xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 147972555f2c70f64cc54182dc18326456e46b92)
1 /*-
2  * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Marshall
6  * Kirk McKusick and Network Associates Laboratories, the Security
7  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
9  * research program
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * Copyright (c) 1982, 1986, 1989, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 4. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	from: @(#)ufs_readwrite.c	8.11 (Berkeley) 5/8/95
60  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
61  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
62  */
63 
64 #include <sys/cdefs.h>
65 __FBSDID("$FreeBSD$");
66 
67 #include <sys/param.h>
68 #include <sys/bio.h>
69 #include <sys/systm.h>
70 #include <sys/buf.h>
71 #include <sys/conf.h>
72 #include <sys/extattr.h>
73 #include <sys/kernel.h>
74 #include <sys/limits.h>
75 #include <sys/malloc.h>
76 #include <sys/mount.h>
77 #include <sys/priv.h>
78 #include <sys/stat.h>
79 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vnode_pager.h>
88 
89 #include <ufs/ufs/extattr.h>
90 #include <ufs/ufs/quota.h>
91 #include <ufs/ufs/inode.h>
92 #include <ufs/ufs/ufs_extern.h>
93 #include <ufs/ufs/ufsmount.h>
94 
95 #include <ufs/ffs/fs.h>
96 #include <ufs/ffs/ffs_extern.h>
97 #include "opt_directio.h"
98 #include "opt_ffs.h"
99 
100 #ifdef DIRECTIO
101 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
102 #endif
103 static vop_fsync_t	ffs_fsync;
104 static vop_lock1_t	ffs_lock;
105 static vop_getpages_t	ffs_getpages;
106 static vop_read_t	ffs_read;
107 static vop_write_t	ffs_write;
108 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
109 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
110 		    struct ucred *cred);
111 static vop_strategy_t	ffsext_strategy;
112 static vop_closeextattr_t	ffs_closeextattr;
113 static vop_deleteextattr_t	ffs_deleteextattr;
114 static vop_getextattr_t	ffs_getextattr;
115 static vop_listextattr_t	ffs_listextattr;
116 static vop_openextattr_t	ffs_openextattr;
117 static vop_setextattr_t	ffs_setextattr;
118 static vop_vptofh_t	ffs_vptofh;
119 
120 
121 /* Global vfs data structures for ufs. */
122 struct vop_vector ffs_vnodeops1 = {
123 	.vop_default =		&ufs_vnodeops,
124 	.vop_fsync =		ffs_fsync,
125 	.vop_getpages =		ffs_getpages,
126 	.vop_lock1 =		ffs_lock,
127 	.vop_read =		ffs_read,
128 	.vop_reallocblks =	ffs_reallocblks,
129 	.vop_write =		ffs_write,
130 	.vop_vptofh =		ffs_vptofh,
131 };
132 
133 struct vop_vector ffs_fifoops1 = {
134 	.vop_default =		&ufs_fifoops,
135 	.vop_fsync =		ffs_fsync,
136 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
137 	.vop_vptofh =		ffs_vptofh,
138 };
139 
140 /* Global vfs data structures for ufs. */
141 struct vop_vector ffs_vnodeops2 = {
142 	.vop_default =		&ufs_vnodeops,
143 	.vop_fsync =		ffs_fsync,
144 	.vop_getpages =		ffs_getpages,
145 	.vop_lock1 =		ffs_lock,
146 	.vop_read =		ffs_read,
147 	.vop_reallocblks =	ffs_reallocblks,
148 	.vop_write =		ffs_write,
149 	.vop_closeextattr =	ffs_closeextattr,
150 	.vop_deleteextattr =	ffs_deleteextattr,
151 	.vop_getextattr =	ffs_getextattr,
152 	.vop_listextattr =	ffs_listextattr,
153 	.vop_openextattr =	ffs_openextattr,
154 	.vop_setextattr =	ffs_setextattr,
155 	.vop_vptofh =		ffs_vptofh,
156 };
157 
158 struct vop_vector ffs_fifoops2 = {
159 	.vop_default =		&ufs_fifoops,
160 	.vop_fsync =		ffs_fsync,
161 	.vop_lock1 =		ffs_lock,
162 	.vop_reallocblks =	ffs_reallocblks,
163 	.vop_strategy =		ffsext_strategy,
164 	.vop_closeextattr =	ffs_closeextattr,
165 	.vop_deleteextattr =	ffs_deleteextattr,
166 	.vop_getextattr =	ffs_getextattr,
167 	.vop_listextattr =	ffs_listextattr,
168 	.vop_openextattr =	ffs_openextattr,
169 	.vop_setextattr =	ffs_setextattr,
170 	.vop_vptofh =		ffs_vptofh,
171 };
172 
173 /*
174  * Synch an open file.
175  */
176 /* ARGSUSED */
177 static int
178 ffs_fsync(struct vop_fsync_args *ap)
179 {
180 	struct vnode *vp;
181 	struct bufobj *bo;
182 	int error;
183 
184 	vp = ap->a_vp;
185 	bo = &vp->v_bufobj;
186 retry:
187 	error = ffs_syncvnode(vp, ap->a_waitfor);
188 	if (error)
189 		return (error);
190 	if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
191 		error = softdep_fsync(vp);
192 		if (error)
193 			return (error);
194 
195 		/*
196 		 * The softdep_fsync() function may drop vp lock,
197 		 * allowing for dirty buffers to reappear on the
198 		 * bo_dirty list. Recheck and resync as needed.
199 		 */
200 		BO_LOCK(bo);
201 		if (vp->v_type == VREG && (bo->bo_numoutput > 0 ||
202 		    bo->bo_dirty.bv_cnt > 0)) {
203 			BO_UNLOCK(bo);
204 			goto retry;
205 		}
206 		BO_UNLOCK(bo);
207 	}
208 	return (0);
209 }
210 
211 int
212 ffs_syncvnode(struct vnode *vp, int waitfor)
213 {
214 	struct inode *ip;
215 	struct bufobj *bo;
216 	struct buf *bp;
217 	struct buf *nbp;
218 	ufs_lbn_t lbn;
219 	int error, wait, passes, noupdate;
220 
221 	noupdate = waitfor & NO_INO_UPDT;
222 	waitfor &= ~NO_INO_UPDT;
223 	ip = VTOI(vp);
224 	ip->i_flag &= ~IN_NEEDSYNC;
225 	bo = &vp->v_bufobj;
226 
227 	/*
228 	 * When doing MNT_WAIT we must first flush all dependencies
229 	 * on the inode.
230 	 */
231 	if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
232 	    (error = softdep_sync_metadata(vp)) != 0)
233 		return (error);
234 
235 	/*
236 	 * Flush all dirty buffers associated with a vnode.
237 	 */
238 	error = 0;
239 	passes = 0;
240 	wait = 0;	/* Always do an async pass first. */
241 	lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
242 	BO_LOCK(bo);
243 loop:
244 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
245 		bp->b_vflags &= ~BV_SCANNED;
246 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
247 		/*
248 		 * Reasons to skip this buffer: it has already been considered
249 		 * on this pass, the buffer has dependencies that will cause
250 		 * it to be redirtied and it has not already been deferred,
251 		 * or it is already being written.
