xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 10f0bcab61ef441cb5af32fb706688d8cbd55dc0)
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/proc.h>
79 #include <sys/resourcevar.h>
80 #include <sys/signalvar.h>
81 #include <sys/stat.h>
82 #include <sys/vmmeter.h>
83 #include <sys/vnode.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vnode_pager.h>
91 
92 #include <ufs/ufs/extattr.h>
93 #include <ufs/ufs/quota.h>
94 #include <ufs/ufs/inode.h>
95 #include <ufs/ufs/ufs_extern.h>
96 #include <ufs/ufs/ufsmount.h>
97 
98 #include <ufs/ffs/fs.h>
99 #include <ufs/ffs/ffs_extern.h>
100 #include "opt_directio.h"
101 #include "opt_ffs.h"
102 
103 #ifdef DIRECTIO
104 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
105 #endif
106 static vop_fsync_t	ffs_fsync;
107 static vop_lock1_t	ffs_lock;
108 static vop_getpages_t	ffs_getpages;
109 static vop_read_t	ffs_read;
110 static vop_write_t	ffs_write;
111 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
112 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
113 		    struct ucred *cred);
114 static vop_strategy_t	ffsext_strategy;
115 static vop_closeextattr_t	ffs_closeextattr;
116 static vop_deleteextattr_t	ffs_deleteextattr;
117 static vop_getextattr_t	ffs_getextattr;
118 static vop_listextattr_t	ffs_listextattr;
119 static vop_openextattr_t	ffs_openextattr;
120 static vop_setextattr_t	ffs_setextattr;
121 static vop_vptofh_t	ffs_vptofh;
122 
123 
124 /* Global vfs data structures for ufs. */
125 struct vop_vector ffs_vnodeops1 = {
126 	.vop_default =		&ufs_vnodeops,
127 	.vop_fsync =		ffs_fsync,
128 	.vop_getpages =		ffs_getpages,
129 	.vop_lock1 =		ffs_lock,
130 	.vop_read =		ffs_read,
131 	.vop_reallocblks =	ffs_reallocblks,
132 	.vop_write =		ffs_write,
133 	.vop_vptofh =		ffs_vptofh,
134 };
135 
136 struct vop_vector ffs_fifoops1 = {
137 	.vop_default =		&ufs_fifoops,
138 	.vop_fsync =		ffs_fsync,
139 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
140 	.vop_vptofh =		ffs_vptofh,
141 };
142 
143 /* Global vfs data structures for ufs. */
144 struct vop_vector ffs_vnodeops2 = {
145 	.vop_default =		&ufs_vnodeops,
146 	.vop_fsync =		ffs_fsync,
147 	.vop_getpages =		ffs_getpages,
148 	.vop_lock1 =		ffs_lock,
149 	.vop_read =		ffs_read,
150 	.vop_reallocblks =	ffs_reallocblks,
151 	.vop_write =		ffs_write,
152 	.vop_closeextattr =	ffs_closeextattr,
153 	.vop_deleteextattr =	ffs_deleteextattr,
154 	.vop_getextattr =	ffs_getextattr,
155 	.vop_listextattr =	ffs_listextattr,
156 	.vop_openextattr =	ffs_openextattr,
157 	.vop_setextattr =	ffs_setextattr,
158 	.vop_vptofh =		ffs_vptofh,
159 };
160 
161 struct vop_vector ffs_fifoops2 = {
162 	.vop_default =		&ufs_fifoops,
163 	.vop_fsync =		ffs_fsync,
164 	.vop_lock1 =		ffs_lock,
165 	.vop_reallocblks =	ffs_reallocblks,
166 	.vop_strategy =		ffsext_strategy,
167 	.vop_closeextattr =	ffs_closeextattr,
168 	.vop_deleteextattr =	ffs_deleteextattr,
169 	.vop_getextattr =	ffs_getextattr,
170 	.vop_listextattr =	ffs_listextattr,
171 	.vop_openextattr =	ffs_openextattr,
172 	.vop_setextattr =	ffs_setextattr,
173 	.vop_vptofh =		ffs_vptofh,
174 };
175 
176 /*
177  * Synch an open file.
178  */
179 /* ARGSUSED */
180 static int
181 ffs_fsync(struct vop_fsync_args *ap)
182 {
183 	int error;
184 
185 	error = ffs_syncvnode(ap->a_vp, ap->a_waitfor);
186 	if (error)
187 		return (error);
188 	if (ap->a_waitfor == MNT_WAIT &&
189 	    (ap->a_vp->v_mount->mnt_flag & MNT_SOFTDEP))
190                 error = softdep_fsync(ap->a_vp);
191 	return (error);
192 }
193 
194 int
195 ffs_syncvnode(struct vnode *vp, int waitfor)
196 {
197 	struct inode *ip = VTOI(vp);
198 	struct bufobj *bo;
199 	struct buf *bp;
200 	struct buf *nbp;
201 	int s, error, wait, passes, skipmeta;
202 	ufs_lbn_t lbn;
203 
204 	wait = (waitfor == MNT_WAIT);
205 	lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
206 	bo = &vp->v_bufobj;
207 
208 	/*
209 	 * Flush all dirty buffers associated with a vnode.
210 	 */
211 	passes = NIADDR + 1;
212 	skipmeta = 0;
213 	if (wait)
214 		skipmeta = 1;
215 	s = splbio();
216 	BO_LOCK(bo);
217 loop:
218 	TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
219 		bp->b_vflags &= ~BV_SCANNED;
220 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
221 		/*
222 		 * Reasons to skip this buffer: it has already been considered
223 		 * on this pass, this pass is the first time through on a
224 		 * synchronous flush request and the buffer being considered
225 		 * is metadata, the buffer has dependencies that will cause
226 		 * it to be redirtied and it has not already been deferred,
227 		 * or it is already being written.
228 		 */
229 		if ((bp->b_vflags & BV_SCANNED) != 0)
230 			continue;
231 		bp->b_vflags |= BV_SCANNED;
232 		if ((skipmeta == 1 && bp->b_lblkno < 0))
233 			continue;
234 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
235 			continue;
236 		BO_UNLOCK(bo);
237 		if (!wait && !LIST_EMPTY(&bp->b_dep) &&
238 		    (bp->b_flags & B_DEFERRED) == 0 &&
239 		    buf_countdeps(bp, 0)) {
240 			bp->b_flags |= B_DEFERRED;
241 			BUF_UNLOCK(bp);
242 			BO_LOCK(bo);
243 			continue;
244 		}
245 		if ((bp->b_flags & B_DELWRI) == 0)
246 			panic("ffs_fsync: not dirty");
247 		/*
248 		 * If this is a synchronous flush request, or it is not a
249 		 * file or device, start the write on this buffer immediately.
