xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 1713e81b9cdf06d2a9a365a7ded13a54fad84798)
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/proc.h>
78 #include <sys/resourcevar.h>
79 #include <sys/signalvar.h>
80 #include <sys/stat.h>
81 #include <sys/vmmeter.h>
82 #include <sys/vnode.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vnode_pager.h>
90 
91 #include <ufs/ufs/extattr.h>
92 #include <ufs/ufs/quota.h>
93 #include <ufs/ufs/inode.h>
94 #include <ufs/ufs/ufs_extern.h>
95 #include <ufs/ufs/ufsmount.h>
96 
97 #include <ufs/ffs/fs.h>
98 #include <ufs/ffs/ffs_extern.h>
99 #include "opt_directio.h"
100 
101 #ifdef DIRECTIO
102 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
103 #endif
104 static int	ffs_fsync(struct vop_fsync_args *);
105 static int	ffs_getpages(struct vop_getpages_args *);
106 static int	ffs_read(struct vop_read_args *);
107 static int	ffs_write(struct vop_write_args *);
108 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
109 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
110 		    struct ucred *cred);
111 static int	ffsext_strategy(struct vop_strategy_args *);
112 static int	ffs_closeextattr(struct vop_closeextattr_args *);
113 static int	ffs_deleteextattr(struct vop_deleteextattr_args *);
114 static int	ffs_getextattr(struct vop_getextattr_args *);
115 static int	ffs_listextattr(struct vop_listextattr_args *);
116 static int	ffs_openextattr(struct vop_openextattr_args *);
117 static int	ffs_setextattr(struct vop_setextattr_args *);
118 
119 
120 /* Global vfs data structures for ufs. */
121 vop_t **ffs_vnodeop_p;
122 static struct vnodeopv_entry_desc ffs_vnodeop_entries[] = {
123 	{ &vop_default_desc,		(vop_t *) ufs_vnoperate },
124 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
125 	{ &vop_getpages_desc,		(vop_t *) ffs_getpages },
126 	{ &vop_read_desc,		(vop_t *) ffs_read },
127 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
128 	{ &vop_write_desc,		(vop_t *) ffs_write },
129 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
130 	{ &vop_deleteextattr_desc,	(vop_t *) ffs_deleteextattr },
131 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
132 	{ &vop_listextattr_desc,	(vop_t *) ffs_listextattr },
133 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
134 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
135 	{ NULL, NULL }
136 };
137 static struct vnodeopv_desc ffs_vnodeop_opv_desc =
138 	{ &ffs_vnodeop_p, ffs_vnodeop_entries };
139 
140 vop_t **ffs_fifoop_p;
141 static struct vnodeopv_entry_desc ffs_fifoop_entries[] = {
142 	{ &vop_default_desc,		(vop_t *) ufs_vnoperatefifo },
143 	{ &vop_fsync_desc,		(vop_t *) ffs_fsync },
144 	{ &vop_reallocblks_desc,	(vop_t *) ffs_reallocblks },
145 	{ &vop_strategy_desc,		(vop_t *) ffsext_strategy },
146 	{ &vop_closeextattr_desc,	(vop_t *) ffs_closeextattr },
147 	{ &vop_deleteextattr_desc,	(vop_t *) ffs_deleteextattr },
148 	{ &vop_getextattr_desc,		(vop_t *) ffs_getextattr },
149 	{ &vop_listextattr_desc,	(vop_t *) ffs_listextattr },
150 	{ &vop_openextattr_desc,	(vop_t *) ffs_openextattr },
151 	{ &vop_setextattr_desc,		(vop_t *) ffs_setextattr },
152 	{ NULL, NULL }
153 };
154 static struct vnodeopv_desc ffs_fifoop_opv_desc =
155 	{ &ffs_fifoop_p, ffs_fifoop_entries };
156 
157 VNODEOP_SET(ffs_vnodeop_opv_desc);
158 VNODEOP_SET(ffs_fifoop_opv_desc);
159 
160 /*
161  * Synch an open file.
162  */
163 /* ARGSUSED */
164 static int
165 ffs_fsync(ap)
166 	struct vop_fsync_args /* {
167 		struct vnode *a_vp;
168 		struct ucred *a_cred;
169 		int a_waitfor;
170 		struct thread *a_td;
171 	} */ *ap;
172 {
173 	struct vnode *vp = ap->a_vp;
174 	struct inode *ip = VTOI(vp);
175 	struct buf *bp;
176 	struct buf *nbp;
177 	int s, error, wait, passes, skipmeta;
178 	ufs_lbn_t lbn;
179 
180 	wait = (ap->a_waitfor == MNT_WAIT);
181 	if (vn_isdisk(vp, NULL)) {
182 		lbn = INT_MAX;
183 		if (vp->v_rdev->si_mountpoint != NULL &&
184 		    (vp->v_rdev->si_mountpoint->mnt_flag & MNT_SOFTDEP))
185 			softdep_fsync_mountdev(vp);
186 	} else {
187 		lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
188 	}
189 
190 	/*
191 	 * Flush all dirty buffers associated with a vnode.
192 	 */
193 	passes = NIADDR + 1;
194 	skipmeta = 0;
195 	if (wait)
196 		skipmeta = 1;
197 	s = splbio();
198 	VI_LOCK(vp);
199 loop:
200 	TAILQ_FOREACH(bp, &vp->v_dirtyblkhd, b_bobufs)
201 		bp->b_vflags &= ~BV_SCANNED;
202 	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
203 		nbp = TAILQ_NEXT(bp, b_bobufs);
204 		/*
205 		 * Reasons to skip this buffer: it has already been considered
206 		 * on this pass, this pass is the first time through on a
207 		 * synchronous flush request and the buffer being considered
208 		 * is metadata, the buffer has dependencies that will cause
209 		 * it to be redirtied and it has not already been deferred,
210 		 * or it is already being written.
211 		 */
212 		if ((bp->b_vflags & BV_SCANNED) != 0)
213 			continue;
214 		bp->b_vflags |= BV_SCANNED;
215 		if ((skipmeta == 1 && bp->b_lblkno < 0))
216 			continue;
217 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
218 			continue;
219 		if (!wait && LIST_FIRST(&bp->b_dep) != NULL &&
220 		    (bp->b_flags & B_DEFERRED) == 0 &&
221 		    buf_countdeps(bp, 0)) {
222 			bp->b_flags |= B_DEFERRED;
223 			BUF_UNLOCK(bp);
224 			continue;
225 		}
226 		VI_UNLOCK(vp);
227 		if ((bp->b_flags & B_DELWRI) == 0)
228 			panic("ffs_fsync: not dirty");
229 		if (vp != bp->b_vp)
230 			panic("ffs_fsync: vp != vp->b_vp");
231 		/*
232 		 * If this is a synchronous flush request, or it is not a
233 		 * file or device, start the write on this buffer immediatly.
