xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 78adacd4eab39a3508bd8c65f0aba94fc6b907ce)
1 /*-
2  * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause)
3  *
4  * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
5  * All rights reserved.
6  *
7  * This software was developed for the FreeBSD Project by Marshall
8  * Kirk McKusick and Network Associates Laboratories, the Security
9  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
10  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
11  * research program
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * Copyright (c) 1982, 1986, 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	from: @(#)ufs_readwrite.c	8.11 (Berkeley) 5/8/95
62  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
63  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
64  */
65 
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
68 
69 #include <sys/param.h>
70 #include <sys/bio.h>
71 #include <sys/systm.h>
72 #include <sys/buf.h>
73 #include <sys/conf.h>
74 #include <sys/extattr.h>
75 #include <sys/kernel.h>
76 #include <sys/limits.h>
77 #include <sys/malloc.h>
78 #include <sys/mount.h>
79 #include <sys/priv.h>
80 #include <sys/rwlock.h>
81 #include <sys/stat.h>
82 #include <sys/sysctl.h>
83 #include <sys/vmmeter.h>
84 #include <sys/vnode.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_param.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vnode_pager.h>
93 
94 #include <ufs/ufs/extattr.h>
95 #include <ufs/ufs/quota.h>
96 #include <ufs/ufs/inode.h>
97 #include <ufs/ufs/ufs_extern.h>
98 #include <ufs/ufs/ufsmount.h>
99 
100 #include <ufs/ffs/fs.h>
101 #include <ufs/ffs/ffs_extern.h>
102 #include "opt_directio.h"
103 #include "opt_ffs.h"
104 
105 #define	ALIGNED_TO(ptr, s)	\
106 	(((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0)
107 
108 #ifdef DIRECTIO
109 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
110 #endif
111 static vop_fdatasync_t	ffs_fdatasync;
112 static vop_fsync_t	ffs_fsync;
113 static vop_getpages_t	ffs_getpages;
114 static vop_getpages_async_t	ffs_getpages_async;
115 static vop_lock1_t	ffs_lock;
116 #ifdef INVARIANTS
117 static vop_unlock_t	ffs_unlock_debug;
118 #endif
119 static vop_read_t	ffs_read;
120 static vop_write_t	ffs_write;
121 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
122 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
123 		    struct ucred *cred);
124 static vop_strategy_t	ffsext_strategy;
125 static vop_closeextattr_t	ffs_closeextattr;
126 static vop_deleteextattr_t	ffs_deleteextattr;
127 static vop_getextattr_t	ffs_getextattr;
128 static vop_listextattr_t	ffs_listextattr;
129 static vop_openextattr_t	ffs_openextattr;
130 static vop_setextattr_t	ffs_setextattr;
131 static vop_vptofh_t	ffs_vptofh;
132 
133 /* Global vfs data structures for ufs. */
134 struct vop_vector ffs_vnodeops1 = {
135 	.vop_default =		&ufs_vnodeops,
136 	.vop_fsync =		ffs_fsync,
137 	.vop_fdatasync =	ffs_fdatasync,
138 	.vop_getpages =		ffs_getpages,
139 	.vop_getpages_async =	ffs_getpages_async,
140 	.vop_lock1 =		ffs_lock,
141 #ifdef INVARIANTS
142 	.vop_unlock =		ffs_unlock_debug,
143 #endif
144 	.vop_read =		ffs_read,
145 	.vop_reallocblks =	ffs_reallocblks,
146 	.vop_write =		ffs_write,
147 	.vop_vptofh =		ffs_vptofh,
148 };
149 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1);
150 
151 struct vop_vector ffs_fifoops1 = {
152 	.vop_default =		&ufs_fifoops,
153 	.vop_fsync =		ffs_fsync,
154 	.vop_fdatasync =	ffs_fdatasync,
155 	.vop_lock1 =		ffs_lock,
156 #ifdef INVARIANTS
157 	.vop_unlock =		ffs_unlock_debug,
158 #endif
159 	.vop_vptofh =		ffs_vptofh,
160 };
161 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1);
162 
163 /* Global vfs data structures for ufs. */
164 struct vop_vector ffs_vnodeops2 = {
165 	.vop_default =		&ufs_vnodeops,
166 	.vop_fsync =		ffs_fsync,
167 	.vop_fdatasync =	ffs_fdatasync,
168 	.vop_getpages =		ffs_getpages,
169 	.vop_getpages_async =	ffs_getpages_async,
170 	.vop_lock1 =		ffs_lock,
171 #ifdef INVARIANTS
172 	.vop_unlock =		ffs_unlock_debug,
173 #endif
174 	.vop_read =		ffs_read,
175 	.vop_reallocblks =	ffs_reallocblks,
176 	.vop_write =		ffs_write,
177 	.vop_closeextattr =	ffs_closeextattr,
178 	.vop_deleteextattr =	ffs_deleteextattr,
179 	.vop_getextattr =	ffs_getextattr,
180 	.vop_listextattr =	ffs_listextattr,
181 	.vop_openextattr =	ffs_openextattr,
182 	.vop_setextattr =	ffs_setextattr,
183 	.vop_vptofh =		ffs_vptofh,
184 };
185 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2);
186 
187 struct vop_vector ffs_fifoops2 = {
188 	.vop_default =		&ufs_fifoops,
189 	.vop_fsync =		ffs_fsync,
190 	.vop_fdatasync =	ffs_fdatasync,
191 	.vop_lock1 =		ffs_lock,
192 #ifdef INVARIANTS
193 	.vop_unlock =		ffs_unlock_debug,
194 #endif
195 	.vop_reallocblks =	ffs_reallocblks,
196 	.vop_strategy =		ffsext_strategy,
197 	.vop_closeextattr =	ffs_closeextattr,
198 	.vop_deleteextattr =	ffs_deleteextattr,
199 	.vop_getextattr =	ffs_getextattr,
200 	.vop_listextattr =	ffs_listextattr,
201 	.vop_openextattr =	ffs_openextattr,
202 	.vop_setextattr =	ffs_setextattr,
203 	.vop_vptofh =		ffs_vptofh,
204 };
205 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2);
206 
207 /*
208  * Synch an open file.
209  */
210 /* ARGSUSED */
211 static int
212 ffs_fsync(struct vop_fsync_args *ap)
213 {
214 	struct vnode *vp;
215 	struct bufobj *bo;
216 	int error;
217 
218 	vp = ap->a_vp;
219 	bo = &vp->v_bufobj;
220 retry:
221 	error = ffs_syncvnode(vp, ap->a_waitfor, 0);
222 	if (error)
223 		return (error);
224 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
225 		error = softdep_fsync(vp);
226 		if (error)
227 			return (error);
228 
229 		/*
230 		 * The softdep_fsync() function may drop vp lock,
231 		 * allowing for dirty buffers to reappear on the
232 		 * bo_dirty list. Recheck and resync as needed.
233 		 */
234 		BO_LOCK(bo);
235 		if ((vp->v_type == VREG || vp->v_type == VDIR) &&
236 		    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
237 			BO_UNLOCK(bo);
238 			goto retry;
239 		}
240 		BO_UNLOCK(bo);
241 	}
242 	if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), 0))
243 		return (ENXIO);
244 	return (0);
245 }
246 
247 int
248 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
249 {
250 	struct inode *ip;
251 	struct bufobj *bo;
252 	struct ufsmount *ump;
253 	struct buf *bp, *nbp;
254 	ufs_lbn_t lbn;
255 	int error, passes;
256 	bool still_dirty, unlocked, wait;
257 
258 	ip = VTOI(vp);
259 	ip->i_flag &= ~IN_NEEDSYNC;
260 	bo = &vp->v_bufobj;
261 	ump = VFSTOUFS(vp->v_mount);
262 
263 	/*
264 	 * When doing MNT_WAIT we must first flush all dependencies
265 	 * on the inode.
