xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 13014ca04aad1931d41958b56f71a2c65b9a7a2c)
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 			lkp = vp->v_vnlock;
365 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
366 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
367 			    ap->a_file, ap->a_line);
368 			if (lkp == vp->v_vnlock || result != 0)
369 				break;
370 			/*
371 			 * Apparent success, except that the vnode
372 			 * mutated between snapshot file vnode and
373 			 * regular file vnode while this process
374 			 * slept.  The lock currently held is not the
375 			 * right lock.  Release it, and try to get the
376 			 * new lock.
377 			 */
378 			(void) _lockmgr_args(lkp, LK_RELEASE, NULL,
379 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
380 			    ap->a_file, ap->a_line);
381 			if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
382 			    (LK_INTERLOCK | LK_NOWAIT))
383 				return (EBUSY);
384 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
385 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
386 			flags &= ~LK_INTERLOCK;
387 		}
388 		break;
389 	default:
390 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
391 	}
392 	return (result);
393 #else
394 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
395 #endif
396 }
397 
398 /*
399  * Vnode op for reading.
400  */
401 /* ARGSUSED */
402 static int
403 ffs_read(ap)
404 	struct vop_read_args /* {
405 		struct vnode *a_vp;
406 		struct uio *a_uio;
407 		int a_ioflag;
408 		struct ucred *a_cred;
409 	} */ *ap;
410 {
411 	struct vnode *vp;
412 	struct inode *ip;
413 	struct uio *uio;
414 	struct fs *fs;
415 	struct buf *bp;
416 	ufs_lbn_t lbn, nextlbn;
417 	off_t bytesinfile;
418 	long size, xfersize, blkoffset;
419 	int error, orig_resid;
420 	int seqcount;
421 	int ioflag;
422 
423 	vp = ap->a_vp;
424 	uio = ap->a_uio;
425 	ioflag = ap->a_ioflag;
426 	if (ap->a_ioflag & IO_EXT)
427 #ifdef notyet
428 		return (ffs_extread(vp, uio, ioflag));
429 #else
430 		panic("ffs_read+IO_EXT");
431 #endif
432 #ifdef DIRECTIO
433 	if ((ioflag & IO_DIRECT) != 0) {
434 		int workdone;
435 
436 		error = ffs_rawread(vp, uio, &workdone);
437 		if (error != 0 || workdone != 0)
438 			return error;
439 	}
440 #endif
441 
442 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
443 	ip = VTOI(vp);
444 
445 #ifdef INVARIANTS
446 	if (uio->uio_rw != UIO_READ)
447 		panic("ffs_read: mode");
448 
449 	if (vp->v_type == VLNK) {
450 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
451 			panic("ffs_read: short symlink");
452 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
453 		panic("ffs_read: type %d",  vp->v_type);
454 #endif
455 	orig_resid = uio->uio_resid;
456 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
457 	if (orig_resid == 0)
458 		return (0);
459 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
460 	fs = ip->i_fs;
461 	if (uio->uio_offset < ip->i_size &&
462 	    uio->uio_offset >= fs->fs_maxfilesize)
463 		return (EOVERFLOW);
464 
465 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
466 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
467 			break;
468 		lbn = lblkno(fs, uio->uio_offset);
469 		nextlbn = lbn + 1;
470 
471 		/*
472 		 * size of buffer.  The buffer representing the
473 		 * end of the file is rounded up to the size of
474 		 * the block type ( fragment or full block,
475 		 * depending ).
476 		 */
477 		size = blksize(fs, ip, lbn);
478 		blkoffset = blkoff(fs, uio->uio_offset);
479 
480 		/*
481 		 * The amount we want to transfer in this iteration is
482 		 * one FS block less the amount of the data before
483 		 * our startpoint (duh!)
484 		 */
485 		xfersize = fs->fs_bsize - blkoffset;
486 
487 		/*
488 		 * But if we actually want less than the block,
489 		 * or the file doesn't have a whole block more of data,
490 		 * then use the lesser number.
491 		 */
492 		if (uio->uio_resid < xfersize)
493 			xfersize = uio->uio_resid;
494 		if (bytesinfile < xfersize)
495 			xfersize = bytesinfile;
496 
497 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
498 			/*
499 			 * Don't do readahead if this is the end of the file.
500 			 */
501 			error = bread(vp, lbn, size, NOCRED, &bp);
502 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
503 			/*
504 			 * Otherwise if we are allowed to cluster,
505 			 * grab as much as we can.
506 			 *
507 			 * XXX  This may not be a win if we are not
508 			 * doing sequential access.
509 			 */
510 			error = cluster_read(vp, ip->i_size, lbn,
511 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
512 		} else if (seqcount > 1) {
513 			/*
514 			 * If we are NOT allowed to cluster, then
515 			 * if we appear to be acting sequentially,
516 			 * fire off a request for a readahead
517 			 * as well as a read. Note that the 4th and 5th
518 			 * arguments point to arrays of the size specified in
519 			 * the 6th argument.
520 			 */
521 			int nextsize = blksize(fs, ip, nextlbn);
522 			error = breadn(vp, lbn,
523 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
524 		} else {
525 			/*
526 			 * Failing all of the above, just read what the
527 			 * user asked for. Interestingly, the same as
528 			 * the first option above.
529 			 */
530 			error = bread(vp, lbn, size, NOCRED, &bp);
531 		}
532 		if (error) {
533 			brelse(bp);
534 			bp = NULL;
535 			break;
536 		}
537 
538 		/*
539 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
540 		 * will cause us to attempt to release the buffer later on
541 		 * and will cause the buffer cache to attempt to free the
542 		 * underlying pages.
