xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
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 #include "opt_ffs.h"
101 
102 #ifdef DIRECTIO
103 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
104 #endif
105 static vop_fsync_t	ffs_fsync;
106 static vop_lock_t	ffs_lock;
107 static vop_getpages_t	ffs_getpages;
108 static vop_read_t	ffs_read;
109 static vop_write_t	ffs_write;
110 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
111 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
112 		    struct ucred *cred);
113 static vop_strategy_t	ffsext_strategy;
114 static vop_closeextattr_t	ffs_closeextattr;
115 static vop_deleteextattr_t	ffs_deleteextattr;
116 static vop_getextattr_t	ffs_getextattr;
117 static vop_listextattr_t	ffs_listextattr;
118 static vop_openextattr_t	ffs_openextattr;
119 static vop_setextattr_t	ffs_setextattr;
120 
121 
122 /* Global vfs data structures for ufs. */
123 struct vop_vector ffs_vnodeops1 = {
124 	.vop_default =		&ufs_vnodeops,
125 	.vop_fsync =		ffs_fsync,
126 	.vop_getpages =		ffs_getpages,
127 	.vop_lock =		ffs_lock,
128 	.vop_read =		ffs_read,
129 	.vop_reallocblks =	ffs_reallocblks,
130 	.vop_write =		ffs_write,
131 };
132 
133 struct vop_vector ffs_fifoops1 = {
134 	.vop_default =		&ufs_fifoops,
135 	.vop_fsync =		ffs_fsync,
136 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
137 };
138 
139 /* Global vfs data structures for ufs. */
140 struct vop_vector ffs_vnodeops2 = {
141 	.vop_default =		&ufs_vnodeops,
142 	.vop_fsync =		ffs_fsync,
143 	.vop_getpages =		ffs_getpages,
144 	.vop_lock =		ffs_lock,
145 	.vop_read =		ffs_read,
146 	.vop_reallocblks =	ffs_reallocblks,
147 	.vop_write =		ffs_write,
148 	.vop_closeextattr =	ffs_closeextattr,
149 	.vop_deleteextattr =	ffs_deleteextattr,
150 	.vop_getextattr =	ffs_getextattr,
151 	.vop_listextattr =	ffs_listextattr,
152 	.vop_openextattr =	ffs_openextattr,
153 	.vop_setextattr =	ffs_setextattr,
154 };
155 
156 struct vop_vector ffs_fifoops2 = {
157 	.vop_default =		&ufs_fifoops,
158 	.vop_fsync =		ffs_fsync,
159 	.vop_lock =		ffs_lock,
160 	.vop_reallocblks =	ffs_reallocblks,
161 	.vop_strategy =		ffsext_strategy,
162 	.vop_closeextattr =	ffs_closeextattr,
163 	.vop_deleteextattr =	ffs_deleteextattr,
164 	.vop_getextattr =	ffs_getextattr,
165 	.vop_listextattr =	ffs_listextattr,
166 	.vop_openextattr =	ffs_openextattr,
167 	.vop_setextattr =	ffs_setextattr,
168 };
169 
170 /*
171  * Synch an open file.
172  */
173 /* ARGSUSED */
174 static int
175 ffs_fsync(struct vop_fsync_args *ap)
176 {
177 	int error;
178 
179 	error = ffs_syncvnode(ap->a_vp, ap->a_waitfor);
180 	if (error)
181 		return (error);
182 	if (ap->a_waitfor == MNT_WAIT &&
183 	    (ap->a_vp->v_mount->mnt_flag & MNT_SOFTDEP))
184                 error = softdep_fsync(ap->a_vp);
185 	return (error);
186 }
187 
188 int
189 ffs_syncvnode(struct vnode *vp, int waitfor)
190 {
191 	struct inode *ip = VTOI(vp);
192 	struct buf *bp;
193 	struct buf *nbp;
194 	int s, error, wait, passes, skipmeta;
195 	ufs_lbn_t lbn;
196 
197 	wait = (waitfor == MNT_WAIT);
198 	lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
199 
200 	/*
201 	 * Flush all dirty buffers associated with a vnode.
202 	 */
203 	passes = NIADDR + 1;
204 	skipmeta = 0;
205 	if (wait)
206 		skipmeta = 1;
207 	s = splbio();
208 	VI_LOCK(vp);
209 loop:
210 	TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs)
211 		bp->b_vflags &= ~BV_SCANNED;
212 	TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
213 		/*
214 		 * Reasons to skip this buffer: it has already been considered
215 		 * on this pass, this pass is the first time through on a
216 		 * synchronous flush request and the buffer being considered
217 		 * is metadata, the buffer has dependencies that will cause
218 		 * it to be redirtied and it has not already been deferred,
219 		 * or it is already being written.
220 		 */
221 		if ((bp->b_vflags & BV_SCANNED) != 0)
222 			continue;
223 		bp->b_vflags |= BV_SCANNED;
224 		if ((skipmeta == 1 && bp->b_lblkno < 0))
225 			continue;
226 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
227 			continue;
228 		VI_UNLOCK(vp);
229 		if (!wait && LIST_FIRST(&bp->b_dep) != NULL &&
230 		    (bp->b_flags & B_DEFERRED) == 0 &&
231 		    buf_countdeps(bp, 0)) {
232 			bp->b_flags |= B_DEFERRED;
233 			BUF_UNLOCK(bp);
234 			VI_LOCK(vp);
235 			continue;
236 		}
237 		if ((bp->b_flags & B_DELWRI) == 0)
238 			panic("ffs_fsync: not dirty");
239 		/*
240 		 * If this is a synchronous flush request, or it is not a
241 		 * file or device, start the write on this buffer immediatly.
242 		 */
243 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
244 
245 			/*
246 			 * On our final pass through, do all I/O synchronously
247 			 * so that we can find out if our flush is failing
248 			 * because of write errors.
249 			 */
250 			if (passes > 0 || !wait) {
251 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
252 					(void) vfs_bio_awrite(bp);
253 				} else {
254 					bremfree(bp);
255 					splx(s);
256 					(void) bawrite(bp);
257 					s = splbio();
258 				}
259 			} else {
260 				bremfree(bp);
261 				splx(s);
262 				if ((error = bwrite(bp)) != 0)
263 					return (error);
264 				s = splbio();
265 			}
266 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
267 			/*
268 			 * If the buffer is for data that has been truncated
269 			 * off the file, then throw it away.
