vfs_default.c revision 120514
1219820Sjeff/* 2219820Sjeff * Copyright (c) 1989, 1993 3219820Sjeff * The Regents of the University of California. All rights reserved. 4219820Sjeff * 5219820Sjeff * This code is derived from software contributed 6219820Sjeff * to Berkeley by John Heidemann of the UCLA Ficus project. 7219820Sjeff * 8219820Sjeff * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project 9219820Sjeff * 10219820Sjeff * Redistribution and use in source and binary forms, with or without 11219820Sjeff * modification, are permitted provided that the following conditions 12219820Sjeff * are met: 13219820Sjeff * 1. Redistributions of source code must retain the above copyright 14219820Sjeff * notice, this list of conditions and the following disclaimer. 15219820Sjeff * 2. Redistributions in binary form must reproduce the above copyright 16219820Sjeff * notice, this list of conditions and the following disclaimer in the 17219820Sjeff * documentation and/or other materials provided with the distribution. 18219820Sjeff * 3. All advertising materials mentioning features or use of this software 19219820Sjeff * must display the following acknowledgement: 20219820Sjeff * This product includes software developed by the University of 21219820Sjeff * California, Berkeley and its contributors. 22219820Sjeff * 4. Neither the name of the University nor the names of its contributors 23219820Sjeff * may be used to endorse or promote products derived from this software 24219820Sjeff * without specific prior written permission. 25219820Sjeff * 26219820Sjeff * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27219820Sjeff * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28219820Sjeff * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29219820Sjeff * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30219820Sjeff * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31219820Sjeff * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32219820Sjeff * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33219820Sjeff * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34219820Sjeff * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35219820Sjeff * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36219820Sjeff * SUCH DAMAGE. 37219820Sjeff */ 38219820Sjeff 39219820Sjeff#include <sys/cdefs.h> 40219820Sjeff__FBSDID("$FreeBSD: head/sys/kern/vfs_default.c 120514 2003-09-27 12:53:33Z phk $"); 41219820Sjeff 42219820Sjeff#include <sys/param.h> 43219820Sjeff#include <sys/systm.h> 44219820Sjeff#include <sys/bio.h> 45219820Sjeff#include <sys/buf.h> 46219820Sjeff#include <sys/conf.h> 47219820Sjeff#include <sys/kernel.h> 48219820Sjeff#include <sys/limits.h> 49219820Sjeff#include <sys/lock.h> 50219820Sjeff#include <sys/malloc.h> 51219820Sjeff#include <sys/mount.h> 52219820Sjeff#include <sys/mutex.h> 53219820Sjeff#include <sys/unistd.h> 54219820Sjeff#include <sys/vnode.h> 55219820Sjeff#include <sys/poll.h> 56219820Sjeff 57219820Sjeff#include <vm/vm.h> 58219820Sjeff#include <vm/vm_object.h> 59219820Sjeff#include <vm/vm_extern.h> 60219820Sjeff#include <vm/pmap.h> 61219820Sjeff#include <vm/vm_map.h> 62219820Sjeff#include <vm/vm_page.h> 63219820Sjeff#include <vm/vm_pager.h> 64219820Sjeff#include <vm/vnode_pager.h> 65219820Sjeff 66219820Sjeffstatic int vop_nolookup(struct vop_lookup_args *); 67219820Sjeffstatic int vop_nostrategy(struct vop_strategy_args *); 68219820Sjeff 69219820Sjeff/* 70219820Sjeff * This vnode table stores what we want to do if the filesystem doesn't 71219820Sjeff * implement a particular VOP. 72219820Sjeff * 73219820Sjeff * If there is no specific entry here, we will return EOPNOTSUPP. 