp; 1488 struct thread *td; 1489 int c, error; 1490 uid_t uid; 1491 gid_t gid; 1492 1493 vap = ap->a_vap; 1494 vp = ap->a_vp; 1495 td = curthread; 1496 if ((vap->va_type != VNON) || 1497 (vap->va_nlink != VNOVAL) || 1498 (vap->va_fsid != VNOVAL) || 1499 (vap->va_fileid != VNOVAL) || 1500 (vap->va_blocksize != VNOVAL) || 1501 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1502 (vap->va_rdev != VNOVAL) || 1503 ((int)vap->va_bytes != VNOVAL) || 1504 (vap->va_gen != VNOVAL)) { 1505 return (EINVAL); 1506 } 1507 1508 de = vp->v_data; 1509 if (vp->v_type == VDIR) 1510 de = de->de_dir; 1511 1512 error = c = 0; 1513 if (vap->va_uid == (uid_t)VNOVAL) 1514 uid = de->de_uid; 1515 else 1516 uid = vap->va_uid; 1517 if (vap->va_gid == (gid_t)VNOVAL) 1518 gid = de->de_gid; 1519 else 1520 gid = vap->va_gid; 1521 if (uid != de->de_uid || gid != de->de_gid) { 1522 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid || 1523 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) { 1524 error = priv_check(td, PRIV_VFS_CHOWN); 1525 if (error) 1526 return (error); 1527 } 1528 de->de_uid = uid; 1529 de->de_gid = gid; 1530 c = 1; 1531 } 1532 1533 if (vap->va_mode != (mode_t)VNOVAL) { 1534 if (ap->a_cred->cr_uid != de->de_uid) { 1535 error = priv_check(td, PRIV_VFS_ADMIN); 1536 if (error) 1537 return (error); 1538 } 1539 de->de_mode = vap->va_mode; 1540 c = 1; 1541 } 1542 1543 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1544 error = vn_utimes_perm(vp, vap, ap->a_cred, td); 1545 if (error != 0) 1546 return (error); 1547 if (vap->va_atime.tv_sec != VNOVAL) { 1548 if (vp->v_type == VCHR) 1549 vp->v_rdev->si_atime = vap->va_atime; 1550 else 1551 de->de_atime = vap->va_atime; 1552 } 1553 if (vap->va_mtime.tv_sec != VNOVAL) { 1554 if (vp->v_type == VCHR) 1555 vp->v_rdev->si_mtime = vap->va_mtime; 1556 else 1557 de->de_mtime = vap->va_mtime; 1558 } 1559 c = 1; 1560 } 1561 1562 if (c) { 1563 if (vp->v_type == VCHR) 1564 vfs_timestamp(&vp->v_rdev->si_ctime); 1565 else 1566 vfs_timestamp(&de->de_mtime); 1567 } 1568 return (0); 1569} 1570 1571#ifdef MAC 1572static int 1573devfs_setlabel(struct vop_setlabel_args *ap) 1574{ 1575 struct vnode *vp; 1576 struct devfs_dirent *de; 1577 1578 vp = ap->a_vp; 1579 de = vp->v_data; 1580 1581 mac_vnode_relabel(ap->a_cred, vp, ap->a_label); 1582 mac_devfs_update(vp->v_mount, de, vp); 1583 1584 return (0); 1585} 1586#endif 1587 1588static int 1589devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td) 1590{ 1591 1592 return (vnops.fo_stat(fp, sb, cred, td)); 1593} 1594 1595static int 1596devfs_symlink(struct vop_symlink_args *ap) 1597{ 1598 int i, error; 1599 struct devfs_dirent *dd; 1600 struct devfs_dirent *de, *de_covered, *de_dotdot; 1601 struct devfs_mount *dmp; 1602 1603 error = priv_check(curthread, PRIV_DEVFS_SYMLINK); 1604 if (error) 1605 return(error); 1606 dmp = VFSTODEVFS(ap->a_dvp->v_mount); 1607 if (devfs_populate_vp(ap->a_dvp) != 0) 1608 return (ENOENT); 1609 1610 dd = ap->a_dvp->v_data; 1611 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen); 1612 de->de_flags = DE_USER; 1613 de->de_uid = 0; 1614 de->de_gid = 0; 1615 de->de_mode = 0755; 1616 de->de_inode = alloc_unr(devfs_inos); 1617 de->de_dir = dd; 1618 de->de_dirent->d_type = DT_LNK; 1619 i = strlen(ap->a_target) + 1; 1620 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK); 1621 bcopy(ap->a_target, de->de_symlink, i); 1622#ifdef MAC 1623 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de); 1624#endif 1625 de_covered = devfs_find(dd, de->de_dirent->d_name, 1626 de->de_dirent->d_namlen, 0); 1627 if (de_covered != NULL) { 1628 if ((de_covered->de_flags & DE_USER) != 0) { 1629 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE); 1630 sx_xunlock(&dmp->dm_lock); 1631 return (EEXIST); 1632 } 1633 KASSERT((de_covered->de_flags & DE_COVERED) == 0, 1634 ("devfs_symlink: entry %p already covered", de_covered)); 1635 de_covered->de_flags |= DE_COVERED; 1636 } 1637 1638 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */ 1639 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */ 1640 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list); 1641 devfs_dir_ref_de(dmp, dd); 1642 devfs_rules_apply(dmp, de); 1643 1644 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp)); 