270 			 */
271 			bremfree(bp);
272 			bp->b_flags |= B_INVAL | B_NOCACHE;
273 			splx(s);
274 			brelse(bp);
275 			s = splbio();
276 		} else
277 			vfs_bio_awrite(bp);
278 
279 		/*
280 		 * Since we may have slept during the I/O, we need
281 		 * to start from a known point.
282 		 */
283 		VI_LOCK(vp);
284 		nbp = TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd);
285 	}
286 	/*
287 	 * If we were asked to do this synchronously, then go back for
288 	 * another pass, this time doing the metadata.
289 	 */
290 	if (skipmeta) {
291 		skipmeta = 0;
292 		goto loop;
293 	}
294 
295 	if (wait) {
296 		bufobj_wwait(&vp->v_bufobj, 3, 0);
297 		VI_UNLOCK(vp);
298 
299 		/*
300 		 * Ensure that any filesystem metatdata associated
301 		 * with the vnode has been written.
302 		 */
303 		splx(s);
304 		if ((error = softdep_sync_metadata(vp)) != 0)
305 			return (error);
306 		s = splbio();
307 
308 		VI_LOCK(vp);
309 		if (vp->v_bufobj.bo_dirty.bv_cnt > 0) {
310 			/*
311 			 * Block devices associated with filesystems may
312 			 * have new I/O requests posted for them even if
313 			 * the vnode is locked, so no amount of trying will
314 			 * get them clean. Thus we give block devices a
315 			 * good effort, then just give up. For all other file
316 			 * types, go around and try again until it is clean.
317 			 */
318 			if (passes > 0) {
319 				passes -= 1;
320 				goto loop;
321 			}
322 #ifdef DIAGNOSTIC
323 			if (!vn_isdisk(vp, NULL))
324 				vprint("ffs_fsync: dirty", vp);
325 #endif
326 		}
327 	}
328 	VI_UNLOCK(vp);
329 	splx(s);
330 	return (ffs_update(vp, wait));
331 }
332 
333 static int
334 ffs_lock(ap)
335 	struct vop_lock_args /* {
336 		struct vnode *a_vp;
337 		int a_flags;
338 		struct thread *a_td;
339 	} */ *ap;
340 {
341 #ifndef NO_FFS_SNAPSHOT
342 	struct vnode *vp;
343 	int flags;
344 	struct lock *lkp;
345 	int result;
346 
347 	switch (ap->a_flags & LK_TYPE_MASK) {
348 	case LK_SHARED:
349 	case LK_UPGRADE:
350 	case LK_EXCLUSIVE:
351 		vp = ap->a_vp;
352 		flags = ap->a_flags;
353 		for (;;) {
354 			/*
355 			 * vnode interlock must be held to ensure that
356 			 * the possibly external lock isn't freed,
357 			 * e.g. when mutating from snapshot file vnode
358 			 * to regular file vnode.
359 			 */
360 			if ((flags & LK_INTERLOCK) == 0) {
361 				VI_LOCK(vp);
362 				flags |= LK_INTERLOCK;
363 			}
364 			lkp = vp->v_vnlock;
365 			result = lockmgr(lkp, flags, VI_MTX(vp), ap->a_td);
366 			if (lkp == vp->v_vnlock || result != 0)
367 				break;
368 			/*
369 			 * Apparent success, except that the vnode
370 			 * mutated between snapshot file vnode and
371 			 * regular file vnode while this process
372 			 * slept.  The lock currently held is not the
373 			 * right lock.  Release it, and try to get the
374 			 * new lock.
375 			 */
376 			(void) lockmgr(lkp, LK_RELEASE, VI_MTX(vp), ap->a_td);
377 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
378 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
379 			flags &= ~LK_INTERLOCK;
380 		}
381 		break;
382 	default:
383 		result = VOP_LOCK_APV(&ufs_vnodeops, ap);
384 	}
385 	return (result);
386 #else
387 	return (VOP_LOCK_APV(&ufs_vnodeops, ap));
388 #endif
389 }
390 
391 /*
392  * Vnode op for reading.
393  */
394 /* ARGSUSED */
395 static int
396 ffs_read(ap)
397 	struct vop_read_args /* {
398 		struct vnode *a_vp;
399 		struct uio *a_uio;
400 		int a_ioflag;
401 		struct ucred *a_cred;
402 	} */ *ap;
403 {
404 	struct vnode *vp;
405 	struct inode *ip;
406 	struct uio *uio;
407 	struct fs *fs;
408 	struct buf *bp;
409 	ufs_lbn_t lbn, nextlbn;
410 	off_t bytesinfile;
411 	long size, xfersize, blkoffset;
412 	int error, orig_resid;
413 	int seqcount;
414 	int ioflag;
415 
416 	vp = ap->a_vp;
417 	uio = ap->a_uio;
418 	ioflag = ap->a_ioflag;
419 	if (ap->a_ioflag & IO_EXT)
420 #ifdef notyet
421 		return (ffs_extread(vp, uio, ioflag));
422 #else
423 		panic("ffs_read+IO_EXT");
424 #endif
425 #ifdef DIRECTIO
426 	if ((ioflag & IO_DIRECT) != 0) {
427 		int workdone;
428 
429 		error = ffs_rawread(vp, uio, &workdone);
430 		if (error != 0 || workdone != 0)
431 			return error;
432 	}
433 #endif
434 
435 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
436 	ip = VTOI(vp);
437 
438 #ifdef DIAGNOSTIC
439 	if (uio->uio_rw != UIO_READ)
440 		panic("ffs_read: mode");
441 
442 	if (vp->v_type == VLNK) {
443 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
444 			panic("ffs_read: short symlink");
445 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
446 		panic("ffs_read: type %d",  vp->v_type);
447 #endif
448 	orig_resid = uio->uio_resid;
449 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
450 	if (orig_resid == 0)
451 		return (0);
452 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
453 	fs = ip->i_fs;
454 	if (uio->uio_offset < ip->i_size &&
455 	    uio->uio_offset >= fs->fs_maxfilesize)
456 		return (EOVERFLOW);
457 
458 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
459 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
460 			break;
461 		lbn = lblkno(fs, uio->uio_offset);
462 		nextlbn = lbn + 1;
463 
464 		/*
465 		 * size of buffer.  The buffer representing the
466 		 * end of the file is rounded up to the size of
467 		 * the block type ( fragment or full block,
468 		 * depending ).