234 		 */
235 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
236 
237 			/*
238 			 * On our final pass through, do all I/O synchronously
239 			 * so that we can find out if our flush is failing
240 			 * because of write errors.
241 			 */
242 			if (passes > 0 || !wait) {
243 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
244 					(void) vfs_bio_awrite(bp);
245 				} else {
246 					bremfree(bp);
247 					splx(s);
248 					(void) bawrite(bp);
249 					s = splbio();
250 				}
251 			} else {
252 				bremfree(bp);
253 				splx(s);
254 				if ((error = bwrite(bp)) != 0)
255 					return (error);
256 				s = splbio();
257 			}
258 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
259 			/*
260 			 * If the buffer is for data that has been truncated
261 			 * off the file, then throw it away.
262 			 */
263 			bremfree(bp);
264 			bp->b_flags |= B_INVAL | B_NOCACHE;
265 			splx(s);
266 			brelse(bp);
267 			s = splbio();
268 		} else
269 			vfs_bio_awrite(bp);
270 
271 		/*
272 		 * Since we may have slept during the I/O, we need
273 		 * to start from a known point.
274 		 */
275 		VI_LOCK(vp);
276 		nbp = TAILQ_FIRST(&vp->v_dirtyblkhd);
277 	}
278 	/*
279 	 * If we were asked to do this synchronously, then go back for
280 	 * another pass, this time doing the metadata.
281 	 */
282 	if (skipmeta) {
283 		skipmeta = 0;
284 		goto loop;
285 	}
286 
287 	if (wait) {
288 		bufobj_wwait(&vp->v_bufobj, 3, 0);
289 		VI_UNLOCK(vp);
290 
291 		/*
292 		 * Ensure that any filesystem metatdata associated
293 		 * with the vnode has been written.
294 		 */
295 		splx(s);
296 		if ((error = softdep_sync_metadata(ap)) != 0)
297 			return (error);
298 		s = splbio();
299 
300 		VI_LOCK(vp);
301 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
302 			/*
303 			 * Block devices associated with filesystems may
304 			 * have new I/O requests posted for them even if
305 			 * the vnode is locked, so no amount of trying will
306 			 * get them clean. Thus we give block devices a
307 			 * good effort, then just give up. For all other file
308 			 * types, go around and try again until it is clean.
309 			 */
310 			if (passes > 0) {
311 				passes -= 1;
312 				goto loop;
313 			}
314 #ifdef DIAGNOSTIC
315 			if (!vn_isdisk(vp, NULL))
316 				vprint("ffs_fsync: dirty", vp);
317 #endif
318 		}
319 	}
320 	VI_UNLOCK(vp);
321 	splx(s);
322 	return (UFS_UPDATE(vp, wait));
323 }
324 
325 
326 /*
327  * Vnode op for reading.
328  */
329 /* ARGSUSED */
330 static int
331 ffs_read(ap)
332 	struct vop_read_args /* {
333 		struct vnode *a_vp;
334 		struct uio *a_uio;
335 		int a_ioflag;
336 		struct ucred *a_cred;
337 	} */ *ap;
338 {
339 	struct vnode *vp;
340 	struct inode *ip;
341 	struct uio *uio;
342 	struct fs *fs;
343 	struct buf *bp;
344 	ufs_lbn_t lbn, nextlbn;
345 	off_t bytesinfile;
346 	long size, xfersize, blkoffset;
347 	int error, orig_resid;
348 	int seqcount;
349 	int ioflag;
350 
351 	vp = ap->a_vp;
352 	uio = ap->a_uio;
353 	ioflag = ap->a_ioflag;
354 	if (ap->a_ioflag & IO_EXT)
355 #ifdef notyet
356 		return (ffs_extread(vp, uio, ioflag));
357 #else
358 		panic("ffs_read+IO_EXT");
359 #endif
360 #ifdef DIRECTIO
361 	if ((ioflag & IO_DIRECT) != 0) {
362 		int workdone;
363 
364 		error = ffs_rawread(vp, uio, &workdone);
365 		if (error != 0 || workdone != 0)
366 			return error;
367 	}
368 #endif
369 
370 	GIANT_REQUIRED;
371 
372 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
373 	ip = VTOI(vp);
374 
375 #ifdef DIAGNOSTIC
376 	if (uio->uio_rw != UIO_READ)
377 		panic("ffs_read: mode");
378 
379 	if (vp->v_type == VLNK) {
380 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
381 			panic("ffs_read: short symlink");
382 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
383 		panic("ffs_read: type %d",  vp->v_type);
384 #endif
385 	orig_resid = uio->uio_resid;
386 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
387 	if (orig_resid == 0)
388 		return (0);
389 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
390 	fs = ip->i_fs;
391 	if (uio->uio_offset < ip->i_size &&
392 	    uio->uio_offset >= fs->fs_maxfilesize)
393 		return (EOVERFLOW);
394 
395 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
396 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
397 			break;
398 		lbn = lblkno(fs, uio->uio_offset);
399 		nextlbn = lbn + 1;
400 
401 		/*
402 		 * size of buffer.  The buffer representing the
403 		 * end of the file is rounded up to the size of
404 		 * the block type ( fragment or full block,
405 		 * depending ).
406 		 */
407 		size = blksize(fs, ip, lbn);
408 		blkoffset = blkoff(fs, uio->uio_offset);
409 
410 		/*
411 		 * The amount we want to transfer in this iteration is
412 		 * one FS block less the amount of the data before
413 		 * our startpoint (duh!)
414 		 */
415 		xfersize = fs->fs_bsize - blkoffset;
416 
417 		/*
418 		 * But if we actually want less than the block,
419 		 * or the file doesn't have a whole block more of data,
420 		 * then use the lesser number.
421 		 */
422 		if (uio->uio_resid < xfersize)
423 			xfersize = uio->uio_resid;
424 		if (bytesinfile < xfersize)
425 			xfersize = bytesinfile;
426 
427 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
428 			/*
429 			 * Don't do readahead if this is the end of the file.
430 			 */
431 			error = bread(vp, lbn, size, NOCRED, &bp);
432 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
433 			/*
434 			 * Otherwise if we are allowed to cluster,
435 			 * grab as much as we can.
