xref: /freebsd/sys/ufs/ffs/ffs_vnops.c (revision 995dc984471c92c03daad19a1d35af46c086ef3e)
1 /*-
2  * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
3  * All rights reserved.
4  *
5  * This software was developed for the FreeBSD Project by Marshall
6  * Kirk McKusick and Network Associates Laboratories, the Security
7  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
8  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
9  * research program
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  * Copyright (c) 1982, 1986, 1989, 1993
33  *	The Regents of the University of California.  All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 4. Neither the name of the University nor the names of its contributors
44  *    may be used to endorse or promote products derived from this software
45  *    without specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
57  * SUCH DAMAGE.
58  *
59  *	from: @(#)ufs_readwrite.c	8.11 (Berkeley) 5/8/95
60  * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
61  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
62  */
63 
64 #include <sys/cdefs.h>
65 __FBSDID("$FreeBSD$");
66 
67 #include <sys/param.h>
68 #include <sys/bio.h>
69 #include <sys/systm.h>
70 #include <sys/buf.h>
71 #include <sys/conf.h>
72 #include <sys/extattr.h>
73 #include <sys/kernel.h>
74 #include <sys/limits.h>
75 #include <sys/malloc.h>
76 #include <sys/mount.h>
77 #include <sys/priv.h>
78 #include <sys/proc.h>
79 #include <sys/resourcevar.h>
80 #include <sys/signalvar.h>
81 #include <sys/stat.h>
82 #include <sys/vmmeter.h>
83 #include <sys/vnode.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_pager.h>
90 #include <vm/vnode_pager.h>
91 
92 #include <ufs/ufs/extattr.h>
93 #include <ufs/ufs/quota.h>
94 #include <ufs/ufs/inode.h>
95 #include <ufs/ufs/ufs_extern.h>
96 #include <ufs/ufs/ufsmount.h>
97 
98 #include <ufs/ffs/fs.h>
99 #include <ufs/ffs/ffs_extern.h>
100 #include "opt_directio.h"
101 #include "opt_ffs.h"
102 
103 #ifdef DIRECTIO
104 extern int	ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
105 #endif
106 static vop_fsync_t	ffs_fsync;
107 static vop_lock1_t	ffs_lock;
108 static vop_getpages_t	ffs_getpages;
109 static vop_read_t	ffs_read;
110 static vop_write_t	ffs_write;
111 static int	ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
112 static int	ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
113 		    struct ucred *cred);
114 static vop_strategy_t	ffsext_strategy;
115 static vop_closeextattr_t	ffs_closeextattr;
116 static vop_deleteextattr_t	ffs_deleteextattr;
117 static vop_getextattr_t	ffs_getextattr;
118 static vop_listextattr_t	ffs_listextattr;
119 static vop_openextattr_t	ffs_openextattr;
120 static vop_setextattr_t	ffs_setextattr;
121 static vop_vptofh_t	ffs_vptofh;
122 
123 
124 /* Global vfs data structures for ufs. */
125 struct vop_vector ffs_vnodeops1 = {
126 	.vop_default =		&ufs_vnodeops,
127 	.vop_fsync =		ffs_fsync,
128 	.vop_getpages =		ffs_getpages,
129 	.vop_lock1 =		ffs_lock,
130 	.vop_read =		ffs_read,
131 	.vop_reallocblks =	ffs_reallocblks,
132 	.vop_write =		ffs_write,
133 	.vop_vptofh =		ffs_vptofh,
134 };
135 
136 struct vop_vector ffs_fifoops1 = {
137 	.vop_default =		&ufs_fifoops,
138 	.vop_fsync =		ffs_fsync,
139 	.vop_reallocblks =	ffs_reallocblks, /* XXX: really ??? */
140 	.vop_vptofh =		ffs_vptofh,
141 };
142 
143 /* Global vfs data structures for ufs. */
144 struct vop_vector ffs_vnodeops2 = {
145 	.vop_default =		&ufs_vnodeops,
146 	.vop_fsync =		ffs_fsync,
147 	.vop_getpages =		ffs_getpages,
148 	.vop_lock1 =		ffs_lock,
149 	.vop_read =		ffs_read,
150 	.vop_reallocblks =	ffs_reallocblks,
151 	.vop_write =		ffs_write,
152 	.vop_closeextattr =	ffs_closeextattr,
153 	.vop_deleteextattr =	ffs_deleteextattr,
154 	.vop_getextattr =	ffs_getextattr,
155 	.vop_listextattr =	ffs_listextattr,
156 	.vop_openextattr =	ffs_openextattr,
157 	.vop_setextattr =	ffs_setextattr,
158 	.vop_vptofh =		ffs_vptofh,
159 };
160 
161 struct vop_vector ffs_fifoops2 = {
162 	.vop_default =		&ufs_fifoops,
163 	.vop_fsync =		ffs_fsync,
164 	.vop_lock1 =		ffs_lock,
165 	.vop_reallocblks =	ffs_reallocblks,
166 	.vop_strategy =		ffsext_strategy,
167 	.vop_closeextattr =	ffs_closeextattr,
168 	.vop_deleteextattr =	ffs_deleteextattr,
169 	.vop_getextattr =	ffs_getextattr,
170 	.vop_listextattr =	ffs_listextattr,
171 	.vop_openextattr =	ffs_openextattr,
172 	.vop_setextattr =	ffs_setextattr,
173 	.vop_vptofh =		ffs_vptofh,
174 };
175 
176 /*
177  * Synch an open file.
178  */
179 /* ARGSUSED */
180 static int
181 ffs_fsync(struct vop_fsync_args *ap)
182 {
183 	int error;
184 
185 	error = ffs_syncvnode(ap->a_vp, ap->a_waitfor);
186 	if (error)
187 		return (error);
188 	if (ap->a_waitfor == MNT_WAIT &&
189 	    (ap->a_vp->v_mount->mnt_flag & MNT_SOFTDEP))
190                 error = softdep_fsync(ap->a_vp);
191 	return (error);
192 }
193 
194 int
195 ffs_syncvnode(struct vnode *vp, int waitfor)
196 {
197 	struct inode *ip = VTOI(vp);
198 	struct buf *bp;
199 	struct buf *nbp;
200 	int s, error, wait, passes, skipmeta;
201 	ufs_lbn_t lbn;
202 
203 	wait = (waitfor == MNT_WAIT);
204 	lbn = lblkno(ip->i_fs, (ip->i_size + ip->i_fs->fs_bsize - 1));
205 
206 	/*
207 	 * Flush all dirty buffers associated with a vnode.
208 	 */
209 	passes = NIADDR + 1;
210 	skipmeta = 0;
211 	if (wait)
212 		skipmeta = 1;
213 	s = splbio();
214 	VI_LOCK(vp);
215 loop:
216 	TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs)
217 		bp->b_vflags &= ~BV_SCANNED;
218 	TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
219 		/*
220 		 * Reasons to skip this buffer: it has already been considered
221 		 * on this pass, this pass is the first time through on a
222 		 * synchronous flush request and the buffer being considered
223 		 * is metadata, the buffer has dependencies that will cause
224 		 * it to be redirtied and it has not already been deferred,
225 		 * or it is already being written.