543 		 */
544 		if (ioflag & IO_DIRECT)
545 			bp->b_flags |= B_DIRECT;
546 
547 		/*
548 		 * We should only get non-zero b_resid when an I/O error
549 		 * has occurred, which should cause us to break above.
550 		 * However, if the short read did not cause an error,
551 		 * then we want to ensure that we do not uiomove bad
552 		 * or uninitialized data.
553 		 */
554 		size -= bp->b_resid;
555 		if (size < xfersize) {
556 			if (size == 0)
557 				break;
558 			xfersize = size;
559 		}
560 
561 		error = uiomove((char *)bp->b_data + blkoffset,
562 		    (int)xfersize, uio);
563 		if (error)
564 			break;
565 
566 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
567 		   (LIST_EMPTY(&bp->b_dep))) {
568 			/*
569 			 * If there are no dependencies, and it's VMIO,
570 			 * then we don't need the buf, mark it available
571 			 * for freeing. The VM has the data.
572 			 */
573 			bp->b_flags |= B_RELBUF;
574 			brelse(bp);
575 		} else {
576 			/*
577 			 * Otherwise let whoever
578 			 * made the request take care of
579 			 * freeing it. We just queue
580 			 * it onto another list.
581 			 */
582 			bqrelse(bp);
583 		}
584 	}
585 
586 	/*
587 	 * This can only happen in the case of an error
588 	 * because the loop above resets bp to NULL on each iteration
589 	 * and on normal completion has not set a new value into it.
590 	 * so it must have come from a 'break' statement
591 	 */
592 	if (bp != NULL) {
593 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
594 		   (LIST_EMPTY(&bp->b_dep))) {
595 			bp->b_flags |= B_RELBUF;
596 			brelse(bp);
597 		} else {
598 			bqrelse(bp);
599 		}
600 	}
601 
602 	if ((error == 0 || uio->uio_resid != orig_resid) &&
603 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
604 	    (ip->i_flag & IN_ACCESS) == 0) {
605 		VI_LOCK(vp);
606 		ip->i_flag |= IN_ACCESS;
607 		VI_UNLOCK(vp);
608 	}
609 	return (error);
610 }
611 
612 /*
613  * Vnode op for writing.
614  */
615 static int
616 ffs_write(ap)
617 	struct vop_write_args /* {
618 		struct vnode *a_vp;
619 		struct uio *a_uio;
620 		int a_ioflag;
621 		struct ucred *a_cred;
622 	} */ *ap;
623 {
624 	struct vnode *vp;
625 	struct uio *uio;
626 	struct inode *ip;
627 	struct fs *fs;
628 	struct buf *bp;
629 	struct thread *td;
630 	ufs_lbn_t lbn;
631 	off_t osize;
632 	int seqcount;
633 	int blkoffset, error, flags, ioflag, resid, size, xfersize;
634 
635 	vp = ap->a_vp;
636 	uio = ap->a_uio;
637 	ioflag = ap->a_ioflag;
638 	if (ap->a_ioflag & IO_EXT)
639 #ifdef notyet
640 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
641 #else
642 		panic("ffs_write+IO_EXT");
643 #endif
644 
645 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
646 	ip = VTOI(vp);
647 
648 #ifdef INVARIANTS
649 	if (uio->uio_rw != UIO_WRITE)
650 		panic("ffs_write: mode");
651 #endif
652 
653 	switch (vp->v_type) {
654 	case VREG:
655 		if (ioflag & IO_APPEND)
656 			uio->uio_offset = ip->i_size;
657 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
658 			return (EPERM);
659 		/* FALLTHROUGH */
660 	case VLNK:
661 		break;
662 	case VDIR:
663 		panic("ffs_write: dir write");
664 		break;
665 	default:
666 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
667 			(int)uio->uio_offset,
668 			(int)uio->uio_resid
669 		);
670 	}
671 
672 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
673 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
674 	fs = ip->i_fs;
675 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
676 		return (EFBIG);
677 	/*
678 	 * Maybe this should be above the vnode op call, but so long as
679 	 * file servers have no limits, I don't think it matters.
680 	 */
681 	td = uio->uio_td;
682 	if (vp->v_type == VREG && td != NULL) {
683 		PROC_LOCK(td->td_proc);
684 		if (uio->uio_offset + uio->uio_resid >
685 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
686 			psignal(td->td_proc, SIGXFSZ);
687 			PROC_UNLOCK(td->td_proc);
688 			return (EFBIG);
689 		}
690 		PROC_UNLOCK(td->td_proc);
691 	}
692 
693 	resid = uio->uio_resid;
694 	osize = ip->i_size;
695 	if (seqcount > BA_SEQMAX)
696 		flags = BA_SEQMAX << BA_SEQSHIFT;
697 	else
698 		flags = seqcount << BA_SEQSHIFT;
699 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
700 		flags |= IO_SYNC;
701 
702 	for (error = 0; uio->uio_resid > 0;) {
703 		lbn = lblkno(fs, uio->uio_offset);
704 		blkoffset = blkoff(fs, uio->uio_offset);
705 		xfersize = fs->fs_bsize - blkoffset;
706 		if (uio->uio_resid < xfersize)
707 			xfersize = uio->uio_resid;
708 		if (uio->uio_offset + xfersize > ip->i_size)
709 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
710 
711                 /*
712 		 * We must perform a read-before-write if the transfer size
713 		 * does not cover the entire buffer.