266 	 */
267 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
268 	    (error = softdep_sync_metadata(vp)) != 0) {
269 		if (ffs_fsfail_cleanup(ump, error))
270 			error = 0;
271 		return (error);
272 	}
273 
274 	/*
275 	 * Flush all dirty buffers associated with a vnode.
276 	 */
277 	error = 0;
278 	passes = 0;
279 	wait = false;	/* Always do an async pass first. */
280 	unlocked = false;
281 	lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
282 	BO_LOCK(bo);
283 loop:
284 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
285 		bp->b_vflags &= ~BV_SCANNED;
286 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
287 		/*
288 		 * Reasons to skip this buffer: it has already been considered
289 		 * on this pass, the buffer has dependencies that will cause
290 		 * it to be redirtied and it has not already been deferred,
291 		 * or it is already being written.
292 		 */
293 		if ((bp->b_vflags & BV_SCANNED) != 0)
294 			continue;
295 		bp->b_vflags |= BV_SCANNED;
296 		/*
297 		 * Flush indirects in order, if requested.
298 		 *
299 		 * Note that if only datasync is requested, we can
300 		 * skip indirect blocks when softupdates are not
301 		 * active.  Otherwise we must flush them with data,
302 		 * since dependencies prevent data block writes.
303 		 */
304 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR &&
305 		    (lbn_level(bp->b_lblkno) >= passes ||
306 		    ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
307 			continue;
308 		if (bp->b_lblkno > lbn)
309 			panic("ffs_syncvnode: syncing truncated data.");
310 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
311 			BO_UNLOCK(bo);
312 		} else if (wait) {
313 			if (BUF_LOCK(bp,
314 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
315 			    BO_LOCKPTR(bo)) != 0) {
316 				bp->b_vflags &= ~BV_SCANNED;
317 				goto next;
318 			}
319 		} else
320 			continue;
321 		if ((bp->b_flags & B_DELWRI) == 0)
322 			panic("ffs_fsync: not dirty");
323 		/*
324 		 * Check for dependencies and potentially complete them.
325 		 */
326 		if (!LIST_EMPTY(&bp->b_dep) &&
327 		    (error = softdep_sync_buf(vp, bp,
328 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
329 			/*
330 			 * Lock order conflict, buffer was already unlocked,
331 			 * and vnode possibly unlocked.
332 			 */
333 			if (error == ERELOOKUP) {
334 				if (vp->v_data == NULL)
335 					return (EBADF);
336 				unlocked = true;
337 				if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
338 				    (error = softdep_sync_metadata(vp)) != 0) {
339 					if (ffs_fsfail_cleanup(ump, error))
340 						error = 0;
341 					return (unlocked && error == 0 ?
342 					    ERELOOKUP : error);
343 				}
344 				/* Re-evaluate inode size */
345 				lbn = lblkno(ITOFS(ip), (ip->i_size +
346 				    ITOFS(ip)->fs_bsize - 1));
347 				goto next;
348 			}
349 			/* I/O error. */
350 			if (error != EBUSY) {
351 				BUF_UNLOCK(bp);
352 				return (error);
353 			}
354 			/* If we deferred once, don't defer again. */
355 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
356 				bp->b_flags |= B_DEFERRED;
357 				BUF_UNLOCK(bp);
358 				goto next;
359 			}
360 		}
361 		if (wait) {
362 			bremfree(bp);
363 			error = bwrite(bp);
364 			if (ffs_fsfail_cleanup(ump, error))
365 				error = 0;
366 			if (error != 0)
367 				return (error);
368 		} else if ((bp->b_flags & B_CLUSTEROK)) {
369 			(void) vfs_bio_awrite(bp);
370 		} else {
371 			bremfree(bp);
372 			(void) bawrite(bp);
373 		}
374 next:
375 		/*
376 		 * Since we may have slept during the I/O, we need
377 		 * to start from a known point.
378 		 */
379 		BO_LOCK(bo);
380 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
381 	}
382 	if (waitfor != MNT_WAIT) {
383 		BO_UNLOCK(bo);
384 		if ((flags & NO_INO_UPDT) != 0)
385 			return (unlocked ? ERELOOKUP : 0);
386 		error = ffs_update(vp, 0);
387 		if (error == 0 && unlocked)
388 			error = ERELOOKUP;
389 		return (error);
390 	}
391 	/* Drain IO to see if we're done. */
392 	bufobj_wwait(bo, 0, 0);
393 	/*
394 	 * Block devices associated with filesystems may have new I/O
395 	 * requests posted for them even if the vnode is locked, so no
396 	 * amount of trying will get them clean.  We make several passes
397 	 * as a best effort.
398 	 *
399 	 * Regular files may need multiple passes to flush all dependency
400 	 * work as it is possible that we must write once per indirect
401 	 * level, once for the leaf, and once for the inode and each of
402 	 * these will be done with one sync and one async pass.
403 	 */
404 	if (bo->bo_dirty.bv_cnt > 0) {
405 		if ((flags & DATA_ONLY) == 0) {
406 			still_dirty = true;
407 		} else {
408 			/*
409 			 * For data-only sync, dirty indirect buffers
410 			 * are ignored.
411 			 */
412 			still_dirty = false;
413 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
414 				if (bp->b_lblkno > -UFS_NDADDR) {
415 					still_dirty = true;
416 					break;
417 				}
418 			}
419 		}
420 
421 		if (still_dirty) {
422 			/* Write the inode after sync passes to flush deps. */
423 			if (wait && DOINGSOFTDEP(vp) &&
424 			    (flags & NO_INO_UPDT) == 0) {
425 				BO_UNLOCK(bo);
426 				ffs_update(vp, 1);
427 				BO_LOCK(bo);
428 			}
429 			/* switch between sync/async. */
430 			wait = !wait;
431 			if (wait || ++passes < UFS_NIADDR + 2)
432 				goto loop;
433 		}
434 	}
435 	BO_UNLOCK(bo);
436 	error = 0;
437 	if ((flags & DATA_ONLY) == 0) {
438 		if ((flags & NO_INO_UPDT) == 0)
439 			error = ffs_update(vp, 1);
440 		if (DOINGSUJ(vp))
441 			softdep_journal_fsync(VTOI(vp));
442 	} else if ((ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)) != 0) {
443 		error = ffs_update(vp, 1);
444 	}
445 	if (error == 0 && unlocked)
446 		error = ERELOOKUP;
447 	return (error);
448 }
449 
450 static int
451 ffs_fdatasync(struct vop_fdatasync_args *ap)
452 {
453 
454 	return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
455 }
456 
457 static int
458 ffs_lock(ap)
459 	struct vop_lock1_args /* {
460 		struct vnode *a_vp;
461 		int a_flags;
462 		char *file;
463 		int line;
464 	} */ *ap;
465 {
466 #if !defined(NO_FFS_SNAPSHOT) || defined(DIAGNOSTIC)
467 	struct vnode *vp = ap->a_vp;
468 #endif	/* !NO_FFS_SNAPSHOT || DIAGNOSTIC */
469 #ifdef DIAGNOSTIC
470 	struct inode *ip;
471 #endif	/* DIAGNOSTIC */
472 	int result;
473 #ifndef NO_FFS_SNAPSHOT
474 	int flags;
475 	struct lock *lkp;
476 
477 	/*
478 	 * Adaptive spinning mixed with SU leads to trouble. use a giant hammer
479 	 * and only use it when LK_NODDLKTREAT is set. Currently this means it
480 	 * is only used during path lookup.