226 		 */
227 		if ((bp->b_vflags & BV_SCANNED) != 0)
228 			continue;
229 		bp->b_vflags |= BV_SCANNED;
230 		if ((skipmeta == 1 && bp->b_lblkno < 0))
231 			continue;
232 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
233 			continue;
234 		VI_UNLOCK(vp);
235 		if (!wait && !LIST_EMPTY(&bp->b_dep) &&
236 		    (bp->b_flags & B_DEFERRED) == 0 &&
237 		    buf_countdeps(bp, 0)) {
238 			bp->b_flags |= B_DEFERRED;
239 			BUF_UNLOCK(bp);
240 			VI_LOCK(vp);
241 			continue;
242 		}
243 		if ((bp->b_flags & B_DELWRI) == 0)
244 			panic("ffs_fsync: not dirty");
245 		/*
246 		 * If this is a synchronous flush request, or it is not a
247 		 * file or device, start the write on this buffer immediatly.
248 		 */
249 		if (wait || (vp->v_type != VREG && vp->v_type != VBLK)) {
250 
251 			/*
252 			 * On our final pass through, do all I/O synchronously
253 			 * so that we can find out if our flush is failing
254 			 * because of write errors.
255 			 */
256 			if (passes > 0 || !wait) {
257 				if ((bp->b_flags & B_CLUSTEROK) && !wait) {
258 					(void) vfs_bio_awrite(bp);
259 				} else {
260 					bremfree(bp);
261 					splx(s);
262 					(void) bawrite(bp);
263 					s = splbio();
264 				}
265 			} else {
266 				bremfree(bp);
267 				splx(s);
268 				if ((error = bwrite(bp)) != 0)
269 					return (error);
270 				s = splbio();
271 			}
272 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
273 			/*
274 			 * If the buffer is for data that has been truncated
275 			 * off the file, then throw it away.
276 			 */
277 			bremfree(bp);
278 			bp->b_flags |= B_INVAL | B_NOCACHE;
279 			splx(s);
280 			brelse(bp);
281 			s = splbio();
282 		} else
283 			vfs_bio_awrite(bp);
284 
285 		/*
286 		 * Since we may have slept during the I/O, we need
287 		 * to start from a known point.
288 		 */
289 		VI_LOCK(vp);
290 		nbp = TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd);
291 	}
292 	/*
293 	 * If we were asked to do this synchronously, then go back for
294 	 * another pass, this time doing the metadata.
295 	 */
296 	if (skipmeta) {
297 		skipmeta = 0;
298 		goto loop;
299 	}
300 
301 	if (wait) {
302 		bufobj_wwait(&vp->v_bufobj, 3, 0);
303 		VI_UNLOCK(vp);
304 
305 		/*
306 		 * Ensure that any filesystem metatdata associated
307 		 * with the vnode has been written.
308 		 */
309 		splx(s);
310 		if ((error = softdep_sync_metadata(vp)) != 0)
311 			return (error);
312 		s = splbio();
313 
314 		VI_LOCK(vp);
315 		if (vp->v_bufobj.bo_dirty.bv_cnt > 0) {
316 			/*
317 			 * Block devices associated with filesystems may
318 			 * have new I/O requests posted for them even if
319 			 * the vnode is locked, so no amount of trying will
320 			 * get them clean. Thus we give block devices a
321 			 * good effort, then just give up. For all other file
322 			 * types, go around and try again until it is clean.
323 			 */
324 			if (passes > 0) {
325 				passes -= 1;
326 				goto loop;
327 			}
328 #ifdef INVARIANTS
329 			if (!vn_isdisk(vp, NULL))
330 				vprint("ffs_fsync: dirty", vp);
331 #endif
332 		}
333 	}
334 	VI_UNLOCK(vp);
335 	splx(s);
336 	return (ffs_update(vp, wait));
337 }
338 
339 static int
340 ffs_lock(ap)
341 	struct vop_lock1_args /* {
342 		struct vnode *a_vp;
343 		int a_flags;
344 		struct thread *a_td;
345 		char *file;
346 		int line;
347 	} */ *ap;
348 {
349 #ifndef NO_FFS_SNAPSHOT
350 	struct vnode *vp;
351 	int flags;
352 	struct lock *lkp;
353 	int result;
354 
355 	switch (ap->a_flags & LK_TYPE_MASK) {
356 	case LK_SHARED:
357 	case LK_UPGRADE:
358 	case LK_EXCLUSIVE:
359 		vp = ap->a_vp;
360 		flags = ap->a_flags;
361 		for (;;) {
362 			/*
363 			 * vnode interlock must be held to ensure that
364 			 * the possibly external lock isn't freed,
365 			 * e.g. when mutating from snapshot file vnode
366 			 * to regular file vnode.
367 			 */
368 			if ((flags & LK_INTERLOCK) == 0) {
369 				VI_LOCK(vp);
370 				flags |= LK_INTERLOCK;
371 			}
372 			lkp = vp->v_vnlock;
373 			result = _lockmgr_args(lkp, flags, VI_MTX(vp),
374 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
375 			    ap->a_file, ap->a_line);
376 			if (lkp == vp->v_vnlock || result != 0)
377 				break;
378 			/*
379 			 * Apparent success, except that the vnode
380 			 * mutated between snapshot file vnode and
381 			 * regular file vnode while this process
382 			 * slept.  The lock currently held is not the
383 			 * right lock.  Release it, and try to get the
384 			 * new lock.
385 			 */
386 			(void) _lockmgr_args(lkp, LK_RELEASE, VI_MTX(vp),
387 			    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT,
388 			    ap->a_file, ap->a_line);
389 			if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
390 				flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
391 			flags &= ~LK_INTERLOCK;
392 		}
393 		break;
394 	default:
395 		result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
396 	}
397 	return (result);
398 #else
399 	return (VOP_LOCK1_APV(&ufs_vnodeops, ap));
400 #endif
401 }
402 
403 /*
404  * Vnode op for reading.