252 		 */
253 		if ((bp->b_vflags & BV_SCANNED) != 0)
254 			continue;
255 		bp->b_vflags |= BV_SCANNED;
256 		/* Flush indirects in order. */
257 		if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR &&
258 		    lbn_level(bp->b_lblkno) >= passes)
259 			continue;
260 		if (bp->b_lblkno > lbn)
261 			panic("ffs_syncvnode: syncing truncated data.");
262 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
263 			continue;
264 		BO_UNLOCK(bo);
265 		if ((bp->b_flags & B_DELWRI) == 0)
266 			panic("ffs_fsync: not dirty");
267 		/*
268 		 * Check for dependencies and potentially complete them.
269 		 */
270 		if (!LIST_EMPTY(&bp->b_dep) &&
271 		    (error = softdep_sync_buf(vp, bp,
272 		    wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
273 			/* I/O error. */
274 			if (error != EBUSY) {
275 				BUF_UNLOCK(bp);
276 				return (error);
277 			}
278 			/* If we deferred once, don't defer again. */
279 		    	if ((bp->b_flags & B_DEFERRED) == 0) {
280 				bp->b_flags |= B_DEFERRED;
281 				BUF_UNLOCK(bp);
282 				goto next;
283 			}
284 		}
285 		if (wait) {
286 			bremfree(bp);
287 			if ((error = bwrite(bp)) != 0)
288 				return (error);
289 		} else if ((bp->b_flags & B_CLUSTEROK)) {
290 			(void) vfs_bio_awrite(bp);
291 		} else {
292 			bremfree(bp);
293 			(void) bawrite(bp);
294 		}
295 next:
296 		/*
297 		 * Since we may have slept during the I/O, we need
298 		 * to start from a known point.
299 		 */
300 		BO_LOCK(bo);
301 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
302 	}
303 	if (waitfor != MNT_WAIT) {
304 		BO_UNLOCK(bo);
305 		if (noupdate)
306 			return (0);
307 		else
308 			return (ffs_update(vp, 0));
309 	}
310 	/* Drain IO to see if we're done. */
311 	bufobj_wwait(bo, 0, 0);
312 	/*
313 	 * Block devices associated with filesystems may have new I/O
314 	 * requests posted for them even if the vnode is locked, so no
315 	 * amount of trying will get them clean.  We make several passes
316 	 * as a best effort.
317 	 *
318 	 * Regular files may need multiple passes to flush all dependency
319 	 * work as it is possible that we must write once per indirect
320 	 * level, once for the leaf, and once for the inode and each of
321 	 * these will be done with one sync and one async pass.
322 	 */
323 	if (bo->bo_dirty.bv_cnt > 0) {
324 		/* Write the inode after sync passes to flush deps. */
325 		if (wait && DOINGSOFTDEP(vp) && noupdate == 0) {
326 			BO_UNLOCK(bo);
327 			ffs_update(vp, MNT_WAIT);
328 			BO_LOCK(bo);
329 		}
330 		/* switch between sync/async. */
331 		wait = !wait;
332 		if (wait == 1 || ++passes < NIADDR + 2)
333 			goto loop;
334 #ifdef INVARIANTS
335 		if (!vn_isdisk(vp, NULL))
336 			vprint("ffs_fsync: dirty", vp);
337 #endif
338 	}
339 	BO_UNLOCK(bo);
340 	error = 0;
341 	if (noupdate == 0)
342 		error = ffs_update(vp, MNT_WAIT);
343 	if (DOINGSUJ(vp))
344 		softdep_journal_fsync(VTOI(vp));
345 	return (error);
346 }
347 
348 static int
349 ffs_lock(ap)
350 	struct vop_lock1_args /* {
351 		struct vnode *a_vp;
352 		int a_flags;
353 		struct thread *a_td;
354 		char *file;
355 		int line;
356 	} */ *ap;
357 {
358 #ifndef NO_FFS_SNAPSHOT
359 	struct vnode *vp;
360 	int flags;
361 	struct lock *lkp;
362 	int result;
363 
364 	switch (ap->a_flags & LK_TYPE_MASK) {
365 	case LK_SHARED:
366 	case LK_UPGRADE:
367 	case LK_EXCLUSIVE:
368 		vp = ap->a_vp;
369 		flags = ap->a_flags;
370 		for (;;) {
371 #ifdef DEBUG_VFS_LOCKS
372 			KASSERT(vp->v_holdcnt != 0,
373 			    ("ffs_lock %p: zero hold count", vp));
374 #endif
375 			lkp = vp->v_vnlock;
376 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
377 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
378 			    ap->a_file, ap->a_line);
379 			if (lkp == vp->v_vnlock || result != 0)
380 				break;
381 			/*
382 			 * Apparent success, except that the vnode
383 			 * mutated between snapshot file vnode and
384 			 * regular file vnode while this process
385 			 * slept.  The lock currently held is not the
386 			 * right lock.  Release it, and try to get the
387 			 * new lock.
388 			 */
389 			(void) _lockmgr_args(lkp, LK_RELEASE, NULL,
390 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
391 			    ap->a_file, ap->a_line);
392 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
393 			    (LK_INTERLOCK | LK_NOWAIT))
394 				return (EBUSY);
395 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
396 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
397 			flags &= ~LK_INTERLOCK;
398 		}
399 		break;
400 	default:
401 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
402 	}
403 	return (result);
404 #else
405 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
406 #endif
407 }
408 
409 /*
410  * Vnode op for reading.
411  */
412 static int
413 ffs_read(ap)
414 	struct vop_read_args /* {
415 		struct vnode *a_vp;
416 		struct uio *a_uio;
417 		int a_ioflag;
418 		struct ucred *a_cred;
419 	} */ *ap;
420 {
421 	struct vnode *vp;
422 	struct inode *ip;
423 	struct uio *uio;
424 	struct fs *fs;
425 	struct buf *bp;
426 	ufs_lbn_t lbn, nextlbn;
427 	off_t bytesinfile;
428 	long size, xfersize, blkoffset;
429 	ssize_t orig_resid;
430 	int error;
431 	int seqcount;
432 	int ioflag;
433 
434 	vp = ap->a_vp;
435 	uio = ap->a_uio;
436 	ioflag = ap->a_ioflag;
437 	if (ap->a_ioflag & IO_EXT)
438 #ifdef notyet
439 		return (ffs_extread(vp, uio, ioflag));
440 #else
441 		panic("ffs_read+IO_EXT");
442 #endif
443 #ifdef DIRECTIO
444 	if ((ioflag & IO_DIRECT) != 0) {
445 		int workdone;
446 
447 		error = ffs_rawread(vp, uio, &workdone);
448 		if (error != 0 || workdone != 0)
449 			return error;
450 	}
451 #endif
452 
453 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
454 	ip = VTOI(vp);
455 
456 #ifdef INVARIANTS
457 	if (uio->uio_rw != UIO_READ)
458 		panic("ffs_read: mode");
459 
460 	if (vp->v_type == VLNK) {
461 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
462 			panic("ffs_read: short symlink");
463 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
464 		panic("ffs_read: type %d",  vp->v_type);
465 #endif
466 	orig_resid = uio->uio_resid;
467 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
468 	if (orig_resid == 0)
469 		return (0);
470 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
471 	fs = ip->i_fs;
472 	if (uio->uio_offset < ip->i_size &&
473 	    uio->uio_offset >= fs->fs_maxfilesize)
474 		return (EOVERFLOW);
475 
476 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
477 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
478 			break;
479 		lbn = lblkno(fs, uio->uio_offset);
480 		nextlbn = lbn + 1;
481 
482 		/*
483 		 * size of buffer.  The buffer representing the
484 		 * end of the file is rounded up to the size of
485 		 * the block type ( fragment or full block,
486 		 * depending ).