250 		 */
251 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
252 
253 			/*
254 			 * On our final pass through, do all I/O synchronously
255 			 * so that we can find out if our flush is failing
256 			 * because of write errors.
257 			 */
258 			if (passes > 0 || !wait) {
259 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
260 					(void) vfs_bio_awrite(bp);
261 				} else {
262 					bremfree(bp);
263 					splx(s);
264 					(void) bawrite(bp);
265 					s = splbio();
266 				}
267 			} else {
268 				bremfree(bp);
269 				splx(s);
270 				if ((error = bwrite(bp)) != 0)
271 					return (error);
272 				s = splbio();
273 			}
274 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
275 			/*
276 			 * If the buffer is for data that has been truncated
277 			 * off the file, then throw it away.
278 			 */
279 			bremfree(bp);
280 			bp->b_flags |= B_INVAL | B_NOCACHE;
281 			splx(s);
282 			brelse(bp);
283 			s = splbio();
284 		} else
285 			vfs_bio_awrite(bp);
286 
287 		/*
288 		 * Since we may have slept during the I/O, we need
289 		 * to start from a known point.
290 		 */
291 		BO_LOCK(bo);
292 		nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
293 	}
294 	/*
295 	 * If we were asked to do this synchronously, then go back for
296 	 * another pass, this time doing the metadata.
297 	 */
298 	if (skipmeta) {
299 		skipmeta = 0;
300 		goto loop;
301 	}
302 
303 	if (wait) {
304 		bufobj_wwait(bo, 3, 0);
305 		BO_UNLOCK(bo);
306 
307 		/*
308 		 * Ensure that any filesystem metatdata associated
309 		 * with the vnode has been written.
310 		 */
311 		splx(s);
312 		if ((error = softdep_sync_metadata(vp)) != 0)
313 			return (error);
314 		s = splbio();
315 
316 		BO_LOCK(bo);
317 		if (bo->bo_dirty.bv_cnt > 0) {
318 			/*
319 			 * Block devices associated with filesystems may
320 			 * have new I/O requests posted for them even if
321 			 * the vnode is locked, so no amount of trying will
322 			 * get them clean. Thus we give block devices a
323 			 * good effort, then just give up. For all other file
324 			 * types, go around and try again until it is clean.
325 			 */
326 			if (passes > 0) {
327 				passes -= 1;
328 				goto loop;
329 			}
330 #ifdef INVARIANTS
331 			if (!vn_isdisk(vp, NULL))
332 				vprint("ffs_fsync: dirty", vp);
333 #endif
334 		}
335 	}
336 	BO_UNLOCK(bo);
337 	splx(s);
338 	return (ffs_update(vp, wait));
339 }
340 
341 static int
342 ffs_lock(ap)
343 	struct vop_lock1_args /* {
344 		struct vnode *a_vp;
345 		int a_flags;
346 		struct thread *a_td;
347 		char *file;
348 		int line;
349 	} */ *ap;
350 {
351 #ifndef NO_FFS_SNAPSHOT
352 	struct vnode *vp;
353 	int flags;
354 	struct lock *lkp;
355 	int result;
356 
357 	switch (ap->a_flags & LK_TYPE_MASK) {
358 	case LK_SHARED:
359 	case LK_UPGRADE:
360 	case LK_EXCLUSIVE:
361 		vp = ap->a_vp;
362 		flags = ap->a_flags;
363 		for (;;) {
364 			/*
365 			 * vnode interlock must be held to ensure that
366 			 * the possibly external lock isn't freed,
367 			 * e.g. when mutating from snapshot file vnode
368 			 * to regular file vnode.
369 			 */
370 			if ((flags & LK_INTERLOCK) == 0) {
371 				VI_LOCK(vp);
372 				flags |= LK_INTERLOCK;
373 			}
374 			lkp = vp->v_vnlock;
375 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
376 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
377 			    ap->a_file, ap->a_line);
378 			if (lkp == vp->v_vnlock || result != 0)
379 				break;
380 			/*
381 			 * Apparent success, except that the vnode
382 			 * mutated between snapshot file vnode and
383 			 * regular file vnode while this process
384 			 * slept.  The lock currently held is not the
385 			 * right lock.  Release it, and try to get the
386 			 * new lock.
387 			 */
388 			(void) _lockmgr_args(lkp, LK_RELEASE, VI_MTX(vp),
389 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
390 			    ap->a_file, ap->a_line);
391 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
392 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
393 			flags &= ~LK_INTERLOCK;
394 		}
395 		break;
396 	default:
397 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
398 	}
399 	return (result);
400 #else
401 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
402 #endif
403 }
404 
405 /*
406  * Vnode op for reading.
407  */
408 /* ARGSUSED */
409 static int
410 ffs_read(ap)
411 	struct vop_read_args /* {
412 		struct vnode *a_vp;
413 		struct uio *a_uio;
414 		int a_ioflag;
415 		struct ucred *a_cred;
416 	} */ *ap;
417 {
418 	struct vnode *vp;
419 	struct inode *ip;
420 	struct uio *uio;
421 	struct fs *fs;
422 	struct buf *bp;
423 	ufs_lbn_t lbn, nextlbn;
424 	off_t bytesinfile;
425 	long size, xfersize, blkoffset;
426 	int error, orig_resid;
427 	int seqcount;
428 	int ioflag;
429 
430 	vp = ap->a_vp;
431 	uio = ap->a_uio;
432 	ioflag = ap->a_ioflag;
433 	if (ap->a_ioflag & IO_EXT)
434 #ifdef notyet
435 		return (ffs_extread(vp, uio, ioflag));
436 #else
437 		panic("ffs_read+IO_EXT");
438 #endif
439 #ifdef DIRECTIO
440 	if ((ioflag & IO_DIRECT) != 0) {
441 		int workdone;
442 
443 		error = ffs_rawread(vp, uio, &workdone);
444 		if (error != 0 || workdone != 0)
445 			return error;
446 	}
447 #endif
448 
449 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
450 	ip = VTOI(vp);
451 
452 #ifdef INVARIANTS
453 	if (uio->uio_rw != UIO_READ)
454 		panic("ffs_read: mode");
455 
456 	if (vp->v_type == VLNK) {
457 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
458 			panic("ffs_read: short symlink");
459 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
460 		panic("ffs_read: type %d",  vp->v_type);
461 #endif
462 	orig_resid = uio->uio_resid;
463 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
464 	if (orig_resid == 0)
465 		return (0);
466 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
467 	fs = ip->i_fs;
468 	if (uio->uio_offset < ip->i_size &&
469 	    uio->uio_offset >= fs->fs_maxfilesize)
470 		return (EOVERFLOW);
471 
472 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
473 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
474 			break;
475 		lbn = lblkno(fs, uio->uio_offset);
476 		nextlbn = lbn + 1;
477 
478 		/*
479 		 * size of buffer.  The buffer representing the
480 		 * end of the file is rounded up to the size of
481 		 * the block type ( fragment or full block,
482 		 * depending ).