481 	 */
482 	if ((ap->a_flags & LK_NODDLKTREAT) != 0)
483 		ap->a_flags |= LK_ADAPTIVE;
484 	switch (ap->a_flags & LK_TYPE_MASK) {
485 	case LK_SHARED:
486 	case LK_UPGRADE:
487 	case LK_EXCLUSIVE:
488 		flags = ap->a_flags;
489 		for (;;) {
490 #ifdef DEBUG_VFS_LOCKS
491 			VNPASS(vp->v_holdcnt != 0, vp);
492 #endif	/* DEBUG_VFS_LOCKS */
493 			lkp = vp->v_vnlock;
494 			result = lockmgr_lock_flags(lkp, flags,
495 			    &VI_MTX(vp)->lock_object, ap->a_file, ap->a_line);
496 			if (lkp == vp->v_vnlock || result != 0)
497 				break;
498 			/*
499 			 * Apparent success, except that the vnode
500 			 * mutated between snapshot file vnode and
501 			 * regular file vnode while this process
502 			 * slept.  The lock currently held is not the
503 			 * right lock.  Release it, and try to get the
504 			 * new lock.
505 			 */
506 			lockmgr_unlock(lkp);
507 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
508 			    (LK_INTERLOCK | LK_NOWAIT))
509 				return (EBUSY);
510 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
511 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
512 			flags &= ~LK_INTERLOCK;
513 		}
514 #ifdef DIAGNOSTIC
515 		switch (ap->a_flags & LK_TYPE_MASK) {
516 		case LK_UPGRADE:
517 		case LK_EXCLUSIVE:
518 			if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
519 				ip = VTOI(vp);
520 				if (ip != NULL)
521 					ip->i_lock_gen++;
522 			}
523 		}
524 #endif	/* DIAGNOSTIC */
525 		break;
526 	default:
527 #ifdef DIAGNOSTIC
528 		if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
529 			ip = VTOI(vp);
530 			if (ip != NULL)
531 				ufs_unlock_tracker(ip);
532 		}
533 #endif	/* DIAGNOSTIC */
534 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
535 		break;
536 	}
537 #else	/* NO_FFS_SNAPSHOT */
538 	/*
539 	 * See above for an explanation.
540 	 */
541 	if ((ap->a_flags & LK_NODDLKTREAT) != 0)
542 		ap->a_flags |= LK_ADAPTIVE;
543 #ifdef DIAGNOSTIC
544 	if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
545 		ip = VTOI(vp);
546 		if (ip != NULL)
547 			ufs_unlock_tracker(ip);
548 	}
549 #endif	/* DIAGNOSTIC */
550 	result =  VOP_LOCK1_APV(&ufs_vnodeops, ap);
551 #endif	/* NO_FFS_SNAPSHOT */
552 #ifdef DIAGNOSTIC
553 	switch (ap->a_flags & LK_TYPE_MASK) {
554 	case LK_UPGRADE:
555 	case LK_EXCLUSIVE:
556 		if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
557 			ip = VTOI(vp);
558 			if (ip != NULL)
559 				ip->i_lock_gen++;
560 		}
561 	}
562 #endif	/* DIAGNOSTIC */
563 	return (result);
564 }
565 
566 #ifdef INVARIANTS
567 static int
568 ffs_unlock_debug(struct vop_unlock_args *ap)
569 {
570 	struct vnode *vp;
571 	struct inode *ip;
572 
573 	vp = ap->a_vp;
574 	ip = VTOI(vp);
575 	if (ip->i_flag & UFS_INODE_FLAG_LAZY_MASK_ASSERTABLE) {
576 		if ((vp->v_mflag & VMP_LAZYLIST) == 0) {
577 			VI_LOCK(vp);
578 			VNASSERT((vp->v_mflag & VMP_LAZYLIST), vp,
579 			    ("%s: modified vnode (%x) not on lazy list",
580 			    __func__, ip->i_flag));
581 			VI_UNLOCK(vp);
582 		}
583 	}
584 #ifdef DIAGNOSTIC
585 	if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && ip != NULL &&
586 	    vp->v_vnlock->lk_recurse == 0)
587 		ufs_unlock_tracker(ip);
588 #endif
589 	return (VOP_UNLOCK_APV(&ufs_vnodeops, ap));
590 }
591 #endif
592 
593 static int
594 ffs_read_hole(struct uio *uio, long xfersize, long *size)
595 {
596 	ssize_t saved_resid, tlen;
597 	int error;
598 
599 	while (xfersize > 0) {
600 		tlen = min(xfersize, ZERO_REGION_SIZE);
601 		saved_resid = uio->uio_resid;
602 		error = vn_io_fault_uiomove(__DECONST(void *, zero_region),
603 		    tlen, uio);
604 		if (error != 0)
605 			return (error);
606 		tlen = saved_resid - uio->uio_resid;
607 		xfersize -= tlen;
608 		*size -= tlen;
609 	}
610 	return (0);
611 }
612 
613 /*
614  * Vnode op for reading.
615  */
616 static int
617 ffs_read(ap)
618 	struct vop_read_args /* {
619 		struct vnode *a_vp;
620 		struct uio *a_uio;
621 		int a_ioflag;
622 		struct ucred *a_cred;
623 	} */ *ap;
624 {
625 	struct vnode *vp;
626 	struct inode *ip;
627 	struct uio *uio;
628 	struct fs *fs;
629 	struct buf *bp;
630 	ufs_lbn_t lbn, nextlbn;
631 	off_t bytesinfile;
632 	long size, xfersize, blkoffset;
633 	ssize_t orig_resid;
634 	int bflag, error, ioflag, seqcount;
635 
636 	vp = ap->a_vp;
637 	uio = ap->a_uio;
638 	ioflag = ap->a_ioflag;
639 	if (ap->a_ioflag & IO_EXT)
640 #ifdef notyet
641 		return (ffs_extread(vp, uio, ioflag));
642 #else
643 		panic("ffs_read+IO_EXT");
644 #endif
645 #ifdef DIRECTIO
646 	if ((ioflag & IO_DIRECT) != 0) {
647 		int workdone;
648 
649 		error = ffs_rawread(vp, uio, &workdone);
650 		if (error != 0 || workdone != 0)
651 			return error;
652 	}
653 #endif
654 
655 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
656 	ip = VTOI(vp);
657 
658 #ifdef INVARIANTS
659 	if (uio->uio_rw != UIO_READ)
660 		panic("ffs_read: mode");
661 
662 	if (vp->v_type == VLNK) {
663 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
664 			panic("ffs_read: short symlink");
665 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
666 		panic("ffs_read: type %d",  vp->v_type);
667 #endif
668 	orig_resid = uio->uio_resid;
669 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
670 	if (orig_resid == 0)
671 		return (0);
672 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
673 	fs = ITOFS(ip);
674 	if (uio->uio_offset < ip->i_size &&
675 	    uio->uio_offset >= fs->fs_maxfilesize)
676 		return (EOVERFLOW);
677 
678 	bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE);
679 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
680 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
681 			break;
682 		lbn = lblkno(fs, uio->uio_offset);
683 		nextlbn = lbn + 1;
684 
685 		/*
686 		 * size of buffer.  The buffer representing the
687 		 * end of the file is rounded up to the size of
688 		 * the block type ( fragment or full block,
689 		 * depending ).
690 		 */
691 		size = blksize(fs, ip, lbn);
692 		blkoffset = blkoff(fs, uio->uio_offset);
693 
694 		/*
695 		 * The amount we want to transfer in this iteration is
696 		 * one FS block less the amount of the data before
697 		 * our startpoint (duh!)
698 		 */
699 		xfersize = fs->fs_bsize - blkoffset;
700 
701 		/*
702 		 * But if we actually want less than the block,
703 		 * or the file doesn't have a whole block more of data,
704 		 * then use the lesser number.