487 		 */
488 		size = blksize(fs, ip, lbn);
489 		blkoffset = blkoff(fs, uio->uio_offset);
490 
491 		/*
492 		 * The amount we want to transfer in this iteration is
493 		 * one FS block less the amount of the data before
494 		 * our startpoint (duh!)
495 		 */
496 		xfersize = fs->fs_bsize - blkoffset;
497 
498 		/*
499 		 * But if we actually want less than the block,
500 		 * or the file doesn't have a whole block more of data,
501 		 * then use the lesser number.
502 		 */
503 		if (uio->uio_resid < xfersize)
504 			xfersize = uio->uio_resid;
505 		if (bytesinfile < xfersize)
506 			xfersize = bytesinfile;
507 
508 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
509 			/*
510 			 * Don't do readahead if this is the end of the file.
511 			 */
512 			error = bread(vp, lbn, size, NOCRED, &bp);
513 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
514 			/*
515 			 * Otherwise if we are allowed to cluster,
516 			 * grab as much as we can.
517 			 *
518 			 * XXX  This may not be a win if we are not
519 			 * doing sequential access.
520 			 */
521 			error = cluster_read(vp, ip->i_size, lbn,
522 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
523 		} else if (seqcount > 1) {
524 			/*
525 			 * If we are NOT allowed to cluster, then
526 			 * if we appear to be acting sequentially,
527 			 * fire off a request for a readahead
528 			 * as well as a read. Note that the 4th and 5th
529 			 * arguments point to arrays of the size specified in
530 			 * the 6th argument.
531 			 */
532 			int nextsize = blksize(fs, ip, nextlbn);
533 			error = breadn(vp, lbn,
534 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
535 		} else {
536 			/*
537 			 * Failing all of the above, just read what the
538 			 * user asked for. Interestingly, the same as
539 			 * the first option above.
540 			 */
541 			error = bread(vp, lbn, size, NOCRED, &bp);
542 		}
543 		if (error) {
544 			brelse(bp);
545 			bp = NULL;
546 			break;
547 		}
548 
549 		/*
550 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
551 		 * will cause us to attempt to release the buffer later on
552 		 * and will cause the buffer cache to attempt to free the
553 		 * underlying pages.
554 		 */
555 		if (ioflag & IO_DIRECT)
556 			bp->b_flags |= B_DIRECT;
557 
558 		/*
559 		 * We should only get non-zero b_resid when an I/O error
560 		 * has occurred, which should cause us to break above.
561 		 * However, if the short read did not cause an error,
562 		 * then we want to ensure that we do not uiomove bad
563 		 * or uninitialized data.
564 		 */
565 		size -= bp->b_resid;
566 		if (size < xfersize) {
567 			if (size == 0)
568 				break;
569 			xfersize = size;
570 		}
571 
572 		error = uiomove((char *)bp->b_data + blkoffset,
573 		    (int)xfersize, uio);
574 		if (error)
575 			break;
576 
577 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
578 		   (LIST_EMPTY(&bp->b_dep))) {
579 			/*
580 			 * If there are no dependencies, and it's VMIO,
581 			 * then we don't need the buf, mark it available
582 			 * for freeing.  For non-direct VMIO reads, the VM
583 			 * has the data.
584 			 */
585 			bp->b_flags |= B_RELBUF;
586 			brelse(bp);
587 		} else {
588 			/*
589 			 * Otherwise let whoever
590 			 * made the request take care of
591 			 * freeing it. We just queue
592 			 * it onto another list.
593 			 */
594 			bqrelse(bp);
595 		}
596 	}
597 
598 	/*
599 	 * This can only happen in the case of an error
600 	 * because the loop above resets bp to NULL on each iteration
601 	 * and on normal completion has not set a new value into it.
602 	 * so it must have come from a 'break' statement
603 	 */
604 	if (bp != NULL) {
605 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
606 		   (LIST_EMPTY(&bp->b_dep))) {
607 			bp->b_flags |= B_RELBUF;
608 			brelse(bp);
609 		} else {
610 			bqrelse(bp);
611 		}
612 	}
613 
614 	if ((error == 0 || uio->uio_resid != orig_resid) &&
615 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
616 	    (ip->i_flag & IN_ACCESS) == 0) {
617 		VI_LOCK(vp);
618 		ip->i_flag |= IN_ACCESS;
619 		VI_UNLOCK(vp);
620 	}
621 	return (error);
622 }
623 
624 /*
625  * Vnode op for writing.
626  */
627 static int
628 ffs_write(ap)
629 	struct vop_write_args /* {
630 		struct vnode *a_vp;
631 		struct uio *a_uio;
632 		int a_ioflag;
633 		struct ucred *a_cred;
634 	} */ *ap;
635 {
636 	struct vnode *vp;
637 	struct uio *uio;
638 	struct inode *ip;
639 	struct fs *fs;
640 	struct buf *bp;
641 	ufs_lbn_t lbn;
642 	off_t osize;
643 	ssize_t resid;
644 	int seqcount;
645 	int blkoffset, error, flags, ioflag, size, xfersize;
646 
647 	vp = ap->a_vp;
648 	uio = ap->a_uio;
649 	ioflag = ap->a_ioflag;
650 	if (ap->a_ioflag & IO_EXT)
651 #ifdef notyet
652 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
653 #else
654 		panic("ffs_write+IO_EXT");
655 #endif
656 
657 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
658 	ip = VTOI(vp);
659 
660 #ifdef INVARIANTS
661 	if (uio->uio_rw != UIO_WRITE)
662 		panic("ffs_write: mode");
663 #endif
664 
665 	switch (vp->v_type) {
666 	case VREG:
667 		if (ioflag & IO_APPEND)
668 			uio->uio_offset = ip->i_size;
669 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
670 			return (EPERM);
671 		/* FALLTHROUGH */
672 	case VLNK:
673 		break;
674 	case VDIR:
675 		panic("ffs_write: dir write");
676 		break;
677 	default:
678 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
679 			(int)uio->uio_offset,
680 			(int)uio->uio_resid
681 		);
682 	}
683 
684 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
685 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
686 	fs = ip->i_fs;
687 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
688 		return (EFBIG);
689 	/*
690 	 * Maybe this should be above the vnode op call, but so long as
691 	 * file servers have no limits, I don't think it matters.