483 		 */
484 		size = blksize(fs, ip, lbn);
485 		blkoffset = blkoff(fs, uio->uio_offset);
486 
487 		/*
488 		 * The amount we want to transfer in this iteration is
489 		 * one FS block less the amount of the data before
490 		 * our startpoint (duh!)
491 		 */
492 		xfersize = fs->fs_bsize - blkoffset;
493 
494 		/*
495 		 * But if we actually want less than the block,
496 		 * or the file doesn't have a whole block more of data,
497 		 * then use the lesser number.
498 		 */
499 		if (uio->uio_resid < xfersize)
500 			xfersize = uio->uio_resid;
501 		if (bytesinfile < xfersize)
502 			xfersize = bytesinfile;
503 
504 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
505 			/*
506 			 * Don't do readahead if this is the end of the file.
507 			 */
508 			error = bread(vp, lbn, size, NOCRED, &bp);
509 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
510 			/*
511 			 * Otherwise if we are allowed to cluster,
512 			 * grab as much as we can.
513 			 *
514 			 * XXX  This may not be a win if we are not
515 			 * doing sequential access.
516 			 */
517 			error = cluster_read(vp, ip->i_size, lbn,
518 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
519 		} else if (seqcount > 1) {
520 			/*
521 			 * If we are NOT allowed to cluster, then
522 			 * if we appear to be acting sequentially,
523 			 * fire off a request for a readahead
524 			 * as well as a read. Note that the 4th and 5th
525 			 * arguments point to arrays of the size specified in
526 			 * the 6th argument.
527 			 */
528 			int nextsize = blksize(fs, ip, nextlbn);
529 			error = breadn(vp, lbn,
530 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
531 		} else {
532 			/*
533 			 * Failing all of the above, just read what the
534 			 * user asked for. Interestingly, the same as
535 			 * the first option above.
536 			 */
537 			error = bread(vp, lbn, size, NOCRED, &bp);
538 		}
539 		if (error) {
540 			brelse(bp);
541 			bp = NULL;
542 			break;
543 		}
544 
545 		/*
546 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
547 		 * will cause us to attempt to release the buffer later on
548 		 * and will cause the buffer cache to attempt to free the
549 		 * underlying pages.
550 		 */
551 		if (ioflag & IO_DIRECT)
552 			bp->b_flags |= B_DIRECT;
553 
554 		/*
555 		 * We should only get non-zero b_resid when an I/O error
556 		 * has occurred, which should cause us to break above.
557 		 * However, if the short read did not cause an error,
558 		 * then we want to ensure that we do not uiomove bad
559 		 * or uninitialized data.
560 		 */
561 		size -= bp->b_resid;
562 		if (size < xfersize) {
563 			if (size == 0)
564 				break;
565 			xfersize = size;
566 		}
567 
568 		error = uiomove((char *)bp->b_data + blkoffset,
569 		    (int)xfersize, uio);
570 		if (error)
571 			break;
572 
573 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
574 		   (LIST_EMPTY(&bp->b_dep))) {
575 			/*
576 			 * If there are no dependencies, and it's VMIO,
577 			 * then we don't need the buf, mark it available
578 			 * for freeing. The VM has the data.
579 			 */
580 			bp->b_flags |= B_RELBUF;
581 			brelse(bp);
582 		} else {
583 			/*
584 			 * Otherwise let whoever
585 			 * made the request take care of
586 			 * freeing it. We just queue
587 			 * it onto another list.
588 			 */
589 			bqrelse(bp);
590 		}
591 	}
592 
593 	/*
594 	 * This can only happen in the case of an error
595 	 * because the loop above resets bp to NULL on each iteration
596 	 * and on normal completion has not set a new value into it.
597 	 * so it must have come from a 'break' statement
598 	 */
599 	if (bp != NULL) {
600 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
601 		   (LIST_EMPTY(&bp->b_dep))) {
602 			bp->b_flags |= B_RELBUF;
603 			brelse(bp);
604 		} else {
605 			bqrelse(bp);
606 		}
607 	}
608 
609 	if ((error == 0 || uio->uio_resid != orig_resid) &&
610 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
611 	    (ip->i_flag & IN_ACCESS) == 0) {
612 		VI_LOCK(vp);
613 		ip->i_flag |= IN_ACCESS;
614 		VI_UNLOCK(vp);
615 	}
616 	return (error);
617 }
618 
619 /*
620  * Vnode op for writing.
621  */
622 static int
623 ffs_write(ap)
624 	struct vop_write_args /* {
625 		struct vnode *a_vp;
626 		struct uio *a_uio;
627 		int a_ioflag;
628 		struct ucred *a_cred;
629 	} */ *ap;
630 {
631 	struct vnode *vp;
632 	struct uio *uio;
633 	struct inode *ip;
634 	struct fs *fs;
635 	struct buf *bp;
636 	struct thread *td;
637 	ufs_lbn_t lbn;
638 	off_t osize;
639 	int seqcount;
640 	int blkoffset, error, flags, ioflag, resid, size, xfersize;
641 
642 	vp = ap->a_vp;
643 	uio = ap->a_uio;
644 	ioflag = ap->a_ioflag;
645 	if (ap->a_ioflag & IO_EXT)
646 #ifdef notyet
647 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
648 #else
649 		panic("ffs_write+IO_EXT");
650 #endif
651 
652 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
653 	ip = VTOI(vp);
654 
655 #ifdef INVARIANTS
656 	if (uio->uio_rw != UIO_WRITE)
657 		panic("ffs_write: mode");
658 #endif
659 
660 	switch (vp->v_type) {
661 	case VREG:
662 		if (ioflag & IO_APPEND)
663 			uio->uio_offset = ip->i_size;
664 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
665 			return (EPERM);
666 		/* FALLTHROUGH */
667 	case VLNK:
668 		break;
669 	case VDIR:
670 		panic("ffs_write: dir write");
671 		break;
672 	default:
673 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
674 			(int)uio->uio_offset,
675 			(int)uio->uio_resid
676 		);
677 	}
678 
679 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
680 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
681 	fs = ip->i_fs;
682 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
683 		return (EFBIG);
684 	/*
685 	 * Maybe this should be above the vnode op call, but so long as
686 	 * file servers have no limits, I don't think it matters.