705 		 */
706 		if (uio->uio_resid < xfersize)
707 			xfersize = uio->uio_resid;
708 		if (bytesinfile < xfersize)
709 			xfersize = bytesinfile;
710 
711 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
712 			/*
713 			 * Don't do readahead if this is the end of the file.
714 			 */
715 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
716 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
717 			/*
718 			 * Otherwise if we are allowed to cluster,
719 			 * grab as much as we can.
720 			 *
721 			 * XXX  This may not be a win if we are not
722 			 * doing sequential access.
723 			 */
724 			error = cluster_read(vp, ip->i_size, lbn,
725 			    size, NOCRED, blkoffset + uio->uio_resid,
726 			    seqcount, bflag, &bp);
727 		} else if (seqcount > 1) {
728 			/*
729 			 * If we are NOT allowed to cluster, then
730 			 * if we appear to be acting sequentially,
731 			 * fire off a request for a readahead
732 			 * as well as a read. Note that the 4th and 5th
733 			 * arguments point to arrays of the size specified in
734 			 * the 6th argument.
735 			 */
736 			u_int nextsize = blksize(fs, ip, nextlbn);
737 			error = breadn_flags(vp, lbn, lbn, size, &nextlbn,
738 			    &nextsize, 1, NOCRED, bflag, NULL, &bp);
739 		} else {
740 			/*
741 			 * Failing all of the above, just read what the
742 			 * user asked for. Interestingly, the same as
743 			 * the first option above.
744 			 */
745 			error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
746 		}
747 		if (error == EJUSTRETURN) {
748 			error = ffs_read_hole(uio, xfersize, &size);
749 			if (error == 0)
750 				continue;
751 		}
752 		if (error != 0) {
753 			brelse(bp);
754 			bp = NULL;
755 			break;
756 		}
757 
758 		/*
759 		 * We should only get non-zero b_resid when an I/O error
760 		 * has occurred, which should cause us to break above.
761 		 * However, if the short read did not cause an error,
762 		 * then we want to ensure that we do not uiomove bad
763 		 * or uninitialized data.
764 		 */
765 		size -= bp->b_resid;
766 		if (size < xfersize) {
767 			if (size == 0)
768 				break;
769 			xfersize = size;
770 		}
771 
772 		if (buf_mapped(bp)) {
773 			error = vn_io_fault_uiomove((char *)bp->b_data +
774 			    blkoffset, (int)xfersize, uio);
775 		} else {
776 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
777 			    (int)xfersize, uio);
778 		}
779 		if (error)
780 			break;
781 
782 		vfs_bio_brelse(bp, ioflag);
783 	}
784 
785 	/*
786 	 * This can only happen in the case of an error
787 	 * because the loop above resets bp to NULL on each iteration
788 	 * and on normal completion has not set a new value into it.
789 	 * so it must have come from a 'break' statement
790 	 */
791 	if (bp != NULL)
792 		vfs_bio_brelse(bp, ioflag);
793 
794 	if ((error == 0 || uio->uio_resid != orig_resid) &&
795 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
796 		UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS);
797 	return (error);
798 }
799 
800 /*
801  * Vnode op for writing.
802  */
803 static int
804 ffs_write(ap)
805 	struct vop_write_args /* {
806 		struct vnode *a_vp;
807 		struct uio *a_uio;
808 		int a_ioflag;
809 		struct ucred *a_cred;
810 	} */ *ap;
811 {
812 	struct vnode *vp;
813 	struct uio *uio;
814 	struct inode *ip;
815 	struct fs *fs;
816 	struct buf *bp;
817 	ufs_lbn_t lbn;
818 	off_t osize;
819 	ssize_t resid;
820 	int seqcount;
821 	int blkoffset, error, flags, ioflag, size, xfersize;
822 
823 	vp = ap->a_vp;
824 	uio = ap->a_uio;
825 	ioflag = ap->a_ioflag;
826 	if (ap->a_ioflag & IO_EXT)
827 #ifdef notyet
828 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
829 #else
830 		panic("ffs_write+IO_EXT");
831 #endif
832 
833 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
834 	ip = VTOI(vp);
835 
836 #ifdef INVARIANTS
837 	if (uio->uio_rw != UIO_WRITE)
838 		panic("ffs_write: mode");
839 #endif
840 
841 	switch (vp->v_type) {
842 	case VREG:
843 		if (ioflag & IO_APPEND)
844 			uio->uio_offset = ip->i_size;
845 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
846 			return (EPERM);
847 		/* FALLTHROUGH */
848 	case VLNK:
849 		break;
850 	case VDIR:
851 		panic("ffs_write: dir write");
852 		break;
853 	default:
854 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
855 			(int)uio->uio_offset,
856 			(int)uio->uio_resid
857 		);
858 	}
859 
860 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
861 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
862 	fs = ITOFS(ip);
863 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
864 		return (EFBIG);
865 	/*
866 	 * Maybe this should be above the vnode op call, but so long as
867 	 * file servers have no limits, I don't think it matters.
868 	 */
869 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
870 		return (EFBIG);
871 
872 	resid = uio->uio_resid;
873 	osize = ip->i_size;
874 	if (seqcount > BA_SEQMAX)
875 		flags = BA_SEQMAX << BA_SEQSHIFT;
876 	else
877 		flags = seqcount << BA_SEQSHIFT;
878 	if (ioflag & IO_SYNC)
879 		flags |= IO_SYNC;
880 	flags |= BA_UNMAPPED;
881 
882 	for (error = 0; uio->uio_resid > 0;) {
883 		lbn = lblkno(fs, uio->uio_offset);
884 		blkoffset = blkoff(fs, uio->uio_offset);
885 		xfersize = fs->fs_bsize - blkoffset;
886 		if (uio->uio_resid < xfersize)
887 			xfersize = uio->uio_resid;
888 		if (uio->uio_offset + xfersize > ip->i_size)
889 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
890 
891 		/*
892 		 * We must perform a read-before-write if the transfer size
893 		 * does not cover the entire buffer.
894 		 */
895 		if (fs->fs_bsize > xfersize)
896 			flags |= BA_CLRBUF;
897 		else
898 			flags &= ~BA_CLRBUF;
899 /* XXX is uio->uio_offset the right thing here? */
900 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
901 		    ap->a_cred, flags, &bp);
902 		if (error != 0) {
903 			vnode_pager_setsize(vp, ip->i_size);
904 			break;
905 		}
906 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
907 			bp->b_flags |= B_NOCACHE;
908 
909 		if (uio->uio_offset + xfersize > ip->i_size) {
910 			ip->i_size = uio->uio_offset + xfersize;
911 			DIP_SET(ip, i_size, ip->i_size);
912 			UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
913 		}
914 
915 		size = blksize(fs, ip, lbn) - bp->b_resid;
916 		if (size < xfersize)
917 			xfersize = size;
918 
919 		if (buf_mapped(bp)) {
920 			error = vn_io_fault_uiomove((char *)bp->b_data +
921 			    blkoffset, (int)xfersize, uio);
922 		} else {
923 			error = vn_io_fault_pgmove(bp->b_pages, blkoffset,
924 			    (int)xfersize, uio);
925 		}
926 		/*
927 		 * If the buffer is not already filled and we encounter an
928 		 * error while trying to fill it, we have to clear out any
929 		 * garbage data from the pages instantiated for the buffer.
930 		 * If we do not, a failed uiomove() during a write can leave
931 		 * the prior contents of the pages exposed to a userland mmap.
932 		 *
933 		 * Note that we need only clear buffers with a transfer size
934 		 * equal to the block size because buffers with a shorter
935 		 * transfer size were cleared above by the call to UFS_BALLOC()
936 		 * with the BA_CLRBUF flag set.
937 		 *
938 		 * If the source region for uiomove identically mmaps the
939 		 * buffer, uiomove() performed the NOP copy, and the buffer
940 		 * content remains valid because the page fault handler
941 		 * validated the pages.