436 			 *
437 			 * XXX  This may not be a win if we are not
438 			 * doing sequential access.
439 			 */
440 			error = cluster_read(vp, ip->i_size, lbn,
441 				size, NOCRED, uio->uio_resid, seqcount, &bp);
442 		} else if (seqcount > 1) {
443 			/*
444 			 * If we are NOT allowed to cluster, then
445 			 * if we appear to be acting sequentially,
446 			 * fire off a request for a readahead
447 			 * as well as a read. Note that the 4th and 5th
448 			 * arguments point to arrays of the size specified in
449 			 * the 6th argument.
450 			 */
451 			int nextsize = blksize(fs, ip, nextlbn);
452 			error = breadn(vp, lbn,
453 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
454 		} else {
455 			/*
456 			 * Failing all of the above, just read what the
457 			 * user asked for. Interestingly, the same as
458 			 * the first option above.
459 			 */
460 			error = bread(vp, lbn, size, NOCRED, &bp);
461 		}
462 		if (error) {
463 			brelse(bp);
464 			bp = NULL;
465 			break;
466 		}
467 
468 		/*
469 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
470 		 * will cause us to attempt to release the buffer later on
471 		 * and will cause the buffer cache to attempt to free the
472 		 * underlying pages.
473 		 */
474 		if (ioflag & IO_DIRECT)
475 			bp->b_flags |= B_DIRECT;
476 
477 		/*
478 		 * We should only get non-zero b_resid when an I/O error
479 		 * has occurred, which should cause us to break above.
480 		 * However, if the short read did not cause an error,
481 		 * then we want to ensure that we do not uiomove bad
482 		 * or uninitialized data.
483 		 */
484 		size -= bp->b_resid;
485 		if (size < xfersize) {
486 			if (size == 0)
487 				break;
488 			xfersize = size;
489 		}
490 
491 		error = uiomove((char *)bp->b_data + blkoffset,
492 		    (int)xfersize, uio);
493 		if (error)
494 			break;
495 
496 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
497 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
498 			/*
499 			 * If there are no dependencies, and it's VMIO,
500 			 * then we don't need the buf, mark it available
501 			 * for freeing. The VM has the data.
502 			 */
503 			bp->b_flags |= B_RELBUF;
504 			brelse(bp);
505 		} else {
506 			/*
507 			 * Otherwise let whoever
508 			 * made the request take care of
509 			 * freeing it. We just queue
510 			 * it onto another list.
511 			 */
512 			bqrelse(bp);
513 		}
514 	}
515 
516 	/*
517 	 * This can only happen in the case of an error
518 	 * because the loop above resets bp to NULL on each iteration
519 	 * and on normal completion has not set a new value into it.
520 	 * so it must have come from a 'break' statement
521 	 */
522 	if (bp != NULL) {
523 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
524 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
525 			bp->b_flags |= B_RELBUF;
526 			brelse(bp);
527 		} else {
528 			bqrelse(bp);
529 		}
530 	}
531 
532 	if ((error == 0 || uio->uio_resid != orig_resid) &&
533 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
534 		ip->i_flag |= IN_ACCESS;
535 	return (error);
536 }
537 
538 /*
539  * Vnode op for writing.
540  */
541 static int
542 ffs_write(ap)
543 	struct vop_write_args /* {
544 		struct vnode *a_vp;
545 		struct uio *a_uio;
546 		int a_ioflag;
547 		struct ucred *a_cred;
548 	} */ *ap;
549 {
550 	struct vnode *vp;
551 	struct uio *uio;
552 	struct inode *ip;
553 	struct fs *fs;
554 	struct buf *bp;
555 	struct thread *td;
556 	ufs_lbn_t lbn;
557 	off_t osize;
558 	int seqcount;
559 	int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
560 
561 	vp = ap->a_vp;
562 	uio = ap->a_uio;
563 	ioflag = ap->a_ioflag;
564 	if (ap->a_ioflag & IO_EXT)
565 #ifdef notyet
566 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
567 #else
568 		panic("ffs_write+IO_EXT");
569 #endif
570 
571 	GIANT_REQUIRED;
572 
573 	extended = 0;
574 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
575 	ip = VTOI(vp);
576 
577 #ifdef DIAGNOSTIC
578 	if (uio->uio_rw != UIO_WRITE)
579 		panic("ffs_write: mode");
580 #endif
581 
582 	switch (vp->v_type) {
583 	case VREG:
584 		if (ioflag & IO_APPEND)
585 			uio->uio_offset = ip->i_size;
586 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
587 			return (EPERM);
588 		/* FALLTHROUGH */
589 	case VLNK:
590 		break;
591 	case VDIR:
592 		panic("ffs_write: dir write");
593 		break;
594 	default:
595 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
596 			(int)uio->uio_offset,
597 			(int)uio->uio_resid
598 		);
599 	}
600 
601 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
602 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
603 	fs = ip->i_fs;
604 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
605 		return (EFBIG);
606 	/*
607 	 * Maybe this should be above the vnode op call, but so long as
608 	 * file servers have no limits, I don't think it matters.
609 	 */
610 	td = uio->uio_td;
611 	if (vp->v_type == VREG && td != NULL) {
612 		PROC_LOCK(td->td_proc);
613 		if (uio->uio_offset + uio->uio_resid >
614 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
615 			psignal(td->td_proc, SIGXFSZ);
616 			PROC_UNLOCK(td->td_proc);
617 			return (EFBIG);
618 		}
619 		PROC_UNLOCK(td->td_proc);
620 	}
621 
622 	resid = uio->uio_resid;
623 	osize = ip->i_size;
624 	if (seqcount > BA_SEQMAX)
625 		flags = BA_SEQMAX << BA_SEQSHIFT;
626 	else
627 		flags = seqcount << BA_SEQSHIFT;
628 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
629 		flags |= IO_SYNC;
630 
631 	for (error = 0; uio->uio_resid > 0;) {
632 		lbn = lblkno(fs, uio->uio_offset);
633 		blkoffset = blkoff(fs, uio->uio_offset);
634 		xfersize = fs->fs_bsize - blkoffset;
635 		if (uio->uio_resid < xfersize)
636 			xfersize = uio->uio_resid;
637 		if (uio->uio_offset + xfersize > ip->i_size)
638 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
639 
640                 /*
641 		 * We must perform a read-before-write if the transfer size
642 		 * does not cover the entire buffer.