469 		 */
470 		size = blksize(fs, ip, lbn);
471 		blkoffset = blkoff(fs, uio->uio_offset);
472 
473 		/*
474 		 * The amount we want to transfer in this iteration is
475 		 * one FS block less the amount of the data before
476 		 * our startpoint (duh!)
477 		 */
478 		xfersize = fs->fs_bsize - blkoffset;
479 
480 		/*
481 		 * But if we actually want less than the block,
482 		 * or the file doesn't have a whole block more of data,
483 		 * then use the lesser number.
484 		 */
485 		if (uio->uio_resid < xfersize)
486 			xfersize = uio->uio_resid;
487 		if (bytesinfile < xfersize)
488 			xfersize = bytesinfile;
489 
490 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
491 			/*
492 			 * Don't do readahead if this is the end of the file.
493 			 */
494 			error = bread(vp, lbn, size, NOCRED, &bp);
495 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
496 			/*
497 			 * Otherwise if we are allowed to cluster,
498 			 * grab as much as we can.
499 			 *
500 			 * XXX  This may not be a win if we are not
501 			 * doing sequential access.
502 			 */
503 			error = cluster_read(vp, ip->i_size, lbn,
504 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
505 		} else if (seqcount > 1) {
506 			/*
507 			 * If we are NOT allowed to cluster, then
508 			 * if we appear to be acting sequentially,
509 			 * fire off a request for a readahead
510 			 * as well as a read. Note that the 4th and 5th
511 			 * arguments point to arrays of the size specified in
512 			 * the 6th argument.
513 			 */
514 			int nextsize = blksize(fs, ip, nextlbn);
515 			error = breadn(vp, lbn,
516 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
517 		} else {
518 			/*
519 			 * Failing all of the above, just read what the
520 			 * user asked for. Interestingly, the same as
521 			 * the first option above.
522 			 */
523 			error = bread(vp, lbn, size, NOCRED, &bp);
524 		}
525 		if (error) {
526 			brelse(bp);
527 			bp = NULL;
528 			break;
529 		}
530 
531 		/*
532 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
533 		 * will cause us to attempt to release the buffer later on
534 		 * and will cause the buffer cache to attempt to free the
535 		 * underlying pages.
536 		 */
537 		if (ioflag & IO_DIRECT)
538 			bp->b_flags |= B_DIRECT;
539 
540 		/*
541 		 * We should only get non-zero b_resid when an I/O error
542 		 * has occurred, which should cause us to break above.
543 		 * However, if the short read did not cause an error,
544 		 * then we want to ensure that we do not uiomove bad
545 		 * or uninitialized data.
546 		 */
547 		size -= bp->b_resid;
548 		if (size < xfersize) {
549 			if (size == 0)
550 				break;
551 			xfersize = size;
552 		}
553 
554 		error = uiomove((char *)bp->b_data + blkoffset,
555 		    (int)xfersize, uio);
556 		if (error)
557 			break;
558 
559 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
560 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
561 			/*
562 			 * If there are no dependencies, and it's VMIO,
563 			 * then we don't need the buf, mark it available
564 			 * for freeing. The VM has the data.
565 			 */
566 			bp->b_flags |= B_RELBUF;
567 			brelse(bp);
568 		} else {
569 			/*
570 			 * Otherwise let whoever
571 			 * made the request take care of
572 			 * freeing it. We just queue
573 			 * it onto another list.
574 			 */
575 			bqrelse(bp);
576 		}
577 	}
578 
579 	/*
580 	 * This can only happen in the case of an error
581 	 * because the loop above resets bp to NULL on each iteration
582 	 * and on normal completion has not set a new value into it.
583 	 * so it must have come from a 'break' statement
584 	 */
585 	if (bp != NULL) {
586 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
587 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
588 			bp->b_flags |= B_RELBUF;
589 			brelse(bp);
590 		} else {
591 			bqrelse(bp);
592 		}
593 	}
594 
595 	if ((error == 0 || uio->uio_resid != orig_resid) &&
596 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
597 		ip->i_flag |= IN_ACCESS;
598 	return (error);
599 }
600 
601 /*
602  * Vnode op for writing.
603  */
604 static int
605 ffs_write(ap)
606 	struct vop_write_args /* {
607 		struct vnode *a_vp;
608 		struct uio *a_uio;
609 		int a_ioflag;
610 		struct ucred *a_cred;
611 	} */ *ap;
612 {
613 	struct vnode *vp;
614 	struct uio *uio;
615 	struct inode *ip;
616 	struct fs *fs;
617 	struct buf *bp;
618 	struct thread *td;
619 	ufs_lbn_t lbn;
620 	off_t osize;
621 	int seqcount;
622 	int blkoffset, error, flags, ioflag, resid, size, xfersize;
623 
624 	vp = ap->a_vp;
625 	uio = ap->a_uio;
626 	ioflag = ap->a_ioflag;
627 	if (ap->a_ioflag & IO_EXT)
628 #ifdef notyet
629 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
630 #else
631 		panic("ffs_write+IO_EXT");
632 #endif
633 
634 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
635 	ip = VTOI(vp);
636 
637 #ifdef DIAGNOSTIC
638 	if (uio->uio_rw != UIO_WRITE)
639 		panic("ffs_write: mode");
640 #endif
641 
642 	switch (vp->v_type) {
643 	case VREG:
644 		if (ioflag & IO_APPEND)
645 			uio->uio_offset = ip->i_size;
646 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
647 			return (EPERM);
648 		/* FALLTHROUGH */
649 	case VLNK:
650 		break;
651 	case VDIR:
652 		panic("ffs_write: dir write");
653 		break;
654 	default:
655 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
656 			(int)uio->uio_offset,
657 			(int)uio->uio_resid
658 		);
659 	}
660 
661 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
662 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
663 	fs = ip->i_fs;
664 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
665 		return (EFBIG);
666 	/*
667 	 * Maybe this should be above the vnode op call, but so long as
668 	 * file servers have no limits, I don't think it matters.