405  */
406 /* ARGSUSED */
407 static int
408 ffs_read(ap)
409 	struct vop_read_args /* {
410 		struct vnode *a_vp;
411 		struct uio *a_uio;
412 		int a_ioflag;
413 		struct ucred *a_cred;
414 	} */ *ap;
415 {
416 	struct vnode *vp;
417 	struct inode *ip;
418 	struct uio *uio;
419 	struct fs *fs;
420 	struct buf *bp;
421 	ufs_lbn_t lbn, nextlbn;
422 	off_t bytesinfile;
423 	long size, xfersize, blkoffset;
424 	int error, orig_resid;
425 	int seqcount;
426 	int ioflag;
427 
428 	vp = ap->a_vp;
429 	uio = ap->a_uio;
430 	ioflag = ap->a_ioflag;
431 	if (ap->a_ioflag & IO_EXT)
432 #ifdef notyet
433 		return (ffs_extread(vp, uio, ioflag));
434 #else
435 		panic("ffs_read+IO_EXT");
436 #endif
437 #ifdef DIRECTIO
438 	if ((ioflag & IO_DIRECT) != 0) {
439 		int workdone;
440 
441 		error = ffs_rawread(vp, uio, &workdone);
442 		if (error != 0 || workdone != 0)
443 			return error;
444 	}
445 #endif
446 
447 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
448 	ip = VTOI(vp);
449 
450 #ifdef INVARIANTS
451 	if (uio->uio_rw != UIO_READ)
452 		panic("ffs_read: mode");
453 
454 	if (vp->v_type == VLNK) {
455 		if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen)
456 			panic("ffs_read: short symlink");
457 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
458 		panic("ffs_read: type %d",  vp->v_type);
459 #endif
460 	orig_resid = uio->uio_resid;
461 	KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
462 	if (orig_resid == 0)
463 		return (0);
464 	KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
465 	fs = ip->i_fs;
466 	if (uio->uio_offset < ip->i_size &&
467 	    uio->uio_offset >= fs->fs_maxfilesize)
468 		return (EOVERFLOW);
469 
470 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
471 		if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
472 			break;
473 		lbn = lblkno(fs, uio->uio_offset);
474 		nextlbn = lbn + 1;
475 
476 		/*
477 		 * size of buffer.  The buffer representing the
478 		 * end of the file is rounded up to the size of
479 		 * the block type ( fragment or full block,
480 		 * depending ).
481 		 */
482 		size = blksize(fs, ip, lbn);
483 		blkoffset = blkoff(fs, uio->uio_offset);
484 
485 		/*
486 		 * The amount we want to transfer in this iteration is
487 		 * one FS block less the amount of the data before
488 		 * our startpoint (duh!)
489 		 */
490 		xfersize = fs->fs_bsize - blkoffset;
491 
492 		/*
493 		 * But if we actually want less than the block,
494 		 * or the file doesn't have a whole block more of data,
495 		 * then use the lesser number.
496 		 */
497 		if (uio->uio_resid < xfersize)
498 			xfersize = uio->uio_resid;
499 		if (bytesinfile < xfersize)
500 			xfersize = bytesinfile;
501 
502 		if (lblktosize(fs, nextlbn) >= ip->i_size) {
503 			/*
504 			 * Don't do readahead if this is the end of the file.
505 			 */
506 			error = bread(vp, lbn, size, NOCRED, &bp);
507 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
508 			/*
509 			 * Otherwise if we are allowed to cluster,
510 			 * grab as much as we can.
511 			 *
512 			 * XXX  This may not be a win if we are not
513 			 * doing sequential access.
514 			 */
515 			error = cluster_read(vp, ip->i_size, lbn,
516 				size, NOCRED, blkoffset + uio->uio_resid, seqcount, &bp);
517 		} else if (seqcount > 1) {
518 			/*
519 			 * If we are NOT allowed to cluster, then
520 			 * if we appear to be acting sequentially,
521 			 * fire off a request for a readahead
522 			 * as well as a read. Note that the 4th and 5th
523 			 * arguments point to arrays of the size specified in
524 			 * the 6th argument.
525 			 */
526 			int nextsize = blksize(fs, ip, nextlbn);
527 			error = breadn(vp, lbn,
528 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
529 		} else {
530 			/*
531 			 * Failing all of the above, just read what the
532 			 * user asked for. Interestingly, the same as
533 			 * the first option above.
534 			 */
535 			error = bread(vp, lbn, size, NOCRED, &bp);
536 		}
537 		if (error) {
538 			brelse(bp);
539 			bp = NULL;
540 			break;
541 		}
542 
543 		/*
544 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
545 		 * will cause us to attempt to release the buffer later on
546 		 * and will cause the buffer cache to attempt to free the
547 		 * underlying pages.
548 		 */
549 		if (ioflag & IO_DIRECT)
550 			bp->b_flags |= B_DIRECT;
551 
552 		/*
553 		 * We should only get non-zero b_resid when an I/O error
554 		 * has occurred, which should cause us to break above.
555 		 * However, if the short read did not cause an error,
556 		 * then we want to ensure that we do not uiomove bad
557 		 * or uninitialized data.
558 		 */
559 		size -= bp->b_resid;
560 		if (size < xfersize) {
561 			if (size == 0)
562 				break;
563 			xfersize = size;
564 		}
565 
566 		error = uiomove((char *)bp->b_data + blkoffset,
567 		    (int)xfersize, uio);
568 		if (error)
569 			break;
570 
571 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
572 		   (LIST_EMPTY(&bp->b_dep))) {
573 			/*
574 			 * If there are no dependencies, and it's VMIO,
575 			 * then we don't need the buf, mark it available
576 			 * for freeing. The VM has the data.
577 			 */
578 			bp->b_flags |= B_RELBUF;
579 			brelse(bp);
580 		} else {
581 			/*
582 			 * Otherwise let whoever
583 			 * made the request take care of
584 			 * freeing it. We just queue
585 			 * it onto another list.
586 			 */
587 			bqrelse(bp);
588 		}
589 	}
590 
591 	/*
592 	 * This can only happen in the case of an error
593 	 * because the loop above resets bp to NULL on each iteration
594 	 * and on normal completion has not set a new value into it.
595 	 * so it must have come from a 'break' statement
596 	 */
597 	if (bp != NULL) {
598 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
599 		   (LIST_EMPTY(&bp->b_dep))) {
600 			bp->b_flags |= B_RELBUF;
601 			brelse(bp);
602 		} else {
603 			bqrelse(bp);
604 		}
605 	}
606 
607 	if ((error == 0 || uio->uio_resid != orig_resid) &&
608 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) {
609 		VI_LOCK(vp);
610 		ip->i_flag |= IN_ACCESS;
611 		VI_UNLOCK(vp);
612 	}
613 	return (error);
614 }
615 
616 /*
617  * Vnode op for writing.
618  */
619 static int
620 ffs_write(ap)
621 	struct vop_write_args /* {
622 		struct vnode *a_vp;
623 		struct uio *a_uio;
624 		int a_ioflag;
625 		struct ucred *a_cred;
626 	} */ *ap;
627 {
628 	struct vnode *vp;
629 	struct uio *uio;
630 	struct inode *ip;
631 	struct fs *fs;
632 	struct buf *bp;
633 	struct thread *td;
634 	ufs_lbn_t lbn;
635 	off_t osize;
636 	int seqcount;
637 	int blkoffset, error, flags, ioflag, resid, size, xfersize;
638 
639 	vp = ap->a_vp;
640 	uio = ap->a_uio;
641 	ioflag = ap->a_ioflag;
642 	if (ap->a_ioflag & IO_EXT)
643 #ifdef notyet
644 		return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
645 #else
646 		panic("ffs_write+IO_EXT");
647 #endif
648 
649 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
650 	ip = VTOI(vp);
651 
652 #ifdef INVARIANTS
653 	if (uio->uio_rw != UIO_WRITE)
654 		panic("ffs_write: mode");
655 #endif
656 
657 	switch (vp->v_type) {
658 	case VREG:
659 		if (ioflag & IO_APPEND)
660 			uio->uio_offset = ip->i_size;
661 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
662 			return (EPERM);
663 		/* FALLTHROUGH */
664 	case VLNK:
665 		break;
666 	case VDIR:
667 		panic("ffs_write: dir write");
668 		break;
669 	default:
670 		panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
671 			(int)uio->uio_offset,
672 			(int)uio->uio_resid
673 		);
674 	}
675 
676 	KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
677 	KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
678 	fs = ip->i_fs;
679 	if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
680 		return (EFBIG);
681 	/*
682 	 * Maybe this should be above the vnode op call, but so long as
683 	 * file servers have no limits, I don't think it matters.