643                  */
644 		if (fs->fs_bsize > xfersize)
645 			flags |= BA_CLRBUF;
646 		else
647 			flags &= ~BA_CLRBUF;
648 /* XXX is uio->uio_offset the right thing here? */
649 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
650 		    ap->a_cred, flags, &bp);
651 		if (error != 0)
652 			break;
653 		/*
654 		 * If the buffer is not valid we have to clear out any
655 		 * garbage data from the pages instantiated for the buffer.
656 		 * If we do not, a failed uiomove() during a write can leave
657 		 * the prior contents of the pages exposed to a userland
658 		 * mmap().  XXX deal with uiomove() errors a better way.
659 		 */
660 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
661 			vfs_bio_clrbuf(bp);
662 		if (ioflag & IO_DIRECT)
663 			bp->b_flags |= B_DIRECT;
664 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
665 			bp->b_flags |= B_NOCACHE;
666 
667 		if (uio->uio_offset + xfersize > ip->i_size) {
668 			ip->i_size = uio->uio_offset + xfersize;
669 			DIP_SET(ip, i_size, ip->i_size);
670 			extended = 1;
671 		}
672 
673 		size = blksize(fs, ip, lbn) - bp->b_resid;
674 		if (size < xfersize)
675 			xfersize = size;
676 
677 		error =
678 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
679 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
680 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
681 			bp->b_flags |= B_RELBUF;
682 		}
683 
684 		/*
685 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
686 		 * if we have a severe page deficiency write the buffer
687 		 * asynchronously.  Otherwise try to cluster, and if that
688 		 * doesn't do it then either do an async write (if O_DIRECT),
689 		 * or a delayed write (if not).
690 		 */
691 		if (ioflag & IO_SYNC) {
692 			(void)bwrite(bp);
693 		} else if (vm_page_count_severe() ||
694 			    buf_dirty_count_severe() ||
695 			    (ioflag & IO_ASYNC)) {
696 			bp->b_flags |= B_CLUSTEROK;
697 			bawrite(bp);
698 		} else if (xfersize + blkoffset == fs->fs_bsize) {
699 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
700 				bp->b_flags |= B_CLUSTEROK;
701 				cluster_write(vp, bp, ip->i_size, seqcount);
702 			} else {
703 				bawrite(bp);
704 			}
705 		} else if (ioflag & IO_DIRECT) {
706 			bp->b_flags |= B_CLUSTEROK;
707 			bawrite(bp);
708 		} else {
709 			bp->b_flags |= B_CLUSTEROK;
710 			bdwrite(bp);
711 		}
712 		if (error || xfersize == 0)
713 			break;
714 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
715 	}
716 	/*
717 	 * If we successfully wrote any data, and we are not the superuser
718 	 * we clear the setuid and setgid bits as a precaution against
719 	 * tampering.
720 	 */
721 	if (resid > uio->uio_resid && ap->a_cred &&
722 	    suser_cred(ap->a_cred, SUSER_ALLOWJAIL)) {
723 		ip->i_mode &= ~(ISUID | ISGID);
724 		DIP_SET(ip, i_mode, ip->i_mode);
725 	}
726 	if (resid > uio->uio_resid)
727 		VN_KNOTE_UNLOCKED(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
728 	if (error) {
729 		if (ioflag & IO_UNIT) {
730 			(void)UFS_TRUNCATE(vp, osize,
731 			    IO_NORMAL | (ioflag & IO_SYNC),
732 			    ap->a_cred, uio->uio_td);
733 			uio->uio_offset -= resid - uio->uio_resid;
734 			uio->uio_resid = resid;
735 		}
736 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
737 		error = UFS_UPDATE(vp, 1);
738 	return (error);
739 }
740 
741 /*
742  * get page routine
743  */
744 static int
745 ffs_getpages(ap)
746 	struct vop_getpages_args *ap;
747 {
748 	int i;
749 	vm_page_t mreq;
750 	int pcount;
751 
752 	GIANT_REQUIRED;
753 
754 	pcount = round_page(ap->a_count) / PAGE_SIZE;
755 	mreq = ap->a_m[ap->a_reqpage];
756 
757 	/*
758 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
759 	 * then the entire page is valid.  Since the page may be mapped,
760 	 * user programs might reference data beyond the actual end of file
761 	 * occuring within the page.  We have to zero that data.
762 	 */
763 	VM_OBJECT_LOCK(mreq->object);
764 	if (mreq->valid) {
765 		if (mreq->valid != VM_PAGE_BITS_ALL)
766 			vm_page_zero_invalid(mreq, TRUE);
767 		vm_page_lock_queues();
768 		for (i = 0; i < pcount; i++) {
769 			if (i != ap->a_reqpage) {
770 				vm_page_free(ap->a_m[i]);
771 			}
772 		}
773 		vm_page_unlock_queues();
774 		VM_OBJECT_UNLOCK(mreq->object);
775 		return VM_PAGER_OK;
776 	}
777 	VM_OBJECT_UNLOCK(mreq->object);
778 
779 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
780 					    ap->a_count,
781 					    ap->a_reqpage);
782 }
783 
784 
785 /*
786  * Extended attribute area reading.
787  */
788 static int
789 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
790 {
791 	struct inode *ip;
792 	struct ufs2_dinode *dp;
793 	struct fs *fs;
794 	struct buf *bp;
795 	ufs_lbn_t lbn, nextlbn;
796 	off_t bytesinfile;
797 	long size, xfersize, blkoffset;
798 	int error, orig_resid;
799 
800 	GIANT_REQUIRED;
801 
802 	ip = VTOI(vp);
803 	fs = ip->i_fs;
804 	dp = ip->i_din2;
805 
806 #ifdef DIAGNOSTIC
807 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
808 		panic("ffs_extread: mode");
809 
810 #endif
811 	orig_resid = uio->uio_resid;
812 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
813 	if (orig_resid == 0)
814 		return (0);
815 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
816 
817 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
818 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
819 			break;
820 		lbn = lblkno(fs, uio->uio_offset);
821 		nextlbn = lbn + 1;
822 
823 		/*
824 		 * size of buffer.  The buffer representing the
825 		 * end of the file is rounded up to the size of
826 		 * the block type ( fragment or full block,
827 		 * depending ).
828 		 */
829 		size = sblksize(fs, dp->di_extsize, lbn);
830 		blkoffset = blkoff(fs, uio->uio_offset);
831 
832 		/*
833 		 * The amount we want to transfer in this iteration is
834 		 * one FS block less the amount of the data before
835 		 * our startpoint (duh!)