669 	 */
670 	td = uio->uio_td;
671 	if (vp->v_type == VREG && td != NULL) {
672 		PROC_LOCK(td->td_proc);
673 		if (uio->uio_offset + uio->uio_resid >
674 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
675 			psignal(td->td_proc, SIGXFSZ);
676 			PROC_UNLOCK(td->td_proc);
677 			return (EFBIG);
678 		}
679 		PROC_UNLOCK(td->td_proc);
680 	}
681 
682 	resid = uio->uio_resid;
683 	osize = ip->i_size;
684 	if (seqcount > BA_SEQMAX)
685 		flags = BA_SEQMAX << BA_SEQSHIFT;
686 	else
687 		flags = seqcount << BA_SEQSHIFT;
688 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
689 		flags |= IO_SYNC;
690 
691 	for (error = 0; uio->uio_resid > 0;) {
692 		lbn = lblkno(fs, uio->uio_offset);
693 		blkoffset = blkoff(fs, uio->uio_offset);
694 		xfersize = fs->fs_bsize - blkoffset;
695 		if (uio->uio_resid < xfersize)
696 			xfersize = uio->uio_resid;
697 		if (uio->uio_offset + xfersize > ip->i_size)
698 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
699 
700                 /*
701 		 * We must perform a read-before-write if the transfer size
702 		 * does not cover the entire buffer.
703                  */
704 		if (fs->fs_bsize > xfersize)
705 			flags |= BA_CLRBUF;
706 		else
707 			flags &= ~BA_CLRBUF;
708 /* XXX is uio->uio_offset the right thing here? */
709 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
710 		    ap->a_cred, flags, &bp);
711 		if (error != 0)
712 			break;
713 		/*
714 		 * If the buffer is not valid we have to clear out any
715 		 * garbage data from the pages instantiated for the buffer.
716 		 * If we do not, a failed uiomove() during a write can leave
717 		 * the prior contents of the pages exposed to a userland
718 		 * mmap().  XXX deal with uiomove() errors a better way.
719 		 */
720 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
721 			vfs_bio_clrbuf(bp);
722 		if (ioflag & IO_DIRECT)
723 			bp->b_flags |= B_DIRECT;
724 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
725 			bp->b_flags |= B_NOCACHE;
726 
727 		if (uio->uio_offset + xfersize > ip->i_size) {
728 			ip->i_size = uio->uio_offset + xfersize;
729 			DIP_SET(ip, i_size, ip->i_size);
730 		}
731 
732 		size = blksize(fs, ip, lbn) - bp->b_resid;
733 		if (size < xfersize)
734 			xfersize = size;
735 
736 		error =
737 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
738 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
739 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
740 			bp->b_flags |= B_RELBUF;
741 		}
742 
743 		/*
744 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
745 		 * if we have a severe page deficiency write the buffer
746 		 * asynchronously.  Otherwise try to cluster, and if that
747 		 * doesn't do it then either do an async write (if O_DIRECT),
748 		 * or a delayed write (if not).
749 		 */
750 		if (ioflag & IO_SYNC) {
751 			(void)bwrite(bp);
752 		} else if (vm_page_count_severe() ||
753 			    buf_dirty_count_severe() ||
754 			    (ioflag & IO_ASYNC)) {
755 			bp->b_flags |= B_CLUSTEROK;
756 			bawrite(bp);
757 		} else if (xfersize + blkoffset == fs->fs_bsize) {
758 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
759 				bp->b_flags |= B_CLUSTEROK;
760 				cluster_write(vp, bp, ip->i_size, seqcount);
761 			} else {
762 				bawrite(bp);
763 			}
764 		} else if (ioflag & IO_DIRECT) {
765 			bp->b_flags |= B_CLUSTEROK;
766 			bawrite(bp);
767 		} else {
768 			bp->b_flags |= B_CLUSTEROK;
769 			bdwrite(bp);
770 		}
771 		if (error || xfersize == 0)
772 			break;
773 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
774 	}
775 	/*
776 	 * If we successfully wrote any data, and we are not the superuser
777 	 * we clear the setuid and setgid bits as a precaution against
778 	 * tampering.
779 	 */
780 	if (resid > uio->uio_resid && ap->a_cred &&
781 	    suser_cred(ap->a_cred, SUSER_ALLOWJAIL)) {
782 		ip->i_mode &= ~(ISUID | ISGID);
783 		DIP_SET(ip, i_mode, ip->i_mode);
784 	}
785 	if (error) {
786 		if (ioflag & IO_UNIT) {
787 			(void)ffs_truncate(vp, osize,
788 			    IO_NORMAL | (ioflag & IO_SYNC),
789 			    ap->a_cred, uio->uio_td);
790 			uio->uio_offset -= resid - uio->uio_resid;
791 			uio->uio_resid = resid;
792 		}
793 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
794 		error = ffs_update(vp, 1);
795 	return (error);
796 }
797 
798 /*
799  * get page routine
800  */
801 static int
802 ffs_getpages(ap)
803 	struct vop_getpages_args *ap;
804 {
805 	int i;
806 	vm_page_t mreq;
807 	int pcount;
808 
809 	pcount = round_page(ap->a_count) / PAGE_SIZE;
810 	mreq = ap->a_m[ap->a_reqpage];
811 
812 	/*
813 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
814 	 * then the entire page is valid.  Since the page may be mapped,
815 	 * user programs might reference data beyond the actual end of file
816 	 * occuring within the page.  We have to zero that data.
817 	 */
818 	VM_OBJECT_LOCK(mreq->object);
819 	if (mreq->valid) {
820 		if (mreq->valid != VM_PAGE_BITS_ALL)
821 			vm_page_zero_invalid(mreq, TRUE);
822 		vm_page_lock_queues();
823 		for (i = 0; i < pcount; i++) {
824 			if (i != ap->a_reqpage) {
825 				vm_page_free(ap->a_m[i]);
826 			}
827 		}
828 		vm_page_unlock_queues();
829 		VM_OBJECT_UNLOCK(mreq->object);
830 		return VM_PAGER_OK;
831 	}
832 	VM_OBJECT_UNLOCK(mreq->object);
833 
834 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
835 					    ap->a_count,
836 					    ap->a_reqpage);
837 }
838 
839 
840 /*
841  * Extended attribute area reading.
842  */
843 static int
844 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
845 {
846 	struct inode *ip;
847 	struct ufs2_dinode *dp;
848 	struct fs *fs;
849 	struct buf *bp;
850 	ufs_lbn_t lbn, nextlbn;
851 	off_t bytesinfile;
852 	long size, xfersize, blkoffset;
853 	int error, orig_resid;
854 
855 	ip = VTOI(vp);
856 	fs = ip->i_fs;
857 	dp = ip->i_din2;
858 
859 #ifdef DIAGNOSTIC
860 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
861 		panic("ffs_extread: mode");
862 
863 #endif
864 	orig_resid = uio->uio_resid;
865 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
866 	if (orig_resid == 0)
867 		return (0);
868 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
869 
870 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
871 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
872 			break;
873 		lbn = lblkno(fs, uio->uio_offset);
874 		nextlbn = lbn + 1;
875 
876 		/*
877 		 * size of buffer.  The buffer representing the
878 		 * end of the file is rounded up to the size of
879 		 * the block type ( fragment or full block,
880 		 * depending ).