942 		 */
943 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
944 		    fs->fs_bsize == xfersize)
945 			vfs_bio_clrbuf(bp);
946 
947 		vfs_bio_set_flags(bp, ioflag);
948 
949 		/*
950 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
951 		 * if we have a severe page deficiency write the buffer
952 		 * asynchronously.  Otherwise try to cluster, and if that
953 		 * doesn't do it then either do an async write (if O_DIRECT),
954 		 * or a delayed write (if not).
955 		 */
956 		if (ioflag & IO_SYNC) {
957 			(void)bwrite(bp);
958 		} else if (vm_page_count_severe() ||
959 			    buf_dirty_count_severe() ||
960 			    (ioflag & IO_ASYNC)) {
961 			bp->b_flags |= B_CLUSTEROK;
962 			bawrite(bp);
963 		} else if (xfersize + blkoffset == fs->fs_bsize) {
964 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
965 				bp->b_flags |= B_CLUSTEROK;
966 				cluster_write(vp, bp, ip->i_size, seqcount,
967 				    GB_UNMAPPED);
968 			} else {
969 				bawrite(bp);
970 			}
971 		} else if (ioflag & IO_DIRECT) {
972 			bp->b_flags |= B_CLUSTEROK;
973 			bawrite(bp);
974 		} else {
975 			bp->b_flags |= B_CLUSTEROK;
976 			bdwrite(bp);
977 		}
978 		if (error || xfersize == 0)
979 			break;
980 		UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE);
981 	}
982 	/*
983 	 * If we successfully wrote any data, and we are not the superuser
984 	 * we clear the setuid and setgid bits as a precaution against
985 	 * tampering.
986 	 */
987 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
988 	    ap->a_cred) {
989 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) {
990 			vn_seqc_write_begin(vp);
991 			UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
992 			DIP_SET(ip, i_mode, ip->i_mode);
993 			vn_seqc_write_end(vp);
994 		}
995 	}
996 	if (error) {
997 		if (ioflag & IO_UNIT) {
998 			(void)ffs_truncate(vp, osize,
999 			    IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
1000 			uio->uio_offset -= resid - uio->uio_resid;
1001 			uio->uio_resid = resid;
1002 		}
1003 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
1004 		error = ffs_update(vp, 1);
1005 		if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error))
1006 			error = ENXIO;
1007 	}
1008 	return (error);
1009 }
1010 
1011 /*
1012  * Extended attribute area reading.
1013  */
1014 static int
1015 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
1016 {
1017 	struct inode *ip;
1018 	struct ufs2_dinode *dp;
1019 	struct fs *fs;
1020 	struct buf *bp;
1021 	ufs_lbn_t lbn, nextlbn;
1022 	off_t bytesinfile;
1023 	long size, xfersize, blkoffset;
1024 	ssize_t orig_resid;
1025 	int error;
1026 
1027 	ip = VTOI(vp);
1028 	fs = ITOFS(ip);
1029 	dp = ip->i_din2;
1030 
1031 #ifdef INVARIANTS
1032 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
1033 		panic("ffs_extread: mode");
1034 
1035 #endif
1036 	orig_resid = uio->uio_resid;
1037 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
1038 	if (orig_resid == 0)
1039 		return (0);
1040 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
1041 
1042 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1043 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
1044 			break;
1045 		lbn = lblkno(fs, uio->uio_offset);
1046 		nextlbn = lbn + 1;
1047 
1048 		/*
1049 		 * size of buffer.  The buffer representing the
1050 		 * end of the file is rounded up to the size of
1051 		 * the block type ( fragment or full block,
1052 		 * depending ).
1053 		 */
1054 		size = sblksize(fs, dp->di_extsize, lbn);
1055 		blkoffset = blkoff(fs, uio->uio_offset);
1056 
1057 		/*
1058 		 * The amount we want to transfer in this iteration is
1059 		 * one FS block less the amount of the data before
1060 		 * our startpoint (duh!)
1061 		 */
1062 		xfersize = fs->fs_bsize - blkoffset;
1063 
1064 		/*
1065 		 * But if we actually want less than the block,
1066 		 * or the file doesn't have a whole block more of data,
1067 		 * then use the lesser number.
1068 		 */
1069 		if (uio->uio_resid < xfersize)
1070 			xfersize = uio->uio_resid;
1071 		if (bytesinfile < xfersize)
1072 			xfersize = bytesinfile;
1073 
1074 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
1075 			/*
1076 			 * Don't do readahead if this is the end of the info.
1077 			 */
1078 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
1079 		} else {
1080 			/*
1081 			 * If we have a second block, then
1082 			 * fire off a request for a readahead
1083 			 * as well as a read. Note that the 4th and 5th
1084 			 * arguments point to arrays of the size specified in
1085 			 * the 6th argument.
1086 			 */
1087 			u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
1088 
1089 			nextlbn = -1 - nextlbn;
1090 			error = breadn(vp, -1 - lbn,
1091 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1092 		}
1093 		if (error) {
1094 			brelse(bp);
1095 			bp = NULL;
1096 			break;
1097 		}
1098 
1099 		/*
1100 		 * We should only get non-zero b_resid when an I/O error
1101 		 * has occurred, which should cause us to break above.
1102 		 * However, if the short read did not cause an error,
1103 		 * then we want to ensure that we do not uiomove bad
1104 		 * or uninitialized data.
1105 		 */
1106 		size -= bp->b_resid;
1107 		if (size < xfersize) {
1108 			if (size == 0)
1109 				break;
1110 			xfersize = size;
1111 		}
1112 
1113 		error = uiomove((char *)bp->b_data + blkoffset,
1114 					(int)xfersize, uio);
1115 		if (error)
1116 			break;
1117 		vfs_bio_brelse(bp, ioflag);
1118 	}
1119 
1120 	/*
1121 	 * This can only happen in the case of an error
1122 	 * because the loop above resets bp to NULL on each iteration
1123 	 * and on normal completion has not set a new value into it.
1124 	 * so it must have come from a 'break' statement
1125 	 */
1126 	if (bp != NULL)
1127 		vfs_bio_brelse(bp, ioflag);
1128 	return (error);
1129 }
1130 
1131 /*
1132  * Extended attribute area writing.
1133  */
1134 static int
1135 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1136 {
1137 	struct inode *ip;
1138 	struct ufs2_dinode *dp;
1139 	struct fs *fs;
1140 	struct buf *bp;
1141 	ufs_lbn_t lbn;
1142 	off_t osize;
1143 	ssize_t resid;
1144 	int blkoffset, error, flags, size, xfersize;
1145 
1146 	ip = VTOI(vp);
1147 	fs = ITOFS(ip);
1148 	dp = ip->i_din2;
1149 
1150 #ifdef INVARIANTS
1151 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1152 		panic("ffs_extwrite: mode");
1153 #endif
1154 
1155 	if (ioflag & IO_APPEND)
1156 		uio->uio_offset = dp->di_extsize;
1157 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1158 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1159 	if ((uoff_t)uio->uio_offset + uio->uio_resid >
1160 	    UFS_NXADDR * fs->fs_bsize)
1161 		return (EFBIG);
1162 
1163 	resid = uio->uio_resid;
1164 	osize = dp->di_extsize;
1165 	flags = IO_EXT;
1166 	if (ioflag & IO_SYNC)
1167 		flags |= IO_SYNC;
1168 
1169 	for (error = 0; uio->uio_resid > 0;) {
1170 		lbn = lblkno(fs, uio->uio_offset);
1171 		blkoffset = blkoff(fs, uio->uio_offset);
1172 		xfersize = fs->fs_bsize - blkoffset;
1173 		if (uio->uio_resid < xfersize)
1174 			xfersize = uio->uio_resid;
1175 
1176 		/*
1177 		 * We must perform a read-before-write if the transfer size
1178 		 * does not cover the entire buffer.