836 		 */
837 		xfersize = fs->fs_bsize - blkoffset;
838 
839 		/*
840 		 * But if we actually want less than the block,
841 		 * or the file doesn't have a whole block more of data,
842 		 * then use the lesser number.
843 		 */
844 		if (uio->uio_resid < xfersize)
845 			xfersize = uio->uio_resid;
846 		if (bytesinfile < xfersize)
847 			xfersize = bytesinfile;
848 
849 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
850 			/*
851 			 * Don't do readahead if this is the end of the info.
852 			 */
853 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
854 		} else {
855 			/*
856 			 * If we have a second block, then
857 			 * fire off a request for a readahead
858 			 * as well as a read. Note that the 4th and 5th
859 			 * arguments point to arrays of the size specified in
860 			 * the 6th argument.
861 			 */
862 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
863 
864 			nextlbn = -1 - nextlbn;
865 			error = breadn(vp, -1 - lbn,
866 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
867 		}
868 		if (error) {
869 			brelse(bp);
870 			bp = NULL;
871 			break;
872 		}
873 
874 		/*
875 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
876 		 * will cause us to attempt to release the buffer later on
877 		 * and will cause the buffer cache to attempt to free the
878 		 * underlying pages.
879 		 */
880 		if (ioflag & IO_DIRECT)
881 			bp->b_flags |= B_DIRECT;
882 
883 		/*
884 		 * We should only get non-zero b_resid when an I/O error
885 		 * has occurred, which should cause us to break above.
886 		 * However, if the short read did not cause an error,
887 		 * then we want to ensure that we do not uiomove bad
888 		 * or uninitialized data.
889 		 */
890 		size -= bp->b_resid;
891 		if (size < xfersize) {
892 			if (size == 0)
893 				break;
894 			xfersize = size;
895 		}
896 
897 		error = uiomove((char *)bp->b_data + blkoffset,
898 					(int)xfersize, uio);
899 		if (error)
900 			break;
901 
902 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
903 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
904 			/*
905 			 * If there are no dependencies, and it's VMIO,
906 			 * then we don't need the buf, mark it available
907 			 * for freeing. The VM has the data.
908 			 */
909 			bp->b_flags |= B_RELBUF;
910 			brelse(bp);
911 		} else {
912 			/*
913 			 * Otherwise let whoever
914 			 * made the request take care of
915 			 * freeing it. We just queue
916 			 * it onto another list.
917 			 */
918 			bqrelse(bp);
919 		}
920 	}
921 
922 	/*
923 	 * This can only happen in the case of an error
924 	 * because the loop above resets bp to NULL on each iteration
925 	 * and on normal completion has not set a new value into it.
926 	 * so it must have come from a 'break' statement
927 	 */
928 	if (bp != NULL) {
929 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
930 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
931 			bp->b_flags |= B_RELBUF;
932 			brelse(bp);
933 		} else {
934 			bqrelse(bp);
935 		}
936 	}
937 
938 	if ((error == 0 || uio->uio_resid != orig_resid) &&
939 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
940 		ip->i_flag |= IN_ACCESS;
941 	return (error);
942 }
943 
944 /*
945  * Extended attribute area writing.
946  */
947 static int
948 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
949 {
950 	struct inode *ip;
951 	struct ufs2_dinode *dp;
952 	struct fs *fs;
953 	struct buf *bp;
954 	ufs_lbn_t lbn;
955 	off_t osize;
956 	int blkoffset, error, flags, resid, size, xfersize;
957 
958 	GIANT_REQUIRED;
959 
960 	ip = VTOI(vp);
961 	fs = ip->i_fs;
962 	dp = ip->i_din2;
963 
964 #ifdef DIAGNOSTIC
965 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
966 		panic("ffs_extwrite: mode");
967 #endif
968 
969 	if (ioflag & IO_APPEND)
970 		uio->uio_offset = dp->di_extsize;
971 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
972 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
973 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
974 		return (EFBIG);
975 
976 	resid = uio->uio_resid;
977 	osize = dp->di_extsize;
978 	flags = IO_EXT;
979 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
980 		flags |= IO_SYNC;
981 
982 	for (error = 0; uio->uio_resid > 0;) {
983 		lbn = lblkno(fs, uio->uio_offset);
984 		blkoffset = blkoff(fs, uio->uio_offset);
985 		xfersize = fs->fs_bsize - blkoffset;
986 		if (uio->uio_resid < xfersize)
987 			xfersize = uio->uio_resid;
988 
989                 /*
990 		 * We must perform a read-before-write if the transfer size
991 		 * does not cover the entire buffer.
992                  */
993 		if (fs->fs_bsize > xfersize)
994 			flags |= BA_CLRBUF;
995 		else
996 			flags &= ~BA_CLRBUF;
997 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
998 		    ucred, flags, &bp);
999 		if (error != 0)
1000 			break;
1001 		/*
1002 		 * If the buffer is not valid we have to clear out any
1003 		 * garbage data from the pages instantiated for the buffer.
1004 		 * If we do not, a failed uiomove() during a write can leave
1005 		 * the prior contents of the pages exposed to a userland
1006 		 * mmap().  XXX deal with uiomove() errors a better way.
1007 		 */
1008 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1009 			vfs_bio_clrbuf(bp);
1010 		if (ioflag & IO_DIRECT)
1011 			bp->b_flags |= B_DIRECT;
1012 
1013 		if (uio->uio_offset + xfersize > dp->di_extsize)
1014 			dp->di_extsize = uio->uio_offset + xfersize;
1015 
1016 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1017 		if (size < xfersize)
1018 			xfersize = size;
1019 
1020 		error =
1021 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1022 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1023 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1024 			bp->b_flags |= B_RELBUF;
1025 		}
1026 
1027 		/*
1028 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1029 		 * if we have a severe page deficiency write the buffer
1030 		 * asynchronously.  Otherwise try to cluster, and if that
1031 		 * doesn't do it then either do an async write (if O_DIRECT),
1032 		 * or a delayed write (if not).
1033 		 */
1034 		if (ioflag & IO_SYNC) {
1035 			(void)bwrite(bp);
1036 		} else if (vm_page_count_severe() ||
1037 			    buf_dirty_count_severe() ||
1038 			    xfersize + blkoffset == fs->fs_bsize ||
1039 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1040 			bawrite(bp);
1041 		else
1042 			bdwrite(bp);
1043 		if (error || xfersize == 0)
1044 			break;
1045 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1046 	}
1047 	/*
1048 	 * If we successfully wrote any data, and we are not the superuser
1049 	 * we clear the setuid and setgid bits as a precaution against
1050 	 * tampering.