714                  */
715 		if (fs->fs_bsize > xfersize)
716 			flags |= BA_CLRBUF;
717 		else
718 			flags &= ~BA_CLRBUF;
719 /* XXX is uio->uio_offset the right thing here? */
720 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
721 		    ap->a_cred, flags, &bp);
722 		if (error != 0)
723 			break;
724 		/*
725 		 * If the buffer is not valid we have to clear out any
726 		 * garbage data from the pages instantiated for the buffer.
727 		 * If we do not, a failed uiomove() during a write can leave
728 		 * the prior contents of the pages exposed to a userland
729 		 * mmap().  XXX deal with uiomove() errors a better way.
730 		 */
731 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
732 			vfs_bio_clrbuf(bp);
733 		if (ioflag & IO_DIRECT)
734 			bp->b_flags |= B_DIRECT;
735 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
736 			bp->b_flags |= B_NOCACHE;
737 
738 		if (uio->uio_offset + xfersize > ip->i_size) {
739 			ip->i_size = uio->uio_offset + xfersize;
740 			DIP_SET(ip, i_size, ip->i_size);
741 		}
742 
743 		size = blksize(fs, ip, lbn) - bp->b_resid;
744 		if (size < xfersize)
745 			xfersize = size;
746 
747 		error =
748 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
749 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
750 		   (LIST_EMPTY(&bp->b_dep))) {
751 			bp->b_flags |= B_RELBUF;
752 		}
753 
754 		/*
755 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
756 		 * if we have a severe page deficiency write the buffer
757 		 * asynchronously.  Otherwise try to cluster, and if that
758 		 * doesn't do it then either do an async write (if O_DIRECT),
759 		 * or a delayed write (if not).
760 		 */
761 		if (ioflag & IO_SYNC) {
762 			(void)bwrite(bp);
763 		} else if (vm_page_count_severe() ||
764 			    buf_dirty_count_severe() ||
765 			    (ioflag & IO_ASYNC)) {
766 			bp->b_flags |= B_CLUSTEROK;
767 			bawrite(bp);
768 		} else if (xfersize + blkoffset == fs->fs_bsize) {
769 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
770 				bp->b_flags |= B_CLUSTEROK;
771 				cluster_write(vp, bp, ip->i_size, seqcount);
772 			} else {
773 				bawrite(bp);
774 			}
775 		} else if (ioflag & IO_DIRECT) {
776 			bp->b_flags |= B_CLUSTEROK;
777 			bawrite(bp);
778 		} else {
779 			bp->b_flags |= B_CLUSTEROK;
780 			bdwrite(bp);
781 		}
782 		if (error || xfersize == 0)
783 			break;
784 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
785 	}
786 	/*
787 	 * If we successfully wrote any data, and we are not the superuser
788 	 * we clear the setuid and setgid bits as a precaution against
789 	 * tampering.
790 	 */
791 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
792 	    ap->a_cred) {
793 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
794 			ip->i_mode &= ~(ISUID | ISGID);
795 			DIP_SET(ip, i_mode, ip->i_mode);
796 		}
797 	}
798 	if (error) {
799 		if (ioflag & IO_UNIT) {
800 			(void)ffs_truncate(vp, osize,
801 			    IO_NORMAL | (ioflag & IO_SYNC),
802 			    ap->a_cred, uio->uio_td);
803 			uio->uio_offset -= resid - uio->uio_resid;
804 			uio->uio_resid = resid;
805 		}
806 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
807 		error = ffs_update(vp, 1);
808 	return (error);
809 }
810 
811 /*
812  * get page routine
813  */
814 static int
815 ffs_getpages(ap)
816 	struct vop_getpages_args *ap;
817 {
818 	int i;
819 	vm_page_t mreq;
820 	int pcount;
821 
822 	pcount = round_page(ap->a_count) / PAGE_SIZE;
823 	mreq = ap->a_m[ap->a_reqpage];
824 
825 	/*
826 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
827 	 * then the entire page is valid.  Since the page may be mapped,
828 	 * user programs might reference data beyond the actual end of file
829 	 * occuring within the page.  We have to zero that data.
830 	 */
831 	VM_OBJECT_LOCK(mreq->object);
832 	if (mreq->valid) {
833 		if (mreq->valid != VM_PAGE_BITS_ALL)
834 			vm_page_zero_invalid(mreq, TRUE);
835 		vm_page_lock_queues();
836 		for (i = 0; i < pcount; i++) {
837 			if (i != ap->a_reqpage) {
838 				vm_page_free(ap->a_m[i]);
839 			}
840 		}
841 		vm_page_unlock_queues();
842 		VM_OBJECT_UNLOCK(mreq->object);
843 		return VM_PAGER_OK;
844 	}
845 	VM_OBJECT_UNLOCK(mreq->object);
846 
847 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
848 					    ap->a_count,
849 					    ap->a_reqpage);
850 }
851 
852 
853 /*
854  * Extended attribute area reading.
855  */
856 static int
857 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
858 {
859 	struct inode *ip;
860 	struct ufs2_dinode *dp;
861 	struct fs *fs;
862 	struct buf *bp;
863 	ufs_lbn_t lbn, nextlbn;
864 	off_t bytesinfile;
865 	long size, xfersize, blkoffset;
866 	int error, orig_resid;
867 
868 	ip = VTOI(vp);
869 	fs = ip->i_fs;
870 	dp = ip->i_din2;
871 
872 #ifdef INVARIANTS
873 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
874 		panic("ffs_extread: mode");
875 
876 #endif
877 	orig_resid = uio->uio_resid;
878 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
879 	if (orig_resid == 0)
880 		return (0);
881 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
882 
883 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
884 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
885 			break;
886 		lbn = lblkno(fs, uio->uio_offset);
887 		nextlbn = lbn + 1;
888 
889 		/*
890 		 * size of buffer.  The buffer representing the
891 		 * end of the file is rounded up to the size of
892 		 * the block type ( fragment or full block,
893 		 * depending ).