692 	 */
693 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
694 		return (EFBIG);
695 
696 	resid = uio->uio_resid;
697 	osize = ip->i_size;
698 	if (seqcount > BA_SEQMAX)
699 		flags = BA_SEQMAX << BA_SEQSHIFT;
700 	else
701 		flags = seqcount << BA_SEQSHIFT;
702 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
703 		flags |= IO_SYNC;
704 
705 	for (error = 0; uio->uio_resid > 0;) {
706 		lbn = lblkno(fs, uio->uio_offset);
707 		blkoffset = blkoff(fs, uio->uio_offset);
708 		xfersize = fs->fs_bsize - blkoffset;
709 		if (uio->uio_resid < xfersize)
710 			xfersize = uio->uio_resid;
711 		if (uio->uio_offset + xfersize > ip->i_size)
712 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
713 
714                 /*
715 		 * We must perform a read-before-write if the transfer size
716 		 * does not cover the entire buffer.
717                  */
718 		if (fs->fs_bsize > xfersize)
719 			flags |= BA_CLRBUF;
720 		else
721 			flags &= ~BA_CLRBUF;
722 /* XXX is uio->uio_offset the right thing here? */
723 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
724 		    ap->a_cred, flags, &bp);
725 		if (error != 0) {
726 			vnode_pager_setsize(vp, ip->i_size);
727 			break;
728 		}
729 		if (ioflag & IO_DIRECT)
730 			bp->b_flags |= B_DIRECT;
731 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
732 			bp->b_flags |= B_NOCACHE;
733 
734 		if (uio->uio_offset + xfersize > ip->i_size) {
735 			ip->i_size = uio->uio_offset + xfersize;
736 			DIP_SET(ip, i_size, ip->i_size);
737 		}
738 
739 		size = blksize(fs, ip, lbn) - bp->b_resid;
740 		if (size < xfersize)
741 			xfersize = size;
742 
743 		error =
744 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
745 		/*
746 		 * If the buffer is not already filled and we encounter an
747 		 * error while trying to fill it, we have to clear out any
748 		 * garbage data from the pages instantiated for the buffer.
749 		 * If we do not, a failed uiomove() during a write can leave
750 		 * the prior contents of the pages exposed to a userland mmap.
751 		 *
752 		 * Note that we need only clear buffers with a transfer size
753 		 * equal to the block size because buffers with a shorter
754 		 * transfer size were cleared above by the call to UFS_BALLOC()
755 		 * with the BA_CLRBUF flag set.
756 		 *
757 		 * If the source region for uiomove identically mmaps the
758 		 * buffer, uiomove() performed the NOP copy, and the buffer
759 		 * content remains valid because the page fault handler
760 		 * validated the pages.
761 		 */
762 		if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
763 		    fs->fs_bsize == xfersize)
764 			vfs_bio_clrbuf(bp);
765 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
766 		   (LIST_EMPTY(&bp->b_dep))) {
767 			bp->b_flags |= B_RELBUF;
768 		}
769 
770 		/*
771 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
772 		 * if we have a severe page deficiency write the buffer
773 		 * asynchronously.  Otherwise try to cluster, and if that
774 		 * doesn't do it then either do an async write (if O_DIRECT),
775 		 * or a delayed write (if not).
776 		 */
777 		if (ioflag & IO_SYNC) {
778 			(void)bwrite(bp);
779 		} else if (vm_page_count_severe() ||
780 			    buf_dirty_count_severe() ||
781 			    (ioflag & IO_ASYNC)) {
782 			bp->b_flags |= B_CLUSTEROK;
783 			bawrite(bp);
784 		} else if (xfersize + blkoffset == fs->fs_bsize) {
785 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
786 				bp->b_flags |= B_CLUSTEROK;
787 				cluster_write(vp, bp, ip->i_size, seqcount);
788 			} else {
789 				bawrite(bp);
790 			}
791 		} else if (ioflag & IO_DIRECT) {
792 			bp->b_flags |= B_CLUSTEROK;
793 			bawrite(bp);
794 		} else {
795 			bp->b_flags |= B_CLUSTEROK;
796 			bdwrite(bp);
797 		}
798 		if (error || xfersize == 0)
799 			break;
800 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
801 	}
802 	/*
803 	 * If we successfully wrote any data, and we are not the superuser
804 	 * we clear the setuid and setgid bits as a precaution against
805 	 * tampering.
806 	 */
807 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
808 	    ap->a_cred) {
809 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
810 			ip->i_mode &= ~(ISUID | ISGID);
811 			DIP_SET(ip, i_mode, ip->i_mode);
812 		}
813 	}
814 	if (error) {
815 		if (ioflag & IO_UNIT) {
816 			(void)ffs_truncate(vp, osize,
817 			    IO_NORMAL | (ioflag & IO_SYNC),
818 			    ap->a_cred, uio->uio_td);
819 			uio->uio_offset -= resid - uio->uio_resid;
820 			uio->uio_resid = resid;
821 		}
822 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
823 		error = ffs_update(vp, 1);
824 	return (error);
825 }
826 
827 /*
828  * get page routine
829  */
830 static int
831 ffs_getpages(ap)
832 	struct vop_getpages_args *ap;
833 {
834 	int i;
835 	vm_page_t mreq;
836 	int pcount;
837 
838 	pcount = round_page(ap->a_count) / PAGE_SIZE;
839 	mreq = ap->a_m[ap->a_reqpage];
840 
841 	/*
842 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
843 	 * then the entire page is valid.  Since the page may be mapped,
844 	 * user programs might reference data beyond the actual end of file
845 	 * occuring within the page.  We have to zero that data.
846 	 */
847 	VM_OBJECT_LOCK(mreq->object);
848 	if (mreq->valid) {
849 		if (mreq->valid != VM_PAGE_BITS_ALL)
850 			vm_page_zero_invalid(mreq, TRUE);
851 		for (i = 0; i < pcount; i++) {
852 			if (i != ap->a_reqpage) {
853 				vm_page_lock(ap->a_m[i]);
854 				vm_page_free(ap->a_m[i]);
855 				vm_page_unlock(ap->a_m[i]);
856 			}
857 		}
858 		VM_OBJECT_UNLOCK(mreq->object);
859 		return VM_PAGER_OK;
860 	}
861 	VM_OBJECT_UNLOCK(mreq->object);
862 
863 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
864 					    ap->a_count,
865 					    ap->a_reqpage);
866 }
867 
868 
869 /*
870  * Extended attribute area reading.