1051 	 */
1052 	if (resid > uio->uio_resid && ucred &&
1053 	    suser_cred(ucred, SUSER_ALLOWJAIL)) {
1054 		ip->i_mode &= ~(ISUID | ISGID);
1055 		dp->di_mode = ip->i_mode;
1056 	}
1057 	if (error) {
1058 		if (ioflag & IO_UNIT) {
1059 			(void)UFS_TRUNCATE(vp, osize,
1060 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1061 			uio->uio_offset -= resid - uio->uio_resid;
1062 			uio->uio_resid = resid;
1063 		}
1064 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1065 		error = UFS_UPDATE(vp, 1);
1066 	return (error);
1067 }
1068 
1069 
1070 /*
1071  * Vnode operating to retrieve a named extended attribute.
1072  *
1073  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1074  * the length of the EA, and possibly the pointer to the entry and to the data.
1075  */
1076 static int
1077 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1078 {
1079 	u_char *p, *pe, *pn, *p0;
1080 	int eapad1, eapad2, ealength, ealen, nlen;
1081 	uint32_t ul;
1082 
1083 	pe = ptr + length;
1084 	nlen = strlen(name);
1085 
1086 	for (p = ptr; p < pe; p = pn) {
1087 		p0 = p;
1088 		bcopy(p, &ul, sizeof(ul));
1089 		pn = p + ul;
1090 		/* make sure this entry is complete */
1091 		if (pn > pe)
1092 			break;
1093 		p += sizeof(uint32_t);
1094 		if (*p != nspace)
1095 			continue;
1096 		p++;
1097 		eapad2 = *p++;
1098 		if (*p != nlen)
1099 			continue;
1100 		p++;
1101 		if (bcmp(p, name, nlen))
1102 			continue;
1103 		ealength = sizeof(uint32_t) + 3 + nlen;
1104 		eapad1 = 8 - (ealength % 8);
1105 		if (eapad1 == 8)
1106 			eapad1 = 0;
1107 		ealength += eapad1;
1108 		ealen = ul - ealength - eapad2;
1109 		p += nlen + eapad1;
1110 		if (eap != NULL)
1111 			*eap = p0;
1112 		if (eac != NULL)
1113 			*eac = p;
1114 		return (ealen);
1115 	}
1116 	return(-1);
1117 }
1118 
1119 static int
1120 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1121 {
1122 	struct inode *ip;
1123 	struct ufs2_dinode *dp;
1124 	struct uio luio;
1125 	struct iovec liovec;
1126 	int easize, error;
1127 	u_char *eae;
1128 
1129 	ip = VTOI(vp);
1130 	dp = ip->i_din2;
1131 	easize = dp->di_extsize;
1132 
1133 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1134 
1135 	liovec.iov_base = eae;
1136 	liovec.iov_len = easize;
1137 	luio.uio_iov = &liovec;
1138 	luio.uio_iovcnt = 1;
1139 	luio.uio_offset = 0;
1140 	luio.uio_resid = easize;
1141 	luio.uio_segflg = UIO_SYSSPACE;
1142 	luio.uio_rw = UIO_READ;
1143 	luio.uio_td = td;
1144 
1145 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1146 	if (error) {
1147 		free(eae, M_TEMP);
1148 		return(error);
1149 	}
1150 	*p = eae;
1151 	return (0);
1152 }
1153 
1154 static int
1155 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1156 {
1157 	struct inode *ip;
1158 	struct ufs2_dinode *dp;
1159 	int error;
1160 
1161 	ip = VTOI(vp);
1162 
1163 	if (ip->i_ea_area != NULL)
1164 		return (EBUSY);
1165 	dp = ip->i_din2;
1166 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1167 	if (error)
1168 		return (error);
1169 	ip->i_ea_len = dp->di_extsize;
1170 	ip->i_ea_error = 0;
1171 	return (0);
1172 }
1173 
1174 /*
1175  * Vnode extattr transaction commit/abort
1176  */
1177 static int
1178 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1179 {
1180 	struct inode *ip;
1181 	struct uio luio;
1182 	struct iovec liovec;
1183 	int error;
1184 	struct ufs2_dinode *dp;
1185 
1186 	ip = VTOI(vp);
1187 	if (ip->i_ea_area == NULL)
1188 		return (EINVAL);
1189 	dp = ip->i_din2;
1190 	error = ip->i_ea_error;
1191 	if (commit && error == 0) {
1192 		if (cred == NOCRED)
1193 			cred =  vp->v_mount->mnt_cred;
1194 		liovec.iov_base = ip->i_ea_area;
1195 		liovec.iov_len = ip->i_ea_len;
1196 		luio.uio_iov = &liovec;
1197 		luio.uio_iovcnt = 1;
1198 		luio.uio_offset = 0;
1199 		luio.uio_resid = ip->i_ea_len;
1200 		luio.uio_segflg = UIO_SYSSPACE;
1201 		luio.uio_rw = UIO_WRITE;
1202 		luio.uio_td = td;
1203 		/* XXX: I'm not happy about truncating to zero size */
1204 		if (ip->i_ea_len < dp->di_extsize)
1205 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1206 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1207 	}
1208 	free(ip->i_ea_area, M_TEMP);
1209 	ip->i_ea_area = NULL;
1210 	ip->i_ea_len = 0;
1211 	ip->i_ea_error = 0;
1212 	return (error);
1213 }
1214 
1215 /*
1216  * Vnode extattr strategy routine for special devices and fifos.
1217  *
1218  * We need to check for a read or write of the external attributes.
1219  * Otherwise we just fall through and do the usual thing.