74219820Sjeff * 75219820Sjeff */ 76219820Sjeff 77219820Sjeffvop_t **default_vnodeop_p; 78219820Sjeffstatic struct vnodeopv_entry_desc default_vnodeop_entries[] = { 79219820Sjeff { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 80219820Sjeff { &vop_advlock_desc, (vop_t *) vop_einval }, 81219820Sjeff { &vop_bmap_desc, (vop_t *) vop_stdbmap }, 82219820Sjeff { &vop_close_desc, (vop_t *) vop_null }, 83219820Sjeff { &vop_createvobject_desc, (vop_t *) vop_stdcreatevobject }, 84219820Sjeff { &vop_destroyvobject_desc, (vop_t *) vop_stddestroyvobject }, 85219820Sjeff { &vop_fsync_desc, (vop_t *) vop_null }, 86219820Sjeff { &vop_getpages_desc, (vop_t *) vop_stdgetpages }, 87219820Sjeff { &vop_getvobject_desc, (vop_t *) vop_stdgetvobject }, 88219820Sjeff { &vop_inactive_desc, (vop_t *) vop_stdinactive }, 89219820Sjeff { &vop_ioctl_desc, (vop_t *) vop_enotty }, 90219820Sjeff { &vop_islocked_desc, (vop_t *) vop_stdislocked }, 91219820Sjeff { &vop_lease_desc, (vop_t *) vop_null }, 92219820Sjeff { &vop_lock_desc, (vop_t *) vop_stdlock }, 93219820Sjeff { &vop_lookup_desc, (vop_t *) vop_nolookup }, 94219820Sjeff { &vop_open_desc, (vop_t *) vop_null }, 95219820Sjeff { &vop_pathconf_desc, (vop_t *) vop_einval }, 96219820Sjeff { &vop_poll_desc, (vop_t *) vop_nopoll }, 97219820Sjeff { &vop_putpages_desc, (vop_t *) vop_stdputpages }, 98219820Sjeff { &vop_readlink_desc, (vop_t *) vop_einval }, 99219820Sjeff { &vop_revoke_desc, (vop_t *) vop_revoke }, 100219820Sjeff { &vop_specstrategy_desc, (vop_t *) vop_panic }, 101219820Sjeff { &vop_strategy_desc, (vop_t *) vop_nostrategy }, 102219820Sjeff { &vop_unlock_desc, (vop_t *) vop_stdunlock }, 103219820Sjeff { NULL, NULL } 104219820Sjeff}; 105219820Sjeff 106219820Sjeffstatic struct vnodeopv_desc default_vnodeop_opv_desc = 107219820Sjeff { &default_vnodeop_p, default_vnodeop_entries }; 108219820Sjeff 109219820SjeffVNODEOP_SET(default_vnodeop_opv_desc); 110219820Sjeff 111219820Sjeff/* 112219820Sjeff * Series of placeholder functions for various error returns for 113219820Sjeff * VOPs. 114219820Sjeff */ 115219820Sjeff 116219820Sjeffint 117219820Sjeffvop_eopnotsupp(struct vop_generic_args *ap) 118219820Sjeff{ 119219820Sjeff /* 120219820Sjeff printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name); 121219820Sjeff */ 122219820Sjeff 123219820Sjeff return (EOPNOTSUPP); 124219820Sjeff} 125219820Sjeff 126219820Sjeffint 127219820Sjeffvop_ebadf(struct vop_generic_args *ap) 128219820Sjeff{ 129219820Sjeff 130219820Sjeff return (EBADF); 131219820Sjeff} 132219820Sjeff 133219820Sjeffint 134219820Sjeffvop_enotty(struct vop_generic_args *ap) 135219820Sjeff{ 136219820Sjeff 137219820Sjeff return (ENOTTY); 138219820Sjeff} 139219820Sjeff 140219820Sjeffint 141219820Sjeffvop_einval(struct vop_generic_args *ap) 142219820Sjeff{ 143219820Sjeff 144219820Sjeff return (EINVAL); 145219820Sjeff} 146219820Sjeff 147219820Sjeffint 148219820Sjeffvop_null(struct vop_generic_args *ap) 149219820Sjeff{ 150219820Sjeff 151219820Sjeff return (0); 152219820Sjeff} 153219820Sjeff 154219820Sjeff/* 155219820Sjeff * Used to make a defined VOP fall back to the default VOP. 156219820Sjeff */ 157219820Sjeffint 158219820Sjeffvop_defaultop(struct vop_generic_args *ap) 159219820Sjeff{ 160219820Sjeff 161219820Sjeff return (VOCALL(default_vnodeop_p, ap->a_desc->vdesc_offset, ap)); 162219820Sjeff} 163219820Sjeff 164219820Sjeff/* 165219820Sjeff * Helper function to panic on some bad VOPs in some filesystems. 