684 	 */
685 	td = uio->uio_td;
686 	if (vp->v_type == VREG && td != NULL) {
687 		PROC_LOCK(td->td_proc);
688 		if (uio->uio_offset + uio->uio_resid >
689 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
690 			psignal(td->td_proc, SIGXFSZ);
691 			PROC_UNLOCK(td->td_proc);
692 			return (EFBIG);
693 		}
694 		PROC_UNLOCK(td->td_proc);
695 	}
696 
697 	resid = uio->uio_resid;
698 	osize = ip->i_size;
699 	if (seqcount > BA_SEQMAX)
700 		flags = BA_SEQMAX << BA_SEQSHIFT;
701 	else
702 		flags = seqcount << BA_SEQSHIFT;
703 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
704 		flags |= IO_SYNC;
705 
706 	for (error = 0; uio->uio_resid > 0;) {
707 		lbn = lblkno(fs, uio->uio_offset);
708 		blkoffset = blkoff(fs, uio->uio_offset);
709 		xfersize = fs->fs_bsize - blkoffset;
710 		if (uio->uio_resid < xfersize)
711 			xfersize = uio->uio_resid;
712 		if (uio->uio_offset + xfersize > ip->i_size)
713 			vnode_pager_setsize(vp, uio->uio_offset + xfersize);
714 
715                 /*
716 		 * We must perform a read-before-write if the transfer size
717 		 * does not cover the entire buffer.
718                  */
719 		if (fs->fs_bsize > xfersize)
720 			flags |= BA_CLRBUF;
721 		else
722 			flags &= ~BA_CLRBUF;
723 /* XXX is uio->uio_offset the right thing here? */
724 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
725 		    ap->a_cred, flags, &bp);
726 		if (error != 0)
727 			break;
728 		/*
729 		 * If the buffer is not valid we have to clear out any
730 		 * garbage data from the pages instantiated for the buffer.
731 		 * If we do not, a failed uiomove() during a write can leave
732 		 * the prior contents of the pages exposed to a userland
733 		 * mmap().  XXX deal with uiomove() errors a better way.
734 		 */
735 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
736 			vfs_bio_clrbuf(bp);
737 		if (ioflag & IO_DIRECT)
738 			bp->b_flags |= B_DIRECT;
739 		if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
740 			bp->b_flags |= B_NOCACHE;
741 
742 		if (uio->uio_offset + xfersize > ip->i_size) {
743 			ip->i_size = uio->uio_offset + xfersize;
744 			DIP_SET(ip, i_size, ip->i_size);
745 		}
746 
747 		size = blksize(fs, ip, lbn) - bp->b_resid;
748 		if (size < xfersize)
749 			xfersize = size;
750 
751 		error =
752 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
753 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
754 		   (LIST_EMPTY(&bp->b_dep))) {
755 			bp->b_flags |= B_RELBUF;
756 		}
757 
758 		/*
759 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
760 		 * if we have a severe page deficiency write the buffer
761 		 * asynchronously.  Otherwise try to cluster, and if that
762 		 * doesn't do it then either do an async write (if O_DIRECT),
763 		 * or a delayed write (if not).
764 		 */
765 		if (ioflag & IO_SYNC) {
766 			(void)bwrite(bp);
767 		} else if (vm_page_count_severe() ||
768 			    buf_dirty_count_severe() ||
769 			    (ioflag & IO_ASYNC)) {
770 			bp->b_flags |= B_CLUSTEROK;
771 			bawrite(bp);
772 		} else if (xfersize + blkoffset == fs->fs_bsize) {
773 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
774 				bp->b_flags |= B_CLUSTEROK;
775 				cluster_write(vp, bp, ip->i_size, seqcount);
776 			} else {
777 				bawrite(bp);
778 			}
779 		} else if (ioflag & IO_DIRECT) {
780 			bp->b_flags |= B_CLUSTEROK;
781 			bawrite(bp);
782 		} else {
783 			bp->b_flags |= B_CLUSTEROK;
784 			bdwrite(bp);
785 		}
786 		if (error || xfersize == 0)
787 			break;
788 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
789 	}
790 	/*
791 	 * If we successfully wrote any data, and we are not the superuser
792 	 * we clear the setuid and setgid bits as a precaution against
793 	 * tampering.
794 	 */
795 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
796 	    ap->a_cred) {
797 		if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) {
798 			ip->i_mode &= ~(ISUID | ISGID);
799 			DIP_SET(ip, i_mode, ip->i_mode);
800 		}
801 	}
802 	if (error) {
803 		if (ioflag & IO_UNIT) {
804 			(void)ffs_truncate(vp, osize,
805 			    IO_NORMAL | (ioflag & IO_SYNC),
806 			    ap->a_cred, uio->uio_td);
807 			uio->uio_offset -= resid - uio->uio_resid;
808 			uio->uio_resid = resid;
809 		}
810 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
811 		error = ffs_update(vp, 1);
812 	return (error);
813 }
814 
815 /*
816  * get page routine
817  */
818 static int
819 ffs_getpages(ap)
820 	struct vop_getpages_args *ap;
821 {
822 	int i;
823 	vm_page_t mreq;
824 	int pcount;
825 
826 	pcount = round_page(ap->a_count) / PAGE_SIZE;
827 	mreq = ap->a_m[ap->a_reqpage];
828 
829 	/*
830 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block,
831 	 * then the entire page is valid.  Since the page may be mapped,
832 	 * user programs might reference data beyond the actual end of file
833 	 * occuring within the page.  We have to zero that data.
834 	 */
835 	VM_OBJECT_LOCK(mreq->object);
836 	if (mreq->valid) {
837 		if (mreq->valid != VM_PAGE_BITS_ALL)
838 			vm_page_zero_invalid(mreq, TRUE);
839 		vm_page_lock_queues();
840 		for (i = 0; i < pcount; i++) {
841 			if (i != ap->a_reqpage) {
842 				vm_page_free(ap->a_m[i]);
843 			}
844 		}
845 		vm_page_unlock_queues();
846 		VM_OBJECT_UNLOCK(mreq->object);
847 		return VM_PAGER_OK;
848 	}
849 	VM_OBJECT_UNLOCK(mreq->object);
850 
851 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
852 					    ap->a_count,
853 					    ap->a_reqpage);
854 }
855 
856 
857 /*
858  * Extended attribute area reading.