871  */
872 static int
873 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
874 {
875 	struct inode *ip;
876 	struct ufs2_dinode *dp;
877 	struct fs *fs;
878 	struct buf *bp;
879 	ufs_lbn_t lbn, nextlbn;
880 	off_t bytesinfile;
881 	long size, xfersize, blkoffset;
882 	ssize_t orig_resid;
883 	int error;
884 
885 	ip = VTOI(vp);
886 	fs = ip->i_fs;
887 	dp = ip->i_din2;
888 
889 #ifdef INVARIANTS
890 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
891 		panic("ffs_extread: mode");
892 
893 #endif
894 	orig_resid = uio->uio_resid;
895 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
896 	if (orig_resid == 0)
897 		return (0);
898 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
899 
900 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
901 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
902 			break;
903 		lbn = lblkno(fs, uio->uio_offset);
904 		nextlbn = lbn + 1;
905 
906 		/*
907 		 * size of buffer.  The buffer representing the
908 		 * end of the file is rounded up to the size of
909 		 * the block type ( fragment or full block,
910 		 * depending ).
911 		 */
912 		size = sblksize(fs, dp->di_extsize, lbn);
913 		blkoffset = blkoff(fs, uio->uio_offset);
914 
915 		/*
916 		 * The amount we want to transfer in this iteration is
917 		 * one FS block less the amount of the data before
918 		 * our startpoint (duh!)
919 		 */
920 		xfersize = fs->fs_bsize - blkoffset;
921 
922 		/*
923 		 * But if we actually want less than the block,
924 		 * or the file doesn't have a whole block more of data,
925 		 * then use the lesser number.
926 		 */
927 		if (uio->uio_resid < xfersize)
928 			xfersize = uio->uio_resid;
929 		if (bytesinfile < xfersize)
930 			xfersize = bytesinfile;
931 
932 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
933 			/*
934 			 * Don't do readahead if this is the end of the info.
935 			 */
936 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
937 		} else {
938 			/*
939 			 * If we have a second block, then
940 			 * fire off a request for a readahead
941 			 * as well as a read. Note that the 4th and 5th
942 			 * arguments point to arrays of the size specified in
943 			 * the 6th argument.
944 			 */
945 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
946 
947 			nextlbn = -1 - nextlbn;
948 			error = breadn(vp, -1 - lbn,
949 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
950 		}
951 		if (error) {
952 			brelse(bp);
953 			bp = NULL;
954 			break;
955 		}
956 
957 		/*
958 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
959 		 * will cause us to attempt to release the buffer later on
960 		 * and will cause the buffer cache to attempt to free the
961 		 * underlying pages.
962 		 */
963 		if (ioflag & IO_DIRECT)
964 			bp->b_flags |= B_DIRECT;
965 
966 		/*
967 		 * We should only get non-zero b_resid when an I/O error
968 		 * has occurred, which should cause us to break above.
969 		 * However, if the short read did not cause an error,
970 		 * then we want to ensure that we do not uiomove bad
971 		 * or uninitialized data.
972 		 */
973 		size -= bp->b_resid;
974 		if (size < xfersize) {
975 			if (size == 0)
976 				break;
977 			xfersize = size;
978 		}
979 
980 		error = uiomove((char *)bp->b_data + blkoffset,
981 					(int)xfersize, uio);
982 		if (error)
983 			break;
984 
985 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
986 		   (LIST_EMPTY(&bp->b_dep))) {
987 			/*
988 			 * If there are no dependencies, and it's VMIO,
989 			 * then we don't need the buf, mark it available
990 			 * for freeing.  For non-direct VMIO reads, the VM
991 			 * has the data.
992 			 */
993 			bp->b_flags |= B_RELBUF;
994 			brelse(bp);
995 		} else {
996 			/*
997 			 * Otherwise let whoever
998 			 * made the request take care of
999 			 * freeing it. We just queue
1000 			 * it onto another list.
1001 			 */
1002 			bqrelse(bp);
1003 		}
1004 	}
1005 
1006 	/*
1007 	 * This can only happen in the case of an error
1008 	 * because the loop above resets bp to NULL on each iteration
1009 	 * and on normal completion has not set a new value into it.
1010 	 * so it must have come from a 'break' statement
1011 	 */
1012 	if (bp != NULL) {
1013 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1014 		   (LIST_EMPTY(&bp->b_dep))) {
1015 			bp->b_flags |= B_RELBUF;
1016 			brelse(bp);
1017 		} else {
1018 			bqrelse(bp);
1019 		}
1020 	}
1021 	return (error);
1022 }
1023 
1024 /*
1025  * Extended attribute area writing.
1026  */
1027 static int
1028 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1029 {
1030 	struct inode *ip;
1031 	struct ufs2_dinode *dp;
1032 	struct fs *fs;
1033 	struct buf *bp;
1034 	ufs_lbn_t lbn;
1035 	off_t osize;
1036 	ssize_t resid;
1037 	int blkoffset, error, flags, size, xfersize;
1038 
1039 	ip = VTOI(vp);
1040 	fs = ip->i_fs;
1041 	dp = ip->i_din2;
1042 
1043 #ifdef INVARIANTS
1044 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1045 		panic("ffs_extwrite: mode");
1046 #endif
1047 
1048 	if (ioflag & IO_APPEND)
1049 		uio->uio_offset = dp->di_extsize;
1050 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1051 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1052 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1053 		return (EFBIG);
1054 
1055 	resid = uio->uio_resid;
1056 	osize = dp->di_extsize;
1057 	flags = IO_EXT;
1058 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1059 		flags |= IO_SYNC;
1060 
1061 	for (error = 0; uio->uio_resid > 0;) {
1062 		lbn = lblkno(fs, uio->uio_offset);
1063 		blkoffset = blkoff(fs, uio->uio_offset);
1064 		xfersize = fs->fs_bsize - blkoffset;
1065 		if (uio->uio_resid < xfersize)
1066 			xfersize = uio->uio_resid;
1067 
1068 		/*
1069 		 * We must perform a read-before-write if the transfer size
1070 		 * does not cover the entire buffer.
1071                  */
1072 		if (fs->fs_bsize > xfersize)
1073 			flags |= BA_CLRBUF;
1074 		else
1075 			flags &= ~BA_CLRBUF;
1076 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1077 		    ucred, flags, &bp);
1078 		if (error != 0)
1079 			break;
1080 		/*
1081 		 * If the buffer is not valid we have to clear out any
1082 		 * garbage data from the pages instantiated for the buffer.
1083 		 * If we do not, a failed uiomove() during a write can leave
1084 		 * the prior contents of the pages exposed to a userland
1085 		 * mmap().  XXX deal with uiomove() errors a better way.
1086 		 */
1087 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1088 			vfs_bio_clrbuf(bp);
1089 		if (ioflag & IO_DIRECT)
1090 			bp->b_flags |= B_DIRECT;
1091 
1092 		if (uio->uio_offset + xfersize > dp->di_extsize)
1093 			dp->di_extsize = uio->uio_offset + xfersize;
1094 
1095 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1096 		if (size < xfersize)
1097 			xfersize = size;
1098 
1099 		error =
1100 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1101 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1102 		   (LIST_EMPTY(&bp->b_dep))) {
1103 			bp->b_flags |= B_RELBUF;
1104 		}
1105 
1106 		/*
1107 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1108 		 * if we have a severe page deficiency write the buffer
1109 		 * asynchronously.  Otherwise try to cluster, and if that
1110 		 * doesn't do it then either do an async write (if O_DIRECT),
1111 		 * or a delayed write (if not).