166219820Sjeff */ 167219820Sjeffint 168219820Sjeffvop_panic(struct vop_generic_args *ap) 169219820Sjeff{ 170219820Sjeff 171219820Sjeff panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name); 172219820Sjeff} 173219820Sjeff 174219820Sjeff/* 175219820Sjeff * vop_std<something> and vop_no<something> are default functions for use by 176219820Sjeff * filesystems that need the "default reasonable" implementation for a 177219820Sjeff * particular operation. 178219820Sjeff * 179219820Sjeff * The documentation for the operations they implement exists (if it exists) 180219820Sjeff * in the VOP_<SOMETHING>(9) manpage (all uppercase). 181219820Sjeff */ 182219820Sjeff 183219820Sjeff/* 184219820Sjeff * Default vop for filesystems that do not support name lookup 185219820Sjeff */ 186219820Sjeffstatic int 187219820Sjeffvop_nolookup(ap) 188219820Sjeff struct vop_lookup_args /* { 189219820Sjeff struct vnode *a_dvp; 190219820Sjeff struct vnode **a_vpp; 191219820Sjeff struct componentname *a_cnp; 192219820Sjeff } */ *ap; 193219820Sjeff{ 194219820Sjeff 195219820Sjeff *ap->a_vpp = NULL; 196219820Sjeff return (ENOTDIR); 197219820Sjeff} 198219820Sjeff 199219820Sjeff/* 200219820Sjeff * vop_nostrategy: 201219820Sjeff * 202219820Sjeff * Strategy routine for VFS devices that have none. 203219820Sjeff * 204219820Sjeff * BIO_ERROR and B_INVAL must be cleared prior to calling any strategy 205219820Sjeff * routine. Typically this is done for a BIO_READ strategy call. 206219820Sjeff * Typically B_INVAL is assumed to already be clear prior to a write 207219820Sjeff * and should not be cleared manually unless you just made the buffer 208219820Sjeff * invalid. BIO_ERROR should be cleared either way. 209219820Sjeff */ 210219820Sjeff 211219820Sjeffstatic int 212219820Sjeffvop_nostrategy (struct vop_strategy_args *ap) 213219820Sjeff{ 214219820Sjeff KASSERT(ap->a_vp == ap->a_bp->b_vp, ("%s(%p != %p)", 215219820Sjeff __func__, ap->a_vp, ap->a_bp->b_vp)); 216219820Sjeff printf("No strategy for buffer at %p\n", ap->a_bp); 217219820Sjeff vprint("vnode", ap->a_vp); 218219820Sjeff vprint("device vnode", ap->a_bp->b_vp); 219219820Sjeff ap->a_bp->b_ioflags |= BIO_ERROR; 220219820Sjeff ap->a_bp->b_error = EOPNOTSUPP; 221219820Sjeff bufdone(ap->a_bp); 222219820Sjeff return (EOPNOTSUPP); 223219820Sjeff} 224219820Sjeff 225219820Sjeff/* 226219820Sjeff * vop_stdpathconf: 227219820Sjeff * 228219820Sjeff * Standard implementation of POSIX pathconf, to get information about limits 229219820Sjeff * for a filesystem. 230219820Sjeff * Override per filesystem for the case where the filesystem has smaller 231219820Sjeff * limits. 232219820Sjeff */ 233219820Sjeffint 234219820Sjeffvop_stdpathconf(ap) 235219820Sjeff struct vop_pathconf_args /* { 236219820Sjeff struct vnode *a_vp; 237219820Sjeff int a_name; 238219820Sjeff int *a_retval; 239219820Sjeff } */ *ap; 240219820Sjeff{ 241219820Sjeff 242219820Sjeff switch (ap->a_name) { 243219820Sjeff case _PC_LINK_MAX: 244219820Sjeff *ap->a_retval = LINK_MAX; 245219820Sjeff return (0); 246219820Sjeff case _PC_MAX_CANON: 247219820Sjeff *ap->a_retval = MAX_CANON; 248219820Sjeff return (0); 249219820Sjeff case _PC_MAX_INPUT: 250219820Sjeff *ap->a_retval = MAX_INPUT; 251219820Sjeff return (0); 252219820Sjeff case _PC_PIPE_BUF: 253219820Sjeff *ap->a_retval = PIPE_BUF; 254219820Sjeff return (0); 255 case _PC_CHOWN_RESTRICTED: 256 *ap->a_retval = 1; 257 return (0); 258 case _PC_VDISABLE: 259 *ap->a_retval = _POSIX_VDISABLE; 260 return (0); 261 default: 262 return (EINVAL); 263 } 264 /* NOTREACHED */ 265} 266 267/* 268 * Standard lock, unlock and islocked functions. 