859  */
860 static int
861 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
862 {
863 	struct inode *ip;
864 	struct ufs2_dinode *dp;
865 	struct fs *fs;
866 	struct buf *bp;
867 	ufs_lbn_t lbn, nextlbn;
868 	off_t bytesinfile;
869 	long size, xfersize, blkoffset;
870 	int error, orig_resid;
871 
872 	ip = VTOI(vp);
873 	fs = ip->i_fs;
874 	dp = ip->i_din2;
875 
876 #ifdef INVARIANTS
877 	if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
878 		panic("ffs_extread: mode");
879 
880 #endif
881 	orig_resid = uio->uio_resid;
882 	KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
883 	if (orig_resid == 0)
884 		return (0);
885 	KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
886 
887 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
888 		if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
889 			break;
890 		lbn = lblkno(fs, uio->uio_offset);
891 		nextlbn = lbn + 1;
892 
893 		/*
894 		 * size of buffer.  The buffer representing the
895 		 * end of the file is rounded up to the size of
896 		 * the block type ( fragment or full block,
897 		 * depending ).
898 		 */
899 		size = sblksize(fs, dp->di_extsize, lbn);
900 		blkoffset = blkoff(fs, uio->uio_offset);
901 
902 		/*
903 		 * The amount we want to transfer in this iteration is
904 		 * one FS block less the amount of the data before
905 		 * our startpoint (duh!)
906 		 */
907 		xfersize = fs->fs_bsize - blkoffset;
908 
909 		/*
910 		 * But if we actually want less than the block,
911 		 * or the file doesn't have a whole block more of data,
912 		 * then use the lesser number.
913 		 */
914 		if (uio->uio_resid < xfersize)
915 			xfersize = uio->uio_resid;
916 		if (bytesinfile < xfersize)
917 			xfersize = bytesinfile;
918 
919 		if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
920 			/*
921 			 * Don't do readahead if this is the end of the info.
922 			 */
923 			error = bread(vp, -1 - lbn, size, NOCRED, &bp);
924 		} else {
925 			/*
926 			 * If we have a second block, then
927 			 * fire off a request for a readahead
928 			 * as well as a read. Note that the 4th and 5th
929 			 * arguments point to arrays of the size specified in
930 			 * the 6th argument.
931 			 */
932 			int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
933 
934 			nextlbn = -1 - nextlbn;
935 			error = breadn(vp, -1 - lbn,
936 			    size, &nextlbn, &nextsize, 1, NOCRED, &bp);
937 		}
938 		if (error) {
939 			brelse(bp);
940 			bp = NULL;
941 			break;
942 		}
943 
944 		/*
945 		 * If IO_DIRECT then set B_DIRECT for the buffer.  This
946 		 * will cause us to attempt to release the buffer later on
947 		 * and will cause the buffer cache to attempt to free the
948 		 * underlying pages.
949 		 */
950 		if (ioflag & IO_DIRECT)
951 			bp->b_flags |= B_DIRECT;
952 
953 		/*
954 		 * We should only get non-zero b_resid when an I/O error
955 		 * has occurred, which should cause us to break above.
956 		 * However, if the short read did not cause an error,
957 		 * then we want to ensure that we do not uiomove bad
958 		 * or uninitialized data.
959 		 */
960 		size -= bp->b_resid;
961 		if (size < xfersize) {
962 			if (size == 0)
963 				break;
964 			xfersize = size;
965 		}
966 
967 		error = uiomove((char *)bp->b_data + blkoffset,
968 					(int)xfersize, uio);
969 		if (error)
970 			break;
971 
972 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
973 		   (LIST_EMPTY(&bp->b_dep))) {
974 			/*
975 			 * If there are no dependencies, and it's VMIO,
976 			 * then we don't need the buf, mark it available
977 			 * for freeing. The VM has the data.
978 			 */
979 			bp->b_flags |= B_RELBUF;
980 			brelse(bp);
981 		} else {
982 			/*
983 			 * Otherwise let whoever
984 			 * made the request take care of
985 			 * freeing it. We just queue
986 			 * it onto another list.
987 			 */
988 			bqrelse(bp);
989 		}
990 	}
991 
992 	/*
993 	 * This can only happen in the case of an error
994 	 * because the loop above resets bp to NULL on each iteration
995 	 * and on normal completion has not set a new value into it.
996 	 * so it must have come from a 'break' statement
997 	 */
998 	if (bp != NULL) {
999 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1000 		   (LIST_EMPTY(&bp->b_dep))) {
1001 			bp->b_flags |= B_RELBUF;
1002 			brelse(bp);
1003 		} else {
1004 			bqrelse(bp);
1005 		}
1006 	}
1007 
1008 	if ((error == 0 || uio->uio_resid != orig_resid) &&
1009 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0) {
1010 		VI_LOCK(vp);
1011 		ip->i_flag |= IN_ACCESS;
1012 		VI_UNLOCK(vp);
1013 	}
1014 	return (error);
1015 }
1016 
1017 /*
1018  * Extended attribute area writing.
1019  */
1020 static int
1021 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1022 {
1023 	struct inode *ip;
1024 	struct ufs2_dinode *dp;
1025 	struct fs *fs;
1026 	struct buf *bp;
1027 	ufs_lbn_t lbn;
1028 	off_t osize;
1029 	int blkoffset, error, flags, resid, size, xfersize;
1030 
1031 	ip = VTOI(vp);
1032 	fs = ip->i_fs;
1033 	dp = ip->i_din2;
1034 
1035 	KASSERT(!(ip->i_flag & IN_SPACECOUNTED), ("inode %u: inode is dead",
1036 	    ip->i_number));
1037 
1038 #ifdef INVARIANTS
1039 	if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1040 		panic("ffs_extwrite: mode");
1041 #endif
1042 
1043 	if (ioflag & IO_APPEND)
1044 		uio->uio_offset = dp->di_extsize;
1045 	KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1046 	KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1047 	if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize)
1048 		return (EFBIG);
1049 
1050 	resid = uio->uio_resid;
1051 	osize = dp->di_extsize;
1052 	flags = IO_EXT;
1053 	if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
1054 		flags |= IO_SYNC;
1055 
1056 	for (error = 0; uio->uio_resid > 0;) {
1057 		lbn = lblkno(fs, uio->uio_offset);
1058 		blkoffset = blkoff(fs, uio->uio_offset);
1059 		xfersize = fs->fs_bsize - blkoffset;
1060 		if (uio->uio_resid < xfersize)
1061 			xfersize = uio->uio_resid;
1062 
1063 		/*
1064 		 * We must perform a read-before-write if the transfer size
1065 		 * does not cover the entire buffer.
1066                  */
1067 		if (fs->fs_bsize > xfersize)
1068 			flags |= BA_CLRBUF;
1069 		else
1070 			flags &= ~BA_CLRBUF;
1071 		error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1072 		    ucred, flags, &bp);
1073 		if (error != 0)
1074 			break;
1075 		/*
1076 		 * If the buffer is not valid we have to clear out any
1077 		 * garbage data from the pages instantiated for the buffer.
1078 		 * If we do not, a failed uiomove() during a write can leave
1079 		 * the prior contents of the pages exposed to a userland
1080 		 * mmap().  XXX deal with uiomove() errors a better way.