1112 		 */
1113 		if (ioflag & IO_SYNC) {
1114 			(void)bwrite(bp);
1115 		} else if (vm_page_count_severe() ||
1116 			    buf_dirty_count_severe() ||
1117 			    xfersize + blkoffset == fs->fs_bsize ||
1118 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1119 			bawrite(bp);
1120 		else
1121 			bdwrite(bp);
1122 		if (error || xfersize == 0)
1123 			break;
1124 		ip->i_flag |= IN_CHANGE;
1125 	}
1126 	/*
1127 	 * If we successfully wrote any data, and we are not the superuser
1128 	 * we clear the setuid and setgid bits as a precaution against
1129 	 * tampering.
1130 	 */
1131 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1132 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1133 			ip->i_mode &= ~(ISUID | ISGID);
1134 			dp->di_mode = ip->i_mode;
1135 		}
1136 	}
1137 	if (error) {
1138 		if (ioflag & IO_UNIT) {
1139 			(void)ffs_truncate(vp, osize,
1140 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1141 			uio->uio_offset -= resid - uio->uio_resid;
1142 			uio->uio_resid = resid;
1143 		}
1144 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1145 		error = ffs_update(vp, 1);
1146 	return (error);
1147 }
1148 
1149 
1150 /*
1151  * Vnode operating to retrieve a named extended attribute.
1152  *
1153  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1154  * the length of the EA, and possibly the pointer to the entry and to the data.
1155  */
1156 static int
1157 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1158 {
1159 	u_char *p, *pe, *pn, *p0;
1160 	int eapad1, eapad2, ealength, ealen, nlen;
1161 	uint32_t ul;
1162 
1163 	pe = ptr + length;
1164 	nlen = strlen(name);
1165 
1166 	for (p = ptr; p < pe; p = pn) {
1167 		p0 = p;
1168 		bcopy(p, &ul, sizeof(ul));
1169 		pn = p + ul;
1170 		/* make sure this entry is complete */
1171 		if (pn > pe)
1172 			break;
1173 		p += sizeof(uint32_t);
1174 		if (*p != nspace)
1175 			continue;
1176 		p++;
1177 		eapad2 = *p++;
1178 		if (*p != nlen)
1179 			continue;
1180 		p++;
1181 		if (bcmp(p, name, nlen))
1182 			continue;
1183 		ealength = sizeof(uint32_t) + 3 + nlen;
1184 		eapad1 = 8 - (ealength % 8);
1185 		if (eapad1 == 8)
1186 			eapad1 = 0;
1187 		ealength += eapad1;
1188 		ealen = ul - ealength - eapad2;
1189 		p += nlen + eapad1;
1190 		if (eap != NULL)
1191 			*eap = p0;
1192 		if (eac != NULL)
1193 			*eac = p;
1194 		return (ealen);
1195 	}
1196 	return(-1);
1197 }
1198 
1199 static int
1200 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1201 {
1202 	struct inode *ip;
1203 	struct ufs2_dinode *dp;
1204 	struct fs *fs;
1205 	struct uio luio;
1206 	struct iovec liovec;
1207 	int easize, error;
1208 	u_char *eae;
1209 
1210 	ip = VTOI(vp);
1211 	fs = ip->i_fs;
1212 	dp = ip->i_din2;
1213 	easize = dp->di_extsize;
1214 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1215 		return (EFBIG);
1216 
1217 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1218 
1219 	liovec.iov_base = eae;
1220 	liovec.iov_len = easize;
1221 	luio.uio_iov = &liovec;
1222 	luio.uio_iovcnt = 1;
1223 	luio.uio_offset = 0;
1224 	luio.uio_resid = easize;
1225 	luio.uio_segflg = UIO_SYSSPACE;
1226 	luio.uio_rw = UIO_READ;
1227 	luio.uio_td = td;
1228 
1229 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1230 	if (error) {
1231 		free(eae, M_TEMP);
1232 		return(error);
1233 	}
1234 	*p = eae;
1235 	return (0);
1236 }
1237 
1238 static void
1239 ffs_lock_ea(struct vnode *vp)
1240 {
1241 	struct inode *ip;
1242 
1243 	ip = VTOI(vp);
1244 	VI_LOCK(vp);
1245 	while (ip->i_flag & IN_EA_LOCKED) {
1246 		ip->i_flag |= IN_EA_LOCKWAIT;
1247 		msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1248 		    0);
1249 	}
1250 	ip->i_flag |= IN_EA_LOCKED;
1251 	VI_UNLOCK(vp);
1252 }
1253 
1254 static void
1255 ffs_unlock_ea(struct vnode *vp)
1256 {
1257 	struct inode *ip;
1258 
1259 	ip = VTOI(vp);
1260 	VI_LOCK(vp);
1261 	if (ip->i_flag & IN_EA_LOCKWAIT)
1262 		wakeup(&ip->i_ea_refs);
1263 	ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1264 	VI_UNLOCK(vp);
1265 }
1266 
1267 static int
1268 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1269 {
1270 	struct inode *ip;
1271 	struct ufs2_dinode *dp;
1272 	int error;
1273 
1274 	ip = VTOI(vp);
1275 
1276 	ffs_lock_ea(vp);
1277 	if (ip->i_ea_area != NULL) {
1278 		ip->i_ea_refs++;
1279 		ffs_unlock_ea(vp);
1280 		return (0);
1281 	}
1282 	dp = ip->i_din2;
1283 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1284 	if (error) {
1285 		ffs_unlock_ea(vp);
1286 		return (error);
1287 	}
1288 	ip->i_ea_len = dp->di_extsize;
1289 	ip->i_ea_error = 0;
1290 	ip->i_ea_refs++;
1291 	ffs_unlock_ea(vp);
1292 	return (0);
1293 }
1294 
1295 /*
1296  * Vnode extattr transaction commit/abort
1297  */
1298 static int
1299 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1300 {
1301 	struct inode *ip;
1302 	struct uio luio;
1303 	struct iovec liovec;
1304 	int error;
1305 	struct ufs2_dinode *dp;
1306 
1307 	ip = VTOI(vp);
1308 
1309 	ffs_lock_ea(vp);
1310 	if (ip->i_ea_area == NULL) {
1311 		ffs_unlock_ea(vp);
1312 		return (EINVAL);
1313 	}
1314 	dp = ip->i_din2;
1315 	error = ip->i_ea_error;
1316 	if (commit && error == 0) {
1317 		ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1318 		if (cred == NOCRED)
1319 			cred =  vp->v_mount->mnt_cred;
1320 		liovec.iov_base = ip->i_ea_area;
1321 		liovec.iov_len = ip->i_ea_len;
1322 		luio.uio_iov = &liovec;
1323 		luio.uio_iovcnt = 1;
1324 		luio.uio_offset = 0;
1325 		luio.uio_resid = ip->i_ea_len;
1326 		luio.uio_segflg = UIO_SYSSPACE;
1327 		luio.uio_rw = UIO_WRITE;
1328 		luio.uio_td = td;
1329 		/* XXX: I'm not happy about truncating to zero size */
1330 		if (ip->i_ea_len < dp->di_extsize)
1331 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1332 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1333 	}
1334 	if (--ip->i_ea_refs == 0) {
1335 		free(ip->i_ea_area, M_TEMP);
1336 		ip->i_ea_area = NULL;
1337 		ip->i_ea_len = 0;
1338 		ip->i_ea_error = 0;
1339 	}
1340 	ffs_unlock_ea(vp);
1341 	return (error);
1342 }
1343 
1344 /*
1345  * Vnode extattr strategy routine for fifos.