269 */ 270int 271vop_stdlock(ap) 272 struct vop_lock_args /* { 273 struct vnode *a_vp; 274 int a_flags; 275 struct thread *a_td; 276 } */ *ap; 277{ 278 struct vnode *vp = ap->a_vp; 279 280#ifndef DEBUG_LOCKS 281 return (lockmgr(vp->v_vnlock, ap->a_flags, VI_MTX(vp), ap->a_td)); 282#else 283 return (debuglockmgr(vp->v_vnlock, ap->a_flags, VI_MTX(vp), 284 ap->a_td, "vop_stdlock", vp->filename, vp->line)); 285#endif 286} 287 288/* See above. */ 289int 290vop_stdunlock(ap) 291 struct vop_unlock_args /* { 292 struct vnode *a_vp; 293 int a_flags; 294 struct thread *a_td; 295 } */ *ap; 296{ 297 struct vnode *vp = ap->a_vp; 298 299 return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE, VI_MTX(vp), 300 ap->a_td)); 301} 302 303/* See above. */ 304int 305vop_stdislocked(ap) 306 struct vop_islocked_args /* { 307 struct vnode *a_vp; 308 struct thread *a_td; 309 } */ *ap; 310{ 311 312 return (lockstatus(ap->a_vp->v_vnlock, ap->a_td)); 313} 314 315/* Mark the vnode inactive */ 316int 317vop_stdinactive(ap) 318 struct vop_inactive_args /* { 319 struct vnode *a_vp; 320 struct thread *a_td; 321 } */ *ap; 322{ 323 324 VOP_UNLOCK(ap->a_vp, 0, ap->a_td); 325 return (0); 326} 327 328/* 329 * Return true for select/poll. 330 */ 331int 332vop_nopoll(ap) 333 struct vop_poll_args /* { 334 struct vnode *a_vp; 335 int a_events; 336 struct ucred *a_cred; 337 struct thread *a_td; 338 } */ *ap; 339{ 340 /* 341 * Return true for read/write. If the user asked for something 342 * special, return POLLNVAL, so that clients have a way of 343 * determining reliably whether or not the extended 344 * functionality is present without hard-coding knowledge 345 * of specific filesystem implementations. 346 * Stay in sync with kern_conf.c::no_poll(). 347 */ 348 if (ap->a_events & ~POLLSTANDARD) 349 return (POLLNVAL); 350 351 return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); 352} 353 354/* 355 * Implement poll for local filesystems that support it. 356 */ 357int 358vop_stdpoll(ap) 359 struct vop_poll_args /* { 360 struct vnode *a_vp; 361 int a_events; 362 struct ucred *a_cred; 363 struct thread *a_td; 364 } */ *ap; 365{ 366 if (ap->a_events & ~POLLSTANDARD) 367 return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events)); 368 return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); 369} 370 371/* 372 * Stubs to use when there is no locking to be done on the underlying object. 373 * A minimal shared lock is necessary to ensure that the underlying object 374 * is not revoked while an operation is in progress. So, an active shared 375 * count is maintained in an auxillary vnode lock structure. 376 */ 377int 378vop_sharedlock(ap) 379 struct vop_lock_args /* { 380 struct vnode *a_vp; 381 int a_flags; 382 struct thread *a_td; 383 } */ *ap; 384{ 385 /* 386 * This code cannot be used until all the non-locking filesystems 387 * (notably NFS) are converted to properly lock and release nodes. 388 * Also, certain vnode operations change the locking state within 389 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 390 * and symlink). Ideally these operations should not change the 391 * lock state, but should be changed to let the caller of the 392 * function unlock them. Otherwise all intermediate vnode layers 393 * (such as union, umapfs, etc) must catch these functions to do 394 * the necessary locking at their layer. Note that the inactive 395 * and lookup operations also change their lock state, but this 396 * cannot be avoided, so these two operations will always need 397 * to be handled in intermediate layers. 