1081 		 */
1082 		if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1083 			vfs_bio_clrbuf(bp);
1084 		if (ioflag & IO_DIRECT)
1085 			bp->b_flags |= B_DIRECT;
1086 
1087 		if (uio->uio_offset + xfersize > dp->di_extsize)
1088 			dp->di_extsize = uio->uio_offset + xfersize;
1089 
1090 		size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1091 		if (size < xfersize)
1092 			xfersize = size;
1093 
1094 		error =
1095 		    uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1096 		if ((ioflag & (IO_VMIO|IO_DIRECT)) &&
1097 		   (LIST_EMPTY(&bp->b_dep))) {
1098 			bp->b_flags |= B_RELBUF;
1099 		}
1100 
1101 		/*
1102 		 * If IO_SYNC each buffer is written synchronously.  Otherwise
1103 		 * if we have a severe page deficiency write the buffer
1104 		 * asynchronously.  Otherwise try to cluster, and if that
1105 		 * doesn't do it then either do an async write (if O_DIRECT),
1106 		 * or a delayed write (if not).
1107 		 */
1108 		if (ioflag & IO_SYNC) {
1109 			(void)bwrite(bp);
1110 		} else if (vm_page_count_severe() ||
1111 			    buf_dirty_count_severe() ||
1112 			    xfersize + blkoffset == fs->fs_bsize ||
1113 			    (ioflag & (IO_ASYNC | IO_DIRECT)))
1114 			bawrite(bp);
1115 		else
1116 			bdwrite(bp);
1117 		if (error || xfersize == 0)
1118 			break;
1119 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1120 	}
1121 	/*
1122 	 * If we successfully wrote any data, and we are not the superuser
1123 	 * we clear the setuid and setgid bits as a precaution against
1124 	 * tampering.
1125 	 */
1126 	if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1127 		if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) {
1128 			ip->i_mode &= ~(ISUID | ISGID);
1129 			dp->di_mode = ip->i_mode;
1130 		}
1131 	}
1132 	if (error) {
1133 		if (ioflag & IO_UNIT) {
1134 			(void)ffs_truncate(vp, osize,
1135 			    IO_EXT | (ioflag&IO_SYNC), ucred, uio->uio_td);
1136 			uio->uio_offset -= resid - uio->uio_resid;
1137 			uio->uio_resid = resid;
1138 		}
1139 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1140 		error = ffs_update(vp, 1);
1141 	return (error);
1142 }
1143 
1144 
1145 /*
1146  * Vnode operating to retrieve a named extended attribute.
1147  *
1148  * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1149  * the length of the EA, and possibly the pointer to the entry and to the data.
1150  */
1151 static int
1152 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac)
1153 {
1154 	u_char *p, *pe, *pn, *p0;
1155 	int eapad1, eapad2, ealength, ealen, nlen;
1156 	uint32_t ul;
1157 
1158 	pe = ptr + length;
1159 	nlen = strlen(name);
1160 
1161 	for (p = ptr; p < pe; p = pn) {
1162 		p0 = p;
1163 		bcopy(p, &ul, sizeof(ul));
1164 		pn = p + ul;
1165 		/* make sure this entry is complete */
1166 		if (pn > pe)
1167 			break;
1168 		p += sizeof(uint32_t);
1169 		if (*p != nspace)
1170 			continue;
1171 		p++;
1172 		eapad2 = *p++;
1173 		if (*p != nlen)
1174 			continue;
1175 		p++;
1176 		if (bcmp(p, name, nlen))
1177 			continue;
1178 		ealength = sizeof(uint32_t) + 3 + nlen;
1179 		eapad1 = 8 - (ealength % 8);
1180 		if (eapad1 == 8)
1181 			eapad1 = 0;
1182 		ealength += eapad1;
1183 		ealen = ul - ealength - eapad2;
1184 		p += nlen + eapad1;
1185 		if (eap != NULL)
1186 			*eap = p0;
1187 		if (eac != NULL)
1188 			*eac = p;
1189 		return (ealen);
1190 	}
1191 	return(-1);
1192 }
1193 
1194 static int
1195 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra)
1196 {
1197 	struct inode *ip;
1198 	struct ufs2_dinode *dp;
1199 	struct fs *fs;
1200 	struct uio luio;
1201 	struct iovec liovec;
1202 	int easize, error;
1203 	u_char *eae;
1204 
1205 	ip = VTOI(vp);
1206 	fs = ip->i_fs;
1207 	dp = ip->i_din2;
1208 	easize = dp->di_extsize;
1209 	if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize)
1210 		return (EFBIG);
1211 
1212 	eae = malloc(easize + extra, M_TEMP, M_WAITOK);
1213 
1214 	liovec.iov_base = eae;
1215 	liovec.iov_len = easize;
1216 	luio.uio_iov = &liovec;
1217 	luio.uio_iovcnt = 1;
1218 	luio.uio_offset = 0;
1219 	luio.uio_resid = easize;
1220 	luio.uio_segflg = UIO_SYSSPACE;
1221 	luio.uio_rw = UIO_READ;
1222 	luio.uio_td = td;
1223 
1224 	error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1225 	if (error) {
1226 		free(eae, M_TEMP);
1227 		return(error);
1228 	}
1229 	*p = eae;
1230 	return (0);
1231 }
1232 
1233 static int
1234 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1235 {
1236 	struct inode *ip;
1237 	struct ufs2_dinode *dp;
1238 	int error;
1239 
1240 	ip = VTOI(vp);
1241 
1242 	if (ip->i_ea_area != NULL)
1243 		return (EBUSY);
1244 	dp = ip->i_din2;
1245 	error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0);
1246 	if (error)
1247 		return (error);
1248 	ip->i_ea_len = dp->di_extsize;
1249 	ip->i_ea_error = 0;
1250 	return (0);
1251 }
1252 
1253 /*
1254  * Vnode extattr transaction commit/abort
1255  */
1256 static int
1257 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1258 {
1259 	struct inode *ip;
1260 	struct uio luio;
1261 	struct iovec liovec;
1262 	int error;
1263 	struct ufs2_dinode *dp;
1264 
1265 	ip = VTOI(vp);
1266 	if (ip->i_ea_area == NULL)
1267 		return (EINVAL);
1268 	dp = ip->i_din2;
1269 	error = ip->i_ea_error;
1270 	if (commit && error == 0) {
1271 		if (cred == NOCRED)
1272 			cred =  vp->v_mount->mnt_cred;
1273 		liovec.iov_base = ip->i_ea_area;
1274 		liovec.iov_len = ip->i_ea_len;
1275 		luio.uio_iov = &liovec;
1276 		luio.uio_iovcnt = 1;
1277 		luio.uio_offset = 0;
1278 		luio.uio_resid = ip->i_ea_len;
1279 		luio.uio_segflg = UIO_SYSSPACE;
1280 		luio.uio_rw = UIO_WRITE;
1281 		luio.uio_td = td;
1282 		/* XXX: I'm not happy about truncating to zero size */
1283 		if (ip->i_ea_len < dp->di_extsize)
1284 			error = ffs_truncate(vp, 0, IO_EXT, cred, td);
1285 		error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1286 	}
1287 	free(ip->i_ea_area, M_TEMP);
1288 	ip->i_ea_area = NULL;
1289 	ip->i_ea_len = 0;
1290 	ip->i_ea_error = 0;
1291 	return (error);
1292 }
1293 
1294 /*
1295  * Vnode extattr strategy routine for fifos.