1179 		 */
1180 		if (fs->fs_bsize > xfersize)
1181 			flags |= BA_CLRBUF;
1182 		else
1183 			flags &= ~BA_CLRBUF;
1184 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1185 		    ucred, flags, &bp);
1186 		if (error != 0)
1187 			break;
1188 		/*
1189 		 * If the buffer is not valid we have to clear out any
1190 		 * garbage data from the pages instantiated for the buffer.
1191 		 * If we do not, a failed uiomove() during a write can leave
1192 		 * the prior contents of the pages exposed to a userland
1193 		 * mmap().  XXX deal with uiomove() errors a better way.
1194 		 */
1195 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1196 			vfs_bio_clrbuf(bp);
1197 
1198 		if (uio->uio_offset + xfersize > dp->di_extsize) {
1199 			dp->di_extsize = uio->uio_offset + xfersize;
1200 			UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
1201 		}
1202 
1203 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1204 		if (size < xfersize)
1205 			xfersize = size;
1206 
1207 		error =
1208 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1209 
1210 		vfs_bio_set_flags(bp, ioflag);
1211 
1212 		/*
1213 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1214 		 * if we have a severe page deficiency write the buffer
1215 		 * asynchronously.  Otherwise try to cluster, and if that
1216 		 * doesn't do it then either do an async write (if O_DIRECT),
1217 		 * or a delayed write (if not).
1218 		 */
1219 		if (ioflag & IO_SYNC) {
1220 			(void)bwrite(bp);
1221 		} else if (vm_page_count_severe() ||
1222 			    buf_dirty_count_severe() ||
1223 			    xfersize + blkoffset == fs->fs_bsize ||
1224 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1225 			bawrite(bp);
1226 		else
1227 			bdwrite(bp);
1228 		if (error || xfersize == 0)
1229 			break;
1230 		UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1231 	}
1232 	/*
1233 	 * If we successfully wrote any data, and we are not the superuser
1234 	 * we clear the setuid and setgid bits as a precaution against
1235 	 * tampering.
1236 	 */
1237 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1238 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) {
1239 			vn_seqc_write_begin(vp);
1240 			UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1241 			dp->di_mode = ip->i_mode;
1242 			vn_seqc_write_end(vp);
1243 		}
1244 	}
1245 	if (error) {
1246 		if (ioflag & IO_UNIT) {
1247 			(void)ffs_truncate(vp, osize,
1248 			    IO_EXT | (ioflag&IO_SYNC), ucred);
1249 			uio->uio_offset -= resid - uio->uio_resid;
1250 			uio->uio_resid = resid;
1251 		}
1252 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1253 		error = ffs_update(vp, 1);
1254 	return (error);
1255 }
1256 
1257 /*
1258  * Vnode operating to retrieve a named extended attribute.
1259  *
1260  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1261  * the length of the EA, and possibly the pointer to the entry and to the data.
1262  */
1263 static int
1264 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name,
1265     struct extattr **eapp, u_char **eac)
1266 {
1267 	struct extattr *eap, *eaend;
1268 	size_t nlen;
1269 
1270 	nlen = strlen(name);
1271 	KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned"));
1272 	eap = (struct extattr *)ptr;
1273 	eaend = (struct extattr *)(ptr + length);
1274 	for (; eap < eaend; eap = EXTATTR_NEXT(eap)) {
1275 		KASSERT(EXTATTR_NEXT(eap) <= eaend,
1276 		    ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1277 		if (eap->ea_namespace != nspace || eap->ea_namelength != nlen
1278 		    || memcmp(eap->ea_name, name, nlen) != 0)
1279 			continue;
1280 		if (eapp != NULL)
1281 			*eapp = eap;
1282 		if (eac != NULL)
1283 			*eac = EXTATTR_CONTENT(eap);
1284 		return (EXTATTR_CONTENT_SIZE(eap));
1285 	}
1286 	return (-1);
1287 }
1288 
1289 static int
1290 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td)
1291 {
1292 	const struct extattr *eap, *eaend, *eapnext;
1293 	struct inode *ip;
1294 	struct ufs2_dinode *dp;
1295 	struct fs *fs;
1296 	struct uio luio;
1297 	struct iovec liovec;
1298 	u_int easize;
1299 	int error;
1300 	u_char *eae;
1301 
1302 	ip = VTOI(vp);
1303 	fs = ITOFS(ip);
1304 	dp = ip->i_din2;
1305 	easize = dp->di_extsize;
1306 	if ((uoff_t)easize > UFS_NXADDR * fs->fs_bsize)
1307 		return (EFBIG);
1308 
1309 	eae = malloc(easize, M_TEMP, M_WAITOK);
1310 
1311 	liovec.iov_base = eae;
1312 	liovec.iov_len = easize;
1313 	luio.uio_iov = &liovec;
1314 	luio.uio_iovcnt = 1;
1315 	luio.uio_offset = 0;
1316 	luio.uio_resid = easize;
1317 	luio.uio_segflg = UIO_SYSSPACE;
1318 	luio.uio_rw = UIO_READ;
1319 	luio.uio_td = td;
1320 
1321 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1322 	if (error) {
1323 		free(eae, M_TEMP);
1324 		return (error);
1325 	}
1326 	/* Validate disk xattrfile contents. */
1327 	for (eap = (void *)eae, eaend = (void *)(eae + easize); eap < eaend;
1328 	    eap = eapnext) {
1329 		eapnext = EXTATTR_NEXT(eap);
1330 		/* Bogusly short entry or bogusly long entry. */
1331 		if (eap->ea_length < sizeof(*eap) || eapnext > eaend) {
1332 			free(eae, M_TEMP);
1333 			return (EINTEGRITY);
1334 		}
1335 	}
1336 	*p = eae;
1337 	return (0);
1338 }
1339 
1340 static void
1341 ffs_lock_ea(struct vnode *vp)
1342 {
1343 	struct inode *ip;
1344 
1345 	ip = VTOI(vp);
1346 	VI_LOCK(vp);
1347 	while (ip->i_flag & IN_EA_LOCKED) {
1348 		UFS_INODE_SET_FLAG(ip, IN_EA_LOCKWAIT);
1349 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1350 		    0);
1351 	}
1352 	UFS_INODE_SET_FLAG(ip, IN_EA_LOCKED);
1353 	VI_UNLOCK(vp);
1354 }
1355 
1356 static void
1357 ffs_unlock_ea(struct vnode *vp)
1358 {
1359 	struct inode *ip;
1360 
1361 	ip = VTOI(vp);
1362 	VI_LOCK(vp);
1363 	if (ip->i_flag & IN_EA_LOCKWAIT)
1364 		wakeup(&ip->i_ea_refs);
1365 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1366 	VI_UNLOCK(vp);
1367 }
1368 
1369 static int
1370 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1371 {
1372 	struct inode *ip;
1373 	struct ufs2_dinode *dp;
1374 	int error;
1375 
1376 	ip = VTOI(vp);
1377 
1378 	ffs_lock_ea(vp);
1379 	if (ip->i_ea_area != NULL) {
1380 		ip->i_ea_refs++;
1381 		ffs_unlock_ea(vp);
1382 		return (0);
1383 	}
1384 	dp = ip->i_din2;
1385 	error = ffs_rdextattr(&ip->i_ea_area, vp, td);
1386 	if (error) {
1387 		ffs_unlock_ea(vp);
1388 		return (error);
1389 	}
1390 	ip->i_ea_len = dp->di_extsize;
1391 	ip->i_ea_error = 0;
1392 	ip->i_ea_refs++;
1393 	ffs_unlock_ea(vp);
1394 	return (0);
1395 }
1396 
1397 /*
1398  * Vnode extattr transaction commit/abort
1399  */
1400 static int
1401 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1402 {
1403 	struct inode *ip;
1404 	struct uio luio;
1405 	struct iovec liovec;
1406 	int error;
1407 	struct ufs2_dinode *dp;
1408 
1409 	ip = VTOI(vp);
1410 
1411 	ffs_lock_ea(vp);
1412 	if (ip->i_ea_area == NULL) {
1413 		ffs_unlock_ea(vp);
1414 		return (EINVAL);
1415 	}
1416 	dp = ip->i_din2;
1417 	error = ip->i_ea_error;
1418 	if (commit && error == 0) {
1419 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1420 		if (cred == NOCRED)
1421 			cred =  vp->v_mount->mnt_cred;
1422 		liovec.iov_base = ip->i_ea_area;
1423 		liovec.iov_len = ip->i_ea_len;
1424 		luio.uio_iov = &liovec;
1425 		luio.uio_iovcnt = 1;
1426 		luio.uio_offset = 0;
1427 		luio.uio_resid = ip->i_ea_len;
1428 		luio.uio_segflg = UIO_SYSSPACE;
1429 		luio.uio_rw = UIO_WRITE;
1430 		luio.uio_td = td;
1431 		/* XXX: I'm not happy about truncating to zero size */
1432 		if (ip->i_ea_len < dp->di_extsize)
1433 			error = ffs_truncate(vp, 0, IO_EXT, cred);
1434 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1435 	}
1436 	if (--ip->i_ea_refs == 0) {
1437 		free(ip->i_ea_area, M_TEMP);
1438 		ip->i_ea_area = NULL;
1439 		ip->i_ea_len = 0;
1440 		ip->i_ea_error = 0;
1441 	}
1442 	ffs_unlock_ea(vp);
1443 	return (error);
1444 }
1445 
1446 /*
1447  * Vnode extattr strategy routine for fifos.