398 */ 399 struct vnode *vp = ap->a_vp; 400 int vnflags, flags = ap->a_flags; 401 402 switch (flags & LK_TYPE_MASK) { 403 case LK_DRAIN: 404 vnflags = LK_DRAIN; 405 break; 406 case LK_EXCLUSIVE: 407#ifdef DEBUG_VFS_LOCKS 408 /* 409 * Normally, we use shared locks here, but that confuses 410 * the locking assertions. 411 */ 412 vnflags = LK_EXCLUSIVE; 413 break; 414#endif 415 case LK_SHARED: 416 vnflags = LK_SHARED; 417 break; 418 case LK_UPGRADE: 419 case LK_EXCLUPGRADE: 420 case LK_DOWNGRADE: 421 return (0); 422 case LK_RELEASE: 423 default: 424 panic("vop_sharedlock: bad operation %d", flags & LK_TYPE_MASK); 425 } 426 vnflags |= flags & (LK_INTERLOCK | LK_EXTFLG_MASK); 427#ifndef DEBUG_LOCKS 428 return (lockmgr(vp->v_vnlock, vnflags, VI_MTX(vp), ap->a_td)); 429#else 430 return (debuglockmgr(vp->v_vnlock, vnflags, VI_MTX(vp), ap->a_td, 431 "vop_sharedlock", vp->filename, vp->line)); 432#endif 433} 434 435/* 436 * Stubs to use when there is no locking to be done on the underlying object. 437 * A minimal shared lock is necessary to ensure that the underlying object 438 * is not revoked while an operation is in progress. So, an active shared 439 * count is maintained in an auxillary vnode lock structure. 440 */ 441int 442vop_nolock(ap) 443 struct vop_lock_args /* { 444 struct vnode *a_vp; 445 int a_flags; 446 struct thread *a_td; 447 } */ *ap; 448{ 449#ifdef notyet 450 /* 451 * This code cannot be used until all the non-locking filesystems 452 * (notably NFS) are converted to properly lock and release nodes. 453 * Also, certain vnode operations change the locking state within 454 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 455 * and symlink). Ideally these operations should not change the 456 * lock state, but should be changed to let the caller of the 457 * function unlock them. Otherwise all intermediate vnode layers 458 * (such as union, umapfs, etc) must catch these functions to do 459 * the necessary locking at their layer. Note that the inactive 460 * and lookup operations also change their lock state, but this 461 * cannot be avoided, so these two operations will always need 462 * to be handled in intermediate layers. 463 */ 464 struct vnode *vp = ap->a_vp; 465 int vnflags, flags = ap->a_flags; 466 467 switch (flags & LK_TYPE_MASK) { 468 case LK_DRAIN: 469 vnflags = LK_DRAIN; 470 break; 471 case LK_EXCLUSIVE: 472 case LK_SHARED: 473 vnflags = LK_SHARED; 474 break; 475 case LK_UPGRADE: 476 case LK_EXCLUPGRADE: 477 case LK_DOWNGRADE: 478 return (0); 479 case LK_RELEASE: 480 default: 481 panic("vop_nolock: bad operation %d", flags & LK_TYPE_MASK); 482 } 483 vnflags |= flags & (LK_INTERLOCK | LK_EXTFLG_MASK); 484 return(lockmgr(vp->v_vnlock, vnflags, VI_MTX(vp), ap->a_td)); 485#else /* for now */ 486 /* 487 * Since we are not using the lock manager, we must clear 488 * the interlock here. 489 */ 490 if (ap->a_flags & LK_INTERLOCK) 491 VI_UNLOCK(ap->a_vp); 492 return (0); 493#endif 494} 495 496/* 497 * Do the inverse of vop_nolock, handling the interlock in a compatible way. 498 */ 499int 500vop_nounlock(ap) 501 struct vop_unlock_args /* { 502 struct vnode *a_vp; 503 int a_flags; 504 struct thread *a_td; 505 } */ *ap; 506{ 507 508 /* 509 * Since we are not using the lock manager, we must clear 510 * the interlock here. 511 */ 512 if (ap->a_flags & LK_INTERLOCK) 513 VI_UNLOCK(ap->a_vp); 514 return (0); 515} 516 517/* 518 * Return whether or not the node is in use. 519 */ 520int 521vop_noislocked(ap) 522 struct vop_islocked_args /* { 523 struct vnode *a_vp; 524 struct thread *a_td; 525 } */ *ap; 526{ 527 528 return (0); 529} 530 531/* 532 * Return our mount point, as we will take charge of the writes. 