894 		 */
895 		size = sblksize(fs, dp->di_extsize, lbn);
896 		blkoffset = blkoff(fs, uio->uio_offset);
897 
898 		/*
899 		 * The amount we want to transfer in this iteration is
900 		 * one FS block less the amount of the data before
901 		 * our startpoint (duh!)
902 		 */
903 		xfersize = fs->fs_bsize - blkoffset;
904 
905 		/*
906 		 * But if we actually want less than the block,
907 		 * or the file doesn't have a whole block more of data,
908 		 * then use the lesser number.
909 		 */
910 		if (uio->uio_resid < xfersize)
911 			xfersize = uio->uio_resid;
912 		if (bytesinfile < xfersize)
913 			xfersize = bytesinfile;
914 
915 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
916 			/*
917 			 * Don't do readahead if this is the end of the info.
918 			 */
919 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
920 		} else {
921 			/*
922 			 * If we have a second block, then
923 			 * fire off a request for a readahead
924 			 * as well as a read. Note that the 4th and 5th
925 			 * arguments point to arrays of the size specified in
926 			 * the 6th argument.
927 			 */
928 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
929 
930 			nextlbn = -1 - nextlbn;
931 			error = breadn(vp, -1 - lbn,
932 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
933 		}
934 		if (error) {
935 			brelse(bp);
936 			bp = NULL;
937 			break;
938 		}
939 
940 		/*
941 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
942 		 * will cause us to attempt to release the buffer later on
943 		 * and will cause the buffer cache to attempt to free the
944 		 * underlying pages.
945 		 */
946 		if (ioflag & IO_DIRECT)
947 			bp->b_flags |= B_DIRECT;
948 
949 		/*
950 		 * We should only get non-zero b_resid when an I/O error
951 		 * has occurred, which should cause us to break above.
952 		 * However, if the short read did not cause an error,
953 		 * then we want to ensure that we do not uiomove bad
954 		 * or uninitialized data.
955 		 */
956 		size -= bp->b_resid;
957 		if (size < xfersize) {
958 			if (size == 0)
959 				break;
960 			xfersize = size;
961 		}
962 
963 		error = uiomove((char *)bp->b_data + blkoffset,
964 					(int)xfersize, uio);
965 		if (error)
966 			break;
967 
968 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
969 		   (LIST_EMPTY(&bp->b_dep))) {
970 			/*
971 			 * If there are no dependencies, and it's VMIO,
972 			 * then we don't need the buf, mark it available
973 			 * for freeing. The VM has the data.
974 			 */
975 			bp->b_flags |= B_RELBUF;
976 			brelse(bp);
977 		} else {
978 			/*
979 			 * Otherwise let whoever
980 			 * made the request take care of
981 			 * freeing it. We just queue
982 			 * it onto another list.
983 			 */
984 			bqrelse(bp);
985 		}
986 	}
987 
988 	/*
989 	 * This can only happen in the case of an error
990 	 * because the loop above resets bp to NULL on each iteration
991 	 * and on normal completion has not set a new value into it.
992 	 * so it must have come from a 'break' statement
993 	 */
994 	if (bp != NULL) {
995 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
996 		   (LIST_EMPTY(&bp->b_dep))) {
997 			bp->b_flags |= B_RELBUF;
998 			brelse(bp);
999 		} else {
1000 			bqrelse(bp);
1001 		}
1002 	}
1003 
1004 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1005 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0 &&
1006 	    (ip->i_flag & IN_ACCESS) == 0) {
1007 		VI_LOCK(vp);
1008 		ip->i_flag |= IN_ACCESS;
1009 		VI_UNLOCK(vp);
1010 	}
1011 	return (error);
1012 }
1013 
1014 /*
1015  * Extended attribute area writing.
1016  */
1017 static int
1018 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1019 {
1020 	struct inode *ip;
1021 	struct ufs2_dinode *dp;
1022 	struct fs *fs;
1023 	struct buf *bp;
1024 	ufs_lbn_t lbn;
1025 	off_t osize;
1026 	int blkoffset, error, flags, resid, size, xfersize;
1027 
1028 	ip = VTOI(vp);
1029 	fs = ip->i_fs;
1030 	dp = ip->i_din2;
1031 
1032 	KASSERT(!(ip->i_flag & IN_SPACECOUNTED), ("inode %u: inode is dead",
1033 	    ip->i_number));
1034 
1035 #ifdef INVARIANTS
1036 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1037 		panic("ffs_extwrite: mode");
1038 #endif
1039 
1040 	if (ioflag & IO_APPEND)
1041 		uio->uio_offset = dp->di_extsize;
1042 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1043 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1044 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1045 		return (EFBIG);
1046 
1047 	resid = uio->uio_resid;
1048 	osize = dp->di_extsize;
1049 	flags = IO_EXT;
1050 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1051 		flags |= IO_SYNC;
1052 
1053 	for (error = 0; uio->uio_resid > 0;) {
1054 		lbn = lblkno(fs, uio->uio_offset);
1055 		blkoffset = blkoff(fs, uio->uio_offset);
1056 		xfersize = fs->fs_bsize - blkoffset;
1057 		if (uio->uio_resid < xfersize)
1058 			xfersize = uio->uio_resid;
1059 
1060 		/*
1061 		 * We must perform a read-before-write if the transfer size
1062 		 * does not cover the entire buffer.