1448  *
1449  * We need to check for a read or write of the external attributes.
1450  * Otherwise we just fall through and do the usual thing.
1451  */
1452 static int
1453 ffsext_strategy(struct vop_strategy_args *ap)
1454 /*
1455 struct vop_strategy_args {
1456 	struct vnodeop_desc *a_desc;
1457 	struct vnode *a_vp;
1458 	struct buf *a_bp;
1459 };
1460 */
1461 {
1462 	struct vnode *vp;
1463 	daddr_t lbn;
1464 
1465 	vp = ap->a_vp;
1466 	lbn = ap->a_bp->b_lblkno;
1467 	if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR)
1468 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1469 	if (vp->v_type == VFIFO)
1470 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1471 	panic("spec nodes went here");
1472 }
1473 
1474 /*
1475  * Vnode extattr transaction commit/abort
1476  */
1477 static int
1478 ffs_openextattr(struct vop_openextattr_args *ap)
1479 /*
1480 struct vop_openextattr_args {
1481 	struct vnodeop_desc *a_desc;
1482 	struct vnode *a_vp;
1483 	IN struct ucred *a_cred;
1484 	IN struct thread *a_td;
1485 };
1486 */
1487 {
1488 
1489 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1490 		return (EOPNOTSUPP);
1491 
1492 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1493 }
1494 
1495 /*
1496  * Vnode extattr transaction commit/abort
1497  */
1498 static int
1499 ffs_closeextattr(struct vop_closeextattr_args *ap)
1500 /*
1501 struct vop_closeextattr_args {
1502 	struct vnodeop_desc *a_desc;
1503 	struct vnode *a_vp;
1504 	int a_commit;
1505 	IN struct ucred *a_cred;
1506 	IN struct thread *a_td;
1507 };
1508 */
1509 {
1510 
1511 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1512 		return (EOPNOTSUPP);
1513 
1514 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1515 		return (EROFS);
1516 
1517 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1518 }
1519 
1520 /*
1521  * Vnode operation to remove a named attribute.
1522  */
1523 static int
1524 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1525 /*
1526 vop_deleteextattr {
1527 	IN struct vnode *a_vp;
1528 	IN int a_attrnamespace;
1529 	IN const char *a_name;
1530 	IN struct ucred *a_cred;
1531 	IN struct thread *a_td;
1532 };
1533 */
1534 {
1535 	struct inode *ip;
1536 	struct extattr *eap;
1537 	uint32_t ul;
1538 	int olen, error, i, easize;
1539 	u_char *eae;
1540 	void *tmp;
1541 
1542 	ip = VTOI(ap->a_vp);
1543 
1544 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1545 		return (EOPNOTSUPP);
1546 
1547 	if (strlen(ap->a_name) == 0)
1548 		return (EINVAL);
1549 
1550 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1551 		return (EROFS);
1552 
1553 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1554 	    ap->a_cred, ap->a_td, VWRITE);
1555 	if (error) {
1556 		/*
1557 		 * ffs_lock_ea is not needed there, because the vnode
1558 		 * must be exclusively locked.
1559 		 */
1560 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1561 			ip->i_ea_error = error;
1562 		return (error);
1563 	}
1564 
1565 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1566 	if (error)
1567 		return (error);
1568 
1569 	/* CEM: delete could be done in-place instead */
1570 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1571 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1572 	easize = ip->i_ea_len;
1573 
1574 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1575 	    &eap, NULL);
1576 	if (olen == -1) {
1577 		/* delete but nonexistent */
1578 		free(eae, M_TEMP);
1579 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1580 		return (ENOATTR);
1581 	}
1582 	ul = eap->ea_length;
1583 	i = (u_char *)EXTATTR_NEXT(eap) - eae;
1584 	bcopy(EXTATTR_NEXT(eap), eap, easize - i);
1585 	easize -= ul;
1586 
1587 	tmp = ip->i_ea_area;
1588 	ip->i_ea_area = eae;
1589 	ip->i_ea_len = easize;
1590 	free(tmp, M_TEMP);
1591 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1592 	return (error);
1593 }
1594 
1595 /*
1596  * Vnode operation to retrieve a named extended attribute.
1597  */
1598 static int
1599 ffs_getextattr(struct vop_getextattr_args *ap)
1600 /*
1601 vop_getextattr {
1602 	IN struct vnode *a_vp;
1603 	IN int a_attrnamespace;
1604 	IN const char *a_name;
1605 	INOUT struct uio *a_uio;
1606 	OUT size_t *a_size;
1607 	IN struct ucred *a_cred;
1608 	IN struct thread *a_td;
1609 };
1610 */
1611 {
1612 	struct inode *ip;
1613 	u_char *eae, *p;
1614 	unsigned easize;
1615 	int error, ealen;
1616 
1617 	ip = VTOI(ap->a_vp);
1618 
1619 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1620 		return (EOPNOTSUPP);
1621 
1622 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1623 	    ap->a_cred, ap->a_td, VREAD);
1624 	if (error)
1625 		return (error);
1626 
1627 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1628 	if (error)
1629 		return (error);
1630 
1631 	eae = ip->i_ea_area;
1632 	easize = ip->i_ea_len;
1633 
1634 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1635 	    NULL, &p);
1636 	if (ealen >= 0) {
1637 		error = 0;
1638 		if (ap->a_size != NULL)
1639 			*ap->a_size = ealen;
1640 		else if (ap->a_uio != NULL)
1641 			error = uiomove(p, ealen, ap->a_uio);
1642 	} else
1643 		error = ENOATTR;
1644 
1645 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1646 	return (error);
1647 }
1648 
1649 /*
1650  * Vnode operation to retrieve extended attributes on a vnode.