1296  *
1297  * We need to check for a read or write of the external attributes.
1298  * Otherwise we just fall through and do the usual thing.
1299  */
1300 static int
1301 ffsext_strategy(struct vop_strategy_args *ap)
1302 /*
1303 struct vop_strategy_args {
1304 	struct vnodeop_desc *a_desc;
1305 	struct vnode *a_vp;
1306 	struct buf *a_bp;
1307 };
1308 */
1309 {
1310 	struct vnode *vp;
1311 	daddr_t lbn;
1312 
1313 	vp = ap->a_vp;
1314 	lbn = ap->a_bp->b_lblkno;
1315 	if (VTOI(vp)->i_fs->fs_magic == FS_UFS2_MAGIC &&
1316 	    lbn < 0 && lbn >= -NXADDR)
1317 		return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1318 	if (vp->v_type == VFIFO)
1319 		return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1320 	panic("spec nodes went here");
1321 }
1322 
1323 /*
1324  * Vnode extattr transaction commit/abort
1325  */
1326 static int
1327 ffs_openextattr(struct vop_openextattr_args *ap)
1328 /*
1329 struct vop_openextattr_args {
1330 	struct vnodeop_desc *a_desc;
1331 	struct vnode *a_vp;
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_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1347 }
1348 
1349 
1350 /*
1351  * Vnode extattr transaction commit/abort
1352  */
1353 static int
1354 ffs_closeextattr(struct vop_closeextattr_args *ap)
1355 /*
1356 struct vop_closeextattr_args {
1357 	struct vnodeop_desc *a_desc;
1358 	struct vnode *a_vp;
1359 	int a_commit;
1360 	IN struct ucred *a_cred;
1361 	IN struct thread *a_td;
1362 };
1363 */
1364 {
1365 	struct inode *ip;
1366 	struct fs *fs;
1367 
1368 	ip = VTOI(ap->a_vp);
1369 	fs = ip->i_fs;
1370 
1371 	if (ap->a_vp->v_type == VCHR)
1372 		return (EOPNOTSUPP);
1373 
1374 	if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY))
1375 		return (EROFS);
1376 
1377 	return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td));
1378 }
1379 
1380 /*
1381  * Vnode operation to remove a named attribute.
1382  */
1383 static int
1384 ffs_deleteextattr(struct vop_deleteextattr_args *ap)
1385 /*
1386 vop_deleteextattr {
1387 	IN struct vnode *a_vp;
1388 	IN int a_attrnamespace;
1389 	IN const char *a_name;
1390 	IN struct ucred *a_cred;
1391 	IN struct thread *a_td;
1392 };
1393 */
1394 {
1395 	struct inode *ip;
1396 	struct fs *fs;
1397 	uint32_t ealength, ul;
1398 	int ealen, olen, eapad1, eapad2, error, i, easize;
1399 	u_char *eae, *p;
1400 	int stand_alone;
1401 
1402 	ip = VTOI(ap->a_vp);
1403 	fs = ip->i_fs;
1404 
1405 	if (ap->a_vp->v_type == VCHR)
1406 		return (EOPNOTSUPP);
1407 
1408 	if (strlen(ap->a_name) == 0)
1409 		return (EINVAL);
1410 
1411 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1412 		return (EROFS);
1413 
1414 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1415 	    ap->a_cred, ap->a_td, IWRITE);
1416 	if (error) {
1417 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1418 			ip->i_ea_error = error;
1419 		return (error);
1420 	}
1421 
1422 	if (ip->i_ea_area == NULL) {
1423 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1424 		if (error)
1425 			return (error);
1426 		stand_alone = 1;
1427 	} else {
1428 		stand_alone = 0;
1429 	}
1430 
1431 	ealength = eapad1 = ealen = eapad2 = 0;
1432 
1433 	eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1434 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1435 	easize = ip->i_ea_len;
1436 
1437 	olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1438 	    &p, NULL);
1439 	if (olen == -1) {
1440 		/* delete but nonexistent */
1441 		free(eae, M_TEMP);
1442 		if (stand_alone)
1443 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1444 		return(ENOATTR);
1445 	}
1446 	bcopy(p, &ul, sizeof ul);
1447 	i = p - eae + ul;
1448 	if (ul != ealength) {
1449 		bcopy(p + ul, p + ealength, easize - i);
1450 		easize += (ealength - ul);
1451 	}
1452 	if (easize > NXADDR * fs->fs_bsize) {
1453 		free(eae, M_TEMP);
1454 		if (stand_alone)
1455 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1456 		else if (ip->i_ea_error == 0)
1457 			ip->i_ea_error = ENOSPC;
1458 		return(ENOSPC);
1459 	}
1460 	p = ip->i_ea_area;
1461 	ip->i_ea_area = eae;
1462 	ip->i_ea_len = easize;
1463 	free(p, M_TEMP);
1464 	if (stand_alone)
1465 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1466 	return(error);
1467 }
1468 
1469 /*
1470  * Vnode operation to retrieve a named extended attribute.
1471  */
1472 static int
1473 ffs_getextattr(struct vop_getextattr_args *ap)
1474 /*
1475 vop_getextattr {
1476 	IN struct vnode *a_vp;
1477 	IN int a_attrnamespace;
1478 	IN const char *a_name;
1479 	INOUT struct uio *a_uio;
1480 	OUT size_t *a_size;
1481 	IN struct ucred *a_cred;
1482 	IN struct thread *a_td;
1483 };
1484 */
1485 {
1486 	struct inode *ip;
1487 	struct fs *fs;
1488 	u_char *eae, *p;
1489 	unsigned easize;
1490 	int error, ealen, stand_alone;
1491 
1492 	ip = VTOI(ap->a_vp);
1493 	fs = ip->i_fs;
1494 
1495 	if (ap->a_vp->v_type == VCHR)
1496 		return (EOPNOTSUPP);
1497 
1498 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1499 	    ap->a_cred, ap->a_td, IREAD);
1500 	if (error)
1501 		return (error);
1502 
1503 	if (ip->i_ea_area == NULL) {
1504 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1505 		if (error)
1506 			return (error);
1507 		stand_alone = 1;
1508 	} else {
1509 		stand_alone = 0;
1510 	}
1511 	eae = ip->i_ea_area;
1512 	easize = ip->i_ea_len;
1513 
1514 	ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1515 	    NULL, &p);
1516 	if (ealen >= 0) {
1517 		error = 0;
1518 		if (ap->a_size != NULL)
1519 			*ap->a_size = ealen;
1520 		else if (ap->a_uio != NULL)
1521 			error = uiomove(p, ealen, ap->a_uio);
1522 	} else
1523 		error = ENOATTR;
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 retrieve extended attributes on a vnode.