1346  *
1347  * We need to check for a read or write of the external attributes.
1348  * Otherwise we just fall through and do the usual thing.
1349  */
1350 static int
1351 ffsext_strategy(struct vop_strategy_args *ap)
1352 /*
1353 struct vop_strategy_args {
1354 	struct vnodeop_desc *a_desc;
1355 	struct vnode *a_vp;
1356 	struct buf *a_bp;
1357 };
1358 */
1359 {
1360 	struct vnode *vp;
1361 	daddr_t lbn;
1362 
1363 	vp = ap->a_vp;
1364 	lbn = ap->a_bp->b_lblkno;
1365 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1366 	    lbn < 0 && lbn >= -NXADDR)
1367 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1368 	if (vp->v_type == VFIFO)
1369 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1370 	panic("spec nodes went here");
1371 }
1372 
1373 /*
1374  * Vnode extattr transaction commit/abort
1375  */
1376 static int
1377 ffs_openextattr(struct vop_openextattr_args *ap)
1378 /*
1379 struct vop_openextattr_args {
1380 	struct vnodeop_desc *a_desc;
1381 	struct vnode *a_vp;
1382 	IN struct ucred *a_cred;
1383 	IN struct thread *a_td;
1384 };
1385 */
1386 {
1387 	struct inode *ip;
1388 	struct fs *fs;
1389 
1390 	ip = VTOI(ap->a_vp);
1391 	fs = ip->i_fs;
1392 
1393 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1394 		return (EOPNOTSUPP);
1395 
1396 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1397 }
1398 
1399 
1400 /*
1401  * Vnode extattr transaction commit/abort
1402  */
1403 static int
1404 ffs_closeextattr(struct vop_closeextattr_args *ap)
1405 /*
1406 struct vop_closeextattr_args {
1407 	struct vnodeop_desc *a_desc;
1408 	struct vnode *a_vp;
1409 	int a_commit;
1410 	IN struct ucred *a_cred;
1411 	IN struct thread *a_td;
1412 };
1413 */
1414 {
1415 	struct inode *ip;
1416 	struct fs *fs;
1417 
1418 	ip = VTOI(ap->a_vp);
1419 	fs = ip->i_fs;
1420 
1421 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1422 		return (EOPNOTSUPP);
1423 
1424 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1425 		return (EROFS);
1426 
1427 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1428 }
1429 
1430 /*
1431  * Vnode operation to remove a named attribute.
1432  */
1433 static int
1434 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1435 /*
1436 vop_deleteextattr {
1437 	IN struct vnode *a_vp;
1438 	IN int a_attrnamespace;
1439 	IN const char *a_name;
1440 	IN struct ucred *a_cred;
1441 	IN struct thread *a_td;
1442 };
1443 */
1444 {
1445 	struct inode *ip;
1446 	struct fs *fs;
1447 	uint32_t ealength, ul;
1448 	int ealen, olen, eapad1, eapad2, error, i, easize;
1449 	u_char *eae, *p;
1450 
1451 	ip = VTOI(ap->a_vp);
1452 	fs = ip->i_fs;
1453 
1454 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1455 		return (EOPNOTSUPP);
1456 
1457 	if (strlen(ap->a_name) == 0)
1458 		return (EINVAL);
1459 
1460 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1461 		return (EROFS);
1462 
1463 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1464 	    ap->a_cred, ap->a_td, VWRITE);
1465 	if (error) {
1466 
1467 		/*
1468 		 * ffs_lock_ea is not needed there, because the vnode
1469 		 * must be exclusively locked.
1470 		 */
1471 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1472 			ip->i_ea_error = error;
1473 		return (error);
1474 	}
1475 
1476 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1477 	if (error)
1478 		return (error);
1479 
1480 	ealength = eapad1 = ealen = eapad2 = 0;
1481 
1482 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1483 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1484 	easize = ip->i_ea_len;
1485 
1486 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1487 	    &p, NULL);
1488 	if (olen == -1) {
1489 		/* delete but nonexistent */
1490 		free(eae, M_TEMP);
1491 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1492 		return(ENOATTR);
1493 	}
1494 	bcopy(p, &ul, sizeof ul);
1495 	i = p - eae + ul;
1496 	if (ul != ealength) {
1497 		bcopy(p + ul, p + ealength, easize - i);
1498 		easize += (ealength - ul);
1499 	}
1500 	if (easize > NXADDR * fs->fs_bsize) {
1501 		free(eae, M_TEMP);
1502 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1503 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1504 			ip->i_ea_error = ENOSPC;
1505 		return(ENOSPC);
1506 	}
1507 	p = ip->i_ea_area;
1508 	ip->i_ea_area = eae;
1509 	ip->i_ea_len = easize;
1510 	free(p, M_TEMP);
1511 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1512 	return(error);
1513 }
1514 
1515 /*
1516  * Vnode operation to retrieve a named extended attribute.
1517  */
1518 static int
1519 ffs_getextattr(struct vop_getextattr_args *ap)
1520 /*
1521 vop_getextattr {
1522 	IN struct vnode *a_vp;
1523 	IN int a_attrnamespace;
1524 	IN const char *a_name;
1525 	INOUT struct uio *a_uio;
1526 	OUT size_t *a_size;
1527 	IN struct ucred *a_cred;
1528 	IN struct thread *a_td;
1529 };
1530 */
1531 {
1532 	struct inode *ip;
1533 	struct fs *fs;
1534 	u_char *eae, *p;
1535 	unsigned easize;
1536 	int error, ealen;
1537 
1538 	ip = VTOI(ap->a_vp);
1539 	fs = ip->i_fs;
1540 
1541 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1542 		return (EOPNOTSUPP);
1543 
1544 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1545 	    ap->a_cred, ap->a_td, VREAD);
1546 	if (error)
1547 		return (error);
1548 
1549 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1550 	if (error)
1551 		return (error);
1552 
1553 	eae = ip->i_ea_area;
1554 	easize = ip->i_ea_len;
1555 
1556 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1557 	    NULL, &p);
1558 	if (ealen >= 0) {
1559 		error = 0;
1560 		if (ap->a_size != NULL)
1561 			*ap->a_size = ealen;
1562 		else if (ap->a_uio != NULL)
1563 			error = uiomove(p, ealen, ap->a_uio);
1564 	} else
1565 		error = ENOATTR;
1566 
1567 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1568 	return(error);
1569 }
1570 
1571 /*
1572  * Vnode operation to retrieve extended attributes on a vnode.