1063                  */
1064 		if (fs->fs_bsize > xfersize)
1065 			flags |= BA_CLRBUF;
1066 		else
1067 			flags &= ~BA_CLRBUF;
1068 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1069 		    ucred, flags, &bp);
1070 		if (error != 0)
1071 			break;
1072 		/*
1073 		 * If the buffer is not valid we have to clear out any
1074 		 * garbage data from the pages instantiated for the buffer.
1075 		 * If we do not, a failed uiomove() during a write can leave
1076 		 * the prior contents of the pages exposed to a userland
1077 		 * mmap().  XXX deal with uiomove() errors a better way.
1078 		 */
1079 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1080 			vfs_bio_clrbuf(bp);
1081 		if (ioflag & IO_DIRECT)
1082 			bp->b_flags |= B_DIRECT;
1083 
1084 		if (uio->uio_offset + xfersize > dp->di_extsize)
1085 			dp->di_extsize = uio->uio_offset + xfersize;
1086 
1087 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1088 		if (size < xfersize)
1089 			xfersize = size;
1090 
1091 		error =
1092 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1093 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1094 		   (LIST_EMPTY(&bp->b_dep))) {
1095 			bp->b_flags |= B_RELBUF;
1096 		}
1097 
1098 		/*
1099 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1100 		 * if we have a severe page deficiency write the buffer
1101 		 * asynchronously.  Otherwise try to cluster, and if that
1102 		 * doesn't do it then either do an async write (if O_DIRECT),
1103 		 * or a delayed write (if not).
1104 		 */
1105 		if (ioflag & IO_SYNC) {
1106 			(void)bwrite(bp);
1107 		} else if (vm_page_count_severe() ||
1108 			    buf_dirty_count_severe() ||
1109 			    xfersize + blkoffset == fs->fs_bsize ||
1110 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1111 			bawrite(bp);
1112 		else
1113 			bdwrite(bp);
1114 		if (error || xfersize == 0)
1115 			break;
1116 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1117 	}
1118 	/*
1119 	 * If we successfully wrote any data, and we are not the superuser
1120 	 * we clear the setuid and setgid bits as a precaution against
1121 	 * tampering.
1122 	 */
1123 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1124 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1125 			ip->i_mode &= ~(ISUID | ISGID);
1126 			dp->di_mode = ip->i_mode;
1127 		}
1128 	}
1129 	if (error) {
1130 		if (ioflag & IO_UNIT) {
1131 			(void)ffs_truncate(vp, osize,
1132 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1133 			uio->uio_offset -= resid - uio->uio_resid;
1134 			uio->uio_resid = resid;
1135 		}
1136 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1137 		error = ffs_update(vp, 1);
1138 	return (error);
1139 }
1140 
1141 
1142 /*
1143  * Vnode operating to retrieve a named extended attribute.
1144  *
1145  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1146  * the length of the EA, and possibly the pointer to the entry and to the data.
1147  */
1148 static int
1149 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1150 {
1151 	u_char *p, *pe, *pn, *p0;
1152 	int eapad1, eapad2, ealength, ealen, nlen;
1153 	uint32_t ul;
1154 
1155 	pe = ptr + length;
1156 	nlen = strlen(name);
1157 
1158 	for (p = ptr; p < pe; p = pn) {
1159 		p0 = p;
1160 		bcopy(p, &ul, sizeof(ul));
1161 		pn = p + ul;
1162 		/* make sure this entry is complete */
1163 		if (pn > pe)
1164 			break;
1165 		p += sizeof(uint32_t);
1166 		if (*p != nspace)
1167 			continue;
1168 		p++;
1169 		eapad2 = *p++;
1170 		if (*p != nlen)
1171 			continue;
1172 		p++;
1173 		if (bcmp(p, name, nlen))
1174 			continue;
1175 		ealength = sizeof(uint32_t) + 3 + nlen;
1176 		eapad1 = 8 - (ealength % 8);
1177 		if (eapad1 == 8)
1178 			eapad1 = 0;
1179 		ealength += eapad1;
1180 		ealen = ul - ealength - eapad2;
1181 		p += nlen + eapad1;
1182 		if (eap != NULL)
1183 			*eap = p0;
1184 		if (eac != NULL)
1185 			*eac = p;
1186 		return (ealen);
1187 	}
1188 	return(-1);
1189 }
1190 
1191 static int
1192 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1193 {
1194 	struct inode *ip;
1195 	struct ufs2_dinode *dp;
1196 	struct fs *fs;
1197 	struct uio luio;
1198 	struct iovec liovec;
1199 	int easize, error;
1200 	u_char *eae;
1201 
1202 	ip = VTOI(vp);
1203 	fs = ip->i_fs;
1204 	dp = ip->i_din2;
1205 	easize = dp->di_extsize;