533 */ 534int 535vop_stdgetwritemount(ap) 536 struct vop_getwritemount_args /* { 537 struct vnode *a_vp; 538 struct mount **a_mpp; 539 } */ *ap; 540{ 541 542 *(ap->a_mpp) = ap->a_vp->v_mount; 543 return (0); 544} 545 546/* Create the VM system backing object for this vnode */ 547int 548vop_stdcreatevobject(ap) 549 struct vop_createvobject_args /* { 550 struct vnode *vp; 551 struct ucred *cred; 552 struct thread *td; 553 } */ *ap; 554{ 555 struct vnode *vp = ap->a_vp; 556 struct ucred *cred = ap->a_cred; 557 struct thread *td = ap->a_td; 558 struct vattr vat; 559 vm_object_t object; 560 int error = 0; 561 562 GIANT_REQUIRED; 563 564 if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE) 565 return (0); 566 567retry: 568 if ((object = vp->v_object) == NULL) { 569 if (vp->v_type == VREG || vp->v_type == VDIR) { 570 if ((error = VOP_GETATTR(vp, &vat, cred, td)) != 0) 571 goto retn; 572 object = vnode_pager_alloc(vp, vat.va_size, 0, 0); 573 } else if (devsw(vp->v_rdev) != NULL) { 574 /* 575 * This simply allocates the biggest object possible 576 * for a disk vnode. This should be fixed, but doesn't 577 * cause any problems (yet). 578 */ 579 object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0); 580 } else { 581 goto retn; 582 } 583 /* 584 * Dereference the reference we just created. This assumes 585 * that the object is associated with the vp. 586 */ 587 VM_OBJECT_LOCK(object); 588 object->ref_count--; 589 VM_OBJECT_UNLOCK(object); 590 vrele(vp); 591 } else { 592 VM_OBJECT_LOCK(object); 593 if (object->flags & OBJ_DEAD) { 594 VOP_UNLOCK(vp, 0, td); 595 msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, 596 "vodead", 0); 597 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 598 goto retry; 599 } 600 VM_OBJECT_UNLOCK(object); 601 } 602 603 KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object")); 604 vp->v_vflag |= VV_OBJBUF; 605 606retn: 607 return (error); 608} 609 610/* Destroy the VM system object associated with this vnode */ 611int 612vop_stddestroyvobject(ap) 613 struct vop_destroyvobject_args /* { 614 struct vnode *vp; 615 } */ *ap; 616{ 617 struct vnode *vp = ap->a_vp; 618 vm_object_t obj = vp->v_object; 619 620 GIANT_REQUIRED; 621 622 if (obj == NULL) 623 return (0); 624 VM_OBJECT_LOCK(obj); 625 if (obj->ref_count == 0) { 626 /* 627 * vclean() may be called twice. The first time 628 * removes the primary reference to the object, 629 * the second time goes one further and is a 630 * special-case to terminate the object. 631 * 632 * don't double-terminate the object 633 */ 634 if ((obj->flags & OBJ_DEAD) == 0) 635 vm_object_terminate(obj); 636 else 637 VM_OBJECT_UNLOCK(obj); 638 } else { 639 /* 640 * Woe to the process that tries to page now :-). 