1651  */
1652 static int
1653 ffs_listextattr(struct vop_listextattr_args *ap)
1654 /*
1655 vop_listextattr {
1656 	IN struct vnode *a_vp;
1657 	IN int a_attrnamespace;
1658 	INOUT struct uio *a_uio;
1659 	OUT size_t *a_size;
1660 	IN struct ucred *a_cred;
1661 	IN struct thread *a_td;
1662 };
1663 */
1664 {
1665 	struct inode *ip;
1666 	struct extattr *eap, *eaend;
1667 	int error, ealen;
1668 
1669 	ip = VTOI(ap->a_vp);
1670 
1671 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1672 		return (EOPNOTSUPP);
1673 
1674 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1675 	    ap->a_cred, ap->a_td, VREAD);
1676 	if (error)
1677 		return (error);
1678 
1679 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1680 	if (error)
1681 		return (error);
1682 
1683 	error = 0;
1684 	if (ap->a_size != NULL)
1685 		*ap->a_size = 0;
1686 
1687 	KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned"));
1688 	eap = (struct extattr *)ip->i_ea_area;
1689 	eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len);
1690 	for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) {
1691 		KASSERT(EXTATTR_NEXT(eap) <= eaend,
1692 		    ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1693 		if (eap->ea_namespace != ap->a_attrnamespace)
1694 			continue;
1695 
1696 		ealen = eap->ea_namelength;
1697 		if (ap->a_size != NULL)
1698 			*ap->a_size += ealen + 1;
1699 		else if (ap->a_uio != NULL)
1700 			error = uiomove(&eap->ea_namelength, ealen + 1,
1701 			    ap->a_uio);
1702 	}
1703 
1704 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1705 	return (error);
1706 }
1707 
1708 /*
1709  * Vnode operation to set a named attribute.
1710  */
1711 static int
1712 ffs_setextattr(struct vop_setextattr_args *ap)
1713 /*
1714 vop_setextattr {
1715 	IN struct vnode *a_vp;
1716 	IN int a_attrnamespace;
1717 	IN const char *a_name;
1718 	INOUT struct uio *a_uio;
1719 	IN struct ucred *a_cred;
1720 	IN struct thread *a_td;
1721 };
1722 */
1723 {
1724 	struct inode *ip;
1725 	struct fs *fs;
1726 	struct extattr *eap;
1727 	uint32_t ealength, ul;
1728 	ssize_t ealen;
1729 	int olen, eapad1, eapad2, error, i, easize;
1730 	u_char *eae;
1731 	void *tmp;
1732 
1733 	ip = VTOI(ap->a_vp);
1734 	fs = ITOFS(ip);
1735 
1736 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1737 		return (EOPNOTSUPP);
1738 
1739 	if (strlen(ap->a_name) == 0)
1740 		return (EINVAL);
1741 
1742 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1743 	if (ap->a_uio == NULL)
1744 		return (EOPNOTSUPP);
1745 
1746 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1747 		return (EROFS);
1748 
1749 	ealen = ap->a_uio->uio_resid;
1750 	if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR))
1751 		return (EINVAL);
1752 
1753 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1754 	    ap->a_cred, ap->a_td, VWRITE);
1755 	if (error) {
1756 		/*
1757 		 * ffs_lock_ea is not needed there, because the vnode
1758 		 * must be exclusively locked.
1759 		 */
1760 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1761 			ip->i_ea_error = error;
1762 		return (error);
1763 	}
1764 
1765 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1766 	if (error)
1767 		return (error);
1768 
1769 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1770 	eapad1 = roundup2(ealength, 8) - ealength;
1771 	eapad2 = roundup2(ealen, 8) - ealen;
1772 	ealength += eapad1 + ealen + eapad2;
1773 
1774 	/*
1775 	 * CEM: rewrites of the same size or smaller could be done in-place
1776 	 * instead.  (We don't acquire any fine-grained locks in here either,
1777 	 * so we could also do bigger writes in-place.)
1778 	 */
1779 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1780 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1781 	easize = ip->i_ea_len;
1782 
1783 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1784 	    &eap, NULL);
1785         if (olen == -1) {
1786 		/* new, append at end */
1787 		KASSERT(ALIGNED_TO(eae + easize, struct extattr),
1788 		    ("unaligned"));
1789 		eap = (struct extattr *)(eae + easize);
1790 		easize += ealength;
1791 	} else {
1792 		ul = eap->ea_length;
1793 		i = (u_char *)EXTATTR_NEXT(eap) - eae;
1794 		if (ul != ealength) {
1795 			bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength,
1796 			    easize - i);
1797 			easize += (ealength - ul);
1798 		}
1799 	}
1800 	if (easize > lblktosize(fs, UFS_NXADDR)) {
1801 		free(eae, M_TEMP);
1802 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1803 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1804 			ip->i_ea_error = ENOSPC;
1805 		return (ENOSPC);
1806 	}
1807 	eap->ea_length = ealength;
1808 	eap->ea_namespace = ap->a_attrnamespace;
1809 	eap->ea_contentpadlen = eapad2;
1810 	eap->ea_namelength = strlen(ap->a_name);
1811 	memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name));
1812 	bzero(&eap->ea_name[strlen(ap->a_name)], eapad1);
1813 	error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio);
1814 	if (error) {
1815 		free(eae, M_TEMP);
1816 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1817 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1818 			ip->i_ea_error = error;
1819 		return (error);
1820 	}
1821 	bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2);
1822 
1823 	tmp = ip->i_ea_area;
1824 	ip->i_ea_area = eae;
1825 	ip->i_ea_len = easize;
1826 	free(tmp, M_TEMP);
1827 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1828 	return (error);
1829 }
1830 
1831 /*
1832  * Vnode pointer to File handle
1833  */
1834 static int
1835 ffs_vptofh(struct vop_vptofh_args *ap)
1836 /*
1837 vop_vptofh {
1838 	IN struct vnode *a_vp;
1839 	IN struct fid *a_fhp;
1840 };
1841 */
1842 {
1843 	struct inode *ip;
1844 	struct ufid *ufhp;
1845 
1846 	ip = VTOI(ap->a_vp);
1847 	ufhp = (struct ufid *)ap->a_fhp;
1848 	ufhp->ufid_len = sizeof(struct ufid);
1849 	ufhp->ufid_ino = ip->i_number;
1850 	ufhp->ufid_gen = ip->i_gen;
1851 	return (0);
1852 }
1853 
1854 SYSCTL_DECL(_vfs_ffs);
1855 static int use_buf_pager = 1;
1856 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1857     "Always use buffer pager instead of bmap");
1858 
1859 static daddr_t
1860 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1861 {
1862 
1863 	return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1864 }
1865 
1866 static int
1867 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1868 {
1869 
1870 	return (blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn));
1871 }
1872 
1873 static int
1874 ffs_getpages(struct vop_getpages_args *ap)
1875 {
1876 	struct vnode *vp;
1877 	struct ufsmount *um;
1878 
1879 	vp = ap->a_vp;
1880 	um = VFSTOUFS(vp->v_mount);
1881 
1882 	if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1883 		return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1884 		    ap->a_rbehind, ap->a_rahead, NULL, NULL));
1885 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1886 	    ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));
1887 }
1888 
1889 static int
1890 ffs_getpages_async(struct vop_getpages_async_args *ap)
1891 {
1892 	struct vnode *vp;
1893 	struct ufsmount *um;
1894 	bool do_iodone;
1895 	int error;
1896 
1897 	vp = ap->a_vp;
1898 	um = VFSTOUFS(vp->v_mount);
1899 	do_iodone = true;
1900 
1901 	if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) {
1902 		error = vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1903 		    ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg);
1904 		if (error == 0)
1905 			do_iodone = false;
1906 	} else {
1907 		error = vfs_bio_getpages(vp, ap->a_m, ap->a_count,
1908 		    ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno,
1909 		    ffs_gbp_getblksz);
1910 	}
1911 	if (do_iodone && ap->a_iodone != NULL)
1912 		ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
1913 
1914 	return (error);
1915 }
1916