881 		 */
882 		size = sblksize(fs, dp->di_extsize, lbn);
883 		blkoffset = blkoff(fs, uio->uio_offset);
884 
885 		/*
886 		 * The amount we want to transfer in this iteration is
887 		 * one FS block less the amount of the data before
888 		 * our startpoint (duh!)
889 		 */
890 		xfersize = fs->fs_bsize - blkoffset;
891 
892 		/*
893 		 * But if we actually want less than the block,
894 		 * or the file doesn't have a whole block more of data,
895 		 * then use the lesser number.
896 		 */
897 		if (uio->uio_resid < xfersize)
898 			xfersize = uio->uio_resid;
899 		if (bytesinfile < xfersize)
900 			xfersize = bytesinfile;
901 
902 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
903 			/*
904 			 * Don't do readahead if this is the end of the info.
905 			 */
906 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
907 		} else {
908 			/*
909 			 * If we have a second block, then
910 			 * fire off a request for a readahead
911 			 * as well as a read. Note that the 4th and 5th
912 			 * arguments point to arrays of the size specified in
913 			 * the 6th argument.
914 			 */
915 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
916 
917 			nextlbn = -1 - nextlbn;
918 			error = breadn(vp, -1 - lbn,
919 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
920 		}
921 		if (error) {
922 			brelse(bp);
923 			bp = NULL;
924 			break;
925 		}
926 
927 		/*
928 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
929 		 * will cause us to attempt to release the buffer later on
930 		 * and will cause the buffer cache to attempt to free the
931 		 * underlying pages.
932 		 */
933 		if (ioflag & IO_DIRECT)
934 			bp->b_flags |= B_DIRECT;
935 
936 		/*
937 		 * We should only get non-zero b_resid when an I/O error
938 		 * has occurred, which should cause us to break above.
939 		 * However, if the short read did not cause an error,
940 		 * then we want to ensure that we do not uiomove bad
941 		 * or uninitialized data.
942 		 */
943 		size -= bp->b_resid;
944 		if (size < xfersize) {
945 			if (size == 0)
946 				break;
947 			xfersize = size;
948 		}
949 
950 		error = uiomove((char *)bp->b_data + blkoffset,
951 					(int)xfersize, uio);
952 		if (error)
953 			break;
954 
955 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
956 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
957 			/*
958 			 * If there are no dependencies, and it's VMIO,
959 			 * then we don't need the buf, mark it available
960 			 * for freeing. The VM has the data.
961 			 */
962 			bp->b_flags |= B_RELBUF;
963 			brelse(bp);
964 		} else {
965 			/*
966 			 * Otherwise let whoever
967 			 * made the request take care of
968 			 * freeing it. We just queue
969 			 * it onto another list.
970 			 */
971 			bqrelse(bp);
972 		}
973 	}
974 
975 	/*
976 	 * This can only happen in the case of an error
977 	 * because the loop above resets bp to NULL on each iteration
978 	 * and on normal completion has not set a new value into it.
979 	 * so it must have come from a 'break' statement
980 	 */
981 	if (bp != NULL) {
982 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
983 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
984 			bp->b_flags |= B_RELBUF;
985 			brelse(bp);
986 		} else {
987 			bqrelse(bp);
988 		}
989 	}
990 
991 	if ((error == 0 || uio->uio_resid != orig_resid) &&
992 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
993 		ip->i_flag |= IN_ACCESS;
994 	return (error);
995 }
996 
997 /*
998  * Extended attribute area writing.
999  */
1000 static int
1001 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1002 {
1003 	struct inode *ip;
1004 	struct ufs2_dinode *dp;
1005 	struct fs *fs;
1006 	struct buf *bp;
1007 	ufs_lbn_t lbn;
1008 	off_t osize;
1009 	int blkoffset, error, flags, resid, size, xfersize;
1010 
1011 	ip = VTOI(vp);
1012 	fs = ip->i_fs;
1013 	dp = ip->i_din2;
1014 
1015 #ifdef DIAGNOSTIC
1016 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1017 		panic("ffs_extwrite: mode");
1018 #endif
1019 
1020 	if (ioflag & IO_APPEND)
1021 		uio->uio_offset = dp->di_extsize;
1022 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1023 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1024 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1025 		return (EFBIG);
1026 
1027 	resid = uio->uio_resid;
1028 	osize = dp->di_extsize;
1029 	flags = IO_EXT;
1030 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1031 		flags |= IO_SYNC;
1032 
1033 	for (error = 0; uio->uio_resid > 0;) {
1034 		lbn = lblkno(fs, uio->uio_offset);
1035 		blkoffset = blkoff(fs, uio->uio_offset);
1036 		xfersize = fs->fs_bsize - blkoffset;
1037 		if (uio->uio_resid < xfersize)
1038 			xfersize = uio->uio_resid;
1039 
1040                 /*
1041 		 * We must perform a read-before-write if the transfer size
1042 		 * does not cover the entire buffer.
1043                  */
1044 		if (fs->fs_bsize > xfersize)
1045 			flags |= BA_CLRBUF;
1046 		else
1047 			flags &= ~BA_CLRBUF;
1048 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1049 		    ucred, flags, &bp);
1050 		if (error != 0)
1051 			break;
1052 		/*
1053 		 * If the buffer is not valid we have to clear out any
1054 		 * garbage data from the pages instantiated for the buffer.
1055 		 * If we do not, a failed uiomove() during a write can leave
1056 		 * the prior contents of the pages exposed to a userland
1057 		 * mmap().  XXX deal with uiomove() errors a better way.
1058 		 */
1059 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1060 			vfs_bio_clrbuf(bp);
1061 		if (ioflag & IO_DIRECT)
1062 			bp->b_flags |= B_DIRECT;
1063 
1064 		if (uio->uio_offset + xfersize > dp->di_extsize)
1065 			dp->di_extsize = uio->uio_offset + xfersize;
1066 
1067 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1068 		if (size < xfersize)
1069 			xfersize = size;
1070 
1071 		error =
1072 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1073 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1074 		   (LIST_FIRST(&bp->b_dep) == NULL)) {
1075 			bp->b_flags |= B_RELBUF;
1076 		}
1077 
1078 		/*
1079 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1080 		 * if we have a severe page deficiency write the buffer
1081 		 * asynchronously.  Otherwise try to cluster, and if that
1082 		 * doesn't do it then either do an async write (if O_DIRECT),
1083 		 * or a delayed write (if not).