1573  */
1574 static int
1575 ffs_listextattr(struct vop_listextattr_args *ap)
1576 /*
1577 vop_listextattr {
1578 	IN struct vnode *a_vp;
1579 	IN int a_attrnamespace;
1580 	INOUT struct uio *a_uio;
1581 	OUT size_t *a_size;
1582 	IN struct ucred *a_cred;
1583 	IN struct thread *a_td;
1584 };
1585 */
1586 {
1587 	struct inode *ip;
1588 	struct fs *fs;
1589 	u_char *eae, *p, *pe, *pn;
1590 	unsigned easize;
1591 	uint32_t ul;
1592 	int error, ealen;
1593 
1594 	ip = VTOI(ap->a_vp);
1595 	fs = ip->i_fs;
1596 
1597 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1598 		return (EOPNOTSUPP);
1599 
1600 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1601 	    ap->a_cred, ap->a_td, VREAD);
1602 	if (error)
1603 		return (error);
1604 
1605 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1606 	if (error)
1607 		return (error);
1608 	eae = ip->i_ea_area;
1609 	easize = ip->i_ea_len;
1610 
1611 	error = 0;
1612 	if (ap->a_size != NULL)
1613 		*ap->a_size = 0;
1614 	pe = eae + easize;
1615 	for(p = eae; error == 0 && p < pe; p = pn) {
1616 		bcopy(p, &ul, sizeof(ul));
1617 		pn = p + ul;
1618 		if (pn > pe)
1619 			break;
1620 		p += sizeof(ul);
1621 		if (*p++ != ap->a_attrnamespace)
1622 			continue;
1623 		p++;	/* pad2 */
1624 		ealen = *p;
1625 		if (ap->a_size != NULL) {
1626 			*ap->a_size += ealen + 1;
1627 		} else if (ap->a_uio != NULL) {
1628 			error = uiomove(p, ealen + 1, ap->a_uio);
1629 		}
1630 	}
1631 	ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1632 	return(error);
1633 }
1634 
1635 /*
1636  * Vnode operation to set a named attribute.
1637  */
1638 static int
1639 ffs_setextattr(struct vop_setextattr_args *ap)
1640 /*
1641 vop_setextattr {
1642 	IN struct vnode *a_vp;
1643 	IN int a_attrnamespace;
1644 	IN const char *a_name;
1645 	INOUT struct uio *a_uio;
1646 	IN struct ucred *a_cred;
1647 	IN struct thread *a_td;
1648 };
1649 */
1650 {
1651 	struct inode *ip;
1652 	struct fs *fs;
1653 	uint32_t ealength, ul;
1654 	int ealen, olen, eapad1, eapad2, error, i, easize;
1655 	u_char *eae, *p;
1656 
1657 	ip = VTOI(ap->a_vp);
1658 	fs = ip->i_fs;
1659 
1660 	if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1661 		return (EOPNOTSUPP);
1662 
1663 	if (strlen(ap->a_name) == 0)
1664 		return (EINVAL);
1665 
1666 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1667 	if (ap->a_uio == NULL)
1668 		return (EOPNOTSUPP);
1669 
1670 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1671 		return (EROFS);
1672 
1673 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1674 	    ap->a_cred, ap->a_td, VWRITE);
1675 	if (error) {
1676 
1677 		/*
1678 		 * ffs_lock_ea is not needed there, because the vnode
1679 		 * must be exclusively locked.
1680 		 */
1681 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1682 			ip->i_ea_error = error;
1683 		return (error);
1684 	}
1685 
1686 	error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1687 	if (error)
1688 		return (error);
1689 
1690 	ealen = ap->a_uio->uio_resid;
1691 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1692 	eapad1 = 8 - (ealength % 8);
1693 	if (eapad1 == 8)
1694 		eapad1 = 0;
1695 	eapad2 = 8 - (ealen % 8);
1696 	if (eapad2 == 8)
1697 		eapad2 = 0;
1698 	ealength += eapad1 + ealen + eapad2;
1699 
1700 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1701 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1702 	easize = ip->i_ea_len;
1703 
1704 	olen = ffs_findextattr(eae, easize,
1705 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1706         if (olen == -1) {
1707 		/* new, append at end */
1708 		p = eae + easize;
1709 		easize += ealength;
1710 	} else {
1711 		bcopy(p, &ul, sizeof ul);
1712 		i = p - eae + ul;
1713 		if (ul != ealength) {
1714 			bcopy(p + ul, p + ealength, easize - i);
1715 			easize += (ealength - ul);
1716 		}
1717 	}
1718 	if (easize > NXADDR * fs->fs_bsize) {
1719 		free(eae, M_TEMP);
1720 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1721 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1722 			ip->i_ea_error = ENOSPC;
1723 		return(ENOSPC);
1724 	}
1725 	bcopy(&ealength, p, sizeof(ealength));
1726 	p += sizeof(ealength);
1727 	*p++ = ap->a_attrnamespace;
1728 	*p++ = eapad2;
1729 	*p++ = strlen(ap->a_name);
1730 	strcpy(p, ap->a_name);
1731 	p += strlen(ap->a_name);
1732 	bzero(p, eapad1);
1733 	p += eapad1;
1734 	error = uiomove(p, ealen, ap->a_uio);
1735 	if (error) {
1736 		free(eae, M_TEMP);
1737 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1738 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1739 			ip->i_ea_error = error;
1740 		return(error);
1741 	}
1742 	p += ealen;
1743 	bzero(p, eapad2);
1744 
1745 	p = ip->i_ea_area;
1746 	ip->i_ea_area = eae;
1747 	ip->i_ea_len = easize;
1748 	free(p, M_TEMP);
1749 	error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1750 	return(error);
1751 }
1752 
1753 /*
1754  * Vnode pointer to File handle
1755  */
1756 static int
1757 ffs_vptofh(struct vop_vptofh_args *ap)
1758 /*
1759 vop_vptofh {
1760 	IN struct vnode *a_vp;
1761 	IN struct fid *a_fhp;
1762 };
1763 */
1764 {
1765 	struct inode *ip;
1766 	struct ufid *ufhp;
1767 
1768 	ip = VTOI(ap->a_vp);
1769 	ufhp = (struct ufid *)ap->a_fhp;
1770 	ufhp->ufid_len = sizeof(struct ufid);
1771 	ufhp->ufid_ino = ip->i_number;
1772 	ufhp->ufid_gen = ip->i_gen;
1773 	return (0);
1774 }
1775