1645} 1646 1647static int 1648devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td) 1649{ 1650 1651 return (vnops.fo_truncate(fp, length, cred, td)); 1652} 1653 1654static int 1655devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, 1656 int flags, struct thread *td) 1657{ 1658 struct cdev *dev; 1659 int error, ioflag, ref; 1660 ssize_t resid; 1661 struct cdevsw *dsw; 1662 struct file *fpop; 1663 1664 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1665 return (EINVAL); 1666 fpop = td->td_fpop; 1667 error = devfs_fp_check(fp, &dev, &dsw, &ref); 1668 if (error != 0) { 1669 error = vnops.fo_write(fp, uio, cred, flags, td); 1670 return (error); 1671 } 1672 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); 1673 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC); 1674 if (ioflag & O_DIRECT) 1675 ioflag |= IO_DIRECT; 1676 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK); 1677 1678 resid = uio->uio_resid; 1679 1680 error = dsw->d_write(dev, uio, ioflag); 1681 if (uio->uio_resid != resid || (error == 0 && resid != 0)) { 1682 vfs_timestamp(&dev->si_ctime); 1683 dev->si_mtime = dev->si_ctime; 1684 } 1685 td->td_fpop = fpop; 1686 dev_relthread(dev, ref); 1687 1688 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF); 1689 return (error); 1690} 1691 1692dev_t 1693dev2udev(struct cdev *x) 1694{ 1695 if (x == NULL) 1696 return (NODEV); 1697 return (cdev2priv(x)->cdp_inode); 1698} 1699 1700static struct fileops devfs_ops_f = { 1701 .fo_read = devfs_read_f, 1702 .fo_write = devfs_write_f, 1703 .fo_truncate = devfs_truncate_f, 1704 .fo_ioctl = devfs_ioctl_f, 1705 .fo_poll = devfs_poll_f, 1706 .fo_kqfilter = devfs_kqfilter_f, 1707 .fo_stat = devfs_stat_f, 1708 .fo_close = devfs_close_f, 1709 .fo_chmod = vn_chmod, 1710 .fo_chown = vn_chown, 1711 .fo_sendfile = vn_sendfile, 1712 .fo_seek = vn_seek, 1713 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE 1714}; 1715 1716static struct vop_vector devfs_vnodeops = { 1717 .vop_default = &default_vnodeops, 1718 1719 .vop_access = devfs_access, 1720 .vop_getattr = devfs_getattr, 1721 .vop_ioctl = devfs_rioctl, 1722 .vop_lookup = devfs_lookup, 1723 .vop_mknod = devfs_mknod, 1724 .vop_pathconf = devfs_pathconf, 1725 .vop_read = devfs_rread, 1726 .vop_readdir = devfs_readdir, 1727 .vop_readlink = devfs_readlink, 1728 .vop_reclaim = devfs_reclaim, 1729 .vop_remove = devfs_remove, 1730 .vop_revoke = devfs_revoke, 1731 .vop_setattr = devfs_setattr, 1732#ifdef MAC 1733 .vop_setlabel = devfs_setlabel, 1734#endif 1735 .vop_symlink = devfs_symlink, 1736 .vop_vptocnp = devfs_vptocnp, 1737}; 1738 1739struct vop_vector devfs_specops = { 1740 .vop_default = &default_vnodeops, 1741 1742 .vop_access = devfs_access, 1743 .vop_bmap = VOP_PANIC, 1744 .vop_close = devfs_close, 1745 .vop_create = VOP_PANIC, 1746 .vop_fsync = devfs_fsync, 1747 .vop_getattr = devfs_getattr, 1748 .vop_link = VOP_PANIC, 1749 .vop_mkdir = VOP_PANIC, 1750 .vop_mknod = VOP_PANIC, 1751 .vop_open = devfs_open, 1752 .vop_pathconf = devfs_pathconf, 1753 .vop_poll = dead_poll, 1754 .vop_print = devfs_print, 1755 .vop_read = dead_read, 1756 .vop_readdir = VOP_PANIC, 1757 .vop_readlink = VOP_PANIC, 1758 .vop_reallocblks = VOP_PANIC, 1759 .vop_reclaim = devfs_reclaim, 1760 .vop_remove = devfs_remove, 1761 .vop_rename = VOP_PANIC, 1762 .vop_revoke = devfs_revoke, 1763 .vop_rmdir = VOP_PANIC, 1764 .vop_setattr = devfs_setattr, 1765#ifdef MAC 1766 .vop_setlabel = devfs_setlabel, 1767#endif 1768 .vop_strategy = VOP_PANIC, 1769 .vop_symlink = VOP_PANIC, 1770 .vop_vptocnp = devfs_vptocnp, 1771 .vop_write = dead_write, 1772}; 1773 1774/* 1775 * Our calling convention to the device drivers used to be that we passed 1776 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 1777 * flags instead since that's what open(), close() and ioctl() takes and 1778 * we don't really want vnode.h in device drivers. 1779 * We solved the source compatibility by redefining some vnode flags to 1780 * be the same as the fcntl ones and by sending down the bitwise OR of 1781 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody 1782 * pulls the rug out under this. 1783 */ 1784CTASSERT(O_NONBLOCK == IO_NDELAY); 1785CTASSERT(O_FSYNC == IO_SYNC); 1786