687 	 */
688 	td = uio->uio_td;
689 	if (vp->v_type == VREG && td != NULL) {
690 		PROC_LOCK(td->td_proc);
691 		if (uio->uio_offset + uio->uio_resid >
692 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
693 			psignal(td->td_proc, SIGXFSZ);
694 			PROC_UNLOCK(td->td_proc);
695 			return (EFBIG);
696 		}
697 		PROC_UNLOCK(td->td_proc);
698 	}
699 
700 	resid = uio->uio_resid;
701 	osize = ip->i_size;
702 	if (seqcount > BA_SEQMAX)
703 		flags = BA_SEQMAX << BA_SEQSHIFT;
704 	else
705 		flags = seqcount << BA_SEQSHIFT;
706 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
707 		flags |= IO_SYNC;
708 
709 	for (error = 0; uio->uio_resid > 0;) {
710 		lbn = lblkno(fs, uio->uio_offset);
711 		blkoffset = blkoff(fs, uio->uio_offset);
712 		xfersize = fs->fs_bsize - blkoffset;
713 		if (uio->uio_resid < xfersize)
714 			xfersize = uio->uio_resid;
715 		if (uio->uio_offset + xfersize > ip->i_size)
716 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
717 
718                 /*
719 		 * We must perform a read-before-write if the transfer size
720 		 * does not cover the entire buffer.
721                  */
722 		if (fs->fs_bsize > xfersize)
723 			flags |= BA_CLRBUF;
724 		else
725 			flags &= ~BA_CLRBUF;
726 /* XXX is uio->uio_offset the right thing here? */
727 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
728 		    ap->a_cred, flags, &bp);
729 		if (error != 0)
730 			break;
731 		/*
732 		 * If the buffer is not valid we have to clear out any
733 		 * garbage data from the pages instantiated for the buffer.
734 		 * If we do not, a failed uiomove() during a write can leave
735 		 * the prior contents of the pages exposed to a userland
736 		 * mmap().  XXX deal with uiomove() errors a better way.
737 		 */
738 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
739 			vfs_bio_clrbuf(bp);
740 		if (ioflag & IO_DIRECT)
741 			bp->b_flags |= B_DIRECT;
742 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
743 			bp->b_flags |= B_NOCACHE;
744 
745 		if (uio->uio_offset + xfersize > ip->i_size) {
746 			ip->i_size = uio->uio_offset + xfersize;
747 			DIP_SET(ip, i_size, ip->i_size);
748 		}
749 
750 		size = blksize(fs, ip, lbn) - bp->b_resid;
751 		if (size < xfersize)
752 			xfersize = size;
753 
754 		error =
755 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
756 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
757 		   (LIST_EMPTY(&bp->b_dep))) {
758 			bp->b_flags |= B_RELBUF;
759 		}
760 
761 		/*
762 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
763 		 * if we have a severe page deficiency write the buffer
764 		 * asynchronously.  Otherwise try to cluster, and if that
765 		 * doesn't do it then either do an async write (if O_DIRECT),
766 		 * or a delayed write (if not).
767 		 */
768 		if (ioflag & IO_SYNC) {
769 			(void)bwrite(bp);
770 		} else if (vm_page_count_severe() ||
771 			    buf_dirty_count_severe() ||
772 			    (ioflag & IO_ASYNC)) {
773 			bp->b_flags |= B_CLUSTEROK;
774 			bawrite(bp);
775 		} else if (xfersize + blkoffset == fs->fs_bsize) {
776 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
777 				bp->b_flags |= B_CLUSTEROK;
778 				cluster_write(vp, bp, ip->i_size, seqcount);
779 			} else {
780 				bawrite(bp);
781 			}
782 		} else if (ioflag & IO_DIRECT) {
783 			bp->b_flags |= B_CLUSTEROK;
784 			bawrite(bp);
785 		} else {
786 			bp->b_flags |= B_CLUSTEROK;
787 			bdwrite(bp);
788 		}
789 		if (error || xfersize == 0)
790 			break;
791 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
792 	}
793 	/*
794 	 * If we successfully wrote any data, and we are not the superuser
795 	 * we clear the setuid and setgid bits as a precaution against
796 	 * tampering.
797 	 */
798 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
799 	    ap->a_cred) {
800 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
801 			ip->i_mode &= ~(ISUID | ISGID);
802 			DIP_SET(ip, i_mode, ip->i_mode);
803 		}
804 	}
805 	if (error) {
806 		if (ioflag & IO_UNIT) {
807 			(void)ffs_truncate(vp, osize,
808 			    IO_NORMAL | (ioflag & IO_SYNC),
809 			    ap->a_cred, uio->uio_td);
810 			uio->uio_offset -= resid - uio->uio_resid;
811 			uio->uio_resid = resid;
812 		}
813 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
814 		error = ffs_update(vp, 1);
815 	return (error);
816 }
817 
818 /*
819  * get page routine
820  */
821 static int
822 ffs_getpages(ap)
823 	struct vop_getpages_args *ap;
824 {
825 	int i;
826 	vm_page_t mreq;
827 	int pcount;
828 
829 	pcount = round_page(ap->a_count) / PAGE_SIZE;
830 	mreq = ap->a_m[ap->a_reqpage];
831 
832 	/*
833 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
834 	 * then the entire page is valid.  Since the page may be mapped,
835 	 * user programs might reference data beyond the actual end of file
836 	 * occuring within the page.  We have to zero that data.
837 	 */
838 	VM_OBJECT_LOCK(mreq->object);
839 	if (mreq->valid) {
840 		if (mreq->valid != VM_PAGE_BITS_ALL)
841 			vm_page_zero_invalid(mreq, TRUE);
842 		vm_page_lock_queues();
843 		for (i = 0; i < pcount; i++) {
844 			if (i != ap->a_reqpage) {
845 				vm_page_free(ap->a_m[i]);
846 			}
847 		}
848 		vm_page_unlock_queues();
849 		VM_OBJECT_UNLOCK(mreq->object);
850 		return VM_PAGER_OK;
851 	}
852 	VM_OBJECT_UNLOCK(mreq->object);
853 
854 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
855 					    ap->a_count,
856 					    ap->a_reqpage);
857 }
858 
859 
860 /*
861  * Extended attribute area reading.
862  */
863 static int
864 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
865 {
866 	struct inode *ip;
867 	struct ufs2_dinode *dp;
868 	struct fs *fs;
869 	struct buf *bp;
870 	ufs_lbn_t lbn, nextlbn;
871 	off_t bytesinfile;
872 	long size, xfersize, blkoffset;
873 	int error, orig_resid;
874 
875 	ip = VTOI(vp);
876 	fs = ip->i_fs;
877 	dp = ip->i_din2;
878 
879 #ifdef INVARIANTS
880 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
881 		panic("ffs_extread: mode");
882 
883 #endif
884 	orig_resid = uio->uio_resid;
885 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
886 	if (orig_resid == 0)
887 		return (0);
888 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
889 
890 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
891 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
892 			break;
893 		lbn = lblkno(fs, uio->uio_offset);
894 		nextlbn = lbn + 1;
895 
896 		/*
897 		 * size of buffer.  The buffer representing the
898 		 * end of the file is rounded up to the size of
899 		 * the block type ( fragment or full block,
900 		 * depending ).