1220  */
1221 static int
1222 ffsext_strategy(struct vop_strategy_args *ap)
1223 /*
1224 struct vop_strategy_args {
1225 	struct vnodeop_desc *a_desc;
1226 	struct vnode *a_vp;
1227 	struct buf *a_bp;
1228 };
1229 */
1230 {
1231 	struct vnode *vp;
1232 	daddr_t lbn;
1233 
1234 	KASSERT(ap->a_vp == ap->a_bp->b_vp, ("%s(%p != %p)",
1235 	    __func__, ap->a_vp, ap->a_bp->b_vp));
1236 	vp = ap->a_vp;
1237 	lbn = ap->a_bp->b_lblkno;
1238 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1239 	    lbn < 0 && lbn >= -NXADDR)
1240 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1241 	if (vp->v_type == VFIFO)
1242 		return (ufs_vnoperatefifo((struct vop_generic_args *)ap));
1243 	panic("spec nodes went here");
1244 }
1245 
1246 /*
1247  * Vnode extattr transaction commit/abort
1248  */
1249 static int
1250 ffs_openextattr(struct vop_openextattr_args *ap)
1251 /*
1252 struct vop_openextattr_args {
1253 	struct vnodeop_desc *a_desc;
1254 	struct vnode *a_vp;
1255 	IN struct ucred *a_cred;
1256 	IN struct thread *a_td;
1257 };
1258 */
1259 {
1260 	struct inode *ip;
1261 	struct fs *fs;
1262 
1263 	ip = VTOI(ap->a_vp);
1264 	fs = ip->i_fs;
1265 	if (fs->fs_magic == FS_UFS1_MAGIC)
1266 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1267 
1268 	if (ap->a_vp->v_type == VCHR)
1269 		return (EOPNOTSUPP);
1270 
1271 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1272 }
1273 
1274 
1275 /*
1276  * Vnode extattr transaction commit/abort
1277  */
1278 static int
1279 ffs_closeextattr(struct vop_closeextattr_args *ap)
1280 /*
1281 struct vop_closeextattr_args {
1282 	struct vnodeop_desc *a_desc;
1283 	struct vnode *a_vp;
1284 	int a_commit;
1285 	IN struct ucred *a_cred;
1286 	IN struct thread *a_td;
1287 };
1288 */
1289 {
1290 	struct inode *ip;
1291 	struct fs *fs;
1292 
1293 	ip = VTOI(ap->a_vp);
1294 	fs = ip->i_fs;
1295 	if (fs->fs_magic == FS_UFS1_MAGIC)
1296 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1297 
1298 	if (ap->a_vp->v_type == VCHR)
1299 		return (EOPNOTSUPP);
1300 
1301 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1302 }
1303 
1304 /*
1305  * Vnode operation to remove a named attribute.
1306  */
1307 static int
1308 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1309 /*
1310 vop_deleteextattr {
1311 	IN struct vnode *a_vp;
1312 	IN int a_attrnamespace;
1313 	IN const char *a_name;
1314 	IN struct ucred *a_cred;
1315 	IN struct thread *a_td;
1316 };
1317 */
1318 {
1319 	struct inode *ip;
1320 	struct fs *fs;
1321 	uint32_t ealength, ul;
1322 	int ealen, olen, eapad1, eapad2, error, i, easize;
1323 	u_char *eae, *p;
1324 	int stand_alone;
1325 
1326 	ip = VTOI(ap->a_vp);
1327 	fs = ip->i_fs;
1328 
1329 	if (fs->fs_magic == FS_UFS1_MAGIC)
1330 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1331 
1332 	if (ap->a_vp->v_type == VCHR)
1333 		return (EOPNOTSUPP);
1334 
1335 	if (strlen(ap->a_name) == 0)
1336 		return (EINVAL);
1337 
1338 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1339 	    ap->a_cred, ap->a_td, IWRITE);
1340 	if (error) {
1341 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1342 			ip->i_ea_error = error;
1343 		return (error);
1344 	}
1345 
1346 	if (ip->i_ea_area == NULL) {
1347 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1348 		if (error)
1349 			return (error);
1350 		stand_alone = 1;
1351 	} else {
1352 		stand_alone = 0;
1353 	}
1354 
1355 	ealength = eapad1 = ealen = eapad2 = 0;
1356 
1357 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1358 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1359 	easize = ip->i_ea_len;
1360 
1361 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1362 	    &p, NULL);
1363 	if (olen == -1) {
1364 		/* delete but nonexistent */
1365 		free(eae, M_TEMP);
1366 		if (stand_alone)
1367 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1368 		return(ENOATTR);
1369 	}
1370 	bcopy(p, &ul, sizeof ul);
1371 	i = p - eae + ul;
1372 	if (ul != ealength) {
1373 		bcopy(p + ul, p + ealength, easize - i);
1374 		easize += (ealength - ul);
1375 	}
1376 	if (easize > NXADDR * fs->fs_bsize) {
1377 		free(eae, M_TEMP);
1378 		if (stand_alone)
1379 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1380 		else if (ip->i_ea_error == 0)
1381 			ip->i_ea_error = ENOSPC;
1382 		return(ENOSPC);
1383 	}
1384 	p = ip->i_ea_area;
1385 	ip->i_ea_area = eae;
1386 	ip->i_ea_len = easize;
1387 	free(p, M_TEMP);
1388 	if (stand_alone)
1389 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1390 	return(error);
1391 }
1392 
1393 /*
1394  * Vnode operation to retrieve a named extended attribute.
1395  */
1396 static int
1397 ffs_getextattr(struct vop_getextattr_args *ap)
1398 /*
1399 vop_getextattr {
1400 	IN struct vnode *a_vp;
1401 	IN int a_attrnamespace;
1402 	IN const char *a_name;
1403 	INOUT struct uio *a_uio;
1404 	OUT size_t *a_size;
1405 	IN struct ucred *a_cred;
1406 	IN struct thread *a_td;
1407 };
1408 */
1409 {
1410 	struct inode *ip;
1411 	struct fs *fs;
1412 	u_char *eae, *p;
1413 	unsigned easize;
1414 	int error, ealen, stand_alone;
1415 
1416 	ip = VTOI(ap->a_vp);
1417 	fs = ip->i_fs;
1418 
1419 	if (fs->fs_magic == FS_UFS1_MAGIC)
1420 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1421 
1422 	if (ap->a_vp->v_type == VCHR)
1423 		return (EOPNOTSUPP);
1424 
1425 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1426 	    ap->a_cred, ap->a_td, IREAD);
1427 	if (error)
1428 		return (error);
1429 
1430 	if (ip->i_ea_area == NULL) {
1431 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1432 		if (error)
1433 			return (error);
1434 		stand_alone = 1;
1435 	} else {
1436 		stand_alone = 0;
1437 	}
1438 	eae = ip->i_ea_area;
1439 	easize = ip->i_ea_len;
1440 
1441 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1442 	    NULL, &p);
1443 	if (ealen >= 0) {
1444 		error = 0;
1445 		if (ap->a_size != NULL)
1446 			*ap->a_size = ealen;
1447 		else if (ap->a_uio != NULL)
1448 			error = uiomove(p, ealen, ap->a_uio);
1449 	} else
1450 		error = ENOATTR;
1451 	if (stand_alone)
1452 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1453 	return(error);
1454 }
1455 
1456 /*
1457  * Vnode operation to retrieve extended attributes on a vnode.