1531  */
1532 static int
1533 ffs_listextattr(struct vop_listextattr_args *ap)
1534 /*
1535 vop_listextattr {
1536 	IN struct vnode *a_vp;
1537 	IN int a_attrnamespace;
1538 	INOUT struct uio *a_uio;
1539 	OUT size_t *a_size;
1540 	IN struct ucred *a_cred;
1541 	IN struct thread *a_td;
1542 };
1543 */
1544 {
1545 	struct inode *ip;
1546 	struct fs *fs;
1547 	u_char *eae, *p, *pe, *pn;
1548 	unsigned easize;
1549 	uint32_t ul;
1550 	int error, ealen, stand_alone;
1551 
1552 	ip = VTOI(ap->a_vp);
1553 	fs = ip->i_fs;
1554 
1555 	if (ap->a_vp->v_type == VCHR)
1556 		return (EOPNOTSUPP);
1557 
1558 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1559 	    ap->a_cred, ap->a_td, IREAD);
1560 	if (error)
1561 		return (error);
1562 
1563 	if (ip->i_ea_area == NULL) {
1564 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1565 		if (error)
1566 			return (error);
1567 		stand_alone = 1;
1568 	} else {
1569 		stand_alone = 0;
1570 	}
1571 	eae = ip->i_ea_area;
1572 	easize = ip->i_ea_len;
1573 
1574 	error = 0;
1575 	if (ap->a_size != NULL)
1576 		*ap->a_size = 0;
1577 	pe = eae + easize;
1578 	for(p = eae; error == 0 && p < pe; p = pn) {
1579 		bcopy(p, &ul, sizeof(ul));
1580 		pn = p + ul;
1581 		if (pn > pe)
1582 			break;
1583 		p += sizeof(ul);
1584 		if (*p++ != ap->a_attrnamespace)
1585 			continue;
1586 		p++;	/* pad2 */
1587 		ealen = *p;
1588 		if (ap->a_size != NULL) {
1589 			*ap->a_size += ealen + 1;
1590 		} else if (ap->a_uio != NULL) {
1591 			error = uiomove(p, ealen + 1, ap->a_uio);
1592 		}
1593 	}
1594 	if (stand_alone)
1595 		ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1596 	return(error);
1597 }
1598 
1599 /*
1600  * Vnode operation to set a named attribute.
1601  */
1602 static int
1603 ffs_setextattr(struct vop_setextattr_args *ap)
1604 /*
1605 vop_setextattr {
1606 	IN struct vnode *a_vp;
1607 	IN int a_attrnamespace;
1608 	IN const char *a_name;
1609 	INOUT struct uio *a_uio;
1610 	IN struct ucred *a_cred;
1611 	IN struct thread *a_td;
1612 };
1613 */
1614 {
1615 	struct inode *ip;
1616 	struct fs *fs;
1617 	uint32_t ealength, ul;
1618 	int ealen, olen, eapad1, eapad2, error, i, easize;
1619 	u_char *eae, *p;
1620 	int stand_alone;
1621 
1622 	ip = VTOI(ap->a_vp);
1623 	fs = ip->i_fs;
1624 
1625 	if (ap->a_vp->v_type == VCHR)
1626 		return (EOPNOTSUPP);
1627 
1628 	if (strlen(ap->a_name) == 0)
1629 		return (EINVAL);
1630 
1631 	/* XXX Now unsupported API to delete EAs using NULL uio. */
1632 	if (ap->a_uio == NULL)
1633 		return (EOPNOTSUPP);
1634 
1635 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1636 		return (EROFS);
1637 
1638 	error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1639 	    ap->a_cred, ap->a_td, IWRITE);
1640 	if (error) {
1641 		if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1642 			ip->i_ea_error = error;
1643 		return (error);
1644 	}
1645 
1646 	if (ip->i_ea_area == NULL) {
1647 		error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1648 		if (error)
1649 			return (error);
1650 		stand_alone = 1;
1651 	} else {
1652 		stand_alone = 0;
1653 	}
1654 
1655 	ealen = ap->a_uio->uio_resid;
1656 	ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1657 	eapad1 = 8 - (ealength % 8);
1658 	if (eapad1 == 8)
1659 		eapad1 = 0;
1660 	eapad2 = 8 - (ealen % 8);
1661 	if (eapad2 == 8)
1662 		eapad2 = 0;
1663 	ealength += eapad1 + ealen + eapad2;
1664 
1665 	eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1666 	bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1667 	easize = ip->i_ea_len;
1668 
1669 	olen = ffs_findextattr(eae, easize,
1670 	    ap->a_attrnamespace, ap->a_name, &p, NULL);
1671         if (olen == -1) {
1672 		/* new, append at end */
1673 		p = eae + easize;
1674 		easize += ealength;
1675 	} else {
1676 		bcopy(p, &ul, sizeof ul);
1677 		i = p - eae + ul;
1678 		if (ul != ealength) {
1679 			bcopy(p + ul, p + ealength, easize - i);
1680 			easize += (ealength - ul);
1681 		}
1682 	}
1683 	if (easize > NXADDR * fs->fs_bsize) {
1684 		free(eae, M_TEMP);
1685 		if (stand_alone)
1686 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1687 		else if (ip->i_ea_error == 0)
1688 			ip->i_ea_error = ENOSPC;
1689 		return(ENOSPC);
1690 	}
1691 	bcopy(&ealength, p, sizeof(ealength));
1692 	p += sizeof(ealength);
1693 	*p++ = ap->a_attrnamespace;
1694 	*p++ = eapad2;
1695 	*p++ = strlen(ap->a_name);
1696 	strcpy(p, ap->a_name);
1697 	p += strlen(ap->a_name);
1698 	bzero(p, eapad1);
1699 	p += eapad1;
1700 	error = uiomove(p, ealen, ap->a_uio);
1701 	if (error) {
1702 		free(eae, M_TEMP);
1703 		if (stand_alone)
1704 			ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1705 		else if (ip->i_ea_error == 0)
1706 			ip->i_ea_error = error;
1707 		return(error);
1708 	}
1709 	p += ealen;
1710 	bzero(p, eapad2);
1711 
1712 	p = ip->i_ea_area;
1713 	ip->i_ea_area = eae;
1714 	ip->i_ea_len = easize;
1715 	free(p, M_TEMP);
1716 	if (stand_alone)
1717 		error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td);
1718 	return(error);
1719 }
1720 
1721 /*
1722  * Vnode pointer to File handle
1723  */
1724 static int
1725 ffs_vptofh(struct vop_vptofh_args *ap)
1726 /*
1727 vop_vptofh {
1728 	IN struct vnode *a_vp;
1729 	IN struct fid *a_fhp;
1730 };
1731 */
1732 {
1733 	struct inode *ip;
1734 	struct ufid *ufhp;
1735 
1736 	ip = VTOI(ap->a_vp);
1737 	ufhp = (struct ufid *)ap->a_fhp;
1738 	ufhp->ufid_len = sizeof(struct ufid);
1739 	ufhp->ufid_ino = ip->i_number;
1740 	ufhp->ufid_gen = ip->i_gen;
1741 	return (0);
1742 }
1743