1206 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1207 		return (EFBIG);
1208 
1209 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1210 
1211 	liovec.iov_base = eae;
1212 	liovec.iov_len = easize;
1213 	luio.uio_iov = &liovec;
1214 	luio.uio_iovcnt = 1;
1215 	luio.uio_offset = 0;
1216 	luio.uio_resid = easize;
1217 	luio.uio_segflg = UIO_SYSSPACE;
1218 	luio.uio_rw = UIO_READ;
1219 	luio.uio_td = td;
1220 
1221 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1222 	if (error) {
1223 		free(eae, M_TEMP);
1224 		return(error);
1225 	}
1226 	*p = eae;
1227 	return (0);
1228 }
1229 
1230 static int
1231 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1232 {
1233 	struct inode *ip;
1234 	struct ufs2_dinode *dp;
1235 	int error;
1236 
1237 	ip = VTOI(vp);
1238 
1239 	if (ip->i_ea_area != NULL)
1240 		return (EBUSY);
1241 	dp = ip->i_din2;
1242 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1243 	if (error)
1244 		return (error);
1245 	ip->i_ea_len = dp->di_extsize;
1246 	ip->i_ea_error = 0;
1247 	return (0);
1248 }
1249 
1250 /*
1251  * Vnode extattr transaction commit/abort
1252  */
1253 static int
1254 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1255 {
1256 	struct inode *ip;
1257 	struct uio luio;
1258 	struct iovec liovec;
1259 	int error;
1260 	struct ufs2_dinode *dp;
1261 
1262 	ip = VTOI(vp);
1263 	if (ip->i_ea_area == NULL)
1264 		return (EINVAL);
1265 	dp = ip->i_din2;
1266 	error = ip->i_ea_error;
1267 	if (commit && error == 0) {
1268 		if (cred == NOCRED)
1269 			cred =  vp->v_mount->mnt_cred;
1270 		liovec.iov_base = ip->i_ea_area;
1271 		liovec.iov_len = ip->i_ea_len;
1272 		luio.uio_iov = &liovec;
1273 		luio.uio_iovcnt = 1;
1274 		luio.uio_offset = 0;
1275 		luio.uio_resid = ip->i_ea_len;
1276 		luio.uio_segflg = UIO_SYSSPACE;
1277 		luio.uio_rw = UIO_WRITE;
1278 		luio.uio_td = td;
1279 		/* XXX: I'm not happy about truncating to zero size */
1280 		if (ip->i_ea_len < dp->di_extsize)
1281 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1282 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1283 	}
1284 	free(ip->i_ea_area, M_TEMP);
1285 	ip->i_ea_area = NULL;
1286 	ip->i_ea_len = 0;
1287 	ip->i_ea_error = 0;
1288 	return (error);
1289 }
1290 
1291 /*
1292  * Vnode extattr strategy routine for fifos.
1293  *
1294  * We need to check for a read or write of the external attributes.
1295  * Otherwise we just fall through and do the usual thing.
1296  */
1297 static int
1298 ffsext_strategy(struct vop_strategy_args *ap)
1299 /*
1300 struct vop_strategy_args {
1301 	struct vnodeop_desc *a_desc;
1302 	struct vnode *a_vp;
1303 	struct buf *a_bp;
1304 };
1305 */
1306 {
1307 	struct vnode *vp;
1308 	daddr_t lbn;
1309 
1310 	vp = ap->a_vp;
1311 	lbn = ap->a_bp->b_lblkno;
1312 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1313 	    lbn < 0 && lbn >= -NXADDR)
1314 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1315 	if (vp->v_type == VFIFO)
1316 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1317 	panic("spec nodes went here");
1318 }
1319 
1320 /*
1321  * Vnode extattr transaction commit/abort
1322  */
1323 static int
1324 ffs_openextattr(struct vop_openextattr_args *ap)
1325 /*
1326 struct vop_openextattr_args {
1327 	struct vnodeop_desc *a_desc;
1328 	struct vnode *a_vp;
1329 	IN struct ucred *a_cred;
1330 	IN struct thread *a_td;
1331 };
1332 */
1333 {
1334 	struct inode *ip;
1335 	struct fs *fs;
1336 
1337 	ip = VTOI(ap->a_vp);
1338 	fs = ip->i_fs;
1339 
1340 	if (ap->a_vp->v_type == VCHR)
1341 		return (EOPNOTSUPP);
1342 
1343 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1344 }
1345 
1346 
1347 /*
1348  * Vnode extattr transaction commit/abort
1349  */
1350 static int
1351 ffs_closeextattr(struct vop_closeextattr_args *ap)
1352 /*
1353 struct vop_closeextattr_args {
1354 	struct vnodeop_desc *a_desc;
1355 	struct vnode *a_vp;
1356 	int a_commit;
1357 	IN struct ucred *a_cred;
1358 	IN struct thread *a_td;
1359 };
1360 */
1361 {
1362 	struct inode *ip;
1363 	struct fs *fs;
1364 
1365 	ip = VTOI(ap->a_vp);
1366 	fs = ip->i_fs;
1367 
1368 	if (ap->a_vp->v_type == VCHR)
1369 		return (EOPNOTSUPP);
1370 
1371 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1372 		return (EROFS);
1373 
1374 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1375 }
1376 
1377 /*
1378  * Vnode operation to remove a named attribute.