1084 		 */
1085 		if (ioflag & IO_SYNC) {
1086 			(void)bwrite(bp);
1087 		} else if (vm_page_count_severe() ||
1088 			    buf_dirty_count_severe() ||
1089 			    xfersize + blkoffset == fs->fs_bsize ||
1090 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1091 			bawrite(bp);
1092 		else
1093 			bdwrite(bp);
1094 		if (error || xfersize == 0)
1095 			break;
1096 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1097 	}
1098 	/*
1099 	 * If we successfully wrote any data, and we are not the superuser
1100 	 * we clear the setuid and setgid bits as a precaution against
1101 	 * tampering.
1102 	 */
1103 	if (resid > uio->uio_resid && ucred &&
1104 	    suser_cred(ucred, SUSER_ALLOWJAIL)) {
1105 		ip->i_mode &= ~(ISUID | ISGID);
1106 		dp->di_mode = ip->i_mode;
1107 	}
1108 	if (error) {
1109 		if (ioflag & IO_UNIT) {
1110 			(void)ffs_truncate(vp, osize,
1111 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1112 			uio->uio_offset -= resid - uio->uio_resid;
1113 			uio->uio_resid = resid;
1114 		}
1115 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1116 		error = ffs_update(vp, 1);
1117 	return (error);
1118 }
1119 
1120 
1121 /*
1122  * Vnode operating to retrieve a named extended attribute.
1123  *
1124  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1125  * the length of the EA, and possibly the pointer to the entry and to the data.
1126  */
1127 static int
1128 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1129 {
1130 	u_char *p, *pe, *pn, *p0;
1131 	int eapad1, eapad2, ealength, ealen, nlen;
1132 	uint32_t ul;
1133 
1134 	pe = ptr + length;
1135 	nlen = strlen(name);
1136 
1137 	for (p = ptr; p < pe; p = pn) {
1138 		p0 = p;
1139 		bcopy(p, &ul, sizeof(ul));
1140 		pn = p + ul;
1141 		/* make sure this entry is complete */
1142 		if (pn > pe)
1143 			break;
1144 		p += sizeof(uint32_t);
1145 		if (*p != nspace)
1146 			continue;
1147 		p++;
1148 		eapad2 = *p++;
1149 		if (*p != nlen)
1150 			continue;
1151 		p++;
1152 		if (bcmp(p, name, nlen))
1153 			continue;
1154 		ealength = sizeof(uint32_t) + 3 + nlen;
1155 		eapad1 = 8 - (ealength % 8);
1156 		if (eapad1 == 8)
1157 			eapad1 = 0;
1158 		ealength += eapad1;
1159 		ealen = ul - ealength - eapad2;
1160 		p += nlen + eapad1;
1161 		if (eap != NULL)
1162 			*eap = p0;
1163 		if (eac != NULL)
1164 			*eac = p;
1165 		return (ealen);
1166 	}
1167 	return(-1);
1168 }
1169 
1170 static int
1171 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1172 {
1173 	struct inode *ip;
1174 	struct ufs2_dinode *dp;
1175 	struct uio luio;
1176 	struct iovec liovec;
1177 	int easize, error;
1178 	u_char *eae;
1179 
1180 	ip = VTOI(vp);
1181 	dp = ip->i_din2;
1182 	easize = dp->di_extsize;
1183 
1184 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1185 
1186 	liovec.iov_base = eae;
1187 	liovec.iov_len = easize;
1188 	luio.uio_iov = &liovec;
1189 	luio.uio_iovcnt = 1;
1190 	luio.uio_offset = 0;
1191 	luio.uio_resid = easize;
1192 	luio.uio_segflg = UIO_SYSSPACE;
1193 	luio.uio_rw = UIO_READ;
1194 	luio.uio_td = td;
1195 
1196 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1197 	if (error) {
1198 		free(eae, M_TEMP);
1199 		return(error);
1200 	}
1201 	*p = eae;
1202 	return (0);
1203 }
1204 
1205 static int
1206 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1207 {
1208 	struct inode *ip;
1209 	struct ufs2_dinode *dp;
1210 	int error;
1211 
1212 	ip = VTOI(vp);
1213 
1214 	if (ip->i_ea_area != NULL)
1215 		return (EBUSY);
1216 	dp = ip->i_din2;
1217 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1218 	if (error)
1219 		return (error);
1220 	ip->i_ea_len = dp->di_extsize;
1221 	ip->i_ea_error = 0;
1222 	return (0);
1223 }
1224 
1225 /*
1226  * Vnode extattr transaction commit/abort
1227  */
1228 static int
1229 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1230 {
1231 	struct inode *ip;
1232 	struct uio luio;
1233 	struct iovec liovec;
1234 	int error;
1235 	struct ufs2_dinode *dp;
1236 
1237 	ip = VTOI(vp);
1238 	if (ip->i_ea_area == NULL)
1239 		return (EINVAL);
1240 	dp = ip->i_din2;
1241 	error = ip->i_ea_error;
1242 	if (commit && error == 0) {
1243 		if (cred == NOCRED)
1244 			cred =  vp->v_mount->mnt_cred;
1245 		liovec.iov_base = ip->i_ea_area;
1246 		liovec.iov_len = ip->i_ea_len;
1247 		luio.uio_iov = &liovec;
1248 		luio.uio_iovcnt = 1;
1249 		luio.uio_offset = 0;
1250 		luio.uio_resid = ip->i_ea_len;
1251 		luio.uio_segflg = UIO_SYSSPACE;
1252 		luio.uio_rw = UIO_WRITE;
1253 		luio.uio_td = td;
1254 		/* XXX: I'm not happy about truncating to zero size */
1255 		if (ip->i_ea_len < dp->di_extsize)
1256 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1257 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1258 	}
1259 	free(ip->i_ea_area, M_TEMP);
1260 	ip->i_ea_area = NULL;
1261 	ip->i_ea_len = 0;
1262 	ip->i_ea_error = 0;
1263 	return (error);
1264 }
1265 
1266 /*
1267  * Vnode extattr strategy routine for fifos.
1268  *
1269  * We need to check for a read or write of the external attributes.
1270  * Otherwise we just fall through and do the usual thing.