901 		 */
902 		size = sblksize(fs, dp->di_extsize, lbn);
903 		blkoffset = blkoff(fs, uio->uio_offset);
904 
905 		/*
906 		 * The amount we want to transfer in this iteration is
907 		 * one FS block less the amount of the data before
908 		 * our startpoint (duh!)
909 		 */
910 		xfersize = fs->fs_bsize - blkoffset;
911 
912 		/*
913 		 * But if we actually want less than the block,
914 		 * or the file doesn't have a whole block more of data,
915 		 * then use the lesser number.
916 		 */
917 		if (uio->uio_resid < xfersize)
918 			xfersize = uio->uio_resid;
919 		if (bytesinfile < xfersize)
920 			xfersize = bytesinfile;
921 
922 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
923 			/*
924 			 * Don't do readahead if this is the end of the info.
925 			 */
926 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
927 		} else {
928 			/*
929 			 * If we have a second block, then
930 			 * fire off a request for a readahead
931 			 * as well as a read. Note that the 4th and 5th
932 			 * arguments point to arrays of the size specified in
933 			 * the 6th argument.
934 			 */
935 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
936 
937 			nextlbn = -1 - nextlbn;
938 			error = breadn(vp, -1 - lbn,
939 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
940 		}
941 		if (error) {
942 			brelse(bp);
943 			bp = NULL;
944 			break;
945 		}
946 
947 		/*
948 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
949 		 * will cause us to attempt to release the buffer later on
950 		 * and will cause the buffer cache to attempt to free the
951 		 * underlying pages.
952 		 */
953 		if (ioflag & IO_DIRECT)
954 			bp->b_flags |= B_DIRECT;
955 
956 		/*
957 		 * We should only get non-zero b_resid when an I/O error
958 		 * has occurred, which should cause us to break above.
959 		 * However, if the short read did not cause an error,
960 		 * then we want to ensure that we do not uiomove bad
961 		 * or uninitialized data.
962 		 */
963 		size -= bp->b_resid;
964 		if (size < xfersize) {
965 			if (size == 0)
966 				break;
967 			xfersize = size;
968 		}
969 
970 		error = uiomove((char *)bp->b_data + blkoffset,
971 					(int)xfersize, uio);
972 		if (error)
973 			break;
974 
975 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
976 		   (LIST_EMPTY(&bp->b_dep))) {
977 			/*
978 			 * If there are no dependencies, and it's VMIO,
979 			 * then we don't need the buf, mark it available
980 			 * for freeing. The VM has the data.
981 			 */
982 			bp->b_flags |= B_RELBUF;
983 			brelse(bp);
984 		} else {
985 			/*
986 			 * Otherwise let whoever
987 			 * made the request take care of
988 			 * freeing it. We just queue
989 			 * it onto another list.
990 			 */
991 			bqrelse(bp);
992 		}
993 	}
994 
995 	/*
996 	 * This can only happen in the case of an error
997 	 * because the loop above resets bp to NULL on each iteration
998 	 * and on normal completion has not set a new value into it.
999 	 * so it must have come from a 'break' statement
1000 	 */
1001 	if (bp != NULL) {
1002 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1003 		   (LIST_EMPTY(&bp->b_dep))) {
1004 			bp->b_flags |= B_RELBUF;
1005 			brelse(bp);
1006 		} else {
1007 			bqrelse(bp);
1008 		}
1009 	}
1010 
1011 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1012 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
1013 	    (ip->i_flag & IN_ACCESS) == 0) {
1014 		VI_LOCK(vp);
1015 		ip->i_flag |= IN_ACCESS;
1016 		VI_UNLOCK(vp);
1017 	}
1018 	return (error);
1019 }
1020 
1021 /*
1022  * Extended attribute area writing.
1023  */
1024 static int
1025 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1026 {
1027 	struct inode *ip;
1028 	struct ufs2_dinode *dp;
1029 	struct fs *fs;
1030 	struct buf *bp;
1031 	ufs_lbn_t lbn;
1032 	off_t osize;
1033 	int blkoffset, error, flags, resid, size, xfersize;
1034 
1035 	ip = VTOI(vp);
1036 	fs = ip->i_fs;
1037 	dp = ip->i_din2;
1038 
1039 	KASSERT(!(ip->i_flag & IN_SPACECOUNTED), ("inode %u: inode is dead",
1040 	    ip->i_number));
1041 
1042 #ifdef INVARIANTS
1043 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1044 		panic("ffs_extwrite: mode");
1045 #endif
1046 
1047 	if (ioflag & IO_APPEND)
1048 		uio->uio_offset = dp->di_extsize;
1049 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1050 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1051 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1052 		return (EFBIG);
1053 
1054 	resid = uio->uio_resid;
1055 	osize = dp->di_extsize;
1056 	flags = IO_EXT;
1057 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1058 		flags |= IO_SYNC;
1059 
1060 	for (error = 0; uio->uio_resid > 0;) {
1061 		lbn = lblkno(fs, uio->uio_offset);
1062 		blkoffset = blkoff(fs, uio->uio_offset);
1063 		xfersize = fs->fs_bsize - blkoffset;
1064 		if (uio->uio_resid < xfersize)
1065 			xfersize = uio->uio_resid;
1066 
1067 		/*
1068 		 * We must perform a read-before-write if the transfer size
1069 		 * does not cover the entire buffer.
1070                  */
1071 		if (fs->fs_bsize > xfersize)
1072 			flags |= BA_CLRBUF;
1073 		else
1074 			flags &= ~BA_CLRBUF;
1075 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1076 		    ucred, flags, &bp);
1077 		if (error != 0)
1078 			break;
1079 		/*
1080 		 * If the buffer is not valid we have to clear out any
1081 		 * garbage data from the pages instantiated for the buffer.
1082 		 * If we do not, a failed uiomove() during a write can leave
1083 		 * the prior contents of the pages exposed to a userland
1084 		 * mmap().  XXX deal with uiomove() errors a better way.
1085 		 */
1086 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1087 			vfs_bio_clrbuf(bp);
1088 		if (ioflag & IO_DIRECT)
1089 			bp->b_flags |= B_DIRECT;
1090 
1091 		if (uio->uio_offset + xfersize > dp->di_extsize)
1092 			dp->di_extsize = uio->uio_offset + xfersize;
1093 
1094 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1095 		if (size < xfersize)
1096 			xfersize = size;
1097 
1098 		error =
1099 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1100 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1101 		   (LIST_EMPTY(&bp->b_dep))) {
1102 			bp->b_flags |= B_RELBUF;
1103 		}
1104 
1105 		/*
1106 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1107 		 * if we have a severe page deficiency write the buffer
1108 		 * asynchronously.  Otherwise try to cluster, and if that
1109 		 * doesn't do it then either do an async write (if O_DIRECT),
1110 		 * or a delayed write (if not).