641 */ 642 vm_pager_deallocate(obj); 643 VM_OBJECT_UNLOCK(obj); 644 } 645 return (0); 646} 647 648/* 649 * Return the underlying VM object. This routine may be called with or 650 * without the vnode interlock held. If called without, the returned 651 * object is not guarenteed to be valid. The syncer typically gets the 652 * object without holding the interlock in order to quickly test whether 653 * it might be dirty before going heavy-weight. vm_object's use zalloc 654 * and thus stable-storage, so this is safe. 655 */ 656int 657vop_stdgetvobject(ap) 658 struct vop_getvobject_args /* { 659 struct vnode *vp; 660 struct vm_object **objpp; 661 } */ *ap; 662{ 663 struct vnode *vp = ap->a_vp; 664 struct vm_object **objpp = ap->a_objpp; 665 666 if (objpp) 667 *objpp = vp->v_object; 668 return (vp->v_object ? 0 : EINVAL); 669} 670 671/* XXX Needs good comment and VOP_BMAP(9) manpage */ 672int 673vop_stdbmap(ap) 674 struct vop_bmap_args /* { 675 struct vnode *a_vp; 676 daddr_t a_bn; 677 struct vnode **a_vpp; 678 daddr_t *a_bnp; 679 int *a_runp; 680 int *a_runb; 681 } */ *ap; 682{ 683 684 if (ap->a_vpp != NULL) 685 *ap->a_vpp = ap->a_vp; 686 if (ap->a_bnp != NULL) 687 *ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize); 688 if (ap->a_runp != NULL) 689 *ap->a_runp = 0; 690 if (ap->a_runb != NULL) 691 *ap->a_runb = 0; 692 return (0); 693} 694 695int 696vop_stdfsync(ap) 697 struct vop_fsync_args /* { 698 struct vnode *a_vp; 699 struct ucred *a_cred; 700 int a_waitfor; 701 struct thread *a_td; 702 } */ *ap; 703{ 704 struct vnode *vp = ap->a_vp; 705 struct buf *bp; 706 struct buf *nbp; 707 int s, error = 0; 708 int maxretry = 100; /* large, arbitrarily chosen */ 709 710 VI_LOCK(vp); 711loop1: 712 /* 713 * MARK/SCAN initialization to avoid infinite loops. 714 */ 715 s = splbio(); 716 TAILQ_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 717 bp->b_vflags &= ~BV_SCANNED; 718 bp->b_error = 0; 719 } 720 splx(s); 721 722 /* 723 * Flush all dirty buffers associated with a block device. 724 */ 725loop2: 726 s = splbio(); 727 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp != NULL; bp = nbp) { 728 nbp = TAILQ_NEXT(bp, b_vnbufs); 729 if ((bp->b_vflags & BV_SCANNED) != 0) 730 continue; 731 bp->b_vflags |= BV_SCANNED; 732 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) 733 continue; 734 VI_UNLOCK(vp); 735 if ((bp->b_flags & B_DELWRI) == 0) 736 panic("fsync: not dirty"); 737 if ((vp->v_vflag & VV_OBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 738 vfs_bio_awrite(bp); 739 splx(s); 740 } else { 741 bremfree(bp); 742 splx(s); 743 bawrite(bp); 744 } 745 VI_LOCK(vp); 746 goto loop2; 747 } 748 749 /* 750 * If synchronous the caller expects us to completely resolve all 751 * dirty buffers in the system. Wait for in-progress I/O to 752 * complete (which could include background bitmap writes), then 753 * retry if dirty blocks still exist. 754 */ 755 if (ap->a_waitfor == MNT_WAIT) { 756 while (vp->v_numoutput) { 757 vp->v_iflag |= VI_BWAIT; 758 msleep((caddr_t)&vp->v_numoutput, VI_MTX(vp), 759 PRIBIO + 1, "fsync", 0); 760 } 761 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 762 /* 763 * If we are unable to write any of these buffers 764 * then we fail now rather than trying endlessly 765 * to write them out. 766 */ 767 TAILQ_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) 768 if ((error = bp->b_error) == 0) 769 continue; 770 if (error == 0 && --maxretry >= 0) { 771 splx(s); 772 goto loop1; 773 } 774 vprint("fsync: giving up on dirty", vp); 775 error = EAGAIN; 776 } 777 } 778 VI_UNLOCK(vp); 779 splx(s); 780 781 return (error); 782} 783 784/* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */ 785int 786vop_stdgetpages(ap) 787 struct vop_getpages_args /* { 788 struct vnode *a_vp; 789 vm_page_t *a_m; 790 int a_count; 791 int a_reqpage; 792 vm_ooffset_t a_offset; 793 } */ *ap; 794{ 795 796 return