1271  */
1272 static int
1273 ffsext_strategy(struct vop_strategy_args *ap)
1274 /*
1275 struct vop_strategy_args {
1276 	struct vnodeop_desc *a_desc;
1277 	struct vnode *a_vp;
1278 	struct buf *a_bp;
1279 };
1280 */
1281 {
1282 	struct vnode *vp;
1283 	daddr_t lbn;
1284 
1285 	vp = ap->a_vp;
1286 	lbn = ap->a_bp->b_lblkno;
1287 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1288 	    lbn < 0 && lbn >= -NXADDR)
1289 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1290 	if (vp->v_type == VFIFO)
1291 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1292 	panic("spec nodes went here");
1293 }
1294 
1295 /*
1296  * Vnode extattr transaction commit/abort
1297  */
1298 static int
1299 ffs_openextattr(struct vop_openextattr_args *ap)
1300 /*
1301 struct vop_openextattr_args {
1302 	struct vnodeop_desc *a_desc;
1303 	struct vnode *a_vp;
1304 	IN struct ucred *a_cred;
1305 	IN struct thread *a_td;
1306 };
1307 */
1308 {
1309 	struct inode *ip;
1310 	struct fs *fs;
1311 
1312 	ip = VTOI(ap->a_vp);
1313 	fs = ip->i_fs;
1314 
1315 	if (ap->a_vp->v_type == VCHR)
1316 		return (EOPNOTSUPP);
1317 
1318 	return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1319 }
1320 
1321 
1322 /*
1323  * Vnode extattr transaction commit/abort
1324  */
1325 static int
1326 ffs_closeextattr(struct vop_closeextattr_args *ap)
1327 /*
1328 struct vop_closeextattr_args {
1329 	struct vnodeop_desc *a_desc;
1330 	struct vnode *a_vp;
1331 	int a_commit;
1332 	IN struct ucred *a_cred;
1333 	IN struct thread *a_td;
1334 };
1335 */
1336 {
1337 	struct inode *ip;
1338 	struct fs *fs;
1339 
1340 	ip = VTOI(ap->a_vp);
1341 	fs = ip->i_fs;
1342 
1343 	if (ap->a_vp->v_type == VCHR)
1344 		return (EOPNOTSUPP);
1345 
1346 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1347 }
1348 
1349 /*
1350  * Vnode operation to remove a named attribute.
1351  */
1352 static int
1353 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1354 /*
1355 vop_deleteextattr {
1356 	IN struct vnode *a_vp;
1357 	IN int a_attrnamespace;
1358 	IN const char *a_name;
1359 	IN struct ucred *a_cred;
1360 	IN struct thread *a_td;
1361 };
1362 */
1363 {
1364 	struct inode *ip;
1365 	struct fs *fs;
1366 	uint32_t ealength, ul;
1367 	int ealen, olen, eapad1, eapad2, error, i, easize;
1368 	u_char *eae, *p;
1369 	int stand_alone;
1370 
1371 	ip = VTOI(ap->a_vp);
1372 	fs = ip->i_fs;
1373 
1374 	if (ap->a_vp->v_type == VCHR)
1375 		return (EOPNOTSUPP);
1376 
1377 	if (strlen(ap->a_name) == 0)
1378 		return (EINVAL);
1379 
1380 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1381 	    ap->a_cred, ap->a_td, IWRITE);
1382 	if (error) {
1383 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1384 			ip->i_ea_error = error;
1385 		return (error);
1386 	}
1387 
1388 	if (ip->i_ea_area == NULL) {
1389 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1390 		if (error)
1391 			return (error);
1392 		stand_alone = 1;
1393 	} else {
1394 		stand_alone = 0;
1395 	}
1396 
1397 	ealength = eapad1 = ealen = eapad2 = 0;
1398 
1399 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1400 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1401 	easize = ip->i_ea_len;
1402 
1403 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1404 	    &p, NULL);
1405 	if (olen == -1) {
1406 		/* delete but nonexistent */
1407 		free(eae, M_TEMP);
1408 		if (stand_alone)
1409 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1410 		return(ENOATTR);
1411 	}
1412 	bcopy(p, &ul, sizeof ul);
1413 	i = p - eae + ul;
1414 	if (ul != ealength) {
1415 		bcopy(p + ul, p + ealength, easize - i);
1416 		easize += (ealength - ul);
1417 	}
1418 	if (easize > NXADDR * fs->fs_bsize) {
1419 		free(eae, M_TEMP);
1420 		if (stand_alone)
1421 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1422 		else if (ip->i_ea_error == 0)
1423 			ip->i_ea_error = ENOSPC;
1424 		return(ENOSPC);
1425 	}
1426 	p = ip->i_ea_area;
1427 	ip->i_ea_area = eae;
1428 	ip->i_ea_len = easize;
1429 	free(p, M_TEMP);
1430 	if (stand_alone)
1431 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1432 	return(error);
1433 }
1434 
1435 /*
1436  * Vnode operation to retrieve a named extended attribute.
1437  */
1438 static int
1439 ffs_getextattr(struct vop_getextattr_args *ap)
1440 /*
1441 vop_getextattr {
1442 	IN struct vnode *a_vp;
1443 	IN int a_attrnamespace;
1444 	IN const char *a_name;
1445 	INOUT struct uio *a_uio;
1446 	OUT size_t *a_size;
1447 	IN struct ucred *a_cred;
1448 	IN struct thread *a_td;
1449 };
1450 */
1451 {
1452 	struct inode *ip;
1453 	struct fs *fs;
1454 	u_char *eae, *p;
1455 	unsigned easize;
1456 	int error, ealen, stand_alone;
1457 
1458 	ip = VTOI(ap->a_vp);
1459 	fs = ip->i_fs;
1460 
1461 	if (ap->a_vp->v_type == VCHR)
1462 		return (EOPNOTSUPP);
1463 
1464 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1465 	    ap->a_cred, ap->a_td, IREAD);
1466 	if (error)
1467 		return (error);
1468 
1469 	if (ip->i_ea_area == NULL) {
1470 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1471 		if (error)
1472 			return (error);
1473 		stand_alone = 1;
1474 	} else {
1475 		stand_alone = 0;
1476 	}
1477 	eae = ip->i_ea_area;
1478 	easize = ip->i_ea_len;
1479 
1480 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1481 	    NULL, &p);
1482 	if (ealen >= 0) {
1483 		error = 0;
1484 		if (ap->a_size != NULL)
1485 			*ap->a_size = ealen;
1486 		else if (ap->a_uio != NULL)
1487 			error = uiomove(p, ealen, ap->a_uio);
1488 	} else
1489 		error = ENOATTR;
1490 	if (stand_alone)
1491 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1492 	return(error);
1493 }
1494 
1495 /*
1496  * Vnode operation to retrieve extended attributes on a vnode.