1379  */
1380 static int
1381 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1382 /*
1383 vop_deleteextattr {
1384 	IN struct vnode *a_vp;
1385 	IN int a_attrnamespace;
1386 	IN const char *a_name;
1387 	IN struct ucred *a_cred;
1388 	IN struct thread *a_td;
1389 };
1390 */
1391 {
1392 	struct inode *ip;
1393 	struct fs *fs;
1394 	uint32_t ealength, ul;
1395 	int ealen, olen, eapad1, eapad2, error, i, easize;
1396 	u_char *eae, *p;
1397 	int stand_alone;
1398 
1399 	ip = VTOI(ap->a_vp);
1400 	fs = ip->i_fs;
1401 
1402 	if (ap->a_vp->v_type == VCHR)
1403 		return (EOPNOTSUPP);
1404 
1405 	if (strlen(ap->a_name) == 0)
1406 		return (EINVAL);
1407 
1408 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1409 		return (EROFS);
1410 
1411 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1412 	    ap->a_cred, ap->a_td, VWRITE);
1413 	if (error) {
1414 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1415 			ip->i_ea_error = error;
1416 		return (error);
1417 	}
1418 
1419 	if (ip->i_ea_area == NULL) {
1420 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1421 		if (error)
1422 			return (error);
1423 		stand_alone = 1;
1424 	} else {
1425 		stand_alone = 0;
1426 	}
1427 
1428 	ealength = eapad1 = ealen = eapad2 = 0;
1429 
1430 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1431 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1432 	easize = ip->i_ea_len;
1433 
1434 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1435 	    &p, NULL);
1436 	if (olen == -1) {
1437 		/* delete but nonexistent */
1438 		free(eae, M_TEMP);
1439 		if (stand_alone)
1440 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1441 		return(ENOATTR);
1442 	}
1443 	bcopy(p, &ul, sizeof ul);
1444 	i = p - eae + ul;
1445 	if (ul != ealength) {
1446 		bcopy(p + ul, p + ealength, easize - i);
1447 		easize += (ealength - ul);
1448 	}
1449 	if (easize > NXADDR * fs->fs_bsize) {
1450 		free(eae, M_TEMP);
1451 		if (stand_alone)
1452 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1453 		else if (ip->i_ea_error == 0)
1454 			ip->i_ea_error = ENOSPC;
1455 		return(ENOSPC);
1456 	}
1457 	p = ip->i_ea_area;
1458 	ip->i_ea_area = eae;
1459 	ip->i_ea_len = easize;
1460 	free(p, M_TEMP);
1461 	if (stand_alone)
1462 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1463 	return(error);
1464 }
1465 
1466 /*
1467  * Vnode operation to retrieve a named extended attribute.
1468  */
1469 static int
1470 ffs_getextattr(struct vop_getextattr_args *ap)
1471 /*
1472 vop_getextattr {
1473 	IN struct vnode *a_vp;
1474 	IN int a_attrnamespace;
1475 	IN const char *a_name;
1476 	INOUT struct uio *a_uio;
1477 	OUT size_t *a_size;
1478 	IN struct ucred *a_cred;
1479 	IN struct thread *a_td;
1480 };
1481 */
1482 {
1483 	struct inode *ip;
1484 	struct fs *fs;
1485 	u_char *eae, *p;
1486 	unsigned easize;
1487 	int error, ealen, stand_alone;
1488 
1489 	ip = VTOI(ap->a_vp);
1490 	fs = ip->i_fs;
1491 
1492 	if (ap->a_vp->v_type == VCHR)
1493 		return (EOPNOTSUPP);
1494 
1495 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1496 	    ap->a_cred, ap->a_td, VREAD);
1497 	if (error)
1498 		return (error);
1499 
1500 	if (ip->i_ea_area == NULL) {
1501 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1502 		if (error)
1503 			return (error);
1504 		stand_alone = 1;
1505 	} else {
1506 		stand_alone = 0;
1507 	}
1508 	eae = ip->i_ea_area;
1509 	easize = ip->i_ea_len;
1510 
1511 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1512 	    NULL, &p);
1513 	if (ealen >= 0) {
1514 		error = 0;
1515 		if (ap->a_size != NULL)
1516 			*ap->a_size = ealen;
1517 		else if (ap->a_uio != NULL)
1518 			error = uiomove(p, ealen, ap->a_uio);
1519 	} else
1520 		error = ENOATTR;
1521 	if (stand_alone)
1522 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1523 	return(error);
1524 }
1525 
1526 /*
1527  * Vnode operation to retrieve extended attributes on a vnode.
1528  */
1529 static int
1530 ffs_listextattr(struct vop_listextattr_args *ap)
1531 /*
1532 vop_listextattr {
1533 	IN struct vnode *a_vp;
1534 	IN int a_attrnamespace;
1535 	INOUT struct uio *a_uio;
1536 	OUT size_t *a_size;
1537 	IN struct ucred *a_cred;
1538 	IN struct thread *a_td;
1539 };
1540 */
1541 {
1542 	struct inode *ip;
1543 	struct fs *fs;
1544 	u_char *eae, *p, *pe, *pn;
1545 	unsigned easize;
1546 	uint32_t ul;
1547 	int error, ealen, stand_alone;
1548 
1549 	ip = VTOI(ap->a_vp);
1550 	fs = ip->i_fs;
1551 
1552 	if (ap->a_vp->v_type == VCHR)
1553 		return (EOPNOTSUPP);
1554 
1555 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1556 	    ap->a_cred, ap->a_td, VREAD);
1557 	if (error)
1558 		return (error);
1559 
1560 	if (ip->i_ea_area == NULL) {
1561 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1562 		if (error)
1563 			return (error);
1564 		stand_alone = 1;
1565 	} else {
1566 		stand_alone = 0;
1567 	}
1568 	eae = ip->i_ea_area;
1569 	easize = ip->i_ea_len;
1570 
1571 	error = 0;
1572 	if (ap->a_size != NULL)
1573 		*ap->a_size = 0;
1574 	pe = eae + easize;
1575 	for(p = eae; error == 0 && p < pe; p = pn) {
1576 		bcopy(p, &ul, sizeof(ul));
1577 		pn = p + ul;
1578 		if (pn > pe)
1579 			break;
1580 		p += sizeof(ul);
1581 		if (*p++ != ap->a_attrnamespace)
1582 			continue;
1583 		p++;	/* pad2 */
1584 		ealen = *p;
1585 		if (ap->a_size != NULL) {
1586 			*ap->a_size += ealen + 1;
1587 		} else if (ap->a_uio != NULL) {
1588 			error = uiomove(p, ealen + 1, ap->a_uio);
1589 		}
1590 	}
1591 	if (stand_alone)
1592 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1593 	return(error);
1594 }
1595 
1596 /*
1597  * Vnode operation to set a named attribute.