vnode_pager_generic_getpages(ap->a_vp, ap->a_m, 797 ap->a_count, ap->a_reqpage); 798} 799 800/* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */ 801int 802vop_stdputpages(ap) 803 struct vop_putpages_args /* { 804 struct vnode *a_vp; 805 vm_page_t *a_m; 806 int a_count; 807 int a_sync; 808 int *a_rtvals; 809 vm_ooffset_t a_offset; 810 } */ *ap; 811{ 812 813 return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count, 814 ap->a_sync, ap->a_rtvals); 815} 816 817/* 818 * vfs default ops 819 * used to fill the vfs function table to get reasonable default return values. 820 */ 821int 822vfs_stdroot (mp, vpp) 823 struct mount *mp; 824 struct vnode **vpp; 825{ 826 return (EOPNOTSUPP); 827} 828 829int 830vfs_stdstatfs (mp, sbp, td) 831 struct mount *mp; 832 struct statfs *sbp; 833 struct thread *td; 834{ 835 return (EOPNOTSUPP); 836} 837 838int 839vfs_stdvptofh (vp, fhp) 840 struct vnode *vp; 841 struct fid *fhp; 842{ 843 return (EOPNOTSUPP); 844} 845 846int 847vfs_stdstart (mp, flags, td) 848 struct mount *mp; 849 int flags; 850 struct thread *td; 851{ 852 return (0); 853} 854 855int 856vfs_stdquotactl (mp, cmds, uid, arg, td) 857 struct mount *mp; 858 int cmds; 859 uid_t uid; 860 caddr_t arg; 861 struct thread *td; 862{ 863 return (EOPNOTSUPP); 864} 865 866int 867vfs_stdsync(mp, waitfor, cred, td) 868 struct mount *mp; 869 int waitfor; 870 struct ucred *cred; 871 struct thread *td; 872{ 873 struct vnode *vp, *nvp; 874 int error, lockreq, allerror = 0; 875 876 lockreq = LK_EXCLUSIVE | LK_INTERLOCK; 877 if (waitfor != MNT_WAIT) 878 lockreq |= LK_NOWAIT; 879 /* 880 * Force stale buffer cache information to be flushed. 881 */ 882 mtx_lock(&mntvnode_mtx); 883loop: 884 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) { 885 /* 886 * If the vnode that we are about to sync is no longer 887 * associated with this mount point, start over. 888 */ 889 if (vp->v_mount != mp) 890 goto loop; 891 892 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 893 894 VI_LOCK(vp); 895 if (TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 896 VI_UNLOCK(vp); 897 continue; 898 } 899 mtx_unlock(&mntvnode_mtx); 900 901 if ((error = vget(vp, lockreq, td)) != 0) { 902 if (error == ENOENT) 903 goto loop; 904 continue; 905 } 906 error = VOP_FSYNC(vp, cred, waitfor, td); 907 if (error) 908 allerror = error; 909 910 mtx_lock(&mntvnode_mtx); 911 if (nvp != TAILQ_NEXT(vp, v_nmntvnodes)) { 912 vput(vp); 913 goto loop; 914 } 915 vput(vp); 916 } 917 mtx_unlock(&mntvnode_mtx); 918 return (allerror); 919} 920 921int 922vfs_stdnosync (mp, waitfor, cred, td) 923 struct mount *mp; 924 int waitfor; 925 struct ucred *cred; 926 struct thread *td; 927{ 928 return (0); 929} 930 931int 932vfs_stdvget (mp, ino, flags, vpp) 933 struct mount *mp; 934 ino_t ino; 935 int flags; 936 struct vnode **vpp; 937{ 938 return (EOPNOTSUPP); 939} 940 941int 942vfs_stdfhtovp (mp, fhp, vpp) 943 struct mount *mp; 944 struct fid *fhp; 945 struct vnode **vpp; 946{ 947 return (EOPNOTSUPP); 948} 949 950int 951vfs_stdinit (vfsp) 952 struct vfsconf *vfsp; 953{ 954 return (0); 955} 956 957int 958vfs_stduninit (vfsp) 959 struct vfsconf *vfsp; 960{ 961 return(0); 962} 963 964int 965vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname, td) 966 struct mount *mp; 967 int cmd; 968 struct vnode *filename_vp; 969 int attrnamespace; 970 const char *attrname; 971 struct thread *td; 972{ 973 if (filename_vp != NULL) 974 VOP_UNLOCK(filename_vp, 0, td); 975 return(EOPNOTSUPP); 976} 977 978/* end of vfs default ops */ 979