1497  */
1498 static int
1499 ffs_listextattr(struct vop_listextattr_args *ap)
1500 /*
1501 vop_listextattr {
1502 	IN struct vnode *a_vp;
1503 	IN int a_attrnamespace;
1504 	INOUT struct uio *a_uio;
1505 	OUT size_t *a_size;
1506 	IN struct ucred *a_cred;
1507 	IN struct thread *a_td;
1508 };
1509 */
1510 {
1511 	struct inode *ip;
1512 	struct fs *fs;
1513 	u_char *eae, *p, *pe, *pn;
1514 	unsigned easize;
1515 	uint32_t ul;
1516 	int error, ealen, stand_alone;
1517 
1518 	ip = VTOI(ap->a_vp);
1519 	fs = ip->i_fs;
1520 
1521 	if (ap->a_vp->v_type == VCHR)
1522 		return (EOPNOTSUPP);
1523 
1524 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1525 	    ap->a_cred, ap->a_td, IREAD);
1526 	if (error)
1527 		return (error);
1528 
1529 	if (ip->i_ea_area == NULL) {
1530 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1531 		if (error)
1532 			return (error);
1533 		stand_alone = 1;
1534 	} else {
1535 		stand_alone = 0;
1536 	}
1537 	eae = ip->i_ea_area;
1538 	easize = ip->i_ea_len;
1539 
1540 	error = 0;
1541 	if (ap->a_size != NULL)
1542 		*ap->a_size = 0;
1543 	pe = eae + easize;
1544 	for(p = eae; error == 0 && p < pe; p = pn) {
1545 		bcopy(p, &ul, sizeof(ul));
1546 		pn = p + ul;
1547 		if (pn > pe)
1548 			break;
1549 		p += sizeof(ul);
1550 		if (*p++ != ap->a_attrnamespace)
1551 			continue;
1552 		p++;	/* pad2 */
1553 		ealen = *p;
1554 		if (ap->a_size != NULL) {
1555 			*ap->a_size += ealen + 1;
1556 		} else if (ap->a_uio != NULL) {
1557 			error = uiomove(p, ealen + 1, ap->a_uio);
1558 		}
1559 	}
1560 	if (stand_alone)
1561 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1562 	return(error);
1563 }
1564 
1565 /*
1566  * Vnode operation to set a named attribute.
1567  */
1568 static int
1569 ffs_setextattr(struct vop_setextattr_args *ap)
1570 /*
1571 vop_setextattr {
1572 	IN struct vnode *a_vp;
1573 	IN int a_attrnamespace;
1574 	IN const char *a_name;
1575 	INOUT struct uio *a_uio;
1576 	IN struct ucred *a_cred;
1577 	IN struct thread *a_td;
1578 };
1579 */
1580 {
1581 	struct inode *ip;
1582 	struct fs *fs;
1583 	uint32_t ealength, ul;
1584 	int ealen, olen, eapad1, eapad2, error, i, easize;
1585 	u_char *eae, *p;
1586 	int stand_alone;
1587 
1588 	ip = VTOI(ap->a_vp);
1589 	fs = ip->i_fs;
1590 
1591 	if (ap->a_vp->v_type == VCHR)
1592 		return (EOPNOTSUPP);
1593 
1594 	if (strlen(ap->a_name) == 0)
1595 		return (EINVAL);
1596 
1597 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1598 	if (ap->a_uio == NULL)
1599 		return (EOPNOTSUPP);
1600 
1601 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1602 	    ap->a_cred, ap->a_td, IWRITE);
1603 	if (error) {
1604 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1605 			ip->i_ea_error = error;
1606 		return (error);
1607 	}
1608 
1609 	if (ip->i_ea_area == NULL) {
1610 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1611 		if (error)
1612 			return (error);
1613 		stand_alone = 1;
1614 	} else {
1615 		stand_alone = 0;
1616 	}
1617 
1618 	ealen = ap->a_uio->uio_resid;
1619 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1620 	eapad1 = 8 - (ealength % 8);
1621 	if (eapad1 == 8)
1622 		eapad1 = 0;
1623 	eapad2 = 8 - (ealen % 8);
1624 	if (eapad2 == 8)
1625 		eapad2 = 0;
1626 	ealength += eapad1 + ealen + eapad2;
1627 
1628 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1629 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1630 	easize = ip->i_ea_len;
1631 
1632 	olen = ffs_findextattr(eae, easize,
1633 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1634         if (olen == -1) {
1635 		/* new, append at end */
1636 		p = eae + easize;
1637 		easize += ealength;
1638 	} else {
1639 		bcopy(p, &ul, sizeof ul);
1640 		i = p - eae + ul;
1641 		if (ul != ealength) {
1642 			bcopy(p + ul, p + ealength, easize - i);
1643 			easize += (ealength - ul);
1644 		}
1645 	}
1646 	if (easize > NXADDR * fs->fs_bsize) {
1647 		free(eae, M_TEMP);
1648 		if (stand_alone)
1649 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1650 		else if (ip->i_ea_error == 0)
1651 			ip->i_ea_error = ENOSPC;
1652 		return(ENOSPC);
1653 	}
1654 	bcopy(&ealength, p, sizeof(ealength));
1655 	p += sizeof(ealength);
1656 	*p++ = ap->a_attrnamespace;
1657 	*p++ = eapad2;
1658 	*p++ = strlen(ap->a_name);
1659 	strcpy(p, ap->a_name);
1660 	p += strlen(ap->a_name);
1661 	bzero(p, eapad1);
1662 	p += eapad1;
1663 	error = uiomove(p, ealen, ap->a_uio);
1664 	if (error) {
1665 		free(eae, M_TEMP);
1666 		if (stand_alone)
1667 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1668 		else if (ip->i_ea_error == 0)
1669 			ip->i_ea_error = error;
1670 		return(error);
1671 	}
1672 	p += ealen;
1673 	bzero(p, eapad2);
1674 
1675 	p = ip->i_ea_area;
1676 	ip->i_ea_area = eae;
1677 	ip->i_ea_len = easize;
1678 	free(p, M_TEMP);
1679 	if (stand_alone)
1680 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1681 	return(error);
1682 }
1683