1111 		 */
1112 		if (ioflag & IO_SYNC) {
1113 			(void)bwrite(bp);
1114 		} else if (vm_page_count_severe() ||
1115 			    buf_dirty_count_severe() ||
1116 			    xfersize + blkoffset == fs->fs_bsize ||
1117 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1118 			bawrite(bp);
1119 		else
1120 			bdwrite(bp);
1121 		if (error || xfersize == 0)
1122 			break;
1123 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1124 	}
1125 	/*
1126 	 * If we successfully wrote any data, and we are not the superuser
1127 	 * we clear the setuid and setgid bits as a precaution against
1128 	 * tampering.
1129 	 */
1130 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1131 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1132 			ip->i_mode &= ~(ISUID | ISGID);
1133 			dp->di_mode = ip->i_mode;
1134 		}
1135 	}
1136 	if (error) {
1137 		if (ioflag & IO_UNIT) {
1138 			(void)ffs_truncate(vp, osize,
1139 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1140 			uio->uio_offset -= resid - uio->uio_resid;
1141 			uio->uio_resid = resid;
1142 		}
1143 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1144 		error = ffs_update(vp, 1);
1145 	return (error);
1146 }
1147 
1148 
1149 /*
1150  * Vnode operating to retrieve a named extended attribute.
1151  *
1152  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1153  * the length of the EA, and possibly the pointer to the entry and to the data.
1154  */
1155 static int
1156 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1157 {
1158 	u_char *p, *pe, *pn, *p0;
1159 	int eapad1, eapad2, ealength, ealen, nlen;
1160 	uint32_t ul;
1161 
1162 	pe = ptr + length;
1163 	nlen = strlen(name);
1164 
1165 	for (p = ptr; p < pe; p = pn) {
1166 		p0 = p;
1167 		bcopy(p, &ul, sizeof(ul));
1168 		pn = p + ul;
1169 		/* make sure this entry is complete */
1170 		if (pn > pe)
1171 			break;
1172 		p += sizeof(uint32_t);
1173 		if (*p != nspace)
1174 			continue;
1175 		p++;
1176 		eapad2 = *p++;
1177 		if (*p != nlen)
1178 			continue;
1179 		p++;
1180 		if (bcmp(p, name, nlen))
1181 			continue;
1182 		ealength = sizeof(uint32_t) + 3 + nlen;
1183 		eapad1 = 8 - (ealength % 8);
1184 		if (eapad1 == 8)
1185 			eapad1 = 0;
1186 		ealength += eapad1;
1187 		ealen = ul - ealength - eapad2;
1188 		p += nlen + eapad1;
1189 		if (eap != NULL)
1190 			*eap = p0;
1191 		if (eac != NULL)
1192 			*eac = p;
1193 		return (ealen);
1194 	}
1195 	return(-1);
1196 }
1197 
1198 static int
1199 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1200 {
1201 	struct inode *ip;
1202 	struct ufs2_dinode *dp;
1203 	struct fs *fs;
1204 	struct uio luio;
1205 	struct iovec liovec;
1206 	int easize, error;
1207 	u_char *eae;
1208 
1209 	ip = VTOI(vp);
1210 	fs = ip->i_fs;
1211 	dp = ip->i_din2;
1212 	easize = dp->di_extsize;
1213 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1214 		return (EFBIG);
1215 
1216 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1217 
1218 	liovec.iov_base = eae;
1219 	liovec.iov_len = easize;
1220 	luio.uio_iov = &liovec;
1221 	luio.uio_iovcnt = 1;
1222 	luio.uio_offset = 0;
1223 	luio.uio_resid = easize;
1224 	luio.uio_segflg = UIO_SYSSPACE;
1225 	luio.uio_rw = UIO_READ;
1226 	luio.uio_td = td;
1227 
1228 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1229 	if (error) {
1230 		free(eae, M_TEMP);
1231 		return(error);
1232 	}
1233 	*p = eae;
1234 	return (0);
1235 }
1236 
1237 static int
1238 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1239 {
1240 	struct inode *ip;
1241 	struct ufs2_dinode *dp;
1242 	int error;
1243 
1244 	ip = VTOI(vp);
1245 
1246 	if (ip->i_ea_area != NULL)
1247 		return (EBUSY);
1248 	dp = ip->i_din2;
1249 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1250 	if (error)
1251 		return (error);
1252 	ip->i_ea_len = dp->di_extsize;
1253 	ip->i_ea_error = 0;
1254 	return (0);
1255 }
1256 
1257 /*
1258  * Vnode extattr transaction commit/abort
1259  */
1260 static int
1261 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1262 {
1263 	struct inode *ip;
1264 	struct uio luio;
1265 	struct iovec liovec;
1266 	int error;
1267 	struct ufs2_dinode *dp;
1268 
1269 	ip = VTOI(vp);
1270 	if (ip->i_ea_area == NULL)
1271 		return (EINVAL);
1272 	dp = ip->i_din2;
1273 	error = ip->i_ea_error;
1274 	if (commit && error == 0) {
1275 		if (cred == NOCRED)
1276 			cred =  vp->v_mount->mnt_cred;
1277 		liovec.iov_base = ip->i_ea_area;
1278 		liovec.iov_len = ip->i_ea_len;
1279 		luio.uio_iov = &liovec;
1280 		luio.uio_iovcnt = 1;
1281 		luio.uio_offset = 0;
1282 		luio.uio_resid = ip->i_ea_len;
1283 		luio.uio_segflg = UIO_SYSSPACE;
1284 		luio.uio_rw = UIO_WRITE;
1285 		luio.uio_td = td;
1286 		/* XXX: I'm not happy about truncating to zero size */
1287 		if (ip->i_ea_len < dp->di_extsize)
1288 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1289 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1290 	}
1291 	free(ip->i_ea_area, M_TEMP);
1292 	ip->i_ea_area = NULL;
1293 	ip->i_ea_len = 0;
1294 	ip->i_ea_error = 0;
1295 	return (error);
1296 }
1297 
1298 /*
1299  * Vnode extattr strategy routine for fifos.
1300  *
1301  * We need to check for a read or write of the external attributes.
1302  * Otherwise we just fall through and do the usual thing.