1598  */
1599 static int
1600 ffs_setextattr(struct vop_setextattr_args *ap)
1601 /*
1602 vop_setextattr {
1603 	IN struct vnode *a_vp;
1604 	IN int a_attrnamespace;
1605 	IN const char *a_name;
1606 	INOUT struct uio *a_uio;
1607 	IN struct ucred *a_cred;
1608 	IN struct thread *a_td;
1609 };
1610 */
1611 {
1612 	struct inode *ip;
1613 	struct fs *fs;
1614 	uint32_t ealength, ul;
1615 	int ealen, olen, eapad1, eapad2, error, i, easize;
1616 	u_char *eae, *p;
1617 	int stand_alone;
1618 
1619 	ip = VTOI(ap->a_vp);
1620 	fs = ip->i_fs;
1621 
1622 	if (ap->a_vp->v_type == VCHR)
1623 		return (EOPNOTSUPP);
1624 
1625 	if (strlen(ap->a_name) == 0)
1626 		return (EINVAL);
1627 
1628 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1629 	if (ap->a_uio == NULL)
1630 		return (EOPNOTSUPP);
1631 
1632 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1633 		return (EROFS);
1634 
1635 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1636 	    ap->a_cred, ap->a_td, VWRITE);
1637 	if (error) {
1638 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1639 			ip->i_ea_error = error;
1640 		return (error);
1641 	}
1642 
1643 	if (ip->i_ea_area == NULL) {
1644 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1645 		if (error)
1646 			return (error);
1647 		stand_alone = 1;
1648 	} else {
1649 		stand_alone = 0;
1650 	}
1651 
1652 	ealen = ap->a_uio->uio_resid;
1653 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1654 	eapad1 = 8 - (ealength % 8);
1655 	if (eapad1 == 8)
1656 		eapad1 = 0;
1657 	eapad2 = 8 - (ealen % 8);
1658 	if (eapad2 == 8)
1659 		eapad2 = 0;
1660 	ealength += eapad1 + ealen + eapad2;
1661 
1662 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1663 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1664 	easize = ip->i_ea_len;
1665 
1666 	olen = ffs_findextattr(eae, easize,
1667 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1668         if (olen == -1) {
1669 		/* new, append at end */
1670 		p = eae + easize;
1671 		easize += ealength;
1672 	} else {
1673 		bcopy(p, &ul, sizeof ul);
1674 		i = p - eae + ul;
1675 		if (ul != ealength) {
1676 			bcopy(p + ul, p + ealength, easize - i);
1677 			easize += (ealength - ul);
1678 		}
1679 	}
1680 	if (easize > NXADDR * fs->fs_bsize) {
1681 		free(eae, M_TEMP);
1682 		if (stand_alone)
1683 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1684 		else if (ip->i_ea_error == 0)
1685 			ip->i_ea_error = ENOSPC;
1686 		return(ENOSPC);
1687 	}
1688 	bcopy(&ealength, p, sizeof(ealength));
1689 	p += sizeof(ealength);
1690 	*p++ = ap->a_attrnamespace;
1691 	*p++ = eapad2;
1692 	*p++ = strlen(ap->a_name);
1693 	strcpy(p, ap->a_name);
1694 	p += strlen(ap->a_name);
1695 	bzero(p, eapad1);
1696 	p += eapad1;
1697 	error = uiomove(p, ealen, ap->a_uio);
1698 	if (error) {
1699 		free(eae, M_TEMP);
1700 		if (stand_alone)
1701 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1702 		else if (ip->i_ea_error == 0)
1703 			ip->i_ea_error = error;
1704 		return(error);
1705 	}
1706 	p += ealen;
1707 	bzero(p, eapad2);
1708 
1709 	p = ip->i_ea_area;
1710 	ip->i_ea_area = eae;
1711 	ip->i_ea_len = easize;
1712 	free(p, M_TEMP);
1713 	if (stand_alone)
1714 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1715 	return(error);
1716 }
1717 
1718 /*
1719  * Vnode pointer to File handle
1720  */
1721 static int
1722 ffs_vptofh(struct vop_vptofh_args *ap)
1723 /*
1724 vop_vptofh {
1725 	IN struct vnode *a_vp;
1726 	IN struct fid *a_fhp;
1727 };
1728 */
1729 {
1730 	struct inode *ip;
1731 	struct ufid *ufhp;
1732 
1733 	ip = VTOI(ap->a_vp);
1734 	ufhp = (struct ufid *)ap->a_fhp;
1735 	ufhp->ufid_len = sizeof(struct ufid);
1736 	ufhp->ufid_ino = ip->i_number;
1737 	ufhp->ufid_gen = ip->i_gen;
1738 	return (0);
1739 }
1740