1458  */
1459 static int
1460 ffs_listextattr(struct vop_listextattr_args *ap)
1461 /*
1462 vop_listextattr {
1463 	IN struct vnode *a_vp;
1464 	IN int a_attrnamespace;
1465 	INOUT struct uio *a_uio;
1466 	OUT size_t *a_size;
1467 	IN struct ucred *a_cred;
1468 	IN struct thread *a_td;
1469 };
1470 */
1471 {
1472 	struct inode *ip;
1473 	struct fs *fs;
1474 	u_char *eae, *p, *pe, *pn;
1475 	unsigned easize;
1476 	uint32_t ul;
1477 	int error, ealen, stand_alone;
1478 
1479 	ip = VTOI(ap->a_vp);
1480 	fs = ip->i_fs;
1481 
1482 	if (fs->fs_magic == FS_UFS1_MAGIC)
1483 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1484 
1485 	if (ap->a_vp->v_type == VCHR)
1486 		return (EOPNOTSUPP);
1487 
1488 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1489 	    ap->a_cred, ap->a_td, IREAD);
1490 	if (error)
1491 		return (error);
1492 
1493 	if (ip->i_ea_area == NULL) {
1494 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1495 		if (error)
1496 			return (error);
1497 		stand_alone = 1;
1498 	} else {
1499 		stand_alone = 0;
1500 	}
1501 	eae = ip->i_ea_area;
1502 	easize = ip->i_ea_len;
1503 
1504 	error = 0;
1505 	if (ap->a_size != NULL)
1506 		*ap->a_size = 0;
1507 	pe = eae + easize;
1508 	for(p = eae; error == 0 && p < pe; p = pn) {
1509 		bcopy(p, &ul, sizeof(ul));
1510 		pn = p + ul;
1511 		if (pn > pe)
1512 			break;
1513 		p += sizeof(ul);
1514 		if (*p++ != ap->a_attrnamespace)
1515 			continue;
1516 		p++;	/* pad2 */
1517 		ealen = *p;
1518 		if (ap->a_size != NULL) {
1519 			*ap->a_size += ealen + 1;
1520 		} else if (ap->a_uio != NULL) {
1521 			error = uiomove(p, ealen + 1, ap->a_uio);
1522 		}
1523 	}
1524 	if (stand_alone)
1525 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1526 	return(error);
1527 }
1528 
1529 /*
1530  * Vnode operation to set a named attribute.
1531  */
1532 static int
1533 ffs_setextattr(struct vop_setextattr_args *ap)
1534 /*
1535 vop_setextattr {
1536 	IN struct vnode *a_vp;
1537 	IN int a_attrnamespace;
1538 	IN const char *a_name;
1539 	INOUT struct uio *a_uio;
1540 	IN struct ucred *a_cred;
1541 	IN struct thread *a_td;
1542 };
1543 */
1544 {
1545 	struct inode *ip;
1546 	struct fs *fs;
1547 	uint32_t ealength, ul;
1548 	int ealen, olen, eapad1, eapad2, error, i, easize;
1549 	u_char *eae, *p;
1550 	int stand_alone;
1551 
1552 	ip = VTOI(ap->a_vp);
1553 	fs = ip->i_fs;
1554 
1555 	if (fs->fs_magic == FS_UFS1_MAGIC)
1556 		return (ufs_vnoperate((struct vop_generic_args *)ap));
1557 
1558 	if (ap->a_vp->v_type == VCHR)
1559 		return (EOPNOTSUPP);
1560 
1561 	if (strlen(ap->a_name) == 0)
1562 		return (EINVAL);
1563 
1564 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1565 	if (ap->a_uio == NULL)
1566 		return (EOPNOTSUPP);
1567 
1568 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1569 	    ap->a_cred, ap->a_td, IWRITE);
1570 	if (error) {
1571 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1572 			ip->i_ea_error = error;
1573 		return (error);
1574 	}
1575 
1576 	if (ip->i_ea_area == NULL) {
1577 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1578 		if (error)
1579 			return (error);
1580 		stand_alone = 1;
1581 	} else {
1582 		stand_alone = 0;
1583 	}
1584 
1585 	ealen = ap->a_uio->uio_resid;
1586 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1587 	eapad1 = 8 - (ealength % 8);
1588 	if (eapad1 == 8)
1589 		eapad1 = 0;
1590 	eapad2 = 8 - (ealen % 8);
1591 	if (eapad2 == 8)
1592 		eapad2 = 0;
1593 	ealength += eapad1 + ealen + eapad2;
1594 
1595 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1596 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1597 	easize = ip->i_ea_len;
1598 
1599 	olen = ffs_findextattr(eae, easize,
1600 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1601         if (olen == -1) {
1602 		/* new, append at end */
1603 		p = eae + easize;
1604 		easize += ealength;
1605 	} else {
1606 		bcopy(p, &ul, sizeof ul);
1607 		i = p - eae + ul;
1608 		if (ul != ealength) {
1609 			bcopy(p + ul, p + ealength, easize - i);
1610 			easize += (ealength - ul);
1611 		}
1612 	}
1613 	if (easize > NXADDR * fs->fs_bsize) {
1614 		free(eae, M_TEMP);
1615 		if (stand_alone)
1616 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1617 		else if (ip->i_ea_error == 0)
1618 			ip->i_ea_error = ENOSPC;
1619 		return(ENOSPC);
1620 	}
1621 	bcopy(&ealength, p, sizeof(ealength));
1622 	p += sizeof(ealength);
1623 	*p++ = ap->a_attrnamespace;
1624 	*p++ = eapad2;
1625 	*p++ = strlen(ap->a_name);
1626 	strcpy(p, ap->a_name);
1627 	p += strlen(ap->a_name);
1628 	bzero(p, eapad1);
1629 	p += eapad1;
1630 	error = uiomove(p, ealen, ap->a_uio);
1631 	if (error) {
1632 		free(eae, M_TEMP);
1633 		if (stand_alone)
1634 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1635 		else if (ip->i_ea_error == 0)
1636 			ip->i_ea_error = error;
1637 		return(error);
1638 	}
1639 	p += ealen;
1640 	bzero(p, eapad2);
1641 
1642 	p = ip->i_ea_area;
1643 	ip->i_ea_area = eae;
1644 	ip->i_ea_len = easize;
1645 	free(p, M_TEMP);
1646 	if (stand_alone)
1647 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1648 	return(error);
1649 }
1650