1303  */
1304 static int
1305 ffsext_strategy(struct vop_strategy_args *ap)
1306 /*
1307 struct vop_strategy_args {
1308 	struct vnodeop_desc *a_desc;
1309 	struct vnode *a_vp;
1310 	struct buf *a_bp;
1311 };
1312 */
1313 {
1314 	struct vnode *vp;
1315 	daddr_t lbn;
1316 
1317 	vp = ap->a_vp;
1318 	lbn = ap->a_bp->b_lblkno;
1319 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1320 	    lbn < 0 && lbn >= -NXADDR)
1321 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1322 	if (vp->v_type == VFIFO)
1323 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1324 	panic("spec nodes went here");
1325 }
1326 
1327 /*
1328  * Vnode extattr transaction commit/abort
1329  */
1330 static int
1331 ffs_openextattr(struct vop_openextattr_args *ap)
1332 /*
1333 struct vop_openextattr_args {
1334 	struct vnodeop_desc *a_desc;
1335 	struct vnode *a_vp;
1336 	IN struct ucred *a_cred;
1337 	IN struct thread *a_td;
1338 };
1339 */
1340 {
1341 	struct inode *ip;
1342 	struct fs *fs;
1343 
1344 	ip = VTOI(ap->a_vp);
1345 	fs = ip->i_fs;
1346 
1347 	if (ap->a_vp->v_type == VCHR)
1348 		return (EOPNOTSUPP);
1349 
1350 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1351 }
1352 
1353 
1354 /*
1355  * Vnode extattr transaction commit/abort
1356  */
1357 static int
1358 ffs_closeextattr(struct vop_closeextattr_args *ap)
1359 /*
1360 struct vop_closeextattr_args {
1361 	struct vnodeop_desc *a_desc;
1362 	struct vnode *a_vp;
1363 	int a_commit;
1364 	IN struct ucred *a_cred;
1365 	IN struct thread *a_td;
1366 };
1367 */
1368 {
1369 	struct inode *ip;
1370 	struct fs *fs;
1371 
1372 	ip = VTOI(ap->a_vp);
1373 	fs = ip->i_fs;
1374 
1375 	if (ap->a_vp->v_type == VCHR)
1376 		return (EOPNOTSUPP);
1377 
1378 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1379 		return (EROFS);
1380 
1381 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1382 }
1383 
1384 /*
1385  * Vnode operation to remove a named attribute.
1386  */
1387 static int
1388 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1389 /*
1390 vop_deleteextattr {
1391 	IN struct vnode *a_vp;
1392 	IN int a_attrnamespace;
1393 	IN const char *a_name;
1394 	IN struct ucred *a_cred;
1395 	IN struct thread *a_td;
1396 };
1397 */
1398 {
1399 	struct inode *ip;
1400 	struct fs *fs;
1401 	uint32_t ealength, ul;
1402 	int ealen, olen, eapad1, eapad2, error, i, easize;
1403 	u_char *eae, *p;
1404 	int stand_alone;
1405 
1406 	ip = VTOI(ap->a_vp);
1407 	fs = ip->i_fs;
1408 
1409 	if (ap->a_vp->v_type == VCHR)
1410 		return (EOPNOTSUPP);
1411 
1412 	if (strlen(ap->a_name) == 0)
1413 		return (EINVAL);
1414 
1415 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1416 		return (EROFS);
1417 
1418 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1419 	    ap->a_cred, ap->a_td, IWRITE);
1420 	if (error) {
1421 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1422 			ip->i_ea_error = error;
1423 		return (error);
1424 	}
1425 
1426 	if (ip->i_ea_area == NULL) {
1427 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1428 		if (error)
1429 			return (error);
1430 		stand_alone = 1;
1431 	} else {
1432 		stand_alone = 0;
1433 	}
1434 
1435 	ealength = eapad1 = ealen = eapad2 = 0;
1436 
1437 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1438 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1439 	easize = ip->i_ea_len;
1440 
1441 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1442 	    &p, NULL);
1443 	if (olen == -1) {
1444 		/* delete but nonexistent */
1445 		free(eae, M_TEMP);
1446 		if (stand_alone)
1447 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1448 		return(ENOATTR);
1449 	}
1450 	bcopy(p, &ul, sizeof ul);
1451 	i = p - eae + ul;
1452 	if (ul != ealength) {
1453 		bcopy(p + ul, p + ealength, easize - i);
1454 		easize += (ealength - ul);
1455 	}
1456 	if (easize > NXADDR * fs->fs_bsize) {
1457 		free(eae, M_TEMP);
1458 		if (stand_alone)
1459 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1460 		else if (ip->i_ea_error == 0)
1461 			ip->i_ea_error = ENOSPC;
1462 		return(ENOSPC);
1463 	}
1464 	p = ip->i_ea_area;
1465 	ip->i_ea_area = eae;
1466 	ip->i_ea_len = easize;
1467 	free(p, M_TEMP);
1468 	if (stand_alone)
1469 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1470 	return(error);
1471 }
1472 
1473 /*
1474  * Vnode operation to retrieve a named extended attribute.
1475  */
1476 static int
1477 ffs_getextattr(struct vop_getextattr_args *ap)
1478 /*
1479 vop_getextattr {
1480 	IN struct vnode *a_vp;
1481 	IN int a_attrnamespace;
1482 	IN const char *a_name;
1483 	INOUT struct uio *a_uio;
1484 	OUT size_t *a_size;
1485 	IN struct ucred *a_cred;
1486 	IN struct thread *a_td;
1487 };
1488 */
1489 {
1490 	struct inode *ip;
1491 	struct fs *fs;
1492 	u_char *eae, *p;
1493 	unsigned easize;
1494 	int error, ealen, stand_alone;
1495 
1496 	ip = VTOI(ap->a_vp);
1497 	fs = ip->i_fs;
1498 
1499 	if (ap->a_vp->v_type == VCHR)
1500 		return (EOPNOTSUPP);
1501 
1502 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1503 	    ap->a_cred, ap->a_td, IREAD);
1504 	if (error)
1505 		return (error);
1506 
1507 	if (ip->i_ea_area == NULL) {
1508 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1509 		if (error)
1510 			return (error);
1511 		stand_alone = 1;
1512 	} else {
1513 		stand_alone = 0;
1514 	}
1515 	eae = ip->i_ea_area;
1516 	easize = ip->i_ea_len;
1517 
1518 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1519 	    NULL, &p);
1520 	if (ealen >= 0) {
1521 		error = 0;
1522 		if (ap->a_size != NULL)
1523 			*ap->a_size = ealen;
1524 		else if (ap->a_uio != NULL)
1525 			error = uiomove(p, ealen, ap->a_uio);
1526 	} else
1527 		error = ENOATTR;
1528 	if (stand_alone)
1529 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1530 	return(error);
1531 }
1532 
1533 /*
1534  * Vnode operation to retrieve extended attributes on a vnode.
1535  */
1536 static int
1537 ffs_listextattr(struct vop_listextattr_args *ap)
1538 /*
1539 vop_listextattr {
1540 	IN struct vnode *a_vp;
1541 	IN int a_attrnamespace;
1542 	INOUT struct uio *a_uio;
1543 	OUT size_t *a_size;
1544 	IN struct ucred *a_cred;
1545 	IN struct thread *a_td;
1546 };
1547 */
1548 {
1549 	struct inode *ip;
1550 	struct fs *fs;
1551 	u_char *eae, *p, *pe, *pn;
1552 	unsigned easize;
1553 	uint32_t ul;
1554 	int error, ealen, stand_alone;
1555 
1556 	ip = VTOI(ap->a_vp);
1557 	fs = ip->i_fs;
1558 
1559 	if (ap->a_vp->v_type == VCHR)
1560 		return (EOPNOTSUPP);
1561 
1562 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1563 	    ap->a_cred, ap->a_td, IREAD);
1564 	if (error)
1565 		return (error);
1566 
1567 	if (ip->i_ea_area == NULL) {
1568 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1569 		if (error)
1570 			return (error);
1571 		stand_alone = 1;
1572 	} else {
1573 		stand_alone = 0;
1574 	}
1575 	eae = ip->i_ea_area;
1576 	easize = ip->i_ea_len;
1577 
1578 	error = 0;
1579 	if (ap->a_size != NULL)
1580 		*ap->a_size = 0;
1581 	pe = eae + easize;
1582 	for(p = eae; error == 0 && p < pe; p = pn) {
1583 		bcopy(p, &ul, sizeof(ul));
1584 		pn = p + ul;
1585 		if (pn > pe)
1586 			break;
1587 		p += sizeof(ul);
1588 		if (*p++ != ap->a_attrnamespace)
1589 			continue;
1590 		p++;	/* pad2 */
1591 		ealen = *p;
1592 		if (ap->a_size != NULL) {
1593 			*ap->a_size += ealen + 1;
1594 		} else if (ap->a_uio != NULL) {
1595 			error = uiomove(p, ealen + 1, ap->a_uio);
1596 		}
1597 	}
1598 	if (stand_alone)
1599 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1600 	return(error);
1601 }
1602 
1603 /*
1604  * Vnode operation to set a named attribute.
1605  */
1606 static int
1607 ffs_setextattr(struct vop_setextattr_args *ap)
1608 /*
1609 vop_setextattr {
1610 	IN struct vnode *a_vp;
1611 	IN int a_attrnamespace;
1612 	IN const char *a_name;
1613 	INOUT struct uio *a_uio;
1614 	IN struct ucred *a_cred;
1615 	IN struct thread *a_td;
1616 };
1617 */
1618 {
1619 	struct inode *ip;
1620 	struct fs *fs;
1621 	uint32_t ealength, ul;
1622 	int ealen, olen, eapad1, eapad2, error, i, easize;
1623 	u_char *eae, *p;
1624 	int stand_alone;
1625 
1626 	ip = VTOI(ap->a_vp);
1627 	fs = ip->i_fs;
1628 
1629 	if (ap->a_vp->v_type == VCHR)
1630 		return (EOPNOTSUPP);
1631 
1632 	if (strlen(ap->a_name) == 0)
1633 		return (EINVAL);
1634 
1635 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1636 	if (ap->a_uio == NULL)
1637 		return (EOPNOTSUPP);
1638 
1639 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1640 		return (EROFS);
1641 
1642 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1643 	    ap->a_cred, ap->a_td, IWRITE);
1644 	if (error) {
1645 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1646 			ip->i_ea_error = error;
1647 		return (error);
1648 	}
1649 
1650 	if (ip->i_ea_area == NULL) {
1651 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1652 		if (error)
1653 			return (error);
1654 		stand_alone = 1;
1655 	} else {
1656 		stand_alone = 0;
1657 	}
1658 
1659 	ealen = ap->a_uio->uio_resid;
1660 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1661 	eapad1 = 8 - (ealength % 8);
1662 	if (eapad1 == 8)
1663 		eapad1 = 0;
1664 	eapad2 = 8 - (ealen % 8);
1665 	if (eapad2 == 8)
1666 		eapad2 = 0;
1667 	ealength += eapad1 + ealen + eapad2;
1668 
1669 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1670 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1671 	easize = ip->i_ea_len;
1672 
1673 	olen = ffs_findextattr(eae, easize,
1674 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1675         if (olen == -1) {
1676 		/* new, append at end */
1677 		p = eae + easize;
1678 		easize += ealength;
1679 	} else {
1680 		bcopy(p, &ul, sizeof ul);
1681 		i = p - eae + ul;
1682 		if (ul != ealength) {
1683 			bcopy(p + ul, p + ealength, easize - i);
1684 			easize += (ealength - ul);
1685 		}
1686 	}
1687 	if (easize > NXADDR * fs->fs_bsize) {
1688 		free(eae, M_TEMP);
1689 		if (stand_alone)
1690 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1691 		else if (ip->i_ea_error == 0)
1692 			ip->i_ea_error = ENOSPC;
1693 		return(ENOSPC);
1694 	}
1695 	bcopy(&ealength, p, sizeof(ealength));
1696 	p += sizeof(ealength);
1697 	*p++ = ap->a_attrnamespace;
1698 	*p++ = eapad2;
1699 	*p++ = strlen(ap->a_name);
1700 	strcpy(p, ap->a_name);
1701 	p += strlen(ap->a_name);
1702 	bzero(p, eapad1);
1703 	p += eapad1;
1704 	error = uiomove(p, ealen, ap->a_uio);
1705 	if (error) {
1706 		free(eae, M_TEMP);
1707 		if (stand_alone)
1708 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1709 		else if (ip->i_ea_error == 0)
1710 			ip->i_ea_error = error;
1711 		return(error);
1712 	}
1713 	p += ealen;
1714 	bzero(p, eapad2);
1715 
1716 	p = ip->i_ea_area;
1717 	ip->i_ea_area = eae;
1718 	ip->i_ea_len = easize;
1719 	free(p, M_TEMP);
1720 	if (stand_alone)
1721 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1722 	return(error);
1723 }
1724 
1725 /*
1726  * Vnode pointer to File handle
1727  */
1728 static int
1729 ffs_vptofh(struct vop_vptofh_args *ap)
1730 /*
1731 vop_vptofh {
1732 	IN struct vnode *a_vp;
1733 	IN struct fid *a_fhp;
1734 };
1735 */
1736 {
1737 	struct inode *ip;
1738 	struct ufid *ufhp;
1739 
1740 	ip = VTOI(ap->a_vp);
1741 	ufhp = (struct ufid *)ap->a_fhp;
1742 	ufhp->ufid_len = sizeof(struct ufid);
1743 	ufhp->ufid_ino = ip->i_number;
1744 	ufhp->ufid_gen = ip->i_gen;
1745 	return (0);
1746 }
1747